id
stringlengths 27
33
| source
stringclasses 1
value | format
stringclasses 1
value | text
stringlengths 13
1.81M
|
|---|---|---|---|
warning/0507/cond-mat0507506.html | ar5iv | text | # Collision-Dependent Atom Tunneling Rate in Bose-Einstein Condensates
## Abstract
We show that the interaction (cross-collision) between atoms trapped in distinct sites of a double-well potential can significantly increase the atom tunneling rate for special trap configurations leading to an effective linear Rabi regime of population oscillation between the trap wells. The inclusion of cross-collisional effects significantly extends the validity of the two-mode model approach allowing it to be alternatively employed to explain the recently observed increase of tunneling rates due to nonlinear interactions.
Atom optics and the physics of ultracold matter waves have witnessed rapid theoretical and experimental progress since the achievement of atomic vapor Bose-Einstein condensates (BECs) BEC . The interest in such a system is quite wide ranged as it has opened new technological frontiers zoller ; bectelep , and has renewed the investigation on many-body physics fundamental issues once BECs trapped in optical lattices offers a powerful toolbox for implementation of condensed matter model systems with highly controllable parameters leggett ; bloch ; qtbox .
The simplest system to exhibit non-trivial many-body phenomena is the two-mode BEC trapped in a double well potential. For this system it was previously predicted corney1 ; raghavan1 a plethora of dynamical regimes, including Josephson oscillation javanainen of atoms between distinct trap sites, and macroscopic self-trapping of atomic populations, where coherent tunneling is suppressed when the number of trapped atoms exceeds a critical value. The first realization of an atomic single BEC Josephson junction has indeed experimentally demonstrated albiez those predictions javanainen ; corney1 ; raghavan1 . Moreover an increase of one order of magnitude of the atomic tunneling rate (from 2 to 25 Hz) due to nonlinear atomic interactions was then observed, which cannot be explained inside those previous models. In those models corney1 ; raghavan1 a two-mode approximation (TMA) is invariably applied, in which each trap well is populated by one single localized condensate mode. However due to a localization assumption the BEC dynamics is considered by regarding only self-collisions (on-site interaction) between atoms of individual condensate modes. Notwithstanding, distinct trap configurations can be as such that cross-collisions (off-site interaction) of atoms in distinct BEC modes cannot be neglected and the observed coherent phenomena are affected by cross-collision induced effects. This situation has been previously considered either by directly integrating a non-polynomial Schrödinger equation albiez ; salasnich or by extending the TMA through Gross-Pitaeviskii equations ananikian .
In this paper we investigate the cross-collisional effects over the quantum dynamics of a BEC trapped in a double-well potential inside the TMA. By appropriate manipulation of the trapping potential the system shows a very rich quantum dynamics related to the cross-collisional rates. In absence of cross-collisions, self-trapping is observed in agreement with previous theoretical predictions corney1 ; raghavan1 . Remarkably, the inclusion of cross-collisional effects can inhibit self-trapping, resulting in an effective coherent oscillation (effective Rabi regime) of atoms between the two wells with an effective Rabi frequency, dependent on the cross-collision rate and the number of atoms in the sample. Moreover cross-collisions increase the range of validity of the TMA (when the many-body interactions produce only slight changes on the ground-state properties of the individual potential sites). The new limit can be as such that the TMA is indeed valid for the number of atoms present in the experiment of Ref. albiez , and thus can be employed for alternative explanation albiez ; ananikian of the observed increase of tunelling rate through off-site interactions.
We consider an atomic BEC trapped in a symmetric double-well potential $`V(𝐫)`$ with minima at $`𝐫_\mathrm{𝟏}`$, and $`𝐫_\mathrm{𝟐}`$, such that $`V(𝐫_{\mathrm{𝟏},\mathrm{𝟐}})=0`$, the many-body Hamiltonian is given by
$`\widehat{H}`$ $`=`$ $`{\displaystyle d^3r\widehat{\psi }^{}(𝐫)\left(\frac{\mathrm{}^2}{2m}^2+V(𝐫)\right)\widehat{\psi }(𝐫)}`$ (1)
$`+{\displaystyle \frac{1}{2}}U_0{\displaystyle d^3r\widehat{\psi }^{}(𝐫)\widehat{\psi }^{}(𝐫)\widehat{\psi }(𝐫)\widehat{\psi }(𝐫)},`$
where $`m`$ is the atomic mass, $`U_0=4\pi \mathrm{}^2a/m`$ measures the strength of the two-body interaction, $`a`$ is the s-wave scattering length, $`\widehat{\psi }(𝐫,t)`$ and $`\widehat{\psi }^{}(𝐫,t)`$ are the Heisenberg picture field operators. We must emphasize that throughout our calculations we assume that the barrier height separating the two wells is sufficiently larger than the condensate chemical potential, $`V_0>\mu _c`$, so that a TMA can be considered to describe the system williams . The derivation of the two-mode Hamiltonian in a double-well trap follows the standard procedures of Ref. corney1 \- The lower energy eigenstates of the global double-well are approximated as the symmetric and anti-symmetric combinations $`u_\pm (𝐫)\frac{1}{\sqrt{2}}[u_1(𝐫)\pm u_2(𝐫)]`$, and the field operators are expanded in terms of the local modes $`u_j(r)`$, $`j=1,2`$ and the Heisenberg picture annihilation operator as $`\widehat{a}_j(t)=d^3𝐫u_j^{}(𝐫)\widehat{\psi }(𝐫,t)`$ so that $`[\widehat{a}_j,\widehat{a}_k^{}]\delta _{jk}`$. The potential expanded around each minimum is $`V(𝐫)=\stackrel{~}{V}^{(2)}(𝐫𝐫_j)+\mathrm{}`$j=1,2, where $`\stackrel{~}{V}^{(2)}(𝐫𝐫_j)`$ is the harmonic approximation to the potential in the vicinity of each minimum. The normalized single-particle ground-state $`u_j(𝐫)`$ of the local potential $`\stackrel{~}{V}^{(2)}(𝐫𝐫_j)`$, with energy $`E_0`$, defines the local mode solutions of the individual wells. The tunneling frequency $`\mathrm{\Omega }`$ between the two minima is then given by the energy level splitting of these two lowest states, $`\mathrm{\Omega }=2R/\mathrm{}`$, where $`R=d^3𝐫u_1^{}(𝐫)[V(r)\stackrel{~}{V}^{(2)}(𝐫𝐫_1)]u_2(𝐫).`$ By noticing that the total number of atoms $`N`$ is a conserved quantity, after some algebraic manipulation, the following Hamiltonian is then obtained in the interaction picture
$`\widehat{H}`$ $`=`$ $`\mathrm{}[2\mathrm{\Lambda }(N1)+\mathrm{\Omega }](\widehat{a}_1^{}\widehat{a}_2+\widehat{a}_2^{}\widehat{a}_1)+\mathrm{}\eta (\widehat{a}_1^{}\widehat{a}_2+\widehat{a}_2^{}\widehat{a}_1)^2`$ (2)
$`+\mathrm{}(\kappa \eta )[(\widehat{a}_1^{})^2(\widehat{a}_1^2)+(\widehat{a}_2^{})^2(\widehat{a}_2^2)]+\mathrm{}\eta N(N2)`$
where $`\kappa =\frac{U_0}{2\mathrm{}V_{eff}}`$ is the self-collision rate and $`\eta =(\frac{U_0}{2\mathrm{}})d^3𝐫u_i^{}u_ju_i^{}u_{0j}`$, and $`\mathrm{\Lambda }=(\frac{U_0}{2\mathrm{}})d^3𝐫u_j^{}u_ju_i^{}u_i,`$ are cross-collisional rates. Here $`V_{eff}^1=d^3𝐫|u_j|^4`$ is the $`j`$-mode effective volume. It is immediate that for $`\mathrm{}\kappa 0`$, the Hamiltonian (2) reduces to that previously investigated in the study of tunneling in condensates javanainen . Also for $`\mathrm{}\eta 0`$ and $`\mathrm{}\mathrm{\Lambda }0`$, the Hamiltonian (2) reduces itself to that discussed in raghavan1 ; corney1 , accounting for tunneling oscillations as well as for population self-trapping. As $`\eta \kappa `$ in Hamiltonian (2) a new dynamical regime of stable long-time tunneling (hereafter called effective Rabi regime) can be attained.
Introducing the Schwinger angular momentum operators $`\widehat{J_x}=(\widehat{a}_2^{}\widehat{a}_2\widehat{a}_1^{}\widehat{a}_1)/2,\widehat{J_y}=i(\widehat{a}_2^{}\widehat{a}_1\widehat{a}_1^{}\widehat{a}_2)/2,\widehat{J_z}=(\widehat{a}_1^{}\widehat{a}_2+\widehat{a}_2^{}\widehat{a}_1)/2,`$ the Hamiltonian (2) can be rewritten as
$$\widehat{H}=\mathrm{}\mathrm{\Omega }^{}\widehat{J}_z+4\mathrm{}\eta \widehat{J}_z^2+2\mathrm{}(\kappa \eta )\widehat{J}_x^2,$$
(3)
where we have neglected constant energy shifts dependent on $`N`$ and have defined a new effective tunneling rate $`\mathrm{\Omega }^{}2\left[\mathrm{\Omega }+2\mathrm{\Lambda }(N1)\right]`$, which is dependent on the number of atoms in the atomic sample and the cross-collisional rate $`\mathrm{\Lambda }`$. In fact $`\mathrm{\Lambda }`$ is very small compared to $`\mathrm{\Omega }`$, but for a sufficiently large number of atoms the additional tunneling term may lead to observable effects whenever $`2\mathrm{\Lambda }(N1)\mathrm{\Omega }`$. Furthermore the third term of Eq. (3) shows that the cross-collisional rate $`\eta `$ competes with the self-collision rate $`\kappa `$ leading to an effective on-site collision rate $`\kappa ^{}\kappa \eta `$. Since $`0\eta \kappa `$, $`\kappa ^{}`$ on Eq. (3) can be disregarded in a trap configuration where $`\eta \kappa `$ with $`\mathrm{\Omega }^{}\kappa ^{}`$. To infer the validity and implications of such a regime we consider a specific trapping potential footnote2 of the form $`V(r)=b\left(x^2\frac{d}{2b}\right)^2+\frac{1}{2}m\omega _t^2(y^2+z^2)`$, where the inter-well coupling occurs along $`x`$, and $`\omega _t`$ is the trap frequency in the y-z plane. This potential has fixed points at $`q_o^2=\frac{d}{2b}`$. The position uncertainty for a harmonic oscillator in the ground state is $`x_0\sqrt{\frac{\mathrm{}}{2m\omega _0}}`$, with $`\omega _0=\sqrt{4d/m}`$. For a suitable choice of the barrier height only two energy eigenstates lie beneath the barrier. Assuming $`\omega _t=\omega _0`$ the local mode on each well is then given by $`u_j(r)=(\frac{1}{2\pi x_0^2})^{\frac{3}{4}}\mathrm{exp}\left(\frac{[(x_jq_0)^2+y_j^2+z_j^2}{4x_0^2}\right).`$ The collision rates may then be evaluated to give $`\kappa =(\frac{U_0}{16\mathrm{}})(\frac{1}{\pi x_0^2})^{3/2}`$, $`\eta =\kappa \mathrm{exp}(q_0^2/2x_0^2)`$, and $`\mathrm{\Lambda }=\kappa \mathrm{exp}(3q_0^2/4x_0^2)`$, while $`\mathrm{\Omega }=\frac{q_0^2\omega _0}{x_0^2}\mathrm{exp}(q_0^2/2x_0^2)`$.
Firstly we shall investigate the validity of the localization of the TMA considered in the present description. It is valid only if the many-body interactions produce small modifications on the ground-state of the individual potential wells footnote3 . In the absence of cross-collisions corney1 the TMA is limited to $`\mathrm{}\omega _0=\frac{\mathrm{}^2}{2mx_0^2}N\mathrm{}\kappa =N\frac{\mathrm{}|U_0|}{2\mathrm{}V_{eff}}`$ and thus the BEC is limited to a few 100 atoms. The inclusion of cross-collision necessarily increases the effective mode volume, which in terms of the effective on-site collision rate $`\kappa ^{}`$ is given by $`V_{eff}^{}V_{eff}/(1e^{q_0^2/2x_0^2})`$. For the trapping potential considered, $`V_{eff}8\pi ^{3/2}x_0^3`$, and the TMA is valid whenever
$$N\frac{2\pi ^{1/2}x_0}{|a|(1e^{q_0^2/2x_0^2})}.$$
(4)
For suitable values of the trap parameters, it is thus possible to have very large numbers at the right of the inequality above. Hence the considered TMA, with cross-collision included, can be appropriately employed for the description of the observed phenomena in present experimental setups albiez for numbers of atoms $`N10^3`$.
Typically $`\eta `$, $`\mathrm{\Lambda }\mathrm{\Omega }`$, but cross-collisional effects can only be neglected if $`2\mathrm{\Lambda }(N1)/\mathrm{\Omega }1`$, i. e. if $`(N1)\frac{a_sx_0}{\pi ^{1/2}q_0^2}e^{q_0^2/4x_0^2}1`$. By considering the experimental values from Ref. albiez for $`N=1300`$, $`a_s=5.3`$ nm for <sup>87</sup>Rb atoms, and $`q_0=2.2\mu `$m, we obtain that only for $`x_00.82q_0`$ is that cross-collisional effects can be neglected. This bound value however depends strongly on the number of atoms in the sample and decreases rapidly as $`N`$ is increased. For example, if the number of atoms is increased in one order of magnitude to $`N=13000`$ then the cross-collisional effects can be neglected only if $`x_00.366q_0`$. Furthermore by equating $`\mathrm{\Omega }^{}`$ to the observed frequency of oscillation of 25 Hz in the experiment performed by Albiez et al. albiez and $`\mathrm{\Omega }`$ to the expected frequency of tunneling of 2 Hz we find with the above parameters that only for $`x_06.56q_0`$ is that the tunneling frequency would significantly be increased due to nonlinear off-site interactions. Of course, inside the applied TMA this situation corresponds to that of two strongly overlapped modes, which would compromise the validity of the localization assumption and the considered gaussian eigenfunctions. However the relation between $`x_0`$ and $`q_0`$ is not linear and depends strongly on $`N`$ and on $`q_0`$ as well. For a fixed $`q_0`$ the $`x_0`$ to $`q_0`$ ratio decreases rapidly as $`N`$ is increased. Again, for $`N=13000`$ the substantial increase in the tunneling frequency would be observed for $`x_00.89q_0`$. We remark that the above analysis could be similarly applied by considering the local modes as approximately given by the excited states of the harmonic oscillator. In such a case the localization conditions in relation to the distance between minima of the trapping potential would be significantly relaxed dassarma .
We now turn to the description of the dynamical regimes attained by the Hamiltonian (3). In Fig. 1 we plot the numerically calculated eigenvalues spectra for a condensate with 1000 atoms and fixed $`\mathrm{\Omega }/\kappa =50`$ ratio. Each curve in the plot represents a specific $`\eta /\kappa `$ ratio as depicted in the figure. For $`\eta /\kappa =1/1000`$ it is immediately verified that the eigenvalues correspond to those of the model without cross-collision discussed in Ref. corney1 . The low-excited eigenstates are close to the eigenstates of the $`\widehat{J}_z`$ operator given by $`|j,j_z`$. Highly excited eigenstates are closer to the eigenstates of the operator $`\widehat{J}_x^2`$, given approximately by $`|j,0_x`$. This last state suggests an ideal situation for occurrence of population self-trapping for large $`N`$. The inflexion point for intermediate eigenvalues represents the frontier for the occurrence of self-trapping. As the ratio $`\kappa /\eta `$ is increased the eigenvalues behavior changes dramatically due to the influence of the term proportional to $`\widehat{J}_z^2`$. An outstanding feature is that the inflexion point is shifted to higher eigenvalues for $`\kappa /\eta =1/1001/10`$ and completely disappears for $`\kappa /\eta 1/5`$. For $`\kappa /\eta 1`$ the term proportional to $`\widehat{J}_x^2`$ is negligible in comparison to the other terms and the Hamiltonian eigenstates get as closer to the $`\widehat{J}_z`$ eigenstates as $`\kappa /\eta 1`$.
The equations of motion for the angular momentum operators are given by
$`\dot{\widehat{J}_x}`$ $`=`$ $`\mathrm{}\mathrm{\Omega }^{}\widehat{J}_y4\mathrm{}\eta [\widehat{J}_y,\widehat{J}_z]_+`$ (5)
$`\dot{\widehat{J}_y}`$ $`=`$ $`\mathrm{}\mathrm{\Omega }^{}\widehat{J}_x2\mathrm{}(\kappa 3\eta )[\widehat{J}_z,\widehat{J}_x]_+`$ (6)
$`\dot{\widehat{J}_z}`$ $`=`$ $`2\mathrm{}(\kappa \eta )[\widehat{J}_y,\widehat{J}_x]_+`$ (7)
One can immediately see that if the cross-collision term is strong enough so that $`\kappa \eta \mathrm{\Omega }^{}`$ then
$$\dot{\widehat{J}_z}0$$
(8)
and thus $`\widehat{J}_z`$ being a constant of motion the set of Eqs. (5, 6, and 8) is solved for $`\widehat{J}_x`$ as
$$\widehat{J}_x(t)=\widehat{J}_x(0)\mathrm{cos}\mathrm{\Omega }^\mathrm{"}t+\widehat{J}_y(0)\mathrm{sin}\mathrm{\Omega }^\mathrm{"}t,$$
(9)
where $`\mathrm{\Omega }^\mathrm{"}\mathrm{\Omega }^{}+8\eta J_z=\left[\mathrm{\Omega }+2\mathrm{\Lambda }(N1)\right]+8\eta J_z`$ is the new Rabi frequency, which now depends on both the number of atoms and the initial condition for $`J_z`$ through the nonlinear cross-collision rates $`\mathrm{\Lambda }`$ and $`\eta `$, respectively. Since $`\eta \mathrm{\Omega }`$, $`J_z`$ value only gives a small shift in $`\mathrm{\Omega }^\mathrm{"}`$. The effective Rabi regime of $`\eta \kappa `$ is justified whenever $`\kappa \eta \mathrm{\Omega }^{\prime \prime }`$.
In Fig. 2 the time evolution of the mean value $`\widehat{J}_x`$ as given by the general system of Eqs. (5-7) for distinct $`\eta /\kappa `$ ratio. We suppose an initial state localized in one well as $`|j,j_x`$. Similarly to Ref. corney1 due to intrinsic quantum fluctuations in the initial condition there are some oscillations of the quantum mean decay andersonb . In Fig. 2(a) the revival of the oscillation that occurs at later times is again due to the discrete spectrum of the many-body Hamiltonian bernstein ; chefles . In Fig. 2(b) the self-trapping of the atomic population is shown. This result should be compared to that of Fig. 2 from Ref.albiez . More interesting though are the new features in the quantum dynamics due entirely to the presence of cross-collisions appearing in Fig. 2(c). Notice that even when cross-collisions are relatively small ($`\kappa /\eta 10`$) coherent tunneling dominates, suppressing the initial self-trapping. As the cross-collisions become stronger the collapse and revival dynamics changes. For large intervals of time there is a total collapse of the mean value, which lasts for large periods of time in such a way that this period decreases as $`\eta `$ gets closer to $`\kappa `$ approaching the limit where it tends to zero when the effective Rabi regime takes place and there is no collapse and revival. Notice that the decreasing amplitude modulation in the collapse and revival is due to nonlinearities introduced in Eqs. (5-7) by cross-collisions. As $`\eta \kappa `$ however, the amplitude of coherent oscillation of the population difference goes to a constant value with oscillating frequency $`\mathrm{\Omega }^\mathrm{"}`$.
To push the discussion presented here further, it is interesting to consider the impact of the cross-collisional terms in related many wells systems qtbox . The dynamics is richer and additional effects may occur as consequence of the overlap of neighboring condensed modes. The results found for the double-well case can be extended and set into the more general framework of hopping in many wells systems. It is possible to render a general discussion of the cross-collisional terms for the derivation of a discrete extended Bose-Hubbard model dassarma and of the Wanier-Stark effect buchleitner based solely on ground state trap solutions. For instance, a $`n`$-mode ansatz of the natural extension of the Hamiltonian in Eq.(2) results in a $`n`$-mode extended Bose-Hubbard Hamiltonian
$`\widehat{H}`$ $`=`$ $`{\displaystyle \underset{l}{}}(\kappa _l\eta _{l,l+1})\widehat{n}_l(\widehat{n}_l1)`$
$`+{\displaystyle \underset{l}{}}\eta _{l,l+1}(\widehat{a}_{l+1}^{}\widehat{a}_l+H.c)^2`$
$`+{\displaystyle \underset{l}{}}[\mathrm{\Omega }_{l,l+1}+2\mathrm{\Lambda }_{l,l+1}(2\widehat{n}_l1)](\widehat{a}_{l+1}^{}\widehat{a}_l+H.c),`$
with all parameters related to the those for the double-well system. The dynamical regimes given by (Collision-Dependent Atom Tunneling Rate in Bose-Einstein Condensates) have in part a similar aspect to the ones presented for the double-well potential but can be extended since all the on-site and off-site parameters can be independently varied. Those issues are being further investigated in the context of phase transitions and shall be addressed elsewhere bhm .
In summary we employed the TMA to model an atomic two-mode BEC trapped in a double well potential considering cross-collisions between the condensed modes. Cross-collisions strongly inhibit the self-trapping phenomenon even for small values of the cross-collision rate $`\eta `$. For a given trap geometry the eigenvalues of the many-body Hamiltonian determine a new dynamics resulting in an effective coherent oscillation (effective Rabi regime) of the population in both wells. In the limit $`\eta \kappa `$ for $`\kappa \eta \mathrm{\Omega }^\mathrm{"}`$ an effective Rabi regime takes place with an effective Rabi frequency, $`\mathrm{\Omega }^\mathrm{"}`$, which is explicitly dependent on the total number of atoms. The limit of validity of such a two-mode approximation increases within the assumption of strong cross-collisions in such a way that it may be in accordance with experimental data albiez observed for the increase of tunneling frequency due to nonlinear interaction. As a last result the collapse and revival dynamics is changed resulting in large periods of total collapse. As $`\eta \kappa `$, however the collapse and revival frequency increases attaining the regime of Rabi population oscillation for $`\eta \kappa `$. It is remarkable that similar results to those previously given in exact numerical calculations albiez ; salasnich ; ananikian could be derived inside the TMA just by adding cross-collisional effects. We hope that the above considerations bring some contribution to experimental investigation.
The authors would like to acknowledge partial financial support from FAPESP under project $`\mathrm{\#}04/146052`$, from CNPq and from FAEPEX-UNICAMP. |
warning/0507/quant-ph0507260.html | ar5iv | text | # Loophole-free Bell’s experiments and two-photon all-versus-nothing violations of local realism
## Abstract
We introduce an extended version of a previous all-versus-nothing proof of impossibility of Einstein-Podolsky-Rosen’s local elements of reality for two photons entangled both in polarization and path degrees of freedom \[A. Cabello, Phys. Rev. Lett. 95, 210401 (2005)\], which leads to a Bell’s inequality where the classical bound is 8 and the quantum prediction is 16. A simple estimation of the detection efficiency required to close the detection loophole using this extended version gives $`\eta >0.69`$. This efficiency is lower than that required for previous proposals.
If, from the result of one experiment, we can predict with certainty the result of a spacelike separated experiment, then, following Einstein, Podolsky, and Rosen (EPR) EPR35 , there must be a local element of reality (LER) corresponding to the latter result. However, some predictions of local realistic theories are in conflict with those of quantum mechanics Bell64 ; CHSH69 . Experiments FC72 ; WJSHZ98 ; RKVSIMW01 have shown an excellent agreement with quantum mechanics and have provided solid evidence against LERs. So far, however, the results of these experiments still admit an interpretation in terms of LERs. A conclusive loophole-free experiment would require spacelike separation between the local experiments and a sufficiently large number of the prepared pairs’ detections; otherwise, the possibility of communication at the speed of light between the particles cannot be excluded (locality loophole Bell81 ), and neither can it be excluded that the nondetections correspond to local instructions like “if experiment $`X`$ is performed, then do not activate the detector” (detection loophole Pearle70 ).
Photons are the best candidates for closing the locality loophole. For instance, the Innsbruck experiment WJSHZ98 with polarization-entangled photons separated $`400`$ m is not subject to the locality loophole; however, its detection efficiency ($`\eta =0.05`$) was not high enough to close the detection loophole ($`\eta 0.83`$ is required CH74 ). The detection efficiency for ions is much higher. For instance, in the Boulder experiment RKVSIMW01 with trapped beryllium ions, $`\eta 0.98`$; however, the distance between ions ($`3\mu `$m) was not enough to close the locality loophole.
There are several proposals for experiments for closing both loopholes LS81 ; however, most of them are very difficult to implement with current technology. The most promising approach for a loophole-free experiment is by using entangled photons and more efficient photodetectors KESC94 . Recent experiments with pairs of entangled photons have achieved $`\eta =0.33`$ Kwiat05 . Closing the detection loophole with maximally entangled states and the Clauser-Horne-Shimony-Holt (CHSH) Bell’s inequality CHSH69 requires a detection efficiency $`\eta >2(\sqrt{2}1)0.83`$ CH74 . By using nonmaximally entangled states and supplementary assumptions, $`\eta `$ can be lowered to $`\eta >0.67`$ Eberhard93 . However, these experiments are based on a different interpretation of EPR’s condition for LERs CG97 .
The detection efficiency required for a loophole-free experiment on Bell’s theorem of impossibility of EPR’s LERs is related with the statistical strength of the proof tested in the experiment (i.e., with the amount of evidence against LERs provided by the corresponding experiment). In this respect, all-versus-nothing (AVN) proofs GHZ89 ; Cabello01a provide stronger evidence against LERs than other proofs VGG03 . Specifically, a loophole-free experiment based on the three-observer version Mermin90a of Greenberger, Horne, and Zeilinger’s proof GHZ89 , would require $`\eta >0.75`$ Larsson98 . The negative side is that it requires three spacelike separated regions. The two-photon version CPZBZ03 ; YZZYZZCP05 ; BCDM05 of the two-observer AVN proof Cabello01a only requires a spacelike separation between two regions, but the detection efficiency needed for a loophole-free test is $`\eta >5/60.83`$ Cabello05a .
In this paper we introduce a new AVN proof for two photons, and its corresponding Bell’s inequality, which requires an efficiency $`\eta >0.69`$ to close the detection loophole. This efficiency, although still higher than that achieved in recent experiments, is lower than that required for any previous proposal for a loophole-free experiment based on bipartite Bell’s inequalities and the usual interpretation of EPR’s condition. The new AVN proof is an extended version of a previous one Cabello05a .
Consider two photons entangled both in polarization and in path degrees of freedom CPZBZ03 ; YZZYZZCP05 ; BCDM05 ; Kwiat97 prepared in the state
$`|\psi `$ $`=`$ $`{\displaystyle \frac{1}{2}}(|Hu_1|Hu_2+|Hd_1|Hd_2`$ (1)
$`+|Vu_1|Vu_2|Vd_1|Vd_2),`$
where $`|H_j`$ and $`|V_j`$ represent horizontal and vertical polarization, and $`|u_j`$ and $`|d_j`$ denote two orthonormal path states for photon-$`j`$. Consider also six local observables on photon-$`j`$: three for polarization degrees of freedom, defined by the operators
$`X_j`$ $`=`$ $`|H_jV|+|V_jH|,`$ (2)
$`Y_j`$ $`=`$ $`i\left(|V_jH||H_jV|\right),`$ (3)
$`Z_j`$ $`=`$ $`|H_jH||V_jV|,`$ (4)
and three for path degrees of freedom,
$`x_j`$ $`=`$ $`|u_jd|+|d_ju|,`$ (5)
$`y_j`$ $`=`$ $`i\left(|d_ju||u_jd|\right),`$ (6)
$`z_j`$ $`=`$ $`|u_ju||d_jd|.`$ (7)
Each of these observables can take two values: $`1`$ or $`1`$. Each observer randomly chooses to measure either a polarization observable, a path observable, or a polarization observable and a path observable on his/her photon. The choice of measurement and the measurement itself on photon-1 are assumed to be spacelike separated from those on photon-2.
We will prove that these observables satisfy EPR’s condition for LER, namely, “if, without in any way disturbing a system, we can predict with certainty (i.e., with probability equal to unity) the value of a physical quantity, then there exists an element of physical reality corresponding to this physical quantity” EPR35 . $`Z_1`$ and $`z_1`$ ($`Z_2`$ and $`z_2`$) are EPR’s LERs because their values can be predicted with certainty from spacelike separated measurements of $`Z_2`$ and $`z_2`$ ($`Z_1`$ and $`z_1`$), respectively, because state (1) satisfies the following equations:
$`Z_1Z_2|\psi `$ $`=`$ $`|\psi ,`$ (8)
$`z_1z_2|\psi `$ $`=`$ $`|\psi .`$ (9)
$`X_1`$ and $`x_1`$ ($`X_2`$ and $`x_2`$) are EPR’s LERs because their values can be predicted with certainty from spacelike separated measurements of $`X_2z_2`$ and $`Z_2x_2`$ ($`X_1z_1`$ and $`Z_1x_1`$), respectively, because state (1) satisfies
$`X_1X_2z_2|\psi `$ $`=`$ $`|\psi ,`$ (10)
$`x_1Z_2x_2|\psi `$ $`=`$ $`|\psi ,`$ (11)
$`X_1z_1X_2|\psi `$ $`=`$ $`|\psi ,`$ (12)
$`Z_1x_1x_2|\psi `$ $`=`$ $`|\psi .`$ (13)
Analogously, $`Y_1`$ and $`y_1`$ ($`Y_2`$ and $`y_2`$) are EPR’s LERs because state (1) satisfies
$`Y_1Y_2z_2|\psi `$ $`=`$ $`|\psi ,`$ (14)
$`y_1Z_2y_2|\psi `$ $`=`$ $`|\psi ,`$ (15)
$`Y_1z_1Y_2|\psi `$ $`=`$ $`|\psi ,`$ (16)
$`Z_1y_1y_2|\psi `$ $`=`$ $`|\psi .`$ (17)
We will prove that two compatible observables on the same photon, like $`X_1`$ and $`z_1`$, are independent EPR’s LERs in the sense that the measurement of one of them does not change the value of the other (and therefore there is no need for any additional assumptions beyond EPR’s condition for LER itself; see Cabello03 for a similar discussion). A suitable measurement of $`X_1`$ does not change $`v(x_1)`$ because $`v(x_1)`$ can be predicted with certainty from a spacelike separated measurement of $`Z_2`$ and $`x_2`$, see Eq. (11), and this prediction is not affected by whether $`X_1`$ is measured before $`x_1`$, or $`X_1`$ and $`x_1`$ are jointly measured. Therefore, EPR’s condition is enough to guarantee that $`x_1`$ has a LER \[i.e., a value $`v(x_1)`$\] which does not change with a measurement of $`X_1`$. A similar reasoning applies to any other local observable involved in the proof.
In addition, state (1) satisfies the following equations:
$`X_1x_1Y_2y_2|\psi `$ $`=`$ $`|\psi ,`$ (18)
$`X_1y_1Y_2x_2|\psi `$ $`=`$ $`|\psi ,`$ (19)
$`Y_1x_1X_2y_2|\psi `$ $`=`$ $`|\psi ,`$ (20)
$`Y_1y_1X_2x_2|\psi `$ $`=`$ $`|\psi .`$ (21)
To be consistent with Eqs. (10)–(21), local realistic theories predict the following relations between the values of the LERs:
$`v(X_1)`$ $`=`$ $`v(X_2)v(z_2),`$ (22)
$`v(x_1)`$ $`=`$ $`v(Z_2)v(x_2),`$ (23)
$`v(X_1)v(z_1)`$ $`=`$ $`v(X_2),`$ (24)
$`v(Z_1)v(x_1)`$ $`=`$ $`v(x_2),`$ (25)
$`v(Y_1)`$ $`=`$ $`v(Y_2)v(z_2),`$ (26)
$`v(y_1)`$ $`=`$ $`v(Z_2)v(y_2),`$ (27)
$`v(Y_1)v(z_1)`$ $`=`$ $`v(Y_2),`$ (28)
$`v(Z_1)v(y_1)`$ $`=`$ $`v(y_2),`$ (29)
$`v(X_1)v(x_1)`$ $`=`$ $`v(Y_2)v(y_2),`$ (30)
$`v(X_1)v(y_1)`$ $`=`$ $`v(Y_2)v(x_2),`$ (31)
$`v(Y_1)v(x_1)`$ $`=`$ $`v(X_2)v(y_2),`$ (32)
$`v(Y_1)v(y_1)`$ $`=`$ $`v(X_2)v(x_2).`$ (33)
However, it is impossible to assign the values $`1`$ or $`1`$ to the observables in a way consistent with Eqs. (22)–(33), and therefore the predictions of quantum mechanics cannot be reproduced by EPR’s LERs. Indeed, the assignation is impossible even for each of eight possible subsets of four equations. For instance, the product of Eqs. (22) and (26) \[or the product of Eqs. (24) and (28)\] leads to $`v(X_1)v(Y_1)=v(X_2)v(Y_2)`$; while the product of Eqs. (30) and (32) \[or the product of Eqs. (31) and (33)\] leads to $`v(X_1)v(Y_1)=v(X_2)v(Y_2)`$. Analogously, the product of Eqs. (23) and (27) \[or the product of Eqs. (25) and (29)\] leads to $`v(x_1)v(y_1)=v(x_2)v(y_2)`$; while the product of Eqs. (30) and (31) \[or the product of Eqs. (32) and (33)\] leads to $`v(x_1)v(y_1)=v(x_2)v(y_2)`$. Note that if we explicitly write down the eight sets, the four Eqs. (30)–(33) would appear twice as frequently as the eight Eqs. (22)–(29).
In a real experiment, measurements are imperfect and the observed correlation functions fail to attain the values assumed in the ideal case. Therefore, it is convenient to translate the contradiction of the AVN proof into a Bell’s inequality. This inequality naturally follows from the observation that the relevant features of the AVN proof derive from the fact that state (1) is an eigenstate of the operator
$`\beta `$ $`=`$ $`X_1X_2z_2+x_1Z_2x_2+X_1z_1X_2+Z_1x_1x_2`$ (34)
$`Y_1Y_2z_2y_1Z_2y_2Y_1z_1Y_2Z_1y_1y_2`$
$`+2X_1x_1Y_2y_2+2X_1y_1Y_2x_2+2Y_1x_1X_2y_2`$
$`+2Y_1y_1X_2x_2.`$
As can be easily checked, in any model based on LERs the expected value of $`\beta `$ must satisfy
$$|\beta |8.$$
(35)
However, the quantum prediction for the state (1) is
$$\psi \left|\beta \right|\psi =16,$$
(36)
which is indeed the maximum possible violation of inequality (35). The difference between the maximal violation of the Bell’s inequality and its upper bound is $`168=8`$ for the inequality presented here, while it is just $`2\sqrt{2}20.8`$ for the CHSH inequality CHSH69 , $`42=2`$ for the three-qubit version of Mermin’s inequality Mermin90c , and $`97=2`$ for the Bell’s inequality derived from the two-observer AVN proof Cabello01a .
The simplest way to estimate the detection efficiency required to avoid the detection loophole for a Bell experiment based on this AVN proof, and a good estimation of the required efficiency for a test of the inequality (35), is to see it as a game in the spirit of Vaidman’s game Vaidman99 and Brassard’s “quantum pseudo-telepathy” Brassard03 . Consider a team of two players, Alice and Bob, each of them isolated in a booth. Each of them is asked one out of eight possible questions: (i) “What are $`v(X)`$ and $`v(z)`$?,” (ii) “What are $`v(Z)`$ and $`v(x)`$?,” (iii) “What are $`v(Y)`$ and $`v(z)`$?,” (iv) “What are $`v(Z)`$ and $`v(y)`$?,” (v) “What are $`v(X)`$ and $`V(x)`$?,” (vi) “What are $`v(X)`$ and $`v(y)`$?,” (vii) “What are $`v(Y)`$ and $`v(x)`$?,” and (viii) “What are $`v(Y)`$ and $`v(y)`$?” If one player is asked a question from (i) to (iv), then the other is asked the same question; if one is asked (v), the other is asked (viii); if one is asked (vi), the other is asked (vii). Therefore, the possible scenarios are (i)-(i), meaning that both Alice and Bob are asked (i), (ii)-(ii), (iii)-(iii), (iv)-(iv), (v)-(viii), (vi)-(vii), (vii)-(vi), and (viii)-(v). Each player must give one of the following answers: “$`1`$ and $`1`$,” “$`1`$ and $`1`$,” “$`1`$ and $`1`$,” or “$`1`$ and $`1`$.” Since $`v(X)`$ represents a LER, Alice’s answer to “What is $`v(X)`$?” must be the same regardless of the scenario in which is asked. The same applies for all 12 LERs used in the game. Alice and Bob win if the product of the answers satisfies the corresponding equation in Eqs. (22)–(33). Let us assume that all questions are asked with the same frequency. This is equivalent to assuming that, from the 12 possible scenarios considered in Eqs. (22)–(33), those of Eqs. (30)–(33) occur twice as frequently than those of Eqs. (22)–(29). Assuming this, it is easy to see that an optimal classical strategy allows the players to win this game in $`3/4`$ of the rounds. For instance, a simple optimal classical strategy is that the players always use the following set of local answers:
$`G:=\left\{\begin{array}{cccc}v(X_1)& v(x_1)& v(X_2)& v(x_2)\\ v(Y_1)& v(y_1)& v(Y_2)& v(y_2)\\ v(Z_1)& v(z_1)& v(Z_2)& v(z_2)\end{array}\right\}=\left\{\begin{array}{cccc}1& 1& 1& 1\\ 1& 1& 1& 1\\ 1& 1& 1& 1\end{array}\right\}.`$ (43)
This strategy always wins except for scenarios (iii)-(iii) and (iv)-(iv) \[i.e., it satisfies all Eqs. (22)–(33), except Eqs. (26)–(29)\]. However, the players can win all the rounds if they share pairs of photons in the state (1) and give as answers the results of the corresponding measurements \[i.e., if one is asked question (i), he/she gives as answers the results of measuring $`X`$ and $`z`$ on his/her photon\].
In a real experiment to test the quantum predictions, the low efficiency of detectors opens the possibility that nondetections correspond to local instructions like “if $`X`$ is measured, then the photon will not activate the detector.” This allows us to construct a model with local instructions which simulates the observed data by taking advantage of those rounds in which one photon goes undetected.
Therefore, to estimate the efficiency required for a loophole-free test consider a modified version of the previous game, including the possibility of each player not answering in a fraction $`1\eta `$ of the rounds. If any of the players gives no answers, that round is not taken into account. This new rule opens the possibility of the players also sharing a fraction of sets of local instructions like
$`B_1:=\left\{\begin{array}{cccc}1& 1& 1& 1\\ 0& 0& 1& 1\\ 1& 1& 1& 1\end{array}\right\},`$ (47)
or
$`B_2:=\left\{\begin{array}{cccc}1& 1& 1& 1\\ 1& 1& 0& 0\\ 1& 1& 1& 1\end{array}\right\},`$ (51)
where the $`0`$s in $`B_1`$ means that, if Alice and Bob are using a set $`B_1`$, Alice will not give any answer to questions which include “What is $`v(Y_1)`$?” or “What is $`v(y_1)`$?,” i.e., to questions (iii), (iv), (vi), (vii), and (viii). Analogously, the $`0`$s in $`B_2`$ means that, if Alice and Bob are using a set $`B_2`$, Bob will not answer questions which include “What is $`v(Y_2)`$?” or “What is $`v(y_2)`$?”
Suppose the players are using sets of predefined answers (i.e., suppose the observed data are adequately described by a local realistic theory). For instance, sets like $`G`$ with a frequency $`1p`$, sets like $`B_1`$ with a frequency $`p/2`$, and sets like $`B_2`$ with a frequency $`p/2`$. This $`p`$ is related to the efficiency of the photodetector corresponding to photon-$`j`$, $`\eta _j`$, by the relation
$$\eta _j=1p+\frac{p}{2}f_j+\frac{p}{2},$$
(52)
where $`f_1`$ ($`f_2`$) is the probability that Alice (Bob) answers \[i.e., she (he) does not get the instruction $`0`$ in her (his) set\] when they are using a $`B_1`$ ($`B_2`$) set. In our case, $`f_j=3/8`$.
Let us calculate the minimum detection efficiency required to discard the possibility that nature is using this particular set of predefined answers. To emulate the quantum probability of winning the game ($`P_Q=1`$ in our case), the minimum $`p`$ must satisfy
$$P_Q=(1p)P_G+\frac{p}{2}P_{B_1}+\frac{p}{2}P_{B_2},$$
(53)
where $`P_G`$ is the probability of winning the game when the players use a $`G`$ set, and $`P_{B_j}`$ is the probability of winning when the players use a $`B_j`$ set and both answer the questions. In our example, $`P_G=3/4`$ and $`P_{B_j}=1`$. Introducing the values in Eqs. (52) and (53), we arrive at the conclusion that our model can simulate the quantum predictions if $`\eta _j11/160.69`$. An exhaustive examination of all possible sets like $`G`$ and $`B_j`$ shows that the previously presented model is indeed optimal and, therefore, we conclude that LERs cannot simulate the quantum predictions if
$$\eta _j>11/16.$$
(54)
If we do a similar analysis for a similar game based only on Eqs. (22), (26), (30), and (32), we arrive at the conclusion that closing the detection hole in this case would require $`\eta _j>3/4`$ Cabello05a .
This work was sparked by some experiments made in Hefei YZZYZZCP05 and Rome BCDM05 and a talk given by P.G. Kwiat in Vienna, and was initiated during the Benasque Center for Science Quantum Information Workshop. The author thanks A. Broadbent, E. Galvão, P.G. Kwiat, A. Lamas-Linares, J.-Å. Larsson, C.-Y. Lu, S. Lu, E. Santos, and H. Weinfurter for useful comments, and acknowledges support by Projects Nos. BFM2002-02815 and FQM-239. |
warning/0507/cond-mat0507373.html | ar5iv | text | # Density functional approach for inhomogeneous star polymers
## Abstract
We propose microscopic density functional theory for inhomogeneous star polymers. Our approach is based on fundamental measure theory for hard spheres, and on Wertheim’s first- and second-order perturbation theory for the interparticle connectivity. For simplicity we consider a model in which all the arms are of the same length, but our approach can be easily extended to the case of stars with arms of arbitrary lengths.
It has been demonstrated that density functional theory (DFT) is a versatile and powerful tool to represent the structural and thermodynamic properties of polymeric fluids Likos ; Mayer ; Woodward91 ; Yeti95 ; Yeti ; Yu . Taking into account the level of physical model, DFT’s of polymers can be divided into two main categories. The first category Likos ; Mayer involves the so-called coarse-graining procedure Louis00 , in which the degrees of freedom of monomers building the polymer coils are integrated out. The resulting effective pairwise potential between the centers of masses of two molecules is then used in further investigations Archer02 . An advantage of these models emerges from the possibility of application of theories of simple fluids to describe the polymers.
It is obvious that coarse-grained models lose some information, e.g. a possibility of evaluation of correlation functions between the monomers. From this point of view, models of the second category Yeti ; Yu ; Yu1 ; Bryk04 ; Forsman:02:1 ; Forsman:04:2 , which explicitly treat the microscopic structure of polymers seem to be superior. Several microscopic DFT approaches for inhomogeneous chain polymers have been proposed in the literature. Woodward and Yethiraj Woodward91 ; Yeti95 ; Yeti developed a theory that combines weighted density approximation, known from theories of simple fluids, with single-chain Monte Carlo simulations. An alternative DFT of inhomogeneous polymer solutions was formulated by Forsman, Woodward and FreasierForsman:02:1 . Their theory is based on the free energy functional resulting from generalized Flory equation of state and was extended to the case of inhomogeneous solutions of star polymers Forsman:04:2 . However, a very convenient from numerical point of view approach was developed by Yu and Wu Yu . This approach is based on Rosenfelds’ fundamental measure theory (FMT) Rosenfeld and on Wertheim’s first-order thermodynamic perturbation theory (TPT1). The theory of Yu and Wu allows for performing quite complex studies because it does not require single-chain Monte Carlo simulations and yields a fully analytical equation of state. This approach Yu has been successfully applied to investigate adsorption, surface phase transitions and capillary condensation in systems involving chain particles Bryk ; Bryk1 ; Bryk2 . It was also extended to the case of inhomogeneous semiflexible and cyclic polyatomic fluids Yu1 , as well as to binary hard–rod-polymer mixtures Bryk03 .
A few years ago Blas and Vega Blas proposed an extension of the associating fluid theory for bulk systems involving branched chain molecules. According to their model, branched molecules are built of chain segments (arms) of tangentially bonded hard spheres connected via articulation vertices, each of them formed by $`f`$ arms. The excess Helmholtz free energy due to the chain connectivity is separated into two contributions, one accounting for the formation of the articulation vertex, and a second one due to the formation of the arms. The first term has been described by the second-order thermodynamic perturbation theory (TPT2), whereas the formation of chain arms – via TPT1. The principal aim of this work is to generalize the bulk theory of Blas and Vega to the case of inhomogeneous systems. We consider the simplest case of molecules with one articulation vertex. The generalization is carried out by utilizing the formalism of Yu and Wu Yu , derived for chain polymers.
We consider an inhomogeneous fluid composed of star molecules, i.e. each molecule is built of a spherical “head” (articulation vertex), and $`f`$ arms tangentially attached to it. Each arm is just a chain of $`M_f`$ tangentially jointed segments. Although the numbers $`M_f`$ can be different, in this note we study the case in which all the arms are of the same length, $`MM_f`$, so that the total number of segments within a molecule is $`N=fM+1`$. All the segments are hard spheres of diameter $`\sigma `$. The bonding potential $`V_b(𝐑)`$ is defined so that $`g(𝐑)=\mathrm{exp}[\beta V_b(𝐑)]`$ is
$$g(𝐑)=\underset{i=1}{\overset{f}{}}\frac{\delta (|𝐫_0𝐫_1^{(i)}|\sigma )}{4\pi \sigma ^2}\underset{j=1}{\overset{M1}{}}\frac{\delta (|𝐫_{j+1}^{(i)}𝐫_j^{(i)}|\sigma )}{4\pi \sigma ^2},$$
where $`𝐑=(𝐫_0,\{𝐫_j^{(i)}\})`$ with $`i=1,2,\mathrm{},f`$ and $`j=1,2,\mathrm{},M`$ denotes the set of segment positions. The articulation vertex is labelled by the subscript $`0`$. All remaining segments are labelled by the superscript (specifying arm) and the subscript (specifying the position within a given arm). The grand potential of the system $`\mathrm{\Omega }`$ is as a functional of the local density of polymers, $`\rho (𝐑)`$,
$$\mathrm{\Omega }[\rho (𝐑)]=F_{id}[\rho (𝐑)]+F_{ex}[\rho (𝐑)]+d𝐑\rho (𝐑)(V_{ext}(𝐑)\mu ),$$
(1)
where $`\mu `$ is the configurational chemical potential, $`V_{ext}`$ is the external potential, $`\beta F_{id}[\rho (𝐑)]=\beta d𝐑\rho (𝐑)V_b(𝐑)+d𝐑\rho (𝐑)[\mathrm{ln}(\rho (𝐑))1]`$ is the ideal part of the free energy and $`F_{ex}`$ is the excess free energy. The external potential is a sum of the potentials acting on each segment, $`V_{ext}(𝐑)=v_0(𝐫_0)+_{i=1}^f_{j=1}^Mv_j^{(i)}(𝐫_j^{(i)})`$. We further assume that the excess free energy is a functional of the average segment local density defined as
$`\rho _s(𝐫)`$ $`=`$ $`\rho _0(𝐫)+{\displaystyle \underset{i=1}{\overset{f}{}}}{\displaystyle \underset{j=1}{\overset{M}{}}}\rho _j^{(i)}(𝐫)={\displaystyle d𝐑\delta (𝐫𝐫_0)\rho (𝐑)}`$ (2)
$`+{\displaystyle \underset{i=1}{\overset{f}{}}}{\displaystyle \underset{j=1}{\overset{M}{}}}{\displaystyle d𝐑\delta (𝐫𝐫_j^{(i)})\rho (𝐑)},`$
where $`\rho _j^{(i)}(𝐫)`$ and $`\rho _0(𝐫)`$ are local densities of segment “$`j`$ within the arm $`i`$” and of the articulation segment, respectively.
Following Yu and Wu Yu we decompose the excess free energy as
$$\beta F_{ex}[\rho _s(𝐫)]=d𝐫\left[\mathrm{\Phi }_{HS}(\{n_\alpha (𝐫)\})+\mathrm{\Phi }_C(\{n_\alpha (𝐫)\})\right],$$
(3)
where $`\mathrm{\Phi }_{HS}`$ results from the hard-sphere repulsion between segments, and $`\mathrm{\Phi }_C`$ is the contribution due to the connectivity. Each of these contributions is a function of four scalar and two vector weighted densities Yu ; Rosenfeld . For the hard sphere contribution $`\mathrm{\Phi }_{HS}`$ we use the White-Bear theory, see Refs. Roth ; Yu:02:2 for the explicit formula.
Wertheim’s perturbation theory for a bulk fluid Wertheim can be naturally incorporated into the DFT framework Chapman . The generalization for inhomogeneous star polymer systems is represented by the expression Blas ; Muller
$$\mathrm{\Phi }_C(\{n_\alpha (𝐫)\})=\mathrm{\Phi }^{arm}(\{n_\alpha (𝐫)\})+\mathrm{\Phi }^{art}(\{n_\alpha (𝐫)\}),$$
(4)
where $`\mathrm{\Phi }^{arm}`$ and $`\mathrm{\Phi }^{art}`$ are the contributions due to the formation of chains within consecutive arms and due to the formation of the articulation vertex. The equation for $`\mathrm{\Phi }^{arm}`$ follows from the theory of Yu and Wu Yu
$$\mathrm{\Phi }^{arm}(\{n_\alpha (𝐫)\})=\frac{1+fN}{N}n_0\zeta \mathrm{ln}[y_{HS}(\sigma ,\{n_\alpha (𝐫)\})],$$
(5)
where $`\zeta =1𝐧_{V2}𝐧_{V2}/(n_2)^2`$, and the contact value of the hard-sphere cavity function, $`y_{HS}(\sigma )`$, results from the Carnahan-Starling equation of state, cf. Eq.(18) of Ref. Yu . Free energy density $`\mathrm{\Phi }^{art}`$ is obtained by generalizing the theory of Blas and Vega Blas ; Muller
$$\mathrm{\Phi }^{art}(\{n_\alpha (𝐫)\})=\mathrm{\Phi }_{TPT1}^{art}(\{n_\alpha (𝐫)\})+\mathrm{\Phi }_{TPT2}^{art}(\{n_\alpha (𝐫)\}),$$
(6)
where $`\mathrm{\Phi }_{TPT1}^{art}`$ and $`\mathrm{\Phi }_{TPT2}^{art}`$ represent the first
$$\mathrm{\Phi }_{TPT1}^{art}(\{n_\alpha (𝐫)\})=\frac{f}{N}n_0\zeta \mathrm{ln}[y_{HS}(\sigma ,\{n_\alpha (𝐫)\})],$$
(7)
and the second-order perturbation terms Blas ; Wertheim
$`\mathrm{\Phi }_{TPT2}^{art}`$ $`=`$ $`\mathrm{ln}\sqrt{1+4\mathrm{\Lambda }}`$
$`\mathrm{ln}{\displaystyle \frac{(1+\sqrt{1+4\mathrm{\Lambda }})^{f+1}(1\sqrt{1+4\mathrm{\Lambda }})^{f+1}}{2^{f+1}}}.`$
In the above $`\mathrm{\Lambda }`$ depends on the number of arms and its evaluation requires the knowledge of the $`f`$-body correlation function for $`f`$ spheres in contact. In the case of $`f=3`$ the application of the superposition approximation yields $`\mathrm{\Lambda }=(1+an_3+bn_3^2)/(1n_3)^31`$, where $`a`$ and $`b`$ are constant that depend on the angles between the arms attached to the articulation vertex Blas ; Wertheim ; Muller . In the case of bulk fluids $`n_3`$ is just the packing fraction. Approximation proposed for inhomogeneous system relies on substitution of the bulk packing fraction by the weighted density $`n_3`$. Note that this approximation is not unique. One can follow the ideas of Yu and WuYu:02:2 and propose an approximation involving, besides scalar, also vector weighted densities. In this work, however, we decided to apply as simple expression, as possible.
Within the TPT1 approach the bulk thermodynamic properties of the star polymers are the same as the properties of chains built of the same number of segments. This is because the first-order perturbation free energy takes into account only the number of segment connections and neglects polymer’s topology. The latter is included within the TPT2 approach, cf. Eq.(15) of Ref. Blas . However, the identity of the bulk thermodynamic properties within the TPT1 theory does not imply that the structure of inhomogeneous fluids of star polymers and of chains with the same number of segments, that results from DFT, is identical.
Minimization of (1) yields
$$\rho (𝐑)=g(𝐑)\mathrm{exp}\left[\beta \mu \beta \lambda _0(𝐫_0)\beta \underset{i=1}{\overset{f}{}}\underset{j=1}{\overset{M}{}}\lambda _j^{(i)}(𝐫_j^{(i)})\right],$$
(9)
where $`\lambda _j^{(i)}(𝐫_j^{(i)})=\delta F_{ex}/\delta \rho _s(𝐫_j^{(i)})+v_j^{(i)}(𝐫_j^{(i)})`$ and $`\lambda _0(𝐫_0)=\delta F_{ex}/\delta \rho _s(𝐫_0)+v_0(𝐫_0)`$. For systems with the density distribution varying only in the $`z`$ direction Eqs. (2) and (9) lead to the following expressions for the segment density profiles
$$\rho _0(z_0)=\mathrm{exp}(\beta \mu )\gamma _0(z_0)\left[G^{M+1}(z_0)\right]^f$$
(10)
and
$$\rho _j^{(i)}(z_j^{(i)})=\mathrm{exp}(\beta \mu )\gamma _j^{(i)}(z_j^{(i)})G^{M+1j}(z_j^{(i)})\stackrel{~}{G}^{j+1}(z_j^{(i)}),$$
(11)
where $`\gamma _j^{(i)}(z)=\mathrm{exp}[\beta \lambda _j^{(i)}(z)]`$; $`\gamma _0(z)\gamma _0^{(i)}(z)`$. The functions $`G^i(z)`$ are defined by recurrence relation Yu
$$G^j(z)=dz^{}\gamma _j^{(i)}(z^{})\frac{\theta (\sigma |zz^{}|)}{2\sigma }G^{j1}(z^{}),$$
(12)
for $`j=2,\mathrm{},M`$ with $`G^i(z)1`$. In the above $`\theta `$ is the unit-step function. The functions $`\stackrel{~}{G}^j(z)`$, however, are given by
$$\stackrel{~}{G}^2(z)=dz^{}\gamma _0(z^{})\frac{\theta (\sigma |zz^{}|)}{2\sigma }[G^{M+1}(z^{})]^{f1},$$
(13)
for $`j=2`$ and
$$\stackrel{~}{G}^j(z)=\mathrm{d}z^{}\gamma _j^{(i)}(z^{})]\frac{\theta (\sigma |zz^{}|)}{2\sigma }\stackrel{~}{G}^{j1}(z^{}),$$
(14)
for $`j>2`$. The equations given above are valid for the stars with arms of identical length. In such a case the profiles $`\rho _j^{(i)}(z)`$ are independent of the arm index $`i`$. A generalization of the theory to the case of stars with arms of different length is straightforward. For example, the integrand in the last equation would involve a product of the functions $`G^{M_1+1}(z^{})G^{M_2+1}(z^{})\mathrm{}`$, instead of $`[G^{M+1}(z^{})]^{f1}`$ (here $`M_i`$’s abbreviate the number of segments within consecutive arms). As a simple application of the theory we calculate density profiles of star molecules built of hard-sphere segments near a hard wall, located at $`z=0`$. The solutions of the density profile equations were obtained by using the standard iterational procedure.
In Fig.1 we compare the average segment density profiles resulting from theory with computer simulations simul for star polymers built of $`f=3`$ arms, each composed of $`M=5`$ segments. The calculations were carried out for bulk segment packing fractions $`\eta _{sb}=\pi \rho _{sb}\sigma ^3/6=0.1,0.2`$ and $`0.3`$, where $`\rho _{sb}`$ is the bulk average segment density. The density profiles in Fig. 1 have been normalized by the bulk density $`\rho _{sb}`$. For $`\eta _{sb}=0.2`$ and 0.3 we show two sets of the DFT results. The first one has been evaluated employing the TPT1 approach (i.e. the term given in Eq. Density functional approach for inhomogeneous star polymers has been neglected), whereas the second set was obtained using TPT1 and TPT2 contributions to the free energy. The differences between the local densities resulting from these two approximations are small and occur only within the region adjacent to the wall. The TPT2 contribution leads to smaller contact values of the average segment local density. This effect is quite obvious, because the TPT2 correction lowers the pressure. The agreement between theoretical predictions and computer simulations is reasonable.
Figure 2 compares the density profiles for three-armed star polymers (resulting from the TPT1 approach) with the profiles of chain polymers built of the same number of segments obtained from the approach of Yu and Wu Yu . The results are for bulk average segment densities $`\rho _{sb}^{}=\rho _{sb}\sigma ^3=0.2`$ and $`0.6`$ and for two model systems with different total number of segments $`N=13`$ and $`N=61`$. For $`N=13`$ each star polymer arm is composed of $`M=4`$ segments, whereas for $`N=61`$ each arm consists of $`M=20`$ segments. In the upper panels we compare the average segment densities normalized to unity. We find that for higher bulk density $`\rho _{sb}^{}=0.6`$ (the upper right-hand side panel) the local densities $`\rho _s(z)`$ of chains and stars are quite similar. The profiles are dominated by packing effects. Larger differences between the star and chain polymer profiles are visible at lower density, $`\rho _{sb}^{}=0.2`$, cf. the upper left-hand side panel. Note that the contact values of $`\rho _s(z)`$ for star and chain polymers are identical in our TPT1 approach. Lower panels of Fig. 2 show the density profiles of selected segments for the same systems. We plot here the profiles of “heads” (in the case of chains the profiles of the first segment) and the profiles of the segments attached directly to the “heads”. The differences between the profiles for the chain and star polymers are now more pronounced, especially for the profiles of the “heads”. We have also inspected the profiles for the segments that are topologically more distanced from the head and have found that the difference between them becomes gradually smaller.
Finally Fig. 3 presents the profiles of the stars built of the same number of segments, but having different number of arms. We have considered the models with $`f=3`$, $`M=20`$ and with $`f=4`$, $`M=15`$. It is not surprising that the difference between the average segment density profiles (cf. Fig. 3a) is now less pronounced than in the case of the profiles displayed in Fig. 2, because the differences in the topology of the particles are now smaller. However, the differences between the individual segment density profiles still persist, cf. Fig. 3b, where we show the profiles of “heads” and the segments directly attached to heads.
In conclusion, in this work we propose density functional theory for inhomogeneous star polymers. Although the theory is written down for the case of arms composed of identical numbers of segments, its generalization for stars with arms of different length is straightforward. Several further extensions are also possible. In particular it would be of interest to consider cases of physically different “head” and “arm” segments in order to mimic the systems involving surfactants. Some of these topics are already under study in our laboratory.
This work was supported by EU under a TOK contract No. 509249. We are grateful to prof. O. Pizio for early discussions and bringing Ref. Blas to our attention. |
warning/0507/cond-mat0507062.html | ar5iv | text | # Solitonic Excitations in Linearly Coherent Channels of Bilayer Quantum Hall Stripes
## I Introduction
It is well known that the ground state of the two-dimensional electron gas (2DEG) in single layer quantum Hall systems near half-odd integer filling factors in Landau levels $`N2`$ i.e. for $`\nu =9/2,11/2,\mathrm{}`$ is a striped state responsible for a strong anisotropy in the conductivity tensor of the 2DEG. This state was predicted on the basis of Hartree-Fock calculationsstripetheory and has been extensively studied experimentally.stripeexperimental
When the interlayer distance, $`d`$, in a bilayer quantum Hall system at filling factor $`\nu `$ is large, one expects the system to behave as two isolated two-dimensional electron gases (2DEG) with filling factor $`\nu /2`$. It is then natural to infer that the ground state of the 2DEG in a bilayer should be a striped state at $`\nu =4N+1`$ at sufficiently large interlayer distances. On the other hand, it is known that, at $`\nu =4N+1`$ interlayer interactions can lead to a homogeneous ground state with spontaneous phase coherence between the layers when the interlayer distance is comparable with the separation between electrons in a single layer. One might then conjecture that, as the interlayer separation is decreased, the striped state acquires a certain degree of coherence due to the interlayer interaction. This conjecture was first studied by Brey and Fertigbreystripes who showed that, as $`d`$ is increased from zero the bilayer ground state goes from a uniform coherent state (UCS) at small interlayer separations to a coherent striped phase (CSP) at $`dd_1`$ and then into a modulated striped state (or anisotropic Wigner crystal) at $`dd_2`$. The interlayer coherence is lost in the modulated stripe state. The range $`[d_1,d_2]`$ increases with $`N`$.Dorra
The coherent striped phase shown in Fig. 1 is a state where charge density waves in the two layers are shifted by $`\xi /2`$ where $`\xi `$ is the period of the stripes in one layer. The most interesting aspect of the CSP is that in the regions where the charge densities in both layers “overlap” (in the plane of the two-dimensional electron gas (2DEG)), the electrons are effectively in a linear superposition of states of the form $`|\psi =\left(|R+|L\right)/\sqrt{2}`$ where $`R,L`$ indicates the right and left wells. The interlayer coherence is then maintained but only along linearly coherent regions (LCR’s) whose width decreases as $`d`$ increases. The CSP is most easily represented in the pseudospin language where an up (down) pseudospin is associated with the right (left) well. The CSP is a pseudospin density wave where the pseudospins oscillates in the $`xz`$ plane and the LCR’s are the one-dimensional regions where the pseudospins lie along the $`x`$ direction in the $`xy`$ plane.
In a previous workcotecspmodes , we have computed the collective excitations of the CSP and showed that the low-energy modes of this phase could be described by an effective pseudospin wave hamiltonian. We have also showncoteparallel that the application of a parallel magnetic field gives rise to a very rich phase diagram for the 2DEG involving commensurate-incommensurate transitions with distinctive signatures in the collective excitations and tunneling $`IV`$. A very exhaustive study of the phase diagram of the 2DEG in the presence of a parallel magnetic field, in higher Landau levels, has also been published by Daw-Wei Wang et al.demlerlongarticle ; demlerarticlecourt .
The band structure of the CSP is shown in Fig. 2. In the Hartree-Fock approximation, the energy gap of this system corresponds to the excitation of an electron-hole pair in a coherent channel (a pseudospin flip in the $`xy`$ plane) and is finite if the tunneling parameter $`t0.`$ An estimate of this gap, taking into account some quantum fluctuations, has been done by E. Papa et al.Papamacdo . However, Brey and Fertigbreystripes pointed out that, in analogy with spin (pseudospin) skyrmion excitation in single (double) layer quantum Hall systems at $`\nu =1`$, the lowest-energy charged excitation should be a pseudospin soliton (or antisoliton) in a coherent channel and the gap should be given by the energy required to create a soliton-antisoliton pair. A pseudospin soliton of charge $`q=e`$ corresponds to a $`2\pi `$ rotation of the pseudospin in the $`xy`$ plane. As for skyrmions or bimerons, the size of these solitons is determined by a competition between tunneling energy (which favors small solitons) and interwell exchange energy and Coulomb interaction which favors slowly varying pseudospin textures (large solitons).
In this work, we compute the energy gap of the CSP due to the excitation of a soliton-antisoliton pair as a function of tunneling and interlayer distance. We use a supercell microscopic unrestricted Hartree-Fock approach to extract the energy of a single soliton from that of a crystal of solitons localized in the LCR’s at filling factor $`\nu =4N+1+\mathrm{\Delta }\nu `$. Our calculation shows that a soliton-antisoliton pair has a lower energy than the electron-hole pair so that these topological excitations will be important in determining the transport properties of the CSP. For completeness, we also compute the energy gap of the CSP using a simple field-theoretic model based on the sine-Gordon Hamiltonian where an exact solution for the pseudospin soliton can be obtained. This model does not contain all the terms included in the microscopic approach, but, for slowly varying pseudospin textures, it should give a fair estimate of the energy gap. We actually improve on this model by taking into account that the channels have a width that depends on the interlayer distance $`d`$ and also by taking into account the interaction of the pseudospins in different channels and the Coulomb interaction between different portions of the topological charge densities.
The paper is organized as follows. In Sec. II, we describe the phase diagram of the 2DEG in the bilayer system at filling factors $`\nu =4N+1`$ and $`\nu =4N+1+\mathrm{\Delta }\nu `$ and define the domain of existence of the soliton crystal from which we want to compute the soliton energy. In Sec. III, we introduce the simple field-theoretic model and the exact solution for the pseudospin sine-Gordon solution. Section IV discusses the supercell method that we use to extract the energy of a single soliton from that of a crystal of solitons. The removal of the soliton-soliton energy is discussed in Sec. V. Section VI discusses our numerical results. We conclude in Sec. VII. Details of the derivation of the microscopic expression for the parameters of the field-theoretic model are given in the appendix.
## II Phase diagram of the 2DEG around $`\nu =4N+1`$
In this section, we review the phase diagram of the 2DEG at filling factor $`\nu =4N+1`$ where the coherent striped state is found and at filling factors slightly above $`\nu =4N+1`$ in order to find the range of interlayer distances where a crystal of solitons localized in the LCR’s is stable. We need the energy of this soliton lattice in order to compute the gap energy as we explained in the introduction. To establish the phase diagram, we compute the energy of different electronic phases in the Hartree-Fock approximation in order to find the one that minimizes the total energy at a given value of $`\nu ,d,`$ and $`t`$. The order parameters for the different phases are the expectation values of the density operator projected onto the Landau level $`N`$ of the partially filled Landau level (the guiding center density), i.e.,
$`\rho _N^{i,j}(𝐪)`$ $`=`$ $`{\displaystyle \frac{1}{N_\varphi }}{\displaystyle \underset{X,X^{}}{}}e^{iq_x(X+X^{})/2}\delta _{X,X^{}q_y\mathrm{}^2}`$
$`\times c_{X,i,N}^{}c_{X^{},j,N},`$
where $`i,j`$ are layer indices and $`X,X^{}`$ are guiding center coordinatescotemethodewc . We make the usual approximation of assuming that the filled levels are inert. We also neglect Landau level mixing and assume that the electron gas in the partially filled level is fully spin polarized. In a crystal phase, $`\rho _N^{i,j}(𝐪)`$ is non zero only for $`𝐪=𝐆`$ where $`𝐆`$ is a reciprocal lattice vector of the crystal. Defining the Hartree and Fock interactions
$$H_{i,j}(N,M;𝐪)=\frac{1}{q\mathrm{}}\mathrm{\Lambda }_{i,j}(𝐪)e^{q^2\mathrm{}^2/2}L_N^0\left(\frac{q^2\mathrm{}^2}{2}\right)L_M^0\left(\frac{q^2\mathrm{}^2}{2}\right),$$
(2)
and
$$X_{i,j}(N,M;𝐪)=\frac{\left[\mathrm{min}(M,N)\right]!}{\left[\mathrm{max}(M,N)\right]!}_0^{\mathrm{}}𝑑y\left(\frac{y^2}{2}\right)^{\left|NM\right|}e^{y^2/2}\left[L_{\mathrm{min}(N,M)}^{\left|NM\right|}\left(\frac{y^2}{2}\right)\right]^2\mathrm{\Lambda }_{i,j}\left(\frac{y}{\mathrm{}}\right)J_0\left(q\mathrm{}y\right),$$
(3)
where $`L_N^M\left(x\right)`$ is a generalized Laguerre polynomial, $`J_0\left(x\right)`$ is the zeroth-order Bessel function of the first kind and the form factor
$$\mathrm{\Lambda }_{i,j}=\{\begin{array}{cc}1,\hfill & \text{if }i=j,\hfill \\ e^{qd},\hfill & \text{if }ij,\hfill \end{array}$$
(4)
the Hartree-Fock energy per electron at total filling factor $`\nu =4N+\stackrel{~}{\nu }`$ can be written as
$$\frac{E}{N_e}=\epsilon \left(\frac{e^2}{\kappa \mathrm{}}\right),$$
(5)
with
$`\epsilon `$ $`=`$ $`{\displaystyle \frac{2\stackrel{~}{t}}{\nu }}\mathrm{Re}\left[\rho _N^{R,L}(0)\right]`$
$`+{\displaystyle \frac{1}{2\nu }}{\displaystyle \underset{i,j}{}}{\displaystyle \underset{𝐆0}{}}H_{i,j}(N,N,𝐆)\rho _N^{i,i}\left(𝐆\right)\rho _N^{j,j}\left(𝐆\right)`$
$`{\displaystyle \frac{1}{2\nu }}{\displaystyle \underset{i,j}{}}{\displaystyle \underset{𝐆}{}}X_{i,j}(N,N,𝐆)\rho _N^{i,j}\left(𝐆\right)\rho _N^{j,i}\left(𝐆\right)`$
$`{\displaystyle \frac{2}{\nu }}{\displaystyle \underset{n<N}{}}{\displaystyle \underset{n^{}<N}{}}X_{R,R}(n,n^{},0)`$
$`{\displaystyle \frac{1}{\nu }}{\displaystyle \underset{n<N}{}}X_{i,i}(n,N,0)\stackrel{~}{\nu }.`$
In this last equation, $`N_e`$ is the total number of electrons in the 2DEG, $`\stackrel{~}{t}`$ is the tunneling strength (in units of $`\left(e^2/\kappa \mathrm{}\right)`$, with $`\kappa `$ the dielectric constant of the host material and $`\mathrm{}=\sqrt{\mathrm{}c/eB}`$ the magnetic length).
The last two terms in Eq. (II) give the interaction between electrons in the filled levels and between electrons in the filled levels and electrons in the partially filled level $`N`$. As we will show later, the filled levels contribute to the quasiparticle energies, but not to the charge gap.
The set of $`\rho _N^{i,j}(𝐆)`$’s corresponding to one particular electronic phase is found by solving the equation of motion for the one-particle Green’s function in the Hartree-Fock approximation. The method is described in detail in Ref. cotemethodewc, .
The band structure of the CSP contains two bands $`E_\pm \left(X\right)`$, as shown in Fig. 2. At exactly $`\nu =4N+1`$, the lowest-energy band is completely filled and the system is gapped even in the absence of tunneling. In fact, in the uniform coherent state that occurs for values of $`d`$ for which stripe ordering had not set in, the band structure consists of two straight lines separated by a gap $`\mathrm{\Delta }_{UCS}=\left(2\stackrel{~}{t}+2X_{R,L}(N,N;0)\right)\left(e^2/\kappa \mathrm{}\right)`$ with $`\mathrm{\Delta }_{UCS}2\stackrel{~}{t}`$ as $`d\mathrm{}.`$ In the CSP, the energy bands are periodically modulated in space with the maxima (minima) of the valence (conduction) band at the locations of the LCR’s. At the Hartree-Fock level, the energy gap is the energy needed to excite an electron from a maximum of the valence band to a minimum of the conduction band. This excitation corresponds to a single spin flip localized in one LCR. The decrease in the HF gap in the CSP is due not so much to the reduction of $`X_{R,L}(N,N;0)`$ with $`d`$ as to the increase in intralayer correlations that increases the with of the modulations in $`E_\pm \left(X\right)`$. As $`d`$ increases, the charge modulations get sharper up to the point where the stripes become square waves at very large $`d`$. Correspondingly, the width of the LCR’s decreases with $`d`$ since interwell coherence and charge modulation compete with each other.
In analogy with the excitations of skyrmions in single quantum well and bimerons in bilayer systems at $`\nu =1`$, Brey and Fertigbreystripes noted that a lower-energy excitation could be achieved by exciting a pseudospin soliton in the LCR instead of a simple electron-hole pair. The pseudospin soliton corresponds to a $`2\pi `$ rotation of the pseudospin in one LCR. A slowly varying pseudospin configuration like that in a soliton has lower exchange energy than a single pseudospin flip but the cost in tunneling energy is increased. As for skyrmions or bimerons, an optimal size for the soliton is obtained at given values of $`\nu ,d`$ and $`t`$. The energy cost for this optimal soliton should be compared with the Hartree-Fock electron-hole pair excitation to determine whether or not these topological excitations are energetically favorable.
In a quantum Hall system, the relation between the charge density of the solitons and their pseudospin texture (at $`\stackrel{~}{\nu }=1`$) is given by the Pontryagian densitymacdobible
$$\delta \rho \left(𝐫\right)=\frac{1}{8\pi N_\varphi }\epsilon _{abc}S_a\left(𝐫\right)\epsilon _{ij}_iS_b\left(𝐫\right)_jS_c\left(𝐫\right),$$
(7)
where $`\epsilon _{ij}`$ and $`\epsilon _{abc}`$ are antisymmetric tensors and $`𝐒\left(𝐫\right)`$ is a classical field with unit modulus representing the pseudospins and $`\delta \rho \left(𝐫\right)`$ is the guiding-center density. If we write a general solution as
$`S_x\left(𝐫\right)`$ $`=`$ $`\mathrm{sin}\theta \left(𝐫\right)\mathrm{cos}\phi \left(𝐫\right),`$ (8)
$`S_y\left(𝐫\right)`$ $`=`$ $`\mathrm{sin}\theta \left(𝐫\right)\mathrm{sin}\phi \left(𝐫\right),`$ (9)
$`S_z\left(𝐫\right)`$ $`=`$ $`\mathrm{cos}\theta \left(𝐫\right),`$ (10)
then the induced density takes the simple form
$$\delta \rho \left(𝐫\right)=\frac{1}{4\pi N_\varphi }\mathrm{sin}\theta \left(𝐫\right)\left[\phi \left(𝐫\right)\times \theta \left(𝐫\right)\right]\widehat{𝐳}.$$
(11)
In a LCR, the polar angle of the pseudospins $`\theta =\pi /2`$. If a soliton is present in this LCR, then $`\phi \left(𝐫\right)`$ rotates by $`\pm 2\pi `$ along the channel (oriented in the $`y`$ direction). As discussed below, this is a generalization of a soliton in the sine-Gordon modelrajaraman . We also have that, in the CSP, $`\theta \left(𝐫\right)0`$ in the LCR’s and so the solitons carry a charge by virtue of Eq. (11).
In the case where pseudospin solitons are the lowest-energy excitations of the CSP, we expect that the ground state at $`\nu =4N+\stackrel{~}{\nu }`$ will be a crystal of solitons localized in the LCR’s. Table I shows that the range of interlayer distances where the CSP is the system’s ground state at $`\nu =4N+1`$ increases with the Landau level index. In this work, we choose to study the phase diagram in Landau level $`N=2`$. We show in Fig. 3 the energy per electron for different electronic phases in $`N=2`$ as a function of interlayer distances and for three values of the tunneling parameter $`\stackrel{~}{t}=0,0.01`$ and $`0.06.`$ The filling factor is $`\nu =9.2`$. The contribution from the filled levels is not included in this calculation since it depends only on $`\nu `$ and is thus the same for all phases. At small interlayer distances, where the ground state at $`\nu =9`$ is a UCS, the ground state at $`\nu =9.2`$ is a one-component hexagonal Wigner crystal (HWC). In this phase, a crystal of electrons of pseudospin $`S_x=1/2`$ and filling $`\stackrel{~}{\nu }=0.2`$ sits on top of a liquid of pseudospins $`S_x=+1/2`$ and filling $`9.0`$. There is no pseudospin texture in that state and, in particular, no bimerons in contrast with the situation in the lowest Landau levelbreybimeron where the ground state is a crystal of bimerons. In fact, we find that bimeron excitations are not relevant in $`N=2`$ even in the limit of vanishing $`\stackrel{~}{t}`$.
For interlayer distances where the CSP is found at $`\nu =9`$, the ground state of the 2DEG at $`\nu =9.2`$ is a centered crystal of pseudospin solitons localized in the LCR’s. We note that there are many possible choices for the lattice structure of this crystal, since solitons may or may not be present in every LCR, depending on the commensuration of the lattice of solitons and the underlying stripe state, and it is likely that there are phase transitions among these different states as the filling factor is varied. For the choice of parameters in this study, the lowest energy state has solitons in every channel. We found however that a similar state with solitons in every second channel but with the same filling factor has very nearly the same energy.
Figure 4 shows an example of the charge distribution as well as the pseudospin texture associated with a centered rectangular soliton crystal. Since the focus of this study is on the energetics of single solitons, we will use only the structure illustrated in Fig. 4 for our quantitative analysis below.
At large interlayer distances, we find that the ground state of the 2DEG at $`\nu =9.2`$ is a superposition of two shifted triangular bubble crystalsstripetheory with partial filling $`\stackrel{~}{\nu }=0.6`$ in each well. Because $`\stackrel{~}{\nu }>0.5`$ the bubbles are clusters of holes and not electrons. We find that the number of holes per bubble is $`M=3`$ in agreement with previous Hartree-Fock calculation in single quantum well systemscotebubble .
## III Field-theoretic model
We use two different approaches to compute the energy gap due to the excitation of soliton-antisoliton pairs. The first one is a field-theoretic calculation valid in the limit of slowly varying pseudospin textures. It is explained in this section. The second one is a microscopic approach where the energy of one soliton is computed from that of a crystal of solitons by removing the soliton-soliton interaction. We call this method the supercell approach. In principle, this second method is not restricted to small gradient of the pseudospin texture and includes terms neglected in the field-theoretic model. We expect it to be more accurate than the field-theoretic approach.
In the field-theoretic approach, we evaluate the energy to create a pseudospin soliton by making a long-wavelength expansion of certain terms in the Hartree-Fock Hamiltonian. We follow the procedure developped in details in Ref. macdobible, . To keep the discussion as brief as possible, we give here only the main results of this model. Full details are provided in the appendix.
There are three main contributions to the energy needed to create a pseudospin texture in a LCR. Since in the ground state the in-plane pseudospin component in a LCR is fully polarized along $`S_x`$, adding a pseudospin texture has a tunnel energy cost when $`t0`$ because of the interaction of the texture with the other channels. A second contribution comes from the interlayer exchange interaction which is responsible for the pseudospin stiffness $`\rho _s`$. As we mentioned above, the exchange interaction favors pseudospin textures that vary slowly in space. A third contribution must be considered in our model in order to get agreement with the microscopic approach. It is the Coulomb interaction between different portions of the soliton in a channel. This interaction favors large solitons.
If the coherent channels are oriented along $`y`$ and are considered as effectively one-dimensional, then the energy cost to make a pseudospin texture on top of the ground state where all pseudospins point in the $`x`$ direction in each channel is
$$\delta E=𝑑y\left[\frac{1}{2}\rho _s\left(\frac{\phi (y)}{y}\right)^2T\left[\mathrm{cos}\phi (y)1\right]\right].$$
(12)
where $`\phi (y)`$ in the azimuthal angle of the pseudospins. Eq. (12) is valid if we ignore the third contribution mentionned above. The parameters $`\rho _s`$ and $`T`$ are the effective stiffness and tunneling parameters. These parameters depend on the precise shape of the LCR’s as well as on the interaction between pseudospins of different channels. In the appendix, we derive a microscopic expression for each of these parameters in terms of the order parameters of the CSP. We show that the effective stiffness is given by
$$\rho _s=\frac{1}{16\pi ^2\mathrm{}^2}\left(\frac{e^2}{\kappa \mathrm{}}\right)𝑑q_x|\mathrm{\Omega }(q_x)|^2\frac{d^2X_{R,L}(N,N;𝐪)}{dq_y^2}|_{q_y0},$$
(13)
where
$$\mathrm{\Omega }(q_x)=\xi \underset{G_x}{}\rho _N^x(G_x)\frac{\mathrm{sin}\left[\left(G_xq_x\right)\xi /4\right]}{\left(G_xq_x\right)\xi /4},$$
(14)
is a form factor that takes into account the shape of the channel centered at $`x=0`$. Also, $`\xi `$ is the interstripe distance in the CSP, $`G_x=2\pi n/\xi `$ with $`n=0,\pm 1,\pm 2,\mathrm{}`$ and $`\rho _N^x(G_x)=\mathrm{Re}\left[\rho _N^{R,L}(G_x)\right].`$ If we define the parameter $`\stackrel{~}{G}_x=4\pi n/\xi `$ and
$$J_{}\left(𝐪\right)=X_{R,L}(N,N;𝐪),$$
(15)
then the parameter $`T`$ can be written as
$$T=\frac{1}{2\pi \mathrm{}^2}\left(\frac{e^2}{\kappa \mathrm{}}\right)\left[\stackrel{~}{t}\mathrm{\Omega }(q_x=0)\frac{1}{\xi }\underset{\stackrel{~}{G}_x}{}J_{}(\stackrel{~}{G}_x,0)|\mathrm{\Omega }(G_x)|^2+\frac{1}{2}\frac{1}{L_x}\underset{q_x}{}J_{}(q_x,0)\left|\mathrm{\Omega }\left(q_x\right)\right|^2\right].$$
(16)
The second and third terms in Eq. (16) come from the fact that, because of the pseudospin stiffness, there is an energy cost to rotate the pseudospins in one channel when the pseudospins in the other channels remain fixed in their ground state position. The contribution of these two terms increases the effective tunneling strength $`T`$. Since the energy cost to create a pseudospin soliton is given by $`E_s=8\sqrt{\rho _sT}`$ we see that this second term keeps $`E_s`$ finite even when $`\stackrel{~}{t}=0`$.
In this field-theoretic model, the energy to create an antisoliton is the same as that needed to create a soliton and the charge gap is simply given by
$$\mathrm{\Delta }=16\sqrt{\rho _sT}.$$
(17)
From the energy functional of Eq. (12), we get that the static solution that minimizes the energy must satisfy the sine-Gordon equation
$$\frac{^2\phi (y)}{y^2}=\frac{T}{\rho _s}\mathrm{sin}\phi (y).$$
(18)
The sine-Gordon (or kink) soliton is a solution of this equation. It is given by
$$\phi (y)=4\mathrm{tan}^1\left[e^{\sqrt[]{\frac{T}{\rho _s}}y}\right].$$
(19)
The length of the soliton can be defined as
$$L_s=\sqrt{\frac{\rho _s}{T}}.$$
(20)
With the energy functional of Eq. (12), we find numerically that both $`\rho _s`$ and $`T`$ decrease rapidly with $`d`$ but the size of the soliton $`L_s`$ decreases with increasing $`d`$. This behavior is opposite to what we obtain in the microscopic calculation where the soliton size increases with $`d`$. As we mentionned above, it is necessary to include the Coulomb interaction between different part of the solitons in order to get the soliton size to increase with $`d`$. This leads to the term (full details are given in the appendix)
$$\delta E_{Coul}=\frac{\mathrm{}^2}{32\pi ^2}𝑑y𝑑y^{}\frac{d\phi \left(y\right)}{dy}V_{\text{eff}}\left(yy^{}\right)\frac{d\phi \left(y^{}\right)}{dy^{}}$$
(21)
Inclusion of this term in in the energy functional introduces a nonlocal non-linear term in the differential equation for the soliton and the resulting equation is very difficult to solve. Following S. Ghosh and R. Rajaramanghosh who use a similar procedure in their calculation of the energy of CP<sup>3</sup> skyrmions in bilayers, we make the following approximation. We insert a pseudospin texture $`\phi (y)=4\mathrm{tan}^1\left[e^{y/L_s^{}}\right]`$ into the total energy functional including the Coulomb integral and evaluate is as a function of $`L_s^{}`$. We then minimize the total energy with respect to the length $`L_s^{}`$ to obtain the energy and length of the soliton. In this way, we find a soliton length that increases with $`d`$ as in the microscopic approach. The procedure is described in details in the appendix.
## IV The supercell microscopic Hartree-Fock method
Let $`\epsilon _{CSP}`$ be the energy per electron in the CSP at $`\nu =4N+1`$ and magnetic field $`B_0`$ in units of $`e^2/\kappa \mathrm{}_0`$. If the number of electrons is kept constant and the magnetic field is decreased (to $`B_1`$) or increased (to $`B_2`$) such that the filling factor becomes $`\nu =4N\pm \stackrel{~}{\nu }`$, then a finite density $`n_{qp}=\left|\stackrel{~}{\nu }1\right|/2\pi \mathrm{}_{1,2}^2`$ of quasiparticles (solitons for $`\stackrel{~}{\nu }>1`$ and antisolitons for $`\stackrel{~}{\nu }<1`$) are created in the CSP. At zero temperature, we expect these quasiparticles to crystallize and to be localized in the LCR’s of the CSP. In the limit where only one quasiparticle is created ($`\stackrel{~}{\nu }1`$), we can define the quasiparticle energy as
$$E_{qp}^\pm =\underset{N_{qp}1}{lim}\frac{\nu }{\left|\stackrel{~}{\nu }1\right|}\left[\epsilon _{SC}\left(\frac{e^2}{\kappa \mathrm{}_{1,2}}\right)\epsilon _{CSP}\left(\frac{e^2}{\kappa \mathrm{}_0}\right)\right],$$
(22)
where $`\epsilon _{SC}`$ is the energy per electron in the soliton crystal (SC) in units of $`e^2/\kappa \mathrm{}`$ with $`N_{qp}`$ solitons and $`E_{qp}^+\left(E_{qp}^{}\right)`$ is the energy to create one soliton (antisoliton).
The quasiparticle energy defined in this way, with the number of electrons kept constant, is refered to as the “proper” quasiparticle energy by Morf and HalperinMorfHalperin . Other definitions are also possible. For example, the “gross” quasiparticle energies (or chemical potentials) are defined by
$`\mu ^+`$ $`=`$ $`E\left(N_e=N_\varphi +1\right)E\left(N_e=N_\varphi \right),`$ (23)
$`\mu ^{}`$ $`=`$ $`E\left(N_e=N_\varphi \right)E\left(N_e=N_\varphi 1\right),`$ (24)
where $`N_\varphi `$ is the degeneracy of the Landau levels at a magnetic field $`B_0`$ such that $`\nu =4N+1.`$ The energy $`E\left(N_\varphi \right)`$ is the total energy of the CSP, and $`E\left(N_\varphi \pm 1\right)`$ is the total energy of the CSP with one more (less) particle in the form of a soliton (antisoliton). In this case, the magnetic field is kept constant while the number of particles changes. At zero temperature, this is precisely the definition of the chemical potentials at filling factors slightly above or below $`\nu =4N+1.`$
The different definitions of the the quasiparticle energies lead to different numerical values. As discussed by MacDonald and Girvinmacdoquasip , however, the numerical value of the gap, $`\mathrm{\Delta },`$ is the same for both definitions so that we can write
$$\mathrm{\Delta }=\mu ^+\mu ^{}=E_{qp}^++E_{qp}^{}.$$
(25)
With the formalism described in Sec. II, we can easily compute the Hartree-Fock energy of a crystal of solitons located in the coherent channels of the bilayer. That is, we can compute $`\epsilon _{SC}`$, find $`E_{qp}^\pm `$ and then the energy gap. However, there are several difficulties with this method that we address in this paper. The first one is that the limit $`n_{qp}1`$ cannot be achieved numerically since that would require infinite matrices in the equation of motion for the single-particle Green’s function. In this work, we have succeeded in computing $`\epsilon _{SC}`$ at filling as small as $`\stackrel{~}{\nu }=1\pm 0.02.`$ The second difficulty is that, when a finite density of quasiparticles is present, $`\epsilon _{SC}`$ includes the interaction energy between quasiparticles. This interaction energy must be computed and removed from $`\epsilon _{SC}.`$ A third difficulty is related to the size of the solitons (antisolitons). In Sec. III, we saw that the soliton size becomes very large when the tunneling energy $`\stackrel{~}{t}0`$ or when $`d`$ is large. In this case the size of the soliton is not given by Eq. (20) but is limited by the lattice constant of the soliton crystal. The quasiparticle energy, then, cannot be computed reliably when the tunneling term is too small or the interlayer distance too big.
We now describe in more details our evaluation of $`E_{qp}^\pm .`$ To avoid computing numerically the energy of the antisoliton crystal as well as that of the soliton crystal, we use the particle-hole symmetry of the Hamiltonian around $`\nu =4N+1`$ to relate the energies of the two crystals with the same filling of quasiparticles. We define state $`0`$ as the CSP at $`\nu =4N+1`$, state $`1`$ as the soliton crystal at $`\nu _1=4N+\stackrel{~}{\nu }_1`$ and state $`2`$ as the crystal of antisolitons at $`\nu _2=4N+\stackrel{~}{\nu }_2`$. The filling factors $`\stackrel{~}{\nu }_2=2\stackrel{~}{\nu }_1`$ so that the lattice constants $`a_1`$ and $`a_2`$ of the two crystals are related by $`\mathrm{}_1/a_1=\mathrm{}_2/a_2`$. The Hartree-Fock energy per electron of the three states are given by Eq. (II) which we rewrite here as
$$\frac{E_m}{N_e}=\left[\left(\frac{\stackrel{~}{\nu }_m}{\nu _m}\right)\epsilon _m\left(\stackrel{~}{\nu }_m\right)+\frac{1}{\nu _m}\mathrm{\Lambda }\left(\stackrel{~}{\nu }_m\right)\right]\left(\frac{e^2}{\kappa \mathrm{}_m}\right).$$
(26)
We have defined
$$\epsilon _m\left(\stackrel{~}{\nu }_m\right)=\frac{2\stackrel{~}{t}}{\stackrel{~}{\nu }_m}\mathrm{Re}\left[\rho _N^{R,L}(0)_m\right]$$
(27)
$$+\frac{1}{2\stackrel{~}{\nu }_m}\underset{i,j}{}\underset{𝐆0}{}H_{i,j}(N,N,𝐆)\rho _N^{i,i}\left(𝐆\right)_m\rho _N^{j,j}\left(𝐆\right)_m$$
$$\frac{1}{2\stackrel{~}{\nu }_m}\underset{i,j}{}\underset{𝐆}{}X_{i,j}(N,N,𝐆)\rho _N^{i,j}\left(𝐆\right)_m\rho _N^{j,i}\left(𝐆\right)_m,$$
which is the energy per electron in the partially filled level. The last term in Eq. (26) is the interaction energy with the filled level with
$$\mathrm{\Lambda }\left(\stackrel{~}{\nu }_i\right)=2\mathrm{\Lambda }_1\mathrm{\Lambda }_2\stackrel{~}{\nu }_i,$$
(28)
where
$`\mathrm{\Lambda }_1`$ $`=`$ $`{\displaystyle \underset{n<N}{}}{\displaystyle \underset{n^{}<N}{}}X_{R,R}(n,n^{},0),`$ (29)
$`\mathrm{\Lambda }_2`$ $`=`$ $`{\displaystyle \underset{n<N}{}}X_{i,i}(n,N,0).`$ (30)
From Eqs. (23) and (24), it is easy to see that the cyclotron and Zeeman energies do not contribute to the transport gap $`\mathrm{\Delta }`$ and so can be ignored in Eq. (26). This is also true of the filled levels since their contribution to the quasiparticle energies are given by
$`\left(E_{qp}^+\right)_{f.l.}`$ $`=`$ $`\underset{N_{qp}1}{lim}{\displaystyle \frac{\nu _1}{\left|\stackrel{~}{\nu }_11\right|}}\left({\displaystyle \frac{e^2}{\kappa \mathrm{}_1}}\right){\displaystyle \frac{1}{\nu _1}}\mathrm{\Lambda }\left(\stackrel{~}{\nu }_1\right)`$
$`\underset{N_{qp}1}{lim}{\displaystyle \frac{\nu _1}{\left|\stackrel{~}{\nu }_11\right|}}{\displaystyle \frac{1}{4N+1}}\mathrm{\Lambda }\left(1\right)\left({\displaystyle \frac{e^2}{\kappa \mathrm{}_0}}\right)`$
$`=`$ $`\left({\displaystyle \frac{e^2}{\kappa \mathrm{}_0}}\right)\left[{\displaystyle \frac{1}{2}}\mathrm{\Lambda }_2+3\mathrm{\Lambda }_1\right],`$
and
$`\left(E_{qp}^{}\right)_{f.l.}`$ $`=`$ $`\underset{N_{qp}1}{lim}{\displaystyle \frac{\nu _2}{\left|\stackrel{~}{\nu }_21\right|}}\left({\displaystyle \frac{e^2}{\kappa \mathrm{}_2}}\right){\displaystyle \frac{1}{\nu _2}}\mathrm{\Lambda }\left(\stackrel{~}{\nu }_2\right)`$
$`\left({\displaystyle \frac{e^2}{\kappa \mathrm{}_0}}\right)\underset{N_{qp}1}{lim}{\displaystyle \frac{\nu _2}{\left|\stackrel{~}{\nu }_21\right|}}{\displaystyle \frac{1}{4N+1}}\mathrm{\Lambda }\left(1\right)`$
$`=`$ $`\left({\displaystyle \frac{e^2}{\kappa \mathrm{}_0}}\right)\left[{\displaystyle \frac{1}{2}}\mathrm{\Lambda }_2+3\mathrm{\Lambda }_1\right],`$
so that $`\left(E_{qp}^+\right)_{f.l.}+\left(E_{qp}^{}\right)_{f.l.}=0.`$ In deriving these two equations, we have used
$$\left(\frac{e^2}{\kappa \mathrm{}_1}\right)=\left(\frac{e^2}{\kappa \mathrm{}_0}\right)\sqrt{\frac{\nu _0}{\nu _1}}.$$
(33)
From the electron-hole symmetry, we get
$$\epsilon _2=\left(\frac{\stackrel{~}{\nu }_1}{2\stackrel{~}{\nu }_1}\right)\left[\epsilon _1+\left(\frac{\stackrel{~}{\nu }_11}{\stackrel{~}{\nu }_1}\right)X\left(0\right)\right],$$
(34)
where
$$X\left(0\right)=X_{R,R}(N,N,\mathrm{𝟎}).$$
(35)
Note that Eq. (34) is exact only in the limit where $`N_{qp}1`$ because the inter-well Hartree and Fock interactions contained in $`\epsilon _m`$ depend on the ratio $`d/\mathrm{}`$ and we have $`d/\mathrm{}_1d/\mathrm{}_2`$.
Combining all results, we have for the energy gap
$`\mathrm{\Delta }`$ $`=`$ $`\underset{\mathrm{\Delta }\nu 0}{lim}{\displaystyle \frac{1}{\mathrm{\Delta }\nu }}\stackrel{~}{\nu }_1\left[\sqrt{{\displaystyle \frac{\nu _0}{\nu _1}}}+\sqrt{{\displaystyle \frac{\nu _0}{\nu _2}}}\right]\epsilon _1\left({\displaystyle \frac{e^2}{\kappa \mathrm{}_0}}\right)`$ (37)
$`+\underset{\mathrm{\Delta }\nu 0}{lim}{\displaystyle \frac{1}{\mathrm{\Delta }\nu }}\left[\sqrt{{\displaystyle \frac{\nu _0}{\nu _2}}}X\left(0\right)2\stackrel{~}{\epsilon }_{CSP}\right]\left({\displaystyle \frac{e^2}{\kappa \mathrm{}_0}}\right),`$
where we have defined
$$\epsilon _{CSP}=\frac{1}{4N+1}\stackrel{~}{\epsilon }_{CSP}.$$
(38)
Simplifying, we get finally
$$\mathrm{\Delta }=\underset{\mathrm{\Delta }\nu 0}{lim}\left[2\frac{\stackrel{~}{\nu }_1}{\mathrm{\Delta }\nu }\epsilon _1\frac{2}{\mathrm{\Delta }\nu }\stackrel{~}{\epsilon }_{CSP}+X\left(0\right)\right]\left(\frac{e^2}{\kappa \mathrm{}_0}\right).$$
(39)
We remark that the change in the magnetic length $`\mathrm{}`$ due to the change in the magnetic field makes no contribution to the energy gap. We could have ignored it in Eq. (26). In fact, the gap defined using Eq. (22) and taking $`e^2/\kappa \mathrm{}_i=e^2/\kappa \mathrm{}_0`$ is the so-called neutral energy gapmacdoquasip and it is equal to the other two gaps that we introduced in this section.
Eq. (39) can also be written as
$$\mathrm{\Delta }=2E_{qp}^++\left[2\epsilon _{CSP}+X\left(0\right)\right]\left(\frac{e^2}{\kappa \mathrm{}_0}\right).$$
(40)
In the lowest Landau level, the energy gap at $`\nu =1`$ is due to the excitation of a bimeron-antibimeron pair and the energy per electron in the UCS is $`\epsilon _{UCS}\left(d\right)=\left[\stackrel{~}{t}\frac{1}{4}\left[X\left(0\right)+\stackrel{~}{X}_d\left(0\right)\right]\right]`$ where $`\stackrel{~}{X}\left(0\right)=X_{R,L}(N,N,\mathrm{𝟎})`$. Eq. (40) can then be written, for this special case, as
$$\mathrm{\Delta }=2E_{qp}^++2\left[\epsilon _{UCS}\left(d\right)\epsilon _{UCS}\left(d=0,t=0\right)\right]\left(\frac{e^2}{\kappa \mathrm{}_0}\right),$$
(41)
which is just the form we used in Ref. breybimeron, .
## V Interaction between quasiparticles
With the simplifications introduced in the preceding section, the energy $`\epsilon _{SC}`$ that enters Eq. (22) and Eq. (39) is given by
$$\epsilon _{SC}=\frac{2\stackrel{~}{t}}{\stackrel{~}{\nu }}\mathrm{Re}\left[\rho ^{R,L}(0)\right]$$
(42)
$$+\frac{1}{2\stackrel{~}{\nu }}\underset{i,j}{}\underset{𝐆0}{}H_{i,j}\left(𝐆\right)\rho ^{i,i}\left(𝐆\right)\rho ^{j,j}\left(𝐆\right)$$
$$\frac{1}{2\stackrel{~}{\nu }}\underset{i,j}{}\underset{𝐆}{}X_{i,j}\left(𝐆\right)\rho ^{i,j}\left(𝐆\right)\rho ^{j,i}\left(𝐆\right),$$
where, to simplify the notation, we have left implicit the index $`N`$ of the Landau level. The soliton crystal is a superposition of a CSP with order parameters $`\left\{\alpha ^{i,j}\left(𝐆\right)\right\}`$ (computed at $`\nu =4N+1`$) and a pure soliton crystal (PSC) with order parameters $`\left\{\beta ^{i,j}\left(𝐆\right)\right\}`$ such that
$$\rho ^{i,j}\left(𝐆\right)=\alpha ^{i,j}\left(𝐆\right)+\beta ^{i,j}\left(𝐆\right).$$
(43)
If we insert this decomposition into Eq. (42), we find
$$\epsilon _{SC}=\epsilon _{CSP}\left(\stackrel{~}{\nu }\right)+\epsilon _{CSPPSC}+\epsilon _{PSC},$$
(44)
where
$$\epsilon _{CSP}\left(\stackrel{~}{\nu }\right)=\frac{2\stackrel{~}{t}}{\stackrel{~}{\nu }}\mathrm{Re}\left[\alpha ^{R,L}(0)\right]$$
(45)
$$+\frac{1}{2\stackrel{~}{\nu }}\underset{i,j}{}\underset{𝐆0}{}H_{i,j}\left(𝐆\right)\alpha ^{i,i}\left(𝐆\right)\alpha ^{j,j}\left(𝐆\right)$$
$$\frac{1}{2\stackrel{~}{\nu }}\underset{i,j}{}\underset{𝐆}{}X_{i,j}\left(𝐆\right)\alpha ^{i,j}\left(𝐆\right)\alpha ^{j,i}\left(𝐆\right)$$
is the energy per electron of the CSP (i.e. $`\epsilon _{CSP}\left(\stackrel{~}{\nu }\right)=\frac{1}{\stackrel{~}{\nu }}\stackrel{~}{\epsilon }_{CSP}`$),
$$\epsilon _{PSC}=\frac{2\stackrel{~}{t}}{\stackrel{~}{\nu }}\mathrm{Re}\left[\beta ^{R,L}(0)\right]$$
(46)
$$+\frac{1}{2\stackrel{~}{\nu }}\underset{i,j}{}\underset{𝐆0}{}H_{i,j}\left(𝐆\right)\beta ^{i,i}\left(𝐆\right)\beta ^{j,j}\left(𝐆\right)$$
$$\frac{1}{2\stackrel{~}{\nu }}\underset{i,j}{}\underset{𝐆}{}X_{i,j}\left(𝐆\right)\beta ^{i,j}\left(𝐆\right)\beta ^{j,i}\left(𝐆\right)$$
is the energy per electron of the PSC and
$$\epsilon _{CSPPSC}=$$
(47)
$$+\frac{1}{\stackrel{~}{\nu }}\underset{i,j}{}\underset{𝐆0}{}H_{i,j}\left(𝐆\right)\alpha ^{i,i}\left(𝐆\right)\beta ^{j,j}\left(𝐆\right)$$
$$\frac{1}{\stackrel{~}{\nu }}\underset{i,j}{}\underset{𝐆}{}X_{i,j}\left(𝐆\right)\alpha ^{i,j}\left(𝐆\right)\beta ^{j,i}\left(𝐆\right)$$
is the interaction energy (per electron) between the CSP and the PSC.
The contribution $`\epsilon _{PSC}`$ causes problem because it contains not only the energy to create the $`N_{qp}`$ solitons but also the interaction energy between the solitons. This interaction energy goes away in the limit $`\mathrm{\Delta }\nu 0`$. As we said, however, we cannot go to arbitrarily small $`\mathrm{\Delta }\nu `$ numerically because solving the equation of motion for the single-particle Green’s function then involves diagonalizing very large matrices. We must then find a way to remove the interaction energy in $`\epsilon _{PSC}`$. Two methods can be used. The first one is to replace $`\epsilon _{PSC}`$ by $`\epsilon _{PSC}\epsilon _{int}`$ where $`\epsilon _{int}`$ is the Madelung energy of the crystal of charged quasiparticles, assuming the quasiparticles to be point particlesbreybimeron . We refer to this method as the “Madelung” method. In the limit $`\mathrm{\Delta }\nu 0`$, the quasiparticles are very far apart and, if they have an isotropic charge distribution, it is a reasonable approximation. In the second method, which we refer to as the “form factor” method, we completely replace $`\epsilon _{PSC}\left(\left\{\beta ^{i,j}\left(𝐆\right)\right\}\right)`$ by the energy $`N_{qp}\epsilon _{PSC}\left(\left\{\beta _{qp}^{i,j}\left(𝐪\right)\right\}\right)`$ where $`\epsilon _{PSC}\left(\left\{\beta _{qp}^{i,j}\left(𝐪\right)\right\}\right)`$ is the energy per electron of a “crystal” of only one quasiparticle. In the case of solitons, which are quite extended and highly anisotropic objects it is necessary to use this second approach.
To evaluate $`\epsilon _{PSC}\left(\left\{\beta _{qp}^{i,j}\left(𝐪\right)\right\}\right)`$, we make use of the fact that, when the quasiparticles are very far apart (limit $`\stackrel{~}{\nu }1`$) so that there is no overlap of the density or spin texture due to different quasiparticles, then we may think of the order parameters in real space as given by
$$\beta ^{i,j}\left(𝐫\right)=\underset{𝐑}{}h_{i,j}\left(𝐫𝐑\right),$$
(48)
where $`𝐑`$ is a lattice site. We know that
$$\beta ^{i,j}\left(𝐫\right)=\frac{1}{V}\underset{𝐆}{}\beta ^{i,j}\left(𝐆\right)e^{i𝐆𝐫},$$
(49)
but it is not possible to get $`h_{i,j}\left(𝐫\right)`$ from this equation. We must make an approximation. Since we work in the low-density limit for the quasiparticles, it is a good approximation to assume that for a “crystal” of one quasiparticle
$$\beta ^{i,j}\left(𝐫\right)_{qp}=\{\begin{array}{ccc}\frac{1}{V}_𝐆\beta ^{i,j}\left(𝐆\right)e^{i𝐆𝐫},& \mathrm{if}& 𝐫v_c\\ 0,& \mathrm{if}& 𝐫v_c\end{array},$$
(50)
where $`v_c`$ is the volume of the unit cell centered at $`𝐫=0.`$ Fourier transforming Eq. (50), we have
$`\beta ^{i,j}\left(𝐪\right)_{qp}`$ $`=`$ $`{\displaystyle _V}𝑑𝐫e^{i𝐪𝐫}\beta ^{i,j}\left(𝐫\right)_{qp}`$
$`=`$ $`{\displaystyle \frac{1}{N_{qp}}}{\displaystyle \underset{𝐆}{}}\beta ^{i,j}\left(𝐆\right)\mathrm{\Lambda }\left(𝐪𝐆\right),`$
where the form factor
$$\mathrm{\Lambda }\left(𝐪𝐆\right)=\frac{1}{v_c}_{v_c}𝑑𝐫e^{i𝐪𝐫}e^{i𝐆𝐫},$$
(52)
depends on the shape of the unit cell of the soliton crystal.
It now remains to compute the Hartree-Fock energy corresponding to the density and pseudospin textures given by the $`\beta _{i,j}\left(𝐪\right)_{qp}`$’s. The energy is still given by an equation similar to Eq. (46) where the summation $`\frac{1}{2\stackrel{~}{\nu }}_𝐆`$ is now replaced by $`\frac{1}{2\stackrel{~}{\nu }}_𝐪`$. To go from the sum to the integral, we use
$`{\displaystyle \frac{1}{2\stackrel{~}{\nu }}}{\displaystyle \underset{𝐪}{}}\left(\mathrm{}\right)`$ $``$ $`{\displaystyle \frac{S}{2\stackrel{~}{\nu }}}{\displaystyle \frac{d𝐪}{\left(2\pi \right)^2}\left(\mathrm{}\right)}`$
$``$ $`{\displaystyle \frac{2\pi \stackrel{~}{N}_e}{2\stackrel{~}{\nu }^2}}{\displaystyle \frac{d𝐪\mathrm{}^2}{\left(2\pi \right)^2}\left(\mathrm{}\right)}.`$
Aso, because $`\beta ^{i,j}\left(\mathrm{𝟎}\right)_{qp}1/N_\phi `$, we introduce a new field $`\mathrm{\Theta }^{i,j}\left(𝐪\right)`$ by the definition
$$\mathrm{\Theta }^{i,j}\left(𝐪\right)=N_\phi \beta ^{i,j}\left(𝐪\right)_{qp}.$$
(54)
With this last definition, we have
$`N_{qp}\epsilon _{PSC}\left(\left\{\beta _{qp}^{i,j}\left(𝐪\right)\right\}\right)`$
$`=`$ $`{\displaystyle \frac{\pi \mathrm{\Delta }\nu }{\stackrel{~}{\nu }}}{\displaystyle \underset{i,j}{}}{\displaystyle \frac{d𝐪\mathrm{}^2}{\left(2\pi \right)^2}H_{i,j}\left(𝐪\right)\mathrm{\Theta }^{i,i}\left(𝐪\right)\mathrm{\Theta }^{j,j}\left(𝐪\right)}`$
$`{\displaystyle \frac{\pi \mathrm{\Delta }\nu }{\stackrel{~}{\nu }}}{\displaystyle \underset{i,j}{}}{\displaystyle \frac{d𝐪\mathrm{}^2}{\left(2\pi \right)^2}X_{i,j}\left(𝐪\right)\mathrm{\Theta }^{i,j}\left(𝐪\right)\mathrm{\Theta }^{j,i}\left(𝐪\right)}.`$
As a test of our “form factor” method, we have computed the energy gap due to the creation of bimeron-antibimeron pairs at $`\nu =1`$ in the lowest Landau level $`N=0.`$ Figure 5 shows the energy gap computed from a triangular lattice of bimerons at $`\nu =1.02`$ and $`\stackrel{~}{t}=0.0025.`$ In this case, the Madelung and form factor methods give identical results at small interlayer distances while the Madelung method slighlty overestimates the energy gap at higher distances. The difference between the two approches at large $`d`$ is due to the fact that the charge density profile of the bimeron becomes more and more anisotropic as $`d`$ increases. Also, the Coulomb interaction is stronger between point particles than between extended particles so that the Madelung approach overestimates the gap energy.
To check the convergence of the supercell approach as the lattice constant gets very large, we show in Fig. 6 the energy gap of the UCS at $`\nu =1`$ computed at different values of $`\nu `$ from a crystal of bimerons. The different curves in this figure are for different values of the tunneling strength. The real gap of the system is, of course, defined for $`\nu 1.`$ We see that the gap converges more rapidly to its $`\nu 1`$ value when the tunneling is stronger. This is understandable since the size of a bimeron decreases when $`\stackrel{~}{t}`$ increases and, for sufficiently strong $`\stackrel{~}{t}`$, this size is independent of the lattice constant even at relatively high $`\nu `$. For smaller $`\stackrel{~}{t}`$ the gap converges to its $`\nu 1`$ value, but only at lower filling $`\nu `$. In the application of the supercell technique to the soliton gap in the next section, we will use the form factor method to remove the interaction energy. As we have just shown, this method is more appropriate in the case where the quasiparticle is highly anisotropic in shape.
## VI Numerical results
We now discuss our numerical results for the energy gap of the CSP. Our calculations are done in Landau level $`N=2`$ around $`\nu =9`$ using the form factor method. Figures 7(a)-(c) contain our main results. Differents gaps are plotted as a function of the interlayer distance for tunnelings (a) $`\stackrel{~}{t}=0.007;`$ (b) $`\stackrel{~}{t}=0.01`$; and (c) $`\stackrel{~}{t}=0.02`$. The filled line is $`\mathrm{\Delta }_{UCS},`$ the energy needed to create an ordinary electron-hole pair from the coherent liquid state at $`\nu =9.`$ At $`\nu =9`$, the liquid state is unstable for $`d>d_1`$ where the coherent striped state is the ground state. The Hartree-Fock gap represented by the curve with the filled squares is given by the energy to create an electron-hole pair in a coherent channel (see Fig. 2 where this gap is defined). The other curves give the energy gap calculated in the supercell method for different filling factors $`\nu `$ and the energy gap calculated with the field-theoretic approach explained in the appendix.
From Fig. 7, it is clear that, in the CSP, the energy needed to create a soliton-antisoliton pair is smaller than that needed to create an electron-hole pair for typical experimental values of the tunneling parameter $`\stackrel{~}{t}`$. The transport gap is thus determined by the creation of these topological excitations (as it was the case for skyrmions in quantum Hall ferromagnet at $`\nu =1`$ or with bimerons in bilayer quantum Hall systems).breybimeron Figures 7(a)-(c) show a rapid decrease of the energy gap near the transition between the coherent liquid and the CSP that should be observable experimentally. The curves corresponding to different filling factors show that the convergence of the supercell method is quite good near the liquid-CSP transition but slow at larger values of interlayer distances. This slow convergence is due to the fact that the size of the soliton increases with interlayer distance as shown in Fig. 8 and the shape of the soliton is then restricted by the lattice constant as we explained previously. As $`d/\mathrm{}`$ increases, it becomes necessary to go to lower filling factors to achieve convergence, something we cannot do numerically. In any case, the soliton gap is always lower than the Hartree-Fock gap at higher values of $`d/\mathrm{}`$ since our approach overestimates the energy gap. Increasing $`\stackrel{~}{t}`$ decreases the size of the solitons, however, so that it is possible to achieve better convergence by increasing the value of the tunneling parameter $`\stackrel{~}{t}`$. This is seen by comparing Fig. 7 (a), (b) and (c). Notice also that, for smaller solitons, the soliton gap is closer to the Hartree-Fock result, as expected.
We also show in Fig. 7 the gap calculated with the field-theoretic method (see Eq. (17)). This gap has the same qualitative behavior with interlayer distance, except at small $`d`$ near the phase transition. It is larger than the gap calculated in the microscopic approach. As we explain in the appendix, the field-theoretic result is incorrect at small $`d`$ or large $`\stackrel{~}{t}`$ (fig. 7(c)) where the stripes are not fully developped. At large $`d`$, we cannot say how different the two gaps (macroscopic and field-theoretic) are because the gap found in the microscopic approach has not yet converged at the lowest filling factor we can achieve.
In the field-theoretic method, the soliton size, $`L_s^{},`$ is obtained by the procedure outlined in Sec. III . When the Coulomb interaction between parts of the soliton is properly included, we find numerically that $`L_s^{}`$ increases with $`d`$ as in the supercell calculation. Both approaches give the same trend for the soliton length. The detailed behaviour with $`d/\mathrm{}`$ is quite different, however. Cearly, the field-theoretic calculation does not capture all the subtleties of the We recall that, as the interlayer distance increases, the width of the LCR’s becomes smaller. The behavior of the soliton size may be understood as arising from the Coulomb energy, which favors spreading the charge of the soliton. Our results are plotted in Fig. 8. In this figure, we see that the supercell and field-theoretic results do not match for large $`\stackrel{~}{t}`$. This is again due to the fact that the stripes are not fully formed at large $`\stackrel{~}{t}`$ so that the expression of Eq. (A.3) for the topological charge is not correct. As expected, Fig. 8 shows that the soliton size decreases with $`\stackrel{~}{t}`$.
We have neglected quantum fluctuations in our calculation. These fluctuations increases in importance as $`d/\mathrm{}`$ increases. They renormalize the pseudospin stiffness and will probably also change the size of the solitons and the quantitative values of the energy gaps. Inclusion of these fluctuations is, however, beyond the scope of this paper.
## VII Conclusion
We have computed the energy gap due to the creation of a soliton-antisoliton pair in the linearly coherent channel of the coherent striped phase found in higher Landau levels in a bilayer quantum Hall system. We have computed this gap using a microscopic unrestricted Hartree-Fock approach as well as a field-theoretic approach valid in the limit of slowly varying pseudospin texture. With both methods, we find that the this energy gap is lower in energy than the Hartree-Fock gap due to the creation of an electron-hole pair in a coherent channel (a single spin flip) so that solitonic excitations must play an important role in the transport properties of the coherent striped phase.
## VIII Acknowledgements
This work was supported by a research grant (for R.C.) and graduate research grants (for C. B. D.) both from the Natural Sciences and Engineering Research Council of Canada (NSERC). H.A.F. acknowledges the support of NSF through Grant No. DMR-0454699.
*
## Appendix A Microscopic expressions for the parameters of the field-theoretic model
In this appendix we present the details of the derivation of the microscopic expressions for the parameters $`\rho _s`$ and $`T`$ used in the field-theoretic model of Sec. III. We drop the Landau level index $`N`$ here since all order parameters are to evaluated in the partially filled level $`N`$. We begin by defining the pseudospin density operators
$`\rho (𝐪)`$ $`=\rho ^{R,R}(𝐪)+\rho ^{L,L}(𝐪),`$ (56)
$`\rho _z(𝐪)`$ $`={\displaystyle \frac{1}{2}}\left[\rho ^{R,R}(𝐪)\rho ^{L,L}(𝐪)\right],`$ (57)
$`\rho _x(𝐪)`$ $`={\displaystyle \frac{1}{2}}\left[\rho ^{R,L}(𝐪)+\rho ^{L,R}(𝐪)\right],`$ (58)
$`\rho _y(𝐪)`$ $`={\displaystyle \frac{1}{2i}}\left[\rho ^{R,L}(𝐪)\rho ^{L,R}(𝐪)\right].`$ (59)
The total Hartree-Fock energy of the electrons in the partially filled level for an unbiased bilayer can be written as
$$E_{HF}=\epsilon \left(\frac{e^2}{\kappa \mathrm{}}\right),$$
(60)
where
$`\epsilon `$ $`=`$ $`2N_\varphi \stackrel{~}{t}\rho _x\left(0\right)`$
$`+{\displaystyle \frac{1}{4}}N_\varphi {\displaystyle \underset{𝐪}{}}\mathrm{{\rm Y}}\left(𝐪\right)\rho \left(𝐪\right)\rho \left(𝐪\right)`$
$`+N_\varphi {\displaystyle \underset{𝐪}{}}J_z\left(𝐪\right)\rho _z\left(𝐪\right)\rho _z\left(𝐪\right)`$
$`+N_\varphi {\displaystyle \underset{𝐪}{}}J_{}\left(𝐪\right)[\rho _x(𝐪)\rho _x\left(𝐪\right)`$
$`+\rho _y(𝐪)\rho _y\left(𝐪\right)].`$
We have introduced the interactions
$$J_z\left(𝐪\right)=H_{R,R}\left(𝐪\right)H_{R,L}\left(𝐪\right)X_{R,R}\left(𝐪\right),$$
(62)
$$\mathrm{{\rm Y}}\left(𝐪\right)=H_{R,R}\left(𝐪\right)+H_{R,L}\left(𝐪\right)X_{R,R}\left(𝐪\right),$$
(63)
and
$$J_{}\left(𝐪\right)=X_{R,L}\left(𝐪\right).$$
(64)
In Eq. (A), $`H_{R,R}\left(0\right)=H_{R,L}\left(0\right)=0`$ because of the interaction between the 2DEG and the positive background of the donors.
We now introduce a unitless and unitary pseudospin field $`S_\alpha (𝐫)`$, with $`\alpha =x,y,z`$ related to the guiding center density operators in the pseudospin formalism by the relation
$$S_\alpha (𝐫)=4\pi \mathrm{}^2N_\varphi \rho _\alpha (𝐫),$$
(65)
and a projectednote1 electron density by the relation
$$n(𝐫)=N_\varphi \rho (𝐫).$$
(66)
Using the definition of the pseudospin operators $`S_\alpha (r)`$ and taking the Fourier transform of Eq. (A), we have
$`\epsilon `$ $`=`$ $`{\displaystyle \frac{\stackrel{~}{t}}{2\pi \mathrm{}^2}}{\displaystyle 𝑑𝐫S_x(𝐫)}`$
$`+{\displaystyle \frac{1}{8\pi \mathrm{}^2}}{\displaystyle 𝑑𝐫𝑑𝐫^{}J_{}(𝐫𝐫^{})𝐒_{}(𝐫)𝐒_{}(𝐫^{})}`$
$`+{\displaystyle \frac{1}{8\pi \mathrm{}^2}}{\displaystyle 𝑑𝐫𝑑𝐫^{}S_z(𝐫)J_z(𝐫𝐫^{})S_z(𝐫^{})}`$
$`+{\displaystyle \frac{\pi \mathrm{}^2}{2}}{\displaystyle 𝑑𝐫𝑑𝐫^{}n(𝐫)\mathrm{{\rm Y}}(𝐫𝐫^{})n(𝐫^{})}.`$
Writing $`S_\alpha (𝐫)`$ in spherical coordinates, it is easy to describe the CSP ground state as
$`S_x(𝐫)_{CSP}`$ $`=\mathrm{sin}\theta (x),`$ (68)
$`S_y(𝐫)_{CSP}`$ $`=0,`$ (69)
$`S_z(𝐫)_{CSP}`$ $`=\mathrm{cos}\theta (x),`$ (70)
while the density $`n(𝐫)=cst`$ is uniform. For a state where there is a spin texture only in the channel centered at $`x=0`$ (channel $`0`$) while the other channels remain in their CSP ground state configuration (we recall that $`\xi `$ is the interstripe distance), we write
$$S_x(𝐫)=\{\begin{array}{cc}\mathrm{sin}\theta (x)\mathrm{cos}\phi (y),\hfill & \text{if }|x|\frac{\xi }{4},\hfill \\ \mathrm{sin}\theta (x),\hfill & \text{if }|x|>\frac{\xi }{4},\hfill \end{array}$$
(71)
$$S_y(𝐫)=\{\begin{array}{cc}\mathrm{sin}\theta (x)\mathrm{sin}\phi (y),\hfill & \text{if }|x|\frac{\xi }{4},\hfill \\ 0,\hfill & \text{if }|x|>\frac{\xi }{4},\hfill \end{array}$$
(72)
$$S_z(𝐫)=S_z(𝐫)_{CSP},$$
(73)
$$n(𝐫)=n(𝐫)_{CSP}+\delta n(𝐫).$$
(74)
In these equations, $`\theta (x)`$ is given by its value in the CSP. Defining
$$J_{i,j}(yy^{})_{C_i}𝑑x_{C_j}𝑑x^{}J_{}(𝐫𝐫^{})\mathrm{sin}\theta (x)\mathrm{sin}\theta (x^{}),$$
(75)
where $`C_i`$ corresponds to the $`i`$-th channel of width $`\xi /2`$ centered at $`x_i`$ and $`_{C_i}=_{x_i\xi /4}^{x_i+\xi /4}`$ , it is easy to show that the energy difference between the this last state and the CSP ground state i.e. the energy to create one soliton in a channel is given by
$$\begin{array}{cc}\hfill \delta \epsilon & =\frac{\stackrel{~}{t}}{2\pi \mathrm{}^2}_{C_0}𝑑x\mathrm{sin}\theta (x)𝑑y\left[\mathrm{cos}\phi (y)1\right]\hfill \\ \hfill +& \frac{1}{4\pi \mathrm{}^2}\underset{i0}{}𝑑y𝑑y^{}J_{i,0}(yy^{})\left[\mathrm{cos}\phi (y^{})1\right]\hfill \\ \hfill +& \frac{1}{8\pi \mathrm{}^2}𝑑y𝑑y^{}J_{0,0}(yy^{})\left[\mathrm{cos}(\phi (y)\phi (y^{}))1\right]\hfill \\ & +\frac{\pi \mathrm{}^2}{2}𝑑𝐫𝑑𝐫^{}\delta n(𝐫)\mathrm{{\rm Y}}(𝐫𝐫^{})\delta n(𝐫^{})\hfill \\ & +\pi \mathrm{}^2𝑑𝐫𝑑𝐫^{}\delta n(𝐫)\mathrm{{\rm Y}}(𝐫𝐫^{})n(𝐫^{})_{CSP}.\hfill \end{array}$$
(76)
The first two terms contribute to the effective tunnelling term $`T`$ while the third term is directly related to the pseudospin stiffness of the system. The fourth term takes into account the Coulomb interaction between different parts of the soliton and the last term is the interaction between the charge of the soliton and that of the CSP. In an antisoliton, this fifth contribution would have exactly the same value but with opposite sign so that this last term does not contribute to the transport gap.
### A.1 Calculation of the pseudospin stiffness $`\rho _s`$
To extract the pseudospin stiffness from the third term of Eq. (76), we make a long-wavelength expansion of the $`\mathrm{cos}(\phi (y)\phi (y^{}))1`$ term. This expansion is possible if the pseudospin texture varies slowly in comparison with $`J_{0,0}(y)`$. We get
$`{\displaystyle \frac{1}{8\pi \mathrm{}^2}}{\displaystyle 𝑑y𝑑y^{}J_{0,0}(yy^{})\left[\mathrm{cos}(\phi (y)\phi (y^{}))1\right]}`$
$`={\displaystyle \frac{1}{16\pi \mathrm{}^2}}\left[{\displaystyle 𝑑y^{}y^{}{}_{}{}^{2}J_{0,0}^{}(y^{})}\right]{\displaystyle 𝑑y\left(\frac{d\phi (y)}{dy}\right)^2}.`$ (77)
Comparing this last result with Eq. (12), we see that
$$\rho _s=\frac{1}{8\pi \mathrm{}^2}𝑑yy^2J_{0,0}(y).$$
(78)
The pseudospin stiffness can be written, more explicitely as
$`\rho _s`$ $`=`$ $`{\displaystyle \frac{1}{8\pi \mathrm{}^2}}{\displaystyle 𝑑yy^2\frac{1}{L_xL_y}\underset{𝐪}{}J_{}(𝐪)e^{iq_yy}}`$
$`\times {\displaystyle _{C_0}}dx{\displaystyle _{C_0}}dx^{}\mathrm{sin}\theta (x)\mathrm{sin}\theta (x^{})e^{iq_x(xx^{})},`$
with $`L_x`$ and $`L_y`$ the length and width of the sample. This allows the integrals over $`x`$ and $`x^{}`$ to be totally decoupled. In fact, defining the form factor
$`\mathrm{\Omega }(q_x)`$ $`=`$ $`{\displaystyle _{C_0}}𝑑x\mathrm{sin}\theta (x)e^{iq_xx}`$
$`=`$ $`\xi {\displaystyle \underset{G_x}{}}\rho _x(G_x){\displaystyle \frac{\mathrm{sin}\left[(G_xq_x)\xi /4\right]}{(G_xq_x)\xi /4}},`$
we can write
$$\rho _s=\frac{1}{16\pi ^2\mathrm{}^2}𝑑q_x|\mathrm{\Omega }(q_x)|^2\frac{d^2J_{}(𝐪)}{dq_y^2}|_{q_y0}.$$
(81)
The form factor $`\mathrm{\Omega }(q_x)`$ takes into account the influence of the shape of the charge modulation along the $`x`$ axis in the CSP phase on the effective pseudospin stiffness in the one dimensional sine-Gordon model.
### A.2 Calculation of the tunneling parameter $`T`$
The effective tunnel coupling $`T`$ can be extracted from the first two terms of Eq. (76). The first term renormalizes the tunnel coupling in the 1D effective theory, taking into account that interlayer coherence exists only in the LCR’s. This first term is simply
$$\frac{\stackrel{~}{t}}{2\pi \mathrm{}^2}\mathrm{\Omega }(0)𝑑y\left[\mathrm{cos}\phi (y)1\right].$$
(82)
The second contribution to the effective tunnel coupling comes from the exchange energy between channel $`0`$ (where a pseudospin texture was created) and the other channels. In these other channels, the in-plane pseudospin component is totally polarized along the $`𝐱`$ direction and the exchange interaction between channel $`i`$ and channel $`0`$ favors a configuration in channel $`0`$ where the pseudospin is also polarized along $`+𝐱`$, just like the simple tunnel coupling $`\stackrel{~}{t}`$. In other words, there is an energy cost, even in the absence of tunneling, to make a rotation of the pseudospins in one channel because of the interaction with the pseudospins in the other channels.
It is possible to extract a simple form for this coupling from the second term of Eq. (76) since
$`{\displaystyle \underset{i0}{}}{\displaystyle 𝑑yJ_{i,0}(y)}`$ $`=`$ $`{\displaystyle \frac{1}{L_x}}{\displaystyle \underset{i0}{}}{\displaystyle \underset{q_x}{}}J_{}(q_x,0)|\mathrm{\Omega }(q_x)|^2e^{iq_x(x_ix_0)}`$ (83)
$`=`$ $`{\displaystyle \frac{1}{L_x}}{\displaystyle \underset{i}{}}{\displaystyle \underset{q_x}{}}J_{}(q_x,0)|\mathrm{\Omega }(q_x)|^2e^{iq_x(x_ix_0)}`$
$`{\displaystyle \frac{1}{L_x}}{\displaystyle \underset{q_x}{}}J_{}(q_x,0)|\mathrm{\Omega }(q_x)|^2,`$
with $`x_nx_0=n\xi /2`$ the center-to-center distance between channels $`n`$ and $`0`$. Because there is a sum over the channels, the sum on the wave-vectors $`q_x`$ reduces to a sum over the reciprocal lattice vectors of a 1D lattice of lattice constant $`\xi /2`$, noted $`\stackrel{~}{G}_x`$, and
$$\begin{array}{cc}& \frac{1}{2}\underset{i0}{}𝑑y𝑑y^{}J_{i,0}(yy^{})\left[\mathrm{cos}\phi (y^{})1\right]\hfill \\ & =\frac{1}{\xi }\underset{\stackrel{~}{G}_x}{}J_{}(\stackrel{~}{G}_x,0)|\mathrm{\Omega }(\stackrel{~}{G}_x)|^2\frac{1}{2}\frac{1}{L_x}\underset{q_x}{}J_{}(q_x,0)|\mathrm{\Omega }(q_x)|^2.\hfill \end{array}$$
(84)
Combining the two terms, we find
$`T`$ $`=`$ $`{\displaystyle \frac{1}{2\pi \mathrm{}^2}}[\mathrm{\Omega }(0)\stackrel{~}{t}{\displaystyle \frac{1}{\xi }}{\displaystyle \underset{\stackrel{~}{G}_x}{}}J_{}(\stackrel{~}{G}_x)|\mathrm{\Omega }\left(\stackrel{~}{G}_x\right)|^2`$
$`+{\displaystyle \frac{1}{2}}{\displaystyle \frac{1}{L_x}}{\displaystyle \underset{q_x}{}}J_{}(q_x,0)\left|\mathrm{\Omega }\left(q_x\right)|^2\right].`$
### A.3 Sine-Gordon soliton and the Coulomb energy
If we combine the tunneling and exchange terms, we find that the energy cost to make one soliton localized in a channel of the CSP is given by Eq. (12). As we mentionned in Sec. III, the static solution that minimizes this energy functional is the sine-Gordon (or kink) soliton $`\phi (y)=4\mathrm{tan}^1\left[e^{\sqrt[]{\frac{T}{\rho _s}}y}\right].`$
We now add to Eq. (12) the Coulomb interaction energy between different parts of the soliton
$$\delta E_{Coul}=\frac{\pi \mathrm{}^2}{2}𝑑𝐫𝑑𝐫^{}\delta n\left(𝐫\right)\mathrm{{\rm Y}}\left(𝐫𝐫^{}\right)\delta n\left(𝐫^{}\right).$$
(86)
To relate $`\delta n(𝐫^{})`$ to the angles $`\theta `$ and $`\phi `$, we use the definition of the topological charge density given in Eq. (11). We assume that, in the one-soliton state, only $`\phi \left(y\right)`$ changes along a channel and that $`\theta \left(𝐫\right)`$ is given by its value in the CSP. We have
$`\delta n\left(𝐫\right)`$ $`=`$ $`{\displaystyle \frac{1}{4\pi }}\phi \left(𝐫\right)\times \left(\mathrm{cos}\theta \left(𝐫\right)\right)\widehat{𝐳}`$
$`=`$ $`{\displaystyle \frac{1}{4\pi }}{\displaystyle \frac{d\phi \left(y\right)}{dy}}{\displaystyle \frac{d}{dx}}\mathrm{cos}\theta \left(x\right).`$
At this point, we must remark that if we use the sine-Gordon solution in Eq. (A.3) and integrate the projected density $`\delta n\left(𝐫\right)`$ in a channel, we find $`_{\xi /4}^{+\xi /4}𝑑x_{\mathrm{}}^+\mathrm{}𝑑y\delta n\left(𝐫\right)=1`$ only if $`\mathrm{cos}\theta \left(x\right)`$ varies from $`1`$ to $`+1`$ in the channel i.e. only in the limit or large interlayer distances where the stripes are fully developped. In consequence, we do not expect our field-theoretic model to be valid near the transition between the UCS and the CSP.
We insert Eq. (A.3) into Eq. (86), and define the form factor (for a channel centered at $`x=0`$)
$`A\left(q_x\right)`$ $`=`$ $`{\displaystyle _{C_0}}𝑑xe^{iq_xx}{\displaystyle \frac{d}{dx}}\mathrm{cos}\theta \left(x\right)`$
$`=`$ $`i\xi {\displaystyle \underset{G_x}{}}\rho _z\left(G_x\right)G_x{\displaystyle \frac{\mathrm{sin}\left(q_xG_x\right)\xi /4}{\left(q_xG_x\right)\xi /4}},`$
and the effective interaction $`V_{\text{eff}}\left(yy^{}\right)`$ in a channel
$$V_{\text{eff}}\left(yy^{}\right)=\frac{1}{S}\underset{𝐪}{}\left|A\left(q_x\right)\right|^2\mathrm{{\rm Y}}\left(𝐪\right)e^{iq_y\left(yy^{}\right)}.$$
(89)
We then find for the Coulomb interaction
$$\delta E_{\text{Coul}}=\frac{\mathrm{}^2}{32\pi ^2}𝑑y𝑑y^{}\frac{d\phi \left(y\right)}{dy}V_{\text{eff}}\left(yy^{}\right)\frac{d\phi \left(y^{}\right)}{dy^{}}.$$
(90)
If we add the contribution $`\delta E_{\text{Coul}}`$ to Eq. (12) and minimize the energy with respect to $`\phi \left(y\right)`$, we find that it introduces a nonlocal term to the sine-Gordon equation. The resulting equation is then very difficult to solve. To get an approximation for the Coulomb energy, we decided to proceed in the following way. We take, as a trial solution, the kink soliton
$$\phi (y)=4\mathrm{tan}^1\left[e^{y/L_s^{}}\right],$$
(91)
where $`L_s^{}`$ is the width of the soliton. The Coulomb energy is then
$$\delta E_{\text{Coul}}\left(L_s^{}\right)=\frac{\pi \mathrm{}^2}{32\pi ^2}𝑑𝐪\left|A\left(q_x\right)\right|^2\mathrm{{\rm Y}}\left(q\right)\text{sech}^2\left(\frac{\pi q_yL_s^{}}{2}\right).$$
(92)
The total energy for the soliton is
$$E=4\frac{\rho _s}{L_s^{}}+4TL_s^{}+\delta E_{\text{Coul}}\left(L_s^{}\right).$$
(93)
We find $`L_s^{}`$ by minimizing numerically the total energy $`E`$. In our numerical calculation, we use $`\mathrm{{\rm Y}}\left(𝐪\right)=H_N\left(𝐪\right)`$ instead of Eq. (63). This is also the interaction considered in similar calculationsmacdobible ,rajaraman . The use of Eq. (63) leads to non-physical results. |
warning/0507/math0507132.html | ar5iv | text | # Laplace transformation updated
## 1 Introduction
The mathematical concept which customarily is addressed by the term Laplace Transformation (LT) was primarily designed as a method for the solution of linear differential equations (DEs), i.e., by a kind of operator calculus. Though the theory of LT dates back to Leibniz, Euler, Laplace, Petzval, and many 20th-century authors, its presently prevalent form, for example in , was essentially worked out by Gustav Doetsch . This version of the theory is in the present article addressed as the traditional theory of LT, TLT.
The theory of Laplace transformation is based on the pair of integral transformations
$$L\{f(t)\}=F(s)=_0^{\mathrm{}}f(t)\mathrm{e}^{st}dt;\text{ }s=\sigma +\mathrm{i}\omega ;\text{ }\sigma >0;$$
(1)
$$L^1\{F(s)\}=\phi (t)=\frac{1}{2\pi \mathrm{i}}_{\sigma \mathrm{i}\mathrm{}}^{\sigma +\mathrm{i}\mathrm{}}F(s)\mathrm{e}^{st}ds,$$
(2)
where $`f(t)`$ is presumed to be a real function of the real variable $`t`$. The L-transform $`F(s)`$ is a complex function of the complex variable $`s=\sigma +\mathrm{i}\omega `$. In typical applications, $`t`$ denotes time and $`\omega `$ denotes circular frequency.
In TLT the behavior of the inverse transform is customarily characterized by
$`\phi (t)`$ $`=`$ $`0\text{ for }t<0,`$ (3)
$`=`$ $`f(t)\text{ for }t>0;`$
at $`t=0`$ it is left undefined. Indeed, in the realm of ordinary mathematical functions – which originally was envisaged by TLT – it is impossible to describe the inverse L-transform’s behavior at $`t=0`$, except if $`f(0)=0`$; cf. Sect. 2.
By contrast, the original function $`f(t)`$ is – and has to be – defined at least for $`0t`$, i.e., including $`t=0`$. If $`f(t)`$ were at $`t=0`$ insufficiently defined the integral (1), i.e., the L-transform $`L\{f(t)\}`$, would not exist.
Hence, in TLT the definition interval of the original function $`f(t)`$ does not match that of the pertinent inverse transform $`\phi (t)`$. In the tradition of Doetsch’s theory of LT it is widely believed that this mismatch of definition intervals is irrelevant, provided that application of Laplace transformation is confined to the interval $`t>0`$ – which in TLT is therefore envisaged. However, this assumption is not tenable. For example, the solution $`y(t)`$ of an inhomogeneous linear DE of the type $`f[y^{}(t),y^{\prime \prime }(t),\mathrm{}]=x(t)`$, even when envisaged only for $`t>0`$, in general depends on the behavior of $`x(t)`$ for $`\mathrm{}<t`$ . Thus, when the solution $`y(t)`$ is to be obtained by LT it is crucial which kind of function is by $`L\{x(t)\}`$ actually represented in the interval $`\mathrm{}<t`$. This fact is in Sect. A.1 illustrated by an example.
The most obvious symptoms of TLT’s inconsistency may be listed, as follows.
a) The mismatch of definition intervals appears to disallow concatenation of L-transforms. As the inverse transform $`\phi (t)`$ is at $`t=0`$ undefined it can, rigorously, not be allowed to be subject to another L-transformation. The fact that concatenation turns out actually to be possible does not make TLT formally consistent in this respect. In Sect. 2.4 it is explained why concatenation of L-transforms is possible.
b) The mismatch of definition intervals appears to exclude impulse functions at $`t=0`$ – such as $`\delta (t)`$ – from L-transformation. The delta-impulse $`\delta (t)`$ is only at $`t=0`$ different from null; thus, according to TLT $`\delta (t)`$ can not exist in the inverse L-transform. The fact that the impulse functions actually can be retrieved from their L-transforms does not cure this formal inconsistency of TLT. In Sect. 2.2 it is explained why impulse functions are LT-consistent.
c) TLT’s derivation theorem
$$L\{f^{}(t)\}=sL\{f(t)\}f(0)$$
(4)
is in conflict with the definition interval of TLT, i.e., $`t>0`$. The real constant $`f(0)`$ that in (4) appears in the L-domain represents a $`t`$-domain function of its own, namely, the delta impulse $`f(0)\delta (t)`$ at $`t=0`$. This impulse is outside TLT’s definition interval and therefore should be regarded as irrelevant . Inconsequently, TLT praises the “initial value” $`f(0)`$ as one of its most advantageous features, as that value eventually appears in the solutions of linear DEs and is utilized to account for a system’s initial state at $`t=0`$. By its inconsequent treatment of the initial value TLT implicitly admits that the behavior at $`t=0`$ of functions and of their L-transforms actually is significant.
d) TLT’s derivation theorem (4) is inconsistent with TLT’s integration theorem
$$L\{f(t)\}=\frac{1}{s}L\{f^{}(t)\}.$$
(5)
The two theorems differ by the initial value $`f(0)`$. In TLT, this unexpected and unexplained discrepancy is customarily tolerated. In Sect. 3.4 the relationships between the theorems for derivation and integration are outlined and the origin of the conflict between Eqs. (4) and (5) is revealed.
e) TLT does not in general keep its promise to provide the solution of linear DEs, i.e., for $`t>0`$. There are discrepancies involved which tend to be disguised by formal pseudo-consistency. This notion just restates what was said above about solution of linear DEs. In Sect. A.1 an example is described of this kind of failure.
f) In particular, TLT’s general solution of the linear DE suffers from the so-called initial-value conflict, In TLT it has become customary to work around this conflict by “patching” the original solution. The initial-value conflict is discussed and explained in Sect. 4.4.
These observations indicate that in TLT certain fundamental aspects of LT’s behavior are not consistently accounted for. The mismatch of definition intervals and its consequences need to be resolved rather than circumvented and/or ignored. Laplace transformation needs to become updated.
The present article offers an outline of a new, alternative approach to LT. The problem of LT-consistency of $`t`$-domain functions is re-inspected and resolved. From the results such obtained there emerge new methods for the solution by LT of both the linear inhomogeneous and the linear homogeneous DE.
Crucial results and observations are noted and emphasized as theorems. Several familiar theorems of LT, such as, e.g. the fundamental
superposition theorem
$`L\{c_1f_1(t)+c_2f_2(t)+\mathrm{}\}`$ $`=`$ $`L\{c_1f_1(t)\}+L\{c_2f_2(t)\}+\mathrm{}`$ (6)
$`=`$ $`c_1L\{f_1(t)\}+c_2L\{f_2(t)\}+\mathrm{},`$
are not affected by the new insights. Another group of theorems become restated and re-justified without assuming a new form. A third group includes theorems that become more or less drastically modified as compared to their familiar form. Eventually, there is a fourth group, i.e., of theorems which may be regarded as new – at least in so far as in TLT they do not play a role.
Several complementary explanations and examples are exiled into an appendix. This article is based on, and complements, earlier related work of the present author .
## 2 Getting Laplace transformation consistent
TLT’s heel of Achilles lies at $`t=0`$. For LT to be consistent it is not sufficient that $`\phi (t)=f(t)`$ for $`t>0`$; rather, the condition
$$\phi (0)=f(0)$$
(7)
must also be fulfilled. To suggest the implications of this requirement, the behavior of the inverse L-transform pertinent to a continual function $`f(t)`$ is illustrated in Fig. 1.
From the figure it becomes apparent that for functions which at $`t=0`$ assume a definite unique value $`f(0)0`$ the inverse L-transform $`\phi (t)`$ includes at $`t=0`$ an abrupt transitional section, i.e., from 0 to $`f(0)`$. As a consequence, such type of function can not meet the criterion (7), because a section of $`\phi (t)`$ at $`t=0`$ can not be equal to the definite unique value $`f(0)`$. This kind of non-LT-consistency applies, in particular, to the prominent class of continual derivable functions. By contrast, if $`f(t)`$ is a priori defined as a causal function, i.e., $`f(t)=0`$ for $`t<0`$, the criterion may actually be met, i.e., for certain conditions which will be discussed below.
For brevity and simplicity, the discussion of LT-consistency is in the present article focussed on the dichotomy between causal functions and d-functions. The term d-function denotes the class of bilateral continual functions that are in the ordinary sense derivable as many times as required. It should be kept in mind, in particular, that any solution of a linear homogeneous DE of finite order is a d-function.
To achieve an explicit account of LT’s behavior at $`t=0`$, the definition interval must obviously be expanded from $`t>0`$ into $`t0`$. As the scope of LT’s formula for inverse transformation (2) already encompasses the entire $`t`$-domain, it is the scope of the formula for L-transformation (1) which has to be expanded. To accomplish this kind of expansion it is not required to challenge the definitions of LT; the bilateral scope is already implied in Eq. (1).
### 2.1 Laplace transformation by Fourier transformation
The implications of the latter notion become apparent when one exploits the intimate relationship that exists between Laplace transformation and Fourier transformation. Equation (1) is equivalent to the following “unilateral” Fourier-transformation formula for the function $`f(t)\mathrm{exp}(\sigma t)`$, i.e.,
$$L\{f(t)\}=F(s)=F(\omega ,\sigma )=_0^{\mathrm{}}f(t)\mathrm{e}^{\sigma t}\mathrm{e}^{\mathrm{i}\omega t}dt.$$
(8)
However, there is not really such a thing as unilateral Fourier transformation; Fourier transformation is inherently bilateral. The actual analysis interval (the “scope”) of the transformation (8), and thus of (1), is not determined by the integral’s limits but by the reciprocal of the frequency spacing $`\mathrm{d}f=\mathrm{d}\omega /(2\pi )`$ of the corresponding Fourier-integral representation . The actual analysis interval of both (1) and (8) encompasses the entire $`t`$-domain $`\mathrm{}<t<+\mathrm{}`$, and the low limit $`t=0`$ of the integral (1) corresponds to the center of the analysis interval. Though these implications of Fourier transformation are fairly elementary, apprehension of them appears to be scarce; cf. Sect. A.2.
Hence, the unilaterality of the integration interval in both (8) and (1) is not equivalent to unilaterality of the analysis interval but rather indicates causality of the transformed function. Equation (8) has to be equivalent to ordinary, i.e., bilateral Fourier transformation of a causal, i.e., bilaterally defined function $`f_\mathrm{c}(t)\mathrm{exp}(\sigma t)`$, such that
$$L\{f(t)\}=F(\omega ,\sigma )=_0^{\mathrm{}}f(t)\mathrm{e}^{\sigma t}\mathrm{e}^{\mathrm{i}\omega t}dt=_{\mathrm{}}^+\mathrm{}f_\mathrm{c}(t)\mathrm{e}^{\sigma t}\mathrm{e}^{\mathrm{i}\omega t}dt.$$
(9)
For LT to be consistent with Fourier transformation it is necessary that the causal function $`f_\mathrm{c}(t)`$ be defined in such a way that the two integrals in (9) are fully equivalent. At first sight this requirement is met, e.g., by the definition $`f_\mathrm{c}(t)=0`$ for $`t<0`$; $`f_\mathrm{c}(t)=f(t)`$ for $`t0`$. However, this definition of the causal function is not actually sufficient, because it leaves the transition from $`f_\mathrm{c}(0)`$ to $`f_\mathrm{c}(+0)`$ undefined. There is another condition involved: The second (bilateral) integral in (9) has to be consistent with the pertinent Fourier-integral representation, i.e., the “inverse Fourier transform”
$$\phi (t)=L^1\{L\{f(t)\}\}=\frac{\mathrm{e}^{\sigma t}}{2\pi }_{\mathrm{}}^+\mathrm{}F(\omega ,\sigma )\mathrm{e}^{\mathrm{i}\omega t}d\omega \text{ for }\mathrm{}<t.$$
(10)
As is suggested in (10), this expression is equivalent to (2). The Fourier-integral representation (10) of $`\phi (t)`$ is continuously defined for $`\mathrm{}<t<+\mathrm{}`$. In particular, while $`\phi (t)`$ is causal, the transition from $`\phi (0)`$ to $`\phi (+0)`$ is not undefined but either includes the so-called connect interval (see below, in particular Sect. 2.3), or its derivative(s), i.e., the impulse functions $`\delta ^{(n)}(t)`$ ($`n=0,1,\mathrm{}`$). The Fourier integral’s evident capability to represent impulse functions proves that neither the inverse Fourier transform nor the inverse L-transform are at $`t=0`$ undefined. As a consequence, for $`\phi (0)=f_\mathrm{c}(0)`$ to hold $`f_\mathrm{c}(0)`$ must not be left insufficiently defined. In TLT this requirement is ignored.
Below, the consistent definition of $`f_\mathrm{c}(t)`$ and thus the consistent expansion of LT’s transformation formula (1), (8) into an equivalent bilateral form, is approached by three steps, advancing from impulse functions to the unit step function and finally to the entire class of causal functions.
### 2.2 Impulse functions
The requirements just outlined for $`f_\mathrm{c}(t)`$ are a priori met by the delta impulse and its derivatives. Thus, Eq. (9) applies to $`f_\mathrm{c}(t)=\delta ^{(n)}(t)`$ and one obtains
$$L\{\delta ^{(n)}(t)\}=s^n;\text{ }n=0,1,\mathrm{}$$
(11)
As the impulse functions are only at $`t=0`$ different from null, inverse transformation is by (10) achieved for $`\mathrm{exp}(st)=1`$, i.e., by inverse Fourier transformation of the function $`(\mathrm{i}\omega )^n`$. One obtains
$$L^1\{L\{\delta ^{(n)}(t)\}\}=\delta ^{(n)}(t);\text{ }n=0,1,\mathrm{}$$
(12)
It is thus established by the theory of Fourier transformation that both the L-transform of $`\delta ^{(n)}(t)`$ and its inverse L-transform exist, and that the latter is identical to $`\delta ^{(n)}(t)`$. The impulse functions are LT-consistent.
As $`\delta ^{(n)}(t)`$ is defined for $`\mathrm{}<t<+\mathrm{}`$, while $`\delta ^{(n)}(t)=0`$ for $`t0`$, existence of the L-transform $`L\{\delta ^{(n)}(t)\}`$ is consistent with the condition that for the “unilateral” transformations (1) and (8) to apply $`f(t)`$ must be defined for $`t0`$. Thus, even when one sticks to the “unilateral” form of the transformation there is no conflict. It is only the inverse L-transform of impulse functions that is in conflict with TLT’s assumptions.
### 2.3 The unit step function and the connect function
Also the unit step function is causal and may be regarded as just another member of the class of impulse functions. The unit step function can be defined as the integral of the delta impulse at $`t=0`$, i.e., by
$$u(t):=_{\mathrm{}}^t\delta (\tau )d\tau ;\text{ }\mathrm{}<t<+\mathrm{}.$$
(13)
Equation (13) may be regarded as an implicit definition of $`u(t)`$. When the unit step function is explicitly defined it is important to preserve its definition at $`t=0`$. This can be accomplished in the form
$`u(t)`$ $`=`$ $`0\text{ for }t<0,`$ (14)
$`=`$ $`u_0(t)\text{ for }t=0;\text{ }u_0(t)=\{0\mathrm{}1\},`$
$`=`$ $`1\text{ for }t>0.`$
This definition accounts for the abrupt transition from $`0`$ to $`1`$ that occurs at $`t=0`$, namely, by inclusion of the so-called unit connect function, $`u_0(t)`$. The pseudo-function $`u_0(t)`$ may be conceived of as an infinite set $`\{0\mathrm{}1\}`$ of real numbers (a distribution) that exists at $`t=0`$.
Inclusion of $`u_0(t)`$ in $`u(t)`$ is indispensable not only for LT-consistency of $`u(t)`$ but also for $`u(t)`$ to be consistent with the concept of the delta impulse. LT-consistency of $`u(t)`$ requires that the condition (7) is met, i.e., that $`L^1\{L\{u(t)\}\}=u(t)`$ at $`t=0`$. When the interval extending from $`u(0)=0`$ to $`u(+0)=1`$ is left undefined, fulfilment of the condition (7) remains undecided.
The existence of the delta impulse, which is conceptualized as the first (non-ordinary) derivative of $`u(t)`$, crucially depends on existence of the connect function $`u_0(t)`$. With respect to (14) there holds
$$\delta (t)=u_0^{}(t)=u^{}(t).$$
(15)
Thus, indeed, inclusion of $`u_0(t)`$ in $`u(t)`$ is indispensable. One can not in earnest conceive the delta impulse to be the derivative of a gap.
As the unit step function is the integral of the delta impulse, it follows from the results depicted in Sect. 2.2 that the unit step function as defined by (13) and (14) is indeed LT-consistent; i.e., there holds
$$L^1\{L\{u(t)\}\}=u(t)\text{ for }\mathrm{}<t<+\mathrm{}.$$
(16)
Notably, the L-transform of the unit step function is equivocal. There holds
$$L\{u(t)\}=L\{1\}=1/s.$$
(17)
With respect to the interval $`t0`$ the functions $`f(t)=u(t)`$ and $`f(t)=1`$ differ by the connect function $`u_0(t)`$ which is included in $`f(t)=u(t)`$ but not in $`f(t)=1`$. The connect function does not become explicitly apparent in the L-transform. As a consequence, from the L-transform $`1/s`$ one can not tell whether it was obtained from $`f(t)=u(t)`$ or from $`f(t)=1`$.
From the fact that both $`u(t)`$ and its derivatives (the impulse functions) are LT-consistent one can conclude that $`u_0(t)`$ invariably is contained in the inverse L-transform. According to (16, 17) there holds
$$L^1\{L\{u(t)\}\}=L^1\{L\{1\}\}=u(t),$$
(18)
and $`u(t)`$ contains $`u_0(t)`$. Thus, for a real function $`f(t)`$ to be LT-consistent it is indispensable that it contains the connect function. This is why $`f(t)=u(t)`$ is LT-consistent whereas $`f(t)=1`$ is not.
The somewhat confusing behavior of $`u_0(t)`$ emerges from the fact that $`u_0(t)`$ is a null-function, i.e.,
$$_T^{+T}u_0(t)dt=0;\text{ }T0;$$
(19)
and, therefore,
$$L\{u_0(t)\}=0.$$
(20)
Whereas $`u_0(t)`$ as such does not become apparent in L-transforms, its derivatives do. The derivatives
$$u_0^{(n)}(t)=u^{(n)}(t)=\delta ^{(n1)}(t);\text{ }n=1,2,\mathrm{}$$
(21)
have the L-transforms
$$L\{u_0^{(n)}(t)\}=L\{u^{(n)}(t)\}=L\{\delta ^{(n1)}(t)\}=s^{n1},\text{ }n=1,2,\mathrm{}$$
(22)
and these are different from null.
The occurrence of the connect function in the inverse L-transform was implicitly pointed out, e.g., by Doetsch . He proved that the inverse L-transform is identical to the original function except for a null-function, i.e., a function whose integral is null. Thus, Doetsch in effect anticipated the involvement of the connect function. However, the fact that the null function does not become apparent in L-transforms led him to define inverse L-transforms that differ only by a null-function to be identical. As a consequence, in TLT the connect function is being ignored. This is a serious mistake, i.e., for the following reasons.
a) Without inclusion of $`u_0(t)`$ one can not obtain a solid conceptual definition of LT-consistency, as was pointed out above; cf. Fig. 1.
b) Without inclusion of $`u_0(t)`$ one can not obtain a solid formal definition of LT-consistency, because the unit-step redundancy theorem $`u(t)u^{(n)}(t)=u^{(n)}(t)`$ does not hold; cf. Sects. 2.4, A.3.
c) Without inclusion of $`u_0(t)`$ neither the derivatives of $`u(t)`$, i.e., the impulse functions $`\delta ^{(n)}(t)`$, nor their L-transforms are sufficiently defined.
### 2.4 Causal functions
Utilizing the consistent definition (13, 14) of the unit step function, finally the causal type of function $`f_\mathrm{c}(t)`$ can be defined in the familiar way, i.e.,
$$f_\mathrm{c}(t)=u(t)f(t).$$
(23)
It is inclusion of the unit connect function $`u_0(t)`$ in the unit step function $`u(t)`$ that makes the expression (23) universal. Equation (23) holds irrespective of whether or not $`f(t)`$ itself is causal. $`f(t)`$ may be either a d-function $`f_\mathrm{d}(t)`$; a causal function of the form $`[u(t)f_\mathrm{d}(t)]`$; or an impulse function $`\delta ^{(n)}(t)`$. This follows from the identities outlined in Sect. A.3 which can be subsumed by the
unit-step redundancy theorem
$$u(t)u^{(n)}(t)=u^{(n)}(t);\text{ }n=0,1,\mathrm{}$$
(24)
By this theorem Eq. (23) remains in effect unchanged when both sides are multiplied by $`u(t)`$; in particular, there holds
$$u(t)f_\mathrm{c}(t)=f_\mathrm{c}(t).$$
(25)
Multiplication by $`u(t)`$ of any kind of causal function is redundant.
On the basis of these observations the bilateral formula for L-transformation envisaged in (9) can be restated, and its equivalence to (1) can be expressed by the
bilaterality theorem
$$L\{f(t)\}=_{\mathrm{}}^+\mathrm{}u(t)f(t)\mathrm{e}^{st}dt.$$
(26)
It is the LT-consistent definition (14) of $`u(t)`$ which renders Eq. (26) consistent, i.e., by consistency with Fourier transformation; and Eq. (1) becomes consistent with Fourier transformation by its equivalence to (26). The mathematical implications of (1) are the same as those of (26).
While the above three-step approach to the formula (26) is helpful by its elucidating implications, it should be noticed that formally the expansion of (1) into (26) can be obtained by one single step, namely,
$$L\{f(t)\}=_0^{\mathrm{}}f(t)\mathrm{e}^{st}dt=\underset{T0}{lim}_{\mathrm{}}^+\mathrm{}r(t,T)f(t)\mathrm{e}^{st}dt,$$
(27)
where $`r(t,T)`$ denotes the unit ramp function of which an example is illustrated in Fig. 2.
By letting $`T0`$ the unit step function $`u(t)`$ emerges from the unit ramp function $`r(t,T)`$, and the connect function $`u_0(t)`$ emerges quite naturally from the ascending part $`r_0(t)`$ of the unit ramp function.
From this approach it becomes apparent that the transition from $`u(0)f(t)=0`$ to $`u(+0)f(t)=f(+0)`$ is not empty but is a vertically ascending continuous function, i.e., $`u_0(t)`$. From this notion there emerges another definition of $`u_0(t)`$, namely,
$$u_0(t)=\underset{T0}{lim}r_0(t,T).$$
(28)
The function $`r_0(t)`$ does not necessarily have to be linear – such as in Fig. 2 – but may be represented by any kind of real continuous function that in the interval $`0T`$ rises monotonically from 0 to 1.
Comparison of (1) to (26) reveals a crucial feature of LT which in TLT is ignored, namely, LT’s inherent ambivalence. There holds the
ambivalence theorem
$$L\{f(t)\}=L\{u(t)f(t)\}=F(s).$$
(29)
From an L-transform $`F(s)`$ one can not tell whether it was obtained from $`f(t)`$ or from $`u(t)f(t)`$. This kind of ambivalence corresponds to the ambivalence of $`L\{u(t)\}`$ that was noted by (17). In both cases the ambivalence is significant by the presence of $`u_0(t)`$ in $`u(t)`$ and by the fact that $`L\{u_0^{(n)}(t)\}=s^{n1}0`$ ($`n=1,2,\mathrm{}`$).
As a consequence of the equivalence of (26) to the formula for Fourier-transformation, the inverse L-transform obtained by (2) is invariably and unequivocally identical to $`u(t)f(t)`$. There holds the
causality theorem
$$\phi (t)=L^1\{L\{f(t)\}\}=L^1\{L\{u(t)f(t)\}\}=u(t)f(t)\text{ for }\mathrm{}<t<+\mathrm{}.$$
(30)
The inverse L-transform is the causal companion of the original function $`f(t)`$.
The latter two theorems warrant consistency of concatenation of L-transformations. From (29) and (30) there follows the
concatenation theorem
$$L\{\phi (t)\}=L\{f(t)\}.$$
(31)
In TLT this theorem can not exist because $`\phi (0)`$ is undefined such that, rigorously, $`L\{\phi (t)\}`$ can not be supposed to exist.
As the behavior of the L-transform L{$`f(t)\}`$ is in every respect characterized by the causal inverse L-transform $`\phi (t)`$, in a sense the L-transform of $`f(t)`$ actually constitutes the L-transform of the causal function $`u(t)f(t)`$. In many contexts – e.g., that of the derivation/integration theorem (cf. Sect. 3) – it is indeed quite helpful to observe the
alias theorem:
The notation $`L\{f(t)\}`$ of an L-transform ultimately is an alias for $`L\{u(t)f(t)\}`$.
From the above insights there emerges a concise proof of the
convolution theorem
$$L\{f_1(t)\}L\{f_2(t)\}=L\left\{_0^tf_1(t\tau )f_2(\tau )d\tau \right\}.$$
(32)
The proof of (32) can be based on the identity
$$_{\mathrm{}}^Tg_1(t)dt_{\mathrm{}}^Tg_2(t)dt=u(t)_0^T_0^tg_1(t\tau )g_2(\tau )d\tau dt;\text{ }T>0;\text{ }tT,$$
(33)
which holds for any pair of causal functions $`g_{1,2}(t)`$. Letting $`g_{1,2}(t)=u(t)f_{1,2}(t)\mathrm{exp}(st)`$ and $`T\mathrm{}`$, one obtains from (33)
$$_0^{\mathrm{}}f_1(t)\mathrm{e}^{st}dt_0^{\mathrm{}}f_2(t)\mathrm{e}^{st}dt=_0^{\mathrm{}}\mathrm{e}^{st}_0^tf_1(t\tau )f_2(\tau )d\tau dt,$$
(34)
where the “factor” $`u(t)`$ is appropriately accounted for. By (1) Eq. (34), indeed, is equivalent to (32).
### 2.5 Testing LT-consistency
For a function $`f(t)`$ to be LT-consistent it is with regard to (30) crucial that
$$u(0)f(0)=f(0).$$
(35)
In this equation neither $`u(0)`$ nor $`f(0)`$ necessarily denote unique definite values. The symbol $`u(0)`$ is just a synonym for the pseudo-function $`u_0(t)`$. As $`f(t)`$ may be any linear combination of d-functions, causal functions, and derivatives, $`f(0)`$ may actually denote any linear combination of definite unique values, connect functions, and derivatives of the latter, i.e., impulse functions. Keeping this in mind, and observing the unit-step redundancy theorem (24), the criterion for LT-consistency of any type of real function $`f(t)`$ can be universally expressed by the
LT-consistency theorem: The function $`f(t)`$ is LT-consistent if, and only if,
$$u(t)f(t)=f(t)\text{ for }\mathrm{}<t<+\mathrm{}.$$
(36)
The criterion (36) provides for the formal proof of the conclusion that was already drawn from Fig. 1, namely, that d-functions $`f_\mathrm{d}(t)`$ are non-LT-consistent: $`u(t)f_\mathrm{d}(t)f_\mathrm{d}(t)`$. By contrast, causal functions $`f_\mathrm{c}(t)`$ are LT-consistent, because by the unit-step redundancy theorem there holds $`u(t)f_\mathrm{c}(t)=f_\mathrm{c}(t)`$.
A useful application of the LT-consistency theorem is verification of LT-consistency of shifted functions. From the formula for inverse L-transformation (2) one obtains in the familiar way the relationship
$$\phi (t\tau )=L^1\{L\{f(t)\}\mathrm{e}^{s\tau }\},$$
(37)
which indicates that multiplication of $`L\{f(t)\}`$ by $`\mathrm{exp}(s\tau )`$ shifts the causal inverse transform by any positive or negative amount $`\tau `$. Utilizing (30) one obtains from (37) by L-transformation the
shifting theorem
$$L\{f(t)\}\mathrm{e}^{s\tau }=L\{u(t\tau )f(t\tau )\};\text{ }\mathrm{}<\tau <+\mathrm{}.$$
(38)
This expression holds for any type of real function $`f(t)`$ and for any $`\tau `$. However, it is only for $`\tau 0`$ that (38) is consistent with the alias theorem, indicating that (38) is only for $`\tau 0`$ LT-consistent. Indeed, for a shifted causal function $`f_\mathrm{c}(t\tau )`$ the criterion (36) reads
$$u(t)f_\mathrm{c}(t\tau )=f_\mathrm{c}(t\tau ),$$
(39)
and this condition is met only for $`\tau 0`$.
The shifting theorem makes particularly apparent that Laplace transformation virtually is confined to causal functions. TLT’s attempt to express the shifting theorem in terms of ordinary derivable functions is awkward and incoherent .
Yet, Laplace transformation of non-causal functions, in particular, d-functions, does not entirely have to be ruled out. L-transformation of d-functions can be sensible and useful; cf. Sects. 3.2, 3.4, 4.2. One just has to keep in mind that the L-transform of a d-function, $`L\{f_\mathrm{d}(t)\}`$, actually is the L-transform of $`u(t)f_\mathrm{d}(t)`$, i.e., $`L\{u(t)f_\mathrm{d}(t)\}`$.
### 2.6 Summary
Laplace transformation of a real function $`f(t)`$ is said to be consistent if the pertinent inverse L-transform is identical to $`f(t)`$. For $`t>0`$ LT is in this sense consistent for any type of real function that can be L-transformed at all. However, such confined consistency is not sufficient for LT to provide a coherent system of operator calculus. Even in the realm of TLT the behavior of functions at $`t=0`$ turns out in effect to be involved. Explicit inclusion of the point $`t=0`$ into LT’s definition interval requires
a) appreciation of the implicit bilaterality of the L-transformation formula (1); and
b) mathematical description of the behavior at $`t=0`$ of both $`f(t)`$ and $`\phi (t)`$, i.e., by the connect function $`u_0(t)`$ and/or its derivatives.
From a) there emerges a bilateral equivalent of the L-transformation formula (1) that makes LT compatible with the theory of Fourier-transformation. This formula in turn is dependent on consistent definition of the unit step function, i.e., according to b). L-transformation turns out to be ambivalent, i.e., there holds $`L\{f(t)\}=L\{u(t)f(t)\}`$. Inverse L-transformation is unequivocal; the inverse transform has the form $`u(t)f(t)`$. Therefore, only causal functions are LT-consistent.
## 3 The derivation/integration theorem
There are two fundamentally different approaches to obtaining LT-theorems such as those for derivation/integration:
a) Determination by (1) of the L-domain operation that corresponds to derivation/integration of the original function $`f(t)`$.
b) Determination by (2) of the L-domain operation that corresponds to derivation/integration of the inverse L-transform $`\phi (t)`$.
L-transformation by (1) or (26) is ambivalent, whereas inverse L-transformation by (2) is unequivocal. Moreover, any kind of application of LT, in particular, to the solution of linear DEs, is in the first place dependent on the inverse L-transform $`\phi (t)`$ – as opposed to the original function $`f(t)`$. Therefore, only the approach b) is adequate.
TLT’s derivation theorem (4) is based on the inadequate approach a). In TLT, the derivative $`f^{}(t)`$ of $`f(t)`$ is presupposed to exist in the ordinary mathematical sense and the L-transform of $`f^{}(t)`$ is expressed by (1). The theorem for the first derivative then emerges from integration by parts:
$`L\{f^{}(t)\}`$ $`=`$ $`{\displaystyle _0^{\mathrm{}}}f^{}(t)\mathrm{e}^{st}dt`$ (40)
$`=`$ $`\left[f(t)\mathrm{e}^{st}\right]_0^{\mathrm{}}+s{\displaystyle _0^{\mathrm{}}}f(t)\mathrm{e}^{st}dt`$
$`=`$ $`sL\{f(t)\}f(0).`$
Although this kind of mathematical reasoning is formally correct, the result is not LT-consistent. The theorem holds only for d-functions $`f(t)=f_\mathrm{d}(t)`$, i.e., functions that at $`t=0`$ are derivable in the ordinary sense; d-functions are non-LT-consistent because $`u(t)f_\mathrm{d}(t)f_\mathrm{d}(t)`$; cf. (36).
Below, the LT-consistent theorems for derivation and integration are obtained by the approach b). These theorems are termed the primary derivation and integration theorem, respectively. They hold for derivatives/integrals of the form $`[u(t)f(t)]^{(\pm n)}`$. There also exist secondary theorems; these hold for $`f^{(\pm n)}(t)`$. TLT’s derivation theorem turns out to be of the secondary type. (In TLT the distinction between $`f^{(\pm n)}(t)`$ and $`[u(t)f(t)]^{(\pm n)}`$ is ignored.) The primary theorems for derivation/integration have the same form such that they can be unified. The unified theorem can be generalized for non-integer order of derivation/integration. The secondary theorems turn out to be virtually irrelevant.
### 3.1 The primary derivation theorem
As the inverse transform $`\phi (t)`$ is a causal and non-ordinary function, its derivatives are non-ordinary and causal, as well. By contrast, the formula (2) can within the integral be derived for $`t`$ in the ordinary sense, and the order of derivation is unlimited, as only the function $`\mathrm{exp}(st)`$ needs to be derived. For the first derivative of $`\phi (t)`$ one obtains
$$\phi ^{}(t)=L^1\{sF(s)\}=L^1\{sL\{f(t)\}\}=L^1\{sL\{u(t)f(t)\}\}.$$
(41)
Observing the concatenation theorem (31), one obtains for the second derivative
$$\phi ^{\prime \prime }(t)=L^1\{sL\{\phi ^{}(t)\}\}=L^1\{s^2L\{\phi (t)\}\}=L^1\{s^2L\{f(t)\}\},$$
(42)
Continuing with this kind of reasoning to obtain higher-order derivatives, and observing (30), one obtains the
primary derivation theorem
$$L\{[u(t)f(t)]^{(n)}\}=s^nL\{f(t)\};\text{ }n=0,1,\mathrm{}$$
(43)
Equation (43) holds for any type of real function $`f(t)`$ and for any $`n`$ for which the inverse transform of $`s^nL\{f(t)\}`$ exists.
For causal functions $`f_\mathrm{c}(t)`$ the primary derivation theorem assumes the form
$$L\{f_\mathrm{c}^{(n)}(t)\}=s^nL\{f_\mathrm{c}(t)\};\text{ }n=0,1,\mathrm{}$$
(44)
The form (44) applies, in particular, to the impulse functions $`f_\mathrm{c}(t)=\delta ^{(n)}(t)`$, and therefore Eqs. (11, 22) are consistent with (44). In contrast to a widespread misconception, the reason why the form (44) holds for impulse functions is not because these are distributions but because they are causal.
The $`t`$-domain implications of the operation $`s^nL\{f(t)\}`$ are quite different for causal functions versus d-functions. As for causal functions $`f_\mathrm{c}(t)`$ there holds $`\phi (t)=f_\mathrm{c}(t)`$, there follows from the above deduction of (43)
$$L^1\{s^nL\{f_\mathrm{c}(t)\}\}=f_\mathrm{c}^{(n)}(t).$$
(45)
For d-functions the corresponding relationship is considerably more complicated. To demonstrate this, functions of the type
$$f_{\mathrm{ud}}(t)=u(t)f_\mathrm{d}(t)$$
(46)
are taken into consideration, where $`f_\mathrm{d}(t)`$ denotes a d-function. The $`n`$-th derivative of the so-called ud-function $`f_{\mathrm{ud}}(t)`$ is depicted by the
dud-theorem
$$f_{\mathrm{ud}}^{(n)}(t)=[u(t)f_\mathrm{d}(t)]^{(n)}=u(t)f_\mathrm{d}^{(n)}(t)+\underset{\nu =0}{\overset{n1}{}}f_\mathrm{d}^{(n1\nu )}(0)\delta ^{(\nu )}(t);\text{ }n=1,2,\mathrm{}$$
(47)
The dud-theorem (derivation of ud-function) expresses the $`n`$-th non-ordinary derivative of $`u(t)f_\mathrm{d}(t)`$ by the ordinary derivatives of $`f_\mathrm{d}(t)`$, i.e., at the expense of getting impulse functions involved as depicted by (47). The dud-theorem is explained in Sect. A.4.
By (47), utilizing (43), the effect of the operation $`s^nL\{f_\mathrm{d}(t)\}`$ gets depicted by
$$L^1\{s^nL\{f_\mathrm{d}(t)\}\}=u(t)f_\mathrm{d}^{(n)}(t)+\underset{\nu =0}{\overset{n1}{}}f_\mathrm{d}^{(n1\nu )}(0)\delta ^{(\nu )}(t);\text{ }n=1,2,\mathrm{}$$
(48)
As an example, consider the operation $`sL\{\mathrm{cos}\omega t\}`$. From (48) one obtains
$$L^1\{sL\{\mathrm{cos}\omega t\}\}=u(t)\omega \mathrm{sin}\omega t+\delta (t).$$
(49)
The same result is obtained from the LT-correspondences (133, 134), i.e.,
$$L^1\{sL\{\mathrm{cos}\omega t\}\}=L^1\left\{s\frac{s}{s^2+\omega ^2}\right\}=L^1\left\{1\frac{\omega ^2}{s^2+\omega ^2}\right\}=\delta (t)u(t)\omega \mathrm{sin}\omega t.$$
(50)
The most important domain of application of the primary derivation theorem is determination of the evoked solution of the inhomogeneous linear DE; cf. Sect. 4.1.
### 3.2 The secondary derivation theorem
Although d-functions and their derivatives are non-LT-consistent, the question is legitimate how the L-transform of $`f_\mathrm{d}^{(n)}(t)`$ can be expressed by the L-transform of $`f_\mathrm{d}(t)`$. The answer is already implied in (48). From that expression one obtains by L-transformation the
secondary derivation theorem
$$L\{f_\mathrm{d}^{(n)}(t)\}=s^nL\{f_\mathrm{d}(t)\}\underset{\nu =0}{\overset{n1}{}}f_\mathrm{d}^{(n1\nu )}(0)s^\nu ;\text{ }n=1,2,\mathrm{}$$
(51)
The secondary derivation theorem (51) turns out to be identical to TLT’s derivation theorem. This is a consequence of TLT’s original endeavour to provide an operator calculus essentially for d-functions; cf. Eq. (40).
The secondary derivation theorm is not a self-contained derivation theorem as it just emerges from application of the primary derivation theorem to d-functions.
### 3.3 The primary integration theorem
According to (2) the inverse transform’s integral function can be depicted by
$$\phi ^{(1)}(t)=L^1\{s^1F(s)\}=L^1\{s^1L\{f(t)\}\}=L^1\{s^1L\{u(t)f(t)\}\}.$$
(52)
As $`\phi (t)`$ is causal and LT-consistent, the integral function $`\phi ^{(1)}(t)`$ is causal and LT-consistent, as well. Thus for the second-order integral there holds
$$\phi ^{(2)}(t)=L^1\{s^1L\{\phi ^{(1)}(t)\}\}=L^1\{s^2\{L\{\phi (t)\}\}=L^1\{s^2L\{f(t)\}\};$$
(53)
and so on for higher-order integrals. Utilizing (30) one obtains the
primary integration theorem
$$L\{[u(t)f(t)]^{(n)}\}=s^nL\{f(t)\};\text{ }n=0,1,\mathrm{}$$
(54)
which for causal functions $`f(t)=f_\mathrm{c}(t)`$ reads
$$L\{f_\mathrm{c}^{(n)}(t)\}=s^nL\{f_\mathrm{c}(t)\}.$$
(55)
The operation $`s^nL\{f(t)\}`$ is equivalent to $`n`$-fold integration of the causal function $`[u(t)f(t)]`$, which implies $`(n1)`$-fold iteration of the $`t`$-domain operation
$$[u(t)f(t)]^{(1)}=_{\mathrm{}}^tu(\tau )f(\tau )d\tau =u(t)_0^tf(\tau )d\tau .$$
(56)
As $`u(t)f(t)`$ and the integral functions are causal, the $`n`$-fold integral can be expressed by the formula
$$[u(t)f(t)]^{(n)}=\frac{u(t)}{(n1)!}_0^t(t\tau )^{n1}f(\tau )d\tau ;\text{ }n=1,2,\mathrm{}$$
(57)
The consistency of this formula with the primary integration theorem (54) can be verified by L-transformation and application of the convolution theorem (32), utilizing the LT-correspondence (130).
The $`t`$-domain implications of the operation $`s^nL\{f(t)\}`$ are just as different for causal functions versus d-functions as was found for derivation. For causal functions $`f_\mathrm{c}(t)`$ one obtains
$$L^1\{s^nL\{f_\mathrm{c}(t)\}\}=f_\mathrm{c}^{(n)}(t).$$
(58)
When $`f(t)=f_\mathrm{d}(t)`$ is a d-function such that $`f_\mathrm{d}(t)`$ is the $`n`$-th ordinary derivative of $`f_\mathrm{d}^{(n)}(t)`$, there holds
$`[u(t)f_\mathrm{d}(t)]^{(1)}`$ $`=`$ $`{\displaystyle _{\mathrm{}}^t}u(t)f_\mathrm{d}(\tau )d\tau =u(t){\displaystyle _0^t}f_\mathrm{d}(\tau )d\tau `$ (59)
$`=`$ $`u(t)[f_\mathrm{d}^{(1)}(t)f_\mathrm{d}^{(1)}(0)].`$
By iteration of (59) one obtains the so-called
iud-theorem (integration of ud-function)
$$f_{\mathrm{ud}}^{(n)}(t)=[u(t)f_\mathrm{d}(t)]^{(n)}=u(t)f_\mathrm{d}^{(n)}(t)u(t)\underset{\nu =0}{\overset{n1}{}}f_\mathrm{d}^{(n+\nu )}(0)\frac{t^\nu }{\nu !};\text{ }n=1,2,\mathrm{}$$
(60)
For the effect of the operation $`s^nL\{f_\mathrm{d}(t)\}`$ one obtains from (54) and (60)
$$L^1\{s^nL\{f_\mathrm{d}(t)\}\}=u(t)f_\mathrm{d}^{(n)}(t)u(t)\underset{\nu =0}{\overset{n1}{}}f_\mathrm{d}^{(n+\nu )}(0)\frac{t^\nu }{\nu !};\text{ }n=1,2,\mathrm{}$$
(61)
As an example, consider the operation $`s^1L\{\mathrm{exp}(at)\}`$. From (61) one obtains
$$L^1\{s^1L\{\mathrm{e}^{at}\}\}=\frac{u(t)}{a}\mathrm{e}^{at}+\frac{u(t)}{a}.$$
(62)
Using the LT-correspondences (131, 132) one obtains the same result, i.e.,
$$L^1\{s^1L\{\mathrm{e}^{at}\}\}=L^1\left\{\frac{1}{s}\frac{1}{s+a}\right\}=\frac{u(t)}{a}(1\mathrm{e}^{at}).$$
(63)
### 3.4 The secondary integration theorem
By analogy to the secondary derivation theorem, Eq. (61) enables for d-functions expression of the L-transform of the $`n`$-th order integral in terms of the L-transform of the d-function itself. By L-transformation of (61), utilizing (130), there emerges the
secondary integration theorem
$$L\{f_\mathrm{d}^{(n)}(t)\}=s^nL\{f_\mathrm{d}(t)\}+\underset{\nu =0}{\overset{n1}{}}f_\mathrm{d}^{(n+\nu )}(0)s^{\nu 1};\text{ }n=1,2,\mathrm{}$$
(64)
The secondary integration theorem is complementary to the secondary derivation theorem. For $`n=1`$ the two theorems are equivalent; indeed, from (64) one obtains
$$L\{f_\mathrm{d}^{(1)}(t)\}=s^1L\{f_\mathrm{d}(t)\}+f_\mathrm{d}^{(1)}(0)s^1.$$
(65)
Taking into account that by definition $`f_\mathrm{d}(t)`$ is the first ordinary derivative of $`f_\mathrm{d}^{(1)}(t)`$, Eq. (65) is identical to the secondary derivation theorem and thus to TLT’s derivation theorem.
As TLT’s derivation theorem (4, 40) is identical to the secondary derivation theorem (51), one would expect TLT’s integration theorem to comply with the secondary integration theorem (64). However, TLT’s integration theorem (5) actually is identical to the primary integration theorem (54). The latter theorem holds only for causal functions; cf. Eq. (55); its application to d-functions – such as is customary in TLT – will in general yield erroneous results.
Thus, it turns out that the alleged mathematical consistency of TLT’s theorems for derivation/integration is delusive. TLT’s derivation theorem holds only for d-functions whereas TLT’s integration theorem holds only for causal functions. In TLT both theorems are ordinarily used in an untenable way: TLT’s derivation theorem is regarded as the derivation theorem of LT although it holds only for the non-LT-consistent d-functions. TLT’s integration theorem is ordinarily utilized for d-functions although it does not apply to this type of function. Eventually, these observations explain why there is a formal conflict between TLT’s theorems for derivation and integration (4, 5).
The secondary integration theorem is not a self-contained integration theorem as it just emerges from application of the primary integration theorem to d-functions.
### 3.5 The generalized derivation/integration theorem
The LT-consistent theorems for derivation and integration, i.e., (43) and (54), can obviously be unified into one formula, i.e.,
$$L\{[u(t)f(t)]^{(n)}\}=s^nL\{f(t)\}\text{ }n=0,\pm 1,\pm 2,\mathrm{}$$
(66)
Thus, ultimately $`t`$-domain integration is in the L-domain merely “reciprocal” to $`t`$-domain derivation. The $`t`$-domain implications of the operation $`s^nL\{f(t)\}`$ that are outlined in Sects. 3.1, 3.3 must be observed.
As a consequence of the formal equivalence of derivation and integration in the L-domain description, the theorem can be generalized for non-integer order of derivation/integration. For any real $`r0`$ there holds (Riemann, Liouville, Cauchy)
$$D^r\{u(t)f(t)\}=\frac{u(t)}{\mathrm{\Gamma }(r)}_0^t(t\tau )^{r1}f(\tau )d\tau ;\text{ }r0.$$
(67)
The operator $`D`$ denotes generalized derivation/integration, while the negative exponent $`r`$ indicates $`r`$-th-order integration. As the integral in (67) is equivalent to the convolution $`t^{r1}f(t)`$, Eq. (67) can by the convolution theorem (32) be L-transformed, and when the LT-correspondence (141) is utilized one obtains
$`L\{D^r\{u(t)f(t)\}\}`$ $`=`$ $`{\displaystyle \frac{1}{\mathrm{\Gamma }(r)}}L\{t^{r1}\}L\{f(t)\}`$ (68)
$`=`$ $`s^rL\{f(t)\};\text{ }r0.`$ (69)
Equation (69) depicts the generalized integration theorem. This theorem suffices to obtain the same kind of generalization for derivation, namely, by concatenation of (69) and (43). The sequence of $`r`$-th order integration and $`n`$-th order derivation yields $`(nr)`$th order derivation or integration, depending on whether $`n>r`$ or $`n<r`$ . Denoting $`nr=\alpha `$ one obtains the
gdi-theorem (generalized derivation/integration theorem)
$$L\{D^\alpha \{u(t)f(t)\}\}=s^\alpha L\{f(t)\};\text{ }\alpha \mathrm{}.$$
(70)
In the form (70) the theorem holds for any type of real function $`f(t)`$. For integer values of $`\alpha `$ Eq. (70) is equivalent to (66).
The concept of fractional calculus, i.e., utilization of derivatives of non-integer order, is essentially based on the formula (67), and this formula holds only for causal functions. In the realm of TLT derivatives of non-integer order can not be consistently expressed at all, as TLT’s derivation theorem neither holds for causal functions nor for non-integer order of derivation. When TLT’s derivation theorem is used anyway, initial values become involved which merely are a nuisance. It is by the present approach – which provides for the generalized theorem (66) – that Laplace transformation becomes a consistent and efficient tool for doing fractional calculus.
Examples for the application of (70), i.e., for $`\alpha =1/2`$, are listed in Sect. A.5.
### 3.6 Summary
LT-consistent theorems for t-domain derivation and integration are obtained by deduction from the behavior of the inverse L-transform. These theorems – the primary theorems – are formally congruent and thus can be unified. The unified primary theorem is generalized for non-integer order of derivation/integration. The generalized theorem (the gdi-theorem) accounts in the L-domain for both integer-order and non-integer-order of $`t`$-domain derivation/integration. As the inverse transform is causal the primary theorems and the gdi-theorem hold for causal functions.
The respective secondary theorems emerge from application of the primary ones to the L-transforms of ordinary derivable functions (d-functions). TLT’s derivation theorem is identical to the secondary derivation theorem. This theorem can not be regarded as the derivation theorem of Laplace transformation because it applies only to the class of d-functions, which is non-LT-consistent. TLT’s integration theorem is identical to the primary integration theorem and therefore holds only for causal functions; its utilization for d-functions – such as in TLT – is untenable.
## 4 Solution of linear differential equations
Below, the new approach to the solution of linear DEs by LT is demonstrated for a fairly general form of the ordinary linear DE, namely,
$$\underset{n=0}{\overset{N}{}}a_\mathrm{n}y^{(n)}(t)=\underset{m=0}{\overset{M}{}}b_\mathrm{m}x^{(m)}(t);\text{ }N=1,2\mathrm{};\text{ }M=0,1\mathrm{}$$
(71)
The coefficients $`a_\mathrm{n}`$, $`b_\mathrm{m}`$ are presupposed to be real constants. Two functions are involved, namely, the excitation function $`x(t)`$ and the response function $`y(t)`$. The form (71) is more general than that accounted for by the TLT method, as in (71) derivatives of the excitation function are allowed to be included, i.e., corresponding to $`M>0`$.
A clear distinction is made between the evoked solution and the spontaneous solution of the inhomogeneous DE . The evoked solution (the pertinent system’s evoked response to $`x(t)`$) is that part of the total response that is elicited by the excitation function $`x(t)`$ alone. The spontaneous solution (the system’s “spontaneous” response), if it exists, is ascribed to a pre-excited initial state of the system. Mathematically, the evoked solution is the DE’s total (general) solution minus the general solution of the pertinent homogeneous DE. The spontaneous solution is identical and synonymous to the general solution of the homogeneous DE.
### 4.1 The evoked response: <br>Particular solution of the inhomogeneous DE
For the excitation function $`x(t)`$ to be LT-consistent it must be defined as a causal function. As a consequence, its derivatives have the form $`[u(t)x(t)]^{(m)}`$. Another consequence is that the evoked response $`y_\mathrm{e}(t)`$ is causal as well; thus, its derivatives implicitly assume the form $`[u(t)y_\mathrm{e}(t)]^{(n)}`$. Therefore, conversion of (71) into the L-domain is governed by the primary derivation theorem (43). The L-domain representation of (71) reads
$$\underset{n=0}{\overset{N}{}}a_\mathrm{n}s^nL\{y_\mathrm{e}(t)\}=\underset{m=0}{\overset{M}{}}b_\mathrm{m}s^mL\{x(t)\};\text{ }N=1,2\mathrm{};\text{ }M=0,1\mathrm{},$$
(72)
and one eventually obtains the evoked solution
$$y_\mathrm{e}(t)=L^1\left\{\frac{\underset{m=0}{\overset{M}{}}b_\mathrm{m}s^m}{_{n=0}^Na_\mathrm{n}s^n}L\{x(t)\}\right\}.$$
(73)
In contrast to the TLT method, the evoked response is by (73) obtained without involvement of initial values, i.e., additional constants. To obtain the evoked response one does not have to pretend that the pertinent system is in a non-preexcited initial state. With respect to the evoked response the system’s initial state is irrelevant.
For the operation $`L^1`$, i.e., inverse L-transformation, there exist two alternatives. The first of them is taking (73) at face value, which implies that the L-transform of $`x(t)`$ has to be included. The second alternative exploits the convolution theorem (32), whereby inverse L-transformation can be confined to the first factor in (73):
$$h(t)=L^1\left\{\frac{\underset{m=0}{\overset{M}{}}b_\mathrm{m}s^m}{_{n=0}^Na_\mathrm{n}s^n}\right\}.$$
(74)
The function $`h(t)`$ depicts the evoked response to the excitation function $`x(t)=\delta (t)`$, i.e., the impulse response. As (73) has the form
$$y_\mathrm{e}(t)=L^1\{L\{h(t)\}L\{x(t)\}\},$$
(75)
by (32) there emerges the familiar convolution formula
$$y_\mathrm{e}(t)=u(t)_0^th(t\tau )x(\tau )d\tau .$$
(76)
For example, for $`N=1`$, $`M=0`$ one obtains the impulse response
$$h(t)=L^1\left\{\frac{b_0}{a_0+a_1s}\right\}=u(t)\beta \mathrm{e}^{\alpha t};\text{ }\alpha =a_0/a_1;\text{ }\beta =b_0/a_1.$$
(77)
Utilizing (77) and (76), one obtains the
evoked solution of the first-order linear DE
$$y_\mathrm{e}(t)=u(t)\beta _0^tx(\tau )\mathrm{e}^{\alpha (t\tau )}d\tau =u(t)\beta \mathrm{e}^{\alpha t}_0^tx(\tau )\mathrm{e}^{\alpha \tau }d\tau .$$
(78)
Though this solution is entirely expressed in the $`t`$-domain, it nevertheless is based on LT, which implies that (78) is correct only for causal excitation functions $`x(t)`$.
It may be noted that it is the independent LT-based expression of the evoked solution (73) that provides a solid basis to operator calculus as it is customarily employed in the theories of linear control systems and electrical circuits.
### 4.2 The spontaneous response: <br>General solution of the homogeneous DE
When the evoked response is by LT determined as just described, for the solution of the homogeneous DE still any method available can be chosen. From the present approach there emerges a new LT-based method which exploits the facts that
a) the spontaneous response $`y_\mathrm{s}(t)`$ and its derivatives are d-functions; their L-transforms have the form $`L\{y_\mathrm{s}^{(n)}(t)\}=L\{u(t)y_\mathrm{s}^{(n)}(t)\}`$;
b) the homogeneous DE can by the dud-theorem be converted into an equivalent inhomogeneous DE.
To get the homogeneous DE made up for conversion into an equivalent inhomogeneous DE, it is multiplied by $`u(t)`$; this yields
$$a_0u(t)y_\mathrm{s}(t)+a_1u(t)y_\mathrm{s}^{}(t)+\mathrm{}+a_Nu(t)y_\mathrm{s}^{(N)}(t)=0.$$
(79)
Because of LT’s ambivalence, the L-transform of (79) is identical to the L-transform of the original homogeneous DE. Application of the dud-theorem (47) converts (79) into the form
$`a_0u(t)y_\mathrm{s}(t)`$ $`+`$ $`a_1[u(t)y_\mathrm{s}(t)]^{}y_\mathrm{s}(0)\delta (t)+a_2[u(t)y_\mathrm{s}(t)]^{\prime \prime }y_\mathrm{s}^{}(0)\delta (t)y_\mathrm{s}(0)\delta ^{}(t)`$ (80)
$`+`$ $`\mathrm{}+a_N[u(t)y_\mathrm{s}(t)]^{(N)}{\displaystyle \underset{\nu =0}{\overset{N1}{}}}y_\mathrm{s}^{(N1\nu )}(0)\delta ^{(\nu )}(t)=0.`$
Equation (80) is equivalent to (79). However, (80) actually is an inhomogeneous DE. The impulse functions included in (80) play the role of virtual excitation functions . This becomes particularly apparent when (80) is expressed in the form
$$\underset{n=0}{\overset{N}{}}a_n[u(t)y_\mathrm{s}(t)]^{(n)}=\underset{\mu =0}{\overset{N1}{}}c_\mu \delta ^{(\mu )}(t),$$
(81)
where the coefficients $`c_\mu `$ are depicted by
$$c_\mu =\underset{\nu =0}{\overset{N1\mu }{}}a_{\mu +\nu +1}y_\mathrm{s}^{(\nu )}(0).$$
(82)
From (81) the solution of the homogeneous DE can be obtained as an evoked solution, i.e., as described in Sect. 4.1. One eventually obtains
$$u(t)y_\mathrm{s}(t)=L^1\left\{\frac{\underset{\mu =0}{\overset{N1}{}}c_\mu s^\mu }{_{n=0}^Na_ns^n}\right\}.$$
(83)
The notation $`u(t)y_\mathrm{s}(t)`$ is not redundant, because $`y_\mathrm{s}(t)`$ is a d-function. Equation (83) in fact depicts the evoked response of a virtual system, i.e., to the excitation function $`\delta (t)`$. The transmission function of that virtual system is defined by the quotient in (83), whose numerator essentially originates from the impulse functions contained in (81).
Finally, the spontaneous solution $`y_\mathrm{s}(t)`$ itself can by extrapolation into $`t0`$ be obtained from (83). As according to the causality theorem (30) the operation $`L^1`$ on the right side of (83) yields either explicitly or implicitly a function of the form $`u(t)y_\mathrm{s}(t)`$, that kind of extrapolation is equivalent to cancellation of $`u(t)`$ on both sides of (83), i.e., after inverse L-transformation.
For example, for $`N=1`$ one obtains from (83) and (82)
$$u(t)y_\mathrm{s}(t)=L^1\left\{\frac{a_1y_\mathrm{s}(0)}{a_0+a_1s}\right\}=u(t)y_\mathrm{s}(0)\mathrm{e}^{\alpha t};\text{ }\alpha =a_0/a_1.$$
(84)
By cancellation of $`u(t)`$ there emerges from (84) the
spontaneous solution of the first-order linear DE
$$y_\mathrm{s}(t)=y_\mathrm{s}(0)\mathrm{e}^{\alpha t}\text{ for }\mathrm{}<t.$$
(85)
The described deduction of (83) ultimately constitutes a new method for obtaining the general solution of the linear homogeneous DE of finite order $`N`$, i.e., by Laplace transformation. The method involves a) determination by (82) of the coefficients $`c_\mu `$, i.e., from the DE’s coefficients $`a_n`$; b) insertion of the coefficients $`c_\mu `$ into (83); c) inverse L-transformation; and d) cancellation of the factor $`u(t)`$. The $`N`$ arbitrary constants which invariably get involved are provided by the initial values at $`t=0`$ of the solution $`y_\mathrm{s}(t)`$ itself and of the latter’s $`N1`$ derivatives. The method is straightforward; yet, mathematical intricacies may occur in step c), i.e., inverse L-transformation.
As a pragmatic alternative to the described deduction, the formula (83) may be obtained in a more immediate way, namely, by using for LT-conversion of the homogeneous DE the secondary derivation theorem (51). This option is enabled by the fact that the secondary derivation theorem accounts for the combination of the dud-theorem with the primary derivation theorem; cf. Sect. 3.2. As the secondary derivation theorem is identical to TLT’s derivation theorem, this option explains why TLT’s total solution of the inhomogeneous DE is quite similar to the new total solution; see the following two sections.
### 4.3 The total solution
Assuming that for the solution of the homogeneous DE the method just described is chosen, the total solution of (71) can be compactly depicted by a formula, i.e.,
$$y(t)=L^1\left\{\frac{\underset{m=0}{\overset{M}{}}b_\mathrm{m}s^m}{_{n=0}^Na_\mathrm{n}s^n}L\{x(t)\}\right\}+L_\mathrm{d}^1\left\{\frac{\underset{\mu =1}{\overset{N}{}}a_\mu \underset{\nu =0}{\overset{\mu 1}{}}y_\mathrm{s}^{(\mu 1\nu )}(0)s^\nu }{_{n=0}^Na_\mathrm{n}s^n}\right\}.$$
(86)
The first term on the right side of (86) depicts the evoked response and is identical to (73). The second term depicts the spontaneous response by one single expression; this term is equivalent to the combination of Eqs.(83) and (82). The operator $`L^1`$ indicates inverse transformation by (2). The operator $`L_\mathrm{d}^1`$ indicates inverse transformation by (2) followed by extrapolation into $`t0`$, i.e., cancellation of the factor $`u(t)`$.
When the inverse L-transform is looked up from a TLT-based table of LT-correspondences it must be observed that in those tables the $`t`$-domain functions are depicted in the non-causal form. In Sect. A.7 examples are listed of the LT-consistent notation of $`t`$-domain functions.
It must be kept in mind that LT-based solutions of linear DEs hold only for causal excitation functions $`x(t)=u(t)x(t)`$. The solution for the “steady-state” may be obtained by asymptotical approach, i.e., for $`t\mathrm{}`$.
An example for application of (86) is depicted in Sect. A.6.
Finally, it should be noticed that for $`N=1`$, $`M=0`$ the LT-based solution of (71) can be entirely expressed in the $`t`$-domain. By superposition of Eqs. (78) and (85) one obtains the
total solution of the first-order linear DE
$$y(t)=y_\mathrm{e}(t)+y_\mathrm{s}(t)=u(t)\beta \mathrm{e}^{\alpha t}_0^tx(\tau )\mathrm{e}^{\alpha \tau }d\tau +y_\mathrm{s}(0)\mathrm{e}^{\alpha t};\text{ }\alpha =a_0/a_1;\text{ }\beta =b_0/a_1.$$
(87)
Thus, for the special case that in (71) there is $`N=1`$, $`M=0`$ one does not actually have to do L-transformation at all. The solution is reduced to evaluation of the integral contained in (87) and it holds for causal excitation functions $`x(t)=u(t)x(t)`$.
### 4.4 TLT’s initial-value conflict
One of the most annoying deficiencies of TLT is potential interference of the DE’s evoked solution with the spontaneous solution. The danger of this kind of interference to occur emerges from the fact that the DE’s solution obtained by TLT includes an inappropriate type of initial values. This is why the phenomenon is termed the initial-value conflict.
When the general solution of a linear DE of the form (71) is worked out by means of TLT – which is possible for $`M=0`$ – one eventually obtains the formula
$$y(t)=L^1\left\{\frac{b_0}{_{n=0}^Na_\mathrm{n}s^n}L\{x(t)\}+\frac{\underset{\mu =1}{\overset{N}{}}a_\mu \underset{\nu =0}{\overset{\mu 1}{}}y^{(\mu 1\nu )}(0)s^\nu }{_{n=0}^Na_\mathrm{n}s^n}\right\}.$$
(88)
This formula is quite similar to (86), i.e., for $`M=0`$. (The constant $`b_0`$ is in (88) included just for formal compatibility with (86); one may in both equations assume $`b_0=1`$.) Letting aside the difference in inverse L-transformation that is apparent by comparison of (88) to (86), there remains the crucial difference that (86) correctly contains the initial values of the spontaneous solution, $`y_\mathrm{s}^{(\nu )}(0)`$, whereas (88) contains the initial values of the total solution, $`y^{(\nu )}(0)`$. The latter, inadequate kind of initial values inevitably emerge from LT-conversion of the DE (71) by TLT’s derivation theorem.
The potential initial-value conflict arises from the fact that the evoked and the spontaneous solutions are superimposed in $`y(t)`$, such that
$$y^{(\nu )}(0)=y_\mathrm{e}^{(\nu )}(0)+y_\mathrm{s}^{(\nu )}(0);\text{ }\nu =0,1,\mathrm{}$$
(89)
The initial values of the evoked solution, $`y_\mathrm{e}^{(\nu )}(0)`$, depend on $`x(t)`$ and on the DE’s coefficients $`a_\mathrm{n}`$, as is depicted by (73). If there happens to be $`y_\mathrm{e}^{(\nu )}(0)0`$, i.e., for all $`\nu =0,1,\mathrm{},N1`$, then there holds $`y^{(\nu )}(0)y_\mathrm{s}^{(\nu )}(0)`$, and (88) will yield the same result as (86), i.e., except for the difference in inverse L-transformation, and for $`M=0`$. However, for many types of system and of excitation function $`y_\mathrm{e}^{(\nu )}(0)`$ may turn out to be different from null, i.e., for at least one value of $`\nu `$. If this occurs then the result obtained by TLT is incorrect; the initial values $`y^{(\nu )}(0)`$ can no longer be freely chosen, i.e., independent of the excitation function; cf. Sect. A.1.
Obviously, (88) can easily be “patched”, i.e., by arbitrarily substituting the values $`y_\mathrm{s}^{(\nu )}(0)`$ for $`y^{(\nu )}(0)`$. Most remarkably, there exists another, less trivial kind of patch, namely, arbitrary substitution in (88) of the values $`y^{(\nu )}(0)`$ for $`y^{(\nu )}(0)`$. The reason why the latter kind of patch works around the initial-value conflict originates from the fact that the evoked response and its derivatives are causal (cf. Sect. 4.1), such that there holds $`y_\mathrm{e}^{(\nu )}(0)0`$, i.e., even when $`y_\mathrm{e}^{(\nu )}(0)0`$. As the spontaneous response and its derivatives are d-functions, there holds $`y_\mathrm{s}^{(\nu )}(0)y_\mathrm{s}^{(\nu )}(0)`$. Thus, one obtains from (89)
$$y^{(\nu )}(0)=y_\mathrm{s}^{(\nu )}(0)\text{ for }\nu =0,1,\mathrm{},N1.$$
(90)
The identity (90) holds for any type of linear DE and of excitation function, From this identity there follows that substitution in (88) of the initial point $`t=0`$ for $`t=0`$ renders (88) compatible with (86), such that TLT’s solution becomes essentially correct, i.e., except for the difference in inverse L-transformation. The “empirical” finding that by this kind of patch TLT’s initial-value conflict actually is worked around, becomes in this way explained.
It should be noticed that the latter kind of patch is just as arbitrary as the former. Even if one had no doubt at all about TLT’s consistency, the mere fact that TLT’s solution (88) in general needs to be patched should have elicited that kind of doubt. One can not in earnest be satisfied with either patch’s working around TLT’s initial-value conflict.
### 4.5 Summary
The total solution of the inhomogeneous linear DE is obtained by two independent algorithms, i.e., one for the evoked response, another for the spontaneous response. As both the excitation function and the evoked response function are causal, the derivatives of these functions assume the form $`[u(t)x(t)]^{(m)}`$ and $`[u(t)y_\mathrm{e}(t)]^{(n)}`$, respectively. When the inhomogeneous DE is by the primary derivation theorem converted into the L-domain one obtains a linear algebraic equation that does not contain any arbitrary constants. This equation is solved in the familiar way to obtain the evoked solution $`y_\mathrm{e}(t)`$ which is defined for $`\mathrm{}<t`$.
For the determination of the spontaneous response $`y_\mathrm{s}(t)`$, i.e., solution of the pertinent homogeneous DE, the fact that the response function and its derivatives are d-functions is a priori taken into account. The L-transforms of these functions assume the form $`L\{y_\mathrm{s}^{(n)}(t)\}=L\{u(t)y_\mathrm{s}^{(n)}(t)\}`$. As a consequence, the homogeneous DE can be multiplied by $`u(t)`$ without affecting its L-transform. The homogeneous DE such modified can by the dud-theorem be converted into an equivalent inhomogeneous DE. The L-transform of the latter DE depicts a virtual linear system whose impulse response equals for $`t>0`$ the spontaneous solution of the original homogeneous DE. By extrapolation into $`t0`$ one obtains from that impulse response the spontaneous solution $`y_\mathrm{s}(t)`$ for $`\mathrm{}<t`$.
For $`M=0`$, i.e., if the excitation function does not include any derivative, the total solution obtained by TLT is formally similar to the correct solution depicted by the new method. However, the TLT-based solution suffers from an initial-value conflict. This conflict can be worked around by manipulation of the initial values. The necessity of patching TLT’s solution is another symptom of TLT’s deficiency.
## 5 Concluding remarks
From the theorems pointed out in Sect. 2 it should be apparent that the key to the consistent LT-theory lies in the hidden bilaterality of the transformation integral (1). By strictly taking that bilaterality into account there emerges a structure of LT-theory that is fundamentally different from that of TLT. Yet, many of the new theory’s theorems and expressions are familiar from TLT. Pronounced formal differences from TLT’s expressions occur, e.g., in the context of the theorems for derivation and integration. The difference between the new formula (86) for the general solution of the linear inhomogeneous DE from the corresponding formula (88) obtained by TLT, though at first sight marginal, actually is crucial, namely, for resolution of TLT’s inherent initial-value conflict.
Mathematical consistency of the new approach to LT is achieved at the expense of non-ordinary $`t`$-domain functions being crucially involved – primarily, the inverse L-transform $`\phi (t)`$, which includes the connect function $`u_0(t)`$ and, possibly, its (non-ordinary) derivatives. This notion brings to mind the aspect of mathematical rigour. If one maintains that there is no rigorous mathematical account for non-ordinary functions and non-ordinary derivatives then the inverse L-transform can not in general be rigorously accounted for and a rigorous theory of Laplace transformation can not exist at all.
Rigorous treatment of non-ordinary functions may be achieved by invoking the theory of distributions. However, what makes LT-theory consistent in the first place is adequate incorporation of non-ordinary functions and non-ordinary derivatives. Mere attachment of distribution theory to TLT can not cure TLT’s inconsistency. Mathematically rigorous treatment of non-ordinary functions and derivatives may be achieved separately and subsequently – which, historically, actually has happened to the delta-impulse.
Making in this sense a distinction between mathematical consistency and rigour, one can say that the new approach to Laplace transformation outlined in the present article is consistent, even though it may not fully qualify as mathematically rigorous. This achievement is distinctly preferable to inconsistency disguised by fake mathematical rigour – which virtually is what is offered by Doetsch and many others.
Once one has developed some scepticism about TLT one may be prepared to realizing that Doetsch’s books are written in a somewhat defensive and dogmatic style, and that Doetsch devoted considerable portions of text to reasoning by plausibility and to explaining away apparent discrepancies. From these observations it may be concluded that Doetsch virtually was aware of TLT’s inherent problems. Indeed, in the preface to he suggested the need for a fundamental redesign of TLT. However, until the end of his life (1977) he stuck to the original layout of TLT, i.e., with some distribution theory as a complement. In the two volumes of his Handbook , Doetsch assembled a wealth of mathematics on LT – which, however, may distract from TLT’s fundamental flaws.
To obtain a more thorough understanding of TLT’s unfortunate history it is probably helpful to take notice of Doetsch’s biography .
## Appendix A APPENDIX
### A.1 Failure of TLT: An example DE
TLT promises to provide a method for obtaining the general solution of the linear inhomogeneous DE, i.e., for any kind of excitation function $`x(t)`$. Below, this promise is challenged by an example, i.e., for the first-order DE
$$ay(t)+y^{}(t)=\widehat{x}\mathrm{sin}\omega t,$$
(91)
where $`\widehat{x}\mathrm{sin}\omega t`$ denotes the excitation function and $`y(t)`$ denotes the response function.
From the theory of linear DE’s one obtains the general solution
$$y(t)=\frac{\widehat{x}}{a^2+\omega ^2}(a\mathrm{sin}\omega t\omega \mathrm{cos}\omega t)+y_\mathrm{s}(0)\mathrm{e}^{at},$$
(92)
where $`y_\mathrm{s}(0)`$ is an arbitrary real constant which specifies any particular solution of the pertinent homogeneous DE.
By the TLT method one obtains the solution
$`y(t)`$ $`=`$ $`L^1\left\{{\displaystyle \frac{\widehat{x}\omega }{s^2+\omega ^2}}{\displaystyle \frac{1}{s+a}}\right\}+L^1\left\{{\displaystyle \frac{y(0)}{s+a}}\right\}`$ (93)
$`=`$ $`{\displaystyle \frac{\widehat{x}}{a^2+\omega ^2}}(\omega \mathrm{e}^{at}+a\mathrm{sin}\omega t\omega \mathrm{cos}\omega t)+y(0)\mathrm{e}^{at};\text{ for }t>0.`$
The result (93) differs from the correct solution (92) in two significant respects:
a) As compared to the correct evoked response
$$y_\mathrm{e}(t)=\frac{\widehat{x}}{a^2+\omega ^2}(a\mathrm{sin}\omega t\omega \mathrm{cos}\omega t)$$
(94)
Eq. (93) includes an extra decaying exponential.
b) As compared to the correct spontaneous response
$$y_\mathrm{s}(t)=y_\mathrm{s}(0)\mathrm{e}^{at}$$
(95)
Eq. (93) includes the initial value $`y(0)`$ of the total response $`y(t)=y_\mathrm{e}(t)+y_\mathrm{s}(t)`$ instead of that of the spontaneous response alone, i.e., $`y_\mathrm{s}(0)`$.
The solutions (92) and (93) both are consistent with the DE (91). Yet only (92) represents the DE’s general solution. This becomes apparent when one attempts formal reconciliation of (93) with (92). Reconciliation can only be obtained by suitable choice of a particular “initial state”. For instance, for $`y_\mathrm{s}(0)=0`$ Eqs. (93, 92) can only be reconciled by setting
$$y(0)=\frac{\widehat{x}\omega }{a^2+\omega ^2}.$$
(96)
Hence, $`y(0)`$ is not really a free parameter of the spontaneous response. While the consistency of (93) with (91) suggests that the TLT-based solution is correct, comparison with (92) reveals that (93) actually is only a particular solution of (91) and that $`y(0)`$ is not a freely assignable constant.
In point of fact, the correct solution of (91) can by L-transformation not be obtained at all, because the L-transform of $`\widehat{x}\mathrm{sin}\omega t`$ is just an alias for $`L\{u(t)\widehat{x}\mathrm{sin}\omega t\}`$; cf. Sect. 2.4. The evoked part of the TLT-based solution (93) inevitably depicts for $`t>0`$ the response to $`u(t)\widehat{x}\mathrm{sin}\omega t`$ instead of to $`\widehat{x}\mathrm{sin}\omega t`$. The above observations reveal that LT actually is sensitive to the difference between $`x(t)`$ and $`u(t)x(t)`$, i.e., if there is $`x(t)u(t)x(t)`$.
From a more general point of view, it should be appreciated that any “causal” physical system which is governed by a linear DE of finite order has a “memory” which becomes manifest in the length of the system’s impulse response. This notion suffices to explain that the evoked response $`y_\mathrm{e}(t)`$ to an excitation function $`x(t)`$ in general depends on the behavior of $`x(t)`$ for $`\mathrm{}<t`$ . When the response is determined by LT, this fact is automatically accounted for, namely. by the implicit bilaterality of (1); cf. Sects. 2.1, 2.4, A.2. The bilaterality of $`\phi (t)`$ is indispensable for the evoked response to be correct, i.e., whether or not $`\phi (t)`$ is causal. L-transformation by (1) just provides for $`\phi (t)`$ being always causal. As a consequence, the evoked response complies with $`x(t)`$ only if $`x(t)`$ was a priori defined to be causal.
### A.2 Analysis interval versus integration interval
In Sect. 2.1 it is pointed out that the t-domain interval which actually is encompassed both by Fourier- and Laplace-transformation invariably is infinite, i.e., $`\mathrm{}<t<+\mathrm{}`$. This interval is termed the respective transformation’s analysis interval; it has to be distinguished from the integration interval, i.e., the interval of $`t`$ which is determined by the transformation-integral’s limits, e.g. in Eqs. (1) and (8). The relationship between these two kinds of interval can be made apparent by regressing to discrete Fourier- and Laplace-transformation. The discrete transformations can be deduced from the Fourier-series representation of a real function $`f(t)`$, i.e.,
$$f(t)=a_0+\underset{n=1}{\overset{\mathrm{}}{}}a_\mathrm{n}\mathrm{cos}\omega _\mathrm{n}t+\underset{n=1}{\overset{\mathrm{}}{}}b_\mathrm{n}\mathrm{sin}\omega _\mathrm{n}t;\text{ }n\text{ integer;}$$
(97)
where
$$a_0=\frac{1}{T}_{T/2}^{+T/2}f(t)dt;$$
(98)
$$a_\mathrm{n}=\frac{2}{T}_{T/2}^{+T/2}f(t)\mathrm{cos}\omega _\mathrm{n}t\mathrm{d}t;$$
(99)
$$b_\mathrm{n}=\frac{2}{T}_{T/2}^{+T/2}f(t)\mathrm{sin}\omega _\mathrm{n}t\mathrm{d}t;$$
(100)
$$\omega _\mathrm{n}=\frac{2\pi n}{T}.$$
(101)
The function $`f(t)`$ such represented is periodic with the period length $`T=2\pi /\omega _1`$.
This set of formulas can be regarded and employed as a complementary pair of transformations . By Eqs. (98-100) a particular section of $`f(t)`$ becomes transformed into an infinite set of real coefficients {$`a_\mathrm{n},b_\mathrm{n}`$} each of which pertains to a discrete frequency $`\omega _\mathrm{n}`$. The section of $`f(t)`$ that extends from $`T/2`$ to $`+T/2`$ plays the role of an analysis interval, $`T`$. The function $`f(t)`$ needs to be defined for $`T/2t+T/2`$. Inverse transformation is achieved by (97). In general, the inverse transform matches the original function $`f(t)`$ only for $`T/2<t<+T/2`$. By the notation chosen in Eqs. (98-100) the center of the analysis interval becomes implicitly denoted $`t=0`$.
To elucidate the implications of this approach it is helpful to express the above formulas in complex notation. Using Euler’s formula, one obtains from Eqs. (99, 100)
$$a_\mathrm{n}\pm \mathrm{i}b_\mathrm{n}=\frac{2}{T}_{T/2}^{+T/2}f(t)\mathrm{e}^{\pm \mathrm{i}\omega _\mathrm{n}t}dt;\text{ }n=1,2,\mathrm{}$$
(102)
By definition of the discrete Fourier transform
$$F(\omega _\mathrm{n},T)=\frac{T}{2}(a_\mathrm{n}\mathrm{i}b_\mathrm{n})$$
(103)
one obtains from (102)
$$F(\omega _\mathrm{n},T)=_{T/2}^{+T/2}f(t)\mathrm{e}^{\mathrm{i}\omega _\mathrm{n}t}dt.$$
(104)
Equation (97) can be expressed in the form
$$f(t)=a_0+\frac{1}{2}\underset{n=1}{\overset{\mathrm{}}{}}(a_\mathrm{n}\mathrm{i}b_\mathrm{n})\mathrm{e}^{\mathrm{i}\omega _\mathrm{n}t}+\frac{1}{2}\underset{n=1}{\overset{\mathrm{}}{}}(a_\mathrm{n}+\mathrm{i}b_\mathrm{n})\mathrm{e}^{\mathrm{i}\omega _\mathrm{n}t}.$$
(105)
Observing that by (98, 104) there holds $`a_0=F(0,T)/T`$, one obtains from (105) and (103) for $`n=\mathrm{},1,0,+1,\mathrm{}`$ the formula for inverse transformation
$$\phi (t)=\frac{1}{T}\underset{n=\mathrm{}}{\overset{+\mathrm{}}{}}F(\omega _\mathrm{n},T)\mathrm{e}^{\mathrm{i}\omega _\mathrm{n}t}=f(t)\text{ for }T/2<t<+T/2.$$
(106)
The transformation (104) is governed by, and confined to, the analysis interval $`T`$. This interval can either be regarded as preset or as being determined by the choice of $`\omega _1`$ and thus of the spacing of the discrete frequencies $`\omega _\mathrm{n}`$, cf. (101). As the center of the analysis interval is for convenience denoted $`t=0`$ there is only one way in which the limits of the integral (104) may become different from $`\pm T/2`$, namely, by $`f(t)`$ being null within a sub-interval of $`T`$ that borders $`t=T/2`$ and/or $`t=+T/2`$. Vice versa, when one or both of the integral’s limits are different from $`\pm T/2`$ this invariably indicates that the function $`f(t)`$ contains a null interval. In particular, when $`f(t)`$ is replaced with the causal function $`f_\mathrm{c}(t)=u(t)f(t)`$ the discrete Fourier transform gets expressed by
$$F(\omega _\mathrm{n},T)=_{T/2}^{+T/2}u(t)f(t)\mathrm{e}^{\mathrm{i}\omega _\mathrm{n}t}dt=_0^{T/2}f(t)\mathrm{e}^{\mathrm{i}\omega _\mathrm{n}t}dt=_0^{T/2}f_\mathrm{c}(t)\mathrm{e}^{\mathrm{i}\omega _{\mathrm{nu}}t}dt.$$
(107)
The difference from $`T/2`$ of the integral’s low limit neither affects the analysis interval $`T`$ nor does it make the transformation unilateral; it just indicates that $`f(t)`$ is causal.
The discrete variant of Laplace transformation emerges from (104) by including in the integral the factor $`\mathrm{exp}(\sigma t)`$ and by setting the integral’s low limit to $`t=0`$. This yields
$$L_\mathrm{T}\{f(t)\}=_0^{T/2}f(t)\mathrm{e}^{s_\mathrm{n}t}dt,$$
(108)
where $`L_\mathrm{T}`$ denotes discrete L-transformation, and
$$s_\mathrm{n}=\sigma +\mathrm{i}\omega _\mathrm{n}.$$
(109)
Regarding the relationship between analysis interval and integration interval, the same applies as was just pointed out above: The integral’s (108) low limit $`t=0`$ indicates that the function actually transformed is causal while the low limit marks the center of the analysis interval.
The inverse discrete L-transform $`\phi (t)`$ is determined by the expression
$$\phi (t)=L_\mathrm{T}^1\{L_\mathrm{T}\{f(t)\}\}=\frac{1}{T}\underset{n=\mathrm{}}{\overset{+\mathrm{}}{}}L_\mathrm{T}\{f(t)\}\mathrm{e}^{s_\mathrm{n}t};\text{ }T/2<t<+T/2.$$
(110)
When (108) is employed for transformation of a non-causal function $`f(t)`$, the inverse transform $`\phi (t)`$ still is defined for $`T/2<t<+T/2`$; however, validity of $`\phi (t)=f(t)`$ is confined to the interval $`0<t<T/2`$. In this sense, the transformation is for non-causal functions inconsistent. For causal functions $`f(t)=f_\mathrm{c}(t)`$ the transformation is consistent, as there holds $`\phi (t)=f_\mathrm{c}(t)`$ for $`T/2<t<+T/2`$.
Laplace transformation as defined by (1, 2) emerges from (108, 110) by letting $`T\mathrm{}`$, implying that the spacing of analysis frequencies becomes infinitesimally small such that the sum (110) becomes an integral, i.e., (2). The transformation integral’s (1) low limit, i.e., $`t=0`$, marks the center of the infinite analysis interval.
### A.3 Redundancy of multiplication by u(t)
When the unit step function is defined without inclusion of the connect function $`u_0(t)`$, i.e., $`u(t)=0`$ for $`t<0`$; $`u(t)=1`$ for $`t0`$, there evidently holds $`u^n(t)=u(t)`$ ($`n=1,2,\mathrm{}`$). Hence, multiplication of $`u(t)`$ by itself is redundant. As $`u(t)`$ actually includes $`u_0(t)`$ it must be verified that this kind of redundancy also holds when $`u(t)`$ is defined according to (14).
For $`u^n(t)=u(t)`$ to hold it is necessary and sufficient that $`u_0^n(t)=u_0(t)`$. As was outlined in Sect. 2.3, the unit connect function $`u_0(t)`$ is sufficiently characterized by saying that it is an infinite set of real numbers {$`0\mathrm{}1`$} that exists at $`t=0`$ and fills the gap which otherwise exists between $`u(0)=0`$ and $`u(+0)=1`$. From this definition it follows that any integer power of $`u_0(t)`$ is characterized by precisely the same criterion: $`u_0^n(t)`$ is for any $`n=2,3,\mathrm{}`$ also an infinite set of real numbers {0…1}. Thus, any power of $`u_0(t)`$ can be renamed $`u_0(t)`$. Therefore, there actually holds the identity
$$u^n(t)=u(t);\text{ }n=1,2,\mathrm{}$$
(111)
Another important question is whether multiplication by $`u(t)`$ of a derivative of $`u(t)`$, i.e., of an impulse function, is also redundant such that there holds
$$u(t)u^{(n)}(t)=u(t)\delta ^{(n1)}(t)=u^{(n)}(t)=\delta ^{(n1)}(t);\text{ }n=1,2,\mathrm{}$$
(112)
The identity (112) turns out to be an inevitable consequence of the fact that the impulse functions are LT-consistent (Sect. 2.2, Eq. (12)), in combination with the causality theorem (30). When in (30) one lets $`f(t)=\delta ^{(n)}(t)`$ one obtains from (12) and (30)
$$L^1\{L\{\delta ^{(n)}(t)\}\}=\delta ^{(n)}(t)=u(t)\delta ^{(n)}(t).$$
(113)
Thus, the identity (112) actually is implied in LT’s fundamental definitions and features. The identity can be made plausible by characterizing $`\delta ^{(n)}(t)`$ as a set of pseudo-functions the type of which depends on $`n`$ and which only at $`t=0`$ are different from 0. Then it may be concluded that multiplication of $`\delta ^{(n)}(t)`$ by $`u_0(t)`$ does not alter the type of function denoted $`\delta ^{(n)}(t)`$ such that, indeed, $`u(t)\delta ^{(n)}(t)`$ can be renamed $`\delta ^{(n)}(t)`$. If more mathematical rigour is desired, the theory of distributions may be invoked.
In Sect. 2.4 the identities (111) and (112) are subsumed by the unit-step redundancy theorem (24).
### A.4 The dud-theorem
The dud-theorem depicts the $`n`$-th derivative of the ud-function $`f_{\mathrm{ud}}(t)`$ such as defined by (46), as follows.
With regard to the definition (14) of the unit step function the first derivative can be expressed by
$`f_{\mathrm{ud}}^{}(t)=[u(t)f_\mathrm{d}(t)]^{}`$ $`=`$ $`0\text{ for }t<0,`$ (114)
$`=`$ $`f_\mathrm{d}(0)u_0^{}(t)\text{ for }t=0,`$
$`=`$ $`f_\mathrm{d}^{}(t)\text{ for }t>0.`$
To obtain the derivatives of higher order, it is helpful that (114) can be converted into a more convenient form, as follows. Utilizing the definition (14) of $`u(t)`$, one can replace the third line on the right side of (114) with the expression
$$u(t)f_\mathrm{d}^{}(t)f_\mathrm{d}^{}(0)u_0(t)\text{ for }\mathrm{}<t.$$
(115)
This converts (114) into a superposition of causal functions which are known to be LT-consistent. As these terms are LT-consistent they can be replaced with the pertinent inverse L-transforms without affecting the validity of $`f_{\mathrm{ud}}^{}(t)`$. This has the effect that the term $`f_\mathrm{d}^{}(0)u_0(t)`$ gets eliminated from (115). As a result, $`f_{\mathrm{ud}}^{}(t)`$ can be expressed in the form
$$f_{\mathrm{ud}}^{}(t)=u(t)f_\mathrm{d}^{}(t)+f_\mathrm{d}(0)u_0^{}(t)\text{ for }\mathrm{}<t.$$
(116)
By comparison of (116) to (46) the scheme becomes apparent according to which the second derivative emerges from the first, the third from the second, and so on. For instance, when (116) is derived to obtain the second derivative of $`f_{\mathrm{ud}}(t)`$, the term $`u(t)f_\mathrm{d}^{}(t)`$ gets converted into $`u(t)f_\mathrm{d}^{\prime \prime }(t)+f_\mathrm{d}^{}(0)u_0^{}(t)`$; and the second term $`f_\mathrm{d}(0)u_0^{}(t)`$ gets converted into $`f_\mathrm{d}(0)u_0^{\prime \prime }(t)`$. By this scheme one eventually obtains for the $`n`$-th derivative the expression
$$f_{\mathrm{ud}}^{(n)}(t)=u(t)f_\mathrm{d}^{(n)}(t)+f_\mathrm{d}^{(n1)}(0)u_0^{}(t)+f_\mathrm{d}^{(n2)}(0)u_0^{\prime \prime }(t)+\mathrm{}+f_\mathrm{d}(0)u_0^{(n)}(t);\text{ }n=1,2,\mathrm{}$$
(117)
Equation (117) is by (47) expressed as the dud-theorem.
### A.5 Derivatives of order 1/2
From the derivatives of non-integer order of causal functions those of order $`1/2`$ are of particular interest . Below, a number of derivatives are listed which were determined by the gdi-theorem (70), using the table of LT-correspondences Sect. A.7.
$$D^{1/2}\{u(t)\}=L^1\left\{\sqrt{s}\frac{1}{s}\right\}=u(t)\frac{1}{\sqrt{\pi t}}$$
(118)
$$D^{1/2}\left\{\frac{u(t)}{\sqrt{t}}\right\}=L^1\left\{\sqrt{s}\sqrt{\pi /s}\right\}=\delta (t)\sqrt{\pi }$$
(119)
$$D^{1/2}\{u(t)\sqrt{t}\}=L^1\left\{\sqrt{s}\frac{\sqrt{\pi }}{2s\sqrt{s}}\right\}=u(t)\frac{\sqrt{\pi }}{2}$$
(120)
$$D^{1/2}\{u(t)\mathrm{e}^t\}=L^1\left\{\sqrt{s}\frac{1}{s1}\right\}=u(t)[1/\sqrt{\pi t}+\mathrm{e}^t\mathrm{erf}(\sqrt{t})]$$
(121)
$$D^{1/2}\{u(t)\mathrm{ln}t\}=L^1\left\{\sqrt{s}\frac{\mathrm{ln}sC_\mathrm{E}}{s}\right\}=u(t)\frac{\mathrm{ln}(4t)}{\sqrt{\pi t}}$$
(122)
$$D^{1/2}\left\{\frac{u(t)}{\sqrt{t}}\mathrm{e}^{a^2/(4t)}\right\}=L^1\{\sqrt{\pi }\mathrm{e}^{a\sqrt{s}}\}=u(t)\frac{a}{2}t^{3/2}\mathrm{e}^{a^2/(4t)}$$
(123)
$$D^{1/2}\{u(t)J_0(2\sqrt{at})\}=L^1\left\{\frac{1}{\sqrt{s}}\mathrm{e}^{a/s}\right\}=u(t)\frac{\mathrm{cos}2\sqrt{at}}{\sqrt{\pi t}}$$
(124)
### A.6 Linear DE that includes a derivative <br>of the excitation function
When the linear first-order DE (91) is modified into the form
$$ay(t)+y^{}(t)=x^{}(t),$$
(125)
the response $`y(t)`$ denotes the electrical current through the capacitor of the electrical circuit shown in Fig. 3, while $`x(t)`$ denotes the current exerted on the circuit by the source Q. The constant $`a`$ equals $`a=1/(RC)`$.
The DE (125) is of the type which by Doetsch was banished from his theory because it includes a derivative of the excitation function. In terms of (71) this DE corresponds to $`N=1`$, $`M=1`$.
The new method, i.e., the formula (86), provides for an algorithmic, straightforward solution. Assuming, as an example, $`x(t)=\widehat{x}u(t)`$, and taking into account the parameters $`N=M=1`$; $`a_0=a`$; $`a_1=1`$; $`b_0=0`$; $`b_1=1`$, one obtains
$`y(t)`$ $`=`$ $`L^1\left\{{\displaystyle \frac{\widehat{x}}{s+a}}\right\}+L_\mathrm{d}^1\left\{{\displaystyle \frac{y_\mathrm{s}(0)}{s+a}}\right\}`$ (126)
$`=`$ $`[\widehat{x}u(t)+y_\mathrm{s}(0)]\mathrm{e}^{at}\text{ for }\mathrm{}<t.`$
Any particular initial state can be freely accounted for by setting $`y_\mathrm{s}(0)`$ accordingly. The two components of the total solution are illustrated in Fig. 4.
### A.7 LT-correspondences
Below, LT-correspondences are listed that are used in the present article. The $`t`$-domain functions are identical to those included in customary tables of TLT, except for the factor $`u(t)`$, which is required to make the correspondences valid in both directions. Notice that the first correspondence (127) holds only in one direction.
$$11/s$$
(127)
$$u(t)1/s$$
(128)
$$\delta ^{(n)}(t)=u^{(n+1)}(t)=u_0^{(n+1)}(t)s^n\text{ }(n=0,1,\mathrm{})$$
(129)
$$u(t)t^n\frac{n!}{s^{n+1}}\text{ }(n=0,1,\mathrm{})$$
(130)
$$u(t)\mathrm{e}^{at}\frac{1}{s+a}$$
(131)
$$\frac{u(t)}{a}(1\mathrm{e}^{at})\frac{1}{s(s+a)}$$
(132)
$$u(t)\mathrm{sin}\omega t\frac{\omega }{s^2+\omega ^2}$$
(133)
$$u(t)\mathrm{cos}\omega t\frac{s}{s^2+\omega ^2}$$
(134)
$$\frac{u(t)}{a^2+\omega ^2}(\omega \mathrm{e}^{at}+a\mathrm{sin}\omega t\omega \mathrm{cos}\omega t)\frac{\omega }{(s+a)(s^2+\omega ^2)}$$
(135)
$$\frac{u(t)}{\sqrt{t}}\sqrt{\frac{\pi }{s}}$$
(136)
$$u(t)\sqrt{t}\frac{\sqrt{\pi /s}}{2s}$$
(137)
$$u(t)\mathrm{e}^t\mathrm{erf}(\sqrt{t})\frac{1}{(s1)\sqrt{s}}$$
(138)
$$u(t)\mathrm{ln}t\frac{\mathrm{ln}sC_\mathrm{E}}{s}\text{ }(C_\mathrm{E}=0.577215\mathrm{})$$
(139)
$$u(t)\frac{\mathrm{ln}t}{\sqrt{t}}\sqrt{\frac{\pi }{s}}(\mathrm{ln}4s+C_\mathrm{E})$$
(140)
$$u(t)t^r\frac{\mathrm{\Gamma }(r+1)}{s^{r+1}}\text{ }(r\mathrm{};\text{ }r>1)$$
(141)
$$\frac{u(t)}{\sqrt{\pi t}}\mathrm{e}^{a^2/(4t)}\frac{1}{\sqrt{s}}\mathrm{e}^{a\sqrt{s}}$$
(142)
$$u(t)at^{3/2}\mathrm{e}^{a^2/(4t)}2\sqrt{\pi }\mathrm{e}^{a\sqrt{s}}$$
(143)
$$\frac{u(t)}{\sqrt{\pi t}}\mathrm{cos}2\sqrt{at}\frac{1}{\sqrt{s}}\mathrm{e}^{a/s}$$
(144)
$$u(t)J_0(2\sqrt{at})\frac{1}{s}\mathrm{e}^{a/s}$$
(145) |
warning/0507/math-ph0507001.html | ar5iv | text | # On the Hamiltonian formulation of Yang–Mills gauge theories
## 1 Introduction
In some of our recent works a new geometrical framework for Yang–Mills field theories and General Relativity in the tetrad–affine formulation has been developed.
The construction of the new geometrical setting started from the observation that even though the Lagrangian densities of the above theories are defined over the first jet–bundle of the configuration space, they only depend of the antisymmetric combination of field derivatives in the space–time indexes. As for the Yang–Mills case, this is the reason for the singularity in the Lagrangian.
The idea consists is considering a suitable quotient of the first jet–bundle, making two sections equivalent when they possess a first order contact with respect to the exterior covariant differentiation, instead of the whole set of derivatives. The fiber coordinates of the resulting quotient bundle are the antisymmetric combinations of the field derivatives that appear in the Lagrangian.
The geometry of the new space has been widely studied, in order to build as many usual geometric structures of the jet–bundle theory as possible, such as contact forms, jet–prolongations of sections, morphisms and vector fields. These are the geometric tools that are needed to implement variational problems in the Poincaré–Cartan formalism. Moreover, particular choices for the fiber coordinates have been shown to be possible: they consist in the components of the strength tensor for Yang–Mills theories and in the torsion and curvature tensors for General Relativity.
This resulted into the elimination of some un–physical degrees of freedom from the theory (represented by un-necessary jet–coordinates) and to even obtain a regular Lagrangian theory in the case of Yang–Mills fields.
This last consideration was the thrust that moved us to write the present work: the presence of a regular Lagrangian allows us to write a Hamiltonian version of the theory proposed in . The advantages arising from the present approach, with respect to the already existing formulations, based on singular Lagrangians (compare, for example, with ), are striking. In fact, the singularity of the Lagrangian is the source of known drawbacks: the equations are defined on a constraint sub–manifold, there exist multiple Hamitonian forms associated with the same Lagrangian and the equivalence between Euler–Lagrange and Hamilton equations is not a direct consequence of Legendre transform any more.
On the contrary, the situation in the new geometrical framework is simpler and more elegant: the “Lagrangian” space and the phase–space have the same dimension and the Legendre transform is a (local) diffeomorphism. This ensures the direct equivalence of Lagrangian and Hamiltonian formulations, both in the Hamilton–De Donder (section 3) and in the Multimomentum Hamiltonian approach (section 4).
Finally, we devoted the last section to study the peculiar $`(3+3)`$ Yang–Mills theory. Starting from the work made in , we showed that a coordinate transformation in the phase–space, together with the Poincaré–Cartan approach in our new formalism, allows to describe a (3+3) Yang–Mills theory by means of Einstein–Cartan like equations in 3 dimensions. In particular, in the case of a free $`(3+3)`$ Yang–Mills field the geometrical construction gives rise to a sort of first–order purely frame–formulation of a General Relativity like theory.
This result is interesting for its further developments: in fact we will show in a subsequent work that an analogous geometrical machinery may be applied to build a first–order purely frame–formulation of General Relativity in four dimensions .
## 2 The geometrical framework
The present section is devoted to revising the geometrical structure that has been introduced in to describe Yang–Mills theories.
Let $`\pi :PM`$ be a principal fiber bundle, with structural group $`G`$ and let $`x^i,g^\mu `$ denote a system of local fibered coordinates on $`P`$. $`J_1(P)`$ denotes the first jet-bundle of $`\pi :PM`$ and it is referred to local coordinates $`x^i,g^\mu ,g_i^\mu \left(\frac{g^\mu }{x^i}\right)`$.
The space of principal connections on $`P`$ is identified with the quotient bundle $`E:=J_1(P)/G`$ with respect to the (jet-prolongation of) the right action $`R_h`$ of the structural group on $`P`$. If $`V_\nu ^\mu (g,h)`$ represents the differential of the right multiplication $`R_h`$ in $`gG`$, a set of local coordinates in the quotient space is provided by $`x^\mu ,a_i^\mu =g_i^\nu V_\nu ^\mu (g,g^1)`$, subject to the following transformation laws:
$$\overline{x}^i=\overline{x}^i(x^j),\overline{a}_i^\mu =\left[Ad(\gamma ^1)_\nu ^\mu a_j^\nu +W_\nu ^\mu (\gamma ^1,\gamma )\frac{\gamma ^\nu }{x^j}\right]\frac{x^j}{\overline{x}^i}$$
(2.1)
where $`Ad_\nu ^\mu `$ and $`W_\nu ^\mu `$ denote respectively the adjoint representation of $`G`$ and the differential of the left multiplication in $`G`$, while $`\gamma :UMG`$ ($`U`$ open set) is an arbitrary smooth map.
As a consequence, the bundle $`EM`$ has the nature of an affine bundle, whose sections represent principal connections over $`PM`$. In fact, every section $`\omega :MJ_1(P)/G`$ yields a connection $`1`$-form on $`P`$, locally described as:
$$\omega (x,g)=\omega ^\mu (x,g)\underset{¯}{e}_\mu :=\left[Ad(g^1)_\nu ^\mu a_i^\nu (x)dx^i+W_\nu ^\mu (g^1,g)dg^\nu \right]\underset{¯}{e}_\mu $$
(2.2)
where $`\underset{¯}{e}_\mu `$ ($`\mu =1,\mathrm{},r`$) indicate a basis of the Lie algebra g of $`G`$.
Finally, let $`\widehat{\pi }:J_1(E)E`$ be the first jet–bundle associated with the bundle $`EM`$, described by the set of local coordinates $`x^i,a_i^\mu ,a_{ij}^\mu \left(\frac{a_i^\mu }{x^j}\right)`$.
In order to provide a better geometrical framework to describe Yang-Mills gauge theories, the following equivalence relation is introduced in $`J_1(E)`$: let $`\omega _1=(x^\mu ,a_i^\mu ,a_{ij}^\mu ),\omega _2=(x^\mu ,a_i^\mu ,\widehat{a}_{ij}^\mu )J_1(E)`$ be such that $`\widehat{\pi }(\omega _1)=\widehat{\pi }(\omega _2)`$, then:
$$\omega _1\omega _2(a_{ij}^\mu a_{ji}^\mu )=(\widehat{a}_{ij}^\mu \widehat{a}_{ji}^\mu )$$
(2.3)
This means that two sections are declared equivalent if their skew–symmetric derivatives are equal. In more geometric terms, being every section of the bundle $`EM`$ represented by a connection 1–form, the first jet–bundle has been constructed assuming that the equivalence between sections having a first–order contact is evaluated through the exterior differentiation (or, equivalently, the covariant exterior differentiation), instead of the whole set of partial derivatives.
Let $`𝒥(E):=J_1(E)/`$ denote the quotient bundle with respect to the above defined equivalence relation and $`\rho :J_1(E)𝒥(E)`$ the canonical (quotient) projection. The bundle $`𝒥(E)`$ is endowed with a set of local fibered coordinates $`x^i,a_i^\mu ,A_{ij}^\mu :=\frac{1}{2}\left(a_{ij}^\mu a_{ji}^\mu \right)(i<j)`$, subject to the following transformation laws:
$$\overline{A}_{ik}^\mu =\frac{x^j}{\overline{x}^i}\frac{x^h}{\overline{x}^k}\left[Ad(\gamma ^1)_\nu ^\mu A_{jh}^\nu +\frac{1}{2}\left(\frac{Ad(\gamma ^1)_\nu ^\mu }{x^h}a_j^\nu \frac{Ad(\gamma ^1)_\nu ^\mu }{x^j}a_h^\nu \right)+\frac{1}{2}\left(\frac{\eta _j^\mu }{x^h}\frac{\eta _h^\mu }{x^j}\right)\right]$$
(2.4)
where $`\eta _j^\mu (x):=W_\nu ^\mu (\gamma ^1(x),\gamma (x))\frac{\gamma ^\nu (x)}{x^j}`$.
This newly defined geometrical framework is endowed with the most common features provided by a standard jet–bundle structure.
$``$ $`𝒥`$-extension of sections. Given a section $`\sigma :ME`$ its $`𝒥`$-extension is defined as $`𝒥\sigma :=\rho J_1\sigma :M𝒥(E)`$, namely projecting the standard jet–prolongation to $`𝒥(E)`$ by means of the quotient map. Conversely, every section $`s:M𝒥(E)`$ will be said to be holonomic if there exists a section $`\sigma :ME`$ such that $`s=𝒥\sigma `$.
$``$ Contact forms. Let us define the following $`2`$-forms on $`𝒥(E)`$:
$$\theta ^\mu :=da_i^\mu dx^i+A_{ij}^\mu dx^idx^j$$
(2.5)
They undergo the transformation laws $`\overline{\theta }^\mu =Ad(\gamma ^1)_\nu ^\mu \theta ^\nu `$. The vector bundle which is locally spanned by the $`2`$-forms (2.5) will be called the contact bundle $`𝒞(𝒥(E))`$ and any section $`\eta :𝒥(E)𝒞(𝒥(E))`$ will be called a contact $`2`$-form. Contact forms are such that $`s^{}(\eta )=0`$ whenever $`s:M𝒥(E)`$ is holonomic.
$``$ $`𝒥`$-prolongations of morphisms and vector fields. A generic morphism $`\mathrm{\Phi }:EE`$, fibered over $`M`$, can be raised to a morphism $`𝒥\mathrm{\Phi }:𝒥(E)𝒥(E)`$ considering its ordinary jet–prolongation and restricting it to $`𝒥(E)`$ through the quotient map, namely:
$$𝒥\mathrm{\Phi }(z):=\rho j_1\mathrm{\Phi }(w)w\rho ^1(z),z𝒥(E)$$
As a matter of fact, not every morphism $`\mathrm{\Phi }:EE`$ commutes with the quotient map and produces a well defined prolongation (i.e. independent of the choice of the representative in the equivalence class), but it has to satisfy the condition:
$$\rho j_1\mathrm{\Phi }(w_1)=\rho j_1\mathrm{\Phi }(w_2)w_1,w_2\rho ^1(z)$$
(2.6)
Referring to for the proof, it is easy to see that the only morphisms satisfying condition (2.5) are necessarily of the form:
$$\{\begin{array}{c}y^i=\chi ^i(x^j)\hfill \\ \\ b_i^\nu =\mathrm{\Phi }_i^\nu (x^j,a_j^\mu )=\mathrm{\Gamma }_\mu ^\nu (x)\frac{x^r}{y^i}a_r^\mu +f_i^\nu (x)\hfill \end{array}$$
(2.7)
where $`\mathrm{\Gamma }_\mu ^\nu (x)`$ and $`f_i^\nu (x)`$ are arbitrary local functions on $`M`$. Their $`𝒥`$prolongation is:
$$\{\begin{array}{c}y^i=\chi ^i(x^k)\hfill \\ \\ b_i^\nu =\mathrm{\Gamma }_\mu ^\nu (x)\frac{x^r}{y^i}a_r^\mu +f_i^\nu (x)\hfill \\ \\ B_{ij}^\nu =\mathrm{\Gamma }_\mu ^\nu A_{ks}^\mu \frac{x^k}{y^i}\frac{x^s}{y^j}+\frac{1}{2}\left[\frac{\mathrm{\Gamma }_\mu ^\nu }{x^k}\left(\frac{x^k}{y^j}\frac{x^r}{y^i}\frac{x^k}{y^i}\frac{x^r}{y^j}\right)a_r^\mu +\frac{f_i^\nu }{x^k}\frac{x^k}{y^j}\frac{f_j^\nu }{x^k}\frac{x^k}{y^i}\right]\hfill \end{array}$$
In a similar way (compare with ), it is easy to prove that the only vector fields of the form
$$X=ϵ^i(x^j)\frac{}{x^i}+\left(\frac{ϵ^k}{x^q}a_k^\mu +D_\nu ^\mu (x^j)a_q^\nu +G_q^\mu (x^j)\right)\frac{}{a_q^\mu }$$
(2.8)
(where $`ϵ^i(x)`$, $`D_\nu ^\mu (x)`$ and $`G_q^\mu (x)`$ are arbitrary local functions on $`M`$) can be $`𝒥`$prolonged to vector fields over $`𝒥(E)`$ as follows:
$$𝒥(X)(z):=\rho _{\rho ^1(z)}(j_1(X))z𝒥(P)$$
(2.9)
The resulting vector field has the form:
$$𝒥(X)=ϵ^i(x^j)\frac{}{x^i}+\left(\frac{ϵ^k}{x^q}a_k^\mu +D_\nu ^\mu (x^j)a_q^\nu +G_q^\mu (x^j)\right)\frac{}{a_q^\mu }+\underset{i<j}{}h_{ij}^\mu \frac{}{A_{ij}^\mu }$$
where
$$h_{ij}^\mu =\frac{1}{2}\left(\frac{D_\nu ^\mu }{x^j}a_i^\nu \frac{D_\nu ^\mu }{x^i}a_j^\nu +\frac{G_i^\mu }{x^j}\frac{G_j^\mu }{x^i}\right)+D_\nu ^\mu A_{ij}^\nu +\left(A_{ki}^\mu \frac{ϵ^k}{x^j}A_{kj}^\mu \frac{ϵ^k}{x^i}\right)$$
Finally, in order to adapt the geometrical framework to the presence of the covariant differentiation induced by connections, it is useful to introduce a set of new fibered local coordinates over $`𝒥(E)`$ of the form:
$$x^i=x^ia_i^\mu =a_i^\mu F_{ij}^\mu =2A_{ji}^\mu +a_j^\nu a_i^\rho C_{\rho \nu }^\mu $$
(2.10)
where $`C_{\rho \nu }^\mu `$ are the structure coefficients of the group $`G`$. The latter are subject to the following transformations laws:
$$\overline{F}_{ik}^\mu =\frac{x^j}{\overline{x}^i}\frac{x^h}{\overline{x}^k}Ad(\gamma ^1)_\nu ^\mu F_{jh}^\nu $$
(2.11)
Using the new coordinates, every Yang–Mills Lagrangian $`m`$–form can be expressed as
$$L=(x^i,a_i^\mu ,F_{ij}^\mu )ds$$
(2.12)
Moreover, it is possible to define a corresponding Poincaré–Cartan $`m`$-form over $`𝒥(E)`$, expressed as
$$\mathrm{\Theta }_L:=ds\frac{1}{2}\theta ^\mu P_\mu $$
(2.13)
where $`P_\mu :=\frac{}{F_{ij}^\mu }ds_{ij}`$, $`ds_{ij}:=\frac{}{x^i}\text{ }\text{ }\frac{}{x^j}\text{ }\text{ }ds`$.
The presence of the Poincarè–Cartan form allows to deduce the evolutions equations for Yang–Mills fields looking for the stationary points of the functional
$$A_L(\gamma ):=_D\gamma ^{}(\mathrm{\Theta }_L)\gamma :DM𝒥(E)$$
(2.14)
The stationarity condition for $`A_L`$ (taking null variations at the boundary of the compact domain $`D`$) is equivalent to the conditions (compare with ):
$$\gamma ^{}(\theta ^\mu )=0$$
(2.15a)
$$\gamma ^{}\left(\frac{}{a_i^\mu }D_j\frac{}{F_{ji}^\mu }\right)=0$$
(2.15b)
The first equation ensures the kinematic admissibility of the critical section $`\gamma `$, while the second represents the field equations of the problem. As a matter of fact, the kinematical admissibility is directly obtained from the variational principle and is not imposed as an a-priori condition, as a consequence of the regularity of the Lagrangian $``$ within the new framework provided by $`𝒥(E)`$.
## 3 The Hamiltonian framework
Let $`\mathrm{\Lambda }^m(E)`$ denote the modulus of $`m`$-forms over $`E`$, and let $`\mathrm{\Lambda }_r^m(E)\mathrm{\Lambda }^m(E)`$ ($`r<m`$) be the sub-bundle consisting of those $`m`$-forms on $`E`$ vanishing when $`r`$ of its given arguments are vertical vectors over the bundle $`EM`$. It is obvious that the above defined bundles form a chain of vector bundles over $`E`$ such that:
$$0\mathrm{\Lambda }_1^m(E)\mathrm{\Lambda }_2^m(E)\mathrm{}\mathrm{\Lambda }_r^m(E)\mathrm{}\mathrm{\Lambda }^m(E)$$
In particular the attention will be focussed on the first two sub-spaces. Given a system of local coordinates over $`E`$ and let $`ds=dx^1\mathrm{}dx^m`$, they can be respectively described as:
$$\mathrm{\Lambda }_1^m(E):=\{\omega \mathrm{\Lambda }^m(E):\omega =pds\}$$
(3.1)
and
$$\mathrm{\Lambda }_2^m(E):=\{\omega \mathrm{\Lambda }^m(E):\omega =pds+\mathrm{\Pi }_\mu ^{ji}da_i^\mu ds_j\}$$
(3.2)
where $`ds_j:=\frac{}{x^j}\text{ }\text{ }ds`$. It is then possible to assume $`\{x^i,a_i^\mu ,p\}`$ as a system of local coordinates on $`\mathrm{\Lambda }_1^m(E)`$, subject to the transformations laws $`p=J\overline{p}`$ (where $`J:=\mathrm{det}\frac{\overline{x}^i}{x^k}`$).
A set of local coordinates for $`\mathrm{\Lambda }_2^m(E)`$ is provided by the functions $`\{x^i,a_i^\mu ,p,\mathrm{\Pi }_\mu ^{ij}\}`$. The latter are subject to a set of transformation laws described by eqs. (2.1), together with:
$$p=J\left(\overline{p}+\overline{\mathrm{\Pi }}_\mu ^{ji}\left(\frac{Ad(\gamma ^1)_\nu ^\mu }{x^q}a_p^\nu +\frac{\eta _p^\mu }{x^q}\right)\frac{x^q}{\overline{x}^j}\frac{x^p}{\overline{x}^i}+\overline{\mathrm{\Pi }}_\mu ^{ij}(Ad(\gamma ^1)_\nu ^\mu a_p^\nu +\eta _p^\mu )\frac{^2x^p}{\overline{x}^j\overline{x}^i}\right)$$
(3.3a)
$$\mathrm{\Pi }_\nu ^{pq}=\overline{\mathrm{\Pi }}_\mu ^{ij}Ad(\gamma ^1)_\nu ^\mu \frac{x^q}{\overline{x}^j}\frac{x^p}{\overline{x}^i}J$$
(3.3b)
The bundle $`\mathrm{\Lambda }_2^m(E)`$ is endowed with the canonical Liouville $`m`$-form, locally expressed as
$$\mathrm{\Theta }:=pds+\mathrm{\Pi }_\mu ^{ji}da_i^\mu ds_j$$
(3.4)
whose differential
$$\mathrm{\Omega }:=d\mathrm{\Theta }=dpds+d\mathrm{\Pi }_\mu ^{ji}da_i^\mu ds_j$$
(3.5)
is a multisymplectic $`(m+1)`$form over $`\mathrm{\Lambda }_2^m(E)`$.
A deeper geometrical insight in the problem can be given observing that eqs. (3.3) make $`\mathrm{\Lambda }_1^m`$ into a vector sub-bundle of $`\mathrm{\Lambda }_2^m`$, thus allowing us to introduce the quotient bundle $`\mathrm{\Lambda }_2^m/\mathrm{\Lambda }_1^m`$. As a consequence of the definition, the latter has the nature of a vector bundle over $`E`$ and is locally described by the system of coordinates $`x^i,a_i^\mu ,\mathrm{\Pi }_\mu ^{ij}`$. It is worth noticing that the transformation law (3.3a) makes $`\pi :\mathrm{\Lambda }_2^m(E)\mathrm{\Lambda }_2^m/\mathrm{\Lambda }_1^m`$ into an affine bundle.
The phase space is defined as the vector sub–bundle $`\mathrm{\Pi }(E)\mathrm{\Lambda }_2^m(E)/\mathrm{\Lambda }_1^m(E)`$ consisting of those elements $`z\mathrm{\Lambda }_2^m(E)/\mathrm{\Lambda }_1^m(E)`$ satisfying the requirement
$$\mathrm{\Pi }_\mu ^{ij}(z)=\mathrm{\Pi }_\mu ^{ji}(z)$$
(3.6)
Condition (3.6) is well–posed because of the transformation laws (3.3). A local system of coordinates for $`\mathrm{\Pi }(E)`$ is provided by $`x^i,a_i^\mu ,\mathrm{\Pi }_\mu ^{ij}(i<j)`$, subject to the same transformation laws (3.3b). Besides, being $`\mathrm{\Pi }(E)`$ a vector sub–bundle, the immersion $`i:\mathrm{\Pi }(E)\mathrm{\Lambda }_2^m(E)/\mathrm{\Lambda }_1^m(E)`$ is well defined and is locally represented by eq. (3.6) itself.
The pull–back bundle $`\widehat{\pi }:(E)\mathrm{\Pi }(E)`$ defined by the following commutative diagram
$$\begin{array}{ccc}(E)& \stackrel{\widehat{i}}{}& \mathrm{\Lambda }_2^m(E)\\ \widehat{\pi }& & \pi & & \\ \mathrm{\Pi }(E)& \stackrel{i}{}& \mathrm{\Lambda }_2^m(E)/\mathrm{\Lambda }_1^m(E)\end{array}$$
(3.7)
will now be taken into account. A local coordinate system for $`(E)`$ is provided by $`x^i,a_i^\mu ,\mathrm{\Pi }_\mu ^{ij}(i<j),p`$, subject to transformation laws (3.3b), together with:
$$p=J\left(\overline{p}+\overline{\mathrm{\Pi }}_\mu ^{ji}\left(\frac{Ad(\gamma ^1)_\nu ^\mu }{x^q}a_p^\nu +\frac{\eta _p^\mu }{x^q}\right)\frac{x^q}{\overline{x}^j}\frac{x^p}{\overline{x}^i}\right)$$
(3.8)
the latter being the antisymmetric part of eq. (3.3a). The above transformation law shows that the bundle $`\widehat{\pi }:(E)\mathrm{\Pi }(E)`$ has the nature of an affine bundle over the phase space. Every section $`h:\mathrm{\Pi }(E)(E)`$ will be called a Hamiltonian section, and will be locally described in the form:
$$h:p=(x^i,a_i^\mu ,\mathrm{\Pi }_\mu ^{ij})$$
(3.9)
The presence of the immersion $`\widehat{i}:(E)\mathrm{\Lambda }_2^m(E)`$, endows $`(E)`$ with the canonical $`m`$-form $`\widehat{i}^{}(\mathrm{\Theta })`$, locally expressed as in eq. (3.4). The latter will be simply denoted as $`\mathrm{\Theta }`$ and will be called the Liouville form on $`(E)`$.
The presence of the $`m`$form $`\mathrm{\Theta }`$ on $`(E)`$, allows to create a correspondence between the Hamiltonian and the Lagrangian viewpoints, based on the existence of a unique diffeomorphism $`\lambda :𝒥(E)(E)`$ fibered over $`E`$ satisfying the requirement:
$$\mathrm{\Theta }_L=\lambda ^{}(\mathrm{\Theta })$$
(3.10)
Such a diffeomorphism will be called the Legendre map. Given a set of local coordinates $`x^i,a_i^\mu ,F_{ij}^\mu (i<j)`$ on $`𝒥(E)`$ and $`x^i,a_i^\mu ,\mathrm{\Pi }_\mu ^{ij}(i<j),p`$ on $`(E)`$, and taking eqs. (2.10) and (2.13) into account, the Poincaré–Cartan m-form can written as
$$\mathrm{\Theta }_L=\left(L\frac{1}{2}\left(F_{kr}^\mu +a_k^\nu a_r^\rho C_{\rho \nu }^\mu \right)\frac{L}{F_{kr}^\mu }\right)ds+\frac{L}{F_{rk}^\mu }da_k^\mu ds_r$$
(3.11)
and the Legendre map defined by eq. (3.10) is such that:
$$\lambda :\{\begin{array}{c}x^i=x^i\hfill \\ \\ a_i^\mu =a_i^\mu \hfill \\ \\ p(x^j,a_j^\alpha ,F_{ij}^\alpha )=L\frac{1}{2}\left(F_{kr}^\mu +a_k^\nu a_r^\rho C_{\rho \nu }^\mu \right)\frac{L}{F_{kr}^\mu }\hfill \\ \\ \mathrm{\Pi }_\mu ^{ij}(x^j,a_j^\alpha ,F_{ij}^\alpha )=\frac{L}{F_{ij}^\mu }\hfill \end{array}$$
(3.12)
The most striking feature of the Legendre transformation between $`𝒥(E)`$ and $`(E)`$ is provided by its regularity, due to the acquired regularity of the Yang–Mills Lagrangian in the space $`𝒥(E)`$. In particular the condition
$$\mathrm{det}\left(\frac{\mathrm{\Pi }_\mu ^{ij}}{F_{rk}^\alpha }\right)0i<j,r<k\alpha ,\mu $$
assures the local invertibility of the last equation (3.12), allowing to obtain the coordinates $`F_{ij}^\alpha `$ as functions $`F_{ij}^\alpha =F_{ij}^\alpha (x^j,a_j^\alpha ,\mathrm{\Pi }_\mu ^{ij})`$. Thus, the Legendre map has the nature of a regular immersion of $`𝒥(E)`$ into $`(E)`$, yielding a submanifold $`\lambda (𝒥(E))(E)`$, locally described by:
$$p(x^j,a_j^\alpha ,\mathrm{\Pi }_\mu ^{ij})=L(x^j,a_j^\alpha ,\mathrm{\Pi }_\mu ^{ij})\frac{1}{2}\left(F_{kr}^\mu (x^j,a_j^\alpha ,\mathrm{\Pi }_\mu ^{ij})+a_k^\nu a_r^\rho C_{\rho \nu }^\mu \right)\mathrm{\Pi }_\mu ^{kr}$$
(3.13)
In accordance with the literature, the function
$$H(x^i,a_i^\mu ,\mathrm{\Pi }_{ij}^\mu )=L(x^i,a_i^\mu ,\mathrm{\Pi }_{ij}^\mu )+\frac{1}{2}F_{kr}^\mu (x^i,a_i^\mu ,\mathrm{\Pi }_\mu ^{ij})\mathrm{\Pi }_\mu ^{kr}$$
(3.14)
will be called the Hamiltonian of the system.
If the phase space $`\mathrm{\Pi }(E)`$ is taken into account, the composition $`\widehat{\lambda }:=\widehat{\pi }\lambda :𝒥(E)\mathrm{\Pi }(E)`$ results to be a (local) diffeomorphism. As a consequence, its (local) inverse map $`\widehat{\lambda }^1:\mathrm{\Pi }(E)𝒥(E)`$ can be considered. Taking the derivatives of eq. (3.14) with respect to $`\mathrm{\Pi }_\mu ^{ij}`$ and using the antisymmetric properties of the coordinates, one gets the coordinate representation for the inverse Legendre map as:
$$\widehat{\lambda }^1:\{\begin{array}{c}x^i=x^i\hfill \\ \\ a_i^\mu =a_i^\mu \hfill \\ \\ F_{ij}^\mu =\frac{H}{\mathrm{\Pi }_\mu ^{ij}}\hfill \end{array}$$
(3.15)
Taking the Legendre map into account, as well as its inverse (3.15), it is easy to see that the image $`\lambda (𝒥(E))`$ defined by eq. (3.13) yields a Hamiltonian section $`h`$, represented by a function $`(x^i,a_i^\mu ,\mathrm{\Pi }_\mu ^{ij})=H(x^i,a_i^\mu ,\mathrm{\Pi }_\mu ^{ij})+\frac{1}{2}a_k^\nu a_r^\rho C_{\rho \nu }^\mu \mathrm{\Pi }_\mu ^{kr}`$.
Now, the presence of the Hamiltonian section allows to perform the pull-back of the Liouville form on $`(E)`$ to the phase space $`\mathrm{\Pi }(E)`$. The result is a Hamiltonian dependent $`m`$form
$$\mathrm{\Theta }_h:=h^{}(\mathrm{\Theta })=H(x^i,a_i^\mu ,\mathrm{\Pi }_\mu ^{ij})ds\mathrm{\Pi }_\mu ^{ij}\left(da_i^\mu ds_j+\frac{1}{2}a_i^\nu a_j^\rho C_{\rho \nu }^\mu ds\right)$$
(3.16)
The variational principle constructed on the phase space $`\mathrm{\Pi }(E)`$ with the $`m`$-form $`\mathrm{\Theta }_h`$ yields the Hamilton equations for the problem. In fact, the solution of the variational problem for the functional
$$A_h(\gamma )=_D\gamma ^{}(\mathrm{\Theta }_h)\mathrm{section}\gamma :\mathrm{D}\mathrm{M}(\mathrm{E})$$
is made of its Euler–Lagrange equations
$$\gamma ^{}(X\text{ }\text{ }d\mathrm{\Theta }_h)=0XV(\mathrm{\Pi }(E),M)$$
(3.17)
where $`V(\mathrm{\Pi }(E),M)`$ denotes the bundle of vectors over $`\mathrm{\Pi }(E)`$ that are vertical with respect to the fibration over $`M`$.
A straightforward calculation shows that eq. (3.17) splits into the following set of equations:
$$\frac{H}{\mathrm{\Pi }_\mu ^{ij}}\frac{a_i^\mu }{x^j}+\frac{a_j^\mu }{x^i}a_i^\nu a_j^\rho C_{\rho \nu }^\mu =0$$
(3.18a)
$$\frac{H}{a_i^\mu }\frac{\mathrm{\Pi }_\mu ^{ji}}{x^j}+\mathrm{\Pi }_\lambda ^{ji}a_j^\gamma C_{\gamma \mu }^\lambda =0$$
(3.18b)
usually referred to as Hamilton–De Donder equations.
The inverse Legendre map (3.15) shows that eq. (3.18a) is the holonomy requirement for the solution, namely:
$$F_{ij}^\mu =+\frac{a_j^\mu }{x^i}\frac{a_i^\mu }{x^j}a_i^\nu a_j^\rho C_{\rho \nu }^\mu $$
On the other hand, eq. (3.18b) can be written in terms of the covariant derivative $`D_j`$ induced by the connection, giving rise to the usual evolution equations for the Yang–Mills fields:
$$D_j\mathrm{\Pi }_\mu ^{ji}=\frac{H}{a_i^\mu }$$
## 4 Multimomentum Hamiltonian formulation
In the previous section a Hamiltonian approach to Yang–Mills field theories has been developed, adapting the already known Hamilton–De Donder formalism developed within the framework of calculus of variations to the new geometrical setting.
Nevertheless, there exists another well–known Hamiltonian approach to field theory, represented by the so–called multimomentum Hamiltonian formalism, where Hamiltonian connections play the same role as Hamiltonian vector fields in symplectic geometry.
The argument has been widely studied in the literature (compare with ), both on the first jet–bundle and on the Legendre bundle (the phase space), in the Lagrangian and Hamiltonian framework. In this section we will show that a multimomentum formulation of the above theory can be built, starting from the Poincarè–Cartan forms (2.13) and (3.16).
We will start extending some definitions and some results about the Legendre bundle of a generic field theory to our space. All the argument will be presented without proofs; the reader is referred to for comments and further developments.
First of all, the canonical monomorphism is introduced as:
$$\mathrm{\Theta }:\mathrm{\Pi }(E)\mathrm{\Lambda }^{m+1}T^{}(E)_MT(M)$$
$$\mathrm{\Theta }:=\mathrm{\Pi }_\mu ^{ji}da_i^\mu ds\frac{}{x^j}$$
(4.1)
The following definitions are strictly associated with monomorphism (4.1).
###### Definition 4.1
The pull–back valued horizontal form $`\mathrm{\Theta }`$, locally described by eq. (4.1) is called multimomentum Liouville form on the phase space $`\mathrm{\Pi }(E)`$.
###### Definition 4.2
The pull–back valued form, defined as
$$\mathrm{\Omega }:=d\mathrm{\Pi }_\mu ^{ji}da_i^\mu ds\frac{}{x^j}$$
(4.2)
will be called the multisymplectic form on $`\mathrm{\Pi }(E)`$.
The relation between the forms (4.1) and (4.2) is described by the following
###### Proposition 4.1
Given a generic 1–form $`\sigma \mathrm{\Lambda }^1(M)`$, the forms (4.1) and (4.2) are such that
$$\mathrm{\Omega }\text{ }\text{ }\sigma =d(\mathrm{\Theta }\text{ }\text{ }\sigma )$$
(4.3)
Let us consider a connection $`\gamma `$ of the bundle $`\mathrm{\Pi }(E)M`$, locally described by the tangent–valued horizontal 1–form
$$\gamma =\left(\frac{}{x^k}+\mathrm{\Gamma }_{kh}^\mu \frac{}{a_h^\mu }+\frac{1}{2}\mathrm{\Gamma }_{k\mu }^{st}\frac{}{\mathrm{\Pi }_\mu ^{st}}\right)dx^k$$
(4.4)
where $`\mathrm{\Gamma }_{k\mu }^{st}=\mathrm{\Gamma }_{k\mu }^{ts}`$.
Then, the following definition can be given:
###### Definition 4.3
A connection $`\gamma `$ of the bundle $`\mathrm{\Pi }(E)M`$, described by eq. (4.4), is called a Hamiltonian connection if the $`(m+1)`$-form $`\gamma \text{ }\text{ }\mathrm{\Omega }`$ is closed.
A straightforward calculation shows that a connection $`\gamma `$ is Hamiltonian if and only if it satisfies the following conditions:
$$\{\begin{array}{c}\frac{\mathrm{\Gamma }_{j\sigma }^{ji}}{a_p^\lambda }=\frac{\mathrm{\Gamma }_{j\lambda }^{jp}}{a_i^\sigma }=0\hfill \\ \\ \frac{\mathrm{\Gamma }_{j\sigma }^{ji}}{\mathrm{\Pi }_\lambda ^{pq}}+\frac{\mathrm{\Gamma }_{pq}^\lambda }{a_i^\sigma }\frac{\mathrm{\Gamma }_{qp}^\lambda }{a_i^\sigma }=0\hfill \\ \\ \frac{\mathrm{\Gamma }_{ji}^\sigma }{\mathrm{\Pi }_\lambda ^{pq}}\frac{\mathrm{\Gamma }_{pq}^\lambda }{\mathrm{\Pi }_\sigma ^{ji}}\frac{\mathrm{\Gamma }_{ij}^\sigma }{\mathrm{\Pi }_\lambda ^{pq}}\frac{\mathrm{\Gamma }_{qp}^\lambda }{\mathrm{\Pi }_\sigma ^{ji}}=0\hfill \end{array}$$
(4.5)
###### Definition 4.4
An m–form $`\eta \mathrm{\Lambda }^1(\mathrm{\Pi }(E))`$ is called a multimomentum Hamiltonian form if for every open set $`U\mathrm{\Pi }(E)`$ there exists a Hamiltonian connection on $`U`$ satisfying the equation
$$\gamma \text{ }\text{ }\mathrm{\Omega }=d\eta $$
(4.6)
Now, we will show that the Poincarè–Cartan form (3.16) is a multimomentum Hamiltonian form. In other words, we will show the existence of Hamiltonian connections $`\gamma `$ satisfying the equation
$$\gamma \text{ }\text{ }\mathrm{\Omega }=d\mathrm{\Theta }_h$$
(4.7)
Moreover such connections will be shown to automatically satisfy the Hamilton–De Donder equations (3.18).
As a matter of fact, given a connection $`\gamma `$ in the form (4.4), we have that
$$\gamma \text{ }\text{ }\mathrm{\Omega }=\mathrm{\Gamma }_{j\sigma }^{ji}da_i^\sigma ds\mathrm{\Gamma }_{ji}^\sigma d\mathrm{\Pi }_\sigma ^{ji}ds+d\mathrm{\Pi }_\sigma ^{ji}da_i^\sigma ds_j$$
(4.8)
Nevertheless, one also as
$$\begin{array}{c}\hfill d\mathrm{\Theta }_h=\frac{H}{a_i^\sigma }da_i^\sigma ds\frac{1}{2}\frac{H}{\mathrm{\Pi }_\sigma ^{ji}}d\mathrm{\Pi }_\sigma ^{ji}ds+d\mathrm{\Pi }_\sigma ^{ji}da_i^\sigma ds_j+\\ \hfill +\frac{1}{2}a_i^\nu a_j^\rho C_{\rho \nu }^\sigma d\mathrm{\Pi }_\sigma ^{ji}ds+\mathrm{\Pi }_\mu ^{ji}C_{\rho \sigma }^\mu a_j^\rho da_i^\sigma ds\end{array}$$
(4.9)
A direct comparison of eqs. (4.8) and (4.9) gives the algebraic expressions satisfied by the components of $`\gamma `$:
$$\mathrm{\Gamma }_{j\sigma }^{ji}+\frac{H}{a_i^\sigma }\mathrm{\Pi }_\mu ^{ji}C_{\rho \sigma }^\mu a_j^\rho =0$$
(4.10a)
$$\mathrm{\Gamma }_{ij}^\sigma \mathrm{\Gamma }_{ji}^\sigma +\frac{H}{\mathrm{\Pi }_\sigma ^{ji}}a_i^\nu a_j^\rho C_{\rho \nu }^\sigma =0$$
(4.10b)
Another direct comparison immediately shows that every integral section of a connection $`\gamma `$ satisfying eqs. (4.10) automatically verify the Hamilton–De Donder eqs. (3.18).
The Lagrangian version of the above multimomentum Hamiltonian formulation is obtained by means of Legendre transform. In fact the Lagrangian multisymplectic form on $`𝒥(E)`$ is defined through the Legendre map as:
$$\mathrm{\Omega }_L:=d\left(\frac{L}{F_{ji}^\mu }\right)da_i^\mu ds\frac{}{x^j}$$
(4.11)
The target connections of the fibration $`𝒥(E)M`$ are of the form
$$\gamma =\left(\frac{}{x^k}+\mathrm{\Gamma }_{kh}^\mu \frac{}{a_h^\mu }+\frac{1}{2}\mathrm{\Gamma }_{kst}^\mu \frac{}{F_{st}^\mu }\right)dx^k$$
(4.12)
with $`\mathrm{\Gamma }_{kst}^\mu =\mathrm{\Gamma }_{kts}^\mu `$, and satisfy the equation
$$\gamma \text{ }\text{ }\mathrm{\Omega }_L=d\mathrm{\Theta }_L$$
(4.13)
Because of the following relation
$$\begin{array}{c}\hfill d\mathrm{\Theta }_L=\frac{L}{a_i^\mu }da_i^\mu ds\frac{1}{2}F_{ij}^\mu d\left(\frac{L}{F_{ij}^\mu }\right)dsd\left(\frac{L}{F_{ij}^\mu }\right)da_i^\mu ds_j+\\ \hfill \frac{1}{2}a_i^\nu a_j^\rho C_{\rho \nu }^\mu d\left(\frac{L}{F_{ij}^\mu }\right)ds\frac{L}{F_{ij}^\mu }C_{\rho \sigma }^\mu a_j^\rho da_i^\sigma ds\end{array}$$
it is easily seen that every $`\gamma `$ solution of eq. (4.13) satisfies the following conditions:
$$\begin{array}{c}\hfill \left(\frac{^2L}{x^jF_{ji}^\sigma }+\mathrm{\Gamma }_{jh}^\mu \frac{^2L}{a_k^\mu F_{ji}^\sigma }+\frac{1}{2}\mathrm{\Gamma }_{jst}^\mu \frac{^2L}{F_{st}^\mu F_{ji}^\sigma }\right)\frac{L}{F_{ji}^\mu }a_j^\gamma C_{\gamma \sigma }^\mu \frac{L}{a_i^\sigma }=0\end{array}$$
(4.14a)
$$F_{ij}^\mu +\mathrm{\Gamma }_{ji}^\mu \mathrm{\Gamma }_{ij}^\mu +a_i^\lambda a_j^\gamma C_{\gamma \lambda }^\mu =0$$
(4.14b)
Once again, it is easy to verify that the integral sections of such a connection $`\gamma `$ automatically satisfy Euler–Lagrange equations (2.15).
## 5 $`3+3`$ Yang–Mills fields
In this section a particular Gauge theory is considered: the base manifold $`M`$ will be taken to be $`3`$-dimensional and the gauge groups can be equivalently chosen between $`G=SO(3)`$ and $`G=SO(2,1)`$.
Under these hypotheses, both space–time and algebra indexes run from $`1`$ to $`3`$. Besides, given a basis $`\{\underset{¯}{e}_\mu \}`$ for the Lie algebra g of $`G`$, we denote by $`K_{\mu \nu }`$ the coefficients of an Ad-invariant metric over g such that the structure coefficients $`C_{\lambda \sigma }^\mu `$ are expressed in the form
$$C_{\lambda \sigma }^\mu =\frac{1}{2}\sqrt{K}K^{\mu \nu }ϵ_{\nu \lambda \sigma }$$
(5.1)
where $`\sqrt{K}=\sqrt{|\mathrm{det}K_{\mu \nu }|}`$, $`K^{\mu \nu }K_{\nu \sigma }=\delta _\sigma ^\mu `$ and $`ϵ_{\nu \lambda \sigma }`$ are the $`3`$-dimensional Levi–Civita permutation symbols.
The use of a dual formulation allows to express such a $`(3+3)`$ gauge theory in terms of a gravity–like theory in purely metric formulation, as proved in .
Now, making an explicit use of the Poincaré–Cartan approach of section 3, we will show that the Hamiltonian version of a $`(3+3)`$ gauge theory has the same shape as an Einstein–Cartan theory. Borrowing from , the central idea consists in performing a local coordinate transformation in the phase space $`\mathrm{\Pi }(E)`$, locally described by the following relations:
$$\{\begin{array}{c}e_p^\nu =\frac{1}{2}K^{\mu \nu }\mathrm{\Pi }_\mu ^{ij}ϵ_{pij}\hfill \\ \\ \omega _{i\beta \alpha }=\frac{1}{2}\sqrt{K}ϵ_{\mu \alpha \beta }a_i^\mu \hfill \end{array}$$
(5.2)
where $`ϵ`$ denotes the usual $`3`$-dimensional Levi–Civita permutation symbol. The inverse transformation of (5.2) is given by
$$\{\begin{array}{c}\mathrm{\Pi }_\mu ^{ij}=K_{\mu \nu }e_p^\nu ϵ^{pij}\hfill \\ \\ a_i^\mu =\frac{1}{\sqrt{K}}ϵ^{\mu \sigma \lambda }\omega _{i\lambda \sigma }\hfill \end{array}$$
(5.3)
It will soon be clear that the coordinates $`e_i^\mu `$ play the role of the triad coordinates, while the coordinates $`\omega _{i\beta \alpha }`$ represent the coefficients of the spin–connection.
It is now easy to see that the Poincaré–Cartan 1–form (3.16) in the new coordinates has the form
$$\mathrm{\Theta }_h=HdsK_{\mu \nu }e_p^\nu ϵ^{pij}\frac{1}{\sqrt{K}}\left(ϵ^{\mu \sigma \lambda }d\omega _{i\lambda \sigma }ds_j+\frac{1}{2}ϵ^{\sigma \alpha \beta }\omega _{j\sigma }^\mu \omega _{i\beta \alpha }ds\right)$$
(5.4)
where $`\omega _{j\sigma }^\mu :=\omega _{j\nu \sigma }K^{\mu \nu }`$.
###### Proposition 5.1
The following identities hold identically:
$$\frac{1}{2}ϵ^{\rho \alpha \beta }\omega _{j\nu \rho }\omega _{i\beta \alpha }=K_{\mu \nu }ϵ^{\mu \sigma \lambda }\omega _{i\lambda \eta }\omega _{j\sigma }^\eta $$
(5.5)
Proof. A direct calculation shows that the left hand side is such that
$$\frac{1}{2}ϵ^{\rho \alpha \beta }\omega _{j\nu \rho }\omega _{i\beta \alpha }=\omega _{j\nu 1}\omega _{i23}\omega _{j\nu 2}\omega _{i13}+\omega _{j\nu 3}\omega _{i12}$$
while the right hand side becomes
$$\begin{array}{c}\hfill K_{\mu \nu }ϵ^{\mu \sigma \lambda }\omega _{i\lambda \eta }\omega _{j\sigma }^\eta =K_{\nu 1}\omega _{i3\eta }\omega _{j\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}2}}^\eta K_{\nu 2}\omega _{i3\eta }\omega _{j\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}1}}^\eta +K_{\nu 3}\omega _{i2\eta }\omega _{j\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}1}}^\eta +\\ \hfill K_{\nu 1}\omega _{i2\eta }\omega _{j\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}3}}^\eta +K_{\nu 2}\omega _{i1\eta }\omega _{j\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}3}}^\eta K_{\nu 3}\omega _{i1\eta }\omega _{j\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}2}}^\eta =\\ \hfill K_{\nu 1}\omega _{i31}\omega _{j\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}2}}^{\mathrm{\hspace{0.33em}\hspace{0.33em}1}}+K_{\nu 1}\omega _{i32}\omega _{j\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}2}}^{\mathrm{\hspace{0.33em}\hspace{0.33em}2}}K_{\nu 2}\omega _{i31}\omega _{j\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}1}}^{\mathrm{\hspace{0.33em}\hspace{0.33em}1}}K_{\nu 2}\omega _{i32}\omega _{j\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}1}}^{\mathrm{\hspace{0.33em}\hspace{0.33em}2}}+\\ \hfill K_{\nu 3}\omega _{i21}\omega _{j\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}1}}^{\mathrm{\hspace{0.33em}\hspace{0.33em}1}}+K_{\nu 3}\omega _{i23}\omega _{j\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}1}}^{\mathrm{\hspace{0.33em}\hspace{0.33em}3}}K_{\nu 1}\omega _{i21}\omega _{j\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}3}}^{\mathrm{\hspace{0.33em}\hspace{0.33em}1}}K_{\nu 1}\omega _{i23}\omega _{j\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}3}}^{\mathrm{\hspace{0.33em}\hspace{0.33em}3}}+\\ \hfill K_{\nu 2}\omega _{i12}\omega _{j\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}3}}^{\mathrm{\hspace{0.33em}\hspace{0.33em}2}}+K_{\nu 2}\omega _{i13}\omega _{j\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}3}}^{\mathrm{\hspace{0.33em}\hspace{0.33em}3}}K_{\nu 3}\omega _{i12}\omega _{j\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}2}}^{\mathrm{\hspace{0.33em}\hspace{0.33em}2}}K_{\nu 3}\omega _{i13}\omega _{j\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}2}}^{\mathrm{\hspace{0.33em}\hspace{0.33em}3}}\end{array}$$
Now we notice that:
$$\omega _{j\nu 1}\omega _{i23}=K_{\nu 1}\omega _{i23}\omega _{j\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}1}}^{\mathrm{\hspace{0.33em}\hspace{0.33em}1}}+K_{\nu 2}\omega _{i23}\omega _{j\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}1}}^{\mathrm{\hspace{0.33em}\hspace{0.33em}2}}+K_{\nu 3}\omega _{i23}\omega _{j\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}1}}^{\mathrm{\hspace{0.33em}\hspace{0.33em}3}}$$
$$\omega _{j\nu 2}\omega _{i13}=K_{\nu 1}\omega _{i13}\omega _{j\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}3}}^{\mathrm{\hspace{0.33em}\hspace{0.33em}1}}+K_{\nu 2}\omega _{i13}\omega _{j\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}3}}^{\mathrm{\hspace{0.33em}\hspace{0.33em}2}}+K_{\nu 3}\omega _{i13}\omega _{j\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}3}}^{\mathrm{\hspace{0.33em}\hspace{0.33em}3}}$$
$$\omega _{j\nu 3}\omega _{i12}=K_{\nu 1}\omega _{i12}\omega _{j\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}2}}^{\mathrm{\hspace{0.33em}\hspace{0.33em}1}}+K_{\nu 2}\omega _{i12}\omega _{j\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}2}}^{\mathrm{\hspace{0.33em}\hspace{0.33em}2}}+K_{\nu 3}\omega _{i12}\omega _{j\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}2}}^{\mathrm{\hspace{0.33em}\hspace{0.33em}3}}$$
whence:
$$\begin{array}{cc}\hfill K_{\mu \nu }ϵ^{\mu \sigma \lambda }\omega _{i\lambda \eta }\omega _{j\sigma }^\eta & =\frac{1}{2}ϵ^{\rho \alpha \beta }\omega _{j\nu \rho }\omega _{i\beta \alpha }+\hfill \\ & K_{\nu 1}\omega _{i23}\omega _{j\mu }^\mu K_{\nu 2}\omega _{i31}\omega _{j\mu }^\mu K_{\nu 3}\omega _{i12}\omega _{j\mu }^\mu \hfill \end{array}$$
The conclusion follows from the trace properties of the coefficients $`\omega _{j\mu }^\mu =0`$.
Taking the identity (5.5) into account, we can write the differential of $`\mathrm{\Theta }_h`$ in the form:
$$\begin{array}{cc}\hfill d\mathrm{\Theta }_h=& \frac{H}{e_i^\lambda }de_i^\lambda ds\frac{1}{2}\frac{H}{\omega _{i\lambda \sigma }}d\omega _{i\lambda \sigma }ds+\frac{1}{\sqrt{K}}ϵ^{pij}ϵ^{\rho \alpha \beta }\omega _{j\rho \nu }e_p^\nu d\omega _{i\beta \alpha }ds\hfill \\ & K_{\mu \nu }ϵ^{pij}de_p^\nu \frac{1}{\sqrt{K}}ϵ^{\mu \sigma \lambda }\left(d\omega _{i\lambda \sigma }ds_j\omega _{i\lambda \eta }\omega _{j\sigma }^\eta ds\right)\hfill \end{array}$$
(5.6)
Now, let $`X=X_p^\nu \frac{}{e_p^\nu }+\frac{1}{2}X_{i\lambda \sigma }\frac{}{\omega _{i\lambda \sigma }}`$ be a vertical vector field, with respect to the fibration $`\mathrm{\Pi }(E)M`$, on the phase space $`\mathrm{\Pi }(E)`$. We calculate the inner product
$$\begin{array}{cc}\hfill X\text{ }\text{ }d\mathrm{\Theta }_h=& \left(\frac{H}{e_p^\nu }dsϵ^{pij}ϵ^{\mu \sigma \lambda }\frac{K_{\mu \nu }}{\sqrt{K}}d\omega _{i\lambda \sigma }ds_j+ϵ^{pij}ϵ^{\mu \sigma \lambda }\frac{K_{\mu \nu }}{\sqrt{K}}\omega _{i\lambda \eta }\omega _{j\sigma }^\eta ds\right)X_p^\nu \hfill \\ \hfill +& \left(\frac{1}{2}\frac{H}{\omega _{i\lambda \sigma }}ds+ϵ^{pij}ϵ^{\mu \sigma \lambda }\frac{K_{\mu \nu }}{\sqrt{K}}de_p^\nu ds_j+\frac{1}{\sqrt{K}}ϵ^{pij}ϵ^{\rho \sigma \lambda }\omega _{j\rho \nu }e_p^\nu ds\right)X_{i\lambda \sigma }\hfill \end{array}$$
(5.7)
The imposition on the Hamilton–De Donder conditions $`\gamma ^{}(X\text{ }\text{ }d\mathrm{\Theta }_h)=0X`$ yields the final equations
$$\frac{H}{e_i^\lambda }ϵ^{pij}ϵ^{\mu \sigma \lambda }\frac{K_{\mu \nu }}{\sqrt{K}}\left(\frac{\omega _{i\lambda \sigma }}{x^j}+\omega _{j\lambda \eta }\omega _{i\sigma }^\eta \right)=0$$
(5.8a)
$$\frac{H}{\omega _{i\lambda \sigma }}+\frac{2K_{\mu \nu }}{\sqrt{K}}ϵ^{pij}ϵ^{\mu \sigma \lambda }\left(\frac{e_p^\nu }{x^j}+\omega _{j\gamma }^\nu e_p^\gamma \right)=0$$
(5.8b)
representing the Hamilton–De Donder equations in the new coordinates.
As it was anticipated at the beginning of the section, eqs. (5.8) have the form of the $`3`$-dimensional Einstein–Cartan equations, where the coordinates $`e_i^\mu `$ and $`\omega _{i\mu \nu }`$ respectively represent the triad components (whenever $`\mathrm{det}e_i^\mu 0`$) and the spin–connection coefficients.
In particular, let us consider a free Yang–Mills field, whose dynamical properties are described by the usual Lagrangian density $`L=\frac{1}{4}F_{ip}^\mu F_{jq}^\nu g^{ij}g^{pq}K_{\mu \nu }\sqrt{g}`$, where $`g_{ij}`$ is a given metric over $`M`$ and $`g:=|\mathrm{det}g_{ij}|`$. Under such circumstances, the Legendre transformation and the Hamiltonian are respectively described by the following equation:
$$\mathrm{\Pi }_\mu ^{ij}=\frac{L}{F_{ij}^\mu }=F_{pq}^\nu g^{ip}g^{jq}K_{\mu \nu }\sqrt{g},H=\frac{1}{4}\frac{1}{\sqrt{g}}\mathrm{\Pi }_\sigma ^{pq}\mathrm{\Pi }_\lambda ^{st}g_{sp}g_{tq}K^{\sigma \lambda }$$
When the new coordinates (5.2) are introduced, the Hamiltonian takes the form:
$$H=\frac{1}{2}G_{kh}g^{kh}\sqrt{g}\sigma (g)(G_{hk}:=e_k^\mu e_h^\nu K_{\mu \nu })$$
with $`\sigma (g)`$ representing the sign of $`\mathrm{det}g_{ij}`$. Since
$$\frac{H}{\omega _{i\lambda \sigma }}=0;\frac{H}{e_p^\nu }=e_k^\mu K_{\mu \nu }g^{kp}\sqrt{g}\sigma (g)$$
eqs. (5.8) take the form
$$2K_{\mu \nu }ϵ^{pij}ϵ^{\mu \sigma \lambda }\left(\frac{e_p^\nu }{x^j}+\omega _{j\gamma }^\nu e_p^\gamma \right)=0$$
(5.9a)
$$\frac{1}{2}e_k^\mu K_{\mu \nu }g^{kp}\sqrt{g}\sigma (g)ϵ^{pij}ϵ_{\nu \lambda \sigma }R_{ij}^{\lambda \sigma }\sqrt{K}\sigma (K)=0$$
(5.9b)
where
$$R_{ij\lambda \sigma }=\frac{\omega _{j\lambda \sigma }}{x^i}\frac{\omega _{i\lambda \sigma }}{x^j}+\omega _{i\lambda \eta }\omega _{j\sigma }^\eta \omega _{j\lambda \eta }\omega _{i\sigma }^\eta ,R_{ij}^{\lambda \sigma }=R_{ij\mu \nu }K^{\mu \lambda }K^{\nu \sigma }$$
and $`\sigma (K)=\mathrm{sign}(\mathrm{det}\mathrm{K}_{\mu \nu }))`$.
Under the hypothesis $`\mathrm{det}e_i^\mu 0`$ eqs. (5.9) have the same form as Einstein equations in the triad–affine formulation. Because of eq. (5.9a), the solution $`\omega _{i\mu \nu }(x)`$ is equal to the (spin–connection associated with) Levi–Civita connection induced by the metric $`G=K_{\mu \nu }e^\mu (x)e^\nu (x)`$, which is a solution of eq. (5.9b).
More in particular, eqs. (5.9) actually describe a first–order purely frame–formulation of a General Relativity like theory in three dimensions.
Infact, we notice that the transformation laws of the coordinates (5.2) are
$$\overline{e}_j^\mu =e_i^\sigma Ad(\gamma ^1)_\sigma ^\mu \frac{x^i}{\overline{x}^j}$$
(5.10a)
and
$$\overline{\omega }_{i\mu \nu }=Ad(\gamma )_\mu ^\sigma Ad(\gamma )_\nu ^\gamma \frac{x^j}{\overline{x}^i}\omega _{j\sigma \gamma }+Ad(\gamma )_\mu ^\eta \frac{Ad(\gamma ^1)_\eta ^\sigma }{x^h}\frac{x^h}{\overline{x}^i}K_{\sigma \nu }$$
(5.10b)
Eqs. (5.10a) are the transition functions of a bundle $`\pi :𝒯M`$, associated with $`P\times _ML(M)`$ ($`L(M)`$ being the frame bundle over $`M`$) through the left action
$$\lambda :(G\times GL(3,\mathrm{}))\times GL(3,\mathrm{})GL(3,\mathrm{}),\lambda (g,J;X):=Ad(g)XJ^1$$
(5.11)
The (local) sections $`e:M𝒯`$ may be identified with (local) triads $`e_i^\mu (x)dx^i`$ on $`M`$; the latter are truly gauge natural objects , sensitive to the changes of trivialization of the structure bundle $`P`$. Each triad $`e^\mu `$ induces a metric on $`M`$ expressed as $`G:=K_{\mu \nu }e^\mu e^\nu `$, which is invariant under transformations (5.10a) by construction.
A new $`𝒥`$-bundle $`\widehat{\pi }:𝒥(𝒯)M`$ can also be constructed by quotienting the first–jet bundle $`j_1(𝒯)`$ of $`\pi :𝒯M`$ with respect to an equivalence relation analogous to (2.3). The bundle $`𝒥(𝒯)`$ is naturally referred to local coordinates $`x^i,e_i^\mu ,E_{ij}^\mu :=\frac{1}{2}\left(e_{ij}^\mu e_{ji}^\mu \right)`$ $`(i<j)`$.
Now the idea is to choose the components of the spin–connections generated by the triads themselves as fiber coordinates on the bundle $`𝒥(𝒯)`$.
Within this framework, let $`z=(x^i,e_i^\mu ,E_{ij}^\mu )`$ be an element of $`𝒥(𝒯)`$, $`x=\widehat{\pi }(z)`$ its projection over $`M`$, $`e^\mu `$ a representative triad belonging to the equivalence class $`z`$ and $`G=K_{\mu \nu }e^\mu e^\nu `$ the metric on $`M`$ induced by the triad $`e^\mu `$; we also denote by $`\mathrm{\Gamma }_{ih}^k`$ the Levi–Civita connection induced by the metric $`G`$ and by $`\omega _{i\nu }^\mu `$ the spin connection associated with $`\mathrm{\Gamma }_{ih}^k`$ through the triad $`e^\mu `$ itself.
The relation between the coefficients $`\mathrm{\Gamma }_{ih}^k`$ and $`\omega _{i\nu }^\mu `$, evaluated in the point $`x=\widehat{\pi }(z)M`$, is expressed by the equation
$$\omega _{i\nu }^\mu (x)=e_k^\mu (x)\left(\mathrm{\Gamma }_{ij}^ke_\nu ^j(x)+\frac{e_\nu ^k(x)}{x^i}\right)$$
(5.12)
If the coefficients $`\mathrm{\Gamma }_{ih}^k`$ are written in terms of the triad $`e^\mu `$ and its derivatives, one gets the well–known expression
$$\omega _{i\nu }^\mu (x):=e_p^\mu (x)\left(\mathrm{\Sigma }_{ji}^p(x)\mathrm{\Sigma }_{ji}^p(x)+\mathrm{\Sigma }_{ij}^p(x)\right)e_\nu ^j(x)$$
(5.13)
where
$$\mathrm{\Sigma }_{ji}^p(x):=e_\lambda ^p(x)E_{ij}^\lambda (x)=e_\lambda ^p(x)\frac{1}{2}\left(\frac{e_i^\lambda (x)}{x^j}\frac{e_j^\lambda (x)}{x^i}\right)$$
(5.14)
the Latin indexes being lowered and raised by means of the metric $`G`$. Equations (5.13) and (5.14) show that the values of the coefficients of the spin–connection $`\omega _{i\nu }^\mu `$, evaluated in $`x=\widehat{\pi }(z)`$, are independent of the choice of the representative $`e^\mu `$ in the equivalence class $`z𝒥(𝒯)`$.
Moreover, the torsion–free condition for the connection $`\omega _{i\nu }^\mu `$ gives a sort of inverse relation of eq. (5.13) in the form
$$2E_{ij}^\mu (x)=\omega _{i\nu }^\mu (x)e_j^\nu (x)\omega _{j\nu }^\mu (x)e_i^\nu (x)$$
(5.15)
Because of the metric compatibility condition $`\omega _{i\mu \nu }:=\omega _{i\nu }^\sigma K_{\sigma \mu }=\omega _{i\nu \mu }`$, there exists a one-to-one correspondence between the values of the antisymmetric part of the derivatives $`E_{ij}^\mu (x)=\frac{1}{2}\left(\frac{e_i^\mu (x)}{x^j}\frac{e_j^\mu (x)}{x^i}\right)`$ and the coefficients of the spin–connection $`\omega _{i\mu \nu }(x)`$ in the point $`x=\widehat{\pi }(z)`$.
The above considerations allow us to take the quantities $`\omega _{i\mu \nu }`$ as fiber coordinates of the bundle $`𝒥(𝒯)`$, looking at the relations (5.13) and (5.15) as coordinate changes in $`𝒥(𝒯)`$.
Finally, it is a straightforward matter to verify that the transformation laws of the spin connection coefficients $`\omega _{i\mu \nu }`$ coincide with eqs. (5.10b), as well as that the $`3`$-form (5.3) is invariant under coordinate transformations (5.10a), (5.10b). |
warning/0507/math0507341.html | ar5iv | text | # A combinatorial generalization of the Boson-Fermion correspondence
## 1. Introduction
The classical Boson-Fermion correspondence is an isomorphism between two representations of the Heisenberg algebra $`H`$: the Bosonic Fock space $`K[H_{}]`$ and the Fermionic Fock space $`^{(0)}`$. It identifies the Schur functions $`s_\lambda (x_1,x_2,\mathrm{})`$ as the images of the basis of semi-infinite wedges $`v_{i_1}v_{i_2}\mathrm{}`$ under this isomorphism. The Boson-Fermion correspondence is an important basic result in mathematical physics; see for example .
The aim of this article is to replace the classical Fermionic Fock space in the Boson-Fermion correspondence by another representation of the Heisenberg algebra, and to obtain other interesting families of symmetric functions instead of the Schur functions. The symmetric functions that we obtain have a tableaux-like definition, and satisfy both Pieri-like identities and a Cauchy-like identity, which we now explain.
Let $`\{F_\lambda (x_1,x_2,\mathrm{})\mathrm{\Lambda }_K:\lambda S\}`$ be a family of symmetric functions with coefficients in a field $`K`$ (usually $``$, $`(q)`$ or $`(q,t)`$), where $`S`$ is some indexing set. Many important families of symmetric functions have the following trio of properties.
1. They can be expressed as the generating functions for a set of “tableaux”, which gives the monomial expansion of $`F_\lambda `$:
$$F_\lambda (x_1,x_2,\mathrm{})=\underset{T}{}s(T)x^{\mathrm{wt}(T)},$$
where the sum is over tableaux $`T`$ with “shape” $`\lambda `$. The composition $`\mathrm{wt}(T)`$ is the weight of $`T`$ and $`s(T)K`$ is some additional parameter associated to $`T`$.
2. Together with a closely related dual family $`\{G_\lambda (x_1,x_2,\mathrm{}):\lambda S\}`$ of symmetric functions, they satisfy a Cauchy identity:
$$\underset{\lambda S}{}F_\lambda (x_1,x_2,\mathrm{})G_\lambda (y_1,y_2,\mathrm{})=\underset{i,j=1}{\overset{\mathrm{}}{}}\left(b_0+b_1x_iy_j+b_2(x_iy_j)^2+\mathrm{}\right),$$
where the coefficients $`b_iK`$.
3. They satisfy a Pieri formula:
$$\stackrel{~}{h}_k(x_1,x_2,\mathrm{})F_\lambda (x_1,x_2,\mathrm{})=\underset{\mu _k\lambda }{}b_{\lambda ,\mu }F_\mu (x_1,x_2,\mathrm{}),$$
where $`k`$ is a positive integer, $`\{\stackrel{~}{h}_1,\stackrel{~}{h}_2,\mathrm{}\}\mathrm{\Lambda }_K`$ is a sequence of symmetric functions and $`b_{\lambda ,\mu }K`$ are coefficients for each pair $`\lambda ,\mu `$ satisfying some condition $`\mu _k\lambda `$.
In all such cases that the author is aware of, the definition of a tableaux involves the condition $`\mu _k\lambda `$ in the Pieri formula. The simplest case is when $`K=`$ and $`F_\lambda =s_\lambda `$, the family of Schur functions. The indexing set $`S=𝒫`$ is the set of partitions. The tableaux are usual semi-standard Young tableaux $`T`$; the statistic $`s(T)`$ is equal to 1 and $`\mathrm{wt}(T)`$ is the usual weight associated to $`T`$. The dual family $`\{G_\lambda =s_\lambda \}`$ is equal to the Schur functions again and in the Cauchy formula, all the coefficients $`b_i=1`$. In the Pieri formula, $`\stackrel{~}{h}_k=h_k`$ are the homogeneous symmetric functions. The condition $`\mu _k\lambda `$ is that $`\mu /\lambda `$ is a horizontal strip of size $`k`$ and all the coefficients $`b_{\lambda ,\mu }=1`$. Recall in particular that a semi-standard Young tableaux is just a chain of partitions forming a sequence of horizontal strips.
Understanding the ubiquity of these three properties in families of symmetric functions was one of the main aims of our work. Our main result is as follows. Given a representation $`V`$ of a Heisenberg algebra $`H`$ with a distinguished basis $`\{v_ssS\}`$, together with a highest vector $`v_b`$ in $`V`$, we define a family $`F_s^V(x_1,x_2,\mathrm{})`$ (and a dual family $`G_s^V`$) of symmetric functions which satisfy a generalized Boson-Fermion correspondence. The definition of $`F_s^V`$ is tableaux-like: for example it gives the monomial expansion of $`F_s^V`$. We show in addition that $`F_s^V`$ satisfy a Pieri rule and a Cauchy identity. Examples of symmetric functions that can be obtained in this way include the Schur functions, Schur $`Q`$-functions, Hall-Littlewood functions and Macdonald polynomials; see .
The motivating example for us was actually a family $`𝒢_\lambda (x_1,x_2,\mathrm{};q)`$ of $`q`$-symmetric functions defined by Lascoux, Leclerc and Thibon combinatorially via ribbon tableaux and algebraically using the action of the Heisenberg algebra on the Fock space of the quantized affine algebra $`U_q(\widehat{𝔰𝔩}_n)`$. In we studied the $`𝒢_\lambda `$ in analogy with Schur functions and discovered ribbon Cauchy and Pieri identities. The current work is an attempt to understand this in a more systematic and general framework. As an application, we now give natural generalizations of the functions $`𝒢_\lambda `$ to Fock spaces of other types and also to higher level Fock spaces. By our main result, these new symmetric functions satisfy Cauchy and Pieri rules as well, and will be the subject of later work.
Our Pieri and Cauchy formulae depend heavily on a sequence $`a_i`$ of parameters defining the relations of the Heisenberg algebra $`H=H[a_i]`$ (see Section 3). On the other hand, as an abstract algebra, the Heisenberg algebra does not depend on the $`a_i`$ (as long as they are non-zero). Thus it is not clear immediately which sequences $`a_i`$ would lead to an interesting theory of symmetric functions.
Our work is also closely related to more combinatorial work of Fomin and of Bergeron and Sottile . Fomin is mostly concerned with Schensted correspondences and the Cauchy identities while Bergeron and Sottile’s work has led to relations with non-commutative symmetric functions and to Hopf algebras. It seems that an interesting non-commutative version of our theory also exists, though we have not attempted to make this precise in the present article.
It would be most interesting to investigate other families of symmetric functions which arise using our correspondence from other representations of Heisenberg algebras which occur naturally.
We now briefly describe the organization of the rest of the paper. In Section 2, we review the theory of Schur functions and symmetric functions. In Section 3, we describe the classical Boson-Fermion correspondence. In Section 4, we explain how to obtain symmetric functions from representations of Heisenberg algebras. In Section 5, we prove our generalized Boson-Fermion correspondence. In Section 6, we prove Pieri and Cauchy formulae for our families of symmetric functions. In Section 7, we prove a partial converse to the theorems of Sections 5 and 6. In Section 8, we give a series of examples beginning with Schur functions, Macdonald polynomials and the behavior when taking direct sums or tensor products of representations. We then explain the example of Lascoux, Leclerc and Thibon’s ribbon functions studied in . Finally, we explain how to generalise ribbon functions to other types and higher levels, following work of Kashiwara, Miwa, Petersen and Yung and Takemura and Uglov .
Acknowledgements. I thank Sergey Fomin for discussions related to this work. This work is part of my Ph.D Thesis at M.I.T., written under the guidance of Richard Stanley.
## 2. Schur functions
We will follow mostly the notation of . Let $`K`$ be a field with characteristic 0. Let $`\mathrm{\Lambda }_K`$ denote the ring of symmetric functions over $`K`$. The ring $`\mathrm{\Lambda }_K`$ should be thought of as the ring of formal power series in countably many variables $`x_1,x_2,\mathrm{}`$, of bounded degree. If the variable set is important then we write $`\mathrm{\Lambda }_K(X)`$ or $`\mathrm{\Lambda }_K(Y)`$. We will let $`h_1,h_2,\mathrm{}`$ denote the homogeneous symmetric functions and $`p_1,p_2,\mathrm{}`$ denote the power sum symmetric functions. Each of these sets forms a set of algebraically independent generators for $`\mathrm{\Lambda }_K`$.
Let $`𝒫`$ denote the set of partitions. Let $`\lambda =(\lambda _1\lambda _2\mathrm{}\lambda _l>0)𝒫`$ be a partition. The size $`|\lambda |`$ of $`\lambda `$ is equal to $`\lambda _1+\mathrm{}+\lambda _l`$ and we write $`\lambda |\lambda |`$. We also write $`l(\lambda )=l`$. We generally do not distinguish between a partition $`\lambda `$ and its Young diagram $`D(\lambda )`$. If $`D(\mu )D(\lambda )`$ then $`\lambda /\mu `$ is a skew shape with size $`|\lambda /\mu |=|\lambda ||\mu |`$.
We let $`h_\lambda :=h_{\lambda _1}h_{\lambda _2}\mathrm{}h_{\lambda _l}`$ and $`p_\lambda :=p_{\lambda _1}p_{\lambda _2}\mathrm{}p_{\lambda _l}`$. The sets $`\{h_\lambda :\lambda 𝒫\}`$ and $`\{p_\lambda :\lambda 𝒫\}`$ are bases of $`\mathrm{\Lambda }_K`$. The homogeneous symmetric functions and the power sum symmetric functions are related by the formula
(1)
$$h_n=\underset{\lambda n}{}z_\lambda ^1p_\lambda $$
where $`z_\lambda =1^{m_1(\lambda )}m_1(\lambda )!2^{m_2(\lambda )}m_2(\lambda )!\mathrm{}`$ and $`m_i(\lambda )=|\{j\lambda _j=i\}|`$.
The monomial symmetric functions are denoted $`m_\lambda `$ and the Schur functions are denoted $`s_\lambda `$. The Schur functions (and more generally skew Schur functions) are the generating functions of Young tableaux:
(2)
$$s_\lambda (x_1,x_2,\mathrm{})=\underset{T}{}x^{\mathrm{wt}(T)},$$
where the sum is over all semistandard Young tableaux $`T`$ of shape $`\lambda `$. Alternatively, $`s_\lambda =_\mu K_{\lambda \mu }m_\mu `$ where the Kostka number $`K_{\lambda \mu }`$ is equal to the number of semistandard Young tableaux of shape $`\lambda `$ and weight $`\mu `$. For the purposes of this paper, a Young tableaux $`T`$ of shape $`\lambda `$ should be thought of as a chain of partitions $`T=(\mathrm{}=\lambda ^0\lambda ^1\mathrm{}\lambda ^l=\lambda )`$ such that each skew shape $`\lambda ^i/\lambda ^{i1}`$ is a horizontal strip. A horizontal strip is a skew shape containing at most one box in each column. The weight of $`T`$ is then the composition $`\mathrm{wt}(T)=(|\lambda ^1/\lambda ^0|,|\lambda ^2/\lambda ^1|,\mathrm{},|\lambda ^l/\lambda ^{l1}|)`$. Similarly a Young tableaux of skew shape $`\lambda /\mu `$ is a chain of partitions $`(\mu =\lambda ^0\lambda ^1\mathrm{}\lambda ^l=\lambda )`$.
The Schur functions satisfy the following Pieri formula, which describes how to write the product of a Schur function and a homogeneous symmetric function in terms of Schur functions:
(3)
$$h_ks_\lambda =\underset{\mu _k\lambda }{}s_\mu ,$$
where here $`\mu _k\lambda `$ means that the skew shape $`\mu /\lambda `$ is a horizontal strip of size $`k`$.
The Schur functions also satisfy the following Cauchy formula, which holds within the ring $`\mathrm{\Lambda }_K(X)\widehat{}_K\mathrm{\Lambda }_K(Y)`$, which is the completion of the tensor product of two copies of the symmetric functions.
(4)
$$\underset{i,j}{}\frac{1}{1x_iy_j}=\underset{\lambda }{}s_\lambda (x_1,x_2,\mathrm{})s_\lambda (y_1,y_2,\mathrm{}).$$
The ring of symmetric functions $`\mathrm{\Lambda }_K`$ possesses a bilinear symmetric form $`.,.:\mathrm{\Lambda }_K\times \mathrm{\Lambda }_KK`$ given by $`s_\lambda ,s_\mu =\delta _{\lambda \mu }`$, or alternatively by $`p_\lambda ,p_\mu =\delta _{\lambda \mu }z_\lambda `$. This inner product is known as the Hall inner product. If $`f\mathrm{\Lambda }_K`$ then $`f^{}\mathrm{End}(\mathrm{\Lambda }_K)`$ denotes the linear operator adjoint to multiplication by $`f`$. As a particular case $`p_k^{}=k\frac{}{p_k}`$ where the differential operator acts on symmetric functions written as polynomials in the power sum symmetric functions.
## 3. The classical Boson-Fermion correspondence
Let $`K`$ be a field with characteristic 0. The Heisenberg algebra $`H=H[a_i]`$ denotes the associative algebra over $`K`$ with 1 generated by $`\{B_k:k\backslash \{0\}\}`$ satisfying
$$[B_k,B_l]=la_l\delta _{k,l},$$
for some non-zero parameters $`a_lK`$ satisfying $`a_l=a_l`$. As an abstract algebra, $`H`$ does not depend on the choice of the elements $`a_l`$, since the generators $`B_k`$ can be re-scaled to force $`a_l=1`$. However, we shall be concerned with representations of $`H`$, and some choices of the generators $`B_k`$ will be more natural.
Let $`K[H_{}]=K[B_1,B_2,\mathrm{}]`$ denote the Bosonic Fock space representation of $`H`$. The action of the Heisenberg algebra on $`K[B_1,B_2,\mathrm{}]`$ is determined by letting $`B_k`$ act by multiplication for $`k<0`$ and setting $`B_k1=0`$ for $`k>0`$.
One can identify $`K[B_1,B_2,\mathrm{}]`$ with the algebra $`\mathrm{\Lambda }_K`$ of symmetric functions over $`K`$ by identifying $`B_k`$ with $`a_kp_k`$ for $`k>0`$. The action of $`H`$ on $`\mathrm{\Lambda }_K`$ is then given by
$$B_k\{\begin{array}{cc}a_kp_k\hfill & \text{for }k1\text{,}\hfill \\ k\frac{}{p_k}\hfill & \text{for }k1\text{.}\hfill \end{array}$$
We will need the following standard lemma later.
###### Lemma 1.
Let $`k1`$ be an integer and $`\lambda `$ be a partition. Then
$$B_kB_\lambda =ka_km_k(\lambda )B_\mu +B_\lambda B_k,$$
where $`\mu `$ is $`\lambda `$ with one less part equal to $`k`$. If $`m_k(\lambda )=0`$ then the first term is just 0.
If $`V`$ is a representation of $`H`$, then a vector $`vV`$ is called a highest weight vector if $`B_kv=0`$ for $`k>0`$. The following result is well known. See for example \[9, Proposition 2.1\].
###### Proposition 2.
Let $`V`$ be an irreducible representation of $`H`$ with non-zero highest weight vector $`vV`$. Then there exists a unique isomorphism of $`H`$-modules $`\varphi :V\stackrel{~}{}K[B_1,B_2,\mathrm{}]`$ such that $`\varphi (v)=1`$.
For the remainder of this section we assume that $`H=H[1]`$ is given by the parameters $`a_l=1`$ for $`l1`$ and $`a_l=1`$ for $`l1`$. Let $`W=_jKv_j`$ be an infinite-dimensional vector space with basis $`\{v_j:jZ\}`$. Let $`^{(0)}`$ denote the vector space with basis given by semi-infinite monomials of the form $`v_{i_0}v_{i_1}\mathrm{}`$ where the indices satisfy:
1. $`i_0>i_1>i_2>\mathrm{}`$
2. $`i_k=k`$ for $`k`$ sufficiently small.
We will call $`^{(0)}`$ the Fermionic Fock space.
###### Remark 1.
Usually $`^{(0)}`$ is considered a subspace of a larger space $`=_m^{(m)}`$. The spaces $`^{(m)}`$ are defined as for $`^{(0)}`$ with the condition (ii) replaced by the condition (ii<sup>(m)</sup>): $`i_k=km`$ for $`k`$ sufficiently small.
Define an action of $`H`$ on $`^{(0)}`$ by
(5)
$$B_k(v_{i_0}v_{i_1}\mathrm{})=\underset{j0}{}v_{i_0}v_{i_1}\mathrm{}v_{i_{j1}}v_{i_jk}v_{i_{j+1}}\mathrm{}.$$
The monomials are to be reordered according to the usual exterior algebra commutation rules so that $`v_{i_0}\mathrm{}v_{i_j}v_{i_{j+1}}\mathrm{}=v_{i_0}\mathrm{}v_{i_{j+1}}v_{i_j}\mathrm{}`$. Thus the sum on the right hand side of (5) is actually finite so the action is well defined. One can check that we indeed do obtain an action of $`H`$. It is also not hard to see that the representation of $`H`$ on $`^{(0)}`$ is irreducible.
The vector $`\overline{v}=v_0v_1\mathrm{}^{(0)}`$ is a highest weight vector for this action of $`H`$. By Proposition 2, there exists an isomorphism $`\sigma :^{(0)}\mathrm{\Lambda }_K`$ sending $`\overline{v}1`$. An algebraic version of the Boson-Fermion correspondence identifies the image of $`v_{i_0}v_{i_1}\mathrm{}`$ under the isomorphism $`\sigma `$.
###### Theorem 3 (\[9, Lecture 6\]).
Let $`\lambda _k=i_k+k`$. Then $`\sigma (v_{i_0}v_{i_1}\mathrm{})=s_\lambda `$.
In , this is called the “second” part of the boson-fermion correspondence. It is important in the study of a family of non-linear differential equations known as the Kadomtzev-Petviashvili (KP) Hierarchy. The “first” part consists of identifying the image of certain vertex operators under $`\sigma `$. The relationship between vertex operators and symmetric function theory have been studied previously in .
Our aim will be to generalise Theorem 3 to representations of Heisenberg algebras with arbitrary parameters $`a_iK`$. We will see that the symmetric functions that one obtains in this manner will always have a tableaux-like definition and satisfy Pieri and Cauchy identities. In our approach, we have ignored the vertex operators, but it would be interesting to see how they are related to our results.
## 4. Symmetric functions from representations of Heisenberg algebras
Let $`H=H[a_i]`$ be the Heisenberg algebra with parameters $`a_iK`$. Define $`B_\lambda :=B_{\lambda _1}B_{\lambda _2}\mathrm{}B_{\lambda _{l(\lambda )}}`$. Let $`D_k:=_{\lambda k}z_\lambda ^1B_\lambda `$ and $`U_k:=_{\lambda k}z_\lambda ^1B_\lambda `$ where $`z_\lambda `$ is as defined in Section 2. Thus $`B_\lambda `$ and $`D_k`$ are related in the same way as $`p_\lambda `$ and $`h_k`$ (see (1)).
Similarly define $`B_\lambda :=B_{\lambda _1}B_{\lambda _2}\mathrm{}B_{\lambda _{l(\lambda )}}`$ and $`U_k:=_{\lambda k}z_\lambda ^1B_\lambda `$. Also let $`S_\lambda H`$ be given by $`S_\lambda :=_\mu z_\mu ^1\chi _\mu ^\lambda B_\mu `$ where the coefficients $`\chi _\mu ^\lambda `$ are the characters of the symmetric group given by $`s_\lambda =_\mu z_\mu ^1\chi _\mu ^\lambda p_\mu `$.
Let $`V`$ be a representation of $`H`$ with distinguished basis $`\{v_s:sS\}`$ for some indexing set $`S`$. For simplicity we will assume that both $`V`$ and $`S`$ are $``$-graded so that $`v_sV`$ are homogeneous elements and $`\mathrm{deg}(v_s)=\mathrm{deg}(s)`$, and that each graded component of $`V`$ is finite-dimensional. We will also assume that the action of $`H`$ is graded in the sense that $`\mathrm{deg}(B_k)=mk`$ for some $`m\backslash \{0\}`$. Define an inner product $`.,.:V\times VK`$ on $`V`$ by requiring that $`\{v_ssS\}`$ forms an orthonormal basis, so that $`v_s,v_s^{}=\delta _{ss^{}}`$.
Let $`s,tS`$. Define the generating functions
(6)
$$F_{s/t}^V(x_1,x_2,\mathrm{})=F_{s/t}(x_1,x_2,\mathrm{}):=\underset{\alpha }{}x^\alpha U_{\alpha _l}U_{\alpha _{l1}}\mathrm{}U_{\alpha _1}t,s,$$
where the sum is over all compositions $`\alpha =(\alpha _1,\alpha _2,\mathrm{},\alpha _l)`$. Similarly define
$$G_{s/t}^V(x_1,x_2,\mathrm{})=G_{s/t}(x_1,x_2,\mathrm{})=\underset{\alpha }{}x^\alpha D_{\alpha _l}D_{\alpha _{l1}}\mathrm{}D_{\alpha _1}s,t.$$
Note that $`F_{s/t}`$ and $`G_{s/t}`$ are homogeneous with degree $`\frac{\mathrm{deg}(s)\mathrm{deg}(t)}{m}`$. So in particular if $`\frac{\mathrm{deg}(s)\mathrm{deg}(t)}{m}`$ is negative or non-integral then the generating functions are 0. For convenience we let $`U_\alpha :=U_{\alpha _l}U_{\alpha _{l1}}\mathrm{}U_{\alpha _1}`$ and $`D_\alpha :=D_{\alpha _l}D_{\alpha _{l1}}\mathrm{}D_{\alpha _1}`$.
The above definitions should be thought of as a tableaux-like definition, as the following example explains.
###### Example 4 (Schur functions).
Let $`H[a_i]=H[1]`$ and $`V=^{(0)}`$. Set $`S=𝒫`$ and $`v_\lambda :=v_{i_0}v_{i_1}\mathrm{}`$, where $`\lambda _k=i_k+k`$. Then we have
$$U_kv_\lambda =\underset{\mu _k\lambda }{}v_\mu ,$$
where the sum is over all horizontal strips $`\mu /\lambda `$ of size $`k`$. So the definition (6) of $`F_{s/t}`$ reduces to (2) – the combinatorial definition of skew Schur functions in terms of Young tableaux.
The following Proposition is immediate from the definition, since $`U_k`$ commutes with $`U_l`$ and $`D_k`$ commutes with $`D_l`$ for all $`k,l`$.
###### Proposition 5.
The generating functions $`F_{s/t}`$ and $`G_{s/t}`$ are symmetric functions.
As before, let $`K[H_{}]H`$ denote the subalgebra of $`H`$ generated by $`\{B_kk<0\}`$ and similarly define $`K[H_+]H`$. The definitions of $`F_{s/t}`$ and $`G_{s/t}`$ can be rephrased in terms of the Heisenberg-Cauchy elements $`\mathrm{\Omega }(H_{},X)`$ and $`\mathrm{\Omega }(H_+,X)`$ which lie in the completed tensor products $`K[H_{}]\widehat{}\mathrm{\Lambda }_K(X)`$ and $`K[H_+]\widehat{}\mathrm{\Lambda }_K(X)`$ respectively:
$$\mathrm{\Omega }(H_{},X):=\underset{\lambda }{}U_\lambda m_\lambda =\underset{\lambda }{}z_\lambda ^1B_\lambda p_\lambda =\underset{\lambda }{}S_\lambda s_\lambda .$$
The last two equalities follow from the classical Cauchy identity. Also define $`\mathrm{\Omega }(H_+,X)K[H_+]\widehat{}\mathrm{\Lambda }_K(X)`$ by $`\mathrm{\Omega }(H_+,X)=_\lambda D_\lambda m_\lambda `$.
Thus for example, one has
$$F_{s/t}(x_1,x_2,\mathrm{})=\mathrm{\Omega }(H_{},X)v_t,v_s$$
and
$$G_{s/t}(x_1,x_2,\mathrm{})=\mathrm{\Omega }(H_+,X)v_s,v_t.$$
One has in particular
(7)
$$G_{s/t}(x_1,x_2,\mathrm{})=\underset{\lambda }{}z_\lambda ^1p_\lambda B_\lambda v_s,v_t.$$
Now let $`bS`$ be such that $`v_b`$ is a highest weight vector for $`H`$. We will write $`F_s:=F_{s/b}`$ and $`G_s:=G_{s/b}`$. The element $`\mathrm{\Omega }(H_{},X)v_bV\widehat{}\mathrm{\Lambda }_K(X)`$ depends only on the choice of $`v_b`$. The symmetric functions $`F_s`$ are the coefficients of $`\mathrm{\Omega }(H_{},X)v_b`$ when it is written in the basis $`\{v_ssS\}`$:
$$\mathrm{\Omega }(H_{},X)v_b=\underset{s}{}v_sF_s(x_1,x_2,\mathrm{}).$$
## 5. Generalization of Boson-Fermion correspondence
Let us suppose that $`bS`$ has been picked so that $`v_bV`$ is a highest weight vector for $`H`$. By Proposition 2, there is a canonical map of $`H`$-modules $`\varphi :Hb\mathrm{\Lambda }_K`$ sending $`v_b1`$. Our choice of inner product for $`V`$ allows us to give a map $`\mathrm{\Phi }:V\mathrm{\Lambda }_K`$.
###### Theorem 6 (Generalized Boson-Fermion correspondence).
The map $`\mathrm{\Phi }:V\mathrm{\Lambda }_K`$ given by $`v_sG_s(x_1,x_2,\mathrm{})`$ is a map of $`H`$-modules.
Recall that $`B_k`$ acts on $`\mathrm{\Lambda }_K`$ by multiplication by $`a_kp_k`$ and $`B_k`$ acts as $`k\frac{}{p_k}`$, for $`k1`$.
###### Proof.
Let us calculate $`B_lG_s`$ and compare with $`\mathrm{\Phi }(B_lv_s)`$. Suppose first that $`l<0`$ and let $`k=l`$. Let $`\lambda `$ be a partition and let $`\mu `$ be $`\lambda `$ with one less part equal to $`k`$. If $`\lambda `$ has no part equal to $`k`$, then $`\mu `$ can be any partition in the following formulae. First write $`B_\lambda B_lv_s,v_b=ka_km_k(\lambda )B_\mu v_s,v_b`$, using a slight variation of Lemma 1 for our $`H`$. Alternatively, one can also compute
$`B_\lambda B_lv_s`$ $`=B_\lambda {\displaystyle \underset{c}{}}B_lv_s,v_cv_c={\displaystyle \underset{c,d}{}}B_lv_s,v_cB_\lambda v_c,v_dv_d`$
so that taking the coefficient of $`v_b`$ we obtain
(8)
$$ka_km_k(\lambda )B_\mu v_s,v_b=\underset{c}{}B_lv_s,v_cB_\lambda v_c,v_b.$$
Now,
$`B_lG_s`$ $`=a_kp_kG_s`$
$`=a_k{\displaystyle \underset{\mu }{}}z_\mu ^1p_kp_\mu B_\mu v_s,v_b`$ $`\text{using (}\text{7}\text{)},`$
$`={\displaystyle \underset{\lambda }{}}z_\lambda ^1p_\lambda \left({\displaystyle \underset{c}{}}B_lv_s,v_cB_\lambda v_c,v_b\right)`$ using (8)
$`={\displaystyle \underset{c}{}}B_lv_s,v_c\left({\displaystyle \underset{\lambda }{}}z_\lambda ^1B_\lambda v_c,v_b\right)`$
$`={\displaystyle \underset{c}{}}B_lv_s,v_cG_c.`$
This shows that $`\mathrm{\Phi }(B_lv_s)=B_l\mathrm{\Phi }(v_s)`$ for $`l<0`$.
Now suppose $`k>0`$, and let $`\lambda `$ and $`\mu `$ be related as before. Then
$`B_kG_s`$ $`=k{\displaystyle \underset{\lambda }{}}z_\lambda ^1{\displaystyle \frac{}{p_k}}p_\lambda B_\lambda v_s,v_b`$
$`=k{\displaystyle \underset{\lambda }{}}z_\lambda ^1m_k(\lambda )p_\mu B_\mu B_kv_s,v_b`$
$`={\displaystyle \underset{\mu }{}}z_\mu ^1p_\mu B_\mu {\displaystyle \underset{c}{}}B_kv_s,v_cv_c,v_b`$
$`={\displaystyle \underset{c}{}}B_kv_s,v_c\left({\displaystyle \underset{\mu }{}}z_\mu ^1p_\mu B_\mu v_c,v_b\right)`$
$`={\displaystyle \underset{c}{}}B_kv_s,v_cG_c.`$
This completes the proof. ∎
When $`V`$ is irreducible, the map $`\mathrm{\Phi }`$ does not depend on the choice of basis, but does depend on $`v_b`$. Since the degree $`\mathrm{deg}(v_b)`$ part of $`V`$ is one dimensional, the image of $`vV`$ is given by the coefficient of the degree $`\mathrm{deg}(v_b)`$ part of $`\mathrm{\Omega }(H_+,X)v`$.
If $`V`$ is not irreducible then the map depends on the inner product $`.,.`$ of $`V`$ (or equivalently, the choice of orthonormal basis).
Note that a different action of $`H`$ on $`\mathrm{\Lambda }_K`$ will allow us to replace the family $`G_s`$ in Theorem 6 by $`F_s`$. More precisely, one can define the adjoint action $`\vartheta :H\mathrm{End}(V)`$ of $`H`$ on $`V`$ by letting the generators $`B_k`$ act according to the formula $`\vartheta (B_k)v_s^{},v_s=v_s^{},B_kv_s`$. With this new representation of $`H`$ on $`V`$, the roles of $`G_s`$ and $`F_s`$ are reversed.
## 6. Pieri and Cauchy identities
Let $`h_k[a_i]`$ denote the image $`\theta (h_k)`$ of $`h_k`$ under the algebra homomorphism $`\theta :\mathrm{\Lambda }\mathrm{\Lambda }_K`$ given by $`\theta (p_k)=a_kp_k`$. Also let $`h_ka_i`$ denote the image $`\kappa (h_k)`$ of $`h_k`$ under the map $`\kappa :\mathrm{\Lambda }_KK`$ given by $`\kappa (p_k)=a_k`$. Note that if all $`\{a_ii1\}`$ are positive (rational) numbers then by (1) so are the numbers $`h_ka_i`$. Let $`h_k^{}`$ be the linear operator on $`\mathrm{\Lambda }_K`$ which is adjoint to multiplication by $`h_k`$ with respect to the Hall inner product.
###### Theorem 7 (Generalized Pieri Rule).
Let $`k1`$. The following identities hold in $`\mathrm{\Lambda }_K`$:
$$h_k[a_i]G_s=\underset{t}{}U_ks,tG_t$$
and
$$h_k[a_i]F_s=\underset{t}{}D_kt,sF_t.$$
The dual identities are:
$$h_k^{}G_s=\underset{t}{}D_ks,tG_t$$
and
$$h_k^{}F_s=\underset{t}{}U_kt,sF_t.$$
###### Proof.
Follows immediately from the definitions of $`U_k,D_k`$ and $`h_k[a_i]`$ together with Theorem 6 and the comments immediately after it. ∎
###### Lemma 8.
The following identity holds as elements of $`H[a_i]`$:
(9)
$$D_bU_a=\underset{j=0}{\overset{m}{}}h_ja_iU_{aj}D_{bj},$$
where $`m=\mathrm{min}(a,b)`$.
###### Proof.
By definition we need to show that
$$\left(\underset{\lambda b}{}z_\lambda ^1B_\lambda \right)\left(\underset{\lambda a}{}z_\lambda ^1B_\lambda \right)=\underset{j=0}{\overset{m}{}}h_ja_i\left(\underset{\lambda aj}{}z_\lambda ^1B_\lambda \right)\left(\underset{\lambda bj}{}z_\lambda ^1B_\lambda \right).$$
Let $`\mu `$ and $`\nu `$ be partitions such that $`|\mu |=aj`$ and $`|\nu |=bj`$. By (1), the coefficient of $`B_\mu B_\nu `$ on the right hand side is equal to $`z_\nu ^1z_\mu ^1_{\lambda j}z_\lambda ^1\theta (p_\lambda )`$. Let $`\rho =\lambda \mu `$ and $`\pi =\lambda \nu `$. We claim that the summand $`z_\nu ^1z_\mu ^1z_\lambda ^1\theta (p_\lambda )`$ is the coefficient of $`B_\mu B_\nu `$ when applying $`[B_k,B_l]=ka_k\delta _{k,l}`$ repeatedly to $`z_\pi ^1z_\rho ^1B_\pi B_\rho `$. This is a straightforward computation, counting the number of ways of picking parts from $`\rho `$ and $`\pi `$ to make the partition $`\lambda `$.
In fact the relation (9), together with the relations $`[U_k,U_l]=[D_k,D_l]=0`$ is equivalent to the defining relations of the Heisenberg algebra $`H[a_i]`$. This is because the sets $`\{B_kk0\}`$ and $`\{U_kk1\}\{D_kk1\}`$ are both generators of $`H[a_i]`$.
###### Theorem 9 (Generalized Cauchy Identity).
We have the following identity in the completion of $`\mathrm{\Lambda }_K(X)\mathrm{\Lambda }_K(Y)`$:
$$\underset{s}{}F_s(x_1,x_2,\mathrm{})G_s(y_1,y_2,\mathrm{})=\underset{j,k}{}\left(1+h_1a_ix_jy_k+h_2a_i(x_jy_k)^2+\mathrm{}\right).$$
More generally, let $`r,tS`$. Then we have
$$\begin{array}{c}\underset{s}{}F_{s/t}(x_1,x_2,\mathrm{})G_{s/r}(y_1,y_2,\mathrm{})=\hfill \\ \hfill \underset{j,k}{}\left(1+h_1a_ix_jy_k+h_2a_i(x_jy_k)^2+\mathrm{}\right)\underset{s}{}F_{r/s}(x_1,x_2,\mathrm{})G_{t/s}(y_1,y_2,\mathrm{}).\end{array}$$
###### Proof.
Let $`U(x):=1+_{i>0}U_ix^i`$ and similarly $`D(x):=1+_{i>0}D_ix^i`$. The identity of Lemma 8 is equivalent to
$$D(y)U(x)=U(x)D(y)\left(1+h_1a_ixy+h_2a_i(xy)^2+\mathrm{}\right).$$
Now notice that by definition we have $`F_{s/t}=\mathrm{}U(x_3)U(x_2)U(x_1)v_t,v_s`$ and $`G_{s/t}=\mathrm{}D(x_3)D(x_2)D(x_1)v_s,v_t`$. The infinite products make sense since in most factors we are picking the term equal to 1. Thus
$`{\displaystyle \underset{s}{}}F_{s/t}(x_1,x_2,\mathrm{})G_{s/r}(y_1,y_2,\mathrm{})`$
$`=\mathrm{}D(y_3)D(y_2)D(y_1)\mathrm{}U(x_3)U(x_2)U(x_1)v_t,v_r`$
$`={\displaystyle \underset{i,j1}{\overset{\mathrm{}}{}}}\left(1+h_1a_ix_iy_j+h_2a_i(x_iy_j)^2+\mathrm{}\right)`$
$`\mathrm{}U(x_3)U(x_2)U(x_1)\mathrm{}D(y_3)D(y_2)D(y_1)v_t,v_r`$
$`={\displaystyle \underset{i,j1}{\overset{\mathrm{}}{}}}\left(1+h_1a_ix_iy_j+h_2a_i(x_iy_j)^2+\mathrm{}\right){\displaystyle \underset{s}{}}G_{t/s}(y_1,y_2,\mathrm{})F_{r/s}(x_1,x_2,\mathrm{}).`$
These manipulations of infinite generating functions make sense since they are well defined when restricted to a finite subset of the variables $`\{x_1,x_2,\mathrm{},y_1,y_2,\mathrm{}\}`$. ∎
###### Remark 2.
It is not clear at this moment which sequences $`a_i`$ and which representations of $`H[a_i]`$ would lead to interesting families of symmetric functions. However, the following may be possible indications:
* Some kind of positivity for the coefficients $`h_ia_i`$; for example if $`K=(q)`$ then we may want $`h_ia_i`$ to have positive coefficients when expanded as a power series in $`q`$.
* A Pieri formula with very few non-zero or with positive coefficients. For example, we may want the coefficients $`U_ks,t`$ and $`D_kt,s`$ to be positive in some sense. This would imply that the definitions of $`F_{t/s}`$ and $`G_{t/s}`$ would also have a positive monomial expansion.
The results of this Section are related to results of Fomin and of Bergeron and Sottile . Fomin studies combinatorial operators on posets and recovers Cauchy style identities similar to ours. His approach is more combinatorial and he focuses on generalizing Schensted style algorithms to these more general situations. Bergeron and Sottile have also made definitions similar to our $`F_{s/t}`$. Their interests have been towards aspects related to Hopf algebras and non-commutative symmetric functions; see also .
###### Remark 3.
An interesting non-commutative version of our theory may exist, where the Heisenberg algebra is replaced with an algebra $`A=B_kk\{0\}`$ with relations
$`[B_k,B_l]`$ $`=0`$ $`\text{if }k\text{ and }l\text{ have opposite sign and }kl,`$
$`[B_k,B_k]`$ $`=ka_k.`$
In this case, the generating functions $`F_{s/t}`$ and $`G_{s/t}`$ will not be symmetric functions but instead be quasi-symmetric functions.
## 7. A partial converse
A partial converse to Theorems 7 and 9 exists. In other words, if a family of symmetric functions satisfies enough properties, then one can conclude that they arise from a generalized Boson-Fermion correspondence as in Theorem 6.
Let $`V`$ be a $`K`$-vector space with a distinguished basis $`\{v_s:sS\}`$. In this section, suppose that $`\{B_k^{}\mathrm{End}(V):k\backslash \{0\}\}`$ are linear operators acting on $`V`$. Suppose further that $`B_k`$ and $`B_l`$ commute if $`k`$ and $`l`$ have the same sign. Let$`D_k^{}:=_{\lambda k}z_\lambda ^1B_\lambda ^{}`$ and $`U_k^{}:=_{\lambda k}z_\lambda ^1B_\lambda ^{}`$. Now we can define $`F_{s/t}^{}(x_1,x_2,\mathrm{}):=_\alpha x^\alpha U_{\alpha _l}^{}U_{\alpha _{l1}}^{}\mathrm{}U_{\alpha _1}^{}t,s`$ and similarly for $`G_{s/t}^{}`$.
###### Theorem 10.
Let $`\{a_kKk0\}`$ be a sequence of non-zero parameters satisfying $`a_k=a_k`$ and suppose that $`\{G_s^{}sS\}`$ are linearly independent. Then the following are equivalent:
1. The operators $`\{B_k^{}\}`$ generate an action of the Heisenberg algebra $`H[a_i]`$ with parameters $`a_i`$.
2. The family $`\{G_s^{}\}`$ satisfies the conclusions of Theorem 7.
3. The families $`\{G_{s/t}^{}\}`$ and $`\{F_{s/t}^{}\}`$ satisfy the conclusions of Theorem 9.
###### Proof.
That (1) implies (2) and (3) is Theorems 7 and 9.
Now suppose (2) holds. Since the family $`\{G_s^{}\}`$ is linearly independent, the action of $`\{U_k^{},D_k^{}\}`$ on $`V`$ is isomorphic to the action of $`\{h_k[a_i],h_k^{}\}`$ on $`\mathrm{span}_K\{G_s^{}\}`$ under the isomorphism $`v_sG_s^{}`$. Thus the action of the operators $`B_k^{}`$ on $`V`$ is isomorphic to the action of $`\{\theta (p_k),p_k^{}\}`$ on $`\mathrm{span}_K\{G_s^{}\}`$ and so generate an action of $`H[a_i]`$. Thus (2) $``$ (1).
Now suppose (3) holds. Then by the argument in the proof of Theorem 9, we must have
$$\left(D^{}(y)U^{}(x)U^{}(x)D^{}(y)\left(1+h_1a_ixy+h_2a_i(xy)^2+\mathrm{}\right)\right)v_t,v_r=0$$
for every $`t,rS`$. This implies that
$$D^{}(y)U^{}(x)=U^{}(x)D^{}(y)\left(1+h_1a_ixy+h_2a_i(xy)^2+\mathrm{}\right)$$
so that we have
$$D_b^{}U_a^{}=\underset{j=0}{\overset{m}{}}h_ja_iU_{aj}^{}D_{bj}^{}.$$
Now reversing the argument in the proof of Lemma 8, we deduce that $`[B_k^{},B_l^{}]=ka_k\delta _{k,l}`$. So (3) $``$ (1).
## 8. Examples
### 8.1. Schur functions
If $`K=`$ and $`V=^{(0)}`$ and $`H=H_{\text{Schur}}=H[1]`$ acts as in Section 3, then Theorem 6 is just Theorem 3, where the indexing set $`S`$ can be identified with the set of partitions $`𝒫`$. In this case, the operators $`B_k`$ and $`B_k`$ are adjoint with respect to $`.,.`$ and so $`F_\lambda =G_\lambda =s_\lambda `$ for every $`\lambda `$. The definition of $`s_{\lambda /\mu }=F_{\lambda /\mu }`$ in terms of the operators $`U_k`$ is exactly the usual combinatorial definition of skew Schur functions in terms of semistandard Young tableaux. The symmetric function $`h_k[a_i]=h_k`$ is the usual homogeneous symmetric function and the coefficients $`U_k\lambda ,\mu `$ are equal to 1 if $`\mu /\lambda `$ is a horizontal strip of size $`k`$ and equal to 0 otherwise. The coefficients $`h_ia_i`$ are all equal to 1 and Theorem 9 reduces to the usual Cauchy identity.
### 8.2. Direct sums
Let $`V_1`$ and $`V_2`$ be two representations of $`H`$ with distinguished bases $`\{v_{s_1}:s_1S_1\}`$ and $`\{v_{s_2}:s_2S_2\}`$ respectively. Then $`V=V_1V_2`$ is a representation of $`H[a_i]`$ with distinguished basis $`\{v_ssS_1S_2\}`$. If $`s,tS_i`$ for some $`i`$ then $`F_{s/t}^V=F_{s/t}^{V_i}`$ otherwise if for example $`sS_1`$ and $`tS_2`$ we have $`F_{s/t}^V=0`$. Thus the family of symmetric functions that we obtain from $`H[a_i]`$ acting on $`V`$ is the union of the families of symmetric functions we obtain from $`V_1`$ and $`V_2`$.
### 8.3. Tensor products
Let $`V_1`$ and $`V_2`$ be two representations of $`H[a_i]`$ with distinguished bases $`\{v_{s_1}:s_1S_1\}`$ and $`\{v_{s_2}:s_2S_2\}`$ respectively, as before. Then $`V_1V_2`$ has a distinguished basis $`\{v_{s_1}v_{s_2}s_1S_1\text{and}s_2S_2\}`$. Let the Heisenberg algebra $`\stackrel{~}{H}:=H[b_i]`$ with generators $`\stackrel{~}{B_k}`$, where $`b_i=2a_i`$, act on $`V_1V_2`$ by defining the action of $`\stackrel{~}{B}_k`$ by
$$\stackrel{~}{B}_kv_1v_2=(B_kv_1)v_2+v_1(B_kv_2).$$
This action is natural when one views $`\mathrm{\Lambda }_K`$ as a Hopf algebra. The action of $`\stackrel{~}{U}_k=_{\lambda k}z_\lambda ^1\stackrel{~}{B}_\lambda `$ is given by
$$\stackrel{~}{U}_kv_1v_2=\underset{i=0}{\overset{k}{}}(U_iv_1)(U_{ki}v_2)$$
and similarly for $`\stackrel{~}{D}_k`$. By definition, one sees that $`F_{s_1s_2/t_1t_2}^{V_1V_2}=F_{s_1/t_1}^{V_1}F_{s_2/t_2}^{V_2}`$ and similarly for the $`G`$-functions. Thus the family of symmetric functions we obtain from $`V=V_1V_2`$ are pairwise products of the symmetric functions we obtain from $`V_1`$ and $`V_2`$.
More generally, the tensor products $`V_1\mathrm{}V_n`$ lead to generating functions which are products $`F_{s_1/t_1}^{V_1}\mathrm{}F_{s_n/t_n}^{V_n}`$ of $`n`$ original generating functions. We will denote the Heisenberg algebra acting on this tensor product by $`H^{(n)}:=H[a_i^{(n)}]`$. The parameters are given by $`a_i^{(n)}=na_i`$.
### 8.4. Macdonald polynomials
Let $`K=(q,t)`$ and let $`P_\lambda (x_1,x_2,\mathrm{};q,t)`$ and $`Q_\lambda (x_1,x_2,\mathrm{};q,t)`$ be the Macdonald polynomials introduced in . Let $`\lambda =(\lambda _1,\lambda _2,\mathrm{})`$ be a partition and $`s=(i,j)\lambda `$ be a square. Then the arm-length of $`s`$ is given by $`a_\lambda (s)=\lambda _ij`$ and the leg-length of $`s`$ is given by $`l_\lambda (s)=\lambda _j^{}i`$. Now let $`s`$ be any square. Define (\[16, Chapter VI, (6.20)\])
$$b_\lambda (s)=b_\lambda (s;q,t)=\{\begin{array}{cc}\frac{1q^{a_\lambda (s)}t^{l_\lambda (s)+1}}{1q^{a_\lambda (s)+1}t^{l_\lambda (s)}}\hfill & \text{if }s\lambda \text{,}\hfill \\ 1\hfill & \text{otherwise.}\hfill \end{array}$$
Now let $`\lambda /\mu `$ be a horizontal strip. Let $`C_{\lambda /\mu }`$ (respectively $`R_{\lambda /\mu }`$) denote the union of columns (respectively rows) that intersect $`\lambda \mu `$. Define (\[16, Chapter VI, (6.24)\])
$$\varphi _{\lambda /\mu }=\underset{sC_{\lambda /\mu }}{}\frac{b_\lambda (s)}{b_\mu (s)},$$
and
$$\psi _{\lambda /\mu }=\underset{sR_{\lambda /\mu }C_{\lambda /\mu }}{}\frac{b_\mu (s)}{b_\lambda (s)}.$$
Let $`V_{\text{Mac}}`$ denote the vector space over $`K`$ with distinguished basis labeled by partitions. Define operators $`\{U_k,D_k:k_{>0}\}`$ by:
$$U_k\lambda =\underset{\mu }{}\varphi _{\mu /\lambda }\mu ,D_k\lambda =\underset{\mu }{}\psi _{\lambda /\mu }\mu ,$$
where the sums are over horizontal strips of size $`|k|`$. Then $`Q_{\lambda /\mu }=F_{\lambda /\mu }`$ and $`P_{\lambda /\mu }=G_{\lambda /\mu }`$, so in particular the operators $`\{U_kk_{>0}\}`$ commute and so do the operators $`\{D_kk_{>0}\}`$. Now we have (\[16, Ex.7.6\])
$$\underset{\rho }{}Q_{\rho /\lambda }(X;q,t)P_{\rho /\mu }(Y;q,t)=(\underset{\sigma }{}Q_{\mu /\sigma }(X;q,t)P_{\lambda /\sigma }(Y;q,t))\underset{i,j}{}\underset{r=0}{\overset{\mathrm{}}{}}\frac{1tx_iy_jq^r}{1x_iy_jq^r}.$$
The product $`_{r=0}^{\mathrm{}}\frac{1ytq^r}{1yq^r}`$ can be written as $`_{n0}g_n(1,0,0,\mathrm{};q,t)y^n`$ where $`g_n`$ is given by (\[16, Chapter VI, (2.9)\])
$$g_n(x_1,x_2,\mathrm{};q,t)=\underset{\lambda n}{}z_\lambda (q,t)^1p_\lambda (x_1,x_2,\mathrm{}),$$
where $`z_\lambda (q,t)=z_\lambda _{i=1}^{l(\lambda )}\frac{1q^{\lambda _i}}{1t^{\lambda _i}}`$. Using Theorem 10, we see that the operators $`\{U_k,D_kk_{>0}\}`$ generate a copy of a Heisenberg algebra $`H_{\text{Mac}}`$. A short calculation shows that the parameters $`a_k(q,t)`$ of this Heisenberg algebra are given by $`a_k=\frac{1t^k}{1q^k}`$. The parameters $`h_ka_i`$ are given by $`h_ka_i=g_n(1,0,0,\mathrm{};q,t)=n_{\lambda n}z_\lambda (q,t)^1`$.
In fact Theorem 10 shows that the Pieri (and dual Pieri) rule for Macdonald polynomials is equivalent to the (generalized) Cauchy identity for Macdonald polynomials.
###### Remark 4.
To obtain the Hall-Littlewood functions, one can just specialize $`q=0`$ in the set up of this section. However, to obtain the Schur $`P`$ and $`Q`$-functions the further specialization $`t=1`$ actually causes some of the $`a_i`$ to be zero. In this case, one should actually consider the subalgebra of the Heisenberg algebra generated by the generators $`B_k`$ where $`k`$ is odd.
### 8.5. Ribbon functions
Let $`n1`$ be a positive integer and $`K=(q)`$. In , a family of symmetric functions $`\{𝒢_\lambda ^{(n)}(x_1,x_2,\mathrm{};q)\}`$ defined in terms of ribbon tableaux, called ribbon functions or LLT-polynomials, were introduced. These symmetric functions arise as the polynomials $`\{F_s^𝐅(x_1,x_2,\mathrm{})\}`$ for the action of a Heisenberg algebra $`H[a_i]`$ on the level one Fock space $`𝐅`$ of $`U_q(\widehat{𝔰𝔩}_n)`$. This Fock space $`𝐅`$ has a basis $`|\lambda `$ naturally labeled by partitions. The parameters are given by $`a_i=\frac{1q^{2nk}}{1q^{2k}}`$ and the action of $`H[a_i]`$, commuting with the action of $`U_q(\widehat{𝔰𝔩}_n)`$, was discovered in . The actions of the generators $`B_k`$ and $`B_k`$ of this Heisenberg algebra are adjoint with respect to the inner product $`|\lambda ,|\mu =\delta _{\lambda \mu }`$, and so the symmetric functions $`F_\lambda `$ and $`G_\lambda `$ for this representation of $`H[a_i]`$ coincide. In , a ribbon Cauchy and Pieri formula for the functions $`𝒢_\lambda ^{(n)}(X;q)`$ was deduced from the action of $`H[a_i]`$ and this is a special (in fact, motivating) case for Theorems 7 and 9.
At $`q=1`$, the Fock space $`𝐅`$ for $`U_q(\widehat{𝔰𝔩}_n)`$ should be thought of as a sum of tensor products:
(10)
$$𝐅\underset{n\text{-cores}}{}(^{(0)})^n$$
where $`^{(0)}`$ is the classical Fermionic Fock space described in Section 3. Combinatorially, the decomposition (10) is given by writing a partition in terms of its $`n`$-core and its $`n`$-quotient; see . As shown in subsection 8.3, the $`F`$-functions we obtain in this way are products of $`n`$ of the $`F`$-functions for $`^{(0)}`$, that is, (skew) Schur functions. This is simply the formula $`𝒢_\lambda (x_1,x_2,\mathrm{};1)=s_{\lambda ^{(0)}}s_{\lambda ^{(1)}}\mathrm{}s_{\lambda ^{(n1)}}`$ observed in . In fact, the $`q=1`$ specialization corresponds to action of the Heisenberg algebra commuting with the action of $`\widehat{𝔰𝔩}_n`$ on $`𝐅`$.
It would be interesting to see whether ribbon functions and Macdonald polynomials can be combined by finding a deformation of the action of $`(H_{\text{Mac}})^{(n)}`$ on $`V_{\text{Mac}}^n`$.
### 8.6. Ribbon functions for other types and other levels
Theorem 6 allows us to define analogues of LLT’s ribbon functions $`𝒢^{(n)}(x_1,x_2,\mathrm{};q)`$ for other (quantized) Fock spaces.
Kashiwara, Miwa, Petersen and Yung have defined (level one) $`q`$-deformed Fock spaces for the affine algebras $`A_n^{(1)}`$, $`A_{2n}^{(2)}`$, $`B_n^{(2)}`$, $`A_{2n1}^{(2)}`$, $`D_n^{(1)}`$ and $`D_{n+1}^{(2)}`$, using a sophisticated construction involving perfect crystals. Let $`\mathrm{\Phi }`$ denote one of these root systems and let $`U_q(𝔤)`$ be the corresponding quantum affine algebra. Let $`^\mathrm{\Phi }`$ be the corresponding $`q`$-deformed Fock space of , which is defined over $`K=(q)`$. The space $`^\mathrm{\Phi }`$ is equipped with an action of an Heisenberg algebra $`H[a_i^\mathrm{\Phi }]`$ commuting with the action of $`U_q(𝔤)`$, where the parameters $`a_i^\mathrm{\Phi }`$ are calculated in . The Fock space $`^\mathrm{\Phi }`$ also has a standard basis indexed by certain semi-infinite products of elements from a perfect crystal for $`U_q(𝔤)`$. We will call this indexing set $`S^\mathrm{\Phi }`$. There is a distinguished highest weight vector $`v_b^\mathrm{\Phi }`$ for some “bottom element” $`bS^\mathrm{\Phi }`$.
###### Definition 11.
Let $`sS^\mathrm{\Phi }`$. The ribbon function of type $`\mathrm{\Phi }`$ is given by $`𝒢_s^\mathrm{\Phi }=F_{s/b}^^\mathrm{\Phi }\mathrm{\Lambda }_K`$.
When $`\mathrm{\Phi }=A_{n1}^{(1)}`$, we recover LLT’s ribbon functions $`𝒢^\mathrm{\Phi }=𝒢^{(n)}(x_1,x_2,\mathrm{};q)`$. The functions $`𝒢^{(n)}(x_1,x_2,\mathrm{};q)`$ have been found to be not only interesting combinatorially (see ) but also to be related to the global basis of the Fock space and to Kazhdan-Lusztig polynomials (see ). One should expect the symmetric functions $`𝒢_s^\mathrm{\Phi }`$ to be interesting as well. Some work in this direction can be found in and will appear separately. Note that it is not known (but in some cases a conjecture) that the action of the generators $`B_k`$ and $`B_k`$ of the Heisenberg algebra on $`^\mathrm{\Phi }`$ are adjoint. This would imply that $`F_{s/b}^^\mathrm{\Phi }=G_{s/b}^^\mathrm{\Phi }`$.
In another direction, Takemura and Uglov have studied Fock spaces $`𝐅^{n,m}`$ for the quantum affine algebra $`U_q(\widehat{𝔰𝔩}_n)`$ of level $`m`$. These Fock spaces also possess a standard basis indexed by partitions and an action of a Heisenberg algebra $`H^{n,m}`$ commuting with the action of $`U_q(\widehat{𝔰𝔩}_n)`$.
###### Definition 12.
Let $`\lambda 𝒫`$. The ribbon function of rank $`n`$ and level $`m`$ is given by $`𝒢_\lambda ^{(n,m)}=F_{\lambda /\mathrm{}}^{𝐅^{n,m}}\mathrm{\Lambda }_K`$.
We have placed the parameters $`n`$ and $`m`$ together in the notation since as explained in there is a level-rank duality in this Fock space. The case $`m=1`$ reduces to LLT’s ribbon functions: $`𝒢_\lambda ^{(n,1)}=𝒢_\lambda ^{(n)}`$. One should expect the functions $`𝒢_\lambda ^{(n,m)}`$ to be interesting as well. The parameters $`a_i`$ for $`H^{n,m}`$ appear to have not yet been calculated, though there are precise conjectures for their values. |
warning/0507/astro-ph0507015.html | ar5iv | text | # Helical Jet in the Gravitationally Lensed Blazar PKS1830-211
## 1 Introduction
The much-studied gravitationally lensed system, PKS1830-211, was identified as such by Rao & Subrahmanyam (1988), and is one of the strongest radio sources in the sky at centimetric wavelengths. On arcsecond scale, it consists of two prominent radio core-knot structures, lensed images of each other, each core-knot axis being roughly perpendicular to the line of separation of the two cores (the two cores being separated by about $`0^\mathrm{"}.98`$ and aligned roughly northeast (NE) and southwest (SW) in the plane of the sky). The two core-knot groups (hereafter NE and SW groups) are point-inversion symmetric with respect to the centre of the system, and show somewhat weaker diffuse emission (Rao & Subrahmanyam 1988, Subrahmanyam et al. 1990); in the radio L-band, the diffuse emission is extended enough that its images actually close round, linking the image groups and making a pseudo-Einstein Ring system (Jauncey et al. 1991), with nearly half of the total flux density of about $`12`$ Jy coming from the cores.The cores themselves have flat radio spectra (Rao & Subrahmanyan 1988). On the scale of milliarcseconds in the radio at 43 GHz, a weaker, inner knot accompanies each of the bright cores, on the same side of each core as the arcsecond-scale knot (which is not detected at this resolution), and separated from it by $`1`$ mas (Garrett et al 1997, Jones et al., Guirado et al. 1999, Jin et al. 2003).
Jin et al. (2003) infer that there are time-dependent positional variations in the radio centroids of emission in the bright cores at 43 GHz, resulting in changes in the separation of the two core components on the scale of $`10`$’s to $`100`$’s $`\mu `$as (Figure 1) between successive observational epochs. Significant radio variability has been reported (Lovell et al. 1996, van Ommen et al. 1995, Hagiwara et al. 1996 and Lovell et al. 1998); in fact, Lovell et al.(1998) report a time delay between the arrival time at the observer of variations in one core image relative to (correlated) variations in the other, of $`26_5^{+4}`$ days. However, Romero et al.(1997) report this system to be non-variable, based on data taken during 1995-1996 in the radio L-band. Lovell et al’s (1998) analysis was based on $`8.6`$ GHz data between 1996 and 1997, during which period dramatic variations occured.
The source has been identified in EGRET observations (Mattox et al., 1997).
In the X-ray, ROSAT obseravtions have revealed absorption against the compact images, in excess of what may be reasonably attributed to our Galaxy, implicating a foreground (lensing) galaxy (Mathur & Nair 1997). ASCA observatons (Oshima et al. 2001) suggest significant X-ray variability in the system, so that the ratio of the NE image’s emission to that of the SW image was a factor of $`7`$ (as compared with a typical factor of $`1`$ to $`1.8`$ in the radio). Given that the source properties resemble those of a blazar, this is very likely a variation intrinsic to the blazar rather than a lens-induced one, though observations of correlated spectral hardness changes would be necessary to confirm that such a phenomenon indeed occurs in this system.
The system that acts as the gravitational lens, now known to be fairly complex, is believed to be in the main a gas-rich massive spiral galaxy, with a measured redshift of $`z_1=0.89`$, through molecular absorption line studies (Wiklind & Combes 1996, Carilli et al. 1998, Chengalur et al. 1999). In the optical, observations are hindered by the location of the system (in Galactic coordinates, $`l=12^o.2,b=5^o.7`$; here optical extinction is experienced of order $`2.7`$ magnitudes, $`cf.`$ Subrahmanyan et al 1990); nonetheless Frye et al. (1999)(I and K band observations), Courbin et al. (1998, 2002) (V, I, J, H & K bands), Lehar et al.(2000) (H and I bands) and Winn et al. (2002)(I and V bands) have presented a wealth of optical and near-IR data, which, suitably image processed, reveal a spiral galaxy lens seen almost face on, with possibly a secondary lens nearby, and perhaps some sign of a lower redshift spiral galaxy about $`2^\mathrm{"}.5`$ southwards of the system. Lidman et al. (1999) determine through infrared spectroscopy a redshift for the source $`(z_s=2.51)`$. HI observations (Lovell et al. 1996) have earlier revealed some absorption at redshift $`z_2=0.19`$. The presence in the optical and IR bands of a foreground low redshift spiral galaxy is highly suggestive of the existence of a secondary lens.
Gravitational lens modeling of the system that has attempted to reproduce a range of image properties (Subrahmanyan et al. 1990, Kochanek & Narayan 1992, Nair et al. 1993), has shown the system to be consistent with an isolated elliptical lens of mass of the order of that of a spiral galaxy. However, in view of the apparently complex optical/IR structure subsequently observed along the foreground and upto the lens redshift, a detailed remodeling is in order.
The present work seeks to provide an explanation for the puzzling observations of Garrett et al.(1998) and Jin et al.(2003), wherein it is shown that structural and temporal variations occur in and around the radio core images in this system in VLBA observations at 43 GHz. These observations probe the system at the highest resolution to date, of considerable interest because of the combination of a high redshift source $`(z_s=2.5)`$ and enhanced resolution due to magnification by lensing. For the purpose of the present study, a full reworking of the lens model is actually inessential; instead, certain lens properties applicable to this system will be either invoked or derived as required.
## 2 Some properties of the lens
Considering the optical and infrared observations of possible multiple lensing bodies along the line of sight to the system (as described in the previous section), only the most elementary assumptions regarding the lens system will be made. At the location of each image in the plane of the sky, it is assumed that it is possible to define a $`2\times 2`$ source-to-image coordinate transformation matrix that is symmetric. This is valid provided one or the other of two circumstances occurs: (a) the dominant lensing agents are roughly at the same distance from the observer, or from the source, so that they can be said to inhabit the same lens plane (cf. Blandford & Narayan $`1986)`$, or (b) save for the main deflectors in the same lens plane, lenses at other redshifts provide only a shear perturbation (however strong or weak) to the main lens’s action. Then the lensing equations reduce to one identical with single-plane lensing, and source-to-image transformation matrices are symmetric as a result (Kovner 1987). The latter assumption is adopted in view of the possible existence of a lens at $`z=0.19`$, which is seen $`2.^\mathrm{"}5`$ south, in the plane of the sky, of the lensed images in the system. In fact, since it has proven fairly successful to model this system with a single elliptical lens (Subrahmanyan et al. 1990, Kochanek & Narayan 1992, Nair et al. 1993), it is possible that the main lens, a spiral galaxy at $`z_1=0.89`$ which is viewed almost face-on, is aided by shear roughly along the north-south direction by the foreground lens, mimicking the behaviour of a single elliptical lens with such an orientation (note that this is close to the direction of the major axis of the elliptical lens model in Nair et al. 1993). That the effect of the foreground $`(z_2=0.19)`$ lens can be treated as a perturbation on the main lens’ action derives some support from the fact that lens models place the effective lens centre between the NE and SW groups of images, and somewhat closer to the negative-parity SW group of images (the parity of which is demonstrated by the fact that temporal variations in its core’s radio flux density lag behind those in the NE core; $`cf.`$ Lovell et al. 1998). This would be as expected for a single lensing galaxy, suggesting that the actual lens is not very far from one.
## 3 A lensed helical jet
### 3.1 Inferred source evolution from observations on the scale of tens of microarcseconds
Figure 1 (from Fig.3 of Jin et al. 2003), plots, for 8 successive epochs of VLBA observations at 43 GHz $`(0.^\mathrm{"}3`$ mas FWHM beam), changes in declination versus changes in right ascension (RA) of the position vector from the centroid of emission of the NE image core to that of the SW image core, all relative to the first epoch ($`t_{\mathrm{ref}}`$) of observation. Formally, the error in estimating these changes is of the order of $`3\mu `$as as both the core images appear in the same field of view in each observation (Jin et al. 2003). The plot is displayed here with straight-line fits to two remarkably linear tracks that can be identified in the plot. In this section, an explanation for this phenomenon will be sought.
Anticipating a physical model to be discussed later in this section, it is assumed that the source, a blazar, emits plasmons, one by one, from its central engine, and that the observed (but essentially unresolved) core radio emission is dominated for a while by that of the most recently emitted plasmon, which propagates away from the central engine with constant velocity before fading from view (its spectral evolution, synchrotron self-absorbed, being as described in, for example, van der Laan 1966). Any changes in the location of the radio emission of the central engine itself are assumed to be undetectable as a function of time $`(3\mu `$as). (The actual position of the central engine then drops out of the following computations, because we consider changes in position relative to a reference epoch).
The radio core of the source and its associated features are lensed into two images, denoted by the subscripts $`\mathrm{A}`$ (the NE group) and $`\mathrm{B}`$ (the SW group) in what follows. Changes occuring in the unlensed source shall be referred to as being in the source plane, and the corresponding effects in the images shall be referred to as occuring in the image plane. For the unlensed source, let the angular distance of an ejected plasmon from the central engine at any given epoch of observation, $`t_k`$, be denoted by $`\overline{\mathrm{\Delta }\mathrm{\Phi }}[t_k]`$, its components along the directions of increasing RA and declination being represented by $`\mathrm{\Delta }\mathrm{\Phi }_\alpha `$ and $`\mathrm{\Delta }\mathrm{\Phi }_\delta `$ respectively.
Now, local to a point in a gravitationally lensed image, positional changes are mapped through a linear transformation of the underlying changes in the source. That is, for small changes $`\overline{\mathrm{\Delta }\mathrm{\Phi }}`$ in the source, the corresponding changes in images A and B are given by:
$$(\mathrm{\Delta }\mathrm{\Phi }_{\alpha ,\mathrm{A}},\mathrm{\Delta }\mathrm{\Phi }_{\delta ,\mathrm{A}})=\left(\begin{array}{cc}a_1& p\\ p& a_4\end{array}\right)(\begin{array}{c}\mathrm{\Delta }\mathrm{\Phi }_\alpha \\ \mathrm{\Delta }\mathrm{\Phi }_\delta \end{array})$$
(1)
$$(\mathrm{\Delta }\mathrm{\Phi }_{\alpha ,\mathrm{B}},\mathrm{\Delta }\mathrm{\Phi }_{\delta ,\mathrm{B}})=\left(\begin{array}{cc}b_1& q\\ q& b_4\end{array}\right)(\begin{array}{c}\mathrm{\Delta }\mathrm{\Phi }_\alpha \\ \mathrm{\Delta }\mathrm{\Phi }_\delta \end{array})$$
(2)
Two comments are in order at this point: (a) It has been assumed that the source-to-image transformation matrices are symmetric, under the circumstances discussed in Section 2 and (b) , when considering observed changes in the images, it is necessary to account for the fact that events in the SW image are behind those in the NE image, by an amount equal to the time delay between the images, $`\tau =26_4^{+5}`$ days (Lovell et al. 1998). Thus the observed changes in the two images map back to two different epochs in the source itself (epochs referring to events in the source will hereinafter be called source epochs, as distinct from observer epochs).
Formally, it is possible to define a relative or image-to-image transformation matrix (in this case, the SW image-to-NE image, or ‘SW2NE’, matrix will be defined):
$$(\mathrm{\Delta }\mathrm{\Phi }_{\alpha ,\mathrm{A}},\mathrm{\Delta }\mathrm{\Phi }_{\delta ,\mathrm{A}})=\left(\begin{array}{cc}T_1& T_2\\ T_3& T_4\end{array}\right)(\begin{array}{c}\mathrm{\Delta }\mathrm{\Phi }_{\alpha ,\mathrm{B}}\\ \mathrm{\Delta }\mathrm{\Phi }_{\delta ,\mathrm{B}}\end{array}),$$
(3)
where $`\eta =(b_1b_4q^2)`$, $`T_1=(a_1b_4pq)/\eta `$, $`T_2=(b_1pa_1q)/\eta `$, $`T_3=(pb_4qa_4)/\eta `$, and $`T_4=(a_4b_1pq)/\eta `$. When evaluating this matrix from the observations, which here include evolving components, care must be taken that appropriate corrections are introduced for the differing source epochs (or alternatively, light arrival time delays) corresponding to each of the images.
Adopting the Hjellming & Johnston (1981) model for SS433 (or the Gower et al. 1982 model, as applied to AGNs) for a precessing helical jet, we assume that a plasmon in the jet is ejected from the central engine of the source with constant velocity $`\overline{v}`$ along the surface of a cone of half-opening angle $`\varphi `$, the symmetry axis of which is the axis of the jet ($`cf.`$ Fig.2). New plasmons are ejected ballistically from time to time. The plasmon ejection velocity vector precesses with angular velocity $`\mathrm{\Omega }`$ about the cone’s surface, making a constant angle $`\varphi `$ with the jet axis, so that each subsequent plasmon is ejected at a different phase of the jet’s precession. Once ejected with a given velocity $`\overline{v}`$, a plasmon continues to propagate away from the central engine with that velocity (or it fades from view before it appears to decelerate or accelerate). Each plasmon is assumed to consist of an isotropically expanding cloud of synchrotron-emitting electrons, so that the centroid of emission is not affected by its evolution. As mentioned earlier, it is assumed that only one plasmon at a time dominates the unresolved radio emission from the radio core. It is also assumed that the line of sight of the observer coincides with the jet precession axis, for simplicity of calculations.
Let a plasmon be ejected from the central engine of the source. The observable proper motion of the unlensed plasmon from the time of its ejection, $`t_{\mathrm{ej}}`$, to the epoch, $`k`$, of observation, $`t_k`$, would be given by its components in the plane of the sky:
$$\mathrm{\Delta }\mathrm{\Phi }_\alpha =\frac{v_\alpha (t_kt_{\mathrm{ej}})}{d(1v_x/c)}$$
(4)
$$\mathrm{\Delta }\mathrm{\Phi }_\delta =\frac{v_\delta (t_kt_{\mathrm{ej}})}{d(1v_x/c)}$$
(5)
Here $`d`$ is the angular diameter distance to the source, and $`c`$ is the velocity of light. The source time interval $`(t_kt_{\mathrm{ej}})`$ is corrected for the motion of the plasmon at relativistic speeds along the line of sight during its propagation. (To convert a source time interval itself to the observed time interval, $`(\delta t)_{\mathrm{observer}}=(\delta t)_{\mathrm{source}}/(1v_x/c)/(1+z_s)`$, where $`z_s`$ is the redshift of the source; here one corrects for cosmological effects as well. This term is included in the angular diameter distance in Eqns. 4 & 5 and therefore need not be specified explicitly).
At epoch $`t_k`$, we observe in image $`\mathrm{A}`$:
$$\mathrm{\Delta }\mathrm{\Phi }_{\alpha ,\mathrm{A}}[t_k]=a_1\mathrm{\Delta }\mathrm{\Phi }_\alpha ^\mathrm{A}+p\mathrm{\Delta }\mathrm{\Phi }_\delta ^\mathrm{A}$$
$$\mathrm{\Delta }\mathrm{\Phi }_{\delta ,\mathrm{A}}[t_k]=p\mathrm{\Delta }\mathrm{\Phi }_\alpha ^\mathrm{A}+a_4\mathrm{\Delta }\mathrm{\Phi }_\delta ^\mathrm{A}$$
where the superscript ‘$`𝐀`$’ is used to indicate that an appropriate correction for the light travel time along the path through image $`\mathrm{A}`$ must be applied. For image $`\mathrm{B}`$:
$$\mathrm{\Delta }\mathrm{\Phi }_{\alpha ,\mathrm{B}}[t_k]=b_1\mathrm{\Delta }\mathrm{\Phi }_\alpha ^\mathrm{B}+q\mathrm{\Delta }\mathrm{\Phi }_\delta ^\mathrm{B}$$
$$\mathrm{\Delta }\mathrm{\Phi }_{\delta ,\mathrm{B}}[t_k]=q\mathrm{\Delta }\mathrm{\Phi }_\alpha ^\mathrm{B}+b_4\mathrm{\Delta }\mathrm{\Phi }_\delta ^\mathrm{B}$$
with a similar use of superscript as before. The convention that will be used in the following is that for events in image $`\mathrm{A}`$, $`(t_kt_{\mathrm{ej}})`$ is written as $`(t_k{}_{\mathrm{A}}{}^{}t_{\mathrm{ej}}^{})`$, and for image $`\mathrm{B}`$, as $`(t_k{}_{\mathrm{B}}{}^{}t_{\mathrm{ej}}^{})=(t_k{}_{\mathrm{A}}{}^{}t_{\mathrm{ej}}^{}\tau )`$, explicitly writing the light travel time difference $`\tau `$ between the paths through images $`\mathrm{A}`$ and $`\mathrm{B}`$ into the expressions.
At this point contact is possible with the observations in Figure 1. What is observed in each image is a change $`\overline{\mathrm{\Delta }\mathrm{\Phi }}`$ from a location $`\overline{\theta }=(\theta _\alpha ,\theta _\delta )`$, the position of the lensed central engine. It is assumed that $`\overline{\theta }`$ for each image remains fixed through the different epochs of observation; since the changes in Figure 1 are plotted relative to a reference epoch, the location $`\overline{\theta }`$ in each image will drop out of all expressions thereafter, hence it will be ignored.
With this simplification, and making the appropriate substitutions, the following quantities are constructed:
$`\delta \mathrm{\Phi }_\alpha [t_k]`$ $`=`$ $`\mathrm{\Delta }\mathrm{\Phi }_{\alpha ,\mathrm{A}}[t_k]\mathrm{\Delta }\mathrm{\Phi }_{\alpha ,\mathrm{B}}[t_k]`$ (6)
$`=`$ $`\{(a_1b_1)v_\alpha ^{}+(pq)v_\delta ^{}\}(t_k{}_{\mathrm{A}}{}^{}t_{\mathrm{ej}}^{})+`$
$`(b_1v_\alpha ^{}+qv_\delta ^{})\tau `$
$`\delta \mathrm{\Phi }_\delta [t_k]`$ $`=`$ $`\mathrm{\Delta }\mathrm{\Phi }_{\delta ,\mathrm{A}}[t_k]\mathrm{\Delta }\mathrm{\Phi }_{\delta ,\mathrm{B}}[t_k]`$ (7)
$`=`$ $`\{(pq)v_\alpha ^{}+(a_4b_4)v_\delta ^{}\}(t_k{}_{\mathrm{A}}{}^{}t_{\mathrm{ej}}^{})+`$
$`(qv_\alpha ^{}+b_4v_\delta ^{})\tau `$
In the above, the prime on quantities $`v_\alpha `$ and $`v_\delta `$ is used to denote the corresponding velocity components divided by $`d(1v_x/c).`$ Also note the implicit assumption that at the epoch of observation, $`t_k`$, the plasmon responsible for the observations is present in both images (i.e., $`t_k{}_{\mathrm{A}}{}^{}t_{\mathrm{ej}}^{}+\tau ).`$ As long as only one plasmon at a time dominates the source emission on this scale of observation, and it expands isotropically (so that the centroid of radio emission is not affected by its evolution), it is possible to neglect the effects of evolution of the plasmon itself over the timescale of observation and deal simply with positional changes of its centroid relative to the core. The presence of a second evolving plasmon plasmon coexisting with the first would considerably complicate this simple picture!
The changes plotted in Figure 1 are relative to a reference epoch $`t_{\mathrm{ref}}`$, so we construct new quantities, which are now the abcissas and ordinates of the plot in Figure 1 :
$`\delta (\delta \mathrm{\Phi }_\alpha )[t_k]`$ $`=`$ $`\delta \mathrm{\Phi }_\alpha [t_k]\delta \mathrm{\Phi }_\alpha [t_{\mathrm{ref}}]`$ (8)
$`=`$ $`(a_1b_1)v_\alpha ^{}+(pq)v_\delta ^{}`$
$`\delta (\delta \mathrm{\Phi }_\delta )[t_k]`$ $`=`$ $`\delta \mathrm{\Phi }_\delta [t_k]\delta \mathrm{\Phi }_\delta [t_{\mathrm{ref}}]`$ (9)
$`=`$ $`(pq)v_\alpha ^{}+(a_4b_4)v_\delta ^{}`$
The slope $`m`$ of a particular linear track in Figure 1 is given by (dropping the primes on $`v_\alpha `$ and $`v_\delta `$ as unnecessary here):
$`m`$ $`=`$ $`{\displaystyle \frac{\delta (\delta \mathrm{\Phi }_\alpha )[t_k]\delta (\delta \mathrm{\Phi }_\alpha )[t_j]}{\delta (\delta \mathrm{\Phi }_\delta )[t_k]\delta (\delta \mathrm{\Phi }_\delta )[t_j]}}`$ (11)
$`=`$ $`{\displaystyle \frac{(a_1b_1)v_\alpha +(pq)v_\delta }{(pq)v_\alpha +(a_4b_4)v_\delta }}`$
From the above expressions it is now apparent that the change in slope of the tracks between the run of epochs 2, 3 & 4 (slope $`m_1=0.85\pm 0.07`$) and that of epochs 5, 6, 7 & 8 (slope $`m_2=4.66\pm 0.48`$) is effected by a change in the quantities $`v_\alpha `$ and $`v_\delta `$, projections of the ejection velocity vector $`\overline{v}`$ of a plasmon, in the plane of the sky. In the context of our model, this corresponds to the dominance of a new plasmon between epochs 4 & 5, which has been ejected at a phase in the precession of the jet which is distinct from that of the previously ejected plasmon. An obvious question at this point is, what is the actual event that causes a discontinuity between epochs 1 & 2, and again between epochs 4 & 5? The appearance of a new plasmon in the NE image precedes that in the SW image by 26 days, or nearly two successive epochs in our 43 GHz observations, so if the discontinuity were to be caused by its appearance in the NE image, there should be another such discontinuity two epochs later, caused by its manifestation in the SW image. The fact that both linear tracks are more than two epochs long suggests that this is not the case. The reason is somewhat subtle and will be discussed in Section 3.6 after deriving some quantities in connection with the source-to-image transformation matrices, but for the moment we identify the discontinuities in these linear tracks as corresponding to the appearance of new plasmons in the SW image (so that, by the time a discontinuity occurs, the same source plasmon is imaged in both the NE and the SW regions).
As an aside, notice that for equal intervals between epochs, the changes $`\delta \overline{\delta \mathrm{\Phi }}`$ should be the same, along a single linear track. That this is only approximately true for each of the two tracks in Figure 1 is an indication that emission near the central core is variable at some level, pulling in or pushing out the centroid of radio emission from time to time. This results in an apparent change in speed of the emergent plasmon, with no significant change in direction.
For later use, the above expression is recast in a slightly different form:
$$\frac{v_\delta }{v_\alpha }=\frac{\stackrel{~}{p}(m\stackrel{~}{a_1})(m\stackrel{~}{b}_1)}{\stackrel{~}{p}(1m\stackrel{~}{a}_4)(1m\stackrel{~}{b}_4)}$$
(13)
where $`\stackrel{~}{p}=p/q`$, $`\stackrel{~}{a_1}=a_1/p`$, $`\stackrel{~}{a}_4=a_4/p`$, $`\stackrel{~}{b}_1=b_1/q`$ and $`\stackrel{~}{b}_4=b_4/q`$. For a given value of slope $`m`$, once the lensing$``$related quantities $`\stackrel{~}{p},\stackrel{~}{a_1},\stackrel{~}{a}_4,\stackrel{~}{b}_1\&\stackrel{~}{b}_4`$ are determined, this expression gives the phase of precession of the jet (modulo $`n\pi `$ radians, where $`n`$ is a natural number), relative to a suitably chosen reference direction, at the time of ejection of the relevant plasmon. Thus, the difference of the phases between the events slopes $`m_1`$ and $`m_2`$, over the known difference in epoch between the observations, yields the angular velocity or rate of precession of the jet’s source, $`\mathrm{\Omega }=\dot{\psi }`$:
$$\mathrm{\Omega }=\frac{(\psi _2\psi _1)}{(t_2t_1)},$$
(14)
where:
$$\psi _1=Tan^1\left(\frac{v_\delta }{v_\alpha }[m_1]\right)$$
$$\psi _2=Tan^1\left(\frac{v_\delta }{v_\alpha }[m_2]\right)$$
and $`t_1`$ and $`t_2`$ are just the dates of the first epoch of observation of each event ($`t_1`$ is epoch 2, 2 Feb 1997, and $`t_2`$ is epoch 5, 21 Mar 1997). The uncertainty in the commencement of each event is regarded as being uniformly distributed, between epochs 1 and 2 for $`t_1`$, and between epochs 4 and 5 for $`t_2`$ (see Fig. 1).
### 3.2 The milliarcsecond$``$scale behaviour
At this point, it is necessary to review some puzzling VLBA observations of PKS1830-211 at 43 GHz which preceded the eight-epoch track of observations of Jin et al. (2003), namely, those of Garrett et al.(1998). These observations revealed dramatic changes in the milliarcsecond scale structure of the two image groups between epochs of observation 30 May 1996 and 14 July 1996. The former epoch of observation showed core-knot structures in both the NE and SW images, as for each of the epochs of observation in Jin et al.(2003), but the latter epoch, quite suddenly and surprisingly, exhibited rich structure in each of the image groups, with as many as six features in the NE image group being potentially identifiable with corresponding features in the SW image group (including the basic core-knot structure, $`cf.`$ Fig. 3). By the time of the beginning of Jin et al.’s (2003) observations (19 January 1997), the system was back to its quiescent phase, with only a simple core-knot structure being imaged. At some point, on a timescale of several weeks to half a year, a dramatic outburst of plasmons occured, which then evolved and faded from view, and did not recur over the three months that followed, during which Jin et al.’s (2003) monitoring observations of the system were in progress.
It is expected that at least some of the plasmons in the NE image will have corresponding entities in the SW image. Obviously related are features A1 and B1, the core images, and A4 and B4, which are seen in both epochs of observation in Figure 3 and appear to be static ‘knots’. Of the short-lived features observed in the maps of 14 July 1996, A3 and B3 are obviously not related to each other, from considerations of image parity. However, A2 and B2, A5 and B5, and A6 and B6 are easily seen to be corresponding images, once the effects of the time delay between the NE and SW images (recalling that phenomena in the NE group of images occur about 26 days earlier than they show up in the SW image group), and the distorting effects of lensing in the two image groups are accounted for. It is conjectured at this point that feature A3 has not as yet manifest a counterpart in the SW group of images (it is younger than 26 days), and feature B3 is so aged that its counterpart in the NE group has already evolved below the map detection limits.
### 3.3 Deriving the relative (SW2NE) image transformation matrix
Assuming that the source-to-image linear transformation matrices remain locally valid across the entire NE or SW image group (on scales of order 1 mas), and that each plasmon in the source is moving ballistically away from the core, on a straight line trajectory, it proves feasible to derive a relative (image-to-image) $`2\times 2`$ transformation matrix from the observations of 14 July 1996 shown in Figure 3.
Recasting the relative image transformation matrix in the notation of Equation 13,
$`(\mathrm{\Delta }\mathrm{\Phi }_{\alpha ,\mathrm{A}},\mathrm{\Delta }\mathrm{\Phi }_{\delta ,\mathrm{A}})`$ $`=`$ $`\left(\begin{array}{cc}T_1& T_2\\ T_3& T_4\end{array}\right)\left(\begin{array}{c}\mathrm{\Delta }\mathrm{\Phi }_{\alpha ,\mathrm{B}}\\ \mathrm{\Delta }\mathrm{\Phi }_{\delta ,\mathrm{B}}\end{array}\right),`$ (19)
with $`T_1=\mathrm{}(\stackrel{~}{a_1}\stackrel{~}{b}_41)`$, $`T_2=\mathrm{}(\stackrel{~}{b}_1\stackrel{~}{a_1})`$, $`T_3=\mathrm{}(\stackrel{~}{b}_4\stackrel{~}{a}_4)`$ and $`T_4=\mathrm{}(\stackrel{~}{a}_4\stackrel{~}{b}_11)`$, and $`\mathrm{}=\stackrel{~}{p}/(\stackrel{~}{b}_1\stackrel{~}{b}_41)`$.
For a given feature $`i`$, its separation from the core position will be denoted by the vector $`(\mathrm{\Delta }\mathrm{\Phi }_{\alpha ,\mathrm{A}_\mathrm{i}},\mathrm{\Delta }\mathrm{\Phi }_{\delta ,\mathrm{A}_\mathrm{i}})`$. From the static features, the cores A1 and B1, and knots A4 and B4, one derives the core-knot separation vector in each image group, yielding two constraints (positional coordinates) on the relative transformation matrix (from Eqn. 19):
$$T_1\mathrm{\Delta }\mathrm{\Phi }_{\alpha ,\mathrm{B}_4}+T_2\mathrm{\Delta }\mathrm{\Phi }_{\delta ,\mathrm{B}_4}\mathrm{\Delta }\mathrm{\Phi }_{\alpha ,\mathrm{A}_4}=0$$
(20)
$$T_3\mathrm{\Delta }\mathrm{\Phi }_{\alpha ,\mathrm{B}_4}+T_4\mathrm{\Delta }\mathrm{\Phi }_{\delta ,\mathrm{B}_4}\mathrm{\Delta }\mathrm{\Phi }_{\delta ,\mathrm{A}_4}=0$$
(21)
The pairs of relationships between A2 and B2, A5 and B5, and A6 and B6 each yield a single constraint, from the shared direction of motion of the plasmon in the source (these are evolving features and their actual locations in the NE and SW images refer back to different source times due to the relative time delay). That is, given the location of, say, plasmon B2 in the the SW group relative to its core image B1, the relative transformation matrix must be such that it reproduces the correct direction for plasmon A2 relative to its core A1 in the NE group of images.
For features A2 and B2, A5 and B5, and A6 and B6, the single constraint equation from each pair reads (with $`i=2,5,6)`$:
$$\frac{\mathrm{\Delta }\mathrm{\Phi }_{\delta ,\mathrm{A}_\mathrm{i}}}{\mathrm{\Delta }\mathrm{\Phi }_{\alpha ,\mathrm{A}_\mathrm{i}}}=\frac{T_3\mathrm{\Delta }\mathrm{\Phi }_{\alpha ,\mathrm{B}_\mathrm{i}}+T_4\mathrm{\Delta }\mathrm{\Phi }_{\delta ,\mathrm{B}_\mathrm{i}}}{T_1\mathrm{\Delta }\mathrm{\Phi }_{\alpha ,\mathrm{B}_\mathrm{i}}+T_2\mathrm{\Delta }\mathrm{\Phi }_{\delta ,\mathrm{B}_\mathrm{i}}}$$
which may be rewritten as:
$`T_1\mathrm{\Delta }\mathrm{\Phi }_{\alpha ,\mathrm{B}_\mathrm{i}}\mathrm{\Delta }\mathrm{\Phi }_{\delta ,\mathrm{A}_\mathrm{i}}+T_2\mathrm{\Delta }\mathrm{\Phi }_{\delta ,\mathrm{A}_\mathrm{i}}\mathrm{\Delta }\mathrm{\Phi }_{\delta ,\mathrm{B}_\mathrm{i}}`$
$`T_3\mathrm{\Delta }\mathrm{\Phi }_{\alpha ,\mathrm{A}_\mathrm{i}}\mathrm{\Delta }\mathrm{\Phi }_{\alpha ,\mathrm{B}_\mathrm{i}}T_4\mathrm{\Delta }\mathrm{\Phi }_{\alpha ,\mathrm{A}_\mathrm{i}}\mathrm{\Delta }\mathrm{\Phi }_{\delta ,\mathrm{B}_\mathrm{i}}`$ $`=0`$ (22)
Thus the observations yield, through Equations 20, 21 and 22, 5 constraint equations for the 4 unknown relative transformation matrix elements, $`T_j,j=1`$ to $`4`$. Note that the unknown matrix elements and the observations determining them are implicitly related, non-linear in the observational quantities, and of the form: $`𝐅(𝐗_a,𝐋_a)=\mathrm{𝟎}`$, where $`𝐗_a`$ is the vector of 4 unknowns (matrix elements) and $`𝐋_a`$ is the vector of 16 observational quantities. $`𝐅`$ itself consists of 5 nonlinear functions. The unknowns in vector X are solved for via a mixed adjustment model (see, for example, A. Leick, 1995), the solution of which is described in the Appendix.
The solution for the elements of the SW2NE transformation matrix are:
$`𝐓_{SW2NE}=`$ $`\left(\begin{array}{cc}1.279& 0.871\\ 0.973& 0.245\end{array}\right)`$ (25)
Its determinant is: $`1.161.`$ This is the NE/SW image flux density ratio, $`k=T_1T_4T_2T_3`$. The covariance (error) matrix is given by:
$`𝐂_𝐓=`$ $`\left(\begin{array}{cccc}0.071& 0.031& 0.106& 0.053\\ 0.031& 0.017& 0.054& 0.026\\ 0.106& 0.054& 0.342& 0.165\\ 0.053& 0.026& 0.165& 0.081\end{array}\right)`$ (30)
From the expressions immediately following Equation 19, it is possible to express the quantities $`\stackrel{~}{a_1}`$, $`\stackrel{~}{a}_4`$, $`\stackrel{~}{b}_1`$ and $`\stackrel{~}{b}_4`$ in terms of $`T_1`$, $`T_2`$, $`T_3`$ and $`T_4`$ and $`\stackrel{~}{p}`$.
$$\stackrel{~}{a_1}=(T_1k/\stackrel{~}{p})/T_3$$
(31)
$$\stackrel{~}{a}_4=(T_4k/\stackrel{~}{p})/T_2$$
(32)
$$\stackrel{~}{b}_1=(\stackrel{~}{p}T_4)/T_3$$
(33)
$$\stackrel{~}{b}_4=(\stackrel{~}{p}T_1)/T_2$$
(34)
The evaluation of these quantities depends on the value of $`\stackrel{~}{p}`$, for which straightforward algebraic methods of determination exist, but which were found to be sensitive to numerical errors of estimate. Hence a different approach has been taken, which will be discussed in Section 3.5.
### 3.4 The directions of ejection for the milliarcsecond-scale features
It is possible, under the explicit assumption that the source velocity for each evolving plasmon remains constant from the time of generation at least upto the epoch represented by the observations of 14 July 1996 in Figure 3, to relate the distance travelled by a given plasmon as seen in the NE image group of that observation, with the distance that is travelled by its corresponding feature, transformed from the SW group of images via the relative image transformation matrix $`𝐓_{SW2NE}`$ and corrected for the lag in time of 26 days for the SW group.
That is, the squared distances are (for $`i=2,5,6`$):
$$\mathrm{D}_{NE}^2=\mathrm{\Delta }\mathrm{\Phi }_{\alpha ,\mathrm{A}_\mathrm{i}}^2+\mathrm{\Delta }\mathrm{\Phi }_{\delta ,\mathrm{A}_\mathrm{i}}^2,$$
(35)
in the NE group of images, and transformed from the SW group to the NE group, are:
$`\mathrm{D}_{SW2NE}^2=`$ $`(T_1\mathrm{\Delta }\mathrm{\Phi }_{\alpha ,\mathrm{B}_\mathrm{i}}+T_2\mathrm{\Delta }\mathrm{\Phi }_{\delta ,\mathrm{B}_\mathrm{i}})^2+`$
$`(T_3\mathrm{\Delta }\mathrm{\Phi }_{\alpha ,\mathrm{B}_\mathrm{i}}+T_4\mathrm{\Delta }\mathrm{\Phi }_{\delta ,\mathrm{B}_\mathrm{i}})^2`$
In the NE group, the velocity of motion is $`\mathrm{D}_{NE}/(t_kt_{ej})`$, and, transformed from the SW group, the corresponding velocity of motion is $`\mathrm{D}_{SW2NE}/(t_kt_{ej}\tau )`$, where $`t_k`$ is the epoch of observation and $`\tau `$ is the time delay between the image groups (assumed to be constant over the span of about a milliarcsecond from the core). If $`𝒬`$ represents the ratio $`\mathrm{D}_{NE}/\mathrm{D}_{SW2NE}`$, then:
$$t_kt_{ej}=\frac{\tau 𝒬}{𝒬1},$$
(37)
with $`𝒬>1.`$
In order to estimate how well the mixed adjustment model-derived relative transformation matrix actually reproduces the position angles of the features in the NE group of images when it is applied to data from the SW group, the quantities $`\mathrm{\Delta }\mathrm{\Phi }_{\delta ,\mathrm{A}_\mathrm{i}}/\mathrm{\Delta }\mathrm{\Phi }_{\alpha ,\mathrm{A}_\mathrm{i}}`$ (for the original position angle of feature $`i`$ in the NE group), and $`(T_3\mathrm{\Delta }\mathrm{\Phi }_{\alpha ,\mathrm{B}_\mathrm{i}}+T_4\mathrm{\Delta }\mathrm{\Phi }_{\delta ,\mathrm{B}_\mathrm{i}})/(T_1\mathrm{\Delta }\mathrm{\Phi }_{\alpha ,\mathrm{B}_\mathrm{i}}+T_2\mathrm{\Delta }\mathrm{\Phi }_{\delta ,\mathrm{B}_\mathrm{i}})`$ (for the position angle of the transformed data relating to feature $`i`$ from the SW image) are calculated. For feature A2 (B2), the original position angle is $`48^\mathrm{o}`$ versus a transformed value of $`30^\mathrm{o}`$ but $`𝒬`$ is $`<1`$ (pointing to some inconsistency for this feature); for feature A5 (B5), the original position angle is $`104^\mathrm{o}`$ as compared with a transformed value of $`113^\mathrm{o}`$, and $`𝒬`$ is 1.272; from Equation 37, a value of 125.2 days is obtained as the ‘age’ of the feature at the time of observation in the NE image on 14 July 1996. For feature A6 (B6), the original position angle is $`159^\mathrm{o}`$, and the transformed position angle is $`158^\mathrm{o}`$; $`𝒬`$ is 6.255 and the ‘age’ of the feature as seen in the NE image is 30.9 days.(Owing to the existence of significant cross-correlations between quantities derived through the mixed adjustment model – $`cf.`$ Equation 30 – formal errors on quantities derived will be given only for the final result). A cross-check on the mapping of the static knot reveals a perfect match for position. The position angle is $`99^\mathrm{o}`$ in the original; transformed, it is $`99^\mathrm{o}`$, and as to the distance from the core, it is 0.99 mas in the original and 0.99 mas in the transformed version. Thus the transformation matrix is heavily influenced by feature A6(B6) and the the location of the static knot feature.
Of course, in order to derive the rate of precession of the jet from these numbers, it is necessary to consider the mappings in the source plane of these features, rather than those in the NE image. For this, it is necessary to calculate the velocities of the various evolving features in the source plane. The elements of the source-to-image transformation matrix in Equation 1 can be rewritten, dividing each element by $`p`$. Then its inverse reads:
$$𝐓_{S2NE}^1=p^2(\stackrel{~}{a_1}\stackrel{~}{a}_41)^2\left(\begin{array}{cc}\stackrel{~}{a}_4& 1\\ 1& \stackrel{~}{a_1}\end{array}\right)$$
(38)
Similarly, for Equation 2, the matrix can be recast and inverted to yield:
$$𝐓_{S2SW}^1=q^2(\stackrel{~}{b}_1\stackrel{~}{b}_41)^2\left(\begin{array}{cc}\stackrel{~}{b}_4& 1\\ 1& \stackrel{~}{b}_1\end{array}\right)$$
(39)
Taking the time-derivative (denoted by a superscribed dot) of the angular separations involved, so that one is now dealing with velocities,
$`\dot{\mathrm{\Delta }\mathrm{\Phi }}_\alpha =`$ $`p^1(\stackrel{~}{a_1}\stackrel{~}{a}_41)^1\left(\stackrel{~}{a}_4\dot{\mathrm{\Delta }\mathrm{\Phi }}_{\alpha ,A}\dot{\mathrm{\Delta }\mathrm{\Phi }}_{\delta ,A}\right)`$ (40)
$`=`$ $`q^1(\stackrel{~}{b}_1\stackrel{~}{b}_41)^1\left(\stackrel{~}{b}_4\dot{\mathrm{\Delta }\mathrm{\Phi }}_{\alpha ,B}\dot{\mathrm{\Delta }\mathrm{\Phi }}_{\delta ,B}\right)`$ (41)
$`\dot{\mathrm{\Delta }\mathrm{\Phi }}_\delta =`$ $`p^1(\stackrel{~}{a_1}\stackrel{~}{a}_41)^1\left(\stackrel{~}{a_1}\dot{\mathrm{\Delta }\mathrm{\Phi }}_{\delta ,A}\dot{\mathrm{\Delta }\mathrm{\Phi }}_{\alpha ,A}\right)`$ (42)
$`=`$ $`q^1(\stackrel{~}{b}_1\stackrel{~}{b}_41)^1\left(\stackrel{~}{b}_1\dot{\mathrm{\Delta }\mathrm{\Phi }}_{\delta ,B}\dot{\mathrm{\Delta }\mathrm{\Phi }}_{\alpha ,B}\right)`$ (43)
Now, from the inverse of the image-to-image transformation matrix,
$$\dot{\mathrm{\Delta }\mathrm{\Phi }}_{\alpha ,B}=k^1(T_4\dot{\mathrm{\Delta }\mathrm{\Phi }}_{\alpha ,A}T_2\dot{\mathrm{\Delta }\mathrm{\Phi }}_{\delta ,A})$$
(44)
$$\dot{\mathrm{\Delta }\mathrm{\Phi }}_{\delta ,B}=k^1(T_1\dot{\mathrm{\Delta }\mathrm{\Phi }}_{\delta ,A}T_3\dot{\mathrm{\Delta }\mathrm{\Phi }}_{\alpha ,A})$$
(45)
so that, substituting for $`\stackrel{~}{a_1}`$ and $`\stackrel{~}{a}_4`$, or $`\stackrel{~}{b}_1`$ and $`\stackrel{~}{b}_4`$ from Equations 31 to 34 and with a little algebra, one obtains:
$$\frac{\dot{\mathrm{\Delta }\mathrm{\Phi }}_\delta }{\dot{\mathrm{\Delta }\mathrm{\Phi }}_\alpha }=\frac{T_2}{T_3}\frac{(\stackrel{~}{p}\dot{\mathrm{\Delta }\mathrm{\Phi }}_{\delta ,\mathrm{B}}\dot{\mathrm{\Delta }\mathrm{\Phi }}_{\delta ,\mathrm{A}})}{(\stackrel{~}{p}\dot{\mathrm{\Delta }\mathrm{\Phi }}_{\alpha ,\mathrm{B}}\dot{\mathrm{\Delta }\mathrm{\Phi }}_{\alpha ,\mathrm{A}})}=\frac{v_\delta }{v_\alpha }$$
(46)
This expression yields (as in the microarcsecond-scale analysis discussed earlier) the phase (modulo $`n\pi `$) of ejection of the relevant plasmon, with a suitable choice of reference direction. The quantities $`\dot{\mathrm{\Delta }\mathrm{\Phi }}_{\alpha ,\mathrm{B}}`$ and $`\dot{\mathrm{\Delta }\mathrm{\Phi }}_{\delta ,\mathrm{B}}`$ are obtained from $`\mathrm{\Delta }\mathrm{\Phi }_{\alpha ,\mathrm{B}}`$ and $`\mathrm{\Delta }\mathrm{\Phi }_{\delta ,\mathrm{B}}`$ by dividing by $`(t_kt_{ej}\tau )`$, and $`\dot{\mathrm{\Delta }\mathrm{\Phi }}_{\alpha ,\mathrm{A}}`$ and $`\dot{\mathrm{\Delta }\mathrm{\Phi }}_{\delta ,\mathrm{A}}`$ are obtained by dividing $`\mathrm{\Delta }\mathrm{\Phi }_{\alpha ,\mathrm{A}}`$ and $`\mathrm{\Delta }\mathrm{\Phi }_{\delta ,\mathrm{A}}`$ by $`(t_kt_{ej})`$. To estimate the rate of precession of the jet from expression 46, the phases of ejection, $`\psi _j`$, are calculated for plasmons $`j`$ (with $`j=5,6`$) and the observed time that has lapsed between the ejection of plasmon 5 and plasmon 6 is calculated as $`(t_kt_{ej,6})(t_kt_{ej,5})`$, where $`t_k`$, the shared epoch of observation, drops out of the estimate. The rest of the calculation follows along the lines described for the microarcsecond scale analysis.
### 3.5 Evaluating the period of precession of the jet from milliarcsecond and microarcsecond scale information
Expressions 13 and 46 are, respectively, the phases of ejection of various plasmons on the microarcsecond and milliarcsecond scales, as derived from observables. Note that both expressions require the evaluation of $`\stackrel{~}{p}`$. In order to do this, it will be assumed that the period of precession of the relativistic jet remains the same when evaluated from data on the scale of tens of microarcseconds, as from data on the scale of milliarseconds (for a jet opening half-angle of, say, $`0^\mathrm{o}.5`$ and assuming a typical linear magnification by the gravitational lens by a factor of 3, this difference of scale corresponds to source-intrinsic times of order $`10^3`$ years, too short to expect a significant evolution in the precession velocity of the jet). Hence a range of values of $`\stackrel{~}{p}`$ is considered, and its actual value is determined from the coincidence on these two scales of calculated values of the rate of precession of the jet, which exercise also determines the rate of precession itself.
The rate of precession is evaluated as that which would have been observed in the absence of the lens, rather than that intrinsic to the source, since the component of the bulk velocity of the jet along the line of sight cannot be determined (but is expected to be close to the velocity of light). It is calculated as described in Sections 3.1 and 3.4, over a range of values $`\stackrel{~}{p}=p/q`$. Recalling that the actual phases of ejection are only determined modulo $`n\pi `$, there is a degeneracy in determining the difference of phases $`\psi _2\psi _1`$ for both the milliarcsecond and the microarcsecond analyses. This means that each separation $`\psi _i\psi _j`$ actually corresponds to four possible branches. Each of these branches is plotted against $`\stackrel{~}{p}`$ in Figure 4. At this point, it is apparent that a value of $`\stackrel{~}{p}`$ of $`0.3`$, and a precession rate of $`0.016`$ radians/day (corresponding to an observed period of 1.08 years) permit a coincidence of the milliarcsecond and microarcsecond scale results, so these values are adopted. To obtain the intrinsic source rotation period, the observed period must be corrected for the cosmological time dilation on account of the source being at a redshift of 2.5, and for the change in time interval on account of the jet velocity in the source. The first correction reduces the period by a factor of 3.5, and the second increases it by a factor of $`1/(1\beta _{los})`$, where $`\beta _{los}=v_x/c`$, with c as the velocity of light. For small jet opening angles, $`\beta _{los}v_{jet}/c`$. Assuming a value of $`\beta =0.99`$, the intrinsic period is calculated to be 30.8 years.
Deriving a period of 1.08 years from the observations raises a question about the authenticity of the results obtained in this paper. We have checked carefully for the possibility of effects related to the annual motion of the Earth around the Sun. The paper of Jin et al. (2003), from which the microarcsecond scale observations are drawn, included a correction for the differential annual aberration between the NE and SW groups of images (separated by $`\mathrm{\hspace{0.33em}1}`$ arcsec), which is typically of the order of tens of microarcseconds. Relative annual parallax between the NE and SW core images is of order $`10^4`$ $`\mu `$as, too small to produce a spurious effect in the observations. (As an aside, another possible source of lensing capable of causing a differential shift in the location of the NE core image relative to the SW one is the Sun itself, which was always more than 15 degrees away from PKS1830-211 in the observations of Jin et al. (2003), thus producing a differential shift in the NE core image relative to the SW core image of much less than a microarcsecond). Another reason that we do not consider our present result to be a consequence of the Earth’s motion around the Sun is the coincidence of this result as obtained from the milliarcsecond scale data of Garrett et al. (1998) with that obtained from the microarcsecond scale data of Jin et al. (2003), as shown in Figure 4. It is difficult to see how the annual motion of the Earth could arrange for this, especially as the milliarcsecond scale data is just one actual epoch of observation (which effectively includes two epochs for the source only because of the existence of two groups of lensed images mapping back to source events separated in time by the time delay of 26 days).
The error in evaluating the precession period is estimated via a Monte Carlo exercise that generates, for each of 60,000 experiments, a random selection of all the observational errors, taking into account the cross-correlations between elements of the SW2NE image-to-image transformation matrix, and then computes the desired precession rate for a value of $`\stackrel{~}{p}=p/q`$ of 0.3. For the microarcsecond scale analysis, the precession rate (in radians per day with $`1\sigma `$ errors) is $`0.0155(0.0166,0.0415)`$, and for the milliarcsecond scale analysis, the corresponding values are $`0.0164(0.575,0.987)`$. The large errors on the milliarcsecond scale result are mainly due to the form of the expression used to evaluate the times of ejection for the two plasmons (see Eqn. 37).
The foregoing results are derived, as mentioned earlier, for an assumed perfect alignment of the line of sight with the jet axis. Typically, this is not the case, although it is expected that the alignment of the two is to within a few degrees for blazar sources. The effect of considering an inclination of the jet axis to the line of sight can be estimated from expressions (1), (2) and (3) for the $`x`$, $`y`$ and $`z`$ velocity components of a plasmon in the precessing jet model of Gower et al. (1982), where axes $`z`$ and $`y`$ are in the plane of the sky and centred on the central engine, and axis $`x`$ is along the line of sight. In general, an additional rotation through an angle $`\chi `$ of Gower et al.’s $`y`$ and $`z`$ axes about the $`x`$ axis is necessary to bring the model into alignment with the right ascension and declination axes. Let the jet half-opening angle be $`\varphi `$, and let, as in earlier expressions, the angle through which the jet has precessed at the time of observation, as measured from a reference time $`t_{ref}`$, be $`\psi `$. Then, with $`i`$ as the inclination angle of the jet axis to the line of sight, the inferred value of $`v_\delta /v_\alpha `$ from, $`e.g.`$, Equation 46, is related to the precession angle $`\psi `$ (for each plasmon considered) by:
$$\frac{Sin\chi +(\frac{v_\delta }{v_\alpha })Cos\chi }{Cos\chi (\frac{v_\delta }{v_\alpha })Sin\chi }=\frac{Sin\varphi Sin\psi }{Sin(i)Cos\varphi Cos(i)Sin\varphi Cos\psi }$$
(47)
From the milliarcsecond scale data, for example, with a jet inclination angle of $`0^\mathrm{o}`$, for feature 5, $`\psi =200^\mathrm{o}`$, and $`t_{ej}t_{ref}=125.2`$ days, and for feature 6, the corresponding values are $`69^\mathrm{o}`$ and 30.9 days, yielding, as mentioned above, a precession period intrinsic to the source of 30.8 years. For non-zero values of $`i`$ and $`\varphi `$, in order to obtain the precession rate or period in the source rest frame, the timescales $`t_{ej}t_{ref}`$ must be in each case corrected by a factor of $`1/(1v_x/c)`$, where $`v_x`$ is the line of sight velocity:
$$\frac{v_x}{c}=\beta \left(Sin\varphi Sin(i)Cos\psi +Cos\varphi Cos(i)\right)$$
(48)
With $`i=8^\mathrm{o}`$, $`\varphi =5^\mathrm{o}`$, $`\chi =60^\mathrm{o}`$ and $`\beta =0.99`$, the precession period in the source frame is 21 years. A choice of $`i=4.^\mathrm{o}5`$ and $`\varphi =3^\mathrm{o}`$ with $`\chi `$ and $`\beta `$ as before, yields a precession period of about 25 years.
### 3.6 Single plasmon events on the microarcsecond scale
From Table 1 of Jin et al. (2003), the ratio of the NE to SW core image flux densities is seen to vary as a function of observational epoch ($`cf.`$ also the crosses plotted in Fig. 5). The variation shows two apparent maxima, indicating source variability (although variations on the timescales of a couple of weeks are possible with a combination of an apparently superluminal source and microlensing by stars in the lens galaxy — $`cf.`$ Gopal-Krishna & Subramanian 1991 — it will be post-justified that this is not the case in the present circumstance). It is tempting to model these variations in terms of the emission of synchrotron self-absorbed plasmons. Following Expression 11 of van der Laan (1966), a series of four succesively emitted plasmons is considered, expanding with constant velocity, with an initial spectral maximum at 43 GHz, and electron power law index of 3/2. Since the model is actually underconstrained by the observations, it is only attempted to seek a consistency between the two. For each image, the emission from the series of four plasmons is computed. In the case of the SW image, this curve is scaled by the NE/SW image flux density ratio $`k=1.161`$, derived from Section 3.3, and delayed in time relative to the NE curve by the time delay of 26 days. The middle two plasmons are constrained to have an observed separation, in terms of time of ejection, of $`(125.230.9)=94.3`$ days, as inferred from Section 3.4. It is found necessary to introduce a constant core flux density, of value comparable to that of the plasmons, to simulate the data. The result of this exercise is shown in Figure 5. Note that corresponding to Epoch 2, the second plasmon has just begun to initiate a rise in the SW curve, and again at Epoch 5, the same phenomenon is observed. Hence the discontinuities in Figure 1 are seen to be a consequence of the observed appearance of a new plasmon in the SW image, as was stated in Section 3.1.
## 4 A massive binary black hole system for a central engine?
A popular model for Active Galactic Nuclei (AGNs) with precessing jets is that they host at their centres massive binary black hole systems (Begelman, Blandford & Rees 1980, Roos 1988, Roos, Kaastra & Hummel 1993, Villata et al. 1998, Villata & Raiteri 1999, Romero et al. 2000, Romero, Fan & Nuza 2003, Rieger 2004, Maness et al. 2004). The system under study in the present paper lends itself to such a possibility. The precession period intrinsic to PKS1830-211 has been derived from Section 3.5 to be about 30.8 years; as was the case for 1928+738 (with a jet precession period of 2.9 years; Roos et al. 1993) and 3C 273 (with an observed jet precession period of 16 years; Romero et al. 2000), this is too rapid to be due to geodetic precession of the jet-emitting black hole (Begelman, Blandford & Rees 1980), without suggesting a lifetime ($`10^3`$ years) for gravitational collapse of the binary system that is too short. One may infer, however, that the precession period is actually a result of the orbital motion of the black hole binary (as did Roos et al. 1993). In this case, the jet velocity is modulated by the orbital motion of the jet-emitting black hole. If $`P`$ is the precession period, $`G`$ the gravitational constant, and $`m`$ and $`M`$ the masses of the two black holes, then, with $`r`$ as their separation from each other:
$$r^3=\frac{P^2}{4\pi ^2}G(m+M)$$
(49)
For a precession period in units of 30.8 years, and scaling the masses by $`10^8\mathrm{M}_{}`$ and the separation by $`10^{16}`$ cm,
$$r_{16}=6.833(P_{30.8})^{2/3}(m_8+M_8)^{1/3}\mathrm{cm}$$
(50)
The gravitational lifetime of the system is given by:
$$t_{grav}=2.9\times 10^5\frac{\frac{M_8}{m_8}r_{16}^4}{(1+\frac{m_8}{M_8})M_8^3}\mathrm{yrs}$$
(51)
In this case, using Equation 50, the lifetime of the system until gravitational collapse is:
$$t_{grav}=6.322\times 10^8\frac{(M_8+m_8)^{1/3}}{m_8M_8}\mathrm{yrs}$$
(52)
The motion of the core of the jet-emitting black hole, orbiting around its companion, would be on the scale of microarcseconds, undetectable in the 43 GHz observations used in this work.
## 5 Discussion and Conclusions
The observations of Jin et al.(2003) permit an analysis of the blazar source of PKS 1830-211 under atypical circumstances. Firstly, changes in the core of the source on scales of tens of microarcseconds could be probed because of the fact that the source is doubly imaged within a single VLBI field of view, and hence uncertainties in the separation of the two core images could be restricted to essentially thermal noise limits. Secondly, there is some degree of magnification due to lensing, which, while its numerical value cannot be determined without the existence of a standard candle/ruler within the source, still provides a closer look than might otherwise have been possible (a typical linear magnification factor would be about 3). Thirdly, this lensed system permits a detailed study of a very distant blazar (redshift $`z_s=2.51`$). These observations, coupled with data from observed changes in the milliarcsecond scale of structure, make it possible to estimate the precession period of the relativistic jet in the blazar source in PKS1830-211. The observed period turns out to be 1.08 years, assuming the jet precession axis to be perfectly aligned with the line of sight to the source. This is one of the highest redshifts for which a blazar jet’s precession period has been actually measured. For PKS1830-211, the typically assumed mass of $`10^8\mathrm{M}_{}`$ or less for each element of the central powering black hole system appears to pose no particular evolutionary problems at this redshift ($`cf.`$, e.g., Yoo & Miralda-Escudé 2004).
Estimating the mass of a possible central black hole system from a measured SED is confounded by the fact that PKS1830-211 is seen through the Galactic plane and is subject to much obscuration in the optical. Moreover, it is lensed, so there remains the possibility of spectral changes due to different scales of the source being magnified to different extents on account of microlensing by stars in the lens galaxy. However, a method such as measuring the timescale of variability of gamma ray emission from the source can be used. Although two timetracks, one time-delayed with respect to the other by 26 days, would be observed together, the timescale of variability is expected to be typically of the order of a day or less (Liang & Liu 2003), making it unnecessary to correct for the time delayed track. Then, following the analysis of Liang & Liu (2003), an upper limit to the mass of the jet-emitting black hole can be derived. (Note that these authors do in fact list a black hole mass of $`10^{9.2}\mathrm{M}_{}`$ and a calculated minimum variability timescale of $`10^{4.5}`$ s in Table 1 of their paper, for the system 1830-210, based on an incorrect source redshift of 1 and the observed EGRET gamma ray flux, which would need to be corrected for lensing; the absolute magnification factor is an unknown. If it is taken as a typical value of 10, one obtains a black hole mass of $`10^{8.7}\mathrm{M}_{}`$ for the source at the observed redshift of 2.51; note also that this mass is determined under the assumption that the gamma ray flux is isotropic or unbeamed. This mass corresponds to a gravitational lifetime of the binary black hole system of order $`10^8`$ years ($`cf.`$ Figure 6)).
In passing, we note that it has been assumed in the present work that the plasmons follow straight line paths along the surface of the cone formed by the jet’s precessing nozzle (following from ballistic ejection of plasmons). On the milliarcsecond scale, the essentially two epochs’ worth of data on the source represented by the single observation of 14 July 1996 (Garrett et al. 1997), which has been used here, permits no investigation of a possible curvature in plasmon motion. Multiepoch radio VLBI studies at, say, 43 GHz, during an active phase for the system, should be able to shed some light on the question of whether we are really seeing ballistically ejected plasmons, or whether the helical appearance of the jet is accompanied by motion of the observed features along helical paths, in which case by the time the jet has reached milliarcsecond scales, observational evidence of curvature in the motion of some features might be expected. More sensitive than the observed motions of individual features perhaps would be information from the observed light curves in each image. Features moving along helical paths would show variable relativistic beaming if the jet axis is even slightly inclined to the line of sight ($`cf.`$, e.g., Schramm et al. 1993), exhibiting a quasi-periodicity in the intensity of emission. (Note, however, that the strict linearity apparent in Figure 1 spanning three and four epochs is a more reliable indication that on the scale of tens of microarcseconds at least, the motion of the plasmons in the present study show linear velocities, with little sign of curvature).
Lastly, there is a question regarding the plasmon lifetimes. Are the plasmons we ‘see’ in Jin et al. (2003) on the scale of tens of microarcseconds likely to be the progenitors of features such as have been seen on the scale of milliarcseconds in Garrett et al. (1998)? We think this is unlikely; longer-lived plasmons would not permit a simple analysis such as has been possible in this work, with almost single plasmon events to be observed on the scale of tens of microarcseconds. It would appear that the plasmon emission mechanism is of a highly variable nature, only occasionally producing long-lived plasmons such as were seen in Garrett et al. (1998), but for most part producing relatively short-lived plasmons in a quiescent phase.
## 6 Acknowledgements
SN wishes to acknowledge Dipankar Bhattacharya and Shiv Sethi for being reliable sounding boards, especially with regard to numerical methods, and Vivek Dhawan for raising a couple of critical questions. We thank an anonymous referee for helpful comments on this work, which led to improvements.
## 7 Appendix: The Mixed Adjustment Model
From Section 3.2, the five constraint equations F, Equations 20, 21 and the three implicit in equation 22, link the four unknowns, the transformation matrix elements $`T_i`$ $`(i=1\mathrm{}4)`$ (constituting vector X in the present analysis), with sixteen observational quantities (positional separations, from their respective core images, of the eight features A4(B4), A2(B2), A5(B5) and A6(B6), as seen in both the NE and the SW images. These sixteen quantities constitute the vector L in this section. The positions of the observational features, with errors of fit, are listed in Table 1. The five equations F overdetermine the quantities in X, hence a numerical method, the Mixed Adjustment Model, is employed to solve for X. Note that the observations and the unknowns are implicitly related. The treatment here follows that described in A.Leick (1995), Section 4.4.
Let in the course of the application of this method, the 16 adjusted observations be denoted by L<sub>a</sub>, and the adjusted vector of unknowns be denoted by X<sub>a</sub>. The mathematical model is then:
$$𝐅(𝐋_𝐚,𝐗_𝐚)=0$$
(53)
Let X<sub>0</sub> be a set of known but approximate values of X, and denote the vector of observations by L<sub>b</sub>. Then Equation 53 may be written as:
$$𝐅(𝐋_𝐛+𝐕,𝐗_\mathrm{𝟎}+𝐗)=\mathrm{𝟎},$$
(54)
where V and X are the residuals at each stage of iteration. This essentially nonlinear form is linearized about the point (L<sub>b</sub>, X<sub>0</sub>):
$$_5𝐁_{16}^{}{}_{16}{}^{}𝐕_1+_5𝐀_4{}_{4}{}^{}𝐗_{1}^{}+_5𝐖_1=\mathrm{𝟎},$$
(55)
where:
$${}_{5}{}^{}𝐁_{16}^{}=\frac{𝐅}{𝐋}_{_{(𝐗_0,𝐋_b)}}$$
(56)
$${}_{5}{}^{}𝐀_{4}^{}=\frac{𝐅}{𝐗}_{_{(𝐗_0,𝐋_b)}}$$
(57)
$${}_{5}{}^{}𝐖_{1}^{}=𝐅(𝐋_𝐛,𝐗_\mathrm{𝟎})$$
(58)
Now a least-squares estimate of X is sought, with the target function V<sup>T</sup>PV, where P is the $`16\times 16`$ weight matrix constructed from the covariance matrix of the observations, $`\sigma _o^2\mathrm{\Sigma }_{\mathrm{L}_\mathrm{b}}^1`$, where $`\sigma _o^2`$ is the a priori variance of unit weight. The target function is minimized subject to the constraints imposed by the linearized mathematical model, Equation 55, with a vector of Lagrangian multipliers, K. Hence, one minimizes with respect to V, K and X, the function:
$$\varphi (𝐕,𝐊,𝐗)=𝐕^T\mathrm{𝐏𝐕}2𝐊^T(\mathrm{𝐁𝐕}+\mathrm{𝐀𝐗}+𝐖),$$
(59)
and the three constraint equations arising from this process are used to calculate V,K and X. Since the actual mathematical model is non-linear, it is necessary to iterate to a solution the actual values of V, K and X. The iterative process is said to converge if:
$$(𝐕^T\mathrm{𝐏𝐕})_i(𝐕^T\mathrm{𝐏𝐕})_{(i1)}<ϵ,$$
(60)
where $`ϵ`$ is a small positive number (here taken as $`10^8`$). In the present case, the calculation converged in five iterations, with a final value of V<sup>T</sup>PV=9. The estimated a posteriori variance of unit weight, $`\sigma _o^2`$, is therefore V<sup>T</sup>PV/(5-4)=9. |
warning/0507/math0507242.html | ar5iv | text | # Addition and subspace theorems for asymptotic large inductive dimension
|
warning/0507/hep-th0507256.html | ar5iv | text | # 1 Introduction
## 1 Introduction
The cosmological implications of superstring theory have been under attentive consideration in the last few years from various viewpoints . This involves the classification and the study of possible time-evolving string backgrounds which amounts to the construction, classification and analysis of supergravity solutions depending only on time or, more generally, on a low number of coordinates including time. In this context a quite challenging and potentially highly relevant phenomenon for the overall interpretation of extra–dimensions and string dynamics is provided by the so named cosmic billiard phenomenon ,,,, ,. This is based on a profound link between the features of time evolution of the cosmological scale factors and the algebraic structure of string theory duality groups. As it is well known, the dualities that unify the various perturbative quantum string models into a unique M-theory are elements of a unified group $`\mathrm{U}()`$ which is the suitable restriction to integers of a corresponding Lie group $`\mathrm{U}()`$ encoded in the low energy limit of superstrings, namely supergravity. The group $`\mathrm{U}\mathrm{U}()`$ appears as isometry group of the scalar manifold $`_{scalar}`$ emerging in compactifications of $`10`$–dimensional supergravity to lower dimensions and crucially depends on the geometry of the compact dimensions and on the number of preserved supersymmetries $`N_Q32`$. For $`N_Q>8`$ the scalar manifold is always a homogeneous space $`\mathrm{U}/\mathrm{H}`$ and what actually happens is that the cosmological scale factors $`a_i(t)`$ associated with the various dimensions of space–time can be interpreted as exponentials of those scalar fields $`h_i(t)`$ which lie in the Cartan subalgebra of $`𝕌`$, while the other scalar fields in $`\mathrm{U}/\mathrm{H}`$ correspond to positive roots $`\alpha >0`$ of the Lie algebra $`𝕌`$. In this way the cosmological evolution is described by the motion of a fictitious ball in the CSA of $`𝕌`$. This space is actually a billiard table whose walls are the hyperplanes orthogonal to the various roots. The fictitious ball bounces on the billiard walls and this means that there are inversions in the time evolution of scale factors. Certain dimensions that were expanding almost suddenly begin to contract and others do the reverse. Such a scenario was introduced by Damour, Henneaux, Julia and Nicolai in , and in a series of papers with collaborators , ,, which generalize classical results obtained in the context of pure General Relativity ,. In this approach the cosmic billiard phenomenon is analyzed as an asymptotic regime in the neighborhood of space-like singularities and the billiard walls are seen as delta function potentials provided by the various $`p`$–forms of supergravity localized at sharp instants of time.
It was observed in that the fundamental mathematical setup underlying the appearance of the billiard phenomenon is the so named Solvable Lie algebra parametrization of supergravity scalar manifolds, pioneered in and later applied to the solution of a large variety of superstring/supergravity problems, including the structure of supersymmetric black-hole solutions ,, the construction of gauged SUGRA potentials, and several other issues (for a comprehensive review see ). Indeed we pointed out in those papers that, thanks to the solvable parametrization, one can establish a precise algorithm to implement the following programme:
1. Reduce the original supergravity in higher dimensions $`D4`$ (for instance $`D=10,11`$) to a gravity-coupled $`\sigma `$–model in $`D3`$ where gravity is non–dynamical and where the original higher dimensional bosonic field equations reduce to geodesic equations for a solvable group-manifold metrically equivalent to a non compact coset manifold $`\mathrm{exp}\left[Solv\left(\mathrm{U}/\mathrm{H}\right)\right]\mathrm{U}/\mathrm{H}`$.
2. Utilize the algebraic structure of the solvable Lie algebra $`Solv\left(\mathrm{U}/\mathrm{H}\right)`$ in order to integrate analytically the geodesic equations. In particular we introduced in a general method of integration, named the $`H`$-compensator method which reduces the geodesic differential equations to a triangular form and hence to quadratures when $`\mathrm{U}/\mathrm{H}`$ is maximally split, which is always the case when supersymmetry is maximal ($`N_Q=32`$)
3. Dimensionally oxide the solutions obtained in this way to exact time dependent solutions of $`D4`$ supergravity. In particular we showed in that the oxidation process is not unique but is algebraically classified by the embedding of Weyl orbits of subalgebras $`𝔾𝕌`$. Indeed the analytic structure of the solution is fully determined only by the algebraic structure of $`𝔾`$. Its physical interpretation varies and depends on the explicit embedding $`𝔾𝕌`$. In this way each solution in $`D3`$ corresponds to an entire orbit of higher dimensional backgrounds, very different from one another, but dual to each other under transformations of the Weyl group $`𝒲Weyl(\mathrm{U})`$.
Within this approach it was proved in that the cosmic billiard phenomenon is indeed a general feature of exact time dependent solutions of supergravity and has smooth realizations. Calling $`𝐡(t)`$ the $`r`$–component vector of Cartan fields (where $`r`$ is the rank of $`𝕌`$) and $`𝐡_\alpha (t)\alpha 𝐡(t)`$ its projection along any positive root $`\alpha `$, a bounce occurs at those instant of times $`t_i`$ such that:
$$\alpha \mathrm{\Delta }_+\backslash \dot{𝐡}_\alpha (t)_{t=t_i}=\mathrm{\hspace{0.17em}0}$$
(1.1)
namely when the Cartan field in the direction of some root $`\alpha `$ inverts its behaviour and begins to shrink if it was growing or viceversa begins to grow if it was shrinking. Since all higher dimensional bosonic fields (off-diagonal components of the metric $`g_{\mu \nu }`$ or $`p`$–forms $`A^{[p]}`$) are, via the solvable parametrization of $`\mathrm{U}/\mathrm{H}`$, in one-to-one correspondence with roots $`\varphi _\alpha \alpha `$, it follows that the bounce on a wall (hyperplane orthogonal to the root $`\alpha `$) is caused by the sudden growing of that particular field $`\varphi _\alpha `$. Indeed we showed in that in exact smooth solutions which we were able to obtain by means of the compensator method, each bounce is associated with a typical bell-shaped behaviour of the root field $`\varphi _\alpha `$ and that the whole process can be interpreted as a temporary localization of the Universe energy density in a lump on a spatial brane associated with the field $`\varphi _\alpha `$.
Although very much encouraging the analysis of was still limited in three respects:
The dimensional reduction process which is responsible for making manifest the duality algebra $`𝕌`$ and hence for creating the whole algebraic machinery utilized in deriving the smooth cosmic billiard solutions was stopped at $`D=3`$, namely at the first point where all the bosonic degrees of freedom can be represented by scalars. In $`D=3`$, $`𝕌`$ is still a finite dimensional Lie algebra and the whole richness of the underlying algebraic structure is not yet displayed. As it is well known , in $`D=2`$ and $`D=1`$, the algebra $`𝕌`$ becomes a Kač–Moody algebra, affine or hyperbolic, respectively. The smooth billiard dynamics has to be reconsidered and extended in view of this.
The constructions of depend, in some crucial points, on the assumption that the coset $`\mathrm{U}/\mathrm{H}`$ corresponds to a pair $`\left\{𝕌,𝕌\right\}`$ of Lie algebra and Lie subalgebra which is maximally split. This is always the case for maximal supersymmetry $`N_Q=32`$ but it is not true for $`N_Q32`$. Extending the $`H`$-compensator method to non maximally split pairs $`\left\{𝕌,𝕌\right\}`$ is necessary in order to discuss billiard dynamics in lower SUSY theories and hence in compactifications of string theory on internal manifolds $`_{internal}`$ with restricted holonomies and $`G`$-structures, with or without fluxes.
The solutions considered in were solutions of the pure $`\sigma `$–model, namely of pure, ungauged supergravity. The extension also to gauged supergravities is mandatory in order to make contact with potentially realistic models, in particular with currently considered flux compactifications.
In a recent paper we have begun to address point $`𝐚)`$ of the above list. There we have shown that the mechanism outlined several years ago by Nicolai as the origin of the Kač Moody extension of the duality algebra which appears in $`D=3`$ when you step down to $`D=2`$, namely the existence of two non-locally related dimensional reduction schemes $`D=4D=2`$, the Ehlers reduction and the Maztner Missner reduction, can be formulated in a general set up which provides a regular scheme of analysis both at the algebraic and at the field theoretical level and which applies to all supergravity theories. In particular we have shown that the $`𝕌_{D=3}`$ algebra emerges from the Ehlers reduction and has the following general decomposition with respect to the $`𝕌_{D=4}`$ algebra:
$$\text{adj}(𝕌_{D=3})=\text{adj}(𝕌_{D=4})\text{adj}(\mathrm{SL}(2,)_\mathrm{E})W_{(2,𝐖)}$$
(1.2)
where $`𝐖`$ is a symplectic representation of $`𝕌_{D=4}`$ determined by the vector fields in the parent $`D=4`$ supergravity and $`\mathrm{SL}(2,)_\mathrm{E}/\mathrm{O}(2)`$ is the target space for a $`\sigma `$-model which encodes the degrees of freedom of pure Einstein gravity. Continuing the Ehlers reduction from $`D=3`$ to $`D=2`$ we obtain a Lagrangian with the same symmetry
$$𝕌_{D=2}^{[E]}=𝕌_{D=3}$$
(1.3)
Alternatively, following the Matzner Missner reduction scheme we obtain a twisted $`\sigma `$-model with symmetry
$$𝕌_{D=2}^{[MM]}=𝕌_{D=4}\mathrm{SL}(2,)_{\mathrm{MM}}$$
(1.4)
where $`\mathrm{SL}(2,)_{\mathrm{MM}}/\mathrm{O}(2)`$ is the target space for a $`\sigma `$-model also encoding the degrees of freedom of pure Einstein gravity. The Matzner Missner $`\mathrm{SL}(2,)_{[\mathrm{MM}]}`$ group, however, is not the same as the Ehlers one $`\mathrm{SL}(2,)_{[\mathrm{E}]}`$ and $`𝕌_{D=2}^{[MM]}`$, $`𝕌_{D=2}^{[E]}`$ are just two different finite dimensional subalgebras of the same infinite dimensional one $`𝕌_{D=2}`$, which is nothing else but the affine Kač–Moody extension of $`𝕌_{D=3}`$:
$$\begin{array}{cc}𝕌_{D=2}^{[E]}& \\ & \\ 𝕌_{D=2}^{[MM]}& \end{array}\}𝕌_{D=2}𝕌_{D=3}^{}$$
(1.5)
Understanding the general pattern for the Kač–Moody extension and mastering its field theoretical realization provides the necessary basis for the construction of smooth billiard solutions which rely on the full fledged Lorentzian signature CSA, lying behind supergravity. This we emphasized and begun to exploit in .
In the present paper we address point $`𝐛)`$ of the list mentioned above.
Our starting point is ungauged supergravity in $`D=4`$, whose bosonic lagrangian takes the following general form:
$`^{(4)}`$ $`=`$ $`\sqrt{\text{det}g}\left[2R[g]{\displaystyle \frac{1}{6}}_{\widehat{\mu }}\varphi ^a^{\widehat{\mu }}\varphi ^bh_{ab}(\varphi )+\text{Im}𝒩_{\mathrm{\Lambda }\mathrm{\Sigma }}F_{\widehat{\mu }\widehat{\nu }}^\mathrm{\Lambda }F^{\mathrm{\Sigma }|\widehat{\mu }\widehat{\nu }}\right]`$ (1.6)
$`+{\displaystyle \frac{1}{2}}\text{Re}𝒩_{\mathrm{\Lambda }\mathrm{\Sigma }}F_{\widehat{\mu }\widehat{\nu }}^\mathrm{\Lambda }F_{\widehat{\rho }\widehat{\sigma }}^\mathrm{\Sigma }ϵ^{\widehat{\mu }\widehat{\nu }\widehat{\rho }\widehat{\sigma }}`$
In eq.(1.6) $`\varphi ^a`$ denotes the whole set of $`n_S`$ scalar fields parametrizing the scalar manifold $`_{scalar}^{D=4}`$ which, for $`N_Q8`$, is necessarily a coset manifold:
$$_{scalar}^{D=4}=\frac{\mathrm{U}_{\mathrm{D}=4}}{\mathrm{H}}$$
(1.7)
For $`N_Q8`$, eq.(1.7) is not obligatory but it is possible. Particularly in the $`𝒩=2`$ case, i.e. for $`N_Q=8`$, a large variety of homogeneous special Kähler or quaternionic manifolds fall into the set up of the present general discussion. The fields $`\varphi ^a`$ have $`\sigma `$–model interactions dictated by the metric $`h_{ab}(\varphi )`$ of $`_{scalar}^{D=4}`$. The theory includes also $`n`$ vector fields $`A_{\widehat{\mu }}^\mathrm{\Lambda }`$ for which
$$_{\widehat{\mu }\widehat{\nu }}^{\pm |\mathrm{\Lambda }}\frac{1}{2}\left[F_{\widehat{\mu }\widehat{\nu }}^\mathrm{\Lambda }\mathrm{i}\frac{\sqrt{\text{det}g}}{2}ϵ_{\widehat{\mu }\widehat{\nu }\widehat{\rho }\widehat{\sigma }}F^{\widehat{\rho }\widehat{\sigma }}\right]$$
(1.8)
denote the self-dual (respectively antiself-dual) parts of the field-strengths. As displayed in eq.(1.6) they are non minimally coupled to the scalars via the symmetric complex matrix
$$𝒩_{\mathrm{\Lambda }\mathrm{\Sigma }}(\varphi )=\mathrm{i}\text{Im}𝒩_{\mathrm{\Lambda }\mathrm{\Sigma }}+\text{Re}𝒩_{\mathrm{\Lambda }\mathrm{\Sigma }}$$
(1.9)
which transforms projectively under $`\mathrm{U}_{\mathrm{D}=4}`$. Indeed the field strengths $`F_{\mu \nu }^\mathrm{\Lambda }`$ plus their magnetic duals fill up a $`2n`$–dimensional symplectic representation of $`𝕌_{\mathrm{D}=4}`$ which we call by the name of $`𝐖`$.
The main point in the analysis of billiard dynamics for the lower SUSY cases is that the pair $`\left\{𝕌_{D=4},𝕌_{D=4}\right\}`$ is generically not maximally split. This implies that $`𝕌_{D=3}`$, whose decomposition with respect to $`U_{D=4}`$ is always given by eq.(1.2) is also not maximally split. This happens since, in these cases, $`𝕌_{D=4,3}`$ is a real section of the corresponding complex Lie algebra $`𝕌()`$ different from the maximally non compact one. Indeed it is only for the maximally non-compact real section that:
1. All Cartan generators $`_i`$ are non compact and belong to the Solvable Lie algebra: $`i`$ $`_iSolv(\mathrm{U}/\mathrm{H})`$.
2. All step operators $`E^\alpha `$ associated with positive roots belong to the solvable algebra: $`\alpha >0`$, $`E^\alpha Solv(\mathrm{U}/\mathrm{H})`$.
3. The maximal compact subalgebra $`𝕌`$ is the span of all generators $`E^\alpha E^\alpha `$, for all positive roots $`\alpha >0`$.
Since items 1-3 in the above list are essential ingredients in the algorithm to derive exact solutions developed by us in , it is evident that our set-up has to be reconsidered carefully in the more general case.
In this paper we make an in depth analysis of a specific example of a non maximally split manifold $`\mathrm{U}_{\mathrm{D}=4}/\mathrm{H}`$, that of $`𝒩=6`$ supergravity, from which we extrapolate a general elegant result which reduces the non-maximally split cases to associated maximally split ones allowing, in this way, the extension of the compensator method to all values of $`N_Q`$ and hence the derivation of exact solutions in all instances.
As we are going to see our present results concerning point $`𝐛`$ are quite relevant also for the appropriate discussion of point $`𝐚`$ as well. Indeed the concept of painted walls that will emerge and that of paint group $`G_{paint}`$ are invariant by dimensional reduction and apply also to the Kač–Moody extensions.
In the next subsection we summarize the main result of our paper.
### 1.1 Tits Satake subalgebras and painted walls
In the case of non maximally non-compact manifolds $`\mathrm{U}/\mathrm{H}`$ the Lie algebra $`𝕌`$ of the numerator group is some appropriate real form
$$𝕌=𝔾_R$$
(1.10)
of a complex Lie algebra $`𝔾()`$ of rank $`r=\text{rank}(𝔾)`$. The Lie algebra $``$ of the denominator $`\mathrm{H}`$ is the maximal compact subalgebra $`𝕌`$ which has typically rank $`r_{compact}>r`$. Denoting, as usual, by $`𝕂`$ the orthogonal complement of $``$ in $`𝔾_R`$:
$$𝔾_R=𝕂$$
(1.11)
and defining as non compact rank or rank of the coset $`\mathrm{U}/\mathrm{H}`$ the dimension of the non compact Cartan subalgebra:
$$r_{nc}=\text{rank}(\mathrm{U}/\mathrm{H})\text{dim}^{n.c.};^{n.c.}\text{CSA}_{𝔾()}𝕂$$
(1.12)
we obtain that $`r_{nc}<r`$. The manifold $`\mathrm{U}/\mathrm{H}`$ is still metrically equivalent to a solvable group manifold $`M_{Solv}\mathrm{exp}[Solv(\mathrm{U}/\mathrm{H})]`$ and the field equations of supergravity still reduce to geodesic equations in $`M_{Solv}`$, which can be reformulated as first order equations by using the constant Nomizu connection (see ):
$$\dot{Y}^A+\underset{Nomizu}{\underset{}{\mathrm{\Gamma }_{BC}^A}}Y^BY^C=\mathrm{\hspace{0.17em}0}$$
(1.13)
but it is the form of the Solvable Lie algebra $`Solv(\mathrm{U}/\mathrm{H})`$, whose structure constants define the Nomizu connection, which is now more complicated and apparently does not allow the immediate use of the compensator method for the solution of equations (1.13). Yet the system (1.13) can be reduced to an equivalent one which is maximally split and can be solved with the methods of . This is a consequence of Tits-Satake theory of non compact cosets and split subalgebras and, within such a mathematical framework of a peculiar universal structure of the solvable algebra $`Solv(\mathrm{U}/\mathrm{H})`$ that, up to our knowledge, had not been observed before. Explicitly we have the following scheme. Splitting the Cartan subalgebra into its compact and non compact subalgebras:
$$\begin{array}{ccccc}\hfill \mathrm{CSA}_{𝔾_R}& =& \mathrm{i}^{comp}& & ^{n.c.}\\ & & & & \\ \hfill \mathrm{CSA}_{𝔾()}& =& ^{comp}& & ^{n.c.}\end{array}$$
(1.14)
every vector in the dual of the full Cartan subalgebra, in particular every root $`\alpha `$ can be decomposed into its parallel and transverse part to $`^{n.c.}`$:
$$\alpha =\alpha _{||}\alpha _{}$$
(1.15)
Setting all $`\alpha _{}=0`$ corresponds to a projection:
$$\mathrm{\Pi }:\mathrm{\Delta }_𝔾\overline{\mathrm{\Delta }}$$
(1.16)
of the original root system $`\mathrm{\Delta }_𝔾`$ onto a new system of vectors living in an euclidean space of dimension equal to the non compact rank $`r_{nc}`$. A priori this is not obvious, but it is nonetheless true that $`\overline{\mathrm{\Delta }}`$ is by itself the root system of a simple Lie algebra $`𝔾_{TS}`$, the Tits-Satake subalgebra of $`𝔾_R`$:
$$\overline{\mathrm{\Delta }}=\text{root system of }𝔾_{TS}𝔾_R$$
(1.17)
The Tits-Satake subalgebra $`𝔾_{TS}𝔾_R`$ is always the maximally non compact real section of its own complexification. For this reason, considering its maximal compact subalgebra $`_{TS}𝔾_{TS}`$ we have a new smaller coset $`\mathrm{G}_{\mathrm{TS}}/\mathrm{H}_{\mathrm{TS}}`$ which is maximally split and whose associated solvable algebra $`Solv(\mathrm{G}_{\mathrm{TS}}/\mathrm{H}_{\mathrm{TS}})`$ has the standard structure utilized in to solve the differential equations (1.13). What is the relation between the two solvable Lie algebras $`Solv(\mathrm{G}_\mathrm{R}/\mathrm{H})`$ and $`Solv(\mathrm{G}_{\mathrm{TS}}/\mathrm{H}_{\mathrm{TS}})`$? The explicit answer to this question and the illustration of its relevance for the solution of the geodesic equations (1.13) is the key result of the present paper. It leads to the concept of billiards with painted walls and can be formulated through the following statements.
* A\] In a projection it can occur that more than one higher dimensional vector maps to the same lower dimensional one. This means that in general there will be several roots of $`\mathrm{\Delta }_𝔾`$ which have the same image in $`\overline{\mathrm{\Delta }}`$. Calling $`\mathrm{\Delta }_𝔾^+`$ and $`\overline{\mathrm{\Delta }}^+`$ the sets of positive roots of the two root systems, it happens that both of them split in two subsets with the following properties.
$`\begin{array}{cccccc}& & \multicolumn{-1}{c}{}& & & & & \\ \hfill 𝔾_R& & & & \hfill 𝔾_{TS}& & & \\ & & \multicolumn{-1}{c}{}& & & & & \\ & & & & & & & \\ \hfill \mathrm{\Delta }_𝔾^+=\mathrm{\Delta }^\eta \mathrm{\Delta }^\delta & & & & \hfill \overline{\mathrm{\Delta }}^+=\overline{\mathrm{\Delta }}^{\mathrm{}}\overline{\mathrm{\Delta }}^s& & & \\ \hfill \eta _1,\eta _2\mathrm{\Delta }^\eta ;& \hfill \eta _1+\eta _2& & \mathrm{\Delta }^\eta \hfill & \hfill \alpha _1^{\mathrm{}},\alpha _2^{\mathrm{}}\overline{\mathrm{\Delta }}^{\mathrm{}};& \hfill \alpha _1^{\mathrm{}}+\alpha _2^{\mathrm{}}& & \overline{\mathrm{\Delta }}^{\mathrm{}}\hfill \\ \hfill \eta \mathrm{\Delta }^\eta ,\delta \mathrm{\Delta }^\delta ;& \hfill \eta +\delta & & \mathrm{\Delta }^\delta \hfill & \hfill \alpha ^{\mathrm{}}\overline{\mathrm{\Delta }}^{\mathrm{}},\alpha ^s\overline{\mathrm{\Delta }}^s;& \hfill \alpha ^{\mathrm{}}+\alpha ^s& & \overline{\mathrm{\Delta }}^s\hfill \\ \hfill \delta _1,\delta _2\mathrm{\Delta }^\delta ;& \hfill \delta _1+\delta _2& & \{\begin{array}{cc}\mathrm{\Delta }^\delta \hfill & \\ \mathrm{\Delta }^\eta \hfill & \end{array}\hfill & \hfill \alpha _1^s,\alpha _2^s\overline{\mathrm{\Delta }}^s;& \hfill \alpha _1^s+\alpha _2^s& & \{\begin{array}{cc}\overline{\mathrm{\Delta }}^{\mathrm{}}\hfill & \\ \overline{\mathrm{\Delta }}^s\hfill & \end{array}\hfill \\ & & & & & & & \end{array}`$ (1.25)
The projection acts on the two different sets in the following way:
$`\mathrm{\Pi }\left[\mathrm{\Delta }^\eta \right]=\overline{\mathrm{\Delta }}^{\mathrm{}}`$
$`\mathrm{\Pi }\left[\mathrm{\Delta }^\delta \right]=\overline{\mathrm{\Delta }}^s`$
$`\alpha ^{\mathrm{}}\overline{\mathrm{\Delta }}^{\mathrm{}};\text{card}\mathrm{\Pi }^1\left[\alpha ^{\mathrm{}}\right]=\mathrm{\hspace{0.17em}1}`$
$`\alpha ^s\overline{\mathrm{\Delta }}^s;\text{card}\mathrm{\Pi }^1\left[\alpha ^s\right]=m`$
$`\text{card}\mathrm{\Delta }_{𝔾_R}^+=\text{card}\overline{\mathrm{\Delta }}^{\mathrm{}}+m\times \text{card}\overline{\mathrm{\Delta }}^s`$ (1.27)
It means that there are two type of roots those which have a distinct image in the projected root system and those which arrange into multiplets with the same projection. The possible multiplicities, however, are only two, either $`1`$ or $`m`$. Because of that we can enumerate the generators of the solvable algebra $`Solv(\mathrm{G}_\mathrm{R}/\mathrm{H})`$ in the following way:
$`H_i`$ $``$ Cartan generators
$`\mathrm{\Phi }_\alpha ^{\mathrm{}}`$ $``$ $`\eta \text{roots}`$
$`\mathrm{\Omega }_{\alpha ^s|I}`$ $``$ $`\delta \text{roots};(I=1,\mathrm{},m)`$ (1.28)
The index $`I`$ enumerating the $`m`$–roots of $`\mathrm{\Delta }_{𝔾_R}`$ that have the same projection in $`\overline{\mathrm{\Delta }}`$ is named the paint index
* B\] There exists a compact subalgebra $`𝔾_{paint}𝔾_R`$ which acts as an algebra of outer automorphisms (i.e. outer derivatives) on the solvable algebra $`Solv_{𝔾_R}Solv(\mathrm{G}_\mathrm{R}/\mathrm{H})𝔾_R`$, namely:
$`[𝔾_{paint},Solv_{𝔾_R}]=Solv_{𝔾_R}`$
(1.29)
* C\] The Cartan generators $`H_i`$ and the generators $`\mathrm{\Phi }_\alpha ^{\mathrm{}}`$ are singlets under the action of $`𝔾_{paint}`$, i.e. each of them commutes with the whole of $`𝔾_{paint}`$:
$$[H_i,𝔾_{paint}]=[\mathrm{\Phi }_\alpha ^{\mathrm{}},𝔾_{paint}]=\mathrm{\hspace{0.17em}0}$$
(1.30)
On the other hand, each of the $`m`$-multiplets of generators $`\mathrm{\Omega }_{\alpha ^s|I}`$ constitutes an orbit under the action of the paint group $`G_{paint}`$, i.e. a linear representation $`𝐃[\alpha ^s]`$ which, for different roots $`\alpha ^s`$ can be different, but has always the same dimension $`m`$ :
$$X𝔾_{paint}:[X,\mathrm{\Omega }_{\alpha ^s|I}]=\left(D^{[\alpha ^s]}[X]\right)_I^J\mathrm{\Omega }_{\alpha ^s|J}$$
(1.31)
* D\] The paint algebra $`𝔾_{paint}`$ contains a subalgebra
$$𝔾_{paint}^0𝔾_{paint}$$
(1.32)
such that with respect to $`𝔾_{paint}^0`$, each $`m`$–dimensional representation $`𝐃[\alpha ^s]`$ branches in the same way as follows:
$$𝐃[\alpha ^s]\stackrel{𝔾_{paint}^0}{}\underset{\text{singlet}}{\underset{}{\mathrm{𝟏}}}\underset{(m1)\text{dimensional}}{\underset{}{𝐉}}$$
(1.33)
Accordingly we can split the range of the paint index $`I`$ as follows:
$$I=\{0,\underset{1,\mathrm{},m1}{\underset{}{x}}\}$$
(1.34)
the index $`0`$ corresponding to the singlet, while $`x`$ ranges over the representation $`𝐉`$
* E\] The tensor product $`𝐉𝐉`$ contains both the identity representation $`\mathrm{𝟏}`$ and the representation $`𝐉`$ itself. Furthermore, there exists, in the representation $`^3𝐉`$ a $`𝔾_{paint}^0`$-invariant tensor $`a^{xyz}`$ such that the two solvable Lie algebras $`Solv_{𝔾_R}`$ and $`Solv_{𝔾_{TS}}`$ can be written as follows
$`\begin{array}{c}& & \multicolumn{-1}{c}{}\\ & Solv_{𝔾_{TS}}\hfill & Solv_{𝔾_R}\hfill \\ & & \multicolumn{-1}{c}{}\\ & [H_i,H_j]=0\hfill & [H_i,H_j]=0\hfill \\ & [H_i,E^\alpha ^{\mathrm{}}]=\alpha _i^{\mathrm{}}E^\alpha ^{\mathrm{}}\hfill & [H_i,\mathrm{\Phi }_\alpha ^{\mathrm{}}]=\alpha _i^{\mathrm{}}\mathrm{\Phi }_\alpha ^{\mathrm{}}\hfill \\ & [H_i,E^{\alpha ^s}]=\alpha _i^sE^{\alpha ^s}\hfill & [H_i,\mathrm{\Omega }_{\alpha ^s|I}]=\alpha _i^s\mathrm{\Omega }_{\alpha ^s|I}\hfill \\ & & \\ \alpha ^{\mathrm{}}+\beta ^{\mathrm{}}\overline{\mathrm{\Delta }}\hfill & [E^\alpha ^{\mathrm{}},E^\beta ^{\mathrm{}}]=0\hfill & [\mathrm{\Phi }_\alpha ^{\mathrm{}},\mathrm{\Phi }_\beta ^{\mathrm{}}]=0\hfill \\ & & \\ \alpha ^{\mathrm{}}+\beta ^{\mathrm{}}\overline{\mathrm{\Delta }}\hfill & [E^\alpha ^{\mathrm{}},E^\beta ^{\mathrm{}}]=N_{\alpha ^{\mathrm{}}\beta ^{\mathrm{}}}E^{\alpha ^{\mathrm{}}+\beta ^{\mathrm{}}}\hfill & [\mathrm{\Phi }_\alpha ^{\mathrm{}},\mathrm{\Phi }_\beta ^{\mathrm{}}]=N_{\alpha ^{\mathrm{}}\beta ^{\mathrm{}}}\mathrm{\Phi }_{\alpha ^{\mathrm{}}+\beta ^{\mathrm{}}}\hfill \\ & & \\ \alpha ^{\mathrm{}}+\beta ^s\overline{\mathrm{\Delta }}\hfill & [E^\alpha ^{\mathrm{}},E^{\beta ^s}]=0\hfill & [\mathrm{\Phi }_\alpha ^{\mathrm{}},\mathrm{\Omega }_{\beta ^s|I}]=0\hfill \\ & & \\ \alpha ^{\mathrm{}}+\beta ^s\overline{\mathrm{\Delta }}^s\hfill & [E^\alpha ^{\mathrm{}},E^{\beta ^s}]=N_{\alpha ^{\mathrm{}}\beta ^s}E^{\alpha ^{\mathrm{}}+\beta ^s}\hfill & [\mathrm{\Phi }_\alpha ^{\mathrm{}},\mathrm{\Omega }_{\beta ^s|I}]=N_{\alpha ^{\mathrm{}}\beta ^s}\mathrm{\Omega }_{\alpha ^{\mathrm{}}+\beta ^s|I}\hfill \\ & & \\ \alpha ^s+\beta ^s\overline{\mathrm{\Delta }}\hfill & [E^{\alpha ^s},E^{\beta ^s}]=0\hfill & [\mathrm{\Omega }_{\alpha ^s|I},\mathrm{\Omega }_{\beta ^s|J}]=0\hfill \\ & & \\ \alpha ^s+\beta ^s\overline{\mathrm{\Delta }}^{\mathrm{}}\hfill & [E^{\alpha ^s},E^{\beta ^s}]=N_{\alpha ^s\beta ^s}E^{\alpha ^s+\beta ^s}\hfill & [\mathrm{\Omega }_{\alpha ^s|I},\mathrm{\Omega }_{\beta ^s|J}]=\delta ^{IJ}N_{\alpha ^s\beta ^s}\mathrm{\Phi }_{\alpha ^s+\beta ^s}\hfill \\ & & \\ \alpha ^s+\beta ^s\overline{\mathrm{\Delta }}^s\hfill & [E^{\alpha ^s},E^{\beta ^s}]=N_{\alpha ^s\beta ^s}E^{\alpha ^s+\beta ^s}\hfill & \{\begin{array}{cc}[\mathrm{\Omega }_{\alpha ^s|0},\mathrm{\Omega }_{\beta ^s|0}]=N_{\alpha ^s\beta ^s}\mathrm{\Omega }_{\alpha ^s+\beta ^s|0}\hfill & \\ & \\ [\mathrm{\Omega }_{\alpha ^s|0},\mathrm{\Omega }_{\beta ^s|x}]=N_{\alpha ^s\beta ^s}\mathrm{\Omega }_{\alpha ^s+\beta ^s|x}\hfill & \\ & \\ [\mathrm{\Omega }_{\alpha ^s|x},\mathrm{\Omega }_{\beta ^s|y}]=N_{\alpha ^s\beta ^s}(\delta ^{xy}\mathrm{\Omega }_{\alpha ^s+\beta ^s|0}\hfill & \\ +a^{xyz}\mathrm{\Omega }_{\alpha ^s+\beta ^s|z})\hfill & \end{array}\hfill \end{array}`$ (1.53)
The existence of the paint group $`G_{paint}`$ and the structure of the solvable Lie algebra displayed in eq. (LABEL:paragonando) imply that we can reduce the geodesic problem on $`G_R/H`$ and hence the supergravity field equations to the geodesic problem on $`G_{TS}/H_{TS}`$ which is maximally split and can be solved with the compensator method introduced in . It suffices to observe that by setting all the components of the tangent vectors in the directions of the generators $`\mathrm{\Omega }_{\alpha ^s|x}`$ to zero we simply reproduce a copy of the solvable Lie algebra of the Tits Satake manifold. Once we have found a solution for this latter, we can extend it to a full fledged solution of the original system by applying rotations of the paint group $`G_{paint}`$ with constant parameters. Physically this means that indeed the billiard table is just the Weyl chamber of the Tits Satake algebra as observed by Damour et al , yet, in the smooth billiard realization the raising and lowering of the walls occurs in paints which specify the precise correspondence with the supergravity fields and hence with the oxidation to higher dimensions.
### 1.2 Content of the paper
In the sequel of this paper we illustrate these general structures by working out in all details a specific example, that arising from $`D=4,𝒩=6`$ supergravity. Our choice is motivated as follows. On one hand, the case $`N_Q=24`$ is the next simplest apart from that of maximal supersymmetry $`N_Q=32`$. Indeed there is just the graviton multiplet, the number of fields is completely fixed and so is the geometric structure of the lagrangian. On the other hand the scalar manifold of $`𝒩=6`$ supergravity is an instance of a special Kähler manifold and the bosonic lagrangian can be reinterpreted as the lagrangian of a particular $`𝒩=2`$ model. In other words we could also reconsider our constructions from an $`𝒩=2`$ viewpoint and interpret the scalar fields we deal with as moduli of an abstract Calabi-Yau compactification. Indeed in a subsequent paper we shall extrapolate from the present example general considerations on billiard dynamics and painted walls in the context of special geometries.
Our paper is organized as follows.
In section 2 we present the in depth analysis of the $`\mathrm{E}_{7(5)}`$ real section: how generators are constructed, how they are subdivided into compact and non compact ones, how the Tits Satake projection works in this case, what is the structure of the solvable Lie algebra generating the coset manifold $`\mathrm{E}_{7(5)}/\mathrm{SO}(12)\times \mathrm{SO}(3)`$ and what is the structure of the paint group. Then in section 3 we derive the Nomizu connection for both the original manifold and its Tits Satake projection and we compare the structure of the two systems of first order equations for the tangent vectors. In section 4 we derive explicit smooth solutions for the maximally split $`\mathrm{F}_4`$ system and we show that they display several bounces: smooth cosmic billiards. In section 5 we uplift the previously found solutions to the original $`\mathrm{E}_{7(5)}`$ system by means of the paint group. Then we discuss the general features of the Tits Satake projection, how it commutes with dimensional reduction and how the paint group is preserved in the reduction. We illustrate these concepts on the specific example. Section 6 contains our conclusions. Then we have two appendices. The first, appendix A, contains the listing and ordering of $`\mathrm{E}_7`$ roots utilized throughout the paper. The second, appendix B, is devoted to the explicit construction of the fundamental $`26`$–dimensional representation of $`\mathrm{F}_{4(4)}`$ which we used in the paper to calculate the needed $`N_{\alpha \beta }`$ matrix.
## 2 The example of $`𝒩`$=6 supergravity
In $`𝒩=6,D=4`$ supergravity there are $`30`$ scalars which span the special Kähler manifold:
$$_{scalar}^{D=4,𝒩=6}=\frac{\mathrm{SO}^{}(12)}{\mathrm{SU}(6)\times \mathrm{U}(1)}$$
(2.1)
and the relevant duality algebra is therefore:
$$𝔾_{D=4}=\mathrm{SO}^{}(12)$$
(2.2)
The $`16`$ graviphotons give rise to $`16`$ electric plus $`16`$ magnetic field strengths that organize into the $`32`$ spinor representation of $`\mathrm{SO}^{}(12)`$ which is symplectic as it should be.
After reduction to $`D=3`$ dimensions and dualization of all the vector fields to scalars we obtain a 3D-gravity coupled $`\sigma `$–model based on the quaternionic symmetric space:
$$_{scalar}^{D=3,N=12}=\frac{\mathrm{E}_{7(5)}}{\mathrm{SO}(3)_\mathrm{R}\times \mathrm{SO}(12)}$$
(2.3)
which is the $`c`$-map of the special Kähler manifold 2.1. In this section we study the structure of the solvable Lie algebra describing the non maximally split non-compact manifold (2.3) and how it is related to its Tits Satake submanifold:
$$_{scalar}^{TitsSatake}=\frac{\mathrm{F}_{4(4)}}{\mathrm{SU}(2)_\mathrm{R}\times \mathrm{Usp}(6)}$$
(2.4)
which is instead maximally split and it is the relevant submanifold defining the cosmic billiard. Our main goal is to show how the solution of the first order equations for the system (2.4) can be used to obtain solutions for the system (2.3). In particular we shall appropriately study how the dynamic walls of the billiard (2.3) are painted copies of the walls associated with the billiard (2.4).
To this effect we have to develop all the algebraic machinery associated with the real form $`\mathrm{E}_{7(5)}`$ of the $`\mathrm{E}_7`$ complex Lie algebra. We begin by spelling out the particular form of the decomposition (1.2)
$$\text{adj}(\mathrm{E}_{7(5)})=\text{adj}(\mathrm{SO}^{}(12))\text{adj}(\mathrm{SL}(2,))(\mathrm{𝟐},\mathrm{𝟑𝟐}_𝐬)$$
(2.5)
where $`\mathrm{𝟑𝟐}_𝐬`$ denotes the spinor representation of $`\mathrm{SO}^{}(12)`$, while $`\mathrm{𝟐}`$ denotes the fundamental representation of $`\mathrm{SL}(2,\mathrm{R})`$. The subgroup $`\mathrm{SO}^{}(12)\times \mathrm{SL}(2,)`$ is regularly embedded and non compact. There is another similar decomposition of the adjoint of $`\mathrm{E}_{7(5)}`$ with respect to its maximal compact subgroup:
$$\text{adj}(\mathrm{E}_{7(5)})=\text{adj}(\mathrm{SO}(12))\text{adj}(\mathrm{SO}(3)_\mathrm{R})(\mathrm{𝟐},\mathrm{𝟑𝟐}_𝐬)$$
(2.6)
where, once again $`\mathrm{𝟑𝟐}_𝐬`$ denotes the spinor representation of the compact $`\mathrm{SO}(12)`$, this time.
The non compact symmetric space (2.3) has $`rank=4`$. This means that of the seven Cartan generators of $`\mathrm{E}_{7(5)}`$, four are non compact and belong to the coset, while three are compact and belong to the compact subalgebra. We proceed to the explicit construction of the involutive automorphism of the complex $`\mathrm{E}_7^{}`$ algebra
$$\sigma :\mathrm{E}_7^{}\mathrm{E}_7^{}$$
(2.7)
which defines the real form $`\mathrm{E}_{7(5)}`$. This given we obtain also the compact subalgebra $``$, the complementary non compact subspace $`𝕂`$ and the solvable Lie algebra $`Solv_{\mathrm{E7}(5)}`$ whose corresponding solvable group manifold is isometrical to the coset manifold (2.3).
### 2.1 The $`\mathrm{E}_7`$ root system, and its projection onto the $`\mathrm{F}_4`$ root system
In order to realize the programme we have just outlined, we begin by choosing an explicit basis of simple roots for $`\mathrm{E}_7`$. In an Euclidean orthonormal basis they are the following ones:
$`\alpha _1`$ $`=`$ $`\{1,1,0,0,0,0,0\}`$
$`\alpha _2`$ $`=`$ $`\{0,1,1,0,0,0,0\}`$
$`\alpha _3`$ $`=`$ $`\{0,0,1,1,0,0,0\}`$
$`\alpha _4`$ $`=`$ $`\{0,0,0,1,1,0,0\},`$
$`\alpha _5`$ $`=`$ $`\{0,0,0,0,1,1,0\},`$
$`\alpha _6`$ $`=`$ $`\{0,0,0,0,1,1,0\},`$
$`\alpha _7`$ $`=`$ $`\{{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{\sqrt{2}}}\}`$ (2.8)
and they are associated with the $`\mathrm{E}_7`$ Dynkin diagram labeled as it is displayed in fig.1:
Next we list all the positive roots of $`\mathrm{E}_7`$ arranged according to their height. They are $`63`$ and they are listed in Appendix A .
The real section $`\mathrm{E}_{7(5)}`$ of the complex Lie algebra $`\mathrm{E}_7^{}`$ is identified by the Tits Satake diagram depicted in fig. 1 where the simple roots $`\alpha _1`$, $`\alpha _3`$, $`\alpha _5`$ are black. This means that in the chosen real form the Cartan generators dual to these three roots $`_{\alpha _{1,3,5}}`$ are compact, while non compact are the Cartan generators in the complementary $`4`$–dimensional subspace. It is fairly easy to describe the space of non-compact Cartan generators $`^{n.c.}`$. It is the span of the four weight vectors $`\lambda _{2,4,5,7}`$ which, by construction, are orthogonal to the roots $`\alpha _{1,3,5}`$. Thus in the chosen euclidean basis we obtain:
$`^{n.c.}`$ $`=`$ $`\text{span}\{\lambda _2,\lambda _4,\lambda _6,\lambda _7\}`$
$`\lambda _2`$ $`=`$ $`\{1,1,0,0,0,0,\sqrt{2}\}`$
$`\lambda _4`$ $`=`$ $`\{1,1,1,1,0,0,2\sqrt{2}\}`$
$`\lambda _6`$ $`=`$ $`\{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},{\displaystyle \frac{3}{\sqrt{2}}}\}`$
$`\lambda _7`$ $`=`$ $`\{0,0,0,0,0,0,\sqrt{2}\}`$
It is now easy to construct an orthogonal basis of four length-two $`7`$-vectors for the space $`^{n.c.}`$ defined by eq.(LABEL:Hspan2467). It is given by:
$`e_1`$ $`=`$ $`\{1,1,0,0,0,0,0\}`$
$`e_2`$ $`=`$ $`\{0,0,1,1,0,0,0\}`$
$`e_3`$ $`=`$ $`\{0,0,0,0,1,1,0\}`$
$`e_4`$ $`=`$ $`\{0,0,0,0,0,0,\sqrt{2}\}`$
Indeed, equivalently to eq.(LABEL:Hspan2467) we can also write:
$$^{n.c.}=\text{span}\{e_1,e_2,e_3,e_4\}$$
(2.11)
We can complete the basis (LABEL:ortonchbas) with other three vectors also of length 2, which are orthogonal to $`e_{1,2,3,4}`$ and also orthogonal among themselves:
$`e_5`$ $`=`$ $`\{1,1,0,0,0,0,0\}`$
$`e_6`$ $`=`$ $`\{0,0,1,1,0,0,0\}`$
$`e_7`$ $`=`$ $`\{0,0,0,0,1,1,0\}`$
The compact Cartan subalgebra is provided by the span of these three vectors:
$$^c=\text{span}\{e_5,e_6,e_7\}$$
(2.13)
The $`\mathrm{E}_7`$ roots are vectors in the dual of the $`7`$-dimensional space which is the direct sum of the four dimensional space $`^{n.c.}`$ plus the three dimensional space $`^c`$:
$$=^{n.c.}^c$$
(2.14)
Hence every root $`\alpha \mathrm{\Delta }_{E_7}`$ can be decomposed as follows:
$$\alpha =\alpha _{||}\alpha _{}$$
(2.15)
where $`\alpha _{||}`$ lies in $`^{n.c.}`$ and $`\alpha _{}`$ is orthogonal to it. The essential point in Tits Satake theory of real forms is that the parallel projections of the roots, namely $`\alpha _{||}`$, are not just arbitrary vectors, rather they are roots of a Lie algebra of rank equal to the dimension of the non compact Cartan Lie algebra which is actually a subalgebra of the original algebra. In our case we have $`rank=\mathrm{\hspace{0.17em}4}`$ and the relevant subalgebra (Tits Satake) is $`\mathrm{F}_{4(4)}E_{7(5)}`$. Indeed the parallel projections $`\alpha _{||}`$ fill the cardinality $`24`$ root-system $`\mathrm{\Delta }_{F_4}`$.
The actual construction of the real form $`\mathrm{E}_{7(5)}`$ involves the careful analysis of the onto projection:
$$\mathrm{\Delta }_{E_7}\stackrel{\pi }{}\mathrm{\Delta }_{F_4}$$
(2.16)
Explicitly, if we decompose the $`63`$ positive roots of $`\mathrm{E}_7`$ along the new orthogonal basis $`e_{1,2,3,4,5,7}`$ we discover the following:
There are just three roots that are orthogonal to the subspace spanned by $`e_{1,2,3,4}`$, namely such that $`\alpha _{||}=0`$. They are precisely the simple roots $`\alpha _1`$, $`\alpha _3`$ and $`\alpha _5`$.
The remaining $`60`$ roots have a projection onto the space spanned by $`e_{1,2,3,4}`$ which takes the form of one of the $`24`$ roots of $`\mathrm{F}_4`$ and all such 24 roots are reproduced in the projection. Namely $`\alpha _{||}\mathrm{\Delta }_{F_4}`$.
The set of $`24`$ roots of $`\mathrm{F}_4`$ is subdivided in two subsets of 12 roots each. The long and the short roots. Each long root appears only once in the projection of $`\mathrm{E}_7`$ roots. Each of the 12 short roots, instead, appears exactly four times as image of four distinct $`\mathrm{E}_7`$ roots. So that we count $`4\times 12+12=60`$
To understand this pattern we have to introduce the $`\mathrm{F}_4`$ root system. The Dynkin diagram of $`\mathrm{F}_4`$ is given in fig. 2
and calling $`y_{1,2,3,4}`$ a basis of orthonormal vectors:
$$y_iy_j=\delta _{ij}$$
(2.17)
a possible choice of simple roots $`\varpi _i`$ which reproduces the Cartan matrix encoded in the Dynkin diagram (2) is the following:
$`\varpi _1`$ $`=`$ $`y_1y_2y_3+y_4`$
$`\varpi _2`$ $`=`$ $`2y_3`$
$`\varpi _3`$ $`=`$ $`y_2y_3`$
$`\varpi _4`$ $`=`$ $`y_1y_2`$ (2.18)
With this basis of simple roots the full root system composed of $`48`$ vectors is given by:
$$\mathrm{\Delta }_{F_4}\underset{\text{24 roots}}{\underset{}{\pm y_i\pm y_j}};\underset{\text{8 roots}}{\underset{}{\pm y_i}};\underset{\text{16 roots}}{\underset{}{\pm y_1\pm y_2\pm y_3\pm y_4}}$$
(2.19)
and one can list the positive roots by height as displayed in table 1.
If we identify the $`\mathrm{E}_7`$ roots with their progressive number as it is defined by their listing in Appendix A, we can reorganize them into the following three subsets according to their projection onto the $`\mathrm{F}_4`$ root space.
First we have the $`\mathrm{\Delta }_\beta `$ set:
$$\beta _1=\alpha _1;\beta _2=\alpha _3;\beta _3=\alpha _5$$
(2.20)
which contains the three roots with vanishing projection onto the $`\mathrm{F}_4`$ root space. As we are going to see, together with their negative and with the compact Cartan generators, these roots define a compact subalgebra $`\mathrm{SO}(3)_1\times \mathrm{SO}(3)_2\times \mathrm{SO}(3)_3`$ with respect to which the generators of the solvable Lie algebra of $`\mathrm{E}_{7(5)}/\mathrm{SO}(12)\times \mathrm{SU}(2)`$ transform covariantly and arrange into representations. Indeed this $`\mathrm{SO}(3)^3`$ is, for the present case, the paint group $`𝔾_{\mathrm{paint}}`$ mentioned in eq. (1.29). The subgroup $`𝔾_{paint}^0𝔾_{\mathrm{paint}}`$ mentioned in eq.s (1.32) and (1.33) is actually the diagonal subgroup $`\mathrm{SO}(3)_{diag}\mathrm{SO}(3)^3`$.
Secondly we have the $`\mathrm{\Delta }_\eta `$ set containing those twelve roots whose projection onto the $`\mathrm{F}_4`$ root space is unique. We organize them according to the height of the $`\mathrm{F}_4`$ root on which they project. The result is displayed in table 2
Thirdly we have the $`\mathrm{\Delta }_\delta `$ set of those $`48`$ $`\mathrm{E}_7`$ roots which arrange into quadruplets having the same projection onto the $`\mathrm{F}_4`$ system. We denote these roots by $`\delta _I^i`$ where $`I=1,\mathrm{},12`$ and $`i=1,2,3,4`$. They are displayed in table 3
### 2.2 The real form $`\mathrm{E}_{7(5)}`$ and its associated solvable Lie algebra
Given these preliminaries we can now introduce the real form $`\mathrm{E}_{7(5)}`$ which follows from the action of a suitable involutive automorphism (2.7) of the complex Lie algebra $`\mathrm{E}_7^{}`$.
Following the general definitions presented in most textbooks on Lie algebra theory (see for instance ), the real form $`𝔾_{}`$ is defined as the subspace of eigenvalue $`1`$ of the relevant automorphism $`\sigma `$, namely we have:
$$\sigma \left(𝔾_{}\right)=𝔾_{}$$
(2.21)
On the other hand, $`\sigma `$ is completely identified by the Tits Satake diagram depicted in fig.1. Indeed the action of $`\sigma `$ is originally defined on the Cartan subalgebra and corresponds to changing the signs of all vectors lying in the compact part while keeping unchanged those lying in the non compact part:
$$\sigma :^{n.c.}^{n.c.};\sigma :^c^c$$
(2.22)
From the Cartan algebra, the action of $`\sigma `$ is canonically extended to the root space. Decomposing each root in its parallel and transverse parts we have:
$$\sigma (\alpha )=\sigma \left(\alpha _{||}+\alpha _{}\right)=\alpha _{||}\alpha _{}$$
(2.23)
If we rewrite all the sixty three $`\mathrm{E}_7`$ roots in the $`e_i`$ basis defined by eq.s (LABEL:ortonchbas) and (LABEL:compbas) we unveil the meaning of our regrouping from the point of view of the automorphism $`\sigma `$. The set $`\mathrm{\Delta }_\eta `$ is composed by all those roots whose transverse part vanishes, namely the components of each $`\eta `$-root along $`e_{5,6,7}`$ are zero:
$$\eta \mathrm{\Delta }_\eta :\eta _{}=0$$
(2.24)
which implies
$$\sigma \left(\eta _I\right)=\eta _I$$
(2.25)
On the other hand the roots in the set $`\mathrm{\Delta }_\delta `$ arrange into pairs such that the transverse part of $`\delta _I^1`$ is the opposite of that of $`\delta _I^4`$ and similarly that of $`\delta _I^2`$ is the opposite of that of $`\delta _I^3`$. Hence we have:
$$\sigma \left(\delta _I^1\right)=\delta _I^4;\sigma \left(\delta _I^4\right)=\delta _I^1;\sigma \left(\delta _I^2\right)=\delta _I^3;\sigma \left(\delta _I^3\right)=\delta _I^2$$
(2.26)
Finally from the root space the automorphism $`\sigma `$ can be lifted to the step operators and hence to the whole algebra. This last step involves the introduction of a set of sign factors. To see this, let $`_i`$ and $`E^\alpha `$ be the Cartan and the step operators, respectively, realized in the maximally non compact, split, real section $`𝔾_{split}`$ of the complex Lie algebra $`𝔾^{}`$ . If regarded as matrices in any of its irreducible representations both $`_i`$ and $`E^\alpha `$ are real matrices. In our case the complex Lie algebra is $`\mathrm{E}_{7}^{}{}_{}{}^{}`$ and the maximally non compact split real section is $`\mathrm{E}_{7(7)}`$. The representation we can focus on is the fundamental $`56`$-dimensional representation and the explicit form of the matrices $`_i`$ and $`E^\alpha `$ we shall utilize was constructed by us in 1997 and it is described in . This fixes the conventions, which is a necessary step, since the definition of the step operators is up to choices of some arbitrary signs. This being set, the lifting of the automorphism $`\sigma `$ from the root space to the complex Lie algebra is defined in the following way. Firstly the complex Lie algebra is defined as the complex span (linear combinations with complex coefficients) of the generators $`_i`$ and $`E^\alpha `$:
$$𝔾^{}=\text{complex span}\{_i,E^\alpha \}$$
(2.27)
Secondly, for each element $`g𝔾^{}`$, we require:
$$\sigma \left(\mathrm{i}g\right)=\mathrm{i}\sigma \left(g\right)$$
(2.28)
where $`\mathrm{i}=\sqrt{1}`$ denotes the imaginary unit. Thirdly the automorphism is fixed by writing its action on the generators:
$$\sigma \left(_i^{||}\right)=_i^{||};\sigma \left(_i^{}\right)=_i^{};\sigma \left(E^\alpha \right)=a_\alpha E^{\sigma (\alpha )}$$
(2.29)
In the above equation $`a_\alpha `$ is a real number whose absolute is immediately fixed to one by consistency with Jacobi identities. Hence $`a_\alpha =\pm 1`$. Yet the choice of these signs is not immediately obvious. Indeed it follows from the original choice of normalizations of the step operators for the split algebra $`𝔾_{split}`$ and therefore it is convention dependent. In a moment we shall resolve this ambiguity relative to the already mentioned choice of conventions, namely those of . First let us observe that once the $`a_\alpha `$ are fixed, the complex linear combinations of split generators forming a complete basis for the real Lie algebra $`𝔾_R`$ are also fixed. As an example let us consider the maximally compact real section $`𝔾_{compact}`$ for which, as it is well known, we always have:
$$𝔾_{compact}=\text{real span}\{\mathrm{i}H_i,\mathrm{i}\frac{1}{2}\left(E^\alpha +E^\alpha \right),\frac{1}{2}\left(E^\alpha E^\alpha \right)\}$$
(2.30)
In view of the previous theory this is easily explained as follows. In this case the whole Cartan subalgebra is compact and hence $`\sigma (_i)=_i`$ for all Cartan generators. From this it follows that $`\alpha _{||}=0`$ for all roots and therefore $`\sigma (\alpha )=\alpha `$. The actual linear combinations displayed in eq.(2.30) follow from the choice $`a_\alpha =1,\alpha `$ which implies:
$$\sigma \left(\mathrm{i}\frac{1}{2}(E^\alpha +E^\alpha )\right)=\mathrm{i}\frac{1}{2}(E^\alpha +E^\alpha );\sigma \left(\frac{1}{2}(E^\alpha E^\alpha )\right)=\frac{1}{2}(E^\alpha E^\alpha )$$
(2.31)
Had we chosen $`a_\alpha =1`$ we would have obtained the same linear combinations but with the $`i`$-factors interchanged: $`\frac{1}{2}\left(E^\alpha +E^\alpha \right)`$ , $`\mathrm{i}\frac{1}{2}\left(E^\alpha E^\alpha \right)`$. Such a choice, however, would be wrong since it does not define an algebra. Indeed the commutator of two generators of type $`\mathrm{i}\frac{1}{2}\left(E^\alpha E^\alpha \right)`$ produces a generator of the same type, but without the $`i`$-factor in front. On the contrary the opposite choice of $`a_\alpha `$, which amounts to the well known choice (2.30) of $`i`$-prefactors consistently defines a subalgebra. This discussion shows that:
1. The choice of the $`a_\alpha `$ factors which completely determines the action the automorphism $`\sigma `$ is fully equivalent to deciding the position of the $`i`$-factors, namely to deciding whether, for each pair $`\alpha `$ and $`\sigma (\alpha )`$ of roots mapped into each other by the automorphism it is
$$\frac{1}{2}(E^\alpha E^{\sigma (\alpha )});\mathrm{i}\frac{1}{2}(E^\alpha +E^{\sigma (\alpha )})$$
(2.32)
or
$$\frac{1}{2}(E^\alpha +E^{\sigma (\alpha )});\mathrm{i}\frac{1}{2}(E^\alpha E^{\sigma (\alpha )})$$
(2.33)
which appear as generators of the algebra $`𝔾_{}`$.
2. The decision whether (2.32) or (2.33) is the right choice is determined by the commutation relations and the closure of the algebra $`𝔾_{}`$ and can be different for different pairs of related roots.
In the case of the $`\mathrm{E}_{7(5)}`$ real section of the $`\mathrm{E}_7^{}`$ complex Lie algebra, using the normalization of step operators derived in we have carefully inspected by computer calculations all the commutation relations and we have derived the assignment of $`i`$-factors displayed in the explicit enumeration of generators of $`\mathrm{E}_{7(5)}`$ displayed in table 4.
### 2.3 The maximal compact subalgebra $`\mathrm{SO}(3)_\mathrm{R}\times \mathrm{SO}(12)`$ and the basis of coset generators
Having explicitly constructed the real Lie algebra $`𝔾_{}=\mathrm{E}_{7(5)}`$ we can now consider its decomposition with respect to its maximal compact subalgebra $`\mathrm{SO}(3)_\mathrm{R}\times \mathrm{SO}(12)`$ which is to us the most relevant issue, since the final goal of our study is the construction of geodesic motions in the manifold (2.3). Being interested in the splitting:
$$𝔾_{}=𝕂$$
(2.34)
we proceed to establishing a canonical basis of generators for $`𝔾_{}`$ organized as it follows:
$$(A=1,\mathrm{},133);T_A=\{\begin{array}{cccc}T_i\hfill & =& _i\hfill & (i=1,\mathrm{},69)\hfill \\ T_{i+69}\hfill & =& 𝕂_i\hfill & (i=1,\mathrm{},64)\hfill \end{array}$$
(2.35)
where $`_i`$ is a basis of generators for the maximal compact subalgebra $`\mathrm{SO}(3)_\mathrm{R}\times \mathrm{SO}(12)`$ and $`𝕂_i`$ is a basis of generators for its orthogonal complement, namely for the tangent space to the manifold (2.3). With reference to table 4 our choice and ordering of the basis $`_i`$ is the following:
$`_{3i2}`$ $`=`$ $`\mathrm{i}{\displaystyle \frac{1}{\sqrt{2}}}H_{\beta _i}(i=1,2,3)`$
$`_{3i1}`$ $`=`$ $`E_{\beta _i}^+(i=1,2,3)`$
$`_{3i}`$ $`=`$ $`E_{\beta _i}^{}(i=1,2,3)`$
$`_{9+I}`$ $`=`$ $`𝐄_I^{}(I=1,\mathrm{},12)`$
$`_{21+A}`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}\left(_A^+\left(_A^+\right)^{}\right)(A=1,\mathrm{},24)`$
$`_{45+A}`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}\left(_A^{}\left(_A^{}\right)^{}\right)(A=1,\mathrm{},24)(A=1,\mathrm{},24)`$
Correspondingly, our choice and ordering for the coset generators $`𝕂_i`$ is displayed below:
$`𝕂_i`$ $`=`$ $`_i^{n.c.}(1,2,3)`$
$`𝕂_4`$ $`=`$ $`_4`$
$`𝕂_{4+I}`$ $`=`$ $`𝐄_I^+(I=1,\mathrm{},12)`$
$`𝕂_{16+A}`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}\left(_A^++\left(_A^+\right)^{}\right)(A=1,\mathrm{},24)`$
$`𝕂_{40+A}`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}\left(_A^{}+\left(_A^{}\right)^{}\right)(A=1,\mathrm{},24)`$
Let us make a few comments. In our ordering of the compact generators $`_i`$, the first nine generate a special subgroup, which we have already identified as the paint group:
$$𝔾_{\mathrm{paint}}=\mathrm{SO}(3)_\beta ^3\mathrm{SO}(3)_{\beta _1}\times \mathrm{SO}(3)_{\beta _2}\times \mathrm{SO}(3)_{\beta _3}\mathrm{SO}(3)_\mathrm{R}\times \mathrm{SO}(12)$$
(2.38)
This latter is associated with the three ”compact roots” defining the real section and will play an important role as automorphism algebra of the Solvable Lie Algebra $`Solv_{7(5)}`$ associated with the coset (2.3). It is appropriate to stress that the subgroup $`\mathrm{SO}(3)_R`$ is none of these three $`\mathrm{SO}(3)_{\beta _i}`$. On the contrary the subgroup $`\mathrm{SO}(3)_\beta ^3`$ sits inside $`\mathrm{SO}(12)`$. The subgroup $`\mathrm{SO}(3)_R`$, which commutes with the whole $`\mathrm{SO}(12)`$, is instead generated by the following uniquely determined linear combinations of the generators $`_i`$:
$`J_1^R`$ $`=`$ $`{\displaystyle \frac{1}{2\sqrt{2}}}\left(_{10}_{13}+_{16}_{21}\right)`$
$`J_2^R`$ $`=`$ $`{\displaystyle \frac{1}{2\sqrt{2}}}\left(_{12}_{14}+_{17}+_{20}\right)`$
$`J_3^R`$ $`=`$ $`{\displaystyle \frac{1}{2\sqrt{2}}}\left(_{11}_{15}+_{18}_{19}\right)`$
which close the standard commutation relations:
$$[J_i^R,J_j^R]=ϵ_{ijk}J_k^R$$
(2.40)
The ordering of the coset generators is obvious from equation (LABEL:Kbasis). First we have have listed the four non-compact Cartans, then non-compact combinations associated with the 12 roots that project onto the long roots of $`\mathrm{F}_4`$ with multiplicity one. Finally we have listed the non-compact combinations associated with the roots that project onto the short roots of $`\mathrm{F}_4`$ in exactly the same order as their compact analogues appear in the listing of $``$-generators. From the point of view of representation theory we know that the $`𝕂`$-space transforms as follows under $`\mathrm{SO}(3)_R\times \mathrm{SO}(12)`$:
$$𝕂=(\mathrm{𝟐},\mathrm{𝟑𝟐}_s)$$
(2.41)
and we could arrange the generators into linear combinations corresponding to the weights of the representation (2.41), yet this is not essential for our present purposes.
### 2.4 Structure of the Solvable Lie algebra
We can now come to the main point of our construction which relates to the solvable Lie algebra $`Solv_{\mathrm{E7}(5)}`$ whose corresponding group manifold is metrically equivalent to the coset manifold (2.3) and to its relation with the solvable Lie algebra $`Solv_{\mathrm{F4}(4)}`$ whose corresponding group manifold is metrically equivalent to the coset manifold (2.4).
First we define the solvable Lie algebra $`Solv_{\mathrm{E7}(5)}`$. This is easily done. Following the general theory recalled, for instance in ,, we know that $`Solv_{\mathrm{E7}(5)}`$ is the linear span of the non-compact Cartan generators plus those linear combinations of the positive root step operators which pertain to the considered real section. In practice this means:
$`Solv_{E7(5)}`$ $`=`$ $`\text{real span}\{_i^{n.c.},E^{\eta _I},_A^+,_A^{}\}`$ (2.42)
$`(i=1,\mathrm{},4;I=1,\mathrm{},12;A=1,\mathrm{},24)`$
As we know the solvable algebra $`Solv_{G/H}`$ associated with any non-compact coset $`G/H`$ has the great advantage that by exponentiation it provides a polynomial parametrization of the coset representative and hence of the scalar fields of supergravity. With respect to the traditional parametrization of cosets in terms of $`\mathrm{exp}(𝕂)`$ the advantages of the solvable parametrization are obtained at one price: while $`𝕂`$ is a representation of $``$, the solvable algebra $`Solv_{G/H}`$ is not. In the non maximally split case something very useful, however, occurs. Although $`Solv_{E7(5)}`$ is not a representation of the full compact group $`\mathrm{SO}(3)_R\times \mathrm{SO}(12)`$ yet it transforms covariantly under the action of a proper compact subgroup, the paint group, specifically $`𝔾_{paint}=\mathrm{SO}(3)_\beta ^3`$, defined in eq.(2.38). Indeed the decomposition of $`Solv_{E7(5)}`$ with respect to $`\mathrm{SO}(3)_\beta ^3`$ is:
$$Solv_{E7(5)}=16\times (\mathrm{𝟏},\mathrm{𝟏},\mathrm{𝟏})\mathrm{\hspace{0.17em}4}\times (\mathrm{𝟐},\mathrm{𝟐},\mathrm{𝟎})\mathrm{\hspace{0.17em}4}\times (\mathrm{𝟐},\mathrm{𝟎},\mathrm{𝟐})\mathrm{\hspace{0.17em}4}\times (\mathrm{𝟎},\mathrm{𝟐},\mathrm{𝟐})$$
(2.43)
and $`\mathrm{SO}(3)_\beta ^3`$ works as an automorphism group of the solvable Lie algebra. The sixteen singlets are the four Cartan generators plus the twelve $`E^{\eta _I}`$ step operators associated with long roots of $`\mathrm{F}_4`$. The forty-eight non-singlets, distributed in irrepses as described in eq.(2.43) are instead the generators associated with the $`\delta `$–roots that project onto the short ones of $`\mathrm{F}_4`$.
This covariant structure of the solvable Lie algebra $`Solv_{E7(5)}`$ is responsible for its relation with $`Solv_{F4(4)}`$ and for the painted billiard phenomenon. Let us see how.
We are interested in the structure constants of the Solvable Lie algebra which in turn determine the Nomizu connection and hence the $`1st`$ order equations for the tangent vector to the geodesic . Calling $`T_\mathrm{\Lambda }`$ a set of generators in the adjoint representation of the algebra we read off the structure constants from the equation:
$$[T_\mathrm{\Sigma },T_\mathrm{\Pi }]=C_{\mathrm{\Sigma }\mathrm{\Pi }}^\mathrm{\Lambda }T_\mathrm{\Lambda }$$
(2.44)
Let us first consider the solvable Lie algebra $`Solv_{F4(4)}`$ associated with the coset (2.4). This is a maximally split case and the structure of $`Solv_{F4(4)}`$ is the canonical one discussed in . We have 4 Cartan generators $`H_a`$ and $`24`$ positive roots that split into two subsets of $`12`$ long roots $`\alpha ^{\mathrm{}}`$ and $`12`$ short roots $`\alpha ^s`$. Calling $`\mathrm{\Delta }_{\mathrm{}}`$ and $`\mathrm{\Delta }_s`$ the two subsets we have the following structure:
$`\alpha ^{\mathrm{}},\beta ^{\mathrm{}}\mathrm{\Delta }_{\mathrm{}}`$ $`:`$ $`\alpha ^{\mathrm{}}+\beta ^{\mathrm{}}=\{\begin{array}{cc}\text{not a root or}\hfill & \\ \gamma ^{\mathrm{}}\mathrm{\Delta }_{\mathrm{}}\hfill & \end{array}`$
$`\alpha ^{\mathrm{}}\mathrm{\Delta }_{\mathrm{}}\text{and}\beta ^s\mathrm{\Delta }_s`$ $`:`$ $`\alpha ^{\mathrm{}}+\beta ^s=\{\begin{array}{cc}\text{not a root or}\hfill & \\ \gamma ^s\mathrm{\Delta }_s\hfill & \end{array}`$
$`\alpha ^s,\beta ^s\mathrm{\Delta }_s`$ $`:`$ $`\alpha ^s+\beta ^s=\{\begin{array}{cc}\text{not a root or}\hfill & \\ \gamma ^s\mathrm{\Delta }_s\text{or}\hfill & \\ \gamma ^{\mathrm{}}\mathrm{\Delta }_{\mathrm{}}\hfill & \end{array}`$ (2.45)
Consequently, let $`H_i`$ be the Cartan generators and $`E^\alpha ^{\mathrm{}}`$, $`E^{\alpha ^s}`$ be the step operators respectively associated with positive long and short roots. This set of operators completes a basis of 28 generators for the solvable Lie algebra $`Solv_{F_{4(4)}}`$. In view of eq.s (2.45) the possible structure constants are:
$$C_{\mathrm{\Sigma }\mathrm{\Pi }}^\mathrm{\Lambda }\{C_{j\beta ^{\mathrm{}}}^\alpha ^{\mathrm{}},C_{\beta ^{\mathrm{}}\gamma ^{\mathrm{}}}^\alpha ^{\mathrm{}},C_{\beta ^s\gamma ^s}^\alpha ^{\mathrm{}},C_{j\beta ^s}^{\alpha ^s},C_{\beta ^s\gamma ^s}^{\alpha ^s},C_{\beta ^{\mathrm{}}\gamma ^s}^{\alpha ^s}\}$$
(2.46)
and we further have:
$`C_{j\beta ^{\mathrm{}}}^\alpha ^{\mathrm{}}`$ $`=`$ $`\delta _\beta ^{\mathrm{}}^\alpha ^{\mathrm{}}\alpha _j^{\mathrm{}}`$
$`C_{j\beta ^s}^{\alpha ^s}`$ $`=`$ $`\delta _{\beta ^s}^{\alpha ^s}\alpha _j^s`$
$`C_{\beta ^{\mathrm{}}\gamma ^{\mathrm{}}}^\alpha ^{\mathrm{}}`$ $`=`$ $`\delta _{\beta ^{\mathrm{}}+\gamma ^{\mathrm{}}}^\alpha ^{\mathrm{}}N_{\beta ^{\mathrm{}}\gamma ^{\mathrm{}}}`$
$`C_{\beta ^{\mathrm{}}\gamma ^s}^{\alpha ^s}`$ $`=`$ $`\delta _{\beta ^{\mathrm{}}+\gamma ^s}^{\alpha ^s}N_{\beta ^{\mathrm{}}\gamma ^s}`$
$`C_{\beta ^s\gamma ^s}^\alpha ^{\mathrm{}}`$ $`=`$ $`\delta _{\beta ^s+\gamma ^s}^\alpha ^{\mathrm{}}N_{\beta ^s\gamma ^s}`$
$`C_{\beta ^s\gamma ^s}^{\alpha ^s}`$ $`=`$ $`\delta _{\beta ^s+\gamma ^s}^{\alpha ^s}N_{\beta ^s\gamma ^s}`$
where the matrix $`N_{\beta \gamma }`$, defined by the standard Cartan-Weyl commutation relations as given in eq.(B.23) of the appendix, or in table LABEL:paragonando, differs from zero only when the sum of the two roots $`\beta `$ and $`\gamma `$ is a root. Hence it suffices to know $`N_{\beta \gamma }`$ and the solvable Lie algebra structure constants are completely determined. In the following three tables (2.48), (2.49), (2.50) we exhibit the values of $`N_{\beta \gamma }`$ for the $`\mathrm{F}_{4(4)}`$ Lie algebra.
$$\underset{N_{\alpha ^{\mathrm{}}\beta ^{\mathrm{}}}}{\underset{}{\begin{array}{ccccccccccccc}& & & & & & & & & & & & \\ \alpha _1^{\mathrm{}}& \alpha _2^{\mathrm{}}& \alpha _3^{\mathrm{}}& \alpha _4^{\mathrm{}}& \alpha _5^{\mathrm{}}& \alpha _6^{\mathrm{}}& \alpha _7^{\mathrm{}}& \alpha _8^{\mathrm{}}& \alpha _9^{\mathrm{}}& \alpha _{10}^{\mathrm{}}& \alpha _{11}^{\mathrm{}}& \alpha _{12}^{\mathrm{}}& \\ & & & & & & & & & & & & \\ & & & & & & & & & & & & \\ 0& \sqrt{2}& 0& 0& \sqrt{2}& 0& 0& \sqrt{2}& 0& \sqrt{2}& 0& 0& \alpha _1^{\mathrm{}}\hfill \\ & & & & & & & & & & & & \\ \sqrt{2}& 0& 0& \sqrt{2}& 0& 0& \sqrt{2}& 0& 0& 0& \sqrt{2}& 0& \alpha _2^{\mathrm{}}\hfill \\ & & & & & & & & & & & & \\ 0& 0& 0& \sqrt{2}& 0& 0& \sqrt{2}& 0& 0& \sqrt{2}& 0& 0& \alpha _3^{\mathrm{}}\hfill \\ & & & & & & & & & & & & \\ 0& \sqrt{2}& \sqrt{2}& 0& 0& 0& 0& \sqrt{2}& \sqrt{2}& 0& 0& 0& \alpha _4^{\mathrm{}}\hfill \\ & & & & & & & & & & & & \\ \sqrt{2}& 0& 0& 0& 0& 0& \sqrt{2}& 0& \sqrt{2}& 0& 0& 0& \alpha _5^{\mathrm{}}\hfill \\ & & & & & & & & & & & & \\ 0& 0& 0& 0& 0& 0& \sqrt{2}& \sqrt{2}& 0& 0& 0& 0& \alpha _6^{\mathrm{}}\hfill \\ & & & & & & & & & & & & \\ 0& \sqrt{2}& \sqrt{2}& 0& \sqrt{2}& \sqrt{2}& 0& 0& 0& 0& 0& 0& \alpha _7^{\mathrm{}}\hfill \\ & & & & & & & & & & & & \\ \sqrt{2}& 0& 0& \sqrt{2}& 0& \sqrt{2}& 0& 0& 0& 0& 0& 0& \alpha _8^{\mathrm{}}\hfill \\ & & & & & & & & & & & & \\ 0& 0& 0& \sqrt{2}& \sqrt{2}& 0& 0& 0& 0& 0& 0& 0& \alpha _9^{\mathrm{}}\hfill \\ & & & & & & & & & & & & \\ \sqrt{2}& 0& \sqrt{2}& 0& 0& 0& 0& 0& 0& 0& 0& 0& \alpha _{10}^{\mathrm{}}\hfill \\ & & & & & & & & & & & & \\ 0& \sqrt{2}& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& \alpha _{11}^{\mathrm{}}\hfill \\ & & & & & & & & & & & & \\ 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& \alpha _{12}^{\mathrm{}}\hfill \end{array}}}$$
(2.48)
$$\underset{N_{\alpha ^{\mathrm{}}\beta ^s}}{\underset{}{\begin{array}{ccccccccccccc}& & & & & & & & & & & & \\ \alpha _1^s& \alpha _2^s& \alpha _3^s& \alpha _4^s& \alpha _5^s& \alpha _6^s& \alpha _7^s& \alpha _8^s& \alpha _9^s& \alpha _{10}^s& \alpha _{11}^s& \alpha _{12}^s& \\ & & & & & & & & & & & & \\ & & & & & & & & & & & & \\ 0& \sqrt{2}& 0& \sqrt{2}& 0& 0& 0& 0& \sqrt{2}& 0& 0& 0& \alpha _1^{\mathrm{}}\hfill \\ 0& 0& \sqrt{2}& 0& 0& \sqrt{2}& 0& \sqrt{2}& 0& 0& 0& 0& \alpha _2^{\mathrm{}}\hfill \\ 0& \sqrt{2}& 0& \sqrt{2}& 0& 0& 0& \sqrt{2}& 0& 0& 0& 0& \alpha _3^{\mathrm{}}\hfill \\ \sqrt{2}& 0& 0& 0& 0& 0& \sqrt{2}& 0& 0& 0& 0& 0& \alpha _4^{\mathrm{}}\hfill \\ \sqrt{2}& 0& 0& 0& 0& \sqrt{2}& 0& 0& 0& 0& 0& 0& \alpha _5^{\mathrm{}}\hfill \\ \sqrt{2}& 0& 0& \sqrt{2}& 0& 0& 0& 0& 0& 0& 0& 0& \alpha _6^{\mathrm{}}\hfill \\ 0& 0& 0& 0& \sqrt{2}& 0& 0& 0& 0& 0& 0& 0& \alpha _7^{\mathrm{}}\hfill \\ 0& 0& \sqrt{2}& 0& 0& 0& 0& 0& 0& 0& 0& 0& \alpha _8^{\mathrm{}}\hfill \\ 0& \sqrt{2}& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& \alpha _9^{\mathrm{}}\hfill \\ 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& \alpha _{10}^{\mathrm{}}\hfill \\ 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& \alpha _{11}^{\mathrm{}}\hfill \\ 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& 0& \alpha _{12}^{\mathrm{}}\hfill \end{array}}}$$
(2.49)
$$\underset{N_{\alpha ^s\beta ^s}}{\underset{}{\begin{array}{ccccccccccccc}& & & & & & & & & & & & \\ \alpha _1^s& \alpha _2^s& \alpha _3^s& \alpha _4^s& \alpha _5^s& \alpha _6^s& \alpha _7^s& \alpha _8^s& \alpha _9^s& \alpha _{10}^s& \alpha _{11}^s& \alpha _{12}^s& \\ & & & & & & & & & & & & \\ 0& 1& 1& 0& 1& 0& 0& \sqrt{2}& \sqrt{2}& \sqrt{2}& 1& 0& \alpha _1^s\hfill \\ & & & & & & & & & & & & \\ 1& 0& \sqrt{2}& 0& \sqrt{2}& 1& 1& 0& 0& 1& 0& \sqrt{2}& \alpha _2^s\hfill \\ & & & & & & & & & & & & \\ 1& \sqrt{2}& 0& 1& \sqrt{2}& 0& 1& 0& 1& 0& 0& \sqrt{2}& \alpha _3^s\hfill \\ & & & & & & & & & & & & \\ 0& 0& 1& 0& 1& \sqrt{2}& \sqrt{2}& 0& 0& 1& \sqrt{2}& 0& \alpha _4^s\hfill \\ & & & & & & & & & & & & \\ 1& \sqrt{2}& \sqrt{2}& 1& 0& 1& 0& 1& 0& 0& 0& \sqrt{2}& \alpha _5^s\hfill \\ & & & & & & & & & & & & \\ 0& 1& 0& \sqrt{2}& 1& 0& \sqrt{2}& 0& 1& 0& \sqrt{2}& 0& \alpha _6^s\hfill \\ & & & & & & & & & & & & \\ 0& 1& 1& \sqrt{2}& 0& \sqrt{2}& 0& 1& 0& 0& \sqrt{2}& 0& \alpha _7^s\hfill \\ & & & & & & & & & & & & \\ \sqrt{2}& 0& 0& 0& 1& 0& 1& 0& \sqrt{2}& \sqrt{2}& 0& 0& \alpha _8^s\hfill \\ & & & & & & & & & & & & \\ \sqrt{2}& 0& 1& 0& 0& 1& 0& \sqrt{2}& 0& \sqrt{2}& 0& 0& \alpha _9^s\hfill \\ & & & & & & & & & & & & \\ \sqrt{2}& 1& 0& 1& 0& 0& 0& \sqrt{2}& \sqrt{2}& 0& 0& 0& \alpha _{10}^s\hfill \\ & & & & & & & & & & & & \\ 1& 0& 0& \sqrt{2}& 0& \sqrt{2}& \sqrt{2}& 0& 0& 0& 0& 0& \alpha _{11}^s\hfill \\ & & & & & & & & & & & & \\ 0& \sqrt{2}& \sqrt{2}& 0& \sqrt{2}& 0& 0& 0& 0& 0& 0& 0& \alpha _{12}^s\hfill \end{array}}}$$
(2.50)
The ordering of long and short roots of the $`\mathrm{F}_4`$ system is that used in tables: 2 and 3. On the other hand the explicit determination of the tensor $`N_{\alpha \beta }`$ which appears in the standard Cartan-Weyl commutation relations was performed via the explicit construction of the fundamental $`26`$-dimensional representation of this Lie algebra. This construction is described in appendix B.
Now the exciting point about the solvable Lie algebra of the full non split coset (2.3) which contains all the degrees of freedom of supergravity is that it can be exhibited in terms of the structure constants of its split Tits Satake submanifold (2.4) by utilizing also the covariance with respect to the compact paint subgroup $`𝔾_{paint}=\mathrm{SO}(3)_\beta ^3`$. This is the result of an essential interplay of impressive elegance between the graded structure of the split Tits Satake algebra, which is non simply laced and for that reason contains a distinction between short and long roots, and the rearrangement of those roots of $`\mathrm{E}_7`$ which project onto the short ones of $`\mathrm{F}_4`$ into representations of the compact paint group $`\mathrm{SO}(3)_\beta ^3`$.
We already emphasized that, under the action of the paint group (2.38), the generators of the Solvable Lie algebra $`Solv_{E_{7(5)}}`$ decompose into the irreducible representations mentioned in eq. (2.43). Let us define the diagonal subgroup of the three $`\mathrm{SO}(3)_\beta `$:
$$𝔾_{\mathrm{paint}}^0\mathrm{SO}(3)_\beta ^{diag}=\text{diagonal}\left[\mathrm{SO}(3)_{\beta _1}\times \mathrm{SO}(3)_{\beta _2}\times \mathrm{SO}(3)_{\beta _3}\right]$$
(2.51)
since
$$\mathrm{𝟐}\mathrm{𝟐}=\mathrm{𝟏}\mathrm{𝟑}$$
(2.52)
holds true for $`\mathrm{SO}(3)`$ representations, it follows that under $`\mathrm{SO}(3)_\beta ^{diag}`$ the Lie algebra $`Solv_{E_{7(5)}}`$ decomposes as follows:
$$Solv_{E_{7(5)}}\stackrel{\mathrm{SO}(3)_\beta ^{diag}}{}\left(\underset{\text{Cartan}}{\underset{}{4}}+\underset{\text{long roots}}{\underset{}{12}}+\underset{\text{short roots}}{\underset{}{12}}\right)\times \mathrm{𝟏}\left(\underset{\text{short roots}}{\underset{}{12}}\right)\times \mathrm{𝟑}$$
(2.53)
Hence the representation $`𝐉`$ mentioned in eq.s (1.33) and (1.34) is $`𝐉=\mathrm{𝟑}`$, the triplet of $`\mathrm{SO}(3)_{diag}`$. The decomposition (2.53) is explicitly exhibited in tables 5, 6 and 7. In table 5, modulo some changes in normalization we list the non compact Cartan generators of $`\mathrm{E}_{7(7)}`$ which correspond to the full set of Cartan generators of $`\mathrm{F}_{4(4)}`$. In table 6 we define a set of generators $`\mathrm{\Phi }[\alpha ^s]`$ associated with the long roots of $`\mathrm{F}_{4(4)}`$, which are obviously given by the step operators $`E^\eta `$ of $`\mathrm{E}_{7(5)}`$ since the $`\eta `$–roots project on such short roots of $`\mathrm{F}_4`$. There are just some suitable $`\pm \sqrt{2}`$ factors in the normalization which are purposely chosen in order to make the relation between the two solvable Lie algebras clean. All these generators are singlets under $`\mathrm{SO}(3)_\beta ^3`$ and therefore also under $`\mathrm{SO}(3)_\beta ^{diag}`$. Finally in table 7 we list a set of four $`\mathrm{E}_{7(5)}`$ generators $`\mathrm{\Omega }_A[\alpha ^s]`$, ($`A=0,X,Y,Z`$), associated with each of the short roots $`\alpha ^s`$ of $`\mathrm{F}_{4(4)}`$. Indeed each such root is the image, in the projection, of four different $`\mathrm{E}_{7(5)}`$ roots, namely, the $`\delta _I^i`$ roots, displayed in table 3. Hence the four $`\mathrm{\Omega }_A[\alpha ^s]`$ operators are, with convenient normalization factors, step operators of $`\mathrm{E}_{7(5)}`$ corresponding to the $`\delta `$-roots. The normalization factors and the precise correspondence is chosen in such a way that the $`\mathrm{\Omega }_0`$ are singlets under $`\mathrm{SO}(3)_\beta ^{diag}`$, while $`\mathrm{\Omega }_{x=X,Y,Z}`$ form a triplet.
If we use these generators and, in order to avoid proliferation of symbols, we denote by the same letter the generator of the algebra and its dual one-form appearing in the Maurer Cartan equations:
$$dT^\mathrm{\Lambda }=\frac{1}{2}C_{\mathrm{\Sigma }\mathrm{\Pi }}^\mathrm{\Lambda }T^\mathrm{\Sigma }T^\mathrm{\Pi }$$
(2.54)
the structure constants of $`Solv_{E_{7(5)}}`$ can be exhibited by writing the following Maurer Cartan equations, which just contain the structure constants of the $`\mathrm{F}_{4(4)}`$ solvable algebra, discussed before, plus the quaternionic structure anticipated in table (LABEL:paragonando) of the introduction.
$`dH_i`$ $`=`$ $`0`$
$`d\mathrm{\Phi }[\alpha ^{\mathrm{}}]`$ $`=`$ $`C_{j\beta ^{\mathrm{}}}^\alpha ^{\mathrm{}}H_j\mathrm{\Phi }[\beta ^{\mathrm{}}]+\frac{1}{2}C_{\beta ^{\mathrm{}}\gamma ^{\mathrm{}}}^\alpha ^{\mathrm{}}\mathrm{\Phi }[\beta ^{\mathrm{}}]\mathrm{\Phi }[\gamma ^{\mathrm{}}]+`$
$`\frac{1}{2}C_{\beta ^s\gamma ^s}^\alpha ^{\mathrm{}}\left(\mathrm{\Omega }_0[\beta ^s]\mathrm{\Omega }_0[\gamma ^s]+\mathrm{\Omega }_x[\beta ^s]\mathrm{\Omega }_x[\gamma ^s]\right)`$
$`d\mathrm{\Omega }_0[\alpha ^s]`$ $`=`$ $`C_{j\beta ^s}^{\alpha ^s}H_j\mathrm{\Omega }_0[\beta ^s]`$
$`+\frac{1}{2}C_{\beta ^s\gamma ^s}^{\alpha ^s}\left(\mathrm{\Omega }_0[\beta ^s]\mathrm{\Omega }_0[\gamma ^s]+\mathrm{\Omega }_x[\beta ^s]\mathrm{\Omega }_x[\gamma ^s]\right)`$
$`+C_{\beta ^{\mathrm{}}\gamma ^s}^{\alpha ^s}\mathrm{\Phi }[\beta ^{\mathrm{}}]\mathrm{\Omega }_0[\gamma ^s]`$
$`d\mathrm{\Omega }_x[\alpha ^s]`$ $`=`$ $`C_{j\beta ^s}^{\alpha ^s}H_j\mathrm{\Omega }_x[\beta ^s]`$
$`+\frac{1}{2}C_{\beta ^s\gamma ^s}^{\alpha ^s}\left(\mathrm{\Omega }_0[\beta ^s]\mathrm{\Omega }_x[\gamma ^s]\mathrm{\Omega }_x[\beta ^s]\mathrm{\Omega }_0[\gamma ^s]ϵ^{xyz}\mathrm{\Omega }_y[\beta ^s]\mathrm{\Omega }_z[\gamma ^s]\right)`$
$`+C_{\beta ^{\mathrm{}}\gamma ^s}^{\alpha ^s}\mathrm{\Phi }[\beta ^{\mathrm{}}]\mathrm{\Omega }_x[\gamma ^s]`$
Equations (LABEL:E7MC) are just a short way of writing all commutation relations and exhibit the interplay between the graded structure of $`Solv_{F_4(4)}`$ and the structure of the paint group representation. Indeed we see the announced quaternionic structure! What actually happens is that the Cartan and long root generators are isomorphic in the two algebras, while the short root generators are promoted to quaternions while going from $`\mathrm{F}_{4(4)}`$ to $`\mathrm{E}_{7(5)}`$. We can write
$$F_{4(4)}E^{\alpha ^s}\mathrm{\Omega }_0[\alpha ^s]+\mathrm{\Omega }_X[\alpha ^s]j^X+\mathrm{\Omega }_Y[\alpha ^s]j^Y+\mathrm{\Omega }_Z[\alpha ^s]j^ZE_{7(5)}$$
(2.56)
where $`j^X,j^Y,j^Z`$ are the three quaternionic imaginary units.
This structure has very relevant consequences for the solution of the differential equations and for the billiard phenomenon. Since the Nomizu connection determining the first order equations for tangent vectors is completely determined by the structure constants of the solvable Lie algebra we can just adopt the following strategy:
Rather than considering the original problem associated with the non split manifold (2.3) we consider the problem associated with the split Tits Satake manifold (2.4), which can be solved along the lines of paper using, in particular, the compensator method to integrate the 1st order differential equations.
Once we have obtained a solution for the system described by the structure constants (2.46-LABEL:cunstanti) we also posses a particular solution for the system (LABEL:E7MC). It just correspond to setting the fields associated with $`\mathrm{\Omega }_{X,Y,Z}`$ to zero.
A large class of solutions of the system (LABEL:E7MC) can be obtained from the general solution of the $`\mathrm{F}_4`$ system with structure constants (2.46-LABEL:cunstanti) by means of global rotations of the paint group $`𝔾_{paint}=\mathrm{SO}(3)_\beta ^3`$.
From the point of view of the billiard picture we know that switching on root fields correspond to the introduction of dynamical walls on which the fictitious cosmic ball will bounce. The structure of the solvable algebra implies that the billiard chamber is just the Weyl chamber of $`\mathrm{F}_{4(4)}`$, yet certain dynamical walls are painted, namely occur in four copies constituting a quaternion. In explicit solutions we see also the color of the actual wall which is excited.
## 3 The first order equations for the tangent vectors
As we showed in , the field equations of the purely time dependent $`\sigma `$-model, which is what we are supposed to solve in our quest for time dependent solutions of supergravity, can be written as follows:
$$\dot{Y}^A+\mathrm{\Gamma }_{BC}^AY^BY^C=\mathrm{\hspace{0.17em}0}$$
(3.1)
where $`Y^A`$ denotes the purely time dependent tangent vectors to the geodesic in an anholonomic basis:
$$Y^A=\{\begin{array}{cc}Y^i=V_I^i\left(\varphi \right)\dot{\varphi }^Ii\text{CSA}\hfill & \\ Y^\alpha =\sqrt{2}V_I^\alpha \left(\varphi \right)\dot{\varphi }^Ii\text{positive root system }\mathrm{\Delta }_+\hfill & \end{array}$$
(3.2)
$`V_I^A\left(\varphi \right)d\varphi ^I`$ being the vielbein of the target manifold we are considering. In eq. (3.1) the symbol $`\mathrm{\Gamma }_{BC}^A`$ denotes the components of the Levi-Civita connection in the chosen anholonomic basis. Explicitly they are related to the components of the Levi Civita connection in an arbitrary holonomic basis by:
$$\mathrm{\Gamma }_{BC}^A=\mathrm{\Gamma }_{JK}^IV_I^AV_B^JV_C^K_K(V_J^A)V_B^JV_C^K,$$
(3.3)
where the inverse vielbein is defined in the usual way:
$$V_I^AV_B^I=\delta _B^A$$
(3.4)
The basic idea of , which was exploited together with the compensator method in order to construct explicit solutions, is the following. As already recalled in eq.(1.13), the connection $`\mathrm{\Gamma }_{BC}^A`$ can be identified with the Nomizu connection defined on a solvable Lie algebra, if the coset representative $`𝕃`$ from which we construct the vielbein is solvable, namely if it is represented as the exponential of the associated solvable Lie algebra $`Solv(\mathrm{U}/\mathrm{H})`$. In fact, as we can read in once we have defined over $`Solv`$ a non degenerate, positive definite symmetric form:
$`,`$ $`:SolvSolv`$
$`X,Y`$ $`=Y,X`$ (3.5)
whose lifting to the manifold produces the metric, the covariant derivative is defined through the Nomizu operator:
$$XSolv:𝕃_X:SolvSolv$$
(3.6)
so that
$$X,Y,ZSolv:\mathrm{\hspace{0.17em}2}Z,𝕃_XY=Z,[X,Y]X,[Y,Z]Y,[X,Z]$$
(3.7)
while the Riemann curvature 2-form is given by the commutator of two Nomizu operators:
$$R_Z^W(X,Y)=W,\left\{[𝕃_X,𝕃_Y]𝕃_{[X,Y]}\right\}Z$$
(3.8)
This implies that the covariant derivative explicitly reads:
$$𝕃_XY=\mathrm{\Gamma }_{XY}^ZZ$$
(3.9)
where
$$\mathrm{\Gamma }_{XY}^Z=\frac{1}{2}\left(Z,[X,Y]X,[Y,Z]Y,[X,Z]\right)\frac{1}{<Z,Z>}X,Y,ZSolv$$
(3.10)
Eq.(3.10) is true for any solvable Lie algebra. In the case of maximally non-compact, split algebras we can write a general form for $`\mathrm{\Gamma }_{XY}^Z`$, namely:
$`\mathrm{\Gamma }_{jk}^i`$ $`=`$ $`0`$
$`\mathrm{\Gamma }_{\alpha \beta }^i`$ $`=`$ $`\frac{1}{2}\left(E_\alpha ,[E_\beta ,H^i]E_\beta ,[E_\alpha ,H^i]\right)=\frac{1}{2}\alpha ^i\delta _{\alpha \beta }`$
$`\mathrm{\Gamma }_{ij}^\alpha `$ $`=`$ $`\mathrm{\Gamma }_{i\beta }^\alpha =\mathrm{\Gamma }_{j\alpha }^i=\mathrm{\hspace{0.17em}0}`$
$`\mathrm{\Gamma }_{\beta i}^\alpha `$ $`=`$ $`{\displaystyle \frac{1}{2}}\left(E^\alpha ,[E_\beta ,H_i]E_\beta ,[H_i,E^\alpha ]\right)=\alpha _i\delta _\beta ^\alpha `$
$`\mathrm{\Gamma }_{\alpha \beta }^{\alpha +\beta }`$ $`=`$ $`\mathrm{\Gamma }_{\beta \alpha }^{\alpha +\beta }={\displaystyle \frac{1}{2}}N_{\alpha \beta }`$
$`\mathrm{\Gamma }_{\alpha +\beta \beta }^\alpha `$ $`=`$ $`\mathrm{\Gamma }_{\beta \alpha +\beta }^\alpha ={\displaystyle \frac{1}{2}}N_{\alpha \beta }`$ (3.11)
where $`N^{\alpha \beta }`$ is defined by the commutator $`[E_\alpha ,E_\beta ]=N_{\alpha \beta }E_{\alpha +\beta }`$, as usual. In the case of $`\mathrm{F}_{4(4)}`$, the coefficients $`N^{\alpha \beta }`$ are read–off from the eq.s (2.49,2.48,2.50). The explicit form (3.11) follows from the choice of the non degenerate metric:
$`_i,_j`$ $`=`$ $`\mathrm{\hspace{0.17em}2}\delta _{ij}`$
$`_i,E_\alpha `$ $`=`$ $`0`$
$`E_\alpha ,E_\beta `$ $`=`$ $`\delta _{\alpha ,\beta }`$ (3.12)
$`_i,_j\mathrm{CSA}`$ and $`E_\alpha `$, step operator associated with a positive root $`\alpha \mathrm{\Delta }_+`$. For any other non split case, as that of $`Solv\left(\mathrm{E}_{7(5)}\right)`$, the Nomizu connection exists nonetheless although it does not take the form (3.11). It follows from eq.(3.10) upon the choice of an invariant positive metric on $`Solv`$ and the use of the structure constants of $`Solv`$ . Given the list of generators in tables 5, 6 and 7, the positive metric on $`Solv`$ is easily defined in full analogy with the definition of . The metric is diagonal and normalized as it follows:
$$\begin{array}{ccccccccc}\hfill H_i,H_j& =& 2\delta _{ij}\hfill & \hfill H_i,\mathrm{\Phi }[\alpha ^{\mathrm{}}]& =& 0\hfill & \hfill H_i,\mathrm{\Omega }_I[\alpha ^s]& =& 0\hfill \\ \hfill \mathrm{\Phi }[\alpha ^{\mathrm{}}],H_i& =& 0\hfill & \hfill \mathrm{\Phi }[\alpha ^{\mathrm{}}],\mathrm{\Phi }[\beta ^{\mathrm{}}]& =& \delta _{\alpha ^{\mathrm{}}\beta ^{\mathrm{}}}\hfill & \hfill \mathrm{\Phi }[\alpha ^{\mathrm{}}],\mathrm{\Omega }_I[\alpha ^s]& =& 0\hfill \\ \hfill \mathrm{\Omega }_I[\alpha ^s],H_i& =& 0\hfill & \hfill \mathrm{\Omega }_I[\alpha ^s],\mathrm{\Phi }[\beta ^{\mathrm{}}]& =& 0\hfill & \hfill \mathrm{\Omega }_I[\alpha ^s],\mathrm{\Omega }_J[\beta ^s]& =& \delta _{IJ}\delta _{\alpha ^s\beta ^s}\hfill \end{array}$$
(3.13)
The Nomizu connection can be explicitly calculated from eq.(3.10) reading the structure constants from the Maurer Cartan equations (LABEL:E7MC). In the case of all split algebras, the first order equations take the general form:
$`\dot{Y}^i`$ $`+`$ $`\frac{1}{2}{\displaystyle \underset{\alpha \mathrm{\Delta }_+}{}}\alpha ^iY_\alpha ^2=\mathrm{\hspace{0.17em}0}`$
$`\dot{Y}^\alpha `$ $`+`$ $`{\displaystyle \underset{\beta \mathrm{\Delta }_+}{}}N_{\alpha \beta }Y^\beta Y^{\alpha +\beta }\alpha _iY^iY^\alpha =\mathrm{\hspace{0.17em}0}`$ (3.14)
which follows from eq.(3.11). For the solvable Lie algebra of $`\mathrm{F}_{4(4)}`$ eq.(3) takes the form:
$`\dot{H}^i+\frac{1}{2}{\displaystyle \underset{\alpha _{\mathrm{}}\mathrm{\Delta }^{\mathrm{}}}{}}\alpha _{\mathrm{}}^i\mathrm{\Phi }[\alpha _{\mathrm{}}]^2+\frac{1}{2}{\displaystyle \underset{\alpha _s\mathrm{\Delta }^s}{}}\alpha _s^i\mathrm{\Omega }[\alpha _s]^2`$ $`=`$ $`0`$
$`\dot{\mathrm{\Phi }}[\alpha _{\mathrm{}}]\alpha _{\mathrm{}}H\mathrm{\Phi }[\alpha _{\mathrm{}}]+{\displaystyle \underset{\beta _{\mathrm{}}\mathrm{\Delta }^{\mathrm{}}}{}}N_{\alpha _{\mathrm{}}\beta _{\mathrm{}}}\mathrm{\Phi }[\beta _{\mathrm{}}]\mathrm{\Phi }[\alpha _{\mathrm{}}+\beta _{\mathrm{}}]`$
$`+{\displaystyle \underset{\beta _s\mathrm{\Delta }^s}{}}N_{\alpha _{\mathrm{}}\beta _s}\mathrm{\Omega }[\beta _s]\mathrm{\Omega }[\alpha _{\mathrm{}}+\beta _s]`$ $`=`$ $`0`$
$`\dot{\mathrm{\Omega }}[\alpha _s]\alpha _sH\mathrm{\Omega }[\alpha _s]+{\displaystyle \underset{\beta _s/\alpha _s+\beta _s\mathrm{\Delta }^{\mathrm{}}}{}}N_{\alpha _s\beta _s}\mathrm{\Omega }[\beta _s]\mathrm{\Phi }[\alpha _s+\beta _s]`$
$`+{\displaystyle \underset{\beta _s/\alpha _s+\beta _s\mathrm{\Delta }^s}{}}N_{\alpha _s\beta _s}\mathrm{\Omega }[\beta _s]\mathrm{\Omega }[\alpha _s+\beta _s]`$ $`=`$ $`0`$ (3.15)
where for notation simplicity we have given to the component $`Y^A`$ of the tangent vector $`\stackrel{}{Y}`$ along a generator $`\mathrm{T}_\mathrm{A}`$ of the solvable Lie algebra the same name as the generator itself.
In the case of the $`\mathrm{E}_{7(5)}`$ Lie algebra the first order equations for the tangent vector take the following form:
$`\dot{H}^i+\frac{1}{2}{\displaystyle \underset{\alpha _{\mathrm{}}\mathrm{\Delta }^{\mathrm{}}}{}}\alpha _{\mathrm{}}^i\mathrm{\Phi }[\alpha _{\mathrm{}}]^2+\frac{1}{2}{\displaystyle \underset{\alpha _s\mathrm{\Delta }^s}{}}\alpha _s^i\left(\mathrm{\Omega }_0[\alpha _s]^2+{\displaystyle \underset{i=1}{\overset{3}{}}}\mathrm{\Omega }_i[\alpha _s]^2\right)`$ $`=`$ $`0`$
$`\dot{\mathrm{\Phi }}[\alpha _{\mathrm{}}]\alpha _{\mathrm{}}H\mathrm{\Phi }[\alpha _{\mathrm{}}]+{\displaystyle \underset{\beta _{\mathrm{}}\mathrm{\Delta }^{\mathrm{}}}{}}N_{\alpha _{\mathrm{}}\beta _{\mathrm{}}}\mathrm{\Phi }[\beta _{\mathrm{}}]\mathrm{\Phi }[\alpha _{\mathrm{}}+\beta _{\mathrm{}}]`$
$`+{\displaystyle \underset{\beta _s\mathrm{\Delta }^s}{}}N_{\alpha _{\mathrm{}}\beta _s}\left(\mathrm{\Omega }_0[\beta _s]\mathrm{\Omega }_0[\alpha _{\mathrm{}}+\beta _s]+{\displaystyle \underset{i=1}{\overset{3}{}}}\mathrm{\Omega }_i[\beta _s]\mathrm{\Omega }_i[\alpha _{\mathrm{}}+\beta _s]\right)`$ $`=`$ $`0`$
$`\dot{\mathrm{\Omega }}_0[\alpha _s]\alpha _sH\mathrm{\Omega }_0[\alpha _s]+{\displaystyle \underset{\beta _s/\alpha _s+\beta _s\mathrm{\Delta }^{\mathrm{}}}{}}N_{\alpha _s\beta _s}\mathrm{\Omega }_0[\beta _s]\mathrm{\Phi }[\alpha _s+\beta _s]`$
$`+{\displaystyle \underset{\beta _s/\alpha _s+\beta _s\mathrm{\Delta }^s}{}}N_{\alpha _s\beta _s}\left(\mathrm{\Omega }_0[\beta _s]\mathrm{\Omega }_0[\alpha _s+\beta _s]+{\displaystyle \underset{i=1}{\overset{3}{}}}\mathrm{\Omega }_i[\beta _s]\mathrm{\Omega }_i[\alpha _s+\beta _s]\right)`$ $`=`$ $`0`$
$`\dot{\mathrm{\Omega }}_i[\alpha _s]\alpha _sH\mathrm{\Omega }_i[\alpha _s]+{\displaystyle \underset{\beta _s/\alpha _s+\beta _s\mathrm{\Delta }^{\mathrm{}}}{}}N_{\alpha _s\beta _s}\mathrm{\Omega }_i[\beta _s]\mathrm{\Phi }[\alpha _s+\beta _s]`$
$`+{\displaystyle \underset{\beta _s/\alpha _s+\beta _s\mathrm{\Delta }^s}{}}N_{\alpha _s\beta _s}(\frac{1}{2}\mathrm{\Omega }_0[\beta _s]\mathrm{\Omega }_i[\alpha _s+\beta _s]+\frac{1}{2}\mathrm{\Omega }_i[\beta _s]\mathrm{\Omega }_0[\alpha _s+\beta _s]`$
$`+ϵ_{ijk}\mathrm{\Omega }_j[\beta _s]\mathrm{\Omega }_k[\alpha _s+\beta _s])`$ $`=`$ $`0`$
As one sees from the above equations the $`\mathrm{E}_{7(5)}`$ differential system (LABEL:E7st) is consistently truncated to the $`\mathrm{F}_{4(4)}`$ system (3.15) by setting $`\mathrm{\Omega }_i[\alpha _s]=0`$, ($`i=1,2,3`$) and identifying $`\mathrm{\Omega }[\alpha _s]=\mathrm{\Omega }_0[\alpha _s]`$. Hence any solution of the $`\mathrm{F}_{4(4)}`$ equations is also a particular solution of the $`\mathrm{E}_{7(5)}`$ ones. On the other hand eq.s (LABEL:E7st) are invariant under the action of the paint group $`G_{paint}=\mathrm{SO}(3)^3`$.
In the next section, we utilize the compensator method to solve the first order equations (3) in the case of $`Solv_{F_{4(4)}}`$ and then we use the paint group to rotate these solutions to general solutions of the first order equations of $`Solv_{\mathrm{E}_{7(5)}}`$.
## 4 Solutions of the $`\mathrm{F}_{4(4)}`$ system by means of the compensator method
As we showed in in the split case the first order equations for the tangent vectors can be solved in the following way. First one considers the decomposition (2.34) of the full algebra and recalls that, in this case, the compact subalgebra is generated by $`E^\alpha E^\alpha `$ for all $`\alpha \mathrm{\Delta }_+`$. Secondly, one writes the decomposition of the left-invariant one–form on the coset manifold $`\mathrm{U}/\mathrm{H}`$ along the compact and non-compact generators:
$$\mathrm{\Omega }=𝕃^1d𝕃=V^A𝕂_A+\omega ^\alpha t_\alpha .$$
(4.1)
where $`V=V^A𝕂_A`$ corresponds to the coset manifold vielbein, while $`\omega =\omega ^\alpha t_\alpha `$ corresponds to the coset manifold $`\mathrm{H}`$–connection. One notes that the condition for the coset representative $`𝕃`$ to be solvable (namely to be the exponential of the solvable algebra) is expressed very simply by:
$$V^\alpha =\sqrt{2}\omega ^\alpha $$
(4.2)
Thirdly one derives the condition to be fulfilled by an H-gauge transformation:
$`𝕃`$ $``$ $`𝕃h=\overline{𝕃}`$
$`h`$ $`=`$ $`\mathrm{exp}\left[\theta ^\alpha t_\alpha \right]`$ (4.3)
in order for the solvable gauge (4.2) to be preserved. This latter reads as follows:
$$\frac{\sqrt{2}}{\text{tr}(t_\alpha ^2)}\text{tr}\left(h^1(\theta )dh(\theta )t_\alpha \right)=V^\beta \left(A(\theta )_\beta ^\alpha +D(\theta )_\beta ^\alpha \right)+V^iD(\theta )_i^\alpha $$
(4.4)
In the above equation the matrix $`A(\theta )`$ is the adjoint representation of $`h\mathrm{H}`$ and $`D(\theta )`$ is the $`D`$–representation of the same group element which acts on the complementary space $`𝕂`$ and which depends case to case:
$`h^1t_\alpha h`$ $`=`$ $`A(\theta )_\alpha ^\beta t_\beta `$
$`h^1𝕂_Ah`$ $`=`$ $`D(\theta )_A^B𝕂_B`$ (4.5)
In our example of $`\mathrm{F}_{4(4)}`$ the compact group is $`\mathrm{H}=\mathrm{SU}(2)_\mathrm{R}\times \mathrm{Usp}(6)`$ and the representation $`D`$ is the $`(\mathrm{𝟏𝟒},\mathrm{𝟐})`$
A simple solution of the first order equations (3) is easily obtained by setting $`Y^\alpha =0`$ and $`Y^i=c^i=\text{const}`$, namely we can begin with a constant vector in the direction of the $`\mathrm{CSA}`$. Such a solution is named the normal form of the tangent vector. In the language of billiard dynamics it corresponds to a fictitious ball that moves on a straight line with a constant velocity. All other solutions of eq.s (3) can be obtained from the normal form solution by means of successive rotations of the compact group, with parameters $`\theta [t]`$ satisfying the differential equation (4.4). The advantage of this method, emphasized in where we introduced it, is that at each successive rotation we obtain an equation which is fully integrable in terms of the integral of the previous ones.
In this paper we just present one solution of the $`\mathrm{F}_{4(4)}`$ system which is fully analytical and already sufficiently complicated to display the billiard dynamics with several bounces. Our solution is obtained by applying $`5`$ successive rotations to a normal form vector that we parametrize in terms of $`4`$ constants. We use an intelligent parametrization which is the following one:
$$𝐘_{\text{nf}}=\{\frac{\omega _5}{2}\omega _6,\frac{\omega _5}{2},\frac{\omega _5}{2}\omega _6+\omega _7,\frac{\omega _{24}}{4}\}$$
(4.6)
The way $`𝐘_{\text{nf}}`$ is parametrized and the name given to the constants $`\omega _{24,7,6,5}`$ anticipate their physical interpretation in the solution we are going to derive. Indeed we obtain our solution by writing:
$$𝐘(t)=𝐘_{\text{nf}}\mathrm{exp}\left[^{24}\theta _{24}(t)\right]\mathrm{exp}\left[^7\theta _7(t)\right]\mathrm{exp}\left[^6\theta _6(t)\right]\mathrm{exp}\left[^5\theta _5(t)\right]\mathrm{exp}\left[^4\theta _4(t)\right]$$
(4.7)
where
$$^n=\frac{1}{2}\left(E^{\varpi [n]}E^{\varpi [n]}\right)$$
(4.8)
are the compact generators associated with the $`\mathrm{F}_4`$ roots numbered as in table 1. The explicit form of the solution of the differential equations (4.4) for the five rotation angles is given below:
$`\theta _4(t)=\mathrm{arccos}\left[{\displaystyle \frac{e^{\left(t\tau _4\right)\left(\omega _5+2\omega _6\omega _7\right)}\sqrt{1+e^{2\left(t\tau _7\right)\omega _7}}}{\sqrt{1+e^{2\left(t\tau _6\right)\omega _6}+e^{2\left(t\tau _7\right)\omega _7}+e^{2\left(t\tau _4\right)\left(\omega _5+2\omega _6\omega _7\right)}\left(1+e^{2\left(t\tau _7\right)\omega _7}\right)}}}\right]`$
$`\theta _5(t)={\displaystyle \frac{\text{arccsc}\left[\sqrt{1+e^{\left(t\tau _5\right)\omega _5}\mathrm{cosh}(\frac{\left(t\tau _{24}\right)\omega _{24}}{2})\text{sech}(\left(t\tau _7\right)\omega _7)^2}\right]}{\sqrt{2}}}`$
$`\theta _6(t)=\text{arccot}\left[{\displaystyle \frac{e^{\left(t\tau _6\right)\omega _6}}{\sqrt{1+e^{2\left(t\tau _7\right)\omega _7}}}}\right]`$
$`\theta _7(t)=\mathrm{arccos}\left[{\displaystyle \frac{e^{\left(t\tau _7\right)\omega _7}}{\sqrt{1+e^{2\left(t\tau _7\right)\omega _7}}}}\right]`$
$`\theta _{24}(t){\displaystyle \frac{\text{arccsc}\left[\sqrt{1+e^{\left(t\tau _{24}\right)\omega _{24}}}\right]}{\sqrt{2}}}`$
(4.9)
Let us briefly mention the explicit form of the eq.s (4.4) from which we obtained the above result. As specified in eq.(4.7) we perform the compact rotations in the order $`24\mathrm{\hspace{0.17em}7}\mathrm{\hspace{0.17em}6}\mathrm{\hspace{0.17em}5}\mathrm{\hspace{0.17em}4}`$. This is not a random choice but it is motivated by the fact that in this way the differential equations (4.4) come up triangular: in other words at each step we just obtain a differential equation for the angle $`\theta _i(t)`$ that depends only on the previously determined angles $`\theta _j(t)`$. The systematic study of triangulization of the differential system (4.4) for general algebras is postponed to a later publication. We just note that one typically has to perform rotations along roots arranged in reverse order with respect to their height but this criterion, although necessary is not yet sufficient in full generality. A complete solution requires a more systematic analysis. It is however fairly easy, by computer calculations, to obtain ordered lists of root-rotations that have the triangular property and hence lead to exact analytic solutions by quadratures. In the case of the present algebra we have already found lists of up to eight successive such rotations and the solution we present with five rotations has just been chosen as an illustrative example of the physical and analytical mechanisms occurring in the differential system (4.4).
This being clarified, we present the differential equations obtained for the angles $`\theta _i(t)`$ in succession. Performing the first rotation around the highest root $`\varpi _{24}`$ we obtain:
$$\frac{\mathrm{sin}\left[2\sqrt{2}\theta _{24}(t)\right]\omega _{24}}{4\sqrt{2}}+\dot{\theta }_{24}(t)=0$$
(4.10)
Performing the second rotation around the root $`\varpi _7`$ we get:
$$\frac{\mathrm{sin}\left[2\theta _7(t)\right]\omega _7}{2}+\dot{\theta }_7(t)$$
(4.11)
which is still an independent equation. Performing the third rotation around the root $`\varpi _6`$ we get a differential equation that depends on the solution of the previous two:
$$\frac{\mathrm{sin}[2\theta _6(t)]\omega _6}{2}\frac{\mathrm{cos}^2\left[\theta _7(t)\right]\mathrm{sin}\left[2\theta _6(t)\right]\omega _7}{2}+\dot{\theta }_6(t)$$
(4.12)
The same happens when we introduce the fourth rotation around $`\varpi _5`$. We obtain:
$`2\sqrt{2}\mathrm{sin}\left[2\sqrt{2}\theta _5(t)\right]\omega _54\sqrt{2}\mathrm{cos}\left[2\theta _7(t)\right]\mathrm{sin}\left[2\sqrt{2}\theta _5(t)\right]\omega _7`$
$`+\sqrt{2}\mathrm{cos}\left[2\sqrt{2}\theta _{24}(t)\right]\mathrm{sin}\left[2\sqrt{2}\theta _5(t)\right]\omega _{24}+16\dot{\theta }_5(t)=\mathrm{\hspace{0.17em}0}`$ (4.13)
Finally, when we perform the $`5`$-th rotation, we get:
$`8\mathrm{sin}\left[2\theta _4(t)\right]\omega _54(3+\mathrm{cos}\left[2\theta _6(t)\right]]\mathrm{sin}\left[2\theta _4(t)\right]\omega _66\mathrm{sin}\left[2\theta _4(t)\right]\omega _7`$
$`+2\mathrm{cos}\left[2\theta _6(t)\right]\mathrm{sin}\left[2\theta _4(t)\right]\omega _7+\mathrm{cos}\left[2\left(\theta _6(t)\theta _7(t)\right)\right]\mathrm{sin}\left[2\theta _4(t)\right]\omega _7`$
$`+2\mathrm{cos}\left[2\theta _7(t)\right]\mathrm{sin}\left[2\theta _4(t)\right]\omega _7+\mathrm{cos}\left[2\theta _6(t)+2\theta _7(t)\right]\mathrm{sin}\left[2\theta _4(t)\right]\omega _7`$
$`+16\dot{\theta }_4(t)=\mathrm{\hspace{0.17em}0}`$ (4.14)
The explicit functions $`\theta _i(t)`$ displayed in eq.(4.9) are the general integral of the system of eq.s (4.10-4.14) where the integration constants are represented by the fixed instants of time $`\tau _i`$. The physical interpretation of these constants becomes clear when we investigate the properties of the scalar fields $`h_i(t)`$ lying in the Cartan subalgebra of $`\mathrm{F}_{4(4)}`$ and eventually representing, after dimensional oxidation the logarithms of the scale factors in the various available dimensions. Following the discussion of we can write
$`h_i(t)`$ $`=`$ $`{\displaystyle H_i(t^{})𝑑t^{}}`$
$`H_i(t)`$ $``$ $`Y^i(t)\text{in CSA}`$ (4.15)
where $`H_i(t)`$ are obtained by inserting the explicit solutions (4.9) into eq.(4.7) and then extracting the first four components of such a vector. We have an analytic although cumbersome expression for the $`H_i(t)`$ in the case of all the considered rotations, yet the next integration to $`h_i(t)`$ can no longer be analytically done if we include all the thetas $`\theta _{24}(t),\theta _7(t),\theta _6(t),\theta _5(t)`$ and $`\theta _4(t)`$. For this reason we prefer to discuss the features of billiard dynamics by considering the simpler solution obtained by including only the first three rotations $`\theta _{24}(t),\theta _7(t)`$ and $`\theta _6(t)`$. This solution is already complicated enough to display the phenomena we want to illustrate yet it still leads to manageable analytic formulae. Explicitly we obtain:
$`H_1(t)`$ $`=`$ $`{\displaystyle \frac{\omega _5}{2}}\omega _6+\mathrm{sin}^2\left[\theta _7(t)\right]\omega _7`$
$`=`$ $`{\displaystyle \frac{\omega _5}{2}}\omega _6+{\displaystyle \frac{\omega _7}{1+e^{2\left(t\tau _7\right)\omega _7}}}`$
$`H_2(t)`$ $`=`$ $`\frac{1}{2}\left[\omega _5+2\mathrm{sin}^2\left[\theta _6(t)\right]\left(\omega _6\mathrm{cos}^2\left[\theta _7(t)\right]\omega _7\right)\right]`$
$`=`$ $`{\displaystyle \frac{\left(1+e^{2\left(t\tau _6\right)\omega _6}+e^{2\left(t\tau _7\right)\omega _7}\right)\omega _5+2\left(\left(1+e^{2\left(t\tau _7\right)\omega _7}\right)\omega _6e^{2\left(t\tau _7\right)\omega _7}\omega _7\right)}{2\left(1+e^{2\left(t\tau _6\right)\omega _6}+e^{2\left(t\tau _7\right)\omega _7}\right)}}`$
$`H_3(t)`$ $`=`$ $`{\displaystyle \frac{\omega _5}{2}}+\mathrm{cos}^2\left[\theta _6(t)\right]\left(\omega _6+\mathrm{cos}^2\left[\theta _7(t)\right]\omega _7\right)`$
$`=`$ $`{\displaystyle \frac{\omega _5}{2}}+{\displaystyle \frac{e^{2\left(t\tau _6\right)\omega _6}\left(\left(e^{2t\omega _7}e^{2\tau _7\omega _7}\right)\omega _6+e^{2t\omega _7}\omega _7\right)}{\left(1+e^{2\left(t\tau _6\right)\omega _6}+e^{2\left(t\tau _7\right)\omega _7}\right)\left(e^{2t\omega _7}+e^{2\tau _7\omega _7}\right)}}`$
$`H_4(t)`$ $`=`$ $`{\displaystyle \frac{1}{4}}\left(\mathrm{cos}\left[2\sqrt{2}\theta _{24}(t)\right]\omega _{24}\right)={\displaystyle \frac{1}{4}}\left(1+{\displaystyle \frac{2}{1+e^{\left(t\tau _{24}\right)\omega _{24}}}}\right)\omega _{24}`$
Considering now the five roots involved in this calculation:
$$\begin{array}{ccccccc}\varpi _{24}& =& \{0,0,0,2\}\hfill & ;& \varpi _7& =& \{1,0,1,0\}\hfill \\ \varpi _6& =& \{0,1,1,0\}\hfill & ;& \varpi _5& =& \{1,1,1,1\}\hfill \\ \varpi _4& =& \{1,1,0,0\}\hfill & & & & \end{array}$$
(4.17)
we can evaluate the five projections of the Cartan fields in the direction of the five relevant roots and we get:
$`h_{\varpi _{24}}(t)`$ $`{\displaystyle \stackrel{}{\varpi }_{24}\stackrel{}{H}(t)}`$ $`=\mathrm{log}(1+e^{\left(t\tau _{24}\right)\omega _{24}})+{\displaystyle \frac{t\omega _{24}}{2}}`$
$`h_{\varpi _7}(t)`$ $`{\displaystyle \stackrel{}{\varpi }_7\stackrel{}{H}(t)}`$ $`=\mathrm{log}(1+e^{2\left(t\tau _7\right)\omega _7})+{\displaystyle \frac{\mathrm{log}(1+e^{2\left(t\tau _6\right)\omega _6}+e^{2\left(t\tau _7\right)\omega _7})}{2}}`$
$`t\omega _6+t\omega _7`$
$`h_{\varpi _6}(t)`$ $`{\displaystyle \stackrel{}{\varpi }_6\stackrel{}{H}(t)}`$ $`={\displaystyle \frac{\mathrm{log}(1+e^{2\left(t\tau _7\right)\omega _7})}{2}}\mathrm{log}(1+e^{2\left(t\tau _6\right)\omega _6}+e^{2\left(t\tau _7\right)\omega _7})+t\omega _6`$
$`h_{\varpi _5}(t)`$ $`{\displaystyle \stackrel{}{\varpi }_5\stackrel{}{H}(t)}`$ $`={\displaystyle \frac{1}{4}}[4\mathrm{log}(1+e^{2\left(t\tau _7\right)\omega _7})2\mathrm{log}(1+e^{\left(t\tau _{24}\right)\omega _{24}})`$
$`2t\omega _54t\omega _7+t\omega _{24}]`$
$`h_{\varpi _4}(t)`$ $`{\displaystyle \stackrel{}{\varpi }_4\stackrel{}{H}(t)}`$ $`={\displaystyle \frac{1}{2}}[\mathrm{log}(1+e^{2\left(t\tau _7\right)\omega _7})+\mathrm{log}(1+e^{2\left(t\tau _6\right)\omega _6}+e^{2\left(t\tau _7\right)\omega _7})`$
$`2t\omega _54t\omega _6+2t\omega _7]`$
In the first of eq.(LABEL:pensata) we can observe the basic building block for the smooth realization of the cosmic billiard behaviour. It is is given by the function:
$$G(t|\omega ,\tau )\mathrm{log}(1+e^{\left(t\tau \right)\omega })+\frac{t\omega }{2}$$
(4.19)
For $`t\tau <<0`$, that is for asymptotically early times the behaviour of $`G(t|\omega ,\tau )`$ is the following one:
$$G(t|\omega ,\tau )\frac{t\omega }{2}$$
(4.20)
which corresponds to the motion of a fictitious ball with constant velocity $`v=\omega /2`$. For asymptotically late times, namely for $`t\tau >>0`$, we have instead:
$$G(t|\omega ,\tau )\frac{t\omega }{2}$$
(4.21)
which corresponds to the motion of a fictitious ball with inverted constant velocity $`v=\omega /2`$. The inversion, namely the bounce occurs in the region $`t\tau 0`$. Hence it appears that the integration constants $`\tau _i`$ introduced in our solution have precisely the meaning of instant of times at which bounces occur. Furthermore each bounce occurs precisely on the the wall orthogonal to each root around which we have made compact rotations while using the compensator algorithm. On the other hand, the components of the normal form solution (4.6) in the CSA direction have the interpretation of components of the velocity vector of the fictitious cosmic ball in the asymptotically early times prior to the first cosmic bounce. Each new rotation introduces a new bounce.
This is illustrated in fig. 3 where the Cartan fields along the five relevant roots are plotted for the solution with three rotations namely in the case of eq.(LABEL:pensata). Here we clearly see three bounces, due to the three rotations introduced.
## 5 Uplifting of $`\mathrm{F}_{4(4)}`$ solutions to $`\mathrm{E}_{7(5)}`$ and painted walls
Now that we have obtained explicit solutions of the first order system (3.15) by means of the compensator method, we can appreciate the role of the paint group, $`\mathrm{G}_{\mathrm{paint}}=\mathrm{SO}(3)^3`$ since rotations of this latter applied to the $`\mathrm{F}_{4(4)}`$ solution generate non trivial solutions of the differential system (LABEL:E7st).
To illustrate the mechanism with an explicit and manageable example we consider the $`\mathrm{F}_{4(4)}`$ solution based on the three rotations angles $`\theta _{24}(t),\theta _7(t),\theta _6(t)`$ for which we have already written the time dependence of the Cartan fields in eq.(LABEL:Hvetti) and we complete it by writing also the time dependence of the root components of the tangent vector. In this case the only non vanishing root fields are $`\mathrm{\Phi }_{12}`$ and $`\mathrm{\Omega }_{3,4,8}`$ respectively associated with the long root $`\alpha _{12}^{\mathrm{}}=2y_4`$ and with the short roots $`\alpha _{3,4,8}^s=y_2+y_3,y_1y_3,y_1+y_2`$. The time dependence of these fields in the considered solution is given by:
$`\mathrm{\Phi }[\alpha _{12}^{\mathrm{}}](t)`$ $`=`$ $`\left({\displaystyle \frac{e^{\frac{\left(t\tau _{24}\right)\omega _{24}}{2}}\omega _{24}}{\sqrt{2}\left(1+e^{\left(t\tau _{24}\right)\omega _{24}}\right)}}\right)`$
$`\mathrm{\Omega }[\alpha _3^s](t)`$ $`=`$ $`{\displaystyle \frac{2e^{\left(t\tau _6\right)\omega _6}\sqrt{1+e^{2\left(t\tau _7\right)\omega _7}}\left(\left(e^{2t\omega _7}+e^{2\tau _7\omega _7}\right)\omega _6e^{2t\omega _7}\omega _7\right)}{\left(1+e^{2\left(t\tau _6\right)\omega _6}+e^{2\left(t\tau _7\right)\omega _7}\right)\left(e^{2t\omega _7}+e^{2\tau _7\omega _7}\right)}}`$
$`\mathrm{\Omega }[\alpha _4^s](t)`$ $`=`$ $`{\displaystyle \frac{2e^{\left(t\tau _7\right)\omega _7}\omega _7}{\left(1+e^{2\left(t\tau _7\right)\omega _7}\right)\sqrt{1+\frac{1+e^{2\left(t\tau _7\right)\omega _7}}{e^{2\left(t\tau _6\right)\omega _6}}}}}`$
$`\mathrm{\Omega }[\alpha _8^s](t)`$ $`=`$ $`{\displaystyle \frac{2e^{\left(t+\tau _6\right)\omega _6+\left(t\tau _7\right)\omega _7},\omega _7}{\sqrt{\left(1+e^{2\left(t\tau _7\right)\omega _7}\right)\left(1+e^{2\left(t+\tau _6\right)\omega _6}\left(1+e^{2\left(t\tau _7\right)\omega _7}\right)\right)}}}`$ (5.1)
and it is displayed in fig.4
Uplifting this solution to an $`\mathrm{E}_{7(5)}`$ solution is done by identifying the Cartan and the long root fields of the two systems and then by identifying:
$`\mathrm{\Omega }[\alpha _3^s](t)`$ $`=`$ $`\mathrm{\Omega }_0[\alpha _3^s](t)`$
$`\mathrm{\Omega }[\alpha _4^s](t)`$ $`=`$ $`\mathrm{\Omega }_0[\alpha _4^s](t)`$
$`\mathrm{\Omega }[\alpha _8^s](t)`$ $`=`$ $`\mathrm{\Omega }_0[\alpha _8^s](t)`$
Next we can rotate the so obtained solution with any element of the nine parameter paint group $`\mathrm{SO}(3)^3`$ whose generators are the first nine operators $`_i`$ described in eq.s(LABEL:Hbasis). For instance we can apply a rotation of a constant angle $`\psi _5`$ along the $`5`$-th generator, namely along $`E_{\beta _2}^+`$. The result putting all-together is given by the Cartan fields in eq.(LABEL:Hvetti) and by the following root fields:
$`\mathrm{\Phi }[\alpha _{12}^{\mathrm{}}](t)`$ $`=`$ $`\left({\displaystyle \frac{e^{\frac{\left(t\tau _{24}\right)\omega _{24}}{2}}\omega _{24}}{\sqrt{2}\left(1+e^{\left(t\tau _{24}\right)\omega _{24}}\right)}}\right)`$
$`\mathrm{\Omega }_0[\alpha _3^s](t)`$ $`=`$ $`{\displaystyle \frac{2e^{\left(t\tau _6\right)\omega _6}\sqrt{1+e^{2\left(t\tau _7\right)\omega _7}}\mathrm{cos}(\frac{\psi _5}{2\sqrt{2}})\left(\left(e^{2t\omega _7}+e^{2\tau _7\omega _7}\right)\omega _6e^{2t\omega _7}\omega _7\right)}{\left(1+e^{2\left(t\tau _6\right)\omega _6}+e^{2\left(t\tau _7\right)\omega _7}\right)\left(e^{2t\omega _7}+e^{2\tau _7\omega _7}\right)}}`$
$`\mathrm{\Omega }_Z[\alpha _3^s](t)`$ $`=`$ $`{\displaystyle \frac{2e^{\left(t\tau _6\right)\omega _6}\sqrt{1+e^{2\left(t\tau _7\right)\omega _7}}\mathrm{sin}(\frac{\psi _5}{2\sqrt{2}})\left(\left(e^{2t\omega _7}+e^{2\tau _7\omega _7}\right)\omega _6e^{2t\omega _7}\omega _7\right)}{\left(1+e^{2\left(t\tau _6\right)\omega _6}+e^{2\left(t\tau _7\right)\omega _7}\right)\left(e^{2t\omega _7}+e^{2\tau _7\omega _7}\right)}}`$
$`\mathrm{\Omega }_0[\alpha _4^s](t)`$ $`=`$ $`{\displaystyle \frac{2e^{\left(t\tau _7\right)\omega _7}\omega _7}{\left(1+e^{2\left(t\tau _7\right)\omega _7}\right)\sqrt{1+\frac{1+e^{2\left(t\tau _7\right)\omega _7}}{e^{2\left(t\tau _6\right)\omega _6}}}}}`$
$`\mathrm{\Omega }_0[\alpha _8^s](t)`$ $`=`$ $`{\displaystyle \frac{2e^{\left(t+\tau _6\right)\omega _6+\left(t\tau _7\right)\omega _7}\mathrm{cos}(\frac{\psi _5}{2\sqrt{2}})\omega _7}{\sqrt{\left(1+e^{2\left(t\tau _7\right)\omega _7}\right)\left(1+e^{2\left(t+\tau _6\right)\omega _6}\left(1+e^{2\left(t\tau _7\right)\omega _7}\right)\right)}}}`$
$`\mathrm{\Omega }_Z[\alpha _8^s](t)`$ $`=`$ $`{\displaystyle \frac{2e^{\left(t+\tau _6\right)\omega _6+\left(t\tau _7\right)\omega _7}\mathrm{sin}(\frac{\psi _5}{2\sqrt{2}})\omega _7}{\sqrt{\left(1+e^{2\left(t\tau _7\right)\omega _7}\right)\left(1+e^{2\left(t+\tau _6\right)\omega _6}\left(1+e^{2\left(t\tau _7\right)\omega _7}\right)\right)}}}`$ (5.3)
Inserting eq.s (LABEL:Hvetti) and eq.s(5.3) into the differential equations (LABEL:E7st) one can patiently verify that they are all satisfied for any value of the angle $`\psi _5`$.
We could continue with more complicated rotations, but the lesson taught by this example should already be sufficiently clear. In this solution the time dependence of $`\mathrm{\Omega }_Z[\alpha _8^s](t)`$ and $`\mathrm{\Omega }_0[\alpha _8^s](t)`$ is exactly the same and the ratio of these two fields is the constant factor $`\mathrm{tan}\left[\frac{\psi _5}{2\sqrt{2}}\right]`$. Similarly for the fields $`\mathrm{\Omega }_Z[\alpha _3^s](t)`$ and $`\mathrm{\Omega }_0[\alpha _3^s](t)`$. Hence it appears that the dynamical walls which raise and lower and cause the bounces of the cosmological factors are just those displayed by the Tits Satake projection of the supergravity scalar manifold, namely, the quaternionic manifold $`\mathrm{F}_{4(4)}/\mathrm{Usp}(6)\times \mathrm{SU}(2)`$, rather than $`\mathrm{E}_{7(5)}/\mathrm{SO}(12)\times \mathrm{SO}(3)`$ in $`D=3`$, and after dynamical oxidation to $`D=4`$, the special Kähler manifold $`\mathrm{Sp}(6,\mathrm{R})/\mathrm{SU}(3)\times \mathrm{U}(1)`$ rather than $`\mathrm{SO}^{}(12)/\mathrm{SU}(6)\times \mathrm{U}(1)`$. Indeed as we have pointed out in and recalled in table 8 taken from , the Tits Satake projection commutes with the c-map produced by the dimensional reduction à la Ehlers and we have the correspondence:
$$\begin{array}{ccc}\hfill \text{adj}(𝕌_{D=3})& =& \text{adj}(𝕌_{D=4})\text{adj}(\mathrm{SL}(2,)_\mathrm{E})W_{(2,W)}\hfill \\ & & \\ \hfill \text{adj}(𝕌_{D=3}^{TS})& =& \text{adj}(𝕌_{D=4}^{TS})\text{adj}(\mathrm{SL}(2,)_\mathrm{E})W_{(2,W^{TS})}\hfill \end{array}$$
(5.4)
where $`\mathrm{SL}(2,)_\mathrm{E}`$ is the Ehlers group coming from the dimensional reduction of pure gravity and $`W`$ denotes the symplectic representation to which vector fields are assigned in $`D=4`$.
What is actually preserved in the c-map is the paint group $`\mathrm{G}_{\mathrm{paint}}`$.
Hence the dynamical walls are those associated with the Tits Satake projected model but they come, in the true supergravity theory, in painted copies, for instance, within the context of our example, the copy $`\mathrm{\Omega }_0`$ and the copy $`\mathrm{\Omega }_Z`$. The paint group rotates these copies into each other. The explicit form taken by the diagram (5.4) in the worked out example studied by the present paper is:
$$\begin{array}{ccc}\hfill \text{adj}(\mathrm{E}_{7(5)})& =& \text{adj}(\mathrm{SO}^{}(12))\text{adj}(\mathrm{SL}(2,)_\mathrm{E})(\mathrm{𝟐},\mathrm{𝟑𝟐}_s)\hfill \\ & & \\ \hfill \text{adj}(\mathrm{F}_{4(4)})& =& \text{adj}(\mathrm{Sp}(6,)\text{adj}(\mathrm{SL}(2,)_\mathrm{E})(\mathrm{𝟐},\mathrm{𝟏𝟒})\hfill \end{array}$$
(5.5)
The representation $`\mathrm{𝟏𝟒}`$ of $`\mathrm{Sp}(6,)`$ is that of an antisymmetric symplectic traceless tensor:
$$\begin{array}{ccc}\text{dim}_{\mathrm{Sp}(6,)}& \begin{array}{c}\\ \stackrel{~}{\text{ }}\end{array}& =\mathbf{\hspace{0.17em}14}\end{array}$$
(5.6)
On the other hand the invariance of the paint group through dimensional reduction and oxidation can be easily checked as follows. First of all we note that $`\mathrm{G}_{\mathrm{paint}}=\mathrm{SO}(3)^3`$ is both a subgroup of $`\mathrm{SO}(12)`$ and of $`\mathrm{SU}(6)`$ as it is easily verified through the subgroup chain:
$$\begin{array}{ccccccc}\mathrm{SO}(12)& & \mathrm{SU}(6)& & \mathrm{SU}(4)& \times & \mathrm{SU}(2)\\ & & & & & & \\ \mathrm{SO}(12)& & \mathrm{SU}(6)& & \mathrm{SO}(6)& \times & \mathrm{SO}(3)\\ & & & & & & \\ \mathrm{SO}(12)& & \mathrm{SU}(6)& & \mathrm{SO}(4)\times \mathrm{SO}(2)& \times & \mathrm{SO}(3)\\ & & & & & & \\ \mathrm{SO}(12)& & \mathrm{SU}(6)& & \mathrm{SO}(3)\times \mathrm{SO}(3)\times \mathrm{SO}(2)& \times & \mathrm{SO}(3)\end{array}$$
(5.7)
Secondly we note that the non maximally split coset manifold (2.1) appearing in $`D=4`$ has dimension $`30`$ and rank $`3`$. This means that out of the $`30`$ positive roots there are three, $`\beta _1`$, $`\beta _2`$ and $`\beta _3`$ that are orthogonal to the $`3`$ non–compact Cartan generators. Together with the three compact Cartan generators they make up the same $`\mathrm{SO}(3)^3`$ paint Lie algebra as in the $`D=3`$ case. Furthermore the remaining $`27`$ non compact roots which together with the $`3`$ non compact Cartans span the solvable Lie algebra of $`Solv\left(\mathrm{SO}^{}(12)/\mathrm{SU}(6)\times \mathrm{U}(1)\right)`$ are accounted for in the following way. The Tits Satake projection of $`\mathrm{SO}^{}(12)`$ is the maximally split Lie algebra $`\mathrm{Sp}(6,)`$. This latter is non simply laced and has $`9`$ positive roots which distribute into $`3`$ long ones ($`\alpha ^{\mathrm{}}=2ϵ_i`$ ($`i=1,\mathrm{},3`$) and $`6`$ short ones ($`\alpha ^s=ϵ_i\pm ϵ_j`$, $`i<j,i,j=1,2,3`$). Just as in $`D=3`$ the long roots of $`\mathrm{Sp}(6,)`$ correspond to roots of $`\mathrm{SO}^{}(12)`$ that are singlets under the paint group, while the short ones correspond to roots of $`\mathrm{SO}^{}(12)`$ which arrange into the following $`12`$ dimensional representation
$$\mathrm{𝟏𝟐}_{paint}=(2,2,0)(2,0,2)(0,2,2)$$
(5.8)
In $`D=3`$ we have $`4`$-copies of the representation $`\mathrm{𝟏𝟐}_{paint}`$ while in $`D=4`$ we just have $`2`$–copies of the same. It is instructive to compare how the total number of roots is retrieved in the two cases:
$$\begin{array}{ccccccc}\text{ }\mathrm{\#}\text{ of }\mathrm{E}_7\text{ roots = }63\hfill & =& \underset{compact}{\underset{}{3}}& +& \underset{long}{\underset{}{12}}& +& 4\times \mathrm{𝟏𝟐}_{paint}\hfill \\ & =& \underset{compact}{\underset{}{3}}& +& \underset{long}{\underset{}{12}}& +& \underset{short}{\underset{}{12}}\times \left(\mathrm{𝟏}+\mathrm{𝟑}\right)_{\mathrm{SO}(3)_{\mathrm{diag}}}\hfill \\ \text{ }\mathrm{\#}\text{ of }\mathrm{SO}^{}(12)\text{ roots = }30\hfill & =& \underset{compact}{\underset{}{3}}& +& \underset{long}{\underset{}{3}}& +& 2\times \mathrm{𝟏𝟐}_{paint}\hfill \\ & =& \underset{compact}{\underset{}{3}}& +& \underset{long}{\underset{}{3}}& +& \underset{short}{\underset{}{6}}\times \left(\mathrm{𝟏}+\mathrm{𝟑}\right)_{\mathrm{SO}(3)_{\mathrm{diag}}}\hfill \end{array}$$
In eq.(5) the second and fourth lines recall that each of the short roots of either $`\mathrm{F}_{4(4)}`$ or $`\mathrm{Sp}(6,)`$ has $`4`$ preimages in the $`D=4`$ algebra which arrange into a triplet plus a singlet with respect to the diagonal subgroup $`\mathrm{SO}(3)_{diag}=𝔾_{paint}^0`$. This shows how the structure of the paint group filters through the dimensional reduction. We can analyze this phenomenon also at the level of the symplectic representation $`𝐖`$ to which the vector fields are assigned. For the full $`𝒩=6`$ supergravity model, this representation is the spinorial $`\mathrm{𝟑𝟐}_s`$ of $`\mathrm{SO}^{}(12)`$. Following the general discussion given in our recent paper , the $`32`$ weights of this representation are in one to one correspondence with the roots of $`\mathrm{E}_7`$ which have non vanishing grading with respect to the highest root $`\psi =\alpha [63]`$ in the numbering of appendix A. This root set subdivides into $`32=8+24`$ where $`8`$ roots are Tits-Satake projected into $`8`$ long roots of $`\mathrm{F}_{4(4)}`$, while $`24`$ are Tits-Satake projected into $`6`$ short roots of the same. The $`14`$-dimensional representation of $`\mathrm{Sp}(6,)`$ is just made by these $`8+6`$ roots of $`\mathrm{F}_{4(4)}`$ which have non vanishing grading with respect to its own highest root $`\psi _{TS}`$. Indeed, as we have noted in the Tits Satake projection of the highest root is the highest root of the target algebra.
The above discussion provides the essential tools to perform the dimensional oxidation of the solutions we have found to full fledged solutions of supergravity models in $`D=4`$ or even in higher dimensions. We do not address this issue in the present paper leaving it for further publications where we also plan to provide a systematic analysis of the Tits Satake projection for all supergravity theories linking it to the properties of the compactification manifolds.
We deem that the present detailed case-study has illustrated the role of the dimensional reduction invariant paint group in reducing the study of billiard dynamics to simpler maximally split cosets.
## 6 Conclusions
In this paper we have considered one of the two necessary extensions of the analysis of smooth cosmic billiards initiated by us in : that to supergravity theories with lesser supersymmetry than the maximal one. The other necessary extension is the further reduction to $`D<3`$ dimensions, which we have recently addressed in by studying the universal field–theoretical mechanism of the affine extension. As displayed in the systematic analysis presented by us in , lesser supersymmetry involves a general new feature: cosets that are not maximally split and correspond to non maximally non compact real sections of their isometry algebra. For these cosets the compensator method devised by us in cannot be directly applied. Yet we have shown in this paper that the dynamical problem can be reduced, also in these cases, to a problem which can be solved with the compensator method. In fact the original system can be reduced to a maximally split one, performing the Tits–Satake projection of the original Lie algebra. The solutions of the projected system (that can be easily found with the compensator method) are also solutions of the complete one. Moreover, we also showed that many other solutions can be obtained from these by global rotations of a suitable compact group that we named *paint group*. Although we do not have the general integral for these cases, we showed how to obtain a large class of solutions that, probably, are the most relevant from the physical point of view.
Tits Satake projection of the original Lie algebra has emerged as a central token in discussing cosmic billiards for lesser supersymmetry. We have illustrated its role by an in depth analysis of a specific example that of $`𝒩=6`$ supergravity. Through this case-study we were able to extrapolate the main general features that apply to all supergravity models and which we plan to study systematically in a future publication. In particular we have elucidated the key role of the $`𝔾_{\mathrm{paint}}`$ group, a notion not yet introduced in the literature and leading to the idea of painted billiard walls. The main property of $`𝔾_{\mathrm{paint}}`$ is that it commutes with the c-map, namely with dimensional reduction. Hence it filters through dimensional oxidation and can be retrieved in higher dimensional supergravity.
The main research line that streams from our results is the analysis of the Tits Satake projection and of its kernel (the paint group) in more general contexts, in particular in the context of generic special Kähler geometry, of which our case study is also an example (see for instance for a review). Furthermore keeping in mind the generic interpretation of the scalar manifold $`_{scalar}`$ as moduli space for the geometry of the compactification manifold, a Calabi Yau $`_{CY}`$ in the case where $`_{scalar}`$ is special Kählerian, it is challenging to obtain the interpretation of the Tits Satake projection at the level of the compact manifold geometry. This, as already stressed, we plan to do in the immediate future.
It is at the same time quite interesting to consider the interplay between the Tits Satake projection and the gauging of supergravity models which is also on agenda.
As we have illustrated in this paper we can easily obtain smooth realizations of the cosmic billiard with several bounces. The number of these bounces, however, is finite, as long as we deal with finite algebras, namely as long as we discuss higher dimensional configurations from a $`D=3`$ perspective. This is so because bounces are created, as we have shown, by compact group rotations along different generators and there is a finite number of them if the number of roots is finite. In order to see infinite bounces and may be chaos we have to have infinitely many roots, namely we have to look at higher dimensional supergravity from a $`D=2`$ or $`D=1`$ perspective. This requires to consider the affine or hyperbolic Kač–Moody extensions which we have addressed in the recent paper . Yet, as touched upon there and readdressed in the present example here, the Tits Satake projection commutes with the affine extension and in general with dimensional reduction, which preserves the structure of the paint group $`𝔾_{\mathrm{paint}}`$. Hence a door has been open how to paint walls and roots also in Kač-Moody algebras.
## Appendix A Listing the positive roots of $`\mathrm{E}_7`$
Listing of all positive roots of $`\mathrm{E}_7`$. The first column gives the Dynkin label, the last gives the euclidean components of the root vectors
$`\begin{array}{ccccc}\alpha [1]& =& \{1,0,0,0,0,0,0\}& =& \{1,1,0,0,0,0,0\}\hfill \\ \alpha [2]& =& \{0,1,0,0,0,0,0\}& =& \{0,1,1,0,0,0,0\}\hfill \\ \alpha [3]& =& \{0,0,1,0,0,0,0\}& =& \{0,0,1,1,0,0,0\}\hfill \\ \alpha [4]& =& \{0,0,0,1,0,0,0\}& =& \{0,0,0,1,1,0,0\}\hfill \\ \alpha [5]& =& \{0,0,0,0,1,0,0\}& =& \{0,0,0,0,1,1,0\}\hfill \\ \alpha [6]& =& \{0,0,0,0,0,1,0\}& =& \{0,0,0,0,1,1,0\}\hfill \\ \alpha [7]& =& \{0,0,0,0,0,0,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [8]& =& \{1,1,0,0,0,0,0\}& =& \{1,0,1,0,0,0,0\}\hfill \\ \alpha [9]& =& \{0,1,1,0,0,0,0\}& =& \{0,1,0,1,0,0,0\}\hfill \\ \alpha [10]& =& \{0,0,1,1,0,0,0\}& =& \{0,0,1,0,1,0,0\}\hfill \\ \alpha [11]& =& \{0,0,0,1,1,0,0\}& =& \{0,0,0,1,0,1,0\}\hfill \\ \alpha [12]& =& \{0,0,0,1,0,1,0\}& =& \{0,0,0,1,0,1,0\}\hfill \\ \alpha [13]& =& \{0,0,0,0,0,1,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [14]& =& \{1,1,1,0,0,0,0\}& =& \{1,0,0,1,0,0,0\}\hfill \\ \alpha [15]& =& \{0,1,1,1,0,0,0\}& =& \{0,1,0,0,1,0,0\}\hfill \\ \alpha [16]& =& \{0,0,1,1,1,0,0\}& =& \{0,0,1,0,0,1,0\}\hfill \\ \alpha [17]& =& \{0,0,1,1,0,1,0\}& =& \{0,0,1,0,0,1,0\}\hfill \\ \alpha [18]& =& \{0,0,0,1,0,1,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [19]& =& \{0,0,0,1,1,1,0\}& =& \{0,0,0,1,1,0,0\}\hfill \\ \alpha [20]& =& \{1,1,1,1,0,0,0\}& =& \{1,0,0,0,1,0,0\}\hfill \\ \alpha [21]& =& \{0,1,1,1,1,0,0\}& =& \{0,1,0,0,0,1,0\}\hfill \\ \alpha [22]& =& \{0,1,1,1,0,1,0\}& =& \{0,1,0,0,0,1,0\}\hfill \\ \alpha [23]& =& \{0,0,1,1,0,1,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [24]& =& \{0,0,1,1,1,1,0\}& =& \{0,0,1,0,1,0,0\}\hfill \\ \alpha [25]& =& \{0,0,0,1,1,1,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [26]& =& \{1,1,1,1,1,0,0\}& =& \{1,0,0,0,0,1,0\}\hfill \\ \alpha [27]& =& \{1,1,1,1,0,1,0\}& =& \{1,0,0,0,0,1,0\}\hfill \\ \alpha [28]& =& \{0,1,1,1,0,1,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [29]& =& \{0,1,1,1,1,1,0\}& =& \{0,1,0,0,1,0,0\}\hfill \\ \alpha [30]& =& \{0,0,1,1,1,1,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [31]& =& \{0,0,1,2,1,1,0\}& =& \{0,0,1,1,0,0,0\}\hfill \end{array}`$
$`\begin{array}{ccccc}\alpha [32]& =& \{1,1,1,1,0,1,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [33]& =& \{1,1,1,1,1,1,0\}& =& \{1,0,0,0,1,0,0\}\hfill \\ \alpha [34]& =& \{0,1,1,1,1,1,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [35]& =& \{0,1,1,2,1,1,0\}& =& \{0,1,0,1,0,0,0\}\hfill \\ \alpha [36]& =& \{0,0,1,2,1,1,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [37]& =& \{1,1,1,1,1,1,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [38]& =& \{1,1,1,2,1,1,0\}& =& \{1,0,0,1,0,0,0\}\hfill \\ \alpha [39]& =& \{0,1,1,2,1,1,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [40]& =& \{0,1,2,2,1,1,0\}& =& \{0,1,1,0,0,0,0\}\hfill \\ \alpha [41]& =& \{0,0,1,2,1,2,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [42]& =& \{1,1,1,2,1,1,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [43]& =& \{1,1,2,2,1,1,0\}& =& \{1,0,1,0,0,0,0\}\hfill \\ \alpha [44]& =& \{0,1,1,2,1,2,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [45]& =& \{0,1,2,2,1,1,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [46]& =& \{1,1,1,2,1,2,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [47]& =& \{1,1,2,2,1,1,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [48]& =& \{1,2,2,2,1,1,0\}& =& \{1,1,0,0,0,0,0\}\hfill \\ \alpha [49]& =& \{0,1,2,2,1,2,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [50]& =& \{1,1,2,2,1,2,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [51]& =& \{1,2,2,2,1,1,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [52]& =& \{0,1,2,3,1,2,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [53]& =& \{1,1,2,3,1,2,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [54]& =& \{1,2,2,2,1,2,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [55]& =& \{0,1,2,3,2,2,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [56]& =& \{1,1,2,3,2,2,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [57]& =& \{1,2,2,3,1,2,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [58]& =& \{1,2,2,3,2,2,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [59]& =& \{1,2,3,3,1,2,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [60]& =& \{1,2,3,3,2,2,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [61]& =& \{1,2,3,4,2,2,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [62]& =& \{1,2,3,4,2,3,1\}& =& \{\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{\sqrt{2}}\}\hfill \\ \alpha [63]& =& \{1,2,3,4,2,3,2\}& =& \{0,0,0,0,0,0,\sqrt{2}\}\hfill \end{array}`$
## Appendix B Explicit construction of the fundamental and adjoint representation of $`\mathrm{F}_{4(4)}`$
The semisimple complex Lie algebra $`\mathrm{F}_4`$ is defined by the Dynkin diagram in figure 2 and a set of simple roots corresponding to such diagram was provided in eq.(2.18). A complete list of the 24 positive roots was given in table 1. The roots were further subdivided into the set of 12 long roots and 12 short roots respectively listed in table 2 and 3 where their correspondence with $`\mathrm{E}_7`$ roots was spelled out. The adjoint representation of $`\mathrm{F}_4`$ is $`52`$–dimensional, while its fundamental representation is $`26`$–dimensional. This dimensionality is true for all real sections of the Lie algebra but the explicit structure of the representation is quite different in each real section. Here we are interested in the maximally split real section $`\mathrm{F}_{4(4)}`$. For such a section we have a maximal, regularly embedded, subgroup $`\mathrm{SO}(5,4)\mathrm{F}_{4(4)}`$. The decomposition of the representations with respect to this particular subgroup is the essential instrument for their actual construction. For the adjoint representation we have the decomposition:
$$\underset{\text{adj}F_{4(4)}}{\underset{}{\mathrm{𝟓𝟐}}}\stackrel{\mathrm{SO}(5,4)}{}\underset{\text{adj}\mathrm{SO}(5,4)}{\underset{}{\mathrm{𝟑𝟔}}}\underset{\text{spinor of }\mathrm{SO}(5,4)}{\underset{}{\mathrm{𝟏𝟔}}}$$
(B.1)
while for the fundamental one we have:
$$\underset{\text{fundamental }F_{4(4)}}{\underset{}{\mathrm{𝟐𝟔}}}\stackrel{\mathrm{SO}(5,4)}{}\underset{\text{vector of }\mathrm{SO}(5,4)}{\underset{}{\mathrm{𝟗}}}\underset{\text{spinor of }\mathrm{SO}(5,4)}{\underset{}{\mathrm{𝟏𝟔}}}\underset{\text{singlet of }\mathrm{SO}(5,4)}{\underset{}{\mathrm{𝟏}}}$$
(B.2)
In view of this, we fix our conventions for the $`\mathrm{SO}(5,4)`$ invariant metric as it follows
$$\eta _{AB}=\text{diag}\{+,+,+,+,+,,,,\}$$
(B.3)
and we perform an explicit construction of the $`16\times 16`$ dimensional gamma matrices which satisfy the Clifford algebra
$$\{\mathrm{\Gamma }_A,\mathrm{\Gamma }_B\}=\eta _{AB}\mathbf{\hspace{0.17em}1}$$
(B.4)
and are all completely real. This construction is provided by the following tensor products:
$`\mathrm{\Gamma }_1`$ $`=`$ $`\sigma _1\sigma _3\mathrm{𝟏}\mathrm{𝟏}`$
$`\mathrm{\Gamma }_2`$ $`=`$ $`\sigma _3\sigma _3\mathrm{𝟏}\mathrm{𝟏}`$
$`\mathrm{\Gamma }_3`$ $`=`$ $`\mathrm{𝟏}\sigma _1\mathrm{𝟏}\sigma _1`$
$`\mathrm{\Gamma }_4`$ $`=`$ $`\mathrm{𝟏}\sigma _1\sigma _1\sigma _3`$
$`\mathrm{\Gamma }_5`$ $`=`$ $`\mathrm{𝟏}\sigma _1\sigma _3\sigma _3`$
$`\mathrm{\Gamma }_6`$ $`=`$ $`\mathrm{𝟏}\mathrm{i}\sigma _2\mathrm{𝟏}\mathrm{𝟏}`$
$`\mathrm{\Gamma }_7`$ $`=`$ $`\mathrm{𝟏}\sigma _1\mathrm{i}\sigma _2\sigma _3`$
$`\mathrm{\Gamma }_8`$ $`=`$ $`\mathrm{𝟏}\sigma _1\mathrm{𝟏}\mathrm{i}\sigma _2`$
$`\mathrm{\Gamma }_9`$ $`=`$ $`\mathrm{i}\sigma _2\sigma _3\mathrm{𝟏}\mathrm{𝟏}`$ (B.5)
where by $`\sigma _i`$ we have denoted the standard Pauli matrices:
$$\sigma _1=\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right);\sigma _2=\left(\begin{array}{cc}0& \mathrm{i}\\ \mathrm{i}& 0\end{array}\right);\sigma _3=\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right)$$
(B.6)
Moreover we introduce the $`C_+`$ charge conjugation matrix, such that:
$`C_+`$ $`=`$ $`\left(C_+\right)^T;C_+^2=\mathrm{𝟏}`$
$`C_+\mathrm{\Gamma }_AC_+`$ $`=`$ $`\left(\mathrm{\Gamma }_A\right)^T`$ (B.7)
In the basis of eq. (B.5) the explicit form of $`C_+`$ is given by:
$$C_+=\mathrm{i}\sigma _2\sigma _1\mathrm{i}\sigma _2\sigma _1$$
(B.8)
Then we define the usual generators $`J_{AB}=J_{BA}`$ of the pseudorthogonal algebra $`\mathrm{SO}(5,4)`$ satisfying the commutation relations:
$$[J_{AB},J_{CD}]=\eta _{BC}J_{AD}\eta _{AC}J_{BD}\eta _{BD}J_{AC}+\eta _{AD}J_{BC}$$
(B.9)
and we construct the spinor and the vector representations by respectively setting:
$$J_{CD}^s=\frac{1}{4}[\mathrm{\Gamma }_C,\mathrm{\Gamma }_D];\left(J_{CD}^v\right)_A^B=\eta _{CA}\delta _D^B\eta _{DA}\delta _C^B$$
(B.10)
In this way if $`v_A`$ denote the components of a vector, $`\xi `$ those of a real spinor and $`ϵ^{AB}=ϵ^{BA}`$ are the parameters of an infinitesimal $`\mathrm{SO}(5,4)`$ rotation we can write the $`\mathrm{SO}(5,4)`$ transformation as follows:
$$\delta _{\mathrm{SO}(5,4)}v_A=2ϵ_{AB}v^B;\delta _{\mathrm{SO}(5,4)}\xi =\frac{1}{2}ϵ^{AB}\mathrm{\Gamma }_{AB}\xi $$
(B.11)
where indices are raised and lowered with the metric (B.3). Furthermore we introduce the conjugate spinors via the position:
$$\overline{\xi }\xi ^TC_+$$
(B.12)
With these preliminaries, we are now a position to write the explicit form of the $`26`$-dimensional fundamental representation of $`\mathrm{F}_{4(4)}`$ and in this way to construct also its structure constants and hence its adjoint representation, which is our main goal.
According to eq.(B.1) the parameters of an $`\mathrm{F}_{4(4)}`$ representation are given by an anti-symmetric tensor $`ϵ_{AB}`$ and a spinor $`q`$. On the other hand a vector in the $`26`$–dimensional representation is specified by a collection of three objects, namely a scalar $`\varphi `$, a vector $`v_A`$ and a spinor $`\xi `$. The representation is constructed if we specify the $`\mathrm{F}_{4(4)}`$ transformation of these objects. This is done by writing:
$$\delta _{F_{4(4)}}\left(\begin{array}{c}\varphi \\ v_A\\ \xi \end{array}\right)\left[ϵ^{AB}T_{AB}+\overline{q}Q\right]\left(\begin{array}{c}\varphi \\ v_A\\ \xi \end{array}\right)=\left(\begin{array}{c}\overline{q}\xi \hfill \\ 2ϵ_{AB}v^B+a\overline{q}\mathrm{\Gamma }_A\xi \hfill \\ \frac{1}{2}ϵ^{AB}\mathrm{\Gamma }_{AB}\mathrm{\hspace{0.17em}3}\varphi q\frac{1}{a}v^A\mathrm{\Gamma }_A\xi \hfill \end{array}\right)$$
(B.13)
where $`a`$ is a numerical real arbitrary but non-null parameter. Eq.(B.13) defines the generators $`T_{AB}`$ and $`Q`$ as $`26\times 26`$ matrices and therefore completely specifies the fundamental representation of the Lie algebra $`\mathrm{F}_{4(4)}`$. Explicitly we have:
$$T_{AB}=\left(\begin{array}{ccc}0& 0& 0\\ & & \\ 0& J_{AB}^v& 0\\ & & \\ 0& 0& J_{AB}^s\end{array}\right)$$
(B.14)
and
$$Q_\alpha =\left(\begin{array}{ccc}0& 0& \delta _\alpha ^\beta \\ & & \\ 0& 0& a\left(\mathrm{\Gamma }_A\right)_\alpha ^\beta \\ & & \\ 3\delta _\alpha ^\beta & \frac{1}{a}\left(\mathrm{\Gamma }_B\right)_\alpha ^\beta & 0\end{array}\right)$$
(B.15)
and the Lie algebra commutation relations are evaluated to be the following ones:
$`[T_{AB},T_{CD}]`$ $`=`$ $`\eta _{BC}T_{AD}\eta _{AC}T_{BD}\eta _{BD}T_{AC}+\eta _{AD}T_{BC}`$
$`[T_{AB},Q]`$ $`=`$ $`\frac{1}{2}\mathrm{\Gamma }_{AB}Q`$
$`[Q_\alpha ,Q_\beta ]`$ $`=`$ $`{\displaystyle \frac{1}{12}}\left(C_+\mathrm{\Gamma }^{AB}\right)_{\alpha \beta }T_{AB}`$ (B.16)
Eq.(B.16), together with eq.s(B.5) and eq.(B.7) provides an explicit numerical construction of the structure constants of the maximally split $`\mathrm{F}_{4(4)}`$ Lie algebra. What we still have to do is to identify the relation between the tensorial basis of generators in eq. (B.16) and the Cartan-Weyl basis in terms of Cartan generators and step operators. To this effect let us enumerate the $`52`$ generators of $`\mathrm{F}_{4(4)}`$ in the tensorial representation according to the following table:
$$\begin{array}{cccccccc}& & & & & & & \\ \hfill \mathrm{\Omega }_1=& T_{12}\hfill & \hfill \mathrm{\Omega }_2=& T_{13}\hfill & \hfill \mathrm{\Omega }_3=& T_{14}\hfill & \hfill \mathrm{\Omega }_4=& T_{15}\hfill \\ \hfill \mathrm{\Omega }_5=& T_{16}\hfill & \hfill \mathrm{\Omega }_6=& T_{17}\hfill & \hfill \mathrm{\Omega }_7=& T_{18}\hfill & \hfill \mathrm{\Omega }_8=& T_{19}\hfill \\ \hfill \mathrm{\Omega }_9=& T_{23}\hfill & \hfill \mathrm{\Omega }_{10}=& T_{24}\hfill & \hfill \mathrm{\Omega }_{11}=& T_{25}\hfill & \hfill \mathrm{\Omega }_{12}=& T_{26}\hfill \\ \hfill \mathrm{\Omega }_{13}=& T_{27}\hfill & \hfill \mathrm{\Omega }_{14}=& T_{28}\hfill & \hfill \mathrm{\Omega }_{15}=& T_{29}\hfill & \hfill \mathrm{\Omega }_{16}=& T_{34}\hfill \\ \hfill \mathrm{\Omega }_{17}=& T_{35}\hfill & \hfill \mathrm{\Omega }_{18}=& T_{36}\hfill & \hfill \mathrm{\Omega }_{19}=& T_{37}\hfill & \hfill \mathrm{\Omega }_{20}=& T_{38}\hfill \\ \hfill \mathrm{\Omega }_{21}=& T_{39}\hfill & \hfill \mathrm{\Omega }_{22}=& T_{45}\hfill & \hfill \mathrm{\Omega }_{23}=& T_{46}\hfill & \hfill \mathrm{\Omega }_{24}=& T_{47}\hfill \\ \hfill \mathrm{\Omega }_{25}=& T_{48}\hfill & \hfill \mathrm{\Omega }_{26}=& T_{49}\hfill & \hfill \mathrm{\Omega }_{27}=& T_{56}\hfill & \hfill \mathrm{\Omega }_{28}=& T_{57}\hfill \\ \hfill \mathrm{\Omega }_{29}=& T_{58}\hfill & \hfill \mathrm{\Omega }_{30}=& T_{59}\hfill & \hfill \mathrm{\Omega }_{31}=& T_{67}\hfill & \hfill \mathrm{\Omega }_{32}=& T_{68}\hfill \\ \hfill \mathrm{\Omega }_{33}=& T_{69}\hfill & \hfill \mathrm{\Omega }_{34}=& T_{78}\hfill & \hfill \mathrm{\Omega }_{35}=& T_{79}\hfill & \hfill \mathrm{\Omega }_{36}=& T_{89}\hfill \\ \hfill \mathrm{\Omega }_{37}=& Q_1\hfill & \hfill \mathrm{\Omega }_{38}=& Q_2\hfill & \hfill \mathrm{\Omega }_{39}=& Q_3\hfill & \hfill \mathrm{\Omega }_{40}=& Q_4\hfill \\ \hfill \mathrm{\Omega }_{41}=& Q_5\hfill & \hfill \mathrm{\Omega }_{42}=& Q_6\hfill & \hfill \mathrm{\Omega }_{43}=& Q_7\hfill & \hfill \mathrm{\Omega }_{44}=& Q_8\hfill \\ \hfill \mathrm{\Omega }_{45}=& Q_9\hfill & \hfill \mathrm{\Omega }_{46}=& Q_{10}\hfill & \hfill \mathrm{\Omega }_{47}=& Q_{11}\hfill & \hfill \mathrm{\Omega }_{48}=& Q_{12}\hfill \\ \hfill \mathrm{\Omega }_{49}=& Q_{13}\hfill & \hfill \mathrm{\Omega }_{50}=& Q_{14}\hfill & \hfill \mathrm{\Omega }_{51}=& Q_{15}\hfill & \hfill \mathrm{\Omega }_{52}=& Q_{16}\hfill \end{array}$$
(B.17)
Then, as Cartan subalgebra we take the linear span of the following generators:
$$CSA\text{span}(\mathrm{\Omega }_5,\mathrm{\Omega }_{13},\mathrm{\Omega }_{20},\mathrm{\Omega }_{26})$$
(B.18)
and furthermore we specify the following basis:
$$\begin{array}{ccccccc}\hfill _1& =& \mathrm{\Omega }_5+\mathrm{\Omega }_{13}\hfill & ;& \hfill _2& =& \mathrm{\Omega }_5\mathrm{\Omega }_{13}\hfill \\ \hfill _3& =& \mathrm{\Omega }_{20}+\mathrm{\Omega }_{26}\hfill & ;& \hfill _4& =& \mathrm{\Omega }_{20}\mathrm{\Omega }_{26}\hfill \end{array}$$
(B.19)
With respect to this basis the step operators corresponding to the positive roots of $`\mathrm{F}_{4(4)}`$ as ordered and displayed in table 1 are those enumerated in table 9.
The steps operators corresponding to negative roots are obtained from those associate with positive ones via the following relation:
$$E^\varpi =𝒞E^\varpi 𝒞$$
(B.20)
where the $`26\times 26`$ symmetric matrix $`𝒞`$ is defined in the following way:
$$𝒞=\left(\begin{array}{ccc}\mathrm{𝟏}& 0& 0\\ & & \\ 0& \eta & 0\\ & & \\ 0& 0& C_+\end{array}\right)$$
(B.21)
A further comment is necessary about the normalizations of the step operators $`E^\varpi `$ which are displayed in table 9. They have been fixed with the following criterion. Once we have constructed the algebra, via the generators (B.14),(B.15), we have the Lie structure constants encoded in eq.(B.16) and hence we can diagonalize the adjoint action of the Cartan generators (B.19) finding which linear combinations of the remaining generators correspond to which root. Each root space is one-dimensional and therefore we are left with the task of choosing an absolute normalization for what we want to call the step operators:
$$E^\varpi =\lambda _\varpi \left(\text{linear combination of }\mathrm{\Omega }\text{.s}\right)$$
(B.22)
The values of $`\lambda _\varpi `$ are now determined by the following non trivial conditions:
1. The differences $`^i=\left(E^{\varpi _i}E^{\varpi _i}\right)`$ should close a subalgebra $`F_{4(4)}`$, the maximal compact subalgebra $`\mathrm{SU}(2)_\mathrm{R}\times \mathrm{Usp}(6)`$
2. The sums $`𝕂^i=\frac{1}{\sqrt{2}}\left(E^{\varpi _i}+E^{\varpi _i}\right)`$ should span a $`28`$-dimensional representation of $``$, namely the aforementioned $`(\mathrm{𝟐},\mathrm{𝟏𝟒})`$ of $`\mathrm{SU}(2)_\mathrm{R}\times \mathrm{Usp}(6)`$
We arbitrarily choose the first four $`\lambda _\varpi `$ associated with simple roots and then all the others are determined. The result is that displayed in table 9. Using the Cartan generators defined by eq.s (B.19) and the step operators enumerated in table 9 one can calculate the structure constants of $`\mathrm{F}_{4(4)}`$ in the Cartan-Weyl basis, namely:
$`[_i,_j]`$ $`=`$ $`0`$
$`[_i,E^\varpi ]`$ $`=`$ $`\varpi ^iE^\varpi `$
$`[E^\varpi ,E^\varpi ]`$ $`=`$ $`\varpi `$
$`[E^{\varpi _i},E^{\varpi _j}]`$ $`=`$ $`𝒩_{\varpi _i,\varpi _j}E^{\varpi _i+\varpi _j}`$ (B.23)
in particular one obtains the explicit numerical value of the coefficients $`𝒩_{\varpi _i,\varpi _j}`$, which, as it is well known, are the only ones not completely specified by the components of the root vectors in the root system. The result of this computation, following from eq.(B.16) is that encoded in eq.s (2.48, 2.49, 2.50) of the main text.
As a last point we can investigate the properties of the maximal compact subalgebra $`\mathrm{SU}(2)\times \mathrm{Usp}(6)\mathrm{F}_{4(4)}`$. As we know a basis of generators for this subalgebras is provided by:
$$H_i=(E^{\varpi _i}E^{\varpi _i});(i=1,\mathrm{},24)$$
(B.24)
but it is not a priori clear which are the generators of $`\mathrm{SU}(2)_\mathrm{R}`$ and which of $`\mathrm{Usp}(6)`$. By choosing a basis of Cartan generators of the compact algebra and diagonalizing their adjoint action this distinction can be established. The generators of $`\mathrm{SU}(2)_\mathrm{R}`$ are the following linear combinations:
$`J_X`$ $`=`$ $`{\displaystyle \frac{1}{4\sqrt{2}}}\left(H_1H_{14}+H_{20}H_{22}\right)`$
$`J_Y`$ $`=`$ $`{\displaystyle \frac{1}{4\sqrt{2}}}\left(H_5+H_{11}H_{18}+H_{23}\right)`$
$`J_Z`$ $`=`$ $`{\displaystyle \frac{1}{4\sqrt{2}}}\left(H_2+H_9H_{16}H_{24}\right)`$ (B.25)
close the standard commutation relations:
$$[J_i,J_j]=ϵ_{ijk}J_k$$
(B.26)
and commute with all the generators of $`\mathrm{Usp}(6)`$. These latter are displayed as follows.
$$\begin{array}{ccc}_1^{(Usp6)}& =& \frac{H_2}{2}\frac{H_9}{2}+\frac{H_{16}}{2}\frac{H_{24}}{2}\\ _2^{(Usp6)}& =& \frac{H_2}{2}+\frac{H_9}{2}+\frac{H_{16}}{2}+\frac{H_{24}}{2}\\ _3^{(Usp6)}& =& \frac{H_2}{2}+\frac{H_9}{2}+\frac{H_{16}}{2}\frac{H_{24}}{2}\end{array}$$
(B.27)
are the Cartan generators. On the other hand the nine pairs of generators which are rotated one into the other by the Cartans with eigenvalues equal to the roots of the compact algebra are the following ones
$$\begin{array}{cccccc}& & & & & \\ \hfill W_1& =& H_{10}\hfill & \hfill Z_1& =& H_7\hfill \\ \hfill W_2& =& H_4\hfill & \hfill Z_2& =& H_{13}\hfill \\ \hfill W_3& =& H_6\hfill & \hfill Z_3& =& H_3\hfill \\ \hfill W_4& =& H_1+H_{14}+H_{20}H_{22}\hfill & \hfill Z_4& =& H_5H_{11}H_{18}+H_{23}\hfill \\ \hfill W_5& =& H_{21}\hfill & \hfill Z_5& =& H_8\hfill \\ \hfill W_6& =& H_1+H_{14}+H_{20}+H_{22}\hfill & \hfill Z_6& =& H_5H_{11}H_{18}H_{23}\hfill \\ \hfill W_7& =& H_1H_{14}+H_{20}+H_{22}\hfill & \hfill Z_7& =& H_5H_{11}+H_{18}+H_{23}\hfill \\ \hfill W_8& =& H_{17}\hfill & \hfill Z_8& =& H_{15}\hfill \\ \hfill W_9& =& H_{12}\hfill & \hfill Z_9& =& H_{19}\hfill \end{array}$$
(B.28) |
warning/0507/nucl-th0507041.html | ar5iv | text | # Erratum: Polarized neutron 𝛽-decay: Proton asymmetry and recoil-order currents [Phys. Rev. C 72, 045501 (2005)]
Bookkeeping errors render the principal result of the paper incorrect. To be more precise, Equation (11) should be replaced by
$$d^2\mathrm{\Gamma }=\frac{2|G_F|^2|g_V|^2}{(2\pi )^3}(m_nR)^4\beta x^2(1x)^2\left(1+3\lambda ^2+C_0^{}(R,x)+C_p\mathrm{cos}\theta _p\right)dE_ed\mathrm{cos}\theta _p.$$
(1)
The $`C_0^{}`$ term consists of isotropic recoil-order terms in this observable and it is given by
$$\begin{array}{c}\hfill C_0^{}=2R[x\lambda (1x\beta ^2x)\lambda ^2(14x\beta ^2x)]4Rf_2\lambda (1x\beta ^2x)\\ \hfill 2Rg_2\lambda (2+x\beta ^2x)+2Rf_3x(1\beta ^2)\end{array}$$
(2)
The proton asymmetry can be found by writing
$$\begin{array}{c}\hfill d^2\mathrm{\Gamma }=f(x)(1+A_p\mathrm{cos}\theta _p)dE_ed\mathrm{cos}\theta _p,\\ \hfill A_p=\frac{C_p}{1+3\lambda ^2+C_0^{}}.\end{array}$$
(3)
Here are the correct expressions for Equations (12) and (13).
$$\begin{array}{c}\text{For }E_e<E_c:\hfill \\ A_p=\frac{2\lambda }{3(1x)^2(1+3\lambda ^2)}\left(3(1+\lambda )(1x)^2+\beta ^2x[23x\lambda (2x)]\right)\hfill \\ \\ +\frac{2R\lambda }{15(1x)^2(1+3\lambda ^2)^2}\left(\begin{array}{c}15(1x)^3(1+\lambda )(\lambda 1)^2\hfill \\ 5\beta ^2x[13x+4x^2+\lambda (35x+4x^2)\hfill \\ \lambda ^2(911x4x^2)+\lambda ^3(53x+4x^2)]\hfill \\ +\beta ^4x^2(\lambda +1)[3x+10\lambda (23x)\lambda ^2(2019x)]\hfill \end{array}\right)\hfill \\ \\ +\frac{4Rf_2\lambda }{15(1x)^2(1+3\lambda ^2)^2}\left(\begin{array}{c}15(1x)^3(\lambda 1)^2\hfill \\ +5\beta ^2x(1x)[4x3+2\lambda +\lambda ^2(1+4x)]\hfill \\ +\beta ^4x^2[3x+\lambda (2030x)\lambda ^2(2019x)]\hfill \end{array}\right)\hfill \\ \\ \frac{2Rg_2}{15(1x)^2(1+3\lambda ^2)^2}\left(\begin{array}{c}15(1x)^2(\lambda 1)[x+\lambda (2+x)2\lambda ^2(1x)]\hfill \\ 5\beta ^2x[12x2\lambda (2x)+\lambda ^2(110x+12x^2)\hfill \\ +2\lambda ^3(13x+2x^2)]\hfill \\ \beta ^4x^2[x+\lambda ^2(2027x)10\lambda ^3(2x)]\hfill \end{array}\right)\hfill \\ \\ \frac{2Rf_3\lambda (\lambda 1)x}{3(1x)^2(1+3\lambda ^2)^2}\left(\begin{array}{c}3(1x)^2(1+3\lambda )\beta ^2(310x+8x^2+3\lambda (36x+4x^2))\hfill \\ \beta ^4x(45x3\lambda x)\hfill \end{array}\right)\hfill \\ \\ +𝒪(R^2)\hfill \end{array}$$
(4)
$$\begin{array}{c}\text{For }E_e>E_c:\hfill \\ A_p=\frac{2\lambda }{3\beta x^2(1+3\lambda ^2)}\left((1x)(13x)\lambda (1x^2)+3\beta ^2x^2(\lambda 1)\right)\hfill \\ \\ +\frac{2R\lambda }{15\beta x^2(1+3\lambda ^2)^2}\left(\begin{array}{c}(1x)[1321x2x^2+\lambda (341x+28x^2)\hfill \\ +\lambda ^2(39103x+34x^2)\lambda ^3(31+3x4x^2)]\hfill \\ 5\beta ^2x(1+\lambda )[3x(12x)+2\lambda (1x)\hfill \\ \lambda ^2(23x+10x^2)]\hfill \\ +30\beta ^4x^3\lambda (1\lambda ^2)\hfill \end{array}\right)\hfill \\ \\ +\frac{4Rf_2\lambda }{15\beta x^2(1+3\lambda ^2)^2}\left(\begin{array}{c}(1x)^2[3+2x+10\lambda (13x)\lambda ^2(1+4x)]\hfill \\ 5\beta ^2x[3x(12x)+2\lambda (1x)\lambda ^2(23x+10x^2)]\hfill \\ 30\beta ^4x^3\lambda (\lambda 1)\hfill \end{array}\right)\hfill \\ \\ \frac{4Rg_2}{15\beta x^2(1+3\lambda ^2)^2}\left(\begin{array}{c}(1x)[24x3x^25\lambda (1+x)\hfill \\ \lambda ^2(413x6x^2)5\lambda ^3(1x^2)]\hfill \\ 5\beta ^2x[xx^23\lambda x\lambda ^2(1x+9x^2)+\lambda ^3(13x+2x^2)]\hfill \\ +15\beta ^4x^3\lambda ^2(\lambda 1)\hfill \end{array}\right)\hfill \\ \\ +\frac{4Rf_3\lambda (\lambda 1)}{3\beta x(1+3\lambda ^2)^2}\left((1\beta ^2)[(12x)(1x)3x\lambda (1x)3\beta ^2x^2]\right)\hfill \\ \\ +𝒪(R^2)\hfill \end{array}$$
(5)
I would like to thank D. Dubbers for bringing this problem to my attention. |
warning/0507/astro-ph0507323.html | ar5iv | text | # A search for changing-look AGN in the Grossan catalog
## 1 Introduction
Recently, a few Seyfert 2 galaxies were discovered to make transitions, on time scales of a few years, from a Compton–thin appearance (when the nuclear radiation is absorbed by material with a line–of–sight column density less than $`\sigma _T^1`$=1.5$`\times 10^{24}`$ cm<sup>-2</sup>) to a reflection–dominated spectrum, and/or viceversa (Guainazzi et al. 2002; Guainazzi 2002; Matt et al. 2003). A reflection–dominated spectrum is recognized by a hard continuum and a prominent (Equivalent Width, EW, $``$1 keV) iron line, and it is commonly assumed to be a signature of Compton–thick absorption (Matt 2000, 2002a). In this hypothesis, the reflection would be due to circumnuclear material, including the part of the absorbing matter which is visible to both the nucleus and the observer, such as the far side of the inner wall of the torus envisaged in Unification Models (Antonucci 1993).
Two possibilities exist for the observed transitions: either a change in the line-of-sight absorbing column, or a ‘switching-off’ of the nucleus, leaving reflection from distant matter as an echo of past activity. While the first hypothesis cannot in most cases be ruled out (and, indeed, seems to be the best explanation in the case of NGC 1365: Risaliti et al. 2005), the second appears more likely (see discussion in Matt et al. 2003) and in at least one case, namely NGC 2992 (Gilli et al. 2000), it is the only tenable. These sources would then be the Seyfert 2 analogs of the Narrow Line Seyfert 1 NGC 4051 (Guainazzi et al. 1998; Uttley et al. 1999). In the following, we will refer to all sources (whether Seyfert 1s or 2s) with the term ‘changing look’, if they temporarily appear to be reflection–dominated.
These objects have been serendipitously discovered; the very nature of this transition makes a systematic study of their properties difficult. In particular, how frequently sources undergo this sort of transition? Moreover, how many Compton–thick Seyfert 2 are really heavily absorbed, and not simply switched–off? It is quite difficult to estimate the fraction of these transitions, due to the lack of a complete and unbiased sample of homogeneously defined Seyfert galaxies with sufficient X-ray temporal and spectroscopic coverage. A first step in this direction was done by Guainazzi et al. (2005a) who reported a typical occurrence rate of a transition every 50 years, on the basis of a sample of 11 optically-selected Seyfert 2 galaxies, whose ASCA and/or BeppoSAX observations suggested Compton-thick obscuration. In this paper, we select a mini-sample of other four ‘changing-look’ candidates with long-term X-ray coverage.
## 2 Observations
### 2.1 The sample
Our sample is extracted from the Grossan (1992) catalog, based on the catalog of X-ray sources from the LASS (Large Area Sky Survey, or HEAO A-1) instrument aboard the HEAO-1 satellite, which included the first scan of the full sky from August 1977 to February 1978 (Wood et al. 1984). The catalog took advantage of the overlapping diamond-shaped error regions of the MC instrument (Modulation Collimator, or A-3 experiment: Gursky et al. 1978), which also flew aboard the HEAO-1 satellite, to identify the 96 AGN that make up the so-called LMA (LASS/MC identified AGN) sample, down to a limiting 2-10 keV flux of about 1.8$`\times 10^{11}`$ erg cm<sup>-2</sup> s<sup>-1</sup>.
Our sample selection criteria aimed at finding good candidates for ‘changing–look’ sources, to be observed with XMM-Newton. The selection was made by taking all the sources in the Grossan catalog which have subsequently been observed in X–rays to be reflection–dominated, or too faint to allow for a detailed spectral analysis, with a flux at least a factor 50 fainter than observed by HEAO A-1. If the optical identifications are correct, these sources could therefore be ‘changing-look’ sources.
These selection criteria resulted in a mini-sample of four targets: NGC 7674, NGC 4968, IRAS 13218+0552 and NGC 1667. We note here that none of these sources belongs to the Piccinotti et al. (1982) HEAO A-2 sample, because they have all 2-10 keV fluxes below its limit.
### 2.2 Data reduction
Table 1 shows the log of all the XMM-Newton observations analysed in this paper. All observations were performed with the EPIC CCD cameras, the pn (Strüder et al. 2001) and the two MOS (Turner et al. 2001), but with different combinations of subframes and filters. Data were reduced with SAS 6.1.0 (Gabriel et al. 2004) and screening for intervals of flaring particle background was done consistently with the choice of extraction radii, in an iterative process based on the procedure to maximize the signal-to-noise ratio described by Piconcelli et al. (2004). No source suffers from pileup problems, so pn spectra were extracted with pattern 0 to 4 and MOS spectra with patterns 0 to 12. When the two MOS observations were performed with the same subframe, their spectra were summed. All spectra were binned in order to oversample the instrumental resolution by at least a factor of 3 and to have no less than 25 counts in each background-subtracted spectral channel. The latter requirement allows us to use the $`\chi ^2`$ statistics. On the other hand, ‘local fits’ were also performed in the 5.25-7.25 keV energy range with the unbinned spectra, using the Cash (1976) statistics, in order to better assess the nature or the presence of the iron lines, like in the cases of NGC 1667 and NGC 7674. We refer the reader to Guainazzi et al. (2005b) for details on this kind of analysis. However, all the EWs reported in the paper refer to the global fits on the binned spectra and with respect to the best fit model.
For two sources, namely NGC 7674 and IRAS 13218+0552, we also analysed previous BeppoSAX observations, in the latter case for the first time, to better compare them to the XMM-Newton results. Event files and spectra were retrieved from the ASDC Multi-Mission Interactive Archive<sup>1</sup><sup>1</sup>1http://www.asdc.asi.it/. MECS spectra for NGC 7674 were also re-extracted from circular regions with different radii, using Xselect. As for the ASCA observations of NGC 4968 and NGC 1667, since they are characterised by detections at the limit of the instrument and different fluxes are reported in literature, we re-analyzed archival data, starting from linearized event lists extracted from the HEASARC archive<sup>2</sup><sup>2</sup>2http://heasarc.gsfc.nasa.gov/db-perl/W3Browse/w3browse.pl.
#### 2.2.1 A note on GINGA observations
Since the GINGA observation of NGC 7674 will be of fundamental importance to understand the nature of this source, we need to consider with some detail the reliability of GINGA fluxes of faint objects. The brightness of the X-ray background (XRB) is known to fluctuate on the sky. With the beam of the GINGA LAC ($`FWHM=1^{}\times 2^{}`$), $`3\sigma `$ fluctuation of the XRB was estimated to be $`2`$ ct s<sup>-1</sup> in the 2-10 keV band (Hayashida et al. 1989; Butcher et al. 1997). Therefore, in the absence of a reliable estimate of the local XRB brightness level, the cosmic variance of the XRB limits the source detection with a pointed GINGA observation.
For some sources, scanning observations were made just before or after pointing at the target object. This will secure the source identification at least in the direction of the scan path within 0.2 degree or less, and reduce the uncertainty due to the cosmic variance of the XRB brightness.
There is a scanning observation for NGC 7674, in which a significant excess was detected at the position of the galaxy, with a count rate of 3.3 ct s<sup>-1</sup> (Awaki 1991). The XRB level was estimated based on the scanning data in Awaki (1991). On the other hand, Smith & Done (1996) did not consider NGC 7674 to be detected because the source count rate they estimated was smaller than the cosmic variance (1.8 ct s<sup>-1</sup>). Awaki (1991) and Smith & Done (1996) used different methods in estimating the detector background. This might introduce some difference in their measure of the source flux. However, we consider the source detection reported in Awaki (1991) to be reliable, as it was based on the scanning observation, which is free from the uncertainty due to cosmic variance in the XRB.
On the other hand, no scanning observation was carried out for NGC 1667 while the local XRB was estimated only based on the data taken at the nearby off-source sky. The source count rate for NGC 1667 reported in Awaki & Koyama (1993) is 0.5 ct s<sup>-1</sup>, which is smaller than the cosmic variance of the XRB. Because of the lack of a scanning observation, this small excess recorded for NGC 1667 cannot be considered as a reliable detection and we take this value as an upper limit. An earlier detection reported in Polletta et al. (1996) came from a detection of a source during a slewing operation of the satellite. The detection is only in the soft band (2–4 keV). An inspection of the ROSAT All-Sky Survey (RASS) image finds brighter soft X-ray sources: one at 15 arcmin to the E and two in 75 arcmin to the N and SE, and they are likely to be confused with NGC 1667.
All spectra were analyzed with Xspec 11.3.1. In the following, errors correspond to the 90% confidence level for one interesting parameter ($`\mathrm{\Delta }\chi ^2=2.71`$), where not otherwise stated. The cosmological parameters used throughout this paper are $`H_0=70`$ km s<sup>-1</sup> Mpc<sup>-1</sup>, $`\mathrm{\Lambda }_0=0.73`$ and $`q_0=0`$.
## 3 Analysis
### 3.1 NGC 7674
NGC 7674 is a Seyfert 2 galaxy with broad H$`\alpha `$ and H$`\beta `$ components in polarized light (Miller & Goodrich 1990; Young et al. 1996). After its LASS 2-10 keV flux of $`2.4\times 10^{11}`$ erg cm<sup>-2</sup> s<sup>-1</sup>, Awaki et al. (1991) reported a flux of $`8\times 10^{12}`$ erg cm<sup>-2</sup> s<sup>-1</sup> with GINGA, with a spectral shape characterized by an absorbed powerlaw (but the derived column density was unconstrained) and an upper limit of 80 eV to the iron line EW. Subsequently, a ROSAT PSPC spectrum derived a 0.5-2 keV flux of $`2\times 10^{13}`$ erg cm<sup>-2</sup> s<sup>-1</sup> (Levenson et al. 2001). The source was then observed by BeppoSAX on November 1996, and found clearly reflection–dominated, with a flux of $`5\times 10^{13}`$ erg cm<sup>-2</sup> s<sup>-1</sup> (Malaguti et al. 1998). If the source was absorbed, instead of switched–off as suggested by the HEAO A-1 and GINGA measurements, the lack of strong excess emission in the PDS instrument permits to put a lower limit to the column density of the absorber of several times 10<sup>24</sup> cm<sup>-2</sup>.
XMM-Newton confirmed the spectral shape observed by BeppoSAX, which we modeled with a bare Compton reflection component (model pexrav in Xspec: Magdziarz & Zdziarski 1995), a neutral iron K$`\alpha `$ line and a steep power law for the soft excess, with photon index $`\mathrm{\Gamma }_s`$. Two further emission lines are required by the data, at $`0.91_{0.03}^{+0.02}`$ keV (Ne ix K$`\alpha `$) and at $`6.97_{0.05}^{+0.26}`$ keV, at the 99.9% and 98% confidence level, respectively, according to F-test. The latter confirms the complex iron line profile observed with BeppoSAX, but is not unambiguously present in the ‘local fit’ (see inset in Fig. 1): a likely explanation is that the feature is not dominated by a single emission line, but is instead a blend of lines, such as the Fe xxvi K$`\alpha `$ and the neutral Fe K$`\beta `$.
The only problem with a Compton-reflection dominated scenario is the EW of the neutral iron line ($`400`$ eV with respect to the reflection component only), which is much lower than the expected one, i.e $`>1`$ keV (see e.g. Matt et al. 1996). It is likely that the measure of the flux of the iron line in NGC 7674 is affected by the above-mentioned prominent feature at higher energies, considering that the EPIC spectra do not have high statistics. In any case, EW as low as $`600`$ eV have been measured in Compton-thick sources and may be due to iron underabundance and/or a small inclination angle of the torus (see e.g. Mrk 3: Bianchi et al. 2005).
The XMM-Newton spectrum is plotted in Fig. 1, while Table 2 summarizes the best fit parameters compared to our re-analysis of the BeppoSAX data. No significant variability is found in any of the spectral parameters. Moreover, we find a somewhat larger BeppoSAX 2-10 keV flux with respect to the one reported by Malaguti et al. (1998) (but they do not quote any error), in better agreement with the $`7\times 10^{13}`$ erg cm<sup>-2</sup> s<sup>-1</sup> measured with XMM-Newton. We note that the BeppoSAX flux should not be contaminated by nearby sources, as the selected extraction radius of 2 arcsec is free of other bright sources (see Table 6). No short-term variability is found in the XMM-Newton lightcurves.
As a final comment, we would like to point out that it is quite difficult to find a physical origin for the very steep powerlaw needed to fit the soft X-ray spectrum of NGC 7674. This is a common problem in the analysis of Compton-thick Seyfert galaxies but, as already suggested by Iwasawa et al. (2002) and Guainazzi et al. (2004), it may be the result of a blending of strong emission lines which mimic a continuum component in low resolution spectra. Indeed, this interpretation turns out to be true in the few objects were high resolution spectra are available, as in NGC 1068 (Kinkhabwala et al. 2002; Brinkman et al. 2002), Circinus (Sambruna et al. 2001) and Mrk 3 (Bianchi et al. 2005; Sako et al. 2000). This seems also to be the case for NGC 7674. A fit of equivalent statistical quality ($`\chi ^2=39/37`$ d.o.f.) is achieved by modelling the soft spectrum with a powerlaw with the same photon index of the primary continuum (now $`\mathrm{\Gamma }1.85`$), plus a number of emission lines, mostly from H- and H-like O, Ne, Mg and Si. Therefore, an interpretation of the soft X-ray spectrum in terms of emission from a gas photoionized by the nuclear continuum is the most likely one also in this object, even if high resolution spectra would be required to finally clarify this issue.
### 3.2 NGC 4968
NGC 4968 is a Seyfert 2 galaxy which was not detected by GINGA (Awaki 1991) or in the RASS, with an upper limit on the PSPC count rate of 0.03, corresponding roughly to a 0.5-2 keV flux of $`10^{13}`$ erg cm<sup>-2</sup> s<sup>-1</sup>, assuming a $`\mathrm{\Gamma }=1.7`$ powerlaw spectrum (Rush et al. 1996). A 2-10 keV flux of a few $`10^{13}`$ erg cm<sup>-2</sup> s<sup>-1</sup> and a poor spectrum consistent with various models characterized the ASCA observation (Turner et al. 1997). Preliminary results on a 2001 XMM-Newton observation were presented by Matt (2002b), clearly showing that the source was reflection-dominated. We present here full results on this XMM-Newton observation, together with a later one. The two spectra are fully compatible each other and are well fitted by the same model adopted for NGC 7674. The only emission line apparent in the data is the neutral Fe K$`\alpha `$ line: Table 3 and Fig. 2 show the best fit parameters and the spectra. No significant variability is found in fluxes or spectral shape between the two observations. Moreover, the XMM-Newton lightcurves of each observation are consistent with a constant source.
### 3.3 IRAS 13218+0552
The source is the only object in our sample with a moderately large redshift and classified as a Type 1. After the HEAO A-1 detection, the first X-ray observation of IRAS 13218+0552 was performed on July 2000 by BeppoSAX, which detected a source with a flux more than 60 times fainter and a low quality spectrum: N<sub>H</sub> and $`\mathrm{\Gamma }`$ are poorly constrained, but the spectrum seems to be steep ($`\mathrm{\Gamma }`$2). Interestingly enough, there seems to be an excess flux in the PDS, suggesting moderately Compton–thick absorption; however, at these count-rate levels ($`0.08\pm 0.05`$ counts s<sup>-1</sup>) confusion is a serious issue and we cannot draw any definite conclusion.
However, the XMM-Newton observation failed to detect IRAS 13218+0552, with an upper limit of $`3.9\times 10^{14}`$ erg cm<sup>-2</sup> s<sup>-1</sup> in the EPIC pn. The closest bright source, located at $`5`$ arcmin from the target, corresponds to the radio source NVSS J132431+053254 (see Table 6). The X-ray spectrum of this source is fully consistent with that of the source observed by BeppoSAX (an unabsorbed powerlaw with $`\mathrm{\Gamma }2`$), even if its flux is significantly lower, being $`5.3\times 10^{14}`$ erg cm<sup>-2</sup> s<sup>-1</sup>. We show in Fig. 3 the contour plot of the MECS observation superimposed on the EPIC pn image, marking the positions of IRAS 13218+0552 and NVSS J132431+053254: it seems likely that the source observed by BeppoSAX is the latter, so that the target object was undetected also in this observation. Moreover, in a ROSAT observation of the same field, no source is apparent at the coordinates of IRAS 13218+0552.
### 3.4 NGC 1667
After the X-ray detection by HEAO A-1 at a flux of $`1.9\times 10^{11}`$ erg cm<sup>-2</sup> s<sup>-1</sup>, the source was not detected by GINGA and presented a very low flux in the ASCA observation (Turner et al. 1997; Pappa et al. 2001). Indeed, our re-analysis of the ASCA data shows that the source is practically undetected above 2 keV. Moreover, only an upper limit was found by the PSPC within the RASS (Rush et al. 1996).
A low flux level similar to the one found by ASCA characterizes the XMM-Newton observation. A fit with a simple powerlaw leads to a very steep photon index ($`\mathrm{\Gamma }3`$), but with large residuals on the softer and the harder parts of the spectrum, resulting in an unacceptable reduced $`\chi ^2`$, greater than 3. On the other hand, the large ratio between the \[OIII\] and the far infrared (IR) fluxes, together with a relatively low X-ray flux, hint to a Compton-thick source (see Tables 1, 5 and Sect. 4.2 for details). Therefore, we tried a model with a pure reflection component and a soft excess: now the fit is perfectly acceptable ($`\chi ^2=17/14`$ d.o.f: see Table 4 and Fig. 4). The model includes an emission line at $`0.87_{0.02}^{+0.03}`$ keV (likely Ne ix K$`\alpha `$), required at the 97% confidence level according to F-test. The presence of a strong iron line is suggested by the ‘local fit’ (see inset in Fig. 4), with a flux of $`1.0_{0.6}^{+1.4}\times 10^6`$ ph s<sup>-1</sup>, consistent with the upper limit found in the global fit. This would correspond to an EW of $`600`$ eV. The observed 2-10 keV flux is $`1\times 10^{13}`$ erg cm<sup>-2</sup> s<sup>-1</sup> and no significant short-term variability trends are present in the lightcurve.
## 4 Discussion
### 4.1 Classification of the objects in the sample
All the sources included in our sample turn out to be likely Compton-thick. While the spectral analysis presented in this paper strongly favour this scenario at least for NGC 7674 and NGC 4968 (see Sect. 3), another test can be done on the basis of their \[OIII\], infrared (IR) and X-ray fluxes, as explained, for example, by Panessa & Bassani (2002). Looking at the diagrams shown in Fig. 5, it is clear that all the four sources populate or are very close to the regions of the Compton-thick Seyfert galaxies. However, we note here that it should be better to define them reflection-dominated objects, since these diagrams cannot distinguish between highly obscured and ‘switched-off’ sources. Indeed, in at least one case, NGC 7674, the diagrams in Fig. 5 show that the source moved from the Compton-thin to the Compton-thick region between two X-ray observations, thus being a good ‘switched off’ candidate. We will discuss this object further in Sect. 4.3.
### 4.2 A re-assessment of LMA identifications
None of the sources in our sample was caught in a high state, comparable to the one measured by HEAO A-1. Therefore, before speculating on large flux variations, we should first re-assess the reliability of the LMA fluxes.
#### 4.2.1 Wrong identifications
A first test is to check the optical identifications given by Grossan (1992) for the four sources in our sample. We reconstructed the rectangular HEAO A-1 error boxes using the coordinates reported by Wood et al. (1984) and overplotted the EPIC pn fields together with the position of the AGN which constitute our sample. The results are shown in Fig. 6. With the only exception of NGC 4968, all the sources fall largely outside of the 95% LASS error box. Note that the actual error region is much smaller than the latter, because the LMA identifications also take into consideration the MC pointings. One of the sources, IRAS 13218+0552, was part of a 22-objects subsample of the LMA analyzed in detail by Remillard et al. (1993), who included a figure very similar to the one presented here. Nonetheless, he suggested that in this case (and in some others) the LASS error boxes should be considered as $`50\%`$ contours. However, it is clear that these identifications are much less robust than for other LMA sources.
If this is the case, we should also look for contamination by nearby sources, which could contribute to the total flux measured by HEAO A-1. The LASS error boxes shown in Fig. 6 are overplotted on the RASS images of the same fields. No bright source lies inside any of the HEAO A-1 boxes. Moreover, we measured fluxes of the brightest objects present in the EPIC fields of each target source. The results are shown in Table 6: none of these sources have X-ray fluxes comparable to the one measured by the LMA and are all (with the exception of NVSS J132431+053254 and SDSS J132442.44+052438.9 in the otherwise empty field of IRAS 13218+0552), much dimmer than the AGN. Even if some of them can be highly variable sources, we have no evidence that their fluxes could have significantly contaminated the LASS count rates.
#### 4.2.2 Wrong fluxes
Even if the optical identification given by Grossan (1992) are very uncertain, there is no clear evidence of contamination from other sources. Therefore, it is possible that the LMA fluxes are wrong. Indeed, there are two reasons to suspect that. The first one is that the fluxes of these sources are among the lowest in the sample and, in the cases of IRAS 13218+0552 and NGC 1667, they are just above the flux limit of the sample itself, with errors around 20$`\%`$. A second issue refers to the fact that HEAO A-1, due to an hardware failure, lost all its spectral resolution, so that it should be considered as a ‘large X-ray photometer over a range of 1-20 keV’, to use the words of Grossan (1992). The measured count rates were then transformed to a flux density at 5 keV adopting an empirical flux conversion derived from a comparison with the HEAO A-2 instrument and assuming a spectral shape of a powerlaw with a photon index of 1.7. The last assumption gives reasonable results for unabsorbed or even moderately absorbed AGN, but can lead to completely wrong fluxes in Compton-thick sources. If the primary continuum is blocked up to 10 keV or higher energies, the measured LASS count rates are totally unrelated to the 2-10 keV flux of the source, because they refer, in an unknown way, to the total band of the instrument, up to 20 keV.
We performed a rough test with Xspec to see if this scenario is consistent with the measured PDS fluxes in the two sources observed by BeppoSAX. We first calculated a 1-20 keV count-rate (arbitrarily normalized) which reproduced the 2-10 keV HEAO flux with a model consisting of an unabsorbed powerlaw with $`\mathrm{\Gamma }=1.7`$. Then, we produced a model consisting of a bare Compton-reflection component (with a 2-10 keV flux like the one observed by XMM-Newton) and a highly obscured (N<sub>H</sub> of several $`10^{24}`$ cm<sup>-2</sup>) powerlaw, yielding the same count-rate. The resulting 10-20 keV flux is, for both NGC 7674 and IRAS 13218+0552, a factor $`100`$ larger than the one observed by BeppoSAX, showing that, at least in these objects, it is unlikely that the count-rate observed by HEAO A-1 can be interpreted in terms of high energy emission of an otherwise obscured object. We note that our test assumes a flat response of the detector in the whole band 1-20 keV. In a more realistic case, where the instrumental effective area decreases at high energies, the difference between the expected flux and the one observed by the PDS would be larger.
In conclusion, even if we do not find any conclusive evidence against the identification or the flux of the sources in our sample, the above-mentioned issues may affect, in principle, all the low-flux objects in the Grossan (1992) catalog, which should be treated with due caution. However, it is worthwhile stressing that the criterion adopted in our sample clearly puts a strong bias towards potentially fake HEAO A-1 identifications or detections, since all the targets were not confirmed at the same flux level by other observations. For our purposes, all the preceding discussion does not allow us to speculate on large X-ray variations of the sources in our sample on the basis of their HEAO A-1 fluxes only. In particular, NGC 4968 is the only object which lies reasonably close to the LMA error box, but, apart from its LMA flux, it does not present any significant variation for several years, similarly to NGC 1667 (see Table 5). On the other hand, if we relax the constraints given by the HEAO error boxes, as suggested by Remillard et al. (1993), in one case, NGC 7674, the LMA flux could be in principle suggestive of a past high flux state of the source, considering the following GINGA flux, which is significantly larger than the BeppoSAX and the XMM-Newton ones. Finally, another object, IRAS 13218+0552, is not detected at all by XMM-Newton. These two sources deserve some further comments.
### 4.3 NGC 7674: a switched-off source?
Even not taking into account the HEAO A-1 observation, the source underwent a flux loss by a factor $`>10`$ in the 7 years separating the GINGA and BeppoSAX observation, then remained fairly stable after other 8 years, when the XMM-Newton observation was performed (see Table 5). Therefore, NGC 7674 may well represent a good candidate for being a switched-off object.
This is also supported by a spectral transition between the two flux states. Indeed, the GINGA spectrum was a featureless powerlaw, with only an upper limit to the EW of the iron line (80 eV) and an unconstrained value for the neutral absorbing column density. On the other hand, the BeppoSAX and XMM-Newton spectral analysis clearly show that the source is actually reprocessing-dominated, with a bare Compton reflection continuum at high energies, an iron line with a EW$`400`$ eV and a soft excess at lower energies (see Fig. 7). Moreover, the source lies in the Compton-thick region in the diagnostic diagrams based on the X-ray, IR and \[OIII\] fluxes (see Fig. 5).
Such a transition between a transmission- to a reprocessing-dominated spectral state would make NGC 7674 the sixth Seyfert 2 galaxy to show this behaviour (see Guainazzi et al. 2005a, and references therein). The time elapsed between the GINGA, BeppoSAX observation and XMM-Newton observations allows us to make an estimate of the distance of the reflector from the Black Hole (BH). If the primary source switched off shortly after the GINGA observation, the detection of a reflection echo 15 years after implies a distance of around 5 pc of the material responsible for it. On the other hand, if, instead, it switched off a little before the BeppoSAX observation, the same flux level measured 8 years later puts a lower limit of almost 3 pc. Since it is impossible to know if in these time intervals the source actually switched on again, care should be taken when considering these lower limits.
Another interesting piece of information comes from the ratio between the normalizations of the primary powerlaw in the Compton-thin state and the Compton reflection component in the reprocessing-dominated state, which leads to an estimate of the covering factor of the reflecting matter. In the case of NGC 7674, this value is approximately $`R=1.5`$, i.e. the subtended solid angle is around $`3\pi `$. We have calculated the same ratio for all the ‘changing-look’ Seyfert 2 sources in the literature and found values ranging from 0.8 to 1.8 (except for the extreme case of NGC 6300, but in this case the primary continuum is very variable and it is difficult to calculate R, Guainazzi 2002). It should be noted that these values must be considered as estimates of the true covering factor, because they are calculated assuming transmission and reflection component fluxes which were not measured simultaneously, thus not necessarily directly linked one to the other. Since in ‘changing-look’ sources the Compton-thick material clearly does not intercept the line of sight, these values of R should be compared to the ones found in Seyfert 1s, which are in the range $`0.51`$ (see e.g. Perola et al. 2002; Bianchi et al. 2004): this is suggestive that the same material (likely the torus envisaged in Unification Models, Antonucci 1993) is responsible for the reflection components observed in the two classes of objects.
### 4.4 A misclassified Compton-thick object: IRAS 13218+0552
While the HEAO A-1 flux is likely the result of a misidentification and the BeppoSAX flux also probably belongs to a nearby source (see previous section), the non-detection of IRAS 13218+0552 by XMM-Newton still makes this source peculiar. Indeed, the upper limit derived for its X-ray flux, together with the IR and \[OIII\] fluxes, clearly puts this object among the Compton-thick sources (see Fig. 5). This is at odds with its current optical classification. Indeed, the presence of emission lines with components broader than $`\mathrm{5\hspace{0.17em}000}`$ km s<sup>-1</sup> and the stellar-like appearance induced Low et al. (1988) to classify the source as a QSO. However, it was soon clear that IRAS 13218+0552 was peculiar and its extremely red IR continuum won it the name of ‘reddest known quasar’ (Low et al. 1989). Moreover, Remillard et al. (1993) reported anomalously broad \[OIII\] emission lines (FWHM $`\mathrm{3\hspace{0.17em}500}`$ km s<sup>-1</sup>), likely the result of some substructure in their profile. An asymmetric \[OIII\] profile, due to a blueshifted component, was also reported by Zheng et al. (2002).
The HST image in the R band is strongly suggestive of a merger between two galaxies in its latest stage, according to Boyce et al. (1996). They also noted that, given its low nuclear luminosity, it should be strictly classified as a Seyfert 1, instead of a QSO, but its colors clearly points toward a very large obscuration, so that it is very likely that it hosts a buried QSO. An extreme velocity outflow (EVOF) component was finally unambiguously found in the H$`\beta `$ and \[OIII\] line profiles, with velocities of the order of $`\mathrm{2\hspace{0.17em}000}`$ km s<sup>-1</sup> (Lípari et al. 2003). This detection, together with the previous results, led these authors to conclude that IRAS 13218+0552 is likely the result of a recent merging process and the nuclear energy is due to the composite activity of a hidden QSO and a starburst, the latter implied also by the tentative detection of Wolf-Rayet features. Indeed, the starburst component could even be the dominant one, as suggested by the optical emission line ratios reported by Kim et al. (1998)<sup>3</sup><sup>3</sup>3The previous classification as a Type 1 object prevented these authors to use, in the case of IRAS 13218+0552, the diagnostic diagrams plotted in Fig. 2 of their paper. However, once the presence of broad lines is excluded, we are allowed to use these line ratios to discriminate between a starburst- or AGN-dominated object..
In conclusion, the original classification as a QSO/Seyfert 1 was probably the result of a misinterpretation of the EVOF component as emission from the Broad Line Region. Therefore, the source should be re-classified as a Type 2 object, Compton-thick in the X-rays and likely surrounded by a massive starburst. Within this scenario, it is possible that the PDS excess detected in the BeppoSAX observation (whose MECS spectrum is instead likely dominated by a nearby source: see Sect. 3.3) refers to the nuclear emission of the buried AGN, which pierces through a very large gas column density. However, if an AGN is actively heating the circumnuclear dust, an infrared compact source should be seen with the high spatial resolution imaging provided by HST at 1.6-2.2 $`\mu `$m, as seen in similar Compton-thick nearby Seyfert galaxies such as NGC 1068 (Thompson et al. 2001).
## 5 Conclusions
We have selected a sample of 4 AGN, included in the Grossan (1992) catalog, which showed in subsequent observations a flux much lower than the one measured with HEAO A-1, thus being good candidates for being ‘changing-look’ sources. None of the sources was caught in a high flux state during the XMM-Newton observations. We have shown that, for all the sources, potential problems with the HEAO A-1 source identification and flux measurement prevent us to be certain that the HEAO A-1 data represent a putative ‘high’ state for these objects.
However, based on the high flux state of its GINGA observation, a factor of ten higher than in the BeppoSAX and XMM-Newton observation, NGC 7674 represents probably the sixth known case of a ‘changing-look’ Seyfert 2 galaxy. This is also supported by a spectral transition between a transmission- to reprocessing-dominated state between the two observations. From the X-ray variability pattern, we can estimate a lower limit of a few parsec to the distance of the reflecting material.
Finally, one of the sources, IRAS 13218+0552, was not detected by XMM-Newton, despite being currently classified as a Seyfert 1 with a large \[OIII\] flux. However, the original classification was likely to be affected by an outflow component in the emission lines. The object likely harbors an highly obscured AGN and should be re-classified as a Type 2 source.
###### Acknowledgements.
We would like to thank the referee, S. Lumsden, for his valuable suggestions. This paper is based on observations obtained with XMM-Newton, an ESA science mission with instruments and contributions directly funded by ESA Member States and the USA (NASA). |
warning/0507/hep-ex0507005.html | ar5iv | text | # 1 Introduction
## 1 Introduction
The decays $`BK\mathrm{}^+\mathrm{}^{}`$ and $`BK^{}\mathrm{}^+\mathrm{}^{}`$, where $`\mathrm{}^+\mathrm{}^{}`$ are the charged lepton pairs $`e^+e^{}`$ or $`\mu ^+\mu ^{}`$ and $`K^{}`$ is the $`K^{}(892)`$ meson, result from $`bs`$ flavor-changing neutral currents (FCNC). In the Standard Model (SM) of electroweak interactions, such $`bs`$ processes are forbidden in tree-level Feynman diagrams; they are allowed at lowest order through one-loop diagrams involving the emission and re-absorption of $`W`$ bosons. Because the lowest order SM diagrams are loops of weakly interacting particles with virtual energies comparable to the electroweak scale, new flavor-changing interactions at the electroweak scale can introduce loop diagrams with comparable amplitudes. The SM predictions of the rates and kinematic distributions of FCNC decays can be significantly modified by a broad class of new physics models, such as a charged Higgs boson , topcolor , weak-scale supersymmetry , fourth-generation fermions , or leptoquarks .
In the SM, three amplitudes contribute at lowest order to the $`bs\mathrm{}^+\mathrm{}^{}`$ process: a photon penguin, a $`Z`$ penguin, and a $`W^+W^{}`$ box diagram (Figure 1). The magnitude of the photon penguin amplitude is well known experimentally from measurements of the rate of the FCNC decay $`bs\gamma `$ and agrees well with SM predictions . The latter two amplitudes are not well known and thus studies of $`bs\mathrm{}^+\mathrm{}^{}`$ provide new information on FCNC processes. The SM decay rate of $`bs\mathrm{}^+\mathrm{}^{}`$ is suppressed relative to other $`b`$ decays, resulting in a predicted total branching fraction of $`(4.2\pm 0.7)\times 10^6`$ , in agreement with experiment .
The most abundant exclusive decays associated with the $`bs\mathrm{}^+\mathrm{}^{}`$ transition, $`BK\mathrm{}^+\mathrm{}^{}`$ and $`BK^{}\mathrm{}^+\mathrm{}^{}`$, are predicted to have branching fractions of $`0.4\times 10^6`$ for $`BK\mathrm{}^+\mathrm{}^{}`$ and about three times that for $`BK^{}\mathrm{}^+\mathrm{}^{}`$ , with a theoretical uncertainty of 30%. The theoretical uncertainty is predominantly due to the uncertainty in the prediction of semileptonic form factors, which model the rate that a $`bs`$ FCNC in a $`B`$ decay results in a single $`K^{()}`$ meson. The partial widths of $`BKe^+e^{}`$ and $`BK\mu ^+\mu ^{}`$ are expected to be identical, because of identical electroweak couplings of electrons and muons. The branching fractions of both $`BK^{}e^+e^{}`$ and $`BK^{}\mu ^+\mu ^{}`$, however, receive a contribution from a pole in the photon penguin amplitude at $`q^2=m_\mathrm{}^+\mathrm{}^{}^20,`$ and the enhancement in the electron mode is significantly larger due to its lower $`q^2`$ threshold. This phase space difference in the pole contribution is expected to reduce the ratio $`\mathrm{\Gamma }(BK^{}\mu ^+\mu ^{})/\mathrm{\Gamma }(BK^{}e^+e^{})`$ from unity to 0.752 . Previous measurements of the exclusive decays are consistent with predictions . In the absence of new physics contributions, improved precision in the exclusive branching fractions will improve experimental constraints of $`BK^{()}`$ form factors.
More precise SM tests can be obtained from rate asymmetries and kinematic distributions of the exclusive decay products. The direct $`CP`$ asymmetries
$$A_{CP}=\frac{\mathrm{\Gamma }(\overline{B}K^{()}\mathrm{}^+\mathrm{}^{})\mathrm{\Gamma }(BK^{()}\mathrm{}^+\mathrm{}^{})}{\mathrm{\Gamma }(\overline{B}K^{()}\mathrm{}^+\mathrm{}^{})+\mathrm{\Gamma }(BK^{()}\mathrm{}^+\mathrm{}^{})}$$
for these decays are expected to be very small in the SM, much less than 1% , whereas new physics at the electroweak scale could enhance $`A_{CP}`$ to values of order one . If one neglects the pole region ($`q^2<0.1\mathrm{GeV}^2/c^4`$) of $`BK^{}e^+e^{}`$, in the SM the ratios
$$R_K=(BK\mu ^+\mu ^{})/(BKe^+e^{}),$$
$$R_K^{}=(BK^{}\mu ^+\mu ^{})/(BK^{}e^+e^{})$$
are expected to be unity with high precision. However, this ratio could be enhanced by corrections of order 10% due to the presence of a supersymmetric neutral Higgs boson with large $`\mathrm{tan}\beta `$ (ratio of vacuum expectation values of the two Higgs doublets) . The Feynman diagram of this process is shown in Figure 2. A large $`\mathrm{tan}\beta `$ would enhance the Higgs/squark coupling, and the Higgs decays to muons will be enhanced relative to decays to electrons because of the large ratio of Yukawa couplings $`m_\mu ^2/m_e^2`$. A measurement of the relative abundance of electrons and muons in exclusive decays is therefore a probe of scalar penguin processes and complements the limits obtained from searches for the rare decay $`B_s\mu ^+\mu ^{}`$ . With sufficiently large samples of $`BK^{}\mathrm{}^+\mathrm{}^{}`$ events, angular asymmetries in the four-particle final state can also accurately gauge the relative phase and magnitude of the three contributing FCNC amplitudes .
## 2 Detector and Datasets
We analyze data collected with the BABAR detector at the PEP-II storage ring at the Stanford Linear Accelerator Center. The data sample comprises 208.0 $`\text{ fb}^1`$ recorded on the $`\mathrm{{\rm Y}}(4S)`$ resonance, yielding $`(229.0\pm 2.5)\times 10^6`$ $`B\overline{B}`$ decays, and an off-resonance sample of 22.1 $`\text{ fb}^1`$ used to study continuum background.
The BABAR detector is described in detail elsewhere . The most important capabilities of the detector for this study are charged-particle tracking and momentum measurement, charged $`\pi /K`$ separation, and lepton identification. Charged particle tracking is provided by a five-layer silicon vertex tracker (SVT) and a 40-layer drift chamber (DCH). The DIRC, a Cherenkov ring-imaging particle-identification system, is used (along with $`\mathrm{d}E/\mathrm{d}x`$ measured in the trackers) to separate charged kaons and pions. Electrons are identified using an electromagnetic calorimeter (EMC), which comprises 6580 thallium-doped CsI crystals. These systems are mounted inside a 1.5 T solenoidal superconducting magnet. Muons are identified in an instrumented flux return (IFR), in which resistive plate chambers are interleaved with the iron plates of the magnet flux return.
Simulated samples of signal $`B`$ decays, charmonium $`B`$ decays, generic $`B\overline{B}`$ decays, and continuum $`e^+e^{}q\overline{q}`$ (for $`q=u`$, $`d`$, $`s`$, or $`c`$) events are used to compute selection efficiencies, optimize event selection, and estimate certain backgrounds, as described below. The simulation is based on GEANT detector emulation software. The model for simulating signal $`B`$ decays is a $`bs\mathrm{}^+\mathrm{}^{}`$ matrix element calculation, which includes $`𝒪(\alpha _s)`$ and $`𝒪(\mathrm{\Lambda }_{\text{QCD}}/m_b)`$ corrections , convolved with $`BK^{()}`$ form factors predicted by light-cone QCD sum rules .
## 3 Event Selection
We select events that include two oppositely charged lepton candidates ($`e^+e^{}`$, $`\mu ^+\mu ^{}`$), a kaon candidate (either $`K^\pm `$ or $`K_S^0`$), and, for the $`BK^{}\mathrm{}^+\mathrm{}^{}`$ modes, a $`\pi ^\pm `$ candidate that, when combined with a kaon candidate, forms a $`K^{}`$ candidate. Electron (muon) candidates are identified by a likelihood (neural-net) based algorithm, and are required to have a minimum momentum $`p>0.3\mathrm{GeV}/c`$ ($`p>0.7\mathrm{GeV}/c`$) in the laboratory frame.
Bremsstrahlung photons from electrons are recovered by combining an electron candidate with up to one photon with $`E_\gamma >30\mathrm{MeV}`$. Recovered photons are restricted to an angular region in the laboratory frame of $`(\theta _\gamma ,\varphi _\gamma )=(\theta _e\pm 35\mathrm{mrad},\varphi _\mathrm{e}\pm 50\mathrm{mrad})`$ around the initial electron direction $`(\theta _e,\varphi _e)`$. Photon conversions and $`\pi ^0`$ Dalitz decays are removed by vetoing all $`e^+e^{}`$ pairs with invariant mass less than $`0.03\mathrm{GeV}/c^2`$, except in $`BK^{}e^+e^{}`$ modes, where we preserve acceptance at low invariant masses by retaining pairs that intersect inside the beam pipe.
Charged kaon candidates are tracks with $`\mathrm{d}E/\mathrm{d}x`$ and DIRC Cherenkov angle consistent with the angle expected for a kaon. $`\pi ^\pm `$ candidates are tracks that do not satisfy the $`K^\pm `$ selection. $`K_S^0`$ candidates are reconstructed from two oppositely charged tracks with an invariant mass (computed assuming they are $`\pi ^+\pi ^{}`$) consistent with the $`K_S^0`$ mass and a common vertex displaced from the average interaction point by at least 1 mm.
True $`B`$ signal decays produce narrow peaks in the distributions of two kinematic variables, which can be fitted to extract the signal and background yields. For a candidate system of $`B`$ daughter particles with total momentum $`𝐩_𝐁`$ in the laboratory frame and energy $`E_B^{}`$ in the $`\mathrm{{\rm Y}}(4S)`$ center-of-mass (CM) frame, we define $`m_{\mathrm{ES}}=\sqrt{(s/2+c^2𝐩_\mathrm{𝟎}𝐩_𝐁)^2/E_0^2c^2p_B^2}`$ and $`\mathrm{\Delta }E=E_B^{}\sqrt{s}/2`$, where $`E_0`$ and $`𝐩_\mathrm{𝟎}`$ are the energy and momentum of the $`\mathrm{{\rm Y}}(4S)`$ in the laboratory frame, and $`\sqrt{s}`$ is the total CM energy of the $`e^+e^{}`$ beams. For signal events, the $`m_{\mathrm{ES}}`$ distribution peaks at the $`B`$ meson mass with resolution $`\sigma 2.5\mathrm{MeV}/c^2`$, and the $`\mathrm{\Delta }E`$ distribution peaks near zero, with a typical width $`\sigma `$ 20 MeV. In $`BK\mathrm{}^+\mathrm{}^{}`$ channels, we perform a two-dimensional unbinned maximum-likelihood fit to the distribution of $`m_{\mathrm{ES}}`$ and $`\mathrm{\Delta }E`$ in the region $`m_{\mathrm{ES}}>5.2\mathrm{GeV}/c^2`$ and $`|\mathrm{\Delta }E|<0.25`$ GeV. In $`BK^{}\mathrm{}^+\mathrm{}^{}`$ decays, we perform a three-dimensional fit to $`m_{\mathrm{ES}}`$, $`\mathrm{\Delta }E`$, in the same regions as for $`BK\mathrm{}^+\mathrm{}^{}`$, and in addition we include in the fit the kaon-pion invariant mass for the region $`0.7<m_{K\pi }<1.1\mathrm{GeV}/c^2`$.
Backgrounds arise from four main sources: (1) random combinations of particles from $`q\overline{q}`$ events produced in the continuum, (2) random combinations of particles from $`\mathrm{{\rm Y}}(4S)B\overline{B}`$ decays, (3) $`B`$ decays to $`s\mathrm{}^+\mathrm{}^{}`$ final states other than the signal mode (“crossfeed”) and (4) $`B`$ decays to topologies similar to the signal modes. The first two (“combinatorial”) backgrounds typically arise from pairs of semileptonic decays of $`D`$ or $`B`$ mesons and produce distributions in $`m_{\mathrm{ES}}`$ and $`\mathrm{\Delta }E`$ which are broadly distributed compared to the signal. The third source has $`m_{\mathrm{ES}}`$ similar to signal, but the peak of the $`\mathrm{\Delta }E`$ distribution is significantly offset from the signal due to the addition of a random particle (“feed-up”) or omission of one of the $`B`$ daughters (“feed-down”). The last source arises from modes such as $`BJ/\psi K^{()}`$ (with $`J/\psi \mathrm{}^+\mathrm{}^{}`$) or $`BK^{()}\pi \pi `$ (with pions misidentified as muons), which have shapes similar to the signal. All selection criteria are optimized with simulated data or with data samples outside the region of the maximum-likelihood fit.
### 3.1 Combinatorial backgrounds
We suppress combinatorial background from continuum processes using a Fisher discriminant , which is a linear combination of variables with coefficients optimized to distinguish between signal and background. The variables used in the Fisher discriminant are the following kinematic quantities computed in the CM frame: (1) the ratio of second- to zeroth-order Fox-Wolfram moments for the event, computed using all charged tracks and neutral energy clusters; (2) the angle between the thrust axis of the $`B`$ candidate and that of the remaining particles in the event; (3) the production angle $`\theta _B`$ of the $`B`$ candidate with respect to the beam axis; and (4) the masses of $`K\mathrm{}`$ pairs with the same charge correlation as a semileptonic $`D`$ decay. The first three variables exploit the differences in event shapes between the jet-like topology of light quark pair production and the spherical shape of $`\mathrm{{\rm Y}}(4S)B\overline{B}`$ production. The fourth variable discriminates between $`D`$ meson decays, which have $`K\mathrm{}`$ mass distributed below the $`D`$ mass, and signal decays, which have a broad distribution in $`K\mathrm{}`$ mass.
We suppress combinatorial backgrounds from $`B\overline{B}`$ events using a likelihood function constructed from (1) the missing energy of the event, computed from all charged tracks and neutral energy clusters; (2) the vertex fit probability of all tracks from the $`B`$ candidate; (3) the vertex fit probability of the two leptons; and (4) the angle $`\theta _B`$. Missing energy provides the strongest suppression of combinatorial $`B\overline{B}`$ background events, which typically contain neutrinos from two semileptonic $`B`$ decays.
The parameters of the Fisher discriminant and the likelihood function are determined separately for each of the eight signal decay modes. The selection criteria for the background suppression variables are optimized simultaneously, and are chosen to minimize signal yield statistical uncertainties in each mode. The efficiencies of the Fisher and likelihood requirements are validated by comparing the efficiencies in data and in simulation using the $`BJ/\psi K^{()}`$ control sample.
### 3.2 Peaking backgrounds
The largest backgrounds that peak in $`m_{ES}`$ and $`\mathrm{\Delta }E`$ are $`B`$ decays to charmonium: $`BJ/\psi K^{()}`$ (with $`J/\psi \mathrm{}^+\mathrm{}^{}`$) and $`B\psi (2S)K^{()}`$ (with $`\psi (2S)\mathrm{}^+\mathrm{}^{}`$). We exclude dilepton pairs consistent with the $`J/\psi `$ mass ($`2.90<m_{e^+e^{}}<3.20\mathrm{GeV}/c^2`$ and $`3.00<m_{\mu ^+\mu ^{}}<3.20\mathrm{GeV}/c^2`$) or with the $`\psi (2S)`$ mass ($`3.60<m_\mathrm{}^+\mathrm{}^{}<3.75\mathrm{GeV}/c^2`$). This veto is applied to $`m_{e^+e^{}}`$ both with and without the inclusion of bremsstrahlung photon recovery. When a lepton radiates or is mismeasured, $`m_\mathrm{}^+\mathrm{}^{}`$ can shift away from the charmonium mass, while $`\mathrm{\Delta }E`$ shifts in a correlated manner. The veto region is extended in the $`(m_\mathrm{}^+\mathrm{}^{},\mathrm{\Delta }E)`$ plane to account for this correlation (Figure 3), removing nearly all charmonium events and simplifying the description of the background in the fit. Because the charmonium events removed by these vetoes are so similar to signal events, these modes provide large control samples (about 13700 events of $`BJ/\psi K^{()}`$ and 1000 events of $`B\psi (2S)K^{()}`$ in all) for studying signal shapes, selection efficiencies, and systematic uncertainties. After the vetoes on $`BJ/\psi K^{()}`$ and $`B\psi (2S)K^{()}`$ decays, the remaining peaking background from these processes is estimated from simulation to be 0.0–1.6 events, depending on the decay mode.
In the muon modes, the pion misidentification rate is significant ($`2\%`$), leading to additional peaking backgrounds from the decay $`B^{}D^0\pi ^{}`$ with $`D^0K^{}\pi ^+`$ or $`D^0K^{}\pi ^+`$, or from $`\overline{B}{}_{}{}^{0}D^+\pi ^{}`$ with $`D^+\overline{K}{}_{}{}^{0}\pi _{}^{+}`$ (and their charge conjugates ). These events are suppressed by vetoing events where the $`K^{()}\mu `$ mass is consistent with a hadronic $`D`$ decay. The remaining background from the charmless decays $`BK^{()}\pi \pi `$, $`BK^{()}K\pi `$, and $`BK^{()}KK`$ is estimated from data. We select control samples of $`BK^{()}h\mu `$ events with the same requirements as signal events, except that muon particle identification is no longer required for the hadron candidate $`h`$ and hadron identification requirements for pions and kaons are used instead. This results in a sample of predominantly hadronic $`B`$ decays. Each event is given a weight corresponding to the muon misidentification rate for the hadron divided by its hadron identification efficiency, and the number of peaking background events from hadronic $`B`$ decays is extracted from the weighted $`m_{\mathrm{ES}}`$ distribution through a maximum-likelihood fit similar to that used to extract a signal. The control sample calculations result in an estimate of 0.4–2.3 background events per decay channel for $`BK^{()}\mu ^+\mu ^{}`$ modes.
Finally, there is a peaking contribution to the electron modes from the rare decays $`BK^{}\gamma `$ (with photon conversion in the detector), $`BK^{()}\pi ^0`$, and $`BK^{()}\eta `$ (with a $`\pi ^0`$ or $`\eta `$ Dalitz decay to $`e^+e^{}\gamma `$). The sum of these backgrounds is estimated from simulation to be 0.0–1.4 per decay channel for the $`BK^{()}e^+e^{}`$ modes.
The number of peaking background events from all sources is shown in Table 1 for the individual decay modes. The peaking backgrounds in the modes with electrons are dominated by processes with real electrons, and the uncertainties are dominated by simulation statistics. The peaking backgrounds for modes with muons are dominated by hadrons misidentified as muons; the dominant uncertainty here is systematic and originates from the unknown $`K/\pi `$ composition of the contributing hadrons.
## 4 Fits
For $`BK\mathrm{}^+\mathrm{}^{}`$, a two-dimensional fit to $`m_{\mathrm{ES}}`$ and $`\mathrm{\Delta }E`$ is performed. For $`BK^{}\mathrm{}^+\mathrm{}^{}`$, the mass of the $`K^{}`$ is added as a third fit variable. The signal shapes are parameterized with separate Crystal Ball functions for $`m_{\mathrm{ES}}`$ and $`\mathrm{\Delta }E`$. Both the $`m_{\mathrm{ES}}`$ and $`\mathrm{\Delta }E`$ shape include a radiative tail, which accounts for the effects of bremsstrahlung of the electrons in the BABAR detector. The $`m_{\mathrm{ES}}`$ shape parameters are additionally assumed to have $`\mathrm{\Delta }E`$ dependence $`c_0+c_2(\mathrm{\Delta }E)^2`$; the variation of the $`m_{\mathrm{ES}}`$ width due to the quadratic term is typically a few percent of $`c_0`$. All signal shape parameters are fixed from the signal simulation, except for the mean and width parameters in $`m_{\mathrm{ES}}`$ and $`\mathrm{\Delta }E`$, which are fixed to values from charmonium data control samples (for the $`m_{\mathrm{ES}}`$ width, $`c_0`$ is fixed from charmonium data and $`c_2`$ is fixed from signal simulation). In the $`BK^{}\mathrm{}^+\mathrm{}^{}`$ channels, the $`K^{}`$ is fitted with a relativistic Breit-Wigner line shape. Adding the mass of the $`K^{}`$ to the likelihood fit increases the precision of the $`BK^{}\mathrm{}^+\mathrm{}^{}`$ branching fraction measurement by approximately 10%.
The background is modeled as the sum of three or four terms: (1) a combinatorial background shape with floating normalization, written as the product of an ARGUS function in $`m_{\mathrm{ES}}`$, a linear term in $`\mathrm{\Delta }E`$, and the product of $`\sqrt{m_{K\pi }m_Km_\pi }`$ and a quadratic function of $`m_{K\pi }`$ for the $`K^{}`$ modes; (2) a peaking background contribution, with the same shape as the signal, but with normalization fixed to estimates of the mean peaking backgrounds (see Table 1); and (3) terms with floating normalization to describe (a) background in $`BK\mathrm{}^+\mathrm{}^{}`$ ($`BK^{}\mathrm{}^+\mathrm{}^{}`$) from $`BK^{}\mathrm{}^+\mathrm{}^{}`$ ($`BK^{}\pi \mathrm{}^+\mathrm{}^{}`$) events with a lost pion, and (b) background in $`BK^{}\mathrm{}^+\mathrm{}^{}`$ from $`BK\mathrm{}^+\mathrm{}^{}`$ events with a randomly added pion. In the $`K^{}`$ modes, we allow an additional background (4) that uses our combinatorial shape in $`m_{\mathrm{ES}}`$ and $`\mathrm{\Delta }E`$, but peaks in $`m_{K\pi }`$ at the $`K^{}`$ mass. The yield of this term is fixed to $`(5\pm 5)\%`$ of the total combinatorial background, as determined from simulation. Because the normalizations for terms (1) and (3) are floating, as are the combinatorial background shape parameters, much of the uncertainty in the background is propagated into the statistical uncertainty on the signal yield obtained from the fit.
The direct $`CP`$ asymmetry $`A_{CP}`$ is also extracted from the fit to the modes $`B^+K^+\mathrm{}^+\mathrm{}^{}`$ and $`BK^{}\mathrm{}^+\mathrm{}^{}`$, where the $`b`$ flavor of signal candidates can be inferred directly from the charges of the final state $`K^{()}`$ hadrons. It is not possible at this time to measure $`A_{CP}`$ in the mode $`B^0K_S^0\mathrm{}^+\mathrm{}^{}`$, as the signal statistics are small and the $`b`$ flavor can only be inferred indirectly from properties of the other $`B`$ meson. The $`CP`$ asymmetry of the combinatorial background is allowed to float in the fit, while the asymmetries of the peaking background and crossfeed background are fixed to 0 and varied from -1 to 1 to evaluate the systematic uncertainty associated with these components.
## 5 Systematic Uncertainties
Table 2 lists the relative systematic uncertainties on the efficiency for each mode. The sources of uncertainty considered are: charged-particle tracking (0.8% per lepton, 1.4% per charged hadron), charged-particle identification (0.5% per electron pair, 1.3% per muon pair, 0.2% per pion, 0.6% per kaon), the continuum suppression cut (0.3%–2.2%, depending on the mode), the $`B\overline{B}`$ suppression cut (0.6%–2.1%), $`K_S^0`$ selection (0.9%), signal simulation statistics (0.4%–0.7%), and the number of $`B\overline{B}`$ events (1.1%). The uncertainty in the signal efficiency due to model dependence of form factors is evaluated for each mode to be the full range of variation from a set of models. The models considered are based on QCD sum rules , light-cone QCD sum rules , and lattice QCD . The model dependence enters through the variation in $`q^2`$ distributions; since the selection efficiency is not highly sensitive to this distribution the efficiency varies by only 4%–7%. For branching fraction measurements which combine modes, the systematic uncertainty is an appropriately weighted sum of correlated and uncorrelated sources from the contributing modes. The total systematic uncertainty in the signal efficiency introduces a systematic uncertainty $`\mathrm{\Delta }_{\text{eff}}`$ in the measured branching fraction.
Systematic uncertainties on the signal yields obtained from the maximum-likelihood fit arise from three sources: uncertainties in the parameters describing the signal shapes, uncertainties in the combinatorial background shape, and uncertainties in the peaking backgrounds. The uncertainties in the means and widths of the signal shapes are obtained by comparing data and simulated data in charmonium control samples. For modes with electrons, we also vary the fraction of signal events in the tail of the $`\mathrm{\Delta }E`$ distribution. To evaluate the uncertainty due to the background shape, we reevaluate the fit yields with three different parameterizations: (1) an exponential shape for $`\mathrm{\Delta }E`$, (2) a quadratic shape for $`\mathrm{\Delta }E`$, and (3) an $`m_{\mathrm{ES}}`$ ARGUS slope parameter $`\zeta `$ , which is linearly correlated with $`\mathrm{\Delta }E`$. The total systematic uncertainty in the fitted signal yield introduces a systematic uncertainty $`\mathrm{\Delta }_{\text{fit}}`$ in the measured branching fraction.
As cross checks, we also test our fit method by measuring the branching fractions and $`A_{CP}`$ of the $`J/\psi `$ $`K^{()}`$and $`\psi (2S)`$ $`K^{()}`$final states using the vetoed charmonium events. The measured branching fractions are in good agreement with the 2004 world average and the recent BABAR measurement . The direct $`CP`$ asymmetries $`A_{CP}`$ are all consistent with zero. We also analyze $`K^{()}`$$`e\mu `$ samples and obtain signal yields consistent with zero.
## 6 Results
The results for the fits to the individual decay modes are shown in Table 3. Branching fraction uncertainties are predominantly statistical, with total systematic uncertainties of about 10% in each decay mode.
To combine the results from the individual modes into the total $`BK\mathrm{}^+\mathrm{}^{}`$ and $`BK^{}\mathrm{}^+\mathrm{}^{}`$ branching fractions, we perform a maximum-likelihood fit where the event yields in all of the modes, after being corrected for selection efficiency and $`K^{()}`$ branching fractions, are constrained to the same value. In this fit we constrain the production rates of charged and neutral $`B`$ meson pairs in the $`\mathrm{{\rm Y}}(4S)`$ decay to be the same. We also constrain the total width ratio $`\mathrm{\Gamma }(B^0)/\mathrm{\Gamma }(B^+)`$ to the world average $`B`$ meson lifetime ratio $`\tau _+/\tau _0=1.086\pm 0.017`$ ; all branching fractions from combined fits are expressed in terms of the $`B^0`$ total width. In $`BK^{}\mathrm{}^+\mathrm{}^{}`$ we perform the fit with the pole region included, adding the constraint:
$$\mathrm{\Gamma }(BK^{}\mu ^+\mu ^{})/\mathrm{\Gamma }(BK^{}e^+e^{})=0.752.$$
As described in Section 1, this originates from the enhanced contribution in $`BK^{}e^+e^{}`$ from the photon penguin amplitude near $`q^2=0`$. The branching fraction for this combined fit is expressed in terms of the $`B^0K^0\mu ^+\mu ^{}`$ channel. We also perform combined fits to the electron and muon channels separately. Table 4 summarizes the results for the combined branching fractions. The combined significance of the signal, including statistical and systematic uncertainties, is $`6.6\sigma `$ and $`5.7\sigma `$ for the $`BK\mathrm{}^+\mathrm{}^{}`$ and $`BK^{}\mathrm{}^+\mathrm{}^{}`$ modes, respectively.
The combined fits to $`BK\mathrm{}^+\mathrm{}^{}`$ and $`BK^{}\mathrm{}^+\mathrm{}^{}`$ are shown in Figures 4 and 5. They correspond to the branching fraction measurements of
$$(BK\mathrm{}^+\mathrm{}^{})=(0.34_{0.07}^{+0.07}\pm 0.03)\times 10^6$$
$$(BK^{}\mathrm{}^+\mathrm{}^{})=(0.78_{0.17}^{+0.19}\pm 0.12)\times 10^6$$
where the first error is statistical and the second is systematic. The satellite peak in the $`\mathrm{\Delta }E`$ distribution at $`0.15`$ $`\mathrm{GeV}`$ for the $`BK\mathrm{}^+\mathrm{}^{}`$ fit arises from the feed-down component of the fit. Examination of events in this region confirms that the addition of a charged or neutral pion results in candidates consistent with $`BK^{}\mathrm{}^+\mathrm{}^{}`$ signal. The effect of such events on the $`BK\mathrm{}^+\mathrm{}^{}`$ signal yield has been studied with fits to simulated samples, and the associated bias to the signal yield is negligible.
For the combined modes we measure the direct $`CP`$ asymmetries
$$A_{CP}(B^+K^+\mathrm{}^+\mathrm{}^{})=0.08\pm 0.22\pm 0.11$$
$$A_{CP}(BK^{}\mathrm{}^+\mathrm{}^{})=0.03\pm 0.23\pm 0.12$$
where the systematic uncertainty is dominated by the unknown asymmetry in the peaking backgrounds.
Table 4 also contains the results from independent fits to the muon and electron channels, with no constraint enforced on the ratio of the two. From these fits we find the ratio of muon to electron branching fractions over the full range of $`q^2`$ to be
$$R_K=1.06\pm 0.48\pm 0.05$$
$$R_K^{}=0.93\pm 0.46\pm 0.06$$
where these are expected in the Standard Model to be 1.00 and 0.75, respectively, with small theoretical uncertainties.
We also perform the fit to the $`BK^{}\mathrm{}^+\mathrm{}^{}`$ channels with the pole region ($`q^2<0.1\mathrm{GeV}^2/c^4`$) excluded, which modifies the Standard Model constraint on the ratio of branching fractions from 0.752 to 1. With the pole region removed, we obtain
$`(BK^{}e^+e^{},q^2>0.1\mathrm{GeV}^2/c^4)`$ $`=`$ $`(0.65_{0.21}^{+0.24}\pm 0.12)\times 10^6`$
$`(BK^{}\mu ^+\mu ^{},q^2>0.1\mathrm{GeV}^2/c^4)`$ $`=`$ $`(0.89_{0.30}^{+0.35}\pm 0.13)\times 10^6`$
$`(BK^{}\mathrm{}^+\mathrm{}^{},q^2>0.1\mathrm{GeV}^2/c^4)`$ $`=`$ $`(0.74_{0.18}^{+0.20}\pm 0.12)\times 10^6`$
From the fits to the $`BK^{}\mathrm{}^+\mathrm{}^{}`$ mode above the pole region, we find the ratio of muon to electron branching fractions
$$R_K^{}(q^2>0.1\mathrm{GeV}^2/c^4)=1.37_{0.74}^{+0.74}\pm 0.11,$$
which is expected to be 1.00 in the Standard Model.
The measured $`BK^{}\mathrm{}^+\mathrm{}^{}`$ branching fraction is consistent with the previously published BABAR result measured with 113$`\text{ fb}^1`$, $`(BK^{}\mathrm{}^+\mathrm{}^{})=(0.88_{0.29}^{+0.33}\pm 0.10)\times 10^6`$. The $`BK\mathrm{}^+\mathrm{}^{}`$ branching fraction is somewhat lower than the previous published BABAR result of $`(BK\mathrm{}^+\mathrm{}^{})=(0.65_{0.13}^{+0.14}\pm 0.04)\times 10^6`$ . Including correlations between events selected, the significance of this difference is equivalent to $`2.2\sigma `$. The ratios $`R_{K^{()}}`$ of muon to electron branching fractions are also consistent with the previously published values.
As a cross check, we have also examined the $`m_{\mathrm{}\mathrm{}}`$ distribution of candidate events in the signal region. Of particular interest would be any evidence for a large excess in the $`m_{\mathrm{}\mathrm{}}`$ spectrum near the lower boundaries of the veto regions, which could indicate $`J/\psi `$ or $`\psi (2S)`$ events escaping the veto. The $`m_{\mathrm{}\mathrm{}}`$ spectrum, shown in Figure 6, exhibits no evidence for such an enhancement. The data points cover the full allowed region in $`m_{\mathrm{}\mathrm{}}`$, including the pole region in $`BK^{}\mathrm{}^+\mathrm{}^{}`$.
Figure 7 summarizes the experimental measurements (points) and their theoretical predictions (boxes). The measurements are in general agreement with the range of rates predicted by the form factor calculations in Ref. . The measured $`B^+K^+\mathrm{}^+\mathrm{}^{}`$ branching fraction is significantly lower than the range estimated in .
## 7 Summary
We have measured the branching fractions and direct $`CP`$ asymmetries $`A_{CP}`$ of the rare FCNC decays $`BK\mathrm{}^+\mathrm{}^{}`$ and $`BK^{}\mathrm{}^+\mathrm{}^{}`$. We find the (lepton-flavor–averaged, $`B`$-charge–averaged) branching fractions
$$(BK\mathrm{}^+\mathrm{}^{})=(0.34\pm 0.07\pm 0.03)\times 10^6$$
$$(BK^{}\mathrm{}^+\mathrm{}^{})=(0.78_{0.17}^{+0.19}\pm 0.12)\times 10^6,$$
consistent with the Standard Model predictions for these modes. We find $`A_{CP}(B^+K^+\mathrm{}^+\mathrm{}^{})`$ and $`A_{CP}(BK^{}\mathrm{}^+\mathrm{}^{})`$ consistent with zero, to a precision of 25%. We have also measured the ratios of the branching fractions of muon pairs to that of electron pairs; these are also consistent with the Standard Model to a precision of 50%. All of the measurements are statistically limited.
## 8 Acknowledgments
We are grateful for the extraordinary contributions of our PEP-II colleagues in achieving the excellent luminosity and machine conditions that have made this work possible. The success of this project also relies critically on the expertise and dedication of the computing organizations that support BABAR. The collaborating institutions wish to thank SLAC for its support and the kind hospitality extended to them. This work is supported by the US Department of Energy and National Science Foundation, the Natural Sciences and Engineering Research Council (Canada), Institute of High Energy Physics (China), the Commissariat à l’Energie Atomique and Institut National de Physique Nucléaire et de Physique des Particules (France), the Bundesministerium für Bildung und Forschung and Deutsche Forschungsgemeinschaft (Germany), the Istituto Nazionale di Fisica Nucleare (Italy), the Foundation for Fundamental Research on Matter (The Netherlands), the Research Council of Norway, the Ministry of Science and Technology of the Russian Federation, and the Particle Physics and Astronomy Research Council (United Kingdom). Individuals have received support from CONACyT (Mexico), the A. P. Sloan Foundation, the Research Corporation, and the Alexander von Humboldt Foundation. |
warning/0507/astro-ph0507240.html | ar5iv | text | # Optical identification of 𝐼𝑆𝑂 far-infrared sources in the Lockman Hole using a deep VLA 1.4 GHz continuum survey
## 1 Introduction
The detection of the far-infrared Cosmic Infrared Background (CIRB) radiation by FIRAS and DIRBE on COBE was the important step toward understanding the physical property of the cosmological far-infrared (FIR) radiation. The CIRB was interpreted as the integrated emission by dust in the distant galaxies (Puget et al., 1996; Fixsen et al., 1998; Hauser et al., 1998) which set a relevant constraint on the evolution of cosmic sources. The CIRB is 10 times brighter than the expected intensity based on the assumption that the infrared emissivity of galaxies does not change with cosmic time (Takeuchi et al., 2001; Franceschini et al., 2001) and has comparable contribution to the total intensity as expected from the optical counts from the Hubble deep field and the Subaru deep field(Totani et al., 2001; Totani & Takeuchi, 2002). Therefore the CIRB’s excess suggests that galaxies in the past were much more active in the FIR and that a substantial fraction of the total energy emitted by high-redshift galaxies were absorbed by dust and re-emitted at long wavelengths.
The SCUBA bolometer camera(Holland et al., 1999) on the sub-millimeter telescope JCMT was able to resolve at least half of the CIRB at wavelength of 850 $`\mu `$m into a population of very luminous infrared galaxies at $`z>1`$ (Smail, Ivision & Blain, 1997; Hughes et al., 1998; Barger et al., 1998; Blain et al., 1999). ISOCAM mid-infrared surveys (Rowan-Robinson et al., 1997; Flores et al., 1999; Mann et al., 2002) have also reported a higher infrared luminosity density (thus a higher star formation rate) at $`0.5<z<1`$ than estimated by previous optical studies.
$`IRAS`$ was able to detect infrared galaxies only to moderate redshifts ($`z0.1`$) with the exception of a few hyperluminous and/or lensed objects such as FSC 10214+4724 (Rowan-Robinson et al., 1991). The improvement of the sensitivity and the extension to the longer wavelength (170 $`\mu `$m) in FIR with ISOPHOT instrument onboard the Infrared Space Observatory ($`ISO`$) provides us with a new tool to study FIR emission from galaxies at the higher-redshifts than $`IRAS`$, and the exploration of the “optically dark side” of the star formation history through a deep FIR survey was the obvious next step.
The spectral energy distribution(SED) of actively star-forming galaxies peaks at $`\lambda 100`$ $`\mu `$m. The “negative” $`k`$-correction makes observations at wavelengths longer than this FIR peak advantageous for detecting high-redshift galaxies. Furthermore, such measurements give the total luminosity without any model-dependent bolometric correction. Therefore several deep surveys were undertaken with the ISOPHOT at 90 $`\mu `$m and/or 170 $`\mu `$m. We performed a deep FIR survey of two fields in the Lockman Hole region in both 90 $`\mu `$m and 170 $`\mu `$m bands as a part of the Japanese/UH cosmology project (Kawara et al., 1998). A 170 $`\mu `$m survey of two fields in the southern Marano area and two fields in the northern ELAIS fields with a combined area of 4 square degrees constitute the FIRBACK program (Puget et al., 1999). A 90 $`\mu `$m survey in ELAIS (Efstathiou et al., 2000) covered 11.6 square degrees.
In order to explore the nature of $`ISO`$ FIR sources in the Lockman Hole fields, we have identified counterparts to the sources at optical and radio wavelengths. We obtained their photometric characteristics and measured their redshifts in order to understand the genuine nature of $`ISO`$ FIR sources in the Lockman Hole fields. Section 2 describes the observations and the data, Section 3 presents the method and results of the source identification, and Section 4 describes the discussions of $`ISO`$ FIR sources in the Lockman Hole fields. The summary is presented in Section 5. Appendix describes the comparison of our catalogs with those in Rodighiero et al. (2003) and Rodighiero & Franceschini (2004), which reduced same data with their own method.
Throughout the paper, a flat Universe with $`H_0=70\mathrm{k}\mathrm{m}\mathrm{s}^1\mathrm{Mpc}^1,\mathrm{\Omega }_M=0.3`$ and $`\mathrm{\Omega }_\lambda =0.7`$ is adopted.
## 2 Data
### 2.1 ISO far-infrared catalogs
Our FIR survey was performed in the ISOPHOT bands C\_90 (90 $`\mu `$m for the reference wavelength) and C\_160 (170 $`\mu `$m; see Kawara et al., 1998; Matsuhara et al., 2000; Kawara et al., 2004) in the Lockman Hole, where the HI column density is measured to be the lowest in the sky (Lockman et al., 1986) and thus the confusion noise due to the infrared cirrus is expected to be the lowest. The survey includes two fields named LHEX and LHNW, each of which covered approximately 44′$`\times `$44′square area. One of the fields, LHEX, was also the target of the ROSAT Lockman Hole ultra-deep HRI survey (Hasinger et al., 1998). The 90 $`\mu `$m and 170 $`\mu `$m observations with $`ISO`$(Kessler et al., 1996) and the data reduction are described in detail by Kawara et al. (1998, 2004). As shown in Figure 1, the ISOPHOT instrument (Lemke et al., 1996) was used to map a total area of $``$0.9 square degrees, consisting of two 44′$`\times `$ 44′ fields.
The $`IRAF`$<sup>1</sup><sup>1</sup>1IRAF is distributed by NOAO, which is operated by AURA, Inc., under contract to the NSF. DAOPHOT package, which has been developed to perform stellar photometry in crowded fields(Stetson, 1987), was used to perform the source extraction from the FIR maps. The survey images are very crowded, with two or more sources frequently appear blended. The FWHM measurements show that most bright sources subtends no more than two detector pixels, implying these sources are detected as point sources. The final 90 $`\mu `$m and 170 $`\mu `$m source catalogs with a signal to noise ratio of three or greater include 223 and 72 sources, respectively (Kawara et al., 2004). UGC 06009<sup>2</sup><sup>2</sup>2If more accurate fluxes are measured in UGC 06009, the fluxes of FIR sources should be re-scaled accordingly., which is the only IRAS source locating within our survey fields, was used for the flux scaling assuming F(90 $`\mu `$m)=1218 mJy and F(170 $`\mu `$m)=1133 mJy. The flux density of the cataloged sources ranges between 40 mJy and 400 mJy at 90 $`\mu `$m and between 90 mJy and 410 mJy at 170 $`\mu `$m.
The surface number densities of $`ISO`$ FIR sources are sufficiently high (10-20 beams per source) that the derived quantities such as flux, position, completeness, and detection limits are significantly affected by source confusion. To evaluate these effects, Kawara et al. (2004) performed a set of simulations by adding artificial sources to the observed FIR maps. These simulations have shown that the measured fluxes of faint sources are indeed significantly overestimated and thus the correction for the flux bias is important. The following expressions are used for the correction:
$`F(90\mu \mathrm{m})=66+0.78\times F^C(90\mu \mathrm{m})+0.0001\times F^C(90\mu \mathrm{m})^2`$ (1)
$`F(170\mu \mathrm{m})=160+0.32\times F^C(170\mu \mathrm{m})+0.0007\times F^C(170\mu \mathrm{m})^2`$ (2)
where $`F`$ and $`F^C`$ denote the observed flux and the flux after correction for the bias effect, respectively. These expressions represent the results from the simulations by Kawara et al. (2004) with 2-3% accuracy for $`F(90\mu \mathrm{m})100\mathrm{m}\mathrm{J}\mathrm{y}`$ or $`F(170\mu \mathrm{m})250\mathrm{m}\mathrm{J}\mathrm{y}`$, and 10% for $`F(170\mu \mathrm{m})<250\mathrm{m}\mathrm{J}\mathrm{y}`$. Thus, the errors resulting from the expressions are small enough to be ignored. In this paper, we refer these corrected flux with Equ. (1) and (2).
Positional uncertainties are also estimated from these simulations. Position error depends on the brightness of FIR sources, and the dispersion in the measured position relative to the input position of the artificial sources, $`\sigma _{pos}`$, is larger toward fainter sources. The derived positional uncertainties from the simulations are $`\sigma _{pos}(90\mu \mathrm{m})20^{\prime \prime }`$ for $`F(90\mu \mathrm{m})100\mathrm{m}\mathrm{J}\mathrm{y}`$ and $`\sigma _{pos}(170\mu \mathrm{m})35^{\prime \prime }`$ for $`F(170\mu \mathrm{m})200\mathrm{m}\mathrm{J}\mathrm{y}`$. Their simulation also show that the completeness rapidly decrease as the flux decrease below 200 mJy; for example, the detection rate is 73% for sources with $`F^C(90\mu \mathrm{m})=43\mathrm{m}\mathrm{J}\mathrm{y}`$, and 64% for $`F^C(170\mu \mathrm{m})=102\mathrm{m}\mathrm{J}\mathrm{y}`$.
### 2.2 Radio observations
The $`ISO`$ positional uncertainties of 20<sup>′′</sup> \- 35<sup>′′</sup> are too large to search for optical counterparts directly because there are always several faint optical sources within each error circle. Thus we have taken a two-steps approach to identify FIR sources. In the first step, deep Very Large Array (VLA) 1.4 GHz imaging was performed and radio counterparts were found utilizing the well-known FIR-radio correlation (Condon, 1992; Yun, Reddy, & Condon, 2001). This approach may introduce a slight bias in favor of star-forming galaxies, but the AGN dominated sources are rare in general among the FIR selected sample (less than a few per cent, see Yun, Reddy, & Condon, 2001). Once radio counterparts are known, then the positional uncertainty is reduced greatly to $`1^{\prime \prime }`$, and the second step of searching for optical counterparts becomes straight forward.
Deep 1.4 GHz continuum images of the LHNW and LHEX fields were obtained using the NRAO<sup>3</sup><sup>3</sup>3The National Radio Astronomy Observatory is a facility of the National Science Foundation operated under cooperative agreement by Associated Universities, Inc. VLA in the B-configuration in February 2000 and March-April 2001, respectively. Four separate pointings cover a $`45^{}\times 45^{}`$ square region centered on the two $`ISO`$ survey fields. The angular resolution of the final images was about $`5^{\prime \prime }`$ (FWHM).
The rms noise achieved in the central $`32^{}\times 32^{}`$ mapping region of the LHEX field is $`1\sigma 15\mu `$Jy. After combining additional data from the archive, the rms noise in the northeast quadrant centered near the ROSAT Deep Survey field (Hasinger et al., 1998) has been improved to $`1\sigma 10\mu `$Jy. Our new VLA 1.4 GHz continuum image has about 2-3 times higher angular resolution and improved sensitivity over the previous observations by de Ruiter et al. (1997, $`\theta 12^{\prime \prime }`$ and $`1\sigma 30\mu `$Jy).
In the LHNW field, four separate pointings were also used. However, the imaging dynamic range in the LHNW field was severely limited by the presence of a bright radio source 3C 244.1 (4.2 Jy at 1.4 GHz) just outside the primary beam. Incomplete subtraction of the time varying PSF response resulted in a strong gradient of additional “noise” across the radio continuum image with a median value of $`1\sigma 30\mu `$Jy in the image center. Therefore the radio source identification is dependent entirely on the local noise level, and the radio source catalog is complete only at the highest flux density level ($`200\mu `$Jy). A more detailed descriptions of these VLA observations will be presented in a later paper (Yun et al., in preparation).
In this paper, we use 387 radio sources brighter than 60 $`\mu `$Jy($`4\sigma `$) in the central $`32^{}\times 32^{}`$ region of the LHEX field and 76 sources in the entire LHNW field.
### 2.3 Optical $`R`$ and $`I`$band imaging
The $`I`$band images of the two Lockman Hole fields were obtained using the 8K CCD Camera (Luppino et al., 1996) attached to the f/10 Cassegrain focus of the University of Hawaii 88<sup>′′</sup> telescope on May 19-24 in 1999. This camera has a 18′$`\times `$18′ field of view. A $`2\times 2`$-pixel binning was used yielding a sampling resolution of 0$`\stackrel{}{\mathrm{.}}`$26 pixel<sup>-1</sup>. Nine pointings were used in each field to cover each field completely. The total exposure times were 14 and 21 minutes for the LHEX and LHNW fields, respectively. The seeing was about $`0\stackrel{}{\mathrm{.}}6`$, and the $`5\sigma `$ detection limit reaches $`I2223`$ depending on the location on the camera. Flux calibration was performed using the observations of SA103 and SA104 (Landolt, 1992) with a systematic uncertainty of $`0.05`$ mag.
The deep $`R`$band imaging of the both fields was carried out using the prime-focus camera, Suprime-Cam (Miyazaki et al., 2002) on the Subaru Telescope on March 19 in 2001. This camera has ten 4K$`\times `$2K CCDs which provides a 34′$`\times `$27′ field of view with 0$`\stackrel{}{\mathrm{.}}`$2/pixel sampling. During our observing run, one CCD on the corner of the focal plane was not available. To cover entire $`ISO`$ survey fields, four different pointings were used in each field. The exposure times were 30 minutes for the two pointings on the west and 25 minutes for the two pointing on the east. The wide-field optical corrector unit of the Suprime-Cam introduces the significant image distortion on the focal plane (e.g., 18<sup>′′</sup> distortion at 20′away from the center). This distortion is corrected using IRAF task GEOTRAN. The $`5\sigma `$ detection limit reaches 26.5 mag under typical seeing of $``$ 0$`\stackrel{}{\mathrm{.}}`$8. Flux calibration was performed using the observations of SA101 (Landolt, 1992) and SA57 (Majewski et al., 1994) fields, with an estimated systematic uncertainty of $`0.08`$ mag.
The astrometry calibration of all optical images was obtained by comparing with the positions of stars in the United States Naval Observatory (USNO) A2.0 catalog (Monet et al., 1998). The astrometric uncertainty is estimated to be less than 1$`\stackrel{}{\mathrm{.}}`$0 for both $`R`$ and $`I`$band imaging.
### 2.4 Optical Spectroscopy
We performed spectroscopy of optical objects identified as radio counterparts to the FIR sources. Keck II telescope at the Mauna Kea Observatory and WIYN<sup>4</sup><sup>4</sup>4The WIYN Observatory is a joint facility of the University of Wisconsin-Madison, Indiana University, Yale University, and the National Optical Astronomy Observatories. 3.5m telescope at the Kitt Peak National Observatory were used with wide wavelength coverage, allowing many emission features for line identification.
#### 2.4.1 Keck Observations
Spectroscopic observations with Keck II telescope were performed preferentially in order of optical brightness on March 30-31 in 2000 and on January 23-24 in 2001, using the Echelle Spectrograph and Imager (ESI:Sheinis et al. 2000) in the low-dispersion mode. The long slit was set to the optical center of the radio sources. The position angle of the slit was adjusted so that radio sources can be observed simultaneously from 3,900Å to 11,000Å. The slit width and length were 1<sup>′′</sup>(6.5 pixels) and 8′, respectively. The spectral resolution ranged from 800 at the blue part to 300 per pixel at the red part. Exposure times used were between 300 sec and 5400 sec, depending on the optical brightness of the target sources. Standard data reductions were carried out using IRAF. Presence of emission or absorption lines was searched by eye. In this paper, we present 15 Keck II spectra of $`ISO`$ FIR galaxies.
#### 2.4.2 WIYN/HYDRA Observations
Spectroscopic observations were performed with the HYDRA multi-object spectrometer on the WIYN 3.5m telescope on February 19 in 2002 and February 6-7 in 2003, using the red cables (2<sup>′′</sup> diameter), the 316 line mm<sup>-1</sup> grating with G5 filter, and the bench camera. The spectra cover the wavelength region 5,020$``$10,000 Å with a spectral resolution of 2.64 Å /pixel.
To cope with faint spectra in fiber-based spectroscopy, beam switching was used to efficiently remove sky emission features such as fringes and time-dependent airglow emission lines which are heavily blended in some cases. The beam switching requires two different pointings of the telescope. At pointing A, one half of 96 fibers available for HYDRA were centered on the targets while the others were set to the sky. At pointing B, the roles of fibers were switched. Beam switching was repeated 7 times every 30-minute, for a total on-source integration time of 210 minutes.
A specialized IRAF package DOHYDRA (Valdes, 1995) was used to perform scattered light removal, spectra extraction, flat fielding, fiber throughput correction, and wavelengths calibration. Sky subtraction and subsequent spectrum co-adding require a careful treatment and are not fully supported by DOHYDRA. We thus developed special IDL routines to handle these tasks. Sky brightness, which is dominated by airglow emission lines, varies with airmass and time. The sky spectra were removed from the “on-target” fibers by using the sky background determined from the preceding and following “sky” fibers. Some scaling of the sky background were required to subtract time-dependent OH airglow emission. After subtracting the sky, the CCD fringes on “on-target” fibers reduced greatly. The resultant sky-subtracted spectra were then coadded to improve the signal-to-noise ratio. Extra-galactic emission lines were searched for by eye and by comparing the source spectra with the airglow spectra and distinguishing the residual airglow from real emission lines.
Spectra were obtained for a total of 50 objects in the two fields. The brightness of the target objects range between $`R=17`$ and $`R=21`$. Nine of the 50 objects are $`ISO`$ FIR sources identified as radio counterparts. The remainder are other faint radio sources identified by the deep radio survey.
## 3 Radio Counterpart Identification
In this section, we illustrate the processes of identifying FIR sources with 1.4 GHz radio sources. Our 1.4 GHz survey has been carried out within the two $`ISO`$ fields in the Lockman Hole. The areas where the present identification work was performed, are 1089 arcmin<sup>2</sup> in LHEX and 1552 arcmin<sup>2</sup> in LHNW. A typical position uncertainty of the radio source is $`1`$<sup>′′</sup>, negligibly smaller than the typical $`ISO`$ FIR position uncertainty.
The original catalogs by Kawara et al. (2004) contain sources with signal-to-noise ratios of 3$`\sigma `$ or better. In the present work, we exclude faint sources which have $`F^C(90\mu \mathrm{m})<43\mathrm{m}\mathrm{J}\mathrm{y}`$ or $`F^C(170\mu \mathrm{m})<102\mathrm{m}\mathrm{J}\mathrm{y}`$, which correspond to the catalog fluxes, $`F(90\mu \mathrm{m})<100\mathrm{m}\mathrm{J}\mathrm{y}`$ or $`(170\mu \mathrm{m})<200\mathrm{m}\mathrm{J}\mathrm{y}`$. Accordingly, we adopt the position uncertainty of $`1\sigma _{POS}(90\mu \mathrm{m})=18^{\prime \prime }`$ and $`1\sigma _{POS}(170\mu \mathrm{m})=30^{\prime \prime }`$ (see Figure 4 of Kawara et al., 2004). Excluding UGC 06009, which is the only $`IRAS`$ source within the survey areas and is used for the flux-calibration, the total numbers of 90 $`\mu `$m and 170 $`\mu `$m sources included in this study are 116 and 20, respectively.
Large FIR error circles and a high number density of faint radio sources ensures that there is a high probability of having confusing multiple radio sources within each FIR error circle. For example, 25 out of 49 ($`50`$%) of the 90 $`\mu `$m sources in the LHEX field have two or more radio sources with in a $`3\sigma `$ error circle (54<sup>′′</sup> radius; see Table 1). Thus the task at hand is to reject confusing radio sources which are within the error circle by chance, by quantifying the reliability of every identification through a statistical approach as discussed below.
### 3.1 Likelihood ratio analysis
The likelihood ratio analysis using cross-association (de Ruiter, Arp, & Willis, 1977; Sutherland & Saunders, 1992; Mann et al., 1997; Rutledge et al., 2000) is a commonly used technique for identifying sources in crowded fields. Here, we adopt the prescription given by Rutledge et al. (2000). It is assumed that each FIR source is physically associated with either one radio source or none. It is also assumed that there are two types of FIR sources – one type has one real radio counterpart while the second type has no real radio counterpart. Because of the source confusion due to the high number density of radio sources, in many cases there are one or more radio sources found within a 3$`\sigma `$ error circle centered at the FIR position.
To assess the reliability of individual identifications, we begin with the likelihood ratio which is described by de Ruiter, Arp, & Willis (1977) and Wolstencroft et al. (1986). Position uncertainty and number density of radio sources are known, and it is assumed that each source has one or more candidate radio counterparts within a $`3\sigma `$ error circle. A dimensionless angular distance $`r`$, between the FIR and radio positions, is defined as,
$$r^2=\left(\frac{\mathrm{\Delta }\theta ^2}{\sigma _{pos}^2+\sigma _R^2}\right)$$
(3)
where $`\mathrm{\Delta }\theta `$ is the positional differences between the FIR and radio sources, and $`\sigma _R`$ is the standard deviation of the radio positions. As already discussed, $`\sigma _R`$ is typically $``$1<sup>′′</sup> and negligible when compared with FIR error $`\sigma _{pos}18`$<sup>′′</sup> or 30<sup>′′</sup>. Thus we adopt $`r=\mathrm{\Delta }\theta /\sigma _{pos}`$. The surface density of radio sources, $`n(f)`$, is the number density of radio sources with 1.4 GHz flux equal to or greater than $`f`$, and it is adopted from Richards (2000).
Assuming that a FIR source and its true radio counterpart are located at the same position, the measured separation $`r`$ is due to the FIR position uncertainty. Then, the probability, $`dp_{id}`$, of having a real radio counterpart at a distance between $`r`$ and $`r+dr`$ from the FIR sources, is $`dp_{id}=r\mathrm{exp}(r^2/2)dr`$. The probability, $`dp_c`$, of finding a first confusing object between $`r`$ and $`r+dr`$ from the FIR source is a product of two probabilities; the probability of not having a counterpart within $`r`$, which is given by $`e^{\pi r^2\sigma ^2n(f)}`$, and the probability of a confusing object between $`r`$ and $`r+dr`$, which is $`1e^{2\pi rdr\sigma ^2n(f)}2\pi r\sigma ^2n(f)dr`$. Thus the probability, $`dp_c`$, is expressed as $`dp_c=2\pi \sigma ^2n(f)rdr\times \mathrm{exp}\left(\pi r^2\sigma ^2n(f)\right)`$.
The likelihood ratio is defined as the ratio of $`dp_{id}/dp_c`$, and
$$LR(r,f)=\frac{1}{2\pi \sigma ^2n(f)}\mathrm{exp}\left(\frac{r^2}{2}+\pi r^2\sigma ^2n(f)\right).$$
(4)
For each FIR source, $`LR`$ is calculated for all radio sources within the 3$`\sigma `$ error circle. Then the frequency distribution $`N(LR)`$ is computed. Figures 2 and 3 show the distribution of $`LR`$ as thick-lined histograms for 90 $`\mu `$m and 170 $`\mu `$m. The $`LR`$ distribution was derived only in the LHEX field because the LHNW field radio data suffers from a non-uniform noise and complex source statistics.
Once the distribution of the likelihood ratio $`LR`$ is determined, a reliability $`R`$ for identification with $`LR`$ can be quantified as
$$R(LR)=\frac{N_{true}(LR)}{N_{true}(LR)+N_{false}(LR)},$$
(5)
where $`N_{true}(LR)`$ and $`N_{false}(LR)`$ are the numbers of $`true`$ and $`false`$ radio counterparts, respectively. In practice, however, it is generally unknown which radio source is the real counterpart or what $`N_{true}`$ and $`N_{false}`$ are.
The denominator in Equ. (5) can be re-written as
$$N_{true}(LR)+N_{false}(LR)=N_{source}(LR),$$
(6)
where $`N_{source}(LR)`$ is the total number of radio sources with $`LR`$ in the field. In general $`N_{true}`$ is not known, but we can estimate $`N_{true}`$ from the difference between $`N_{source}`$ and $`N_{ran}`$, where $`N_{source}`$ and $`N_{ran}`$ are the numbers of radio sources in regions containing FIR sources and containing no FIR sources, respectively. There is not enough area free from $`ISO`$ FIR sources because of the high source density and large positional uncertainty of the $`ISO`$ FIR sources. We estimated $`N_{ran}(LR)`$ by randomly assigning positions to radio sources in the catalog. These simulations are repeated 100 times and averaged. The resultant $`N_{ran}(LR)`$ is shown as shaded histograms in Figures 2 and 3. Comparing the observed distributions $`N_{source}(LR)`$ with the randomly generated $`N_{ran}(LR)`$, it is clear that $`N_{source}(LR)`$ has excess over $`N_{ran}(LR)`$ for $`\mathrm{log}(LR)>0`$.
Our radio catalog consists of both kinds of radio sources: “real” and “false” counterparts of FIR sources. Thus, $`N_{ran}(LR)`$ should be greater than $`N_{false}(LR)`$. Using $`N_{ran}(LR)`$, $`N_{true}(LR)`$ is represented as;
$$N_{true}(LR)N_{source}(LR)N_{ran}(LR).$$
(7)
Here we introduce the modified reliability $`\stackrel{~}{R}(LR)`$, as defined by
$$\stackrel{~}{R}(LR)=\frac{N_{source}(LR)N_{ran}(LR)}{N_{source}(LR)}\frac{N_{true}(LR)}{N_{source}(LR)}=R(LR).$$
(8)
$`\stackrel{~}{R}`$ is plotted on Figures 4 and 5 together with quadratic approximations of $`\stackrel{~}{R}`$ as a function of log($`LR`$).
As seen in Eqa. (8), $`\stackrel{~}{R}(LR)`$ is less than the true reliability $`R(LR)`$. The typical ratio $`R/\stackrel{~}{R}`$ for $`log(LR)>0`$ can be estimated as follows. According to Table 1, the number of radio sources is 387, and 27 of them are identified with 90 $`\mu `$m sources. Thus, the fraction of true radio counterpart is 27/387. Because of randomly assigned radio positions in the simulation, this fraction should be independent of $`LR`$. In Figure 2, $`N_{source}(log(LR)>0)`$ = 69 and $`N_{ran}(log(LR)>0)`$ = 28. Then, $`N_{false}(log(LR)>0)`$ is $`28\times (1.27/387)`$. Substituting these into Eqa. (5) and (8), $`R=[6928\times (1.27/387)]/69=0.62`$ and $`\stackrel{~}{R}=[6928]/69=0.59`$. Thus, $`R(LR)/\stackrel{~}{R}(LR)`$ = 1.05 for $`log(LR)>0`$, and the difference between the true and modified reliability $`R`$ and $`\stackrel{~}{R}`$ is small.
Where more than one radio sources are found within a 3$`\sigma `$ error radius, the sum of $`R`$ of individual sources frequently exceeds unity. Thus, it is necessary to normalize $`R`$ so that the sum of $`R`$ plus the probability of having no radio counterpart is unity. Suppose that there are $`M`$ radio candidates, one of which or none of which is a true radio counterpart. $`P_{id,i}`$ and $`P_{noid}`$ denote the probability that the $`ith`$ candidate is the uniquely true radio counterpart and none of radio candidates is the real radio counterpart. Then the probability of the $`ith`$ candidate to be “true” and “false” are $`\stackrel{~}{R}_i`$ and $`(1\stackrel{~}{R}_i)`$, respectively. Thus, $`P_{noid}`$ and $`P_{id,i}`$ are:
$$P_{noid}=\frac{\mathrm{\Pi }_{j=1}^M(1\stackrel{~}{R}_j)}{S},$$
(9)
and
$$P_{id,i}=\frac{\left[\stackrel{~}{R}_i\mathrm{\Pi }_{ji}^M(1\stackrel{~}{R}_j)\right]}{S},$$
(10)
where $`S`$ is a normalization factor, specific to each $`ISO`$ FIR source. Setting the sum of all the probabilities to unity gives
$$P_{noid}+\underset{i=1}{\overset{M}{}}P_{id,i}=1.$$
(11)
The normalization factor $`S`$ is derived as,
$$S=\underset{i=1}{\overset{M}{}}\stackrel{~}{R}_i\mathrm{\Pi }_{ji}^M(1\stackrel{~}{R}_j)+\mathrm{\Pi }_{j=1}^M(1\stackrel{~}{R}_j).$$
(12)
### 3.2 Identification
The main criterion adopted for a true radio counterpart identification is the condition $`P_{id}>P_{noid}`$. The results are summarized in Table 1. In the LHEX field, 27 of 49 90 $`\mu `$m sources are identified with radio sources. The identification rate is higher in bright sources than in fainter sources as shown in Fig 6. The rate is 81% (13/16) for 90 $`\mu `$m sources brighter than $`100\mathrm{m}\mathrm{J}\mathrm{y}`$ while it is reduced to 42% (14/33) for sources fainter than $`100\mathrm{m}\mathrm{J}\mathrm{y}`$. This is due to poorer positional accuracy (Kawara et al., 2004) for fainter sources. Seven of nine 170 $`\mu `$m sources in the LHEX field meet our criterion. Five of them have the same identification as their 90 $`\mu `$m counterpart. One exception is 1EX085/2EX068, for which the 90 $`\mu `$m identification is given priority for the reason described at the end of this section. While one additional source, 2EX016, does not meet our identification criterion, the identified radio source for the associated 90 $`\mu `$m source, 1EX034, is regarded as the true counterpart. In summary, a total of eight 170 $`\mu `$m sources are identified.
Because of the higher effective noise in the radio continuum image of the LHNW field, there are only a few cases where two or more radio candidates are found. To proceed with the same source identification criterion, the reliability functions derived from the LHEX field were applied to the LHNW field as well. As a result, 15 of 67 sources and 5 of 11 sources in 90 $`\mu `$m and 170 $`\mu `$m are identified with radio sources. The identification rate is 48% (11/23) for 90 $`\mu `$m sources brighter than $`100\mathrm{m}\mathrm{J}\mathrm{y}`$ and 9% (4/44) for fainter than $`100\mathrm{m}\mathrm{J}\mathrm{y}`$.
Optical identification was made by searching optical objects within a 2<sup>′′</sup> radius of the radio counterpart on the Subaru $`R`$-band image. If no objects are found, bright objects ($`R<20`$) are searched within a 5<sup>′′</sup> radius from the radio counterpart. All of the radio counterparts are identified with an optical object. The dispersion of separation between radio and optical sources is only 0$`\stackrel{}{\mathrm{.}}`$6. All of the optical counterparts are galaxies except for 1EX030, which is a point-like source showing broad emission lines characteristic of a quasar. The $`R`$-band images are shown together with 1.4 GHz contours in Figure 7.
Individual radio counterparts are listed together with the optical data in Table 2 and 3 for the 90 $`\mu `$m and 170 $`\mu `$m sources. In columns 1 & 2, names of the 90 and 170 $`\mu `$m sources are given. In column 3, a name of radio sources are given in order of appearance. In columns 4 & 5, the radio coordinates are given in the J2000 system. In columns 6 & 7, the FIR fluxes are given including the correction for the bias effect(Equ. (1) and (2)). In column 8, the 1.4 GHz flux is given in $`\mu `$Jy. In columns 9 & 10, $`R`$ and $`I`$band magnitudes are given in Vega system. Column 11 lists the angular distance of the radio counterpart from the position of the FIR source in arcsec. In columns 12-14, $`LR`$, $`P_{id}`$, and $`P_{noid}`$ are given. If more than one radio sources meet the identification criterion $`P_{id}>P_{noid}`$, then all of radio sources are listed in the tables. Table 4 summarizes the optical and radio properties of the identified $`ISO`$ FIR sources.
Source identification is complicated in several cases, and additional details are discussed below.
1EX076 has three radio candidates: two bright galaxies (REX11,REX13) and one bright radio source (REX12). REX11, REX12 and REX13 are shown at the $`R`$band image in Figure 11(a) REX11 and REX13 are a interacting galaxies pair at the same redshift of $`z=0.073`$. Moreover, REX11 is the closest to 1EX076 and its reliability is the highest among the three objects. REX11 is thus the most likely optical counterpart of this FIR source although there might also be a significant contribution from REX13.
Although 2EX016 does not have any radio candidate that meets our identification criterion, 1EX034 is identified with a radio source. This radio source is thus regarded as the counterpart of 1EX034/2EX036.
As shown in Figure 11(b), 1EX269 and 2EX047 have three common candidates (REX21, 22, & 23). REX21 is regarded as the counterpart because REX21 has the highest reliability among the three.
2EX068 has two radio candidates (REX07, REX32) meeting the criterion, while 1EX085 has only one candidate (REX07) meeting the identification criterion(Figure 11(c)). Thus, 2EX068 is identified with REX07.
1NW025($`F^c(90\mu \mathrm{m})=91`$mJy; RA=10:35:16.2, Dec.=+57:33:19 at J2000) does not have any candidates meeting the criterion while 2NW006($`F^c(170\mu \mathrm{m})=158`$mJy; RA=10:35:17.0, Dec.=+57:33:22 at J2000) is identified with a relatively bright radio source (F(1.4 GHz)=2587 $`\mu `$Jy; RA=10:35:23.30, Dec.=+57:32:49.6 at J2000). The separation between 1NW025 and 2NW006 is only 8<sup>′′</sup>. Since the identified source is too far (3.6 $`\sigma _{pos}`$) from the 1NW025 position. We regard 1NW025/2NW006 as unidentified.
## 4 Discussion
### 4.1 Redshift and FIR Luminosity
Spectroscopic redshifts of 29 out of 44 FIR sources in Table 4 are available. Twenty five optical spectra of FIR sources were obtained with the KECK II and WIYN telescopes while redshifts of four objects (1NW062, 1NW100, 1NW092 and 1NW044) were kindly provided to us by A. Barger (private communication). The 25 optical spectra are shown in Figure 12.
The redshift distributions of all FIR sources and 170 $`\mu `$m-detected FIR sources are shown in Figure 16 by open and filled histograms. Although more distant sources are expected at 170 $`\mu `$m than at 90 $`\mu `$m because of the strong k-correction brightening for dust emission, there are no significant differences in the redshift distribution between the sources detected at 170 $`\mu `$m and the sources detected only at 90 $`\mu `$m. Seven out of twelve (58%) 170 $`\mu `$m-detected source and nineteen out of twenty seven (70%) 90 $`\mu `$m-only-detected sources lie at $`z<0.3`$. This may be attributed to the small number statistics. Patris et al. (2003) reported that 95% (20/21) of 170 $`\mu `$m sources brighter than 200 mJy found in their FIRBACK southern Marano fields are at $`z<0.3`$. Our 170 $`\mu `$m sources are slightly more distant than theirs although their flux limit and their cumulative number density down to this limit are similar to ours. Their radio survey is shallower using a larger observing beam, and these and other aspects of their source identification procedure might have introduced a bias toward optically bright foreground sources.
The redshift versus IR color relation is plotted in Figure 17 for our $`ISO`$ FIR sources together with expectation from greybodies with $`\lambda ^1`$ and $`\lambda ^2`$ emissivities at $`z=010`$. Most of $`ISO`$ FIR sources with 170 $`\mu `$m detection have a dust temperature ranging from 20-30 K for $`\lambda ^1`$ and 15-25 K for $`\lambda ^2`$, which is consistent with the 170 $`\mu `$m/90 $`\mu `$m color temperature distribution of 74 ELAIS sources reported by Héraudeau et al. (2004).
The FIR flux, $`F_{FIR}(40\mu \mathrm{m}500\mu \mathrm{m})`$, can be estimated from the gray body fitting with the observed $`F^C(90\mu m)`$ and $`F^C(170\mu m)`$, the temperature from Figure 17 and the assumption of $`\lambda ^2`$ emissivity. Adopting the different dust emissivity, $`\lambda ^1`$, the FIR flux, $`F_{FIR}`$, increase 15 percents on average. This assumption will not change the main conclusion of this paper. It should be noted that the detection limits (43 mJy at 90 $`\mu `$m and 102 mJy at 170 $`\mu `$m after the correction for the flux bias) are substituted into the undetected band for the objects detected only in one band.
The FIR luminosity, $`L_{FIR}`$, is then obtained as,
$$L_{FIR}=4\pi D_L^2\times F_{FIR},$$
(13)
where $`D_L`$ is the luminosity distance. The resultant FIR luminosity is given in Table 4 and plotted in Figure 18 as a function of redshift. Our sample consists of 24 sources with $`L_{FIR}<10^{12}\mathrm{L}_{}`$, four with $`L_{FIR}=10^{1213}\mathrm{L}_{}`$, and one with $`L_{FIR}>10^{13}\mathrm{L}_{}`$. In the $`IRAS`$ bright source catalog, only six out of 324 sources have $`L_{FIR}>10^{12}\mathrm{L}_{}`$ (Soifer et al., 1987). Thus, a fraction of $`L_{FIR}>10^{12}\mathrm{L}_{}`$ sources are 10 times greater in our $`ISO`$ sample than in the IRAS bright source catalog. All the $`IRAS`$ bright sources lie within $`z0.1`$, while, only 10 of 29 sources are at $`z0.1`$ in our $`ISO`$ sample. It is noted here that our spectroscopy project is not complete yet, and all the sources with $`L(90\mu \mathrm{m})/L(R)>50`$ are left unobserved. It is generally agreed that sources with $`L(90\mu \mathrm{m})/L(R)>50`$ belong to a population of ULIRGs with $`L_{FIR}>10^{12}\mathrm{L}_{}`$. Hence, our sample should contain a greater fraction of $`L_{FIR}>10^{12}\mathrm{L}_{}`$ sources than that derived from our current spectroscopic knowledge.
### 4.2 Luminosity functions
Here we derive luminosity functions of our $`ISO`$ FIR sources having $`F^C(90\mu \mathrm{m})85`$ mJy. These new criterion is more strict than those used to identify sources having $`F^C(90\mu \mathrm{m})43`$ mJy or $`F^C(170\mu \mathrm{m})102`$ mJy. The reason for using the new criteria is to control the detection limit of our $`ISO`$ FIR sources simply and to avoid from the large correction for the completeness of the sample. 27 out of 44 galaxies identified with $`ISO`$ FIR sources meet the new criterion, and their redshifts of 21 sources are available.
The luminosity function, $`d\mathrm{\Phi }(L_{FIR})/dL_{FIR}`$, (i.e., the volume density of galaxies per unit luminosity range) is derived by following the the $`1/V_{max}`$ method as described in Schmidt (1968) and Eales (1993). The volume density of galaxies with luminosity between $`L`$ and $`L+dL`$ is defined as,
$$\frac{d\mathrm{\Phi }(L_{FIR})}{dL_{FIR}}dL_{FIR}=\underset{j}{}\frac{1}{p(F^C)V_j},$$
(14)
where $`p(F^C)`$ is the detection probability for source with corrected flux $`F^C`$, and the summation is over all sources with luminosity between $`L`$ and $`L+dL`$ in the sample. $`p(F^C)`$ is obtained by combining the detection rate given in Figure 3 by Kawara et al. (2004) with Equ. (1) for transforming the observed flux $`F`$ to the corrected $`F^C`$. $`V_j`$ is defined as,
$$V_j=_\mathrm{\Omega }_{z_{min}}^{z_{max}}\frac{d^2V}{d\mathrm{\Omega }dz}𝑑z𝑑\mathrm{\Omega }.$$
(15)
where $`\mathrm{\Omega }`$ is the solid angle in this survey and $`d^2V/d\mathrm{\Omega }dz`$ is the comoving volume element. $`z_{max}`$ is the maximum redshift defined by the limiting fluxes at the faintest end, namely, $`F^C(90\mu \mathrm{m})=85`$ mJy, while $`z_{min}`$ is the minimum redshift by the flux limits at the brightest end which is set to 1 Jy for 90 $`\mu `$m. The luminosity functions are derived for three redshift bins; $`z=0.030.10`$, $`z=0.100.30`$ and $`z=0.300.60`$. These redshift bins correspond to the different FIR luminosities, $`L_{FIR}=0.77.4\times 10^{10}\mathrm{L}_{\mathrm{}}`$, $`L_{FIR}=5.428\times 10^{10}\mathrm{L}_{\mathrm{}}`$, and $`L_{FIR}=51240\times 10^{10}\mathrm{L}_{\mathrm{}}`$, respectively. This reason is that the luminosity of the $`ISO`$ FIR sources is following as a function of redshift (See Fig. 18). The numbers of objects in each bin are 6, 7 and 5 and three sources which have $`F^C(90\mu \mathrm{m})85\mathrm{m}\mathrm{J}\mathrm{y}`$ are out of these three bins: two of them are at $`z<0.03`$ and the other is at $`z=1.1`$. The luminosity function requires the following additional conditions; $`z_{max}`$ = $`z_u`$ for $`z_{max}>z_u`$ and $`z_{min}`$ = $`z_l`$ for $`z_{min}<z_l`$ where $`z_u`$ and $`z_l`$ are the maximum and minimum redshifts of the specific redshift bin.
The luminosity function of our $`ISO`$ FIR sources sample is given in Table 5. Figure 19 compares the $`ISO`$ FIR sources sample with other galaxy samples. These result is calculated by summing up $`1/p(F^C)V_j`$ of the sources having the redshift(Equ. (14)). Our spectroscopic redshifts are obtained for 78%(21/27) of this sample and no correction by a factor of 1.3 (27/21) was not applied in Table 5 and Figure 19. The six $`ISO`$ FIR sources without redshift are not expected to be the same populations as our 21 $`ISO`$ FIR sources having redshift, because the six sources without redshift tend to have higher $`L(90\mu \mathrm{m})/L(R)`$ than those in redshift-measured source; sources without redshift have the range $`30<L(90\mu \mathrm{m})/L(R)<1400`$, while the 20 redshift-measured sources have $`1<L(90\mu \mathrm{m})/L(R)<20`$ and one source at z$`=0.469`$ have $`L(90\mu \mathrm{m})/L(R)=43.7`$.
The comparison includes the $`IRAS`$ bright galaxy sample by Soifer et al. (1987) with the mean redshift of $`<z>0.04`$, the $`IRAS`$ 1Jy ULIRG sample by Kim & Sanders (1998) with $`<z>0.15`$, and the SCUBA galaxy sample by Barger, Cowie & Richards (2000) at $`z=13`$. Here we assume that the infrared luminosity, $`L_{IR}(81000\mu \mathrm{m})`$, is nearly equal with the FIR luminosity, $`Ł_{FIR}(40500\mu \mathrm{m})`$. The comparison shows a clear trend of the evolutionary effect; at a given luminosity, a greater density of galaxies for a higher redshift. It is particularly clear that there are a rapid evolution in the ULIRG population toward high redshift; the space densities are $`1\times 10^7\mathrm{Mpc}^3`$ at $`<z>0.04`$ (Soifer et al., 1987), $`5\times 10^7\mathrm{Mpc}^3`$ at $`<z>0.15`$ (Kim & Sanders, 1998), and $`4.6\times 10^5\mathrm{Mpc}^3`$ at $`z=0.30.6`$. In other words, relative to the local universe, the space densities are $``$ 5 times greater at $`<z>0.15`$ and $``$ 460 times greater at $`z=0.30.6`$. At the highest end of the FIR luminosity, the space densities are 1000 times greater at $`z=13`$ than the local Universe.
There is uncertainty of flux calibration of our $`ISO`$ FIR sources. Our flux calibration was done with one IRAS source(UGC 06009) which have $`50`$ percent flux errors(Kawara et al., 2004). This uncertainty brought systematic luminosity shift to 0.3 dex in Figure 19. However even if there were the 0.3 dex shift to lower luminosity, the evolution of $`ISO`$ FIR sources, especially at $`z=0.30.6`$ still exists. In addition, we compared the 90 $`\mu `$m luminosity function of the ELAIS(Serjeant et al., 2004). Serjeant et al. (2004) presents that the $`ISO`$ 90 $`\mu `$m luminosity function of the ELAIS is consistent with the IRAS with the assumption of pure luminosity evolution of $`(1+z)^3`$. If we apply their pure luminosity evolution of $`(1+z)^3`$ in our data, at least our luminosity function at $`z=0.030.1`$ and $`0.10.3`$ are consistent with that of ELAIS and IRAS Bright galaxy sample. On the other hand, the luminosity function at $`z=0.30.6`$ show an luminosity excess by a factor of $``$2 of that of IRAS Bright galaxy sample after luminosities reduced by a factor of $`(1+z)^3`$. This suggests much stronger evolution to the $`ISO`$ FIR sources at this redshift range where ELAIS can barely observe. It is noted again that our spectroscopic observations miss almost all sources with $`L(90\mu \mathrm{m})/L(R)>30`$, many of which would belong to a population of ULIRGs with $`L_{FIR}>10^{12}\mathrm{L}_{}`$. The space density of $`ISO`$ FIR sources with $`L_{FIR}>10^{12}`$ ,which is derived here, must be significantly underestimated.
### 4.3 Nature of $`ISO`$ FIR sources
The fraction of $`ISO`$ FIR sources associated with an AGN is estimated from optical emission lines, radio continuum emission, and X-ray activity. The excitation diagnosis by Veilleux & Osterbrock (1987) is applied to the spectra of the $`ISO`$ FIR sources for the optical spectral classification. One object (1EX030) is a quasar with $`L_{FIR}=10^{13.1}L_{\mathrm{}}`$ at $`z=1.11`$. Two objects (1EX025 & 1NW133) are type II Seyferts or LINERs with $`L_{FIR}=10^{10.810.9}L_{\mathrm{}}`$ at $`z=0.110.16`$. Two objects (1EX026 & 1EX076) are HII galaxies with $`L_{FIR}=10^{9.410.4}L_{\mathrm{}}`$ at $`z=0.030.07`$. The remaining galaxies are left unclassified because of the insufficient wavelength coverage, spectral resolution, or quality of the spectrum. Optical images indicate that four galaxies and possibly another two are interacting systems. Their optical properties are summarized in Table 4.
The relationship between 1.4 GHz and FIR emission is examined on the $`L(1.4GHz)/L(90\mu \mathrm{m})`$ versus $`F^C(170\mu \mathrm{m})/F^C(90\mu \mathrm{m})`$ diagram as shown in Figure 20<sup>5</sup><sup>5</sup>5$`L(1.4GHz)`$, $`L(90\mu \mathrm{m})`$ and $`L(R)`$ are monochromatic luminosities and are defined by $`4\pi D_L^2\nu f_\nu `$ where $`D_L`$ denotes the luminosity distance to the object.. Open and filled squares represent sources lying in the LHEX and LHNW fields, respectively. The dashed and dotted lines are the redshift loci of two different types of ULIRGs, NGC 6240 (with an AGN) and Arp 220 (without an AGN) (e.g. Smith et al., 1998), while the dash dotted line shows that of starburst galaxy M82. This figure indicates that most of the $`ISO`$ FIR sources are pure star forming or star-formation dominated galaxies as they lie near the loci of Arp 220 and M82. Three optically classified AGNs, 1EX030, 1EX025 & 1NW133, also appear in the same area occupied by star-formation dominated objects. Three FIR sources, 1EX100, 1EX047/2EX036, and 1NW272/2NW009, are found near the locus of NGC 6240. 1NW272/2NW009 is classified as type II AGN with $`L_{FIR}=10^{12.2}L_{\mathrm{}}`$ at $`z=0.47`$. No spectroscopic data are available for the two other sources. The source 2EX015, which has the highest $`L(1.4GHz)/L(90\mu \mathrm{m})`$, is identified with a powerful radio galaxy at $`z=0.710`$.
As discussed so far, our sample contains at least seven galaxies hosting an AGN and five ULIRGs. Three AGN-host galaxies have FIR luminosity characteristic of ULIRGs. Thus, 60% (3/5) of our ULIRGs are AGN galaxies, in agreement with that found in the IRAS sample (Sanders & Mirabel, 1996), and this suggests that the fraction of AGN galaxies with the ULIRG luminosity does not change much from the local Universe to $`z0.5`$. The deep X-ray survey have been conducted in the LHEX field by ROSAT and XMM-Newton satellites (e.g. Lehmann et al., 2001; Mainieri et al., 2002). If our seven AGNs have the X-ray to FIR luminosity ratio $`L_X/L_{FIR}`$ similar to those of AGN-associated ULIRGs such as NGC 6240 and Mrk 231, $`2\times 10^{15}`$ erg cm<sup>-2</sup> s<sup>-1</sup> and $`6\times 10^{15}`$ erg cm<sup>-2</sup> s<sup>-1</sup> are expected at the 0.5-2 keV and 2-10 keV energy bands, respectively, for objects having 100 mJy at 100 $`\mu `$m. Out of our seven AGN galaxies, only one source, 1EX030, identified as a quasar, has been detected in the X-ray. Two undetected sources lie within the ROSAT deep survey area with a flux limit of $`10\times 10^{15}`$ erg cm<sup>-2</sup> s<sup>-1</sup>, and it is not surprising that these galaxies were not detected by ROSAT. The powerful radio galaxy 2EX015 at $`z=0.710`$ and another AGN candidate 1EX110 (which has no optical spectrum yet) are found within the XMM-Newton survey area with a limiting flux of $`1.4\times 10^{15}`$ erg cm<sup>-2</sup> s <sup>-1</sup> in the 2-10 keV energy bands. The X-ray non-detection of these two objects may be attributed to a Compton thick absorber with a column density of $`3\times 10^{24}`$ cm<sup>-2</sup>, which is equivalent to $`A_V1200`$ mag. The two remaining AGN sources are outside the X-ray survey area. It is interesting that the fraction of AGN host galaxies detected by the deep X-ray surveys is rather small. This is also a nice demonstration that the combination of FIR and radio observations offers a powerful tool to find heavily obscured AGNs which might be difficult to be found in the X-ray.
The 90 $`\mu `$m luminosity relative to the optical R-band luminosity, $`L(90\mu \mathrm{m})/L(R)`$, is plotted against optical color , $`RI`$, in Figure 21. Two large symbols (square and diamond) are FIRBACK 170 $`\mu `$m sources, FN1-40 at $`z=0.449`$ and FN1-64 at $`z=0.907`$ (Chapman et al., 2002). The loci expected from the SEDs of NGC 6240, Arp 220, and M82 at $`z=02`$ are shown using dashed and dotted lines. This figure demonstrates that the $`ISO`$ succeeded in detecting objects with a wide range of $`L(90\mu \mathrm{m})/L(R)`$ ratio, from 1 to 1000, and show a bimodal distribution of $`L(90\mu \mathrm{m})/L(R)`$ for our $`ISO`$ FIR sources. One peak with $`1<L(90\mu \mathrm{m})/L(R)<3`$ appears to represent the population of normal star forming galaxies in nearby universe because of their brightness in the optical. The other peak with FIR excess represents infrared dominant sources like Arp 220 whose luminosity is nearly entirely reprocessed by dust. In Rodighiero et al. (2005) which reduced same ISO data in the LHEX and made identification with radio and 15 $`\mu `$m sources(See Appendix for details), nine of 11 sources were fitted with the SEDs of M82 or M51 which are FIR moderate galaxies, while two others have a Arp 220 SED. The number of Arp 220-like sources seem to be small, however their SED-template fitting was done for the sources whose redshifts are available. Thus there might be a bias to optically bright sources.
Figure 21 also shows that there is possibly a third class of objects with an extreme excess of FIR luminosity, $`L(90\mu \mathrm{m})/L(R)>500`$. Although ULIRGs are known to have the largest intrinsic excess of the FIR luminosity relative to the optical luminosity, their extreme optical and FIR colors cannot be adequately explained by simply redshifting the observed SEDs of ULIRGs or starburst galaxies. Therefore, the five such objects, one from FIRBACK and four from our survey, may represent a new population of extreme FIR-excess galaxies. If the extreme values of $`L(90\mu \mathrm{m})/L(R)`$ are intrinsic, the small optical luminosity may imply that their stellar system has not been fully developed or their entire stellar structures are heavily obscured by dust.
Four such objects in the Lockman Hole fields are located on the outside of the field where the submillimeter and millimeter observations with the SCUBA, Bolocam and MAMBO(Scott et al., 2002; Greve et al., 2004; Laurent et al., 2005) were performed. In future, the measurements in such wavelength will give the additional information to them. In addition, one of the Spitzer Space Telescope Legacy Programs, SWIRE(Lonsdale et al., 2003), cover these fields in the wavelength from 3.6 $`\mu `$m to 160 $`\mu `$m and will release their data and help to understand four such objects.
## 5 Summary
By exploiting the FIR-radio correlation, we have performed the Likelihood-Ratio analysis to identify the far-infrared sources that have been found in an area of $`0.9`$ deg<sup>2</sup> during the $`ISO`$ deep far-infrared survey in the Lockman Hole. New observations have been conducted to construct the catalogs of radio and optical objects, which include a deep VLA 1.4 GHz observations, optical $`R`$ & $`I`$band imaging on the Subaru 8m and UH2.2m telescopes, and optical spectroscopy on the Keck II 10m and WIYN 3.5m telescopes. A summary of the results presented in this paper is as follows:
* Our samples of 116 and 20 sources are selected with the criteria of $`F^C(90\mu \mathrm{m})43`$ mJy and $`F^C(170\mu \mathrm{m})102`$ mJy, respectively. Our 1.4 GHz radio sample includes a total of 463 sources.
* In order to remove positional coincidence by chance, we calculate the Likelihood-Ratio and their reliability. As a results, 44 FIR sources are identified with radio sources.
* Optical identification of the 44 FIR/radio association is conducted using accurate radio positions. The dispersion in the difference between the radio and optical position is 0$`\stackrel{}{\mathrm{.}}`$6.
* Dust temperature derived from the FIR color ranges between 15 and 30 K.
* Spectroscopic redshifts have been obtained for 29 out of 44 identified sources. There are no significant differences in the redshift between 170 $`\mu `$m-detected sources and 90 $`\mu `$m-only-detected sources.
* 24 (out of 29) FIR galaxies with redshifts have $`L_{FIR}<10^{12}\mathrm{L}_{}`$, four with $`L_{FIR}=10^{1213}\mathrm{L}_{}`$, and one $`L_{FIR}>10^{13}\mathrm{L}_{}`$.
* The luminosity functions are calculated using the $`1/V_{max}`$ method. The space density of the our sample galaxies at $`z=0.30.6`$ is $`4.6\times 10^5`$ Mpc<sup>-3</sup>, which is 460 times as high as that at the local universe. A rapid evolution in the ULIRG population is suggested.
* Most of $`ISO`$ FIR sources have $`L(1.4GHz)/L(90\mu \mathrm{m})`$ similar to star-forming galaxies Arp 220 and M82, indicating star formation is the dominant source of their FIR and radio emission in these galaxies.
* Our FIR sample contains at least seven AGNs, which are classified either from optical emission lines, excess in radio emission, or X-ray activity.
* 60%(3/5) of our ULIRGs are AGN galaxies, implying that the percentage of AGN galaxies with ULIRG luminosity does not change significantly between $`z=0`$ and $`z0.6`$.
* Five of the seven AGN galaxies are within the ROSAT X-ray survey field, and two are within the XMM-Newton survey fields. X-ray emission has been detected in only one source, 1EX030, which is optically classified quasar. If our AGN galaxies have $`L_X/L_{FIR}`$ similar to NGC 6240, a ULIRG hosting an AGN, then none of our AGN galaxies would have been detected by ROSAT. The non-detection by XMM-Newton 2-10 keV band implies a very thick absorption column density of $`3\times 10^{24}\mathrm{cm}^2`$ or $`A_V1200`$ mag obscuring the central source of the two AGN galaxies. The combination of FIR and radio observations would provide a powerful tool to find heavily obscured AGNs which might be difficult to be found in the X-ray.
* Several sources show an extreme FIR luminosity relative to the optical $`R`$-band, $`L(90\mu \mathrm{m})/L(R)>500`$. Such extreme values cannot be explained from the redshifted SEDs of ULIRGs and may imply a new population of galaxies where an extreme activity of star formation in an undeveloped stellar system. If so, we might be looking at the formation of bulges or ellipticals.
We wish to thank the staff of the Subaru Observatory, NOAO, NRAO, Keck Observatory and the UH88 telescope for their assistance and hospitality during the several observing runs in which collected data for this paper. This research made use of the NASA/IPAC Extragalactic Database (NED) which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration. This paper is based on observations with $`ISO`$, an ESA project with instruments funded by ESA member states and with the participation of ISAS and NASA.
## Appendix A Comparison with Rodighiero et al.
Rodighiero et al. (2003) and Rodighiero & Franceschini (2004) have reduced our 90 $`\mu `$m data in the LHEX and LHNW fields, by using their own method, a parametric algorithm that fits the signal time history of each detector pixel. They then identified the singularities induced by cosmic ray impacts and by transient effects in the detectors and extract real sky sources. The numbers of their sources with signal-to-noise ratio S/N $`>`$ 3 are 36 and 28 in the LHEX and LHNW, respectively, while our catalogs consist of 116 and 107 sources having S/N $`>`$ 3(Kawara et al., 2004). Both catalogs are made with the different method. Thus the reason of this difference is expected to be in the reduction and source extracting methods, but is unclear.
In Table 6, their 15 LHEX and 11 LHNW sources have a counterpart within 30<sup>′′</sup> in our catalogs. Figure 22 compares the Rodighiero et al. flux with ours. One IRAS source, UGC 06009, is reported as ex003(577 $`\pm `$ 110 mJy) in Rodighiero et al. (2003), while we used this source for the flux calibration of the IRAS 100 $`\mu `$m measurement(1218 mJy). The difference of a factor of 2.1 between ours and Rodighiero et al. might be expected and the mean ratio of ours with those in Rodighiero et al. (2003) and Rodighiero & Franceschini (2004) is 1.8. As shown in Figure 22, there are some sources in our catalog having $``$100 mJy which show much fainter fluxes in theirs, and these sources could make the deviation of the flux ratio bigger.
Rodighiero et al. (2005) have done the identification work of their 36 90 $`\mu `$m sources(Rodighiero et al., 2003) in the LHEX field using the association with the shallow radio(de Ruiter et al., 1997) and 15 $`\mu `$m sources (Fadda et al., 2004; Rodighiero et al., 2004). 17 of them are identified with radio and 15 $`\mu `$m sources, all of which are optically identified. 14 of the 17 optically identified sources are found in our catalog(Kawara et al., 2004) and we made same optical identification in this paper. However only one(ex002/1EX100) of their 19 optically unidentified sources is found in Kawara et al. (2004) and is succeeded to identify in this work. It is unclear why most of their unidentified sources are not found in Kawara et al. (2004). |
warning/0507/cond-mat0507431.html | ar5iv | text | # Energy-resolved inelastic electron scattering off a magnetic impurity
## I Introduction
Scattering of an electron off a magnetic impurity embedded in a conductor is known to be anomalously strong Abrikosov-textbook . The origin of the anomaly is rooted in the degeneracy of the localized spin states. This degeneracy, being removed by a weak exchange interaction with the itinerant electrons in a metal, gives rise to the strong scattering of electrons with low energy – the Kondo effect. Perturbation theory in the exchange interaction constant $`J`$ is singular. The second-order contribution in $`J`$ to the scattering amplitude diverges logarithmically Kondo if the electron energy $`E`$ (measured from the Fermi level) and temperature $`T`$ are approaching zero. It is important to notice that the logarithmically divergent contribution to the amplitude corresponds to an elastic process. Indeed, this contribution comes from the change of state of one electron; states of all other itinerant electrons are the same in the beginning and end of the scattering process. Therefore, the energies of the electron before and after the scattering is unchanged.
The divergence noticed by Kondo is not unique to the second order of the perturbation theory. Its higher orders ($`n>2`$) also contain divergent terms of the type $`J^n\mathrm{ln}^{n1}(D/\epsilon )`$, where $`\epsilon =\mathrm{max}(E,T)`$, and $`D`$ is some ultraviolet energy cut-off, whose value depends on the specific model; $`D\epsilon `$. These leading logarithmic terms may be summed up by diagrammatic method Abrikosov1965 or by means of the “Poor man’s scaling” Anderson renormalization group (RG), yielding for the scattering amplitude
$$A_{k,\sigma ,Sk^{},\sigma ^{},S^{}}=\frac{1}{\mathrm{ln}(\epsilon /T_K)}𝐬_{\sigma ,\sigma ^{}}𝐒_{s,s^{}}.$$
(1)
where $`𝐬_{\sigma ,\sigma ^{}}`$ and $`𝐒_{s,s^{}}`$ are the spin operators of the conduction electrons and the impurity, respectively. The so-called Kondo temperature is given in terms of the cut-off, $`D`$, and the exchange interaction, $`J`$, as $`T_K=De^{1/(J\nu )}`$, where $`\nu `$ is the density of states. Like the lowest-order perturbation theory result, the leading-logarithmic approximation Eq. (1) corresponds to purely elastic electron scattering.
The leading-logarithmic approximation is adequate at $`\epsilon T_K`$, but it fails at low temperatures. A convenient phenomenological description of the low-energy behavior of a single-channel Kondo model is given by Nozières’ effective Fermi liquid theory. In this theory, a scattering problem can be formulated, too. It is clearly seen Nozieres , however, that the scattering is not purely elastic at $`\epsilon T_K`$. At $`T=0`$, for example, the inelastic contribution to the electron scattering cross-section scales as $`(E/T_K)^2`$, and becomes comparable to the elastic part at $`ET_K`$.
The Kondo effect is a crossover phenomenon, rather than a phase transition. The measurable characteristics, such as the contribution to the susceptibility or resistivity due to magnetic impurities depend smoothly on temperature. Similarly, the electron scattering off a magnetic impurity, which is deeply inelastic at $`\epsilon T_K`$, must have some inelastic component at any energy $`E`$. In this paper we investigate in detail the inelastic scattering of a high-energy electron off a magnetic impurity.
A study of the energy-resolved, differential cross-section, $`\sigma (E,\omega )`$, is interesting in its own right, but it can, in principle, also be measured, e.g., in a modification of the experiments of Pothier et al.Pothier . Further motivation to study $`\sigma (E,\omega )`$ beyond perturbation theory comes from the recent theoretical work of Zaránd et al. Natan . In Ref. Natan, the energy dependence of the total scattering cross-section, $`\sigma _{\mathrm{tot}}(E)=𝑑\omega \sigma (E,\omega )`$, was addressed. With the help of the optical theorem, the total cross-section $`\sigma _{\mathrm{tot}}(E)`$ was compared with the elastic part of it. The conclusion reached in Ref. Natan, regarding the energy domain $`ET_K`$ is striking: at $`T=0`$ the scattering is deeply inelastic; the elastic part turns out to be negligibly small. This seemingly contradicts the leading-logarithmic result for the scattering amplitude Eq. (1). The physical explanation of this phenomenon however remained unclear in Ref. Natan, and motivates us to revisit the problem of inelastic scattering. The dependence of the differential cross-section $`\sigma (E,\omega )`$ on $`\omega `$, which we consider in this paper, clarifies the issue, as we are able to determine the distribution of energy losses in the inelastic electron scattering off a magnetic impurity.
The separation of the electron scattering cross-section in the Kondo effect into elastic and inelastic parts at $`ET_K`$ was not addressed for decades, as it does not affect the routinely measured quantity, the resistivity. The anomalously fast electron energy relaxation in some mesoscopic metallic wires Pothier , which was discovered in the last decade, prompted a search for relaxation mechanisms driven by impurities with internal degrees of freedom. A viable mechanism of energy relaxation was suggested first in Ref. Kaminski, , and was associated with the electron-electron scattering mediated by exchange interaction of electrons with magnetic impurities. The removal of degeneracy of the localized spin states by the exchange interaction results in an anomaly of the electron-electron scattering cross-section at small energy transfers Kaminski ; the collision of two electrons with energies $`E,E^{}T_K`$ leads to a re-distribution of the energies between the two particles, $`E,E^{}E\omega ,E^{}+\omega `$, and has cross-section $`K(\omega ,E,E^{})J^4/\omega ^2`$ in the lowest-order perturbation theory zavsol . The $`1/\omega ^2`$ dependence of $`K`$ allowed the experimental observations Pothier to be explained qualitatively. Later experiments Anthore performed in a magnetic field sufficient for the Zeeman splitting of impurity energy levels did confirm the origin Kaminski of the inelastic electron-electron scattering, and indicated the irrelevance of more exotic mechanisms, which assumed a generic non-Fermi liquid behavior introduced by impurities Kroha .
The existence of energy exchange between electrons mediated by their interaction with a magnetic impurity indicates the inelastic nature of the electron scattering off a magnetic impurity. Indeed, using the Fermi Golden rule we find
$`\sigma (E,\omega ){\displaystyle \frac{J^4}{\omega ^2}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑E^{}f(E^{})(1f(E^{}+\omega )){\displaystyle \frac{J^4}{\omega }}`$ (2)
at $`\omega T`$. So already in the simplest perturbation theory it becomes clear that there is an inelastic contribution to scattering. As long as $`E,\omega T`$, temperature does not affect the inelastic cross-section in this order. It is not clear, however, what the relation is between the inelastic cross-section $`\sigma (E,\omega )`$ and the leading-logarithmic result Eq. (1): On one hand, $`\sigma (E,\omega )J^4`$ is parametrically smaller than the scattering cross-section following from Eq. (1). On the other hand, the total inelastic cross-section obtained from Eq. (2), $`\sigma _{\mathrm{tot}}(E)=𝑑\omega \sigma (E,\omega )`$, diverges at $`\omega 0`$ indicating the inapplicability of the lowest-order perturbation results at small energy transfers.
The lowest-order perturbation theory for $`K(\omega ,E,E^{})`$ can be controllably improved in two respects. First, at $`E,E^{}T_K`$ and $`|\omega |E,E^{}`$ the four constants $`J`$ entering as a product in the perturbative result, may be replaced Kaminski by the properly renormalized Anderson quantities kroha1 . Second, the divergence at $`\omega 0`$ is cut off due to the dynamics of localized spin. An adequate theory may be developed for high temperatures, $`TT_K`$, where the cut-off occurs due to the Korringa relaxation Kaminski . These improvements allow one to see that $`\sigma _{\mathrm{tot}}(E)`$ is finite, but are insufficient to investigate the details of the $`\sigma (E,\omega )`$ dependence.
In this paper we concentrate on the differential cross-section, $`\sigma (E,\omega )`$, of inelastic scattering of a highly excited electron with energy $`ET_K`$. Despite the many-body nature of the Kondo effect, this quantity is well-defined at $`\omega E`$. We show in Section II that in the limit $`ET_K`$ the differential cross-section is related to the dissipative part of the impurity spin susceptibility, $`\chi ^{\prime \prime }`$. From the low- and high-frequency asymptotes of $`\chi ^{\prime \prime }`$ we extract in Sections III and IV the behavior of the differential cross-section $`\sigma (E,\omega )`$ in the absence and presence of a Zeeman energy, respectively. The analytical asymptotes thus obtained are complemented by results of the numerical renormalization groupWilson (NRG), which allows us to access also the intermediate range of frequencies and magnetic fields. The connection between the result of Ref. Natan, and the leading-logarithmic approximation for the scattering amplitude (1) describing only elastic scattering will be explained in detail. Finally, in Section V we discuss possible hot-electron experiments in metallic mesoscopic wires and in a semiconductor quantum-dot setup in order to measure the differential scattering cross-section $`\sigma (E,\omega )`$.
## II Relation between inelastic scattering cross-section and susceptibility
The relation between the scattering cross-section of a “foreign” spin-carrying particle and the spin-spin correlation function of a magnetic medium is well-known from the theory of neutron scattering MagnetismReference . Here we derive a similar relation for scattering off a magnetic impurity of a high-energy electron belonging itself to the Fermi liquid hosting the magnetic impurity.
The exchange interaction between the impurity spin and spins of electrons forming the Fermi liquid,
$$H_{\mathrm{int}}=J\underset{𝐤,𝐤^{}}{}𝐒𝐬_{\alpha \alpha ^{}}c_{𝐤\alpha }^{}c_{𝐤^{}\alpha ^{}}^{}$$
(3)
gives rise to the Kondo effect. Here $`J`$ is the constant of exchange interaction between the impurity spin and itinerant electrons with energies $`ϵ_{𝐤\sigma }`$ (measured from the Fermi level) confined to some energy band, $`|ϵ_{𝐤\sigma }|<D`$. Here $`𝐬_{\alpha \alpha ^{}}`$ is $`\frac{1}{2}`$ times the vector of Pauli matrices. The Kondo problem allows for a logarithmic renormalization: the low-energy properties of the system described by the Hamiltonian (3) coincide with those for a Hamiltonian defined in a narrower band, say $`|ϵ_{𝐤\sigma }|<E`$, upon the proper renormalization of the exchange constant,
$$J(E)=\frac{J(D)}{1\nu J(D)\mathrm{ln}(D/E)},J(D)=J,$$
(4)
where $`\nu `$ is the density of states of itinerant electrons. The perturbative renormalization Eq. (4) is valid as long as the running energy \[$`E`$ in the case of Eq. (4)\] significantly exceeds the Kondo temperature $`T_K`$. An important property of the logarithmic renormalization is that only exponentially wide energy intervals $`(\epsilon _1,\epsilon _2)`$, such that $`\nu J|\mathrm{ln}(\epsilon _1/\epsilon _2)|1`$ contribute significantly to the renormalization. That allows us to “skip” some relatively narrow strip of energies, say, $`(E\mathrm{\Delta }E,E+\mathrm{\Delta }E)`$, with $`\mathrm{\Delta }EE`$, in the renormalization process, yielding a Hamiltonian
$`H_{\mathrm{int}}`$ $`=J(\stackrel{~}{D}){\displaystyle \underset{|ϵ_{𝐤\alpha }|,|ϵ_{𝐤^{}\alpha ^{}}|<\stackrel{~}{D}}{}}𝐒𝐬_{\alpha \alpha ^{}}c_{𝐤\alpha }^{}c_{𝐤^{}\alpha ^{}}^{}`$
$`+J(E){\displaystyle \underset{E\mathrm{\Delta }E<ϵ_{𝐤\alpha },ϵ_{𝐤^{}\alpha ^{}}<E+\mathrm{\Delta }E}{}}𝐒𝐬_{\alpha \alpha ^{}}c_{𝐤\alpha }^{}c_{𝐤^{}\alpha ^{}}^{}`$ (5)
$`+J(E){\displaystyle \underset{\begin{array}{c}E\mathrm{\Delta }E<ϵ_{𝐤\alpha }<E+\mathrm{\Delta }E,|ϵ_{𝐤^{}\alpha ^{}}|<\stackrel{~}{D};\\ E\mathrm{\Delta }E<ϵ_{𝐤^{}\alpha ^{}}<E+\mathrm{\Delta }E,|ϵ_{𝐤\alpha }|<\stackrel{~}{D}\end{array}}{}}𝐒𝐬_{\alpha \alpha ^{}}c_{𝐤\alpha }^{}c_{𝐤^{}\alpha ^{}}^{},`$ (8)
with $`\stackrel{~}{D}E\mathrm{\Delta }E`$. The renormalized exchange constants here may be expressed in terms of the Kondo temperature, $`\nu J(\epsilon )=1/\mathrm{ln}(\epsilon /T_K)`$. There is no need to distinguish between $`J(E\mathrm{\Delta }E)`$, $`J(E)`$ or $`J(E+\mathrm{\Delta }E)`$ as long as $`ET_K`$.
If the scattering of an electron with initial energy $`E`$ leaves it in the energy domain $`(E\mathrm{\Delta }E,E+\mathrm{\Delta }E)`$, then the corresponding cross-section, within the lowest-order perturbation theory in $`J(E)`$, can be evaluated with the help of the Hamiltonian (II). The first line of Eq. (II) plays the role of the Hamiltonian of a magnetic medium in the neutron scattering problem, and the second line describes the interaction of the energetic particle (we deal with an electron rather than with a neutron though) with the medium. The remaining part of the Hamiltonian does not contribute to the scattering cross-section in the lowest-order calculation.
Consider such a scattering of an energetic electron with energy $`E`$ and spin $`\sigma `$ in the initial and $`E\omega `$ and $`\sigma ^{}`$, respectively, in the final state with $`\omega E`$ such that $`E\omega (E\mathrm{\Delta }E,E+\mathrm{\Delta }E)`$. The state of the remaining system before and after scattering may be characterized by the wave functions $`\mathrm{\Psi }_i`$ and $`\mathrm{\Psi }_f`$, respectively. The initial and final state of the total system is then given by the product states
$`\begin{array}{cc}\text{initial state:}\hfill & |i=|E,\sigma |\mathrm{\Psi }_i\hfill \\ \text{final state:}\hfill & |f=|E\omega ,\sigma ^{}|\mathrm{\Psi }_f.\hfill \end{array}`$ (11)
The differential cross-section of inelastic scattering $`\sigma _{\sigma ^{}\sigma }(E,\omega )`$ is determined by the probability $`P_{\sigma ^{}\sigma }(E,\omega )d\omega `$ of scattering of an electron with initial energy $`E`$ and spin $`\sigma `$ into a state within interval of energies $`(E\omega ,E\omega d\omega )`$ and spin $`\sigma ^{}`$,
$`P_{\sigma ^{}\sigma }(E,\omega )d\omega =v_F\sigma _{\sigma ^{}\sigma }(E,\omega )d\omega ,`$ (12)
where $`v_F`$ denotes the Fermi velocity. By energy conservation, $`\omega =\xi _f\xi _i`$, where energies $`\xi _i`$, $`\xi _f`$ are associated respectively with the functions $`\mathrm{\Psi }_i`$ and $`\mathrm{\Psi }_f`$ involving the states in the domain $`|ϵ_{𝐤\sigma }|<\stackrel{~}{D}`$. In the absence of a magnetic field, energy $`E`$ is the orbital energy in the initial state, and $`\omega `$ is the change in the orbital energy resulting from scattering. In the presence of Zeeman splitting, the initial energy and the energy transfer include the orbital and Zeeman parts, e.g. $`E=ϵ_𝐤+\sigma g_e\mu _BB/2`$.
The standard application of the lowest-order perturbation theory in the interaction of the energetic electron with the remaining system yields for the scattering probability
$$w_{fi}=|J(E)𝐬_{\sigma \sigma ^{}}\mathrm{\Psi }_f|𝐒|\mathrm{\Psi }_i|^22\pi \nu \delta (\xi _i\xi _f+\omega ),$$
where $`\nu `$ is the density of states for the energetic ($`ϵE`$) electron. After the summation over the final states and proper thermal averaging over the initial states, we are able to relate $`w_{fi}`$ with $`\sigma _{\sigma ^{}\sigma }(E,\omega )`$, and obtain the differential scattering cross-section
$`\sigma _{\sigma ^{}\sigma }(E,\omega )={\displaystyle \frac{\nu }{4v_F}}J^2(E)`$ (13)
$`\times \left(\delta _{\sigma ^{}\sigma }𝒮_{zz}(\omega )+𝐬_{\sigma ^{}\sigma }^+𝒮_+(\omega )+𝐬_{\sigma ^{}\sigma }^{}𝒮_+(\omega )\right),`$
where $`𝐬_{\sigma ^{}\sigma }^\pm =𝐬_{\sigma ^{}\sigma }^x\pm i𝐬_{\sigma ^{}\sigma }^y`$. As in the theory of neutron scattering MagnetismReference , the cross-section involves a spin-spin correlation function. Here it is the correlation function of the local magnetic impurity spin,
$`𝒮_{ab}(\omega )={\displaystyle \underset{\mathrm{}}{\overset{\mathrm{}}{}}}𝑑te^{i\omega t}S^a(t)S^b(0)`$ (14)
$`={\displaystyle \underset{\{|\mathrm{\Psi }_i,|\mathrm{\Psi }_f\}}{}}{\displaystyle \frac{e^{\beta \xi _i}}{Z}}\mathrm{\Psi }_i|S^a|\mathrm{\Psi }_f\mathrm{\Psi }_f|S^b|\mathrm{\Psi }_i2\pi \delta (\xi _i\xi _f+\omega ).`$
We thus reduced the scattering cross-section to an expression where its dependence on the energy of the scattering hot electron, $`E`$, separates from the dependence on the energy loss $`\omega `$. The dependence on the energy loss is determined by the dynamics of the impurity spin characterized by the correlation function $`𝒮`$. The spin correlator is related to the dissipative part of the impurity susceptibility via the fluctuation-dissipation theorem,
$`(g\mu _B)^2𝒮_{ab}(\omega )={\displaystyle \frac{2}{1e^{\beta \omega }}}\chi _{ab}^{\prime \prime }(\omega ).`$ (15)
Here $`\mu _B`$ is the Bohr magneton, and $`g`$ is the impurity $`g`$-factor. The behavior of $`\chi ^{\prime \prime }`$ in various limits will be discussed in the following sections. The spin dynamics is thus included in a non-perturbative fashion. It will allow us to investigate the behavior of the cross-section at any energy transfer; at $`\omega T_K`$ we apply effective Fermi liquid theory, and the region of intermediate energies, $`\omega T_K`$, is covered with the help of NRG calculations.
However, it is important to note that the total scattering cross-section is fixed by the sum rule for the spin correlation function, $`𝒮_{ab}(\omega )`$. Consider the total cross-section obtained after averaging over the initial electronic spin configurations, $`\sigma `$, summing over the final ones, $`\sigma ^{}`$, and integrating over the energy transfer $`\omega `$,
$`\sigma _{\mathrm{tot}}(E)={\displaystyle \frac{1}{2}}{\displaystyle \underset{\sigma ,\sigma ^{}}{}}{\displaystyle \underset{\mathrm{}}{\overset{\mathrm{}}{}}}𝑑\omega \sigma _{\sigma ^{}\sigma }(E,\omega )={\displaystyle \frac{3\pi }{8}}{\displaystyle \frac{1}{\nu v_F}}{\displaystyle \frac{1}{\mathrm{ln}^2\frac{E}{T_K}}}.`$ (16)
We substituted the explicit form for the energy dependent exchange interaction, $`J(E)=1/(\nu \mathrm{ln}(E/T_K))`$. The total scattering cross-section will be used throughout the rest of the paper as a convenient basic unit of measurement for the differential cross-section discussed below.
As we are mainly interested in the dependence of the scattering probability on the energy transfer, $`\omega `$, we will confine ourselves in the following to an analysis of the scattering cross-section averaged over the initial electronic spin configurations, $`\sigma `$, and summed over the final ones, $`\sigma ^{}`$,
$`\sigma (E,\omega )=\sigma _{\mathrm{tot}}(E){\displaystyle \frac{2}{3\pi }}\left[𝒮_{zz}(\omega )+{\displaystyle \frac{1}{2}}\left(𝒮_+(\omega )+𝒮_+(\omega )\right)\right].`$ (17)
Note that a Zeeman energy of electrons forming the Fermi sea was already incorporated in the definition of the energies $`E`$ and $`\omega `$. The generalization of our results to spin-resolved scattering is straightforward.
## III Inelastic electron scattering in the absence of Zeeman splitting
In the absence of a magnetic field the expression for the scattering cross-section (17) simplifies considerably since the impurity spin correlator is diagonal, $`𝒮(\omega )𝒮_{zz}(\omega )=\frac{1}{2}𝒮_+(\omega )`$,
$`\sigma (E,\omega )=\sigma _{\mathrm{tot}}(E){\displaystyle \frac{2}{\pi }}𝒮(\omega ).`$ (18)
Let us first establish the relation between Eq. (18) and the well-known result of the leading-logarithmic approximation Abrikosov1965 ; Anderson . For that, we need to substitute in Eq. (18) the function $`𝒮(\omega )`$ evaluated in the zeroth order in the exchange interaction $`J(\stackrel{~}{D})`$. In this order $`𝒮^{(0)}(\omega )=\frac{\pi }{2}\delta (\omega )`$, which yields the well-known result Abrikosov1965 ; Anderson for the cross-section,
$`\sigma ^{(0)}(E,\omega )=\sigma _{\mathrm{tot}}(E)\delta (\omega ),`$ (19)
i.e., scattering is elastic in the leading-logarithmic approximation. The elasticity breaks down, however, if one accounts for $`J(\stackrel{~}{D})0`$. Indeed, the exchange interaction $`J(\stackrel{~}{D})`$ leads to some dynamics of the impurity spin. The delta-function in Eq. (19) gets broadened, and spectral weight is transfered to finite energies $`\omega 0`$. The shape of the broadened peak is related to the character of the spin dynamics, which is different in the limits of high, $`TT_K`$, and low, $`TT_K`$ temperatures. We study the shape of the peak in these limits below. However, note that the broadening does not affect the total cross-section which is fixed by the sum rule and remains the same as for the elastic scattering, Eq. (19), evaluated in the leading logarithmic approximation.
### III.1 Inelastic electron scattering at $`TT_K`$
At $`TT_K`$, the local spin exhibits relaxational dynamics. The Bloch equations for the impurity spin in the absence of a magnetic field,
$`{\displaystyle \frac{}{t}}S^a={\displaystyle \frac{1}{\tau _K}}S^a,`$ (20)
imply the following form for the imaginary part of the susceptibilityGoetze71 , $`\chi _{ab}^{\prime \prime }(\omega )=\delta _{ab}\chi ^{\prime \prime }(\omega )`$ with
$`\chi ^{\prime \prime }(\omega )=\chi _0(T){\displaystyle \frac{\omega /\tau _K}{\omega ^2+(1/\tau _K)^2}}.`$ (21)
It involves the static susceptibility which is given by $`\chi _0(T)=(g\mu _B)^2/(4T)`$. The decay time, $`\tau _K`$, in the Bloch equations is the Korringa relaxation time Korringa , $`1/\tau _K=\pi (\nu J(T))^2T`$. Inserting the scale dependent exchange interaction, $`J(T)`$, the Korringa relaxation rate reads explicitely
$`{\displaystyle \frac{1}{\tau _K}}={\displaystyle \frac{\pi T}{\mathrm{ln}^2\frac{T}{T_K}}}.`$ (22)
It is parametrically smaller than $`T`$ at temperatures $`TT_K`$.
Expression (21) adequately accounts for the behavior of $`\chi ^{\prime \prime }`$ at low frequencies, $`\omega T`$, but fails at higher frequencies. For $`\omega T`$, the susceptibility can be evaluated within the lowest-order perturbation theory in the exchange constantKoller , $`J(\stackrel{~}{D})`$,
$`(g\mu _B)^2\chi ^{\prime \prime }(\omega )={\displaystyle \frac{\pi }{4}}{\displaystyle \frac{1}{\omega \mathrm{ln}^2\frac{|\omega |}{T_K}}}.`$ (23)
The additional logarithmic frequency dependence arises from the logarithmic enhancement of the exchange interaction due to the perturbative RG, which is now cut-off at a band width $`\stackrel{~}{D}\omega `$.
The resulting differential cross-section $`\sigma (E,\omega )`$ can be found with the help of Eq. (15). It is symmetric in $`\omega `$ at small energy transfers. It shows a narrow peak at $`\omega =0`$ and falls off significantly within the region of energies $`|\omega |T`$:
$$\sigma (E,\omega )=\sigma _{\mathrm{tot}}(E)\delta _\mathrm{\Gamma }(\omega ),$$
(24)
where we introduced a “broadened delta-function”, which is a Lorentzian with linewidth $`\mathrm{\Gamma }=1/\tau _K`$,
$$\delta _\mathrm{\Gamma }(\omega )=\frac{1}{\pi }\frac{1/\tau _K}{\omega ^2+(1/\tau _K)^2}.$$
(25)
As $`1/\tau _KT`$, see Eq. (22), almost the full weight of the total cross-section is accounted for by Eqs. (24) and (25), see Fig. 1.
At higher energy transfers, $`|\omega |T`$, the cross-section is asymmetric in $`\omega `$,
$$\sigma (E,\omega )=\sigma _{\mathrm{tot}}(E)\frac{1}{1e^{\omega /T}}\frac{1}{\omega \mathrm{ln}^2(|\omega |/T_K)}.$$
(26)
The probability for the scattered electron to acquire energy ($`\omega <0`$) is exponentially suppressed. Although the contribution of Eq. (26) to the total cross-section is parametrically small, $`1/\mathrm{ln}(T/T_K)`$, it is worth noting that its decay with $`\omega `$ is remarkably slow.
The slow decay of $`\sigma (E,\omega )`$ vs $`\omega `$ is related to the dependence on the transfered energy of the cross-section for inelastic electron-electron scattering mediated by a magnetic impurity Kaminski . The probability for such an inelastic scattering between two electrons with initial energies $`E`$ and $`E^{}`$ and final energies $`E\omega `$ and $`E^{}+\omega `$ was calculated in Ref. Kaminski, ; all of these four energies were assumed to be large compared to $`T_K`$. According to Ref. Kaminski, \[see also Eq. (9) of Ref. Ujsaghy, \], the contribution $`K(\omega ;E,E^{})`$ of a single magnetic impurity to this probability in the limit of high energy, $`E|\omega |`$, reads
$$K(\omega ;E,E^{})=\frac{3\pi }{8\nu }\frac{1}{\mathrm{ln}^2\frac{|E|}{T_K}}\frac{4}{\left(\mathrm{ln}\frac{|E^{}|}{T_K}+\mathrm{ln}\frac{|E^{}+\omega |}{T_K}\right)^2}\frac{1}{\omega ^2}.$$
(27)
The differential cross-section (26) can be obtained by integrating $`K(\omega ;E,E^{})`$ over the available phase space volume of one of the scattering electrons,
$`v_F\sigma (E,\omega )={\displaystyle 𝑑E^{}f(E^{})(1f(E^{}+\omega ))K(\omega ;E,E^{})}.`$ (28)
The Fermi functions in Eq. (28) confine the energy $`E^{}`$ to an interval $`\omega E^{}0`$. This includes a regime where the arguments of the $`E^{}`$–dependent logarithmic factors are not meaningful anymore and should be replaced by temperature or the Korringa relaxation rate. After such cut-off, integration over $`E^{}`$ is easily performed, yielding $`\mathrm{ln}^2|\omega |/T_K`$ within logarithmic accuracy. This way, starting from the collision integral kernel of Ref. Kaminski, , one recovers Eq. (26).
### III.2 Inelastic electron scattering at $`TT_K`$
When the temperature is below the Kondo temperature the picture differs drastically from the zeroth order result (19). For $`TT_K`$, the low-frequency behavior of the scattering cross-section is beyond perturbation theory. Nevertheless, the cross-section for small energy transfers, $`|\omega |T_K`$, may be found with the help of the Shiba relation Shiba for the susceptibility,
$$(g\mu _B)^2\chi ^{\prime \prime }(\omega )=2\pi \omega \left[\chi _0(T=0)\right]^2.$$
(29)
The zero-temperature static susceptibility $`\chi _0(0)`$ is used conventionallyHewson to define the pre-exponential factor of the Kondo temperature, $`\chi _0(0)=[(g\mu _B)^2W]/(4T_K)`$; here $`W=0.413\mathrm{}`$ is Wilson’s number. (We present a convenient derivation of the Shiba relation in Appendix A.) The corrections to the Shiba relation are of order $`𝒪(\omega T^2/T_K^2,\omega ^3/T_K^2)`$ and are sub-leading. We thus obtain for the cross-section at $`|\omega |,TT_K`$:
$$\sigma (E,\omega )=\sigma _{\mathrm{tot}}(E)\frac{W^2}{2}\frac{1}{1e^{\omega /T}}\frac{\omega }{T_K^2}.$$
(30)
The high-frequency limit, $`|\omega |T_K`$, of the scattering cross-section can still be obtained perturbatively and is given by Eq. (26).
Comparing the results Eq. (26) and Eq. (30) we see that for temperatures $`TT_K`$ the differential cross-section $`\sigma (E,\omega )`$ peaks at energy transfers of the order of $`\omega T_K`$. It then decreases linearly upon further decrease of $`\omega `$, until it crosses over (at $`|\omega |T`$) into the exponential tail for $`\omega <0`$, see inset of Fig. 2. At zero temperature the factor containing $`\mathrm{exp}(\omega /T)`$ in Eq. (30) becomes a step function which forbids any energy gain from the Kondo system,
$$\sigma (E,\omega )=\sigma _{\mathrm{tot}}(E)\frac{W^2}{2}\mathrm{\Theta }(\omega )\frac{\omega }{T_K^2};$$
(31)
here $`\mathrm{\Theta }(x)=1`$ if $`x>0`$ and $`0`$ if $`x<0`$.
The region between the asymptotes given in Eqs. (26) and (30) can be bridged by calculations performed with the NRG method. In this method, after the logarithmic discretization of the conduction band one maps the Kondo Hamiltonian onto a semi-infinite chain with the impurity at the end. As a consequence of the logarithmic discretization, the hopping along the chain decreases exponentially, $`t_n\mathrm{\Lambda }^{n/2}`$, where $`\mathrm{\Lambda }>1`$ is the discretization parameter and $`n`$ is the site index. (We have used $`\mathrm{\Lambda }=2`$ throughout the calculations presented in the paper.) The separation of energy scales provided by the exponential decay of the hopping rate allows us to diagonalize the Hamiltonian iteratively and keep the eigenstates with the lowest energy as most relevant ones. Since we know the energy eigenvalues and eigenstates, we are able to calculate the impurity spin correlation function directly (see Eq.(14)) given that the Dirac delta function appearing in the Lehman representation must be broadened when performing a numerical calculationCosti ; Bulla . The result of the NRG calculation is shown in Fig. 2.
To summarize this Section we demonstrated that the dynamics of the impurity spin leads to inelastic electron scattering at all temperatures. The main contribution to the total scattering cross-section comes from $`\omega T_K`$ or $`|\omega |1/\tau _K`$ at $`TT_K`$ and $`TT_K`$ respectively. The total scattering cross-section is fixed by the sum rule for the impurity spin correlation function, see Eq. (16), and is thus determined by the effective exchange constant, $`J(E)`$, evaluated within the leading logarithmic approximationAbrikosov1965 .
## IV Zeeman effect in the electron scattering
We now address the case when the degeneracy of the impurity spin is lifted by a magnetic field. The Zeeman splitting of the impurity spin is described by the Hamiltonian
$`H_{\mathrm{Zeeman}}=g\mu _B𝐒^zB.`$ (32)
In the presence of the Zeeman splitting the scattering electron has to pay Zeeman energy in order to transfer spin to the Kondo system. The resonance structure for electron scattering involving a spin-flip will therefore differ from the one of non-spin-flip scattering. Evaluating the impurity spin correlator in zeroth order in $`J(\stackrel{~}{D})`$ we obtain for the scattering cross-section (17) in the leading logarithmic approximation
$`\sigma ^{(0)}(E,\omega )=\sigma _{\mathrm{tot}}(E){\displaystyle \frac{2}{1+e^{\beta \omega }}}`$ (33)
$`\times {\displaystyle \frac{1}{3}}\left\{\delta (\omega )+\delta (\omega \omega _Z(B))+\delta (\omega +\omega _Z(B))\right\}`$
The single delta-function for $`B=0`$, Eq. (19), is now split into three contributions. In addition to a delta-function at zero frequency, which is due to non-spin-flip scattering, there are two Zeeman satellites at $`\omega =\pm \omega _Z(B)`$. In the limit of low temperatures, $`TB`$, the satellite at negative Zeeman energy corresponding to an energy gain of the scattering electron is exponentially small as it is clear from Eq. (33).
The Zeeman energy $`\omega _Z(B)`$ depends on the renormalized $`g`$-factor, which is different from its bare value $`g`$ appearing in the Zeeman Hamiltonian (32). When we derived the effective interaction Hamiltonian (II), we integrated out a finite band of electronic degrees of freedom which lead to a renormalization of the exchange interaction $`J`$. The Zeeman term (32) is not invariant under this perturbative renormalization of the Kondo model. Similar to the exchange interaction $`J`$, the $`g`$-factor is also renormalized when the band is reduced from $`D`$ to $`\stackrel{~}{D}`$. As explained in Appendix B, the scale-dependent $`g`$-factor in the leading logarithmic order is given by
$`{\displaystyle \frac{g(\stackrel{~}{D})}{g}}=\left(1{\displaystyle \frac{1}{2\mathrm{ln}\stackrel{~}{D}/T_K}}\right).`$ (34)
To find the observable value of $`g`$-factor, one needs to set $`\stackrel{~}{D}=\mathrm{max}\{T,g\mu _BB\}`$. The position of the Zeeman resonances, to the leading logarithmic order, is given by Moore00
$`\omega _Z(B)=g\left(1{\displaystyle \frac{1}{2\mathrm{ln}\left(\mathrm{max}\{T,g\mu _BB\}/T_K\right)}}\right)\mu _BB.`$ (35)
Beyond the leading logarithmic approximation the dynamics of the local spin is characterized by a further redistribution of the spectral weight of the scattering cross-section (33). However, a striking feature of the presence of a magnetic field is that a finite weight of the delta-resonance at $`\omega =0`$ will still survive after accounting for the coupling of the impurity spin to the low-energy degrees of freedom of the Fermi sea. In other words, at any ratio $`T/T_K`$ a part of the scattering becomes elastic if a magnetic field $`B0`$ is turned on. This can be best understood by considering the longitudinal spin correlation function in time. For $`B0`$ this correlation function will not fully decay with time but rather saturate at a value given by the finite expectation value of the impurity spin, $`S^z(t)S^z(0)S^z^2`$ for $`t\mathrm{}`$. This finite saturation value leads to a finite weight of the delta-function $`\delta (\omega )`$ in its Fourier transform and in Eq. (13). Let us decompose $`\sigma (E,\omega )`$ into the elastic and inelastic parts,
$`\sigma (E,\omega )=\sigma _{\mathrm{el}}(E,\omega )+\sigma _{\mathrm{inel}}(E,\omega ).`$ (36)
The elastic part will be determined by the magnetization of the impurity spin
$`\sigma _{\mathrm{el}}(E,\omega )=\sigma _{\mathrm{tot}}(E){\displaystyle \frac{4}{3}}S^z^2\delta (\omega ).`$ (37)
Being a thermodynamic quantity, $`S^z`$ has a well-studied field and temperature dependence bethe-ansatz ; Hewson . In the scaling regime, $`f(t,b)=S^z`$ is a function of $`t=T/T_K`$ and $`b=g\mu _BB/T_K`$. The asymptote of $`f(t,b)`$ at $`\mathrm{max}(t,b)1`$ is with logarithmic accuracy given by
$`f(t,b)={\displaystyle \frac{1}{2}}\left(1{\displaystyle \frac{1}{2\mathrm{ln}[\mathrm{max}(t,b)]}}\right)`$ (38)
$`\times \mathrm{tanh}\left[{\displaystyle \frac{b}{2t}}\left(1{\displaystyle \frac{1}{2\mathrm{ln}[\mathrm{max}(t,b)]}}\right)\right].`$
Note that in the limit $`t=0`$, $`b1`$, Eq. (38) yields the ground-state value of $`S^z`$ in the perturbative regime. In the opposite limit of a weak field, $`b1t`$, spin polarization is small according to the Curie law, $`fb/4t`$. In the developed Kondo regime, $`\mathrm{max}(t,b)1`$, the average spin is $`f(t,b)=(W/4)b`$.
The weight of the elastic scattering, Eq. (37), evaluated with NRG is shown in Fig. 3. In the limit of small magnetic fields this weight increases as $`B^2`$. The saturation of the weight to its large-field limit, $`1/3`$, is remarkably slow due to the logarithmic correction to the magnetizationbethe-ansatz , see Eq. (38).
The inelastic part of the scattering cross-section, $`\sigma _{\mathrm{inel}}(E,\omega )`$, accounts for the remaining spectral weight. Note however that the total scattering cross-section, i.e. the total spectral weight, is independent of the magnetic field; its value being fixed by the sum rule for the impurity spin correlator.
### IV.1 Dissipative part of magnetic susceptibility
To analyze the inelastic scattering-cross section in more detail for the two limiting cases $`TT_K`$ and $`TT_K`$, we start from presenting the proper details regarding the frequency dependence of the dissipative parts of longitudinal and transversal impurity spin susceptibilities ($`\chi _{zz}^{\prime \prime }`$ and $`\chi _+^{\prime \prime }`$, respectively).
At $`TT_K`$, one may treat the exchange interaction $`J(\stackrel{~}{D})`$ perturbatively at any field $`B`$. The effect of $`B`$ on $`\chi ^{\prime \prime }`$ is negligible as long as the Zeeman splitting $`\omega _Z(B)`$ is smaller than the Korringa relaxation rate $`1/\tau _K`$ , see Eq. (22). At higher fields, the susceptibility becomes anisotropic, $`\chi _{zz}^{\prime \prime }\frac{1}{2}\chi _+^{\prime \prime }`$, and its frequency dependence acquires a well-resolved structure. The dissipative part of the susceptibility can be found from the Bloch equations MagnetismReference . The transversal part takes the form
$`\chi _+^{\prime \prime }(\omega )`$ $`=2\chi _T{\displaystyle \frac{\omega /T_2}{[\omega \omega _Z(B)]^2+(1/T_2)^2}},`$ (39)
where the static transversal differential susceptibility can be expressed with the help of Eq. (38) as $`\chi _T=(g\mu _B)f(t,b)/B`$. The longitudinal part reads
$`\chi _{zz}^{\prime \prime }(\omega )`$ $`=\chi _L^D{\displaystyle \frac{\omega /T_1}{\omega ^2+(1/T_1)^2}},`$ (40)
where $`\chi _L^D`$ is given by
$`\chi _L^D`$ $`={\displaystyle \frac{(g\mu _B)^2\left(1\frac{1}{\mathrm{ln}\left(\mathrm{max}\{T,g\mu _BB\}/T_K\right)}\right)}{4T\mathrm{cosh}^2\frac{\omega _Z(B)}{2T}}}.`$ (41)
The factor $`\chi _L^D`$ can be understood as the contribution to the static suceptibility which originates from the response of the occupation factors of the two Zeeman levels to a varying magnetic field, $`\chi _L^D=g_{\mathrm{eff}}\mu _Bn_+n_{}/B`$; here $`g_{\mathrm{eff}}`$, see Eq. (34), is the appropriately renormalized $`g`$-factor. Note that only in the limit $`\omega _Z(B)T`$ when the renormalized $`g`$-factor (34) is insensitive to the magnetic field, $`\chi _L^D`$ does coincide with the full static longitudinal differential susceptibility $`\chi _L=g\mu _Bf(t,b)/B`$.
In the case of a moderately high field, $`1/\tau _K\omega _Z(B)T`$, the relaxation times, $`T_1`$ and $`T_2`$, equal each other MagnetismReference and are given by Eq. (22); $`T_1=T_2=\tau _K`$.
At even higher fields, $`\omega _Z(B)T`$, the peak structure in $`\chi _+^{\prime \prime }(\omega )`$ is still described by a Lorentzian form of Eq. (39) but the corresponding relaxation time is determined now by the Zeeman splitting rather than by temperatureGoetze71 ,
$`{\displaystyle \frac{1}{T_2}}={\displaystyle \frac{\pi }{4}}{\displaystyle \frac{\omega _Z(B)}{\mathrm{ln}^2\frac{\omega _Z(B)}{T_K}}}.`$ (42)
The frequency dependence of the longitudinal susceptibility, however, requires additional discussion.
Generally, the susceptibility, $`\chi _{ij}(\omega )`$, describes the response of the magnetic impurity to a local magnetic field that oscillates with frequency $`\omega `$. At low frequencies, the variation of the dissipative part of the longitudinal component $`\chi _{zz}^{\prime \prime }(\omega )`$ given by Eq. (40) can be understood in the framework of the Debye mechanism Debye of relaxational losses: At $`\omega =0`$, relaxation caused by the exchange interaction between the local magnetic moment and itinerant electrons establishes equilibrium Gibbs occupation factors for the two Zeeman-split levels. At finite but small frequency $`\omega `$, the Zeeman splitting, which is caused by the sum of a constant and a slowly-varying magnetic field, changes with time slowly, and the relaxation acts to adjust the occupation factors to the instant values of the Zeeman splitting. The adjustment occurs via the emission (or absorption) of particle-hole pairs with energy $`\epsilon _{ph}\omega _Z(B)`$ by flips of the local spin. It is the time variation of the occupation factors of the Zeeman-split levels that leads to dissipation. In the limit $`\omega 0`$, the leading term in $`\chi _{zz}^{\prime \prime }(\omega )`$, according to Eq. (40), is
$`\chi _{zz}^{\prime \prime }(\omega )|_{\mathrm{Debye}}`$ $`=\chi _L^DT_1\omega .`$ (43)
As was already mentioned, in the weak-field case $`1/T_1`$ is given by Eq. (22). In the limit $`\omega _Z(B)T`$, the time $`T_1`$ was found Goetze71 to be $`T_1=T_2/2`$ with $`T_2`$ of Eq. (42). The contribution (43) to $`\chi _{zz}^{\prime \prime }`$ from the Debye relaxational losses is valid at arbitrary ratio $`\omega _Z(B)/T`$. Note however that despite the fact that Eq. (43) describes dissipation at low frequency the Debye mechanism is associated with the emission of particle-hole pairs with a comparatively high energy $`\epsilon _{ph}\omega _Z(B)`$. In the limit $`\omega _Z(B)T`$, the Debye mechanism thus yields only an exponentially small contribution to dissipation,
$$\chi _{zz}^{\prime \prime }(\omega )|_{\mathrm{Debye}}=\frac{2}{\pi }\frac{(g\mu _B)^2\omega }{T}\frac{\mathrm{ln}^2\frac{\omega _Z(B)}{T_K}}{\omega _Z(B)}\mathrm{exp}\left[\frac{\omega _Z(B)}{T}\right].$$
(44)
The exponential smallness of $`\chi _L^D`$ comes from the small probability of the thermal occupation of the highly-excited state, corresponding to the upper of the two Zeeman-split levels. Temporal variations in this exponentially small quantity leads to an exponentially small contribution to $`\chi _{zz}^{\prime \prime }(\omega )`$.
Under these conditions, a second contribution, originating from the low-energy part of the spectrum, $`|ϵ|\mathrm{max}[\omega ,T]`$, becomes important. The processes contributing here do not involve real impurity spin-flip processes (which are exponentially suppressed), but only virtual transitions. The starting point is the observation that the impurity magnetization locally polarizes the Fermi sea. If the Zeeman splitting of the impurity is slowly varied with a small frequency $`\omega `$, the magnetic polarization of the Fermi sea will adjust itself to the instantaneous adiabatic value of the impurity magnetization. Since the spectrum of the particle-hole pairs is continuous this adjustment results in dissipation via the emission of pairs with small frequency $`\epsilon _{ph}\omega `$, which is in contrast to the Debye-mechanism where the emitted particle-hole pairs carry a large energy of order the Zeeman splitting. As shown in Appendix A, this contribution to the susceptibility can be obtained by applying Nozières’ Fermi liquid theory and is adequately accounted for by the generalized Shiba relation Eq. (65). Evaluating $`dS^z/dB`$ with the help of Eq. (38) at $`Tg\mu _BB`$, we find for the dissipative part of the longitudinal susceptibility
$`\chi _{zz}^{\prime \prime }(\omega )={\displaystyle \frac{\pi }{8}}{\displaystyle \frac{(g\mu _B)^2\omega }{\omega _Z^2(B)}}{\displaystyle \frac{1}{\mathrm{ln}^4\frac{\omega _Z(B)}{T_K}}},\omega \omega _Z(B).`$ (45)
Comparing Eq. (45) with the result for the Debye mechanism, we see that the strong-field asymptote Eq. (44) for the latter mechanism is important only in a narrow interval of temperatures $`\omega _Z(B)T\omega _Z(B)/6`$, as for all practical purposes $`\mathrm{ln}\mathrm{ln}(\omega _Z/T_K)1`$. Dispensing with that interval, we will use for the dissipative part of the longitudinal susceptibility Eq. (40) with $`T_1=\tau _K`$ in the case $`\omega _Z(B)T`$, and Eq. (45) in the case $`\omega _Z(B)T`$.
At low temperatures, $`TT_K`$, there is little effect of the magnetic field on $`\chi ^{\prime \prime }(\omega )`$ for weak fields, $`g\mu _BBT_K`$. In the strong-field regime, $`\omega _Z(B)T_KT`$, the main contribution to the transversal part of the dissipative susceptibility is given by Eq. (39) with the relaxation time $`T_2`$ of Eq. (42). The longitudinal part is described by Eq. (45) at $`\omega \omega _Z(B)`$. Equation (39) adequately describes the non-monotonic behavior of $`\chi _+^{\prime \prime }(\omega )`$, but fails at higher frequencies; similarly the linear dependence in $`\chi _{zz}^{\prime \prime }(\omega )`$ does not stretch beyond $`\pm \omega _Z(B)`$. In the limit $`|\omega |\omega _Z(B)`$, the magnetic field does not affect significantly the dissipation, and Eq. (23) is applicable.
### IV.2 Elastic and inelastic components of electron scattering
The coupling of the impurity spin to the low-energy degrees of freedom of the Fermi seas will lead to a broadening and redistribution of the spectral weight of the three delta-functions in Eq. (33).
#### IV.2.1 High temperatures: $`TT_K`$
At high temperature, $`TT_K`$, and weak magnetic field, $`\omega _Z(B)T`$, the spin polarization is weak, and the elastic component of the scattering is small. Using Eqs. (37) and (38) we find
$`\sigma _{\mathrm{el}}(E,\omega )=`$ (46)
$`=\sigma _{\mathrm{tot}}(E){\displaystyle \frac{4}{3}}\left[1{\displaystyle \frac{2}{\mathrm{ln}(T/T_K)}}\right]\left[{\displaystyle \frac{g\mu _BB}{4T}}\right]^2\delta (\omega ).`$
The major contribution to the scattering cross-section comes from the inelastic processes. At fields satisfying the condition $`\omega _Z(B)\tau _K1`$, which still belongs to the domain of weak fields $`\omega _Z(B)T`$, the single maximum in the $`\omega `$-dependence of the cross-section, see Eq. (24), splits into three:
$`\sigma _{\mathrm{inel}}(E,\omega )`$ $`\sigma _{\mathrm{tot}}(E)`$ (47)
$`\times `$ $`{\displaystyle \frac{1}{3}}\left[\delta _\mathrm{\Gamma }(\omega )+\delta _\mathrm{\Gamma }(\omega \omega _Z(B))+\delta _\mathrm{\Gamma }(\omega +\omega _Z(B))\right]`$
The broadened delta-function was defined in Eq. (25) with a relaxation rate, $`\mathrm{\Gamma }`$, given by the inverse Korringa time, $`\mathrm{\Gamma }=1/\tau _K`$. (We neglected a small part of the spectral weight which moved to the elastic component of the scattering cross-section).
With the increase of the ratio $`\omega _Z(B)/T`$, the intensity of the elastic scattering increases, and in the strong-field limit we find
$$\sigma _{\mathrm{el}}(E,\omega )=\sigma _{\mathrm{tot}}(E)\frac{1}{3}\left[1\frac{1}{\mathrm{ln}(g\mu _BB/T_K)}\right]\delta (\omega ).$$
(48)
Simultaneously, the maximum of $`\sigma _{\mathrm{inel}}(E,\omega )`$ at negative $`\omega `$ gets suppressed, and the structure at $`|\omega |\omega _Z(B)`$ broadens and becomes asymmetric. In the limit $`\omega _Z(B)/T1`$, only a single maximum at positive $`\omega `$ remains in the inelastic cross-section,
$`\sigma _{\mathrm{inel}}(E,\omega )`$ $`=`$ $`\sigma _{\mathrm{tot}}(E){\displaystyle \frac{2}{3\pi }}{\displaystyle \frac{1}{1e^{\omega /T}}}{\displaystyle \frac{1}{\omega _Z(B)}}`$ (49)
$`\times `$ $`{\displaystyle \frac{\omega /T_2}{[\omega \omega _Z(B)]^2+(1/T_2)^2}}.`$
Here the relaxation time $`T_2`$ is defined by Eq. (42). This main contribution to the inelastic scattering is proportional to $`\chi _+^{\prime \prime }(\omega )`$ and comes from the spin-flip processes. The comparison of Eqs. (44) and (45) with Eq. (39) shows that at $`\omega _Z(B)T_K`$ the effect of the dissipative part of longitudinal susceptibility is small starting from $`\omega _Z(B)/T4`$. Under this condition, $`\chi _{zz}^{\prime \prime }(\omega )`$ yields a contribution to $`\sigma (E,\omega )`$ which is small compared to Eq. (49).
The high-frequency tail, $`|\omega |\mathrm{max}[T_K,g\mu _BB,T]`$, is unaffected by the Zeeman splitting and still given by Eq. (26).
#### IV.2.2 Low temperatures: $`TT_K`$
We turn now to the opposite limit of small temperature, $`TT_K`$. At weak magnetic field, $`g\mu _BBT_K`$, the low-frequency behavior of the scattering cross-section is beyond perturbation theory. In this regime, the electron scatters from a fully developed, many-body Kondo singlet. Here we can use the Shiba relation Eq. (30) to access the low-frequency tail of the cross-section. In the presence of a magnetic field there are additional corrections to the Shiba relation of order $`𝒪(\omega (g\mu _BB)^2/T_K^2)`$ which are sub-leading and are neglected in the following. We get for the low-frequency part $`|\omega |T_K`$
$`\sigma (E,\omega )`$ $`=\sigma _{\mathrm{tot}}(E){\displaystyle \frac{W^2}{2}}{\displaystyle \frac{1}{1e^{\omega /T}}}`$ (50)
$`\times \left[{\displaystyle \frac{1}{6}}\left({\displaystyle \frac{g\mu _BB}{T_K}}\right)^2\delta (\omega )+{\displaystyle \frac{\omega }{T_K^2}}\right].`$
where $`W`$ is again Wilson’s number Hewson . The scattering cross-section decreases linearly with frequency. At $`\omega T`$ the linear decrease crosses over into an exponential tail which extends to negative frequencies. In Fig. 4 NRG results at $`T=0`$ for the inelastic cross-section at small magnetic fields are compared with the NRG data at $`B=0`$. In finite field the slope in the linear low-frequency regime is reduced. The difference in slope is of order $`𝒪(g\mu _BB/T_K)^2`$, a correction alluded to but neglected in Eq. (50). This difference however accounts for the reduction of the inelastic scattering weight. The weight of order $`𝒪(g\mu _BB/T_K)^2`$ is transfered from the inelastic to the elastic scattering contribution leading to a delta peak at $`\omega =0`$, as sketched in the inset of Fig. 4. In contrast to the case of high temperatures ($`TT_K,g\mu _BB`$) the elastic scattering contribution now does not sit on top of a large Lorentz-peak but is rather located within the scattering pseudogap. Although its weight is small, here it is easily distinguishable from the background. The crossover from the linear dependence on $`\omega `$ to the high-frequency behavior occurs at $`\omega T_K`$, where the inelastic scattering cross-section has a maximum. The high-frequency tail is still given by the perturbative expression (26).
When the magnetic field is increased above the Kondo temperature, $`g\mu _BBT_K`$, the elastic and inelastic components of the scattering cross-section are given by Eqs. (48) and (49), respectively. The elastic peak at $`\omega =0`$ now exhausts almost the full spectral weight of the longitudinal correlator, i.e. it accounts for approximately $`1/3`$ of the total scattering cross-section, see Fig. 3. The remaining $`2/3`$ of the total spectral weight are to be found in the extended structure of the Zeeman satellite (49) centered at $`\omega =\omega _Z(B)`$. The effect of Zeeman spitting on the cross-section is confined to the region of energies $`|\omega |\omega _Z(B)`$. At $`|\omega |\mathrm{max}[T_K,g\mu _BB,T]`$ the behavior of $`\sigma (E,\omega )`$ is again given by Eq. (26).
In Fig. 5 the inelastic cross-section is shown in the limit of large magnetic fields, $`g\mu _BBT_K`$, as given by Eq. (49). The inset compares the result with the NRG. The low-frequency and high-frequency asymptotes are reproduced in the numerical calculation fairly well. The deviation in the width of the Zeeman peak, however, demonstrates the limitation of the NRG method. Due to the logarithmic frequency resolution the NRG tends to overbroaden any peak in the spectral function centered around a nonzero frequency.
## V Possible Experiments
As we have shown above, the differential scattering cross-section of the magnetic impurity shows a rich structure in frequency space. In the following we suggest two experiments that are sensitive to the dynamics of a Kondo impurity and from which, in principle, the energy-resolved scattering cross-section can be extracted.
### V.1 Mesoscopic wires
Inelastic scattering off magnetic impurities has been identified to be at the origin of an anomalously large energy relaxation in mesoscopic metallic wires Pothier . We propose a modification of the original experiment performed by Pothier et al. Pothier that allows in principle to access the scattering cross-section considered in this paper. We assume that the wire is connected to the reservoirs on one end by an open contact and on the other via a tunnel junction, see Fig. 6. In the limit of small transparency of the tunnel junction the wire is almost in equilibrium; only a small amount of energetic quasi-particles tunnel into the wire and relax their energy during scattering processes on magnetic impurities. In lowest order in the transparency of the tunnel junction we can treat this relaxation mechanism in terms of the differential scattering cross-section, $`\sigma (E,\omega )`$ of a test particle coming with energy $`E`$ in an otherwise equilibrium system.
Consider a mesoscopic wire of length $`L`$. The equilibrium distribution in the right and left reservoir is given by a Fermi function, $`f_\mathrm{F}(E)`$ and $`f_\mathrm{F}(EeU)`$, respectively, where the energy $`E`$ is measured with respect to the chemical potential of the right reservoir. The voltage drop across the wire is $`U`$. Within the wire the distribution function, $`f(E;x,U)`$, will depend on the position across the wire, $`x[0,L]`$. It is determined by the relaxation mechanisms and carries information on the differential scattering cross-section $`\sigma (E,\omega )`$. This distribution function is probed by an additional tunnel contact that is attached to the wire at a certain position $`x_T[0,L]`$ and connects it to a conductor with a sharp feature in the density of states. Measurement of a small tunneling current through this auxiliary contact as a function of voltage $`V`$, see Fig. 6, allows one Pothier to probe the electron energy distribution. This way, the distribution function $`f(E;x_T,U)`$ in the wire at some point $`x_T`$ in the presence of a bias $`U`$ applied across the wire was investigated Pothier ; Anthore . The sharp feature in the electron density of states in the probe was due to its superconducting state Pothier (the BCS anomaly) or due to the Coulomb interaction in a low-dimensional diffusive electron system Anthore (zero-bias anomaly). In the following we show that measurement of the derivative $`f(E;x_T,U)/U`$ in a modified (compared to Ref. Pothier, ) setup of Fig. 6 allows one to access the inelastic scattering cross-section $`\sigma (E,\omega )`$.
The distribution function within the wire is governed by the diffusive Boltzmann equation Nagaev
$`D{\displaystyle \frac{^2f(E;x,U)}{x^2}}=I[f]`$ (51)
where $`D`$ is the diffusion coefficient of the wire. The collision integral is local in space,
$`I[f]`$ $`=c_{\mathrm{imp}}v_F`$ (52)
$`\times {\displaystyle _{\mathrm{}}^{\mathrm{}}}d\omega [f(E)(1f(E\omega ))\sigma (E,\omega )`$
$`(1f(E))f(E\omega )\sigma (E\omega ,\omega )].`$
where $`c_{\mathrm{imp}}`$ is the impurity concentration within the wire and $`\sigma (E,\omega )`$ is the differential cross-section of a single magnetic impurity, and for notational convenience the dependence of the distribution function, $`f`$, on $`x`$ and $`U`$ has been omitted. The boundary condition at the open contact to the right reservoir is simply $`f(E;x=L,U)=f_\mathrm{F}(E)`$. The boundary condition at the tunnel contact, which connects the wire to the left reservoir, is determined by current conservation
$`g_T(f_\mathrm{F}(EeU)f(E;x=0,U))=`$ (53)
$`=\nu D{\displaystyle \frac{f(E;x=0,U)}{x}}`$
where $`g_T`$ is the dimensionless conductance of the tunneling contact and $`\nu `$ is the density of states of the wire.
In zeroth order in the collision integral we obtain the solution
$`f^{(0)}(E;x,U)=f_\mathrm{F}(E){\displaystyle \frac{L_0+x}{L_0+L}}+f_\mathrm{F}(EeU){\displaystyle \frac{Lx}{L_0+L}},`$ (54)
where we introduced the length $`L_0`$; the relation of $`L_0`$ to the length of the wire, $`L`$, is determined by the ratio of conductances of the wire and the tunneling contact, $`L_0/L=g_w/g_T`$, with $`g_w=\nu D/L`$. In the limit of large transparency of the tunneling contact, $`L_0/L1`$, the obtained solution reduces to the well-known formula for the distribution function of a diffusive wire with open contacts Nagaev . However, we are focusing on the other limit of a large tunneling barrier, $`L/L_01`$, where we get $`f^{(0)}(E;x,U)=f_\mathrm{F}(E)+\delta f^{(0)}(E;x,U)`$ with
$`\delta f^{(0)}(E;x,U)=`$ (55)
$`=\left(f_\mathrm{F}(EeU)f_\mathrm{F}(E)\right){\displaystyle \frac{Lx}{L_0}}+𝒪\left({\displaystyle \frac{L}{L_0}}\right)^2.`$
The deviation of the energy distribution in the wire from the one in the right reservoir is of first order in the small parameter $`L/L_0`$.
In the following we consider the correction to the distribution function in lowest order in the collision integral and in the small parameter $`L/L_0`$. We get in leading order in $`L/L_0`$
$`I[f^{(0)}]`$ $`=c_{\mathrm{imp}}v_F{\displaystyle \frac{Lx}{L_0}}\left(f_\mathrm{F}(EeU)f_\mathrm{F}(E)\right)`$ (56)
$`\times {\displaystyle _{\mathrm{}}^{\mathrm{}}}d\omega \sigma (E,\omega )(1e^{\beta \omega })`$
$`\left(f_\mathrm{F}(E\omega eU)f_\mathrm{F}(E\omega )\right).`$
Returning now to Eq. (51), we are able to find the correction to the distribution function. The energy dependence of $`f(E;x,U)/U`$ within the interval $`0<E<eU`$ is caused by electron energy relaxation. At $`T=0`$ it is given by
$`{\displaystyle \frac{f(E;x,U)}{(eU)}}=`$ (57)
$`={\displaystyle \frac{c_{\mathrm{imp}}v_F}{DL_0}}\left({\displaystyle \frac{x^3}{6}}+{\displaystyle \frac{Lx^2}{2}}{\displaystyle \frac{L^2x}{3}}\right)\sigma (eU,eUE).`$
The structure of the distribution function in this energy interval is directly related to the differential scattering cross-section $`\sigma (E,\omega )`$. The simple relation between $`f(E;x,U)/U`$ and the cross-section holds as long as the events of scattering off magnetic impurities occur rarely over the time limited by the diffusion of an electron across the wire. Note however that in addition to Eq. (57), there is a sharp contribution at the edge of the energy interval, $`E=eU`$, resulting from the zeroth order contribution (55) to the non-equilibrium distribution function. At finite temperature, this limits the experimental accessibility of $`\sigma (E,\omega )`$ for $`\omega T`$.
### V.2 Quantum dot in the Kondo regime
A second experimental possibility is very similar in spirit to the first one but considers a quantum dot setup. The starting point is a semiconductor-based ballistic wire that has on its right hand side contact with a large reservoir, see Fig. 7. On the other end, the wire is connected to a quantum dot in the Kondo regime. In this regime, the spin of the dot forms a many-body ground state with the electrons in the wire. Electrons injected at a bias $`U`$ into the wire through the quantum dot form a non-equilibrium distribution, which is probed via an auxiliary weak contact having potential $`V`$. The auxiliary contact consists of a second quantum dot, marked RL in Fig. 7, which is tuned to the resonant tunneling regime. In the case of a sharp resonance, the setup of Fig. (7) allows one to measure the electron energy distribution in the quantum wire. This non-equilibrium distribution, in turn, is sensitive to the inelastic transport through the Kondo dot and bears signatures of the differential scattering cross-section of the Kondo spin.
If the left reservoir is disconnected from the Kondo dot the electrons within the ballistic wire have a equilibrium Fermi distribution, $`f^{(0)}(E)=f_F(E)`$. The injection of hot electrons from the left reservoir will lead to a non-equilibrium correction to the distribution function of the right-movers within the wire, $`f(E)=f_F(E)+\delta f(E)`$. We obtain in lowest order in tunneling between the left lead and the dot:
$`v_F\delta f(E)=v_F{\displaystyle }d\xi \sigma (\xi ,\xi E)(f_F(\xi eU)(1f_F(E))`$
$`e^{\beta (\xi E)}f_F(E)(1f_F(\xi eU))),`$ (58)
where we used already the detailed balance relation, $`\sigma (E,\omega )e^{\beta \omega }=\sigma (E\omega ,\omega )`$. After taking the derivative with respect to $`U`$, the above equation simplifies considerably at $`T=0`$, and we get
$`{\displaystyle \frac{f(E)}{(eU)}}=\sigma (eU,eUE).`$ (59)
The measurement of this quantity with help of the auxiliary contact thus yields direct access to the differential inelastic scattering cross-section of a Kondo system.
## VI Summary
We analyzed inelastic scattering of energetic electrons off a magnetic impurity. For such scattering, the dependence of the differential cross-section, $`\sigma (E,\omega )`$, on energy $`E`$ of the incoming electron is logarithmically weak at $`ET_K`$, and arises from the renormalization of the exchange coupling. In the leading logarithmic approximation, the total cross-section $`\sigma _{\mathrm{tot}}=𝑑\omega \sigma (E,\omega )`$ is proportional to $`1/\mathrm{ln}^2(E/T_K)`$, in agreement with Ref. Abrikosov1965, . More interestingly, the electron scattering is inelastic, and the dependence of $`\sigma (E,\omega )`$ on the energy transfer, $`\omega `$, is determined by the spin-spin correlation function of the impurity or, equivalently, by the dissipative part of the impurity spin susceptibility, $`\chi ^{\prime \prime }`$. In the absence of magnetic field, the elastic component of scattering appears only in order $`1/\mathrm{ln}^4(E/T_K)`$.
Our findings confirm and quantify the conclusion of Ref. Natan, regarding the inelastic nature of Kondo scattering, and also provide a clear physical picture of the mechanism of inelastic scattering. In the absence of magnetic field, the inelastic scattering cross-section is parametrically larger than the elastic one. The typical energy transfer $`|\omega |`$ in an inelastic scattering event is however small compared to $`E`$. At high temperatures, $`TT_K`$, the characteristic energy transfer is determined by the Korringa relaxation rate of the magnetic impurity, and at low temperatures, it is defined by the value of $`T_K`$. In the high-temperature limit, the cross-section is maximal at $`\omega =0`$, and at $`TT_K`$ it reaches its maximum at $`\omega T_K`$. The decrease of the cross-section in the domain $`\omega T_K`$ is remarkably slow, $`\sigma (E,\omega )[\omega \mathrm{ln}^2(\omega /T_K)]^1`$. The domain of intermediate energy transfers, $`\omega T_K`$, is covered by NRG calculations. The numerical results fit well with the analytically evaluated asymptotes at $`\omega T_K`$ and $`\omega T_K`$. In the presence of an external magnetic field, the Zeeman splitting of the magnetic impurity levels results in the appearance of an elastic component of electron scattering already in the leading logarithmic order (in $`E/T_K`$).
Finally, we proposed possible hot-electron experiments with a metallic mesoscopic wire and with a semiconductor quantum-dot device which in principle allow one to access the differential scattering cross-section of a localized magnetic moment.
###### Acknowledgements.
This work was supported by NSF Grants DMR02-37296 and EIA02-10736 (L.G.), by the Deutsche Forschungsgemeinschaft (DFG) grant GA 1072/1-1 (M.G.), the DFG ”Center for Functional Nanostructures” (P.W.), DFG SFB631 and the European ”Spintronics” RTN (J.v.D.), and by Hungarian Grants OTKA D048665, T048782, T046303 (L.B.). L.B. is a grantee of the Bolyai Janos Scholarship.
## Appendix A Derivation of Shiba relation
Here we provide a simple derivation of Shiba relation Shiba , using Nozières’ idea of a low-temperature Fermi liquid description of the Kondo problem.
Within Nozières’ theory, at $`TT_K`$ the effect of a weak ($`g\mu _BBT_K`$) magnetic field applied to a Kondo impurity is described by a local-field Hamiltonian,
$$H_B=\frac{2\chi _0}{g\mu _B\nu }B\underset{k,k^{},\sigma }{}s_{\sigma \sigma }^z\psi _{k\sigma }^{}\psi _{k^{}\sigma }.$$
(60)
Here $`\nu `$ is the density of states at the Fermi level, $`\chi _0=[W(g\mu _B)^2]/(4T_K)`$ is the linear susceptibility, summation over $`k`$ and $`k^{}`$ occurs within a shell of states $`\mathrm{\Delta }k`$ sufficiently close to the Fermi level ($`\mathrm{\Delta }kT_K/v_F`$ with $`v_F`$ being the Fermi velocity), and field $`B`$ is applied along the $`z`$-axis. One may easily check that the action of the field described by the Hamiltonian (60) indeed results in a local magnetization $`M=\chi _0B`$. For that one starts with the evaluation of the spin-dependent scattering phase $`\delta _\sigma `$ off the local perturbation Eq. (60) using the Born approximation,
$$\delta _\sigma =\pi \sigma \nu \frac{\chi _0}{g\mu _B\nu }B,\sigma =\pm 1.$$
(61)
Having the phase difference $`\delta _+\delta _{}`$, we evaluate the magnetization using the Friedel sum rule,
$$M=\frac{g\mu _B}{2}\frac{\delta _+\delta _{}}{\pi }=\chi _0B.$$
(62)
Having the right form of the local perturbation, we now allow for a slow variation of the field, $`B=B_0\mathrm{cos}(\omega t)`$, assuming that the frequency $`\omega T_K`$. Next we evaluate the energy absorption rate $`w`$ caused by such time-dependent perturbation. Using the Fermi Golden rule, we arrive at
$`w`$ $`=\pi \omega \left[{\displaystyle \frac{\chi _0B_0}{g\mu _B\nu }}\right]^2\nu ^2{\displaystyle 𝑑ϵf(ϵ)[f(ϵ\omega )f(ϵ+\omega )]}`$
$`=\pi \omega ^2\left[{\displaystyle \frac{\chi _0B_0}{g\mu _B}}\right]^2.`$ (63)
In the last line, we discarded corrections of order $`𝒪(e^{T_K/T})`$ arising from the boundaries of the energy integral. Recalling finally that $`w=\frac{1}{2}\omega \chi ^{\prime \prime }(\omega )B_0^2`$, we arrive at the Shiba relation Eq. (29).
Using the framework of the above derivation, it is straightforward to generalize the Shiba relation to the case of a weak slowly varying field applied to the local moment on top of a time-independent field $`B`$ of arbitrary strength. In the generalized relation, $`\chi _0`$ is the static differential susceptibility, and the relation is applicable in the regime $`\omega ,T\mathrm{max}\{B,T_K\}`$.
We assume that the basis of the effective low-energy Hamiltonian has been chosen such that it incorporates already the effect of the time-independent local magnetic field $`B`$. Consider now a small perturbation to this effective Hamiltonian induced by a small change in the applied local magnetic field $`B+\delta B`$,
$$H_{\delta B}=\frac{2}{g\mu _B\nu }\frac{M}{B}\delta B\underset{k,k^{},\sigma }{}s_{\sigma \sigma }^zc_{k\sigma }^{}c_{k^{}\sigma }.$$
(64)
The summation over $`k`$ and $`k^{}`$ is bounded by $`|k|,|k^{}|\mathrm{max}\{g\mu _BB,T_K\}/v_F`$. The prefactor can be determined in the same way as before. In contrast to the limit $`B=0`$, here the resulting phase shift yields information about the change in magnetization $`M(B+\delta B)M(B)=(M/B)\delta B`$, where $`M/B`$ is the differential susceptibility. The same arguments as above will yield the generalized Shiba relation
$$\chi _{zz}^{\prime \prime }(\omega )=2\pi \omega \left[\frac{S^z}{B}\right]^2$$
(65)
Here $`S^z`$ is the equilibrium average spin value in the presence of field $`B`$. In the perturbative regime, the average is given in Eq. (38).
In Fig. 8 the prediction of the generalized Shiba relation (65) is illustrated with the NRG result. The susceptibility, $`\chi _{zz}^{\prime \prime }`$, and the non-linear static susceptibility, $`\chi _0=g\mu _BS^z/B`$, have been independently evaluated with the NRG. The dashed line in Fig. 8 is plotted with the help of Eq. (65) and compares well with the low-frequency asymptote of $`\chi _{zz}^{\prime \prime }`$.
## Appendix B RG equation for the impurity $`g`$-factor
We present a derivation of the two-loop RG equation for the impurity $`g`$-factor of the Kondo model and, in particular, explain the different roles played by the impurity and conduction electron $`g`$-factor in the renormalization process. To this end, we will use Abrikosov’s pseudo-fermion representationAbrikosov1965 for the impurity spin, $`𝐒=f^{}\frac{1}{2}𝝈f`$, where $`f^{}=(f_{}^{},f_{}^{})`$ in a compact spinor notation and $`𝝈`$ is the vector of Pauli matrices. We will need the action of the Kondo model, which consists of three parts, $`𝒮=𝒮_s+𝒮_d+𝒮_K`$. The quadratic part of the Abrikosov pseudo-fermions reads
$`𝒮_d`$ $`={\displaystyle _0^\beta }𝑑\tau f^{}(\tau )\left[_\tau \lambda _0g\mu _B{\displaystyle \frac{1}{2}}𝝈^aB_a\right]f(\tau ),`$ (66)
where $`g`$ is the impurity $`g`$-factor and $`B^a`$ is the magnetic field, which is taken to point in the z-direction $`B^a=B\delta _{az}`$. In order to enforce the Hilbert space constraint, $`f^{}f=1`$, a chemical potential, $`\lambda _0\mathrm{}`$, is introduced Abrikosov1965 . The Kondo interaction is given by
$`𝒮_K`$ $`={\displaystyle _0^\beta }𝑑\tau \left(\mathrm{\Psi }^{}(\tau ){\displaystyle \frac{1}{2}}𝝈^a\mathrm{\Psi }(\tau )\right)J_{ab}\left(f^{}(\tau ){\displaystyle \frac{1}{2}}𝝈^bf(\tau )\right),`$ (67)
where the local electron operator at the impurity site is $`\mathrm{\Psi }^{}=\frac{dk}{2\pi v_\mathrm{F}}(c_k^{},c_k^{})`$. We allow for different values of the exchange interaction in the direction orthogonal and perpendicular to the magnetic field, $`(J_{ab})=`$diag$`\{J_{},J_{},J_{}\}`$. Finally, the quadratic part of the s-electrons reads
$`𝒮_s={\displaystyle _0^\beta }𝑑\tau {\displaystyle \underset{D}{\overset{D}{}}}{\displaystyle \frac{dk}{2\pi v_\mathrm{F}}}c_{k\sigma }^{}(\tau )\left[_\tau +k\right]c_{k\sigma }^{}(\tau ).`$ (68)
In the presence of a Zeeman energy for the s-electrons, the Fermi sea of the spin-up and -down electrons are shifted with respect to each other giving rise to a finite Pauli magnetization. In Eq. (68) we assumed that the band has already been symmetrized around the respective Fermi energies by integrating out a finite number of electronic degrees of freedom. This process results in a perturbative renormalization of the impurity $`g`$-factor due to the so-called Knight shift. The first-order Knight-shift diagram is shown in Fig. 9a. The $`g`$-factor, $`g`$, appearing in (66) is therefore understood to be already the Knight-shifted impurity $`g`$-factor,
$`g=g_i{\displaystyle \frac{J_{}\nu }{2}}g_e+𝒪(J_{}\nu )^2,`$ (69)
where $`g_i`$ and $`g_e`$ are the bare impurity- and electronic $`g`$-factors, respectively, and the density of states is $`\nu =1/(2\pi v_\mathrm{F})`$. As is clear from Eq. (69) the electronic $`g_e`$ can be absorbed in an effective impurity $`g`$-factor. The Knight shift is thus only a perturbative phenomenon and, in particular, is not enhanced by logarithmic renormalizations. This is expected since the Pauli magnetization affects only electronic states far away from the Fermi edge deep inside the Fermi sea.
The field theory can be renormalized ZinnJustin with a wave-function, impurity $`g`$-factor and Kondo-coupling renormalization (in addition to a counterterm absorbing a shift in the unphysical chemical potential $`\lambda _0`$),
$`f=\sqrt{Z}f^R,g={\displaystyle \frac{g^R}{Z}},J_{ab}={\displaystyle \frac{J_{ab}^R}{Z}}.`$ (70)
We compute the renormalization of the Kondo coupling to one-loop and the renormalization of the wave-function and $`g`$-factor to two-loop order. The corresponding diagrams are shown in Fig. 9b and c. The resulting RG equations for the Kondo vertex are the well-known poor man’s scaling equations Anderson
$`{\displaystyle \frac{d(J_{}\nu )}{d\mathrm{ln}D}}`$ $`=(J_{}\nu )(J_{}\nu ),{\displaystyle \frac{d(J_{}\nu )}{d\mathrm{ln}D}}=(J_{}\nu )^2.`$ (71)
For the wave-function renormalization we obtain,
$`{\displaystyle \frac{d\mathrm{ln}Z}{d\mathrm{ln}D}}`$ $`={\displaystyle \frac{1}{8}}\left((J_{}\nu )^2+2(J_{}\nu )^2\right).`$ (72)
Finally, the main result is the RG equation for the $`g`$-factor,
$`{\displaystyle \frac{dg}{d\mathrm{ln}D}}`$ $`={\displaystyle \frac{1}{2}}g(J_{}\nu )^2.`$ (73)
Solving this equation in the isotropic case, $`J_{}=J_{}=J`$, and expanding the result in leading logarithmic order we get
$`g(D)=g_i\left(1{\displaystyle \frac{1}{2\mathrm{ln}\frac{D}{T_K}}}+\left(1{\displaystyle \frac{g_e}{g_i}}\right){\displaystyle \frac{J\nu }{2}}\right)`$ (74)
where we already substituted the Knight-shifted $`g`$-factor (69). In the scaling limit, $`J0`$ while $`T_K`$ is held fixed, any dependence on the electronic $`g`$-factor, $`g_e`$, vanishes, and we obtain the result cited in the body of the paper, Eq. (34). In particular, note that in the absence of a Knight shift, $`g_e=0`$, the perturbative correction to the $`g`$-factor starts only in second order in the exchange coupling $`J`$ but, nevertheless, after renormalization group improvement leads to a correction which is of leading logarithmic order. At zero temperature the RG equation for the $`g`$-factor coincides with the RG equation for the impurity magnetization, which was already determined in Ref. Abrikosov70, . |
warning/0507/hep-th0507167.html | ar5iv | text | # 1 Introduction
## 1 Introduction
It is generally believed that in the very early universe, at times smaller than the Planck time, quantum gravity effects play a crucial role. Since the physics in the Planck era prepares the initial conditions for the subsequent classical evolution of the universe and, as a result, is likely to be responsible for many of its features we observe today it is clearly very desirable to gain some understanding of the quantum gravitational processes which took place immediately after the Big Bang.
The construction of a consistent quantum theory of gravity which is indispensable for investigations of this kind is a major challenge, of course. In trying to set up such a theory the first question to ask is what are the degrees of freedom which properly describe the gravitational field at the quantum level. In view of the successes of classical general relativity (GR) the most natural working hypothesis is that they are given by the gauge invariant content of the metric tensor $`g_{\mu \nu }`$. Then, trying to construct a quantum field theory based upon $`g_{\mu \nu }`$, a second question arises: what is the (bare) action describing its dynamics? The obvious first option to try is the Einstein-Hilbert action underlying classical general relativity, but it is well-known that its perturbative quantization leads to a non-renormalizable theory. The next logical possibility is to stay within perturbation theory but to use a different action. For example, one could try to add terms quadratic in the curvature to the Einstein-Hilbert term; this yields a theory which is indeed perturbatively renormalizable but nevertheless has to be discarded because of unitarity problems.
Thus it seems likely that a satisfactory quantum field theory of the continuum metric, if it exists, requires us to abandon both the classical action and perturbation theory. Weinberg’s “asymptotic safety” scenario is an approach in precisely this category<sup>1</sup><sup>1</sup>1Approaches which (typically) retain the Einstein-Hilbert form of the action but instead abandon the traditional continuum $`g_{\mu \nu }`$ include loop gravity and random triangulations .. Here the basic idea is to define, non-perturbatively, a theory which is based upon a very special bare action, namely one which is infinitesimally close to a fixed point $`\mathrm{\Gamma }_{}[g_{\mu \nu }]`$ of the Wilsonian renormalization group (RG). If it exists, the fixed point allows us to take the limit of an infinite ultraviolet (UV) cutoff in a controlled, or “safe” way.
The theory thus constructed will be referred to as Quantum Einstein Gravity or “QEG”. Using the technique of the exact RG equations in the continuum \- a considerable amount of evidence has been collected by now - which indicates that an appropriate fixed point is indeed likely to exist on the “theory space” consisting of the diffeomorphism invariant functionals of $`g_{\mu \nu }`$. Practical calculations typically involve a truncation of this theory space. A conceptually quite different approximation leading to similar conclusions is the (exact) quantization of the restricted class of metrics admitting two Killing vectors .
Most of the investigations using the effective average action formalism \- for implementing the RG “coarse graining” employ the so-called Einstein-Hilbert truncation which retains only Newton’s constant $`G(k)`$ and the cosmological constant $`\mathrm{\Lambda }(k)`$ as running parameters. Here $`k`$ denotes the variable infrared (IR) cutoff, the momentum scale down to which the quantum fluctuations of the metric are integrated out. If one introduces the dimensionless couplings $`g(k)k^2G(k)`$ and $`\lambda (k)\mathrm{\Lambda }(k)/k^2`$ the flow equations in this truncation consist of an autonomous system of two coupled ordinary differential equations:
$$k_kg=\beta _g(g,\lambda ),k_k\lambda =\beta _\lambda (g,\lambda ).$$
(1.1)
The beta-functions $`\beta _g`$ and $`\beta _\lambda `$ were first derived in , and the resulting equations were studied numerically in ref. . The flow on the $`g`$-$`\lambda `$-plane displays two fixed points: a Gaussian fixed point (GFP) at the origin, and a non-Gaussian fixed point (NGFP) at $`(g_{},\lambda _{})`$ with $`g_{}`$ and $`\lambda _{}`$ both positive. At least according to what can be said on the basis of the Einstein-Hilbert truncation, this NGFP has all the properties necessary for asymptotic safety.
Ref. contains a complete classification of all types of RG trajectories occurring in the Einstein-Hilbert truncation. Particularly important are those of Type Ia, IIa, and IIIa which, when $`k`$ is lowered, run towards negative, vanishing, and positive values of the cosmological constant, respectively. In ref. the very special trajectory which is realized in Nature has been identified and its parameters were determined. This trajectory is of Type IIIa; it is sketched schematically in fig. 1.
For $`k\mathrm{}`$ it starts infinitesimally close to the NGFP. Then, lowering $`k`$, the trajectory spirals about the NGFP and approaches the “separatrix”, the distinguished trajectory which ends at the GFP. It runs almost parallel to the separatrix for a very long “RG time”; only in the “very last moment” before reaching the GFP, at the turning point T, it gets driven away towards larger values of $`\lambda `$. In fig. 1 the points P<sub>1</sub> and P<sub>2</sub> symbolize the beginning and the end of the regime in which classical general relativity is valid (“GR regime”). This section of the trajectory is virtually identical to a canonical one for which $`G`$ and $`\mathrm{\Lambda }`$ are $`k`$-independent. The classical regime starts soon after the turning point T which is passed at the scale $`k_\mathrm{T}10^{30}m_{\mathrm{Pl}}`$ corresponding to the (macroscopic!) length $`k_\mathrm{T}^110^3`$ cm.<sup>2</sup><sup>2</sup>2Since it is only the cosmological constant which shows a significant running between $`k=m_{\mathrm{Pl}}`$ and $`k=k_\mathrm{T}`$, and since $`\mathrm{\Lambda }`$ cannot probably be measured in a millimeter-size laboratory experiment, one nevertheless would not expect to see violations of classical general relativity immediately above $`k_\mathrm{T}`$ .
In ref. it was argued that there starts a regime of strong IR renormalization effects to the right of the point P<sub>2</sub> which might become visible at astrophysical and cosmological length scales. In fact, within the Einstein-Hilbert approximation, trajectories of Type IIIa cannot be continued to the extreme IR ($`k0`$). They terminate at a non-zero value of $`k`$ as soon as the trajectory reaches $`\lambda =1/2`$. (Close to the question mark in fig. 1.) Before it starts becoming invalid and has to be replaced by a more precise treatment, the Einstein-Hilbert approximation suggests that $`G`$ will increase, while $`\mathrm{\Lambda }`$ decreases, as $`\lambda 1/2`$ .
The Type IIIa trajectory of QEG which Nature has selected is highly special or “unnatural” in the following sense. It is fine-tuned in such a way that it gets extremely close to the GFP before “turning left”. In it was shown that the coordinates $`g_\mathrm{T}`$ and $`\lambda _\mathrm{T}`$ of the turning point are both very small: $`g_\mathrm{T}=\lambda _\mathrm{T}10^{60}`$. In the GR regime, $`g`$ decreases from $`g(k)=10^{70}`$ at a typical terrestrial length scale of $`k^1=1`$ meter to $`g(k)=10^{92}`$ at the solar system scale of $`k^1=1`$ astronomical unit. Extrapolating to cosmological scales one finally has $`g(k)=10^{120}`$ when $`k`$ equals the present Hubble constant $`H_0`$.
In this analysis the two free parameters which uniquely characterize a Type IIIa trajectory where derived from the measured values of $`G(k)`$ for laboratory values of $`k`$, and $`\mathrm{\Lambda }(k)`$ at $`kH_0`$. It can be argued that the IR renormalization effects, if they exist, could not change $`G`$ and $`\mathrm{\Lambda }`$ by many orders of magnitude between $`P_2`$ (at solar system scales, say) and cosmological scales.
Indeed, the present Hubble parameter $`k=H_0`$ is approximately the scale where the Einstein-Hilbert trajectory becomes unreliable. The observations indicate that today the cosmological constant is of the order $`H_0^2`$. Interpreting this value as the running $`\mathrm{\Lambda }(k)`$ at the scale $`k=H_0`$, the dimensionless $`\lambda (k)`$, at this scale, is of the order unity: $`\lambda (H_0)\mathrm{\Lambda }(H_0)/H_0^2=𝒪(1)`$. So it is quite precisely near the present Hubble scale where the Einstein-Hilbert truncation becomes insufficient for a description of the trajectory Nature has chosen.
The “unnaturalness” of Nature’s gravitational RG trajectory has an important consequence. Because it gets so extremely close to the GFP it spends a very long RG time in its vicinity because the $`\beta `$-functions are small there. As a result, the termination of the trajectory at $`\lambda =1/2`$ is extremely delayed, by 60 orders of magnitude, compared to a generic trajectory where this happens for $`k`$ near the Planck mass. In ref. it was argued that this non-generic feature of the trajectory is a necessary condition for a long classical regime with $`G,\mathrm{\Lambda }=const`$ to emerge, and any form of classical physics to be applicable.
It was also shown that for any trajectory which actually does admit a long classical regime the cosmological constant is automatically small. Nevertheless, the fine-tuning behind the “unnatural” trajectory Nature has selected is of a much more general kind than the traditional cosmological constant problem : the primary issue is the emergence of a classical space-time; once this is achieved, the extreme smallness of the observed $`\mathrm{\Lambda }`$ (compared to $`m_{\mathrm{Pl}}^2`$) comes for free. (See for a detailed discussion.)
Knowing at least the qualitative features of the RG running of the gravitational parameters one can try to use this information for investigating how quantum gravity effects modify the classical Friedmann-Robertson-Walker (FRW) cosmology. An immediate consequence of an approach of this kind is that the cosmological constant becomes a time-dependent quantity so that in principle it should be possible to understand its (tiny) value today as the result of a dynamical evolution process. In this respect the RG approach has certain features in common with the quintessence models . The main difference is that the evolution is driven by the vacuum fluctuations of the gravitational field itself, and no extra quintessence field needs to be introduced.
In ref. a first application of the RG flow from QEG to the cosmology of the early universe has been described. The tool used there was a kind of gravitational “RG improvement” \- . The running couplings $`G(k)`$ and $`\mathrm{\Lambda }(k)`$ where converted to functions of the cosmological time $`t`$ by identifying the RG scale $`k`$ with $`1/t`$ or, what was the same in this context, the Hubble parameter $`H(t)`$. Then the resulting $`G(t)`$ and $`\mathrm{\Lambda }(t)`$ were inserted into the Einstein equations for a homogeneous and isotropic universe. In the fixed point regime where
$$G(k)=g_{}/k^2,\mathrm{\Lambda }(k)=\lambda _{}k^2,$$
(1.2)
and $`G(t)t^2`$, $`\mathrm{\Lambda }(t)1/t^2`$ it was then possible to find exact analytic solutions for the Robertson-Walker scale factor $`a(t)`$. Provided this improvement of the cosmological evolution equations does indeed encapsulate the leading quantum effects, the fixed point solution should describe the universe at times much earlier than the Planck time since (1.2) is valid only as long as $`km_{\mathrm{Pl}}`$, the Planck mass being the lower boundary of the asymptotic scaling region. A strong argument in favor of the validity of the approach is that gravity becomes weakly coupled for $`t0`$ since $`G(k)`$, vanishing for $`k\mathrm{}`$, is an asymptotically free coupling.
In particular for the equation of state $`p=\rho /3`$ and flat 3-sections ($`K=0`$) the fixed point cosmology has various remarkable properties.<sup>3</sup><sup>3</sup>3In ref. also fixed point solutions for the more general equation of state $`p=w\rho `$ and (pseudo)spherical 3-spaces ($`K=\pm 1`$) were obtained. In a different context similar cosmologies were investigated in . Fixed point cosmologies of the late universe are discussed in . It is completely scale-free; the scale factor and the matter energy density behave as $`a(t)t`$ and $`\rho (t)1/t^4`$, respectively. For every RG trajectory interpolating between the fixed point behavior (1.2) and the classical $`G,\mathrm{\Lambda }=const`$, the fixed point solution is a universal attractor in the space of improved FRW cosmologies in the sense that they all approach the fixed point solution for $`t0`$. Stated the other way around: the fixed point cosmology prepares the initial conditions for the subsequent cosmological evolution which is essentially classical. Those initial conditions are such that the energy density today equals precisely the critical one, $`\rho _{\mathrm{crit}}`$. No fine-tuning is necessary here, only a discrete choice, picking flat rather than (pseudo)spherical 3-spaces . Furthermore, as a result of the linear expansion $`a(t)t`$ at early times, the RG improved cosmology has no particle horizon. This is another feature which could be of phenomenological importance.
Up to now the solution to the coupled system of RG- and cosmological evolution equations is known only in the fixed point regime. The purpose of the present paper is to analyze the complete solution all the way from the Big Bang to scales of the order of the present Hubble constant. Using mostly numerical methods we shall discuss in detail whether or not the Einstein-Hilbert truncation is sufficient in order to connect the UV fixed point regime (Planck era) to the present era of the universe.
The remaining sections of this paper are organized as follows. In Section 2 we briefly review some properties of the coupled RG- and cosmological evolution equation, point out a conceptual difficulty related to the fact that this system is overdetermined for any “rigid” cutoff identification, and propose a solution to this problem. Then, in Section 3, we present the RG equations to be used in the following. Section 4 contains an investigation of all those properties of the RG improved cosmology which do not depend on the detailed form of the RG trajectory but rather reflect certain general properties of the “theory space” on which the RG flow takes place. In Section 5 we explore the actual cosmological time evolution; in doing so we advocate the point of view that this evolution is primarily an evolution with respect to the scale $`k`$. Section 6 contains a brief summary and discussion of the results.
## 2 RG improved Einstein’s equation
Following we investigate homogeneous and isotropic cosmologies described by a standard Robertson-Walker metric containing the scale factor $`a(t)`$ and the parameter $`K=0,\pm 1`$ which distinguishes the three possible types of maximally symmetric 3-spaces of constant cosmological time $`t`$. The dynamics is governed by Einstein’s equation $`G_{\mu \nu }=\mathrm{\Lambda }g_{\mu \nu }+8\pi GT_{\mu \nu }`$ where $`T_\mu ^\nu \mathrm{diag}[\rho ,p,p,p]`$ is a conserved energy momentum tensor for which the equation of state $`p(t)=w\rho (t)`$ with an arbitrary constant $`w>1`$ is assumed.
To start with, we assume that we have fixed a certain “cutoff identification” $`k=k(t)`$. Then every RG trajectory $`k(G(k),\mathrm{\Lambda }(k))`$ obtained by solving the RG equations gives rise to time dependent functions $`G(t)G(k=k(t))`$ and $`\mathrm{\Lambda }(t)\mathrm{\Lambda }(k=k(t))`$. Replacing the constants $`G`$ and $`\mathrm{\Lambda }`$ in Einstein’s equation with these functions, and specializing for a Robertson-Walker metric, the field equations boil down to
$`\left({\displaystyle \frac{\dot{a}}{a}}\right)^2+{\displaystyle \frac{K}{a^2}}={\displaystyle \frac{1}{3}}\mathrm{\Lambda }+{\displaystyle \frac{8\pi }{3}}G\rho ,`$ (2.1a)
$`\dot{\rho }+3(1+w){\displaystyle \frac{\dot{a}}{a}}\rho =0,`$ (2.1b)
$`\dot{\mathrm{\Lambda }}+8\pi \rho \dot{G}=0.`$ (2.1c)
Here the “dot” denotes the derivative with respect to the cosmological time $`t`$. Eq. (2.1a) is the familiar Friedmann equation, albeit with a time dependent $`G`$ and $`\mathrm{\Lambda }`$, and eq. (2.1b) is the conservation law $`D_\mu T_\nu ^\mu =0`$. Eq. (2.1c) represents an additional consistency condition which is necessary for the integrability of Einstein’s equation. If (2.1c) holds true, its RHS has a vanishing covariant divergence, as has its LHS by virtue of Bianchi’s identity.
For $`G(t)`$ and $`\mathrm{\Lambda }(t)`$ fixed, the equations (2.1a,b,c) constitute a system of 3 equations for only two unknowns, $`a(t)`$ and $`\rho (t)`$. In general this system is overdetermined and, for generic $`G(t)`$ and $`\mathrm{\Lambda }(t)`$, will not admit any solution for $`a(t)`$ and $`\rho (t)`$. This can be seen as follows. The time dependence of $`\rho `$ follows directly from (2.1c),
$$\rho (t)=\frac{1}{8\pi }\frac{\dot{\mathrm{\Lambda }}(t)}{\dot{G}(t)},$$
(2.2)
and integrating (2.1b) yields $`\rho a^{3+3w}=/(8\pi )`$, with $``$ a constant of integration. Combining this relation with (2.2) we find the time dependence of the scale factor:
$$a(t)=\left[\frac{\dot{G}(t)}{\dot{\mathrm{\Lambda }}(t)}\right]^{1/(3+3w)}.$$
(2.3)
Already at this point we know $`a(t)`$ and $`\rho (t)`$, but we have not used the improved Friedmann equation (2.1a) yet. Checking whether (2.2) and (2.3) also solve (2.1a), one usually discovers that this is not the case: the three equations (2.1a,b,c) are not integrable for generic $`G(t)`$ and $`\mathrm{\Lambda }(t)`$.
At this point there are two logical possibilities: either the improvement of the field equations is not an adequate way of exploiting the RG information, or one can ultimately arrive at a description which is self-consistent within this framework by modifying $`G(t)`$ and $`\mathrm{\Lambda }(t)`$ in such a way that all three equations (2.1a,b,c) are satisfied. As for the second possibility, there are two obvious ways of modifying the time dependence of $`G`$ and $`\mathrm{\Lambda }`$. On can change the cutoff scheme , i.e., the operator $`_k(D^2)`$ which suppresses the “slow” modes in the path integral, thus slightly changing the trajectory $`k(G(k),\mathrm{\Lambda }(k))`$, or one can change the “cutoff identification” $`k=k(t)`$. The details of both the trajectory<sup>4</sup><sup>4</sup>4The critical exponents of the fixed points, for instance, are universal (i.e., $`_k`$-independent) features of the RG flow, but not the precise shape of the trajectories. and the cutoff identification are non-universal, i.e., unphysical, but in combining $`G(k)`$ with $`k=k(t)`$ to obtain $`G(t)G(k=k(t))`$ the non-universalities can cancel to some extent,<sup>5</sup><sup>5</sup>5See ref. for a particularly transparent example in a black hole context. provided $`k(t)`$ is chosen appropriately.
The length scale set by the IR cutoff<sup>6</sup><sup>6</sup>6See ref. for a more detailed discussion., $`\mathrm{}k^1`$, can be visualized as the variable resolution of the “microscope” with which space-time is observed since the effective average action $`\mathrm{\Gamma }_k`$ describes the dynamics of fields averaged over volumes of linear extension $`\mathrm{}`$ . In cosmological applications it is sensible to readjust the resolution of the microscope as the universe becomes larger. In which way precisely $`\mathrm{}`$ is increased as the scale factor grows depends on the interpretation one wants to give to the coarse grained picture of the expanding universe. In the ansatz $`k1/t`$ was used, motivated by the fact that when the age of the universe is $`t`$, no process with frequency smaller than $`1/t`$ can have occurred yet. Also the Hubble scale $`kH(t)`$ would be a natural choice since in cosmology the Hubble length $`\mathrm{}_H1/H(t)`$ measures the size of the “Einstein elevator” outside which curvature effects become appreciable.
In the present paper we shall adopt the following strategy for choosing $`k(t)`$. In order to circumvent the problem of the overdetermined equations (2.1a,b,c) we shall not a priori fix the cutoff identification $`k=k(t)`$ in a rigid way, but rather derive it from the field equations themselves, imposing the condition that those equations should be integrable. On the RG side, we shall stick to a fixed cutoff scheme. Then, choosing initial conditions for the trajectory, we obtain uniquely defined functions $`G(k)`$ and $`\mathrm{\Lambda }(k)`$. Next it is important to note that, given the $`k`$-dependence of $`G`$ and $`\mathrm{\Lambda }`$, we immediately know $`a`$ and $`\rho `$ as functions of $`k`$:
$`a(t(k))`$ $`=\left[{\displaystyle \frac{G^{}(k)}{\mathrm{\Lambda }^{}(k)}}\right]^{1/(3+3w)},`$ (2.4a)
$`\rho (t(k))`$ $`={\displaystyle \frac{1}{8\pi }}{\displaystyle \frac{\mathrm{\Lambda }^{}(k)}{G^{}(k)}}.`$ (2.4b)
On the LHS of these relations $`t(k)`$ is the functional inverse of the cutoff identification $`k=k(t)`$. The eqs. (2.4a,b) follow from (2.2) and (2.3), respectively, since $`\dot{G}=\left(dk/dt\right)G^{}`$, $`\dot{\mathrm{\Lambda }}=\left(dk/dt\right)\mathrm{\Lambda }^{}`$, and the derivative of $`k(t)`$ drops out from the ratio $`\dot{G}/\dot{\mathrm{\Lambda }}`$. (Here and in the following the prime denotes a derivative with respect to $`k`$.)
With (2.4a,b) the second and the third of the eqs. (2.1a,b,c) are satisfied. We can now try to make the whole system consistent by allowing the Friedmann equation (2.1a) to fix the relationship between $`k`$ and $`t`$. Inserting (2.4a,b) into eq. (2.1a) the latter yields the following differential equation<sup>7</sup><sup>7</sup>7Here we have chosen the sign of the square root such that $`t`$ increases with decreasing $`k`$, i.e., late cosmological times will be associated with small $`k`$-values on the RG-trajectory. for $`k=k(t)`$
$$\frac{dk}{dt}=\frac{a}{a^{}}\left[\frac{\mathrm{\Lambda }}{3}+\frac{8\pi }{3}G\rho \right]^{1/2},$$
(2.5)
which can be solved (numerically) for any given trajectory.<sup>8</sup><sup>8</sup>8A similar strategy was used in ref. . The RHS of (2.5) is completely specified in terms of RG data: $`G(k)`$ and $`\mathrm{\Lambda }(k)`$ obtained directly from the trajectory, and $`a`$ and $`\rho `$ are given by (2.4a,b).
This method provides us with a solution $`\{a(t),\rho (t),k(t)\}`$ for any trajectory $`(G(k),\mathrm{\Lambda }(k))`$, but it is not clear a priori whether this solution is physically sensible. It can have a meaningful interpretation only if the continuous readjustment of the “microscope’s” resolution described by $`k=k(t)`$ is correlated with the expansion of the universe in a transparent and, in particular, monotonic way. This will indeed turn out to be the case: We shall see that the function $`k(t)`$ obtained dynamically is reasonably close to $`k(t)H(t)`$ during all epochs of the history of the universe.
## 3 The RG equations
The RG equations of QEG in the Einstein-Hilbert truncation are given by (1.1). The pertinent theory space is the $`g`$-$`\lambda `$-plane coordinatized by the dimensionless couplings $`g`$ and $`\lambda `$. The (dimensionless, $`k`$-independent) beta-functions $`\beta _g`$ and $`\beta _\lambda `$ are the components of a vector field on the theory space. In ref. they were obtained for $`d`$ space-time dimensions.
In $`d`$ dimensions, the dimensionful couplings are given by $`G(k)k^{2d}g(k)`$ and $`\mathrm{\Lambda }(k)k^2\lambda (k)`$. Their $`k`$-dependence is governed by the system of equations
$$k\frac{d}{dk}G(k)=\beta _G(G,\mathrm{\Lambda },k),k\frac{d}{dk}\mathrm{\Lambda }(k)=\beta _\mathrm{\Lambda }(G,\mathrm{\Lambda },k).$$
(3.1)
The (dimensionful, $`k`$-dependent) beta-functions $`\beta _G`$ and $`\beta _\mathrm{\Lambda }`$ read :
$`\beta _G(G,\mathrm{\Lambda },k)=`$ $`\eta _NG,`$ (3.2a)
$`\beta _\mathrm{\Lambda }(G,\mathrm{\Lambda },k)=`$ $`\eta _N\mathrm{\Lambda }+{\displaystyle \frac{1}{2}}(4\pi )^{1d/2}k^dG[2d(d+1)\mathrm{\Phi }_{d/2}^1(2\mathrm{\Lambda }/k^2)`$ (3.2b)
$`8d\mathrm{\Phi }_{d/2}^1(0)d(d+1)\eta _N\stackrel{~}{\mathrm{\Phi }}_{d/2}^1(2\mathrm{\Lambda }/k^2)].`$
Here $`\eta _N`$ denotes the anomalous dimension of $`\sqrt{g}R`$; it can be expressed as
$$\eta _N=\frac{gB_1(\lambda )}{1gB_2(\lambda )}|_{g=k^{d2}G,\lambda =\mathrm{\Lambda }/k^2}$$
(3.3)
with the abbreviations
$$\begin{array}{cc}\hfill B_1(\lambda )& \frac{1}{3}(4\pi )^{1d/2}[d(d+1)\mathrm{\Phi }_{d/21}^1(2\lambda )6d(d1)\mathrm{\Phi }_{d/2}^2(2\lambda )\hfill \\ & 4d\mathrm{\Phi }_{d/21}^1(0)24\mathrm{\Phi }_{d2}^2(0)],\hfill \\ \hfill B_2(\lambda )& \frac{1}{6}(4\pi )^{1d/2}\left[d(d+1)\stackrel{~}{\mathrm{\Phi }}_{d/21}^1(2\lambda )6d(d1)\stackrel{~}{\mathrm{\Phi }}_{d/2}^2(2\lambda )\right].\hfill \end{array}$$
(3.4)
The “threshold functions” $`\mathrm{\Phi }_n^p`$ and $`\stackrel{~}{\mathrm{\Phi }}_n^p`$ are given by certain integrals which depend on the cutoff scheme, i.e., on $`R_k(p^2)`$. In Appendix A we discuss them for the cutoff schemes used in the present paper: the sharp cutoff , the exponential cutoff and Litim’s optimized cutoff . Note that the system of differential equations (1.1) is an autonomous one, while (3.1) is not.
In this paper we shall solve (3.1) for $`d=4`$ numerically and use the resulting RG trajectories $`(G(k),\mathrm{\Lambda }(k))`$ for the improvement procedure outlined in Section 2. Since the cosmological equations contain matter we should, in principle, use the beta-functions of gravity coupled to the corresponding matter system. Since we are interested in a qualitative understanding only, we shall use the flow equations of pure gravity, however. As in ref. , our analysis is based upon the explicit assumption that the matter fields do not change the gross qualitative features of the pure gravity RG flow. Unless we know with certainty what the matter fields in Nature are we cannot anyhow decide on a theoretical basis whether or not this assumption is really correct.
## 4 Cosmology on theory space
The cosmological evolution with respect to the Robertson-Walker time $`t`$ is related to a scale-evolution via the cutoff identification $`k=k(t)`$. Therefore we may think of the history of the universe as a curve in the truncated theory space, $`k(g(k),\lambda (k))`$.
Remarkably, certain properties of the universe at a given $`t`$ depend only on the point $`(g,\lambda )`$ in theory space where the universe happens to “sit” at that time, but not on $`k`$ or on the form of the trajectory. Examples include the matter and vacuum energy densities (divided by the critical density) and the deceleration parameter.
The present section is devoted to those properties of the RG improved cosmology which are directly related to the theory space (the $`g`$-$`\lambda `$-plane) and can be analyzed without first solving for the RG flow.
There exist two curves on the $`g`$-$`\lambda `$-plane which are important for a qualitative understanding of the improved cosmologies: the “$`\mathrm{\Omega }`$-line” at which $`\beta _\mathrm{\Lambda }=0`$, and a line on which $`\eta _N`$ diverges.
### 4.1 The $`\mathrm{\Omega }`$-line
By combining eqs. (2.4b) and (3.1) we can express the matter energy density directly in terms of beta-functions:
$$\rho (t(k))=\frac{1}{8\pi }\frac{\beta _\mathrm{\Lambda }(G(k),\mathrm{\Lambda }(k),k)}{\beta _G(G(k),\mathrm{\Lambda }(k),k)}.$$
(4.1)
Let us apply this formula to the class of cosmologies defined by the following properties: (i) They are not re-contracting, i.e., at least for $`t\mathrm{}`$ their scale factor $`a(t)`$ increases monotonically with $`t`$, and $`\rho 1/a^{3+3w}`$ decreases correspondingly. (ii) The beta-function $`\beta _G`$ does not diverge during the evolution. We shall argue later on that the cosmology realized in Nature is likely to belong to this class.
For these cosmologies we have $`a(t\mathrm{})\mathrm{}`$ and $`\rho (t\mathrm{})0`$. Hence their formal endpoints have $`\rho =0`$ at “$`t=\mathrm{}`$” due to an eternal dilution of matter. In view of eq. (4.1) this means that $`\beta _\mathrm{\Lambda }`$ vanishes for $`t\mathrm{}`$ since, by assumption, $`\beta _G`$ is always finite:
$$\underset{t\mathrm{}}{lim}\beta _\mathrm{\Lambda }(G(k(t)),\mathrm{\Lambda }(k(t)),k(t))=0.$$
(4.2)
In order to analyze the implication of this condition we first consider the more general equation
$$\beta _\mathrm{\Lambda }(G(k),\mathrm{\Lambda }(k),k)=0$$
(4.3)
where no cutoff identification is invoked and $`k`$ is considered the independent variable. Remarkably, when re-expressed in terms of the dimensionless couplings $`gk^2G`$ and $`\lambda \mathrm{\Lambda }/k^2`$, eq. (4.3) becomes $`k`$-independent. (Here and in the following we specialize to $`d=4`$.) Using eq. (3.2b), we see that $`\beta _\mathrm{\Lambda }=0`$ is equivalent to
$$\eta _N(g,\lambda )\lambda +\frac{1}{2\pi }g\left[10\mathrm{\Phi }_2^1(2\lambda )8\mathrm{\Phi }_2^1(0)5\eta _N\stackrel{~}{\mathrm{\Phi }}_2^1(2\lambda )\right]=0.$$
(4.4)
This provides a condition on $`g`$ and $`\lambda `$. It defines a curve on the $`g`$-$`\lambda `$-plane which we shall refer to as the “$`\mathrm{\Omega }`$-line”. It is the locus of all possible endpoints (“$`\mathrm{\Omega }`$-points”) for the RG trajectories $`k(g(k),\lambda (k))`$ belonging to the cosmologies with eternal expansion and $`\beta _G<\mathrm{}`$.
What will turn out crucial for the cosmologies of this class is that the $`\mathrm{\Omega }`$-line coincides by no means with the boundary of the theory space, but is a well defined, albeit scheme dependent curve which lies at least partly in its interior. In fig. 2 it is plotted for the three cutoffs discussed in Appendix A.
Let us assume we apply the method outlined in Section 2 to a RG trajectory of Type IIIa such as the one depicted in fig. 1. Then it can happen that even after an infinite cosmological time the RG trajectory on the $`g`$-$`\lambda `$-plane does not reach the regime of possibly strong IR effects ($`\lambda 1/2`$) but rather stops at some point on the $`\mathrm{\Omega }`$-line. In this case the $`\mathrm{\Omega }`$-line is “screening” the region of strong IR renormalizations. We shall see that, because of this mechanism, there are cosmologies which would never experience the conjectured strong IR quantum effects and remain essentially classical for $`t\mathrm{}`$.
However, the situation is somewhat involved since for most cutoff schemes the actual boundary of theory space is not precisely the straight line at $`\lambda =1/2`$, but rather a complicated curve slightly left of it on which $`\eta _N`$ diverges already. (The sharp cutoff is an exception in the sense that $`\eta _N`$ diverges at $`\lambda =1/2`$ only.) Therefore, in order to find out whether there can be a cosmological era with non-trivial IR effects, we must know the relative position of the $`\mathrm{\Omega }`$-line and the boundary line with $`|\eta _N|=\mathrm{}`$. This will be the topic of the next subsection.
### 4.2 Diverging anomalous dimension
Eqs. (3.2a,b) show that $`G`$ and $`\mathrm{\Lambda }`$ are strongly renormalized when the anomalous dimension $`\eta _N`$ is large. This happens close to certain curves on theory space on which $`\eta _N`$ diverges. As long as $`|\eta _N|`$ is only moderately large we tend to believe the predictions of the Einstein-Hilbert truncation.
The anomalous dimension (3.3) can diverge in two ways:
1. The numerator of the RHS of (3.3) can diverge, $`gB_1(\lambda )\mathrm{}`$. This happens only along the line $`\lambda =1/2`$.
2. The denominator on the RHS of (3.3) can vanish, $`1gB_2(\lambda )=0`$.
Analyzing the second possibility for the cutoff schemes from Appendix A we find that, for the sharp cutoff, this condition is never fulfilled for $`g0`$. For the exponential and the optimized cutoff, however, it is satisfied along a curve on the $`g`$-$`\lambda `$-plane which lies to the left of $`\lambda =1/2`$ (for $`g0`$). This line can be found numerically; it is displayed in fig. 3.
As a result the boundary of the theory space is the straight line $`\lambda =1/2`$ for the sharp cutoff, and the curved lines of fig. 3 for the other two cutoffs. On these boundaries, $`\eta _N\mathrm{}`$, and no RG trajectory can be integrated beyond them. (See also .)
In fig. 3 the solid lines correspond to the locus where $`\eta _N`$ diverges, while the dashed lines indicate the position of the $`\mathrm{\Omega }`$-line. For the Type IIIa trajectories coming from the left, the $`\mathrm{\Omega }`$-line generically screens the singularity at $`\lambda =1/2`$, but it does not provide a general screening mechanism for the other divergences which arise from a vanishing denominator in $`\eta _N`$. We observe that for a certain value $`g=g_{\mathrm{int}}`$ the $`\mathrm{\Omega }`$-line and the locus of $`\eta _N`$-singularities intersect. If $`g>g_{\mathrm{int}}`$ the $`\eta _N`$-singularities occur at smaller values of $`\lambda `$ than the $`\mathrm{\Omega }`$-line. In this regime we find no screening of the $`\eta _N`$-singularity.
For $`g<g_{\mathrm{int}}`$, however, the situation reverses and the $`\mathrm{\Omega }`$-line occurs at smaller values of $`\lambda `$ than the $`\eta _N`$ singularity. In this case, both types of divergences in $`\eta _N`$, the ones coming from $`\lambda =1/2`$ and the vanishing of the denominator in $`\eta _N`$ are shielded for all three cutoff schemes. This implies that a Type IIIa trajectory which is sufficiently close to the $`g=0`$-axis in the IR will always be stopped at the $`\mathrm{\Omega }`$-line before the anomalous dimension $`\eta _N`$ diverges, independently of the cutoff scheme employed. In view of the tiny value $`g=𝒪(10^{120})`$ for $`\lambda =𝒪(1/2)`$ this should in particular be the case for the trajectory realized in Nature.
### 4.3 Energy densities and deceleration parameter
In order to study the evolution of the energy densities associated to the matter and the cosmological constant, it is useful to introduce the relative matter and vacuum energy densities:
$$\rho _{\mathrm{crit}}\frac{3H^2}{8\pi G},\mathrm{\Omega }_M\frac{\rho }{\rho _{\mathrm{crit}}}=\frac{1}{3H^2}\frac{\mathrm{\Lambda }^{}}{G^{}/G},\mathrm{\Omega }_\mathrm{\Lambda }\frac{\rho _\mathrm{\Lambda }}{\rho _{\mathrm{crit}}}=\frac{\mathrm{\Lambda }}{3H^2}.$$
(4.5)
We assume a spatially flat universe ($`K=0`$) and therefore have $`\mathrm{\Omega }_{\mathrm{tot}}\mathrm{\Omega }_M+\mathrm{\Omega }_\mathrm{\Lambda }=1`$ as a consequence of the system (2.1a,b,c) . Using the Friedmann equation (2.1a), we can eliminate the Hubble parameter from the equations for $`\mathrm{\Omega }_M`$ and $`\mathrm{\Omega }_\mathrm{\Lambda }`$:
$$\frac{1}{\mathrm{\Omega }_M}=1\frac{G^{}/G}{\mathrm{\Lambda }^{}/\mathrm{\Lambda }},\frac{1}{\mathrm{\Omega }_\mathrm{\Lambda }}=1\frac{\mathrm{\Lambda }^{}/\mathrm{\Lambda }}{G^{}/G}.$$
(4.6)
This motivates defining the function
$$Y(k)\frac{\mathrm{\Lambda }^{}/\mathrm{\Lambda }}{G^{}/G},$$
(4.7)
which completely determines the evolution of $`\mathrm{\Omega }_M`$ and $`\mathrm{\Omega }_\mathrm{\Lambda }`$:
$$\mathrm{\Omega }_M=\frac{Y(k)}{1Y(k)},\mathrm{\Omega }_\mathrm{\Lambda }=\frac{1}{1Y(k)}.$$
(4.8)
Here $`Y(k)`$ is understood as being evaluated along a particular solution of the RG equations. The above relations imply that, for $`Y(k)=0`$, the total energy density is completely dominated by the cosmological constant ($`\mathrm{\Omega }_M=0,\mathrm{\Omega }_\mathrm{\Lambda }=1`$), while $`Y(k)=\pm \mathrm{}`$ corresponds to complete matter domination ($`\mathrm{\Omega }_M=1,\mathrm{\Omega }_\mathrm{\Lambda }=0`$).
These relations can be used to express the deceleration parameter
$$q(t)\frac{\ddot{a}a}{\dot{a}^2}$$
(4.9)
in terms of $`Y(k)`$. From the improved system (2.1a,b,c) one derives that, as in the classical case,
$$q=\frac{1}{2}(3w+1)\mathrm{\Omega }_M\mathrm{\Omega }_\mathrm{\Lambda }.$$
(4.10)
Substituting (4.8) we obtain
$$q=\frac{1}{Y1}\left[\frac{3w+1}{2}Y+1\right].$$
(4.11)
Writing
$$Y(k)=y(g(k),\lambda (k))$$
(4.12)
in terms of the dimensionless couplings $`g,\lambda `$ it turns out that
$$y(g,\lambda )=1+\frac{1}{2\pi \eta _N(g,\lambda )}\frac{g}{\lambda }\left[10\mathrm{\Phi }_2^1(2\lambda )8\mathrm{\Phi }_2^1(0)5\eta _N\stackrel{~}{\mathrm{\Phi }}_2^1(2\lambda )\right]$$
(4.13)
has no explicit $`k`$-dependence. By definition, $`Y`$ is a function of $`k`$, while $`y`$ is a function of $`g`$ and $`\lambda `$. Hence $`\mathrm{\Omega }_M`$, $`\mathrm{\Omega }_\mathrm{\Lambda }`$, and $`q`$ are completely determined by the values of $`g`$ and $`\lambda `$:
$$\mathrm{\Omega }_M=\frac{y(g,\lambda )}{1y(g,\lambda )},\mathrm{\Omega }_\mathrm{\Lambda }=\frac{1}{1y(g,\lambda )},q=\frac{1+(3w+1)y(g,\lambda )/2}{y(g,\lambda )1}.$$
(4.14)
Thus it is important to know the properties of the function $`y(g,\lambda )`$.
We start by comparing eq. (4.13) to the condition determining the $`\mathrm{\Omega }`$-line, eq. (4.4). This shows that $`y(g,\lambda )=0`$ at the $`\mathrm{\Omega }`$-line, so that the energy density of a universe on this line is completely dominated by the cosmological constant, $`\mathrm{\Omega }_\mathrm{\Lambda }=1,\mathrm{\Omega }_M=0`$.
In our further analysis we first focus on the $`g=0`$-limit of $`y(g,\lambda )`$. This is motivated by the fact that the trajectory realized in Nature has $`g1`$ so that this limit provides a good approximation. Setting $`g=0`$ in (4.13) we then obtain:
$$y_0(\lambda )\underset{g0}{lim}y(g,\lambda )=1+\frac{1}{2\pi \lambda B_1(\lambda )}\left[10\mathrm{\Phi }_2^1(2\lambda )8\mathrm{\Phi }_2^1(0)\right].$$
(4.15)
Using the sharp cutoff, this function is displayed in fig. 4.
Looking at the left diagram, we observe that the domain of definition of $`y_0`$, ($`\mathrm{}<\lambda 1/2`$), contains the following special points and intervals:
$$\begin{array}{cc}\hfill \lambda \mathrm{}& y_0(\lambda )1,\hfill \\ \hfill \mathrm{}<\lambda <0& y_0(\lambda )>1,\hfill \\ \hfill \lambda =0^+& y_0(\lambda )=\mathrm{},\hfill \\ \hfill 0<\lambda \lambda _{\mathrm{\Omega }\text{line}}& y_0(\lambda )0,\hfill \\ \hfill \lambda _{\mathrm{\Omega }\text{line}}<\lambda <1/2& 0<y_0(\lambda )<1,\hfill \\ \hfill \lambda =1/2& y_0(\lambda )=1.\hfill \end{array}$$
(4.16)
Fig. 5 shows the full function $`y(g,\lambda )`$ on the $`g`$-$`\lambda `$-plane. This figure illustrates that the behavior described above also extends to the region where $`g>0`$ so that the discussion given below will also apply to a general trajectory with $`g>0`$. In particular, close to the $`\mathrm{\Omega }`$-line, $`y`$ is an approximately linear function of $`\lambda `$.
In terms of $`\mathrm{\Omega }_M`$ and $`\mathrm{\Omega }_\mathrm{\Lambda }`$, the features of $`y`$ found above have the following interpretation. The region $`\mathrm{}<\lambda <0`$ is associated to the IR behavior of RG-trajectories of Type Ia. A cosmology evolving along such a trajectory will asymptote to $`\mathrm{\Omega }_\mathrm{\Lambda }\mathrm{}`$, $`\mathrm{\Omega }_M+\mathrm{}`$ with the sum $`\mathrm{\Omega }_\mathrm{\Lambda }+\mathrm{\Omega }_M=1`$.
Next we have the separatrix which, in the IR, ends at the GFP $`g_{}=0,\lambda _{}=0`$. At this point $`y_0(\lambda =0^+)=\mathrm{}`$, so that the cosmology asymptotes to $`\mathrm{\Omega }_M=1,\mathrm{\Omega }_\mathrm{\Lambda }=0`$. The resulting universe will then be completely matter dominated at late times.
In the region $`0<\lambda \lambda _{\mathrm{\Omega }\text{line}}`$, we find the RG-trajectories of Type IIIa, discussed in the previous section. In this case the cosmological evolution ends at the $`\mathrm{\Omega }`$-line where $`y(g,\lambda )=0`$. Hence these cosmologies lead to a domination of the cosmological constant at late times: $`\mathrm{\Omega }_M=0,\mathrm{\Omega }_\mathrm{\Lambda }=1`$.
An interesting property of this latter region is that it splits into two rather different parts. In the first part, where $`y(g,\lambda )`$ increases rapidly, we observe matter domination while in the linear regime $`\mathrm{\Omega }_\mathrm{\Lambda }`$ dominates.
In the region $`\lambda _{\mathrm{\Omega }\text{line}}<\lambda <1/2`$ we have $`\mathrm{\Omega }_\mathrm{\Lambda }>1`$ and $`\mathrm{\Omega }_M<0`$, which is (probably) unphysical.
For later use we note the value of the function $`y`$ at the NGFP $`(g_{},\lambda _{})`$:
$$y(g_{},\lambda _{})=1$$
(4.17)
This result is most easily proven by simply inserting $`\mathrm{\Lambda }(k)=\lambda _{}k^2`$ and $`G(k)=g_{}/k^2`$ into eq. (4.7). By virtue of eq. (4.14), this implies that $`\mathrm{\Omega }_M^{}=\mathrm{\Omega }_\mathrm{\Lambda }^{}=1/2`$ and $`q^{}=(3w1)/4`$ at the fixed point.
## 5 Evolution in time and scale
The proper way of thinking about an RG improved cosmological history is to visualize it as a curve on theory space, the RG trajectory, which, by means of eqs. (2.1a,b,c), induces both a Robertson-Walker scale factor $`a(t)`$ and a cutoff identification $`k=k(t)`$. In this section we “switch on” this $`k`$\- or $`t`$-evolution and study, mostly by numerical methods, the coupled system (2.1a,b,c) after having first obtained a trajectory $`(g(k),\lambda (k))`$, or rather $`(G(k),\mathrm{\Lambda }(k))`$, by solving the flow equations (3.1). Because of their special relevance we shall consider trajectories of Type IIIa only. Since the crucial qualitative features such as the existence of an $`\mathrm{\Omega }`$-line are the same in all cutoff schemes, we shall use the technically convenient sharp cutoff throughout.
### 5.1 Initial conditions and RG trajectories
To investigate the properties of the solutions $`(g(k),\lambda (k))`$ numerically we choose our initial conditions for the dimensionless couplings $`g,\lambda `$ along the line connecting the turning points $`(g_T,\lambda _T)`$ of the trajectories. (See fig. 6.) For the sharp cutoff and close to the GFP this line is given by
$$\lambda _T=\frac{\phi _2}{2\pi }g_T.$$
(5.1)
For the numerics we supplement this relation with
$$k_{\mathrm{init}}=1,t_{\mathrm{init}}=0,a_{\mathrm{init}}=1.$$
(5.2)
This gives rise to a one-parameter family of cosmological solutions which, loosely speaking, are characterized by their distance to the GFP at $`\lambda =0,g=0`$. By definition, the trajectories pass the turning point at the scale $`k_T`$, i.e., $`g(k_T)=g_T`$ and $`\lambda (k_T)=\lambda _T`$. Thus, adopting the relations (5.2) amounts to expressing all dimensionful quantities in terms of appropriate powers of $`k_T`$. (In this parameterization the Big Bang occurs at $`t<0`$.)
Starting with these initial values we evolve the solutions in two directions:
* Towards the UV ($`k>k_T`$) where the RG-Flow will be attracted towards the NGFP.
* Towards the IR ($`k<k_T`$) where the cosmology will run towards the $`\mathrm{\Omega }`$-line.
For explicitness we consider the three sample solutions with
$$g_T=10^1,g_T=10^2,\text{and}g_T=10^3.$$
(5.3)
These examples are sufficient to illustrate the general trend for decreasing $`g_T`$ and to understand the qualitative properties of the RG-trajectory realized in Nature which has an extremely tiny $`g_T`$. The RG-trajectories obtained by numerically solving (3.1) with these initial conditions are shown in fig. 6.
We see that decreasing $`g_T`$ results in squeezing the solution into the corner between the separatrix (connecting the NGFP and the GFP) and the $`g=0`$-axis. In fact, within the resolution of fig. 6 the two trajectories with $`g_T=10^2`$ and $`g_T=10^3`$ run virtually on top of the separatrix (for $`k>k_T`$) and the horizontal axis (for $`k<k_T`$).
All three trajectories are of Type IIIa and terminate at the boundary $`\lambda =1/2`$. For $`g_T=10^1,10^2`$, and $`10^3`$ the termination scales are found to be $`k_{\mathrm{term}}/k_T=0.178,0.0612`$, and $`0.0196`$, respectively. These numbers are consistent with the rough estimate $`k_{\mathrm{term}}/k_T=𝒪(\sqrt{g_T})`$ derived in .<sup>9</sup><sup>9</sup>9The proportionality constant is $`k_{\mathrm{term}}/k_T0.612\sqrt{g_T}`$. The first case $`g_T=10^1`$ does not quite fit into the pattern, since this trajectory is not yet sufficiently deep in the GR regime. They confirm that the closer the trajectory approaches the GFP the later (in $`k`$) it terminates.
### 5.2 UV vs. IR branch
The two branches of the trajectory, $`k>k_T`$ and $`k<k_T`$, give rise to two corresponding branches of the cosmological solution which we shall refer to as the “UV cosmology” and “IR cosmology”, respectively. In the UV cosmology we are going to use the equation of state with $`w=1/3`$ (“radiation”), while we employ $`w=0`$ (“dust”) for the IR cosmology. In reality the universe has passed the scale $`k=k_T`$ well inside the radiation dominated era, but since we are mostly interested in qualitative features of RG improved cosmologies this aspect is inessential.
Next we shall apply the method described in Section 2 to the trajectories of the previous subsection and discuss their UV- and IR-branches in turn. In all examples the Hubble parameter is calculated using eq. (B.7) derived in Appendix B, and the deceleration parameter is determined with the help of eq. (4.14).
### 5.3 RG-improved UV cosmologies
The numerical results for the UV cosmologies resulting from the $`g_T`$-values (5.3) are displayed in figs. 7 \- 10. Fig. 7 shows the scale factor $`a(t)`$, and fig. 8 the corresponding Hubble constant $`H\dot{a}/a`$ and deceleration parameter $`q`$. In fig. 9 Newton’s constant $`G(t)G(k(t))`$ and likewise the cosmological constant is plotted as a function of time. Here $`t`$ is measured in units of $`k_T^1`$, and the origin $`t=0`$ corresponds to $`k=k_T`$.
#### 5.3.1 The regimes of the UV cosmologies
We observe that the cosmologies possess an initial singularity (“Big Bang”) at a time $`t_B<0`$ where $`a(t_B)=0`$ and $`H(tt_B)\mathrm{}`$. For $`t>t_B`$ we can distinguish the following regimes:
(A) The NGFP regime
Immediately after the Big Bang, for $`km_{\mathrm{Pl}}`$, the RG trajectory is well approximated by the constant functions $`g(k)=g_{}`$ and $`\lambda (k)=\lambda _{}`$. They correspond to the $`k`$-dependence (1.2) for which the system (2.1a,b,c) can be solved analytically. This yields the fixed point solution found in ref. . For $`w=1/3`$ it yields $`a(tt_B)`$, $`H=1/(tt_B)`$, $`q=0`$, $`G(tt_B)^2`$, and $`\mathrm{\Lambda }1/(tt_B)^2`$. The numerical solutions do indeed show this behavior close to $`t_B`$. In particular, $`G`$ vanishes and $`\mathrm{\Lambda }`$ diverges at the Big Bang.
(B) The linear regime of the NGFP
At slightly later time, corresponding to smaller scales $`k`$, the RG flow can be linearized about the NGFP . The linearized trajectories $`(g(k),\lambda (k))`$ are spirals about the NGFP, characterized by two critical exponents $`\theta ^{}`$ and $`\theta ^{\prime \prime }`$ . The most prominent cosmological signature of this regime is the oscillatory behavior of $`q(t)`$ seen in fig. 8 and, in a slightly magnified way, in fig. 10. Directly at the NGFP one has $`q=0`$; when the trajectory starts circling about the NGFP there are phases with both $`q>0`$ (deceleration) and $`q<0`$ (acceleration). In fact, there exists an infinite sequence of phases with either sign. When one approaches the Big Bang from above ($`tt_B`$), $`q`$ decreases from about $`q1`$, has a first zero, becomes negative, has another zero, becomes positive for a period shorter than the previous one, then again becomes negative for a time span shorter than the previous one, and so on. The Big Bang is approached by an infinite sequence of oscillations in $`q`$, of decreasing amplitude and decreasing duration.
(C) Crossover and linear regime of the GFP
At still lower scales, or later times, the RG trajectories leave the linear regime of the NGFP and very quickly “cross over” towards the GFP. Linearizing about the GFP, the resulting cosmologies are again easily understood analytically. One finds that approximately $`G(k)=const`$ and $`\mathrm{\Lambda }(k)=\mathrm{\Lambda }(k_T)[1+(k/k_T)^4]/2`$. If $`k`$ is related to $`t`$ in a monotonic way, we expect $`G`$ to stay constant and $`\mathrm{\Lambda }`$ to decrease towards its value at $`k_T`$ (where it becomes constant, too). This is exactly what we observe in fig. 9.
In the linear regime of the GFP, for the time $`t<0`$, i.e., before the turning point of the trajectory, the cosmological constant is not important for the three solutions. At least for $`g_T`$ small they have $`\mathrm{\Omega }_\mathrm{\Lambda }0`$ and $`\mathrm{\Omega }_M1`$ there, yielding $`q1`$. This is exactly the plateau-value of $`q(t)`$ which is approached in the second plot of fig. 8.
#### 5.3.2 Dynamical determination of $`k=k(t)`$
As we explained in Section 2, our method determines the cutoff identification $`k=k(t)`$ dynamically in such a way that the field equations are integrable by construction, the risk being that the $`k(t)`$ thus obtained does not lend itself for a clear interpretation of the resulting coarse-grained cosmology.
In the first diagram of fig. 11 we display the relationship between $`k`$ and $`t`$ by plotting $`(t(k)k)^1`$ as a function of $`\mathrm{ln}(k/k_T)`$. The motivation for this presentation is as follows. If this function is $`k`$-independent, $`t(k)k=const`$, we have $`t(k)1/k`$ or vice versa $`k(t)1/t`$ which is the rigid cutoff identification used in . In order to get the $`(t(k)k)^1`$ vs. $`\mathrm{ln}(k/k_T)`$ plot, we have shifted $`t(k)`$ such that the Big Bang ($`k=\mathrm{}`$) corresponds to $`t(k=\mathrm{})=0`$.<sup>10</sup><sup>10</sup>10This is different in figs. 8 and 9 where the UV cosmology corresponds to $`t<0`$. At $`t=0k=k_T=1`$, the expression $`(t(k)k)^1`$ develops a pole, which is avoided by shifting the Big Bang to $`t=0`$. In an analogous fashion the second plot in fig. 11 shows $`H(k)/k`$ as a function of $`k`$; this function would be constant for the identification $`k(t)H(t)`$.
From fig. 11 we see that both $`kH(k)`$ and $`k1/t(k)`$ provide a valid approximation to the true cutoff-identification in the UV domain ($`k>k_T`$) if one is interested in the overall qualitative features only. In this domain the approximation $`k1/t(k)`$ performs somewhat better than $`kH(k)`$.
#### 5.3.3 Oscillatory inflation
Finally we return to the phenomenon of the $`q(t)`$-oscillations in the NGFP-scaling regime which we described in 5.3.1 (B). In order to analyze the sign-flips of the deceleration parameter we recall from (4.14) that $`q=q(g,\lambda )`$ can be regarded as a function on theory space, the $`t`$-dependence arising by inserting the trajectory with $`k=k(t)`$, i.e., $`q(t)q(g(k(t)),\lambda (k(t)))`$.
As for the function $`q(g,\lambda )`$, there are regions on the $`g`$-$`\lambda `$-plane where it is positive and others where it is negative, the common boundary being the “$`q=0`$-line” shown in fig. 6. By virtue of (4.14) this line is given by the implicit condition $`1+(3w+1)y(g,\lambda )/2=0`$. To obtain the diagram in fig. 6 the equation for $`w=1/3`$, $`y(g,\lambda )=1`$, was solved numerically. To the right (left) of the $`q=0`$-line one has $`q<0`$ ($`q>0`$).
Now it is important to observe that precisely for $`w=1/3`$ the $`q=0`$-line runs exactly through the NGFP. In fact, according to (4.17) the NGFP has $`y(g_{},\lambda _{})=1`$, i.e., it lies on the $`q=0`$-line pertaining to $`w=1/3`$. As a consequence, when the RG trajectory spirals around the NGFP it crosses the $`q=0`$-line an infinite number of times in either direction. This explains the $`q`$-oscillations about $`q=0`$ found in the numerical solutions.
Thus the RG-improvement predicts epochs of accelerated expansion (“inflation”) in the early universe and one might wonder whether there can be any relation to the traditional models of cosmological inflation. To analyze this question we compute the number of “$`e`$-folds” the universe expands during the periods with $`q<0`$. We consider only the last such period because it has the longest duration and the most negative $`q(t)`$, see fig. 10. Note that even there $`q(t)`$ never becomes as small as $`1`$ which would correspond to a de Sitter-like phase.
We first derive a general formula which can be used to calculate the number of $`e`$-folds of expansion, $`N`$, occurring along any RG-trajectory. This formula will then be used to determine $`N`$ for the last period of accelerated expansion occurring before the solution leaves the linear regime of the NGFP (see fig. 6).
Our starting point is the usual definition of $`N`$,
$$N\mathrm{ln}\left[\frac{a(t_\mathrm{e})}{a(t_\mathrm{b})}\right],$$
(5.4)
where $`t_\mathrm{b}`$ and $`t_\mathrm{e}`$ denote the cosmological times where the acceleration begins and ends, respectively. Using (2.4a) we can relate $`N`$ directly to the RG-trajectory:
$$N=\frac{1}{3+3w}\mathrm{ln}\left[\frac{G_\mathrm{e}^{}}{\mathrm{\Lambda }_\mathrm{e}^{}}\frac{\mathrm{\Lambda }_\mathrm{b}^{}}{G_\mathrm{b}^{}}\right].$$
(5.5)
Here and in the following the subscripts $`\mathrm{b}`$, $`\mathrm{e}`$ denote the corresponding quantity at the beginning and end of the acceleration period. Eq. (5.5) can be evaluated further by re-expressing the quantities $`G^{}`$, $`\mathrm{\Lambda }^{}`$ using eq. (B.4). Writing the result in terms of the dimensionless couplings $`g,\lambda `$ we find
$$N=\frac{4}{3+3w}\mathrm{ln}\left[\frac{k_\mathrm{b}}{k_\mathrm{e}}\right]+\frac{1}{3+3w}\mathrm{ln}\left[\frac{A(g_\mathrm{e},\lambda _\mathrm{e})}{A(g_\mathrm{b},\lambda _\mathrm{b})}\right],$$
(5.6)
where we define
$$A(g,\lambda )g\left[\lambda +\frac{g}{2\pi \eta ^{\mathrm{sc}}}\left(10\mathrm{ln}(12\lambda )+2\zeta (3)\frac{5}{2}\eta ^{\mathrm{sc}}\right)\right]^1.$$
(5.7)
In terms of the RG-time $`\tau \mathrm{ln}(k/k_T)`$, and for $`w=1/3`$, this becomes simply
$$N=\tau _\mathrm{b}\tau _\mathrm{e}+\frac{1}{4}\mathrm{ln}\left[\frac{A(g_\mathrm{e},\lambda _\mathrm{e})}{A(g_\mathrm{b},\lambda _\mathrm{b})}\right].$$
(5.8)
Looking at fig. 6 we see that the coordinates of the last and the second to last $`q=0`$-crossing, ($`g_\mathrm{e},\lambda _\mathrm{e}`$) and ($`g_\mathrm{b},\lambda _\mathrm{b}`$), are to a very good precision equal for all Type IIIa trajectories, so the last term in (5.8) gives rise to an almost “universal” contribution.
We then evaluate $`N`$ along the last period of accelerated expansion, the semi-circle in fig. 6. The corresponding values are summarized in table 1.
From this table we observe that $`A_\mathrm{b}`$ and $`A_\mathrm{e}`$ are indeed independent of $`g_T`$ as suggested by fig. 6. Furthermore, we find that while the values of $`\tau _\mathrm{b}`$ and $`\tau _\mathrm{e}`$ increase monotonically when decreasing $`g_T`$, the difference $`\tau _\mathrm{b}\tau _\mathrm{e}`$, too, is independent of the initial condition for $`g_T`$. This implies that all trajectories give rise to essentially the same number of $`e`$-folds: $`N0.9`$. This number is far smaller than the 60 $`e`$-folds occurring in the “usual” inflationary scenarios, so that there is clearly no direct correspondence.
### 5.4 RG-improved IR cosmologies
In this section we investigate the “IR cosmologies” related to the $`k<k_T`$ branch of the RG trajectories. For the numerical illustration we shall use the initial conditions (5.3) again, this time integrating the RG equations towards smaller $`k`$.
#### 5.4.1 The running of $`G`$ and $`\mathrm{\Lambda }`$
In fig. 12 we have plotted $`G(k)`$ and $`\mathrm{\Lambda }(k)`$ for the three sample trajectories.
Going downward from $`k=1`$ (in $`k_T`$-units, as always) the cosmological constant has a weak scale dependence at first but then assumes a constant value over a wide range of $`k`$-values. For $`G`$, this plateau behavior has set in before the turning point even. These plateaus correspond to the classical GR regime where, by definition, $`G,\mathrm{\Lambda }const`$. At even smaller $`k`$, $`G(k)`$ starts increasing until it finally develops a vertical tangent at the termination scale $`k_{\mathrm{term}}`$. (For the real RG trajectory with $`g_T10^{60}`$ the increase is less abrupt than it appears in fig. 12.) For $`kk_{\mathrm{term}}`$, as long as $`G(k)`$ is not too different from its classical plateau value, the Einstein-Hilbert truncation should be reliable still. It is natural to ask, therefore, whether the IR increase of $`G(k)`$ leads to cosmological consequences, as long as we can trust the approximation. (Note that $`\mathrm{\Lambda }(k)`$ remains regular at the termination point; $`\lambda (k_{\mathrm{term}})=1/2`$ entails $`\mathrm{\Lambda }(k_{\mathrm{term}})=k_{\mathrm{term}}^2/2`$.)
#### 5.4.2 The cosmological history
Applying the method of Section 2 to the functions $`G(k)`$ and $`\mathrm{\Lambda }(k)`$ of fig. 12 we obtain the cosmological solutions displayed in the figs. 13 (scale factor), 14 (Hubble and deceleration parameter), 15 (time dependence of $`G`$ and $`\mathrm{\Lambda }`$), and 16 (scale-time relationship).
The gross features of all three of the typical cosmologies considered are as follows. Near $`kk_T`$ the universe is matter dominated $`(\mathrm{\Omega }_M1,\mathrm{\Omega }_\mathrm{\Lambda }0)`$ and has a deceleration parameter $`q1/2`$ therefore. According to fig. 14, this value decreases towards $`q=1`$ at asymptotically late times. This asymptotic regime is completely $`\mathrm{\Lambda }`$-dominated: $`\mathrm{\Omega }_M0,\mathrm{\Omega }_\mathrm{\Lambda }1`$. Decreasing $`g_T`$ the cosmological time passing during this transition from matter to vacuum dominance increases. The value $`q0.55`$, for instance, which roughly corresponds to the deceleration parameter observed today, is reached at increasingly late cosmological times as $`g_T`$ is becoming smaller. The Hubble parameter approaches a constant value for $`t\mathrm{}`$, indicating that the space-time approaches de Sitter space asymptotically. The kink in the $`a(t)`$-curves of fig. 14 marks the onset of the de Sitter behavior.
The transition form matter to $`\mathrm{\Lambda }`$-dominance occurs in classical cosmology as well. To what extent are the improved cosmologies modified by quantum gravity effects? Fig. 15 shows that $`\mathrm{\Lambda }(t)`$ is decreasing as long as $`k`$ is not too far below $`k_T`$, while $`G(t)`$ is essentially constant.
For $`t5k_T^1`$, say, both $`\mathrm{\Lambda }(t)`$ and $`G(t)`$ attain constant values which remain unaltered for $`t\mathrm{}`$ (at least within the resolution of fig. 15). In the next subsection we shall demonstrate that the asymptotic values of $`\mathrm{\Lambda }`$ and $`G`$ are exactly those predicted by the flow equation linearized about the GFP provided $`g_T`$ is small enough. This implies that, for $`g_T`$ sufficiently small, i.e., for trajectories getting close to the GFP, there are no significant renormalization effects at late times.
#### 5.4.3 Asymptotic vs. laboratory values and the cosmological constant problem
Let us come back to the plots of $`G(t)`$ and $`\mathrm{\Lambda }(t)`$ shown in fig. 15. This figure illustrates that decreasing $`g_T`$ decreases the asymptotic values of $`\mathrm{\Lambda }(t)`$ and $`G(t)`$ for $`t\mathrm{}`$. Table 2 summarizes these asymptotic values of $`G`$ and $`\mathrm{\Lambda }`$ for various choices of $`g_T`$. This table also contains the corresponding “laboratory” values $`G_{\mathrm{lab}}`$ and $`\mathrm{\Lambda }_{\mathrm{lab}}`$. Their interpretation is as follows .
Provided the trajectory gets sufficiently close to the GFP it contains a long classical regime. (In fig. 1 this “GR-regime” corresponds to the segment of the trajectory between the points P<sub>1</sub> and P<sub>2</sub>.) In this regime the dimensionful quantities $`G`$ and $`\mathrm{\Lambda }`$ are approximately $`k`$-independent. Linearizing about the GFP one finds for their values, expressed in terms of $`k_T`$,
$`G_{\mathrm{lab}}`$ $`=g_T/k_T^2,`$ (5.9a)
$`\mathrm{\Lambda }_{\mathrm{lab}}`$ $`={\displaystyle \frac{1}{2}}\lambda _Tk_T^2=(\phi _2/4\pi )g_Tk_T^2.`$ (5.9b)
In table 2 we compare the asymptotic values of $`G`$ and $`\mathrm{\Lambda }`$ to their “lab” values computed from (5.9a,b). We find that for $`g_T`$ of the order $`10^2`$ or smaller the plateau values at asymptotically late times virtually coincide with those measured in a “laboratory” in which classical GR is known to apply. At $`g_T=10^1`$, instead, $`G_{\mathrm{asym}}`$ is slightly larger than $`G_{\mathrm{lab}}`$ while $`\mathrm{\Lambda }_{\mathrm{asym}}`$ is slightly smaller than $`\mathrm{\Lambda }_{\mathrm{lab}}`$.
The interpretation of this result is as follows. For $`g_T`$ sufficiently small, the universe does not experience significant IR renormalization effects; in the limit $`t\mathrm{}`$ it basically keeps its values of $`G`$ and $`\mathrm{\Lambda }`$ from the GR regime. Only for comparatively large values of $`g_T`$ we find deviations between the GR-formulas (5.9a,b) and the numerically computed asymptotic values. Those deviations occur because the RG trajectory does not get sufficiently close to the GFP to give rise to a long classical regime where (5.9a,b) would apply.
The conventional Planck mass is defined in terms of Newton’s constant measured in a classical “laboratory”: $`m_{\mathrm{Pl}}1/\sqrt{G_{\mathrm{lab}}}`$. This definition, together with (5.9a) leads to the following important relation between $`m_{\mathrm{Pl}}`$ and $`k_T`$:
$$k_T^2=g_Tm_{\mathrm{Pl}}^2.$$
(5.10)
Obviously $`g_T1`$ leads to a large hierarchy $`k_T/m_{\mathrm{Pl}}1`$. In table 3 we use (5.10) in order to express the asymptotic and laboratory values of $`G`$ and $`\mathrm{\Lambda }`$ in terms of the more familiar Planckian units.
As for the cosmological constant problem, the crucial point to be noted is that $`\mathrm{\Lambda }_{\mathrm{lab}}\mathrm{\Lambda }_{\mathrm{asym}}`$ is suppressed by a factor $`g_T^2`$ relative to its “natural” value $`m_{\mathrm{Pl}}^2`$. As was discussed in detail in ref. , fine-tuning the RG-trajectory in such a way that it spends a long RG time near the GFP by picking an “unnaturally” small value of $`g_T`$ leads to a long (in $`k`$-units) classical regime on the trajectory. Once this is achieved, the solution of the cosmological constant problem is for free:
$$\mathrm{\Lambda }_{\mathrm{lab}}/m_{\mathrm{Pl}}^2=(\phi _2/4\pi )g_T^21.$$
(5.11)
The numbers in table 3 illustrate this important phenomenon. The smaller is $`g_T`$, the closer the trajectory approaches the GFP, the smaller is the cosmological constant in the classical GR laboratory further downstream on the trajectory.
The relative importance of the cosmological constant can be summarized as follows. In the very early universe, in the NGFP regime, one has $`\mathrm{\Omega }_M=\mathrm{\Omega }_\mathrm{\Lambda }=1/2`$ showing that the vacuum and matter energy densities drive the expansion on an exactly equal footing. Near the GFP, in particular in the classical GR regime, one automatically is lead to $`\mathrm{\Omega }_\mathrm{\Lambda }0`$ and $`\mathrm{\Omega }_M1`$, a conventional purely matter dominated (flat) FRW cosmology. At infinitely late times $`\mathrm{\Lambda }`$ takes over again and with $`\mathrm{\Omega }_\mathrm{\Lambda }1`$ and $`\mathrm{\Omega }_M0`$ a de Sitter like behavior sets in.
#### 5.4.4 $`\mathrm{\Omega }`$-line mechanism and dynamical cutoff identification
In fig. 12 we saw that $`G(k)`$ and $`\mathrm{\Lambda }(k)`$ show a strong IR running for small $`k`$. It therefore comes as a surprise perhaps that $`G(t)`$ and $`\mathrm{\Lambda }(t)`$, according to fig. 15, show no sign of such IR renormalizations for $`t\mathrm{}`$; according to the previous subsection, the late cosmology is essentially classical.
This apparent contradiction is resolved in fig. 16 which displays the relationship between the scale $`k`$ and the cosmological time $`t`$. As expected, $`t=t(k)`$ is a monotonically decreasing function of $`k`$. However, an infinitely old universe (“$`t=\mathrm{}`$”) does not correspond to $`k=0`$, but rather to a non-zero asymptotic scale
$$k_{\mathrm{asym}}k(t\mathrm{})>0.$$
(5.12)
The function $`t=t(k)`$ has a singularity at this scale: $`t(kk_{\mathrm{asym}})\mathrm{}`$. Stated differently, the inverse function $`k=k(t)`$ is bounded below by $`k_{\mathrm{asym}}`$ and, even for arbitrarily late times, does not reach arbitrarily small scales: $`k(t)>k_{\mathrm{asym}}`$ for all $`t`$.
The absence of visible IR effects in $`G(t)G(k=k(t))`$ and $`\mathrm{\Lambda }(t)\mathrm{\Lambda }(k=k(t))`$ is explained by the fact that, for the examples considered, $`k_{\mathrm{asym}}`$ is much larger that the scale $`k_{\mathrm{term}}`$ near which $`G(k)`$ and $`\mathrm{\Lambda }(k)`$ get renormalized strongly. This is precisely the $`\mathrm{\Omega }`$-line mechanism we discussed already earlier: for $`t\mathrm{}`$ the dimensionless $`g(k(t))`$ and $`\lambda (k(t))`$ approach a point on the $`\mathrm{\Omega }`$-line. The endpoint of the cosmology, “$`a=\mathrm{}`$” or $`\rho =0`$, corresponds to this point on the $`g`$-$`\lambda `$-plane; it is still far away from the $`|\eta _N|=\mathrm{}`$-line close to which the IR renormalizations would become strong. Therefore, once the universes have entered the GR regime at about $`t5k_T^1`$, say, they remain classical for all later times. In this sense the $`\mathrm{\Omega }`$-line separates the cosmologically accessible parts of theory space from those with strong IR running.
The asymptotic de Sitter phase of the universe corresponds to the approximately time independent scale $`k(t)const=k_{\mathrm{asym}}k_{\mathrm{term}}`$.
Let us look more closely at the cutoff identification $`k=k(t)`$ which was generated dynamically by our system of equations. In a slightly different presentation it is displayed as the $`k<k_T`$-branch in fig. 11. In this figure $`(t(k)k)^1`$ and $`H(t(k))/k`$ are plotted vs. $`\mathrm{ln}(k/k_T)`$. If the actual (obviously rather complicated) cutoff identification was close to $`k1/t`$ and $`kH(t)`$, respectively, those plots would show a constant function for all $`k`$.
It is clear, and also confirmed by the plots, that in the IR the real $`k(t)`$ is quite different from $`k1/t`$, the reason being that the real $`k(t)`$ is constant, equal to $`k_{\mathrm{asym}}`$, while $`1/t`$ decreases monotonically for $`t\mathrm{}`$.
On the other hand, $`H(t(k))/k`$ is seen to be approximately $`k`$-independent within a factor of less than about 2. Thus, at least at a qualitative or “semi-quantitative” level, the actual cutoff identification can be approximated by $`kH`$.<sup>11</sup><sup>11</sup>11As an ansatz, $`kH`$ has also been employed in refs. in a different context. This identification would indeed associate a constant scale $`k_{\mathrm{asym}}`$ to the de Sitter final state, proportional to the time independent $`H_{\mathrm{asym}}=\sqrt{\mathrm{\Lambda }_{\mathrm{asym}}/3}`$.
From these observations it is clear how the cosmologies are to be interpreted. Since approximately $`kH`$ the length scale characterizing the averaging volume is the Hubble radius $`\mathrm{}_HH^1`$. The cosmological parameters computed correspond to averages over the volume $`\mathrm{}_H^4`$. The scale $`\mathrm{}_H`$ characterizes the radius of curvature of the four-dimensional space-time, the size of the “Einstein elevator”. The asymptotic de Sitter phase makes it particularly clear that the temporal proper distance to the Big Bang, $`t`$, would not lend itself as a coarse graining scale within the improved equations approach. For the UV cosmology, in particular near the NGFP , both identifications are equivalent, though.
### 5.5 The RG-trajectory realized by Nature
In ref. the observational (supernova, CMBR, etc.) data were interpreted under the assumption that the gravitational RG trajectory which Nature realizes has the qualitative features of a Type IIIa trajectory of QEG. It was found that the turning point of this trajectory is extremely close to the GFP, and that it is passed at a scale very far below the Planck scale:
$$g_T=𝒪(10^{60}),\lambda _T=𝒪(10^{60}),k_T=𝒪(10^{30}m_{\mathrm{Pl}}).$$
(5.13)
These order of magnitude estimates do not depend on whether the late universe is, or is not affected by IR renormalization effects. The analysis shows that even if there was no $`\mathrm{\Omega }`$-mechanism the data would not afford a ratio $`G_{\mathrm{cosmological}}/G_{\mathrm{lab}}`$ larger than about one order of magnitude .
The universe passes the turning point a fraction of a second after the Big Bang, at a Hubble radius of the order $`\mathrm{}_Hk_T^110^{30}l_{\mathrm{Pl}}10^3`$ cm. After that $`g(k(t))`$ decreases from $`10^{60}`$ to its present value $`g_{\mathrm{today}}`$ of about $`10^{120}`$.
Since $`g(k)1`$ during most epochs of the cosmological evolution, in particular in the late (IR) universe, we may use the $`g0`$ limit of the corresponding beta-functions there. This allows for a simple determination of the point on the $`g`$-$`\lambda `$-plane where the universe resides today. Assuming we know $`\mathrm{\Omega }_{\mathrm{\Lambda }0}\mathrm{\Omega }_\mathrm{\Lambda }(t_{\mathrm{today}})`$ we can solve the second equation of (4.14) for the present value of $`\lambda `$, $`\lambda _{\mathrm{today}}`$:
$$1y(0,\lambda _{\mathrm{today}})=\mathrm{\Omega }_{\mathrm{\Lambda }0}^1.$$
(5.14)
Here we approximated $`g_{\mathrm{today}}0`$. Note that the precise value of $`\lambda _{\mathrm{today}}`$ is scheme dependent.
For a rough estimate of $`\lambda _{\mathrm{today}}`$ we may assume that the classical FRW cosmology is essentially correct at late times. Analyzing the observational data within this framework yields $`\mathrm{\Omega }_{\mathrm{\Lambda }0}0.7`$. For this value, and the $`y`$-function corresponding to the sharp cutoff, we find
$$\lambda _{\mathrm{today}}0.320.$$
(5.15)
Looking at fig. 2 confirms that the point $`(g_{\mathrm{today}},\lambda _{\mathrm{today}})(10^{120},0.32)`$ indeed lies to the left of the $`\mathrm{\Omega }`$-line for the sharp cutoff, as it should be.
#### 5.5.1 The age of the RG-universe
As a check of our beta-function formalism, and in order to confirm the conclusions of subsection 5.4, we next derive a relationship between the age of the universe, $`t_{\mathrm{today}}`$, the cosmological constant $`\mathrm{\Lambda }_{\mathrm{lab}}\mathrm{\Lambda }_{\mathrm{asym}}`$, and the measured $`\mathrm{\Omega }_{\mathrm{\Lambda }0}`$.
First we use the value of $`\mathrm{\Omega }_{\mathrm{\Lambda }0}`$ in order to solve (5.14) for $`\lambda _{\mathrm{today}}(\mathrm{\Omega }_{\mathrm{\Lambda }0})`$. Next we rewrite (the reciprocal) eq. (2.5) in terms of the dimensionless couplings $`g=k^2G`$, $`\lambda =\mathrm{\Lambda }/k^2`$ and expand its RHS up to the leading order in $`g`$:
$$\begin{array}{cc}\hfill \frac{dt}{dk}=& \frac{\sqrt{3}k^2}{\sqrt{\lambda }(1y_0)^{1/2}}\left[\frac{(\stackrel{~}{B}_1^{\mathrm{sc}})^{}}{B_1^{\mathrm{sc}}}\frac{\pi \lambda (\stackrel{~}{B}_1^{\mathrm{sc}})^{}\left(10\mathrm{ln}(12\lambda )4\zeta (3)+\frac{20\lambda }{12\lambda }\right)}{\pi \lambda B_1^{\mathrm{sc}}\left(5\mathrm{ln}(12\lambda )2\zeta (3)\right)}\right]\hfill \\ & +𝒪(g).\hfill \end{array}$$
(5.16)
Here
$$(\stackrel{~}{B}_1^{\mathrm{sc}})^{}=\frac{4}{3\pi }\lambda \left[\frac{5}{12\lambda }+\frac{18}{(12\lambda )^2}\right].$$
(5.17)
This expression can be converted to a differential equation for $`dt/d\lambda `$ by exploiting that in the GR regime we have the relation $`\lambda =\mathrm{\Lambda }_{\mathrm{lab}}/k^2`$. Its use is legitimate here since the age of the universe for trajectories with sufficient “squeezing” is dominated by their GR regime. Substituting it into (5.17) we obtain the desired general relationship:
$$t_{\mathrm{today}}=\frac{1}{\sqrt{\mathrm{\Lambda }_{\mathrm{lab}}}}I(\lambda _{\mathrm{today}}).$$
(5.18)
Here $`I(\lambda _{\mathrm{today}})`$ is given by the dimensionless integral
$$\begin{array}{cc}\hfill I(\lambda _{\mathrm{today}})& _0^{\lambda _{\mathrm{today}}}𝑑\lambda \frac{\sqrt{3}}{2\lambda (1y_0)^{1/2}}\hfill \\ & \times \left[\frac{(\stackrel{~}{B}_1^{\mathrm{sc}})^{}}{B_1^{\mathrm{sc}}}\frac{\pi \lambda (\stackrel{~}{B}_1^{\mathrm{sc}})^{}\left(10\mathrm{ln}(12\lambda )4\zeta (3)+\frac{20\lambda }{12\lambda }\right)}{\pi \lambda B_1^{\mathrm{sc}}\left(5\mathrm{ln}(12\lambda )2\zeta (3)\right)}\right].\hfill \end{array}$$
(5.19)
As an example, we evaluate this integral numerically for the $`\lambda _{\mathrm{today}}`$ given in eq. (5.15). We find
$$I(\lambda _{\mathrm{today}}=0.32)=4.19$$
(5.20)
Using this value in (5.19), along with $`\sqrt{\mathrm{\Lambda }_{\mathrm{lab}}}\sqrt{\mathrm{\Lambda }_{\mathrm{asym}}}10^{60}m_{\mathrm{Pl}}`$ we obtain the age $`t_{\mathrm{today}}=4.19\times \mathrm{\hspace{0.17em}10}^{60}t_{\mathrm{Pl}}`$. As expected, this is essentially the same age one obtains from the classical FRW equations if one uses the same input data.
#### 5.5.2 The deceleration parameter
We now calculate $`q(t)`$ along the trajectory realized in Nature below the turning point. As in the previous subsection we will work in the approximation $`g0`$. We parameterize the trajectory by its $`\lambda `$-values $`\lambda [\lambda _T,\lambda _{\mathrm{\Omega }\mathrm{line}}]`$, i.e., $`k`$ is thought of as a function of $`\lambda `$, with the inverse $`\lambda =\lambda (k)`$. In this parameterization we can use (5.18) to obtain $`t(\lambda (k))=1/\sqrt{\mathrm{\Lambda }_{\mathrm{lab}}}I(\lambda (k))`$, the cosmological time passing during the evolution along the trajectory, while the corresponding deceleration parameter $`q(\lambda (k))`$ is obtained from eq. (4.11).
The function $`q(t)`$ along the trajectory can then be obtained as the parametric curve $`\lambda (t(\lambda ),q(\lambda ))`$, $`\lambda [\lambda _T,\lambda _{\mathrm{\Omega }\mathrm{line}}]`$.
The resulting $`q(t)`$ is shown in fig. 17. Note that the time $`t`$ displayed on the horizontal axis is in units $`10^{60}t_{\mathrm{Pl}}`$, indicating that the “crossover” from $`q(0)=1/2`$ to $`q(t\mathrm{})=1`$ indeed occurs at cosmological time spans of the order of the age of our present universe. Indeed, the present universe corresponds to the age $`t_{\mathrm{today}}4.2\times 10^{60}t_{\mathrm{Pl}}`$ and the deceleration parameter $`q_{\mathrm{today}}0.55`$.
## 6 Summary and Discussion
In this paper we presented a coherent renormalization group based framework which allows for the inclusion of potential quantum gravity effects during the entire cosmological history, from the epoch after the initial singularity to the scales of the present universe. We demonstrated that the strategy of RG improving the field equations is physically viable during all stages of the cosmological evolution. While from a mathematical point of view it is always possible to fix the cutoff scale in a way which renders the modified Einstein equations consistent, the result of the dynamical determination of $`k(t)`$ is not a priori guaranteed to lead to a useful interpretation of the coarse grained cosmology. However, we found that the dynamical cutoff identification corresponds to a time dependent averaging scale $`\mathrm{}(t)`$ which, at any time, is approximately equal to the Hubble radius $`\mathrm{}_H(t)`$. Therefore we may conclude that the “microscope” whose observations are described by the RG improved cosmology has exactly the right “resolving power” to see the large scale structure typical of cosmology.
One of the most interesting features of the early cosmology was the discovery of a kind of “oscillatory inflation”. Running time backwards, the initial singularity is approached by an infinite sequence of increasingly short time intervals during which the universe accelerates and decelerates, respectively. This behavior is due to the fact that the non-Gaussian fixed point has complex critical exponents. The oscillatory approach of the Big Bang is reminiscent of the classical Belinskii-Khalatnikov-Lifshitz scenario but contrary to the latter the quantum effect found here occurs even in isotropic universes.
The most interesting aspect of the improved cosmologies at late times is the “$`\mathrm{\Omega }`$-mechanism” which we investigated in detail. At small scales the RG trajectories of the relevant type leave the domain of validity of the Einstein-Hilbert truncation, and it has been argued that in the IR there could be a regime where the RG effects become strong again. Remarkably, even if this regime exists, the cosmologies we found would never get affected by it. Even at arbitrarily late times they remain essentially classical since the underlying RG trajectory is never probed beyond a certain “$`\mathrm{\Omega }`$-point” at which $`\eta _N`$ is very small still.
At this point we must ask whether this shielding of this $`\eta _N`$-divergence is a reliable prediction. Does the universe as a whole really never enter the regime where $`|\eta _N|`$ is large? Our argument was based upon two crucial facts: (i) The consistency condition for the RG improved Einstein’s equation has the simple form (2.1c) which leads directly to the relationship $`\rho \mathrm{\Lambda }^{}(k)/G^{}(k)`$ of eq. (2.4b); (ii) The RG flow is such that the zero of $`\mathrm{\Lambda }^{}(k)`$ occurs at a higher value of $`k`$ than the divergence of $`G^{}(k)`$. As a consequence of (ii), $`\rho `$ vanishes and $`a\rho ^{1/(3+3w)}`$ reaches infinity before $`|\eta _N|`$ has become large.
As for the validity of (ii), even though we analyzed the Einstein-Hilbert truncation only, this prediction was perfectly stable under a change of the cutoff scheme. Because of this robustness, it is likely to be actually correct and not a truncation artifact.
The much more subtle issue is (i) which is not related to the RG flow per se but to the improvement procedure. It is important to note that, at least in this particular form, the zero of $`\mathrm{\Lambda }^{}(k)`$ implies an infinite scale factor only in the approach of improving the *field equations*. It would not happen if one improves the solutions of the classical equations by replacing $`GG(t)`$, $`\mathrm{\Lambda }\mathrm{\Lambda }(t)`$ there, or if one improves the action functional. The latter approach has been investigated in detail in ref. . Performing the substitution under the space-time integral of the Einstein-Hilbert action one obtains a different consistency condition:<sup>12</sup><sup>12</sup>12This form of the consistency condition corresponds to the choice $`\mathrm{\Theta }_{\mu \nu }=0`$, see ref. .
$$\dot{\mathrm{\Lambda }}+8\pi \rho \dot{G}+3H\left(\frac{\dot{G}}{G}\right)^2+3\left(\frac{\dot{G}}{G}\right)\left(\frac{\ddot{a}}{a}\right)=0.$$
(6.1)
We observe that, in this framework, $`\dot{\mathrm{\Lambda }}\mathrm{\Lambda }^{}0`$ by no means implies that $`\rho 0`$ and $`a\mathrm{}`$. The improved equation and action approaches are (almost) equivalent in the UV, but can lead to different predictions in the IR. There they effectively represent certain non-local terms of the ordinary effective action $`\mathrm{\Gamma }_{k=0}`$ generated during the RG running but not taken into account explicitly . The additional terms in (6.1) which are not present in (2.1c) are due to the fact that in the improved action approach the scalar fields $`G`$ and $`\mathrm{\Lambda }`$ carry energy and momentum which is neglected in the improved equation approach.
For this reason it cannot be excluded for the time being that the shielding of the $`\eta _N`$-divergence is an artifact of the improvement scheme since, at least via the mechanism discussed above, the shielding at $`\mathrm{\Lambda }^{}=0`$ occurs only if one neglects the energy and momentum carried by the scalars $`G`$ and $`\mathrm{\Lambda }`$. Clearly more work is needed in order to settle this issue. It should be emphasized, though, that whatever is the final answer about RG effects in late cosmology it bears no simple relationship to the conjectured modifications of gravity at galactic scales. The corresponding cutoff identifications mimicking the essential IR-relevant invariants in both cases are extremely difficult to guess beforehand, and in particular in a Lorentzian setting where the actual cutoff is (a generalization of) “virtuality” rather than “inverse distance” one certainly cannot expect that the visibility of the quantum effects increases with distance in a naive way. For this reason the results of the present paper have no direct relevance to the scenario of modified galactic dynamics in .
Acknowledgements
We would like to thank A. Bonanno, J. Moffat, and T. Prokopec for helpful discussions.
## Appendix A Threshold functions for various cutoff schemes
In this appendix we collect the relevant properties of the threshold functions $`\mathrm{\Phi }`$ and $`\stackrel{~}{\mathrm{\Phi }}`$ for the cutoff schemes used in the main part of the paper. We employ different cutoff schemes in order to gain insight about whether properties of our solutions are independent of the scheme used to suppress the IR-modes in the path integral when deriving the beta-functions (3.2a,b). This cutoff-scheme dependence is contained in the threshold functions
$$\begin{array}{cc}\hfill \mathrm{\Phi }_n^p(w)& \frac{1}{\mathrm{\Gamma }(n)}_0^{\mathrm{}}𝑑zz^{n1}\frac{R^{(0)}(z)zR^{(0)}(z)}{\left[z+R^{(0)}(z)+w\right]^p},\hfill \\ \hfill \stackrel{~}{\mathrm{\Phi }}_n^p(w)& \frac{1}{\mathrm{\Gamma }(n)}_0^{\mathrm{}}𝑑zz^{n1}\frac{R^{(0)}(z)}{\left[z+R^{(0)}(z)+w\right]^p},\hfill \end{array}$$
(A.1)
(defined for $`p=1,2,3,\mathrm{}`$ and $`n>0`$) through the choice of the dimensionless cutoff function $`R^{(0)}(z)`$. Concretely, we employed three different type A cutoffs , the exponential cutoff, the optimized cutoff , and the sharp cutoff . The latter two have the advantage that the integrals appearing in (A.1) can be evaluated analytically. This provides a considerable simplification when solving the RG-equations as for these cases the RG-flow of $`G(k),\mathrm{\Lambda }(k)`$ is governed by simple first order differential equations.
### A.1 The optimized cutoff
For the optimized cutoff the dimensionless cutoff function $`R^{(0)}(z)`$ is given by
$$R^{(0)}(z)^{\mathrm{opt}}(1z)\mathrm{\Theta }(1z).$$
(A.2)
Substituting this expression into (A.1) and carrying out the integrals we obtain
$$\mathrm{\Phi }_n^p(w)^{\mathrm{opt}}=\frac{1}{\mathrm{\Gamma }(n+1)}\frac{1}{(1+w)^p},\stackrel{~}{\mathrm{\Phi }}_n^p(w)^{\mathrm{opt}}=\frac{1}{\mathrm{\Gamma }(n+2)}\frac{1}{(1+w)^p}.$$
(A.3)
These threshold functions are related through the relations
$$\begin{array}{cc}\hfill \mathrm{\Phi }_n^{p+1}(w)^{\mathrm{opt}}=& \frac{1}{p}\frac{}{w}\mathrm{\Phi }_n^p(w)^{\mathrm{opt}},\mathrm{\Phi }_{n+1}^p(w)^{\mathrm{opt}}=\frac{1}{n+1}\mathrm{\Phi }_n^p(w)^{\mathrm{opt}},\hfill \\ \hfill \stackrel{~}{\mathrm{\Phi }}_n^p(w)^{\mathrm{opt}}=& \frac{1}{n+1}\mathrm{\Phi }_n^p(w)^{\mathrm{opt}}.\hfill \end{array}$$
(A.4)
### A.2 The exponential cutoff
The exponential cutoff is given by the following one-parameter family of cutoff functions
$$R^{(0)}(z;s)^{\mathrm{Exp}}\frac{sz}{\mathrm{exp}(sz)1},s>0,$$
(A.5)
where $`s`$ is a shape parameter. In this case the integrals (A.1) cannot be evaluated analytically. Therefore we have to rely on a numerical integration when solving the RG equations. In order to have a reasonable convergence in the threshold functions we choose the shape parameter $`s=2`$.
### A.3 The sharp cutoff
The sharp cutoff is defined via the cutoff function
$$R_k(p^2)^{\mathrm{sc}}k^2R^{(0)}(z)^{\mathrm{sc}}\widehat{R}\mathrm{\Theta }(1z),$$
(A.6)
where the limit $`\widehat{R}\mathrm{}`$ is taken after substituting this function into the integrals defining the threshold functions. The resulting threshold functions have been determined in :
$$\begin{array}{cc}\hfill \mathrm{\Phi }_n^1(w)^{\mathrm{sc}}=& \frac{1}{\mathrm{\Gamma }(n)}\mathrm{ln}(1+w)+\phi _n,\text{for}p=1,\hfill \\ \hfill \mathrm{\Phi }_n^p(w)^{\mathrm{sc}}=& \frac{1}{\mathrm{\Gamma }(n)}\frac{1}{p1}\frac{1}{(1+w)^{p1}},\text{for}p>1,\hfill \\ \hfill \stackrel{~}{\mathrm{\Phi }}_n^1(w)^{\mathrm{sc}}=& \frac{1}{\mathrm{\Gamma }(n+1)},\text{for}p=1,\hfill \\ \hfill \stackrel{~}{\mathrm{\Phi }}_n^1(w)^{\mathrm{sc}}=& 0,\text{for}p>1.\hfill \end{array}$$
(A.7)
The $`\phi _n`$’s are a priori undetermined constants of integration. We choose them as
$$\phi _n\mathrm{\Phi }_n^1(0)^{\mathrm{Exp}(s=1)}=n\zeta (n+1),$$
(A.8)
where $`\zeta (n)`$ denotes the Riemann $`\zeta `$-function.
## Appendix B $`H`$ as a function of the RG scale
In this appendix we rewrite the Hubble parameter as a function of $`k,G(k)`$, and $`\mathrm{\Lambda }(k)`$. Our starting point is eq. (2.1a) which (for $`K=0`$) yields
$$H\frac{\dot{a}}{a}=\pm \frac{1}{\sqrt{3}}\left[\mathrm{\Lambda }+8\pi G\rho \right]^{1/2},$$
(B.1)
with the plus (minus) sign corresponding to an expanding (contracting) universe. We then substitute eq. (2.4b) to obtain $`H`$ in terms of $`G,\mathrm{\Lambda }`$ and their $`k`$-derivatives
$$H=\pm \left[\frac{1}{3G^{}}\left(G^{}\mathrm{\Lambda }G\mathrm{\Lambda }^{}\right)\right]^{1/2}.$$
(B.2)
The $`k`$ derivatives are conveniently expressed through the modified beta-functions
$$\stackrel{~}{\beta }_G\frac{1}{k}\beta _G,\stackrel{~}{\beta }_\mathrm{\Lambda }\frac{1}{k}\beta _\mathrm{\Lambda }.$$
(B.3)
With these new functions the derivatives $`G^{}`$, $`\mathrm{\Lambda }^{}`$ are simply
$$G^{}=\stackrel{~}{\beta }_G(k,G,\mathrm{\Lambda }),\mathrm{\Lambda }^{}=\stackrel{~}{\beta }_\mathrm{\Lambda }(k,G,\mathrm{\Lambda }).$$
(B.4)
Substituting the sharp cutoff (A.7) into these equations we find
$$\begin{array}{cc}\hfill \stackrel{~}{\beta }_G& =\frac{kG^2B_1^{\mathrm{sc}}}{1B_2^{\mathrm{sc}}k^2G}\hfill \\ \hfill \stackrel{~}{\beta }_\mathrm{\Lambda }& =\frac{kG\mathrm{\Lambda }B_1^{\mathrm{sc}}}{1B_2^{\mathrm{sc}}k^2G}+\frac{1}{\pi }k^3G\left[5\mathrm{ln}\left(1\frac{2\mathrm{\Lambda }}{k^2}\right)+2\zeta (3)\frac{5}{4}\eta ^{\mathrm{sc}}\right].\hfill \end{array}$$
(B.5)
where
$$\begin{array}{cc}\hfill B_1^{\mathrm{sc}}& =\frac{1}{3\pi }\left[5\mathrm{ln}\left(1\frac{2\mathrm{\Lambda }}{k^2}\right)+\zeta (2)\frac{18}{1\frac{2\mathrm{\Lambda }}{k^2}}6\right],\hfill \\ \hfill B_2^{\mathrm{sc}}& =\frac{5}{6\pi },\hfill \\ \hfill \eta ^{\mathrm{sc}}& =\frac{k^2GB_1^{\mathrm{sc}}}{1B_2^{\mathrm{sc}}k^2G}.\hfill \end{array}$$
(B.6)
Eq. (B.2) can then be evaluated along any given RG-trajectory $`k(G(k),\mathrm{\Lambda }(k))`$. Thus, in terms of (B.5) with (B.6), the Hubble parameter for an expanding universe is given by
$$H(k,G,\mathrm{\Lambda })=\left[\frac{\mathrm{\Lambda }\stackrel{~}{\beta }_G(k,G,\mathrm{\Lambda })G\stackrel{~}{\beta }_\mathrm{\Lambda }(k,G,\mathrm{\Lambda })}{3\stackrel{~}{\beta }_G(k,G,\mathrm{\Lambda })}\right]^{1/2}.$$
(B.7) |
warning/0507/cond-mat0507452.html | ar5iv | text | # Bosonic versus fermionic pairs of topological spin defects in monolayered high-𝑇_"c" superconductors
## I Introduction
Since the discovery of high-$`T_\text{c}`$ superconductivity (HTSC) in La<sub>2</sub>CuO<sub>4</sub>,Bednorz a vast amount of work has been done on slightly doped quasi-two dimensional (2D) antiferromagnets.Barnes ; Dagotto94 ; Orenstein ; NCYeh ; Norman It has been observed that these materials display very unusual properties, with a rich variety of temperature-doping phases diagram. Specifically, away from the overdoped side, the cuprates do not appear to be a Landau Fermi liquid. For instance, they should be considered as doped Mott insulators. However, the theoretical status of the field has been largely phenomenological and controversial.NCYeh ; Norman As far as we know, there is no consensus on the origin of the superconductivity nor on the pseudogap phase. Therefore, finding a microscopic mechanism for HTSC still is an open problem.
Local-density approximation (LDA) and generalized gradient approximation (GGA)Perdew to density functional theory have been used so far to rationalize the electronic structure of HTSCs. Although LDA is a useful technique for some materials, it has been shown that both LDA or GGA are not appropriate for antiferromagnetic materials because they tend to yield a metallic ground state with incorrect delocalized spin density and band ordering.Pickett ; Moreira This is attributed to an extreme nonanalytic and nonlocal behavior of density functional theory as the particle number is changed,Godby ; Hybertsen86 implying the need for a self-energy correction, or at least an orbital dependent potential to obtain a realistic description of band gaps. To overcome such a problem, different semiempirical corrections to LDA have been proposed so far as, i.e., LDA+SIC (Refs. S-Gunnarsson, ; Svane, ; Szotek, and LDA+$`U`$.Czyzyk ; Anisimov ; Wei
An alternative approach to the electronic structure of the HTSCs is based on the use of model Hamiltonians that aim to incorporate the essential physics into a few parameters. It is generally accepted that electron correlation is important for HTSC. Furthermore, it is well known that the (covalent-structure) valence-bond (VB) model or, equivalently, the Heisenberg Hamiltonian includes most of the electron correlation. Thence, early in 1987, AndersonAnderson87 proposed that the important features of the undoped HTSC parent compounds can be described by a Heisenberg Hamiltonian on a two-dimensional square lattice with one electron per site. Meanwhile, EmeryEmery87 ; Emery88 proposed a three-band Hubbard model. Unfortunately, the number of parameters of a three-band Hubbard model turns to be too large. Therefore, Zhang and RiceZhang proposed a simplification of the three-band Hubbard model into the well known $`t`$-$`J`$, which implicitly includes the O($`p`$)-Cu($`d`$) hybridization and recovers the initial effective one-band description of Anderson. Since an appropriate parametrization is essential for the predictive capability of model Hamiltonians, much progress has been achieved on the high-level *ab initio* computation of reliable appropriate parameters using only the crystal structure as external input.CPL ; Calzado ; Munoz
Stimulated by Anderson’s suggestion,Anderson87 a renewed interest of low-dimensional quantum spin-1/2 antiferromagnetic systems emerged. According to the Lieb and Mattis theoremLieb-Mattis the ground state for the undoped half-filled bipartite system must be a singlet. Therefore, the appropriate ground-state wave function could have a resonating-valence-bond (RVB) character. It was soon pointed out that short-range RVB wave functions exhibit topological long-range order.KivelsonRS ; Kivelson ; PRB-91 ; KleinZV Furthermore, recentlyHansson topological order for superconductors has also been claimed, away from truly microscopic models, making use of bosonic theories of the quantum Ginzburg-Landau form. In Ref. PRB-91, Klein and collaborators investigated the short-range RVB wave functions within a dimer coverings approximation for the square lattice and found that the dimer-coverings show a type of long-range spin-pairing order (LRSPO). Using arguments based on the LRSPO they predicted a per-site energy $`\epsilon \delta ^2`$, where $`\delta `$ is the deviation of the local LRSPO with respect to the LRSPO of the ground state. Furthermore, topological spin defects (TSDs), namely a site that is not spin paired to a singlet, or a hole in hole-doped superconductors, or a doubly occupied site in electron-doped materials, were assimilated to Bloch walls separating phases with a difference in LRSPO of $`\pm 1`$. It was argued that, at a longitudinal distance $`\rho `$ past the TSD on the less stabilized side, the defect should also presumably have only spread out a transverse distance $`\rho `$, so that $`\delta 1/\rho `$, and $`\epsilon 1/\rho ^2`$. Therefore, the energy contribution from all the sites of the given longitudinal distance past the TSD is $`\rho \mathrm{\Delta }\epsilon 1/\rho `$. Summation over all the sites up to a given distance thence gives an energy cost of $`\mathrm{ln}\rho `$. When the TSD are charged, it was also suggested that this long flat attraction $`\mathrm{ln}\rho `$ along with the screened repulsion $`\mathrm{exp}\{\alpha \rho \}/\rho `$ could lead to a weakly bound pair. Recently,CPL a linear relationship between $`T_c`$ and the $`J/t`$ ratio, as obtained from high-level *ab initio* calculations, was found. It was argued that such a linear relation arises from the LRSPO mechanism previously suggested.PRB-91 Furthermore, the so-called $`t`$-$`J`$ Hamiltonian for the cuprates seems to be pointing to the right direction.
The existence of a LRSPO for more general RVB wave functions has been proven for ladderlike quantum spin-1/2 antiferromagnetic systems.EPJ ; capitolVB Most of the considerations associated to the existence of this LRSPO for the ladderlike quantum spin-1/2 antiferromagnetic systems are readily applicable to the square lattice. In particular, bound pairs of TSD are predicted to occur. However, as far as we know, the energy of such a pair of TSD as a function of the distance has not been obtained yet. Even more, arguments based on the LRSPO alone cannot decide if vacancies (doubly occupied sites), let’s say *charge-wearing* TSDs, organize themselves as bound pairs of two charge-wearing TSDs, as bosonic-character pairs, or each charge-wearing TSD would bind to a non spin-paired spin, let us say a *spin-wearing* TSD, leading to a fermionic-character pair.
Here focus is on the energy associated with these bosonic and fermionic pairs as described by symmetry-adapted extended wave functions. We find that the fermionic pairs are favored for low doping levels, but the Fermi level increases with doping while the energy of the bosonic pairs lowers. At a critical doping the energy of the bosonic pair goes below the energy of two fermionic pairs at the Fermi level, suggesting the pairing of charge-wearing TSDs, and hence providing a microscopic mechanism for HTSC.
The description of these bosonic and fermionic pairs is based on a $`t`$-$`t^{}`$-$`J`$-like model Hamiltonian $`H=H_I+H_J+H_t+H_t^{}`$, where $`H_I`$ is the energy associated with the ionization potential for hole-doped materials, or the energy associated with the electron affinity for the electron-doped systems. The $`H_J`$ is the well known nearest-neighbor Heisenberg Hamiltonian,
$$H_J=J\underset{𝐑,𝐑^{}}{}𝐒_𝐑𝐒_𝐑^{},$$
(1)
where $`𝐒_𝐑`$ is the spin operator for the spin on the site $`𝐑`$, and $`𝐑,𝐑^{}`$ means that $`𝐑`$ and $`𝐑^{}`$ are nearest neighbors. The nearest- and next-nearest-neighbor hopping contributions to the Hamiltonian are, respectively,
$`H_t`$ $`=`$ $`t{\displaystyle \underset{𝐑,𝐑^{}}{}}{\displaystyle \underset{\sigma }{}}\left(c_{𝐑\sigma }^{}c_{𝐑^{}\sigma }+c_{𝐑^{}\sigma }^{}c_{𝐑\sigma }\right)`$ (2)
$`\times `$ $`\left(1\widehat{n}_{𝐑\overline{\sigma }}\right)\left(1\widehat{n}_{𝐑^{}\overline{\sigma }}\right),`$
$`H_t^{}`$ $`=`$ $`{\displaystyle \underset{𝐑,𝐑^{}}{}}t_{𝐑,𝐑^{}}^{}{\displaystyle \underset{\sigma }{}}\left(c_{𝐑\sigma }^{}c_{𝐑^{}\sigma }+c_{𝐑^{}\sigma }^{}c_{𝐑\sigma }\right)`$ (3)
$`\times `$ $`\left(1\widehat{n}_{𝐑\overline{\sigma }}\right)\left(1\widehat{n}_{𝐑^{}\overline{\sigma }}\right),`$
where $`c_{𝐑\sigma }^{}`$ ($`c_{𝐑\sigma }`$) creates (destroys) an electron on site $`𝐑`$ with spin $`\sigma =\alpha `$, $`\beta `$. The double occupancy is avoided by the factors $`1\widehat{n}_{𝐑\overline{\sigma }}`$, where $`\widehat{n}_{𝐑\overline{\sigma }}`$ is the number operator on site $`𝐑`$ with spin $`\overline{\sigma }=\beta `$, $`\alpha `$. The summation on $`𝐑,𝐑^{}`$ means that $`𝐑`$ and $`𝐑^{}`$ are next-nearest neighbors. The hopping integral $`t_{𝐑,𝐑^{}}^{}`$ depends on the number of holes within the plaquette. Such a model is known to reproduce the low-energy spectrum of the three-band Hubbard model.Hybertsen90 Here, we use the parameters obtained by high-level *ab initio* calculations using only the crystal structure as external input.CPL ; Calzado We approximate the screened electrostatic repulsion within the bosonic pair by the Yukawa potential,Ashcroft ; Madelung the screening agent being the gas of the fermionic pairs. The three-dimensional (3D\] character of the screening is taken into account by an interlayer hopping integral $`t_{}`$. The Heisenberg part of the energy associated with a pair of static TSDs is estimated by the dimer-covering-*counting* approximationSeitz ; Klein-86 ; Zivkovic on $`w\times L`$ antiferromagnetic quantum spin-1/2 square lattice, with $`w=4,6,\mathrm{},20`$, $`L\mathrm{}`$, and cyclic boundary conditions in both directions. Counting of the dimer-covering configurations has been achieved by a transfer-matrix technique.EPJ ; capitolVB
This paper is organized as follows: In Sec. II we review the main concepts about LRSPO,EPJ ; capitolVB describing the scenario where the TSDs are located. In Sec. III the energy per site of the half-filled ground state, and the gain of the Heisenberg energy associated with a static pair of TSDs is estimated. In Sec. IV the symmetry-adapted extended wave-functions appropriate for bosonic and fermionic pairs moving in a CuO<sub>2</sub> layer will be defined, and the energy bands will be obtained. From the two-fluids equilibrium condition, the critical doping $`p_c`$ for the onset of pairing to bosonic pairs among charge-wearing TSDs is obtained. Finally a summary and the conclusions are given in Sec. V.
## II Long-range spin-pairing order and topological spin defects
From the Lieb and Mattis theoremLieb-Mattis it is well known that for bipartite spin systems a *maximally-spin-paired* ground state is expected. In particular, at half filling, for ladder-like systems, with equal number of sites in the $`𝒜`$ and $``$ sublattices, the ground state is a singlet. Singlet states can be achieved by configuration interaction (CI) among covalent VB configurations or RVB. For instance, a linearly independent set of VB configurations can be achieved by pairing to a singlet each spin in the sublattice $`𝒜`$ to a spin in the sublattice $``$, not necessarily one of its nearest neighbors (see Fig. 1).
It is knownEPJ ; capitolVB that any (covalent) VB configuration exhibits a LRSPO related to the local (at boundary) array of SPs penetrating any boundary $`f_n`$ (see, for instance, Fig. 1). The parameter associated with the LRSPO, $`D`$, can take $`w+1`$ different relevant values. The shape of the boundary selected to define the different $`w+1`$ values of $`D`$ is quite arbitrary, though when $`w`$=even and the boundary is chosen to run parallel to rungs, the $`w+1`$ different values of $`D`$ are:
$$D=0,\pm 1,\mathrm{},\pm \frac{w}{2}.$$
(4)
This LRSPO allows to separate the set of VB configurations in different subsets. Since two singlets from different subsets must be different repeatedly at every position along the ladder, they are asymptotically orthogonal and non interacting via any interaction mediated by a few-particle operator. Then the matrix of the Hamiltonian asymptotically block-diagonalizes, so configurations belonging to different subsets do not mix in the CI sense. Thus $`D`$ may be taken as a long-range order parameter labelling the eigenstates of the $`D`$ block. Under low-frustration conditions, the expected ordering of the lowest-lying energy $`E_D`$ from the different blocks is
$$E_0<E_1<\mathrm{}<E_{w/2},$$
(5)
with $`E_D=E_D`$.
Now, half-filled excited states or slightly doped states are analyzed via TSDs. There are different types of excitations conceivable from a *maximally-spin-paired* ground state. Let us say, preserving half filling (one electron per site), there are primarily spin excitations. In this case, two spin-wearing TSDs, one in the sublattice $`𝒜`$ and the other in the sublattice $``$, are obtained by breaking one SP to form a triplet state. Away from half-filling, there are low-energy spin and charge excitations. For instance, removing (adding) one electron produces two sites that cannot be SP, a charge-wearing TSD and a spin-wearing TSD, one in the sublattice $`𝒜`$ and the other in the sublattice $``$, the ladder becoming a doublet. In this case hopping terms must be retained in the Hamiltonian and the so-called $`t`$-$`J`$ model or different extensions that incorporate either next-nearest-neighbor hopping $`t^{}`$ or electrostatic repulsion have been employed so far. Thence, the doublet is a weighted superposition of VB configurations with a spin-wearing TSD and a charge-wearing TSD lying in different sublattices. Still, going up in the hierarchy of Hamiltonians, the Hubbard or even a more general Hamiltonian must be considered. In this case, still another type of excitations (though presumably of higher energy if a Heisenberg-type Hamiltonian is assumed to govern the lowest-lying region of the spectrum) can be produced relaxing the single-occupancy constrain. This leads to the *ionic* states, i.e., states with at least a pair of sites, one doubly occupied and the other empty, namely one negatively charge-wearing TSD and one positively charge-wearing TSD.
Of special relevance here is how the LRSPO is disrupted by a TSD (see Fig. 2). For instance, a TSD in a site $`[n,i]`$, $`n`$ indicating the rung and $`i`$ the leg, can be seen as a domain wall on the rung $`n`$ which separates the lattice in two sectors with associated left, $`D_l`$, and right, $`D_r`$, order parameters. When we choose the sublattice $`𝒜`$ as formed by the set of sites $`[m,j]`$ with $`m+j`$=even,
$$D_r=D_l(1)^{n+i}.$$
(6)
Thence, to fulfill boundary conditions TSDs must appear by pairs, one TSD in the sublattice $`𝒜`$ and the other in the sublattice $``$, to ensure $`\mathrm{\Delta }D=0`$ from the left to the right of the pair. Such a pair define an intervening region with $`\mathrm{\Delta }D=\pm 1`$ with respect to the LRSPO $`D`$ of the host (see Fig. 2). Then, away from half-filling, it may be conceivable an intervening region limited by two charge-wearing TSDs, or a charge-wearing TSD and a spin-wearing TSD (provided that the doping is not so strong as to preclude a maximally-spin-paired ground state). In particular, when placing a pair of TSDs above the ground state ($`D`$=0), the order parameter of the intervening region will be $`|D_p|`$=1, which from Eq. (5) is expected to have higher associated energy per site. This indicates that the pair of TSDs should try to remain as close as possible. Thus, bound pairs of TSDs are predicted to occur. To show that this is the case is one of the concerns of the present paper.
## III Heisenberg energy of a static pair of TSDs
Within the dimer-covering-*counting* approximation the *resonance energy*, $`E_r(w,D)`$ in units of $`J`$, i.e., the energy correction below the energy of a single dimer-covering structure, depends on the configuration interaction among the different dimer-covering configurations with LRSPO $`D`$. When an equally weighted wave function is considered, it has been arguedSeitz ; Klein-86 ; Zivkovic that one might consider this energy lowering to depend solely on the dimension of the space spanned by the appropriate dimer-covering configurations. Let $`𝒩_D(w)`$ be the number of linearly independent dimer-covering configurations with the LRSPO $`D`$. Since $`𝒩_D(w)`$ is multiplicative in terms of a break up into subsystems while the energy is additive, such a functional dependence should be of the form
$$E_r(w,D)C\mathrm{ln}𝒩_D(w),$$
(7)
The values $`𝒩_D(w)`$ can be easily obtained by a transfer-matrix technique.EPJ Let us start computing $`𝒩_D(w)`$ for a maximally spin-paired half-filled system. Let us analyze from a local point of view the dimer-covering singlets. We can identify the dimer-covering local states according to which legs have a pairing across the $`f_n`$ boundary. In the present case it can be seen that, for each boundary, there are $`2^w`$ different local states, $`|e_{nI})`$ ($`I`$ ranging), which can be classified according to the value of $`D`$, $`|e_{nI}^D)`$.
Proceeding from the left to the right, from the boundary $`f_{n1}`$ to $`f_n`$, a dimer-covering-*counting* matrix, $`𝒦_n`$, is defined as $`(e_{n1I}|𝒦_n|e_{nJ})`$ the number of different ways $`|e_{nJ})`$ can succeed $`|e_{n1I})`$. Then, the number of dimer-covering states in a $`D`$ subspace is
$$𝒩_D(w)=\underset{e_{0I}^D}{}(e_{0I}^D|𝒦_1𝒦_2\mathrm{}𝒦_L|e_{0I}^D).$$
(8)
Since $`D_n=D_{n+1}`$ for any dimer-covering singlet, $`𝒦_n`$ is a block-diagonal symmetric matrix that does not depend on $`n`$ and we can omit this subindex. For $`L\mathrm{}`$, the highest eigenvalue $`\mathrm{\Lambda }_{wD}`$ of the $`D`$ block $`𝒦_D`$ dominates, and
$$𝒩_D(w)\mathrm{\Lambda }_{wD}^L.$$
(9)
Therefore,
$$E_D(w)wL\epsilon _0+E_r(w,D)wL\epsilon _0CL\mathrm{ln}\mathrm{\Lambda }_{wD},$$
(10)
where $`\epsilon _0=0.375`$ is the energy per spin of a single dimer-covering configuration. SinceEPJ ; capitolVB
$$\mathrm{\Lambda }_{wD}>\mathrm{\Lambda }_{wD^{}}\text{when}|D|<|D^{}|,$$
(11)
with $`\mathrm{\Lambda }_{wD}=\mathrm{\Lambda }_{w|D|}`$, the Heisenberg energy for the half-filled ground state belongs to the subspace $`D`$=0, as suggested by Eq. (5), and can be approximated (in units of $`J`$) by
$$E_0(w)wL\epsilon _0CL\mathrm{ln}\mathrm{\Lambda }_{w0}.$$
(12)
$`C`$ is a fitting parameter independent of the structure to some degree. The value of $`C`$ for the nearest-neighbor isotropic Heisenberg model has been determined for a class of benzenoid hydrocarbonsSeitz (with $`C`$=0.5667) and for finite square-lattice fragmentsZivkovic (with $`C`$=0.75), by fitting the logarithm of the dimer-covering count to the resonance energy calculated from an equally weighted dimer-covering wave function. For the spin-1/2 square lattice, more general RVB approximations suggestEPJ a rough estimate of $`C=0.94\pm 0.19`$. Here $`C`$ is fixed to yield a reasonably good estimate of the ground-state Heisenberg energy of the half-filled square lattice. Table 1 summarizes the ground-state energy per site for $`w=4,6,\mathrm{},20`$ and its extrapolation to $`w\mathrm{}`$. We use $`C=1`$ from here on, since $`C1.0083`$ yields the ground-state energy of Liang *et al.*Liang
When adding a TSD to a CuO<sub>2</sub> layer the transfer matrix $`𝒦`$ across the defect must be substituted by the appropriate $`𝒦_𝐑`$, where $`𝐑`$ is the vector position of the TSD. Therefore, the number of dimer-covering configurations when adding a pair of TSDs to the half-filled ground state, located, respectively, at $`\mathrm{𝟎}`$ and $`[m,j]`$, with non-negative $`m`$ and $`j`$, with $`m+j`$=odd, is
$`𝒩_{[m,j]}(w)`$ $`=\left(\mathrm{\Lambda }_{w0}|𝒦_\mathrm{𝟎}𝒦^{m1}𝒦_{[m,j]}𝒦^{Lm1}|\mathrm{\Lambda }_{w0}\right)`$ (13)
$`\mathrm{\Lambda }_{w0}^{Lm1}(\mathrm{\Lambda }_{w0}|𝒦_\mathrm{𝟎}𝒦^{m1}𝒦_{[m,j]}|\mathrm{\Lambda }_{w0}).`$
Thence, the Heisenberg energy (in units of $`J`$) associated with a pair of static TSDs separated $`[m,j]`$, $`m+j`$=odd, with respect to the energy of the half-filled ground state is
$$\epsilon _{[m,j]}(w)2\epsilon _0+\mathrm{ln}\frac{\mathrm{\Lambda }_{w0}^{m+1}}{(\mathrm{\Lambda }_{w0}|𝒦_\mathrm{𝟎}𝒦^{m1}𝒦_{[m,j]}|\mathrm{\Lambda }_{w0})}.$$
(14)
Table 2 summarizes the energies $`\epsilon _{[m,j]}(w)`$ from $`[m,j]=[1,0]`$ to $`[7,4]`$ and $`w=4`$, 6, $`\mathrm{}`$, 20. The $`w\mathrm{}`$ limit, $`\epsilon _{[m,j]}`$, is obtained by fitting $`\epsilon _{[m,j]}(w)`$ by a power series in $`1/w`$.
For moderate to long distances, our results indicate that the Heisenberg energy of such a static excitation increases as $`\mathrm{ln}\rho `$, as predicted in Ref. PRB-91, . Nevertheless, a tiny deviation from this behavior is observed for small values of $`\rho `$. This is because details of the lattice are more important for short distances, as also is expected from the form of the denominator in Eq. (14). See, for instance, Fig. 3. Therefore, it is expected that the TSDs of a pair will try to remain as close as possible. However, this is not enough to decide whether charge-wearing TSDs organize themselves as bound pairs of two charge-wearing TSDs, with bosonic character, or each charge-wearing TSD would bind to a spin-wearing TSD, leading to a fermionic-character pair.
## IV Two-dimensional extended wave functions
The bosonic or the fermionic pairs are far from being static. The hopping terms of the Hamiltonian allow any charge-wearing TSD to move while the exchange part mixes up all the VB configurations. Therefore, the appropriate wave function must be a weighted superposition of all possible static configurations, fulfilling translational and point group symmetry conditions. Thence, the wave functions for both bosonic and fermionic extended pairs of TSDs should be invariant under the operations of the factor group isomorphic to the $`C_{4v}`$ (4 mm) group, as obtained by factorizing the full group into the translational subgroup and the planar subgroup. Thence, there can be conceivable extended wave functions with symmetry $`𝒮`$,
$$𝒮=\{\begin{array}{cc}𝒜_1,\hfill & \text{totally symmetric},(x^2+y^2),\hfill \\ 𝒜_2,\hfill & \text{antisymmetric under the four reflections},\hfill \\ _1,\hfill & \text{antisymmetric under }C_4^\pm \text{}\sigma _{x\pm y},(x^2y^2),\hfill \\ _2,\hfill & \text{antisymmetric under }C_4^\pm \text{}\sigma _x\text{}\sigma _y,(xy).\hfill \end{array}$$
(15)
Therefore, symmetry-adapted extended wave functions for both the fermionic pairs and bosonic pairs can be written as
$$\varphi _{[m,j]}^𝒮(𝐤)N_{[m,j]}\underset{𝐑}{\overset{}{}}\text{e}^{\text{i}𝐤𝐑}\underset{𝝆_{[m,j]}}{}|𝐑,𝐑+𝝆_{[m,j]}\chi _{𝝆_{[m,j]}}^𝒮,$$
(16)
where $`N_{[m,j]}`$ is the normalization; $`𝝆_{[m,j]}`$ is a vector obtained by any operation of the point group acting on $`[m,j]`$, with $`0j<m`$ and $`m+j`$=odd; $`|𝐑,𝐑+𝝆_{[m,j]}`$ is the state with a static pair of TSDs, let’s say, a charge-wearing TSD lying on site $`𝐑`$, and a spin-wearing TSD (a second charge-wearing TSD) on $`𝐑+𝝆_{[m,j]}`$ for the fermionic (bosonic) pairs; $`\chi _{𝝆_{[m,j]}}^𝒮`$ is the appropriate character of the irreducible representation $`𝒮`$, with $`\chi _{[m,j]}1`$. Finally, $``$ is the square lattice ($`=𝒜`$) for the fermionic (bosonic) pair. Then, care must be taken with the allowed values for $`𝐤`$. For instance, when dealing with the fermionic pairs, $`𝐤`$ belongs to the Brillouin zone of a square lattice with lattice constant $`a`$. On the other hand, for bosonic pairs $`|k_x|,|k_y|\pi /2a`$, because the summation is restricted to run on the sublattice $`𝒜`$.
It can be readily seen that only the $`𝒜_1`$ and $`_1`$ symmetries are allowed for $`j`$=0. Since different symmetries do not mix, here we restrict ourselves to $`𝒜_1`$ and $`_1`$ symmetries even when $`j0`$.
### IV.1 Energy of the fermionic pairs
The expectation values given by the wave functions of Eq. (16) are
$`\zeta _{[m,j]}^𝒮(𝐤)`$ $``$ $`I{\displaystyle \frac{1}{2}}t\left(\mathrm{cos}k_xa+\mathrm{cos}k_ya\right)\delta _{[m,j]}^{[1,0]}`$ (17)
$`+`$ $`\delta _{j,m1}\left(1+\delta _{[m,j]}^{[1,0]}\right)t_1^{}\chi _{[j,m]}^𝒮\mathrm{cos}k_xa\mathrm{cos}k_ya`$
$`+`$ $`J\left(\epsilon _{[m,j]}+\gamma _{[m,j]}\right),`$
where $`I`$ is the ionization potential (electron affinity) for hole-doped (electron-doped) materials. $`t`$ and $`t_i^{}`$ are nearest- and next-nearest-neighbor hopping integrals, with $`i=1`$ when there is only one charge-wearing TSD in the plaquette, and $`i=2`$ when there are two nearest-neighbor charge-wearing TSDs in the plaquette; $`\gamma _{[m,j]}`$ arises from the Heisenberg terms involving the spin-wearing TSD (see fig. 4),
$$\gamma _{[m,j]}\{\begin{array}{cc}\frac{3}{4}\frac{1}{3}\chi _{[0,1]}^𝒮,\hfill & m=1,\hfill \\ 1\frac{1}{8}\delta _{j,1}\frac{1}{4}\chi _{[j,m]}^𝒮\delta _{j,m1},\hfill & \text{otherwise}.\hfill \end{array}$$
(18)
There are two families of nonzero off-diagonal matrix elements of the Hamiltonian. When $`mm^{}=\pm 1`$ and $`jj^{}=\pm 1`$,
$$H_{[m,j]}^{[m^{},j^{}]}=\lambda \left(t_1^{}\mathrm{cos}k_xa\mathrm{cos}k_ya\frac{1}{4}J\frac{1}{8}J\delta _{j,m1}\chi _{[j,m]}^𝒮\right).$$
(19)
When $`m^{}m=\pm 2`$ and $`j^{}=j`$ or $`m^{}=m`$ and $`j^{}j=\pm 2`$,
$$H_{[m,j]}^{[m^{},j^{}]}=\{\begin{array}{cc}\sqrt{3}J/12,\hfill & [m^{},j^{}]\text{ or }[m,j]=[1,0],\hfill \\ \lambda J/8,\hfill & \text{otherwise},\hfill \end{array}$$
(20)
with $`\lambda =\sqrt{2}`$ when either $`j`$ or $`j^{}`$ is zero (but not both), and $`\lambda =1`$ otherwise.
The zero-order lowest-lying fermionic pairs are the $`[1,0]`$. Close to $`\mathrm{\Gamma }`$ the energy of these fermionic pairs is
$$\zeta _{[1,0]}^𝒮(𝐤)\zeta _\mathrm{\Gamma }^𝒮+\frac{\mathrm{}^2k_{}^2}{2m_{}^𝒮}$$
(21)
with
$`\zeta _\mathrm{\Gamma }^𝒮`$ $``$ $`It+2t_1^{}\chi _{[0,1]}^𝒮+J\left(\epsilon _{[1,0]}+{\displaystyle \frac{3}{4}}{\displaystyle \frac{1}{3}}\chi _{[0,1]}^𝒮\right)`$
$`{\displaystyle \frac{\mathrm{}^2}{2m_{}^𝒮}}`$ $``$ $`\left({\displaystyle \frac{1}{4}}tt_1^{}\chi _{[0,1]}^𝒮\right)a^2`$ (22)
Thence, for $`t_1^{}>J/6`$, the zero-order lowest-lying band has $`x^2y^2`$ symmetry. This is the case for the La<sub>2-x</sub>Sr<sub>x</sub>CuO<sub>4</sub> (Ref. Calzado, ) (LSCO). For the monolayered HTSC of Table 1 in Ref. CPL, we know that $`J/60.019\text{}0.030`$ eV. On the other hand, strong differences on hopping integrals among the different HTSC are not expected, as suggested by the small variations observed on the nearest-neighbor hopping integral, $`t`$. Therefore, we expect that the zero-order lowest-lying band will show $`_1`$ symmetry for all of these HTSC.
We are now concerned whether admixing wave functions with different $`[m,j]`$ to the $`\varphi _{[1,0]}^𝒮(𝐤)`$ would be relevant or even if the ordering of the lowest-lying $`𝒜_1`$ and $`_1`$ bands could be reversed. We have obtained the corrections to $`\zeta _{[1,0]}^{𝒜_1}(𝐤)`$ and $`\zeta _{[1,0]}^_1(𝐤)`$ by diagonalizing the matrix of the Hamiltonian in the basis of the two, three and four lowest-lying wave functions with the appropriate $`𝒜`$ or $``$ symmetries. Thence, making use of the parameters for the LSCO,Calzado when the number of fermionic pairs per Cu is $`p=0.05\text{}0.07`$ we obtain corrections to the zero-order Fermi energy of $`\mathrm{\Delta }^{(2)}=16.98`$ to $`14.56`$ meV, $`\mathrm{\Delta }^{(3)}=1.45`$ to $`1.33`$ meV, and $`\mathrm{\Delta }^{(4)}=0.26`$ to $`0.20`$ meV for the $`_1`$ band, while for the $`𝒜_1`$ the corrections are $`\mathrm{\Delta }^{(2)}=7.21`$ to $`5.1`$ meV, $`\mathrm{\Delta }^{(3)}=3.15`$ to $`2.7`$ meV, and $`\mathrm{\Delta }^{(4)}=0.15\text{ to }0.1`$ meV. For the monolayered HTSC of Table 1 in Ref. CPL, , assuming $`t_1^{}0.2t`$, these corrections are slightly decreasing with $`J/t`$. We observe that, up to meV, the correction to the $`𝒜_1`$ energy is smaller than the correction to the $`_1`$. Therefore, it is expected that the lowest-lying band still has $`x^2y^2`$ symmetry. Furthermore, the band with symmetry $`𝒜_1`$ would not start filling until a critical doping of $`p0.20\text{}0.22`$ holes per CuO<sub>2</sub> unit, provided that all the charge-wearing TSDs organize as fermionic pairs. At this doping, the corrections $`\mathrm{\Delta }^{(n)}`$ to the energies are still smaller than those referred above. Since this doping is out of the range of our interest, we restrict ourselves to consider only the band with $`x^2y^2`$ symmetry. Also, since the error in the parameters of the Hamiltonian are of the order of meV, we neglect corrections to the energy smaller than 1 meV. Therefore, we consider only the $`[1,0]`$, $`[2,1]`$, and $`[3,0]`$ wave functions to describe the lowest-lying band of the fermionic pairs.
### IV.2 Energy of the bosonic pairs
When dealing with $`t`$-$`J`$-like model Hamiltonians, the electrostatic repulsion is generally neglected, although with some exceptions.Kivelson90 ; Dagotto92 Since the *screened* electrostatic repulsion between the charge-wearing TSDs in a pair, $`V_{[m,j]}`$, may be relevant, here it is included in the diagonal terms of the Hamiltonian,
$$H_{[m,j]}^𝒮=J\epsilon _{[m,j]}+2I+V_{[m,j]}+H_t+H_t^{}_{[m,j]}^𝒮.$$
(23)
It is not difficult to obtain $`H_t_{[m,j]}^𝒮=0`$, and
$$H_t^{}_{[m,j]}^𝒮=\kappa _{[m,j]}\chi _{[j,m]}^𝒮\left(1+\mathrm{cos}k_xa\mathrm{cos}k_ya\right),$$
(24)
with
$$\kappa _{[m,j]}=\{\begin{array}{cc}2t_2^{},\hfill & [m,j]=[1,0],\hfill \\ t_1^{},\hfill & mj=1,m>1,\hfill \\ 0,\hfill & \text{otherwise}.\hfill \end{array}$$
(25)
The electrostatic repulsion within a bosonic pair is expected to be screened by the gas of the fermionic pairs. However, the fermionic pairs have been considered so far as moving in a two-dimensional square lattice. It is generally accepted that the $`c`$ axis effect is simply to tune the electronic structure of the CuO<sub>2</sub> planes. Nevertheless, screening is a three-dimensional effect that could be taken into account by an interlayer hopping integral $`t_{}`$. Considering a nonzero $`t_{}`$ would imply a correction to the energy $`t_{}k_{}^2c^2`$, where $`c`$ is the lattice constant perpendicular to the $`ab`$ layers. Since $`t_{}`$ is rather small,Zha it can be neglected for the energy-balance considerations, but it is essential for screening purposes. Then, for the electrostatic repulsion within a bosonic pair we take as a first approximation the Yukawa potentialAshcroft ; Madelung as the dominant term,
$$V_\rho \frac{q^2}{\rho }\mathrm{exp}\left\{\left(4\pi e^2g_F\right)^{1/2}\rho \right\},$$
(26)
where $`g_F`$ is the density of states at the Fermi level of the fermionic pairs per unit of volume of the solid,
$$g_F\left(\frac{3m_{}m_{}^2\nu p}{\pi ^4\mathrm{}^6a^2c}\right)^{1/3},$$
(27)
$`p`$ being the number of fermionic pairs per site, and $`\nu `$ is the number of square-lattice layers cutting a unit cell. Therefore, close to $`\mathrm{\Gamma }`$, the diagonal terms of the Hamiltonian are
$`H_{[m,j]}^𝒮`$ $``$ $`J\epsilon _{[m,j]}+2I+{\displaystyle \frac{e^2}{\rho }}\mathrm{exp}\left\{\beta \rho (\nu p)^{1/6}\right\}`$ (28)
$`+`$ $`\kappa _{[m,j]}\chi _{[j,m]}^𝒮\left(2{\displaystyle \frac{1}{2}}a^2k^2\right),`$
with
$$\beta \frac{2e}{a}\left(\frac{6}{\pi t_{}(t+4t_1^{})^2c^3}\right)^{1/6}.$$
(29)
The non-zero off-diagonal elements of the Hamiltonian can also be readily obtained,
$$H_{[m,j]}^{[m^{},j^{}]}=\lambda t_1^{}(1+\mathrm{cos}k_xa\mathrm{cos}k_ya);|mm^{}|,|jj^{}|=1,$$
(30)
with $`\lambda =\sqrt{2}`$ when either $`j`$ or $`j^{}`$ is zero, and $`\lambda =1`$ otherwise.
Since the nonzero off-diagonal elements of the Hamiltonian are important as compared to the differences among the diagonal elements, the energy of the bosonic pairs must be obtained by diagonalizing the matrix of the Hamiltonian.
Since the screened electrostatic repulsion decays faster than exponentially, while $`\epsilon _{[m,j]}`$ increases logarithmically, there must be a minimum in the energy and confinement is expected to occur. Furthermore, from Eq. (28) and assuming that $`t_1^{}`$, $`t_2^{}>0`$,Calzado it is expected that the bosonic pairs will also show $`x^2y^2`$ symmetry as it is generally accepted.Dagotto94 At this point, it is worth noting that if the next-nearest-neighbor hopping is neglected and the Hamiltonian is reduced to the $`t`$-$`J`$ model, it turns out that the $`𝒜_1`$ and $`_1`$ symmetries would be degenerate.
### IV.3 Two-fluids equilibrium condition
At $`T=0`$, the question now is whether this lowest-lying bosonic pairs would have lower energy than two fermionic pairs in its Fermi level, so the bosonic pairs would be favored. At low doping level, it is expected that the fermionic pairs will be favored. Nevertheless, the Fermi level, $`k_F\sqrt{2\pi p}/a`$, increases linearly with $`p`$, while the electrostatic repulsion among the two charge-wearing TSDs in the bosonic pair is exponentially reduced with $`p^{1/6}`$, so its ground-state energy is lowered. Therefore, we wonder whether there exist a critical value $`p_c`$ such that the ground-state of the bosonic pairs, as measured with respect twice the energy of a fermionic pair at its Fermi level, $`\mathrm{\Delta }(p)`$, is zero.
To explore such a possibility, we have diagonalized the matrix of the Hamiltonian for $`𝐤=0`$, and increasing values of $`p`$. To reach corrections to the ground-state energy within the order of meV, we have considered up to a $`12\times 12`$ matrix involving all the states which would contribute to perturbation theory truncated to tenth order. For a certain regime of parameters, at low enough doping level, the fermionic pairs are favorable. As $`p`$ increases there exist a critical value of doping, $`p_c`$, such that $`\mathrm{\Delta }(p_c)=0`$ for the lowest-lying bosonic pairs.
Doping above $`p_c`$ yields bosonic pairs. In this case, the bosonic pairs are expected also to contribute to the screening and the electrostatic repulsion could become negligible. If so, there would be a cascade process of pairing among the fermionic pairs until a new equilibrium between the two fluids is reached at $`p_f<p_c`$. Thence, we expect the number of bosonic pairs at $`p>p_f`$ to be $`p_b(pp_f)/2`$. There is a lower limit of $`p_f`$ such that $`\mathrm{\Delta }_0(p_f)=0`$, as obtained when the electrostatic repulsion is completely neglected.
For the sake of estimating the order of magnitude of $`p_c`$ and $`p_f`$ for a generic HTSC, let us make use of the parameters appropriate for LSCO. We take $`a3.8`$ Å and $`c/a3.47`$ from Ref. Dagotto94, . For the Hamiltonian parameters, we take $`J=0.144`$ eV and $`t=0.549`$ eV from Ref. CPL, , and $`t_2^{}=0.130`$ eV and $`t_1^{}=0.112`$ eV from Ref. Calzado, . All of these parameters were obtained from high-level *ab initio* calculations with the geometry as the only external input, being the errors within the meV. Since there is no high-level *ab initio* calculation for the interlayer hopping integral, we use the low-doping $`t_{}0.7`$ meV obtained from experimental results by Zha, Cooper, and Pines.Zha Within these parameters regime we get $`\beta 7.8/a`$. Computing $`\mathrm{\Delta }(p)`$ for increasing values of $`p`$ we find that $`\mathrm{\Delta }(p)`$ is changing its sign at $`p_c0.0524`$ (see, for instance, Fig. 5).
At $`p_c`$, the mean distance between the two holes of the pair is $`\rho _c9.08`$ Å with a standard deviation $`\sigma _c5.08`$ Å. Identifying in a rather loose way the spatial extent of the pair wave function ($`\rho _c+\sigma _c`$) with the coherence length $`\xi `$, we obtain $`\xi 14.16`$ Å, in good agreement with the in-plane value ($`\xi 14`$-$`15`$ Å)Barnes ; Dagotto94 suggested from experimental findings.
On the other hand, when the electrostatic repulsion is neglected, $`\mathrm{\Delta }_0(p)`$ changes its sign at $`p_f0.0505`$, and the mean distance between the two holes is $`\rho _f8.78`$ Å with a standard deviation $`\sigma _f4.99`$ Å. Therefore, the estimated coherence length is $`\xi 13.77`$ Å. Again, it is worth noting that if the next-nearest-neighbor hopping is neglected and the Hamiltonian is reduced to the $`t`$-$`J`$ model, the value of the critical doping ($`p_c0.28`$) is out of the range where the superconductivity is observed. In addition, at so high doping the validity of such model Hamiltonians could be questioned.
As far as we know, for other monolayered cuprate superconductors only the $`t`$ and $`J`$ parameters have been obtained from high-level *ab initio* calculations. Nevertheless, for the purpose of estimating how $`p_c`$ and $`p_f`$ vary with $`J/t`$, let us assume that the ratios $`t_1^{}/t`$ and $`t_2^{}/t`$, as well $`a\beta `$ do not change very much among them, and use the values appropriate for LSCO. If so, we find that $`p_f`$ and $`p_c`$ increase as $`J/t`$ decreases. See, for instance, Fig. 6, where we also include the $`p_f`$ and $`p_c`$ values for LSCO as a function of $`J/t`$. This result suggests that for low $`J/t`$ the onset of superconductivity would be located at a too high level of doping such that it could be beyond the validity of the present approximation. Therefore, the superconductivity could be suppressed.
Furthermore, it is worth noting here that the parameters that characterize a superconductor are taken as independent of doping. Nevertheless, $`J`$ as well as the hopping integral do depend locally on doping, as suggested by the high-level emphab initio calculations of Calzado and Malrieu.Calzado For instance, their calculations suggest that $`J`$ decreases, while $`t`$ increases, with doping. Therefore, it is expected that $`J/t`$ decreases with doping, probably not linearly (see, for instance, the dotted lines of Fig. 6). Consequently, $`p_b`$ would decrease, eventually down to zero at a critical doping $`p_c^{}`$ such that the $`J/t`$ as a function of the doping crosses again the $`p_f`$ function. Thence, the superconductivity would be suppressed in the overdoped regime for a doping $`p>p_c^{}`$. Therefore, a better knowledge of the parameters is essential to fully understand the phenomenon of HTSC.
## V Summary and Conclusions
We have shown that the Heisenberg energy associated with a pair of TSD in a spin=1/2 square lattice increases logarithmically with distance. Therefore, a charge-wearing TSD (either a hole or a doubly occupied site) in a spin=1/2 square lattice binds to another TSD, either to a spin-wearing TSD or to another charge-wearing TSD.
We have constructed symmetry-adapted extended wave functions for both a fermionic pair of a charge-wearing TSD and a spin-wearing TSD, and a bosonic pair of two charge-wearing TSD. The energy associated with such fermionic and bosonic pairs has been obtained.
For the lowest-lying fermionic band and the lowest-lying bosonic pairs the symmetry turns to be $`x^2y^2`$ when $`t_1^{}>J/6`$ and $`t_1^{},t_2^{}>0`$, respectively. Since these conditions are fulfilled for monolayered HTSC, we obtain that the symmetry of the bosonic pairs is $`x^2y^2`$ for these materials, as it is generally accepted.Dagotto94
For the LSCO compound, we find a critical doping for bosonic pairing $`p_c0.0524`$ and $`p_f0.0505`$ (when the electrostatic repulsion is completely neglected). This finding could be related to the onset of High $`T_\text{c}`$ superconductivity, the superconducting state being a Bose condensate. This is also compatible with the existence of pairs above $`T_\text{c}`$, a forerunner of the pseudogap physics of the cuprates. At the critical doping, we find a mean distance between the two holes of the pair $`\rho _c9.08`$ Å ($`\rho _f8.78`$ Å), and an estimated coherence length $`\xi 14.16`$ Å ($`\xi 13.77`$ Å). These features are in good agreement with the experimental result of $`p_c0.05`$ (Ref. Yamada, ) and $`\xi 14`$-$`15`$ Å.Barnes ; Dagotto94
For the monolayered cuprate superconductors of Table 1 in Ref. CPL, , we have obtained $`p_f`$ and $`p_c`$ as a function of $`a`$ and $`J/t`$, while keeping fixed $`a\beta 7.8`$ and $`t_i^{}/t`$. See, for instance, Fig. 6. It can be observed that $`p_f`$ and $`p_c`$ increase as $`J/t`$ is lowered.
Extensions of the present work towards $`T0`$, and to other possible charge-wearing TSDs self-organization are in progress. For instance, it is of interest to explore the parameter regime for bosonic pairs with $`xy`$ symmetry, phase separation, and stripe formation.
###### Acknowledgements.
This research was supported by the Spanish DGI (Project No. PFM2002-02629). the author thanks Professor D.J. Klein, Professor A. Labarta, Professor F. Illas and Dr. I. de P. R. Moreira for valuable suggestions. |
warning/0507/hep-th0507184.html | ar5iv | text | # Indice
## Ringraziamenti
Questa lavoro di Tesi è stato svolto presso il Dipartimento di Fisica dell’Università di Roma “Tor Vergata”, e presso il CPHT dell’Ecole Polytechnique di Parigi, in cui ho avuto la fortuna di trovare ambienti di ricerca estremamente stimolanti. Vorrei ringraziare in particolare il Prof. A. Sagnotti per essere essere stato una guida autorevole e disponibile nel vasto mondo della Teoria delle Stringhe e per avermi offerto la preziosa possibilità di collaborare a questo progetto di ricerca. Vorrei ringraziare anche il Prof. Emilian Dudas, con cui ho avuto modo di lavorare a Parigi, per la disponibilità e l’istruttiva collaborazione.
Infine vorrei esprimere la mia riconoscenza ai dottorandi del Dipartimento di Fisica dell’Università di Roma “Tor Vergata” per l’ambiente piacevole e sereno che mi hanno fatto trovare in questi mesi di studio e di ricerca nella mia “nuova” Università. Un particolare ringraziamento va al Dr. Marco Nicolosi per le utili discussioni e consigli.
This work was supported in part by INFN, by the MIUR-COFIN Contract 2003-023852, by the EU contracts MRTN-CT-2004-503369 and MRTN-CT-2004-512194, by the INTAS contract 03-51-6346, and the NATO grant PST.CLG.978785. The author would like to thank the CPhT of the Ecole Poytechnique, and in particular Prof. E. Dudas, for the kind hospitality extended to him while this Thesis was being completed.
## Introduzione
### Il cammino dell’unificazione
La nostra attuale conoscenza delle leggi fondamentali dell’universo poggia su due teorie, il Modello Standard e la Relatività Generale. Entrambe risultano descrivere l’universo correttamente nei limiti delle verifiche empiriche oggi permesse, ma allo stesso tempo esse sono due teorie inconciliabili.
Nel corso degli ultimi cento anni la fisica è avanzata superando di volta in volta le contraddizioni emerse fra le teorie esistenti. Il tentativo di rendere l’elettromagnetismo di Maxwell compatibile con la relatività Galileana portò Einstein alla formulazione della Relatività Ristretta. In seguito, il tentativo di conciliare gravitazione Newtoniana e Relatività Speciale lo portò a sviluppare la Teoria della Relatività Generale. Infine, la Teoria Quantistica dei Campi nacque dal bisogno di rendere compatibili la Meccanica Quantistica e la Relatività Speciale.
Allo stesso tempo lo sviluppo storico della fisica è stato segnato da una straordinaria unificazione nella descrizione di fenomeni apparentemente molto diversi. Il primo grande passo fu l’unificazione dei fenomeni elettrici e magnetici. La teoria elettrostatica, formulata da Cavedish nel periodo dal 1771 al 1773 e completata da Coulomb nel 1785, fu messa in rapporto con i fenomeni magnetici dai lavori di Oersted, che osservò la deflessione dell’ago di una bussola a causa della presenza di correnti (1819), di Biot-Savart e Ampère, che stabilirono le regole con cui le correnti elettriche producono campi magnetici (1820-25), e di Faraday, che mostrò che campi magnetici variabili generano campi elettrici (1873). L’unificazione dei fenomeni elettrici e magnetici fu portata a termine da Maxwell, che costruì un sistema coerente di equazioni in grado di descrivere tutti i fenomeni noti (1831). Le equazioni di Maxwell portarono alla previsione delle onde elettromagnetiche. Una seconda unificazione di tipo differente, ma di grandissima portata, fu prodotta dalla teoria della Relatività Ristretta, formulata nel 1905 da Albert Einstein, che mostrò come lo spazio e il tempo non siano entità separate e assolute, come nella Meccanica Newtoniana, ma formino piuttosto un continuo spazio-temporale che è l’arena in cui i fenomeni fisici si dipanano. Allo stesso modo la teoria, portava all’unificazione dei concetti di massa ed energia, fino ad allora ben distinti. Sulla strada dell’unificazione, un radicale cambiamento di paradigma fu introdotto dalla formulazione della Meccanica Quantistica. I lavori di molti scienziati e in particolare di Planck, Schrödinger, Dirac, Heinsenberg e Einstein, mostrarono come una corretta descrizione dei fenomeni microscopici necessitasse dell’unificazione dei concetti di onda e particella, in alcuni casi la quantizzazione dei possibili valori degli osservabili, l’individuazione di osservabili che non possono essere misurati simultaneamente per i quali valgono relazioni di indeterminazione, e l’abbandono di una meccanica deterministica in favore di leggi intrinsecamente probabilistiche. La Meccanica Quantistica ha inoltre portato all’introduzione di un momento intrinseco delle particelle, detto *spin* e alla distinzione delle particelle in due famiglie con differenti proprietà statistiche, i *fermioni* e i *bosoni*, rispettivamente di spin semintero e intero.
La nostra attuale visione del mondo poggia sull’identificazione di quattro forze fondamentali e di diverse profonde relazioni fra tre di queste. La forza fondamentale che storicamente è stata identificata per prima è la *forza di gravità*, descritta prima dalla Gravitazione Newtoniana e poi dalla Relatività Generale.
La seconda forza fondamentale è la *forza elettromagnetica*. La descrizione delle forze elettromagnetiche in termini della teoria classica di campo di Maxwell è stata superata con la formulazione della prima teoria quantistica relativistica di campo, l’Elettrodinamica Quantistica (QED). Lo schema concettuale della Teoria Quantistica dei Campi (QFT), realizza pienamente la dualità onda-particella, associando le particelle a *quanti di energia* di corrispondenti campi d’onda (nel caso della QED i quanti del campo sono i fotoni), in maniera tale da rendere evidente l’indistinguibilità di tutte le particelle di uno stesso tipo. Nella visione tradizionale della QFT, le particelle fondamentali sono i quanti di energia dei campi che entrano nella lagrangiana della teoria fondamentale. La descrizione in termini di campi assegna alle particelle un numero limitato di attributi (*numeri quantici*), tra i quali la massa, lo spin e uno o più tipi di carica. Le interazioni tra particelle sono mediate da scambi di altre particelle: ad esempio le interazioni elettromagnetiche sono associate allo scambio di fotoni.
La terza forza fondamentale è la *forza debole*, responsabile dei processi di decadimento $`\beta `$, fra i quali il più noto è il decadimento di un protone in un neutrone, un elettrone e un antineutrino. I decadimenti $`\beta `$ sono noti sin dalla fine del diciannovesimo secolo, ma l’identificazione di questa forza fondamentale è avvenuta solo nella prima metà del ventesimo secolo. Le interazioni deboli sono sensibilmente più flebili delle interazioni elettromagnetiche, e la loro prima descrizione è dovuta a Fermi. Negli anni ’60, i lavori di Glashow, Weinberg e Salam hanno portato a formulare una teoria in grado di descrivere le forze deboli e elettromagnetiche all’interno di uno schema unificato. La teoria quantistica di campo che descrive le forze deboli e elettromagnetiche è nota come *teoria elettrodebole*.
Infine la quarta forza fondamentale è detta *forza forte*, o forza di colore. La forza forte è responsabile di diversi stati aggregati come i neutroni, i protoni, i pioni e di molte altre particelle subnucleari. I costituenti elementari di questi stati vengono identificati nei *quarks*, che non possono propagarsi liberamente, ma solo in stati aggregati, a causa delle peculiari proprietà della forza forte. La teoria quantistica di campo che descrive questa forza è nota come Cromodinamica Quantistica (QCD). La teoria elettrodebole e la QCD formano insieme il Modello Standard della fisica delle particelle.
Il Modello Standard è quindi una teoria quantistica di campo, e costituisce una straordinaria sintesi delle nostre conoscenze di tre delle quattro forze fondamentali: le forze elettromagnetiche, le forze deboli e le forze forti. La descrizione di queste interazioni avviene in termini di una teoria di gauge di Yang-Mills con gruppo di gauge $`SU\left(3\right)\times SU\left(2\right)\times U\left(1\right)`$. Una teoria di gauge è una generalizzazione della Teoria di Maxwell dell’elettromagnetismo, i cui pontenziali sono in generale matrici che soddisfano equazioni di campo non lineari, anche in assenza di materia.
In particolare, la simmetria di gauge $`SU\left(2\right)\times U\left(1\right)`$ è associata alle interazioni deboli, mediate dai bosoni massivi di spin $`1`$, $`W^\pm `$ e $`Z^0`$, e alle interazioni elettromagnetiche mediate dai fotoni, bosoni di massa nulla di spin $`1`$; il gruppo di gauge $`SU\left(3\right)`$ è invece associato alle interazioni forti, mediate da otto bosoni di massa nulla, detti gluoni. Ci sono poi diverse particelle fondamentali di “materia”, fermioni di spin $`1/2`$, divisi in due gruppi: i *leptoni* e i *quarks*. I leptoni includono elettroni $`e^{}`$, muoni $`\mu ^{}`$, e tauoni $`\tau ^{}`$ con i rispettivi neutrini di massa quasi nulla $`\nu _e`$, $`\nu _\mu `$, $`\nu _\tau `$, e sono organizzati in tre doppietti,
$$Leptoni:(\nu _e,e^{}),(\nu _\mu ,\mu ^{}),(\nu _\tau ,\tau ^{}).$$
(1)
Considerando anche le rispettive antiparticelle, si ha quindi un totale di dodici leptoni.
I quarks hanno carica forte (colore), debole e elettrica. Esistono sei differenti tipi di quarks, detti *sapori*: *up* e *down*, *charmed* e *strange*, e *top* e *bottom*. I quarks sono organizzati in tre doppietti, che sono la ripetizione di massa crescente del primo doppietto, formato dai costituenti della materia ordinaria, i quarks $`u`$ e $`d`$,
$$Quarks:(u,d),(c,s),(t,b).$$
(2)
La carica forte si presenta in tre differenti colori. Considerando anche le antiparticelle, si hanno $`6\times 3\times 2=36`$ quarks, e il Modello Standard ha quindi un totale di 48 particelle di materia fermioniche e 12 bosoni di gauge.
Il Modello Standard è completato dal meccanismo di *rottura spontanea della simmetria di gauge*, detto comunemente meccanismo di Higgs, ma dovuto a R. Brout, F. Englert e P. Higgs, che permette di avere una teoria rinormalizzabile con bosoni di gauge massivi, come richiesto dalle interazioni deboli. La rottura di simmetria ha luogo La lagrangiana elettrodebole descrive con quattro bosoni di gauge di massa nulla, una proprietà essenziale ai fini della sua rinormalizzabilità, mentre il processo di rottura di simmetria genera le masse dei bosoni vettoriali responsabili delle interazioni deboli: $`W^+`$, $`W^{}`$, e $`Z^0`$. La particella che rimane di massa nulla è il fotone. La rottura di simmetria di gauge produce un ulteriore effetto fisicamente: la comparsa di uno scalare massivo, il bosone di Higgs. Ad oggi, la verifica sperimentale dell’esistenza di questo bosone rimane il tassello mancante nella verifica delle previsioni del Modello Standard.
Il comportamento delle forze elettromagnetiche e di quelle deboli e forti è estremamente diverso. I fotoni non portano carica elettromagnetica, e l’interazione fra particelle cariche risentirà quindi solo delle correzioni radiative originate dalla creazione di coppie di particella-antiparticella nel vuoto. Al contrario, le forze deboli e forti sono associate a bosoni di gauge carichi, e l’autointerazione dei bosoni di gauge genera un effetto di anti-schermo. Per questa ragione, mentre le forze elettromagnetiche sono libere per piccoli impulsi e grandi distanze, e quindi possono essere studiate perturbativamente, al contrario le forze forti sono *asintoticamente libere* ad alte energie. Questo comportamento è alla base del fenomeno del confinamento dei quarks, suggerito dalla crescita dell’interazione al diminuire dell’energia, che spiega come mai non esistano apparentemente particelle libere con carica di colore. A causa del meccanismo di Higgs, che da massa ai bosoni di gauge, le forze deboli risultano inoltre effettivamente deboli per energie minori di $`M_W100GeV`$.
Come si è detto, il Modello Standard incorpora tutte le proprietà conosciute delle interazioni forti, deboli e elettromagnetiche e risulta in straordinario accordo con i dati sperimentali. D’altra parte, esso è per molti aspetti una teoria poco soddisfacente: ha un numero molto alto di parametri arbitrari (circa una ventina) che devono essere opportunamente fissati e non incorpora in alcun modo le interazioni gravitazionali. Queste due osservazioni danno ragione degli innumerevoli sforzi profusi nel corso degli ultimi trenta anni per individuare una teoria più fondamentale che incorpori la gravità e che possibilmente non contenga parametri liberi.
È importante anche notare come il Modello Standard non sia realmente una teoria di unificazione delle forze fondamentali, sebbene i settori elettrodebole e forte non siano del tutto separati (esistono molte particelle soggette ad entrambe le forze). Negli anni ci sono stati molti tentativi di formulare una *Teoria di Grande Unificazione* (GUT) in grado di unire questi due settori del Modello Standard. Ad oggi, nonostante i molti indizi incoraggianti, specie, come si vedrà, per le estensioni supersimmetriche del Modello Standard, il programma di unificazione è tutt’altro che completo.
Prima di discutere le forze gravitazionali, è utile accennare ai problemi di rinormalizzabilità delle teorie quantistiche di campo. Sin dalla nascita della Teoria Quantistica dei Campi è stato chiaro che lo sviluppo perturbativo, sin dalle prime correzioni quantistiche all’ordine ad un loop, porta in generale ad avere quantità divergenti per gli ordinari osservabili fisici. La ragione di queste divergenze è da ricercare nel passaggio da un numero finito di gradi di libertà nella Meccanica Quantistica, ad un numero infinito nella Teoria Quantistica dei Campi. Questo porta a una somma continua su un numero infinito di modi interni negli integrali di loop, che genera quantità divergenti. Evitando i dettagli tecnici, l’idea chiave della rinormalizzazione è che i parametri “nudi” che compaiono nella lagrangiana di campo, come le costanti di accoppiamento e le masse, siano divergenti ma comunque non misurabili. Inoltre le divergenze di questi parametri possono essere scelte in modo da cancellare gli infiniti negli osservabili della teoria. Assorbendo questi infiniti nei parametri “nudi” della teoria, è possibile definire nuovi parametri “vestiti”, finiti e misurabili. Una teoria si dice rinormalizzabile se è resa non divergente da un numero finito di queste ridefinizioni.
La forza gravitazionale è estremamente debole rispetto alle altre forze fondamentali ma, a differenza di queste, è solo attrattiva e quindi risulta dominante nella dinamica su grande scala dell’universo. La forza gravitazionale è descritta, in termini geometrici, dalla Relatività Generale di Einstein. In questa teoria, a differenza della Relatività Speciale, la metrica è una struttura dinamica e le forze gravitazionali nascono dalla curvatura dello spazio-tempo. La Relatività Generale è una teoria di campo classica e, ad oggi, non esiste alcuna sua formulazione quantistica consistente.
Come si è visto, una delle idee guida della Teoria Quantistica dei Campi associa una particella ad ogni campo, sia di forza che di materia. Le particelle che mediano la trasmissione delle interazioni si muovono nello spazio tempo fra oggetti che portano le cariche dell’interazione. Nel caso della gravità, l’idea va incontro ad alcune difficoltà concettuali. La forza di gravità è infatti associata alla dinamica dello stesso spazio-tempo, mentre si vorrebbe che le forze gravitazionali fossero mediate da particelle (i *gravitoni*) che si propaghino su un background spazio-temporale definito. Il modo più naturale per ottenere questa descrizione è linearizzare la teoria, separando il tensore metrico in una parte di background, ad esempio dato da una metrica Minkowskiana $`\eta _{\mu \nu }\mathrm{diag}(,+,+,+,\mathrm{})`$, e una fluttuazione dipendente dalla posizione $`h_{\mu \nu }\left(x\right)`$, che deve essere piccola , $`\left|h_{\mu \nu }\left(x\right)\right|1`$,
$$g_{\mu \nu }=\eta _{\mu \nu }+h_{\mu \nu }\left(x\right).$$
(3)
In questo modo $`h_{\mu \nu }\left(x\right)`$ può essere associata ad una particella di spin due, il gravitone che si propaga su un background di riferimento. Questo è il punto di partenza necessario per formulare una teoria quantistica della gravità. Le difficoltà però non si limitano a questo.
La difficoltà maggiore nel definire una teoria quantistica di campo che descriva tutte le forze fondamentali risiede nell’apparente impossibilità nello scrivere una teoria quantistica della gravità che sia rinormalizzabile .
Per dare un’idea del problema possiamo considerare un processo di scambio di gravitoni. In analogia con la teoria elettromagnetica in cui le interazioni sono pesate dalla costante di struttura fine $`\alpha =q^2/\mathrm{}c`$, è possibile definire un accoppiamento adimensionale per le interazioni gravitazionali come $`\alpha _G=G_NE^2/\mathrm{}c^5`$, dove $`G_N`$ è la costante di Newton. In unità naturali $`\mathrm{}=c=1`$ si ha $`G_N=M_P^2`$, dove la massa di Planck $`M_P=1.2\times 10^{19}GeV`$. La figura 1 mostra due particelle che si propagano liberamente e le correzioni al processo dovute allo scambio di uno o due gravitoni. La prima correzione sarà proporzionale a $`\alpha _G`$ e quindi a $`E^2/M_P^2`$, dove $`E`$ è naturalmente l’energia caratteristica del processo. Questa correzione diventa quindi rilevante per alte energie con $`E>M_P`$. A queste energie i diagrammi di scambio di due gravitoni sono di ordine
$$M_P^4^{\mathrm{}}𝑑E^{}E^3,$$
(4)
dove $`E^{}`$ e l’energia degli stati virtuali intermedi, e quindi la teoria presenta delle divergenze. Agli ordini successivi le divergenze, come si capisce facilmente, sono più gravi e questo rende la teoria perturbativa inservibile. Questo problema di *divergenze ad alte energie* rende la teoria non rinormalizzabile. In realtà la teoria di Einstein è priva di divergenze ad un loop per ragioni di simmetria non evidenti, mentre le prime divergenze si manifestano a due loop .
Per spiegare le divergenze che si trovano in gravità quantistica vengono avanzate alcune ipotesi, tra le quali quelle comunemente considerate più ragionevoli sono due. La prima è che la teoria abbia un punto fisso non banale ultravioletto, ovvero che le divergenze abbiano origine dall’espansione perturbativa in potenze dell’accoppiamento, mentre una risoluzione esatta della teoria risulterebbe perfettamente consistente. La seconda è che alla scala di Planck sia presente “nuova fisica” e che pertanto la Relatività Generale sia soltanto una teoria effettiva di bassa energia di un teoria più profonda. Questa seconda ipotesi è storicamente fondata nel passaggio dalla teoria di Fermi delle interazioni deboli alla teoria elettrodebole.
Ci si aspetta che le divergenze debbano scomparire in una teoria fondamentale che sia in grado di distribuire nello spazio tempo l’interazione. Il modo più ovvio di farlo è operare una discretizzazione dello spazio-tempo, ma questo distrugge l’invarianza di Lorentz al di sotto di una certa scala, e rende molto difficile mantenere l’invarianza per trasformazioni locali di coordinate, che è propria della Relatività Generale. Come si vedrà, l’alternativa fornita dalla Teoria delle Stringhe è associare le particelle ad oggetti unidimensionali, stringhe, piuttosto che puntiformi come nella Teoria Quantistica dei Campi.
Nel discutere di possibili generalizzazioni delle teorie che oggi conosciamo, è utile accennare a altri due problemi non risolti della Fisica Teorica. Uno di questi è il *problema di gerarchia*. La nostra conoscenza delle leggi fondamentali non è, infatti, in grado di spiegare i diversi ordini di grandezza che riscontriamo in natura. Un primo problema di gerarchia è la grande differenza fra le scale degli accoppiamenti delle diverse forze. Ad esempio, il rapporto fra la costante di Fermi $`G_F`$ e la costante di Newton, che determinano le scale delle interazioni deboli e gravitazionali a bassa energia, è $`G_F/G_N10^{35}\mathrm{}^2c^2`$. Allo stesso modo l’attuale fisica teorica può solo registrare ma non spiegare le gerarchie presenti nei parametri del Modello Standard, ad esempio fra le masse dei fermioni.
Un secondo problema aperto è il *problema di costante cosmologica*, ovvero la determinazione teorica del suo valore in accordo con i dati sperimentali. La costante cosmologica è legata alla densità dell’energia di vuoto e determina la curvatura media dell’universo. Le stime teoriche prodotte dalla QFT ($`\rho _{th}M_P^4c^5/\mathrm{}^3`$) sono in forte disaccordo con i valori osservati in natura ($`\rho _{exp}H^2c^2/G_N`$), dove $`H`$ è la costante di Hubble. La stima teorica risulta di 120 ordini di grandezze più grande di quella sperimentale, e questo è il più grande disaccordo tra teoria ed esperimenti nella storia della scienza.
### Oltre il Modello Standard e la Relatività Generale
Nella ricerca di completamenti del Modello Standard una delle idee teoriche più affascinati è la *supersimmetria*. La supersimmetria è una simmetria spazio-temporale fra particelle bosoniche e fermioniche apparentemente non realizzata alle energie ordinarie. La supersimmetria è stata inizialmente sviluppata alla fine degli anni ’60 nei tentativi di individuazione di un *Master Group* che combinasse i gruppi di simmetria interni e il gruppo di Lorentz in maniera non banale (Myazawa, 1966). ed è stata riscoperta nel 1971 a partire da due diversi filoni di ricerca: da un lato nell’ambito della teoria delle stringhe, dall’altro nella ricerca di generalizzazioni dell’usuale algebra spazio-temporale. La prima azione supersimmetrica in quattro dimensioni è stata proposta nel 1974 da Wess e Zumino.
Nonostante la supersimmetria non risulti in apparenza verificata, sono molte le ragioni che inducono allo studio di possibili estensioni supersimmetriche della fisica che conosciamo. La prima è sicuramente legata alle difficoltà che si incontrano nel costruire modelli di teorie di campo unificate. Il problema è che, nel costruire un gruppo unificato che combini il gruppo di Lorentz e un gruppo di Lie compatto, si incorre nel Teorema di Coleman-Mandula che stabilisce che un gruppo di questo tipo non può avere rappresentazioni finito-dimensionali. Questo indica che non è possibile costruire un *master group* che combini sia la gravità che lo spettro di particelle.
La supersimmetria è un modo per evitare il Teorema di Coleman-Mandula, che non riguarda appunto una simmetria non banale che mescoli campi fermionici e bosonici e che ponga entrambi in uno stesso multipletto.
In teorie supersimmetriche è presente un operatore $`Q`$ che trasforma stati bosonici $`|B`$ in stati fermionici $`|F`$,
$$Q|B=|F,$$
(5)
e una sua conseguenza diretta è la possibilità di avere multipletti in cui compaiano fermioni e bosoni.
Un altro aspetto di estremo interesse di questa teoria è che rendendo la supersimmetria una simmetria locale, si giunge necessariamente ad una teoria della gravità. La nuova teoria chiamata Supergravità, e scoperta da Ferrara, Freedman e van Nieuwenhuizen nel 1976, anche se non finita, risulta meno divergente della gravità ordinaria. Come si vedrà essa ha un ruolo centrale nelle ricerche in corso attualmente.
Uno degli aspetti di maggiore interesse nello studio della supersimmetria è che questa conduce naturalmente a tentativi di unificazione della fisica delle particelle e della gravità. In teorie supersimmetriche gli accoppiamenti di gauge delle tre forze del Modello Standard realizzano l’unificazione, con una buona approssimazione, a energie intorno a $`2\times 10^{16}GeV`$. L’accoppiamento adimensionale della gravità, come si è visto, a differenza di quanto avviene per le altre interazioni, dipende fortemente dalla scala di energia. La figura 2 mostra come in teorie supersimmetriche ci siano incoraggianti indicazioni di un’unificazione delle quattro forze fondamentali.
Infine la supersimmetria sembra avere un ruolo per il problema della costante cosmologica. Fermioni e bosoni contribuiscono infatti all’energia di vuoto con segni opposti, e una teoria supersimmetrica nello spazio piatto da quindi luogo a una costante cosmologica nulla. Un teoria realistica e semplicemente verificabile dovrebbe prevedere la rottura della supersimmetria a scale di energie prossime a quelle dei nuovi acceleratori. Modelli di questo tipo portano però a stime della costante cosmologica migliori di quelle della QFT standard, ma che restano lontane dai valori osservati.
Un’altra idea di grande interesse, introdotta nei tentativi di realizzazione di una teoria di grande unificazione è quella delle *dimensioni extra*. È possibile che, su scale dell’ordine della lunghezza di Planck ($`\mathrm{}_P=10^{33}cm`$) o anche maggiori, lo spazio-tempo evidenzi oltre alle quattro dimensioni estese ordinarie (le tre dimensioni spaziali e il tempo), diverse altre dimensioni spaziali piccole (*compattificate*). Naturalmente queste dimensioni sfuggirebbero ad una rilevazione con ogni tipo di sonde di lunghezze d’onda grandi rispetto alla scala delle dimensioni compatte.
Quest’idea, a prima vista singolare, risulta invece piuttosto naturale. Il numero di dimensioni spazio-temporali non è in alcun modo vincolato dalla Teoria dei Campi, mentre una teoria più fondamentale potrebbe avere il pregio di fissarlo univocamente per ragioni di consistenza interna (come si vedrà nel caso della Teoria delle Stringhe).
Un primo argomento sulla ragionevolezza delle dimensioni extra è di tipo cosmologico. L’attuale modello cosmologico contempla una fase iniziale di espansione dell’universo. Quelle che oggi sono dimensioni estese sono state, nelle fasi iniziali dell’universo, piccole e fortemente curvate. Non sembrerebbe pertanto irragionevole se le dimensioni fossero più di quattro e se solo alcune di queste avessero subito il processo di espansione, fino alla struttura dello spazio-tempo che conosciamo, dal momento che non c’è alcuna ragione per supporre un processo di espansione isotropo. In natura sono presenti numerose simmetrie spontaneamente rotte, e un principio di rottura di simmetria simile potrebbe quindi interessare anche le simmetrie spazio-temporali. Il gruppo di simmetria ordinario di Lorentz $`SO(3,1)`$, potrebbe essere, in questo caso, il gruppo di simmetria residua di un gruppo più grande $`SO(d,1)`$, spontaneamente rotto, $`d>3`$. La simmetria verrebbe rotta dalla compattificazione delle dimensioni extra.
L’interesse per la possibile esistenza per ulteriori dimensioni è giustificato dal fatto che queste potrebbero essere responsabili di alcuni aspetti della fisica osservata. A titolo di esempio, si poò considerare un modello molto simile al quello proposto nel 1919 da T. Kaluza, che diede orgine alle teorie con dimensioni extra, anche note come *teorie di Kaluza-Klein*. L’idea è che, oltre alle 4 dimensioni ordinarie, esista una quinta dimensione, compattificata nel modo più semplice possibile, e cioè su un cerchio, tramite l’identificazione
$$\varphi \left(x_5\right)\varphi \left(x_5+2\pi R\right),$$
(6)
dove $`R`$ è il raggio di compattificazione della quinta dimensione. Espandendo il campo $`\varphi `$,
$$\varphi \left(x\right)=\underset{n}{}\varphi _ne^{ipx},$$
(7)
si trova che la relazione di periodicità impone la quantizzazione dei momenti $`p_5=n/r`$ in termini di un intero $`n`$. Per raggi di compattificazione dell’ordine della lunghezza di Planck, i modi più alti con $`n0`$ sono particolarmente massivi $`m10^{19}GeV`$. Nel limite di bassa energia questi modi possono essere ignorati, tenendo conto solo del modo $`n=0`$. Questo significa che in questa approssimazione, il campo $`\varphi \left(x\right)`$ perde la sua dipendenza dalla quinta coordinata,
$$_5\varphi \left(x\right)0.$$
(8)
Si può così decomporre la Relatività Generale in 5-dimensioni in campi quadridimensionali, scrivendo il tensore metrico nella forma
$$g_{AB}=\left[\begin{array}{ccc}& & \\ g_{\mu \nu }& & A_\mu \\ & & \\ & & \\ & & \\ A_\nu & & \varphi ^{}\end{array}\right].$$
(9)
Considerando le equazioni di Einstein in cinque dimensioni e scrivendole in termini delle componenti (9), si trovano sia le equazioni di Einstein per la metrica quadridimensionale $`g_{\mu \nu }`$ che le equazioni di Maxwell per il potenziale vettore $`A_\mu `$. L’ultimo elemento della matrice è un campo scalare $`\varphi ^{}`$, noto come dilatone, che si accoppia sia al campo elettromagnetico che alla metrica e che avremo modo di incontrare spesso nel seguito di questa Tesi.
In maniera simile, per spazi con dimensioni più alte, dall’equazione di Dirac è possibile ritrovare differenti generazioni di quarks e leptoni: un singolo spinore nello spazio con molte dimensioni porta a molti campi spinoriali in quattro dimensioni. Il problema è però, in generale, realizzare la struttura chirale delle interazioni deboli.
### La Teoria delle Stringhe come teoria di unificazione
La Teoria delle Stringhe, nata alla fine degli anni ’60 dal tentativo di descrivere le forze nucleari forti, è oggi il candidato più promettente per una teoria unificata di tutte le forze fondamentali . Dal momento che per le interazioni forti appariva impossibile far ricorso alla teoria perturbativa dei campi, si cercavano al tempo esempi concreti di “matrici S”, ovvero espressioni in grado di determinare direttamente le probabilità di diverse interazioni sotto condizioni assegnate. Nel 1968 Veneziano propose un’ampiezza che sembrava appunto descrivere un’interazione fra particelle scalari risultante dallo scambio di infinite particelle di massa e spin crescente . L’ampiezza aveva anche sorprendenti proprietà di simmetria fra le variabili di Mandelstam $`s`$ e $`t`$ legate alla descrizione degli impulsi in gioco. Poco dopo il lavoro di Veneziano, Shapiro e Virasoro proposero una nuova “matrice S” con una simmetria più ampia tra le variabili che descrivono gli impulsi . Nel 1970 le ampiezze furono reinterpretate, principalmente grazie ai lavori di Nambu e Susskind, come ampiezze d’interazione di oggetti unidimensionali, stringhe appunto. Nel 1971 l’inclusione dei gradi di libertà fermionici portò alla scoperta delle stringhe supersimmetriche .
Lo studio di questi modelli, detti allora “modelli duali”, continuò fino allo sviluppo della QCD, che si dimostrò presto la corretta teoria delle interazioni forti. Nel 1974 però Scherk e Schwarz, e indipendentemente Yoneya proposero che la Teoria delle Stringhe poteva descrivere le interazioni gravitazionali e poteva essere per questo un candidato come teoria di unificazione .
Il periodo 1984-85 vide importanti risultati che convinsero un’ampia parte della comunità scientifica del potenziale della Teoria delle Stringhe in relaziona al problema dell’unificazione. In particolare, nel 1984 Green e Schwarz mostrarono che la teoria di superstringhe aperte e chiuse di Tipo I è priva di anomalie, e quindi quantisticamente consistente, grazie ad un meccanismo di cancellazione del tutto nuovo, se il suo gruppo di gauge è SO(32) .
Nella Teoria delle Stringhe le forze sono unificate in maniera molto profonda, dal momento che le particelle sono unificate. Mentre nella Teoria Quantistica dei Campi le particelle fondamentali sono considerate puntiformi, in Teoria delle Stringhe esse sono identificate con i modi vibrazionali di oggetti fondamentali unidimensionali, le stringhe.
Lo studio di una teoria quantistica relativistica di oggetti unidimensionali rivela una ricchezza sorprendente, e molte delle caratteristiche attese da una teoria di unificazione:
* La Teoria di Stringhe include naturalmente la gravità. Ogni teoria consistente di stringhe contiene uno stato vibrazionale di massa nulla e spin 2 che può essere identificato con il gravitone, dal momento che a basse energie la sua dinamica è descritta dalle equazioni di Einstein.
* Almeno in teoria delle perturbazioni, la teoria è una teoria quantistica della gravità priva di divergenze.
* E’ possibile in maniera molto naturale introdurre grandi gruppi di gauge in grado di contenere, almeno in linea di principio, il Modello Standard, come atteso dalle Teorie di Grande Unificazione.
* La quantizzazione della Teoria delle Stringhe porta a fissare univocamente la dimensione dello spazio tempo, per ragioni di consistenza. La dimensionalità dello spazio-tempo si trova essere per la stringa bosonica $`D=26`$, e per la superstringa $`D=10`$.
* La supersimmetria può essere inclusa in maniera molto naturale nella Teorie delle Stringhe, e questo produce notevoli semplificazioni.
* La Teoria non ha parametri liberi. In particolare, la costante di accoppiamento di stringa, $`g_s`$, è determinata dinamicamente dal campo del dilatone, $`g_s=e^\varphi `$.
Le stringhe hanno una scala dimensionale naturale che può essere stimata con l’analisi dimensionale. Dal momento che la teoria delle stringhe è una teoria quantistica che descrive anche la gravità, essa deve coinvolgere le costanti fondamentali $`c`$ (velocità della luce), $`\mathrm{}`$ (costante di Planck), e $`G_N`$ (costante di Newton). Da queste costanti naturali, si può formare una lunghezza, la lunghezza di Plack, a cui si è già fatto cenno,
$$\mathrm{}_P=\sqrt{\alpha ^{}}=\left(\frac{\mathrm{}G_N}{c^3}\right)^{3/2}=1.6\times 10^{33}cm,$$
(10)
in maniera simile si definisce la massa di Planck come
$$m_P=\left(\frac{\mathrm{}c}{G_N}\right)^{1/2}=1.2\times 10^{19}GeV/c^2.$$
(11)
Esperimenti ad energie molto inferiori all’energia di Planck, come quelli possibili attualmente, non possono risolvere distanze dell’ordine della lunghezza di Planck. Alle energie accessibili le stringhe possono quindi essere approssimate efficacemente da particelle puntiformi. Questo giustifica, dal punto di vista della Teoria delle Stringhe, il successo della QFT nel descrivere la fisica che conosciamo.
Nella sua evoluzione temporale, una stringa disegna una superficie bidimensionale nello spazio-tempo, detta *superficie d’universo* (*world-sheet*), che può essere parametrizzata con coordinate ($`\sigma `$, $`\tau `$). La superficie d’universo è l’equivalente per una stringa della traiettoria descritta da una particella puntiforme. La “storia” di una stringa nello spazio-tempo $`D`$-dimensionale è descritta dalla sua coordinata $`X^\mu (\sigma ,\tau )`$, e l’azione generica di stringa è della forma
$$S=T𝑑A,$$
(12)
dove $`T`$ indica la tensione di stringa e $`dA`$ è l’elemento d’area spazzato dalla stringa nel suo cammino. Come si vedrà in dettaglio nel prossimo capitolo, si tratta della naturale generalizzazione dell’azione di particella relativistica.
Nella formulazione perturbativa della Teoria Quantistica dei Campi i contributi delle ampiezze sono associati ai diagrammi di Feynman, che raffigurano tutte le possibile configurazioni delle traiettorie. In particolare, le interazioni corrispondono alle giunzioni delle traiettorie. Allo stesso modo, la teoria perturbativa di stringa coinvolge superfici di universo di diverse topologie. L’esistenza delle interazioni è però una conseguenza della topologia della superficie d’universo piuttosto che di una singolarità locale (vedi figura 3). Questa differenza rispetto alla teoria dei campi, ha due importanti implicazioni. La prima è che in Teoria delle Stringhe la struttura delle interazioni è unicamente determinata dalla teoria libera, mentre la seconda è che, dal momento che le superfici di universo sono lisce, le ampiezze in Teoria delle Stringhe non hanno divergenze ultraviolette (almeno nel caso di stringhe chiuse). La comparsa di divergenze nella Teoria dei Campi Quantistica è invece riconducibile al fatto che le interazioni sono localizzate in punti dello spazio-tempo.
Esistono due tipi possibili di stringhe, aperte e chiuse. Le stringhe aperte hanno due estremità, mentre quelle chiuse non ne hanno alcuna. Si possono avere teorie di sole stringhe chiuse o di stringhe chiuse e aperte. La ragione intuitiva dell’impossibilità di scrivere teorie di sole stringhe aperte e quindi senza gravità (che ha sempre origine da stringhe chiuse), è che le stringhe aperte possono chiudersi dando luogo a stringhe chiuse.
Una seconda divisione è tra stringhe bosoniche e superstringhe. Le stringhe bosoniche descrivono solo particelle di spin intero, e per questo motivo non appaiono realistiche. Al contrario, le superstringhe descrivono anche fermioni, e ci si aspetta che una descrizione realistica unificata delle forze fondamentali possa sorgere dalle teorie di superstringa.
Un aspetto cruciale della teoria delle stringhe è legato allo studio delle anomalie. Un’anomalia è una violazione quantistica di una simmetria posseduta da una teoria a livello classico. Le anomalie sono ben note anche nello studio della Teoria dei Campi. Nel caso la violazione riguardi una simmetria globale, questa non è dannosa ma comporta la comparsa di effetti nella fenomenologia della teoria. Ad esempio, nel Modello Standard le anomalie globali sono importanti nella determinazione della vita media del $`\pi ^0`$ e della massa della particella $`\eta ^{}`$. Al contrario, se l’anomalia riguarda una simmetria di gauge questo porta in generale ad una inconsistenza delle teoria. Infatti le simmetrie di gauge sono necessarie per la cancellazione dei gradi di libertà non fisici, e l’impossibilità di farlo porta ad una perdita di unitarietà della teoria. In realtà, per una teoria bidimensionale come il modello chirale di Schwinger (una teoria di gauge $`U\left(1\right)`$ accoppiata ad un fermione di massa nulla), le anomalie di gauge non sono dannose e i gradi di libertà aggiuntivi possono essere inclusi ottenendo una teoria consistente, ma non si è ancora in grado di generalizzare questa procedura in dimensioni più alte. Per formulare una teoria delle stringhe consistente è quindi necessario che tutte le anomalie locali (di gauge, gravitazionali, miste) si cancellino. In presenza di stringhe aperte la cancellazione delle anomalie di gauge si ottiene, come si vedrà in dettaglio, imponendo un’opportuna cancellazione nel settore di Ramond-Ramond (R-R), che è uno dei settori dello spettro di superstringa. Questa condizione equivale a richiedere che lo spazio-tempo compattificato sia globalmente neutro, rispetto alle cariche corrispondenti, e fissa necessariamente il gruppo di gauge della superstringa di Tipo I in dieci dimensioni. La cancellazione delle anomalie del diagramma di esagono all’ordine ad un loop, l’analogo del diagramma a triangolo in quattro dimensioni, avviene grazie al contributo del diagramma di propagazione al livello ad albero della 2-forma: questo è noto come meccanismo di Green e Schwarz . Infine, la cancellazione dell’anomalia di Weyl porta a determinare univocamente la dimensione $`D`$ dello spazio-tempo.
La nostra comprensione della Teoria delle Stringhe è stata per molti anni limitata agli aspetti perturbativi. Dalla teoria dei campi è però noto che molti importanti effetti dinamici sorgono quando si hanno molti gradi di libertà in accoppiamento forte, quali il confinamento e il meccanismo di Higgs, che sono necessari alla comprensione della fisica della teoria. Fortunatamente, a partire dalla metà degli anni ’90, si è ottenuto un notevole progresso nella comprensione delle teorie supersimmetriche nel limite di accoppiamento forte, che ha avuto come effetto una sorprendente unificazione delle teorie di superstringa note.
### Relazioni di dualità e M-teoria
La famiglia delle teorie di superstringa contiene cinque differenti modelli con supersimmetria spazio-temporale in $`10`$ dimensioni: Tipo I $`SO\left(32\right)`$, Tipo IIA, Tipo IIB, eterotica $`SO\left(32\right)`$ (HO), eterotica $`E_8\times E_8`$ (HE), oltre ad alcuni altri modelli non supersimmetrici. Negli ultimi dieci anni molti sforzi sono stati spesi nello studio delle relazioni che collegano i differenti modelli. Si è così giuti a comprendere che le diverse teorie supersimmetriche sono legate fra loro da alcune trasformazioni, dette di *dualità*, e che esse possono essere interpretate come differenti limiti di un’unica teoria sottostante in $`11`$ dimensioni, detta M-teoria, i cui gradi di libertà fondamentali non sono però stringhe .
Le teorie di Tipo II A e IIB sono teorie di superstringhe chiuse orientate, mentre la Tipo I è una teoria di superstringhe aperte e chiuse non orientate. Nel limite di basse energie, in cui la teoria di stringhe può essere interpretata come una teoria di campo, queste tre teorie descrivono altrettanti modelli di supergravità in $`10`$ dimensioni, rispettivamente i modelli di Tipo IIA, IIB, di Tipo I, quest’ultima accoppiata ad una teoria con supersimmetrica di Yang-Mills con gruppo di gauge $`SO\left(32\right)`$.
La stringa eterotica è un modello di sole stringhe chiuse. Dal momento che, come si vedrà, per una stringa chiusa i modi destri e sinistri sono indipendenti, è possibile considerare modelli i cui modi destri sono i modi di una superstringa in 10 dimensioni, mentre quelli sinistri sono i modi di una stringa bosonica in 26 dimensioni. Naturalmente occorre compattificare le dimensioni bosoniche chirali in eccesso. Questa costruzione, a prima vista singolare, porta alla formulazione di una teoria supersimmetrica consistente e introduce in modo naturale i gradi di libertà interni di una teoria di gauge senza dover considerare stringhe aperte. Le condizioni di consistenza selezionano solo due possibili gruppi di gauge, $`SO\left(32\right)`$ e $`E_8\times E_8`$.
Della M-teoria si conosce la teoria di basse energie, trovata alla fine degli anni ’70 da Cremmer, Julia e Scherk, che risulta essere l’unica teoria di supergravità in 11 dimensioni . In particolare, si vede che uno spazio-tempo 11-dimensionale è il più grande in cui sia possibile formulare una teoria di supergravità. La supergravità in undici dimensioni è una teoria non rinormalizzabile, e quindi non può essere considerata come una teoria fondamentale. Al contrario, la corretta interpretazione sembra essere quella di una teoria effettiva di bassa energia della M-teoria. La M-teoria non è direttamente legata ad una teoria di stringa, dal momento che il suo limite di bassa energia non contiene il potenziale di oggetti unidimensionali.
Fra le cinque teorie di superstringa e la M-teoria esiste una fitta rete di relazioni, nota come “rete di dualità”. In generale una dualità è una relazione di equivalenza fra sistemi fisici apparentemente distinti, che connette gli stati di una teoria con quelli di un’altra (o diversi stati della stessa teoria di partenza) preservando le interazioni e le simmetrie.
Le relazioni di dualità sono ben note in teoria dei campi. Un esempio particolarmente importante di questo tipo di relazioni è la dualità elettro-magnetica. Una sorprendente caratteristica delle equazioni di Maxwell è la simmetria dei termini sinistri delle equazioni sotto gli scambi del campo elettrico e del campo magnetico $`𝐄𝐁`$ e $`𝐁𝐄`$. Questa simmetria porta all’idea suggestiva che possano esistere cariche magnetiche oltre che cariche elettriche, e una conseguenza di questa ipotesi è una relazione di quantizzazione del prodotto tra cariche elettriche e magnetiche trovata da Dirac. La ricerca di evidenze sperimentali di questa congettura, che ha solide basi nelle teorie di grande unificazione e di superstringa, è oggi un attivo campo di ricerca.
Come si è detto, le relazioni di dualità sono di straordinario interesse anche in teoria delle stringhe. Un primo esempio è la S-dualità che collega settori di una teoria con accoppiamento forte a settori con accoppiamento debole
$$S:g_s\frac{1}{g_s}.$$
(13)
L’importanza di questa dualità risiede nel fatto che, come si è già avuto modo di sottolineare, mentre per accoppiamento forte non è possibile lo studio perturbativo della teoria, al contrario questo diventa possibile in accoppiamento debole.
Si trova che la S-dualità identifica il limite di accoppiamento debole della teoria di stringa eterotica $`SO\left(32\right)`$, con il limite di accoppiamento forte della Tipo I $`SO\left(32\right)`$. Inoltre si trova che la teoria Tipo IIB è auto-duale sotto questa trasformazione. La simmetria tra accoppiamento forte e debole per S-dualità è esplicita nelle teorie di basse energie, mentre rimane una congettura per le teorie complete, anche se sono molti gli indizi che spingono a ritenerla vera.
In teoria delle stringhe la S-dualità scambia fra di loro diversi tipi di oggetti, in particolare stringhe e alcuni oggetti estesi conosciuti come *$`D`$-brane* , che risultano fondamentali per avere il giusto conteggio dei gradi di libertà della teoria prima e dopo la dualità. Le $`D`$-brane corrispondono a configurazioni solitoniche la cui tensione è proporzionale all’inverso della costante d’accoppiamento di stringa, e sono mappate dalla S-dualità in stati di stringa. Nel limite perturbativo ($`g_s`$ piccolo) le $`D`$-brane si presentano come oggetti essenzialmente rigidi e privi di dinamica, mentre al contrario nel limite di accoppiamento forte le $`D`$-brane si rivelano oggetti dinamici che possono muoversi e curvarsi. Una $`D`$-brana è caratterizzata, oltre che dalla tensione, anche dalla sua carica di Ramond-Ramond. Naturalmente esisteranno anche anti-brane, con carica opposta e stessa tensione.
Una seconda relazione di dualità è la T-dualità, che identifica teorie compattificate su un cerchio di raggio $`R`$, con teorie compattificate su un raggio $`1/R`$,
$$T:R\frac{\alpha ^{}}{R}.$$
(14)
La T-dualità identifica le due teorie di tipo II, e le due teorie di stringa eterotica. A differenza della S-dualità, la T-dualità è una simmetria perturbativa .
Esiste una ulteriore relazione che unifica la teoria di Tipo I con la IIB, conosciuta come proiezione di orientifold $`\mathrm{\Omega }`$ . L’azione di $`\mathrm{\Omega }`$ scambia modi sinistri e destri di una stringa chiusa, e i punti fissi di questa proiezione corrispondono nello spazio-tempo a oggetti estesi non dinamici detti piani di orientifold o $`O`$-piani. Gli $`O`$-piani risultano essere carichi e avere tensione che, a differenza di quella delle $`D`$-brane, può avere valori anche negativi.
Le ultime relazioni utili a completare il diagramma delle dualità sono quelle che collegano la M-teoria alle teorie IIA e HE. Si può vedere che, partendo dalla teoria di supergravità 11-dimensionale, e compattificando l’undicesima dimensione su un cerchio $`S^1`$, si trova la supergravità di Tipo IIA. Al contrario, compattificandola su un segmento $`S^1/_2`$, si ritrova la teoria di bassa energia della stringa eterotica $`E_8\times E_8`$ .
Complessivamente le relazioni di dualità ricostruiscono i sei lati dell’esagono di dualità in figura 4. Questa mostra la profonda unità delle teorie di superstringa in 10 dimensioni, che possono essere pensate come limiti di una teoria sottostante che viene identificata con la M-teoria. La teoria delle stringhe rivela quindi, sorprendentemente, di non essere realmente una teoria di oggetti unidimensionali.
### Teorie realistiche e problemi legati alla loro verifica
Si è visto che la Teoria delle Stringhe sembra essere buon candidato per una teoria di unificazione. Ci si aspetta quindi che sia possibile far emergere, insieme alla gravitazione quantistica, il Modello Standard dalla teoria delle stringhe nel limite di bassa energia. Si è visto che la teoria delle stringhe possiede strutture in grado di accogliere tutte le particelle e interazioni del Modello Standard, ma allo stato attuale della ricerca non si è ancora riusciti ad identificare un modello che nel limite di bassa energia che includa tutti i dettagli del Modello Standard.
I modelli di superstringa, che contengono sia bosoni che fermioni, sono definiti in $`D=10`$. La costruzione di modelli realistici in quattro dimensioni, in grado di descrivere l’universo che conosciamo, deve necessariamente affrontare la compattificazione delle sei dimensioni extra e la rottura della supersimmetria, che come abbiamo detto non è osservata in natura. Come si vedrà i due problemi sono in parte collegati.
Il modo più semplice di compattificare le ulteriori dimensioni è considerare un toro $`T^n`$, dove $`n`$ indica la sua dimensione. Si tratta di un’estensione molto semplice del modello di Kaluza-Klein, che nel caso di stringhe chiuse porta ad una fisica più ricca, grazie alla possibilità che le stringhe si avvolgano intorno alle dimensioni compattificate. Una generalizzazione di questo tipo di compattificazioni sono le compattificazioni su orbifold , che si ottengono identificando punti della varietà interna, sotto l’azione di un gruppo discreto. Queste identificazioni lasciano, in generale, dei punti fissi che in teoria delle stringhe non producono singolarità. Per questo tipo di compattificazioni è possibile determinare analiticamente sia lo spettro al livello ad albero che le interazioni. Infine, una classe di compattificazioni di grande interesse sono gli spazi di Calabi-Yau che, a differenza delle compattificazioni di orbifold, sono varietà lisce ma che in alcuni limiti si riducono ad orbifold. Un Calabi-Yau è, in generale, una varietà $`n`$-dimensionale complessa, compatta, con metrica di Kähler Ricci-piatta, e con gruppo di olonomia $`SU\left(d\right)`$, con $`d=n/2`$ . La riduzione del gruppo di olonomia da $`SO\left(n\right)`$ a $`SU\left(d\right)`$ è una conseguenza della richiesta che una supersimmetria sia presente in $`D=4`$.
La rottura delle supersimmetria può essere ottenuta anche in compattificazioni toroidali o su orbifold. Nel caso di compattificazioni toroidali è possibile generalizzare alle stringhe il meccanismo di Scherk-Schwarz , che per teorie supersimmetriche di campo con compattificazioni consiste nell’introdurre degli shift dei momenti di Kaluza-Klein dei vari campi proporzionali alle loro cariche, introducendo in generale differenze di massa fra fermioni e bosoni che rompono la supersimmetria. Questo meccanismo in Teorie delle Stringhe si arricchisce della possibilità di introdurre shift non solo nei momenti ma anche nei winding , che sono i numeri quantici associati al numero di avvolgimenti della stringa intorno alla direzione compatta. La supersimmetria è conseguentemente rotta alla scala $`1/R`$ dove $`R`$ è la scala tipica della dimensione compatta. Si tratta di un meccanismo di rottura spontanea di simmetria, regolato da un parametro continuo, e nel limite di decompattificazione la supersimmetria viene ripristinata. Nel caso di stringhe chiuse l’introduzione degli shift nei momenti o nei winding porta essenzialmente allo stesso fenomeno. Questo è dovuto alle proprietà delle teorie di stringa chiusa sotto T-dualità. Al contrario, per la teoria di Tipo I l’effetto dei due tipi di shift sulla fisica del modello è molto diverso, e i due meccanismi risultanti sono detti *Scherk-Schwarz supersymmetry breaking* ed *M-theory breaking* . Mentre nel primo caso la supersimmetria è rotta sia nello spazio tempo che sulle brane, per il modello di M-theory breaking si ha un interessante fenomeno detto *brane supersymmetry*, ovvero le eccitazioni di bassa energia di brane immerse in uno spazio-tempo non supersimmetrico possono essere supersimmetriche. Questo è vero a livello classico, ma la supersimmetria delle brane viene rotta da correzioni radiative.
In modelli con compattificazioni su orbifold è possibile introdurre la rottura della supersimmetria sulle brane in due diversi modi. Il primo modo detto *brane supersymmetry breaking* si realizza in modelli che richiedono configurazioni con la presenza simultanea di brane e antibrane di diversi tipi. In questo caso il settore chiuso è supersimmetrico, anche se in generale è diverso dal settore chiuso di una teoria non compattificata, mentre nel settore aperto la supersimmetria è rotta alla scala di stringa. Questo tipo di configurazioni, che si studieranno in dettaglio nei prossimi capitoli, risultano stabili ovvero privi di tachioni. Il secondo modo consiste nel deformare uno spettro aperto supersimmetrico con un sistema di coppie separate di brane e antibrane dello stesso tipo. Questo tipo di configurazione risultano instabili a causa delle forze attrattive fra brane e antibrane.
L’ultimo metodo conosciuto per rompere la supersimmetria in Teoria delle Stringhe consiste nell’introdurre campi magnetici all’interno delle dimensioni compatte che equivale, nella rappresentazione che si ottiene con una T-dualità, all’introduzione di brane ruotate . La rottura di simmetria avviene perchè gli estremi delle stringhe aperte sono carichi e possono accoppiarsi ai campi magnetici, dando luogo a degli shift nelle masse degli stati di stringa differenti a seconda dei loro spin. Le differenze di massa che si producono in questo modo possono portare alla rottura della simmetria fra fermioni e bosoni (un’opportuna scelta dei campi magnetici può anche preservare la supersimmetria).
Uno dei principali problemi nello studio di teorie di stringa realistiche è che la teoria ha un enorme numero di vuoti approssimativamente stabili, che corrispondono alle diverse scelte per possibili per le dimensioni compattificate. La fisica osservata risente in modo cruciale dalla scelta del vuoto, dal momento che i parametri di bassa energia in quattro dimensioni dipendono da alcuni *moduli* continui e discreti che codificano il tipo di compattificazione. D’altra parte la Teoria delle Stringhe condivide apparentemente con la Relatività di Einstein la mancanza di un principio globale di minimo che permetta di selezionare globalmente le configurazioni energeticamente più favorevoli. In cosmologia solitamente si fa ricorso ad argomenti basati sulle condizioni iniziali, le simmetrie, la semplicità. In Teoria delle Stringhe questo significa che non è possibile selezionare uno dei vuoti stabili fissando arbitrariamente i moduli. I moduli si presentano nella teoria di bassa energia come campi scalari con accoppiamenti esclusivamente derivativi e quindi con potenziali piatti e valori di vuoto indeterminati. In questo modo la Teoria delle Stringhe, la cui formulazione è priva di parametri liberi, viene a generare un numero molto alto di parametri discreti e continui attraverso le sue soluzioni. Una teoria è naturalmente tanto più predittiva quanto più basso è il numero delle sue soluzioni. In Teoria delle Stringhe, la presenza di molti vuoti è oggi il maggiore ostacolo alla possibilità di estrarre i parametri del Modello Standard. Questo problema è noto come *problema dei moduli*.
La ricerca di uno (o più) modelli di stringa in grado di prevedere tutti i parametri del Modello Standard è uno dei principali obiettivi oggi perseguiti al fine di giustificare l’adozione della Teoria delle Stringhe come teoria delle interazioni fondamentali. D’altra parte è di grande interesse anche l’esplorazione della teoria per la gravità quantistica. Grandi progressi sono stati compiuti nello studio delle situazioni in cui gli effetti quantistici della gravità diventano rilevanti. Il più celebre di questi è forse la precisa interpretazione statistica della termodinamica di Bekenstein dei buchi neri, per una vasta classe di questi oggetti legati in modo diretto alla supersimmetria. Lo studio semi-classico della gravità quantistica, dove campi quantistici vengono studiati in un background classico di buco nero, ha portato Hawking all’inizio degli anni ’70 ad ipotizzare l’emissione di radiazione termica da parte dei buchi neri. Precedentemente Bekenstein aveva proposto una formula per l’entropia dei buchi neri associata alla loro area. Questa formula, che va sotto il nome di entropia di buco nero di Bekenstein-Hawking, è semplicemente
$$S=\frac{A}{4G_N},$$
(15)
costituisce il cuore della delle leggi termodinamiche dei buchi neri. Quando fu formulata, questa legge non era sostenuta da nessuna teoria della gravità quantistica che spiegasse in termini statistici la relazione fra l’entropia e le proprietà dei buchi neri. La teoria delle stringhe ha permesso, almeno per una classe di buchi neri, di giustificare questa formula in funzione di gradi di libertà microscopici, eccitazioni di $`D`$-brane . Naturalmente una teoria completa della gravitazione quantistica sarebbe di straordinaria importanza nello studio della cosmologia delle fasi iniziali di formazione dell’universo. In queste fasi infatti la Relatività Generale non è assolutamente in grado di fornire informazioni e la nostra attuale conoscenza della Teoria delle Stringhe appare insufficiente.
È utile infine accennare a due campi di ricerca sperimentale che potrebbero fornire utili indizi sulla Teoria delle Stringhe, la ricerca delle ulteriori dimensioni e della supersimmetria. Si è precedentemente indicata la lunghezza di Planck, $`\mathrm{}_P10^{33}`$ cm, come scala naturale delle ulteriori dimensioni. Questa scelta porta le ulteriori dimensioni a scale lontane dalle nostre possibilità di verifica sperimentale, in quanto gli attuali acceleratori sono infatti in grado di esplorare distanze solo fino a $`10^{16}`$ cm. Anche se questo scenario appare probabile, la Teoria delle Stringhe non esclude la presenza di dimensioni extra “grandi”, dell’ordine, ad esempio, di $`10^{18}`$ cm. In questo caso una verifica sperimentale sarebbe possibile e qualora avvenisse costituirebbe una spettacolare evidenza per la plausibilità della Teoria. Allo stesso modo, se si dovessero trovare evidenze sperimentali della realizzazione della supersimmetria ad alte energie, questo costituirebbe un indizio del fatto che la Teoria delle Stringhe si sta muovendo nella giusta direzione.
### Oltre un loop
Come si è detto, lo studio sistematico delle proprietà di vuoti con supersimmetria rotta in quattro dimensioni è attualmente uno dei principali ostacoli da superare per giungere ad una comprensione soddisfacente del legame tra la Teoria delle Stringhe e il Modello Standard delle interazioni fondamentali. Questa Tesi tratta di alcuni aspetti tecnici legati a questo confronto, e in particolare discute la struttura di alcune correzioni radiative in modelli con supersimmetria rotta.
Al momento esistono diverse tecniche per il calcolo di diagrammi ad un loop nella Teoria delle Stringhe, e il loro studio ha rivelato una serie di proprietà sorprendenti, tra cui nuovi meccanismi per la cancellazione delle anomalie e per la rottura della supersimmetria. Inoltre, alcuni dei risultati di questa analisi (correzioni di soglia, l’analogo nella Teoria delle Stringhe delle funzioni beta del gruppo di rinormalizzazione) sono alla base dei tentativi di confronto con la Fisica delle Particelle Elementari. Non esistono invece tecniche generali oltre un loop, e per molto tempo la stessa definizione dei diagrammi ha presentato difficoltà ritenute insormontabili. Recentemente E. D’Hoker e D.H. Phong hanno proposto una definizione operativa per il contributo a due loop in una classe di superstringhe chiuse e orientate in D=10. Oggetto di questo lavoro di Tesi è l’estensione di questo risultato ad altri tipi di superstringhe in D=10, in vista anche della costruzione di un’espressione generale per le correzioni di soglia a due loops.
Nel primo capitolo vengono discusse le proprietà fondamentali e le lagrangiane delle stringhe relativistiche aperte e chiuse, sia nel caso bosonico che in quello di superstringa, lo spettro delle loro eccitazioni e i metodi di quantizzazione canonica e nel cono di luce.
Il secondo capitolo contiene una breve introduzione alla formulazione funzionale della teoria delle stringhe (integrale di Polyakov). In questa formulazione le ampiezze di stringa sono definite da una somma su tutte le possibili “storie” di stringa che interpolino tra stati iniziali e finali. L’integrale di Polyakov dà quindi luogo ad una espansione perturbativa della teoria in potenze della “costante di accoppiamento di stringa”, ordinata dalla caratteristica di Eulero delle superfici di Riemann coinvolte.
Il terzo capitolo contiene una breve rassegna sulle ampiezze di stringa bosonica, e mostra in dettaglio alcune ampiezze di stringa aperta e chiusa all’ordine ad albero e ad un loop.
Il quarto capitolo contiene i risultati principali dello studio delle ampiezze di vuoto ad un loop, che in Teoria delle Stringhe sono interpretabili come funzioni di partizioni e permettono di estrarre un gran numero di informazioni sulla struttura dei modelli e di definirne le regole costitutive.
Il quinto capitolo è dedicato ai più semplici meccanismi di compattificazione delle dimensioni aggiuntive della teoria, su tori e orbifold. Queste compattificazioni rivelano una simmetria perturbativa molto profonda, nota come T-dualità, che gioca un ruolo centrale anche nella definizione di altri oggetti estesi, le $`D`$-brane e gli $`O`$-piani.
Il sesto capitolo è dedicato all’introduzione dei modelli con supersimmetria spazio-temporale rotta sia in maniera esplicita (Tipo $`0`$) che spontaneamente.
L’ultimo capitolo è dedicato alle ampiezze di superstringa e ai problemi relativi alla loro definizione a genere più alto del primo. Il lavoro originale di questa Tesi consiste nella generalizzazione di risultati ottenuti recentemente da D’Hoker e Phong per le ampiezze di genere due per i modelli di superstringa chiusa ed orientata ad altri casi con supersimmetria rotta (modelli di Tipo $`0`$ e modelli con “brane supersymmetry breaking” in dieci dimensioni), e in risultati preliminari sulle loro correzioni di soglia. Questo studio rappresenta il punto di partenza per lo studio sistematico delle ridefinizioni di vuoto introdotte a due loops dalla rottura della supersimmetria. Il contenuto di questo capitolo è frutto di una collaborazione in corso con il Dr. Carlo Angelantonj dell’Università di Torino, il Prof. Emilian Dudas dell’Ecole Polytechnique di Parigi, e il mio relatore di Tesi.
## Capitolo 1 Stringhe Relativistiche
### 1.1 Particella relativistica
Una particella relativistica puntiforme che viva in uno spazio $`D`$-dimensionale, che consideriamo con metrica $`\eta _{\mu \nu }=\mathrm{diag}(,+,+,\mathrm{})`$, descrive nel suo moto una traiettoria (*una superficie d’universo* 1-dimensionale) parametrizzata dal tempo proprio $`\tau `$. L’azione più semplice per una particella massiva è data dalla lunghezza del cammino percorso nello spazio-tempo. La lunghezza infinitesima è
$$d\mathrm{}=\left(ds^2\right)^{1/2}=\left(dX^\mu dX^\nu \eta _{\mu \nu }\right)^{1/2}=\left(dX^\mu dX_\mu \right)^{1/2},$$
(1.1)
dove si ha $`\left(ds^2\right)>0`$ per una particella massiva e l’azione è quindi
$$S_{pr}=m𝑑\mathrm{}=m𝑑\tau \sqrt{\eta _{\mu \nu }\dot{X}^\mu \dot{X}^\nu },$$
(1.2)
dove il punto indica la derivata rispetto a $`\tau `$. Variando l’azione si trova
$$\delta S_{pr}=m𝑑\tau \left(\frac{\dot{X}^\mu \delta \dot{X}_\mu }{\sqrt{\dot{X}^\mu \dot{X}_\mu }}\right)=m𝑑\tau \left(\frac{\ddot{X}^\nu }{\sqrt{\dot{X}^\mu \dot{X}_\mu }}\right)\delta X_\nu ,$$
(1.3)
e pertanto per una variazione $`\delta X_\nu `$ arbitraria, si ottiene l’equazione del moto
$$\frac{d^2X^\mu }{d\tau ^2}=0.$$
(1.4)
L’azione (1.2), oltre a contenere una radice che rende complicata la quantizzazione, non può descrivere particelle di massa nulla. È possibile scrivere una nuova azione equivalente alla (1.2) nel caso massivo, ma in grado di descrivere anche il caso di particelle di massa nulla,
$$S_{pr}^{}=\frac{1}{2}𝑑\tau \left(e^1\dot{X}^\mu \dot{X}_\mu em^2\right),$$
(1.5)
dove $`e\left(\tau \right)`$ è un moltiplicatore di Lagrange, un termine che non ha dinamica. Da un punto di vista fisico si può immaginare $`e\left(\tau \right)`$ come legato ad una possibile ‘metrica’ $`\gamma _{\tau \tau }`$ definita sulla traiettoria della particella
$$e\left(\tau \right)=\sqrt{\gamma _{\tau \tau }\left(\tau \right)},ds^2=\gamma _{\tau \tau }d\tau d\tau .$$
(1.6)
Si tratta naturalmente di una descrizione ridondante nel caso di una particella puntiforme, che descrive traiettorie unidimensionali, ma nel caso di stringa sarà particolarmente utile. Variando la (1.5) rispetto ad $`e`$ si ottiene
$$\delta S_{pr}^{}=\frac{1}{2}𝑑\tau \left[e^2\dot{X}^\mu \dot{X}_\mu m^2\right]\delta e,$$
(1.7)
e considerando variazioni arbitrarie $`\delta e`$ si trova l’equazione del moto per $`e`$
$$\dot{X}^\mu \dot{X}_\mu +e^2m^2=0,$$
(1.8)
che costituisce un vincolo per la teoria. Risolvendo l’equazione trovata si ottiene
$$e=\frac{1}{m}\sqrt{\dot{X}^\mu \dot{X}_\mu },$$
(1.9)
che sostituita nell’azione (1.5) da
$$S_{pr}^{}=\frac{1}{2}𝑑\tau \left[m\sqrt{\dot{X}^\mu \dot{X}_\mu }+\sqrt{\dot{X}^\mu \dot{X}_\mu }m^1m^2\right]=S_{pr},$$
(1.10)
mostrando l’equivalenza a livello classico della due azioni.
La nuova azione ha due simmetrie interessanti. La prima è l’invarianza sotto trasformazioni di Poincaré dello spazio-tempo ambiente,
$$X^\mu X^\mu =\mathrm{\Lambda }_\nu ^\mu X^\nu +a^\mu ,$$
(1.11)
dove $`\mathrm{\Lambda }_\nu ^\mu `$ è una matrice di $`SO(1,D1)`$ e $`a^\mu `$ un vettore arbitrario $`D`$-dimensionale. Questa è una simmetria globale comune ad entrambe le azioni scritte. La seconda simmetria è propria solo della seconda azione, ed è una simmetria locale o *di gauge*, definita sulla linea di universo e dovuta alla presenza di $`e\left(\tau \right)`$. L’azione è infatti invariante sotto riparametrizzazioni della forma
$$\delta X=\zeta \left(\tau \right)\frac{dX\left(\tau \right)}{d\tau },\delta e=\frac{d}{d\tau }\left[\zeta \left(\tau \right)e\left(\tau \right)\right],$$
(1.12)
per un arbitrario parametro $`\zeta \left(\tau \right)`$.
### 1.2 La Stringa Bosonica
Una stringa è un oggetto esteso unidimensionale che descrive nel suo moto una superficie di universo (*world-sheet*) bidimensionale che può essere parametrizzata con coordinate $`(\tau ,\sigma )`$. La prima può essere vista come il tempo proprio e la seconda come la coordinata spaziale che corre lungo la stringa, scegliendo $`0\sigma \pi `$. L’evoluzione della stringa nello spazio-tempo è descritta dalle funzioni $`X^\mu (\tau ,\sigma )`$, con $`\mu =0,\mathrm{},D1`$, che descrivono l’immersione della superficie di universo nello spazio-tempo.
#### 1.2.1 Azione di Stringa Bosonica
L’azione e tutte le quantità fisiche, come nel caso di particella puntiforme, devono essere indipendenti dalla parametrizzazione della superficie d’universo. La più semplice azione che possiamo scrivere è proporzionale all’area del world-sheet spazzato dalla stringa. Per esprimere l’azione in termini di $`X^\mu (\tau ,\sigma )`$, definiamo la metrica indotta $`h_{ab}`$, i cui indici $`a,b`$ corrono sui valori $`(\tau ,\sigma )`$:
$$h_{ab}=_aX^\mu _bX_\mu .$$
(1.13)
L’azione di stringa che si ottiene è l’azione di Nambu-Goto, ovvero
$$S_{NG}=\frac{1}{2\pi \alpha ^{}}_M𝑑\tau 𝑑\sigma \sqrt{deth_{ab}},$$
(1.14)
dove M indica la superficie di universo, e $`\alpha ^{}`$ è la *pendenza di Regge*. Non è difficile rendersi conto che l’integrale è l’area spazzata dalla stringa nel suo moto, mentre la costante moltiplicativa
$$T=\frac{1}{2\pi \alpha ^{}}$$
(1.15)
è dimensionalmente una forza fratto una lunghezza, e può essere interpretata come la tensione di stringa. In forma più esplicita l’azione si scrive
$$S_{NG}=\frac{1}{2\pi \alpha ^{}}_M𝑑\tau 𝑑\sigma \sqrt{\left(\frac{X^\mu }{\sigma }\frac{X^\mu }{\tau }\right)^2\left(\frac{X^\mu }{\sigma }\right)^2\left(\frac{X_\mu }{\tau }\right)^2}.$$
(1.16)
Come nel caso di particella puntiforme, è possibile scrivere un’azione classicamente equivalente introducendo una metrica indipendente $`\gamma _{ab}(\sigma ,\tau )`$, che scegliamo con segnatura Lorentziana $`(,+)`$, sulla superficie d’universo $`M`$. La nuova azione, derivata in origine da Brink, Di Vecchia, Howe, Deser e Zumino , ma conosciuta come azione di Polyakov , che ne ha studiato in dettaglio la quantizzazione, è
$`S_P`$ $`=`$ $`{\displaystyle \frac{1}{4\pi \alpha ^{}}}{\displaystyle _M}𝑑\sigma 𝑑\tau \left(\gamma \right)^{1/2}\gamma ^{ab}_aX^\mu _bX^\nu \eta _{\mu \nu }`$ (1.17)
$`=`$ $`{\displaystyle \frac{1}{4\pi \alpha ^{}}}{\displaystyle _M}𝑑\tau 𝑑\sigma \left(\gamma \right)^{1/2}\gamma ^{ab}h_{ab},`$
dove $`\gamma =det\gamma _{ab}`$. Per verificare l’equivalenza delle azioni a livello classico si utilizzano, come nel caso di particella puntiforme, le equazioni del moto che si ottengono variando la metrica. La variazione dell’azione è
$$\delta _\gamma S_P=\frac{1}{4\pi \alpha ^{}}_M𝑑\tau 𝑑\sigma \left(\gamma \right)^{1/2}\left\{\frac{1}{2}\delta \gamma \gamma ^{ab}h_{ab}+\delta \gamma ^{ab}h_{ab}\right\},$$
(1.18)
ed utilizzando la variazione del determinante $`\delta \gamma =\gamma \gamma ^{ab}\delta \gamma _{ab}=\gamma \gamma _{ab}\delta \gamma ^{ab}`$, diventa
$$\delta _\gamma S_P=\frac{1}{4\pi \alpha ^{}}_M𝑑\tau 𝑑\sigma \left(\gamma \right)^{1/2}\delta \gamma ^{ab}\left\{h_{ab}\frac{1}{2}\gamma _{ab}\gamma ^{cd}h_{cd}\right\}.$$
(1.19)
da cui si trova l’equazione del moto
$$h_{ab}=\frac{1}{2}\gamma _{ab}\gamma ^{cd}h_{cd}.$$
(1.20)
Dividendo ciscun membro di questa equazione per la radice quadrata di meno il suo determinante si ottiene
$$h_{ab}\left(h\right)^{1/2}=\gamma _{ab}\left(\gamma \right)^{1/2},$$
(1.21)
una relazione di proporzionalità fra $`\gamma _{ab}`$ e la metrica indotta, e sostituendo infine nell’azione di Polyakov si ritrova l’azione di Nambu-Goto.
Le azioni di Polyakov e di Nambu-Goto hanno diverse simmetrie. In particolare, entrambe sono invarianti sotto le trasformazioni dello spazio tempo del gruppo Poincaré
$`X^\mu (\tau ,\sigma )`$ $`=`$ $`\mathrm{\Lambda }_\nu ^\mu X^\nu (\tau ,\sigma )+a^\mu ,`$
$`\gamma _{ab}^{}(\tau ,\sigma )`$ $`=`$ $`\gamma _{ab}(\tau ,\sigma )`$ (1.22)
e sotto i diffeomorfismi, trasformazioni generali delle coordinate sul world-sheet,
$`X^\mu (\tau ^{},\sigma ^{})`$ $`=`$ $`X^\mu (\tau ,\sigma ),`$
$`{\displaystyle \frac{\sigma ^c}{\sigma ^a}}{\displaystyle \frac{\sigma ^d}{\sigma ^b}}\gamma _{cd}^{}(\tau ^{},\sigma ^{})`$ $`=`$ $`\gamma _{ab}(\tau ,\sigma ),`$ (1.23)
con $`(\tau ^{}(\tau ,\sigma ),\sigma ^{}(\tau ,\sigma ))`$ le nuove coordinate. L’azione di Polyakov ha però un’ulteriore invarianza sotto trasformazioni bidimensionali di Weyl
$`X^\mu (\tau ,\sigma )`$ $`=`$ $`X^\mu (\tau ,\sigma ),`$
$`\gamma _{ab}^{}(\tau ,\sigma )`$ $`=`$ $`e^{2\omega (\tau ,\sigma )}\gamma _{ab}(\tau ,\sigma ),`$ (1.24)
per $`\omega (\tau ,\sigma )`$ arbitrario. Si può comprendere questa ulteriore simmetria osservando che l’equazione del moto (1.21) che lega l’azione di Polyakov a quella di Nambu-Goto determina $`\gamma _{ab}`$ solo a meno di un riscalamento. Tutte le metriche collegate da trasformazioni di Weyl corrispondono quindi alla medesima metrica indotta, e quindi alla stessa descrizione nello spazio-tempo in termini di $`X^\mu (\tau ,\sigma )`$.
L’azione di Polyakov può essere generalizzata aggiungendo termini polinomiali nelle derivate che abbiano tutte le simmetrie dell’azione scritta. L’unico termine invariante sotto trasformazioni di Poincaré, diffeomorfismi e trasformazioni di Weyl è l’azione di Einstein-Hilbert in due dimensioni
$$\chi =\frac{1}{4\pi }_M𝑑\tau 𝑑\sigma \left(\gamma \right)^{1/2}R+\frac{1}{2\pi }_M𝑑sk,$$
(1.25)
dove $`R`$ è lo scalare di Ricci costruito da $`\gamma _{ab}`$ e $`k`$ è la traccia del tensore di curvatura geodesica sul bordo della superficie d’universo. Sotto una trasformazione di Weyl $`\left(\gamma \right)^{1/2}e^{2\omega }\left(\gamma \right)^{1/2}`$ e $`Re^{2\omega }\left(R2^2\omega \right)`$ l’azione scritta è invariante, mentre un termine di costante cosmologica
$$\mathrm{\Theta }=\frac{1}{4\pi \alpha ^{}}_M𝑑\tau 𝑑\sigma \left(\gamma \right)^{1/2},$$
(1.26)
non è al contrario invariante sotto trasformazioni di Weyl.
L’azione di Polyakov generalizzata è quindi
$`S_P^{}`$ $`=`$ $`S_P\lambda \chi `$ (1.27)
$`=`$ $`{\displaystyle _M}𝑑\sigma 𝑑\tau \left(\gamma \right)^{1/2}\left\{{\displaystyle \frac{1}{4\pi \alpha ^{}}}\gamma ^{ab}_aX^\mu _bX^\nu \eta _{\mu \nu }+{\displaystyle \frac{\lambda }{4\pi }}R\right\}{\displaystyle \frac{\lambda }{2\pi }}{\displaystyle _M}𝑑sk.`$
Il termine di Einstein-Hilbert in due dimensioni e il corrispondente temine contenente $`k`$ non portano però dinamica alla metrica, dal momento che risultano essere derivate totali, e il loro contributo ad $`S`$ dipende per questo solo dalla topologia del world-sheet.
#### 1.2.2 Tensore energia-impulso ed equazioni del moto
La variazione dell’azione (1.17) rispetto alla metrica definisce il tensore energia-impulso,
$`T^{ab}(\tau ,\sigma )`$ $`=`$ $`4\pi \left(\gamma \right)^{1/2}{\displaystyle \frac{\delta }{\delta \gamma _{ab}}}S_P`$ (1.28)
$`=`$ $`{\displaystyle \frac{1}{\alpha ^{}}}\left(^aX^\mu ^bX_\mu {\displaystyle \frac{1}{2}}\gamma ^{ab}^cX^\mu _cX_\mu \right),`$
che per l’equazione del moto (1.21) è identicamente nullo, ovvero $`T^{ab}=0`$. Dall’invarianza sotto trasformazioni di Weyl si ottiene anche $`T_a^a=\gamma _{ab}T^{ab}=0`$.
Per ottenere le equazioni del moto dei campi occorre variare l’azione rispetto ad $`X^\mu `$
$`\delta S_P`$ $`=`$ $`{\displaystyle \frac{1}{2\pi \alpha ^{}}}{\displaystyle 𝑑\tau 𝑑\sigma _a\left\{\sqrt{\gamma }\gamma ^{ab}_bX_\mu \right\}\delta X^\mu }`$ (1.29)
$``$ $`{\displaystyle \frac{1}{2\pi \alpha ^{}}}{\displaystyle 𝑑\tau \sqrt{\gamma }_\sigma X_\mu \delta X^\mu }|_{\sigma =0}^{\sigma =\pi },`$
che porta all’equazione
$$_a\left(\sqrt{\gamma }\gamma ^{ab}_bX_\mu \right)=\sqrt{\gamma }^2X^\mu =0,$$
(1.30)
dove $``$ denota la derivata covariante. L’equazione trovata deve essere accompagnata da condizioni che cancellino i termini di bordo e che siano consistenti con l’invarianza sotto trasformazioni di Poincarè. Si possono imporre condizioni al bordo di Neumann
$$StringheAperte:\{\begin{array}{c}X^\mu (\tau ,0)=0\\ X^\mu (\tau ,\pi )=0\end{array},$$
che portano a definire stringhe aperte con estremi liberi di muoversi nello spazio-tempo, oppure condizioni di periodicità
$$StringheChiuse:\{\begin{array}{c}X^\mu (\tau ,0)=X^\mu (\tau ,\pi )\\ X^\mu (\tau ,0)=X^\mu (\tau ,\pi )\\ \gamma _{ab}(\tau ,0)=\gamma _{ab}(\tau ,\pi )\end{array}.$$
che portano a definire stringhe chiuse.
Per risolvere le equazioni del moto (1.30) in maniera diretta si può semplificare il problema utilizzando le simmetrie di gauge dell’azione. La metrica $`\gamma _{ab}`$ è una matrice simmetrica $`2\times 2`$ ed è specificata da tre funzioni indipendenti. Fissando le due riparametrizzazioni delle coordinate sulla superficie di universo e l’invarianza di Weyl, si sceglie
$$\gamma _{ab}=\eta _{ab}e^\varphi =\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right)e^\varphi .$$
(1.31)
Questa scelta della gauge, in cui la metrica bidimensionale è piatta a meno di un fattore conforme, è detta *gauge conforme*. Le equazioni del moto divengono quindi
$$\left(\frac{^2}{\sigma ^2}\frac{^2}{\tau ^2}\right)X^\mu (\tau ,\sigma )=0,$$
(1.32)
che si riconosce essere l’equazione delle onde in due dimensioni, la cui soluzione più generale è
$$X^\mu (\tau ,\sigma )=X_L^\mu \left(\tau +\sigma \right)+X_R^\mu \left(\tau \sigma \right).$$
(1.33)
Imponendo sulla soluzione generale dell’equazione del moto le condizioni al bordo (1.2.2) si ottiene, nel caso di stringa aperta, l’espansione nei modi di vibrazione
$$X^\mu =x^\mu +2\alpha ^{}p^\mu \tau +i\sqrt{2\alpha ^{}}\underset{n0}{}\frac{\alpha _n^\mu }{n}e^{in\tau }\mathrm{cos}n\sigma ,$$
(1.34)
mentre imponendo le condizioni di periodicità (1.2.2) si trova per una stringa chiusa
$$X^\mu =x^\mu +2\alpha ^{}p^\mu \tau +i\frac{\sqrt{2\alpha ^{}}}{2}\underset{n0}{}\left(\frac{\alpha _n^\mu }{n}e^{2in\left(\tau \sigma \right)}+\frac{\stackrel{~}{\alpha }_n^\mu }{n}e^{2in\left(\tau +\sigma \right)}\right),$$
(1.35)
dove, per avere una soluzione reale, si impone $`\alpha _n^\mu =\left(\alpha _n^\mu \right)^{}`$ e $`\stackrel{~}{\alpha }_n^\mu =\left(\stackrel{~}{\alpha }_n^\mu \right)^{}`$. Si osserva che $`x^\mu `$ e $`p^\mu `$ sono rispettivamente la posizione e l’impulso del centro di massa della stringa. Si può identificare $`p^\mu `$ come il modo zero dell’espansione:
$`stringaaperta:\alpha _0^\mu `$ $`=`$ $`\left(2\alpha ^{}\right)^{1/2}p^\mu ,`$
$`stringachiusa:\alpha _0^\mu `$ $`=`$ $`\left({\displaystyle \frac{\alpha ^{}}{2}}\right)^{1/2}p^\mu .`$ (1.36)
Dal punto di vista fisico l’espansione nei modi della stringa chiusa è quella di coppie di onde indipendenti che viaggiano lungo la stringa in direzioni opposte mentre, nel caso aperto si hanno onde stazionarie, dal momento che le condizioni al bordo impongono ai modi destri e sinistri di riflettersi gli uni negli altri. I vincoli sul tensore energia impulso sono
$`T_{\tau \sigma }`$ $`=`$ $`T_{\sigma \tau }={\displaystyle \frac{1}{\alpha ^{}}}\dot{X}^\mu X_\mu ^{}=0,`$
$`T_{\sigma \sigma }`$ $`=`$ $`T_{\tau \tau }={\displaystyle \frac{1}{2\alpha ^{}}}\left(\dot{X}^\mu \dot{X}_\mu +X^\mu X_\mu ^{}\right)=0,`$ (1.37)
che possono anche essere scritti come
$$\left(\dot{X}\pm X^{}\right)^2=0$$
(1.38)
Introducendo le nuove coordinate $`\sigma ^\pm =\tau \pm \sigma `$ si ha $`X^\mu (\tau ,\sigma )=X_L^\mu \left(\sigma ^+\right)+X_R^\mu \left(\sigma ^{}\right)`$. La metrica diventa $`ds^2=d\tau ^2+d\sigma ^2d\sigma ^+d\sigma ^{}`$ e quindi $`\eta _+=\eta _+=1/2`$, $`\eta ^+=\eta ^+=2`$ e $`\eta _{++}=\eta _{}=\eta ^{++}=\eta ^{}=0`$. Le derivate vengono scomposte come $`_\tau =_++_{}`$ e $`_\sigma =_+_{}`$. I vincoli sul tensore energia impulso nelle nuove coordinate sono
$`T_{++}`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left(T_{\tau \tau }+T_{\tau \sigma }\right)={\displaystyle \frac{1}{\alpha ^{}}}_+X^\mu _+X_\mu =0,`$
$`T_{}`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left(T_{\tau \tau }T_{\tau \sigma }\right)={\displaystyle \frac{1}{\alpha ^{}}}_{}X^\mu _{}X_\mu =0.`$ (1.39)
Nelle nuove coordinate è immediato rendersi conto che la scelta della gauge conforme (1.31) non fissa completamente la simmetria locale. Infatti, per due trasformazioni indipendenti delle coordinate sulla superficie d’universo del tipo
$$\sigma ^+\sigma ^+=f\left(\sigma ^+\right),\sigma ^{}\sigma ^{}=g\left(\sigma ^{}\right),$$
(1.40)
si ha una trasformazione della metrica
$$\gamma _+^{}=\left(\frac{f\left(\sigma ^+\right)}{\sigma ^+}\frac{g\left(\sigma ^{}\right)}{\sigma ^{}}\right)^1\gamma _+$$
(1.41)
che può essere facilmente rissorbita con una trasformazione di Weyl della forma
$$\gamma _+^{}=e^{2\omega _L\left(\sigma ^+\right)+2\omega _R\left(\sigma ^{}\right)}\gamma _+$$
(1.42)
per $`e^{2\omega _L\left(\sigma ^+\right)}=_+f\left(\sigma ^+\right)`$ e $`e^{2\omega _R\left(\sigma ^{}\right)}=_{}g\left(\sigma ^{}\right)`$. La teoria di stringhe, nella gauge conforme, definisce una teoria di campo conforme bidimensionale .
#### 1.2.3 Dinamica Hamiltoniana
La densità lagrangiana, per la scelta fatta della metrica, è
$$=\frac{1}{4\pi \alpha ^{}}\left(_\sigma X^\mu _\sigma X_\mu _\tau X^\mu _\tau X_\mu \right),$$
(1.43)
da cui si può derivare il momento coniugato di $`X^\mu `$
$$\mathrm{\Pi }^\mu =\frac{\delta }{\delta \left(_\tau X^\mu \right)}=\frac{1}{2\pi \alpha ^{}}\dot{X}^\mu .$$
(1.44)
Si hanno classicamente le parentesi di Poisson a tempi uguali:
$`\{X^\mu \left(\sigma \right),\mathrm{\Pi }^\nu \left(\sigma ^{}\right)\}_{PB}`$ $`=`$ $`\eta ^{\mu \nu }\delta \left(\sigma \sigma ^{}\right),`$
$`\{X^\mu \left(\sigma \right),X^\nu \left(\sigma ^{}\right)\}_{PB}`$ $`=`$ $`0,`$
$`\{\mathrm{\Pi }^\mu \left(\sigma \right),\mathrm{\Pi }^\nu \left(\sigma ^{}\right)\}_{PB}`$ $`=`$ $`0.`$ (1.45)
Da cui si derivano facilmente le relazioni sugli oscillatori, sull’impulso e sulla coordinata del centro di massa
$`\{\alpha _m^\mu ,\alpha _n^\nu \}_{PB}`$ $`=`$ $`\{\stackrel{~}{\alpha }_m^\mu ,\stackrel{~}{\alpha }_n^\nu \}_{PB}=im\delta _{m+n}\eta ^{\mu \nu },`$
$`\{x^\mu ,p^\nu \}_{PB}`$ $`=`$ $`\eta ^{\mu \nu },`$
$`\{\alpha _m^\mu ,\stackrel{~}{\alpha }_n^\nu \}_{PB}`$ $`=`$ $`0.`$ (1.46)
La densità hamiltoniana è
$$=\dot{X}^\mu \mathrm{\Pi }_\mu =\frac{1}{4\pi \alpha ^{}}\left(_\sigma X^\mu _\sigma X_\mu +_\tau X^\mu _\tau X_\mu \right),$$
(1.47)
da cui si ricava l’Hamiltoniana integrando lungo la stringa
$`H`$ $`=`$ $`{\displaystyle _0^\pi }𝑑\sigma \left(\sigma \right)={\displaystyle \frac{1}{2}}{\displaystyle \underset{\mathrm{}}{\overset{\mathrm{}}{}}}\alpha _n\alpha _nstringheaperte,`$
$`H`$ $`=`$ $`{\displaystyle _0^{2\pi }}𝑑\sigma \left(\sigma \right)={\displaystyle \frac{1}{2}}{\displaystyle \underset{\mathrm{}}{\overset{\mathrm{}}{}}}\left(\alpha _n\alpha _n+\stackrel{~}{\alpha }_n\stackrel{~}{\alpha }_n\right)stringhechiuse.`$ (1.48)
Dalle (1.2.2) possiamo definire gli operatori di Virasoro come i modi di Fourier del tensore energia-impulso. Per la stringa chiusa essi sono
$`L_m`$ $`=`$ $`{\displaystyle \frac{1}{\pi \alpha ^{}}}{\displaystyle _0^{2\pi }}𝑑\sigma T_{}e^{im\left(\tau \sigma \right)}={\displaystyle \frac{1}{2}}{\displaystyle \underset{\mathrm{}}{\overset{\mathrm{}}{}}}\alpha _{mn}\alpha _n,`$
$`\overline{L}_m`$ $`=`$ $`{\displaystyle \frac{1}{\pi \alpha ^{}}}{\displaystyle _0^{2\pi }}𝑑\sigma T_{++}e^{im\left(\tau +\sigma \right)}={\displaystyle \frac{1}{2}}{\displaystyle \underset{\mathrm{}}{\overset{\mathrm{}}{}}}\stackrel{~}{\alpha }_{mn}\stackrel{~}{\alpha }_n,`$ (1.49)
e soddisfano le condizioni di realità
$$L_m^{}=L_me\overline{L}_m^{}=\overline{L}_m$$
(1.50)
Nel caso di stringa aperta non c’è differenza fra oscillatori destri e sinistri, e
$`L_m`$ $`=`$ $`{\displaystyle \frac{1}{\pi \alpha ^{}}}{\displaystyle _0^\pi }𝑑\sigma \left\{T_{}e^{im\left(\tau \sigma \right)}+T_{++}e^{im\left(\tau +\sigma \right)}\right\}`$ (1.51)
$`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \underset{\mathrm{}}{\overset{\mathrm{}}{}}}\alpha _{mn}\alpha _n.`$
L’Hamiltoniana si può riscrivere in termini di operatori di Virasoro:
$`H`$ $`=`$ $`L_0stringheaperte,`$
$`H`$ $`=`$ $`L_0+\overline{L}_0stringhechiuse.`$ (1.52)
I vincoli (1.2.2), in questo formalismo, equivalgono a richiedere per tutti i modi $`L_m=0`$, $`\overline{L}_m=0`$, per ogni $`m`$. Usando le parentesi di Poisson per gli oscillatori (1.2.3) si trova l’algebra di Virasoro:
$`\{L_m,L_n\}_{PB}`$ $`=`$ $`i\left(mn\right)L_{m+n},`$
$`\{\overline{L}_m,\overline{L}_n\}_{PB}`$ $`=`$ $`i\left(mn\right)\overline{L}_{m+n},`$
$`\{L_m,\overline{L}_n\}_{PB}`$ $`=`$ $`0.`$ (1.53)
### 1.3 Quantizzazione della Stringa Bosonica
Ci sono diversi approcci alla quantizzazione delle stringhe. Come in teoria dei campi è possibile quantizzare la teoria classica sia in maniera canonica che nel formalismo dell’integrale sui cammini. La quantizzazione canonica consiste nel sostituire le variabili classiche con operatori quantistici e nel sostituire le parentesi di Poisson con relazioni di commutazione fra operatori
$$\{,\}_{PB}i[,].$$
(1.54)
Come si è visto si sono introdotti dei vincoli nello studio della stringa classica. Nella teoria quantistica, l’introduzuone di questi vincoli è possibile con due approcci differenti. Il primo, conosciuto come *quantizzazione canonica covariante*, consiste nel quantizzare le varibili classiche senza considerare i vincoli per poi imporli sullo spazio di Hilbert degli stati: in questo modo si preserva l’invarianza di Lorentz esplicita della teoria. Il secondo approccio, la *quantizzazione nel cono di luce*, consiste nel risolvere esplicitamente i vincoli al livello della teoria classica, nella *gauge del cono di luce*, per poi quantizzare. In questo caso l’invarianza di Lorentz esplicita viene persa.
Nel formalismo funzionale dell’integrale sui cammini il metodo di Faddeev-Popov viene associato a tecniche BRST e si trova uno spazio degli stati manifestamente Lorentz invariante che però contiene anche dei campi aggiuntivi non fisici chiamati campi di *ghost*. Il formalismo dell’integrale funzionale, che in teoria delle stringhe è conosciuto come integrale di Polyakov, sarà studiato in dettaglio nel prossimo capitolo.
#### 1.3.1 Quantizzazione canonica covariante
La quantizzazione (1.54) conduce alle relazioni di commutazione
$`[X^\mu \left(\sigma \right),\mathrm{\Pi }^\nu \left(\sigma ^{}\right)]`$ $`=`$ $`i\eta ^{\mu \nu }\delta \left(\sigma \sigma ^{}\right),`$
$`[\alpha _m^\mu ,\alpha _n^\nu ]`$ $`=`$ $`[\stackrel{~}{\alpha }_m^\mu ,\stackrel{~}{\alpha }_n^\nu ]=m\delta _{m+n}\eta ^{\mu \nu },`$
$`[x^\nu ,p^\mu ]`$ $`=`$ $`i\eta ^{\mu \nu },[\alpha _m^\mu ,\stackrel{~}{\alpha }_n^\nu ]=0,`$ (1.55)
mentre le condizioni di realtà divengono condizioni di hermiticità degli operatori. Definendo nuovi operatori $`\sqrt{m}\alpha _{\pm m}^\mu `$ si trovano le relazioni di commutazione di $`D`$ coppie di operatori di creazione e di distruzione di un oscillatore quantistico. Gli operatori di Virasoro $`L_m`$ sono stati costruiti dagli operatori di creazione e distruzione, come in Teoria dei Campi, e la quantizzazione richiede che siano *normalmente ordinati*
$$L_m=\frac{1}{2}\underset{\mathrm{}}{\overset{\mathrm{}}{}}:\alpha _{mn}\alpha _n:+a\delta _{m,0},$$
(1.56)
cioé che tutti gli operatori di distruzione siano sulla destra.
In Teoria dei Campi, la prescrizione dell’ordinamento normale è necessaria per ottenere operatori quantistici correttamente definiti, ovvero che abbiano autovalori finiti sugli stati fisici. In generale un operatore in una teoria di campo quantistica è un prodotto a punti uguali di campi fondamentali. Non è difficile accorgersi che un prodotto di campi a punti uguali è formalmente divergente dal momento che si hanno delle somme infinite di prodotti di operatori di creazione di distruzione. La prescrizione di spostare gli operatori di creazione a sinistra e quelli di distruzione a destra elimina le divergenze.
Nel nostro caso, l’ordinamento normale non comporta problemi, date le relazioni di commutazione (1.3.1), eccetto che per $`L_0`$, dal momento che $`\alpha _n^\mu `$ e $`\alpha _m^\mu `$ non commutano. Si trova
$$L_0=\frac{1}{2}\alpha _0^2+\underset{n=1}{\overset{\mathrm{}}{}}\alpha _n\alpha _n+D\underset{n=1}{\overset{\mathrm{}}{}}n,$$
(1.57)
il secondo termine è una costante divergente che, come si vedrà in dettaglio nel prossimo paragrafo, può essere regolata lasciando un termine finito che corrisponde all’energia di punto zero degli oscillatori, che indichiamo con $`a`$. Calcolando l’algebra di Virasoro per la teoria quantistica, facendo attenzione all’ordinamento normale, si trova che l’algebra classica è modificata dalla presenza di un termine centrale:
$`[L_m,L_n]`$ $`=`$ $`\left(mn\right)L_{m+n}+{\displaystyle \frac{D}{12}}m\left(m^21\right)\delta _{m+n},`$
$`[\overline{L}_m,\overline{L}_n]`$ $`=`$ $`\left(mn\right)\overline{L}_{m+n}+{\displaystyle \frac{D}{12}}m\left(m^21\right)\delta _{m+n},`$
$`[L_m,\overline{L}_n]`$ $`=`$ $`0.`$ (1.58)
Lo spazio di Hilbert degli stati può essere costruito a partire dallo stato di vuoto $`|0;k`$, che definiamo come lo stato annichilato da tutti gli operatori di distruzione, e dove $`k`$ indica l’impulso del centro di massa della stringa. Uno stato generico $`|\varphi `$ sarà costruito agendo con gli operatori di creazione sul vuoto. Non è difficile rendersi conto che lo spazio degli stati contiene anche stati non fisici, a norma negativa, dal momento che per $`\mu =0`$ si hanno relazioni di commutazione (1.3.1) con segno opposto alle altre. Ad esempio
$$\left|\alpha _1^0|0;k\right|^2=k;0\left|\alpha _1^0\alpha _1^0\right|0;k=1.$$
(1.59)
Si devono pertanto imporre dei vincoli sullo spazio degli stati:
$`\left(L_0a\right)|\varphi `$ $`=`$ $`0,L_m|\varphi \mathrm{𝑝𝑒𝑟}m>0`$
$`\left(\overline{L}_0a\right)|\varphi `$ $`=`$ $`0,\overline{L}_m|\varphi \mathrm{𝑝𝑒𝑟}m>0.`$ (1.60)
Si noti che i vincoli sono imposti in maniera “debole” (in modo tale che il valore di aspettazione sugli stati sia zero), ovvero solo per $`m>0`$, dal momento che, a causa dell’estensione centrale dell’algebra di Virasoro, si avrebbe altrimenti un’inconsistenza:
$$0=\varphi \left|[L_m,L_m]\right|\varphi =2m\varphi \left|L_0\right|\varphi +\frac{D}{12}m\left(m^21\right)\varphi |\varphi 0.$$
(1.61)
La massa degli stati viene studiata definendo un operatore di massa, $`M^2=p^\mu p_\mu `$. Ricordano le definizioni (1.2.2), dalla (1.57) si ha, nel caso di stringa chiusa,
$`L_0`$ $`=`$ $`{\displaystyle \frac{\alpha ^{}}{4}}p^\mu p_\mu +{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}\alpha _n\alpha _n=`$
$`=`$ $`{\displaystyle \frac{\alpha ^{}}{4}}M^2+{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}\alpha _n\alpha _n,`$
$`\overline{L}_0`$ $`=`$ $`{\displaystyle \frac{\alpha ^{}}{4}}p^\mu p_\mu +{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}\stackrel{~}{\alpha }_n\stackrel{~}{\alpha }_n=`$ (1.62)
$`=`$ $`{\displaystyle \frac{\alpha ^{}}{4}}M^2+{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}\stackrel{~}{\alpha }_n\stackrel{~}{\alpha }_n,`$
da cui utilizzando i vincoli (1.3.1) si trovano le formule di massa
$`M^2`$ $`=`$ $`{\displaystyle \frac{4}{\alpha ^{}}}\left({\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}\alpha _n\alpha _na\right),`$
$`M^2`$ $`=`$ $`{\displaystyle \frac{4}{\alpha ^{}}}\left({\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}\stackrel{~}{\alpha }_n\stackrel{~}{\alpha }_na\right).`$ (1.63)
Possiamo riscrivere in forma simmetrica la formula di massa come semisomma e semidifferenza delle due formule trovate. Si ottiene
$`M^2={\displaystyle \frac{2}{\alpha ^{}}}\left({\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}\alpha _n\alpha _n+{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}\stackrel{~}{\alpha }_n\stackrel{~}{\alpha }_n2a\right),`$
$`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}\alpha _n\alpha _n={\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}\stackrel{~}{\alpha }_n\stackrel{~}{\alpha }_n,`$ (1.64)
dove la seconda condizione è un vincolo sullo spettro di massa (level matching condition). Per la stringa aperta si trova più semplicemente
$$M^2=\frac{1}{\alpha ^{}}\left(\underset{n=1}{\overset{\mathrm{}}{}}\alpha _n\alpha _na\right).$$
(1.65)
Definiamo due operatori numero che contino il numero pesato degli oscillatori sinistri e destri: $`N=\alpha _n\alpha _n=nN_n`$ e $`\overline{N}=\stackrel{~}{\alpha }_n\stackrel{~}{\alpha }_nn\overline{N}_n`$, nei termini dei quali le formule di massa acquistano una forma particolarmente compatta:
$`M^2={\displaystyle \frac{2}{\alpha ^{}}}\left(N+\overline{N}2a\right),N=\overline{N},stringhechiuse`$ (1.66)
$`M^2={\displaystyle \frac{1}{\alpha ^{}}}(Na).stringheaperte`$ (1.67)
Gli stati fisici della teoria sono gli stati costruiti con gli operatori di creazione che rispettino le condizioni (1.3.1). Oltre a questi stati si trovano degli stati
$$|\mathrm{spur}=L_n|,$$
(1.68)
detti *spuri*, ortogonali a tutti gli stati fisici. Esistono anche stati che sono sia fisici che spuri, ma che si possono eliminare dallo spazio di Hilbert fisico dal momento che si può vedere che corrispondono a stati nulli. Lo studio dettagliato dello spettro culmina in un celebre teorema (*no ghost theorem*) che stabilisce che per $`D=26`$ lo spettro fisico ottenuto dalle (1.3.1) contiene solo stati a norma positiva.
È utile, a questo punto, proseguire lo studio dello spettro della stringa bosonica introducendo la quantizzazione nel cono di luce.
#### 1.3.2 Quantizzazione nel cono di luce
Le coordinate del cono di luce sono definite come
$$X^\pm =\frac{X^0\pm X^1}{\sqrt{2}},$$
(1.69)
mentre le coordinate $`X^i`$ per $`i0,1`$ rimangono invariate. Nelle nuove coordinate la metrica si vede facilmente essere
$`a^\mu b_\mu =a^+b^{}a^{}b^++a^ib^i,`$
$`a_{}=a^+,a_+=a^{},a_i=a^i.`$ (1.70)
La simmetria locale residua nella gauge conforme permette di fissare
$$X^+(\sigma ,\tau )=x^++2\alpha ^{}p^+\tau ,$$
(1.71)
che è detta gauge del cono di luce. I vincoli (1.2.2) possono essere risolti esprimendo le coordinate $`X^{}`$ in funzione di quelle trasverse, scrivendo
$`_+X^\mu _+X_\mu =2_+X^+_+X_{}+\left(_+X^i\right)^2=0,`$
$`_{}X^\mu _{}X_\mu =2_{}X^+_{}X_{}+\left(_{}X^i\right)^2=0.`$ (1.72)
Dal momento che
$$_\pm X^+=\alpha ^{}p^+,$$
(1.73)
sostituendo si trova
$`2\alpha ^{}p^+_+X_{}=\left(_+X^i\right)^2,`$
$`2\alpha ^{}p^+_{}X_{}=\left(_+X^i\right)^2.`$ (1.74)
Sostituendo l’espansione nei modi delle coordinate $`X^\mu `$ si trova per gli oscillatori (stringa chiusa)
$`\alpha _m^{}={\displaystyle \frac{1}{\sqrt{2\alpha ^{}}p^+}}{\displaystyle \underset{n}{}}\alpha _{mn}^i\alpha _n^i,`$ (1.75)
e un’espressione analoga per $`\stackrel{~}{\alpha }^{}`$, mentre nel caso di stringa aperta cambiano solo le normalizzazioni.
Dall’espressione trovata si giunge facilmente alla formula di massa. Vediamolo in dettaglio per il caso di stringa chiusa, dal momento che il caso di stringa aperta cambia solo per le definizioni degli oscillatori. Ricordando le espressioni degli zero modi (1.2.2), dall’espressione (1.75) si ottiene
$$\alpha ^{}p^+p^{}=\underset{n}{}\alpha _n^i\alpha _n^i=\frac{\alpha ^{}}{2}\left(p^i\right)^2+\underset{n\left\{0\right\}}{}\alpha _n^i\alpha _n^i,$$
(1.76)
insieme all’analoga espressione in termini di $`\stackrel{~}{\alpha }_n^i`$. Dal momento che $`M^2=p^\mu p_\mu `$, si trovano l’espressione classica della formula di massa e la level matching condition in funzione dei soli coefficienti degli oscillatori trasversi
$`M^2={\displaystyle \frac{1}{\alpha ^{}}}\left({\displaystyle \underset{n\left\{0\right\}}{}}\alpha _n^i\alpha _n^i+{\displaystyle \underset{n\left\{0\right\}}{}}\stackrel{~}{\alpha }_n^i\stackrel{~}{\alpha }_n^i\right),`$
$`{\displaystyle \underset{n\left\{0\right\}}{}}\alpha _n^i\alpha _n^i={\displaystyle \underset{n\left\{0\right\}}{}}\stackrel{~}{\alpha }_n^i\stackrel{~}{\alpha }_n^i.`$ (1.77)
Per quantizzare occorre imporre le relazioni di commutazione canoniche (1.3.1), e l’ordinamento normale che porta, come si è visto nel paragrafo precedente, ad un termine costante infinito nella formula di massa
$$\frac{D2}{2}\underset{n}{\overset{\mathrm{}}{}}n,$$
(1.78)
dove il fattore $`\left(D2\right)`$ viene dalla somma sulle direzioni trasverse. L’energia di punto zero può essere regolata cancellando con un contro-termine la parte divergente. Inseriamo nella sommatoria un regolatore (*cutoff*) esponenziale in modo da valutarne il termine finito come
$`{\displaystyle \frac{D2}{2}}{\displaystyle \underset{n}{\overset{\mathrm{}}{}}}ne^{ϵn}`$ $`=`$ $`{\displaystyle \frac{D2}{2}}{\displaystyle \frac{d}{dϵ}}{\displaystyle \underset{n}{\overset{\mathrm{}}{}}}e^{ϵn}={\displaystyle \frac{D2}{2}}{\displaystyle \frac{d}{dϵ}}\left({\displaystyle \frac{1}{1e^ϵ}}\right)`$ (1.79)
$`=`$ $`{\displaystyle \frac{D2}{2}}\left({\displaystyle \frac{1}{ϵ^2}}{\displaystyle \frac{1}{12}}+O\left(ϵ\right)\right).`$
Il primo termine può essere cancellato con un controtermine del tipo $`d^2\sigma \left(\gamma \right)^{1/2}`$, lasciando nel limite $`ϵ0`$, la costante
$$a=\frac{D2}{24}.$$
(1.80)
Introducendo gli operatori numero trasversi $`N^{}`$, $`\overline{N}^{}`$ in cui le somme coninvolgono solo operatori relativi alle dimensioni trasverse, si ha
$`M^2={\displaystyle \frac{2}{\alpha ^{}}}\left(N^{}+\overline{N}^{}2a\right),N^{}=\overline{N}^{},stringhechiuse`$ (1.81)
$`M^2={\displaystyle \frac{1}{\alpha ^{}}}(N^{}a).stringheaperte`$ (1.82)
A questo punto possiamo studiare lo spettro di massa. Iniziamo dal caso di stringa aperta, un cui stato generico può essere costruito agendo sul vuoto $`|0;k`$ con gli operatori di creazione,
$$|N;k=\left[\underset{i=2}{\overset{D1}{}}\underset{n=1}{\overset{\mathrm{}}{}}\frac{\left(\alpha _n^i\right)^{N_{in}}}{\left(n^{N_{in}}N_{in}!\right)^{1/2}}\right]|0;k,$$
(1.83)
dove $`k`$ è il momento del centro di massa e $`N_{in}`$ sono i numeri di occupazione di ciascun modo. Ogni scelta dei numeri d’occupazione rappresenta, dal punto di vista dello spazio-tempo, una diversa particella o stato di spin. Lo stato più basso in massa è
$$|0;k,M^2=\frac{2D}{24\alpha ^{}},$$
(1.84)
la cui massa quadrata è negativa per $`D>2`$: si tratta quindi di un tachione. Poiché in Teoria dei Campi l’energia potenziale di un campo scalare libero è $`\frac{1}{2}m^2\varphi `$, un valore di massa quadrata negativo indica che il vuoto è uno stato instabile, come ad esempio il vuoto simmetrico nel caso di una teoria con rottura spontanea di simmetria. La presenza di un tachione in teoria delle stringhe è quindi un’indicazione dell’instabilità del vuoto. Si tratta di un problema di definizione della stringa bosonica non del tutto risolto allo stato attuale.
I primi stati eccitati si ottengono eccitando uno solo dei modi $`n=1`$
$$\alpha _1^i|0;k,M^2=\frac{26D}{24\alpha ^{}}.$$
(1.85)
Per uno stato massivo ci si può sempre porre in un sistema di riferimento in cui la particella sia a riposto $`p^\mu =(m,0,\mathrm{},0)`$. Gli stati interni formano una rappresentazione del gruppo delle rotazioni spaziali $`SO\left(D1\right)`$. Per una particella di massa nulla non esiste un sistema di riferimento in cui sia a riposo, ma si può scegliere un sistema di riferimento in cui $`p^\mu =(E,E,\mathrm{},0)`$, e gli stati corrispondenti formano rappresentazioni del gruppo $`SO\left(D2\right)`$. L’invarianza di Lorentz richiede quindi che si abbiano, in $`D`$ dimensioni, $`D1`$ stati di spin per una particella vettoriale massiva e $`D2`$ per una particella vettoriale a massa nulla. Dal momento che il primo stato eccitato ha solo $`D2`$ stati di spin, deve essere un vettore a massa nulla $`A_\mu `$. Si trova pertanto che l’invarianza di Lorentz fissa la dimensione dello spazio-tempo della teoria di stringa bosonica a $`D=26`$.
Nel caso della stringa chiusa, a partire dallo stato più basso in massa $`|0,0;k`$, si può costruire, con gli operatori di creazione dei modi destri e sinistri, uno stato generico
$$|N,\overline{N};k=\left[\underset{i=2}{\overset{D1}{}}\underset{n=1}{\overset{\mathrm{}}{}}\frac{\left(\alpha _n^i\right)^{N_{in}}\left(\stackrel{~}{\alpha }_n^i\right)^{\overline{N}_{in}}}{\left(n^{N_{in}}N_{in}!n^{\overline{N}_{in}}\overline{N}_{in}!\right)^{1/2}}\right]|0,0;k,$$
(1.86)
dove i numeri di occupazione devono essere tali da rispettare il vincolo $`N=\overline{N}`$. Lo stato più basso in massa è ancora un tachione
$$|0,0;k,M^2=\frac{2D}{6\alpha ^{}}.$$
(1.87)
I primi stati eccitati sono
$$\alpha _1^i\stackrel{~}{\alpha }_1^j|0,0;k,M^2=\frac{26D}{6\alpha ^{}},$$
(1.88)
come nel caso aperto questi stati non completano una rappresentazione di $`SO\left(D1\right)`$ e devono essere a massa nulla, fissando ancora $`D=26`$, e quindi $`a=1`$. Lo stato risultante a massa nulla è una rappresentazione tensoriale di $`SO\left(D2\right)`$, può essere decomposto in un tensore simmetrico a traccia nulla, in un tensore antisimmetrico, e uno scalare. Infatti, in generale, un tensore può essere scritto come
$$T^{ij}=\frac{1}{2}\left(T^{ij}+T^{ji}\frac{2}{D2}\delta ^{ij}T^{kk}\right)+\frac{1}{2}\left(T^{ij}T^{ji}\right)+\frac{1}{D2}\delta ^{ij}T^{kk},$$
(1.89)
in cui i tre termini non si mischiano sotto rotazione. I tre stati possono essere interpretati come un gravitone $`G_{\mu \nu }`$, un tensore antisimmetrico $`B_{\mu \nu }`$ e un dilatone $`\varphi `$.
#### 1.3.3 Azione di Stringa in un background curvo
Per giustificare l’identificazione degli stati a massa nulla, occorre fare una piccola digressione discutendo una generalizzazione dell’azione di Polyakov che descrive il moto di una stringa chiusa in un background coerente dei propri stati a massa nulla,
$$S_\sigma =\frac{1}{4\pi \alpha ^{}}_Md^2\sigma g^{1/2}\left[\left(g^{ab}G_{\mu \nu }\left(X\right)+iϵ^{ab}B_{\mu \nu }\left(X\right)\right)_aX^\mu _bX^\nu +\alpha ^{}R\mathrm{\Phi }\left(X\right)\right].$$
(1.90)
L’azione (1.90) è essenzialmente una approssimazione di bassa energia della descrizione di una stringa che si propaghi in un background dovuto alla presenza di un campo di stringa, di cui si manifestano solo gli stati a massa nulla. Questo tipo di azione è ben nota anche per alcuni modelli di Teoria dei Campi ed è conosciuta come *modello sigma non lineare*.
La presenza della metrica $`G_{\mu \nu }\left(X\right)`$ in sostituzione della metrica piatta $`\eta _{\mu \nu }`$ può essere interpretata come l’introduzione di un background curvo che correttamente risulta essere uno “stato coerente” di gravitoni. Infatti come sarà più chiaro nel formalismo funzionale dell’integrale di Polyakov, la presenza di $`G_{\mu \nu }\left(X\right)`$ nell’azione equivale ad inserire un operatore di vertice del gravitone. A questo punto dovrebbe risultare abbastanza naturale, nel linguaggio di stringa, l’inclusione nell’azione (1.90) anche degli altri campi a massa nulla nel background, $`B_{\mu \nu }\left(X\right)`$ e $`\mathrm{\Phi }\left(X\right)`$.
Nella nuova teoria gli accoppiamenti sono quindi tutti determinati dagli stati a massa nulla della teoria stessa. In particolare, come si avrà modo di vedere in dettaglio nel prossimo capitolo, la costante di accoppiamento di stringa è essenzialmente data dall’esponenziale del termine di Eulero, che nella (1.90) è moltiplicato per il campo del dilatone $`\mathrm{\Phi }\left(X\right)`$. Si trova così un risultato di grande interesse, il fatto che la costante d’accoppiamento di stringa è determinata dalla parte costante del campo del dilatone (valore di aspettazione),
$$g_s=e^{<\mathrm{\Phi }>}.$$
(1.91)
Perchè l’azione (1.90) definisca una teoria quantistica di stringa consistente occorre imporre la cancellazione di eventuali anomalie per l’invarianza di Weyl che equivale a richiedere che il tensore bidimensionale energia-impulso sia a traccia nulla. Questo porta ad imporre l’annullamento di tre funzioni $`\beta `$, legate ai tre campi a massa nulla di background, che risultano essere
$`\beta _{\mu \nu }^G`$ $`=`$ $`\alpha ^{}\left(R_{\mu \nu }+2_\mu _\nu \mathrm{\Phi }{\displaystyle \frac{1}{4}}H_{\mu \kappa \sigma }H_\nu ^{\kappa \sigma }\right)+O\left(\alpha ^2\right),`$
$`\beta _{\mu \nu }^B`$ $`=`$ $`\alpha ^{}\left({\displaystyle \frac{1}{2}}^\kappa H_{\kappa \mu \nu }+^\kappa \mathrm{\Phi }H_{\kappa \mu \nu }\right)+O\left(\alpha ^2\right),`$
$`\beta ^\mathrm{\Phi }`$ $`=`$ $`\alpha ^{}\left({\displaystyle \frac{D26}{6\alpha ^{}}}{\displaystyle \frac{1}{2}}^2\mathrm{\Phi }+_\kappa \mathrm{\Phi }^\kappa \mathrm{\Phi }{\displaystyle \frac{1}{24}}H_{\kappa \mu \nu }H^{\kappa \mu \nu }\right)+O\left(\alpha ^2\right),`$ (1.92)
con $`H_{\mu \nu \kappa }_\mu B_{\nu \kappa }+_\nu B_{\kappa \mu }+_\kappa B_{\mu \nu }`$. Queste funzioni sono essenzialmente le funzioni beta del gruppo di rinormalizzazione degli accoppiamenti. Le equazioni
$$\beta _{\mu \nu }^G=\beta _{\mu \nu }^B=\beta ^\mathrm{\Phi }=0.$$
(1.93)
hanno la struttura di equazioni del moto per i campi di background, e possono essere infatti derivate da un’azione di bassa energia nello spazio-tempo,
$`\mathrm{S}`$ $`=`$ $`{\displaystyle \frac{1}{2\kappa _0^2}}{\displaystyle }d^DX(G)^{1/2}e^{2\mathrm{\Phi }}[R+4_\mu \mathrm{\Phi }^\mu \mathrm{\Phi }{\displaystyle \frac{1}{12}}H_{\mu \nu \lambda }H^{\mu \nu \lambda }`$ (1.94)
$`{\displaystyle \frac{2\left(D26\right)}{3\alpha ^{}}}+O\left(\alpha ^{}\right)].`$
In particolare, al primo ordine perturbativo, l’equazione $`\beta _{\mu \nu }^G=0`$ è l’equazione di Einstein con termini di sorgente dovuti al tensore antisimmetrico e al dilatone, mentre l’equazione $`\beta _{\mu \nu }^B=0`$ è la generalizzazione delle equazioni di Maxwell per una sorgente unidimensionale. Agli ordini successivi compaiono delle correzioni di stringa alle equazioni di bassa energia.
È abbastanza sorprendente che l’equazione di Einstein compaia come condizione dell’invarianza di Weyl della teoria bidimensionale. In questo modo la gravità emerge naturalmente dalla richiesta di consistenza della teoria, giustificando l’identificazione di $`G_{\mu \nu }`$ con il gravitone.
### 1.4 La Superstringa
La teoria di stringa bosonica, discussa fin qui, ha il pregio di mettere in luce, in un contesto relativamente semplice, alcuni dei punti di forza di una reinterpretazione delle particelle fondamentali in termini di oscillazioni di oggetti unidimensionali, ma d’altra parte non descrive fermioni e presenta nello spettro il tachione come indice dell’instabilità del vuoto. Si è quindi portati a cercare generalizzazioni dell’azione bosonica a partire dall’inclusione di gradi di libertà fermionici.
Un’idea particolarmente feconda è quella di introdurre una supersimmetria sul world-sheet che leghi le coordinate spazio-temporali $`X^\mu (\tau ,\sigma )`$, che si sono viste essere campi bosonici sul world-sheet, ad un partner fermionico $`\psi _\alpha ^\mu (\tau ,\sigma )`$. L’indice $`\mu `$ indica che, dal punto di vista dello spazio-tempo, la coordinata fermionica trasforma come un vettore, le cui componenti sono spinori sul world-sheet. La teoria ottenuta in questo modo è detta teoria di superstringa, e l’azione corrispondente è
$`S`$ $`=`$ $`{\displaystyle \frac{1}{4\pi \alpha ^{}}}{\displaystyle }d\sigma d\tau \sqrt{\gamma }[\gamma ^{ab}_aX^\mu _bX^\nu i\overline{\psi }^\mu \mathrm{\Gamma }^a_a\psi ^\nu `$ (1.95)
$`i\overline{\chi }_a\mathrm{\Gamma }^b\mathrm{\Gamma }^a\psi ^\mu (_bX^\nu {\displaystyle \frac{i}{4}}\overline{\chi }_b\psi ^\nu )]\eta _{\mu \nu },`$
dove $`\chi _a`$ è un gravitino di Majorana e, come $`\sqrt{\gamma }\gamma ^{ab}`$, è un moltiplicatore di Lagrange senza dinamica. Infatti mentre, come si è gia detto, il termine dinamico della metrica bidimensionale di Einstein-Hilbert è un termine topologico, il termine cinetico del gravitino dovrebbe essere l’azione di Rarita-Schiwinger che, essendo proporzionale ad un tensore a tre indici $`\gamma _{abc}`$, in due dimensioni è identicamente nulla. Per accoppiare i campi spinoriali alla gravità bidimensionale si è introdotto il *formalismo del vielbein*, ovvero una base nello spazio tangente della varietà definita dalla metrica $`\gamma _{ab}`$,
$`\gamma _{ab}=\eta _{mn}e_a^me_b^n,`$
$`\eta ^{mn}=\gamma ^{ab}e_a^me_b^n.`$ (1.96)
Le matrici $`\mathrm{\Gamma }^a`$ sono definite come $`\mathrm{\Gamma }^ae_m^a\mathrm{\Gamma }^m`$, dove le $`\mathrm{\Gamma }^m`$ sono matrici di Dirac ordinarie, mentre la derivata covariante per i campi spinoriali, $`_a`$, è definita come $`_a_a\frac{i}{4}\omega _a^{mn}\sigma _{mn}`$.
Oltre a possedere le simmetrie note l’azione scritta è invariante sotto trasformazioni locali di supersimmetria
$`\delta \gamma _{ab}`$ $`=`$ $`2i\mathrm{\Gamma }_a\chi _b,`$
$`\delta \chi _a`$ $`=`$ $`2_aϵ,`$
$`\delta \psi ^\mu `$ $`=`$ $`\mathrm{\Gamma }^a\left(_aX^\mu {\displaystyle \frac{i}{2}}\chi _a\psi ^\mu \right)ϵ,`$
$`\delta X^\mu `$ $`=`$ $`iϵ\psi ^\mu ,`$ (1.97)
dove il parametro $`ϵ`$ è uno spinore di Majorana. L’azione può essere semplificata per un’opportuna scelta della gauge. Scegliendo l’equivalente della gauge conforme del caso bosonico, detta gauge superconforme,
$$\gamma _{ab}=\eta _{ab}e^\varphi ,\chi _a=\mathrm{\Gamma }_a\zeta ,$$
(1.98)
ed utilizzando l’identità delle matrici gamma bidimensionali $`\mathrm{\Gamma }_a\mathrm{\Gamma }^b\mathrm{\Gamma }^a=0`$, l’azione si riduce a
$$S=\frac{1}{4\pi \alpha ^{}}𝑑\sigma 𝑑\tau \left(\eta ^{ab}_aX^\mu _bX^\nu i\overline{\psi }^\mu \mathrm{\Gamma }^a_a\psi ^\nu \right)\eta _{\mu \nu },$$
(1.99)
l’azione di D campi scalari e D campi fermionici liberi. Procedendo come nel caso della stringa bosonica si trovano due correnti conservate, il tensore energia-impulso, ottenuto dalla variazione della metrica sulla superficie d’universo
$`T_{ab}`$ $`=`$ $`{\displaystyle \frac{1}{\alpha ^{}}}\left(_aX^\mu _bX_\mu +{\displaystyle \frac{i}{4}}\overline{\psi }^\mu \left(\mathrm{\Gamma }_a_b+\mathrm{\Gamma }_b_a\right)\psi _\mu \right)`$ (1.100)
$`+{\displaystyle \frac{1}{2\alpha ^{}}}\eta _{ab}\left(^cX^\mu _cX_\mu +{\displaystyle \frac{1}{2}}\overline{\psi }^\mu \mathrm{\Gamma }\psi _\mu \right)=0,`$
e la supercorrente, ottenuta variando il gravitino,
$$J^a=\frac{1}{2\alpha ^{}}\mathrm{\Gamma }^b\mathrm{\Gamma }^a\psi ^\mu _bX_\mu =0.$$
(1.101)
Come nel caso bosonico, le due equazioni sono vincoli della teoria, che in questo caso prendendo il nome di vincoli di super-Virasoro.
Variando l’azione rispetto ai campi $`X^\mu `$ e $`\psi ^\mu `$ e imponendo l’annullamento dei termini di bordo
$$\delta _{bordo}\left[X_\mu ^{}\delta X^\mu +i\left(\psi _+\delta \psi _+\psi _{}\delta \psi _{}\right)\right]_0^\pi =0,$$
(1.102)
si trovano le equazioni di Klein-Gordon e di Dirac:
$`\left({\displaystyle \frac{^2}{\sigma ^2}}{\displaystyle \frac{^2}{\tau ^2}}\right)X^\mu (\tau ,\sigma )`$ $`=`$ $`0,`$
$`i\mathrm{\Gamma }^a_a\psi ^\mu `$ $`=`$ $`0.`$ (1.103)
Le matrici $`\mathrm{\Gamma }^a`$ in due dimensioni possono essere scelte puramente immaginarie
$$\mathrm{\Gamma }^0=\sigma _2=\left(\begin{array}{cc}0& i\\ i& 0\end{array}\right),\mathrm{\Gamma }^1=i\sigma _1=\left(\begin{array}{cc}0& i\\ i& 0\end{array}\right),$$
(1.104)
e l’operatore di chiralità è
$$\mathrm{\Gamma }^3=\mathrm{\Gamma }^0\mathrm{\Gamma }^1=\sigma _3=\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right).$$
(1.105)
L’operatore di Dirac in questa base è reale,
$$i\mathrm{\Gamma }^a_a\psi ^\mu =\left(\begin{array}{cc}0& \left(_\tau _\sigma \right)\\ \left(_\tau +_\sigma \right)& 0\end{array}\right)\left(\genfrac{}{}{0pt}{}{\psi _+^\mu }{\psi _{}^\mu }\right)$$
(1.106)
e quindi le componenti di $`\psi _a^\mu `$ sul world-sheet possono essere scelte reali. Dalle equazioni del moto si vede che i campi $`\psi _\pm ^\mu `$ sono funzioni delle coordinate $`\sigma ^\pm =\tau \pm \sigma `$. Infine, la matrice di chiralità mostra come $`\psi _\pm ^\mu `$ siano due spinori di Majorana-Weyl. Come nel caso bosonico le condizioni di annullamento dei termini di bordo, compatibili con l’invarianza di Lorentz, definiscono le stringhe aperte e chiuse. Per le stringhe aperte insieme alle condizioni (1.2.2) occorre richiedere che
$`\psi _+^\mu \left(0\right)`$ $`=`$ $`\psi _{}^\mu \left(0\right),`$
$`\psi _+^\mu \left(\pi \right)`$ $`=`$ $`\pm \psi _{}^\mu \left(\pi \right).`$ (1.107)
Le condizioni che si ottengono in corrispondenza del segno relativo positivo sono dette di Ramond, mentre quelle corrispondenti al segno negativo sono dette di Neveu-Schwarz . Allo stesso modo per la superstringa chiusa le condizioni possibili insieme alle (1.2.2) sono
$`\psi _i^\mu \left(\pi \right)`$ $`=`$ $`+\psi _i^\mu \left(0\right),condizionidiRamond\left(R\right)`$
$`\psi _i^\mu \left(\pi \right)`$ $`=`$ $`\psi _i^\mu \left(0\right).condizionidiNeveuSchwarz\left(NS\right)`$ (1.108)
Per la stringa chiusa si hanno quindi quattro possibili combinazioni di condizioni al bordo, che definiscono quattro distiniti settori: R-R, NS-R, R-NS, NS-NS.
Le equazioni del moto (1.4) possono essere ora risolte imponendo sulle soluzioni le condizioni al bordo trovate. Per il campo bosonico si ottengono le soluzioni già note (1.34, 1.35). Le soluzioni dell’equazione di Dirac, nel caso di stringa chiusa, sono per i modi destri e sinistri:
$`\psi _+`$ $`=`$ $`\sqrt{2\alpha ^{}}{\displaystyle \underset{r}{}}\stackrel{~}{\psi }_r^\mu e^{2ir\left(\tau +\sigma \right)}r\mathrm{oppure}r+{\displaystyle \frac{1}{2}}`$
$`\psi _{}`$ $`=`$ $`\sqrt{2\alpha ^{}}{\displaystyle \underset{r}{}}\psi _r^\mu e^{2ir\left(\tau \sigma \right)}r\mathrm{oppure}r+{\displaystyle \frac{1}{2}},`$ (1.109)
dove l’indice $`r`$ corre sui seminteri per condizioni di Ramond, e sugli interi per condizioni di Neveu-Schwarz. Le soluzioni di stringa aperta che si ottengono sono della stessa forma di quelle di stringa chiusa, ma con le frequenze di oscillazione dimezzate, normalizzazione $`\sqrt{\alpha ^{}}`$ e coefficienti dei modi destri e sinistri identificati.
È utile riscrivere, nelle nuove coordinate $`\sigma ^\pm =\tau \pm \sigma `$, il tensore energia impulso e la supercorrente separando le parti olomorfe, che dipendono da $`\sigma ^+`$, da quelle antiolomorfe, che dipendono da $`\sigma ^{}`$:
$`T_{\pm \pm }`$ $`=`$ $`_\pm X^\mu _\pm X_\mu +{\displaystyle \frac{i}{2}}\psi _\pm _\pm \psi _\pm ,`$
$`J_\pm \left(\xi ^\pm \right)`$ $`=`$ $`\psi _\pm ^\mu _\pm X_\mu .`$ (1.110)
#### 1.4.1 Quantizzazione canonica
La quantizzazione canonica della superstringa si ottiene imponendo le relazioni di commutazione usuali per campi di spin intero e semintero:
$`[\dot{X}^\mu (\sigma ,\tau ),X^\nu (\sigma ^{},\tau )]`$ $`=`$ $`i\delta \left(\sigma \sigma ^{}\right)\eta ^{\mu \nu },`$
$`\{\psi _\pm ^\mu (\sigma ,\tau ),\psi _\pm ^\nu (\sigma ^{},\tau )\}`$ $`=`$ $`\delta \left(\sigma \sigma ^{}\right)\eta ^{\mu \nu },`$ (1.111)
da cui si ottengono per gli oscillatori le relazioni di commutazione
$`[\alpha _m^\mu ,\alpha _n^\nu ]`$ $`=`$ $`m\delta _{m+n}\eta ^{\mu \nu },`$
$`\{\psi _r^\mu ,\psi _s^\nu \}`$ $`=`$ $`\eta ^{\mu \nu }\delta _{r+s},`$ (1.112)
e le medesime relazioni per gli oscillatori destri. Come nel caso bosonico si riconoscono, dopo un riscalamento, le regole di commutazione degli operatori di creazione e distruzione. Uno stato generico $`|\varphi `$ sarà ottenuto dall’azione degli operatori bosonici e fermionici di creazione sul vuoto.
La quantizzazione covariante nel caso della superstringa richiede che si impongano i vincoli
$$T_{++}=T_{}=J_+=J_{}=0,$$
(1.113)
che, come nel caso bosonico, possono essere decomposti in modi normali. Per la stringa chiusa si ha
$`L_m|\varphi `$ $`=`$ $`\overline{L}_m|\varphi =0,m>0,`$
$`\left(L_0a\right)|\varphi `$ $`=`$ $`\left(\overline{L}_0\overline{a}\right)|\varphi =0,`$
$`G_r|\varphi `$ $`=`$ $`\overline{G}_r|\varphi =0,r>0,`$ (1.114)
dove
$`L_m`$ $`=`$ $`{\displaystyle \frac{1}{\pi \alpha ^{}}}{\displaystyle _0^{2\pi }}𝑑\sigma T_{}e^{im\sigma }={\displaystyle \frac{1}{2}}{\displaystyle \underset{n}{}}:\alpha _{mn}^\mu \alpha _{\mu ,n}:+{\displaystyle \frac{1}{4}}{\displaystyle \underset{r}{}}\left(2rm\right):\psi _{mr}^\mu \psi _{\mu ,r}:+a\delta _{m,0},`$
$`G_r`$ $`=`$ $`{\displaystyle \frac{2}{\pi \alpha ^{}}}{\displaystyle _0^{2\pi }}𝑑\sigma J_{}e^{ir\sigma }={\displaystyle \underset{n}{}}:\alpha _n\psi _{rn}:,`$ (1.115)
In maniera analoga si trovano gli operatori di super-Virasoro nel caso di stringa aperta. L’algebra degli operatori è
$`[L_m,L_n]`$ $`=`$ $`\left(mn\right)L_{m+n}+{\displaystyle \frac{c}{12}}\left(m^3m\right)\delta _{m+n},`$
$`\{G_r,G_s\}`$ $`=`$ $`2L_{r+s}+{\displaystyle \frac{c}{12}}\left(4r^21\right)\delta _{r+s},`$
$`[L_m,G_r]`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left(m2r\right)G_{m+r},`$ (1.116)
dove $`c=D+D/2`$ con $`D`$ il contributo dei gradi di libertà bosonici e $`D/2`$ il contributo dei gradi di libertà fermionici. La super-algebra è nota come algebra di Ramond per $`r,s`$ interi e di Neveu-Schwarz per $`r,s`$ seminteri. I campi antiolomorfi portano una seconda copia di questa algebra.
Prima di discutere in dettaglio lo spettro, vediamo brevemente la quantizzazione nel formalismo del cono di luce. Le coordinate di cono di luce sono $`X^\pm =\left(X^0\pm X^1\right)/\sqrt{2}`$ e in maniera analoga definiamo $`\psi _\pm `$. La guage di cono di luce è
$`X^+`$ $`=`$ $`x^++2\alpha ^{}p^+\tau `$
$`\psi _\pm ^+`$ $`=`$ $`0,`$ (1.117)
e risolvendo come nel caso bosonico i vincoli (1.113), si ottiene per gli oscillatori
$`\alpha _n^{}`$ $`=`$ $`{\displaystyle \frac{2}{\sqrt{2\alpha ^{}}p^+}}[{\displaystyle \underset{i=1}{\overset{D2}{}}}{\displaystyle \frac{1}{2}}{\displaystyle \underset{m}{}}:\alpha _{nm}^i\alpha _m^i:`$
$`+{\displaystyle \underset{i=1}{\overset{D2}{}}}{\displaystyle \frac{1}{2}}{\displaystyle \underset{r}{}}(r{\displaystyle \frac{n}{2}}):\psi _{nr}^i\psi _r^i:+a\delta _n],`$
$`\psi _r^{}`$ $`=`$ $`{\displaystyle \frac{2}{\sqrt{2\alpha ^{}}p^+}}{\displaystyle \underset{i=1}{\overset{D2}{}}}{\displaystyle \underset{s}{}}\alpha _{rs}^i\psi _s^i,`$ (1.118)
dove $`r,s`$ nel settore di Ramond e $`r,s+1/2`$ nel settore di Neveu-Schwarz mentre $`a`$ è la parte finita del termine divergente dovuto all’ordinamento normale che varia a seconda del settore. Le energie di punto zero possono essere determinate, sul modello di quanto già fatto, studiando l’andamento della funzione $`\zeta `$ di Riemann
$$\zeta _\alpha (1,ϵ)=\underset{n=1}{\overset{\mathrm{}}{}}\left(n+\alpha \right)e^{\left(n+\alpha \right)ϵ}\stackrel{ϵ0}{}\frac{1}{12}\left[6\alpha \left(\alpha 1\right)+1\right]+O\left(\frac{1}{ϵ}\right),$$
(1.119)
da cui si ottiene che ogni grado di libertà bosonico contribuisce all’energia di vuoto con $`a=\frac{1}{24}`$, mentre ogni grado di libertà fermionico con $`a=+\frac{1}{24}`$ per condizioni periodiche e con $`a=\frac{1}{48}`$ per condizioni antiperiodiche. Si ha quindi $`a^R=0`$ e $`a^{NS}=\frac{D2}{16}`$, e anche per la superstringa la dimensione dello spazio tempo è fissata dall’invarianza di Lorentz, in questo caso a $`D=10`$, per cui $`a^R=0`$ e $`a^{NS}=\frac{1}{2}`$.
Definendo gli operatori numero bosonici e fermionici
$`N_B^{}`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{D2}{}}}{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}\alpha _n^i\alpha _n^i,`$
$`N_F^{}`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{D2}{}}}{\displaystyle \underset{r}{\overset{\mathrm{}}{}}}r\psi _r^i\psi _r^i,`$ (1.120)
dagli zero modi si trova per la formula di massa nel caso chiuso
$$M^2=\frac{2}{\alpha ^{}}\left[N_B^{}+\overline{N}_B^{}+N_F^{}+\overline{N}_F^{}+a+\overline{a}\right],N_B^{}+N_F^{}+a=\overline{N}_B^{}+\overline{N}_F^{}+\overline{a},$$
(1.121)
mentre per la superstringa aperta
$$M^2=\frac{2}{\alpha ^{}}\left[N_B^{}+N_F^{}+a\right].$$
(1.122)
#### 1.4.2 Spettro di Superstringa
La teoria descritta fin qui, è in realtà inconsistente come teoria quantistica anche per $`D=10`$; per ottenere una teoria consistente lo spettro deve essere opportunamente troncato come proposto inizialmente da Gliozzi, Scherk e Olive (proiezione GSO). Una prima ragione è la presenza nello spettro di un tachione nel settore NS dovuto al contributo negativo all’energia di vuoto di $`a^{NS}`$. Un seconda ragione risiede nella particolarità dello spettro di avere operatori anticommutanti $`\psi _r^\mu `$ che mappano bosoni in bosoni: dato uno stato bosonico $`|\varphi `$, lo stato ottenuto applicando un operatore anticommuntante $`\psi _r^\mu |\varphi `$, è ancora uno stato bosonico. Questo viola il teorema di spin-statistica, si tratta di un effetto poco naturale e quindi indesiderabile. La terza ragione che ci spinge a definire una proiezione dello spettro è che in questo modo è possibile ottenere uno spettro supersimmetrico. Infatti, se la supersimmetria sulla superficie bidimensionale d’universo è già presente nella teoria, dal punto di vista dello spazio-tempo 10 dimensionale, la supersimmetria va implementata con un opportuno troncamento dello spettro.
Nel settore NS, il tachione scalare $`0_{NS}`$ può essere eliminato defininendo un operatore di proiezione $`P_{GSO}^{\left(NS\right)}=\frac{1()^F}{2}`$, dove $`F`$ conta il numero di operatori fermionici. Lo stato fondamentale dello spettro proiettato diventa $`\psi _{\frac{1}{2}}^i0_{NS}`$. Nel settore R, dal momento che gli operatori $`\psi _0^i`$ soddisfano l’algebra di Clifford (a meno di un riscalamento)
$$\{\psi _0^i,\psi _0^j\}=\delta ^{ij},$$
(1.123)
lo stato fondamentale è una rappresentazione spinoriale dell’algebra di $`SO\left(8\right)`$, in particolare uno spinore di Majorana a massa nulla. Introducendo l’operatore di proiezione $`P_{GSO}^{\left(R\right)}=\frac{1+\left(1\right)^F\mathrm{\Gamma }_9}{2}`$, dove $`\mathrm{\Gamma }_9`$ è la matrice di chiralità definita nello spazio trasverso, lo stato fondamentale diventa uno spinore di Majorana-Weyl di chiralità definita e gli stati eccitati costruiti da questo con operatori di creazione fermionici avranno, alternativamente, chiralità destra e sinistra.
Lo spettro di bassa energia diella superstringa aperta GSO proiettato (teoria di Tipo I), contiene un vettore a massa nulla nel settore NS e un fermione di Majorana-Weyl nel settore R, che insieme formano un multipletto di super Yang-Mills in $`D=10`$. La teoria di tipo I risulta priva di anomalie .
Nelle stringhe chiuse la proiezione GSO porta ad uno spettro di bassa energia di $`𝒩=2`$ supergravità. La teoria che si ottiene scegliendo le chiralità dei vuoti destro e sinistro concordi, è detta di Tipo $`IIB`$, mentre la teoria con vuoti discordi è detta di Tipo $`IIA`$. Dal momento che gli spin sono additivi, i settori NS-NS e R-R dello spettro conterranno bosoni e i settori NS-R e R-NS fermioni. In termini di rappresentazioni di massa nulla si ha:
$`TipoIIA:\left(\mathrm{𝟠}_𝕧\mathrm{𝟠}_𝕤\right)\left(\mathrm{𝟠}_𝕧\mathrm{𝟠}_𝕔\right),`$
$`TipoIIB:\left(\mathrm{𝟠}_𝕧\mathrm{𝟠}_𝕤\right)\left(\mathrm{𝟠}_𝕧\mathrm{𝟠}_𝕤\right).`$ (1.124)
Decomponendo questi prodotti in termini di rappresentazioni del gruppo di Lorentz $`SO\left(8\right)`$, si trova il contenuto in termini di stati. Nel settore NS-NS si ottiene
$$\left(\mathrm{𝟠}_𝕧\mathrm{𝟠}_𝕧\right)=\mathrm{𝟙}\mathrm{𝟚𝟠}\mathrm{𝟛𝟝}=\mathrm{\Phi }+B_{\mu \nu }+h_{\mu \nu },$$
(1.125)
che identifichiamo con il dilatone, il tensore antisimmetrico e il gravitone. Allo stesso modo nel settore R-R si trova, per la $`IIB`$, una $`2`$-forma e una $`4`$-forma con una curvatura autoduale e, per la $`IIA`$ un vettore e una $`3`$-forma. Nei settori misti si hanno due gravitini e due spinori (detti dilatini). Nella $`IIA`$ i due gravitini sono di chiralità opposta e così anche i due dilatini mentre, nella $`IIB`$ i due gravitini hanno la stessa chiralità e chiralità opposta rispetto ai due dilatini. Si può vedere che, sebbene la IIB abbia spettro chirale, la teoria è priva di anomalie gravitazionali .
## Capitolo 2 Integrale di Polyakov
### 2.1 L’integrale funzionale di Polyakov
L’integrale sui cammini di Feynman è un metodo molto naturale per definire una teoria di campo. Anche in Teoria delle Stringhe è possibile definire un integrale sui “cammini” di stringa, ovvero una somma su tutte le possibile *storie* della stringa (world-sheet o superfici d’universo) che interpolino tra stati iniziali e finali. Le storie sono pesate da un fattore
$$e^{\left(iS/\mathrm{}\right)},$$
(2.1)
dove $`S`$ indica l’azione corrispondente. L’unico modo di implementare le interazioni in teoria di stringhe in maniera consistente con le simmetrie, è considerare come permesse le sole interazioni già implicite nella somma sui world-sheet. In questo modo si può ottenere una teoria consistente, priva di divergenze e unitaria. È interessante osservare come le interazioni siano indotte dalla topologia globale del world-sheet, mentre localmente le proprietè del world-sheet sono indistinguibili da quelle di una teoria libera. Si può osservare come, data una sezione di un diagramma di interazione, questa sembra aver luogo in un dato punto dello spazio-tempo, ma in realtà la sezione è arbitraria. È infatti possibile definire una nuova sezione del diagramma, che corrisponde ad effettuare una trasformazione di Lorentz, il cui risultato è spostare l’interazione in un punto differente dello spazio-tempo.
A seconda delle topologie che si considerano nella somma sui world-sheet, si definiscono differenti teorie di stringa. Disegnando un diagramma di stringa si hanno due tipi di “bordi”: i bordi definiti dalle stringhe iniziali e finali e quelli che corrispondono alle world-line degli eventuali estremi di stringa aperta. Limitiamoci a considerare i secondi per le definizioni seguenti. Esistono quattro possibili scelte per definire la somma sui world-sheet, che corrispondono a quattro distinte teorie di stringa:
1. stringhe chiuse orientate: tutti i world-sheet sono orientabili senza bordi;
2. stringhe chiuse non orientate: tutti i world-sheet sono senza bordi;
3. stringhe aperte e chiuse orientate: tutti i world sheet sono orientabili con bordi;
4. strighe aperte e chiuse non orientate: tutti i world sheet sono ammessi.
Non è possibile definire, invece, teorie di sole stringhe aperte. Consideriamo infatti un world-sheet con topologia di anello: questo è un diagramma di vuoto di stringa aperta, ma è anche interpretabile come diagramma di propagazione di stringa chiusa (figura). Quindi, partendo da una teoria di sole stringhe aperte, la somma sui world-sheet porterà ad includere necessariamente processi in cui lo scattering di stringhe aperte produce stringhe chiuse.
Sviluppiamo l’idea della somma sui world-sheet . Il punto di partenza è la teoria di stringa Euclidea, in cui la metrica Minkowkiana sul world-sheet $`\gamma _{ab}`$ è sostituita da una metrica Euclidea $`g_{ab}(\sigma ^1,\sigma ^2)`$.Consideriamo per semplicità una teoria bosonica. L’integrale funzionale che definisce la teoria è su tutte le metriche Euclidee e su tutte le immersioni $`X^\mu (\sigma ^1,\sigma ^2)`$ del world-sheet nello spazio-tempo di Minkowski:
$$\left[dX\right]\left[dg\right]e^{\left(S\right)}.$$
(2.2)
L’azione euclidea è
$$S=S_X+\lambda \chi ,$$
(2.3)
dove
$$S_X=\frac{1}{4\pi \alpha ^{}}_M𝑑\sigma ^2g^{1/2}g^{ab}_aX^\mu _bX_\mu ,$$
(2.4)
$$\chi =\frac{1}{4\pi }_M𝑑\sigma ^2g^{1/2}R+\frac{1}{2\pi }_M𝑑sk,$$
(2.5)
e $`k`$ è la curvatura geodetica dei bordi. Il vantaggio dell’integrale funzionale Euclideo è che l’integrale sulle metriche è ben definito. Sui world-sheet non banali a cui abbiamo accennato si possono avere metriche Euclidee non singolari, ma non sempre metriche Minkowskiane non singolari. Si può dimostrare inoltre che in due dimensioni il passaggio all’euclideo è ben giustificato e, la teoria è equivalente a quella Minkowkiana.
Il termine $`\chi `$ è localmente una derivata totale in due dimensioni e, quindi, dipende solo dalla topologia del world-sheet; è detto *numero di Eulero* del world-sheet. Il fattore $`e^{\left(\lambda \chi \right)}`$ nell’integrale funzionale pesa le differenti topologie nella somma sui world-sheet. Quindi il termine di Eulero nell’azione genera le potenze della costante di accoppiamento della teoria di stringa che possiamo definire come
$$g_o^2g_se^\lambda .$$
(2.6)
La costante $`\lambda `$ non è un parametro libero della teoria, ma è legata al valore di vuoto di un campo detto dilatone. Differenti valori di $`\lambda `$ non corrispondono quindi a diverse teorie ma a differenti *background* di una sola teoria.
### 2.2 Metodo di Faddeev-Popov
L’integrale funzionale (2.2) è mal definito, dal momento che l’integrazione avviene su infinite copie delle configurazioni fisiche collegate tra loro da trasformazioni di simmetria del tipo $`diff\times Weyl`$. Occorre dividere l’integrale funzionale per il volume del gruppo di simmetria locale,
$$Z\frac{\left[dX\right]\left[dg\right]}{V_{diff\times Weyl}}e^S.$$
(2.7)
Questo equivale ad integrare sulle sole configurazioni fisiche inequivalenti. Per farlo si deve fissare opportunamente un gauge, restringendo l’integrazione su una sezione dello spazio delle configurazioni $`(X,g)`$ che intersechi una sola volta ciascuna classe di equivalenza di gauge.
Nel capitolo precedente si è visto come sia possibile fissare localmente la metrica sul world-sheet nella forma diagonale
$$\widehat{g}_{ab}\left(\sigma \right)=\delta _{ab},$$
(2.8)
passando quindi da una metrica curva ad una piatta. Questo dal momento che in due dimensioni il tensore di Weyl si annulla identicamente. La trasformazione di Weyl dello scalare di Ricci è
$$g^{1/2}R^{}=g^{1/2}\left(R2^2\omega \right).$$
(2.9)
Risolvendo l’equazione $`2^2\omega =2`$ per $`\omega `$ si fissa $`R^{}=0`$. Dal momento che in due dimensioni il tensore di Riemann è proporzionale allo scalare di Ricci
$$R_{abcd}=\frac{1}{2}\left(g_{ac}g_{bd}g_{ad}g_{bc}\right)R,$$
(2.10)
si fissa a zero anche il tensore di Riemann, che equivale a scegliere una metrica piatta. Come si è visto la teoria con la metrica fissata conserva una invarianza conforme residua. Rivediamolo brevente per la teoria euclidea. A tal scopo introduciamo coordinate euclidee complesse $`z=\sigma ^1+i\sigma ^2`$, con metrica $`ds^2=dzd\overline{z}`$. Una trasformazione olomorfa di $`z`$,
$$z^{}\sigma ^1+i\sigma ^2=f\left(z\right),$$
(2.11)
combinata con una trasformazione di Weyl, manda la metrica in
$$ds^2=e^{2\omega }\left|_zf\right|^2dz^{}d\overline{z}^{},$$
(2.12)
e quindi scegliendo $`\omega =ln\left|_zf\right|`$, la metrica è invariante. Si è trovato che il gruppo di simmetria conforme è il sottogruppo delle trasformazioni $`diff\times Weyl`$, che lasciano invariata la metrica unitaria. È un risultato di grande valore, dal momento che indica la possibilità di utilizzare in Teoria delle Stringhe gli strumenti tipici delle Teorie di Campo conformi bidimensionali.
Dopo aver fissato la metrica l’integrale funzionale corre lungo una sezione parametrizzata da $`X^\mu `$ soltanto. Per ottenere la misura corretta occorre seguire, come nelle usuali teorie di campo, la procedura di Faddeev-Popov. L’idea è separare l’integrale funzionale in un integrale sul gruppo di gauge moltiplicato per un integrale lungo la sezione di gauge, in modo da poter normalizzare il primo termine. Il determinante di Faddeev e Popov è lo Jacobiano di questo cambio di variabili.
Consideriamo la combinazione di una trasformazione di coordinate e di una trasformazione di Weyl, che indichiamo come $`\zeta `$,
$$\zeta :g_{ab}\left(\sigma \right)g_{ab}^\zeta \left(\sigma ^{}\right)=e^{2\omega \left(\sigma \right)}\frac{\sigma ^c}{\sigma ^a}\frac{\sigma ^d}{\sigma ^b}g_{cd}\left(\sigma \right).$$
(2.13)
La definizione consueta della misura di Faddeev-Popov $`\mathrm{\Delta }_{FP}`$ è
$$1=\mathrm{\Delta }_{FP}\left(g\right)\left[d\zeta \right]\delta \left(g\widehat{g}^\zeta \right),$$
(2.14)
dove $`\widehat{g}_{ab}`$ indica la metrica in cui ci si vuole portare e si è indicata con $`\left[d\zeta \right]`$ la misura di integrazione gauge invariante sul gruppo $`diff\times Weyl`$. Non è difficile dimostrare che il determinante di Faddeev-Popov è gauge invariante:
$`\mathrm{\Delta }_{FP}\left(g^\zeta \right)^1`$ $`=`$ $`={\displaystyle \left[d\zeta ^{}\right]\delta \left(g^\zeta \widehat{g}^\zeta ^{}\right)}={\displaystyle \left[d\zeta ^{}\right]\delta \left(g\widehat{g}^{\zeta ^1\zeta ^{}}\right)}`$ (2.15)
$`=`$ $`{\displaystyle \left[d\zeta ^{\prime \prime }\right]\delta \left(g\widehat{g}^{\zeta ^{\prime \prime }}\right)}=\mathrm{\Delta }_{FP}\left(g\right)^1,`$
dove $`\zeta ^{\prime \prime }=\zeta ^1\zeta ^{}`$, nella seconda uguaglianza si è fatto uso dell’invarianza di gauge della funzione delta, e nella terza dell’invarianza della misura.
Inserendo l’identità (2.14) nell’integrale funzionale, si ottiene
$$Z\left[\widehat{g}\right]=\frac{\left[d\zeta \right]\left[dX\right]\left[dg\right]}{V_{diff\times Weyl}}\mathrm{\Delta }_{FP}\left(g\right)\delta \left(g\widehat{g}^\zeta \right)e^{S[X,g]}.$$
(2.16)
Integrando su $`g_{ab}`$ e rinominando la variabile di integrazione $`XX^\zeta `$, si ottiene
$$Z\left[\widehat{g}\right]=\frac{\left[d\zeta \right]\left[dX^\zeta \right]}{V_{diff\times Weyl}}\mathrm{\Delta }_{FP}\left(\widehat{g}^\zeta \right)e^{S[X^\zeta ,\widehat{g}^\zeta ]}.$$
(2.17)
Ricordando l’invarianza di gauge di $`\mathrm{\Delta }_{FP}`$, dell’azione e della misura dell’integrale funzionale si scrive
$$Z\left[\widehat{g}\right]=\frac{\left[d\zeta \right]\left[dX\right]}{V_{diff\times Weyl}}\mathrm{\Delta }_{FP}\left(\widehat{g}\right)e^{S[X,\widehat{g}]},$$
(2.18)
dove l’integrando non dipende da $`\zeta `$. L’integrazione su $`\zeta `$ porta il volume del gruppo che si cancella con il fattore al denominatore lasciando
$$Z\left[\widehat{g}\right]=\left[dX\right]\mathrm{\Delta }_{FP}\left(\widehat{g}\right)e^{S[X,\widehat{g}]}.$$
(2.19)
Per calcolare esplicitamente il determinante di Faddeev-Popov consideriamo la gauge completamente fissata dalla procedura di gauge fixing, in modo tale che per un solo valore di $`\zeta `$ la funzione $`\delta \left(g\widehat{g}^\zeta \right)`$ sia diversa da zero (in questo modo non si avranno i problemi legati alle *copie di Gribov*). Questo è possibile dal momento che la simmetria globale residua non modifica le proprietà locali del world-sheet. Consideriamo trasformazioni infinitesime di gauge: la variazione infinitesima della (2.13) risulta essere
$$\delta g_{ab}=2\delta \omega g_{ab}_a\delta \sigma _b_b\delta \sigma _a.$$
(2.20)
Definendo un operatore differenziale $`P_1`$ che mandi vettori in tensori a due indici a traccia nulla
$$\left(P_1\delta \sigma \right)_{ab}=\frac{1}{2}\left(_a\delta \sigma _b+_b\delta \sigma _ag_{ab}_c\delta \sigma ^c\right),$$
(2.21)
la 2.20 può essere riscritta nella forma
$$\delta g_{ab}=\left(2\delta \omega _c\delta \sigma ^c\right)g_{ab}2\left(P_1\delta \sigma \right)_{ab}.$$
(2.22)
Per trasformazioni $`\zeta `$ vicine all’identità, il determinante di Faddeev e Popov si scrive quindi
$$\mathrm{\Delta }_{FP}\left(\widehat{g}\right)^1=\left[d\delta \omega d\delta \sigma \right]\delta \left[\left(2\delta \omega _c\delta \sigma ^c\right)g_{ab}+2\left(P_1\delta \sigma \right)_{ab}\right],$$
(2.23)
e introducendo la rappresentazione esponenziale della funzione $`\delta `$ si ottiene
$$\mathrm{\Delta }_{FP}\left(\widehat{g}\right)^1=\left[d\delta \omega d\delta \sigma d\beta \right]e^{2\pi i{\scriptscriptstyle d^2\sigma \widehat{g}^{1/2}\beta ^{ab}\left[\left(2\delta \omega _c\delta \sigma ^c\right)g_{ab}+2\left(P_1\delta \sigma \right)_{ab}\right]}},$$
(2.24)
dove $`\beta ^{ab}`$ è un campo tensoriale simmetrico. Integrando su $`\delta \omega `$ si ottiene una funzione $`\delta `$ che forza $`\beta ^{ab}`$ ad avere traccia nulla. Denotando il tensore a traccia nulla con $`\beta ^{ab}`$ possiamo scrivere
$$\mathrm{\Delta }_{FP}\left(\widehat{g}\right)^1=\left[d\beta ^{}d\delta \omega \right]e^{4\pi i{\scriptscriptstyle d^2\sigma \widehat{g}^{1/2}\beta ^{ab}\left(\widehat{P}_1\delta \sigma \right)_{ab}}},$$
(2.25)
dove il cappuccio sull’operatore differenziale indica la presenza della metrica $`\widehat{g}_{ab}`$. È possibile, ricordando le proprietà delle variabili anticommutanti, invertire l’integrale funzionale sostituendo a ciascun campo bosonico da integrare un corrispondente campo anticommutante. Questi campi anticommuntanti prendono il nome di campi *ghost*. Quindi definendo i campi anticommutanti
$`\delta \sigma ^ac^a,`$
$`\beta _{ab}^{}b_{ab},`$ (2.26)
dove $`b_{ab}`$ è a traccia nulla, possiamo scrivere
$$\mathrm{\Delta }_{FP}\left(\widehat{g}\right)=\left[dbdc\right]e^{S_g}.$$
(2.27)
L’azione $`S_g`$ dei campi ghost, scegliendo una normalizzazione conveniente, è
$$S_g=\frac{1}{2\pi }d^2\sigma \widehat{g}^{1/2}b_{ab}\widehat{}^ac^b=\frac{1}{2\pi }d^2\sigma \widehat{g}^{1/2}b_{ab}\left(\widehat{P}_1c\right)^{ab}.$$
(2.28)
La procedura di gauge fixing introduce nell’integrale di Polyakov l’azione di campi ghost ‘non fisici’. Localmente l’integrale funzionale è
$$Z\left[\widehat{g}\right]=\left[dXdbdc\right]e^{S_XS_g}.$$
(2.29)
L’azione è quadratica nei campi e l’integrale può essere calcolato esattamente con le regole degli integrali gaussiani. Il calcolo dell’integrale porta a scrivere l’integrale di Polyakov come
$$Z\left[\widehat{g}\right]\left(det\widehat{}^2\right)^{D/2}det\widehat{P}_1,$$
(2.30)
dove il primo determinante ha origine dall’integrazione sui campi $`X`$ e il secondo dall’integrazione sui campi ghost.
Nella gauge conforme, in coordinate olomorfe $`z,\overline{z}`$, l’azione dei campi ghost diventa
$`S_g`$ $`=`$ $`{\displaystyle \frac{1}{2\pi }}{\displaystyle d^2z\left(b_{zz}_{\overline{z}}c^z+b_{\overline{z}\overline{z}}_zc^{\overline{z}}\right)}`$ (2.31)
$`=`$ $`{\displaystyle \frac{1}{2\pi }}{\displaystyle d^2z\left(b_{zz}_{\overline{z}}c^z+b_{\overline{z}\overline{z}}_zc^{\overline{z}}\right)}.`$
Le derivate covarianti in $`\overline{z}`$ di tensori con soli indici $`z`$ si riducono a derivate ordinarie, e vice versa, per via delle proprietà della connessione.
Nel caso di stringa aperta si deve anche considerare la presenza di bordi sul world sheet. È conveniente considerare l’integrale funzionale definito su campi presenti in regioni fissate. In questo modo l’invarianza sotto diffeomorfismi del world sheet deve essere limitata a cambi di coordinate che lascino i bordi invariati. La variazione di coordinate sulla superficie di universo $`\delta \sigma ^a`$ deve avere una componente normale nulla,
$$n_a\delta \sigma ^a=0.$$
(2.32)
Questa proprietà viene naturalmente trasmessa al corrispondente campo anticommutante, per cui si ha
$$n_ac^a=0.$$
(2.33)
Il campo $`c^a`$ è quindi propozionale al vettore tangente $`t^a`$. Le equazioni del moto impongono una condizione di bordo su $`b_{ab}`$
$$_M𝑑sn^ab_{ab}\delta c^b=0,$$
(2.34)
che insieme alla condizione di bordo su $`c^a`$ implica
$$n_at_bb^{ab}=0.$$
(2.35)
L’integrale funzionale nella forma (2.29) è ben definito dal momento che non presenta problemi di multiplo conteggio e può essere calcolato esplicitamente. La simmetria di gauge è stata fissata, ma l’integrale funzionale contiene una simmetria globale residua, *l’invarianza BRST*, come si vedrà in dettaglio nei prossimi paragrafi.
### 2.3 Quantizzazione BRST
Prima di discutere la quantizzazion BRST nel caso della stringa, passeremo brevemente in rassegna le sue proprietà in generale. Consideriamo una teoria con campi $`\varphi _i`$, che abbia una qualche simmetria di gauge, in modo tale che le trasformazioni di gauge soddisfino l’algebra<sup>1</sup><sup>1</sup>1Questa non è l’algebra più generale possibile, ma è sufficiente per una discussione introduttiva.
$$[\delta _\alpha ,\delta _\beta ]=f_{\alpha \beta }^\gamma \delta _\gamma .$$
(2.36)
Il gauge può essere fissato imponendo una condizione appropiata,
$$F^A\left(\varphi _i\right)=0.$$
(2.37)
Usando il metodo di Faddeev-Popov, l’integrale funzionale si scrive come
$`{\displaystyle \frac{\left[d\varphi \right]}{V_{gauge}}e^{S_0}}`$ $``$ $`{\displaystyle \left[d\varphi \right]\left[db_A\right]\left[dc^\alpha \right]\delta \left(F^A\left(\varphi \right)\right)e^{S_0{\scriptscriptstyle b_A\left(\delta _\alpha F^A\right)c^\alpha }}}`$ (2.38)
$``$ $`{\displaystyle \left[d\varphi \right]\left[dB_A\right]\left[db_A\right]\left[dc^\alpha \right]e^{S_0{\scriptscriptstyle B_AF^A\left(\varphi \right)}{\scriptscriptstyle b_A\left(\delta _\alpha F^A\right)c^\alpha }}}`$
$`=`$ $`{\displaystyle \left[d\varphi \right]\left[dB_A\right]\left[db_A\right]\left[dc^\alpha \right]e^S},`$
dove
$$S=S_0+S_1+S_2,S_1=iB_AF^A\left(\varphi \right),S_2=b_A\left(\delta _\alpha F^A\right)c^\alpha .$$
(2.39)
È utile osservare come l’indice $`\alpha `$ associato al campo ghost $`c_\alpha `$ è in corrispondenza uno ad uno con i parametri delle trasformazioni di gauge (2.36), mentre l’indice $`A`$ del campo di ghost $`b_A`$ è legato alle condizioni di gauge.
L’azione complessiva S è invariante sotto le trasformazioni di Becchi-Rouet-Stora-Tyutin (BRST) ,
$`\delta _{BRST}\varphi _i`$ $`=`$ $`iϵc^\alpha \delta _\alpha \varphi _i,`$
$`\delta _{BRST}b_A`$ $`=`$ $`ϵB_A,`$
$`\delta _{BRST}c^\alpha `$ $`=`$ $`{\displaystyle \frac{1}{2}}ϵc^\beta c^\gamma f_{\beta \gamma }^\alpha ,`$
$`\delta _{BRST}B_A`$ $`=`$ $`0,`$
dove il parametro $`ϵ`$ deve essere una costante anticommutante. La prima trasformazione è semplicemente la trasformazione di gauge originaria su $`\varphi _i`$ con il parametro di gauge sostituito con il campo ghost $`c_\alpha `$. C’è un numero di ghost conservato che è $`+1`$ per $`c_\alpha `$, $`1`$ per $`b_A`$ e $`ϵ`$, e $`0`$ per gli altri campi. Non è difficile accorgersi che i termini aggiuntivi dell’azione dovuti al gauge fixing in (2.38) possono essere scritti in termini di una trasformazione BRST
$$\delta _{BRST}\left(b_AF^a\right)=ϵ\left[B_AF^A\left(\varphi \right)+b_Ac^\alpha \delta _\alpha F^A\left(\varphi \right)\right].$$
(2.41)
La simmetria BRST è un’estensione della simmetria di gauge che sopravvive alla procedura di scelta della gauge. Consideriamo ora l’effetto di una piccola variazione della condizione di gauge $`\delta F^A`$ su un ampiezza fisica
$$ϵ\delta _F\psi |\psi ^{}=i\psi \left|\delta _{BRST}\left(b_A\delta F^A\right)\right|\psi ^{}=\psi \left|\{Q_{BRST},b_A\delta F^A\}\right|\psi ^{},$$
(2.42)
dove $`Q_{BRST}`$ è la carica conservata legata alla simmetria BRST. L’ampiezza corrispondente ad una osservabile non deve cambiare per variazioni della condizione di gauge. Dal momento che i campi $`c^\alpha `$ e $`b_A`$ sono legati rispettivamente al parametro di gauge $`ϵ^\alpha `$ e al moltiplicatore di Lagrange $`B_A`$, è naturale supporre che essi siano reali. Questo porta ad assumere che $`Q_{BRST}^{}=Q_{BRST}`$, e insieme alla (2.42) porta direttamente al risultato
$$Q_{BRST}|\varphi =0,$$
(2.43)
Ovvero tutti gli stati fisici $`|\varphi `$ devono essere BRST invarianti.
Per verificare se la carica BRST è effettivamente conservata, occorre verificare che essa commuti con la variazione dell’Hamiltoniana conseguente alla variazione della condizione di gauge,
$`0`$ $`=`$ $`[Q_{BRST},\delta H]=[Q_{BRST},\delta _B\left(b_A\delta F^A\right)]`$ (2.44)
$`=`$ $`[Q_{BRST},\{Q_{BRST},b_A\delta F^A\}]=[Q_{BRST}^2,b_A\delta F^A].`$
Dal momento che deve essere verificata per variazioni arbitrarie della condizione di gauge, si deve avere
$$Q_{BRST}^2=0,$$
(2.45)
ovvero la carica deve essere nilpotente per avere una descrizione quantistica consistente. Il caso $`Q_{BRST}^2=cost.`$ è escluso dal momento che $`Q_{BRST}^2`$ ha numero di ghost $`+2`$. Non è difficile verificare direttamente che, agendo due volte sui campi con le trasformazioni BRST, tutti i campi sono invarianti.
Vediamo le conseguenze fisiche delle proprietà di $`Q_{BRST}`$. Uno stato $`Q_{BRST}|\chi `$ sarà annichilato da $`Q_{BRST}`$ per qualsiasi $`|\chi `$ ed è quindi fisico. Eppure questo stato è ortogonale a tutti gli stati fisici, compreso se stesso. Tuttte le ampiezze fisiche che conivolgono questi stati sono quindi nulle, e stati di questo tipo sono per questo detti *stati nulli*. Due stati fisici che differiscano per uno stato nullo,
$$|\psi ^{}=|\psi +Q_{BRST}|\chi $$
(2.46)
avranno quindi gli stessi prodotti interni con tutti gli stati fisici, e saranno pertanto fisicamente indistinguibili. Come nel caso della quantizzazione covariante tradizionale, i veri stati fisici corrispondono all’insieme delle classi di equivalenza composte da stati che differiscono fra loro per stati nulli. Questa è una costruzione naturale per operatori nilpotenti, ed è conosciuta come la coomologia di $`Q_{BRST}`$. In coomologia ci si riferisce agli stati annichilati da $`Q_{BRST}`$ come stati *chiusi*, mentre a quelli della forma $`Q_{BRST}|\chi `$ come stati *esatti*. Lo spazio fisico degli stati è quindi
$$_{BRST}=\frac{_{chiusi}}{_{esatti}}.$$
(2.47)
#### 2.3.1 Quantizzazione BRST in Teoria delle Stringhe
Vediamo le trasformazioni BRST in teoria delle stringhe. Consideriamo oltre all’azione di Polyakov e dei campi ghost il termine di gauge fixing
$$\frac{i}{4\pi }d^2\sigma g^{1/2}B^{ab}\left(\delta _{ab}g_{ab}\right).$$
(2.48)
Integrando su $`B_{ab}`$ e utilizzando le equazioni del moto di $`g_{ab}`$ per eliminare $`B_{ab}`$ dalle trasformazioni, si ottengono le trasformazioni BRST nella forma
$`\delta _{BRST}X^\mu `$ $`=`$ $`iϵ\left(c+\stackrel{~}{c}\overline{}\right)X^\mu ,`$
$`\delta _Bb`$ $`=`$ $`iϵ\left(T^X+T^g\right),\delta _B\stackrel{~}{b}=iϵ\left(\stackrel{~}{T}^X+\stackrel{~}{T}^g\right),`$
$`\delta _Bc`$ $`=`$ $`iϵ\left(c+\stackrel{~}{c}\overline{}\right)c,\delta _B\stackrel{~}{c}=iϵ\left(c+\stackrel{~}{c}\overline{}\right)\stackrel{~}{c}.`$ (2.49)
Come si è visto, il ghost di Weyl forza $`b_{ab}`$ ad essere a traccia nulla. Dal teorema di Noether si trova la corrente della simmetria
$`j_{BRST}`$ $`=`$ $`cT^X+{\displaystyle \frac{1}{2}}:cT^g:+{\displaystyle \frac{3}{2}}^2c,`$ (2.50)
$`=`$ $`cT^X+:bcc:+{\displaystyle \frac{3}{2}}^2c:,`$
e la corrispondente corrente $`\stackrel{~}{j}_B`$. L’ultimo termine è una derivata totale e non contribuisce per questo alla carica BRST, ma viene aggiunto per garantire che $`j_{BRST}`$ trasformi come una corrente (tensore di rango 1). Le espansioni OPE per le correnti BRST con i campi dei ghost e con un generico tensore di materia sono
$`j_{BRST}b\left(0\right)`$ $``$ $`{\displaystyle \frac{3}{z^3}}+{\displaystyle \frac{1}{z^2}}j^g\left(0\right)+{\displaystyle \frac{1}{z}}T^{X+g}\left(0\right),`$
$`j_{BRST}c\left(0\right)`$ $``$ $`{\displaystyle \frac{1}{z}}cc\left(0\right),`$
$`j_{BRST}𝒪^m\left(0\right)`$ $``$ $`{\displaystyle \frac{h}{z^2}}c𝒪^m\left(0\right)+{\displaystyle \frac{1}{z}}\left[h\left(c\right)𝒪^m\left(0\right)+c𝒪^m\left(0\right)\right].`$ (2.51)
L’operatore di carica è
$$Q_{BRST}=\frac{1}{2\pi i}\left(dzj_{BRST}+d\overline{z}\stackrel{~}{j}_{BRST}\right).$$
(2.52)
Dall’azione di ghost si possono ricavare le equazioni del moto dei campi
$$_{\overline{z}}b_{zz}=_zb_{\overline{z}\overline{z}}=_{\overline{z}}c^z=_zc^{\overline{z}}=0.$$
(2.53)
Imponendo le condizioni di periodicità nel caso di stringa chiusa o le condizioni ai bordi nel caso di stringa aperta, si possono espandere i campi ghost in termini di modi di Fourier.
L’espansione OPE porta a trovare
$$\{Q_{BRST},b_m\}=L_m^X+L_m^g.$$
(2.54)
In termini di modi dei ghost si trova
$`Q_{BRST}`$ $`=`$ $`{\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}\left(c_nL_n^m+\stackrel{~}{c}_n\stackrel{~}{L}_n^m\right)`$ (2.55)
$`+`$ $`{\displaystyle \underset{m,n=\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle \frac{\left(mn\right)}{2}}:\left(c_mc_nb_{mn}+\stackrel{~}{c}_m\stackrel{~}{c}_n\stackrel{~}{b}_{mn}\right):+a^B\left(c_0+\stackrel{~}{c}_0\right).`$
La costante $`a^B=a^g`$ è fissata a $`1`$ calcolando l’anticommutatore
$$\{Q_{BRST},b_0\}=L_0^X+L_0^g.$$
(2.56)
Per verificare la proprietà di nilpotenza della carica BRST occorre calcolare l’anticommutatore
$$\{Q_{BRST},Q_{BRST}\},$$
(2.57)
che risulta essere uguale a $`0`$ solo per la dimesione critica $`D=26`$. Si tratta di un risultato atteso. La quantizzazione della stringa mostra infatti che c’è un anomalia nella simmetria di gauge (anomalia di Weyl) che si cancella solo per $`D=26`$.
In generale, in presenza di un gruppo di simmetria residuo, che non sia stato fissato dalle condizioni di gauge (in teoria delle stringhe si tratta del gruppo conforme i cui generatori sono $`L_m`$ e $`\stackrel{~}{L}_m`$), si deve richiedere che i suoi generatori si annullino negli elementi di matrice fisici. Dato un gruppo di simmetria residuo di generatori $`G_I`$, questi formeranno un’algebra
$$[G_I,G_J]=ig_{IJ}^KG_K.$$
(2.58)
Associata ad ogni generatore c’è una coppia di ghost, $`b_I`$ e $`c^I`$, con le relazioni di anticommutazione
$$\{c^I,b_J\}=\delta _I^J,\{c^I,c^J\}=\{b_I,b_J\}=0,$$
(2.59)
La forma generale della carica BRST, come può vedere dalla (2.55), è
$$Q_{BRST}=c^IG_I^m\frac{i}{2}g_{IJ}^Kc^Ic^Jb_K=c^I\left(G_I^m+\frac{1}{2}G_I^g\right),$$
(2.60)
dove $`G_I^m`$ è la parte di materia di $`G_I`$ e
$$G_I^g=ig_{IJ}^Kc^Jb_k,$$
(2.61)
è la parte dovuta ai campi ghost. Sia $`G_I^m`$ che $`G_I^g`$ soddisfano l’algebra (2.58). Usando i commutatori (2.58) e (2.59), si trova
$$Q_{BRST}^2=\frac{1}{2}\{Q_{BRST},Q_{BRST}\}=\frac{1}{2}g_{IJ}^Kg_{KL}^Mc^Ic^Jc^Lb_M=0.$$
(2.62)
L’uguaglianza a $`0`$ è dovuta all’identità di Jacobi dell’algebra dei $`G_I`$ che richiede che $`g_{IJ}^Kg_{KL}^M`$ si annulli antisimmetrizzando gli indici $`IJK`$. Il termine di carica centrale che si è ignorato deve essere introdotto opportunamente. Questa generalizzazione permetterà di trattare il caso di superstringa in maniera diretta.
#### 2.3.2 Spettro di stringa in quantizzazione BRST
Come visto nella discussione precedente, gli stati fisici sono quelli annichilati dall’operatore $`Q_{BRST}`$ che non siano esatti, ovvero non della forma $`Q_{BRST}|\chi `$. C’è una condizione addizionale che gli stati fisici devono rispettare,
$$b_0|\varphi =0.$$
(2.63)
Questa condizione, nota come *gauge di Siegel*, è dovuta al fatto che quando si calcolano le ampiezze d’interazione degli stati fisici, i propagatori contengono sempre fattori $`b_0`$, che effettivamente proiettano gli stati fisici sul sottospazio che soddisfa la (2.63) dal momento che $`b_0^2=0`$.
Lo spazio di Hilbert degli stati sarà dato dal prodotto tensoriale fra lo spazio di Hilbert degli stati creati dagli oscillatori di $`X^\mu `$ e quello creato dagli oscillatori dei campi ghost. Il vuoto degli stati di ghost è definito come lo stato annichilato dai modi di oscillazione dei ghost positivi
$$b_{n>0}|ghostvacuum=0,c_{n>0}|ghostvacuum=0.$$
(2.64)
C’è una sottigliezza dovuta alla presenza di zero modi $`b_0`$ e $`c_0`$ che soddisfano le condizioni $`b_0^2=0`$ $`c_0^2=0`$ e $`\{b_0,c_0\}=1`$. Queste relazioni definiscono l’algebra delle matrici $`\gamma `$ in due dimensioni. Questa rappresentazione contiene due stati: uno stato “spin alto” e uno stato “spin basso” tali che
$`b_0|`$ $`=`$ $`0,b_0|=|,`$
$`c_0|`$ $`=`$ $`0,c_0|=|.`$ (2.65)
Imporre anche la condizione (2.63) implica che il corretto vuoto di ghost sia lo stato $`|`$. Gli stati vengono creati a partire da questo vuoto agendo con i modi negativi dei ghost $`b_m`$, $`c_n`$. Non si può però agire con $`c_0`$ dal momento che lo stato ottenuto non rispetterebbe la condizione (2.63).
Vediamo ora i primi livelli dello spettro di stringa di aperta. Al livello zero si ha un solo stato, il prodotto scalare dei vuoti dei due spazi di Hilbert che indichiamo come $`|,k^\mu `$. Si deve avere
$$Q_{BRST}|,k^\mu =0,$$
(2.66)
dal momento che tutti i termini di $`Q_{BRST}`$ contengono $`L_0`$ e operatori di annichilazione. Questo mostra anche che non esistono a questo livello stati esatti, e quindi ogni stato invariante corrisponde a una classe di coomologia. Si tratta degli stati del tachione.
Al primo livello i possibili operatori sono $`\alpha _1^\mu `$, $`b_1`$ e $`c_1`$. Lo stato più generale che si può costruire è
$$|\psi _1=\left(\zeta \alpha _1+\xi _1c_1+\xi _2c_1\right)|,k^\mu ,$$
(2.67)
che contiene 28 parametri: i 26 parametri del vettore di polarizzazione $`\zeta _\mu `$ e le due costanti $`\xi _1`$ e $`\xi _2`$. La condizione BRST richiede
$`Q_{BRST}|\psi _1`$ $`=`$ $`\left(c_1k\alpha _1+c_1k\alpha _1\right)|\psi _1`$ (2.68)
$`=`$ $`2\left(\left(k\zeta \right)c_1+\xi _1k\alpha _1\right)|,k^\mu =0.`$
I termini proporzionali a $`c_0`$ sommano a zero per la condizione di mass-shell ($`k^2=0`$), e sono stati per questo omessi. L’uguaglianza a zero vale solo per $`k\zeta =\xi _1=0`$, e quindi rimangono solo $`26`$ stati linearmente indipendenti. Occorre controllare che questi stati non siano Q-esatti. Gli stati invarianti con norma nulla vengono creati da $`c_1`$ e $`k\alpha _1`$. Il più generale stato Q-esatto a questo livello, per $`k^2=0`$, è
$`Q_{BRST}|\chi =2\left(\left(k\zeta ^{}\right)c_1+\xi _{1}^{}{}_{}{}^{}k\alpha _1\right)|,k^\mu .`$ (2.69)
Quindi lo stato $`c_1|0;k`$ è Q-esatto, mentre il vettore di polarizzazione è trasverso con la relazione $`\zeta _\mu \zeta _\mu +\xi _1^{}k_\mu `$. Questo lascia i 24 stati di polarizzazione di una particella vettoriale a massa nulla, in accordo con quanto trovato con gli altri metodi di quantizzazione. Non è difficile verificare che gli stati fisici trovati sono a norma positiva.
La stessa procedura può essere seguita per i livelli più alti. Nel caso di stringhe chiuse occorre tenere conto degli operatori dei modi destri e sinistri.
### 2.4 Ampiezze d’urto
L’idea di considerare come proprie della teoria le sole interazioni implicite nella somma sui world-sheet che abbiano come bordi gli stati iniziali e finali di stringa è estremamente semplice e naturale. In generale è però molto complicato definire stati iniziali e finali consistenti con le simmetrie locali della superficie d’universo. Nel limite in cui si considerano sorgenti di stringa infinitamente lontane dalla regione di interazione, la definizione delle ampiezze di interazione è notevolmente semplificata. Per assegnati stati iniziali e finali, le ampiezze d’interazione di questi processi asintotici definiscono elementi della matrice S.
Non è difficile dare alcuni argomenti di tipo euristico sul tipo di superfici di universo legate ai processi asintotici. Si può immaginare che lontano dalla zona d’interazione le stringhe si propaghino liberamente. Come è facile immaginare, una striga chiusa che si propaghi liberamente descrive un cilindo, che può essere rappresentato in coordinate complesse $`w`$ come
$$2\pi tIm\left(w\right)0,ww+2\pi ,$$
(2.70)
dove l’estremo inferiore del cilindro $`Im\left(w\right)=2\pi t`$ è dato dalla sorgente esterna, mentre l’estremo superiore si congiunge al resto della superficie d’universo nella regione di interazione. Il limite in cui si hanno processi asintotici si ha, naturalmente, per $`t\mathrm{}`$. Il world-sheet complessivo di un processo di interazione di stringhe chiuse sarà quindi una superficie chiusa (la regione di interazione) su cui si inseriscono i tubi di propagazione delle stringhe entranti e uscenti. Nel limite $`t\mathrm{}`$, tenendo conto della simmetria conforme della teoria, i tubi di propagazione possono essere immaginati infinitamente lunghi e infinitamente stretti, e il world sheet d’interazione si riduce quindi ad una superficie chiusa con una “puntura” per ciscun stato esterno. Nel caso più semplice la superficie d’interazione è una sfera. In temini più precisi si può considerare la descrizione conformemente equivalente del tubo di propagazione che si ha in termini della coordinata $`z`$,
$$z=e^{iw},e^{2\pi t}\left|z\right|1,$$
(2.71)
che mappa i punti del cilindro nel disco unitario. In questa descrizione lo stato iniziale è il cerchio piccolo di raggio $`e^{2\pi t}`$, nel limite $`t\mathrm{}`$ il cerchio piccolo si riduce ad un punto.
Le considerazioni fatte valgono anche nel caso di stringhe aperte. Una stringa aperta propagandosi liberamente descrive un ‘nastro’, ovvero una superficie bidimensionale con bordi, che può essere rappresentato come
$$2\pi tIm\left(w\right)0,0Re\left(w\right)\pi ,$$
(2.72)
dove $`Im\left(w\right)=2\pi t`$ è la sorgente e $`Re\left(w\right)=0,\pi `$ sono i bordi descritti dagli estremi della stringa. La mappa $`z=e^{iw}`$,
$$e^{2\pi t}\left|z\right|1,Im\left(z\right)0,$$
(2.73)
manda il nastro nel semicerchio di raggio unitario posizionato all’origine nel semipiano superiore, a meno di un piccolo semicerchio di raggio $`e^{2\pi t}`$ dato dallo stato iniziale. Nel limite $`t\mathrm{}`$ le sorgenti si riducono ad un punto sul bordo del semipiano.
Come si è visto, ogni sorgente di stringa diventa una “perturbazione” locale sul world-sheet. Ad ogni stringa entrante o uscente di momento D-dimensionale $`k^\mu `$ e stato interno $`j`$, si può far corrispondere un *operatore di vertice* locale $`𝒱_j\left(k\right)`$. Il segno di $`k^0`$ distingue gli stati entranti da quelli uscenti, $`k^\mu =\pm (E,k)`$ dove il segno meno indica gli stati entranti.
Si è visto che l’integrale di Polyakov è una somma su tutte le superfici che collegano gli stati iniziali a quelli finali. Considerando solo superfici d’universo compatte con punture che rappresentino gli stati esterni, l’integrale funzionale porta naturalmente a definire gli elementi della matrice di interazione come
$$S_{j_1,\mathrm{},j_n}(k_1,\mathrm{},k_n)=\underset{\stackrel{topologie}{compatte}}{}\frac{\left[dXdg\right]}{V_{diff\times Weyl}}e^{S_X\lambda \chi }\underset{i=1}{\overset{n}{}}d^2\sigma _ig\left(\sigma _i\right)^{1/2}𝒱_{j_i}(k_i,\sigma _i).$$
(2.74)
Gli operatori di vertice, che saranno discussi nei prossimi capitoli, sono integrati sul world-sheet per rendere i loro contributi invarianti sotto diffeomorfismi.
I processi di interazione possono naturalmente coinvolgere anche topologie non connesse, ma questo tipo di ampiezze possono essere pensate come fattorizzazione di processi con topologie connesse: sarà quindi sufficiente limitarsi a questi ultimi. In due dimensioni la classificazione delle topologie compatte e connesse è ben nota. Ogni superficie compatta connessa orientata senza bordi è equivalente ad una sfera con $`h`$ manici (vedi figura 2.1). Si è visto che il numero di Eulero $`\chi `$ determina il peso delle superfici di world-sheet nell’espansione di Polyakov. I modelli di sole stringhe chiuse hanno la particolarità di avere un solo tipo di contributo ad ogni ordine della teoria perturbativa, con
$$\chi =22h,$$
(2.75)
e il loro sviluppo perturbativo è una somma su superfici chiuse di Riemann con un numero crescente di manici *h*.
Come si è detto, nelle teorie di stringhe aperte e chiuse non orientate lo sviluppo coinvolge anche superfici di Riemann non orientabili e con bordi che, quindi, possono contenere un numero variabile di due nuove strutture, bordi, *b*, e crosscap, *c*. In questo caso
$$\chi =22hbc,$$
(2.76)
e la serie perturbativa ora include potenze sia pari che dispari di $`g_s`$. Il genere *g* di una superficie viene definito come:
$$g=h+\frac{b}{2}+\frac{c}{2}.$$
(2.77)
Alla luce di queste osservazioni, emerge immediatamente la relativa semplicità dell’espansione perturbativa della Teoria delle Stringhe. Il numero di distinte topologie è molto piccolo comparato al numero di grafici di Feynman distinti in Teoria dei Campi: un singolo grafico di di stringa contiene molti grafici di campo dell’ordine perturbativo corrispondente (si vedano ad esempio all’ordine ad un loop i grafici di campo e di stringa, vedi figura 2.2). Questa caratteristica è stata usata anche in tecniche di calcolo dei diagrammi di Teoria dei Campi.
### 2.5 Spazio dei moduli
Si è visto che l’ampiezze di stringa (2.74) sono date, in generale, da un integrale funzionale sullo *spazio delle metriche*, $`𝒢_r`$, per una topologia $`r`$ del world-sheet. In una teoria di stringhe chiuse orientate, come si è visto, $`r`$ può essere pensato come il numero di manici $`h`$. Quozientando lo spazio delle metriche per il gruppo delle simmetrie di gauge, si ottiene lo *spazio dei moduli*,
$$_r=\frac{𝒢_r}{\left(diff\times Weyl\right)_r}.$$
(2.78)
In generale lo spazio dei moduli è parametrizzato da un numero finito di *moduli*.
La scelta di una metrica non fissa completamente la gauge. Infatti può esistere un sottogruppo di $`diff\times Weyl`$ che lasci la metrica invariante, detto *Gruppo Conforme di Killing* $`\left(CKG\right)`$.
In presenza di operatori di vertice nell’integrale funzione, è utile considerare la loro posizione sulla superficie d’interazione come moduli della superficie, alla stessa stregua dei *moduli della metrica*. In questo modo lo spazio dei moduli su cui si definisce l’integrale di Polyakov sarà dato da
$$_{r,n}=\frac{𝒢_r\times ^n}{\left(diff\times Weyl\right)_r},$$
(2.79)
dove $`^n`$ è lo spazio dei moduli che definscono le posizioni dei vertici di interazione. Il gruppo di simmetria residuo $`CKG`$, può essere fissato, assegnando ad un numero opportuno di vertici la posizione sul world-sheet.
In generale, $`diff_r`$ non è connesso, il quoziente del gruppo dei diffeomorfismi rispetto alla componente connessa che contiene l’identità,
$$\frac{diff_r}{diff_{r0}},$$
(2.80)
è detto *gruppo modulare*.
Riassumendo, per studiare in dettaglio lo spazio delle metriche in seguito al gauge fixing, occorre tener conto da un lato dei parametri della metrica che non possono essere rimossi dalle simmetrie (*moduli*), e dall’altro delle simmetrie residue codificate nel gruppo di Killing conforme.
Bisogna cercare le variazioni infinitesime della metrica che non siano equivalenti a trasformazioni di gauge $`diff\times Weyl`$ e quindi corrispondono a variazioni dei moduli. Occorre anche cercare le trasformazioni infinitesime di $`diff\times Weyl`$ che non cambino la metrica, che definiscono i vettori conformi di Killing $`CKV`$, elementi di $`CKG`$.
Una trasformazione infinitesima del gruppo di gauge si è vista essere
$$\delta g_{ab}=2\left(P_1\delta \sigma \right)_{ab}+\left(2\delta \omega \delta \sigma \right)g_{ab},$$
(2.81)
dove l’operatore simmetrico a traccia nulla $`P_1`$ è stato definito nella (2.21). I moduli corrispondono a variazioni della metrica $`\delta ^{}g_{ab}`$ che siano ortogonali a tutte le variazioni della forma (2.81),
$`0`$ $`=`$ $`{\displaystyle d^2\sigma g^{1/2}\delta ^{}g_{ab}\left[2\left(P_1\delta \sigma \right)^{ab}+\left(2\delta \omega \delta \sigma \right)g^{ab}\right]}`$ (2.82)
$`=`$ $`{\displaystyle d^2\sigma g^{1/2}\left[2\left(P_1^T\delta ^{}g\right)_a\delta \sigma ^a+\delta ^{}g_{ab}g^{ab}\left(2\delta \omega \delta \sigma \right)g^{ab}\right]}.`$
L’operatore trasposto $`P_1^T=^bu_{ab}`$, che manda tensori a due indici a traccia nulla in vettori, è definito come
$$(u,P_1v)=d^2\sigma g^{1/2}uP_1v=(P_1^Tu,v)=d^2\sigma g^{1/2}P_1^Tuv.$$
(2.83)
Perchè l’integrale (2.82) sia nullo per $`\delta \omega `$ e $`\delta \sigma `$ arbitrari si deve avere
$`g^{ab}\delta ^{}g_{ab}`$ $`=`$ $`0,`$
$`\left(P_1^T\delta ^{}g\right)_a`$ $`=`$ $`0.`$ (2.84)
La prima condizione impone che $`g_{ab}^{}`$ sia a traccia nulla. Per ogni soluzione distinta di queste equazioni si avrà un modulo della metrica.
I CKV sono trasformazioni (2.81) tali che $`\delta g_{ab}=0`$. Prendendo la traccia di questa equazione si fissa $`\delta \omega `$. L’equazione per i vettori conformi di Killing si riduce a
$$\left(P_1\delta \sigma \right)_{ab}=0.$$
(2.85)
Ponendosi in gauge conforme le equazioni (2.5) e (2.85) prendono forma semplice
$`_{\overline{z}}\delta ^{}g_{zz}`$ $`=`$ $`_z\delta ^{}g_{\overline{z}\overline{z}}=0,`$
$`_{\overline{z}}\delta z`$ $`=`$ $`_z\delta \overline{z}=0,`$ (2.86)
le variazioni dei moduli corrispondono a differenziali olomorfi quadratici e i $`CKV`$ a campi vettoriali olomorfi.
I moduli metrici corrispondono al kernel di $`P_1^T`$, e i CKV al kernel di $`P_1`$. Il teorema di Riemann-Roch collega il numero di moduli (reali) della metrica $`\mu =dimkerP_1^T`$, il numero di vettori di Killing conformi $`\kappa =dimkerP_1`$, e il numero di Eulero $`\chi `$ (2.75, 2.76),
$$\mu \kappa =3\chi .$$
(2.87)
Con una trasformazione di Weyl è sempre possibile porsi in una metrica in cui $`R`$ sia costante. In questo modo il segno di $`R`$ determina il segno di $`\chi `$ (2.5). Dal momento che $`P_1^TP_1=\frac{1}{2}^2\frac{1}{4}R`$, si ha
$`{\displaystyle d^2\sigma g^{1/2}\left(P_1\delta \sigma \right)_{ab}\left(P_1\delta \sigma \right)^{ab}}`$ $`=`$ $`{\displaystyle d^2\sigma g^{1/2}\delta \sigma _a\left(P_1^TP_1\delta \sigma \right)^a}`$ (2.88)
$`=`$ $`{\displaystyle d^2\sigma g^{1/2}\left(\frac{1}{2}_a\delta \sigma _b^a\delta \sigma ^b\frac{R}{4}\delta \sigma _a\delta \sigma ^a\right)}.`$
Per $`\chi `$ negativo, il termine di destra dell’equazione è strettamente positivo, quindi $`P_1\delta \sigma `$ non può essere mai nullo. In maniera simile si può dimostrare che $`P_1^T\delta ^{}\sigma `$ non può essere mai nullo per superfici con $`\chi `$ positivo. Riassumendo si è trovato che
$`\chi >0`$ $`:`$ $`\kappa =3\chi ,\mu =0,`$
$`\chi <0`$ $`:`$ $`\kappa =0,\mu =3\chi .`$ (2.89)
### 2.6 Superfici di Riemann
#### 2.6.1 Superfici di Riemann e varietà Riemanniane
Data una varietà è possibile definirne un ricoprimento con un insieme di aperti che si sovrappongano. All’interno degli aperti definiamo delle coordinate $`\sigma _m^a`$, dove $`m`$ indicizza gli aperti e $`a`$ le dimensioni della varietà. Quando due aperti $`m`$ e $`n`$ si sovrappongono, le rispettive coordinate saranno legate da *funzioni di trasizione*, $`f_{mn}`$, differenziabili
$$\sigma _m^a=f_{mn}^a\left(\sigma _n\right).$$
(2.90)
Per una varietà Riemanniana saranno definite metriche $`g_{m,ab}\left(\sigma _m\right)`$ in ogni aperto, collegate, nelle zone di sovrapposizione degli aperti, dalla legge di trasformazione dei tensori.
Per una varietà complessa, si avranno coordinate complesse $`z_m^a`$ in ogni aperto. In questo caso l’indice $`a`$ prende i valori da uno a $`d/2`$, dove $`d`$ è la dimensione della varietà. Le funzioni di transizione dovranno essere funzioni olomorfe,
$$z_m^a=f_{mn}^a\left(z_n\right).$$
(2.91)
È possibile definire funzioni olomorfe sulla varietà dal momento che questa proprietà non dipende dalle coordinate $`z_m^a`$ usate. Due varietà complesse sono equivalenti se esiste una mappa olomofa biunivoca fra loro. Un cambio di coordinate olomorfo in ogni aperto definisce per questo una superficie equivalente.
Nel caso di una dimensione complessa (due dimensioni reali), le varietà differenziali sono dette *superfici di Riemann*, e si ha la corrispondenza
$$superficidiRiemannvarieta^{}RiemannianemodWeyl.$$
(2.92)
Per verificare questo isomofismo, si può partire da una varietà Rimanniana e porsi, come già visto per la gauge conforme, in ogni aperto, in sistemi di coordinate $`z_m`$ tali che
$$ds^2dz_md\overline{z}_m.$$
(2.93)
Le metriche di aperti vicini non devono essere necessariamente uguali, ma le funzioni di transizione devono essere olomorfe per preservare la forma delle metriche (2.93). Questo prova la prima parte dell’isomorfismo. Per la mappa inversa, si può scegliere una metrica $`dz_md\overline{z}_m`$ nell’aperto $`m`$-esimo congiungendola con continuità alle metriche degli aperti sovrapposti per ottenere una varietà Riemanniana.
#### 2.6.2 Alcune superfici di Riemann
Vediamo in dettaglio le superfici di genere più basso che, come si vedrà sono cruciali nella determinazione delle ampiezza ai primi ordini della teoria perturbativa di stringa.
La sfera $`\chi =2`$, $`g=0`$, il disco e il piano proiettivo $`\left(\chi =1,g=1/2\right)`$ sono le uniche superfici di Riemann con numero di Eulero positivo.
##### La sfera
La Sfera $`S_2`$ può essere ricoperta con due aperti coordinati che si sovrappongano. Consideriamo due dischi $`\left|z\right|<\rho `$ e $`\left|u\right|<\rho `$ con $`\rho >1`$ uniti tramite l’identificazione $`u=1/z`$. Nel limite $`\rho \mathrm{}`$, la coordinata $`z`$ è ben definita ovunque tranne che per $`u=0`$. Si può quindi lavorare nel disco $`z`$, ricordandosi di verificare che nel punto di singolarità tutto vada come deve.
La sfera quindi può essere studiata come una superficie piatta, dotandola di metriche piatte nei due aperti connesse da trasformazioni conformi (trasformazioni di gauge della simmetria residua). Una metrica conforme, come si è visto, è della forma
$$ds^2=e^{2\omega (z,\overline{z})}dzd\overline{z},$$
(2.94)
Differenziando la relazione di identificazione, si ha $`dzd\overline{z}=\left|z\right|^4dud\overline{u}`$, la condizione di nonsingolarità della metrica in $`u=0`$ è che il fattore esponenziale sia almeno dell’ordine $`\left|z\right|^4`$ nel limite $`z\mathrm{}`$.
Ricordando i risultati precedentemente derivati, la sfera non ha moduli ma ha 6 vettori di Killing conformi. Le metriche saranno quindi tutte equivalenti a meno di diffeomorfismi e di trasformazioni di Weyl. Le equazioni infinitesime (2.5), nel caso della caso della sfera, devono valere in entrambi gli aperti coordinati. Dal momento che
$`\delta u`$ $`=`$ $`{\displaystyle \frac{u}{z}}\delta z=z^2\delta z,`$
$`\delta g_{uu}`$ $`=`$ $`\left({\displaystyle \frac{u}{z}}\right)^2\delta g_{zz}=z^4\delta g_{zz}.`$ (2.95)
I differenziali quadratici olomorfi $`\delta g_{zz}`$, devono essere olomofi in $`z`$ per le (2.5), e andare come $`\left|z\right|^4`$ all’infinito, in modo da verificare le equazioni anche nell’aperto $`u`$. Allo stesso modo i CKV devono crescere per $`z\mathrm{}`$ al più come $`z^2`$. La soluzione per $`\delta z`$ è della forma
$`\delta z`$ $`=`$ $`a_0+a_1z+a_2z^2,`$
$`\delta \overline{z}`$ $`=`$ $`a_0^{}+a_1^{}\overline{z}+a_2^{}\overline{z}^2,`$ (2.96)
dove gli $`a_i`$, sono i sei parametri reali (tre complessi), previsti dal teorema di Rimann-Roch. In queste trasformazioni si può riconoscere la forma infinitesima del gruppo $`PSL(2,)`$. Una trasformazione generale di questo gruppo è della forma
$$zz^{}=\frac{\alpha z+\beta }{\gamma z+\delta },$$
(2.97)
con $`\alpha ,\beta ,\delta ,\gamma `$ complessi, tali che $`\alpha \delta \beta \gamma =1`$. Naturalmente cambiando segno contemporaneamente a tutti i parametri complessi, la trasformazione è invariata.
##### Il disco
Ai fini della derivazione delle ampiezze è utile costruire il disco $`D_2`$ identificando i punti della sfera sotto riflessione. La relazione di riflessione più semplice è
$$z^{}=\overline{z},$$
(2.98)
che permette di considerare come regione fondamentale il semipiano superiore. L’asse reale è il luogo dei punti fissi rispetto alla relazione di riflessione e diventa quindi il bordo del disco. Una relazione di riflessione più complessa è
$$z^{}=\frac{1}{\overline{z}}.$$
(2.99)
Considerando coordinate polari $`z=re^{i\varphi }`$, si comprende che questa riflessione inverte il raggio lasciano l’angolo fissato. La regione fondamentale è il disco unitario.
Anche il disco non ha moduli. I vettori di Killing Conformi saranno il sottogruppo delle trasformazioni infinitesime di $`PSL(2,)`$, che lascino il bordo invariato. Queste trasformazioni, ricordando la relazione di riflessione, saranno della forma (2.97) ma con parametri reali.
##### Il Piano Proiettivo
Il piano proiettivo $`RP_2`$ può essere pensato come un’identificazione della sfera sotto $`_2`$. L’identificazione antipodale
$$z^{}=\frac{1}{\overline{z}},$$
(2.100)
non ha punti fissi e quindi non definisce bordi. Lo spazio risulta non orientato. La regione fondamentale può essere fissata sia nel semipiano superiore, sia nel disco unitario $`\left|z\right|1`$. La linea definita dall’equatore è detta crosscap ed è responsabile della non orientabilità della superficie. Il piano proiettivo reale non ha moduli, mentre i CKV sono il sottogruppo di $`PSL(2,)`$ che rispetta l’identificazione (2.100), e corrisponde al gruppo delle rotazioni $`SO\left(3\right)`$.
Studiamo ora le superfici con $`\chi =0`$, $`g=1`$ che, come si vedrà, sono legate alle ampiezze di vuoto in teoria delle stringhe. Le possibili superfici sono il *toro* ($`h=1,c=0,b=0`$), la *bottiglia di Klein* ($`h=0,c=0,b=2`$), l’anello ($`h=0,c=0,b=2`$) e il *nastro di M$`\ddot{o}`$bius* ($`h=0,c=1,b=1`$). In generale queste superfici possono essere aperte su un piano con un numero opportuno di tagli. In particolare, per superfici con numero di Eulero nullo è possibile definire una metrica euclidea sul piano.
##### Il toro
Il toro $`T^2`$ è una superficie chiusa orientabile e può essere immaginato come un cilindo i cui estremi vengano sovrapposti. Nel piano complesso il toro è definito dalle identificazioni
$$zz+n\lambda _1+m\lambda _2,$$
(2.101)
Dove $`\lambda _1`$ e $`\lambda _2`$ sono vettori nel piano complesso e $`m`$ e $`n`$ son due numeri interi. Queste identificazioni definiscono un reticolo la cui cella fondamentale è un parallelogramma con i lati opposti identificati, e con un opportuno riscalamento si può sempre scegliere il lato orizzontale di lunghezza unitaria. Il toro è quindi definito dall’assegnazione di un unico parametro complesso $`\tau =\tau _1+i\tau _2`$ con parte immaginaria positiva $`\tau _2`$, uguale al rapporto fra il lato obliquo e quello orizzontale della cella fondamentale, che viene detto *parametro di Teichmüller*, o modulo del toro. Non tutti i moduli $`\tau `$ definiscono tori inequivalenti. È infatti possibile ridefinire la cella fondamentale, ottendo tori equivalenti, traslando il lato orizzontale superiore di multipli della lunghezza orizzontale, o scambiando il lato orizzontale e quello obliquo della cella. Queste due operazioni sono rispettivamente generate dalle trasformazioni
$$T:\tau \tau +1,S:\tau \frac{1}{\tau }.$$
(2.102)
T e S sono i generatori del gruppo modulare $`PSL(2,)=SL(2,)/_2`$, la cui azione su $`\tau `$ è
$$\tau \tau ^{}=\frac{a\tau +b}{c\tau +d}conadbc=1ea,b,c,d,$$
(2.103)
dove il quoziente $`_2`$ è dovuto al fatto che invertendo i segni di $`a,b,c`$ e $`d`$ $`\tau `$ è invariante. Queste trasformazioni possono essere viste come ‘grandi’ diffeomorfismi sul toro, scrivendole nella forma
$`\left[\begin{array}{cc}\hfill \sigma ^1& \\ \hfill \sigma ^2& \end{array}\right]=\left[\begin{array}{cc}\hfill d& \hfill b\\ \hfill c& \hfill a\end{array}\right]\left[\begin{array}{cc}\hfill \sigma ^1& \\ \hfill \sigma ^2& \end{array}\right].`$ (2.110)
Usando le trasformazioni modulari (2.103), si può mostrare che ogni $`\tau `$ è equivalente ad un solo punto nella regione
$$=\{\frac{1}{2}<\tau _1\frac{1}{2},\tau 1\}$$
(2.111)
i cui bordi sono identificati come in figura. La regione del $`\tau `$-piano $``$ è detta *regione fondamentale* dello spazio dei moduli per il toro.
Per concludere questa breve presentazione del toro occorre ricordare la presenza di due vettori di Killing conformi che corrispondono a traslazioni rigide sulla superficie bidimensionale
$$\sigma ^a\sigma ^a+v^a.$$
(2.112)
Oltre a questo sottogruppo del gruppo $`diff\times Weyl`$, anche le trasformazioni discrete
$$\sigma ^a\sigma ^a$$
(2.113)
lasciano la metrica sul toro invariata.
##### La bottiglia di Klein
La bottiglia di Klein $`K_2`$ è una superficie chiusa non orientabile, e può essere vista come un cilindro i cui estremi siano congiunti dopo una trasformazione di parità $`\mathrm{\Omega }`$. Corrisponde al piano complesso con le identificazioni
$$zz+n\overline{z}+it,$$
(2.114)
dove l’unico modulo $`t`$ è definito sull’intevallo $`[0,\mathrm{}]`$. $`K_2`$ non ha alcun ricoprimento nello spazio tridimensionale Euclideo privo di auto-intersezioni, ma è ottenibile dal toro doppiamente ricoprente di modulo puramente immaginario $`\tau =2it`$ con l’identificazione
$$z^{}=\overline{z}+it.$$
(2.115)
C’è una seconda possibilità nella scelta del poligono fondamentale, ottenuta dimezzando il lato orizzontale e raddoppiando quello verticale, lasciando quindi l’area della cella inalterata. Il risultato è una rappresentazione equivalente di $`K_2`$ come un tubo che termina su due crosscap. Il tubo è dato dalla regione intena al poligono i cui lati orizzontali hanno ora la stessa orientazione, mentre i crosscap sono i due lati verticali, dove i punti che differiscono per traslazioni di metà della lunghezza dei lati sono identificati a coppie.
##### L’anello
L’anello $`C_2`$ è una superficie orientabile con due bordi. Nel piano complesso è la regione
$$0Re\left(z\right)1,zz+it,$$
(2.116)
una striscia di larghezza $`1`$ e lunghezza $`it`$ con i lati orizzontali identificati. Il modulo $`t`$ è definito sull’intevallo $`[0,\mathrm{}]`$, e anche in questo caso non c’è invarianza modulare. Il ricoprimento doppio del cilindro è il toro di modulo puramente immaginario $`\tau =it`$, da cui esso può essere ottenuto con l’identificazione
$$z^{}=\overline{z},$$
(2.117)
una riflessione rispetto all’asse immaginario. Le linee $`\sigma ^1=0,1`$ sono fisse rispetto all’involuzione e corrispondono ai bordi.
##### Il nastro di Möbius
L’ultima superficie con $`g=1`$ è il nastro di Möbius $`M_2`$, una superficie non orientabile con un bordo che può essere vista come una striscia chiusa con un rivolgimento indotto da una parità $`\mathrm{\Omega }`$,
$$0Re\left(z\right)1,z\overline{z}+1+it.$$
(2.118)
Si noti che in questo caso i due lati verticali della cella fondamentale descrivono differenti porzioni di un unico bordo. Anche per questa superficie è possibile ottenere una rappresentazione diversa, molto importante, con una ridefinizione della cella fondamentale. Raddoppiando il lato verticale e dimezzando quello orizzontale, $`M_2`$ è infatti rappresentabile come un tubo fra un bordo e un crosscap. Infatti nel nuovo poligono fondamentale uno dei lati verticali corrisponde all’unico bordo del nastro di Möbius, mentre nell’altro i punti sono identificati a coppie dopo la traslazione verticale tramite l’involuzione (2.118), ed è quindi un crosscap. È importante osservare che in questo caso, a differenza dei precedenti, si ha un toro doppiamente ricoprente ma con un modulo non puramente immaginario
$$\tau =\frac{1}{2}+\frac{1}{2}i\tau _2$$
(2.119)
Per superfici di genere più alto vale una relazione di equivalenza: tre crosscap possono essere sostituiti con un manico e un crosscap. Questo limita l’espansione di Polyakov a superfici con numero arbitrario di manici $`h`$ e di bordi $`b`$, ma solo con 0,1 o 2 crosscap $`c`$.
#### 2.6.3 Descrizioni equivalenti dello spazio dei moduli
La descrizione più naturale dello spazio dei moduli si ha scegliendo una sezione dello spazio delle metriche in modo da avere una sola metrica per ogni classe di equivalenza. Si definisce in questo modo una famiglia di metriche $`\widehat{g}_{ab}(t,\sigma )`$ parametrizzate dai moduli $`t^k`$. Una descrizione equivalente si ha considerando la metrica fissata e codificando i moduli nalla regione delle coordinate.
Nel caso del toro si può definire la regione di coordinate
$$0\sigma ^11,0\sigma ^21,$$
(2.120)
e quindi
$$(\sigma ^1,\sigma ^2)(\sigma ^1,\sigma ^2)+(m,n).$$
(2.121)
In questo sitema di coordinate non è possibile fissare una metrica unitaria preservando le condizioni di periodicità (2.121), ma è possibile sceglire la metrica della forma
$$ds^2=\left|d\sigma ^1+\tau d\sigma ^2\right|^2,$$
(2.122)
dove $`\tau `$ è una costante complessa.
Possiamo passare ad una descrizione equivalente in cui la metrica sia unitaria definendo nuove coordinate
$$\stackrel{~}{\sigma }^a\stackrel{~}{\sigma }^a+\left(mu^a+nv^a\right),$$
(2.123)
con periodicità definita dai due vettori arbitrari $`u^a`$ e $`v^a`$. Possiamo fissare $`u=(1,0)`$ ruotando e riscalando il sistema di coordinate. Rimangono non fissati i due parametri di $`v`$. Definendo $`w=\stackrel{~}{\sigma }^1+i\stackrel{~}{\sigma }^2`$, la metrica è $`ds^2=dwd\overline{w}`$, e la periodicità è
$$ww+\left(m+n\tau \right),$$
(2.124)
con $`\tau =v^1+iv^2`$. Naturalmente per $`w=\sigma ^1+\tau \sigma ^2`$ si ritorna alla descrizione (2.121). La descrizione del toro in termini di coordinate complesse $`w`$ illustra l’idea di una varietà complessa. Si pensi ad esempio di segliere un solo aperto un po’ più largo del poligono fondamentale del toro. In questo modo le condizioni di periodicità sono le funzioni di transizione nelle zone di sovrapposizioni fra i bordi opposti dell’aperto.
Nel definire l’integrale funzionale sulle metriche è più semplice considerare la metrica come una funzione di coordinate fissate come nel caso (2.121). Al contrario per studiare una teoria quantistica di campo su una data superficie è più semplice adottare una descrizione in cui la metrica sia unitaria con funzioni di transizione dipendenti dai moduli.
## Capitolo 3 Ampiezze di stringa bosonica
Partendo dall’idea della somma sulle “storie” di stringa, si è giunti a definire la formula (2.74) per le ampiezze di interazione. Per studiare in dettaglio le ampiezze di stringa per una data topologia, e quindi per un dato ordine perturbativo, la strategia è quella di ridurre l’integrale funzionale sulle metriche ad un integrale sui moduli che descrivono lo spazio delle metriche inequivalenti e sulle orbite di del gruppo di gauge $`diff\times Weyl`$. Questo si ottiene scegliendo un gauge e utilizzando la procedura di Faddeev-Popov.
Si può dimostrare che per uno spazio tempo piatto euclideo, in dimensione critica, le ampiezze si riducono a integrali ben definiti sullo spazio dei moduli. In particolare, in questo capitolo si svilupperà il formalismo funzionale per il calcolo della ampiezze e si calcoleranno esplicitamente le ampiezze all’ordine ad albero, e ad un loop per la stringa bosonica.
### 3.1 Ampiezze come integrali sullo spazio dei moduli
Data una teoria conforme con generici campi di materia $`\varphi _i`$ (per i quali $`c=\stackrel{~}{c}=26`$), l’integrale di Polyakov euclideo per la matrice S è
$$S_{j_1,\mathrm{},j_n}(k_1,\mathrm{},k_n)=\underset{\chi }{}\frac{\left[d\varphi dg\right]}{V_{diff\times Weyl}}e^{S_m\lambda \chi }\underset{i=1}{\overset{n}{}}d^2\sigma _ig\left(\sigma _i\right)^{1/2}𝒱_{j_i}(k_i,\sigma _i),$$
(3.1)
Dove la somma è sulle diverse caratteristiche di Eulero delle superfici. Vogliamo ora mostrare come, con un’opportuna scelta del gauge, l’integrale funzionale sulle metriche non viene completamente eliminato, ma si riduce ad un integrale finito-dimensionale sullo spazio dei moduli della superficie. In particolare, vogliamo mostrare che l’integrale sulle metriche e sulle posizioni dei vertici diventa un integrale sul gruppo di gauge, sui moduli e sulle posizioni (di una parte) dei vertici,
$$\left[dg\right]d^{2n}\sigma \left[d\zeta \right]d^\mu td^{2n\kappa }\sigma .$$
(3.2)
Dopo aver fattorizzato il volume del gruppo di gauge, lo Jacobiano per questa trasformazione fornisce precisamente la misura sullo spazio dei moduli.
Il determinante di Faddeev-Popov può essere definito più precisamente come
$$1=\mathrm{\Delta }_{FP}(g,\sigma )_Fd^\mu t_{diff\times Weyl}\left[d\zeta \right]\delta \left(g\widehat{g}\left(t\right)^\zeta \right)\underset{(a,i)f}{}\delta \left(\sigma _i^a\widehat{\sigma }_i^{\zeta a}\right).$$
(3.3)
Ogni metrica è genericamente equivalente, a meno di una trasformazione $`Diff\times Weyl`$, alla metrica di riferimento $`\widehat{g}\left(t\right)`$, per un solo valore di $`t`$ e $`\zeta `$, o, nel caso in cui siano presenti simmetrie residue discrete, per un numero finito $`n_R`$ di valori distinti.
Inserendo l’espressione (3.3) in (3.1) e invertendo l’ordine delle integrazioni si ottiene
$`S_{j_1,\mathrm{},j_n}(k_1,\mathrm{},k_n)={\displaystyle \underset{\stackrel{topologie}{compatte}}{}}{\displaystyle _F}d^\mu t\mathrm{\Delta }_{FP}(\widehat{g}\left(t\right),\widehat{\sigma }){\displaystyle \left[d\varphi \right]\underset{(a,i)f}{}d\sigma _i^a}`$
$`\times e^{S_m[\varphi ,\widehat{g}\left(t\right)]\lambda \chi }{\displaystyle \underset{i=1}{\overset{n}{}}}\left[\widehat{g}\left(\sigma _i\right)^{1/2}𝒱_{j_i}(k_i;\sigma _i)\right].`$ (3.4)
L’integrale è ora sullo spazio dei moduli $`F`$ e sulle coordinate dei vertici rimaste non fissate. Le simmetrie di gauge residue, legate al $`CKG`$, sono state utilizzate per fissare $`\kappa `$ posizioni dei vertici di interazione.
Valutiamo ora il determinante $`\mathrm{\Delta }_{FP}`$. Le funzioni $`\delta `$, come si è osservato possono essere diverse da zero in un numero finito $`n_R`$ di punti legati da simmetrie discrete residue. Per calcolare la misura di Faddeev-Popov, si può espandere nell’intorno di uno di questi punti, tenendo conto eventualmente della simmetria discreta e dividendo per $`n_R`$. La variazione generale della metrica è una variazione di gauge più una variazione nel modulo $`t^\mu `$,
$$\delta g_{ab}=\underset{k=1}{\overset{\mu }{}}\delta t^k_{t^k}\widehat{g}_{ab}2\widehat{g}_{ab}2\left(\widehat{P}_1\delta \sigma \right)_{ab}+\left(2\delta \omega \widehat{}\delta \sigma \right)\widehat{g}_{ab}.$$
(3.5)
Scrivendo le funzioni e i funzionali $`\delta `$ in forma esponenziale come integrali su $`\sigma ^a`$ e su $`\beta _{ab}`$, e integrando quindi su $`\delta \omega `$ per ottenere il tensore $`\beta _{ab}^{}`$ a traccia nulla per il determinate di Faddeev-Popov si ottiene l’espressione
$`\mathrm{\Delta }_{FP}(\widehat{g},\widehat{\sigma })^1=`$ $`n_R`$ $`{\displaystyle d^\mu \delta t\left[d\delta \sigma d\delta \omega \right]\underset{(a,i)f}{}\delta \left(\delta \sigma ^a\left(\widehat{\sigma }_i\right)\right)}`$
$`=`$ $`n_R`$ $`{\displaystyle d^\mu \delta td^\kappa x\left[d\beta ^{}d\delta \sigma \right]}`$ (3.6)
$`\times `$ $`\mathrm{exp}\left[2\pi i(\beta ^{},2\widehat{P}_1\delta \sigma \delta t^k_k\widehat{g})+2\pi i{\displaystyle \underset{(a,i)f}{}}x_{ai}\delta \sigma ^a\left(\widehat{\sigma }_i\right)\right].`$
Si possono ora sostituire le variabili bosoniche con variabili grassmaniane per invertire il determinante,
$`\delta \sigma `$ $``$ $`c^a`$
$`\beta ^{ab}`$ $``$ $`b_{ab}`$
$`x_{ai}`$ $``$ $`\eta _{ai}`$
$`\delta t^k`$ $``$ $`\xi ^k.`$ (3.7)
Scegliendo una normalizzazione opportuna per i nuovi campi, ottiene infine
$`\mathrm{\Delta }_{FP}(\widehat{g},\widehat{\sigma })`$ $`=`$ $`{\displaystyle \frac{1}{n_R}}{\displaystyle \left[dbdc\right]d^\mu \xi d^\kappa \eta }`$ (3.8)
$`\times exp\left[{\displaystyle \frac{1}{4\pi }}(b,2\widehat{P}_1c\xi ^k_k\widehat{g})+{\displaystyle \underset{(a,i)f}{}}\eta _{ai}c^a\left(\widehat{\sigma }_i\right)\right]`$
$`=`$ $`{\displaystyle \frac{1}{n_R}}{\displaystyle \left[dbdc\right]e^{S_g}\underset{k=1}{\overset{\mu }{}}\frac{1}{4\pi }(b,_k\widehat{g})\underset{(a,i)f}{}c^a\left(\widehat{\sigma }_i\right)}.`$
dopo aver effettuato l’integrazione sulle variabili grassmaniane $`\eta _{ai}`$ e $`\xi ^k`$.
Nel sostituire l’espressione trovata nell’integrale funzionale si può fissare il segno complessivo in modo tale da ottenere un risultato positivo. L’espressione finale per la matrice S è pertanto
$`S_{j_1,\mathrm{},j_n}(k_1,\mathrm{},k_n)`$ $`=`$ $`{\displaystyle \underset{\stackrel{topologie}{compatte}}{}}{\displaystyle _F}{\displaystyle \frac{d^\mu t}{n_R}}{\displaystyle \left[d\varphi dbdc\right]e^{S_mS_g\lambda \chi }}`$
$`\times `$ $`{\displaystyle \underset{(a,i)f}{}}{\displaystyle 𝑑\sigma _i^a\underset{k=1}{\overset{\mu }{}}\frac{1}{4\pi }(b,_k\widehat{g})\underset{(a,i)f}{}c^a\left(\widehat{\sigma }_i\right)\underset{i=1}{\overset{n}{}}\widehat{g}\left(\sigma _a\right)^{1/2}𝒱_{j_i}(k_i,\sigma _i)}.`$
(3.9)
Questo risultato può essere esteso a tutte le costruzioni di stringa bosonica, estendendo eventualmente la classe delle superfici incluse e variando il tipo di vertici permessi. È importante notare che ogni coodinata fissata utilizzando le simmetrie di gauge residue associate ai $`CKV`$, l’integrazione sulla posizione di un vertice venga sostituita con un fattore $`c_i^a`$, mentre ogni modulo metrico da luogo ad un inserzione di $`b`$.
#### 3.1.1 Calcolo della misura di Faddeev-Popov
L’espressione (3.1) per le ampiezze di interazione può essere ulteriormente ridotta, calcolando direttamente il termine di Faddeev-Popov nell’integrale funzionale.
Definiamo una base completa di funzioni su cui espandere i campi di ghost. Ricordando le definizioni del prodotto scalare (2.83), l’azione di ghost può essere scritta in forme equivalenti,
$$S_g=\frac{1}{2\pi }(b,P_1c)=\frac{1}{2\pi }(P_1^Tb,c).$$
(3.10)
Dal momento che l’operatore $`P_1`$ manda vettori in tensori a due indici, non è possibile diagonalizzarlo. Si possono però diagonalizzare $`P_1^TP_1`$ e $`P_1P_1^T`$,
$$P_1^TP_1C_J^a=\nu _{}^{}{}_{J}{}^{2}C_J^a,P_1P_1^TB_{Kab}=\nu _K^2B_{Kab}.$$
(3.11)
Le autofunzioni corrispondenti possono essere scelte reali e normalizzate in relazione ai rispettivi prodotti scalari,
$`(C_J,C_J^{})`$ $`=`$ $`{\displaystyle d^2\sigma g^{1/2}C_J^aC_{J^{}a}}=\delta _{JJ^{}},`$
$`(B_K,B_K^{})`$ $`=`$ $`{\displaystyle d^2\sigma g^{1/2}B_{Kab}B_K^{}^{ab}}=\delta _{KK^{}}.`$ (3.12)
Si può inoltre osservare che
$$\left(P_1P_1^T\right)P_1C_J=P_1\left(P_1^TP_1\right)C_J=\nu _{}^{}{}_{J}{}^{2}P_1C_J,$$
(3.13)
e questo comporta che $`P_1C_J`$ è una autofunzione di $`P_1P_1^T`$. Nello stesso modo si dimostra che $`P_1^TB_K`$ è una autofunzione di $`P_1^TP_1`$. Esiste quindi una corrispondenza biunivoca fra le autofunzioni dei due operatori, eccetto nei casi $`P_1C_J=0`$ e $`P_1^TB_K=0`$, ovvero per gli autovalori nulli degli operatori $`P_1^TP_1`$ e $`P_1P_1^T`$. Questi autovalori sono i $`\kappa `$ vettori di Killing conformi e i $`\mu `$ differenziali quadratici olomorfi. Indichiamo le autofunzioni di autovalore nullo come $`C_{0j}`$ e $`B_{0k}`$, e le restanti autofunzioni con $`C_J`$, $`B_K`$ per $`J,K=1,2,\mathrm{}`$. Le autofunzioni corrispondenti ad autovalori non nulli sono quindi legate dalle relazioni
$$B_{Jab}=\frac{1}{\nu _J}\left(P_1C_J\right)_{ab},\nu _J=\nu _J^{}0.$$
(3.14)
I campi di ghost possono essere sviluppati nelle basi di autofunzioni appena definite,
$$c^a\left(\sigma \right)=\underset{J}{}c_JC_J^a\left(\sigma \right),b_{ab}=\underset{K}{}b_KB_{Kab}\left(\sigma \right),$$
(3.15)
e in termini dei modi l’integrale funzionale $`\mathrm{\Delta }_{FP}`$ sui ghost diventa
$$\underset{k=1}{\overset{\mu }{}}db_{0k}\underset{j=1}{\overset{\kappa }{}}dc_{0j}\underset{J}{}db_Jdc_Je^{\frac{\nu _Jb_Jc_J}{2\pi }}\underset{k=1}{\overset{\mu }{}}\frac{1}{4\pi }(b,_k\widehat{g})\underset{(a,i)f}{}c^a\left(\sigma _i\right).$$
(3.16)
Gli integrali sulle variabili grassmaniane sono diversi da zero solo quando la variabile compare nell’integrando. I modi nulli $`c_{0j}`$ e $`b_{0k}`$ non compaiono nell’azione, ma solo nelle inserzioni, e correttamente il numero di inserzioni di ciascun tipo di ghost è uguale rispettivamente a $`\kappa `$ e $`\mu `$, esattamente i numeri di modi zero. L’integrale funzionale risultante fattorizza nella forma
$`\mathrm{\Delta }_{FP}={\displaystyle \underset{k=1}{\overset{\mu }{}}db_{0k}}`$ $`{\displaystyle \underset{k^{}=1}{\overset{\mu }{}}}\left[{\displaystyle \underset{k^{\prime \prime }=1}{\overset{\mu }{}}}{\displaystyle \frac{b_{0k^{\prime \prime }}}{4\pi }}(B_{0k^{\prime \prime }},_k^{}\widehat{g})\right]`$ (3.17)
$`\times {\displaystyle }{\displaystyle \underset{j=1}{\overset{\kappa }{}}}dc_{0j}{\displaystyle \underset{(a,i)f}{}}\left[{\displaystyle \underset{j^{}=1}{\overset{\kappa }{}}}c_{0j^{}}C_{0j^{}}^a\left(\sigma _i\right)\right]`$
$`\times {\displaystyle }{\displaystyle \underset{J}{}}db_Jdc_Je^{\frac{\nu _Jb_Jc_J}{2\pi }}.`$
La saturazione degli integrali sui modi zero dei due campi ghost con le variabili grassmaniane da luogo a due determinanti finito-dimensionali. I modi restanti danno invece luogo ad un prodotto infinito che ricostruisce un determinante funzionale. Si ha quindi
$$\mathrm{\Delta }_{FP}=det\frac{(B_{0k},_k^{}\widehat{g})}{4\pi }detC_{0j}^a\left(\sigma _i\right)\stackrel{}{det}\left(\frac{P_1^TP_1}{4\pi ^2}\right)^{1/2},$$
(3.18)
dove il determinante “primato” indica l’assenza degli zero modi e $`C_{0j}^a`$ è correttamente una matrice quadrata dal momento che sia gli indici $`(a,i)f`$ che gli indici $`j`$ corrono sui $`\kappa `$ valori.
#### 3.1.2 Differenziali di Beltrami e misura di integrazione
Per derivare la misura di Faddeev-Popov anche nella rappresentazione in cui i moduli siano codificati nelle funzioni di transizione introduciamo il concetto di *differenziali di Beltrami*. I differenziali di Beltrami sono una base nello spazio duale a quello dei differenziali olomorfi quadratici, rispetto ai quali possono essere integrati.
Data una superficie di Riemann, consideriamo un suo ricoprimento di aperti con coordinate complesse $`z_m`$, dove l’indice $`m`$ identifica gli aperti, e funzioni di transizioni olomorfe. Dato un punto $`t_0`$ dello spazio dei moduli, sia definita in queste coordinate una metrica $`\widehat{g}\left(t_0\right)`$, equivalente a meno di trasformazioni di Weyl a $`dz_md\overline{z}_m`$.
Vogliamo descrivere l’effetto della variazione dei moduli prima nella rappresentazione in cui la metrica dipenda esplicitamente da essi per passare quindi nella rappresentazione in cui la metrica è fissata e le funzioni di transizione dipendono esplicitamente dai moduli.
Definiamo il differenziale di Beltrami nella prima descrizione come
$$\mu _{ka}^b=\frac{1}{2}\widehat{g}^{bc}_k\widehat{g}_{ac},$$
(3.19)
e le inserzione del campo $`b`$ sono
$$\frac{1}{2\pi }(b,\mu _k)=\frac{1}{2\pi }d^2z\left(b_{zz}\mu _{k\overline{z}}^z+b_{\overline{z}\overline{z}}\mu _{kz}^{\overline{z}}\right).$$
(3.20)
Nella seconda descrizione dopo una variazione dei moduli $`\delta t^k`$, si avranno le variazioni di coordinate
$$z_m^{}=z_m+\delta t^kv_{km}^{z_m}(z_m,\overline{z}_m).$$
(3.21)
Le metriche risultanti devono essere equivalenti, a meno di trasformazioni di Weyl, per i punti dello spazio dei moduli $`t_0^k`$ e $`t_0^k+\delta t^k`$,
$$dz_md\overline{z}_mdz_md\overline{z}_m+\delta t^k\left(\mu _{kz_m}^{\overline{z}_m}dz_mdz_m+\mu _{k\overline{z}_m}^{z_m}d\overline{z}_md\overline{z}_m\right).$$
(3.22)
La variazione delle coordinate può quindi essere espressa in termini del differenziale di Beltrami,
$$\mu _{kz_m}^{\overline{z}_m}=_{z_m}v_{kz_m}^{\overline{z}_m},\mu _{k\overline{z}_m}^{z_m}=_{\overline{z}_m}v_{kz_m}^{z_m}.$$
(3.23)
In questa forma si risconosce la versione infinitesimale dell’*equazione di Beltrami*. In realtà, queste equazioni non determinano completamente $`v_{kz_m}^{\overline{z}_m}`$ e $`v_{kz_m}^{z_m}`$. La parte olomorfa di $`v_{kz_m}^{\overline{z}_m}`$ e la parte antiolomorfa di $`v_{kz_m}^{z_m}`$, che restano non fissate, corrispondono alla possibilità di operare riparametrizzazioni olomorfe. Sostituendo nella (3.20) e integrando per parti si trova
$$\frac{1}{2\pi }(b,\mu _k)=\frac{1}{2\pi i}\underset{m}{}_{C_m}\left(dz_mv_{km}^{\overline{z}_m}b_{z_mz_m}d\overline{z}_mv_{km}^{\overline{z}_m}b_{\overline{z}_m\overline{z}_m}\right),$$
(3.24)
dove l’integrazione è in senso antiorario lungo i contorni $`C_m`$ che circondano gli aperti.
Ricordando la (3.21), la derivata delle funzioni di trasizione rispetto ai moduli risulta essere
$`{\displaystyle \frac{z_m}{t^k}}|_{z_n}`$ $`=`$ $`{\displaystyle \frac{dz_m}{dt^k}}{\displaystyle \frac{z_m}{z_n}}|_t{\displaystyle \frac{dz_n}{dt^k}}=`$ (3.25)
$`=`$ $`v_{km}^{z_m}{\displaystyle \frac{z_m}{z_n}}|_tv_{kn}^{z_n}=v_{km}^{z_m}v_{kn}^{z_n}.`$
I contorni di integrazione degli aperti vicini si posso combinare in modo da scrivere
$$\frac{1}{2\pi }(b,\mu _k)=\frac{1}{2\pi i}\underset{mn}{}_{C_{mn}}\left(dz_m\frac{z_m}{t^k}|_{z_n}b_{z_mz_m}d\overline{z}_m\frac{\overline{z}_m}{t^k}|_{z_n}b_{\overline{z}_m\overline{z}_m}\right).$$
(3.26)
La somma corre su tutte le coppie di aperti che si sovrappongano. I contorni $`C_{mn}`$ sono definiti intorno alle sovrapposzioni degli aperti $`m`$ e $`n`$, in senso antiorario dal punto di vista di $`m`$. Le inserzioni sono ora espresse in termini delle funzioni di transizione.
### 3.2 Operatori di vertice
Come si è visto, la costruzione dell’ampiezze di stringa richiede che gli stati esterni vengano mappati in un numero finito di punti. In ognuno di questi punti in cui una stringa entra o esca devono comparire operatori locali con i giusti numeri quantici degli stati di interesse. Si è quindi portati all’idea di associare ad ogni stato della stringa un qualche operatore locale della teoria di campo conforme bidimensionale. Questi operatori sono detti *operatori di vertice*, e una richiesta importante è che gli operatori di vertice abbiano dimensione conforme $`(1,1)`$ per le stringhe chiuse e $`1`$ per le stringhe aperte. Questo garantisce infatti la loro integrabilità sul world-sheet.
Un operatore di vertice di uno stato fisico di momento fissato $`k`$ deve obbedire ad alcune condizioni di covarianza. Essenzialmente, esso deve essere consistente con tutte le simmetrie della corrispondente teoria. Per teorie di stringa bosonica gli operatori di vertice devono quindi avere seguenti proprietà:
1. invarianza per traslazioni spazio-temporali. La dipendenza dal campo $`X^\mu `$ può quindi essere indotta solo da fattori della forma $`\mathrm{exp}\left(ikX\right)`$ o da derivate di $`X^\mu `$;
2. invarianza per trasformazioni di Lorentz. Gli indici spazio-temporali $`\mu `$ devono essere tutti saturati;
3. invarianza per riparametrizzazioni della superficie di universo. Gli indici bidimensionali devono essere tutti contratti e deve comparire un fattore di volume $`g^{1/2}`$;
4. invarianza per trasformazioni di Weyl.
L’operatore di vertice più generale è quindi della forma della forma
$$𝒱(ϵ,k)=_Md^2\sigma g^{1/2}W(ϵ,X,R)e^{ikX},$$
(3.27)
dove $`W`$ è un polinomio scalare e $`ϵ`$ è un tensore di polarizzazione. Ad esempio l’operatore di vertice per un tachione di stringa chiusa è
$`V_0`$ $`=`$ $`2g_c{\displaystyle d^2\sigma g^{1/2}e^{ikX}}`$ (3.28)
$`=`$ $`g_c{\displaystyle d^2z}:e^{ikX}:,`$
dove abbiamo anche incluso la costante di accoppiamento di stringa chiusa $`g_c`$. In maniera analoga, per il primo stato eccitato si ottiene
$$V_1=d^2z:X^\mu \overline{}X^\nu e^{ikX}:.$$
(3.29)
La richiesta dell’invarianza di Weyl per i vertici porta alla condizione di mass-shell sugli impulsi. In Teorie delle Stringhe è quindi possibile soltanto definire operatori di vertice *on-shell*.
L’estensione al caso di stringa aperta coinvolge integrali sui bordi, ad esempi il vertice del tachione è
$$V_0=g_o_M𝑑s\left[e^{ikX}\right]_r,$$
(3.30)
che risulta invariante per trasformazioni di Weyl se la condizione di mass-shell $`k^2=1/\alpha ^{}`$ è verificata. Per il vertice del fotone si trova invece
$$V_1=i\frac{g_o}{\left(2\alpha ^{}\right)^{1/2}}e_\mu _M𝑑s\left[\dot{X}e^{ikX}\right]_r,$$
(3.31)
che risulta invariante per trasformazioni di Weyl per $`k^2=0`$.
### 3.3 Ampiezze al livello ad albero
All’ordine più basso le ampiezze di interazione coinvolgono superfici con caratteristica di Eulero positiva e numeri variabili di vertici di interazione. A questo livello le ampiezze, dette *ampiezze ad albero*, corrispondono alla teoria classica. Le correzioni quantistiche si ottengono tenendo conto, nello sviluppo perturbativo, delle correzioni a loop.
#### 3.3.1 Funzioni di correlazione
Prima di specializzare i calcoli dei valori di aspettazione al caso delle tre superfici che compaiono nelle ampiezze ad albero, facciamo alcune considerazioni generali per una superficie compatta arbitraria $`M`$. Il punto di partenza, come in Teoria dei Campi, è un funzionale generatore, che indichiamo in generale come
$$Z\left[J\right]=e^{i{\scriptscriptstyle d^2\sigma J\left(\sigma \right)X\left(\sigma \right)}},$$
(3.32)
dove la notazione $`\mathrm{}`$ indica l’integrale funzionale che definisce la teoria, con l’inserzione di opportuni campi. Consideramo un set $`𝒳_I\left(\sigma \right)`$ di autosoluzioni dell’operatore cinetico $`^2`$, opportunamente normalizzate,
$`^2𝒳_I`$ $`=`$ $`\omega _I^2𝒳_I,`$
$`{\displaystyle d^\sigma g^{1/2}𝒳_I𝒳_I^{}}`$ $`=`$ $`\delta _{II^{}}.`$ (3.33)
Espandendo i campi $`X^\mu \left(\sigma \right)`$ e $`J^\mu \left(\sigma \right)`$ in termini delle autosoluzioni
$`X^\mu \left(\sigma \right)`$ $`=`$ $`{\displaystyle \underset{I}{}}x_I𝒳_I\left(\sigma \right),`$
$`J^\mu \left(\sigma \right)`$ $`=`$ $`{\displaystyle \underset{I}{}}J\mu _I𝒳_I\left(\sigma \right),`$ (3.34)
l’integrale funzionale diventa
$$Z\left[J\right]=\underset{I,\mu }{}𝑑x_I^\mu e^{\frac{\omega _I^2x_I^\mu x_{I\mu }}{4\pi \alpha ^{}}+ix_I^\mu J_{I\mu }}.$$
(3.35)
Si ottiene quindi un prodotto di integrali gaussiani, che si possono esprimere in termini degli autovalori, eccetto che per il modo nullo $`𝒳_0`$, per il quale l’azione è nulla, che da luogo ad una funzione delta,
$`Z\left[J\right]`$ $`=`$ $`i\left(2\pi \right)^d\delta ^d\left(J_0\right){\displaystyle \underset{I0}{}}\left({\displaystyle \frac{4\pi ^2\alpha ^{}}{\omega _I^2}}\right)^{d/2}e^{\frac{\pi \alpha ^{}J_IJ_I}{\omega _I^2}}`$ (3.36)
$`=`$ $`i\left(2\pi \right)^d\delta ^d\left(J_0\right)\left(\stackrel{}{det}{\displaystyle \frac{^2}{4\pi ^2\alpha ^{}}}\right)^{d/2}e^{\frac{1}{2}{\scriptscriptstyle d^2\sigma d^2\sigma ^{}J\left(\sigma \right)J\left(\sigma ^{}\right)G^{}(\sigma ,\sigma ^{})}}.`$
Nell’espressione ottenuta, per riscrivere l’esponenziale, è stata utilizzata l’espressione dei coefficienti di espansione della corrente $`J^\mu `$
$$J_I^\mu =d^2\sigma J^\mu \left(\sigma \right)𝒳_I\left(\sigma \right).$$
(3.37)
Per ottenere degli integrali gaussiani con segno corretto, si sono ruotati i modi di tipo tempo, effettuando la consueta scelta di un tempo immaginario $`x_I^0ix_I^d`$ per $`I0`$. Per mantenere la corretta funzione $`\delta `$, si è poi contro-ruotato il modo zero di tipo tempo, $`x_0^0`$, producendo il fattore moltiplicativo $`i`$. La funzione primata
$$G^{}(\sigma ,\sigma ^{})=\underset{I0}{}\frac{2\pi \alpha ^{}}{\omega _I^2}𝒳_I\left(\sigma \right)𝒳_I\left(\sigma ^{}\right),$$
(3.38)
dove la somma è su tutti i modi eccetto il modo nullo, è la funzione di Green, e soddisfa l’equazione differenziale
$`{\displaystyle \frac{1}{2\pi \alpha ^{}}}^2G(\sigma ,\sigma ^{})`$ $`=`$ $`{\displaystyle \underset{I0}{}}𝒳_I\left(\sigma \right)𝒳_I\left(\sigma ^{}\right)`$ (3.39)
$`=`$ $`g^{1/2}\delta ^2\left(\sigma \sigma ^{}\right)𝒳_0^2,`$
in virtù della proprietà di completezza delle funzioni $`𝒳_I`$. La funzione di Green ordinaria per una sorgente puntiforme, in cui compare solo la funzione delta, nel caso di stringa è corretta dalla presenza del termine $`𝒳_0^2`$. Questo può essere interpretato come un contributo di carica di background che neutralizza una carica localizzata, che come tale non può essere presente su una superficie compatta.
#### 3.3.2 Ampiezze sulla sfera
Specializziamo la soluzione trovata alla sfera. La soluzione dell’equazione differenziale (3.39) sulla sfera è
$$G^{}(\sigma _1,\sigma _2)=\frac{\alpha ^{}}{2}\mathrm{ln}\left|z_1z_2\right|^2+f(z_1,\overline{z}_1)+f(z_2,\overline{z}_2),$$
(3.40)
dove la funzione $`f(z,\overline{z})`$ che, come si vedrà, non compare nelle ampiezze, è della forma
$$f(z,\overline{z})=\frac{\alpha ^{}𝒳_0^2}{4}d^2z^{}e^{2\omega (z^{},\overline{z}^{})}\mathrm{ln}\left|zz^{}\right|^2+k,$$
(3.41)
dove $`k`$ è una costante.
Consideriamo l’ampiezza d’urto per $`n`$ tachioni, inserendo i vertici nell’integrale funzionale,
$$A_{S_2}^n=\left[e^{ik_1X\left(\sigma _1\right)}\right]_r\left[e^{ik_2X\left(\sigma _2\right)}\right]_r\mathrm{}\left[e^{ik_nX\left(\sigma _n\right)}\right]_r_{S_2}.$$
(3.42)
Questo corrisponde a
$$J\left(\sigma \right)=\underset{i=1}{\overset{n}{}}k_i\delta ^2\left(\sigma \sigma _i\right),$$
(3.43)
e sostituendo nell’ampiezza (3.36) si ottiene
$`A_{S_2}^n(k,\sigma )`$ $`=`$ $`i𝒳_0^d\left(\stackrel{}{det}{\displaystyle \frac{^2}{4\pi ^2\alpha ^{}}}\right)_{S_2}^{d/2}\left(2\pi \right)^d\delta ^d\left({\displaystyle \underset{i}{}}k_i\right)`$ (3.44)
$`\times `$ $`\mathrm{exp}\left({\displaystyle \underset{\stackrel{i,j=1}{i<j}}{\overset{n}{}}}k_ik_jG^{}(\sigma _i,\sigma _j){\displaystyle \frac{1}{2}}{\displaystyle \underset{i=1}{\overset{n}{}}}k_i^2G_r^{}(\sigma _i,\sigma _i)\right).`$
È importante notare come la funzione $`\delta `$ ora garantisca opportunamente la conservazione dell’impulso. Il determinante può essere regolarizzato, e la funzione di Green è stata rinormalizzata, introducendo un contro-termine
$$G_r^{}(\sigma ,\sigma ^{})=G^{}(\sigma ,\sigma ^{})\frac{\alpha ^{}}{2}\mathrm{ln}d^2(\sigma ,\sigma ^{}),$$
(3.45)
dove $`d^2(\sigma ,\sigma ^{})`$ è la distanza fra punti lungo le geodetiche. Per brevi distanze, in metrica conforme, si ha $`d^2(\sigma ,\sigma ^{})\left(\sigma \sigma ^{}\right)^2e^{2\omega \left(\sigma \right)}`$, quindi si ha per la funzione rinormalizzata,
$$G_r^{}(\sigma ,\sigma )=2f(z,\overline{z})+\alpha ^{}\omega (z,\overline{z}).$$
(3.46)
Sostituendo nell’ampiezza sulla sfera si ha
$$A_{S_2}^n(k,\sigma )=iC_{S_2}^X\left(2\pi \right)^d\delta ^d\left(\underset{i}{}k_i\right)\mathrm{exp}\left(\frac{\alpha ^{}}{2}\underset{i}{}k_i^2\omega \left(\sigma _i\right)\right)\underset{\stackrel{i,j=1}{i<j}}{\overset{n}{}}\left|z_{ij}\right|^{\alpha ^{}k_ik_j},$$
(3.47)
in cui, come atteso, non compare la funzione $`f`$ e la costante è
$$C_{S_2}^X=𝒳_0^d\left(\stackrel{}{det}\frac{^2}{4\pi ^2\alpha ^{}}\right)_{S_2}^{d/2}.$$
(3.48)
L’ampiezza d’urto di stati eccitati di stringa sarà del tipo
$$\underset{i=1}{\overset{n}{}}\left[e^{ik_iX(z_i,\overline{z}_i)}\right]_r\underset{j=1}{\overset{p}{}}X_j^\mu \left(z_j^{}\right)\underset{k=1}{\overset{q}{}}\overline{}X^{\nu _k}\left(\overline{z}_k^{\prime \prime }\right)_{S_2},$$
(3.49)
dal momento che i vertici sono esponenziali per derivate di $`X^\mu `$. L’ampiezza si calcola sommando su tutte le contrazioni, dove ogni $`X`$ e $`\overline{}X`$ deve essere contratta sia con gli esponenziali che con le altre derivate di $`X`$. Il risultato finale è
$`iC_{S_2}^X\left(2\pi \right)^d\delta ^d\left({\displaystyle \underset{i}{}}k_i\right)\mathrm{exp}\left({\displaystyle \frac{\alpha ^{}}{2}}{\displaystyle \underset{i}{}}k_i^2\omega \left(\sigma _i\right)\right){\displaystyle \underset{\stackrel{i,j=1}{i<j}}{\overset{n}{}}}\left|z_{ij}\right|^{\alpha ^{}k_ik_j}`$
$`\times {\displaystyle \underset{j=1}{\overset{p}{}}}\left[v^{\mu _j}\left(z_j^{}\right)+q^{\mu _j}\left(z_j^{}\right)\right]{\displaystyle \underset{k=1}{\overset{q}{}}}\left[\stackrel{~}{v}^{\nu _k}\left(\overline{z}_k^{\prime \prime }\right)+\stackrel{~}{q}^{\nu _k}\left(\overline{z}_k^{\prime \prime }\right)\right]_{S_2},`$ (3.50)
dove
$$v^\mu =i\frac{\alpha ^{}}{2}\underset{i=1}{\overset{n}{}}\frac{k_i^\mu }{zz_i},\stackrel{~}{v}^\mu \left(\overline{z}\right)=i\frac{\alpha ^{}}{2}\underset{i=1}{\overset{n}{}}\frac{k_i^\mu }{\overline{z}\overline{z}_i},$$
(3.51)
provengono dalle contrazioni con gli esponenziali.
Il contributo dei ghost delle ampiezze, come abbiamo visto, è esprimibile in termini del determinante (3.18). Ricordando la struttura delle ampiezze, il numero di CKV della sfera impone che l’unico valore di aspettazione indipendente dei campi ghost, che compare nelle ampiezze ad albero, sia
$$c\left(z_1\right)c\left(z_2\right)c\left(z_3\right)\stackrel{~}{c}\left(\overline{z}_4\right)\stackrel{~}{c}\left(\overline{z}_5\right)\stackrel{~}{c}\left(\overline{z}_6\right)_{S_2},$$
(3.52)
I sei vettori di Killing della sfera compongono una base complessa non ortonormale
$`(1,0),(z,0),(z^2,0),(0,1),(0,\overline{z}),(0,\overline{z}^2).`$ (3.53)
In questa base il determinante degli zero modi del campi $`c`$ fattorizza in due blocchi
$$detC_{0j}^a=C_{S_2}^gdet\left|\begin{array}{ccc}1& 1& 1\\ z_1& z_2& z_3\\ z_1^2& z_2^2& z_3^2\end{array}\right|det\left|\begin{array}{ccc}1& 1& 1\\ \overline{z}_1& \overline{z}_2& \overline{z}_3\\ \overline{z}_1^2& \overline{z}_2^2& \overline{z}_3^2\end{array}\right|=C_{S_2}^gz_{12}z_{13}z_{23}\overline{z}_{45}\overline{z}_{46}\overline{z}_{56},$$
(3.54)
dove la notazione $`z_{ij}`$ sta per $`z_iz_j`$ e la costante $`C_{S_2}^g`$ è uno Jacobiano finito dimensionale, dovuto alla scelta di una base non ortonormale.
#### 3.3.3 Ampiezze sul disco
Per scrivere l’ampiezza sul disco, partendo dal caso della sfera, è necessario introdurre una riflessione. L’identificazione $`z^{}=\overline{z}`$ porta a restringere $`z`$ al semipiano superiore del piano complesso, e introducendo al contempo nella (3.39) una carica immagine, che porta a una funzione di Green del tipo
$$G^{}(\sigma _1,\sigma _2)=\frac{\alpha ^{}}{2}\mathrm{ln}\left|z_1z_2\right|^2\frac{\alpha ^{}}{2}\mathrm{ln}\left|z_1\overline{z}_2\right|^2.$$
(3.55)
L’ampiezza d’urto fra tachioni diventa
$`{\displaystyle \underset{i=1}{\overset{n}{}}}\left[e^{ik_iX(z_i,\overline{z}_i)}\right]_r_{D_2}`$ $`=`$ $`iC_{D_2}^X\left(2\pi \right)^d\delta ^d\left({\displaystyle \underset{i}{}}k_i\right){\displaystyle \underset{i=1}{\overset{n}{}}}\left|z_i\overline{z}_i\right|^{\alpha ^{}k_i^2/2}`$ (3.56)
$`\times `$ $`{\displaystyle \underset{\stackrel{i,j=1}{i<j}}{\overset{n}{}}}\left|z_iz_j\right|^{\alpha ^{}k_ik_j}\left|z_i\overline{z}_j\right|^{\alpha ^{}k_ik_j},`$
mentre per ampiezze d’urto fra stati eccitati di stringa occorre nuovamente sommare su tutte le contrazioni.
Per vertici sul bordo del disco, i due termini della funzione di Green a punti uguali sono entrambi divergenti, e sottraendo il consueto controtermine del primo termine, la funzione resta divergente. Occorre pertanto definire un ordinamento normale di bordo raddoppiando il controtermine,
$$\stackrel{}{}X^\mu \left(y_1\right)X^\nu \left(y_2\right)\stackrel{}{}=X^\mu \left(y_1\right)X^\nu \left(y_2\right)+2\alpha ^{}\eta ^{\mu \nu }\mathrm{ln}|y_1y_2|,$$
(3.57)
dove i punti $`y`$ sono sull’asse reale.
Per calcolare le ampiezze dei ghost nel caso del disco, il modo più semplice è utilizzare nuovamente la “duplicazione”. In analogia con il caso precedente, si considera pertanto il caso della sfera, con le opportune restrizioni sui campi
$$\stackrel{~}{b}\left(\overline{z}\right)=b\left(z^{}\right),\stackrel{~}{c}\left(\overline{z}\right)=c\left(z^{}\right),z^{}=\overline{z},Im\left(z\right)>0.$$
(3.58)
e il risultato è
$$c\left(z_1\right)c\left(z_2\right)c\left(z_2\right)_{D_2}=C_{D_2}^gz_{12}z_{13}z_{23}.$$
(3.59)
#### 3.3.4 Ampiezze sul piano proiettivo
Con il metodo delle cariche immagine, la funzione di Green risulta essere in questo caso
$$G^{}(\sigma _1,\sigma _2)=\frac{\alpha ^{}}{2}\mathrm{ln}\left|z_1z_2\right|^2\frac{\alpha ^{}}{2}\mathrm{ln}\left|1+z_1\overline{z}_2\right|^2,$$
(3.60)
e l’ampiezza tachionica è quindi
$`{\displaystyle \underset{i=1}{\overset{n}{}}}\left[e^{ik_iX(z_i,\overline{z}_i)}\right]_r_{D_2}`$ $`=`$ $`iC_{RP_2}^X\left(2\pi \right)^d\delta ^d\left({\displaystyle \underset{i}{}}k_i\right){\displaystyle \underset{i=1}{\overset{n}{}}}\left|1+z_i\overline{z}_i\right|^{\alpha ^{}k_i^2/2}`$ (3.61)
$`\times `$ $`{\displaystyle \underset{\stackrel{i,j=1}{i<j}}{\overset{n}{}}}\left|z_iz_j\right|^{\alpha ^{}k_ik_j}\left|1+z_i\overline{z}_j\right|^{\alpha ^{}k_ik_j}.`$
Anche le caso del piano proiettivo si può utilizzare la stessa tecnica di “duplicazione” per calcolare la parte di ampiezza dovuta ai campi di ghost. L’involuzione $`z^{}=\overline{z}^1`$ porta a richiedere
$$\stackrel{~}{b}\left(\overline{z}\right)=\left(\frac{z^{}}{\overline{z}}\right)^2b\left(z^{}\right)=z^4b\left(z^{}\right),\stackrel{~}{c}\left(\overline{z}\right)=\left(\frac{z^{}}{\overline{z}}\right)^1c\left(z^{}\right)=z^2c\left(z^{}\right),$$
(3.62)
e l’ampiezza si trova essere
$$c\left(z_1\right)c\left(z_2\right)c\left(z_2\right)_{RP_2}=C_{RP_2}^gz_{12}z_{13}z_{23}.$$
(3.63)
### 3.4 Ampiezze per stringhe chiuse e aperte
#### 3.4.1 Ampiezza di Veneziano
Studiamo, a titolo di esempio, l’ampiezza di interazione di quattro tachioni di stringhe aperte. L’ampiezza all’ordine più basso ha la topologia del disco, che possiamo rappresentare come il semipiano superiore, indicando le coordinare reali del bordo con $`y`$. Dopo aver fissato la metrica si possono utilizzare, come si è visto, tre dei CKV per fissare tre dei quattro vertici di interazione in posizioni $`y_1`$, $`y_2`$, $`y_3`$. Ogni coordinata di vertice di interazione fissata viene sostituita dall’inserzione di un campo di ghost $`c`$. Ogni vertice di interazione porta un fattore di accoppiamento di stringa $`g_o`$. Tenendo conto dei due possibili differenti ordinamenti dei vertici fissati, e prendendo il modulo del contributo dei ghost, l’ampiezza d’urto si scrive
$`S_{D_2}(k_1,k_2,k_3,k_4)`$ $`=`$ (3.64)
$`=`$ $`g_o^4e^\lambda {\displaystyle _{\mathrm{}}^{\mathrm{}}}dy_4{\displaystyle \underset{i=1}{\overset{3}{}}}\stackrel{}{}c^1\left(y_i\right)e^{ik_iX\left(y_i\right)}\stackrel{}{}\stackrel{}{}e^{ik_4X\left(y_4\right)}\stackrel{}{}+(k_2k_3)`$
$`=`$ $`ig_o^4C_{D_2}\left(2\pi \right)^{26}\delta ^{26}\left({\displaystyle \underset{i}{}}k_i\right)\left|y_{12}y_{13}y_{23}\right|{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑y_4{\displaystyle \underset{i<j}{}}\left|y_{ij}\right|^{2\alpha ^{}k_ik_j}`$
$`+(k_2k_3).`$
Convenzionalmente si sceglie $`y_1=0`$, $`y_2=1`$ e $`y_3\mathrm{}`$ e si adottano le variabili di Mandelstam
$$s=\left(k_1+k_2\right)^2,t=\left(k_1+k_3\right)^2,u=\left(k_1+k_4\right)^2,$$
(3.65)
che sono legate dalla conservazione del momento e dalla condizione di mass-shell
$$s+t+u=\underset{i}{}m_i=\frac{4}{\alpha ^{}}.$$
(3.66)
Utilizzando la relazione $`2\alpha ^{}k_ik_j=2+\alpha ^{}\left(k_i+k_j\right)^2`$, e introducendo le variabili di Mandelstam, l’ampiezza (3.64) diventa
$`S_{D_2}(k_1,k_2,k_3,k_4)`$ $`=`$ $`ig_o^4C_{D_2}\left(2\pi \right)^{26}\delta ^{26}\left({\displaystyle \underset{i}{}}k_i\right)`$
$`\times `$ $`\left[{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑y_4\left|y_4\right|^{\alpha ^{}u2}\left|1y_4\right|^{\alpha ^{}t2}+\left(ts\right)\right].`$
L’espressione trovata può essere riscritta in forma più elegante, mettendo in evidenza la simmetria nei diversi canali. L’integrale si divide in tre regioni: $`\mathrm{}<y_4<0`$, $`0<y_4<1`$, e $`0<y_4<\mathrm{}`$, in cui si hanno diversi ordinamenti per i vertici. Si possono utilizzare le trasformazioni (2.97) a parametri reali in modo tale da avere contributi solo nella seconda regione d’integrazione, in questo modo l’ampiezza diventa
$$S_{D_2}(k_1,k_2,k_3,k_4)=ig_o^4C_{D_2}\left(2\pi \right)^{26}\delta ^{26}\left(\underset{i}{}k_i\right)\left[I(s,t)+I(t,u)+I(u,s)\right],$$
(3.68)
con
$$I(s,t)=_0^1𝑑yy^{\alpha ^{}s2}\left(1y\right)^{\alpha ^{}t2}.$$
(3.69)
L’ampiezza (3.68) può essere riscritta utilizzando la funzione beta di Eulero
$$B(a,b)=_0^1𝑑yy^{a1}\left(1y\right)^{b1},$$
(3.70)
che può essere espressa in termini di funzioni $`\mathrm{\Gamma }`$. Definendo infatti una nuova variabile $`y=v/w`$, a $`w`$ fissato, si ha
$$w^{a+b1}B(a,b)=_0^w𝑑vv^{a1}\left(wv\right)^{b1},$$
(3.71)
e moltiplicando entrambi i lati dell’equazione per $`_0^{\mathrm{}}𝑑we^w`$, si trova
$$\mathrm{\Gamma }\left(a+b\right)B(a,b)=_0^{\mathrm{}}𝑑vv^{a1}e^v_0^{\mathrm{}}d\left(wv\right)\left(wv\right)^{b1}e^{\left(wv\right)}=\mathrm{\Gamma }\left(a\right)\mathrm{\Gamma }\left(b\right),$$
(3.72)
da cui segue immediatamente che
$$B(a,b)=\frac{\mathrm{\Gamma }\left(a\right)\mathrm{\Gamma }\left(b\right)}{\mathrm{\Gamma }\left(a+b\right)}.$$
(3.73)
Osservando che
$$I(s,t)=B(\alpha ^{}s1,\alpha ^{}t1),$$
(3.74)
e fissando la costante richiedendo l’unitarietà della matrice S, l’ampiezza può essere scritta nella forma
$`S_{D_2}(k_1,k_2,k_3,k_4)`$ $`=`$ $`ig_o^2C_{D_2}\left(2\pi \right)^{26}\delta ^{26}\left({\displaystyle \underset{i}{}}k_i\right)`$ (3.75)
$`\times \left[{\displaystyle \frac{\mathrm{\Gamma }\left(\alpha ^{}s1\right)\mathrm{\Gamma }\left(\alpha ^{}t1\right)}{\mathrm{\Gamma }\left(\alpha ^{}s\alpha ^{}t2\right)}}+\left(tu\right)+\left(su\right)\right],`$
dove
$$A(s,t)=\frac{\mathrm{\Gamma }\left(\alpha ^{}s1\right)\mathrm{\Gamma }\left(\alpha ^{}t1\right)}{\mathrm{\Gamma }\left(\alpha ^{}s\alpha ^{}t2\right)}$$
(3.76)
è la celebre *Ampiezza di Veneziano*, derivata originariamente per descrivere le interazioni forti .
L’ampiezza di Veneziano ha due proprietà notevoli: la simmetria manifesta sotto lo scambio di $`s`$ con $`t`$, detta *dualità planare*, e la proprietà che i residui nei poli in $`s`$ siano polinomi in $`t`$ e viceversa. Entrambe le proprietà sono caratteristiche importanti della Teoria delle Stringhe. In Teoria dei Campi, infatti, le ampiezze ad albero associate ai diagrammi di Feynman hanno individualmente solo un polo in uno dei canali.
La dualità planare dell’ampiezza di Veneziano riflette la deformabilità del disco, che consente di avvicinare a coppie i vertici di interazione, modificando il processo. Questo comporta che l’ampiezza completa (3.75) condensa, nel caso di stringa, diversi diagrammi d’urto di Feynman, che corrispondono ai diversi canali $`s`$, $`t`$ e $`u`$ in cui si può avere il processo d’interazione (vedi figura 3.1).
La funzione $`\mathrm{\Gamma }\left(x\right)`$ ha la proprietà di avere infiniti poli sul semiasse reale negativo, in corrispondenza degli interi. Nell’intorno di questi poli, per $`xn`$, si ha
$$\mathrm{\Gamma }\left(x\right)\stackrel{xn}{}\frac{()^n}{n!}\frac{1}{x+n},$$
(3.77)
mentre dal momento che $`\mathrm{\Gamma }\left(x+1\right)=x\mathrm{\Gamma }\left(x\right)`$, si trova
$$\frac{\mathrm{\Gamma }\left(y\right)}{\mathrm{\Gamma }\left(yn\right)}=\left(y1\right)\left(y2\right)\left(\mathrm{}\right)\left(yn\right).$$
(3.78)
La struttura dei residui dell’ampiezza sarà quindi del tipo
$$A(s,t)\stackrel{xn}{}\frac{()^n}{n!}\frac{\left(y1\right)\left(y2\right)\left(\mathrm{}\right)\left(yn\right)}{x+n}.$$
(3.79)
L’ampiezza di Veneziano, considerando i giusti argomenti delle funzioni gamma, ha quindi infiniti poli e, ad per esempio, per il canale $`s`$, in corrispondenza di $`\alpha ^{}s=1,0,1,2,\mathrm{}`$. L’ampiezza di Veneziano descrive quindi, un processo d’urto fra particelle scalari che scambiano tra loro infinite particelle con spin e massa crescente.
#### 3.4.2 Ampiezza di Shapiro-Virasoro
Il calcolo dell’ampiezza d’urto di stringhe chiuse procede in maniera simile al caso di stringhe aperte. Per quattro stringhe tachioniche si trova,
$$S_{S_2}(k_1,k_2,k_3,k_4)=g_c^4e^{2\lambda }_{}d^2z_4\underset{i=1}{\overset{3}{}}:\stackrel{~}{c}ce^{ik_iX}(z_i,\overline{z}_i)::e^{ik_4X}(z_4,\overline{z}_4):_{S_2},$$
(3.80)
dove l’integrazione avviene su tutto il piano complesso $``$. Calcolando i valori di aspettazione, e fissando i vertici di interazione nei punti $`z_1=0`$, $`z_2=1`$ e $`z_3=\mathrm{}`$, l’ampiezza si riduce a
$`S_{S_2}(k_1,k_2,k_3,k_4)`$ $`=`$ $`g_c^4C_{S_2}\left(2\pi \right)^{26}\delta ^{26}\left({\displaystyle \underset{i}{}}k_i\right)`$ (3.81)
$`\times {\displaystyle _{}}d^2z_4\left|z_4|^{\alpha ^{}u/24}\right|1z_4|^{\alpha ^{}t/24}.`$
In questo caso le variabili di Mandelstam sono soggette al vincolo $`s+t+u=16/\alpha ^{}`$. Questa ampiezza ha poli nella variabile $`u`$ per $`\mathrm{}<z_4<0`$, nella variabile $`t`$ per $`0<z_4<1`$ e in $`s`$ per $`1<z_4<\mathrm{}`$. I poli si hanno per i valori
$$\alpha ^{}s,\alpha ^{}t,\alpha ^{}u=4,0,4,8,\mathrm{},$$
(3.82)
che sono correttamente i valori dei quadrati delle masse degli stati di stringa chiusa.
Anche l’ampiezza di striga chiusa può essere espressa in temini di funzioni gamma,
$$S_{S_2}(k_1,k_2,k_3,k_4)=\frac{8\pi ig_c^2}{\alpha ^{}}\left(2\pi \right)^{26}\delta ^{26}\left(\underset{i}{}k_i\right)C(1\alpha ^{}t/4,1\alpha ^{}u/4),$$
(3.83)
dove
$$C(a,b)=2\pi \frac{\mathrm{\Gamma }\left(a\right)\mathrm{\Gamma }\left(b\right)\mathrm{\Gamma }\left(c\right)}{\mathrm{\Gamma }\left(a+b\right)\mathrm{\Gamma }\left(a+c\right)\mathrm{\Gamma }\left(b+c\right)},a+b+c=1.$$
(3.84)
e utilizzando gli argomenti corretti si ha
$$C\frac{\mathrm{\Gamma }\left(1\alpha ^{}t/4\right)\mathrm{\Gamma }\left(1\alpha ^{}u/4\right)\mathrm{\Gamma }\left(1\alpha ^{}s/4\right)}{\mathrm{\Gamma }\left(2\alpha ^{}t/4\alpha ^{}u/4\right)\mathrm{\Gamma }\left(2\alpha ^{}t/4\alpha ^{}s/4\right)\mathrm{\Gamma }\left(2\alpha ^{}u/4\alpha ^{}s/4\right)}.$$
(3.85)
Questa è l’*ampiezza di Shapiro-Virasoro*. In questa ampiezza la simmetria fra i canali $`s`$, $`t`$ e $`u`$ è completa.
#### 3.4.3 Fattori di Chan-Paton e interazioni di gauge
È possibile introdurre una generalizzazione particolarmente interessante delle teorie di stringa aperta, che ne rispetta le simmetrie, ammettendo la presenza agli estremi della stringa di gradi di libertà aggiuntivi non dinamici, noti come *cariche di Chan-Paton* . Uno stato di stringa aperto può così essere scritto nella forma
$$|N;k;ij,$$
(3.86)
dove gli indici $`i`$ e $`j`$ associati agli estremi della stringa vanno da $`1`$ a $`n`$. Dal momento che le cariche di Chan-Paton vengono introdotte senza termini dinamici nella Lagrangiana, tutte le simmetrie della teoria sono preservate. È possibile definire una base degli stati del tipo
$$|N;k;a=\underset{i,j=1}{\overset{n}{}}\lambda _{ij}^a|N;k;ij$$
(3.87)
con $`\lambda _{ij}^a`$ una opportuna rappresentazione di un gruppo di simmetria. Questo è un modo molto naturale di introdurre gruppi di simmetria, anche non abeliani, in Teoria delle Stringhe.
Le ampiezze di stringa aperta per uno stesso tipo bosoni esterni possono essere generalizzate definendo ampiezze “rivestite”
$$A(1,\mathrm{},n)\mathrm{tr}\left(\lambda _1^a\mathrm{}\lambda _2^a\right),$$
(3.88)
dove $`A(1,\mathrm{},n)`$ indica l’ampiezza calcolata senza cosiderare i gradi di libertà aggiuntivi. Questa è una scelta abbastanza naturale dal momento che le cariche di Chan-Paton non evolvono, e quindi l’interazione fra gli estremi delle stringhe deve avvenire fra estremi che si trovino nello stesso stato di carica. A livello ad albero, nell’ampiezza di disco, ciscun estremo destro di stringa deve avere alla sua sinistra l’estremo di un’altra stringa nello stesso stato.
Le ampiezze modificate devono inoltre rispettare i vincoli di unitarietà della teoria. La risoluzione di questi vincoli ha dimostrato la possibilità di introdurre in teoria di stringhe tutti i gruppi classici, $`SO\left(N\right)`$, $`Sp\left(N\right)`$, $`U\left(N\right)`$. I due estremi delle stringhe aperte devono avere valori nelle rappresentazioni fondamentali dei gruppi classici.
### 3.5 Ampiezze ad un loop
I primi diagrammi a loop che entrano nell’espansione perturbativa della Teoria delle Stringhe sono associati a superfici di Riemann con numero di Eulero nullo. In particolare in questa sezione discutiamo il diagramma di toro che è associato, nelle teorie di stringa chiusa, al primo ordine perturbativo. Altre ampiezze, presenti in teorie aperte e non orientate, saranno studiate in dettaglio, nel caso in cui non si abbiano inserzioni di vertici, nel capitolo 4.
Nel caso del toro la funzione di Green (3.39) deve essere periodica in entrambe le direzioni definite sulla superficie. Ci si aspetta anche che sia la somma di un contributo olomorfo e di uno antiolomorfo per analogia con il caso di genere zero e per le proprietà del laplaciano. Le funzioni theta di Jacobi (si veda l’Appendice A) rispondono bene a queste proprietà, e un buon candidato è
$$G^{}(w,\overline{w};w^{},\overline{w}^{})\frac{\alpha ^{}}{2}\mathrm{ln}|\vartheta _1\left(\frac{ww^{}}{2\pi }\right|\tau )|^2,$$
(3.89)
dove $`\vartheta _1`$ è una funzione theta di Jacobi che tende a zero linearmente quando il suo primo argomento tende all’origine. Il logaritmo è somma di un contributo olomorfo e antiolomorfo, è quindi annullato da l’azione di $`\overline{}`$. La funzione proposta non è però doppiamente periodica, a causa delle proprieta di trasformazione delle funzioni $`\vartheta `$: sotto una trasformazione $`ww+2\pi \tau `$, si genera infatti un termine $`\alpha ^{}\left[Im\left(ww^{}\right)+\pi \tau _2\right]`$. Inoltre occorre introdurre, come si è già avuto modo di notare, anche un termine di carica di background. Tendendo conto di queste due osservazioni, la funzione di Green si trova essere
$$G^{}(w,\overline{w};w^{},\overline{w}^{})=\frac{\alpha ^{}}{2}\mathrm{ln}|\vartheta _1\left(\frac{ww^{}}{2\pi }\right|\tau )|^2+\alpha ^{}\frac{\left[Im\left(ww^{}\right)\right]^2}{4\pi \tau _2}+k(\tau ,\tau ),$$
(3.90)
dove la funzione $`k`$ si determina per ortogonalità rispetto a $`𝒳_0`$, ma come nel caso della sfera non contribuisce nelle ampiezze. Il valore di aspettazione per un prodotto di operatori di vertici si trova come nel caso della sfera, e risulta essere
$`{\displaystyle \underset{i=1}{\overset{n}{}}}:e^{ik_iX(z_i,\overline{z}_i)}:_{T_2}`$ $`=`$ $`iC_{T_2}^X\left(\tau \right)\left(2\pi \right)^d\delta ^d\left({\displaystyle \underset{i}{}}k_i\right)`$ (3.91)
$`\times `$ $`{\displaystyle \underset{i<j}{}}|{\displaystyle \frac{2\pi }{_\nu \vartheta _1\left(0|\tau \right)}}\vartheta _1\left({\displaystyle \frac{w_{ij}}{2\pi }}\right|\tau )\mathrm{exp}[{\displaystyle \frac{\left(Imw_{ij}\right)^2}{4\pi \tau _2}}]|^{\alpha ^{}k_ik_j}.`$
L’ampiezza di Toro può essere ricavata dall’espressione (3.36), tenendo conto della presenza di due moduli reali e di due vettori di Killing conformi. La sua forma generale è
$$S_{T^2}(1;2;\mathrm{};n)=\frac{1}{2}_{}𝑑\tau 𝑑\overline{\tau }B\stackrel{~}{B}c\stackrel{~}{c}𝒱_1(w_1,\overline{w}_2)\underset{i=2}{\overset{n}{}}𝑑w_i𝑑\overline{w}_i𝒱_i(w_i,\overline{w}_i)_{T_2},$$
(3.92)
dove si sono utilizzati i CKV per fissare uno dei vertici di interazione. L’integrazione avviene sulla regione fondamentale dello spazio dei moduli, mentre il fattore moltiplicativo $`1/2`$ tiene conto dell’ulteriore simmetria discreta $`_2`$ che, come si è visto, caratterizza il toro.
L’inserzione dei campi di ghost è in questo caso
$$B=\frac{1}{4\pi }(b,_\tau g)=\frac{1}{2\pi }d^2wb_{ww}\left(w\right)_\tau g_{\overline{w}\overline{w}}.$$
(3.93)
Se consideriamo una variazione della metrica $`\delta g_{ww}=ϵ^{}`$, la nuova metrica è
$`ds^2`$ $`=`$ $`dwd\overline{w}+ϵ^{}dw^2+ϵd\overline{w}^2`$ (3.94)
$`=`$ $`\left(1+ϵ^{}+ϵ\right)d\left[w+ϵ\left(\overline{w}w\right)\right]d\left[\overline{w}+ϵ^{}\left(w\overline{w}\right)\right]+O\left(ϵ^2\right),`$
che risulta equivalente (a meno di una trasformazione di Weyl) ad una metrica della forma $`dw^{}d\overline{w}^{}`$ con $`w^{}=w+ϵ\left(\overline{w}w\right)`$, di periodicità
$$w^{}w^{}+2\pi w^{}+2\pi \left(\tau 2i\tau _2ϵ\right).$$
(3.95)
Quindi la variazione della metrica equivale ad una trasformazione del modulo
$$\delta \tau =2i\tau _2ϵ.$$
(3.96)
Tornando al calcolo delle inserzioni, il risultato (3.96) permette di calcolare esplicitamente la derivata della metrica rispetto a $`\tau `$, e quindi
$$B=\frac{i}{4\pi \tau _2}d^2wb_{ww}\left(w\right)=2\pi ib_{ww}\left(0\right).$$
(3.97)
L’indipendenza dalla posizione del campo di ghost è dovuta al fatto che sul toro i differenziali quadratici sono costanti, mentre l’integrale d’area è semplicemente $`2𝑑\sigma ^1𝑑\sigma ^2=2\left(2\pi \right)^2\tau _2`$.
Si può ulteriormente manipolare l’espressione delle ampiezze per porre tutti i vertici sullo stesso piano. Dal momento che i $`CKV`$ sono costanti, il valore di aspettazione dei campi $`c`$, è indipendente dalla posizione. Si può quindi fissare la posizione dei campi di ghost in maniera arbitraria, e l’operatore di vertice può essere reso libero di muoversi sulla superficie del toro, ricordandosi però di mediare su tutte le sue possibili traslazioni. Per farlo si introduce nell’integrale funzionale l’integrale
$$\frac{dwd\overline{w}}{2\left(2\pi \right)^2\tau _2},$$
(3.98)
dove il denominatore è opportunamente l’area del toro che corrisponde al volume dei CKV. Raccogliendo i risultati trovati, l’ampiezza diventa
$$S_{T^2}(1;2;\mathrm{};n)=_{}\frac{d\tau d\overline{\tau }}{4\tau _2}b\left(0\right)\stackrel{~}{b}\left(0\right)c\left(0\right)\stackrel{~}{c}\left(0\right)\underset{i=1}{\overset{n}{}}𝑑w_i𝑑\overline{w}_i𝒱_i(w_i,\overline{w}_i)_{T_2},$$
(3.99)
e in particolare l’ampiezza di vuoto è
$$Z_{T^2}=_{}\frac{d\tau d\overline{\tau }}{4\tau _2}b\left(0\right)\stackrel{~}{b}\left(0\right)c\left(0\right)\stackrel{~}{c}\left(0\right)_{T_2},$$
(3.100)
che calcoleremo esplicitamente nel formalismo del cono di luce e che, come si vedrà, contiene molte informazioni importanti sulla teoria.
## Capitolo 4 Funzioni di partizione
### 4.1 Ampiezze di vuoto ad un loop
In Teoria dei campi le ampiezze di vuoto ad un loop sono completamente determinate dallo spettro della teoria e non contengono molte informazioni a parte la loro relazione con la costante cosmologica. Al contrario in Teoria delle Stringhe le ampiezze di vuoto permettono di studiare lo spettro perturbativo della teoria libera e di estrarre condizioni di consistenza che la rendano priva di divergenze e di anomalie.
Per calcolare le ampiezze ad un loop per stringhe chiuse e aperte è utile partire dalla Teoria dei Campi. Iniziamo dal caso più semplice, calcolando l’energia di vuoto per un campo scalare massivo in D-dimensioni
$$S_{\left(E\right)}=d^Dx\frac{1}{2}\left(_\mu \varphi ^\mu \varphi +M^2\varphi ^2\right).$$
(4.1)
L’integrale funzionale della teoria euclidea, ottenuto dopo una rotazione di Wick, è
$$Z\left[J\right]=\left[𝒟\varphi \right]e^{S_{\left(E\right)}J\varphi }=e^{W\left[J\right]},$$
(4.2)
dove W\[J\] è il funzionale generatore delle funzioni di Green connesse. L’*azione effettiva* è definita da una trasformazione di Legendre:
$$\mathrm{\Gamma }\left[\overline{\varphi }\right]=W\left[J\right]+d^DxJ\overline{\varphi },$$
(4.3)
e $`\mathrm{\Gamma }\left[\overline{\varphi }\right]`$ è il funzionale generatore delle funzioni irriducibili ad una particella. Nel limite classico $`\mathrm{}0`$ si trova, sostituendo l’espressione (4.3) in (4.2),
$$\mathrm{\Gamma }\left[\overline{\varphi }\right]=S\left[\overline{\varphi }\right]+O\left(\mathrm{}\right).$$
(4.4)
Dalla (4.3) si può calcolare la derivata di $`\mathrm{\Gamma }\left[\overline{\varphi }\right]`$ a J fissato
$$\frac{\delta \mathrm{\Gamma }\left[\overline{\varphi }\right]}{\delta \overline{\varphi }}=J,$$
(4.5)
e quindi per $`J=0`$ si ottengono i punti estremali dell’azione effettiva che possono essere interpretati come energie dei vuoti corrispondenti. Nel caso della teoria libera (4.1) si ha quindi per l’energia di vuoto l’espressione
$$e^\mathrm{\Gamma }=\left[𝒟\varphi \right]e^{S_{\left(E\right)}}\left[\mathrm{det}\left(\mathrm{}+M^2\right)\right]^{\frac{1}{2}},$$
(4.6)
che si ottiene dall’integrale gaussiano generalizzato, e quindi
$$\mathrm{\Gamma }=\frac{1}{2}\mathrm{tr}\left[\mathrm{ln}\left(\mathrm{}+M^2\right)\right].$$
(4.7)
Per estrarre la dipendenza dalla massa M è utile l’identità
$$\mathrm{tr}\left(\mathrm{ln}A\right)=_ϵ^{\mathrm{}}\frac{dt}{t}\mathrm{tr}\left(e^{tA}\right),$$
(4.8)
dove $`ϵ`$ è un cutoff ultravioletto e $`t`$ è un parametro di Schwinger. La traccia è facilmente calcolabile usando una base che diagonalizzi l’operatore cinetico, ovvero un set completo di autostati dell’impulso
$$\mathrm{\Gamma }=\frac{V}{2}_ϵ^{\mathrm{}}\frac{dt}{t}\frac{d^Dp}{\left(2\pi \right)^D}e^{tp^2}e^{tM^2},$$
(4.9)
dove V è il volume spazio-temporale. Integrando sugli impulsi si trova quindi l’energia di vuoto ad un loop per un grado di libertà bosonico:
$$\mathrm{\Gamma }=\frac{V}{2\left(4\pi \right)^{\frac{D}{2}}}_ϵ^{\mathrm{}}\frac{dt}{t^{\frac{D}{2}+1}}e^{tM^2}.$$
(4.10)
Ripetendo il calcolo per un fermione di Dirac, le regole per l’integrazione di variabili anticommutanti introducono un segno $`()`$, e ricordando che in $`D`$ dimensioni gli spinori di Dirac hanno $`2^{\frac{D}{2}}`$ gradi di libertà si ottiene
$$\mathrm{\Gamma }=\frac{V2^{\frac{D}{2}}}{2\left(4\pi \right)^{\frac{D}{2}}}_ϵ^{\mathrm{}}\frac{dt}{t^{\frac{D}{2}+1}}e^{tM^2}.$$
(4.11)
Per poter usare le formule trovate per calcolare l’energia di vuoto ad un loop in teoria delle stringhe occorre un’opportuna generalizzazione. Si è visto che $`\mathrm{\Gamma }`$ dipende solo dalle masse dei modi fisici che sono presenti nel loop di vuoto, e quindi nel caso in cui siano presenti più particelle (o più eccitazioni di stringa), si avrà una traccia sulle masse. Si deve infine tenere conto del segno e della molteplicità dei contributi fermionici. Queste osservazioni possono essere raccolte nell’espressione generale
$$\mathrm{\Gamma }=\frac{V}{2\left(4\pi \right)^{\frac{D}{2}}}_ϵ^{\mathrm{}}\frac{dt}{t^{\frac{D}{2}+1}}\mathrm{Str}\left(e^{tM^2}\right),$$
(4.12)
dove la supertraccia $`\mathrm{Str}`$ è
$$\mathrm{Str}=\underset{bosoni}{}\underset{fermioni}{}.$$
(4.13)
Prima di applicare questa formula al calcolo delle ampiezze di vuoto di stringa torniamo per un momento alle superfici di Riemann di genere $`g=1`$. Il toro rappresenta una stringa chiusa orientata che si propaga in un loop, e il suo modulo $`\tau `$ può essere interpretato fisicamente come il tempo proprio della stringa nel loop. L’invarianza modulare indica che c’è un’infinità di scelte equivalenti nella scelta del tempo sul world-sheet. Nella teoria questa invarianza è peculiare, dal momento che introduce un cut-off naturale ultravioletto, come si è visto definendo la regione fondamentale nel $`\tau `$-piano. Al contrario, nelle altre superfici di genere $`g=1`$ sono possibili due scelte del tempo proprio legate sostanzialmente dalle trasformazioni $`S`$, che danno luogo a diverse interpretazioni dei diagrammi. Non c’è quindi alcuna simmetria che protegga dalle divergenze, che vanno cancellate imponendo condizioni di cancellazione.
La bottiglia Klein, come si è visto, può essere interpretata alternativamente come un diagramma di vuoto di stringa chiusa non orientata o come un diagramma di propagazione ad albero di stringa chiusa fra due crosscap. Le due rappresentazioni corrispondono a scelte differenti del tempo proprio, rispettivamente il lato verticale e quello orizzontale dei due poligoni fondamentali. Anche nell’anello sono possibili due differenti scelte del tempo sul world-sheet. Scegliendo la direzione verticale si interpreta l’anello come un diagramma di vuoto di una stringa aperta, mentre scegliendo il tempo sull’asse reale, si ha un diagramma di propagazione ad albero di una stringa chiusa fra due bordi.
La striscia di Möbius, scegliendo il tempo “verticale”, descrive una stringa aperta non orientata in un diagramma di vuoto. Al contrario, ridefinendo il poligono fondamentale, come si è visto, e, scegliendo un tempo proprio “orizzontale”, si ha la propagazione ad albero di una stringa chiusa fra un bordo e un crosscap. Solitamente si fa riferimento alle ampiezze di vuoto come *ampiezze nel canale diretto*, e alle stesse nella rappresentazione ad albero come *ampiezze nel canale trasverso*.
### 4.2 Funzioni di partizione della stringa bosonica
#### 4.2.1 Teoria chiusa orientata nel formalismo del cono di luce
Applichiamo l’espressione (4.12) per una teoria di sola stringa bosonica chiusa in dimensione critica $`D=26`$, il cui spettro di massa è dato da
$$M^2=\frac{2}{\alpha ^{}}\left[N^{}+\overline{N}^{}2\right],$$
(4.14)
con la “level matching condition”
$$N^{}\overline{N}^{}=0.$$
(4.15)
Sostituendo nella (4.12) con $`D=26`$ la (4.14) e imponendo il vincolo di level-matching introducendo una $`\delta `$-function, si ottiene
$$\mathrm{\Gamma }=\frac{V}{2\left(4\pi \right)^{13}}_{\frac{1}{2}}^{\frac{1}{2}}𝑑s_ϵ^{\mathrm{}}\frac{dt}{t^{14}}\mathrm{tr}\left(e^{\frac{2}{\alpha ^{}}\left[N^{}+\overline{N}^{}2\right]t}e^{2\pi i\left[N^{}\overline{N}^{}\right]s}\right),$$
(4.16)
che, definendo un parametro di Schwinger complesso $`\tau =s+i\frac{t}{\pi \alpha ^{}}`$ e definendo $`q=e^{2\pi i\tau }`$, $`\overline{q}=e^{2\pi i\overline{\tau }}`$, può essere scritta nella forma elegante
$$\mathrm{\Gamma }=\frac{V}{2\left(4\alpha ^{}\pi ^2\right)^{13}}_{\frac{1}{2}}^{\frac{1}{2}}𝑑\tau _1_ϵ^{\mathrm{}}\frac{d\tau _2}{\tau _2^{14}}\mathrm{tr}\left(q^{N^{}1}\overline{q}^{\overline{N}^{}1}\right).$$
(4.17)
Questa è l’ampiezza per una stringa chiusa che si propaghi in un loop, e l’integrazione sulla varibile complessa $`\tau `$ dovrebbe equivalere ad integrare sullo spazio dei moduli del toro, dal momento che il calcolo dell’integrale sui cammini, richiede che si sommi su tutte le metriche rappresentanti tutte le superfici topologicamente inequivalenti, e quindi nel nostro caso su tutti i possibili tori. Si è visto infatti che il toro è univocamente determinato, assegnata una metrica piatta, dal modulo complesso $`\tau `$ con parte immaginaria positiva. In realtà il calcolo seguito fin qui porta un problema di multiplo conteggio, dal momento che non tutti i punti del $`\tau `$-piano rappresentano tori inequivalenti. Occorre, come si è detto, restringersi nell’integrazione alla regione fondamentale $``$, e a meno di una costante di normalizzazione, l’ampiezza del toro è quindi
$$𝒯=_{}\frac{d^2\tau }{\tau _2^2}\frac{1}{\tau _2^{12}}\mathrm{tr}\left(q^{N^{}1}\overline{q}^{\overline{N}^{}1}\right).$$
(4.18)
Possiamo ora riprodurre il calcolo dell’ampiezza di vuoto per una stringa chiusa. Cosideriamo la cella fondamentale di un toro di modulo $`\tau `$ nel piano complesso e interpretiamo l’asse verticale come dimensione temporale e quello orizziontale come dimensione spaziale. Immaginiamo una stringa di lunghezza unitaria che al tempo $`t=0`$ giaccia sull’asse orizzontale, e dopo un tempo $`t=\tau _2`$ sia propagata raggiungendo il lato superiore della cella, traslando di $`x=Re\left(\tau \right)=\tau _1`$. Gli operatori per le traslazioni temporali e spaziali sono rispettivamente l’Hamiltoniana $`H=L_0^{}+\overline{L}_0^{}2`$ e l’impulso $`P=L_0^{}\overline{L}_0^{}`$. Si ritrova così il risultato precedente
$$Z=\mathrm{Tr}\left\{e^{2\pi \tau _2H}e^{2\pi i\tau _1P}\right\}=\mathrm{Tr}\left\{q^{L_0^{}1}\overline{q}^{\overline{L}_0^{}1}\right\},$$
(4.19)
dove Tr è una somma sulle variabile discrete e un’integrazione su quelle continue.
A questo punto calcoliamo esplicitamente l’ampiezza di toro (4.18), partendo dalla traccia (la somma sull’indice $`i`$ delle dimensioni è sottintesa)
$$\mathrm{tr}q^{N^{}1}=\frac{1}{q}\mathrm{tr}\left(q^{_{n=1}^{\mathrm{}}\alpha _n^i\alpha _n^i}\right),$$
(4.20)
che possiamo scrivere esplicitamente in una base di tutti gli stati di stringa della forma $`\alpha _n|0`$, $`\left(\alpha _n\right)^2|0`$, ecc. in cui si ha:
$$q^{\alpha _n\alpha _n}=\left(\begin{array}{ccccc}1& & & & \\ & q^n& & & \\ & & q^{2n}& & \\ & & & q^{3n}& \\ & & & & \mathrm{}\end{array}\right)$$
(4.21)
e quindi si ha
$$\mathrm{tr}\left(q^{_{n=1}^{\mathrm{}}\alpha _n^i\alpha _n^i}\right)=\underset{i=1}{\overset{24}{}}\underset{n=1}{\overset{\mathrm{}}{}}\underset{k}{}q^{nk}=\frac{1}{_{n=1}\left(1q^n\right)^{24}}.$$
(4.22)
Usando la funzione $`\eta `$ di Dedekind , definita come:
$$\eta \left(\tau \right)=q^{\frac{1}{24}}\underset{n=1}{\overset{\mathrm{}}{}}\left(1q^n\right),$$
(4.23)
possiamo riscrivere l’ampiezza di toro con i risultati ottenuti nella forma
$$𝒯_{bosonica}=\frac{d^2\tau }{\tau _2^2}\frac{1}{\tau _2^{12}\left(\eta \overline{\eta }\right)^{24}}.$$
(4.24)
$`𝒯`$ è invariante modulare, dal momento che la misura d’integrazione è invariante modulare e, usando le proprietà di trasformazione della funzione $`\eta `$,
$$T:\eta \left(\tau \right)\eta (\tau +1)=e^{\frac{i\pi }{12}}\eta \left(\tau \right),S:\eta \left(\tau \right)\eta (\frac{1}{\tau })=\sqrt{i\tau }\eta \left(\tau \right),$$
(4.25)
si verifica che anche il fattore $`\tau _2^{1/2}\left(\eta \overline{\eta }\right)`$ è invariante.
Si è detto che le ampiezze di vuoto scritte meritano il nome di funzioni di partizione, la ragione è che, espandendo l’integrando in potenze di $`q`$ e $`\overline{q}`$ e moltiplicando per un fattore $`4/\alpha ^{}`$ si ottengono i livelli delle eccitazioni in $`M^2`$, moltiplicati per coefficienti che forniscono le loro degenerazioni. La degenerazione è in questo caso il numero di partizioni del livello in numeri interi. Ad esempio, al livello tre si avrebbe degenerazione tre dal momento che si hanno le tre possibilità $`\alpha _3`$, $`\alpha _1\alpha _2`$ e $`\alpha _1\alpha _1\alpha _1`$. Per il toro, in $`D=26`$ nei livelli più bassi si ha
$$𝒯:\frac{1}{\left(\eta \overline{\eta }\right)^{24}}\frac{1}{q\overline{q}}[1+24(q+\overline{q})+\left(24\right)^2q\overline{q}+\mathrm{}],$$
(4.26)
dove il termine $`\left(q\overline{q}\right)^1`$ è la base della torre degli stati e indica che a livello zero si ha un tachione. Eliminando gli stati che non rispettano la condizione di level-matching, restano infiniti termini del tipo $`q^{w_1}\overline{q}^{w_2}`$ che indicano la presenza di campi massivi di “peso” $`(w_1,w_2)`$, con $`w_1=w_2`$. Ad esempio a livello uno gli stati di massa nulla hanno molteplicità $`24^2`$, e sono, come atteso, gli stati del multipletto ($`G_{\mu \nu }`$, $`B_{\mu \nu }`$, $`\mathrm{\Phi }`$).
#### 4.2.2 Discendenti aperti
La costruzione di teorie consistenti di stringhe chiuse e aperte non orientate a partire da teorie di sole stringhe chiuse è basata sulla proiezione di orientifold . L’idea è costruire dalle ampiezze di vuoto ad un loop di stringhe orientate, le ampiezze di stringhe non orientate per mezzo dell’operatore $`\mathrm{\Omega }`$ di world-sheet. Per una stringa chiusa si vede che lo spettro è invariante sotto parità del world-sheet, $`\sigma \sigma `$, dove l’azione naturale di $`\mathrm{\Omega }`$ scambia i modi sinistri e destri
$$\mathrm{\Omega }:\alpha _n^\mu \stackrel{~}{\alpha }_n^\mu .$$
(4.27)
Per ottenere stringhe non orientate si devono identificare i modi destri e sinistri della stringa. Questo corrisponde a proiettare lo spettro degli stati su uno dei due autospazi associati ai due autovalori di $`\mathrm{\Omega }`$, $`\pm 1`$. La proiezione deve essere consistente con le interazioni di stringa. Dal momento che l’urto di due stringhe chiuse antisimmetriche sotto lo scambio dei loro modi sinistri e destri darebbe luogo ad uno stato simmetrico, la sola opzione in questo caso è proiettare sugli stati simmetrici. Dal punto di vista del calcolo delle ampiezze di vuoto la proiezione si ottiene inserendo nella traccia sugli stati in (4.24) l’operatore
$$P=\frac{1+\mathrm{\Omega }}{2}.$$
(4.28)
L’azione del proiettore equivale a $`𝒯𝒯/2+𝒦`$, con $`𝒦`$ la funzione di partizione per la bottiglia di Klein
$$𝒦=\frac{1}{2}__𝒦\frac{d^2\tau }{\tau _2^{14}}\mathrm{tr}\left(q^{N^{}1}\overline{q}^{\overline{N}^{}1}\mathrm{\Omega }\right),$$
(4.29)
dove
$$\underset{L,R}{}L,Rq^{N^{}1}\overline{q}^{\overline{N}^{}1}\mathrm{\Omega }L,R=\underset{L,R}{}L,Rq^{N^{}1}\overline{q}^{\overline{N}^{}1}R,L,$$
(4.30)
e l’ortogonalità degli stati riduce la somma al solo sottospazio diagonale, identificando $`N^{}`$ con $`\overline{N}^{}`$,
$$\underset{L}{}LR\left(q\overline{q}\right)^{N^{}1}LR=\frac{1}{\eta \left(2i\tau _2\right)^{24}},$$
(4.31)
dal momento che $`q\overline{q}=e^{4\pi \tau _2}=e^{2\pi i\left(2i\tau _2\right)}`$. Si è trovata un’espressione che dipende naturalmente dal modulo del toro doppiamente ricoprente della bottiglia di Klein. Per fissare il dominio di integrazione $`_𝒦`$ occorre ricordare che, come si è osservato, non c’è invarianza modulare e, quindi, occorre integrare $`\tau _2`$ su tutto il dominio di definizione, che coincide con il semiasse immaginario positivo. In conclusione si ottiene
$$𝒦=\frac{1}{2}_0^{\mathrm{}}\frac{d^2\tau }{\tau _2^{14}}\frac{1}{\eta ^{24}\left(2i\tau _2\right)}.$$
(4.32)
L’espansione dell’ampiezza di vuoto della bottiglia di Klein (4.32) in funzione di $`q`$ e $`\overline{q}`$ è
$$𝒦:\frac{1}{\left(\eta ^{24}\right)\left(2i\tau _2\right)}\frac{1}{q\overline{q}}[1+24\left(q\overline{q}\right)+\mathrm{}],$$
(4.33)
e per conoscere lo spettro della teoria proiettata occorre sommare il contributo della bottiglia di Klein (4.33) a quello del toro (4.26) opportunamente dimezzato e ridotto ai termi che soddisfano la condizione di “level matching”:
$$\frac{1}{2}𝒯+𝒦:\frac{1}{q\overline{q}}[1+\frac{24\left(24+1\right)}{2}\left(q\overline{q}\right)+\mathrm{}].$$
(4.34)
Si è ottenuto il risultato atteso: sono scomparsi tra gli stati di massa nulla i gradi di libertà del tensore antisimmetrico $`B_{\mu \nu }`$.
Si è visto che sono possibili due diverse scelte naturali per il “tempo” sul world-sheet e che a queste corrispondono due diverse rappresentazioni diagrammatiche dell’ampiezza di bottiglia di Klein. Le due rappresentazioni sono legate da una trasformazione $`S`$, e comportano una ridefinzione della cella fondamentale. Per passare nel canale trasverso è sufficiente raddoppiare il modulo su cui si integra $`t=2\tau _2`$ generando un fattore moltipicativo $`2^{13}`$, ed effettuare quindi una trasformazione $`S`$ : $`t_21/\mathrm{}`$, ottendo
$$\stackrel{~}{𝒦}=\frac{2^{13}}{2}_0^{\mathrm{}}𝑑\mathrm{}\frac{1}{\eta ^{24}\left(i\mathrm{}\right)}.$$
(4.35)
L’ampiezza $`\stackrel{~}{𝒦}`$ è interpretata come l’ampiezza di propagazione di una stringa chiusa fra due crossacap in un tempo “orizzontale” $`\mathrm{}`$, ed è pertanto un diagramma “ad albero”. Come si è già detto, non c’è per la bottiglia di Klein invarianza sotto l’azione del gruppo modulare. Per ottenere una teoria che sia priva di divergenze, occorre per questo aggiungere alla teoria i settori di stringa aperta per poter poi imporre condizioni di cancellazione delle divergenze fissando i gradi di libertà delle cariche di Chan-Paton.
Per calcolare l’ampiezza di Anello nel canale diretto sostituiamo l’operatore di massa di stringa aperta $`M=\alpha ^{}\left(N^{}1\right)`$ nella (4.12)
$$𝒜=\frac{1}{2}_0^{\mathrm{}}\frac{d\tau _2}{\tau _2^{14}}\mathrm{tr}q^{\frac{1}{2}\left(N^{}1\right)},$$
(4.36)
dove $`q=e^{2\pi \tau _2}`$ e, la traccia è su stati del tipo $`\lambda _{ij}k;ij`$. Gli indici $`i`$ e $`j`$ sono riferiti alle cariche di Chan-Paton di moltiplicità N agli estremi della stringa aperta. Calcolando la traccia e tenendo conto della presenza di gradi di libertà aggiuntivi, si ottiene
$$𝒜=\frac{N^2}{2}_0^{\mathrm{}}\frac{d\tau _2}{\tau _2^{14}}\frac{1}{\eta ^{24}\left(\frac{1}{2}i\tau _2\right)}.$$
(4.37)
Anche in questo caso c’è una dipendenza naturale dal modulo del toro doppiamente ricoprente $`\tau =\frac{1}{2}i\tau _2`$. L’ampiezza del canale trasverso si ottiene utilizzando come varibile di integrazione il modulo del toro doppiamente ricoprente $`t=\tau _2/2`$, che porta un fattore $`2^{13}`$, e quindi effettuando una trasformazione $`S`$ : $`t_21/\mathrm{}`$, ottenendo infine
$$\stackrel{~}{𝒜}=\frac{2^{13}N^2}{2}_0^{\mathrm{}}𝑑\mathrm{}\frac{1}{\eta ^{24}\left(i\mathrm{}\right)}.$$
(4.38)
L’interpretazione di questa ampiezza al livello ad albero come propagazione di una stringa chiusa fra due bordi, porta ad associare N al coefficiente di riflessione sui bordi.
L’ultimo settore da studiare è quello di stringa aperta non orientata. Come nel caso di stringa chiusa, le ampiezze non orientate si ottengono inserendo nella traccia di anello il proiettore
$$P=\frac{\left(1+ϵ\mathrm{\Omega }\right)}{2},$$
(4.39)
e tenendo conto che nel caso di stringhe aperte la parità sul world-sheet ha un’azione leggermente differente dal caso chiuso, $`\mathrm{\Omega }:\sigma \pi \sigma `$, e quindi sugli oscillatori
$$\mathrm{\Omega }:\alpha _n^\mu (1)^n\alpha _n^\mu .$$
(4.40)
L’azione su uno stato generico deve essere scritta tenendo conto della struttura aggiuntiva introdotta con le cariche di Chan Paton. In generale ci si aspetta che l’azione di $`\mathrm{\Omega }`$ inverta le cariche agli estremi, ma si deve anche tenere conto della sua azione sui fattori di Chan Paton. Questa ulteriore libertà permette di lasciare indeterminato il segno $`ϵ`$ nella proiezione (4.39), che come si vedrà è fissato dalla simmetria delle matrici di Chan Paton e quindi dalla scelta del gruppo di gauge.
Consideriamo ora l’ampiezza del nastro di Möbius, che definiamo come
$$=\frac{ϵ}{2}_0^{\mathrm{}}\frac{d\tau _2}{\tau _2^{14}}\mathrm{tr}\left(q^{\frac{1}{2}\left(N^{}1\right)}\mathrm{\Omega }\right).$$
(4.41)
Nel calcolare la traccia si deve tener conto dell’operatore $`\mathrm{\Omega }`$, che restringe la somma sugli indici $`i`$, $`j`$ agli stati diagonali e quindi porta un fattore N, mentre sugli oscillatori, come visto nella (4.40), genera un segno $`\left(1\right)^n`$,
$$\mathrm{tr}\left(q^{\frac{1}{2}_{n=1}^{\mathrm{}}\alpha _n^i\alpha _n^i}\mathrm{\Omega }\right)=N\underset{i=1}{\overset{24}{}}\underset{n=1}{\overset{\mathrm{}}{}}\underset{k}{}()^{nk}q^{\frac{nk}{2}}=\frac{N}{_{n=1}\left(1()^nq^{\frac{n}{2}}\right)^{24}}.$$
(4.42)
Nell’espressione trovata si riconosce la dipendenza dal modulo del toro doppiamente ricoprente, osservando che $`q^{\frac{n}{2}}()^n=e^{2\pi in\left(1/2+i\tau _2/2\right)}`$. Definendo un funzione $`\widehat{\eta }`$ come
$$\widehat{\eta }\left(\frac{1}{2}i\tau _2+\frac{1}{2}\right)=\left(\sqrt{q}\right)^{\frac{1}{24}}\underset{n=1}{\overset{\mathrm{}}{}}\left(1()^nq^{\frac{n}{2}}\right),$$
(4.43)
che differisce dalla $`\eta `$ di Dedekind per una fase, l’ampiezza di nastro di Möbius è quindi
$$=\frac{ϵN}{2}_0^{\mathrm{}}\frac{d\tau _2}{\tau _2^{14}}\frac{1}{\widehat{\eta }^{24}\left(\frac{1}{2}i\tau _2+\frac{1}{2}\right)}.$$
(4.44)
In generale, come si vedrà anche in altri casi, il passaggio al canale trasverso per il nastro di Möbius richiede qualche accortezza, a causa del modulo complesso del toro doppiamente ricoprente. La trasformazione su $`\tau `$ che definisce il passaggio all’ampiezza ad albero è
$$P:\frac{1}{2}+i\frac{\tau _2}{2}\frac{1}{2}+i\frac{1}{2\tau _2},$$
(4.45)
che può essere ottenuta da una sequenza di trasformazioni S e T,
$$P=TST^2S,$$
(4.46)
Per $`\widehat{\eta }`$ la ridefinizione di fase introduce un’operazione di coniugio, e si ha
$$P=T^{1/2}ST^2ST^{1/2}.$$
(4.47)
Nel nostro caso per passare al canale trasverso ridefiniamo $`\tau _21/t`$, che corrisponde ad effetturare una trasformazione P. È semplice mostrare che
$$\widehat{\eta }\left(\frac{i}{2t}+\frac{1}{2}\right)=\sqrt{t}\widehat{\eta }\left(\frac{it}{2}+\frac{1}{2}\right)$$
(4.48)
e quindi
$$\stackrel{~}{}=\frac{ϵN}{2}_0^{\mathrm{}}𝑑t\frac{1}{\widehat{\eta }^{24}\left(\frac{1}{2}it+\frac{1}{2}\right)},$$
(4.49)
e infine, con $`\mathrm{}=t/2`$, si ha
$$\stackrel{~}{}=2\frac{ϵN}{2}_0^{\mathrm{}}𝑑\mathrm{}\frac{1}{\widehat{\eta }^{24}\left(i\mathrm{}+\frac{1}{2}\right)}.$$
(4.50)
A questo punto vediamo per il canale diretto, dove si hanno stringhe aperte, l’effetto dell’operatore di proiezione sviluppando in potenze di di $`\sqrt{q}`$ le ampiezze di anello e di Möbius
$$𝒜+:\frac{N^2ϵN}{2\sqrt{q}}+24\frac{N^2ϵN}{2}+\mathrm{}.$$
(4.51)
Quindi per $`ϵ=+1`$ si hanno $`N\left(N1\right)/2`$ vettori di massa nulla che completano la rappresentazione aggiunta di un gruppo $`SO\left(N\right)`$, mentre per $`ϵ=1`$ si hanno gli $`N\left(N+1\right)/2`$ vettori di massa nulla dell’aggiunta di un gruppo $`USp\left(N\right)`$.
La teoria costruita fin qui presenta divergenze ultraviolette nel canale diretto nel limite $`\tau _20`$: infatti fatta eccezione per il toro, nelle altre ampiezze di genere $`g=1`$ l’integrazione conivolge, come già evidenziato, tutto il semiasse positivo. L’insorgere delle divergenze può essere compreso meglio studiando il canale trasverso, dove esse compaiono nel limite infrarosso $`\mathrm{}\mathrm{}`$ e sono dovute alla propagazione di tachioni e di stati di stringa chiusa di massa nulla. I contributi degli stati generici di massa $`M_i`$ e con degenerazione $`c_i`$ che si propagano nel canale trasverso sono proporzionali a
$$\underset{i}{}c_i_0^{\mathrm{}}𝑑\mathrm{}e^{\mathrm{}M_i^2}=\underset{\mathrm{}\mathrm{}}{lim}\left(c_T\frac{e^{\left|M_T^2\right|\mathrm{}}}{M_T^2}\right)+\underset{i}{}c_i\frac{1}{p_i^2+M_i^2}|_{p_i^2=0}.$$
(4.52)
Nel limite $`\mathrm{}\mathrm{}`$ il tachione e gli stati a massa nulla danno origine ai contributi dominanti. La divergenza tachionica può essere regolata formalmente, e in ogni caso è assente nelle superstringhe. Al contrario, i contributi divergenti dovuti agli stati di massa nulla sono genericamente inevitabili e hanno implicazioni fisiche, che avremo modo di studiare, e sono dovuti al divergere dei loro propagatori nel limte di impulso nullo, come si è messo in evidenza nella (4.52). È possibile cancellare queste divergenze notando che nel limite $`\mathrm{}\mathrm{}`$ l’ampiezza trasversa fattorizza nella somma di contributi scrivibili come prodotti di due funzioni ad un punto di stati di massa nulla su un bordo o un crosscap e di un propagatore. Si può quindi eliminare le divergenze dovute ai poli nei propagatori, imponendo che la somma dei residui dei diversi contributi sia nulla, una condizione detta *condizione di tadpole* (vedi figura 4.1). Alla luce della discussione precedente dovrebbe essere chiaro che questo porta a fissare il segno $`ϵ`$ e la dimensione $`N`$, ovvero a fissare il gruppo di gauge di Chan-Paton.
Nel caso della stringa bosonica, raccogliendo i termini di massa nulla degli sviluppi delle ampiezze (4.35), (4.38) e (4.50), si ha
$$\stackrel{~}{𝒦}+\stackrel{~}{𝒜}+\stackrel{~}{}2^{13}+2^{13}N^22ϵN=2^{13}\left(Nϵ\mathrm{\hspace{0.17em}2}^{13}\right)^2,$$
(4.53)
le condizioni di tadpole fissa $`ϵ=+1`$ e $`N=2^{13}=8192`$, e quindi il gruppo di gauge $`SO\left(8192\right)`$ . Posticipiamo per il momento la discussione sul significato fisico della condizione di tadpole, che avremo modo di affrontare in maniera più generale, limitandoci a dire che essa equivale ad eliminare, nella teoria bosonica effettiva di basse energie, un potenziale per il dilatone $`\phi `$,
$$V\left(Nϵ2^{13}\right)d^{10}x\sqrt{g}e^\phi .$$
(4.54)
### 4.3 Funzioni di partizione di Superstringa
#### 4.3.1 Superstringhe di Tipo II
Per scrivere la funzione di partizione del toro nel caso della Superstringa in $`D=10`$, utilizziamo la formula di massa, scritta in termini di $`N^{}=N_B+N_F`$ and $`\overline{N}^{}=\overline{N}_B+\overline{N}_F`$
$$M^2=\frac{2}{\alpha ^{}}\left[N^{}+\overline{N}^{}+a+\overline{a}\right],$$
(4.55)
dove $`a=1/2`$ nel settore NS e $`a=0`$ nel settore R. Sostituendo nella (4.12) e introducendo, come nel caso bosonico, una funzione $`\delta `$ che tenga conto della condizione di level-matching, si può scrivere
$$𝒯=_{}\frac{d^2\tau }{\tau _2^2}\frac{1}{\tau _2^4}\mathrm{Str}\left(q^{N^{}+a}\overline{q}^{\overline{N}^{}+\overline{a}}\right).$$
(4.56)
Per calcolare la supertraccia occorre tener conto che lo spettro di superstringa è somma diretta dei quattro settori distinti NS-NS, R-R, NS-R e R-NS. Procediamo per passi, iniziando con il calcolare le tracce dei settori NS e R
$$\mathrm{tr}_{NS}q^{N_B+N_F\frac{1}{2}}=\frac{1}{\sqrt{q}}\mathrm{tr}\left(q^{_{n=1}^{\mathrm{}}\alpha _n^i\alpha _n^i}\right)\mathrm{tr}\left(q^{_{r=\frac{1}{2}}^{\mathrm{}}rb_r^ib_r^i}\right),$$
(4.57)
$$\mathrm{tr}_Rq^{N_B+N_F}=\mathrm{tr}\left(q^{_{n=1}^{\mathrm{}}\alpha _n^i\alpha _n^i}\right)\mathrm{tr}\left(q^{_{r=1}^{\mathrm{}}rb_r^ib_r^i}\right),$$
(4.58)
dove la somma sugli indici $`i=1\mathrm{}8`$ è sottintesa. La traccia bosonica è stata già calcolata nella (4.22), e nel calcolare invece la traccia fermionica occorre tener conto che, per il principio di esclusione di Pauli, i possibili numeri di occupazione sono solo $`0`$ e $`1`$, per cui nel settore NS si ha
$$\mathrm{tr}\left(q^{_{r=\frac{1}{2}}^{\mathrm{}}r\psi _r^i\psi _r^i}\right)=\underset{i=1}{\overset{8}{}}\underset{r=1/2}{\overset{\mathrm{}}{}}\left(1+q^r\right).$$
(4.59)
Nel settore R si deve tener conto che l’indice $`r`$ prende valori interi e che, per la presenza dei modi zero anticommutanti
$$\{\psi _0^i,\psi _0^j\}=\delta ^{ij},$$
(4.60)
il vuoto è una rappresentazione dell’algebra di Clifford in $`SO\left(8\right)`$. Di conseguenza la traccia avrà un fattore moltiplicativo $`2^{\left(D2\right)/2}=16`$ che ne conta la degenerazione.
Mettendo insieme i risultati per la parte bosonica e fermionica e ridefinendo l’indice $`r`$ nel settore NS si può scrivere
$$\mathrm{tr}_{NS}\left(q^{N_B+N_F\frac{1}{2}}\right)=\frac{_{n=1}^{\mathrm{}}\left(1+q^{n1/2}\right)^8}{\sqrt{q}_{n=1}^{\mathrm{}}\left(1q^n\right)^8},$$
(4.61)
$$\mathrm{tr}_R\left(q^{N_B+N_F}\right)=16\frac{_{n=1}^{\mathrm{}}\left(1+q^n\right)^8}{_{n=1}^{\mathrm{}}\left(1q^n\right)^8}.$$
(4.62)
Lo spettro della superstringa deve essere proiettato usando i proiettori GSO. Nel settore di Ramond si ha che $`\mathrm{tr}_R\left[\mathrm{\Gamma }_9\left(1\right)^F\right]=0`$ dal momento che ogni stato è accompagnato da un altro di chiralità opposta, e quindi
$$\mathrm{tr}_{NS}\left[q^{N_B+N_F\frac{1}{2}}\frac{1\left(1\right)^F}{2}\right]=\frac{_{n=1}^{\mathrm{}}\left(1+q^{n1/2}\right)^8_{n=1}^{\mathrm{}}\left(1q^{n1/2}\right)^8}{2\sqrt{q}_{n=1}^{\mathrm{}}\left(1q^n\right)^8},$$
(4.63)
$$\mathrm{tr}_R\left[q^{N_B+N_F}\frac{1\pm \mathrm{\Gamma }_9\left(1\right)^F}{2}\right]=16\frac{_{n=1}^{\mathrm{}}\left(1+q^n\right)^8}{_{n=1}^{\mathrm{}}\left(1q^n\right)^8}.$$
(4.64)
Le espressioni trovate possono essere scritte in termini di funzioni $`\theta `$ di Jacobi di argomento $`z=0`$ e caratteristiche $`\alpha `$ e $`\beta `$ uguali a 0 o a $`\frac{1}{2}`$, dove le funzioni $`\theta `$ di Jacobi sono definite come
$$\vartheta \left[\genfrac{}{}{0pt}{}{\alpha }{\beta }\right]\left(z|\tau \right)=\underset{n}{}q^{\frac{1}{2}\left(n+\alpha \right)^2}e^{2\pi i\left(n+\alpha \right)\left(z+\beta \right)},$$
(4.65)
o in forma di prodotto come
$`\vartheta \left[\genfrac{}{}{0pt}{}{\alpha }{\beta }\right]\left(z|\tau \right)`$ $`=`$ $`e^{2i\pi \alpha \left(z+\beta \right)}q^{\alpha ^2/2}{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}\left(1q^n\right){\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}\left(1+q^{n+\alpha 1/2}e^{2i\pi \left(z+\beta \right)}\right)`$ (4.66)
$`\times {\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}(1+q^{n\alpha 1/2}e^{2i\pi \left(z+\beta \right)}).`$
Utilizzando le definizioni (4.66) e (4.23) si possono costruire
$`{\displaystyle \frac{\vartheta ^4\left[\genfrac{}{}{0pt}{}{1/2}{1/2}\right]\left(0|\tau \right)}{\eta ^{12}\left(\tau \right)}}={\displaystyle \frac{\vartheta _1^4\left(0|\tau \right)}{\eta ^{12}\left(\tau \right)}}=0,`$ (4.67)
$`{\displaystyle \frac{\vartheta ^4\left[\genfrac{}{}{0pt}{}{1/2}{0}\right]\left(0|\tau \right)}{\eta ^{12}\left(\tau \right)}}={\displaystyle \frac{\vartheta _2^4\left(0|\tau \right)}{\eta ^{12}\left(\tau \right)}}=16{\displaystyle \frac{_{n=1}^{\mathrm{}}\left(1+q^n\right)^8}{_{n=1}^{\mathrm{}}\left(1q^n\right)^8}},`$ (4.68)
$`{\displaystyle \frac{\vartheta ^4\left[\genfrac{}{}{0pt}{}{0}{0}\right]\left(0|\tau \right)}{\eta ^{12}\left(\tau \right)}}={\displaystyle \frac{\vartheta _3^4\left(0|\tau \right)}{\eta ^{12}\left(\tau \right)}}={\displaystyle \frac{_{n=1}^{\mathrm{}}\left(1+q^{n1/2}\right)^8}{q^{1/2}_{n=1}^{\mathrm{}}\left(1q^n\right)^8}},`$ (4.69)
$`{\displaystyle \frac{\vartheta ^4\left[\genfrac{}{}{0pt}{}{0}{1/2}\right]\left(0|\tau \right)}{\eta ^{12}\left(\tau \right)}}={\displaystyle \frac{\vartheta _4^4\left(0|\tau \right)}{\eta ^{12}\left(\tau \right)}}={\displaystyle \frac{_{n=1}^{\mathrm{}}\left(1q^{n1/2}\right)^8}{q^{1/2}_{n=1}^{\mathrm{}}\left(1q^n\right)^8}},`$ (4.70)
che come si è visto compaiono nelle (4.63), che possono essere riscritte come
$`\mathrm{tr}_{NS}\left[q^{N_B+N_F\frac{1}{2}}{\displaystyle \frac{1\left(1\right)^F}{2}}\right]`$ $`=`$ $`{\displaystyle \frac{1}{\eta ^8}}{\displaystyle \frac{\vartheta _3^4\left(0|\tau \right)\vartheta _4^4\left(0|\tau \right)}{2\eta ^4\left(\tau \right)}},`$
$`\mathrm{tr}_R\left[q^{N_B+N_F}{\displaystyle \frac{1\pm \mathrm{\Gamma }_9\left(1\right)^F}{2}}\right]`$ $`=`$ $`{\displaystyle \frac{1}{\eta ^8}}{\displaystyle \frac{\vartheta _2^4\left(0|\tau \right)\pm \vartheta _1^4\left(0|\tau \right)}{2\eta ^4\left(\tau \right)}},`$ (4.71)
dove il segno nel settore di Ramond indica la chiralità del vuoto. A questo punto siamo quindi in grado di scrivere esplicitamente la (4.56), e tenendo conto del segno negativo per i contributi fermionici
$`𝒯={\displaystyle _{}}{\displaystyle \frac{d^2\tau }{\tau _2^6}}{\displaystyle \frac{1}{\eta ^8\left(\tau \right)\eta ^8\left(\overline{\tau }\right)}}({\displaystyle \frac{\vartheta _3^4\left(0|\tau \right)\vartheta _4^4\left(0|\tau \right)}{2\eta ^4\left(\tau \right)}}{\displaystyle \frac{\vartheta _2^4\left(0|\tau \right)\pm \vartheta _1^4\left(0|\tau \right)}{2\eta ^4\left(\tau \right)}})\times `$ (4.72)
$`\left({\displaystyle \frac{\vartheta _3^4\left(0|\overline{\tau }\right)\vartheta _4^4\left(0|\overline{\tau }\right)}{2\eta ^4\left(\overline{\tau }\right)}}{\displaystyle \frac{\vartheta _2^4\left(0|\overline{\tau }\right)\pm \vartheta _1^4\left(0|\overline{\tau }\right)}{2\eta ^4\left(\overline{\tau }\right)}}\right).`$
I due segni $``$ vengono fissati indipendentemente con la scelta della chiralità del vuoto R nei settori sinistro e destro, e si hanno due differenti funzioni di partizione a seconda che i due segni siano concordi o discordi. Le due teorie hanno spettri supersimmetrici: infatti l’identità
$$\vartheta _3^4\vartheta _4^4\vartheta _2^4=0,$$
(4.73)
conosciuta come aequatio identica satis abstrusa di Jacobi , alla luce delle considerazioni fatte, implica che ad ogni livello lo spettro contiene lo stesso numero di gradi di libertà bosonici e fermionici.
Per scrivere le funzioni di partizione delle due teorie supersimmetriche introduciamo la notazione dei caratteri, che è un utile modo per codificare l’intero contenuto di una rappresentazione. Nelle Teorie Conformi un carattere può essere scritto come
$$\chi \left(q\right)=q^{hc/24}\underset{k}{}d_kq^k,$$
(4.74)
dove $`h`$ è il peso conforme del campo primario e $`c`$ la carica centrale. I caratteri dell’estensione affine dell’algebra di $`so\left(8\right)`$ di livello $`k=1`$ sono esprimibili come
$`O_8`$ $`=`$ $`{\displaystyle \frac{\vartheta _3^4+\vartheta _4^4}{2\eta ^4}},V_8={\displaystyle \frac{\vartheta _3^4\vartheta _4^4}{2\eta ^4}},\left(NS\right)`$
$`S_8`$ $`=`$ $`{\displaystyle \frac{\vartheta _2^4+\vartheta _1^4}{2\eta ^4}},C_8={\displaystyle \frac{\vartheta _2^4\vartheta _1^4}{2\eta ^4}}.\left(R\right)`$ (4.75)
Se si tiene conto anche dei gradi di libertà bosonici, dividendo i caratteri di $`so\left(8\right)`$ per $`\eta ^8`$ si trova che gli sviluppi in $`q`$ di $`O_8`$, $`V_8`$, $`S_8`$ e $`C_8`$ contengono all’ordine più basso rispettivamente un tachione, un vettore, uno spinore sinistro e uno destro di Majorana-Weyl.
Le funzioni di partizione di superstringa chiusa possono ora essere scritte in forma molto semplice. Scegliendo i vuoti di Ramond destro e sinistro concordi si ottiene la funzione di partizione della teoria IIB
$$𝒯_{IIB}=\frac{d^2\tau }{\tau _2^2}\frac{1}{\tau _2^4\left(\eta \overline{\eta }\right)^8}\left|V_8S_8\right|^2,$$
(4.76)
altrimenti si ottiene la teoria IIA
$$𝒯_{IIA}=\frac{d^2\tau }{\tau _2^2}\frac{1}{\tau _2^4\left(\eta \overline{\eta }\right)^8}\left(\overline{V}_8\overline{S}_8\right)\left(V_8C_8\right).$$
(4.77)
Lo spettro di basse energie della IIA e della IIB può essere facilmente ritrovato dalle funzioni di partizione ricordando le proprietà dei caratteri. Il termine $`V_8\overline{V}_8`$ del settore NS-NS è comune alle due teorie e contiene il gravitone $`G_{\mu \nu }`$, il dilatone $`\varphi `$ e la due forma $`B_{\mu \nu }`$. Per la IIA nel settore R-R $`S_8\overline{C}_8`$ porta un vettore abeliano e una 3-forma, mentre i temini NS-R e R-NS, $`V_8\overline{C}_8`$ e $`S_8\overline{V}_8`$ due gravitini e due dilatini di chiralità opposta. Nella teoria IIB dal settore RR si ha un secondo scalare, un’altra 2-forma e una 4-forma con curvatura autoduale da $`S_8\overline{S}_8`$; due gravitini e due dilatini di stessa chiralità dai termini misti $`V_8\overline{S}_8+S_8\overline{V}_8`$.
Per verificare l’invarianza modulare delle funzioni di partizione trovate è utile determinare le matrici che implementano le trasformazioni $`T`$ ed $`S`$ sui caratteri dell’algebra di $`so\left(8\right)`$. Per farlo occorre partire dalle trasformazioni modulari sulle funzioni $`\vartheta `$
$$\vartheta \left[\genfrac{}{}{0pt}{}{\alpha }{\beta }\right]\left(z|\tau +1\right)=e^{i\pi \alpha \left(\alpha 1\right)}\vartheta \left[\genfrac{}{}{0pt}{}{\alpha }{\beta +\alpha 1/2}\right]\left(z|\tau \right),$$
(4.78)
$$\vartheta \left[\genfrac{}{}{0pt}{}{\alpha }{\beta }\right]\left(\frac{z}{\tau }|\frac{1}{\tau }\right)=\left(i\tau \right)^{1/2}e^{2i\pi \alpha \beta +i\pi z^2/\tau }\vartheta \left[\genfrac{}{}{0pt}{}{\beta }{\alpha }\right]\left(z|\tau \right).$$
(4.79)
Sostituendo nelle (4.3.1) si trova
$$T=e^{i\pi /3}\mathrm{diag}(1,1,1,1),$$
(4.80)
$`S={\displaystyle \frac{1}{2}}\left(\begin{array}{cccc}\hfill 1& \hfill 1& \hfill 1& \hfill 1\\ \hfill 1& \hfill 1& \hfill 1& \hfill 1\\ \hfill 1& \hfill 1& \hfill 1& \hfill 1\\ \hfill 1& \hfill 1& \hfill 1& \hfill 1\end{array}\right).`$ (4.85)
Ricordando che, come si è visto per la stringa bosonica, la misura di integrazione e il fattore $`\tau _2^4\left(\eta \overline{\eta }\right)^8`$ sono invariati modulari, è immediato, usando le trasformazioni (4.80) e (4.85), verificare l’invarianza modulare delle funzioni di partizione trovate.
#### 4.3.2 Superstringa di Tipo I
Come si è visto nel caso bosonico, si può costruire una teoria di stringhe aperte e chiuse non orientate proiettando su stati invarianti sotto $`\mathrm{\Omega }`$. Per avere una teoria consistente occorre che la teoria di partenza sia invariante sotto $`\mathrm{\Omega }`$, e questo avviene solo per la IIB, dove i settori destro e sinistro hanno la stessa proiezione GSO . La teoria risultante è detta superstringa di tipo I. Al contrario la teoria IIA non può essere proiettata rispettando l’invarianza di Lorentz in $`D=10`$.
Procedendo esattamente come nel caso bosonico e ricordando i calcoli fatti per le tracce fermioniche, si trovano nel canale diretto le ampiezze
$$𝒦=\frac{1}{2}_0^{\mathrm{}}\frac{d\tau _2}{\tau _2^6}\frac{\left(V_8S_8\right)\left(2i\tau _2\right)}{\eta ^8\left(2i\tau _2\right)},$$
(4.86)
$$𝒜=\frac{N^2}{2}_0^{\mathrm{}}\frac{d\tau _2}{\tau _2^6}\frac{\left(V_8S_8\right)\left(\frac{1}{2}i\tau _2\right)}{\eta ^8\left(\frac{1}{2}i\tau _2\right)},$$
(4.87)
$$=\frac{ϵN}{2}_0^{\mathrm{}}\frac{d\tau _2}{\tau _2^6}\frac{\left(\widehat{V}_8\widehat{S}_8\right)\left(\frac{1}{2}i\tau _2+\frac{1}{2}\right)}{\widehat{\eta }^8\left(\frac{1}{2}i\tau _2+\frac{1}{2}\right)}.$$
(4.88)
Nella (4.88) per avere un integrando reale si sono introdotti caratteri $`\widehat{\chi }`$ definiti in generale come
$$\widehat{\chi }\left(i\tau _2+\frac{1}{2}\right)=q^{hc/24}\underset{k}{}\left(1\right)^kd_kq^k,q=e^{2\pi \tau _2},$$
(4.89)
che differiscono da $`\chi \left(i\tau _2+1/2\right)`$ per una fase $`e^{i\pi \left(hc/24\right)}`$.
La bottiglia di Klein simmetrizza il settore $`NSNS`$ e antisimmetrizza il settore $`RR`$. Lo spettro di massa nulla nel settore chiuso risultante dalla proiezione $`𝒯𝒯/2+𝒦`$ si ottiene dalla IIB eliminando la $`2`$-forma dal settore $`NSNS`$ e lo scalare e la $`4`$-forma autoduale dal settore $`RR`$, e dimezzando i settori misti. Gli stati che sopravvivono alla proiezione formano un multipletto $`𝒩=1`$ di supergravità minimale in $`D=10`$ dimensioni. Nel settore aperto proiettato si hanno $`\left(N^2ϵN\right)/2`$ vettori di massa nulla e i corrispettivi partner supersimmetrici, fermioni di Majorana-Weyl. Si ha quindi un multipletto di $`N=1`$ super Yang-Mills nell’aggiunta del gruppo $`SO\left(N\right)`$ per $`ϵ=1`$, e nell’aggiunta di $`USp\left(N\right)`$ per $`ϵ=1`$.
Il passaggio nel canale trasverso nel caso della bottiglia di Klein e dell’anello non comporta nessuna novità rispetto al caso bosonico, dal momento che come si è visto la combinazione $`\left(V_8S_8\right)`$ è invariante sotto S. Le ampiezze trasverse sono quindi:
$$\stackrel{~}{𝒦}=\frac{2^5}{2}_0^{\mathrm{}}𝑑\mathrm{}\frac{\left(V_8S_8\right)\left(i\mathrm{}\right)}{\eta ^8\left(i\mathrm{}\right)}.$$
(4.90)
$$\stackrel{~}{𝒜}=\frac{2^5N^2}{2}_0^{\mathrm{}}𝑑\mathrm{}\frac{\left(V_8S_8\right)\left(i\mathrm{}\right)}{\eta ^8\left(i\mathrm{}\right)}.$$
(4.91)
I coefficienti dei caratteri $`V_8`$ e $`S_8`$ sono da interpretare in termini di quadrati delle funzioni ad un punto su un bordo. Il passaggio al canale trasverso per la Möbius richiede qualche precisazione. La ridefinzione dei caratteri (4.89) porta a ridefinire anche la trasformazione P, che come già osservato diventa
$$P=T^{1/2}ST^2ST^{1/2}.$$
(4.92)
Sulla base
$$\frac{O_8}{\tau _2^4\eta ^8},\frac{V_8}{\tau _2^4\eta ^8},\frac{S_8}{\tau _2^4\eta ^8},\frac{C_8}{\tau _2^4\eta ^8}$$
(4.93)
la matrice T agisce in maniera semplice $`T=\mathrm{diag}(1,1,1,1)`$, e dal momento che in generale per una teoria conforme vale
$$S^2=\left(ST\right)^3=𝒞,$$
(4.94)
dove si è indicata con $`𝒞`$ la matrice di coniugazione, che essendo le rappresentazioni di $`so\left(8\right)`$ autoconiugate, nel nostro caso coincide con l’identità, dalla (4.92) si ottiene
$$P=T=\mathrm{diag}(1,1,1,1).$$
(4.95)
Quindi l’ampiezza di Möbius nel canale trasverso risulta essere
$$\stackrel{~}{}=2\frac{ϵN}{2}_0^{\mathrm{}}𝑑\mathrm{}\frac{\left(\widehat{V}_8\widehat{S}_8\right)\left(i\mathrm{}+\frac{1}{2}\right)}{\widehat{\eta }^8\left(i\mathrm{}+\frac{1}{2}\right)}.$$
(4.96)
Le condizioni di tadpole della Tipo I per i settori $`NSNS`$ e $`RR`$, in conseguenza della supersimmetria, risultano nell’unica condizione
$$\frac{2^5}{2}+\frac{2^5N^2}{2}+2\frac{ϵN}{2}=\frac{2^5}{2}\left(N+32ϵ\right)^2=0,$$
(4.97)
che seleziona i valori (N=32, $`ϵ=1)`$, e quindi il gruppo di gauge $`SO\left(32\right)`$ .
È importante notare che, sebbene per la superstringa di Tipo I i tadpole NS-NS e R-R vengano cacellati simultaneamente, concettualmente le condizioni nei due settori hanno un significato fisico molto differente. Quello che si può vedere è che, dal punto di vista spazio-temporale, i bordi del worldsheet tracciati dalle estremità delle stringhe aperte sono mappati in oggetti aperti, $`D9`$-brane, che riempiono completamente lo spazio tempo, mentre i crosscap vengono mappati in oggetti non dinamici gli $`O`$-piani. In generale sia le $`Dp`$-brane che gli $`Op`$-piani hanno tensione e portano cariche R-R con potenziali che sono $`\left(p+1\right)`$-forme $`C_{p+1}`$. Per una $`D`$-brana tensione e carica sono entrambe positive, mentre come si avrà modo di vedere, nel vuoto perturbativo della Tipo I sono presenti due tipi di $`O`$-piani: gli $`O_+`$-piani con carica e tensione negative e gli $`O_{}`$-piani con carica e tensione positiva. In più si hanno $`D`$-antibrane e $`O`$-antipiani (indicate con $`\overline{D}`$-brane e $`\overline{O}`$-piani) con stessa tensione e cariche R-R opposte.
A questo punto dovrebbe essere chiaro che mentre la cancellazione dei tadpole R-R è una condizione di di neutralità della totale della carica, necessaria in presenza di compattificazioni dal momento che le linee di Farady del potenziale $`C_{p+1}`$ si trovano ad essere confinate. Al contrario, la condizione di tadpole NS-NS, come si è visto da luogo ha una correzione all’energia di vuoto dipendente dal dilatone,
$$V_\varphi T𝑑x\sqrt{detG}e^\varphi ,$$
(4.98)
che può anche non essere cancellata . Difatti nei modelli con supersimmetria rotta, i tadpole NS-NS non possono essere, in generale, cancellati e sono accompagnati dall’insorgere di divergenze infrarosse. Questo indica che il background Minkowskiano non è più una soluzione della teoria e che è necessaria una ridefinizione del vuoto .
### 4.4 Proiezione di orientifold
Si è visto sia nel caso di stringa bosonica sia in quello di superstringa come sia possibile costruire, con la proiezione di orientifold, da una teoria di stringhe chiuse orientate, teorie consistenti di stringhe aperte e chiuse non orientate. Diamo ora una formulazione più generale della costruzione. L’ampiezza di toro è in generale scrivibile, lasciando sottintesa l’integrazione, come (4.99)
$$𝒯=\underset{i,j}{}\overline{\chi }_iX_{ij}\chi _j,$$
(4.99)
con $`X_{ij}`$ una matrice generica di numeri interi (nei modelli razionali è finto dimensionale). Si può richiedere per semplicità $`X_{ij}=0,1`$, dal momento che altrimenti si avrebbe una ambiguità da risolvere nella proiezione. Per avere invariaza modulare si devono imporre i vincoli
$$S^{}XS=X,T^{}XT=X.$$
(4.100)
La bottiglia di Klein si ottiene identificando modi destri e sinistri, e quindi si propagheranno solo i settori che nel toro siano simmetrici sotto $`\mathrm{\Omega }`$, quindi con $`X_{ii}0`$
$$𝒦=\frac{1}{2}\underset{i}{}𝒦^i\chi _i,𝒦^i=\pm X_{ii}.$$
(4.101)
Il segno dei $`𝒦^i`$ è indeterminato, dal momento nell’ampiezza di toro $`X_{ij}`$ era associato ai moduli quadri dei caratteri. Nel canale trasverso si può scrivere
$$\stackrel{~}{𝒦}=\frac{1}{2}\underset{i}{}\left(\mathrm{\Gamma }^i\right)^2\chi _i,$$
(4.102)
dove i coefficienti $`\mathrm{\Gamma }^i`$ sono funzioni ad un punto (ovvero coefficienti di riflessione) dei caratteri sui crosscap. I segni di $`𝒦^i`$ quindi devono essere scelti in modo da avere una teoria interagente consistente ovvero con coefficienti positivi $`\left(\mathrm{\Gamma }^i\right)^2=K_jS_{ij}`$ nel canale trasverso.
L’ampiezza di anello nel canale trasverso ha la forma
$$\stackrel{~}{𝒜}=\frac{1}{2}\underset{i}{}\chi _i\left(\underset{a}{}B_a^in^a\right)^2,$$
(4.103)
dove $`_aB_a^in^a`$ sono le funzioni ad un punto sui bordi e le $`n_a`$ le molteplicità associate alle cariche di Chan-Paton. Dal momento che un settore che si rifletta su di un bordo subisce una coniugazione di carica, nell’anello trasverso si propagheranno solo i caratteri $`\chi _i`$ che compaiano nel toro nella forma $`\overline{\chi }_i^C\chi _i`$. Nel canale diretto l’ampiezza di anello può essere scritta come
$$𝒜=\frac{1}{2}\underset{i,a,b}{}𝒜_{ab}^in^an^b\chi _i,$$
(4.104)
con $`𝒜_{ab}^i=0,1`$ come nel caso del toro.
Il nastro di Möbius nel canale trasverso si è visto essere un tubo di propagazione fra un bordo e un crosscap, e pertanto i coefficenti dei caratteri possono essere scritti come prodotti delle rispettive funzioni ad un punto su un bordo e su un crosscap che compaiono nell’anello e nella bottiglia di Klein. Un fattore moltiplicativo due tiene conto delle due possibili configurazioni. Riassumendo l’ampiezza trasversa risulta
$$\stackrel{~}{}=\frac{2}{2}\underset{i}{}\widehat{\chi }_i\mathrm{\Gamma }^i\left(\underset{a}{}B_a^in^a\right),$$
(4.105)
mentre nel canale diretto si ha
$$=\frac{1}{2}\underset{i,a}{}^i{}_{a}{}^{}n_{}^{a}\widehat{\chi }_i.$$
(4.106)
Si può verificare la consistenza della costruzione controllando che $``$ sia la corretta proiezione di $`𝒜`$, ovvero che si abbia $`^i{}_{a}{}^{}=\pm 𝒜_{aa}^i`$.
## Capitolo 5 Compattificazioni e T-dualità
I modelli di superstringa sono consistenti in $`D=10`$, mentre la stringa bosonica ha dimensione critica $`D=26`$. Dal momento che la nostra esperienza fisica mostra l’esistenza di sole tre dimensioni spaziali estese e del tempo, per costruire modelli realistici occorre introdurre meccanismi che rendano compatte e “microscopiche” le dimensioni extra. In altri termini occorre pensare ad uno spazio-tempo Minkowskiano spontaneamente rotto,
$$^D=^4\times K^{D4},$$
(5.1)
dove $`K`$ è una varietà compatta. Il modo più semplice di introdurre compattificazioni è quello di identificare periodicamente $`n`$ coordinate su di un toro di dimensione $`n`$. Si tratta di una estensione molto semplice del modello di Kaluza-Klein, che però nel caso di stringa comporta l’insorgere di fenomeni del tutto nuovi rispetto alla teoria dei campi come la presenza degli stati di winding, l’innalzamento delle simmetrie di gauge e la T-dualità . Si è inoltre indotti poi ad introdurre oggetti dinamici estesi, le $`D`$-brane e gli $`O`$-piani .
Una generalizzazione di questo tipo di compattificazioni sono le compattificazioni su orbifold , che si ottengono identificando i punti della varietà interna , sotto l’azione di un gruppo discreto. Un orbifold non è una varietà liscia dal momento, che in generale, si avranno punti fissi sotto l’azione del gruppo discreto che però possono essere rimossi ottenendo varietà lisce di Calabi-Yau.
### 5.1 Compattificazioni toroidali
#### 5.1.1 Compattificazione su $`S^1`$ per stringhe chiuse
Partiamo dal caso più semplice, quello di una teoria di stringa bosonica compattificata su un cerchio unidimensionale $`S^1`$ di raggio R. Si devono imporre condizioni di periodicità su un singolo campo scalare,
$$X(\tau ,\sigma )X(\tau ,\sigma )+2\pi R,$$
(5.2)
e la periodicità ha due effetti. Anzitutto, dal momento che l’operatore di traslazione deve essere univocamente definito
$$e^{ipX}e^{ip\left(X+2\pi R\right)},$$
(5.3)
l’impulso del centro di massa è quantizzato:
$$p=\frac{n}{R}.$$
(5.4)
Il secondo effetto è peculiare della Teoria delle Stringhe: una stringa chiusa può avvolgersi intorno alla dimensione compatta,
$$X(\tau ,\sigma +\pi )=X(\tau ,\sigma )+2\pi Rw,w.$$
(5.5)
Il numero intero $`w`$ è il *winding number* ed è conservato nelle interazioni di stringa. Le osservazioni fatte possono essere raccolte scrivendo l’espasione del campo $`X`$ in oscillatori
$$X=x+2\alpha ^{}\frac{n}{R}\tau +2wR\sigma +i\frac{\sqrt{2\alpha ^{}}}{2}\underset{k0}{}\left(\frac{\alpha _k}{k}e^{2ik\left(\tau \sigma \right)}+\frac{\stackrel{~}{\alpha }_k}{k}e^{2ik\left(\tau +\sigma \right)}\right).$$
(5.6)
Che può essere scritta come somma dei modi destri e sinistri $`X=X_L+X_R`$, con
$$X_{L,R}=\frac{1}{2}x+\alpha ^{}p_{L,R}\left(\tau \sigma \right)+\left(oscillatori\right),$$
(5.7)
dove si sono definiti gli impulsi sinistro e destro come
$$p_L=\frac{n}{R}+\frac{wR}{\alpha ^{}},p_R=\frac{n}{R}\frac{wR}{\alpha ^{}}.$$
(5.8)
La massa degli stati di stringa va riscritta tenendo conto che, dal punto di vista di un osservatore che viva in $`\left(D1\right)`$-dimensioni, $`M^2=p^2`$, dove p è il momento degli stati in 25 dimensioni. In altri termini, i momenti interni contribuiscono all’energia a riposo della stringa, e la formula di massa è quindi
$$M^2=\frac{2}{\alpha ^{}}\left[\frac{\alpha ^{}}{4}p_L^2+\frac{\alpha ^{}}{4}p_R^2+N^{}+\overline{N}^{}2\right],$$
(5.9)
con il vincolo
$$\frac{\alpha ^{}}{4}p_R^2+N^{}\left(\frac{\alpha ^{}}{4}p_L^2+\overline{N}^{}\right)=0.$$
(5.10)
##### Spettro compattificato ed allargamento della simmetria di gauge
Usando le (5.8), possiamo riscrivere il vincolo di massa e la condizione di level matching come
$$M^2=\frac{n^2}{R^2}+\frac{w^2R^2}{\alpha ^2}+\frac{2}{\alpha ^{}}\left[N^{}+\overline{N}^{}2\right],0=nw+N^{}\overline{N}^{}.$$
(5.11)
Studiamo i modi di massa nulla, cosiderando una compattificazione lungo la direzione $`X^{25}`$. A valori generici di R, per avere stati di stringa a massa nulla, deve aversi $`m=w=0`$ e $`N=\overline{N}=1`$. Si hanno $`24^2`$ stati di massa nulla come per la teoria non compatta, che adesso possiamo scrivere in maniera conveniente, dal punto di vista 25 dimensionale come
$$\alpha _1^\mu \stackrel{~}{\alpha }_1^\nu |0,0,\alpha _1^\mu \stackrel{~}{\alpha }_1^{25}|0,0,\alpha _1^{25}\stackrel{~}{\alpha }_1^\mu |0,0,\alpha _1^{25}\stackrel{~}{\alpha }_1^{25}|0,0.$$
(5.12)
Il primo si decompone in un gravitone $`G_{\mu \nu }`$, un tensore antisimmetrico $`B_{\mu \nu }`$ ed un dilatone $`\varphi `$ in 25 dimensioni. Il secondo e il terzo sono vettori di Kaluza-Klein
$$A{}_{\mu }{}^{}\left(R\right)\frac{1}{2}(GB)_{\mu ,25},A{}_{\mu }{}^{}\left(L\right)\frac{1}{2}(G+B)_{\mu ,25}$$
(5.13)
che portano una simmetria di gauge $`U\left(1\right)_L\times U\left(1\right)_R`$ . L’ultimo è uno scalare che possiamo scrivere come
$$\varphi ^{}\frac{1}{2}\mathrm{log}G_{25,25},$$
(5.14)
ed è il modulo del raggio effettivo di compattificazione, dal momento che la radice quadrata della componente della metrica $`G_{25,25}`$ è proprio la misura del raggio della direzione compattificata $`X^{25}`$. Il suo valore di aspettazione non può essere fissato con un principio di minimo e rimane indeterminato.
Un nuovo effetto tipicamente di stringa si ha al valore del raggio $`R=\sqrt{\alpha ^{}}`$, dove il vincolo di massa può essere scritto come
$$M^2=\frac{2}{\alpha ^{}}\left(n+w\right)^2+\frac{4}{\alpha ^{}}\left(N^{}1\right)=\frac{2}{\alpha ^{}}\left(nw\right)^2+\frac{4}{\alpha ^{}}\left(\overline{N}^{}1\right),$$
(5.15)
con i vincoli
$$\left(n+w\right)^2+4N^{}=4,\left(nw\right)^2+4\overline{N}^{}=4.$$
(5.16)
In aggiunta alle soluzioni di massa nulla già discusse per $`n=w=0`$ con $`N=\overline{N}=1`$, si hanno anche
$$n=w=\pm 1,N^{}=0,\overline{N}^{}=1;n=w=\pm 1,N^{}=1,\overline{N}^{}=0;$$
(5.17)
$$n=\pm 2,w=0,N^{}=\overline{N}^{}=0;n=0,w=\pm 2,N^{}=\overline{N}^{}=0.$$
(5.18)
Che in termini di stati corrispondono a 8 nuovi scalari
$$\stackrel{~}{\alpha }_1^{25}|\pm 1,1,\alpha _1^{25}|\pm 1,\pm 1,|\pm 2,0,|0,2,$$
(5.19)
e 4 nuovi vettori
$$\stackrel{~}{\alpha }_1^\mu |\pm 1,1,\alpha _1^\mu |\pm 1,\pm 1.$$
(5.20)
In totale si hanno quindi 9 particelle scalari, 3 vettori destri e 3 vettori sinistri, il gruppo di simmetria abeliano $`U\left(1\right)_L\times U\left(1\right)_R`$ è stato promosso a $`SU\left(2\right)_L\times SU\left(2\right)_R`$.
Spostando R dal valore $`\sqrt{\alpha ^{}}`$ i bosoni di gauge extra acquistano massa
$$M=\frac{\left|R^2\alpha ^{}\right|}{R\alpha ^{}}\frac{2}{\alpha ^{}}\left|R\alpha ^{1/2}\right|.$$
(5.21)
Come noto, è possibile in Teoria dei Campi (e quindi anche nella teoria effettiva di basse energie di stringa) dare massa a bosoni di gauge attraverso una rottura spontanea di simmetria. Infatti per $`R=\sqrt{\alpha ^{}}`$ ci sono 10 scalari di massa nulla, alcuni dei quali vengono riassorbiti per dare massa a 4 dei sei vettori di gauge.
##### T-dualità per stringhe chiuse
Dalla formula di massa (5.11) si vede che nel limite $`R\mathrm{}`$ gli stati di winding diventano infinitamente massivi mentre lo spettro degli impulsi diventa continuo, e pertanto nel limite di decompattificazione si ritrova la fisica conosciuta. Al contrario nel limite $`R0`$ gli stati di momento compattificato diventano infinitamente massivi, mentre lo spettro di winding diventa continuo. Si trova quindi che anche nel limite di riduzione dimensionale lo spettro di stringa sembra nuovamente diventare quello di una dimensione non compatta, ovvero i limiti $`R\mathrm{}`$ e $`R0`$ sono fisicamente identici. Lo spettro di massa è infatti invariante sotto
$$T:RR^{}=\frac{\alpha ^{}}{R},nw.$$
(5.22)
La trasformazione definita, detta *T-dualità*, agisce sui momenti come
$$T:p_Lp_L,p_Rp_R\mathrm{ovvero}T:\alpha _0\alpha _0,\stackrel{~}{\alpha }_0\stackrel{~}{\alpha }_0.$$
(5.23)
Si può definire l’azione sugli oscillatori generalizzando quella sugli zero modi, si vede in questo modo che la T-dualità agisce come una trasformazione di parità sul solo settore destro:
$$T:X(\tau ,\sigma )=X_L\left(\tau \sigma \right)+X_R\left(\tau +\sigma \right)X^T(\tau ,\sigma )=X_L\left(\tau \sigma \right)X_R\left(\tau +\sigma \right)$$
(5.24)
Si è detto che i limti $`R\mathrm{}`$ e $`R0`$ sono fisicamente equivalenti per la stringa bosonica, e quindi lo spazio delle teorie inequivalenti è definito dai dominii $`R\alpha ^{1/2}`$ oppure $`0R\alpha ^{1/2}`$. La prima scelta è più naturale, dal momento che è più intuitivo ragionare in termini di momenti continui piuttosto che di windings.
Dalla teoria effettiva di bassa energia si può vedere che la T-dualità è anche una simmetria della teoria interagente, ma la sua azione è non banale sul dilatone (e più in generale sui campi di background), e quindi modifica la costante di accoppiamento di stringa:
$$T:\varphi \varphi ^{}=\varphi +log\sqrt{\alpha ^{}}R.$$
(5.25)
Si può vedere che al raggio autoduale $`R=\sqrt{\alpha ^{}}`$, in cui si ha l’enhancement del gruppo di gauge, la simmetria $`_2`$ di T-dualità diventa parte della simmetria continua $`SU\left(2\right)\times SU\left(2\right)`$. Questo è particolarmente importante in quanto che indica che la T-dualità non è una simmetria solo della teoria perturbativa, ma più in generale della teoria esatta.
##### Funzione di partizione
Calcoliamo l’ampiezza di vuoto di stringa chiusa su $`S^1`$ a partire dalla (4.9), dove l’integrazione va fatta sui momenti dello spazio non compatto ($`D1`$)dimensionale, per cui si ottiene una differente potenza del modulo $`\tau _2`$
$$\frac{1}{\tau _2^{D/2+1}}\frac{1}{\tau _2^{\left(D1\right)/2+1}}.$$
(5.26)
Utilizzando la formula di massa (5.9) e il vincolo (5.10) si ottiene inoltre la traccia
$$\mathrm{tr}\left(q^{N^{}+\frac{\alpha ^{}}{4}p_R^21}\overline{q}^{\overline{N}^{}+\frac{\alpha ^{}}{4}p_L^21}\right)=\frac{_{m,n}q^{\frac{\alpha ^{}}{4}p_R^2}\overline{q}^{\frac{\alpha ^{}}{4}p_L^2}}{\left[\eta \left(\tau \right)\eta \left(\overline{\tau }\right)\right]^{24}},$$
(5.27)
che è stata calcolata sugli stati $`|k;n,w`$, e la funzione di partizione del toro è quindi
$$𝒯=_{}\frac{d^2\tau }{\tau _2^2}\frac{_{m,n}q^{\frac{\alpha ^{}}{4}p_R^2}\overline{q}^{\frac{\alpha ^{}}{4}p_L^2}}{\tau _2^{121/2}\left[\eta \left(\tau \right)\eta \left(\overline{\tau }\right)\right]^{24}}.$$
(5.28)
L’ampiezza trovata è invariante modulare. Sotto T, essa prende una fase $`\mathrm{exp}\left(i\pi \left(p_L^2p_R^2\right)\right)`$ che si vede essere unitaria ricordando $`p_L^2p_R^2=2nw`$. Per verificare l’invarianza sotto S occorre la formula di rissomazione di Poisson
$$\underset{\left\{n_i\right\}}{}e^{\pi n^\mathrm{T}An+2i\pi b^\mathrm{T}n}=\frac{1}{\sqrt{\mathrm{det}\left(A\right)}}\underset{\left\{m_i\right\}}{}e^{\pi \left(mb\right)^\mathrm{T}A^1\left(mb\right)},$$
(5.29)
e dopo una trasformazione $`S`$ la sommatoria in (5.28) usando le espressioni (5.8) si scrive
$$\underset{n,w}{}e^{\pi n^2\left(\frac{\alpha ^{}\tau _2}{\left|\tau \right|^2R^2}\right)}e^{2\pi in\left(\frac{w\tau _1}{\left|\tau \right|^2}\right)}e^{\frac{\pi \tau _2w^2R^2}{\alpha ^{}\left|\tau \right|^2}}.$$
(5.30)
Si può usare la formula di risommazione per $`n`$ con $`A=\alpha ^{}\tau _2/|\tau |^2R^2/`$ e $`b=w\tau _1/\left|\tau \right|^2`$ ottenendo
$$\frac{\left|\tau \right|R}{\sqrt{\alpha ^{}\tau _2}}\underset{n^{},w}{}e^{\frac{\pi \left|\tau \right|^2R^2}{\alpha ^{}\tau _2}\left(n^{}\frac{w\tau _1}{\left|\tau \right|^2}\right)^2}e^{\frac{\pi \tau _2w^2R^2}{\alpha ^{}\left|\tau \right|^2}},$$
(5.31)
che può essere riscritta nella forma
$$\frac{\left|\tau \right|R}{\sqrt{\alpha ^{}\tau _2}}\underset{n^{},w}{}e^{\pi w^2\left(\frac{R^2}{\alpha ^{}\tau _2}\right)}e^{2\pi in^{}\frac{iR^2w\tau _1}{\alpha ^{}\tau _2}}e^{\pi n^2\frac{\left|\tau \right|^2R^2}{\alpha ^{}\tau _2}}.$$
(5.32)
Utilizzando nuovamente la formula di risommazione rispetto a $`w`$ con $`A=R^2/\alpha ^{}\tau _2`$ e $`b=iR^2n\tau _1/\alpha ^{}\tau _2`$ si arriva a scrivere
$$\left|\tau \right|\underset{n^{},w^{}}{}e^{\pi \frac{\alpha ^{}\tau _2}{R^2}\left(w^{}+i\frac{R^2n^{}\tau _1}{\alpha ^{}\tau _2}\right)^2}e^{\pi \frac{\left|\tau \right|^2R^2n^2}{\alpha ^{}\tau _2}},$$
(5.33)
che si vede subito essere l’espressione di partenza con winding e momenti scambiati e con un fattore moltiplicativo $`\left|\tau \right|`$ che viene riassorbito dalla trasformazione S su $`1/\eta \overline{\eta }`$.
#### 5.1.2 Compattificazione su $`S^1`$ per stringhe aperte
Le stringhe aperte non possono allacciarsi lungo una direzione compatta, e non hanno quindi numero di winding, e pertanto il loro comportamento nel limite $`R0`$ deve per questo essere differente da quello delle stringhe chiuse e più simile a quello di una particella. Per $`R0`$ gli stati di momento diventano infinitamente massivi ma, mancando uno spettro di winding che diventi continuo, si ha una riduzione dimensionale. Si è visto che una teoria di stringhe aperte deve contenere anche stringhe chiuse, e sembrerebbe quindi che nel limite di “riduzione dimensionale” si trovi una teoria in cui le stringhe aperte vivono in $`D1`$ dimensioni e quelle chiuse in $`D`$. In realtà i punti che distinguono una stringa chiusa da una aperta sono gli estremi. Con questa osservazione si può arrivare a capire che nel limite $`R0`$ entrambe le stringhe oscillano in spazi $`D`$-dimensionali, ma gli estremi di stringa aperta si trovano vincolati ad un iperpiano $`D1`$-dimensionale.
Scrivendo l’espansione dei modi di una stringa aperta nella forma $`X=X_L+X_R`$ con,
$`X_L`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left(x+q\right)+\alpha ^{}{\displaystyle \frac{n}{R}}\left(\tau +\sigma \right)+i{\displaystyle \frac{\sqrt{2\alpha ^{}}}{2}}{\displaystyle \underset{k}{}}{\displaystyle \frac{\alpha _k}{k}}e^{ik\left(\tau +\sigma \right)},`$
$`X_R`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left(x+q\right)+\alpha ^{}{\displaystyle \frac{n}{R}}\left(\tau \sigma \right)+i{\displaystyle \frac{\sqrt{2\alpha ^{}}}{2}}{\displaystyle \underset{k}{}}{\displaystyle \frac{\alpha _k}{k}}e^{ik\left(\tau \sigma \right)},`$ (5.34)
dove $`q`$ indica un termine arbitrario che si cancella nella somma, e ricordando la definizione della T-dualità sui campi X nella (5.24), per la coordinata duale si ottiene
$$X^T=X_LX_R=q+2\alpha ^{}\frac{n}{R}\sigma +i\left(\sqrt{2\alpha ^{}}\right)\underset{k}{}\frac{\alpha _k}{k}e^{ik\tau }\mathrm{sin}\left(k\sigma \right).$$
(5.35)
Nella coordinata T-duale non c’è più dipendenza da $`\tau `$ negli zero modi, e per questo il momento del centro di massa è nullo. Infatti, dal momento che i termini oscillanti si annullano per $`\sigma =0,\pi `$, si vede che la stringa ha estremi fissati nella direzione compattificata:
$$X^T\left(\pi \right)=X^T\left(0\right)=\frac{2\pi \alpha ^{}n}{R}=2\pi nR^{}.$$
(5.36)
Si hanno pertanto condizioni di Dirichlet agli estremi $`_\tau X=0`$, in luogo delle usuali condizioni di Neumann $`_\sigma X=0`$ . È immediato rendersi conto di cosa è avvenuto osservando che
$`_\sigma X`$ $`=`$ $`X_L^{}\left(\tau +\sigma \right)X_R^{}\left(\tau \sigma \right)=_\tau X^T,`$
$`_\tau X`$ $`=`$ $`X_L^{}\left(\tau +\sigma \right)+X_R^{}\left(\tau \sigma \right)=_\sigma X^T.`$ (5.37)
Naturalmente questo vale solo nella coordinata T-dualizzata, mentre gli estremi rimangono liberi di muoversi liberamente nelle altre dimensioni, che definiscono un iperpiano detto $`D`$-brana . Le $`D`$-brane, come si vedrà, sono in realtà oggetti dinamici.
##### Cariche di Chan-Paton e linee di Wilson
Con le cariche di Chan-Paton, la T-dualità di stringa aperta si arricchisce di nuovi aspetti. È infatti possibile rompere il gruppo di gauge delle cariche introducendo *linee di Wilson* preservandone però il rango . Consideriamo, per semplicità, il caso della stringa bosonica orientata con gruppo di gauge $`U\left(N\right)`$. Un campo costante di background, della direzione compattificata $`X^{25}`$, può essere diagonalizzato nella forma
$$A_{25}=diag\{\theta _1,\theta _2,\mathrm{},\theta _N\}/2\pi R.$$
(5.38)
In questa forma il campo si trova nel sottogruppo diagonale $`U\left(1\right)^n`$ di $`U\left(N\right)`$. Localmente è un campo di pura gauge, la cui curvatura è nulla e quindi le equazioni del campo sono soddisfatte in maniera banale da
$$A_{25}=i\mathrm{\Lambda }^1\mathrm{\Lambda },\mathrm{\Lambda }=diag\{e^{iX^{25}\theta _1/2\pi R},e^{iX^{25}\theta _2/2\pi R},\mathrm{},e^{iX^{25}\theta _N/2\pi R}\}.$$
(5.39)
Dal momento che il parametro di gauge $`\mathrm{\Lambda }`$ non è periodico nella dimensione compatta si hanno interessanti effetti fisici. Infatti, anche ponendo il campo di background a zero con una trasformazione di gauge $`\mathrm{\Lambda }^1`$, si ha che i campi carichi sotto una trasformazione $`X^{25}X^{25}+2\pi R`$ acquistano una fase
$$diag\{e^{i\theta _1},e^{i\theta _2},\mathrm{},e^{i\theta _N}\}.$$
(5.40)
Si può definire una quantità gauge invariante detta linea di Wilson, che tenga conto degli effetti fisici introdotti dal campo di background
$$W=\mathrm{exp}\left(iq𝑑X^{25}A_{25}\right)=\mathrm{exp}\left(iq\theta \right).$$
(5.41)
L’azione di stringa viene modificata dall’introduzione di $`A_{25}`$ con la comparsa di termini di bordo
$$S=\frac{T}{2}𝑑\tau 𝑑\sigma \left(\dot{X}^2X^2\right)+q_i𝑑\tau A_{25}\dot{X}^{25}|_{\sigma =0}+q_j𝑑\tau A_{25}\dot{X}^{25}|_{\sigma =0}.$$
(5.42)
Il momento canonico risulta “shiftato”, per una stringa $`|ij`$ che ha carica $`+1`$ sotto $`U\left(1\right)_i`$ , $`1`$ sotto $`U\left(1\right)_j`$ si ha
$$p_{25}=\frac{\left(2\pi n\theta _j+\theta _i\right)}{2\pi R},$$
(5.43)
e lo spettro di massa diventa
$$M^2=\frac{\left(2\pi n\theta _j+\theta _i\right)^2}{4\pi ^2R^2}+\frac{1}{\alpha ^{}}\left(N1\right).$$
(5.44)
Per i bosoni di gauge a massa nulla, con n=0 e N=1
$$M^2=\frac{\left(\theta _j\theta _i\right)^2}{4\pi ^2R^2},$$
(5.45)
se tutti i $`\theta _i`$ sono distinti, gli unici vettori di massa nulla sono quelli con $`i=j`$, e il gruppo di gauge è rotto a $`U\left(1\right)^n`$. Se invece $`r`$ dei $`\theta _i`$ sono uguali, la corrispondente matrice $`r\times r`$ di vettori è di massa nulla, e il gruppo di simmetria è rotto a $`U\left(r\right)\times U\left(1\right)^{Nr}`$.
Per capire cosa stia succedendo passiamo nella picture T-duale, dove, dal momento che gli impulsi diventano winding, si avranno winding frazionari ovvero,
$$X^{25,T}\left(\pi \right)X^{25,T}\left(0\right)=\left(2\pi n+\theta _i\theta _j\right)R^{},$$
(5.46)
e, a meno di costanti additive, l’estremo della stringa nello stato $`i`$ è nella posizione
$$X^{25,T}=\theta _iR^{}=2\pi \alpha ^{}A_{25,ii}.$$
(5.47)
Gli estremi della stringa giacciono quindi su due iperpiani ($`D`$-brane) che si trovano in posizioni differenti. Nella picture T-duale i $`\theta _i`$ sono quindi angoli che definiscono nella dimensione compattificata la posizione degli iperpiani. In generale si hanno N iperpiani in posizioni differenti lungo il cerchio di compattificazione, e in questo caso gli stati di massa nulla sono gli scalari $`\alpha _1^{25}|i,i`$ e i vettori $`\alpha _1^i|i,i`$ nell’aggiunta di U(1). Quando si fanno coincidere $`r`$ delle N $`D`$-brane, si hanno $`r^2`$ vettori di massa nulla (e lo stesso numero di scalari) dati dagli stati $`\alpha _1^i|i,j`$ per $`i,j1,\mathrm{}r`$, che corrispondono agli stati di stringhe con gli estremi attaccati a brane coincidenti.
#### 5.1.3 T-dualità per stringhe non orientate
Consideriamo una teoria di stringhe chiuse non orientate. Per ottenerla si è proiettato lo spazio degli stati della teoria orientata sugli autovettori di $`\mathrm{\Omega }`$ con autovalore $`+1`$. La T-dualità si è vista essere una parità sui soli modi destri della stringa. Quindi nella picture T-duale, l’azione di $`\mathrm{\Omega }`$ è
$$\mathrm{\Omega }:X^T=\left(X_LX_R\right)\left(X_RX_L\right)=X^T.$$
(5.48)
Dal momento che, proiettando lo spettro con $`\mathrm{\Omega }`$ si sono ottenute *teorie non orientate*, mentre, come si vedrà nei prossimi paragrafi, proiettando con una riflessione spazio temporale $`_2`$ si ottengono gli *orbifold*, il risultato combinato prende il nome di *orientifold*.
Nella picture T-duale il cerchio di compattificazione di raggio R’ è mappato nel segmento $`[0,\pi R^{}]`$, i cui estremi sono punti fissi dell’involuzione. Questi piani detti $`O`$-piani sono all’origine della natura non orientata dello spettro. Al contrario, lontano dagli $`O`$-piani, la fisica locale è quella delle stringhe orientate. A differenza della $`D`$-brane, gli $`O`$-piani non sono oggetti dinamici. Nel caso di una teoria con $`k`$ dimensioni compatte, lo spazio tempo duale sarà naturalmente il toro $`T^k`$ con identificazioni $`_2`$ nelle direzioni compatte, ovvero un ipercubo con $`2^k`$ O(25-k)-piani sui vertici.
Nel caso di stringhe aperte, cosiderando una sola dimensione compatta, si hanno due $`O`$-piani nei punti $`X=0,\pi R^{}`$. Introducendo cariche di Chan-Paton con simmetria di gauge SO(2N) (il caso del gruppo simplettico è identico), una linea di Wilson generica può essere ridotta alla forma diagonale per 2N autovalori
$$W=diag\{e^{i\theta _1},e^{i\theta _1},e^{i\theta _2},e^{i\theta _2},\mathrm{},e^{i\theta _N},e^{i\theta _N},\}.$$
(5.49)
Si hanno cioè nella picture T-duale $`N`$ $`D`$-brane sul segmento fondamentale $`[0,\pi R^{}]`$ ed altre $`N`$ $`D`$-brane immagine nei punti identificati dall’involuzione di orientifold. Le stringhe si possono stendere fra le $`D`$-brane acquistando massa, e il gruppo di gauge si rompe, nel caso più generale, a $`U\left(1\right)^n`$. Come nel caso orientato, un gruppo di $`r`$ $`D`$-brane coincidenti ricompone un fattore $`U\left(r\right)`$. Adesso però c’è una nuova possibilità: che $`r`$ $`D`$-brane coincidano con un $`O`$-piano e quindi con le loro brane immagine: in questo modo si ottiene l’ampliamento della simmetria di gauge a $`SO\left(2r\right)`$. La simmetria massimale si recupera naturalmente per tutte le brane coincidenti con uno dei due $`O`$-piani, e in questo caso il gruppo è $`SO\left(2N\right)`$. La rottura della simmetria di gauge, introdotta con questo meccanismo preserva il rango.
#### 5.1.4 Compattificazioni di Superstringa su $`S^1`$
##### Funzioni di partizione di Superstringa
Per illustrare la compattificazione toroidale nel caso di superstringa studiamo il caso della IIB e la costruzione di orientifold della Tipo I. Nel caso bosonico si è visto che, se una dimensione viene compattificata, l’integrazione sui momenti interni è sostituita da una sommatoria sul $`n`$ e $`w`$. L’effetto della compattificazione sull’ampiezza di vuoto può essere riassunto con la sostituzione
$$\frac{1}{\eta \overline{\eta }\sqrt{\tau _2}}\underset{n,w}{}\frac{q^{\frac{\alpha ^{}}{4}p_R^2}\overline{q}^{\frac{\alpha ^{}}{4}p_L^2}}{\eta \overline{\eta }},$$
(5.50)
ed è immediato rendersi conto che tale sostituzione è valida anche nel caso di superstringa. La funzione di partizione della IIB, dando per intesa l’integrazione su $`\tau _2`$ (e la misura di integrazione) e i gradi di libertà bosonici non compatti, si scrive
$$𝒯=\left|V_8S_8\right|^2\underset{n,w}{}\frac{q^{\alpha ^{}p_R^2/4}\overline{q}^{\alpha ^{}p_L^2/4}}{\eta \left(\tau \right)\eta \left(\overline{\tau }\right)}.$$
(5.51)
Per costruire i discendenti aperti iniziamo con la bottiglia di Klein. Nella bottiglia di Klein possono propagarsi solo gli stati simmetrici sotto scambio dei modi sinistri e destri, che, come si vede facilmente dalle (5.8), sono gli stati con $`w=0`$
$$𝒦=\frac{1}{2}\left(V_8S_8\right)\left(2i\tau _2\right)\underset{n}{}\frac{\left(e^{2\pi \tau _2}\right)^{\alpha ^{}n^2/2R^2}}{\eta \left(2i\tau _2\right)}.$$
(5.52)
C’è un’altra possibilità nella definizione di $`\mathrm{\Omega }`$, perché è possibile assegnare differenti autovalori agli stati di momento pari e dispari, compatibilmente con la teoria interagente, ottenendo
$$𝒦^{}=\frac{1}{2}\left(V_8S_8\right)\left(2i\tau _2\right)\underset{n}{}()^n\frac{\left(e^{2\pi \tau _2}\right)^{\alpha ^{}n^2/2R^2}}{\eta \left(2i\tau _2\right)},.$$
(5.53)
L’ampiezza nel canale trasverso la si ottiene con la sostituzione $`t=2\tau _2`$, effettuando una trasformazione modulare $`S`$ ed infine utilizzando la formula di risommazione di Poisson (5.29)
$$\stackrel{~}{𝒦}=\frac{2^5}{2}\frac{R}{\sqrt{\alpha ^{}}}\left(V_8S_8\right)\left(i\mathrm{}\right)\underset{w}{}\frac{\left(e^{2\pi \mathrm{}}\right)^{\left(2wR\right)^2/4\alpha ^{}}}{\eta \left(i\mathrm{}\right)},$$
(5.54)
$$\stackrel{~}{𝒦^{}}=\frac{2^5}{2}\frac{R}{\sqrt{\alpha ^{}}}\left(V_8S_8\right)\left(i\mathrm{}\right)\underset{w}{}\frac{\left(e^{2\pi \mathrm{}}\right)^{\left(2w+1\right)^2R^2/4\alpha ^{}}}{\eta \left(i\mathrm{}\right)},$$
(5.55)
dove le potenze di 2 si ottengono considerando l’ampiezza completa, con il fattore $`\tau _2^{41/2}`$. E’interessante notare come nel canale trasverso i momenti siano diventati winding. Inoltre la formula di Possion ha dato origine ad un fattore $`R/\sqrt{\alpha ^{}}`$ che tiene conto del volume della dimensione trasversa.
Nella seconda proiezione si vede come il fattore $`()^n`$ dopo la risommazione porti uno shift del winding. In questo modello, nella bottiglia di Klein non si propagano stati di massa nulla, dal momento che lo shift nei winding fa si che manchi il termine $`q^0`$, e non ci siano pertanto divergenze. Si tratta di un modello consistente di sole stringhe chiuse, mentre il primo modello richiede l’introduzione dei settori aperti. Si è detto che gli stati che si riflettono su di un bordo subiscono una coniugazione, e nell’anello trasverso si propagheranno quindi solo stati di momento nullo, quelli per i quali $`p_L=p_R`$
$$\stackrel{~}{𝒜}=\frac{2^5}{2}N^2\frac{R}{\sqrt{\alpha ^{}}}\left(V_8S_8\right)\left(i\mathrm{}\right)\underset{w}{}\frac{\left(e^{2\pi \mathrm{}}\right)^{w^2R^2/4\alpha ^{}}}{\eta \left(i\mathrm{}\right)}.$$
(5.56)
L’ampiezza di anello diretto si trova, come di consueto, con una trasformazione S e una risommazione di Poisson
$$𝒜=\frac{1}{2}N^2\left(V_8S_8\right)\left(\frac{1}{2}i\tau _2\right)\underset{n}{}\frac{\left(e^{2\pi \tau _2}\right)^{\alpha ^{}n^2/2R^2}}{\eta \left(\frac{1}{2}i\tau _2\right)},$$
(5.57)
dalle ampiezze trasverse di anello e bottiglia di Klein si può scrivere l’ampiezza della striscia di Möbius nel canale trasverso
$$\stackrel{~}{}=\frac{2}{2}N\frac{R}{\sqrt{\alpha ^{}}}\left(\widehat{V}_8\widehat{S}_8\right)\left(i\mathrm{}+\frac{1}{2}\right)\underset{w}{}\frac{\left(e^{2\pi \mathrm{}}\right)^{\left(2wR\right)^2/4\alpha ^{}}}{\widehat{\eta }\left(i\mathrm{}+\frac{1}{2}\right)},$$
(5.58)
ed infine operando una trasformazione $`P`$ ed una risommazione, si ottiene l’ampiezza nel canale diretto
$$=\frac{1}{2}N\left(\widehat{V}_8\widehat{S}_8\right)\left(\frac{1}{2}i\tau _2+\frac{1}{2}\right)\underset{n}{}\frac{\left(e^{2\pi \tau _2}\right)^{\alpha ^{}n^2/2R^2}}{\widehat{\eta }\left(\frac{1}{2}i\tau _2+\frac{1}{2}\right)}.$$
(5.59)
Le condizioni di tadpole NS-NS e R-R sono entrambe soddisfatte fissando $`N=32`$, $`ϵ=+1`$, ottenendo il gruppo di gauge $`SO\left(32\right)`$. In altri termini si hanno 32 $`D9`$-brane su un $`09_+`$-piano, e questo garantisce la cancellazione delle divergenze.
Si è detto che le stringhe aperte permettono l’introduzione di linee di Wilson che rompono il gruppo di simmetria delle cariche di Chan-Paton preservandone il rango. Possiamo vedere un esempio di questo fenomeno nella superstringa Tipo I. Una linea di Wilson produce uno shift sui momenti $`nn+a_i+a_j`$, l’ampiezza di anello nel canale diretto si scrive
$$𝒜=\frac{1}{2}\left(V_8S_8\right)\left(\frac{1}{2}i\tau _2\right)\underset{n,i,j}{}\frac{\left(e^{2\pi \tau _2}\right)^{\alpha ^{}\left(n+a_i+a_j\right)^2/2R^2}}{\eta \left(\frac{1}{2}i\tau _2\right)}.$$
(5.60)
Nel canale trasverso, come si vede dalla formula di risommazione di Possion, lo shift sui momenti si traduce in termini lineari in $`w`$ nell’esponenziale e, di conseguenza, in una fase che può essere scritta in maniera conveniente definendo una matrice costante diagonale
$$W=\mathrm{diag}(\mathrm{e}^{2\pi \mathrm{ia}_1},\mathrm{e}^{2\pi \mathrm{ia}_2},\mathrm{},\mathrm{e}^{2\pi \mathrm{ia}_{32}}),$$
(5.61)
con $`a_2=a_1,a_4=a_3,\mathrm{}a_{32}=a_{31}`$, $`0<\left|a_i\right|<1`$. Non è difficile rendersi conto che i coefficienti cercati corrispondono a
$$\left(\mathrm{trW}^\mathrm{w}\right)^2=\underset{\mathrm{i},\mathrm{j}}{}\mathrm{e}^{2\pi \mathrm{iw}\left(\mathrm{a}_\mathrm{i}+\mathrm{a}_\mathrm{j}\right)}.$$
(5.62)
L’ampiezza di anello nel canale trasverso si scrive
$$\stackrel{~}{𝒜}=\frac{2^5}{2}\frac{R}{\sqrt{\alpha ^{}}}\left(V_8S_8\right)\left(i\mathrm{}\right)\underset{w}{}\frac{\left(\mathrm{trW}^\mathrm{w}\right)^2\left(e^{2\pi \mathrm{}}\right)^{w^2R^2/4\alpha ^{}}}{\eta \left(i\mathrm{}\right)},$$
(5.63)
in le fasi introdotte dalle linea di Wilson corrispondono a coefficienti di riflessione diferenti nei diversi settori di winding. L’ampiezza Möbius trasversa si ricava come di solito da quella di anello e di Klein. Nella Klein trasversa si propagavano solo i settori con winding pari, quindi si ottiene
$$\stackrel{~}{}=\frac{2}{2}\frac{R}{\sqrt{\alpha ^{}}}\left(\widehat{V}_8\widehat{S}_8\right)\left(i\mathrm{}+\frac{1}{2}\right)\underset{w}{}\frac{\mathrm{trW}^{2\mathrm{w}}\left(\mathrm{e}^{2\pi \mathrm{}}\right)^{\left(2\mathrm{w}\mathrm{R}\right)^2/4\alpha ^{}}}{\widehat{\eta }\left(i\mathrm{}+\frac{1}{2}\right)}.$$
(5.64)
Passando al canale diretto si vede che si propagano solo i settori con $`a_i=a_j`$, che è un risultato atteso tenendo conto della non orientabilità delle superficie
$$=\frac{1}{2}\left(\widehat{V}_8\widehat{S}_8\right)\left(\frac{1}{2}i\tau _2+\frac{1}{2}\right)\underset{n,i}{}\frac{\left(e^{2\pi \tau _2}\right)^{\alpha ^{}\left(n+2a_i\right)^2/2R^2}}{\widehat{\eta }\left(\frac{1}{2}i\tau _2+\frac{1}{2}\right)}.$$
(5.65)
Studiando lo spettro di massa nulla di anello di vede che, in generale, sono presenti 16 vettori che corrispondono a prendere $`n=0`$, $`a_i=a_j`$. Il gruppo di gauge è quindi rotto a $`U\left(1\right)^{16}`$. Nella rappresentazione T-duale questa scelta corrisponde a posizionare tutte le brane in punti differenti.
Naturalmente ponendo a zero un certo numero di $`a_i`$ e segliendo i rimanenti $`a_1\mathrm{}a_{2M}`$ tutti con valori differenti si ha un parziale recupero della simmetria di gauge: $`SO\left(322M\right)\times U\left(1\right)^M`$. Questo nel mondo T-dualizzato corrisponde ad avere $`M`$ D$`8`$-brane e le rispettive brane-immagine separate e $`\left(322M\right)`$ D$`8`$-brane coincidenti con l’$`O`$-piano nell’origine della dimensione compatta.
Specializziamo ore le ampiezze trovate in un caso di particolare interesse, per $`a_1=a_3=\mathrm{}a_{2M1}=A`$, $`a_2=a_4=\mathrm{}a_{2M}=A`$, e $`a_{2M+1}=\mathrm{}a_{32}=0`$. Definendo $`N=322M`$, si può scrivere
$`𝒜`$ $`=`$ $`(V_8S_8)\left(\frac{1}{2}i\tau _2\right){\displaystyle \underset{n}{}}\{(M\overline{M}+{\displaystyle \frac{1}{2}}N^2){\displaystyle \frac{q^{\alpha ^{}n^2/2R^2}}{\eta \left(\frac{1}{2}i\tau _2\right)}}`$ (5.66)
$`+MN{\displaystyle \frac{q^{\alpha ^{}\left(n+A\right)^2/2R^2}}{\eta \left(\frac{1}{2}i\tau _2\right)}}+\overline{M}N{\displaystyle \frac{q^{\alpha ^{}\left(nA\right)^2/2R^2}}{\eta \left(\frac{1}{2}i\tau _2\right)}}`$
$`+{\displaystyle \frac{1}{2}}M^2{\displaystyle \frac{q^{\alpha ^{}\left(n+2A\right)^2/2R^2}}{\eta \left(\frac{1}{2}i\tau _2\right)}}+{\displaystyle \frac{1}{2}}\overline{M}^2{\displaystyle \frac{q^{\alpha ^{}\left(n2A\right)^2/2R^2}}{\eta \left(\frac{1}{2}i\tau _2\right)}}\},`$
$``$ $`=`$ $`(\widehat{V}_8\widehat{S}_8)(\frac{1}{2}i\tau _2+\frac{1}{2}){\displaystyle \underset{n}{}}\{{\displaystyle \frac{1}{2}}N{\displaystyle \frac{q^{\alpha ^{}n^2/2R^2}}{\widehat{\eta }\left(\frac{1}{2}i\tau _2+\frac{1}{2}\right)}}`$ (5.67)
$`+{\displaystyle \frac{1}{2}}M{\displaystyle \frac{q^{\alpha ^{}\left(n+2A\right)^2/2R^2}}{\widehat{\eta }\left(\frac{1}{2}i\tau _2+\frac{1}{2}\right)}}+{\displaystyle \frac{1}{2}}\overline{M}{\displaystyle \frac{q^{\alpha ^{}\left(n2A\right)^2/2R^2}}{\widehat{\eta }\left(\frac{1}{2}i\tau _2+\frac{1}{2}\right)}}\}.`$
Naturalmente $`M`$ e $`\overline{M}`$ sono numeri uguali, ma la notazione è utile per sottolineare che si riferiscono a cariche coniugate. Si hanno $`M\overline{M}`$ e $`N^2/2`$ stati di massa nulla nell’ampiezza di anello e $`N/2`$ dalla Möbius, che formano la rappresentazione aggiunta del gruppo di gauge $`SO\left(N\right)\times U\left(M\right)`$. Nel canale trasverso
$`\stackrel{~}{𝒜}`$ $`=`$ $`{\displaystyle \frac{2^5}{2}}\left(V_8S_8\right)\left(i\mathrm{}\right){\displaystyle \frac{R}{\sqrt{\alpha ^{}}}}{\displaystyle \underset{w}{}}{\displaystyle \frac{q^{w^2R^2/4\alpha ^{}}}{\eta \left(i\mathrm{}\right)}}\left(N+Me^{2i\pi Aw}+\overline{M}e^{2i\pi Aw}\right)^2,`$
$`\stackrel{~}{}`$ $`=`$ $`{\displaystyle \frac{2}{2}}\left(\widehat{V}_8\widehat{S}_8\right)\left(i\mathrm{}+\frac{1}{2}\right){\displaystyle \frac{R}{\sqrt{\alpha ^{}}}}{\displaystyle \underset{w}{}}{\displaystyle \frac{q^{\left(2wR\right)^2/4\alpha ^{}}}{\eta \left(i\mathrm{}+\frac{1}{2}\right)}}\left(N+Me^{4i\pi Aw}+\overline{M}e^{4i\pi Aw}\right),`$
e le condizioni di tadpole fissano $`N=322M`$. Nella picture T-duale si hanno $`322M`$ D$`8`$-brane sull’$`O`$-piano posto nell’origine e le altre $`M`$ brane e le loro immagini in un punto arbitrario della dimensione compatta. Anche in questo caso si può verificare un fenomeno di estensione della simmetria di gauge. Per la scelta $`A=1/2`$, $`M`$ e $`\overline{M}`$ hanno uguali coefficienti di riflessione nel canale trasverso, mentre nel canale diretto l’ampiezza dipende solo dalla loro somma. Questa scelta delle $`a_i`$ corrisponde nella descrizione T-duale ad avere N $`D8`$-brane coincidenti con l’$`O`$-piano all’origine ed M $`D8`$-brane coincidenti con l’$`O`$-piano posto a $`\pi R^{}`$.
##### T-dualità delle superstringhe di Tipo II
Nel caso delle superstringhe di Tipo II la T-dualità mostra l’esistenza di una relazione molto profonda che lega le due teorie IIA e IIB. Compattificare una singola coordinata $`X^9`$ in una delle due teorie considerarne il limite $`R0`$ equivale a considerare il limite $`R\mathrm{}`$ nella coordinata duale, la cui componente destra è riflessa
$$X_R^{9,T}=X_R^9,$$
(5.69)
come nel caso bosonico. Per l’invarianza superconforme si deve avere anche
$$\stackrel{~}{\psi }^{9,T}=\stackrel{~}{\psi }^9,$$
(5.70)
questo implica che la chiralità del vuoto del settore destro di Ramond è invertita. Se ad esempio si parte dalla teoria IIA, si compattifica una direzione e si prende il limite di $`R`$ piccolo, si ottiene esattamente la teoria IIB con $`R`$ grande, e viceversa.
#### 5.1.5 Compattificazioni toroidali in più dimensioni
Il caso della compattificazione sul cerchio può essere generalizzato a compattificazioni su tori d-dimensionali $`T^d(S^1)^d`$. Denotando con $`X^m`$ le dimensioni compatte, con $`m=1,\mathrm{},d`$, la condizione di periodicità è data da
$$X^mX^m+2\pi R^{\left(m\right)}n^m,$$
(5.71)
dove $`n^m`$ sono interi e $`R^{\left(m\right)}`$ è il raggio del m-esimo cerchio. La metrica sul toro può essere scritta, introducendo i vielbein $`e_m^a`$ nella forma diagonale
$$g_{mn}=\delta _{ab}e_m^ae_n^b,$$
(5.72)
ed è conveniente definire coordinate nello spazio tangente $`X^a=X^me_m^a`$. Più in generale con una matrice $`e_m^a`$ generica la relazione di equivalenza diventa
$$X^aX^a+2\pi e_m^an^m.$$
(5.73)
Si è quindi definito un reticolo $`\mathrm{\Lambda }=\left\{e_m^an^m,n^m\right\}`$. Il toro può pertanto essere pensato come quoziente dello spazio piatto d-dimensionale sul reticolo
$$T\frac{^d}{2\pi \mathrm{\Lambda }}.$$
(5.74)
I momenti coniugati delle coordinate $`X^a`$, che indichiamo con $`p^a`$ saranno quantizzati. Infatti spostandoci da un punto del reticolo ad un altro, producendo una variazione di $`X`$ di $`\delta X2\pi \mathrm{\Lambda }`$, si deve avere
$$e^{ipX}=e^{ip\left(X+\delta X\right)},$$
(5.75)
che impone $`p\delta X2\pi `$, ovvero,
$$p^n=g^{mn}n_m,$$
(5.76)
dove $`n_m`$ sono interi. Moltiplicando la (5.76) per $`e_n^a`$, si trova che gli impulsi $`p^a`$ vivono nel reticolo duale
$$\mathrm{\Lambda }^{}\left\{e^{am}n_m,n_m\right\},$$
(5.77)
dove i vielbein inversi sono definiti con la metrica inversa
$$e^{am}e_m^ag^{mn},e^{am}e_m^b=\delta ^{ab}.$$
(5.78)
Naturalmente si possono avere anche settori di winding. Sul modello di quanto visto nel caso unidimensionale, definiamo un momento destro ed uno sinistro
$$p_L^a=p^a+\frac{w^aR^{\left(a\right)}}{\alpha ^{}}e^{am}n_m+\frac{1}{\alpha ^{}}e_m^aw^m,$$
(5.79)
$$p_R^a=p^a\frac{w^aR^{\left(a\right)}}{\alpha ^{}}e^{am}n_m\frac{1}{\alpha ^{}}e_m^aw^m.$$
(5.80)
Per studiare le proprietà del reticolo definiamo una base “più grande” che contenga i due reticoli distinti $`\mathrm{\Lambda },\mathrm{\Lambda }^{}`$:
$$\widehat{e}_m=\frac{1}{\alpha ^{}}\left(\genfrac{}{}{0pt}{}{e_m^a}{e_m^a}\right),\widehat{e}^m=\left(\genfrac{}{}{0pt}{}{e^{am}}{e^{am}}\right).$$
(5.81)
Nella nuova base si ha
$$\widehat{p}=\left(\genfrac{}{}{0pt}{}{p_L^a}{p_R^a}\right)=\widehat{e}_mw^m+\widehat{e}^mn_m.$$
(5.82)
Si è definito un reticolo in $`\left(d+d\right)`$ dimensioni che chiamiamo $`\mathrm{\Gamma }_{d,d}`$. Possiamo definire una metrica Lorentziana di segnatura $`(d,d)`$
$$G=\left(\genfrac{}{}{0pt}{}{\delta _{ab}}{0}\genfrac{}{}{0pt}{}{0}{\delta _{ab}}\right),$$
(5.83)
da cui si vede che
$`\widehat{e}_m\widehat{e}_n=0=\widehat{e}^m\widehat{e}^n,`$
$`\widehat{e}_m\widehat{e}^n={\displaystyle \frac{2}{\alpha ^{}}}\delta _n^m,`$ (5.84)
che mostrano che il reticolo è *autoduale* dal momento che, a meno di un riscalamento, si ha $`\mathrm{\Gamma }_{d,d}^{}=\mathrm{\Gamma }_{d,d}`$. Inoltre, il prodotto interno fra due momenti è
$$\left(\widehat{e}_mw^m+\widehat{e}^mn_m\right)\left(\widehat{e}_nw^n+\widehat{e}^nn_n^{}\right)=\frac{2}{\alpha ^{}}\left(w^mn_m^{}+n_mw^m\right),$$
(5.85)
e quindi il reticolo è *pari*. Queste due proprietà del reticolo sono necessarie per garantire l’invarianza modulare della teoria. Consideriamo infatti la generalizzazione della funzione di partizione per una compattificazione in $`d`$ dimensioni, che può essere scritta nella forma (per i gradi di libertà compatti)
$$𝒯=\frac{_{\mathrm{\Gamma }_{d,d}}q^{\frac{\alpha ^{}}{4}p_Lp_L}\overline{q}^{\frac{\alpha ^{}}{4}p_Rp_R}}{\left[\eta \left(\tau \right)\eta \left(\overline{\tau }\right)\right]^d},.$$
(5.86)
Sotto T si ha un fattore di fase nella funzione di partizione $`\mathrm{exp}\left(i\pi \alpha ^{}\left(p_Lp_Lp_Rp_R\right)/2\right)`$, che è uguale ad 1 per la parità del reticolo, mentre sotto una trasformazione S, generalizzando i calcoli visti nel caso della funzione di partizione di una compattificazione unidimensionale, si trova
$$𝒯_\mathrm{\Gamma }\left(\frac{1}{\tau }\right)=vol\left(\mathrm{\Gamma }^{}\right)𝒯_\mathrm{\Gamma }^{}\left(\tau \right).$$
(5.87)
Nel caso di reticoli autoduali il volume della cella fondamentale è unitario, infatti in generale vale $`vol\left(\mathrm{\Gamma }\right)vol\left(\mathrm{\Gamma }^{}\right)=1`$ e, per i reticoli autoduali, deve aversi anche ovviamente $`vol\left(\mathrm{\Gamma }\right)=vol\left(\mathrm{\Gamma }^{}\right)`$. La funzione di partizione risulta quindi invariante modulare.
Ogni reticolo con metrica Lorentziana con segnatura $`(d,d)`$ può essere trasformato in un altro reticolo Lorenziano mediante una trasformazione di $`O(d,d,)`$. Il gruppo $`O(d,d,R)`$ non è una simmetria della teoria perché cambia lo spettro. Lo spettro di massa dipende però dai prodotti scalari $`p_Lp_L`$ e $`p_Rp_R`$, ed è quindi invariante sotto $`O(d,)\times O(d,)`$. Le teorie inequivalenti sono date dalle descritte
$$\frac{O(d,d,)}{O(d,)\times O(d,)},$$
(5.88)
il cui numero di parametri è
$$\frac{2d\left(2d1\right)}{2}d\left(d1\right)=d^2.$$
(5.89)
Dal punto di vista micorscopico i $`d^2`$ gradi di libertà possono essere identificati con la combinazione del tensore simmetrico $`g_{mn}`$ e di quello antisimmetrico $`B_{mn}`$. Infatti è possibile considerare una naturale generalizzazione dell’azione di stringa in cui si considerano come campi di background oltre ai gravitoni anche i tensori antisimmetrici e i dilatoni tutti a valori costanti. Per gli impulsi, in questo caso, si ha, scalando il vielbein,
$$p_{L,m}=n_m+\frac{1}{\alpha ^{}}\left(g_{mn}B_{mn}\right)w^n,$$
(5.90)
$$p_{R,m}=n_m\frac{1}{\alpha ^{}}\left(g_{mn}+B_{mn}\right)w^n.$$
(5.91)
Il gruppo di trasformazioni (5.88) deve essere diviso anche per il gruppo di trasformazioni di T-dualità, che è molto più ampio rispetto al caso di compattificazione su $`S^1`$, e corrisponde a $`O(d,d,)`$:
$$\frac{O(d,d,)}{O(d,)\times O(d,)\times O(d,d,)},$$
(5.92)
Questo gruppo contiene, rispetto ai campi di background, trasformazioni di dualità sui singoli assi $`R\alpha ^{}/R`$, shift del tensore antisimmetrico $`B_{mn}B_{mn}+N_{mn}`$ con $`N_{mn}`$ interi e trasformazioni che rispettano la periodicità della forma $`x^{}=L_n^mx^n`$, con L matrice a valori interi e determinante unitario.
La generalizzazione delle funzioni di partizione aperte nel caso di compattificazioni su $`S^1`$ al caso $`T^d`$ è abbastanza immediata, ma esiste un’importante novità. È interessante infatti notare come la costruzione di orientifold, imponga la quantizzazione del tensore antisimmetrico $`B_{mn}`$. Imponendo la simmetria sotto $`\mathrm{\Omega }`$ della teoria si trova infatti la condizione
$$n_m+\frac{1}{\alpha ^{}}\left(g_{mn}B_{mn}\right)w^n=n_m^{}\frac{1}{\alpha ^{}}\left(g_{mn}+B_{mn}\right)w^n,$$
(5.93)
che porta a richiedere $`n=n^{}`$, $`w=w^{}`$ e
$$\frac{2}{\alpha ^{}}B_{mn}.$$
(5.94)
La presenza del tensore antisimmetrico quantizzato permette di rompere il gruppo di gauge riducendone il rango . Essa può essere interpretata come una manifestazione della presenza simultanea di due tipi di $`O`$-piani, $`O_+`$ e $`O_{}`$, con cariche e tensioni opposte.
### 5.2 Compattificazioni su orbifold
Gli orbifold toroidali sono una classe di compattificazioni di grande interesse in teoria delle stringhe, dal momento che introducono in modo naturale rotture di simmetrie. Prima di vedere alcuni esempi, diamone una formulazione geometrica.
In generale un orbifold è lo spazio quoziente di una varietà $``$ su un gruppo discreto $`G`$, la cui azione sia definita sulla varietà $`G:`$. Lo spazio quoziente $`\mathrm{\Gamma }/𝒢`$ è quindi costruito identificando i punti con la relazione di equivalenza $`xgx`$ per tutti gli elementi del gruppo $`gG`$. In generale l’azione del gruppo definisce dei punti fissi, ovvero punti che non trasformano sotto l’azione di $`G`$: dato un punto $`x`$ si ha $`gx=x`$ per $`gG`$. I punti fissi sono singolari e un orbifold in generale non è una varietà. Da un punto di vista fisico nei punti singolari (si comportano genericamente come il vertice di un cono) la dinamica delle particelle è mal definita, ma nel caso di stringhe chiuse orientate l’invarianza modulare determina completamete gli spettri. È anche possibile rimuovere le singolarità (tecnica di “blow up”) ottenendo varietà lisce di Calabi-Yau e loro generalizzazioni.
Vediamo la costruzione di una teoria di stringa su un orbifold . Consideriamo una teoria invariante modulare, il cui spazio di Hilbert ammetta una simmetria discreta G. A partire da questa teoria se ne può definire un’altra i cui stati siano invarianti sotto l’azione del gruppo e che sia ancora invariante modulare. Per costruire la nuova funzione di partizione si comincia dal definire funzioni di partizione “twistate”, che chiameremo $`Z_{g,h}`$, ovvero si impongono sui campi condizioni al bordo definite come
$$\varphi \left(z+1\right)=g\varphi \left(z\right),\varphi \left(z+\tau \right)=h\varphi \left(z\right),$$
(5.95)
dove $`g,hG`$ e $`\left|G\right|`$ è il numero di elementi del gruppo. La funzione di partizione della nuova teoria invariante modulare si può a questo punto scrivere nella forma
$$Z_{orb}=\frac{1}{\left|G\right|}\underset{g,hG}{}Z_{g,h},$$
(5.96)
dove nel caso di gruppi non abeliani la somma coinvolge in realtà solo coppie $`(g,h)`$ commutanti. La somma su $`g`$ è riferita ai vari settori “twistati” dello spazio di Hilbert della teoria, cioè a fissate scelte del tempo $`\tau `$ sul world-sheet, mentre la somma su $`b`$ produce una proiezione invariante sotto l’azione del gruppo in ogni settore. La (5.96) è effettivamente invariante modulare, come si verifica facilmente dal momento che
$$T:Z_{g,h}Z_{g,gh},S:Z_{g,h}Z_{h,g},$$
(5.97)
da cui si comprende bene che la richiesta di commutatività degli elementi $`(g,h)`$ garantisce che l’azione di $`T`$ sia ben definita.
In forma operativa la costruzione di un orbifold avviene con una proiezione dello spazio di Hilbert degli stati su un sottospazio G-invariante, per mezzo del proiettore
$$P=\frac{1}{\left|G\right|}\underset{hG}{}h,$$
(5.98)
da cui si ottiene la funzione di partizione
$$Z_{proj}=\frac{1}{\left|G\right|}\underset{hG}{}Z_{1,h},$$
(5.99)
che è chiaramente non invariante modulare, dal momento che ad esempio una trasformazione S manda $`Z_{1,h}`$ in $`Z_{h,1}`$. L’invarianza modulare viene recuperata sommando su tutti i possibili “twist” spaziali $`g`$ che commutino con $`h`$, e in questo modo si ottiene la (5.96).
#### 5.2.1 Orbifold $`S^1/_2`$ di stringa bosonica
Studiamo, come caso introduttivo, una teoria di stringa bosonica compattificata su un orbifold $`S^1/_2`$. Il cerchio $`S^1`$, parametrizzato da una coordinata $`X`$, ha una simmetria $`_2`$, $`g:XX`$. Questa simmetria si estende allo spettro degli stati e agli operatori della teoria di stringa. Alcuni stati sono pari sotto $`g`$, mentre altri sono dispari. Come nel caso di $`\mathrm{\Omega }`$, si può definire una nuova teoria proiettando gli stati sul settore pari. Questa operazione è il punto di partenza per definire una teoria di stringa sullo spazio di orbifold $`S^1/_2`$. Lo spazio in cui vive la stringa ora è un segmento i cui estremi sono punti fissi dell’azione di $`_2`$, $`X=0,\pi R`$. Per scrivere la funzione di partizione dobbiamo inserire nella traccia sugli stati della funzione di partizione il proiettore
$$P=\frac{\left(1+g\right)}{2}.$$
(5.100)
L’azione di $`g`$ sugli operatori è naturalmente
$$g:\alpha _k\alpha _k,g:\stackrel{~}{\alpha }_k\stackrel{~}{\alpha }_k.$$
(5.101)
mentre su uno stato generico di momento $`n`$ e winding $`w`$, l’azione di g è
$$g\underset{i=1}{\overset{N}{}}\alpha _{k_i}\underset{j=1}{\overset{\overline{N}}{}}\overline{\alpha }_{k_j}|n,w=()^{N+\overline{N}}\underset{i=1}{\overset{N}{}}\alpha _{k_i}\underset{j=1}{\overset{\overline{N}}{}}\overline{\alpha }_{k_j}|n,w.$$
(5.102)
Nel calcolare la traccia sopraviveranno solo gli stati con $`n=w=0`$, e denotando con $`N_k`$ il numero di oscillatori con frequenza $`k`$, si ha
$$\frac{1}{\left(q\overline{q}\right)^{\frac{1}{24}}}\mathrm{tr}\left(q^{N^{}+\frac{\alpha ^{}}{4}p_R^2}g\right)=\frac{1}{\left(q\overline{q}\right)^{\frac{1}{24}}}\underset{k=1}{}\underset{N_k=0}{}\left(q^k\right)^{N_k}=\frac{1}{\left(q\overline{q}\right)^{\frac{1}{24}}}\underset{k=1}{}\frac{1}{1+q^k}.$$
(5.103)
La funzione di partizione nel settore non twistato è, dando ancora una volta per intesi l’integrazione e la sua misura e i gradi di libertà non compatti,
$`𝒯_{untwisted}`$ $`=`$ $`{\displaystyle \frac{1}{2}}𝒯_{S^1}\left(R\right)+{\displaystyle \frac{1}{2}}{\displaystyle \frac{1}{\left(q\overline{q}\right)^{\frac{1}{24}}}}{\displaystyle \frac{1}{_{k=1}\left(1+q^k\right)\left(1+\overline{q}^k\right)}}`$ (5.104)
$`=`$ $`{\displaystyle \frac{1}{2}}𝒯_{S^1}\left(R\right)+\left|{\displaystyle \frac{\eta }{\vartheta _2}}\right|,`$
dove naturalmente si è chiamata $`𝒯_{S^1}`$ la funzioni di partizione della stringa sul cerchio (5.28). Ora occorre completare la funzione di partizione con i settori “twistati” per recuperare l’invarianza conforme. Il settore twistato è quello in cui (stringa di lunghezza $`\pi `$)
$$X\left(\sigma +\pi \right)=X\left(\sigma \right),$$
(5.105)
la condizione di antiperiodicità porta ad una espansione in modi semi-interi:
$$X=x_0+i\sqrt{\frac{\alpha ^{}}{2}}\underset{k}{}\left(\frac{\alpha _{k+1/2}}{k+1/2}e^{2i\left(k+1/2\right)\left(\tau \sigma \right)}+\frac{\stackrel{~}{\alpha }_{k+1/2}}{k+1/2}e^{2i\left(k+1/2\right)\left(\tau +\sigma \right)}\right),$$
(5.106)
dove $`x_0=0`$ o $`\pi R`$. L’assenza di zero modi indica che il momento è nullo e non ci sono winding. L’energia di punto zero è corrisondentemente spostata da $`1/24`$ per un bosone periodico a $`+1/48`$ per un bosone antiperiodico, e quindi l’effetto netto è $`+1/16`$. Le condizioni di massa nel settore twisted sono
$$M^2=\frac{4}{\alpha ^{}}\left(N^{}\frac{15}{16}\right),N^{}=\overline{N}^{}.$$
(5.107)
Calcoliamo il contributo del settore twistato alla funzione di partizione
$`\left(q\overline{q}\right)^{\frac{1}{48}}\mathrm{tr}_{\mathrm{twisted}}\left(q^N^{}\overline{q}^{\overline{N}^{}}{\displaystyle \frac{\left(1+g\right)}{2}}\right)`$ $`=`$ $`\left(q\overline{q}\right)^{\frac{1}{48}}\left[{\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{\left|1q^{k1/2}\right|^2}}+{\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{\left|1+q^{k1/2}\right|^2}}\right]`$ (5.108)
$`=`$ $`\left|{\displaystyle \frac{\eta }{\vartheta _4}}\right|+\left|{\displaystyle \frac{\eta }{\vartheta _3}}\right|.`$
La funzione di partizione dell’orbifold $`S^1/_2`$ è quindi data da
$$𝒯_{S^1/_2}=\frac{1}{2}\left(𝒯_{S^1}\left(R\right)+2\left|\frac{\eta }{\vartheta _2}\right|+2\left|\frac{\eta }{\vartheta _4}\right|+2\left|\frac{\eta }{\vartheta _3}\right|\right),$$
(5.109)
che può essere riscritta, utilizzando l’identità $`\vartheta _2\vartheta _3\vartheta _4=2\eta ^3`$, come
$$𝒯_{S^1/_2}=\frac{1}{2}\left(𝒯_{S^1}\left(R\right)+\frac{\left|\vartheta _3\vartheta _4\right|}{\eta \overline{\eta }}+\frac{\left|\vartheta _2\vartheta _3\right|}{\eta \overline{\eta }}+\frac{\left|\vartheta _2\vartheta _4\right|}{\eta \overline{\eta }}\right).$$
(5.110)
La funzione di partizione è ora invariante modulare, come si può verificare direttamente, o ricordando che è stata costruita come somma di termini con tutte le possibili condizioni al bordo periodiche e antiperiodiche.
#### 5.2.2 Orbifold $`T^4/_2`$ di superstringa
Studiamo ora due modelli in sei dimensioni, la Tipo IIB e la Tipo I su un orbifold $`T^4/_2`$. È bene distinguere i gradi di libertà fermionici che vivono nelle dimensioni compatte da quelli definiti sullo spazio-tempo esteso 6-dimensionale. Per far questo è opportuno decomporre i caratteri di $`SO\left(8\right)`$ in rappresentazioni di $`SO\left(4\right)\times SO\left(4\right)`$, riferendo il primo $`SO\left(4\right)`$ al cono di luce di $`_6`$ ed il secondo alla varietà interna:
$`V_8`$ $`=V_4O_4+O_4V_4,O_8`$ $`=O_4O_4+V_4V_4,`$
$`S_8`$ $`=C_4C_4+S_4S_4,C_8`$ $`=S_4C_4+C_4S_4.`$ (5.111)
Definiamo anche le seguenti combinazioni, che torneranno presto utili:
$`Q_o=V_4O_4C_4C_4,`$ $`Q_v=O_4V_4S_4S_4,`$ (5.112)
$`Q_s=O_4C_4S_4O_4,`$ $`Q_c=V_4S_4C_4V_4.`$
Per avere consistenza fra l’azione di $`_2`$ e la supersimmetria di world-sheet, $`O_4`$ e $`C_4`$ devono essere pari mentre $`V_4`$ e $`S_4`$ dispari sotto $`_2`$ . Quindi $`Q_o`$ e $`Q_c`$ sono gli autovettori positivi e $`Q_v`$ e $`Q_s`$ quelli negativi.
L’ampiezza di toro con una compattificazione su $`T^4`$ è
$$𝒯_{++}=\left|V_8S_8\right|^2\mathrm{\Sigma }_{n,w},$$
(5.113)
dove si sono dati per sottintesi i gradi di libertà non compatti e si è indicata con $`\mathrm{\Sigma }_{n,w}`$, la sommatoria sul reticolo delle varietà interne di metrica $`g`$
$$\mathrm{\Sigma }_{n,w}=\underset{n,w}{}\frac{q^{\frac{\alpha ^{}}{4}p_L^\mathrm{T}g^1p_L}\overline{q}^{\frac{\alpha ^{}}{4}p_R^\mathrm{T}g^1p_R}}{\eta ^4\overline{\eta }^4}.$$
(5.114)
Utilizzando le (5.112) e separando nella sommatoria i modi zero dagli stati più alti in momento e winding l’ampiezza di toro (5.113) diventa
$$𝒯_{++}=\left|Q_oQ_v\right|^2\left[\mathrm{\Sigma }_{n,w}^{}+\frac{1}{\left(\eta \overline{\eta }\right)^4}\right],$$
(5.115)
si è indicato con $`\mathrm{\Sigma }^{}`$ la sommatoria senza gli zero modi. Come si è visto nel caso $`S^1/_2`$, solo gli zero modi vengono proiettati mentre gli altri stati vengono dimezzati.
Per scrivere la funzione di partizione di orbifold si può proiettare l’ampiezza (5.115) con $`_2`$, effettuare una trasformazione S e poi proiettare ancora con $`_2`$. Infatti la prima proiezione pone condizioni antiperiodiche nella direzione ‘temporale’ sul worldsheet mentre la trasformazione $`S`$ e la nuova proiezione $`_2`$ individuano, rispettivamente, il settore con condizioni antiperiodiche nella direzione ‘spaziale’ ed il settore con condizioni antiperiodiche sia nella direzione ‘spaziale’ che ‘temporale’.
Come si è già trovato nel caso $`S^1/_2`$, si ha
$$\frac{1}{\eta ^4}\stackrel{_2}{}\left(\frac{2\eta }{\vartheta _2}\right)^2\stackrel{S}{}4\left(\frac{\eta }{\vartheta _4}\right)^2\stackrel{_2}{}4\left(\frac{\eta }{\vartheta _3}\right)^2.$$
(5.116)
Utilizzando la trasformazione S definita sui caratteri di $`SO\left(4\right)`$
$`S={\displaystyle \frac{1}{2}}\left(\begin{array}{cccc}\hfill 1& \hfill 1& \hfill 1& \hfill 1\\ \hfill 1& \hfill 1& \hfill 1& \hfill 1\\ \hfill 1& \hfill 1& \hfill 1& \hfill 1\\ \hfill 1& \hfill 1& \hfill 1& \hfill 1\end{array}\right),`$ (5.121)
si trova
$$Q_o+Q_v\stackrel{_2}{}Q_oQ_v\stackrel{S}{}Q_s+Q_c\stackrel{_2}{}Q_sQ_c.$$
(5.122)
La funzione di partizione di toro può essere scritta nella forma
$$𝒯=\frac{1}{2}\left[\left|Q_o+Q_v\right|^2\mathrm{\Sigma }_{n,w}+\left|Q_oQ_v\right|^2\left|\frac{2\eta }{\vartheta _2}\right|^4+16\left|Q_s+Q_c\right|^2\left|\frac{\eta }{\vartheta _4}\right|^4+16\left|Q_sQ_c\right|^2\left|\frac{\eta }{\vartheta _3}\right|^4\right],$$
dove il fattore $`16=2^4`$ tiene conto del numero di punti fissi dell’orbifold.
Lo spettro della teoria chiusa è organizzato in multipletti di supersimmetria $`𝒩(2,0)`$ in $`D=6`$. Si ha un multipletto gravitazionale, che contiene il gravitone, cinque 2-forme autoduali e due gravitini sinistri e 21 multipletti tensoriali, 16 dei quali dal settore twisted, contenenti una 2-forma anti-autoduale, cinque scalari e due spinori destri. Questo è l’unico spettro di questo tipo privo di anomalie.
Costruiamo ora i discendenti aperti, inziando come di consueto dalla bottiglia di Klein. L’azione di $`\mathrm{\Omega }`$ scambia $`p_L`$ e $`p_R`$, ma grazie alla identificazione $`XX`$ si ha anche $`p_{L,R}p_{L,R}`$. Questo comporta che nella bottiglia di Klein si propaghino non solo gli stati di momento ma anche quelli di winding. Con le solite regole si scrive
$$𝒦=\frac{1}{4}\left[\left(Q_o+Q_v\right)\left(\underset{n}{}\frac{q^{\frac{\alpha ^{}}{2}n^\mathrm{T}g^1n}}{\eta \left(q\right)^4}+\underset{w}{}\frac{q^{\frac{1}{2\alpha ^{}}w^\mathrm{T}gw}}{\eta \left(q\right)^4}\right)+2\times 16\left(Q_s+Q_c\right)\left(\frac{\eta }{\vartheta _4}\right)^2\right],$$
(5.123)
dove si è risolta in maniera diagonale l’ambiguità nel settore twistato. Nel canale trasverso si trova
$$\stackrel{~}{𝒦}=\frac{2^5}{4}\left[\left(Q_o+Q_v\right)\left(v_4\underset{w}{}\frac{q^{\frac{1}{4\alpha ^{}}w^\mathrm{T}gw}}{\eta \left(i\mathrm{}\right)^4}+\frac{1}{v_4}\underset{n}{}\frac{q^{\frac{\alpha ^{}}{4}n^\mathrm{T}g^1n}}{\eta \left(i\mathrm{}\right)^4}\right)+2\left(Q_oQ_v\right)\left(\frac{2\eta }{\vartheta _2}\right)^2\right],$$
(5.124)
dove $`v_4=\sqrt{\mathrm{det}g/\left(\alpha ^{}\right)^4}`$ è proporzionale al volume delle dimensioni compatte.
Lo spettro di massa nulla si ottiene da quello del toro eliminando gli stati antisimmetrici sotto $`\mathrm{\Omega }`$ nel settore NS-NS e gli stati simmetrici nel settore R-R, mentre i settori misti vengono dimezzati. Lo spettro di massa nulla risultante corrisponde ad un multipletto gravitazionale, contenente il gravitone, una 2-forma autoduale e un gravitino sinistro; ad un multipletto tensoriale, contenente una 2-forma anti-autoduale, uno scalare e uno spinore destro, e a 20 ipermultipletti dei quali 16 dal settore twistato, contenenti 4 scalari e uno spinore destro.
Al livello di massa nulla la bottiglia di Klein trasversa è data da
$$\stackrel{~}{𝒦}_0=\frac{2^5}{4}\left[Q_o\left(\sqrt{v_4}+\frac{1}{\sqrt{v_4}}\right)^2+Q_v\left(\sqrt{v_4}\frac{1}{\sqrt{v_4}}\right)^2\right],$$
(5.125)
da cui si legge il contenuto in termini di $`O`$-piani, la cui tensione e carica di R-R può essere letta da $`Q_o=V_4O_4C_4C_4`$. Si vede subito che, insieme ai soliti $`O9`$-piani, sono presenti anche degli $`O5`$-piani. Quattro T-dualità lungo le direzioni compatte mappano $`O9`$-piani, in $`O5`$-piani e $`v_4`$ in $`1/v_4`$.
Sono possibili anche altre proiezioni di Klein; è possibile infatti proiettare momenti e windings pari e dispari in maniera diversa, ottenendo
$`𝒦^{}`$ $`=`$ $`{\displaystyle \frac{1}{4}}[(Q_o+Q_v)({\displaystyle \underset{n}{}}(1)^n{\displaystyle \frac{q^{\frac{\alpha ^{}}{2}n^\mathrm{T}g^1n}}{\eta ^4}}+{\displaystyle \underset{w}{}}(1)^w{\displaystyle \frac{q^{\frac{1}{2\alpha ^{}}w^\mathrm{T}gw}}{\eta ^4}})`$ (5.126)
$`+2\times (88)(Q_s+Q_c)\left({\displaystyle \frac{\eta }{\vartheta _4}}\right)^2],`$
dove $`\left(1\right)^n`$ e $`\left(1\right)^w`$, stanno ad indicare simbolicamente diverse possibili scelte disponibili per introdurre, in uno o più tori, segni alternati. In questo caso si ha una teoria consistente di sole stringhe chiuse non orientate, dal momento che i segni nel canale trasverso si traducono in shift dei momenti e dei winding che eliminano i modi a massa nulla. Una terza possibilità di grande interesse è
$$𝒦^{\prime \prime }=\frac{1}{4}\left[\left(Q_o+Q_v\right)\left(\underset{n}{}\frac{q^{\frac{\alpha ^{}}{2}n^\mathrm{T}g^1n}}{\eta \left(q\right)^4}+\underset{w}{}\frac{q^{\frac{1}{2\alpha ^{}}w^\mathrm{T}gw}}{\eta \left(q\right)^4}\right)2\times 16\left(Q_s+Q_c\right)\left(\frac{\eta }{\vartheta _4}\right)^2\right],$$
(5.127)
dal momento che porta un settore aperto non supersimmetrico. Si avrà modo di studiare in dettaglio questo modello nel capitolo sulla rottura di supersimmetria.
La scelta più semplice per l’ampiezza di anello risulta essere
$`𝒜`$ $`=`$ $`{\displaystyle \frac{1}{4}}[(Q_o+Q_v)(N^2{\displaystyle \underset{n}{}}{\displaystyle \frac{q^{\frac{\alpha ^{}}{2}n^\mathrm{T}g^1n}}{\eta \left(q\right)^4}}+D^2{\displaystyle \underset{w}{}}{\displaystyle \frac{q^{\frac{1}{2\alpha ^{}}w^\mathrm{T}gw}}{\eta \left(q\right)^4}})`$ (5.128)
$`+\left(R_N^2+R_D^2\right)\left(Q_oQ_v\right)\left({\displaystyle \frac{2\eta }{\vartheta _2}}\right)^2`$
$`+2ND(Q_s+Q_c)\left({\displaystyle \frac{\eta }{\vartheta _4}}\right)^2+2R_NR_D(Q_sQ_c)\left({\displaystyle \frac{\eta }{\vartheta _3}}\right)^2],`$
dove $`N`$ e $`D`$ contano la molteplicità degli estremi della stringa aperta, rispettivamente con condizioni di Neumann e di Dirichlet, e $`R_D`$ e $`R_N`$ definiscono l’azione dell’orbifold sulle cariche di Chan-Paton . Nel canale trasverso si ha
$`\stackrel{~}{𝒜}`$ $`=`$ $`{\displaystyle \frac{2^5}{4}}[(Q_o+Q_v)\left(N^2v_4{\displaystyle \underset{w}{}}{\displaystyle \frac{q^{\frac{1}{4\alpha ^{}}w^\mathrm{T}gw}}{\eta \left(i\mathrm{}\right)^4}}{\displaystyle \frac{D^2}{v_4}}{\displaystyle \underset{n}{}}{\displaystyle \frac{q^{\frac{\alpha ^{}}{4}n^\mathrm{T}g^1n}}{\eta \left(i\mathrm{}\right)^4}}\right)`$ (5.129)
$`+2ND\left(Q_oQ_v\right)\left({\displaystyle \frac{2\eta }{\vartheta _2}}\right)^2`$
$`+16\left(R_N^2+R_D^2\right)\left(Q_s+Q_c\right)\left({\displaystyle \frac{\eta }{\vartheta _4}}\right)^2`$
$`2\times 4R_NR_D(Q_sQ_c)\left({\displaystyle \frac{\eta }{\vartheta _3}}\right)^2],`$
che porta al contributo alla condizione di tadpole
$`\stackrel{~}{𝒜}_0`$ $`=`$ $`{\displaystyle \frac{2^5}{4}}\{Q_o(N\sqrt{v_4}+{\displaystyle \frac{D}{\sqrt{v_4}}})^2+Q_v(N\sqrt{v_4}{\displaystyle \frac{D}{\sqrt{v_4}}})^2`$ (5.130)
$`+`$ $`Q_s[15R_N^2+(R_N4R_D)^2]+Q_c[15R_N^2+(R_N+4R_D)^2]\},`$
in cui si distinguono i contributi delle stringhe aperte con le diverse possibili combinazioni di condizioni al bordo di Neumann e di Dirichlet che corrispondono alle configurazioni $`D9`$-$`D9`$, $`D9`$-$`D5`$ e $`D5`$-$`D9`$, $`D5`$-$`D5`$.
L’ampiezza della striscia di Möbius nel canale trasverso, calcolata come di consueto, porta a
$`\stackrel{~}{}`$ $`=`$ $`{\displaystyle \frac{2}{4}}[(\widehat{Q}_o+\widehat{Q}_v)(Nv_4{\displaystyle \underset{w}{}}{\displaystyle \frac{q^{\frac{1}{4\alpha ^{}}w^\mathrm{T}gw}}{\eta \left(i\mathrm{}\right)^4}}+D{\displaystyle \frac{1}{v_4}}{\displaystyle \underset{n}{}}{\displaystyle \frac{q^{\frac{\alpha ^{}}{4}n^\mathrm{T}g^1n}}{\eta \left(i\mathrm{}\right)^4}})`$ (5.131)
$`+(N+D)(\widehat{Q}_o\widehat{Q}_v)\left({\displaystyle \frac{2\widehat{\eta }}{\widehat{\vartheta }_2}}\right)^2],`$
dove si sono fissati i segni in maniera opportuna per cancellare il tadpole R-R. Il contributo di massa nulla è
$`\stackrel{~}{}_0`$ $`=`$ $`{\displaystyle \frac{2}{4}}[\widehat{Q}_o(\sqrt{v_4}+{\displaystyle \frac{1}{\sqrt{v_4}}})(N\sqrt{v_4}+{\displaystyle \frac{D}{\sqrt{v_4}}})`$ (5.132)
$`+\widehat{Q}_v(\sqrt{v_4}{\displaystyle \frac{1}{\sqrt{v_4}}})(N\sqrt{v_4}{\displaystyle \frac{D}{\sqrt{v_4}}})].`$
Per passare nel canale diretto occorre definire una trasformazione $`P`$ sui caratteri di $`SO\left(4\right)`$, che si trova essere $`P=\mathrm{diag}(\sigma _1,\sigma _1)`$. L’azione di $`P`$ quindi scambia $`\widehat{Q}_v`$ con $`\widehat{Q}_o`$, si ottiene
$``$ $`=`$ $`{\displaystyle \frac{1}{4}}[(\widehat{Q}_o+\widehat{Q}_v)(N{\displaystyle \underset{n}{}}{\displaystyle \frac{q^{\frac{\alpha ^{}}{2}n^\mathrm{T}g^1n}}{\eta \left(q\right)^4}}+D{\displaystyle \underset{w}{}}{\displaystyle \frac{q^{\frac{1}{2\alpha ^{}}w^\mathrm{T}gw}}{\eta \left(q\right)^4}})`$ (5.133)
$`(N+D)(\widehat{Q}_o\widehat{Q}_v)\left({\displaystyle \frac{2\widehat{\eta }}{\widehat{\vartheta }_2}}\right)^2].`$
La condizione di tadpole impone la cancellazione della somma dei termini $`\stackrel{~}{𝒜}_0`$, $`\stackrel{~}{𝒦}_0`$, $`\stackrel{~}{}_0`$. Dal momento che i termini $`R_D`$ e $`R_N`$ compaiono solo in $`\stackrel{~}{𝒜}_0`$, occorre che siano entrambe nulle:
$$R_N=R_D=0,$$
(5.134)
e si deve inoltre avere
$$\left(N\sqrt{v_4}\pm D\frac{1}{\sqrt{v_4}}\right)=32\left(\sqrt{v_4}\pm \frac{1}{\sqrt{v_4}}\right),$$
(5.135)
che fissa
$$N=32,D=32.$$
(5.136)
La corretta parametrizzazione per le molteplicità di Chan-Paton è
$`N`$ $`=`$ $`n+\overline{n},n=\overline{n}=16,`$
$`D`$ $`=`$ $`d+\overline{d},d=\overline{d}=16,`$ (5.137)
che in accordo con (5.134) fissa
$$R_N=i\left(n\overline{n}\right),R_D=i\left(d\overline{d}\right).$$
(5.138)
Con la parametrizzazione data, le ampiezze del settore aperto, per il livello di massa nulla si scrivono
$`𝒜_0`$ $`=`$ $`\left(n\overline{n}+d\overline{d}\right)Q_0+\frac{1}{2}\left(n^2+\overline{n}^2+d^2+\overline{d}^2\right)Q_v+\left(n\overline{d}+\overline{n}d\right)Q_s`$
$`_0`$ $`=`$ $`\frac{1}{2}\left(n+\overline{n}+d+\overline{d}\right)\widehat{Q}_v,`$ (5.139)
da cui si legge lo spettro di bassa energia, che risulta privo di anomalie . $`Q_o`$ porta un multipletto di gauge di $`𝒩=(1,0)`$, un vettore e uno spinore sinistro, nella rappresentazione aggiunta di $`U\left(16\right)_{D9}\times U\left(16\right)_{D5}`$; $`Q_v`$ contribuisce con ipermultipletti nella $`(16\times 15/2,1)`$ e $`(1,16\times 15/2)`$ e nelle loro complesse coniugate; $`Q_s`$ contribuisce con metà di un ipermultipletto nella $`(16,\overline{16})`$ e con il suo coniugato nella $`(\overline{16},16)`$, e pertanto ancora con un ipermultipletto completo.
## Capitolo 6 Rottura di Supersimmetria
La costruzione di modelli di stringa che riproducano a basse energie la fisica conosciuta richiede l’introduzione di opportuni meccanismi di rottura della supersimmetria, dal momento che la degenerazione fra bosoni e fermioni tipica della supersimmetria esatta non è osservata nei processi ordinari.
In teoria delle stringhe esistono differenti possibilità di implementare la rottura di supersimmetria. In questo capitolo si vedranno prima le teorie in dieci dimensioni di Tipo 0, che forniscono esempi di rottura “esplicita” della supersimmetria, e in seguito modelli con rottura “spontanea”, nei quali la simmetria può essere recuperata scegliendo opportunamente un parametro continuo: *Scherk-Schwarz supersymmetry breaking* , *Brane supersymmetry breaking* e modelli in cui la rottura è indotta da deformazioni magnetiche.
### 6.1 Tipo 0
È possibile costruire, a partire dalla forma generale della funzione di partizione (4.99) due modelli in dieci dimensioni invarianti modulari, e non supersimmetrici, detti di Tipo 0 . Lasciando intesi i gradi di libertà bosonici e l’integrazione, le funzioni di partizione corrispondenti sono
$`𝒯_{0A}`$ $`=`$ $`\left|O_8\right|^2+\left|V_8\right|^2+\overline{S}_8C_8+\overline{C}_8S_8,`$
$`𝒯_{0B}`$ $`=`$ $`\left|O_8\right|^2+\left|V_8\right|^2+\left|S_8\right|^2+\left|C_8\right|^2.`$ (6.1)
Ricordando il contenuto dei caratteri coinvolti in termini di particelle non è difficile studiare lo spettro di bassa energia delle due teorie. I due modelli non contengono fermioni, dal momento che essi non hanno settori misti NS-R e R-NS. Nel settore NS-NS il termine $`\left|O_8\right|^2`$ rende le due teorie tachioniche, mentre $`\left|V_8\right|^2`$ porta in entrambi i casi un gravitone, un tensore antisimmetrico e un dilatone ($`G_{\mu \nu },B_{\mu \nu },\varphi `$). Il settore di R-R distingue le due teorie. Nella 0A si trova $`\overline{S}_8C_8`$ e $`\overline{C}_8S_8`$, che portano due vettori $`A_\mu `$ e due 3-forme $`C_{\mu \nu \rho }`$, mentre nella OB $`\left|S_8\right|^2`$ e $`\left|C_8\right|^2`$ danno due scalari, altri due tensori antisimmetrici ed una 4-forma completa $`D_{\mu \nu \rho \sigma }`$.
#### 6.1.1 Discendenti aperti
##### Orientifold della OA
Vediamo la costruzione di orientifold della OA . L’ampiezza di bottiglia di Klein nel canale diretto si ottiene, in base alle regole note, dai soli termini simmetrici sotto lo scambio dei modi sinistri e destri. L’ampiezza nel canale traverso è invece ottenuta tramite una trasformazione S da quella diretta, dopo essere passati al modulo del toro doppiamente ricoprente. Si può quindi scrivere
$$𝒦=\frac{1}{2}\left(O_8+V_8\right),$$
(6.2)
$$\stackrel{~}{𝒦}=\frac{2^5}{2}\left(O_8+V_8\right).$$
(6.3)
La bottiglia di Klein proietta via il tensore antisimmetrico dal settore NS-NS e dimezza il settore R-R.
Ricordando che in $`so\left(8\right)`$ le rappresentazioni sono autoconiugate, nell’anello trasverso si propagheranno solo i termini che compaiono nella (6.1) in forma diagonale, ovvero $`O_8`$ e $`V_8`$. Si possono associare ai due caratteri differenti coefficienti di riflessione, e quindi
$$\stackrel{~}{𝒜}=\frac{2^5}{2}\left[\left(n_b+n_f\right)^2V_8+\left(n_bn_f\right)^2O_8\right].$$
(6.4)
Dopo aver ridefinito il modulo, si ottiene l’ampiezza nel canale diretto con una trasformazione S
$$𝒜=\frac{1}{2}\left[\left(n_b^2+n_f^2\right)\left(O_8+V_8\right)2n_bn_f\left(S_8+C_8\right)\right].$$
(6.5)
L’ampiezza di Möbius trasversa si trova ricordando che i suoi coefficienti dei caratteri sono medie geometriche dei rispettivi coefficienti in $`\stackrel{~}{𝒦}`$ e $`\stackrel{~}{𝒜}`$ moltiplicati per un fattore combinatorio 2
$$\stackrel{~}{}=ϵ\frac{2}{2}\left[\left(n_b+n_f\right)\widehat{V}_8+\left(n_bn_f\right)\widehat{O}_8\right].$$
(6.6)
L’ampiezza di Möbius nel canale diretto si trova con un riscalamento del modulo e una trasformazione P
$$=ϵ\frac{1}{2}\left[\left(n_b+n_f\right)\widehat{V}_8\left(n_bn_f\right)\widehat{O}_8\right].$$
(6.7)
Nelle ampiezze del canale trasverso non compaiono stati di R-R, e l’unico tadpole che compare è quello di NS-NS. Se si decide di imporre la tadopole condition, si trova $`ϵ=1`$ e $`n_b+n_f=32`$, selezionando il gruppo di gauge $`SO\left(n_b\right)\times SO\left(n_f\right)`$. Lo spettro aperto contiene un vettore nell’aggiunta del gruppo, un fermione di Majorana ($`S_8+C_8`$) nella bifondamentale $`(n_b,n_f)`$ e tachioni nella $`(\frac{n_b^2+n_b}{2},1)`$ e nella $`(1,\frac{n_f^2n_f}{2})`$. Se invece non si impone la condizione di tadpole NS-NS, si può anche segliere il segno $`ϵ=+1`$, selezionando il gruppo di gauge $`USp\left(n_b\right)\times USp\left(n_f\right)`$.
##### Orientifold della OB
Nel caso della OB sono possibili tre differenti scelte per l’ampiezza di bottiglia di Klein . La prima scelta simmetrizza tutti i caratteri (considerando una base che tenga conto del segno dei fermioni $`O_8,V_8,S_8,C_8`$)
$$𝒦_1=\frac{1}{2}\left(O_8+V_8S_8C_8\right),$$
(6.8)
e lo spettro chiuso risultante contiene un tachione, il gravitone, e il dilatone nel settore NS-NS ed una coppia di 2-forme nel settore di R-R. Passando al modulo del toro doppiamente ricoprente e facendo una trasformazione modulare si ottiene l’ampiezza nel canale trasverso
$$\stackrel{~}{𝒦}_1=\frac{2^6}{2}V_8.$$
(6.9)
Nell’anello trasverso si propagheranno tutti i caratteri dal momento che sono autoconiugati ($`𝒞=1`$), e compaiono nel toro con il proprio carattere coniugato. Si può quindi scrivere
$`\stackrel{~}{𝒜}_1`$ $`=`$ $`{\displaystyle \frac{2^6}{2}}[(n_o+n_v+n_s+n_c)^2V_8+(n_o+n_vn_sn_c)^2O_8`$ (6.10)
$`(n_o+n_v+n_sn_c)^2S_8(n_o+n_vn_s+n_c)^2C_8],`$
e nel canale diretto si trova
$`𝒜_1`$ $`=`$ $`\frac{1}{2}[(n_o^2+n_v^2+n_s^2+n_c^2)V_8+2(n_on_v+n_sn_c)O_8`$ (6.11)
$`2(n_vn_s+n_on_c)S_82(n_vn_c+n_on_s)C_8].`$
L’ampiezza di anello trovata corrisponde all’*ansatz di Cardy* . Nei casi in cui, nella ampiezza di toro (4.99), $`X_{ij}=𝒞`$, tutti i caratteri si riflettono sui bordi. Si hanno quindi tante condizioni al bordo quanti sono i settori nello spettro, con una corrispondenza uno ad uno tra indici di bordo e indici dei settori nel bulk. Le regole di fusione, ovvero le regole che determinano il prodotto fra caratteri, si scrivono nella forma
$$\left[\chi _i\right]\times \left[\chi _j\right]=\underset{k}{}𝒩_{ij}^k\left[\chi _k\right],$$
(6.12)
dove si è indicato con $`\chi _i`$ il generico carattere. L’ansatz di Cardy consiste nell’utilizzare la matrice delle regole di fusione $`𝒩_{ij}^k`$, espressa in termini della matrice $`S`$ dalle formula di Verlinde
$$𝒩_{ij}{}_{}{}^{k}=\underset{l}{}\frac{S_i^lS_j^lS^{}_l^k}{S_1^l},$$
(6.13)
per definire il contenuto del $`k`$-esimo stato con condizioni al bordo $`i`$ e $`j`$. In questa forma l’espressione dell’ampiezza di anello è
$$𝒜=\frac{1}{2}\underset{i,j,k}{}𝒩_{ij}^kn^in^j\chi _k.$$
(6.14)
Per i caratteri di $`so\left(8\right)`$ ($`O_8,V_8,C_8,S_8`$) l’algebra di fusione dice che un generico carattere fonde con $`V_8`$ dando se stesso, $`V_8`$ è quindi l’identità dell’algebra, mentre la fusione di $`C_8`$ e $`S_8`$ da $`O_8`$. Si ritrova così dalla (6.14) l’ampiezza di anello (6.11). Dalle ampiezze trasverse $`\stackrel{~}{𝒦}_1`$ e $`\stackrel{~}{𝒜}_1`$ si ottiene l’ampiezza di Möbius trasversa
$$\stackrel{~}{}_1=\frac{2}{2}\left(n_o+n_v+n_s+n_c\right)\widehat{V}_8,$$
(6.15)
e da quest’ultima con una trasformazione $`P`$, l’ampiezza nel canale diretto
$$_1=\frac{1}{2}\left(n_o+n_v+n_s+n_c\right)\widehat{V}_8,$$
(6.16)
che è una corretta simmetrizzazione dell’anello. Il segno della Möbius è stato fissato per imporre le condizioni di tadpole che, per i tre settori contenenti modi di massa nulla, $`V_8,S_8`$ e $`C_8`$ risultano essere
$`n_o+n_v+n_s+n_c`$ $`=`$ $`64,`$
$`n_on_vn_s+n_c`$ $`=`$ $`0,`$
$`n_on_v+n_sn_c`$ $`=`$ $`0,`$ (6.17)
e fissano $`n_0=n_v`$ e $`n_s=n_c`$, determinando il gruppo di gauge $`SO\left(n_o\right)\times SO\left(n_v\right)\times SO\left(n_s\right)\times SO\left(n_c\right)`$. Lo spettro di basse energie ha vettori nell’aggiunta, tachioni nelle diverse rappresentazioni bifondamentali $`(n_o,n_v,1,1)`$, $`(1,1,n_s,n_c)`$, fermioni sinistri nella $`(1,n_v,n_s,1)`$ e nella $`(n_o,1,1,n_c)`$, fermioni destri nella $`(1,n_v,1,n_c)`$ e nella $`(n_o,1,n_s,1)`$. Lo spettro è chirale in ragione delle diverse rappresentazioni dei fermioni destri e sinistri, ma è possibile verificare che le condizioni di tadpole RR eliminano tutte le anomalie di gauge.
Vediamo ora le altre due possibili scelte per l’ampiezza di bottiglia di Klein
$`𝒦_2`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left(O_8+V_8+S_8+C_8\right),`$
$`𝒦_3`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left(O_8+V_8+S_8C_8\right),`$ (6.18)
che nel canale trasverso portano
$`\stackrel{~}{𝒦}_2`$ $`=`$ $`{\displaystyle \frac{2^6}{2}}O_8,`$
$`\stackrel{~}{𝒦}_3`$ $`=`$ $`{\displaystyle \frac{2^6}{2}}\left(C_8\right).`$ (6.19)
La seconda proiezione simmetrizza il settore NS-NS dando lo stesso spettro del caso precedente e antisimmetrizza il settore di R-R, dando uno scalare complesso e una 4-forma completa. La terza ampiezza di Klein elimina dallo spettro il tachione ed il tensore antisimmetrico dal settore NS-NS, mentre nel settore R-R proietta via uno scalare e una 4-forma autoduale dagli stati di $`\left|C_8\right|^2`$ ed il tensore antisimmetrico da quelli di $`\left|S_8\right|^2`$. Lo spettro è naturalmente chirale a causa delle diverse proiezioni su $`C_8`$ e $`S_8`$.
Le ampiezze di anello e Möbius compatibili con la seconda proiezione possono essere ottenute da quella del primo modello fondendo i vari termini con $`O_8`$, e infatti non è difficile rendersi conto che questa operazione, nel caso della bottiglia di Klein, porta a derivare il secondo modello dal primo. Si ottengono così le ampiezze nel canale diretto
$`𝒜_2`$ $`=`$ $`\frac{1}{2}[(n_o^2+n_v^2+n_s^2+n_c^2)O_8+2(n_on_v+n_sn_c)V_8`$ (6.20)
$`2(n_vn_s+n_on_c)C_82(n_vn_c+n_on_s)S_8]`$
$`_2`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left(n_o+n_vn_sn_c\right)\widehat{O}_8,`$ (6.21)
che nel canale trasverso diventano
$`\stackrel{~}{𝒜}_2`$ $`=`$ $`{\displaystyle \frac{2^6}{2}}[(n_o+n_v+n_s+n_c)^2V_8+(n_o+n_vn_sn_c)^2O_8`$ (6.22)
$`+(n_on_v+n_sn_c)^2C_8+(n_on_vn_s+n_c)^2S_8]`$
$`\stackrel{~}{}_2`$ $`=`$ $`\pm {\displaystyle \frac{2}{2}}\left(n_o+n_vn_sn_c\right)\widehat{O}_8.`$ (6.23)
Il segno della Möbius rimane indeterminato, dal momento che non viene fissato da nessuna condizione di tadpole. Visto che in $``$ non compare $`V_8`$, possiamo interpretare le cariche in termini di gruppi unitari, richiedendo che $`n_b=n_o`$, $`\overline{n}_b=n_v`$, $`n_f=n_s`$ e $`\overline{n}_f=n_c`$. Il tadpole di R-R, che si trova da $`\stackrel{~}{𝒜}_2`$, porta le condizioni $`n_b=\overline{n}_b`$ e $`n_f=\overline{n}_f`$. Il gruppo di gauge è quindi $`U\left(n_b\right)\times U\left(n_f\right)`$, e lo spettro di bassa energia contiene vettori nell’aggiuta del gruppo, fermioni sinistri di Majorana-Weyl nella $`(1,\overline{n}_b,1,\overline{n}_f)`$ e nella $`(n_b,1,n_f,1)`$, fermioni destri di Majorana-Weyl nella $`(1,\overline{n}_b,n_f,1)`$ e nella $`(n_b,1,1,\overline{n}_f)`$ e tachioni in diverse rappresentazioni destre e sinistre. Lo spettro risulta chirale ma privo di anomalie.
Le rimanenti ampiezze del terzo modello si trovano fondendo i vari termini del primo modello con il carattere $`C_8`$,
$`𝒜_3`$ $`=`$ $`{\displaystyle \frac{1}{2}}[(n_o^2+n_v^2+n_s^2+n_c^2)C_82(n_on_v+n_sn_c)S_8`$ (6.24)
$`+2(n_vn_s+n_on_c)V_8+2(n_vn_c+n_on_s)O_8],`$
$`_3`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left(n_on_vn_s+n_c\right)\widehat{C}_8,`$ (6.25)
che nel canale trasverso divengono
$`\stackrel{~}{𝒜}_3`$ $`=`$ $`{\displaystyle \frac{2^6}{2}}[(n_o+n_v+n_s+n_c)^2V_8(n_o+n_vn_sn_c)^2O_8`$ (6.26)
$`(n_on_vn_s+n_c)^2C_8+(n_on_v+n_sn_c)^2S_8],`$
$`\stackrel{~}{}_3`$ $`=`$ $`{\displaystyle \frac{2}{2}}\left(n_on_vn_s+n_c\right)\widehat{C}_8.`$ (6.27)
Anche in questo caso il gruppo di gauge è unitario, dal momento che in $``$ non compare $`V_8`$. Ponendo $`n_v=n`$, $`n_s=\overline{n}`$, $`n_o=m`$ e $`n_c=\overline{m}`$, la condizione di tadpole R-R su $`S_8`$ fissa $`m=\overline{m}`$ e $`n=\overline{n}`$, mentre quella su $`C_8`$ fissa $`mn=32`$, determinando il gruppo di gauge $`U\left(m\right)\times U\left(n\right)`$. La particolare scelta $`n=0`$ elimina i tachioni dal settore aperto, e il modello ottenuto è noto come $`0^{}B`$. Il gruppo di gauge diventa in realtà $`SU\left(32\right)`$, e dal momento che il fattore $`U\left(1\right)`$ è anomalo e il corrispondente vettore diventa massivo . Lo spettro di massa nulla contiene un vettore nell’aggiunta e un fermione destro nelle rappresentazioni antisimmetriche $`\frac{m\left(m1\right)}{2}`$ e nella $`\frac{\overline{m}\left(\overline{m}1\right)}{2}`$.
È possibile leggere dalle ampiezze dei tre modelli discussi il contenuto in termini di $`O`$-piani e $`D`$-brane. Dal momento che si sono introdotte due differenti cariche di R-R, si hanno due tipi differenti di $`O`$-piani e $`D`$-brane che hanno la stessa tensione ma cariche di R-R opposte (vedi Tabella 6.1). Si vede che le ampiezze trasverse di Klein contengono le seguenti combinazioni di $`O`$-piani
$$\stackrel{~}{𝒦}_1\mathrm{O9}_\pm ^{\left(1\right)}\mathrm{O9}_\pm ^{\left(2\right)}\overline{\mathrm{O9}}_\pm ^{\left(1\right)}\overline{\mathrm{O9}}_\pm ^{\left(2\right)},$$
(6.28)
$$\stackrel{~}{𝒦}_2\mathrm{O9}_{}^{\left(1\right)}\mathrm{O9}_\pm ^{\left(2\right)}\overline{\mathrm{O9}}_{}^{\left(1\right)}\overline{\mathrm{O9}}_\pm ^{\left(2\right)},$$
(6.29)
$$\stackrel{~}{𝒦}_3\mathrm{O9}_{}^{\left(1\right)}\mathrm{O9}_\pm ^{\left(2\right)}\overline{\mathrm{O9}}_\pm ^{\left(1\right)}\overline{\mathrm{O9}}_{}^{\left(2\right)},$$
(6.30)
dove la diversa scelta del segno è possibile modificando il segno della Möbius e non cambiando le condizioni di tadpole nel settore R-R. Da $`\stackrel{~}{𝒜}_1`$ si può identificare direttamente il tipo di $`D`$-brane, ottenendo
$$n_0\overline{\mathrm{D9}}^{\left(1\right)},n_vD9^{\left(1\right)},n_sD9^{\left(2\right)},n_c\overline{\mathrm{D9}}^{\left(2\right)}.$$
(6.31)
Il secondo ed il terzo caso sono meno immediati, dal momento che le loro brane sono sovrapposizioni di quelle del primo modello. Nel secondo modello si vede che, per avere coefficienti positivi in $`\stackrel{~}{𝒜}_2`$ per i caratteri $`S_8`$ e $`C_8`$, occorre assorbire il segno negativo nel quadrato delle cariche. In questo modo $`n_b`$ ed $`n_f`$ devono riferirisi ad oggetti con cariche rispettivamente $`(1,1,e^{i\pi /2},e^{i\pi /2})`$ e $`(1,1,e^{i\pi /2},e^{i\pi /2})`$. Questo nel caso di una scelta di segno $`+`$ per $`\stackrel{~}{}_2`$. Si ottengono queste proprietà combinando con coefficienti complessi $`n_o`$ $`\overline{\mathrm{D9}}^{\left(1\right)}`$ e $`n_v`$ D9<sup>(1)</sup> $`\left(n_o=n_v\right)`$, per dare
$$n_b=\frac{n_oe^{i\pi /4}+n_ve^{i\pi /4}}{\sqrt{2}}\mathrm{and}\overline{n_b}=\frac{n_oe^{i\pi /4}+n_ve^{+i\pi /4}}{\sqrt{2}},$$
(6.32)
e $`n_s`$ D9<sup>(2)</sup> con $`n_c`$ $`\overline{\mathrm{D9}}^{\left(2\right)}`$ $`\left(n_s=n_c\right)`$, ottenendo
$$n_f=\frac{n_se^{i\pi /4}+n_ce^{i\pi /4}}{\sqrt{2}}\mathrm{and}\overline{n_f}=\frac{n_se^{i\pi /4}+n_ce^{+i\pi /4}}{\sqrt{2}}.$$
(6.33)
Nel terzo modello si può vedere che le giuste combinazioni (e le loro coniugate) sono invece
$$n=\frac{n_ve^{i\pi /4}+n_ce^{i\pi /4}}{\sqrt{2}}\mathrm{and}\mathrm{m}=\frac{\mathrm{n}_\mathrm{o}\mathrm{e}^{\mathrm{i}\pi /4}+\mathrm{n}_\mathrm{s}\mathrm{e}^{\mathrm{i}\pi /4}}{\sqrt{2}}.$$
(6.34)
### 6.2 Deformazioni di Scherk-Schwarz
In compattificazioni di teorie di campo supersimmetriche è possibile introdurre shift dei momenti di Kaluza-Klein dei vari campi proporzionali alle loro cariche, producendo differenze di massa fra fermioni e bosoni e rompendo quindi la supersimmetria. Questo meccanismo, detto di Scherk-Schwarz , in teorie delle stringhe si arricchisce della possibilità di introdurre shift non solo nei momenti ma anche nei winding . Per teorie di stringhe chiuse orientate le due possibilità sono legate da una T-dualità e descrivono essenzialmente lo stesso fenomeno.
Costruendo discendenti aperti si ottengono invece risultati molto diversi. Ci si riferisce a questi due fenomeni come *Scherk-Schwarz supersymmetry breaking* ed *M-theory breaking*, dal momento che il secondo può essere collegato al primo via T-dualità lungo l’undicesima coordinata . Il modello di M-theory breaking presenta un interessante aspetto detto *brane supersymmetry*, ovvero le eccitazioni di bassa energia di brane immerse in un bulk non supersimmetrico possono essere supersimmetriche. In questo caso la rottura di supersimmetria viene considerata a meno di correzioni radiative. Mentre la scala di rottura della supersimmetria via deformazioni di Scherk-Schwarz è data dall’inverso del raggio di compattificazione, nel caso di *brane supersymmetry* ci si aspetta contributi radiativi dell’ordine $`\frac{1}{R^2}\frac{1}{M_Pl}`$. L’analisi di questo fenomeno è la motivazione principale degli ultimi capitoli di questa Tesi.
#### 6.2.1 Scherk-Schwarz supersymmetry breaking
Partiamo dalla teoria IIB e consideriamone un orbifold, proiettando lo spettro con i generatori di $`_2`$ dati da $`()^F\delta `$, dove $`F=F_L+F_R`$ conta i fermioni spazio-temporali e l’azione di $`\delta `$ è lo shift $`\delta :XX+\pi R`$ lungo la direzione spaziale compattificata su $`S^1`$ con raggio R. Scrivendo la somma su impulsi e su windings sul cerchio come
$$\mathrm{\Lambda }_{n+a,w+b}=\underset{n,w}{}\frac{q^{\frac{\alpha ^{}}{4}\left(\frac{\left(n+a\right)}{R}+\frac{\left(w+b\right)R}{\alpha ^{}}\right)^2}\overline{q}^{\frac{\alpha ^{}}{4}\left(\frac{\left(n+a\right)}{R}\frac{\left(w+b\right)R}{\alpha ^{}}\right)^2}}{\eta \left(q\right)\eta \left(\overline{q}\right)},$$
(6.35)
la funzione di partizione della IIB è
$$𝒯_{IIB}=\left|V_8S_8\right|^2\mathrm{\Lambda }_{n,w}.$$
(6.36)
Studiamo per passi l’azione dell’orbifold sulla funzione di partizione. Nel settore untwisted, l’azione $`\left(1+\delta \right)/2`$ manda $`\mathrm{\Lambda }_{n,w}\left(\mathrm{\Lambda }_{n,w}+()^m\mathrm{\Lambda }_{n,w}\right)/2`$. Lo shift $`\delta `$ identifica le due metà del cerchio di compattificazione, e dal punto di vista dell’impulso equivale ad avere un raggio R/2, ovvero a sommare sui soli momenti pari mantenendo invariato il raggio (non così per il winding). Per recuperare l’invarianza modulare occorre introdurre anche settori twisted, procedendo come già fatto nel caso dell’orbifold $`S^1/_2`$, e si trova che
$$\mathrm{\Lambda }_{n,w}\stackrel{\delta }{}\mathrm{\Lambda }_{n+\frac{1}{2},w}=()^m\mathrm{\Lambda }_{n,w}\stackrel{S}{}\mathrm{\Lambda }_{n,w+\frac{1}{2}}\stackrel{\delta }{}\mathrm{\Lambda }_{n+\frac{1}{2},w+\frac{1}{2}}=()^m\mathrm{\Lambda }_{n,w+\frac{1}{2}}.$$
(6.37)
L’operatore $`()^F\delta `$ agisce sui caratteri invertendo il segno dei settori fermionici, e anche in questo caso occorre considerare l’azione della trasformazione modulare S (4.85) per ricostruire una funzione di partizione invariante modulare
$$\left|V_8S_8\right|^2\stackrel{\delta }{}\left|V_8+S_8\right|^2\stackrel{S}{}\left|O_8C_8\right|^2\stackrel{\delta }{}\left|O_8+C_8\right|^2.$$
(6.38)
La funzione di partizione invariante modulare corrispondente alla proiezione $`()^F\delta `$ è quindi
$`𝒯_{\mathrm{KK}}`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left[\right|V_8S_8|^2\mathrm{\Lambda }_{n,w}+|V_8+S_8|^2(1)^n\mathrm{\Lambda }_{n,w}`$ (6.39)
$`+|O_8C_8|^2\mathrm{\Lambda }_{n,w+\frac{1}{2}}+|O_8+C_8|^2(1)^n\mathrm{\Lambda }_{n,w+\frac{1}{2}}],`$
che può essere riscritta nella forma
$`𝒯_{\mathrm{KK}}`$ $`=`$ $`\left(V_8\overline{V}_8+S_8\overline{S}_8\right)\mathrm{\Lambda }_{2n,w}+\left(O_8\overline{O}_8+C_8\overline{C}_8\right)\mathrm{\Lambda }_{2n,w+\frac{1}{2}}`$ (6.40)
$`\left(V_8\overline{S}_8+S_8\overline{V}_8\right)\mathrm{\Lambda }_{2n+1,w}\left(O_8\overline{C}_8+C_8\overline{O}_8\right)\mathrm{\Lambda }_{2n+1,w+\frac{1}{2}}.`$
Nel limite di decompattificazione si ritrova la funzione di partizione della $`IIB`$., mentre per raggi piccoli ($`R𝒪\left(\sqrt{\alpha ^{}}\right)`$) si sviluppano instabilità tachioniche. Per $`n=w=0`$ il carattere $`O_8`$, che parte con $`h=1/2`$, è moltiplicato per un fattore $`q^{\frac{\alpha ^{}}{4}\left(\frac{R}{2\alpha ^{}}\right)^2}`$. Si ha quindi un tachione nello spettro per $`\frac{1}{2}+\frac{R^2}{16\alpha ^{}}<0`$ ovvero $`R<2\sqrt{2\alpha ^{}}`$. Nella discussione che segue ci limiteremo però al caso di spettro non tachionico.
Studiamo i discendenti aperti. Nella bottiglia di Klein si propagano solo i settori di impulso nullo, e indicando le somme sui momenti e sugli impulsi come
$$P_n\left(q\right)=\underset{n}{}\frac{q^{\frac{\alpha ^{}}{2}n^\mathrm{T}g^1n}}{\eta \left(q\right)^4},W_w\left(q\right)=\underset{w}{}\frac{q^{\frac{1}{2\alpha ^{}}w^\mathrm{T}gw}}{\eta \left(q\right)^4},$$
(6.41)
l’ampiezza si scrive
$$𝒦_{\mathrm{KK}}=\frac{1}{2}\left(V_8S_8\right)P_{2n},$$
(6.42)
che nel canale trasverso diventa
$$\stackrel{~}{𝒦}_{\mathrm{KK}}=\frac{2^5}{4}v\left(V_8S_8\right)W_w,$$
(6.43)
dove $`v=R/\sqrt{\alpha ^{}}`$. Nel canale trasverso dell’anello si propagano solo windings, e quindi
$`\stackrel{~}{𝒜}_{\mathrm{KK}}`$ $`=`$ $`{\displaystyle \frac{2^5}{4}}v\{[(n_1+n_2+n_3+n_4)^2V_8(n_1+n_2n_3n_4)^2S_8]W_w`$
$`+[(n_1n_2+n_3n_4)^2O_8(n_1n_2n_3+n_4)^2C_8]W_{w+\frac{1}{2}}\},`$
dove si sono parametrizzati come di solito i coefficienti di riflessione dei caratteri. Dai segni relativi degli $`n_i`$ si vede che $`n_1`$ e $`n_2`$ contano il numero di D9-brane mentre $`n_3`$ e $`n_4`$ contano le $`\overline{\mathrm{D9}}`$-brane. Una trasformazione modulare S e una risommazione di Poisson danno l’ampiezza nel canale diretto
$`𝒜_{\mathrm{KK}}`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left(n_1^2+n_2^2+n_3^2+n_4^2\right)\left[V_8P_{2n}S_8P_{2n+1}\right]+`$ (6.45)
$`+\left(n_1n_2+n_3n_4\right)\left[V_8P_{2n+1}S_8P_{2n}\right]`$
$`+\left(n_1n_3+n_2n_4\right)\left[O_8P_{2n}C_8P_{2n+1}\right]`$
$`+\left(n_1n_4+n_2n_3\right)\left[O_8P_{2n+1}C_8P_{2n}\right].`$
Infine da $`\stackrel{~}{𝒜}_{\mathrm{KK}}`$ e $`\stackrel{~}{𝒦}_{\mathrm{KK}}`$ si trova
$$\stackrel{~}{}_{\mathrm{KK}}=\frac{v}{2}\left[\left(n_1+n_2+n_3+n_4\right)\widehat{V}_8W_w\left(n_1+n_2n_3n_4\right)\widehat{S}_8\left(1\right)^wW_w\right],$$
(6.46)
dove i segni $`()^w`$ per la sommatoria sui winding davanti a $`\widehat{S}_8`$ sono stati scelti per avere nel canale diretto un termine del tipo $`P_{2n+1}\widehat{S}_8`$. Si trova quindi
$$_{\mathrm{KK}}=\frac{1}{2}\left(n_1+n_2+n_3+n_4\right)\widehat{V}_8P_{2n}+\frac{1}{2}\left(n_1+n_2n_3n_4\right)\widehat{S}_8P_{2n+1},$$
(6.47)
un risultato consistente con l’anello, dove i termini misti nelle cariche corrispondono a contributi orientati che, pertanto, non vengono proiettati. Le condizioni di tadpole sono
NS-NS: $`n_1+n_2+n_3+n_4=`$ $`32,`$ (6.48)
R-R: $`n_1+n_2n_3n_4=`$ $`32.`$ (6.49)
La condizione di tadpole R-R fissa il numero totale di brane. Imponendo anche la condizione di tadpole NS-NS vengono rimosse le anti-brane $`\left(n_3=n_4=0\right)`$, e si ottiene in questo modo la famiglia di gruppi di gauge $`SO\left(n_1\right)\times SO\left(32n_1\right)`$, con vettori nell’aggiunta e spinori chirali nella bi-fondamentale. Nel settore chiuso la supersimmetria è rotta alla scala di compattificazione e lo stesso avviene nel settore aperto.
#### 6.2.2 M-theory breaking e “brane supersymmetry”
Studiamo ora il caso in cui ci sia uno shift sui winding. Ci si aspetta che, rispetto alle eccitazioni di bassa energia nella direzione ortogonale alle D8-brane (T-duali delle D9-brane nel primo modello), lo shift non produca effetti e che quindi la supersimmetria sia preservata nel settore aperto. In realtà quello che succede è che la supersimmetria viene rotta nelle eccitazioni massive e, attraverso correzioni radiative, anche nel settore aperto di massa nulla.
L’ampiezza di toro della IIB con shift nei winding può essere ricavata semplicemente scambiando momenti e winding nella (6.40), e quindi si ottiene
$`𝒯_\mathrm{W}`$ $`=`$ $`\left(V_8\overline{V}_8+S_8\overline{S}_8\right)\mathrm{\Lambda }_{n,2w}+\left(O_8\overline{O}_8+C_8\overline{C}_8\right)\mathrm{\Lambda }_{n+\frac{1}{2},2w}`$ (6.50)
$`\left(V_8\overline{S}_8+S_8\overline{V}_8\right)\mathrm{\Lambda }_{n,2w+1}\left(O_8\overline{C}_8+C_8\overline{O}_8\right)\mathrm{\Lambda }_{n+\frac{1}{2},2w+1}.`$
Anche in questo caso per $`R\sqrt{\alpha ^{}}`$ si sviluppano instabilità tachioniche, mentre nel limite di riduzione dimensionale $`R0`$ si ritrova lo spettro supersimmetrico. Anche in questo caso ci limiteremo a trattare il caso non tachionico. Nell’ampiezza di bottiglia di Klein nel canale diretto si propagano solo i modi con $`w=0`$
$$𝒦_\mathrm{W}=\frac{1}{2}\left(V_8S_8\right)P_n+\frac{1}{2}\left(O_8C_8\right)P_{n+\frac{1}{2}},$$
(6.51)
e da questa si trova l’ampiezza nel canale trasverso
$$\stackrel{~}{𝒦}_\mathrm{W}=\frac{2^5}{2}\mathrm{\hspace{0.17em}2}v\left(V_8W_{4w}S_8W_{4w+2}\right).$$
(6.52)
Si vede che nel canale trasverso gli unici modi di massa nulla vengono dal settore NS-NS, e ci si aspetta per questo che la carica di R-R totale sia nulla e che il modello contenga $`O9`$-piani e $`\overline{\mathrm{O9}}`$-piani. Al contrario nell’anello trasverso si propagheranno solo stati di momento nullo, che sono associati ai caratteri $`V_8`$ e $`S_8`$. Parametrizzando i differenti coefficienti di riflessione si ottiene
$`\stackrel{~}{𝒜}_\mathrm{W}`$ $`=`$ $`{\displaystyle \frac{2^5}{2}}\mathrm{\hspace{0.17em}2}v\{[(n_1+n_2+n_3+n_4)^2V_8(n_1+n_2n_3n_4)^2S_8]W_{4w}`$
$`+[(n_1n_2+n_3n_4)^2V_8(n_1n_2n_3+n_4)^2S_8]W_{4w+2}\},`$
dove si è fattorizzata la somma sui winding nella forma $`W_{2w}=W_{4w}+W_{4w+2}`$ per poter poi agevolmente comparare $`\stackrel{~}{𝒜}_\mathrm{W}`$ con $`\stackrel{~}{𝒦}_\mathrm{W}`$. Anche in questo caso $`n_1`$ e $`n_2`$ contano il numero di D9-brane, mentre $`n_3`$ e $`n_4`$ contano il numero di $`\overline{\mathrm{D9}}`$-brane. Come di consueto l’ampiezza nel canale diretto si ottiene tramite una trasformazione modulare S
$$𝒜_\mathrm{W}=\frac{1}{2}\left(n_1^2+n_4^2\right)\left(V_8S_8\right)\left(P_n+P_{n+\frac{1}{2}}\right)+n_1n_4\left(O_8C_8\right)\left(P_{n+\frac{1}{4}}+P_{n+\frac{3}{4}}\right),$$
(6.54)
e a questo punto non è difficile ricavare l’ampiezza trasversa del nastro di Möbius da $`\stackrel{~}{𝒦}_\mathrm{W}`$ e $`\stackrel{~}{𝒜}_\mathrm{W}`$, che risulta essere
$$\stackrel{~}{}_\mathrm{W}=2v\left[\left(n_1+n_2+n_3+n_4\right)\widehat{V}_8W_{4w}\left(n_1n_2n_3+n_4\right)\widehat{S}_8W_{4w+2}\right],$$
(6.55)
e da questa, con una trasformazione modulare P,
$$_\mathrm{W}=\frac{1}{2}\left(n_1+n_4\right)\left[\left(\widehat{V}_8\widehat{S}_8\right)P_n+\left(\widehat{V}_8+\widehat{S}_8\right)P_{n+\frac{1}{2}}\right].$$
(6.56)
Le condizioni di tadpole risultano essere
$$\text{NS-NS:}n_1+n_2+n_3+n_4=32,\text{R-R:}n_1+n_2=n_3+n_4.$$
(6.57)
Nel limite di riduzione dimensionale $`R0`$, si sviluppano ulteriori tadpole dovuti al fatto che in questo limite gli stati di winding in $`W_{4n+2}`$ diventano a massa nulla. Le nuove condizioni sono
$$\text{NS-NS:}n_1n_2+n_3n_4=0,\text{R-R:}n_1n_2n_3+n_4=32.$$
(6.58)
Tutte le condizioni sono risolte per $`n_1=n_4=16`$ e $`n_2=n_3=0`$, fissando così il gruppo di gauge a $`SO\left(16\right)\times SO\left(16\right)`$.
Dalla somma di $`𝒜_\mathrm{W}`$ e $`_\mathrm{W}`$ si vede che nel settore aperto lo spettro aperto di bassa energia è supersimmetrico, dal momento che contiene uno spinore e un vettore nella rappresentazione aggiunta del gruppo. La supersimmetria invece è rotta nel settore aperto massivo e nel bulk, dove lo shift nei winding alza in massa il gravitino.
### 6.3 Brane supersymmetry breaking
In modelli con compattificazioni e proiezioni $`_2`$ si hanno due possibilità per rompere la supersimmetria. La prima consiste nell’invertire tensione e carica dell’O9-piano e dell’O5-piano, ottenendo l’ampiezza standard di bottiglia di Klein, e rompendo la supersimmetria nel settore aperto ma non in quello chiuso. La seconda possibilità consiste nell’invertire carica e tensione solo di un tipo di $`O`$-piano, ad esempio l’O5-piano. Si ottiene in tal modo una diversa proiezione di Klein e si devono introdurre le corrisponenti D5-antibrane. La supersimmetria, preservata dalle D9-brane, viene così rotta alla scala di stringa.
È possibile studiare questo fenomeno in un modello relativamente semplice, l’orbifold $`T^4/_2`$. Si è già accennato alla possibile proiezione di Klein
$$𝒦=\frac{1}{4}\left[\left(Q_o+Q_v\right)\left(P_m+W_n\right)2\times 16\left(Q_s+Q_c\right)\left(\frac{\eta }{\vartheta _4}\right)^2\right].$$
(6.59)
Al livello di massa nulla, lo spettro di stringa chiusa coincide nel settore untwisted con quello del modello supersimmetrico studiato, mentre nel settore twisted il settore NS-NS è antisimmetrizzato e quello R-R è simmetrizzato. Si hanno cioè 16 scalari NS-NS, 16 2-forme anti-autoduali e 16 spinori destri. Lo spettro ha ancora supersimmetria $`𝒩=(1,0)`$ e contiene il solito multipletto del gravitone, 17 multipletti tensoriali, 16 dei quali dal settore twisted e 4 iper-multipletti.
Nel canale trasverso si trova
$$\stackrel{~}{𝒦}=\frac{2^5}{4}\left[\left(Q_o+Q_v\right)\left(v_4W_n^{\left(e\right)}+\frac{1}{v_4}P_m^{\left(e\right)}\right)2\left(Q_oQ_v\right)\left(\frac{2\eta }{\vartheta _2}\right)^2\right],$$
(6.60)
il cui limite di bassa energia è
$$\stackrel{~}{𝒦}_0=\frac{2^5}{4}\left[Q_o\left(\sqrt{v_4}\frac{1}{\sqrt{v_4}}\right)^2+Q_v\left(\sqrt{v_4}+\frac{1}{\sqrt{v_4}}\right)^2\right].$$
(6.61)
Rispetto al modello supersimmetrico precedentemente studiato si vede che c’è un segno relativo fra $`\sqrt{v_4}`$ e $`1/\sqrt{v_4}`$ nel coefficiente associato a $`Q_o`$. Sviluppando il quadrato, questo introduce un segno negativo davanti al termine misto e indica la presenza della configurazione annuciata con $`O9_+`$ e $`O5_{}`$ piani (o in alternativa la configurazione T-duale con $`O9_{}`$ e $`O5_+`$ piani). Nel primo caso ci si attende che nel modello compaiano anche $`\overline{\mathrm{D5}}`$-brane che neutralizzino la carica di R-R, e quindi che nell’ampiezza di anello trasverso compaia un segno relativo fra le cariche di R-R delle $`D9`$-brane e delle $`\overline{\mathrm{D5}}`$ brane.
L’ampiezza d’anello trasverso corrispondente è
$`\stackrel{~}{𝒜}`$ $`=`$ $`{\displaystyle \frac{2^5}{4}}[(Q_o+Q_v)(N^2v_4W_n+{\displaystyle \frac{D^2}{v_4}}P_m)+16(R_N^2+R_D^2)(Q_s+Q_c)\left({\displaystyle \frac{\eta }{\vartheta _4}}\right)^2`$ (6.62)
$`+2ND\left(V_4O_4+C_4C_4O_4V_4S_4S_4\right)\left({\displaystyle \frac{2\eta }{\vartheta _2}}\right)^2`$
$`2\times 4R_NR_D(O_4C_4+S_4O_4V_4S_4C_4V_4)\left({\displaystyle \frac{\eta }{\vartheta _3}}\right)^2],`$
che nel limite di basse energie diventa
$$\stackrel{~}{𝒜}_0=\frac{2^5}{4}\left[\left(V_4O_4S_4S_4\right)\left(N\sqrt{v_4}+\frac{D}{\sqrt{v_4}}\right)^2+\left(O_4V_4C_4C_4\right)\left(N\sqrt{v_4}\frac{D}{\sqrt{v_4}}\right)^2\right].$$
(6.63)
Si vede che il prodotto delle tensioni, che è legato al coefficiente di $`O_4V_4`$, è sempre positivo mentre le cariche di R-R, che si leggono dal coefficiente di $`C_4C_4`$, sono opposte. Nel canale diretto l’ampiezza di anello diventa
$`𝒜`$ $`=`$ $`{\displaystyle \frac{1}{4}}[(Q_o+Q_v)(N^2P_m+D^2W_n)+(R_N^2+R_D^2)(Q_oQ_v)\left({\displaystyle \frac{2\eta }{\vartheta _2}}\right)^2`$ (6.64)
$`+2ND\left(O_4S_4C_4O_4+V_4C_4S_4V_4\right)\left({\displaystyle \frac{\eta }{\vartheta _4}}\right)^2`$
$`+2R_NR_D(O_4S_4C_4O_4+V_4C_4+S_4V_4)\left({\displaystyle \frac{\eta }{\vartheta _3}}\right)^2].`$
Infine l’ampiezza di Möbius trasversa si trova essere
$`\stackrel{~}{}`$ $`=`$ $`{\displaystyle \frac{1}{2}}[v_4NW_n^{\left(e\right)}(\widehat{V}_4\widehat{O}_4+\widehat{O}_4\widehat{V}_4\widehat{S}_4\widehat{S}_4\widehat{C}_4\widehat{C}_4)`$ (6.65)
$`+{\displaystyle \frac{1}{v_4}}DP_m^{\left(e\right)}\left(\widehat{V}_4\widehat{O}_4\widehat{O}_4\widehat{V}_4\widehat{S}_4\widehat{S}_4\widehat{C}_4\widehat{C}_4\right)`$
$`+N\left(\widehat{V}_4\widehat{O}_4+\widehat{O}_4\widehat{V}_4\widehat{S}_4\widehat{S}_4+\widehat{C}_4\widehat{C}_4\right)\left({\displaystyle \frac{2\widehat{\eta }}{\widehat{\vartheta }_2}}\right)^2`$
$`+D(\widehat{V}_4\widehat{O}_4\widehat{O}_4\widehat{V}_4\widehat{S}_4\widehat{S}_4+\widehat{C}_4\widehat{C}_4)\left({\displaystyle \frac{2\widehat{\eta }}{\widehat{\vartheta }_2}}\right)^2],`$
che nel limite di basse energie diventa
$`\stackrel{~}{}_0`$ $`=`$ $`{\displaystyle \frac{1}{2}}[\widehat{V}_4\widehat{O}_4(\sqrt{v_4}{\displaystyle \frac{1}{\sqrt{v_4}}})(N\sqrt{v_4}+{\displaystyle \frac{D}{\sqrt{v_4}}})`$ (6.66)
$`+\widehat{O}_4\widehat{V}_4\left(\sqrt{v_4}+{\displaystyle \frac{1}{\sqrt{v_4}}}\right)\left(N\sqrt{v_4}{\displaystyle \frac{D}{\sqrt{v_4}}}\right)`$
$`\widehat{C}_4\widehat{C}_4\left(\sqrt{v_4}{\displaystyle \frac{1}{\sqrt{v_4}}}\right)\left(N\sqrt{v_4}{\displaystyle \frac{D}{\sqrt{v_4}}}\right)`$
$`\widehat{S}_4\widehat{S}_4(\sqrt{v_4}+{\displaystyle \frac{1}{\sqrt{v_4}}})(N\sqrt{v_4}+{\displaystyle \frac{D}{\sqrt{v_4}}})],`$
ed infine nel canale diretto, tramite una trasformazione P, si trova
$``$ $`=`$ $`{\displaystyle \frac{1}{4}}[NP_m(\widehat{O}_4\widehat{V}_4+\widehat{V}_4\widehat{O}_4\widehat{S}_4\widehat{S}_4\widehat{C}_4\widehat{C}_4)`$ (6.67)
$`DW_n\left(\widehat{O}_4\widehat{V}_4+\widehat{V}_4\widehat{O}_4+\widehat{S}_4\widehat{S}_4+\widehat{C}_4\widehat{C}_4\right)`$
$`N\left(\widehat{O}_4\widehat{V}_4\widehat{V}_4\widehat{O}_4\widehat{S}_4\widehat{S}_4+\widehat{C}_4\widehat{C}_4\right)\left({\displaystyle \frac{2\widehat{\eta }}{\widehat{\vartheta }_2}}\right)^2`$
$`+D(\widehat{O}_4\widehat{V}_4\widehat{V}_4\widehat{O}_4+\widehat{S}_4\widehat{S}_4\widehat{C}_4\widehat{C}_4)\left({\displaystyle \frac{2\widehat{\eta }}{\widehat{\vartheta }_2}}\right)^2].`$
Dal momento che nell’ampiezza di Möbius compare, nello spettro di basse energie, un vettore di massa nulla, si hanno cariche reali di Chan-Paton che vengono parametrizzate nella forma
$`N=n_1+n_2,D=d_1+d_2,`$
$`R_N=n_1n_2,R_D=d_1d_2,`$ (6.68)
e in questo modo lo spettro di massa nulla nel settore aperto risulta
$`𝒜_0+_0`$ $`=`$ $`{\displaystyle \frac{n_1\left(n_11\right)+n_2\left(n_21\right)+d_1\left(d_1+1\right)+d_2\left(d_2+1\right)}{2}}V_4O_4`$ (6.69)
$`{\displaystyle \frac{n_1\left(n_11\right)+n_2\left(n_21\right)+d_1\left(d_11\right)+d_2\left(d_21\right)}{2}}C_4C_4`$
$`+\left(n_1n_2+d_1d_2\right)\left(O_4V_4S_4S_4\right)+\left(n_1d_2+n_2d_1\right)O_4S_4`$
$`\left(n_1d_1+n_2d_2\right)C_4O_4.`$
Le condizioni di tadpole R-R, che si trovano sommando i contributi di massa nulla di $`\stackrel{~}{𝒜}_0`$, $`\stackrel{~}{𝒦}_0`$ e $`\stackrel{~}{}_0`$, fissano
$$n_1=n_2=d_1=d_2=16.$$
(6.70)
e quindi
$$N=D=32,R_N=R_D=0,$$
(6.71)
e quindi il gruppo di gauge è $`\left[SO\left(16\right)\times SO\left(16\right)\right]_9\times \left[USp\left(16\right)\times USp\left(16\right)\right]_5`$, dove i primi due fattori si riferiscono alle $`D9`$-brane e i secondi alle $`\overline{\mathrm{D5}}`$ brane.
Il settore con condizioni al bordo NN ha uno spettro di massa nulla ancora supersimmetrico, con un multipletto di gauge nella rappresentazione aggiunta di $`SO\left(16\right)\times SO\left(16\right)`$ ed un ipermultipletto nella $`(16,16,1,1)`$. Al contrario, il settore DD è chiaramente non supersimmetrico, dal momento che contiene vettori nell’aggiunta del gruppo $`USp\left(16\right)\times USp\left(16\right)`$, 4 scalari nella $`(1,1,16,16)`$, fermioni di Weyl destri nella $`(1,1,120,1)`$ e nella $`(1,1,1,120)`$, e fermioni di Weyl sinistri nella $`(1,1,16,16)`$. Anche i settori con condizioni al bordo miste ND sono non supersimmetrici e comprendono un doppietto di scalari nella $`(16,1,1,16)`$ e nella $`(1,16,16,1)`$, e fermioni di Majorana-Weyl nella $`(16,1,16,1)`$ e nella $`(1,16,1,16)`$. Questi fermioni sono peculiari dello spazio-tempo in sei dimensioni dove i fermioni di Weyl possono essere soggetti ad una condizione addizionale di Majorana, implementata con una matrice di coniugazione di carica pseudoreale.
La rottura di supersimmetria lascia non cancellati i tadpole nel settore NS-NS, che risultano essere
$$\left[\left(N32\right)\sqrt{v}_4+\frac{D+32}{\sqrt{v}_4}\right]^2V_4O_4+\left[\left(N32\right)\sqrt{v}_4\frac{D+32}{\sqrt{v}_4}\right]^2O_4V_4.$$
(6.72)
Per questa ragione si genera un potenziale del dilatone localizzato sulle $`\overline{\mathrm{D5}}`$ brane. I coefficienti di $`V_4O_4`$ e di $`O_4V_4`$ sono proporzionali ai quadrati delle funzioni ad un punto sui bordi e sui crosscap. Dal punto di vista dell’azione di basse energie i due contributi NS-NS hanno origine dai termini
$$\mathrm{\Delta }S\left(N32\right)\sqrt{v_4}d^6x\sqrt{g}e^{\phi _6}+\frac{D+32}{\sqrt{v_4}}d^6x\sqrt{g}e^{\phi _6},$$
(6.73)
e in particolare dalle variazioni del dilatone in sei dimensioni $`\phi _6`$ e del volume interno $`v_4`$ rispetto ai loro valori di background.
$$\phi _6\phi _6\delta \phi _6,\sqrt{v}_4\left(1+\delta h\right)\sqrt{v}_4.$$
(6.74)
Nell’azione scritta il primo termine si riferisce al contributo del sistema di D9-brane e dell’O9-piano, e, come atteso, si cancella in conseguenza della condizione di tadpole nel settoe R-R, che fissa $`N=32`$ preservando la supersimmetria. Al contrario, il secondo termine è dovuto ai contributi delle $`\overline{\mathrm{D5}}`$ brane e dell’O5<sub>-</sub>-piano che non viene cancellato.
### 6.4 Deformazioni magnetiche
L’introduzione di campi magnetici interni (nello spazio tempo compatto) da luogo genericamente alla rottura della supersimmetria . La ragione risiede nel fatto che gli estremi carichi delle stringhe aperte si accoppiano al campo, producendo shift nelle masse degli stati di stringa differenti a seconda dei loro momenti di dipolo.
#### 6.4.1 Campi abeliani e stringhe bosoniche aperte
Un campo abeliano che viva in un sottospazio (una $`D`$-brana in uno spazio esteso o compattificato), che indichiamo con gli indici $`m,n=0,\mathrm{}d`$ dove $`d`$ è la dimensione del sottospazio, si accoppia alle cariche di Chan-Paton di una stringa aperta. L’azione di stringa (consideriamo il caso bosonico) sarà modificata dall’aggiunta di termini al bordo di accoppiamento minimale
$$S=𝑑\tau 𝑑\sigma (\dot{X},X^{})+q_\mathrm{R}𝑑\tau A_m_\tau X^m|_{\sigma =\pi }+q_\mathrm{L}𝑑\tau A_m_\tau X^m|_{\sigma =0}.$$
(6.75)
Limitiamoci a considerare un campo costante che possiamo scrivere nella forma
$$A_m=\frac{1}{2}F_{mn}X^n,$$
(6.76)
per cui l’azione diventa
$`S={\displaystyle 𝑑\tau 𝑑\sigma (\dot{X},X^{})}`$ $`+`$ $`{\displaystyle \frac{1}{2}}q_\mathrm{R}{\displaystyle 𝑑\tau F_{mn}X^n_\tau X^m}|_{\sigma =\pi }`$ (6.77)
$`+`$ $`{\displaystyle \frac{1}{2}}q_\mathrm{L}{\displaystyle 𝑑\tau F_{mn}X^n_\tau X^m}|_{\sigma =0}.`$
Le equazioni del moto che si ottengono dalla lagrangiana risultano invariate dal momento che non vengono influenzate dalla presenza dei termini di bordo. Cambiano, invece, le condizioni al bordo, che diventano
$$_\sigma X_m\pm 2\pi \alpha ^{}q_{\mathrm{L},\mathrm{R}}F_{mn}_\tau X^n=0,\sigma =0,\pi ,$$
(6.78)
e sono quindi intermedie tra condizioni di Neumann e di Dirichlet. Consideriamo ora in caso di un campo magnetico con tutte le componenti nulle ad eccezione $`F_{23}=F_{32}H`$. Le coordinate generiche $`\mu `$ avranno condizioni al bordo di Neumann, mentre $`X^2`$ e $`X^3`$ devono soddisfare
$`_\sigma X_2\pm 2\pi \alpha ^{}q_{\mathrm{L},\mathrm{R}}H_\tau X^3`$ $`=`$ $`0,\sigma =0,\pi `$
$`_\sigma X_32\pi \alpha ^{}q_{\mathrm{L},\mathrm{R}}H_\tau X^2`$ $`=`$ $`0,\sigma =0,\pi .`$ (6.79)
Se la coordinata $`X^3`$ è compattificata, è possibile effettuare una trasformazione di T-dualità modificando le condizioni al bordo con il risultato che $`_\sigma _\tau `$. In questo modo si ottiene
$`_\sigma \left(X^2\pm 2\pi \alpha ^{}q_{\mathrm{L},\mathrm{R}}HX^{3,T}\right)`$ $`=`$ $`0,\sigma =0,\pi ,`$
$`_\tau \left(X^{3,T}2\pi \alpha ^{}q_{\mathrm{L},\mathrm{R}}HX^2\right)`$ $`=`$ $`0,\sigma =0,\pi .`$ (6.80)
Introducendo gli angoli definiti da $`\mathrm{tan}\left(\theta _{L,R}\right)=2\pi \alpha ^{}q_{L,R}H`$, la (6.4.1) si scrive
$`_\sigma \left(\mathrm{cos}\left(\theta _{L,R}\right)X^2\mathrm{sin}\left(\theta _{L,R}\right)X^{3,T}\right)`$ $`=`$ $`0,\sigma =0,\pi ,`$
$`_\tau \left(\mathrm{sin}\left(\theta _{L,R}\right)X^2\pm \mathrm{cos}\left(\theta _{L,R}\right)X^{3,T}\right)`$ $`=`$ $`0,\sigma =0,\pi ,`$ (6.81)
che suggerisce un’interpretazione geometrica: l’introduzione di campi magnetici ruota le $`D`$-brane nella picture T-duale. Nelle nuove coordinate
$`Y^1=\mathrm{cos}\left(\theta _{L,R}\right)X^2\mathrm{sin}\left(\theta _{L,R}\right)X^{3,T},`$ (6.82)
$`Y^2=\mathrm{sin}\left(\theta _{L,R}\right)X^2\pm \mathrm{cos}\left(\theta _{L,R}\right)X^{3,T},`$ (6.83)
si hanno condizioni di Neumann nelle direzioni tangenziali e di Dirchlet in quelle ortogonali alle brane. Nel caso di stringhe con cariche uguali ed opposte agli estremi la rotazione delle brane non ha effetti sulla massa, ma se le cariche differenti, le brane su cui sono posti gli estremi vengono ruotate di angoli differenti e i modi delle stringhe acquistano massa.
Si può a questo punto quantizzare la stringa. È utile introdurre le coordinate complesse
$$X_\pm =\frac{1}{\sqrt{2}}\left(X^1\pm iX^2\right),$$
(6.84)
i cui momenti coniugati sono
$$P_{}(\tau ,\sigma )=\frac{1}{2\pi \alpha ^{}}\left\{_\tau X_{}(\tau ,\sigma )+iX_{}(\tau ,\sigma )2\pi \alpha ^{}H\left[q_L\delta \left(\sigma \right)+q_R\delta \left(\pi \sigma \right)\right]\right\}.$$
(6.85)
Risolvendo le equazioni del moto si ottengono espansioni nei modi che dipenderanno dalla carica totale $`q_L+q_R`$. Se $`q_L+q_R0`$, $`X_+(\tau ,\sigma )`$, $`X_{}=X_+^{}`$, si trova
$$X_+(\tau ,\sigma )=x_++i\sqrt{2\alpha ^{}}\left[\underset{n=1}{\overset{\mathrm{}}{}}a_n\psi _n(\tau ,\sigma )\underset{m=0}{\overset{\mathrm{}}{}}b_m^{}\psi _m(\tau ,\sigma )\right],$$
(6.86)
con
$$\psi _n(\tau ,\sigma )=\frac{1}{\sqrt{\left|nz\right|}}\mathrm{cos}\left[\left(nz\right)\sigma +\gamma \right]e^{i\left(nz\right)\tau },$$
(6.87)
dove $`z`$, $`\gamma `$ e $`\gamma ^{}`$ definite come
$$z=\frac{1}{\pi }\left(\gamma +\gamma ^{}\right),\gamma =\mathrm{tan}^1\left(2\pi \alpha ^{}q_LH\right),\gamma ^{}=\mathrm{tan}^1\left(2\pi \alpha ^{}q_RH\right).$$
(6.88)
Per $`a_n,a_m^{}`$ e $`b_n,b_m^{}`$ si hanno le solite relazioni di commutazione mentre, gli zero modi non commutano
$$[x_+,x_{}]=\frac{1}{H\left(q_L+q_R\right)}.$$
(6.89)
Essi sono infatti l’analogo degli usuali operatori di creazione e di distruzione per i livelli di Landau che, nel limite di campi deboli, danno correzioni alla massa della forma
$$\mathrm{\Delta }M^2=\left(2n+1\right)\left(q_L+q_R\right)H.$$
(6.90)
Nel caso di carica totale nulla si ha invece $`z=0`$, e gli oscillatori non risentono più del campo magnetico, il cui effetto si manifesta negli zero modi
$$X_+(\tau ,\sigma )=\frac{x_++p_{}\left[\tau i2\pi \alpha ^{}qH\left(\sigma \frac{1}{2}\pi \right)\right]}{\sqrt{1+\left(2\pi \alpha ^{}qH\right)^2}}+i\sqrt{2\alpha ^{}}\underset{n=1}{\overset{\mathrm{}}{}}\left[a_n\psi _n(\tau ,\sigma )b_n^{}\psi _n(\tau ,\sigma )\right].$$
(6.91)
Consideriamo ora un campo magnetico costante che viva in uno spazio compatto. Ad esempio, compattifichiamo due dimensioni su un toro $`T^2`$. Se $`2\pi R_1`$ e $`2\pi R_2`$ sono i due lati della cella fondamentale, la degenerazione di Landau $`k`$ è data da
$$k=2\pi R_1R_2qH=2\pi \alpha ^{}vqH$$
(6.92)
dove si è definito $`v=R_1R_2/\alpha ^{}`$. Si può vedere che $`k`$ così definito è anche il numero intero che compare nella condizione di quantizzazione di Dirac, dovuta alla natura di monopolo del campo magnetico uniforme sul toro. Definiamo sulla cella fondamentale un potenziale vettore
$$A_2=a_1,A_3=a_2+HX_2,$$
(6.93)
tale che $`F_{23}=H`$. Come si è visto, le costanti $`a_{1,2}`$ sono legate all’introduzione di linee di Wilson, e in questo caso le possiamo porre a zero, concentrandoci sul termine che introduce una curvatura. Le trasformazioni di gauge
$$A_i=A_iie^{i\phi }_ie^{i\phi },i=1,2$$
(6.94)
per
$$\phi =2\pi R_1HX_2,$$
(6.95)
permettono di spostarsi con continuità da $`X_1=0`$ a $`X_1=2\pi R_1`$. Imponendo la monodromia della funzione $`q\phi `$, si trova la condizione di quantizzazione di Dirac $`2\pi \alpha ^{}vqH=k`$, da cui si vede chiaramente che l’intero $`k`$ è lo stesso che definisce la degenerazione dei livelli di Landau. Nella picture T-duale $`k`$ viene interpretato come il numero di avvolgimenti della $`D`$-brana ruotata sul toro e, come si vede dalla definizione dell’angolo di rotazione, usando $`R_2^T=\alpha ^{}/R_2`$,
$$\mathrm{tan}\theta =k\frac{R_2^T}{R_1}.$$
(6.96)
#### 6.4.2 Stringhe aperte su orbifold magnetizzati
Consideriamo l’azione effettiva di bassa energia di una D9-brana immersa in un campo abeliano di background,
$`S_9`$ $`=`$ $`T_9{\displaystyle \underset{a=1}{\overset{32}{}}}{\displaystyle _{_{10}}}d^{10}Xe^\varphi \sqrt{\mathrm{det}\left(g_{10}+2\pi \alpha ^{}q_aF\right)}`$ (6.97)
$`\mu _9{\displaystyle \underset{p,a}{}}{\displaystyle _{_{10}}}e^{2\pi \alpha ^{}q_aF}C_{p+1},`$
dove $`a`$ identifica le cariche di Chan-Paton che si accoppiano ai campi magnetici. Il primo termine è l’azione di Born Infeld ed il secondo è il termine di Wess-Zumino di accoppiamento con i campi di R-R $`C_{p+1}`$. $`T_9`$ e $`\mu _9`$ sono rispettivamente la tensione e la carica di R-R della brana, che per una generica brana BPS sono legati dalla relazione
$$T_p=\left|\mu _p\right|=\sqrt{\frac{\pi }{2k^2}}\left(2\pi \sqrt{\alpha ^{}}\right)^{3p},$$
(6.98)
dove $`k^2=8\pi G_N^{\left(10\right)}`$ definisce la costante di Newton in $`10`$ dimensioni. Introducendo una compattificazione dello spazio su due doppi tori in cui vivano due campi magnetici abeliani costanti
$$_{10}=_6\times T^2\left(H_1\right)\times T^2\left(H_2\right),$$
(6.99)
e utilizzando (6.98) si ottiene per l’azione
$`S_9`$ $`=`$ $`T_9{\displaystyle _{_{10}}}d^{10}Xe^\varphi {\displaystyle \underset{a=1}{\overset{32}{}}}\sqrt{g_6}\sqrt{\left(1+2\pi \alpha ^{}q_aH_1^2\right)\left(1+2\pi \alpha ^{}q_aH_2^2\right)}`$ (6.100)
$`32\mu _9{\displaystyle _{_{10}}}C_{10}\mu _5v_1v_2H_1H_2{\displaystyle \underset{a=1}{\overset{32}{}}}\left(2\pi q_a\right)^2{\displaystyle __6}C_6,`$
dove $`v_i=R_i^1R_i^2/\alpha ^{}`$ sono i volumi dei due tori di di raggi $`R_i^1`$ e $`R_i^2`$. Il termine lineare nell’espansione dell’azione di Wess-Zumino non compare, perché il generatore del gruppo abeliano $`U\left(1\right)`$ è a traccia nulla, dal momento che nel nostro caso è un sottogruppo del gruppo di gauge totale $`SO\left(32\right)`$ definito sulle D9-brane. Fissando $`H_1=\pm H_2`$ e usando le condizioni di quantizzazione di Dirac $`k_i=2\pi \alpha ^{}v_iqH_i`$ per entrambi i campi magnetici, l’azione si semplifica notevolmente e diventa
$`S_9`$ $`=`$ $`32{\displaystyle _{_{10}}}\left(d^{10}X\sqrt{g_6}e^\varphi T_9+\mu _9C_{10}\right)`$ (6.101)
$`{\displaystyle \underset{a=1}{\overset{32}{}}}\left({\displaystyle \frac{q_a}{q}}\right)^2{\displaystyle __6}\left(d^6X\sqrt{g_6}\left|k_1k_2\right|T_5e^\varphi +k_1k_2\mu _5C_6\right).`$
L’azione trovata indica che una D-9 brana magnetizzata mima $`\left|k_1k_2\right|`$ D5 brane se $`k_1k_2>0`$ ($`H_1=+H_2`$), ovvero $`\overline{\mathrm{D5}}`$ brane se $`k_1k_2<0`$ ($`H_1=H_2`$). Questo fenomeno si ritrova nei modelli di orbifold di stringa, in cui la presenza di O5-piani che riassorbano la carica delle D5 brane dovuta alle D-9 brane magnetizzate, permette la costruzione di modelli supersimmetrici .
Un esempio di questo fenomeno è dato illustrato modello $`_6\times \left[T^2\left(H_1\right)\times T^2\left(H_2\right)\right]/_2`$, in cui si hanno due campi magnetici abeliani nei due tori $`T^2`$. Per scrivere le ampiezze di vuoto occorre partire dal modello supersimmetrico $`T^4/_2`$. Decomponiamo i caratteri interni in rappresentazioni di $`SO\left(2\right)\times SO\left(2\right)`$
$`Q_o(z_1;z_2)`$ $`=`$ $`V_4\left(0\right)\left[O_2\left(z_1\right)O_2\left(z_2\right)+V_2\left(z_1\right)V_2\left(z_2\right)\right]`$
$`C_4\left(0\right)\left[S_2\left(z_1\right)C_2\left(z_2\right)+C_2\left(z_1\right)S_2\left(z_2\right)\right],`$
$`Q_v(z_1;z_2)`$ $`=`$ $`O_4\left(0\right)\left[V_2\left(z_1\right)O_2\left(z_2\right)+O_2\left(z_1\right)V_2\left(z_2\right)\right]`$
$`S_4\left(0\right)\left[S_2\left(z_1\right)S_2\left(z_2\right)+C_2\left(z_1\right)C_2\left(z_2\right)\right],`$
$`Q_s(z_1;z_2)`$ $`=`$ $`O_4\left(0\right)\left[S_2\left(z_1\right)C_2\left(z_2\right)+C_2\left(z_1\right)S_2\left(z_2\right)\right]`$
$`S_4\left(0\right)\left[O_2\left(z_1\right)O_2\left(z_2\right)+V_2\left(z_1\right)V_2\left(z_2\right)\right],`$
$`Q_c(z_1;z_2)`$ $`=`$ $`V_4\left(0\right)\left[S_2\left(z_1\right)S_2\left(z_2\right)+C_2\left(z_1\right)C_2\left(z_2\right)\right]`$ (6.102)
$`C_4\left(0\right)\left[V_2\left(z_1\right)O_2\left(z_2\right)+O_2\left(z_1\right)V_2\left(z_2\right)\right],`$
dove gli argomenti $`z_i`$ sono gli shift nei modi dovuti all’introduzione dei campi magnetici, che si sono trovati essere
$$z_i=\frac{1}{\pi }\left(\mathrm{tan}^1\left(2\pi \alpha ^{}q_LH_i\right)+\mathrm{tan}^1\left(2\pi \alpha ^{}q_RH_i\right)\right),$$
(6.103)
e i caratteri di livello 1 dell’estensione affine di $`O\left(2n\right)`$ sono legati alle quattro funzioni theta di Jacobi dalle relazioni
$`O_{2n}\left(z\right)`$ $`=`$ $`{\displaystyle \frac{1}{2\eta ^n\left(\tau \right)}}\left[\vartheta _3^n\left(z|\tau \right)+\vartheta _4^n\left(z|\tau \right)\right],`$
$`V_{2n}\left(z\right)`$ $`=`$ $`{\displaystyle \frac{1}{2\eta ^n\left(\tau \right)}}\left[\vartheta _3^n\left(z|\tau \right)\vartheta _4^n\left(z|\tau \right)\right],`$
$`S_{2n}\left(z\right)`$ $`=`$ $`{\displaystyle \frac{1}{2\eta ^n\left(\tau \right)}}\left[\vartheta _2^n\left(z|\tau \right)+i^n\vartheta _1^n\left(z|\tau \right)\right],`$
$`C_{2n}\left(z\right)`$ $`=`$ $`{\displaystyle \frac{1}{2\eta ^n\left(\tau \right)}}\left[\vartheta _2^n\left(z|\tau \right)i^n\vartheta _1^n\left(z|\tau \right)\right].`$ (6.104)
Il settore chiuso non viene alterato dall’introduzione dei campi magnetici sulle $`D`$-brane che interagiscono solo con le cariche di Chan-Paton del settore aperto. L’ampiezza di toro continua quindi ad assumere la forma usuale
$$𝒯=\frac{1}{2}\left[\left|Q_o+Q_v\right|^2\mathrm{\Sigma }_{n,w}+\left|Q_oQ_v\right|^2\left|\frac{2\eta }{\vartheta _2}\right|^4+16\left|Q_s+Q_c\right|^2\left|\frac{\eta }{\vartheta _4}\right|^4+16\left|Q_sQ_c\right|^2\left|\frac{\eta }{\vartheta _3}\right|^4\right].$$
e indicando le somme sui momenti e sui winding per i due tori come $`P_i`$ e $`W_i`$, l’ampiezza di Klein è
$$𝒦=\frac{1}{4}\left\{\left(Q_o+Q_v\right)(0;0)\left[P_1P_2+W_1W_2\right]+16\times 2\left(Q_s+Q_c\right)(0;0)\left(\frac{\eta }{\vartheta _4\left(0\right)}\right)^2\right\},$$
(6.105)
Nel settore aperto, per gruppi di gauge unitari come nel caso del modello originario supersimmetrico $`T^4/_2`$, indichiamo il numero di D9 brane neutre con $`N_0=n+\overline{n}`$, mentre $`m`$ e $`\overline{m}`$ contano il numero di D9 brane magnetizzate con cariche $`U\left(1\right)`$ uguali a $`+1`$ o $`1`$. Si hanno poi le D5 brane, con la loro molteplicità $`d+\overline{d}`$. L’ampiezza di anello coinvolge quindi diversi tipi di stringhe aperte: le stringhe “dipolari”, con molteplicità di Chan-Paton $`m\overline{m}`$; quelle scariche con molteplicità indipendenti da $`m`$ e $`\overline{m}`$; quelle con una singola carica, con moltpelicità linerari in $`m`$ e $`\overline{m}`$; infine quelle doppiamente cariche con molteplicità $`m^2`$, $`\overline{m}^2`$. Per le stringhe “dipolari” si è visto che l’introduzione di campi magnetici non introduce uno shift, ma modifica gli zero modi (6.91). I momenti devono quindi essere quantizzati in unità di $`1/R\sqrt{1+\left(2\pi \alpha ^{}H_i\right)^2}`$. Per queste stringhe quindi si deve sostituire la somma $`P_1P_2`$ con $`\stackrel{~}{P}_1\stackrel{~}{P}_2`$, definita in termini dei momenti $`m_i/R\sqrt{1+\left(2\pi \alpha ^{}H_i\right)^2}`$. Alla luce delle osservazioni fatte, l’ampiezza di anello si trova essere
$`𝒜`$ $`=`$ $`{\displaystyle \frac{1}{4}}\{(Q_o+Q_v)(0;0)[(n+\overline{n})^2P_1P_2+(d+\overline{d})^2W_1W_2+2m\overline{m}\stackrel{~}{P}_1\stackrel{~}{P}_2]`$
$``$ $`2\left(m+\overline{m}\right)\left(n+\overline{n}\right)\left(Q_o+Q_v\right)(z_1\tau ;z_2\tau ){\displaystyle \frac{k_1\eta }{\vartheta _1\left(z_1\tau \right)}}{\displaystyle \frac{k_2\eta }{\vartheta _1\left(z_2\tau \right)}}`$
$``$ $`\left(m^2+\overline{m}^2\right)\left(Q_o+Q_v\right)(2z_1\tau ;2z_2\tau ){\displaystyle \frac{2k_1\eta }{\vartheta _1\left(2z_1\tau \right)}}{\displaystyle \frac{2k_2\eta }{\vartheta _1\left(2z_2\tau \right)}}`$
$``$ $`\left[\left(n\overline{n}\right)^22m\overline{m}+\left(d\overline{d}\right)^2\right]\left(Q_oQ_v\right)(0;0)\left({\displaystyle \frac{2\eta }{\vartheta _2\left(0\right)}}\right)^2`$
$``$ $`2\left(m\overline{m}\right)\left(n\overline{n}\right)\left(Q_oQ_v\right)(z_1\tau ;z_2\tau ){\displaystyle \frac{2\eta }{\vartheta _2\left(z_1\tau \right)}}{\displaystyle \frac{2\eta }{\vartheta _2\left(z_2\tau \right)}}`$
$``$ $`\left(m^2+\overline{m}^2\right)\left(Q_oQ_v\right)(2z_1\tau ;2z_2\tau ){\displaystyle \frac{2\eta }{\vartheta _2\left(2z_1\tau \right)}}{\displaystyle \frac{2\eta }{\vartheta _2\left(2z_2\tau \right)}}`$
$`+`$ $`2\left(n+\overline{n}\right)\left(d+\overline{d}\right)\left(Q_s+Q_c\right)(0;0)\left({\displaystyle \frac{\eta }{\vartheta _4\left(0\right)}}\right)^2`$
$`+`$ $`2\left(m+\overline{m}\right)\left(d+\overline{d}\right)\left(Q_s+Q_c\right)(z_1\tau ;z_2\tau ){\displaystyle \frac{\eta }{\vartheta _4\left(z_1\tau \right)}}{\displaystyle \frac{\eta }{\vartheta _4\left(z_2\tau \right)}}`$
$``$ $`2\left(n\overline{n}\right)\left(d\overline{d}\right)\left(Q_sQ_c\right)(0;0)\left({\displaystyle \frac{\eta }{\vartheta _3\left(0\right)}}\right)^2`$
$``$ $`2(m\overline{m})(d\overline{d})(Q_sQ_c)(z_1\tau ;z_2\tau ){\displaystyle \frac{\eta }{\vartheta _3\left(z_1\tau \right)}}{\displaystyle \frac{\eta }{\vartheta _3\left(z_2\tau \right)}}\},`$
e l’ampiezza di Möbius risulta
$``$ $`=`$ $`{\displaystyle \frac{1}{4}}[(\widehat{Q}_o+\widehat{Q}_v)(0;0)[(n+\overline{n})P_1P_2+(d+\overline{d})W_1W_2]`$
$``$ $`\left(m+\overline{m}\right)\left(\widehat{Q}_o+\widehat{Q}_v\right)(2z_1\tau ;2z_2\tau ){\displaystyle \frac{2k_1\widehat{\eta }}{\widehat{\vartheta }_1\left(2z_1\tau \right)}}{\displaystyle \frac{2k_2\widehat{\eta }}{\widehat{\vartheta }_1\left(2z_2\tau \right)}}`$
$``$ $`\left(n+\overline{n}+d+\overline{d}\right)\left(\widehat{Q}_o\widehat{Q}_v\right)(0;0)\left({\displaystyle \frac{2\widehat{\eta }}{\widehat{\vartheta }_2\left(0\right)}}\right)^2`$
$``$ $`(m+\overline{m})(\widehat{Q}_o\widehat{Q}_v)(2z_1\tau ;2z_2\tau ){\displaystyle \frac{2\widehat{\eta }}{\widehat{\vartheta }_2\left(2z_1\tau \right)}}{\displaystyle \frac{2\widehat{\eta }}{\widehat{\vartheta }_2\left(2z_2\tau \right)}}],`$
dove si sono raggruppati i termini con cariche $`U\left(1\right)`$ opposte, e con argomenti $`z_i`$ opposti, utilizzando la simmetria delle funzioni theta di Jacobi e, si sono indicati il modulo di $`𝒜`$ e di $``$ con $`\tau `$. Si può anche notare come le stringhe con uno o due estremi carichi sono associate rispettivamente a funzioni di argomenti $`z_i`$ e $`2z_i`$.
Lo spettro non è più supersimmetrico, e in generale può sviluppare modi tachionici (instabilità di Nielsen-Olesen) . Nel settore untwisted, nel limite di campi deboli la formula di massa riceve correzioni della forma
$$\mathrm{\Delta }M^2=\frac{1}{2\pi \alpha ^{}}\underset{i=1,2}{}\left[\left(2n_i+1\right)\left|2\pi \alpha ^{}\left(q_\mathrm{L}+q_\mathrm{R}\right)H_i\right|+4\pi \alpha ^{}\left(q_\mathrm{L}+q_\mathrm{R}\right)\mathrm{\Sigma }_iH_i\right],$$
dove il primo termine è il contributo dei livelli di Landau e il secondo è l’accoppiamento dei momenti magnetici di spin $`\mathrm{\Sigma }_i`$ ai campi magnetici. Si vede chiaramente nella formula di $`\mathrm{\Delta }M^2`$ che, per valori generici dei campi, l’accoppiamento dei vettori interni può abbassare l’energia di Landau del livello più basso generando tachioni. I modi fermionici di spin semintero possono al limite compensare il contributo dei vettori. Nel settore twistato non ci sono livelli di Landau ma, mentre la parte fermionica di $`Q_s`$, $`S_4O_4`$ non sviluppa tachioni dal momento che i caratteri interni sono scalari e gli scalari non hanno accoppiamento magnetico, la parte bosonica $`O_4C_4`$ ha accoppiamento magnetico e sviluppa tachioni.
Si vede però che un’opprtuna scelta dei campi magnetici rende lo spettro privo di tachioni. Infatti ponendo $`H_1=H_2`$ si eliminano tutte le instabilità tachioniche. Inoltre di può vedere che questa scelta porta ad avere ampiezze di anello e di Möbius identicamente nulle, un segnale che una supersimmetria residua è presente nello spettro completo di stringa.
Completiamo questa breve rassegna del modello studiandone le condizioni di tadpole. Nel settore R-R untwisted, per il termine $`C_4S_2C_2`$, si ha
$$\left[n+\overline{n}+m+\overline{m}32+\left(2\pi \alpha ^{}q\right)^2H_1H_2\left(m+\overline{m}\right)\right]\sqrt{v_1v_2}+\frac{1}{\sqrt{v_1v_2}}\left[d+\overline{d}32\right]=0.$$
(6.108)
Si può vedere che le altre condizioni di tapole di R-R nel settore untwisted sono compatibili con questa o si cancellano dopo l’identificazion $`n=\overline{n}`$, $`m=\overline{m}`$, $`d=\overline{d}`$. La condizione (6.108) è legata al termine di Wess-Zumino nell’azione di basse energie. Imponendo la condizione di quantizzazione di Dirac su entrambi i due tori
$$2\pi \alpha ^{}qH_iv_i=k_i\left(i=1,2\right)$$
(6.109)
si ottiene
$`m+\overline{m}+n+\overline{n}=32,`$
$`k_1k_2\left(m+\overline{m}\right)+d+\overline{d}=32,`$ (6.110)
da cui si vede che le D9 brane magnetizzate acquisiscono la carica di R-R di $`\left|k_1k_2\right|`$ D5 brane se $`k_1k_2>0`$ o altrettante $`\overline{\mathrm{D5}}`$ antibrane se $`k_1k_2<0`$.
Il settore NS-NS untwisted contiene in generale si hanno condizioni di tadpole non cancellate. Per il tadpole del dilatone, da $`V_4O_2O_2`$ si ottiene
$`\left[n+\overline{n}+\left(m+\overline{m}\right)\sqrt{\left(1+\left(2\pi \alpha ^{}q\right)^2H_1^2\right)\left(1+\left(2\pi \alpha ^{}q\right)^2H_2^2\right)}32\right]\sqrt{v_1v_2}`$
$`+{\displaystyle \frac{1}{\sqrt{v_1v_2}}}\left[d+\overline{d}32\right],`$ (6.111)
che può essere legato alle derivate del termine di Born-Infeld nell’azione di bassa energia rispetto al campo del dilatone. Scegliendo $`H_1=H_2`$ e utilizzando la relazione di quantizzazione di Dirac si ritrova la forma della (6.110), e in questo caso quindi il tadpole di NS-NS si annulla identicamente imponendo le condizioni di tadpole nel settore R-R.
Per il termine $`O_4V_2O_2`$ la condizione di tapole risulta essere
$`\left[n+\overline{n}+\left(m+\overline{m}\right){\displaystyle \frac{1\left(2\pi \alpha ^{}qH_1\right)^2}{\sqrt{1+\left(2\pi \alpha ^{}qH_1\right)^2}}}\sqrt{1+\left(2\pi \alpha ^{}qH_2\right)^2}32\right]\sqrt{v_1v_2}`$
$`{\displaystyle \frac{1}{\sqrt{v_1v_2}}}\left[d+\overline{d}32\right],`$ (6.112)
che corrisponde anche alla condizione di tadpole per $`O_4O_2V_2`$ scambiando $`H_1`$ e $`H_2`$. Questi termini sono legati alle derivate del termine di Born-Infeld rispetto al volume dei due tori interni. Nelle condizioni di tadpole trovate non compaiono quadrati perfetti, a causa del comportamento del campo magnetico sotto inversione temporale. Le ampiezze del canale trasverso, della forma $`𝒯\left(B\right)\left|q^{L_0}\right|B`$, coinvolgono un’operazione di inversione temporale $`𝒯`$ sotto la quale il campo magnetico è dispari, questo introduce segni nelle ampiezze che impediscono la formazione di forme sesquilineari. Si può recuperare la corretta struttura dell’ampiezza di anello aggiungendo all’ampiezza di Möbius anche i contributi $`𝒯\left(B\right)\left|q^{L_0}\right|C`$ e $`𝒯\left(C\right)\left|q^{L_0}\right|B`$.
Anche in questo caso la scelta $`H_1=H_2`$, insieme alle condizioni di quantizzazione di Dirac, porta alla cancellazione dei tadpole.
La condizione di tadpole R-R nel settore twisted per $`S_4O_2O_2`$ è
$$15\left[\frac{1}{4}\left(m\overline{m}+n\overline{n}\right)\right]^2+\left[\frac{1}{4}\left(m\overline{m}+n\overline{n}\right)\left(d\overline{d}\right)\right]^2,$$
(6.113)
che riflette il fatto che le D5 brane sono tutte coincidenti con il medesimo punto fisso. Si annulla identificando le molteplicità coniugate. Il corrispondente tadpole NS-NS
$$\frac{2\pi \alpha ^{}q\left(H_1H_2\right)}{\sqrt{\left(1+\left(2\pi \alpha ^{}qH_1\right)^2\right)\left(1+\left(2\pi \alpha ^{}qH_2\right)^2\right)}},$$
(6.114)
e come nei casi precedenti si annulla solo per la scelta $`H_1=H_2`$.
## Capitolo 7 Superstringhe oltre un loop
### 7.1 Oltre un loop
Al livello ad albero e ad un loop le ampiezze per i diversi modelli di superstringa sono state calcolate da tempo, ma una formulazione operativa delle ampiezze a genere più alto anche, per i casi più semplici, è mancata a lungo. Recentemente E. D’Hoker e D.H. Phong hanno ottenuto una formulazione gauge invariante molto esplicita per le ampiezze di genere due per le superstringhe di Tipo II ed Eterotiche .
La complicazione fondamentale a genere più alto è l’emergere nella procedura di gauge fixing di supermoduli dispari grassmaniani che invece sono del tutto assenti al livello ad albero e ad un loop per strutture di spin pari. Per strutture di spin dispari ad un loop compare invece un modulo dispari ma che non introduce eccessive complicazioni.
Nel corso degli ultimi dieci anni molti sforzi sono stati spesi nel tentativo di dare una formulzione consistente per le ampiezze a genere superiore al primo. In particolare Friedan, Martinec e Shenker hanno proposto un primo approccio al problema basato sulla Teoria di Campo Conforme sul world-sheet, l’invarianza BRST e l’operatore di *picture changing*, in cui gli effetti dei supermoduli dispari vengono riassunti in termini di inserzioni dell’operatore di picture changing su un world-sheet puramente bosonico definito da soli moduli bosonici . Calcoli diretti hanno però dimostrato che questo produce ampiezze a due loop dipendenti dalla scelta dell’orbita di gauge.
Per superare questo ostacolo si sono cercate a lungo formulazioni differenti a partire da una grande varietà di principi quali l’invarianza modulare, il gauge del cono di luce, la geometria globale dello spazio di Teichmuller, la gauge unitaria, il formalismo operatoriale, metodi gruppali, fattorizzazioni e geometria algebrica. In nessuno di questi modi si è però riusciti a derivare ampiezze gauge indipendenti. La difficoltà nel definire ampiezze di superstringa a genere più alto del primo ha portato addirittura a valutare l’ipotesi che queste ampiezze fossero intrinsecamente ambigue.
Una formulazione delle ampiezze di superstringa di Tipo II e Eterotica, come integrali sullo spazio dei supermoduli, è stata proposta infine da D’Hoker e Phong a partire dallo formalismo di superspazio del world-sheet. In particolare è stata trovata una procedura consistente di gauge fixing e di separazione chirale sullo spazio dei supermoduli che permette la costruzione delle ampiezze. I lavori di D’Hoker e Phong consentono la costruzione operativa delle ampiezze in termini di integrali sui soli moduli bosonici.
Inoltre D’Hoker e Phong hanno mostrato che l’inconsistenza della fomulazione mediante inserzioni di operatori di picture changing evidenziata da Verlinde e Verlinde nasce da un’eliminazione dei supermoduli anticommutanti inconsistente con la supersimmetria locale sul world-sheet. Per superare questo problema, i due autori hanno proposto una nuova procedura di gauge-fixing basata sulla proiezione delle supergeometrie sulla loro matrice super-periodica, in luogo di quella effettuata sulle loro soggiacenti geometrie bosoniche. A differenza di quest’ultima, la proiezione sulla matrice super-periodica si è dimostrata essere invariante sotto supersimmetria locale del world-sheet.
Le prime sezioni di questo capitolo contengono una breve rassegna sulle proprietà degli spinori sulle superfici di Riemann e sulla formulazione delle ampiezze di superstringa. Un’introduzione più dettagliata alla supergeometria e alla formulazione dell’azione di superstringa in termini di supercampi è data nell’Appendice B, insieme ad una breve raccolta di ampiezze di superstringa al livello ad albero e ad un loop. La parte centrale del capitolo è dedicata alla discussione dei problemi legati alla definizione di ampiezze di superstringa di genere più alto del primo, in particolare alle difficoltà della costruzione mediante picture changing operator, e ai risultati di D’Hoker e Phong per le superstringhe di Tipo II e Eterotiche. Le ultime sezioni sono infine dedicate al lavoro originale di questa Tesi che consiste nella generalizzazione dei risultati ottenuti da D’Hoker e Phong ad altri casi con supersimmetria rotta (modelli di tipo 0 e modelli con “brane supersymmetry breaking” in dieci dimensioni).
### 7.2 Spinori su superfici di Riemann
In superfici di topologia non banale gli spinori devono essere definiti in maniera opportuna. A differenza di un vettore, uno spinore acquista in generale delle fasi a seguito del trasporto parallelo lungo curve chiuse e si hanno pertanto delle ambiguità. Come si vedrà, per superfici chiuse orientate di genere $`g`$, e quindi per teorie di stringhe chiuse orientate, esistono $`2^{2g}`$ scelte consistenti di fasi per spinori reali. Ognuna di queste scelte è detta *struttura di spin*.
Un aspetto importante nella definizione di una teoria perturbativa per le superstringhe è quindi l’assegnazione delle strutture di spin. Nel formalismo funzionale la proiezione GSO per le stringhe di tipo II si ottiene separando gli spinori di chiralità destra e sinistra, assegnando a ciascun gruppo strutture di spin indipendenti $`\nu `$ e $`\overline{\nu }`$, e sommando su tutte le strutture di spin. Questa è anche la prescrizione più naturale per evitare la comparsa di anomalie, dal momento che nessuna delle possibili strutture di spin viene privilegiata. La scelta di assegnare a tutti gli spinori di uno stesso gruppo la stessa struttura di spin è una richiesta necessaria per preservare l’invarianza spazio-temporale di Lorentz.
In realtà il principio di separazione dei gradi di libertà destri e sinistri richiede di far fronte e diversi inconvenienti. Come si è visto, la formulazione funzionale richiede che le azioni formulate con segnatura minkowskiana siano analiticamente continuate a segnature euclidee. Nello spazio di Minkowski $`\psi ^\mu `$ e $`\chi _m`$ sono spinori di Majorana-Weyl, ma nella segnatura Euclidea non esistono spinori di Majorana-Weyl: le due componenti chirali di uno spinore di Majorana sono l’una la complessa coniugata dell’altra, e devono quindi avere la medesima struttura di spin. Per aggirare questa difficoltà si parte da uno spinore reale (somma di due spinore di Weyl con chiralità opposta) separando i contributi solo nella quantizzazione. In questo modo, ogni fattore può essere pensato come il contributo di un fermione di Majorana-Weyl. Una difficoltà maggiore viene dal campo $`X^\mu `$ e dal termine $`\chi \overline{\chi }\psi _+\psi _{}`$ dell’azione di superstringa (1.95), che si dovrà opportunamente separare. Nei prossimi paragrafi si vedrà che la separazione chirale dei contributi associati ai momenti interi $`p_I^\mu `$ può essere implementata in virtù di un teorema (vedi Appendice B).
Per definire in maniera appropiata le strutture di spin è utile richiamare alcune nozioni della teoria delle superfici di Riemann. Il *primo gruppo di omologia* di una superficie compatta $`M`$ senza bordi e con $`h`$ manici è
$$H^1\left(M\right)=^{2h}.$$
(7.1)
Una base canonica per questo gruppo è costituite da curve chiuse $`A_I`$, $`B_I`$, $`I=1,2,\mathrm{},h`$, con forma di intersezione definita da
$$\mathrm{\#}(A_I,A_J)=0,\mathrm{\#}(A_I,B_J)=\delta _{IJ},\mathrm{\#}(B_I,B_J)=0,$$
(7.2)
dove $`\mathrm{\#}(A_I,B_J)=\mathrm{\#}(B_I,A_J)`$ (si veda la figura 7.1). La scelta di una base non è unica: data una base canonica $`(A_I,B_I)`$, si può definire una nuova base canonica $`(A_I^{},B_I^{})`$ come
$$B_I^{}=B_{IJ}A_J+A_{IJ}B_J,A_I^{}=D_{IJ}A_J+C_{IJ}B_J,$$
(7.3)
dove la matrice $`\left(2h\times 2h\right)`$
$$M=\left(\begin{array}{cc}A& B\\ C& D\end{array}\right),$$
(7.4)
appartiene al gruppo simplettico con coefficienti interi $`Sp(2h,)`$, che viene anche detto *gruppo modulare di Siegel*, che preserva la forma d’intersezione. Usando una decomposizione omologica canonica in cicli $`A`$ e $`B`$ si può aprire la superficie, come si è visto nel caso del toro, ottenendo una sua rappresentazione come regione semplicemente connessa del piano con metrica opportuna (piatta o di Poincaré a seconda che il genere sia uno o maggiore) i cui lati siano identificati a coppie.
Per definire gli spinori sulla superficie di Riemann è conveniente fissare una struttura di spin di riferimento $`\nu `$. Per fermioni reali il fattore di fase dovuto al trasporto lungo ciascun ciclo della base canonica $`A_I`$ e $`B_I`$, di ogni altra struttura di spin, differirà da quello di $`\nu `$ per $`0`$ o $`\pi `$, e si avranno complessivamente $`2^{2g}`$ differenti strutture di spin.
Ogni struttura di spin definisce una distinta classe di spinori ed un corrispondente operatore di Dirac. Esiste una classsificazione naturale delle strutture di spin in pari e dispari, rispetto alla parità del numero di zero modi dell’operatore di Dirac. In generale, il numero di zero modi dell’operatore di Dirac è o $`0`$ o $`1`$. Diffeomorfismi sul world-sheet $`M`$ possono collegare strutture di spin distinte $`\nu `$ e $`\nu ^{}`$, ma preserveranno la parità della struttura di spin, dal momento che la parità degli zero modi di Dirac è invariante.
### 7.3 Ampiezze di Superstringa
Come si è visto nel primo capitolo, nella formulazione di Ramond–Neveu–Schwarz delle superstringhe i gradi di libertà sono la posizione bosonica $`X^\mu `$ e la sua controparte fermionica $`\psi ^\mu `$, campi sul world-sheet $`\mathrm{\Sigma }`$ che trasformano come vettori sotto le trasformazioni di Lorentz dello spazio piatto Minkowskiano. In questa fomulazione compaiono anche la metrica sulla superficie di universo $`g_{mn}`$ e il campo del gravitino $`\chi _m`$, che risultano non dinamici. L’azione costruita in (1.95) è invariante sotto diffeomorfismi, trasformazioni di supersimmetria $`N=1`$, trasformazioni di Weyl e di super Weyl del world-sheet.
Alla luce del ruolo chiave giocato dalla supersimmetria locale, è conveniente riformulare l’azione in termini di un supercampo di materia $`𝐗^\mu `$ e di una supergeometria specificata dal riferimento locale $`E_M^A`$ e dal supercampo di connessione $`U\left(1\right)`$, $`\mathrm{\Omega }_M`$. La relazione fra i supercampi e i campi ordinari è
$`𝐗^\mu `$ $``$ $`X^\mu +\theta \psi _+^\mu +\overline{\theta }\psi _{}^\mu +i\theta \overline{\theta }F^\mu ,`$
$`E_m^a`$ $``$ $`e_m{}_{}{}^{a}+\theta \gamma ^a\chi _m{\displaystyle \frac{i}{2}}\theta \overline{\theta }Ae_m{}_{}{}^{a},`$ (7.5)
dove $`A`$ e $`F`$ sono campi ausiliari. L’azione (1.95) può essere riscritta in termini di supercampi in forma estremamente compatta
$`S={\displaystyle \frac{1}{4\pi }}{\displaystyle _\mathrm{\Sigma }}d^{2|2}𝐳E𝒟_+𝐗^\mu 𝒟_{}𝐗^\mu ,E\mathrm{sdet}E_M{}_{}{}^{A},`$ (7.6)
dove $`𝒟_\pm `$ sono derivate supercovarianti la cui forma esplicita può essere trovata nell’Appendice B ma non è necessaria alla presente esposizione.
Nella formulazione perturbativa di stringa alla Polyakov le ampiezze sono della forma
$$𝐀_𝒪=\underset{g=0}{\overset{\mathrm{}}{}}\frac{D\left(E\mathrm{\Omega }\right)\delta \left(T\right)}{\mathrm{Vol}\left(\mathrm{Symm}\right)}DX^\mu 𝒪e^S,$$
(7.7)
dove l’operatore $`𝒪`$ rappresenta sinteticamente l’inserzione di un numero arbitrario di operatori di vertice (figura 7.2). Per stringhe critiche in $`D=10`$, le simmetrie classiche sono tutte preservate anche a livello quantistico:
$$\mathrm{Symm}=\mathrm{sDiff}\left(\mathrm{\Sigma }\right)\times \mathrm{sWeyl}\left(\mathrm{\Sigma }\right)\times \mathrm{sU}\left(1\right)\left(\mathrm{\Sigma }\right).$$
(7.8)
Nella formulazione di Polyakov delle ampiezze, la misura di integrazione viene opportunamente divisa per il volume di questo gruppo di simmetria. Infine $`\delta \left(T\right)`$ implementa il vincolo sulla torsione della supergeometria $`N=1`$.
### 7.4 Gauge fixing sul superspazio e separazione chirale
Una procedura di gauge fixing consistente può essere derivata a partire da principi primi riducendo l’integrale su tutte le supergeometrie ad un integrale finito dimensionale sullo *spazio dei supermoduli* che è definito come il quoziente delle supergeometrie per tutte le simmetrie locali (7.8). Le dimensioni dello spazio dei supermoduli sono
$`s_h`$ $``$ $`\{E_M{}_{}{}^{A},\mathrm{\Omega }_M+\mathrm{torsion}\mathrm{constraints}\}/\mathrm{sDiff}\times \mathrm{sWeyl}\times sU\left(1\right)`$
$`\mathrm{dim}\left(s_h\right)`$ $`=`$ $`\{\begin{array}{cc}\left(0|0\right)\hfill & h=0\text{ ,}\hfill \\ \left(1|0\right)_e\mathrm{or}\left(1|1\right)_o\hfill & h=1\text{ ,}\hfill \\ \left(3h3|2h2\right)\hfill & h2\text{ ,}\hfill \end{array}`$ (7.9)
dove $`e`$ e $`o`$ indicano rispettivamente i casi di strutture di spin pari e dispari.
Lo spazio dei supermoduli è uno spazio quoziente, e quindi non ammette una parametrizzazione canonica. Occorre scegliere un’orbita di gauge $`𝒮`$ della stessa dimensione di $`s_h`$, che intersechi tutte le orbite del gruppo di simmetria (7.8). Per superfici di genere $`h2`$, si parametrizza $`𝒮`$ con $`m^A=\left(m^a|\zeta ^\alpha \right)`$, dove $`a=1,\mathrm{},3h3`$ identifica i supermoduli pari e $`\alpha =1,\mathrm{},2h2`$ identifica i supermoduli dispari. La procedura di gauge fixing è stata proposta da E. Verlinde e H. Verlinde e derivata da principi primi da D’Hoker e Phong , e coinvolge i supercampi di ghost $`B`$ e $`C`$ la cui espansione in campi ordinari è
$`B`$ $``$ $`\beta +\theta b+\mathrm{campi}\mathrm{ausiliari},`$
$`C`$ $``$ $`c+\theta \gamma +\mathrm{campi}\mathrm{ausiliari},`$ (7.10)
insieme ai loro complessi coniugati. L’espressione per le ampiezze in seguito alla procedura di gauge fixing è
$$𝐀_𝒪=_s\left|dm^A\right|^2D\left(𝐗BC\right)\left|\underset{A}{}\delta \left(H_A|B\right)\right|^2𝒪e^S,$$
(7.11)
dove l’azione di campi di materia e di ghost è
$$I\frac{1}{2\pi }_\mathrm{\Sigma }d^{2|2}𝐳E\left(\frac{1}{2}𝒟_+X^\mu 𝒟_{}X_\mu +B𝒟_{}C+\overline{B}𝒟_+\overline{C}\right).$$
(7.12)
I differenziali di super-Beltrami sono i vettori tangenti all’orbita di gauge $`𝒮`$ e sono definiti come
$$\left(H_A\right)_{}{}_{}{}^{z}()^{A\left(M+1\right)}E_{}{}_{}{}^{M}\frac{E_M^z}{m^A}=\overline{\theta }(\mu _A\theta \chi _A)|_{\mathrm{WZ}}.$$
(7.13)
La formula delle ampiezze (7.11) è stata ricavata a partire da una formulazione euclidea sul world-sheet dell’azione, come naturale per la formulazione alla Polyakov della teoria perturbativa. In questa fomulazione i gradi di libertà destri e sinistri sono collegati da un operazione di coniugazione. Al contrario, nella teoria originaria con segnatura Minkowskiana sul word-sheet i fermioni destri e sinistri erano indipendenti. Questa indipendenza è cruciale nella definizione delle teorie di stringa chiusa, dal momento che le strutture di spin destre e sinistre sono indipendenti e la proiezione GSO deve avvenire in maniera indipendente sui gradi di libertà con chiralità destra e sinistra.
Per recuperare l’indipendenza dei gradi di libertà occorre applicare una procedura di separazione chirale. Sebbene a prima vista l’azione (1.95) sembrebbe non consentire una procedura di separazione, dal momento che il temine quartico fermionico accoppia chiralità differenti e gli zero modi del campo scalare $`X^\mu `$ non possano essere separati, questa risulta possibile all’interno di ogni blocco conforme caratterizzato da un momento interno di loop $`p_I^\mu `$, $`I=1,\mathrm{},h`$. È conveniente a tal fine scegliere una base canonica per la prima omologia della superficie in termini dei cicli $`A_I`$ e $`B_I`$, per $`I=1,\mathrm{},h`$ (vedi fig. 7.1). I momenti di loop possono essere così pensati come i momenti che attraversano i cicli $`A_I`$.
La precedura di separazione chirale può essere riassunta in termini di regole operative. Le funzioni di correlazioni per il supercampo scalare puossono essere scritte come
$$\underset{i=1}{\overset{N}{}}V_i(k_i,ϵ_i)_{X^\mu }=𝑑p_I^\mu \left|\underset{i=1}{\overset{N}{}}V_i^{chi}(k_i,ϵ_i;p_I^\mu )_+\right|^2,$$
(7.14)
dove $`\mathrm{}_+`$ indica che si sono usate le regole effettive per le contrazioni degli operatori di vertice $`V_i^{chi}(k_i,ϵ_i;p_I^\mu )`$ date nella tabella 7.1.
Nella tabelle $`E(z,w)`$ è la forma prima e $`S_\delta (z,w)`$ è il kernel di Szegö. Il punto delle regole effettive è che queste coinvolgono solo oggetti meromorfi, a differenza dell’usuale propagatore $`x^\mu \left(z\right)x^\nu \left(w\right)`$ che è dato dalla funzione di Green scalare $`\delta ^{\mu \nu }G(z,w)`$. Per le ampiezze si ottiene
$$𝐀_𝒪\left[\delta \right]=\left|\underset{A}{}dm^A\right|^2𝑑p_I^\mu \left|e^{i\pi p_I^\mu \widehat{\mathrm{\Omega }}_{IJ}p_J^\mu }𝒜_𝒪\left[\delta \right]\right|^2,$$
(7.15)
dove $`𝒜_𝒪\left[\delta \right]`$ è il correlatore chirale effettivo
$$𝒜_𝒪\left[\delta \right]=\underset{A}{}\delta \left(H_A|B\right)𝒪_+\mathrm{exp}\left\{_\mathrm{\Sigma }\frac{d^2z}{2\pi }\chi _{\overline{z}}{}_{}{}^{+}S\left(z\right)\right\}_+,$$
(7.16)
e $`S\left(z\right)`$ è la supercorrente totale
$$S\left(z\right)=\frac{1}{2}\psi _+^\mu _zx_+^\mu +\frac{1}{2}b\gamma \frac{3}{2}\beta _zc\left(_z\beta \right)c,$$
(7.17)
La matrice $`\widehat{\mathrm{\Omega }}_{IJ}`$ è detta *matrice dei super periodi* e può essere definita ad ogni genere $`h>0`$. A genere 2 la sua espressione è piuttosto semplice,
$$\widehat{\mathrm{\Omega }}_{IJ}=\mathrm{\Omega }_{IJ}\frac{i}{8\pi }_\mathrm{\Sigma }d^2z_\mathrm{\Sigma }d^2w\omega _I\left(z\right)\chi _{\overline{z}}S_\delta (z,w)\chi _{\overline{w}}\omega _J\left(w\right),$$
(7.18)
dove $`\mathrm{\Omega }_{IJ}`$ è la matrice dei periodi corrispondente alla struttura complessa della metrica $`g_{mn}`$. Le funzioni $`\omega _I\left(z\right)`$ sono una base di differenziali olomorfi abeliani duali agli $`A_I`$-cicli, tali che
$$_{A_I}\omega _J=\delta _{IJ},_{B_I}\omega _J=\mathrm{\Omega }_{IJ}.$$
(7.19)
In analogia con i differenziali Abeliani ordinari si possono introdurre forme $`1/2`$ superolomorfe $`\widehat{\omega }_I`$, che possono essere normalizzate canonicamente sui cicli $`A_I`$, e che integrate sui cicli $`B_I`$ danno la matrice dei super periodi,
$$𝒟_{}\widehat{\omega }_I=0,_{A_I}\widehat{\omega }_J=\delta _{IJ},_{B_I}\widehat{\omega }_J=\widehat{\mathrm{\Omega }}_{IJ}.$$
(7.20)
L’ampiezza per la superstringa di Tipo II si ottiene tenendo conto dei contributi dei modi destri e sinistri, utilizzando la stessa matrice dei periodi e i momenti interni, ma strutture di spin indipendenti. L’ampiezza per la Tipo II risulta essere
$$𝐀_{II𝒪}=𝑑p_I^\mu \underset{\delta ,\overline{\delta }}{}\eta _{\delta ,\overline{\delta }}_{s_h}\left|dm^A\right|^2\left|\mathrm{exp}\left\{i\pi p_I^\mu \widehat{\mathrm{\Omega }}_{IJ}p_J^\mu \right\}\right|𝒜_𝒪\left[\delta \right]\left(\widehat{\mathrm{\Omega }}\right)\overline{𝒜}_𝒪\left[\overline{\delta }\right]\left(\widehat{\mathrm{\Omega }}^{}\right),$$
(7.21)
dove $`_h`$ indica lo spazio dei moduli bosonici delle superfici di Riemann di genere $`h`$. Le fasi $`\eta _{\delta ,\overline{\delta }}`$ devono essere scelte in modo da essere consistenti con l’invarianza modulare, e sono necessarie per introdurre la proiezione GSO indipendentemente sui modi sinistri e destri. Le ampiezze per la superstringa eterotica si possono scrivere in maniera analoga.
### 7.5 Calcolo della misura chirale
#### 7.5.1 Costruzione della misura chirale
La misura chirale nella formulazione di D’Hoker e Phong fa uso dei supermoduli supersimmetrici $`m^A=(\widehat{\mathrm{\Omega }}_{IJ},\zeta ^\alpha )`$. Tutte le quantità calcolate originariamente per la metrica $`g_{mn}`$ con struttura complessa $`\mathrm{\Omega }_{IJ}`$ devono essere riespresse in termini della matrice dei super periodi $`\widehat{\mathrm{\Omega }}_{IJ}`$. Nelle funzioni di correlazione, questo cambiamento si ottiene inserendo il tensore energia-impulso,
$$\mathrm{\Omega }_{IJ}\widehat{\mathrm{\Omega }}_{IJ}\{\begin{array}{ccc}g& & \widehat{g}=g+\widehat{\mu }\\ _{\overline{z}}& & \widehat{}_{\overline{z}}=_{\overline{z}}+\widehat{\mu }_z\\ \mathrm{}\left(g\right)& =& \mathrm{}\left(\widehat{g}\right)+\widehat{\mu }T\mathrm{}\left(\widehat{g}\right)\end{array},$$
(7.22)
dove il differenziale di Beltrami è associato alle deformazioni della struttura complessa $`\mathrm{\Omega }_{IJ}`$ in $`\widehat{\mathrm{\Omega }}_{IJ}`$. Alla luce della relazione che lega le due matrici periodiche, $`\widehat{\mu }`$ è dato da
$$_\mathrm{\Sigma }\widehat{\mu }\omega _I\omega _J=\frac{1}{8\pi }_\mathrm{\Sigma }d^2z_\mathrm{\Sigma }d^2w\omega _I\left(z\right)\chi _{\overline{z}}S_\delta (z,w)\chi _{\overline{w}}\omega _J\left(w\right).$$
(7.23)
La matrice dei super periodi è invariante sotto variazioni dei supermoduli dispari $`\zeta `$, e questo implica che nessuna delle componenti del super differziale di Beltrami è nulla,
$$\delta _\zeta \widehat{\mathrm{\Omega }}_{IJ}=0\{\begin{array}{c}H_A=\overline{\theta }\left(\mu _A\theta \chi _A\right)\\ \mu _A0\&\chi _A0\end{array}.$$
(7.24)
Gli oggetti duali dei super differenziali di Beltrami sono 3/2 forme superolomorfe di tipo dispari $`\mathrm{\Phi }_{IJ}`$ e pari $`\mathrm{\Phi }_\alpha `$. La loro formula esplicita per $`\mathrm{\Phi }_{IJ}`$ è
$$\mathrm{\Phi }_{IJ}=\frac{i}{2}\left(\widehat{\omega }_I𝒟_+\widehat{\omega }_J+\widehat{\omega }_J𝒟_+\widehat{\omega }_I\right),$$
(7.25)
che è normalizzata in modo da soddisfare le relazioni
$$H_a|\mathrm{\Phi }_{IJ}=\delta _{a,IJ}\mathrm{e}H_\alpha |\mathrm{\Phi }_{IJ}=0.$$
(7.26)
#### 7.5.2 Cambio di base per i super differenziali di Beltrami
Prima di procedere al calcolo diretto occorre introdurre un cambio di base per i super differenziali di Beltrami. Questo è necessario dal momento che l’utilizzo di moduli bosonici supersimmetrici forza tuttte le componenti di $`H_A`$ ad essere non nulle come indicato nella (7.24). Senza operare il cambio di base il prodotto dei fattori $`\delta \left(H_A|B\right)`$ produce una forma per le funzioni di correlazione eccessivamente complicata e impossibile da gestire.
Per operatori di vertice che siano indipendenti dai super campi di ghost $`B`$ (come nel caso di vertici NS), $`H_A`$ risulta essere effettivamente accoppiato con $`B`$. Si può quindi effettuare un cambio di base da $`H_A`$ ad un nuovo super differenziale di Beltrami $`H_A^{}`$, scelto per semplicità della forma
$`H_a^{}`$ $`=`$ $`\overline{\theta }\delta (z,p_a)a=1,2,3,`$
$`H_\alpha ^{}`$ $`=`$ $`\overline{\theta }\theta \delta (z,q_\alpha )\alpha =1,2.`$ (7.27)
Considerando un insieme completo arbitrario di 3/2 forme superolomorfe pari e dispari $`\mathrm{\Phi }_C`$, si ha
$$\underset{A}{}\delta \left(H_A|B\right)=\frac{\mathrm{sdet}H_A|\mathrm{\Phi }_C}{\mathrm{sdet}H_A^{}|\mathrm{\Phi }_C}\underset{a}{}b\left(p_a\right)\underset{\alpha }{}\delta \left(\beta \left(q_\alpha \right)\right)$$
(7.28)
Questa formula è chiaramente indipendente dalla scelta delle $`\mathrm{\Phi }_C`$. È di grande utilità che tutte le funzioni di correlazione siano ora espresse in termini di inserzioni di campi ordinari e che tutte le complicazioni presenti in $`H_A`$ sono confinate al fattore moltiplicativo.
Esistono due scelte naturali della base per $`\mathrm{\Phi }_C`$. La prima, che indichiamo con $`\mathrm{\Phi }_C`$, è duale di $`H_A`$, mentre la seconda, che indichiamo con $`\mathrm{\Phi }_C^{}`$, è duale di $`H_A^{}`$,
$$H_A|\mathrm{\Phi }_C=H_A^{}|\mathrm{\Phi }_C^{}=\delta _{AC}$$
(7.29)
Mentre la forma esplicita di $`\mathrm{\Phi }_C^{}`$ è nota, per $`\mathrm{\Phi }_C`$ la forma esplicita è nota solamente per le componenti dispari (7.25). Per le componenti pari non si ha un espressione canonica. In generale esse possono essere espresse come una combinazione lineare
$$\mathrm{\Phi }_\gamma \left(𝐳\right)=\mathrm{\Phi }_ϵ^{}\left(𝐳\right)C^ϵ{}_{\gamma }{}^{}+\mathrm{\Phi }_{IJ}\left(𝐳\right)D^{IJ}{}_{\gamma }{}^{},$$
(7.30)
dove $`C`$ e $`D`$ sono matrici indipendenti da $`𝐳`$ ma dipendenti dai moduli. Accoppiando con $`H_\alpha `$ e usando il fatto che $`H_\alpha |\mathrm{\Phi }_{IJ}=0`$, si ha $`detC\times detH_\alpha |\mathrm{\Phi }_\gamma ^{}=1`$, e tenendo conto di tutti i fattori
$$\frac{\mathrm{sdet}H_A|\mathrm{\Phi }_C}{\mathrm{sdet}H_A^{}|\mathrm{\Phi }_C}=\frac{1}{det\mathrm{\Phi }_{IJ}\left(p_a\right)\times detH_\alpha |\mathrm{\Phi }_\gamma ^{}}.$$
(7.31)
Le componenti $`\mathrm{\Phi }_{IJ}\left(p_a\right)`$ sono note in forma esplicita, e le componenti $`\mu _\alpha `$ e $`\chi _\alpha `$ di $`H_\alpha =\overline{\theta }\left(\mu _\alpha \theta \chi _\alpha \right)`$ sono anch’esse note. L’oggetto $`\chi _\alpha `$ rappresenta la scelta dell’orbita dei gravitini sul worldsheet, ed è pertanto fissato nella procedura di gauge fixing (le ampiezze dovranno risultare indipendenti da questa scelta). L’oggeto $`\mu _\alpha `$ si può dimostrare essere dato da $`\mu _\alpha =\widehat{\mu }/\zeta ^\alpha `$. Tutti i fattori della formula in cui si è fissato un gauge sono quindi noti in forma esplicita, e si può scrivere la completa misura chirale nella forma
$$𝒜\left[\delta \right]=\frac{\underset{a}{}b\left(p_a\right)\underset{\alpha }{}\delta \left(\beta \left(q_\alpha \right)\right)}{det\mathrm{\Phi }_{IJ+}\left(p_a\right)detH_\alpha |\mathrm{\Phi }_\beta ^{}}\left\{1+\frac{1}{2\pi }\widehat{\mu }T\frac{1}{8\pi ^2}_\mathrm{\Sigma }_\mathrm{\Sigma }\chi \chi SS\right\}$$
(7.32)
Il calcolo diretto dimostra che questa espressione è invariante sotto trasformazioni locali di supersimmetria sul world-sheet, come atteso .
#### 7.5.3 Il calcolo in componenti
Per ottenere una formula operativa per il calcolo delle ampiezze si può considerare per il gravitino un’orbita con supporto su due punti arbitrari $`x_1`$ e $`x_2`$,
$$\chi _\alpha \left(z\right)=\delta (z,x_\alpha ).$$
(7.33)
La misura chirale può essere così espressa interamente in termini di quantità che sono meromorfe sul worldsheet, quali la forma prima $`E(z,w)`$, il kernel di Szegö $`S_\delta (z,w)`$, la funzione di Green per i ghost $`bc`$ $`G_2(z,w)`$ (che è definita in modo da annullarsi quando $`z=p_1,p_2,p_3`$ alla luce delle inserzioni di $`b`$ a $`p_a`$) la funzione di Green di superghost $`G_{3/2}(z,w)`$ (che si annulla quando $`z=q_1,q_2`$ in ragione dell’inserzione $`\delta \left(\beta \right)`$ a $`q_\alpha `$). Esistono inoltre alcuni differenziali olomorfi, $`\psi _\alpha ^{}\left(z\right)`$ e $`\overline{\psi }_\alpha \left(z\right)`$, 3/2 forme olomorfe normalizzate in modo da avere $`\psi _\alpha ^{}\left(q_\beta \right)=\overline{\psi }_\alpha \left(x_\beta \right)=\delta _{\alpha \beta }`$, e la quantità $`\varpi _a(z,w)`$ che induce una mappa uno ad uno fra le due forme olomorfe in una variabile e le forme olomorfe di due variabili ciascuna di peso 1. Quest’ultima è normalizzata in modo da avere $`\varpi _a(p_b,p_b)=\delta _{ab}`$.
La misura chirale si scrive nella forma
$$𝒜\left[\delta \right]=\frac{\underset{a}{}b\left(p_a\right)\underset{\alpha }{}\delta \left(\beta \left(q_\alpha \right)\right)}{det\omega _I\omega _J\left(p_a\right)det\psi _\beta ^{}\left(x_\alpha \right)}\left\{1+\frac{\zeta ^1\zeta ^2}{16\pi ^2}\underset{i=1}{\overset{6}{}}𝒳_i\right\}.$$
(7.34)
Le quantità $`𝒳_i`$ sono definite come
$`𝒳_1`$ $`=`$ $`10S_\delta (x_1,x_2)_{x_1}_{x_2}\mathrm{ln}E(x_1,x_2),`$
$`3_{x_2}G_2(x_1,x_2)G_{3/2}(x_2,x_1)2G_2(x_1,x_2)_{x_2}G_{3/2}(x_2,x_1)(12),`$
$`𝒳_2`$ $`=`$ $`S_\delta (x_1,x_2)\omega _I\left(x_1\right)\omega _J\left(x_2\right)_I_J\mathrm{ln}\left({\displaystyle \frac{\vartheta \left[\delta \right]\left(0\right)^5\vartheta \left(p_1+p_2+p_33\mathrm{\Delta }\right)}{\vartheta \left[\delta \right]\left(q_1+q_22\mathrm{\Delta }\right)}}\right),`$
$`𝒳_3`$ $`=`$ $`2S_\delta (x_1,x_2){\displaystyle \underset{a}{}}\varpi _a(x_1,x_2)\left[B_2\left(p_a\right)+B_{3/2}\left(p_a\right)\right],`$
$`𝒳_4`$ $`=`$ $`2S_\delta (x_1,x_2){\displaystyle \underset{a}{}}_{p_a}_{x_1}\mathrm{ln}E(p_a,x_1)\varpi _a(p_a,x_2)(12),`$ (7.35)
$`𝒳_5`$ $`=`$ $`{\displaystyle \underset{a}{}}S_\delta (p_a,x_1)_{p_a}S_\delta (p_a,x_2)\varpi _a(x_1,x_2)(12),`$
$`𝒳_6`$ $`=`$ $`3_{x_2}G_2(x_1,x_2)G_{3/2}(x_2,x_1)+2f_{3/2}\left(x_1\right)G_2(x_1,x_2)\overline{\psi }_1\left(x_2\right)(12)`$
$`+2G_{3/2}(x_2,x_1)G_2(x_1,x_2)\overline{\psi }_2\left(x_2\right)+_{x_2}G_2(x_2,x_1)\overline{\psi }_2\left(x_1\right)(12),`$
dove si è utilizzata la seguente notazione,
$`f_n\left(w\right)`$ $`=`$ $`\omega _I\left(w\right)_I\mathrm{ln}\vartheta \left[\delta \right]\left(D_n\right)+_w\mathrm{ln}\left({\displaystyle \underset{i}{}}\sigma \left(w\right)E(w,z_i)\right),`$
$`B_2\left(w\right)`$ $`=`$ $`27T_1\left(w\right)+{\displaystyle \frac{1}{2}}f_2\left(w\right)^2{\displaystyle \frac{3}{2}}_wf_2\left(w\right)2{\displaystyle \underset{a}{}}_{p_a}_w\mathrm{ln}E(p_a,w)\varpi _a(p_a,w),`$
$`B_{3/2}\left(w\right)`$ $`=`$ $`12T_1\left(w\right){\displaystyle \frac{1}{2}}f_{3/2}\left(w\right)^2+_wf_{3/2}\left(w\right)+{\displaystyle \frac{3}{2}}_{x_1}G_2(w,x_1)+{\displaystyle \frac{3}{2}}_{x_2}G_2(w,x_2)`$ (7.36)
$`{\displaystyle \frac{3}{2}}_wG_{3/2}(x_1,w)\overline{\psi }_1\left(w\right){\displaystyle \frac{3}{2}}_wG_{3/2}(x_2,w)\overline{\psi }_2\left(w\right){\displaystyle \frac{1}{2}}G_{3/2}(x_1,w)\overline{\psi }_1\left(w\right)`$
$`{\displaystyle \frac{1}{2}}G_{3/2}(x_2,w)\overline{\psi }_2\left(w\right)+G_2(w,x_1)\overline{\psi }_1\left(x_1\right)+G_2(w,x_2)\overline{\psi }_2\left(x_2\right).`$
L’espressione ottenuta per la misura chirale è una somma di termini che sono tutti manifestamente funzioni scalari, meromorfe, di $`x_\alpha `$, $`q_\alpha `$ e $`p_a`$. Si può dimostrare tramite calcolo diretto che il risultato finale è indipendente da tutti questi punti, e questo dimostra la consistenza dell’espressione trovata .
### 7.6 Problemi con la definizione precedente dell’ampiezza
Possiamo a questo punto accennare ai problemi della formulazione via picture changing e quantizzazione BRST. Un’assunzione centrale in questo approccio è la scelta di fissare una metrica indipendente dai supermoduli dispari, mentre la forma del gravitino (preso a supporto puntiforme nei punti $`z_\alpha `$) caratterizza l’orbita di gauge per i supermoduli dispari
$$g_{mn}\left(m^a\right),\chi =\underset{\alpha =1,2}{}\zeta ^\alpha \chi _\alpha \left(m^a\right).$$
(7.37)
Questa scelta porta ad avere, nell’integrale dell’ampiezza, un termine di integrazione sulle variabili grassmaniane del tipo
$$𝒪\underset{a=1}{\overset{3h3}{}}\left(\mu _a|b\right)\underset{\alpha =1}{\overset{2h2}{}}Y\left(z_\alpha \right)\underset{a=1}{\overset{3h3}{}}dm^a,$$
(7.38)
dove $`Y\left(z_\alpha \right)`$ è l’operatore di picture changing. Come si è detto un calcolo esplicito dimostra che in quest’ampiezza c’è una dipendenza residua dai punti di inserzione $`z_\alpha `$.
Le ragioni del fallimento di questo approccio risiedono nella procedura di gauge fixing sullo spazio dei super moduli. I super moduli anticommutanti possono essere pensati come fibre sui supermoduli pari. L’operazione di integrazione dei supermoduli, che permette di avere delle ampiezze di superstringa espresse come integrali sui soli moduli pari, equivale ad una proiezione lungo le fibre dello spazio dei supermoduli sulle loro basi pari.
L’originale scelta problematica del gauge equivale alla proiezione
$`(g_{mn},\chi _m)`$ $``$ $`(g_{mn}^{},\chi _m^{})\mathrm{sotto}\mathrm{SUSY}`$
$``$ $``$
$`g_{mn}`$ $`/`$ $`g_{mn}^{}\mathrm{sotto}\mathrm{Diff}\times \mathrm{Weyl}`$ (7.39)
La scelta di fissare i moduli dispari solo in funzione del gravitino produce un’inconsistenza nelle ampiezze, come si può capire osservando che, una trasformazione di supersimmetria sulla metrica,
$$\delta g_{mn}=2\xi ^+\chi _{\{m}{}_{}{}^{+}e_{n\}}^{}^{\overline{z}}$$
(7.40)
modifica anche i moduli $`m^a`$ definiti in precedenza. La scelta operata per i supermoduli anticommutanti non è quindi invariante sotto supersimmetria. Questo spiega le ambiguità trovate nella definizione delle ampiezze.
Una corretta proiezione deve quindi essere tale da ottenere dei supermoduli $`m^a`$ definiti in maniera invariante sotto l’azione della supersimmetria locale,
$`(g_{mn},\chi _m)`$ $``$ $`(g_{mn}^{},\chi _m^{})\mathrm{sotto}\mathrm{SUSY}`$
$``$ $``$
$`\widehat{g}_{mn}\left(m^a\right)`$ $``$ $`\widehat{g}_{mn}^{}\left(m^a\right)\mathrm{sotto}\mathrm{Diff}\times \mathrm{Weyl}.`$ (7.41)
Come si è visto nell’approccio alla D’Hoker e Phong questo avviene grazie all’utilizzo di una matrice dei super periodi $`\widehat{\mathrm{\Omega }}_{IJ}`$ invariante sotto supersimmetria locale.
### 7.7 Formule esplicite in termini di funzioni $`\vartheta `$
L’indipendenza della misura da tutti i punti consente di fissare $`x_\alpha =q_\alpha `$. Dal momento che i termini $`𝒳_2`$, $`𝒳_3`$, $`𝒳_4`$ sono proporzionali a $`S_\delta (x_1,x_2)`$ essi si annullano scegliendo la *split gauge* $`S_\delta (q_1,q_2)=0`$. Questa scelta risulta particolarmente naturale dal momento che implica $`\widehat{\mathrm{\Omega }}_{IJ}=\mathrm{\Omega }_{IJ}`$.
È infine vantaggioso scegliere come punti $`p_a`$ i tre zeri di una 3/2 forma olomorfa $`\psi _A\left(z\right)`$. Questa scelta porta ad una forma particolarmente utile per la funzione di Green $`G_2`$ dei campi $`bc`$ in termini di $`\psi _A`$ e del kernel di Szegö,
$$G_2(z,w)=S_\delta (z,w)\psi _A\left(z\right)/\psi _A\left(w\right),$$
(7.42)
tale che in *split gauge* $`G_2(q_1,q_2)=0`$. Combinando tutti i contributi, si ottiene $`𝒳_1+𝒳_6=𝒳_2=𝒳_3=𝒳_4=0`$ mentre il solo $`𝒳_5`$ è diverso da zero. Il calcolo esplicito di $`𝒳_5`$ è piuttosto complesso è può essere trovato in .
La formula finale della misura chirale di superstringa
$$d\mu \left[\delta \right]\left(\mathrm{\Omega }\right)=\frac{\mathrm{\Xi }_6\left[\delta \right]\left(\mathrm{\Omega }\right)\vartheta \left[\delta \right]^4(0,\mathrm{\Omega })}{16\pi ^6\mathrm{\Psi }_{10}\left(\mathrm{\Omega }\right)}d^3\mathrm{\Omega }_{IJ},$$
(7.43)
è invece estremamente semplice ed è espressa in termini una forma modulare, di funzioni $`\vartheta `$ a genere due, e di una funzione $`\mathrm{\Xi }_6\left[\delta \right]\left(\mathrm{\Omega }\right)`$ su cui torneremo a breve.
A genere due esistono 16 strutture di spin, che possono essere descritte mediante caratteristiche semi-intere
$$\kappa =\left(\kappa ^{}|\kappa ^{\prime \prime }\right)\kappa ^{},\kappa ^{\prime \prime }(0,\frac{1}{2})^2.$$
(7.44)
in questa notazione le due componenti di $`\kappa ^{}`$ si riferiscono alle strutture di spin sui cicli $`A_I`$, mentre le componenti di $`\kappa ^{\prime \prime }`$ si riferiscono a quelli sui cicli $`B_I`$. Le strutture di spin pari e dispari possono essere distinte in base al valore pari o dispari del prodotto $`4\kappa ^{}\kappa ^{\prime \prime }`$. Si trova così che le 16 strutture di spin sono divise in 10 strutture pari, indicate genericamente con $`\delta `$, e 6 dispari, indicate con $`\nu `$. Ogni struttura pari, a genere 2, può essere scritta in due diversi modi come somma di tre distinte strutture di spin dispari,
$$\delta =\nu _{i_1}+\nu _{i_2}+\nu _{i_3}=\nu _{i_4}+\nu _{i_5}+\nu _{i_6}.$$
(7.45)
Date due strutture di spin $`\kappa `$ e $`\rho `$ la loro segnatura è definita come
$$\kappa |\rho \mathrm{exp}\left\{4\pi i\left(\kappa ^{}\rho ^{\prime \prime }\rho ^{}\kappa ^{\prime \prime }\right)\right\},$$
(7.46)
che assume valori $`\pm 1`$.
Le funzioni $`\vartheta `$ di genere due con caratteristica $`\kappa `$ sono definite come
$$\vartheta \left[\kappa \right](v,\mathrm{\Omega })\underset{n\kappa ^{}𝐙^2}{}\mathrm{exp}\left\{i\pi n^t\mathrm{\Omega }n+2\pi in^t\left(v+\kappa ^{\prime \prime }\right)\right\},$$
(7.47)
e sono funzioni di $`v𝐂^2`$ pari o dispari a seconda che $`\kappa `$ sia una struttura di spin pari o dispari. Di grande importanza sono le $`\vartheta `$-costanti, definite come
$$\vartheta \left[\delta \right]\vartheta \left[\delta \right](0,\mathrm{\Omega }),$$
(7.48)
dove $`\delta `$ è una struttura di spin pari. Per strutture di spin dispari, in analogia con il caso di genere 1 si ha,
$$\vartheta \left[\nu \right](0,\mathrm{\Omega })0$$
(7.49)
L’oggetto indicato con $`\mathrm{\Psi }_{10}`$ è una forma modulare di genere due definita come
$$\mathrm{\Psi }_{10}\left(\mathrm{\Omega }\right)\underset{\delta \mathrm{even}}{}\vartheta \left[\delta \right]^2(0,\mathrm{\Omega }),$$
(7.50)
mentre $`\mathrm{\Xi }_6\left[\delta \right]\left(\mathrm{\Omega }\right)`$ è definita come
$$\mathrm{\Xi }_6\left[\delta \right]\left(\mathrm{\Omega }\right)\underset{1i<j3}{}\nu _i|\nu _j\underset{k=4,5,6}{}\vartheta \left[\nu _i+\nu _j+\nu _k\right]^4(0,\mathrm{\Omega }).$$
(7.51)
È importante osservare che $`\mathrm{\Xi }_6\left[\delta \right]\left(\mathrm{\Omega }\right)`$ non è una forma modulare. Nella definizione di $`\mathrm{\Xi }_6\left[\delta \right]\left(\mathrm{\Omega }\right)`$, la struttura di spin $`\delta `$ è scritta come somma di tre distinte strutture di spin dispari $`\delta =\nu _1+\nu _2+\nu _3`$ mentre $`\nu _4`$, $`\nu _5`$ e $`\nu _6`$ indicano le rimanenti strutture di spin dispari.
### 7.8 Proprietà modulari
Le trasformazioni modulari a cui si è precedentemente accennato sono definite come le trasformazioni che lasciano la matrice canonica di intersezione invariante
$$\left(\begin{array}{cc}A& B\\ C& D\end{array}\right)\left(\begin{array}{cc}0& I\\ I& 0\end{array}\right)\left(\begin{array}{cc}A& B\\ C& D\end{array}\right)^t=\left(\begin{array}{cc}0& I\\ I& 0\end{array}\right)\left(\begin{array}{cc}A& B\\ C& D\end{array}\right)Sp(4,𝐙)$$
(7.52)
e formano a genere due il gruppo $`Sp(4,𝐙)`$. I generatori di questo gruppo sono,
$`M_i`$ $`=`$ $`\left(\begin{array}{cc}I& B_i\\ 0& I\end{array}\right)B_1=\left(\begin{array}{cc}1& 0\\ 0& 0\end{array}\right)B_2=\left(\begin{array}{cc}0& 0\\ 0& 1\end{array}\right)B_3=\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right)`$
$`S`$ $`=`$ $`\left(\begin{array}{cc}0& I\\ I& 0\end{array}\right)`$
$`\mathrm{\Sigma }`$ $`=`$ $`\left(\begin{array}{cc}\sigma & 0\\ 0& \sigma \end{array}\right)\sigma =\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right)`$
$`T`$ $`=`$ $`\left(\begin{array}{cc}\tau _+& 0\\ 0& \tau _{}\end{array}\right)\tau _+=\left(\begin{array}{cc}1& 1\\ 0& 1\end{array}\right)\tau _{}=\left(\begin{array}{cc}1& 0\\ 1& 1\end{array}\right).`$ (7.53)
L’azione sulle strutture di spin è data da
$$\left(\begin{array}{c}\stackrel{~}{\kappa }^{}\\ \stackrel{~}{\kappa }^{\prime \prime }\end{array}\right)=\left(\begin{array}{cc}D& C\\ B& A\end{array}\right)\left(\begin{array}{c}\kappa ^{}\\ \kappa ^{\prime \prime }\end{array}\right)+\frac{1}{2}\mathrm{diag}\left(\begin{array}{c}CD^T\\ AB^T\end{array}\right),$$
(7.54)
dove diag$`\left(M\right)`$ indica genericamente, per una matrice $`M`$ $`n\times n`$, un vettore colonna $`1\times n`$ i cui valori siano i valori diagonali di $`M`$. Sulla matrice periodica la trasformazione agisce come
$$\stackrel{~}{\mathrm{\Omega }}=\left(A\mathrm{\Omega }+B\right)\left(C\mathrm{\Omega }+D\right)^1,$$
(7.55)
Mentre le funzioni $`\vartheta `$ trasformano come
$$\vartheta \left[\stackrel{~}{\kappa }\right](\left\{\left(C\mathrm{\Omega }+D\right)^1\right\}^tv,\stackrel{~}{\mathrm{\Omega }})=ϵ(\kappa ,M)det\left(C\mathrm{\Omega }+D\right)^{\frac{1}{2}}e^{i\pi v^t\left(C\mathrm{\Omega }+D\right)^1Cv}\vartheta \left[\kappa \right](v,\mathrm{\Omega }).$$
(7.56)
Il fattore di fase $`ϵ(\kappa ,M)`$ dipende sia da $`\kappa `$ che dalla trasformazione modulare $`M`$, ed è tale che $`ϵ(\kappa ,M)^8=1`$.
Utilizzando le relazioni date, si trovano facilmente le leggi di trasformazione
$`d^3\stackrel{~}{\mathrm{\Omega }}_{IJ}`$ $`=`$ $`det\left(C\mathrm{\Omega }+D\right)^3d^3\mathrm{\Omega }_{IJ},`$
$`\vartheta \left[\stackrel{~}{\delta }\right]^4(0,\stackrel{~}{\mathrm{\Omega }})`$ $`=`$ $`ϵ^4det\left(C\mathrm{\Omega }+D\right)^2\vartheta \left[\delta \right]^4(0,\mathrm{\Omega }),`$
$`\mathrm{\Xi }_6\left[\stackrel{~}{\delta }\right]\left(\stackrel{~}{\mathrm{\Omega }}\right)`$ $`=`$ $`ϵ^4det\left(C\mathrm{\Omega }+D\right)^6\mathrm{\Xi }_6\left[\delta \right]\left(\mathrm{\Omega }\right),`$
$`\mathrm{\Psi }_{10}\left(\stackrel{~}{\mathrm{\Omega }}\right)`$ $`=`$ $`det\left(C\mathrm{\Omega }+D\right)^{10}\mathrm{\Psi }_{10}\left(\mathrm{\Omega }\right),`$ (7.57)
da cui si vede chiaramente che $`\mathrm{\Xi }_6\left[\delta \right]\left(\mathrm{\Omega }\right)`$ non trasforma come una forma modulare.
La legge di trasformazione modulare per la misura chirale è quindi
$$d\mu \left[\stackrel{~}{\delta }\right]\left(\stackrel{~}{\mathrm{\Omega }}\right)=det\left(C\mathrm{\Omega }+D\right)^5d\mu \left[\delta \right]\left(\mathrm{\Omega }\right).$$
(7.58)
Il peso $`5`$ è legato alla dimensione critica $`D=10`$, in maniera analoga a quanto avviene a genere uno. Questo risulta evidente dopo l’integrazione sui momenti interni, che porta alla comparsa di un fattore $`det\mathrm{Im}\mathrm{\Omega }`$, la cui legge di trasformazione modulare risulta essere
$$det\mathrm{Im}\stackrel{~}{\mathrm{\Omega }}=\left|det\left(C\mathrm{\Omega }+D\right)\right|^2det\mathrm{Im}\mathrm{\Omega }.$$
(7.59)
Si trova così che la misura completa per il doppio toro, tenendo conto dei fattori destri e sinistri è, come atteso, covariante sotto trasformazioni modulari,
$$\left(det\mathrm{Im}\stackrel{~}{\mathrm{\Omega }}\right)^5d\mu \left[\stackrel{~}{\delta }\right]\left(\stackrel{~}{\mathrm{\Omega }}\right)\times \overline{d\mu \left[\stackrel{~}{\overline{\delta }}\right]\left(\stackrel{~}{\mathrm{\Omega }}\right)}=\left(det\mathrm{Im}\mathrm{\Omega }\right)^5d\mu \left[\delta \right]\left(\mathrm{\Omega }\right)\times \overline{d\mu \left[\overline{\delta }\right]\left(\mathrm{\Omega }\right)}.$$
(7.60)
### 7.9 Proiezione GSO e funzione di partizione della Tipo II
Per implementare la proiezione GSO occorre sommare le misure chirali sulle strutture di spin con opportune fasi $`\eta _\delta `$ come nella (7.21),
$$d\mu \left(\mathrm{\Omega }\right)=\underset{\delta }{}\eta _\delta d\mu \left[\delta \right]\left(\mathrm{\Omega }\right),$$
(7.61)
in modo da ottenere una misura completa invariante sotto l’azione di $`Sp(4,𝐙)`$. Considerando le leggi di trasformazione delle misure chirali, di det Im($`\mathrm{\Omega }`$) e della misura $`d^3\mathrm{\Omega }_{IJ}`$, è possibile vedere che per la superstringa di Tipo II tutte le fasi devono essere uguali.
A questo punto possiamo scrivere il contributo al secondo ordine alla la funzione di partizione della Tipo II e quindi alla costante cosmologica, che risulta essere
$$Z_{\mathrm{II}}=__2\frac{\left|d^3\mathrm{\Omega }\right|^2}{\left(det\mathrm{Im}\mathrm{\Omega }\right)^5}\times \frac{\left|\underset{\delta }{}\mathrm{\Xi }_6\left[\delta \right]\left(\mathrm{\Omega }\right)\vartheta \left[\delta \right]^4(0,\mathrm{\Omega })\right|^2}{2^8\pi ^{12}\left|\mathrm{\Psi }_{10}\left(\mathrm{\Omega }\right)\right|^2}.$$
(7.62)
Si può dimostrare con considerazioni generali sulle proprietà delle forme modulari che l’integrando del contributo al secondo ordine alla costante cosmologica si annulla identicamente.
### 7.10 Degenerazioni delle ampiezze di genere 2
Nel calcolo delle ampiezze di superstringa un test importante è che queste obbediscano alle corrette fattorizzazioni negli stati fisici quando il world-sheet degenera. A genere due esistono due casi inequivalenti, a seconda che la degenerazione separi la superficie in due parti sconnesse o meno. Scegliendo un base di cicli canonici di omologia come in figura 7.1, e usando la seguente parametrizzazione della matrice dei periodi in questa base
$$\mathrm{\Omega }=\left(\begin{array}{cc}\tau _1& \tau \\ \tau & \tau _2\end{array}\right),$$
(7.63)
si vede facilmente che la *degenerazione separante* corrisponde al limite $`\tau 0`$, per $`\tau _1`$ e $`\tau _2`$ fissati, mentre la *degenerazione non separante* avviene nel limite $`\tau _2i\mathrm{}`$, per $`\tau _1`$ e $`\tau `$ fissati (si veda la figura 7.3).
#### 7.10.1 Degenerazione separante
L’andamento asintotico della misura d’integrazione si trova studiando l’effetto delle degenerazioni sulle funzioni $`\vartheta `$ a genere due. In questo limite è utile distinguere due casi. Il primo contiene 9 delle 10 strutture di spin $`\delta `$ per le quali le strutture di spin $`\mu _1`$ e $`\mu _2`$ sulle componeti connesse di genere 1 del doppio toro sono entrambe pari, si tratta del caso NS-NS. Il secondo è l’unico caso in cui entrambe le strutture di spin $`\nu _0`$ sono dispari, si tratta del caso R-R.
Per studiare la degenerazione si usa l’espansione di Taylor delle funzioni $`\vartheta `$ di genere 2 intorno a $`\tau =0`$, data in termini di funzioni $`\vartheta `$ di genere 1, che indichiamo per semplicità in questa sezione genericamente come $`\vartheta _1`$. Le espansioni nei due casi risultano essere
$`\vartheta \left[\begin{array}{c}\mu _1\\ \mu _2\end{array}\right](0,\mathrm{\Omega })`$ $`=`$ $`{\displaystyle \underset{p=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{\left(2\tau \right)^{2p}}{\left(2p\right)!}}_{\tau _1}^p\vartheta _1\left[\mu _1\right](0,\tau _1)_{\tau _2}^p\vartheta _1\left[\mu _2\right](0,\tau _2),`$
$`\vartheta \left[\begin{array}{c}\nu _0\\ \nu _0\end{array}\right](0,\mathrm{\Omega })`$ $`=`$ $`{\displaystyle \frac{1}{4\pi i}}{\displaystyle \underset{p=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{\left(2\tau \right)^{2p+1}}{\left(2p+1\right)!}}_{\tau _1}^p\vartheta _1^{}\left[\nu _0\right](0,\tau _1)_{\tau _2}^p\vartheta _1^{}\left[\nu _0\right](0,\tau _2).`$ (7.64)
I termini dominanti di queste espansioni per le $`\vartheta `$-costanti sono quindi
$`\vartheta \left[\begin{array}{c}\mu _1\\ \mu _2\end{array}\right](0,\mathrm{\Omega })`$ $`=`$ $`\vartheta _1\left[\mu _1\right](0,\tau _1)\vartheta _1\left[\mu _2\right](0,\tau _2)+𝒪\left(\tau ^2\right),`$
$`\vartheta \left[\begin{array}{c}\nu _0\\ \nu _0\end{array}\right](0,\mathrm{\Omega })`$ $`=`$ $`2i\pi \tau \eta \left(\tau _1\right)^3\eta \left(\tau _2\right)^3+𝒪\left(\tau ^3\right).`$ (7.65)
Si trova così che i limiti di $`\mathrm{\Psi }_{10}\left(\mathrm{\Omega }\right)`$ e $`\mathrm{\Xi }_6\left[\delta \right]\left(\mathrm{\Omega }\right)`$ sono dati da
$`\mathrm{\Psi }_{10}\left(\mathrm{\Omega }\right)`$ $`=`$ $`\left(2\pi \tau \right)^22^{12}\eta \left(\tau _1\right)^{24}\eta \left(\tau _2\right)^{24}+𝒪\left(\tau ^4\right),`$
$`\mathrm{\Xi }_6\left[\begin{array}{c}\mu _1\\ \mu _2\end{array}\right]\left(\mathrm{\Omega }\right)`$ $`=`$ $`2^8\mu _1|\nu _0\mu _2|\nu _0\eta \left(\tau _1\right)^{12}\eta \left(\tau _2\right)^{12}+𝒪\left(\tau ^2\right),`$
$`\mathrm{\Xi }_6\left[\begin{array}{c}\nu _0\\ \nu _0\end{array}\right]\left(\mathrm{\Omega }\right)`$ $`=`$ $`32^8\eta \left(\tau _1\right)^{12}\eta \left(\tau _2\right)^{12}+𝒪\left(\tau ^2\right).`$ (7.66)
Considerando tutti i contributi si trova che il limite della misura è
$`\mathrm{NS}\mathrm{NS}d\mu \left[\begin{array}{c}\mu _1\\ \mu _2\end{array}\right]`$ $`=`$ $`{\displaystyle \frac{d^3\tau }{\tau ^2}}{\displaystyle \underset{i=1,2}{}}{\displaystyle \frac{\mu _i|\nu _0\vartheta \left[\mu _i\right]^4\left(\tau _i\right)}{32\pi ^4\eta \left(\tau _i\right)^{12}}}+𝒪\left(\tau ^0\right),`$
$`\mathrm{R}\mathrm{R}d\mu \left[\begin{array}{c}\nu _0\\ \nu _0\end{array}\right]`$ $`=`$ $`{\displaystyle \frac{3\tau ^2d^3\tau }{2^6\pi ^4}}+𝒪\left(\tau ^4\right).`$ (7.67)
Il caso NS-NS riproduce correttamente i fattori ad un loop, incluse le fasi GSO introdotte dal limite di $`\mathrm{\Xi }_6\left[\delta \right]`$. Il prefattore $`\tau ^2`$ indica la presenza di uno stato intermedio tachionico. Combinando questa misura con la corrispondente misura destra e tenendo conto anche degli effetti dei momenti interni si trova che anche uno stato a massa nulla intermedio è presente. Una risommazione GSO parziale in una o nell’altra delle due parti sconnesse ad 1-loop cancella il tachione.
#### 7.10.2 Degenerazione non separante
Il caso non separante si studia in maniera analoga a quanto fatto fin qui. Si trova che
$`d\mu \left[\begin{array}{c}\mu _i\\ 00\end{array}\right]`$ $`=`$ $`+V_i(\tau ,\tau _1){\displaystyle \frac{d^3\tau }{q}}+𝒪\left(q^0\right),`$
$`d\mu \left[\begin{array}{c}\mu _i\\ 0\frac{1}{2}\end{array}\right]`$ $`=`$ $`V_i(\tau ,\tau _1){\displaystyle \frac{d^3\tau }{q}}+𝒪\left(q^0\right),`$
$`d\mu \left[\begin{array}{c}\mu _i\\ \frac{1}{2}0\end{array}\right]`$ $`=`$ $`𝒪\left(q^0\right),`$
$`d\mu \left[\begin{array}{c}\nu _0\\ \nu _0\end{array}\right]`$ $`=`$ $`𝒪\left(q^0\right).`$ (7.68)
Nelle prime tre linee, $`\mu _i`$ rappresenta una delle tre strutture di spin pari di genere 1 e $`V_i(\tau ,\tau _1)`$ è la funzione a due punti per il tachione sulla superficie degenere di genere 1. La singolarità $`q^1`$ corrisponde al tachione che attraverso il ciclo di omologia $`A_2`$ quando la struttura di spin nel secondo manico è o $`\left[00\right]`$ oppure $`\left[\mathrm{0\; 1}/2\right]`$, che corrispondono a condizioni al bordo NS. Una risommazione parziale sulle strutture di spin nel solo settore NS elimina la singolarità tachionica. D’altra parte, come atteso, nessun tachione appare per condizioni al bordo R.
### 7.11 Superstringa di Tipo 0B a genere 2
È possibile generalizzare i risultati ottenuti da D’Hoker e Phong per la superstringa di Tipo II, alla superstringa di Tipo 0 è possibile ricordando che $`\mathrm{\Xi }_6\left[\delta \right]`$ implementa le corrette fasi GSO per la Tipo II, come si vede chiaramente dalle (7.10.1). Per ottenere la funzione di partizione della Tipo $`0B`$, che come si ricorderà è una teoria di superstringa con rottura di supersimmetria esplicita, in analogia con quanto fatto a genere uno si deve prendere il modulo dei diversi oggetti nella sommatoria nella funzione di partizione della Tipo II (7.62). La funzione di partizione della Tipo 0B si può scrivere nella forma
$$Z_{0\mathrm{B}}^{g=2}=__2\frac{\left|d^3\mathrm{\Omega }\right|^2}{\left(det\mathrm{Im}\mathrm{\Omega }\right)^5}\times \frac{\underset{\delta }{}\left|\mathrm{\Xi }_6\left[\delta \right]\left(\mathrm{\Omega }\right)\right|^2\left|\vartheta \left[\delta \right]^4(0,\mathrm{\Omega })\right|^2}{2^8\pi ^{12}\left|\mathrm{\Psi }_{10}\left(\mathrm{\Omega }\right)\right|^2}.$$
(7.69)
Per verificare la correttezza della funzione di partizione proposta si può studiare la degenerazione separante, nel limite $`\tau 0`$. In questo limite ci si attende di ottenere due funzioni di partizione di genere uno della Tipo 0B, che come si ricorderà sono della forma,
$$Z_{0B}^{g=1}=\frac{d^2\tau }{\tau _2^2}\frac{\left|O_8\right|^2+\left|V_8\right|^2+\left|S_8\right|^2+\left|C_8\right|^2}{\tau _2^4\left(\eta \overline{\eta }\right)^8},$$
(7.70)
Scriviamo esplicitamente le strutture di spin, utlizzando una notazione leggermente differente da quella utilizzata da D’Hoker e Phong,
$$\kappa =\left(\begin{array}{cc}\alpha _1& \alpha _2\\ \beta _1& \beta _2\end{array}\right),$$
(7.71)
dove $`\alpha _i`$ e $`\beta _i`$ sono riferite alle strutture di spin dell’$`i`$-esima componente connessa del doppio toro. Le strutture di spin dispari sono,
$`\nu _1`$ $`=`$ $`\left(\begin{array}{cc}0& 1/2\\ 0& 1/2\end{array}\right),\nu _2=\left(\begin{array}{cc}1/2& 0\\ 1/2& 0\end{array}\right),\nu _3=\left(\begin{array}{cc}0& 1/2\\ 1/2& 1/2\end{array}\right),`$
$`\nu _4`$ $`=`$ $`\left(\begin{array}{cc}1/2& 0\\ 1/2& 1/2\end{array}\right),\nu _5=\left(\begin{array}{cc}1/2& 1/2\\ 0& 1/2\end{array}\right),\nu _6=\left(\begin{array}{cc}1/2& 1/2\\ 1/2& 0\end{array}\right),`$ (7.72)
mentre quelle dispari sono
$`\delta _1`$ $`=`$ $`\left(\begin{array}{cc}0& 0\\ 0& 0\end{array}\right),\delta _2=\left(\begin{array}{cc}0& 0\\ 0& 1/2\end{array}\right),\delta _3=\left(\begin{array}{cc}0& 0\\ 1/2& 0\end{array}\right),`$
$`\delta _4`$ $`=`$ $`\left(\begin{array}{cc}0& 0\\ 1/2& 1/2\end{array}\right),\delta _5=\left(\begin{array}{cc}0& 1/2\\ 0& 0\end{array}\right),\delta _6=\left(\begin{array}{cc}0& 1/2\\ 1/2& 0\end{array}\right),`$
$`\delta _7`$ $`=`$ $`\left(\begin{array}{cc}1/2& 0\\ 0& 0\end{array}\right),\delta _8=\left(\begin{array}{cc}1/2& 0\\ 0& 1/2\end{array}\right),\delta _9=\left(\begin{array}{cc}1/2& 1/2\\ 0& 0\end{array}\right),`$
$`\delta _0`$ $`=`$ $`\left(\begin{array}{cc}1/2& 1/2\\ 1/2& 1/2\end{array}\right).`$ (7.73)
Si vede quindi che, nella notazione usuale, la degenerazione delle funzioni $`\vartheta `$ a genere 2 nel limite $`\tau 0`$ si scrive
$`\vartheta \left[\delta _1\right](0,\mathrm{\Omega })`$ $`=`$ $`\vartheta _3(0,\tau _1)\vartheta _3(0,\tau _2)+𝒪\left(\tau ^2\right),`$
$`\vartheta \left[\delta _2\right](0,\mathrm{\Omega })`$ $`=`$ $`\vartheta _3(0,\tau _1)\vartheta _4(0,\tau _2)+𝒪\left(\tau ^2\right),`$
$`\vartheta \left[\delta _3\right](0,\mathrm{\Omega })`$ $`=`$ $`\vartheta _4(0,\tau _1)\vartheta _3(0,\tau _2)+𝒪\left(\tau ^2\right),`$
$`\vartheta \left[\delta _4\right](0,\mathrm{\Omega })`$ $`=`$ $`\vartheta _4(0,\tau _1)\vartheta _4(0,\tau _2)+𝒪\left(\tau ^2\right),`$
$`\vartheta \left[\delta _5\right](0,\mathrm{\Omega })`$ $`=`$ $`\vartheta _3(0,\tau _1)\vartheta _2(0,\tau _2)+𝒪\left(\tau ^2\right),`$
$`\vartheta \left[\delta _6\right](0,\mathrm{\Omega })`$ $`=`$ $`\vartheta _4(0,\tau _1)\vartheta _2(0,\tau _2)+𝒪\left(\tau ^2\right),`$
$`\vartheta \left[\delta _7\right](0,\mathrm{\Omega })`$ $`=`$ $`\vartheta _2(0,\tau _1)\vartheta _3(0,\tau _2)+𝒪\left(\tau ^2\right),`$
$`\vartheta \left[\delta _8\right](0,\mathrm{\Omega })`$ $`=`$ $`\vartheta _2(0,\tau _1)\vartheta _4(0,\tau _2)+𝒪\left(\tau ^2\right),`$
$`\vartheta \left[\delta _9\right](0,\mathrm{\Omega })`$ $`=`$ $`\vartheta _2(0,\tau _1)\vartheta _2(0,\tau _2)+𝒪\left(\tau ^2\right),`$
$`\vartheta \left[\delta _0\right](0,\mathrm{\Omega })`$ $`=`$ $`\vartheta _1(0,\tau _1)\vartheta _1(0,\tau _2)2i\pi \tau \eta \left(\tau _1\right)^3\eta \left(\tau _2\right)^3+𝒪\left(\tau ^2\right),`$ (7.74)
dove naturalmente nell’ultimo termine $`\vartheta _1(0,\tau _1)=0`$. Osservando che il modulo di $`\mathrm{\Xi }_6\left[\delta _i\right]\left(\mathrm{\Omega }\right)`$ cancella le fasi,
$$\left|\mathrm{\Xi }_6\left[\delta _i\right]\left(\mathrm{\Omega }\right)\right|^2=\left|2^8\eta \left(\tau _1\right)^{12}\eta \left(\tau _2\right)^{12}\right|^2+𝒪\left(\tau ^4\right),$$
(7.75)
dove nel caso di $`\delta _0`$ si è operato un riscalamento del fattore 3 di $`\mathrm{\Xi }_6`$, e ricordando che si era trovato
$$\mathrm{\Psi }_{10}\left(\mathrm{\Omega }\right)=\left(2\pi \tau \right)^22^{12}\eta \left(\tau _1\right)^{24}\eta \left(\tau _2\right)^{24}+𝒪\left(\tau ^4\right),$$
(7.76)
si può calcolare facilmente la degenerazione della funzione di partizione. Nel limite nel limite $`\tau 0`$ si trova in questo modo, il prodotto di due funzioni di partizione di Tipo 0B a genere 1. Questo dimostra la correttezza della funzione di partizione a genere 2 proposta.
### 7.12 Discendenti Aperti della Tipo 0’B a genere $`3/2`$
#### 7.12.1 Ampiezze di genere $`3/2`$
Come si è visto, le funzioni di partizione di stringa aperta ricevono contributi da superfici con bordi e crosscap. Questo tipo di superfici ammettono sempre come ricoprimento doppio superfici chiuse e orientabili. L’immersione nel ricoprimento doppio è descritta per mezzo di involuzioni che generalizzano relazioni di riflessione. Queste involuzioni invertono la matrice di intersezione e per questo sono dette anti-conformi, e agiscono per mezzo di matrici che, con un’opportuna scelta di base, possono essere sempre scritte nella forma
$$I=\left(\begin{array}{cc}I& 0\\ \mathrm{\Delta }& I\end{array}\right),$$
(7.77)
con $`\mathrm{\Delta }`$ una matrice simmetrica. Il caso di genere 1, discusso in dettaglio, è illuminante. L’anello, la bottiglia di Klein e il nastro di Möbius, come si è visto, ammettono il toro come ricoprimento doppio, via involuzioni anti-conformi con $`\mathrm{\Delta }=0`$ nei primi due casi e $`\mathrm{\Delta }=1`$ nell’ultimo. Questo risultato si estende a tutti i generi e permette un trattamento unificato delle varie superfici di interesse .
Partendo da una base canonica di omologia di cicli definita dai cicli $`A_I`$ e $`B_I`$, con matrice di intersezione
$$J=\left(\begin{array}{cc}0& I\\ I& 0\end{array}\right),$$
(7.78)
la matrice di involuzione soddisferà la relazione
$$I^TJI=J,$$
(7.79)
dal momento che inverte l’orientazione della superficie ricoprente. In generale la matrice di involuzione per una superficie di genere $`g`$ ha forma a blocchi,
$$I=\left(\begin{array}{cc}A& B\\ C& D\end{array}\right),$$
(7.80)
con $`A`$, $`B`$, $`C`$ e $`D`$ matrici $`g\times g`$ di interi che soddisfano le condizioni
$$AB^T=BA^T,CD^T=DC^T,eAD^TBC^T=1,$$
(7.81)
che risultano dalla (7.79). Non è difficile mostrare che una matrice dei periodi $`\mathrm{\Omega }`$ compatibile con l’involuzione I deve soddisfare la condzione
$$\overline{\mathrm{\Omega }}=I\left(\mathrm{\Omega }\right)=\left(C+D\mathrm{\Omega }\right)\left(A+B\mathrm{\Omega }\right)^1.$$
(7.82)
Una scelta opportuna della base di omologia semplifica la forma della matrice dei periodi, e si può dimostrare che è sempre possibile porre la matrice di involuzione nella forma (7.77), . La matrice $`\mathrm{\Delta }`$ è molto importante dal momento che determina la forma della parte reale della matrice dei periodi $`\mathrm{\Omega }`$.
Per ottenere questa forma della matrice di involuzione occorre scegliere dei cicli $`a_i`$ che siano lasciati invarianti dall’involuzione. A tal fine è utile ricordare che l’equivalenza fra tre crosscap e un manico e un crosscap riduce le superfici in tre classi con rispettivamente 0, 1 e 2 crosscap. Nel caso in cui non si abbiano crosscap, l’involuzione lascia fissi un certo numero di bordi. I cicli $`a_i`$ posssono essere scelti in parte come i cicli lungo questi bordi e in parte come somme di cicli di coppie di manici scambiati dall’involuzione. I cicli $`b_i`$ sono i cicli lungo i manici che contribuiscono alla $`\mathrm{\Delta }`$, dal momento che trasformano in maniera non banale sotto l’involuzione. Nel caso di superfici con crosscap, ricordando che un crosscap e un bordo con punti opposti identificati, si capisce che un ciclo che sia identificato con un crosscap induce una rivoluzione intorno a se stesso nella trasformazione del ciclo omologico coniugato. Nel caso della striscia di Möbius questo argomento porta alla matrice dei periodi nota, non puramente immaginaria.
È possibile operare trasformazioni della base in omologia mantenendo la forma triangolare della matrice di involuzione. In generale una trasformazione di questo tipo sarà della forma
$$M=\left(\begin{array}{cc}A& 0\\ C& \left(A^T\right)^1\end{array}\right),$$
(7.83)
con $`detA=\pm 1`$ e
$$2C=\mathrm{\Delta }A\left(A^T\right)^1\mathrm{\Delta },$$
(7.84)
una matrice di interi pari. Trasformazioni di quiesto tipo sono banali a genere 1, ma non a genere più alto, dove sostanzialmente scambiano tori vicini.
In questa nuova base di omologia, detta *base identità*, la matrice dei periodi può essere scritta nella forma particolarmente semplice,
$$\mathrm{\Omega }=\frac{\mathrm{\Delta }}{2}+i\mathrm{\Omega }_2$$
(7.85)
con $`\mathrm{\Omega }_2`$ una matrice a valori reali.
Esistono cinque superfici di genere $`3/2`$ con ricoprimento doppio di genere due, che possono essere identificate per mezzo di triple di numeri che indicano nell’ordine il numero di manici, di bordi e di crosscap. Ad esempio, una superficie con tre bordi è indicata come $`\left(030\right)`$, mentre un toro con un bordo con (110). Nella base identià le matrici $`\mathrm{\Delta }`$ che definiscono le involuzioni sono
$`\mathrm{\Delta }_{\left(110\right)}`$ $`=`$ $`\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right),\mathrm{\Delta }_{\left(101\right)}=\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right),\mathrm{\Delta }_{\left(030\right)}=\left(\begin{array}{cc}0& 0\\ 0& 0\end{array}\right),`$
$`\mathrm{\Delta }_{\left(012\right)}`$ $`=`$ $`\left(\begin{array}{cc}1& 1\\ 1& 0\end{array}\right),\mathrm{\Delta }_{\left(012\right)}^{}=\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right),\mathrm{\Delta }_{\left(012\right)}^{\prime \prime }=\left(\begin{array}{cc}0& 1\\ 1& 1\end{array}\right),`$
$`\mathrm{\Delta }_{\left(021\right)}`$ $`=`$ $`\left(\begin{array}{cc}1& 0\\ 0& 0\end{array}\right),\mathrm{\Delta }_{\left(021\right)}^{}=\left(\begin{array}{cc}1& 1\\ 1& 1\end{array}\right),\mathrm{\Delta }_{\left(021\right)}^{\prime \prime }=\left(\begin{array}{cc}0& 0\\ 0& 1\end{array}\right),`$ (7.86)
dove per la bottiglia di Klein con un bordo $`\left(012\right)`$ si distinguono tre casi, collegati da trasfortmazioni modulari in relazione alla posizione del bordo rispetto ai crosscap. Allo stesso nel caso della striscia di Möbius con un bordo $`\left(021\right)`$ si distinguono tre casi in relazione alla posizione del crosscap rispetto ai bordi.
#### 7.12.2 Contributi alla funzione di partizione a genere $`3/2`$
Per studiare la sistematica delle ampiezze di genere $`3/2`$ in un caso con supersimmetria spazio-temporale rotta, un esempio interessante è fornito dal modello non tachionico 0’B, costruito a genere 1 a partire dall’ampiezza di Klein, indicata come $`𝒦_3`$. In particolare si utilizzerà questo modello nel caso particolarmente semplice in cui si fissi $`n_v=N`$, $`n_s=\overline{N}`$, $`n_o=0`$, $`n_c=0`$, ottenendo nel canale diretto le ampiezze di genere 1,
$`𝒦_3`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left(O_8+V_8+S_8+C_8\right),`$
$`𝒜_3`$ $`=`$ $`N\overline{N}V_8{\displaystyle \frac{1}{2}}\left(N^2+\overline{N}^2\right)C_8,`$
$`_3`$ $`=`$ $`{\displaystyle \frac{N+\overline{N}}{2}}\widehat{C}_8,`$ (7.87)
dove come di solito la notazione lascia intesi i gradi di libertà bosonici e la misura di integrazione. Nel canale trasverso si ottiene,
$`\stackrel{~}{𝒦}_3`$ $`=`$ $`{\displaystyle \frac{2^6}{2}}C_8,`$
$`\stackrel{~}{𝒜}_3`$ $`=`$ $`{\displaystyle \frac{2^6}{2}}\left[\left(N^2+\overline{N}^2\right)\left(V_8C_8\right)\left(N\overline{N}\right)^2\left(O_8S_8\right)\right],`$
$`\stackrel{~}{}_3`$ $`=`$ $`\left(N+\overline{N}\right)\widehat{C}_8.`$ (7.88)
Per calcolare le correzioni dovute ai diagrammi di genere 3/2 di questo modello, occorre partire da una generica ampiezza di genere 3/2 che, può essere ottenuta con semplici considerazioni a partire dall’ampiezza di stringa chiusa di genere 2. In particolare si considerano solo metà dei gradi di libertà, in analogia con il caso di genere uno, e si introducono dei coefficienti generici $`C_\delta ^{g=3/2}`$ che devono essere opportunamente fissati in relazione alla diverse ampiezze,
$$Z^{g=3/2}=d^3\mathrm{\Omega }\frac{1}{\mathrm{\Psi }_{10}\left(\mathrm{\Omega }\right)}\underset{\delta }{}C_\delta ^{g=3/2}\mathrm{\Xi }_6\left[\delta \right]\left(\mathrm{\Omega }\right)\vartheta \left[\delta \right]^4(0,\mathrm{\Omega }).$$
(7.89)
Questa è un ampiezza di genere 3/2 nel canale di stringhe chiuse (il canale trasverso), dal momento che si è scelto di non introdurre potenze della matrice dei periodi al denominatore, ancora in analogia con il canale trasverso di genere 1.
Il caso più semplice è quello dell’ampiezza a tre bordi. Per fissare i coefficienti consideriamo l’ampiezza (7.89) nel limte $`\tau i\mathrm{}`$. In questo limite ci si aspetta che la funzione di partizione fattorizzi in due ampiezze trasverse di anello. Utilizzando il comportamento asintotico per $`\tau i\mathrm{}`$ delle funzioni $`\vartheta `$ di genere 2 (7.11), di $`\mathrm{\Psi }_{10}\left(\mathrm{\Omega }\right)`$ e di $`\mathrm{\Xi }_6\left[\delta \right]\left(\mathrm{\Omega }\right)`$ si ricava il limite della (ref32open) che può essere confrontato con il contenuto in termini di funzioni $`\vartheta `$ di genere 1 del generico prodotto $`\stackrel{~}{𝒜}\left(\tau _1\right)\stackrel{~}{𝒜}\left(\tau _2\right)`$. Si trova così che
$`C_{\delta _1}^{\left(030\right)}`$ $`=`$ $`C_{\delta _5}^{\left(030\right)}=C_{\delta _7}^{\left(030\right)}=C_{\delta _9}^{\left(030\right)}=4N\overline{N}M\overline{M},`$
$`C_{\delta _2}^{\left(030\right)}`$ $`=`$ $`C_{\delta _8}^{\left(030\right)}=2N\overline{N}\left(M^2+\overline{M}^2\right),`$
$`C_{\delta _3}^{\left(030\right)}`$ $`=`$ $`C_{\delta _6}^{\left(030\right)}=2M\overline{M}\left(N^2+\overline{N}^2\right),`$
$`C_{\delta _4}^{\left(030\right)}`$ $`=`$ $`C_{\delta _0}^{\left(030\right)}=\left(N^2+\overline{N}^2\right)\left(M^2+\overline{M}^2\right),`$ (7.90)
Per verificare la correttezza dell’identificazione si può studiare il comportamento dell’ampiezza con tre bordi trovata sotto una trasformazione modulare S (7.8) e calcolarne il limite di degenerazione separante. Dal momento che questa ampiezza ammette una matrice dei periodi puramente immaginaria, l’azione di $`S`$ sarà
$$S:\mathrm{\Omega }\mathrm{\Omega }^1=\frac{i}{\tau _1\tau _2\tau ^2}\left(\begin{array}{cc}\tau _1& \tau \\ \tau & \tau _2\end{array}\right),$$
(7.91)
dove adesso i $`\tau _i`$ sono reali. Nel limite $`\tau _2\mathrm{}`$ questa trasformazione equivale a due trasformazioni S su ciascuno dei due tori di genere 1,
$$\mathrm{\Omega }\left(\begin{array}{cc}\frac{i}{\tau _1}& 0\\ 0& \frac{i}{\tau _2}\end{array}\right).$$
(7.92)
Ci si aspetta quindi di ottenere la corretta fattorizzazione di due ampiezze di anello nel canale diretto.
Per procedere nel calcolo occorre specializzare le leggi di trasformazione modulari date nei paragrafi precedenti al caso di S. Per le strutture di spin si trova,
$`S:\{\begin{array}{ccc}\stackrel{~}{\delta }_1=\delta _1& \stackrel{~}{\delta }_2=\delta _5& \stackrel{~}{\delta }_3=\delta _7\\ \stackrel{~}{\delta }_4=\delta _9& \stackrel{~}{\delta }_5=\delta _2& \stackrel{~}{\delta }_6=\delta _8\\ \stackrel{~}{\delta }_7=\delta _3& \stackrel{~}{\delta }_8=\delta _6& \stackrel{~}{\delta }_9=\delta _4\\ \stackrel{~}{\delta }_0=\delta _0& & \end{array}`$ (7.93)
Le trasformazioni per gli altri oggetti che compaiono nell’ampiezza con tre bordi sono
$`S:\{\begin{array}{ccc}d^3\stackrel{~}{\mathrm{\Omega }}_{IJ}& =& det\left(\mathrm{\Omega }\right)^3d^3\mathrm{\Omega }\\ \vartheta \left[\stackrel{~}{\delta }\right]^4(0,\stackrel{~}{\mathrm{\Omega }})& =& det\left(\mathrm{\Omega }\right)^2\vartheta \left[\delta \right]^4(0,\mathrm{\Omega })\\ \mathrm{\Xi }_6\left[\stackrel{~}{\delta }\right]\left(\stackrel{~}{\mathrm{\Omega }}\right)& =& det\left(\mathrm{\Omega }\right)^6\mathrm{\Xi }_6\left[\delta \right]\left(\mathrm{\Omega }\right)\\ \mathrm{\Psi }_{10}\left(\stackrel{~}{\mathrm{\Omega }}\right)& =& det\left(\mathrm{\Omega }\right)^{10}\mathrm{\Psi }_{10}\left(\mathrm{\Omega }\right)\end{array}.`$ (7.94)
Si può quindi operare una trasformazione S dell’ampiezze e calcolarne successivamente il limite nella degenerazione separante con le regole note,
$`Z^{\left(030\right)}`$ $``$ $`{\displaystyle }{\displaystyle \frac{d\tau }{\left(2\pi \right)^2\tau ^2}}{\displaystyle }{\displaystyle \frac{d\tau _1d\tau _2}{\tau _1^5\tau _2^5}}{\displaystyle \frac{1}{\eta ^8\left(\tau _1\right)\eta ^8\left(\tau _2\right)}}\{N\overline{N}M\overline{M}\left[{\displaystyle \frac{\vartheta _3^4\vartheta _4^4}{2\eta ^4}}\right]\left(\tau _1\right)\left[{\displaystyle \frac{\vartheta _3^4\vartheta _4^4}{2\eta ^4}}\right]\left(\tau _2\right)`$ (7.95)
$`+`$ $`N\overline{N}{\displaystyle \frac{M^2+\overline{M}^2}{2}}\left[{\displaystyle \frac{\vartheta _3^4\vartheta _4^4}{2\eta ^4}}\right]\left(\tau _1\right)\left[{\displaystyle \frac{\vartheta _2^4\vartheta _1^4}{2\eta ^4}}\right]\left(\tau _2\right)`$
$`+`$ $`M\overline{M}{\displaystyle \frac{N^2+\overline{N}^2}{2}}\left[{\displaystyle \frac{\vartheta _2^4\vartheta _1^4}{2\eta ^4}}\right]\left(\tau _1\right)\left[{\displaystyle \frac{\vartheta _3^4\vartheta _4^4}{2\eta ^4}}\right]\left(\tau _2\right)`$
$``$ $`{\displaystyle \frac{N^2+\overline{N}^2}{2}}{\displaystyle \frac{M^2+\overline{M}^2}{2}}\left[{\displaystyle \frac{\vartheta _2^4\vartheta _1^4}{2\eta ^4}}\right]\left(\tau _1\right)\left[{\displaystyle \frac{\vartheta _2^4\vartheta _1^4}{2\eta ^4}}\right]\left(\tau _2\right)\}.`$
L’espressione trovata corrisponde come atteso al prodotto di due ampiezze di anello nel canale diretto.
Lo studio delle altre ampiezze risulta più complesso in ragione della loro matrice dei periodi non puramente immaginaria, ma il metodo per l’identificazione dei coefficienti resta sostanzialmente invariato.
La superficie con un bordo e un manico (figura 7.4), che si ottiene tramite una semplice involuzione di riflessione dal doppio toro (figura 7.5) può essere studiata sulla scorta di quanto già fatto. La funzione di partizione viene scritta nella forma
$$Z^{\left(110\right)}=d^3\mathrm{\Omega }\frac{1}{\mathrm{\Psi }_{10}\left(\mathrm{\Omega }\right)}\underset{\delta }{}C_\delta ^{\left(110\right)}\mathrm{\Xi }_6\left[\delta \right]\left(\mathrm{\Omega }\right)\vartheta \left[\delta \right]^4(0,\mathrm{\Omega }).$$
(7.96)
In una base canonica di omologia, l’involuzione può essere scritta nella forma
$$I=\left(\begin{array}{cc}\sigma _1& 0\\ 0& \sigma \end{array}\right),$$
(7.97)
come ci si rende conto facilmente osservando la figura 7.5. Il vincolo 7.82 permette di scrivere la matrice dei periodi nella forma
$$\mathrm{\Omega }=\left(\begin{array}{cc}\tau _1& i\tau \\ i\tau & \overline{\tau }_1\end{array}\right),$$
(7.98)
con $`\tau `$ è reale. La forma della matrice dei periodi è tale che, nel limite di degenerazione separante $`\tau 0`$ in cui il digramma si presenta come un toro con un tadpole su un bordo, si ritrovino, come desiderato, i settori di stringa chiusa della funzione di partizione del toro della OB. Per rendersene conto si osservi che in questo limite
$`\vartheta \left[\begin{array}{cc}\alpha _1& \alpha _2\\ \beta _1& \beta _2\end{array}\right](0,\mathrm{\Omega })`$ $`=`$ $`\vartheta \left[\begin{array}{c}\alpha _1\\ \beta _1\end{array}\right](0,\tau _1)\left[\begin{array}{c}\alpha _2\\ \beta _2\end{array}\right](0,\overline{\tau }_1)+𝒪\left(\tau ^2\right)`$ (7.99)
$`=`$ $`\vartheta \left[\begin{array}{c}\alpha _1\\ \beta _1\end{array}\right](0,\tau _1)\vartheta \left[\begin{array}{c}\alpha _2\\ \beta _2\end{array}\right](0,\overline{\tau }_1)+𝒪\left(\tau ^2\right)`$
$`=`$ $`\vartheta \left[\begin{array}{c}\alpha _1\\ \beta _1\end{array}\right](0,\tau _1)\overline{\vartheta \left[\begin{array}{c}\alpha _2\\ \beta _2\end{array}\right](0,\tau _1)}+𝒪\left(\tau ^2\right),`$
come è possibile verificare a partire dalla definizione della funzione $`\vartheta `$.
In questo limite ci si aspetta che siano presenti solo i settori NS-NS, dal momento che sul toro per funzioni ad un punto i settori R-R sono zero,
$$Z\left(N+\overline{N}\right)\left(V_8^2+O_8^2\right).$$
(7.100)
Si possono fissare in questo modo i coefficienti della ampiezza trovando
$$C_{\delta _1}^{\left(110\right)}=C_{\delta _4}^{\left(110\right)}=N+\overline{N},\mathrm{tutti}\mathrm{gli}\mathrm{altri}C_{\delta _i}^{\left(110\right)}=0.$$
(7.101)
Con considerazioni analoghe si può studiare il diagramma con un manico e un crosscap (101), che equivale ad un diagramma con tre crosscap.
Gli ultimi diagrammi su cui, allo stato attuale della ricerca, sono in grado di fare alcune considerazioni sono la prima e la terza versione del diagramma con due bordi e un crosscap (021) (il caso con un crosscap centrale presenta sottigliezze nella degenerazione separante). Questo tipo di diagramma può essere studiato in stretta analogia con quanto fatto per il diagramma con tre bordi: se ne fissano i coefficienti studiandone la fattorizzazione nel limite di degenerazione separante e poi se ne verifica la corretta attribuzione passando al canale diretto e studiandone nuovamente la degenerazione separante.
In questo caso ci si aspetta di trovare il prodotto di una striscia di Möbius per un anello. Consideriamo il primo caso. Studiando il limite $`\tau \mathrm{}`$ nel canale diretto e confrontando con il prodotto $`\stackrel{~}{}\times \stackrel{~}{𝒜}`$, si fissano i coefficienti
$`C_{\delta _0}^{\left(021\right)}`$ $`=`$ $`\left(N^2+\overline{N}^2\right)\left(M+\overline{M}\right),C_{\delta _7}^{\left(021\right)}=\left(N^2+\overline{N}^2\right)\left(M+\overline{M}\right),`$
$`C_{\delta _8}^{\left(021\right)}`$ $`=`$ $`\left(N^2+\overline{N}^2\right)\left(M+\overline{M}\right),C_{\delta _9}^{\left(021\right)}=2N\overline{N}\left(M+\overline{M}\right),`$
$`C_{\delta _1}^{\left(021\right)}`$ $`=`$ $`C_{\delta _2}^{\left(021\right)}=C_{\delta _3}^{\left(021\right)}=C_{\delta _4}^{\left(021\right)}=C_{\delta _5}^{\left(021\right)}=0.`$ (7.102)
La complicazione di questo diagramma rispetto al caso con tre bordi risiede nella corretta individuzione della matrice che permette il passaggio dal canale trasverso a quello diretto. Ricordando che nel canale trasverso $`I\left(a_i\right)=a_i`$ mentre nel canale diretto $`I\left(a_i\right)=a_i`$. Si può individuare la trasformazione che collega le due rappresentazioni che, nella base identità di omologia, risulta essere
$$M=\left(\begin{array}{cccc}1& 0& 0& 0\\ 0& 0& 0& 1\\ 2& 0& 1& 0\\ 0& 1& 0& 0\end{array}\right).$$
(7.103)
Studiando l’azione di questa trasformazione sulle strutture di spin si trova
$`M:\{\begin{array}{ccc}\stackrel{~}{\delta }_1=\delta _1& \stackrel{~}{\delta }_2=\delta _5& \stackrel{~}{\delta }_3=\delta _3\\ \stackrel{~}{\delta }_4=\delta _6& \stackrel{~}{\delta }_5=\delta _2& \stackrel{~}{\delta }_6=\delta _4\\ \stackrel{~}{\delta }_7=\delta _7& \stackrel{~}{\delta }_8=\delta _9& \stackrel{~}{\delta }_9=\delta _8\\ \stackrel{~}{\delta }_0=\delta _0& & \end{array}`$ (7.104)
Ricordando le leggi di trasformazione di $`\mathrm{\Psi }_{10}\left(\mathrm{\Omega }\right)`$, di $`\mathrm{\Xi }_6\left[\delta \right]\left(\mathrm{\Omega }\right)`$ e di $`d^3\mathrm{\Omega }`$, è possibile verificare la correttezza della scelta dei coefficienti.
### 7.13 Un ansatz per i campi magnetici a genere 2
A titolo di esempio delle possibili applicazioni della tecnica sviluppata consideriamo una proposta di ampiezza con tre bordi con l’introduzione di campi magnetici interni. Questo modello tutt’ora oggetto di studio, presenta sottigliezze non ancora del tutto chiarite. L’ansatz per l’ampiezza, in analogia con il caso noto di genere 1, è
$`Z_B^{\left(030\right)}`$ $``$ $`{\displaystyle \underset{a,b,c}{}}{\displaystyle \frac{\pi ^2\left(q_a+q_c\right)\left(q_b+q_c\right)B^2}{\sqrt{1+\pi ^2q_c^2B^2}}}{\displaystyle d^3\mathrm{\Omega }}`$ (7.105)
$`{\displaystyle \frac{_{z_1}\vartheta \left[\begin{array}{cc}1/2& 0\\ 1/2& 0\end{array}\right]\left(0|\mathrm{\Omega }\right)_{z_2}\vartheta \left[\begin{array}{cc}0& 1/2\\ 0& 1/2\end{array}\right]\left(0|\mathrm{\Omega }\right)}{\vartheta \left[\begin{array}{cc}1/2& 0\\ 1/2& 0\end{array}\right](ϵ_a+ϵ_c,0|\mathrm{\Omega })\vartheta \left[\begin{array}{cc}0& 1/2\\ 0& 1/2\end{array}\right](0,ϵ_b+ϵ_c|\mathrm{\Omega })}}\times `$
$`{\displaystyle \underset{\delta }{}}{\displaystyle \frac{\mathrm{\Xi }_6\left[\delta \right]\left(\mathrm{\Omega }\right)\vartheta \left[\delta \right](ϵ_a+ϵ_c,ϵ_b+ϵ_c|\mathrm{\Omega })\vartheta ^3\left[\delta \right](0,0|\mathrm{\Omega })}{\mathrm{\Psi }_{10}\left(\mathrm{\Omega }\right)}}.`$
Ci si aspetta, nel limite di degenerazione non separante $`\tau _2i\mathrm{}`$, in cui si allontana il bordo $`b`$ all’infinito, di ottenere un espressione della forma,
$$\left(q_a+q_c\right)B\underset{\alpha ,\beta }{}\frac{\eta _{\alpha \beta }\vartheta \left[\begin{array}{c}\alpha _2\\ \beta _2\end{array}\right]\left(ϵ_a+ϵ_c|\tau _1\right)\vartheta ^3\left[\begin{array}{c}\alpha _2\\ \beta _2\end{array}\right]\left(0|\tau _1\right)}{\vartheta _1\left[\begin{array}{c}\alpha _2\\ \beta _2\end{array}\right]\left(0|\tau _1\right)\eta ^9\left(\tau _1\right)}\sqrt{1+\pi ^2q_b^2B^2}$$
(7.106)
I calcoli fatti, che risultano molto complicati, e richiedono un utilizzo intenso delle proprietà delle funzioni $`\vartheta `$, mostrano che nel limite $`\tau _2i\mathrm{}`$ si trova una espressione simile alla (7.106), insieme però a termini in cui campo magnetico compare in forma esponenziale che risultano piuttosto sottili da interpretare.
## Capitolo 8 Conclusioni
In questo lavoro di Tesi si è iniziato il lavoro di generalizzazione dei risultati ottenuti da D’Hoker e Phong per le superstrighe chiuse di Tipo II e Eterotiche agli altri altri modelli di stringhe con supersimmetria rotta (modelli di tipo 0 e modelli con “brane supersymmetry breaking” in dieci dimensioni). In particolare si è riusciti a studiare la superstringa 0’B individuando tecniche utili alla trattazione degli altri modelli.
Il lavoro svolto, frutto di una collaborazione in atto con il Dr. Carlo Angelantonj dell’Università di Torino, il Prof. Emilian Dudas dell’Ecole Polytechnique e il mio relatore di Tesi, mostra la consistenza della costruzione di orientifold a genere 3/2 e rappresenta il punto di partenza per lo studio sistematico delle ridefinizioni di vuoto introdotte a due loops dalla rottura della supersimmetria, in vista anche della costruzione di un’espressione generale per le correzioni di soglia a due loops.
## Appendice A Funzioni Theta
##### Definizione
Le funzioni $`\vartheta `$ di Jacobi sono definite in termini di una somma infinita
$$\vartheta \left[\genfrac{}{}{0pt}{}{\alpha }{\beta }\right]\left(z|\tau \right)=\underset{n}{}q^{\frac{1}{2}\left(n+\alpha \right)^2}e^{2\pi i\left(n+\alpha \right)\left(z+\beta \right)},$$
(A.1)
dove $`\alpha ,\beta `$ e $`q=e^{2\pi i\tau }`$, o in maniera equivalente come
$`\vartheta \left[\genfrac{}{}{0pt}{}{\alpha }{\beta }\right]\left(z|\tau \right)`$ $`=`$ $`e^{2i\pi \alpha \left(z+\beta \right)}q^{\alpha ^2/2}{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}\left(1q^n\right){\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}\left(1+q^{n+\alpha 1/2}e^{2i\pi \left(z+\beta \right)}\right)`$ (A.2)
$`\times {\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}(1+q^{n\alpha 1/2}e^{2i\pi \left(z+\beta \right)}),`$
in particolare
$`\vartheta \left[\genfrac{}{}{0pt}{}{0}{0}\right]\left(z|\tau \right)=\vartheta _3\left(z|\tau \right)={\displaystyle \underset{m=1}{\overset{\mathrm{}}{}}}\left(1q^m\right)\left(1+e^{2\pi iz}q^{m1/2}\right)\left(1+e^{2\pi iz}q^{m1/2}\right),`$
$`\vartheta \left[\genfrac{}{}{0pt}{}{0}{1/2}\right]\left(z|\tau \right)=\vartheta _4\left(z|\tau \right)={\displaystyle \underset{m=1}{\overset{\mathrm{}}{}}}\left(1q^m\right)\left(1e^{2\pi iz}q^{m1/2}\right)\left(1e^{2\pi iz}q^{m1/2}\right),`$
$`\vartheta \left[\genfrac{}{}{0pt}{}{1/2}{0}\right]\left(z|\tau \right)=\vartheta _2\left(z|\tau \right)=2q^{1/8}\mathrm{cos}\left(\pi z\right)`$
$`\times {\displaystyle \underset{m=1}{\overset{\mathrm{}}{}}}(1q^m)(1+e^{2\pi iz}q^m)(1+e^{2\pi iz}q^m),`$
$`\vartheta \left[\genfrac{}{}{0pt}{}{1/2}{1/2}\right]\left(z|\tau \right)=\vartheta _1\left(z|\tau \right)=2q^{1/8}\mathrm{sin}\left(\pi z\right)`$
$`\times {\displaystyle \underset{m=1}{\overset{\mathrm{}}{}}}(1q^m)(1e^{2\pi iz}q^m)(1e^{2\pi iz}q^m).`$ (A.3)
##### Trasformazioni modulari
$$\vartheta \left[\genfrac{}{}{0pt}{}{\alpha }{\beta }\right]\left(z|\tau +1\right)=e^{i\pi \alpha \left(\alpha 1\right)}\vartheta \left[\genfrac{}{}{0pt}{}{\alpha }{\beta +\alpha 1/2}\right]\left(z|\tau \right),$$
(A.4)
$$\vartheta \left[\genfrac{}{}{0pt}{}{\alpha }{\beta }\right]\left(\frac{z}{\tau }|\frac{1}{\tau }\right)=\left(i\tau \right)^{1/2}e^{2i\pi \alpha \beta +i\pi z^2/\tau }\vartheta \left[\genfrac{}{}{0pt}{}{\beta }{\alpha }\right]\left(z|\tau \right).$$
(A.5)
in particolare si ha
$`\vartheta _3\left(z|\tau +1\right)=\vartheta _4\left(z|\tau +1\right),`$
$`\vartheta _4\left(z|\tau +1\right)=\vartheta _3\left(z|\tau +1\right),`$
$`\vartheta _2\left(z|\tau +1\right)=q^{1/8}\vartheta _2\left(z|\tau +1\right),`$
$`\vartheta _1\left(z|\tau +1\right)=q^{1/8}\vartheta _1\left(z|\tau +1\right),`$ (A.6)
$`\vartheta _3\left(z/\tau |1/\tau \right)=\sqrt{i\tau }e^{\pi iz^2/\tau }\vartheta _3\left(z|\tau \right),`$
$`\vartheta _4\left(z/\tau |1/\tau \right)=\sqrt{i\tau }e^{\pi iz^2/\tau }\vartheta _2\left(z|\tau \right),`$
$`\vartheta _2\left(z/\tau |1/\tau \right)=\sqrt{i\tau }e^{\pi iz^2/\tau }\vartheta _4\left(z|\tau \right),`$
$`\vartheta _1\left(z/\tau |1/\tau \right)=i\sqrt{i\tau }e^{\pi iz^2/\tau }\vartheta _1\left(z|\tau \right).`$
##### Identità utili
Aequatio identica satis abstrusa di Jacobi
$$\vartheta _3^4\vartheta _4^4\vartheta _2^4=0,$$
(A.8)
$$\vartheta _3^4\left(z\right)+\vartheta _1^4\left(z\right)\vartheta _4^4\left(z\right)\vartheta _2^4\left(z\right)=0,$$
(A.9)
Altre identità utili sono
$$\vartheta _3\left(z\right)\vartheta _3^3\vartheta _4\left(z\right)\vartheta _4^3\vartheta _2\left(z\right)\vartheta _2^3=2\vartheta _1^4\left(z/2\right),$$
(A.10)
insieme con le identità per $`z=0`$
$$\vartheta _3^{\prime \prime }\vartheta _3\vartheta _2^2\vartheta _2^{\prime \prime }\vartheta _2\vartheta _3^2=4\pi ^2\eta ^6\vartheta _4^2,$$
(A.11)
$$\vartheta _4^{\prime \prime }\vartheta _4\vartheta _3^2\vartheta _3^{\prime \prime }\vartheta _3\vartheta _4^2=4\pi ^2\eta ^6\vartheta _2^2.$$
(A.12)
##### Funzione $`\eta `$ di Dedekind
La funzione $`\eta `$ di Dedekind è definita come
$$\eta \left(\tau \right)=q^{1/24}\underset{n=1}{\overset{\mathrm{}}{}}\left(1q^n\right),$$
(A.13)
e le sue trasformazioni modulari sono
$`\eta \left(\tau +1\right)=e^{i\pi /12}\eta \left(\tau \right),`$
$`\eta \left(1/\tau \right)=\sqrt{i\tau }\eta \left(\tau \right).`$ (A.14)
Infine
$$\vartheta \left(0|\tau \right)=0,$$
(A.15)
mentre la derivata prima di $`\vartheta _1`$ all’origine è
$$\vartheta _1^{}\left(0|\tau \right)=2\pi \eta ^3.$$
(A.16)
## Appendice B Alcuni risultati sulle ampiezze di Superstringa ad albero e ad un loop
Per generalizzare al caso di Superstringa la teoria perturbativa sviluppata per le stringhe bosoniche è utile introdurre il *formalismo dei supercampi*. La supersimmetria locale N=1 sul worldsheet non è manifesta nel formalismo in componenti che si è utilizzato nel primo capitolo. Al contrario, il formalismo $`N=1`$ dei supercampi, formulato nel linguaggio $`N=1`$ della *supergeometria*, pone le trasformazioni di supersimmetria e i diffeomorfismi sullo stesso piano, rendendo entrambe le simmetrie manifeste. La loro azione combinata forma il gruppo dei *super-diffeomorfismi* sulla super-varietà del world-sheet. In questo capitolo si introdurranno brevemente i concetti fondamentali della supergeometria, il formalismo dei supercampi e la costruzione delle ampiezze d’interazione di superstringa a partire dall’integrale funzionale (per una rassegna dettagliata sull’argomento si veda ).
### B.1 Supergeometria
Localmente, una supervarietà $`\mathrm{\Sigma }`$ ($`N=1`$) di dimensione $`\left(2|2\right)`$ è parametrizzata da coordinate $`\xi ^M=(\xi ^m,\theta ^\mu )`$, $`m=1,2`$, $`\mu =1,2`$ con regole di commutazione
$$\xi ^M\xi ^N=()^{MN}\xi ^N\xi ^M$$
(B.1)
dove $`()^{MN}`$ è $`1`$ tranne nel caso $`M=\mu `$, $`N=\nu `$, in cui vale $`1`$. Introduciamo un riferimento (*super-zweibain*) ortonormale locale sulla supervarietà,
$$E^Ad\xi ^ME_M{}_{}{}^{A}A=(a,\alpha )a=z,\overline{z};\alpha =+,,$$
(B.2)
dove le regole di commutazione associate ad indici greci e latini sono le stesse che valgono per le coordinate. Indicando il riferimento inverso come $`E_A^M`$, si ha $`E_A{}_{}{}^{M}E_{M}^{}{}_{}{}^{B}=\delta _A^B`$ e $`E_M{}_{}{}^{A}E_{A}^{}{}_{}{}^{N}=\delta _M^N`$.
Si può introdurre un gruppo di trasformazioni di gauge $`U\left(1\right)`$, sotto il quale i campi del riferimento $`E^z`$, $`E^{\overline{z}}`$, $`E^+`$, $`E^{}`$ hanno pesi $`1`$, $`+1`$, $`\frac{1}{2}`$, $`+\frac{1}{2}`$. Un tensore generico $`V`$ che trasformi come $`\left(E^z\right)^n`$ avrà pertanto peso $`n`$ nel gruppo $`U\left(1\right)`$. Introducendo un campo di gauge $`U\left(1\right)`$, detto connessione,
$$\mathrm{\Omega }=d\xi ^M\mathrm{\Omega }_M$$
(B.3)
si può costruire una super-derivata covariante per il gruppo di simmetria $`U\left(1\right)`$
$$𝒟_A^{\left(n\right)}VE_A{}_{}{}^{M}(_M+in\mathrm{\Omega }_M)V.$$
(B.4)
I supercampi tensoriali di *torsione* $`T_{AB}^C`$ e di *curvatura* $`R_{AB}`$ sono definiti dalle relazioni di commutazione (e di anticommutazione per gli indici spinoriali)
$$[𝒟_A,𝒟_B]V=T_{AB}{}_{}{}^{C}𝒟_{C}^{}V+inR_{AB}V.$$
(B.5)
La supergeometria può essere specificata imponendo alcuni vincoli sui campi di torsione di curvatura,
$$T_{\alpha \beta }^\gamma =0,T_{\alpha \beta }^c=2\gamma _{\alpha \beta }^c,R_{++}=R_{}=0,$$
(B.6)
dove le $`\gamma ^c`$ sono matrici bidimensionali di Dirac. I vincoli (B.6) possono essere scritti in forma equivalente come
$$𝒟_+^2=𝒟_z,𝒟_{}^2=𝒟_{\overline{z}}\{𝒟_+,𝒟_{}\}V=inR_+V,$$
(B.7)
dove $`V`$ è un generico supercampo tensoriale di peso $`n`$. Le rimanenti derivate covarianti possono essere trovate utilizzando le identità di Jacobi. L’unica quantità libera nella supergeometria è il supercampo di curvatura $`R_+=R_+R`$.
La supergeometria è invariante sotto trasformazioni che preservino i vincoli sulla torsione (B.6, B.7). Le simmetrie della supergeometria sono le trasformazioni locali $`sU\left(1\right)`$, generate da un campo locale $`L`$,
$`E_M^\pm `$ $`=`$ $`e^{\pm \left(i/2\right)L}\widehat{E}_M^\pm ,𝒟_+^{\left(n\right)}e^{i\left(n+1/2\right)L}\widehat{𝒟}_+^{\left(n\right)}e^{inL},`$
$`E_M^z`$ $`=`$ $`e^{iL}\widehat{E}_M^z,𝒟_{}^{\left(n\right)}e^{i\left(n1/2\right)L}\widehat{𝒟}_{}^(n)e^{inL},`$
$`E_M^{\overline{z}}`$ $`=`$ $`e^{iL}\widehat{E}_M^{\overline{z}},\mathrm{\Omega }_M=\widehat{\mathrm{\Omega }}_M+_ML;`$ (B.8)
le super-riparametrizzazioni, che formano il gruppo $`sDiff\left(M\right)`$, la cui versione infinitesima è generata da un supercampo vettoriale $`\delta V^M`$,
$$E_A^M\delta E_M^B=𝒟_A\delta V^B\delta V^CT_{CA}^B+\delta V^C\mathrm{\Omega }_CE_A^B;$$
(B.9)
le super-trasformazioni di Weyl, generate da un supercampo reale scalare,
$$E_a^M=e^\mathrm{\Sigma }\widehat{E}_M^a,E_a^M=e^{\mathrm{\Sigma }/2}\left[\widehat{E}_M^a+\widehat{E}_M^a\left(\gamma _a\right)^{\alpha \beta }\widehat{𝒟}_\beta \mathrm{\Sigma }\right],$$
(B.10)
che per la superconnessione, la supercurvatura e le superderivate risultano essere
$`\mathrm{\Omega }_M`$ $`=`$ $`\widehat{\mathrm{\Omega }}_M+\widehat{E}_M^aϵ_a^b\widehat{𝒟}_b\mathrm{\Sigma }+\widehat{E}_M^\alpha \left(\gamma _5\right)_\alpha ^\beta \widehat{𝒟}_\beta \mathrm{\Sigma },`$
$`R_+`$ $`=`$ $`e^\mathrm{\Sigma }\left(\widehat{R}_+2i\widehat{𝒟}_+\widehat{𝒟}_{}\mathrm{\Sigma }\right),`$
$`𝒟_+^{\left(n\right)}`$ $`=`$ $`e^{\left(n1/2\right)\mathrm{\Sigma }}\widehat{𝒟}_+^{\left(n\right)}e^{n\mathrm{\Sigma }},`$
$`𝒟_{}^{\left(n\right)}`$ $`=`$ $`e^{\left(n+1/2\right)\mathrm{\Sigma }}\widehat{𝒟}_{}^{\left(n\right)}e^{+n\mathrm{\Sigma }}.`$ (B.11)
È utile combinare le trasformazioni $`sWeyl`$ e $`sU\left(1\right)`$ in una trasformazione complessa di Weyl, data in termini di un supercampo complesso $`\mathrm{\Lambda }`$, definito come
$$\mathrm{\Lambda }\mathrm{\Sigma }iL,\mathrm{\Lambda }^{}\mathrm{\Sigma }+iL.$$
(B.12)
Le superderivate trasformano sotto $`sWeyl\times sU\left(1\right)`$ come
$$𝒟_+^{\left(n\right)}=e^{n\mathrm{\Lambda }1/2\mathrm{\Lambda }^{}}\widehat{𝒟}_+^{\left(n\right)}e^{n\mathrm{\Lambda }},𝒟_{}^{\left(n\right)}=e^{n\mathrm{\Lambda }^{}1/2\mathrm{\Lambda }}\widehat{𝒟}_{}^{\left(n\right)}e^{n\mathrm{\Lambda }^{}}.$$
(B.13)
e la supercurvatura trasforma come
$$R_+=e^\mathrm{\Sigma }\left(\widehat{R}_+2i\widehat{𝒟}_+\widehat{𝒟}_{}\mathrm{\Sigma }\right).$$
(B.14)
Definiamo le coordinate complesse $`(\xi ,\overline{\xi },\theta ,\overline{\theta })`$ come
$$\xi =1/\sqrt{2}\left(\xi ^1+i\xi ^2\right),\theta =1/\sqrt{2}\left(\theta ^1+i\theta ^2\right),$$
(B.15)
con $`\overline{\xi }`$ e $`\overline{\theta }`$ le rispettive coordinate complesse coniugate. Localmente ogni supergeometria $`N=1`$, in analogia con quanto succede nel caso bosonico, è equivalente a meno di trasformazioni $`sWeyl\times sU\left(1\right)`$ ad una supergeometria piatta (Euclidea) in cui $`R_+=0`$. Nel superspazio piatto si ha per il superzweibein
$`E_m^a`$ $`=`$ $`\delta _m^a,E_m^\alpha =0,`$
$`E_\mu ^a`$ $`=`$ $`\left(\gamma ^a\right)_\mu ^\beta \theta _\beta ,E_\mu ^\alpha =\delta _\mu ^\alpha ,`$ (B.16)
mentre le superderivate prendono una forma estremamente semplice
$`𝒟_+`$ $`=`$ $`{\displaystyle \frac{}{\theta }}+\theta {\displaystyle \frac{}{\xi }},𝒟_z=𝒟_+^2={\displaystyle \frac{}{\xi }},`$
$`𝒟_{}`$ $`=`$ $`{\displaystyle \frac{}{\overline{\theta }}}+\overline{\theta }{\displaystyle \frac{}{\overline{\xi }}},𝒟_{\overline{z}}=𝒟_{}^2={\displaystyle \frac{}{\overline{\xi }}}.`$ (B.17)
La simmetria residua del gruppo $`sDiff\times sU\left(1\right)\times sWeyl`$ che lascia invariante la supergeometria piatta è il gruppo delle trasformazioni *superconformi*, che comprende i diffeomorfismi superanalitici
$$\xi \xi ^{}(\xi ,\theta ),\overline{\xi }\overline{\xi }^{}(\overline{\xi },\overline{\theta }),\theta \theta ^{}(\xi ,\theta ),\overline{\theta }\overline{\theta }^{}(\overline{\xi },\overline{\theta }),$$
(B.18)
combinati con trasformazioni di $`sWeyl`$ e $`sU\left(1\right)`$.
Per ottenere una comprensione più profonda della supergeometria è utile studiare la formulazione *in componenti*, che richiede l’eliminazione di alcune componenti dei supercampi. Per farlo si può fissare il gauge di Wess-Zumino per il superzweibein, definita dalle condizioni
$`E_\mu ^\alpha `$ $``$ $`\delta _\mu ^\alpha +\theta ^\nu e_{\nu \mu }^\alpha ,E_\mu ^a\theta ^\nu e_{\nu \mu }^a,`$
$`e_{\nu \mu }^\alpha `$ $`=`$ $`e_{\mu \nu }^\alpha ,e_{\nu \mu }^a=e_{\mu \nu }^a,`$ (B.19)
sull’espansione in potenze di $`\theta `$, a meno di termini di ordine più alto. Il superzweibein può sempre essere ridotto in questa forma con una super-riparametrizzazione e i campi indipendenti rimanenti sono $`e_m^a`$, $`\chi _m^\alpha `$ e un campo ausiliario $`A`$. Per le altre componenti si trova
$`E_m^a`$ $`=`$ $`e_m{}_{}{}^{a}+\theta \gamma ^a\chi _m{\displaystyle \frac{1}{2}}\theta \overline{\theta }e_m{}_{}{}^{a}A,`$
$`E_m^\alpha `$ $`=`$ $`{\displaystyle \frac{1}{2}}\chi _m{}_{}{}^{\alpha }{\displaystyle \frac{i}{2}}\theta ^\beta \left(\gamma ^a\right)_\beta {}_{}{}^{\alpha }e_{m}^{}{}_{}{}^{a}A+\mathrm{},`$
$`E_\mu ^a`$ $`=`$ $`\left(\gamma ^a\right)_\mu {}_{}{}^{\beta }\theta _{\beta }^{},`$
$`E_\mu ^\alpha `$ $`=`$ $`\delta _\mu {}_{}{}^{\alpha }(1+{\displaystyle \frac{i}{4}}\theta \overline{\theta }A).`$
L’integrazione sulla supervarietà $`\mathrm{\Sigma }`$ richiede la definzione di una super-misura $`d\mu _E`$, invariante sotto super-diffeomorfismi, definita come
$$d\mu _EsdetE_M{}_{}{}^{A}d\xi d\overline{\xi }d\theta d\overline{\theta },$$
(B.21)
dove il super determinante è
$$sdetE_M{}_{}{}^{A}det(E_m{}_{}{}^{a}E_M{}_{}{}^{\alpha }(E_\mu {}_{}{}^{\alpha })_{}^{1}E_\mu {}_{}{}^{a})(detE_\mu {}_{}{}^{\alpha })^1.$$
(B.22)
Il prodotto scalare fra due supercampi $`\mathrm{\Phi }_1`$ e $`\mathrm{\Phi }_2`$ di peso $`n`$ viene definto come
$$\mathrm{\Phi }_1,\mathrm{\Phi }_2=_\mathrm{\Sigma }𝑑\mu _E\mathrm{\Phi }_1^{}\mathrm{\Phi }_2.$$
(B.23)
### B.2 Integrale funzionale di Superstringa
Per definire l’integrale funzionale di superstringa è utile riscrivere l’azione di superstringa e l’azione dei campi di ghost in termini di supercampi. Introduciamo un supercampo di coordinata di stringa, di peso $`0`$ rispetto all’azione della simmetria $`sU\left(1\right)`$, che definisca l’immersione della supervarietà nello spazio-tempo dieci-dimensionale,
$$𝐗:\mathrm{\Sigma }^{10},$$
(B.24)
che in termini di componenti scriviamo come
$$𝐗^\mu =X^\mu +\theta ^\mu \psi _+{}_{}{}^{\mu }+\overline{\theta }\psi _{}^\mu +\theta \overline{\theta }F^\mu \mu =0,1,\mathrm{},9.$$
(B.25)
L’azione della superstringa in termini di supercampi si ottiene accoppiando i supercampi di posizione alla supergravità $`N=1`$ bidimensionale,
$$S_𝐗=\frac{1}{4\pi }_\mathrm{\Sigma }𝑑\mu _E𝒟_{}𝐗𝒟_+𝐗,$$
(B.26)
che scritta in componenti si riduce all’azione di superstringa nota, a meno di un termine non dinamico $`F^2`$ che che coinvolge il campo ausiliario. L’azione scritta possiede le simmetrie locali $`sU\left(1\right)\times sWeyl\times sDiff`$, ed è evidentemente invariante sotto trasformazioni spazio temporali di Poincaré. Si può dimostrare che l’azione scritta è l’unica possibile che possegga tutte le simmetrie volute.
L’azione dei supercampi di ghost $`B`$ e $`C`$, il cui peso sotto le trasformazioni di $`sU\left(1\right)`$ può essere scelto rispettivamente $`n`$ e $`\left(\frac{1}{2}n\right)`$, viene scritta come
$$S_{BC}=\frac{1}{2\pi }_\mathrm{\Sigma }𝑑\mu _E\left(B𝒟_{}^{\left(n\right)}C+\overline{B}𝒟_+^{\left(n\right)}\overline{C}\right).$$
(B.27)
Per ritrovare l’azione nota del sistema $`(b,c)`$, scomponiamo i campi $`B`$, $`C`$ in componenti
$`B`$ $`=`$ $`\beta +\theta b+\overline{\theta }b^{}+\theta \overline{\theta }\beta ^{},`$
$`C`$ $`=`$ $`c+\theta \gamma +\overline{\theta }\gamma ^{}+\theta \overline{\theta }c^{}.`$ (B.28)
I campi $`b^{}`$, $`\beta ^{}`$, $`\gamma ^{}`$ e $`c^{}`$ sono campi ausiliari, i cui pesi per $`sU\left(1\right)`$ possono essere ricavati ricordando che $`\theta `$ e $`\overline{\theta }`$ hanno pesi $`\frac{1}{2}`$ e $`\frac{1}{2}`$. Si trova che i campi hanno pesi $`\beta `$, $`3/2`$; $`b`$, $`2`$; $`\gamma `$, $`1/2`$; $`c`$, $`1`$.
In questo modo si riconosce il sistema $`(b,c)`$ noto, e un sistema $`(\beta ,\gamma )`$ di campi di ghost commutanti (detti superghost) che ha il giusto peso per essere il patner supersimmetrico dei campi di ghost. I campi $`(\beta ,\gamma )`$ sono spinori sul world-sheet e la loro struttura di spin deve essere la stessa dei campi $`\psi _+^\mu `$ in quanto campi di ghost di una supersimmetria $`N=1`$. L’azione $`BC`$ può essere scritta in componenti nel gauge di Wess-Zumino,
$$S_{BC}=\frac{1}{2\pi }_\mathrm{\Sigma }d^2z[b_{\left(1\right)}^zc+\beta _{\left(1/2\right)}^z\gamma \chi _{\overline{z}}{}_{}{}^{+}S_{+}^{\overline{z}}+c.c.],$$
(B.29)
e l’azione totale
$$S=S_𝐗+S_{BC},$$
(B.30)
è invariante superconforme. Procedendo come nel caso bosonico è possibile identificare i supermoduli. Lo spazio delle supergeometrie di genere $`g`$ che soddisfano i vincoli sulla torsione (B.6) che siano inequivalenti sotto i gruppi di simmetria $`sDiff\left(M\right)`$, $`sWeyl\left(M\right)`$ e $`sU\left(1\right)`$ definisce lo spazio dei supermoduli
$$s_g\frac{\left\{E_M{}_{}{}^{A},\mathrm{\Omega }_M\text{che soddisfino la (}\text{B.6}\text{)}\right\}}{sDiff\times sWeyl\times sU\left(1\right)}.$$
(B.31)
La dimensione dello spazio dei supermoduli si trova essere
$$dims_g=\{\begin{array}{cc}\left(0|0\right)& g=0\\ \left(1|0\right)& g=1\text{struttura di spin pari}\\ \left(1|1\right)& g=1\text{struttura di spin dispari}\\ \left(3g3|2g2\right)& g2\end{array}$$
(B.32)
Lo spazio olomorfo cotangente di $`s_g`$ è generato dai differenziali super-olomorfi $`3/2`$, $`\mathrm{\Phi }_K`$, che soddisfano
$$𝒟_{}^{\left(3/2\right)}\mathrm{\Phi }_K=0K=1,\mathrm{},dims_g.$$
(B.33)
I differenziali di super-Beltrami sono definiti come duali di $`\mathrm{\Phi }_K`$ e possono essere parametrizzati dalla variazione dei supermoduli per $`E_M^A`$
$$\mu _KE_{}{}_{}{}^{M}\frac{E_M^z}{m_K}K=1,\mathrm{},dims_g,$$
(B.34)
che nel gauge di Wess-Zumino diventano
$$\mu _K=\overline{\theta }\left(e_{\overline{z}}{}_{}{}^{n}\frac{e_n^z}{m_K}\theta \frac{\chi _{\overline{z}}^+}{m_K}\right).$$
(B.35)
Il primo termine è il differenziale di Beltrami consueto, per un modulo pari $`m_K`$, mentre il secondo termine è un differenziale di Beltrami dispari per un modulo dispari $`m_K`$.
Tralasciando per un momento i problemi di separazione chirale degli spinori, possiamo introdurre la definizione generale delle ampiezze d’interazione. Definendo in analogia con il caso bosonico l’integrale funzionale a partire dall’idea della somma sulle superfici di universo, la forma generale delle ampiezze si scrive come
$$A_g=\left[dE\right]\left[d\mathrm{\Omega }\right]\delta \text{(vincoli)}\left[d𝐗\right]V_1\mathrm{}V_Ne^{S_𝐗},$$
(B.36)
dove i $`V_i`$ sono operatori di vertice di superstringa da definire. Nella dimensione critica $`D=10`$ e per operatori di vertice super-conformemente invarianti, si può dimostrare che
$`A_g={\displaystyle _{s_g}}{\displaystyle \underset{K}{}}dm_K{\displaystyle \left[d𝐗\right]\left[dBd\overline{B}\right]\left[dCd\overline{C}\right]V_1\mathrm{}V_Ne^{S_𝐗S_{BC}}}`$
$`{\displaystyle \underset{K}{}}\delta \left(\mu _\kappa ,B\right)\delta \left(\overline{\mu }_\kappa ,\overline{B}\right).`$ (B.37)
#### B.2.1 Quantizzazione BRST
Come nel caso bosonico, scegliendo un gauge e introducendo i campi di ghost di Faddeev-Popov, la simmetria originaria sopravvive in forma di simmetria BRST. L’azione totale $`S=S_𝐗+S_{BC}`$ è invariante sotto le trasformazioni dei supercampi
$`\delta 𝐗^\mu `$ $`=`$ $`\lambda C𝒟_+^2X^\mu {\displaystyle \frac{1}{2}}\lambda 𝒟_+C𝒟_+𝐗^\mu +c.c.,`$
$`\delta C`$ $`=`$ $`\lambda C𝒟_+^2C{\displaystyle \frac{1}{4}}\lambda 𝒟_+C𝒟_+C,`$
$`\delta B`$ $`=`$ $`\lambda T,`$ (B.38)
dove $`\lambda `$ è un parametro grassmaniano costante e $`T`$ è il super tensore di energia-impulso. Per il supercampo $`𝐗`$ si trova
$`T^{\left(𝐗\right)}`$ $`=`$ $`{\displaystyle \frac{1}{2}}𝒟_+𝐗𝒟_+^2𝐗`$ (B.39)
$`=`$ $`{\displaystyle \frac{1}{2}}\psi _+_zX+\theta \left({\displaystyle \frac{1}{2}}_zX_z𝐗{\displaystyle \frac{1}{2}}_z\psi _+\psi _+\right),`$
mentre per il sistema $`(B,C)`$ il supertensore $`T`$ è
$`T^{\left(BC\right)}`$ $`=`$ $`C𝒟_+^2B+{\displaystyle \frac{1}{2}}𝒟_+C𝒟_+B{\displaystyle \frac{3}{2}}𝒟_+^2CB`$ (B.40)
$`=`$ $`c_z\beta +{\displaystyle \frac{1}{2}}\gamma b{\displaystyle \frac{3}{2}}_zc\beta +\theta \left(T^{\left(bc\right)}+T^{\left(\beta \gamma \right)}\right).`$
La corrente BRST associata alla simmetria è
$$J_{BRST}=C\left(T^{\left(𝐗\right)}+\frac{1}{2}T^{\left(BC\right)}\right)\frac{3}{4}𝒟_+\left(C𝒟_+CB\right).$$
(B.41)
Come già visto nel caso bosonico, la carica associata alla corrente $`BRST`$
$$Q_{BRST}=𝑑z𝑑\theta J_{BRST},$$
(B.42)
è conservata e nilpotente.
### B.3 Teorema di separazione chirale
In Teoria delle Stringhe esiste un risultato molto potente sulle proprietà di olomorfia delle funzioni di correlazione per operatori di vertice fattorizzati di stati fisici (stati *on-shell*), che va sotto il nome di *Teorema di Separazione Chirale*. Un generico operatore di vertice della forma
$$V_i=d^2z_id^2\theta _iW_i(z_i,\theta _i,\overline{z}_i,\overline{\theta }_i),$$
(B.43)
è fattorizzato se $`W_i`$ è prodotto di un fattore olomorfo e di un fattore anti-olomorfo
$$W_i(z_i,\theta _i,\overline{z}_i,\overline{\theta }_i)=W_i(z_i,\theta _i)\stackrel{~}{W}_i(\overline{z}_i,\overline{\theta }_i).$$
(B.44)
Un generico vertice di interazione può essere scritto come una combinazione lineare di operatori di vertice fattorizzati. Definendo i $`g`$ momenti interni $`p_I^\mu `$, associati agli 1-cicli di una base canonica $`A_I`$, $`B_I`$ (B.1), $`I=1,\mathrm{},g`$ di una superficie di genere $`g`$, come
$$p_I^\mu =_{A_I}𝑑z𝑑\theta 𝒟_+𝐗^\mu ,$$
(B.45)
le funzioni di correlazione di operatori di vertice non integrati sulla supergeometria, per momenti interni $`p_I^\mu `$ e supermoduli fissati, sono
$`W_1\mathrm{}W_N_E\left(p_I\right)`$ $`{\displaystyle \left[dX\right]\left[dBdC\right]W_1\mathrm{}W_N}`$
$`{\displaystyle \underset{I=1}{\overset{g}{}}}\delta \left(p_I^\mu {\displaystyle _{A_I}}𝑑z𝑑\theta 𝒟_+𝐗^\mu \right){\displaystyle \underset{K=1}{\overset{dims_g}{}}}\left|\mu _K,B\right|^2e^{S_𝐗S_{BC}}`$
(B.46)
Il Teorema di separazione chirale comprende due risultati. Il primo è la fattorizzazione delle ampiezze di superstringa a fissati momenti interni. Le ampiezze risultano scrivibili nella forma
$$W_1\mathrm{}W_N_E\left(p_I\right)=\delta \left(k\right)𝒞_\nu ^F\overline{𝒞}_\nu ^F,$$
(B.47)
con $`k`$ gli impulsi esterni e dove $`𝒞_\nu {}_{}{}^{F}=𝒞_\nu {}_{}{}^{F}(z_i,\theta _i;\zeta _i,;m_i;p_I,k_i)`$ è una funzione analitica complessa dei supermoduli $`m_K`$, $`K=1,\mathrm{},dims_g`$, dei punti di inserzione $`(z_i,\theta _i)`$, e dei fattori sinistri del tensore di polarizzazione $`\zeta _{i\mu }`$. Per semplicità si sono considerate strutture di spin uguali per gli spino di chiralità destra e sinistra. $`\overline{𝒞}_\nu ^F`$ è quindi il complesso coniugato di $`𝒞_\nu ^F`$, con la stessa struttura di spin $`\nu `$,
$$\overline{𝒞}_\nu ^F(\overline{z}_i,\overline{\theta }_i;\overline{\zeta }_i;\overline{m}_K;p_I,k_i)=𝒞_\nu {}_{}{}^{F}(z_i,\theta _i;\zeta _i;m_Kp_I,k_i)_{}^{}.$$
(B.48)
L’espressione delle funzioni $`𝒞_\nu ^F`$ può essere trovata in forma esplicita utilizzando le funzioni di Green e il kernel di Szegö.
Il secondo risultato del Teorema di separazione chirale fissa la forma delle ampiezze di superstringa. Si trova per le teorie di tipo II
$$A_h=\delta \left(k\right)\underset{\overline{\nu }}{}Q_{\nu \overline{\nu }}_{^{10g}}d^{10}p_I_{s_h}𝑑m_K𝑑\overline{m}_K\underset{i=1}{\overset{N}{}}_\mathrm{\Sigma }d^2z_id^2\theta _i𝒞_\nu ^F\overline{𝒞}_{\overline{\nu }}^F,$$
(B.49)
dove $`Q_{\nu \overline{\nu }}=\pm 1`$ sono fattori che dipendono dalla struttura di spin e che realizzano la proiezione GSO.
### B.4 Operatori di vertice
Per stringhe chiuse si hanno quattro classi di operatori di vertice: $`NSNS`$, $`RNS`$, $`NSR`$, $`RR`$. Iniziamo dagli operatori di vertice $`NSNS`$, senza dipendenza dai campi ghost. Le parti destra e sinistra possono essere ricavate utilizzando la separazione chirale, e in analogia con i vertici di stringa bosonica si ha per un vertice generico, in una supergeometria $`N=1`$ piatta,
$$V(ϵ,k)=_\mathrm{\Sigma }𝑑\mu _EP_n(ϵ,𝒟_+X,𝒟_{}𝐗,𝒟_+^2𝐗,\mathrm{})e^{ik𝐗},$$
(B.50)
dove $`P_n`$ è una somma di termini con un numero totale di $`n`$ derivate dei campi $`𝐗`$. Nel caso di supergeometrie non piatte, l’operatore di vertice dipenderà anche dalla supercurvatura $`R_+`$.
Per essere invarianti sotto l’azione del gruppo $`sU\left(1\right)`$, i vertici devono avere un ugual numero $`n`$ di derivate $`𝒟_+`$ e $`𝒟_{}`$. I vertici di questa forma sono invarianti sotto l’azione di $`sDiff`$ per costruzione, mentre l’invarianza per $`sWeyl`$ può essere verificata senza difficoltà.
L’operatore di vertice degli stati a massa nulla è
$$V(ϵ,k)=ϵ_{\mu \nu }\left(k\right)_\mathrm{\Sigma }𝑑\mu _E𝒟_+𝐗^\mu 𝒟_{}𝐗^\nu e^{ik𝐗}.$$
(B.51)
Si può vedere che l’invariaza per $`sWeyl`$ richiede $`ϵ_{\mu \nu }\left(k\right)k^\mu =ϵ_{\mu \nu }\left(k\right)k^\nu =0`$, e si ritrovano in questo modo in particolare, i vertici del dilatone, del gravitone e del campo $`B_{\mu \nu }`$.
La parte sinistra dell’operatore di vertice (e allo stesso modo la parte destra) può essere scritta in componenti come
$$𝑑\theta 𝒟_+𝐗_L^\mu e^{ik𝐗_L}=\left(_zX_L^\mu i\psi _+^\mu \psi _+^\nu k_\nu \right)e^{ikX_L}.$$
(B.52)
La definizione dei vertici di Ramond introduce diverse complicazioni . Non è infatti possibile scrivere un operatore di vertice per i fermioni a massa nulla a partire dal solo campo spinoriale, ma si deve opportunamente tener conto del contributo dei campi ghost. Consideriamo la parte sinistra di Ramond di un operatore di vertice, che scriviamo come
$$W_{}(z,u,k)=u^\alpha \left(k\right)𝒪\left(z\right)S_\alpha \left(z\right)e^{ikX_L\left(z\right)},$$
(B.53)
dove $`u^\alpha `$ è uno spinore che costiutisce l’analogo del tensore di polarizzazione $`ϵ`$ dei vertici NS. Perchè il vertice sia Weyl invariante, l’operatore $`𝒪\left(z\right)`$, che dipende dai campi ghost $`\beta `$ e $`\gamma `$, deve essere un campo primario di peso $`3/8`$, in modo che $`W_{}`$ abbia peso $`1`$ per i campi a massa nulla.
Il sistema di campi di ghost $`(\beta ,\gamma )`$ può essere rappresentato in termini di due campi bosonici liberi $`\varphi `$ e $`\chi `$, tali che
$$\gamma =e^{\varphi \chi },\beta =e^{\varphi +\chi }_z\chi ,$$
(B.54)
con
$`\varphi \left(z\right)\varphi \left(w\right)`$ $``$ $`\mathrm{ln}\left(zw\right),`$
$`\chi \left(z\right)\chi \left(w\right)`$ $``$ $`\mathrm{ln}\left(zw\right),`$ (B.55)
L’operatore $`𝒪\left(z\right)`$ può essre scritto come esponenziale
$$𝒪\left(z\right)=e^{\frac{1}{2}\varphi \left(z\right)}.$$
(B.56)
L’operatore di vertice destro di Ramond si trova essere
$$W_+(z,u,k)=e^{\varphi /2}u^\alpha \left(\gamma _\mu \right)_{\alpha \beta }S^{}{}_{}{}^{\beta }\left(z\right)(_zx^\mu i\psi _+^\mu k\psi _+)e^{ikx}$$
(B.57)
### B.5 Ampiezze ad albero
Limitandoci alle ampiezze di stringa chiusa, al livello ad albero il world-sheet ha la topologia della sfera $`g`$, e in questo caso non si hanno supermoduli. Consideriamo lo scattering di stati a massa nulla $`NSNS`$. L’operatore di vertice (B.51) può essere riscritto, fattorizzando $`ϵ_{\mu \nu }\left(k\right)=\zeta _\mu \left(k\right)\overline{\zeta }_\nu \left(k\right)`$ con $`\zeta `$ e $`\overline{\zeta }`$ parametri Grassmaniani, nella forma
$$V(\zeta ,\overline{\zeta };k)=_\mathrm{\Sigma }d^2zd^2\theta e^{ik𝐗+\xi 𝒟_+𝐗+\overline{\xi }𝒟_{}𝐗},$$
(B.58)
limitandoci a considerare i contributi lineari in $`\zeta `$ e $`\overline{\zeta }`$. Il propagatore di $`𝐗`$ si trova essere
$$X^\mu (z,\theta )X^\nu (z^{},\theta ^{})=\mathrm{ln}\left|zz^{}\theta \theta ^{}\right|^2\eta ^{\mu \nu }.$$
(B.59)
I determinanti associati ai campi $`x`$, $`\psi `$, $`b`$, $`c`$ e $`\beta `$, $`\gamma `$, costituiscono una costante moltiplicativa.
In questo caso, il teorema di separazione chirale è un’ovvia conseguenza della mancanza di momenti interni per la sfera. La funzione di correlazione dei vertici non integrati porta ad un prodotto di una funzione $`𝒞^F`$ complessa analitica in $`z_i`$, $`\theta _i`$ per la sua complessa coniugata. Consideriamo gli operatori chirali di vertice che sono
$$W_L(z_i,\theta _i)=e^{ik_iX_L(z_i,\theta _i)+\zeta _i𝒟_+X_L(z_i,\theta _i)},$$
(B.60)
e il corrispondente operatore destro. La contrazione dei campi chirali risulta essere
$$X_L^\mu (z,\theta )X_L^\nu (z^{},\theta ^{})=\mathrm{ln}\left(zz^{}\theta \theta ^{}\right)\eta ^{\mu \nu },$$
(B.61)
da cui si trova per il valore di aspettazione dei vertici chirali
$$W_L(z_1,\theta _1)\mathrm{}W_L(z_N,\theta _N)=𝒞^F(z_i,\theta _i),$$
(B.62)
con
$$𝒞^F=\mathrm{exp}\underset{ij}{\overset{N}{}}\left\{ik_i\zeta _j\frac{\theta _{ij}}{z_{ij}}+\frac{1}{2}\zeta _i\zeta _j\frac{1}{z_{ij}}+\frac{1}{2}k_ik_j\mathrm{ln}z_{ij}\right\}.$$
(B.63)
e l’espressione analoga per i vertici con chiralità opposta. La notazione usata è quella standard $`\theta _{ij}=\theta _i\theta _j`$, $`z_{ij}=z_iz_j\theta _i\theta _j`$.
#### B.5.1 Trasformazioni superconformi
Per la superstringa a livello ad albero si ha invarianza sotto il gruppo di automorfismi superconforme $`OSp(1,2)`$. La misura di integrazione deve tenere conto di questa simmetria ulteriore. Per definire il gruppo di trasformazioni, consideriamo una terna di coordinate $`(v,w,\psi )`$ dove le varibili latine (greche) descrivono variabili commutanti (anticommutanti). Su questa tripletta, si ha l’azione naturale di $`GL\left(2|1\right)`$, $`T:WTW`$,
$$W=\left(\begin{array}{c}v\\ w\\ \psi \end{array}\right),T=\left(\begin{array}{ccc}a& b& \alpha \\ c& d& \beta \\ \gamma & \delta & A\end{array}\right),$$
(B.64)
Per stabilire un contatto con il superspazio $`N=1`$, introduciamo le coordinate proiettive
$$z=\frac{v}{w},\theta =\frac{\psi }{w},$$
(B.65)
sulle quali l’azione di $`GL\left(2|1\right)`$ è
$$z\frac{az+b+\alpha \theta }{cz+d+\beta \theta }\theta \frac{\gamma z+\delta +A\theta }{cz+d+\beta \theta }$$
(B.66)
Per ottenere una trasformazione superconforme, l’elemento lineare $`dz=dz+\theta d\theta `$ deve trasformare in se stesso a meno di un riscalamento conforme, in maniera equivalente, la forma quadratica
$$z_{12}=z_1z_2\theta _1\theta _2=\frac{v_1w_2v_2w_1\psi _1\psi _2}{w_1w_2}$$
(B.67)
deve trasformare in se stessa a meno di un riscalamento conforme. Questo si ottiene imponendo su $`T`$ il vincolo,
$$T^TKT=K$$
(B.68)
dove K è la matrice
$$\left(\begin{array}{ccc}0& +1& 0\\ 1& 0& 0\\ 0& 0& 1\end{array}\right),$$
(B.69)
La legge di trasformazione della forma quadratica si deriva facilmente, e si trova essere
$$T:z_{12}\frac{z_{12}}{\left(cz_1+d+\beta \theta _1\right)\left(cz_2+d+\beta \theta _2\right)}.$$
(B.70)
Allo stesso modo per l’elemento lineare, si trova
$$dz\frac{dz}{\left(cz+d+\beta \theta \right)^2},$$
(B.71)
ed infine la legge di trasformazione dell’elemento di volume è
$$dzd\theta \frac{dzd\theta }{\left(cz+d+\beta \theta \right)}.$$
(B.72)
Gli elementi di $`OSp(1,2)`$ sono in corrispondenza biunivoca con una tripletta di punti del superpiano $`(z_1,\theta _1)`$, $`(z_2,\theta _2)`$, $`(z_3,\theta _3)`$, vincolata da un equazione a valori grassmaniani. Il conteggio dei parametri funziona perchè $`OSp(1,2)`$ ha 3 parametri commutanti e 2 anticommutanti. Il vincolo è una funzione invariante del gruppo $`OSp(1,2)`$ a valori grassmaniani, che dipende dai 3 punti,
$$\mathrm{\Delta }=\frac{z_{12}\theta _3+z_{31}\theta _2+z_{23}\theta _1+\theta _1\theta _2\theta _3}{\left(z_{12}z_{23}z_{31}\right)^{1/2}}.$$
(B.73)
Ponendo $`\mathrm{\Delta }`$ a zero, si fissa uno dei parametri $`\theta `$. In questo modo si può vedere che c’è una corrsipondenza tra triplette di punti che soddisfino il vincolo e elementi di $`OSp(1,2)`$. L’elemento di volume invariante indotto su $`OSp(1,2)`$ si ottiene moltiplicando l’elemento di volume ordinario per una funzione delta che tenga conto del vincolo,
$$d\mu =\frac{dz_1dz_2dz_3d\theta _1d\theta _2d\theta _3}{\left(z_{12}z_{23}z_{31}\right)^{1/2}}\delta \left(\mathrm{\Delta }\right).$$
(B.74)
#### B.5.2 Ampiezze
Le funzioni a zero-, uno- e due-punti di superstringa sono tutte nulle. Il modo più rapido per capirlo è osservare che il sottogruppo del gruppo superconforme, che lascia 0, 1, o 2 punti fissati ha un volume infinito.
La funzione a tre punti, calcolata a partire dal valore di aspettazione dei vertici (B.63), si trova essere
$$V(ϵ_1,k_1)V\left(ϵ_2k_2\right)V\left(ϵ_3k_3\right)=4\left(2\pi \right)^{10}\delta \left(k\right)ϵ_1^{\mu _1\overline{\mu }_1}ϵ_2^{\mu _2\overline{\mu }_2}ϵ_3^{\mu _3\overline{\mu }_3}K_{\mu _1\mu _2\mu _3}K_{\overline{\mu }_1\overline{\mu }_2\overline{\mu }_3},$$
(B.75)
con
$$K_{\mu _1\mu _2\mu _3}=\eta _{\mu _1\mu _2}k_{1\mu _3}+\eta _{\mu _2\mu _3}k_{2\mu _1}+\eta _{\mu _3\mu _1}k_{3\mu _2}.$$
(B.76)
Si può vedere che a causa della trasversalità del tensore di polarizzazione, la funzione a tre-punti per particelle a massa nulla è zero.
Per calcolare la funzione a quattro punti, occorre fissare l’invarianza superconforme fissando i verti di interazione. Scegliendo $`z=z_1`$, $`z_2=0`$, $`z_3=1`$, $`z_4=\mathrm{}`$, $`\theta _1,\theta _2,\theta _3=\theta _4=0`$, si trova dalla (B.63),
$`\mathrm{exp}𝒢_4^\zeta `$ $`=`$ $`\left\{\zeta _1\zeta _2\zeta _3\zeta _4{\displaystyle \frac{1}{z_{12}z_{34}}}+\zeta _1\zeta _3\zeta _2\zeta _4{\displaystyle \frac{1}{z_{13}z_{24}}}+\zeta _1\zeta _4\zeta _2\zeta _3{\displaystyle \frac{1}{z_{14}z_{23}}}\right\}`$ (B.77)
$`+`$ $`\left\{\zeta _1\zeta _2\left(k_1\zeta _3k_2\zeta _4{\displaystyle \frac{\theta _2\theta _1}{z_{12}z_{13}z_{24}}}+k_1\zeta _4k_2\zeta _3{\displaystyle \frac{\theta _2\theta _1}{z_{12}z_{14}z_{23}}}\right)+\text{perm.}\right\}.`$
Per trovare l’ampiezza si dovrebbe moltiplicare questa espressione per l’equivalente espressione in $`\overline{\zeta }`$, integrare in $`z`$ e $`\theta `$ e raggruppare i termini. Il calcolo è però enormemente semplificato dalle proprietà di fattorizzazione degli integrali di Veneziano.
Per un integrale ordinario si ha
$$\frac{d^2z}{\pi }z^A\overline{z}^{\stackrel{~}{A}}\left(1z\right)^B\left(1\overline{z}\right)^{\stackrel{~}{B}}=\frac{\mathrm{\Gamma }\left(1\stackrel{~}{A}\stackrel{~}{B}\right)}{\mathrm{\Gamma }\left(\stackrel{~}{A}\right)\mathrm{\Gamma }\left(\stackrel{~}{B}\right)}\frac{\mathrm{\Gamma }\left(1+A\right)\mathrm{\Gamma }\left(1+B\right)}{\mathrm{\Gamma }\left(A+B+2\right)},$$
(B.78)
per $`A\stackrel{~}{A}`$ and $`B\stackrel{~}{B}`$ interi. Non è difficile verificare che questa formula è simmetrica sotto lo scambio $`(A,B)(\stackrel{~}{A},\stackrel{~}{B})`$. Una formula analoga può essere derivata per i super-integrali
$`{\displaystyle \frac{d^2z_1}{\pi }d^2\theta _2\left[\theta _1\theta _2\right]^a\left[\overline{\theta }_1\overline{\theta }_2\right]^{\stackrel{~}{a}}z_{12}^A\overline{z}_{12}^{\stackrel{~}{A}}\left(1z_1\right)^B\left(1\overline{z}_1\right)^{\stackrel{~}{B}}}`$ (B.79)
$`=\left(2i\right)^{1a}\left(+2i\right)^{1\stackrel{~}{a}}{\displaystyle \frac{\mathrm{\Gamma }\left(\stackrel{~}{a}\stackrel{~}{A}\stackrel{~}{B}\right)}{\mathrm{\Gamma }\left(\stackrel{~}{A}\right)\mathrm{\Gamma }\left(\stackrel{~}{B}\right)}}{\displaystyle \frac{\mathrm{\Gamma }\left(1+A\right)\mathrm{\Gamma }\left(1+B\right)}{\mathrm{\Gamma }\left(A+B+1+a\right)}}`$
con $`a`$ e $`\stackrel{~}{a}`$ a valori $`0`$ o $`1`$. L’integrale è simmetrico per $`\left(aAB\right)\left(\stackrel{~}{a}\stackrel{~}{A}\stackrel{~}{B}\right)`$. Utilizzando questa proprietà, si può ottenere l’espressione completa delle ampiezze limitandosi a considerare il contributo dipendende da $`z`$.
Per l’ampiezza a quattro punti si trova in questo modo
$`<V(ϵ_1,k_1)V(ϵ_2,k_2)V(ϵ_3,k_3)V(ϵ_4,k_4)>`$ (B.80)
$`=`$ $`\left(2\pi \right)^{10}\delta \left(k\right)g^4{\displaystyle d^2z_1d^2\theta _2\left|z_{12}\right|^s\left|z_11\right|^ue^{𝒢_4^\zeta +𝒢_4^{\overline{\zeta }}}}`$ (B.81)
$`=`$ $`\pi \left(2\pi \right)^{10}\delta \left(k\right)g^4{\displaystyle \frac{\mathrm{\Gamma }\left(s/2\right)\mathrm{\Gamma }\left(t/2\right)\mathrm{\Gamma }\left(u/2\right)}{\mathrm{\Gamma }\left(1+\frac{s}{2}\right)\mathrm{\Gamma }\left(1+\frac{t}{2}\right)\mathrm{\Gamma }\left(+\frac{u}{2}\right)}}ϵ^{1\overline{1}}ϵ^{2\overline{2}}ϵ^{3\overline{3}}ϵ^{4\overline{4}}K_{1234}K_{\overline{1}\overline{2}\overline{3}\overline{4}},`$
dove, usando l’abbreviazione $`i`$ per $`\mu _i`$ ($`K_{\mu _1\mu _2\mu _3\mu _4}=K_{1234}`$), si è definito il tensore
$`K_{1234}`$ $`=`$ $`\left(st\eta _{13}\eta _{24}su\eta _{14}\eta _{23}tu\eta _{12}\eta _{34}\right)`$ (B.82)
$``$ $`s\left(k_1^4k_3^2\eta _{24}+k_2^3k_4^1\eta _{13}k_1^3k_4^2\eta _{23}k_2^4k_3^1\eta _{14}\right)`$
$`+`$ $`t\left(k_2^1k_4^3\eta _{13}+k_3^4k_1^2\eta _{24}k_2^4k_1^3\eta _{34}k_3^1k_4^2\eta _{12}\right)`$
$``$ $`u\left(k_1^2k_4^3\eta _{23}+k_3^4k_2^1\eta _{14}k_1^4k_2^3\eta _{34}k_3^2k_4^1\eta _{12}\right).`$
### B.6 Ampiezze ad un loop
Per calcolare l’ampiezza di superstringa ad un loop, occorre utilizzare il teorema di separazione chirale per determinare le ampiezze chirali $`𝒞_\nu ^F`$. Cosideriamo per semplicità strutture di spin sinistra e destra uguali.
Per strutture di spin $`\nu `$ pari non si hanno moduli e non ci sono zero modi di Dirac. Consideriamo la solita rappresentazione del toro come parallelogrammo di lati $`0`$, $`1`$, $`\tau `$, $`1+\tau `$, con $`\tau `$ appartenente al semipiano complesso superiore. La metrica può quindi essere scelta come $`g=2\left|dz\right|^2`$.
L’azione scritta in componenti è
$`S_X+S_{BC}`$ $`=`$ $`{\displaystyle \frac{1}{4\pi }}{\displaystyle _\mathrm{\Sigma }}d^2z(_zX_{\overline{z}}X\psi _+_{\overline{z}}\psi _+\psi _{}_z\psi _{}`$ (B.83)
$`+2b_{\overline{z}}c+2\overline{b}_z\overline{c}+2\beta _{\overline{z}}\gamma +2\overline{\beta }_z\overline{\gamma }),`$
e le inserzioni dei campi di ghost e di anti-ghost si riducono a $`b\overline{b}c\overline{c}`$, dal momento che per una struttura di spin pari non ci sono zero modi.
Nel settore $`NSNS`$ i vertici non contengono campi di ghost. In questo caso quindi l’ampiezza si semplifica e si possono integrare i campi $`b`$, $`c`$, $`\overline{b}`$, $`\overline{c}`$, $`\beta `$, $`\gamma `$, $`\overline{\beta }`$ e $`\overline{\gamma }`$. Tenendo conto anche dei determinanti provenienti dall’integrazione sui campi $`X`$ e $`\psi _\pm `$ e utilizzando il teorema di separazione chirale si può scrivere
$$\delta \left(k\right)𝒞_\nu 𝒞_\nu =M_\nu \overline{M}_\nu _\nu _\nu ,$$
(B.84)
dove $`M_\nu \overline{M}_\nu `$ sono i termini dovuti ai determinanti e $`_\nu \overline{}_\nu `$ sono i termini delle funzioni di correlazione degli operatori di vertice. In forma esplicita, si trova,
$$M_\nu \overline{M}_\nu =\left(\frac{\stackrel{}{det}\mathrm{\Delta }_0}{\left[Im\left(\tau \right)\right]^2}\right)^5\left(det/D_+\right)_\nu ^5\left(det/D_{}\right)_\nu ^5\left(\frac{\stackrel{}{det}\mathrm{\Delta }_1}{\left[Im\left(\tau \right)\right]^2}\right)^{+1}\left(det\mathrm{\Delta }_{1/2}^{}\right)_\nu ^1.$$
(B.85)
I determinanti posso essere calcolati esplicitamente, e
$`{\displaystyle \frac{\stackrel{}{det}\mathrm{\Delta }_0}{\left[Im\left(\tau \right)\right]^2}}`$ $`=`$ $`{\displaystyle \frac{\stackrel{}{det}\mathrm{\Delta }_1}{\left[Im\left(\tau \right)\right]^2}}=\left|\eta \left(\tau \right)\right|^4,`$
$`\left(det/D_+/D_{}\right)_\nu `$ $`=`$ $`\left(det\mathrm{\Delta }_{1/2}^{}\right)_\nu =\left|{\displaystyle \frac{\vartheta \left[\nu \right]\left(0|\tau \right)}{\eta \left(\tau \right)}}\right|^2.`$ (B.86)
Si ottiene in questo modo
$$M_\nu =\frac{\vartheta \left[\nu \right]\left(0|\tau \right)^4}{\eta \left(\tau \right)^{12}},\overline{M}_\nu =\frac{\overline{\vartheta \left[\nu \right]\left(0|\tau \right)^4}}{\overline{\eta \left(\tau \right)}^{12}},$$
(B.87)
a meno di un fattore di fase indipendente da $`\tau `$, ma che può dipendere da $`\nu `$. I contributi dei vertici possono essere similmente calcolati, ottenendo
$`_\nu (z_i,\theta _i,\zeta ,k,p)`$ $`=`$ $`\mathrm{exp}\{i\pi p^2\tau +2\pi p{\displaystyle \underset{i}{}}(\zeta _i\theta _i+ik_iz_i)`$ (B.88)
$`{\displaystyle \frac{1}{2}}{\displaystyle \underset{ij}{}}(k_ik_jG_\nu (z_i,\theta _i;z_j,\theta _j)+\zeta _i\zeta _j𝒟_+^i𝒟_+^jG_\nu `$
$`+2ik_i\zeta _j𝒟_+^jG_\nu )\},`$
dove la funzione di correlazione $`G_\nu `$ è data da
$`G_\nu (z_i,\theta _i;z_j,\theta _j)`$ $`=`$ $`𝐗_L(z_i,\theta _i)𝐗_L(z_j,\theta _j)`$ (B.89)
$`=`$ $`\mathrm{ln}E(z_i,z_j)+\theta _i\theta _jS_\nu (z_i,z_j).`$
$`E(z,w)`$ è la forma prima, e $`S_\nu `$ è il kernel di Szegö. Per il toro queste funzioni sono
$`E(z,w)`$ $`=`$ $`{\displaystyle \frac{\vartheta _1\left(zw|\tau \right)}{\vartheta _1^{}\left(0|\tau \right)}},`$
$`S_\nu \left(zw\right)`$ $`=`$ $`{\displaystyle \frac{\vartheta \left[\nu \right]\left(zw|\tau \right)\vartheta _1^{}\left(0|\tau \right)}{\vartheta \left[\nu \right]\left(0|\tau \right)\vartheta _1\left(zw|\tau \right)}}.`$ (B.90)
Per strutture di spin dispari c’è un modulo complesso dispari $`\chi `$, e $`10`$ zero modi di Dirac, uno per ogni direzione spazio-temporale. In questo caso, oltre all’azione $`S_X+S_{BC}`$, c’è un’inserzione della supercorrente sul world-sheet, $`\chi S_{z+}`$,
$$S_{z+}=\left(\psi _+_zX\frac{1}{2}b\gamma +\frac{3}{2}\beta _zc+_z\beta c\right)+c.c.$$
(B.91)
L’integrazione sul modulo $`\chi `$ porta ad avere un’inserzione di termini $`\psi _+_zx`$ e $`\psi _{}_{\overline{z}}x`$. Considerando i campi ghost e anti-ghost, l’inserzione totale è
$$b\overline{b}c\overline{c}\delta \left(\beta \right)\delta \left(\overline{\beta }\right)\delta \left(\gamma \right)\delta \left(\overline{\gamma }\right)S_{z+}S_{\overline{z}}.$$
(B.92)
Consideriamo ora le funzioni a uno-, due-, tre-, quattro-punti per stati a massa nulla $`NSNS`$. Le strutture di spin dispari contribuiscono nei diagrammi con almeno $`5`$ operatori di vertice. Quindi per queste ampiezze basta considerare stutture di spin pari, e non c’è pertanto differenza fra le stringhe di tipo $`IIA`$ e $`IIB`$.
Iniziamo dal determinare i fattori $`Q_{\nu \overline{\nu }}`$. Ricordando le proprietà delle funzioni $`\eta `$ e $`\theta `$, si ha che
$`M_{ab}\left(\tau +1\right)`$ $`=`$ $`e^{i\pi a}M_{a\left(b+a+1\right)}\left(\tau \right)`$
$`M_{ab}\left({\displaystyle \frac{1}{\tau }}\right)`$ $`=`$ $`{\displaystyle \frac{1}{\tau ^4}}M_{ba}\left(\tau \right).`$ (B.93)
Le funzioni theta sono definite come
$`\vartheta _{00}(z,\tau )`$ $`=`$ $`\vartheta _3(z,\tau )=\vartheta (z,\tau ),`$
$`\vartheta _{01}(z,\tau )`$ $`=`$ $`\vartheta _4(z,\tau )=\vartheta (z+{\displaystyle \frac{1}{2}},\tau ),`$
$`\vartheta _{10}(z,\tau )`$ $`=`$ $`\vartheta _2(z,\tau )=e^{i\pi \tau /4+\pi iz}\vartheta (z+{\displaystyle \frac{\tau }{2}},\tau ),`$
$`\vartheta _{11}(z,\tau )`$ $`=`$ $`\vartheta _1(z,\tau )=e^{i\pi \tau /4+\pi iz}\vartheta (z+{\displaystyle \frac{1}{2}}+{\displaystyle \frac{\tau }{2}},\tau ).`$ (B.94)
e la richiesta di invarianza modulare dell’ampiezza, impone
$`{\displaystyle \underset{(a,b)}{}}Q_{\left(ab\right)\overline{\nu }}M_{\left(ab\right)}\left(\tau +1\right)`$ $`=`$ $`{\displaystyle \underset{(a,b)}{}}Q_{(a,b)\overline{\nu }}M_{\left(ab\right)}\left(\tau \right),`$
$`{\displaystyle \underset{(a,b)}{}}Q_{\left(ab\right)\overline{\nu }}M_{\left(ab\right)}\left(1/\tau \right)`$ $`=`$ $`{\displaystyle \frac{1}{\tau ^4}}{\displaystyle \underset{(a,b)}{}}Q_{(a,b)\overline{\nu }}M_{ab}\left(\tau \right).`$ (B.95)
Si può in questo modo fissare la fase dell’ampiezza.
Possiamo ora calcolare in maniera diretta la funzione di vuoto. Dal momento che non si hanno vertici l’ampiezza è semplicemente
$$A_1=Vol\left(M\right)__1\frac{d^2\tau }{\tau _2^4}\underset{\nu \overline{\nu }}{}Q_{\nu \overline{\nu }}M_\nu M_{\overline{\nu }},$$
(B.96)
che risulta annullarsi, per una relazione di identità delle funzioni $`\theta `$ appartenente alla serie di identità di Riemann. In maniera simile, con l’aiuto di queste relazioni, si trova che anche le funzioni a uno-, due e tre-punti si annullano.
Il significato fisico di questo risultato è che lo spazio tempo piatto di Minkowski è una soluzione delle equazioni di superstringa al livello ad un loop, e che non ci sono rinormalizzazioni della massa e degli accoppiametni a questo ordine. Gli ultimi due risultati vanno sotto il nome di *teoremi di non rinormalizzazione*.
Rimane da discutere la funzione a quattro-punti per gli stati a massa nulla $`NSNS`$. Chiaramente questa ampiezza non può annullarsi in una teoria interagente. In questo contesto ci limitiamo a fornire il risultato, rimandando alle referenze per i dettagli del calcolo . Si trova
$$A_1(k_i,ϵ_i)=\delta \left(k\right)ϵ^{1\overline{1}}ϵ^{2\overline{2}}ϵ^{3\overline{3}}ϵ^{4\overline{4}}K^{1234}K^{\overline{1}\overline{2}\overline{3}\overline{4}}A_1(s,t),$$
(B.97)
dove
$`A_1(s,t)={\displaystyle \frac{1}{2}}{\displaystyle __1}{\displaystyle \frac{d^2\tau }{\tau _2^2}}{\displaystyle _\mathrm{\Sigma }}{\displaystyle \frac{d^2z_1}{\tau _2}}{\displaystyle _\mathrm{\Sigma }}{\displaystyle \frac{d^2z_2}{\tau _2}}{\displaystyle _\mathrm{\Sigma }}{\displaystyle \frac{d^2z_3}{\tau _2}}`$ $`e^{{}_{\frac{s}{2}}{}^{}(G_{12}+G_{34}G_{13}G_{24})}`$ (B.98)
$`e^{+\frac{t}{2}\left(G_{23}+G_{14}G_{13}G_{24}\right)}.`$
I propagatori dei campi $`X`$ ad un loop sono
$$G_{ij}=G(z_iz_j)\left|\tau \right)\mathrm{ln}\left|\frac{\vartheta _1\left(z_iz_j|\tau \right)}{\theta _1^{}\left(0|\tau \right)}\right|^2+\frac{2\pi }{\tau _2}Im(z_iz_j)^2.$$
(B.99) |
warning/0507/cond-mat0507697.html | ar5iv | text | # Exchange interactions and temperature dependence of the magnetization in half-metallic Heusler alloys
## I Introduction
During the last decade the half-metallic ferromagnets have become one of the most studied classes of materials. The existence of a gap in the minority-spin band structure leads to 100% spin-polarization of the electron states at the Fermi level and makes these systems attractive for applications in the emerging field of spintronics. Zutic In half-metals the creation of a fully spin-polarized current should be possible that should maximize the efficiency of magnetoelectronics devices.deBoeck
The half-metallicity was first predicted by de Groot and collaborators in 1983 when studying the band structure of a half-Heusler alloy NiMnSb.deGroot They found that the spin-down channel is semiconducting. In 2002 Galanakis et al. have shown that the gap arises from the interaction between the $`d`$-orbitals of Ni and Mn creating bonding and antibonding states separated by a gap. GalanakisHalf Ishida and collaborators have proposed that also the full-Heusler compounds of the type Co<sub>2</sub>MnZ, where Z stands for Si and Ge, are half-metals.Ishida In these compounds the origin of half-metallicity is more complex than in the half-Heusler alloys because of the presence of the states located entirely at the Co sites.GalanakisFull Several other Heusler alloys have been predicted to be half-metals.various Akinaga and collaborators Akinaga were able to crystallize a CrAs thin film in the zinc-blende structure, that is similar to the lattice of the Heusler alloys. The magnetic moment per formula unit was found to be close to 3$`\mu _B`$ that corresponds to the integer value characteristic for half-metals. A number of further half-metallic materials are CrO<sub>2</sub> in a metastable cubic phase, Fe<sub>3</sub>O<sub>4</sub>, the manganites (e.g La<sub>0.7</sub>Sr<sub>0.3</sub>MnO<sub>3</sub>)Soulen , the diluted magnetic semiconductors (e.g. Mn impurities in Si or GaAs). Freeman ; Akai
Besides strong spin polarization of the charge carriers in the ground state the spintronics materials must possess a high Curie temperature to allow the applications in the devices operating at room temperature. Available experimental information shows that the Heusler alloys are promising systems also in this respect. Webster Up to now the main body of the theoretical studies was devoted to the properties of the half-metallic gap.Nanda Recently, Chioncel and collaborators studied the influence of the correlation effects on the electron structure of CrAs.Chioncel They found that the spin-magnon interaction leads to the appearance of non-quasiparticle states in the spin-minority channel. The states are shown to lie above the Fermi level and to be sensitive to the value of the lattice constant. For a number of Heusler alloys it was shown that half-metallicity is preserved under tetragonalization of the crystal lattice Block and application of the hydrostatic pressure Picozzi-Gala . Mavropoulos et al. studied the influence of the spin-orbit coupling on the spin-polarization at the Fermi level and found the effect to be very small Mavropoulos that is in agreement with a small orbital moment calculated by Galanakis. GalanakisOrbit Larson et al.Larson have shown that the structure of Heusler alloys is stable with respect to the interchange of atoms and Orgassa and collaborators and Picozzi and collaborators have demonstrated that a small degree of disorder does not destroy the half-metallic gap.Orgassa ; Picozzi2004 Dowben and Skomski have shown that at non-zero temperatures the spin-wave excitations lead to the presence at the Fermi level of the electron states with opposite spin projections leading to decreasing spin-polarization of the charge carriers. Dowben
Despite very strong interest to the half-metallic ferromagnetism in Heusler alloys the number of theoretical studies of exchange interactions and Curie temperature in Heusler alloys is still very small. The first contribution to the density functional theory of the exchange interactions in these systems was made in an early paper by Kübler et al. Kubler83 where the microscopic mechanisms of the magnetism of Heusler alloys were discussed on the basis of the comparison of the ferromagnetic and antiferromagnetic configurations of the Mn moments. Recently, the studies of the inter-atomic exchange interactions in several Heusler compounds were reported by the present authors and Kurtulus et al. Sasioglou2004 ; Sasioglou2005 ; Kurtulus . Şaşıog̃lu et al. studied the exchange interactions in non-half-metallic Ni<sub>2</sub>MnZ (Z=Ga,In,Sn,Sb) and half-metallic Mn<sub>2</sub>VZ (Z=Al,Ge). The importance of the inter-sublattice exchange interaction has been demonstrated. For example, in the case of Mn<sub>2</sub>VZ (Z=Al,Ge) it was shown that the antiferromagnetic coupling between the V and Mn moments stabilizes the ferromagnetic alignment of the Mn moments. KüblerKubler2003 estimated $`T_C`$ of NiMnSb to be 601 K to 701 K depending on the approach used in the calculations. These values are in good correlation with experimental value of 730 K.Webster
The main task of the present contribution is the study of the exchange interactions in both half- and full-Heusler alloys. We use the calculated exchange parameters to estimate the Curie temperature in both the random phase ($`T_C^{RPA}`$) and the mean field approximations ($`T_C^{MFA}`$). In Section II we briefly discuss the formalism employed in the calculations. In Section III we present the results on the spin magnetic moments and the density of states (DOS) for four compounds studied: NiMnSb, CoMnSb, Co<sub>2</sub>MnSi and Co<sub>2</sub>CrAl. In Section IV we discuss the calculated exchange interactions and Curie temperatures. Section V is devoted to the consideration of the temperature dependence of magnetization. The films of Heusler alloys grown on different substrates can have different lattice parameters and, as a result, noticeable variation of the electron structure. Section V contains the summary. In the Appendix, we present the formalism for the calculation of the Curie temperature of a multi-sublattice ferromagnet within the framework of the random phase approximation.
## II Calculational Method
Half- and full-Heusler alloys crystallize in the $`C1_b`$ and $`L2_1`$ structures respectively (see Fig. 1). The lattice consists from 4 interpenetrating fcc lattices. In the case of the half-Heusler alloys (XYZ) one of the four sublattices is vacant. The Bravais lattice is in both cases fcc. In full-Heusler alloy the atomic basis consists of four atoms. For example, in Co<sub>2</sub>MnSi the positions of the basis atoms in Wyckoff coordinates are the following: Co atoms at $`(\mathrm{0\hspace{0.25em}0\hspace{0.25em}0})`$ and $`(\frac{1}{2}\frac{1}{2}\frac{1}{2})`$, Mn at $`(\frac{1}{4}\frac{1}{4}\frac{1}{4})`$, Si at $`(\frac{3}{4}\frac{3}{4}\frac{3}{4})`$. The Co atoms at the two different sublattices have the same local environment rotated by 90<sup>o</sup> with respect to the axis. In half-Heusler compounds the position $`(\frac{1}{2}\frac{1}{2}\frac{1}{2})`$ is vacant.
The calculations are carried out with the augmented spherical waves method (ASW)asw within the atomic-sphere approximation (ASA).asa The exchange-correlation potential is chosen in the generalized gradient approximation. gga A dense Brillouin zone (BZ) sampling $`30\times 30\times 30`$ is used. The radii of all atomic spheres are chosen equal. In the case of half-Heusler alloys we introduce an empty sphere located at the unoccupied site.
### II.1 Exchange parameters
The method for the calculation of exchange constants has been presented elsewhere.Sasioglou2004 Here we give a brief overview.
We describe the interatomic exchange interactions in terms of the classical Heisenberg Hamiltonian
$$H_{eff}=\underset{\mu ,\nu }{}\underset{\begin{array}{c}^{𝐑,𝐑^{}}\\ ^{(\mu 𝐑\nu 𝐑^{})}\end{array}}{}J_{\mathrm{𝐑𝐑}^{}}^{\mu \nu }𝐞_𝐑^\mu 𝐞_𝐑^{}^\nu $$
(1)
In Eq. (1), the indices $`\mu `$ and $`\nu `$ number different sublattices and $`𝐑`$ and $`𝐑^{}`$ are the lattice vectors specifying the atoms within sublattices, $`𝐞_𝐑^\mu `$ is the unit vector pointing in the direction of the magnetic moment at site $`(\mu ,𝐑)`$.
We employ the frozen-magnon approach to calculate interatomic Heisenberg exchange parameters.magnon The calculations involve few steps. In the first step, the exchange parameters between the atoms of a given sublattice $`\mu `$ are computed. The calculation is based on the evaluation of the energy of the frozen-magnon configurations defined by the following atomic polar and azimuthal angles
$$\theta _𝐑^\nu =\theta ,\varphi _𝐑^\nu =𝐪𝐑+\varphi ^\nu .$$
(2)
The constant phase $`\varphi ^\nu `$ is chosen equal to zero. The magnetic moments of all other sublattices are kept parallel to the z axis. Within the Heisenberg model (1) the energy of such configuration takes the form
$$E^{\mu \mu }(\theta ,𝐪)=E_0^{\mu \mu }(\theta )+\mathrm{sin}^2\theta J^{\mu \mu }(𝐪)$$
(3)
where $`E_0^{\mu \mu }`$ does not depend on q and the Fourier transform $`J^{\mu \nu }(𝐪)`$ is defined by
$$J^{\mu \nu }(𝐪)=\underset{𝐑}{}J_{0𝐑}^{\mu \nu }\mathrm{exp}(i𝐪𝐑).$$
(4)
In the case of $`\nu =\mu `$ the sum in Eq. (4) does not include $`𝐑=0`$. Calculating $`E^{\mu \mu }(\theta ,𝐪)`$ for a regular $`𝐪`$-mesh in the Brillouin zone of the crystal and performing back Fourier transformation one gets exchange parameters $`J_{0𝐑}^{\mu \mu }`$ for sublattice $`\mu `$. The determination of the exchange interactions between the atoms of two different sublattices $`\mu `$ and $`\nu `$ is discussed in Ref. Sasioglou2004, .
### II.2 Curie temperature
The Curie temperature is estimated within two different approaches: the mean–field approximation (MFA) and random phase approximation (RPA). The MFA for a multi-sublattice material requires solving the system of coupled equationsSasioglou2004 ; Anderson
$$e^\mu =\frac{2}{3k_BT}\underset{\nu }{}J_0^{\mu \nu }e^\nu $$
(5)
where $`e^\nu `$ is the average $`z`$ component of $`𝐞_𝐑^\nu `$ and $`J_0^{\mu \nu }_𝐑J_{0𝐑}^{\mu \nu }`$. Eq. 5 can be represented in the form of eigenvalue matrix-problem
$$(𝚯T𝐈)𝐄=0$$
(6)
where $`\mathrm{\Theta }_{\mu \nu }=\frac{2}{3k_B}J_0^{\mu \nu }`$, $`𝐈`$ is a unit matrix and $`𝐄`$ is the vector of $`e^\nu `$. The largest eigenvalue of matrix $`\mathrm{\Theta }`$ gives the value of $`T_C^{MFA}`$.Anderson
A more consequent method for the study of the thermodynamics of Heisenberg systems is provided by the RPA approach. tyablikov ; Callen The RPA technique is intensively used for studies of both single-sublattice pajda ; bouzerar and multi-sublatticeantiferro ; ferri ; magnetite ; Azaria ; t\_layer ; m\_layer\_1 ; m\_layer\_2 ; Nolting ; Turek2005 systems. In the case that only the exchange interactions within one sublattice are important the Curie temperature within the RPA is given by the relationpajda
$$\frac{1}{k_BT_C^{RPA}}=\frac{3}{2}\frac{1}{N}\underset{q}{}\frac{1}{J(\mathrm{𝟎})J(𝐪)},$$
(7)
We use the RPA approach to study the temperature dependence of the magnetization in the temperature interval from 0 K to $`T_C`$. The RPA technique for a multi-sublattice system is briefly presented in Appendix.
## III DOS and magnetic moments
### III.1 NiMnSb and CoMnSb
In this section we report the calculation of DOS and magnetic moments at different lattice parameters for NiMnSb and for CoMnSb compound that has one electron per formula unit less than NiMnSb. The electronic structure of both compounds has been extensively studied earlier and the reader is referred to the review GalanakisReview, and references therein for detailed discussion. Here we present a brief description of the calculational results aiming to provide the basis for further considerations and to allow the comparison with previous work.
In Table 1 we collect the atomic and total spin moments for three different lattice parameters. The investigation of the influence of the value of the lattice parameter on the properties of the Heusler alloys is important since the samples grown on different substrates can have different lattice spacings. The first calculation is performed for the experimental bulk lattice constant.Webster The calculated densities of states (DOS) for this case are presented in the upper panel of Fig. 2. For both NiMnSb and CoMnSb the Fermi level lies in the low-energy part of the half-metallic gap. The compression of the lattice pushes the majority $`p`$ states to higher energies that results in increased energy position of the Fermi level with respect to the half-metallic gap. At the lattice parameter $`a_{II}`$ the Fermi level coincides with the upper edge of the gap (Fig. 2). In the next step we further contracted the lattice constant by 1% (lattice parameter $`a_{III}`$, bottom panel in Fig. 2). In this case the Fermi level is slightly above the gap and the total spin moment is slightly smaller than the integer values of 3 and 4 $`\mu _\mathrm{B}`$ for CoMnSb and NiMnSb respectively.
The contraction of the lattice leads to an increase of the hybridization between the $`d`$ orbitals of different transition-metal atoms. This results in a decrease of the spin moment of Mn. In the case of NiMnSb this change is small: the reduction of the Mn spin moment under lattice contraction from the experimental lattice parameter to $`a_{II}`$ is $``$0.2 $`\mu _B`$. The Ni spin moment increases by about the same value to preserve the integer value of the total spin moment of 4 $`\mu _\mathrm{B}`$.
In CoMnSb, the half-metallic gap is larger than in NiMnSb. As a result, the transition of the Fermi level to the upper gap-edge requires a large lattice contraction of 11% (Table 1). This leads to a strong decrease of the Mn moment by 0.84$`\mu _\mathrm{B}`$. To compensate this decrease the Co moment changes its sign transforming the magnetic structure from ferrimagnetic to ferromagnetic.
The influence of the lattice contraction on the exchange interactions and Curie temperature is discussed in the next Section.
### III.2 Co<sub>2</sub>CrAl and Co<sub>2</sub>MnSi
The second group of materials studied in the paper is formed by the full-Heusler compounds Co<sub>2</sub>MnSi and Co<sub>2</sub>CrAl. The electronic structure of these systems has been studied earlier. GalanakisFull Compared to half-Heusler systems, the presence of two Co atoms per formula unit results in an increased coordination number of Co atoms surrounding Mn atoms (eight instead of four in CoMnSb). This leads to an increased hybridization between the $`3d`$ orbitals of the Mn and Co atoms. The spin moment of Co in Co<sub>2</sub>MnSi is about 1 $`\mu _\mathrm{B}`$ that is considerably larger than the Co moment in CoMnSb. In Co<sub>2</sub>CrAl the Co moment is about 1/3rd smaller than in Co<sub>2</sub>MnSi that reflects a smaller value of the Cr moment compared to the Mn moment (Table 1).
As in the case of the half-Heusler compounds discussed above, the variation of the lattice parameter leads to the change in the position of the Fermi level. At the experimental lattice parameter the Fermi level of Co<sub>2</sub>CrAl lies in the lower part of the half-metallic gap while for Co<sub>2</sub>MnSi it is close to the middle of the gap (Fig. 2b). The contraction of the lattice needed to place the Fermi level at the upper edge of the gap is smaller than for CoMnSb. As a result, the change in the magnetic moments is also relatively weak (Table 1).
## IV Exchange parameters and Curie temperature
### IV.1 NiMnSb and CoMnSb
In Fig. 3 we present the exchange constants calculated for various lattice spacings. The Co-Co, Ni-Ni exchange interactions as well as the exchange interactions between the moments of the 3d atoms and the induced moments of Sb atoms are very weak and are not shown. The weakness of the effective Co-Co and Ni-Ni exchange interactions can be explained by a relatively large distance between atoms (Fig. 1) and relatively small atomic moments.
On the other hand, each Ni(Co) atom is surrounded by four Mn atoms as nearest neighbors that results in strong Mn-Ni(Co) exchange interaction (Fig. 3). Also the exchange interaction between large Mn moments is strong.
The ferromagnetic Mn-Mn interactions are mainly responsible for the stable ferromagnetism of these materials. For both systems and for all lattice spacings studied the leading Mn-Mn exchange interaction is strongly positive. In NiMnSb, the Mn-Ni interaction of the nearest neighbors is positive for all three lattice parameters leading to the parallel orientation of the spins of the Mn and Ni atoms. In CoMnSb the situation is different. At the experimental lattice parameter the leading Mn-Co interaction is negative resulting in the ferrimagnetism of the system. For the contracted lattices the interaction changes sign resulting in the ferromagnetic ground state of the alloy.
The analysis of the strength of the exchange interaction as a function of the lattice parameter shows that in CoMnSb the contraction leads to a strong increase of both leading Mn-Co and Mn-Mn interactions. On the other hand, in NiMnSb the increase of the Mn-Ni interaction is accompanied by a decrease of the leading Mn-Mn interaction. Simultaneously, the interaction between the second-nearest Mn atoms increases with contraction in the case of NiMnSb staying almost unchanged in CoMnSb. This complexity of the behavior reflects the complexity of the electronic structure of the systems.
The interatomic exchange parameters are used to evaluate the Curie temperature within two different approaches: MFA and RPA. In Table 2 we present the values of the Curie temperature obtained, first, by taking into account the Mn-Mn interactions only and, second, with account for both Mn-Mn and Mn-Ni(Co) interactions. The contribution of the inter-sublattice interactions to the Curie temperature appears to be less than 5 % for both compounds and the Curie temperature is mainly determined by the intra-sublattice Mn-Mn interaction.
The MFA and RPA estimations of the Curie temperature differ rather strongly (Table 2). The relative difference of two estimations is about 20%. The reason behind this difference will be discussed in the following section. For the systems considered here the RPA estimations of the Curie temperatures are in good agreement with the experiment, somewhat overestimating the experimental values.
Recently Kübler Kubler2003 reported estimations of the Curie temperature of NiMnSb. His approach is based on the evaluation of the non-uniform magnetic susceptibility on the basis of the Landau-type expansion for the free energy. Within some approximations the parameters used in the study of the thermodynamical properties can be expressed in terms of the quantities evaluated within the first-principles DFT calculations. The estimated values of the Curie temperature are 601 K for a static approach and 701 K if the frequency dependence of the susceptibility is taken into account. These estimations are somewhat lower than the value of 880 K given by the RPA approach (Table 2). A detailed comparative analysis of the two calculational schemes is needed to get an insight in the physical origin of this difference.
The contraction of the lattice in the case of the NiMnSb compound leads to an increase of the Mn-Ni interactions (Fig. 3). This results in increased difference between the Curie temperatures calculated with the Mn-Mn interactions only and with both Mn-Mn and Ni-Mn interactions taken into account (Table 2). For CoMnSb, the leading exchange interactions of both Mn-Mn and Mn-Co types increase in the value under transition from the experimental lattice constant to $`a_{II}`$ (Fig. 3). As a result, the Curie temperature increases with contraction by about 50%.
### IV.2 Co<sub>2</sub>CrAl and Co<sub>2</sub>MnSi
The presence of an extra Co atom in the full-Heusler alloys makes the interactions more complex than in the case of the half-Heusler alloys. In CoMnSb the important interactions arise between nearest Mn atoms (Mn-Mn interactions) and between nearest Mn and Co atoms (Mn-Co interaction). In the case of Co<sub>2</sub>MnSi (Fig. 4) the interactions between Co atoms at the same sublattice (Co-Co) and between Co atoms at different sublattices (Co<sup>1</sup>-Co<sup>2</sup>) must be taken into account. The cobalt atoms at different sublattices have the same local environment rotated by 90<sup>o</sup> about the axis. The leading interaction responsible for the stability of the ferromagnetism is the Mn-Co interaction between Mn atoms and eight nearest Co atoms (Fig. 4). This interaction changes weakly with the contraction of the lattice. Our exchange parameters agree well with the parameters of Kurtulus et al. (Fig. 4) who also found the Co-Mn exchange interaction to be leading. Kurtulus
The interaction between nearest Co atoms at different sublattices (empty squares in Fig. 4) favors the ferromagnetism also and is stronger than the ferromagnetic interaction between the nearest Mn atoms (filled spheres). Although the spin moment of Mn atoms is larger than the moment of Co atoms (Table 1) the opposite relation between exchange parameters can be the consequence of the smaller distance between the Co atoms: $`a/2`$ between the Co atoms and $`\sqrt{2}a/2`$ between the Mn atoms. An interesting feature of the intra-sublattice Mn-Mn and Co-Co interactions is different signs of the exchange parameters for different distances between atoms. This leads to a RKKY-like oscillations of the parameters (Fig. 4).
In Co<sub>2</sub>CrAl the leading Cr-Co interactions (filled triangles) are much smaller than corresponding Mn-Co interactions in Co<sub>2</sub>MnSi. On the other hand, the leading inter-sublattice ferromagnetic Co-Co interactions are comparable in both systems. The compression of the lattice leads to an increase of the magnitude of the inter-sublattice Co-Cr and Co<sup>1</sup>-Co<sup>2</sup> coupling. The intra-sublattice Cr-Cr and Co-Co interactions oscillate with varying inter-atomic distances.
The difference in the properties of the exchange parameters of the half- and full-Heusler alloys is reflected in the calculated Curie temperatures (Table 2). In contrast to CoMnSb where the Mn-Mn exchange interactions are dominant, in Co<sub>2</sub>MnSi they play a secondary role. The T$`{}_{}{}^{MFA(RPA)}{}_{C}{}^{}`$ calculated taking into account these interactions only is much smaller than the Curie temperature calculated with all inter-atomic exchange interactions taken into account (Table 2). The same conclusion is valid for Co<sub>2</sub>CrAl where the Cr-Cr interactions give about half of the Curie temperature obtained with all interactions included into consideration.
A striking feature of the full Heusler compound Co<sub>2</sub>CrAl that differs it strongly from the half-Heusler systems considered in the previous Section is a very small difference between the $`T_C`$ values calculated within the MFA and RPA approaches. A similar behavior was obtained for the Curie temperatures of the zincblende MnSi and MnC.stability\_zb In Co<sub>2</sub>MnSi, the relative difference of the MFA and RPA estimations assumes an intermediate position between the half-Heusler systems and Co<sub>2</sub>CrAl.
To understand the origin of the strong variation of the relative difference of the MFA and RPA estimations of the Curie temperature we compare in Fig. 5 the frozen magnon dispersions for two compounds. The magnons correspond to the Mn sublattice in the case of NiMnSb and to the Cr sublattice in the case of Co<sub>2</sub>CrAl. As seen from Table 2 the MFA and RPA estimations obtained with the use of these dispersions differ by 20% for NiMnSb and by 5% for Co<sub>2</sub>CrAl.
The Curie temperature is given by the average value of the magnon energies. In MFA this is the arithmetic average while in RPA this is harmonic average. Therefore we need to understand why for Co<sub>2</sub>CrAl these two averages are much closer than for NiMnSb. The following properties of the averages are important for us. The arithmetic average takes all the magnon values with equal weight whereas in the harmonic average the weight decreases with increasing energy of the magnon.pajda ; bouzerar ; stability\_zb ; above\_room ; sabiryanov It is an arithmetic property that the MFA estimation is larger than the RPA one or equal to it if all numbers to be averaged are equal to each other. In terms of magnon energies, $`T_C^{MFA}`$ is equal to $`T_C^{RPA}`$ in the case that the magnon spectrum is dispersion-less.
Considering the frozen-magnon dispersions from the viewpoint of these properties we indeed can expect that the arithmetic and harmonic averages will be closer for Co<sub>2</sub>CrAl. In Fig. 5 both curves are scaled to have almost the same maximal value. It is seen that the Co<sub>2</sub>CrAl dispersion has smaller relative contribution of the low-energy magnons because of the steeper increase of the curve at small wave vectors. It has also smaller contribution of the magnons with the largest energies because the maxima have the form of well-defined peaks opposite to NiMnSb where we get a plateau. Thus the main contribution in the case of Co<sub>2</sub>CrAl comes from intermediate energies that makes the MFA and RPA estimations closer.
In Fig. 6 we present the calculated spin-wave spectra for NiMnSb and Co<sub>2</sub>CrAl. The spin-wave energies are obtained by the diagonalization of the matrix of exchange parameters that contains all important intra- and inter-sublattice interactions. The number of branches in the spectrum is equal to the number of magnetic atoms in the unit cell: two in NiMnSb and three in Co<sub>2</sub>CrAl. One of the branches is acoustic and has zero energy for zero wave vector. Also in the spin-wave spectra, we see strong difference between two systems. In NiMnSb, the acoustic branch is predominantly of the Ni type stemming from the weak interaction between Mn and Ni magnetic moments (see Fig. 3). On the other hand, the optical branch is of predominantly the Mn type. The strong hybridization between two sublattices is obtained only about $`\text{q}=0`$. In Co<sub>2</sub>CrAl, the energy scale of the branches differs much smaller and the hybridization between sublattices is stronger than for NiMnSb.
Coming back to the considerations of the Curie temperatures, we conclude that, in general, the Curie temperatures of Co<sub>2</sub>MnSi and Co<sub>2</sub>CrAl calculated within both MFA and RPA are in good agreement with experiment while the MFA values in the case of NiMnSb and CoMnSb overestimate the Curie temperature strongly.
The lattice contraction leads in both compounds to an enhancement of the Mn-Co(Cr-Co) exchange constants that results in an increase of the Curie temperature.
Kurtulus and collaborators have calculated the Curie temperature for Co<sub>2</sub>MnSi within MFA and found the value of 1251 K that is considerably larger than our MFA estimate of 857 K. This difference is unexpected since the values of the exchange parameters obtained by Kurtulus et al. agree well with our parameters (Fig.4). To reveal the origin of the discrepancy we performed the MFA calculation of the Curie temperature with the exchange parameters of Kurtulus et al. and obtained the $`T_C`$ value of 942 K which is in reasonable agreement with our estimate. Apparently the reason for the inconsistency is in the procedure of the solving of the multiple-sublattice MFA problem used by Kurtulus et al. that should deviate from the standard one. Anderson
## V Temperature dependence of the magnetization
The study of the temperature dependence of the magnetic properties of itinerant ferromagnets is one of the fundamental problems of ongoing researches. Although density functional theory can formally be extended to the finite temperatures Mermin , it is rarely used because of the lack of suitable exchange-correlation potentials for magnetic systems at finite temperatures. Statistical mechanics treatment of model Hamiltonians is usually employed. In this section we will present the results of the calculation of the temperature dependence of magnetization that is based on the consideration of the Heisenberg hamiltonian with exchange parameters calculated within a parameter-free DFT approach (Sect. II).
To calculate the temperature dependence of the magnetization we use the RPA method as described in appendix. We consider both classical-spin and quantum-spin cases.
In the classical-spin calculations the calculated values of the magnetic moments (Table 1) are used. To perform quantum-mechanical RPA calculation we assign integer values to the atomic moments. In the semi Heusler compounds we ignore the induced moments on Ni and Co atoms and assign the whole moment per formula unit to the Mn atom: 4$`\mu _B`$ ($`S=2`$) in NiMnSb and 3$`\mu _B`$ ($`S=3/2`$) in CoMnSb. In Co<sub>2</sub>MnSi we take the values of 3$`\mu _B`$ ($`S=3/2`$) and 1$`\mu _B`$ ($`S=1/2`$) for Mn and Co atoms respectively. This assignment preserves the value of the total spin moment per chemical unit. In Co<sub>2</sub>CrAl we use in the quantum-RPA calculations the atomic moment of 2$`\mu _B`$ ($`S=1`$) for Cr and 1$`\mu _B`$ ($`S=1/2`$) for Co.
In Fig. 7(a), we present in the normalized form the calculated temperature dependence of the magnetization for both families of Heusler compounds. The calculations are performed for the experimental lattice parameter. For comparison, the experimental curves are presented. The nature of the spin (quantum or classical) influences the form of the curves considerably. The classical curve lies lower than the quantum one. This results from a faster drop of the magnetization in the low-temperature region in the case of classical spins. In general, the quantum consideration gives better agreement of the form of the temperature dependence of the magnetization with experiment.
In Fig. 7(b) we present the temperature dependence of the magnetization of individual sublattices. As expected from the previous discussions in half-Heusler systems the main contribution to the magnetization comes from the Mn sublattice while for the full-Heusler systems both 3d atoms contribute substantially.
Considering the calculated Curie temperatures we notice that the value of $`T_C`$ calculated within the quantum-mechanical RPA is substantially larger than the corresponding classical estimation (see Fig. 7). This property is well-known and has its mathematical origin in the factor $`(S+1)/S`$ entering the RPA expression for the Curie temperature (Eq. 21). In Fig. 8 we show the dependence of the Curie temperature calculated within the quantum mechanical RPA approach on the value of $`S`$. The exchange parameters are kept unchanged in these calculations. We see that the dependence has a monotonous character tending to a classical limit for large $`S`$.
Presently we do not have an explanation why quantum-mechanical calculations give better form of the temperature dependence while the classical calculation provides better value of the Curie temperature. We can suggest the following arguments. The quantum treatment is more appropriate than the classical one in the low-temperature region. At high temperatures characterized by strong deviation of the atomic spins from the magnetization axis the quantum treatment gives too slow decrease of the magnetization. It is worth noting that the consequent theory should take into account not only the orientational disorder of the atomic moments but also the single-particle (Stoner-type) excitations leading to the decrease of atomic moments. Another important aspect is related to the fact that the exchange parameters used in the calculations are estimated within the picture of classical atomic moments described above. It is possible that the values of the exchange parameters must be modified for the use in the quantum-mechanical case. These questions belong to fundamental problems of the quantum-mechanical description of the magnetic systems with itinerant electrons.
## VI Summary and conclusions
We studied the electronic structure of several Heusler alloys using the augmented spherical waves method in conjunction with the generalized gradient approximation to the exchange and correlation potential. Using the frozen-magnon approximation we calculated inter-atomic exchange parameters that were used to estimate the Curie temperature. The Curie temperature was estimated within both mean-field and random-phase approximation techniques.
For the half-Heusler alloys NiMnSb and CoMnSb the dominant interaction is between the Mn atoms. The lattice compression results in considerable change of the exchange parameters and Curie temperature.
The magnetic interactions are more complex in full-Heusler alloys Co<sub>2</sub>MnSi and Co<sub>2</sub>CrAl. In both cases the ferromagnetism is stabilized by the inter-sublattice interactions between the Mn(Cr) and Co atoms and between Co atoms belonging to different sublattices. Both the random phase and mean field approximations slightly underestimate the values of the Curie temperature. Compression of the lattice constant has little effect on the magnetic properties of the full-Heusler alloys.
We study the temperature dependence of the magnetization within the quantum mechanical and classical RPA. The quantum-mechanical approach gives the form of the temperature dependence that is in good agreement with experiment. The value of the Curie temperature is, however, overestimated in the quantum-mechanical calculation.
###### Acknowledgements.
The financial support of Bundesministerium für Bildung und Forschung is acknowledged. I.G. is a fellow of the Greek State Scholarship Foundation. We thank the authors of Ref. Turek2005, for making the manuscript available before publication.
## Appendix A The random phase approximation for multi-sublattice Heisenberg Hamiltonian
The Green function approach is a powerful tool in the study of the magnetism of complex systems. (See, e.g., the application of the method to antiferromagnetantiferro , ferrimagnetsferri ; magnetite , random alloysAzaria , layered systems t\_layer ; m\_layer\_1 ; m\_layer\_2 , disordered dilute magnetic systems Nolting , multi-sublattice ferromagnets Turek2005 .) In this appendix we briefly overview the formalism to study the temperature dependence of the magnetization of multi-sublattice systems within the random phase approximation.
We start with the Heisenberg Hamiltonian for quantum spins
$$H=\underset{ij}{}\underset{\mu \nu }{}J_{ij}^{\mu \nu }𝐞_{i,\mu }𝐞_{j,\nu }$$
(8)
where $`𝐞_{i,\mu }=(\widehat{s}_{i,\mu }^x,\widehat{s}_{i,\mu }^y,\widehat{s}_{i,\mu }^z)/(S_\mu )`$ is the normalized spin operator corresponding to site ($`i,\mu `$).
In terms of the creation and destruction operators $`\widehat{s}_{i,\mu }^{}=\widehat{s}_{i,\mu }^x\widehat{s}_{i,\mu }^y`$ the Hamiltonian can be written in the form
$$H=\underset{ij}{}\underset{\mu \nu }{}\stackrel{~}{J}_{ij}^{\mu \nu }[\widehat{s}_{i,\mu }^+\widehat{s}_{j,\nu }^{}+\widehat{s}_{i,\mu }^z\widehat{s}_{j,\nu }^z]$$
(9)
where $`\stackrel{~}{J}_{ij}^{\mu \nu }=J_{ij}^{\mu \nu }/S_\mu S_\nu `$.
Following Callen Callen let us introduce Green function
$$G_{ij}^{\mu \nu }(\tau )=\frac{i}{\mathrm{}}\theta (\tau )[\widehat{s}_{i,\mu }^+(\tau ),\mathrm{exp}(\eta \widehat{s}_{j,\nu }^z)\widehat{s}_{j,\nu }^{}]$$
(10)
where $`\eta `$ is a parameter, $`\theta (\tau )`$ is the step function ($`\theta (\tau )=1`$ for $`\tau 0`$), $`[\mathrm{}]`$ denotes the commutator and $`\mathrm{}`$ is the thermal average over the canonical ensemble, ie., $`F=\text{Tr}[\mathrm{exp}(\beta H)F]/\text{Tr}[\mathrm{exp}(\beta H)]`$ with $`\beta =1/k_BT`$
Writing the equation of motion for $`G_{ij}^{\mu \nu }(\tau )`$ we obtain
$`{\displaystyle \frac{}{\tau }}G_{ij}^{\mu \nu }(\tau )`$ $`=`$ $`{\displaystyle \frac{i}{h}}\delta (\tau )[\widehat{s}_{i,\mu }^+(\tau ),\mathrm{exp}(\eta \widehat{s}_{j,\nu }^z)\widehat{s}_{j,\nu }^{}]{\displaystyle \frac{1}{\mathrm{}^2}}\theta (\tau )`$ (11)
$`\times [[\widehat{s}_{i,\mu }^+(\tau ),\widehat{H}],\mathrm{exp}(\eta \widehat{s}_{j,\nu }^z)\widehat{s}_{j,\nu }^{}]`$
The last commutator term in Eq. (11) generates higher-order Green functions. These functions can be reduced to lower-order functions by using Tyablikov decoupling (random phase approximation) schemetyablikov :
$$[\widehat{s}_{i,\mu }^+(\tau )\widehat{s}_{k,\mu }^z,\widehat{s}_{j,\nu }^{}]\widehat{s}_{k,\mu }^z[\widehat{s}_{i,\mu }^+(\tau ),\widehat{s}_{j,\nu }^{}]$$
(12)
Applying this decoupling procedure to Eq. (11) we get
$`{\displaystyle \frac{}{\tau }}G_{ij}^{\mu \nu }(\tau )`$ $`=`$ $`{\displaystyle \frac{i}{h}}\delta (\tau )[\widehat{s}_{i,\mu }^+(\tau ),\mathrm{exp}(\eta \widehat{s}_{j,\nu }^z)\widehat{s}_{j,\nu }^{}]`$ (13)
$`+{\displaystyle \frac{2i}{\mathrm{}}}{\displaystyle \underset{k,\xi }{}}\stackrel{~}{J}_{i,k}^{\mu \xi }[\widehat{s}_{i,\mu }^zG_{kj}^{\xi \nu }(\tau )`$
$`\widehat{s}_{k,\xi }^zG_{ij}^{\mu \nu }(\tau )]`$
After a Fourier transformation to energy and momentum space \[$`g(𝐪,\omega )=\frac{1}{2\pi }_l𝑑\omega e^{i\mathrm{𝐪𝐑}_l}G_{l0}(\tau )`$\] we obtain
$`\mathrm{}\omega g_{\mu \nu }(𝐪,\omega )`$ $`=`$ $`{\displaystyle \frac{1}{2\pi }}[\widehat{s}_\mu ^+,\mathrm{exp}(\eta \widehat{s}_\nu ^z)\widehat{s}_\nu ^{}]\delta _{\mu \nu }`$ (14)
$`2{\displaystyle \underset{\xi }{}}\{\stackrel{~}{J}_{\mu \xi }(𝐪)\widehat{s}_{i,\mu }^zg_{\xi \nu }(𝐪,\omega )`$
$`\stackrel{~}{J}_{\mu \xi }(\mathrm{𝟎})\widehat{s}_{k,\xi }^zg_{\mu \nu }(𝐪,\omega )\}`$
Eq. (14) can be written in a compact matrix form
$$[\mathrm{}\omega \text{I}\text{M}(𝐪)]\text{g}(𝐪,\omega )=\text{u}$$
(15)
where $`\text{g}(𝐪,\omega )`$ is a symmetric square matrix, I is a unit matrix and the inhomogeneity matrix $`𝐮`$ is expressed by
$$u_{\mu \nu }=\frac{1}{2\pi }[\widehat{s}_\mu ^+,\mathrm{exp}(\eta \widehat{s}_\nu ^z)\widehat{s}_\nu ^{}]\delta _{\mu \nu },$$
(16)
matrix $`\text{M}(𝐪)`$ is defined by
$$M_{\mu \nu }(𝐪)=\left\{\underset{\xi }{}2\stackrel{~}{J}_{\mu \xi }(\mathrm{𝟎})\widehat{s}_\xi ^z\right\}\delta _{\mu \nu }2\stackrel{~}{J}_{\mu \nu }(𝐪)\widehat{s}_\mu ^z$$
(17)
Next, we introduce the transformation which diagonalizes matrix $`𝐌(𝐪)`$:m\_layer\_2
$$𝐋(𝐪)𝐌(𝐪)𝐑(𝐪)=\mathrm{\Omega }(𝐪)$$
(18)
where $`\mathrm{\Omega }(𝐪)`$ is the diagonal matrix whose elements give the spin wave energies $`\omega _\mu (𝐪)`$. The number of branches in the spin wave spectrum is equal to the number of magnetic atoms in the unit cell. The transformation matrix $`𝐑(𝐪)`$ and its inverse $`𝐑^1(𝐪)=𝐋(𝐪)`$ are obtained from the right eigenvectors of $`𝐌(𝐪)`$ as columns and from the left eigenvectors as rows, respectively.
Using the spectral theorem and Callen’s technique Callen one obtains the thermal averages of the sublattice magnetizations:
$$\widehat{s}_\mu ^z=\frac{(S_\mu \mathrm{\Phi }_\mu )(1+\mathrm{\Phi }_\mu )^{2S_\mu +1}+(S_\mu +1+\mathrm{\Phi }_\mu )\mathrm{\Phi }_\mu ^{2S_\mu +1}}{(1+\mathrm{\Phi }_\mu )^{2S_\mu +1}(\mathrm{\Phi }_\mu )^{2S_\mu +1}}$$
(19)
where $`\mathrm{\Phi }_\mu `$ is an auxiliary function given by
$$\mathrm{\Phi }_\mu =\frac{1}{N}\underset{q}{}\underset{\nu }{}L_{\mu \nu }(𝐪)\frac{1}{e^{\beta \omega _\nu (𝐪)}1}R_{\mu \nu }(𝐪)$$
(20)
In Eq. (20), $`N`$ is the number of $`𝐪`$ points in the first BZ.
Eq. (19) is the central equation for the calculation of the sublattice magnetizations. It must be solved self-consistently. The Curie temperature $`T_C`$ is determined as the point where all sublattice magnetizations vanish.
Near $`T_C`$ ($`\mathrm{\Phi }_\mu \mathrm{}`$ and $`\widehat{s}_\mu ^z0`$) Eq. (19) can be simplified. Expanding in $`\mathrm{\Phi }_\mu `$ and using Eq. (20) one obtains
$$\widehat{s}_\mu ^z=\frac{(S_\mu +1)}{3S_\mu }\left\{\frac{1}{S_\mu ^2N}\underset{q,\nu }{}L_{\mu \nu }(𝐪)\frac{1}{e^{\beta \omega _\nu (𝐪)}1}R_{\mu \nu }(𝐪)\right\}^1.$$
(21)
From Eq. (21), it follows that for spin-independent Heisenberg exchange parameters \[Eq. (8)\] the dependence of the Curie temperature on the spin value is defined by the factor $`(S_\mu +1)/S_\mu `$.
The classical limit can be obtained by letting $`S_\mu \mathrm{}`$ in Eqs. (19) and (21).Turek2005 Factor $`(S_\mu +1)/S_\mu `$ in Eq. (21) becomes in this case unity. The temperature dependence of the magnetization can be calculated using a semiclassical analog of Eq. (19) given by Turek2005 ; Callen2
$$e_\mu ^z=\left(\left\{\frac{1}{N}\underset{q,\nu }{}L_{\mu \nu }(𝐪)\frac{1}{e^{\beta \omega _\nu (𝐪)}1}R_{\mu \nu }(𝐪)\right\}^1\right)$$
(22)
where $`(x)=\mathrm{coth}(x)1/x`$ is the Langevin function and $`𝐞_\mu `$ is the angular momentum vector of size one. |
warning/0507/gr-qc0507041.html | ar5iv | text | # 1 Introduction
## 1 Introduction
A satisfactorily empirical corroboration of a fundamental theory requires that as many independent experiments as possible are conducted by different scientists in different laboratories. Now, the general relativistic gravitomagnetic Lense-Thirring (LT) effect is difficult to test, also in the weak-field and slow-motion approximation, valid in our Solar System, both because such a relativistic effect is very small and the competing classical signals are often quite larger. Until now, a $`6\%`$ LT test has recently been conducted in the gravitational field of Mars by using the data of the Mars Global Surveyor (MGS) spacecraft; other tests accurate to about<sup>1</sup><sup>1</sup>1Other estimates point towards a $`1520\%`$ error. $`10\%`$ have been performed by a different team in the gravitational field of the Earth by analyzing the data of the LAGEOS and LAGEOS II artificial satellites (see Section 1.2). We think it is worthwhile to further extend these efforts trying to use different laboratories, i.e. other gravitational fields, even if the outcomes of such tests should be less accurate than those conducted so far. To this aim, in this paper we investigate the LT effect induced by the Sun on the orbital motion of the inner planets of the Solar System in the context of the latest results in the planetary ephemerides field.
### 1.1 The Lense-Thirring effect
According to Einstein, the action of the gravitational potential $`U`$ of a given distribution of mass-energy is described by the coefficients $`g_{\mu \nu },\mu ,\nu =0,1,2,3`$, of the space-time metric tensor. They are determined, in principle, by solving the fully non-linear field equations of the Einsteinian General Theory of Relativity (GTR) for the considered mass-energy content. These equations can be linearized in the weak-field ($`U/c^2<<1`$, where $`c`$ is the speed of light in vacuum) and slow-motion ($`v/c<<1`$) approximation (Mashhoon 2001; Ruggiero and Tartaglia 2002), valid throughout the Solar System, and look like the equations of the linear Maxwellian electromagnetism. Among other things, a noncentral, Lorentz-like force
$$𝑭_{\mathrm{LT}}=2m\left(\frac{𝒗}{c}\right)\times 𝑩_g$$
(1)
acts on a moving test particle of mass $`m`$. It is induced by the post-Newtonian component $`𝑩_g`$ of the gravitational field in which the particle moves with velocity $`𝒗`$. $`𝑩_g`$ is related to the mass currents of the mass-energy distribution of the source and comes from the off-diagonal components $`g_{0i},i=1,2,3`$ of the metric tensor. Thanks to such an analogy, the ensemble of the gravitational effects induced by mass displacements is also named gravitomagnetism. For a central rotating body of mass $`M`$ and proper angular momentum $`𝑳`$ the gravitomagnetic field is
$$𝑩_g=\frac{G[3𝒓(𝒓𝑳)r^2𝑳]}{cr^5}.$$
(2)
One of the consequences of eq. (1) and eq. (2) is a gravitational spin–orbit coupling. Indeed, if we consider the orbital motion of a particle in the gravitational field of a central spinning mass, it turns out that the longitude of the ascending node $`\mathrm{\Omega }`$ and the argument of pericentre $`\omega `$ of the orbit of the test particle are affected by tiny secular advances $`\dot{\mathrm{\Omega }}_{\mathrm{LT}}`$, $`\dot{\omega }_{\mathrm{LT}}`$ (Lense and Thirring 1918, Barker and O’Connell 1974, Cugusi and Proverbio 1978, Soffel 1989, Ashby and Allison 1993, Iorio 2001)
$$\dot{\mathrm{\Omega }}_{\mathrm{LT}}=\frac{2GL}{c^2a^3(1e^2)^{\frac{3}{2}}},\dot{\omega }_{\mathrm{LT}}=\frac{6GL\mathrm{cos}i}{c^2a^3(1e^2)^{\frac{3}{2}}},$$
(3)
where $`a,e`$ and $`i`$ are the semimajor axis, the eccentricity and the inclination, respectively, of the orbit and $`G`$ is the Newtonian gravitational constant. Note that in their original paper Lense and Thirring (1918) used the longitude of pericentre $`\varpi \mathrm{\Omega }+\omega `$.
The gravitomagnetic force may have strong consequences in many astrophysical and astronomical scenarios involving, e.g., accreting disks around black holes (Thorne et al. 1986; Stella et al. 2003), gravitational lensing and time delay (Sereno 2003; 2005a; 2005b). Unfortunately, in these contexts the knowledge of the various competing effects is rather poor and makes very difficult to reliably extract the genuine gravitomagnetic signal from the noisy background. E.g., attempts to measure the LT effect around black holes are often confounded by the complexities of the dynamics of the hot gas in their accretion disks. On the contrary, in the solar and terrestrial space environments the LT effect is weaker but the various sources of systematic errors are relatively well known and we have the possibility of using various artificial and natural orbiters both to improve our knowledge of such biases and to design suitable observables circumventing these problems, at least to a certain extent.
### 1.2 The performed and ongoing tests
Up to now, all the performed and ongoing tests of gravitomagnetism were implemented in the weak-field and slow-motion scenarios of the Earth and Mars gravitational fields.
As far as the Earth is concerned, in April 2004 the GP-B spacecraft (Everitt 1974; Fitch et al. 1995; Everitt et al. 2001) was launched. Its aim is the measurement of another gravitomagnetic effect, i.e. the precession of the spins (Pugh 1959; Schiff 1960) of four superconducting gyroscopes carried onboard. The level of accuracy obtained so far is about $`256128\%`$ (Muhlfelder et al. 2007), with the hope of reaching<sup>2</sup><sup>2</sup>2See on the WEB StanfordNews 4/14/07 downloadable at http://einstein.stanford.edu/. $`13\%`$ in December 2007.
In regard to the LT effect on the orbit of a test particle, the idea of using the LAGEOS satellite and, more generally, the Satellite Laser Ranging (SLR) technique to measure it in the terrestrial gravitational field with the existing artificial satellites was put forth for the first time by Cugusi and Proverbio (1978). Attempts to practically implement such a strategy began in 1996 with the LAGEOS and LAGEOS II satellites (Ciufolini et al. 1996). The latest test was performed by Ciufolini and Pavlis (2004). They analyzed the data of LAGEOS and LAGEOS II by using an observable independently proposed by Pavlis (2002), Ries et al. (2003a, 2003b) and Iorio and Morea (2004). The error claimed by Ciufolini and Pavlis (2004) is 5-10$`\%`$ at 1-3 sigma, respectively. The assessment of the total accuracy of such a test raised a debate (Iorio 2005a; 2005b; 2006a; 2007; Ciufolini and Pavlis 2005; Lucchesi 2005).
Recently, a $`6\%`$ LT test on the orbit of the Mars Global Surveyor (MGS) spacecraft in the gravitational field of Mars has been reported (Iorio 2006b); indeed, the predictions of general relativity are able to accommodate, on average, about $`94\%`$ of the measured residuals in the out-of-plane part of the MGS orbit over 5 years.
Finally, it must be noted that, according to Nordtvedt (2003), the multi-decade analysis of the Moon’orbit by means of the Lunar Laser Ranging (LLR) technique yields a comprehensive test of the various parts of order $`𝒪(c^2)`$ of the post-Newtonian equation of motion. The existence of the gravitomagnetic interaction as predicted by GTR would, then, be inferred from the high accuracy of the lunar orbital reconstruction. A 0.1$`\%`$ test was recently reported (Murphy et al. 2007): a critical discussion of the real sensitivity of LLR to gravitomagnetism can be found in (Kopeikin 2007). Also the radial motion of the LAGEOS satellite would yield another indirect confirmation of the existence of the gravitomagnetic interaction (Nordtvedt 1988).
## 2 The solar gravitomagnetic field
The action of the solar gravitomagnetic field on the Mercury’s longitude of perihelion was calculated for the first time by de Sitter (1916) who, by assuming an homogenous and uniformly rotating Sun, found a secular advance of $`0.01`$ arcseconds per century ( <sup>′′</sup> cy<sup>-1</sup> in the following). This value is also quoted by Soffel (1989). Cugusi and Proverbio (1978) yield $`0.02`$ <sup>′′</sup> cy<sup>-1</sup> for the argument of perihelion of Mercury. Instead, recent determinations of the Sun’s proper angular momentum $`L_{}=(190.0\pm 1.5)\times 10^{39}`$ kg m<sup>2</sup> s<sup>-1</sup> from helioseismology (Pijpers 1998; 2003), accurate to $`0.8\%`$, yield a precessional effect one order of magnitude smaller for Mercury (Ciufolini and Wheeler 1995; Iorio 2005c). See Table 1 for the gravitomagnetic precessions of the four inner planets.
As can be seen, they are of the order of $`10^310^5`$ <sup>′′</sup> cy<sup>-1</sup>.
So far, the LT effect on the orbits of the Sun’s planets was believed to be too small to be detected (Soffel 1989). Iorio (2005c) preliminarily investigated the possibility of measuring such tiny effects in view of recent important developments in the planetary ephemerides generation. It is remarkable to note that the currently available estimate of $`L_{}`$ is accurate enough to allow, in principle, a genuine test of GTR. Moreover, it was determined in a relativity-free fashion from astrophysical techniques which do not rely on the dynamics of planets in the gravitational field of the Sun. Thus, there is no any a priori ‘memory’ effect of GTR itself in the adopted value of $`L_{}`$.
## 3 Compatibility of the determined extra-precessions of planetary perihelia with the LT effect
### 3.1 The Keplerian orbital elements
The Keplerian orbital elements like $`\varpi `$ are not directly observable quantities like right ascensions, declinations, ranges and range-rates which can be measured from optical observations, radiometric measurements, meridian transits, etc. They can only be computed from a state vector in rectangular Cartesian coordinates which also allows to compute predicted values of the observations. In this sense, speaking of an “observed” time series of a certain Keplerian element would mean that it has been computed from the machinery of the data reduction of the real observations<sup>3</sup><sup>3</sup>3In practice, this procedure cannot be always performed at all times because the observations collected at a given time not always allow for the computation of the entire state vector in rectangular Cartesian coordinates. . Keeping this in mind, it would be possible, in principle, to extract the LT signal from the planetary motions by taking the difference between two suitably computed time-series of the Keplerian elements in such a way that it fully accounts for the gravitomagnetic signature. Such ephemerides, which should share the same initial conditions, would differ in the fact that one would be based on the processing of the real data, which are presumed to fully contain also the LT signal, and the other one would, instead, be the result of a purely numerical propagation. The dynamical force models with which the data are to be processed and the numerical ephemeris propagated do not contain the gravitomagnetic force itself: only the general relativistic gravitoelectric terms must be present. Moreover, the astronomical parameters entering the perturbations which can mimic the LT signature should not be fitted in the data reduction process: they should be kept fixed to some reference values, preferably obtained in a relativity-independent way so to avoid ‘imprinting’ effects. Thus, in the resulting “residual” time series $`\mathrm{\Delta }\varpi _{\mathrm{obs}}(t)`$, the LT signature should be entirely present.
### 3.2 The EPM2004 ephemerides
A somewhat analogous procedure was recently implemented with the Ephemerides of Planets and the Moon EPM2004 (Pitjeva 2005a; 2005b) produced by the Institute of Applied Astronomy (IAA) of the Russian Academy of Sciences (RAS). They are based on a data set of more than 317,000 observations (1913-2003) including radiometric measurements of planets and spacecraft, astrometric CCD observations of the outer planets and their satellites, and meridian and photographic observations. Such ephemerides were constructed by the simultaneous numerical integration of the equations of motion for all planets, the Sun, the Moon, 301 largest asteroids, rotations of the Earth and the Moon, including the perturbations from the solar quadrupolar mass moment $`J_2^{}`$ and asteroid ring that lies in the ecliptic plane and consists of the remaining smaller asteroids. In regard to the post-Newtonian dynamics, only the gravitoelectric terms, in the harmonic gauge, were included (Newhall et al. 1983).
### 3.3 The measured extra-precessions of the planetary perihelia and the Lense-Thirring effect
As a preliminary outlook on the measurability of the Lense-Thirring perihelion precessions, let us make the following considerations. The magnitude of the gravitomagnetic shift of the Mercury’s perihelion over a 90-years time span like that covered by the EPM2004 data amounts to 0.0018 <sup>′′</sup>. The accuracy in determining the secular motion of Mercury’s perihelion can be inferred from the results for the components of the eccentricity vector $`k=e\mathrm{cos}\varpi `$ and $`h=e\mathrm{sin}\varpi `$ reported in Table 4 by Pitjeva (2005b). Indeed, the formal standard deviations of $`k`$ and $`h`$ are 0.123 and 0.099 milliarcseconds, respectively. Thus, the formal error in measuring $`\varpi `$ is about 0.0007 <sup>′′</sup>. An analogous calculation for the Earth yields an error in $`\varpi `$ of $`8\times 10^5`$ <sup>′′</sup>.
The EPM2004 ephemerides were used to determine corrections $`\mathrm{\Delta }\dot{\varpi }_{\mathrm{obs}}`$ to the secular precessions of the longitudes of perihelia of the inner planets as fitted parameters of a particular solution. In Table 3 by Pitjeva (2005a), part of which is reproduced in Table 2, it is possible to find their values obtained by comparing the model observations computed using the constructed ephemerides with actual observations. Note that in determining such extra-precessions the PPN parameters (Will 1993) $`\gamma `$ and $`\beta `$ and the solar even zonal harmonic coefficient $`J_2^{}`$ were not fitted; they were held fixed to their GTR values, i.e. $`\gamma =\beta =1`$, and to $`J_2^{}=2\times 10^7`$. Note also that the unit values of $`\beta `$ and $`\gamma `$ were measured in a variety of approaches which are independent of the gravitomagentic force itself.
Although the original purpose<sup>4</sup><sup>4</sup>4The goal by Pitjeva (2005a) was to make a test of the quality of the previously obtained general solution in which certain values of $`\beta ,\gamma `$ and $`J_2`$, were obtained. If the construction of the ephemerides was satisfactory, very small “residual” effects due to such parameters should have been found. She writes: “At present, as a test, we can determine \[…\] the corrections to the motions of the planetary perihelia, which allows us to judge whether the values of $`\beta ,\gamma ,`$ and $`J_2`$ used to construct the ephemerides are valid.”. The smallness of the extra-perihelion precessions found in her particular test-solution is interpreted by Pitjeva as follows: “Table 3 shows that the parameters $`\beta =1,\gamma =1,`$ and $`J_2=2\times 10^7`$ used to construct the EPM2004 ephemerides are in excellent agreement with the observations.” of the determination of such corrections was not the measurement of the LT effect, the results of Table 3 by Pitjeva (2005a) can be used to take first steps towards an observational corroboration of the existence of the solar gravitomagnetic force. Indeed, the uncertainties in the predicted values of the LT precessions induced by the error in $`L_{}`$ (Pijpers 1998; 2003) amount to $`1\times 10^5`$ <sup>′′</sup> cy<sup>-1</sup> for Mercury, $`7\times 10^7`$ <sup>′′</sup> cy<sup>-1</sup> for the Earth and $`2\times 10^7`$ <sup>′′</sup> cy<sup>-1</sup> for Mars: they are far smaller than the errors in Table 2, so that a genuine comparison with the measured precessions make sense.
By comparing Table 1 and Table 2 of this paper it turns out that the predictions of GTR for the LT effect are compatible with the small determined corrections to the secular motions of the planetary perihelia for<sup>5</sup><sup>5</sup>5In the case of Venus the discrepancy between the predicted and the measured values is slightly larger than the measurement error. For such a planet the perihelion is not a good observable because of the small eccentricity of its orbit ($`e_{\mathrm{Venus}}=0.0066`$). Mercury ($`0.0086`$<sup>′′</sup> cy<sup>-1</sup>$`<0.0020`$<sup>′′</sup> cy<sup>-1</sup>$`<0.0014`$ <sup>′′</sup> cy<sup>-1</sup>), the Earth ($`0.0006`$<sup>′′</sup> cy<sup>-1</sup>$`<0.0001`$<sup>′′</sup> cy<sup>-1</sup>$`<0.0002`$ <sup>′′</sup> cy<sup>-1</sup>) and Mars (-0.0004 <sup>′′</sup> cy<sup>-1</sup>$`<`$ -$`3\times 10^5`$ <sup>′′</sup> cy<sup>-1</sup>$`<`$ 0.0006 <sup>′′</sup> cy<sup>-1</sup>). In normalized units $`\mu `$ ($`\mu _{\mathrm{GTR}}=1`$) we have $`\mu _{\mathrm{obs}}^{\mathrm{Mercury}}=1.8\pm 2.5`$, $`\mu _{\mathrm{obs}}^{\mathrm{Earth}}=2\pm 4`$ and $`\mu _{\mathrm{obs}}^{\mathrm{Mars}}=3.3\pm 16`$. Figure 1 summarizes the obtained results.
The discrepancies between the predicted and the determined values are reported in Table 3 of this paper. They are smaller than the measurement uncertainties, so that a $`\chi ^2=\left(\frac{PD}{E}\right)^2=0.2`$ can be obtained. It must be noted that the determined extra-precessions of Table 2 are also compatible with zero, but at a worse level. Indeed, $`\chi ^2=0.8`$ in this case.
A way to improve the robustness and reliability of such a test would be to vary the adopted values for the solar oblateness within the currently accepted ranges and investigate the changes in the fitted values of the extra-precessions. Moreover, it would also be important to produce an analogous set of solutions with $`\beta ,\gamma `$ and $`J_2^{}`$ fixed in which the extra-precessions of the nodes are determined; in this way it would be possible to use only Mercury.
### 3.4 Some possible systematic errors due to other competing effects
In order to check our conclusion that the LT effect is the main responsible for the observed secular corrections to the planetary perihelia $`\mathrm{\Delta }\dot{\varpi }_{\mathrm{obs}}`$ let us focus on Mercury and on the known perturbations which could induce a secular extra-perihelion advance due to their mismodelling.
The major sources of secular advances of the perihelia are the Schwarzschild gravitoelectric part of the solar gravitational field and the quadrupolar mass moment $`J_2^{}`$ of the Sun. Their nominal effects on the longitudes of perihelion of the inner planets are quoted in Table 4 and Table 5 of this paper; the analytical expressions are
$`\dot{\varpi }_{\mathrm{GE}}`$ $`=`$ $`{\displaystyle \frac{3nGM}{c^2a(1e^2)}},`$ (4)
$`\dot{\varpi }_{J_2}`$ $`=`$ $`{\displaystyle \frac{3}{2}}{\displaystyle \frac{nJ_2}{\left(1e^2\right)^2}}\left({\displaystyle \frac{R}{a}}\right)^2\left(1{\displaystyle \frac{3}{2}}\mathrm{sin}^2i\right)\dot{\varpi }_{.2}J_2,`$ (5)
where $`n=\sqrt{GM/a^3}`$ is the Keplerian mean motion and $`R`$ is the mean equatorial radius of the central body.
In view of their large size with respect to the LT effect, one could legitimately ask if the determined extra-precessions are due to the systematic errors in such competing secular rates. An a-priori analytical analysis shows that it should not be the case.
#### 3.4.1 The impact of the solar oblateness
In regard to $`J_2^{}`$, which is still rather poorly known, only values measured in such a way that no a priori ‘imprinting’ effects occurred should be considered for our purposes.
E.g., the most recent determinations of the solar oblateness based on astrophysical techniques yield values close to $`2.2\times 10^7`$ (Paternò et al. 1996; Pijpers 1998; Mecheri et al. 2004) with discrepancies between the various best estimates of the same order of magnitude of their errors, i.e. $`10^9`$. Let us see if such determinations are compatible with the determined extra-advances of perihelia. By assuming a correction of $`10\%`$ of the adopted reference value by Pitjeva, i,e. $`\delta J_2^{}=2\times 10^8`$, the resulting residual precession due to the solar oblateness would amount to $`+0.0025`$<sup>′′</sup> cy<sup>-1</sup> for Mercury. It falls outside the measured range.
A way to a priori cancel out any possible impact of the uncertainty in the solar oblateness consists in suitably combining the perihelia advances of two planets so to de-correlate by construction the LT effect and the precessions due to $`J_2`$. This approach allows to extract the gravitomagnetic signal independently of the solar quadrupolar mass moment. On the other hand, it is also possible to measure a correction $`\delta J_2^{}`$ to it independently of the LT effect. In turn, such value for $`\delta J_2^{}`$ can be used to check if it can accommodate the determined extra-precessions of Table 2. Let us write
$$\{\begin{array}{ccc}\mathrm{\Delta }\dot{\varpi }_{\mathrm{obs}}^{\mathrm{Mercury}}=\dot{\varpi }_{.2}^{\mathrm{Mercury}}\delta J_2^{}+\dot{\varpi }_{\mathrm{LT}}^{\mathrm{Mercury}}\mu ,\hfill & & \\ & & \\ \mathrm{\Delta }\dot{\varpi }_{\mathrm{obs}}^{\mathrm{Earth}}=\dot{\varpi }_{.2}^{\mathrm{Earth}}\delta J_2^{}+\dot{\varpi }_{\mathrm{LT}}^{\mathrm{Earth}}\mu .\hfill & & \end{array}$$
(6)
By solving eq. (6) with respect to $`\mu `$ it is possible to obtain
$$\mu =\frac{\mathrm{\Delta }\dot{\varpi }_{\mathrm{obs}}^{\mathrm{Mercury}}+c_1\mathrm{\Delta }\dot{\varpi }_{\mathrm{obs}}^{\mathrm{Earth}}}{\dot{\varpi }_{\mathrm{LT}}^{\mathrm{Mercury}}+c_1\dot{\varpi }_{\mathrm{LT}}^{\mathrm{Earth}}},$$
(7)
with
$$c_1=\frac{\dot{\varpi }_{.2}^{\mathrm{Mercury}}}{\dot{\varpi }_{.2}^{\mathrm{Earth}}}\left(\frac{a^{\mathrm{Earth}}}{a^{\mathrm{Mercury}}}\right)^{7/2}\left(\frac{1e_{\mathrm{Earth}}^2}{1e_{\mathrm{Mercury}}^2}\right)^2=30.1930,$$
(8)
and
$$\dot{\varpi }_{\mathrm{LT}}^{\mathrm{Mercury}}+c_1\dot{\varpi }_{\mathrm{LT}}^{\mathrm{Earth}}=0.0013^{\prime \prime }\mathrm{cy}^1.$$
(9)
The combination of eq. (7) is not affected by the solar oblateness whatever its real value is: indeed,
$$\dot{\varpi }_{.2}^{\mathrm{Mercury}}+c_1\dot{\varpi }_{.2}^{\mathrm{Earth}}=0,J_2^{}.$$
(10)
By inserting the value of eq. (8), the figures of Table 1 and the results of Table 2 in eq. (7) one obtains for such a combination<sup>6</sup><sup>6</sup>6The error has been evaluated by summing in quadrature the errors of Table 2 according to eq. (7). $`\mu _{\mathrm{obs}}=1.8\pm 10`$. As can be noted, the best estimate for $`\mu `$ does not change with respect to the case of Mercury’s perihelion only, as if departures of the solar oblateness from the adopted reference value were of little importance. Indeed, if we solve eq. (6) with respect to the correction to the Sun’s quadrupolar mass moment the equation
$$\delta J_2^{}=\frac{\mathrm{\Delta }\dot{\varpi }_{\mathrm{obs}}^{\mathrm{Mercury}}+d_1\mathrm{\Delta }\dot{\varpi }_{\mathrm{obs}}^{\mathrm{Earth}}}{\dot{\varpi }_{.2}^{\mathrm{Mercury}}+d_1\dot{\varpi }_{.2}^{\mathrm{Earth}}},$$
(11)
with
$$d_1=\frac{\dot{\varpi }_{\mathrm{LT}}^{\mathrm{Mercury}}}{\dot{\varpi }_{\mathrm{LT}}^{\mathrm{Earth}}}\left(\frac{a_{\mathrm{Earth}}}{a_{\mathrm{Mercury}}}\right)^3\left(\frac{1e_{\mathrm{Earth}}^2}{1e_{\mathrm{Mercury}}^2}\right)^{3/2}=18.3864,$$
(12)
is obtained. Eq. (11) allows to measure the correction to the adopted value of $`J_2^{}`$, by construction, independently of the LT effect in the sense that
$$\dot{\varpi }_{\mathrm{LT}}^{\mathrm{Mercury}}+d_1\dot{\varpi }_{\mathrm{LT}}^{\mathrm{Earth}}=0,L_{}.$$
(13)
The result is<sup>7</sup><sup>7</sup>7Note that Pitjeva (2005a) derived a negative correction $`\delta J_2^{}`$ of order $`10^8`$ from the determined extra-advance of Mercury’s perihelion only, without taking into account the biasing impact of the LT effect which amounts to $`8\%`$ for Mercury, as can be inferred from Table 1 and Table 5.
$$\delta J_2^{}=(+0.01\pm 0.47)\times 10^7.$$
(14)
Such value can be considered as a dynamical measurement of the solar oblateness independent of the general relativistic gravitomagnetic features of motion. It induces a “residual” precession of $`+0.0001`$ <sup>′′</sup> cy<sup>-1</sup> on Mercury’s perihelion, which is smaller than its observed extra-advance and the related error. For the Earth the “residual” effect of eq. (14) would amount to $`+4\times 10^6`$ <sup>′′</sup> cy<sup>-1</sup>.
It is important to note that the combination of eq. (7) yields $`\chi ^2=0.007`$, while, by assuming zero extra-precessions, one has $`\chi ^2=0.03`$: also in this case, the relativistic prediction of LT is in better agreement with data than the zero-effect hypothesis.
#### 3.4.2 The post-Newtonian gravitoelectric precessions
Although the large nominal values of their precessions, the post-Newtonian gravitoelectric terms do not represent a problem. Indeed, they are fully included in the dynamical force models of EPM2004 in terms of the PPN parameters $`\beta `$ and $`\gamma `$ which are presently known at a $`10^410^5`$ level (Pitjeva 2005a; Bertotti et al. 2003). Moreover, theoretical deviations from the GTR values are expected at a $`10^610^7`$ level (Damour and Nordtvedt 1993).
#### 3.4.3 The impact of the asteroids
As already noted, the dynamical force models adopted in EPM2004 also include the action of the major asteroids and of the ecliptic ring which accounts for the other minor bodies. Indeed, it has recently pointed out that their impact limits the accuracy of the inner planets’ ephemerides over time-scales of a few decades (Standish and Fienga 2002) in view of the relatively high uncertainty in their masses (Krasinsky et al. 2002; Pitjeva 2005b) Recently, Fienga and Simon (2005) have shown that also Mercury’s orbit is affected to a detectable level by secular perturbations due to the most important asteroids.
May it happen that the mismodelled part of such secular precessions could explain the observed $`\mathrm{\Delta }\dot{\varpi }_{\mathrm{obs}}^{\mathrm{Mercury}}`$?
From Table 3 by Fienga and Simon (2005) the nominal amplitude of the secular perturbations on $`\varpi ^{\mathrm{Mercury}}`$ due to 295 major asteroids can be calculated. It turns out to be $`0.0004`$ <sup>′′</sup> cy<sup>-1</sup>; even assuming a conservative $`10\%`$ uncertainty (Pitjeva 2005b), it is clear that the asteroids are not the cause of the determined extra-shift of Mercury’s perihelion.
#### 3.4.4 The impact of non-Einsteinian effects
In regard to other possible sources of extra-secular precessions of the planetary perihelia outside the scheme of the Newton-Einstein gravity, recently it has been shown by Lue and Starkman (2003) that the multidimensional braneworld gravity model by Dvali, Gabadadze and Porrati (2000) predicts also a secular perihelion shift in addition to certain cosmological features. By postulating that the current cosmic acceleration is entirely caused by the late-time self-acceleration, constraints from Type 1A Supernovæ data yield a value of $`0.0005`$ <sup>′′</sup> cy<sup>-1</sup> for the Lue-Starkman planetary precessions. Also this effect is too small to accommodate the determined additional perihelion advance of Mercury.
### 3.5 Analysis of other independent data
The shift of the perihelion yields a variation of the planet’s range that can be expressed as (Nordtvedt 2000) $`\mathrm{\Delta }r=ea\mathrm{\Delta }\varpi `$. In the case of Mercury the centennial variation due to the Lense-Thirring precession amounts to -115 m.
Fienga et al. (2005) used their numerical ephemerides<sup>8</sup><sup>8</sup>8They include a complete suit of improved dynamical models, apart from just the post-Newtonian gravitomagnetic forces whose effects are, thus, fully present in the determined residuals. INPOP, recently produced at the Institute of mécanique céleste et de calcul des éphémérides (IMCCE), to fit different kinds of observations including also the radar-ranging data to the inner planets. They found for the Mercury’s range residual the value $`\delta r_{\mathrm{meas}}=95.6\pm 784`$ m. It is worth noting that in obtaining these results also the impact of the asteroids on Mercury’s orbital motion was accounted for. As it can be noted, also in this case the errors are large, but the general relativistic prediction for the Lense-Thirring effect are in better agreement with the data than the hypothesis of null effect (\[$`(\mathrm{P}\mathrm{D})/\mathrm{E}]^2=6\times 10^4`$ and \[$`(\mathrm{P}\mathrm{D})/\mathrm{E}]^2=1\times 10^2`$, respectively).
## 4 Constraints on a Yukawa-like fifth force
The differences between the determined extra-precessions and the predicted LT rates of Table 3 of this paper can also be used to strongly constrain, at planetary length-scales $`10^{10}10^{11}`$ m, departures from the inverse-square-law phenomenologically parameterized in terms of the magnitude $`|\alpha |`$ of the strength of a Yukawa-like fifth force (Fischbach et al. 1986; Adelberger et al. 2003). Indeed, a potential
$$U_{\mathrm{Yukawa}}=\frac{GM}{r}\left[1+\alpha \mathrm{exp}\left(\frac{r}{\lambda }\right)\right],$$
(15)
where $`\lambda `$ is the range of such a hypothesized force, can produce a secular perihelion advance over scales $`\lambda `$ comparable to $`a`$ (Lucchesi 2003)
$$\dot{\varpi }_{\mathrm{Yukawa}}\frac{\alpha n}{e}.$$
(16)
By using the figures of Table 3 it is possible to constrain $`\alpha `$ to $`10^{12}10^{13}`$ level at $`r=\lambda 1`$ A.U. The most recently published constraints in the planetary range are at $`10^910^{10}`$ level (Bertolami and Paramos 2005; Reynaud and Jaekel 2005).
## 5 Discussion and conclusions
In this paper we discussed the possibility of performing new tests of post-Newtonian gravity in the Solar System. To this aim, we analyzed the estimated corrections to the secular rates of the perihelia of the inner planets of the Solar System recently determined by E.V. Pitjeva (Institute of Applied Astronomy, Russian Academy of Sciences). She used the EPM2004 ephemerides with a wide range of observational data spanning almost one century; in a particular solution, she solved also for the secular motions of the perihelia by keeping fixed the PPN parameters $`\beta `$ and $`\gamma `$ and the solar quadrupole mass moment $`J_2^{}`$, and neglecting the gravitomagnetic force in the dynamical force models.
It turns out that the post-Newtonian LT secular precessions predicted by GTR are compatible with the determined extra-precessions for Mercury, the Earth and, to a lesser extent, Mars: in normalized units ($`\mu =1`$ in GTR) we have $`\mu _{\mathrm{obs}}=1.8\pm 2.5`$ for Mercury, $`\mu _{\mathrm{obs}}=2\pm 4`$ for the Earth and $`\mu _{\mathrm{obs}}=3.3\pm 16.6`$ for Mars. A suitable combination of the perihelia of Mercury and the Earth, which cancels out any possible bias by $`J_2^{}`$, yields $`\mu _{\mathrm{obs}}=1.8\pm 10`$. It must be noted that the errors are still large and the data are compatible also with the hypothesis of zero extra-precessions, but at a worse level with respect to the relativistic LT prediction. If confirmed by further, more extensive and robust data analysis by determining, e.g., the extra-precessions of the nodes as well, it would be the first observational evidence of the solar gravitomagnetic field. The processing of further amounts of data, along with those expected in future from the forthcoming planetary mission BepiColombo and, perhaps, Messenger and Venus Express as well, although to a lesser extent, will further improve the accuracy in determining the orbital motion of these planets and, consequently, the precision of the LT tests.
A by-product of the present analysis is represented by new, strong constraints ($`10^{12}10^{13}`$) on the strength of a Yukawa-like fifth force at scales of about one Astronomical Unit.
## Acknowledgements
I thank E.V. Pitjeva for helpful clarifications about her measured extra-precessions, J.-F. Pascual-S$`\stackrel{´}{\mathrm{a}}`$nchez, O. Bertolami and G. Melki for useful comments and references. I am grateful to the anonymous referees whose comments greatly improved the paper. |
warning/0507/hep-th0507220.html | ar5iv | text | # SUM RULES FOR THE DIRAC SPECTRUM OF THE SCHWINGER MODEL
## I Introduction
The Schwinger model Schwinger , electrodynamics in $`1+1`$ dimension is one of the few quantum field theories that can be solved analytically Lowenstein:1971fc ; Jayewardena:1988td ; SW . Since the model contains nonperturbative features that are also found in QCD, such as a nonzero chiral condensate and instantons, it has added significantly to our understanding of gauge theories in general. However, most interesting quantum field theories cannot be solved analytically, and one has to rely on approximate techniques to investigate them. For nonperturbative phenomena, two approaches which both have been tested in the Schwinger model, have been very successful: lattice simulations and chiral perturbation theory. In particular, a large number of lattice studies have benefited from the Schwinger model lat-kogut ; lat-az ; lat-dil ; lat-neu ; lat-gatt ; lat-far ; lat-shai ; lat-hoel ; lat-kis ; lat-biet ; biet-dil ; durr-prd71 ; lat-durr ; lat-poul . Among others, we mention the study of chiral fermions on the lattice lat-neu ; lat-far ; lat-gatt ; lat-biet , the analysis of the continuum limit for staggered fermions lat-durr , and the effect of quenching on the Dirac spectrum lat-durrsh ; lat-poul .
For theories with a mass gap and a nonzero chiral condensate the mass dependence of the partition function can be obtained from a chiral Lagrangian. In particular, in the domain where the Compton wavelength of the Goldstone modes is much larger than the size of the box only the constant fields have to be included in the partition function so that its mass dependence simply follows from the mass term of the chiral Lagrangian LS . Based on this result, it was shown in LS that the Dirac spectrum has to satisfy consistency relations in the form of sum rules for inverse eigenvalues. These relations are not sufficient to determine the Dirac spectrum. However, it turned out that the complete low-lying Dirac spectrum can be obtained from a modified chiral Lagrangian Vplb ; OTV ; DOTV , or equivalently, from chiral random matrix theory SV ; V . In the Schwinger model, and also for QCD with $`N_f=1`$, there are no low-lying excitations and the only contribution to the low-energy limit of the partition function is the vacuum energy. Starting from this result, it is possible to derive sum rules for the inverse Dirac eigenvalues of the Schwinger model LS ; smvac . However, to obtain a deeper understanding of these sum rules in the context of quantum field theory, it of is interest to derive them directly from the microscopic field theory. This was achieved in smvac for the simplest sum rule in the sector of zero topological charge.
In this paper we give a microscopic derivation of the simplest sum rule in the sector of arbitrary topological charge $`\nu `$. This requires a detailed understanding of scalar correlation functions in the Schwinger model which were derived before both in the plane Manias:1989bu and on the torus SW ; us ; steele-VZ ; steele ; Azakov:2001pz . It turns out that the clustering property of the scalar correlation functions alone is sufficient to derive the (gauge field averaged) sum rules. Below we will give a derivation based on this property. A third way to obtain Leutwyler-Smilga sum rules is to start from the bosonized form of the massive Schwinger model. In the microscopic derivation of sum rules we first derive a more informative quantity – the same sum rules for an arbitrary fixed external gauge field (which was also obtained for $`\nu =0`$ in smvac ). The universal Leutwyler-Smilga sum rules then follow after averaging over the gauge fields.
Analytical results obtained for the Schwinger model have been tested elaborately by means of Monte-Carlo simulations. The numerical value of the chiral condensate has been reproduced accurately lat-dil ; lat-durr ; lat-poul . The index theorem was confirmed and the fermionic would-be zero modes were identified lat-gatt ; lat-far . Leutwyler-Smilga sum rules have been obtained both for zero and nonzero topological charge lat-far . The distributions of the smallest eigenvalues of the Dirac operator for the Schwinger model lat-dil ; lat-far ; lat-durr ; lat-poul agree with analytical results obtained by means of chiral Lagrangians LS ; smvac and random matrix theory SV ; V ; indiv .
A theory that is closely related to the Schwinger model are random Dirac fermions where the electromagnetic interaction is replaced by a random gauge field interaction. Starting with the work of Gade and Wegner gade the behavior of the spectral density near zero has attracted a great deal of attention in particular with applications to the quantum Hall effect in mind ludwig ; simons . An important difference with the Schwinger model is that in the condensed matter literature the physically interesting case is the quenched model, sometimes defined as the limit of zero flavors. The consensus is that the density of states diverges for strong disorder as is the case for the quenched Schwinger model doel ; kenway ; smvac ; lat-poul . For weak disorder, on the other hand, the spectral density vanishes. We will discuss the Dirac spectrum of this model and relate it to the chiral condensate and the simplest Leutwyler-Smilga sum rule.
The organization of this paper is as follows. In an introductory chapter we introduce the Leutwyler-Smilga sum rules and the Schwinger model and discuss its main properties. A combinatorial derivation of sum rules based on the clustering property of scalar correlators is given in section III. Section IV contains a microscopic derivation of the simplest sum rule, and a derivation from the bosonized massive Schwinger model is given in section V. It is explained in section VI that modified sum rules for two or more flavors can be related to modified clustering properties of the scalar correlators. A random matrix derivation of sum rules is given in section VII. In this section we also study the fluctuations of the fermion determinant. In the first part of section VIII we discuss the effect of topology on the Dirac spectrum and the mechanism of the formation of a chiral condensate. In the second part of section VIII we discuss the behavior of the Dirac spectrum for random fermions. Concluding remarks are made in section IX.
## II General Framework
### II.1 Sum Rules
In this paper we study the eigenvalues of the Dirac operator by means of sum rules for its inverse eigenvalues introduced in LS . The eigenvalues of the anti-hermitian Dirac operator are defined by
$`D\varphi _k=i\lambda _k\varphi _k.`$ (1)
In sector with topological charge $`\nu `$ the Dirac operator has exactly $`\nu `$ zero eigenvalues that are not paired. All other eigenvalues occurs in pairs $`\pm \lambda _k`$. In the sector of topological charge $`\nu `$ the partition function is defined by
$`Z^\nu (m)=m^\nu {\displaystyle 𝑑A_\mu ^\nu \stackrel{}{det}[iD]e^{S(A_\mu ^\nu )}},`$ (2)
where the integral is over gauge fields in the sector of topological charge $`\nu `$ weighted by the gauge field action $`S(A_\mu )`$. As indicated by the prime,
$`\stackrel{}{det}[iD]={\displaystyle \underset{\lambda _n0}{}}\lambda _n^2,`$ (3)
the zero eigenvalues are not included in the determinant resulting in the pre-factor $`m^\nu `$. At fixed $`\theta `$ angle the partition function is given by
$`Z(m,\theta )={\displaystyle \underset{\nu =\mathrm{}}{\overset{\mathrm{}}{}}}e^{i\nu \theta }Z^\nu (m).`$ (4)
Below we will use two types of averages of an operator $`O`$ denoted by single and double brackets. The first type is defined by (notice that $`Z^0(m=0)=Z(m=0,\theta )`$)
$`O^\nu ={\displaystyle \frac{1}{Z^0(m=0)}}{\displaystyle DA_\mu ^\nu O[A_\mu ^\nu ]\stackrel{}{det}(iD)e^{S(A_\mu ^\nu )}},`$ (5)
and the second type by LS
$`O^\nu ={\displaystyle \frac{m^\nu }{Z^\nu (m)}}|_{m=0}{\displaystyle DA_\mu ^\nu O(A_\mu ^\nu )\stackrel{}{det}(iD)e^{S(A_\mu ^\nu )}}.`$ (6)
To obtain sum rules for the inverse Dirac eigenvalues in a sector of topological charge $`\nu `$ we expand the fermion determinant in powers of $`m`$,
$`\stackrel{}{det}[i\overline{)}D+m]=\stackrel{}{det}[i\overline{)}D]{\displaystyle \underset{n0}{}}\left(1+m^2{\displaystyle \underset{n>0}{}}{\displaystyle \frac{1}{\lambda _n^2}}+m^4{\displaystyle \underset{\genfrac{}{}{0pt}{}{n_i>0}{n_1n_2}}{}}{\displaystyle \frac{1}{\lambda _{n_1}^2\lambda _{n_2}^2}}+\mathrm{}\right).`$ (7)
For the small mass expansion of the partition function we thus obtain
$`{\displaystyle \frac{Z^\nu (m)}{Z^0(m=0)}}`$ $`=`$ $`m^\nu \stackrel{}{det}[iD]^\nu +m^{\nu +2}{\displaystyle \underset{n>0}{}}{\displaystyle \frac{1}{\lambda _n^2}}^\nu +m^{\nu +4}{\displaystyle \underset{\genfrac{}{}{0pt}{}{n_i>0}{n_1n_2}}{}}{\displaystyle \frac{1}{\lambda _{n_1}^2\lambda _{n_2}^2}}^\nu +\mathrm{}`$ (8)
$`=`$ $`m^\nu \stackrel{}{det}[iD]^\nu \left[1+m^2{\displaystyle \underset{n>0}{}}{\displaystyle \frac{1}{\lambda _n^2}}^\nu +m^4{\displaystyle \underset{\genfrac{}{}{0pt}{}{n_i>0}{n_1n_2}}{}}{\displaystyle \frac{1}{\lambda _{n_1}^2\lambda _{n_2}^2}}^\nu +\mathrm{}\right].`$
where we have used the definition (6) for the average in the sector of topological charge $`\nu `$. Therefore, sum rules for the inverse Dirac eigenvalues follow from the sub-leading terms in the small mass expansion of the partition function in the sector of topological charge $`\nu `$. If
$`{\displaystyle \frac{Z^\nu (m)}{Z^0(m=0)}}=m^\nu (a_0+a_2m^2+\mathrm{}),`$ (9)
the simplest sum rule is given by
$`{\displaystyle \underset{n>0}{}}{\displaystyle \frac{1}{\lambda _n^2}}^\nu ={\displaystyle \frac{a_2}{a_0}}.`$ (10)
In LS general arguments where given that, for a gauge theory interacting with fermions according to the Dirac operator with a nonzero chiral condensate, this ratio is given by
$`{\displaystyle \frac{\mathrm{\Sigma }^2V^2}{4(N_f+\nu )}}.`$ (11)
For $`N_f=1`$ this argument is particularly simple. The $`\theta `$ dependence of the partition function, as can be observed from (2) and (4), is obtained from the replacement $`mm\mathrm{exp}(i\theta )`$. Given that the vacuum energy is equal to $`\mathrm{Re}(mV\mathrm{\Sigma })`$, this results in the large volume partition function LS
$`Z(m,\theta )=e^{mV\mathrm{\Sigma }\mathrm{cos}\theta }.`$ (12)
By inverting (4) we find that the partition function in sector $`\nu `$ is given by LS
$`Z_\nu =I_\nu (mV\mathrm{\Sigma })={\displaystyle \frac{1}{\nu !2^\nu }}(mV\mathrm{\Sigma })^\nu (1+{\displaystyle \frac{(mV\mathrm{\Sigma })^2}{4(|\nu |+1)}}+\mathrm{}).`$ (13)
This results immediately in the sum rule (11). In this paper we will derive this sum rule by a microscopic calculation in the Schwinger model.
### II.2 The Schwinger Model
In this section we give a brief review of the Schwinger Model Schwinger which is massless QED in two dimensions. The Euclidean Lagrangian is defined by
$`={\displaystyle \frac{1}{2}}F_{01}^2\overline{\psi }[i\overline{)}g\overline{)}A]\psi ,`$ (14)
where the electric field strength is given by
$`F_{01}=_0A_1_1A_0,`$ (15)
and
$`i\overline{)}g\overline{)}A=\gamma _\mu (i_\mu gA_\mu ),`$ (16)
with Euclidean gamma matrices defined by
$`\gamma _0=\sigma _1,\gamma _1=\sigma _2,\gamma _5=\sigma _3.`$ (17)
The Lagrangian of this model has a chiral symmetry which is broken by the $`U(1)`$ axial anomaly. As a consequence the “photon” becomes massive with mass given by $`\mu =g/\sqrt{\pi }`$. In this model, a local external charge is screened by massless fermions. Therefore, the asymptotic states contain no fermions, and the Schwinger boson is the only physical particle. The theory is super-renormalizable with a coupling constant $`g`$ (which has the dimension of mass) that does not run.
The (massless) Schwinger model, being equivalent to a noninteracting gas of bosons, is exactly solvable and was solved many times, by different techniques (operator language, path integral, bosonization) and on different manifolds Lowenstein:1971fc ; Casher:1974vf ; Danilov:1980ez ; Manton:1985jm ; Hetrick:1988yg ; Jayewardena:1988td ; SW ; Joos:1990km ; Roskies . Also, many authors considered more specific features of the model, such as certain correlation functions at zero and at finite temperature bardakci ; Rothe:1978hx ; Manias:1989bu ; us ; steele-VZ ; steele ; Grignani:1995cw ; Durr:1996im ; Azakov:2001pz , the fermion determinant, zero modes, the index theorem and instantons Nielsen:1976hs ; Nielsen:1977aw ; Ansourian:1977qe ; Patrascioiu:1979xj ; Hortacsu:1979fg ; Hortacsu:1980kv ; Seiler:1980yx ; SW ; Fry:1992qz ; Smilga:1993sn ; Adam:1993fc , the multi-flavor Schwinger model weis ; shrock ; Gatt-multi ; shi-smil-multi ; SmV ; Hos-multi .
The vector potential $`A_\mu `$ can be decomposed as
$`A_\mu =ϵ_{\mu \nu }^\nu \varphi +_\mu \lambda ,`$ (18)
with the last term being a pure gauge. The topological charge, $`\nu `$, of the gauge fields is equal to the difference of the number of right handed and left handed zero modes of the Dirac operator and is therefore necessarily quantized. It is given by
$`\nu `$ $`=`$ $`{\displaystyle \frac{g}{4\pi }}{\displaystyle d^2xϵ_{\mu \nu }F_{\mu \nu }}={\displaystyle \frac{g}{2\pi }}{\displaystyle d^2x_\alpha ^2\varphi }.`$ (19)
By Stokes theorem the topological charge is determined by the asymptotic behavior of $`\varphi `$. One easily shows that the large distance asymptotics,
$`\varphi (x){\displaystyle \frac{\nu }{2g}}\mathrm{log}x^2+\mathrm{const}.,|x|\mathrm{},`$ (20)
results in a topological charge equal to $`\nu `$. The boundary conditions (20) imply that the plane is compactified at infinity, for example by stereographic projection to a sphere with radius $`R`$ Nielsen:1977aw ; Ansourian:1977qe ; Hortacsu:1979fg .
Because of a particular property of the 2d Dirac algebra,
$`\gamma _\mu \gamma _5=iϵ_{\mu \nu }\gamma _\nu ,`$ (21)
the Dirac operator can be written in in the form weis
$`i\overline{)}D_\varphi =e^{g\varphi \gamma _5}i\overline{)}e^{g\varphi \gamma _5}.`$ (22)
Using this representation one easily finds the following explicit expressions for the zero modes in the sector of topological charge $`\nu `$:
$`\psi _p(x)={\displaystyle \frac{1}{\sqrt{2\pi }}}(x^+)^pe^{g\varphi (x)}\left(\begin{array}{c}1\\ 0\end{array}\right),p=0\mathrm{}\nu 1,x^\pm =x_0\pm ix_1.`$ (25)
Note that these zero modes are not normalized. From the asymptotic behavior (20), it is easy to see that for $`p=0,\mathrm{},\nu 2`$ the zero modes are normalizable. However, the zero mode for $`p=\nu 1`$ is not normalizable on the plane (its norm diverges logarithmically). It has to be included nevertheless because it is normalizable on the compactified plane as required by the index theorem Nielsen:1977aw ; Ansourian:1977qe ; Hortacsu:1979fg ; bardakci ; Jackiw:1984ji . The pre-factor $`1/\sqrt{2\pi }`$ in the zero modes also follows from matching with compact manifolds Adam:1993fc .
The unpaired zero modes explicitly break the $`U_A(1)`$ symmetry of the partition function. In the massless limit the chiral condensate $`\overline{\psi }\psi `$ comes from the zero modes in the sectors $`\nu =\pm 1`$. It has a particularly simple form on the plane Lowenstein:1971fc ; SW ; smcon ; Adam:1993fc ,
$`\mathrm{\Sigma }\overline{\psi }\psi ={\displaystyle \frac{\mu }{2\pi }}e^\gamma ,`$ (26)
where $`\gamma `$ is the Euler constant. As a consequence of the Banks-Casher formula BC the average density of the low-lying Dirac eigenvalues is given by $`\mathrm{\Sigma }V/\pi `$. This also implies that if we take the thermodynamic limit before the chiral limit, we find a nonzero value for the condensate from the nonzero modes durr-prd71 .
### II.3 Effective Action
Using the action (14) and the decomposition (18), the partition function in the sector of topological charge $`\nu `$ is given by
$`Z^\nu (m)`$ $`=`$ $`{\displaystyle D\varphi ^\nu e^{S(\varphi ^\nu )}D\overline{\psi }D\psi e^{{\scriptscriptstyle d^2x\overline{\psi }[i\overline{)}D+m]\psi }}}`$ (27)
$`=`$ $`{\displaystyle D\varphi ^\nu e^{S(\varphi ^\nu )}det[iD+m]}`$
$`=`$ $`{\displaystyle D\varphi ^\nu e^{S(\varphi ^\nu )}m^\nu \underset{n>0}{}\lambda _n^2}.`$
Here we have used the chiral symmetry of the nonzero Dirac eigenvalues. The infinite product over the eigenvalues has to be regularized in the UV. This procedure is well known Fujikawa , and results in an anomalous contribution to the effective action. The result is Hortacsu:1979fg ; Roskies ; SW
$`{\displaystyle \underset{n>0}{}}\lambda _n^2=𝒞det𝒩\mathrm{exp}\left({\displaystyle \frac{\mu ^2}{2}}{\displaystyle d^2x\varphi (x)\mathrm{\Delta }\varphi (x)}\right),`$ (28)
where $`𝒞`$ is an (infinite) normalization constant which drops out from all final results and $`𝒩`$ is the norm matrix of the (unnormalized) zero modes,
$`𝒩_{pq}={\displaystyle d^2x\psi _p^{}(x)\psi _q(x)}.`$ (29)
The bosonic partition function in the sector of topological charge $`\nu `$ is thus given by (to the lowest order in $`m`$)
$`Z^\nu (m)=m^\nu {\displaystyle D\varphi ^\nu det𝒩e^{\mathrm{\Gamma }[\varphi ^\nu ]}},`$ (30)
with the effective action $`\mathrm{\Gamma }`$ equal to
$`\mathrm{\Gamma }(\varphi )={\displaystyle \frac{1}{2}}{\displaystyle d^2x\varphi (x)[\mathrm{\Delta }^2\mu ^2\mathrm{\Delta }]\varphi (x)}.`$ (31)
This effective action $`\mathrm{\Gamma }`$ defines the propagator
$`{\displaystyle \frac{1}{p^4+p^2\mu ^2}}={\displaystyle \frac{1}{\mu ^2}}\left({\displaystyle \frac{1}{p^2}}{\displaystyle \frac{1}{p^2+\mu ^2}}\right).`$ (32)
In the coordinate space,
$`𝒢(x)={\displaystyle \frac{1}{4g^2}}\mathrm{log}x^2{\displaystyle \frac{1}{2g^2}}K_0(\mu |x|)+\mathrm{const}.,`$ (33)
where $`K_0(\mu |x|)`$ is a modified Bessel function of the second kind, which is exponentially small at large distances. The large distance behavior of this propagator is therefore determined by the first term which is a massless propagator
$`𝒢(x)={\displaystyle \frac{1}{4g^2}}\mathrm{log}x^2+\mathrm{const}..`$ (34)
The arbitrary constant in the equation (33) can only be obtained after a suitable infrared regularization of the theory. For example, this constant can be determined by matching with the result on compact manifolds Smilga:1993sn ; Adam:1993fc .
### II.4 Partition Function in the Sector of Topological Charge $`\nu `$
To evaluate the partition function in the sector of topological charge $`\nu `$ (which we take positive for notational convenience; of course, the partition function does not depend on the sign of the topological charge), we first rewrite the norm matrix. Using the explicit expressions for the zero modes we find
$`det𝒩`$ $`={\displaystyle \frac{1}{(2\pi )^\nu }}`$ $`{\displaystyle d^2x_1\mathrm{}d^2x_\nu \frac{1}{\nu !}\underset{\sigma \pi }{}\mathrm{sg}(\sigma \pi )x_1^{\sigma (0)}x_1^{\pi (0)}\mathrm{}x_\nu ^{\sigma (\nu 1)}x_\nu ^{\pi (\nu 1)}e^{2g_{q=1}^\nu \varphi (x_q)}}.`$ (35)
The sums over permutations $`\sigma `$ and $`\pi `$ can be rewritten as the product of two determinants,
$`det𝒩`$ $`=`$ $`{\displaystyle \frac{1}{\nu !(2\pi )^\nu }}{\displaystyle d^2x_1\mathrm{}d^2x_\nu e^{2g_{q=1}^\nu \varphi (x_q)}\left|\begin{array}{cccc}1& x_1& \mathrm{}& x_1^{\nu 1}\\ \mathrm{}& \mathrm{}& & \mathrm{}\\ 1& x_\nu & \mathrm{}& x_{\nu 1}^\nu \end{array}\right|^2}=`$ (39)
$`=`$ $`{\displaystyle \frac{1}{\nu !(2\pi )^\nu }}{\displaystyle d^2x_1\mathrm{}d^2x_\nu e^{2g_{q=1}^\nu \varphi (x_q)}\underset{i>j}{\overset{\nu }{}}|x_ix_j|^2},`$ (40)
where a well-known property of Vandermonde determinant was used. Using this expression in (2), we find for the partition function to the lowest order in $`m`$,
$`Z^\nu (m)={\displaystyle \frac{m^\nu }{\nu !(2\pi )^\nu }}{\displaystyle d^2x_1\mathrm{}d^2x_\nu \underset{i>j}{\overset{\nu }{}}|x_ix_j|^2D\varphi e^{\mathrm{\Gamma }(\varphi )}e^{2g_{q=1}^\nu \varphi (x_q)}}.`$ (41)
Since the effective action is Gaussian the path integral is simply given by
$`{\displaystyle \frac{1}{Z^0(m=0)}}{\displaystyle D\varphi e^{\mathrm{\Gamma }(\varphi )}e^{2g_{q=1}^\nu \varphi (x_q)}}=e^{2\nu g^2𝒢(0)+4g^2_{i>j}^\nu 𝒢(x_ix_j)}.`$ (42)
Using the explicit expression for $`𝒢`$ on the plane (33) we observe that the factor $`_{i>j}^\nu |x_ix_j|^2`$ is canceled by the asymptotic behavior of the Greens function. Thus, in the limit of large volume
$`{\displaystyle \frac{Z^\nu (m)}{Z^0(m=0)}}={\displaystyle \frac{m^\nu }{\nu !}}\left({\displaystyle \frac{e^{2g^2𝒢(0)}V}{2\pi }}\right)^\nu +O(m^{\nu +1}).`$ (43)
The chiral condensate can be expressed as
$`\overline{\psi }\psi ={\displaystyle \frac{1}{Z^0(m=0)}}{\displaystyle \frac{1}{V}}\underset{m0}{lim}{\displaystyle \frac{Z^1(m)+Z^1(m)}{m}}.`$ (44)
Since $`Z^1(m)=Z^1(m)`$, the chiral condensate is given by
$`\mathrm{\Sigma }={\displaystyle \frac{1}{\pi }}e^{2g^2𝒢(0)},`$ (45)
which can be used to eliminate $`𝒢(0)`$ from the partition function. Naively, eliminating $`𝒢(0)`$ in favor of the condensate is equivalent to fixing the constant in (33). Notice that only the product of the square of the normalization constant of the zero modes and $`\mathrm{exp}(2g^2𝒢(0))`$ is fixed by this condition.
## III Combinatorial Derivation of Sum Rules
In the sector of topological charge $`\nu `$ we find for the small mass expansion of the partition function in a finite but large volume $`V`$
$`{\displaystyle \frac{Z^\nu (m)}{Z^0(m=0)}}={\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{m^n}{n!}}{\displaystyle d^dx_1\mathrm{}d^dx_n\overline{\psi }\psi (x_1)\mathrm{}\overline{\psi }\psi (x_n)^\nu }.`$ (46)
The average is over field configurations with topological charge $`\nu `$ weighted by the gauge field action and the determinant of the Dirac operator in the space of nonzero modes. For one flavor, the terms in the expansion with $`n<\nu `$ vanish, since the number of zero modes in the correlator is not sufficient to compensate the zeros of fermion determinant. For $`n=\nu `$ we have
$`\overline{\psi }\psi (x_1)\mathrm{}\overline{\psi }\psi (x_n)^\nu =\overline{\psi }_R\psi _L(x_1)\mathrm{}\overline{\psi }_R\psi _L(x_n)^\nu .`$ (47)
For massless quarks, this expectation value is called a minimal correlator and saturated by zero modes bardakci ; manias-cor ; steele . If we use the explicit expressions for the zero modes and take into account their (anti-)symmetrization, we recover (41) and what followed. However, since we are only interested in the leading large-volume behavior, there is a simpler way - to use the clustering property: for all $`|x_ix_j|\mathrm{}`$,
$`\overline{\psi }_R\psi _L(x_1)\mathrm{}\overline{\psi }_R\psi _L(x_n)^\nu \left({\displaystyle \frac{\mathrm{\Sigma }}{2}}\right)^\nu ,`$ (48)
so that, in agreement with (43) Adam:1997wt
$`{\displaystyle \frac{Z^\nu (m)}{Z^0|_{m=0}}}={\displaystyle \frac{m^\nu }{\nu !}}\left({\displaystyle \frac{\mathrm{\Sigma }V}{2}}\right)^\nu +O(m^{\nu +1}).`$ (49)
Let us consider sub-leading orders in $`m`$ in (46). For $`n>\nu `$, the non-vanishing contributions have $`n\nu `$ contractions from the nonzero mode part of the propagator. Since the massless propagator connects only states with the same chirality the expectation value $`\overline{\psi }\psi (x_1)\mathrm{}\overline{\psi }\psi (x_{\nu +1})^\nu =0`$. The first non-vanishing sub-leading term in (46) is
$`{\displaystyle \frac{m^{\nu +2}}{(\nu +2)!}}{\displaystyle d^2x_1\mathrm{}d^2x_nd^2xd^2y\overline{\psi }\psi (x_1)\mathrm{}\overline{\psi }\psi (x_n)\overline{\psi }\psi (x)\overline{\psi }\psi (y)^\nu }.`$ (50)
This correlator is sometimes called a non-minimal correlator and was calculated in manias-cor ; steele ; Azakov:2001pz . However, for our purpose we will only need the large distance limit of this correlator, which again follows from the clustering property of the correlators. By decomposing the field into left handed and right handed components we find
$`{\displaystyle d^2x_1\mathrm{}d^2x_\nu d^2xd^2y\overline{\psi }\psi (x_1)\mathrm{}\overline{\psi }\psi (x_\nu )\overline{\psi }\psi (x)\overline{\psi }\psi (y)_\nu }`$ (51)
$`=(\nu +2){\displaystyle d^2x_1\mathrm{}d^2x_\nu d^2xd^2y\overline{\psi }_L\psi _R(x_1)\mathrm{}\overline{\psi }_L\psi _R(x_\nu )\overline{\psi }_L\psi _R(x)\overline{\psi }_R\psi _L(y)^\nu }.`$
The combinatorial factor $`\nu +2`$ arises because the fermion bilinear with opposite chiralities can be at each of the $`\nu +2`$ points. Using the clustering property for the expectation value we obtain for the large volume limit of (46)
$`{\displaystyle \frac{m^{\nu +2}}{(\nu +2)!}}(\nu +2)\left({\displaystyle \frac{\mathrm{\Sigma }}{2}}\right)^{\nu +2}={\displaystyle \frac{m^{\nu +2}}{(\nu +1)!}}\left({\displaystyle \frac{\mathrm{\Sigma }}{2}}\right)^{\nu +2}.`$ (52)
The mass dependence of the partition function to this order is thus given by
$`{\displaystyle \frac{Z^\nu (m)}{Z^0|_{m=0}}}={\displaystyle \frac{m^\nu }{\nu !}}\left({\displaystyle \frac{\mathrm{\Sigma }V}{2}}\right)^\nu +{\displaystyle \frac{m^{\nu +2}}{(\nu +1)!}}\left({\displaystyle \frac{\mathrm{\Sigma }V}{2}}\right)^{\nu +2}+O(m^{\nu +4}).`$ (53)
Comparing this result with (8), we arrive at the sum rule
$`{\displaystyle \underset{n0}{}}{\displaystyle \frac{1}{\lambda _n^2}}^\nu ={\displaystyle \frac{1}{|\nu |+1}}{\displaystyle \frac{\mathrm{\Sigma }^2V^2}{2}},`$ (54)
which also incorporates the case of negative $`\nu `$. In LS this result was obtained from the mass dependence of the partition function at nonzero vacuum angle. In our derivation we only used the clustering property of expectation values. Therefore our derivation is also valid for QCD with one massless flavor. For the Schwinger model a derivation of the sum rule for zero topological charge was given in smvac .
The term of the order $`m^{\nu +2l},l=2,3,_{\mathrm{}}`$ is not much more complicated. Again $`\nu `$ of the fermion bilinears are saturated by zero modes. The remaining ones are contracted by the nonzero mode Green’s function, and therefore we have to choose $`l`$ of the $`\nu +2l`$ fermion bilinears with opposite chirality Adam:1997wt ,
$`{\displaystyle d^2x_1\mathrm{}d^2x_{\nu +2l}\overline{\psi }\psi (x_1)\mathrm{}\overline{\psi }\psi (x_{\nu +2l})^\nu }`$ (55)
$`=`$ $`\left(\begin{array}{c}\nu +2l\\ l\end{array}\right){\displaystyle }d^2x_1\mathrm{}d^2x_{\nu +2l}\overline{\psi }_L\psi _R(x_1)\mathrm{}\overline{\psi }_L\psi _R(x_\nu )\overline{\psi }_L\psi _R(x_{\nu +1}\mathrm{}\overline{\psi }_R\psi _L(x_{\nu +2l})^\nu .`$ (58)
Again using the clustering property we obtain for the large volume limit of the $`m^{\nu +2l}`$ order term in (46),
$`{\displaystyle \frac{m^{\nu +2l}}{(\nu +2l)!}}{\displaystyle \frac{(\nu +2l)!}{(\nu +l)!l!}}\left({\displaystyle \frac{\mathrm{\Sigma }V}{2}}\right)^{\nu +2l}={\displaystyle \frac{m^{\nu +2l}}{(\nu +l)!l!}}\left({\displaystyle \frac{\mathrm{\Sigma }V}{2}}\right)^{\nu +2l}.`$ (59)
The term of order $`m^{2l}`$ in the expansion of (7) is given by
$`m^{2l}\stackrel{}{det}[i\overline{)}D]{\displaystyle \underset{\genfrac{}{}{0pt}{}{n_i>0}{n_1\mathrm{}n_l}}{}}{\displaystyle \frac{1}{\lambda _{n_1}^2\mathrm{}\lambda _{n_l}^2}}.`$ (60)
Gathering all the pieces, we obtain the final result smvac ; LS
$`{\displaystyle \underset{\genfrac{}{}{0pt}{}{n_i0}{n_1\mathrm{}n_l}}{}}{\displaystyle \frac{1}{\lambda _{n_1}^2\mathrm{}\lambda _{n_l}^2}}^\nu ={\displaystyle \frac{|\nu |!}{2^ll!(|\nu |+l)!}}\left(\mathrm{\Sigma }V\right)^{2l}.`$ (61)
This sum rule is valid for any theory with the above clustering property. In particular, it is valid for QCD with one flavor LS .
## IV Microscopic Derivation of Sum Rules in the Schwinger Model
A microscopic derivation of the sum rule (54) in the sector of topological charge $`\nu =0`$ was given in smvac . In this section we generalize this derivation to arbitrary nonzero topological charge. The main idea of the derivation may be summarized in the following identity
$`{\displaystyle \underset{\lambda _n0}{}}{\displaystyle \frac{1}{\lambda _n^2}}=\mathrm{Tr}\left[G^{\nu \mathrm{\hspace{0.17em}2}}\right],`$ (62)
where $`G^\nu `$ is the Green’s function of the Dirac operator with external field $`A_\mu =ϵ_{\mu \rho }_\rho \varphi `$. The equality (62) follows from the spectral representation of the Green’s function
$`G^\nu (x,y)={\displaystyle \underset{n0}{}}{\displaystyle \frac{\psi _n^\varphi (x)\psi _n^\varphi ^{}(y)}{i\lambda _n}},`$ (63)
and the fact that excited Dirac states are normalized to 1 (the trace is both over Dirac indexes and spatial coordinates). Therefore, we need the Green’s function $`G^\nu (x,y)`$ for an arbitrary background field $`\varphi `$ in the sector of topological charge $`\nu `$ (which is taken $`\nu >0`$ for convenience). Due to the index theorem, the Dirac equation has exactly $`\nu `$ right-handed zero modes for a generic field configuration in this class. The Green’s function $`G^\nu (x,y)`$ is thus defined by brown
$`i\overline{)}D_xG^\nu (x,y)=\delta (xy)\gamma _5^+P^\nu (x,y),`$ (64)
where
$`P^\nu (x,y)={\displaystyle \underset{p=1}{\overset{\nu }{}}}{\displaystyle \frac{\psi _p(x)\psi _p^{}(y)}{d^2z\psi _p^{}(z)\psi _p(z)}},`$ (65)
is the projector on the subspace of zero modes, $`\psi _p(x),p=1,\mathrm{},\nu `$, and
$`\gamma _5^+={\displaystyle \frac{1}{2}}(1+\gamma _5).`$ (66)
In other words, $`G_\nu (x,y)`$ is the Green’s function in the space of the nonzero modes. The explicit solution of (64) is given by:
$`G^\nu (x,y)=(1\gamma _5^+P^\nu )\stackrel{~}{G}^\nu (1\gamma _5^+P^\nu ),`$ (67)
where, because $`i\overline{)}D_\varphi =e^{g\varphi \gamma _5}i\overline{)}e^{g\varphi \gamma _5}`$,
$`\stackrel{~}{G}^\nu (x,y)e^{g\varphi (x)\gamma _5}G_0(xy)e^{g\varphi (y)\gamma _5}.`$ (68)
The free two-dimensional Dirac propagator $`G_0(x,y)`$ is given by
$`G_0(x,y)={\displaystyle \frac{1}{2\pi }}{\displaystyle \frac{\gamma _\mu (x_\mu y_\mu )}{(xy)^2}}.`$ (69)
In the representation $`\psi _1(x)=e^{g\varphi (x)}`$ and $`\{\psi _k,k=2,\mathrm{},\nu \}`$ is an orthogonal set perpendicular to $`\psi `$, one can easily show that (67) is a solution of (64). Using that $`\gamma _5^+\stackrel{~}{G}^\nu \gamma _5^+=0`$ we find the sum rule
$`{\displaystyle \underset{\lambda _n0}{}}{\displaystyle \frac{1}{\lambda _n^2}}`$ $`=`$ $`\mathrm{Tr}\left[(1(1+\gamma _5)P^\nu )[G^\nu ]^2\right]`$ (70)
$`=`$ $`\mathrm{Tr}\left[(1+\gamma _5)(1P^\nu )[G^\nu ]^2\right].`$
In the second equality we have used that in the term proportional to the identity, both helicities give the same result. Using the explicit representation for the Green’s function and carrying out the trace over the $`\gamma `$ matrices, we find
$`{\displaystyle \underset{\lambda _n0}{}}{\displaystyle \frac{1}{\lambda _n^2}}`$ $`=`$ $`{\displaystyle d^2xd^2yd^2z(\delta (xy)P^\nu (x,y))tr\left[(1+\gamma _5)e^{g\varphi (y)\gamma 5}G_0(y,z)e^{2g\varphi (z)\gamma _5}G_0(z,x)e^{g\varphi (x)\gamma _5}\right]}`$ (71)
$`=`$ $`{\displaystyle \frac{2}{(2\pi )^2}}{\displaystyle d^2xd^2yd^2z(\delta (xy)P^\nu (x,y))e^{g\varphi (y)g\varphi (x)+2g\varphi (z)}}`$
$`\times {\displaystyle \frac{(xz)_\mu (zy)_\mu iϵ_{\mu \nu }(yz)_\mu (zx)_\nu }{(xz)^2(zy)^2}}`$
$`=`$ $`{\displaystyle \frac{1}{(2\pi )^2}}{\displaystyle d^2xd^2yd^2z(\delta (xy)P^\nu (x,y))e^{g\varphi (y)g\varphi (x)+2g\varphi (z)}}`$
$`\times \left[{\displaystyle \frac{(xy)^22iϵ_{\mu \nu }(yz)_\mu (zx)_\nu }{(xz)^2(zy)^2}}{\displaystyle \frac{1}{(xz)^2}}{\displaystyle \frac{1}{(yz)^2}}\right].`$
In the last equality we have used that
$`(xz)_\mu (zy)_\mu ={\displaystyle \frac{1}{2}}\left((xy)^2(xz)^2(zy)^2\right).`$ (72)
The last two terms in the square brackets are independent of either $`y`$ or $`x`$ and vanish after integration as can be seen from the representation (65)
$`{\displaystyle d^2yP^\nu (x,y)e^{g\varphi (y)}}=e^{g\varphi (x)}.`$ (73)
Using that the integral over the $`\delta `$-function also vanishes, we find the sum rule
$`{\displaystyle \underset{\lambda _n0}{}}{\displaystyle \frac{1}{\lambda _n^2}}`$ $`=`$ $`{\displaystyle \frac{1}{(2\pi )^2}}{\displaystyle d^2xd^2yd^2zP^\nu (x,y)e^{g\varphi (y)g\varphi (x)+2g\varphi (z)}\frac{(xy)^22iϵ_{\mu \nu }(yz)_\mu (zx)_\nu }{(xz)^2(zy)^2}}.`$
The projector formula (65) is written in terms of orthogonal modes. However, since the zero modes (25) are not orthogonal, we require a projector formula that is valid for an arbitrary set of non-orthogonal modes. This is easily achieved by using a Lagrange interpolation such that the projector is equal unity for each of the zero modes. The required form is given by
$`P^\nu (x,y)={\displaystyle \frac{1}{(\nu 1)!}}{\displaystyle \frac{1}{det𝒩}}{\displaystyle \underset{\sigma }{}}\mathrm{sgn}(\sigma )\left|\begin{array}{cccc}\psi _1^{}(y)\psi _{\sigma (1)}(x)& (\psi _1,\psi _{\sigma (2)})& \mathrm{}& (\psi _1,\psi _{\sigma (\nu )})\\ \mathrm{}& \mathrm{}& & \mathrm{}\\ \psi _\nu ^{}(y)\psi _{\sigma (1)}(x)& (\psi _\nu ,\psi _{\sigma (2)})& \mathrm{}& (\psi _\nu ,\psi _{\sigma (\nu )})\end{array}\right|,`$ (78)
where the normalization factor is the familiar determinant of the norm matrix, given by (40). The projector can be expressed in terms of Slater determinants as
$`P^\nu (x,y)=\nu {\displaystyle \frac{d^2x_2\mathrm{}d^2x_\nu \chi (x,x_2,\mathrm{},x_\nu )\chi ^{}(y,x_2,\mathrm{},x_\nu )}{d^2x_1\mathrm{}d^2x_\nu |\chi (x_1,\mathrm{},x_\nu )|^2}}.`$ (79)
Here, $`\chi (x_1,\mathrm{},x_\nu )`$ is the Slater determinant of fermionic zero modes. In this representation, it is particularly transparent that the projector is not only independent of the choice of basis, but (unlike $`det𝒩`$), is also independent of the normalization of the zero modes, as of course it should be. Using the explicit representation of the fermionic zero modes we find the projector
$`\nu {\displaystyle \frac{𝑑x_2\mathrm{}𝑑x_\nu _{k=2}^\nu (x^+x_k^+)(y^{}x_k^{})\mathrm{\Delta }(\{x_2^+,\mathrm{},x_\nu ^+\})\mathrm{\Delta }(\{x_2^{},\mathrm{},x_\nu ^{}\})e^{g(\varphi (x)+\varphi (y))2g(\varphi (x_2)+\mathrm{}+\varphi (x_\nu ))}}{𝑑x_1\mathrm{}𝑑x_\nu \mathrm{\Delta }(\{x_1^+,\mathrm{},x_\nu ^+\})\mathrm{\Delta }(\{x_1^{},\mathrm{},x_\nu ^{}\})e^{2g(\varphi (x_1)+\mathrm{}+\varphi (x_\nu ))}}},`$
where $`\mathrm{\Delta }`$ is a Vandermonde determinant. Renaming $`x`$ by $`x_1`$ and $`y`$ by $`x_{\nu +1}`$, the sum rule in the sector of topological charge $`\nu `$ can be written as
$`{\displaystyle \underset{\lambda _n0}{}}{\displaystyle \frac{1}{\lambda _n^2}}`$ $`=`$ $`{\displaystyle \frac{2}{(2\pi )^{\nu +2}(\nu 1)!}}{\displaystyle \frac{1}{det𝒩}}{\displaystyle d^2z𝑑x_1\mathrm{}𝑑x_{\nu +1}e^{2g(\varphi (x_1)+\mathrm{}+\varphi (x_{\nu +1}))+2g\varphi (z)}}`$
$`\times {\displaystyle \underset{k=2}{\overset{\nu }{}}}(x_1^+x_k^+)(x_{\nu +1}^{}x_k^{})\mathrm{\Delta }(\{x_2^+,\mathrm{}x_\nu ^+\})\mathrm{\Delta }(\{x_2^{},\mathrm{}x_\nu ^{}\})`$
$`\times {\displaystyle \frac{(x_1x_{\nu +1})^22iϵ_{\alpha \beta }(x_1z)_\alpha (zx_{\nu +1})_\beta }{(x_1z)^2(zx_{\nu +1})^2}}.`$
Only the symmetric part of the expression in the second line of this equation contributes to the integrand. Since the expression is already symmetric in $`x_2,\mathrm{},x_\nu `$, we only need to symmetrize with respect to $`x_1`$ and $`x_{\nu +1}`$. Then we get,
$`𝒮{\displaystyle \underset{k=2}{\overset{\nu }{}}}(x_1^+x_k^+)(x_{\nu +1}^{}x_k^{})\mathrm{\Delta }(\{x_2^+,\mathrm{}x_\nu ^+\})\mathrm{\Delta }(\{x_2^{},\mathrm{}x_\nu ^{}\})`$
$`\times {\displaystyle \frac{(x_1x_{\nu +1})^2/2iϵ_{\mu \nu }(x_1z)_\mu (zx_{\nu +1})_\nu }{(x_1z)^2(zx_{\nu +1})^2}}`$
$`={\displaystyle \frac{\mathrm{\Delta }(\{x_1^+,\mathrm{}x_{\nu +1}^+\})\mathrm{\Delta }(\{x_1^{},\mathrm{}x_{\nu +1}^{}\})}{\nu (\nu +1)}}{\displaystyle \underset{pq}{}}{\displaystyle \frac{(x_px_q)^2/2iϵ_{\mu \nu }(x_pz)_\mu (zx_q)_\nu }{_{lp}(x_p^{}x_l^{})_{lq}(x_q^+x_l^+)(x_pz)^2(zx_q)^2}}.`$
(82)
The term with $`p=q`$ vanishes and can be included in the sum. Using that
$`{\displaystyle \underset{p}{}}{\displaystyle \frac{1}{_{kp}(x_px_k)}}=0,`$ (83)
we can replace the numerator
$`(x_px_q)^2/2iϵ_{\mu \nu }(x_pz)_\mu (zx_q)_\nu `$ (84)
in (82) by
$`(x_px_q)^2/2(x_pz)^2/2(x_qz)^2/2iϵ_{\mu \nu }(x_pz)_\mu (zx_q)_\nu `$ (85)
$`=`$ $`(x_p^+z^+)(x_q^{}z^{}).`$
After simplifying the numerator we find for (82)
$`{\displaystyle \frac{2\mathrm{\Delta }(\{x_1^+,\mathrm{}x_{\nu +1}^+\})\mathrm{\Delta }(\{x_1^{},\mathrm{}x_{\nu +1}^{}\})}{\nu +1}}{\displaystyle \underset{p,q}{}}{\displaystyle \frac{1}{_{lp}(x_p^{}x_l^{})_{lq}(x_q^+x_l^+)(x_p^{}z^{})(x_q^+z^+)}}.`$
The sum in this equation factorizes into the product of two sums. Each sum can be expressed as a determinant
$`{\displaystyle \underset{p}{}}{\displaystyle \frac{\mathrm{\Delta }(\{x_1^{},\mathrm{}x_{\nu +1}^{}\})}{_{lp}(x_p^{}x_l^{})(x_p^{}z^{})}}`$ $`=`$ $`\left|\begin{array}{ccccc}1/(x_1^{}z^{})& 1& x_1^{}& \mathrm{}& x_1^{\nu 1}\\ \mathrm{}& & & & \mathrm{}\\ 1/(x_{\nu +1}^{}z^{})& 1& x_{\nu 1}^{}& \mathrm{}& x_{\nu +1}^{\nu 1}\end{array}\right|`$ (90)
$`=`$ $`{\displaystyle \frac{\mathrm{\Delta }(\{x_1^{},\mathrm{}x_{\nu +1}^{}\})}{_p(x_p^{}z^{})}}.`$ (91)
The determinant is easily evaluated by multiplying each row by a factor $`(x_k^{}z^{})`$. Exactly the same simplification can be made for the $`+`$ variables. The sum rule therefore simplifies to
$`{\displaystyle \underset{\lambda _n0}{}}{\displaystyle \frac{1}{\lambda _n^2}}`$ $`=`$ $`{\displaystyle \frac{2}{(2\pi )^{\nu +2}(\nu +1)!}}{\displaystyle \frac{1}{det𝒩}}{\displaystyle d^2z𝑑x_1\mathrm{}𝑑x_{\nu +1}e^{2g(\varphi (x_1)+\mathrm{}+\varphi (x_{\nu +1}))+2g\varphi (z)}}`$ (92)
$`\times {\displaystyle \frac{\mathrm{\Delta }(\{x_1^+,\mathrm{}x_{\nu +1}^+\})\mathrm{\Delta }(\{x_1^{},\mathrm{}x_{\nu +1}^{}\})}{_{q=1}^{\nu +1}(z^+x_q^+)(z^{}x_q^{})}}.`$
This sum rule is valid for a arbitrary external gauge field $`A_\mu =ϵ_{\mu \rho }_\rho \varphi `$ in the sector of topological charge $`\nu `$. The Leutwyler-Smilga sum rules (10) are obtained after averaging over the fields $`\varphi (x)`$ with the effective action $`det𝒩e^{\mathrm{\Gamma }[\varphi ]}`$, where $`\mathrm{\Gamma }[\varphi ]`$ is defined in (31). Notice that effective action in the nontrivial topological sectors, due to the presence of $`det𝒩`$, is not only non-Gaussian in $`\varphi `$ but also non-local in $`\varphi `$. However, the factor $`det𝒩`$ cancels in the average of the sum rule. The resulting path integral is Gaussian in $`\varphi `$ and the average over the $`\varphi `$ fields is given by a lowest order cumulant expansion. This results in
$`{\displaystyle \underset{\lambda _n0}{}}{\displaystyle \frac{1}{\lambda _n^2}}`$ $`=`$ $`{\displaystyle \frac{2}{(2\pi )^{\nu +2}(\nu +1)!}}{\displaystyle d^2z𝑑x_1\mathrm{}𝑑x_{\nu +1}e^{2(\nu +2)g^2G(0)+4g^2_{k<l}^{\nu +1}G(x_k,x_l)4g^2_{k=1}^{\nu +1}G(x_k,z)}}`$ (93)
$`\times {\displaystyle \frac{\mathrm{\Delta }(\{x_1^+,\mathrm{}x_{\nu +1}^+\})\mathrm{\Delta }(\{x_1^{},\mathrm{}x_{\nu +1}^{}\})}{_{q=1}^{\nu +1}(z^+x_q^+)(z^{}x_q^{})}}.`$
The second line of (93) is canceled by the asymptotic behavior of the exponentiated Green’s functions in the first line of (93) (see (34)). The large volume limit of the sum rule is therefore given by
$`{\displaystyle \underset{\lambda _n0}{}}{\displaystyle \frac{1}{\lambda _n^2}}`$ $`=`$ $`{\displaystyle \frac{2}{(2\pi )^{\nu +2}(\nu +1)!}}V^{\nu +2}e^{2(\nu +2)g^2G(0)}={\displaystyle \frac{2}{(\nu +1)!}}\left({\displaystyle \frac{\mathrm{\Sigma }V}{2}}\right)^{\nu +2},`$ (94)
where we used (45). Now, using (49), and the definition (6), we finally obtain for the Leutwyler-Smilga sum rule,
$`{\displaystyle \underset{\lambda _n0}{}}{\displaystyle \frac{1}{\lambda _n^2}}_\nu `$ $`=`$ $`{\displaystyle \frac{m^\nu Z^0(m)}{Z^\nu (m)}}|_{m=0}{\displaystyle \underset{\lambda _n0}{}}{\displaystyle \frac{1}{\lambda _n^2}}`$ (95)
$`=`$ $`{\displaystyle \frac{m^\nu Z^0(m)}{Z^\nu (m)}}|_{m=0}{\displaystyle \frac{2}{(2\pi )^{\nu +2}(\nu +1)!}}V^{\nu +2}e^{2(\nu +2)g^2G(0)}={\displaystyle \frac{1}{2(\nu +1)}}\mathrm{\Sigma }^2V^2.`$
The result (95) is in agreement with the sum rule obtained by Leutwyler and Smilga LS and with the result obtained from Random Matrix Theory SV .
## V Bosonization and Sum Rules
In this section we derive sum rules for the inverse Dirac eigenvalues starting from a bosonic description of the massive Schwinger Model. In the particle physics literature, abelian bosonization goes back to the work of Lowenstein and Swieca Lowenstein:1971fc , who essentially bosonized massless Schwinger Model, and Coleman Coleman and Mandelstam Mandelstam . The bosonized form of the massive Schwinger Model was studied in Coleman-Jackiw ; Coleman-multi . Most earlier works used an operator language. The path integral approach to bosonization, which we will use below, was developed later, in many works, in particular in GamboaSaravi:1983xw ; Naon:1984zp .
It is well-known Coleman-Jackiw ; Coleman-multi that the action of the massive Schwinger model can be written as
$`S[\phi ]={\displaystyle d^2x\left[\frac{1}{2}\left(_\nu \phi \right)^2+\frac{\mu ^2}{2}\phi ^2cm\mathrm{cos}(2\sqrt{\pi }\phi \theta )\right]}.`$ (96)
with the constant $`c=\mu e^{\gamma _E}/(2\pi )|\mathrm{\Sigma }|`$. The value of this constant is such that the chiral condensate is equal to its known value (26). At small $`m`$ this is a theory of a weakly self-coupled massive scalar field. The physics of this model is discussed in detail in Coleman-Jackiw ; Coleman-multi . In the mean field limit, where $`\varphi =0`$, we obtain (12) from the action (96). This directly leads to the Leutwyler-Smilga sum rules.
The first step in the microscopic derivation of sum rules is to eliminate the mass term of the $`\phi `$ field by
$`e^{{\scriptscriptstyle d^2x{\scriptscriptstyle \frac{\mu ^2}{2}}\phi ^2}}={\displaystyle D(\mathrm{\Delta }\varphi )e^{{\scriptscriptstyle d^2x[{\scriptscriptstyle \frac{1}{2}}(\mathrm{\Delta }\varphi )^2+i\mathrm{\Delta }\varphi \phi ]}}}.`$ (97)
After partial integration of $`(\phi )^2`$ and shifting $`\phi `$ by $`i\mu \varphi `$ we obtain the partition function
$`Z(m,\theta )={\displaystyle D\phi D\varphi e^{S_{\mathrm{eff}}[\phi ,\varphi ]}},`$ (98)
with effective action given by
$`S_{\mathrm{eff}}[\varphi ,\phi ]={\displaystyle d^2x\left[\frac{1}{2}\varphi [\mathrm{\Delta }^2\mu ^2\mathrm{\Delta }]\varphi +\frac{1}{2}\left(_\nu \phi \right)^2\frac{1}{2}cm\left[e^{2g\varphi +i2\sqrt{\pi }\phi +i\theta }+e^{2g\varphi i2\sqrt{\pi }\phi i\theta }\right]\right]}.`$
(99)
The integral $`D\varphi `$ is only over modes with $`\mathrm{\Delta }\varphi `$ not identically equal zero. The partition function in the sector of topological charge $`\nu `$ is given by the Fourier transform of the partition function,
$`Z^\nu (m)={\displaystyle _0^{2\pi }}{\displaystyle \frac{d\theta }{2\pi }}e^{i\theta \nu }Z(m,\theta ).`$ (100)
The integral over $`\theta `$ is easily performed by expanding the integrand in powers of $`e^{i\theta }`$,
$`Z^\nu (m)`$ $`=`$ $`{\displaystyle _0^{2\pi }}{\displaystyle \frac{d\theta }{2\pi }}e^{i\theta \nu }{\displaystyle D\phi D\varphi e^{S_{\mathrm{eff}}(\phi ,\varphi )}\underset{k,l=0}{\overset{\mathrm{}}{}}\frac{(cm/2)^{k+l}}{k!l!}d^2x_1\mathrm{}d^2x_kd^2y_1\mathrm{}d^2y_l}`$ (101)
$`\times \left[{\displaystyle \underset{q=1}{\overset{k}{}}}e^{2g\varphi (x_q)i2\sqrt{\pi }\phi (x_q)i\theta }\right]\left[{\displaystyle \underset{p=1}{\overset{l}{}}}e^{2g\varphi (y_p)+i2\sqrt{\pi }\phi (y_p)+i\theta }\right].`$
It is clear that only the terms with $`k=l+\nu `$ contribute to the integral. However, we can also perform the integral after shifting
$`\varphi \varphi {\displaystyle \frac{i}{2g}}\theta .`$ (102)
In the effective action this results in the extra term
$`i\theta {\displaystyle \frac{g}{2\pi }}{\displaystyle d^2x\mathrm{\Delta }\varphi }.`$ (103)
The integral over $`\theta `$ therefore vanishes unless
$`{\displaystyle \frac{g}{2\pi }}{\displaystyle d^2x\mathrm{\Delta }\varphi }=\nu ,`$ (104)
i.e. the topological charge of the fields $`\varphi `$ is equal to $`\nu `$.
The path integral over $`\phi `$ is Gaussian, and is given by
$`{\displaystyle D\phi \left[\underset{p=1}{\overset{l}{}}e^{i2\sqrt{\pi }\phi (y_p)}\right]\left[\underset{q=1}{\overset{\nu +l}{}}e^{i2\sqrt{\pi }\phi (x_q)}\right]e^{S_0[\phi ]}}={\displaystyle \frac{_{k<r}^l(y_ky_r)^2_{i<j}^{\nu +l}(x_ix_j)^2}{_{q=1}^{\nu +l}_{p=1}^l(y_px_q)^2}}{\displaystyle D\phi e^{S_0[\phi ]}}.`$
with <sup>1</sup><sup>1</sup>1It is assumed that an infrared regulator has been introduced for the massless scalar field.
$`S_0[\phi ]={\displaystyle \frac{1}{2}}{\displaystyle d^2x\left(_\mu \phi \right)^2}.`$ (106)
Thus we end up with
$`Z^\nu (m)`$ $`=`$ $`{\displaystyle D\phi e^{S_0[\phi ]}D\varphi e^{\mathrm{\Gamma }[\varphi ]}\underset{l=0}{\overset{\mathrm{}}{}}\frac{(cm/2)^{2l+\nu }}{(\nu +l)!l!}d^2y_1\mathrm{}d^2y_ld^2x_1\mathrm{}d^2x_{\nu +l}}`$ (107)
$`\times {\displaystyle \frac{_{k<r}^l(y_ky_r)^2_{i<j}^{\nu +l}(x_ix_j)^2}{_{q=1}^{\nu +l}_{p=1}^l(y_px_q)^2}}e^{2g[\varphi (y_1)+\mathrm{}+\varphi (y_l)\varphi (x_1)\mathrm{}\varphi (x_{\nu +l})]}.`$
The path integral over $`\varphi `$ can be performed by a leading order cumulant expansion. The Vandermonde determinants are again canceled by the large distance asymptotic behavior of the Green’s functions. In the large volume limit we thus obtain
$`{\displaystyle \frac{Z_\nu (m)}{Z_0(m=0)}}={\displaystyle \underset{l=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{m^{\nu +2l}}{(\nu +l)!l!}}\left({\displaystyle \frac{\mathrm{\Sigma }V}{2}}\right)^{\nu +2l},`$ (108)
which is equal to the result obtained from the clustering assumption The sum rules are the same as derived in section VI,
$`{\displaystyle \underset{\genfrac{}{}{0pt}{}{n_i0}{n_1\mathrm{}n_l}}{}}{\displaystyle \frac{1}{\lambda _{n_1}^2\mathrm{}\lambda _{n_l}^2}}^\nu ={\displaystyle \frac{|\nu |!}{2^ll!(|\nu |+l)!}}\left(\mathrm{\Sigma }V\right)^{2l}.`$ (109)
From this result one easily derives that $`Z(m,\theta )=\mathrm{exp}(mV\mathrm{\Sigma }\mathrm{cos}\theta )`$ (see also LS ; Adam:1997wt ).
We end this section with a conjecture for sum rules in a fixed external $`\varphi `$field. By now we realize that the field $`\varphi `$ introduced in (97) is indeed the same as in previous sections. We can extract the sum rules from comparison with the partition function in eigenvalue representation. To that end, we rewrite the partition function as
$`Z^\nu (m)`$ $`=`$ $`m^\nu {\displaystyle D\varphi e^{S[\varphi ]}\stackrel{}{det}[i\overline{)}D]\frac{det^{}[i\overline{)}D+m]}{det^{}[i\overline{)}D]}}`$ (110)
$`=`$ $`{\displaystyle D\varphi e^{\mathrm{\Gamma }[\varphi ]}m^\nu det𝒩^\nu \underset{l=0}{\overset{\mathrm{}}{}}m^{2l}\underset{\genfrac{}{}{0pt}{}{n_i0}{n_1\mathrm{}n_l}}{}\frac{1}{\lambda _{n_1}^2\mathrm{}\lambda _{n_l}^2}},`$
where the formulas (7) and (60) have been used. We remind that $`det𝒩^\nu `$ is determinant of the norm matrix (40). We conjecture that the integrands of (107) and (110) are equal:
$`{\displaystyle \underset{\genfrac{}{}{0pt}{}{n_i0}{n_1\mathrm{}n_l}}{}}{\displaystyle \frac{1}{\lambda _{n_1}^2\mathrm{}\lambda _{n_l}^2}}`$ $`=`$ $`{\displaystyle \frac{𝒞}{(\nu +l)!l!det𝒩^\nu }}{\displaystyle d^2y_1\mathrm{}d^2y_ld^2x_1\mathrm{}d^2x_{\nu +l}}`$
$`\times {\displaystyle \frac{_{k<r}^l(y_ky_r)^2_{i<j}^{\nu +l}(x_ix_j)^2}{_{q=1}^{\nu +l}_{p=1}^l(y_px_q)^2}}e^{2g[\varphi (y_1)+\mathrm{}+\varphi (y_l)\varphi (x_1)\mathrm{}\varphi (x_{\nu +1})]}.`$
For $`l=1`$ this conjecture was shown to be valid in in section IV.
## VI Clustering for Two of More Flavors in ChPT
In multi-flavor QCD, chiral symmetry is broken spontaneously, with the appearance of massless Goldstone bosons. This results in long range correlations that will modify the clustering property of scalar correlators and will affect the sum rules for the inverse Dirac eigenvalues. Starting from a chiral Lagrangian we will propose a modified clustering relation.
Let us first show that factorization leads to an incorrect result. To leading order in the quark masses the partition function in the sector of topological charge $`\nu `$ is given by
$`{\displaystyle \frac{Z^\nu (m_1,\mathrm{},m_{N_f})}{Z^0(m_i=0)}}={\displaystyle \frac{(m_1\mathrm{}m_{N_f})^\nu }{[\nu !]^{N_f}}}{\displaystyle \underset{i,j}{}dx_i^{(j)}\overline{\psi }\psi (x_1^{(1)})\mathrm{}\overline{\psi }\psi (x_\nu ^{(1)})\mathrm{}\overline{\psi }\psi (x_1^{(N_f)})\mathrm{}\overline{\psi }\psi (x_\nu ^{(N_f)})^\nu }.`$
(112)
If we assume clustering as in the case of one flavor we obtain
$`\overline{\psi }\psi (x_1^{(1)})\mathrm{}\overline{\psi }\psi (x_\nu ^{(1)})\mathrm{}\overline{\psi }\psi (x_1^{(N_f)})\mathrm{}\overline{\psi }\psi (x_\nu ^{(N_f)})^\nu \left({\displaystyle \frac{\mathrm{\Sigma }}{2}}\right)^{\nu N_f},`$ (113)
resulting in the partition function
$`{\displaystyle \frac{Z^\nu (m_1,\mathrm{},m_{N_f})}{Z^0(m_i=0)}}={\displaystyle \frac{1}{[\nu !]^{N_f}}}\left[{\displaystyle \frac{m_1\mathrm{\Sigma }V}{2}}\right]^\nu \mathrm{}\left[{\displaystyle \frac{m_{N_f}\mathrm{\Sigma }V}{2}}\right]^\nu .`$ (114)
However, instead of $`1/[\nu !]^{N_f}`$, the correct pre-factor LS is given by $`_{k=0}^{N_f1}k!/(\nu +k)!`$.
The reason for the incorrect pre-factor in (114) is that we did not account for the spontaneous symmetry breaking of $`SU_A(N_f)`$, which results in massless Goldstone bosons and long range correlations. As a consequence, the clustering property does not hold directly even for the correct ground state (i.e. at fixed $`\theta `$). It is possible, as we will show below, to write down a modified clustering property.
The clustering property follows from the large distance behavior of the correlators. This is determined by the zero momentum term in the chiral Lagrangian, i.e. by
$`Z_{\mathrm{Low}}(M,\theta )=𝒩{\displaystyle 𝒟Ue^{V\mathrm{\Sigma }\mathrm{ReTr}(MU^{}e^{i\theta /N_f})}},`$ (115)
with normalization constant $`𝒩`$ such that the partition function is equal to unity for $`M=0`$. For a diagonal mass matrix we obtain the expansion
$`Z_{\mathrm{Low}}(m_f,\theta )={\displaystyle \underset{k_1,\mathrm{},k_{N_f}=0}{\overset{\mathrm{}}{}}}{\displaystyle _{USU(N_f)}}𝑑U{\displaystyle \underset{l=1}{\overset{N_f}{}}}\left[\left({\displaystyle \frac{m_lV\mathrm{\Sigma }}{2}}\right)^{k_l}{\displaystyle \frac{(U_{ll}^{}e^{i\theta /N_f}+U_{ll}e^{i\theta /N_f})^{k_l}}{k_l!}}\right],`$ (116)
which should be compared to the large volume limit of the QCD partition function given by
$`Z^{\mathrm{QCD}}(m_f,\theta )={\displaystyle \underset{k_1,\mathrm{},k_{N_f}=0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{l=1}{\overset{N_f}{}}}{\displaystyle \frac{\left(m_lV\right)^{k_l}}{k_l!}}{\displaystyle \underset{l=1}{\overset{N_f}{}}}(\overline{\psi }_l\psi _l)_l^k_\theta .`$ (117)
We thus obtain the clustering relation
$`{\displaystyle \underset{l=1}{\overset{N_f}{}}}(\overline{\psi }_l\psi _l)^{k_l}_\theta ={\displaystyle _{USU(N_f)}}𝑑U{\displaystyle \underset{l=1}{\overset{N_f}{}}}{\displaystyle \frac{\mathrm{\Sigma }}{2}}(U_{ll}^{}e^{i\theta /N_f}+U_{ll}e^{i\theta /N_f})^{k_l}.`$ (118)
If we isolate the phase $`\xi _i`$ of the diagonal matrix element $`U_{ii}`$ this can be rewritten more suggestively as
$`{\displaystyle \underset{l=1}{\overset{N_f}{}}}(\overline{\psi }_l\psi _l)^{k_l}_\theta ={\displaystyle _{USU(N_f)}}𝑑U{\displaystyle \underset{l=1}{\overset{N_f}{}}}|U_{ll}|\overline{\psi }\psi _{(\theta /N_f)\xi _l}^{k_l},`$ (119)
with $`\overline{\psi }\psi _{(\theta /N_f)\xi _l}`$ the chiral condensate for one flavor and vacuum angle equal to $`(\theta /N_f)\xi _l`$. Notice that only the diagonal matrix elements of $`U`$ enter in this expression. Since $`|U_{ll}|1`$ we could interpret the factors $`|U_{ll}|`$ as the projection of the one flavor condensates onto certain directions in group space.
In general, the integral in (118) cannot be evaluated analytically. However, the situation simplifies enormously for the minimal correlators in the sector of fixed topological charge $`\nu `$. In this case we have $`n_1=n_2=\mathrm{}=n_{N_f}=\nu `$ and the integral over $`\theta `$ is only nonzero if no factors $`U_{ll}^{}`$ occur. This results in
$`{\displaystyle \underset{l=1}{\overset{N_f}{}}}(\overline{\psi }_l\psi _l)^\nu ^\nu =\left({\displaystyle \frac{\mathrm{\Sigma }}{2}}\right)^{\nu N_f}{\displaystyle _{USU(N_f)}}𝑑U{\displaystyle \underset{l=1}{\overset{N_f}{}}}U_{ll}^\nu .`$ (120)
This integral can be evaluated analytically (see appendix A)
$`{\displaystyle 𝑑UU_{11}^\nu U_{22}^n\mathrm{}U_{N_fN_f}^\nu }=\left[\nu !\right]^{N_f}{\displaystyle \underset{i=0}{\overset{N_f1}{}}}{\displaystyle \frac{i!}{(i+\nu )!}}.`$ (121)
The leading mass dependence of the partition function is obtained by substituting the result (120) into equation (112),
$`Z^\nu (m_1,\mathrm{},m_{N_f})=\left[{\displaystyle \frac{m_1\mathrm{\Sigma }V}{2}}\right]^\nu \mathrm{}\left[{\displaystyle \frac{m_{N_f}\mathrm{\Sigma }V}{2}}\right]^\nu {\displaystyle \underset{i=0}{\overset{N_f1}{}}}{\displaystyle \frac{i!}{(i+\nu )!}}.`$ (122)
This explains the suppression of the scalar correlator for $`N_f2`$.
## VII Relations with Random Matrix Theory
The chiral random matrix model in the sector of topological charge $`\nu `$ and $`N_f`$ flavors with equal quark mass $`m`$ is defined by SV ; V
$`Z_{N_f}^\nu (m)=C_n^{(1)}{\displaystyle 𝑑W\stackrel{N_f}{det}(D+m)e^{n\mathrm{\Sigma }^2\mathrm{Tr}WW^{}}},`$ (123)
where
$`D=\left(\begin{array}{cc}0& iW\\ iW^{}& 0\end{array}\right),`$ (126)
and $`W`$ is a complex $`n\times (n+\nu )`$ matrix. The eigenvalues of this model are distributed according to a semicircle. The largest eigenvalue is equal to $`1/\mathrm{\Sigma }`$. We use this eigenvalue to normalize the partition function such that it becomes dimensionless. If we also normalize the partition function to unity for $`N_f=0`$ we obtain the normalization constant
$`C_n^{(1)}=e^{nN_f}\left({\displaystyle \frac{n}{\pi }}\right)^{n(n+\nu )}\mathrm{\Sigma }^{2n(n+\nu )}.`$ (127)
We also have included the factor $`e^{nN_f}`$ which eliminates the constant vacuum energy for $`N_f0`$. We can evaluate the partition function (123) in the limit $`m0`$ in two different ways. First, using an eigenvalue representation of $`W`$, and second, using a $`\sigma `$-model representation of the large $`n`$ limit of the partition function. Since the $`\sigma `$-model is exactly the partition function (115), we only discuss the first approach in the next subsection.
### VII.1 Eigenvalue Representation
To obtain an eigenvalue representation of the partition function we use the polar decomposition
$`W=U\mathrm{\Lambda }V^1,`$ (128)
where $`UU(n+\nu )/U^n(1)\times U(\nu )`$ and $`VU(n)`$. The Jacobian of this transformation is given by
$`J=\mathrm{\Delta }(\{\lambda _k^2\})2^n{\displaystyle \underset{k=1}{\overset{n}{}}}(\lambda _k)^{2\nu +1}.`$ (129)
The eigenvalues of $`\mathrm{\Lambda }`$ are positive or zero, and the integration is over an ordered sequence of eigenvalues which will be accounted for by a factor $`1/n!`$ below. To lowest non-vanishing order in $`m`$, the eigenvalue representation of the partition function is thus given by
$`Z_{N_f}^\nu (m)=C_n^{(1)}C_n^{(2)}m^{\nu N_f}{\displaystyle \frac{1}{n!}}{\displaystyle \underset{k=1}{\overset{n}{}}2d\lambda _k\lambda _k^{2(\nu +N_f)+1}\mathrm{\Delta }^2(\{\lambda _k^2\})e^{n\mathrm{\Sigma }^2_k\lambda _k^2}},`$ (130)
where we have introduced a second constant,
$`C_n^{(2)}=2^{n^2n\nu }{\displaystyle \frac{\mathrm{vol}(U(n+\nu ))\mathrm{vol}(U(n))}{\mathrm{vol}(U(\nu ))\mathrm{vol}^n(U(1))}},`$ (131)
and the volume of $`U(n)`$ is given by
$`\mathrm{vol}(U(n))={\displaystyle \frac{(2\pi )^{n(n+1)/2}}{_{k=0}^{n1}k!}}.`$ (132)
We rewrite the partition function in terms of the integration variables
$`x_k=2n\mathrm{\Sigma }^2\lambda _k^2.`$ (133)
This results in
$`Z_{N_f}^\nu (m)=C_n^{(3)}C_n^{(2)}m^{\nu N_f}{\displaystyle \frac{1}{n!}}{\displaystyle \underset{k=1}{\overset{n}{}}dx_kx_k^{\nu +N_f}\mathrm{\Delta }^2(\{x_k\})e^{x_k/2}},`$ (134)
where
$`C_n^{(3)}=(2n)^{[n(n+\nu )+nN_f]}\pi ^{n(n+\nu )}e^{nN_f}.`$ (135)
The integral over the $`x_k`$ was calculated by Forrester forrester-book . Our final result for the partition function is thus given by
$`Z_{N_f}^\nu (m)=m^{\nu N_f}\mathrm{\Sigma }^{2nN_f}C_n^{(3)}C_n^{(2)}2^{n(\nu +N_f+n)}{\displaystyle \underset{j=0}{\overset{n1}{}}}j!(\nu +N_f+j)!.`$ (136)
Collecting all factors we find the partition function
$`Z_{N_f}^\nu (m)=m^{\nu N_f}\mathrm{\Sigma }^{2nN_f}n^{nN_f}e^{nN_f}{\displaystyle \underset{j=0}{\overset{n1}{}}}{\displaystyle \frac{(\nu +N_f+j)!}{(\nu +j)!}}.`$ (137)
Using the Stirling formula, we find in the thermodynamic limit
$`Z_{N_f}^\nu (m)=m^{\nu N_f}\mathrm{\Sigma }^{2nN_f}{\displaystyle \frac{(2\pi )^{N_f/2}n^{N_f(N_f+2\nu )/2}}{_{k=0}^{N_f1}(\nu +k)!}}.`$ (138)
For $`N_f=1`$ we find the ratio
$`{\displaystyle \frac{Z_{N_f=1}^{\nu =1}(m)}{Z_{N_f=1}^{\nu =0}(m)}}=m(n+1).`$ (139)
According to (44) the condensate is given by
$`\underset{V\mathrm{}}{lim}\underset{m0}{lim}{\displaystyle \frac{1}{mV}}\left[{\displaystyle \frac{Z_{N_f=1}^{\nu =1}(m)}{Z_{N_f=1}^{\nu =0}(m)}}+{\displaystyle \frac{Z_{N_f=1}^{\nu =1}(m)}{Z_{N_f=1}^{\nu =0}(m)}}\right].`$ (140)
In random matrix theory the volume is identified as $`V=2n`$ such that (139) results in a condensate equal to unity. The reason for this incorrect result is that we have evaluated the ratio of two products with the same number of nonzero eigenvalues. To obtain a result with the correct dimensionality we have to regularize the determinants by only including eigenvalues below a given energy <sup>2</sup><sup>2</sup>2For overlap fermions neuberger achieved by pairing the zero eigenvalues with the largest eigenvalues. In our case, the largest eigenvalue is equal to $`1/\mathrm{\Sigma }`$. If we extend the Dirac operator for $`n1`$ modes and $`\nu =1`$ with this eigenvalue, we find the ratio (for $`N_f=1`$) $`Z_{N_f=1}^{\nu =1}(m)|_{nn1}/Z_{N_f=1}^{\nu =0}(m)|_n=mn\mathrm{\Sigma },`$ where we have used (138). This result also explains that in the case of overlap fermions the correct chiral condensate is obtained in the Schwinger model if the limit $`m0`$ is taken first lat-durr . . For $`\nu =1`$ we have on average one eigenvalue less with absolute value below a given energy than for $`\nu =0`$. This results in an overall factor of $`1/p\mathrm{\Delta }`$, where $`p`$ is the number of eigenvalues below the cut-off energy and $`\mathrm{\Delta }`$ is the spacing between the eigenvalues. Instead of using a fixed energy cut-off, we calculate the ratio of the partition functions for a fixed total number of eigenvalues meaning that in the sector of topological charge $`\nu `$ we make the replacement $`nn\nu /2`$ in (138). We finally find the ratio
$`{\displaystyle \frac{Z_{N_f}^\nu (m)}{Z_{N_f}^0(m=0)}}=(mn\mathrm{\Sigma })^{\nu N_f}{\displaystyle \underset{k=0}{\overset{N_f1}{}}}{\displaystyle \frac{k!}{(k+\nu )!}},`$
(141)
in agreement with general arguments given in LS . This result also follows from the fact that the large $`N`$ limit of the random matrix model is given by the nonlinear sigma model (115) SV . As was already found in SV , the result (141) confirms that the chiral anomaly is contained in random matrix theory.
Since the finite volume partition function for $`N_f=1`$ is a Bessel function, it is tempting to identify its zeros with the average position of the eigenvalues. In the random matrix model it can be easily checked whether this identification is justified. Numerically, it turns out that the ratio obtained from the average eigenvalues is a factor $`2/\pi `$ smaller than the result given in (139). A second question we have asked is whether the average determinant is mainly given by the product of the average eigenvalues or whether it is due to the fluctuations of the eigenvalues. It turns out that the product of the average eigenvalues (with the determinant included in the weight) is an order of magnitude larger than the average determinant.
## VIII Discussion
### VIII.1 Spectral Duality
The chiral condensate can be obtained in two ways. First, by taking the chiral limit after the thermodynamic limit,
$`\mathrm{\Sigma }^{(1)}=\underset{m0}{lim}\underset{V\mathrm{}}{lim}{\displaystyle \frac{1}{V}}{\displaystyle \frac{1}{i\lambda _k+m}},`$ (142)
and, second, by inverting the two limits,
$`\mathrm{\Sigma }^{(2)}=\underset{V\mathrm{}}{lim}\underset{m0}{lim}{\displaystyle \frac{1}{V}}{\displaystyle \frac{1}{i\lambda _k+m}}.`$ (143)
In the first case, the condensate arises as a consequence of spontaneous symmetry breaking, and in the second case it is due to the anomaly or instantons. The two condensates are not necessarily equal. For example, if we restrict the partition function to gauge field configurations with zero topological charge, the second definition gives zero. By spectral duality we mean that the two condensates are equal for the $`\theta `$ \- vacuum sum-poul . In the second case, when the contribution to the condensate comes from the sector with topological charge equal to one, the eigenvalues for $`\nu =1`$ are shifted such that the condensate is equal to the one obtained from the first definition. The first definition implies the Banks-Casher formula BC
$`\mathrm{\Sigma }^{(1)}={\displaystyle \frac{\pi \rho _0}{V}}.`$ (144)
This implies that the average spacing of the eigenvalues is given by
$`\mathrm{\Delta }={\displaystyle \frac{1}{\rho (0)}}={\displaystyle \frac{\pi }{\mathrm{\Sigma }V}}.`$ (145)
If we assume that the position of the eigenvalue $`\lambda _n`$ is given by $`n\mathrm{\Delta }`$, we obtain the sum rule
$`{\displaystyle \underset{n>0}{}}{\displaystyle \frac{1}{n^2\mathrm{\Delta }^2}}={\displaystyle \frac{1}{6}}\mathrm{\Sigma }^2V^2,`$ (146)
which explains the functional dependence of the sum rule.
As we will show next, the large $`\nu `$ limit of the sum rule can be obtained from the large $`\nu `$ limit of the average position of the eigenvalues. In the sector of topological charge $`\nu `$ the “average” position of the eigenvalues can be expressed in terms of the zeros, $`j_{\nu ,k}`$, of the Bessel functions LS ,
$`\lambda _k=\pm {\displaystyle \frac{j_{\nu ,k}}{\mathrm{\Sigma }V}}.`$ (147)
For large $`k`$ at fixed $`\nu `$ one one finds
$`\lambda _kj_{\nu ,k}^{\mathrm{as}}\mathrm{\Delta }(k+{\displaystyle \frac{\nu }{2}}\delta )\mathrm{\Delta }\mathrm{with}\delta ={\displaystyle \frac{1}{4}}.`$ (148)
Here, $`\mathrm{\Delta }1/\rho (0)`$ is the average spacing of the eigenvalues. This formula does not accurately give the position of the eigenvalues for large $`\nu `$ and finite $`k`$. In this case one can use a uniform asymptotic expansion of the zeros given by
$`j_{\nu ,k}\nu z_k,`$ (149)
with $`z_k`$ implicitly defined by $`(4k1)\pi =4\nu (\sqrt{z_k^21}\mathrm{arccos}(1/z_k)).`$ This results in the correct large $`\nu `$ limit of the sum rule
$`{\displaystyle \underset{k}{}}{\displaystyle \frac{1}{\lambda _k^2}}`$ $``$ $`\mathrm{\Sigma }^2V^2{\displaystyle _0^{\mathrm{}}}𝑑k{\displaystyle \frac{1}{j_{\nu ,k}^2}}={\displaystyle \frac{\mathrm{\Sigma }^2V^2}{\pi \nu }}{\displaystyle _1^{\mathrm{}}}𝑑z{\displaystyle \frac{\sqrt{z^21}}{z^3}}={\displaystyle \frac{\mathrm{\Sigma }^2V^2}{4\nu }}.`$ (150)
Let us now calculate the condensate according to the second definition
$`\mathrm{\Sigma }^{(2)}={\displaystyle \frac{1}{V}}{\displaystyle \frac{_k^{}(i\lambda _k+m)_{\nu =1}}{_k(i\lambda _k+m)_{\nu =0}}}+{\displaystyle \frac{1}{V}}{\displaystyle \frac{_k^{}(i\lambda _k+m)_{\nu =1}}{_k(i\lambda _k+m)_{\nu =0}}}.`$ (151)
Using the expression (13) for the partition function, this can be rewritten as
$`\mathrm{\Sigma }^{(2)}=\underset{m0}{lim}{\displaystyle \frac{1}{V}}\left[{\displaystyle \frac{I_1(mV\mathrm{\Sigma })/m}{I_0(mV\mathrm{\Sigma })}}+{\displaystyle \frac{I_1(mV\mathrm{\Sigma })/m}{I_0(mV\mathrm{\Sigma })}}\right]=\mathrm{\Sigma }.`$ (152)
The ratio in (151) can be interpreted as the ratio of the expectation values of the fermion determinant in the topological sector $`\nu =1`$ and $`\nu =0`$.
We will calculate the averages in (151) by replacing the eigenvalues according to (147)
$`\underset{m0}{lim}{\displaystyle \frac{1}{V}}{\displaystyle \frac{_k^{}(i\lambda _k+m)_{\nu =1}}{_k(i\lambda _k+m)_{\nu =0}}}`$ $``$ $`\underset{m0}{lim}{\displaystyle \frac{1}{V}}{\displaystyle \frac{_k^{}(ij_{\nu =1,k}/(\mathrm{\Sigma }V)+m)}{_k(ij_{\nu =0,k}/(\mathrm{\Sigma }V)+m)}}`$
$`=`$ $`{\displaystyle \frac{1}{V}}{\displaystyle \frac{_k^{}ij_{\nu =1,k}\mathrm{\Delta }/\pi }{_k^{}ij_{\nu =1,k}^{as}\mathrm{\Delta }/\pi }}{\displaystyle \frac{_kij_{\nu =0,k}^{\mathrm{as}}\mathrm{\Delta }/\pi }{_kij_{\nu =0,k}\mathrm{\Delta }/\pi }}{\displaystyle \frac{_k^{}ij_{\nu =1,k}^{\mathrm{as}}\mathrm{\Delta }/\pi }{_kij_{\nu =0,k}^{\mathrm{as}}\mathrm{\Delta }/\pi }}.`$
The first two ratios are finite, but the third ratio has to be regularized which we will do by a $`\zeta `$-function regularization (see Appendix B). For $`m=0`$ we find the result (see (201)
$`{\displaystyle \frac{1}{V}}{\displaystyle \frac{_k^{}ij_{\nu =1,k}^{\mathrm{as}}\mathrm{\Delta }/\pi }{_kij_{\nu =0,k}^{\mathrm{as}}\mathrm{\Delta }/\pi }}={\displaystyle \frac{1}{V\mathrm{\Delta }}}{\displaystyle \frac{\mathrm{\Gamma }^2(1\delta )}{\mathrm{\Gamma }^2(\frac{3}{2}\delta )}}={\displaystyle \frac{\mathrm{\Sigma }^{(1)}}{\pi }}{\displaystyle \frac{\mathrm{\Gamma }^2(1\delta )}{\mathrm{\Gamma }^2(\frac{3}{2}\delta )}}.`$ (154)
The ratios $`_k^{}ij_{\nu ,k}/_k^{}ij_{\nu =1,k}^{as}`$ can be easily evaluated numerically. We find
$`{\displaystyle \frac{_kij_{\nu =0,k}}{_kij_{\nu =0,k}^{as}}}=\mathrm{\Gamma }^2({\displaystyle \frac{3}{4}})/\sqrt{2},{\displaystyle \frac{_k^{}ij_{\nu =1,k}}{_k^{}ij_{\nu =1,k}^{as}}}=\pi \mathrm{\Gamma }^2(1.25)/2\sqrt{2}.`$ (155)
Therefore,
$`\mathrm{\Sigma }^{(2)}=\mathrm{\Sigma }^{(1)}.`$ (156)
Another way of evaluating the ratio (154) is from the limit
$`\underset{n\mathrm{}}{lim}{\displaystyle \frac{1}{n}}{\displaystyle \frac{_{k=n}^{n}{}_{}{}^{}ij_{\nu =1,k}^{\mathrm{as}}\mathrm{\Delta }/\pi }{_{k=n}^nij_{\nu =0,k}^{\mathrm{as}}\mathrm{\Delta }/\pi }}=\underset{n\mathrm{}}{lim}{\displaystyle \frac{1}{n}}\left[{\displaystyle \frac{_{k=0}^{n1}(k+\frac{5}{4})}{_{k=0}^{n1}(k+\frac{3}{4})}}\right]^{1/2}`$ (157)
Using the infinite product representation of the $`\mathrm{\Gamma }`$-function we find that this limit is given by $`\mathrm{\Gamma }^2(\frac{3}{4})/\mathrm{\Gamma }^2(\frac{5}{4}).`$ We thus have
$`{\displaystyle \frac{_k^{}ij_{\nu =1,k}^{\mathrm{as}}\mathrm{\Delta }/\pi }{_kij_{\nu =0,k}^{\mathrm{as}}\mathrm{\Delta }/\pi }}|_\zeta ={\displaystyle \frac{1}{\mathrm{\Delta }}}\underset{n\mathrm{}}{lim}{\displaystyle \frac{1}{n}}{\displaystyle \frac{_{k=0}^{n}{}_{}{}^{}ij_{\nu =1,k}^{\mathrm{as}}\mathrm{\Delta }/\pi }{_{k=0}^nij_{\nu =0,k}^{\mathrm{as}}\mathrm{\Delta }/\pi }},`$ (158)
so that the definition (157) of the infinite product would have resulted in an incorrect value for the chiral condensate. This explains why in the random matrix calculation of previous section the incorrect chiral condensate is obtained from (139). The fact that the two results differ by the eigenvalue at the cut-off is in agreement with the interpretation that $`\zeta `$-function regularization includes the product of the eigenvalues up to a fixed energy.
### VIII.2 Random Gauge Field
.
In the condensed matter literature Dirac fermions in random gauge fields have received a considerable amount of attention. Among the different types of models that have been considered, we mention the random magnetic field problem Casher:1984tb , the random flux model simons , the random mass model ludwig and the random gauge field model ludwig . In this section we only consider the latter model which is the quenched Schwinger model with the gauge field action is replaced by
$`{\displaystyle \frac{1}{4}}F_{\mu \nu }^2{\displaystyle \frac{1}{2}}{\displaystyle \frac{g^2}{\sigma ^2}}A_\mu A_\mu .`$ (159)
The factor $`g^2`$ is introduced to have the same normalization as in ludwig . Also in this case we can make a Hodge decomposition of the gauge field bernard ,
$`A_\mu =ϵ_{\mu \nu }_\nu \varphi +_\mu \mathrm{\Lambda }.`$ (160)
Let us consider the action of an instanton configuration given by
$`\varphi ={\displaystyle \frac{1}{2g}}\mathrm{log}(x^2+\rho ^2).`$ (161)
In the Schwinger model the action of this configuration is given by
$`S_{\mathrm{inst}}={\displaystyle \frac{1}{2}}{\displaystyle d^2xF_{01}^2}={\displaystyle \frac{1}{2}}{\displaystyle d^2x^2\varphi ^2\varphi }={\displaystyle \frac{\pi }{3g^2\rho ^2}},`$ (162)
which is infrared finite. Also the anomalous contribution to the effective action in the Schwinger model is infrared finite,
$`{\displaystyle d^2x\frac{\mu ^2}{2}\varphi ^2\varphi }={\displaystyle \frac{1}{2}}\left(\mathrm{log}\rho ^2+1\right).`$ (163)
For the random gauge model, on the other hand, we obtain the action
$`{\displaystyle \frac{1}{2}}{\displaystyle \frac{g^2}{\sigma ^2}}{\displaystyle _{|x|<R}}d^2xA_\mu ^2`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \frac{g^2}{\sigma ^2}}{\displaystyle _{|x|<R}}d^2x_\mu \varphi _\mu \varphi `$ (164)
$`=`$ $`{\displaystyle \frac{\pi }{2\sigma ^2}}\left[\mathrm{log}({\displaystyle \frac{R^2+\rho ^2}{\rho ^2}}){\displaystyle \frac{R^2}{R^2+\rho ^2}}\right].`$
which diverges for $`R\mathrm{}`$ which implies that instantons are suppressed in the thermodynamic limit. Therefore the chiral condensate of the massless theory is zero, and by spectral duality, the eigenvalue density of the Dirac operator vanishes independent of the topological charge. For $`\sigma ^2=\pi `$ the suppression factor is $`1/V`$, which exactly as in the Schwinger model for $`N_f=1`$, results in a finite spectral density for $`E0`$. Notice that the difference between the anomalous contribution in effective action of the Schwinger model and the random gauge action is only in the boundary term, which diverges for $`R\mathrm{}`$.
Let us evaluate the spectral density of a random gauge model with $`N_f`$ massless flavors and $`n`$ flavors with mass $`m`$. Because of the above remark, we restrict ourselves to the trivial topological charge sector, where the boundary term that gives the diverging result in (164) is absent. Using that the pure gauge term decouples from the partition function, the effective action of this model is given by
$`S_{\mathrm{RG}}={\displaystyle d^2x\left[\frac{1}{2}\varphi [\frac{g^2}{\sigma ^2}\mathrm{\Delta }+\mu ^2(N_f+n)\mathrm{\Delta }]\varphi i\frac{\theta g}{2\pi }\mathrm{\Delta }\varphi +\underset{\alpha }{}\overline{\psi }_\alpha \left[i\overline{)}+me^{2g\gamma _5\varphi }\right]\psi _\alpha \right]}.`$ (165)
The massless fermion fields result in an overall constant which can be ignored. The action of the massive fermion fields will be replaced by their bosonized form given by
$`S_\mathrm{F}(\varphi ,\phi _\alpha )={\displaystyle \frac{1}{2}}{\displaystyle \underset{\alpha =1}{\overset{n}{}}}(_\mu \phi _\alpha )^2cm{\displaystyle \underset{\alpha =1}{\overset{n}{}}}\mathrm{cos}(2\phi _\alpha \sqrt{\pi }2ig\varphi ).`$ (166)
If we define
$`\mathrm{\Gamma }(\varphi )={\displaystyle \frac{1}{2}}\varphi ({\displaystyle \frac{g^2}{\sigma ^2}}^2+\mu ^2(N_f+n)^2)\varphi i{\displaystyle \frac{\theta g}{2\pi }}\mathrm{\Delta }\varphi ,`$ (167)
the total bosonic action is given by
$`S_{\mathrm{RG}}={\displaystyle d^2x[\mathrm{\Gamma }(\varphi )+S_F(\varphi ,\phi _\alpha )]}.`$ (168)
After shifting the $`\phi _\alpha `$ fields by
$`\phi _\alpha \phi _\alpha +i\mu \varphi {\displaystyle \frac{\theta }{2\sqrt{\pi }}},`$ (169)
we obtain the action density
$`{\displaystyle \frac{1}{2}}{\displaystyle \underset{\alpha =0}{\overset{n}{}}}(_\mu \phi _\alpha )^2+{\displaystyle \frac{1}{2}}({\displaystyle \frac{g^2}{\sigma ^2}}+\mu ^2N_f)(_\mu \varphi )^2cm{\displaystyle \underset{\alpha =1}{\overset{n}{}}}\mathrm{cos}(2\phi _\alpha \sqrt{\pi }\theta )i\mu _\mu \varphi {\displaystyle \underset{\alpha =1}{\overset{n}{}}}_\mu \phi _\alpha .`$ (170)
The final action density is obtained after performing the Gaussian integral over $`\varphi `$,
$`{\displaystyle \frac{1}{2}}\zeta {\displaystyle \underset{\alpha =1}{\overset{n}{}}}{\displaystyle \underset{\beta =1}{\overset{n}{}}}_\mu \phi _\alpha _\mu \phi _\beta +{\displaystyle \frac{1}{2}}{\displaystyle \underset{\alpha =1}{\overset{n}{}}}(_\mu \phi _\alpha )^2cm{\displaystyle \underset{\alpha =1}{\overset{n}{}}}\mathrm{cos}(2\phi _\alpha \sqrt{\pi }\theta ),`$ (171)
where
$`\zeta ={\displaystyle \frac{\mu ^2}{\mu ^2N_f+g^2/\sigma ^2}}.`$ (172)
Considering the cosine term as a vertex, the propagator of the $`\phi _\alpha `$ fields is given by
$`G_{\alpha \beta }(p)={\displaystyle \frac{1}{p^2}}(\delta _{\alpha \beta }{\displaystyle \frac{\zeta }{1+n\zeta }}).`$ (173)
Using this result one can easily derive the dependence of the chiral condensate on the infrared cutoff which is taken to be equal to the size of the box. Following the derivation of kenway and using the lattice spacing $`a`$ as ultraviolet cutoff we find for $`a/L1`$ that
$`G_{\alpha \beta }(|x|0)={\displaystyle \frac{1}{2\pi }}(\delta _{\alpha \beta }{\displaystyle \frac{\zeta }{1+n\zeta }})\mathrm{log}{\displaystyle \frac{a}{L}}`$ (174)
The chiral condensate is given by
$`\mathrm{\Sigma }`$ $`=`$ $`{\displaystyle \frac{1}{n}}{\displaystyle \frac{1}{V}}_m\mathrm{log}Ze^{2i\sqrt{\pi }\phi _1}`$ (175)
$`=`$ $`e^{2\pi G_{11}(|x|0)}\left({\displaystyle \frac{a}{L}}\right)^{1\zeta /(1+n\zeta )}`$
In the replica limit, $`n0`$, we obtain the condensate
$`\mathrm{\Sigma }{\displaystyle \frac{1}{L^{1\zeta }}},`$ (176)
According to the Banks-Casher the smallest nonzero eigenvalue can be estimated as
$`\lambda _{\mathrm{min}}{\displaystyle \frac{\pi }{\mathrm{\Sigma }V}}{\displaystyle \frac{1}{L^{1+\zeta }}}.`$ (177)
On the other hand, if the eigenvalue density is given by
$`\rho (\lambda )V\lambda ^\alpha ,`$ (178)
the smallest nonzero eigenvalue follows from
$`{\displaystyle _0^{\lambda _{\mathrm{min}}}}\rho (\lambda )𝑑\lambda 1.`$ (179)
Therefore,
$`\lambda _{\mathrm{min}}{\displaystyle \frac{1}{L^{2/(\alpha +1)}}},`$ (180)
so that
$`\alpha ={\displaystyle \frac{1\zeta }{1+\zeta }}.`$ (181)
This result has also been derived from a more sophisticated renormalization group analysis ludwig ; fukui ; carp . For the action (171) we obtain
$`\alpha ={\displaystyle \frac{g^2+\mu ^2\sigma ^2(N_f1)}{g^2+\mu ^2\sigma ^2(N_f+1)}}={\displaystyle \frac{1+\sigma ^2(N_f1)/\pi }{1+\sigma ^2(N_f+1)/\pi }}.`$ (182)
In the quenched case ($`N_f=0`$), this result agrees with the result obtained in ludwig , and in the limit $`\sigma \mathrm{}`$ we recover the result for the $`N_f`$ flavor Schwinger model smcon ; smvac . For $`N_f=1`$ the spectral density is qualitatively different from the Schwinger model. A diverging spectral density for $`N_f=0`$ has been observed in recent lattice simulations lat-poul . If the spectral density vanishes, it still possible to derive sum rules for the inverse Dirac eigenvalues poul-multi ; janik ; gernot-multi ; janik-multi . It could be that such sum rules are easier obtained for lattice simulation of random Dirac fermions than for lattice simulations of the Schwinger model with two massless flavors lat-poul .
A special case where the sum rules for the inverse Dirac eigenvalues can be evaluated is $`N_f=0`$ and $`n=1`$. Since the sum rule follows from the coefficient of $`m^2`$ in the expansion of the partition function in powers of $`m`$, the spectral density corresponding to this sum rule is the one of the Schwinger model with one massless flavor.
For $`n=1`$ and $`N_f=0`$ the partition function after the transformation $`\phi _1\mu \varphi `$ is given by
$`Z(m,\theta )={\displaystyle D\phi e^{\frac{1}{2}F^2{\scriptscriptstyle d^2x\varphi \mathrm{\Delta }\varphi }+cm\frac{1}{2}{\scriptscriptstyle d^2x[e^{i\theta +2ig\varphi \left(x\right)}+e^{i\theta 2ig\varphi \left(x\right)}]}}},`$
(183)
where
$`F^2=\mu ^2+{\displaystyle \frac{\mu ^2\sigma ^2}{\pi }}.`$ (184)
Let us evaluate the simplest sum rule in the sector of zero topological charge. To this end we expand the partition function to second order in $`m`$ and integrate over $`\theta `$. We obtain
$`_m^2\mathrm{log}Z^{\nu =0}(m)={\displaystyle d^2xd^2ye^{4g^2G^{\mathrm{RG}}(x,y)4g^2G^{\mathrm{RG}}(0,0)}},`$ (185)
where the random gauge field Green’s function is given by
$`G^{\mathrm{RG}}(x,y)={\displaystyle \frac{1}{4\pi F^2}}\mathrm{log}(xy)^2.`$ (186)
The volume dependence of the sum rule is thus given by
$`_m^2\mathrm{log}Z^{\nu =0}(m)V^{2\mu ^2/F^2}.`$ (187)
If the eigenvalue density scales as
$`\rho (E)VE^\alpha ,`$ (188)
the volume dependence of the sum rule is given by
$`{\displaystyle \underset{k}{}}{\displaystyle \frac{1}{\lambda _k^2}}V^{\frac{2}{1+\alpha }},`$ (189)
so that the exponent $`\alpha `$ is equal to
$`\alpha ={\displaystyle \frac{\mu ^2}{2F^2\mu ^2}}={\displaystyle \frac{1}{1+2\sigma ^2/\pi }}.`$ (190)
Indeed, this result agrees with (182) for $`N_f=1`$.
## IX Conclusions
In conclusion, we have analyzed Leutwyler-Smilga sum rules in the Schwinger model and found that they are in agreement with the universal result from chiral perturbation theory and random matrix theory. Instead of relying on general symmetry arguments, we have performed a microscopic calculation of the sum of the inverse square Dirac eigenvalues in the sector of arbitrary topological charge $`\nu `$, generalizing a result by Smilga obtained in the sector of zero topological charge.
We have shown that validity of the sum rules follows from the clustering property of the scalar correlation functions. This argument also applies to QCD with one flavor. For QCD with several flavors, the naive clustering argument fails due to the presence of Goldstone bosons. A modified clustering property is obtained from the chiral Lagrangian that corresponds to the low energy limit of QCD partition function.
The dependence of the sum rules on the topological charge $`\nu `$ is consistent with a shift of the Dirac eigenvalues by $`\nu /2`$ times the average level spacing. Such shift of the Dirac spectrum exactly results in the chiral condensate for massless quarks in the sector of topological charge $`\nu =1`$. However, obtaining this result, requires a regularization of the fermion determinant with a fixed energy cutoff such as for example the $`\zeta `$-function regularization.
In the microscopic derivation, the sum rules where obtained from more general sum rules valid for a fixed external gauge field with topological charge $`\nu `$. The most general sum rules in this class were conjectured based a simplified derivation of Leutwyler-Smilga sum rules starting from the bosonized Schwinger model. It would be interesting to probe such sum rules directly in lattice simulations.
The Schwinger model is part of a larger class of models known under the name of random Dirac fermions which have been investigated in the context of the quantum Hall effect. The key difference between the two models is that instantons are suppressed by the random gauge field action. Therefore, the chiral condensate of the massless theory vanishes in the domain below a critical value of the disorder. In these models we expect modified sum rules that would be a useful probe for the scaling behavior of the density of states near zero.
Acknowledgements. We would like to thank Andreas Ludwig, Andreas Smilga and Pierre van Baal for useful discussions. This work was supported in part by US DOE Grant DE-FG-88ER40388.
## X Appendix A
In this appendix we calculate the integral given in (121). We start from the result creutz ,
$`W(J)`$ $`=`$ $`{\displaystyle _{SU(N_f)}}𝑑U\mathrm{exp}(Tr(JU))`$ (191)
$`=`$ $`{\displaystyle \underset{i=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{2!\mathrm{}(N_f1)!}{i!\mathrm{}(i+N_f1)!}}\left(detJ\right)^i.`$
We are interested in the integral
$`{\displaystyle _{SU(N_f)}}𝑑UU_{11}^\nu U_{22}^\nu \mathrm{}U_{N_fN_f}^\nu =\left[\left({\displaystyle \frac{d}{dJ_{11}}}\right)^\nu \mathrm{}\left({\displaystyle \frac{d}{dJ_{N_fN_f}}}\right)^\nu W(J)\right]|_{J=0}.`$ (192)
On the other hand, we can choose $`J=\mathrm{diag}[J_{11},J_{22},\mathrm{},J_{N_fN_f}]`$, so that
$`detJ=J_{11}J_{22}\mathrm{}J_{N_fN_f}.`$ (193)
Using this result, we arrive at $`N_f`$ independent differentiations over diagonal matrix elements of $`J`$. The only term in the sum that contributes is the one with $`i=\nu `$ resulting in
$`{\displaystyle _{SU(N_f)}}𝑑UU_{11}^\nu U_{22}^n\mathrm{}U_{N_fN_f}^\nu `$ $`=`$ $`\left[\nu !\right]^{N_f}{\displaystyle \frac{2!\mathrm{}(N_f1)!}{\nu !\mathrm{}(\nu +N_f1)!}}`$ (194)
$`=`$ $`\left[\nu !\right]^{N_f}{\displaystyle \underset{i=0}{\overset{N_f1}{}}}{\displaystyle \frac{i!}{(i+\nu )!}}.`$
## XI Appendix B
In this appendix we compute the determinant of the Dirac operator in $`\zeta `$-function regularization with eigenvalues given by the asymptotic values of the zeros of Bessel functions,
$`j_{\nu ,k}^{\mathrm{as}}=k+{\displaystyle \frac{\nu }{2}}\delta \mathrm{with}\delta ={\displaystyle \frac{1}{4}}.`$ (195)
In a $`\zeta `$ function regularization the determinant of the Dirac operator in the sector of topological charge $`\nu `$ is given by
$`\mathrm{log}{\displaystyle \underset{k}{}^{}}(i\mathrm{\Delta }j_{\nu ,k}^{\mathrm{as}}/\pi +m)={\displaystyle \frac{d}{ds}}|_{s=0}\left[{\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{(k+\frac{\nu }{2}\delta +im/\mathrm{\Delta })^s\mathrm{\Delta }^s}}+{\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{(k+\frac{\nu }{2}\delta im/\mathrm{\Delta })^s\mathrm{\Delta }^s}}\right].`$
The function in between the brackets is known as the Hurwitz zeta function,
$`\zeta (s,a){\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{(k+a)^s}}.`$ (197)
For convenience we only calculate the product for $`m=0`$. For $`\nu =1`$ we find
$`\mathrm{log}{\displaystyle \underset{k}{}^{}}i{\displaystyle \frac{\mathrm{\Delta }}{\pi }}j_{\nu =1,k}^{\mathrm{as}}`$ $`=`$ $`2{\displaystyle \frac{d}{ds}}|_{s=0}[\zeta (s,{\displaystyle \frac{3}{2}}\delta )\mathrm{\Delta }^s]`$ (198)
$`=`$ $`2\zeta ^{}(0,{\displaystyle \frac{3}{2}}\delta )+2\zeta (0,{\displaystyle \frac{3}{2}}\delta )\mathrm{log}\mathrm{\Delta }.`$
Combining this with the result for $`\nu =0`$,
$`\mathrm{log}{\displaystyle \underset{k}{}^{}}i{\displaystyle \frac{\mathrm{\Delta }}{\pi }}j_{\nu =0,k}^{\mathrm{as}}`$ $`=`$ $`2{\displaystyle \frac{d}{ds}}|_{s=0}[\zeta (s,1\delta )\mathrm{\Delta }^s]`$ (199)
$`=`$ $`2\zeta ^{}(0,1\delta )+2\zeta (0,1\delta )\mathrm{log}\mathrm{\Delta },`$
and using that
$`\zeta (0,a)={\displaystyle \frac{1}{2}}a,\zeta ^{}(0,a)=\mathrm{log}\mathrm{\Gamma }({\displaystyle \frac{1}{2}}\mathrm{log}(2\pi ),`$ (200)
we find the ratio,
$`{\displaystyle \frac{1}{V}}{\displaystyle \frac{_k^{}i\frac{\mathrm{\Delta }}{\pi }j_{\nu =1,k}^{\mathrm{as}}}{_ki\frac{\mathrm{\Delta }}{\pi }j_{\nu =0,k}^{\mathrm{as}}}}={\displaystyle \frac{1}{V\mathrm{\Delta }}}{\displaystyle \frac{\mathrm{\Gamma }^2(1\delta )}{\mathrm{\Gamma }^2(\frac{3}{2}\delta )}}={\displaystyle \frac{\mathrm{\Sigma }^{(1)}}{\pi }}{\displaystyle \frac{\mathrm{\Gamma }^2(1\delta )}{\mathrm{\Gamma }^2(\frac{3}{2}\delta )}}.`$ (201) |
warning/0507/math0507579.html | ar5iv | text | # Estimates of potential kernel and Harnack’s inequality for anisotropic fractional Laplacian
## 1 Main results and background
Let $`\alpha (0,2)`$ and $`d\{1,2,\mathrm{}\}`$. We consider an arbitrary Lévy measure on $`^d\{0\}`$ which is symmetric, homogeneous: $`\nu (rB)=r^\alpha \nu (B)`$, and nondegenerate (for definitions see Section 2). $`\nu `$ yields a convolution semigroup of probability measures $`\{P_t,t>0\}`$ on $`^d`$. Each $`P_t`$ has a smooth density $`p_t`$. We consider the corresponding potential measure $`𝕍=_0^{\mathrm{}}P_t𝑑t`$ and the potential kernel
$$V(x)=_0^{\mathrm{}}p_t(x)𝑑t,x^d.$$
$`V(x)=|x|^{\alpha d}V(x/|x|)`$, but it may be infinite in some directions (\[18, pp. 148-149\]). It is of interest to study continuity of $`V`$ on the unit sphere $`𝕊`$ in $`^d`$ under specific assumptions on $`\nu `$ (see (13)).
###### Theorem 1
If $`d>\alpha `$ and $`\nu `$ is a $`\gamma `$-measure on $`𝕊`$ with $`\gamma >d2\alpha `$ then $`V`$ is continuous on $`𝕊`$.
The following partial converse shows that the threshold $`d2\alpha `$ is exact.
###### Theorem 2
If $`𝕍`$ is a $`\kappa `$-measure on $`𝕊`$ then $`\nu `$ is a $`(\kappa 2\alpha )`$-measure on $`𝕊`$.
In particular, if $`V`$ is bounded on $`𝕊`$ then $`\nu `$ is a $`(d2\alpha )`$–measure on $`𝕊`$.
We define an operator $`𝒜`$ on smooth functions $`\phi `$ with compact support in $`^d`$, $`\phi C_c^{\mathrm{}}(^d)`$, by
$`𝒜\phi (x)`$ $`=`$ $`{\displaystyle \underset{^d}{}}\left(\phi (x+y)\phi (x)y\phi (x)\mathbf{\hspace{0.33em}1}_{|y|<1}\right)\nu (dy)`$
$`=`$ $`\underset{\epsilon 0^+}{lim}{\displaystyle \underset{|y|>\epsilon }{}}\left(\phi (x+y)\phi (x)\right)\nu (dy).`$
$`𝒜`$ is a restriction of the infinitesimal generator of $`\{P_t\}`$ \[35, Example 4.1.12\], and what we refer to as the anisotropic fractional Laplacian in the title of the paper. In this connection we recall that in the special case of $`\nu (dy)=c|y|^{d\alpha }dy`$ one obtains the fractional Laplacian $`\mathrm{\Delta }^{\alpha /2}`$. For properties of $`\mathrm{\Delta }^{\alpha /2}`$ and a discussion of equivalent definitions of its harmonic functions we refer the reader to .
Harmonic functions corresponding to $`𝒜`$, or $`\nu `$, are defined by the mean value property with respect to an appropriate family of harmonic measures, see Section 4. The main goal of the paper is to characterize those operators $`𝒜`$ for which Harnack’s inequality holds, i.e., there is a constant $`C=C(\alpha ,\nu )`$ such that for every function $`u`$ which is harmonic in the unit ball and nonnegative in $`^d`$
(1)
$$u(x_1)Cu(x_2),|x_1|<1/2,|x_2|<1/2.$$
To this end we use the relative Kato condition (RK) meaning that there is a constant $`K`$ such that
(2)
$$\underset{B(y,1/2)}{}|yv|^{\alpha d}\nu (dv)K\nu (B(y,1/2)),y^d.$$
###### Theorem 3
Harnack’s inequality holds for $`𝒜`$ if and only if (RK) holds for $`\nu `$.
Theorem 3 is a strengthening of \[18, Theorem 1\], where an additional technical assumption was made: $`\nu (dy)c|y|^{d\alpha }dy`$, to guarantee the boundedness of $`V`$ on $`𝕊`$. We now drop the assumption and the boundedness is obtained as the sole consequence of (2) via Theorem 1. We also adapt some of our previous techniques from to handle measures $`\nu `$ which are not absolutely continuous with respect to the Lebesgue measure on $`^d`$ (see, e.g., (27)).
Our estimates of the semigroup in Section 3 are based in part on ideas of , which concerns more complicated non-convolutional semigroups. Another, recent paper gives involved estimates of our convolution semigroup $`\{P_t\}`$ in individual directions (see also in this connection). Here we only need isotropic estimates of $`\{P_t\}`$ from above, and our considerations become simpler than those of and .
In Section 4, 5 and 6 we develop the methods of . That (2) implies (1) is proved by using a maximum principle for a Dynkin-type version of the operator $`𝒜`$ to explicitly estimate its Green function $`G(x,v)`$ for the unit ball, see Proposition 1 below. Noteworthy, our proof of the estimate is exclusive to non-local operators, of which $`𝒜`$ is an illustrative special case. In particular it turns out that $`G(x,v)`$ has the singularity at the pole comparable to that of the Riesz kernel: $`|vx|^{\alpha d}`$. The singularity influences the magnitude of the corresponding Poisson kernel of the ball, $`P(x,y)`$, as given by the Ikeda-Watanabe formula (27). The influence is critical if and only if (2) fails to hold. This relates (2) to (1). Such a direct influence of the singularity of the potential kernel on the Poisson kernel does not occur for second order elliptic operators, which is why we can expect analogues of Theorem 3 only for nonlocal operators.
The recent development in the study of Harnack’s inequality for general integro-differential operators similar to $`𝒜`$ was initiated in , see also . The class of considered operators gradually extended, see , , , , , and the references given there. We note that the operators dealt with in these papers are not translation invariant nor are they homogeneous. On the other hand the papers focus on sufficient conditions for Harnack’s inequality and they are restricted by certain isotropic estimates of the operator’s kernel from below.
Our confinement to translation invariant homogeneous operators $`𝒜`$ results in part from the fact that the problem of the construction of the semigroup from a general nonlocal operator satisfying the positive maximum principle does not have a final solution yet. We refer the reader to , , , and . A general survey of the subject and more references can be found in . We refer the reader to for an account of the related potential theory of second order elliptic operators. We like to point out that while a symmetric second order elliptic operator with constant coefficients is merely a linear transformation of the Laplacian, the operators $`𝒜`$ and their harmonic functions considered here are very diverse ().
The remainder of the paper is organized as follows. First definitions are given in Section 2. In Section 3 we estimate the semigroup (see (17) below) and the potential measure $`𝕍`$ and we prove our first two theorems. In Section 4 we give preliminaries needed for the proof of Theorem 3, which is presented in Section 5 and 6. In Section 6 we also recall after two explicit examples to show how irregular the Lévy measure $`\nu `$ can be for Harnack’s inequality to hold or to fail for $`𝒜`$.
At the end of the paper we mention some remaining open problems.
## 2 Preliminaries
For $`x^d`$ and $`r>0`$ we let $`|x|=\sqrt{_{i=1}^dx_i^2}`$ and $`B(x,r)=\{y^d:|yx|<r\}`$. We denote $`𝕊=\{x^d:|x|=1\}`$. All the sets, functions and measures considered in the sequel will be Borel. For a measure $`\lambda `$ on $`^d`$, $`|\lambda |`$ denotes its total mass. For a function $`f`$ we let $`\lambda (f)=f𝑑\lambda `$, whenever the integral makes sense. When $`|\lambda |<\mathrm{}`$ and $`n=1,2,\mathrm{}`$ we let $`\lambda ^n`$ denote the $`n`$-fold convolution of $`\lambda `$ with itself:
$$\lambda ^n(f)=f(x_1+x_2+\mathrm{}+x_n)\lambda (dx_1)\lambda (dx_2)\mathrm{}\lambda (dx_n).$$
We also let $`\lambda ^0=\delta _0`$, the evaluation at $`0`$. We call $`\lambda `$ degenerate if there is a proper linear subspace $`M`$ of $`^d`$ such that $`supp(\lambda )M`$; otherwise we call $`\lambda `$ nondegenerate.
In what follows we will consider measures $`\mu `$ concentrated on $`𝕊`$. We will assume that $`\mu `$ is positive, finite, nondegenerate (in particular $`\mu 0`$), and symmetric:
$$\mu (D)=\mu (D),D^d.$$
We will call $`\mu `$ the spectral measure. We let
(3)
$$\nu (D)=_𝕊_0^{\mathrm{}}\mathrm{𝟏}_D(r\xi )r^{1\alpha }𝑑r\mu (d\xi ),D^d,$$
where $`\mathrm{𝟏}_D`$ is the indicator function of $`D`$. Note that $`\nu `$ is symmetric. It is a Lévy measure on $`^d`$, i.e.
$$_^d\mathrm{min}(|y|^2,\mathrm{\hspace{0.17em}1})\nu (dy)<\mathrm{}.$$
For $`r>0`$ and a function $`\phi `$ on $`^d`$ we consider its dilation $`\phi _r(y)=\phi (y/r)`$, and we note that $`\nu (\phi _r)=r^\alpha \nu (\phi )`$. In particular $`\nu `$ is homogeneous: $`\nu (rB)=r^\alpha \nu (B)`$ for $`B^d`$. Similarly, if $`\phi C_c^{\mathrm{}}(^d)`$, then $`𝒜(\phi _r)=r^\alpha (𝒜\phi )_r`$. This is the homogeneity of $`𝒜`$. In connection with the rest of our statement in Abstract we recall that every operator $`A`$ on $`C_c^{\mathrm{}}(^d)`$, which satisfies the positive maximum principle:
$$\underset{y^d}{sup}\phi (y)=\phi (x)0impliesA\phi (x)0,$$
is given uniquely in the form
$`A\phi (x)`$ $`=`$ $`{\displaystyle \underset{i,j=1}{\overset{d}{}}}a_{ij}(x)D_{x_i}D_{x_j}\phi (x)+b(x)\phi (x)c(x)\phi (x)`$
$`+{\displaystyle \underset{^d}{}}\left(\phi (x+y)\phi (x)y\phi (x)\mathbf{\hspace{0.33em}1}_{|y|<1}\right)\nu (x,dy).`$
Here $`y\phi `$ is the scalar product of $`y`$ and the gradient of $`\phi `$ and, for every $`x`$, $`a(x)=(a_{ij}(x))_{i,j=1}^n`$ is a nonnegative definite real symmetric matrix, the vector $`b(x)=(b_i(x))_{i=1}^d`$ has real coordinates, $`c(x)0`$, and $`\nu (x,)`$ is a Lévy measure. This description is due to Courrége, see \[33, Proposition 2.10\], \[49, Chapter 2\] or \[35, Chapter 4.5\]. For translation invariant operators $`A`$ the characteristics $`a`$, $`b`$, $`c`$, and $`\nu `$ are independent of $`x`$. If $`A`$ is symmetric:
$$_^dA\phi (x)\varphi (x)𝑑x=_^dA\varphi (x)\phi (x)𝑑x\text{for}\phi ,\varphi C_c^{\mathrm{}}(^d),$$
then $`b=0`$ and $`\nu `$ is necessarily symmetric (see, e.g., \[35, p. 251\] and \[33, Corollary 2.14\]). If $`A`$ is homogeneous but not a local operator () then $`a=0`$ and $`\nu `$ must be homogeneous, hence (3) holds with some $`\alpha (0,2)`$ (note that $`𝒜\phi (0)=\nu (\phi )`$ if $`\phi C_c^{\mathrm{}}(^d\{0\})`$).
We now construct the corresponding semigroup (for a more axiomatic introduction to convolution semigroups we refer the reader to ). For $`\epsilon >0`$ we let $`\widehat{\nu }_\epsilon =\mathrm{𝟏}_{B(0,\epsilon )^c}\nu `$, i.e. $`\widehat{\nu }_\epsilon (f)=\nu (\mathrm{𝟏}_{B(0,\epsilon )^c}f)`$, and we let $`\stackrel{~}{\nu }_\epsilon =\mathrm{𝟏}_{B(0,\epsilon )}\nu `$. We consider the probability measures
$`\widehat{P}_t^\epsilon `$ $`=`$ $`\mathrm{exp}(t(\widehat{\nu }_\epsilon |\widehat{\nu }_\epsilon |\delta _0))={\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{t^n(\widehat{\nu }_\epsilon |\widehat{\nu }_\epsilon |\delta _0))^n}{n!}}`$
$`=`$ $`e^{t|\widehat{\nu }_\epsilon |}{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{t^n\widehat{\nu }_\epsilon ^n}{n!}},t>0.`$
Here $`\widehat{\nu }_\epsilon ^n=(\widehat{\nu }_\epsilon )^n`$. $`\widehat{P}_t^\epsilon `$ form a convolution semigroup:
$$\widehat{P}_t^\epsilon \widehat{P}_s^\epsilon =\widehat{P}_{s+t}^\epsilon ,s,t>0.$$
The Fourier transform of $`\widehat{P}_t^\epsilon `$ is
$$(\widehat{P}_t^\epsilon )(u)=e^{iuy}\widehat{P}_t^\epsilon (dy)=\mathrm{exp}\left(t(e^{iuy}1)\widehat{\nu }_\epsilon (dy)\right),u^d.$$
The measures $`\widehat{P}_t^\epsilon `$ weakly converge to a probability measure $`P_t`$ as $`\epsilon 0`$ (this essentially depends on (6) below). $`\{P_t,t>0\}`$ is also a convolution semigroup and $`(P_t)(u)=\mathrm{exp}(t\mathrm{\Phi }(u))`$, where
$`\mathrm{\Phi }(u)`$ $`=`$ $`{\displaystyle \left(e^{iuy}1iuy\mathrm{𝟏}_{B(0,1)}(y)\right)\nu (dy)}`$
$`=`$ $`{\displaystyle \left(\mathrm{cos}(uy)1\right)\nu (dy)}={\displaystyle \frac{\pi }{2\mathrm{sin}\frac{\pi \alpha }{2}\mathrm{\Gamma }(1+\alpha )}}{\displaystyle _𝕊}|u\xi |^\alpha \mu (d\xi ).`$
Since $`\mu `$ is finite and nondegenerate,
(5)
$$\mathrm{\Phi }(u)=|u|^\alpha \mathrm{\Phi }(u/|u|)|u|^\alpha .$$
We call $`\nu `$ the Lévy measure of the semigroup $`\{P_t,t0\}`$ .
By a similar limiting procedure we construct the semigroup $`\{\stackrel{~}{P}_t^\epsilon ,t>0\}`$ such that
$$(\stackrel{~}{P}_t^\epsilon )(u)=\mathrm{exp}\left(t(e^{iuy}1iuy\mathrm{𝟏}_{B(0,1)}(y))\stackrel{~}{\nu }_\epsilon (dy)\right).$$
Note that
(6)
$$_^d|y|^2\stackrel{~}{P}_t^\epsilon (dy)=t_^d|y|^2\stackrel{~}{\nu }_\epsilon (dy).$$
The Lévy measures of $`\{\stackrel{~}{P}_t^\epsilon \}`$ and $`\{\widehat{P}_t^\epsilon \}`$ are $`\stackrel{~}{\nu }_\epsilon `$ and $`\widehat{\nu }_\epsilon `$, respectively, and we have
(7)
$$P_t=\stackrel{~}{P}_t^\epsilon \widehat{P}_t^\epsilon .$$
The measures $`P_t`$ and $`\stackrel{~}{P}_t^\epsilon `$ have rapidly decreasing Fourier transform hence they are absolutely continuous with bounded smooth densities denoted $`p_t(x)`$ and $`\stackrel{~}{p}_t^\epsilon (x)`$, respectively. Of course,
(8)
$$p_t=\stackrel{~}{p}_t^\epsilon \widehat{P}_t^\epsilon .$$
By using (5) we obtain the scaling property of $`\{p_t\}`$:
(9)
$$p_t(x)=t^{d/\alpha }p_1(t^{1/\alpha }x),x^d.$$
In particular,
(10)
$$p_t(x)ct^{d/\alpha }.$$
We define the potential measure of the semigroup $`\{P_t\}`$:
$$𝕍(D)=_0^{\mathrm{}}P_t(D)𝑑t,D^d.$$
By (10), $`𝕍`$ is finite on bounded subsets of $`^d`$ if $`d>\alpha `$. Let
(11)
$$V(x)=_0^{\mathrm{}}p_t(x)𝑑t,x^d,$$
so that
$$𝕍(D)=_DV(x)𝑑x,D^d.$$
We call $`V(x)`$ the potential kernel of the stable semigroup. By (9)
(12)
$$V(x)=|x|^{\alpha d}V\left(x/|x|\right),x0,$$
and $`𝕍(rD)=r^\alpha 𝕍(D)`$ for $`r>0`$, $`D^d`$.
If $`d=1`$ then up to a constant there is only one measure $`\nu `$ to consider: $`\nu (dy)=|y|^{1\alpha }dy`$, corresponding to $`𝒜=c\mathrm{\Delta }^{\alpha /2}`$. This case is not excluded from our considerations but it is sometimes trivial. In particular, if $`d=1\alpha `$ then $`V\mathrm{}`$ (\[7, Example 14.30\]). We refer to for more information and references on the case $`d=1\alpha `$.
Constants in this paper mean positive real numbers. We often write $`fg`$ to indicate that there is $`c=c(\alpha ,\mu )`$, i.e. a constant $`c`$ depending only on $`\alpha `$ and $`\mu `$, such that $`c^1fgcf`$.
## 3 Estimates of semigroup and potential measure
A general reference to the potential theory of convolution semigroups is (see also ).
We consider an auxiliary scale of smoothness for $`\nu `$.
###### Definition 1
We say that $`\nu `$ is a $`\gamma `$-measure on $`𝕊`$ if
(13)
$$\nu (B(x,r))cr^\gamma ,|x|=1,\mathrm{\hspace{0.33em}0}<r<1/2.$$
Since $`\nu (drd\theta )=r^{1\alpha }dr\mu (d\theta )`$, it is at least a $`1`$-measure and at most a $`d`$-measure on $`𝕊`$. If $`\nu `$ is a $`\gamma `$-measure with $`\gamma >1`$, then $`\mu `$ has no atoms. $`\nu `$ is a $`d`$-measure if and only if it is absolutely continuous with respect to the Lebesgue measure and has a density function which is locally bounded on $`^d\{0\}`$. We refer the reader to and for considerations related to this case.
In the remainder of this section we fix $`1\gamma d`$ and we assume that $`\nu `$ is a $`\gamma `$-measure on $`𝕊`$.
We will first estimate individual terms in the series in (2).
###### Lemma 1
There exists $`C=C(\alpha ,\mu )`$ such that for $`\epsilon >0`$ and $`n=1,2,\mathrm{}`$ we have
(14)
$$\widehat{\nu }_\epsilon ^n(B(x,r))C^nr^\gamma \epsilon ^{(n1)\alpha },|x|=1,$$
provided $`0<r<\mathrm{max}(\epsilon /3,1/5^n)`$.
Proof. We proceed by induction. Note that (14) holds for $`n=1`$ by (13). Let $`c_0`$ and $`n`$ be such that (14) is satisfied with $`C=c_0`$. We first assume that $`r<\epsilon /3`$. For every $`x𝕊`$ by homogeneity of $`\nu `$ and (13) we have
$`\widehat{\nu }_\epsilon ^{n+1}(B(x,r))`$ $`=`$ $`{\displaystyle \underset{|xy|>2\epsilon /3}{}}\widehat{\nu }_\epsilon (B(xy,r))\widehat{\nu }_\epsilon ^n(dy)`$
$``$ $`{\displaystyle \underset{|xy|>2\epsilon /3}{}}\nu (B(xy,r))\widehat{\nu }_\epsilon ^n(dy)`$
$`=`$ $`{\displaystyle \underset{|xy|>2\epsilon /3}{}}|xy|^\alpha \nu (B({\displaystyle \frac{xy}{|xy|}},{\displaystyle \frac{r}{|xy|}}))\widehat{\nu }_\epsilon ^n(dy)`$
$``$ $`c_1r^\gamma {\displaystyle \underset{|xy|>2\epsilon /3}{}}|xy|^{\alpha \gamma }\widehat{\nu }_\epsilon ^n(dy)`$
(note that $`r/|xy|<1/2`$ provided $`|xy|>2\epsilon /3`$). Now let $`\epsilon /3r<1/5^{n+1}`$. Then $`2r+\epsilon <1/5^n`$ and by induction
$`{\displaystyle \underset{|xy|<2r+\epsilon }{}}\widehat{\nu }_\epsilon (B(xy,r))\widehat{\nu }_\epsilon ^n(dy)`$ $``$ $`|\widehat{\nu }_\epsilon |\widehat{\nu }_\epsilon ^n(B(x,2r+\epsilon ))`$
$``$ $`{\displaystyle \frac{|\mu |}{\alpha }}\epsilon ^\alpha c_0^n(2r+\epsilon )^\gamma \epsilon ^{(n1)\alpha }`$
$``$ $`c_0^nc_2r^\gamma \epsilon ^{n\alpha },`$
for some $`c_2=c_2(\alpha ,\mu )`$ ; and by homogeneity of $`\nu `$ and (13) we get
$`{\displaystyle \underset{|xy|>2r+\epsilon }{}}\widehat{\nu }_\epsilon (B(xy,r))\widehat{\nu }_\epsilon ^n(dy)`$ $``$ $`{\displaystyle \underset{|xy|>2r+\epsilon }{}}\nu (B(xy,r))\widehat{\nu }_\epsilon ^n(dy)`$
$``$ $`{\displaystyle \underset{|xy|>2r+\epsilon }{}}c_1r^\gamma |xy|^{\alpha \gamma }\widehat{\nu }_\epsilon ^n(dy)`$
$``$ $`c_1r^\gamma {\displaystyle \underset{|xy|>2\epsilon /3}{}}|xy|^{\alpha \gamma }\widehat{\nu }_\epsilon ^n(dy).`$
From the above we have
(15)
$$\widehat{\nu }_\epsilon ^{n+1}(B(x,r))c_1r^\gamma \underset{|xy|>2\epsilon /3}{}|xy|^{\alpha \gamma }\widehat{\nu }_\epsilon ^n(dy)+c_0^nc_2r^\gamma \epsilon ^{n\alpha },$$
for all $`0<r<\mathrm{max}(\epsilon /3,1/5^{n+1})`$.
Let $`L_\epsilon =\mathrm{log}_5(3/2\epsilon )`$. If $`2\epsilon /3<1/5^n`$ then we get by induction
$$\begin{array}{ccc}\underset{2\epsilon /3<|xy|<1/5^n}{}|xy|^{\alpha \gamma }\widehat{\nu }_\epsilon ^n(dy)\hfill & & \underset{k=n}{\overset{L_\epsilon }{}}\underset{1/5^{k+1}<|xy|<1/5^k}{}|xy|^{\alpha \gamma }\widehat{\nu }_\epsilon ^n(dy)\hfill \\ & & \underset{k=n}{\overset{L_\epsilon }{}}(5^{k+1})^{\alpha +\gamma }\widehat{\nu }_\epsilon ^n(B(x,1/5^k))\hfill \\ & & c_0^n5^{\alpha +\gamma }\epsilon ^{(n1)\alpha }\underset{k=1}{\overset{L_\epsilon }{}}5^{k\alpha }\hfill \\ & & c_0^nc_3\epsilon ^{n\alpha },\hfill \end{array}$$
where $`c_3=c_3(\alpha ,\mu )`$. Also,
$$\begin{array}{ccc}\underset{|xy|>1/5^n}{}|xy|^{\alpha \gamma }\widehat{\nu }_\epsilon ^n(dy)\hfill & & (5^{\alpha +\gamma })^n|\widehat{\nu }_\epsilon ^n|\hfill \\ & =& (5^{\alpha +\gamma }\frac{|\mu |}{\alpha })^n\epsilon ^{n\alpha }\hfill \\ & & c_0^n\epsilon ^{n\alpha },\hfill \end{array}$$
by taking large $`c_0`$. We get
$$\underset{|xy|>2\epsilon /3}{}|xy|^{\alpha \gamma }\widehat{\nu }_\epsilon ^n(dy)c_0^n\epsilon ^{n\alpha }(c_3+1),$$
and (15) yields
$$\widehat{\nu }_\epsilon ^{n+1}(B(x,r))c_0^{n+1}r^\gamma \epsilon ^{n\alpha }.$$
###### Corollary 2
There exists $`C=C(\alpha ,\mu )`$ such that
(16)
$$\widehat{\nu }_\epsilon ^n(B(x,\lambda \epsilon ))C^n\lambda ^\gamma (1+\lambda ^\alpha )\epsilon ^{\gamma (n1)\alpha },\lambda >0,\epsilon >0,|x|=1.$$
Proof. Lemma 1 yields (16) for $`\lambda \epsilon <1/5^n`$. For $`\lambda \epsilon 1/5^n`$ we have
$$\widehat{\nu }_\epsilon ^n(B(x,\lambda \epsilon ))|\widehat{\nu }_\epsilon ^n|=\frac{|\mu |^n}{\alpha ^n}\epsilon ^{n\alpha }(\frac{|\mu |}{\alpha }5^{\alpha +\gamma })^n\lambda ^{\alpha +\gamma }\epsilon ^{\gamma (n1)\alpha }.$$
In what follows we denote $`\widehat{P}_t=\widehat{P}_t^{t^{1/\alpha }}`$ and $`\stackrel{~}{P}_t=\stackrel{~}{P}_t^{t^{1/\alpha }}`$.
###### Corollary 3
There exists $`C=C(\alpha ,\mu )`$ such that
$$\widehat{P}_t(B(x,\lambda t^{1/\alpha }))C\lambda ^\gamma (1+\lambda ^\alpha )t^{1+\frac{\gamma }{\alpha }},\lambda >0,t>0,|x|=1.$$
Proof. Corollary 2 yields
$`\widehat{P}_t(B(x,\lambda t^{1/\alpha }))`$ $`=`$ $`e^{|\mu |/\alpha }{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{t^n\widehat{\nu }_{t^{1/\alpha }}^n(B(x,\lambda t^{1/\alpha }))}{n!}}`$
$``$ $`e^{|\mu |/\alpha }{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{c^n\lambda ^\gamma (1+\lambda ^\alpha )t^{1+\frac{\gamma }{\alpha }}}{n!}}`$
$`=`$ $`e^{c|\mu |/\alpha }\lambda ^\gamma (1+\lambda ^\alpha )t^{1+\frac{\gamma }{\alpha }}.`$
###### Corollary 4
$`\widehat{P}_1(B(y,\lambda ))C\lambda ^\gamma (1+\lambda ^\alpha )|y|^{\alpha \gamma }`$ for $`y^d`$ and $`\lambda >0`$.
Proof. Let $`y^d\{0\}`$ and $`x=y/|y|`$, $`t=|y|^\alpha `$. By scaling and Corollary 3 we have
$`\widehat{P}_1(B(y,\lambda ))=\widehat{P}_t(B(x,\lambda t^{1/\alpha }))c\lambda ^\gamma (1+\lambda ^\alpha )|y|^{\alpha \gamma }.`$
We note that for every $`q>0`$ we have that $`|y|^q\stackrel{~}{P}_1(dy)<\mathrm{}`$, because the support of $`\stackrel{~}{\nu }_1`$ is bounded (). A simple reasoning based on this and the boundedness of the derivative of $`\stackrel{~}{p}_1`$ yields
$$\stackrel{~}{p}_1(y)c_q(1+|y|)^q,q>0,y^d,$$
see \[43, Lemma 9\].
###### Lemma 5
For every $`q>0`$ there exists $`C=C(\alpha ,\mu ,q)`$ such that
$$\stackrel{~}{P}_1(B(z,\rho ))C(1+|z|)^q\rho ^d,\rho 1,z^d.$$
Proof. If $`|z|<2`$ then $`\stackrel{~}{P}_1(B(z,\rho ))=_{B(z,\rho )}\stackrel{~}{p}_1(y)𝑑yc\rho ^dc(1+|z|)^q\rho ^d`$.
If $`|z|2`$ then $`\stackrel{~}{P}_1(B(z,\rho ))c(1+|z|/2)^q\rho ^dc(1+|z|)^q\rho ^d`$.
The proof of the following lemma is a simplification of the proof of \[43, Theorem 3\].
###### Lemma 6
$`P_1(B(z,\rho ))C|z|^{\alpha \gamma }\rho ^d`$ for $`z^d`$ and $`0<\rho 1`$.
Proof. By (7), Lemma 5, and Corollary 4
$`P_1(B(z,\rho ))`$ $`=`$ $`\stackrel{~}{P}_1\widehat{P}_1(B(z,\rho ))={\displaystyle _^d}\stackrel{~}{P}_1(B(zy,\rho ))\widehat{P}_1(dy)`$
$`=`$ $`{\displaystyle _0^1}\widehat{P}_1(\{y:\stackrel{~}{P}_1(B(zy,\rho ))>s\})𝑑s`$
$``$ $`{\displaystyle _0^1}\widehat{P}_1(\{y:c(1+|zy|)^q\rho ^d>s\})𝑑s`$
$``$ $`{\displaystyle _0^{c\rho ^d}}\widehat{P}_1(B(z,c^{1/q}s^{1/q}\rho ^{d/q}))𝑑s`$
$``$ $`c{\displaystyle _0^{c\rho ^d}}(c^{1/q}s^{1/q}\rho ^{d/q})^\gamma (1+(c^{1/q}s^{1/q}\rho ^{d/q})^\alpha )|z|^{\gamma \alpha }𝑑s`$
$`=`$ $`c|z|^{\gamma \alpha }\left[\rho ^{d\gamma /q}{\displaystyle _0^{c\rho ^d}}s^{\gamma /q}𝑑s+\rho ^{d(\gamma +\alpha )/q}{\displaystyle _0^{c\rho ^d}}s^{(\gamma +\alpha )/q}𝑑s\right]`$
$`=`$ $`c|z|^{\gamma \alpha }\left[\rho ^{d\gamma /q}(\rho ^d)^{1\gamma /q}+\rho ^{d(\gamma +\alpha )/q}(\rho ^d)^{1(\gamma +\alpha )/q}\right]=c|z|^{\gamma \alpha }\rho ^d.`$
The following two corollaries are our main estimates of the semigroup. Corollary 8 is an analogue of \[43, Theorem 3\], while (17) corresponds to .
###### Corollary 7
$`P_1(B(z,\rho ))C(1+|z|)^{\alpha \gamma }\rho ^d`$ if $`\mathrm{\hspace{0.33em}0}\rho <|z|/2`$.
Proof. We recall that $`p_1(y)=P_1(dy)/dy`$ is bounded and so Lemma 6 yields
(17)
$$p_1(y)c(1+|y|)^{\gamma \alpha },y^d.$$
If $`0\rho <|z|/2`$ then $`P_1(B(z,\rho ))c_{B(z,\rho )}(1+|y|)^{\gamma \alpha }𝑑y(1+|z|)^{\alpha \gamma }\rho ^d.`$
###### Corollary 8
$`P_t(B(x,\rho ))Ct^{1+\frac{\gamma d}{\alpha }}\rho ^d`$, provided $`|x|=1`$, $`t>0`$, and $`0\rho t^{1/\alpha }`$.
Proof. By scaling and Lemma 6 we have $`P_t(B(x,\rho ))=P_1(B(xt^{1/\alpha },\rho t^{1/\alpha }))ct^{1+\frac{\gamma d}{\alpha }}\rho ^d`$.
Proof of Theorem 1. Let $`|x|=1`$, $`0\rho <1/2`$. By scaling and Corollary 7
$`𝕍(B(x,\rho ))`$ $`=`$ $`{\displaystyle _0^{\mathrm{}}}P_t(B(x,\rho ))𝑑t={\displaystyle _0^{\mathrm{}}}P_1(B(xt^{1/\alpha },\rho t^{1/\alpha }))𝑑t`$
$``$ $`c\rho ^d{\displaystyle _0^{\mathrm{}}}(1+t^{1/\alpha })^{\gamma \alpha }t^{d/\alpha }𝑑t.`$
The integral is finite because $`d/\alpha <1`$ and $`(\gamma +\alpha d)/\alpha >1`$. Let $`y^d\{0\}`$, $`x=y/|y|`$. By scaling, a change of variable, and (17)
$`V(y)`$ $`=`$ $`{\displaystyle _0^{\mathrm{}}}t^{d/\alpha }p_1(yt^{1/\alpha })𝑑t=|y|^{\alpha d}{\displaystyle _0^{\mathrm{}}}t^{d/\alpha }p_1(xt^{1/\alpha })𝑑t`$
$``$ $`|y|^{\alpha d}{\displaystyle _0^{\mathrm{}}}t^{d/\alpha }(1+t^{1/\alpha })^{\gamma \alpha }𝑑tc|y|^{\alpha d}.`$
The first integral above is locally uniformly convergent on $`^d\{0\}`$ hence $`V`$ is continuous there.
We now proceed to our converse, Theorem 2. We propose a general approach based on a simple study of generator $`𝒜`$. We first note that
(18)
$$p_t(x)>0,x^d(t>0),$$
see () or \[43, Lemma 5\]. In fact, (18) easily follows from (8), (2), continuity of $`\stackrel{~}{p}_t^\epsilon `$, and the fact that $`supp(\nu )+supp(\nu )+\mathrm{}+supp(\nu )`$ ($`d`$ times) equals $`^d`$.
By (18), (12), and continuity of $`p_t`$ for $`t>0`$, there is a constant $`c=c(\alpha ,\mu )`$ such that
(19)
$$V(x)c|x|^{\alpha d},x^d.$$
###### Lemma 9
Let $`d>\alpha `$. For all $`\phi C_c^{\mathrm{}}(^d)`$ we have
$$\underset{^d}{}𝒜\phi (xy)𝕍(dy)=\phi (x),x^d,$$
where the integral is absolutely convergent.
This is well-known (see, e.g., \[36, Theorem 3.5.78\]). We only note that $`|𝒜\phi (x)|c(1+|x|)^{1\alpha }`$. The absolute convergence follows from this and the homogeneity of $`𝕍`$.
Proof of Theorem 2. If $`d=1\alpha `$ then $`V\mathrm{}`$ and there is nothing to prove. Thus we assume that $`d>\alpha `$. We fix a function $`\varphi C_c^{\mathrm{}}(^d)`$ such that $`\varphi 0`$, $`\mathrm{supp}\varphi B(0,1/2)`$ and $`\varphi =1`$ on $`B(0,1/3)`$. Let $`r>0`$. Put $`\varphi _r(x)=\varphi (x/r)`$ and $`\mathrm{\Lambda }_r(x)=𝒜\varphi _r(x)`$. Homogeneity of $`𝒜`$ yields $`\mathrm{\Lambda }_r(x)=r^\alpha \mathrm{\Lambda }_1(x/r)`$. Note that $`𝒜\varphi =\mathrm{\Lambda }_1`$ is bounded, hence there is a constant $`c`$ such that
$$\mathrm{\Lambda }_r(x)cr^\alpha .$$
If $`|x|r/2`$ then $`\mathrm{\Lambda }_r(x)0`$, and in fact $`\mathrm{\Lambda }_r(x)\nu (B(x,r/3))`$. Let $`|x|>r`$. From Lemma 9 we have
$`0`$ $`=`$ $`{\displaystyle \underset{^d}{}}\mathrm{\Lambda }_r(xy)𝕍(dy){\displaystyle \underset{B(x,r/2)}{}}\mathrm{\Lambda }_r(xy)𝕍(dy)+{\displaystyle \underset{B(0,r/4)}{}}\mathrm{\Lambda }_r(xy)𝕍(dy)`$
$``$ $`cr^\alpha 𝕍((B(x,r/2))+{\displaystyle \underset{B(0,r/4)}{}}\nu (B(xy,r/3))𝕍(dy)`$
$``$ $`cr^\alpha 𝕍((B(x,r/2))+𝕍(B(0,r/4))\nu (B(x,r/12)).`$
Since $`𝕍(B(0,r/4))=r^\alpha 𝕍(B(0,1/4))`$ and $`𝕍(B(0,1/4))<\mathrm{}`$ we get
(20)
$$\nu (B(x,r/12))cr^{2\alpha }𝕍((B(x,r/2)),|x|>r.$$
We note that similar results can also be derived from the lower bounds for the semigroup as given in \[53, Theorem 1.1\].
## 4 Harnack’s inequality: preliminaries
The general references for this section are , , , or . The Lévy measure $`\nu `$ yields a standard symmetric stable Lévy process $`(X_t,P^x)`$ with generating triplet $`(0,\nu ,0)`$. Namely, the transition probabilities of the process $`(X_t,P^x)`$ are $`P(t,x,A)=P_t(Ax)`$, $`t>0`$, $`x^d`$, $`A^d`$, and $`P(0,x,A)=\mathrm{𝟏}_A(x)`$, where $`\{P_t,t0\}`$ is the stable semigroup of measures introduced in Preliminaries. The process is strong Markov with respect to the so-called standard filtration.
The process conveniently leads to a definition of harmonic measures $`\omega _D^x`$, and their properties (21) and (24) below. For an analytic definition of these, called the fundamental family, we refer to (see also ).
For open $`U^d`$ we denote $`\tau _U=inf\{t0;X_tU\}`$, the first exit time of $`U`$. We write $`\omega _D^x`$ for the harmonic measure of (open) $`D`$:
$$\omega _D^x(A)=P^x(\tau _D<\mathrm{},X_{\tau _D}A),x^d,A^d.$$
By the strong Markov property
(21)
$$\omega _D^x(A)=\omega _D^y(A)\omega _U^x(dy),\text{ if }UD.$$
We say that a function $`u`$ on $`^d`$ is harmonic in open $`D^d`$ if
(22)
$$u(x)=E^xu(X_{\tau _U})=_{U^c}u(y)\omega _U^x(dy),x^d,$$
for every bounded open set $`U`$ with the closure $`\overline{U}`$ contained in $`D`$. It is called regular harmonic in $`D`$ if (22) holds for $`U=D`$. If $`D`$ is unbounded then $`E^xu(X_{\tau _D})=E^x[\tau _D<\mathrm{};u(X_{\tau _D})]`$ by a convention. Under (22) it will be only assumed that the expectation in (22) is well defined (but not necessarily finite). Regular harmonicity implies harmonicity, and it is inherited by subsets $`UD`$. This follows from (21).
We denote by $`p_t^D(x,v)`$ the transition density of the process killed at the first exit from $`D`$:
$$p_t^D(x,v)=p(t,x,v)E^x[\tau _D<t;p(t\tau _D,X_{\tau _D},v)],t>0,x,v^d.$$
Here $`p(t,x,v)=p_t(vx)`$. For convenience we will assume in the sequel that $`D`$ is regular: $`P^x[inf\{t>0:X_tD\}=0]=1`$ for $`xD^c`$ (see ). Then $`p_t^D`$ is symmetric: $`p_t^D(x,v)=p_t^D(v,x)`$, $`x,vD`$ (see, e.g., ). The strong Markov property yields
(23)
$$p(t,x,v)=E^x[p(t\tau _D,X_{\tau _D},v);\tau _D<t],xD,vD^c.$$
In particular, $`p_t^D(x,v)=0`$ if $`xD`$, $`vD^c`$. We let
$$G_D(x,v)=\underset{0}{\overset{\mathrm{}}{}}p_t^D(x,v)𝑑t,$$
and we call $`G_D(x,v)`$ the Green function for $`D`$. If $`V`$ is continuous on $`^d\{0\}`$, so that $`V(x)c|x|^{\alpha d}`$, then the strong Markov property yields for $`x,vD`$
(24)
$$G_D(x,v)=V(x,v)E^xV(X_{\tau _D},v)=V(x,v)_{D^c}V(z,v)\omega _D^x(dz).$$
Here $`V(x,v)=V(vx)`$. The Green function is symmetric: $`G_D(x,v)=G_D(v,x)`$, continuous in $`D\times D\{(x,v):x=v\}`$, and it vanishes if $`xD^c`$ or $`vD^c`$.
Note that $`V(x,v)`$ is harmonic in $`x`$ on $`^d\{v\}`$. Indeed, if $`xD`$ and $`dist(D,v)>0`$ then by (23)
$`V(x,v)`$ $`=`$ $`{\displaystyle \underset{0}{\overset{\mathrm{}}{}}}E^x[p(t\tau _D,X_{\tau _D},v);\tau _D<t]𝑑t=E^xV(X_{\tau _D},v).`$
Similarly, the Green function $`vG_D(x,v)`$ is harmonic in $`D\{x\}`$.
By Ikeda–Watanabe formula we have
(25)
$$\omega _D^x(A)=_DG_D(x,v)\nu (Av)𝑑v,\text{ if }dist(A,D)>0.$$
We note here that translation–invariance of the Lebesgue measure and Fubini–Tonelli theorem yield
(26)
$$\mathrm{\Phi }(v)\mathrm{\Psi }(v+z)m(dz)𝑑v=\mathrm{\Phi }(v+z)\mathrm{\Psi }(v)m(dz)𝑑v,$$
for every symmetric measure $`m`$ and nonnegative functions $`\mathrm{\Phi }`$ and $`\mathrm{\Psi }`$. In particular, taking $`m=\nu `$, $`\mathrm{\Phi }(v)=G_D(x,v)`$ and $`\mathrm{\Psi }(v)=\mathrm{𝟏}_A(v)`$ we get
$$_DG_D(x,v)\nu (Av)𝑑v=_A_{D+v}G_D(x,vz)\nu (dz)𝑑v.$$
If the boundary of $`D`$ is smooth or even Lipschitz then
$$\omega _D^x(D)=0,xD,$$
see (see also , ). In this case $`\omega _D^x`$ is absolutely continuous with respect to the Lebesgue measure on $`D^c`$. Its density function, or the Poisson kernel, is given by the formula
(27)
$$P_D(x,y)=_{yD}G_D(x,yz)\nu (dz),xD.$$
Note that such $`D`$ are regular, because of (18) and scaling. In particular the above considerations apply to $`D=B(0,1)`$.
It follows from (9) that for every $`r>0`$ and $`x^d`$ the $`P_x`$ distribution of $`\{X_t,t0\}`$ is the same as the $`P_{rx}`$ distribution of $`\{r^1X_{r^\alpha t},t0\}`$. In particular,
(28)
$$\omega _D^x(A)=\omega _{rD}^{rx}(rA).$$
We call (28) scaling, too. It yields that for $`u`$ harmonic on $`D`$, the dilation, $`u_r`$, is harmonic on $`rD`$. A similar remark concerns translations.
By (26) we also obtain
$$_{B(0,1/2)}|y|^{\alpha d}\nu (Ay)𝑑y=_A_{B(y,1/2)}|yz|^{\alpha d}\nu (dz)𝑑y,A^d,$$
and
$$_{B(0,1/2)}\nu (Ay)𝑑y=_A\nu (B(y,1/2))𝑑y,A^d.$$
Therefore we can express the relative Kato condition (RK) in an equivalent form:
(29)
$$_{B(0,1/2)}|y|^{\alpha d}\nu (Ay)𝑑yK_{B(0,1/2)}\nu (Ay)𝑑y,A^d.$$
We remark that (RK) is a local condition at infinity: the inequality in (2) only needs to be verified for large $`y^d`$. In particular, if it holds for $`|y|>1`$ then it holds for all $`y^d`$, possibly with a different constant, see . Noteworthy, the reverse of (2) (and (29)) always holds, so actually (RK) means comparability of both sides of (2) (and (29)).
In what follows we let $`G=G_{B(0,1)}`$, $`P=P_{B(0,1)}`$ and we define
$$s(x)=E^x\tau _{B(0,1)}=_{B(0,1)}G(x,v)𝑑v.$$
Explicit formulas for these functions for $`\nu (dy)=|y|^{d\alpha }dy`$ are known and may give some insight into the general situation. They are essentially due to M. Riesz, see, e.g., , , , , . In particular (for isotropic $`\nu `$) we have
(30)
$$P(x,y)=C_\alpha ^d\left[\frac{1|x|^2}{|y|^21}\right]^{\alpha /2}|xy|^d,|x|<1,|y|>1.$$
The following two lemmas are consequences of symmetry and nondegeneracy of the spectral measure $`\mu `$. They can be proved similarly as Lemma 4 and Lemma 10 of , so we skip the proofs.
###### Lemma 10
There exist $`\epsilon =\epsilon (\alpha ,\mu )(0,1)`$ and $`C=C(\alpha ,\mu )`$ such that
(31)
$$\nu (B(x,1\epsilon ))C,$$
provided $`1\epsilon <|x|<1`$.
###### Lemma 11
There exists $`C=C(\alpha ,\mu )`$ such that
$$s(x)C(1|x|^2)^{\alpha /2},|x|<1.$$
## 5 Necessity of relative Kato condition
In this short section we assume that Harnack’s inequality (1) holds. We make no further assumptions on $`\nu `$ beyond these in Section 2. In particular our considerations do not depend on the estimates in Section 3.
###### Lemma 12
Harnack’s inequality implies the relative Kato condition.
Proof. We first consider the case $`d>\alpha `$. We claim that
(32)
$$V(x)|x|^{\alpha d},x^d.$$
Indeed, for every $`|x|=1`$, $`𝕍(B(x,1/4))=_{B(x,1/4)}V(v)𝑑v𝕍(B(0,2))<\mathrm{}`$, so there exists $`vB(x,1/4)`$ such that $`V(v)𝕍(B(0,2))/|B(0,1/4)|`$. By Harnack’s inequality $`V(x)cV(v)`$. The estimate (32) follows from (12) and (19).
Let $`g(v)=\mathrm{min}(G(0,v),1)`$. We claim that
(33)
$$G(x,v)g(v)|vx|^{\alpha d}if|x|<1/2and|v|<1.$$
Indeed, by (32) and (24) for small $`\delta >0`$ we have:
$$G(x,v)|vx|^{\alpha d},|x|<1/2,|xv|<\delta .$$
Harnack’s inequality yields that $`G(x,v)|vx|^{\alpha d}`$ provided $`|x|<1/2`$ and $`|v|<3/4`$, and also $`G(x,v)G(0,v)`$ if $`|x|<1/2`$ and $`|v|>3/4`$. Note that $`g`$ is locally bounded from below on $`B(0,1)`$. This completes the proof of (33).
For every $`A^d`$ the function $`x\omega _{B(0,1)}^x(A)`$ is nonnegative on $`^d`$ and regular harmonic in $`B(0,1)`$. Harnack’s inequality (1), (25), (33), and Fubini-Tonelli yield
$`\omega _{B(0,1)}^0(A)`$ $``$ $`{\displaystyle _{B(0,1/2)}}\omega _{B(0,1)}^x(A)𝑑x`$
$``$ $`{\displaystyle _B}{\displaystyle _{B(0,1/2)}}g(v)|vx|^{\alpha d}\nu (Av)𝑑v𝑑x`$
$``$ $`{\displaystyle _B}g(v)\nu (Av)𝑑v.`$
This and (25) yield
$$_Bg(v)|v|^{\alpha d}\nu (Av)𝑑v_Bg(v)\nu (Av)𝑑v.$$
To this “approximate equality” we add the following one:
$$_{BB(0,3/4)}|v|^{\alpha d}\nu (Av)𝑑v_{BB(0,3/4)}\nu (Av)𝑑v,$$
and we obtain
$$_B|v|^{\alpha d}\nu (Av)𝑑v_B\nu (Av)𝑑v,AB^c.$$
A change of variable: $`v=2u`$ yields (29) and (2).
In the case $`d\alpha `$ we have $`d=1`$, and so $`\nu (dy)=c|y|^{1\alpha }dy`$, which satisfies (RK).
## 6 Sufficiency of relative Kato condition
In what follows we assume that (RK) holds for $`\nu `$. We will also assume that $`d>\alpha `$ unless stated otherwise.
The key step in the proof of Harnack’s inequality is the following estimate for the Green function of the ball, which we prove after a sequence of lemmas. We note that it is essentially the same inequality as (33), but proved under explicit assumptions on $`\nu `$ rather than by stipulating Harnack’s inequality. The estimate was suggested by the sharp estimates of the Green function of Lipschitz domains for the isotropic $`\nu `$ (see also ). We also refer the reader to for more explicit estimates for smooth domains and to, e.g., for explicit formulas for the ball.
###### Proposition 1
$`G(x,v)s(v)|vx|^{\alpha d}`$ provided $`|x|<1/2`$ and $`|v|<1`$.
###### Lemma 13
$`\nu `$ is a $`(d\alpha )`$-measure on $`𝕊`$.
Proof. Indeed, for $`|x|=1`$, $`0<r<1/2`$ by (2) we obtain
$$\nu (B(x,r))r^{d\alpha }\underset{B(x,1/2)}{}|xz|^{\alpha d}\nu (dz)K\nu (B(0,1/2)^c)r^{d\alpha }.$$
Theorem 1 yields that $`V`$ is continuous on $`^d\{0\}`$. Consequently, $`V(x)|x|^{\alpha d}`$ and $`G(x,y)`$ is continuous on $`B\times B\{(x,y):x=y\}`$.
###### Lemma 14
$`G(x,v)|vx|^{\alpha d},\text{ if }|x|<1/2,|v|<3/4.`$
We skip the proof as it is the same as the one of Lemma 6 in .
We note that $`\underset{xz}{lim}G(x,v)=0`$ for every $`vB(0,1)`$ and every point $`z𝕊`$ because the measures $`\omega _{B(0,1)}^x`$ weakly converge to $`\delta _z`$. This is related to the regularity of $`B(0,1)`$, and it follows, e.g., from the estimate
$$\omega _{B(x,1|x|)}^x(B(0,1)^c)c,$$
which is a consequence of scaling, nondegeneracy of $`\nu `$ (compare (31)), and (25).
We will employ the operator
$$𝒰_r\varphi (x)=\frac{E^x\varphi (X_{\tau _{B(x,r)}})\varphi (x)}{E^x\tau _{B(x,r)}},$$
whenever the expression is well defined for given $`\varphi `$, $`r>0`$ and $`x`$. We note that $`𝒰_r`$ is implicitly used in \[7, Chapter III §17\]. Clearly, if $`h`$ is harmonic in $`D`$, $`xD`$, and $`r<dist(x,D^c)`$, then $`𝒰_rh(x)=0`$. We note that
$$𝒰\varphi (x)=\underset{r0}{lim}𝒰_r\varphi (x)$$
is the Dynkin characteristic operator, which was used in in a similar way.
We record the following observation (maximum principle).
###### Lemma 15
If there is $`r>0`$ such that $`𝒰_rh(x)>0`$ then $`h(x)<\underset{y^d}{sup}h(y)`$.
###### Lemma 16
There exists $`C=C(\alpha ,\mu )`$ such that
$$G(x,v)<Cs(v),|x|<1/2,3/4<|v|<1.$$
Proof. By the strong Markov property we have
$`s(v)`$ $`=`$ $`E^v\tau _B=E^v(\tau _A+\tau _{B(0,1)}\theta _{\tau _A})=E^v\tau _A+E^vE^{X_{\tau _A}}\tau _{B(0,1)}`$
$`=`$ $`E^v\tau _A+E^vs(X_{\tau _A}),v^d,AB(0,1),`$
which yields $`𝒰_rs(v)=1`$ for $`vB(0,1)`$ and $`r<1|v|`$.
For $`n\{1,2,\mathrm{}\}`$ and $`xB(0,1/2)`$ we let $`g(v)=G(x,v)`$ and $`g_n(v)=\mathrm{min}(G(x,v),n)`$. For $`vB(x,1/8)^c`$ we have that $`G(x,v)c_1|xv|^{\alpha d}`$ hence $`g_n(v)=G(x,v)`$ provided $`nc_18^{d\alpha }`$. By harmonicity of $`g`$ on $`B(0,1)\{x\}`$, scaling property, (25) and (2) we obtain that for $`vB(0,1)B(0,3/4)`$ and $`r<\mathrm{min}(1|v|,1/16)`$ it holds
$`𝒰_rg_n(v)`$ $`=`$ $`𝒰_r(g_ng)(v)`$
$`=`$ $`{\displaystyle \frac{1}{E^0\tau _{B(0,1)}}}{\displaystyle \underset{B(0,1)}{}}G(0,w){\displaystyle (g_ng)(v+rw+z)\nu (dz)𝑑w}`$
$``$ $`{\displaystyle \frac{c_2}{s(0)}}{\displaystyle \underset{B(0,1)}{}}G(0,w){\displaystyle \underset{B(xvrw,1/8)}{}}|xvrwz|^{\alpha d}\nu (dz)𝑑w`$
$``$ $`{\displaystyle \frac{c_2K}{s(0)}}{\displaystyle \underset{B(0,1)}{}}G(0,w)\nu (B(xvrw,1/8))𝑑wc_3.`$
If $`a>c_3`$ then
$$𝒰_r(asg_n)(v)=a𝒰_rg_n(v)a+c_3<0.$$
By scaling
$`s(v)`$ $``$ $`E^v\tau _{B(v,1|v|)}=(1|v|)^\alpha E^0\tau _{B(0,1)}`$
$``$ $`4^\alpha E^0\tau _{B(0,1)},|v|<3/4.`$
Since $`g_n(v)n`$, we see that $`as(v)g_n(v)>0`$ for $`vB(0,3/4)`$ provided $`a>n/(4^\alpha E^0\tau _{B(0,1)})`$.
Let $`a_0=\mathrm{max}[c_3,n/(4^\alpha E^0\tau _{B(0,1)})]+1`$ and $`h(v)=a_0s(v)g_n(v)`$. We have $`h(v)0`$ for $`v\overline{B(0,3/4)}`$, $`h(v)=0`$ for $`vB(0,1)^c`$ and $`𝒰_rh(v)<0`$ for $`vB(0,1)B(0,3/4)`$, $`r<\mathrm{min}(1|v|,1/16)`$. Lemma 15 and continuity of $`h`$ yields $`h(v)0`$ in $`B(0,1)`$. Since $`g_n=g`$ on $`B(0,3/4)^c`$, the lemma follows.
Lemma 16 and 11 yield the following conclusion:
(35)
$$G(x,v)C(1|v|)^{\alpha /2},|x|<1/2,\mathrm{\hspace{0.33em}3}/4<|v|<1.$$
###### Lemma 17
There is $`C=C(\alpha ,\mu )`$ such that $`G(x,v)Cs(v)`$ provided $`|x|<1/2`$ and $`|v|<1`$.
Proof. Let $`xB(0,1/2)`$. We fix $`\epsilon `$ such that (31) is satisfied. Lemma 14 yields that $`G(x,v)c_1>0`$ for $`vB(0,1\epsilon )`$. Let $`n\{1,2,\mathrm{}\}`$ be such that $`c_12/n`$. By (35) there is $`\eta >0`$ such that $`G(x,v)1/n`$ for $`vB(0,1)B(0,1\eta )`$. Let $`g(v)=G(x,v)`$ and $`g_n(v)=\mathrm{min}(g(v),1/n)`$. We have
$$g_n(v)=g(v),vB(0,1)B(0,1\eta ),$$
and
$$g(v)g_n(v)2/n1/n=1/n,vB(0,1\epsilon ),$$
hence by Lemma 10 for $`vB(0,1)\overline{B(0,1\eta )}`$ and $`r<\mathrm{min}(1|v|,(\epsilon \eta )/2)`$ we obtain
$`𝒰_rg_n(v)`$ $`=`$ $`𝒰_r(g_ng)(v)`$
$`=`$ $`{\displaystyle \frac{1}{s(0)}}{\displaystyle \underset{B(0,1)}{}}G_{B(0,1)}(0,w){\displaystyle (g_ng)(v+rw+z)\nu (dz)𝑑w}`$
$``$ $`{\displaystyle \frac{1}{n}}{\displaystyle \frac{1}{s(0)}}{\displaystyle \underset{B(0,1)}{}}G_{B(0,1)}(0,w)\nu (B(v+rw,1\epsilon ))𝑑w{\displaystyle \frac{c_2}{n}}.`$
For $`a>0`$ we have
$$𝒰_r(ag_ns)(v)c_2a/n+1,vB(0,1)B(0,1\eta ).$$
This is negative if $`a>n/c_2`$. Furthermore $`s(v)c_3`$ for $`vB(0,1)`$ and $`g_n(v)c_4>0`$ for $`vB(0,1\eta )`$. Thus $`ag_n(v)s(v)ac_4c_3>0`$ for $`vB(0,1\eta )`$ if only $`a>c_3/c_4`$. Note that our estimates do not depend on $`x`$, provided $`|x|<1/2`$. Let $`a_0=\mathrm{max}(c_3/c_4,n/c_2)+1`$ and $`h(v)=a_0g_n(v)s(v)`$. We have $`h(v)0`$ for $`v\overline{B(0,1\eta )}`$ and $`𝒰_rh(v)<0`$ for $`vB(0,1)B(0,1\eta )`$. By Lemma 15 and the continuity of $`h`$ we get $`h(v)0`$ in $`B(0,1)`$ and the lemma follows.
Proof of Proposition 1. The estimate is a consequence of (6), Lemma 14, 16, and 17.
Maciej Lewandowski has informed us that he recently proved the converse of the inequality in Lemma 11. This implies
(36)
$$G(x,v)(1|v|^2)^{\alpha /2}|vx|^{\alpha d},|x|<1/2,|v|<1.$$
We will not use (36) in the sequel; the less explicit estimate in Lemma 17 suffices for our purposes. Note that the asymptotic of $`G`$ at the pole is different when $`d=1\alpha `$, see, e.g., .
###### Lemma 18
(RK) implies Harnack’s inequality for all $`d\{1,2,\mathrm{}\}`$ and $`\alpha (0,2)`$.
Proof. By translation and scaling invariance of the class of harmonic functions and by a covering argument we only need to verify that
$$u(0)cu(x),|x|<1/2,$$
whenever $`u`$ is nonnegative on $`^d`$ and regular harmonic on $`B(0,1)`$. For this to hold it is sufficient to have, with the same constant $`c`$,
(37)
$$P(0,y)cP(x,y),|x|<1/2,|y|>1.$$
If $`d=1`$, (37) follows from (30). Thus we only need to examine the case $`d>\alpha `$. By the decomposition $`B(0,1)=B(0,1/2)[B(0,1)B(0,1/2)]`$, (27), Proposition 1, (2), and the fact that $`s`$ is bounded away from zero on compact subsets of $`B(0,1)`$ (comp. (6)), we obtain
$`P(0,y)`$ $``$ $`{\displaystyle _{B(y,1)}}s(yv)|yv|^{\alpha d}\nu (dv){\displaystyle _{B(y,1)}}s(yv)\nu (dv)`$
$``$ $`c{\displaystyle _{B(y,1)}}s(yv)|yvx|^{\alpha d}\nu (dv)`$
$``$ $`P(x,y),|x|<1/2,|y|>1.`$
Proof of Theorem 3. See Lemma 12 and Lemma 18.
We conclude with a few remarks and open problems.
By translation and dilation invariance of the class of considered harmonic functions, and by a covering argument Harnack’s inequality holds for every compact subset of every connected domain of harmonicity. We note that : (1) it does not generally hold for disconnected open sets, as the support of $`yP(x,y)`$ may be smaller than $`B(0,1)^c`$ (see 25), (2) it does hold for all open sets if $`\nu `$ is isotropic (this follows from (30), or see ).
We consider the following examples of measures $`\nu `$. (RK) holds for $`\nu _1(dy)|y|^{d\alpha }dy`$ (both sides of (2) may be explicitly estimated). Next, let $`\xi 𝕊`$, $`0<r<\sqrt{2}`$, and $`C=𝕊[B(\xi ,r)B(\xi ,r)]`$. (RK) holds for $`\nu _2(dy)=\mathrm{𝟏}_C(y/|y|)|y|^{d\alpha }dy`$, see .
On the other hand, consider balls $`B_nB_n^{}`$ centered at $`𝕊`$, with radii $`4^n`$ and $`2^n`$, respectively, and such that $`\{B_n^{}\}`$ are pairwise disjoint. Let $`C=_{nn_0}B_n`$ and let $`\nu _3(dy)=\mathrm{𝟏}_C(y/|y|)|y|^{d\alpha }dy`$. If $`d1>\alpha `$ then (RK) does not hold for $`\nu _3`$ () even though it is bounded by $`\nu _1`$.
Let $`B_{\xi ,r}=B(\xi ,r)𝕊`$. By integrating in polar coordinates we can give the characterization of relative Kato condition in terms of its spectral measure $`\mu `$ and $`B_{\xi ,r}`$ (comp. ). Let $`d\alpha >1`$. (RK) holds for $`\nu `$ if and only if
(38)
$$_{B_{\xi ,r}}(|\eta \xi |/r)^{\alpha (d1)}\mu (d\eta )c\mu (B_{\xi ,r}),\xi 𝕊,\mathrm{\hspace{0.33em}0}<r<c.$$
In the case $`d=2`$, $`\alpha =1`$, (RK) is equivalent to
(39)
$$_{B_{\xi ,r}}\mathrm{log}(2r/|\eta \xi |)\mu (d\eta )c\mu (B_{\xi ,r}),\xi 𝕊,\mathrm{\hspace{0.33em}0}<r<c.$$
In the case of $`d=2`$ and $`\alpha >1`$ (RK) is always satisfied. We omit the proofs.
###### Corollary 19
If $`d1<\alpha `$ then Harnack’s inequality holds for $`𝒜`$.
This may be extended as follows. We will say that $`\nu `$ is a strict $`\gamma `$-measure if
(40)
$$\nu (B(x,r))r^\gamma ,\text{ provided }xsupp\nu ,|x|=1,\mathrm{\hspace{0.33em}0}<r<1/2,$$
compare (13). Of course, if $`\nu `$ is a (strict) $`\gamma `$ measure on $`𝕊`$ than $`\mu `$ is a (strict) $`(\gamma 1)`$-measure (on $`𝕊`$). This observation and (38) yield the following conclusion, which we state without proof.
###### Corollary 20
If $`\nu `$ is a strict $`\gamma `$-measure with $`\gamma >d\alpha `$, then Harnack’s inequality holds for $`𝒜`$.
The example of $`\nu _3`$ shows the importance of the strictness assumption. We interpret (RK) as a property of balance or firmness of $`\nu `$. As such it is close to the reverse Hölder condition with exponent $`q>d/\alpha `$, see .
If $`\mu (\xi )>0`$ for some $`\xi 𝕊`$ then $`\nu `$ is a $`1`$-measure only. By Theorem 2 the potential kernel $`V`$ is unbounded on $`𝕊`$ if $`1>d2\alpha `$ (in fact, if $`1d2\alpha `$, see \[53, Theorem 1.1\], ). That $`V`$ may be infinite on rays emanating from the origin shows that harmonic functions cannot be defined pointwise by means of $`𝒜`$. In general they even lack finiteness in the domain of harmonicity (but see and in this connection). Thus the potential-theoretic properties of the operators $`𝒜`$ are very diverse among considered measures $`\nu `$. This is in sharp contrast with the fact that the exponents $`\mathrm{\Phi }`$ (see (5)) are all comparable and the same is true of the corresponding Dirichlet forms (, see also ). The boundary potential theory of $`𝒜`$ will generally be very different from that of the fractional Laplacian (see \[51, p. 199\] for a simple remark on this subject).
We like to mention a number of further interesting problems and references: (1) characterization of continuity and higher order regularity of $`V`$ on $`𝕊`$ (), (2) the boundary Harnack principle (comp. ), the corresponding approximate factorization of $`G(x,v)`$ for all $`x,vB(0,1)`$ (comp. and Proposition 1 above), and related boundary problems (comp. ), (3) study of other Lévy measures which are in the form of a product in polar coordinates, (4) study of similar nonlocal operators $`𝒜`$ which are not translation invariant ().
K. Bogdan (corresponding author), Institute of Mathematics, Polish Academy of Sciences, Institute of Mathematics, Wrocław University of Technology,
E-mail address: bogdan@pwr.wroc.pl
P. Sztonyk, Institute of Mathematics, Wrocław University of Technology, Wybrzeże Wyspiańskiego 27, 50–370 Wrocław, Poland
E-mail address: sztonyk@im.pwr.wroc.pl |
warning/0507/cond-mat0507250.html | ar5iv | text | # Ground state of the Kagomé-like S=1/2 antiferromagnet, Volborthite Cu3V2O7(OH)₂∙2H2O
## Abstract
Volborthite compound is one of the very few realizations of S=1/2 quantum spins on a highly frustrated kagomé-like lattice. Low-T SQUID measurements reveal a broad magnetic transition below 2 K which is further confirmed by a peak in the <sup>51</sup>V nuclear spin relaxation rate ($`1/T_1`$) at 1.4 K$`\pm `$0.2 K. Through <sup>51</sup>V NMR, the ground state (GS) appears to be a mixture of different spin configurations, among which 20% correspond to a well defined short range order, possibly of the $`\sqrt{3}\times \sqrt{3}`$ type. While the freezing involve all the Cu<sup>2+</sup> spins, only 40% of the copper moment is actually frozen which suggests that quantum fluctuations strongly renormalize the GS.
preprint: APS/123-QED
In the quest for novel states of condensed matter, frustration has emerged as a key concept Ramirez (2001). In magnetic systems, competitive interactions resulting from the geometry of the lattice, especially for corner sharing networks, can lead to a macroscopic entropy at T=0 K and could favor a novel spin liquid state. Theoretically, Heisenberg S=1/2 antiferromagnet on a 2D kagome lattice is predicted to lead to such an exotic quantum state Misguich and Lhuillier (2003); Mila (1998). Actually very few real systems approach this model. So far, the Cr<sup>3+</sup> (S=3/2) kagomé bilayer compounds SrCr<sub>9p</sub>Ga<sub>12-9p</sub>O<sub>19</sub> (SCGO) and Ba<sub>2</sub>Sn<sub>2</sub>ZnGa<sub>10-7p</sub>Cr<sub>7p</sub>O<sub>22</sub> ($`p<0.97`$) have been the experimental archetypes of Heisenberg frustrated antiferromagnets. In this context, the recent rediscovery of Volborthite Cu<sub>3</sub>V<sub>2</sub>O<sub>7</sub>(OH)$`{}_{2}{}^{}2`$H<sub>2</sub>Lafontaine et al. (1990); Hiroi et al. (2001) is a major step towards a realization of the theoretical model. The magnetic lattice of this natural antiferromagnet consists of quantum S=1/2 (Cu<sup>2+</sup>) spins sitting at the vertices of well separated ($`7.2\AA `$) kagome-like planes. The lattice displays a monoclinic distortion, possibly yielding two Cu-Cu interaction constants $`J_1J_2`$, which does not seem to impact on the characteristic fingerprints of frustration. Indeed, despite strong antiferromagnetic interactions ($`J90`$ K), no transition towards an ordered state has been detected down to 1.8 K, neither in susceptibility nor in heat capacity measurements Hiroi et al. (2001). Instead, a maximum in both these quantities is observed around 20 K, probably reflecting the enhancement of short range correlations, and defines a new low energy scale as a result of frustration Mendels et al. (2000); Bono et al. (2004a). Besides, as in kagomé bilayers Uemura et al. (1994), muon spin relaxation ($`\mu `$SR) experiments detected no sign of static spin freezing but rather temperature independent spin fluctuations persisting down to 50 mK, indicative of a fluctuating quantum GS Fukaya et al. (2003). Only in one ESR study the existence of an internal field was evidenced at 1.8 K which was interpreted as short range order Okubo et al. (2001).
In this Letter, we show, for the first time, that volborthite undergoes, around 1.3 K, a transition to a frozen state which we characterize using <sup>51</sup>V NMR as a local probe of magnetism. Note that, so far, NMR investigation of the GS in the kagomé bilayers had proved to be impossible because of a wipe out of the NMR intensity at low T Limot et al. (2002). The volborthite powder samples were prepared by refluxing an aqueous suspension of V<sub>2</sub>O<sub>5</sub> and a basic copper (II) carbonate salt Cu(OH)2-Cu(CO3) for several days. No spurious phase was detected in X-ray measurements. SQUID and NMR measurements above 2 K and $`\mu `$SR experiments on these samples Bert et al. (2004) were similar to the above described published data.
The low T static susceptibility of volborthite is presented in Fig. 1 (top panel). Below 2 K, the separation between the field cooled (FC) and zero field cooled (ZFC) susceptibilities reveals a spin freezing process. At 1.2 K, ”ageing” effects Vincent et al. (1997) have been observed (not shown here) which ascertains the glassy nature of the GS. At variance with textbook spin glass transition, the FC susceptibility does not level off below the transition and the hardly detectable broad maximum on the ZFC branch occurs below the FC/ZFC separation <sup>1</sup><sup>1</sup>1The up-turn at $`T<0.1`$ K, is due to few ppm magnetic impurities in the grease used to ensure thermalization.. This suggests a rather broad distribution of freezing temperatures as, for instance, in super-paramagnets where spin clusters get progressively blocked on lowering T.
To address the important issue of the intrinsic character of this glassy low-T phase, the susceptibility of two zinc substituted samples is also reproduced in Fig. 1. Zn atoms enter Volborhtite lattice without noticeable distortion. Since Zn<sup>2+</sup> is non-magnetic, the Zn/Cu substitution results in diluting the kagome magnetic lattice. The freezing temperature $`T_g`$, defined here as the temperature of the ZFC maxima, are plotted in the inset of Fig. 1 as a function of the magnetic lattice dilution and compared to the same data for SCGO compounds Limot et al. (2002). In marked contrast with SCGO, $`T_g`$ for Volborthite is strongly affected by dilution. This reminds us of the drastic reduction upon dilution of the dynamical plateau value seen in $`\mu `$SR experiments Bert et al. (2004) and suggests that the bilayer topology likely better accommodates defects than the single layer one in volborthite. Nonetheless, in both systems, the random dilution of the magnetic network reduces $`T_g`$ and hence simple magnetic dilution as a source of disorder cannot explain, alone, the spin-glass like transition.
Microscopic insight into this ground state is provided by our low-T <sup>51</sup>V NMR experiments. The non-magnetic V<sup>5+</sup> ions are situated above or below the center of the stars which constitutes the kagomé lattice. They thus probe symmetrically the magnetism of six Cu<sup>2+</sup> ions belonging to the same hexagon through an hyperfine coupling, estimated to be $`A7.7`$ kOe from susceptibility versus NMR shift measurements (see Ref. Bert et al. (2004)) above 90 K in agreement with the value estimated in Ref. Hiroi et al. (2001).
The nuclear spin lattice relaxation has been measured by the saturation-recovery method and fitted to a sum, with T-independent coefficients, of four exponential terms with relaxation rates proportional to $`1/T_1`$ as expected for a S=7/2 nuclear spin Narath (1967) in the case of partial saturation of the NMR line. The divergence of $`1/T_1`$ at 1.4 K$`\pm `$0.2 K (Fig. 3, inset) is a strong evidence that the vanadium probe indeed feels a transition in agreement with our susceptibility measurements. Following the analysis of Ref. Fukaya et al. (2003) of the zero field $`\mu `$SR relaxation rate below 1 K ($`\lambda =5\mu `$s<sup>-1</sup> in our sample Bert et al. (2004)), we get 3.5 MHz for the copper spin fluctuation rate. From the hyperfine coupling $`A`$, we would then expect the corresponding <sup>51</sup>V $`1/T_1`$ to be $`20`$ ms<sup>-1</sup>, much higher than the value actually measured. This suggests that spin fluctuations, seen in $`\mu `$SR, are efficiently filtered by the nuclear probe. We shall come back to this point later.
Characteristic <sup>51</sup>V ($`\gamma /2\pi =11.1923`$ MHz/T) NMR spectra measured at $`\nu _0=20.733`$ MHz, below 5 K, are presented in Fig. 2. At these low T, the quadrupolar splitting of the S=7/2 vanadium NMR line is masked by a large magnetic broadening which reflects the width of the field distribution at the vanadium site. On lowering T, the NMR line broadens drastically around 1.5 K and then saturates below 0.6 K. At a lower 12.548 MHz irradiating frequency, we checked that this saturation is field independent, and therefore reflects, as expected, a frozen field distribution in the GS. Upon closer inspection of the T-dependence of the lineshape, we note first the appearance of a broad background feature below 1.5 K. In order to track qualitatively this broad component we chose rather arbitrarily to plot the half width at 1/5<sup>th</sup> of the maximum in Fig. 3. Then the main central line, which width is roughly the width at half maximum of the spectra (open symbols in Fig. 3), starts to broaden rapidly and below 1 K, its lineshape changes from lorentzian-like to gaussian-like. The two distinct features, broad background and central line, in the low-T spectra suggest that the NMR line, below 1.5 K, is no more homogenous but results from at least two different types of magnetic environments of vanadium. This is probably related, as well, to the progressive freezing observed in macroscopic susceptibility. It is an important finding of this study as it demands a special magnetic ordering in the GS leading to two different V sites.
We now focus on the T=0.35 K spectra (Fig. 4) in the well established GS of volborthite. By comparison of the integrated intensity with $`T>2K`$ data, we checked that all sites are detected at this low-T and we therefore probe the bulk properties of the sample. The top and bottom spectra have been obtained in the same conditions except for a different duration $`\tau `$ in the pulse sequence $`\pi /2\tau \pi /2`$. This contrast procedure allows us to separate the two components of the spectrum. They indeed prove to have different spin-spin relaxation times $`T_2`$ which we determined by standard $`T_2`$ measurements at $`H=1.845`$ T and $`H=1.955`$ T.
In the long time spectrum (top panel), the slowly relaxing ($`T_2=105\mu `$s) broad background clearly stands out, without the fast relaxing ($`T_2=60\mu `$s) gaussian-like component and appears to be rectangular shaped. An unshifted narrower line also appears on this spectrum. However, because of its much longer $`T_2=240\mu `$s, it eventually represents less than 1$`\%`$ of the total sample and probably arises from some small impurity phase. The rectangular lineshape is a clear signature of the presence of one well defined frozen field $`H_f`$ at the vanadium site arising from the neighboring copper moments. The resonance condition writes $`H_0=2\pi \nu _0/\gamma =𝐇+𝐇_f`$ where $`H`$ is the applied field. Due to powder distribution, $`𝐇_f`$ is randomly oriented with respect to $`𝐇`$. In the limit $`H_fH_0`$ one would expect a rectangular lineshape with cut-offs at $`H_0\pm H_f`$. An exact derivation of the NMR line $`P(H)`$ yields
$$P(H)=(H_0^2+H^2H_f^2)/(4H_fH^2)$$
(1)
for $`|HH_0|<H_f`$. The component labelled ”ordered” in Fig. 4 is a fit with such a model with $`H_f=0.16`$ T and a narrow distribution (0.02 T HWHM) of this frozen field. The solid line in the top panel is a global fit to the NMR line, including the impurity phase. In order to produce a well-defined amplitude of the frozen field at the vanadium site, short-range order of the six neighboring Cu<sup>2+</sup> moments must exist. In a classical picture, the energy is minimized on the kagomé lattice, when the spins are at 120 from each other on each triangle. Following most of the theoretical works Reimers and Berlinsky (1993), we assume a $`\sqrt{3}\times \sqrt{3}`$ type short range order, sketched in the inset of Fig. 4, which corresponds to alternating chiralities (+ and - signs) on neighboring triangles. Such a short range order indeed leads to a well defined $`H_f=3H_{Cu}`$ resulting field at the vanadium site, where $`H_{Cu}`$ is the frozen field arising from each Cu<sup>2+</sup>. On the contrary, a uniform chirality order ($`𝐪=0`$) would lead to a null field at the vanadium site which cannot reproduce our NMR data. In addition, the classical $`\sqrt{3}\times \sqrt{3}`$ order favors local collective excitations made of a coherent out-of-plane rotation of the six spins belonging to a same hexagon as depicted in the inset of Fig. 4. It is remarkable that such excitations, which cost zero energy, do not affect the resulting field at the vanadium site, neither by hyperfine nor dipolar coupling, and are, therefore, ”filtered” out at the symmetric vanadium site. This could explain why we measured a small $`1/T_1`$ which, besides, decreases at low temperature, as in usual static phases, whereas, in $`\mu `$SR experiments, muons which sit in less symmetric positions, still feel a strong dynamics which persists below $`T_g`$ Fukaya et al. (2003); Bert et al. (2004). From $`H_f`$ and the hyperfine coupling constant, we get $`0.41\mu _B`$ for the frozen Cu<sup>2+</sup> moment contributing to this slow relaxing component. This small value, compared to 1$`\mu _B`$ expected for a spin 1/2, demonstrates that zero point quantum fluctuations strongly affect the volborthite GS.
In the bottom panel, combining the formerly discussed slow relaxing components with a frozen gaussian-like component (”disordered” label) results in the solid line which reproduces well the short time experimental spectra. Taking into account the different $`T_2`$ values, we evaluate the corresponding sample fractions, in the limit $`\tau 0`$, to be $``$80% for the gaussian like component and $``$20% for the rectangular-shaped one. A gaussian-like frozen component is easily obtained provided that the Cu<sup>2+</sup> frozen moments belonging to a same hexagon are randomly oriented as in the case of a spin glass. In this random picture, we extract also a small $`0.44\mu _B`$ value for the frozen copper moment consistent with the previous one. A gaussian like component could also come from several frozen field values, smaller than in the $`\sqrt{3}\times \sqrt{3}`$ case and slightly distributed, arising from various other spin configurations. A crude illustration of this scenario can be obtained if one assumes a random distribution of the chiralities on the kagomé lattice. One finds then that 4% of the hexagons are in the $`\sqrt{3}\times \sqrt{3}`$ configurations while the other configurations lead to either $`H_f=0`$ (54%) or $`H_f=\sqrt{3}H_{Cu}`$ (41%). Our experimental value already indicates that, within this classical framework, the $`\sqrt{3}\times \sqrt{3}`$ configuration has to be favored by some mechanism. An exact quantum calculation of the possible states in a finite kagomé spin 1/2 cluster would certainly allow a better quantitative understanding of this gaussian-like component in this scenario. At the nanoscopic scale of the NMR probe, it is difficult to decide whether these two vanadium sites in the GS reveal different domains or appear as a mixture in a single phase. However, both configurations freeze at approximatively the same temperature and, for both, we found a similar frozen fraction of the Cu<sup>2+</sup> moments. Both arguments favor an original single phase description.
The disorder of the GS is a challenge to our understanding of highly frustrated magnets, given that volborthite is probably the purest model system known so far. From a comparison with Zn diluted samples, we estimated an upper limit of 1% spin vacancy like defects in the pure volborthite sample. Theoretical works Chandra et al. (1993); Ferrero et al. (2003) have put forward the possibility of an intrinsic ”topological” spin glass state arising from frustration alone contrary to usual spin glasses where both frustration and disorder are responsible for glassiness. Alternatively, in Ref. Bono et al. (2004b), it was argued that a small amount of disorder could induce a spin glass like state provided that a coherent RVB background couples very efficiently the defect centers. This scenario naturally explains the simultaneous occurrence of the glassy-like transition and the dynamical plateau in $`\mu `$SR experiments and is consistent with our finding of a two component NMR signal, arising from vanadium nuclei far and close to a defect.
In conclusion , volborthite which presents all the well established signatures of a spin liquid, namely no freezing at $`T\theta _{CW}`$ and a dynamical plateau at $`T0`$, allows, for the first time, a detailed NMR local investigation of the GS. This enabled us to study the internal field configurations and their dynamics. Further, this opens new avenues for refined theoretical calculations which are necessary to reveal the influence of the possible dissymmetry of the interactions and the actual texture of the GS. |
warning/0507/astro-ph0507549.html | ar5iv | text | # A possible black hole in the gamma-ray microquasar LS 5039
## 1 Introduction
The third EGRET catalog contains 271 sources detected at energies above 100 MeV (Hartman et al., 1999). Apart from extragalactic sources at high galactic latitudes and some galactic pulsars and supernova remnants, the majority of these sources, $``$168 or $``$62 per cent, still remains unidentified. Among them, there are 72 sources located at low galactic latitudes, having $`|b|`$$`<`$10$`\mathrm{°}`$, which represents around 45 per cent of the unidentified sources. Therefore, several of these objects are presumably of galactic nature. The existence of similar properties between some of these sources, indicate that there are at least three different groups of galactic populations: a group of young stellar objects and star-forming regions (Romero, 2001), the sources forming a halo around the galactic center, and finally a group of sources correlated with the Gould Belt (Grenier, 2000) (see Romero et al. 2004 and Grenier 2004 for recent updates). The identification of their nature is of prime importance in high energy astrophysics (see Cheng & Romero 2005 for a recent compilation).
On the other hand, microquasars are galactic X-ray binaries with relativistic collimated jet emission. The compact object (a neutron star or a black hole) accretes matter from a companion star, which can be a massive early type star or a low-mass late type star. Since the first discoveries, one decade ago, microquasars have raised a strong astrophysical interest because they provide small scale counterparts of the highly energetic outflows seen in AGNs and quasars. Their short characteristic time-scales allow us to study the physics of accretion and outflow in a much faster way than in AGNs. Furthermore, because of their binary nature, one can derive accurate orbital parameters and masses for the accreting compact object and its donor star, which may be correlated with jet properties. There are currently around 15 microquasars known (Ribó 2005) from a population of $``$300 X-ray binaries (Liu, van Paradijs & van den Heuvel 2000, 2001), but only a few with well determined system parameters. Detailed reviews on microquasars can be found in Mirabel & Rodríguez (1999) and Fender (2005).
LS 5039 is the optical counterpart of the X-ray source RX J1826.2$``$1450 and it was proposed as a High Mass X-ray Binary (HMXB) by Motch et al. (1997). Based on archival NVSS data and their own VLA observations Martí, Paredes & Ribó (1998) identified its radio counterpart, which appeared to be a persistent non-thermal radio source. Subsequent VLBA observations, performed by Paredes et al. (2000), revealed relativistic radio jets at milliarcsecond scales, which qualified LS 5039 as a microquasar. In the same work, Paredes et al. draw a possible connection with the unidentified gamma-ray source 3EG J1824$``$1514, and suggest, for the first time, that microquasars could be sources of high-energy $`\gamma `$-ray emission above 100 MeV (see Paredes 2005 and Ribó, Combi & Mirabel 2005 for reviews on other associations). Moreover, LS 5039 has been associated very recently with the very high energy gamma-ray source HESS J1826$``$148, detected at energies above 250 GeV, reinforcing the association with the EGRET source (Aharonian et al., 2005). The optical component of the LS 5039 system has been classified as a O6.5 V((f)) by Clark et al. (2001) and McSwain et al. (2001, 2004; hereafter M01 and M04, respectively) derived the first orbital parameters, with a proposed orbital period of 4.4267 days and a highly eccentric orbit of $`e=0.48\pm 0.06`$. In addition, evidence for intrinsic polarization at the 3 percent level was presented in Combi et al. (2004) and interpreted as Thomson scattering in the stellar envelope. The present paper reports novel spectroscopic observations of LS 5039 that support a revised orbital period of 3.9060 days, a higher mass function and significantly lower eccentricity than previous claims. Our results have important implications for the nature of the compact object, the presence of CNO products in the optical companion, the mass loss of the supernova (SN) explosion, the recoil velocity and runaway nature of LS 5039 (see Ribó et al. 2002 and M04), as well as for the interpretation of the existing X-ray data (see Reig et al. 2003, Bosch-Ramon et al. 2005, and references therein) and very high energy gamma-ray data (Aharonian et al., 2005).
The observational details are outlined in section 2. Section 3.1 presents the analysis of the radial velocities and orbital parameters while section 3.2 focuses on the rotational velocity calculation of the optical star. The determination of the stellar parameters and chemical abundances are described in section 3.3. In section 4 we present the implications of our new orbital solution for the nature of the compact object. Based on all the updated parameters we conduct a general discussion in section 5.
## 2 Observations and Data Reduction
We observed LS 5039 using the Intermediate Dispersion Spectrograph (IDS) attached to the 2.5-m Isaac Newton Telescope (INT) at the Observatorio del Roque de Los Muchachos on the nights of 23–31 July 2002 and 1–10 July 2003. A total of 196 spectra were obtained with the combination of the 235 mm camera and the R900V grating which provided a useful wavelength coverage (free from vignetting) of $`\lambda \lambda `$3900–5500. We used integration times of 300–900 s. The seeing was variable (1–2 arcsec) during our runs and we used a 1.2-arcsec slit which resulted in a resolution of 83 km s<sup>-1</sup> (FWHM). In addition, we also obtained five spectra of LS 5039 with the holographic grating H2400B with the aim of measuring the rotational broadening ($`v\mathrm{sin}i`$) of the optical companion’s absorption lines. The spectral resolution of these spectra was 30 km s<sup>-1</sup> and we varied the central wavelength in order to fully cover the range $`\lambda \lambda `$3900–5100, where most of the prominent He i and He ii lines lie. The standard star HD 168075 B, of spectral type O7 V((f)), was also observed with the same instrumental configurations for the purpose of rotational broadening analysis. Furthermore, we obtained 14 H$`\alpha `$ spectra at 58 km s<sup>-1</sup> resolution on the nights of 4–7 July 2003 to derive the mass-loss rate in the optical star. These spectra were also taken simultaneously with RXTE observations with the aim of studying the correlation between X-ray flux and mass-loss rate proposed by Reig et al. (2003) and confirmed by M04. These results are discussed in Bosch-Ramon et al. (2005). A full observing log is presented in Table LABEL:log.
The images were de-biased and flat-field corrected, and the spectra subsequently extracted using conventional optimal extraction techniques in order to maximize the signal-to-noise ratio of the output (Horne 1986). CuAr and CuNe comparison lamp images were obtained every 15–30 minutes, and the $`\lambda `$-pixel scale was derived through 4/6th-order polynomial fits to 53/86 lines (depending on the set-up), resulting in an rms scatter $`<0.04`$ Å. The calibration curves were interpolated linearly in time.
## 3 Analysis
### 3.1 Radial Velocities and Period Search
All the spectra were rectified, by fitting a low-order spline to the continuum, and re-binned into a uniform velocity scale of 83 km s<sup>-1</sup>. We show in Fig. 1 the averaged spectrum of LS 5039 along with the template star HD 168075 B of spectral type O7 V((f)). The ratio He i$`\lambda `$4471/He ii$`\lambda `$4541 is somewhat steeper for LS 5039, which suggests a slightly earlier spectral type than O7 V, in good agreement with the O6.5 V spectral type proposed by Clark et al. (2001). Therefore, a synthetic O6.5 V template was generated for the cross-correlation analysis. This was computed using the NLTE library OSTAR2002 (Lanz & Hubeny, 2003) for $`T_{\mathrm{eff}}=\mathrm{39\hspace{0.17em}000}`$ K, $`\mathrm{log}g=3.85`$, $`v\mathrm{sin}i=113`$ km s<sup>-1</sup> (see Sects. 3.2 and 3.3) and degraded to our instrument resolution of 83 km s<sup>-1</sup>. Every R900V spectrum of LS 5039 was then cross-correlated with the synthetic spectrum, after subtracting the continuum. The main IS absorption lines and bands at $`\lambda `$3934 (Ca ii K), $`\lambda `$4430, $`\lambda `$4501, $`\lambda `$4726, $`\lambda `$4762, $`\lambda `$4885, $`\lambda `$5449 and $`\lambda \lambda `$5487–5550 were masked, leaving all Balmer and He lines for the analysis. We present the resulting velocities in Fig. 2. Note the night-to-night variability which supports a $``$4-day periodicity, in line with the results of M04. Some nights also show evidence for superimposed short time-scale ($``$hr) variability that is likely produced by a contaminating wind component.
In order to obtain the orbital period of LS 5039 we have performed a timing analysis on a database including our radial velocities (196 data points) and the ones reported in M01 and M04 (54 data points), to get advantage of the long baseline and to break the 1-year alias of our window function. We have noted a small systematic offset of $``$6 km s<sup>-1</sup> between McSwain’s velocities and ours (obtained by computing the running mean of the two databases) and this has been corrected before the analysis. Because the orbital modulation may not be sinusoidal due to a possible eccentricity (see M01 and M04), we have employed the phase dispersion minimization (PDM) algorithm (Stellingwerf 1978), that is better suited for non-sinusoidal periodicities (see, e.g., Otazu et al. 2002, 2004). For completeness we have also used the standard Scargle Fourier Transform (Scargle 1982) and the CLEAN algorithm (Roberts, Lehár & Dreher 1987). All the periodograms were computed in the frequency range $`\nu =0.05`$$`3`$ cycle d<sup>-1</sup> with resolution $`1\times 10^5`$ cycle d<sup>-1</sup>. We show in Fig. 3 the power spectra obtained by the three methods and they all provide significant peaks at a frequency of 0.2560$`\pm `$0.0001 cycle d<sup>-1</sup>, corresponding to 3.906$`\pm `$0.001 d. We plot in Fig. 4 all the radial velocities folded on the 4.4267 d period of M04 and our favoured 3.906 d period. We note that although our radial velocity measurements taken on July 2002 were compatible with the ephemeris of M04, as it is stated in their paper, the new measurements acquired on July 2003 are not, therefore ruling out the 4.4267 d period. There are several INT data points that deviate from the general trend of the radial velocity curve with the 3.906 d period around phase 0.45, which correspond to the excursion around HJD 2 452 827.5 visible in Fig. 2. The spurious data point of McSwain around phase 0.4 corresponds to the first value reported in M04, which was obtained from a single spectrum in that night and could be the result of a short-term excursion as the one quoted above. Since the radial velocity curve is not sinusoidal, we subsequently decided to fit an eccentric orbit model (see Wilson & Devinney 1971 and van Hamme & Wilson 2003<sup>1</sup><sup>1</sup>1The original Wilson & Devinney code has suffered major upgrades since its first release, including significant improvements in the underlying physical models. The most recent version of the code together with the relevant documentation can be found in ftp://astro.ufl.edu/pub/wilson/lcdc2003.) to our database. As McSwain’s velocities have no error bars we assigned equal weight to all the velocities in the fit. Our best solution yields $`P_{\mathrm{orb}}`$ = 3.90603$`\pm `$0.00017 d and $`T_0`$=HJD 2 451 943.09$`\pm `$0.10 (i.e. time of periastron passage). The rest of the orbital elements are listed in the first column of Table LABEL:esolutions.
Since hydrogen lines in early-type stars may be contaminated by wind emission (Puls et al., 1996), we decided to treat different line species separately. Therefore, we extracted sets of radial velocities for the H i, He i and He ii lines from our INT spectra. Figure 5 presents the radial velocity curves for the three groups of lines folded on the 3.90603 d period and $`T_0`$=HJD 2 451 943.09. We have also fitted these velocity curves with eccentric orbital models, but using the data errors for the weighting scheme and fixing $`P_{\mathrm{orb}}`$ and $`T_0`$ to the above values. Table LABEL:esolutions lists the best-fitting parameters for the separated groups of lines, and the corresponding orbital solutions can be seen overplotted in Fig. 5. We note a trend in the $`\gamma `$-velocities towards redder values as one moves from Balmer to the higher excitation He i and He ii lines. Blue-shifted velocities are usually considered a signature of stellar wind, due to P-Cygni contamination of the photospheric profiles, and can be conspicuous in the low members of the Balmer series and some He i lines, such as $`\lambda `$4471. Furthermore, the fitted solutions to the Balmer and He i lines present significantly larger rms scatter than the He ii solution. This is also consistent with a tentative scenario of variable wind contamination in Balmer and He i lines. Therefore, we decided to give more credit to the He ii solution, which will be adopted as the true orbital elements of LS 5039 for the remainder of the paper. As a test, we have also computed a combined solution using the weighted mean of every orbital parameter and we find this to be fully consistent with the He ii solution within errors, except for the systemic velocity which is obviously on the blue side. Our $`K`$-velocity yields a larger $`f(M)`$ and compact object’s mass range than in previous works (M01, M04). To better illustrate the resulting orbit, we show in Fig. 6 the relative motion of the compact object around the optical companion as seen from above (i.e., for an observer with $`i=0\mathrm{°}`$).
### 3.2 Rotational Broadening
As a first approach, we have followed the technique applied to 1A 0620$``$00 by Marsh, Robinson & Wood (1994) and described in their paper. Essentially, we subtract different broadened versions of the O7 V((f)) template HD 168075 B from our Doppler corrected average spectrum of LS 5039 and perform a $`\chi ^2`$ test on the residuals. The O7 V((f)) spectrum was broadened by convolution with the rotational profile of Gray (1992) which assumes a linearized limb darkening coefficient $`ϵ`$. We have taken $`ϵ=0.23`$ which is appropriate for the stellar parameters of our star ($`T_{\mathrm{eff}}=\mathrm{39\hspace{0.17em}000}`$ K, $`\mathrm{log}g=3.85`$, see Sect. 3.3) in the $`B`$-band. We performed the analysis independently for the two groups of spectra at different resolutions, 83 and 30 km s<sup>-1</sup>, and find minimum $`\chi ^2`$ for broadenings of $`105\pm 2`$ and $`126\pm 12`$ km s<sup>-1</sup>, respectively. Since the template also has a rotational velocity of 79 km s<sup>-1</sup> (Penny 1996), we need to sum this quadratically to get the true rotational velocity of LS 5039. This yields $`v\mathrm{sin}i=131\pm 2`$ and $`149\pm 12`$ km s<sup>-1</sup> for the two resolutions. These numbers can be compared with previous determinations of $`131\pm 6`$ km s<sup>-1</sup> (M01) and $`140\pm 8`$ km s<sup>-1</sup> (M04).
Alternatively, we attempted a refined determination following the Fourier technique described by Gray (1992); for a recent application see Royer et al. (2002). This technique takes advantage of the fact that the Fourier transform of the rotational profile has zeroes at regular positions that depend on the projected rotational velocity. The whole line profile, being a convolution of different contributing profiles in the wavelength domain, conserves these zeroes in the Fourier domain. While the natural profile does not introduce additional zeroes, other broadening mechanisms may do, although at the large rotational velocities of the O stars the firsts zeroes of the transformed profile are expected to be due to rotation. However, as the Stark effect strongly dominates the line profiles of Balmer and He ii lines, we have only used the He i lines for the $`v\mathrm{sin}i`$ determination. This method yields $`v\mathrm{sin}i=113\pm 8`$ km s<sup>-1</sup>, with the error obtained from the dispersion of the individual lines. This is a significantly lower projected rotational velocity than any of the previously obtained ones. However, this lower value is consistent with the finding that a part of the line broadening usually attributed to rotation is probably due to some kind of turbulence (see for example Ryans et al. 2002) and, therefore, we give more credit to this latter determination. We note that for the fit to the line profiles (next section) we have used a projected rotational velocity of 150 km s<sup>-1</sup> in good agreement with M04.
### 3.3 Stellar Parameters
We have determined the stellar parameters using the latest version of FASTWIND (Santolaya-Rey, Puls & Herrero 1997; Repolust, Puls & Herrero 2004; Puls et al. 2005). This is a NLTE, spherical, mass-losing model atmosphere code particularly optimized for the analysis of massive OBA stars. We have fitted simultaneously the following lines: H$`\alpha `$, H$`\beta `$, H$`\gamma `$, H$`\delta `$, He i $`\lambda `$4387, $`\lambda `$4471, $`\lambda `$4922, and He ii $`\lambda `$4200, $`\lambda `$4541, $`\lambda `$4686. From the analysis we obtain $`T_{\mathrm{eff}}=\mathrm{39\hspace{0.17em}000}\pm \mathrm{1\hspace{0.17em}000}`$ K and $`\mathrm{log}g=3.85\pm 0.10`$. While our temperature is consistent with that of M04 our gravity is slightly lower than the one adopted by these authors (although adopted errors allow for a marginal agreement). Note, however, that their gravity, which was determined by model fitting only to the wings of the H$`\gamma `$ profile, is likely biased upward by the blend of N iii lines (see details in M04). We also confirm that LS 5039 is contaminated by CNO products (see M04), with strong N and weak C lines, as compared to stars of similar spectral type. It can therefore be classified as a member of the ON type group stars, i.e., its complete spectral type is ON6.5 V((f)), as given by M04.
Adopting the best-fitting extinction parameters given by M04 (i.e. $`E(BV)=1.28\pm 0.02`$ and $`R=3.18\pm 0.07`$ which leads to $`A_\mathrm{V}=4.07\pm 0.11`$, different from what they quote), the updated $`M_\mathrm{V}=4.77\pm 0.15`$ for an O6.5 V star from the new calibration by Martins, Schaerer & Hillier (2005), and $`V=11.33\pm 0.02`$ and $`V=11.32\pm 0.01`$ mag from Clark et al. (2001) we derive a distance of $`2.54\pm 0.04`$ kpc. We have also considered the slightly different values of $`V`$ reported in the literature, ranging from 11.20 to 11.39 mag with typical 1$`\sigma `$ uncertainties of 0.03 mag (Drilling 1991; Lahulla & Hilton 1992; Martí et al. 1998; Martí et al. 2004), and computed the corresponding distances. The weighted mean and standard deviation of all these values provides a more realistic distance estimate of $`d=2.5\pm 0.1`$ kpc, that will be adopted for the remainder of the paper. With these numbers we get a radius of $`R_\mathrm{O}=9.3_{0.6}^{+0.7}`$ R (see, e.g. Herrero, Puls & Najarro 2002). From this radius we derive a luminosity of $`\mathrm{log}(L_\mathrm{O}/`$L$`{}_{}{}^{})=5.26\pm 0.06`$ and a mass of $`M_\mathrm{O}=22.9_{2.9}^{+3.4}`$ M.
The mass-loss rate has been derived from the H$`\alpha `$ profile, which appears to be variable on a secular time-scale (Reig et al. 2003; M04; Bosch-Ramon et al. 2005). Our 2003 data show a very stable equivalent width with a mean value of $`EW=2.8\pm 0.1`$ Å and no variations above a 5 per cent level over 1 orbital cycle. Despite this, we have selected two extreme profiles showing minimum and maximum $`EW`$ for the calculation, that we have performed by using again the latest version of FASTWIND (Puls et al. 2005; see examples of use in Repolust et al. 2004). For the lower state (i.e., largest absorption) we obtain a wind mass-loss rate of $`3.7\times 10^7`$ M yr<sup>-1</sup>. While we can put an upper limit for this state of $`5.0\times 10^7`$ M yr<sup>-1</sup>, the lower limit is much more uncertain, as the sensitivity of H$`\alpha `$ to mass-loss rate falls off rapidly at such low values. For the high state, we obtain a best value of $`7.5\times 10^7`$ M yr<sup>-1</sup>, i.e., a factor of two if we interpret the small difference in the profiles as due to the stellar wind and not to the low signal-to-noise ratio. Again, $`5.0\times 10^7`$ M yr<sup>-1</sup> can be considered a lower limit for this state, while we can set an absolute upper limit to the high state mass-loss rate of $`1.0\times 10^6`$ M yr<sup>-1</sup>. With these values, the Modified Wind Momentum of the upper state coincides with the values quoted by Repolust et al. (2004), being a factor of two lower for the low state, but still within their uncertainties. We also note that our values are a factor $``$4–8 higher than those derived by M04 from data acquired in 2002, but compare well with average mass-loss rates in O6.5 V stars (Howarth & Prinja, 1989).
## 4 Mass of the compact object
Our revised mass function and stellar parameters provide new constraints on the mass of the compact object in LS 5039. First of all, considering the adopted value of $`f(M)=0.0053\pm 0.0009`$ M and using an inclination of $`i=90\mathrm{°}`$ we obtain a strict lower limit for the mass of the compact object as a function of $`M_\mathrm{O}`$. This limit is represented by the lowest solid line in Fig. 7, where we also plot similar relationships for other inclination angles. For the range of possible masses of the optical companion, $`22.9_{2.9}^{+3.4}`$ M, we obtain a lower limit on the mass of the compact object in the range 1.34–1.61 M, with an additional linear 1$`\sigma `$ uncertainty of 0.11 M due to the mass function error bar.
Another constraint comes from the fact that no X-ray eclipses are seen in LS 5039. With the new ephemeris reported here, the BeppoSAX observations performed by Reig et al. (2003) cover phases between 0.969 and 0.205, while an eventual X-ray eclipse should be centered at phase 0.058 (see Fig. 6). In fact, periastron takes place at $`t=(1.0\pm 0.9)\times 10^4`$ s in figure 1 of Reig et al. (2003), while superior conjunction of the compact object takes place at $`t=(3.0\pm 0.9)\times 10^4`$ s, where the errors come from the uncertainty in our $`T_0`$. The X-ray flux is nearly constant during this last interval, implying that X-ray eclipses can be definitively ruled out in LS 5039. This condition has been used to compute the lower limit on $`M_\mathrm{X}`$ as a function of $`M_\mathrm{O}`$ (which in turn is also a function of $`R_\mathrm{O}`$). This is represented as a dashed line in Fig. 7, which yields $`i<64.6\mathrm{°}`$ and $`M_\mathrm{X}>1.49`$ M for the case of $`M_\mathrm{O}=20.0`$ M, and $`i<63.3\mathrm{°}`$ and $`M_\mathrm{X}>1.81`$ M for $`M_\mathrm{O}=26.3`$ M. Therefore, this condition provides a compact object mass above $`1.49\pm 0.11`$ M (the errorbar due to the $`f(M)`$ uncertainty). This value is still below the 1.75–2.44 M neutron star mass in Vela X-1 (Quaintrell et al., 2003), or the (95 per cent confidence) lower limit of 1.68 M for at least one of the two pulsars in binaries with eccentric orbits within the globular cluster Terzan 5 (Ransom et al., 2005). Therefore, all the previously obtained values are compatible with a massive or even a canonical neutron star.
A further constraint can be obtained by assuming that the optical companion star is pseudo-synchronized, i.e. its rotational and orbital angular velocities are synchronized at periastron passage (Kopal 1978; Claret & Giménez 1993). We have estimated the synchronization time-scale of the optical component of LS 5039 using the formalism by Zahn (1989) (as formulated in Claret & Cunha 1997) and the tidal evolution parameters of the stellar models of Claret (2004). We find a synchronization time-scale of about 1 Myr. This value is similar to the (rough) upper limit to the age of LS 5039 (Ribó et al., 2002) and, therefore, the system appears to have had time to reach orbital pseudo-synchronism. Note, however, that both the synchronization timescale and the age are only accurate as an order of magnitude estimate and, consequently, we can only state that both quantities do not exclude each other. On the other hand, the theoretical orbital circularization time-scale comes out to be a much larger value of 10 Myr, compatible with the observation of an eccentric orbit. Therefore, assuming pseudo-synchronization (P-S) we can combine our determination of $`v\mathrm{sin}i`$ and radius to estimate the binary inclination $`i_{\mathrm{P}\mathrm{S}}`$, i.e.
$$v\mathrm{sin}i=2\pi FR_\mathrm{O}\mathrm{sin}i_{\mathrm{P}\mathrm{S}}P_{\mathrm{orb}}^1,$$
where $`F=(1+e)^{1/2}/(1e)^{3/2}=2.22\pm 0.17`$, and we find values of $`i_{\mathrm{P}\mathrm{S}}`$ in the range 26.8–23.2°, depending on $`M_\mathrm{O}`$. These inclinations set strong constraints to the compact object’s mass, as is displayed by the thicker solid line in Fig. 7. The possible values of $`M_\mathrm{X}`$ range from 3.14 to 4.35 M. The 1$`\sigma `$ uncertainty region (obtained through propagating errors in $`i_{\mathrm{P}\mathrm{S}}`$ and $`f(M)`$) is marked by the dark-grey area in Fig. 7. This leads to compact object masses in the range 2.75–5.00 M. The 3$`\sigma `$ uncertainty region is also indicated in Fig. 7 with a light-grey area and yield masses in the range 2.14–6.78 M. The central value of $`M_\mathrm{O}=22.9`$ M implies $`i_{\mathrm{P}\mathrm{S}}=24.9\pm 2.8\mathrm{°}`$ (1$`\sigma `$ uncertainty) and $`M_\mathrm{X}=3.7_{1.0}^{+1.3}`$ M (1$`\sigma `$ uncertainty in all involved parameters), and is indicated by a cross in Fig. 7. Therefore, if the plausible assumption of pseudo-synchronization is correct, the compact object is consistent with being a black hole.
Since there is no significant H$`\alpha `$ emission and the binary system does not exhibit strong X-ray outbursts during the periastron passage (see, e.g., Bosch-Ramon et al. 2005), the donor star is not expected to overflow its Roche lobe at any orbital phase. To check this, we have generated a grid of solutions in the following way: for each possible optical companion mass in Fig. 7 we have iterated for all possible inclination angles and obtained the corresponding $`M_\mathrm{X}`$ (and its error) through the mass function. Then $`M_\mathrm{X}`$ is used to compute the Roche lobe at periastron and its uncertainty (using the semi-analytical expression in Eggleton 1983). We find that, for the possible range of optical star masses $`M_\mathrm{O}=20.0`$$`26.3`$ M and within 1$`\sigma `$, the Roche lobe is not overflown at periastron for $`M_\mathrm{X}<8`$ M (or $`i>13.4\mathrm{°}`$).
A final constraint on the inclination angle comes from the rotational velocity. The break-up speed, or critical rotational velocity, can be obtained by means of $`v_{\mathrm{crit}}=\sqrt{(2/3)GM_\mathrm{O}/R_\mathrm{O}}`$ (see Porter 1996). Considering the limits of our likely values for the mass and radius of the optical companion we obtain $`v_{\mathrm{crit}}=540`$–580 km s<sup>-1</sup>. Finally, by using our measured rotational velocity of $`v\mathrm{sin}i=113\pm 8`$ km s<sup>-1</sup>, we can constrain the inclination angle to be above 11–12°, which in turn provides upper limits for the compact object mass of 8–10 M (depending on $`M_\mathrm{O}`$).
In summary, the lack of X-ray eclipses, Roche lobe overflow, and break-up speed constrain the inclination angle to be in the range 13–64°, and the compact object mass to 1.5–8 M. If the primary in LS 5039 is pseudo-synchronized the inclination angle is $`24.9\pm 2.8\mathrm{°}`$, and the compact object mass $`M_\mathrm{X}=3.7_{1.0}^{+1.3}`$ M.
The inclination angle could be tested through accurate photometry near the periastron passage. We have simulated the light curve of the tidally distorted optical component using our orbital parameters and predict a $``$0.02 mag increase in brightness near the periastron for $`i=45\mathrm{°}`$, which could be revealed by 2–3 mmag accuracy photometry. If this variability is not detected or is below 0.01 mag, the inclination angle must be below 30°, and the mass of the compact object above 3.0 M, confirming its black hole nature. We have looked for this effect in our existing photometry (Martí et al., 2004) but the results are not conclusive due to noise and scarce phase sampling. More observations are currently underway.
## 5 Discussion
In this section we evaluate the implications of all the updated parameters on several properties of LS 5039.
### 5.1 Optical companion
The contamination of the optical companion by CNO products may be a consequence of mass transfer of CNO-processed material from the supernova progenitor prior to the explosion (M04). Alternatively, it might be caused by mixing processes (induced by rotation) which brings CNO processed material from the inner regions into the stellar atmosphere (Heger, Langer & Woosley, 2000). Indeed, if the system is pseudo-synchronized, then our measured $`v\mathrm{sin}i=113\pm 8`$ km s<sup>-1</sup> translates into a large rotational velocity of $`v=268\pm 34`$ km s<sup>-1</sup>, which is $``$0.5$`v_{\mathrm{crit}}`$. Therefore, although we cannot discard the hypothesis of mass transfer prior to the SN explosion, we favour the mixing scenario to explain the presence of CNO products. We also note that the optical companion in LS 5039 is slightly undermassive for its spectral type, since from the calibration of Martins et al. (2005) an O6.5 V star should have $`M_{\mathrm{spec}}=29.0`$ M, while our measured mass of $`M_\mathrm{O}=22.9_{2.9}^{+3.4}`$ M covers spectral types between O7 V and O8.5 V in their calibration. This effect has been observed in other HMXB (see, e.g., van den Heuvel et al. 1983).
### 5.2 X-ray variability and accretion/ejection energetic balance
An interesting consequence of our orbital solution is the smaller eccentricity, $`e=0.35\pm 0.04`$, compared to McSwain’s solution, $`e=0.48\pm 0.06`$. A lower eccentricity is easier to reconcile with the weak orbital X-ray modulation, with the X-ray flux varying a factor of 2.5, observed by Bosch-Ramon et al. (2005). In a simplistic Bondi-Hoyle scenario (i.e. spherical accretion through winds) our eccentricity, together with the updated masses, provides an X-ray variability with a factor of 12 between maximum and minimum flux, which is still a factor $``$5 higher than observed (see Reig et al. 2003 for details about the method). Even when considering that all possible uncertainties conspire in one or the opposite direction, the expected variability is in the range 7–24, which is a factor 3–10 higher than observed. This led Bosch-Ramon et al. (2005) to propose the existence of an accretion disc that would partly screen the compact object from direct wind accretion.
We can now compare the luminosity released in the vicinity of the compact object (i.e., excluding the contribution from the companion which dominates in the optical domain) with the available accretion luminosity. The first one includes the radio, X-ray, high energy, and very high energy (VHE) gamma-ray domains. It is maximum in the high energy gamma-ray domain since, using our new distance estimate of 2.5 kpc, we have: $`L_{\mathrm{radio},0.1100\mathrm{GHz}}=8\times 10^{30}`$ (Martí et al., 1998), $`L_{\mathrm{X},330\mathrm{keV}}=1.0`$$`2.5\times 10^{34}`$ (Bosch-Ramon et al. 2005, after removing the diffuse background contamination), $`L_{\gamma ,>100\mathrm{MeV}}=2.7\times 10^{35}`$ (Hartman et al., 1999), and $`L_{\mathrm{VHE},>250\mathrm{GeV}}=4\times 10^{33}`$ erg s<sup>-1</sup> (Aharonian et al., 2005). Regarding the accretion luminosity, we have used the Bondi-Hoyle scenario to obtain a rough estimate of the accreted matter and the energy that can be released from the accretion process. By using the updated parameters ($`M_\mathrm{O}=22.9`$ M, $`R_\mathrm{O}=9.3`$ R, $`M_\mathrm{X}=3.7`$ M, $`R_\mathrm{X}=R_{\mathrm{Sch}}=2GM_\mathrm{X}/c^2=2.95(M_\mathrm{X}/\mathrm{M}_{})`$ km, $`P_{\mathrm{orb}}=3.90603`$ d, $`e=0.35`$, $`\dot{M}=5\times 10^7`$ M yr<sup>-1</sup>), $`v_{\mathrm{inf}}=2440`$ km s<sup>-1</sup> from M04 and $`\beta =0.8`$ we obtain an accretion luminosity averaged through the orbit of $`<L_{\mathrm{acc}}>=8\times 10^{35}`$ erg s<sup>-1</sup>. Therefore, around 1/3 of the accreted luminosity would be radiated within the relativistic jet of LS 5039, while the remaining available luminosity could be lost by the advection of matter towards the black hole. Of course, the scenario is not so simple. First of all, we should include the kinetic luminosity of the jet, given by $`L_\mathrm{k}=(\mathrm{\Gamma }1)\dot{M}_{\mathrm{jet}}c^2`$ (where $`\mathrm{\Gamma }`$ is the bulk Lorentz factor of the jet, and we have neglected the energy needed to abandon the potential well). For a jet with a velocity $`\beta =v/c=0.2`$ (from $`\beta \mathrm{cos}\theta =0.17\pm 0.05`$ found by Paredes et al. 2002 and assuming $`\theta =i_{\mathrm{P}\mathrm{S}}25\mathrm{°}`$), or $`\mathrm{\Gamma }=1.02`$, we obtain $`L_\mathrm{k}=1.0\times 10^{36}`$ erg s<sup>-1</sup> (see details on how to estimate $`\dot{M}_{\mathrm{jet}}`$ in Paredes et al. 2000). This luminosity is slightly higher than the accretion luminosity, although this discrepancy could be solved by increasing the mass of the compact object up to 5 M and/or the mass-loss rate of the primary up to $`10^6`$ M yr<sup>-1</sup>. However, the Bondi-Hoyle accretion scenario is an oversimplification of the real accretion processes that take place in this source, because: 1) the presence of a thick accretion disc around the compact object seems to be necessary to launch the relativistic jets in microquasars (see, e.g. Fender 2005 and references therein); 2) the presence of a disc also appears to be necessary to explain the weak orbital X-ray variability in LS 5039 (Bosch-Ramon et al., 2005). Moreover, we note that due to fast rotation of the optical star and its proximity to fill the Roche lobe during periastron passage, we cannot discard additional mass-loss in the equatorial plane of the binary system, which could provide the needed amount of additional accretion luminosity. We note that we have neglected the energy stored in magnetic fields through all this discussion. In any case, we can say that the current estimates of accretion/ejection luminosities agree quite well by using our new set of parameters. We note that if the compact object were a neutron star with $`M_\mathrm{X}=1.4`$ M and $`R_\mathrm{X}=10`$ km we would have $`<L_{\mathrm{acc}}>=5\times 10^{34}`$ erg s<sup>-1</sup>. This value is more than one order of magnitude lower than for the case of a 3.7 M black hole, and $``$2.5 times smaller than the gamma-ray luminosity of LS 5039.
### 5.3 Before and after the SN explosion
Using our new radial velocity of the binary system $`\gamma =17.2\pm 0.7`$ km s<sup>-1</sup> and an updated proper motions estimate of $`\mu _{\alpha \mathrm{cos}\delta }=4.8\pm 0.8`$ mas yr<sup>-1</sup>, $`\mu _\delta =10.9\pm 0.9`$ mas yr<sup>-1</sup> (including a new radio position obtained with VLA+Pie Town observations; Martí, Ribó & Paredes, in preparation), we can recompute the total systemic velocity of LS 5039. We find $`v_{\mathrm{sys}}=126\pm 9`$ km s<sup>-1</sup> (see details on the method in Ribó et al. 2002). On the other hand, our improved masses and eccentricity have an impact on the formation history of LS 5039. Tidal forces act to circularize the binary orbit and hence the current eccentricity $`e=0.35\pm 0.04`$ can be taken as a lower limit to $`e_{\mathrm{post}\mathrm{SN}}`$, the post-SN eccentricity. In the context of a symmetric SN explosion, $`e_{\mathrm{post}\mathrm{SN}}`$ is related to the mass lost in the SN event $`\mathrm{\Delta }M`$ through $`\mathrm{\Delta }M=e_{\mathrm{post}\mathrm{SN}}\times (M_\mathrm{X}+M_\mathrm{O})`$ which yields $`\mathrm{\Delta }M>9\pm 2`$ M and $`P_{\mathrm{re}\mathrm{circ}}=3.2\pm 0.2`$ d (using our current values for $`P_{\mathrm{orb}}`$ and $`e`$). With these numbers, and using the equations in Nelemans et al. (1999), we obtain a theoretical recoil velocity of $`130\pm 20`$ km s<sup>-1</sup>, in good agreement with our new space velocity. The previous discrepancy between these two values reported by M04 vanishes thanks, mainly, to the lower value of $`e`$, since the equations are only slightly sensitive to $`M_\mathrm{X}`$. Therefore a high mass loss of $``$9 M during the SN explosion could provide both the eccentricity and the space velocity that we currently observe in LS 5039. We note that our predicted orbital period prior to the SN explosion of 1.8 d yields a Roche lobe radius of $`8.7`$ R, which is compatible within errors to the O star radius. Finally, the relatively large mass of the SN progenitor, around 13 M, is compatible with the $`10`$–15 M upper limit found by Fryer & Kalogera (2001).
The kinetic energy of the binary system is $`4.2\pm 0.8\times 10^{48}`$ erg, i.e., merely $`4\times 10^3`$ times the energy of a typical SN, although nearly one order of magnitude higher than in GRO J1655$``$40 (Mirabel et al., 2002). On the other hand, from the pulsar birth velocity value of $`400\pm 40`$ km s<sup>-1</sup> (Hobbs et al., 2005), and assuming a pulsar mass of 1.4 M, we obtain that their typical linear momentum is $`560\pm 56`$ M km s<sup>-1</sup>. In contrast, the linear momentum of the LS 5039 binary system is $`\mathrm{3\hspace{0.17em}350}\pm 450`$ M km s<sup>-1</sup>, well above the previous value and above the value of any individual pulsar. To our knowledge, this value is also significantly higher than in all other binary systems with measured velocities, being a possible exception 4U 1700$``$37. For this system, moving at $`70\pm 5`$ km s<sup>-1</sup>, and using different masses from Ankay et al. (2001) and Clark et al. (2002), we obtain a linear momentum in the range 2300–4200 M km s<sup>-1</sup>, encompassing the value found for LS 5039.
We note that once the orbit is re-circularized, with a period of 3.2 d, the expected Roche radius will be 15 R, well above the current radius of the star. After leaving the main sequence, the star will fill its Roche lobe, leading to unstable mass transfer that will probably turn-off the X-ray binary and microquasar phase of LS 5039 (Frank, King & Raine, 2002). From the evolutionary tracks by Meynet et al. (1994) we know that the main sequence lifetime is between 4.5 and 6.5 Myr (assuming a solar metallicity of $`Z=0.02`$ and for stars with initial masses between 40 and 25 M). On the other hand, the LS 5039 trip from the galactic midplane to its current position would take $``$0.5 Myr (explained in Ribó et al. 2002 and independent of the distance to the source). Consequently, the X-ray binary could survive as a microquasar up to 4–6 Myr from now, placing the system at galactic latitudes of $`10`$ to $`13\mathrm{°}`$. Therefore, relatively nearby systems similar to LS 5039 could be the counterparts of unidentified EGRET sources with $`|b|>5\mathrm{°}`$.
### 5.4 Orbital behaviour of the TeV counterpart
Aharonian et al. (2005) have recently reported the detection of a very high energy gamma-ray source at TeV energies, namely HESS J1826$``$148, with a position consistent at the 3$`\sigma `$ level with that of LS 5039. Moreover, the spectral energy distribution at high energies makes the association between the HESS and EGRET sources virtually certain. These authors state that no periodic variations are apparent when folding the data using the orbital ephemeris of M04. We show in Fig. 8 the HESS data folded using our new orbital period. Despite the large error bars, we notice a possible flux variation of a factor $``$3 with the orbital period, with a quasi-sinusoidal pattern and maximum around phase 0.9. This behaviour is reminiscent to the one recently found in X-rays in RX J1826.2$``$1450/LS 5039 by Bosch-Ramon et al. (2005), with a flux variation of a factor $``$2.5 and a maximum around phase 0.8. Although further HESS observations would be necessary to confirm this orbital variability, this similarity reinforces the association between LS 5039 and HESS J1826$``$148 and, therefore, practically confirms the association between the microquasar and the EGRET source.
## 6 Summary
We have reported new optical spectroscopy of the microquasar LS 5039 and obtained a new orbital solution. In particular, we find $`P_{\mathrm{orb}}=3.90603\pm 0.00017`$ d, $`e=0.35\pm 0.04`$, systemic velocity $`\gamma =17.2\pm 0.7`$ km s<sup>-1</sup> and a mass function for the compact object $`f(M)=0.0053\pm 0.0009`$ M, significantly different from previous results. We have also derived a new distance estimate of $`d=2.5\pm 0.1`$ kpc and a mass of the optical companion of $`M_\mathrm{O}=22.9_{2.9}^{+3.4}`$ M. Using this information and assuming pseudo-synchronization we obtain an inclination of $`i=25\pm 3\mathrm{°}`$, which yields to $`M_\mathrm{X}=3.7_{1.0}^{+1.3}`$ M. This strongly suggests that the compact object in LS 5039 is a black hole. With our new orbital parameters there is a good agreement between the accretion and ejection luminosities around the compact object. The space velocity of the binary system is also in good agreement with the theoretical recoil velocity in a symmetric SN explosion with a mass loss of $``$9 M. Finally, the orbital variability of the TeV counterpart is reminiscent to the one seen in X-rays, reinforcing the association between LS 5039, HESS J1826$``$148 and 3EG J1824$``$1514.
## Acknowledgments
We acknowledge I. Negueruela for help in the observations and obtaining some spectra in the 2002 and 2003 campaigns. We also thank V. Bosch-Ramon, E. Körding, and L. J. Pellizza for useful discussions. J. C. acknowledges support from the Spanish MCYT grant AYA2002-0036. M. R., J. M. P. and J. M. acknowledge partial support by DGI of the Spanish Ministerio de Educación y Ciencia (former Ministerio de Ciencia y Tecnología) under grants AYA2001-3092, AYA2004-07171-C02-01 and AYA2004-07171-C02-02, as well as additional support from the European Regional Development Fund (ERDF/FEDER). M. R. acknowledges support from the French Space Agency (CNES) and by a Marie Curie Fellowship of the European Community programme Improving Human Potential under contract number HPMF-CT-2002-02053. I. R. acknowledges support from the Spanish Ministerio de Ciencia y Tecnología through a Ramón y Cajal fellowship. J. M. is also supported by the Junta de Andalucía (Spain) under project FQM322. A. H. acknowledges support from the Spanish Ministerio de Educación y Ciencia grant AYA2004-08271-02-01. MOLLY and DOPPLER software developed by T. R. Marsh is gratefully acknowledged. The INT is operated on the island of La Palma by the Royal Greenwich Observatory in the Spanish Observatorio del Roque de Los Muchachos of the Instituto de Astrofísica de Canarias. This research has made use of the NASA’s Astrophysics Data System Abstract Service and of the SIMBAD database, operated at CDS, Strasbourg, France. |
warning/0507/cond-mat0507010.html | ar5iv | text | # Absence of single particle Bose-Einstein condensation at low densities for bosons with correlated-hopping
## Abstract
Motivated by the physics of mobile triplets in frustrated quantum magnets, the properties of a two dimensional model of bosons with correlated-hopping are investigated. A mean-field analysis reveals the presence of a pairing phase without single particle Bose-Einstein condensation (BEC) at low densities for sufficiently strong correlated-hopping, and of an Ising quantum phase transition towards a BEC phase at larger density. The physical arguments supporting the mean-field results and their implications for bosonic and quantum spin systems are discussed.
The models of interacting bosons (with or without disorder) have been a subject of active research. They are studied for a variety of reasons, coming from different experimental systems, such as Josephson junction arrays Geerligs et al. (1989), <sup>4</sup>He in porous media Chan et al. (1988), disordered films with superconducting and insulating phases Haviland et al. (1989), or more recently in the context of atoms trapped on an optical lattice Greiner et al. (2002). The interplay of interaction, disorder and kinetic energy leads to the ground states that can be a superfluid, a Bose glass, a Mott insulator or a supersolid Fisher et al. (1989); Klien et al. (2001); Alet and Sørensen (2004); Freericks and Monien (1996); Sheshadri et al. (1995); Sengupta et al. (2005); van Otterlo and Wagenblast (1994). In the context of spin models too, the Schwinger boson mean-field theories provide a useful description of magnetism in the bosonic language Arovas and Auerbach (1988); Sarkar et al. (1989); Chandra et al. (1990); Mila et al. (1991).
Over the last decade, bosons have also been used in the context of quantum magnetism to describe the magnetization process of gapped systems with a singlet ground state such as spin ladders, the triplets induced by the magnetic being treated as hard-core bosons. These bosons may condense, leading to the ordering of the transverse component of the spins, but they might as well undergo a superfluid-insulator transition, leading to magnetization plateaux Rice (2002). For pure SU(2) interactions, and without disorder, the common belief is that the only alternative, not realized so far in quantum magnets, is a supersolid, i.e. a coexistence of these phases.
In this paper, we propose that there is another possibility, namely a pairing phase without single particle Bose condensation. Our starting point is the observation that the effective bosonic model of a frustrated quantum magnet such as SrCu<sub>2</sub>(BO<sub>3</sub>)<sub>2</sub> Kageyama et al. (1999) contains, in addition to the usual kinetic and potential terms, a correlated-hopping term where a boson can hop only if there is another boson nearby, and that this term can be the dominant source of kinetic energy in geometries such as the orthogonal dimer model realized in SrCu<sub>2</sub>(BO<sub>3</sub>)<sub>2</sub> Momoi and Totsuka (2000); Miyahara and Ueda (2003). While the possibility of bound state formation was already pointed out in that context, the consequences of the presence of such a term on the phase diagram at finite densities have not been worked out yet.
For clarity, we concentrate in this paper on a minimal version of the model, but we have checked that the conclusions apply to the more realistic model derived for SrCu<sub>2</sub>(BO<sub>3</sub>)<sub>2</sub> Bendjama et al. . This model is defined on a square lattice by the Hamiltonian
$`H`$ $`=`$ $`t{\displaystyle \underset{𝐫}{}}{\displaystyle \underset{\delta =\pm x,\pm y}{}}b_{𝐫+\delta }^{}b_𝐫\mu {\displaystyle \underset{𝐫}{}}n_𝐫`$ (1)
$`t^{}{\displaystyle \underset{𝐫}{}}{\displaystyle \underset{\delta =\pm x}{}}{\displaystyle \underset{\delta ^{}=\pm y}{}}n_𝐫\{b_{𝐫+\delta }^{}b_{𝐫+\delta ^{}}+h.c.\}`$
where $`b_𝐫^{},b_𝐫`$ are boson operators and $`n_𝐫=b_𝐫^{}b_𝐫`$. $`t`$ and $`t^{}`$ are the measures of single particle and correlated-hopping respectively. A hard core constraint that excludes multiple occupancy should in principle be included. However, we will concentrate on the low density limit, where this constraint is expected to be irrelevant. So in the following we work with regular (soft core) bosons.
Since the correlated-hopping term in Eq. (1) is quartic in the single particle boson operators, the simplest thing to do is a mean-field theory. Since the system gains energy through correlated-hopping by having two particles nearby, a natural choice for a mean-field is the pairing amplitude. The particle density and the kinetic amplitudes are the other choices for the mean-fields. In the following, we formulate a mean-field theory in terms of these order parameters defined by:
$`\mathrm{\Delta }`$ $`=`$ $`b_𝐫^{}b_{𝐫\pm \delta }^{}(\text{pairing amplitude})`$
$`\kappa `$ $`=`$ $`b_𝐫^{}b_{𝐫\pm \delta },\kappa ^{}=b_{𝐫\pm \delta }^{}b_{𝐫\pm \delta ^{}}(\text{kinetic amplitudes})`$
$`n`$ $`=`$ $`b_𝐫^{}b_𝐫(\text{particle density})`$
where $`\delta \delta ^{}`$, and $`\delta ,\delta ^{}=x,y`$. The particle density is taken to be uniform, and the kinetic amplitudes real. In principle we can allow for an internal phase in the pairing amplitude (a non-zero phase between $`x`$ and $`y`$ direction bonds) like in the mean-field theory of t-J model in the context of the high-T<sub>c</sub> cuprates. Here, we take the internal phase to be zero (the extended s-wave pairing). The corresponding mean-field Hamiltonian has the following form:
$$H_{MF}=_0+\underset{𝐤}{}\{\xi _𝐤b_𝐤^{}b_𝐤\mathrm{\Delta }_𝐤[b_𝐤^{}b_𝐤^{}+h.c.]\}$$
(2)
where $`_0`$, $`\xi _𝐤`$ and $`\mathrm{\Delta }_𝐤`$ are given by
$`_0`$ $`=`$ $`8t^{}L[\mathrm{\Delta }^2+\kappa ^2+n\kappa ^{}]`$
$`\xi _𝐤`$ $`=`$ $`2(t+4t^{}\kappa )(\mathrm{cos}k_x+\mathrm{cos}k_y)`$
$`8t^{}n\mathrm{cos}k_x\mathrm{cos}k_y(\mu +8t^{}\kappa ^{})`$
$`\mathrm{\Delta }_𝐤`$ $`=`$ $`4t^{}\mathrm{\Delta }(\mathrm{cos}k_x+\mathrm{cos}k_y)`$
The Hamiltonian $`H_{MF}`$ can easily be diagonalized using Bogoliubov transformation for bosons. The canonical free energy density for $`H_{MF}`$ is given as:
$`f`$ $`=`$ $`\lambda \left(n+{\displaystyle \frac{1}{2}}\right)+8t^{}\left(\mathrm{\Delta }^2+\kappa ^2\right)`$ (3)
$`+{\displaystyle \frac{1}{2L}}{\displaystyle \underset{𝐤}{}}E_𝐤+{\displaystyle \frac{1}{\beta L}}{\displaystyle \underset{𝐤}{}}\mathrm{log}\left(1\mathrm{e}^{\beta E_𝐤}\right)`$
where $`E_𝐤=\sqrt{\xi _𝐤^24\mathrm{\Delta }_𝐤^2}`$ is the quasi-particle dispersion, $`\beta =1/k_BT`$, and $`\lambda =\mu +8t^{}\kappa ^{}`$ is the effective chemical potential. Re-defining the chemical potential in this way makes $`\kappa ^{}`$ a redundant order parameter in the mean-field theory. Note that $`\mu `$ and $`\kappa ^{}`$ appear with right combination to give $`\lambda `$ as the new chemical potential in $`\xi _𝐤`$. Hence, for a given $`n`$, $`f`$ is purely a function of $`\lambda `$, $`\kappa `$ and $`\mathrm{\Delta }`$.
The self-consistent equations for the order parameters can be written
$`n`$ $`=`$ $`n_c{\displaystyle \frac{1}{2}}+{\displaystyle \frac{1}{2L}}{\displaystyle \underset{𝐤}{}}{\displaystyle \frac{\xi _𝐤}{E_𝐤}}\mathrm{coth}{\displaystyle \frac{\beta E_𝐤}{2}}`$ (4)
$`\kappa `$ $`=`$ $`n_c+{\displaystyle \frac{1}{4L}}{\displaystyle \underset{𝐤}{}}{\displaystyle \frac{\xi _𝐤(\mathrm{cos}k_x+\mathrm{cos}k_y)}{E_𝐤}}\mathrm{coth}{\displaystyle \frac{\beta E_𝐤}{2}}`$ (5)
$`\mathrm{\Delta }`$ $`=`$ $`n_c+{\displaystyle \frac{2t^{}\mathrm{\Delta }}{L}}{\displaystyle \underset{𝐤}{}}{\displaystyle \frac{(\mathrm{cos}k_x+\mathrm{cos}k_y)^2}{E_𝐤}}\mathrm{coth}{\displaystyle \frac{\beta E_𝐤}{2}}`$ (6)
where $`n_c`$ is the condensate density (the occupancy of the zero energy mode if any). Since the model is two-dimensional, $`n_c=0`$ for $`T>0`$. At $`T=0`$, $`n_c`$ may or may not be zero, and solutions must be searched with two strategies: assume $`n_c=0`$ and solve these equations for the unknowns ($`\kappa `$, $`\mathrm{\Delta }`$, $`\lambda `$), or assume there is a zero energy mode (which fixes $`\lambda `$), and solve for the unknowns ($`\kappa `$, $`\mathrm{\Delta }`$, $`n_c`$). In that case, the wave-vector corresponding to the zero-energy mode ($`𝐤=0`$ here) must be excluded from the sum. If several solutions are found for a given density, the one with lowest energy should be chosen. In practice, we only found one solution for a given density. In general, these equations are solved by simple iteration. Note however that, when $`n_c=0`$, Eqs. (4) and (5) can be still solved by iteration, but Eq. (6) needs to be solved for $`\mathrm{\Delta }`$ by some numerical method at each step of the iteration.
Let us first discuss the $`T=0`$ results. The most remarkable feature is that it turned out to be impossible to find a solution with a non-zero condensate at low enough density unless $`t^{}`$ is very small. In other words, as soon as the correlated-hopping is not too small, there is no single-particle BEC at low density. The critical density $`n^{}`$ below which this is the case is plotted as a function of $`\alpha =t^{}/(t+t^{})`$ in Fig. 2. This figure calls for some comments. First of all, $`t^{}/t`$ need not to be large for the effect to be observable, which ensures the relevance of the present discussion for the quantum magnets such as SrCu<sub>2</sub>(BO<sub>3</sub>)<sub>2</sub>. Besides, the critical value $`n^{}`$ is quite small even when $`t^{}`$ dominates, and our approximation to treat triplets as soft-core bosons is expected to be good in the whole range of Fig. 2. Finally, densities of a few percent are definitely accessible in the context of quantum magnets, the density being equivalent to the magnetization relative to the saturation value.
Next we turn to the nature of this non-BEC phase. Clearly this cannot be a commensurate insulating phase of the type observed before since it occurs for a range of densities. In fact, its nature is best revealed by looking at the order parameters. While $`n_c=0`$ in the pairing phase and E$`{}_{gap}{}^{}=0`$ in the single particle BEC phase, the mean-field solution for $`\kappa `$ and $`\mathrm{\Delta }`$ is non-zero on both sides. In Fig. 3 the behavior of $`n_c`$, $`\kappa `$ and $`\mathrm{\Delta }`$ is shown as a function of $`n`$ for $`\alpha =0.99`$. It is not surprising that $`\mathrm{\Delta }`$ is non-zero in the single particle BEC phase. In fact, the single particle BEC state means $`b^{}0`$, which further implies that $`b^{}b^{}b^{}^20`$. Hence $`\mathrm{\Delta }`$ will always be non-zero in the single particle BEC phase. The correct measure of the existence of the pairing (independent of the contribution from the single particle BEC) is $`\mathrm{\Delta }n_c`$. We know from the calculation \[see Fig. 4\] that for the non-interacting Bose gas ($`\alpha =0`$), $`\mathrm{\Delta }=n_c`$, as it should be. However, for any finite $`\alpha `$ we find $`\mathrm{\Delta }n_c>0`$. Thus, for arbitrarily small values of the correlated-hopping, the system develops a tendency towards pair formation. However, it does not suppress the single particle BEC in favor of a purely pairing phase until sufficiently strong $`\alpha `$ is reached for sufficiently small $`n`$.
The results of our mean-field calculation are similar to those obtained on a different problem in the context of the atomic gases Romans et al. (2004). These are studies regarding the transition from a purely molecular condensate (MC) to an atomic condensate with a non-zero fraction of the molecular condensate present (AC+MC) accross the Feshbach resonance. Our pairing phase is like the MC, and the single particle BEC phase is analogous to the AC+MC.
The nature of the transition based on symmetry considerations is also similar. The mean-field Hamiltonian, $`H_{MF}`$, explicitly breaks the U(1) gauge symmetry, however it is still invariant under global $`Z_2`$ symmetry, that is under $`b_𝐫b_𝐫`$. In other words gauge symmetry, $`b_𝐫^{}b_𝐫^{}e^{i\varphi }`$ leaves $`H_{MF}`$ invariant for $`\varphi =\pi `$. This residual Ising like symmetry will also be broken if there is single particle BEC (because $`b0`$). What we have in Fig. 2 is such an Ising symmetry breaking quantum phase transition, where $`n_c`$ is the relevant order parameter. The pairing phase respects this $`Z_2`$ symmetry while the single particle BEC phase breaks it spontaneously.
The temperature dependence of various quantities in the mean-field theory is shown in Fig. 5. The temperature at which $`\mathrm{\Delta }`$ becomes zero is called $`T_c`$. This quantifies the mean-field phase transition from a normal Bose gas at high temperatures to a pairing phase below $`T_c`$. The inset of Fig. 5 shows $`T_c`$ as a function of $`n`$. Remarkably, there is no detectable anomaly upon going through the critical density $`n^{}`$. Since we are in 2D, we do not expect to have a true BEC of pairs, but rather a Kosterlitz-Thouless (KT) transition. These results suggest that the system should undergo one KT transition whatever the density, followed by an Ising transition if $`n>n^{}`$.
To check the validity of the mean-field approximation, hence of our conclusions, it would be very useful to have unbiased numerical results on the model of Eq. (1). However, we have good reasons to believe that the predictions of the present mean-field theory are physically relevant. Mathematically, the structure of the mean-field equations and the results of the calculations are similar to the Schwinger boson mean-field theory of the quantum spin system Sarkar et al. (1989); Mila et al. (1991). In that context, the single particle BEC phase implies an ordered phase in the spin variables, while the pairing phase denotes a disordered phase. Now the physical relevance of these disordered phases is well established in the context of quantum magnets Chakravarty et al. (1989); Sachdev (1999), and we expect the same to be true here.
The model studied in the present paper has similarities with the ring exchange model of bosonic Cooper pairs introduced some time ago by Paramekanti and collaborators Paramekanti et al. (2002). In their model, the pairs of bosons hop on the opposite corners of a plaquette. The low temperature physics is significantly different however. In their model, the number of bosons is conserved on each row and column of the square lattice, leading to the Luttinger liquid like physics and critical correlations in the ground state. In our model, the correlated-hopping does not sustain any such conservation law, and the ground state is expected to develop a true long range order.
Finally, let us briefly discuss the physical implications of these results for the magnetization process of gapped quantum magnets. The thermodynamics was already discussed in the boson language: we expect to observe a KT transition for any magnetization, followed by an Ising transition if the magnetization is larger than a critical value. This will remain essentially true for 3D systems, the KT transition being replaced by a true phase transition toward an ordered phase. However, we also expect very significant differences between the zero temperature phases. Single particle BEC means magnetic long-range order, and the system is expected to have gapless transverse spin waves. In other words, the gap detected in spectroscopies such as inelastic neutron scattering or NMR will vanish. However, at low magnetization, we only have pair BEC. The order implied by this pair BEC will be of nematic type since the transverse components of the spins within a pair can be flipped without changing the correlations. But more importantly, there is a gap to single particle excitations, i.e. to single spin flips. Although the system is gapless in this phase, we thus expect to observe a gap in neutron scattering or NMR, the gapless excitations appearing only in the channel $`\mathrm{\Delta }S=2`$.
In summary, we have shown that the correlated-hopping can change drastically the properties of bosons, leading at low densities to a pairing phase without single-particle BEC, and with gapped quasi-particle excitations. In the context of quantum frustrated magnets, this leads to the prediction of an Ising phase transition (for low magnetization) as a function of the magnetic field, for systems where frustration reduces direct hopping of triplets, thus making correlated-hopping the dominant process of kinetic energy.
Beyond frustrated magnets, these results will have implications on all systems where correlated-hopping may be the dominant source of kinetic energy. One such class of systems are the atomic gases, where different external parameters control the hopping and the Coulomb terms. Whether instabilities of the kind described here can be induced in these systems by reducing the single-particle kinetic energy is left for the future investigation.
###### Acknowledgements.
We thank A. Georges, S. Miyahara, D. Poilblanc and M. Troyer for stimulating discussions on different aspects of this work. We also acknowledge the Swiss National Funds and the MaNEP for the financial support. |
warning/0507/physics0507011.html | ar5iv | text | # Two Models for Bolometer and Microcalorimeter Detectors with Complex Thermal Architectures
## 1 Introduction
The operational principle of simple microcalorimeters and bolometers is based on three components. An absorber where the incoming power or energy is dissipated and converted into a change in temperature, a sensor that reads the change in temperature, and a thermal link from the detector to the heat sink that brings the system back to equilibrium after a measurement. The sensor is usually a resistor whose resistance depends strongly on temperature around the working point. In this case a change in resistance can be measured as a change in voltage or a change in current using a current or voltage bias.
In 1982 J.C Mather presented a complete non-equilibrium theory for the noise in simple bolometers with ideal resistive thermometers and in 1984 it was extended to microcalorimeter performance . At temperatures below 200 mK the ideal assumptions are no longer valid and complex thermal architectures are needed to understand the behavior of these devices. At these low temperatures the thermal fluctuations between the thermometer lattice and its electron system, or between the thermometer and the absorber are no longer negligible and therefore these components must be considered as separate entities in the model. These non-ideal effects are called electron decoupling and absorber decoupling. Another consequence of working at low temperatures is the increased dependence of the thermometer resistance on the readout power, making the ideal resistance-temperature relationship inaccurate. Galeazzi and McCammon constructed a general procedure for developing bolometer and microcalorimeter models for these complex thermal architectures using block diagram formalism of control theory.
To quantify the relation between incoming power or energy and the measured change in voltage or current, including non-ideal effects, this paper follows the modeling procedure of Galeazzi and McCammon . The first step in this modeling procedure is to set up the temperature equations, then apply a Taylor expansion to derive a linear model for small temperature deviations from equilibrium. Afterward, Fourier transforms are used to express the equations in the frequency domain, and finally the coupled equations are solved using block diagram algebra. This procedure yields closed form solutions for the responsivity and dynamic impedance of the model, including noise contributions.
The two models developed in this paper are modifications of the absorber decoupling model obtained in . Model 1 was developed to describe the new generation of transition edge sensor detectors where the absorber is not electrically isolated from the thermometer by a gluing agent but rather the two are deposited one on top of the other . Model 2 describes microcalorimeters that have the heat sink connected to the absorber instead of to the thermometer. This may occur when the absorber is much bigger than the thermometer and therefore it is necessary to connect the heat sink to the absorber rather than to the thermometer. These two models will help optimize the next generation of detectors, and because of the analytical results of the modeling procedure the relations between the detector’s resolution and the different parameters included in the model should be clear.
## 2 Model 1
This model is suitable to describe TES detectors with an integrated absorber. In this model both the electron system and the absorber are detached from the lattice. The lattice is connected to the heat sink by a thermal conductivity $`G`$, the electron system is connected to the lattice by a thermal conductivity $`G_{el}`$, and the absorber is connected to the electron system by a thermal conductivity $`G_a`$ ( see Fig. 1). The absorber is directly connected to the electron system instead of the lattice because with integrated absorbers the absorber-lattice thermal coupling is expected to be negligible compared to that of the absorber-electron system.
### 2.1 Responsivity S($`\omega `$)
The following equations determine the temperature for each of the three components in the model:
$$C_a\frac{d(T_a^{})}{dt}+_{T_e^{}}^{T_a^{}}G_a(T^{})𝑑T^{}=W$$
(1)
$$C_e\frac{d(T_e^{})}{dt}+_{T_a^{}}^{T_e^{}}G_a(T^{})𝑑T^{}+_{T_l^{}}^{T_e^{}}G_{el}(T^{})𝑑T^{}=P(T_e^{})$$
(2)
$$C_l\frac{d(T_l^{})}{dt}+_{T_e^{}}^{T_l^{}}G_{el}(T^{})𝑑T^{}+_{T_s}^{T_l^{}}G(T^{})𝑑T^{}=0,$$
(3)
where $`C_a`$, $`C_e`$, and $`C_l`$ are the heat capacities of the absorber, the electron system, and the lattice system respectively, and $`T_a^{}`$, $`T_e^{}`$, and $`T_l^{}`$ are the corresponding temperatures. $`W`$ is the incoming outside power to be measured and $`P(T_e^{})`$ is the Joule power dissipated into the sensor by the bias current/voltage. In the case of microcalorimeters $`W=E\delta (t_o)`$, where $`E`$ is the photon energy and $`\delta (t_o)`$ is the delta function.
The equilibrium conditions of the system are obtained by setting the outside power to zero ($`W=0`$), and $`d(T_x^{})/dt=0`$ ($`x`$= $`a`$, $`e`$, or $`l`$) since the equilibrium temperatures are independent of time. Therefore the equilibrium temperatures $`T_a`$ of the absorber, $`T_e`$ of the electron system, and $`T_l`$ of the lattice are given by the integrals in the previous three equations. For example the integral in Eq. 1 must equal zero at equilibrium, which implies that the thermal equilibrium temperature of the absorber is the same as that of the electron system. We are interested in small deviations about the equilibrium temperatures, therefore we set $`T_x^{}=T_x+\mathrm{\Delta }T_x`$, where $`T_x`$ is the equilibrium temperature for each component of the model, and $`\mathrm{\Delta }T_x`$ is the small temperature deviation from equilibrium:
$$C_a\frac{d(T_a+\mathrm{\Delta }T_a)}{dt}+_{T_e+\mathrm{\Delta }T_e}^{T_a+\mathrm{\Delta }T_a}G_a(T^{})𝑑T^{}=W$$
(4)
$$C_e\frac{d(T_e+\mathrm{\Delta }T_e)}{dt}+_{T_a+\mathrm{\Delta }T_a}^{T_e+\mathrm{\Delta }T_e}G_a(T^{})𝑑T^{}+_{T_l+\mathrm{\Delta }T_l}^{T_e+\mathrm{\Delta }T_e}G_{el}(T^{})𝑑T^{}=P(T_e+\mathrm{\Delta }T_e)$$
(5)
$$C_l\frac{d(T_l+\mathrm{\Delta }T_l)}{dt}+_{T_e+\mathrm{\Delta }T_e}^{T_l+\mathrm{\Delta }T_l}G_{el}(T^{})𝑑T^{}+_{T_s}^{T_l+\mathrm{\Delta }T_l}G(T^{})𝑑T^{}=0.$$
(6)
In the small signal limit $`\mathrm{\Delta }T_x`$ is small compared to $`T_x`$, and a Taylor expansion up to the first $`\mathrm{\Delta }T_x`$ term is appropriate. The results are the equations that determine small temperature deviations about equilibrium:
$$C_a\frac{d(\mathrm{\Delta }T_a)}{dt}+G_a\mathrm{\Delta }T_a=W+G_a\mathrm{\Delta }T_e$$
(7)
$$C_e\frac{d(\mathrm{\Delta }T_e)}{dt}+G_a\mathrm{\Delta }T_e+G_{el}(T_e)\mathrm{\Delta }T_e=\mathrm{\Delta }P+G_a\mathrm{\Delta }T_a+G_{el}(T_l)\mathrm{\Delta }T_l$$
(8)
$$C_l\frac{d(\mathrm{\Delta }T_l)}{dt}+G_{el}(T_l)\mathrm{\Delta }T_l+G\mathrm{\Delta }T_l=G_{el}(T_e)\mathrm{\Delta }T_e,$$
(9)
where $`\mathrm{\Delta }P=P(T_e+\mathrm{\Delta }T_e)P(T_e)`$, and for simplicity we used $`G_a=G_a(T_a)`$ and $`G=G(T_l)`$.
These are coupled differential equations which are difficult to solve directly; instead they are transformed into coupled algebraic equations using Fourier transforms. The quantity $`\mathrm{\Delta }P`$ represents what is known as the electro-thermal feedback term and can be written as $`\mathrm{\Delta }P=G_{ETF}\mathrm{\Delta }T_e`$, where $`G_{ETF}=P(RR_L)\alpha /T_eR(R_L+R)`$; (see reference ). Converting Eqs. 7, 8, and 9 into the frequency domain using Fourier transforms we obtain:
$$j\omega C_a\mathrm{\Delta }T_a+G_a\mathrm{\Delta }T_a=W+G_a\mathrm{\Delta }T_e$$
(10)
$$j\omega C_e\mathrm{\Delta }T_e+(G_a+G_{el}(T_e)+G_{ETF})\mathrm{\Delta }T_e=G_a\mathrm{\Delta }T_a+G_{el}(T_l)\mathrm{\Delta }T_l$$
(11)
$$j\omega C_l\mathrm{\Delta }T_l+(G_{el}(T_l)+G)\mathrm{\Delta }T_l=G_{el}(T_e)\mathrm{\Delta }T_e.$$
(12)
With these equations it is possible to solve for $`\mathrm{\Delta }T_e`$, which can be related to the measured quantities $`\mathrm{\Delta }I`$ or $`\mathrm{\Delta }V`$, using the typical detector readout circuit of Fig. 2:
$$\mathrm{\Delta }V=V\frac{\alpha }{T_e}\frac{R_L}{R_L+R}\mathrm{\Delta }T_e$$
(13)
$$\mathrm{\Delta }I=I\frac{\alpha }{T_e}\frac{R}{R_L+R}\mathrm{\Delta }T_e,$$
(14)
where $`\alpha =T_e/R\times dR/dT_e`$ is the sensitivity of the detector, $`R_L`$ is the load resistance, $`R`$ is the resistance of the detector, $`V`$ is the voltage across $`R`$, and $`I`$ is the current flowing through $`R`$.
To simplify the notation let $`X`$ be either $`V`$ or $`I`$, and introduce the quantity $`A_{tr}=R/X\times dX/dR`$ to be deduced from the previous two equations. Then Eqs. 13 and 14 can be summarized as:
$$\frac{\mathrm{\Delta }X}{X}=\alpha A_{tr}\frac{\mathrm{\Delta }T_e}{T_e}.$$
(15)
Equations 10, 11, 12, and 15 can be solved using the block diagram of Fig. 3. To set up the block diagram consider the left hand side of Eqs.10, 11, 12, and 15 as the response function of the absorber system, electron system, lattice system, and circuit readout respectively. The right hand side of these equations corresponds to the input to each system. Connecting the response functions with their appropriate inputs leads to the block diagram of Fig. 3.
To solve the block diagram in Fig. 3 for $`\mathrm{\Delta }X(\omega )`$ we used the procedure and simplification rules of the block diagram formalism described in . This result is then used to find the responsivity, which is defined as $`S(\omega )=\mathrm{\Delta }X(\omega )/W(\omega )`$. The following responsivity characterizes the response of Model 1 detectors:
$$S(w)=\frac{G_a}{(G_a+j\omega C_a)\left[(G_a+G_{el}(T_e)+G_{ETF}+j\omega C_e)\frac{G_{el}(T_l)G_{el}(T_e)}{j\omega C+G_{el}(T_l)+G}\right]G_a^2}\frac{X\alpha A_{tr}}{T_e}.$$
(16)
### 2.2 Dynamic Impedance
A detector can also be described by its complex dynamic impedance $`Z(\omega )=dV(\omega )/dI(\omega )`$. The dynamic impedance differs from the detector resistance due to the effect of the electro-thermal feedback. When the current changes, the power dissipated into the detector changes too, therefore the temperature and the detector’s resistance change. The dynamic impedance is a useful parameter because it is easily measured experimentally. To find the dynamic impedance we use $`G_{ETF}\mathrm{\Delta }T_e=\mathrm{\Delta }P`$ in Eqs. 11, and use Eqs. 10, 11, and 12 to find $`\mathrm{\Delta }T_e`$ in terms of $`\mathrm{\Delta }P`$, $`\omega `$, the heat capacity of each of the three components, and the three thermal conductivities:
$$dT_e=\frac{dP}{\left(j\omega C_e+G_a+G_{el}(T_e)\frac{G_a^2}{G_a+j\omega C_a}\frac{G_{el}(T_l)G_{el}(T_e)}{j\omega C+G_{el}(T_l)+G}\right)}.$$
(17)
Differentiating Ohm’s law ($`V=IR`$) and using the definition of sensitivity $`\alpha `$ we obtain:
$$dV=RdI+I\frac{\alpha RdT}{T}.$$
(18)
Substituting Eq. 17 into Eq 18 and using the fact that $`dP=VdI+IdV`$, it is possible to solve for $`dV/dI`$ and obtain the following result for the dynamic impedance:
$$Z(\omega )=\frac{dV}{dI}=R\frac{\left[\alpha P+T\left(j\omega C_e+G_a+G_{el}(T_e)\frac{G_a^2}{G_a+j\omega C_a}\frac{G_{el}(T_l)G_{el}(T_e)}{j\omega C+G_{el}(T_l)+G}\right)\right]}{\left[\alpha P+T\left(j\omega C_e+G_a+G_{el}(T_e)\frac{G_a^2}{G_a+j\omega C_a}\frac{G_{el}(T_l)G_{el}(T_e)}{j\omega C+G_{el}(T_l)+G}\right)\right]}.$$
(19)
### 2.3 Noise
The effect of noise on a detector’s performance is quantified by the Noise Equivalent Power (NEP). It corresponds to the power that would be required as input to the detector in order to generate an output equal to the signal generated by the noise. The NEP can therefore be calculated as the ratio between the output generated by the noise and the responsivity of the detector:
$$NEP_y=\frac{\mathrm{\Delta }X_y}{S(\omega )}.$$
(20)
The variable $`y`$ stands in for any of the possible noise terms: $`amp`$=amplifier noise, $`j`$=Johnson noise, $`R_L`$=load resistor noise, $`1/f`$=$`1/f`$ noise, $`a`$=absorber-electron system thermal noise, $`th`$=heat sink-lattice thermal noise or $`he`$=electron system-lattice thermal noise.
To obtain the Noise Equivalent Power for each term, the noise contributions $`e_{amp}`$, $`e_j`$, $`e_{R_L}`$, $`P_{R_L}`$, $`(\mathrm{\Delta }R/R)_{1/f}`$, $`P_a`$, $`P_{th}`$, and $`P_{he}`$ must be added to the block diagram of Fig. 3. Figure 4 shows where each noise term should be added to the block diagram (for more details see ). Solving the noise block diagram for each noise term independently and dividing by the responsivity obtained in Eq. 16 we obtain the following NEP’s:
$$NEP_a=P_a(\omega )j\omega \tau _a$$
(21)
$$NEP_{R_L}=P_{R_L}(\omega )(1+j\omega \tau _a)+\frac{e_{R_L}}{S(\omega )}$$
(22)
$$NEP_{amp}=\frac{e_{amp}}{S(\omega )}$$
(23)
$$NEP_{he}=P_{he}\frac{(1+j\omega \tau _a)(G+j\omega C_l)}{(G+G_{el}(T_l)+j\omega C_l)}$$
(24)
$$NEP_{th}=P_{th}\frac{G_{el}(T_l)(1+j\omega \tau _a)}{(G+G_{el}(T_l)+j\omega C_l)}$$
(25)
$`NEP_{e_j}=e_j(\omega ){\displaystyle \frac{T_e}{IR\alpha }}(1+j\omega \tau _a)(G_a+G_{el}(T_e)+j\omega C_e`$
$`{\displaystyle \frac{G_{el}(T_l)G_{el}(T_e)}{j\omega C+G_{el}(T_l)+G}}{\displaystyle \frac{G_a^2}{G_a+j\omega C_a}})`$ (26)
$`NEP_{1/f}=\left({\displaystyle \frac{\mathrm{\Delta }R(\omega )}{R}}\right)_{1/f}{\displaystyle \frac{T_e}{\alpha }}(1+j\omega \tau _a)(G_a+G_{el}(T_e)+j\omega C_e`$
$`{\displaystyle \frac{G_{el}(T_l)G_{el}(T_e)}{j\omega C+G_{el}(T_l)+G}}{\displaystyle \frac{G_a^2}{G_a+j\omega C_a}}).`$ (27)
Where $`\tau _a=C_a/G_a`$.
## 3 Model 2
In experiments involving dark matter detectors and double-beta decay detectors the absorber size is significant and can have a mass up to almost 1 Kg . For mechanical reasons these large absorbers must be mechanically connected to the heat sink. The thermal link between the detector and the heat sink can therefore also be through the absorber rather than through the sensor. Model 2 reflects this condition by having the absorber connected to the heat sink through a thermal conductivity $`G`$. Model 2 also takes into account absorber decoupling and electron decoupling by connecting the lattice system to the absorber through a thermal conductivity $`G_a`$ and by having the electron system connected to the lattice system through a thermal conductivity $`G_{el}`$ (see Fig. 5).
Applying the same procedure previously used for Model 1 we obtain for Model 2 the block diagram of Fig. 6.Solving the block diagram for the responsivity, the dynamic impedance, and all the noise contributions, we obtain the following results:
$$S\left(\omega \right)={\scriptscriptstyle \frac{G_aG_{el}(T_l)}{(G_{el}(T_e)+G_{ETF}+j\omega C_e)\left[(G_a+G_{el}(T_l)+j\omega C_l)(G_a+G+j\omega C_a)G_a^2\right]G_{el}(T_l)G_{el}(T_e)(G_a+G+j\omega C_a)}}{\scriptscriptstyle \frac{X\alpha A_{tr}}{T_e}}$$
(28)
$$Z(\omega )=R\frac{\alpha P+T_e\left[j\omega C_e+G_{el}(T_e)G_{el}(T_l)\left(\frac{G_{el}(T_e)(j\omega C_a+G_a+G)}{(j\omega C_a+G_a+G)(j\omega C_l+G_a+G_{el}(T_l))G_a^2}\right)\right]}{\alpha P+T_e\left[j\omega C_e+G_{el}(T_e)G_{el}(T_l)\left(\frac{G_{el}(T_e)(j\omega C_a+G_a+G)}{(j\omega C_a+G_a+G)(j\omega C_l+G_a+G_{el}(T_l))G_a^2}\right)\right]}$$
(29)
$$NEP_{th}=P_{th}$$
(30)
$$NEP_{amp}=\frac{e_{amp}}{S(\omega )}$$
(31)
$$NEP_a=P_a\left[\frac{G+j\omega C_a}{G_a}\right]$$
(32)
$$NEP_{he}=P_{he}\left[\frac{(G_a+G+j\omega C_a)(G_a+G_{el}(T_l)+j\omega C_l)G_a^2}{G_aG_{el}(T_l)}\frac{G+G_a+j\omega C_a}{G_a}\right]$$
(33)
$$NEP_{R_L}=P_{R_L}\left[\frac{(G_a+G+j\omega C_a)(G_a+G_{el}(T_l)+j\omega C_l)G_a^2}{G_aG_{el}(T_l)}\right]+\frac{e_{RL}}{S(\omega )}$$
(34)
$`NEP_{e_j}=e_j{\displaystyle \frac{T_e}{IR\alpha }}{\displaystyle \frac{(G_a+G+j\omega C_a)[(G_a+G_{el}(T_l)+j\omega C_l)\frac{G_{el}(T_l)G_{el}(T_e)}{j\omega C_e+G_{el}(T_e)}]G_a^2}{G_aG_{el}(T_l)}}`$
$`\times (j\omega C_e+G_{el})`$ (35)
$`NEP_{\frac{1}{f}}=\left({\displaystyle \frac{\mathrm{\Delta }R(\omega )}{R}}\right)_{\frac{1}{f}}{\displaystyle \frac{(G_a+G+j\omega C_a)[(G_a+G_{el}(T_l)+j\omega C_l)\frac{G_{el}(T_l)G_{el}(T_e)}{j\omega C_e+G_{el}(T_e)}]G_a^2}{G_aG_{el}(T_l)}}`$
$`\times {\displaystyle \frac{T_e}{\alpha }}(j\omega C_e+G_{el}).`$ (36)
## 4 Examples of Energy Resolution and Time constant Results from the Models
In experimental setups of detectors some of the parameters are fixed while others can vary. With the freedom to vary a few parameters the goal is to optimize the detectors characteristics. The equations derived in this paper offer the flexibility and power to perform such operations. An example of such applications is reported in fig. 7. The characteristics of interest are energy resolution and the time constant of the detector. The energy resolution is calculated following equation 69 in reference . The time constant is calculated as the inverse of the first turn frequency of the responsivity. The variable parameters in the examples of fig. 7 are: heat capacity of the absorber ($`C_a`$) vs. thermal conductivity between absorber and thermometer ($`G_a`$) and thermal conductivity between detector and heat sink ($`G`$) vs. thermal conductivity between absorber and thermometer ($`G_a`$). The fixed parameters are reported in Table 1. By inputing the fixed parameter into the two models of this paper and the model found in and letting $`C_a`$, $`G`$ , and $`G_a`$ vary we obtain the twelve contour plots in fig. 7. The first six plots predict energy resolution while the last six plots refer to the detectors time constant. Each column of contour plots belongs to one of the three different models.
## 5 Conclusions
To improve the performance of microcalorimeters and bolometers it is important to accurately understand how this depends on the fabrications parameters. Significant improvements in detectors performance have, in fact, been achieved by optimizing the design based on an accurate model of the detector . In this paper we derived detailed theoretical models to describe the behavior of two different detector architectures. The use of block diagram algebra has allowed us to present the results in an analytical form that can be easily and immediately utilized by investigators to improve the design of their detectors. Furthermore, the contour plots in fig. 7 provide an example of how the equations derived in this paper can be utilized to predict the characteristics of detectors.
## Tables
Table 1: Values of the fixed parameters needed to produce the contour plots of fig. 7
## Figures
FIG. 1: Thermal architecture of Model 1.
FIG. 2: Typical readout circuit. Notice that if $`R_L<<R`$ the detector is voltage biased, if $`R_L>>R`$ the detector is current biased.
FIG. 3: Block diagram representing Model 1.
FIG. 4: Block diagram including noise contributions for Model 1.
FIG. 5: Thermal architecture of Model 2.
FIG. 6: Block diagram including noise contributions for Model 2.
FIG. 7: Contour plots of how the energy resolution and time constant of each model change with respect to the heat capacity $`C_a`$ and the thermal conductivities $`G`$ and $`G_a`$. These plots were constructed using the fixed parameters in Table 1.
## Table 1: Values of the fixed parameters needed to produce the contour plots in fig 7
| Parameter | Value |
| --- | --- |
| $`R`$ | $`5m\mathrm{\Omega }`$ |
| $`R_L`$ | $`0.2m\mathrm{\Omega }`$ |
| $`V_{bias}`$ | $`1.0848\times 10^7V`$ |
| $`\alpha `$ | $`100`$ |
| $`C_a`$ | $`1pJ/K`$ |
| $`C_l`$ | $`4.911\times 10^5pJ/K`$ |
| $`C_e`$ | $`0.154pJ/K`$ |
| $`T_s`$ | $`0.1K`$ |
| $`G_{el}(T_l)`$ | $`5\times 10^{10}W/K`$ |
| $`G_{el}(T_e)`$ | $`5.87\times 10^{10}W/K`$ |
| $`G`$ | $`1\times 10^{10}W/K`$ | |
warning/0507/math-ph0507046.html | ar5iv | text | # Towards analytical solutions of the alloy solidification problem.
## 1 Introduction.
A general methodolgy of achieving analytical solutions of the alloy solidification problem is presented in this manuscript. There is an analytical solution for pure substance (Stefan’s problem) and few analytical and semi-analytical solutions for alloys . We suggest a general methodology which can provide wide range solutions to test different numerical schemes .
We consider the case when physical properties $`\mathrm{\Phi }`$ (density, heat capacity, or heat conductivity) in solid and in liquid are constant. Within mushy zone these properties and the enthalpy depend on temperature as follows (i.e. obey the lever rule):
$$\mathrm{\Phi }(T)=\left[1\lambda (T)\right]\mathrm{\Phi }_s+\lambda (T)\mathrm{\Phi }_l,$$
(1)
where $`\mathrm{\Phi }_s`$ and $`\mathrm{\Phi }_l`$ are properties in solid and liquid, respectively, $`\lambda (T)`$ is volumetric liquid fraction. Then we rewrite heat transfer equation in the full enthalpy term $`H(T)`$. The key idea of the present work is the mushy heat diffusivity requirement to be constant.
$$\alpha (T)=\frac{\kappa (T)}{\frac{dH(T)}{dT}}=\alpha _{sl}=const,$$
(2)
where $`\kappa (T)`$ is heat conductivity. From this condition we can find liquid fraction $`\lambda (T)`$ by means of which we are able to linearize an initial energy conservation equation. Thus, the following methodology is
1. To rewrite of the heat equation in the full enthalpy term.
2. To require of the thermal diffusivity to be constant in the solid, mushy and liquid zone.
3. Condition $`\alpha (T)=\alpha _{sl}=const`$ is ordinary differential equation for liquid fraction $`\lambda =\lambda (T)`$. Additionaly we require $`\lambda (T_l)=1`$.
4. To solve this ODE and find $`\lambda =\lambda (T,\alpha _{sl})`$.
5. To impose additional condition $`\lambda (T_s,\alpha _{sl})=\lambda _0`$. For $`\lambda _0=0`$ we have noneutectic alloy, and for $`\lambda _00`$ – eutectic. From this condition we find $`\alpha _{sl}`$.
6. Now we have the heat equation with constant-peace coefficients and we can it solve easy.
It needs to note, that this problem cannot be solved with well defined (predefined) $`\lambda (T)`$ function, instead of the function $`\lambda =\lambda (T)`$ is determined from linearisation conditions.
## 2 The linearisation of the heat equation.
We will solve an energy conservation equation
$`{\displaystyle \frac{H}{t}}=div\left(k(T)gradT\right),`$ (3)
where full enthalpy $`H`$ is
$$H(T)=\rho (T)\left[1\lambda (T)\right]\underset{0}{\overset{T}{}}C_s(\zeta )𝑑\zeta +\rho (T)\lambda (T)\underset{0}{\overset{T}{}}C_l(\zeta )𝑑\zeta +\rho (T)\lambda (T)L,$$
(4)
where $`C_s`$, $`C_l`$ is specific heat in solid and in liquid, $`L`$ is latent heat of fusion, $`\rho =\rho _s=\rho _l`$ is density, which all are constants. We express the heat conductivity in the ”mixture” form
$$\kappa (T)=\left[1\lambda (T)\right]\kappa _s+\lambda (T)\kappa _l,$$
(5)
where $`\kappa _s`$ and $`\kappa _l`$ are constant heat conductivity in solid and liquid, respectively.
Taking into account the expression
$$gradT=\frac{gradH}{\frac{dH}{dT}}$$
(6)
the Eq. (3) can be rewritten in the general form
$$\frac{H}{t}=div\left(\alpha (T)gradH\right),$$
(7)
where $`\alpha (H)`$ is thermal diffusivity, which defined as
$$\alpha (T)=\frac{\kappa (T)}{\frac{dH(T)}{dT}},$$
(8)
If $`\kappa (T)`$ and $`dH(T)/dT`$ depend on temperature arbitrary manner then Equation (7) is nonlinear. To achieve an analitical solution we need to require the thermal diffusivity to be constant in all regions (solid, mushy and liquid).
$$\alpha (T)=\{\begin{array}{cc}\alpha _s=const\hfill & \text{for }T<T_s\hfill \\ \alpha _{sl}=const\hfill & \text{for }T_sTT_l\hfill \\ \alpha _l=const\hfill & \text{for }T>T_l\hfill \end{array}$$
(9)
In our case $`\alpha _s=\kappa _s\rho C_s`$ and $`\alpha _l=\kappa _l/\rho C_l`$ are constant by definition. For the derivation of mushy enthalpy (the apparent capacity $`\times `$ density) we get:
$$\frac{1}{\rho }\frac{dH(T)}{dT}=\left[1\lambda \right]C_s+\lambda C_l+\left[(C_lC_s)T+L\right]\frac{d\lambda (T)}{dT}.$$
(10)
then from the mushy part of Eq. (9) we obtain an ordinal differential equation for $`\lambda (T)`$
$$\left[1+pT\right]\frac{d\lambda (T)}{dT}+a\lambda (T)+b=0,$$
(11)
where we denote
$`a={\displaystyle \frac{a_{sl}\rho (C_lC_s)(\kappa _l\kappa _s)}{a_{sl}\rho L}},`$ (12)
$`b={\displaystyle \frac{a_{sl}\rho C_s\kappa _s}{a_{sl}\rho L}},`$ (13)
$`p={\displaystyle \frac{C_lC_s}{L}}.`$ (14)
We require
$$\lambda (T_l)=1,$$
(15)
where $`T_l`$ is a liquidus temperature. Solution of Eqs. (11) and (15) is
$$\lambda (T)=\frac{b}{a}+\frac{a+b}{a}\left(\frac{1+pT_l}{1+pT}\right)^{\frac{a}{p}}.$$
(16)
It needs to determine an additional condition for $`\lambda (T)`$ function, namely to define the liquid fraction value at solidus temperature
$$\lambda (T_s)=\{\begin{array}{cc}0\hfill & \text{for noneutectic alloy}\hfill \\ \lambda _0\hfill & \text{for eutectic alloy}\hfill \end{array}$$
(17)
To obtain the analytical solution of Eq. (7) we need to solve Eq.(17) to find root $`a_{sl}`$. Then we need to solve Eq. (7) with suitable initial and boundary conditions.
The enthalpy of the system (4) we may design
$$\frac{H(T)}{\rho }=\{\begin{array}{cc}C_sT\hfill & \text{for }T<T_s\hfill \\ C_sT+\lambda (T)\left[(C_lC_s)T+L\right]\hfill & \text{for }T_sTT_l\hfill \\ C_lT+L\hfill & \text{for }T>T_l\hfill \end{array}.$$
(18)
In future we need the value $`dT/dH`$:
$$\rho \frac{dT}{dH}=\{\begin{array}{cc}\frac{1}{C_s}\hfill & \text{for }T<T_s\hfill \\ \frac{1}{\left[1\lambda \right]C_s+\lambda C_l+\left[(C_lC_sT+L)\right]\frac{d\lambda }{dT}}\hfill & \text{for }T_sTT_l\hfill \\ \frac{1}{C_l}\hfill & \text{for }T>T_l\hfill \end{array}.$$
(19)
It should be note that some expressions with $`d\lambda /dT`$ may be written in more simplified form versus $`\lambda `$, for example:
$$\frac{d\lambda }{dT}=\frac{a\lambda +b}{1+pT}=\frac{a\lambda +b}{(C_lC_s)T+L}L,$$
(20)
and the combination
$$\left[(C_lC_s)T+L\right]\frac{d\lambda }{dT}=\left(a\lambda +b\right)L.$$
(21)
## 3 An analytical solution for enthalpy.
We will examine the simple problem
$`{\displaystyle \frac{H}{t}}={\displaystyle \frac{}{x}}\left(\alpha (H){\displaystyle \frac{H}{x}}\right),`$ (22)
$`H(t=0)=H(T_{init}),`$ (23)
$`H|_{x=0}=H_{out}=H(T_{out}),`$ (24)
$`H|_{x=\mathrm{}}=H_{init}=H(T_{init}).`$ (25)
The solution of these equations with constant-piece function $`\alpha (H)`$ can be easy find . To solve this equation we divide whole region $`[0,\mathrm{})`$ into three subintervals $`[0,X_s)`$, $`[X_s,X_l]`$ and $`(X_l,\mathrm{})`$ ($`X_s`$ and $`X_l`$ are solidus and liquidus positions, respectively). Moreover, we assume that (the similarity solution):
$$X_s(t)=k_s\sqrt{t},X_l(t)=k_l\sqrt{t},$$
(26)
where $`k_s`$ and $`k_l`$ are constants. Solutions on the subintervals are:
$$H(x,t)=H_{out}+(H_sH_{out})\frac{erf\left(\frac{x}{2\sqrt{\alpha _st}}\right)}{erf\left(\frac{k_s}{2\sqrt{\alpha _s}}\right)},x[0,X_s),$$
(27)
$$H(x,t)=\frac{(H_lH_s)erf\left(\frac{x}{2\sqrt{\alpha _{sl}t}}\right)+H_serf\left(\frac{k_l}{2\sqrt{\alpha _{sl}}}\right)H_lerf\left(\frac{k_s}{2\sqrt{\alpha _{sl}}}\right)}{erf\left(\frac{k_l}{2\sqrt{\alpha _{sl}}}\right)erf\left(\frac{k_s}{2\sqrt{\alpha _{sl}}}\right)},x[X_s,X_l],$$
(28)
$$H(x,t)=H_{init}(H_{init}H_l)\frac{erfc\left(\frac{x}{2\sqrt{\alpha _lt}}\right)}{erfc\left(\frac{k_l}{2\sqrt{\alpha _l}}\right)},x(X_l,\mathrm{}),$$
(29)
where we defined
$`H_s=H(T_s)=\rho C_sT_s,`$ (30)
$`H_l=H(T_l)=\rho (C_lT_l+L).`$ (31)
By using the two conditions at the interfaces (the first one from which is Stefan’s condition at the solidus (eutectic) point):
$$\alpha _s\frac{H}{x}|_{x=X_s0}=\alpha _{sl}\frac{H}{x}|_{x=X_s+0}+\rho \lambda _0L\frac{dX_s(t)}{dt},$$
(32)
$$\alpha _{sl}\frac{H}{x}|_{x=X_l0}=\alpha _l\frac{H}{x}|_{x=X_l+0},$$
(33)
we derive the following two equations from which to evaluate $`k_s`$ and $`k_l`$:
$$\frac{\sqrt{\alpha _s}(H_sH_{out})exp\left(\frac{k_s^2}{4\alpha _s}\right)}{erf\left(\frac{k_s}{2\sqrt{\alpha _s}}\right)}\frac{\sqrt{\alpha _{sl}}(H_lH_s)exp\left(\frac{k_s^2}{4\alpha _{sl}}\right)}{erf\left(\frac{k_l}{2\sqrt{\alpha _{sl}}}\right)erf\left(\frac{k_s}{2\sqrt{\alpha _{sl}}}\right)}=\frac{\sqrt{\pi }}{2}\rho \lambda _0Lk_s,$$
(34)
$$\frac{\sqrt{\alpha _{sl}}(H_lH_s)exp\left(\frac{k_l^2}{4\alpha _{sl}}\right)}{erf\left(\frac{k_l}{2\sqrt{\alpha _{sl}}}\right)erf\left(\frac{k_s}{2\sqrt{\alpha _{sl}}}\right)}\frac{\sqrt{\alpha _l}(H_{init}H_l)exp\left(\frac{k_l^2}{4\alpha _l}\right)}{erfc\left(\frac{k_l}{2\sqrt{\alpha _l}}\right)}=0.$$
(35)
## 4 What we can get from the exact solution?
Usualy we have numerical scheme which gives us the temperature, but not enthalpy. Below we write down formulas for temperature evaluation versus enthalpy and some other parameters.
### 4.1 Solidus and liquidus velocities.
From Eq. (26) we get the front velocities
$$v_l(t)=\frac{dX_l(t)}{dt}=\frac{k_l}{2\sqrt{t}},v_s(t)=\frac{dX_s(t)}{dt}=\frac{k_s}{2\sqrt{t}}.$$
(36)
### 4.2 Temperature curves.
From Eq. (18) and Eqs. (27) - (29) we can easy to find:
In the solid ($`x<k_s\sqrt{t}`$):
$$T(x,t)=\frac{1}{\rho C_s}\left[H_{out}+(H_sH_{out})\frac{erf\left(\frac{x}{2\sqrt{\alpha _st}}\right)}{erf\left(\frac{k_s}{2\sqrt{\alpha _s}}\right)}\right].$$
(37)
In the mushy zone ($`k_s\sqrt{t}xk_l\sqrt{t}`$) we need to solve nonlinear equation to get $`T=T(x,t)`$:
$`C_sT+\lambda (T)\left[(C_lC_s)T+L\right]`$
$`={\displaystyle \frac{1}{\rho }}{\displaystyle \frac{(H_lH_s)erf\left(\frac{x}{2\sqrt{\alpha _{sl}t}}\right)+H_serf\left(\frac{k_l}{2\sqrt{\alpha _{sl}}}\right)H_lerf\left(\frac{k_s}{2\sqrt{\alpha _{sl}}}\right)}{erf\left(\frac{k_l}{2\sqrt{\alpha _{sl}}}\right)erf\left(\frac{k_s}{2\sqrt{\alpha _{sl}}}\right)}}.`$ (38)
In the liquid ($`x>k_l\sqrt{t}`$):
$$T(x,t)=\frac{1}{\rho C_l}\left[H_{init}(H_{init}H_l)\frac{erfc\left(\frac{x}{2\sqrt{\alpha _lt}}\right)}{erfc\left(\frac{k_l}{2\sqrt{\alpha _l}}\right)}\right]\frac{L}{C_l}.$$
(39)
### 4.3 Temperature gradients.
From equation
$$\frac{T}{x}=\frac{dT}{dH}\frac{H}{x}$$
end from Eq. (19) easy to achive the expressions for temperature gradients:
$$\frac{T}{x}=\{\begin{array}{cc}\frac{1}{\rho C_s}\frac{H}{x}\hfill & T<T_s\hfill \\ \frac{1}{\rho }\frac{1}{\left[1\lambda \right]C_s+\lambda C_l+\left[(C_lC_s)T+L\right]\frac{d\lambda }{dT}}\frac{H}{x}\hfill & T_sTT_l\hfill \\ \frac{1}{\rho C_l}\frac{H}{x}\hfill & T>T_l\hfill \end{array}$$
(40)
For enthalpy gradients from Eqs. (27) - (29) we get
$$\frac{H}{x}=\{\begin{array}{cc}\frac{H_sH_{out}}{erf\left(\frac{k_s}{2\sqrt{\alpha _s}}\right)}\frac{e^{\frac{x^2}{4\alpha _st}}}{\sqrt{\pi \alpha _st}}\hfill & H<H_s\hfill \\ \frac{H_lH_s}{erf\left(\frac{k_l}{2\sqrt{\alpha _{sl}}}\right)erf\left(\frac{k_s}{2\sqrt{\alpha _{sl}}}\right)}\frac{e^{\frac{x^2}{4\alpha _{sl}t}}}{\sqrt{\pi \alpha _{sl}t}}\hfill & H_sHH_l\hfill \\ \frac{H_{init}H_l}{erfc\left(\frac{k_l}{2\sqrt{\alpha _l}}\right)}\frac{e^{\frac{x^2}{4\alpha _lt}}}{\sqrt{\pi \alpha _lt}}\hfill & H>H_l\hfill \end{array}$$
(41)
Thus finally we have:
$$\frac{T}{x}=\{\begin{array}{cc}\frac{1}{\rho C_s}\frac{H_sH_{out}}{erf\left(\frac{k_s}{2\sqrt{\alpha _s}}\right)}\frac{e^{\frac{x^2}{4\alpha _st}}}{\sqrt{\pi \alpha _st}}\hfill & x<k_s\sqrt{t}\hfill \\ \frac{(H_lH_s)/\left(erf\left(\frac{k_l}{2\sqrt{\alpha _{sl}}}\right)erf\left(\frac{k_s}{2\sqrt{\alpha _{sl}}}\right)\right)}{\rho \left[(1\lambda )C_s+\lambda C_l+\left[(C_lC_s)T+L\right]\frac{d\lambda }{dT}\right]}\frac{e^{\frac{x^2}{4\alpha _{sl}t}}}{\sqrt{\pi \alpha _{sl}t}}\hfill & k_s\sqrt{t}xk_l\sqrt{t}\hfill \\ \frac{1}{\rho C_l}\frac{H_{init}H_l}{erfc\left(\frac{k_l}{2\sqrt{\alpha _l}}\right)}\frac{e^{\frac{x^2}{4\alpha _lt}}}{\sqrt{\pi \alpha _lt}}\hfill & x>k_l\sqrt{t}\hfill \end{array}$$
(42)
There is a very interesting parameters as temperature gradient in liquid phase at the liquidus point $`G_l`$. For it we can write down
$$G_l(t)=\frac{T(x,t)}{t}|_{x=X_l+0}=\frac{1}{\rho C_l}\frac{H_{init}H_l}{erfc\left(\frac{k_l}{2\sqrt{\alpha _l}}\right)}\frac{e^{\frac{k_l^2}{4\alpha _l}}}{\sqrt{\pi \alpha _lt}}.$$
(43)
This parameter controls the type of solidification microstructure.
### 4.4 Cooling rate.
As we can see $`G_l(t)1/\sqrt{t}`$. Liquidus velocity (36) varies with time alse as $`v_l(t)1/\sqrt{t}`$, then cooling rate given by $`G_lv_l1/t`$. It is easy to show that is so. The cooling rate is defined as
$$\dot{T}(x,t)=\frac{T(x,t)}{t}=\frac{dT}{dH}\frac{H}{t}$$
(44)
$$\frac{T}{t}=\{\begin{array}{cc}\frac{1}{\rho C_s}\frac{H_sH_{out}}{erf\left(\frac{k_s}{2\sqrt{\alpha _s}}\right)}\frac{\alpha _sxe^{\frac{x^2}{4\alpha _st}}}{2\sqrt{\pi }(\alpha _st)^{3/2}}\hfill & x<k_s\sqrt{t}\hfill \\ \frac{(H_lH_s)/\left(erf\left(\frac{k_l}{2\sqrt{\alpha _{sl}}}\right)erf\left(\frac{k_s}{2\sqrt{\alpha _{sl}}}\right)\right)}{\rho \left[(1\lambda )C_s+\lambda C_l+\left[(C_lC_s)T+L\right]\frac{d\lambda }{dT}\right]}\frac{\alpha _{sl}xe^{\frac{x^2}{4\alpha _{sl}t}}}{2\sqrt{\pi }(\alpha _{sl}t)^{3/2}}\hfill & k_s\sqrt{t}xk_l\sqrt{t}\hfill \\ \frac{1}{\rho C_l}\frac{H_{init}H_l}{erfc\left(\frac{k_l}{2\sqrt{\alpha _l}}\right)}\frac{\alpha _lxe^{\frac{x^2}{4\alpha _lt}}}{2\sqrt{\pi }(\alpha _lt)^{3/2}}\hfill & x>k_l\sqrt{t}\hfill \end{array}$$
(45)
Thus, cooling rate at the liquidus point is given by
$$\dot{T}_l=\dot{T}|_{x=X_l+0}=\frac{1}{\rho C_l}\frac{H_{init}H_l}{erfc\left(\frac{k_l}{2\sqrt{\alpha _l}}\right)}\frac{k_le^{\frac{k_l^2}{4\alpha _l}}}{2\sqrt{\pi \alpha _l}t}\frac{1}{t}.$$
(46)
The value $`\dot{T}_l`$ is very important, because it defines secondary arm dendrite spacing . The another important expression is $`G_l^{1/2}v_l^{1/4}`$, which defines primary arm dendrite spacing . From Eqs. (36) and (43) we can show that primary arm spacing varies versus time like $`t^{3/8}`$. However, as it’s very known, after some critical gradient and velocity at the liquidus point will be take a place columnar to eqiaxed transition .
### 4.5 Local solidification time
For directional solidificcation, the local solidification time $`t_{ls}`$ can be estimated from the following equation:
$$t_{ls}(x)=\left[\frac{1}{k_s^2}\frac{1}{k_l^2}\right]x^2,$$
(47)
where quadratic increasing with $`x`$ of $`t_{ls}`$ we have, because the mushy zone lehgth increases versus $`x`$ and solidus/liquidus velocities decrease. The local solidification time controls the some segragation processes in the mushy zone.
## 5 Numerical scheme.
We will solve heat transfer equation
$`\rho C(T){\displaystyle \frac{T}{t}}={\displaystyle \frac{}{x}}\left(\kappa (T){\displaystyle \frac{T}{x}}\right)`$ (48)
which concerns with Eq. (23). Here $`\kappa (T)`$ edfined by Eq. (5) and mushy heat capacity (so-called apparent capacity):
$$C(T)=\left[1\lambda (T)\right]C_s+\lambda (T)C_l+\left[(C_lC_s)T+L\right]\frac{d\lambda (T)}{dT}.$$
(49)
The first, we draw the grid with spatial step $`h=x_{i+1}x_i`$, where $`i=0,1,\mathrm{},N`$. The second, integrating the Eq. (48) over the $`x[(i\frac{1}{2})h,(i+\frac{1}{2})h]`$ we can write heat balance equation as follows
$$\rho C(T_i)\frac{T_i}{t}h=\left(\kappa (T)\frac{T}{x}\right)|_{x=(i\frac{1}{2})h}^{x=(i+\frac{1}{2})h}\kappa _{i+\frac{1}{2}}\frac{T_{i+1}T_i}{h}\kappa _{i\frac{1}{2}}\frac{T_iT_{i1}}{h}.$$
(50)
The left part of this equation we express as
$$\rho C(T_i)\frac{T_i}{t}\rho C(T_i^n)\frac{T_i^{n+1}T_i^n}{\tau },$$
where $`n`$ is time index, $`\tau `$ is time step. Thus the Eq. (48) we can write down in the discrete form as
$$\left[\frac{\tau \kappa _{i\frac{1}{2}}}{h^2}\right]T_{i1}^{n+1}\left[\frac{\tau }{h^2}\left(\kappa _{i\frac{1}{2}}+\kappa _{i+\frac{1}{2}}\right)+\rho C_i\right]T_i^{n+1}+\left[\frac{\tau \kappa _{i+\frac{1}{2}}}{h^2}\right]T_{i+1}^{n+1}=\rho C_iT_i^n,$$
(51)
where
$$\kappa _{i\frac{1}{2}}=\frac{2\kappa _{i1}\kappa _i}{\kappa _{i1}+\kappa _i},\kappa _{i+\frac{1}{2}}=\frac{2\kappa _i\kappa _{i+1}}{\kappa _i+\kappa _{i+1}}.$$
We note, that $`C_i`$ etc are calculated at time $`t_n`$, i.e.
$$C_i=C(T_i^n),\kappa _{i1}=\kappa (T_{i1}^n)etc.$$
Boundary conditions are
$$T_0=T_{out},T_N=T_{init}.$$
(52)
Because the Eq. (51) is tri-diagonal linear system, then its solving is trivial and we do not discuss this issue here.
## 6 Binary alloy solidification: numerical treatment.
In this section we consider the solidification of noneutectic titanium-based alloy, which we can treat as pseudo-binary alloy. Physical properties of titanium alloy VT3-1 (Ti-6.5Al-2.5Mo-1.5Cr-0.5Fe-0.3Si) are present in the Table 1. These parameters we are used for numerical simulation of liquid pool profiles during vacuum arc remelting process .
A solution of the Eq. (17) with $`\lambda _0=0`$ is $`a_{sl}=2.26891\times 10^7m^2/s`$. Figure 1 shows the temperature dependence of the liquid fraction. Additionaly Figure 1 shows the function
$$\lambda _t(T)=1\frac{T_mT_S}{T_lT_s}\frac{T_lT}{T_mT},$$
(53)
which we used for VT3-1 alloy . The difference between $`\lambda _T`$ and $`\lambda _t(T)`$ is small, then we have nearly realistic problem. If $`g(T)`$ approximated with a power function
$$\lambda _n=\left(\frac{TT_s}{T_lT_s}\right)^n,$$
then we get $`n1.5`$. To test very simple numerical apparent capacity-based method we carried out simulations with following parameters: $`a_s=3.7037\times 10^6m^2/s`$, $`a_l=6.48148\times 10^6m^2/s`$, $`T_{out}=800^oC`$, $`T_{init}=1650^oC`$, $`H_{out}=2.16\times 10^9J/m^3`$, $`H_{init}=10.5057\times 10^9J/m^3`$, $`H_s=4.185\times 10^9J/m^3`$, $`H_l=10.3455\times 10^9J/m^3`$.
Solutions of Eqs. (34)-(35) are $`k_s=0.00134109m/s^{1/2}`$, $`k_l=0.00206009m/s^{1/2}`$.
The numerical model parameters are chosen as: length of domain $`d=0.5m`$, nodes number $`N=500`$, time step $`\tau =0.1s`$. A numerical model can provide excellent agreement with obtained analytical solution. The results obtained are in a Figure 2: movement of both the solidus and the liquidus front (26). We used linear interpolation between $`T_i`$ and $`T_{i+1}`$ for estimating position of the fronts, whereas $`T_{s,l}[T_i,T_{i+1}]`$.
The errors in the positions of solidus/liquidus
$$\epsilon _x(t)=\frac{X_{num}(t)X_{exact}(t)}{X_{exact}(t)}100\%$$
are presented in the Figure 5. Moreover Figure 4 shows the temperature profiles after 20 and 500 seconds under the same numerical conditions.
Figure 5 shows temperaure profiles errors defined by
$$\epsilon _T(x)=\frac{T_{num}(x)T_{exact}(x)}{T_{exact}(x)}100\%.$$
We would like to underline that the purpose of this work is to achieve the exact analytical solution on alloy solidification. Due to this, advantages and disatvantages different numerical algorithms can be done in future. Due to this, we don’t study the process of solidification of Ti-6.5Al-2.5Mo-1.5Cr-0.5Fe-0.3Si alloy, but only use the thermo-physical properties of this alloy to show as the model works.
## 7 Conclusions.
In this paper, analytical solution of alloy solidification problem is presented. We developed a special method to obtain an exact analytical solution for mushy zone problem. The main requirement of the method is thermal diffusivity to be constant in the mushy zone. Due to such condition ordinary differential equation for liquid fraction function is achieved. Thus the present method can be examined as ”a model” way to get analytical solution of some unrealistic problems.
An example of solutions is given – the noneutectic titanium-based alloy solidification. We provide the comparison of the simple numerical simulation results with obtained exact solutions. We show that very our numerical apparent capacity-based scheme provides a good agreement with exact solutions for solidus/liquidus position and for temperatures profiles in different moments of solidification time.
Once again we would like to underline that the main goal of this paper to provide the benchmark for binary alloy solidification problem, but not in the analysis of used numerical scheme.
If predefined $`\lambda (T)`$ function is to be examined, we can use another suggestions. For example, we can require to heat conductivity (from experiment, e.g.) to be proportional to the apparent capacity, i.e.
$$\kappa (T)=a_{sl}\rho \left(C_s+(C_lC_s)\lambda (T)+\left[(C_lC_s)T+L\right]\frac{d\lambda (T)}{dT}\right).$$
Or, for the second example, we require to apparent capacity (from experiment) to be proportional to mushy heat conductivity, i.e.
$$\frac{dH(T)}{dT}=\frac{\kappa _s+(\kappa _l\kappa _s)\lambda (T)}{a_{sl}}.$$
Moreover, we may use the Bäcklund’s transformation to make mushy heat equation linearisation. In this case we get nonlinear condition
$$\frac{H^2(T)\lambda (T)}{\frac{dH(T)}{dT}}=const.$$
These linearization methods will provide us with some additional analytical solutions of alloy solidification problem.
The author thanks to Prof. V. Alexiades from ORNL for very usefull discussion.
## References |
warning/0507/astro-ph0507569.html | ar5iv | text | # Rotational velocities of the giants in symbiotic stars: I. D’–type symbiotics based on observations obtained in ESO programs 073.D-0724A and 074.D-0114
## 1 Introduction
The Symbiotic stars (SSs – thought to comprise a white dwarf (WD) accreting from a cool giant or Mira) represent the extremum of the interacting binary star classification. They offer a unique and exciting laboratory in which to study such important processes as (i) mass loss from cool giants and the formation of Planetary Nebulae; (ii) accretion onto compact objects, (iii) photoionisation and radiative transfer in gaseous nebulae, and (iv) nonrelativistic jets and bipolar outflows (i.e. Kenyon 1986; Corradi et al. 2003).
Soker (2002) has shown theoretically that the cool companions in symbiotic systems are likely to rotate much faster than isolated cool giants or those in wide binary systems. However, there are no systematic investigations of $`v\mathrm{sin}i`$ of the mass donors in SS.
On the basis of their IR properties, SS have been classified into stellar continuum (S) and dusty (D or D’) types. The D–type systems contain Mira variables as mass donors. The D’–type are characterized by an earlier spectral type (F-K) of the cool component. and lower dust temperatures. Among 188 objects in the latest catalogue of symbiotic stars (Belczyński et al. 2000), there are only seven that are classified as D’–type: Wray15-157, AS 201, V417 Cen, HD 330036, AS 269, StH$`\alpha `$ 190 and Hen 3-1591 (Hen 3-1591 can be S or D’). Three of these have been studied using model atmospheres and all display rapid rotation and s-process elemental over-abundances (see Pereira et al. 2005).
Our aims here are: (1) to measure the projected rotational velocities ($`v\mathrm{sin}i`$) and the rotational periods (P<sub>rot</sub>) of the giants in D’ symbiotic stars, using a cross correlation function (CCF) approach; (2) to test the theoretical predictions that the mass donors in SSs are faster rotators than the isolated giants or those in wide binary systems; (3) to provide pointers to the determination of binary periods (assuming co-rotation). This is the first of a series of papers exploring the rotation velocities of the mass donating (cool) components of SSs.
## 2 Observations
The observations were performed with FEROS at the 2.2m telescope (ESO, La Silla). FEROS is a fibre-fed Echelle spectrograph, providing a high resolution of $`\lambda /\mathrm{\Delta }\lambda =`$48000, a wide wavelength coverage from about 4000 Å to 8000 Å in one exposure and a high overall efficiency (Kaufer et al. 1999). The 39 orders of the Echelle spectrum are registered on a 2k$`\times `$4k EEV CCD. All spectra are reduced using the dedicated FEROS data reduction software implemented in the ESO-MIDAS system.
## 3 $`v\mathrm{sin}i`$ measurement technique
Radial velocities and projected rotational velocities have been derived by cross-correlating the observed spectra with a K0-type numerical mask yielding a cross-correlation function (CCF) whose centre gives the radial velocity and whose width is related to the broadening mechanisms affecting the whole spectra such as stellar rotation and turbulence. Details of the cross-correlation procedure are given in Melo et al. (2001).
In order to use the width ($`\sigma `$) of the CCF as an estimate of $`v\mathrm{sin}i`$ one needs to subtract the amount of broadening contributing to $`\sigma `$ unrelated to stellar rotation (e.g., convection, instrumental profile, etc.), i.e., $`\sigma _0`$. Melo et al. (2001) calibrated $`\sigma _0`$ as a function of $`(BV)`$ for FEROS spectra of stars with $`0.6<(BV)<1.2`$. The $`v\mathrm{sin}i`$ measured from our CCFs for a set of standard stars within this $`BV`$ range are in good agreement with the literature values. Therefore, for all 4 giants in Table 1, the Melo et al. (2001) calibration has been adopted.
For $`v\mathrm{sin}i`$ greater than about 30 km s<sup>-1</sup> the shape of the CCF becomes gradually closer to the Gray rotational profile (Gray 1976). Therefore in order to correctly fit the CCF, a different approach is needed as described in Melo (2003). The CCF is then fitted by a family of functions $`CCF_{V\mathrm{sin}i}=CD[g_0G(V\mathrm{sin}i)]`$ which is the result of the convolution of the CCF of a non-rotating star $`g_0`$, which can be fairly approximated by a gaussian, and the Gray (Gray 1976) rotational profile computed for several rotational velocities $`G(V\mathrm{sin}i)`$. For each function $`CCF_{V\mathrm{sin}i}`$ we found the radial velocity $`V_r`$, the depth $`D`$ and the continuum $`C`$ for minimizing the quantity $`\chi _{V\mathrm{sin}i}^2`$ which is the traditional $`\chi ^2`$ function with $`\sigma _i=1`$, where $`\sigma _i`$ is the measurement error (see Fig. 1 of Melo 2003, for an example of the procedure). The CCFs of our objects are plotted in Fig. 1.
For $`v\mathrm{sin}i`$$`>30`$ km s<sup>-1</sup> the typical error of our $`v\mathrm{sin}i`$ measurements is 10%. For $`10<`$$`v\mathrm{sin}i`$$`30`$ km s<sup>-1</sup> the error is about 1.5 km s<sup>-1</sup>. Our measurements are given in Table 1.
## 4 Rotation of the mass donors
### 4.1 Individual objects
HD 330036 : Pereira et al. (2005) obtained L=650 L for the cool component (with possible uncertainties 160$`<L<`$3000 L), T$`{}_{eff}{}^{}=`$6200$`\pm `$150 K, $`\mathrm{log}g=2.4`$$`\pm `$0.7. This imply $`R_g=22`$R (using $`L=4\pi \sigma _{SB}R_g^2T_{eff}^4`$), M$`{}_{g}{}^{}=`$4.46 $`M_{}`$ (using R<sub>g</sub> and $`\mathrm{log}g`$), and P$`{}_{rot}{}^{}10.4`$$`\pm `$2.4 d (using P$`_{rot}`$$`v\mathrm{sin}i`$$`2\pi R_g`$).
Hen 3-1591 : Medina Tanco & Steiner (1995) give spectral type K1 for the cool component. We assume that it is luminosity class III. The average radius of a K1III star is 23.9$`\pm `$3 R and the average T$`{}_{eff}{}^{}=`$4280$`\pm `$200 K (van Belle et al. 1999). A K1III star would have a mass of 3.9$`\pm `$0.3 $`M_{}`$ (Allen 1973). The uncertainties correspond to $`\pm `$0.5 spectral types. This will result in L=172 L($`\pm `$20%) and P$`{}_{rot}{}^{}51.0`$$`\pm `$11.3 d.
StH$`\alpha `$ 190 : The cool component is of type G4 III/IV with T$`{}_{eff}{}^{}=`$5300$`\pm `$150 K, $`\mathrm{log}g=3.0`$$`\pm `$0.5, and L=45 L (Smith et al. 2001). This implies R$`{}_{g}{}^{}=7.9\pm 0.4`$ R and P$`{}_{rot}{}^{}3.8`$$`\pm `$1.2 days (upper limit calculated for R$`{}_{g}{}^{}=8.3`$ R and $`\mathrm{sin}i=1.0`$). The upper limit for P<sub>rot</sub> is considerably shorter than the supposed orbital periods of 171 d (Munari et al. 2001) or 38 d (Smith et al. 2001).
V417 Cen : Van Winckel et al. (1994) detected a photometric period of 245 days. For the cool component they obtained G9 Ib-II, $`\mathrm{log}`$ L/L$`{}_{\mathrm{}}{}^{}=`$3.5, T$`{}_{eff}{}^{}=`$5000 K, $`\mathrm{log}g=1.5`$$`\pm `$0.5. This implies $`R_g=75`$R and P$`{}_{rot}{}^{}50.6`$$`\pm `$10.2 days. P<sub>rot</sub> is considerably shorter than the period obtained from photometry. However, the photometric period is not confirmed with radial velocity measurements and we do not know whether this is the orbital period.
AS 201 : Following Pereira et al. (2005), the cool component is of type F9III with T$`{}_{eff}{}^{}=`$6000$`\pm `$100 K, L=700 L (with possible uncertainties $`300<L<1200`$ L), and $`\mathrm{log}g=2.3`$$`\pm `$0.3. This imply R$`{}_{g}{}^{}=24.5`$ R and P$`{}_{rot}{}^{}49.5`$$`\pm `$11.0 days.
### 4.2 Projected rotational velocities $`v\mathrm{sin}i`$\- comparision with catalogues
There are no systematic investigations of the rotation of the mass donors in SSs. The rotational velocities of 13 S–type systems listed in Fekel et al. (2003) are between $`v\mathrm{sin}i`$= 3.6 - 10.4 km s<sup>-1</sup>. All D’–type systems so far observed (see Table 2) rotate with $`v\mathrm{sin}i`$$`>`$20 km s<sup>-1</sup>.
The catalog of de Medeiros et al. (2002) of $`v\mathrm{sin}i`$ of Ib supergiant stars contains 16 objects from spectral type G8-K0 Ib-II. All they have $`v\mathrm{sin}i`$ in the range 1-20 km s<sup>-1</sup>. It means that V417 Cen is an extreme case of very fast rotation for this spectral class.
The catalogue of rotational velocities for evolved stars (de Medeiros et al. 1999) lists $``$100 K1III stars, and 90% of them rotate with $`v\mathrm{sin}i`$$`<`$8 km s<sup>-1</sup>. There are only 5 with $`v\mathrm{sin}i`$$`>`$20 km s<sup>-1</sup>. This means that Hen 3-1591 is a very fast rotator (in the top 5%). The same catalog contains 5 objects from spectral type F8III-F9III. They rotate with $`v\mathrm{sin}i`$ of 10-35 km s<sup>-1</sup>. AS 201 is well within in this range. However HD 330036 is an extremely fast rotator. The same catalog lists 60 objects from spectral type G3,G4,G5 III-IV. They all rotate with $`v\mathrm{sin}i`$ $`<24`$ km s<sup>-1</sup>. Again, this means that StH$`\alpha `$ 190 is an extremely fast rotator.
Thus overall, 4 out of 5 D’–type SSs in our survey are very fast rotators.
### 4.3 Critical speed of rotation
There is a natural upper limit for rotation speeds, where the centripetal acceleration balances that due to gravitational attraction, often named the “critical speed”, where $`v_{\mathrm{crit}}=\sqrt{GM/1.5R}=357\sqrt{M/R}`$ km s<sup>-1</sup> (the factor of 1.5 appears from the assumption that at critical rotation speeds the equatorial radius is 1.5 times the polar radius, $`R`$). The calculated $`v_{\mathrm{crit}}`$ is included in Table 2. No star can rotate faster than its critical speed, however we can see that at least three D’–type SSs are rotating at a substantial fraction of their critical speeds: $`\frac{v\mathrm{sin}i}{v_{\mathrm{crit}}}`$ $`0.67`$ (HD 330036), $`0.54`$ (StH$`\alpha `$ 190), $`0.71`$ (V417 Cen). Probably for these three SSs the orbital inclination is high $`i50^0`$. For the remaining two objects we can not exclude the possibility that they also rotate very fast but are observed at low inclination ($`i30^0`$).
## 5 Synchronization and binary periods
### 5.1 Synchronization in SS
The physics of tidal synchronization for stars with convective envelopes has been analyzed several times (e.g. Zahn 1977 and also the discussion in Chapter 8 of Tassoul 2000). There are some differences in the analysis of different authors, leading to varying synchronization time-scales. Here, we use the estimate from Zahn (1977, 1989): the synchronization time-scale in terms of the period is
$$\tau _{\mathrm{syn}}800\left(\frac{M_gR_g}{L_g}\right)^{1/3}\frac{M_g^2(\frac{M_g}{M_{WD}}+1)^2}{R_g^6}P_{orb}^4\mathrm{years}$$
(1)
where $`M_g`$ and $`M_{WD}`$ are the masses of the giant and white dwarf respectively, and $`R_g`$ and $`L_g`$ are the radius and luminosity of the giant (all in Solar units). The orbital period $`P`$ is measured in days.
The S–type SSs are very likely synchronized (Schild et al. 2001; Schmutz et al. 1994). Other proof of this supposition is that most of the SSs with derived orbital parameters (see Mikołajewska 2003) have orbital eccentricity $`e`$0. Because the circularization time of the orbits is $``$10 times longer than the synchronization time (Schmutz et al. 1994 and the references therein), if a SS’s orbit is circularized it will very likely be synchronized too.
A typical D’–type SS would have $`R_g20R_{},L_g500L_{},M_g3M_{}`$, and white dwarf mass $`M_{WD}1M_{}`$. For a period of 100 days for a typical D’-type SS, we find $`\tau _{\mathrm{syn}}9\times 10^4`$ years.
For the individual systems, we calculated the synchronization time ($`\tau _{syn}`$) for two cases: $`\tau _{syn}`$ is derived assuming P$`{}_{orb}{}^{}=`$P<sub>rot</sub>, and $`\tau _{syn}(100)`$ assuming P$`{}_{orb}{}^{}=`$100 days. These are given in Table 2. Depending on the individual parameters, the synchronization time can be from $`<`$100 yr up to $`>`$10<sup>7</sup> yr. This means that it is possible for a D’–type SS to be synchronized if the orbital period is short (P$`{}_{orb}{}^{}`$P$`{}_{rot}{}^{})`$.
### 5.2 Evolutionary status of the mass donors
The mass of the mass donors in S–type SSs with known parameters are in the range 0.6 - 3.2 $`M_{}`$ (Mikołajewska 2003). The calculated masses of the mass donors in D’–type SSs are larger. As can be seen in Table 2, they range from 2.2 up to 6.5 $`M_{}`$.
Masses of 8 M are generally considered the upper limit for evolution to planetary nebula nuclei and white dwarfs, after heavy mass loss, especially during their AGB phases. Following our calculations for the mass of the giants (Table 2) and assuming M$`{}_{WD}{}^{}1.4M_{}`$, the total mass of the binary is about (M$`{}_{g}{}^{}+`$M$`{}_{WD}{}^{})3.58.0M_{}`$, in agreement with the above upper limit of 8 M for the WD progenitor.
The position of the mass donors on the H-R diagram is presented in Fig.2, assuming near solar chemical composition and stellar parameters as given in Sect.4.1 (see also Pereira 2005). The evolutionary tracks of Shaller et al. (1992) have been used. The donors appeared in a wide mass interval – from 2.5 to $`7M_{}`$. The derived evolutionary masses are in good agreement with those obtained from R<sub>g</sub> and $`\mathrm{log}g`$. Three of them appeared crossing the Hertzsprung gap (HD 330036,V417 Cen and AS 201), StH$`\alpha `$ 190 is situated near the base of the red giant branch, and He 3-1591 is already evolving on the red giant branch.
The relevant time for a $`5M_{}`$ star to cross the Hertzsprung gap is $``$8$`\times `$10<sup>5</sup> yr and its life time on the red giant branch is $``$5$`\times `$10<sup>5</sup> yr (Iben 1991). For a $`1.5M_{}`$ star these times are 1.5$`\times `$10<sup>8</sup> yr and 1.57$`\times `$10<sup>8</sup> yr (Iben 1991). These lifetimes are longer than the calculated $`\tau _{\mathrm{syn}}`$ and comparable with $`\tau _{\mathrm{syn}}(100)`$. This means that the rotation of the mass donors in D’–type SSs could be synchronized for these lifetimes.
### 5.3 Clues regarding the orbital periods
Because the orbital periods of the majority of SSs are unknown, an indirect method to obtain P<sub>orb</sub> is to measure $`v\mathrm{sin}i`$. If the mass donors in SSs are co-rotating (P$`{}_{rot}{}^{}=`$P<sub>orb</sub>), we can find clues for the orbital periods via the simple relation P<sub>orb</sub> v<sub>rot</sub> = 2$`\pi `$R<sub>g</sub>, where P<sub>orb</sub> is the orbital period, v<sub>rot</sub> and R<sub>g</sub> are the rotational velocity and radius of the giant, respectively. It is very useful in the case of eclipsing binaries, where $`\mathrm{sin}i1`$ and $`v\mathrm{sin}i`$$``$v<sub>rot</sub>. If the inclination is unknown, we can only obtain an upper limit for P<sub>orb</sub>.
Using Kepler’s third law \[$`4\pi a^3=G(M_g+M_{WD})P^2`$\], we can calculate the semimajor axis of the systems. These are given in Table 2. The values in the brackets (for P<sub>rot</sub> and $`a`$) correspond to the estimation of the upper limit of these parameters, assuming 10% errors in $`v\mathrm{sin}i`$ and R<sub>g</sub>.
Up to now, from 188 SSs, the orbital elements and binary periods are well known for $``$40 objects only (and they are all S–type). The orbital periods are in the range 200 - 2000 days (Mikołajewska 2003). As can be seen in Table 2, if D’–type SSs are synchronized, their orbital periods would be relatively short (4-60 days) and the distance between the WD and the mass donor would be 2-5 R<sub>g</sub>.
## 6 Discussion
The observation of fast rotation of D’–type SSs raises two further questions: first, what is the evolutionary history of these stars which has produced such high rotation, and second, what does the effect of high rotation have on their mass loss and subsequent dust formation?
We see three possible reasons for the fast rotation of mass donors in D’–type SSs:
(i) the rotation is synchronized with the binary period (P$`{}_{rot}{}^{}=`$ P<sub>orb</sub>). In this case their orbital periods should be short $``$50 days.
However, it also could be, that they are not synchronously rotating. The possibility that P$`{}_{rot}{}^{}>`$ P<sub>orb</sub> has to be excluded because the orbital separations would be unreasonably small, and the synchronization time would be extremely short (they will be synchronized in 30-9000 years; see $`\tau _{syn}`$ in Table 2). If P$`{}_{rot}{}^{}<`$ P<sub>orb</sub>, reasons for their fast rotation could be:
(ii) the current giants have been spun-up from the transfer of angular momentum. Jeffries & Stevens (1996) proposed a mechanism in which accretion of a slow massive wind from the AGB progenitor of the current white dwarf can transfer sufficient angular momentum. This also explains the chemical enrichment in s-process elements in D’ SSs, that were present in the AGB wind (see also Pereira et al. 2005). Mass transfer via L<sub>1</sub>, when the current white dwarf was the mass donor, can also spin-up the companion, as in millisecond radio pulsars (van den Heuvel 1984).
(iii) planet swallowing to spin up the giant. Rough estimation gives angular momentum transfer during a collision of a planet with mass $`m_p`$ and velocity $`v_p`$ to a giant of mass $`M_g`$, of the order of $`\mathrm{\Delta }\mathrm{\Gamma }=m_pv_pR_g`$. This causes a change in the giant’s rotational velocity $`\mathrm{\Delta }v_g^{rot}=m_pv_p/const.M_g`$, where const.=$`J_g/M_gR_g^2`$ depends on the internal structure of the giant. ($`const.=0.4`$ for a solid uniform density sphere, and less for a star-like centrally condensed sphere). Assuming a centrally condensed star (giant) such that $`const.=0.01`$, $`v_p=\sqrt{GM_g/R}10100`$ km s<sup>-1</sup>, $`M_g=210M_{}`$ and $`m_p=0.01M_{}`$ we estimate $`\mathrm{\Delta }v_g^{rot}150`$ km s<sup>-1</sup>, showing that in the right circumstances, the planet could spin up the giant to the rotational velocities observed in D’–type SSs.
Fast rotation, i.e. $`v_{rot}0.5v_{\mathrm{crit}}`$, may change a spherical star with a spherical wind into an equatorially flattened system, with both the radius of the star and stellar wind parameters depending on the stellar latitude. Such stars will have an equatorial radius significantly larger than the polar one, and equatorially enhanced mass loss (see Lamers 2004 and references therein).
Because it seems that the majority of the giants in D’–type SSs are rapid rotators (Sect. 4.3), we expect that:
(1) they have a larger mass loss rate than the slower rotating giants;
(2) their mass loss is enhanced in the equatorial regions;
(3) they could be even equatorially flattened.
It is possible, that the dusty environment in D’–type SSs is connected with rapid rotation of the mass donors. Intense mass loss in the equatorial regions can lead to the formation of an excretion disc in which the higher gas density enhances dust formation and growth. The broad IR excess in D’–type SSs can be due to the temperature stratification in the dust from such an excretion disc (Van Winckel et al. 1994). Other possible explanations for the presence of dust can be that it is left over from the formation of a planetary system, or it is a relic from a strong dusty mass loss when the present day white dwarf was on the AGB. However, we consider it is more likely that it originates in the current outflow and that this is enhanced in the equatorial regions by rapid rotation.
## 7 Conclusions
Our main results are:
(1) We have measured the rotational velocities of the mass donors in D’–type symbiotic stars, using the CCF approach.
(2) Four of the five objects appeared to be very fast rotators compared with the catalogues of $`v\mathrm{sin}i`$ for the corresponding spectral types. At least three of them rotate at a substantial fraction ($`0.5`$) of the critical velocity. This means that at least in D’–type SSs, the cool components rotate faster than isolated giants (as predicted by Soker 2002).
(3) If they are tidally locked, the orbital periods should be as short as $``$10-50 days.
(4) As a result of the rapid rotation, they must have larger mass loss rates than the more slowly rotating giants, and their mass loss is probably enhanced in the equatorial regions.
To understand better these objects, we need their binary periods to be derived from radial velocity measurements and the inclination determined. In subsequent papers, we plan to explore the projected rotational velocity of the cool giants in the other types of symbiotic stars and to compare their rotational velocities with that of the isolated giants with similar mass and evolutionary stage.
## Acknowledgments
This research has made use of MIDAS, IRAF, SIMBAD and Starlink. RZ is supported by a PPARC Research Assistantship and MFB is a PPARC Senior Fellow. We also acknowledge the vital contribution made by Dr John Porter to our successful telescope time proposals and to this paper, which was underway at the time of his tragic death. |
warning/0507/math0507595.html | ar5iv | text | # Untitled Document
Nilpotents, Integral Closure and Equisingularity conditions
Terence Gaffney
Introduction
The motivation for this note comes from the fact that if you want an equisingularity theory for general singularities parallel to the complete intersection case, you must admit nilpotents into your structure sheaves.
Here is a simple example. Consider a family whose members are 2-planes in 4-space meeting transversally at a single point, with one plane fixed and the other moving. In example 1 we show that if we take a hyperplane section of a member of this family we get a non-reduced space; if we take a hyperplane section of the total space of this family, which contains the parameter space, we again get a space with nilpotents.
We could take the reduced structure on the intersection, but the fat point says something interesting. It exists because of the possibilty of pulling the two lines which make up the hyperplane section apart by a flat deformation. So the question arises, how do we have to change the theory to take nilpotents into account? In , we gave an algebraic condition for the Whitney conditions to hold in terms of the integral closure of the Jacobian module associated to the defining equations of the analytic set. Assume $`X`$ is equidimensional as a set. In this note, in section 1 we show that we don’t need to take the reduced struture for $`X`$ in checking this algebraic criterion, we can use the generators for any ideal which will give the reduced structure on $`X`$ at smooth points of $`X`$.
This is useful in applications where often the set of equations constructed by some geometric process are only known to give the reduced structure at the set of smooth points.
So the integral closure approach is blind to nilpotents just as the Whitney conditions are. As a consequence, in section 2, in Theorem 2.2, we can prove easily that the Whitney conditions are preserved by intersection with a generic hyperplane using the theory of integral closure, and we can describe precisely when a hyperplane is generic in this sense.
In the course of studying what it means for a hyperplane to be generic for this situation, we are led to prove a result in the case of families of complete intersections with isolated singularities, relating the limiting tangent hyperplanes to a complex analytic set $`X`$ and the fibers of $`X`$ over $`Y`$ (Theorem 2.6). This allows an improvement of Theorem 2.2 in this case.
The author thanks Steven Kleiman for helpful conversations. This project grew out of a collaboration with David Trotman and Leslie Wilson; the author thanks both of his collaborators for their interest and support.
1. Equisingularity and Nilpotents
We start with a motivating example.
Example (1.1) Let $`X`$ be the germ of an analytic set defined at the origin in $`𝐂^5`$ with reduced structure by
$$\{x(w+ty)=0,xz=0,y(w+ty)=0,yz=0\}.$$
The structure is reduced because the collection $`\{x,y,z,w+ty\}`$ is an R-sequence in $`𝒪_5`$, the local ring of germs of analytic functions at the origin in $`𝒪_5`$, so the only relations between these elements are the Kozul relations. The projection from $`𝐂^5`$ to the $`t`$-axis, makes $`X`$ into a family of sets; the fiber over $`0`$ , $`X(0)`$ is defined by
$$\{xw=0,xz=0,yw=0,yz=0\},$$
and is also reduced for the same reason. So, $`X`$ is a family consisting of 2 two-planes, one plane fixed and the other moving. The $`t`$-axis is the parameter space of the family and is also the singular set of $`X`$.
However, if we intersect with a hyperplane of the form $`x=ay+bz+cw`$, $`c0`$, the simplest structure to use on the intersections is that given by adding the equation of $`H`$ to the equations defining $`X`$ and $`X(0)`$. If we use the equation for $`H`$ to eliminate $`x`$, then this structure on $`XH`$ contains the nilpotent $`(b/c)z+w+ty`$, while the structure on $`X(0)H`$ contains the nilpotent $`(b/c)z+w`$. (If we eliminate $`x`$, the ”simple structure“ is defined by the product of the ideals $`I_1=(bz+cw,y)`$ and $`I_2=(w+ty,z)`$, while the reduced structure is defined by the intersection of the two ideals, and the intersection contains the element $`(b/c)z+w+ty`$ which is not in the product.)
As we shall see, many equisingularity conditions are described using information from the Jacobian module of $`X`$. If $`X,x𝐂^n`$ is defined as $`F^1(0)`$ where $`F:𝐂^n,x𝐂^p,0`$, then the Jacobian module of $`X`$ with the structure defined by $`F`$, is just the submodule of $`𝒪_{X,x}^p`$ generated by the partial derivatives of $`F`$ and denoted $`JM(F)`$. Sometimes we have $`X,x𝐂^k\times 𝐂^n`$, with $`y`$ a set of coordinates on $`𝐂^k`$ and $`z`$ a set of coordinates on $`𝐂^n`$. In this case, we let $`JM_z(F))`$ denote the submodule of $`𝒪_{X,x}^p`$ generated by the partial derivatives of $`F`$ with respect to $`z`$, $`JM_y(F)`$ denotes the submodule of $`𝒪_{X,x}^p`$ generated by the partial derivatives of $`F`$ with respect to $`y`$. The example of the previous paragraph shows that if we wish to pass information from the Jacobian module of $`X`$ to that of $`XH`$ easily, we must deal with nilpotent structures on both the total space and the members of the family.
We now introduce the notion of integral closure, and begin to consider the connection between nilpotents and this idea.
Recall that if $`R`$ is a commutative, unitary ring, $`I`$ an ideal in $`R`$, $`f`$ an element of $`R`$, then $`f`$ is integrally dependent on $`I`$ if there exists a positive integer $`k`$ and elements $`a_j`$ in $`I^j,`$ so that $`f`$ satisfies the relation $`f^k+a_1f^{k1}+\mathrm{}+a_{k1}f+a_k=0`$ in $`R.`$
The integral closure of an ideal $`I`$, denoted $`\overline{I}`$, consists of all elements integrally dependent on it.
Proposition (1.2)Supose $`I`$ is an ideal in a commutative, unitary ring $`R`$. If $`h`$ is a nilpotent element of $`R`$, then $`h\overline{I}`$.
Proof. Since $`h`$ is nilpotent it satisfies an equation of the form $`T^k=0`$ for some $`k`$.
Proposition (1.3)Suppose $`R`$ as above, and $`R_r`$ is the ring gotten by modding out by nilpotents, p the projection map. Suppose $`I`$ is an ideal in $`R`$, $`h`$ an element of $`R`$, then $`\overline{I}`$ contains $`h`$ iff $`\overline{(p(I))}`$ contains $`p(h)`$.
Proof. If $`\overline{I}`$ contains $`h`$, then we just reduce mod nilpotents and the result follows. Suppose $`\overline{(p(I))}`$ contains $`p(h)`$. Then $`p(h)`$ satisfies an equation of integral dependence. Choosing representatives of the coefficients of the equation we have $`h`$ satisfies an equation of the form $`P(h)=g`$ where $`g`$ is nilpotent. Then consider $`P^k(h)=g^k`$. For $`k`$ large enough, $`g^k=0`$, and $`P^k`$ is a polynomial of the desired form, because it is a product of polynomials of the desired form.
We can also talk about the integral closure of a module. Suppose $`M`$ is a submodule of $`E=R^k`$, $`k`$ at least 1. Let $`\rho :ESE`$ be the inclusion of $`E`$ into its symmetric algebra; then $`hE`$ is integrally dependent on $`M`$ if $`\rho (h)`$ is integrally dependent on the ideal generated by $`\rho (M)`$.
If $`NM`$ are a pair of modules such that $`\overline{N}=\overline{M}`$, then we say that $`N`$ is a reduction of $`M`$.
Corollary (1.4)Suppose $`R`$ is a commutative, unitary ring, $`R_r`$ is the ring gotten by modding out by nilpotents, p the projection map from $`R^k`$ to $`R_r^k`$, suppose $`M`$ is a submodule in $`R^k`$, $`h`$ an element of $`R^k`$, then $`\overline{M}`$ contains $`h`$ iff $`\overline{(p(M))}`$ contains $`p(h)`$.
Proof. Translate from modules to ideals and use proposition 1.2.
There is a useful criterion (curve criterion) for checking if an element is in the integral closure of a module, which also allows us to define the notion of strict dependence. Suppose $`X,x`$ is a complex analytic germ, $`M`$ a submodule of $`𝒪_{X,x}^p`$. Then $`h𝒪_{X,x}^p`$ is in the integral closure of $`M`$ (resp. strictly dependent on $`M`$) iff for all $`\varphi :𝐂,0X,x`$, $`h\varphi (\varphi ^{}M)𝒪_1`$ (resp. if for all $`\varphi :𝐂,0X,x`$ we have $`h\varphi m_1\varphi ^{}M`$, where $`m_1`$ is the maximal ideal in $`𝒪_1`$). (For a discussion of the curve criterion cf. .)
We denote the set of elements strictly dependent on $`M`$ by $`M^{}`$.
There is also a version of Nakayama’s lemma for integral closure based on the curve criterion.
Proposition (1.5)Suppose $`NM𝒪_{X,x}^p`$, and $`\overline{N+M^{}}=\overline{M}`$, then $`N`$ is a reduction of $`M`$.
Proof. Use the curve criterion and Nakayama’s lemma.
The theorem of Rees is known to hold if even if the local ring is not reduced. We have:
Proposition (1.6)Suppose $`X`$ is the germ of an analytic space, such that $`X_{red}`$ is equidimensional, $`I`$ and $`J`$ two ideals in the local ring of $`X`$. Suppose $`e(I)=e(J)`$, $`IJ`$, then $`J`$ in $`\overline{I}`$.
Proof. Cf
In fact, the multiplicity of $`I`$ is the same as the multiplicity of $`(p(I))`$, when the structure on $`R`$ is generically reduced. Here is an argument by Steven Kleiman proving this fact.
We have the map of Rees algebras $`R(I)R(p(I))`$ is surjective and has a nilpotent kernel, so the blowup of Spec$`(R_r)`$ is simply the reduction of the blowup of Spec$`(R)`$, and the exceptional divisor of the first induces to that of the second; the equation now follows from the projection formula applied to the inclusion of the reduced scheme in the nonreduced one.
For the various stratification conditions we still need to do some work, because these depend on the Jacobian module, which means we must differentiate our functions which are giving us the non-reduced structure.
The setup for our theorems is as follows: $`X,0𝐂^{k+n},0`$ is the germ of a complex analytic set at the origin, $`Y=𝐂^k\times 0X`$, $`(y_1,\mathrm{},y_k)`$ coordinates on $`𝐂^k`$, $`(z_1,\mathrm{},z_n)`$ coordinates on $`𝐂^n`$. $`X_0`$ denotes the smooth points of $`X`$, $`X`$ with reduced structure. We denote the ideal sheaf generated by the $`(z_1,\mathrm{},z_n)`$ by $`m_Y`$. We assume $`X_{red}`$ defined by an ideal $`J`$, $`I`$ an ideal such that $`V(I)=X`$, and $`I`$ gives the reduced structure on $`X`$ off the singular set of $`X`$. Let $`f`$ and $`g`$ be map germs whose components are a set of generators for $`J`$ and $`I`$. Let $`F:X,Y𝐂,0`$.
Theorem (1.7)Suppose $`X,0𝐂^{n+k},0`$ is the germ of an analytic space, such that $`X_{red}`$ is equidimensional, $`X`$, $`I`$ and $`J`$, $`Y`$, $`f`$, $`g`$ as in the set-up above then:
$$\frac{f}{y_i}\overline{JM_z(f)}\mathrm{iff}\frac{g}{y_i}\overline{JM_z(g)}.$$
Proof. We will see shortly that the implication $`\frac{f}{y_i}\overline{JM_z(f)}`$ implies $`\frac{g}{y_i}\overline{JM_z(g)}`$ is trivial, so we will concentrate on the other implication. We know that the inclusion $`\frac{g}{y_i}\overline{JM_z(g)}`$ holds with the reduced structure, and we will show the desired implication using the curve criterion. We need only use curves whose image (except for the origin) lies in the smooth part of $`X`$. This is because inclusion at the module level is equivalent to inclusion of the corresponding Fitting ideals, and to check this we only need the kind of curves we are considering.
Since $`IJ`$, we can form the quotient sheaf; this is supported on the singular set of $`X`$ by hypothesis, so $`IK^rJ`$ for some $`r`$ where $`K`$ defines the singular locus of X, by the Nullstellensatz.
Suppose the number of components of $`f`$ is $`p`$ while that of $`g`$ is $`p^{}`$. Then there is a $`p^{}\times p`$ matrix $`H_0`$ such that $`H_0(f)=(g)`$. Since $`IK^rJ`$, there exist matrices $`H_1`$, $`H_2`$ such that $`H_1(g)=H_2(f)`$, where the entries of $`H_2(f)`$ are $`q_{(j1)p+t}=k_j^rf_t`$, $`k_j`$ a set of generators of $`K`$, $`1tp`$. The matrix $`H_2`$, whose entries are $`h_{(j1)p+t,t}=k_j^r`$, $`1tp`$ has maximal rank at all smooth points of $`X`$.
If we differentiate each of our column vectors we obtain:
$$Dg=H_0Df$$
$$H_1Dg=H_2Df$$
working over the local ring of $`X_{red}`$. Now we pull back by a curve $`\varphi `$. Since the rest of the argument is independent of the dimension of $`Y`$ we assume dim $`Y=1`$ for notational simplicity.
Now the inclusion $`\frac{g}{y}\overline{JM_z(g)}`$ implies that $`\frac{g}{y}\varphi JM_z(g)\varphi `$. This is equivalent to the existence of $`(1,v(t))`$ such that
$$(Dg\varphi )(1,v)=0.$$
It’s now clear that $`\frac{f}{y}\overline{JM_z(f)}`$ implies $`\frac{g}{y}\overline{JM_z(g)}`$ as $`0=(Df\varphi )(1,v)`$ implies $`0=H_0\varphi (Df\varphi )(1,v)=(Dg\varphi )(1,v)=0`$.
Suppose $`(Dg\varphi )(1,v)=0`$. Then $`H_1\varphi (Dg\varphi )(1,v)=H_2\varphi (Df\varphi )(1,v)`$. Since $`H_2\varphi `$ is invertible for $`t0`$, it follows that $`0=(Df\varphi )(1,v)`$.
It is now easy to see how to improve our theorems relating integral closure and the various stratification conditions.
Theorem (1.8)Suppose $`X,0𝐂^{n+k},0`$ is the germ of an analytic space, such that $`X_{red}`$ is equidimensional, $`X`$, $`I`$ and $`J`$, $`Y`$, $`f`$, $`g`$, $`F`$ as in the set-up above then:
$$1)\frac{f}{y_i}\overline{m_YJM_z(f)}i\mathrm{iff}\frac{g}{y_i}\overline{m_YJM_z(g)}i\mathrm{iffthepair}(X_0,Y)\mathrm{satisfies}W.$$
$$2)\frac{(f,F)}{y_i}JM_z(f,F)^{}i\mathrm{iff}\frac{(g,F)}{y_i}JM_z(g,F)^{}i$$
$$\mathrm{iffthepair}(X_0,Y)\mathrm{satisfies}\text{A}\text{F}.$$
$$3)\frac{(f,F)}{y_i}\overline{m_YJM_z(f,F)}i\mathrm{iff}\frac{(g,F)}{y_i}\overline{m_YJM_z(g,F)}i$$
$$\mathrm{iffthepair}(X_0,Y)\mathrm{satisfies}\text{W}\text{F}.$$
Proof. Here is the argument for the first implication in 1).We know $`Dg=H_0Df`$, $`H_1Dg=H_2Df`$. By linearity we know $`z_i\frac{g}{z_j}=H_0z_i\frac{f}{z_j}`$, so
$$H_1H_0[\frac{f}{y},[z_i\frac{f}{z_j}]]=H_2[\frac{f}{y},[z_i\frac{f}{z_j}]],$$
and now the argument proceeds as in Theorem 6. The equivalence of $`\frac{f}{y_i}\overline{m_YJM_z(f)}`$ and W is Theorem 2.5 of p309.
For the A<sub>f</sub>case, define $`\widehat{H_0}`$ to be the matrix such that the lower right corner entry is $`1`$, the other entries in the last row and column 0, and the rest of the matrix is $`H_0`$. Then $`\widehat{H_0}(f,F)=(g,F)`$, and differentiating we get
$$\widehat{H_0}D(f,F)=D(g,F).$$
Extend $`H_2`$ to $`\widehat{H_2}`$, $`H_1`$ to $`\widehat{H_1}`$ as we extended $`H_0`$ to $`\widehat{H_0}`$. It is easy to see that
$$\widehat{H_1}D(g,F)=\widehat{H_2}D(f,F).$$
Now the argument goes as before. (Though we need strict dependence for A<sub>f</sub>, this is not a problem, because we can assume $`v`$ vanishes at the origin.) The W<sub>f</sub>argument combines the last two arguments. The equivalences with the stratification conditions comes from Lemma 5.1 p565 of in the A<sub>F</sub> case and from Proposition 2.1 p36 .
2. Generic plane sections and the Whitney conditions
In this section, we apply the results on nilpotents to the study of sections of $`(X_0,Y)`$.
Setup: Given $`X`$, $`Y`$ as in the setup before Theorem 7, assume the pair $`(X_0,Y)`$ satisfies condition $`W`$. Consider a sequence of hyperplanes $`H_1,H_2,\mathrm{},H_n`$, such that $`P_i=\underset{j=1}{\overset{i}{}}H_j`$ is a plane of codimension $`i`$ for all $`i`$, all $`H_iY`$, and $`H_i`$ defined by a linear form $`F_i`$. Note the set of hyperplanes in $`Y\times 𝐂^n`$ which contain $`Y`$ are parametrised by $`𝐏^{n1}`$.
We want to characterize those sequences for which $`(X_i=P_iX_0,Y)`$ satisfies $`W`$, and in which $`(X_i=P_iX_0,Y)`$ is generic in a certain sense, which we will develop.
To study the relation of the hyperplanes and the pair $`(X_0,Y)`$, we use a variant of the Grassman modification (). In $`𝐂^k\times 𝐂^n\times \widehat{𝐏}^{n1}`$ consider the incidence variety $`\stackrel{~}{𝐂^k\times 𝐂^n}=\{((y,z),H)|(y,z)H,H𝐏^{n1}\}`$. Denote the projection to $`𝐂^k\times 𝐂^n`$ by $`\beta `$, define $`\stackrel{~}{X}`$, to be $`\beta ^1(X)`$. We can assume that we will be working with hyperplanes with equations $`z_n=a_1z_1+\mathrm{}+a_{n1}z_{n1}`$. Then we can use $`(y,z_1,\mathrm{},z_{n1},a_1,\mathrm{}a_{n1})`$ as coordinates on $`\stackrel{~}{𝐂^k\times 𝐂^n}`$, and in these coordinates $`\beta (y,z,a)=(y,z,a_1z_1+\mathrm{}+a_{n1}z_{n1})`$. In these coordinates, we give $`\stackrel{~}{X}`$ the scheme structure defined by $`\beta ^{}(I)`$ where $`I`$ defines $`X`$. Since $`\beta `$ is a submersion off $`𝐂^k\times 0\times \widehat{𝐏}^{n1}`$, this structure will be reduced off of $`Y\times \widehat{𝐏}^{n1}`$, assuming $`I`$ gives the reduced structure off $`Y`$. (Otherwise it will be reduced off $`\beta ^1S`$, where $`S`$ is the set of points where $`I`$ fails to give the reduced structure.) It is known that $`(\stackrel{~}{X}_0,Y\times \widehat{𝐏}^{n1)}`$ satisfies W off some Z-open dense subset $`V`$ of $`Y\times \widehat{𝐏}^{n1}`$. Let $`U\widehat{𝐏}^{n1}=V0\times \widehat{𝐏}^{n1}`$. If $`H`$ is an element of $`U`$, it follows that $`(\stackrel{~}{X}_0,Y\times \widehat{𝐏}^{n1)}`$ satisfies W at $`(0,H)`$. We say such $`H`$ are W-generic hyperplanes for $`(X_0,Y)`$. A sequence of hyperplanes as in the setup is W-generic if each $`H_i`$ is W-generic for $`(X_{i1},Y)`$.
Since the curve criterion is so helpful in dealing with integral closure questions, it is helpful to know when a curve on $`X,0`$ lifts to $`\stackrel{~}{X},0,H`$, $`H`$ a hyperplane containing $`Y`$. Given a curve $`\varphi :𝐂,0X,0`$ the limiting $`Y`$-secant of $`\varphi `$ is the limit as $`t`$ tends to $`0`$ of the line determined by $`<\varphi _{k+1}(t),\mathrm{},\varphi _{k+n}(t)>`$.
Proposition (2.1)Suppose $`X,Y,0`$ is the germ of a pair of complex analytic sets, $`(X,Y)`$ as in the setup before theorem 1.6. Then $`\varphi :𝐂,0X,0`$, where the image of $`\varphi `$ does not lie in $`Y`$, lifts to $`\stackrel{~}{X},0,H`$, $`H`$ a hyperplane containing $`Y`$, if and only if $`Hl`$, where $`l`$ is the limiting $`Y`$-secant of $`\varphi `$.
Proof. Suppose the equation of $`H`$ is $`z_n=_{i=1}^{n1}a_iz_i`$. Such a $`\varphi `$ has an extension if and only if there exists $`a_i(t)`$, $`a_i(0)=a_i`$, such that $`\varphi _{k+n}(t)=_{i=1}^{n1}a_i(t)\varphi _{k+i}(t)`$. It is easy to see that this last equation holds iff
$$\underset{t0}{lim}(1/t^k)(\varphi _{k+n}(t)_ia_i\varphi _{k+i}(t))=0$$
where k is the minimum of the order of the first non-vanishing term in $`\{\varphi _{k+i}(t)\}`$.
We will give two characterizations of the W-generic hyperplanes for $`(X_0,Y)`$, one geometric, one algebraic. If $`H`$ is a hyperplane on $`𝐂^k\times 𝐂^n`$ denote the submodule of $`JM(f)`$ generated by the directional derivatives of $`f`$ in directions tangent to $`H`$ by $`JM(f)_H`$. If $`H`$ contains $`Y`$, then let $`JM_z(f)_H`$ denote the submodule of $`JM_z(f)`$ generated by directional derivatives of $`f`$ in directions tangent to $`H𝐂^n`$.
Theorem (2.2)In the setup of this section, $`H`$ is W-generic for $`(X_0,Y)`$ iff either (hence both) of the following conditions hold
1) $`H`$ is not a limiting tangent hyperplane to $`X`$ at the origin.
2) $`JM(f)_H`$ is a reduction of $`JM(f)`$ as $`𝒪_X`$ modules.
Proof. 1) and 2) are equivalent by . Further, if $`H`$ is not a limiting tangent hyperplane, then it is clear that $`(XH)_0=X_0H`$. For $`(XH)_0X_0H`$, while if there is a curve of singular points on $`X_0H`$, then $`H`$ is a tangent hyperplane at such points, so $`H`$ is a limiting tangent hyperplane.
The proof from here is similar to that of Theorem 2.9 of .
Denote $`f\beta `$ by $`G`$. A calculation using the chain rule shows that:
$$G/a_i=z_if/z_n\beta $$
$$G/y_i=f/y_i\beta $$
$$JM_z(G)=(f/z_j\beta +a_jf/z_n\beta )$$
By Theorem 1.8 1), $`H`$ is generic if and only if
$$G/a_i,G/y_i\overline{\beta ^{}(m_Y)JM_z(G)}.$$
Using the curve criterion, and Nakayama’s lemma, we see that $`z_if/z_n\beta \overline{(\beta ^{}(m_Y)JM_z(G)}`$, if and only if $`f/z_n\varphi \varphi ^{}JM_z(f)_H`$, for $`\varphi `$ any curve on $`X_0`$, whose limiting $`Y`$-secant is in $`H`$. (These are the curves on $`X_0,0`$ which lift to $`\stackrel{~}{X}_0,(0,H)`$ by 2.1)
Suppose $`H`$ is not a limiting tangent hyperplane. Then $`JM(f)_H`$ is a reduction of $`JM(f)`$. Since $`(X_0,Y)`$ satisfies W at the origin, we have that $`JM_y(f)JM(f)^{}`$
This implies that
$$\overline{JM_z(f)_H+JM(f)^{}}=\overline{JM(f)}$$
so, $`JM_z(f)_H`$ is a reduction of $`JM(f)`$ by 1.5. Then $`JM_y(f)\overline{m_YJM_z(f)_H}`$ and $`\overline{JM_z(f)_H}=\overline{JM_z(f)}`$ imply that $`z_if/z_n\beta \overline{(\beta ^{}(m_Y)JM_z(G)}`$ and $`G/y_i\overline{\beta ^{}(m_Y)JM_z(G)}`$, which shows $`H`$ is W-generic.
Suppose $`H`$ is W-generic. Then $`JM_y(f),f/z_n\varphi \varphi ^{}JM_z(f)_H`$ holds for all $`\varphi `$,whose limiting $`Y`$-secant is in $`H`$, again by the curve criterion and Nakayama’s lemma. This checks the condition of 2) for all curves whose limiting $`Y`$-secant is in $`H`$. In particular $`H`$ cannot be a limit of tangent hyperplanes along such a curve. Now suppose $`H`$ is a limiting tangent hyperplane along some curve $`\varphi `$; then since the pair $`(X_0,Y)`$ satisfy W and hence Whitney B, the limiting secant line of $`\varphi `$ must be in the limiting tangent plane, which gives a contradiction.
Corollary (2.3)Suppose $`H`$ is a W-generic hyperplane for $`(X_0,Y)`$, then $`(X_0H,Y)`$ satisfies condition W.
Proof. By the above proof $`JM_z(f)_H`$ is a reduction of $`JM(f)`$, this implies that $`JM_y(f)\overline{m_YJM_z(f)_H}`$. Let $`G=(f,H)`$; $`G`$ defines $`XH`$ with possibly non-reduced structure. Then the inclusion $`JM_y(f)\overline{m_YJM_z(f)_H}`$ continues to hold when we restrict to $`𝒪_{XH}^p`$. This implies the inclusion $`JM_y(G)\overline{JM_z(G}`$ since $`H`$ is independent of $`y`$. The result follows from 1.8 1).
Remark (2.4) The implication $`H`$ not a limit of tangent hyperplanes implies W-generic was proved by Lê and Teissier (), using the aureole and the theory of polar varieties. They also worked with the integral closures of Fitting ideals, as the theory of the integral closure of modules was not available then.
Corollary (2.5)Suppose $`(X_0,Y)`$ satisfy W, $`\{H_i\}`$ are a sequence of hyperplanes as in the set up for this section, then $`\{H_i\}`$ is a W-generic sequence if and only if each $`H_i`$ is not a limiting tangent plane at the origin for $`X_{i1}`$.
Proof. Apply Theorem 2 inductively to each $`X_{i1}`$.
Existing technology allows for a significant improvement if $`X`$ is a family of complete intersections with isolated singularities (ICIS).
Given an equidimensional complex analytic germ $`XC^k\times C^n,0`$ containing $`Y=C^k\times 0`$, suppose that the pair $`X_0,Y`$ satisfies condition WA at the origin. This means that every limit of tangent hyperplanes at the origin contains $`Y`$. There is always a bijection $`\iota `$ which takes hyperplanes containing $`Y`$ to hyperplanes through the origin in $`C^n`$, given by intersecting these hyperplanes with $`0\times C^n`$. We can ask does this bijection induce a 1-1 correspondence between the limiting tangent hyperplanes to $`X`$ at the origin, and the limiting tangent hyperplanes to $`X(0)𝐂^n`$ at the origin? An affirmative answer means we can reduce the study of the limiting tangent hyperplanes to $`X`$ to a study of the limiting tangent hyperplanes of the fibers of the family.
If $`X`$ is an ICIS, then there is a numerical criterion for $`H`$ not to be a limiting tangent hyperplane, namely that $`e(JM(f)_H)=e(JM(f))`$ (Cf. prop. 2.6 ).
Theorem (2.6)Suppose $`X`$ is a family of ICIS, $`(X_0,Y)`$ satisfies WA, and suppose $`e(JM(X(y)))`$ independent of $`Y`$. Then $`\iota `$ induces a 1-1 correspondence between the limiting tangent hyperplanes to $`X`$ at $`0`$ and the limiting tangent hyperplanes in $`𝐂^n`$ to $`X(0\times 𝐂^n)`$.
Proof. Suppose $`H`$ is not a limiting tangent hyperplane to $`X`$, $`HY`$. A careful examination of the proof of Theorem 2.2 shows that $`JM_z(f)_H`$ is still a reduction of $`JM(f)`$, since the WA condition implies that $`JM_y(f)JM(f)^{}`$. Restricting to $`0\times 𝐂^n`$, shows that $`\iota (H)`$ is not a limiting tangent hyperplane to $`X(0)`$, because $`JM_z(f)_H`$ remains a reduction of $`JM_z(f)`$.
Suppose $`H𝐂^n`$ is not a limiting tangent hyperplane to $`X(0)`$ at the origin. Then in $`𝒪_{X(0),0}^p`$, $`JM(f_0)_{\iota (H)}`$ is a reduction of $`JM(f_0)`$, hence the multiplicity of the two modules are the same. The multiplicity of the family of modules $`JM(f_y)_{\iota (H)}`$ must be constant, because $`e(M(f_y)_{\iota (H)},0)e(JM(f_y),0)`$, $`e(M(f_y)_{\iota (H)},0)`$ is upper semicontinuous and $`e(JM(f_y),0)`$ is constant by hypothesis. This means that $`JM(f_y)_{\iota (H)}`$ is a reduction of $`JM(f_y)`$ at the origin for all $`y`$, hence $`\iota (H)`$ is not a limiting tangent hyperplane to $`X(y)`$ at the origin.
Thus we have that the multiplicity of $`JM(f_y)_{\iota (H)}`$ at the origin is independent of $`y`$, and $`JM_z(f)`$ is fiberwise integrally dependent on $`JM_z(f)_H`$. The principle of specialization of integral dependence for modules (PSID) () then implies that $`JM_z(f)_H`$ is a reduction of $`JM_z(f)`$ in $`𝒪_{X,0}`$. Whitney A then implies $`JM_z(f)_H`$ is a reduction of $`JM(f)`$, hence $`H`$ is not a limiting tangent hyperplane. Since $`\iota `$ induces a 1-1 correspondence between hyperplanes which are not tangent hyperplanes, it then follows that it induces a 1-1 correspondence between limiting tangent hyperplanes, which finishes the proof.
The condition of the theorem that $`e(JM(X(y)))`$ independent of $`Y`$ has a geometric interpretation–it means that there is no relative polar variety of $`X`$ of dimension equal to $`Y`$.
With Theorem 2.6, we can improve 2.2, 2.3 and 2.5.
Corollary (2.7)Suppose $`X`$ is a family of ICIS, $`(XY,Y)`$ satisfies W, then the following are equivalent
1) $`\{H_i\}`$ is a W-generic sequence.
2)Each $`\iota (H_i)`$ is not a limiting tangent hyperplane at the origin of $`X_{i1}(0)`$.
3) $`e(JM(f_0),0)=e(JM(f)_{\iota (H)},0)`$.
Proof. If $`X`$ is a family of ICIS, with $`XY`$ smooth, $`XY,Y`$ Whitney, it follows that the multiplicity of $`JM(f_y)`$ is independent of $`y`$ at the origin. The proof then proceeds by induction on $`i`$. By Theorem 2.6, the equivalent conditions 2) and 3) imply that $`H_i`$ is not a limiting tangent hyperplane of $`X_{i1}`$. Theorem 2.2 implies $`H_i`$ is generic for $`X_{i1}`$, while $`H_i`$ is not a limiting tangent hyperplane of $`X_{i1}`$ also implies $`S(X_i)Y`$.
If we assume 1), then Theorem 2.2 implies that $`H`$ is not a limiting tangent hyperplane to $`X`$ at the origin, and 2.6 implies 2) and 3).
In there is a numerical criterion for $`H`$ not to be a limit of tangent hyperplanes which holds for equidimensional spaces. It is based on the author’s extension of the Buchsbaum-Rim multiplicity to modules of non-finite colength. Using this criterion and the generalization of the PSID in , it is reasonable to expect that the analogue of 2.7 holds in general.
References
G0 T. Gaffney, Aureoles and integral closure of modules, Stratifications, singularities and differential equations, II (Marseille, 1990; Honolulu, HI, 1990), 55–62, Travaux en Cours, 55, Hermann, Paris, 1997.
G1 T. Gaffney, Integral closure of modules and Whitney equisingularity, Invent. Math. 107 (1992), 301–22
G2 T. Gaffney Equisingularity of plane sections, $`t_1`$ condition and the integral closure of modules, Real and complex singularities (Sao Carlos, 1994), Pitman Res. Notes Math. Ser., Longman, Harlow, vol. 333, 1995, p 95–111
G3 T. Gaffney, Generalized Buchsbaum-Rim multiplcities and a theorem of Rees, Special issue in honor of Steven L. Kleiman. Comm. Algebra 31 (2003), no. 8, 3811–3827.
G4 T. Gaffney, The Multiplicity-Polar Formula and Equisingularity, in preparation
GK T. Gaffney and S. Kleiman. Specialization of integral dependence for modules. Inventiones math. 137, 541-574 1999
GK2 T. Gaffney and S. L. Kleiman, W<sub>f</sub>and integral dependence, Real and complex singularities (S o Carlos, 1998), 33–45, Chapman and Hall/CRC Res. Notes Math., 412, Chapman and Hall/CRC, Boca Raton, FL, 2000.
KT1 S. Kleiman and A. Thorup, A geometric theory of the Buchsbaum--Rim multiplicity, J. Algebra 167 (1994), 168–231
LT D.T. Lê and B. Teissier, Limites d’espaces tangents en géométrie analytique, Comm. Math. Helv., vol 63, 1988, p. 540–578 |
warning/0507/math0507179.html | ar5iv | text | # Contents
## 0 Introduction
### 0.1 Almost complex manifolds with non-degenerate Nijenhuis tensor
In geometry, two kinds of plane distributions often arise. There are integrable ones: complex structures, foliations, CR-structures. On the other hand, there are “maximally non-integrable” distributions, such as the contact structures, where the obstruction to integrability is nowhere degenerate. Looking at almost complex structures in dimension 3, one finds that the obstruction to the integrability, so-caled Nijenhuis tensor
$$N:\mathrm{\Lambda }^2T^{1,0}(M)T^{0,1}(M)$$
maps a 3-dimensional bundle to a 3-dimensional bundle. It is only natural to study the class of complex 3-manifolds such that $`N`$ is nowhere degenerate.
Given such a manifold $`M`$, it is possible to construct a nowhere degenerate, positive volume form $`detN^{}\overline{detN^{}}`$ on $`M`$ (for details, see (1.2)).
We study extrema of this volume form, showing that these extrema correspond to an interesting geometric structure (2.4).
In Hermitian geometry, one often encounters a special kind of almost complex Hermitian manifolds, called strictly nearly Kähler (NK-) manifolds, or Gray manifolds, after Alfred Gray (4.1). These manifolds can be characterized in terms of the $`G_2`$-structure on their Riemannian cone, or in terms of a special set of equations reminiscent of Calabi-Yau equations (Subsection 4.4).
We prove that a strictly nearly Kähler 3-manifold is uniquely determined by its almost complex structure (3). Moreover, such manifolds are extrema of the volume functional associated with the Nijenhuis tensor (2.4). This reminds of a construction of Hitchin’s functional on the space of all $`SL(3,)`$-structures on a manifold, having extrema on Calabi-Yau manifolds (\[Hi2\]).
### 0.2 Contents
This paper has the following structure.
$``$ In Section 1, we introduce the class of 3-manifolds with nowhere degenerate Nijenhuis tensor, and describe the basic structures associated with these manifolds. We give a sketch of a proof of existence of a Hermitian connection with totally antisymmetric torsion, due to Friedrich and Ivanov, and show that such a Hermitian metric is uniquely determined by the almost complex structure, if the Nijenhuis tensor is nowhere degenerate.
$``$ In Section 2, we introduce the nearly Kähler manifolds, giving several versions of their definition and listing examples.
$``$ In Section 3 we apply the results about connections with totally antisymmetric torsion to nearly Kähler geometry, showing that an almost complex structure determines the Hermitian structure on such a manifold uniquely, up to a constant multiplier.
$``$ In Section 4, we give several additional versions of a definition of a nearly Kähler manifold, obtaining an explicit description of a Nijenhuis tensor in terms of an orthonormal frame. We also interpret the nearly Kähler structure on a manifold in terms of $`G_2`$-geometry of its Riemannian cone. This is used to show that an NK-structure on a manifold $`M`$ is uniquely determined by its metric, unless $`M`$ is locally isometric to a 6-sphere (4.4).
$``$ In Section 5, we study infinitesimal variations of an almost complex structure. We prove that NK-manifolds are extrema of an intrinsic volume functional described earlier. A partial converse result is also obtained. Given an almost complex manifold $`M`$ with nowhere degenerate Nijenhuis tensor, admitting a Hermitian connection with totally antisymmetric torsion, $`M`$ is an extremum of the intrinsic volume functional if and only if $`M`$ is nearly Kähler.
## 1 Almost complex manifolds with non-degenerate Nijenhuis tensor
### 1.1 Nijenhuis tensor on 6-manifolds
Let $`(M,I)`$ be an almost complex manifold. The Nijenhuis tensor maps two $`(1,0)`$-vector fields to the $`(0,1)`$-part of their commutator. This map is $`C^{\mathrm{}}`$-linear, and vanishes, as the Newlander-Nirenberg theorem implies, precisely when $`I`$ is integrable. We write the Nijenhuis tensor as
$$N:\mathrm{\Lambda }^2T^{1,0}(M)T^{0,1}(M).$$
The dual map
$$N^{}:\mathrm{\Lambda }^{0,1}(M)\mathrm{\Lambda }^{2,0}(M)$$
(1.1)
is also called the Nijenhuis tensor. Cartan’s formula implies that $`N^{}`$ acts on $`\mathrm{\Lambda }^1(M)`$ as the $`(2,1)`$-part of the de Rham differential.
When one studies the distributions, one is usually interested in integrable ones (such as $`T^{1,0}(M)TM`$ for complex or CR-manifolds) or ones where the obstruction to integrability is nowhere degenerate (such as a contact distribution).
For the Nijenhuis tensor in complex dimension $`>3`$, non-degeneracy does not make much sense, because the space $`\mathrm{Hom}(\mathrm{\Lambda }^{0,1}(M),\mathrm{\Lambda }^{2,0}(M))`$ becomes quite complicated. However, for $`n=3`$, both sides of (1.1) are 3-dimensional, and we can define the non-degeneracy as follows.
Definition 1.1: Let $`(M,I)`$ be an almost complex manifold of real dimension 6, and $`N:\mathrm{\Lambda }^2T^{1,0}(M)T^{0,1}(M)`$ the Nijenhuis tensor. We say that $`N`$ is non-degenerate if $`N`$ is an isomorphism everywhere. Then $`(M,I)`$ is called an almost complex 6-manifold with nowhere degenerate Nijenhuis tensor.
Remark 1.2: Such manifolds were investigated by R. Bryant. His results were presented at a conference \[Br1\], but never published. The present author did not attend (unfortunately) and was not aware of his work.
The first thing one notices is that the determinant $`detN^{}`$ gives a section
$$detN^{}\mathrm{\Lambda }^{3,0}(M)^2\mathrm{\Lambda }^{3,0}(M)^{}.$$
Taking
$$detN^{}\overline{detN^{}}\mathrm{\Lambda }^{3,0}(M)\mathrm{\Lambda }^{0,3}(M)=\mathrm{\Lambda }^6(M)$$
(1.2)
we obtain a nowhere degenerate real volume form $`\mathrm{Vol}_I`$ on $`M`$. This form is called the canonical volume form associated with the Nijenhuis tensor. This gives a functional $`I\stackrel{\mathrm{\Psi }}{}_M\mathrm{Vol}_I`$ on the space of almost complex structures. One of the purposes of this paper is to investigate the critical points of the functional $`\mathrm{\Psi }`$, in the spirit of Hitchin’s work (\[Hi2\], \[Hi3\]).
### 1.2 Connections with totally antisymmetric torsion
Let $`(M,g)`$ be a Riemannian manifold, $`:TMTM\mathrm{\Lambda }^1M`$ a connection, and $`T\mathrm{\Lambda }^2MTM`$ its torsion. Identifying $`TM`$ and $`\mathrm{\Lambda }^1M`$ via $`g`$, we may consider $`T`$ as an element in $`\mathrm{\Lambda }^2M\mathrm{\Lambda }^1M`$, that is, a 3-form on $`TM`$. If $`T`$ is totally skew-symmetric as a 3-form on $`TM`$, we say that $``$ is a connection with totally skew-symmetric (or totally antisymmetric) torsion. If, in addition, $`M`$ is Hermitian, and $``$ preserves the Hermitian structure, we say that $``$ is a Hermitian connection with totally antisymmetric torsion.
Connections with totally skew-symmetric torsion are extremely useful in physics and differential geometry. An important example of such a connection is provided by a theorem of Bismut (\[Bi\]).
Theorem 1.3: Let $`(M,I)`$ be a complex manifold, and $`g`$ a Hermitian metric. Then $`M`$ admits a unique connection with totally skew-symmetric torsion preserving $`I`$ and $`g`$.
Connections with totally skew-symmetric torsion were studied at great length by Friedrich, Ivanov and others (see e.g. \[FI\], \[F\], \[AF\]). Bismut’s theorem requires the base manifold to be complex. Motivated by string theory, Friedrich and Ivanov generalized Bismut’s theorem to non-integrable almost complex manifolds (\[FI\]). For completeness, we sketch a proof of their theorem below.
Theorem 1.4: Let $`(M,I,\omega )`$ be an almost complex Hermitian manifold, and
$$N:\mathrm{\Lambda }^2T^{1,0}(M)T^{0,1}(M).$$
the Nijenhuis tensor. Consider the 3-linear form $`\rho :T^{1,0}(M)\times T^{1,0}(M)\times T^{1,0}(M)`$,
$$\rho (x,y,z):=\omega (N(x,y),z)$$
(1.3)
Then $`M`$ admits a connection $``$ with totally skew-symmetric torsion preserving $`(\omega ,I)`$ if and only if $`\rho `$ is skew-symmetric. Moreover, such a connection is unique.
Sketch of a proof: 1.2 is proven essentially in the same way as one proves Bismut’s theorem and existence and uniqueness of a Levi-Civita connection. Let $`(M,I,g)`$ be a Hermitian manifold, and $`_0`$ a Hermitian connection. Then all Hermitian connections can be obtained by taking $`(A):=_0+A`$, where $`A`$ is a 1-form with coefficients in the algebra $`𝔲(TM)`$ of all skew-Hermitian endomorpisms. The torsion $`T_A`$ of $`(A)`$ is written as
$$T_A=T_0+\mathrm{Alt}_{12}(A),$$
where $`T_0`$ is a torsion of $`_0`$, and $`\mathrm{Alt}_{12}`$ denotes the antisymmetrization of
$$\mathrm{\Lambda }^1(M)𝔲(TM)\mathrm{\Lambda }^1(M)\mathrm{\Lambda }^1(M)TM$$
over the first two indices. We identify $`𝔲(TM)`$ with $`\mathrm{\Lambda }^{1,1}(M)`$ in a standard way. Then 1.2 can be reinterpreted as a statement about linear-algebraic properties of the operator
$$\mathrm{Alt}_{12}:\mathrm{\Lambda }^1(M)\mathrm{\Lambda }^{1,1}(M)\left(\mathrm{\Lambda }^2(M)\mathrm{\Lambda }^1(M)\right)^{(2,1)+(1,2)}.$$
(1.4)
(the superscript $`(\mathrm{})^{(2,1)+(1,2)}`$ means taking $`(2,1)+(1,2)`$-part with respect to the Hodge decomposition), as follows.
By definition, the Nijenhuis tensor $`N`$ is a section of $`\mathrm{\Lambda }^{2,0}T^{0,1}`$. Identifying $`T^{0,1}`$ with $`\mathrm{\Lambda }^{1,0}`$ via $`g`$, we can consider $`N`$ as an element of $`\mathrm{\Lambda }^{2,0}\mathrm{\Lambda }^{1,0}`$. By Cartan’s formula, $`N`$ is equal to the $`(3,0)`$-part of the torsion. Therefore, existence of a connection with totally skew-symmetric torsion implies that (1.3) is skew-symmetric.
Conversely, assume that (1.3) is skew-symmetric. Since (1.4) maps $`\mathrm{\Lambda }^1(M)\mathrm{\Lambda }^{1,1}(M)`$ to $`(2,1)(1,2)`$-tensors, the $`(3,0)`$ and $`(0,3)`$-parts of torsion stay skew-symmetric if we modify the connection by adding $`A\mathrm{\Lambda }^1𝔲(TM)`$. Denote by $`T_1`$ the $`(2,1)(1,2)`$-part of the torsion $`T_0`$. To prove 1.2, we need to find $`A\mathrm{\Lambda }^1(M)\mathrm{\Lambda }^{1,1}(M)`$ such that $`T_1\mathrm{Alt}_{12}(A)`$ is totally skew-symmetric.
The map $`\mathrm{Alt}_{12}:\mathrm{\Lambda }^1(M)\mathrm{\Lambda }^2(M)\mathrm{\Lambda }^2(M)\mathrm{\Lambda }^1(M)`$ is an isomorphism, as a dimension count implies (this map has no kernel, which is easy to see). Therefore, (1.4) is injective. Using dimension count again, we find that cokernel of (1.4) projects isomorphically into
$$\mathrm{\Lambda }^{2,1}(M)\mathrm{\Lambda }^{1,2}(M)\mathrm{\Lambda }^2(M)\mathrm{\Lambda }^1(M).$$
Therefore, for any $`T_1`$ in $`(2,1)(1,2)`$-part of $`\mathrm{\Lambda }^2(M)\mathrm{\Lambda }^1(M)`$ there exists $`A\mathrm{\Lambda }^1(M)\mathrm{\Lambda }^{1,1}(M)`$ and $`B\mathrm{\Lambda }^{2,1}(M)\mathrm{\Lambda }^{1,2}(M)`$ such that $`T_1=\mathrm{Alt}_{12}(A)+B`$.
### 1.3 Connections with antisymmetric torsion on almost complex 6-manifolds
Let $`(M,I)`$ be an almost complex manifold, $`N`$ its Nijenhuis tensor. To obtain all Hermitian connections with totally skew-symmetric torsion on $`(M,I)`$, one needs to find all metrics $`g`$ for which the tensor $`\omega (N(x,y),z)`$ is skew-symmetric. As 1.2 implies, these metrics are precisely those for which such a connection exists.
We also prove the following proposition
Proposition 1.5: Let $`(M,I)`$ be an almost complex 6-manifold with Nijenhuis tensor which is non-degenerate in a dense subset of $`M`$, and $`g`$ a Hermitian metric admitting a connection with totally antisymmetric torsion. Then $`g`$ is uniquely determined by $`I`$, up to conformal equivalence. Moreover, the Riemannian metric $`g`$ determines $`I`$ uniquely, unless $`(M,g)`$ is locally isometric to a 6-sphere.
Proof: This is 3 and 4.4.
### 1.4 Correspondence with the results of R. Bryant
Since the first version of this paper was written, the previously unpublished results of R. Bryant appeared in a fundamental and important preprint \[Br2\]. There is a significant overlap with our research, although the presentation and terminology is different. The property (1.3) (which is equivalent to existence of Hermitian connection with totally antisymmetric curvature) is called “Nijenhuis tensor of real type” in \[Br2\]. The main focus of \[Br2\] is the so-called “quasi-integrable almost complex manifold”: manifolds with Nijenhuis tensor of real type, which is at every point of $`M`$ either non-degenerate (of constant signature) or zero. Examples of such structures are found. In particular, all twistor spaces of Kaehler surfaces with sign-definite holomorphic bisectional curvature are shown to be quasi-integrable. A variant of 2.4 is also proven. It is shown (\[Br2\], Proposition 8) that nearly Kaehler manifolds are critical points of the functional $`\mathrm{Vol}_I`$.
## 2 Nearly Kähler manifolds: an introduction
Nearly Kähler manifolds (also known as $`K`$-spaces or almost Tachibana spaces) were defined and studied by Alfred Gray (\[Gr1\], \[Gr2\], \[Gr3\], \[Gr4\]) in a general context of intrinsic torsion of $`U(n)`$-structures and weak holonomies. An almost complex Hermitian manifold $`(M,I)`$ is called nearly Kähler if $`_X(I)X=0`$, for any vector fields $`X`$ ($``$ denotes the Levi-Civita connection). In other words, the tensor $`\omega `$ must be totally skew-symmetric, for $`\omega `$ the Hermitian form on $`M`$. If $`_X(\omega )0`$ for any non-zero vector field $`X`$, $`M`$ is called strictly nearly Kähler.
In this section, we give an overview of known results and “folk theorems” of nearly Kähler geometry. Most of this theory was known (in a different context) since 1980-ies, when the study of Killing spinors was initiated (\[BFGK\]).
### 2.1 Splitting theorems for nearly Kähler manifolds
As V. F. Kirichenko proved, nearly Kähler manifolds admit a connection with totally antisymmetric, parallel torsion (\[K\]). This observation was used to prove a splitting theorem for nearly Kähler manifolds: any nearly Kähler manifold is locally a Riemannian product of a Kähler manifold and a strictly nearly Kähler one (\[Gr4\], \[N1\]).
A powerful classification theorem for Riemannian manifolds admitting an orthogonal connection with irreducible connection and parallel torsion was obtained by R. Cleyton and A. Swann in \[CS\]. Cleyton and Swann proved that any such manifold is either locally homogeneous, has vanishing torsion, or has weak holonomy $`G_2`$ (in dimension 7) or $`SU(3)`$ (in dimension 6).
Using Kirichenko theorem, this result can be used to obtain a classification of nearly Kähler manifolds. P.-A. Nagy (\[N2\]) has shown that that any strictly nearly Kähler manifold is locally a product of locally homogeneous manifolds, strictly nearly Kähler 6-manifolds, and twistor spaces of quaternionic Kähler manifolds of positive Ricci curvature, equipped with the Eells-Salamon metric.
These days the term “nearly Kähler” usually denotes strictly nearly Kähler 6-manifolds. In sequel we shall follow this usage, often omitting “strictly” and “6-dimensional”.
In dimension 6, a manifold is (strictly) nearly Kähler if and only if it admits a Killing spinor (\[Gru\]). Therefore, such a manifold is Einstein, with positive Einstein constant.
As one can easily show (see 4.1), strictly nearly Kähler 6-manifolds can be defined as 6-manifolds with structure group $`SU(3)`$ and fundamental forms $`\omega \mathrm{\Lambda }_{}^{1,1}(M)`$, $`\mathrm{\Omega }\mathrm{\Lambda }^{3,0}(M)`$, satisfying $`d\omega =3\lambda \mathrm{Re}\mathrm{\Omega }`$, $`d\mathrm{Im}\mathrm{\Omega }=2\lambda \omega ^2`$. An excellent introduction to nearly Kähler geometry is found in \[MNS\].
The most puzzling aspect of nearly Kähler geometry is a complete lack of non-homogeneous examples. With the exception of 4 homogeneous cases described below (Subsection 2.3), no other compact examples of strictly nearly Kähler 6-manifolds are known to exist.
### 2.2 Nearly Kähler manifolds in $`G_2`$-geometry and physics
Nearly Kähler manifolds have many uses in geometry and physics. Along with Calabi-Yau manifolds, nearly Kähler manifolds appear as target spaces for supersymmetric sigma-models, solving equations of type II string theory. These manifolds are the only 6-manifolds admitting a Killing spinor. This implies that a Riemannian cone $`C(M)`$ of a nearly Kähler manifold has a parallel spinor.
Let $`(M,g)`$ be a Riemannian manifold. Recall that the Riemannian cone of $`(M,g)`$ is a product $`M\times ^{>0}`$, with a metric $`gt^2\lambda dt^2`$, where $`t`$ is a unit parameter on $`^{>0}`$, and $`\lambda `$ a constant. It is well known that $`M`$ admits a real Killing spinor if and only if $`C(M)`$ admits a parallel spinor (for appropriate choice of $`\lambda `$). Then, $`C(M)`$ has restricted holonomy, for any nearly Kähler 6-manifold. It is easy to check that in fact $`C(M)`$ has holonomy $`G_2`$. This explains a tremendous importance that nearly Kähler manifolds play in $`G_2`$-geometry.
We give a brief introduction of $`G_2`$-geometry, following \[Hi2\] and \[J2\]. Let $`V^7`$ be a 7-dimensional real vector space. The group $`GL(7,)`$ acts on $`\mathrm{\Lambda }^3(V^7)`$ with two open orbits. For $`\nu `$ in one of these orbits, its stabilizer $`St(\nu )GL(7,)`$ is 14-dimensional, as a dimension count insures. It is easy to check that $`St(\nu )`$ is a real form of a Lie group $`G_2`$. For one of these orbits, $`St(\nu )`$ is a compact form of $`G_2`$, for another one it is non-compact. A 3-form $`\nu \mathrm{\Lambda }^3(V^7)`$ is called stable if its stabilizer is a compact form of $`G_2`$.
A 7-manifold $`X`$ equipped with a 3-form $`\rho `$ is called a $`G_2`$-manifold if $`\rho `$ is stable everywhere in $`X`$. In this case, the structure group of $`X`$ is reduced to $`G_2`$. Also, $`X`$ is equipped with a natural Riemannian structure:
$$x,y_X(\rho \mathrm{}x)(\rho \mathrm{}x)\rho (x,yTM).$$
(2.1)
A $`G_2`$-manifold is called parallel if $`\rho =0`$, where $``$ is the Levi-Civita connection associated with this Riemannian structure.
Isolated singularities of $`G_2`$-manifolds are of paramount importance in physics (\[AG\], \[AW\]). A simplest example of an isolated singular point is a conical singularity.
A metric space $`X`$ with marked points $`x_1,\mathrm{}x_n`$ is called a space with isolated singularities, if $`X\backslash \{x_1,\mathrm{}x_n\}`$ is a Riemannian manifold. Consider a space $`(X,x)`$ with a single singular point. The singularity $`xX`$ is called conical if $`X`$ is equipped with a flow acting on $`X`$ by homotheties and contracting $`X`$ to $`x`$. In this case, $`X\backslash x`$ is isomorphic to a Riemannian cone of a Riemannian manifold $`M`$.
It is easy to check that the cone $`C(M)`$ of a nearly Kähler manifold is equipped with a parallel $`G_2`$-structure, and, conversely, every conical singularity of a parallel $`G_2`$-manifold is obtained as $`C(M)`$, for some nearly Kähler manifold $`M`$ (\[Hi3\], \[IPP\]). For completeness’ sake, we give a sketch of a proof of this result in 4.3.
The idea of this correspondence is quite clear. Let $`X=C(M)`$ be a parallel $`G_2`$-manifold, and $`\omega _C`$ its 3-form. Unless $`X`$ is flat, we may assume that $`X`$ has holonomy which is equal to $`G_2`$ and not its proper subgroup. Indeed, if holonomy of $`X`$ is less than $`G_2`$, by Berger’s classification of irreducible holonomies $`X`$ is represented (as a Riemannian manifold) as a product of manifolds of smalled dimension. However, the singular point of the metric completion $`\overline{X}`$ is isolated, and this precludes such a decomposition, unless $`\overline{X}`$ is smooth. In the latter case, $`X`$ is flat.
Since holonomy of $`X`$ is (strictly) $`G_2`$, the 3-form can be reconstructed from the Riemannian structure uniquely. After rescaling, we may assume that the Riemannian structure structure on $`X=C(M)`$ is homogeneous of weight 2, with respect to the action of $`^{>0}`$ on $`C(M)`$. Then $`\omega _C`$ is homogeneous of weight 3. Homogeneous $`G_2`$-structures on $`C(M)`$ correspond naturally to $`SU(3)`$-structures on $`M`$. We write $`\omega _C`$ as $`t^2\pi ^{}\omega dt+t^3\pi ^{}\rho `$, where $`\rho `$, $`\omega `$ are forms on $`M`$, and $`\pi :C(M)M`$ is the standard projection. From a local coordinate expression of a $`G_2`$-form, we find that $`\omega `$ is a Hermitian form corresponding to an almost complex structure $`I`$, and $`\rho =\mathrm{Re}\mathrm{\Omega }`$ for a nowhere degenerate $`(3,0)`$-form $`\mathrm{\Omega }`$ on $`(M,I)`$.
The converse is proven by the same computation: given an $`SU(3)`$-manifold $`(M,I,\omega ,\mathrm{\Omega })`$, we write a 3-form
$$\omega _C:=t^2\pi ^{}\omega dt+t^3\pi ^{}\rho ,$$
(2.2)
on $`C(M)`$, and show that it is a $`G_2`$-structure, using a coordinate expression for a $`G_2`$-form.
As Fernandez and Gray proved in \[FG\], a $`G_2`$-manifold $`(X,\omega _C)`$ is parallel if and only if $`\omega _C`$ is harmonic. For the form (2.2), $`d\omega _C=0`$ is translated into
$$d\omega =3\rho .$$
(2.3)
Since $`\rho =I\rho `$ and $`\omega =\omega ^2`$, the condition $`d^{}\omega _C=0`$ becomes $`dI\rho =2\omega ^2`$. After an appropriate rescaling, we find that this is precisely the condition defining the nearly Kähler structure (4.1). Therefore, $`C(M)`$ is a $`G_2`$-manifold if and only if $`M`$ is nearly Kähler (see Subsection 4.3 for a more detailed argument).
The correspondence between conical singularities of $`G_2`$-manifolds and nearly Kähler geometry can be used further to study the locally conformally parallel $`G_2`$-manifolds (see also \[IPP\]). Locally conformally parallel $`G_2`$-manifold is a 7-manifold $`M`$ with a covering $`\stackrel{~}{M}`$ equipped with a parallel $`G_2`$-structure, with the deck transform acting on $`\stackrel{~}{M}`$ by homotheties. Since homotheties preserve the Levi-Civita connection $`\stackrel{~}{}`$ on $`\stackrel{~}{M}`$, $`\stackrel{~}{}`$ descends to a torsion-free connection on $`M`$, which is no longer orthogonal, but preserves the conformal class of a metric. Such a connection is called a Weyl connection, and a conformal manifold of dimension $`>2`$ equipped with a torsion-free connection preserving the conformal class a Weyl manifold. The Weyl manifolds are a subject of much study in conformal geometry (see e.g. \[DO\] and the reference therein).
The key theorem of Weyl geometry is proven by P. Gauduchon (\[Ga1\]). He has shown that any compact Weyl manifold is equipped with a privileged metric in its conformal class. This metric (called a Gauduchon metric now) is defined as follows.
Let $`(M,[g],)`$ be a compact Weyl manifold, where $`[g]`$ is a conformal class, and $`g[g]`$ any metric within this conformal class. Since $`[g]=0`$, we have $`(g)=g\theta `$, where $`\theta `$ is a 1-form, called a Lee form. A metric $`g`$ is called Gauduchon if $`\theta `$ satisfies $`d^{}\theta =0`$. A Gauduchon metric is unique (up to a complex multiplier).
Let now $`(M,,[g])`$ be a Weyl manifold with a Ricci-flat connection $``$. In \[Ga2\], Gauduchon has shown that the Lee form $`\theta `$ of the Gauduchon metric on $`M`$ is parallel with respect to the Levi-Civita connection associated with this metric.
Applying this argument to a compact locally conformally parallel $`G_2`$-manifold $`M`$, we obtain that the Lee form is parallel. From this one infers that the parallel $`G_2`$-covering $`\stackrel{~}{M}`$ of $`M`$ is a cone over some Riemannian manifold $`S`$ (see e.g. \[V\], Proposition 11.1, also see \[KO\] and \[GOP\]). Using the argument stated above, we find that this manifold is in fact nearly Kähler. Therefore, $`S`$ is Einstein, with positive Ricci curvature. Since $`M`$ is compact, $`S`$ is complete, and by Myers theorem, $`S`$ is actually compact (see \[V\], Remark 10.7). Now, the argument which proves Theorem 12.1 of \[V\] can be used to show that $`dimH^1(M,)=1`$, and $`M=C(S)/`$. This gives the following structure theorem, which is proven independently in \[IPP\].
Theorem 2.1: Let $`M`$ be a compact locally conformally parallel $`G_2`$-manifold. Then $`M=C(S)/`$, where $`S`$ is a nearly Kähler manifold, and the $``$-action on $`C(S)S\times ^{>0}`$ is generated by a map $`(x,t)(\phi (x),qt)`$, where $`|q|>1`$ is a real number, and $`\phi :SS`$ an automorphism of nearly Kähler structure.
### 2.3 Examples of nearly Kähler manifolds
Just as the conical singularities of parallel $`G_2`$-manifolds correspond to nearly Kähler manifolds, the conical singularities of $`Spin(7)`$-manifolds correspond to the so-called “nearly parallel” $`G_2`$-manifolds (see \[I\]). A $`G_2`$-manifold $`(M,\omega )`$ is called nearly parallel if $`d\omega =c\omega `$, where $`c`$ is some constant. The analogy between nearly Kähler 6-manifolds and nearly parallel $`G_2`$-manifolds is almost perfect. These manifolds admit a connection with totally antisymmetric torsion and have weak holonomy $`SU(3)`$ and $`G_2`$ respectively. N. Hitchin realized nearly Kähler 6-manifolds and nearly parallel $`G_2`$-manifolds as extrema of a certain functional, called Hitchin functional by physicists (see \[Hi3\]).
However, examples of nearly parallel $`G_2`$-manifolds are found in profusion (every 3-Sasakian manifold is nearly parallel $`G_2`$), and compact nearly Kähler manifolds are rare.
Only 4 compact examples are known (see the list below); all of them homogeneous. In \[Bu\] it was shown that any homogeneous nearly Kähler 6-manifold belongs to this list.
1. The 6-dimensional sphere $`S^6`$. Since the cone $`C(S^6)`$ is flat, $`S^6`$ is a nearly Kähler manifold, as shown in Subsection 2.2. The almost complex structure on $`S^6`$ is reconstructed from the octonion action, and the metric is standard.
2. $`S^3\times S^3`$, with the complex structure mapping $`\xi _i`$ to $`\xi _i^{}`$, $`\xi _i^{}`$ to $`\xi _i`$, where $`\xi _i`$, $`\xi _i^{}`$, $`i=1,2,3`$ is a basis of left invariant 1-forms on the first and the second component.
3. Given a self-dual Einstein Riemannian 4-manifold $`M`$ with positive Einstein constant, one defines its twistor space $`\mathrm{Tw}(M)`$ as a total space of a bundle of unit spheres in $`\mathrm{\Lambda }_{}^2(M)`$ of anti-self-dual 2-forms. Then $`\mathrm{Tw}(M)`$ has a natural Kähler-Einstein structure $`(I_+,g)`$, obtained by interpreting unit vectors in $`\mathrm{\Lambda }_{}^2(M)`$ as complex structure operators on $`TM`$. Changing the sign of $`I_+`$ on $`TM`$, we obtain an almost complex structure $`I_{}`$ which is also compatible with the metric $`g`$ (\[ES\]). A straightforward computation insures that $`(\mathrm{Tw}(M),I_{},g)`$ is nearly Kähler (\[M\]).
As N. Hitchin proved (\[Hi1\]), there are only two compact self-dual Einstein 4-manifolds: $`S^4`$ and $`P^2`$. The corresponding twistor spaces are $`P^3`$ and the flag space $`F(1,2)`$. The almost complex structure operator $`I_{}`$ induces a nearly Kähler structure on these two symmetric spaces.
### 2.4 Nearly Kähler manifolds are extrema of volume on almost complex manifolds with nowhere degenerate Nijenhuis tensor
Let $`(M,I,\omega )`$ be a nearly Kähler manifold, and $`N^{}:\mathrm{\Lambda }^{0,1}(M)\mathrm{\Lambda }^{2,0}(M)`$ the Nijenhuis tensor. By Cartan’s formula, $`N^{}`$ is the $`(2,1)`$-part of the de Rham differential (with respect to the Hodge decomposition). In 4.1 it is shown that $`d\omega `$ is a real part of a nowhere degenerate $`(3,0)`$-form $`\mathrm{\Omega }`$. Therefore, the 3-form
$$\omega (N(x,y),z)=d\omega (x,y,z)=\mathrm{Re}\mathrm{\Omega }(x,y,z)$$
is nowhere degenerate on $`T^{1,0}(M)`$. We obtain that the Nijenhuis tensor $`N`$ is nowhere degenerate.
The main result of this paper is the following theorem, which is analogous to \[Hi3\].
Theorem 2.2: Let $`(M,I)`$ be a compact almost complex 6-manifold with nowhere degenerate Nijenhuis tensor admitting a Hermitian connection with totally antisymmetric torsion. Consider the functional
$$I_M\mathrm{Vol}_I$$
(2.4)
on the space of such manifolds constructed in Subsection 1.1. Then (2.4) has a critical point at $`I`$ if and only if $`(M,I)`$ admits a nearly Kähler metric.
Proof: Follows from 5.2 and 4.1.
Remark 2.3: As follows from 3, the nearly Kähler metric on $`(M,I)`$ is uniquely determined by the almost complex structure.
## 3 Almost complex structures and connections <br>with totally antisymmetric torsion
Let $`(M,I)`$ be a 6-dimensional almost complex manifold, and
$$N^{}:\mathrm{\Lambda }^{0,1}(M)\mathrm{\Lambda }^{2,0}(M)$$
its Nijenhuis tensor. Given a point $`xM`$, the operator $`N^{}|_{_{\mathrm{\Lambda }_x^{0,1}(M)}}`$ can a priori take any value within $`\mathrm{Hom}(\mathrm{\Lambda }^{0,1}(M),\mathrm{\Lambda }^{2,0}(M))`$. For $`N^{}|_{_{\mathrm{\Lambda }_x^{0,1}(M)}}`$ generic, the stabilizer $`St(N_x^{})`$ of $`N_x^{}`$ within $`GL(T_xM)`$ is 2-dimensional. If we fix a complex parameter, the eigenspaces of $`N_x^{}`$ (taken in apporpriate sense) define a frame in $`TM`$. Thus, a geometry of a “very generic” 6-dimensional almost complex manifold is rather trivial.
However, for a $`N_x^{}`$ inside a 10-dimensional subspace
$$W_0\mathrm{Hom}(\mathrm{\Lambda }^{0,1}(M),\mathrm{\Lambda }^{2,0}(M)),$$
(3), the stabilizer $`St(N_x^{})`$ contains $`SU(3)`$, and the geometry of $`(M,I)`$ becomes more interesting.
Proposition 3.1: Let $`(M,I)`$ be an almost complex 6-manifold with Nijenhuis tensor which is non-degenerate in a dense set. Assume that $`(M,I)`$ admits a Hermitian structure $`\omega `$ and a Hermitian connection with totally antisymmetric torsion. Then $`\omega `$ is uniquely determined by $`I`$, up to conformal equivalence.
Proof: Consider the map
$$C:=\mathrm{Id}N^{}:\mathrm{\Lambda }^{1,1}(M)\mathrm{\Lambda }^{1,0}(M)\mathrm{\Lambda }^{2,0}(M)$$
(3.1)
obtained by acting with the Nijenhuis tensor $`N^{}:\mathrm{\Lambda }^{0,1}(M)\mathrm{\Lambda }^{2,0}(M)`$ on the second tensor multiplier of $`\mathrm{\Lambda }^{1,1}(M)\mathrm{\Lambda }^{1,0}(M)\mathrm{\Lambda }^{0,1}(M)`$. Then $`C`$ maps $`\omega `$ to a 3-form
$$x,y,z\omega (N(x,y),z).$$
As 1.2 implies, $`(M,I,\omega )`$ admits a Hermitian connection with totally antisymmetric torsion if and only if $`C(\omega )`$ lies inside a 1-dimensional space
$$\mathrm{\Lambda }^{3,0}(M)\mathrm{\Lambda }^{1,0}(M)\mathrm{\Lambda }^{2,0}(M).$$
However, $`C`$ is an isomorphism in a dense subset of $`M`$, hence, all $`\omega `$ which satisfy the conditions of 1.2 are proportional.
Remark 3.2: The same argument proves that an almost complex manifold admits a Hermitian connection with totally antisymmetric torsion if and only if $`C^1(\mathrm{\Lambda }^{3,0}(M))`$ contains a Hermitian form. This is the space $`W_0`$ alluded to in the beginning of this section.
3 leads to the following corollary.
Corollary 3.3: Let $`(M,I)`$ be an almost complex 6-manifold. Then $`(M,I)`$ admits at most one strictly nearly Kähler metric, up to a constant multiplier.
Proof: Let $`\omega _1`$ and $`\omega _2`$ be nearly Kähler metrics on $`(M,I)`$. Since $`(M,I,\omega _i)`$ is strictly nearly Kähler, the 3-form $`C(\omega _i)\mathrm{\Lambda }^{3,0}(M)`$ is nowhere degenerate (see (3.1)). Therefore, $`(M,I)`$ has nowhere degenerate Nijenhuis tensor. Then, by 3, $`\omega _i`$ are proportional: $`\omega _1=f\omega _2`$. However, $`d\omega _i^2=0`$ on any nearly Kähler 3-manifold (see e.g. 4.1 (ii)). Then $`2fdf\omega _2^2=0`$. This implies $`df=0`$, because the map $`\eta \eta \omega _2^2`$ is an isomorphism on $`\mathrm{\Lambda }^1(M)`$.
Remark 3.4: The converse is also true: unless $`(M,g)`$ is locally isometric to a 6-sphere, the Riemannian metric $`g`$ determines the nearly Kähler almost complex structure $`I`$ uniquely (4.4).
## 4 Nearly Kähler geometry and Hermitian connections with totally antisymmetric torsion
### 4.1 Hermitian structure on $`\mathrm{\Lambda }^{3,0}(M)`$ and nearly Kähler manifolds
Let $`(M,I)`$ be an almost complex 6-manifold, and $`\mathrm{\Omega }\mathrm{\Lambda }^{3,0}(M)`$ a non-degenerate $`(3,0)`$-form. Then $`\mathrm{\Omega }\overline{\mathrm{\Omega }}`$ is a positive volume form on $`M`$. This gives a $`\mathrm{Vol}(M)`$-valued Hermitian structure on $`\mathrm{\Lambda }^{3,0}(M)`$. If $`M`$ is in addition Hermitian, then $`M`$ is equipped with a natural volume form $`\mathrm{Vol}_h`$ associated with the metric, and the map
$$\mathrm{\Omega }\frac{\mathrm{\Omega }\overline{\mathrm{\Omega }}}{\mathrm{Vol}_h}$$
can be considered as a Hermitian metric on $`\mathrm{\Lambda }^{3,0}(M)`$. This metric agrees with the usual Riemann-Hodge pairing known from algebraic geometry, when $`I`$ is integrable. The following definition is a restatement of the classical one, see Subsection 2.
Definition 4.1: Let $`(M,I,\omega )`$ be an almost complex Hermitian manifold, and $``$ the Levi-Civita connection. Then $`(M,I,\omega )`$ is called nearly Kähler if the tensor $`\omega `$ is totally antisymmetric:
$$\omega \mathrm{\Lambda }^3(M).$$
The following theorem is a main result of this section.
Theorem 4.2: Let $`(M,I,\omega )`$ be an almost complex Hermitian 6-manifold equipped with a $`(3,0)`$-form $`\mathrm{\Omega }`$. Assume that $`\mathrm{\Omega }`$ satisfies $`3\lambda \mathrm{Re}\mathrm{\Omega }=d\omega `$, and $`|\mathrm{\Omega }|_\omega =1`$, where $`\lambda `$ is a constant, and $`||_\omega `$ is the Hermitian metric on $`\mathrm{\Lambda }^{3,0}(M)`$ constructed above. Then the following conditions are equivalent.
$`M`$ admits a Hermitian connection with totally antisymmetric torsion.
$`d\mathrm{\Omega }=2\sqrt{1}\lambda \omega ^2`$
$`(M,I,\omega )`$ is nearly Kähler, and $`d\omega =\omega `$.
The equivalence of (ii) and (iii) is known (see e.g. \[Hi3\], the second part of the proof of Theorem 6).
The existence of Hermitian connections with totally antisymmetric torsion on nearly Kähler manifolds is also well known (see Section 2). This connection is written as $`_{NK}=+T`$, where $``$ is the Levi-Civita connection on $`M`$, and $`T`$ the operator obtained from the 3-form $`3\lambda \mathrm{Im}\mathrm{\Omega }`$ by raising one of the indices. The torsion of $`_{NK}`$ is totally antisymmetric by construction (it is equal $`T`$). Also by construction, we find that $`T(\omega )=3\lambda \mathrm{Re}\mathrm{\Omega }`$, hence $`_{NK}(\omega )=0`$. Therefore, $`_{NK}`$ is a Hermitian connection with totally antisymmetric torsion. This takes care of the implication (iii) $``$ (i).
To prove 4.1, it remains to prove that (i) implies (ii); we do that in Subsection 4.2. For completeness’ sake, we sketch the proof of the implication (ii) $``$ (iii) in Subsection 4.4.
Remark 4.3: As 3 shows, a non-Kähler nearly Kähler metric on $`M`$ is uniquely determined by the almost complex structure $`I`$.
### 4.2 Connections with totally antisymmetric torsion and Nijenhuis tensor
Lemma 4.4: In assumptions of 4.1, (i) implies (ii).
Proof.
Step 1: We show that $`d\mathrm{\Omega }\mathrm{\Lambda }^{2,2}(M)`$.
Were $`(M,I)`$ integrable, the differential $`d`$ would have only (0,1)- and (1,0)-part with respect to the Hodge decomposition: $`d=d^{1,0}+d^{0,1}`$. For a general almost complex manifold, $`d`$ splits onto 4 parts:
$$d=d^{2,1}+d^{1,0}+d^{0,1}+d^{1,2}.$$
This follows immediately from the Leibniz rule. However,
$$0=d^2\omega =d(\mathrm{\Omega }+\overline{\mathrm{\Omega }})=d\mathrm{\Omega }+d\overline{\mathrm{\Omega }}.$$
(4.1)
Since $`\mathrm{\Lambda }^{p,q}(M)`$ vanishes for $`p`$ or $`q>3`$, we also have
$$d\mathrm{\Omega }+d\overline{\mathrm{\Omega }}=d^{0,1}\mathrm{\Omega }+d^{1,2}\mathrm{\Omega }+d^{2,1}\overline{\mathrm{\Omega }}+d^{1,0}\overline{\mathrm{\Omega }}$$
(4.2)
The four terms on the right hand side of (4.2) have Hodge types $`(3,1)`$, $`(2,2)`$, $`(2,2)`$ and $`(1,3)`$. Since their sum vanishes by (4.1), we obtain
$$d^{0,1}\mathrm{\Omega }=0,d^{1,0}\overline{\mathrm{\Omega }}=0,d^{2,1}\overline{\mathrm{\Omega }}=d^{1,2}\mathrm{\Omega }.$$
Then (4.2) gives
$$d\mathrm{\Omega }=d^{2,1}\overline{\mathrm{\Omega }}=d^{1,2}\mathrm{\Omega }.$$
(4.3)
Step 2:
$$d^{2,1}|{}_{_{\mathrm{\Lambda }^{1,1}(M)}}{}^{}=\mathrm{Id}N^{},$$
(4.4)
where $`N^{}:\mathrm{\Lambda }^{0,1}(M)\mathrm{\Lambda }^{2,0}(M)`$ is the Nijenhuis tensor,
$$\mathrm{Id}N^{}:\mathrm{\Lambda }^{1,1}(M)\mathrm{\Lambda }^{2,0}(M)\mathrm{\Lambda }^{1,0}(M)$$
acts as $`N^{}`$ on the second multiplier of $`\mathrm{\Lambda }^{1,1}(M)\mathrm{\Lambda }^{1,0}(M)\mathrm{\Lambda }^{0,1}(M)`$, and $``$ denotes the exterior product. (4.4) is immediately implied by the Cartan’s formula for the de Rham differential.
Step 3: From the existence of Hermitian connection with totally antisymmetric torsion we obtain that the form
$$\omega (N(x,y),z):T^{1,0}M\times T^{1,0}M\times T^{1,0}M$$
is totally antisymmetric (see 1.2). From (4.4) it follows that
$$\omega (N(x,y),z)=d\omega =3\lambda \mathrm{Re}\mathrm{\Omega }.$$
(4.5)
Consider an orthonormal frame $`dz_1,dz_2,dz_3`$ in $`\mathrm{\Lambda }^{1,0}(M)`$, satisfying $`\mathrm{\Omega }=dz_1dz_2dz_3`$ (such a frame exists because $`|\mathrm{\Omega }|_\omega =1`$). Then (4.5) gives
$$N^{}(d\overline{z}_i)=\lambda d\stackrel{ˇ}{z}_i,$$
(4.6)
where $`d\stackrel{ˇ}{z}_1=dz_2dz_3`$, $`d\stackrel{ˇ}{z}_2=dz_1dz_3`$, $`d\stackrel{ˇ}{z}_3=dz_1dz_2`$.
Step 4: Using Cartan’s formula as in Step 2, we express $`d^{1,2}\mathrm{\Omega }`$ through the Nijenhuis tensor. Then (4.3) can be used to write $`d\mathrm{\Omega }=d^{1,2}\mathrm{\Omega }`$ in terms of $`N^{}`$. Finally, (4.6) allows to write $`d^{1,2}\mathrm{\Omega }`$ in coordinates, obtaining $`d\mathrm{\Omega }=2\sqrt{1}\lambda \omega ^2`$.
### 4.3 $`G_2`$-structures on cones of Hermitian 6-manifolds
Proposition 4.5: Let $`(M,I,\omega )`$ be an almost complex Hermitian manifold, $`\mathrm{\Omega }\mathrm{\Lambda }^{3,0}(M)`$ a $`(3,0)`$-form which satisfies $`d\omega =3\lambda \mathrm{Re}\mathrm{\Omega }`$, for some real constant, and $`|\mathrm{\Omega }|_\omega =1`$. Assume, in addition, that $`d\mathrm{\Omega }=2\sqrt{1}\lambda \omega ^2`$. Consider the cone $`C(M)=M\times ^{>0}`$, equipped with a 3-form $`\rho =3t^2\omega dt+t^3d\omega `$, where $`t`$ is the unit parameter on the $`^{>0}`$-component. Then $`(C(M),\rho )`$ is a parallel, $`G_2`$-manifold (see Subsection 2.2). Moreover, any parallel $`G_2`$-structure $`\rho ^{}`$ on $`C(M)`$ is obtained this way, assuming that $`\rho ^{}`$ is homogeneous of weight 3 with respect to the the natural action of $`^{>0}`$ on $`C(M)`$.
Proof: As Fernandez and Gray has shown (\[FG\]), to show that a $`G_2`$-structure $`\rho `$ is parallel it suffices to prove that $`d\rho =d^{}\rho =0`$. Clearly, $`d\rho =0`$, because
$$d\rho =3t^2d\omega dt+3t^2dtd\omega =0.$$
On the other hand, $`(\omega dt)=\frac{1}{2}t^2\omega ^2`$, and $`d\omega =3dtI(d\omega )`$, where $``$ is taken with respect to the cone metric on $`C(M)`$. This is clear, because $`(\omega ,\mathrm{\Omega })`$ defines an $`SU(3)`$-structure on $`M`$, and $`d\omega =3\lambda \mathrm{Re}\mathrm{\Omega }`$. Then
$$\rho =\frac{3}{2}t^4\omega ^23t^3dtI(d\omega ).$$
(4.7)
Since $`d\mathrm{\Omega }=2\sqrt{1}\lambda \omega ^2`$ and $`3\lambda d\mathrm{Re}\mathrm{\Omega }=d^2\omega =0`$, we obtain $`d\mathrm{Im}\mathrm{\Omega }=2\lambda \omega ^2`$. This gives $`dI(d\omega )=2\omega ^2`$, because $`\lambda I(d\omega )=\mathrm{Im}\mathrm{\Omega }`$. Then (4.7) implies
$$d(\rho )=6t^3dt\omega ^2+3t^3dtdI(d\omega )=6t^3dt\omega ^26t^3dt\omega ^2=0.$$
We proved that $`C(M)`$ is a parallel $`G_2`$-manifold. The converse statement is straightforward.
In Subsection 2.2 it is shown that the holonomy of $`C(M)`$ is strictly $`G_2`$, unless it is flat (in the latter case, $`M`$ is locally isometric to a sphere). Therefore, 4.3 implies the following corollary.
Corollary 4.6: In assumptions of 4.3, the almost complex structure is uniquely determined by the metric, unless $`M`$ is locally isometric to a 6-sphere.
### 4.4 Near Kählerness obtained from $`G_2`$-geometry
Now we can conclude the proof of 4.1, implying 4.1 (iii) from 4.1 (ii). Let $`M`$ be a 6-manifold satisfying assumptions of 4.1 (ii). Consider the cone $`C(M)`$ equipped with a parallel $`G_2`$-structure $`\rho `$ as in 4.3. Let $`g_0`$ be a cone metric on $`C(M)`$. From the argument used to prove 4.3, it is clear that $`g_0`$ is a metric induced by the 3-form $`\rho `$ as in (2.1).
Consider the map $`C(M)\stackrel{\tau }{}M\times `$ induced by $`(m,t)(m,\mathrm{log}t)`$, and let $`g_1=\tau ^{}g_\pi `$ be induced by the product metric $`g_\pi `$ on $`M\times `$. Denote by $`_0`$, $`_1`$ the corresponding Levi-Civita connections. We know that $`_0(\rho )=0`$, and we need to show that
$$_1(\omega )=d\omega .$$
(4.8)
The metrics $`g_0`$, $`g_1`$ are proportional: $`g_1=g_0e^t`$. This allows one to relate the Levi-Civita connections $`_1`$ and $`_0`$ (see e.g. \[Or\]):
$$_1=_0+\frac{1}{2}A,$$
where $`A:TM\mathrm{End}(\mathrm{\Lambda }^1(M))`$ is an $`\mathrm{End}(\mathrm{\Lambda }^1(M))`$-valued 1-form mapping $`XTM`$ to
$$(\theta ,X)\mathrm{Id}X\theta +X^{\mathrm{}}\theta ^{\mathrm{}}$$
(4.9)
and $`\theta `$ the 1-form defined by $`_0(g_1)=g_1\theta `$, $`X\theta `$ the tensor product of $`X`$ and $`\theta `$ considered as an endomorphism of $`\mathrm{\Lambda }^1(M)`$, and $`X^{\mathrm{}}\theta ^{\mathrm{}}`$ the dual endomorphism.
From (4.9) and $`_0(\rho )=0`$ we obtain
$$(_1)_X(\rho )=(X,\theta )\rho (\rho \mathrm{}X)\theta +(\rho \mathrm{}\theta ^{\mathrm{}})X^{\mathrm{}}.$$
(4.10)
Since $`\theta =\frac{dt}{t}`$, we have $`_1(\theta )=0`$, and $`_1`$ preserves the decomposition $`\mathrm{\Lambda }^{}(C(M))\mathrm{\Lambda }^{}(M)dt\mathrm{\Lambda }^{}(M)`$. Restricting ourselves to the $`dt\mathrm{\Lambda }^{}(M)`$-summand of this decomposition and applying (4.10), we find
$$(_1)_X(t^3\omega \theta )=t^3(d\omega \mathrm{}X)\theta .$$
for any $`X`$ orthogonal to $`dt`$. Since $`g_1`$ is a product metric on $`C(M)M\times `$, this leads to $`\omega =d\omega `$, where $``$ is the Levi-Civita connection on $`M`$. This implies (4.8). We deduced 4.1 (iii) from 4.1 (ii). The proof of 4.1 is finished.
Using 4.3, we also obtain the following useful proposition.
Proposition 4.7: Let $`(M,I,g)`$ be a nearly Kähler manifold. Then the almost complex structure is uniquely determined by the Riemannian structure, unless $`M`$ is locally isometric to a 6-sphere.
## 5 Almost complex structures on 6-manifolds and their infinitesimal variations
### 5.1 Hitchin functional and the volume functional
Let $`(M,I)`$ be an almost complex 6-manifold with nowhere degenerate Nijenhuis tensor $`N`$, and $`\mathrm{Vol}_I=detN^{}\overline{detN^{}}`$ the corresponding volume form (see (1.2)). In this section we study the extrema of the functional $`I\stackrel{\mathrm{\Psi }}{}_M\mathrm{Vol}_I`$.
A similar functional was studied by N. Hitchin for 6- and 7-manifolds equipped with a stable 3-form (see \[Hi3\]). Since then, this functional acquired a pivotal role in string theory and M-theory, under the name “Hitchin functional”.
Our first step is to describe the variation of $`\mathrm{\Psi }`$. We denote by $`𝔐`$ the space of all almost complex structures with nowhere degenerate Nijenhuis tensor on $`M`$.
Let $`(M,I,\omega )`$ be an almost complex manifold with nowhere degenerate Nijenhuis tensor
$$N\mathrm{Hom}(\mathrm{\Lambda }^2T^{1,0}(M),T^{0,1}(M)),$$
$`\delta T_I𝔐`$ an infinitesimal variation of $`I`$, and
$$N_\delta \mathrm{Hom}(\mathrm{\Lambda }^2T^{1,0}(M),T^{0,1}(M))$$
the corresponding variation of the Nijenhuis tensor. Consider the form $`\rho :=\omega (N(x,y),z)`$ associated with the Hermitian structure on $`M`$ as in 1.2. After rescaling $`\omega `$, we assume that
$$|\rho |_\omega =1.$$
(5.1)
Since the Nijenhuis tensor is nowhere degenerate, $`\rho `$ is also nowhere degenerate. Therefore, $`\rho `$ can be used to identify $`T^{0,1}(M)`$ and $`\mathrm{\Lambda }^2T^{1,0}(M)`$, and we may consider $`N_\delta `$ as an endomorphism of $`\mathrm{\Lambda }^{0,1}(M)`$. Notice that this identification maps $`N`$ to the identity automorphism of $`\mathrm{\Lambda }^{0,1}(M)`$.
Claim 5.1: In these assumptions,
$$\frac{d\mathrm{\Psi }}{dI}(\delta )=2\mathrm{Re}_M\mathrm{Tr}N_\delta \mathrm{Vol}_I.$$
(5.2)
Proof: It is well known that
$$\frac{d(detA)}{d}t=detA\mathrm{Tr}\left(A^1\frac{dA}{dt}\right)$$
for any matrix $`A`$. Applying that to the map
$$N^{}\overline{N}^{}:\mathrm{\Lambda }^{1,0}(M)\mathrm{\Lambda }^{0,1}(M)\mathrm{\Lambda }^{0,2}(M)\mathrm{\Lambda }^{2,0}(M),$$
we obtain that
$$\frac{d(det(N^{}\overline{N}^{}))}{dI}(\delta )=\mathrm{Tr}\left(\frac{(N_\delta ^{}\overline{N}^{}+N^{}\overline{N}_\delta ^{})}{(N^{}\overline{N}^{})}\right)det(N^{}\overline{N}^{})$$
(5.3)
However, after we identify $`\mathrm{\Lambda }^{1,0}(M)`$ and $`\mathrm{\Lambda }^{0,2}(M)`$ as above, $`N`$ becomes an identity, and (5.3) gives
$$\frac{d(det(N^{}\overline{N}^{}))}{dI}(\delta )=2\mathrm{Re}\mathrm{Tr}N_\delta \mathrm{Vol}_I$$
(5.4)
Remark 5.2: We find that the extrema of the functional $`\mathrm{\Psi }(M,I)=_M\mathrm{Vol}_I`$ are precisely those almost complex structures for which $`\mathrm{Re}\mathrm{Tr}N_\delta =0`$ for any infinitesimal variation $`\delta `$ of $`I`$.
### 5.2 Variations of almost complex structures and the Nijenhuis tenor
It is convenient, following Kodaira and Spencer, to consider infinitesimal variantions of almost complex structures as tensors $`\delta \mathrm{\Lambda }^{0,1}(M)T^{1,0}(M)`$. Indeed, a complex structure on a vector space $`V`$, $`dim_{}V=2d`$, can be considered as a point of the Grassmanian of $`d`$-dimensional planes in $`V`$. The tangent space to a Grassmanian at a point $`WV`$ is given by $`\mathrm{Hom}(W,V/W)`$.
Consider the $`(0,1)`$-part $`^{0,1}`$ of the Levi-Civita connection
$$^{0,1}\delta \mathrm{\Lambda }^{0,1}(M)T^{1,0}(M)\mathrm{\Lambda }^{0,1}(M),$$
and let $`\overline{}:\mathrm{\Lambda }^{0,1}(M)T^{1,0}(M)\mathrm{\Lambda }^{0,2}(M)T^{1,0}(M)`$ denote the composiion of $`^{0,1}`$ with the exterior multiplication map
$$\mathrm{\Lambda }^{0,1}(M)T^{1,0}(M)\mathrm{\Lambda }^{0,1}(M)\mathrm{\Lambda }^{0,2}(M)T^{1,0}(M).$$
The following claim is well known.
Claim 5.3: Let $`(M,I)`$ be an almost complex manifold, and
$$\delta \mathrm{\Lambda }^{0,1}(M)T^{1,0}(M)$$
an infinitesimal variation of almost complex structure. Denote by $`N_\delta \mathrm{\Lambda }^{2,0}(M)T^{0,1}(M)`$ the corresponding infinitesimal variation of the Nijenhuis tensor (see Subection 5.1). Then $`\overline{N}_\delta =\overline{}\delta `$, where
$$\overline{}:\mathrm{\Lambda }^{0,1}(M)T^{1,0}(M)\mathrm{\Lambda }^{0,2}(M)T^{1,0}(M)$$
is the differential operator defined above.
Proof: The proof of 5.2 follows from a direct computation (see e.g. \[KS\]).
5.2 can be used to study the deformation properties of the functional $`I\stackrel{\mathrm{\Psi }}{}_M\mathrm{Vol}_I`$ constructed above (see Subsection 5.1). Indeed, from 5.1 it follows that $`\mathrm{\Psi }`$ has an extremum at $`I`$ if and only if $`\mathrm{Re}\mathrm{Tr}N_\delta =0`$ for any $`\delta \mathrm{\Lambda }^{0,1}(M)T^{1,0}(M)`$. Using the identification $`T^{1,0}(M)\mathrm{\Lambda }^{2,0}(M)`$, provided by the non-degenerate $`(3,0)`$-form as above, we can consider $`\delta `$ as a $`(2,1)`$-form on $`M`$. Then
$$\overline{}\delta \mathrm{\Lambda }^{0,2}(M)\mathrm{\Lambda }^{2,0}(M)=\mathrm{\Lambda }^{2,2}(M)$$
is the $`(2,2)`$-part of $`d\delta `$. Under these identifications, and using $`|\rho |_\omega =1`$ from (5.1), we can express $`\mathrm{Tr}\overline{N}_\delta `$ as
$$\mathrm{Tr}\overline{N}_\delta =\frac{\overline{}\delta \omega }{\mathrm{Vol}_I},$$
(5.5)
where $`\overline{}`$ is a $`(0,1)`$-part of the de Rham differential. This gives the following claim.
Claim 5.4: Let $`(M,I,\omega )`$ be an almost complex Hermitian 6-manifold with nowhere degenerate Nijenhuis tensor. Assume that the corresponding 3-form $`\rho `$ satisfies $`|\rho |_\omega =1`$ (see (5.1)). Consider the functional $`\mathrm{\Psi }(I)=_M\mathrm{Vol}_I`$ on the space of such almost complex structures. Then
$$\frac{d\mathrm{\Psi }}{dI}(\delta )=2\mathrm{Re}_M\overline{}\delta \omega ,$$
(5.6)
where $`\delta \mathrm{\Lambda }^{0,1}(M)T^{1,0}(M)`$ is an infinitesimal deformation of an almost complex structure $`I`$, considered as a $`(2,1)`$-form on $`M`$.
Proof: 5.2 is implied immediately by (5.5) and 5.1.
Comparing 5.2 with 5.1, we find the following
Corollary 5.5: In assumptions of 5.2, $`I`$ is an extremum of $`\mathrm{\Psi }`$ if and only if
$$\mathrm{Re}_M\overline{}\delta \omega =0$$
(5.7)
for any $`\delta \mathrm{\Lambda }^{2,1}(M)`$.
Integrating by parts, we find that (5.7) is equivalent to
$$\mathrm{Re}_M\delta \overline{}\omega =0$$
and to $`\overline{}\omega =0`$. This gives the following proposition
Proposition 5.6: Let $`(M,I,\omega )`$ be an almost complex Hermitian 6-manifold with nowhere degenerate Nijenhuis tensor. Consider the functional $`\mathrm{\Psi }(I)=_M\mathrm{Vol}_I`$ on the space of such almost complex structures on $`M`$. Then $`I`$ is an extremum of $`\mathrm{\Psi }`$ if and only if $`d\omega `$ lies in $`\mathrm{\Lambda }^{3,0}(M)\mathrm{\Lambda }^{0,3}(M)`$.
Now, 5.2 together with 4.1 implies 2.4. Notice that by 3, the nearly Kähler Hermitian structure on $`(M,I)`$ is (up to a constant multiplier) uniquely determined by $`I`$.
Acknowledgements: I am grateful to Robert Bryant, Nigel Hitchin, Paul-Andi Nagy and Uwe Semmelmann for valuable advice and consultations. P.-A. Nagy also suggested adding 4.4. Much gratitude to the referee for useful suggestions and the invaluable help of finding a multitude of minor errors.
Misha Verbitsky
University of Glasgow, Department of Mathematics,
15 University Gardens, Glasgow G12 8QW, Scotland.
Institute of Theoretical and Experimental Physics
B. Cheremushkinskaya, 25, Moscow, 117259, Russia
verbit@maths.gla.ac.uk, verbit@mccme.ru |
warning/0507/cond-mat0507002.html | ar5iv | text | # Itinerancy and Hidden Order in 𝑈𝑅𝑢₂𝑆𝑖₂
## I Introduction
The possibility of exotic particle-hole pairing leading to quadrupolar and orbital charge currents has been discussed extensively in the context of the two-dimensional Hubbard model.Halperin68 ; Affleck88 ; Kotliar88 ; Nersesyan89 ; Schulz89 More recently d-wave charge-density wave states, both orderedChakravarty01 and fluctuating,Lee04 have been proposed to explain the pseudogap phase in the underdoped cuprates and ground-states of doped two-leg Hubbard and $`tJ`$ ladders.Fjaerestad02 ; Wu04 In this paper we discuss related anisotropic particle-hole pairing in a different setting, namely that of three-dimensional Fermi liquids. We believe that such pairing may occur in the heavy fermion metal $`URu_2Si_2`$, and here we provide theoretical support for our earlier publications (with J.A. Mydosh) on this topic.Chandra02a ; Chandra02b ; Mydosh03 ; Chandra03 Though the initial motivation for our orbital antiferromagnetism (OAFM) proposal in $`URu_2Si_2`$ was primarily experimental, here we observe that coexistence of large electron-electron repulsion and antiferromagnetic fluctuations favours node formation in particle-hole pairing and hence the formation of anisotropic charge-density wave states. After presenting technical details behind specific predictions for neutron scattering and for NMR, we turn towards a microscopic description of orbital antiferromagnetism. We start by presenting the generalised Landau parameters associated with this anisotropic pairing. Next we study a toy model where this instability occurs. We end with a discussion of these results in the light of more recent measurements, and also suggest further experiments to test our ideas.
The heavy fermion metal $`URu_2Si_2`$ displays a classic second-order phase transition (see Figure 1) at $`T_0=17.5K`$, and yet the nature of the associated order parameter remains elusive nearly two decades after its discovery. This phase transition is characterised by a large entropy lossPalstra85 and sharp anomalies in the linearPalstra85 and the nonlinear susceptibilities,Miyako91 ; Ramirez92 the thermal expansion,deVisser86 and the resistivity,Palstra86 where standard mean-field relations between measured thermodynamic quantities are satisfied.Chandra94 At the transition, neutron scattering experiments observe gapped, propagating magnetic excitationsWalter86 ; Mason91 ; Broholm87 ; Broholm91 that suggest the formation of a spin density wave. However, subsequent neutron scattering measurementsBroholm87 ; Broholm91 indicate that the staggered magnetic moment ($`m_0=0.03\mu _B`$ per U atom), is too small to account for the entropy loss at the transition,Buyers96 which has been attributed to the development of an enigmatic hidden order.
There is strong experimental evidence that the antiferromagnetism and the hidden order in $`URu_2Si_2`$ are phase-separated and thus develop independently.Chandra03 High-field measurementsMentink96 ; vanDijk97 ; Bourdarot03 indicate that the bulk anomalies survive up to 40 Tesla ($`T`$), while the staggered moment is destroyedMason95 by comparatively modest fields of 15 $`T`$. Furthermore, the staggered magnetic moment grows linearly with pressureAmitsuka99 while bulk anomalies associated with the hidden order remain relatively pressure-independent.Fisher90 Phase separation is also indicated by muon spin resonance $`(\mu SR)`$ experiments.Luke94 ; Amitsuka03 The most direct evidence has come from recent NMR pressure-dependent measurements (see Figure 2):Matsuda01 for $`T<T_0`$ the existence of distinct antiferromagnetic and paramagnetic (hidden order) phases is clearly observed in samples with less than $`10\%`$ of the volume magnetic ($`m_{spin}0.3\mu _B`$) at ambient pressure. The observed increase of the staggered magnetic moment with pressureAmitsuka99 is then simply a volume-fraction effect.Matsuda01 The magnetic order develops independently from the hidden order through a first order transition,Chandra02a and the associated temperature-pressure phase diagram has been determined using thermal expansion measurements.Motoyama02
The mysterious phase transition at $`T_0`$ has features that have both local and itinerant electronic natures, and these coexisting dual characteristics make its description quite challenging. For example, the development of a sharp propagating mode just below $`T_0`$ observed by inelastic neutron scattering Walter86 ; Broholm87 ; Mason91 emphasises the importance of local crystal-field excitations at the transition. Nevertheless a purely local picture cannot provide a straightforward explanation for the observed elastic anomaliesLuthi93 near $`T_0`$ that are distinct from those of typical uniaxial antiferromagnetsMelcher70 both due to their (weak) magnitudes and due to the absence of precursor effects for $`T>T_0`$.
The sharp mean-field nature of the phase transition at $`T_0`$, together with the magnitude of the condensation entropy and the observed development of gap in the excitation spectrum all suggest the development of density-wave order within a fluid of itinerant quasiparticles.Ramirez92 ; Chandra94 ; Chandra02b ; Virosztek02 Itineracy is implicated by the sharpness of the transition while gap formation and the large entropy of condensation speak in favour of an order parameter at a finite wavevector. However, a dissenting view on this last point, involving p-wave ferromagnetism, has recently been proposed.Varma05 We note that within the itinerant perspective presented here, there are problems matching details of the excitation spectra as observed in inelastic neutron scattering experiments.Broholm91 On the other hand, a purely local scenarioBroholm91 ; Santini94 (with anticipated corrections for itinerant fermions) simply can not be reconciled with the almost complete quenching of the local moments, implicated by the paramagnetic (as opposed to Curie-like) susceptibility (see inset Fig. 1b.) and the large linear specific capacity, normally associated with well-formed heavy electrons (Fig. 1a). There are addition inconsistencies with a local picture: for example, the gap $`\mathrm{\Delta }`$ used in the local singlet schemeBroholm91 to explain the dispersing magnetic mode has a different field-dependence from that of the bulk $`\mathrm{\Delta }`$ associated with thermodynamic quantities.Santini00 A strict adherence to a local scheme requires consideration of many additional crystal-field levelsSantini00 evolving differently in an applied field.
A proper theoretical description of the transition at $`T_0`$ in $`URu_2Si_2`$ must therefore encompass both local and itinerant features of the problem. More specifically, the observed Fermi liquid properties for $`T>T_0`$ (e.g. Fig. 1) combined with the large entropy loss and the sharp nature of the transition indicate that the underlying quasiparticle excitations are itinerant, presumably composite objects formed from the 5f spin and orbital degrees of freedom of the $`U`$ ions. Local physics (e.g. Kondo physics, spin-orbit coupling, crystal-field schemes) plays a key role in their development.
We have just outlined a number of general considerations that we believe are crucial features of the hidden order in $`URu_2Si_2`$. Given these criteria, we (with J.A. Mydosh) have proposed that it can be described by a general density wave whose form factor is constrained by experimental observation and is ultimately determined by underlying local excitations.Mydosh03 We note that a number of proposals for the hidden order that fit into this general framework have been made.Chandra94 ; Virosztek02 ; Amitsuka02 ; Kiss04 ; Mineev04 ; Varma05 We argue that the large entropy loss at the transition can only be understood if the density-wave involves the polarisation of a significant fraction of the quasiparticle band, a condition that discounts a conventional spin-density wave due to the small size of the observed magnetic moment. Taking our cue from ambient-pressure Si NMR measurements (see Figure 3) that indicate broken time-reversal symmetry in the hidden ordered phase,Bernal01 we (with J.A. Mydosh) have proposed that $`UR_2Si_2`$ becomes an incommensurate orbital antiferromagnet at $`T=T_0`$ with charge currents circulating between the uranium ions.Chandra02b Here the modulation wavevector is chosen to fit the observed isotropic field distribution at the silicon sites. The resulting real-space fields can then be Fourier transformed to calculate a neutron scattering structure factor with a ring of possible q-vectors. Though these results have been presented elsewhere,Chandra02b in this paper (Section II) we provide supporting technical details and further discussion. We also determine the NMR linewidths at the Ru sites. Detailed comparison with recent experiment puts constraints on the allowed incommensurate wavevectors, allowing us to make more specific predictions for neutron scattering measurements.
In the second part of this paper, we turn towards an underlying microscopic picture of orbital antiferromagnetism. More specifically, in Section III we explore particle-hole pairings in anisotropic incompressible Fermi liquids with specific application to $`URu_2Si_2.`$ Next (Section IV) we introduce a simple $`tJ`$ model with a single heavy band and weak antiferromagnetic spin fluctuations (AFMSF). We note that this particular Hamiltonian was originally introducedMiyake86 to describe the AFMSF-mediated transition in $`URu_2Si_2`$ at $`1.2K`$. We show that this same toy model also supports particle-hole pairings associated with incommensurate orbital antiferromagnetism and quadrupolar charge-density wave formation. We end (Section IV) the paper with a summary, and then discuss our results in the context of recent high-field and thermal transport measurements.
## II Phenomenology and Experimental Predictions
In this Section we review the experimental motivation for incommensurate orbital antiferromagnetism as the hidden order in $`URu_2Si_2`$. We develop the phenomenology of this proposal, independent of microscopic details. The magnitude and the ordering wavevector of the orbital currents are fittedChandra02b to the observed isotropic field-distribution at the silicon sites as measured by nuclear magnetic resonance (NMR).Bernal01 The real-space fields produced by the orbital charge currents at all points in the sample volume are then determined, and we use this information to make specific predictions for neutron scattering structure factors and for NMR at non-silicon sites to test this proposal.
### II.1 Incommensurate Orbital Antiferromagnetism as the Hidden Order in $`URu_2Si_2`$
We begin our phenomenological discussion by reviewing the case for incommensurate orbital antiferromagnetism as the hidden order in $`URu_2Si_2`$. There have been many proposals for the primary order parameter in this material,Virosztek02 ; Amitsuka02 ; Kiss04 ; Mineev04 ; Varma05 ; Shah00 and until recently it was assumed that the spin antiferromagnetism and the hidden order are coupled and homogenous. However pressure-dependent NMR measurements,Matsuda01 supported by muon spin resonanceLuke94 ; Amitsuka03 and thermal expansionMotoyama02 data, indicate that the hidden and the magnetic orders are phase-separated and thus are completely independent.Chandra03
We believe that an important clue to the nature of the hidden order in $`URu_2Si_2`$ is provided by Si NMR measurements at ambient pressureBernal01 that indicate that at $`TT_0`$ the paramagnetic (non-split) silicon NMR line-width develops a field-independent, isotropic component whose temperature-dependent magnitude is proportional to that of the hidden order parameter. These results imply an isotropic field distribution at the silicon sites whose root-mean square value is proportional to the hidden order ($`\psi `$)
$$B^\alpha (i)B^\beta (j)=A^2\psi ^2\delta _{\alpha \beta },$$
(1)
and is $`10`$ Gauss at $`T=0`$. This field magnitude is too small to be explained by the observed momentBroholm87 that induces a field $`B_{spin}=\frac{8\pi }{3}\frac{M}{a^3}=100`$ Gauss where $`a`$ is the $`UU`$ bond length ($`a=4\times 10^8`$ cm). Furthermore this moment is aligned along the $`c`$-axis, and thus cannot account for the isotropic nature of the local field distribution detected by NMR. These measurements indicate that as the hidden order develops, a static isotropic magnetic field develops at each silicon site. This is strong evidence that the hidden order parameter breaks time-reversal invariance.
The magnetic fields at the silicon nuclei have two possible origins:Schlichter78 the conduction electron-spin interaction and the orbital shift that is due to current densities. In $`URu_2Si_2`$, the observed Knight shiftBernal01 indicates a strong Ising anisotropy of the conduction electron fluid along the $`c`$ axis; therefore the electron-spin coupling is unlikely to be responsible for the measured isotropic field distribution at the Si sites. It thus seems natural that these local fields are produced by orbital currents that develop at $`T_0`$, and thus we attribute the observed isotropic linewidth to the orbital shift.
It is this line of reasoning that led us (with J.A. Mydosh) to proposeChandra02b that $`URu_2Si_2`$ is an incommensurate orbital antiferromagnet at $`T=T_0`$ with charge currents circulating between the uranium ions. The planar tetragonal structure of $`URu_2Si_2`$ presents a natural setting for an anisotropic charge instability of this type. We can estimate the local fields at the silicon sites that are produced by the orbital currents. On dimensional grounds, the current along the uranium-uranium bond is given by $`I\frac{e\mathrm{\Delta }}{\mathrm{}}`$ where $`\mathrm{\Delta }`$ is the gap associated with the formation of the hidden order at $`T_0`$; we note that this expression also emerges from an analysis of the Hubbard model.Affleck88 If this orbital charge current is flowing around a uranium plaquette of side length $`a`$, then the magnetic field produced at a height $`a`$ above it is given by Ampere’s Law to be $`B\frac{2}{ac}\frac{e\mathrm{\Delta }}{\mathrm{}}=11G`$, in good agreement with the observed field strength;Bernal01 here we have used the experimental valuePalstra85 $`\mathrm{\Delta }=110K`$. Note that the resulting orbital moment, $`m_{OAFM}=0.02\mu _B`$ ($`m_{OAFM}=Ia^2`$), is comparable to the effective spin moment at ambient pressure. We emphasise that an orbital moment produces a field an order of magnitude less than that associated with a spin moment of the same value; the low field strengths observed at the silicon sites are quantitatively consistent with our proposal that they originate from charge currents.
This orbital moment, $`m_{OAFM}=0.02\mu _B`$, can also account for the entropy loss at the transition. We emphasise that its large value suggests that the amplitude of any proposed density-wave must be a significant fraction of its maximally allowed value, and will proceed to show that this is the case for the OAFM. In a metal the change in the entropy is given by $`\mathrm{\Delta }S=\mathrm{\Delta }\gamma _nT_0`$ where $`\mathrm{\Delta }\gamma _n`$ is the change in the linear specific heat coefficient resulting from the gapping of the Fermi surface. $`\mathrm{\Delta }\gamma _n`$ is inversely proportional to the Fermi energy $`ϵ_F`$ of the gapped Fermi surface, so that in general the change in entropy per unit cell is given by $`\mathrm{\Delta }𝒮\frac{\mathrm{\Delta }S}{k_B}(\frac{k_BT_0}{ϵ_F})`$. Since the transition at $`T_0`$ is mean-field in nature,Chandra94 we have $`\mathrm{\Delta }T_0`$ so that $`\mathrm{\Delta }𝒮\frac{\mathrm{\Delta }}{ϵ_F}`$. Now we recall that the orbital magnetic moment is
$$m_{OAFM}=Ia^2=\left(\frac{e}{\mathrm{}}\right)a^2\mathrm{\Delta }0.02\mu _B$$
(2)
such that it is saturated when $`\mathrm{\Delta }ϵ_F`$
$$m_{OAFM}^{}\left(\frac{e}{\mathrm{}}\right)a^2ϵ_F\left(\frac{a}{a_0}\right)^2\left(\frac{ϵ_F}{ϵ_H}\right)\mu _B0.1\mu _B$$
(3)
analogous to the saturation value of the electron spin $`\mu _B=\left(\frac{e}{\mathrm{}}\right)a_0^2ϵ_H`$ where $`a_0`$ and $`ϵ_H`$ are the Bohr radius and the energy of the Hydrogen atom respectively; here we have used $`\frac{a}{a_0}10^2`$ and $`\frac{ϵ_F}{ϵ_H}\frac{M_H}{M^{}}10^3`$ where $`M_H`$ and $`M^{}`$ refer to the mass of hydrogen and of $`URu_2Si_2`$ respectively. Then the change in entropy at the transition ($`\mathrm{\Delta }𝒮\frac{\mathrm{\Delta }}{ϵ_F}`$) due to the development of orbital antiferromagnetism can be expressed as
$$\mathrm{\Delta }𝒮_{OAFM}\left(\frac{m_{OAFM}}{m_{OAFM}^{}}\right)0.02\left(\frac{\mu _B}{m_{OAFM}^{}}\right)0.2$$
(4)
which is a number ($`0.2=0.3\mathrm{ln}2`$) in good agreement with experiment.Palstra85 We also note that the critical field for suppressing the thermodynamic anomalies is distinct from its spin counterpart: the ratio $`\frac{H_c^{orb}}{H_c^{spin}}\frac{\mu _b}{m_{OAFM}^{}}10`$ is qualitatively consistent with the observed critical field associated with the destruction of hidden order.Mentink96 ; vanDijk97 We emphasise that the sizable entropy loss associated with the development of orbital antiferromagnetism in $`URu_2Si_2`$ is a direct consequence of its renormalised electron mass ($`\frac{M^{}}{M}\frac{ϵ_H}{ϵ_F}`$). More generally the orbital moment is a larger fraction of its saturation value than is its spin counterpart, and this leads to the large entropy loss.
Orbital antiferromagnetism can therefore account for the local field magnitudes at the silicon ions and for the large entropy loss at the transition. Our next step is to tune the ordering wavevector to fit the isotropic distribution at these sites and then to determine the real-space fields throughout the sample volume. This can then be Fourier transformed to make predictions for neutron scattering.Chandra02b We note that it has been suggestedVarma05 that the isotropic nature of the field distributions at the silicon sites may be due to impurity-broadening. Though disorder is certainly present in these samples, we believe that the incommensurate nature of the density wave is the origin of this isotropy. Towards proving this point, we have determined the anisotropic field distributions at non-silicon sites; their observation via NMR would certainly not be possible if there were significant disorder-smearing.
Before proceeding with this program, let us comment briefly on the current experimental situation regarding the proposal of incommensurate orbital antiferromagnetism in $`URu_2Si_2`$. We admit that our proposal is closely linked to the ambient-pressure NMR experiments,Bernal01 which are the only direct evidence of broken time-reversal symmetry in the hidden ordered phase and have not been reproduced by other groups. We note that muon spin resonance measurementsLuke94 ; Amitsuka03 support the emergence of local fields with the same temperature-dependence as that associated with NMR, but their overall amplitudes are two orders of magnitude less than that seen in the NMR measurements. This is a point to which we return in the discussion. Although incommensurate peaks have been seen in inelastic neutron scattering measurements,Broholm91 ; Bull02 ; Wiebe04 , these are due to excitations above the partly gapped Fermi surface and are not directly related to the orbital antiferromagnetism. Current experimental resolution for elastic scattering - a direct probe of the incommensurate orbital antiferromagnetic order - is not yet good enough to confirm or deny the OAFM scenario. Here we present technical support for previous predictions for neutron structure factors,Chandra02b while also making specific suggestions for measurements where the signal should be sufficiently strong to be observed practically.
### II.2 Predictions for Neutron Scattering
In order to calculate the neutron cross section for scattering by incommensurate orbital antiferromagnetic order, we use the Born scattering formula,
$$\frac{d\sigma }{d\mathrm{\Omega }}=\left(\frac{g_Ne^2}{8\pi \mathrm{}c}\right)^2|𝐁(𝐪)|^2=r_0^2S(𝐪),$$
(5)
where $`g_N`$ is the neutron gyromagnetic ratio, $`𝐪`$ the scattering wavevector of the neutrons, $`|𝐁(𝐪)|^2`$ is the structure factor of the magnetic fields produced by the orbital currents and $`S(𝐪)=|𝐁(𝐪)|^2/(4\pi \mu _B)^2`$ is the structure factor measured in units of the Bohr magneton $`(\mu _B)`$.
We shall compute the magnetic field as the curl of the vector potential, $`𝐁(𝐱)=\times 𝐀`$. The procedure will be to compute the vector potential produced by the circulating current around a given plaquette. We shall denote the co-ordinate of the centre of plaquette j by $`𝐗_j`$. The corners of this plaquette are located at sites $`𝐱_j^{(r)}`$, ($`r=1,4`$) where
$$𝐱_j^{(r)}=𝐗_j+𝐱^{(r)},(r=1,4),$$
as shown in Fig. 4 (a). The circulating current around plaquette $`j`$ is then taken to be
$$I_C(𝐗_𝐣)=I_0e^{i𝐐𝐗_𝐣}+\mathrm{H}.\mathrm{c}$$
(6)
Using Ampere’s law, link 1-2 will will produce a contribution to the vector potential given by
$$𝐀^{12}(𝐱)=\frac{1}{c}\underset{j}{}_{𝐱_j^{(1)}}^{𝐱_j^{(2)}}𝑑x^{}\frac{I_C(𝐗_j)\widehat{𝐱}_{12}}{|𝐱𝐱_{}^{}{}_{j}{}^{}|},$$
(7)
where $`\widehat{𝐱}_{12}`$ is the unit vector pointing along the bond from $`1`$ to $`2`$. Writing $`𝐱_{}^{}{}_{j}{}^{}`$ as
$$𝐱_{}^{}{}_{j}{}^{}=𝐱_j^{(1)}+w(𝐱^{(2)}𝐱^{(1)}),$$
where $`0<w<1`$ defines the position along the link, we have
$$𝐀^{12}(𝐱)=\frac{a}{c}\underset{j}{}_0^1𝑑w\frac{I_C(𝐗_j)\widehat{𝐱}_{12}}{|𝐱\{𝐱_j^{(1)}+w(𝐱^{(2)}𝐱^{(1)})\}|}.$$
(8)
for the vector potential where $`a`$ is the U-U bond length in the $`ab`$ plane.
We now compute $`𝐁^{12}=\times 𝐀^{12}`$, and take the Fourier transform to obtain
$`𝐁^{12}(𝐪)`$ $`=`$ $`{\displaystyle \frac{a}{c}}{\displaystyle \underset{j}{}}{\displaystyle _0^1}𝑑w{\displaystyle d^3xe^{i𝐪.𝐱}I_C(𝐗_j)\widehat{𝐱}_{12}\times \frac{1}{|𝐱\{𝐗_j+𝐱^{(1)}+w(𝐱^{(2)}𝐱^{(1)})\}|}}`$ (9)
$`=`$ $`{\displaystyle \frac{ia}{c}}{\displaystyle \underset{j}{}}I_C(𝐗_j)\widehat{𝐱}_{12}\times 𝐪{\displaystyle _0^1}𝑑w{\displaystyle d^3xe^{i𝐪.𝐱}\frac{1}{|𝐱\{𝐱_j+𝐱^{(1)}+w(𝐱^{(2)}𝐱^{(1)})\}|}}.`$
Using
$$d^3e^{i𝐪𝐱}\frac{1}{|𝐱𝐚|}=\frac{4\pi }{q^2}e^{i𝐪𝐚},$$
we obtain
$`𝐁^{12}(𝐪)`$ $`=`$ $`{\displaystyle \frac{i4\pi a}{q^2c}}{\displaystyle \underset{j}{}}I_C(𝐗_j)\widehat{𝐱}_{12}\times 𝐪{\displaystyle _0^1}dw\mathrm{exp}[i𝐪.(𝐱_j+𝐱^{(1)}+w(𝐱^{(2)}𝐱^{(1)}))]`$ (10)
$`=`$ $`{\displaystyle \frac{4\pi a}{q^2c}}{\displaystyle \underset{j}{}}e^{i𝐪.𝐗_j}I_C(𝐗_j){\displaystyle \frac{\widehat{𝐱}_{12}\times 𝐪}{𝐪.(𝐱^{(2)}𝐱^{(1)})}}(e^{i𝐪.𝐱^{(2)}}e^{i𝐪.𝐱^{(1)}})`$
$``$ $`{\displaystyle \frac{4\pi I_0}{q^2c}}\left(𝐅^{12}(𝐪)\times 𝐪\right){\displaystyle \underset{j}{}}e^{i(𝐐𝐪).𝐗_j},`$ (11)
where we have replaced $`I_C(𝐗_𝐣)=I_0e^{i𝐐𝐗_j}`$ and
$$𝐅^{12}(𝐪)=\frac{\widehat{𝐱}_{12}}{𝐪\widehat{𝐱}_{12}}(e^{i𝐪𝐱^{(1)}}e^{i𝐪𝐱^{(2)}})$$
(12)
is the form-factor associated with link $`12`$ in the plaquette centred about $`𝐗_j`$. To sum over all of the links around the plaquette, we must add together the form factors
$`𝐅(𝐪)`$ $`=`$ $`𝐅^{12}(𝐪)+𝐅^{23}(𝐪)+𝐅^{34}(𝐪)+𝐅^{41}(𝐪)`$
$`=`$ $`[{\displaystyle \frac{\widehat{𝐱}}{𝐪.\widehat{𝐱}}}\{e^{i𝐪.(\widehat{𝐱}+\widehat{𝐲})a/2}e^{i𝐪.(\widehat{𝐱}\widehat{𝐲})a/2}`$
$`+e^{i𝐪.(\widehat{𝐱}+\widehat{𝐲})a/2}e^{i𝐪.(\widehat{𝐱}+\widehat{𝐲})a/2}\}\widehat{𝐱}\widehat{𝐲}]`$
$`=`$ $`4\mathrm{sin}\left({\displaystyle \frac{q_xa}{2}}\right)\mathrm{sin}\left({\displaystyle \frac{q_ya}{2}}\right)\left\{{\displaystyle \frac{\widehat{𝐲}}{𝐪.\widehat{𝐲}}}{\displaystyle \frac{\widehat{𝐱}}{𝐪.\widehat{𝐱}}}\right\}.`$ (14)
We let $`𝐐`$ be the wavevector for the orbital order so that $`I(𝐱_j)=I_0\mathrm{exp}[i𝐐.𝐱_j]`$.
Replacing $`𝐅^{12}𝐅`$ in Eq. 11, we we obtain the complete Fourier transform of the magnetic field:
$$𝐁(𝐪)=\underset{j}{}\mathrm{exp}[i(𝐐𝐪).𝐱_j]\mathrm{sin}\left(\frac{q_xa}{2}\right)\mathrm{sin}\left(\frac{q_ya}{2}\right)\{\frac{\widehat{𝐲}}{𝐪\widehat{𝐲}}\frac{\widehat{𝐱}}{𝐪.\widehat{𝐱}}\}\times 𝐪.$$
(15)
The U sites $`𝐱_j`$ can be written as
$$𝐱_j=a(j_1,j_2,0)+\frac{c}{2}(0,0,j_3)+\frac{1}{2}(1(1)^{j_3})(\frac{a}{2},\frac{a}{2},0),$$
(16)
where $`c`$ is the separation between even or odd numbered U planes. The unit cell has lattice vectors $`(a,0,0),(0,a,0),(0,0,c)`$. For an isotropic distribution of magnetic fields at the Si sites, we can reasonably expect $`𝐐`$ to be staggered between successive U layers. We permit $`𝐐`$ to be incommensurate in the $`ab`$ plane:
$$𝐐=(Q_x,Q_y,0)+\frac{2\pi }{c}(0,0,1).$$
(17)
Summing over the lattice sites in Eq.(15) we find
$`𝐁_{\mathrm{𝐎𝐀𝐅𝐌}}(𝐪)`$ $`=`$ $`{\displaystyle \frac{8\pi I_0}{q^2c}}{\displaystyle \underset{𝐆}{}}\delta _{𝐪,𝐐+𝐆}[1+e^{i𝐆(a/2,a/2,c/2)}]\times `$ (18)
$`\times `$ $`\mathrm{sin}\left({\displaystyle \frac{q_xa}{2}}\right)\mathrm{sin}\left({\displaystyle \frac{q_ya}{2}}\right)\left\{{\displaystyle \frac{\widehat{𝐱}}{𝐪.\widehat{𝐱}}}{\displaystyle \frac{\widehat{𝐲}}{𝐪.\widehat{𝐲}}}\right\}\times 𝐪,`$
where $`𝐆=2\pi [n_1/a,n_2/a,n_3/c]`$ is a reciprocal lattice vector.
Equation (18) should be contrasted with the corresponding expression if the order parameter were a spin density wave instead of an orbital antiferromagnet:
$`𝐁_{SDW}(𝐪)`$ $`=`$ $`{\displaystyle \frac{4\pi }{c}}{\displaystyle \underset{𝐆}{}}\delta _{𝐪,𝐐+𝐆}[1+e^{i𝐆(a/2,a/2,c/2)}]\times `$ (19)
$`\times \left\{\widehat{𝐪}\times (𝐌\times \widehat{𝐪})\right\}.`$
We note that a major difference between the two cases is that $`𝐁_{\mathrm{𝐎𝐀𝐅𝐌}}(𝐪)`$ decreases rapidly as $`q^2`$ while $`𝐁_{SDW}(𝐪)`$ is constant. This makes OAFM much harder to detect in neutron scattering experiments than its SDW counterpart. Second, the term $`(𝐪\times (𝐌\times 𝐪))`$ in $`𝐁_{OAFM}(q)`$ indicates that scattering is suppressed for $`𝐪=𝐐`$ since for an SDW along the $`c`$ axis, $`𝐌𝐐=\frac{2\pi }{c}(0,0,1)`$. here is no such term in $`𝐁_{OAFM}(q)`$. Thus the presence of a finite scattering amplitude at this particular wavevector in $`URu_2Si_2`$ would be a “smoking gun” confirmation of incommensurate orbital antiferromagnetism as the hidden order.
Next we turn to obtaining the structure factor $`|𝐁(𝐪)|^2`$. Neutrons couple to the orbital currents via their magnetic moment ($`𝝁_N=g_N\mu _B𝐒`$) as $`E=𝝁_N.𝐁`$. For incoherent neutrons, $`|𝐁(𝐪)|^2`$ is the modulus squared of $`𝐐`$ averaged over the orientation. the neutrons. Thus
$`S(𝐪)`$ $`=`$ $`{\displaystyle \frac{|𝐁(𝐪)|^2}{(4\pi \mu _B)^2}}=\left({\displaystyle \frac{NI_0a^2}{c\mu _B}}\right)^2{\displaystyle \underset{𝐆_{n_1,n_2,n_3}}{}}\delta _{𝐪,𝐐+𝐆}\left\{j_0\left[{\displaystyle \frac{q_xa}{2}}\right]j_0\left[{\displaystyle \frac{q_ya}{2}}\right]\right\}^2\times `$ (20)
$`\times \left[{\displaystyle \frac{1+\mathrm{cos}[\pi (n_1+n_2+n_3)]}{2}}\right]^2{\displaystyle \frac{q_x^2+q_y^2}{q_x^2+q_y^2+q_z^2}},`$
where $`j_0(x)=\frac{\mathrm{sin}x}{x}`$ and $`N`$ is the number of U sites. From this expression, we find that the maximum scattering intensity is predictedChandra02b to lie in a ring $`\stackrel{}{Q}=\stackrel{}{Q_0}+\stackrel{}{q}`$ of radius $`|q|0.2`$ centred on the wavevector $`\stackrel{}{Q_0}=(001)`$ where $`\stackrel{}{q}`$ lies in the $`ab`$ plane. Once again, we emphasise that scattering in the vicinity of $`\stackrel{}{Q_0}`$ is forbidden for the case of ordered spins along the $`c`$-axis; thus the observed presence of neutron scattering intensity at this particular wavevector would be a “smoking gun” confirmation of orbital antiferromagnetism as the hidden order.
In general the structure factor can be written as a product
$$S(q)=f(q)g(q)$$
(21)
where $`g(q)`$ is a function periodic in the reciprocal lattice vector but $`f(q)`$ is not. For the case of orbital antiferromagnetism, the calculated structure factor yields an asymptotic form for the form factor $`f(q)\frac{1}{q^4}`$. This power-law decay of the intensity peaks is due to the extended nature of the scattering source in contrast to the exponentially decaying structures observed for point-like spin antiferromagnetism.
It is tempting to state that such power-law peaks will be a clear signature of orbiting charge currents, but we still need to determine whether the overall intensities are observable. We can estimate the strength of the predicted OAFM neutron signal compared to that associated with spin magnetism at ambient pressure. Our calculations indicate that a fifth of the total integrated weight of $`S(q)`$ (TIWSQ) resides in the first Brillouin zone for the OAFM. Using the sum rule that relates the total ISWQ (integrated weight of $`S(q)`$ to the square of the moment, we have
$`(IWSQ)_{BZ1}`$ $`=`$ $`{\displaystyle \frac{1}{5}}(TIWSQ)_{OAFM}={\displaystyle \frac{1}{5}}(m_{OAFM})^2`$ (22)
$`=`$ $`{\displaystyle \frac{1}{500}}(m_{spin})^2={\displaystyle \frac{1}{500}}(TIWSQ)_{spin}`$ (23)
where we have used $`m_{OAFM}=0.2\mu _B`$ and $`m_{spin}=0.3\mu _B`$. Since the magnetic region occupies roughly a tenth of the sample at ambient pressure we then write
$$(IWSQ)_{BZ1}=\frac{1}{50}\mathrm{Measured}(TIWSQ)_{spin}$$
(24)
which indicates that the scattering peaks in the first Brillouin zone due to orbital ordering should have roughly 1/50 the intensity of the analogous spin peaks at ambient pressure. There have been two exploratory neutron studiesBull02 ; Wiebe04 but neither was conclusive due to issues of resolution. In particular the more recent elastic measurementsWiebe04 were not performed at the predicted wavevector $`\stackrel{}{Q_p}=(\tau _p,\tau _p,1)`$ where there should be no dipole scattering; please recall that here the form factor $`(\stackrel{}{q}\times \stackrel{}{m}`$) and $`\stackrel{}{m}`$ is aligned with the c-axis. More specifically the scattering intensity should be a factor of twenty higher than at $`\stackrel{}{Q_e}=(1+\tau _x,\tau _y,0)`$ where the experiments were performed, and the experimental resolution should be good enough then to prove/refute the orbital antiferromagnetism proposal.
### II.3 Nuclear Magnetic Resonance Linewidth at the Si and the Ru Sites
Nuclear Magnetic Resonance (NMR) is a local probe of the strength and the local distribution of the magnetic field distribution in the material. We use experimental NMR results to determine the ordering wavevector associated with the orbital antiferromagnetism, which can then be included in the structure factor calculated above. Thus neutron scattering and NMR are complementary. Eq.(8) gives the vector potential at a point $`𝐱`$ due to a current in link $`12`$ of a plaquette centred at $`𝐗_j`$. Contributions from other links in the plaquette may be similarly written out (see Fig.4). The magnetic field at any point $`𝐱`$ can be obtained using $`𝐁=\times 𝐀`$, where $`𝐀`$ is the total vector potential obtained by summing contributions from all links and plaquettes. We give detailed expressions for $`𝐀`$ in the appendix.
For the sake of completeness, we reviewChandra02a ; Chandra02b our arguments regarding the Si NMR measurementsBernal01 and the ordering wavevector of the orbital antiferromagnetism. We note that the silicon atoms in $`URu_2Si_2`$ are located at low-symmetry sites above and below the uranium plaquettes, so that the fields there do not cancel. Therefore the proposed OAFM must have an incommensurate $`𝐐(\pi ,\pi )`$ in order to produce isotropic field distributions at the silicon sites. If the order parameter in the hidden order phase is OAFM, then such a magnetic field distribution at the Si sites would be possible if the wavevector for orbital ordering were incommensurate,Chandra02a ; Chandra02b
$$𝐐=\frac{2\pi }{a}(0.22\mathrm{cos}\varphi ,0.22\mathrm{sin}\varphi ,a/c).$$
(25)
Fig.(5) shows the distribution of the magnetic field lines about the $`ab`$ plane for an incommensurate $`𝐐`$ corresponding to $`\varphi =\pi /4`$ in Eq.(25), and viewed in the direction.
A convenient definition of the anisotropy in the magnetic field at a given site is
$$\zeta =|(B_{}B_{})/(B_{}+B_{})|,$$
Fig.(6) shows the anisotropy as a function of the $`𝐐`$ vector. While the field distribution at the Si sites is isotropic, that need not be the case at other sites such as Ru; furthermore the anisotropic nature of the field distribution at the Ru sites would indicate that disorder-averaging is not at play here. If we take as the origin any uranium atom in the lattice, we find the Ru sites at coordinates
$$𝐗_{\text{Ru}}=\frac{a}{2}(ij+1,i+j,0)+\frac{c}{2}(0,0,k+1/2),$$
(26)
where $`i,j,k`$ are integers. Fig.7 shows the anisotropy of the magnetic field distribution at the Ru sites.
Recent Ru NMR measurementsBernal04 report a local magnetic field anisotropy of around 0.3. Values of $`𝐐`$ deduced from our OAFM model using the Ru NMR data should of course be consistent with Si NMR. The anisotropy of the magnetic field at the Ru sites calculated from our model shows strong variations as the orientation of the incommensurate wavevector given in Eq.(25) is varied. Anisotropy of field at the Ru sites for OAFM ordering wavevectors given by Eq.(25) varies from about 0.7 along the $`\varphi =0,\pi /2`$ directions to nearly unity along $`\varphi =\pi /4`$. Thus the most likely incommensurate wavevector $`𝐐`$ for OAFM lies close to the $`\varphi =0,\pi ,\pm \pi /2`$ directions.
Neutron scattering measurementsWiebe04 show enhanced scattering for $`T>T_0`$ at the incommensurate wavevectors
$$𝐐_{exp}=(2\pi /a)(n_1+0.4\mathrm{cos}\varphi ,n_2+0.4\mathrm{cos}\varphi ,n_3),$$
(27)
where $`n_1+n_2+n_3`$ is an odd integer. Below $`T_0`$, the ring of excitations seems to collapse toward the $`x`$ and $`y`$ directions, decreasing in intensity. The structure-factor predicted in Eq.(20) could not be verified/refuted due to issues of resolution.Wiebe04 According to Eq.(20), the structure factor measured near $`𝐐_{exp}=(2\pi /a)(1.4,0,0)`$, as was done in the most recent experimentWiebe04 has a scattering intensity that is smaller than that at $`𝐐=(2\pi /a)(0.4,0,a/c)`$ by a factor of more than five. In an earlier experiment,Broholm91 enhanced scattering was observed at $`𝐐=(2\pi /a)(1.4,0,0)`$ above the transition temperature $`T_0`$. The scattering intensity was sharply enhanced for $`T<T_0`$, and furthermore, the scattering linewidth decreased to resolution-limited values. More work is needed to verify whether the incommensurate peak observed in neutron scattering measurements is related to Eq.(25) deduced from Si and Ru NMR data using our model of orbital antiferromagnetism, and we strongly suggest elastic neutron scattering measurements at the wavevector predicted to have the greatest intensity ($`𝐐=(2\pi /a)(0.4,0,a/c)`$) to test OAFM as hidden order.
## III Towards a Microscopic Description of the Hidden Order
We now turn to a more microscopic approach to the hidden order. As we have already noted, a proper theoretical description of $`URu_2Si_2`$ must encompass both local and itinerant features of the problem. A general duality scheme for heavy electron systems has been proposed.Kuramoto90 In this model, the itinerant excitations are constructed from the low-lying crystal-field multiplets of the uranium atom. The quasiparticles associated with the heavy Fermi liquid are composite objects formed from the localised orbital and spin degrees of freedom of the U ions and the conduction electron fields. The phase transition in this model is then a Fermi-surface instability of these composite itinerant f-electrons. This approach has been adaptedOkuno98 to describe the coexistence of hidden order with a small moment in $`URu_2Si_2`$. With the more recent understanding that the hidden ordered phase does not contain a staggered magnetisation, we have revisited this duality schemeMydosh03 and, guided by experiment, now discuss its implications for the nature of the mysterious order that develops at $`T=T_0`$.
### III.1 Possible symmetries for particle-hole pairing
We begin with the assumption that all the excitations of $`URu_2Si_2`$ that condense into the hidden ordered state are of itinerant character. More specifically, we will assume that all of the system’s local physics (e.g. local moment character of the f-electrons) has been absorbed into the formation of composite quasiparticles. Given this premise, it then follows that key aspects of the (hidden) order parameter will be expressed through its matrix elements between quasiparticle states. If we denote it by the operator $`\widehat{\mathrm{\Psi }}`$, then its general matrix element between quasiparticle states is
$$𝐤+𝐐/\mathrm{𝟐},\sigma |\widehat{\mathrm{\Psi }}|𝐤𝐐/\mathrm{𝟐},\sigma ^{}=A_𝐤^{\sigma \sigma ^{}}(𝐐)$$
(28)
where $`𝐐`$ is the ordering wave-vector and $`|𝐤\sigma `$ is the quasiparticle state of momentum $`𝐤`$. Microscopically we would have to characterise $`\widehat{\mathrm{\Psi }}`$ in terms of the detailed crystal-field split states of the U ion, but for the purposes of characterising the phase transition, quasiparticle matrix elements should suffice. Within the Hilbert space of the mobile f-electrons, the order parameter can then be written
$$\widehat{\mathrm{\Psi }}A_𝐤^{\sigma \sigma ^{}}(𝐐)𝐜_{𝐤+𝐐/\mathrm{𝟐},\sigma }^{}𝐜_{𝐤𝐐/\mathrm{𝟐},\sigma ^{}}.$$
(29)
where $`A_𝐤^{\sigma \sigma ^{}}(𝐐)`$ is a general function of spin and momentum.
We are therefore considering a class of density waves with the most general pairing in the particle-hole channel characterised by $`A_𝐤^{\sigma \sigma ^{}}(𝐐)`$. We now categorise the possible particle-hole pairingsMydosh03 in $`URu_2Si_2`$. Assuming that the hidden order develops between the uranium atoms in each basal plane, we restrict our attention to nearest-neighbour pairings on a two-dimensional square lattice, and display the five resulting possibilities in Table 1 in Eq.(29). We emphasise that each of these pairing choices will partially gap the Fermi surface, accounting for the large entropy loss and the observed anomalies in several bulk quantities.Chandra02b In conventional charge- and spin-density waves (CDWs and SDWs respectively), the quantity $`A_𝐤(𝐐)`$ is an isotropic function of momentum. However in more general cases $`A_𝐤(𝐐)`$ will develop a nodal structure which leads to anisotropy (Table 1) that is favoured by strong Coulomb interaction, as we shall discuss in the next section.
### III.2 General Discussion of Anisotropic Charge Instabilities in Fermi Liquids
At low temperatures, heavy electron materials form almost incompressible Landau Fermi liquids in which the residual interactions between heavy quasiparticles are driven by strong, low-lying antiferromagnetic spin fluctuations. This harshly renormalised electronic environment is conducive to the development of instabilities in which electrons or holes form bound-states that contain nodes in their pair wavefunction.
Such arguments are well established in the context of anisotropic Cooper pairing.Miyake86 ; Emery87 Here we extend these ideas, arguing that an almost incompressible Fermi liquid is highly susceptible to the formation of anisotropic density waves, where the staggered electron-hole condensate contains a node in the pair wavefunction. This issue first arose in the context of orbital ordering in cuprate superconductorsNayak00 . Here it has been emphasised that strong Coulomb interactions suppress electron-hole bound-state formation in CDWs, unless the bound-state contains a node.Chakravarty01 Heavy electron fluids provide a unique opportunity to apply these arguments to three-dimensional systems. Furthermore there is no controversy associated with the Landau-Fermi liquid of their normal states, a situation in distinct contrast to the situation in the cuprates.
In a heavy electron fluid, the density of states is severely renormalised so that the ratio of the quasiparticle and the bare band-structure density of states
$$\frac{N^{}(0)}{N(0)}\frac{1}{Z}$$
is typically at least a factor of ten. In these systems the magnetic susceptibility, given in Landau Fermi liquid theory by
$$\chi =\frac{N^{(0)}}{1+F_o^a}$$
is weakly enhanced. By contrast, the charge susceptibility is severely depressed by strong coulomb interactions and is essentially given by the unrenormalised band-structure value
$$\chi _c=\frac{N^{(0)}}{1+F_o^s}N(0)$$
which is why the fluid is characterised as “almost incompressible”. It is this basic effect that rules out the formation of isotropic charge density wave order and s-wave superconductivity.
Response functions that contain an anisotropic form factor are unaffected by the strong Coulomb interactions. The key point here is that the strong interaction effects are local and thus they do not affect the higher Landau parameters, due to the nodes in the corresponding spherical harmonics. For example if we consider a “chemical potential” which couples anisotropically to the Fermi surface in the l-th angular momentum channel, then the corresponding susceptibility is given by
$$\chi _c^{(l)}\frac{N^{}(0)}{1+F_l^s}N^{}(0)$$
provided the higher Landau parameters are not much larger than unity. From this discussion, we see that large mass renormalisation and strong Coulomb repulsion suppresses isotropic CDW formation but that analogueous instabilities can form in higher angular momentum channels.
### III.3 Anisotropic Pairings: The Contenders
We have just argued that the large Coulomb repulsion between the heavy fermion quasiparticles (incompressibility) in $`URu_2Si_2`$ discourages isotropic pairing in the CDW channel. This expectation is confirmed by experiment, for charge density wave formation is expected to produce a lattice distortion, yet none is observed to develop URu<sub>2</sub>Si<sub>2</sub> below the 17K phase transition. Similarly neutron scattering is inconsistent with the presence of an isotropic spin density wave in the hidden ordered phase.Broholm87 ; Broholm91 Thus, mainly due to the incompressibility of the heavy Fermi liquid, we are left with three remaining anisotropic particle-hole pairing states (see Table 1).
The possibility of d-spin density waves as the hidden order in $`URu_2Si_2`$ has been raised by several authors.Ramirez92 ; Ikeda98 In a Stoner analysis, d-SDWs require ferromagnetic exchange interactions of neighbouring spins. In particular, for antiferromagnetic interactions, a Stoner analysis reveals that the d-SDW has a lower transition temperature than competing quadrupolar CDW (q-CDW) or spin-density waves.Kiselev99 Thus a d-SDW scenario favours ferromagnetic fluctuations in $`URu_2Si_2`$; by contrast, its transition at $`T^{}=1.2K`$ to a d-wave superconductor indicates the importance of antiferromagnetic fluctuations at $`T>T^{}`$.
Before discussing the two remaining options presented within the framework of Table 1, we want to mention two recent proposals for the hidden order parameter that both lead to quasiparticle matrix elements similar to those of a higher-order SDW. In the first one, the authorsMineev04 argue that consistency with experiment can be maintained for an SDW that develops predominantly in the p- or s- bands whose neutron form-factor at the Bragg peaks is significantly smaller than that of f-electrons. Here the key conceptual difficulty is that the matrix element of the order parameter in the f-bands would have to be small; yet the large entropy of condensation observed at $`T=T_0`$ is almost certainly associated with these same f-electrons. It has also been suggestedKiss04 the hidden order results from octupolar crystal-field states. In the quasiparticle basis, such an order parameter behaves like a spin-density wave with a small $`g`$ factor. At present, the viability of this approach awaits more detailed predictions regarding the magnetic distributions within the sample that then, like for the OAFM scenario, could be tested by NMR and neutron measurements.
Returning to the table of possible pairing symmetries (Table 1) we therefore have two remaining options: the quadrupolar charge density waveAmitsuka02 (Fig. 8 (a)) and the orbital antiferromagnet(Fig. 8 (b)).Chandra02a ; Chandra02b , where both scenarios are consistent with our picture of $`URu_2Si_2`$ as an incompressible Fermi liquid with strong antiferromagnetic fluctuations. Each order parameter has nodes, so that neither couples directly to the local charge density. Furthermore both incommensurate density waves couples weakly to uniform strain, and thus are both consistent with the observed insensitivityLuthi93 of the elastic response at $`T_0`$. Recent uniaxial stress measurements suggests that the hidden order is sensitive to the presence of local tetragonal symmetry,Yokoyama02 a feature that can be explained within both frameworks for completely different reasons. In the orbital antiferromagnet the currents are equal in each basal directionChandra02b , whereas within the quadrupolar scenario it is known that some of the crystal-field states with tetragonal symmetry are quadrupolar.Santini00 Unfortunately the diamagnetic response cannot be used to discriminate between these two scenario, as the contribution from orbital antiferromagnetism is small compared to that associated with the gapping of the Fermi surface ($`\frac{\chi _{Pauli}}{\chi _{diam}}100`$).
At present, the key factor distinguishing the orbital antiferromagnet from the quadrupolar charge-density wave scenarios is the absence or present of time-reversal breaking. Because the local field distributions and strengths measured by NMR have not yet been observed by other methods, there is still uncertainty about these results. We note that it has been arguedKiss04 that the observation of a stress-induced momentYokoyama02 implies that the hidden-order breaks time-reversal symmetry; much as we would like to believe this, we note that this result can be attributed to a volume-fraction effect and thus is inconclusive. Both the quadrupolar charge density wave and the orbital antiferromagnet have nodes in their respective gaps, which should in principle be observable via photoemission and/or scanning tunnelling microscopy, though issues associated with the nature of the surface of this material remain to be resolved. However the quadrupolar charge density wave is not expected to lead to magnetic neutron scattering, and therefore detailed elastic measurements are critical for resolving the nature of the hidden order parameter.
## IV Toy model for Anisotropic Particle-Hole Pairing
Next we explore a simple $`tJ`$ model for heavy electrons with antiferromagnetic spin fluctuations, and explore different orderings. We are motivated by experiment in our choice of the model. URu<sub>2</sub>Si<sub>2</sub> undergoes a phase transition to a d-wave superconducting state at $`T_0=0.8K`$, and the pairing is understood to be mediated by antiferromagnetic spin fluctuations. The same $`tJ`$ model also encompasses orbital antiferromagnetism, quadrupolar CDW, and isotropic SDW.
We consider a simplified model for the heavy Fermi liquid, described by $`H=H_0+H_I`$, where
$$H_0=\underset{𝐤}{}ϵ_𝐤c_{𝐤\sigma }^{}c_{𝐤\sigma }$$
describes the band of heavy electrons and
$$H_I=\underset{𝐪}{}J(𝐪)𝐒(𝐪).𝐒(𝐪),$$
(30)
is the interaction between them. Here, $`𝐒(𝐪)=\frac{1}{2}c_{𝐤+𝐪\alpha }^{}𝝈_{\alpha \beta }c_{𝐤\beta }`$ is the Fourier transform of the local spin operator. . In this simplified model, we consider the indices $`\sigma `$ to represent the pseudo-spin indices of the spin-orbit-coupled, heavy-electron band. We recall that we are working in an itinerant basis where the local physics (e.g. spin-orbit coupling) is absorbed into the composite quasiparticle states. Using the completeness relation $`\stackrel{}{\sigma }_{\alpha \beta }\stackrel{}{\sigma }_{\gamma \delta }+\delta _{\alpha \beta }\delta _{\gamma \eta }=2\delta _{\alpha \eta }\delta _{\gamma \beta }`$ we may rewrite this interaction as
$$H_I=\frac{1}{2}\underset{ij,\sigma \sigma ^{}}{}J_{ij}\left(c_{i\sigma }^{}c_{i\sigma ^{}}c_{j\sigma ^{}}^{}c_{j\sigma }\frac{1}{2}n_in_j\right).$$
Here we have rewritten the electron operators in a local basis, so that $`c_{j\sigma }=\frac{1}{\sqrt{N}}_𝐤c_{𝐤\sigma }^{}e^{i𝐤𝐱_j}`$ is the electron creation operator at site $`j`$, $`N`$ is the number of sites in the lattice and $`J_{ij}=\frac{1}{N}_𝐪J(𝐪)e^{i𝐪(𝐱_i𝐱_j}`$ is the spin interaction between sites $`i`$ and $`j`$. We shall ignore the second term, which involves the heavily suppressed fluctuations in quasiparticle occupation at each site. The first term can be decoupled as
$`H_I={\displaystyle \frac{1}{2N}}{\displaystyle \underset{𝐪,𝐤,𝐩,\sigma \sigma ^{}}{}}J(𝐤𝐩)c_{\sigma ,𝐤_+}^{}c_{\sigma ,𝐤_{}}c_{\sigma ^{},𝐩_{}}^{}c_{\sigma ^{},𝐩_+},`$ (31)
where $`𝐤_\pm =𝐤\pm \frac{1}{2}𝐪`$, etc. The interaction potential $`J(𝐪)`$ can be expanded into partial waves,
$$V_l=2_0^1𝑑xxP_l(12x^2)J(2px),$$
(32)
where $`x=\mathrm{sin}(\theta /2)`$ and $`pp_F`$. We require that the $`l=0`$ (isotropic) component be large and negative, reflecting strong on-site quasiparticle repulsion. This has the effect of suppressing isotropic particle-hole pairing. However this potential $`V_l`$, for $`l>0`$, could be attractive, which would then favour particle-hole pairing in higher angular momentum channels. Such higher angular momentum components are present due to anisotropy of the interaction $`J(𝐪)`$, which occurs at sufficiently large values of $`𝐪`$ where the underlying symmetry of the crystal becomes important.
For the purposes of a toy model, we shall assume in URu<sub>2</sub>Si<sub>2</sub>, nearest neighbour antiferromagnetic spin fluctuations (AFMSF) predominate, so that
$$J(𝐪)2J_1\gamma _𝐪^1,$$
(33)
where the form factor $`\gamma _𝐪^1=\mathrm{cos}(q_xa)+\mathrm{cos}(q_ya)`$. With this approximation, the interaction in Eq.(31) is separable:
$`H_I={\displaystyle \frac{J_1}{N}}{\displaystyle \underset{𝐪\frac{1}{2}\mathrm{BZ},k,p;\mathrm{\Gamma }=1,4}{}}(\gamma _𝐩^\mathrm{\Gamma }\rho _𝐩(𝐪))^{}\gamma _𝐤^\mathrm{\Gamma }\rho _𝐤(𝐪),`$ (34)
where
$$\rho _𝐤(𝐪)=\underset{\sigma }{}c_{𝐤+\frac{1}{2}𝐪\sigma }^{}c_{𝐤\frac{1}{2}𝐪\sigma }$$
(35)
are the particle-hole operators and
$`\gamma _𝐤^{1,2}`$ $`=`$ $`\mathrm{cos}(k_xa)\pm \mathrm{cos}(k_ya)`$ (36)
$`\gamma _𝐤^{3,4}`$ $`=`$ $`i(\mathrm{sin}(k_xa)\pm \mathrm{sin}(k_ya))`$ (37)
are form factors that transform under the point-group symmetry of the lattice. Since $`\rho _𝐤(𝐪)=\rho _𝐤^{}(𝐪)`$, the quantity inside the summation is symmetric under $`𝐪𝐪`$, and so, by doubling the prefactor and restricting the sum over $`𝐪`$ to one-half the Brillouin zone, we assure that every term in the $`𝐪`$ sum is independent.
This interaction is attractive and of equal magnitude in the four anisotropic channels. $`\gamma _𝐤^1`$, $`\gamma _𝐤^2`$, $`\gamma _𝐤^{3,4}`$ have s-like, d-like and p-like symmetry respectively. Notice that bond-variables $`_\sigma c_{i\sigma }^{}c_{j\sigma }`$ are invariant under time reversal, $`_\sigma c_{i\sigma }^{}c_{j\sigma }=_\sigma c_{j\sigma }^{}c_{i\sigma }^{}`$ and the imaginary pre-factors in $`\gamma _𝐤^{3,4}`$ have been chosen so that the form-factors respect this symmetry, i.e $`\gamma _𝐤^\mathrm{\Gamma }=(\gamma _𝐤^\mathrm{\Gamma })^{}`$.
By carrying out a “Hubbard Stratonovich” decoupling of $`H_I`$, we obtain
$`H_I`$ $``$ $`{\displaystyle \underset{𝐪\frac{1}{2}\mathrm{BZ},𝐤;\mathrm{\Gamma }=1,4}{}}\left[\mathrm{\Delta }_𝐪^\mathrm{\Gamma }\gamma _𝐤^\mathrm{\Gamma }\rho _𝐤(𝐪)+\overline{\mathrm{\Delta }}_𝐪^\mathrm{\Gamma }(\gamma _𝐤^\mathrm{\Gamma })^{}\rho _𝐤^{}(𝐪)\right]`$ (38)
$`+`$ $`{\displaystyle \frac{N}{2J_1}}{\displaystyle \underset{𝐪\frac{1}{2}\mathrm{BZ};\mathrm{\Gamma }=1,4}{}}\overline{\mathrm{\Delta }}_𝐪^\mathrm{\Gamma }\mathrm{\Delta }_𝐪^\mathrm{\Gamma }.`$
Now the mean-field solution to this expression is determined by the saddle-point condition
$$\mathrm{\Delta }_𝐪^\mathrm{\Gamma }=\frac{J_1}{N}\underset{𝐤}{}(\gamma _𝐤^\mathrm{\Gamma })^{}\rho _𝐤(𝐪)$$
(39)
In general, the density wave will condense at a primary wavevector $`𝐪=𝐐`$. For a realistic model, $`𝐐`$ may well be incommensurate, in which case, it will be accompanied by a family of corresponding $`𝐐^{}`$ that form a “star” of q-vectors under the point group. There will in general also be higher harmonics of $`𝐐`$. To illustrate the key ideas however, we shall assume a simple model in which a single $`𝐐`$ dominates the density wave, i.e.
$$\mathrm{\Delta }_𝐪^\mathrm{\Gamma }=\mathrm{\Delta }^\mathrm{\Gamma }\delta _{𝐪,𝐐}+\overline{\mathrm{\Delta }}^\mathrm{\Gamma }\delta _{𝐪,𝐐}$$
For this discussion, we shall also assume that the Fermi surface is “almost nested”, so that the Fermi surface can be divided into two equal parts or reduced Brillouin zones (RBZ): region I in which $`|ϵ_{𝐤\frac{1}{2}𝐐}||ϵ_{𝐤+\frac{1}{2}𝐐}|`$ and region II in which $`|ϵ_{𝐤\frac{1}{2}𝐐}||ϵ_{𝐤+\frac{1}{2}𝐐}|`$. In perfectly nested Fermi surfaces $`ϵ_𝐤=ϵ_{𝐤+𝐐}`$ are perfectly degenerate. For a square lattice and $`𝐐=(\pi ,\pi )`$ the reduced Brillouin zone is the diamond-shaped region bounded by $`\pi k_y\pi ,\pi +|k_y|k_x\pi |k_y|`$.
The mean-field Hamiltonian is then
$`H_{MFT}`$ $`=`$ $`{\displaystyle \underset{𝐤RBZ}{}}(c_{𝐤^+}^{},c_𝐤^{}^{})\left[\begin{array}{cc}ϵ_{𝐤^+}& \mathrm{\Delta }_𝐤\\ \overline{\mathrm{\Delta }}_𝐤& ϵ_𝐤^{}\end{array}\right]\left(\begin{array}{c}c_{𝐤^+}\\ c_𝐤^{}\end{array}\right)`$ (44)
$`+`$ $`N{\displaystyle \underset{\mathrm{\Gamma }=1,4}{}}{\displaystyle \frac{\overline{\mathrm{\Delta }}^\mathrm{\Gamma }\mathrm{\Delta }^\mathrm{\Gamma }}{J_1}}`$ (45)
where $`𝐤^\pm =𝐤\pm 𝐐/2`$ and $`\mathrm{\Delta }_𝐤=_\mathrm{\Gamma }\mathrm{\Delta }^\mathrm{\Gamma }\gamma _𝐤^\mathrm{\Gamma }`$. Here, $`𝐐`$ is the wave vector for particle-hole pairing. Diagonalising the electronic part of the total Hamiltonian yields two bands,
$`E_𝐤^{(\pm )}`$ $`=`$ $`{\displaystyle \frac{1}{2}}(ϵ_{𝐤+\frac{1}{2}𝐐}+ϵ_{𝐤\frac{1}{2}𝐐})`$ (46)
$`\pm \sqrt{{\displaystyle \frac{1}{4}}(ϵ_{𝐤+\frac{1}{2}𝐐}ϵ_{𝐤\frac{1}{2}𝐐})^2+|\mathrm{\Delta }_𝐤|^2},.`$
The mean field solution for pairing density $`\rho _𝐩(𝐐)`$ in Eq.(39) is obtained by setting the variation of the free energy $`F`$,
$`F`$ $`=`$ $`T{\displaystyle \underset{𝐤}{\overset{RBZ}{}}}\mathrm{ln}[(1+\mathrm{exp}(\beta E_𝐤^{(+)}))(1+\mathrm{exp}(\beta E_𝐤^{()}))]`$ (47)
$`+N{\displaystyle \underset{\mathrm{\Gamma }=1,4}{}}{\displaystyle \frac{\overline{\mathrm{\Delta }}^\mathrm{\Gamma }\mathrm{\Delta }^\mathrm{\Gamma }}{J_1}}`$
with respect to $`\overline{\mathrm{\Delta }}^\mathrm{\Gamma }`$ to zero. This yields the gap equation,
$`{\displaystyle \frac{\mathrm{\Delta }^\mathrm{\Gamma }}{J_1}}={\displaystyle \frac{1}{N}}{\displaystyle \underset{𝐤,\mathrm{\Gamma }}{\overset{RBZ}{}}}\mathrm{\Delta }_𝐤(\gamma _𝐤^\mathrm{\Gamma })^{}{\displaystyle \frac{f(E_𝐤^{()})f(E_𝐤^{(+)})}{E_𝐤^{(+)}E_𝐤^{()}}}.`$ (48)
In the special case where condensation occurs in a single channel $`\mathrm{\Gamma }=\mathrm{\Gamma }_0`$, this simplifies to
$`{\displaystyle \frac{1}{J_1}}={\displaystyle \frac{1}{N}}{\displaystyle \underset{𝐤}{\overset{RBZ}{}}}|\gamma _𝐤^{\mathrm{\Gamma }_0}|^2{\displaystyle \frac{f(E_𝐤^{()})f(E_𝐤^{(+)})}{E_𝐤^{(+)}E_𝐤^{()}}}.`$ (49)
At $`T=T_0`$, Equation (48) is essentially a Stoner criterion $`J_1\chi _{0\psi }(0)=1`$ where
$`\chi _{0\psi }(𝐪)={\displaystyle \frac{1}{N}}{\displaystyle \underset{𝐤}{\overset{RBZ}{}}}|\gamma _𝐤^{\mathrm{\Gamma }_0}|^2{\displaystyle \frac{f(ϵ_{𝐤^{}𝐪/2})f(ϵ_{𝐤^++𝐪/2})}{ϵ_{𝐤^++𝐪/2}ϵ_{𝐤^{}𝐪/2}}},`$ (50)
is the susceptibility associated with the hidden order parameter $`\psi `$, measured at a wave vector $`𝐪+𝐐`$.
Without details of the band-structure we can not predict which of the four order parameters will dominate. Some general comments are however in order. Although the pairing equation (48) does not involve any isotropic order parameter, the extended-s wave order parameter $`\gamma _𝐤^1`$ does have the same point-group symmetry as a pure s-wave and if it condenses, will tend to induce charge modulation. In a real heavy electron system, the effects of Coulomb interaction will renormalise the effective coupling constant for this channel, eliminating this order parameter from consideration. Of the remaining cases, $`\gamma _𝐤^2`$ corresponds to a q-CDW order and $`\gamma _𝐤^{3,4}`$ can be associated with spontaneous orbital or line currents between the Uranium atoms, as we shall now show.
Let us consider the current
$$j_{ij}=\frac{iet}{\mathrm{}}\underset{\sigma }{}(c_{j\sigma }^{}c_{i\sigma }c_{i\sigma }^{}c_{j\sigma })$$
(51)
from $`i`$ to $`j`$ along bond $`ij`$. Orbital order corresponds to a non-vanishing circulation of the current in a plaquette:
$$I_C=\frac{1}{4a}𝐣.d𝐥0$$
and, therefore, is of the form (Fig. 4 (a) ).
$`I_C(𝐗)={\displaystyle \frac{1}{4}}\left[j_{12}+j_{23}+j_{34}+j_{41}\right],`$ (52)
where $`𝐗)`$ is the position of the centre of the plaquette, and the indices $`(14)`$ label the corners of the plaquette, taking the sense of rotation to be anti-clockwise.
Now consider the evaluation of the bond variable $`_\sigma c_\sigma ^{}(𝐱+𝐚/2)c_\sigma (𝐱𝐚/2)`$. Taking the Fourier transform each electron field, we obtain
$`{\displaystyle \underset{\sigma }{}}c_\sigma ^{}(𝐱+𝐚/2)c_\sigma (𝐱𝐚/2)`$ (53)
$`=`$ $`{\displaystyle \frac{1}{N}}{\displaystyle \underset{𝐤,𝐤^{},\sigma }{}}c_{𝐤\sigma }^{}c_{𝐤^{}\sigma }e^{i[𝐤(𝐱+𝐚/2)𝐤^{}(𝐱𝐚/2)]}`$
$`=`$ $`{\displaystyle \frac{1}{N}}{\displaystyle \underset{𝐤,𝐤^{}}{}}\rho _{(𝐤+𝐤^{})/2}(𝐤𝐤^{})e^{i[(𝐤𝐤^{})𝐱+\stackrel{𝐩}{\stackrel{}{\frac{1}{2}(𝐤+𝐤^{})}}𝐚]}`$ (54)
$`=`$ $`e^{i𝐐𝐱}{\displaystyle \frac{1}{N}}{\displaystyle \underset{𝐩}{}}\rho _𝐩(𝐐)e^{i𝐩𝐚}+(𝐐𝐐)`$ (55)
where we assumed $`\rho _𝐤(𝐪)=\delta _{𝐤,𝐐}\rho _𝐤(𝐐)+(𝐐𝐐)`$. The current along a given bond is therefore
$`j(𝐱+𝐚/2,𝐱𝐚/2)`$ (56)
$`=`$ $`e^{i𝐐𝐱}{\displaystyle \frac{et}{N\mathrm{}}}{\displaystyle \underset{𝐤}{}}\rho _𝐤(𝐐)2\mathrm{sin}(𝐤𝐚)+(\mathrm{H}.\mathrm{c})`$ (57)
Averaging the currents anti-clockwise around a plaquette centred at $`𝐗`$, we arrive at
$$I_C(𝐗)=I_C\mathrm{exp}[i𝐐.𝐗]+\mathrm{H}.\mathrm{c}$$
(58)
where
$`I_C(𝐗)`$ $`=`$ $`i{\displaystyle \frac{et}{N\mathrm{}}}{\displaystyle \underset{𝐤}{}}\rho _𝐤(𝐐)[s_x\mathrm{sin}(Q_ya/2)s_y\mathrm{sin}(Q_xa/2)`$ (59)
$`=`$ $`{\displaystyle \frac{et}{N\mathrm{}}}{\displaystyle \underset{𝐤}{}}\rho _𝐤(𝐐)[\alpha _+\gamma _𝐤^4+\alpha _{}\gamma _𝐤^3],`$ (60)
where we have used the notation $`s_{x,y}\mathrm{sin}(k_{x,y}a)`$ and
$$\alpha _\pm =\frac{1}{2}\left[\mathrm{sin}(Q_ya/2)\pm \mathrm{sin}(Q_xa/2)\right]$$
(notice the ordering of the y and x terms). The form factor for orbital current order is thus a weighted mixture of $`\gamma _𝐤^4`$ and $`\gamma _𝐤^3`$. Using Eq.(39) to simplify Eq. (60), we obtain a relation between the orbital current and gap,
$$I=\frac{e\mathrm{\Delta }_C}{\mathrm{}}.\frac{t}{J_1}.$$
(61)
where
$$\mathrm{\Delta }_C=\alpha _+\mathrm{\Delta }^4+\alpha _{}\mathrm{\Delta }^3.$$
In actual fact, the relative weight of the two channels in the orbital antiferromagnet is not an adjustable parameter. If we calculate the divergence of the current at a given node in the lattice, we find that
$`𝐣(𝐫)`$ (62)
$`=`$ $`j(𝐫+a\widehat{𝐱},𝐫)j(𝐫a\widehat{𝐱},𝐫)+j(𝐫+a\widehat{𝐲},𝐫)j(𝐫+a\widehat{𝐲},𝐫)`$ (63)
$`=`$ $`{\displaystyle \frac{4et}{\mathrm{}J_1}}\left(\alpha _+\mathrm{\Delta }^3\alpha _{}\mathrm{\Delta }^4\right)=0`$ (64)
so the choice of $`𝐐`$ vector determines the mix of $`\gamma ^3`$ and $`\gamma ^4`$ symmetry in the orbital antiferromagnet.
In an itinerant model the condition for instability into the hidden order phase will be given by the Stoner criterion already discussed ($`J_1\chi _{0\psi }=1`$) where $`\chi _{0\psi }\frac{1}{t}`$ so that typically at the transition $`I=\frac{\beta e\mathrm{\Delta }}{\mathrm{}}`$ where $`\beta O(1)`$ is a constant; this is the relation we used for the current in our earlier phenomenological treatment. The form factor $`\gamma _𝐤^2=\mathrm{cos}(k_x)\mathrm{cos}(k_y)`$ corresponds to a quadrupolar charge density wave (q-CDW). The particular details of the conduction electron spectrum $`ϵ_𝐤`$ determine which order parameter has a higher critical temperature. For instance, if $`(Q_x,Q_y)=(\pi ,\pi )`$, and $`ϵ_𝐤=t(\mathrm{cos}(k_x)+\mathrm{cos}(k_y))`$, which corresponds to a nested Fermi surface, then from Equation (49) the relation for $`T_0`$ is
$`1`$ $`=`$ $`{\displaystyle \frac{J_1}{8\pi ^2t}}{\displaystyle _\pi ^\pi }𝑑k_y{\displaystyle _{\pi +|k_y|}^{\pi |k_y|}}𝑑k_x|\gamma _𝐤^i|^2{\displaystyle \frac{\text{tanh}\left(\frac{E_𝐤}{2T_0^i}\right)}{E_𝐤}}.`$ (65)
Here we used $`E_𝐤(\mathrm{\Delta }=0)ϵ_{𝐤+𝐐/2}=t(\mathrm{sin}(k_x)+\mathrm{sin}(k_y))`$. In this particular limit, we can explicitly verify that orbital antiferromagnetism has a higher $`T_0`$ than q-CDW. In the real material, the spectrum may differ greatly from this simple form, which may result in a preference of q-CDW over OAFM.
Our discussion in this section is based on a weak-coupling treatment of orbital antiferromagnetism, which is technically only valid in the vicinity of a nesting instability. Real heavy electron systems involve interactions of a size comparable with the band-width, in which the vicinity to nesting will no longer be a requirement. Practical modelling of these situations will require alternate strong-coupling methods, such as methods based on a Kondo lattice model. It is however interesting to note that that both symmetry and microscopic toy treatments appear to point to quadrupolar charge density wave and orbital antiferromagnetism as the leading contenders for hidden order in $`URu_2Si_2`$.
## V Fluctuations and Nesting
The sharpness of the phase transition in URu<sub>2</sub>Si<sub>2</sub> indicates that fluctuations do not make a significant contribution to thermodynamic properties. From the observed specific heat anomaly, the region of fluctuations is certainly smaller than $`\mathrm{\Delta }T0.1K`$, so that $`t_g=\mathrm{\Delta }T/T_0<\frac{1}{200}`$. This is an unusual situation in the general context of local moment magnetism, where broad fluctuation regions are generally seen in the specific heat anomaly. This result is sometimes taken to indicate that the hidden order involves a nested Fermi surface.Ikeda98 ; Sikkema96 However band structure calculations have not revealed any signs of a nested Fermi surface,Norman88 and it is difficult to see how such a condition might occur naturally in the complex band-structure of an f-electron system. Sharp mean-field transitions are generally taken as an indication of a large coherence length scale associated with fluctuations. In insulating systems (e.g. ferroelectrics) this arises from the long-range nature of the interaction. In superconductors and in nested charge density wave systems, the long coherence length $`\xi _0=v_F/\mathrm{\Delta }`$ is a consequence of the non-local order parameter response of the itinerant electron fluid.
So can the sharpness of the specific heat transition can be used in $`URu_2Si_2`$ be used to infer the presence of nesting $`URu_2Si_2`$? In fact, as we shall now see, a careful examination of the Ginzburg criterion for this system shows that while we may confirm that the ordering is itinerant in nature, the small size of the heavy electron Fermi energy means that we do not need to invoke nesting to understand that sharpness of the transition.
The Ginzburg criterion for a phase transition is given by
$$t_G=\frac{1}{[(\xi _0/a)^d(\delta S/k_B)]^{2/(4d)}},$$
(66)
or in three dimensions,
$$t_G=\frac{1}{(\xi _0/a)^6(\delta S/k_B)^2},(d=3)$$
(67)
Here $`a`$ is the lattice spacing, $`\delta S`$ is the entropy associated with the phase transition and $`\xi _0`$ is the coherence length of the order parameter. Microscopically, $`\xi _0`$ is determined from the Gaussian fluctuation term in the order-parameter expansion of the free energy,
$$\mathrm{\Delta }F\frac{1}{2}_𝐪\alpha |\mathrm{\Psi }_𝐪|^2(t+q^2\xi _0^2)$$
(68)
where $`t=\frac{T}{T_0}1`$ and $`\alpha `$ is a normalisation constant. The Gaussian coefficient in the integral is directly related to the static susceptibility of the order parameter
$$\chi _\psi ^1(q)=\alpha (t+q^2\chi _0^2)$$
(69)
The relationship between the coherence length $`\chi _0`$ and microscopic quantities depends markedly on the underlying physics. In insulators, $`\xi _0`$ tends to be determined by the range of interaction of the order parameter, but in itinerant systems, it is determined by the non-local order-parameter polarisation that develops in the electron fluid.
For example, in a local-moment antiferromagnet, with interaction $`H=\frac{1}{2}_qS_qS_q`$,
$$\chi _q^1=\mu _B^2(T+J_q)$$
(70)
When we expand around the unstable $`q`$ vector, $`q=Q_0`$
$$J(q)=\theta _C(1\kappa ^2(\stackrel{}{q}\stackrel{}{Q}_0)^2)$$
(71)
where $`\theta _C=T_0`$ is the Curie constant and $`\kappa ^1`$ the effective range of the interaction. With this form, we see that for insulating systems, the coherence length $`\xi _o=\kappa ^1`$ becomes the range of the interaction. For short-range interactions, this reason, the breadth of fluctuation region is generally large. In insulating systems, narrow fluctuation regimes are therefore associated with long-range interactions.
By contrast, in itinerant electron systems the order-parameter susceptibility generally takes the form
$$\chi _\psi ^1(q)=g+[\chi _{0\psi }(q)]^1$$
(72)
where $`g`$ is the strength of short-range interaction between electrons in the channel corresponding to the order parameter and $`\chi _{0\psi }`$ takes the form given in (50). It is the momentum dependence of $`\chi _0(q)`$ that determines the Ginzburg criterion in itinerant systems. To understand the role of nesting, let us consider a Fermi surface in which the departure from dispersion is measured by an energy scale $`\mu `$, (e.g $`ϵ_𝐤=2t(\mathrm{cos}k_x+\mathrm{cos}k_y)\mu `$) then the dispersion satisfies
$$ϵ_{𝐤𝐐}=ϵ_𝐤+2\mu $$
(73)
so that the bare susceptibility (50) is given by
$`\chi _{0\psi }(𝐪)={\displaystyle \underset{𝐤}{\overset{RBZ}{}}}|\gamma _{𝐤𝐐/2}^\mathrm{\Gamma }|^2{\displaystyle \frac{f(ϵ_𝐤^{})f(ϵ_{𝐤^+})}{ϵ_{𝐤+𝐪/2}ϵ_{𝐤𝐪/2}+2\mu }},`$ (74)
For $`\mu =0`$ and $`q=0`$, this integral is logarithmically divergent at $`T=0`$, and given by $`\chi _{0\psi }(q=0)\rho \overline{|\psi ^\mathrm{\Gamma }|^2}\mathrm{ln}\left(\frac{D}{T}\right)`$ at finite temperatures. Finite $`𝐪`$ modifies the Fermi functions in (74), so that
$`[\chi _\psi ^0(q)]`$ $``$ $`\left(1{\displaystyle \frac{(v_Fq)^2}{4}}{\displaystyle \frac{^2}{T^2}}\right)\chi _{0\psi }(T)`$ (75)
$``$ $`\rho \left(\mathrm{ln}\left({\displaystyle \frac{D}{T}}\right){\displaystyle \frac{(v_Fq)^2}{T^2}}\right)`$ (76)
so that
$$g+\chi _{0\psi }^1=\frac{g}{\mathrm{ln}(D/T_0)}\left[t+\left(\frac{v_Fq}{T_0}\right)^2\right]$$
(77)
and the free energy expansion takes the form
$$\mathrm{\Delta }F_𝐪|\mathrm{\Psi }_𝐪|^2(\left(\frac{\delta T}{T_0}\right)+\left(\frac{v_Fq}{T_0}\right)^2)^2)$$
(78)
so by comparing with (68), we see that for a nested system, the coherence length takes the “BCS” form
$$\xi _0\frac{v_F}{T_0}.$$
(79)
When $`|\mu |>>T_0`$, then we must replace $`T_0|\mu |`$ in the Landau-Ginzburg expansion, i.e.
$`\mathrm{\Delta }F`$ $``$ $`{\displaystyle _𝐪}|\mathrm{\Psi }_𝐪|^2\left(\left({\displaystyle \frac{\delta T}{|\mu |}}\right)+\left({\displaystyle \frac{v_Fq}{\mu }}\right)^2\right)^2`$ (80)
$`=`$ $`\left({\displaystyle \frac{T_0}{|\mu |}}\right){\displaystyle _𝐪}|\mathrm{\Psi }_𝐪|^2\left(\left({\displaystyle \frac{\delta T}{T_0}}\right)+\left({\displaystyle \frac{v_Fq}{|\mu |T_0}}\right)^2\right)^2`$ (81)
from which we see that the coherence length is given by
$$\xi _0\frac{v_F}{\sqrt{T_0|\mu |}}\sqrt{\xi _{nested}a}$$
(82)
where we have replaced $`\frac{v_F}{|\mu |}a`$, so loosely speaking, the absence of nesting replaces the coherence length by the geometric mean of the BCS coherence length $`v_F/T_0`$ and the lattice spacing.
Let us now return to our case, $`URu_2Si_2`$. Here, using the three dimensional form of the Ginzburg criterion, and taking $`URu_2Si_2`$, $`\delta S0.1k_B`$, so that
$$t_G\frac{100}{2(\xi _0/a)^6}.$$
(83)
Suppose the fluctuation region is less than $`0.1K`$, i.e. $`t_G<(0.1K/20K)1/200`$, then a lower bound for the coherence length is
$$\xi _0/a(t_G/100)^{\frac{1}{6}}=(2\times 10^4)^{1/6}5$$
Clearly, the presence of the sixth power in the Ginzburg criterion, means that only modest coherence length is required to account for experiments. Were the hidden order strictly associated with the local moments, then we would expect $`\xi _0/a1`$, and clearly, the absence of fluctuations is sufficient to rule this case out. The high pressure magnetic phase transition does in fact show clear signs of Ising fluctuations, and in this region, it would appear that the ordering transition is indeed local in nature. However, we can account for the coherence length of the hidden order transition by appealing to itineracy, without nesting. By assuming that $`v_F/\mathrm{\Delta }\frac{ϵ_F}{\mathrm{\Delta }}a25a`$. Taking $`ϵ_F10^3K`$, consistent with the heavy mass $`m^{}/m_e60`$ and $`\mathrm{\Delta }10^2K`$, we are clearly in the right range. From these arguments, we see that a correlation length of order $`5`$ lattice spacings is fully consistent with a system that is un-nested but itinerant. We conclude that the sharpness of the hidden order phase transition in $`URu_2Si_2`$ only implies itineracy. Indeed, there are a number of heavy electron systems with sharp thermodynamic transitions and commensurate magnetic order, indicative of un-nested Fermi surfaces, such as $`U_2Zn_{17}`$Ott84 and $`UPd_2Al_3`$.Geibel91 In each of these cases, it is most likely the itineracy alone that is responsible for the narrow fluctuation regime.
## VI Discussion
The observed Fermi liquid behaviour for $`T>T_0`$, the sharp nature of the transition and the large entropy loss point to the hidden order as a general density-wave with itinerant excitations formed from the local spin and orbital degrees of freedom of the uranium ions and f-electrons. Motivated by nuclear magnetic resonance measurements, we have expanded on our proposal (with J.A. Mydosh) of the hidden order as incommensurate orbital antiferromagnetism and have provided technical details for our predictions for elastic neutron scattering. Next we have turned to a microscopic description of the hidden order. After discussing symmetries and allowed particle-hole pairings in general terms, we studied the developing of these ordering in the setting of a toy single-band $`tJ`$ model within a weak-coupling approach. Within this framework, selection between q-CDW and OAFM ordering is not possible, though the situation may be different in the (experimentally relevant) strong-coupling regime. As discussed in Sec.III.2, density wave instabilities such as q-CDW and OAFM can account for the large entropy loss observed at the transition ($`\delta Sk_B^2T_0N^{}(0)`$) if the density of states at the Fermi surface, $`N^{}(0)`$, is large (as is the case in a heavy Fermi-liquid), and there is a substantial gapping of the Fermi surface.
The weak-coupling model we considered requires the nesting of a significant part of the Fermi surface. This requirement can be relaxed if the coupling is strong. Indeed it seems that a strong coupling description might be more appropriate for URu<sub>2</sub>Si<sub>2</sub>, since the transition temperature $`T_0`$ is an order of magnitude smaller than the gap $`\mathrm{\Delta }`$ unlike a weak coupling description where $`T_0`$ is more comparable with $`\mathrm{\Delta }`$. Unfortunately here it is difficult to perform controlled calculations in this regime, and thus experiment is crucial for discerning between these two competing scenarios of quadrupolar charge density wave order and orbital antiferromagnetism. In Sec. II we studied the consequences of OAFM for the neutron scattering structure-factor $`𝐒(𝐪)`$ and NMR at the Si and Ru sites. No particular microscopic model was assumed here, so the analysis is applicable for any coupling. NMR observations were used with our OAFM model to predict an incommensurate wavevector for orbital ordering which may be verified by neutron scattering measurements. To date, these predictions remain untested, as current experimental resolution is insufficient to observe the anticipated signal level from an OAFM.Bull02 ; Wiebe04 Here we identify a region of momentum space where elastic neutron scattering probes will clearly be able to distinguish between a spin density wave and an OAFM with current signal-noise levels. This prediction for orbital anferromagnetism remains a challenge for future experiments.
Our proposal of orbital antiferromagnetism is strongly motivated by the inhomogeneous line-broadening observed in ambient pressure NMR,Bernal01 and there are questions associated with this experiment that concern us greatly. In particular, the local fields measured via NMR in epoxied powdered samples are an order of magnitude larger than those probed by muon spin resonance or nuclear magnetic resonance in single-crystal ones. One interesting possibility is “motional narrowing”. The proposed orbital antiferromagnetic order is incommensurate and quite similar in its current patterns to a flux lattice of core-less vortices, where the absence of vortex cores weakens the pinning effect of disorder. In single domain crystals an incommensurate orbital antiferromagnet should then be weakly pinned, giving rise to large thermal motion.Blatter94 The probed local-fields will then be “motionally narrowed”, i.e their time-average will be significantly reduced in magnitude relative to their its static counterpart. One of the predictions of this scenario, is that the the muon or NMR linewidth will increase systematically with disorder - an effect that might be tested using radiation damage to systematically tune the disorder in a single sample.
Since we started working on this project, there have been a number of new experiments which may place further constraints on the nature of the hidden order in $`URu_2Si_2`$. In particular, recent magnetotransport measurementsBahnia05 indicate an unusually large Nernst signal in $`URu_2Si_2`$ that develops at $`T=T_0`$. This kind of behaviour has also been seen in the pseudogap phase of underdoped cuprate superconductors. In the case of the cuprate superconductors, this is most likely an effect of the Magnus force on the pre-formed pairs in the pseudogap. However, here, the absence of any superconductivity makes it far more likely that the giant Nernst effect seen here is a property of the quasiparticles in the presence of the hidden order parameter. These new results clearly place an important constraint on the microscopic nature of the order parameter.
Recent high magnetic field studiesKim03 have raised additional questions about the hidden order in $`URu_2Si_2`$. Application of high magnetic fields confirms that the hidden order persists to significantly higher values than does the remnant antiferromagnetism, affirming the two-phase scenario.Chandra02a Moreover the application of still higher fields leads to a profusion of new hidden order phases that may well cloak a field-induced quantum critical point. At the current time, it is not yet clear whether the proposed quantum critical point is a consequence of the loss of hidden order, or whether it might arise from the close vicinity to a quantum critical end point (as is the caseMillis02 with $`SrRu_2O_4`$.)
The hidden order mystery in Uranium Ruthenium-2 Silicon-2 can be regarded as part of a much broader set of longstanding problems that our community faces in the context of highly correlated materials. Coexisting forms of hidden order, novel metallic states manifested by unusual resistance and magnetotransport properties, field-induced quantum phase transitions- each of these features, present in $`URu_2Si_2`$, are manifest themselves in a wide range of other strongly correlated materials, such as the cuprate superconductors, strontium ruthenate, magneto-resistance materials, and many other heavy electron systems. $`URu_2Si_2`$ offers an alternative perspective on these problems, and optimistically, its ultimate solution will provide part of the key to understanding these broader questions.
## Appendix A Spatial distribution of vector potential due to OAFM
Here we calculate the vector potential $`𝐀(𝐱)`$ due to orbital order by summing up contributions from currents in all links. Consider first the contribution $`𝐀^{12}`$ defined in Eq.(8) for links along $`12`$. Performing the integral over $`w`$ gives
$`𝐀^{12}(𝐱)`$ $`=`$ $`\widehat{𝐱}{\displaystyle \frac{I_0}{c}}{\displaystyle \underset{j}{}}e^{i𝐐.𝐗_j}[\mathrm{sinh}^1\left({\displaystyle \frac{a/2+X_jx}{\sqrt{(yY_j+a/2)^2+(zZ_j)^2}}}\right)`$ (84)
$`\mathrm{sinh}^1\left({\displaystyle \frac{a/2+X_jx}{\sqrt{(yY_j+a/2)^2+(zZ_j)^2}}}\right)].`$
where we have used the notation $`𝐗_j=(X_j,Y_j,Z_j)`$ to denote the co-ordinates of the centre of the plaquette. For the link $`43`$ shown in Fig.4 we have
$`𝐀^{43}(𝐱)`$ $`=`$ $`\widehat{𝐱}{\displaystyle \frac{I_0}{c}}{\displaystyle \underset{j}{}}e^{i𝐐.𝐗_j}[\mathrm{sinh}^1\left({\displaystyle \frac{a/2+X_jx}{\sqrt{(yY_ja/2)^2+(zZ_j)^2}}}\right)`$ (85)
$`\mathrm{sinh}^1\left({\displaystyle \frac{a/2+X_jx}{\sqrt{(yY_ja/2)^2+(zZ_j)^2}}}\right)].`$
The $`x`$ component of the vector potential is then $`A_x(𝐱)=A^{12}(𝐱)+A^{43}(𝐱)`$. Similarly the vector potential in the links $`14`$ and $`23`$,
$`𝐀^{14}(𝐱)`$ $`=`$ $`\widehat{𝐲}{\displaystyle \frac{I_0}{c}}{\displaystyle \underset{j}{}}e^{i𝐐.𝐗_j}[\mathrm{sinh}^1\left({\displaystyle \frac{a/2+Y_jy}{\sqrt{(xX_j+a/2)^2+(zZ_j)^2}}}\right)`$ (87)
$`\mathrm{sinh}^1\left({\displaystyle \frac{a/2+Y_jy}{\sqrt{(xX_j}+a/2)^2+(zZ_z)^2}}\right)],`$
$`𝐀^{23}(𝐱)`$ $`=`$ $`\widehat{𝐲}{\displaystyle \frac{I_0}{c}}{\displaystyle \underset{j}{}}e^{i𝐐.𝐗_j}[\mathrm{sinh}^1\left({\displaystyle \frac{a/2+Y_jy}{\sqrt{(xX_ja/2)^2+(zZ_j)^2}}}\right)`$ (88)
$`\mathrm{sinh}^1\left({\displaystyle \frac{a/2+X_{jy}y}{\sqrt{(xX_ja/2)^2+(zZ_j)^2}}}\right)],`$
yield the $`y`$ component of the vector potential $`A_y(𝐱)=A^{14}(𝐱)+A^{23}(𝐱)`$. The magnetic field follows straightforwardly from $`𝐁=\times 𝐀`$.
###### Acknowledgements.
We are very grateful to J.A. Mydosh for innumerable discussions on $`URu_2Si_2`$ which have shaped our view on this subject. We have also benefitted greatly from interactions with W.J.L. Buyers particularly related to the experimental neutron scattering situation and the Ginzburg criterion in itinerant systems. Discussions with K. McEwen, B. Maple, B. Marston, P. Ong and I. Usshikin are also acknowledged. VT is supported by a Junior Research Fellowship from Trinity College, Cambridge. P. Chandra and P. Coleman are supported by the grants NSF DMR 0210575 and NSF DMR 0312495 respectively. P. Coleman acknowledges the hospitality of the Kavli Institute for Theoretical Physics, where part of this work was performed. |
warning/0507/cond-mat0507039.html | ar5iv | text | # Pattern reconstruction and sequence processing in feed-forward layered neural networks near saturation
## I Introduction
Models of attractor neural networks for processing sequences of patterns, as a realization of a temporal association, have been of great interest following Hopfield’s pioneering work \[1-7\] and renewed interest has come through both, the availability of new theoretical dynamic approaches to study the evolution of disordered systems, in particular neural networks near saturation \[8-12\] and experimental findings. Among the latter are the results of Miyashita et al. who showed that serial positions of stimuli to which monkeys were exposed during training are converted to spatial correlations of neural activities MC88 ; Mi88 .
A neural network model with a symmetric learning rule that consists of a Hebbian part and a pair of pattern sequences has been proposed to interpret the experiments GTA93 ; Br94 . The analysis of the equilibrium states for both finite and extensive loading of patterns, revealed the presence of correlated attractors with a decreasing correlation as the separation of the patterns in the sequence increases, in apparent support of the experimental work. Phase diagrams were obtained where correlated attractors compete with Hopfield-like attractors and with symmetric states Cu93 ; CT94 . The dynamical evolution to the stationary states has been studied recently by means of dynamical replica theory UHO04 .
Neural network models for sequence processing with asymmetric interactions are more natural from a biological point of view. Due to the lack of microscopic detailed balance, however, the equilibrium states of the network cannot be obtained by means of the statistical mechanics approach AGS87 , and one has to resort to a dynamical study. Analytic and numerical studies of the stationary states and some aspects of the dynamics for the competition between pattern reconstruction and asymmetric sequence processing in the case of finite loading of patterns, that is in the absence of stochastic noise, where the ratio $`\alpha =p/N`$ between the number of patterns $`p`$ and units $`N`$ is zero, appeared in works by Coolen et al. CS92 ; WSC95 . Phase diagrams for the stationary states of a parallel dynamics exhibit either stable Hopfield-like or symmetric mixture states characterized by fixed-point solutions and stable limit cycles described by periodic fixed points. Non-stationary solutions of the dynamics also appear for increasingly large number of patterns CS92 . More recent works deal with stationary limit cycles in asymmetric sequence processing without pattern reconstruction, for extensive loading of patterns \[12,23-26\].
It is important to consider the effects of extensive loading of patterns and to study the stability of the states that appear in the competition between pattern reconstruction and asymmetric sequence processing to the presence of stochastic noise (finite $`\alpha `$). This noise could be due to the presence of a previously learnt macroscopic number of patterns following a specific learning rule. Since the problem is a dynamical one, it is also important to study the dependence of the network performance on initial conditions.
Those are issues that, apparently, have not been considered before and the purpose of the present paper is to study them on a tractable dynamical feed-forward layered neural network model with no feed-back loops DKM89 . This is an extensively used model that consists of identical layers of $`N`$ non-interacting units on each layer, with synaptic interactions only between units on consecutive layers. The feed-forward nature of the model and the updating of the units on each layer endow the network of a dynamics in which the layer index becomes a discrete time. We consider an interaction matrix that yields either a static Hebbian noise or a dynamic Hebbian plus-sequential noise and restrict the work to binary units and patterns. In the finite loading case, the model becomes identical to that of Coolen and Sherrington for a parallel dynamics CS92 .
The outline of the paper is the following. In Sec. 2 we present the model and the relevant order parameters. In Sec. 3 we derive the macroscopic dynamics of the network and in Sec. 4 we present the results in the form of phase diagrams of stationary states and regions of non-stationary solutions. We end with a discussion and conclusions in Sec. 5.
## II The model
The network model consists of $`L`$ layers with $`N`$ binary Ising units (neurons) on each layer $`l`$ in a microscopic state $`𝝈(l)=\{\sigma _1(l),\mathrm{},\sigma _N(l)\}`$, in which each $`\sigma _i(l)=\pm 1`$. The state $`+1`$ represents a firing neuron and the state $`1`$ a neuron at rest. The microscopic dynamics of the network is generated as follows: given a configuration on the first layer, $`𝝈(1)`$, all units on layer $`l+1`$ are updated simultaneously according to the alignment of each unit $`i`$ to its local field
$$h_i(l+1)=\underset{j=1}{\overset{N}{}}J_{ij}(l)\sigma _j(l),$$
(1)
due to the states of the units on the previous layer $`l`$, following a stochastic law with probability
$$\mathrm{Prob}(\sigma _\mathrm{i}(\mathrm{l}+1)|𝝈(\mathrm{l}))=\frac{\mathrm{exp}[\beta \sigma _\mathrm{i}(\mathrm{l}+1)\mathrm{h}_\mathrm{i}(\mathrm{l}+1)]}{2\mathrm{cosh}[\beta \mathrm{h}_\mathrm{i}(\mathrm{l}+1))]}.$$
(2)
Thus, the network has a parallel dynamics with no feed-back loops in which the layer indices may be associated with discrete time steps. Here, $`J_{ij}(l)`$ is the synaptic connection defined below between unit $`j`$ on layer $`l`$ and unit $`i`$ on layer $`l+1`$. The parameter $`\beta =T^1`$ controls the synaptic noise such that the dynamics of the network becomes a deterministic one when $`T0`$ and fully random when $`T\mathrm{}`$. In the former case, the state of a unit is given by the deterministic form
$$\sigma _i(l+1)=\mathrm{sgn}[h_i(l+1)].$$
(3)
A macroscopic set of $`p=\alpha N`$ independent and identically distributed random patterns $`𝝃^\mu (l)=\{\xi _1^\mu (l),\mathrm{},\xi _N^\mu (l)\}`$, $`\mu =1,\mathrm{},p`$, and each $`\xi _i^\mu (l)=\pm 1`$ with probability $`\frac{1}{2}`$ are stored in the learning stage on the units of each layer $`l`$, independently of other layers. Since we are interested in the retrieval of one or a small number $`c`$ of patterns and in the recognition of a finite sequence, we assume a learning rule of the form
$`J_{ij}(l)`$ $`=`$ $`{\displaystyle \frac{1}{N}}{\displaystyle \underset{\mu ,\rho =1}{\overset{c}{}}}\xi _i^\mu (l+1)A_{\mu \rho }\xi _j^\rho (l)`$
$`+`$ $`{\displaystyle \frac{1}{N}}{\displaystyle \underset{\mu ,\rho =c+1}{\overset{p}{}}}\xi _i^\mu (l+1)B_{\mu \rho }\xi _j^\rho (l),`$
where
$`A_{\mu \rho }`$ $`=`$ $`\nu \delta _{\mu \rho }+(1\nu )S_{\mu \rho },\mu \mathrm{mod}\mathrm{c},`$ (5)
$`B_{\mu \rho }`$ $`=`$ $`b\delta _{\mu \rho }+(1b)S_{\mu \rho },\mu \mathrm{mod}\mathrm{p},`$
in which $`B_{\mu \rho }`$ is non-zero only for $`\mu c+1`$. These are the (layer independent) elements of matrices $`𝐀`$ and $`𝐁`$, with continuous independent $`\nu `$ and $`b`$ ($`0\nu ,b1`$), in which $`\delta _{\mu \rho }=1`$ if $`\mu =\rho `$, and zero otherwise, while $`S_{\mu \rho }=\delta _{\mu ,\rho +1}`$ is a permutation matrix element. The signal-to-noise analysis carried out below depends on the assumption that the patterns can be separated into $`c`$ ”low” or condensed ones, with finite overlaps with the states of the network in the $`N\mathrm{}`$ limit and ($`pc`$) ”high” or non-condensed patterns with overlaps of $`O(1/\sqrt{N})`$ with the states. The assumption is justified in the case of a block-matrix interaction (here, a diagonal two-block matrix) and the specific form we choose is inspired in an earlier proposal CT94 . In that spirit we assume here not necessarily the same weights, $`\nu `$ and $`b`$. This is in order to contemplate different simple possibilities and to avoid an eventual closure problem in the noise term discussed below. The first part of $`J_{ij}(l)`$ represents a finite cycle which is a superposition of a Hebbian learning (the diagonal part of $`𝐀`$) of single patterns and of patterns in a sequence (the non-diagonal part). Again, in the spirit of that reference, one may think of the condensed part in the presence of an infinite cycle (in the large $`p`$ limit) of a previously learnt similar superposition of Hebbian and sequential patterns represented by the second part of $`J_{ij}(l)`$, which acts as a noise. Eventually, one could have $`0\nu 1`$ and a Hebbian noise with $`b=1`$ CT94 , which is a rather convenient choice.
The learning stage of the network is a dynamical process that involves patterns on two consecutive layers. The Hebbian part of the rule may be thought as a static process that reinforces the learning of the same pattern on every pair of consecutive layers (times) whereas the sequential part of the rule is a dynamic process in which the synaptic interaction is due to a pattern at a given time with the following pattern in the sequence at the next discrete time. It has been found before, in studying the finite loading case, that the competing static process has a stabilizing effect on the dynamic sequential process leading to a phase of symmetric states, which are the only ordered stable states within a wide range of relative synaptic strengths under an appropriate amount of synaptic noise parameter $`T`$. Eventually, the static process may fail to lock the transitions in the dynamic process and non-stationary quasi-periodic solutions may appear which have already been found in the case of finite loading CS92 . These are natural features of the model that are enhanced by stochastic noise (finite $`\alpha `$) as will be seen and discussed in this work.
The properties of $`𝐀`$, in particular its eigenvalues and the complete set of orthogonal eigenvectors, lead to symmetries of the solutions of the non-linear parallel dynamics in the form of a $`\nu /(1\nu )`$ duality, that has been discussed for finite loading of patterns CS92 . We find here an additional $`b/(1b)`$ duality. For the case of continuous bifurcations of solutions, the eigenvalues of the matrix yield the transition temperatures to the ordered states and the eigenvectors give the symmetry directions of the macroscopic overlaps between the states of the network and the patterns.
We define the macroscopic overlap between the configuration $`𝝈(l)`$ of the network on layer $`l`$ and one or more condensed key patterns $`𝝃^\mu (l)`$, $`\mu =1,\mathrm{},c`$, on that layer as the large-N limit, $`m^\mu (l)`$, of
$$m_N^\mu (l)=\frac{1}{N}\underset{i=1}{\overset{N}{}}\xi _i^\mu (l)\sigma _i(l),$$
(6)
where the brackets denote a thermal average with Eq.(2). Since the number of condensed patterns is finite, one may use the self-averaging property to write $`m^\mu (l)=\xi _i^\mu (l)\sigma _i(l)`$, where the outer bracket here and below denotes a configurational average over the patterns. Similarly, we define the overlap between the same configuration and a given non-condensed pattern $`𝝃^\mu (l)`$, $`\mu =c+1,\mathrm{},p`$, on layer $`l`$,
$$M_N^\mu (l)=\frac{1}{N}\underset{i=1}{\overset{N}{}}\xi _i^\mu (l)\sigma _i(l).$$
(7)
Assuming that a given configuration of the first layer has a finite overlap $`m^\mu (1)=O(1)`$ with one or more condensed patterns and overlaps $`M_N^\mu (1)=O(1/\sqrt{N})`$ with the non-condensed patterns, the dynamic evolution of the network will yield overlaps $`m^\mu (l)=O(1)`$ and $`M_N^\mu (l)=O(1/\sqrt{N})`$ on the following layers. We consider next the evolution equations for the overlaps.
## III Dynamics of the network
Adapting the standard procedure for the layered network to our model, we write the local field at a unit on layer $`l+1`$ due to the overlaps with all patterns on layer $`l`$, in the large-N limit DKM89 ,
$$h_i(l+1)=\underset{\mu ,\rho =1}{\overset{c}{}}\xi _i^\mu (l+1)A_{\mu \rho }m^\rho (l)+z_i(l),$$
(8)
where the first term is the signal and $`z_i(l)`$ is the large-N limit of the noise
$$R_i(l)=\underset{\mu ,\rho =c+1}{\overset{p}{}}\xi _i^\mu (l+1)B_{\mu \rho }M_N^\rho (l),$$
(9)
due to the overlaps of the states with the non-condensed patterns. This is a random quantity in both the patterns on layer $`l+1`$ and the implicit dependence on thermal and configurational randomness of the overlaps $`M_N^\mu (l)`$ on layer $`l`$ due to the previous layers.
The noise gives a finite contribution to the local field. Indeed, due to the fact that $`M_N^\mu (l)=O(1/\sqrt{N})`$ and that $`R_i(l)`$ is a sum of a large number of statistically independent random variables, one can first apply the central limit theorem and then the law of large numbers to conclude that $`z_i(l)`$ follows a Gaussian distribution with mean zero and a variance $`\mathrm{\Delta }^2(l)`$ given by the large-N limit of
$$\mathrm{\Delta }_N^2(l)=\underset{\mu =c+1}{\overset{p}{}}[bM_N^\mu (l)+(1b)M_N^{\mu 1}(l)]^2.$$
(10)
We used the fact that the patterns are unbiased and uncorrelated random variables with $`\xi _i^\mu (l+1)=0`$ and $`\xi _i^\mu (l+1)\xi _i^\nu (l+1)=\delta _{\mu \nu }`$.
We make use now of the local field to derive the recursion relations for the vector overlaps with the condensed patterns, $`𝒎(l)=\{m^\mu (l)\},\mu =1,\mathrm{},c`$, and for the variance of the noise. For the former we obtain
$$𝒎(l+1)=𝝃(l+1)Dz\mathrm{tanh}\{\beta [𝝃(l+1).𝑨𝒎(l)+\mathrm{\Delta }(l)z]\},$$
(11)
where $`Dz=e^{z^2/2}dz/\sqrt{2\pi }`$ and the brackets denote an average over the explicit patterns. The variance of the noise requires recursion relations not only for the average squared non-condensed overlaps, $`M_N^\mu (l)^2`$ and $`M_N^{\mu 1}(l)^2`$, which can be derived in the usual way DKM89 , but also for the correlation of two consecutive overlaps $`M_N^\mu (l)M_N^{\mu 1}(l)`$. This generates, in turn, correlations between next-to-consecutive overlaps, $`M_N^\mu (l)M_N^{\mu 2}(l)`$ and so on, which requires to keep track of a general form
$`C_n^2(l)`$ $`=`$ $`{\displaystyle \underset{\mu =c+1}{\overset{p}{}}}[bM_N^\mu (l)+(1b)M_N^{\mu 1}(l)]`$
. $`[bM_N^{\mu n}(l)+(1b)M_N^{\mu n1}(l)]`$
leading altogether to $`pc+1`$ recursion relations,
$`\mathrm{\Delta }^2(l+1)`$ $`=`$ $`\stackrel{~}{b}^2(\alpha +\beta ^2I^2\mathrm{\Delta }^2(l))`$
$`+`$ $`2b(1b)\beta ^2I^2C_1^2(l),`$
$`C_1^2(l+1)`$ $`=`$ $`\stackrel{~}{b}^2\beta ^2I^2C_1^2(l)`$
$`+`$ $`b(1b)[\alpha +\beta ^2I^2(\mathrm{\Delta }^2(l)+C_2^2(l))],`$
$`C_n^2(l+1)`$ $`=`$ $`\stackrel{~}{b}^2\beta ^2I^2C_n^2(l)`$
$`+`$ $`b(1b)\beta ^2I^2(C_{n1}^2(l)+C_{n+1}^2(l)),`$
for $`n=2,\mathrm{},pc1`$ and $`C_{pc}=\mathrm{\Delta }`$, where
$`\stackrel{~}{b}^2`$ $`=`$ $`b^2+(1b)^2`$ (16)
$`I(l)`$ $`=`$ $`1q(l).`$
Here, $`q(l)=\sigma (l)^2`$ is the spin-glass order parameter given by
$$q(l)=Dz\mathrm{tanh}^2\{\beta [𝝃(l+1).𝑨𝒎(l)+\mathrm{\Delta }(l)z]\}.$$
(17)
Note that the Gaussian noise is symmetric under the change $`b`$ $`(1b)`$ and that in the case where either $`b`$ is one or zero, that is for purely Hebbian or sequential noise, respectively, the recursion relation for the variance reduces to the simple form
$$\mathrm{\Delta }^2(l+1)=\alpha +\beta ^2I^2(l)\mathrm{\Delta }^2(l).$$
(18)
Otherwise, one has to face the full set of recursion relations which may become quite a task for an asymptotically large $`p`$. Indeed, the system of relations may not form a closed and finite set and there is no guaranty that this is not, in general, the case. Fortunately, by working out numerically the equations for the dynamics in all the cases we were interested in this work we found that the set of recursion relations is practically finite and one proceeds as follows.
Given an initial overlap $`𝒎(1)=\{m^\mu (1)\},\mu =1,\mathrm{},c`$, on the first layer, and taking $`\mathrm{\Delta }^2(1)=C_n^2(1)=\alpha `$, for all $`n`$, Eqs. (11)-(17) describe the dynamics of the network and its stationary states. The latter are given by the fixed-point solutions $`𝒎^{}=𝒎(l+1)=𝒎(l)`$, $`\mathrm{\Delta }^{}=\mathrm{\Delta }(l+1)=\mathrm{\Delta }(l)`$ and $`C_n^{}=C_n(l+1)=C_n(l)`$, for each $`n=1,\mathrm{},pc1`$. Expressing $`C_1^{}`$ in terms of all higher $`C_n^{}`$’s yields $`(\mathrm{\Delta }^{})^2\alpha `$ and in the case of $`b`$ either one or zero this reduces to the simple form
$$(\mathrm{\Delta }^{})^2=\frac{\alpha }{1\beta ^2(1q)^2},$$
(19)
where $`q`$ is the fixed-point value of $`q(l)`$.
There may also exist other solutions, as non-stationary states, as will be seen in the next section. Solving for $`T=0`$, searching for fixed-point solutions and for non-stationary states one finds that only a small number of the $`C_n^{}`$’s on the $`(pc)`$-cycle of non-condensed overlaps are clearly non-zero and all others vanish. The ones that survive become smaller with increasing $`n`$ indicating the vanishing of the averaged overlaps $`M_N^\mu (l)M_N^{\mu n}(l)`$.
In the case of finite loading of patterns, where $`\alpha =0`$, the variance of the noise vanishes and the equations for the overlaps and the spin-glass order parameter become disconnected and we recover the equations for the overlaps found in previous work CS92 .
## IV Results
We show and discuss next our results and use the solutions for finite loading as a guide. We are mainly interested in the network performance in the case of training with a macroscopic number of patterns and on the impact of stochastic noise on the dynamics and the stationary states. Among the important features are the critical loading level, the relative size of the regions with meaningful information processing, the presence of stable symmetric mixture states and quasi-periodic non-stationary solutions. We also want to find out the specific dependence, if any, of these properties on the form of the stochastic noise and we restrict ourselves to Hebbian or Hebbian plus sequential noise.
The phase diagrams discussed below are obtained from the dynamical equations for the macroscopic order parameters and they depend naturally on initial conditions. Quite different sets of these conditions may reflect distinct basins of attraction of the stationary states of the model (specified by a given pair of values of $`\nu `$ and $`b`$) for given $`T`$ and $`\alpha `$. It will be seen that non-stationary states can be reached from certain initial overlaps.
We concentrate on a small number of macroscopic condensed overlaps, specifically $`c=3`$ and $`c=4`$, since the results depend on whether $`c`$ is even or odd, with the remaining $`pc`$ patterns as noise in the case of extensive loading. This already illustrates the typical results. For specific results on higher values of $`c`$ and the differences between an even and odd number of patterns in the case of finite loading we refer the reader to earlier work CS92 . Some of these results will be used below to infer the changes that one may expect for extensive loading.
### IV.1 Finite loading
First we reconsider the solutions for finite loading, where $`\alpha =0`$, in order to show the phases that appear, the periodic features of the cyclic phase and the quasi-periodic solutions in the region of unstable fixed points.
The $`(T,\nu )`$ phase diagram of stationary solutions for $`\alpha =0`$ and $`c=4`$ is shown in Fig.1 for the best initial overlaps that favor single-pattern reconstruction (Hopfield or Hopfield-like solutions, see below). There is a paramagnetic phase (P) where $`𝒎^{}=0`$ and $`q=0`$ above $`T=1`$. This is a line of continuous bifurcations to a phase of stable symmetric fixed-point solutions (S), $`m_1^{}=m_2^{}=m_3^{}=m_4^{}0`$, that ends to the left at a boundary (dashed line) obtained analytically beyond which the symmetric states become unstable. This is a region of non-stationary quasi-periodic solutions marked q-p, as will be seen below, that is included with the warning that it is not a true part of the phase diagram of stationary states. The same applies to all other phase diagrams in this paper where the q-p solutions are shown. It is worth noting that the symmetric states are the only stable states in the $`S`$ region due to the stabilizing effect of the static condensed patterns in the learning rule and that this is a characteristic feature of the model, as discussed in Sec. 2. Note that the symmetric states are enhanced by synaptic noise.
Eventual alterations of the learning rule aimed at eliminating what has been termed as ”spurious” (here not at all) symmetric mixture states, say by means of the introduction of a bias in the distribution of the patterns AGS97/2 , would leave nothing but the non-stationary q-p states over a large region of the phase diagram.
The upper region of the phase diagram is a phase of stable Hopfield-like fixed-point solutions (H), with the four components $`m_1^{},\mathrm{},m_4^{}`$ not all equal and different from zero, except at $`T=0`$ where one component may be one and the other ones zero. Using this zero $`T`$ result as an initial solution we construct by numerical iteration of the overlaps the upper and lower phase boundaries indicated by solid lines. All other initial conditions yield phase boundaries above the upper or below the lower line and larger regions of q-p solutions. Those lines are boundaries of first-order transitions for all values of $`\nu `$, except $`\nu =1`$ or zero. The first-order transition is to the region of q-p solutions with remanent finite overlap components. The location of the transition for $`c=4`$ is, as expected, close to that for higher $`c`$ obtained in ref. CS92 . We also find, as expected, larger discontinuities at the first-order transition and shorter remanent overlaps in the q-p region, with our smaller number of condensed patterns than in that reference.
The lower region of the phase diagram is the phase of period-4 stable stationary cyclic solutions (C). These are solutions in which $`m_i(l+4)=m_i(l)0`$ for $`i=1,\mathrm{},4`$, with one overlap close to one and the other ones near zero in each time step after a transient. This fact characterizes sequence processing in which the network makes a transition from one pattern to the next at each layer. The upper and lower first-order phase boundaries appear symmetrically in the phase diagram due to the $`\nu /(1\nu )`$ duality and both are equally sensitive to initial values for the overlaps.
The nature of the cyclic and quasi-periodic solutions for $`\alpha =0`$ and $`c=4`$ is best illustrated by the power spectra shown in Figs. 2 (a) and (b), respectively. The first one is for a typical $`\nu =0.1`$ and $`T=0.15`$, within the phase of cyclic states, and the other one is for a typical $`\nu =0.30`$ and $`T=0.35`$ in the region of non-stationary states. The power spectrum $`S(\omega )`$, where $`\omega `$ is the frequency conjugate to the layer index (a discrete time) may be defined in extension to the case of a continuous time ER85 , as
$$S(\omega )=(\mathrm{const})\mathrm{lim}_L\mathrm{}\frac{1}{L}|\underset{l=0}{\overset{L}{}}e^{i\omega l}m_1(l)|^2,$$
(20)
where $`m_1(l)`$ is any one of the components of the overlaps.
The spectrum in Fig. 2a clearly exhibits a period-4 solution. In contrast, the spectrum of non-stationary solutions shown in Fig. 2b indicates quasi-periodic solutions. These are solutions characterized by main frequencies that are linear combinations of four recognizable basic frequencies.
Consider next the $`(T,\nu )`$ phase diagram of stable stationary states for $`\alpha =0`$ and $`c=3`$, again with initial conditions $`m_1=1,m_2=m_3=0`$ that favor single-pattern reconstruction, shown in Fig. 3. The symmetric states are now stable over the whole diagram, up to $`T=1`$, except for an upper and a symmetrically placed lower region of unstable states, where Hopfield-like (H) and cyclic (C) solutions, respectively as indicated, are the only stable solutions. The stability of the symmetric solutions for all $`\nu `$ at low $`T`$ is a characteristic feature of the phase diagrams for an odd number of condensed patterns CS92 . There is also a pair of tiny q-p solutions, that do not appear on the scale of the figure, between the exclusive H and C regions and the exclusive S phase for high $`T`$. The H or C solutions are stable everywhere above or below the upper or the lower solid lines, respectively. The symmetric states compete for stability with Hopfield-like or cyclic solutions in the regions $`H+S`$ or $`C+S`$, depending on initial values of the overlaps. The symmetric states in the S phase are, again, the only stable states in the intermediate region between the two first-order transitions. Other initial more symmetric overlaps yield, again, smaller H and C phases and pairs of q-p solutions between the S and the H and C phases.
### IV.2 Extensive loading
For extensive loading, with $`\alpha 0`$, one has to solve the full set of Eqs. (11)-(17). We still expect to have three ordered phases: H (Hopfield-like), S and C, with specific $`𝒎^{}0`$ and $`q0`$. Also, a disordered spin-glass phase (SG) should appear with $`𝒎^{}=0`$ and $`q0`$ at the end of the ordered phases and this phase should survive for all higher $`T`$. The latter is a property of the layered network. All the transitions are now expected to be discontinuous and all the phase boundaries have to be obtained numerically.
One may expect the $`\nu /(1\nu )`$ duality also to hold for non-zero $`\alpha `$. Indeed, the Gaussian noise due to the non-condensed patterns does not change neither under the set of linear transformations that keep the probability distribution of the condensed patterns invariant nor under the permutation index matrix $`𝑺=\{S_{\mu \rho }\}`$. These are the two basic transformations which lead to a one-to-one correspondence between every state in the upper part $`(\nu >1/2)`$ in the $`(T,\nu )`$ phase diagram and a state in the lower part CS92 . The duality serves to check the symmetry between the numerically constructed phase boundaries.
The new feature of the $`(T,\nu )`$ phase diagrams for both $`c=3`$ and $`c=4`$, not shown here, that are reached from Hopfield type initial overlaps that favor single-pattern reconstruction, $`m_1=1,m_2=m_3=0`$ and $`m_1=1,m_2=m_3=m_4=0`$, respectively, is that the symmetric states become stable at low temperatures, including $`T=0`$, within a finite range of intermediate values of $`\nu `$ as soon as $`\alpha `$ is non-zero. These states remain as the only stable states in that region, and there are H and C phases for large and small values of $`\nu `$, respectively, and a SG phase at high $`T`$. The phase boundaries to the H and C phases are nearly the same for both $`c=3`$ and $`c=4`$, and there appear no q-p solutions. The resulting phase diagrams are very similar for either a Hebbian noise or full (Hebbian plus sequential) noise with $`b=\nu `$, the main difference being that the symmetric phase becomes somewhat enlarged at high $`T`$ in the latter case.
Other more symmetric initial overlaps, which favor symmetric mixture states, lead to somewhat different results for $`c=3`$ and $`c=4`$. For the latter, initial overlaps $`m_1=0.21,m_2=m_3=m_4=0.20`$ reduce the size of the H and C regions and lead now to intermediate regions of q-p solutions, as shown in Fig. 4 for $`\alpha =0.008`$ and Hebbian noise. A very similar diagram is obtained in the case of full noise with $`b=\nu `$, and analysis of the power spectrum confirms the nature of the q-p states as non-stationary quasi-periodic solutions. In contrast, with initial overlaps $`m_1=0.21,m_2=m_3=0.20`$, in the case of $`c=3`$, the symmetric states become the only stable states at low $`T`$ for all values of $`\nu `$. Regions of stable Hopfield-like and cyclic states appear at higher $`T`$ and, again, there appear non-stationary q-p solutions between these phases and the $`S`$ phase.
All components of the overlap in the H phase are different from zero for non-zero $`\alpha `$ at $`T0`$, although some of them may be small. In contrast, some components may be zero when $`T=0`$, as in the case of finite loading CS92 . Also, for non-zero $`\alpha `$, the stable states in the C phase are still period-4 cyclic solutions. We checked this explicitly for $`\alpha =0.008`$ and values for $`\nu `$ and $`T`$ within that phase.
To see the overall role of extensive loading and to find the critical storage ratio $`\alpha _c`$ for a given superposition of Hebbian and sequential learning, we consider next the $`(\alpha ,\nu )`$ phase diagram of stable states at $`T=0`$ for $`c=4`$, shown in Fig. 5, for Hebbian noise and initial overlaps $`m_1=1,m_2=m_3=m_4=0`$ that favor single-pattern reconstruction. Note that the critical storage ratio for pure Hebbian learning, that is for $`\nu =1`$, is $`\alpha _c=0.269`$, in accordance with the known result for the layered network model DKM89 . Due to the $`\nu /(1\nu )`$ duality, it is also seen to be the critical ratio for pure sequential learning of the condensed patterns, as it should be, since a pure sequential noise is a Hebbian noise. The critical storage ratio $`\alpha _c`$ for the retrieval of Hopfield-like or cyclic states is given by a point on the phase boundaries where the H or C phases end. In order to check our results obtained from numerical iterations, we performed numerical simulations to locate a few points on the first-order transitions between the ordered phases, at $`T=0`$ and we obtained results in good agreement. As will be seen below (cf. Fig. 7), somewhat different results are obtained in the case of full noise. Also, more symmetric initial conditions, say $`m_1=0.21,m_2=m_3=m_4=0.20`$, lead to further q-p solutions.
We consider next the $`(\alpha ,\nu )`$ phase diagram of stable states at $`T=0`$ for $`c=3`$ and Hebbian noise. For the initial overlaps $`m_1=1,m_2=m_3=0`$ one obtains a similar diagram to that shown in Fig. 5 for $`c=4`$. The first-order transitions between the H and the C phase with the SG phase are close to the case of $`c=4`$. Instead, for more symmetric initial overlaps, say $`m_1=0.4,m_2=0.3,m_3=0.2`$, one finds a quite different phase diagram with a symmetric phase almost everywhere below a transition to the SG phase and largely suppressed stable Hopfield-like and cyclic states, as shown in Fig. 6. A similar diagram, except for an enlarged S phase, is obtained in the case of full noise with $`b=\nu `$.
We look now at the effects of Hebbian plus sequential noise for $`c=4`$ and finite $`\alpha `$ at $`T=0`$. We do this again for $`b=\nu `$ and consider also the case of a matrix in which the high components have equally favored Hebbian and sequential parts, in order to explore the symmetry of the model around $`b=0.5`$, independently of $`\nu `$. In each case we consider two kinds of initial conditions: one favoring single pattern and cycle retrieval, with $`m_1=1,m_2=m_3=m_4=0`$, and another more symmetric with $`m_1=0.21,m_2=m_3=m_4=0.20`$.
The results for $`b=\nu `$ and $`b=0.5`$ for the first initial condition are shown in Figs. 7(a) and (b), respectively. In both cases there is now a finite region of non-stationary q-p solutions below the phase of stable symmetric states, that was not present for purely Hebbian noise. This is a feature of complete (Hebbian plus sequential) noise and in the case of $`b=0.5`$ there is also a considerable improvement of the critical $`\alpha _c`$ for Hopfield-like and cyclic retrieval. Indeed, we find the quite higher $`\alpha _c=0.6438`$ shown in Fig. 7(b). We also checked with other choices for $`b`$ and found that $`b=0.5`$ seems to be the optimal case.
Coming back to the more symmetric initial condition for $`c=4`$, one finds quite larger regions of stable symmetric states and of non-stationary q-p solutions and particularly smaller regions of $`H`$ and $`C`$ phases for both $`b=\nu `$ and for $`b=0.5`$.
## V Discussion and conclusions
We studied in this work the dynamics of the competition between pattern reconstruction and asymmetric sequence processing in an exactly solvable feed-forward layered neural network model for both finite and for extensive loading of patterns. The strictly feed-forward nature of the network makes it an ideal system to study the dynamics as a discrete-time evolution. Given the initial overlaps as inputs on the sites of the first layer, the dynamics is generated by random local fields at every site on the next layer which depend on the synaptic interaction between units on the two layers. In turn, the local fields determine the probability of updating of units at the sites of that layer. The asymmetry of the sequence processing in this work refers to the sequential part of the learning rule that consists of a synaptic interaction connecting a pattern on a layer with the consecutive pattern in the sequence on the next layer. A symmetric sequence processing, not considered in this work, would involve an additional synaptic interaction of the same strength connecting a pattern on a layer with the previous pattern in the sequence on the next layer. We come back to this issue below.
The model studied here is based on the superposition of a Hebbian and a sequential learning rule for a finite cycle of condensed patterns and a stochastic noise due to a previously learnt macroscopic set of either single patterns that follow a Hebbian rule or a superposition of sequential patterns in a cycle with single patterns. The superposition we consider is suitable for the study of the competition between pattern reconstruction and asymmetric sequence processing. We were especially interested in the effects of finite synaptic noise ($`\alpha 0`$) on the performance phase diagrams that appear in the case of extensive loading of patterns. Naively, one may expect that the form of the noise does not make a great difference, as we found in most of this work, but this is not always the case.
New dynamic equations for the overlaps and for the variance of the Gaussian noise were obtained in the form of discrete-time recursion relations for extensive loading of patterns. In the case of a Hebbian plus sequential learning rule with a macroscopic number of patterns $`p=\alpha N`$, the variance of the noise depends on a macroscopic number of correlations between overlaps with non-condensed patterns, $`M_N^\mu (l)M_N^\rho (l)`$. All these correlations could be relevant for the dynamics but when it comes to the fixed points (and stationary cycles) only a small number of them survives making the sequence processing a tractable and solvable problem, at least in the relevant cases we considered in this work. In all the cases we studied the correlations between overlaps with non-condensed patterns formed, practically, a finite and tractable set.
Explicit phase diagrams of stable states and regions of non-stationary solutions were obtained in this work that show the effects of stochastic noise due to a macroscopic number of learnt patterns, and different behaviors may be obtained depending on a variety of relevant parameters. We also considered, briefly, the optimal case where $`b=0.5`$ in the synapses that generate the noise. The reason why this yields a larger $`\alpha _c`$ for $`\nu =1`$ or zero is that the symmetric phase is considerably enhanced towards extreme values of $`\nu `$ combined with the fact that Hopfield-like retrieval in this model is, in general, more robust to stochastic noise than the stable symmetric states for small or large $`\nu `$.
Some general conclusions that may be drawn from the phase diagrams are the following. First, as expected, the retrieval quality of Hopfield-like and cyclic states is gradually reduced with an increase in the storage ratio $`\alpha `$. Less obvious, there is also a reduction in nearly single-pattern retrieval or in cycle retrieval as $`\nu `$ or $`1\nu `$ are decreased, respectively. The change may be either to a symmetric state or to a spin-glass state. Eventually, depending on the form of the noise due to the previously learnt patterns, the transition could be to non-stationary quasi-periodic states. The symmetric states appear as a locking of the sequential transitions by the static Hebbian part of the learning rule.
A further result of our work is that, for a Hopfield type initial condition $`m_1=1,m_2=m_3=m_4=0`$, the first-order phase boundaries in the ($`T,\nu `$) phase diagrams where the Hopfield-like phase ends is practically the same for small $`\alpha `$ and either Hebbian or complete noise, for both $`c=4`$ and for $`c=3`$. That is also the case for the ($`\alpha ,\nu `$) phase diagrams at $`T=0`$.
Turning now to predictions for somewhat larger, finite values of $`c`$, either even or odd for Hopfield type initial conditions, we expect that the non-stationary solutions that appear in the finite loading case for small $`T`$ CS92 should disappear as soon as $`\alpha `$ becomes non-zero, allowing for the presence of a symmetric phase down to $`T=0`$, leaving a phase diagram with essentially no q-p solutions.
It may be pointed out that there is a similarity between our ($`\alpha ,\nu `$) phase diagrams at $`T=0`$, with initial Hopfield type overlaps, and the corresponding phase diagram for pattern reconstruction and symmetric sequence processing referred to above obtained by means of statistical mechanics CT94 . The case of pure sequential processing is not considered in that work and, instead of a single-pattern retrieval region found there, we have here a Hopfield-like region (a difference already noted in the context of finite loading CS92 ). The boundaries of either of these regions have a similar dependence with the ratio $`\nu /(1\nu )`$. Moreover, there are similar regions of symmetric phases and, in contrast to our work, there is also a region of correlated states.
Based on the experience with the feed-forward layered network to process information with a Hebbian learning rule, one may expect qualitatively similar results to those obtained here to apply for a recurrent network with a Hebbian-plus-sequential learning rule. Most of the work reported here deals with the dynamics near stationary states. Since our recursion relations for the overlaps and for the noise are general for the layered network model, they may also be employed to study the transients of the dynamics, in particular to find out what makes the correlations between distant (in pattern space) high components of the overlaps vanish. One may also study the slowing down of the dynamics due to the macroscopic number of non-condensed patterns. Work along some of these lines is currently in progress.
As an extension of this work one may study the performance of this model with a number of sequences. There is recent work only with sequential learning, for an infinite number of limit cycles MKO03 . We point out that there is renewed interest in sequence learning from a biological point of view Ni03 and that asymmetric networks have been found to be computationally superior LD99 .
## VI Acknowledgments
The work of one of the authors (WKT) was financially supported, in part, by CNPq (Conselho Nacional de Desenvolvimento Científico e Tecnológico), Brazil. Grants from CNPq and FAPERGS (Fundação de Amparo à Pesquisa do Estado de Rio Grande do Sul), Brazil, to the same author are gratefully acknowledged. F. L. Metz acknowledges a fellowship from CNPq. |
warning/0507/quant-ph0507137.html | ar5iv | text | # Quantum Phase Gate Operation Based on Nonlinear Optics: Full Quantum Analysis
## Abstract
We present a full quantum treatment of a five-level atomic system coupled to two quantum and two classical light fields. The two quantum fields undergo a cross-phase modulation induced by electromagnetically induced transparency. The performance of this configuration as a two-qubit quantum phase gate for travelling single-photons is examined. A trade-off between the size of the conditional phase shift and the fidelity of the gate is found. Nonetheless, a satisfactory gate performance is still found to be possible in the transient regime, corresponding to a fast gate operation.
Single photons are natural candidates for the implementation of quantum information processing systems chuang95 . This is due to the photon’s robustness against decoherence and the availability of single-qubit operations. However, it is difficult to realize the necessary two-qubit operations since the interaction between photons is very small. A possible solution is the enhancement of photon-photon interaction either in cavity QED configurations turch or in dense atomic media exhibiting electromagnetically induced transparency (EIT) eit . In this latter case, optical nonlinearities can be produced when EIT is disturbed, either by introducing additional energy level(s) Schmidt96 ; Wang01 , or by mismatching the probe and control field frequencies Grangier98 ; Matsko03 .
In this letter, we address the feasibility of EIT-based systems for the implementation of a two-qubit quantum phase gate (QPG) for travelling single photons Lukin00 ; Ottaviani03 ; Masalas04 , by means of a *full quantum* treatment of the system dynamics. In a QPG, one qubit gets a phase conditional to the other qubit state according to the transformation Lloyd95 ; NielsenChuang $`|i_1|j_2\mathrm{exp}\left\{i\varphi _{ij}\right\}|i_1|j_2`$ where $`\{i,j\}=0,1`$ denote the logical qubit bases. This gate is universal when the conditional phase shift (CPS)
$$\varphi =\varphi _{11}+\varphi _{00}\varphi _{10}\varphi _{01},$$
(1)
is nonzero, and it is equivalent to a CNOT gate up to local unitary transformations when $`\varphi =\pi `$ Lloyd95 ; NielsenChuang . The existing literature focused only on the evaluation of the CPS and on the best conditions for achieving $`\varphi =\pi `$ Lukin00 ; Ottaviani03 ; Masalas04 , while the gate fidelity, which is the main quantity for estimating the efficiency of a gate, has been never evaluated. In this letter we calculate *both the fidelity and the CPS* of the QPG, enabling us to discover a general *trade–off* between a large CPS and a gate fidelity close to one, hindering the QPG operation. However, we show that this trade-off can be bypassed in the transient regime, which has never been considered before in EIT situations, still allowing a satisfactory gate performance.
The qubits are given by polarized single–photon wave packets with different frequencies, and the phase shifts $`\varphi _{ij}`$ are generated when these two pulses cross an atomic ensemble in a five-level “M” configuration (see Fig. 1). The population is assumed to be initially in the ground state $`|3`$. From this ground state, it could be excited by either the single–photon probe field, coupling to transition $`|3|2`$, or by the single–photon trigger field, coupling to transition $`|3|4`$. If the five levels are Zeeman sub-levels of an alkali atom, and both pulses have a sufficiently narrow bandwidth, the Zeeman splittings can be chosen so that the atomic medium is coupled only to a given circular polarization of either the probe or trigger field, while it is transparent for the orthogonally polarized mode, which crosses the gas undisturbed Ottaviani03 . In this way, the logical basis for each qubit practically coincides with the two lowest Fock states of the mode with the “right” polarization, $`|0_j`$ and $`|1_j`$ ($`j=p,t`$).
When the probe (trigger) is on two–photon resonance with the classical pump field with Rabi frequency $`\mathrm{\Omega }_1(\mathrm{\Omega }_4)`$, i.e. $`\delta _1=\delta _2(\delta _3=\delta _4)`$ (see Fig. 1 for a definition of the detunings), the system exhibits EIT for probe and trigger simultaneously. In fact, the scheme can be seen as formed by two adjacent $`\mathrm{\Lambda }`$ systems, perfectly symmetric between probe and trigger. A nonzero CPS occurs whenever a nonlinear cross-phase modulation (XPM) between probe and trigger is present. This cross-Kerr interaction takes place if the two-photon resonance condition is violated. For small frequency mismatch $`ϵ_{12}=\delta _1\delta _2`$ and $`ϵ_{34}=\delta _3\delta _4`$ (both chosen to be within the EIT window), absorption remains negligible and the cross-Kerr interaction between probe and trigger photons may be strong. The consequent CPS may become large, of the order of $`\pi `$, if the probe and trigger pulse simultaneously cross the atomic medium and interact for sufficient time. This is achieved when the group velocities of the two pulses are small and equal (to $`v_g`$, see Refs. Lukin00 ; Ottaviani03 ), so that the interaction time is given by $`t_{int}=L/v_g`$, $`L`$ being the length of the gas cell cigar . The inherent symmetry of this scheme guarantees perfect group velocity matching whenever $`\delta _1=\delta _4`$, $`\delta _2=\delta _3`$ and $`g_p/\mathrm{\Omega }_1=g_t/\mathrm{\Omega }_4`$, where $`g_j=\mu _j\sqrt{\omega _j/2\mathrm{}ϵ_0V}`$ ($`j=p,t`$) is the coupling constant between the probe (trigger) quantum mode with frequency $`\omega _j`$ and the corresponding transition with electric dipole moment $`\mu _j`$. These features are shared by all the proposals for an EIT-based, nonlinear two-qubit quantum gate Lukin00 ; Ottaviani03 . They essentially differ only in the way in which group velocity matching is achieved.
The scope of this paper is to find the ultimate *physical* limits imposed on QPG operations in systems with EIT-based optical nonlinearities. To this end, we neglect all the possible technical limitations and experimental imperfections. First, we assume perfect spatial mode matching between the input single-photon pulses entering the gas cell and the optical modes excited by the driven atomic medium, and which are determined by the geometrical properties of the gas cell and of the pump beams Duan02 . This allows us to describe the probe and trigger fields in terms of *single* travelling optical modes, with annihilation operators $`\widehat{a}_{p,t}`$. Next, we assume that the pulses are tailored in such a way that they simultaneously enter the gas cell and completely overlap with it during the interaction. This means that their length (compressed due to group velocity reduction) is of the order of the cell length $`L`$ and their beam waist is of the order of the cell radius. In this way, the two pulses interact with *all* $`N_a`$ atoms in the cell, and moreover one can ignore spatial aspects of pulse propagation. With these assumptions, and neglecting dipole-dipole interactions, the interaction picture Hamiltonian may be written as
$`H=\mathrm{}ϵ_{12}\widehat{S}_{11}+\mathrm{}\delta _2\widehat{S}_{22}+\mathrm{}\delta _3\widehat{S}_{44}+\mathrm{}ϵ_{34}\widehat{S}_{55}`$ (2)
$`+\mathrm{}\mathrm{\Omega }_1\sqrt{N_a}\left(\widehat{S}_{21}+\widehat{S}_{12}\right)+\mathrm{}g_p\sqrt{N_a}\left(\widehat{a}_p\widehat{S}_{23}+\widehat{S}_{32}\widehat{a}_p^{}\right)`$
$`+\mathrm{}g_t\sqrt{N_a}\left(\widehat{a}_t\widehat{S}_{43}+\widehat{S}_{34}\widehat{a}_t^{}\right)+\mathrm{}\mathrm{\Omega }_4\sqrt{N_a}\left(\widehat{S}_{45}+\widehat{S}_{54}\right),`$
where we have defined the collective atomic operators $`\widehat{S}_{kl}=_{i=1}^{N_a}\sigma _{kl}^i/\sqrt{N_a}`$, $`kl=1,\mathrm{},5`$, and $`\widehat{S}_{kk}=_i\sigma _{kk}^i`$, with $`\sigma _{kl}^i|k_il|`$ referring to the $`i`$th atom. Since the initial state
$`|\psi _{in}={\displaystyle \underset{i=1}{\overset{N_a}{}}}|3_i(c_{00}|0_p|0_t+c_{01}|0_p|1_t`$
$`+c_{10}|1_p|0_t+c_{11}|1_p|1_t)`$ (3)
contains at most two excitations, the time evolution driven by Eq. (2) is simple and restricted to a finite-dimensional Hilbert space involving few symmetric collective atomic states. In fact, each component of the initial state of Eq. (3) evolves independently in a different subspace. Defining $`|e_3^{(n_p,n_t)}=_{i=1}^{N_a}|3_i|n_p|n_t`$, the component with no photon in Eq. (3), $`|e_3^{(0,0)}`$, is an eigenstate of $`H`$ and does not evolve. The component $`|e_3^{(0,1)}`$ evolves in a three-dimensional Hilbert space spanned also by the two states $`|e_4^{(0,0)}`$ and $`|e_5^{(0,0)}`$, where we have defined, for $`r=1,2,4,5`$, the symmetric collective states
$$|e_r^{(n_p,n_t)}=\frac{1}{\sqrt{N_a}}\underset{i=1}{\overset{N_a}{}}|3_1,3_2,\mathrm{},r_i,\mathrm{},3_{N_a}|n_p|n_t.$$
(4)
In a similar fashion, the component with only one probe photon, $`|e_3^{(1,0)}`$, evolves in a three-dimensional Hilbert space spanned also by the two states $`|e_1^{(0,0)}`$ and $`|e_2^{(0,0)}`$. Finally, the component $`|e_3^{(1,1)}`$ evolves in the five dimensional subspace spanned also by the four collective states $`|e_1^{(0,1)}`$, $`|e_2^{(0,1)}`$, $`|e_4^{(1,0)}`$, and $`|e_5^{(1,0)}`$. What is relevant is that the dynamics remain simple and restricted within a finite-dimensional Hilbert space even when we include spontaneous emission, so that time evolution is described by the following master equation for the system density matrix $`\rho `$,
$$\dot{\rho }=\frac{i}{\mathrm{}}[H,\rho ]+\underset{kl}{}\frac{\gamma _{kl}}{2}\underset{j=1}{\overset{N_a}{}}\left(2\sigma _{kl}^j\rho \sigma _{kl}^j\sigma _{kl}^j\sigma _{kl}^j\rho \rho \sigma _{kl}^j\sigma _{kl}^j\right),$$
(5)
where $`\gamma _{kl}`$ denotes the decay rate from the excited states $`l=2,4`$ to the ground states $`k=1,3,5`$ dephasing . Spontaneous emission seems to complicate the system dynamics. However, the Hamiltonian evolution involves only the *singly excited* symmetric atomic states of Eq. (4). This means that these collective states decay with a rate equal to the single-atom decay rate $`\gamma _{kl}`$, and that spontaneous emission involves only a restricted number of additional collective atomic states in the dynamics. To state it in an equivalent way, the atomic medium behaves as an effective *single* $`5`$-level atom, with a collectively enhanced coupling with the optical modes $`g_j\sqrt{N_a}`$, but with the same single-atom decay rates $`\gamma _{kl}`$, Rabi frequencies $`\mathrm{\Omega }_i`$, and detunings $`\delta _i`$ (see also Ref. Duan01 ).
Spontaneous emission causes the four independent Hilbert subspaces corresponding to the four initial state components to become coupled. Moreover, the joint effect of the “cross” decay channels $`|4|1`$ and $`|2|5`$ together with the Hamiltonian dynamics couples the above-mentioned collective states with six new states, $`|e_1^{(1,0)}`$, $`|e_2^{(1,0)}`$, $`|e_3^{(2,0)}`$ (populated if $`\gamma _{41}0`$), and $`|e_5^{(0,1)}`$, $`|e_4^{(0,1)}`$, $`|e_3^{(0,2)}`$ (populated if $`\gamma _{25}0`$). Therefore Eq. (5) actually describes dynamics in a Hilbert space of dimension $`18`$, which we have numerically solved in order to establish the performance of the QPG.
This analysis allows us to fully characterize the QPG operation, by calculating *both* the CPS $`\varphi `$ of Eq. (1) and the fidelity of the gate, at variance with former treatments Lukin00 ; Ottaviani03 ; Masalas04 . The accumulated CPS as a function of $`t_{int}`$ is obtained by using the fact that the phase shifts $`\varphi _{ij}`$ of Eq. (1) are given by combinations of the phases of the off-diagonal matrix elements (in the Fock basis) of the reduced density matrix of the probe and trigger modes, $`\rho _f(t_{int})`$. The gate fidelity is given by NielsenChuang
$$(t_{int})=\sqrt{\overline{\psi _{id}(t_{int})\left|\rho _f(t_{int})\right|\psi _{id}(t_{int})}},$$
(6)
where $`|\psi _{id}(t_{int})=c_{00}\mathrm{exp}\{i\varphi _{00}(t_{int})\}|0_p,0_t+c_{01}\mathrm{exp}\{i\varphi _{01}(t_{int})\}|0_p,1_t+c_{10}\mathrm{exp}\{i\varphi _{10}(t_{int})\}|1_p,0_t+c_{11}\mathrm{exp}\{i\varphi _{11}(t_{int})\}|1_p,1_t`$ is the ideally evolved state from the initial condition (3), with phases $`\varphi _{ij}(t_{int})`$ evaluated from $`\rho _f(t_{int})`$ as discussed above. The overbar denotes the average over all initial states (i.e., over the $`c_{ij}`$, see Ref. Poyatos97 ). The above fidelity characterizes the performance of the QPG as a deterministic gate. However, one could also consider the QPG as a *probabilistic* gate, whose operation is considered only when the number of output photons is equal to the number of input photons. The performance of this probabilistic QPG could be experimentally studied by performing a conditional detection of the phase shifts, and it is characterized by the *conditional* fidelity $`^c(t_{int})`$, similar to that of Eq. (6), but with $`\rho _f(t_{int})`$ replaced by $`\rho _f^c(t_{int})=\mathrm{Tr}_{atom}\{|\psi _{nj}(t_{int})\psi _{nj}(t_{int})|\}/\psi _{nj}(t_{int})|\psi _{nj}(t_{int})`$, where $`|\psi _{nj}(t_{int})`$ is the (non-normalized) evolved atom-field state conditioned to the detection of no quantum jumps Carmichael93 , i.e., of no spontaneous emission.
The conditional fidelity is always larger than the unconditional one, but they become equal (and both approach $`1`$) for an ideal QPG in which the number of photons is conserved and all the atoms remain in state $`|3`$. This ideal condition is verified in the limit of large detunings $`\delta _j\gamma _{kj}`$ (to significantly suppress spontaneous emission) and very small couplings $`g_j\sqrt{N_a}\mathrm{\Omega }_j`$. In this limit, each component of the initial state of Eq. (3) practically coincides with the dark state of the four independent Hamiltonian dynamics discussed above. The four phase shifts $`\varphi _{ij}`$ can be evaluated as a fourth-order perturbation expansion of the corresponding eigenvalue, multiplied by $`t_{int}`$, obtaining the following CPS
$`\varphi ={\displaystyle \frac{g_p^2g_t^2N_a^2t_{int}}{(ϵ_{34}\delta _3\mathrm{\Omega }_4^2)(ϵ_{12}\delta _1\mathrm{\Omega }_1^2)}}`$ (7)
$`\times \left[{\displaystyle \frac{ϵ_{34}(ϵ_{12}^2+\mathrm{\Omega }_1^2)}{(ϵ_{12}\delta _1\mathrm{\Omega }_1^2)}}+{\displaystyle \frac{ϵ_{12}(ϵ_{34}^2+\mathrm{\Omega }_4^2)}{(ϵ_{34}\delta _3\mathrm{\Omega }_4^2)}}\right].`$
This prediction is verified by the numerical solution of Eq. (5) in the limit of large detunings and small couplings. However the resulting CPS is too small, even for very long interaction times (i.e., long gas cells): for example, for $`g_{p,t}\sqrt{N_a}=0.5`$ MHz, $`ϵ_{12,34}=1.9`$ MHz, $`\mathrm{\Omega }_{1,4}=65`$ MHz and $`\delta _{1,3}=1.9`$ GHz, we obtain a tiny CPS of only $`3\times 10^4`$ radians when $`t_{int}=10^4`$ s. This is not surprising because this limit corresponds to a dispersive regime far from EIT, and one has to explore the non-perturbative regime of larger couplings in order to exploit EIT and achieve a satisfactory QPG operation.
We have found good QPG performance for the following parameters, corresponding to a gas cell of $`N_a10^8`$ <sup>87</sup>Rb atoms: $`\gamma _{kl}=\gamma =2\pi \times 6`$ MHz, $`\delta _1=\delta _3=15\gamma `$, $`ϵ_{12}=ϵ_{34}=0.01\gamma `$, $`g_p=g_t=0.0022\gamma `$, $`\mathrm{\Omega }_1=\mathrm{\Omega }_4=4\gamma `$. The results are shown in Figs. 2 and 3, where we see that a CPS of $`\pi `$ radians is obtained in the transient regime for $`t_{int}0.4/\gamma 10`$ ns, corresponding to a *fast operation* of the gate. At the same interaction time, the unconditional gate fidelity (Fig. 3, full line) is about $`94\%`$, while the conditional gate fidelity reaches the value of $`99\%`$ (Fig. 3, dashed line), in correspondence with a success probability of the gate equal to $`0.94`$. The probe and trigger group velocity is $`v_g3\times 10^6`$ ms<sup>-1</sup>, yielding a gas cell length $`L=v_gt_{int}3.1`$ cm. The value of $`g_j`$ yields an interaction volume $`V210^3`$ cm<sup>3</sup>, corresponding to a gas cell diameter of about $`330`$ $`\mu `$m and to an atomic density $`N_a/V510^{10}`$ cm<sup>-3</sup>.
EIT is a stationary phenomenon, while the above results are obtained in the transient regime where $`\gamma t_{int}<1`$. However we can attribute these results to a sort of “non-stationary”, EIT process. This is suggested by the reduction of $`v_g`$ (by a factor $`100`$), which has been estimated by evaluating the “instantaneous” susceptibility from the reduced atomic density matrix given by Eq. (5) and then averaging over the time interval between $`0`$ and $`t_{int}`$. This “non-stationary” $`v_g`$ is one order of magnitude smaller than the conventional $`v_g`$ obtained from the steady-state susceptibility corresponding to the above parameters. The presence of a moderate EIT process is also confirmed by the fact that in a numerical study of the three-level ladder atomic scheme, yielding XPM without EIT Schmidt96 , we have found a slower accumulation of the CPS and a smaller conditional fidelity ($`78\%`$) for a corresponding set of parameters.
Our study of Eq. (5) also shows that it is not possible to achieve a comparable QPG performance in the steady-state regime $`\gamma t_{int}1`$. In fact, we have found at best a CPS of $`\pi `$ in correspondence with fidelities $`(t_{int})`$ and $`^c(t_{int})`$ equal to $`77\%`$ and $`83\%`$, respectively. This is due to the general presence of a trade-off between the size of the CPS and of the gate fidelity. In fact, we have seen that both gate fidelities approach $`1`$ in the small perturbation limit, but with a CPS which becomes appreciable only for unrealistically long gas cells. A larger CPS requires a larger ratio $`g_j\sqrt{N_a}/\mathrm{\Omega }_j`$. This condition however increases the population of atomic states $`|1`$ and $`|5`$ at the expense of the initial atomic state $`|3`$, unavoidably decreasing the gate fidelity. Similar conclusions hold for other options, such as increased detunings $`\delta _j`$, or adjusting two-photon detunings $`ϵ_{ij}`$. This trade-off is present also at large ratios $`g_j\sqrt{N_a}/\mathrm{\Omega }_j`$ in the transient regime, where, however, it may be less effective. In fact, in this case one has significant oscillations of the atomic populations, but it is possible to find appropriate interaction times $`t_{int}`$ at which high fidelities are achieved (see Fig. 3), simultaneously with a CPS of about $`\pi `$.
In conclusion, our study shows that the implementation of efficient EIT-based nonlinear two-qubit gates for travelling single-photons is possible. In fact, even if there is a trade-off between the size of the CPS and the fidelity of the gate in the stationary regime, it is possible to have a satisfactory gate performance in the transient regime, where a fast gate operation and fidelities equal to $`0.99`$ are achievable. The experimental realization might be challenging, but the implementation of this quasi-deterministic two-qubit gate would be extremely useful, not only for quantum computation, but also for quantum communication purposes: for example, a QPG allows a complete Bell-state discrimination for single-photon polarization qubits Vitali00 . We expect that these considerations apply to all EIT-based crossed-Kerr schemes Lukin00 ; Ottaviani03 , regardless of the specific level scheme considered. Finally, we note that our analysis does not apply to situations where the nonlinearity comes from independent processes such as atomic collisions or dipole-dipole interactions Masalas04 .
We acknowledge enlightening discussions with G. Di Giuseppe. |
warning/0507/astro-ph0507023.html | ar5iv | text | # New optical polarization measurements of quasi-stellar objects. The data Based on observations collected at the European Southern Observatory (ESO, La Silla and Paranal), Table 4 is only available in electronic form at CDS via anonymous ftp to cdsarc.u-strasbg.fr (130.79.128.5) or http://cdsweb.u-strasbg.fr/cgi-bin/qcat?J/A+A/433/757
## 1 Introduction
Based on a large sample of quasi-stellar objects with measured optical polarization, Hutsemékers (HUT98a (1998)) discovered that there exists regions in the sky where the QSO polarization vectors appear concentrated along preferential directions, on scales up to $``$ 1 Gpc. New data enabled Hutsemékers & Lamy (HUT01 (2001)) to confirm this effect. In order to obtain an accurate and complete description of this intriguing phenomenon, new QSO polarization measurements are badly needed. The present paper provides a new set of polarimetric data for 94 QSOs located in the North Galactic Cap and for 109 QSOs located in the South Galactic Pole region, with details on the observations, data reduction, and measurements. The comprehensive analysis and the interpretation of the full sample will be reported elsewhere.
This sample may be used for a variety of other studies such as investigating the relation between the optical polarization properties of QSOs and their optical spectra (e.g. Stockman et al. STO84 (1984)) or their radio properties (e.g. Berriman et al. BER90 (1990), Rusk RUS90 (1990), Visvanathan & Wills VIS98 (1998)). It also enables one to study the polarization properties of sub-classes of QSOs such as Broad Absorption Line (BAL) quasi-stellar objects (e.g. Hutsemékers et al. HUT98b (1998), Schmidt & Hines SCH99 (1999), Lamy & Hutsemékers LAM04 (2004)) or near-infrared selected QSOs (e.g. Smith et al. SMI02 (2002)).
## 2 The observations
The polarimetric observations were carried out during 5 runs at the European Southern Observatory, La Silla, in August 2000, March 2002, May 2002, August 2003 and October 2003, using the 3.6m telescope equipped with EFOSC2. Two additional objects were observed on April 21, 2002 with EFOSC2 and three objects on February 25, 2003 in service mode with the VLT UT1 equipped with the FORS1 camera. The CCD#40 mounted on EFOSC2 is a 2k$`\times `$2k CCD with a pixel size of 15 $`\mu `$m corresponding to 0.158″ on the sky in the 1$`\times `$1 binning mode. The standard resolution mode for the 2k$`\times `$2k Tektronix CCD detector of FORS1 has a pixel size of 24 $`\mu `$m corresponding to 0.2″on the sky.
With both the EFOSC2 and FORS1 instruments, linear polarimetry is performed by inserting in the parallel beam a Wollaston prism which splits the incoming light rays into two orthogonally polarized beams. Each object in the field has therefore two orthogonally polarized images on the CCD detector, separated by 20$`\mathrm{}`$ for EFOSC2 and 22$`\mathrm{}`$ for FORS1. To avoid image overlapping, one puts at the telescope focal plane a special mask made of alternating transparent and opaque parallel strips whose width corresponds to the splitting. The object is positioned at the center of a transparent strip which is imaged on a region of the CCD free of defects. The final CCD image then consists of alternate orthogonally polarized strips of the sky, two of them containing the polarized images of the object itself (di Serego Alighieri SER89 (1989), SER97 (1997); Lamy & Hutsemékers Lam99 (1999), hereafter Lam99). Note that the polarization measurements do not depend on variable transparency or seeing since the two orthogonally polarized images of the object are simultaneously recorded.
In order to derive the two normalized Stokes parameters $`q`$ and $`u`$ which characterize the linear polarization, frames must be obtained with at least two different orientations of the Wollaston. In practice, the Wollaston is not rotated but a half-wave plate (HWP) is inserted in the optical path and four frames are obtained with the HWP at 4 different position angles (0$`\mathrm{°}`$, 22.5$`\mathrm{°}`$, 45$`\mathrm{°}`$, and 67.5$`\mathrm{°}`$). Even if only two different orientations of the HWP are sufficient to retrieve the linear polarization (Melnick et al. MEL89 (1989)), two additional orientations make it possible to remove most of the instrumental polarization (di Serego Alighieri SER89 (1989)).
Targets were selected from the Véron catalogue (Véron-Cetty & Véron VER01 (2001)), from the Sloan Digital Sky Survey Early Data Release (Schneider et al. SCH02 (2002, 2003), Reichard et al. REI03 (2003)), from Becker et al. (BEC00 (2000), BEC02 (2002)), Menou et al. (MEN01 (2001)), Barkhouse & Hall (BAR01 (2001)), Hall et al. (HAL02 (2002)) and Smith et al. (SMI02 (2002)), mostly according to their position on the sky i.e. with right ascensions and declinations corresponding to the region of polarization vector alignments defined in Hutsemékers (HUT98a (1998)). Bright objects were preferred, as well as BAL, radio-loud and red QSOs which are usually more polarized.
All but two observations were obtained through the Bessel V filter with typical exposure times between 1 and 10 minutes per frame. Polarized and unpolarized standard stars were observed in the Bessel V, R, and gunn $`i`$ filters in order to unambiguously fix the zero-point of the polarization position angle and to check the whole observing and reduction process. In August 2000 and October 2003 the sky was clear and the seeing around 1$`\stackrel{}{.}`$0, while in August 2003 conditions were not as good with cirrus and seeing around 1$`\stackrel{}{.}`$5. During the other runs, the seeing was always between 1$`\stackrel{}{.}`$1 and 1$`\stackrel{}{.}`$5, and the sky covered at worst with thin cirrus. A few observations were obtained with a high Moon fraction illumination ($`>0.7`$) and are of lower quality. Indeed, high levels of sky background induce larger errors in the sky subtraction process and the subsequent polarization measurements. This is especially relevant when the polarization of the target is low.
## 3 Data reduction
The $`q`$ and $`u`$ normalized Stokes parameters are computed from the measurement of the integrated intensity of the orthogonally polarized upper and lower images of the object, for the 4 different orientations of the HWP. They are calculated with respect to the instrumental reference frame using the following formulae:
$`q`$ $`=`$ $`{\displaystyle \frac{R_q1}{R_q+1}}\text{where}R_q^2={\displaystyle \frac{I_0^\mathrm{u}/I_0^\mathrm{l}}{I_{45}^\mathrm{u}/I_{45}^\mathrm{l}}},`$
$`u`$ $`=`$ $`{\displaystyle \frac{R_u1}{R_u+1}}\text{where}R_u^2={\displaystyle \frac{I_{22.5}^\mathrm{u}/I_{22.5}^\mathrm{l}}{I_{67.5}^\mathrm{u}/I_{67.5}^\mathrm{l}}},`$
where $`I^\mathrm{u}`$ and $`I^\mathrm{l}`$ respectively refer to the intensities integrated over the upper and lower orthogonally polarized images of the object. The combination of four frames obtained with different HWP orientations not only removes most of the instrumental polarization, but is also essential for correcting the effects of image distortions introduced by the HWP (Lam99).
In order to measure levels of polarization as small as 0.6 % with 0.2 % uncertainty, it is mandatory to achieve photometry with a very high accuracy. For this purpose, the data were first corrected for bias and flat-fielded. The photometric measurements for each image were done using the MIDAS procedure developed by Lam99. The different steps of this procedure are the following : (1) several regions of the background close to the target are interactively chosen; a plane is fitted to their mean values and subtracted from each image individually; (2) the position of the object in each strip is measured at subpixel precision by fitting a 2D gaussian profile. The flux is subsequently integrated in circles centered at the fitted positions; (3) the Stokes parameters are then computed for any reasonable value of the aperture radius. Since they are found to be stable against radius variation, it was decided to always measure them inside a fixed aperture radius of $`3.0\times [(2\mathrm{ln}2)^{1/2}`$ HWHM\] <sup>1</sup><sup>1</sup>1For He 1304-1157, the radius was $`2.5\times [(2\mathrm{ln}2)^{1/2}`$ HWHM\] due to the presence of a cosmic-ray hit at larger radii. where HWHM represents the mean half-width at half-maximum of the gaussian profile. This empirical choice, seeing independent, is unsensitive to image distorsions (Lam99).
The uncertainties $`\sigma _q`$ and $`\sigma _u`$ on the normalized Stokes $`q`$ and $`u`$ are evaluated by computing the errors on the intensities $`I^\mathrm{u}`$ and $`I^\mathrm{l}`$ from the read-out noise and from the photon noise in the object and the sky background (after converting the counts in electrons), and by propagating these errors. Uncertainties are around 0.15% for both $`q`$ and $`u`$. For a few faint objects listed in Table 1 we were not able to derive reliable measurements, namely due to a higher than usual sky background. These measurements are rejected from the sample presented in Table 4.
A zero-point angle offset correction, filter dependent, is then applied to the QSO normalized Stokes parameters $`q`$ and $`u`$ in order to convert the polarization angle measured in the instrumental reference frame into the equatorial reference frame. This angle offset is determined from polarized standard stars observed each night and listed in Table 2. These stars have been selected to have polarization angles distributed in the full \[0°, 180°\] range. For all stars observed during a given run (and between the runs themselves), the values of the angle offset do agree within 1$`\mathrm{°}`$ standard deviation <sup>2</sup><sup>2</sup>2Except the measurements for HD251204 which disagree in both polarization degree and position angle from tabulated data, possibly indicating polarization variability (see also Weitenbeck WEI99 (1999)).
We have also observed several unpolarized standard stars in the V, R, and $`i`$ filters (Table 2). For these stars we measure $`p=0.12\pm 0.05`$%, $`p=0.08\pm 0.04`$% and $`p=0.11\pm 0.04`$% in August 2000, August 2003 and October 2003, respectively, indicating that the residual instrumental polarization is small, a result in agreement with the expectation that most of the instrumental polarization is removed by the observing procedure. No difference between the three filters has been noticed.
Since on most CCD frames field stars are simultaneously recorded, one can use them to estimate the residual instrumental polarization and/or interstellar polarization. While a frame-by-frame correction of the QSO Stokes parameters is in principle possible, it is nevertheless hazardous since we are never sure that the polarization of field stars correctly represents the interstellar polarization which could affect distant QSOs. For example, we have found stars located in the same field with significantly different polarization degrees and angles. Also, the field stars are often fainter than the QSO such that a frame-by-frame correction would introduce uncertainties on the QSO polarization larger than the correction itself. We then compute the weighted average ($`\overline{q}_{}`$ and $`\overline{u}_{}`$) and dispersion ($`\overline{\sigma }_{}`$) of the normalized Stokes parameters of field stars considering the $`n_{}`$ frames with suitable stars obtained during a given run. These quantities are given in Table 3. The small values and dispersions of the residual polarization confirm the small level of uncorrected instrumental polarization. They also indicate that, on average, the interstellar polarization is small, in agreement with the fact that all objects in the sample are at high galactic latitudes<sup>3</sup><sup>3</sup>3Except FIRST J0809+2753 with $`|b_{\mathrm{II}}|=28.33\mathrm{°}`$ ($`|b_{\mathrm{II}}|30\mathrm{°}`$). To minimize the systematic errors, this residual polarization is conservatively taken into account by subtracting the systematic $`\overline{q}_{}`$ and $`\overline{u}_{}`$ from the measured QSO $`q`$ and $`u`$, and by adding quadratically the errors. Since $`\overline{q}_{}`$ and $`\overline{u}_{}`$ are nearly identical in March and May 2002, only the mean values $`\overline{q}_{}=0.06`$%, $`\overline{u}_{}=+0.06`$% and $`\overline{\sigma }_{}=0.12`$% are used. For those objects observed with FORS1 in February 2003, no correction is done, while the correction used for the March and May 2002 runs is applied to the April 2002 data, obtained with the same instrumental setup.
In order to better understand the nature and the effect of this correction, we illustrate in Fig. 1 (bottom panel) the distribution of the polarization position angles we have measured for field stars<sup>4</sup><sup>4</sup>4We consider a single star per frame/field. In some cases this star is made up of the combination of several fainter stars from a given frame. at high galactic latitude ($`|b_{\mathrm{II}}|30^{}`$). Polarization data reported in Lamy & Hutsemékers (Lam00 (2000), hereafter Lam00) and obtained with the same instrumentation are also included. After removing bad quality measurements and redundancies, this leads to a total sample of 204 field star measurements at $`|b_{\mathrm{II}}|30^{}`$, of which about half have polarization angles with $`\sigma _\theta 14^{}`$. The polarization angles of the stars from the Heiles catalogue (Heiles HEI00 (2000)) are illustrated in the top panel of Fig. 1. The distributions are very similar to ours<sup>5</sup><sup>5</sup>5There is also a good agreement between the polarization degrees (typically around 0.2–0.3% , cf. Fig. 2), provided that one considers distant stars in the Heiles catalogue, i.e. stars at distances $`100200`$ pc., including definite concentrations of the polarization angles around two main directions: $``$70 towards the NGP and $``$135 towards the SGP. The existence of these two major directions in the interstellar polarization towards the North and the South Galactic Poles has also been reported by Berdyugin et al. (BER04 (2004)), considering distant stars at high galactic latitudes. This similarity in the polarization angle distributions suggests that a significant part of the polarization we measure for field stars is interstellar in origin. If we average the values of the residual polarization for the NGP and the SGP separately, we get from Table 3 and Lam00, $`\overline{q}_{}`$ = $`0.05`$% and $`\overline{u}_{}`$ = $`+0.08`$% for the NGP, and $`\overline{q}_{}`$ = $`0.01`$% and $`\overline{u}_{}`$ = $`0.09`$% for the SGP, which correspond to the polarization angles $`\overline{\theta }_{}`$ = 61$`\mathrm{°}`$ and $`\overline{\theta }_{}`$ = 133$`\mathrm{°}`$, respectively, in agreement with the trend seen in Fig. 1. Note that a small contribution due to instrumental polarization cannot be excluded given the differences in the values of $`\overline{q}_{}`$ and $`\overline{u}_{}`$ for the various runs (Table 3, Lam00). Although small, the correction by the systematic $`\overline{q}_{}`$ and $`\overline{u}_{}`$<sup>6</sup><sup>6</sup>6Different $`\overline{q}_{}`$ and $`\overline{u}_{}`$ are used for each observing run. Ideally, as suggested by the results in Fig. 1, one should have also computed $`\overline{q}_{}`$ and $`\overline{u}_{}`$ for $`b_{\mathrm{II}}30^{}`$ and $`b_{\mathrm{II}}30^{}`$ separately. However this makes little difference since nearly all objects observed in a given run were either at $`b_{\mathrm{II}}30^{}`$ or at $`b_{\mathrm{II}}30^{}`$. then removes the bias in the distribution of the polarization angles observed in Fig. 1, at least from the statistical, systematic, point of view.
## 4 The results
Table 4 summarizes the measurements. The first eight columns give the QSO name, type, equatorial coordinates (J2000) and redshift $`z`$, the date of observation and the normalized Stokes parameters $`q`$ and $`u`$ corrected for the systematic residual polarization given in Table 3. The normalized Stokes parameters are given in the equatorial reference frame. The QSO name is the one used in the Véron catalogue (Véron-Cetty and Véron VER01 (2001)) when given, and in the NASA/IPAC Extragalactic Database (NED) otherwise. The name is followed by the object classification. The following notation has been adopted : R if known radio emitter, B if known BAL, RB if both, and U otherwise.
Then, from these values, the polarization degree is evaluated with $`p=(q^2+u^2)^{1/2}`$ and the associated error with $`\sigma _p=(\sigma ^2+\overline{\sigma }_{}^2)^{1/2}`$ where $`\sigma \sigma _q\sigma _u`$. In addition, $`p`$ must be corrected for the statistical bias inherent to the fact that $`p`$ is always a positive quantity. For this purpose, we used the Wardle & Kronberg estimator (WAR74 (1974)) which was found to be a reasonably good estimator of the true polarization degree (Simmons & Stewart SIM85 (1985)). The debiased value $`p_0`$ of the polarization degree is reported in column 11. The polarization position angle $`\theta `$ is obtained by solving the equations $`q=p\mathrm{cos}2\theta `$ and $`u=p\mathrm{sin}2\theta `$. The uncertainty of the polarization position angle $`\theta `$ is estimated from the standard Serkowski (SER62 (1962)) formula where the debiased value $`p_0`$ is conservatively used instead of $`p`$, i.e. $`\sigma _\theta =28\stackrel{}{.}65\sigma _p/p_0`$. Note that due to the HWP chromatism over the V band, an additional error $`23\mathrm{°}`$ on $`\theta `$ should be accounted for (cf. the wavelength dependence of the polarization angle offset in di Serego Alighieri SER97 (1997)).
Fig. 2 illustrates the distribution of the field star polarization measured on the QSO frames<sup>7</sup><sup>7</sup>7We were also able to measure the polarization of 10 field galaxies with a reasonable accuracy. Within the uncertainties, their polarization (both in degree and angle) does not differ from the polarization of the field stars. This suggests that the contamination by interstellar polarization in our Galaxy is not significantly higher for objects at extragalactic distances., for the NGP and the SGP regions separately. For both regions of the sky, the polarization degree is small, most often $``$ 0.3%. The distributions are very similar although stars with polarization between 0.3% and 0.5% seem slightly more numerous in the SGP. The median polarization is 0.11% in the NGP region and 0.15% in the SGP region. The overall distribution suggests that virtually every quasi-stellar object with a polarization higher than 0.6% is intrinsically polarized, in agreement with previous studies (Berriman et al. BER90 (1990), Lam00). In only five cases (flagged in Table 4), the QSO polarization is both significant ($`0.6`$ %) and comparable in degree and angle to the polarization of field stars, suggesting a probable contamination. Four objects with marginal contamination are also indicated in that Table.
### 4.1 Testing for possible biases in the data
Since the goal of our polarization measurements is to study concentrations of QSO polarization position angles along preferred directions, it is important to verify that the distribution of polarization angles is not significantly contaminated by either the instrumental polarization or the interstellar polarization in our Galaxy.
#### 4.1.1 Instrumental polarization
To verify that the polarization angles are not affected by an instrumental bias, a first test was done using standard stars with intrinsic polarization angles distributed in the full \[0°, 180°\] range (Sect. 2 and Lam00). The excellent agreement –most often within 1$`\mathrm{°}`$– between the polarization angles we measure and those values published in the literature demonstrates the absence of such a bias, at least for highly polarized objects.
In order to perform a similar test at lower polarization levels, typically around 1%, we have measured the polarization of a sample of 13 QSOs previously observed by Berriman et al. (BER90 (1990)), Schmidt & Hines (SCH99 (1999)) and Smith et al. (SMI02 (2002)) using different telescopes and instruments. The targets were selected to have polarization angles distributed in the full \[0°, 180°\] range (Table 6). Fig. 3 shows the observed polarization angle $`\theta `$ (from Table 4) versus the published one $`\theta _l`$ (from Table 6). A good overall agreement is observed, suggesting the absence of an instrumental bias. However, the correlation is not as good as one could have expected given the formal uncertainties. This is due to the fact that we compare unfiltered, white light, measurements to V-band polarization angles (with some additional noise from possible time variability). Indeed, while basically constant with wavelength, a slight dependence of the polarization angle on wavelength (i.e. about ten degrees between the blue and red parts of the spectrum) is often observed in low polarization QSOs (Webb et al. WEB93 (1993), Antonucci et al. ANT96 (1996)). For two objects, V-band polarization angle measurements are available in the literature, $`\theta `$ = 63 $`\pm `$ 3° for PG~1004+130 (Webb et al. WEB93 (1993)) and $`\theta `$ = 23 $`\pm `$ 2° for 3C~323.1 (Schmidt & Smith SCH00 (2000)). They are in much better agreement with our V-band measurements than with the white light data. A comparison of polarization measurements obtained in the same filter is then needed to test more accurately for possible instrumental contamination at low polarization levels.
#### 4.1.2 Interstellar polarization
In order to estimate the effect of the interstellar polarization on significantly polarized ($`p0.6`$%) objects, and namely whether it can introduce a bias in the distribution of the QSO polarization angles, we subtract frame by frame (when possible) the Stokes parameters of the field stars from the QSO ones, both uncorrected for the systematic $`\overline{q}_{}`$ and $`\overline{u}_{}`$. New QSO polarization degrees ($`p^{}`$) and angles ($`\theta ^{}`$) are then derived. With such a test, we implicitely assume that the polarization of field stars correctly represents the interstellar polarization affecting the QSOs.
In Fig. 4 & 5 we compare the new polarization angle $`\theta ^{}`$ to the uncorrected one $`\theta _0`$, for the NGP and the SGP regions separately. Only significantly polarized QSOs are considered, i.e. those objects with a polarization degree $`0.6\%`$. The distribution of the final polarization angle $`\theta `$ given in Table 4 is also illustrated. Although a systematic effect may be noticed, it is small enough to only slightly modify the distribution of the QSO polarization angles. This is due to the fact that the field star polarization is most often small (Fig. 2) such that only a few significantly polarized QSOs are affected (and, as done in Table 4, the most discordant objects may be flagged as contaminated).
This test demonstrates that the distribution of the polarization angles of polarized ($`p0.6`$%) QSOs is not significantly contaminated by the interstellar polarization, and more particularly after subtracting the systematic correction. But this conclusion is only valid if the interstellar polarization which affects the QSOs is not significantly larger than the polarization measured from field stars.
### 4.2 Time variability of the polarization
For some targets of our sample, several polarimetric measurements do exist. MARK~877, the BAL QSO Q2208-1720, and the binary QSO PKS~1145-071 A&B have been observed at two epochs (March and May 2002) and do not show any evidence for significant polarization variability. Also, the measurements for MARK~877 and the BAL QSO J2359-12 are in excellent agreement with the values reported by Berriman et al. (BER90 (1990)) and Brotherton et al. (BRO01 (2001)), respectively. The four radio-emitters PKS 2203-215, PKS 2204-54, PKS 2240-260 and PKS~1136-13 were previously observed in white light by Fugmann & Meisenheimer (FUG88 (1988)) and Impey & Tapia (IMP90 (1990)). Within the uncertainties and given the different wavelength coverages, we find a good agreement with our polarimetric measurements, i.e. no evidence for a significant variability, including for the highly polarized quasar PKS 2240-260.
On the contrary, the polarization level of PKS~1222+037 grew up to 2.4 % between March and May 2002, indicating that this object might be variable. Our observations also suggest the variability (in both polarization level and angle) of the BL Lac candidate PKS~1216-010 (Londish et al. LON02 (2002)). Indeed, Visvanathan & Wills (VIS98 (1998)) report for this object a polarization level $`p=6.9\pm 0.8`$% and a polarization angle $`\theta =8\pm 3.3\mathrm{°}`$ which are significantly different from our values ($`p_0=11.1\pm 0.19`$% and $`\theta =100\pm 0.5\mathrm{°}`$). Such a variability is common among highly polarized quasars and BL Lac objects (see e.g. Rieke et al. RIE77 (1977), Valtaoja et al. VAL91 (1991)).
## 5 Conclusions
New polarization measurements have been obtained for a sample of 203 quasi-stellar objects located in the NGP and the SGP regions with a final median uncertainty $``$ 0.25% on the polarization degree. 184 measurements are first time measurements. Among these ones, half of the 42 BAL QSOs show a level of polarization $`>`$ 1% and 12 QSOs have a polarization level higher than 3% (including the BAL QSO SDSS~J1409+0048).
Based on previous measurements found in the literature, we find evidence for a variation of both the polarization level and the polarization angle of the BL Lac candidate PKS~1216-010. We also report a significant variability in a two month period of the polarization of PKS~1222+037. Such a variability is not surprising since this object is reported to have a flat radio spectrum (Teraesranta et al. TER01 (2001)) which means, in unified schemes, a pole-on view of the AGN with a relativistically beamed polarized emission (Urry & Padovani URR95 (1995), Rusk RUS90 (1990), Antonucci & Ulvestad ANT85 (1985)).
The pair of quasi-stellar objects PKS~1145-071 A&B has been observed at two different epochs. They have polarization levels $`>0.6`$% and different polarization angles.
This sample also includes the gravitationally lensed QSO J11319-1231 (Sluse et al. SLU03 (2003)). The lensed nature of this object has been serendipitously unveiled on the polarimetric images obtained during the May 2002 observing run. The global polarization level of this object is smaller than 0.3%, confirming that gravitationally lensed QSOs are not more polarized than other quasi-stellar objects (cf. Hutsemékers et al. HUT98b (1998) and Lam00).
We have also shown that, if the effect of the interstellar polarization in our Galaxy onto distant objects may be adequately represented by the polarization measured from field stars, the significantly polarized ($`p0.6`$%) QSOs show little contamination in the distribution of their polarization angles.
###### Acknowledgements.
Dominique Sluse acknowledges support from an ESO studentship in Santiago and PRODEX (Gravitational lens studies with HST). Hervé Lamy would like to thank Prof. J. Lemaire and BIRA-IASB for giving him the opportunity to observe in La Silla in March 2002. Hernan Quintana acknowledges partial support from the FONDAP Centro de Astrophysics. This research has made use of the NASA/IPAC Extragalactic Database (NED), which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration. |
warning/0507/quant-ph0507130.html | ar5iv | text | # Extremal quantum cloning machines
## I Introduction
The impossibility of preparing several exact copies of an unknown quantum state, encapsulated by the no-cloning theorem no-cloning , is one of the most remarkable features of quantum mechanics. In addition to being of fundamental interest, it is also a pivotal ingredient in many practical applications, first among all quantum cryptography, where the impossibility of perfect cloning crucially poses limitations to eavesdropping.
From the discovery of the no-cloning theorem to now, a main research focus in the literature has been to find the best approximation of ideal quantum cloning with physical transformations allowed by quantum mechanics. Many relevant cases have been studied, and, depending on the set of states to be cloned, different optimal machines have been found BuzHill ; GisMass ; Wern ; CerIpe ; CerIb ; PhaseDM . In particular, much attention has been devoted to the situation in which the set of states to be cloned is invariant under a group of unitary transformations, the so-called *group covariant cloning* covar .
Despite the variety of cloning transformations that are known today, it is remarkable that the overwhelming majority of optimal covariant cloning machines share some common features, which relate their structure to a particular superposition of double-Bell states. This observation, which was originally formulated in an ansatz cerfansatz1 ; cerfansatz2 , has since then been often exploited to find optimal cloners along with their physical realizations (see e.g. CBKG ; qutrit ; PhaseCL ). Although the double-Bell ansatz has been shown to be correct in many cases, no general proof has been provided yet of its validity, and the common features of these optimal cloning machines are still just a surprising coincidence.
The aim of this paper is to provide a formal proof of this double-Bell ansatz in a covariant context, analyzing the physical meaning of the related implicit assumptions. This analysis a posteriori explains in a general way the appearance of double-Bell states in the optimal one-to-two covariant cloners, and also allows us to connect several cloning problems (e.g. the cloning of the four states involved in BB84 to the phase-covariant cloning of equatorial states).
In Section II, we set the problem of cloning an invariant set of states in the language of quantum operations, and define the covariance and strong covariance conditions. In Section III, we characterize the set of extremal covariant cloners, and show that it includes the set of strongly-covariant cloners. In Section IV, we analyze the special case of covariant cloners under the discrete Weyl-Heisenberg group, and show that all extremal covariant cloners are then necessarily also strongly covariant. This result is shown to imply the double-Bell ansatz, which is then used to derive the optimal cloners in various settings for qubits, $`d`$-dimensional, or infinite-dimensional states. Finally, the conclusions are drawn in Section V.
## II Cloning as a quantum operation
### II.1 Cloning an invariant set of states
Consider a machine $``$ that takes states in the Hilbert space $``$ of a quantum system to states in $``$. The task of the cloning machine is to provide two approximate copies of a state picked up from a given set of density matrices $`𝖲()`$ which is invariant under the action of some group of symmetry transformations. The action of the group—call it $`𝐆`$—is specified by a unitary representation $`\{U_g|g𝐆\}`$, and the set of states $`𝖲`$ enjoys the invariance property
$$U_g𝖲U_g^{}=𝖲g𝐆,$$
(1)
where $`U_g𝖲U_g^{}=\{U_g\rho U_g^{}|\rho 𝖲\}`$. It is important to stress that here, in contrast to the usual definition, we do not require the set $`𝖲`$ to be the group orbit of a fixed input state $`\rho _0𝖲`$, that is $`𝖲=\{U_g\rho _0U_g^{}|g𝐆\}`$. In fact, in what follows, the sole invariance of the set $`𝖲`$ will be sufficient.
The quality of the cloning machine is judged by introducing a figure of merit, usually the Uhlmann fidelity Uhlmann , which measures how close the joint output state $`(\rho )`$ is to two exact copies of the input state $`\rho `$. Sometimes, instead, it is more interesting to evaluate the single-clone fidelity, which measures how close the state of each clone is to the input state $`\rho `$. The results we are going to present hold for both kinds of fidelity and, more generally, for any figure of merit $`F[\rho ,(\rho )]`$ satisfying the invariance property
$$F[U_g\rho U_g^{},U_g^2(\rho )U_g^2]=F[\rho ,(\rho )],$$
(2)
for any $`g𝐆`$.
In this setting, the optimization problem is to maximize the average value of the figure of merit,
$$F=_S\mathrm{d}\mu (x)F[\rho _x,(\rho _x)],$$
(3)
where $`x`$ parametrizes the input states and $`\mathrm{d}\mu (x)`$ is an invariant probability distribution over the set of input states, i.e.,
$$\mathrm{d}\mu (gx)=\mathrm{d}\mu (x)g𝐆,x𝖲.$$
(4)
### II.2 Covariance condition
As a consequence of the invariance of the set of input states (1), of the figure of merit (2), and of the probability distribution (4), there is no loss of generality in assuming the cloning machine $``$ to be *covariant*, that is
$$(U_g\rho U_g^{})=U_g^2(\rho )U_g^2g𝐆,\rho .$$
(5)
In fact, for any non-covariant cloning machine $`𝒩`$, there is always a covariant one which has the same average fidelity, namely $`=\mathrm{d}gU_g^2𝒩(U_g\rho U_g^{})U_g^2`$, where $`\mathrm{d}g`$ is the normalized Haar measure on the group.
A convenient tool for the study of optimal cloning is the formalism of quantum operations (QO). A cloning machine is described by a completely-positive trace-preserving map $``$ that takes states in an Hilbert space $``$ to states in the Hilbert space $``$. According to Jam ; OperatorR , this map $``$ can be put in one-to-one correspondence with a positive operator $`R`$ on $`_1_2_3`$, where the indices 1 and 2 stand for the two output clones, while index 3 stands for the input system (all spaces are isomorphic to $``$). Specifically, by fixing a basis $`=\{|n|n=1,\mathrm{},d\}`$ for the $`d`$-dimensional Hilbert space $``$, the correspondence is given by
$$R=(𝟙)|𝟙𝟙|,$$
(6)
where $`|𝟙^\mathrm{𝟚}`$ is (up to normalization) the maximally entangled state $`|𝟙=_{𝕟=\mathrm{𝟙}}^𝕕|𝕟|𝕟`$. In terms of the operator $`R`$, the action of the QO on states is given by
$$M(\rho )=\mathrm{Tr}_3\left[𝟙_\mathrm{𝟙}𝟙_\mathrm{𝟚}\rho _\mathrm{𝟛}^𝕋\right],$$
(7)
where $`T`$ denotes transposition with respect to the fixed basis $``$.
Notice that, since the map $``$ is completely positive, the operator $`R`$ defined by Eq. (6) is positive. Moreover, according to Eq. (7), the trace-preservation condition $`\mathrm{Tr}[(\rho )]=1\rho `$ becomes
$$\mathrm{Tr}_{1,2}[R]=𝟙_\mathrm{𝟛},$$
(8)
that is, the trace of $`R`$ over the two output spaces gives the identity in the input space. Finally, the covariance condition (5) translates into OperatorR
$$[R,U_gU_gU_g^{}]=0,g𝐆,$$
(9)
with $``$ denoting complex conjugation with respect to the fixed basis $``$.
### II.3 Strong covariance condition
In this paragraph, we introduce a stronger requirement than simple covariance, which we will call strong covariance. This requirement concerns the unitary realization of the cloning machine with an ancilla, and corresponds to imposing that the ancilla transforms under the action of the group as the time-reversed of the transformation undergone by the two clones.
The explicit form of the strong covariance condition can be introduced by purifying the QO describing the cloning machine. The operator $`R`$ introduced in Eq. (6) is (up to normalization) the output state resulting from the application of the map $``$ on a maximally entangled state. Such an output state is not pure in general, but it can always be purified by introducing an ancillary system. In this way, the QO is realized as a unitary transformation (isometry) on the extended Hilbert space. Let us define $`|\mathrm{\Psi }^4`$ as the (normalized) pure state of the two clones, the input system, and the ancilla after the cloning transformation. The operator $`R`$ of Eq. (6) is then given by
$$R=d\mathrm{Tr}_4[|\mathrm{\Psi }\mathrm{\Psi }|],$$
(10)
the index 4 denoting the ancilla.
We say that the unitary realization of a cloning machine is strongly covariant if the joint output state $`|\mathrm{\Psi }`$ satisfies the property navez
$$U_gU_gU_g^{}U_g^{}|\mathrm{\Psi }=|\mathrm{\Psi },g𝐆.$$
(11)
In other words, a strongly covariant realization of cloning requires that i) the ancilla transforms under the group with the time-reversed unitary $`U_g^{}`$, and ii) the joint output state is invariant under the action of the group. From a physical point of view, this corresponds intuitively to assuming a kind of “conservation law” in the cloning process, where the ancilla undergoes a time-reversed transformation in order to balance the corresponding transformation of the two clones.
We will name strongly covariant a map that admits a strongly-covariant unitary realization. It is easy to see that a strongly covariant map is always covariant, but the converse is not necessarily true. The puzzle is now that all the known optimal covariant cloners satisfy this additional property. In the following, we will investigate the meaning of this strong covariance condition, showing in particular that the strongly-covariant maps coincide with the extremal covariant maps in the case of the (discrete or continuous) Weyl-Heisenberg group, which happens to be a symmetry of the set of input states in the vast majority of cloners considered in the literature.
## III Extremal covariant cloning machines
### III.1 Characterization of extremal covariant QOs
The set of covariant QO is a convex set, namely the convex combination of two such QO is still a covariant QO. In the same way, the set of positive operators $`R`$ defined by (6) and satisfying the relations (8) and (9) is a convex set. We will call $`𝒞`$ such a convex set of “covariant operators”.
Since for a pure input state the Uhlmann fidelity—either global or single-clone—is a linear functional of the QO, the search for the optimal covariant cloner can be restricted without loss of generality to the extremal points of this convex set, i.e. those QOs that cannot be written as convex combinations of other QOs. The convex structure of the set of covariant QOs then greatly simplifies the optimization problem. Although finding a characterization of the extremal covariant maps is, in general, a rather complicated issue Scutaru ; KeylWern ; ExtrPovmAndQo , here we can give a simple characterization of the extremal covariant maps in the special case where the representation $`\{U_g|g𝐆\}`$ acting on the input states is irreducible.
In order to deal with the covariance condition (9) it is useful to decompose the Hilbert space $`^3`$ into irreducible subspaces:
$$^3=\underset{\mu 𝖣}{}\underset{i=1}{\overset{m_\mu }{}}_i^{(\mu )}.$$
(12)
Here the index $`\mu `$ runs over the set $`𝖣`$ of the inequivalent representations that show up in the Clebsch-Gordan decomposition of the representation $`\{U_gU_gU_g^{}\}`$, while the index $`i`$ distinguishes $`m_\mu `$ different subspaces carrying equivalent representations. We recall that, by definition, two irreducible subspaces $`_i^{(\mu )}`$ and $`_j^{(\mu )}`$ of a given representation $`\{V_g\}`$ carry equivalent representations if and only if there exists an isomorphism $`T_{ij}^{(\mu )}:_j^{(\mu )}_i^{(\mu )}`$ such that $`[T_{ij}^{(\mu )},V_g]=0,g𝐆`$.
Using Schur’s lemma, it is possible to prove (see, e.g., OperatorR ) that the general expression of a positive operator satisfying the commutation relation (9) is
$$R=\underset{\mu 𝖣}{}\underset{i,j}{}r_{ij}^{(\mu )}T_{ij}^{(\mu )},$$
(13)
where each $`r^{(\mu )}`$ is a positive $`m_\mu \times m_\mu `$ matrix. Moreover, by diagonalizing the matrix $`r^{(\mu )}`$, we can write
$$R=\underset{\mu 𝖣}{}\underset{i}{}\lambda _i^{(\mu )}P_i^{(\mu )},$$
(14)
where $`\lambda _i^{(\mu )}0`$, and $`P_i^{(\mu )}`$ is the projection onto an irreducible subspace $`𝒦_i^{(\mu )}`$ carrying the representation $`\mu `$. The diagonalization of the matrix $`r_{ij}^{(\mu )}`$ corresponds to switching from the decomposition (12) to a new decomposition of the Hilbert space $`^3`$
$$^3=\underset{\mu 𝖣}{}\underset{i=1}{\overset{m_\mu }{}}𝒦_i^{(\mu )},$$
(15)
where $`\{𝒦_i^{(\mu )}\}`$ is a new set of irreducible subspaces. In fact, due to the presence of equivalent representations, there is a freedom in the choice of irreducible subspaces that decompose the Hilbert space MlMeasurements .
###### Theorem 1
If the representation $`\{U_g\}`$ is irreducible, then a covariant operator $`R𝒞`$ is extremal if and only if it is proportional to a projection onto an irreducible subspace, namely
$$R=\frac{d}{d_\mu }P_i^{(\mu )}.$$
(16)
where $`P_i^{(\mu )}`$ is the projection onto the irreducible subspace $`𝒦_i^{(\mu )}`$ whose dimension is $`d_\mu `$.
Proof. Let be $`R`$ a covariant operator in $`𝒞`$. Since $`R`$ is a positive operator commuting with the group action (9), it has the form (14) with a suitable decomposition of the Hilbert space. On the other hand, any projection $`P_i^{(\mu )}`$ in the sum satisfies $`[P_i^{(\mu )},U_g^2U_g^{}]=0g`$, therefore its partial trace $`\mathrm{Tr}_{1,2}[P_i^{(\mu )}]`$ commutes with the irreducible representation $`\{U_g^{}\}`$. By Schur’s lemma, the partial trace is proportional to the identity in $`_3`$, namely $`\mathrm{Tr}_{1,2}[P_i^{(\mu )}]=k_\mu 𝟙_\mathrm{𝟛}`$. Taking traces on both sides, we can evaluate the proportionality constant $`k_\mu =\frac{d_\mu }{d}`$. As a consequence, any positive operator defined by $`R_i^{(\mu )}=\frac{d}{d_\mu }P_i^{(\mu )}`$ satisfies both (8) and (9), whence it is itself a covariant operator in $`𝒞`$. On the other hand, Eq. (14) yields the convex decomposition of $`R`$ in terms of the extremal points $`\{R_i^{(\mu )}\}`$ proportional to the orthogonal projectors $`P_i^{(\mu )}`$.$`\mathrm{}`$
Remark. When the set of input states is invariant under an irreducible representation, Theorem 1 greatly simplifies the search for optimal cloners, since one just needs to find the irreducible subspaces $`𝒦_i^{(\mu )}`$ of $`^3`$ and find out which operator $`R_i^{(\mu )}`$ projecting on $`𝒦_i^{(\mu )}`$ maximizes the fidelity.
### III.2 Characterization of strongly covariant QOs
Theorem (1) allows to understand the meaning of the strong covariance condition in the case where the group representation $`\{U_g\}`$ is irreducible. In this case, we will show that the strongly covariant maps form a special subset of the set of extremal covariant QOs.
###### Theorem 2
Denote by $`\omega `$ the irreducible representation $`\{U_g\}`$ transforming the input states. Then, the strong covariance condition amounts to restricting to extremal QOs of the form
$$R=P_i^{(\omega )}.$$
(17)
In other words, the strongly covariant maps are the extremal maps with $`\mu =\omega `$ in Eq. (16). (Notice that, by definition, $`d/d_\omega =1`$.) To find such maps, one has to select among the irreducible subspaces of $`^3`$ those carrying a representation equivalent to $`\{U_g\}`$ (the representation transforming the input states).
Proof. Consider a pure joint state $`|\mathrm{\Psi }^4`$ satisfying the strong covariance condition (11). Since any $`P_i^{(\mu )}(^3)`$ in (14) commutes with the representation $`\{U_g^2U_g^{}\}`$, the vector $`|\mathrm{\Psi }_i^{(\mu )}=(P_i^{(\mu )}𝟙)|\mathbb{\Psi }`$ also satisfies the strong covariance condition, namely
$$U_g^2U_g^2|\mathrm{\Psi }_i^{(\mu )}=|\mathrm{\Psi }_i^{(\mu )}g𝐆.$$
(18)
On the other hand $`|\mathrm{\Psi }_i^{(\mu )}`$ transforms with the representation $`\mu \omega ^{}`$, corresponding to $`P_i^{(\mu )}(U_g^2U_g^{})P_i^{(\mu )}`$ for $`\mu `$ and $`U_g^{}`$ for $`\omega ^{}`$. Therefore, the Clebsch-Gordan series of $`\mu \omega ^{}`$ must contain the trivial representation $`\mu _0`$, where the action of any group element is given by multiplication by the number $`1`$. In terms of the characters $`\chi _\mu (g),\chi _\omega (g),`$ and $`\chi _{\mu _0}(g)1`$ of the three representations, this amounts to say that the character of the trivial representation is not orthogonal to the character of the tensor product $`\mu \omega ^{}`$, namely
$$\chi _{\mu _0},\chi _\mu \chi _\omega ^{}=_𝐆\mathrm{d}g\chi _\mu (g)\chi _\omega ^{}(g)0.$$
(19)
Since the characters of irreducible representations are orthonormal, the value of the integral (19) is the Kronecker delta $`\delta _{\mu \omega }`$. Therefore, the tensor product $`\mu \omega ^{}`$ contains the trivial representation $`\mu _0`$ if and only if $`\mu =\omega `$. According to this, the operator $`R=d\mathrm{Tr}_4[|\mathrm{\Psi }\mathrm{\Psi }|]`$ must have a special block form
$$R=\underset{i}{}\lambda _i^{(\omega )}P_i^{(\omega )},$$
(20)
that is, the sum (14) runs only on the projections with $`\mu =\omega `$. Finally, we can prove that $`R`$ is also extremal. Since $`R=d\mathrm{Tr}_4[|\mathrm{\Psi }\mathrm{\Psi }|]`$, the rank of $`R`$ is the Schmidt number of the pure state $`|\mathrm{\Psi }`$ with respect to the bipartition ancilla vs clones+input, whence it cannot be larger than the dimension of the ancilla, that is, $`\mathrm{rank}(R)d`$. On the other hand, from Eq. (20), we have $`\mathrm{rank}(R)=dn`$, where $`n`$ is the number of blocks in the direct sum. By comparison, we obtain $`n=1`$, i.e., $`R`$ is proportional to just one irreducible projection. Exploiting the characterization of Theorem 1, we know that such an operator is extremal. $`\mathrm{}`$
Remark. Theorem 2 thus implies that imposing strong covariance instead of covariance corresponds to considering a special class of extremal covariant QOs. In general, an extremal covariant map with respect to some group is not necessarily strongly covariant with respect to that group. However, strong covariance becomes simply equivalent to covariance together with extremality in the special case of the discrete Weyl-Heisenberg group. This is the topic of the next Section.
## IV Extremal cloners for the Weyl-Heisenberg group
### IV.1 Covariance vs strong covariance
Let us consider the class of cloning machines characterized by the fact that the set of states $`𝖲`$ to be cloned is invariant under the discrete Weyl-Heisenberg group, namely the set of unitary operators
$$U_{pq}=\underset{k=0}{\overset{d1}{}}e^{\frac{2\pi i}{d}kq}|kpk|,p,q=0,\mathrm{},d1,$$
(21)
where $`\{|k|k=0,\mathrm{},d1\}`$ is an orthonormal basis of a $`d`$-dimensional Hilbert space, and $``$ denotes the addition modulo $`d`$. This class includes for instance the universal cloning machines Wern , the Fourier-covariant cloning machines CBKG , or the phase-covariant cloning machines NiuGriff ; PhaseCinc ; PhaseDM ; PhaseBDM ; PhaseCL , as well as these three cases for generic asymmetry between the clones. Indeed, in all these cases, due to the invariance of the set of input states, one can assume without loss of generality that the cloner is covariant under the Weyl-Heisenberg group.
###### Theorem 3
For the discrete Weyl-Heisenberg group, all extremal covariant cloners are also strongly covariant.
Proof. Since the action of the discrete Weyl-Heisenberg group is irreducible in the $`d`$-dimensional Hilbert space $``$, we can exploit the characterization of Theorem 1. The decomposition (12) of the Hilbert space $`^3`$ into irreducible subspaces of the representation $`\{U_{pq}U_{pq}U_{pq}^{}\}`$ now reads
$$^3=\underset{r,s=0}{\overset{d1}{}}_{rs}$$
(22)
where
$$_{rs}=|U_{rs}.$$
(23)
Here $`|U_{rs}`$ denotes the subspace of vectors of the form $`|\psi |U_{rs}`$, where $`|\psi `$ and
$$|U_{rs}=\underset{k=0}{\overset{d1}{}}e^{\frac{2\pi i}{d}ks}|kr|k$$
(24)
is the $`d`$-dimensional Bell states. The orthogonal subspaces $`_{rs}`$ all carry the same representation, namely for any couple of spaces $`_{rs}`$ and $`_{r^{}s^{}}`$, one has the isomorphism
$$T_{rs,r^{}s^{}}=\frac{1}{d}U_{rs}^{}U_{r^{}s^{}}|U_{rs}U_{r^{}s^{}}|$$
(25)
that commutes with the representation $`\{U_{pq}^2U_{pq}^{}\}`$. Moreover, since $`U_{pq}U_{pq}^{}|𝟙=|𝟙,𝕡,𝕢`$, the space $`_{00}=|𝟙`$ carries the representation $`\{U_{pq}\}`$. Summarizing, all irreducible subspaces in the decomposition of $`^3`$ carry the same representation, which is equivalent to $`\{U_{pq}\}`$, the representation acting on the input states. Therefore, all the extremal maps in Theorem 1 are also strongly covariant, according to Theorem 2. $`\mathrm{}`$
The result of Theorem 3 shows that, if the set of input states is invariant with respect to the discrete Weyl-Heisenberg group, then one can assume strong covariance without loss of generality, since it provides a parametrization of all extremal covariant QO. Moreover, in the following we will see that the the strongly covariant cloning machines (w.r.t. the discrete Weyl-Heisenberg group) can be parametrized in terms of “double-Bell” states, thus explaining with a general argument the presence of a recurrent structure that characterizes the known optimal cloners.
### IV.2 Parametrization with double-Bell states
Using Theorem 3, we can parametrize explicitly all the extremal quantum cloning transformations that are covariant with respect to the discrete Weyl-Heisenberg group. Since the operator $`R`$ associated to an extremal map is the projection onto an irreducible subspace (see Eq. (17)), it is enough to write the most general form of such a projection, which has the form
$$P_𝐚=\underset{r,s,r^{},s^{}=0}{\overset{d1}{}}a_{rs}a_{r^{}s^{}}^{}T_{rs,r^{}s^{}}$$
(26)
with $`𝐚=\{a_{rs}\}`$ such that $`_{r,s}|a_{rs}|^2=1`$. Remarkably, the irreducible projections are in one to one correspondence with the pure states in $``$. As a matter of fact, the convex structure of covariant QOs is exactly the same as the convex structure of states on $``$.
By inserting Eq. (25) in Eq. (26), we obtain
$$R=\underset{r,s,r^{},s^{}=0}{\overset{d1}{}}\frac{a_{rs}a_{r^{}s^{}}}{d}U_{rs}^{}U_{r^{}s^{}}|U_{rs}U_{r^{}s^{}}|,$$
(27)
thus giving the explicit parametrization of a generic extremal covariant map. Finally, by purifying $`R`$ we can characterize the (strongly covariant) unitary realization of the extremal cloning machine with the pure output state of the “double Bell” form
$$|\mathrm{\Psi }=\underset{r,s=0}{\overset{d1}{}}a_{rs}\frac{|U_{rs}^{}_{1,4}}{\sqrt{d}}\frac{|U_{rs}_{2,3}}{\sqrt{d}}.$$
(28)
This proves the “double Bell” ansatz cerfansatz1 ; cerfansatz2 , which captures the characteristic feature of all the above-mentioned optimal cloners CBKG ; Wern ; NiuGriff ; PhaseCinc ; PhaseDM ; PhaseBDM ; PhaseCL . The expression (28) for the optimal cloner can be then assumed without loss of generality whenever the set of input states is invariant under the Weyl-Heisenberg group. Indeed, such an invariance is very common, whence the form (28) covers most of the one-to-two cloning machines considered in the literature. Moreover, Theorem 3 and the double-Bell form can be extended in a direct way to the case of the continuous Weyl-Heisenberg group in infinite dimension (see Subsection IV.5).
### IV.3 Optimal qubit cloners
In this Subsection we review the main examples of qubit cloners in the framework drawn in the previous sections. Theorem 3 greatly simplifies the search of optimal cloners, and explains some interesting relations among different cloning machines.
#### IV.3.1 Cloning of the BB84 states
The study of the optimal cloning as a possible cryptographic attack is crucial for the security analysis of the BB84 cryptographic protocol. In this case, the aim of an eavesdropper is to clone with the same fidelity two mutually unbiased bases, corresponding to the eigenvectors of the Pauli matrices $`\sigma _x`$ and $`\sigma _y`$. Such discrete set of states describes a square in the equatorial plane of the Bloch sphere, and it is clearly invariant under the action of the discrete Weyl-Heisenberg group, which in dimension 2 is just the Pauli group,
$$U_{0,0}=𝟙,𝕌_{\mathrm{𝟘},\mathrm{𝟙}}=\sigma _𝕫,𝕌_{\mathrm{𝟙},\mathrm{𝟘}}=\sigma _𝕩,𝕌_{\mathrm{𝟙},\mathrm{𝟙}}=𝕚\sigma _𝕪.$$
(29)
Using the double-Bell form (28), and optimizing coefficients, one finds the optimal asymmetric cloner of Ref. qutrit
$$\begin{array}{cc}\hfill |\mathrm{\Psi }& =\frac{1}{2}\{F_B|𝟙_{\mathrm{𝟙},\mathrm{𝟜}}|𝟙_{\mathrm{𝟚},\mathrm{𝟛}}+(\mathrm{𝟙}𝔽_𝔹)|\sigma _𝕫_{\mathrm{𝟙},\mathrm{𝟜}}|\sigma _𝕫_{\mathrm{𝟚},\mathrm{𝟛}}\hfill \\ & +\sqrt{F_B(1F_B)}(|\sigma _x_{1,4}|\sigma _x_{2,3}+|\sigma _y_{1,4}|\sigma _y_{2,3})\}.\hfill \end{array}$$
(30)
Here $`F_B`$ is the fixed fidelity of Bob’s clone (Hilbert space $`_2`$). The fidelity of Eve’s clone is given by $`F_E=1/2+\sqrt{F_B(1F_B)}`$, so that the symmetric cloner has a fidelity $`1/2+1/\sqrt{8}`$.
#### IV.3.2 Phase-covariant qubit cloning
The general theory allows us to assume again the double-Bell expression of Eq. (28), since the equatorial states $`1/\sqrt{2}(|0+e^{i\varphi }|1)`$ are invariant under the action of the Pauli group. This implies that the asymmetric cloning obtained in Ref. qutrit is actually optimal, and in particular, the popular conjecture that phase-covariant equatorial cloning PhaseCinc is indeed equivalent to the BB84-states cloning qutrit is now proved. Clearly, the double-Bell form is exactly the same as in Eq. (30).
#### IV.3.3 Six states cloning
This cloning problem is linked to the security of the six-state quantum cryptographic protocol sixst . The states to be cloned are the six eigenstates of the three Pauli matrices, which are invariant under the Pauli group (i.e. the discrete Weyl-Heisenberg group in dimension 2). Therefore, one can use again the double-Bell form, and the expression for the optimal asymmetric cloning is cerfansatz2
$$\begin{array}{cc}\hfill |\mathrm{\Psi }=& \frac{1}{2}\{\sqrt{\frac{3F_B1}{2}}|𝟙_{\mathrm{𝟙},\mathrm{𝟜}}|𝟙_{\mathrm{𝟚},\mathrm{𝟛}}+\hfill \\ & \sqrt{\frac{1F_B}{2}}(\underset{i=1}{\overset{3}{}}|\sigma _i_{1,4}|\sigma _i_{2,3})\},\hfill \end{array}$$
(31)
where $`F_B`$ is the fixed fidelity of Bob’s clone. The fidelity of Eve’s clone is then given by $`F_E=1F_B/2+\sqrt{(3F_B1)(1F_B)}/2`$, so that the symmetric cloner has the fidelity 5/6.
#### IV.3.4 Universal cloning
In the case of universal cloning, it is straightforward to see that the set of input states (the whole surface of the Bloch sphere) is invariant under the Pauli group. Similarly to the case of phase-covariant cloning, using the double-Bell form (28), we obtain the same optimal cloner as in the case of the six states, thus proving the equivalence between the six-states cloning and the universal cloning. Accordingly, the double-Bell expression for the optimal universal cloner is the same as in Eq. (31).
#### IV.3.5 Cubic cloning
Using the present method, we can analyze easily all cloning problems with the set of input states invariant under the Pauli group, which in the Bloch sphere corresponds to invariance under $`\pi `$-rotation around the 3 reference axes. As a new example, let us consider the cloning of eight pure states forming a cube in the Bloch sphere. By performing a suitable rotation, we can always bring the vertexes of the cube in the positions specified by the Bloch vectors $`\{\pm 1/\sqrt{3},\pm 1/\sqrt{3},\pm 1/\sqrt{3}\}`$, so that the states to be cloned become
$$\rho =\frac{1}{2}(𝟙\pm \frac{\mathrm{𝟙}}{\sqrt{\mathrm{𝟛}}}\sigma _𝕩\pm \frac{\mathrm{𝟙}}{\sqrt{\mathrm{𝟛}}}\sigma _𝕪\pm \frac{\mathrm{𝟙}}{\sqrt{\mathrm{𝟛}}}\sigma _𝕫).$$
(32)
This set of states is clearly invariant under the Pauli group. Starting from a general double-Bell form
$$|\mathrm{\Psi }=\frac{1}{2}\underset{i=0}{\overset{3}{}}a_i|\sigma _i_{1,4}|\sigma _i_{2,3},$$
(33)
where $`\sigma _0=𝟙`$ and $`_i|a_i|^2=1`$, one gets the following expressions for the fidelities of the two clones
$$F_B=|a_0|^2+\frac{1}{3}\underset{i=1}{\overset{3}{}}|a_i|^2,F_A=\frac{2}{3}+\frac{1}{3}\left|\underset{i=0}{\overset{3}{}}a_i\right|^2.$$
(34)
It is clear that one can take all the coefficients $`a_i`$ as nonnegative without affecting $`F_B`$, and seek the maximum of $`F_A`$ only for $`a_i0`$. Using the method of Lagrange multipliers one can then maximize $`F_B`$ for fixed $`F_B`$, thus obtaining
$$a_0=\sqrt{\frac{3F_B1}{2}},a_i=\sqrt{\frac{1F_B}{2}}.$$
(35)
Comparing these values with the corresponding ones in Eq. (31), we see that the optimal cloning of a cube in the Bloch sphere is performed by the same machine that gives the optimal cloning of the six-states and the optimal universal cloning.
### IV.4 Optimal $`d`$-dimensional cloners
#### IV.4.1 Cloning of two Fourier-transformed bases
The $`d`$-dimensional generalization of the cloning of BB84 states gives rise to the problem of cloning two bases that are Fourier-transformed, namely the computational basis $`\{|m\}`$ and the dual basis $`\{|e_m\}`$, where
$$|e_m=\frac{1}{\sqrt{d}}\underset{p=0}{\overset{d1}{}}e^{\frac{2\pi imp}{d}}|p.$$
(36)
The invariance of $`𝖲`$ under the action of the discrete Heisenberg group is straightforward, and the optimal asymmetric cloning corresponds to the following double-Bell form CBKG
$$\begin{array}{cc}\hfill |\mathrm{\Psi }=& \frac{1}{d}\{F_B|𝟙|𝟙+\frac{\mathrm{𝟙}𝔽_𝔹}{𝕕\mathrm{𝟙}}\underset{𝕡,𝕢=\mathrm{𝟙}}{\overset{𝕕\mathrm{𝟙}}{}}|𝕌_{𝕡𝕢}^{}|𝕌_{𝕡𝕢}+\hfill \\ & \sqrt{\frac{F_B(1F_B)}{d1}}\underset{p=1}{\overset{d1}{}}(|U_{p0}^{}|U_{p0}+|U_{0p}^{}|U_{0p})\}.\hfill \end{array}$$
(37)
The fidelity of Eve’s clone is given by
$$F_E=\frac{F_B}{d}+\frac{(d1)(1F_B)}{d}+\frac{2}{d}\sqrt{(d1)F(1F)}$$
(38)
so that the symmetric cloner has the fidelity $`(1+1/\sqrt{d})/2`$.
#### IV.4.2 Multiple phase-covariant cloning
The optimal cloning of states of the form $`\frac{1}{\sqrt{d}}(|0+_{k=1}^{d1}e^{i\varphi _k}|k)`$ fits the constraints for the validity of the double-Bell form, since the set $`𝖲`$ is clearly invariant under the discrete Heisenberg group. For the double-Bell form for the optimal cloner, see PhaseCL .
#### IV.4.3 Universal cloning
In this case, the set $`𝖲`$ of states to be cloned is the whole set of pure states in a $`d`$-dimensional Hilbert space, which is clearly invariant under all the unitaries in the discrete Weyl-Heisenberg group. The optimal universal cloning BuzHill ; Wern corresponds indeed to the following double Bell form
$$\begin{array}{cc}\hfill |\mathrm{\Psi }=& \frac{1}{d}\{\sqrt{\frac{(d+1)F_B1}{d}}|𝟙_{\mathrm{𝟙},\mathrm{𝟜}}|𝟙_{\mathrm{𝟚},\mathrm{𝟛}}+\hfill \\ & \sqrt{\frac{1F_B}{d(d1)}}\underset{(p,q)(0,0)}{}|U_{p,q}^{}_{1,4}|U_{pq}_{2,3}\},\hfill \end{array}$$
(39)
as derived in Ref. cerfansatz2 . The fidelity of Eve’s clone is given by
$`F_E=1{\displaystyle \frac{(d^22)F_B+2d}{d^2}}`$ (40)
$`+{\displaystyle \frac{2\sqrt{d1}}{d^2}}\sqrt{(1F_B)[(d+1)F_B1]}`$
so that the symmetric cloner has a fidelity $`F=1/2+1/(d+1)`$.
### IV.5 Cloning of continuous variables
Theorem (3) and the double-Bell form can be extended to the continuous-variable case, where the set of states to be cloned lies in an infinite dimensional Hilbert space and is invariant under the Weyl-Heisenberg representation of the displacements in the complex plane, i.e. under the set of unitaries
$$\{D(\alpha )=e^{\alpha a^{}\overline{\alpha }a}|\alpha \},$$
(41)
where $`[a,a^{}]=1`$. The Weyl-Heisenberg representation can be regarded indeed as the continuous-variable version of the discrete Weyl-Heisenberg group, where the couple of integers $`(p,q)`$ is replaced by the complex number $`\alpha `$. In this case, one can decompose the Hilbert space $`^3`$ (two clones + input system) by substituting formally the direct sum (22) with a direct integral
$$^3=_{}\mathrm{d}^2\alpha _\alpha ,$$
(42)
where
$$_\alpha =|D(\alpha ),$$
(43)
and $`|D(\alpha )=_{m,n=0}^{\mathrm{}}m|D(\alpha )|n|m|n`$ for a fixed orthonormal basis $`\{|n|n=0,1,\mathrm{}\}`$. The subspaces $`_\alpha `$ are orthogonal in the Dirac sense and carry all the same representation. The continuous variable version of the isomorphism (25) is
$$T_{\alpha \beta }=\frac{1}{\pi }D(\alpha )^{}D(\beta )|D(\alpha )D(\beta )|.$$
(44)
According to the characterization of Theorem (1) and generalizing (26), an extremal QO is then represented by
$$R=_{}\mathrm{d}^2\alpha _{}\mathrm{d}^2\beta \varphi (\alpha )\varphi ^{}(\beta )T_{\alpha \beta },$$
(45)
where $`_{}\mathrm{d}^2\alpha |\varphi (\alpha )|^2=1`$. Again, the convex structure of covariant QO is the same as the convex structure of states on $``$. Moreover, it is still possible to give the purification of the cloning machine as
$$|\mathrm{\Psi }=_{}\mathrm{d}^2\alpha \varphi (\alpha )\frac{|D(\alpha )^{}_{1,4}}{\sqrt{\pi }}\frac{|D(\alpha )_{2,3}}{\sqrt{\pi }},$$
(46)
according to the continuous-variable version of the double-Bell ansatz. This special form of the unitary realization is indeed the unifying feature of the known continuous-variable cloners (CerIpe ; CerIb ).
## V Conclusion
We have analyzed the problem of cloning a set of states that is invariant under the action of a given symmetry group. If we use a figure of merit that is invariant with respect to this group, such as the Uhlman fidelity, then the optimal cloning transformation (i.e., the transformation that maximizes the average fidelity over the set of input states) can be chosen to be group covariant. We have shown that substituting this covariance condition with a strong covariance condition implies that the resulting cloning transformation is extremal. The converse is not true in general, that is, an extremal covariant transformation is not necessarily strongly covariant. However, when the considered invariance group is the (discrete or continuous) Weyl-Heisenberg group, the converse also holds, so that the set of strongly-covariant cloners is equivalent to the set of extremal covariant cloners. Since the covariant cloners form a convex set, and since the fidelity is linear in the cloning transformation, this equivalence greatly simplifies the search for optimal cloners: it is sufficient to search among the set of extremal cloners. Luckily, the set of strongly-covariant (hence extremal) cloners with respect to the Weyl-Heisenberg group can be parametrized in a very compact form, which coincides with the so-called double-Bell ansatz. In this form, the cloner only depends on $`d^2`$ real parameters for a $`d`$-dimensional input state. As a consequence of the simplification of the optimization problem, one can easily derive a large variety of optimal cloning transformations. As an illustration of the power of the method, we proved the optimality of several cloners that have been described in the literature, including the continuous-variable cloners. As a side result, we proved that the optimal cloner of the four states involved in the BB84 protocol (six states involved in the 6-state protocol) is the phase-covariant (universal) cloner. We also showed that the optimal cloner of any eight states forming a cube on the Bloch sphere is the universal cloner.
###### Acknowledgements.
This work has been supported by INFM under PRA-2002-CLON, and has been co-funded by the EC and Ministero Italiano dell’Università e della Ricerca (MIUR) through the co-sponsored ATESIT project IST-2000-29681 and Cofinanziamento 2003. G.M.D. acknowledges partial support from the Multiple Universities Research Initiative (MURI) program administered by the U.S. Army Research Office under Grant No. DAAD1900-1-0177. N.J.C. also acknowledges hospitality of the QUIT group, as well as partial support from MIUR, from the Action de Recherche Concertée de la Communauté Française de Belgique, from the IUAP program of the Belgian Federal Governement under grant V-18, and from the EC through projects RESQ (IST-2001-37559) and SECOQC (IST-2003-506813). |
warning/0507/astro-ph0507584.html | ar5iv | text | # The NN2 Flux Difference Method for Constructing Variable Object Light Curves
## 1 Introduction
The astronomical time domain provides unique insight into a range of astrophysical phenomena. Studies of variable stars yield information about stellar structure and evolution as well as help to set the extra-galactic distance scale. Active Galactic Nuclei (AGN) reveal the high-energy phenomena associated with the super-massive black holes that reside at the centers of most galaxies. Supernovae (SNe) and Gamma-Ray Bursts (GRBs) provide a glimpse of the fantastic energies released during the violent death throes of several types of stars. Type Ia supernovae (SNe Ia) are of particular interest because their use as “standard candles” has revealed the acceleration of the expansion of the universe from an inferred cosmological constant-like force (Riess et al., 1998; Perlmutter et al., 1999).
Studies of variable sources require specific analysis methods that are not necessary for non-variable sources. Since it is often difficult to detect variation in an object by simply inspecting images, the standard procedure is to subtract images taken at different times to remove objects with constant flux. Photometrically variable objects are then obvious. For the case of SNe, one typically obtains a pair of observations separated in time to allow for SNe not present in the first image to reach observable brightness in the second (see, e.g., Perlmutter et al. 1995 and Schmidt et al. 1998 for a description of the method). After detection, additional observations are made to obtain the complete light curve necessary for cosmological analysis (see Phillips, 1993; Riess, Press, & Kirshner, 1996). In order to construct the light curve, it is necessary for at least one observation (the “template” image) to contain no SN flux. Often this is the initial image used during discovery of the SN. In many instances, however, the SN is present at a faint level in this image, so an additional observation, taken after the SN has faded from view, is required. The light curve is then calculated by measuring the flux levels in subtractions of each image from the designated template using, for example, the subtraction procedure described by Alard & Lupton (1998). This, the “single-template method,” is the typical means of constructing light curves of SNe and other variable sources.
However, this method has certain drawbacks. The primary flaw is that the quality of any subtraction depends greatly upon the two images involved. If the template is of a poor quality caused, for instance, by poor seeing or a low signal-to-noise ratio (S/N), then $`every`$ subtraction will be degraded, with a corresponding increase in the measured flux uncertainty, even if all other images are of high quality. Any flaw in the template creates a systematic error for the entire light curve that is not detectable from internal consistency checks or through comparison with another SN light curve.
In order to alleviate this problem, we have developed a new method for constructing light curves of photometrically variable objects. Given $`N`$ observations there are a total of $`N(N1)/2`$ pairs of images that can be subtracted together, only $`N1`$ of which are performed in the single-template method. A matrix of flux differences can be constructed from these subtractions and used to determine the flux at each individual epoch. This process removes the dependence on any single observation, because all observations are treated equally as a “template.” We refer to this method as the “N(N-1)/2” method (hereafter abbreviated NN2; see Novicki & Tonry 2000 for an initial description).
Section 2 describes the mathematical underpinnings of NN2. In Section 3 we demonstrate the efficacy of the method using simulated SNe inserted into images used during an actual high-redshift SN survey. Section 4 gives our conclusions.
## 2 Mathematical Basis of the NN2 Method
We assume that we start with $`N`$ observations of an object, so that one may construct from all pairs of subtractions an $`N\times N`$ antisymmetric matrix $`A`$ of flux differences that we wish to analyze as a “vector-term difference.” In other words, we want to find an $`N`$-vector $`V`$ of fluxes such that
$$A_{ij}=V_jV_i.$$
(1)
We also assume that we have a symmetric $`N\times N`$ error matrix $`E`$ that expresses our uncertainty in each term of $`A`$. As we shall see, this matrix may not be easy to generate, and its interpretation may be somewhat ambiguous. However, one can imagine generating an error matrix by the following procedure.
In each of the difference images, we measure a flux for the object in question. In general this measurement consists of fitting a fixed point-spread function (PSF) profile at the location of the object by adjusting the amplitude of the PSF (both positive or negative) and the local background level. The PSF profile may be obtained from a suitable star in the original image while the location of the variable source may be determined by summing all the difference images (adjusted to keep the sign of the object positive) and fitting the location in this sum. Once we have a flux measurement, we can insert copies of the object at nearby empty regions of the difference image and repeat the procedure. The mean of the recovered fluxes indicates whether there is a bias in the measurement, and the scatter may be used as a term $`E_{ij}`$ in the error matrix.
The crux of the NN2 method is the distillation of the photometric measurements from the full set of $`N(N1)/2`$ subtractions to a lightcurve, $`V`$, that represents our best understanding of the behavior of the object under consideration. As long as it is consistent, the exact procedure for measuring the flux on the difference images is not central to the NN2 method we present here.
In order to find an optimal $`V`$, we wish to minimize the quantity
$$\chi ^2=\underset{i,j;i<j}{}\frac{(A_{ij}+V_jV_i)^2}{E_{ij}^2}$$
(2)
This construction may not be entirely appropriate depending on the errors in the flux differences $`A_{ij}`$. Ideally, if we possessed an extremely high-quality template with no SN flux present and applied an optimal subtraction procedure, the errors would be primarily due to photon counting statistics (see Alard & Lupton 1998 for a discussion). We would expect these errors to be uncorrelated and would simply wish to employ the single-template method to construct the SN light curve. As mentioned in the introduction, in practice there are nearly always imperfections associated with the template image that remove us from this idealized regime. These template errors introduce correlations in the individual flux measurement errors that are difficult to quantify and are typically assumed to be negligible in SN light-curve analysis. The use of the NN2 method, however, will introduce further correlations as a result of the common images in the various subtractions (for instance, the error in $`V_1V_2`$ will be anti-correlated with the error in $`V_2V_3`$ due to the common error in $`V_2`$). Although we believe that the use of the NN2 method will improve the ability to accurately recover variable object light curves, one should recognize that the NN2 method is expected to introduce these additional correlations to the fluxes measured from the various subtraction images, and so the $`\chi ^2`$ given above is not technically appropriate. Errors due to systematics in the subtraction procedure, such as those associated with template or software imperfections, would be expected to be effectively uncorrelated, and if they were dominant then Eq. 2 would indeed represent the proper $`\chi ^2`$. With these caveats in mind, we will proceed to use the definition of $`\chi ^2`$ as given in Eq. 2 as the basis of the NN2 method. Tests of its ability to recover accurate light-curve information in the following section demonstrate its effectiveness in practice.
However, we need to make one minor modification to our $`\chi ^2`$ because the $`\chi ^2`$ defined in Eq. 2 is degenerate to the addition of a constant to the $`V`$ vector—geometrically, $`\chi ^2`$ is constant along the line $`\widehat{i}`$. In order to lift this degeneracy and permit us to solve for $`V`$, we add a term to $`\chi ^2`$ that is quadratic in the degenerate direction, so that
$$\chi ^2=\underset{i,j;i<j}{}\frac{(A_{ij}+V_jV_i)^2}{E_{ij}^2}+\frac{(\underset{i}{}V_i)^2}{E^2},$$
(3)
where $`E`$ is a suitable typical uncertainty; for example,
$$\frac{1}{E^2}=\frac{2}{N(N1)}\underset{i,j;i<j}{}\frac{1}{E_{ij}^2}.$$
(4)
Our solution will therefore have $`_iV_i=0`$. This construction explicitly forces one to determine an accurate zero flux level at a later stage. In the single-template method this zero flux level is generally implicitly determined by assuming that the object of interest has zero flux in the template image. This same assumption can similarly be used in the NN2 method, but more sophisticated methods involving comparisons of many different images can also be invoked. If the absolute brightness of the variation being studied is important, the NN2 method clearly cannot free one from the requirement of having a fiducial image to measure the zero flux level. This is a fundamental limitation of any differential photometry method as the information is simply not available without such a fiducial image. However, even in the absence of a fiducial image, the NN2 method will produce a sensible and accurate relative lightcurve.
We now seek to solve for our lightcurve vector $`V`$ by minimizing $`\chi ^2`$ with respect to $`V`$:
$`0`$ $`=`$ $`{\displaystyle \frac{\chi ^2}{V_k}}`$ (5)
$`=`$ $`2{\displaystyle \underset{i,j;i<j}{}}{\displaystyle \frac{(A_{ij}+V_jV_i)}{E_{ij}^2}}(\delta _{jk}\delta _{ik})+2{\displaystyle \underset{i}{}}{\displaystyle \frac{V_i}{E^2}}.`$ (6)
Exploiting the antisymmetry of $`A`$ and the symmetry of $`E`$ we can rewrite Eq. 6 as
$$0=2\underset{i;ik}{}\frac{(A_{ik}+V_kV_i)}{E_{ik}^2}+2\underset{i}{}\frac{V_i}{E^2}.$$
(7)
These $`N`$ equations can be solved for $`V`$ by inverting a matrix $`C`$:
$$\underset{i;ik}{}\frac{A_{ik}}{E_{ik}^2}=\underset{i}{}C_{ik}V_i$$
(8)
where
$$C_{ik}=\frac{1}{E_{ik}^2}+\underset{j}{}\frac{1}{E_{kj}^2}\delta _{ik}+\frac{1}{E^2}.$$
(9)
The inverse of this curvature (Hessian) matrix $`C`$ yields uncertainties for $`V`$ from the square root of the diagonal elements as well as covariances from normalizing the off-diagonal elements by the two diagonal terms (under the assumption that the error matrix truly does represent Gaussian, independent uncertainties for each of the terms of the antisymmetric difference matrix).
An alternative approach to calculating uncertainties in $`V`$ stems from assuming that there is a vector $`\sigma `$ such that
$$E_{ij}^2=\sigma _i^2+\sigma _j^2.$$
(10)
Under this assumption, we seek to minimize
$$\chi _e^2=\underset{i,j;i<j}{}\left(E_{ij}^2+\sigma _i^2+\sigma _j^2\right)^2.$$
(11)
The minimization condition is
$`0`$ $`=`$ $`{\displaystyle \frac{\chi _e^2}{\sigma _k^2}}`$ (12)
$`=`$ $`2{\displaystyle \underset{i,j;i<j}{}}\left(E_{ij}^2+\sigma _i^2+\sigma _j^2\right)\left(\delta _{ik}+\delta _{jk}\right)`$ (13)
$`=`$ $`2{\displaystyle \underset{i;ik}{}}\left(E_{ik}^2+\sigma _i^2+\sigma _k^2\right).`$ (14)
These $`N`$ equations are solved by inverting a matrix $`D`$
$$\underset{i;ik}{}E_{ik}^2=\underset{i}{}D_{ik}\sigma _i^2$$
(15)
where
$$D_{ik}=1+(N2)\delta _{ik}.$$
(16)
After solving for $`V`$ and $`\sigma `$, we can evaluate the quality of the fit by comparing $`\chi ^2`$ to the number of degrees of freedom,
$$N_{\mathrm{d}of}=\frac{N(N1)}{2}(N1).$$
(17)
This $`N_{\mathrm{d}of}`$ comes from the number of data points, $`N(N1)/2`$, minus the number of model parameters, $`N1`$. Recall that we’ve explicitly required $`_iV_i=0`$, so that the number of model parameters is $`N1`$ rather than $`N`$.
Having outlined the basic method, we now discuss a fundamental uncertainty in the NN2 process. We can imagine two types of error that will cause $`V`$ to differ from the true flux values. The first, which we term “external error,” is intrinsic to the images themselves. For example, if the object has a positive statistical fluctuation in flux in one image or is corrupted by a cosmic ray that happens to coincide with the position of the object on the detector, this error will propagate through the entire differencing and analysis procedure. It is possible to obtain an antisymmetric difference matrix that is an exact vector-term difference ($`\chi ^2=0`$), but the solution vector will still contain errors. The second type of error, which we call “internal error,” is caused by the procedure of generating the antisymmetric matrix. One might imagine a set of images that have no flux error whatsoever, but through errors in convolving, differencing, or flux fitting, an antisymmetric matrix may be created that is not a perfect vector-term difference and for which $`\chi ^2>0`$.
Roughly speaking, one might expect that if the error matrix $`E`$ consists entirely of external errors the resulting $`\sigma `$ terms will all be approximately $`E/\sqrt{2}`$, since $`E`$ is the quadrature sum of two $`\sigma `$ terms. Alternatively, if the error matrix is purely internal error the $`\sigma `$ terms might be expected to be approximately $`E/\sqrt{N}`$, since each term in $`V`$ comes from comparison with $`N1`$ other images. In the case of external errors, the uncertainties derived from the $`\chi _e^2`$ analysis are correct. In the internal error case the uncertainties obtained from the covariance matrix derived from the $`\chi ^2`$ analysis are likewise appropriate.
It is not clear how to disentangle these different sorts of errors. The procedure suggested above of dropping copies of the object into each difference image and evaluating the scatter of the result will be sensitive to each sort of error, but it is possible to imagine cases where this procedure is unsatisfactory. We suggest that the errors provided in the $`E`$ matrix be interpreted as external errors and taken seriously as such. Thus, the vector $`V`$ is assigned an external uncertainty equal to the $`\sigma `$ vector. However, in order to handle a situation where $`\chi ^2/N_{\mathrm{d}of}`$ is much greater than 1 (i.e., where the antisymmetric matrix is simply not well represented as a vector-term difference), we suggest also creating an internal uncertainty vector $`\tau `$ that is obtained from the diagonal terms of the covariance matrix, scaled by $`\chi ^2/N_{\mathrm{d}of}`$:
$$\tau _k=\left(C_{kk}^1\frac{\chi ^2}{N_{\mathrm{d}of}}\right)^{1/2}$$
(18)
The total uncertainty is then the quadrature sum of $`\sigma `$ and $`\tau `$. Note that this approach implicitly assumes that the internal and external errors are uncorrelated and are proportional to one another as well as the provided $`E`$ matrix.
For problems where $`\chi ^2/N_{\mathrm{d}of}`$ is near unity without adjustment, the $`\tau `$ vector will be smaller than the $`\sigma `$ vector by approximately $`\sqrt{2/N}`$ and will make a fairly small contribution to the total uncertainty. When $`\chi ^2/N_{\mathrm{d}of}1`$ (i.e., the antisymmetric matrix is very closely represented by the vector-term difference), the $`\tau `$ vector will be negligible. However, when $`\chi ^2/N_{\mathrm{d}of}1`$, the $`\tau `$ vector will act to correct $`\chi ^2/N_{\mathrm{d}of}`$ to approximately unity, and this procedure will provide reasonable uncertainties, even though $`E`$ may be much too small.
## 3 Demonstration of Improved Accuracy in Recovering Variable Object Light Curves
The first extensive use of the NN2 method we have developed here occurred during the SN-search component of the IfA Deep Survey (Barris et al., 2004), although we also employed it to a limited extent in a previous SN survey by Tonry et al. (2003). The IfA Deep Survey was undertaken primarily with Suprime-Cam (Miyazaki et al., 1998) on the Subaru 8.2-m telescope and was supplemented with the 12K camera (Cuillandre et al., 1999) on the Canada-France-Hawaii 3.6-m telescope. Scores of high-redshift SN candidates were discovered (Barris et al., 2001, 2002) with 23 confirmed as SNe Ia. We here present several tests we performed to demonstrate the improved performance of NN2 relative to the single-template method.
In order to make a controlled test of the effectiveness of NN2 vs. the single-template method, we inserted artificial SNe into the survey images. The light curves of these objects consisted of a linear ramp-up and ramp-down in brightness over the time period covered by the survey observations. The timing of the light-curve maxima were selected at random and could lie within or outside of the survey period. The simulated SNe were laid down in a regular grid across the survey area, and all pairs of images were subtracted. Object-detection software was then run on all subtraction images to detect photometrically variable objects (both real objects and the artificial SNe) and to construct the NN2 flux difference matrix. The positions of both real and artificial SNe were fit by the object-detection software as described in Section 2. Since we knew the true light-curve properties used to create the synthetic SNe, we could calculate the root-mean-square scatter (RMS) around this artificial light curve using both the NN2 flux calculation and the single-template method with every individual observation as the template (this latter is equivalent to taking the flux values from a single column of the NN2 flux difference matrix).
We inserted approximately 2000 simulated SNe into the $`I`$-band observations of each of four $``$0.5 square-degree fields from the IfA Deep Survey, spanning a peak magnitude range of approximately $`m_I=21`$$`25`$. We used a predefined grid of positions to insert the simulated SNe, without taking into consideration the presence of actual objects nearby that would cause problems for detection and accurate photometric measurements. The small fraction that were so affected were accordingly not used in the final analysis.
Figure 1 shows the percentage improvement in the cumulative distributions of the RMS (in flux units) from the NN2 method over the set of all RMS values from the single-template method from the four survey fields (RMS values are calculated from flux measurements scaled so that a value of $`\mathrm{flux}=1`$ corresponds to a magnitude of $`25`$). The cumulative fraction for the NN2 method is larger than that for the single-template method distribution at all values of RMS, indicating that the NN2 method does indeed tend to yield $`smaller`$ RMS values. The NN2 method more accurately recovers the actual light curve of these variable objects.
Since one could imagine that certain templates of very high quality could outperform the NN2 method while the collection of all single-template measurements, as shown in Fig. 1, does not, we next examine the relationship between the NN2 RMS and the single-template RMS for individual observations. To illustrate this comparison, we will concentrate on only one of the survey fields, although details of our investigation of the entire survey area can be found in Barris (2004). Table 1 contains relevant information for the 16 observations of the selected field, f0438. This field is representative of the entire survey area, though it is notable that it contains an observation that was quite strongly affected by clouds (Observation 4), as seen by its unusually bright zero-point magnitude. Also noteworthy is Observation 11, taken in poor seeing conditions. We would expect the performance of the single-template method using these observations to be poor in comparison to the NN2 procedure.
In Table 1 we demonstrate that the typical RMS obtained with the NN2 method for the set of 1775 simulated SNe is smaller than the single-template method RMS using $`every`$ observation of the selected field. The improvement is generally fairly small, ranging from $`510`$%. We demonstrate below that these differences are statistically significant. The use of either Observations 4 and 11 as single templates, as expected, produces substantially worse results relative to the NN2 method than the other observations. For these observations the improvement due to the NN2 method is substantially larger than 10%. Figure 2 shows graphically the percentage difference in the cumulative RMS distributions, similar to Figure 1, for each individual observation of f0438.
Having demonstrated the improved performance of the NN2 method, we can test the statistical significance of the differences between the NN2 RMS values and those calculated via the single-template method and examine whether these differences indicate an actual difference in the distributions of the results from the two methods. To do so we use the non-parametric Kolmogorov-Smirnov (K-S) test, with results given in Table 2. For the sample of all single-template RMS values compared to NN2, the K-S probability value is $`5\times 10^9`$, indicating with strong confidence that the distributions are different. We also divide the sample into magnitude bins, since the relative behavior of NN2 RMS to single-template RMS is expected to be sensitive to the object’s S/N and hence to the magnitude for a given sensitivity.
The K-S probability values for three approximately equal magnitude bins show that for each of the subsamples the difference between NN2 and the single-template method is statistically significant, increasingly so at fainter magnitudes. Finally, we compare each observation individually with the NN2 method and see again that the observed improvement in RMS with NN2 is highly significant for nearly all observations (the only obvious potential exception is observation 8). These K-S probability values demonstrate that the NN2 method is not distributed identically to the single-template method, and the differences in median values given in Table 1 are indeed indicative of statistically significant differences in the distributions. This test confirms that the NN2 method truly does produce improved results in generating differential light curves.
## 4 Conclusions
We have described the mathematical foundation of a new method for constructing the light curves of photometrically variable objects. This method uses all $`N(N1)/2`$ possible subtractions involving $`N`$ images in order to calculate a vector of fluxes of the variable object and offers a powerful alternative to the single-template method that is in standard use for studying variable sources.
If one has a data set with a limited number of good fiducial observations, the NN2 method will outperform any single-template subtraction approach. For cases where a large number of fiducial observations are available to construct a deep template image, the NN2 method and the single-template approach using this deep template should yield comparable results. In this situation we would encourage the use of both methods to provide additional checks and constraints on the differential light curve.
We have tested the performance of the NN2 method by inserting artificial SNe into images from the IfA Deep Survey and comparing the RMS scatter from flux measurements using the two different methods. We find that the RMS from the NN2 method is better than the single-template RMS for the large majority (typically 65%-72%) of the SNe for every possible template. The median values for the ratio of NN2 RMS (in flux units) to single-template RMS measurements are typically $`0.93`$$`0.96`$, demonstrating that the NN2 method results in a $`5`$% improvement in the accuracy of the recovered light curve for these observations.
Using Kolmogorov-Smirnov statistics, we have demonstrated that these differences are significant, reflecting an actual difference between the performance of the two methods. We find extremely high probabilities that the NN2 RMS is distributed significantly differently from the single-template RMS values. This difference and improvement in RMS holds even for the very high quality templates that would be considered ideal for the single-template method.
We therefore make the following conclusions:
1. For the IfA Deep Survey observations, use of the NN2 method typically results in a 5-10% improvement in the RMS of the recovered light curve in comparison to the single-template method.
2. For observations that have a large external error, such as those taken under poor conditions, the NN2 method results in a substantial improvement ($`10\%`$) over the single-template method.
3. When working with high-quality observations, with small external error, the internal errors (such as those due to implementation of the subtraction process) dominate. If these errors are large, the NN2 method should outperform the single-template method to a large degree. If these errors are kept small, as we believe is possible based our extensive experience with SN surveys, then the NN2 method will result in a modest but significant improvement in accuracy of light-curve recovery.
In summary, the NN2 method we present here maximizes the time variability information contained in a series of observations by using the relative differences between all pairs of images to construct the optimal differential light curve. references
The source code for our implementation of the NN2 method presented here is available at http://www.ctio.noao.edu/essence/nn2/.
This work was supported in part by grant AST-0443378 from the United States National Science Foundation. |
warning/0507/quant-ph0507213.html | ar5iv | text | # A Method To Find Quantum Noiseless Subsystems
## Abstract
We develop a structure theory for decoherence-free subspaces and noiseless subsystems that applies to arbitrary (not necessarily unital) quantum operations. The theory can be alternatively phrased in terms of the superoperator perspective, or the algebraic noise commutant formalism. As an application, we propose a method for finding all such subspaces and subsystems for arbitrary quantum operations. We suggest that this work brings the fundamental passive technique for error correction in quantum computing an important step closer to practical realization.
Introduction. — The problem of controlling and maintaining properties of quantum systems which are in contact with an environment has received considerable recent attention. Primarily these investigations have been driven by the need to better understand the special features of evolving quantum systems that distinguish the quantum computing paradigm. Of central importance in the field of “quantum error correction” is the requirement for techniques to avoid and overcome the degrading effects of decoherence. Early work in quantum error correction included the realization that many physical error models contain symmetries induced by the system-environment interplay. This led to the discovery of “decoherence-free subspaces” (DFS) and “noiseless subsystems” (NS) Palma et al. (1996); Duan and Guo (1997); Zanardi and Rasetti (1997); Lidar et al. (1998); Knill et al. (2000); Zanardi (2001); Kempe et al. (2001). (On occasion we shall refer to both notions jointly as “NS”.) In these schemes, subspaces – and subsystems in the more abstract case – are identified within a system Hilbert space, with the property that all initial states encoded therein remain immune to the errors of a quantum operation of interest. Experimental efforts Kwiat et al. (2000); Kielpinski et al. (2001); Fortunato et al. (2002); Viola et al. (2003) have affirmed the viability of this “passive quantum error correction” (PQEC) technique.
It is also becoming clear that the NS formalism is applicable beyond the realm of quantum error correction. In quantum communication and cryptography, for instance, NS have been used as vehicles for avoiding noise Boileau et al. (2003); this may lead to practical applications of NS in the near future. Further, NS are ideal for determining how to achieve distributed quantum information processing in the absence of shared reference frames BRS03 . The NS concept has also arisen in recent analysis of black holes DMS04 , and quantum gravity KM05 , where NS are used to identify the relational symmetry-invariant physical degrees of freedom in the quantum causal history framework.
There is an obvious advantage to PQEC in the context of quantum computing. If a quantum operation (or channel) is found to possess NS, then, by taking care at the initial encoding stage, the need for active error correction after the fact is minimized. However, this protocol has a notable drawback. While substantial analysis has been carried out in important special cases, the protocol lacks a general method to find NS for arbitrary quantum operations. It is our belief that the PQEC approach will play a substantive role in quantum computing devices, and in applications beyond quantum computing, only if some sort of general approach for finding NS is derived.
In this paper we propose such a method. Specifically, we develop a structure theory that shows precisely how properties of a quantum operation as a superoperator determine its NS structure. Moreover, if an operator-sum decomposition of Kraus (or “error”) operators $`=\{E_a\}`$ for a channel $``$ is known, we show how algebraic properties of the operators $`E_a`$ determine this structure. This information naturally leads to the aforementioned method. Our analysis utilizes the framework for NS recently introduced under the umbrella of “operator quantum error correction” in KLP05 ; KLPL05 (see also Seife (2005)).
As a consequence of this work, we suggest that the fundamental passive technique for error correction in quantum computing has been brought an important step closer to practical realization. Let us discuss these points further through a pair of illustrative examples, full details of the theory will be provided below.
First we consider a simple example. Let $`=^2^2`$ be the combined system Hilbert space for two spin-$`\frac{1}{2}`$ particles. Let $`\{|00,|01,|10,|11\}`$ be the associated basis. Consider the channel $`=\{Z_1,Z_2\}`$ where $`Z_1=Z1\mathrm{l}_2`$ and $`Z_2=1\mathrm{l}_2Z`$, with the Pauli matrix $`Z=|00||11|`$. Then the action of $``$ on a density matrix $`\rho `$ on $``$ is given by $`(\rho )=\frac{1}{2}(Z_1\rho Z_1^{}+Z_2\rho Z_2^{})`$. This channel has no non-trivial NS. The key point is that the “noise commutant” $`𝒜^{}=\{Z_1,Z_2\}^{}`$, which is the set of all operators on $``$ that commute with both $`Z_1`$ and $`Z_2`$, only contains the diagonal matrices with respect to the standard basis; i.e., the matrices corresponding to classical states. On the other hand, suppose our channel is $`=\{U_1,U_2\}`$, where $`U_k=UZ_k`$, $`k=1,2`$, and $`U=1\mathrm{l}_42|1111|`$. In this case, the noise commutant $`𝒜^{}=\{U_1,U_2\}^{}`$ contains a single qubit NS (in fact it is a DFS). Indeed, any operator of the form $`\sigma =a|0000|+b|0011|+c|0011|+d|1111|`$, $`a,b,c,d`$, belongs to $`𝒜^{}`$ and satisfies $`(\sigma )=\sigma `$. As discussed below, that this NS is also fixed by $``$ follows from the fact that $``$ is unital, or bistochastic; i.e., $`(1\mathrm{l})=1\mathrm{l}`$.
As a new example of NS, and one that will also be discussed further below, we consider an error model first discussed in Knill and Laflamme (1997) in the context of active error correction. In this case the channel $`=\{E_0,E_1,E_2\}`$ acts on 2-qubit space and has three Kraus operators given by
$`E_0`$ $`=`$ $`\alpha (|0000|+|1111|)+|0101|+|1010|,`$ (1)
$`E_1`$ $`=`$ $`\beta (|0000|+|1000|+|0111|+|1111|),`$ (2)
$`E_2`$ $`=`$ $`\beta (|0000||1000||0111|+|1111|),`$ (3)
where $`q`$ is a scalar $`0<q<1`$ with $`\alpha =\sqrt{12q}`$ and $`\beta =\sqrt{q/2}`$. One can check that $`(1\mathrm{l})=_{i=0}^2E_iE_i^{}1\mathrm{l}`$, and hence $``$ is non-unital. As we show below, the noise commutant here $`𝒜^{}=\{E_0,E_1,E_2\}^{}`$ supports a single qubit NS that is not fixed by the action of $``$. Further, there is another NS for the channel, in fact a DFS, that is not contained in the noise commutant. In particular, if we define the projector $`P=|0101|+|1010|`$, then all operators supported by $`P`$ are fixed by $``$; that is, $`(\sigma )=\sigma `$ for all $`\sigma =P\sigma P`$. However, these operators do not belong to the noise commutant. For instance, notice that $`E_iP=0PE_i`$ for $`i=1,2`$. Hence, this error model has a NS inside its noise commutant that is not fixed, and a fixed DFS that is not contained in its noise commutant.
Thus, one can ask, what is the underlying phenomena that produces noiseless subsystems? The previous example indicates that we must consider more than the noise commutant and fixed point set for the map. As it turns out, the structure theory we derive for NS can be phrased in terms of more general operator algebras obtained in the same spirit as the noise commutant, and, alternatively, in terms of modified fixed point sets for the map. Therefore, our approach has the advantage of either being set in an algebraic context, or strictly in terms of properties of the superoperator.
The rest of the paper is organized as follows. We next recall the NS framework. We follow this by proving a theorem that yields the structure theory, and then show precisely how it may be used to find NS. Optimality of the method is then established, and this is followed with a conclusion on possible future work and limitations.
Noiseless Subsystem Framework. — Given a quantum operation (or “channel”), represented by a completely positive, trace preserving superoperator $`:()()`$ on a (finite dimensional) Hilbert space $``$, the NS protocol Palma et al. (1996); Duan and Guo (1997); Zanardi and Rasetti (1997); Lidar et al. (1998); Knill et al. (2000); Zanardi (2001); Kempe et al. (2001); KLP05 ; KLPL05 seeks subsystems $`^B`$ (with $`dim^B>1`$) of the full system Hilbert space $`=(^A^B)𝒦`$ such that
$`\sigma ^A\sigma ^B,\tau ^A:(\sigma ^A\sigma ^B)=\tau ^A\sigma ^B.`$ (4)
Here we have written $`\sigma ^A`$ (resp. $`\sigma ^B`$) for operators in $`(^A)`$ (resp. $`(^B)`$). In terms of partial traces, Eq. (4) can be equivalently phrased as,
$`(\mathrm{Tr}_A)(\sigma )=\mathrm{Tr}_A(\sigma ),\sigma =\sigma ^A\sigma ^B.`$ (5)
Thus, to be precise, $`B`$ is said to encode a noiseless subsystem (or decoherence-free subspace in the case $`dim^A=1`$) for $`:()()`$ when Eq. (4) is satisfied.
The basic questions we address are the following: $`(1)`$ Is there a structure theory for such subsystems? $`(2)`$ If so, can it be applied to derive a canonical method to find such subsystems for arbitrary quantum operations? Our answer to the first question is yes, and for the second we make a proposal that lends itself to the possibility of a computational algorithm.
Structure Theorem. — Let $`:()()`$ be a quantum operation. We shall write $`=\{E_a\}`$ when an error model for $``$ is known; i.e., the operation elements $`E_a`$ determine $``$ through the familiar operator-sum representation $`(\sigma )=_aE_a\sigma E_a^{}`$ Choi (1975); Kraus (1971).
The (full) noise commutant $`𝒜^{}`$ for $``$ is the set of all operators in $`()`$ that commute with the operators $`E_a`$ and $`E_a^{}`$. The $``$-algebra $`𝒜`$ generated by the $`E_a`$ is called the interaction algebra associated with $``$. In the unital case ($`(1\mathrm{l})=1\mathrm{l}`$) it is obvious that every $`\sigma 𝒜^{}`$ satisfies $`(\sigma )=\sigma `$, and, in fact, every operator that is fixed by $``$ belongs to $`𝒜^{}`$ Kribs (2003). Of course, in the general case the operator $`(1\mathrm{l})`$ may not be so well behaved, and all that can be said for operators $`\sigma 𝒜^{}`$ is that they satisfy $`(\sigma )=\sigma (1\mathrm{l})=(1\mathrm{l})\sigma `$. This equation is suggestive of the more general phenomena that must be analyzed to obtain NS for arbitrary quantum operations. Given a projection $`P`$ in $`()`$, we shall make the natural identification of the subalgebra $`P()P`$ of $`()`$ with the algebra $`(P)`$.
###### Theorem 1
Let $`=\{E_a\}`$ be a quantum operation on $`()`$. Suppose $`P`$ is a projection on $``$ such that
$`(P)=P(P)P.`$ (6)
Then $`E_aP=PE_aP`$, $`a`$. Define
$`𝒜_P^{}:=\{\sigma (P):[\sigma ,PE_aP]=0=[\sigma ,PE_a^{}P]\}.`$
and,
$`\mathrm{Fix}_P():=`$ $`\{\sigma (P):(\sigma )=\sigma (P)=(P)\sigma ,`$
$`(\sigma ^{}\sigma )=\sigma ^{}(P)\sigma ,(\sigma \sigma ^{})=\sigma (P)\sigma ^{}\},`$
Then $`\mathrm{Fix}_P()`$ is a $``$-algebra inside $`(P)`$ that coincides with the algebra $`𝒜_P^{}`$; that is,
$`\mathrm{Fix}_P()=𝒜_P^{}.`$ (7)
Proof. Let $`P`$ be a projection that satisfies Eq. (6). Then
$$0P^{}E_aPE_a^{}P^{}P^{}(P)P^{}=0a.$$
Hence $`P^{}E_aP=0`$, or equivalently $`E_aP=PE_aP`$, $`a`$. Let $`E_{a,P}:=PE_aP=E_aP`$, $`a`$. It is clear that $`\mathrm{Fix}_P()`$ contains the commutant (taken inside $`(P)`$) of the operators $`\{E_{a,P},E_{a,P}^{}\}`$. Let $`\sigma 𝒜_P^{}`$. We are required to show that $`\sigma `$ commutes with the operators $`E_{a,P}`$ and $`E_{a,P}^{}`$.
The properties $`(\sigma )=\sigma (P)=(P)\sigma `$, $`\sigma =P\sigma P`$ and Eq. (6) are seen through a calculation to imply that
$$(\sigma ^{}\sigma )\sigma ^{}(P)\sigma =\underset{a}{}[\sigma ,E_{a,P}^{}]^{}[\sigma ,E_{a,P}^{}]0.$$
(This inequality may be regarded as a generalization of the Schwarz inequality for completely positive maps from Choi (1974); Davis (1957).) Thus, given $`(\sigma )=\sigma (P)=(P)\sigma `$, and so $`(\sigma ^{})=(\sigma )^{}=\sigma ^{}(P)=(P)\sigma ^{}`$, it follows that
$`(\sigma ^{}\sigma )=\sigma ^{}\sigma (P)`$ $`\mathrm{iff}`$ $`\sigma E_{a,P}^{}=E_{a,P}^{}\sigma ,a,`$
$`(\sigma \sigma ^{})=\sigma \sigma ^{}(P)`$ $`\mathrm{iff}`$ $`\sigma E_{a,P}=E_{a,P}\sigma ,a.`$
This completes the proof. $`\mathrm{}`$
Observe that the maximally mixed state $`P=1\mathrm{l}`$ trivially satisfies Eq. (6), and the algebra $`𝒜_{1\mathrm{l}}^{}`$ coincides with the full noise commutant $`\{E_a,E_a^{}\}`$. However, as discussed above, the operator $`(1\mathrm{l})`$ may not have many nice properties. In general there may be other projections $`P`$ that support larger noiseless subsystems.
Noiseless Subsystems. — Let $`P`$ be a projection that satisfies Eq. (6). The structure theory for $``$-algebras Davidson (1996) yields a unitary $`U`$ on $`P`$ such that
$`U𝒜_P^{}U^{}={\displaystyle \underset{k}{}}(1\mathrm{l}_{m_k}M_{n_k}),`$ (8)
for a unique (up to reordering) family of positive integers $`m_k,n_k1`$. We have used $`M_{n_k}`$ to denote the operator algebra $`(^{n_k})`$, represented as matrices with respect to some orthonormal basis. Note that the algebra $`𝒜_P^{}`$ may be regarded as a subalgebra of $`()`$ simply by taking a direct sum $`𝒜_P^{}O_{n_P}`$ of $`𝒜_P^{}`$ together with the “zero algebra” $`0_{n_P}`$ of $`n_P\times n_P`$ matrices on $`P^{}`$, where $`n_P=\mathrm{dim}_km_kn_k`$.
The algebra structure Eq. (8) induces a decomposition of the subspace $`P`$ as
$`P={\displaystyle \underset{k}{}}(^{A_k}^{B_k}),`$ (9)
where $`\mathrm{dim}(^{A_k})=m_k`$ and $`\mathrm{dim}(^{B_k})=n_k`$. Observe that the positive operator $`(P)`$ belongs to the commutant inside $`(P)`$ of $`𝒜_P^{}`$ by definition. As this commutant has structure $`𝒜_P:=𝒜_P^{\prime \prime }=_k(M_{m_k}1\mathrm{l}_{n_k})`$, it follows that there are operators $`\sigma _k(^{A_k})=M_{m_k}`$ such that $`(P)=_k\sigma _k1\mathrm{l}_{n_k}`$.
Now let $`\rho =1\mathrm{l}^{A_k}\rho ^{B_k}`$ belong to the subalgebra $`1\mathrm{l}^{A_k}(^{B_k})`$ of $`𝒜_P^{}=_k(1\mathrm{l}^{A_k}(^{B_k}))`$. Then we have
$`(1\mathrm{l}^{A_k}\rho ^{B_k})=(\rho )=\rho (P)=\sigma _k\rho ^{B_k}.`$ (10)
But Eq. (4) holds if and only if it holds for $`\sigma ^A=1\mathrm{l}^A`$ KLP05 ; KLPL05 . Therefore, it follows from Eq. (10) that each of the subsystems $`^{B_k}`$ is noiseless for $``$ and the following result is established.
###### Theorem 2
Let $``$ be a quantum operation on $`()`$. Let $`P`$ be a projection on $``$ that satisfies Eq. (6) and let $`P=_k(^{A_k}^{B_k})`$ be the decomposition of $`P`$ induced by the $``$-algebra structure of $`𝒜_P^{}=\mathrm{Fix}_P()`$. Then the subsystems $`^{B_k}`$, with $`dim^{B_k}>1`$, are each noiseless subsystems for $``$.
In fact, it follows that if the input states are restricted to the subspace $`^{A_k}^{B_k}`$, then the corresponding restriction of $``$ satisfies $`(P_k()P_k)=_k\mathrm{id}_{B_k}`$ where $`P_k`$ is the projection of $``$ onto $`^{A_k}^{B_k}`$, $`_k`$ is a quantum operation on $`(^{A_k})`$, and $`\mathrm{id}_{B_k}`$ is the identity channel on $`(^{B_k})`$.
The NS structure for a number of unital channels have been analyzed in detail. An extensively studied class of channels arise from “collective noise”, which has a number of physical interpretations (see Holbrook, et al. (2005); Junge et al. (2005) for a detailed analysis of this and related NS structures). We note a connection with Bartlett et al. (2004) which includes a decomposition for collective noise channels of the form $`=_k(𝒟_k\mathrm{id}_k)(P_k\sigma P_k)`$, where the $`P_k`$ are projections associated with a decomposition of the system Hilbert space induced by underlying representation theory and the $`𝒟_k`$ are depolarizing channels (see Eq. (22) of Bartlett et al. (2004)). Interestingly, this may now be seen as a special case of the general form derived here.
Let us return to the non-unital example discussed in the Introduction. A computation shows in this case that the full noise commutant satisfies $`𝒜_{1\mathrm{l}}^{}=\{E_0,E_1,E_2\}^{}1\mathrm{l}_2M_2`$, and thus supports a single qubit NS. Indeed, if $`\sigma 𝒜_{1\mathrm{l}}^{}`$ is written as $`1\mathrm{l}_2\sigma _0`$, $`\sigma _0M_2`$, with respect to this unitary equivalence, then a calculation shows that
$$(\sigma )=(1\mathrm{l}_2\sigma _0)=\left(\begin{array}{cc}1q& 0\\ 0& 1+q\end{array}\right)\sigma _0.$$
But recall that the projection $`P=|0101|+|1010|`$ defines a DFS for $``$; specifically, $`(\sigma )=\sigma `$, $`\sigma =P\sigma P`$. Now we can see precisely how this DFS arises. Namely, $`(P)=P`$ satisfies Eq. (6) and thus we find a single qubit NS for $``$, with $`|\psi :=|01`$ and $`|\varphi :=|10`$, given by
$`𝒜_P^{}=\mathrm{span}\{|\psi \psi |,|\psi \varphi |,|\varphi \psi |,|\varphi \varphi |\}M_2.`$
Notice that $`𝒜_P^{}`$ is not contained in $`𝒜_{1\mathrm{l}}^{}`$, and thus this DFS would not be detected through an analysis of the full noise commutant alone. Further, while $`𝒜_{1\mathrm{l}}^{}`$ and $`𝒜_P^{}`$ have the same “size” from an encoding viewpoint (i.e., a single qubit) Kup03 , it is perhaps more convenient to work with $`𝒜_P^{}M_2`$, as it can be more easily isolated within the full system Hilbert space.
In fact, this example gives an indication as to how active and passive techniques for quantum error correction can be combined to combat noise. Indeed, we have just noted that the subspace $`\{|01,|10\}`$ determines a DFS for $``$. On the other hand, in Knill and Laflamme (1997) it was shown that active error correction may be used to overcome corruption by $``$ of the code subspace $`\{|00,|11\}`$.
Optimality of the Method. — The previous two sections yield a canonical method to compute noiseless subsystems for a given quantum operation $``$ which can be succinctly stated as follows:
* Compute the projections $`P`$ such that Eq. (6) holds.
* Compute the structure of the algebras $`𝒜_P^{}=\mathrm{Fix}_P()`$ as in Eq. (8).
Then, in the notation above, the subspaces $`^{B_k}`$, with $`dim^{B_k}>1`$, encode noiseless subsystems for $``$ via the operator algebras $`1\mathrm{l}^{A_k}(^{B_k})`$
A crucial final step in the process is to determine if this scheme captures all noiseless subsystems for $``$. We next show that this is indeed the case.
###### Theorem 3
Let $``$ be a quantum operation on $`()`$. Suppose that $`=(^A^B)𝒦`$ and that $`^B`$ is a noiseless subsystem for $``$ as in Eq. (4). Let $`P`$ be the projection of $``$ onto $`^A^B`$. Then $`(P)=P(P)P`$ and the algebra $`𝒜_P^{}`$ contains $`1\mathrm{l}^A(^B)`$ as a simple, unital $``$-subalgebra.
Proof. Let $`\{|\alpha _k\}`$ be an orthonormal basis for $`^A`$, and let $`\{P_{kl}=|\alpha _k\alpha _l|1\mathrm{l}^B\}`$ be the corresponding matrix units inside $`(^A)1\mathrm{l}^B`$. It was proved in KLP05 ; KLPL05 that $`^B`$ is noiseless for $`=\{E_a\}`$ as in Eq. (4) precisely when $`E_aP=PE_aP`$ and there are scalars $`\{\lambda _{akl}\}`$ such that
$`P_{kk}E_aP_{ll}=\lambda _{akl}P_{kl}a,k,l.`$ (11)
Note that the projection $`P`$ is given by $`P=_kP_k`$, where we have written $`P_k`$ for $`P_{kk}`$. Thus we have
$`(P)`$ $`=`$ $`{\displaystyle \underset{a,k}{}}E_aP_kE_a^{}={\displaystyle \underset{a,k,l,l^{}}{}}P_lE_aP_kE_a^{}P_l^{}`$
$`=`$ $`{\displaystyle \underset{a,k,l,l^{}}{}}\lambda _{alk}\overline{\lambda }_{al^{}k}P_{lk}P_{kl^{}}={\displaystyle \underset{a,k,l,l^{}}{}}\lambda _{alk}\overline{\lambda }_{al^{}k}P_{ll^{}}.`$
Let $`\sigma =1\mathrm{l}^A\sigma ^B1\mathrm{l}^A(^B)`$. Then since the $`P_{kl}`$ commute with $`\sigma =P\sigma P`$ we have
$`(\sigma )`$ $`=`$ $`(P\sigma P)=P(P\sigma P)P`$
$`=`$ $`{\displaystyle \underset{a,k,k^{},l,l^{}}{}}P_kE_aP_k^{}\sigma P_l^{}E_a^{}P_l`$
$`=`$ $`{\displaystyle \underset{a,k,k^{},l,l^{}}{}}\lambda _{akk^{}}\overline{\lambda }_{all^{}}P_{kk^{}}\sigma P_{l^{}l}`$
$`=`$ $`\sigma (P)=(P)\sigma .`$
In particular, this implies (with $`\sigma ^B=1\mathrm{l}^B`$) that $`(P)=P(P)P`$ and that the algebra $`1\mathrm{l}^A(^B)`$ is contained in $`𝒜_P^{}`$. It is clear that $`𝒜_P^{}`$ and $`1\mathrm{l}^A(^B)`$ have the same unit $`P`$, and that $`1\mathrm{l}^A(^B)`$ is a simple (i.e., contains no non-trivial ideals) $``$-subalgebra of $`𝒜_P^{}`$. $`\mathrm{}`$
We finish with a consequence for the unital case. The class of unital channels includes numerous physical error models such as collective noise, randomized unitary channels, etc. It is important to note that the full noise commutant captures all NS in this case. In particular, this means algebras $`𝒜_P^{}`$ may not be contained inside $`𝒜_{1\mathrm{l}}^{}`$ only in the non-unital case.
###### Corollary 1
Let $``$ be a unital quantum operation on $`()`$. If $`𝔄=1\mathrm{l}^A(^B)`$ is the algebra determined by a noiseless subsystem for $``$ as in Eq. (4), then $`𝔄`$ is a subalgebra of the full noise commutant $`𝒜_{1\mathrm{l}}^{}`$.
Proof. Let $`P`$ be the projection of $``$ onto $`^A^B`$. By Eq. (4), there is a $`\tau ^A`$ such that $`(P)=(1\mathrm{l}^A1\mathrm{l}^B)=\tau ^A1\mathrm{l}^B`$. Since $``$ is a unital completely positive map, we know that $`\tau ^A`$ is a contraction operator, and hence $`(P)P`$. Then in fact $`(P)=P`$ by Lemma 2.3 from Kribs (2003). Thus, it follows from Theorem 3, and the definition of $`𝒜_P^{}`$, that $`𝔄𝒜_P^{}`$ is a subalgebra of $`\mathrm{Fix}()`$. $`\mathrm{}`$
Conclusion. — We have derived a structure theory for decoherence-free subspaces and noiseless subsystems that applies to arbitrary quantum operations. As an application, we have proposed a method to compute NS for any given operation. We expect that the method could be formalized into a computational algorithm, as suggested by recent literature Holbrook, et al. (2004); Zarikian (2005) which includes algorithms written to calculate operator algebra structures, but there are still details to work through. We plan to undertake this investigation elsewhere.
We discussed a non-unital example in which the maximally mixed state and a smaller projection support different single qubit noiseless subsystems. We suggest that this work motivates reconsideration of the quantum channels that appear in the literature, for the possible existence of noiseless subsystems. We wonder about possible experimental implications of this work. It would also be interesting to investigate connections with other recent noiseless subsystem related efforts such as DMS04 ; LS05 ; KM05 .
Acknowledgements. We thank Dietmar Bisch for asking a question that partly motivated this work. We are grateful to John Holbrook, Raymond Laflamme, Rob Spekkens, and Karol Zyczkowski for helpful comments. D.W.K. would also like to thank other colleagues at UofG, IQC and Perimeter Institute for interesting discussions. This work was partially supported by NSERC. |
warning/0507/cond-mat0507444.html | ar5iv | text | # Synchronous versus sequential updating in the three-state Ising neural network with variable dilution
## 1 Introduction
The dynamics and the storage and retrieval properties of multi-state attractor neural networks have been studied over some time now and numerous results are available (see, e.g., and references therein). The majority of the results obtained on the storage and retrieval properties concern sequential updating of the neurons.
Recently, it has been realized that synchronous updating of the spins in disordered systems can lead to different physics -. For example, for binary spins, it is known that the phase diagram of the sequential and synchronous Little-Hopfield neural network in the replica-symmetric approximation are different , whereas the phase diagrams of the Sherrington-Kirkpatrick model are the same . For the three-state Ising ferromagnet the same stationary solutions appear except for negative couplings, while for the Blume-Emery-Griffiths ferromagnet the phase diagram for synchronous updating is much richer .
For the three-state Ising neural network the possible different physics between sequential and synchronous updating has not yet been studied. Looking at the literature we see that using sequential updating the equilibrium properties of the $`Q`$-Ising model for the fully connected architecture have been studied in and making a replica symmetric ansatz. The results for the extremely diluted architecture appeared in and have been extended later to the whole dilution range in . The layered architecture with variable dilution has been examined recently . Nothing has been reported on the equilibrium properties for the recurrent architectures, however, when synchronous updating is used. On the other hand, concerning dynamics no calculations were done for the $`Q`$-Ising model with sequential updating. For synchronous updating the work of on the exactly solvable extremely diluted asymmetric Little-Hopfield model has been extended to the $`Q`$-Ising model in . Later on, the dynamics for the fully connected and the extremely diluted symmetric $`Q`$-Ising architecture have been solved in and, respectively, . The latter studies make use of the so-called signal-to-noise analysis (see, e.g., for references on this method). An extension to the whole dilution range, in analogy with has not yet been given.
The aim of this work is precisely to fill the gaps mentioned above with a report on the study of the $`Q=3`$-Ising network with synchronous updating and variable dilution and a detailed comparison of the results obtained with the ones for sequential updating. First, the thermodynamic and retrieval properties are examined using replica symmetric mean-field theory. Capacity-temperature phase diagrams are derived for several values of the pattern activity and different gradations of dilution. Apart from the appearance of cycles the asymptotic behaviour is almost identical to the one for sequential updating. The spin-glass region is visibly enhanced, while the retrieval region, however, is only marginally enhanced. Only the addition of self-couplings can enlarge the retrieval region substantially, especially in the case of strong dilution. A calculaton of the information content shows that both for synchronous and sequential updating the three-state networks are robust against the interference of static noise coming from random dilution. Next, the dynamics of the model is studied using the generating function technique . As an illustration some typical flow diagrams for the overlap order parameter are presented. It is possible to extract the result for sequential updating from the one for synchronous updating. As in the Hopfield model one can argue that the signal-to-noise analysis used before in the literature is a short memory approximation correct up to the third time step. And it can also be shown that the signal-to-noise analysis can be made exact.
The rest of the paper is organized as follows. In Section 2 the three-state Ising neural network with synchronous updating and variable dilution is introduced. Section 3 reports on the replica symmetric mean field theory calculation of the free energy and the fixed-point equations for the relevant order parameters. In Section 4 the phase diagrams and retrieval properties are discussed for arbitrary temperatures as a function of the gain parameter, the amount of dilution and the strength of the self-coupling. Section 5 discusses the dynamics for the model using the generating functional analysis and comments on the relation with the signal-to-noise analysis. In Section 6 some concluding remarks are given. Finally, the appendix presents the explicit saddle-point equations.
## 2 The three-state Ising neural network
Consider a network of N neurons, $`𝝈=\{\sigma _1,\mathrm{},\sigma _N\}`$, which can take values from the set $`𝒮=\{1,0,1\}`$. In this network we want to store $`p=\alpha N`$ patterns, $`\{\xi _i^\mu \}`$, $`i=1,\mathrm{},N`$ and $`\mu =1,\mathrm{},p`$. They are supposed to be independent identically distributed random variables (i.i.d.r.v.) with respect to $`i`$ and $`\mu `$, drawn from a probability distribution given by
$$\text{P}(\xi _i^\mu )=a\delta (1(\xi _i^\mu )^2)+(1a)\delta (\xi _i^\mu ),$$
(1)
with $`a`$ the pattern activity defined by the expectation value
$$\text{E}((\xi _i^\mu )^2)=a.$$
(2)
The neurons are updated synchronously according to the transition probability
$`\text{W}\left[𝝈(t+1)|𝝈^{\mathbf{}}(t)\right]={\displaystyle \underset{i=1}{\overset{N}{}}}\text{Pr}\left(\sigma _i(t+1)=s𝒮|𝝈^{\mathbf{}}(t)\right)`$ (3)
$`\text{Pr}\left(\sigma _i(t+1)=s𝒮|𝝈^{\mathbf{}}(t)\right)={\displaystyle \frac{\mathrm{exp}[\beta ϵ_i(s|𝝈^{\mathbf{}}(t))]}{{\displaystyle \underset{s𝒮}{}}\mathrm{exp}[\beta ϵ_i(s|𝝈^{\mathbf{}}(t))]}}`$ (4)
with $`\beta `$ the inverse temperature and $`ϵ_i(s|𝝈)`$ an effective single site energy function given by
$$ϵ_i(s|𝝈)=\left[\frac{1}{2}h_i(𝝈)sbs^2\right],$$
(5)
where $`b`$ is the gain parameter of the system suppressing or enhancing the zero state of the neurons. The random local fields are defined by
$$h_i(𝝈)=\underset{j=1}{\overset{N}{}}J_{ij}^c\sigma _j.$$
(6)
The couplings $`J_{ij}^c`$ are taken to be of the form
$$J_{ij}^c=\frac{c_{ij}}{c}J_{ij},$$
(7)
where the probability distribution of the $`\{c_{ij}\}`$ is given by
$$\text{P}(c_{ij})=c\delta (c_{ij}1)+(1c)\delta (c_{ij}).$$
(8)
Hence, they allow for a diluted architecture. The $`J_{ij}`$ are determined via the Hebb rule
$$J_{ij}=\frac{1}{aN}\underset{\mu =1}{\overset{p}{}}\xi _i^\mu \xi _j^\mu $$
(9)
In order for the model to satisfy detailed balance (see,e.g., ), the dilution has to be symmetric ($`c_{ij}=c_{ji}`$). In the case of extreme dilution, when $`c=0`$, the average number of connections per neuron, $`cN`$, is still infinite
$$\underset{N\mathrm{}}{lim}\frac{1}{cN}=0.$$
(10)
Finally, we recall that the detailed balance property for synchronous updating is not destroyed by the presence of self-couplings, i.e., couplings of the form $`J_{ii}^c`$. Hence, we do allow for this type of couplings and redefine $`J_{ii}^cJ_0J_{ii}^c=\alpha J_0`$, with $`J_0`$ a parameter and $`\alpha =p/cN`$ the capacity.
The long-time behaviour is governed by the Hamiltonian
$$H(𝝈)=\frac{1}{\beta }\underset{i=1}{\overset{N}{}}\mathrm{ln}\left[\underset{s𝒮}{}\mathrm{exp}(\beta [h_i(𝝈)sbs^2])\right]+b\underset{i=1}{\overset{N}{}}\sigma _i^2.$$
(11)
In addition, when evaluating traces over spins in the calculation of, e.g., the partition function, one realizes that the system is equivalent to one with a Hamiltonian involving a set of duplicate Ising spins (see, e.g., ), which can be written as
$$H(𝝈,𝝉)=\underset{i,j}{}J_{ij}^c\sigma _i\tau _j+b\underset{i}{}[\sigma _i^2+\tau _i^2]$$
(12)
such that $`\underset{𝝈}{\text{Tr}}\mathrm{exp}[\beta H(𝝈)]=\underset{𝝈}{\text{Tr}}\underset{𝝉}{\text{Tr}}\mathrm{exp}[\beta H(𝝈,𝝉)]`$.
It is well-known that the equilibrium behaviour can be fixed-points and/or cycles of period 2, i.e., $`\sigma _i(t)=\sigma _i(t+2),i`$.
In the next Section we study the thermodynamic and retrieval properties of this model starting from the free energy.
## 3 Replica mean-field theory
In order to calculate the free energy we use the replica method as applied to dilute systems ,,-. Since this method is really standard by now, at least for sequential updating, we refrain from giving any detailed calculations but concentrate on the main results and the differences between sequential and synchronous updating. Indeed, some complications arise due to the symmetry between the two types of spins in the Hamiltonian.
Starting from the replicated partition function, performing the dilution average and the average over the condensed ($`\mu =1`$) and non-condensed ($`\mu >1`$) patterns we obtain for the replicated free energy density
$`f_n=a𝒎\stackrel{~}{𝒎}+{\displaystyle \frac{\alpha c}{2\beta }}\text{Tr\hspace{0.17em}log}\left(𝑰\beta 𝑨\right)`$
$`(1c){\displaystyle \frac{\alpha \beta }{2}}\text{Tr}\left(𝒒𝒑+𝒓^2\right)\alpha (J_0c)\text{Tr}\left(𝒓\right)`$
$`+{\displaystyle \frac{\alpha \beta }{2}}\text{Tr}\left(𝒒\widehat{𝒒}+𝒑\widehat{𝒑}+2𝒓\widehat{𝒓^{}}\right){\displaystyle \frac{1}{\beta }}\mathrm{log}\underset{𝝈𝝉}{\text{Tr}}\mathrm{exp}\left(\beta \stackrel{~}{H}(𝝈,𝝉)\right)_\xi ,`$ (13)
with
$$\stackrel{~}{H}(𝝈,𝝉)=\xi (𝒎𝝈+\stackrel{~}{𝒎}𝝉)\frac{\alpha \beta }{2}\left(𝝈\widehat{𝒒}𝝈^{}+𝝉\widehat{𝒑}𝝉^{}+2𝝈\widehat{𝒓}𝝉^{}\right)\beta b\left(𝝈^2+𝝉^2\right).$$
(14)
The symbol $`𝑰`$ denotes the unit matrix in replica space and
$$𝑨=\left(\begin{array}{cc}𝒒& i𝒓\\ i𝒔& 𝒑\end{array}\right).$$
Hereby the usual order parameters are introduced as replica matrices ($`n`$ is the replica index) with elements
$`m_\alpha ^\mu ={\displaystyle \frac{1}{aN}}{\displaystyle \underset{i}{}}\xi _i^\mu \sigma _i^\alpha ,\stackrel{~}{m}_\alpha ^\mu ={\displaystyle \frac{1}{aN}}{\displaystyle \underset{i}{}}\xi _i^\mu \tau _i^\alpha ,`$ (15)
$`q_{\alpha \beta }={\displaystyle \frac{1}{N}}{\displaystyle \underset{i}{}}\sigma _i^\alpha \sigma _i^\beta ,p_{\alpha \beta }={\displaystyle \frac{1}{N}}{\displaystyle \underset{i}{}}\tau _i^\alpha \tau _i^\beta ,`$ (16)
$`r_{\alpha \beta }={\displaystyle \frac{1}{N}}{\displaystyle \underset{i}{}}\sigma _i^\alpha \tau _i^\beta ,s_{\alpha \beta }={\displaystyle \frac{1}{N}}{\displaystyle \underset{i}{}}\tau _i^\alpha \sigma _i^\beta .`$ (17)
Their conjugate variables are denoted with a hat. The free energy should be interpreted as being extremised with respect to $`𝒎`$, $`\widehat{𝒎}`$, $`𝒒`$, $`𝒑`$ and $`𝒓`$. We remark that the corresponding result for sequential dynamics (see, e.g., ) is recovered by assuming $`𝝈=𝝉`$ and rescaling the temperature with a factor 2. Furthermore, we notice that compared with sequential updating, the replicated free energy (13) is more involved. It is clear that a priori the matrices appearing in this expression are not necessarily symmetric.
As mentioned in the introduction it is known that synchronous updating can lead to two-cycles as stationary solutions. Results for the Hopfield model and the Blume-Emery-Griffiths model show that such cycles do not appear in the retrieval region of the corresponding phase diagrams. Therefore, cycles have been neglected in the replica calculation of the retrieval properties of these models. In analogy, for the three-state Ising model discussed here we make a similar ansatz. We neglect cycles in the sequel of the replica calculation and we assume that the two sets of spins behave in a completely symmetric way. Consequently, $`𝒒=𝒑`$, $`𝒓=𝒓^{}`$, and similarly for the conjugate variables. The same symmetry applies to $`𝒎`$ and $`\stackrel{~}{𝒎}`$. The free energy then becomes
$`f=a𝒎^2+{\displaystyle \frac{\alpha c}{2\beta }}\text{Tr\hspace{0.17em}log}\left((𝑰\beta 𝒓)^2\beta ^2𝒒^2\right)`$
$`(1c){\displaystyle \frac{\alpha \beta }{2}}\text{Tr}\left(𝒒^2+𝒓^2\right)\alpha (J_0c)\text{Tr}\left(𝒓\right)`$
$`+\alpha \beta \text{Tr}\left(𝒒\widehat{𝒒}+𝒓\widehat{𝒓}\right){\displaystyle \frac{1}{\beta }}\mathrm{log}\underset{𝝈𝝉}{\text{Tr}}\mathrm{exp}\left(\beta \stackrel{~}{H}(𝝈,𝝉)\right)_\xi ,`$ (18)
with
$$\stackrel{~}{H}(𝝈,\tau )=\xi 𝒎(𝝈+𝝉)\frac{\alpha \beta }{2}\left(𝝈\widehat{𝒒}𝝈^{}+𝝉\widehat{𝒒}𝝉^{}+2𝝈\widehat{𝒓}𝝉^{}\right)\beta b\left(𝝈^2+𝝉^2\right).$$
(19)
Next, we take the replica symmetry (RS) ansatz
$$m_\alpha =m,q_{\alpha \beta }=q_1,q_{\alpha \alpha }=q_0,r_{\alpha \beta }=r_1,r_{\alpha \alpha }=r,\alpha \beta $$
(20)
and similarly for the conjugated variables. Due to the non-cycle ansatz, we have in addition
$$r_1=r_{\alpha \beta }=q_{\alpha \beta }=q_1$$
(21)
such that the RS free energy reads
$`f=am^2+{\displaystyle \frac{\alpha c}{2\beta }}\mathrm{log}\left[(1\chi _r)^2\chi ^2\right]\alpha c{\displaystyle \frac{q}{1\chi _r\chi }}`$
$`(1c){\displaystyle \frac{\alpha \beta }{2}}\left(q_0^22q^2+r^2\right)+\alpha \beta \left(\widehat{q}_0q_02\widehat{q_1}q_1+\widehat{r}r\right)`$
$`+\alpha (cJ_0)r{\displaystyle \frac{1}{\beta }}{\displaystyle 𝒟z\mathrm{log}\underset{\sigma \tau }{\text{Tr}}\mathrm{exp}\left(\beta \stackrel{~}{H}(\sigma ,\tau |z)\right)}_\xi ,`$ (22)
with the effective Hamiltonian
$$\stackrel{~}{H}(\sigma ,\tau |z)=\left(\xi m+\sqrt{\alpha \widehat{q}_1}z\right)(\sigma +\tau )+\left(b\frac{1}{2}\alpha \widehat{\chi }\right)(\sigma ^2+\tau ^2)\alpha \widehat{\chi }_r\sigma \tau $$
(23)
and $`𝒟z`$ the gaussian measure $`𝒟z=dz(2\pi )^{1/2}\mathrm{exp}(z^2/2)`$. We have also defined the susceptibilities
$$\chi =\beta (q_0q_1),\chi _r=\beta (rq_1),$$
(24)
and their conjugate expressions
$$\widehat{\chi }=\beta (\widehat{q}_0\widehat{q}_1),\widehat{\chi }_r=\beta (\widehat{r}\widehat{q}_1).$$
(25)
The phase structure of the network is then determined by the solution of the following set of saddle-point equations
$`m={\displaystyle \frac{1}{a}}\xi {\displaystyle 𝒟z\sigma _z}_\xi ,r={\displaystyle 𝒟z\sigma \tau _z}_\xi ,`$ (26)
$`q_0={\displaystyle 𝒟z\sigma ^2_z}_\xi ,q_1={\displaystyle 𝒟z\sigma _z^2}_\xi ,`$ (27)
where the average $`_z`$ is defined with respect to the effective Hamiltonian (23) and
$`\widehat{q}_1=q_1\left[(1c)+c{\displaystyle \frac{1}{(1\chi _r\chi )^2}}\right],`$ (28)
$`\widehat{\chi }=(1c)\chi +c{\displaystyle \frac{\chi }{(1\chi _r)^2\chi ^2}},`$ (29)
$`\widehat{\chi }_r=(1c)\chi _r+c{\displaystyle \frac{1\chi _r}{(1\chi _r)^2\chi ^2}}+(Jc).`$ (30)
Compared with sequential updating we notice the extra equations for $`r`$ and $`\stackrel{~}{\chi }_r`$ expressing that $`\sigma `$ and $`\tau `$ can be affected differently by thermal fluctuations and the fact that the equations for the conjugate parameters are different. In the appendix we present the explicit forms for (26)-(27).
## 4 Phase diagrams and retrieval properties
First, we look at the special case of low loading ($`\alpha =0`$). In that case the effective Hamiltonian (23) simplifies to
$$\stackrel{~}{H}(\sigma ,\tau |z)=\xi m(\sigma +\tau )+b(\sigma ^2+\tau ^2),$$
(31)
which practically means that the two spin variables occurring in the expression for the free energy (22) become independent yielding $`f_{\text{par}}=2f_{\text{seq}}`$ as found before for other models . Consequently, the saddle point equations are the same for synchronous and sequential updating and both yield the same stationary states. An explicit calculation shows that the low loading results correspond to those of the $`3`$-Ising ferromagnet. For the relevant phase diagram we refer to (Fig. 2 for $`J>0`$).
Next, we turn to finite loading and solve the saddle point equations (26)-(30) numerically for arbitrary temperature $`T`$. We present the phase diagrams for some representative values of the parameters $`c`$, $`b`$ and $`J_0`$. We focuss on uniformly distributed patterns ($`a=2/3`$). Results for sequential updating of the network are included for comparison.
In Figure 1 we present the RS $`\alpha T`$ phase diagrams with ($`J_0=1`$) and without ($`J_0=0`$) self-coupling for the fully connectivity architecture ($`c=1`$) and gain parameter $`b=0.2`$ and $`b=0.5`$. The retrieval transition is discontinuous in all cases. For a fixed $`T`$, the retrieval capacity of the synchronously updated network with self-coupling is slightly larger than the one for the network without self-coupling, and both are slightly larger than the one for the sequentially updated network. However, the enhancement stays marginal, also for growing self-coupling. The spin glass transition is always continuous. For $`b=0.5`$, contrary to $`b=0.2`$, the spin glass phase is not stable at small $`\alpha `$. But the retrieval phase is stable. This means that for $`b=0.5`$, small $`\alpha `$ and low $`T`$ the retrieval phase co-exists only with the paramagnetic phase, and not with the spin glass phase.
In Figure 2 we show the RS phase diagrams for the symmetrically diluted networks with $`c=0.01`$. The other network parameters are kept identical to those in Figure 1. The retrieval (spin-glass) transitions remain discontinuous (continuous). The enhancement of the retrieval capacity is again marginal for the network with synchronous updating and without self-coupling, compared to the one for sequential updating. However, increasing the self-coupling shows a substantial improvement, except for $`\alpha =0`$. Also the spin-glass region is visibly enlarged in that case. The remark concerning the stability of the spin-glass phase for $`b=0.5`$ and low $`T`$ also applies to the diluted network, but with a small change: one remarks (see the bottom left corner of Figure 2 (b)) that the spin-glass phase becomes stable at small $`\alpha `$ and low $`T`$, although there remains a region where only the retrieval and paramagnetic phases are stable. Finally, we notice a strong re-entrant behaviour in the retrieval region. This is related to the replica-symmetric approximation and is also seen in the Hopfield model . Consequently, we conjecture that the fact that for $`b=0.5`$, e.g., the maximal critical capacity for the network with self-coupling is larger compared to the one for the network without self-coupling, is an effect of this approximation.
The RS phase diagrams for the extremely diluted symmetric ($`c=0`$) network are shown in Figure 3. The other parameters are kept identical to those in Figures 1 and 2. The results for sequential and synchronous updating without self-coupling coincide. All transitions are continuous. Again, we notice a strong re-entrance behaviour in the retrieval region, in agreement with earlier results for this extremely diluted limit . Increasing the self-coupling makes the synchronously updated network more robust against temperature, although, as expected, it has no effect on the critical temperature at zero $`\alpha `$. The re-entrance point is reached for the same value of $`\alpha `$.
Finally, we briefly study the robustness of these three-state networks against the interference of static noise coming from random dilution. In Fig. 4 we show the information content of the model, being the product of the loading capacity and the mutual information , for synchronous updating and several amounts of dilution $`c`$ with and without self-coupling.
We see that for the fully connected network ($`c=1.0`$, left figure), the results for $`J_0=0`$ and $`J_0=1`$ do practically coincide, in agreement with Fig. 1. We remark that for b=0.2, two solutions co-exist at small $`\alpha `$ ($`0\alpha <0.007`$). One of them corresponds to perfect retrieval, with $`m=1.0`$, $`q_1=q_0=a,\chi =0`$. The corresponding information content is very close to that for $`b=0.5`$. For the diluted network ($`c=0.01`$, right figure), in agreement with Fig. 2, self-coupling mostly leads to a higher information content for the optimal gain parameter $`b=0.5`$. (Again, for $`b=0.2`$ there exist two solutions for small $`\alpha `$ but the difference is very small and can hardly be seen on the scale of the figure.)
Next, by comparing with Fig. 5 for sequential updating, one notices that the results for sequential and synchronous updating (without self-coupling) are almost coincident. In all cases, we find that the quality of the retrieval properties is affected very little, unless the amount of dilution is high.
## 5 Dynamics
As mentioned in the introduction most studies in the literature on the dynamics of the $`Q`$-Ising model are based upon the signal-to-noise analysis. Except for the extremely diluted asymmetric and layered architectures, which can be solved exactly in closed form, the dynamics for the other architectures, only examined for synchronous updating, is obtained in the form of a recursive scheme. Recently, by a comparison with the generating functional analysis (GFA), it has been found for the Hopfield model that the application of this method in the study of the dynamics involves a short-memory approximation implying that the results are only exact up to the third time step, although they stay very accurate in the retrieval region for further time steps, but not so in the spin-glass region. It has also been shown that the signal-to-noise analysis can be made exact, leading to the same results as the generating functional analysis.
Therefore, it is interesting to reconsider the dynamics for the $`Q=3`$-state Ising model using this generating functional technique, which was introduced in to the field of statistical mechanics and, by now, is part of many textbooks. We closely follow the derivation in , extend it to variable dilution and indicate the differences for the model at hand. Both sequential and synchronous updating are discussed. In fact, the discussion is made for general $`Q`$. Since the method itself has become rather standard, we restrict ourselves to a presentation of the main arguments.
The idea of the GFA approach to study dynamics is to look at the probability to find a certain microscopic path in time. The basic tool to study the statistics of these paths is the generating functional
$$Z[𝝍]=\underset{𝝈(0),\mathrm{},𝝈(t)}{}P[𝝈(0),\mathrm{},𝝈(t)]\underset{i=1}{\overset{N}{}}\underset{t^{}=0}{\overset{t}{}}e^{i\psi _i(t^{})\sigma _i(t^{})}$$
(32)
with $`P[𝝈(0),\mathrm{},𝝈(t)]`$ the probability to have a certain path in phase space
$$P[𝝈(0),\mathrm{},𝝈(t)]=P[𝝈(0)]\underset{t^{}=0}{\overset{t1}{}}W[𝝈(t^{}+1)|𝝈(t^{})]$$
(33)
and $`W[𝝈(t^{}+1)|𝝈(t^{})]`$ the transition probabilities from $`𝝈^{}`$ to $`𝝈`$. Synchronous updating is expressed by (recall eq. (4)).
$$W[𝝈(t^{}+1)|𝝈(t^{})]=\underset{i=1}{\overset{N}{}}\frac{\mathrm{exp}\left(\beta \sigma _i(t^{}+1)ϵ_i(\sigma _i(t^{}+1)|𝝈(t^{}))\right)}{\underset{𝜎}{\text{Tr}}\mathrm{exp}\left(\beta \sigma ϵ_i(\sigma |𝝈(t^{}))\right)}$$
(34)
where $`ϵ_i(\sigma _i(t^{}+1)|𝝈(t^{}))`$ now includes a time-dependent external field $`\theta _i(s)`$ in order to define a response function
$$ϵ_i(\sigma _i(t^{}+1)|𝝈(t^{}))=\left[\frac{1}{2}h_i(𝝈(t^{}))\sigma (t^{}+1)b(\sigma (t^{}+1))^2\right]+\theta _i(t^{}).$$
(35)
One can find all the relevant order parameters, i.e., the overlap $`m(t)`$, the correlation function $`C(t,t^{})`$ and the response function $`G(t,t^{})`$, by calculating appropriate derivatives of the above functional (32) and letting $`𝝍=\{\psi _i\}`$ tend to zero afterwards
$`m(t)`$ $`=`$ $`i\underset{𝝍0}{lim}{\displaystyle \frac{1}{aN}}{\displaystyle \underset{i}{}}\xi _i{\displaystyle \frac{\delta Z[𝝍]}{\delta \psi _i(t)}},`$ (36)
$`G(t,t^{})`$ $`=`$ $`i\underset{𝝍0}{lim}{\displaystyle \frac{1}{N}}{\displaystyle \underset{i}{}}{\displaystyle \frac{\delta ^2Z[𝝍]}{\delta \psi _i(t)\delta \theta _i(t^{})}},`$ (37)
$`C(t,t^{})`$ $`=`$ $`\underset{𝝍0}{lim}{\displaystyle \frac{1}{N}}{\displaystyle \underset{i}{}}{\displaystyle \frac{\delta ^2Z[𝝍]}{\delta \psi _i(t)\delta \psi _i(t^{})}}.`$ (38)
A difference with the Little-Hopfield model is the presence of the factor $`1/a`$ in (36). First, due to the proper scaling of the couplings with this factor (recall eq. (9)) the average over the non-condensed patterns does not introduce an extra factor. Secondly, in the saddle point, terms containing this extra factor vanish. As a consequence, the factor $`1/a`$ does not appear explicitly in the expressions at all. The only point to keep in mind is the occurrence of the term proportional to the factor $`b`$ in the dynamics (recall eq. (5)).
Another difference with respect to the treatment in is that one needs to average over the dilution. This is done before averaging over the non-condensed patterns by using the fact that the diluted couplings $`J_{ij}^c`$ are of order $`𝒪((cN)^{1/2})`$ and that the $`J_{ij}𝒪(N^{1/2})`$. Noting that the distribution for the dilution eq. (8) is i.i.d.r.v. for $`i<j`$ and expanding the exponential containing the $`J_{ij}^c`$ up to order $`𝒪(N^{3/2})`$ makes this average then straightforward.
In the thermodynamic limit one expects the physics of the problem to be independent of the quenched disorder and, therefore, one is interested in derivatives of $`\overline{Z[𝝍]}`$, whereby the overline denotes the average over this disorder, i.e., over all pattern realizations. This results in an effective single spin local field given by
$$h(t)=\xi m(t)+\alpha \underset{t^{}=0}{\overset{t1}{}}R(t,t^{})\sigma (t^{})+\sqrt{\alpha }\eta (t)$$
(39)
with $`\eta (t)`$ temporally correlated noise with zero mean and correlation matrix
$$𝐃=c(𝑰𝐆)^1𝐂(𝑰𝐆^{})^1+(1c)𝐂$$
(40)
and the retarded self-interactions
$$𝐑=c(𝑰𝐆)^1+(1c)𝐆+(J_0c)𝑰.$$
(41)
The order parameters defined above can be written as
$`m(t)`$ $`=`$ $`\xi \sigma (t)_{},`$ (42)
$`C(t,t^{})`$ $`=`$ $`\sigma (t)\sigma (t^{})_{},`$ (43)
$`G(t,t^{})`$ $`=`$ $`{\displaystyle \frac{\sigma (t)}{\theta (t^{})}}_{}.`$ (44)
The average over the effective path measure $``$ is given by
$$f_{}=\text{Tr}_{\{\sigma (1),\mathrm{},\sigma (t)\}}𝑑𝜼P(𝜼)P(𝝈|𝜼)f,$$
(45)
where $`d𝜼=_t^{}\eta (t^{})`$ and with
$`P(𝝈|𝜼)={\displaystyle \underset{t^{}=0}{\overset{t1}{}}}{\displaystyle \frac{\mathrm{exp}(\beta \sigma (t^{}+1)h(t^{})\beta b\sigma ^2(t^{}+1))}{\underset{𝜎}{\text{Tr}}\left(\mathrm{exp}(\beta \sigma (t^{}+1)h(t^{})\beta b\sigma ^2(t^{}+1))\right)}}`$ (46)
$`P(𝜼)={\displaystyle \frac{1}{\sqrt{det(2\pi 𝐃)}}}\mathrm{exp}\left({\displaystyle \frac{1}{2}}{\displaystyle \underset{t^{},t^{\prime \prime }=0}{\overset{t1}{}}}\eta (t^{})𝐃^1(t^{},t^{\prime \prime })\eta (t^{\prime \prime })\right).`$ (47)
where $`h(t^{})=\xi m(t^{})+\alpha _pR(t^{},p)\sigma (p)+\eta (t^{})`$. The average denoted by the double brackets in (44) is over the condensed pattern and initial conditions. The set of eqs. (39)-(41) and (45)-(47) represent an exact dynamical scheme for the evolution of the network from which all relevant order parameters can be obtained at all time steps.
The order parameters are given by (42)-(44). To acquire a more intuitive expression for the response function we note that also
$$G(t,t^{})=\frac{\sigma (t)}{\eta (t^{})}_{}.$$
(48)
We remark that $`G(t^{},t^{\prime \prime })=0`$ for $`t^{}t^{\prime \prime }`$ and $`D(t^{},t^{\prime \prime })=D(t^{\prime \prime },t^{})`$ and that for all $`t^{}<t`$
$$G(t,t^{})=\beta \sigma (t)\left[\sigma (t^{}+1)\frac{\underset{𝜎}{\text{Tr}}\sigma \mathrm{exp}\left(\beta ϵ_i(\sigma _i(t^{}+1)|𝝈(t^{}))\right)}{\underset{𝜎}{\text{Tr}}\mathrm{exp}\left(\beta \sigma ϵ_i(\sigma |𝝈(t^{}))\right)}\right]_{}$$
(49)
where $`h(t^{})`$ appearing in $`ϵ_i`$ is given by (39).
For sequential updating we have to start from the stochastic process
$$p_{s+1}(𝝈)=\underset{𝝈^{}}{}W_s[𝝈;𝝈^{}]p_s(𝝈^{})$$
(50)
with $`p_{s+1}(𝝈)`$ the probability to be in a state $`𝝈`$ at time $`s+1`$ and $`W_s[𝝈;𝝈^{}]`$ given by
$$W_s[𝝈;𝝈^{}]=\frac{1}{N}\underset{i}{}\left\{w_i(𝝈)\delta _{𝝈,𝝈^{}}+w_i(F_i𝝈)\delta _{𝝈,G_i𝝈^{}}+w_i(G_i𝝈)\delta _{𝝈,F_i𝝈^{}}\right\},$$
(51)
with the shorthand $`w_i(𝝈)\text{Pr}\{\sigma _i(s+1)=\sigma _i|𝝈(s)\}`$ and where $`F_i`$ and $`G_i`$ are cyclic spin-flip operators between the spin states. Each time step a randomly chosen spin is updated. In the thermodynamic limit the dynamics becomes continuous because the characteristic time scale is $`N^1`$. The standard procedure is then to update a random spin according to (4) and (5) with time intervals $`\mathrm{\Delta }`$ that are Poisson distributed with mean $`N^1`$ . We can then write a continuous master equation in the thermodynamic limit
$`{\displaystyle \frac{d}{ds}}p_s(𝝈)`$ $``$ $`\underset{\mathrm{\Delta }0}{lim}{\displaystyle \frac{p_{s+\mathrm{\Delta }}(𝝈)p_s(𝝈)}{\mathrm{\Delta }}}`$
$`=`$ $`{\displaystyle \underset{i}{}}\left\{(w_i(𝝈)1)p_s(𝝈)+w_i(F_i𝝈)p_s(F_i𝝈)+w_i(G_i𝝈)p_s(G_i𝝈)\right\}`$
In that case the effective path average reads
$$f_{}=𝑑𝜼P(𝜼)[f]_𝜼,$$
(53)
where $`[f]_𝜼`$ is an average over the (effective) stochastic process conditioned to the noise $`𝜼`$ generated by the dynamics (5) and the distribution $`P(𝜼)`$ is Gaussian with correlation matrix $`𝐃`$
$$P(𝜼)=\frac{1}{\sqrt{det(2\pi 𝐃)}}\mathrm{exp}\left(\frac{1}{2}𝑑t^{}𝑑t^{\prime \prime }\eta (t^{})𝐃^1(t^{},t^{\prime \prime })\eta (t^{\prime \prime })\right).$$
(54)
This result is very similar to the one for synchronous dynamics (recall eq. 47). The main reason is that one does not need the explicit form of the transition rates of the Markovian process in order to derive the effective path average. If only the initial conditions factorize over the site index $`i`$ then all sites become independent in the thermodynamic limit. In general, this is a characteristic feature of mean-field systems.
Specialising this discussion to three Ising states $`\{1,0,+1\}`$ we have that for synchronous updating the trace in eq. (46) is given by
$$\underset{𝜎}{\text{Tr}}(\mathrm{exp}(\beta \sigma (t^{}+1)h(t^{})\beta b\sigma ^2(t^{}+1))=1+2\mathrm{cosh}(\beta h(t^{}))\mathrm{exp}(\beta b).$$
(55)
For sequential updating the spin-flip operators in eq. (51) are defined by
$`F_i\mathrm{\Phi }(𝝈)=\mathrm{\Phi }(\sigma _1,\mathrm{},\sigma _{i1},{\displaystyle \frac{3\sigma _i^2\sigma _i+2}{2}},\sigma _{i+1},\mathrm{},\sigma _N)`$
$`G_i\mathrm{\Phi }(𝝈)=F_i(F_i\mathrm{\Phi }(𝝈)).`$ (56)
We have solved the dynamics numerically using the Eissfeller-Opper method . The idea thereby is to replicate the system into $`M`$ copies of the effective spin and let each of them evolve independently in time according to their own stochastic path with its own noise variable. Averages over the effective path measure (recall eqs. (45) and (53)) are replaced by averages over the ensemble of copies for large $`M`$. We have taken $`M=10^5`$.
As an illustration we show some typical flow diagrams for the overlap order parameter in Figure 6 for some fully connected ($`c=1`$, top figures) and diluted ($`c=0.01`$, bottom figures) models. The model parameters are $`\alpha =0.01`$ and $`b=0.2,T=0.3`$ (top left), $`b=0.5,T=0.1`$ (top right), $`b=0.2,T=0.5`$ (bottom left) and $`b=0.5,T=0.2`$ (bottom right). Self-couplings are set to zero.
Several remarks are in order. All situations shown are in the retrieval phase of the corresponding phase diagrams Figures 1 and 2. They all converge to the corresponding RS equilibrium results provided the initial overlap order parameter is large enough. This gives us also an idea of the basins of attraction. These basins are bigger for $`b=0.2`$ since the relevant model parameters are chosen deeper in the retrieval region, i.e., closer to the line of thermodynamic stability.
For completeness we end this Section on the discussion of the dynamics employing the GFA approach with a couple of remarks. First, this technique can also be applied to asymmetric dilution ($`c_{ij}c_{ji}`$) and we then get for the correlation matrix and retarded self-interactions
$`𝐑`$ $`=`$ $`c(𝑰𝐆)^1+\mathrm{\Gamma }(1c)𝐆+(J_0c)𝑰,`$ (57)
$`𝐃`$ $`=`$ $`c(𝑰𝐆)^1𝐂(𝑰𝐆^{})^1+(1c)𝐂.`$ (58)
with the asymmetry factor $`\mathrm{\Gamma }`$ given by
$$\mathrm{\Gamma }=\frac{c_{ij}c_{ji}c^2}{c(1c)}.$$
(59)
Secondly, one can compare the GFA technique with the signal-to-noise analysis for the three-state Ising network . This comparison is formally completely analogous to the one for the Hopfield model discussed in detail in . The only point at which one has to be careful is in deriving the recursion relations for the coloured noise $`𝜼`$. The diagonal of $`𝐂`$, namely, is no longer equal to 1 but depends on time. Taking this into account one can take the results for the Hopfield model and extend it to $`Q>2`$-Ising models. The outcome is that, even for general $`Q`$, the signal-to-noise analysis is a short memory approximation to the real effective dynamics giving results correct up to the third time step. Like for the other models, this approximation is extremely good in the retrieval region but fails in the spin glass region. It is then a rather straightforward exercise to extend the revisited signal-to-noise analysis, proposed in to the $`Q`$-Ising case. Since the final results are completely equivalent to those of the GFA eqs. (39)-(41), we refrain from repeating further explicit details and refer to the work mentioned above ().
## 6 Concluding remarks
We have studied the statics and dynamics of the three-state Ising neural network model with synchronous updating of the neurons and variable dilution. We have followed as closely as possible the discussion for the Hopfield model adapting the methods to the multi-state nature of the neurons. Although the equations that describe both statics and dynamics are formally rather similar to those of the Hopfield model, we have seen that the physics behind it is quite different.
We have examined the thermodynamic and retrieval properties for this model using replica symmetric mean-field theory and have made a detailed comparison of the results with those for sequential updating. Capacity-temperature phase diagrams are derived for several values of the pattern activity and different gradations of dilution. Apart from the appearance of cycles the asymptotic behaviour is almost identical to the one for sequential updating. The spin-glass region is visibly enhanced, while the retrieval region, however, is only marginally enhanced. Only the addition of self-couplings can enlarge the retrieval region substantially, especially in the case of strong dilution. A calculation of the information content shows that both for synchronous and sequential updating the three-state networks are robust against the interference of static noise coming from random dilution.
Concerning the dynamics, we have extended the generating functional analysis to the study of $`Q`$-state spins. As an illustration some typical flow diagrams for the overlap order parameter are presented in the case $`Q=3`$. It is possible to extract the result for sequential updating from the one for synchronous updating. A comparison with the signal-to-noise approach is made. As for the Hopfield model one can argue that the signal-to-noise analysis used before in the literature is a short memory approximation correct up to the third time step. It can also be shown that this signal-to-noise analysis can be made exact.
## Acknowledgments
The authors would like to thank Jordi Busquets Blanco, Gyoung Moo Shim and Walter K. Theumann for useful discussions. R. E. thanks the kind hospitality and the support of the Instituut voor Theoretische Fysica, KULeuven. This work has been supported in part by the Fund of Scientific Research, Flanders-Belgium.
## Appendix
We write down explicitly the saddle-point equations discussed in Section 3. Therefore we need the averages $`\sigma _z`$, $`\sigma ^2_z`$ and $`\sigma \tau _z`$ with respect to the effective Hamiltonian (23)
$$\sigma _z=\frac{2}{𝒵}\left\{\mathrm{sinh}(2\beta M)\mathrm{e}^{\beta \left[2\left(b+\frac{\alpha }{2}\widehat{\chi }\right)+\alpha \widehat{\chi }_r\right]}+\mathrm{sinh}(\beta M)\mathrm{e}^{\beta \left[b+\frac{\alpha }{2}\widehat{\chi }\right]}\right\},$$
(60)
$$\sigma \tau _z=\frac{2}{𝒵}\left\{\mathrm{cosh}(2\beta M)\mathrm{e}^{\beta \left[2\left(b+\frac{\alpha }{2}\widehat{\chi }\right)+\alpha \widehat{\chi }_r\right]}\mathrm{e}^{\beta \left[2\left(b+\frac{\alpha }{2}\widehat{\chi }\right)\alpha \widehat{\chi }_r\right]}\right\},$$
(61)
$`\sigma ^2_z`$ $`=`$ $`{\displaystyle \frac{2}{𝒵}}\{\mathrm{cosh}(2\beta M)\mathrm{e}^{\beta \left[2\left(b+\frac{\alpha }{2}\widehat{\chi }\right)+\alpha \widehat{\chi }_r\right]}`$ (62)
$`+`$ $`\mathrm{cosh}(\beta M)\mathrm{e}^{\beta \left[b+\frac{\alpha }{2}\widehat{\chi }\right]}+\mathrm{e}^{\beta \left[2\left(b+\frac{\alpha }{2}\widehat{\chi }\right)\alpha \widehat{\chi }_r\right]}\},`$
with
$$M=\xi m+\sqrt{\alpha q_1}z.$$
(63)
Using the symmetry properties of these functions, we average over $`\xi `$ to obtain the saddle-point equations for the retrieval state of one pattern at finite temperature. We get
$$m=𝒟z\sigma _z(m+\sqrt{\alpha \widehat{q}_1}z),$$
(64)
$$r=a𝒟z\sigma \tau _z(m+\sqrt{\alpha \widehat{q}_1}z)+(1a)𝒟z\sigma \tau _z(\sqrt{\alpha \widehat{q}_1}z),$$
(65)
$$q_0=a𝒟z\sigma ^2_z(m+\sqrt{\alpha \widehat{q}_1}z)+(1a)𝒟z\sigma ^2_z(\sqrt{\alpha \widehat{q}_1}z),$$
(66)
$$q_1=a𝒟z\left[\sigma _z\right]^2(m+\sqrt{\alpha \widehat{q}_1}z)+(1a)𝒟z\left[\sigma _z\right]^2(\sqrt{\alpha \widehat{q}_1}z),$$
(67)
that need to be solved together with the algebraic equations (28)-(30). |
warning/0507/astro-ph0507439.html | ar5iv | text | # Heavy Element Production in Inhomogeneous Big Bang Nucleosynthesis
## I INTRODUCTION
What happened in the early universe has a great influence on the history of the universe because they determined the initial conditions. It is very important to check whether our standard model of cosmology is correct or not as theories and observations develop. Baryogenesis and BBN should be checked because they determine the history of the chemical evolution.
In the standard model of elementary particle physics, the baryogenesis is possible only through the electro-weak spharelon process. In the supersymmetric standard model, it is much easier to explain the baryon number asymmetry because there are many scalar fields which have baryon number. One of the most striking property of supersymmetric theories is that they have flat directions in potentials. Some of them have baryon number and if fields condensate in these directions, it is possible to produce large baryon number. This is the basic idea of the Affleck-Dine baryogenesis affl-dine . Usually baryon number production is assumed to be taken place homogeneously all over the space. This is natural because we know that the universe is homogeneous and if baryogenesis is inhomogeneous in large scale it contradicts observations cmb . Of course resolution ability of observations is limited and small scale inhomogeneity is not excluded by observations. Though it seems to be unnatural to consider such small scale inhomogeneity, recent observations force us to reconsider the possibility of inhomogeneous baryogenesis.
It has become clear that the evolution of matter started earlier than we have known before. For example, Wilkinson Microwave Anisotropy Probe (WMAP) data suggests that reionization began when $`z`$20 WMAP . According to Refs. Barth:2003 ; Dietrich:2002 , star formation activity started when $`z`$ 10. In addition, it is known that the quasar metallicity did not significantly change from the time of high redshift to the present time Boksenberg:2003 . Recently a galaxy at $`z`$=10.0 was observed Pello . Other evidences of heavy elements from the high redshifts are given in Pichon:2003 ; Cohen ; Songaila ; Prochaska ; Freuding ; Pettini
Motivated by these observational evidences, we investigate the possibility that inhomogeneous baryogenesis produced very high baryon density in small fraction of the universe and in these regions some fraction of heavy elements were already synthesized during BBN.
Heavy elements production during BBN itself is not a new idea. Previous researches on the inhomogeneous big bang nucleosynthesis are given in IBBN1 ; IBBN2 . Heavy elements production is also mentioned in jedam . These works, however, do not include very heavy elements IBBN1 or they create neutron rich regions and calculate the nucleosynthesis in those regions IBBN2 . These are not suitable for our present purpose.
For the production of heavy elements during BBN, high baryon density is necessary. However if we simply increase the baryon-photon ratio all over the universe homogeneously, it would apparently contradict the observed light element abundances light1 and CMBR cmb . Instead, we assume that the baryon density of the universe is inhomogeneous before and during BBN. In most part of the universe $`\eta `$ is small ($`\eta 6\times 10^{10}`$) as observed while small fraction of the universe is occupied with very high baryon density, $`\eta 𝒪(1)`$. Because our aim is to see how BBN goes in the high baryon density regions and not make the precise adjustment between BBN and CMBR, we neglect the baryon diffusion. In this case, the baryon density in high density regions can be treated almost free parameter without contradicting observations. (It is a complicated problem whether we can treat $`\eta `$ as a free parameter in realistic models. See, for example, neutrino .)
In section II we explain the theoretical aspects of our model Dolgov:1993si . In section III, we explain our network and what kind of effects we take into account. Section IV is the main results of our numerical study. In this section we explain BBN is the p-process like and simultaneously the r-process like. And also BBN can produce very heavy elements including proton rich nuclei such as $`{}_{}{}^{92}\mathrm{Mo},^{94}\mathrm{Mo},^{96}\mathrm{Ru},\mathrm{and}^{98}\mathrm{Ru}`$.
## II THEORETICAL BACKGROUND
Theoretical background of this model is inhomogeneous baryogenesis Dolgov:1993si . We are going to explain basic aspects of this model. For more detail, see Dolgov:1993si ; matsu .
The basic idea of the model Dolgov:1993si is a modified version of the Affleck-Dine baryogenesis affl-dine . Assume that the interaction Lagrangian has the general renormalizable form
$$\begin{array}{cc}\hfill _{int}& =\lambda |\varphi |^2\mathrm{\Phi }^2+g|\varphi |^2\mathrm{\Phi }\hfill \\ & =\lambda (\mathrm{\Phi }\mathrm{\Phi }_1)^2|\varphi |^2\lambda \mathrm{\Phi }_1^2|\varphi |^2,\hfill \end{array}$$
(1)
where $`\varphi `$ is Affleck-Dine (AD) field, $`\mathrm{\Phi }`$ is the inflaton field, $`g`$ and $`\lambda `$ are the coupling constants and $`\mathrm{\Phi }_1=g/2\lambda `$.
In a simplest case, the effective mass of the AD field $`\varphi `$ can be written by
$$(m_{eff}^\varphi )^2=m_0^2+\lambda (\mathrm{\Phi }\mathrm{\Phi }_1)^2,$$
(2)
where $`m_0^2`$ is the vacuum mass of $`\varphi `$.
The vacuum expectation value of $`\mathrm{\Phi }`$ is assumed to evolve from very large value, i.e. , $`\mathrm{\Phi }\mathrm{\Phi }_1`$, decreases to zero. As $`\mathrm{\Phi }`$ goes down to $`\mathrm{\Phi }_1`$, the effective mass square of $`\varphi `$ becomes negative and the phase transition takes place. When $`\mathrm{\Phi }`$ is far from $`\mathrm{\Phi }_1`$ the mass square is positive. If the duration of $`\mathrm{\Phi }\mathrm{\Phi }_1`$ is short, the transition would take place only in a small fraction of space. Consequently in the dominant part of the universe baryon asymmetry is small as observed $`\eta =𝒪(10^9)`$, while in a small part of the universe the baryon asymmetry can be very large, even close to unity. In this simple model, the signature of barionic charge is not fixed. Baryonic chaege can become both positive and negative B-charge . However, the high density regions are very small compared to cosmological scale, high density anti-matter regions would disappear by pair-annihilation while late time inflation can prevent the annihilation domain .
Because we are not very interested in the detail of the shape of the bubbles and the effect of diffusion in this paper, we assume that the bubble sizes are large enough to neglect the diffusion effects. Also, bubbles are not large so as to contradict the observations cmb .
In this case, the BBN calculation can be treated as that of homogeneous big bang nucleosynthesis. In the following section, we present the results of the calculations and their physical interpretations.
At first sight, in the high baryon density regions the reaction seems to proceed along the proton rich side because BBN occurs in proton rich environment matsu . However, surprisingly it is not correct. BBN proceeds along the proton rich and the neutron rich side.
## III NUMERICAL CALCULATIONS
The basic method of our calculation is the same as that of the homogeneous big bang nucleosynthesis.
We solve the Friedmann equation
$$\left(\frac{\dot{a}}{a}\right)^2=\frac{8\pi G\rho }{3}$$
(3)
where $`\rho =\rho _\gamma +(\rho _e^{}+\rho _{e^+})+\rho _\nu +\rho _b`$, and $`a`$ is the scale factor. The energy conservation law is
$$\frac{d}{dt}(\rho a^3)+\frac{p}{c^2}\frac{d}{dt}(a^3)=0$$
(4)
for the time evolution of the temperature and the baryon density.
Abundance change in the region is evaluated with a nuclear reaction network, which includes 4463 nuclei from neutron, proton to Americium (Z = 95, A = 292). Nuclear data, such as reaction rates, nuclear masses, and partition functions, are same as in fujimoto . It should be emphasized that both proton-rich and neutron-rich nuclei are produced in a high density region (Fig. 1). Therefore it is required using a large network to calculate abundances in the high $`\eta `$ region.
## IV RESULTS
We have calculated BBN for various values of $`\eta `$, from $`10^{10}`$ to $`10^2`$. It is known that in the standard, low baryon density BBN, nuclei heavier than Boron are hardly synthesized. Fig. 1 represents the synthesized nuclei for $`\eta =1\times 10^6`$ at the epoch $`T=1\times 10^7`$K. We can see that heavier nuclei such as Ca ($`10^{14}`$ in mass fraction) are synthesized.
Naturally as $`\eta `$ becomes large, heavier nuclei are synthesized. However we find there is drastic change in the nucleosynthesis around $`\eta 3\times 10^4`$. To see this transition, we pick up two values of $`\eta =10^4`$ and $`10^3`$, and investigate what is going on during the nucleosynthesis.
Fig. 2 represents how many nuclei are synthesized at the temperature $`T=3\times 10^9`$K, and $`\eta =10^4`$.
Red (blue) color represents more (less) synthesized nuclei. (Stable nuclei also plotted in Fig. 23, and 5, with gray color.) We can see that the reaction goes along stable line. It is well known that nuclei whose neutron and proton numbers are special values (for example 20, 28, 50, 82 etc.), the magic numbers, are especially stable. At these points, the reactions are stagnated and the reaction paths are bent. Especially, the stagnation at $`N`$ = 82 is one of the biggest factors that prevent the reaction to proceed further beyond the mass number 190. As the temperature goes down, the locus of the reaction begins to bent to different directions (Fig. 3).
For lighter nuclei (mass number $`A100`$), proton captures are very active and the locus moves to proton rich direction. For nuclei whose mass numbers are between 100 and 120, the locus is across the stable nuclei from proton rich side to neutron rich side. For heavier nuclei for $`A120`$, neutron capture is more efficient. This suggests that both the r-process and p-process occur simultaneously in BBN.
Physical interpretation of this situation is as follows. The environment in BBN is proton rich, i.e., the electron fraction $`Y`$e ranges from 0.8 to 0.9 matsu . Naive expectation of BBN is the p-process. For relatively light heavy nuclei, proton capture is active. However proton capture processes become exponentially difficult as the proton number increases because of their coulomb barrier. On the other hand, neutrons are still not consumed out during heavy nuclei are synthesized as shown in Fig. 4. Very heavy nuclei captures neutrons and the locus of the reaction changes toward the neutron rich side. Transition point from proton rich side to neutron side depends on the baryon-photon ratio $`\eta `$. The transition occurs at larger mass number for larger $`\eta `$. The reactions depend on the abundances of the seed nuclei. The higher baryon density follows many seeds which lead to proton captures on heavier nuclei. Fig. 4 shows the time evolution of mass fraction for nuclei whose mass number is 90 and 158. Let us see the time evolution of the mass number 90. <sup>90</sup>Zr is a stable nucleus and <sup>90</sup>Mo is a proton rich one. First, <sup>90</sup>Zr is synthesized and later <sup>90</sup>Mo is synthesized while the amount of <sup>90</sup>Zr decreases. This represents that the reaction first proceeds along the stable line and later move to the proton rich side. At the late stage, <sup>90</sup>Zr increases while <sup>90</sup>Mo decreases because unstable proton rich nuclei decay to stable nuclei such as <sup>90</sup>Mo $``$ <sup>90</sup>Nb $``$ <sup>90</sup>Zr. The lifetimes of <sup>90</sup>Nb and <sup>90</sup>Mo are $`14.6`$ h = $`5.33\times 10^4`$ sec and $`5.67`$ h = $`2.04\times 10^4`$ sec, respectively.
For heavier nuclei of the mass number 158, the situation is different. <sup>158</sup>Gd is a stable nucleus and <sup>158</sup>Eu is a neutron rich nucleus, instead of proton rich one. First the stable nucleus <sup>158</sup>Gd is synthesized and later neutron rich <sup>158</sup>Eu is synthesized. This shows that in heavier nuclei region, the stable nuclei are produced first as $`{}_{}{}^{90}Zr`$, but later the neutron rich nuclei are produced instead of proton rich nuclei. The decrease in <sup>158</sup>Eu in the late stage is the same as <sup>90</sup>Mo, $`\beta `$ decay to stable nuclei. The abundances of neutron rich nuclei of the mass number 90 and those of proton rich nuclei of the mass number 158 are very small and not drawn in this figure. We can also see that neutrons are still left when heavy nuclei are synthesized.
Now let us see the case $`\eta `$ =$`10^3`$. Fig. 5 shows the locus of the reaction at the temperature $`T=1.8\times 10^9`$ K. It is apparently different from the results of $`\eta `$= $`10^4`$. For, in this case, the reactions first proceeds along the stable line. However, the reactions directly proceeds to the proton rich region. Another important difference is that very heavy nuclei of $`A80`$ are not synthesized. The physical interpretation is as follows. In a high baryon density region, the seeds for the reactions to proceed are abundant. The nuclear reaction proceeds promptly and all neutrons are consumed by light nuclei as shown in Fig. 6. This prevents the nucleosynthesis from proceeding to the large mass number region. In Fig. 6 we only draw the abundance having $`A=90`$. Heavier nuclei are not synthesized enough. When heavy nuclei are synthesized, neutrons are almost consumed out.
In Fig. 7, we show the relation between the mass number and the number fraction relative to the solar abundances. As the baryon density becomes larger, the heavier nuclei are synthesized for $`\eta `$ less than $`1\times 10^4`$. However, when $`\eta 1\times 10^4`$, the maximum mass number decreases as $`\eta `$ becomes larger.
The number fraction ratios of p-nuclei to the solar abundances is listed in Table. 1.
For $`\eta =10^3`$, <sup>92</sup>Mo, <sup>94</sup>Mo, <sup>96</sup>Ru and <sup>98</sup>Ru drastically increase, due to the change of the loci of the reaction flows. This suggests that highly inhomogeneous BBN would have a large influence on the abundances of solar p-nuclei.
The observed abundances seem not to be explained by BBN with only a single $`\eta `$. However this does not exclude the possibility of our assumption. It is unnecessary for BBN abundances to match exactly to the solar abundances because produced nuclei in high $`\eta `$ regions would have mixed with nuclei synthesized in low density regions and also there should be nuclei synthesized in star activities.
Basic feature of BBN at each $`\eta `$ are classified as follows. For $`\eta 10^6`$, synthesized nuclei are limitted to $`A40`$. Nuclei whose $`A`$ are around 20 are less synthesized even at large value of $`\eta `$. For $`\eta `$ from $`10^5`$ to $`10^4`$, the abundances of nuclei whose mass number of $`30A56`$ grow rapidly with $`A`$. Abundances of nuclei $`A56`$ suddenly decrease but again slowly increase. At around $`A=`$ 140, they turn to decrease.
After the rapid decrease in the abundances of $`A56`$ for $`\eta =10^3`$ and $`10^2`$, the abundance profiles are rather different.
For $`\eta =10^3`$, the abundances do not drastically change from $`A=`$ 64 to around 86. They rapidly decrease for $`A`$ above 100 and maximum $`A`$ synthesized is 114. For $`\eta =10^2`$, right side wing of Fe peak is similar to the solar abundances. There is a peak around $`A=`$ 72 and the abundance production decreases rapidly above beyond the peak until $`A98`$. The maximum $`A`$ synthesized is 98.
To compare our results with observations such as metal poor star abundances, we need to take into account dynamical mixing after the epoch of BBN. This depends on a model significantly and will be a future work.
We should examine the idea presented in this paper with more realistic model, and determine whether heavy elements were really synthesized in BBN or not. The former is to take into account the diffusion effects before and during BBN and also lepton asymmetry. The latter is to calculate the nucleosynthesis in supermassive stars. This is because supermassive stars are generally thought to have synthesized first heavy elements in the universe. We need to know whether heavy elements observed in high redshift were synthesized in BBN or supermassive stars. Nucleosynthesis in supermassive stars and BBN in high baryon density region is similar. It would be a problem how to distinguish these two nucleosynthesis from observations.
## V Conclusion
We have investigated BBN in high baryon density region. In these regions, not only light elements which are synthesized in standard BBN but also very heavy elements are produced. We found BBN is both the p-process like and the r-process like. The transition from the p-process to the r-process is due to the Coulomb barriers of proton-rich nuclei and the amounts of neutrons when heavy elements begin to be synthesized. The loci of the reaction flows change drastically above $`\eta =10^3`$. Above $`\eta =10^3`$, a lot of seed nuclei cause active proton capture and the reaction flows end before very heavy elements are synthesized.
Our calculations demonstrate that very heavy elements can be synthesized in BBN, including proton-rich nuclei. These nuclei will be related to the origin of the solar abundances, heavy elements observed in high redshifts and early star formations via cooling effects.
For more realistic models in BBN, we need to include diffusion effects. Comparison with the nucleosynthesis in supermassive stars is also important. We leave these issues for future study.
## VI Acknowledgements
S.M. thanks Kazuhiro Yahata, Yuuiti Sendouda, Shigehiro Nagataki, Naoki Yoshida, Mamoru Shimizu, Kohji Yoshikawa, Atsunori Yonehara, Shinya Wanajo, Tomoya Takiwaki, Naoyuki Itagaki, Koji Higashiyama, Satoshi Honda for useful discussions. This research was supported in part by Grants-in-Aid for Scientific Research provided by the Ministry of Education, Science and Culture of Japan through Research Grant No.S 14102004, No.14079202. S.M.’s work was supported in part by JSPS(Japan Society for the Promotion of Science). |
warning/0507/astro-ph0507213.html | ar5iv | text | # “PAH” emission bands in selected planetary nebulae: a study of the behaviour with gas phase C/O ratio11footnote 1based on observations with ISO, an ESA project with instruments funded by ESA member states (especially the PI countries: France, Germany, the Netherlands and the United Kingdom) and with the participation of ISAS and NASA
## 1 Introduction
Emission features occur near 3.3, 6.2, 7.7, 8.7, 11.3 $`\mu `$m in infrared (IR) nebular spectra (Russell, Soifer & Merrill 1977; Russell, Soifer & Willner 1977; Sellgren, Werner & Dinerstein 1983; Aitken & Roche 1982; Roche & Aitken 1986). Airborne spectroscopy (Cohen et al. 1986,1989) shows these form a generic spectrum, with the most intense peaks unobservable from the ground (6.2,7.7 $`\mu `$m); all band intensities are correlated; and the 7.7-$`\mu `$m band tightly correlates with gas phase C/O ratio in planetaries (i.e., for newly formed dust, in a circumstellar environment). The features are attributed to vibrationally excited polycyclic aromatic hydrocarbons (PAHs) and related materials (Duley & Williams 1981; Leger & Puget 1984; Allamandola, Tielens & Barker 1985; Peeters et al. 2002). The PAHs are regarded as permeating almost every phase of the interstellar medium (ISM)(Allamandola, Hudgins, & Sandford 1999). Various components are recognized (Allamandola, Tielens & Barker 1989; Tielens 1993): the narrow features are carried by molecular-sized PAHs (50 C-atoms); larger PAH clusters (500 C-atoms) yield the plateaus underlying the narrow features; while the 25 and 60-$`\mu `$m “cirrus” emission originates from amorphous carbon grains, perhaps built from PAHs, of size 5000 to 50,000 C-atoms. This hypothesis is supported by the obvious link between carbon abundance and the bands (Barlow 1983; Cohen et al. 1986,1989; Casassus et al. 2001), and the tight correlation between the 6.2 and 7.7-$`\mu `$m bands (both from C=C skeletal modes). The 11.3-$`\mu `$m feature is due to out-of-plane bending vibrations of peripheral H atoms attached to aromatic molecular units; its precise frequency depends on the number of adjacent H atoms on each edge ring (Bellamy 1958). The observed 11.3 and 12.7-$`\mu `$m bands and the 10.5-14 $`\mu `$m plateau of emission (Cohen, Tielens & Allamandola 1985) are characteristic of aromatic rings with nonadjacent, or 2 or 3 adjacent, peripheral H atoms (see also van Diedenhoven et al. 2004). With the advent of ISO (Kessler et al. 1996) spectra, additional features attributed to PAHs have been identified, such as the 16.4-$`\mu `$m peak and plateau, attributed by Moutou et al. (2000) to C-triple bonds.
Using exclusively the $`IRAS`$ Low Resolution Spectrometer (LRS) data base, Volk & Cohen (1990) sought a correlation of the 11.3-$`\mu `$m band with C/O, finding an apparently constant value of 11.3-$`\mu `$m strength for C/O$`>`$1, with an abrupt transition near C/O = 1, the boundary between O- and C-rich nebulae. Any relationship between the 7.7-$`\mu `$m or 11.3-$`\mu `$m bands and gas phase C/O may also hold clues to the mechanisms of dust formation in planetaries, and might even elucidate the role of peripheral hydrogen atoms.
Understanding carbon dust formation has wide significance in astronomy, because AGB stars, the precursors of C-rich PNe, are the dominant formation sites known for refractory grains
subsequently injected into the ISM. Models for soot formation in flames have been proposed, based on neutral radicals, ions, PAHs, polyacetylenic chains, or fullerenes as intermediaries (Hucknall 1985; Barnard & Bradley 1985; Curl & Smalley 1988; Ugarte 1992,1995; Hecht 1986; Iglesias-Groth 2004). Each route could probably lead to C-soot, depending on specific physical conditions. Our goal is to explore these routes, using C-rich planetaries as laboratories to seek the precursors to C-dust grains. Strong nebular UV fields excite the PAHs and the resulting emission serves as a direct probe of the molecular gas content.
## 2 The sample of Planetary Nebulae
There are two ways to investigate the dependence of PAHs on gas-phase C/O in PNe: using a single well-studied band from the ground; or measuring the strongest bands from airborne and space-borne observatories. An example of the former approach is the survey of the 3.3-$`\mu `$m band by Roche et al. (1996), who used common instrumentation for their entire sample of PNe. Previous space-based and airborne work on the relationships between the PAH bands and nebular C/O has depended upon the use of measurements made with a variety of instruments, with different apertures, and at different spectral resolutions. This approach encounters the problem of poor sensitivity to either the 11.3-$`\mu `$m band from the Kuiper Airborne Observatory (KAO) or to the 3.3-$`\mu `$m band from the ISO SWS. It is clearly important to cover a wide range of IR wavelengths with common instrumentation for this work, so that any trends found are not compromised. ISO afforded a unique opportunity to secure the requisite measurements.
However, such observations were clearly predicated primarily on the IR brightness of the nebulae, without specific regard to their C/O ratios. Therefore, the PNe we have selected have well-determined C/O values, with broad agreement among several authors and different analytical methodologies (e.g. Kingsburgh & Barlow 1994; Rola & Stasinska 1994; Zuckerman & Aller 1986). Further, we have isolated precisely those nebulae whose C/O values bear most directly on the investigation of the trends with PAH band strength found in the old airborne data.
Table 1 summarizes the set of PNe assembled for this study; objects are listed alphabetically. For each PN spectrum we tabulate the name, TDT (ISO’s observation identification number), the astronomical observing template (AOT), date, integration time, whether an LWS01 spectrum was obtained in an “off” position, the angular diameter in arcsec, and a reference for the adopted diameter. We initially selected all PNe for which AOTs SWS01 and LWS01 (low-resolution spectra) were obtained by ISO. The prerequisite for PN selection was the existence of a published value of C/O (eliminating such objects as M 2-43 with its bright PAH emission but no C/O abundance ratio). The SWS aperture varies with wavelength. From 2-12 $`\mu `$m, the size is between 14<sup>′′</sup> and 20<sup>′′</sup> across the several bands; from 12-29 $`\mu `$m, between 14<sup>′′</sup> and 27<sup>′′</sup>; and from 29-45 $`\mu `$m, between 20<sup>′′</sup> and 33<sup>′′</sup>. The sample includes three large PNe that extend substantially outside the SWS apertures. One of these is even larger than the LWS apertures (NGC 5189). For several nebulae of interest (i.e. with sizeable C/O) no LWS01 spectra were taken by ISO because of far-infrared (FIR) faintness (e.g. M 4-18) yet the PAHs were well-detected by the SWS. Therefore, we extended the sample to PNe for which SWS01 and $`IRAS`$ 60/100-$`\mu `$m photometry were available, but which lacked any usable LWS01 spectra (i.e. absent, or too noisy).
The restriction to PNe for which SWS01 spectra are available through the ISO Data Archive leads to a bimodal sample of nebulae. Both PNe with bright emission lines and very weak IR continua, and PNe of types known to exhibit PAH emission bands with strong IR continua (e.g. the \[WCL\] sequence whose central stars show Wolf-Rayet-like emission spectra) were almost equally frequently targeted by the observing community. The integration times used for the sample varied by a factor of six, depending on the goals of the original observers. Consequently, we can offer no statement as to the completeness of our sample based upon any physical characteristic.
For two PNe multiple data sets exist because those objects had been designated as LWS wavelength calibrators (NGC 7027 (26 spectra) and NGC 6543 (94)). For these objects, a single representative LWS01 spectrum was chosen. NGC 6543 was also observed on five occasions in the SWS01 AOT but we coadded all five spectra using inverse-variance weighting to enhance the signal-to-noise ratio.
## 3 The available spectra
The two relevant ISO spectroscopic instruments are the SWS (de Graauw et al. 1996) and LWS (Clegg et al. 1996). In each case we sought low-resolution data from the AOTs “SWS01” (covering the range from 2.38 to 45.2 $`\mu `$m), and “LWS01” (43-197 $`\mu `$m).
The SWS01 data include all the known mid-infrared PAH emission bands and emission plateaus (3.2-3.6, 6-9, 10.5-15, and 16-21 $`\mu `$m) between 3 and 21 $`\mu `$m. The SWS apertures were generally well-matched to the optical and radio diameters of the selected PNe. Typical achieved resolving powers were several hundred for AOTs taken at the fastest speed (i.e. shortest observing time $``$1100s), although the data archive includes higher resolution spectra for some of our chosen PNe that were observed at the slowest speeds (e.g. resolving power $``$1500 for $``$6500s observing time).
LWS01 AOTs serve to assess the luminosity of the commonly found thermal emission from cool dust grains in PNe. The LWS01 spectra have a resolution of 0.3 $`\mu `$m from 43–93 $`\mu `$m, and 0.6 $`\mu `$m from 84–197 $`\mu `$m. Observing times with the LWS ranged from 640 to 3400s, through an aperture with an effective diameter of 66–86<sup>′′</sup>, depending on wavelength.
The Spitzer Space Telescope (SST) and its infrared spectrometer (IRS) are able to observe PNe. However, the great sensitivity of this observatory and its instruments implies that many of the known PNe with PAH emission are too bright for spectroscopy with the SST. Further, the small slits of the IRS cannot accommodate most of our target PNe. Consequently, the ISO Data Archive offers an opportunity to revisit many of the PNe known to show PAH emission, with apertures that generally are sufficiently large to capture essentially all the PAH emission across these objects (Smith, Aitken & Roche 1989) or from their surrounding photodissociation regions (PDRs) (Aitken & Roche 1983).
We visually inspected all SWS01 spectra for PNe that survived our selection criteria, seeking detections of the PAH features. Nebulae clearly showing the bands were investigated first. Subsequently we examined the remaining PNe with the goal of setting quantitative upper limits on their PAH emission.
## 4 Band integrals and the join of SWS and LWS spectra
The PAH bands are low-resolution features. Therefore, to enhance the detectability of the bands with SWS spectra, we interactively used boxcar-smoothing (with widths of 50 or 75 points) better to define the overall continuum and PAH features in noisy data. Figures 1 and 2 illustrate the results of this smoothing applied to eight PNe. The 7.7-$`\mu `$m and 11.3-$`\mu `$m PAH bands were integrated after interactively defining a single underlying continuum in each PN across the $`515\mu `$m range. The wavelengths selected to define these continua (by cubic splines) and the regions chosen for the band integrals were those described by Cohen et al. (1986,1989) and Volk & Cohen (1990), for which choices the original relationships between integrated 7.7 or 11.3-$`\mu `$m intensity and C/O were found. Care was taken to remove any influence of smoothed emission lines on these band integrals. For example, the 7.46-$`\mu `$m Pf$`\alpha `$ line was cut out of those nebulae in which it was detected, by replacing it by a local smooth continuum at the base of the line in the original, unsmoothed, SWS spectra. Likewise, in high-excitation PNe, we expunged the \[Nevii\] line at 7.65 $`\mu `$m prior to smoothing and integrating the emission bands. In the absence of any visual recognition of the PAH features, formal upper limits were set by one of two methods. We measured the nearest positive emission hump lying within the wavelength interval for definition of a PAH integration. In addition we computed the formal integral over the same wavelength range of the mean spectrum plus 3$`\times `$ the standard deviation calculated over the interval. Differences arose between these two methods primarily when a spectrum had negative values or when the splined continuum locally exceeded the measured spectrum. We adopted whichever approach yielded the more conservative upper limit.
To evaluate the total, integrated, observed IR energy, I(IR), we assembled complete SWS+LWS energy distributions by splicing together the two data sets for each PN in the overlap region from 40-45 $`\mu `$m. From past experience of this procedure (e.g. Cohen et al. 2002) we have determined that LWS01 spectra are better absolutely calibrated in this region than SWS. Moreover, if a PN extends slightly outside the SWS aperture, the larger LWS aperture will more reliably assess the total nebular continuum. Therefore, we visually examined the overlap and rescaled each SWS spectrum upwards to achieve the best match to the LWS when necessary. This approach assumes that the PN emission lying outside the SWS Band-4 aperture is spectrally identical to that sampled within the aperture.
Given the complete spliced energy distribution, each spectrum was examined overall to eliminate any remaining specious features at the long end of the SWS band-4 range and/or the beginning of the LWS detector SW1 range. The resulting cleaned and spliced spectra were then merged. When a PN had multiple spectra these were combined using inverse-variance weighting, before splicing the resulting SWS and LWS data.
For the ten PNe lacking adequate LWS01 spectroscopy, but having $`IRAS`$ FIR photometry, we used 60 and 100-$`\mu `$m flux densities as a substitute for absent, or an alternative to noisy, LWS01 data to estimate the FIR contribution to I(IR). We summed the products of flux density and bandwidth at 60/100 $`\mu `$m, using the synthetic contiguous bandwidths of Emerson (1988) to avoid the uncertainties involved in color correction for PNe due to emission lines. Upper limits at either 60 or 100 $`\mu `$m were treated as actual flux densities to assess the calculated FIR component of the IR luminosity. (These ten PNe are identified in Figures 3 and 4 by having a large plus sign through their plot symbols, and by asterisks in Table 2.) Tests were made on several PNe for which we had LWS01 data, comparing I(IR) derived by the two methods. These indicated that using $`IRAS`$ data in this fashion produced values of I(IR) within 25 percent of the actual integrated SWS+LWS spectra.
## 5 Results
Table 2 presents the results, listing PN name, gas-phase C/O abundance ratio, I(IR) (in W cm<sup>-2</sup>), I(7.7)/I(IR), I(11.3)/I(IR), and a reference for the C/O value adopted. Whenever possible, we have selected abundance ratios derived from collisionally excited forbidden lines (CELs) rather than from optical recombination lines (ORLs). PNe often yield substantially different abundance ratios when determined by the two methods (e.g. Liu et al. 2000,2001a; Tsamis et al. 2004). Liu et al. (2000) have argued for the existence of H-deficient ionized clumps in PNe, cool enough to suppress forbidden CELs but produce strong ORL emission. Most of the PNe in Table 1 have CEL C/O ratios. When these are unavailable, we have used the ORL line ratio (e.g., for Vy 2-2). We have never mixed CEL and ORL abundance determinations to create a C/O value.
In the first studies of the multi-band PAH spectra of PNe from the KAO it was recognised that the ratio of I(7.7)/I(IR) ranged over two orders of magnitude. To avoid confining the distribution of the bulk of the PNe detected to the lower left corner of any figure against C/O by use of a linear plot, the logarithmic ordinate was selected. We have followed this precedent. CPD-568032 and He 2-113 are well-measured objects with extreme values of C/O, yet they both emit comparable fractions of the IR flux in the PAH bands to those emitted by other PNe with much more modest C/O ratios. Consequently, we exclude them from our sample in terms of any regression analysis.
Fig. 3 illustrates the 7.7-$`\mu `$m results, distinguishing between PNe detected in the 7.7-$`\mu `$m band, the subset of nebulae with \[WCL\] central stars, and the 26 objects with only upper limits. The solid line plotted in this figure indicates the formal regression line for all detected PNe with the exceptions of CPD-568032 and He 2-113. The intercept is $`2.16\pm 0.06`$, the slope is 1.47$`\pm `$0.33, and the Pearson correlation coefficient for these 15 PNe is r=0.76. We identify this dashed regression line as the ISO version of the relationship reported by Cohen et al. (1989: their Figure 20) based on airborne spectroscopy. (Note that, in their figure, the point for CPD-568032 was misplotted too high by a factor of 10 in ordinate. It should have appeared at -1.1.)
The regression line for the six \[WCL\] objects (after excluding CPD-568032 and He 2-113) with the 7.7-$`\mu `$m band is insignificantly different (at the 1$`\sigma `$ level) from the line for the 15 PNe. M 4-18 has the largest value of I(7.7)/I(IR). Ground-based spectra (Aitken & Roche 1982; Rinehart et al. 2002) clearly show its 11.3-$`\mu `$m band although its 8.7-$`\mu `$m feature is convincingly shown only by the Aitken & Roche (1982) spectrum from Mauna Kea. M 4-18’s SWS spectrum is very noisy, although the PAH features are unquestionably detected. However, the spectral levels of the PAH bands in this SWS spectrum match those in the spectrum by Rinehart et al. (2002) to within 10 percent, confirming our estimate of I(7.7)/I(IR) for this object. In the small sample of PNe studied by Cohen et al. (1989) NGC 6302 substantively helped to define the trend at 7.7 $`\mu `$m because of its low C/O ratio. With our enlarged set of PNe NGC 6302 no longer controls the existence of a trend in this diagram. The correlation coefficient remains r=0.76 even if one were to exclude NGC 6302 along with CPD-568032 and He 2-113. There is no cause to reject this PN; we simply make the point that a single PN does not strongly influence the regression line in this sample at 7.7 $`\mu `$m.
One immediately sees that the \[WCL\] PNe dominate the plot by contributing almost all the large ratios of I(7.7)/I(IR). Only NGC 6369, among the \[WCL\] PNe, was not detected in PAH emission. Roche, Aitken & Whitmore (1983) observed this PN from the ground at very low resolving power ($``$40), in a 20<sup>′′</sup> beam which accommodated about half the ionized gas distribution. At best there are possible suggestions of emission humps near 8.7 and 11.3-$`\mu `$m emission in their spectrum. The object has only a single SWS01 spectrum, as opposed to several of the \[WCL\] nebulae which have higher IR surface brightness.
All the 7.7-$`\mu `$m non-detections have relatively short exposure (fast speed) SWS01 observations (1062-1912 s), suggesting that they were spectra with poor signal-to-noise. Almost all PNe with C/O ratios $``$0.85 (log$`_{10}`$C/O $``$-0.07) and I(7.7)/I(IR) $`>`$0.4 percent (log<sub>10</sub> C/O $``$-2.45) were detected in this PAH feature. Seven PNe, lying in the lower-left quadrant of Fig. 3 below these limiting values, were not detected in the 7.7-$`\mu `$m band. Only NGC 6302 is detected outside the above limits of C/O and I(7.7)/I(IR). This detection was certainly aided by the use of the slowest speed SWS01 AOT (6528 s) because the PAHs are weak in this PN and I(IR) is large.
Fig. 4 similarly presents the 11.3-$`\mu `$m results. A single regression line is again plotted for the 12 detected PNe, excluding CPD-568032 and He 2-113. The intercept is -2.59$`\pm `$0.08 and slope 2.12$`\pm `$0.49. The influence of NGC 6302 on this regression line is again minimal: r drops from 0.71 to 0.67 if one were to exclude this object from the regression analysis. Clearly there is no justification for doing this. The PN furthest from the line, with the lowest I(11.3)/I)IR) of the sample, is IC 418 although there is likewise no reason to exclude it. Several nebulae show a broad hump between 10.6 and 12.4 $`\mu `$m attributed to emission by SiC grains. In these objects the 11.3-$`\mu `$m PAH band is a very small emission feature near the peak of the SiC structure, requiring careful definition of a local continuum to extract the PAH feature. Perhaps this accounts for its unusual weakness in IC 418, whose SiC emission is very bright. Overall, I(11.3)/I(IR) rises abruptly at low C/O and maintains this rise at least as far as C/O $`2`$, exactly as found by Volk & Cohen (1990: their Figure 3). Only the two extreme PNe are observed beyond this value of C/O.
From their analysis of the 3.3-$`\mu `$m band, Roche et al. (1996) suggested that the equivalent width (EW) of this band argued for a cut-off below C/O = 0.6. For high quality near-infrared ground-based spectra, EWs are appropriate. For space-based spectra the sample is still limited by signal-to-noise. This can lead to poor estimates of the local continuum required to define an EW. Therefore, I(7.7)/I(IR) is a far more robust measure of PAH strength than EW for our sample of PN spectra. However, in the interest of trying to determine whether our sample supports a similar conclusion, we have reformulated the 7.7-$`\mu `$m data as EWs. But this lack of robustness means that we must exclude several PNe from the sample. These are precisely those objects in which portions of the overall splined continua are so noisy that negative continua result at some wavelengths within the broad 7.7-$`\mu `$m feature, or else very large EWs ($`>`$$`\mu `$m) are formally derived. We again exclude CPD-568032 and He 2-113 but must also reject Sw St 1, NGC 40, NGC 6302, Cn 1-5, and M 4-18. The final sample includes 10 PNe. We have carried out a careful estimation of the errors in the derived EWs by adding in quadrature the uncertainties in each integrated 7.7-$`\mu `$m intensity with those associated with the definition of the underlying continuum. Fractional uncertainties of 30% have been assigned to the observed C/O values. Fig. 5 presents the EWs for these PNe linearly against C/O. The solid line is the best fitting relationship. The formal detectable onset of 7.7-$`\mu `$m emission is for C/O = 0.56$`{}_{0.41}{}^{}{}_{}{}^{0.21}`$, which includes the value of 0.6 found by Roche et al. (1996), at which C/O value they estimate that 3.3-$`\mu `$m PAH emission first appears.
Comparing the behaviour of CPD-568032 and He 2-113 with the regressions in Figures 3 and 4 suggests that perhaps these two objects are so extreme because some process “saturates” so that further increase of C/O no longer yields more intense PAH emission relative to the overall dust continuum emission. For both these PNe, essentially all of the stellar ultraviolet (UV) photons are being absorbed and reradiated in the IR by the PAHs and dust particles (Aitken et al. 1980), which is not the case for most of the other objects in our sample. Around these two objects, there could therefore be a significant stratification with radius of both the spectral quality of the ambient UV radiation field, with harder photons being absorbed closer in, and of the characteristic grain temperature, so that strong PAH emission closer in may be offset by weak PAH and strong far-IR grain emission further out. Therefore, the best guides to any underlying correlations between band strengths and C/O ratios are those provided by our PN samples that omit CPD-568032 and He 2-113.
The 6.2-$`\mu `$m PAH band is seen in 14 PNe and their relative strengths, I(6.2)/I(IR), are given in Table 3. In most of these nebulae this band is measured rather poorly because high noise in the SWS spectra below 5-6 $`\mu `$m adversely affects the definition of the underlying continua that we subtract. There is no meaningful correlation between I(6.2)/I(IR) and C/O.
## 6 Conclusions
PAHs were not detected in two of the three largest nebulae in our sample, either because the circumnebular PDRs that ought to contain PAHs are not sampled by the modest sizes of the SWS apertures, or else because these lower density nebulae are optically thin to ionizing radiation and have no PDRs. This could explain the absence of detectable 7.7-$`\mu `$m emission bands in NGC 5189 and NGC 6720. PAHs were detected by the SWS in NGC 6537, despite its diameter of 70<sup>′′</sup>. However, this object is bipolar and has a similarly high excitation class and morphology to NGC 6302, in which PAHs are also detected. This morphology implies a central concentration of material that is lacking in large classical PNe like NGC 6720. Indeed, the strong 3.3 and 3.4-$`\mu `$m bands seen in a 3<sup>′′</sup> slit spectrum of NGC 6537 by Roche et al. (1996) offer confirmation that the PAH emission in at least this nebula arises in the compact core.
Using a sample of PNe more than twice as large as that previously available from airborne spectroscopy, we have vindicated the existence of a relationship between I(7.7)/I(IR) and gas phase C/O. The fraction of total IR luminosity emitted by the 7.7-$`\mu `$m band is observed to be roughly linearly proportional to C/O for abundance ratios up to $``$3. From a sample of PNe about the same size as that used in the LRS study by Volk & Cohen (1990) but based on higher signal-to-noise SWS spectra, we have confirmed that I(11.3)/I(IR) rises rapidly with C/O to at least C/O $``$2, growing roughly as the square of the gas-phase abundance ratio. For nebulae with C/O beyond a value of 2-3 this fractional growth of the PAH emission bands apparently ceases. The difference between the logarithmic intercepts of these two relationships (-2.59 and -2.16) indicates that, for PNe with C/O$``$1, the intensity of the 11.3-$`\mu `$m band is 37 percent of the 7.7-$`\mu `$m band’s intensity. Note that the use of logarithmic plots greatly reduces the dependence of this ratio of band strengths on abundance ratio because the centroid of our sample is C/O$``$1. This is in excellent agreement with the average of 36 percent measured by Cohen et al. (1986, their Table 5) for the generic spectrum of astrophysical PAHs. The 1986 sample included only 7 PNe for which band ratios relative to I(7.7) were definable and included Hii regions and reflection nebulae too. We have, therefore, redetermined I(6.2)/I(7.7) and I(11.3)/I(7.7) purely from our new sample of PNe, and find a mean ratio and standard error of the mean of 0.62$`\pm `$0.13 (13 PNe) and 0.40$`\pm `$0.06 (14 PNe), respectively.
These empirical correlations between PAH band strength and nebular C/O abundance ratio could yield insights into the carbon dust condensation process because PAHs have been suggested to be the likely molecular precursors of C grains (Crawford, Tielens & Allamandola 1985; Allamandola, Tielens & Barker 1989). I(7.7)/I(IR) essentially measures the far-ultraviolet absorption cross sections of PAHs with respect to that of the larger C-grains, believed to dominate the continuum emission in planetaries. One obvious interpretation of the observed correlations in Figs. 3 and 4 is that n(PAH)/n(carbon grains) increases with increasing C/O ratio. If, on the other hand, n(PAH)/n(carbon grains) is largely invariant, one could interpret the correlations of PAH band strengths with C/O as due to growth of the grain component (higher C/O leading to bigger particles with smaller UV cross sections per unit volume, so that PAHs absorb more UV energy relative to the grains).
## 7 Acknowledgements
We thank the anonymous referee for comments that have helped improve the paper. MC thanks NASA for supporting the early portion of this work through grant NAG5-4884, and later through contract JPL961501, with Berkeley. It is a pleasure to thank the Physics & Astronomy Dept. at UCL for hosting a number of short-term visits that have enabled the completion of this effort. |
warning/0507/nlin0507062.html | ar5iv | text | # 1 Introduction
## 1 Introduction
Recently, the celebrated shallow-water equation obtained by Camassa and Holm
$$u_tu_{txx}=3uu_x+2u_xu_{xx}+uu_{xxx}\kappa u_x,\kappa =\mathrm{const}$$
(1)
was extended in by adding on the right hand side a term $`\rho \rho _x`$ with a new variable $`\rho `$, which satisfies the continuity equation $`\rho _t+(u\rho )_x=0`$. The model resulting from the above generalization first appeared in the study of deformations of the bihamiltonian structure of hydrodynamic type and was coined 2-component Camassa-Holm equation. Soon after its derivation the model was identified with the first negative flow of the AKNS hierarchy .
Another well-known integrable partial differential equation of interest to our study is the Dym-type equation :
$$u_{xxt}=2u_xu_{xx}+uu_{xxx}\kappa u_x.$$
(2)
It can also be extended to two-component version by adding a term $`\rho \rho _x`$ on the right hand side of (2). The resulting two variable system is shown here to be equivalent to the negative flow of one of the extensions of the AKNS model. It is also equivalent to the special limiting procedure of deformations of the bihamiltonian structure of hydrodynamic type.
In this paper the following is accomplished: First, we explore the Schrödinger spectral problem of second order describing both 2-component Camassa-Holm and Dym type equation for different values of the deformation parameter $`\mu `$. We show that this Schrödinger spectral problem can be cast into the linear $`2\times 2`$ matrix spectral problem. Using sl$`(2)`$ gauge invariance we transform the time evolution flow of the linear spectral problem into the AKNS first negative flow. We should point out that there exist several ways to extend the AKNS hierarchy to incorporate negative flows. These extensions are parametrized by a single parameter identified with $`\mu `$. We associate two different constructions of the negative flows of the AKNS hierarchy to the 2-component Camassa-Holm (for $`\mu =1`$) and Dym type equation (for $`\mu =0`$). The result concerning the 2-component Camassa-Holm equation constitutes an algebraic version of the proof given in . Using connections of the AKNS and deformed Sinh-Gordon models to the 2-component Camassa-Holm and Dym type equations, respectively, we are able to find explicit soliton solutions given in hodographic variables. The relation to the AKNS models allows us to construct a new chain of charges conserved with respect to equations of motion of two-component Camassa-Holm and two-component Dym type equations. For both hierarchies the modified AKNS Hamiltonians provide a tower of positive order Hamiltonians obtained via the underlying Lenard relations of the Poisson brackets of hydrodynamic type from the Casimir of the second bracket.
In section 2, we briefly review the algebraic approach to the AKNS model and show how to extend the model in two different ways to negative time flows based on the zero-curvature identities. In section 3, we set up a class of two-component Schrödinger spectral problems parametrized by $`\mu `$. In the next section 4, we transform the Schrödinger spectral problem by the reciprocal transformation and linearize it. The resulting linear $`2\times 2`$ matrix spectral problem is then transformed by an sl$`(2)`$ gauge transformation into the AKNS Lax spectral problem. The time flows of 2-component Camassa-Holm and Dym type equation are shown to coincide with two different negative flows of the extended AKNS model. Our construction allows us to find, in subsection 4.1, explicit soliton solutions for various values of $`\mu `$. In section 5, we reproduce equations of motion for $`\mu 0`$ and $`\mu =0`$ cases in the setting of deformations of the bihamiltonian structure of hydrodynamic type. Remarkably, the Hamiltonians governing positive evolution flows of the AKNS hierarchy define conserved charges for the 2-component Camassa-Holm and Dym type equations. Also the conserved charges induced by the AKNS model satisfy among themselves the Lenard relations of the bihamiltonian structure of hydrodynamic type. Thus, the Hamiltonians of the bihamiltonian structure of hydrodynamic type connected to 2-component Camassa-Holm and Dym type equations split into two chains, one of the positive order induced by the AKNS hierarchy and one of the negative order containing generators of the equations of motion defining both hierarchies.
## 2 Extended AKNS model
First, let us present the AKNS hierarchy in the setting of the sl$`(2)`$ loop algebra endowed with homogeneous gradation defined by the operator $`\lambda d/d\lambda `$. A variable $`\lambda `$ plays a double role of a loop parameter of the loop algebra and a spectral parameter of the underlying hierarchy. The matrix Lax operator $`L`$ for the AKNS hierarchy reads:
$$L=\frac{}{y}\left[\begin{array}{cc}\lambda & 0\\ 0& \lambda \end{array}\right]\left[\begin{array}{cc}0& q\\ r& 0\end{array}\right],$$
(3)
where $`/y`$ is the derivative with respect to “space” variable $`y`$. The matrix Lax operator can be compactly written as $`L=/yEA_0`$, with $`E=\lambda \sigma _3`$ and the matrix $`A_0=q\sigma _++r\sigma _{}`$, where $`\sigma _3`$ is the Pauli matrix and $`\sigma _\pm `$ are given in terms of other Pauli matrices $`\sigma _1,\sigma _2`$
$$\sigma _{}=\frac{1}{2}\left(\sigma _1\mathrm{i}\sigma _2\right)=\left[\begin{array}{cc}0& 0\\ 1& 0\end{array}\right],\sigma _+=\frac{1}{2}\left(\sigma _1+\mathrm{i}\sigma _2\right)=\left[\begin{array}{cc}0& 1\\ 0& 0\end{array}\right].$$
We work within an algebraic approach to the integrable models based on the linear spectral problem $`L(\mathrm{\Psi })=0`$, which simplifies considerably under a dressing transformation:
$$\mathrm{\Theta }^1\left(\frac{}{y}EA_0\right)\mathrm{\Theta }=\frac{}{y}E,$$
(4)
where the dressing matrix $`\mathrm{\Theta }=\mathrm{exp}\left(_{i<0}\lambda ^i\theta ^{(i)}\right)`$ is an exponential in negative powers of the spectral parameter $`\lambda `$ on a formal loop space of sl$`(2)`$. Similarly, for higher flows we obtain
$$\mathrm{\Theta }^1\left(\frac{}{t_n}E^{(n)}\underset{i=0}{\overset{n1}{}}\lambda ^iD_n^{(i)}\right)\mathrm{\Theta }=\frac{}{t_n}E^{(n)},$$
(5)
where $`E^{(n)}=\lambda ^n\sigma _3`$ and terms $`D_n^{(i)}`$ are obtained from projection $`(\mathrm{\Theta }E^{(n)}\mathrm{\Theta }^1)_+`$ of $`\mathrm{\Theta }E^{(n)}\mathrm{\Theta }^1`$ on the positive powers of $`\lambda `$ via expansion relation:
$$\left(\mathrm{\Theta }E^{(n)}\mathrm{\Theta }^1\right)_+=E^{(n)}+\underset{i=0}{\overset{n1}{}}\lambda ^iD_n^{(i)}.$$
These dressing relations give rise to the zero-curvature conditions for the positive flows of the AKNS hierarchy
$$[\frac{}{y}EA_0,\frac{}{t_n}E^{(n)}\underset{i=0}{\overset{n1}{}}D_n^{(i)}]=\mathrm{\Theta }[\frac{}{y}E,\frac{}{t_n}E^{(n)}]\mathrm{\Theta }^1=0.$$
(6)
In particular, for $`n=2`$ we obtain the second flow of the AKNS hierarchy:
$$\frac{r}{t_2}=\frac{1}{2}r_{yy}+qr^2;\frac{q}{t_2}=\frac{1}{2}q_{yy}q^2r,$$
(7)
which reproduces the familiar vector non-linear Schrödinger equation.
According to , the Hamiltonian densities of the AKNS model are defined as
$$_n=tr\left(E^{(0)}A^{(n)}\right)=\frac{1}{2}\underset{k=0}{\overset{n1}{}}tr\left(A^{(k)}A^{(1+kn)}\right),$$
(8)
where $`A^{(n)}`$ are given by
$$\mathrm{\Theta }_y\mathrm{\Theta }^1=\underset{k=1}{\overset{\mathrm{}}{}}A^{(k)}\lambda ^k,$$
where the symbol $`tr`$ in expression (8) denotes a sl$`(2)`$ trace. We list the first two Hamiltonians. Inserting $`n=1`$ in (8) we obtain:
$$_1=tr(E^{(0)}A^{(1)})=\frac{1}{2}tr(A_0^2)=rq.$$
(9)
Similarly, for $`n=2`$ we obtain
$$_2=qr_y.$$
(10)
Next, we extend the AKNS model to include negative grade time evolution equations governed by the zero-curvature equations
$$[\frac{}{y}EA_0,\frac{}{t_n}D^{(1)}D^{(2)}\mathrm{}D^{(n)}]=0.$$
(11)
Here, we only consider the first negative flow with $`n=1`$ and set for brevity $`s=t_1`$. In this case the compatibility equation (11) reduces to
$`(A_0)_sD_y^{(1)}+[E+A_0,D^{(1)}]=0.`$ (12)
A general solution of the compatibility equation (12) is given by
$$D^{(1)}=B^{(1)}B^1,A_0=B_yB^1,$$
(13)
in terms of the zero-grade group element, $`B`$, of $`\mathrm{SL}(2)`$, that satisfies equation:
$$(B_yB^1)_s=[B^{(1)}B^1,E]$$
(14)
or, equivalently,
$$(B^1B_s)_y=[^{(1)},B^1EB].$$
(15)
Here $`^{(1)}`$ is an element of sl$`(2)`$ algebra of $`1`$ grade.
Remarkably, the compatibility of the $`t_1`$ flow with positive $`t_n,n1`$ flows does not require that the matrix $`^{(1)}`$ commutes with $`E=\lambda \sigma _3`$, as pointed out in and . It turns out that all possible cases are parametrized by a parameter $`\mu `$ and fall into two main classes depending on whether $`\mu `$ takes non-zero or zero value. The corresponding generic choices of $`^{(1)}`$ are:
$$^{(1)}=\{\begin{array}{cc}\mu \sigma _3/4\lambda \hfill & \text{for}\mu 0\hfill \\ \sigma _+/\lambda \hfill & \text{for}\mu =0.\hfill \end{array}$$
(16)
Note that the value of determinant of $`^{(1)}`$ is equal to $`\mu ^2/16\lambda ^2`$ and $`0`$, respectively. There exist other choices of $`^{(1)}`$ for these values of the determinant but they only lead to the gauge equivalent copies of hierarchies derived from the choice (16).
## 3 A class of two-component Schrödinger spectral problems
Consider a linear system
$`\psi _{xx}`$ $`=`$ $`\left({\displaystyle \frac{\mu ^2}{4}}m\lambda +\rho ^2\lambda ^2\right)\psi ,`$ (17)
$`\psi _t`$ $`=`$ $`({\displaystyle \frac{1}{2\lambda }}+u)\psi _x+{\displaystyle \frac{1}{2}}u_x\psi ,`$ (18)
for some arbitrary constant $`\mu `$ (see for $`\mu =1`$ and for $`\mu =0`$). Compatibility condition for the above system, yields three independent equations
$`\rho _t`$ $`=`$ $`\left(u\rho \right)_x,`$ (19)
$`m_t`$ $`=`$ $`2mu_xm_xu+\rho \rho _x,`$ (20)
$`m_x`$ $`=`$ $`\mu ^2u_xu_{xxx},`$ (21)
corresponding to coefficients of $`\lambda ^2`$, $`\lambda `$ and $`\lambda ^0`$ in the expansion of $`\psi _{xxt}\psi _{txx}=0`$. Equation (21) can be integrated to yield:
$$m=\mu ^2uu_{xx}+\frac{1}{2}\kappa .$$
(22)
Here, $`\kappa `$ is an integration constant. For $`\mu 0`$ that integration constant can be removed by transforming the system by Galilean transformation:
$$x^{}=x+vt,t^{}=t,\frac{}{x^{}}=\frac{}{x},\frac{}{t}=v\frac{}{x^{}}+\frac{}{t^{}}.$$
In the primed system equation (20) becomes:
$$vm_x^{}+m_t^{}=2mu_x^{}m_x^{}u+\rho \rho _x^{}=2\left(\mu ^2uu_{x^{}x^{}}+\frac{1}{2}\kappa \right)u_x^{}m_x^{}u+\rho \rho _x^{}$$
Next, performing a shift $`uuv`$ and choosing velocity $`v`$ such that $`v=\kappa /2\mu ^2`$ eliminates the linear terms in $`u_x`$ and $`u_{xxx}`$ from the above equation. Clearly, the above argument works only for $`\mu 0`$ and from now on we put the integration constant $`\kappa `$ to zero as long as $`\mu 0`$.
Note, that the positive constant $`\mu ^2`$, that is different from one can be absorbed by appropriately redefining fields and derivatives. Defining $`\stackrel{~}{\lambda }=\lambda /\mu ^2`$, $`\stackrel{~}{\rho }=\rho \mu `$, $`\stackrel{~}{u}=u\mu ^2`$ and new variables $`\stackrel{~}{x}`$ and $`\stackrel{~}{t}`$ such that $`_x=\mu _{\stackrel{~}{x}}`$, $`_t=(1/\mu )_{\stackrel{~}{t}}`$ allows us to rewrite a linear system (17)-(18) as
$`\psi _{\stackrel{~}{x}\stackrel{~}{x}}`$ $`=`$ $`\left({\displaystyle \frac{1}{4}}m\stackrel{~}{\lambda }+\stackrel{~}{\rho }^2\stackrel{~}{\lambda }^2\right)\psi ,`$ (23)
$`\psi _{\stackrel{~}{t}}`$ $`=`$ $`({\displaystyle \frac{1}{2\stackrel{~}{\lambda }}}+\stackrel{~}{u})\psi _{\stackrel{~}{x}}+{\displaystyle \frac{1}{2}}\stackrel{~}{u}_{\stackrel{~}{x}}\psi .`$ (24)
with $`m=\mu ^2uu_{xx}+\kappa /2=\stackrel{~}{u}\stackrel{~}{u}_{\stackrel{~}{x}\stackrel{~}{x}}+\kappa /2`$. Thus, for $`\mu ^21`$ the spectral system has been transformed to the canonical system with $`\mu ^2=1`$.
In case of a negative $`\mu ^2`$ (imaginary $`\mu `$) we make the changes as above but with $`|\mu ^2|`$ instead of $`\mu ^2`$ and arrive at
$$\psi _{\stackrel{~}{x}\stackrel{~}{x}}=\left(\frac{1}{4}m\stackrel{~}{\lambda }+\stackrel{~}{\rho }^2\stackrel{~}{\lambda }^2\right)\psi .$$
(25)
Thus, only three cases of $`\mu ^2=1,0,1`$ need to be considered separately, as concerns equations of motion.
The case of $`\mu ^2=1`$ corresponds to the 2-component Camassa-Holm model with $`m=uu_{xx}+\kappa /2`$, introduced in , while the case of $`\mu ^2=1`$ corresponds to equation (20) with $`m=uu_{xx}+\kappa /2`$, which for $`\rho =0`$ is well-known to possess the compacton solutions .
For $`\mu =0`$, we obtain from (22) $`m=u_{xx}+\kappa /2`$. Inserting this value of $`m`$ into equation (20) yields
$$u_{xxt}=2u_xu_{xx}uu_{xxx}+\kappa u_x\rho \rho _x.$$
(26)
For $`\rho =0`$ this is the Dym type equation (2).
After one integration (and ignoring the integration constant) we obtain from (26)
$$\begin{array}{cc}\hfill 0& =u_{xt}+uu_{xx}\kappa u+\frac{1}{2}u_x^2+\frac{1}{2}\rho ^2\hfill \\ & =\left(u_t+uu_x\right)_x\kappa u+\frac{1}{2}\left(u_x^2+\rho ^2\right).\hfill \end{array}$$
(27)
In terms of a new function
$$v=\frac{1}{2}\left(u_x^2+\rho ^2\right)$$
(28)
we can cast equation (27) in the following form
$$\left(u_t+uu_x\right)_x\kappa u+v=0.$$
(29)
In addition, it follows from equations (27) and (19) that $`v`$ defined by relation (28) also satisfies
$$v_t+\left(u(v+u\kappa /2)\right)_x=0,$$
(30)
which becomes a continuity equation in the $`\kappa =0`$ limit. The linear system corresponding to equations (29) and (30) takes a form
$$\begin{array}{cc}\hfill \psi _{xx}& =\left((u_{xx}\kappa /2)\lambda +\left(2v+u_x^2\right)\lambda ^2\right)\psi ,\hfill \\ \hfill \psi _t& =(\frac{1}{2\lambda }+u)\psi _x+\frac{1}{2}u_x\psi .\hfill \end{array}$$
(31)
## 4 Transformation to the first order spectral problem. Algebraic Connection to the AKNS model
Now, for an arbitrary $`\mu `$ we perform a reciprocal transformation $`(x,t)(y,s)`$ defined by relations
$$dy=\rho dx\rho udt,ds=dt,$$
(32)
and
$$\frac{}{x}=\rho \frac{}{y},\frac{}{t}=\frac{}{s}\rho u\frac{}{y}.$$
(33)
The commutativity of derivatives with respect to $`s`$ and $`y`$ variables is ensured by the continuity equation (19). Applying the reciprocal transformation and then redefining $`\psi `$ by $`\phi =\sqrt{\rho }\psi `$ as in leads from the spectral problem (17)-(18) to :
$`\phi _{yy}`$ $`=`$ $`\left(\lambda ^2P\lambda Q\right)\phi ,`$ (34)
$`\phi _s`$ $`=`$ $`{\displaystyle \frac{\rho }{2\lambda }}\phi _y+{\displaystyle \frac{\rho _y}{4\lambda }}\phi ,`$ (35)
where
$$\begin{array}{cc}\hfill P& =\frac{m}{\rho ^2}\hfill \\ \hfill Q& =\frac{\mu ^2}{4\rho ^2}\frac{\rho _{yy}}{2\rho }+\frac{\rho _y^2}{4\rho ^2}.\hfill \end{array}$$
(36)
Our main point in this section is that we can rewrite the second-order spectral problem (34)-(35) as a first-order linear problem:
$`\left[\begin{array}{c}\phi \\ \eta \end{array}\right]_y`$ $`=`$ $`A\left[\begin{array}{c}\phi \\ \eta \end{array}\right]`$ (37)
$`\left[\begin{array}{c}\phi \\ \eta \end{array}\right]_s`$ $`=`$ $`D\left[\begin{array}{c}\phi \\ \eta \end{array}\right],`$ (38)
involving $`\mathrm{sl}(2)`$ matrices:
$$\begin{array}{cc}\hfill A& =\left[\begin{array}{cc}g& \lambda \\ \lambda P& g\end{array}\right]=\lambda \sigma _1+g\sigma _3P\sigma _{},\hfill \\ \hfill D& =\frac{1}{\lambda }D_0\frac{1}{2}\rho \sigma _1,D_0=\frac{\mu }{4}\sigma _3+\frac{1}{4}\left(P\rho 2g_s\right)\sigma _{}.\hfill \end{array}$$
(39)
Note that determinant of $`D_0`$ is equal to $`\mathrm{det}D_0=\mu ^2/16`$ and, therefore, the matrix $`D_0`$ becomes singular for $`\mu =0`$. Eliminating $`\eta `$ from the linear spectral problem (37)-(38) reproduces equations (34)-(35) for $`\phi `$ providing that function $`g(y,s)`$ appearing in (39) satisfies the Riccati equation
$$Q=g^2g_y$$
for $`Q`$ given in equation (36). Remarkably, the solution to the above Riccati equation takes a local form of
$$g(y,s)=\frac{\mu }{2\rho }+\frac{\rho _y}{2\rho }.$$
(40)
The zero-curvature equation
$$A_sD_y+[A,D]=0,$$
(41)
can easily be derived from the linear spectral problem (37)-(38). It is equivalent to equations:
$$P_s=\rho _y,Q_s+\frac{1}{2}P_y\rho +P\rho _y=0.$$
(42)
These equations were found in directly from compatibility of equations (34)-(35). It follows from the first of the above equations that there exists a function $`f(y,s)`$ such that $`P=f_y`$ and $`\rho =f_s`$.
By plugging $`\rho =f_s`$ and $`P=f_y`$ into the second relation in (42) one obtains as in :
$$\mu ^2\frac{f_{ss}}{2f_s^3}+f_{sy}f_y+\frac{1}{2}f_sf_{yy}\frac{f_{ssyy}}{2f_s}+\frac{f_{ssy}f_{sy}}{2f_s^2}+\frac{f_{ss}f_{syy}}{2f_s^2}\frac{f_{ss}f_{sy}^2}{2f_s^3}=0.$$
(43)
This appears to be the only condition, which the function $`f`$ has to satisfy in order to be a solution of the model.
For $`\mu =0`$ equation (43) simplifies to
$$\left(f_s^2f_yf_{ssy}+\frac{f_{ss}f_{sy}}{f_s}\right)_y=0.$$
(44)
Integrating the above equation once and setting the integration constant to $`\kappa /2`$ (see explanation below) yields
$$\frac{f_{ssy}}{f_s^2}\frac{f_{ss}f_{sy}}{f_s^3}+\frac{\frac{1}{2}\kappa }{f_s^2}=f_y,$$
(45)
or
$$\left(\mathrm{ln}f_s\right)_{sy}+\frac{1}{2}\kappa /f_s=f_sf_y.$$
Indeed, multiplying both sides of equation (45) by $`f_s^2`$ and taking a derivative with respect to $`y`$ yields (44). It remains to be shown that the choice of $`\kappa /2`$ as the integration constant in equation (45) was consistent with equations of motion. To do this we start by recalling that $`P=m/\rho ^2`$ with
$$m=u_{xx}+\frac{1}{2}\kappa =\rho (\rho u_y)_y+\frac{1}{2}\kappa $$
in the $`\mu =0`$ case. The continuity equation (19) reads in the hodographic variables $`\rho _s=\rho ^2u_y`$. Accordingly, substituting $`u_y=(1/\rho )_s`$ into $`P`$ we get
$$P=\frac{(\rho _s/\rho )_y}{\rho }+\frac{\frac{1}{2}\kappa }{\rho ^2}=f_y,$$
which is precisely equation (45).
Let us turn our attention back to the zero-curvature equation (41). This equation is invariant under the $`\mathrm{sl}(2)`$ gauge transformation:
$$AUAU^1+U_yU^1,DUDU^1+U_sU^1.$$
This invariance will be used in what follows to cast the linear spectral problem (37)-(38) in the standard form of the first positive and first negative flow equations of the $`\mathrm{sl}(2)`$ AKNS hierarchy.
As a first step we gauge away the term $`\frac{1}{2}\rho \sigma _1`$ of order $`\lambda ^0`$ in the expression for $`D`$ in equation (39) by choosing
$$U=\mathrm{exp}\left(\frac{1}{2}f(y,s)\sigma _1\right)=\mathrm{cosh}\frac{f}{2}+\sigma _1\mathrm{sinh}\frac{f}{2}$$
so that $`U_sU^1\frac{1}{2}\rho \sigma _1=0`$, due to $`f_s=\rho `$. Consequently,
$$\begin{array}{cc}\hfill A& UAU^1+U_yU^1=U\left(\lambda \sigma _1+g\sigma _3P\sigma _{}\right)U^1+\frac{1}{2}P\sigma _1\hfill \\ & =\lambda \sigma _1+\sigma _3\left(g\mathrm{cosh}f\frac{1}{2}P\mathrm{sinh}f\right)\mathrm{i}\sigma _2\left(g\mathrm{sinh}f\frac{1}{2}P\mathrm{cosh}f\right)\hfill \\ \hfill D& \frac{1}{\lambda }UD_0U^1=\frac{1}{4\lambda }[(\sigma _1\mathrm{cosh}f\mathrm{i}\sigma _2\mathrm{sinh}f)\mu \hfill \\ & +(P\rho 2g_s)(\sigma _3+\sigma _1\mathrm{cosh}f+\mathrm{i}\sigma _2\mathrm{sinh}f)]\hfill \end{array}$$
Note that the gauge transformed of the matrix $`D`$ is now proportional to $`1/\lambda `$.
Next, we define the constant matrix $`\mathrm{\Omega }=\frac{1}{\sqrt{2}}\left(\sigma _1+\sigma _3\right)`$, that by a similarity transformation maps $`\sigma _1`$ to $`\sigma _3`$, $`\mathrm{\Omega }\sigma _1\mathrm{\Omega }^1=\sigma _3`$. Note also that $`\mathrm{\Omega }\sigma _2\mathrm{\Omega }^1=\sigma _2`$. The combined gauge transformations first by $`U`$ and then by $`\mathrm{\Omega }`$ produce the final result
$$\begin{array}{cc}\hfill A& E+A_0=\mathrm{\Omega }\left[UAU^1+U_yU^1\right]\mathrm{\Omega }^1\hfill \\ & =\lambda \sigma _3+\sigma _1\left(g\mathrm{cosh}f\frac{1}{2}P\mathrm{sinh}f\right)+\mathrm{i}\sigma _2\left(g\mathrm{sinh}f\frac{1}{2}P\mathrm{cosh}f\right)\hfill \\ \hfill D& D^{(1)}=\frac{1}{\lambda }\mathrm{\Omega }UD_0U^1\mathrm{\Omega }^1=\frac{1}{4\lambda }[(P\rho 2g_s)\sigma _3.\hfill \\ & .+\sigma _1((P\rho 2g_s)\mathrm{sinh}f\mu \mathrm{cosh}f)+\mathrm{i}\sigma _2((P\rho 2g_s)\mathrm{cosh}f\mu \mathrm{sinh}f)]\hfill \end{array}$$
(46)
In the above, we re-introduced $`E=\lambda \sigma _3`$ and $`A_0=r\sigma _{}+q\sigma _+`$. Comparing with the right hand side of equation (46) we find that
$$q=e^f\left(g\frac{P}{2}\right),r=e^f\left(g+\frac{P}{2}\right).$$
(47)
Furthermore, defining matrix entries of $`D^{(1)}`$ as
$$D^{(1)}=\frac{1}{\lambda }\left[\begin{array}{cc}\alpha & \beta \\ \gamma & \alpha \end{array}\right],$$
(48)
we find from (46), that $`\alpha ,\beta `$ and $`\gamma `$ are given by
$$\begin{array}{cc}\hfill \alpha & =\frac{1}{4}\left(P\rho 2g_s\right)\hfill \\ \hfill \beta & =e^f\left(\alpha \frac{\mu }{4}\right)\hfill \\ \hfill \gamma & =e^f\left(\alpha +\frac{\mu }{4}\right).\hfill \end{array}$$
(49)
They satisfy the determinant formula $`\alpha ^2+\beta \gamma =\mu ^2/16`$. The matrix entries of $`A_0`$ and $`D^{(1)}`$ enter the following simple identities
$`2\alpha `$ $`=\beta e^f\gamma e^f,`$ $`{\displaystyle \frac{\mu }{2}}`$ $`=\beta e^f+\gamma e^f`$ (50)
$`P`$ $`=re^fqe^f,`$ $`g`$ $`={\displaystyle \frac{1}{2}}\left(re^f+qe^f\right).`$ (51)
It follows that the linear spectral problem (37)-(38) has been transformed by the above gauge transformation to:
$`\mathrm{\Psi }_y`$ $`=`$ $`\left(E+A_0\right)\mathrm{\Psi }=\left[\begin{array}{cc}\lambda & 0\\ 0& \lambda \end{array}\right]\mathrm{\Psi }+\left[\begin{array}{cc}0& q\\ r& 0\end{array}\right]\mathrm{\Psi }`$ (52)
$`\mathrm{\Psi }_s`$ $`=`$ $`D^{(1)}\mathrm{\Psi }={\displaystyle \frac{1}{\lambda }}\left[\begin{array}{cc}\alpha & \beta \\ \gamma & \alpha \end{array}\right]\mathrm{\Psi }`$ (53)
for some two-component object $`\mathrm{\Psi }`$.
We recognize in (52) the spectral problem $`L\left(\mathrm{\Psi }\right)=0`$ with the AKNS Lax operator given by equation (3). It also follows easily that equation (12) is the compatibility equation of the spectral equations (52)-(53). The compatibility equation (12) yields
$$q_s=2\beta ,r_s=2\gamma .$$
(54)
when projected on zero grade, and
$$\begin{array}{cc}\hfill \alpha _y& =\frac{1}{2}(rq)_s=q\gamma r\beta \hfill \\ \hfill \beta _y& =2\alpha q\hfill \\ \hfill \gamma _y& =2\alpha r,\hfill \end{array}$$
(55)
when projected on $`1`$ grade. Equations (55) can also be directly derived from equations of motion (42).
### 4.1 Examples and Solutions
Let us recall that a general solution of the compatibility equation (12) is given by expressions from equation (13). It is convenient to parametrize the $`\mathrm{SL}(2)`$ group element $`B`$ appearing in expressions in (13) by the $`\mathrm{sl}(2)`$ algebra elements through the Gauss decomposition:
$$B=e^{\chi \sigma _{}}e^{R\sigma _3}e^{\varphi \sigma _+}.$$
(56)
#### 4.1.1 The case of $`\mu 0`$
As an example, we first consider $`\mu ^2=4`$ with $`^{(1)}=\sigma _3/2\lambda `$ according to equation (16). As in in order to match the number of independent modes in the matrix $`A_0`$ we impose two “diagonal” constraints $`Tr\left(B_yB^1\sigma _3\right)=0`$ and $`Tr\left(B^1B_s\sigma _3\right)=0`$, which effectively eliminate $`R`$ in terms of $`\varphi `$ and $`\chi `$.
From $`B_yB^1=q\sigma _++r\sigma _{}`$ we obtain the following representation for $`q`$ and $`r`$:
$$q=\frac{h_y}{\mathrm{\Delta }}e^R;r=\overline{h}_ye^R$$
(57)
where
$$h=\varphi e^R;\overline{h}=\chi e^R;\mathrm{\Delta }=1+h\overline{h}$$
(58)
with a non-local field $`R`$ being determined in terms $`h`$ and $`\overline{h}`$ from the “diagonal” constraints:
$`Tr\left(B_yB^1\sigma _3\right)`$ $`=`$ $`0R_y={\displaystyle \frac{\overline{h}h_y}{\mathrm{\Delta }}},`$ (59)
$`Tr\left(B^1B_s\sigma _3\right)`$ $`=`$ $`0R_s={\displaystyle \frac{h\overline{h}_s}{\mathrm{\Delta }}}.`$ (60)
The zero curvature equations are in this parameterization
$`q_s`$ $`=`$ $`\left({\displaystyle \frac{h_y}{\mathrm{\Delta }}}e^R\right)_s=2he^R,`$ (61)
$`r_s`$ $`=`$ $`\left(\overline{h}_ye^R\right)_s=2\overline{h}\mathrm{\Delta }e^R.`$ (62)
The two-parameter solution to the above equations can be deduced from a method combining dressing and vertex techniques as described e.g. in . The explicit expression is found to be given by:
$$\begin{array}{cc}\hfill h& =\frac{b\mathrm{exp}\left(s/\gamma _2y\gamma _2\right)}{1+\mathrm{\Gamma }\mathrm{exp}\left(s\left(\frac{1}{\gamma _2}\frac{1}{\gamma _1}\right)y\left(\gamma _2\gamma _1\right)\right)}\hfill \\ \hfill \overline{h}& =\frac{a\mathrm{exp}\left(s/\gamma _1+y\gamma _1\right)}{1+\mathrm{\Gamma }\mathrm{exp}\left(s\left(\frac{1}{\gamma _2}\frac{1}{\gamma _1}\right)y\left(\gamma _2\gamma _1\right)\right)}\hfill \\ \hfill e^R& =\frac{1+\frac{\gamma _1}{\gamma _2}\mathrm{\Gamma }\mathrm{exp}\left(s\left(\frac{1}{\gamma _2}\frac{1}{\gamma _1}\right)y\left(\gamma _2\gamma _1\right)\right)}{1+\mathrm{\Gamma }\mathrm{exp}\left(s\left(\frac{1}{\gamma _2}\frac{1}{\gamma _1}\right)y\left(\gamma _2\gamma _1\right)\right)},\hfill \end{array}$$
(63)
where
$$\mathrm{\Gamma }=\frac{ab\gamma _1\gamma _2}{(\gamma _1\gamma _2)^2}$$
is given in terms of four arbitrary constants $`a,b,\gamma _1,\gamma _2`$. Higher multi-soliton solutions can be obtained using the same straightforward procedure.
Comparing with equations (54) we find that
$$he^R=e^f\alpha _{},\overline{h}\mathrm{\Delta }e^R=e^f\alpha _+$$
(64)
where we introduced the notation
$$\alpha _\pm =\alpha \pm \frac{1}{2}.$$
By multiplying the above two relations we find that
$$h\overline{h}\mathrm{\Delta }=\left(\mathrm{\Delta }1\right)\mathrm{\Delta }=\alpha _+\alpha _{}=\alpha ^2\frac{1}{4}.$$
Solutions to this quadratic equation are
$$\mathrm{\Delta }=\pm \alpha _\pm ,h\overline{h}=\mathrm{\Delta }1=\pm \alpha _{}.$$
(65)
Adding two relations in (64) we get
$$2\alpha =\alpha _++\alpha _{}=hx+\frac{1}{x}\overline{h}\mathrm{\Delta },x=e^{Rf}.$$
Solving this quadratic equation yields:
$$f_\pm =R+\mathrm{ln}h\mathrm{ln}\alpha _\pm .$$
(66)
Equation (65) contains two solutions. The first one, namely, $`\alpha _+=\mathrm{\Delta }`$, $`\alpha _{}=\mathrm{\Delta }1`$, when inserted into equation (66) yields
$$f_+=R+\mathrm{ln}h\mathrm{ln}\mathrm{\Delta },f_{}=R+\mathrm{ln}h\mathrm{ln}(\mathrm{\Delta }1)=R\mathrm{ln}\overline{h}.$$
while the second solution, $`\alpha _+=(\mathrm{\Delta }1)`$, $`\alpha _{}=\mathrm{\Delta }`$, leads to
$$f_+=R+\mathrm{ln}h\mathrm{ln}(\mathrm{\Delta }+1),f_{}=R+\mathrm{ln}h\mathrm{ln}(\mathrm{\Delta }).$$
Thus, the Bäcklund transformation
$$f_+=f_{}+ϵ\mathrm{ln}\left(\frac{\mathrm{\Delta }}{\mathrm{\Delta }1}\right),ϵ=\pm 1$$
relates the two values $`f_+`$ and $`f_{}`$. In the reduced case of sinh-Gordon equation with $`h=\overline{h}`$ we find that $`\mathrm{\Delta }=1+h^2`$ and
$$R=\frac{1}{2}\mathrm{ln}\left(1+h^2\right)=\frac{1}{2}\mathrm{ln}\mathrm{\Delta },\mathrm{ln}(\mathrm{\Delta }1)=2\mathrm{ln}h.$$
It follows that all the above values of $`f_+,f_{}`$ can be summarized as:
$$f_ϵ=\frac{ϵ}{2}\mathrm{ln}\frac{1+h^2}{h^2},ϵ=\pm 1.$$
#### 4.1.2 The case of $`\mu =0`$
In the case of $`\mu =0`$, the matrix $`D^{(1)}`$ from equation (48) takes a simple form
$$D^{(1)}=\frac{\alpha }{\lambda }\left[\begin{array}{cc}1& e^f\\ e^f& 1\end{array}\right],$$
(67)
which according to definition (16) is reproduced by
$$\frac{1}{\lambda }B\sigma _+B^1=\frac{1}{\lambda }\left[\begin{array}{cc}\chi & 1\\ \chi ^2& \chi \end{array}\right]e^{2R}$$
for
$$\begin{array}{cc}\hfill \chi & =e^f\hfill \\ \hfill e^{2R}& =\alpha e^f\hfill \end{array}$$
(68)
or $`R=\left(f+\mathrm{ln}\alpha \right)/2`$.
From equation (55) we find for $`\mu =0`$ that $`\alpha _y=2\alpha g`$. Therefore
$$g=\frac{\alpha _y}{2\alpha }=\frac{1}{2}(\mathrm{ln}\alpha )_y$$
(69)
and by comparing with definition (40) we conclude that
$$\left(\rho \alpha \right)_y=0.$$
(70)
Next, we calculate
$$B_yB^1=\left(R_y\chi \varphi _ye^{2R}\right)\sigma _3+\varphi _ye^{2R}\sigma _++\left(\chi _y+2\chi R_y\chi ^2\varphi _ye^{2R}\right)\sigma _{}.$$
(71)
Imposing condition $`Tr\left(B_yB^1\sigma _3\right)=0`$ implies $`R_y\chi \varphi _ye^{2R}=0`$ or
$$\varphi _y=R_y/\alpha =\frac{1}{2\alpha }\left(f_y+\frac{\alpha _y}{\alpha }\right)=\frac{1}{\alpha }\left(gP/2\right).$$
What remains of expression (71) is now given by
$$B_yB^1=\varphi _ye^{2R}\sigma _++\left(\chi _y+\chi R_y\right)\sigma _{}=(R_y/\chi )\sigma _++\left(\chi _y+\chi R_y\right)\sigma _{}.$$
Recalling relations (68) and (69) we obtain the desired results
$$q=\varphi _ye^{2R}=\left(g\frac{1}{2}P\right)e^f,r=\chi _y+\chi R_y=\left(g+\frac{1}{2}P\right)e^f,$$
that reproduce expressions (47).
The compatibility equation
$$\left(B_yB^1\right)_s=[B\sigma _+B^1,\sigma _3]=\left[\begin{array}{cc}0& 2\\ 2\chi ^2& 0\end{array}\right]e^{2R}$$
yields the following equations of motion
$`\left({\displaystyle \frac{R_y}{\chi }}\right)_s`$ $`=`$ $`2e^{2R}`$ (72)
$`\chi _{ys}+2\chi _sR_y`$ $`=`$ $`0\left(\chi _se^{2R}\right)_y=0.`$ (73)
Equation (73) implies that
$$\chi _s=c_3(s)e^{2R},$$
(74)
where $`c_3(s)`$ is a an arbitrary function of $`s`$ only.
From equations (68) we find that
$$\chi _s=f_se^f=\rho e^f=c_3(s)\alpha ^1e^f$$
and therefore $`\rho =c_3(s)/\alpha `$ in agreement with relation (70). It follows that $`c_3(s)`$ has to be different from zero for consistency of the model with $`\rho 0`$.
Integrating relations (74) and (72) leads to
$`\chi `$ $`=`$ $`{\displaystyle ^s}c_3(s)e^{2R}ds+c_2(y)`$ (75)
$`{\displaystyle \frac{R_y}{\chi }}`$ $`=`$ $`2{\displaystyle ^s}e^{2R}ds+c_1(y),`$ (76)
where $`c_1,c_2`$ depend at most on $`y`$. Combining these two equations and setting $`c_3(s)`$ to be a constant $`c_3`$ we get the deformed sinh-Gordon equation for $`R`$ :
$$R_y=2c_3^se^{2R}ds^se^{2R}ds2c_2^se^{2R}ds+c_1c_3^se^{2R}ds+c_1c_2$$
(77)
or
$$R_{ys}=2c_3\left(^se^{2R}ds^se^{2R}ds\right)_s2c_2e^{2R}+c_1c_3e^{2R}.$$
The one soliton solution to the above equation with $`c_1=c_2=0`$ is given by (see also )
$$R(s,y)=\frac{s}{2}+2c_3y+\mathrm{ln}\left(\frac{k_0e^{k_1s+k_2y}+k_1+1}{k_0e^{k_1s+k_2y}k_1+1}\right),$$
(78)
where
$$k_2=8c_3\frac{k_1}{k_1^21}$$
and $`k_0`$, $`k_1`$ are real constants. The corresponding expression for $`\chi `$ is
$$\chi (s,y)=e^f=e^{s4c_3y}\frac{(k_11)^2/(k_1+1)+k_0e^{k_1s+k_2y}}{k_0e^{k_1s+k_2y}+k_1+1}c_3.$$
(79)
The above function $`\chi `$ together with $`R`$ from (78) solve equations (72)-(73). The function $`f`$ defined by equation (79) provides a one-soliton solution to equation (44). It satisfies
$$f_s=\frac{c_3}{\alpha }$$
with
$$\alpha =\frac{\left((k_11)^2/(k_1+1)+k_0e^{k_1s+k_2y}\right)\left(k_0e^{k_1s+k_2y}+k_1+1\right)}{\left(k_0e^{k_1s+k_2y}k_1+1\right)^2}c_3.$$
(80)
Eliminating $`\alpha `$ from equation (72) using (68) we get
$$\left(\mathrm{ln}f_s\right)_{ys}+f_sf_y=4\alpha =\frac{4c_3}{f_s}.$$
Therefore, comparing with (45), we see that $`\kappa =8c_3`$ and $`f`$ given in (79) satisfies (45) and therefore also (44). One-soliton solution $`R(s,y)`$ given by expression (78) satisfies therefore:
$$R_y=\frac{\kappa }{4}^se^{2R}ds^se^{2R}ds.$$
## 5 Bihamiltonian structure
### 5.1 Bihamiltonian structure of the 2-component Camassa-Holm model
As in , we consider the following bihamiltonian structure:
$$\begin{array}{cc}\hfill \{w_1(x),w_1(x^{})\}_1& =\{w_2(x),w_2(x^{})\}_1=0,\hfill \\ \hfill \{w_1(x),w_2(x^{})\}_1& =\delta ^{}(xx^{})\frac{1}{\mu }\delta ^{\prime \prime }(xx^{}).\hfill \\ \hfill \{w_1(x),w_1(x^{})\}_2& =2\delta ^{}(xx^{}),\hfill \\ \hfill \{w_1(x),w_2(x^{}))\}_2& =w_1(x)\delta ^{}(xx^{})+w_1^{}(x)\delta (xx^{}),\hfill \\ \hfill \{w_2(x),w_2(x^{})\}_2& =w_2(x)\delta ^{}(xx^{})+_x\left(w_2(x)\delta (xx^{})\right),\hfill \end{array}$$
(81)
where $`1/\mu `$ now plays a role of the deformation parameter. There exists an hierarchy of Hamiltonians related through Lenard relations
$$\{w_i(x),H_j\}_2=j\{w_i(x),H_{j1}\}_1,j=1,2,3,\mathrm{}.$$
(82)
The flows of the bihamiltonian hierarchy are generated by the Hamiltonians $`H_j,j<0`$ via:
$$\frac{w_i}{t_{j+2}}=\{w_i(x),H_j\}_1,j=3,\mathrm{},i=1,2.$$
(83)
The lower Hamiltonians $`H_j`$ for $`j>1`$ can be obtained recursively from the Casimir $`H_1=w_2(x)𝑑x`$ of the first Poisson bracket applying the Lenard relations (82). Following , we introduce objects $`\phi _1,\phi _2`$ defined by $`w_1=\phi _1\phi _{1,x}/\mu ,w_2=\phi _2+\phi _{2,x}/\mu `$. Then
$$\{\phi _1(x),w_2(x^{})\}_1=\delta ^{}(xx^{}),\{w_1(x),\phi _2(x^{})\}_1=\delta ^{}(xx^{})$$
and the Lenard relations yield
$`H_2`$ $`=`$ $`{\displaystyle [\phi _2\left(\phi _1\phi _{1,x}/\mu \right)]dx}`$
$`H_3`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle [\phi _2^2+\phi _2\phi _1\left(\phi _1\phi _{1,x}/\mu \right)]dx}.`$
Plugging the above $`H_3`$ into equation of motion (83) for $`j=3`$ we obtain:
$$\begin{array}{cc}\hfill (w_1)_t& =\left(\phi _2+\frac{1}{2}\phi _1^2\frac{1}{2\mu }\phi _1\phi _{1,x}\right)_x,\hfill \\ \hfill (w_2)_t& =\left(\phi _1\phi _2+\frac{1}{2\mu }\phi _1\phi _{2,x}\right)_x,\hfill \end{array}$$
(84)
where $`t=t_1`$. Defining $`u`$ such that $`\phi _1=2u`$ and $`\rho `$ such that $`w_2=\rho ^2/\mu ^2+w_1^2/4`$ or $`\rho ^2=w_1^2\mu ^2/4w_2\mu ^2`$ we can rewrite the above system of equations after a transformation $`tt`$ as
$`u_tu_{xxt}/\mu ^2`$ $`=`$ $`\rho \rho _x/\mu ^23uu_x+2u_xu_{xx}/\mu ^2+uu_{xxx}/\mu ^2`$ (85)
$`\rho _t`$ $`=`$ $`(u\rho )_x,`$ (86)
which agrees with the 2-component Camassa-Holm equation. Multiplying equation (85) by $`\mu ^2`$ and taking $`\mu ^20`$ yields
$$u_{xxt}=\rho \rho _x+2u_xu_{xx}+uu_{xxx}$$
(87)
corresponding to eq. (26) with $`\kappa =0`$.
In order to take the $`\mu 0`$ limit of the Poisson structure (81) it is convenient to change the variables from $`w_1`$, $`w_2`$ to $`m`$ and $`\rho `$ defined as
$$\begin{array}{cc}\hfill m& =\frac{1}{2}\mu ^2\left(w_1(x)+w_{1,x}/\mu \right)=\mu ^2uu_{xx}\hfill \\ \hfill \rho ^2& =\mu ^2\left(w_1^2/4w_2\right).\hfill \end{array}$$
(88)
In terms of $`m`$ and $`\rho `$ the Poisson bracket structure (81) turns into:
$$\begin{array}{cc}\hfill \{m(x),m(x^{})\}_1& =0,\hfill \\ \hfill \{\rho ^2(x),\rho ^2(x^{})\}_1& =\mu ^2\left(2m(x)\delta ^{}(xy)+m^{}(x)\delta (xy)\right),\hfill \\ \hfill \{\rho ^2(x),m(x^{})\}_1& =\frac{1}{2}\mu ^2\left(\mu ^2\delta ^{}(xx^{})+\delta ^{\prime \prime \prime }(xx^{})\right),\hfill \\ \hfill \{m(x),m(x^{})\}_2& =\frac{1}{2}\mu ^2\left(\mu ^2\delta ^{}(xx^{})\delta ^{\prime \prime \prime }(xx^{})\right),\hfill \\ \hfill \{\rho (x),\rho (x^{}))\}_2& =\frac{1}{2}\mu ^2\delta ^{}(xx^{}),\hfill \\ \hfill \{\rho (x),m(x^{})\}_2& =0,\hfill \end{array}$$
(89)
### 5.2 The $`\mu 0`$ limit and the Dym type hierarchy.
Redefining the brackets as follows
$$\{,\}_j\mu ^2\{,\}_j$$
and taking $`\mu 0`$ limit in equation (89) we find for the first and second bracket structure in terms of $`u`$ and $`\rho `$ (see also ):
$$\begin{array}{cc}\hfill \{u(x),u(x^{})\}_1& =0,\hfill \\ \hfill \{\rho ^2(x),\rho ^2(x^{})\}_1& =2u^{\prime \prime }(x^{})\delta ^{}(xx^{})u^{\prime \prime \prime }(x^{})\delta (xx^{})\hfill \\ \hfill \{\rho ^2(x),u(x^{})\}_1& =\frac{1}{2}\delta ^{}(xx^{}),\hfill \\ \hfill \{u(x),u(x^{})\}_2& =\frac{1}{2}_x^1\delta (xx^{}),\hfill \\ \hfill \{\rho (x),\rho (x^{})\}_2& =\frac{1}{2}_x\delta (xx^{}),\hfill \\ \hfill \{u(x),\rho (x^{})\}_2& =0,\hfill \end{array}$$
(90)
The first bracket in (90) has the Casimir:
$$H_1^{(1)}=[\rho ^2(x)u_x^2(x)]dx.$$
This Casimir leads via Lenard relation (82) to the Hamiltonian:
$$H_2=2(\rho ^2u_x^2)udx,$$
which in turn generates equations of motion of $`\mu =0`$ case via equations (83) and (82):
$$\begin{array}{cc}\hfill \frac{u_x}{t}& =\frac{1}{2}\{u_x(x),H_2\}_2=\frac{1}{2}(u_x)^2uu_{xx}\frac{1}{2}\rho ^2\hfill \\ \hfill \frac{\rho }{t}& =\frac{1}{2}\{\rho (x),H_2\}_2=\left(u\rho \right)_x\hfill \end{array}$$
(91)
This Hamiltonian structure can be extended by an additional term:
$$\overline{H}_2=2\kappa u^2dx.$$
Adding this term to $`H_2`$ will lead via relations (91) to correct equations of motion (29)-(30).
### 5.3 Hamiltonians of positive order
There exists another class of conserved charges, different from the chain of Hamiltonians $`H_j,j=1,2,\mathrm{}`$ of negative order discussed above. These are the Hamiltonians of positive order originating from the Casimir
$$H_1^{(2)}=2\rho (x)dx$$
(92)
of the second Poisson bracket (89). We now employ Lenard relations (82) to construct higher order Hamiltonians. The first recurrence step:
$$\{,\rho (x)dx\}_1=\{,H_0\}_2,$$
where “$``$” stands for phase space variables $`m(x)`$ and $`\rho (x)`$, leads to a new Hamiltonian:
$$H_0=\frac{m(x)}{\rho (x)}dx.$$
(93)
in agreement with expression found in . The integrand of $`H_0`$ can be rewritten as
$$\frac{m}{\rho }=\rho \frac{m}{\rho ^2}=\rho P=\rho f_y=f_x$$
and therefore $`H_0`$ appears to be a surface term that would vanish if $`f`$ would be a local field.
On the next level we find from the Lenard relations:
$$\{,H_0\}_1=\{,H_1\}_2,$$
with
$$H_1=\left[\frac{1}{4\rho ^3}\left(\rho _x^2m^2\right)+\frac{\mu ^2}{4\rho }\right]dx$$
(94)
This recurrence process can be continued to yield higher order Hamiltonians. Technical calculations involved in obtaining higher order Hamiltonians become increasingly tedious. Remarkably, we can bypass these difficulties by relying on the underlying AKNS structure governing higher positive flows. We recall the Hamiltonian densities $`_n`$ (8) of the AKNS model generating the positive flows of the model. Their conservation law with respect to the negative flow $`s`$ takes a form
$$\left(_n\right)_s=X_y,$$
(95)
where $`s`$ and $`y`$ are “reciprocal” variables describing time and space of the AKNS model and $`X`$ is some local quantity. The above relation ensures that $`(_ndy)_s=0`$ (for $`X`$ local in $`u`$ and $`\rho `$) and thus the integral $`_ndy`$ is conserved.
In terms of the original $`t,x`$ variables the conservation laws (95) read
$$\left(\frac{}{t}+\rho u\frac{}{y}\right)_n=\frac{}{y}X$$
or
$$\frac{}{t}_n=\rho u\frac{}{y}_n+\frac{}{y}X=u\frac{}{x}_n+\frac{1}{\rho }\frac{}{x}X,$$
where we used that $`/y=\rho /x`$. It follows that
$$\frac{}{t}\left(\rho _n\right)=(u\rho )_x_nu\rho _{nx}+X_x=\frac{}{x}\left(Xu\rho H_n\right).$$
(96)
Thus, the quantities $`\rho _n`$ are conserved charges of the 2-component Camassa-Holm and 2-component Dym type models.
The first two Hamiltonian densities $`rq`$ and $`rq_y`$ of the AKNS model, given by relations (9) and (10), give rise, after use of definitions (47),(36) and (40), to the following conserved charges:
$$\begin{array}{cc}\hfill H_1& =\rho _1dx=\rho rqdx=\left[\frac{1}{4\rho ^3}\left(\rho _x^2m^2\right)+\frac{\mu ^2}{4\rho }\right]dx,\hfill \\ \hfill H_2& =\frac{1}{2}\rho _2dx=\frac{1}{2}\rho rq_ydx=\frac{1}{2}rq_xdx\hfill \\ & =\frac{m}{2\rho ^2}\left[\frac{\mu ^2}{4\rho }\frac{3}{4}\frac{\rho _x^2}{\rho ^3}+\frac{\rho _{xx}}{2\rho ^2}\frac{m^2}{4\rho ^3}\right]dx,\hfill \end{array}$$
(97)
which we have verified explicitly to be conserved under equations of motion (19)-(20) for $`m=\mu ^2uu_{xx}`$ for $`\mu 0`$ and $`m=u_{xx}+\kappa /2`$ for $`\mu =0`$.
We recognize in $`\rho _1`$ the Hamiltonian $`H_1`$ derived in (94) from the Casimir of the second bracket via Lenard recursion relations. Furthermore, we have shown that $`H_1=\rho _1dx`$ and $`H_2=\rho _2dx/2`$ also are interrelated via the Lenard relation:
$$\{,H_1\}_1=\{,H_2\}_2.$$
Therefore, the conclusion is that the AKNS induced Hamiltonians $`\rho _n`$ form the sequence of positive order Hamiltonians of the 2-component Camassa-Holm and 2-component Dym type hierarchies. The formula (8) given in section 2 can be used to systematically derive all the Hamiltonians governing positive flows of this model.
## Acknowledgments
We thank A. Das and L.A. Ferreira for helpful discussions. H.A. acknowledges support from Fapesp and thanks the IFT-UNESP Institute for its hospitality. JFG and AHZ thank CNPq for a partial support. |
warning/0507/hep-th0507010.html | ar5iv | text | # Scalar-Induced Compactifications in Higher Dimensional Supergravities
## 1 Introduction
A central issue in all higher-dimensional theories is compactification. Various mechanisms have been proposed and the general idea is to construct vacuum spacetimes of $`M^4\times X`$ geometry, where $`M^4`$ is 4D Minkowski spacetime and $`X`$ is a compact internal space. In string theory for example, one way of constructing appropriate vacua is to look for classical supergravity solutions. As supergravity is the low-energy limit of string theory, supergravity solutions describe accordingly low-energy string vacua. There will be $`\alpha ^{}`$–corrections as well as string-loop corrections to these solutions but, nevertheless, these solutions will still be valid in some appropriate limits. These vacua are constructed by solving the classical field equations with appropriate fields turned on. Usually such fields are antisymmetric $`p`$–forms as well as various scalars like the dilaton, axion etc., which appear in almost all supergravity theories. The four-dimensional Plank mass $`M_P`$ is proportional to the volume $`V(X)`$ of $`X`$
$$M_P^2=M_s^8V(X),$$
(1)
where $`M_s^21/\alpha ^{}`$ is the string-mass scale. Propagating gravity therefore exists in four dimensions if the volume of $`X`$ is finite. This is always the case for a smooth compact space $`X`$.
It should be stressed, however, that non-compact spaces may also be employed. Such spaces have been considered in the Kaluza-Klein programme ,,. Adopting the proposal of a non-compact internal space, we are facing a new problem. A smooth non-compact space has infinite volume so that the four-dimensional Planck mass $`M_P^2`$ will be infinite. As a result, gravitational interactions will be actually higher-dimensional and not four-dimensional as we want. The solution here is to assume that the non-compact space has finite volume. In this case we expect singularities and several pathologies like continuous spectra, violation of conservation laws for energy, momentum, angular momentum etc. caused by possible leakage from the singular points. Thus, in order for our proposal to be viable, all these pathologies should be avoided.
This is the case, for example, in the tear-drop solution ,. In this solution, the scalars of the type IIB supergravity are non-vanishing and the 10D spacetime is compactified to a space diffeomorphic to the $`\frac{SU(1,1)}{U(1)}`$ scalar manifold. This compactification is triggered by a non-trivial scalar field configuration. The spacetime develops a naked singularity which, however, is harmless and does not lead to any physically unacceptable situation as all physical quantities, like energy, momentum and angular momentum are conserved. It should be noted that a similar solution is also the stringy cosmic string where, in addition, the non-perturbative $`SL(2,)`$ symmetry of type IIB is employed for the compactification .
Here, we will show that tear-drop-like solutions are actually quite generic, using $`D=9,8,7`$ minimal supergravities as a concrete example. The supergravities under consideration contain numerous scalar fields, combinations of which may trigger compactifications of the sort described in the preceding paragraph. Adopting a convenient parameterization of the scalar sector of these theories in terms of solvable Lie algebras, we can write the scalar Lagrangian and the equations of motion of the theory in a compact form. We can then identify solutions of the equations of motion, in which the non-vanishing scalars form $`\frac{SL(2,)}{U(1)}`$ submanifolds of the original scalar manifold, inducing tear-drop-like compactifications. To be specific, there are singular and non-singular solutions. The former are like the tear-drop ,, while the latter are like the stringy cosmic string . Moreover, in the case of $`D=8`$ supergravity, there is also a compactification to 4D, achieved by two vector multiplets coupled to the supergravity multiplet. The compactifications found preserve half the supersymmetries of the original theory in all cases.
In the following section 2, in order to set up our notation, we introduce an appropriate parametrization for the $``$ spaces. In section 3, we briefly review the $`D=9,8,7`$ minimal supergravities and, after deriving the field equations with scalar fields turned on, we present their solutions. In section 4, we discuss the supersymmetric properties of the solutions. Finally, in section 5, we comment on our findings.
## 2 Scalar Coset Manifolds in Minimal Supergravities
The scalar fields of minimal supergravities coupled to matter in $`D=9,8,7,5,4`$ (apart from the dilaton contained in the supergravity multiplet), parameterize non-compact coset manifolds of $`\frac{SO(10D,n)}{SO(10D)\times SO(n)}`$ type<sup>1</sup><sup>1</sup>1In the special case $`D=5`$, there are actually more possibilities for the scalar manifold.. To see how these manifolds arise, we first note that in these theories the supergravity multiplet contains $`10D`$ vector fields while the vector multiplet contains $`10D`$ scalars. These two types of fields carry indices of the R-symmetry group of the supersymmetry algebra which, for the specific dimensions, is isomorphic to $`SO(10D)`$. By combining the supergravity multiplet with $`n`$ vector multiplets, the total $`10D+n`$ vectors fall into the defining representation of $`SO(10D,n)`$, so that the latter is identified as a global symmetry group of the theory. On the other hand, the theory describing the $`n(10D)`$ real scalars from the vector multiplets is invariant under $`SO(n)`$ rotations between the scalars of the same R-symmetry index in different multiplets as well as under $`SO(10D)`$ R-symmetry rotations within each vector multiplet. Therefore, the scalar manifold, i.e. the space of inequivalent points parameterized by the scalars, is the coset space $`=\frac{SO(10D,n)}{SO(10D)\times SO(n)}`$. The scalar isometry group $`SO(10D,n)`$ is a non-compact real form of $`D_{\mathrm{}}`$ (for $`10D+n=2\mathrm{}`$) or of $`B_{\mathrm{}}`$ (for $`10D+n=2\mathrm{}+1`$).
Here, we will first review the parameterizations of scalar cosets in both the coset-manifold and the group-manifold approaches. Using the second approach, we will then give the general form of the scalar Lagrangian. Finally, we will specialize to the $`\frac{SO(10D,n)}{SO(10D)\times SO(n)}`$ coset spaces for the dimensions $`D=9,8,7`$ and we will discuss the further simplifications that occur in the structure of the scalar Lagrangian.
### 2.1 The coset and the group manifold approaches
Consider a general non-compact coset manifold $`=G/H`$, where $`G`$ is some group associated with a non-compact real form $`𝔤_{nc}`$ of a complex Lie algebra $`𝔤`$ and $`H`$ is its subgroup associated with the maximal compact subalgebra $`𝔥`$ of $`𝔤_{nc}`$. Such a manifold admits two different descriptions, which we will outline below.
The first is the usual coset-manifold description, based on the Cartan decomposition $`𝔤_{nc}=𝔥𝔨`$, where $`𝔨`$ is the non-compact complementary subspace of $`𝔥`$ within $`𝔤_{nc}`$; this subspace is not an algebra. In this description, a coset representative is defined as $`L_K=\text{Exp}(k)`$ with $`k𝔨`$. Since $`𝔨`$ is not an algebra, $`L_K`$ is not a group element but, in general, includes an $`H`$–valued part. This description is the usual one employed in the construction of supergravity theories. In particular, the decomposition of the Maurer-Cartan form of a coset representative into an $`𝔥`$–valued and a $`𝔨`$–valued part gives the composite connections and the composite coset vielbeins respectively in a form that may be directly used in the supersymmetry transformation laws. However, the explicit form of the scalar Lagrangian in this description is quite complicated.
The second description is the group-manifold description, which is based on solvable Lie algebras. This description is based on the Iwasawa decomposition which ensures that, for any non-compact real form $`𝔤_{nc}`$ of $`𝔤`$, there exists a solvable Lie algebra $`\mathrm{Solv}(𝔤_{nc})`$ such that $`𝔤_{nc}`$ may be decomposed as the direct sum $`𝔤_{nc}=𝔥\mathrm{Solv}(𝔤_{nc})`$. The solvable Lie algebra is constructed as follows. We first consider a Cartan-Weyl basis for the generators of $`𝔤`$, denoting the $`\mathrm{}`$ generators of the Cartan subalgebra $`𝔞`$ by $`\{H_I\}`$ and the positive-root generators by $`\{E_A\}`$; the set of the positive roots is denoted as $`\mathrm{\Phi }^+`$. The solvable Lie algebra is then given by the direct sum
$$\mathrm{Solv}(𝔤_{nc})=𝔞_{nc}𝔫.$$
(2)
Here $`𝔞_{nc}`$ is the non-compact part of the Cartan subalgebra $`𝔞`$ of $`𝔤`$,
$$𝔞_{nc}=𝔞𝔨,$$
(3)
generated by an appropriate subset $`\{H_i\}`$ of the Cartan generators and $`𝔫`$ is the algebra which is constructed from the set $`\{E_\alpha \}`$ of the positive-root generators of $`𝔤`$ that do not commute with all $`H_i`$’s according to the relation
$$𝔫=(\underset{\alpha \mathrm{\Delta }^+}{}E_\alpha )𝔤_{nc},$$
(4)
where $`\mathrm{\Delta }^+`$ is the subset of $`\mathrm{\Phi }^+`$ containing the positive roots associated with $`\{E_\alpha \}`$. The intersection symbol in (4) denotes that the $`\{E_\alpha \}`$ should be arranged in suitable linear combinations that belong to the non-compact real form $`𝔤_{nc}`$. We note that, in the special case where $`G`$ is a split group (i.e. where $`=G/H`$ is maximally non-compact), the solvable Lie algebra of $`𝔤_{nc}`$ coincides with its Borel subalgebra generated by all $`H_I`$’s and all $`E_A`$’s; in the case of $`G=SO(10D,n)`$, this happens only when $`n=10D`$ or $`n=10D\pm 1`$. In the group-manifold description, a coset representative is defined as $`L_S=\text{Exp}(s)`$ with $`s\mathrm{Solv}(𝔤_{nc})`$ and, unlike the case in the coset-manifold description, it *is* an element of $`\mathrm{Solv}(𝔤_{nc})`$, since the latter is an algebra (for a given point $`P`$, the representatives $`L_S(P)`$ and $`L_K(P)`$ are equivalent up to a right-multiplication by an element of $`H`$). Specifically, the Iwasawa decomposition implies that a general element $`𝐠G`$ can be uniquely decomposed as
$$𝐠=\text{Exp}(h)\text{Exp}(a)\text{Exp}(n);h𝔥,a𝔞_{nc},n𝔫.$$
(5)
As a result, $`𝐠`$ can be written as the product of elements obtained by exponentiation of the maximal compact subalgebra $`𝔥`$, the non-compact Cartan subalgebra $`𝔞_{nc}`$ and the subalgebra $`𝔫`$ associated with the positive roots in $`\mathrm{\Delta }^+`$. According to this decomposition, a coset representative is obtained from (5) by discarding the $`H`$–valued factor and is thus given by
$$L_S=\text{Exp}(a)\text{Exp}(n);a𝔞_{nc},n𝔫.$$
(6)
The advantage of the group-manifold description of the scalar coset in supergravity theories is that, unlike the coset-manifold description, it leads to a natural one-to-one correspondence between the scalar fields of the theory with the generators $`\{H_i\}`$ and $`\{E_\alpha \}`$ which form an algebra. For that reason, the explicit form of the scalar coset Lagrangian simplifies considerably through the use of group-theoretical methods. The application of solvable Lie algebras in supergravity has been extensively studied in ,,,,.
### 2.2 The scalar Lagrangian
Here, we will review the construction of scalar coset sigma-model Lagrangians using the group-manifold approach; the detailed procedure can be found e.g. in ,,,. According to the remarks of the previous paragraph, one may use the Iwasawa decomposition (6) to parameterize a coset representative. In the context of a sigma model, the Lie-algebra-valued quantities $`a𝔞_{nc}`$ and $`n𝔫`$ are taken to be functions of the spacetime coordinates $`x^M`$. They can be expressed in terms of the fields $`\{\varphi ^i(x)\}`$ (dilatons, corresponding to $`\{H_i\}`$) and $`\{\chi ^\alpha (x)\}`$ (axions, corresponding to $`\{E_\alpha \}`$) as follows
$$a(x)=\frac{1}{2}\varphi ^i(x)H_i,n(x)=\chi ^\alpha (x)E_\alpha .$$
(7)
So, the Iwasawa decomposition reads
$$L=e^{\frac{1}{2}\varphi ^iH_i}e^{\chi ^\alpha E_\alpha }.$$
(8)
We stress again that the root-space generators $`\{E_\alpha \}`$ are in fact restricted to enter (7) and (8) only through the appropriate linear combinations that belong to the chosen real form of the isometry algebra. This does not alter at all the discussion that follows.
The Lagrangian of the sigma model coupled to gravity is expressed in terms of $`L`$ by
$$e^1_s=\frac{1}{4}\text{Tr}\left[(_MLL^1)(^MLL^1)^\mathrm{\#}+(_MLL^1)(^MLL^1)\right],$$
(9)
where, “$`\mathrm{\#}`$” denotes the generalized transpose, defined as the transformation induced by the Cartan involution on the group elements. It is easily shown that the Maurer-Cartan form appearing in (9) has the explicit form
$$_MLL^1=\frac{1}{2}_M\varphi ^iH_i+e^{\frac{1}{2}\alpha _i\varphi ^i}F_M^\alpha E_\alpha ,$$
(10)
with $`\alpha _i`$ the $`i`$-the component of the root $`\alpha `$. Also, $`F_M^\alpha `$ is the field strength associated with $`\chi ^\alpha `$, given by
$$F_M^\alpha =_M\chi ^\alpha +\frac{1}{2!}(\chi ^\gamma C_{\gamma \beta }^\alpha )_M\chi ^\beta +\frac{1}{3!}(\chi ^\gamma C_{\gamma \epsilon }^\alpha )(\chi ^\delta C_{\delta \beta }^\epsilon )_M\chi ^\beta +\mathrm{}$$
(11)
where $`C_{\alpha \beta }^\gamma `$ are the structure constants in $`[E_\alpha ,E_\beta ]=C_{\alpha \beta }^\gamma E_\gamma `$, i.e $`C_{\alpha \beta }^\gamma =N_{\alpha ,\beta }`$ if $`\alpha +\beta =\gamma `$ and zero otherwise. Using the parameterization (10), one can show that the scalar Lagrangian takes the simple form
$$e^1_s=\frac{1}{4}\underset{i}{}(_M\varphi ^i)^2\frac{1}{2}\underset{\alpha }{}e^{\alpha _i\varphi ^i}(F_M^\alpha )^2,$$
(12)
where we employed a normalization with $`\text{Tr}H_iH_j=2\delta _{ij}`$ and $`\text{Tr}E_\alpha E_\alpha =2`$.
### 2.3 The $`\frac{SO(10D,n)}{SO(10D)\times SO(n)}`$ scalar coset
After this general discussion, let us specialize to the $`\frac{SO(10D,n)}{SO(10D)\times SO(n)}`$ case of interest; here we will always assume that $`n10D`$. To construct the $`D_{\mathrm{}}`$ or $`B_{\mathrm{}}`$ root vectors associated with $`SO(10D,n)`$, we define $`ϵ_i`$ as the $`\mathrm{}`$–dimensional vector whose $`i`$–th element is unity with all other elements zero. Then, the positive roots of $`D_{\mathrm{}}`$ or $`B_{\mathrm{}}`$ are given by
$`ϵ_i\pm ϵ_j`$ ; $`i<j=1,\mathrm{}\mathrm{},`$
$`ϵ_i`$ ; $`i=1,\mathrm{}\mathrm{}\text{(only for }B_{\mathrm{}}\text{)}.`$ (13)
and so the associated generators are $`E_{ϵ_i\pm ϵ_j}`$ and $`E_{ϵ_i}`$.
To construct the solvable Lie algebra, we choose our conventions so that the non-compact Cartan generators $`\{H_i\}`$ of $`𝔞_{nc}`$ are given by $`\{H_{ϵ_i}\}`$, $`i=1,\mathrm{},10D`$. The generators of the nilpotent subalgebra $`𝔫`$ are then found by considering the subset of the generators $`\{E_{ϵ_i\pm ϵ_j},E_{ϵ_i}\}`$ that do not commute with all of the $`H_i`$’s and taking appropriate linear combinations that belong to the non-compact real form of $`D_{\mathrm{}}`$ or $`B_{\mathrm{}}`$ appropriate for the space of interest. Explicit constructions will be shown below.
In what follows, we will further specialize to the cases of interest, namely $`D=9,8,7`$. In particular, we will apply the above procedure to construct the solvable Lie algebra and then, examining its structure, we will see how the scalar Lagrangian simplifies.
* $`D=9`$. In the $`D=9`$ case, corresponding to $`\frac{SO(1,n)}{SO(n)}`$, there is only one non-compact Cartan generator which, in our conventions is taken to be
$$H_{ϵ_1}.$$
(14)
The generators of the solvable algebra are then given by the positive-root generators , which already belong to the $`𝔰𝔬(1,n)`$ real form
$`E_{ϵ_1+ϵ_i},E_{ϵ_1ϵ_i};i=2,\mathrm{}\mathrm{}`$
$`E_{ϵ_1},\text{(only for }B_{\mathrm{}}\text{)},`$ (15)
In this case, the structure of the scalar Lagrangian is very simple. Indeed, all structure constants associated with positive-root generators vanish,
$$C_{\beta \gamma }^\alpha =0,$$
(16)
and so the field strengths (11) are simply $`F_M^\alpha =_M\chi ^\alpha `$.
* $`D=8`$. In the $`D=8`$ case, the scalar coset manifold is $`\frac{SO(2,n)}{SO(2)\times SO(n)}`$, with $`n2`$. There are two non-compact Cartan generators which can be chosen as
$$H_{ϵ_1},H_{ϵ_2}.$$
(17)
The generators of the solvable algebra are then given by the positive-root generators
$`E_{ϵ_1+ϵ_2},E_{ϵ_1ϵ_2},`$
$`E_{ϵ_1},E_{ϵ_2},\text{(only for }B_{\mathrm{}}\text{)},`$ (18)
that are already in the $`𝔰𝔬(2,n)`$ real form, plus the linear combinations ,
$`{\displaystyle \frac{1}{\sqrt{2}}}(E_{ϵ_1+ϵ_i}+E_{ϵ_1ϵ_i}),{\displaystyle \frac{\text{i}}{\sqrt{2}}}(E_{ϵ_1+ϵ_i}E_{ϵ_1ϵ_i});i=3,\mathrm{}\mathrm{},`$
$`{\displaystyle \frac{1}{\sqrt{2}}}(E_{ϵ_2+ϵ_i}+E_{ϵ_2ϵ_i}),{\displaystyle \frac{\text{i}}{\sqrt{2}}}(E_{ϵ_2+ϵ_i}E_{ϵ_2ϵ_i});i=3,\mathrm{}\mathrm{}.`$ (19)
To see the simplifications that occur regarding the coset Lagrangian, we are finding that the only nonzero structure constants $`C_{\beta \gamma }^\alpha `$(11) are given by
$`C_{ϵ_1ϵ_2,ϵ_2\pm ϵ_i}^{ϵ_1\pm ϵ_i},C_{ϵ_1\pm ϵ_i,ϵ_2ϵ_i}^{ϵ_1+ϵ_2}`$
$`C_{ϵ_1,ϵ_2}^{ϵ_1+ϵ_2},C_{ϵ_1ϵ_2,ϵ_2}^{ϵ_1}\text{(only for }B_{\mathrm{}}\text{)}.`$ (20)
It follows then that the only nonzero terms of Eq. (11) quadratic in the structure constants involve the combinations
$`C_{ϵ_2\pm ϵ_i,ϵ_1ϵ_i}^{ϵ_1+ϵ_2}C_{ϵ_1ϵ_2,ϵ_2ϵ_i}^{ϵ_1ϵ_i},C_{ϵ_2ϵ_i,ϵ_1\pm ϵ_i}^{ϵ_1+ϵ_2}C_{ϵ_2\pm ϵ_i,ϵ_1ϵ_2}^{ϵ_1\pm ϵ_i},`$
$`C_{ϵ_2,ϵ_1}^{ϵ_1+ϵ_2}C_{ϵ_1ϵ_2,ϵ_2}^{ϵ_1},C_{ϵ_2,ϵ_1}^{ϵ_1+ϵ_2}C_{ϵ_2,ϵ_1ϵ_2}^{ϵ_1}\text{(only for }B_{\mathrm{}}\text{)},`$ (21)
and that terms of cubic or higher order vanish since there are no available indices to contract; this implies that, in this case, Eq. (11) actually contains no further terms. The important fact that we will need later on is that there are no nonzero structure constants with $`ϵ_1+ϵ_2`$ as a lower index.
* $`D=7`$. In the $`D=7`$ case, the scalar coset manifold is $`\frac{SO(3,n)}{SO(3)\times SO(n)}`$, with $`n3`$. There are three non-compact Cartan generators which can be chosen as
$$H_{ϵ_1},H_{ϵ_2},H_{ϵ_3}.$$
(22)
In a similar manner as before, the solvable generators are given by the positive-root generators
$`E_{ϵ_1+ϵ_2},E_{ϵ_1ϵ_2},E_{ϵ_1+ϵ_3},E_{ϵ_1ϵ_3},E_{ϵ_2+ϵ_3},E_{ϵ_2ϵ_3},`$
$`E_{ϵ_1},E_{ϵ_2},E_{ϵ_3},\text{(only for }B_{\mathrm{}}\text{)},`$ (23)
and the combinations
$`{\displaystyle \frac{1}{\sqrt{2}}}(E_{ϵ_1+ϵ_i}+E_{ϵ_1ϵ_i}),{\displaystyle \frac{\text{i}}{\sqrt{2}}}(E_{ϵ_1+ϵ_i}E_{ϵ_1ϵ_i});`$ $`i=4,\mathrm{}\mathrm{},`$ (24)
$`{\displaystyle \frac{1}{\sqrt{2}}}(E_{ϵ_2+ϵ_i}+E_{ϵ_2ϵ_i}),{\displaystyle \frac{\text{i}}{\sqrt{2}}}(E_{ϵ_2+ϵ_i}E_{ϵ_2ϵ_i});`$ $`i=4,\mathrm{}\mathrm{},`$
$`{\displaystyle \frac{1}{\sqrt{2}}}(E_{ϵ_3+ϵ_i}+E_{ϵ_3ϵ_i}),{\displaystyle \frac{\text{i}}{\sqrt{2}}}(E_{ϵ_3+ϵ_i}E_{ϵ_3ϵ_i});`$ $`i=4,\mathrm{}\mathrm{}.`$
In this case, the structure of the Lagrangian is more complicated since there exist structure-constant combinations of cubic and quartic (in the $`D_{\mathrm{}}`$ case) order. As in the previous case, no nonzero structure constants with an $`ϵ_1+ϵ_2`$ lower index arise.
## 3 Scalar-induced compactifications in $`D=9,8,7`$ minimal supergravities
As stated in the intruduction, the mechanism of scalar-induced compactification admits generalizations to supergravities of diverse dimensions. In this section, we demonstrate the existence of such solutions for the case of the minimal supergravities in $`D=9,8,7`$. We start by discussing the basic aspects of these minimal supergravities. Next, we use the parameterization of the previous section to write down the field equations for a vacuum configuration containing only gravity and scalars. Finally, we show that these equations are consistent with a particular ansatz for the scalars and we present the resulting solutions. Having described the basic aspects of the scalar coset manifolds, w will describe next the general aspects of the $`D=9,8,7`$ supergravities of interest.
### 3.1 Minimal supergravities in $`D=9,8,7`$
#### 3.1.1 $`D=9`$ supergravity
The field content of the massless representations of the $`D=9`$, $`𝒩=2`$ supersymmetry algebra consists of the following multiplets
Supergravity multiplet $`:`$ $`(g_{MN},B_{MN},A_M,\sigma ,\psi _M,\chi ),`$
Vector multiplet $`:`$ $`(A_M,\phi ,\lambda ).`$ (25)
where all spinors are pseudoMajorana. A general $`D=9`$, $`𝒩=2`$ supergravity theory is constructed by combining the supergravity multiplet with $`n`$ vector multiplets. This leads to the reducible multiplet
$$(g_{MN},B_{MN},A_M^I,\phi ^{\overline{\alpha }},\sigma ,\psi _M,\chi ,\lambda ^{\overline{a}}).$$
(26)
where $`\overline{\alpha }=1,\mathrm{},n`$ labels the scalars, $`\overline{a}=1,\mathrm{},n`$ labels the gauginos and $`I=1,\mathrm{},n+1`$ labels the vectors (here, we employ barred indices, e.g. $`\overline{\alpha }`$, in order to avoid confusion with the indices appearing in section 2).
As mentioned before, the $`n`$ scalars $`\phi ^{\overline{\alpha }}`$ parameterize the non-compact coset manifold $`\frac{SO(1,n)}{SO(n)}`$ (more precisely, they parameterize the space $`H^n=\frac{SO_0(1,n)}{SO(n)}`$, which is the upper sheet of a hyperboloid). To parameterize the scalar coset, one introduces a coset representative $`L=\{L_I^A\}`$ given by $`(n+1)\times (n+1)`$ matrix in the vector representation<sup>2</sup><sup>2</sup>2In the coset-manifold approach, the usual choice is to combine the $`n`$ scalars into a column vector $`\mathrm{\Phi }`$ and define $`L=\mathrm{exp}\left(\begin{array}{cc}0& \mathrm{\Phi }\\ \mathrm{\Phi }^T& 0\end{array}\right)`$. (here, $`A`$ is the curved index analog to $`I`$). The inverse matrix, given by $`L^1=\{L_A^I\}`$, satisfies
$$L_A^IL_I^B=\delta _A^B.$$
(27)
The elements of $`L`$ and its inverse can be decomposed as
$$L_I^A=(L_I,L_I^{\overline{a}}),L_A^I=(L^I,L_{\overline{a}}^I).$$
(28)
Also, $`L`$ satisfies the $`SO(1,n)`$ orthogonality condition
$$\eta _{AB}L_I^AL_J^B=\eta _{IJ},$$
(29)
where $`\eta _{AB}=\eta _{IJ}=\text{diag}(1,+1,\mathrm{},+1)`$ is the $`SO(1,n)`$ invariant tensor. A quantity of interest constructed out of $`L`$ is the tensor
$$a_{IJ}=L_IL_J+L_I^{\overline{a}}L_J^{\overline{a}},$$
(30)
which contracts $`I`$ and $`J`$ so as to yield a ghost-free kinetic term for the vectors.
To construct scalar kinetic terms and covariant derivatives in the coset-manifold approach, one considers the Maurer-Cartan form<sup>3</sup><sup>3</sup>3Note that, due to the particular parameterization of the coset, we have to use right-invariant Maurer-Cartan forms instead of the left-invariant ones usually employed in the supergravity literature. of $`L`$, given by $`_MLL^1`$. This matrix-valued one-form decomposes into the coset vielbein $`P_M^{\overline{a}}`$ and the $`SO(n)`$ composite connection $`Q_{M\overline{a}}^{\overline{b}}`$. In the standard vector representation of $`SO(1,n)`$, this decomposition has the form
$$_MLL^1=\left(\begin{array}{cc}0& P_M^{\overline{a}}\\ P_{M\overline{a}}& Q_{M\overline{a}}^{\overline{b}}\end{array}\right),$$
(31)
The Lagrangian of the theory was first constructed in . Its bosonic part is given by
$`e^1`$ $`=`$ $`{\displaystyle \frac{1}{2}}R{\displaystyle \frac{1}{12}}e^{2\sigma }G_{MNP}^2{\displaystyle \frac{1}{4}}e^\sigma a_{IJ}F_{MN}^IF^{JMN}{\displaystyle \frac{7}{16}}(_M\sigma )^2{\displaystyle \frac{1}{4}}P_M^{\overline{a}}P_{\overline{a}}^M`$ (32)
where the field strengths $`G_{MNP}`$ and $`F_{MN}^I`$ are defined as
$$G_{MNP}=3\left(_{[M}B_{NP]}+\eta _{IJ}F_{[MN}^IA_{P]}^J\right),F_{MN}^I=2_{[M}A_{N]}^I.$$
(33)
#### 3.1.2 $`D=8`$ supergravity
The massless representations of the $`D=8`$, $`𝒩=1`$ supersymmetry algebra form the following multiplets
Supergravity multiplet $`:`$ $`(g_{MN},B_{MN},A_M^{\overline{i}},\sigma ,\psi _M,\chi ),`$
Vector multiplet $`:`$ $`(A_M,\phi ^{\overline{i}},\lambda ).`$ (34)
where all spinors are pseudoMajorana and the index $`\overline{i}=1,2`$ refers to the $`SO(2)`$ R-symmetry group. A general $`D=8`$, $`𝒩=1`$ supergravity theory is constructed by combining the supergravity multiplet with $`n`$ vector multiplets. This leads to the reducible multiplet
$$(g_{MN},B_{MN},A_M^I,\phi ^{\overline{\alpha }},\sigma ,\psi _M,\chi ,\lambda ^{\overline{a}}).$$
(35)
where $`\overline{\alpha }=1,\mathrm{},2n`$ labels the scalars, $`\overline{a}=1,\mathrm{},n`$ labels the gauginos and $`I=1,\mathrm{},n+2`$ labels the vectors.
The $`2n`$ scalars $`\phi ^{\overline{\alpha }}`$ parameterize the non-compact coset manifold $`\frac{SO(2,n)}{SO(2)\times SO(n)}`$. The coset representative $`L`$ and its inverse, now given by $`(n+2)\times (n+2)`$ matrices, are defined in a similar manner as before, they can be decomposed as
$$L_I^A=(L_I^{\overline{i}},L_I^{\overline{a}}),L_A^I=(L_{\overline{i}}^I,L_{\overline{a}}^I).$$
(36)
They satisfy the orthogonality condition
$$\eta _{AB}L_I^AL_J^B=\eta _{IJ},$$
(37)
where $`\eta _{AB}=\eta _{IJ}=\text{diag}(1,1,+1,\mathrm{},+1)`$ is the $`SO(2,n)`$ invariant tensor. The tensor needed for the contraction of $`I`$ and $`J`$ in the vector kinetic term is now
$$a_{IJ}=L_I^{\overline{i}}L_J^{\overline{i}}+L_I^{\overline{a}}L_J^{\overline{a}}$$
(38)
To proceed, we consider the Maurer-Cartan form of $`L`$, which contains the coset vielbein $`P_{M\overline{i}}^{\overline{a}}`$ and the $`SO(2)`$ and $`SO(n)`$ composite connections $`Q_{M\overline{i}}^{\overline{j}}`$ and $`Q_{M\overline{a}}^{\overline{b}}`$. In the vector representation, the decomposition has the form
$$_MLL^1=\left(\begin{array}{cc}Q_{M\overline{i}}^{\overline{j}}& P_{M\overline{i}}^{\overline{a}}\\ P_{M\overline{a}}^{\overline{i}}& Q_{M\overline{a}}^{\overline{b}}\end{array}\right),$$
(39)
For later convenience, it is also useful to define the quantities
$$\widehat{P}_{M\overline{a}}P_{M\overline{a}}^1+i\gamma _{(9)}P_{M\overline{a}}^2,Q_MQ_{M1}^2.$$
(40)
The Lagrangian of the theory was first constructed in . Its bosonic part is given by
$$e^1=\frac{1}{2}R\frac{1}{12}e^{2\sigma }G_{MNP}^2\frac{1}{4}e^\sigma a_{IJ}F_{MN}^IF^{JMN}\frac{3}{8}(_M\sigma )^2\frac{1}{2}P_{M\overline{i}}^{\overline{a}}P_{\overline{a}}^{M\overline{i}},$$
(41)
with the field strengths $`G_{MNP}`$ and $`F_{MN}^I`$ defined as in (33).
#### 3.1.3 $`D=7`$ supergravity
The field content of the massless representations of the $`D=7`$, $`𝒩=2`$ supersymmetry algebra consists of the following multiplets
Supergravity multiplet $`:`$ $`(g_{MN},B_{MN},A_{M\overline{j}}^{\overline{i}},\sigma ,\psi _M^{\overline{i}},\chi ^{\overline{i}}),`$
Vector multiplet $`:`$ $`(A_M,\phi _{\overline{j}}^{\overline{i}},\lambda ^{\overline{i}}).`$ (42)
where all spinors are symplectic Majorana and the index $`\overline{i}=1,2`$ labels the fundamental representation of the $`Sp(1)SU(2)SO(3)`$ R-symmetry group. A general $`D=7`$, $`𝒩=2`$ supergravity theory is constructed by combining the supergravity multiplet with $`n`$ vector multiplets. This leads to the reducible multiplet
$$(g_{MN},B_{MN},A_M^I,\phi ^{\overline{\alpha }},\sigma ,\psi _\mu ^{\overline{i}},\chi ^{\overline{i}},\lambda ^{\overline{a}\overline{i}}),$$
(43)
where $`\overline{\alpha }=1,\mathrm{},3n`$ labels the scalars, $`\overline{a}=1,\mathrm{},n`$ labels the gauginos, and $`I=1,\mathrm{},n+3`$ labels the vectors resulting from the combination of $`A_{\mu \overline{j}}^{\overline{i}}`$ and $`A_\mu ^{\overline{a}}`$. Our notation and conventions are as in ,.
The $`3n`$ scalars $`\phi ^{\overline{\alpha }}`$ parameterize the non-compact coset space $`\frac{SO(3,n)}{SO(3)\times SO(n)}`$. Its representative $`L`$ and its inverse are $`(n+3)\times (n+3)`$ matrices, they are decomposed as
$$L_I^A=(L_{I\overline{j}}^{\overline{i}},L_I^{\overline{a}}),L_A^I=(L_{\overline{i}}^{\overline{j}I},L_{\overline{a}}^I),$$
(44)
and they satisfy
$$\eta _{AB}L_I^AL_J^B=\eta _{IJ},$$
(45)
with $`\eta _{AB}=\eta _{IJ}=\text{diag}(1,1,1,+1,\mathrm{},+1)`$.
The Maurer-Cartan form of $`L`$ contains the coset vielbein $`P_{M\overline{a}\overline{i}}^{\overline{j}}`$ and the $`SO(3)`$ and $`SO(n)`$ composite connections $`Q_{M\overline{i}}^{\overline{j}}`$ and $`Q_{M\overline{a}}^{\overline{b}}`$. The decomposition is as follows
$$_MLL^1=\left(\begin{array}{cc}Q_{M\overline{i}}^{\overline{j}}& P_{M\overline{j}}^{\overline{a}\overline{i}}\\ P_{M\overline{a}\overline{i}}^{\overline{j}}& Q_{M\overline{a}}^{\overline{b}}\end{array}\right).$$
(46)
We note that here we have followed the usual conventions by replacing the $`SO(3)`$ vector index $`\overline{X}=1,2,3`$ naturally appearing in the vector representation, by a pair of symmetric $`SO(3)`$ spinor (or $`Sp(1)`$ fundamental) indices $`\overline{i}\overline{j}`$. The transformation between the two types of notation will be shown and utilized in §4.3.
The bosonic Lagrangian of the theory is given by
$$e^1=\frac{1}{2}R\frac{1}{12}e^{2\sigma }G_{MNP}^2\frac{1}{4}e^\sigma a_{IJ}F_{MN}^IF^{JMN}\frac{5}{8}(_M\sigma )^2\frac{1}{2}P_{M\overline{j}}^{\overline{a}\overline{i}}P_{\overline{a}\overline{i}}^{M\overline{j}}.$$
(47)
### 3.2 Field equations
Our next task is to derive and solve the field equations, which follow from the $`D=9,8,7`$ dimensional Lagrangians (32,41,47), respectively. We will assume a configuration where the only nonzero fields besides gravity are the scalars $`\phi ^{\overline{\alpha }}`$. Employing the decomposition $`\phi ^{\overline{\alpha }}=(\varphi ^i,\chi ^\alpha )`$ of section 2, and appropriately writing the scalar Lagrangian as
$$e^1_s=\frac{1}{8}\underset{i=1}{\overset{\mathrm{}}{}}(_M\varphi ^i)^2\frac{1}{4}\underset{\alpha }{}e^{\alpha _i\varphi ^i}(F_M^\alpha )^2,$$
(48)
one easily sees that the non-trivial equations of motion, in $`D=9,8,7`$ are the Einstein equation
$$R_{MN}=\frac{1}{4}\underset{i=1}{\overset{\mathrm{}}{}}_M\varphi ^i_N\varphi ^i+\frac{1}{2}\underset{\alpha }{}e^{\alpha _i\varphi ^i}F_M^\alpha F_N^\alpha ,$$
(49)
the scalar equations for $`\varphi ^i`$,
$$\mathrm{}\varphi ^i=\underset{\alpha }{}\alpha _ie^{\alpha _i\varphi ^i}(F_M^\alpha )^2,$$
(50)
and the scalar equations for $`\chi ^\alpha `$,
$$_M(e^{\alpha _i\varphi ^i}F^{\alpha M})=\underset{\beta \gamma =\alpha }{}N_{\beta ,\gamma }e^{\gamma _i\varphi ^i}F_M^\beta F^{\gamma M}.$$
(51)
Next, we will proceed with finding an appropriate embedding of an $`\frac{SL(2,)}{U(1)}`$ submanifold in the scalar cosets. The embedding proceeds by finding a subset $`(h,e_+)`$ of the solvable Lie algebra generators (or suitable linear combinations) so that the closed set $`(h,e_+,e_{})`$, with $`e_{}`$ being the negative-root generator corresponding to $`e_+`$, is normalized so as to satisfy the $`𝔰𝔩(2,)`$ algebra. For the $`SO(2,n)`$ and $`SO(3,n)`$ cases, it is known that such a set is given by $`(\alpha H,E_\alpha )`$ where $`\alpha `$ can be any of the long roots; for the $`SO(1,n)`$ case, this requires a minor modification. The $`\frac{SL(2,)}{U(1)}`$ submanifold will then be parameterized by the two fields that correspond to $`h`$ and $`e_+`$. For the three cases under consideration, the explicit embeddings are shown below.
* $`D=9`$. In the $`D=9`$ case, the desired embedding in $`SO(1,n)/SO(n)`$ is found by observing that the generators
$$h=2H_{ϵ_1},e_\pm =E_{\pm (ϵ_1+ϵ_2)}+E_{\pm (ϵ_1ϵ_2)},$$
(52)
satisfy the $`𝔰𝔩(2,)`$ algebra
$$[h_,e_\pm ]=\pm 2e_\pm ,[e_+,e_{}]=h.$$
(53)
The fields $`\varphi `$ and $`\chi `$ parameterizing the $`\frac{SL(2,)}{U(1)}`$ submanifold are the fields along $`h`$ and $`e_+`$ respectively, namely $`\varphi \frac{1}{2}\varphi ^1`$ and $`\chi \frac{1}{2}(\chi ^{ϵ_1+ϵ_2}+\chi ^{ϵ_1ϵ_2})`$. In this case, it is readily shown that the scalar field equations are consistent with the configuration
$`\varphi ^i=0`$ $`\text{except}\varphi (x)={\displaystyle \frac{1}{2}}\varphi ^1(x)`$
$`\chi ^\alpha =0`$ $`\text{except}\chi (x)={\displaystyle \frac{1}{2}}\left(\chi ^{ϵ_1+ϵ_2}(x)+\chi ^{ϵ_1ϵ_2}(x)\right),`$ (54)
i.e., with a configuration where all fields except $`\varphi `$ and $`\chi `$ are zero. This can be seen by recalling that, in this particular case, all structure constants $`C_{\beta \gamma }^\alpha `$ vanish. It follows that (i) all axion field strengths are simply $`F_M^\alpha =_M\chi ^\alpha `$ and (ii) there are no nonzero $`N_{\beta ,\gamma }`$’s on the RHS of (51) since that would require that $`\alpha +\beta =\gamma `$ i.e. that there exists a nonzero $`C_{\alpha \beta }^\gamma `$ for some $`\gamma `$. So, the RHS of (51) is identically zero for all $`\alpha `$ and thus one is allowed to set any axion fields to zero.
* $`D=8`$. In this case, the desired embedding in $`\frac{SO(2,n)}{SO(2)\times SO(n)}`$ is identified by noticing that the generators
$$h=H_{ϵ_1}+H_{ϵ_2}=H_{ϵ_1+ϵ_2},e_\pm =E_{\pm (ϵ_1+ϵ_2)},$$
(55)
satisfy the $`𝔰𝔩(2,)`$ algebra. The fields parameterizing the $`\frac{SL(2,)}{U(1)}`$ subspace are accordingly given by $`\varphi \frac{1}{2}(\varphi ^1+\varphi ^2)`$ and $`\chi \chi ^{ϵ_1+ϵ_2}`$; the field strength of $`\chi `$ will be denoted as $`F_M`$. Similarly to the $`D=9`$ case, to show that the configuration
$`\varphi ^i=0`$ $`\text{except}\varphi (x)={\displaystyle \frac{1}{2}}\left(\varphi ^1(x)+\varphi ^2(x)\right)`$
$`\chi ^\alpha =0`$ $`\text{except}\chi (x)=\chi ^{ϵ_1+ϵ_2}(x)`$ (56)
where only $`\varphi `$ and $`\chi `$ are nonzero is consistent with the equations of motion, we proceed as follows. Recalling that there are no no nonzero structure constants with lower index $`ϵ_1+ϵ_2`$, we immediately see that the only field strength containing $`\chi `$ is $`F_M`$ and has the form
$$F_M=_M\chi +\mathrm{}$$
(57)
where the omitted terms are independent of $`\chi `$. It follows that, for our configuration, we have $`F_M=_M\chi `$ and $`F^{\alpha ϵ_1+ϵ_2}=0`$. Using these facts, we first consider the equation of motion (50) for the second dilaton $`\varphi ^{}=\frac{1}{2}(\varphi _1\varphi _2)`$. It is easily seen that the RHS of this equation vanishes,
$$\frac{1}{2}\underset{\alpha }{}(\alpha _1\alpha _2)e^{\alpha _i\varphi ^i}(F_M^\alpha )^2=\frac{1}{2}[(ϵ_1+ϵ_2)_1(ϵ_1+ϵ_2)_2]e^{2\varphi }(F_M)^2=0,$$
(58)
and so this equation is consistent with $`\varphi ^{}=0`$. Second, we consider the equation of motion (51) for the axions $`\chi ^{\alpha ϵ_1+ϵ_2}`$. For our configuration, the LHS of this equation equals zero. The only possible case in which the RHS would be nonzero is the case where $`\beta =\gamma =ϵ_1+ϵ_2`$ so that the combination $`(F_M)^2`$ would appear; however, this would require that $`\alpha =(\beta \gamma )=0`$ which cannot be satisfied. Therefore, the configuration under consideration satisfies the scalar equations of motion.
* $`D=7`$. In this last case, the embedding of $`\frac{SL(2,)}{U(1)}`$ in $`\frac{SO(3,n)}{SO(3)\times SO(n)}`$ is again given by the generators $`h`$ and $`e_\pm `$ of (55) and it is parametrized by the associated fields $`\varphi \frac{1}{2}(\varphi ^1+\varphi ^2)`$ and $`\chi \chi ^{ϵ_1+ϵ_2}`$.
Again, the configuration (3.2) is consistent with the equations of motion. For $`\varphi ^{}`$ and $`\chi ^{\alpha ϵ_1+ϵ_2}`$, the proof proceeds exactly as before. As for the extra dilaton $`\varphi ^3`$ present in this case, the same reasoning leading to Eq. (59) implies that the RHS of the corresponding equation of motion is given by
$$\underset{\alpha }{}\alpha _3e^{\alpha _i\varphi ^i}(F_M^\alpha )^2=(ϵ_1+ϵ_2)_3e^{2\varphi }(F_M)^2=0,$$
(59)
and so it is again consistent to set $`\varphi ^3=0`$.
### 3.3 The solution
As we have seen, the configurations (3.2),(3.2) in $`D=9`$ and $`D=8,7`$, respectively are consistent with the scalar field equations. It remains now to solve the Einstein equations (49), as well as the equations (50,51) for the remaining scalars $`\varphi ,\chi `$. By assembling $`\varphi `$ and $`\chi `$ into the complex combination
$$\tau =\tau _1+\text{i}\tau _2=\chi +\text{i}e^\varphi ,$$
(60)
the Einstein and scalar field equations may be written as
$$R_{MN}=\frac{1}{4\tau _2^2}(_M\tau _N\overline{\tau }+_M\overline{\tau }_N\tau )$$
(61)
and
$$^2\tau \frac{2_M\tau ^M\tau }{\overline{\tau }\tau }=0,$$
(62)
respectively. To solve these equations, we split the spacetime coordinates as $`(x^\mu ,y^m)`$ where $`x^\mu ,\mu =0,\mathrm{},D3`$ parametrize a $`(D2)`$–dimensional spacetime and $`y^m,m=1,2`$ parametrize a two-dimensional surface. Writing the internal coordinates $`y^m`$ in the complex basis $`(z,\overline{z})`$, we use the ansatz
$$ds_D^2=g_{\mu \nu }dx^\mu dx^\nu +e^{2\mathrm{\Omega }(z,\overline{z})}dzd\overline{z},\tau =\tau (z,\overline{z})$$
(63)
where the scalars depend only on the internal coordinates. The scalar equation (62) for this ansatz is written as
$$\overline{}\tau \frac{2\tau \overline{}\tau }{\tau \overline{\tau }}=0,$$
(64)
and is thus solved for any holomorphic (antiholomorphic) $`\tau =\tau (z)\left(\tau (\overline{z})\right)`$. Passing to the Einstein equation, its $`(\mu \nu )`$ components are given by
$$R_{\mu \nu }=0,$$
(65)
which implies that we may take the $`(D2)`$–dimensional spacetime to be Minkowski spacetime. As for the $`(mn)`$ components of the Einstein equation, they lead, for holomorphic $`\tau `$, to the equation
$$2\overline{}\mathrm{\Omega }=\frac{\tau \overline{}\overline{\tau }}{4\tau _2^2},$$
(66)
which is solved by
$$\mathrm{\Omega }=\frac{1}{2}\mathrm{ln}\tau _2+f(z)+\overline{f}(\overline{z}),$$
(67)
where $`f(z)`$ can be any holomorphic function. Thus, the solution for the metric and the scalars reads
$$ds_D^2=\eta _{\mu \nu }dx^\mu dx^\nu +d\sigma ^2,d\sigma ^2=\tau _2(z,\overline{z})|F(z)|^2dzd\overline{z},\tau =\tau (z),$$
(68)
where $`F(z)=\mathrm{exp}\left(f(z)\right)`$. Note that, for any holomorphic $`\tau =\tau (z)`$, there exists a corresponding 2D metric $`d\sigma ^2`$ (68), which may be singular or non-singular.
#### 3.3.1 Singular solutions
A homomorphic $`\tau `$, which leads to a singular solution is
$$\tau =\text{i}\frac{R^b+iz^b}{R^biz^b}.$$
(69)
This field gives rise to the 2D metric
$$d\sigma ^2=\left(1\left|\frac{z}{R}\right|^{2b}\right)dzd\overline{z}$$
(70)
after choosing $`F(z)=1`$ for the metric to be regular around $`z=0`$. This solution, for $`b=1`$, reduces to the tear-drop , and it is singular at $`|z|=R`$ where it has a naked singularity. However, the singularity is harmless since it does not lead to any physically unacceptable situation. It can be proven for example that energy, momentum and angular momentum are conserved ,, whereas, it may be relevant for the solution of the cosmological constant problem . Moreover, the volume of the singular transverse space is finite, which leads to finite 4D Planck constant and thus to conventional 4D gravitational interactions.
#### 3.3.2 Non-singular solutions
There are also non-singular solutions to the equations (64,66). For the construction of these solutions, we recall that the field equations (61,62) are invariant under the $`SL(2,)`$ transformation
$$\tau \frac{a\tau +b}{c\tau +d},adcb=1,$$
(71)
for real $`a,b,c,d`$. In fact, the symmetry is reduced to the modular group $`SL(2,)`$, when non-perturbative effects are taken into account. To proceed, we note that the energy per unit $`(D2)`$–volume is
$$E=\frac{\text{i}}{2}d^2z\overline{}\mathrm{ln}\tau _2.$$
(72)
In order to find finite energy solutions one has to restrict $`\tau `$ to the fundamental domain of $`SL(2,)`$ . Then, $`\tau `$ has discontinuous jumps done by the $`SL(2,)`$ transformations $`\tau \tau +1`$ as we go around the singularities at $`z=z_i`$. These jumps and the requirement of holomorphicity imply that, near the location of the singularities, we must have
$$\tau \frac{1}{2\pi \text{i}}\mathrm{ln}(zz_i).$$
(73)
The energy in this case is indeed finite and it turns out to be proportional to the volume of the fundamental domain $`_1`$,
$$E=\frac{\pi }{6}n,$$
(74)
where $`n`$ is the number of times the z-plane covers $`_1`$.
Since the fundamental domain of $`SL(2,)`$ is mapped to the complex sphere in the $`j`$–plane through the modular $`j`$–function, we may express the solution for $`\tau `$ as the pull-back of $`j(\tau )`$. Thus we may write
$$j(\tau )=\frac{P(z)}{Q(z)},$$
(75)
where $`P(z),Q(z)`$ are polynomials of degree $`p`$ and $`q`$, respectively. If $`pq`$, $`j`$ approaches a constant value as $`|z|\mathrm{}`$ and $`n=q`$ in this case. There exist $`q`$ points at which $`Q(z)`$ has zeroes and these points are singular. As has been shown in , there are singularities at the zeroes of $`P(z)`$ as well, which an be avoided however, by choosing $`P(z)=\text{const}.`$.
Recalling now that $`SL(2,)`$ is generated by
$$\tau \frac{1}{\tau },\tau \tau +1,$$
(76)
the metric (68) is clearly not modular invariant. However, we may use the freedom to choose the holomorphic function $`F(z)`$ to make the metric non-degenerate as well as modular invariant. These two conditions specify $`F(z)`$ to be
$$F(z)=\eta (\tau )^2\underset{i=1}{\overset{n}{}}(zz_i)^{1/12},$$
(77)
where $`\eta (\tau )=q^{1/24}_{r>0}(1q^r)`$ is Dedekind’s $`\eta `$–function ($`q=e^{2\pi i\tau }`$). Then, the metric turns out to be
$$ds_D^2=\eta _{\mu \nu }dx^\mu dx^\nu +\tau _2\eta (\tau )^2\overline{\eta }(\overline{\tau })\left|\underset{i=1}{\overset{n}{}}(zz_i)^{1/12}\right|^2dzd\overline{z}.$$
(78)
The asymptotic form of (78) is
$$ds_D^2\eta _{\mu \nu }dx^\mu dx^\nu +(z\overline{z})^{n/12}dzd\overline{z},$$
(79)
and one recognizes a deficit angle $`\delta =\pi n/6`$. With $`n=12`$ strings the deficit angle becomes $`\delta =2\pi `$ and the transverse space is asymptotically a cylinder while $`n=24`$ strings produce a deficit angle $`\delta =4\pi `$ and the transverse space is a compact $`S^2`$. As a result, the $`\tau `$–field configurations defined implicitly by
$$j(\tau (z))=\underset{i=1}{\overset{24}{}}\frac{1}{zz_i}$$
(80)
compactifies the D-dimensional space-time to $`M^{D2}\times S^2`$ with metric given in eq.(78) for $`n=24`$.
We should mention here that there are some special cases which may further compactify $`M^D`$. For example, one may consider two vector multiplets coupled to the gravity multiplet in $`D=8`$ supergravity. In this case, the scalar manifold is $`\frac{SO(2,2)}{SO(2)\times SO(2)}`$, which is actually $`\frac{SL(2,)}{U(1)}\times \frac{SL(2,)}{U(1)}`$. Then, in this case, compactification to four dimensions may be achieved and the $`D=8`$ vacuum is of the form $`M^4\times S^2\times S^2`$ with metric
$`ds_D^2`$ $`=`$ $`\eta _{\mu \nu }dx^\mu dx^\nu `$
$`+\tau _2\eta (\tau )^2\overline{\eta }(\overline{\tau })\left|{\displaystyle \underset{i=1}{\overset{n}{}}}(zz_i)^{1/12}\right|^2dzd\overline{z}+\sigma _2\eta (\sigma )^2\overline{\eta }(\overline{\sigma })\left|{\displaystyle \underset{i=1}{\overset{n}{}}}(ww_i)^{1/12}\right|^2dwd\overline{w}.`$
where $`\tau ,\sigma `$ parametrize the two $`\frac{SL(2,)}{U(1)}`$ factors of the scalar manifold and $`z,w`$are the complex coordinates on the transverse 4D space.
## 4 Supersymmetry
The last issue we intend to address is how many of the supersymmetries of the original theories are preserved in our scalar-induced compactifications. In what follows, we will show that the compactification of the $`D=9,8,7`$ theories preserve one half of the initial supersymmetries, leading to effective theories with $`D=7`$, $`𝒩=2`$, $`D=6`$, $`𝒩=1`$ and $`D=5`$, $`𝒩=2`$ supersymmetry respectively.
### 4.1 $`D=9`$
In the $`D=9`$ case, all we need to do is consider the gravitino variation which, in a background where all fields except gravity and $`\tau `$ are zero, is given by
$$\delta \psi _M=_Mϵ=(_M+\omega _M)ϵ,$$
(82)
where<sup>4</sup><sup>4</sup>4Here and in what follows, the flat-space gamma matrices are denoted as $`\gamma ^A`$ and the curved-space gamma matrices are denoted as $`\mathrm{\Gamma }^M=e_A^M\gamma ^A`$. $`\omega _M\frac{1}{4}\omega _{MAB}\gamma ^{AB}`$. Due to the specific form of the metric in (68), the only non-trivial spin-connection terms are $`\omega _m`$. Using conformal flatness of the internal metric, we find
$$\omega _m=\frac{1}{2}\gamma _m^n_n\mathrm{\Omega },$$
(83)
and, thus, for
$$\gamma ^{78}ϵ=\text{i}ϵ$$
(84)
we get for the gravitino shifts
$$\delta \psi _z=\left(+\frac{1}{2}\mathrm{\Omega }\right)ϵ,\delta \psi _{\overline{z}}=\left(\overline{}\frac{1}{2}\overline{}\mathrm{\Omega }\right)ϵ.$$
(85)
It easy to see that the vanishing of the gravitino shift $`\delta \psi _{\overline{z}}`$ specify $`ϵ_+`$ to be
$$ϵ_+=e^{\frac{1}{2}\mathrm{\Omega }}ϵ_0$$
(86)
where $`ϵ_0`$ is a constant spinor. Then, $`\delta \psi _z`$ is then non-vanishing and the solution preserves half of the supersymmetries. It can easily be checked that $`\delta \psi _z`$ is normalizable as it should.
### 4.2 $`D=8`$
To check supersymmetry of our compactification in the $`D=8`$ case, we have to consider the non-trivial supersymmetry variations of the fields. These are the variation of the gravitino, which now has the form
$$\delta \psi _M=D_Mϵ=\left(_M+\omega _M\frac{\text{i}}{2}Q_M\right)ϵ,$$
(87)
and the variation of the gauginos, given by
$$\delta \lambda _{\overline{a}}=\frac{\text{i}}{2}\mathrm{\Gamma }^M\widehat{P}_{M\overline{a}}ϵ,$$
(88)
where $`Q_M`$ and $`\widehat{P}_{M\overline{a}}`$ are given in (40). To check whether the above equations are satisfied in our background, we have to compute $`\omega _m`$, $`Q_m`$ and $`\widehat{P}_{ma}`$ (since all $`\mu `$ components vanish). Regarding the spin connection, we find
$$\omega _z=\frac{\text{i}}{2}\gamma ^{67}\mathrm{\Omega },\omega _{\overline{z}}=\frac{\text{i}}{2}\gamma ^{67}\overline{}\mathrm{\Omega }.$$
(89)
Regarding $`Q_m`$ and $`\widehat{P}_{m\overline{a}}`$, they may be read off from the Maurer-Cartan form in the standard $`SO(2,n)`$ basis according to (39). To this end, we take for definiteness the two non-compact Cartan generators of $`SO(2,n)`$ as $`H_{ϵ_1}=T_{13}`$ and $`H_{ϵ_2}=T_{24}`$, where $`(T_{IJ})_{KL}\delta _{IK}\delta _{JL}+\delta _{IL}\delta _{JK}`$. In this basis, the generators $`H_{ϵ_1+ϵ_2}`$ and $`E_{ϵ_1+ϵ_2}`$ are given by ,
$$H_{ϵ_1+ϵ_2}=\left(\begin{array}{ccccc}0& 0& 1& 0& \mathrm{}\\ 0& 0& 0& 1& \mathrm{}\\ 1& 0& 0& 0& \mathrm{}\\ 0& 1& 0& 0& \mathrm{}\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\end{array}\right),E_{ϵ_1+ϵ_2}=\frac{1}{2}\left(\begin{array}{ccccc}0& 1& 0& 1& \mathrm{}\\ 1& 0& 1& 0& \mathrm{}\\ 0& 1& 0& 1& \mathrm{}\\ 1& 0& 1& 0& \mathrm{}\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\end{array}\right).$$
(90)
Inserting the above expressions into (10), now given by $`_MLL^1=\frac{1}{2}_M\varphi H_{ϵ_1+ϵ_2}+e^\varphi _M\chi E_{ϵ_1+ϵ_2}`$, we find the Maurer-Cartan form
$$_MLL^1=\frac{1}{2}\left(\begin{array}{cccccc}0& e^\varphi _M\chi & _M\varphi & e^\varphi _M\chi & 0& \mathrm{}\\ e^\varphi _M\chi & 0& e^\varphi _M\chi & _M\varphi & 0& \mathrm{}\\ _M\varphi & e^\varphi _M\chi & 0& e^\varphi _M\chi & 0& \mathrm{}\\ e^\varphi _M\chi & _M\varphi & e^\varphi _M\chi & 0& 0& \mathrm{}\\ 0& 0& 0& 0& 0& \mathrm{}\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\end{array}\right).$$
(91)
From this expression, we read off the $`SO(2)`$ connection
$$Q_m=\frac{1}{2}e^\varphi _m\chi =\frac{_m\tau _1}{2\tau _2},$$
(92)
and the vielbein
$`\widehat{P}_{m1}={\displaystyle \frac{\text{i}}{2\tau _2}}(\gamma _{(9)}_m\tau _1\text{i}_m\tau _2),\widehat{P}_{m2}={\displaystyle \frac{1}{2\tau _2}}(_m\tau _1\text{i}\gamma _{(9)}_m\tau _2).`$ (93)
Using the above expressions, we can determine the supersymmetry of the background. Starting from the gravitino variation (87), we write its $`z`$ and $`\overline{z}`$ components as
$`\delta \psi _z=(+{\displaystyle \frac{\text{i}}{4}}{\displaystyle \frac{\tau _1+\gamma ^{67}\tau _2}{\tau _2}}+{\displaystyle \frac{\text{i}}{2}}\gamma ^{67}f)ϵ,\delta \psi _{\overline{z}}=(\overline{}+{\displaystyle \frac{\text{i}}{4}}{\displaystyle \frac{\overline{}\tau _1\gamma ^{67}\overline{}\tau _2}{\tau _2}}{\displaystyle \frac{\text{i}}{2}}\gamma ^{67}\overline{}\overline{f})ϵ`$ (94)
Unlike the previous case, both components of the gravitino variation can be made to vanish for nontrivial $`ϵ`$. Indeed, we immediately see that, if $`ϵ`$ is subject to the condition
$$\gamma ^{67}ϵ=\text{i}ϵ,$$
(95)
then the $`SO(2)`$ connection cancels the $`\tau _2`$–dependent part of the spin connection for holomorphic $`\tau `$. Then, the variation (94) vanishes if $`ϵ`$ is given by
$$ϵ=e^{\text{i}f_2(z,\overline{z})}ϵ_0,$$
(96)
where $`ϵ_0`$ is a constant spinor subject to (95) and $`f_2(z,\overline{z})=\text{Im}f(z)`$. Turning to the gaugino variation (88), we note that the supersymmetry spinor $`ϵ`$ satisfies
$$\gamma _{(9)}ϵ=ϵ.$$
(97)
Then, noticing that (95) implies that $`(\gamma ^6\text{i}\gamma ^7)ϵ=0`$, we easily find that the gaugino variation vanishes as well,
$`\delta \lambda _{\overline{1}}={\displaystyle \frac{1}{4\tau _2}}\mathrm{\Gamma }^m_m\overline{\tau }ϵ=0,\delta \lambda _{\overline{2}}={\displaystyle \frac{\text{i}}{4\tau _2}}\mathrm{\Gamma }^m_m\overline{\tau }ϵ=0.`$ (98)
The constraint imposed by (95) on the spinor $`ϵ`$ projects out half its components and amounts to a chirality projection. To see this, we note that the 8D and 6D chirality operators $`\gamma _{(9)}`$ and $`\gamma _{(7)}`$ are related by
$$\gamma _{(9)}=\text{i}\gamma _{(7)}\gamma ^{67},$$
(99)
so that Eq. (97) is equivalent to a chirality projection in 6D,
$$\gamma _{(7)}ϵ=ϵ.$$
(100)
Therefore, the background (68) preserves half the supersymmetries of the original theory, leading to a chiral 6D effective theory with $`𝒩=1`$ supersymmetry.
### 4.3 $`D=7`$
Let us finally check the supersymmetry of our compactification in the $`D=7`$ case. Here, the non-trivial supersymmetry variations of the fields are given by the gravitino variation
$$\delta \psi _{M\overline{i}}=D_Mϵ_{\overline{i}}=(_M+\omega _M)ϵ_{\overline{i}}+\frac{1}{2}Q_{M\overline{i}}^{\overline{j}}ϵ_{\overline{j}},$$
(101)
and the gaugino variation
$$\delta \lambda _{\overline{a}\overline{i}}=\text{i}\sqrt{2}\mathrm{\Gamma }^MP_{M\overline{a}\overline{i}}^{\overline{j}}ϵ_{\overline{j}}.$$
(102)
Proceeding as before, we find the spin connection $`\omega _z=\frac{\text{i}}{2}\gamma ^{56}\mathrm{\Omega }`$. Passing employing a basis where $`H_{ϵ_1}=T_{24}`$ and $`H_{ϵ_2}=T_{35}`$, we find the Maurer-Cartan form
$$_MLL^1=\frac{1}{2}\left(\begin{array}{ccccccc}0& 0& 0& 0& 0& 0& \mathrm{}\\ 0& 0& e^\varphi _M\chi & _M\varphi & e^\varphi _M\chi & 0& \mathrm{}\\ 0& e^\varphi _M\chi & 0& e^\varphi _M\chi & _M\varphi & 0& \mathrm{}\\ 0& _M\varphi & e^\varphi _M\chi & 0& e^\varphi _M\chi & 0& \mathrm{}\\ 0& e^\varphi _M\chi & _M\varphi & e^\varphi _M\chi & 0& 0& \mathrm{}\\ 0& 0& 0& 0& 0& 0& \mathrm{}\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\end{array}\right).$$
(103)
From this, we can immediately read off the $`SO(3)`$ connection and the vielbein in the forms $`\stackrel{~}{Q}_{m\overline{X}}^{\overline{Y}}`$ and $`\stackrel{~}{P}_{m\overline{a}}^{\overline{X}}`$ that involve the *triplet* index $`\overline{X}=1,2,3`$ for $`SO(3)`$ (the use of this index is emphasized by the tildes). We find
$$\stackrel{~}{Q}_{m\overline{2}}^{\overline{3}}=\stackrel{~}{Q}_{m\overline{3}}^{\overline{2}}=\frac{_m\tau _1}{2\tau _2}$$
(104)
and
$$\stackrel{~}{P}_{m\overline{1}}^{\overline{2}}=\stackrel{~}{P}_{m\overline{2}}^{\overline{3}}=\frac{_m\tau _2}{2\tau _2},\stackrel{~}{P}_{m\overline{1}}^{\overline{3}}=\stackrel{~}{P}_{m\overline{2}}^{\overline{2}}=\frac{_m\tau _1}{2\tau _2}$$
(105)
To insert these expressions in the supersymmetry transformations (101) and (102), we have to switch from the triplet notation to the doublet notation. For this, we use the transformations
$$Q_{m\overline{i}}^{\overline{j}}=\frac{\text{i}}{2}ϵ_{\overline{X}\overline{Y}\overline{Z}}(\sigma ^{\overline{X}})_{\overline{i}}^{\overline{j}}\stackrel{~}{Q}_m^{\overline{Y}\overline{Z}},P_{m\overline{a}\overline{i}}^{\overline{j}}=\frac{1}{\sqrt{2}}\stackrel{~}{P}_{m\overline{a}}^{\overline{X}}(\sigma ^{\overline{X}})_{\overline{i}}^{\overline{j}},$$
(106)
which yield
$$Q_{m\overline{1}}^{\overline{2}}=Q_{m\overline{2}}^{\overline{1}}=\text{i}\frac{_m\tau _1}{2\tau _2},$$
(107)
and
$`P_{m\overline{1}\overline{1}}^{\overline{1}}=P_{m\overline{1}\overline{2}}^{\overline{2}}={\displaystyle \frac{_m\tau _1}{2\sqrt{2}\tau _2}},P_{m\overline{1}\overline{1}}^{\overline{2}}=P_{m\overline{1}\overline{2}}^{\overline{1}}=\text{i}{\displaystyle \frac{_m\tau _2}{2\sqrt{2}\tau _2}},`$
$`P_{m\overline{2}\overline{1}}^{\overline{1}}=P_{m\overline{2}\overline{2}}^{\overline{2}}={\displaystyle \frac{_m\tau _2}{2\sqrt{2}\tau _2}},P_{m\overline{2}\overline{1}}^{\overline{2}}=P_{m\overline{2}\overline{2}}^{\overline{1}}=\text{i}{\displaystyle \frac{_m\tau _1}{2\sqrt{2}\tau _2}}.`$ (108)
Using the above expressions, we can determine the supersymmetry of the background. Indeed, considering the gravitino shifts (101) we find that, $`\delta \psi _{M\overline{i}}=0`$ if the spinors $`ϵ_{\overline{1}}`$ and $`ϵ_{\overline{2}}`$ are subject to the condition
$$\gamma ^{56}ϵ_{\overline{1}}=\text{i}ϵ_{\overline{2}}.$$
(109)
Then the $`SO(3)`$ connection cancels the $`\tau _2`$–dependent part of the spin connection for holomorphic $`\tau `$. The resulting equations form a system of coupled differential equations whose solution is concisely written as
$$ϵ_{\overline{1}}=e^{\gamma ^{56}f_2(z,\overline{z})}ϵ_{0,\overline{1}},ϵ_{\overline{2}}=e^{\gamma ^{56}f_2(z,\overline{z})}ϵ_{0,\overline{2}},$$
(110)
where $`ϵ_{0,\overline{1}}`$ and $`ϵ_{0,\overline{2}}`$ are constant spinors satisfying (109). Then, it can easily be shown that the gaugino shifts vanish as well. Noting that the condition (109) implies that $`\gamma ^6ϵ_{\overline{1},\overline{2}}=\text{i}\gamma ^5ϵ_{\overline{2},\overline{1}}`$ and using holomorphicity of $`\tau `$, one sees that this variation vanishes as well. For example, for $`\delta \lambda _{\overline{1}\overline{1}}`$, we find
$`\delta \lambda _{\overline{1}\overline{1}}`$ $`=`$ $`{\displaystyle \frac{\text{i}}{2\tau _2^{3/2}}}\left[\gamma ^5(_5\tau _1ϵ_{\overline{1}}\text{i}_5\tau _2ϵ_{\overline{2}})+\gamma ^6(_6\tau _1ϵ_{\overline{1}}\text{i}_6\tau _2ϵ_{\overline{2}})\right]`$ (111)
$`=`$ $`{\displaystyle \frac{\text{i}}{2\tau _2^{3/2}}}\gamma ^5\left[(_5\tau _1_6\tau _2)ϵ_{\overline{1}}\text{i}(_6\tau _1+_5\tau _2)ϵ_{\overline{2}}\right]`$
$`=`$ $`{\displaystyle \frac{\text{i}}{2\tau _2^{3/2}}}\gamma ^5\left[(\overline{}\tau +\overline{\tau })ϵ_{\overline{1}}(\overline{}\tau \overline{\tau })ϵ_{\overline{2}}\right]=0,`$
and similarly for the other gauginos.
The condition (109), correlating the two symplectic-Majorana 7D supersymmetry spinors, again halves their degrees of freedom. So, the background (68) preserves half the supersymmetries of the original theory, leading to a 5D effective theory with $`𝒩=2`$ supersymmetry.
## 5 Conclusions
In this work, we have shown how scalars may trigger compactification of $`D=9,8`$ and $`D=7`$ supergravities. This has previously be shown to work for the tear-drop solution , in 10D type IIB supegravity. In this case, the complex scalar of type IIB theory, which parametrize $`\frac{SL(2,)}{U(1)}`$ is used to curl up two of the space-time coordinates leading to an internal space diffeomorphic to the scalar manifold. There is a naked singularity in this construction, which however is harmless as it does not lead to any violation of conservation laws. In other words, although the presence of the singularity is annoying, there is no any leakage of energy, momentum, or angular momentum through it. Moreover, the volume of the transverse external space is finite leading to a finite 4D Planck mass and, consequently, to conventional 4D gravity.
Here, we studied a similar possible mechanism for the case of higher-dimensional supergravities. The compactification mechanism we were after is triggered by the many scalars which exist in higher-dimensional supergravities. We have considered $`D=9,8,7`$ minimal supergravities coupled to vector multiplets. The vector multiplets contain scalars which, together with the scalars of the gravity multiplet, form scalar manifolds of the general form $`\frac{SO(10D,n)}{SO(10D)\times SO(n)}`$ in the presence of $`n`$ vector multiplets. We have shown here that there are solutions to the supergravity field equations where all but two of the scalars are non zero. These two non-trivial scalars parametrize the $`\frac{SL(2,)}{U(1)}`$ coset. We then presented solutions which are either singular, like the tear-drop prototype ,, or non-singular like the stringy-cosmic string . The latter compactify spacetime into $`M^{D2}\times S^2`$, where for the particular case of $`D=8`$ with two vector multiplets, compactification to $`M^4\times S^2\times S^2`$ may be achieved. Finally, we have shown that these compactifications are supersymmetric.
Acknowledgements
We would like to thank S. D. Avramis for helpful discussions. This work is co-funded by the European Social Fund (75%) and National Resources (25%) - (EPEAEK II) - PYTHAGORAS. |
warning/0507/cond-mat0507061.html | ar5iv | text | # Coulomb Interactions and Ferromagnetism in Pure and Doped Graphene
## I Introduction
The ferromagnetic instability due to the exchange interaction in a three dimensional (3D) electron gas attracted attention since the early days of quantum mechanics Bloch (1929) and has been studied in great detail Iwamoto and Sawada (1962); Misawa (1965). Recent Monte Carlo calculations Ceperley (1978); Ceperley and Alder (1980) have confirmed the presence of ferromagnetism in the phase diagram of the 3D electron gas at low doping. Similar studies have also suggested the existence of a ferromagnetic phase in the diluted two dimensional (2D) electron gas Attaccalite et al. (2002) with a first order transition from a paramagnetic phase to a ferromagnetic phase with full polarization. As the electron density is reduced, electron-electron interactions become stronger and dynamical screening disappear. At the extreme limit of zero density the electron gas should crystalize into a Wigner solid where the electrons feel the unscreened Coulomb interaction. The elusive ferromagnetic phase of the electron gas lurks between the Wigner crystal and the Fermi liquid state that exists at higher doping when electron-electron interactions are fully screened Attaccalite et al. (2002); Ceperley (1999).
In recent years, the experimental search for the ferromagnetic phase of the diluted electron gas has not been succesful Ceperley (1999); Fisk et al. (2002); Young et al. (1999). Nevertheless, there has been strong experimental indications on the existence of ferromagnetism in highly disordered graphite samples Esquinazi et al. (2003); dis . The origin of this phase is still unclear, and a number of different mechanisms have been proposed Ovchinnikov and Shamovsky (1991); Harigaya (2001); Lehtinen et al. (2004); Vozmediano et al. (2005). Nevertheless, there is no final word on the origin of ferromagnetism in graphite. Graphite is a layered material made out of graphene layers (a honeycomb lattice with one electron per $`\pi `$ orbital, that is, a half-filled band). The traditional view of graphite based on band-structure calculations assumes coherent hopping between graphene layers, and describes graphite as a low density metal with almost compensated electron and hole pockets, with $`10^4`$ to $`10^5`$ electrons per Carbon Brandt et al. (1988). This traditional picture, however, completely disregards the strong and unscreened interactions between electrons that should exist at low densities. In fact, recent experiments in true 2D graphene systems Novoselov et al. (2004); Zhang et al. (2004, 2005); Berger et al. (2004) show that electron-electron interactions and disorder have to be taken into account in order to obtain a fully consistent picture of graphene nun . Recent theoretical results nun raise questions on the wisdom of thinking of strongly correlated layered system such as graphite, as truly 3D. The claim is that the full 2D nature of graphene has to be taken into account before graphene planes are coupled by weak van der Waals interactions in order to form the 3D solid.
One of the most striking features of the electronic structure of perfect graphene planes is the linear relationship between the electronic energy, $`E_𝒌`$, with the two-dimensional momentum, $`𝒌=(k_x,k_y)`$, that is: $`ϵ(𝒌)=\pm \mathrm{}v_\mathrm{F}|𝒌|`$, where $`v_\mathrm{F}`$ is the Dirac-Fermi velocity. This singular dispersion relation is a direct consequence of the honeycomb lattice structure that can be seen as two interpenetrating triangular sublattices. In ordinary metals and semiconductors the electronic energy and momentum are related quadratically via the so-called effective mass, $`m^{}`$, ($`E_𝒌=\mathrm{}^2𝒌^2/(2m^{})`$), that controls much of their physical properties. Because of the linear dispersion relation, the effective mass in graphene is zero, leading to an unusual electrodynamics. In fact, graphene can be described mathematically by the 2D Dirac equation, whose elementary excitations are particles and holes (or anti-particles), in close analogy with systems in particle physics. In a perfect graphene sheet the chemical potential crosses the Dirac point and, because of the dimensionality, the electronic density of states vanishes at the Fermi energy. The vanishing of the effective mass or density of states has profound consequences. It has been shown, for instance, that the Coulomb interaction, unlike in an ordinary metal, remains unscreened and gives rise to an inverse quasi-particle lifetime that increases linearly with energy or temperature González et al. (1996), in contrast with the usual metallic Fermi liquid paradigm, where the inverse lifetime increases quadratically with energy.
As mentioned above, its is well known that direct exchange interactions can lead to a ferromagnetic instability in a dilute electron gas Bloch (1929); Stoner (1947). In this work we generalize the analysis of the exchange instability of the electron gas to pure and doped 2D graphene sheets. Although pure graphene should be a half-filled system, we have recently shown nun that extended defects such as dislocations, disclinations, edges, and micro-cracks can lead to the phenomenon of self-doping where charge is transfered to/from defects to the bulk in the presence of particle-hole asymmetry. The extended defects are unavoidable in graphene because there can be no long-range positional Carbon order at finite temperatures in 2D (the Hohenberg-Mermin-Wagner theorem). Furthermore, we have also shown that although extended defects lead to self-doping, they do not change the transport and electronic properties. Life-time effects are actually introduced by localized disorder such as vacancies and ad-atoms. Thus, we have also considered the influence of disorder in the generation of ferromagnetism. It is worth noting that the possibility of other instabilities in a graphene plane, related to the Coulomb interaction have also been studied in the literature Khveshchenko (2001); Gorbar et al. (2002). The nature of the exchange instability in a system with many bands is also interesting on its own right Goñi et al. (2002), and it has not been studied extensively. Furthermore, graphene is the basic material for the synthesis of other compounds with sp<sup>2</sup> bonding: graphite is obtained by the stacking of graphene planes, Carbon nanotubes are synthesized by the wrapping of graphene along certain directions, and fullerenes ”buckyballs” are generated from graphene by the creation of topological defects with five and seven fold symmetry. Therefore, the understanding of the ferromagnetic instability in graphene can have impact on a large class of systems. Finally, we also mention that a simple analysis using the standard Stoner criterium for ferromagnetism fails in graphene, as the density of states of undoped graphene vanishes at the Fermi level Peres et al. (2004).
The electron-electron interaction in graphene can lead to other instabilities at low temperatures, in addition to the ferromagnetic phase considered here. A local on site repulsive term can lead to an antiferromagnetic phase, when its value exceeds a critical thresholdSorella and Tossatti (1992); Peres et al. (2004). In the following, we will concentrate on the role of the ferromagnetic exchange instability, which, as already mentioned, is important in electronic systems with a low density of carriers, and which has not been considered in the literature so far.
Our main results can be summarized by the zero-temperature phase diagram $`g`$versus $`n`$ (where $`n`$ is the doping away from half-filling) shown in Fig. 1. The strength of the electron-electron interactions in graphene is parameterized by the dimensionless coupling constant, $`g`$, defined as:
$`g={\displaystyle \frac{e^2/ϵ_0}{\mathrm{}v_\mathrm{F}}},`$ (1)
where $`e`$ is the charge of the electron, and $`ϵ_0`$ the dielectric constant of the system. Notice that $`g`$ is exactly the ratio between the Coulomb to the kinetic energy of the electron system. This coupling constant replaces the well-known parameter $`r_s(e^2/ϵ_0)/[\mathrm{}^2k_F/m^{}]`$ of the non-relativistic electron gas (where $`k_F`$ is the Fermi momentum). In the pure compound ($`n=0`$) the paramagnetic-ferromagnetic transition is of first order with partial polarization and occurs at a critical value of $`g=g_c5.3`$. As the doping is increased, the ferromagnetic transition is suppressed (a larger value of $`g_c`$ is required) up to around $`n0.2`$ where the first order line ends at a tri-critical point a line of second order transitions emerges with a fully polarized ferromagnetic phase. A unique feature of the ferromagnetism in these systems, unlike the ordinary 2D and 3D electron gases, is the fact that there are two types of ferromagnetic phases, one that has only one type of carrier (either electron or hole) and a second phase with two types of carriers (electrons and holes).
The paper is organized as follows: in the next section we present the model for a graphene plane in the continuum limit taken into account the Dirac fermion spectrum and the long-range Coulomb interactions; in Section III we discuss the exchange energy for graphene through a variational wavefunction calculation in three different situations: Dirac fermions without a gap; Dirac fermions with a gap; and Dirac fermions with disorder treated within the coherent potential approximation (CPA) approximation; Section IV contains our conclusions. We also have included two appendixes with the details of the calculations.
## II The model for a graphene layer
The valence and conducting bands in graphene are formed by Carbon $`\pi `$ orbitals which are arranged in an honeycomb lattice (a non-Bravais lattice). The extrema of these bands lie at the $`\mathrm{\Gamma }`$ point and at the two inequivalent corners of the hexagonal Brillouin Zone. When the filling is close to one electron per Carbon atom, the Fermi energy lies close to the corners. Near these points, a standard long wavelength expansion gives for the kinetic part of the Hamiltonian the expression,
$$_{\mathrm{kin}}(𝒌)\mathrm{}v_\mathrm{F}\left(\begin{array}{cc}0& k_x+ik_y\\ k_yik_y& 0\end{array}\right),$$
(2)
which leads to the dispersion relation,
$$ϵ(𝒌)=\pm \mathrm{}v_\mathrm{F}|𝒌|.$$
(3)
In a tight-binding description of the graphene plane with nearest neighbor hopping energy $`t`$ the Dirac-Fermi velocity is given by:
$`\mathrm{}v_\mathrm{F}={\displaystyle \frac{3}{2}}ta`$ (4)
where $`a`$ is the Carbon-Carbon distance ($`t2.5`$ eV and $`a=1.42\AA `$) Brandt et al. (1988). The eigenstates of (2) can be written as:
$`\mathrm{\Psi }_{𝒌,\alpha ,\sigma }(𝒓)`$ $``$ $`\left(\begin{array}{c}\psi _a(𝒓)\\ \psi _b(𝒓)\end{array}\right)\chi _\sigma ,`$ (7)
$`=`$ $`{\displaystyle \frac{e^{i𝒌𝒓}}{\sqrt{2}}}\left(\begin{array}{c}e^{i\varphi _𝒌/2}\\ \alpha e^{i\varphi _𝒌/2}\end{array}\right)\chi _\sigma ,`$ (10)
where $`a`$ and $`b`$ label the two sublattices of the honeycomb lattice, $`\varphi _𝒌=\mathrm{arctan}(k_y/k_x)`$ is a phase factor, $`\alpha =\pm 1`$ labels the electron and hole-like bands, and $`\chi _\sigma `$ is the spin part of the wavefunction. The dispersion and the wavefunctions are the solutions of the 2D Dirac equation. This approach in the continuum requires the introduction of a cut-off in momentum space, $`k_c`$, in such a way that all momenta, $`𝒌`$, are defined such that: $`0|𝒌|k_c`$, where $`k_c`$ is chosen so as to keep the number of states in the Brillouin zone is fixed, that is, $`\pi k_c^2=(2\pi )^2/A_0`$, and $`A_0`$ is the area of the unit cell in the honeycomb lattice.
It is easy to show that with the dispersion given in (3) the single particle density of states, $`\rho (E)`$, vanishes linearly with energy at the Dirac point, $`\rho (E)|E|`$. In this case, there is no electronic screening DiVincenzo and Mele (1984) and the electrons interact through long-range Coulomb forces. The electron-electron interactions can be written in terms of the field operators, $`\widehat{\mathrm{\Psi }}(𝒓)`$, as:
$$_\mathrm{I}=\frac{1}{2}𝑑𝒓_1𝑑𝒓_2\widehat{\mathrm{\Psi }}^{}(𝒓_1)\widehat{\mathrm{\Psi }}^{}(𝒓_2)V(𝒓_1𝒓_2)\widehat{\mathrm{\Psi }}(𝒓_2)\widehat{\mathrm{\Psi }}(𝒓_1),$$
(11)
where $`V(𝒓)=e^2/(ϵ_0r)`$ is the bare Coulomb interaction. One can now expand the field operators in the basis of states given in (10), that is,
$`\widehat{\mathrm{\Psi }}(𝒓)={\displaystyle \frac{1}{\sqrt{A}}}{\displaystyle \underset{𝒌,\alpha ,\sigma }{}}\mathrm{\Psi }_{𝒌,\alpha ,\sigma }(𝒓)a_{𝒌,\alpha ,\sigma }`$ (12)
where $`a_{𝒌,\alpha ,\sigma }`$ ($`a_{𝒌,\alpha ,\sigma }^{}`$) is the annihilation (creation) operator for an electron with momentum $`𝒌`$, band $`\alpha `$, and spin $`\sigma `$ ($`\sigma =,`$ and $`A`$ is the area of the system). In this case, the Coulomb interaction reads:
$`_\mathrm{I}`$ $`=`$ $`{\displaystyle \frac{2\pi e^2}{8ϵ_0A}}{\displaystyle \underset{𝒌,𝒑,𝒒}{}}{\displaystyle \underset{\alpha _1,\mathrm{},\alpha _4}{}}{\displaystyle \underset{\sigma ,\sigma ^{}}{}}{\displaystyle \frac{1}{q}}[\alpha _2\alpha _3e^{i[\varphi ^{}(𝒑)\varphi (𝒑+𝒒)]}+1][\alpha _1\alpha _4e^{i[\varphi ^{}(𝒌)\varphi (𝒌+𝒒)]}+1]a_{𝒌,\alpha _1,\sigma _1}^{}a_{𝒑,\alpha _2,\sigma _2}^{}a_{𝒑+𝒒,\alpha _3,\sigma _2}a_{𝒌𝒒,\alpha _4,\sigma _1}.`$
It is easy to see that the Coulomb interaction induces scattering between bands (inter-band) and also within each band (intra-band). Furthermore, the $`1/q`$ dependence of the interaction (that comes from the Fourier transform of the $`1/r`$ potential in 2D) provides an electron-electron scattering that is stronger than in 3D, allowing for the possibility of a ferromagnetic transition at weaker coupling. As in the case of the Hund’s coupling in atomic systems, the spin polarized state is always preferred when long-range interactions are present since, by the Pauli’s exclusion principle, both kinetic and Coulomb energies are minimized simultaneously. This should be contrast with ultra-short range interactions of the Hubbard type that almost always benefit anti-ferromagnetic coupling via a kinetic exchange mechanism.
## III Exchange energy of a graphene plane.
In what follows we examine the required conditions for a ferromagnetic ground state in graphene. Our purpose in this work is not to obtain exact values for the critical couplings, that may required more sophisticated approaches, but instead our aim is to show that a ferromagnetic ground state in graphene is possible in principle. In order to study the ferromagnetic instability we use a variational procedure that respects all the symmetries of the problem. We assume that: (i) the ferromagnetic instability only affects states close to the Dirac points in the region at the edge of the Brillouin zone (that is, long wavelength approximation is still valid); (ii) in the ferromagnetic state the electronic bands are shifted rigidly (hence, self-energy effects such as Dirac-Fermi velocity renormalizations are neglected); (iii) even when the bands are shifted, and a finite density of states is produced at the Fermi energy, the Coulomb interaction remains unscreened (this assumption is equivalent to assume that the chemical potential shift is always small and that the screening length is larger than the inter-particle distance); (iv) the ferromagnetic state is uniform and translational invariant. Besides considering the case of a gapless system, we have also studied the case where a gap $`\mathrm{\Delta }`$ opens in the Dirac spectrum (that is, when the dispersion relation becomes $`E_𝒌=\pm \sqrt{\mathrm{\Delta }+\mathrm{}^2v_\mathrm{F}^2}`$). The gapped case is interesting because it allows the study of the crossover between the Dirac case when $`\mathrm{\Delta }=0`$ to the standard 2D case with a finite effective mass $`m^{}\mathrm{\Delta }`$ (see details ahead). We also briefly the discuss the effects of disorder on the stabilization of the ferromagnetic state via a CPA approximation in order to point out that disorder may be fundamental for the realization of a ferromagnetic phase in graphite.
### III.1 Gapless system
#### III.1.1 Exchange energy. Inter- and intraband contributions.
The possible ferromagnetic instability arises from the gain in exchange energy when the system is polarized. A finite spin polarization, on the other hand, leads to an increase in kinetic energy. Thus, there are two competing energies in the problem: the exchange energy that is minimized by polarization and the kinetic energy that is increased by it. The variational states that we consider in our approach are Slater determinants of the wave-functions given by (10) in the configurations shown in Fig.2.
As function of the Fermi wave vector, $`k_\mathrm{F}`$, the kinetic energy of the unpolarized state is:
$$_{\mathrm{kin}}=K=\frac{A}{3\pi }v_\mathrm{F}\mathrm{}(k_c^3k_\mathrm{F}^2),$$
(14)
and the exchange energy, for any doping, as determined from Eq.(LABEL:exchange) can be written as
$`E_{ex}=`$ $``$ $`{\displaystyle \frac{A}{(2\pi )^2}}{\displaystyle \frac{e^2}{4ϵ_0}}{\displaystyle \underset{\sigma }{}}{\displaystyle \underset{\alpha _a,\alpha _b}{}}{\displaystyle _0^{2\pi }}𝑑\theta {\displaystyle kp𝑑k𝑑p}`$ (15)
$`{\displaystyle \frac{1+\alpha _a\alpha _b\mathrm{cos}\theta }{|𝒌𝒑|}}n_\mathrm{F}^{\sigma ,\alpha _a}(𝒌)n_\mathrm{F}^{\sigma ,\alpha _b}(𝒑),`$
where $`n_\mathrm{F}^{\sigma ,\alpha _a(\alpha _b)}(𝒌)`$ is the Fermi occupation function, $`a(b)`$ is the band indice, and $`\alpha _a,\alpha _b=\pm 1`$.
In the ferromagnetic state the degeneracy of the spin states is lifted and the Fermi momentum of the up and down spin states becomes $`k_{}`$ and $`k_{}`$, respectively. Depending on the values of $`k_\mathrm{F},k_{}`$ and $`k_{}`$, we can define the three cases shown in Fig. 2. For a doping, $`\delta `$ per unit area, the number of electrons per Carbon away from half-filling, $`n`$, can be written as:
$`n=\delta A_0.`$ (16)
Because of the different values of $`k_{}`$ and $`k_{}`$ the system acquires a spin magnetization, $`\mu =g_s\mu _Bm`$, where $`g_s2`$ is the electron gyromagnetic factor, $`\mu _B`$ is the Bohr magneton, and $`m=sA_0`$ with $`s=n_{}n_{}`$, is the spin polarization. Notice that the maximum polarization allowed is $`m=22n`$ since each added (subtracted) electron leads to a doubly (empty) Carbon $`\pi `$ orbital.
The total exchange energy, eq.(15), can be split into intra- and inter-band contributions. In many band systems where the different bands arise from different atomic orbitals, the overlap integral between Bloch states corresponding to different bands can be neglected, and, consequently, there are no inter-band contributions to the exchange energy. An analogous effect arises when the different bands are localized at different sites of the lattice, as in the gapful case to be considered below. There are also situations where the different bands arise from the same orbitals at the same sites, but their phases in a region much larger than the unit cell are such that the overlap integral vanishes. This is the case for the two different Dirac cones which can be defined in the honeycomb lattice. We do not need to include in eq.(15) terms due to interactions between electrons near different Dirac points of the Brillouin Zone.
The case studied here, where the overlap between Bloch states in different bands cannot be neglected, and a corresponding term in the exchange energy has to the included is generic to narrow gap semiconductors, and this term may be important in lightly doped materials.
It is worth noting that these inter-band exchange effect arise from the non local nature of the exchange interaction. They cannot be studied when the exchange energy is approximated by a local term which only depends on the total charge density.
#### III.1.2 Undoped case: $`n=0`$
The Fermi level in the paramagnetic case is at $`ϵ_\mathrm{F}=0`$, and the bands are half-filled. Then, in the paramagnetic state one has $`k_{}=k_{}`$. When the system polarizes the magnetization is such that $`k_{}=\sqrt{2\pi s}`$ and the change in energy relative to the paramagnetic state is given by:
$`\mathrm{\Delta }E`$ $`=`$ $`\mathrm{\Delta }K+\mathrm{\Delta }E_{\mathrm{ex}}={\displaystyle \frac{A_0}{3\pi }}\mathrm{}v_\mathrm{F}k_{}^3`$ (17)
$``$ $`{\displaystyle \frac{A_0}{(2\pi )^2}}{\displaystyle \frac{e^2}{4ϵ_0}}\left[2k_{}^3R_1(1)4k_ck_{}^2R_0\left({\displaystyle \frac{k_{}}{k_c}}\right)\right],`$
where the functions $`R_n(x)`$ are defined in the Appendix A. Unfortunately it is not possible to find an analytical expression (using elementary functions) for the energy change as a function of the electron polarization $`s=k_{}^2/(2\pi )`$. For $`k_{}k_c`$, the leading contribution comes from the expansion of function $`R_0(x)x\mathrm{ln}(x)`$ for $`x1`$ (see Appendix A). Hence, the exchange energy increases as the polarization increases, and a ferromagnetic state with small magnetization is not favored. This effect can be cast as a logarithmic renormalization of the Fermi energy, which reduces the density of states near the Fermi level, and suppresses the tendency toward ferromagnetismGonzález et al. (1999).
At large magnetizations, $`k_c^2/s1`$, the kinetic energy contribution tends to a term proportional to $`v_\mathrm{F}k_c^3`$ and the exchange contribution becomes negative and proportional to $`(e^2k_c^3)/ϵ_0`$. The exchange term dominates, and the system undergoes a discontinuous transition to a state with polarization of order unity when:
$$g_c=\frac{e^2}{\mathrm{}v_\mathrm{F}ϵ_0}\frac{16\pi }{6R_1(1)12R_0(1)}5.3,$$
(18)
which gives the critical coupling $`g_c(n=0)5.3`$ for the appearance of ferromagnetism in the clean system, as shown in Fig.1.
#### III.1.3 Doped case, $`n0`$, one type of carrier in the ferromagnetic phase
In this case the doping, $`\delta `$, and magnetization, $`s`$, are such that $`k_\mathrm{F}=\sqrt{2\pi \delta }`$ in the paramagnetic paramagnetic phase, and $`k_{}=\sqrt{2\pi (s+\delta )}`$ and $`k_{}=\sqrt{2\pi (s\delta )}`$ in the ferromagnetic phase. In this phase there is only one type of carriers, either electrons or holes. The change in energy between the paramagnetic and ferromagnetic phase is:
$`\mathrm{\Delta }E`$ $`=`$ $`\mathrm{\Delta }K+\mathrm{\Delta }E_{\mathrm{ex}}={\displaystyle \frac{A_0}{6\pi }}v_F\mathrm{}(k_{}^3+k_{}^32k_\mathrm{F}^3)`$ (19)
$``$ $`{\displaystyle \frac{A_0}{(2\pi )^2}}{\displaystyle \frac{e^2}{ϵ_0}}[k_{}^3R_1(1)+k_{}^3R_1(1)2k_\mathrm{F}^3R_1(1)`$
$`+`$ $`2k_ck_{}^2R_2\left({\displaystyle \frac{k_{}}{k_c}}\right)+2k_ck_{}^2R_2\left({\displaystyle \frac{k_{}}{k_c}}\right)`$
$``$ $`4k_ck_\mathrm{F}^2R_2\left({\displaystyle \frac{k_\mathrm{F}}{k_c}}\right)].`$
The behavior of the energy change as a function of the spin polarization is shown in left hand pannel in Fig.3 for points $`1`$ and $`2`$ of the phase diagram in Fig.1. Notice that the transition between the paramagnetic phase (point $`2`$) to the ferromagnetic phase (point $`1`$) is discontinuous with full polarization, $`m=22n`$. In this case analytical expansion when $`s\delta `$ is now possible. For $`k_\mathrm{F},k_{},k_{}k_c`$ the value of the exchange contribution is dominated by the expansion of $`R_2(x)`$ (see Appendix A). The contribution of the exchange interaction to the term proportional $`s^2`$ is positive at low doping, and a continuous ferromagnetic transition is not possible. This contribution becomes negative only for $`n=\delta A_00.059`$. As in the previous case, we can also analyze the system energy for large values of the magnetization. We obtain an instability to a ferromagnetic state with full polarization ($`m=22n`$), which for $`n0`$ leads to a state with both electron and hole carriers with different Fermi surface areas. The dependence of the coupling constant $`g_c`$ on $`n`$ is given in Fig.1 by the dashed line. (See more on the conclusions about a speculative scenario for the origin of electrons and hole pockets in graphite.)
#### III.1.4 Doped case, $`n0`$, two types of carriers in the ferromagnetic phase
In this case the calculation is analogous to the previous one. The change in energy in this case is given by:
$`\mathrm{\Delta }E`$ $`=`$ $`\mathrm{\Delta }K+\mathrm{\Delta }E_{\mathrm{ex}}={\displaystyle \frac{A_0}{6\pi }}v_F\mathrm{}(k_{}^3k_{}^32k_\mathrm{F}^3)`$ (20)
$``$ $`{\displaystyle \frac{A_0}{(2\pi )^2}}{\displaystyle \frac{e^2}{ϵ_0}}[k_{}^3R_1(1)+k_{}^3R_1(1)2k_\mathrm{F}^3R_1(1)`$
$``$ $`2k_ck_{}^2R_1\left({\displaystyle \frac{k_{}}{k_c}}\right)+2k_ck_{}^2R_2\left({\displaystyle \frac{k_{}}{k_c}}\right)`$
$``$ $`4k_c^2R_2\left({\displaystyle \frac{k_\mathrm{F}}{k_c}}\right)].`$
As in the two previous cases, the leading term when $`k_\mathrm{F},k_{},k_{}k_c`$ is due to the expansion of the function $`R_2(x)`$, which leads to an increase in the exchange energy, which is detrimental for ferromagnetism. The energy change as a function of $`m`$ is shown in the right hand panel of Fig.3. We show the energy at points $`3`$ (paramagnetic) and $`4`$ (ferromagnetic) of Fig. 1. The transition in this case is second order with only partial polarization, $`m<n`$. As a consequence only one type of carries exist. The dependence of the coupling constant $`g_c`$ on $`n`$ is given in Fig.1 by the solid line.
### III.2 Gapful system
A gap can open in the Dirac spectrum when the two sites in the unit cell of the honeycomb lattice model become inequivalent equivalent. In this case, the kinetic energy Hamiltonian, Eq.(2) changes to:
$$_{\mathrm{kin}}(𝒌)\left(\begin{array}{cc}\mathrm{\Delta }& v_\mathrm{F}\mathrm{}(k_x+ik_y)\\ v_\mathrm{F}\mathrm{}(k_yk_y)& \mathrm{\Delta }\end{array}\right),$$
(21)
which leads to the modified dispersion relation,
$$ϵ_𝒌=\pm \sqrt{\mathrm{\Delta }^2+(\mathrm{}v_\mathrm{F}|𝒌|)^2}.$$
(22)
For wavevectors such that $`\mathrm{}v_\mathrm{F}|𝒌|\mathrm{\Delta }`$ the energies and wavefunctions are essentially the ones found in the absence of the gap, as discussed previously. If the filling is such that the Fermi wavevector satisfies this conditions, but $`k_\mathrm{F}k_c`$ the analysis presented earlier remains valid. At sufficiently low fillings, $`\mathrm{}v_\mathrm{F}|k_\mathrm{F}|\mathrm{\Delta }`$, the dispersion relation, Eq.(21) can be approximated by:
$$ϵ_𝒌\pm \mathrm{\Delta }\pm \frac{(\mathrm{}v_\mathrm{F}|𝒌|)^2}{2\mathrm{\Delta }},$$
(23)
and the bands depend quadratically on the wave vector and we can define an effective mass $`m^{}=\mathrm{\Delta }/v_\mathrm{F}^2`$. Hence, the contribution of the kinetic energy to the polarization energy is formally similar to that obtained for an 2D electron gas with parabolic dispersion discussed extensively in the literature. In this case, the spinor wave function becomes:
$$\mathrm{\Psi }_{𝒌,\sigma }(𝒓)\left(\begin{array}{c}e^{i𝒌𝒓}\\ 0\end{array}\right)\chi _\sigma ,$$
(24)
for the upper sub-band, while the weight of the spinor is concentrated on $`\psi _b`$, Eq.(10), for the lower sub-band. This change modifies significantly the spinor overlap factor in the calculation of the exchange integral, Eq.(15). The overlap between Bloch states in different bands for momenta near the Fermi points vanishes (see the discussion at the end of Section III.A.1). These states do not give rise to inter-band contributions. The only inter-band contributions which need to be included are due to interactions between states far from the chemical potential among themselves, and between these states at the bottom of the lower band and those at the Fermi level. These terms are not modified when the system is polarized, and they do not contribute to the exchange instability. The remaining intraband term is equivalent to that derived for the electron gas with parabolic dispersion relation. The change in energy when the polarized state is formed can be written as
$`\mathrm{\Delta }E`$ $`=`$ $`\mathrm{\Delta }K+\mathrm{\Delta }E_{\mathrm{ex}}={\displaystyle \frac{A_0}{8\pi }}{\displaystyle \frac{v_\mathrm{F}^2}{2\mathrm{\Delta }}}(k_{}^4+k_{}^42k_\mathrm{F}^4)`$ (25)
$``$ $`{\displaystyle \frac{A_0}{(2\pi )^2}}{\displaystyle \frac{e^2}{ϵ_0}}{\displaystyle \frac{4}{3}}(k_{}^3+k_{}^32k_\mathrm{F}^3),`$
As in the usual case of the 2D electron gas, the system shows an instability toward a ferromagnetic state when $`k_\mathrm{F}(16\mathrm{\Delta }e^2)/(\pi v_\mathrm{F}^2ϵ_0)`$. In agreement with the previous discussion, this instability vanishes when $`\mathrm{\Delta }0`$.
### III.3 The effect of disorder
We approximate the effects of disorder on the average electronic structure by means of the CPA Soven (1967). This approximation describes the effects of disorder on the electronic structure by means of a local self energy, $`\mathrm{\Sigma }(\omega )`$ which is calculated self consistently. While CPA cannot describe localization effects, it still gives very good results for the physical properties of graphene nun .
The total energy, including the exchange contribution, can be expressed in terms of single particle Green’s functions, which are calculated within the CPA. The main steps of the calculation are sketched in Appendix B. We assume that the disorder is induced by vacancies, as likely to occur in samples treated by proton bombardment. The amount of disorder is parametrized by the concentration o vacancies, $`n_{\mathrm{vac}}`$. The CPA leads to a density of states which is finite at $`\omega =0`$, and decays for $`\omega v_\mathrm{F}n_{\mathrm{vac}}^{1/2}`$nun .
Assuming that $`lim_{\omega 0}\mathrm{Im}\mathrm{\Sigma }(\omega )=\mathrm{\Sigma }_0(\mathrm{}v_\mathrm{F})/l`$, where $`l`$ is the average distance between vacancies nun the calculations in Appendix B admit some simplifications. If the concentration of vacancies is small, $`\mathrm{\Sigma }_0\mathrm{}v_\mathrm{F}|k_c|`$. At large energies the CPA result vanishes quite fast as a function of energy, $`lim_{\omega \pm \mathrm{}v_\mathrm{F}|k_c|}\mathrm{\Sigma }(\omega )=0`$. Disorder only changes significantly the results obtained for a clean plane if $`ϵ_\mathrm{F}\mathrm{\Sigma }_0`$. This regime corresponds to electronic densities such that $`|n|n_0=(\mathrm{\Sigma }_0/\mathrm{}v_\mathrm{F})^2/2\pi `$.
In this limit, we can approximately write
$$n_\stackrel{}{𝐤}^\pm \{\begin{array}{cc}0\hfill & \hfill \mathrm{}v_\mathrm{F}|𝒌|\mathrm{\Sigma }_0,\\ 1/2+\frac{ϵ_\mathrm{F}}{\pi \mathrm{\Sigma }_0}\hfill & \hfill \mathrm{}v_\mathrm{F}|𝒌|\mathrm{\Sigma }_0,\end{array}.$$
(26)
where the $`\pm `$ index refers to the two subbands of the noninteracting system (see Appendix B).
The total density of carriers is obtained by integrating this expression over $`\stackrel{}{𝐤}`$ (see Appendix B). Finally, we can also calculate the density of states per unit area and unit energy, which, for $`|\omega |\mathrm{\Sigma }_0`$, becomes a constant:
$$D(\omega )=D_0\frac{1}{2\pi }\frac{\mathrm{\Sigma }_0}{v_\mathrm{F}^2}\mathrm{log}\left(\frac{\mathrm{}v_\mathrm{F}k_c}{\mathrm{\Sigma }_0}\right),|\omega |,|ϵ_\mathrm{F}|\mathrm{\Sigma }_0,$$
(27)
A constant density of states implies that the total number of carriers scales as $`nD_0ϵ_\mathrm{F}`$, instead of the relation $`nϵ_\mathrm{F}^2`$ obtained for the clean system.
From equations (26) and (27) we can infer that both the kinetic energy and the exchange energy depend quadratically on the density of carriers, since $`K(n)K(0)`$ and $`E_{\mathrm{exch}}(n)E_{\mathrm{exch}}(0)`$ scale as $`ϵ_\mathrm{F}^2(n)n^2`$. In addition, we know that for $`nn_0`$ the values of $`K(n)`$ and $`E_{\mathrm{exch}}(n)`$ should be comparable to those obtained in the absence of disorder. Then, we can write:
$`K(n)`$ $``$ $`c_{\mathrm{kin}}{\displaystyle \frac{2A_0\mathrm{\Sigma }_0^3}{3\pi \mathrm{}^2v_\mathrm{F}^2}}\left({\displaystyle \frac{n}{n_0}}\right)^2,`$
$`E_{\mathrm{exch}}(n)`$ $``$ $`c_{\mathrm{exch}}{\displaystyle \frac{A_0e^2\mathrm{\Sigma }_0^3}{3\pi ^2ϵ_0\mathrm{}^3v_\mathrm{F}^3}}\left({\displaystyle \frac{n}{n_0}}\right)^2,`$ (28)
where $`c_{\mathrm{kin}}`$ and $`c_{\mathrm{exch}}`$ are numerical constants of order unity. In a spin polarized system, we have:
$$E_{\mathrm{tot}}(n,m)=\frac{1}{2}\left[K(n+m)+K(nm)+E_{\mathrm{exch}}(n+m)+E_{\mathrm{exch}}(nm)\right],$$
(29)
so that:
$$\mathrm{\Delta }E=\mathrm{\Delta }K+\mathrm{\Delta }E_{\mathrm{exch}}=c_{\mathrm{kin}}\frac{2A_0\mathrm{\Sigma }_0^3}{3\pi \mathrm{}^2v_\mathrm{F}^2}\left(\frac{m}{n_0}\right)^2c_{\mathrm{exch}}\frac{A_0e^2\mathrm{\Sigma }_0^3}{3\pi ^2ϵ_0\mathrm{}^3v_\mathrm{F}^3}\left(\frac{m}{n_0}\right)^2.$$
(30)
The ferromagnetic phase is stable provided that:
$$g_{c,\mathrm{disorder}}=\frac{e^2}{ϵ_0\mathrm{}v_\mathrm{F}}>\frac{2\pi c_{\mathrm{kin}}}{c_{\mathrm{exch}}},$$
(31)
This result implies that, if $`nn_0`$ the critical coupling is independent of the amount of disorder.
We have estimated the ratio $`c_{\mathrm{exch}}/c_{\mathrm{kin}}`$ performing numerically the calculation described in Appendix B for suficiently low carrier concentration and density of vacancies. We find:
$$g_{c,\mathrm{disorder}}=\frac{e^2}{ϵ_0\mathrm{}v_\mathrm{F}}3.8,$$
(32)
indicating that in the case of disorder ferromagnetism is stabilized at a smaller value of the Coulomb interaction. Thus, we can conclude that, at least in CPA, ferromagnetism will be enhanced when disorder is present, in agreement with the experimental data Esquinazi et al. (2003); dis .
The enhancement of the tendency towards ferromagnetism in the presence of disorder is due to the increase in the density of states at low energies. The existence of these states implies that a finite polarization can be achieved with a smaller cost in kinetic energy, in a qualitatively similar way to the Stoner criterium which explains itinerant ferromagnetism in the presence of short range interactions.
## IV Discussion and Conclusions
We have analyzed the ferromagnetic instabilities induced by the exchange interaction in a system where the electronic structure can be approximated by the 2D Dirac equation, as it is the case for isolated graphene planes.
In pure graphene we have found that, as a function of doping, a ferromagnetic transition is possible when the coupling constant is sufficiently large. Our findings are summarized in the zero temperature phase diagram presented in Fig. 1. In this figure we represent the critical coupling $`g_c`$ as function of the doping $`n`$. There are two different regions in the phase diagram. For small doping, $`n<0.2`$ the transition is first order, leading to a ferromagnetic phase with spin polarization $`m=22n`$ and two types of carriers (electrons and holes). For doping larger than $`n>0.2`$ the transition becomes of second order with a magnetization smaller than the doping $`n`$ and one type of carrier (electrons or holes). The connection between the magnetization and the carrier type is unique to the Dirac fermion problem. We should emphasize that our calculation for the Dirac fermion problem is at the same level of the one performed by Bloch, and therefore it is to be expected that an exact solution of this problem will modify quantitatively the phase diagram analyzed here. It is also worth remarking that the electronic structure shown in panel of (c) Fig. 2 shows that, in the ferromagnetic phase, a nominally half filled system has electron and hole pockets. The existence of these pockets does not depend on the presence of intarlayer coherence, however
We have also analyzed the effect of the exchange interaction in disordered systems using the CPA. A continuous transition into a ferromagnetic phase is possible, and the coupling required for its existence is reduced with respect to the clean case. This tendency can be qualitatively explained by noting that the disorder leads to an increase of the density of states at low energies, making the system more polarizable. This explanation is rather general, and it should not depend on the way the effects of disorder are approximated.
Finally, one would ask how our results can be translated for the experiments in disordered graphite Esquinazi et al. (2003); dis . If we naively think of graphite as a stacking of isolated graphene planes we can estimate the value of the coupling constant for graphite to be $`g2.8`$ (for $`ϵ_01`$) Brandt et al. (1988), and therefore far away from the ferromagnetic region (corresponding to the dotted line in Fig.1). The presence of disorder will definitely bring the value of the critical coupling to lower values and according to our calculations $`g_{c,\mathrm{disorder}}3.8`$ would put dirty graphite at the borderline of a ferromagnetic instability.
Nevertheless, the picture of graphene as a non-interacting stacking of graphene planes is certainly incorrect. Because of the absence of screening, long-range forces will play a major role, and the graphene planes will interact via van der Waals interactions. The problem of ferromagnetism in graphite still depends on the better understanding of the coupling between graphene planes. More work has to be developed in order to understand the problem of ferromagnetism in graphite. In any case, our results here are valid for single graphene planes and it would be very interesting to investigate whether graphitic devices Novoselov et al. (2004); Zhang et al. (2004, 2005); Berger et al. (2004) studied recently can sustain any form of ferromagnetism.
## V Acknowledgments
N.M.R.P and F. G. are thankful to the Quantum Condensed Matter Visitor’s Program at Boston University. A.H.C.N. was partially supported through NSF grant DMR-0343790. N. M. R. Peres would like to thank Fundação para a Ciência e Tecnologia for a sabbatical grant partially supporting his sabbatical leave.
## Appendix A Calculation of the exchange integral
The three dimensional integral in Eq.(15) can be written as a combination of integrals of the form:
$$R_n(a)=_0^{2\pi }𝑑\alpha _0^1x𝑑x_0^1y𝑑y\frac{\mathrm{sign}(n)(1)^n\mathrm{cos}\alpha }{\sqrt{x^2+y^2a^22xya\mathrm{cos}\alpha }},$$
(33)
where: $`n=0,1,2`$, $`\mathrm{sign}(n)`$ gives the sign of $`n`$ and $`\mathrm{sign}(0)=0`$. The values of the functions $`R_n(a)`$, for $`a=0`$, are $`R_0(0)=0`$ and $`R_1(0)=R_2(0)=\pi `$. We also have:
$`R_0(1)`$ $`=`$ $`{\displaystyle \frac{2}{3}}\left(2+\pi (\mathrm{ln}2+1/2)+4𝒞\pi (1+\mathrm{ln}4)/2\right)`$
$``$ $`1.109,`$
$`R_1(1)`$ $`=`$ $`8/3+R_0(1)3.776,`$ (34)
where $`𝒞0.915966`$ is the Catalan constant.
Assuming that $`0a1`$ we define:
$$R_n(a)=_0^{2\pi }𝑑\alpha [\mathrm{sign}(n)1(1)^n\mathrm{cos}\alpha ]K(\alpha ,a),$$
(35)
where $`K(\alpha ,a)`$ is given by:
$`K(\alpha ,a)`$ $`=`$ $`{\displaystyle \frac{1}{3a^2}}[(1+a^3)+(1+a^2)\sqrt{1+a^22a\mathrm{cos}\alpha }(1+a^3)\mathrm{cos}\alpha \mathrm{ln}(1\mathrm{cos}\alpha )a^3\mathrm{cos}\alpha \mathrm{ln}a`$ (36)
$`+`$ $`\mathrm{cos}\alpha \mathrm{ln}(a\mathrm{cos}\alpha +\sqrt{1+a^22a\mathrm{cos}\alpha })a^3\mathrm{cos}\alpha \mathrm{ln}(1a\mathrm{cos}\alpha +\sqrt{1+a^22a\mathrm{cos}\alpha })]`$
This expression allows us to obtain the expansions:
$`R_0(a)`$ $``$ $`{\displaystyle \frac{\pi }{3}}[a\mathrm{ln}a+S_0(a)],`$ (37)
$`R_n(a)`$ $``$ $`{\displaystyle \frac{\pi }{3}}[3+(1)^na\mathrm{ln}a+S_n(a)],`$ (38)
for $`n=1,2`$ and
$`S_0(a)`$ $`=`$ $`\left(2\mathrm{ln}2{\displaystyle \frac{1}{6}}\right)a{\displaystyle \frac{9}{80}}a^3{\displaystyle \frac{45}{1792}}a^5`$ (39)
$``$ $`{\displaystyle \frac{175}{18432}}a^7,`$
$`S_n(a)`$ $`=`$ $`(1)^n\left(2\mathrm{ln}2{\displaystyle \frac{1}{6}}\right)a{\displaystyle \frac{3}{8}}a^2+(1)^n{\displaystyle \frac{9}{80}}a^3`$ (40)
$``$ $`{\displaystyle \frac{3}{64}}a^4+(1)^n{\displaystyle \frac{45}{1792}}a^5{\displaystyle \frac{15}{1024}}a^6`$
$`+`$ $`(1)^n{\displaystyle \frac{175}{18432}}a^7.`$
Note that $`R_1(a)R_2(a)=2R_0(a)`$ is always satisfied.
## Appendix B Calculation of the exchange energy in the presence of disorder.
We write the one-electron energies in the absence of interactions and disorder as:
$$ϵ_𝒌^\pm =\pm v_\mathrm{F}(𝒌),$$
(41)
up to some cutoff $`k_c`$, where the two signs correspond to the two bands in the electronic spectrum. Using the CPA, the one electron Green’s function can be written as:
$$G^\pm (𝒌,\omega )=\frac{1}{\omega \mathrm{\Sigma }(\omega )ϵ_𝒌^\pm }.$$
(42)
The occupancy of a given state at fixed chemical potential, $`ϵ_\mathrm{F}`$, is:
$$n_𝒌^\pm =_{\omega _c}^{ϵ_\mathrm{F}}\frac{1}{\pi }\mathrm{Im}G^\pm (𝒌,\omega )𝑑\omega ,$$
(43)
where a frequency cutoff, $`\omega _c`$ is also defined.
The total number o electrons, $`n`$, and the kinetic energy can be written as:
$`n`$ $`=`$ $`{\displaystyle \underset{\alpha =\pm }{}}{\displaystyle \frac{2}{\pi }}{\displaystyle _0^{k_c}}n_{|𝒌|}^\alpha k𝑑k,`$
$`K`$ $`=`$ $`{\displaystyle \underset{\alpha =\pm }{}}{\displaystyle \frac{2}{\pi }}{\displaystyle _0^{k_c}}\alpha ϵ_{|𝒌|}n_{|𝒌|}^\alpha k𝑑k.`$ (44)
These one-dimensional integrals are calculated numerically. Finally, the exchange energy is:
$$E_{\mathrm{exch}}=\frac{e^2}{4\pi ^4}d^2𝒌_1d^2𝒌_2\frac{[n^+(𝒌_1)+n^{}(𝒌_2)]^2+[n^+(𝒌_1)n^{}(𝒌_2)]^2\mathrm{cos}[\varphi (𝒌_1)\varphi (𝒌_2)]}{|𝒌_1𝒌_2|},$$
(45)
and:
$$\varphi (𝒌)=\mathrm{arctan}\left(\frac{k_y}{k_x}\right).$$
(46)
This expression can be reduced to a three-dimensional integral, which is calculated numerically.
The total energy, $`E_{\mathrm{tot}}(n)=K(n)+E_{exch}(n)`$, can be written as:
$$E_{\mathrm{tot}}(n)=E_{\mathrm{tot}}(n_{})+E_{\mathrm{tot}}(n_{}).$$
(47)
The exchange instability towards ferromagnetism implies that:
$$E_{\mathrm{tot}}(n/2\delta n)+E_{\mathrm{tot}}(n/2+\delta n)<2E_{\mathrm{tot}}(n/2),$$
(48)
so that:
$$\frac{^2E_{\mathrm{tot}}}{n^2}|_{n/2}<0.$$
(49) |
warning/0507/gr-qc0507090.html | ar5iv | text | # Conformally Flat Metric, Position-Dependent Mass and Cold Dark Matter
## 1 Introduction
The maximal acceleration hypothesis was first conjectured by Caianiello . Different aspects, formulation and inferences concerning possible existence of the limiting value for the proper acceleration of a particle have been considered in various works on the classical and quantum bases (see, for example and the references therein). Despite the fact that there are many arguments supporting the existence of MA, its actual status is still open to dispute. Specifically, it is not clear whether the numerical value of MA should be considered as a universal constant, similar to the speed of light, or as a parameter depending on the individual mass due to the action of force (Mass-Dependent Maximal Acceleration - MDMA). It was shown that the effective conformally flat metric can arise directly from the existence of maximal acceleration . This metric reveals interesting confinement aspects. Namely, the Lorentz-scalar potential and damping normalization factor introduced in particular relativistic phenomenological quarkonium models to provide quark confinement occur (in this case) purely geometrically as a consequence of the existence of the effective conformally flat static metric . As it was first shown by A.K.Gorbatsevich and L.M.Tomilchik (see also -), and somewhat later independently by M.Gasperini , such a metric leads to the appearance of the coordinate-dependent rest mass. The position-dependent mass was introduced by several authors in nonrelativistic quantum mechanical models developed to study the electronic properties of condensed media (see, for example ), including rather interesting attempts of pure geometrical interpretation of such a dependence via the constant curvature space .
The apparent efficiency of the quantum-mechanical models based on the ad hoc hypothesis of a coordinate-dependent rest mass suggests that this concept has universal nature. Therefore, it seems natural to extend it to the classical (nonquantum) objects as well as to find the general principles that could naturally lead to such a dependence. It is likely that a search for kinematic restrictions involving the maximal-acceleration hypothesis has considerable promise.
In any case, it is natural to expect that the observable effects possible due to the existence of MA should appear already at a level of the point-like particle classical dynamics.
In the present paper new arguments in favor of MDMA are put forward. The Lagrangian and Hamiltonian classical dynamics of a point-like particle with the coordinate-dependent mass is considered. Under the special choice of such dependence, the effective Lagrangian for the pure gravitational interaction is proposed. It is shown that within this model the peculiar form of the corresponding rotation curve is as a whole reproduced without the use of the cold dark matter concept. It is demonstrated that the canonical quantization of this model leads directly to the Dirac oscillator model for a particle with Plank’s mass.
## 2 Some Heuristic Considerations
It is well known that the existence of a maximal transmission velocity for the signal synchronizing clocks separated spatially in each fixed inertial reference frame leads to the appearance of four-dimensional space-time with the pseudo-Euclidean structure.
But according to Einstein and Poincare, the procedure of clock synchronization suggests the transmission of an instantaneous signal, i.e. a signal whose duration can be made as small as is wished. On the other hand, a synchronizing signal can be transmitted and accepted only due to the exchange of any finite portion of energy $`\delta E`$ (and hence momentum $`\delta p`$). If $`E`$ and $`p`$ are certain time-dependent functions, we have evident relations
$$\delta E=\frac{dE}{dt}\delta t,\delta p=\frac{dp}{dt}\delta t$$
relating the formation duration of a synchronizing signal ($`\delta t`$) to the values of the energy and momentum carried by the transmitted signal. In principle, the quantities $`\delta E`$ and $`\delta p`$ should be finite, whereas the interval $`\delta t`$ should always tend to zero, implicitly suggesting fulfillment of the conditions
$$\underset{\delta t0}{lim}\frac{dE}{dt}=\underset{\delta t0}{lim}\frac{dp}{dt}=\mathrm{}.$$
(1)
However, if the quantities $`(\frac{dE}{dt})_{lim}`$ and $`(\frac{dp}{dt})_{lim}`$, respectively representing maximal power and maximal force, are assumed to be limiting, a nonzero extra time retardation occurs:
$$\delta t_{extra}=(\frac{dE}{dt})_{lim}^1\delta E,(or(\frac{dp}{dt})_{lim}^1\delta p)$$
that should be taken into account in clock synchronization. It is clear that the smaller the distance between the synchronized clocks the more evident the corrections (recall that the Einstein clock is point-like by definition). Basing on this reasoning, it is suggested that there exists a constant $`\kappa _0`$ representing the upper limit for possible values of the proper energy changing with time . In this paper, it has been proposed to combine the infinitesimal intervals of the Minkowski space $`ds`$ and momentum space $`dp`$ in accordance with the Born Reciprocity Principle :
$$dS^2=ds^2+\frac{1}{\kappa _0^2}dp^2=dx^\mu dx_\mu +\frac{1}{\kappa _0^2}dp^\mu dp_\mu .$$
(2)
Using the customary definition $`ds=cd\tau `$ ($`\tau `$ is the proper time), we obtain
$$dS^2=ds^2\{1+\frac{1}{F_0^2}\frac{dp^\mu }{d\tau }\frac{dp_\mu }{d\tau }\}$$
(3)
where the parameter $`F_0=\kappa _0c`$ is a constant with the dimension of force. The model proposed by Caianiello and his coworkers (see ,, ) to include the effects of maximal acceleration into the particle dynamics consisted in extension of the space - time manifold to the eight-dimensional space - time tangent bundle.
The fundamental infinitesimal interval for a particle is represented by the following eight-dimensional line element:
$$dS^2=dx^\mu dx_\mu +\frac{c^2}{A^2}d\dot{x}^\mu d\dot{x}_\mu $$
(4)
where $`\dot{x}_\mu =\frac{dx_\mu }{d\tau }`$. $`A`$ is a parameter with the dimension of acceleration. Assuming the Minkowski background metric, i.e. $`g_{\mu \nu }=\eta _{\mu \nu }=diag(1,1,1,1)`$, and taking into account that $`d\dot{x}^\mu =d\ddot{x}^\mu d\tau `$, we obtain from (4) that
$$dS^2=\left(1+\frac{\ddot{x}^\mu \ddot{x}_\mu }{A^2}\right)ds^2.$$
(5)
Here $`\ddot{x}^\mu =\frac{d^2x^\mu }{d\tau ^2}`$, $`ds^2=dx^\mu dx_\mu =c^2d\tau ^2`$, and $`\ddot{x}^\mu `$ is the space-like vector (i.e. $`\ddot{x}^\mu \ddot{x}_\mu <0`$). The explicit form of $`\ddot{x}^\mu \ddot{x}_\mu `$ in the noncovariant notation is
$$\ddot{x}^\mu \ddot{x}_\mu =\gamma ^3\left\{\underset{¯}{W}^2\frac{1}{c^2}(\underset{¯}{W},\underset{¯}{V})^2\right\}$$
(6)
where $`\underset{¯}{\beta }=\frac{d\underset{¯}{r}}{cdt}=\frac{\underset{¯}{\overset{\dot{}}{r}}}{c}`$, $`\underset{¯}{W}=\frac{d^2\underset{¯}{r}}{d^2t}=\underset{¯}{\overset{¨}{r}}`$, $`\gamma =(1\underset{¯}{\beta }^2)^{\frac{1}{2}}`$ . In case of $`\dot{r}=0`$ we obtain from (5), (6) the formula
$$dS^2=c^2d\tau ^2\left(1\frac{\underset{¯}{W}^2}{A^2}\right)$$
(7)
demonstrating the limiting role of $`A`$. Taking the interval (3) for a material point, i.e. assuming $`\frac{dE}{dt}=\frac{1}{c}\left(\underset{¯}{V}\frac{d\underset{¯}{p}}{dt}\right)`$, we obtain the expression
$$dS^2=ds^2\left\{1\left(1\frac{\underset{¯}{\overset{\dot{}}{r}}^2}{c^2}\right)^1\frac{1}{F_o^2}\left(\frac{d\underset{¯}{p}}{dt}\right)^2\left(1\frac{\underset{¯}{\overset{\dot{}}{r}}^2}{c^2}\mathrm{cos}^2\varphi \right)\right\}$$
(8)
where $`\varphi `$ is an angle between $`\underset{¯}{\overset{\dot{}}{r}}`$ and $`\underset{¯}{\overset{\dot{}}{p}}=\frac{d\underset{¯}{p}}{dt}`$. From this formula it follows that using the same assumptions as in the derivation of relation (7) and with $`\frac{d\underset{¯}{p}}{dt}=\underset{¯}{f}`$ we obtain the expression
$$dS=cd\tau \left(1\frac{f^2}{F_o^2}\right)^{1/2}$$
(9)
from whence the constant $`F_o`$ plays the limiting role. It is natural to identify this constant as maximal force (MF). We assume that this constant has purely classical nature, i.e. it is not related to a minimum value of the action $`a_{min}=\mathrm{}`$ and represents the inverse of the Einstein gravitation constant, being numerically determined as
$$F_o=\frac{c^4}{G}.$$
(10)
Postulating this universal constant as $`\kappa _o(F_o=c\kappa _o)`$, we come to the representation of the mass-dependent maximal acceleration (MDMA)
$$\left(\frac{dV}{dt}\right)_{max}=W_{max}=A=\frac{F_o}{m}=\frac{c^4}{mG}=\frac{2c^2}{r_g(m)}$$
(11)
where $`r_g(m)=\frac{2mG}{c^2}`$ is the gravitation radius corresponding to the (point-like!) mass $`m`$.
It is seen from (10) and (11 that MDMA has an obvious classical, Newtonian meaning of the centripetal acceleration of a test point-like particle rotating uniformly in a circle whose radius is equal to the ”radius” of the Schwarzschild sphere.
Needless to say that this descriptive pattern should not be considered literally. Nevertheless, the idea that the Schwarzschild parameter $`r_g`$ is related to the MDMA scheme holds much promise. A model for hyperbolic motion of the point-like mass in the Special Relativity also indicates the existence of this relation. Here, as it is known, the corresponding world line is given by the equation $`\underset{¯}{r}^2c^2t^2=r_0^2`$, where $`r_0`$ is a fixed parameter having the dimension of length. If the initial velocity is equal to zero, for the absolute value of the three-dimensional acceleration $`\underset{¯}{W}`$ we obtain the expression $`W={\displaystyle \frac{c^2}{r_0}}\left(1+\left({\displaystyle \frac{ct}{r_0}}\right)^2\right)^{\frac{3}{2}}`$, from which it is seen that the quantity $`W_0={\displaystyle \frac{c^2}{r_0}}`$ represents acceleration at the initial instant of time, i.e. it is the MDMA for the given mass $`m`$. If $`W_0`$ is determined by formula (11), the parameter $`r_0`$ is described as $`r_0={\displaystyle \frac{mG}{c^2}}=\frac{1}{2}r_g`$.
It may be inferred intuitively that, because of the existence of maximal acceleration, deviations of the mechanical motion of a point-like particle from the standard dynamics should manifest themselves under conditions when its acceleration tends to $`W_0`$ (or is comparable to it). It is clear that in case when $`A`$ is determined by expression (11), the ”kinematic” manifestations of MDMA will be more evident for the material points with greater mass. Let us estimate the quantity $`\frac{W}{A}`$ for a case of electromagnetic interactions. To this end, we use the expression for the force of the static Coulomb interaction of two charges $`e`$ and $`Ze`$ (i. e. $`mW=\frac{e^2Z}{r^2}`$ ). In this case the expression is as follows:
$$\frac{W}{A}=\frac{f}{F_o}=\frac{e^2}{r^2}\frac{Z}{F_o}\frac{e^2G}{c^4}\frac{Z}{r^2}=\frac{r_o^2}{r^2}$$
where $`r_o^2=\frac{e^2GZ}{C^4}`$ and hence the corrections to unity in (9) will be equal to $`(\frac{r_o}{r})^4`$.
As it is seen, $`r_o^2\frac{1}{2}r_cr_g`$, where $`r_c=\frac{e^2}{mc^2}`$, $`r_g=\frac{2mG}{c^2}`$ are the classical and gravitation radii of the point-like charged mass for the interaction of two equal charges .
For all the elementary particles, the quantity $`r_o`$ vanishes, being equal to $`r_o(10^{68})^{\frac{1}{2}}10^{34}`$ cm.
Needless to say that such reasoning gives only a rough estimate, since the use of the classical parameters of electromagnetic interaction is inadequate in real situation ( it is well known that for such interactions the quantum effects become significant at distances on the order of the atomic dimensions). Nevertheless, a fixed value of the parameter $`r_0`$ ($`r_0`$ is of the order of Plank’s length) points to the fact that the effects related to MDMA can influence the known elementary particles only at distances comparable to Plank’s length (i.e. at energies of $`10^{19}`$Gev) associated with the Plank energy scales. The same may be valid for strong interactions too.
A distinct situation is observed in case of gravitation interactions.
First, note that the concepts of force and acceleration are used in modern physics only in the classical context. This being so, it is not imperative, in our opinion, to relate the numerical value of maximal acceleration to the constant $`\mathrm{}`$, as this is made commonly (see ). The existence of the maximal force interpreted purely classically, however, should exclude a fall at the center for the pattern of mutual attraction of two point-like particles, or should lead to the appearance of some effective repulsion. From this standpoint, the situation is qualitatively similar to the well-known effects of quantum dynamics caused by the Plank’s quantum of action and hence noncommutativity of the canonically conjugate coordinates and momenta.
To illustrate the ”repulsion” effect caused by the classical maximal force, we use the elementary concepts based on the Newtonian law of gravitation. If there exists the above-postulated maximal force, there should exist a minimum distance $`r_0`$ between two attracting point-like masses. This distance may be determined from the condition
$$\frac{mMG}{r_0^2}=\frac{c^4}{G}$$
(12)
whence the expression for parameter $`r_0`$ may be found
$$r_0^2=\frac{1}{4}r_gR_g$$
(13)
where $`r_g,R_g`$ are the corresponding Schwarzschild radii.
This result correlates well with the conclusion that the Schwarzschild sphere is principally impenetrable for the test classical particle obtained in on the basis of the solution of the motion problem in the Schwarzschild field with regard to maximal acceleration.
It is obvious that the absence of fall at the center in such a two-particle problem means that there exists some nonzero angular momentum. Actually, from the standpoint of a ”naive” model, a minimum distance, where the centers of two small balls with masses $`m`$ and $`M`$ and hence radii $`r_g=\frac{2mG}{c^2},R_g=\frac{2MG}{c^2}`$ can come of each other, is equal to $`r_g+R_g`$. We easily calculate that the moment of inertia of such a system with respect to its center of mass (disregarding the proper rotation of the balls) is equal to
$$I_0=(m+M)r_gR_g.$$
(14)
For such a system there should exist a maximal frequency of rotation around the center of mass. An approximate estimation of the quantity $`\omega _{max}`$ within the scope of the model considered gives the expression
$$\omega _{max}=\frac{c}{R_g}.$$
(15)
Proceeding from this expression, for a minimal value of the angular momentum we obtain
$$L_{min}=I_0\omega _{max}.$$
(16)
Actually, this nonzero angular momentum cannot be attributed to either of the two particles but belongs to the system as a whole. The situation is similar to Extra Spin in the electric charge - magnetic monopole system.
Now consider the applicability of the model under study to classical systems. Apart from such a fairly perfect classical theory of gravitation as the Standard General Relativity, there is quite a number of purely gravitational, comparatively isolated systems, the observed mechanical properties of which may be described in reality on the basis of the models correlating with the ordinary Newton approximation. First, consider rotation of the most abundant, typical spiral galaxies, in a sense, belonging to the simplest astrophysical objects. There is a reason to believe that such systems may be considered within the framework of the Newton approximation: the observed intragalaxy velocities do not exceed several thousandth of the speed of light, and the corresponding Schwarzschild radius measures fraction of one parsec (this value is by a factor of $`10^410^5`$ smaller than the characteristic dimensions of a galaxy, even having a mass on the order of $`10^{12}`$ of the Sun mass). Therefore, modeling of the observed rotation of stars around the galactic center by the nonrelativistic movement of a material point in a centrally-symmetric field, representing a combination of the Newton attraction and gravitational attraction potential linearly increasing with the distance, seems to be wholly warranted.
At the same time, the observed asymptotic behavior of the rotational curves for the typical spiral galaxies is in drastic contradiction with such a theoretical representation. The most popular idea that should eliminate this contradiction is currently associated with the cold-dark-matter concept. Alternative explanation schemes based on the attempts to modify the Newton model of gravitational interactions (see and the literature therein) are also available. In our opinion, using the coordinate dependence of the rest mass offers great promise here. As it will be shown below, the use of a simple classical model enables one to reproduce the general shape of the rotational curve for the typical spiral galaxy practically over the whole range of distances from its center.
When using the conception of phase space and Hamiltonian dynamics, it is convenient to introduce, a la M.Born , the parameters $`q_0`$, $`p_0`$ having the dimensions of length and momentum, respectively. For each specific physical system with a finite action these parameters are assumed to be determined by the relations $`q_0p_0=a,p_0/q_0=\kappa _0`$ ,where $`\kappa _0`$ is a universal constant,($`\kappa _0={\displaystyle \frac{c^3}{G}}`$),and $`a`$ is the parameter with the dimension of action determined for each classical or quantum system. Its minimum value is equal to the Planck universal constant $`\mathrm{}`$. Thus, the following definitions are true for the parameters: $`q_0=(a\kappa _0^1)^{\frac{1}{2}},p_0=(a\kappa _0)^{\frac{1}{2}}`$.
It is evident that for $`a_{min}=\mathrm{}`$ the parameters $`q_0`$, $`p_0`$ are equal to the corresponding Planck’s quantities $`l_p`$, $`p_p`$. And the ratio $`{\displaystyle \frac{p_0}{q_0}}`$ is independent of $`a`$, having the same value both for classical and quantum systems <sup>1</sup><sup>1</sup>1Note that from the geometrical standpoint, the constant $`\kappa _0=p_0/q_0`$ determines maximal deformation (”Prokrust strain”) of a given phase area (including an elementary phase cell)..
Let us consider the relation between maximal acceleration and conformal transformation.
It is well known that attempts to interpret the special conformal transformation (SCT) in the Minkowski Space were made even by L.Page and N.I.Adams (see and papers cited there). This transformation may be written in the following form:
$$x^\mu =\sigma (x,b)(x^\mu +b^\mu x^2)$$
(17)
where
$$\sigma (x,b)=(1+2bx+b^2x^2)^1,$$
$$bx=b^\mu x_\mu =b_\mu x^\mu ,$$
$`b^\mu `$ is the four-vector parameter with the dimension of $`(length)^1`$. The parameter $`b^\mu `$ is traditionally related to the constant relative four-dimensional acceleration of the reference frame.
Using this interpretation, we write $`b^\mu =c^2A^\mu `$ , where $`A^\mu `$ denotes the constant relative four-dimensional acceleration. Besides, it is required that the obtained expression be coincident, in the limit, with the path formula for the uniformly accelerated movement along the $`x`$ axis. The vector $`b^\mu `$ should be space-like.
Let us write $`b^\mu `$ in the following form:
$$b^\mu =\{0,c^2W,0,0\}$$
(18)
where $`W`$ is the $`x`$ \- component of the three-dimensional acceleration.
For world lines of the point-like particles only the interior of the light cone is available. Specifically, it is assumed that $`x^\mu =\{ct,0,0,0\}`$. Then we obtain
$$x^2=c^2t^2,bx=0,b^2=\frac{w^2}{4c^4},\sigma (x,b)=1\left(\frac{Wt}{2c}\right)^2.$$
Consequently, transformations (17) take the following form:
$$x^\mu =\frac{Wt^2}{2}\left(1\left(\frac{Wt}{2c}\right)^2\right)^1,t^{}=t\left(1\left(\frac{Wt}{2c}\right)^2\right)^1.$$
(19)
If the term $`\left(\frac{Wt}{2c}\right)^2`$ in the denominator is ignored, as it might be expected, we obtain nonrelativistic expressions $`x^{}=\frac{Wt^2}{2}`$, $`t^{}=t`$.
It is seen that higher values of $`\frac{Wt}{2}`$ are limited by $`c`$.
Evidently, the condition $`\frac{Wt}{2c}<1`$ suggests two variants of maximal values for the acceleration and corresponding time interval
$$(a)W_{min}\mathrm{\Delta }t_{max}=2c,$$
$$(b)W_{max}\mathrm{\Delta }t_{min}=2c.$$
It may be conclusively advocated that the existence of maximal acceleration requires the existence of a minimal time interval, and conversely if there is a maximal time interval, there should exist a certain minimal acceleration.
It seems probable that, in reality, both these opportunities should be taken into account. In principle, simultaneous existence of any large and small time intervals is a necessary condition for any model realization of a physical clock per se. To illustrate, consider the circle of a clock dial and the primes on it as well as, in the general case, large and small periods and the possibility of comparing them to a set of integers.
For a uniform and isotropic model of the Universe as a whole there are obvious candidates for $`\mathrm{\Delta }t_{min}`$ and $`\mathrm{\Delta }t_{max}`$: Planck’s time $`\tau _p`$ and the Universe age (inverse of the Hubble constant).
It is interesting that the existence of $`\tau _{max}`$ (in combination with the assumption that there exists a maximal force $`F_o=\frac{c^4}{G}`$ ) should lead to the upper limit of the total action in the metagalaxy (assuming $`\tau _{max}=H_o^1`$)
$$S_{max}=F_oc\tau _{max}^2=\frac{c^5}{GH_o^2}$$
that at a given experimental value of $`H_o^110^{18}s`$ results in $`a_{max}=10^{95}ergs`$.
Since the quantum of action $`a_{min}=\mathrm{}10^{27}ergs`$, the total number of quanta is equal to
$$N10^{122}e^{280}.$$
## 3 Classical Dynamics of a Particle with the Position-Dependent Mass
As it has been shown in (see also -), a geodesic equation in the conformally flat metric
$$g_{\mu \nu }=U^2(x)\eta _{\mu \nu },\eta _{\mu \nu }=diag(1,1,1,1)$$
(20)
for the static case with $`U(x)=U(\underset{¯}{r}),\frac{(U^2)}{t}=0`$ may be written in the following form:
$$\frac{d\underset{¯}{p}}{dt}+\frac{p_0^2}{2m}grad(U^2)=0,\frac{dm}{dt}=0.$$
(21)
Here $`\underset{¯}{p}=m\underset{¯}{\overset{\dot{}}{r}},m=c^1p_0U(\underset{¯}{r})(1(\frac{1}{c}\underset{¯}{\overset{\dot{}}{r}})^2)^{\frac{1}{2}}`$ and $`p_0`$ is a parameter with the dimension of momentum. This equation is formally coincident with a nonrelativistic equation of motion for a ”particle” of ”mass” $`m`$ in a ”potential field”
$$\underset{¯}{f}=\frac{p_0^2}{2m}grad(U^2).$$
(22)
The solution of equation (21) gives an essentially new result: the existence of the peculiar parametric dependence of ”mass” $`m`$ on the initial conditions due to its appearance in (21) as an integral of motion rather than as a numerical constant. The momentum $`\underset{¯}{p}`$ is related to the energy $`E`$ by the standard relation
$$E^2c^2\underset{¯}{p}^2=c^2p_0^2U^2(\underset{¯}{r}),$$
(23)
demonstrating that the quantity $`c^1p_0U(\underset{¯}{r})`$ plays a role of the coordinate-dependent rest mass.
The same result can be obtained using the standard variational procedure with an action determined by the linear element
$$ds=(g_{\mu \nu }dx^\mu dx^\nu )^{1/2}$$
(24)
that is defined by the metric (20). The associated Lagrangian is of the form
$$L=cp_0U(r)(1\frac{\underset{¯}{\overset{\dot{}}{r}}^2}{c^2})^{1/2}.$$
(25)
The form of the corresponding Hamiltonian is as follows:
$$H=c(\underset{¯}{p}^2+p_0^2U(r)^2)^{\frac{1}{2}}.$$
(26)
In centrally symmetric case (apart from the energy $`E`$) there exists a vector integral of motion (angular momentum)
$$\underset{¯}{M}=\underset{¯}{r}\times \underset{¯}{p}=Ec^2\underset{¯}{L},\underset{¯}{L}=\underset{¯}{r}\times \underset{¯}{\overset{\dot{}}{r}}.$$
(27)
In this case, equation of motion (21) takes the form
$$\underset{¯}{\overset{¨}{r}}+\frac{c^2}{2}ϵ^2r^1\frac{dU^2}{dr}\underset{¯}{r}=0$$
(28)
where $`ϵ=(cp_0)^1E=U(r)(1(\frac{1}{c}\underset{¯}{\overset{\dot{}}{r}})^2)^{\frac{1}{2}}`$ is the reduced energy. A motion will be finite when $`U(r)`$ is an increasing function of $`r`$. It is assumed that this function has no singularities throughout the domain of its definition. Since equation (28) is formally coincident with the nonrelativistic dynamics equation for a point-like particle in a centrally symmetric potential field, the motion is investigated in a standard way. In this case the radial velocity $`\dot{r}`$ is determined by the following expression:
$$\dot{r}^2=c^2(1\frac{U^2(r)}{ϵ^2})\frac{\underset{¯}{L}^2}{r^2}$$
(29)
where $`ϵ`$ and $`\underset{¯}{L}`$ are the integrals of motion defined above. The turning points are obtained from the equation $`\dot{r}=0`$. The motion will be finite if the function $`U(r)`$ is such that this equation has two real positive roots and, in exceptional cases, the trajectories are closed.
Consider a particular model. To this end, we choose the expression for the function $`U(r)`$ of the form
$$U(r)=\left(1+\frac{r^2}{R_0^2}\right)^{1/2}$$
(30)
where $`R_0`$ is a parameter having the dimension of length. The reasoning in favor of this choice is heuristic in character:
. the problem is exactly solvable;
. at small $`r(R_0)`$ the form of the potential corresponding to the field of a gravitating mass, continuously distributed with a constant density, is reproduced by $`U^2(r)`$;
. the required asymptotic behavior of the velocity is provided at large $`r(R_0)`$;
. the model is reciprocally symmetric in the sense of M. Born.
Furthermore, we assume that the Lagrangian contains the constant $`\overline{C}`$ representing the asymptotic limit of the velocity of mechanical motion, for a test particle in the closed system considered, rather than the speed of light. In other words, we deal with a model defined by the effective Lagrangian of the form
$$L_{eff}=m_0\overline{C}^2\left(1+\frac{r^2}{R_0^2}\right)^{1/2}\left(1\frac{\underset{¯}{\overset{\dot{}}{r}}^2}{\overline{C}^2}\right)^{1/2}$$
(31)
where $`\overline{C}`$ and $`R_0`$ are experimentally determined parameters. $`m_0`$ is the mass of the test particle that, as will be shown later, disappears in the final result. In this case, the condition $`\dot{r}=0`$ leads to a biquadratic equation
$$r^4R_0^2(ϵ^21)r^2+R_0^2(\overline{C})^2ϵ^2\underset{¯}{L}^2=0$$
(32)
whose solution determines the semiaxes of the elliptic trajectories as follows:
$$r_\pm =\frac{R_0}{\sqrt{2}}(ϵ^21)^{1/2}\{1\pm (14\underset{¯}{L}^2ϵ^2R_0^2(\overline{C})^2(ϵ^21))^2)^{1/2}\}^{1/2}.$$
(33)
The circular orbits correspond to the equality $`r_+=r_{}`$, namely the condition $`R_0^2\overline{C}^2(ϵ^21)^2=4ϵ^2\underset{¯}{L}^2`$ that, considering $`\underset{¯}{L}^2=\underset{¯}{r}^2\times \underset{¯}{\overset{\dot{}}{r}}^2`$, leads to an explicit expression relating the velocity $`v=(\underset{¯}{\overset{\dot{}}{r}}^2)^{1/2}`$ of the circular orbital motion of a point-like particle to the orbital radius
$$v(r)=\overline{C}\left(2+\frac{R_0^2}{r^2}\right)^{1/2}.$$
(34)
As can be seen, the function (34) reproduces remarkably well the general shape of the rotation curve for a spiral galaxy. Figure 1 shows experimental data characteristic for the rotation curve of the NGC 3198 galaxy (this figure has been taken from ).
Figure 2 demonstrates the rotation curve $`v(r)`$ calculated by formula (34) using the following parameters:
$$\overline{C}=212.132km/s,R_0=3.182kpc.$$
(35)
Note that such a numerical value of the parameter $`R_0`$ seems to be reasonable as regards the MDMA concept proposed by us. The problem considered is, to a certain extent, a classical analog of the quantum oscillator problem. As it is known, here the characteristic parameter with the dimension of length is determined as $`x_0^2=\frac{\mathrm{}}{m\omega _0}`$, where $`m`$ is the mass of an oscillating particle and $`\omega _0`$ is the eigenfrequency of the oscillator. In the case under study the angular momentum $`L_{min}`$ determined by formula (16) serves as the minimal action. On the other hand, for the quantity $`\omega _0`$ the following choice seems to be natural in this case: $`\omega _0=(\rho _mG)^{1/2}`$. Here $`\rho _m`$ is a constant (position-independent) mass density determined as $`\rho _mM/R^3`$, where $`M`$ is the total mass of the substance found within a region with the linear dimensions $`R`$. Then, for $`R_0^2=\frac{L_{min}}{m\omega _0}`$, we obtain accurately to the constant on the order of unity
$$R_0=(1+r_g/R_g)^{1/2}(R/R_g)^{3/4}R_g$$
(36)
where $`r_g`$ and $`R_g`$ are the Schwarzschild radii of a galaxy and star, respectively, and $`R`$ is the linear dimension of a galaxy. For a typical galaxy with $`M10^{10}`$ mass of the Sun and $`R10^5pc`$, the value of $`R_0`$ determined from (36) is about one kpc (kiloparsec).
Note that, in the order of magnitude, the product of the empirical parameters $`\overline{C}^2R_0`$ equals to $`c^2R_g`$, where $`c`$ is the speed of light and $`R_g`$ is the Schwarzschild radius corresponding to the galaxy mass. This makes it possible to suppose that there exists some scale invariance necessitating special investigation from the standpoint of the conformally-symmetric dynamics.
## 4 Quantization: Dirac Oscillator Model for Plankeon
Let us show that the quantization procedure based on the function $`U(r)`$ of the form (30) leads directly to the well-known Dirac oscillator model . In this case the operator $`\widehat{H}^2`$ corresponding to the Hamilton function (26) takes the form
$$\widehat{H}^2=E_0^2(\underset{¯}{P}^2+\underset{¯}{\rho }^2+1)$$
(37)
where $`P_k=i\frac{}{\xi _k},\rho _k=\xi _k=\frac{x_k}{q_0},k=1,2,3`$, and $`E_0=E_p=(c^5\mathrm{}G^1)^{1/2},q_0=l_p=(c^3\mathrm{}G)^{1/2}`$ are the corresponding Plank’s quantities.
Standard linearization, in accordance with Dirac’s procedure, gives the following expression(in noncovariant notation):
$$\widehat{H}=E_0\{(\underset{¯}{\overset{^}{\alpha }}\underset{¯}{P})+\widehat{\beta }(1+\underset{¯}{\rho }^2)^{1/2}\}$$
(38)
where $`\underset{¯}{\overset{^}{\alpha }},\widehat{\beta }=\widehat{\rho }_3`$ are the standard Dirac matrices.
The operator $`\widehat{U}^2=1+\underset{¯}{\rho }^2`$ suggests obvious factorization $`\widehat{U}^2=\widehat{U}_+\widehat{U}_{}=\widehat{U}_{}\widehat{U}_+`$, where
$$\widehat{U}_\pm =1\pm i(\underset{¯}{\overset{^}{\alpha }}\underset{¯}{\rho })(\widehat{U}_\pm ^+=\widehat{U}_{})$$
(39)
are normal mutually-conjugate Hermitian operators. In this case $`\widehat{U}_\pm =\gamma _5\widehat{U}_{}\gamma _5`$ (in the given representation,$`\gamma _5=\widehat{\rho }_2`$). Substituting (39) into (38), we obtain two Hermitian operators
$$\widehat{H}_\pm =E_0\{(\underset{¯}{\overset{^}{\alpha }}\underset{¯}{P})\pm i\beta (\underset{¯}{\overset{^}{\alpha }}\underset{¯}{\rho })+\widehat{\beta }\}=E_0\{\underset{¯}{\overset{^}{\alpha }}(\underset{¯}{P}i\widehat{\beta }\underset{¯}{\rho })+\widehat{\beta }\}.$$
(40)
As seen, operators (40) are exactly coincident with Hamiltonian of the Dirac oscillator $`(\widehat{H}_+)`$ and its supersymmetric partner $`(\widehat{H}_{})`$ in the noncovariant representation (see, for example, ). It is significant that in the version being considered the model describes a particle with Plank’s mass $`m_p=(\mathrm{}cG^1)^{1/2}`$. This model will be discussed in the context of the gravity quantization problem in a separate paper.
## 5 Concluding Remarks
In our opinion, the concept of the coordinate-dependent rest mass (CDRM), along with the hypothesis that there exists the mass-dependent maximal acceleration (MDMA), may be effectively used in the field of quantum and classical dynamics. Gravitational interactions represent an area, where the models of the classical Lagrangian and Hamiltonian dynamics with CDRM may be applied. The case in point is the description of an intermediate region requiring no recourse to the strict general relativity and making the use of the Newton approximation insufficient. There is reason to believe that such an intermediate region is due to the rotation of large quasi-stationary cosmological objects, primarily typical spiral galaxies.
It is clear that such phenomenological models should be substantiated from the standpoint of the standard general relativity, necessitating special investigation. On the other hand, the development of such models can help to solve a number of problems of the relativistic cosmology, e.g., the well-known singularity problem. It is interesting that the use of the same phenomenological model makes it possible to give a description of the behavior of the rotation curves representing the typical spiral galaxies, in the classical version, and an exactly solvable Dirac oscillator model for a spinor particle with Plank’s mass, in the quantum version. It is not improbable that this enables construction of a theory for the behavior of fermions against the background of extremely strong gravitation fields.
Note that the existence of the universal constant with the dimension of momentum/length, postulated by us, allows hyperbolic geometry to be introduced in each phase plane (and in the eight-dimensional phase space QTPH) calling for further studies in subsequent papers.
And, finally, we make some heuristic and philosophical remarks.
Our opinion is that the modern situation with the dark matter is similar to the situation preceding the development of the special relativity. At that time, all attempts of elimination of numerous paradoxes generated by the ether concept based on the dynamic and ontological principles were unsuccessful. Actually, solution of the problem has been found by changing the geometry of the four-dimensional space-time manifold, i.e. owing to a change in the kinematics. It is probable that the dark matter will repeat the lot of the ether. We ventured to suggest that the dark matter is a peculiar factor resultant from the use of inadequate geometry of the eight-dimensional manifold, representing an extended phase space, and hence the use of inadequate kinematics.
Section 3 of this paper contains the results obtained jointly by both coauthors. And all the remaining is on personal responsibility of the first author, so that all deficiencies in the text belong to him <sup>2</sup><sup>2</sup>2We took the liberty to borrow this phrase from the Introduction to the excellent book ”Relativity: The General Theory” by J.L. Synge (L.M.T.).
Unfortunately we were not familiar with publications by G.W. Gibbons and C. Schiller before July 2005.
## 6 Acknowledgements
The authors are grateful to: Professors E.A.Tolkachev, E.V.Doktorov, A.K.Gorbatsevi h, Yu.A.Kurochkin, S.Ya.Kilin; Dr’s V.A.Mossolov, J.G. Suarez, A.E.Shalyt-Margolin, Yu.P.Vyblyi for the support and fruitful discussions on the problems under study. This work is partially supported by the Belarusian Foundation for Fundamental Research. |
warning/0507/hep-th0507176.html | ar5iv | text | # 1 Introduction
## 1 Introduction
Supersymmetric field theories in dimensions higher than the four accessible in our everyday experiences have been contemplated for many years now. Besides being forced on us by our current understanding of superstring theory, it has also proven to be of possible phenomenological importance in which discipline such theories are known as supersymmetric “brane-world” models . Currently, this type of theory is being used by various groups in attempts to implement supersymmetry breaking in a manner consistent with the stringent bounds coming from flavor changing neutral currents. One particular application uses a supersymmetric gravitational theory in a five-dimensional spacetime of which the “extra” spacial dimension is a compact interval of some length<sup>1</sup><sup>1</sup>1An elegant way of constructing such a space is to start with a 5D Minkowski space and toroidally compactify one of the spacial directions, producing a space of topology $`^4\times S^1`$. One subsequently defines a non-free action of $`_2`$ on the circle which has two antipodal fixed points (i.e. a reflection through a “diameter”) and mods the circle by this action. Since the action was not free, the quotient $`S^1/_2`$ is not a smooth manifold (figure 1). Nevertheless it is a manifold-with-boundary diffeomorphic to the closed finite interval $`[0,1]`$. The “orbifold” $`^4\times (S^1/_2)`$ has as boundary two hyperplanes (the “orbifold fixed planes”), each isometric to a 4D Minkowski space, at the fixed points of the $`_2`$ action.
$`\mathrm{}`$ . At each end of the interval is a copy of 4D Minkowski spacetime (the branes). Brane world models consist of postulating that the standard model of particle physics is localized on one of these branes (the “visible” or “infra-red” brane) while other fields propagate in the interior (“bulk”) of the 5D spacetime or on the other “hidden” (or “ultraviolet”) brane (see figure 2).
In the particular model under consideration, supersymmetry is broken on the hidden brane by allowing the $`F`$\- (or $`D`$-term ) of some field localized on that brane to acquire a vacuum expectation value. This is then communicated to our brane by super-graviton loops in the bulk which induce the breakdown of supersymmetry on the standard model brane. The final result is a four-dimensional effective action for the standard model fields on the visible brane.
Needless to say, calculations such as that of gravitational loop corrections are difficult to perform in components (although it was done in at one loop). On the other hand, it is commonly believed that the development of a full-fledged 5D superspace formulation has the annoying drawback that the result, which is desired to be a four-dimensional effective action, is given in a complicated form. More exactly, the output of such an effort is manifestly supersymmetric in five dimensions and must be dimensionally reduced in the final stages of the calculation. For this reason a “hybrid” formalism was developed for supergravity in five dimensions which keeps manifest only 4D, $`𝒩=1`$ super-Poincaré invariance . This hybrid is given in terms of supergravity prepotentials which allows one to apply the powerful supergraph techniques necessary for perturbative quantum calculations. Indeed, in it was used to compute, in a more econmical way than in the component approach of , the leading gravity loop contribution to supersymmetry breakdown described above in a very simple way.
Although the formalism was successfully extended to allow a “warping” of the extra dimension and the gravity-mediation scenario investigated in this background , it has a major drawback which arises as follows. This approach is essentially a superfield Noether procedure in which one starts with a linearized supergravity action, and then tries to reconstruct interaction terms, order by order, by consistently deforming the gauge transformations, etc. Usually the Noether procedure can be completed if it requires a finite number of iterations, as is the case with polynomial actions. But superfield supergravity is a highly nonlinear theory in terms of its prepotentials (see for reviews). As a result, the limitations of this hybrid approach are called into question. More importantly, it turns out to be difficult to discover the rules governing the coupling of this theory to other matter fields in the bulk. In the end, we are forced to turn to the known (full-fledged) off-shell formulations for 5D simple supergravity,<sup>2</sup><sup>2</sup>2We prefer to use the term “5D simple supergravity,” since in the literature 5D simple supersymmetry is called sometimes $`𝒩=1`$ and sometimes $`𝒩=2`$, depending upon taste and background. with or without supersymmetric matter, in the hope of deducing a useful superfield formulation.
Off-shell 5D simple supergravity was sketched in superspace, a quarter century ago, by Breitenlohner and Kabelschacht and Howe (building on a related work ). More recently, it was carefully elaborated by Zucker at the component level, and finally perfected in within the superconformal tensor calculus.
Using the results of the 5D superconformal tensor calculus for supergravity-matter systems, one can develope a hybrid $`𝒩=1`$ superspace formalism by fitting the component multiplets into superfields. Such a program has been carried out in . Although useful for tree-level phenomenological applications, we believe this approach is not the optimum (economical) formulation for doing supergraph loop calculations. The point is that the superconformal tensor calculus usually corresponds to a Wess-Zumino gauge in superfield supergravity. But such gauge conditions are impractical as far as supergraph calculations are concerned.
When comparing the superconformal tensor calculus for 5D simple supergravity with that for 4D, $`𝒩=2`$ and 6D, $`𝒩=(1,0)`$ supergravities (see for the relevant references), it is simply staggering how similar these formulations are, modulo some fine details. From the point of view of a superspace practitioner, the reason for this similarity is that the three versions of superconformal calculus are generated from (correspond to a Wess-Zumino gauge for) a harmonic superspace formulation for the corresponding supergravity theory, and such harmonic superspace formulations<sup>3</sup><sup>3</sup>3The harmonic superspace formulation for 4D, $`𝒩=2`$ supergravity is reviewed in the book . For the case of 6D, $`𝒩=(1,0)`$ supergravity, such a formulation was constructed in , and it can be used to derive a relevant formulation for 5D simple supergravity by dimensional reduction. look almost identical in the space-time dimensions 4, 5 and 6, again modulo fine details. For example, independent of the space-time dimension, the Yang-Mills supermultiplet is described by (formally) the same gauge superfield $`𝒱^{++}`$, with the same gauge freedom $`\delta 𝒱^{++}=𝒟^{++}\lambda `$, and with the same Wess-Zumino-type gauge
$$\mathrm{i}𝒱^{++}=\theta ^+\mathrm{\Gamma }^m\theta ^+A_m(x)+\theta ^+\mathrm{\Gamma }^5\theta ^+A_5(x)+\theta ^+\mathrm{\Gamma }^6\theta ^+A_6(x)+O(\theta ^3),$$
where $`\theta ^+`$ is a four-component anticommuting spinor variable, and $`m=0,1,2,3`$.
The concept of harmonic superspace was originally developed for 4D, $`𝒩=2`$ supersymmetric theories including supergravity , and by now it has become a textbook subject<sup>4</sup><sup>4</sup>4The book contains a list of relevant publications in the context of harmonic superspace. . Actually it can be argued that harmonic superspace is a natural framework for all supersymmetric theories with eight supercharges, both at the classical and quantum levels. In the case of four space-time dimensions, probably the main objection to this approach was the issue that theories in harmonic superspace are often difficult to reduce to $`𝒩=1`$ superfields (the kind of reduction which brane-world practitioners often need). But this objection has been lifted since the advent and subsequent perfection of 4D, $`𝒩=2`$ projective superspace which allows a nice reduction to $`𝒩=1`$ superfields and which appears to be a truncated version of the harmonic superspace .
What is the difference between harmonic superspace and projective superspace? In five space-time dimensions (to be concrete), they make use of the same supermanifold $`^{5|8}\times S^2`$, with $`^{5|8}`$ the conventional 5D simple superspace. In harmonic superspace, one deals with so-called Grassmann analytic (also known as twised chiral) superfields that are chosen to be smooth tensor fields on $`S^2`$. In projective superspace, one also deals with Grassmann analytic superfields that are holomorphic functions on an open subset of $`S^2`$. It is clear that the harmonic superspace setting is more general. Actually, many results originally obtained in projective superspace can be reproduced from harmonic superspace by applying special truncation procedures . The remarkable features of projective superspace are that (i) the projective supermultiplets are easily represented as a direct sum of standard 4D, $`𝒩=1`$ superfields; (ii) this approach provides simple rules to construct low-energy effective actions that are easily expressed in terms of 4D, $`𝒩=1`$ superfields. Of course, one could wonder why both harmonic and projective superspaces should be introduced? The answer is that, in many respects, they are complementary to each other. (This is analogous to the relation between the theorems of existence of solutions for differential equations and concrete techniques to solve such equations.)
To avoid technicalities, in this paper we do not consider 5D superfield supergravity at all, and concentrate only on developing a 5D simple superspace approach to globally supersymmetric gauge theories. One of our main objectives is to demonstrate that 5D superspace may be useful, even in the context of 4D effective theories with an extra dimension. Here we develop manifestly 5D supersymmetric techniques which, on the one hand, allow us to construct many of the 5D supersymmetric models originally developed within the “hybrid” formulation. One the other hand, these techniques make it possible to construct very interesting supersymmetric nonlinear sigma-models whose construction is practically beyond the scope of the “hybrid” formulation. Examples of such 5D supersymmetric sigma-models are constructed for the first time below. We therefore believe that the paper should be of some interest to both superspace experts and newcomers.
It is worth saying a few words about the global structure of this paper. We are aiming at (i) elaborating 5D off-shell matter supermultiplets and their superfield descriptions; (ii) developing various universal procedures to construct manifestly 5D supersymmetric action functionals, and then applying them to specific supermultiplets; (iii) elaborating on techniques to reduce such super-actions to 4D, $`𝒩=1`$ superfields. New elements of 5D superfield formalism are introduced only if they are essential for further consideration. For example, the Yang-Mills off-shell supermultiplet can be realized in 5D conventional superspace in terms of constrained superfields. In order to solve the constraints, however, one has to introduce the concept of harmonic superspace.
This paper is organized as follows: In section 2 we describe, building on earlier work , the 5D Yang-Mills supermultiplet and its salient properties, both in the conventional and harmonic superspaces. We also describe several off-shell realizations for the 5D hypermultiplet. In section 3 we present two procedures to construct 5D manifestly supersymmetric actions for multiplets with and without intrinsic central charge, and give several examples. Section 4 is devoted to 5D supersymmetric Chern-Simons theories. Their harmonic superspace actions are given in a new form, as compared with , which allows a simple reduction to the projective superspace. We also uncover the 5D origin for the superfield constraints describing the so-called 4D, $`𝒩=2`$ nonlinear vector-tensor multplet. In section 5, some of the results developed in the previous sections are reduced to a “hybrid” formulation which keeps manifest only 4D, $`𝒩=1`$ super Poincaré symmetry. Section 6 introduces 5D simple projective superspace and projective multiplets. Here we also present two families of 5D off-shell supersymmetric nonlinear sigma-models which are formulated, respectively, in terms of a (i) 5D tensor multiplet; (ii) 5D polar mutiplet. Section 7 deals with the vector multiplet in projective superspace. A brief conclusion is given in section 8. This paper also includes three technical appendices. Appendix A contains our 5D notation and conventions, inspired by those in , as well as some important identities. Appendix B is devoted to a review of the well-known one-to-one correspondence between smooth tensor fields on $`S^2=\mathrm{SU}(2)/\mathrm{U}(1)`$ and smooth scalar functions over SU(2) with definite U(1) charges. Finally, in appendix C we briefly demonstrate, mainly following , how to derive the projective superspace action (6.16) from the harmonic superspace action (3.2).
## 2 5D Supersymmetric Matter
### 2.1 Vector multiplet in conventional superspace
To describe a Yang-Mills supermultiplet in 5D simple superspace $`^{5|8}`$ parametrized by coordinates $`z^{\widehat{A}}=(x^{\widehat{a}},\theta _i^{\widehat{\alpha }})`$ we introduce gauge-covariant derivatives<sup>5</sup><sup>5</sup>5Our notation and conventions are collected in Appendix A.
$$𝒟_{\widehat{A}}=(𝒟_{\widehat{a}},𝒟_{\widehat{\alpha }}^i)=D_{\widehat{A}}+\mathrm{i}𝒱_{\widehat{A}}(z),[𝒟_{\widehat{A}},𝒟_{\widehat{B}}\}=T_{\widehat{A}\widehat{B}}{}_{}{}^{\widehat{C}}𝒟_{\widehat{C}}^{}+C_{\widehat{A}\widehat{B}}\mathrm{\Delta }+\mathrm{i}_{\widehat{A}\widehat{B}},$$
(2.1)
with $`D_{\widehat{A}}=(_{\widehat{a}},D_{\widehat{\alpha }}^i)`$ the flat covariant derivatives obeying the anti-commutation relations (A.27), $`\mathrm{\Delta }`$ the central charge, and $`𝒱_{\widehat{A}}`$ the gauge connection taking its values in the Lie algebra of the gauge group. The connection is chosen to be inert under the central charge transformations, $`[\mathrm{\Delta },𝒱_{\widehat{A}}]=0`$. The operators $`𝒟_{\widehat{A}}`$ possess the following gauge transformation law
$$𝒟_{\widehat{A}}\mathrm{e}^{\mathrm{i}\tau (z)}𝒟_{\widehat{A}}\mathrm{e}^{\mathrm{i}\tau (z)},\tau ^{}=\tau ,[\mathrm{\Delta },\tau ]=0,$$
(2.2)
with the gauge parameter $`\tau (z)`$ being arbitrary modulo the reality condition imposed. The gauge-covariant derivatives are required to obey some constraints such that
$`\{𝒟_{\widehat{\alpha }}^i,𝒟_{\widehat{\beta }}^j\}`$ $`=`$ $`2\mathrm{i}\epsilon ^{ij}\left((\mathrm{\Gamma }^{\widehat{c}}){}_{\widehat{\alpha }\widehat{\beta }}{}^{}𝒟_{\widehat{c}}^{}+\epsilon _{\widehat{\alpha }\widehat{\beta }}(\mathrm{\Delta }+\mathrm{i}𝒲)\right),[𝒟_{\widehat{\alpha }}^i,\mathrm{\Delta }]=0,`$
$`[𝒟_{\widehat{a}},𝒟_{\widehat{\beta }}^j]`$ $`=`$ $`\mathrm{i}(\mathrm{\Gamma }_{\widehat{a}}){}_{\widehat{\beta }}{}^{}{}_{}{}^{\widehat{\gamma }}𝒟_{\widehat{\gamma }}^{j}𝒲,[𝒟_{\widehat{a}},𝒟_{\widehat{b}}]={\displaystyle \frac{1}{4}}(\mathrm{\Sigma }_{\widehat{a}\widehat{b}})^{\widehat{\alpha }\widehat{\beta }}𝒟_{\widehat{\alpha }}^i𝒟_{\widehat{\beta }i}𝒲=\mathrm{i}_{\widehat{a}\widehat{b}},`$ (2.3)
with the matrices $`\mathrm{\Gamma }_{\widehat{a}}`$ and $`\mathrm{\Sigma }_{\widehat{a}\widehat{b}}`$ defined in Appendix A. Here the field strength $`𝒲`$ is hermitian, $`𝒲^{}=𝒲`$, and obeys the Bianchi identity (see e.g. )
$$𝒟_{\widehat{\alpha }}^{(i}𝒟_{\widehat{\beta }}^{j)}𝒲=\frac{1}{4}\epsilon _{\widehat{\alpha }\widehat{\beta }}𝒟^{\widehat{\gamma }(i}𝒟_{\widehat{\gamma }}^{j)}𝒲,$$
(2.4)
and therefore
$$𝒟_{\widehat{\alpha }}^{(i}𝒟_{\widehat{\beta }}^j𝒟_{\widehat{\gamma }}^{k)}𝒲=0.$$
(2.5)
The independent component fields contained in $`𝒲`$ are:
$`\phi =𝒲||,\mathrm{i}\mathrm{\Psi }_{\widehat{\alpha }}^i=𝒟_{\widehat{\alpha }}^i𝒲||,4\mathrm{i}F_{\widehat{\alpha }\widehat{\beta }}=𝒟_{(\widehat{\alpha }}^i𝒟_{\widehat{\beta })i}𝒲||,4\mathrm{i}X^{ij}=𝒟^{\widehat{\alpha }(i}𝒟_{\widehat{\alpha }}^{j)}𝒲||.`$ (2.6)
Here and in what follows, $`U||`$ denotes the $`\theta `$-independent component of a superfield $`U(x,\theta )`$. It is worth noting that
$$F_{\widehat{a}\widehat{b}}=_{\widehat{a}\widehat{b}}||.$$
(2.7)
### 2.2 Vector multiplet in harmonic superspace
The most elegant way to solve the constraints encoded in the algebra (2.3) is to use the concept of harmonic superspace originally developed for 4D, $`𝒩=2`$ supersymmetric theories (related ideas appeared in ). In this approach, the conventional superspace $`^{5|8}`$ is embedded into $`^{5|8}\times S^2`$, where the two-sphere $`S^2=\mathrm{SU}(2)/\mathrm{U}(1)`$ is parametrized by so-called harmonic $`u_i^{}`$ and $`u_i^+`$, that is group elements
$$(u_i{}_{}{}^{},u_i{}_{}{}^{+})\mathrm{SU}(2),u_i^+=\epsilon _{ij}u^{+j},(u^{+i})^{}=u_i^{},u^{+i}u_i^{}=1.$$
(2.8)
As is well-known, tensor fields over $`S^2`$ are in a one-to-one correspondence with functions over SU(2) possessing definite harmonic U(1) charge (see for a review). A function $`\mathrm{\Psi }^{(p)}(u)`$ is said to have harmonic U(1) charge $`p`$ if
$`\mathrm{\Psi }^{(p)}(\mathrm{e}^{\mathrm{i}\alpha }u^+,\mathrm{e}^{\mathrm{i}\alpha }u^{})=\mathrm{e}^{\mathrm{i}p\alpha }\mathrm{\Psi }^{(p)}(u^+,u^{}),|\mathrm{e}^{\mathrm{i}\alpha }|=1.`$ (2.9)
Such functions, extended to the whole harmonic superspace $`^{5|8}\times S^2`$, that is $`\mathrm{\Psi }^{(p)}(z,u)`$, are called harmonic superfields. Introducing the harmonic derivatives
$`D^{++}=u^{+i}{\displaystyle \frac{}{u^i}},D^{}=u^i{\displaystyle \frac{}{u^{+i}}},D^0`$ $`=`$ $`u^{+i}{\displaystyle \frac{}{u^{+i}}}u^i{\displaystyle \frac{}{u^i}},`$
$`[D^0,D^{\pm \pm }]=\pm 2D^{\pm \pm },[D^{++},D^{}]`$ $`=`$ $`D^0,`$ (2.10)
one can see that $`D^0`$ is the operator of harmonic U(1) charge, $`D^0\mathrm{\Psi }^{(p)}(z,u)=p\mathrm{\Psi }^{(p)}(z,u)`$. Defining
$$𝒟_𝐀(𝒟_{\widehat{A}},𝒟^{++},𝒟^{},𝒟^0),𝒟^{\pm \pm }=D^{\pm \pm },𝒟^0=D^0,$$
(2.11)
one observes that the operators $`𝒟_𝐀`$ possess the same transformation law (2.2) as $`𝒟_{\widehat{A}}`$.
Introduce a new basis for the spinor covariant derivatives: $`𝒟_{\widehat{\alpha }}^+=𝒟_{\widehat{\alpha }}^iu_i^+`$ and $`𝒟_{\widehat{\alpha }}^{}=𝒟_{\widehat{\alpha }}^iu_i^{}`$. Then, eq. (2.3) leads to
$`\{𝒟_{\widehat{\alpha }}^+,𝒟_{\widehat{\beta }}^+\}`$ $`=`$ $`0,[𝒟^{++},𝒟_{\widehat{\alpha }}^+]=0,`$
$`\{𝒟_{\widehat{\alpha }}^+,𝒟_{\widehat{\beta }}^{}\}`$ $`=`$ $`2\mathrm{i}\left(𝒟_{\widehat{\alpha }\widehat{\beta }}+\epsilon _{\widehat{\alpha }\widehat{\beta }}(\mathrm{\Delta }+\mathrm{i}𝒲)\right),`$ (2.12)
$`[𝒟^{++},𝒟_{\widehat{\alpha }}^{}]`$ $`=`$ $`𝒟_{\widehat{\alpha }}^+,[𝒟^{},𝒟_{\widehat{\alpha }}^+]=𝒟_{\widehat{\alpha }}^{}.`$
In harmonic superspace, the integrability condition $`\{𝒟_{\widehat{\alpha }}^+,𝒟_{\widehat{\beta }}^+\}=0`$ is solved by
$$𝒟_{\widehat{\alpha }}^+=\mathrm{e}^{\mathrm{i}\mathrm{\Omega }}D_{\widehat{\alpha }}^+\mathrm{e}^{\mathrm{i}\mathrm{\Omega }},$$
(2.13)
for some Lie-algebra valued harmonic superfield $`\mathrm{\Omega }=\mathrm{\Omega }(z,u)`$ of vanishing harmonic U(1) charge, $`D^0\mathrm{\Omega }=0`$. This superfield is called the bridge. The bridge possesses a richer gauge freedom than the original $`\tau `$-group (2.2)
$$\mathrm{e}^{\mathrm{i}\mathrm{\Omega }(z,u)}\mathrm{e}^{\mathrm{i}\lambda (z,u)}\mathrm{e}^{\mathrm{i}\mathrm{\Omega }(z,u)}\mathrm{e}^{\mathrm{i}\tau (z)},D_{\widehat{\alpha }}^+\lambda =0,[\mathrm{\Delta },\lambda ]=0.$$
(2.14)
The $`\lambda `$\- and $`\tau `$-transformations generate, respectively, the so-called $`\lambda `$\- and $`\tau `$-groups.
One can now define covariantly analytic superfields constrained by
$$𝒟_{\widehat{\alpha }}^+\mathrm{\Phi }^{(p)}=0.$$
(2.15)
Here $`\mathrm{\Phi }^{(p)}(z,u)`$ carries U(1)-charge $`p`$, $`D^0\mathrm{\Phi }^{(p)}=p\mathrm{\Phi }^{(p)}`$, and can be represented as follows
$$\mathrm{\Phi }^{(p)}=\mathrm{e}^{\mathrm{i}\mathrm{\Omega }}\varphi ^{(p)},D_{\widehat{\alpha }}^+\varphi ^{(p)}=0,$$
(2.16)
with $`\varphi ^{(p)}(\zeta )`$ being an analytic superfield – that is, a field over the so-called analytic subspace of the harmonic superspace parametrized by
$$\zeta \{𝒙^{\widehat{a}},\theta ^{+\widehat{\alpha }},u_i^+,u_j^{}\},$$
(2.17)
where
$$𝒙^{\widehat{a}}=x^{\widehat{a}}+\mathrm{i}(\mathrm{\Gamma }^{\widehat{a}})_{\widehat{\beta }\widehat{\gamma }}\theta ^{+\widehat{\beta }}\theta ^{\widehat{\gamma }},\theta _{\widehat{\alpha }}^\pm =\theta _{\widehat{\alpha }}^iu_i^\pm .$$
(2.18)
In particular, the gauge parameter $`\lambda `$ in (2.14) is an unconstrained analytic superfield of vanishing harmonic U(1) charge, $`D^0\lambda =0`$. It is clear that the superfields $`\mathrm{\Phi }^{(p)}`$ and $`\varphi ^{(p)}`$ describe the same matter multiplet but in different frames (or, equivalently, representations), and they transform under the $`\tau `$\- and $`\lambda `$-gauge groups, respectively.
$$\mathrm{\Phi }^{(p)}(z,u)\mathrm{e}^{\mathrm{i}\tau (z)}\mathrm{\Phi }^{(p)}(z,u),\varphi ^{(p)}(z,u)\mathrm{e}^{\mathrm{i}\lambda (z,u)}\varphi ^{(p)}(z,u).$$
(2.19)
By construction, the analytic subspace (2.17) is closed under the supersymmetry transformations. Unlike the chiral subspace, it is real with respect to the generalized conjugation (often called the smile-conjugation) $`\stackrel{˘}{}`$ defined to be the composition of the complex conjugation (Hermitian conjugation in the case of Lie-algebra-valued superfields) with the operation acting on the harmonics only
$`(u_i^+)^{}=u_i^{},(u_i^{})^{}=u_i^+(u_i^\pm )^{}=u_i^\pm ,`$ (2.20)
hence
$$(u^{+i})\stackrel{˘}{}=u_i^+(u_i^{})\stackrel{˘}{}=u^i.$$
(2.21)
The analytic superfields of even U(1) charge may therefore be chosen to be real. In particular, the bridge $`\mathrm{\Omega }`$ and the gauge parameter $`\lambda `$ are real.
The covariant derivatives in the $`\lambda `$-frame are obtained from those in the $`\tau `$-frame, eq. (2.11), by applying the transformation
$$𝒟_𝐀\mathrm{e}^{\mathrm{i}\mathrm{\Omega }}𝒟_𝐀\mathrm{e}^{\mathrm{i}\mathrm{\Omega }}.$$
(2.22)
Then, the gauge transformation of the covariant derivatives becomes
$$𝒟_𝐀\mathrm{e}^{\mathrm{i}\lambda (\zeta )}𝒟_𝐀\mathrm{e}^{\mathrm{i}\lambda (\zeta )},\stackrel{˘}{\lambda }=\lambda ,[\mathrm{\Delta },\lambda ]=0.$$
(2.23)
In the $`\lambda `$-frame, the spinor covariant derivatives $`𝒟_{\widehat{\alpha }}^+`$ coincide with the flat ones, $`𝒟_{\widehat{\alpha }}^+=D_{\widehat{\alpha }}^+`$, while the harmonic covariant derivatives acquire connections,
$$𝒟^{\pm \pm }=D^{\pm \pm }+\mathrm{i}𝒱^{\pm \pm }.$$
(2.24)
The real connection $`𝒱^{++}`$ is seen to be an analytic superfield, $`D_{\widehat{\alpha }}^+𝒱^{++}=0`$, of harmonic U(1) charge plus two, $`D^0𝒱^{++}=2𝒱^{++}`$. The other harmonic connection $`𝒱^{}`$ turns out to be uniquely determined in terms of $`𝒱^{++}`$ using the zero-curvature condition
$$[𝒟^{++},𝒟^{}]=D^0D^{++}𝒱^{}D^{}𝒱^{++}+\mathrm{i}[𝒱^{++},𝒱^{}]=0,$$
(2.25)
as demonstrated in . The result is
$`𝒱^{}(z,u)={\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}(\mathrm{i})^{n+1}{\displaystyle du_1\mathrm{}du_n\frac{𝒱^{++}(z,u_1)𝒱^{++}(z,u_2)\mathrm{}𝒱^{++}(z,u_n)}{(u^+u_1^+)(u_1^+u_2^+)\mathrm{}(u_n^+u^+)}},`$ (2.26)
with $`(u_1^+u_2^+)=u_1^{+i}u_2^+_i`$, and the harmonic distributions on the right of (2.26) defined, e.g., in . Integration over the group manifold SU(2) is normalized according to
$$du\mathrm{\hspace{0.17em}1}=1,duu_{(i_1}^+\mathrm{}u_{i_n}^+u_{j_1}^{}\mathrm{}u_{j_m)}^{}=0,n+m>0.$$
(2.27)
As far as the connections $`𝒱_{\widehat{\alpha }}^{}`$ and $`𝒱_{\widehat{a}}`$ are concerned, they can be expressed in terms of $`𝒱^{}`$ with the aid of the (anti-)commutation relations
$`[𝒟^{},𝒟_{\widehat{\alpha }}^+]`$ $`=`$ $`𝒟_{\widehat{\alpha }}^{},\{𝒟_{\widehat{\alpha }}^+,𝒟_{\widehat{\beta }}^{}\}=2\mathrm{i}\left(𝒟_{\widehat{\alpha }\widehat{\beta }}+\epsilon _{\widehat{\alpha }\widehat{\beta }}(\mathrm{\Delta }+\mathrm{i}𝒲_\lambda )\right).`$ (2.28)
In particular, one obtains
$$𝒲_\lambda =\frac{\mathrm{i}}{8}(\widehat{D}^+)^2𝒱^{},(\widehat{D}^+)^2=D^{+\widehat{\alpha }}D_{\widehat{\alpha }}^+,$$
(2.29)
where $`𝒲_\lambda `$ stands for the field strength in the $`\lambda `$-frame. Therefore, $`𝒱^{++}`$ is the single unconstrained analytic prepotential of the theory. With the aid of (2.25) one can obtain the following useful expression
$`𝒲={\displaystyle \frac{\mathrm{i}}{8}}{\displaystyle du(\widehat{D}^{})^2𝒱^{++}}+O\left((𝒱^{++})^2\right).`$ (2.30)
In the Abelian case, only the first term on the right survives. In what follows, we do not distinguish between $`𝒲`$ and $`𝒲_\lambda `$.
With the notation $`(\widehat{𝒟}^+)^2=𝒟^{+\widehat{\alpha }}𝒟_{\widehat{\alpha }}^+`$, the Bianchi identity (2.4) takes the form
$$𝒟_{\widehat{\alpha }}^+𝒟_{\widehat{\beta }}^+𝒲=\frac{1}{4}\epsilon _{\widehat{\alpha }\widehat{\beta }}(\widehat{𝒟}^+)^2𝒲𝒟_{\widehat{\alpha }}^+𝒟_{\widehat{\beta }}^+𝒟_{\widehat{\gamma }}^+𝒲=0.$$
(2.31)
Using the Bianchi identity (2.31), one can readily construct a covariantly analytic descendant of $`𝒲`$
$$\mathrm{i}𝒢^{++}=𝒟^{+\widehat{\alpha }}𝒲𝒟_{\widehat{\alpha }}^+𝒲+\frac{1}{4}\{𝒲,(\widehat{𝒟}^+)^2𝒲\},𝒟_{\widehat{\alpha }}^+𝒢^{++}=𝒟^{++}𝒢^{++}=0.$$
(2.32)
### 2.3 Vector multiplet in components
The gauge freedom
$$\delta 𝒱^{++}=𝒟^{++}\lambda =D^{++}\lambda +\mathrm{i}[𝒱^{++},\lambda ]$$
(2.33)
can be used to choose a Wess-Zumino gauge of the form
$`𝒱^{++}(𝒙,\theta ^+,u)`$ $`=`$ $`\mathrm{i}(\widehat{\theta }^+)^2\phi (𝒙)\mathrm{i}\theta ^+\mathrm{\Gamma }^{\widehat{m}}\theta ^+A_{\widehat{m}}(𝒙)+4(\widehat{\theta }^+)^2\theta ^{+\widehat{\alpha }}u_i^{}\mathrm{\Psi }_{\widehat{\alpha }}^i(𝒙)`$ (2.34)
$``$ $`{\displaystyle \frac{3}{2}}(\widehat{\theta }^+)^2(\widehat{\theta }^+)^2u_i^{}u_j^{}X^{ij}(𝒙),`$
where
$$(\widehat{\theta }^+)^2=\theta ^{+\widehat{\alpha }}\theta _{\widehat{\alpha }}^+,\theta ^+\mathrm{\Gamma }^{\widehat{m}}\theta ^+=\theta ^{+\widehat{\alpha }}(\mathrm{\Gamma }^{\widehat{m}})_{\widehat{\alpha }}{}_{}{}^{\widehat{\beta }}\theta _{}^{+}{}_{\widehat{\beta }}{}^{}=(\mathrm{\Gamma }^{\widehat{m}})_{\widehat{\alpha }\widehat{\beta }}\theta ^{+\widehat{\alpha }}\theta ^{+\widehat{\beta }}.$$
(2.35)
In this gauge, the expression (2.26) simplifies considerably
$`𝒱^{}(z,u)`$ $`=`$ $`{\displaystyle du_1\frac{𝒱^{++}(z,u_1)}{(u^+u_1^+)^2}}+{\displaystyle \frac{\mathrm{i}}{2}}{\displaystyle du_1du_2\frac{[𝒱^{++}(z,u_1),𝒱^{++}(z,u_2)]}{(u^+u_1^+)(u_1^+u_2^+)(u_2^+u^+)}}`$ (2.36)
$`+`$ $`\text{terms of third- and fourth-order in}𝒱^{++}.`$
Here the explicit form of the cubic and quartic terms is not relevant for our consideration. One of the important properties of the Wess-Zumino gauge is
$$𝒟_{\widehat{m}}||_{\widehat{m}}+\mathrm{i}𝒱_{\widehat{m}}||=_{\widehat{m}}+\mathrm{i}A_{\widehat{m}}(x).$$
(2.37)
The component fields of $`𝒲`$ and $`𝒱^{++}`$ can be related to each other using the identity
$$F_2^+=(u_1^+u_2^+)F_1^{}(u_1^{}u_2^+)F_1^+,F^\pm =F^iu_i^\pm ,$$
(2.38)
and the analyticity of $`𝒱^{++}`$. (The latter property implies, for instance, $`D^+𝒱^{++}(z,u_1)=(u^+u_1^+)D_1^{}𝒱^{++}(z,u_1)`$.) Thus one gets
$`𝒲||={\displaystyle \frac{\mathrm{i}}{8}}(\widehat{D}^+)^2𝒱^{}||={\displaystyle \frac{\mathrm{i}}{8}}{\displaystyle }\mathrm{d}u_1(\widehat{D}_1^{})^2𝒱^{++}(z,u_1)||`$ $`=`$ $`\phi (x),`$
$`𝒟_{\widehat{\alpha }}^+𝒲||={\displaystyle \frac{\mathrm{i}}{8}}{\displaystyle }\mathrm{d}u_1(u^+u_1^+)D_{1\widehat{\alpha }}^{}(\widehat{D}_1^{})^2𝒱^{++}(z,u_1)||`$ $`=`$ $`\mathrm{i}\mathrm{\Psi }_{\widehat{\alpha }}^i(x)u_i^+,`$ (2.39)
$`(\widehat{𝒟}^+)^2𝒲||={\displaystyle \frac{\mathrm{i}}{8}}{\displaystyle }\mathrm{d}u_1(u^+u_1^+)^2(\widehat{D}_1^{})^2(\widehat{D}_1^{})^2𝒱^{++}(z,u_1)||`$ $`=`$ $`4\mathrm{i}X^{ij}(x)u_i^+u_j^+.`$
Finally, eq. (2.37) implies that the component field $`F_{\widehat{\alpha }\widehat{\beta }}=F_{(\widehat{\alpha }\widehat{\beta })}`$ in (2.6) is (the bispinor form of) the gauge-covariant field strength $`F_{\widehat{m}\widehat{n}}`$ generated by the gauge field $`A_{\widehat{m}}`$.
### 2.4 Fayet-Sohnius hypermultiplet
Following the four-dimensional $`𝒩=2`$ supersymmetric construction due to Fayet and Sohnius , an off-shell hypermultiplet with intrinsic central charge, which is coupled to the Yang-Mills supermultiplet, can be described by a superfield $`𝒒^i(z)`$ and its conjugate $`\overline{𝒒}_i(z)`$, $`\overline{𝒒}_i=(𝒒^i)^{}`$, subject to the constraint
$$𝒟_{\widehat{\alpha }}^{(i}𝒒^{j)}=0.$$
(2.40)
Introducing $`𝒒^+(z,u)=𝒒^i(z)u_i^+`$ and $`\stackrel{˘}{𝒒}{}_{}{}^{+}(z,u)=\overline{𝒒}^i(z)u_i^+`$, the constraint (2.40) can be rewritten in the form
$$𝒟_{\widehat{\alpha }}^+𝒒^+=𝒟_{\widehat{\alpha }}^+\stackrel{˘}{𝒒}{}_{}{}^{+}=0,𝒟^{++}𝒒^+=𝒟^{++}\stackrel{˘}{𝒒}{}_{}{}^{+}=0.$$
(2.41)
Thus $`𝒒^+`$ is a constrained analytic superfield. Using the algebra of gauge-covariant derivatives, the constraints can be shown to imply<sup>6</sup><sup>6</sup>6By analogy with the four-dimensional case , the operator $`\stackrel{}{\mathrm{}}`$ can be called the covariant analytic d’Alembertian. Given a covariantly analytic superfield $`\mathrm{\Phi }^{(q)}`$, the identity $`\stackrel{}{\mathrm{}}\mathrm{\Phi }^{(q)}=\frac{1}{64}(\widehat{𝒟}^+)^2(\widehat{𝒟}^+)^2(𝒟^{})^2\mathrm{\Phi }^{(q)}`$ holds, and therefore $`\stackrel{}{\mathrm{}}`$ preserves analyticity.
$`\stackrel{}{\mathrm{}}𝒒^+=0,`$ (2.42)
$`\stackrel{}{\mathrm{}}=𝒟^{\widehat{a}}𝒟_{\widehat{a}}+(𝒟^{+\widehat{\alpha }}𝒲)𝒟_{\widehat{\alpha }}^{}{\displaystyle \frac{1}{4}}(\widehat{𝒟}^{+\widehat{\alpha }}𝒟_{\widehat{\alpha }}^+𝒲)𝒟^{}+{\displaystyle \frac{1}{8}}[𝒟^{+\widehat{\alpha }},𝒟_{\widehat{\alpha }}^{}]𝒲+(\mathrm{\Delta }+\mathrm{i}𝒲)^2.`$
Therefore, the requirement of a constant central charge, $`\mathrm{\Delta }𝒒^+=m𝒒^+`$, with $`m`$ a constant mass parameter, is equivalent to an equation of motion for the hypermultiplet.
Independent component fields of $`𝒒^i(z)`$ can be chosen as
$`C^i=𝒒^i||,\lambda _{\widehat{\alpha }}={\displaystyle \frac{\mathrm{i}}{\sqrt{8}}}𝒟_{\widehat{\alpha }}^i𝒒_i||,F^i=\mathrm{\Delta }𝒒^i||.`$ (2.43)
All other components can be related to these and their derivatives. For example,
$`(\widehat{𝒟}^{})^2𝒒^+`$ $`=`$ $`8\mathrm{i}\mathrm{\Delta }𝒒^{}8𝒲𝒒^{},`$ (2.44)
$`𝒟_{\widehat{\alpha }}^{}\mathrm{\Delta }𝒒^+`$ $`=`$ $`𝒟_{\widehat{\alpha }}{}_{}{}^{\widehat{\beta }}𝒟_{\widehat{\beta }}^{+}𝒒^{}+\mathrm{i}𝒲𝒟_{\widehat{\alpha }}^+𝒒^{}+2\mathrm{i}(𝒟_{\widehat{\alpha }}^+𝒲)𝒒^{},`$ (2.45)
$`(\widehat{𝒟}^{})^2\mathrm{\Delta }𝒒^+`$ $`=`$ $`8\mathrm{i}𝒟^{\widehat{a}}𝒟_{\widehat{a}}𝒒^{}+8𝒲\mathrm{\Delta }𝒒^{}+8\mathrm{i}𝒲^2𝒒^{}+4\mathrm{i}(𝒟^{\widehat{\alpha }}𝒲)𝒟_{\widehat{\alpha }}^+𝒒^{}`$ (2.47)
$`+2\mathrm{i}(𝒟^{\widehat{\alpha }}𝒟_{\widehat{\alpha }}^+𝒲)𝒒^{}2\mathrm{i}(𝒟^{\widehat{\alpha }}𝒟_{\widehat{\alpha }}^{}𝒲)𝒒^+.`$
### 2.5 Off-shell hypermultiplets without central charge
One of the main virtues of the harmonic superspace approach is that it makes possible an off-shell formulation for a charged hypermultiplet (transforming in an arbitrary representation of the gauge group) without central charge. Such a $`q^+`$-hypermultiplet is described by an unconstrained analytic superfield $`q^+(z,u)`$ and its conjugate $`\stackrel{˘}{q}^+(z,u)`$,
$$𝒟_{\widehat{\alpha }}^+q^+=0,\mathrm{\Delta }q^+=0.$$
(2.48)
In this approach, the requirement that $`q^+`$ be a holomorphic spinor field over the two-sphere, $`𝒟^{++}q^+=0`$, is equivalent to an equation of motion.<sup>7</sup><sup>7</sup>7The equation of motion for the massless Fayet-Sohnius hypermultiplet, which is characterised by the kinematic constraint $`𝒟^{++}𝒒^+=0`$, can be shown to be $`\mathrm{\Delta }𝒒^+=0`$, if the dynamics is generated by the Lagrangian (3.16) with $`m=0`$. The harmonic dependence of the $`q^+`$-hypermultiplet is non-trivial. One can represent $`q^+(z,u)`$ by a convergent Fourier series of the form (B.9) with $`p=1`$. The corresponding Fourier coefficients $`q^{i_1\mathrm{}i_{2n+1}}(z)`$, where $`n=0,1,\mathrm{}`$, obey some constraints that follow from the analyticity condition in (2.48).
Given a hypermultiplet that transforms in a real representation of the gauge group, it can be described by a real anaytic superfied $`\omega (z,u)`$,
$$𝒟_{\widehat{\alpha }}^+\omega =0,\mathrm{\Delta }\omega =0,$$
(2.49)
called the $`\omega `$-hypermultiplet . The gauge parameter $`\lambda `$ in (2.23) is of this superfield type. It is then clear that the $`\omega `$-hypermultiplet can be used, for instance, to formulate a gauge-invariant Stückelberg description for massive vector multiplets.
## 3 Supersymmetric Actions
In the case of vanishing central charge, $`\mathrm{\Delta }=0`$, it is easy to construct manifestly supersymmetric actions within the harmonic superspace approach . Given a scalar harmonic superfield $`L(z,u)`$ and a scalar analytic superfield $`L^{(+4)}(\zeta )`$, supersymmetric actions are:
$`S_\mathrm{H}`$ $`=`$ $`{\displaystyle }\mathrm{d}^5x\mathrm{d}^8\theta \mathrm{d}uL={\displaystyle }\mathrm{d}^5x\mathrm{d}u(\widehat{D}^{})^4(\widehat{D}^+)^4L\left|\right|,`$ (3.1)
$`S_\mathrm{A}`$ $`=`$ $`{\displaystyle }\mathrm{d}\zeta ^{(4)}L^{(+4)}={\displaystyle }\mathrm{d}^5x\mathrm{d}u(\widehat{D}^{})^4L^{(+4)}\left|\right|,𝒟_{\widehat{\alpha }}^+L^{(+4)}=0,`$ (3.2)
where
$$(\widehat{D}^\pm )^4=\frac{1}{32}(\widehat{D}^\pm )^2(\widehat{D}^\pm )^2.$$
(3.3)
As follows from (3.1), any integral over the full superspace can be reduced to an integral over the analytic subspace,
$$\mathrm{d}^5x\mathrm{d}^8\theta duL=d\zeta ^{(4)}L^{(+4)},L^{(+4)}=(\widehat{D}^+)^4L.$$
(3.4)
The massless $`q^+`$-hypermultiplet action is
$$S=d\zeta ^{(4)}\stackrel{˘}{q}{}_{}{}^{+}𝒟_{}^{++}q^+.$$
(3.5)
This action also describes a massive hypermutliplet if one assumes that (i) the gauge group is $`G\times \mathrm{U}(1)`$, and (ii) the U(1) gauge field $`𝒱_0^{++}`$ possesses a constant field strength $`𝒲_0=\mathrm{const}`$, $`|𝒲_0|=m`$, see for more details.<sup>8</sup><sup>8</sup>8A different approach to formulate massive hypermultiplets was proposed in .
Similarly to the chiral scalar in 4D, $`𝒩=1`$ supersymmerty, couplings for the $`q^+`$-hypermultiplet are easy to construct. For example, one can consider a Lagrangian of the form
$`L^{(+4)}=\stackrel{˘}{q}{}_{}{}^{+}𝒟_{}^{++}q^++\lambda (\stackrel{˘}{q}{}_{}{}^{+}q_{}^{+})^2+\stackrel{˘}{q}{}_{}{}^{+}\{\sigma _1(\widehat{𝒟}^+)^2𝒲+\mathrm{i}\sigma _2𝒢^{++}\}q^+,`$ (3.6)
with the quartic self-coupling first introduced in . Consistent couplings for the Fayet-Sohnius hypermultiplet are much more restrictive, as a result of a non-vanishing intrinsic central charge.
### 3.1 Four-derivative vector multiplet action
As another example of supersymemtric action, we consider four-derivative couplings that may occur in low-energy effective actions for a U(1) vector multiplet.
$`S_{\mathrm{four}\mathrm{deriv}}`$ $`=`$ $`{\displaystyle d\zeta ^{(4)}𝒢^{++}\left\{\kappa _1𝒢^{++}+\mathrm{i}\kappa _2(\widehat{D}^+)^2𝒲\right\}}+{\displaystyle \mathrm{d}^5x\mathrm{d}^8\theta H(𝒲)},`$ (3.7)
with $`\kappa _{1,2}`$ coupling constants, the analytic superfield $`𝒢^{++}`$ given by (2.32), and $`H(𝒲)`$ an arbitrary function. The third term on the right is a natural generalization of the four-derivative terms in 4D, $`𝒩=2`$ supersymmetry first introduced in .
### 3.2 Multiplets with intrinsic central charge
For 5D off-shell supermultiplets with $`\mathrm{\Delta }0`$, the construction of supersymmetric actions is based on somewhat different ideas developed in for the case of 4D, $`𝒩=2`$ supersymmetric theories.
When $`\mathrm{\Delta }0`$, there exists one more useful representation (in addition to the $`\tau `$-frame and $`\lambda `$-frame) for the covariant derivatives
$$𝒟_𝐀\mathbf{}_𝐀=\mathrm{e}^{\mathrm{i}(\mathrm{\Omega }+\mathrm{\Sigma })}𝒟_𝐀\mathrm{e}^{\mathrm{i}(\mathrm{\Omega }+\mathrm{\Sigma })}𝐃_𝐀+\mathrm{i}𝒱_𝐀,\mathrm{\Sigma }=\theta ^{\widehat{\alpha }}\theta _{\widehat{\alpha }}^+\mathrm{\Delta }.$$
(3.8)
For the operators $`𝐃_𝐀=\mathrm{e}^{\mathrm{i}\mathrm{\Sigma }}D_𝐀\mathrm{e}^{\mathrm{i}\mathrm{\Sigma }}`$ one obtains
$`\mathbf{}_{\widehat{\alpha }}^+`$ $`=`$ $`𝐃_{\widehat{\alpha }}^+={\displaystyle \frac{}{\theta ^{\widehat{\alpha }}}},`$ (3.9)
$`𝐃^{++}`$ $`=`$ $`D^{++}+\mathrm{i}(\widehat{\theta }^+)^2\mathrm{\Delta },D^{++}=u^{+i}{\displaystyle \frac{}{u^i}}+\mathrm{i}(\mathrm{\Gamma }^{\widehat{a}})_{\widehat{\beta }\widehat{\gamma }}\theta ^{+\widehat{\beta }}\theta ^{+\widehat{\gamma }}{\displaystyle \frac{}{𝒙^{\widehat{a}}}}+\theta ^{+\widehat{\alpha }}{\displaystyle \frac{}{\theta ^{\widehat{\alpha }}}},`$
where
$$(\widehat{\theta }^+)^2=\theta ^{+\widehat{\alpha }}\theta _{\widehat{\alpha }}^+.$$
(3.10)
As is seen, in this frame the spinor gauge-covariant derivative $`\mathbf{}_{\widehat{\alpha }}^+`$ coincides with partial derivatives with respect to $`\theta ^{\widehat{\alpha }}`$, while the analyticity-preserving gauge-covariant derivative $`\mathbf{}^{++}=𝐃^{++}+\mathrm{i}𝒱^{++}`$ acquires a term proportional to the central charge.
In accordance with , the supersymmetric action involves a special gauge-invariant analytic superfield $`𝐋^{++}`$
$$𝐃_{\widehat{\alpha }}^+𝐋^{++}=0,𝐃^{++}𝐋^{++}=0.$$
(3.11)
The action is
$$S=\mathrm{i}d\zeta ^{(4)}(\widehat{\theta }^+)^2𝐋^{++}.$$
(3.12)
Although $`S`$ involves naked Grassmann variables, it turns out to be supersymmetric,due to the constraints imposed on $`𝐋^{++}`$. Its invariance under the supersymmetry transformations can be proved in complete analogy with the four-dimensional case . The action (3.12) possesses another nice representation obtained in Appendix C, eq. (C.13).
One can transform $`𝐋^{++}`$ to the $`\tau `$-frame in which
$$L^{++}(z,u)=\mathrm{e}^{\mathrm{i}\mathrm{\Sigma }}𝐋^{++}=L^{ij}(z)u_i^+u_j^+.$$
(3.13)
This gauge-invariant superfield obeys the constrains
$$D_{\widehat{\alpha }}^+L^{++}=0,D^{++}L^{++}=0.$$
(3.14)
Doing the Grassmann and harmonic integrals in (3.12) gives
$$S=\frac{\mathrm{i}}{12}\mathrm{d}^5x\widehat{𝒟}^{ij}L_{ij}\left|\right|,\widehat{𝒟}^{ij}=𝒟^{\widehat{\alpha }(i}𝒟_{\widehat{\alpha }}^{j)},$$
(3.15)
where we have replaced, for convenience, ordinary spinor covariant derivatives by gauge-covariant ones (this obviously does not change the action, for $`L_{ij}`$ is gauge invariant).
In four space-time dimensions, the super-action (3.15) was postulated by Sohnius several years before the discovery of harmonic superspace. It is quite remarkable that only within the harmonic superspace approach, this super-action can be represented as a superspace integral having a transparent physical interpretation. To wit, the factor $`\mathrm{i}(\widehat{\theta }^+)^2`$ in (3.12) can be identified with a vacuum expectation value $`𝒱_\mathrm{\Delta }^{++}`$ of the central-charge gauge superfield $`𝒱_\mathrm{\Delta }^{++}`$ (compare with (2.3)). With such an interpretation, the super-action admits simple generalizations to the cases of (i) rigid supersymmetric theories with gauged central charge , and (ii) supergravity-matter systems .
The super-action (3.12), and its equivalent form (3.15), can be used for superymmetric theories without central charge; an example will be given below. It is only the constraints (3.14) which are relevant in the above construction.
### 3.3 Fayet-Sohnius hypermultiplet
An example of a theory with non-vanishing central charge is provided by the Fayet-Sohnius hypermultiplet. The Fayet-Sohnius hypermultiplet coupled to a Yang-Mills supermultiplet is described by the Lagrangian
$$L_{\mathrm{FS}}^{++}=\frac{1}{2}\stackrel{˘}{𝒒}^+\stackrel{}{\mathrm{\Delta }}𝒒^+\mathrm{i}m\stackrel{˘}{𝒒}^+𝒒^+,$$
(3.16)
with $`m`$ the hypermultiplet mass.
To compute the component action that follows from (3.16), one should use the definitions (2.6) and (2.43) for the component fields of $`𝒲`$ and $`𝒒^i`$, respectively.
$`S_{\mathrm{FS}}`$ $`=`$ $`{\displaystyle }\mathrm{d}^5x\{𝒟_a\overline{C}_k𝒟^aC^k\mathrm{i}\overline{\lambda }\overline{)𝒟}\lambda +\overline{F}_kF^k+m\overline{\lambda }\lambda +\overline{\lambda }\phi \lambda {\displaystyle \frac{\mathrm{i}}{2}}\overline{C}_kX^k{}_{\mathrm{}}{}^{}C_{}^{\mathrm{}}`$ (3.17)
$`{\displaystyle \frac{1}{2}}\overline{C}_k\phi ^2C^k(\mathrm{i}m\overline{F}_kC^k{\displaystyle \frac{1}{\sqrt{8}}}\overline{\lambda }\mathrm{\Psi }_kC^k+\mathrm{i}\overline{F}_k\phi C^k+\mathrm{c}.\mathrm{c}.)\}.`$
### 3.4 Vector multiplet
The Yang-Mills supermultiplet is described by the Lagrangian<sup>9</sup><sup>9</sup>9In terms of the analytic prepotential $`𝒱^{++}`$, the super Yang-Mills action is non-polynomial .
$$L_{\mathrm{YM}}^{++}=\frac{1}{4}\mathrm{tr}𝒢^{++},\mathrm{\Delta }L_{\mathrm{YM}}^{++}=0,$$
(3.18)
with $`𝒢^{++}`$ given in (2.32). The corresponding equation of motion can be shown to be
$$(\widehat{𝒟}^+)^2𝒲=0𝒟_{\widehat{\alpha }}^+𝒟_{\widehat{\beta }}^+𝒲=0.$$
(3.19)
It follows from this that
$$𝒟^{\widehat{a}}𝒟_{\widehat{a}}𝒲=\frac{1}{2}\{𝒟_i^{\widehat{\alpha }}𝒲,𝒟_{\widehat{\alpha }}^i𝒲\}.$$
(3.20)
In the Abelian case, eq. (3.19) reduces to
$$D_{\widehat{\alpha }}^+D_{\widehat{\beta }}^+𝒲=0^{\widehat{a}}_{\widehat{a}}𝒲=0.$$
(3.21)
From the point of view of 4D, $`𝒩=2`$ supersymmetry, this can be recognized as the off-shell superfield constraints describing the so-called linear vector-tensor multiplet discovered by Sohnius, Stelle and West and re-vitalized fifteen years later in the context of superstring compactifications .
The Yang-Mills action with components defined by (2.34) is
$`S_{\mathrm{YM}}={\displaystyle }\mathrm{d}^5x\mathrm{tr}\{{\displaystyle \frac{1}{4}}F_{\widehat{a}\widehat{b}}F^{\widehat{a}\widehat{b}}`$ $``$ $`{\displaystyle \frac{1}{2}}𝒟_{\widehat{a}}\phi 𝒟^{\widehat{a}}\phi +{\displaystyle \frac{1}{4}}X^{ij}X_{ij}`$ (3.22)
$`+`$ $`{\displaystyle \frac{\mathrm{i}}{2}}\mathrm{\Psi }^k\overline{)𝒟}\mathrm{\Psi }_k{\displaystyle \frac{1}{2}}\mathrm{\Psi }^k[\phi ,\mathrm{\Psi }_k]\}.`$
## 4 Chern-Simons Couplings
Consider two vector multiplets: (i) a U(1) vector multiplet $`𝒱_\mathrm{\Delta }^{++}`$; (ii) a Yang-Mills vector multilpet $`𝒱_{\mathrm{YM}}^{++}`$. They can be coupled to each other, in a gauge-invariant way, using the interaction
$$S_{\mathrm{int}}=d\zeta ^{(4)}𝒱_\mathrm{\Delta }^{++}\mathrm{tr}𝒢_{\mathrm{YM}}^{++},$$
(4.1)
where $`𝒢_{\mathrm{YM}}^{++}`$ corresponds to the non-Abelian multiplet and is defined as in eq. (2.32). Invariance of $`S_{\mathrm{int}}`$ under the U(1) gauge transformations
$$\delta 𝒱^{++}=D^{++}\lambda ,D_{\widehat{\alpha }}^+\lambda =0,$$
(4.2)
follows from the constraints (2.32) to which $`𝒢_{\mathrm{YM}}^{++}`$ is subject.
Let us assume that the physical scalar field in $`𝒱_\mathrm{\Delta }^{++}`$ possesses a non-vanishing expectation value (such a situation occurs, for instance, when $`𝒱_\mathrm{\Delta }^{++}`$ is the vector multiplet gauging the central charge symmetry). In accordance with , this condition is expressed as $`𝒲_\mathrm{\Delta }(z)=\mu 0`$; then, there exists a gauge fixing such that
$$𝒱_\mathrm{\Delta }^{++}(\zeta ,u)=\mathrm{i}\mu (\widehat{\theta }^+)^2,\mu =\mathrm{const}.$$
(4.3)
Now, combining the interaction (4.1) with the gauge-invariant kinetic terms for $`𝒱_\mathrm{\Delta }^{++}`$ and $`𝒱_{\mathrm{YM}}^{++}`$, the complete action becomes
$$S=d\zeta ^{(4)}𝒱_\mathrm{\Delta }^{++}\left\{g_\mathrm{\Delta }^2𝒢_\mathrm{\Delta }^{++}+g_{\mathrm{YM}}^2\mathrm{tr}𝒢_{\mathrm{YM}}^{++}\right\},$$
(4.4)
with $`g_\mathrm{\Delta }`$ and $`g_{\mathrm{YM}}`$ coupling constants. A different form for this action was given in . The theory (4.4) is superconformal at the classical level. It would be interesting to compute, for instance, perturbative quantum corrections.
Let us consider the special case of a single Abelian gauge field, $`𝒱_\mathrm{\Delta }^{++}=𝒱_{\mathrm{YM}}^{++}𝒱^{++}`$. The equations of motion for the corresponding Chern-Simons theory,
$$S_{\mathrm{CS}}=\frac{1}{12g^2}d\zeta ^{(4)}𝒱^{++}𝒢^{++},$$
(4.5)
can be shown to be
$$\mathrm{i}𝒢^{++}=𝒟^{+\widehat{\alpha }}𝒲𝒟_{\widehat{\alpha }}^+𝒲+\frac{1}{2}𝒲(\widehat{𝒟}^+)^2𝒲=0.$$
(4.6)
Using the Bianchi identity (2.31), one can rewrite this in the form
$$D_{\widehat{\alpha }}^+D_{\widehat{\beta }}^+𝒲=\frac{1}{2}\epsilon _{\widehat{\alpha }\widehat{\beta }}\frac{D^{+\widehat{\gamma }}𝒲D_{\widehat{\gamma }}^+𝒲}{𝒲}.$$
(4.7)
From the point of view of 4D, $`𝒩=2`$ supersymmetry, this can be recognized as the off-shell superfield constraint describing the so-called nonlinear vector-tensor multiplet<sup>10</sup><sup>10</sup>10The nonlinear vector-tensor multiplet was discovered in . . Resorting to the two-component spinor notation, eq. (4.7) leads to
$`D_\alpha ^+\overline{D}_{\stackrel{\text{.}}{\alpha }}^+𝒲=0,D^{+\alpha }D_\alpha ^+𝒲={\displaystyle \frac{1}{𝒲}}\left(D^{+\alpha }𝒲D_\alpha ^+𝒲\overline{D}_{\stackrel{\text{.}}{\alpha }}^+𝒲\overline{D}^{+\stackrel{\text{.}}{\alpha }}𝒲\right).`$ (4.8)
In the case of the dynamical system (4.4), the equation of motion for the Abelian gauge field is
$$\frac{1}{\kappa }𝒢_\mathrm{\Delta }^{++}=\mathrm{tr}𝒢_{\mathrm{YM}}^{++},$$
(4.9)
with $`\kappa `$ a coupling constant. With properly defined dimensional reduction 5D $``$ 4D, this can be recognized as the superfield constraint describing the Chern-Simons coupling of a nonlinear vector-tensor to an external $`𝒩=2`$ Yang-Mills supermultiplet .
The super Chern-Simons actions can be readily reduced to components in the Wess-Zumino gauge (2.34) for the Abelian gauge field $`𝒱^{++}`$. If $`L^{++}(z,u)=L^{ij}(z)u_i^+u_j^+`$ is a real analytic superfield of the type (3.14), then
$`S`$ $`=`$ $`{\displaystyle d\zeta ^{(4)}𝒱^{++}L^{++}}`$
$`=`$ $`{\displaystyle }\mathrm{d}^5x\{X^{ij}L_{ij}+{\displaystyle \frac{\mathrm{i}}{12}}\phi \widehat{𝒟}^{ij}L_{ij}+{\displaystyle \frac{\mathrm{i}}{12}}A_{\widehat{a}}(𝒟^i\mathrm{\Gamma }^{\widehat{a}}𝒟^j)L_{ij}{\displaystyle \frac{2}{3}}\mathrm{\Psi }^{i\widehat{\alpha }}𝒟_{\widehat{\alpha }}^jL_{ij}\}\left|\right|.`$
The Abelian supersymmetric Chern-Simons theory (4.5) leads to the following component action:
$`S_{\mathrm{CS}}`$ $`=`$ $`{\displaystyle \frac{1}{2g^2}}{\displaystyle }\mathrm{d}^5x\{{\displaystyle \frac{1}{3}}ϵ^{\widehat{a}\widehat{b}\widehat{c}\widehat{d}\widehat{e}}A_{\widehat{a}}F_{\widehat{b}\widehat{c}}F_{\widehat{d}\widehat{e}}{\displaystyle \frac{1}{2}}\phi F_{\widehat{a}\widehat{b}}F^{\widehat{a}\widehat{b}}\phi _{\widehat{a}}\phi ^{\widehat{a}}\phi +{\displaystyle \frac{1}{2}}\phi X_{ij}X^{ij}`$ (4.11)
$`{\displaystyle \frac{\mathrm{i}}{2}}F_{\widehat{a}\widehat{b}}(\mathrm{\Psi }^k\mathrm{\Sigma }^{\widehat{a}\widehat{b}}\mathrm{\Psi }_k)+\mathrm{i}\phi (\mathrm{\Psi }^k\overline{)}\mathrm{\Psi }_k){\displaystyle \frac{\mathrm{i}}{2}}X_{ij}(\mathrm{\Psi }^i\mathrm{\Psi }^j)\}.`$
## 5 5D Supermultiplets in Reduced Superspace
Some of the results described in the previous sections can easily be reduced to a “hybrid” formulation which keeps manifest only 4D, $`𝒩=1`$ super Poincaré symmetry. As the 5D superfields depend on two sets of 4D anticommuting Majorana spinors, $`(\theta _{\underset{¯}{1}}^\alpha ,\overline{\theta }_{\stackrel{\text{.}}{\alpha }}^{\underset{¯}{1}})`$ and $`(\theta _{\underset{¯}{2}}^\alpha ,\overline{\theta }_{\stackrel{\text{.}}{\alpha }}^{\underset{¯}{2}})`$, such a hybrid formulation is equivalent to integrating out, say, the second set and keeping intact the first set of variables
$$\theta ^\alpha =\theta _{\underset{¯}{1}}^\alpha ,\overline{\theta }_{\stackrel{\text{.}}{\alpha }}=\overline{\theta }_{\stackrel{\text{.}}{\alpha }}^{\underset{¯}{1}}.$$
(5.1)
In this approach, one deals with reduced (or $`𝒩=1`$) superfields $`U|`$, $`D_\alpha ^{\underset{¯}{2}}U|`$, $`\overline{D}_{\underset{¯}{2}}^{\stackrel{\text{.}}{\alpha }}U|,\mathrm{}`$ (of which not all are usually independent) and 4D, $`𝒩=1`$ spinor covariant derivatives $`D_\alpha `$ and $`\overline{D}^{\stackrel{\text{.}}{\alpha }}`$ defined in the obvious way:
$$U|=U(x,\theta _i^\alpha ,\overline{\theta }_{\stackrel{\text{.}}{\alpha }}^i)|_{\theta _{\underset{¯}{2}}=\overline{\theta }^{\underset{¯}{2}}=0},D_\alpha =D_\alpha ^{\underset{¯}{1}}|_{\theta _{\underset{¯}{2}}=\overline{\theta }^{\underset{¯}{2}}=0},\overline{D}^{\stackrel{\text{.}}{\alpha }}=\overline{D}_{\underset{¯}{1}}^{\stackrel{\text{.}}{\alpha }}|_{\theta _{\underset{¯}{2}}=\overline{\theta }^{\underset{¯}{2}}=0}.$$
(5.2)
Our consideration below naturally reproduces many of the 5D supersymmetric models originally derived in the hybrid formulation .
### 5.1 Vector multiplet
Let us introduce reduced gauge covariant derivatives
$$\{𝓓_\alpha ,\overline{𝓓}^{\stackrel{\text{.}}{\alpha }},𝓓_a,𝓓_5\}=\{𝒟_\alpha ^{\underset{¯}{1}},𝒟_{\underset{¯}{1}}^{\stackrel{\text{.}}{\alpha }},𝒟_a,𝒟_5\}|.$$
(5.3)
As follows from (2.3), their algebra is
$`\{𝓓_\alpha ,𝓓_\beta \}=\{\overline{𝓓}_{\stackrel{\text{.}}{\alpha }},\overline{𝓓}_{\stackrel{\text{.}}{\beta }}\}=0,\{𝓓_\alpha ,\overline{𝓓}_{\stackrel{\text{.}}{\beta }}\}=2\mathrm{i}𝓓_{\alpha \stackrel{\text{.}}{\beta }},`$
$`[𝓓_\alpha ,𝓓_{\beta \stackrel{\text{.}}{\beta }}]=2\mathrm{i}\epsilon _{\alpha \beta }\overline{𝒲}_{\stackrel{\text{.}}{\beta }},[\overline{𝓓}_{\stackrel{\text{.}}{\alpha }},𝓓_{\beta \stackrel{\text{.}}{\beta }}]=2\mathrm{i}\epsilon _{\stackrel{\text{.}}{\alpha }\stackrel{\text{.}}{\beta }}𝒲_\beta ,`$
$`[𝓓_\alpha ,𝓓_5+]=0,[\overline{𝓓}_{\stackrel{\text{.}}{\alpha }},𝓓_5]=0,`$ (5.4)
where
$`=𝒲|,𝒲_\alpha =𝒟_\alpha ^{\underset{¯}{2}}𝒲|.`$ (5.5)
It can be seen that the field strengths $``$, $`𝒲_\alpha `$ and $`\overline{𝒲}^{\stackrel{\text{.}}{\alpha }}`$ are the only independent $`𝒩=1`$ descendants of $`𝒲`$.
The strengths $``$ and $`𝒲_\alpha `$ obey some constraints which follow from the Bianchi identities (2.4) and (2.5). Consider first the constraint (2.4) with two derivatives of $`𝒲`$. Taking the part with $`(i,j,\widehat{\alpha },\widehat{\beta })=(\underset{¯}{1},\underset{¯}{1},\alpha ,\stackrel{\text{.}}{\alpha })`$ gives the “$`𝒩=1`$ chirality” of $`𝒲_\alpha `$
$`\overline{𝓓}_{\stackrel{\text{.}}{\alpha }}𝒲_\alpha =0.`$ (5.6)
Taking instead the part with $`(i,j,\widehat{\alpha },\widehat{\beta })=(\underset{¯}{1},\underset{¯}{2},\alpha ,\beta )`$ gives the familiar Bianchi identity
$`𝓓^\alpha 𝒲_\alpha \overline{𝓓}_{\stackrel{\text{.}}{\alpha }}\overline{𝒲}^{\stackrel{\text{.}}{\alpha }}=0.`$ (5.7)
Next, the $`(i,j,\widehat{\alpha },\widehat{\beta })=(\underset{¯}{1},\underset{¯}{1},\alpha ,\beta )`$ and $`(i,j,\widehat{\alpha },\widehat{\beta })=(\underset{¯}{1},\underset{¯}{2},\alpha ,\stackrel{\text{.}}{\alpha })`$ parts, respectively, give
$`\overline{𝒟}_{\underset{¯}{2}\stackrel{\text{.}}{\gamma }}\overline{𝒟}_{\underset{¯}{2}}^{\stackrel{\text{.}}{\gamma }}𝒲|=𝓓^2,𝒟_\alpha ^{\underset{¯}{2}}\overline{𝒟}_{\underset{¯}{2}\stackrel{\text{.}}{\beta }}𝒲|=𝓓_\alpha \overline{𝓓}_{\stackrel{\text{.}}{\alpha }}.`$ (5.8)
The latter identities support the statement that $``$, $`𝒲_\alpha `$ and $`\overline{𝒲}^{\stackrel{\text{.}}{\alpha }}`$ are the only independent $`𝒩=1`$ descendants of $`𝒲`$. Finally, decomposing the second constraint (2.5) with $`(i,j,k)=(\underset{¯}{2},\underset{¯}{2},\underset{¯}{1})`$ and $`(\widehat{\alpha },\widehat{\beta },\widehat{\gamma })=(\dot{\alpha },\dot{\beta },\gamma )`$ gives
$`{\displaystyle \frac{1}{4}}\overline{𝓓}^2𝓓_\alpha +𝓓_5𝒲_\alpha [,𝒲_\alpha ]=0.`$ (5.9)
In accordance with (3.18), the super Yang-Mills action is
$$S_{\mathrm{YM}}=\frac{\mathrm{i}}{12}\mathrm{d}^5x\widehat{𝒟}_{ij}L_{\mathrm{YM}}^{ij}\left|\right|,L_{\mathrm{YM}}^{ij}=\frac{\mathrm{i}}{4}\mathrm{tr}(𝒟^{\widehat{\alpha }i}𝒲𝒟_{\widehat{\alpha }}^j𝒲+\frac{1}{4}\{𝒲,\widehat{𝒟}^{ij}𝒲\}).$$
(5.10)
Its reduced form can be shown to be
$`S_{\mathrm{YM}}`$ $`=`$ $`\mathrm{tr}{\displaystyle \mathrm{d}^5x\left\{\frac{1}{4}\mathrm{d}^2\theta 𝒲^\alpha 𝒲_\alpha +\frac{1}{4}\mathrm{d}^2\overline{\theta }\overline{𝒲}_{\stackrel{\text{.}}{\alpha }}\overline{𝒲}^{\stackrel{\text{.}}{\alpha }}+\mathrm{d}^4\theta ^2\right\}}.`$ (5.11)
Here the Grassmann measures $`\mathrm{d}^2\theta `$ and $`\mathrm{d}^4\theta `$ are part of the chiral and the full superspace measures, respectively, in 4D, $`𝒩=1`$ supersymmetric field theory.
It is instructive to solve the constraints encoded in (5.4). A general solution to the equations $`\{𝓓_\alpha ,𝓓_\beta \}=[𝓓_\alpha ,𝓓_5+]=0`$ is
$$𝓓_\alpha =\mathrm{e}^\mathrm{\Xi }D_\alpha \mathrm{e}^\mathrm{\Xi },𝓓_5+=\mathrm{e}^\mathrm{\Xi }\left(_5+\mathrm{\Phi }^{}\right)\mathrm{e}^\mathrm{\Xi },D_\alpha \mathrm{\Phi }^{}=0,$$
(5.12)
for some Lie-algebra-valued prepotentials $`\mathrm{\Xi }`$ and $`\mathrm{\Phi }^{}`$, of which $`\mathrm{\Xi }`$ is complex unconstrained and $`\mathrm{\Phi }^{}`$ antichiral. Similarly, the constraints $`\{\overline{𝓓}_{\stackrel{\text{.}}{\alpha }},\overline{𝓓}_{\stackrel{\text{.}}{\beta }}\}=[\overline{𝓓}_{\stackrel{\text{.}}{\alpha }},𝓓_5]=0`$ are solved by
$$\overline{𝓓}_{\stackrel{\text{.}}{\alpha }}=\mathrm{e}^\mathrm{\Xi }^{}\overline{D}_{\stackrel{\text{.}}{\alpha }}\mathrm{e}^\mathrm{\Xi }^{},𝓓_5=\mathrm{e}^\mathrm{\Xi }^{}\left(_5\mathrm{\Phi }\right)\mathrm{e}^\mathrm{\Xi }^{},\overline{D}_{\stackrel{\text{.}}{\alpha }}\mathrm{\Phi }=0.$$
(5.13)
The prepotentials introduced possess the following gauge transformations
$`\mathrm{e}^\mathrm{\Xi }^{}`$ $``$ $`\mathrm{e}^{\mathrm{i}\tau (z)}\mathrm{e}^\mathrm{\Xi }^{}\mathrm{e}^{\mathrm{i}\lambda (z)},\mathrm{\Phi }\mathrm{e}^{\mathrm{i}\lambda (z)}\left(\mathrm{\Phi }_5\right)\mathrm{e}^{\mathrm{i}\lambda (z)}1,\overline{D}_{\stackrel{\text{.}}{\alpha }}\lambda =0.`$ (5.14)
Here the $`\lambda `$-gauge group occurs as a result of solving the constraints in terms of the unconstrained prepotentials.
By analogy with the 4D $`𝒩=1`$ super Yang-Mills case, one can introduce a chiral representation defined by applying a complex gauge transformation with $`\tau =\mathrm{\Xi }^{}`$. This gives
$`𝓓_\alpha =\mathrm{e}^VD_\alpha \mathrm{e}^V,`$ $`\overline{𝓓}_{\stackrel{\text{.}}{\alpha }}=\overline{D}_{\stackrel{\text{.}}{\alpha }},`$
$`𝓓_5+=\mathrm{e}^V\left(_5+\mathrm{\Phi }^{}\right)\mathrm{e}^V`$ $`𝓓_5=_5\mathrm{\Phi },`$ (5.15)
where
$`\mathrm{e}^V=\mathrm{e}^\mathrm{\Omega }\mathrm{e}^\mathrm{\Omega }^{},V^{}=V.`$
Here the real Lie-algebra valued superfield $`V`$ is the standard $`𝒩=1`$ super Yang-Mills prepotential. For $``$ we obtain
$$2=\mathrm{\Phi }+\mathrm{e}^V\mathrm{\Phi }^{}\mathrm{e}^V+\mathrm{e}^V(_5\mathrm{e}^V).$$
(5.16)
We have thus reproduced the results obtained by Hebecker within the hybrid approach .
### 5.2 Fayet-Sohnius hypermultiplet
The Fayet-Sohnius hypermultiplet $`𝒒^i`$ generates two independent $`𝒩=1`$ superfields transforming in the same representation of the gauge group,
$`\stackrel{~}{Q}^{}`$ $`=`$ $`𝒒^{\underset{¯}{1}}|,Q=𝒒^{\underset{¯}{2}}|,`$ (5.17)
and obeying the constraints
$$𝓓_\alpha \stackrel{~}{Q}^{}=0,\overline{𝓓}_{\stackrel{\text{.}}{\alpha }}Q=0.$$
(5.18)
These constraints follow from (2.40). Thus $`Q`$ and $`\stackrel{~}{Q}^{}`$ are covariantly chiral and antichiral, respectively. The central charge transformation of these superfields is:
$`\mathrm{i}\mathrm{\Delta }Q`$ $`=`$ $`{\displaystyle \frac{1}{4}}\overline{𝓓}^2\stackrel{~}{Q}^{}+(𝓓_5)Q,\mathrm{i}\mathrm{\Delta }\stackrel{~}{Q}^{}={\displaystyle \frac{1}{4}}𝓓^2Q+(+𝓓_5)\stackrel{~}{Q}^{}.`$ (5.19)
In accordance with (3.16), the action for the Fayet-Sohnius hypermultiplet is
$`S_{\mathrm{FS}}`$ $`=`$ $`{\displaystyle \frac{\mathrm{i}}{12}}{\displaystyle }\mathrm{d}^5x\widehat{𝒟}_{ij}L_{\mathrm{FS}}^{ij}|,L_{\mathrm{FS}}^{ij}=({\displaystyle \frac{1}{2}}\overline{𝒒}^{(i}\stackrel{}{\mathrm{\Delta }}𝒒^{j)}\mathrm{i}m\overline{𝒒}^{(i}𝒒^{j)}).`$ (5.20)
It can be shown to reduce to the following $`𝒩=1`$ action
$`S_{\mathrm{FS}}`$ $`=`$ $`{\displaystyle }\mathrm{d}^5x\{{\displaystyle }\mathrm{d}^4\theta (Q^{}Q+\stackrel{~}{Q}\stackrel{~}{Q}^{})+({\displaystyle }\mathrm{d}^2\theta \stackrel{~}{Q}(𝓓_5+m)Q+\mathrm{c}.\mathrm{c}.)\}.`$ (5.21)
As follows from (5.4), the operator $`𝓓_5`$ preserves chirality.
## 6 Projective Superspace and Dimensional Reduction
Throughout this section, we consider only 5D supermultiplets without central charge, $`\mathrm{\Delta }=0`$. However, many results below can be extended to include the case $`\mathrm{\Delta }0`$.
### 6.1 Doubly punctured harmonic superspace
Let $`𝝍^{(p)}(z,u)`$ be a harmonic superfield of non-negative U(1) charge $`p`$. Here we will be interested in solutions to the equation
$$D^{++}𝝍^{(p)}=0D^{++}D_{\widehat{\alpha }}^+𝝍^{(p)}=0,p0.$$
(6.1)
If $`𝝍^{(p)}(z,u)`$ is globally defined and smooth over $`^{5|8}\times S^2`$, it possesses a convergent Fourier series of the form (B.9). If $`𝝍^{(p)}(z,u)`$ is further constrained to obey the equation (6.1), then its general form becomes
$$𝝍^{(p)}(z,u)=𝝍^{i_1\mathrm{}i_p}(z)u_{i_1}^+\mathrm{}u_{i_p}^+.$$
(6.2)
Therefore, such a globally defined harmonic superfield possesses finitely many component fields, and this can thought of as a consequence of the Riemann-Roch theorem specified to the case of $`S^2`$. A more interesting situation occurs if one allows $`𝝍^{(p)}(z,u)`$ to have a few singularities on $`S^2`$.
For further consideration, it is useful to cover $`S^2`$ by two charts and introduce local complex coordinates in each chart, as defined in Appendix B. In the north chart (parametrized by the complex variable $`w`$ and its conjugate $`\overline{w}`$) we can represent $`𝝍^{(p)}(z,u)`$ as follows
$$𝝍^{(p)}(z,u)=(u^{+\underset{¯}{1}})^p\psi (z,w,\overline{w}).$$
(6.3)
If $`𝝍^{(p)}(z,u)`$ is globally defined over $`^{5|8}\times S^2`$, then $`\psi (z,w,\overline{w})𝝍_\mathrm{N}^{(p)}(z,w,\overline{w})`$ is given as in eq. (B.10). It is a simple exercise to check that
$$D^{++}𝝍^{(p)}(z,u)=(u^{+\underset{¯}{1}})^{p+2}(1+w\overline{w})^2_{\overline{w}}\psi (z,w,\overline{w}),p0,$$
(6.4)
and therefore
$`D^{++}𝝍^{(p)}=0,p0_{\overline{w}}\psi =0.`$ (6.5)
Assuming that $`𝝍^{(p)}(z,u)`$ may possess singularities only at the north and south poles of $`S^2`$, we then conclude that
$$\psi (z,w)=\underset{n=\mathrm{}}{\overset{+\mathrm{}}{}}\psi _n(z)w^n.$$
(6.6)
Now, consider an analytic superfield $`\mathit{\varphi }^{(p)}`$ obeying the constraint (6.1).
$$D_{\widehat{\alpha }}^+\mathit{\varphi }^{(p)}=0,D^{++}\mathit{\varphi }^{(p)}=0,p0.$$
(6.7)
We assume that $`\mathit{\varphi }^{(p)}(z,u)`$ is non-singular outside the north and south poles of $`S^2`$. Then, representing $`\mathit{\varphi }^{(p)}(z,u)=(u^{+\underset{¯}{1}})^p\varphi (z,w,\overline{w})`$ and defining
$$D_{\widehat{\alpha }}^+=u^{+\underset{¯}{1}}_{\widehat{\alpha }}(w),_{\widehat{\alpha }}(w)=D_{\widehat{\alpha }}^iw_i,w_i=(w,1),$$
(6.8)
eq. (6.7) is solved as
$$_{\widehat{\alpha }}(w)\varphi (z,w)=0,\varphi (z,w)=\underset{n=\mathrm{}}{\overset{+\mathrm{}}{}}\varphi _n(z)w^n.$$
(6.9)
These relations define a projective superfield, following the four-dimensional terminology . Since the supersymmetry transformations act simply as the identity transformation on $`S^2`$, the above consideration clearly defines supermultiplets. Such supermultiplets turn out to be most suited for dimensional reduction.
The projective analogue of the smile-conjugation (2.21) is
$`\stackrel{˘}{\varphi }(z,w)={\displaystyle \underset{n=\mathrm{}}{\overset{+\mathrm{}}{}}}(1)^n\overline{\varphi }_n(z)w^n,_{\widehat{\alpha }}(w)\stackrel{˘}{\varphi }(z,w)=0.`$ (6.10)
If $`\stackrel{˘}{\varphi }(z,w)=\varphi (z,w)`$, the projective superfield is called real. The projective conjugation (6.10) can be derived from the smile-conjugation (2.21), see for details.
There are several types of projective superfields . A real projective superfield of the form (7.11) is called a tropical multiplet. A real projective superfield of the form
$$\varphi (z,w)=\underset{n}{\overset{+n}{}}\varphi _n(z)w^n,\stackrel{˘}{\varphi }=\varphi ,n$$
(6.11)
is called a real O($`2n`$) multiplet.<sup>11</sup><sup>11</sup>11One can also introduce complex O($`2n`$+1) multiplets . A projective superfield $`\mathrm{{\rm Y}}(z,w)`$ of the form (6.29) is called an arctic multiplet, and its conjugate, $`\stackrel{˘}{\mathrm{{\rm Y}}}(z,w)`$, an antarctic multiplet. The $`\mathrm{{\rm Y}}(z,w)`$ and $`\stackrel{˘}{\mathrm{{\rm Y}}}(z,w)`$ constitute a polar multiplet. More general projective superfields occur if one multiplies any of the considered superfields by $`w^n`$, with $`n`$ an integer.
At this stage, it is useful to break the manifest 5D Lorentz invariance by switching from the four-component spinor notation to the two-component one. Representing
$`_{\widehat{\alpha }}(w)=\left(\begin{array}{c}_\alpha (w)\\ \overline{}^{\stackrel{\text{.}}{\alpha }}(w)\end{array}\right),_\alpha (w)wD_\alpha ^{\underset{¯}{1}}D_\alpha ^{\underset{¯}{2}},\overline{}^{\stackrel{\text{.}}{\alpha }}(w)\overline{D}_{\underset{¯}{1}}^{\stackrel{\text{.}}{\alpha }}+w\overline{D}_{\underset{¯}{2}}^{\stackrel{\text{.}}{\alpha }},`$ (6.14)
the constraints (6.10) can be rewritten in the component form
$$D_\alpha ^{\underset{¯}{2}}\varphi _n=D_\alpha ^{\underset{¯}{1}}\varphi _{n1},\overline{D}_{\underset{¯}{2}}^{\stackrel{\text{.}}{\alpha }}\varphi _n=\overline{D}_{\underset{¯}{1}}^{\stackrel{\text{.}}{\alpha }}\varphi _{n+1}.$$
(6.15)
In accordance with (A.32) and (A.33), one can think of the operators $`D_A=(_a,D_\alpha ^i,\overline{D}_i^{\stackrel{\text{.}}{\alpha }})`$, where $`a=0,1,2,3`$, as the covariant derivatives of 4D, $`𝒩=2`$ central charge superspace, with $`x^5`$ being the central charge variable. The relations (6.15) imply that the dependence of the component superfields $`\varphi _n`$ on $`\theta _{\underset{¯}{2}}^\alpha `$ and $`\overline{\theta }_{\stackrel{\text{.}}{\alpha }}^{\underset{¯}{2}}`$ is uniquely determined in terms of their dependence on $`\theta _{\underset{¯}{1}}^\alpha `$ and $`\overline{\theta }_{\stackrel{\text{.}}{\alpha }}^{\underset{¯}{1}}`$. In other words, the projective superfields depend effectively on half the Grassmann variables which can be choosen to be the spinor coordinates of 4D, $`𝒩=1`$ superspace (5.1). In other words, it is sufficient to work with reduced superfields $`\varphi (w)|`$ and 4D, $`𝒩=1`$ spinor covariant derivatives $`D_\alpha `$ and $`\overline{D}^{\stackrel{\text{.}}{\alpha }}`$ defined in (5.2).
If the series in (6.9) is bounded from below (above), then eq. (6.15) implies that the two lowest (highest) components in $`\varphi (w)|`$ are constrained $`𝒩=1`$ superfields. For example, in the case of the arctic multiplet, eq. (6.29), the leading components $`\mathrm{\Phi }=\mathrm{{\rm Y}}_0|`$ and $`\mathrm{\Gamma }=\mathrm{{\rm Y}}_1|`$ obey the constraints (6.30).
Given a real projective superfield $`L(z,w)`$, one can construct a supersymmetric invariant
$$S=\frac{1}{2\pi \mathrm{i}}_C\frac{\mathrm{d}w}{w}\mathrm{d}^5x\mathrm{d}^4\theta L(w)|\frac{1}{2\pi \mathrm{i}}_C\frac{\mathrm{d}w}{w}S(w),$$
(6.16)
with $`C`$ a contour around the origin (in what follows, such a contour is always assumed). For $`S(w)`$ there are several equivalent forms:
$`S(w)={\displaystyle \frac{1}{16}}{\displaystyle }\mathrm{d}^5xD^2\overline{D}^2L(z,w)\left|\right|={\displaystyle \frac{1}{16}}{\displaystyle }\mathrm{d}^5x(D^{\underset{¯}{1}})^2(\overline{D}_{\underset{¯}{1}})^2L(z,w)\left|\right|`$ (6.17)
assuming only that total space-time derivatives do not contribute. The invariance of $`S(w)`$ under arbitrary SUSY transformations is easy to demonstrate. Defining
$$D^4=\frac{1}{16}(D^{\underset{¯}{1}})^2(\overline{D}_{\underset{¯}{1}})^2,$$
(6.18)
one can argue as follows:
$`\delta S(w)`$ $`=`$ $`\mathrm{i}{\displaystyle }\mathrm{d}^5x(\epsilon _i^\alpha Q_\alpha ^i+\overline{\epsilon }_{\stackrel{\text{.}}{\alpha }}^i\overline{Q}_i^{\stackrel{\text{.}}{\alpha }})D^4L(z,w)\left|\right|={\displaystyle }\mathrm{d}^5x(\epsilon _i^\alpha D_\alpha ^i+\overline{\epsilon }_{\stackrel{\text{.}}{\alpha }}^i\overline{D}_i^{\stackrel{\text{.}}{\alpha }})D^4L(z,w)\left|\right|`$ (6.19)
$`=`$ $`{\displaystyle }\mathrm{d}^5x(\epsilon _{\underset{¯}{2}}^\alpha D_\alpha ^{\underset{¯}{2}}+\overline{\epsilon }_{\stackrel{\text{.}}{\alpha }}^{\underset{¯}{2}}\overline{D}_{\underset{¯}{2}}^{\stackrel{\text{.}}{\alpha }})D^4L(z,w)\left|\right|={\displaystyle }\mathrm{d}^5xD^4(\epsilon _{\underset{¯}{2}}^\alpha D_\alpha ^{\underset{¯}{2}}+\overline{\epsilon }_{\stackrel{\text{.}}{\alpha }}^{\underset{¯}{2}}\overline{D}_{\underset{¯}{2}}^{\stackrel{\text{.}}{\alpha }})L(z,w)\left|\right|`$
$`=`$ $`{\displaystyle }\mathrm{d}^5xD^4(\epsilon _2^\alpha D_\alpha ^{\underset{¯}{1}}w\overline{\epsilon }_{\stackrel{\text{.}}{\alpha }}^{\underset{¯}{2}}\overline{D}_{\underset{¯}{1}}^{\stackrel{\text{.}}{\alpha }}w^1)L(z,w)\left|\right|=0,`$
with $`Q_\alpha ^i`$ and $`\overline{Q}_i^{\stackrel{\text{.}}{\alpha }}`$ the supersymmetry generators.
### 6.2 Tensor multiplet and nonlinear sigma-models
The tensor multiplet (also called O(2) multiplet) is described by a constrained real analytic superfield $`\mathrm{\Xi }^{++}`$:
$$D_{\widehat{\alpha }}^+\mathrm{\Xi }^{++}=0,D^{++}\mathrm{\Xi }^{++}=0.$$
(6.20)
The corresponding projective superfield $`\mathrm{\Xi }(z,w)`$ is defined by $`\mathrm{\Xi }^{++}(z,u)=\mathrm{i}u^{+\underset{¯}{1}}u^{+\underset{¯}{2}}\mathrm{\Xi }(z,w)`$. Without distinguishing between $`\mathrm{\Xi }(z,w)`$ and $`\mathrm{\Xi }(z,w)|`$, we have
$$\mathrm{\Xi }(w)=\mathrm{\Phi }+wGw^2\overline{\mathrm{\Phi }},\overline{G}=G,$$
(6.21)
where the component superfields obey the constraints
$$\overline{D}^{\stackrel{\text{.}}{\alpha }}\mathrm{\Phi }=0,\frac{1}{4}\overline{D}^2G=_5\mathrm{\Phi }.$$
(6.22)
Here we consider a 5D generalization of the 4D, $`𝒩=2`$ supersymmetric nonlinear sigma-model<sup>12</sup><sup>12</sup>12The construction given in has recently been reviewed and extended in . studied in and related to the so-called $`c`$-map . Let $`F`$ be a holomorphic function of $`n`$ variables. Associated with this function is the following supersymmetric action
$$S=\mathrm{d}^5x\mathrm{d}^4\theta [\frac{1}{2\pi \mathrm{i}}\frac{\mathrm{d}w}{w}\frac{F(\mathrm{\Xi }^I(w))}{w^2}+\mathrm{c}.\mathrm{c}.].$$
(6.23)
Since
$`F(\mathrm{\Xi }^I(w))`$ $`=`$ $`F\left(\mathrm{\Phi }^I+wG^Iw^2\overline{\mathrm{\Phi }}^I\right)`$
$`=`$ $`F(\mathrm{\Phi })+wF_I(\mathrm{\Phi })G^Iw^2\left(F_I(\mathrm{\Phi })\overline{\mathrm{\Phi }}^I{\displaystyle \frac{1}{2}}F_{IJ}(\mathrm{\Phi })G^IG^J\right)+O(w^3),`$
the contour integral is trivial to do. The action is equivalent to
$$S[\mathrm{\Phi },\overline{\mathrm{\Phi }},G]=\mathrm{d}^5x\mathrm{d}^4\theta \left\{K(\mathrm{\Phi },\overline{\mathrm{\Phi }})\frac{1}{2}g_{I\overline{J}}(\mathrm{\Phi },\overline{\mathrm{\Phi }})G^IG^J\right\},$$
(6.24)
where
$$K(\mathrm{\Phi },\overline{\mathrm{\Phi }})=\overline{\mathrm{\Phi }}^IF_I(\mathrm{\Phi })+\mathrm{\Phi }^I\overline{F}_I(\overline{\mathrm{\Phi }}),g_{I\overline{J}}(\mathrm{\Phi },\overline{\mathrm{\Phi }})=F_{IJ}(\mathrm{\Phi })+\overline{F}_{IJ}(\overline{\mathrm{\Phi }}).$$
(6.25)
The Kähler potential $`K(\mathrm{\Phi },\overline{\mathrm{\Phi }})`$ generates the so-called rigid special Kähler geometry .
Let us work out a dual formulation for the theory (6.24). Introduce a first-order action
$`S[\mathrm{\Phi },\overline{\mathrm{\Phi }},G]+{\displaystyle }\mathrm{d}^5x\{{\displaystyle }\mathrm{d}^2\theta \mathrm{\Psi }_I(_5\mathrm{\Phi }^I+{\displaystyle \frac{1}{4}}\overline{D}^2G^I)+\mathrm{c}.\mathrm{c}.\}`$ (6.26)
$`=`$ $`S[\mathrm{\Phi },\overline{\mathrm{\Phi }},G]{\displaystyle }\mathrm{d}^5x\{{\displaystyle }\mathrm{d}^4\theta (\mathrm{\Psi }_I+\overline{\mathrm{\Psi }}_I)G^I+({\displaystyle }\mathrm{d}^2\theta \mathrm{\Psi }_I_5\mathrm{\Phi }^I+\mathrm{c}.\mathrm{c}.)\},`$
where the superfield $`G^I`$ is now real unconstrained, while $`\mathrm{\Psi }_I`$ is chiral, $`\overline{D}_{\stackrel{\text{.}}{\alpha }}\mathrm{\Psi }_I=0`$. In this action we can integrate out $`G^I`$ using the corresponding equations of motion. This gives
$`S[\mathrm{\Phi },\overline{\mathrm{\Phi }},\mathrm{\Psi },\overline{\mathrm{\Psi }}]={\displaystyle }\mathrm{d}^5x\{{\displaystyle }\mathrm{d}^4\theta H(\mathrm{\Phi },\overline{\mathrm{\Phi }},\mathrm{\Psi },\overline{\mathrm{\Psi }})+({\displaystyle }\mathrm{d}^2\theta \mathrm{\Psi }_I_5\mathrm{\Phi }^I+\mathrm{c}.\mathrm{c}.)\},`$ (6.27)
where
$$H(\mathrm{\Phi },\overline{\mathrm{\Phi }},\mathrm{\Psi },\overline{\mathrm{\Psi }})=K(\mathrm{\Phi },\overline{\mathrm{\Phi }})+\frac{1}{2}g^{I\overline{J}}(\mathrm{\Phi },\overline{\mathrm{\Phi }})(\mathrm{\Psi }_I+\overline{\mathrm{\Psi }}_I)(\mathrm{\Psi }_J+\overline{\mathrm{\Psi }}_J).$$
(6.28)
The potential $`H(\mathrm{\Phi },\overline{\mathrm{\Phi }},\mathrm{\Psi },\overline{\mathrm{\Psi }})`$ is the Kähler potential of a hyper Kähler manifold. By construction, this potential is generated by another Kähler potential, $`K(\mathrm{\Phi },\overline{\mathrm{\Phi }})`$, which is associated with the holomorphic function $`F(\mathrm{\Phi })`$ defining the rigid special Kähler geometry . The correspondence $`K(\mathrm{\Phi },\overline{\mathrm{\Phi }})H(\mathrm{\Phi },\overline{\mathrm{\Phi }},\mathrm{\Psi },\overline{\mathrm{\Psi }})`$ is called the rigid $`c`$-map .
### 6.3 Polar hypermultiplet and nonlinear sigma-models
According to , the polar hypermultiplet is generated by projective superfields
$$\mathrm{{\rm Y}}(z,w)=\underset{n=0}{\overset{\mathrm{}}{}}\mathrm{{\rm Y}}_n(z)w^n,\stackrel{˘}{\mathrm{{\rm Y}}}(z,w)=\underset{n=0}{\overset{\mathrm{}}{}}(1)^n\overline{\mathrm{{\rm Y}}}_n(z)\frac{1}{w^n}.$$
(6.29)
The projective superfields $`\mathrm{{\rm Y}}`$ and $`\stackrel{˘}{\mathrm{{\rm Y}}}`$ are called arctic and antarctic , respectively. The constraints (6.15) imply that the leading components $`\mathrm{\Phi }=\mathrm{{\rm Y}}_0|`$ and $`\mathrm{\Gamma }=\mathrm{{\rm Y}}_1|`$ are constrained
$$\overline{D}^{\stackrel{\text{.}}{\alpha }}\mathrm{\Phi }=0,\frac{1}{4}\overline{D}^2\mathrm{\Gamma }=_5\mathrm{\Phi }.$$
(6.30)
The other components of $`\mathrm{{\rm Y}}(w)`$ are complex unconstrained superfields, and they appear to be non-dynamical (auxiliary) in models with at most two space-time derivatives at the component level.
Here we consider a 5D generalization of the 4D, $`𝒩=2`$ supersymmetric nonlinear sigma-model studied in . It is described by the action
$$S[\mathrm{{\rm Y}},\stackrel{˘}{\mathrm{{\rm Y}}}]=\mathrm{d}^5x\mathrm{d}^4\theta \left[\frac{1}{2\pi \mathrm{i}}\frac{\mathrm{d}w}{w}K(\mathrm{{\rm Y}}(w),\stackrel{˘}{\mathrm{{\rm Y}}}(w))\right].$$
(6.31)
This 5D supersymmetric sigma-model respects all the geometric features of its 4D, $`𝒩=1`$ predecessor
$$S[\mathrm{\Phi },\overline{\mathrm{\Phi }}]=\mathrm{d}^4x\mathrm{d}^4\theta K(\mathrm{\Phi },\overline{\mathrm{\Phi }}),$$
(6.32)
where $`K(A,\overline{A})`$ is the Kähler potential of some manifold $``$. The Kähler invariance of (6.32)
$$K(\mathrm{\Phi },\overline{\mathrm{\Phi }})K(\mathrm{\Phi },\overline{\mathrm{\Phi }})+\left(\mathrm{\Lambda }(\mathrm{\Phi })+\overline{\mathrm{\Lambda }}(\overline{\mathrm{\Phi }})\right)$$
(6.33)
turns into
$$K(\mathrm{{\rm Y}},\stackrel{˘}{\mathrm{{\rm Y}}})K(\mathrm{{\rm Y}},\stackrel{˘}{\mathrm{{\rm Y}}})+\left(\mathrm{\Lambda }(\mathrm{{\rm Y}})+\overline{\mathrm{\Lambda }}(\stackrel{˘}{\mathrm{{\rm Y}}})\right)$$
(6.34)
for the model (6.31). A holomorphic reparametrization $`A^If^I\left(A\right)`$ of the Kähler manifold has the following counterparts
$$\mathrm{\Phi }^If^I\left(\mathrm{\Phi }\right),\mathrm{{\rm Y}}^I(w)f^I\left(\mathrm{{\rm Y}}(w)\right)$$
(6.35)
in the 4D and 5D cases, respectively. Therefore, the physical superfields of the 5D theory
$$\mathrm{{\rm Y}}^I(w)|_{w=0}=\mathrm{\Phi }^I,\frac{\mathrm{d}\mathrm{{\rm Y}}^I(w)}{\mathrm{d}w}|_{w=0}=\mathrm{\Gamma }^I,$$
(6.36)
should be regarded, respectively, as a coordinate of the Kähler manifold and a tangent vector at point $`\mathrm{\Phi }`$ of the same manifold. That is why the variables $`(\mathrm{\Phi }^I,\mathrm{\Gamma }^J)`$ parametrize the tangent bundle $`T`$ of the Kähler manifold $``$.
The auxiliary superfields $`\mathrm{{\rm Y}}_2,\mathrm{{\rm Y}}_3,\mathrm{}`$, and their conjugates, can be eliminated with the aid of the corresponding algebraic equations of motion
$$dww^{n1}\frac{K(\mathrm{{\rm Y}},\stackrel{˘}{\mathrm{{\rm Y}}})}{\mathrm{{\rm Y}}^I}=0,n2.$$
(6.37)
Their elimination can be carried out using the ansatz
$`\mathrm{{\rm Y}}_n^I={\displaystyle \underset{p=o}{\overset{\mathrm{}}{}}}G^I{}_{J_1\mathrm{}J_{n+p}\overline{L}_1\mathrm{}\overline{L}_p}{}^{}(\mathrm{\Phi },\overline{\mathrm{\Phi }})\mathrm{\Gamma }^{J_1}\mathrm{}\mathrm{\Gamma }^{J_{n+p}}\overline{\mathrm{\Gamma }}^{\overline{L}_1}\mathrm{}\overline{\mathrm{\Gamma }}^{\overline{L}_p},n2.`$ (6.38)
Upon elimination of the auxiliary superfields,<sup>13</sup><sup>13</sup>13As explained in , the auxiliary superfields can be eliminated only perturbatively for general Kähler manifolds. This agrees with a theorem proved in that, for a Kähler manifold $``$, a canonical hyper-Kähler structure exists, in general, on an open neighborhood of the zero section of the cotangent bundle $`T^{}`$. It was further demonstrated in that the auxiliary superfields can be eliminated in the case of compact Kähler symmetric spaces. the action (6.31) takes the form
$`S_{\mathrm{tb}}[\mathrm{\Phi },\overline{\mathrm{\Phi }},\mathrm{\Gamma },\overline{\mathrm{\Gamma }}]`$ $`=`$ $`{\displaystyle }\mathrm{d}^5x\mathrm{d}^4\theta \{K(\mathrm{\Phi },\overline{\mathrm{\Phi }})g_{I\overline{J}}(\mathrm{\Phi },\overline{\mathrm{\Phi }})\mathrm{\Gamma }^I\overline{\mathrm{\Gamma }}^{\overline{J}}`$ (6.39)
$`+{\displaystyle \underset{p=2}{\overset{\mathrm{}}{}}}_{I_1\mathrm{}I_p\overline{J}_1\mathrm{}\overline{J}_p}(\mathrm{\Phi },\overline{\mathrm{\Phi }})\mathrm{\Gamma }^{I_1}\mathrm{}\mathrm{\Gamma }^{I_p}\overline{\mathrm{\Gamma }}^{\overline{J}_1}\mathrm{}\overline{\mathrm{\Gamma }}^{\overline{J}_p}\},`$
where the tensors $`_{I_1\mathrm{}I_p\overline{J}_1\mathrm{}\overline{J}_p}`$ are functions of the Riemann curvature $`R_{I\overline{J}K\overline{L}}(\mathrm{\Phi },\overline{\mathrm{\Phi }})`$ and its covariant derivatives. Each term in the action contains equal powers of $`\mathrm{\Gamma }`$ and $`\overline{\mathrm{\Gamma }}`$, since the original model (6.31) is invariant under rigid U(1) transformations
$$\mathrm{{\rm Y}}(w)\mathrm{{\rm Y}}(\mathrm{e}^{\mathrm{i}\alpha }w)\mathrm{{\rm Y}}_n(z)\mathrm{e}^{\mathrm{i}n\alpha }\mathrm{{\rm Y}}_n(z).$$
(6.40)
For the theory with action $`S_{\mathrm{tb}}[\mathrm{\Phi },\overline{\mathrm{\Phi }},\mathrm{\Gamma },\overline{\mathrm{\Gamma }}]`$, we can develop a dual formulation involving only chiral superfields and their conjugates as the dynamical variables. Consider the first-order action
$`S_{\mathrm{tb}}[\mathrm{\Phi },\overline{\mathrm{\Phi }},\mathrm{\Gamma },\overline{\mathrm{\Gamma }}]{\displaystyle }\mathrm{d}^5x\{{\displaystyle }\mathrm{d}^2\theta \mathrm{\Psi }_I(_5\mathrm{\Phi }^I+{\displaystyle \frac{1}{4}}\overline{D}^2\mathrm{\Gamma }^I)+\mathrm{c}.\mathrm{c}.\}`$ (6.41)
$`=`$ $`S_{\mathrm{tb}}[\mathrm{\Phi },\overline{\mathrm{\Phi }},\mathrm{\Gamma },\overline{\mathrm{\Gamma }}]+{\displaystyle }\mathrm{d}^5x\{{\displaystyle }\mathrm{d}^4\theta \mathrm{\Psi }_I\mathrm{\Gamma }^I{\displaystyle }\mathrm{d}^2\theta \mathrm{\Psi }_I_5\mathrm{\Phi }^I+\mathrm{c}.\mathrm{c}.\},`$
where the tangent vector $`\mathrm{\Gamma }^I`$ is now complex unconstrained, while the one-form $`\mathrm{\Psi }_I`$ is chiral, $`\overline{D}_{\stackrel{\text{.}}{\alpha }}\mathrm{\Psi }_I=0`$. Upon elimination of $`\mathrm{\Gamma }`$ and $`\overline{\mathrm{\Gamma }}`$, with the aid of their equations of motion, the action turns into $`S_{\mathrm{cb}}[\mathrm{\Phi },\overline{\mathrm{\Phi }},\mathrm{\Psi },\overline{\mathrm{\Psi }}]`$. Its target space is the cotangent bundle $`T^{}`$ of the Kähler manifold $``$.
It is instructive to consider a free hypermultiplet described by the Kähler potential $`K_{\mathrm{free}}(A,\overline{A})=\overline{A}A`$. Then
$`S_{\mathrm{free}}[\mathrm{{\rm Y}},\stackrel{˘}{\mathrm{{\rm Y}}}]={\displaystyle \mathrm{d}^5x\mathrm{d}^4\theta \underset{n=0}{\overset{\mathrm{}}{}}(1)^n\overline{\mathrm{{\rm Y}}}_n(z)\mathrm{{\rm Y}}_n}={\displaystyle \mathrm{d}^5x\mathrm{d}^4\theta \left(\overline{\mathrm{\Phi }}\mathrm{\Phi }\overline{\mathrm{\Gamma }}\mathrm{\Gamma }\right)}+\mathrm{}`$ (6.42)
Here the dots stand for the auxiliary superfields’ contribution. Now, eliminating the auxiliary superfields and dualizing $`\mathrm{\Gamma }`$ into a chiral scalar, one obtains the action for the free Fayet-Sohnius hypermultiplet, equation (5.21).
## 7 Vector Multiplet in Projective Superspace
In the Abelian case, the gauge transformation (2.33) simplifies
$$\delta 𝒱^{++}=D^{++}\lambda ,D_{\widehat{\alpha }}^+\lambda =0,\stackrel{˘}{\lambda }=\lambda .$$
(7.1)
The field strength (2.30) also simplifies
$$𝒲=\frac{\mathrm{i}}{8}du(\widehat{D}^{})^2𝒱^{++}.$$
(7.2)
It is easy to see that $`𝒲`$ is gauge invariant.
The gauge freedom (7.1) can be used to choose the supersymmetric Lorentz gauge
$$D^{++}𝒱^{++}=0.$$
(7.3)
In other words, in this gauge $`𝒱^{++}`$ becomes a real O(2) multiplet,
$$𝒱^{++}=\mathrm{i}u^{+\underset{¯}{1}}u^{+\underset{¯}{2}}V(z,w),V(z,w)=\frac{1}{w}\phi (z)+V(z)w\overline{\phi }(z).$$
(7.4)
Since $`𝒲`$ is gauge invariant, for its evaluation one can use any potential $`𝒱^{++}`$ from the same gauge orbit, in particular the one obeying the gauge condition (7.3). This Lorentz gauge is particularly useful for our consideration. Using the relation (C.6) and noting that $`|u^{+\underset{¯}{1}}|^2=(1+w\overline{w})^1`$, we can rewrite $`𝒲`$ in the form
$$𝒲=\frac{1}{2}du𝒫(w)V(z,w).$$
(7.5)
This can be further transformed to
$$𝒲=\frac{1}{4\pi \mathrm{i}}\frac{\mathrm{d}w}{w}𝒫(w)V(z,w).$$
(7.6)
Indeed, the consideration in Appendix C justifies the following identity
$`\underset{R\mathrm{}}{lim}\underset{ϵ0}{lim}{\displaystyle du\varphi _{R,ϵ}(u)}={\displaystyle \frac{1}{2\pi \mathrm{i}}}{\displaystyle \frac{\mathrm{d}w}{w}\varphi (w)},`$ (7.7)
with the regularization $`\varphi _{R,ϵ}(u)=\varphi _{R,ϵ}(w,\overline{w})`$ of a function $`\varphi (w)`$ holomorphic on $`^{}`$ defined according to (C.2). Since the integrand on the right of (7.5) is, by construction, a smooth scalar field on $`S^2`$, we obvoiusly have
$$du𝒫(w)V(z,w)=\underset{R\mathrm{}}{lim}\underset{ϵ0}{lim}du𝒫(w)V_{R,ϵ}(z,u).$$
(7.8)
The representation (7.6) allows one to obtain a new formulation for the vector multiplet. Let $`\mathrm{\Lambda }(z,w)`$ be an arctic multiplet
$$\mathrm{\Lambda }(z,w)=\underset{n=0}{\overset{\mathrm{}}{}}\mathrm{\Lambda }_n(z)w^n,_{\widehat{\alpha }}(w)\mathrm{\Lambda }(z,w)=0,$$
(7.9)
and $`\stackrel{˘}{\mathrm{\Lambda }}(z,w)`$ its smile-conjugate. It then immediately follows that
$$\frac{\mathrm{d}w}{w}𝒫(w)\mathrm{\Lambda }(z,w)=\frac{\mathrm{d}w}{w}𝒫(w)\stackrel{˘}{\mathrm{\Lambda }}(z,w)=0.$$
(7.10)
Now, introduce a real tropical multiplet $`V(z,w)`$,
$$V(z,w)=\underset{n=\mathrm{}}{\overset{+\mathrm{}}{}}V_n(z)w^n,_{\widehat{\alpha }}(w)V(z,w)=0,\overline{V}_n=(1)^nV_n,$$
(7.11)
possessing the gauge freedom
$$\delta V(z,w)=\mathrm{i}\left(\stackrel{˘}{\mathrm{\Lambda }}(z,w)\mathrm{\Lambda }(z,w)\right).$$
(7.12)
With such gauge transformations, eq. (7.6) defines a gauge invariant field strength. Next, in accordance with the superfield structure of the tropical and arctic multiplets, the gauge freedom can be used to turn $`V(z,w)`$ into a real O(2) multiplet, i.e. to bring $`V(z,w)`$ to the form (7.4). We thus arrive at the projective superspace formulation<sup>14</sup><sup>14</sup>14An alternative procedure to deduce the projective superspace formulation for the 4D, $`𝒩=2`$ vector multiplet from the corresponding harmonic superspace formulation can be found in . for the vector multiplet .
Now, we are in a position to evaluate the $`𝒩=1`$ field strengths (5.5) in terms of the prepotentials $`V_n`$. It follows from (C.6) that
$`=𝒲|={\displaystyle \frac{1}{4\pi \mathrm{i}}}{\displaystyle }{\displaystyle \frac{\mathrm{d}w}{w}}𝒫(w)V(w)|={\displaystyle \frac{1}{2}}(\mathrm{\Phi }+\overline{\mathrm{\Phi }}+_5V),`$ (7.13)
where we have defined
$$\mathrm{\Phi }=\frac{1}{4}\overline{D}^2V_1|,V=V_0|.$$
(7.14)
The spinor field strength $`𝒲_\alpha `$ is given by
$`𝒲_\alpha (z)=D_\alpha ^{\underset{¯}{2}}𝒲|={\displaystyle \frac{1}{4\pi \mathrm{i}}}{\displaystyle }{\displaystyle \frac{\mathrm{d}w}{w}}([D_\alpha ^{\underset{¯}{2}},𝒫(w)]+𝒫(w)D_\alpha ^{\underset{¯}{2}})V(w)|.`$ (7.15)
However, as $`[D_\alpha ^{\underset{¯}{2}},𝒫(w)]=w_5D_\alpha ^{\underset{¯}{1}}`$ and given that for any projective superfield $`\varphi (w)`$ we have $`D_\alpha ^{\underset{¯}{2}}\varphi (w)=wD_\alpha ^{\underset{¯}{1}}\varphi (w)`$, this expression reduces to
$`𝒲_\alpha ={\displaystyle \frac{1}{4\pi \mathrm{i}}}{\displaystyle }{\displaystyle \frac{\mathrm{d}w}{w}}\left({\displaystyle \frac{1}{4}}\overline{D}^2D_\alpha \right)V(w)|={\displaystyle \frac{1}{8}}\overline{D}^2D_\alpha V.`$ (7.16)
It can be seen that the gauge transformation (7.12) acts on the superfields in (7.14) as follows:
$$\delta V=\mathrm{i}(\overline{\mathrm{\Lambda }}\mathrm{\Lambda }),\delta \mathrm{\Phi }=\mathrm{i}_5\mathrm{\Lambda }\mathrm{\Lambda }=\mathrm{\Lambda }_1|.$$
(7.17)
The approach presented in this section can be applied to reformulate the supersymmetric Chern-Simons theory (4.5) in projective superspace, and the possibility for this is based on the following observation. Let $`^{++}`$ be a linear multiplet, that is a real analytic superfield obeying the constraint $`D^{++}^{++}=0`$. Then, the functional
$$d\zeta ^{(4)}𝒱^{++}^{++}$$
is invariant under the gauge transformations (7.1). We can further represent $`^{++}=(\mathrm{i}u^{+\underset{¯}{1}}u^{+\underset{¯}{2}})L(z,w)`$, where $`(z,w)`$ is a real O(2) multiplet. Then the functional
$$\frac{1}{2\pi \mathrm{i}}\mathrm{d}^5x\mathrm{d}^4\theta \frac{\mathrm{d}w}{w}V(w)L(w)|$$
is invariant under the gauge transformations (7.12).
In the case of Chern-Simons theory (4.5), the role of $`^{++}`$ is played by the gauge-invariant superfield $`(12g^2)^1𝒢^{++}`$, with $`𝒢^{++}`$ defined in (2.32). With the real O(2) multiplet $`G(z,w)`$ introduced by
$$𝒢^{++}=(\mathrm{i}u^{+\underset{¯}{1}}u^{+\underset{¯}{2}})G(z,w),G(w)=\frac{1}{w}\mathrm{\Psi }+K+w\overline{\mathrm{\Psi }},$$
(7.18)
the Chern-Simons theory (4.5) is equivalently described by the action
$$12g^2S_{\mathrm{CS}}=\frac{1}{2\pi \mathrm{i}}\mathrm{d}^5x\mathrm{d}^4\theta \frac{\mathrm{d}w}{w}V(w)G(w)|12g^2\mathrm{d}^5x𝓛_{\mathrm{CS}}.$$
(7.19)
Direct evaluation of $`\mathrm{\Psi }`$ and $`K`$ gives
$`\mathrm{\Psi }`$ $`=`$ $`𝒲^\alpha 𝒲_\alpha +{\displaystyle \frac{1}{2}}\overline{D}^2(^2),`$
$`K`$ $`=`$ $`D^\alpha 𝒲_\alpha 2(D^\alpha )𝒲_\alpha +\mathrm{c}.\mathrm{c}.+2_5(^2).`$ (7.20)
These results lead to
$`12g^2𝓛_{\mathrm{CS}}={\displaystyle \mathrm{d}^2\theta \mathrm{\Phi }𝒲^\alpha 𝒲_\alpha }`$ $`+`$ $`{\displaystyle \mathrm{d}^4\theta V\left[D^\alpha 𝒲_\alpha +2(D^\alpha )𝒲_\alpha \right]}+\mathrm{c}.\mathrm{c}.`$ (7.21)
$`+`$ $`4{\displaystyle \mathrm{d}^4\theta ^3}.`$
Here we have chosen to present the answer in the form $`(\mathrm{potential})\times (\mathrm{fieldstrength})\times (\mathrm{fieldstength})`$ analogously to the standard representation of the bosonic Chern-Simons action.<sup>15</sup><sup>15</sup>15The result presented here was given previously in the first reference of . In comparing the results one should keep in mind that terms such as $`\mathrm{d}^4\theta V\left[\frac{1}{2}(\mathrm{\Phi }+\overline{\mathrm{\Phi }})D^\alpha 𝒲_\alpha +D^\alpha \mathrm{\Phi }𝒲_\alpha \right]+\mathrm{c}.\mathrm{c}.`$ can be rewritten to look like $`\mathrm{d}^2\theta \mathrm{\Phi }𝒲^\alpha 𝒲_\alpha +\mathrm{c}.\mathrm{c}.`$, thereby changing the appearance of the action.
The structure of the superspace action obtained is the following. The first and second line of (7.21) are separately invariant under the gauge transformation (7.17) up to surface terms as is easily seen. The relative factor of 4 is fixed by five-dimensional Lorentz invariance. This could be derived either from the component projection or, less painfully, by checking the five-dimensional mass-shell condition on the super-fieldstrengths using their equations of motion together with their Bianchi identities. Finally, under the shift $`\mathrm{\Phi }\mathrm{\Phi }+1`$, the action shifts by $`S_{\mathrm{CS}}S_{\mathrm{CS}}+S_{\mathrm{YM}}+\mathrm{surface}\mathrm{term},`$ where $`S_{\mathrm{YM}}`$ is the 5D Yang-Mills action (5.11) with the proper normalization.
For completeness, we also present here projective superspace extensions of the vector multiplet mass term and the Fayet-Iliopoulos term (their harmonic superspace form is given in ). The vector multiplet mass term is
$$m^2\mathrm{d}\zeta ^{(4)}(𝒱^{++})^2\frac{m^2}{2\pi \mathrm{i}}\mathrm{d}^5x\mathrm{d}^4\theta \frac{\mathrm{d}w}{w}V^2(w)|.$$
(7.22)
The gauge invariant Fayet-Iliopoulos term is
$$\mathrm{d}\zeta ^{(4)}c^{++}𝒱^{++}\frac{1}{2\pi \mathrm{i}}\mathrm{d}^5x\mathrm{d}^4\theta \frac{\mathrm{d}w}{w}c(w)V(w)|,$$
(7.23)
where $`c^{++}=c^{ij}u_i^+u_j^+`$, with a constant real iso-vector $`c^{ij}`$. Defining $`c^{++}=\mathrm{i}u^{+\underset{¯}{1}}u^{+\underset{¯}{2}}c(w)`$, with $`c(w)=w^1\overline{\xi }_{}+\xi _{}w\xi _{}`$, the FI action then reduces to
$`\xi _{}{\displaystyle \mathrm{d}^5x\mathrm{d}^4\theta V}+2\mathrm{R}\mathrm{e}\left(\xi _{}{\displaystyle \mathrm{d}^5x\mathrm{d}^2\theta \mathrm{\Phi }}\right).`$ (7.24)
So far the considerations in this section have been restricted to the Abelian case. It is necessary to mention that the projective superspace approach can be generalized to provide an elegant description of 5D super Yang-Mills theories, which is very similar to the well-known description of 4D, $`𝒩=1`$ supersymmetric theories. In particular, the Yang-Mills supermultiplet is described by a real Lie-algebra-valued tropical superfield $`V(z,w)`$ with the gauge transformation
$$\mathrm{e}^{V(w)}\mathrm{e}^{\mathrm{i}\stackrel{˘}{\mathrm{\Lambda }}(w)}\mathrm{e}^{V(w)}\mathrm{e}^{\mathrm{i}\mathrm{\Lambda }(w)},$$
(7.25)
which is the non-linear generalization of the Abelian gauge transformation (7.12). The hypermultiplet sector is described by an arctic superfield $`\mathrm{{\rm Y}}(z,w)`$ and its conjugate, with the gauge transformation
$$\mathrm{{\rm Y}}(w)\mathrm{e}^{\mathrm{i}\mathrm{\Lambda }(w)}\mathrm{{\rm Y}}(w).$$
(7.26)
The hypermultiplet gauge–invariant action is
$$S[\mathrm{{\rm Y}},\stackrel{˘}{\mathrm{{\rm Y}}},V]=\frac{1}{2\pi \mathrm{i}}\frac{\mathrm{d}w}{w}\mathrm{d}^5x\mathrm{d}^4\theta \stackrel{˘}{\mathrm{{\rm Y}}}(w)\mathrm{e}^{V(w)}\mathrm{{\rm Y}}(w).$$
(7.27)
## 8 Conclusion
In the present paper we have developed the manifestly supersymmetric approach to five-dimensional globally supersymmetric gauge theories. It is quite satisfying that 5D superspace techniques provide a universal setting to formulate all such theories in a compact, transparent and elegant form, similarly to the four-dimensional $`𝒩=1`$ and $`𝒩=2`$ theories. We believe that these techniques are not only elegant but, more importantly, are useful. In particular, these techniques may be useful for model building in the context of supersymmetric brane-world scenarios. The two examples of supersymmetric nonlinear sigma-models, which were constructed in section 6, clearly demonstrate the power of the 5D superspace approach.
Five-dimensional super Yang-Mills theories possess interesting properties at the quantum level . Further insight into their quantum mechanical structure may be obtained by carrying out explicit supergraph calculations. Supersymmetric Chern-Simons theories (4.4) are truly interesting in this respect.
Note Added: After this paper was posted to the hep-th archive, we were informed of a related interesting work on 6D, $`𝒩=(1,0)`$ supersymmetric field theories .
Acknowledgements:
SMK is grateful to Jim Gates and the Center for String and Particle Theory at the University of Maryland, where this project was conceived, for hospitality. The work of SMK is supported in part by the Australian Research Council. The work of WDL is supported by the University of Maryland Center for String and Particle Theory.
## Appendix A 5D Notation and Conventions
Our 5D notation and conventions are very similar to those introduced in .
The 5D gamma-matrices $`\mathrm{\Gamma }_{\widehat{m}}=(\mathrm{\Gamma }_m,\mathrm{\Gamma }_5)`$, with $`m=0,1,2,3`$, defined by
$$\{\mathrm{\Gamma }_{\widehat{m}},\mathrm{\Gamma }_{\widehat{n}}\}=2\eta _{\widehat{m}\widehat{n}}\mathbf{\hspace{0.17em}1},(\mathrm{\Gamma }_{\widehat{m}})^{}=\mathrm{\Gamma }_0\mathrm{\Gamma }_{\widehat{m}}\mathrm{\Gamma }_0$$
(A.1)
are chosen in accordance with
$`(\mathrm{\Gamma }_m){}_{\widehat{\alpha }}{}^{}{}_{}{}^{\widehat{\beta }}=\left(\begin{array}{cc}0& (\sigma _m)_{\alpha \stackrel{\text{.}}{\beta }}\\ (\stackrel{~}{\sigma }_m)^{\stackrel{\text{.}}{\alpha }\beta }& 0\end{array}\right),(\mathrm{\Gamma }_5){}_{\widehat{\alpha }}{}^{}{}_{}{}^{\widehat{\beta }}=\left(\begin{array}{cc}\mathrm{i}\delta _\alpha ^\beta & 0\\ 0& \mathrm{i}\delta ^{\stackrel{\text{.}}{\alpha }}_{\stackrel{\text{.}}{\beta }}\end{array}\right),`$ (A.6)
such that $`\mathrm{\Gamma }_0\mathrm{\Gamma }_1\mathrm{\Gamma }_2\mathrm{\Gamma }_3\mathrm{\Gamma }_5=\mathrm{𝟏}`$. The charge conjugation matrix, $`C=(\epsilon ^{\widehat{\alpha }\widehat{\beta }})`$, and its inverse, $`C^1=C^{}=(\epsilon _{\widehat{\alpha }\widehat{\beta }})`$ are defined by
$`C\mathrm{\Gamma }_{\widehat{m}}C^1=(\mathrm{\Gamma }_{\widehat{m}}){}_{}{}^{\mathrm{T}},\epsilon ^{\widehat{\alpha }\widehat{\beta }}=\left(\begin{array}{cc}\epsilon ^{\alpha \beta }& 0\\ 0& \epsilon _{\stackrel{\text{.}}{\alpha }\stackrel{\text{.}}{\beta }}\end{array}\right),\epsilon _{\widehat{\alpha }\widehat{\beta }}=\left(\begin{array}{cc}\epsilon _{\alpha \beta }& 0\\ 0& \epsilon ^{\stackrel{\text{.}}{\alpha }\stackrel{\text{.}}{\beta }}\end{array}\right).`$ (A.11)
The antisymmetric matrices $`\epsilon ^{\widehat{\alpha }\widehat{\beta }}`$ and $`\epsilon _{\widehat{\alpha }\widehat{\beta }}`$ are used to raise and lower the four-component spinor indices.
A Dirac spinor, $`\mathrm{\Psi }=(\mathrm{\Psi }_{\widehat{\alpha }})`$, and its Dirac conjugate, $`\mathrm{\Psi }=(\overline{\mathrm{\Psi }}^{\widehat{\alpha }})=\mathrm{\Psi }^{}\mathrm{\Gamma }_0`$, look like
$`\mathrm{\Psi }_{\widehat{\alpha }}=\left(\begin{array}{c}\psi _\alpha \\ \overline{\varphi }^{\stackrel{\text{.}}{\alpha }}\end{array}\right),\overline{\mathrm{\Psi }}^{\widehat{\alpha }}=(\varphi ^\alpha ,\overline{\psi }_{\stackrel{\text{.}}{\alpha }}).`$ (A.14)
One can now combine $`\overline{\mathrm{\Psi }}^{\widehat{\alpha }}=(\varphi ^\alpha ,\overline{\psi }_{\stackrel{\text{.}}{\alpha }})`$ and $`\mathrm{\Psi }^{\widehat{\alpha }}=\epsilon ^{\widehat{\alpha }\widehat{\beta }}\mathrm{\Psi }_{\widehat{\beta }}=(\psi ^\alpha ,\overline{\varphi }_{\stackrel{\text{.}}{\alpha }})`$ into a SU(2) doublet,
$$\mathrm{\Psi }_i^{\widehat{\alpha }}=(\mathrm{\Psi }_i^\alpha ,\overline{\mathrm{\Psi }}_{\stackrel{\text{.}}{\alpha }i}),(\mathrm{\Psi }_i^\alpha )^{}=\overline{\mathrm{\Psi }}^{\stackrel{\text{.}}{\alpha }i},i=\underset{¯}{1},\underset{¯}{2},$$
(A.15)
with $`\mathrm{\Psi }_{\underset{¯}{1}}^\alpha =\varphi ^\alpha `$ and $`\mathrm{\Psi }_{\underset{¯}{2}}^\alpha =\psi ^\alpha `$. It is understood that the SU(2) indices are raised and lowered by $`\epsilon ^{ij}`$ and $`\epsilon _{ij}`$, $`\epsilon ^{\underset{¯}{1}\underset{¯}{2}}=\epsilon _{\underset{¯}{2}\underset{¯}{1}}=1`$, in the standard fashion: $`\mathrm{\Psi }^{\widehat{\alpha }i}=\epsilon ^{ij}\mathrm{\Psi }_j^{\widehat{\alpha }}`$. The Dirac spinor $`\mathrm{\Psi }^i=(\mathrm{\Psi }_{\widehat{\alpha }}^i)`$ satisfies the pseudo-Majorana condition $`\overline{\mathrm{\Psi }}_i{}_{}{}^{\mathrm{T}}=C\mathrm{\Psi }_i`$. This will be concisely represented as
$$(\mathrm{\Psi }_{\widehat{\alpha }}^i)^{}=\mathrm{\Psi }_i^{\widehat{\alpha }}.$$
(A.16)
With the definition $`\mathrm{\Sigma }_{\widehat{m}\widehat{n}}=\mathrm{\Sigma }_{\widehat{n}\widehat{m}}=\frac{1}{4}[\mathrm{\Gamma }_{\widehat{m}},\mathrm{\Gamma }_{\widehat{n}}]`$, the matrices $`\{\mathrm{𝟏},\mathrm{\Gamma }_{\widehat{m}},\mathrm{\Sigma }_{\widehat{m}\widehat{n}}\}`$ form a basis in the space of $`4\times 4`$ matrices. The matrices $`\epsilon _{\widehat{\alpha }\widehat{\beta }}`$ and $`(\mathrm{\Gamma }_{\widehat{m}})_{\widehat{\alpha }\widehat{\beta }}`$ are antisymmetric, $`\epsilon ^{\widehat{\alpha }\widehat{\beta }}(\mathrm{\Gamma }_{\widehat{m}})_{\widehat{\alpha }\widehat{\beta }}=0`$, while the matrices $`(\mathrm{\Sigma }_{\widehat{m}\widehat{n}})_{\widehat{\alpha }\widehat{\beta }}`$ are symmetric. Given a 5-vector $`V^{\widehat{m}}`$ and an antisymmetric tensor $`F^{\widehat{m}\widehat{n}}=F^{\widehat{n}\widehat{m}}`$, we can equivalently represent them as the bi-spinors $`V=V^{\widehat{m}}\mathrm{\Gamma }_{\widehat{m}}`$ and $`F=\frac{1}{2}F^{\widehat{m}\widehat{n}}\mathrm{\Sigma }_{\widehat{m}\widehat{n}}`$ with the following symmetry properties
$`V_{\widehat{\alpha }\widehat{\beta }}`$ $`=`$ $`V_{\widehat{\beta }\widehat{\alpha }},\epsilon ^{\widehat{\alpha }\widehat{\beta }}V_{\widehat{\alpha }\widehat{\beta }}=0,F_{\widehat{\alpha }\widehat{\beta }}=F_{\widehat{\beta }\widehat{\alpha }}.`$ (A.17)
The two equivalent descriptions $`V_{\widehat{m}}V_{\widehat{\alpha }\widehat{\beta }}`$ and and $`F_{\widehat{m}\widehat{n}}F_{\widehat{\alpha }\widehat{\beta }}`$ are explicitly described as follows:
$`V_{\widehat{\alpha }\widehat{\beta }}=V^{\widehat{m}}(\mathrm{\Gamma }_{\widehat{m}})_{\widehat{\alpha }\widehat{\beta }},`$ $`V_{\widehat{m}}={\displaystyle \frac{1}{4}}(\mathrm{\Gamma }_{\widehat{m}})^{\widehat{\alpha }\widehat{\beta }}V_{\widehat{\alpha }\widehat{\beta }},`$
$`F_{\widehat{\alpha }\widehat{\beta }}={\displaystyle \frac{1}{2}}F^{\widehat{m}\widehat{n}}(\mathrm{\Sigma }_{\widehat{m}\widehat{n}})_{\widehat{\alpha }\widehat{\beta }},`$ $`F_{\widehat{m}\widehat{n}}=(\mathrm{\Sigma }_{\widehat{m}\widehat{n}})^{\widehat{\alpha }\widehat{\beta }}F_{\widehat{\alpha }\widehat{\beta }}.`$ (A.18)
These results can be easily checked using the identities (see e.g. ):
$`\epsilon _{\widehat{\alpha }\widehat{\beta }\widehat{\gamma }\widehat{\delta }}`$ $`=`$ $`\epsilon _{\widehat{\alpha }\widehat{\beta }}\epsilon _{\widehat{\gamma }\widehat{\delta }}+\epsilon _{\widehat{\alpha }\widehat{\gamma }}\epsilon _{\widehat{\delta }\widehat{\beta }}+\epsilon _{\widehat{\alpha }\widehat{\delta }}\epsilon _{\widehat{\beta }\widehat{\gamma }},`$
$`\epsilon _{\widehat{\alpha }\widehat{\gamma }}\epsilon _{\widehat{\beta }\widehat{\delta }}\epsilon _{\widehat{\alpha }\widehat{\delta }}\epsilon _{\widehat{\beta }\widehat{\gamma }}`$ $`=`$ $`{\displaystyle \frac{1}{2}}(\mathrm{\Gamma }^{\widehat{m}})_{\widehat{\alpha }\widehat{\beta }}(\mathrm{\Gamma }_{\widehat{m}})_{\widehat{\gamma }\widehat{\delta }}+{\displaystyle \frac{1}{2}}\epsilon _{\widehat{\alpha }\widehat{\beta }}\epsilon _{\widehat{\gamma }\widehat{\delta }},`$ (A.19)
and therefore
$$\epsilon _{\widehat{\alpha }\widehat{\beta }\widehat{\gamma }\widehat{\delta }}=\frac{1}{2}(\mathrm{\Gamma }^{\widehat{m}})_{\widehat{\alpha }\widehat{\beta }}(\mathrm{\Gamma }_{\widehat{m}})_{\widehat{\gamma }\widehat{\delta }}+\frac{1}{2}\epsilon _{\widehat{\alpha }\widehat{\beta }}\epsilon _{\widehat{\gamma }\widehat{\delta }},$$
(A.20)
with $`\epsilon _{\widehat{\alpha }\widehat{\beta }\widehat{\gamma }\widehat{\delta }}`$ the completely antisymmetric fourth-rank tensor.
Complex conjugation gives
$$(\epsilon _{\widehat{\alpha }\widehat{\beta }})^{}=\epsilon ^{\widehat{\alpha }\widehat{\beta }},(V_{\widehat{\alpha }\widehat{\beta }})^{}=V^{\widehat{\alpha }\widehat{\beta }},(F_{\widehat{\alpha }\widehat{\beta }})^{}=F^{\widehat{\alpha }\widehat{\beta }},$$
(A.21)
provided $`V^{\widehat{m}}`$ and $`F^{\widehat{m}\widehat{n}}`$ are real.
The conventional 5D simple superspace $`^{5|8}`$ is parametrized by coordinates $`z^{\widehat{A}}=(x^{\widehat{a}},\theta _i^{\widehat{\alpha }})`$. Then, a hypersurface $`x^5=\mathrm{const}`$ in $`^{5|8}`$ can be identified with the 4D, $`𝒩=2`$ superspace $`^{4|8}`$ parametrized by
$$z^A=(x^a,\theta _i^\alpha ,\overline{\theta }_{\stackrel{\text{.}}{\alpha }}^i),(\theta _i^\alpha )^{}=\overline{\theta }^{\stackrel{\text{.}}{\alpha }i}.$$
(A.22)
The Grassmann coordinates of $`^{5|8}`$ and $`^{4|8}`$ are related to each other as follows:
$`\theta _i^{\widehat{\alpha }}=(\theta _i^\alpha ,\overline{\theta }_{\stackrel{\text{.}}{\alpha }i}),\theta _{\widehat{\alpha }}^i=\left(\begin{array}{c}\theta _\alpha ^i\\ \overline{\theta }^{\stackrel{\text{.}}{\alpha }i}\end{array}\right).`$ (A.25)
Interpreting $`x^5`$ as a central charge variable, one can view $`^{5|8}`$ as a 4D, $`𝒩=2`$ central charge superspace (see below).
The flat covariant derivatives $`D_{\widehat{A}}=(_{\widehat{a}},D_{\widehat{\alpha }}^i)`$ obey the algebra
$$\{D_{\widehat{\alpha }}^i,D_{\widehat{\beta }}^j\}=2\mathrm{i}\epsilon ^{ij}\left((\mathrm{\Gamma }^{\widehat{c}}){}_{\widehat{\alpha }\widehat{\beta }}{}^{}_{\widehat{c}}^{}+\epsilon _{\widehat{\alpha }\widehat{\beta }}\mathrm{\Delta }\right),[D_{\widehat{\alpha }}^i,_{\widehat{b}}]=[D_{\widehat{\alpha }}^i,\mathrm{\Delta }]=0,$$
(A.26)
or equivalently
$$[D_{\widehat{A}},D_{\widehat{B}}\}=T_{\widehat{A}\widehat{B}}{}_{}{}^{\widehat{C}}D_{\widehat{C}}^{}+C_{\widehat{A}\widehat{B}}\mathrm{\Delta },$$
(A.27)
with $`\mathrm{\Delta }`$ the central charge. The spinor covariant derivatives are
$$D_{\widehat{\alpha }}^i=\frac{}{\theta _i^{\widehat{\alpha }}}\mathrm{i}(\mathrm{\Gamma }^{\widehat{b}}){}_{\widehat{\alpha }\widehat{\beta }}{}^{}\theta _{}^{\widehat{\beta }i}_{\widehat{b}}\mathrm{i}\theta _{\widehat{\alpha }}^i\mathrm{\Delta }.$$
(A.28)
One can relate the operators
$`D^i(D_{\widehat{\alpha }}^i)=\left(\begin{array}{c}D_\alpha ^i\\ \overline{D}^{\stackrel{\text{.}}{\alpha }i}\end{array}\right),\overline{D}_i(D_i^{\widehat{\alpha }})=(D_i^\alpha ,\overline{D}_{\stackrel{\text{.}}{\alpha }i})`$ (A.31)
to the 4D, $`𝒩=2`$ covariant derivatives $`D_A=(_a,D_\alpha ^i,\overline{D}_i^{\stackrel{\text{.}}{\alpha }})`$ where
$`D_\alpha ^i`$ $`=`$ $`{\displaystyle \frac{}{\theta _i^\alpha }}+\mathrm{i}(\sigma ^b)_{\alpha \dot{\beta }}\overline{\theta }^{\stackrel{\text{.}}{\beta }i}_b\mathrm{i}\theta _\alpha ^i(\mathrm{\Delta }+\mathrm{i}_5),`$
$`\overline{D}_{\stackrel{\text{.}}{\alpha }i}`$ $`=`$ $`{\displaystyle \frac{}{\overline{\theta }^{\stackrel{\text{.}}{\alpha }i}}}\mathrm{i}\theta _i^\beta (\sigma ^b)_{\beta \stackrel{\text{.}}{\alpha }}_b\mathrm{i}\overline{\theta }_{\stackrel{\text{.}}{\alpha }i}(\mathrm{\Delta }\mathrm{i}_5).`$ (A.32)
These operators obey the anti-commutation relations
$`\{D_\alpha ^i,D_\beta ^j\}`$ $`=`$ $`2\mathrm{i}\epsilon ^{ij}\epsilon _{\alpha \beta }(\mathrm{\Delta }+\mathrm{i}_5),\{\overline{D}_{\stackrel{\text{.}}{\alpha }i},\overline{D}_{\stackrel{\text{.}}{\beta }j}\}=2\mathrm{i}\epsilon _{ij}\epsilon _{\stackrel{\text{.}}{\alpha }\stackrel{\text{.}}{\beta }}(\mathrm{\Delta }\mathrm{i}_5),`$
$`\{D_\alpha ^i,\overline{D}_{\stackrel{\text{.}}{\beta }j}\}`$ $`=`$ $`2\mathrm{i}\delta _j^i(\sigma ^c)_{\alpha \stackrel{\text{.}}{\beta }}_c,`$ (A.33)
which correspond to the 4D, $`𝒩=2`$ supersymmetry algebra with a complex central charge (see also ).
In terms of the operators (A.31), the operation of complex conjugation acts as follows
$$(D^iF)^{}\mathrm{\Gamma }_0=(1)^{ϵ(F)}\overline{D}_iF^{},$$
(A.34)
with $`F`$ an arbitrary superfield and $`ϵ(F)`$ its Grassmann parity. This can be concisely represented as
$$(D_{\widehat{\alpha }}^iF)^{}=(1)^{ϵ(F)}D_i^{\widehat{\alpha }}F^{}.$$
(A.35)
## Appendix B Tensor Fields on the Two-Sphere
In this appendix we recall, following , the well-known one-to-one correspondence between smooth tensor fields on $`S^2=\mathrm{SU}(2)/\mathrm{U}(1)`$ and smooth scalar functions over SU(2) with definite U(1) charges. The two-sphere is obtained from SU(2) by factorization with respect to the equivalence relation
$$u^{+i}\mathrm{e}^{\mathrm{i}\phi }u^{+i}\phi .$$
(B.1)
We start by introducing two open charts forming an atlas on SU(2) which, upon identificationon (B.1), leads to a useful atlas on $`S^2`$. The north patch is defined by
$$u^{+\underset{¯}{1}}0,$$
(B.2)
and here we can represent
$`u^{+i}=u^{+\underset{¯}{1}}w^i,`$ $`w^i=(1,u^{+\underset{¯}{2}}/u^{+\underset{¯}{1}})=(1,w),`$
$`u_i^{}=\overline{u^{+\underset{¯}{1}}}\overline{w}_i,`$ $`\overline{w}_i=(1,\overline{w}),|u^{+\underset{¯}{1}}|^2=(1+w\overline{w})^1.`$ (B.3)
The south patch is defined by
$$u^{+\underset{¯}{2}}0,$$
(B.4)
and here we have
$`u^{+i}=u^{+\underset{¯}{2}}y^i,`$ $`y^i=(u^{+\underset{¯}{1}}/u^{+\underset{¯}{2}},1)=(y,1),`$
$`u_i^{}=\overline{u^{+\underset{¯}{2}}}\overline{y}_i,`$ $`\overline{y}_i=(\overline{y},1),|u^{+\underset{¯}{2}}|^2=(1+y\overline{y})^1.`$ (B.5)
In the overlap of the two charts we have
$$u^{+i}=\frac{\mathrm{e}^{\mathrm{i}\alpha }}{\sqrt{(1+w\overline{w})}}w^i=\frac{\mathrm{e}^{\mathrm{i}\beta }}{\sqrt{(1+y\overline{y})}}y^i,$$
(B.6)
where
$$y=\frac{1}{w},\mathrm{e}^{\mathrm{i}\beta }=\sqrt{\frac{w}{\overline{w}}}\mathrm{e}^{\mathrm{i}\alpha }.$$
(B.7)
The variables $`w`$ and $`y`$ are seen to be local complex coordinates on $`S^2`$ considered as the Riemann sphere, $`S^2=\{\mathrm{}\}`$; the north chart $`U_\mathrm{N}=`$ is parametrized by $`w`$ and the south patch $`U_\mathrm{S}={}_{}{}^{}\{\mathrm{}\}`$ is parametrized by $`y`$.
Along with $`w^i`$ and $`\overline{w}_i`$, we often use their counterparts with lower (upper) indices
$$w_i=\epsilon _{ij}w^j=(w,1),\overline{w}^i=\epsilon ^{ij}\overline{w}_j=(\overline{w},1),\overline{w_i}=\overline{w}^i,$$
(B.8)
and similar for $`y_i`$ and $`\overline{y}^i`$.
Let $`\mathrm{\Psi }^{(p)}(u)`$ be a smooth function on SU(2) with U(1)-charge $`p`$ chosen, for definiteness, to be non-negative, $`p0`$. Such a function possesses a convergent Fourier series of the form
$$\mathrm{\Psi }^{(p)}(u)=\underset{n=0}{\overset{\mathrm{}}{}}\mathrm{\Psi }^{(i_1\mathrm{}i_{n+p}j_1\mathrm{}j_n)}u_{i_1}^+\mathrm{}u_{i_{n+p}}^+u_{j_1}^{}\mathrm{}u_{j_n}^{},p0.$$
(B.9)
In the north patch we can write
$`\mathrm{\Psi }^{(p)}(u)`$ $`=`$ $`(u^{+\underset{¯}{1}})^p\mathrm{\Psi }_\mathrm{N}^{(p)}(w,\overline{w}),`$
$`\mathrm{\Psi }_\mathrm{N}^{(p)}(w,\overline{w})`$ $`=`$ $`{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}\mathrm{\Psi }^{(i_1\mathrm{}i_{n+p}j_1\mathrm{}j_n)}{\displaystyle \frac{w_{i_1}\mathrm{}w_{i_{n+p}}\overline{w}_{j_1}\mathrm{}\overline{w}_{j_n}}{(1+w\overline{w})^n}}.`$ (B.10)
In the south patch we have
$`\mathrm{\Psi }^{(p)}(u)`$ $`=`$ $`(u^{+\underset{¯}{2}})^p\mathrm{\Psi }_\mathrm{S}^{(p)}(y,\overline{y}),`$
$`\mathrm{\Psi }_\mathrm{S}^{(p)}(y,\overline{y})`$ $`=`$ $`{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}\mathrm{\Psi }^{(i_1\mathrm{}i_{n+p}j_1\mathrm{}j_n)}{\displaystyle \frac{y_{i_1}\mathrm{}y_{i_{n+p}}\overline{y}_{j_1}\mathrm{}\overline{y}_{j_n}}{(1+y\overline{y})^n}}.`$ (B.11)
Finally, in the overlap of the two charts $`\mathrm{\Psi }_\mathrm{N}^{(p)}`$ and $`\mathrm{\Psi }_\mathrm{S}^{(p)}`$ are simply related to each other
$$\mathrm{\Psi }_\mathrm{S}^{(p)}(y,\overline{y})=\frac{1}{w^p}\mathrm{\Psi }_\mathrm{N}^{(p)}(w,\overline{w}),$$
(B.12)
If we redefine
$$\widehat{\mathrm{\Psi }}_\mathrm{N}^{(p)}(w,\overline{w})=\mathrm{e}^{\mathrm{i}p\pi /4}\mathrm{\Psi }_\mathrm{N}^{(p)}(w,\overline{w}),\stackrel{ˇ}{\mathrm{\Psi }}_\mathrm{S}^{(p)}(y,\overline{y})=\mathrm{e}^{\mathrm{i}p\pi /4}\mathrm{\Psi }_\mathrm{S}^{(p)}(y,\overline{y}),$$
the above relation takes the form
$$\stackrel{ˇ}{\mathrm{\Psi }}_\mathrm{S}^{(p)}(y,\overline{y})=\left(\frac{y}{w}\right)^{p/2}\widehat{\mathrm{\Psi }}_\mathrm{N}^{(p)}(w,\overline{w})$$
(B.13)
and thus defines a smooth tensor field on $`S^2`$.
## Appendix C Projective Superspace Action
In this appendix we briefly demonstrate, following , how to derive the projective superspace action (6.16) from the harmonic superspace action (3.2). More details can be found in .
Consider an arbitrary projective superfield $`\varphi (z,w)`$, eq. (6.10), which is allowed to be singular only at $`w=0`$ and $`w=\mathrm{}`$ (i.e. $`\varphi (z,w)`$ is holomorphic on the doubly punctured sphere $`S^2\backslash \{NS\}`$). It is possible to promote $`\varphi (z,w)`$ to a smooth analytic superfield over $`S^2`$ by smearing (regularizing) its singularities with functions used in the construction of the partition of unity in differential geometry.
Consider a smooth cut-off function $`F_{R,ϵ}(x)`$ sketched in figure 3.
This function extrapolates smoothly from unit magnitude to zero in a small region between $`R`$, with is assumed to be large number, and $`R+ϵ`$ where $`ϵ`$ is small. The derivative of this function localizes whatever it multiplies to this region and is normalized so that in the limit
$`\underset{ϵ0}{lim}F_{R,ϵ}^{}(x)=\delta (xR)`$ (C.1)
as a distribution. Now, we can regularize the projective superfield $`\varphi (z,w)`$ as follows:
$$\varphi (z,w)\varphi _{R,ϵ}(z,w,\overline{w})=F_{R,ϵ}(|w|^1)\varphi (z,w)F_{R,ϵ}(|w|),$$
(C.2)
and the result is a a smooth neutral analytic superfield over the harmonic superspsace. If $`\varphi (z,w)`$ is regular at $`w=0`$ or $`w=\mathrm{}`$, then the factor $`F_{R,ϵ}(|w|^1)`$ or $`F_{R,ϵ}(|w|)`$ on the right of (C.2) can be removed.
The above procedure can also be used to generate charged analytic superfields. For instance, if $`\mathrm{\Lambda }(z,w)`$ is a real projective superfield, $`\stackrel{˘}{\mathrm{\Lambda }}=\mathrm{\Lambda }`$, then the following superfields
$`L_{R,ϵ}^{++}(z,u)`$ $`=`$ $`\mathrm{i}u^{+\underset{¯}{1}}u^{+\underset{¯}{2}}F_{R,ϵ}(|w|^1)L(z,w)F_{R,ϵ}(|w|)\mathrm{i}u^{+\underset{¯}{1}}u^{+\underset{¯}{2}}L_{R,ϵ}(z,w,\overline{w}),`$ (C.3)
$`L_{R,ϵ}^{(+4)}(z,u)`$ $`=`$ $`(u^{+\underset{¯}{1}}u^{+\underset{¯}{2}})^2F_{R,ϵ}(|w|^1)L(z,w)F_{R,ϵ}(|w|)(u^{+\underset{¯}{1}}u^{+\underset{¯}{2}})^2L_{R,ϵ}(z,w,\overline{w})`$ (C.4)
are real analytic superfields of charge $`+2`$ and $`+4`$, respectively. One can use $`L_{R,ϵ}^{(+4)}(z,u)`$ in the role of Lagrangian in (3.2). In the final stages we will remove the regulator by taking first $`ϵ0`$ and then $`R\mathrm{}`$.
As is seen from (3.2) and (3.3), the analytic action involves a square of $`(\widehat{D}^{})^2`$, and therefore we sould express the operators $`(\widehat{D}^{})^2`$ in local coordinates. What actually we need here is this operator acting on analytic or projective superfields $`\mathrm{\Phi }`$ such that $`_\alpha (w)\mathrm{\Phi }=\overline{}^{\stackrel{\text{.}}{\alpha }}(w)\mathrm{\Phi }=0`$, with operators $`_\alpha `$ and $`\overline{}^{\stackrel{\text{.}}{\alpha }}(w)`$ defined in (6.14). The analyticity allows us to move all $`𝒟_\alpha ^{\underset{¯}{2}}`$ and $`\overline{𝒟}_{\underset{¯}{2}}^{\stackrel{\text{.}}{\alpha }}`$ derivatives onto $`\mathrm{\Phi }`$ and rewrite them in terms of $`D_\alpha ^{\underset{¯}{1}}`$ and $`\overline{D}_{\underset{¯}{1}}^{\stackrel{\text{.}}{\alpha }}`$. When this is done, we find in local coordinates for an analytic $`\mathrm{\Phi }`$
$`(\widehat{D}^{})^2\mathrm{\Phi }`$ $`=`$ $`4(\overline{u^{+\underset{¯}{1}}})^2{\displaystyle \frac{(1+\overline{w}w)^2}{w}}𝒫(w)\mathrm{\Phi },`$ (C.5)
where we have defined the projective differential operator
$`𝒫(w)={\displaystyle \frac{1}{4w}}(\overline{D}_{\underset{¯}{1}})^2+_5{\displaystyle \frac{w}{4}}(D^{\underset{¯}{1}})^2.`$ (C.6)
It is worth pointing out that eq. (C.5) also holds in the presence of a non-vanishing central charge $`\mathrm{\Delta }`$. Using the analyticity of $`\mathrm{\Phi }`$ again, it is easy to show that
$`(\widehat{D}^{})^4\mathrm{\Phi }`$ $`=`$ $`(\overline{u^{+\underset{¯}{1}}})^4{\displaystyle \frac{(1+\overline{w}w)^4}{w^2}}D^4\mathrm{\Phi }+\text{total derivatives},`$ (C.7)
with the $`D^4`$ operator defined by (6.18). The latter operator determines the projective superspace measure, see eq. (6.17). Finally making use of the identity
$$\mathrm{d}u=\frac{\mathrm{d}^2w}{\pi (1+w\overline{w})^2},$$
(C.8)
one obtains (note $`|u^{+\underset{¯}{1}}|^2=(1+w\overline{w})^1`$)
$$\mathrm{d}\zeta ^{(4)}L_{R,ϵ}^{(+4)}(z,u)=\frac{1}{\pi }\mathrm{d}^5x\frac{\mathrm{d}^2w}{(1+w\overline{w})^2}D^4L_{R,ϵ}(z,w,\overline{w})\left|\right|.$$
(C.9)
Representing here
$$\frac{1}{(1+w\overline{w})^2}=\frac{1}{w}_{\overline{w}}\frac{1}{(1+w\overline{w})}$$
and integrating by parts, one can then show
$$\underset{R\mathrm{}}{lim}\underset{ϵ0}{lim}\mathrm{d}\zeta ^{(4)}L_{R,ϵ}^{(+4)}(z,u)=\frac{1}{2\pi \mathrm{i}}\mathrm{d}^5x\frac{\mathrm{d}w}{w}D^4L(z,w)\left|\right|.$$
(C.10)
This is exactly the projective action.
The formalism developed in this appendix can be applied to obtain a nice representation for the supersymmetric action (3.12) which is equivalent to
$`S`$ $`=`$ $`{\displaystyle \frac{\mathrm{i}}{4}}{\displaystyle }\mathrm{d}^5x{\displaystyle }\mathrm{d}u(\widehat{D}^{})^2L^{++}\left|\right|,D_{\widehat{\alpha }}^+L^{++}=0,D^{++}L^{++}=0.`$ (C.11)
Representing $`L^{++}=\mathrm{i}u^{+\underset{¯}{1}}u^{+\underset{¯}{2}}L(z,w)`$ and using eq. (C.5), we obtain
$`S`$ $`=`$ $`{\displaystyle }\mathrm{d}^5x{\displaystyle }\mathrm{d}u𝒫(w)L(z,w)\left|\right|.`$ (C.12)
Finally, making use of (7.7) gives
$`S`$ $`=`$ $`{\displaystyle \frac{1}{8\pi \mathrm{i}}}{\displaystyle }\mathrm{d}^5x{\displaystyle }{\displaystyle \frac{\mathrm{d}w}{w}}[{\displaystyle \frac{1}{w}}\overline{D}^2wD^2]L(z,w)\left|\right|.`$ (C.13)
As an example of the usefulness of such a form, we can consider the super Yang-Mills Lagrangian (3.18). A trivial contour integration in (C.13) then immediately reproduces the action for this theory in reduced superspace (5.11). |
warning/0507/physics0507059.html | ar5iv | text | # A modelling approach towards Epidermal homoeostasis control
## I Introduction
### I.1 Methodology
In many applications of mathematical modelling, continuum equations are used to describe the evolution of discrete biological systems such as tumours, epithelia, animal populations murray2002a etc. Such equations can be solved comparably efficiently and have helped to understand many qualitative and quantitative features of tumour growth byrne2003b ; schaller2006a . In this modelling approach, the discrete cell numbers are approximated by continuous functions. However, the inclusion of discrete effects even in simple models may lead to qualitatively different results bettelheim2001 ; louzoun2003 . This problem becomes even more important when modelling the initial evolution of cancer, which is thought to originate within a single cell gatenby2003a . For systems containing a few cells of a certain species only, the mathematical foundations of approximating the species dynamics by a continuous distribution function are rather shaky, and especially for the complicated interactions between several cell species including birth and death processes, one may expect a qualitatively different behaviour to arise from agent-based approaches.
Within the class of agent-based (individual-based) models, cellular tissues are modelled as a set of strongly-interacting discrete objects. At present, it seems reasonable to consider the cell as the smallest entity in these models, though this is not stringent (compare e. g. the extended Potts model graner1992 ; turner2002 ). For agent-based models, the concept of the cellular automaton neumann1966 ; gardner1970 has proven very useful, since it allows to describe cells by simple interaction rules. Tumour growth for example has widely been modelled with cellular automata smolle1998a ; dormann2002 ; ferreira2002 . Most of the current agent-based models are implemented on a lattice. In models where a single lattice site is occupied by a single cell, volume-nonconserving events such as proliferation require far-reaching configuration changes on the lattice, which in such models comes along with rupture of ”intercellular bindings” on a large scale meineke2001 . This leads to the necessity of deriving effective interaction rules, which can often not be easily related to physical laws. Consequently, the avoided lattice artifacts often come at the price of ending up with parameters that principally cannot be determined by independent experiments and predictive power of the model is lost. A physical motivation for cell-cell interactions can be included by allowing for shape fluctuations. In lattice models, this can be achieved by increasing the resolution of the lattice (i. e., by representing a single cell by a variable number of lattice nodes as e. g. in the Potts model turner2002 or the hyphasma model meyer\_hermann2004 ). Likewise, one can also in off-lattice models introduce further degrees of freedom per cell (such as e. g. the dynamics of the cell boundary as in weliky1990 ; weliky1991 ). However, especially in the realistic case with three spatial dimensions, such models with an increased resolution use a dramatically increased number of dynamic variables and thereby the necessary calculation time increases (for example, in cickovski2005a about 13000 cells are considered with about 120 lattice nodes per cell in a three-dimensional setup). As an advantage however, physically-motivated interactions can be used to calculate the cellular dynamics. Off-lattice centre-based models drasdo1995 ; dallon2004 allow for physical interactions to be included without the disadvantage of the large number of dynamic variables. As a drawback, they assume an intrinsic cellular equilibrium shape (e. g. spherical or ellipsoidal) and treat all deviations from this form as small perturbations. Thus, they represent a compromise between speed of numerical calculation and model accuracy.
We have previously applied an off-lattice centre-based model to the qualitatively simple problem of multicellular tumour spheroid growth schaller2005a . In comparison to an analogous continuum approach schaller2006a it turned out that agent-based models have greater potential to describe specific biological systems realistically. Keeping in mind that continuum models also arise from averaging (thereby deleting information), this is not a big surprise. However, it must be said that the degree of model complexity should always be limited by the experimental signature and for many experimental signatures continuum models will suffice.
In this article, we set up an off-lattice centre-based model for the epidermis that is intrinsically consistent and is also based on physical interactions as far as possible. This has the advantage that the model can be falsified.
### I.2 The Epidermis
The epidermis is a stratified squamous epithelial tissue. It does not contain separate blood vessels and is therefore dependent on diffusion of nutrients from the dermis situated below. The epidermis can be divided into several layers montagna1992 :
The innermost stratum germinativum or stratum basale (basal layer) is a monolayer, in which most cell divisions occur. It is separated from the dermis below by a basal membrane. A fraction of the cells created there by cell division travels upwards into the stratum spinosum. Within this layer, the process of cornification begins: The cytoplasm looses water and is filled with keratin filaments. Within the stratum granulosum, cells die off and their shape flattens. This special pathway of cell death is also called anoikis. Completely cornified cells mark the stratum corneum, which is clearly distinguishable from the layers below. This layer does not contain viable cells and constitutes an efficient barrier for water and many of its solutes. The thickness of this layer varies strongly for different regions of the skin montagna1992 . In the upper part of this layer, the cellular material detaches by dissolving intercellular contacts.
Within this article, only three layers will be distinguished: The term stratum medium will be used as a combination for all layers not belonging to the stratum germinativum or the stratum corneum.
The cell types encountered in the epidermis are keratinocytes, melanocytes, Langerhans cells, and Merkel cells. Of these, the dominant fraction is constituted by the keratinocytes with roughly 75000 cells per square mm hoath2003 ; bauer2001 .
Keratinocytes are produced in the stratum germinativum by cell division. In order to maintain epidermal homoeostasis, in average one of the two daughter cells must leave the basal layer and travel upwards. The keratinocyte remaining in the basal layer will be termed stem cell in this article. The cell travelling upwards transforms into a fully-differentiated keratinocyte – possibly undergoing transit amplifying proliferations barthel2000 – and reaches the surface after about 12 to 14 days. During this passage, the keratinocytes follow the process of cornification described above.
Melanocytes are dendritic cells that are distributed within the basal layer, and their density is approximately 2000 cells per square mm montagna1992 ; moncrieff2001 ; hoath2003 . They adhere to the basal membrane via hemi-desmosomes. Their normal function is to provide keratinocytes in the skin with melanin. Tumours arising from cancerous melanocytes are called melanoma. Since most cancerous melanocytes still produce melanin, such tumours have a characteristic black colour.
The Langerhans cells are dendritic cells of the immune system and it is believed that Merkel cells play a role in sensation. Since neither effects of the immune system nor the mechanisms of sensation will be studied here, the latter two cell types will not be contained in the in silico representation and will not be discussed further.
The diffusional properties of the skin have important implications on medication applied to this tissue. With an observed strong increase of the manifestation of melanoma moncrieff2001 , studies of melanoma development are of huge importance. Especially their early diagnosis bears the potential to improve the chances of recovery. Within this article, some basic questions related to the initial stages of melanoma growth will be addressed.
## II The model
In view of the complicated matter in reality, any model will inevitably simplify the system by neglecting properties that we consider to be less important. Starting from this perspective, the following approximation may seem reasonable:
Within the model, cells are represented as adhesively and elastically interacting (compressible) spheres of time-dependent radii. Processes such as cell proliferation and cell death correspond to insertion or removal of spheres to the system, respectively. The cell growth determines the time-dependent cell radii. Accordingly, cells consume nutrients and their presence also changes the diffusive properties within the tissue. In analogy to the cell cycle, the model cells can assume different internal states.
Note that the simplifying assumption of spherical cell shapes facilitates the use of adjacency detection algorithms such as the Delaunay triangulation schaller2004 . The basic specifics and consequences of this paradigm will be examined in the following, whereas the detailed technical discussion can be found in the appendix.
### II.1 Continuous model variables
The problem of describing the dynamics of cell-cell contacts is currently not solved. In this article we have chosen the Johnson-Kendall-Roberts (JKR) model johnson1958 ; johnson1971 – supplemented with viscous shear and normal forces and delta-correlated random forces. The JKR model includes elastic and adhesive normal forces and has been experimentally verified for soft materials such as rubber. The simultaneous treatment of elastic and adhesive properties goes beyond some previous modelling approaches kreft2004a ; picioreanu2004a ; schaller2005a . Cell movement within solution (and even more in tissue) can be regarded as highly overdamped, which is why viscous effects cannot be neglected (in fact, they are known to be dominant). A detailed discussion of these interactions can be found in appendix A for the JKR-model, appendix B for the random forces, and appendix C for the arising equations of motion including dampening force contributions.
Under the assumption of dominant friction and neglecting the contribution of torque to dampening, the equations of motion for $`N`$ cells can be written as a large $`3N\times 3N`$ matrix $`A(t)`$ acting on the $`3N`$-dimensional cell position vector $`𝒙=[x_1(t),y_1(t),z_1(t),\mathrm{},x_N(t),y_N(t),z_N(t)]`$
$`A(t)\dot{𝒙}=𝒃,`$ (1)
where the vector $`𝒃(t)`$ on the right hand side includes all non-viscous forces and the matrix $`A(t)`$ contains the dampening contributions, see also the example in the appendix D. Since there is no analytic form of these quantities, equation (1) has to be solved numerically.
As the friction matrix $`A(t)`$ has neither a diagonal (non-isotropic friction) nor another simple structure, its inverse cannot be easily calculated. The fact that only cells in contact can contribute to friction however leads to a sparse population of the dampening matrix. In addition, since more than one spatial dimension is considered, the dampening matrix is not even block diagonal. Therefore, the inverse of such a sparse matrix is not necessarily sparse as well and for the systems considered in this paper, the inverse matrix of $`A(t)`$ would even not fit into main memory of normal PC’s. Therefore, we have used an iterative procedure – the method of conjugate gradients, see appendix D – that does not necessitate the explicit calculation of $`A^1(t)`$ to solve above equation for the position vector $`𝒙(t)`$.
In addition, one has to solve the problem of the dynamics of diffusible signals such as e. g. nutrients. If processes such as convection and flow transport are comparably small, these are well described by reaction-diffusion equations. Such types of equations arise naturally from the continuum equation, if one assumes that the flow is always proportional to the concentration gradient and tends to even out all gradients. The discretisation of reaction-diffusion equations on a regular lattice leads to a large linear systems as above. However, in this case the matrix possesses more symmetry properties and well-adapted algorithms exist for the solution, see appendix D. Note that an alternative approach circumventing a grid discretization would be to use the method of Green functions following newman2004 .
Within the JKR model, adhesion is described by an adhesion energy density parameter $`\sigma `$. Assuming spatially uniform receptor ($`C_i^{\mathrm{rec}}`$) and ligand ($`C_i^{\mathrm{lig}}`$) densities on the cell membranes of cells $`i`$ and $`j`$, this parameter can be expressed as $`\sigma _{ij}=\frac{\sigma ^{\mathrm{max}}}{2}\left[C_i^{\mathrm{rec}}C_j^{\mathrm{lig}}+C_i^{\mathrm{lig}}C_j^{\mathrm{rec}}\right]`$, where $`\sigma ^{\mathrm{max}}`$ determines the maximum adhesion energy density (model parameter). A measure for the total cellular binding strength can then be derived from the sum of all binding energies with the next neighbours
$`\mathrm{\Sigma }_i(t)={\displaystyle \underset{j𝒩𝒩(i)}{}}\sigma _{ij}(t)A_{ij}(t).`$ (2)
Assuming that cells with low binding are shed off the skin surface, we remove necrotic and cornified cells from the simulation as soon as their binding strength falls below a critical value $`\mathrm{\Sigma }_i\mathrm{\Sigma }^{\mathrm{min}}`$ (model parameter). Note that this choice also has the consequence that non-viable cells without contact to other cells are also removed from the simulation. These cells do not have anchorage and would be shed off in the realistic epidermis. In the case of cornified and necrotic cells we have assumed an exponential decay of receptor and ligand molecules with a given rate $`\alpha `$, i. e.,
$`\dot{C}_i^{\mathrm{rec}/\mathrm{lig}}=\alpha C_i^{\mathrm{rec}/\mathrm{lig}}.`$ (3)
In the JKR model \[compare equation (8)\], the resulting decreased cell-cell adhesion would – with unchanged elastic parameters – lead to a perturbation of the equilibrium distance. Therefore, we have chosen to adapt the elastic cell modulus simultaneously. To maintain for similar cells a constant equilibrium distance, this implies a decreasing elastic modulus according to
$`\dot{E}_i=2\alpha E_i.`$ (4)
Although the loss of receptors and ligands as well as decreasing cell elasticity may be reasonable assumptions for cornified and necrotic cells, the overall time course may be quite different in reality. We have tried other forms of necrotic cell removal. For example, one could think of removing non-viable cells randomly at a constant rate as was done in schaller2005a . This possibility however did significantly disturb the layered structure of the stratum corneum. Holes in this protective layer in turn did lead to sudden loss of the proliferation-moderating soluble substance in the epidermis and thereby to irregular proliferative behaviour and considerable oscillations in epidermal thickness. The same problem occurred when assuming a (Gaussian-distributed) cell-specific time after which non-viable cells were removed from the simulation.
Thus, the assumptions regarding continuous model variables can be summarised as follows:
* cell shape : spherical with slight perturbations
* cell-cell interactions : elastic, adhesive, and viscous non-isotropic dampening
* approximations : dominating friction and neglect of the back-reaction of angular velocities on the kinetics
* necrotic and cornified cells : loss of receptor and ligands leading to a reduced intercellular adhesion
### II.2 Discrete Model Variables
Without representations of internal cellular states, the model would merely calculate the mechanical interaction between a number of adhesively and elastically interacting deformable spheres. However, it is well known that the different states in the cell cycle yield a different cell behaviour. This should also be reflected in the model. Extending a previous agent-based modelling approach schaller2005a to the necessities of the epidermis, we distinguish between the following internal states in the model: M-phase, $`\mathrm{G}_1`$-phase, $`\mathrm{S}/\mathrm{G}_2`$-phase, $`\mathrm{G}_0`$-phase, necrotic, cornified. The first four states are illustrated in figure 1 left panel.
During $`\mathrm{G}_1`$-phase, the cell volume grows at a constant rate $`r_\mathrm{V}`$, i. e., the radius increases according to $`\dot{R}=\left(4\pi R^2\right)^1r_\mathrm{V}`$, until the cell reaches its final mitotic radius $`R^{(\mathrm{m})}`$. The volume growth rate $`r_\mathrm{V}`$ is deduced from the minimum observed cycle time $`\tau ^{\mathrm{min}}`$ and the durations of the $`\mathrm{S}/\mathrm{G}_2`$-phase and the M-phase. Afterwards, no further cell growth is performed.
At the end of the $`\mathrm{G}_1`$-phase, a checkpointing mechanism is performed. At this checkpoint, the cell can either enter the $`\mathrm{G}_0`$-phase or the $`\mathrm{S}/\mathrm{G}_2`$-phase: In potten2000 it has been suspected that a diffusible substance produced at the basal layer might moderate the cellular proliferation. It is well known that the stratum corneum constitutes an effective barrier against the loss of water and its solutes as well as other substances barthel2000 ; bashir2001 ; kasting2003 . Its removal leads to a proliferative response. Hence, we use this correlation to establish a causal connection as a hypothesis. As the simplest model assumption, we will here assume the distribution of mobile water to influence cell proliferation. By mobile water we mean the fraction of the water content of the tissue that is not bound in intra- or extracellular cavities. Note that evidently the concentration of this water fraction can change in time and space. If the local mobile water concentration is below a critical value (model parameter), the cell will directly continue the cell cycle, whereas in the other case the cell cycle will be prolonged by the cell switching into the $`\mathrm{G}_0`$-phase. Cells leave the $`\mathrm{G}_0`$-phase to enter the $`\mathrm{S}/\mathrm{G}_2`$-phase if either the local water concentration falls below the threshold or after an individual maximum time has passed (drawn from a random number generator, see subsection II.3). Note that the assumption of a different moderating diffusible signal would not significantly change the model as long as it is not created or consumed by the cells in the epidermis themselves.
Within this article the S-phase and $`\mathrm{G}_2`$-phase are not distinguished, their inclusion would be a mere technical aspect. After leaving the $`\mathrm{S}/\mathrm{G}_2`$-phase (see subsection II.3), the cells deterministically enter mitosis. Keratinocytes and healthy melanocytes underly an exception at this point: After the fourth cell generation, keratinocytes cornify (enter anoikis) meineke2001 ; potten2000 ; barthel2000 . Healthy melanocytes simply remain at the end of $`\mathrm{S}/\mathrm{G}_2`$-phase.
Within the model, the difference between the $`\mathrm{S}/\mathrm{G}_2`$-phase and the $`\mathrm{G}_0`$-phase is that the duration of the first is determined by an individual duration that can be derived from experiments, whereas the duration of the latter is also controlled by the spatio-temporal evolution of the concentration of the moderating substance (in this case, mobile water).
One should be aware that our classification of internal cellular states may not directly correspond to the realistic biological system. However, the only net effect of the existence of the $`\mathrm{G}_0`$-phase is the prolongation of the cell cycle time: Cells in $`\mathrm{G}_0`$-phase can serve as a reservoir of cells ready to start proliferating as soon as the local water concentration falls below a critical threshold. A different terminology or a placement of the $`\mathrm{G}_0`$-phase after or within the $`\mathrm{S}/\mathrm{G}_2`$-phase would therefore not significantly change the model.
At the beginning of the mitotic phase – which lasts for about half an hour for most cell types – a mother cell is replaced by two daughter cells. The radii of the daughter cells are decreased $`R^{(\mathrm{d})}=R^{(\mathrm{m})}2^{1/3}`$ to ensure conservation of the target volume during M-phase. In addition, they are placed at distance $`d_{ij}^0=2R^{(\mathrm{m})}(12^{1/3})`$ to ensure that in this first discontinuous step the daughter cells do not leave the region occupied by the mother cell, see figure 1 right panels. In most cases, the daughter cells have the same cell type as the mother cell. The only exception is given by the keratinocyte stem cells which divide asymmetrically: By model assumption the upper daughter cell differentiates to a keratinocyte. The new cells are subject to their initially dominating repulsive forces (8). Note that an adaptive timestep derived from a maximum spatial stepsize ensures that the mitotic partners do not loose contact. Afterwards, the daughter cells enter the $`\mathrm{G}_1`$-phase thus closing the cell cycle.
Viable cells can enter necrosis at any time in the cell cycle as soon as the nutrient concentration at the cellular position falls below a cell-type specific critical threshold (model parameter). As the dominant pathway to cell death, keratinocytes in contrast enter anoikis after completing $`\mathrm{S}/\mathrm{G}_2`$-phase in the fourth generation. Naturally, necrotic or cornified cells do not consume any nutrients.
The corresponding assumptions on discrete model variables can be summarised as follows
* cell proliferation
+ cellular states: M-phase, $`\mathrm{G}_1`$-phase, $`\mathrm{S}/\mathrm{G}_2`$-phase, $`\mathrm{G}_0`$-phase, necrotic, cornified
+ local mobile water concentration can prolong the duration of $`G_0`$ state for keratinocytes
+ conservation of target volume during M-phase
+ for stem cells: upper cell differentiates to a keratinocyte, lower cell remains a stem cell
+ keratinocytes can undergo a maximum of four transit proliferations, whereas stem cells divide ad infinitum
+ healthy melanocytes do not proliferate, whereas malignant melanocytes can divide ad infinitum
* cell growth: growth of cell volume at constant rate during $`\mathrm{G}_1`$
* cell death
+ low local nutrient concentration induces necrosis
+ keratinocytes undergo cornification after fourth generation
+ cornified cells without contact to others are removed from the simulation immediately
### II.3 Stochastic Elements
It is an empirical fact that processes in biological systems underly significant stochastic deviations: For example, biofilm cell populations starting from a single cell desynchronise proliferation after about five generations kreft1998 . Such a behaviour can not be explained by processes such as contact inhibition or nutrient depletion, as these are not active for small systems with only $`2^5`$ cells.
In the model, this is represented by stochastic elements that can be derived from a pseudo-random number generator matpack\_manual . The involved stochastic elements are the delta-correlated random forces (see appendix B) acting on every cell, the initial direction of the displacement vector at mitosis, and the durations of some cell cycle phases such as the M-phase, the $`\mathrm{S}/\mathrm{G}_2`$-phase, and the $`\mathrm{G}_0`$-phase.
As was done in previous models for biofilms kreft2001a ; picioreanu2004a , the initial direction of mitosis is determined from a random distribution, which is unifom on the unit sphere. This is the simplest modelling assumption that did not induce artifacts. In addition, it should be noted that during M-phase configuration changes are still possible due to interactions with the neighbouring cells.
In order to yield a sufficiently fast desynchronisation of the cell cycles, the individual duration times for the M-phase and the $`\mathrm{S}/\mathrm{G}_2`$-phase as well as the maximum duration time for the $`\mathrm{G}_0`$-phase are drawn from a normally-distributed random number generator matpack\_manual with a given mean and width. Without these stochastic elements, the model exhibits artificial oscillations around a steady state even in later stages. Technically, the duration of each phase is determined at the beginning of the phase. Naturally, the parameters on the random number generators can be set individually for every cell type.
### II.4 Computer Platform
The computer code was written following the paradigm of object-oriented programming in $`\mathrm{C}^{++}`$ and was compiled with the GNU compiler gcc version 3.3. The code was executed on an AMD Athlon(tm) MP Processor 1800+ with 1 GByte of RAM on a Linux platform.
### II.5 Simulation setup
As the computational domain, a rectangular box of dimensions $`200\mu \mathrm{m}\times 200\mu \mathrm{m}\times 400\mu \mathrm{m}`$ has been considered. Since epidermal tissue is anisotropic, the boundary conditions have to be chosen non-homogeneous as well. Note that the cellular kinetics is described with a system of ordinary differential equations (26). Therefore, the term “boundary condition” refers to the special interactions of cells with the boundary of the computational domain. It is known that a realistic epidermis exhibits a ruffled basal layer montagna1992 . However, in order to treat the microenvironment of epidermal tissue as simple as possible, the basal layer has been implemented here as a static planar boundary at the bottom with normal vector $`𝒆_z`$. With using the JKR model (8), the interaction with such a planar boundary can be well implemented by assuming contact with a cell of infinitely large radius. Specifically, the $`z`$-boundary has been assumed to be of infinite elasticity $`E_{\mathrm{bound}}=\mathrm{}`$. Since the inverse elastic moduli enter additively in the JKR model in equation (6), this choice does not sensitively change the global model behaviour but merely shifts the equilibrium distance between basal membrane and bottom cell layer. The corresponding adhesive anchorage in the basal layer has been made dependent on the cell type (see the discussion below). In order to minimise the boundary effects in $`x`$ and $`y`$ direction, periodic boundary conditions could be used for the cell cell interaction. This however would necessitate a rather tedious mirroring of cells close to the boundary. In addition, one would have to use periodic boundary conditions on the associated reaction-diffusion equations as well to avoid additional artifacts. Therefore, here a different (mirror cell) approach has been chosen: Every cell in contact with a $`x`$ or $`y`$ boundary is assumed to be in contact with a cell of the same type, size, receptor and ligand equipment, etc. In short, it interacts with a virtual mirror copy of itself, where the contact area is situated within the boundary plane. In upper $`z`$-direction there are no boundary conditions on the cells – recall that necrotic or cornified cells are removed eventually. In comparison to a static boundary this procedure also has the additional advantage that drag force artifacts are reduced.
The boundary conditions on the cells have their counterpart in the reaction-diffusion equations for the mobile water concentration and the nutrients: The concentrations at the lower $`z`$-boundary have been fixed to the maximum value (Dirichlet boundary conditions), and above the cell layers (dynamic thickness, a stratum corneum need not always exist during the simulations) both concentrations are fixed to 0. Technically, this has been implemented by setting the concentrations to vanish at all grid volume elements not containing any cells: The resolution of the reaction-diffusion grid was low enough to prevent the emergence of artificial sink terms in intercellular cavities throughout all simulations (such problems could – in principle – also be avoided completely by using Green functions newman2004 ). At the $`x`$ and $`y`$ boundaries, no-flux von Neumann boundary conditions have been used, i. e., $`_xu=0`$ and $`_yu=0`$. Note that this is equivalent to the corresponding boundary conditions on the cells: The boundary is impenetrable for both cells and nutrients. Thus, for an in $`x`$ and $`y`$ directions homogeneous cell distribution, the problem would effectively reduce to a one-dimensional one.
The initial conditions have been determined as follows: A monolayer of keratinocyte stem cells was distributed on the basal membrane. Afterwards, the position of the cells in the cell cycle has been randomised uniformly to avoid initial artifacts. This configuration could for example mimic a severely perturbed epidermis, where suddenly not only the stratum corneum but also the stratum medium was removed. Consequently, a strong proliferative response should be expected.
After establishment of a steady-state flow equilibrium, different perturbations have been performed. These will be discussed in the next section.
## III Results
### III.1 Flow equilibrium
Our first question was whether the proposed control mechanism of the water-concentration-induced prolongation of the cell cycle time could actually produce the macroscopically observed flow equilibrium of skin. In particular, we asked whether
* a steady-state flow equilibrium is established, and
* whether this equilibrium is stable against perturbations such as complete removal of the stratum corneum that is performed for example in tape-stripping experiments barthel2000 .
These questions can be interpreted as a sanity check of the model assumptions and it turns out that both have an affirmative answer (see figure 2).
Starting from a monolayer of cells, the local water concentration above the basal membrane is quite low such that no cell enters cell cycle prolongation. The net effect is an initial exponential growth phase (top left panel). After four generations, cornification of the first keratinocytes begins, followed by the formation of a stratum corneum with a considerably decreased diffusion coefficient for water. This in turn leads to an increased water concentration and thereby a greater fraction of cells residing in $`\mathrm{G}_0`$-phase: The initial exponential growth slows down and then the cell number decreases, since the cornified cells shed off the skins surface. Afterwards, the dynamics equilibrates. After 35 days, a tape-stripping experiment has been performed: All cornified cells are suddenly removed from the simulation. This leads again to a proliferative response. However, since this time the cornified layer quickly re-establishes due to reservoir cells in $`\mathrm{G}_0`$-phase, the proliferative response is considerably smaller than initially. Note that the dominant contribution to the rapid formation of the cornified layer in the model results from the fraction of $`\mathrm{G}_0`$-keratinocytes that have already reached their fourth generation. Interestingly, the oscillations around the equilibrium value are remarkably strong. The number of cells displays a slight (but decelerating) upward tendency, but 15 days after the disturbance (last vertical line), saturation is nearly reached. The final cell numbers correspond well to observed densities of keratinocytes (75000 cells per square mm skin at the breast bauer2001 ). In the top right panel of figure 2 it is demonstrated that with an intact stratum corneum (second and last frame), the water concentration is large in the lower layers of the epidermis and then falls rapidly. In figure 2 bottom row it becomes visible in the latest frame that the cornified layer exhibits a small hole (cells in dark grey). Due to a considerable loss of water, this causes many distant keratinocytes to leave their cell cycle arrest (cells in light grey changing to cells in dark grey) and thereby leads to a perturbation of the equilibrium.
In potten2000 the authors had hypothesised a diffusible substance that moderates cellular proliferation times. The present model does not contradict the hypothesis that this substance could simply be the moisture of the epidermis but other diffusible substances would presumably lead to equivalent model behaviour. Therefore, a confirmation/falsification of this model hypothesis would require more data on the candidate substances (diffusion coefficients, reaction rates, concentrations) and the associated processes.
### III.2 Melanocyte anchorage
Another question is how the degree of anchorage to the basal layer influences the ability of cancerous melanocytes to persist within the skin. It is well-known that most human melanoma cell lines have decreased or no expression of cadherins and exhibit a decreased ability to adhere to keratinocytes tang1994 . Therefore, this question is especially interesting from a clinical point of view. At first, we suspected that increased basal adhesion would lead to an increased fraction of melanocytes bound to the basal membrane and thereby a smaller fraction that is shed to regions where the nutrient supply falls below necrosis-inducing levels. Thus, the total number of melanocytes should decrease with decreasing anchorage. In order to test this, a single (non-proliferating) melanocyte was placed at the basal layer in the centre of the computational domain, and the system was evolved until flow equilibrium was established. Then, the melanocyte was turned cancerous by suddenly allowing for proliferation with a much larger rate than keratinocytes. In addition, we concomitantly reduced the anchorage to the basal layer. Starting from experience with multicellular tumour spheroids schaller2005a , we assumed the cycle time of cancerous melanocytes to be in the order of 15 hours. Surprisingly, it turned out that the overall growth dynamics was hardly dependent on the anchorage to the basal layer, see figure 3 left panel. Initially, the growth of melanocytes follows an exponential growth law, which is soon slowed down since the melanocytes reach distant regions from the basal layer, where nutrient support is poor. Since due to nutrient depletion the total number of viable cells already indicates saturation, also the total number of melanocytes must saturate eventually. Even with no adhesion to the basal membrane, comparable numbers of tumour cells were produced. Direct observation of the cross-sections (not shown) revealed the reason: With the given melanocyte proliferation rate of 15 hours, exponential growth was always faster than the epidermal flow induced by the turnover on the basal layer.
Consequently, we varied the proliferation rate of the cancerous cells in combination with complete loss of basal membrane anchorage (see figure 3 right panel). It turned out that there is a region of proliferation rates, where the melanocytes do not persist within the epidermis. This region is separated from the region of melanoma persistence by a comparably large domain where stochastic effects become important. Interestingly, in this case the period of coexistence of healthy skin and transformed cells may be remarkably long, which may give time for further malignant transformations in the realistic epidermis. It should be stressed that in this region the melanocyte proliferation rate is still much larger than the keratinocyte proliferation (their cycle prolongation is active for small melanoma). In addition, in the absence of death processes the growth law of keratinocytes follows the equation $`\dot{n}_{\mathrm{ker}}=16\alpha _{\mathrm{stem}}n_{\mathrm{stem}}`$, if one neglects the retardation induced by the four transient keratinocyte proliferations. This leads to linear growth only (with a fixed number of stem cells $`n_{\mathrm{stem}}`$), whereas the growth law of malignant melanocytes will be exponential $`n_{\mathrm{mel}}=n_{\mathrm{mel}}^0e^{\alpha _{\mathrm{mel}}t}`$ in the absense of death processes.
We further examined the region that separates melanoma persistence and complete shed-off of cancerous melanocytes by changing the melanocyte cycle time to $`\tau _{\mathrm{mel}}=(44.44\pm 5.56)`$ h. One finds that the usual spherical form one observes for in vitro tumour spheroids is considerably deformed for this system to cylinder-shaped or cone-shaped, compare figure 4 left panel. This is due to the pre-existent flow-equilibrium of the surrounding tissue and the effective one-dimensional diffusional constraint. Note also that the boundaries of the tumours are rather diffuse. Initially, a thin column of cancerous melanocytes is formed. Then, in the example in figure 4, left panel, first row, the melanocytes can persist within the life-sustaining zone until their growth velocity outweighs the upward-directed flow velocity and direct contact with the basal membrane is re-established. Afterwards, in the middle of the column of cancerous cells the upward forces are decreased, since for the interior cells there is no direct contact with keratinocytes moving upwards. In the simulations in figure 4 left panel, the thickness of the epidermis increases in those simulations where the tumour has re-established contact with the basal membrane. This is due to the displacement of keratinocytes – which are constrained in $`x`$ and $`y`$ dimensions – and also to the loss of the protective cornified layer, which leads to enlarged keratinocyte proliferation rates. It may be speculated that the cross-sections correspond to initial stages of a highly aggressive nodular melanoma moncrieff2001 that has not yet become clinically manifest. It may also be hypothesised that the micrometastases sometimes observed around primary melanoma in skin may correspond to branches of melanoma clones that have separated from the main clone during the upward flow. Interestingly, the shapes of these structures appear to be dynamically changing in these initial phases.
Using different initial seed values for the random number generator, we have performed several simulations with otherwise equal parameters. It turns out that completely different outcomes may occur in this region of melanocyte proliferation rates (figure 4 left panel and thick curves in the right panel). The stochastic effects result from stochastic forces, the randomly chosen mitotic direction, and the randomly distributed duration times of the cell cycle. In this in silico experiment, the different seed value did already lead to different configurations before the malignant transformation. More specific, the initial conditions for the growth of cancerous melanocytes had also been varied by employing stochastic elements before. In order to separate these effects, we started another series of simulations with equal initial seed values. In contrast to the previous simulations, the seed value of the random number generators was reset to different values right at the time of the malignant transformation. Thus, the initial environment of the cancerous melanocyte was the same in these simulations. It turned out that the variance of the outcomes narrowed considerably (thin grey curves in figure 4 right panel) but still exhibit large variations in the cell number (logarithmic plot). Thus, it can be concluded that the initial environment of cancerous melanocytes contributes significantly to the final outcome. Note that this does not only refer to the spatial cellular position, but also to the local proliferative state and thereby to the local upward flow velocity: The upward drag forces will be larger if the cancerous cell is surrounded by many proliferating keratinocytes with a net upward flow velocity.
In conclusion, stochastic effects generally play an important role in the initial phases of in silico melanoma development, since for the small cell numbers in the initial phases, they do not average out completely. In addition, their secondary consequences, i. e., the variation of the initial local environment by stochastic influences, are relevant.
## IV Model parameters
Reasonable dynamics has been achieved with the parameters in table 1.
The viscosity of the extracellular matrix $`\eta `$ determines the friction on loosely bound cells. Large viscosities lead to increased friction. Since viscous friction due to the cytoskeleton is assumed to be small ($`\gamma _{}0`$, compare appendix C) – also dominates friction in directions normal to the cell contact surfaces. Thereby, also the initial speed of cell division in M-phase is dominantly dependent on $`\eta `$. As long as the force relaxation occurs on a shorter timescale than the total cell doubling time, this does not have macroscopic effects on the evolution of the tissue. This is different when $`\gamma _{}`$ does not vanish. If it has the same order of magnitude as $`\gamma _{}`$, it will dominate the contribution inflicted by the viscosity $`\eta `$. However, if the magnitude of the total drag force coefficient does not change (marked by $`\gamma ^2=\gamma _{}^2+\gamma _{}^2`$), we have found by comparing the three extreme cases (that is, $`\gamma _{}=0,\gamma _{}=\gamma `$ and $`\gamma _{}=\gamma ,\gamma _{}=0`$ and $`\gamma _{}=\gamma _{}=\gamma /\sqrt{2}`$) that the differences in the overall population dynamics are rather small. It may be speculated that this is because in the present calculations the relaxation speed has no direct back-reaction on the number of cells, as in contrast to galle2005a ; schaller2005a contact inhibition has not been included in the model. As here absence of perpendicular friction has been assumed, the tangential friction coefficient $`\gamma _{}`$ dominantly determines the speed of relaxation within the tissue. The chosen value led to reasonable dynamics and has been estimated from galle2005a .
The adhesion energy density $`\sigma ^{\mathrm{max}}`$ determines the cell-cell equilibrium distance and the binding strength, which was a marker for the removal of necrotic or cornified cells. Generally, this value will in reality be time-dependent, compare also the discussion at the end of subsection II.1. Therefore, the binding energy density has been derived from the observed equilibrium distance chu2004 solving (8) instead. With this procedure, the equilibrium distances are in a physiological regime. Note that larger adhesion will lead to smaller equilibrium distances (with moderately increased contact surfaces and drag forces) but also to longer persistence times of dead cells, which results in an increased thickness of the stratum corneum. However, due to equation (3) this latter effect only enters logarithmically. When both the adhesion energy $`\sigma ^{\mathrm{max}}`$ and the minimum anchorage $`\mathrm{\Sigma }^{\mathrm{min}}`$ are decreased, one will still have to decrease the maximum stepsize in the numerical solution to maintain the level of accuracy. This is due to the fact that for decreased adhesion, the equilibrium distance and the contact distance $`d_{ij}^{\mathrm{contact}}=R_i+R_j`$ are closer together.
The equilibrium thickness of the cornified layer is strongly dependent on the receptor loss rate $`\alpha `$ and the minimum anchorage $`\mathrm{\Sigma }^{\mathrm{min}}`$. In addition, it will be sensitive to the cycle times of stem cells and keratinocytes, since these determine the number of keratinocytes finally undergoing cornification.
The elastic parameters correspond to approximate physiologic values for cells mahaffy2000 ; maniotis1997 ; galle2005a . However, it is known that – depending on the cell type and individual cytoskeleton – significant deviations may occur. With the given drag forces, mechanical relaxation occurs on a shorter scale than the cell cycle times, such that changes in physiologic windows have only small macroscopic consequences. It should be noted however that already for moderately changed Young moduli (and/or reduced Poisson moduli) the equilibrium distance between cells will be shifted, which might make decreased maximum spatial stepsizes necessary in the numerical solution to avoid unphysiological losses of contact.
As has already been discussed above, the stochastic elements may have significant influence on melanoma development. These can be divided into stochastic forces, randomly chosen durations of the cell cycle phases, and the random direction of mitosis.
Stochastic forces contribute to the detachment of cornified and necrotic cells, since these do neither advance in the cell cycle nor proliferate. We have found that small variations in the strength of stochastic forces in physiologic regimes only change the fluctuations in the epidermal thickness around the unchanged equilibrium value. On a technical level, the existence of a planar basal layer in combination with completely absent stochastic forces sometimes led to planar cell configurations at the basal layer, which is unfavourable for the Delaunay triangulation schaller2004 . As may be expected, considerably larger stochastic forces have a strong influence on the thickness of the stratum corneum, since loosely bound cells are removed much faster and the protective layer is lost easily. This in turn leads to loss of water and on-going reactions of keratinocytes that leave $`G_0`$-phase.
The values of the durations of M-phase $`\tau ^{(\mathrm{m})}`$, the $`S/G_2`$-phase $`\tau ^{\mathrm{S}/\mathrm{G}_2}`$ and the prolongation of the cell cycle $`\tau ^{(\mathrm{G}_0)}`$ influence the relative distribution of cells within the cell cycle, whereas the sum of their squared widths primarily determines the speed of desynchronisation of cell division (compare figures 3 and 4 right panels). Due to missing data, the durations of these cellular states have been fixed from a previous publication schaller2005a . The shortest observed cycle time determines the proliferation time for keratinocytes when the water concentration is below the critical threshold and has been estimated from experimental observations potten2000 . The system is most sensitive to the $`\mathrm{G}_0`$-phase prolongation time $`\tau ^{\mathrm{G}_0}`$, which has been estimated from barthel2000 to yield reasonable dynamics.
Without the modelling constraint that on division of keratinocyte stem cells, only the upper cell becomes a differentiating keratinocyte, the basal layer would loose more and more stem cells in the model. In other cell divisions, the simple assumption of a randomly distributed initial mitotic direction did not lead to numerical artifacts. However, it can be expected that the configuration of the neighbour cells soon changes the initial direction of the mitotic doublet.
The average cell volume of keratinocytes varies from $`425\mu \mathrm{m}^3`$ for cornified cells to $`800\mu \mathrm{m}^3`$ for stratum granulosum keratinocytes norlen2004 . Therefore, with the intrinsic assumption of spherical shape, the maximum cell radius has been fixed to $`R^{(\mathrm{m})}=5.0\mu \mathrm{m}`$, which also influences the time-dependent target volume. Note however, that within the stratum corneum the cornified cells flatten considerably and the realistic intrinsic cell shape cannot be regarded as spherical anymore.
The glucose uptake rate for cancerous melanocytes $`\lambda ^{(\mathrm{mel})}`$ has been chosen considerably larger than the glucose uptake rate of keratinocytes $`\lambda ^{(\mathrm{ker})}`$. This is motivated by the observation that cancerous cells have a considerably increased metabolism. The actual values are in the range observed for other tumour cells wehrle2000 . The minimum nutrient concentration $`U_{\mathrm{gluc}}^{\mathrm{crit}}`$, below which for melanocytes necrosis is induced, has been chosen to be in the order of $`1\mathrm{mM}`$, since necrosis of cancer cells becomes visible at these nutrient concentrations in vitro freyer1986a ; freyer1986b . Thereby, the combination of melanocyte nutrient uptake rate and minimum glucose concentration define a region, where melanocytes can survive.
For simplicity we have assumed that as a net effect the cells do not consume or secrete mobile water. A possible model extension could incorporate such effects by including cellular swelling during hydration. The critical mobile water concentration $`U_{\mathrm{H}_20}^{\mathrm{crit}}`$ has been adjusted to obtain a reasonable equilibrium thickness of the stratum medium with $`𝒪\left(5\right)`$ cell layers.
The apparent diffusivity of the mobile water $`D_{\mathrm{H}_20}^{\mathrm{strat}.\mathrm{med}.}`$ in stratum medium as well as in stratum corneum $`D_{\mathrm{H}_20}^{\mathrm{strat}.\mathrm{corn}.}`$ has been determined experimentally by various studies. Though strong variances exist, all of them predict a strong decline of the apparent diffusion coefficient kasting2003 ; bashir2001 ; schwindt1998 . Roughly speaking, the local water diffusion coefficients influence the gradient of the mobile water concentration: Large diffusion coefficients correspond to a small gradient. Therefore, for an intact stratum corneum the water concentration is approximately constant throughout the stratum medium and then falls rapidly, compare also figure 2 right panel.
The same general features hold true for the glucose diffusion coefficient $`D_{\mathrm{gluc}}^{\mathrm{tissue}}`$, which has specifically been determined for the human skin tuchin2001 . The glucose concentration at the basal layer $`U_{\mathrm{gluc}}^{\mathrm{bound}}`$ has been fixed to values that are normal for blood carvalho2004 . However, it should be noted that in reality the blood glucose concentration may vary significantly – for example after a meal. Since within the model for normal parameter sets anoikis is the predominant pathway for keratinocytes and dominantly the cancerous melanocytes consume glucose at large rates in the model, the glucose concentration strongly influences the chances of melanocyte survival here. An improved model could for example include an intracellular glucose reservoir to average out the time-dependent supply.
In order not to loose stem cells at the basal layer migrating upwards to the stratum corneum, the basal adhesion energy has been chosen to be twice the maximum adhesion energy density $`\sigma ^{\mathrm{max}}`$. This did suffice to disable loss of stem cells. For non-proliferating melanocytes, the basal adhesion has been chosen similarly.
## V Discussion
It had been demonstrated already in schaller2005a that with the aid of kinetic and dynamic weighted Delaunay triangulations agent-based models can treat up to $`10^5\mathrm{}10^6`$ cells. In the present contribution, it has been shown that with a more complete treatment of the equations of motion, such models can still handle $`10^4\mathrm{}10^5`$ cells.
Apart from these technicalities, from a biological point of view a diffusible substance can serve as a moderator on cellular proliferation in the epithelium. The parameters used do not contradict that a simple candidate of this substance could be the mobile water in the tissue. The homoeostasis was found to be roughly stable against perturbations such as tape-stripping experiments, which can serve as a sanity check on the model implementation.
Independently, the consequences of a varying basal adhesion of cancerous melanocytes have been studied. It turned out that these are strongly interlinked with the balance of proliferative melanocyte and keratinocyte activities. In addition, it has been shown that in some regions of parameter space, stochastic effects and especially their consequences on the initial state on the environment play an important role in the in silico representation of melanoma growth. Evidently, the model behaviour has been found under the precondition of several explicit and implicit approximations. These do of course limit the generality of the model and we want to summarise some shortcomings of the model below:
From our point of view, a significant macroscopic shortcoming of our approach is the failure of the model to explain the reduced thickness of the stratum corneum. This is at least partly due to the fact that the inherent cell shape is spherical, whereas cornified cells flatten and form polarised adhesive bindings montagna1992 . In reality, this will lead to a greater stability of the stratum corneum in comparison to the model, which would also imply a smoother evolution around the steady-state flow equilibrium than exhibited in figure 2 left panel. Possibly, choosing ellipsoids in contrast to spheres as the intrinsic cell shape dallon2004 may provide an alternative. Another possibility would be to use boundary-based models such as e. g. the extended Potts model savill2003 .
From the theoretical point of view, the model could be significantly improved by deriving a contact model valid for two-body interactions that also include non-normal forces and do not underlie the constraints of only small cell deformations. Also, for in vitro cell populations that are not fixed to a substrate, the effects of torque may become important. These refined theories however require much better experimental resolution than currently provided. It appears questionable whether centre-based models are able to cope with the increasing degree of complexity resulting from these improvements.
The basal layer has been approximated with a plane boundary condition in this article. Its replacement by a corrugated structure would significantly enlarge the region where water and nutrients are provided in abundance and thereby lead to a far greater cell reservoir that is able to start a proliferative response in case of injury. It may be speculated that this is one of the reasons why the ruffled basal layer has developed in skin. In addition, one would expect that a ruffled basal layer will also lead to a ruffled skin surface. Especially for the clinical question of melanoma invasion depth, the plane boundary condition should be replaced by a boundary that can be penetrated by malignant melanocytes. This would allow to study the time-course of initial invasion and to compare the invasion depth with clinical melanoma classifications.
The dynamics of the nutrients and of water has been described with a reaction-diffusion approach here. However, due to the cellular movement, there will also be a convective and a transport contribution that is completely neglected in the current simulations. With the large diffusion coefficient for water and nutrients in viable tissues, this approximation is presumably valid within the viable layers but may be questionable in the stratum corneum. Note that the polarised structure of the cornified cells in the stratum corneum may also give rise to non-isotropic diffusion, where the diffusion coefficient is not a scalar value anymore. To a first approximation however, this effect may be well absorbed into the apparent diffusion coefficient as is done in the experimental measurements.
The cell cycle has been approximated here by a small number of internal cellular states only. It may also be questioned whether a subdivision into discrete substates makes sense. One may also expect a much smoother reaction of the epidermis to the removal of all keratinocytes if transition into and out of $`\mathrm{G}_0`$-phase would not depend on a threshold water concentration, but would be determined by transition probabilities that may continuously depend on the water concentration. This may be judged with quantified experimental data.
The model also uses comparably many parameters but all of them have a distinct physical counterpart. This makes it in principle possible to determine these parameters by independent experiments. Despite of all the previously-mentioned shortcomings (most of these being valid for lattice-based approaches as well), off-lattice agent-based models also have important advantages over most lattice models: They have the intrinsic potential to use physical (realistic) parameters with a moderate increase in computational effort. This opens the possibility to gain knowledge about the system by falsifying the model using independent experiments. Therefore, quantified experiments on well-defined experimental systems are of urgent interest to constrain the uncontrolled growth in the number of theoretical models on cellular tissue.
## VI Acknowledgements
G. S. is indebted to J. Galle and T. Beyer for valuable discussions on contact models, physiologic parameters, and numerical algorithms. FIAS is supported by the ALTANA AG.
## Appendix A The JKR contact model
Already the dynamics of rigid bodies in contact is a difficult problem, as the local geometry at the contact region will strongly influence the involved forces. Therefore, most contact models applied in practice are not motivated by microscopic assumptions but rather mimic the realistic behaviour.
The JKR-model includes elastic and adhesive (but not viscous) interaction of solid spheres. It is often used in a biological context to estimate cellular parameters from experimental observations (JKR-test, verdier2003 ). Thus, one can at least on short time scales hope, that even though the parameters derived from such measurements moy1999 will not yield a correct description of the cytoskeleton (which is known to be viscoplastic), their usage in the model will at least lead to dynamics similar to that observed in the experiments.
The characteristics of the JKR contact model relevant for our considerations can be summarised as follows: Two spheres $`i`$ and $`j`$ placed at positions $`𝒙_i`$ and $`𝒙_j`$, having radii $`R_{i/j}`$, Young moduli $`E_{i/j}`$, Poisson moduli $`\nu _{i/j}`$, and contact surface energy density $`\sigma _{ij}`$ underlie the interaction force johnson1971
$`F_{ij}^{\mathrm{JKR}}`$ $`=`$ $`\left[{\displaystyle \frac{K_{ij}a_{ij}^3}{R_{ij}}}\sqrt{6\pi \sigma _{ij}K_{ij}a_{ij}^3}\right],`$ (5)
where $`a_{ij}`$ denotes the radius of the circular contact area between the deformed spheres, $`R_{ij}`$ the reduced radius, and $`K_{ij}`$ incorporates the combined elastic properties
$`R_{ij}={\displaystyle \frac{R_iR_j}{R_i+R_j}},K_{ij}={\displaystyle \frac{1}{\frac{3}{4}\left(\frac{1\nu _i^2}{E_i}+\frac{1\nu _j^2}{E_j}\right)}}.`$ (6)
For vanishing adhesive properties ($`\sigma _{ij}=0`$) one recovers the purely elastic Hertz model hertz1882 ; landau1959 . The contact radius $`a_{ij}`$ is related to the indentation or overlap (see figure 5 left panel) $`h_{ij}=R_i+R_j\left|𝒙_i𝒙_j\right|`$ via johnson1971 ; brilliantov2004
$`h_{ij}={\displaystyle \frac{a_{ij}^2}{R_{ij}}}\sqrt{{\displaystyle \frac{8\pi \sigma _{ij}}{3K_{ij}}}}\sqrt{a_{ij}},`$ (7)
which may have – depending on the value of $`h_{ij}`$ – none, one, or two solutions with $`a_{ij}>0`$. For relatively small adhesion $`\sigma _{ij}/(K_{ij}R_{ij})1`$, the second term on the right hand side can be neglected, and the solution $`a_{ij}\sqrt{h_{ij}R_{ij}}`$ can be inserted into equation (5) to yield an approximate force-distance relationship brilliantov2004
$`F_{ij}^{\mathrm{JKR}}`$ $``$ $`\left[K_{ij}R_{ij}^2\left({\displaystyle \frac{h_{ij}}{R_{ij}}}\right)^{3/2}\sqrt{6\pi \sigma _{ij}K_{ij}R_{ij}^3}\left({\displaystyle \frac{h_{ij}}{R_{ij}}}\right)^{3/4}\right],`$ (8)
which has been used as the JKR force throughout this article. The force is negative (adhesive) for small overlaps $`h_{ij}`$ and becomes positive (repulsive) for larger overlaps. Note that, independent on the approximation of small adhesion in (7), the adhesive force has the maximum magnitude
$`F_{ij}^{\mathrm{adh}}={\displaystyle \frac{3}{2}}\pi \sigma _{ij}R_{ij},`$ (9)
which is also independent on the elastic cell properties and thus allows an estimate of $`\sigma _{ij}`$ from cell-doublet-rupture experiments such as e. g. benoit2000 ; chu2004 . Since in reality the spheres underlie deformation, the resulting approximate sphere contact surface in JKR theory
$`A_{ij}^{\mathrm{JKR}}=\pi a_{ij}^2\pi h_{ij}R_{ij}`$ (10)
is in general different from the virtual contact surface that would follow intuitively from the sphere overlap region (figure 5 left panel). The above contact surface has been chosen in the model to make it intrinsically consistent. In the following, the short hand notations $`R_{\mathrm{min}}=\mathrm{min}\{R_i,R_j\}`$ and $`R_{\mathrm{max}}=\mathrm{max}\{R_i,R_j\}`$ will be used with suppressed indices.
For the approximate theory (8) one can introduce a two-body interaction potential via
$`F_{ij}^{\mathrm{JKR}}={\displaystyle \frac{V^{\mathrm{JKR}}}{d_{ij}}}=+{\displaystyle \frac{V^{\mathrm{JKR}}}{h_{ij}}}={\displaystyle \frac{1}{R_{ij}}}{\displaystyle \frac{V^{\mathrm{JKR}}}{h_{ij}/R_{ij}}},`$ (11)
which leads for our case to
$`V_{ij}^{\mathrm{JKR}}\left(h_{ij}/R_{ij}\right)={\displaystyle \frac{2}{5}}K_{ij}R_{ij}^3\left({\displaystyle \frac{h_{ij}}{R_{ij}}}\right)^{5/2}{\displaystyle \frac{4}{7}}\sqrt{6\pi \sigma _{ij}K_{ij}R_{ij}^5}\left({\displaystyle \frac{h_{ij}}{R_{ij}}}\right)^{7/4},`$ (12)
which is a special case of the Lennard-Jones potential (compare figure 5). However, here the parameters have either been linked to cellular properties that are accessible by independent experiments or been fixed by microscopic assumptions.
The quantity $`h_{ij}/R_{ij}`$ describes the relative position of both spheres and is related to the orthogonal sphere distance (compare edelsbrunner1996 )
$`\pi (\widehat{𝒙},\widehat{𝒚})=\left(𝒙𝒚\right)^2R_x^2R_y^2`$ (13)
for the spheres $`\widehat{𝒙}=(𝒙,R_x^2)`$ and $`\widehat{𝒚}=(𝒚,R_y^2)`$ via
$`\pi (\widehat{𝒓}_i,\widehat{𝒓}_j)=\left({\displaystyle \frac{h_{ij}}{R_{ij}}}\right)^2R_{ij}^22\left({\displaystyle \frac{h_{ij}}{R_{ij}}}1\right)R_iR_j,`$ (14)
compare also figure 6 left panel.
The full JKR-theory has several shortcomings:
1. It is only valid for small deformations $`h_{ij}/R_{ij}1`$, since the linear elastic theory assumed in the derivation of (5) is not valid for large deformations landau1959 . In addition, it approximates the cytoskeleton as a homogeneous solid, which is not the case verdier2003 . Regarding the numerical solution of the interacting particle system, this has the consequence that some cells may be completely covered by others, since the JKR force (8) does not diverge at complete overlaps. To circumvent this, a modified interaction potential has been used, which displays this divergence
$`V(x)`$ $`=`$ $`f(x)V^{\mathrm{JKR}}(x),`$
$`f(x)`$ $`=`$ $`\{\begin{array}{ccc}\frac{(x_\mathrm{d}x_\mathrm{m})^2}{(x_\mathrm{d}2x_\mathrm{m})(x_\mathrm{d}x)}\frac{x}{x_\mathrm{d}2x_\mathrm{m}}& :& x_\mathrm{m}xx_\mathrm{d}\\ 1& :& \text{else}\end{array},`$ (17)
if one chooses as matching point $`x_\mathrm{m}=1`$ and as point for divergence $`x_\mathrm{d}=2+2R_{\mathrm{min}}/R_{\mathrm{max}}`$ (compare figure 6). The choice of this modified potential only led to significantly different growth dynamics for $`𝒪\left(10^4\right)`$ cells if cellular growth was constrained by static boundaries galle2005b , which indicates that the used drag forces (see appendix C) were small enough to enable fast relaxation.
2. The JKR-theory does neither include viscous effects arising from the cytoskeleton nor dissipation occurring in the extracellular matrix. Therefore, the model has been supplemented with additional drag forces, which are specified in appendix C.
3. The original result (5) has been derived as a pure two-body interaction johnson1971 , which is also the case for its purely elastic precursor hertz1882 ; landau1959 . However, for many adhering spheres already for small individual deformations additional forces will come into play, since
* the spheres are pre-stressed,
* the contact regions of various cells may overlap.
Thus, the JKR model does not correctly describe cellular compression for multiple overlaps. The extent of this shortcoming will critically depend on the current adjacency topology which makes an analytical approach infeasible. For numerical ease and due to missing estimates in this article the following practitioners approach has been chosen. Below the target cell volume $`V_{i/j}^{\mathrm{target}}`$ the cell experiences additional repulsive – isotropic – forces due to compression of the cytoskeleton. Then, the resulting additional repulsive force acts in the direction of the neighbours $`j`$ with magnitude
$`F_{ij}^{\mathrm{comp}}`$ $`=`$ $`A_{ij}\left[{\displaystyle \frac{E_i}{3(12\nu _i)}}\left(1{\displaystyle \frac{V_i}{V_i^{\mathrm{target}}}}\right)+{\displaystyle \frac{E_j}{3(12\nu _j)}}\left(1{\displaystyle \frac{V_j}{V_j^{\mathrm{target}}}}\right)\right],`$ (18)
where $`V_{i/j}`$ denote the current cellular sphere volumes (reduced by the overlaps with neighbouring spheres) and $`A_{ij}`$ the circular JKR contact surfaces in equation (10). Note that owing to model simplicity, neither volume nor surface corrections schaller2005a are calculated with the Voronoi tessellation in this article.
4. Whereas the forces in the approximate model (8) only depend on the actual relative cellular positions, a more realistic scenario would have to include hysteresis effects, as adhesive intercellular bonds form after contact chu2004 . This however would require to keep track of the time evolution of cellular adjacencies. In part, the time evolution can be incorporated into the time dependence of the adhesive parameters
$`\sigma _{ij}={\displaystyle \frac{\sigma ^{\mathrm{max}}}{2}}\left[C_i^{\mathrm{rec}}(t)C_j^{\mathrm{lig}}(t)+C_i^{\mathrm{lig}}(t)C_j^{\mathrm{rec}}(t)\right],`$ (19)
where the $`0C_{i/j}^{\mathrm{rec}/\mathrm{lig}}(t)1`$ represent the receptor or ligand densities on the cell membrane – normalised relative to a maximum density, and $`\sigma ^{\mathrm{max}}`$ is the maximum adhesion energy, respectively. It must be noted that also the cytoskeleton reorganises and thereby the intrinsic cell shape will not remain spherical after contact. A full description of these effects would therefore not only require time-dependent elastic parameters ($`E_i`$, $`\nu _i`$), but also the implementation of a dynamically changing intrinsic equilibrium cell-shape, which is presumably not within the reach of a centre-based model drasdo2003 ; kreft2001a .
5. In addition, the derivation of the JKR model relies on the fact that only normal forces act. For cell doublets with friction, shear forces will in reality exist. It is assumed here that the net effect of shear forces on the validity of the JKR approach can be neglected, such that they can be independently included in the drag forces.
At least for keratinocytes the application of the JKR model to cell doublet rupture experiments chu2004 leads to discrepancies between the visual equilibrium distance and the equilibrium distance predicted by the full JKR-model (5): If one derives via equation (9) the maximum adhesion energy density from the maximum rupture force recorded in chu2004 , the resulting equilibrium distance predicted by (5) is considerably different than observed in the figures of the same publication: The indentation $`h_{ij}`$ resulting from equation (7) becomes negative (pointing to extrapolation of JKR theory beyond the region of its validity), whereas for the approximate JKR model, the limiting condition $`\sigma _{ij}/(K_{ij}R_{ij})1`$ is certainly violated, which would lead to considerably smaller equilibrium distances (larger indentations) than in reality. For example, for cell-cell contact times smaller than 30 seconds, average rupture forces of 20 nN have been measured chu2004 . Assuming $`K_{ij}=1000`$ Pa and $`R_{ij}=2.5\mu \mathrm{m}`$ one would thereby find from equation (9) an adhesion energy density of $`\sigma ^{\mathrm{max}}1.7\mathrm{nN}/\mu \mathrm{m}`$. However, then the equilibrium distances resulting from equations (5) or (8), respectively, are inconsistent with the equilibrium distances in chu2004 . This indicates that the JKR model is not directly applicable to strongly adhesive cells. For larger times, the discrepancy becomes even worse.
However, we expect that all these shortcomings are not major sources of error if one aims at analysing control mechanisms. An improved contact model could generally be included in such simulations, but it should be reasonably motivated by microscopic theories or experimental data first.
## Appendix B Random Forces
Due to thermal fluctuations, any particle in a solution will be subject to random forces (Brownian motion). In addition, some cell types exhibit intrinsic (active) movement which sometimes appear to be of random nature and thus follow the same mathematics as Brownian motion. For systems with these characteristics, the time-dependent stochastic forces $`𝑭(t)`$ modelling the random behaviour have to fulfil two conditions ma2003 :
1. their mean vanishes $`𝑭(t)=0`$ and
2. the forces are not correlated, i. e., $`𝑭(t_1)𝑭(t_2)=3\xi ^2\delta (t_1t_2)`$.
The parameter $`\xi `$ thereby quantifies the strength of the stochastic fluctuations. The movement of single cells in a solution is highly overdamped dallon2004 , and any stochastic force fulfilling the above conditions will lead in the Langevin equation to a diffusion-like evolution of cellular distribution, i. e., in the absence of additional forces the squared displacement will be given by
$`\left[𝒓(t)𝒓_0\right]^2=3{\displaystyle \frac{\xi ^2}{\gamma ^2}}t=6Dt,`$ (20)
where $`D`$ is the corresponding diffusion coefficient and $`\gamma `$ is a dampening constant, which effectively describes the strength of friction. The above identity is also known as the fluctuation dissipation theorem, since it connects the fluctuations ($`\xi `$) with the dissipation ($`\gamma `$). If this dynamics is observed for free spherical cells in medium, the friction constant can for highly-damped dynamics be well approximated by the Stokes friction $`\gamma _0=6\pi \eta R`$, where $`R`$ represents the radius of the cell and $`\eta `$ the viscosity of the surrounding medium. Evidently, with the same random forces applied, cellular movement will be much smaller if drag forces due to cellular bindings are at work. For numerical implementations, a stochastic force fulfilling the above conditions can be given by ma2003
$`F_i(t)`$ $`=`$ $`{\displaystyle \frac{\xi }{\sqrt{\mathrm{\Delta }t}}}\chi _{0,1}^{\mathrm{GAUSS}},`$ (21)
where $`\mathrm{\Delta }t`$ describes the width of the timestep, and $`\chi _{0,1}^{\mathrm{GAUSS}}`$ is a random number drawn from a Gaussian distribution matpack\_manual with mean $`\mu _\chi =0`$ and width $`\sigma _\chi =1`$.
It should be noted however, that active random eucaryotic movement in reality usually occurs with pseudopods fletcher2004 : The cell attaches protrusions to neighbouring cells (or the extracellular matrix) and randomly pulls towards them. This has two further implications
* the stochastic forces become two-body forces, i. e., the neighbour cell that the pseudopod is attached to, is subject to the corresponding negative force. Also, the forces act into the direction of the normals galle2005a .
* Since the pseudopods do not enable pushing, the average stochastic force component into the direction of a given neighbour cell will not in general vanish. For example, at interfaces of dense tissue (where the pseudopods find resistance) and fluid (where no net force can be generated) one cannot expect the contributions into the different directions to compensate each other.
Since the intrinsic logic behind active cellular movement following Brownian mathematics is not fully understood and also active movement with pseudopods is not quantified for the cell types considered in the simulation (keratinocytes and melanocytes), we have chosen to implement stochastic forces via equation (21) as acting randomly on every cell that reacts passively to these in return.
## Appendix C Equations of Motion
For $`N`$ spherical cells with positions $`𝒙_i(t)`$ and radii $`R_i`$ subject to cell-cell as well as cell-medium and cell-substrate interactions, the equations of motion can in the reference frame of motionless medium and boundaries be summarised as (compare also ferrez2001 )
$`m_i\ddot{x}_i^\alpha `$ $`=`$ $`F_i^\alpha +{\displaystyle \underset{J𝒩(i)}{}}F_{iJ}^\alpha +{\displaystyle \underset{j𝒩𝒩(i)}{}}F_{ij}^\alpha {\displaystyle \underset{\beta }{}}\gamma _i^{\alpha \beta }\dot{x}_i^\beta `$
$`{\displaystyle \underset{J𝒩(i)}{}}{\displaystyle \underset{\beta }{}}\gamma _{iJ}^{\alpha \beta }\left[\dot{x}_i^\beta +R_i\left(𝒏_{iJ}\times 𝝎_i\right)^\beta \right]{\displaystyle \underset{\beta }{}}{\displaystyle \underset{j𝒩𝒩(i)}{}}\gamma _{ij}^{\alpha \beta }\left[\dot{x}_i^\beta \dot{x}_j^\beta +R_i\left(𝒏_{ij}\times 𝝎_i\right)^\beta +R_j\left(𝒏_{ij}\times 𝝎_j\right)^\beta \right],`$
$`I_i\dot{\omega }_i^\alpha `$ $`=`$ $`T_i^\alpha +{\displaystyle \underset{J𝒩(i)}{}}R_i\left(𝒏_{iJ}\times 𝑭_{iJ}\right)^\alpha +{\displaystyle \underset{j𝒩𝒩(i)}{}}R_i\left(𝒏_{ij}\times 𝑭_{ij}\right)^\alpha {\displaystyle \underset{\beta }{}}\mathrm{\Gamma }_i^{\alpha \beta }\omega _i^\beta `$
$`{\displaystyle \underset{J𝒩(i)}{}}{\displaystyle \underset{\beta }{}}\mathrm{\Gamma }_{iJ}^{\alpha \beta }\left[\dot{x}_i^\beta +R_i\left(𝒏_{iJ}\times 𝝎_i\right)^\beta \right]{\displaystyle \underset{\beta }{}}{\displaystyle \underset{j𝒩𝒩(i)}{}}\mathrm{\Gamma }_{ij}^{\alpha \beta }\left[\dot{x}_i^\beta \dot{x}_j^\beta +R_i\left(𝒏_{ij}\times 𝝎_i\right)^\beta +R_j\left(𝒏_{ij}\times 𝝎_j\right)^\beta \right],`$
where $`\alpha ,\beta \{0,1,2\}`$ denote the Cartesian indices and $`i,j\{0,1,\mathrm{},N1\}`$ the cellular indices. The first equation describes the evolution of the cell positions $`𝒙_i(t)`$, whereas the second equation accounts for the evolution of the cellular spin velocities $`𝝎_i(t)`$. The terming $`𝒩𝒩(i)`$ denotes all cells having direct contact with cell $`i`$, whereas $`𝒩(i)`$ refers to all boundaries in direct contact with cell $`i`$. Such a set of neighbouring cells can be efficiently determined as a subset of all neighbours in the weighted Delaunay triangulation of the set of spheres. (We had developed and applied such a triangulation module previously in schaller2004 ; schaller2005a .) Since for most problems few boundary conditions will be given, these are hard-wired in the code for every specific problem individually. Note that the back-reaction of the cells on the boundaries is neglected implicitly assuming that the boundaries are stationary.
The first term on the right-hand side of the first equation $`F_i^\alpha `$ may generally include deterministic (for example, crawling forces on a substrate) and stochastic (e. g. Brownian motion) forces on a single cell, whereas the second and third terms $`_{j𝒩(i)}F_{iJ}^\alpha `$ and $`_{j𝒩𝒩(i)}F_{ij}^\alpha `$ include the cell-boundary and intercellular two-body forces (e. g. stochastic two-body forces or the deterministic JKR-force, compare subsections B and A), respectively. The fourth term $`_\beta \gamma _i^{\alpha \beta }\dot{x}_i^\beta `$ denotes cell-medium friction, whereas the last two terms denote dampening due to friction with the boundaries ($`\gamma _{iJ}^{\alpha \beta }`$) and with neighbouring cells ($`\gamma _{ij}^{\alpha \beta }`$), respectively. Note that the dampening forces can be divided in a contribution proportional to a relative cell velocity and a contribution arising from the angular velocities of both cells, where the normal vector $`𝒏_{ij}`$ is understood to point from cell $`i`$ towards cell $`j`$ (which restores the apparently violated antisymmetry of the dampening forces under exchange of $`i`$ and $`j`$).
The quantity $`I_i`$ on the left-hand side of the second equation denotes the inertial momentum ($`I_i=\frac{2}{5}m_iR_i^2`$ for rigid homogeneous spheres). In analogy with the forces, the first term on the right-hand side of the second equation $`T_i^\alpha `$ describes an intrinsic torque of the $`i^{\mathrm{th}}`$ cell, whereas the second and third terms describe the torques generated by cell-boundary and cell-cell interactions. In contrast to the forces, the dampening constant $`\mathrm{\Gamma }_i^{\alpha \beta }`$ (capital coefficients denote rotational dampening) does not only incorporate friction with the surrounding fluid but also the dissipation of rotational energy into internal degrees of freedom of the cell (i. e., finally heat). As with the forces, the last terms describe the rotational dampening due to cell-boundary and cell-cell interaction, respectively.
Note the equation for the rotation may generally back-react onto the first equation via other channels as well. For example, the interaction forces could depend on respective angular momenta. Moreover, the terms describing the influence of the torques on the angular velocity implicitly assume that the cell is a rigid body, which is not the case. Although already sophisticated enough, the above equations should therefore be regarded a simple possible ansatz.
In the over-damped approximation
$`m_i\ddot{x}_i^\alpha 0\text{and}I_i\dot{\omega }_i^\alpha 0,`$ (23)
which is widely used to describe cell movements in fluids, the interaction forces and torques are always balanced by the friction forces and torques, respectively. Concerning the friction torques we assume that the internal friction of the cell is dominant ($`\mathrm{\Gamma }_i^{\alpha \beta }\mathrm{\Gamma }_{ij}^{\alpha \beta },\mathrm{\Gamma }_{iJ}^{\alpha \beta }`$) and that there is no intrinsic torque generated by the cell types we consider ($`T_i^\alpha =0`$). In addition, we assume that the dominant torque dampening is approximately isotropic ($`\mathrm{\Gamma }_i^{\alpha \beta }=\delta ^{\alpha \beta }\mathrm{\Gamma }_i`$). Then, one obtains for the angular velocity
$`\mathrm{\Gamma }_i\omega _i^\alpha {\displaystyle \underset{J𝒩(i)}{}}R_i\left(𝒏_{iJ}\times 𝑭_{iJ}\right)^\alpha +{\displaystyle \underset{j𝒩𝒩(i)}{}}R_i\left(𝒏_{ij}\times 𝑭_{ij}\right)^\alpha .`$ (24)
If one inserts the above expression into the first equation of (C), one observes that the friction terms describing the influence of the torque on the cellular force dampening is suppressed by prefactors of $`\gamma _{iJ}^{\alpha \beta }R_i^2/\mathrm{\Gamma }_i`$ and $`\gamma _{ij}^{\alpha \beta }R_i^2/\mathrm{\Gamma }_i`$. Whether these terms can be neglected, is dominantly related to the structure of the cytoskeleton. We assume here that the cytoskeleton does not transmit shear forces well.
With these approximations, the relevant equations of motion take the form
$`0`$ $``$ $`F_i^\alpha +{\displaystyle \underset{J𝒩(i)}{}}F_{iJ}^\alpha +{\displaystyle \underset{j𝒩𝒩(i)}{}}F_{ij}^\alpha {\displaystyle \underset{\beta }{}}\gamma _i^{\alpha \beta }\dot{x}_i^\beta {\displaystyle \underset{J𝒩(i)}{}}{\displaystyle \underset{\beta }{}}\gamma _{iJ}^{\alpha \beta }\dot{x}_i^\beta {\displaystyle \underset{\beta }{}}{\displaystyle \underset{j𝒩𝒩(i)}{}}\gamma _{ij}^{\alpha \beta }\left(\dot{x}_i^\beta \dot{x}_j^\beta \right).`$ (25)
Note that solving the remaining equation for the cellular spin velocities (24) is not necessary, since its back-reaction on the cell movement has been neglected and any snapshot of a rotating sphere cannot be distinguished from a motionless sphere. The above equation can be rewritten to yield
$`{\displaystyle \underset{k,\beta }{}}\left\{\left[\gamma _k^{\alpha \beta }+{\displaystyle \underset{j}{}}\gamma _{kj}^{\alpha \beta }+{\displaystyle \underset{J}{}}\gamma _{kJ}^{\alpha \beta }\right]\delta _{ik}\gamma _{ik}^{\alpha \beta }\right\}\dot{x}_k^\beta =F_i^\alpha +{\displaystyle \underset{j}{}}F_{ij}^\alpha .`$ (26)
From the properties of the friction coefficients it can also be deduced that the linear system defined by this equation is symmetric and also diagonally dominant as long as $`\gamma _i^{\alpha \alpha }+_J\gamma _{iJ}^{\alpha \alpha }>0i,\alpha `$. In addition, it must be noted that the system will be extremely sparsely populated as the friction coefficients vanish for all cells not being in direct contact, see the appendix D for an example.
A usual choice for cell-medium friction is the well-known Stokes-relation $`\gamma _i^{\alpha \beta }=6\pi \eta R_i\delta ^{\alpha \beta }`$ joos1989 introduced in subsection B. The friction coefficients and two-body forces fulfil the following conditions:
$`\gamma _{ij}^{\alpha \beta }=\gamma _{ji}^{\alpha \beta },F_{ij}^\alpha `$ $`=`$ $`F_{ji}^\alpha \text{(Newton’s third axiom)},`$
$`\gamma _{ij}^{\alpha \beta }`$ $`=`$ $`\gamma _{ij}^{\beta \alpha }\text{(projection operator property)},`$
$`\gamma _i^{\alpha \beta }`$ $`=`$ $`\gamma _i\delta ^{\beta \alpha }\text{(isotropy)},`$
$`\gamma _{ii}^{\alpha \beta }`$ $`=`$ $`0\text{(no self-friction)}.`$ (27)
In a strict sense, Newton’s third axiom only applies to the total two-body force. However, here the model should consistently include contact forces $`F_{ij}^\alpha `$ and drag forces $`\gamma _{ij}^{\alpha \beta }\left(\dot{x}_i^\beta \dot{x}_j^\beta \right)`$, which may act independently from each other. Therefore, actio et reactio has been assumed to act separately. The symmetry in $`(\alpha ,\beta )`$ of the friction coefficients also arises from the symmetry properties of the projection operators: The drag forces expressed by the friction coefficients may be divided in normal drag forces and tangential (shear) drag forces. Assuming that they are proportional to the effective contact area between two cells $`i`$ and $`j`$
$`A_{ij}^{\mathrm{eff}}=A_{ij}{\displaystyle \frac{1}{2}}\left[C_i^{\mathrm{rec}}(t)C_j^{\mathrm{lig}}(t)+C_i^{\mathrm{lig}}(t)C_j^{\mathrm{rec}}(t)\right]`$ (28)
and to the normal or tangential projection of the velocity differences, respectively, the friction coefficients take the form
$`\gamma _{ij}^{\alpha \beta }`$ $`=`$ $`A_{ij}^{\mathrm{eff}}\left(\gamma _{}𝒫_{ij,}^{\alpha \beta }+\gamma _{}𝒫_{ij,}^{\alpha \beta }\right)\left(1\delta _{ij}\right),`$
$`\gamma _{iJ}^{\alpha \beta }`$ $`=`$ $`A_{iJ}\left(\gamma _{}𝒫_{iJ,}^{\alpha \beta }+\gamma _{}𝒫_{iJ,}^{\alpha \beta }\right).`$ (29)
The friction constant $`\gamma _{}`$ predominantly describes internal friction within the cytoskeleton galle2005a , since force contributions for movements normal to the cell-cell contact surface are already contained within the JKR interaction model. The friction constant $`\gamma _{}`$ is set to vanish within this article thereby implicitly assuming that dampening due to friction within the cytoskeleton is much smaller than dampening due to cell-cell bindings. In contrast, the tangential friction constant $`\gamma _{}`$ describes drag forces resulting from broken bindings during movements tangential to the intercellular contact plane dallon2004 ; galle2005a . For model consistency, the used contact surfaces are chosen identical with the JKR contact surface (10). Since over a wide range of physiological overlaps this relates to the spherical overlap that would result from undeformed spheres by about a factor of two, a different choice of the contact surface could be compensated by appropriately changed friction parameters.
The intercellular tangential and perpendicular projectors are given by
$`𝒫_{ij,}^{\alpha \beta }=\delta ^{\alpha \beta }n_{ij}^\alpha n_{ij}^\beta ,𝒫_{ij,}^{\alpha \beta }=n_{ij}^\alpha n_{ij}^\beta ,`$ (30)
and the cell-boundary projectors
$`𝒫_{iJ,}^{\alpha \beta }=\delta ^{\alpha \beta }n_{iJ}^\alpha n_{iJ}^\beta ,𝒫_{iJ,}^{\alpha \beta }=n_{iJ}^\alpha n_{iJ}^\beta ,`$ (31)
respectively. In the above projection operators, $`𝒏_{ij}`$ represents the normal vector pointing from cell $`i`$ towards cell $`j`$ (compare also figure 7 left panel), whereas $`𝒏_{iJ}`$ denotes the normal vector of the boundary $`J`$ at the contact point with cell $`i`$. Note that with these projection operators, the conditions on the friction coefficients (C) are automatically fulfilled.
## Appendix D Numerical Solution
An example including cell-cell and cell-boundary friction is illustrated in figure 7. Indeed, for this special example all non-isotropic friction coefficients vanish except $`\gamma _{03}^{\alpha \beta },\gamma _{04}^{\alpha \beta },\gamma _{23}^{\alpha \beta },\gamma _{34}^{\alpha \beta },\mathrm{\Gamma }_0^{\alpha \beta },\mathrm{\Gamma }_4^{\alpha \beta }`$.
Consequently, for this example the system (26) would assume the form
$`\left(\begin{array}{ccccc}𝚪_0+𝜸_0+𝜸_{03}+𝜸_{04}& 𝕆& 𝕆& 𝜸_{03}& 𝜸_{04}\\ 𝕆& 𝜸_1& 𝕆& 𝕆& 𝕆\\ 𝕆& 𝕆& 𝜸_2+𝜸_{23}& 𝜸_{23}& 𝕆\\ 𝜸_{03}& 𝕆& 𝜸_{23}& 𝜸_3+𝜸_{03}+𝜸_{34}& 𝜸_{34}\\ 𝜸_{04}& 𝕆& 𝕆& 𝜸_{34}& 𝜸_4+𝜸_4+𝜸_{04}+𝜸_{34}\end{array}\right)\left(\begin{array}{c}\dot{𝒙}_0\hfill \\ \dot{𝒙}_1\hfill \\ \dot{𝒙}_2\hfill \\ \dot{𝒙}_3\hfill \\ \dot{𝒙}_4\hfill \end{array}\right)=`$ (42)
$`\left(\begin{array}{c}𝑭_0+𝑭_{03}+𝑭_{04}+𝑭_{0\mathrm{B}}\hfill \\ 𝑭_1\hfill \\ 𝑭_2+𝑭_{23}\hfill \\ 𝑭_3+𝑭_{30}+𝑭_{32}+𝑭_{34}\hfill \\ 𝑭_4+𝑭_{40}+𝑭_{42}+𝑭_{43}+𝑭_{4\mathrm{B}}\hfill \end{array}\right),`$ (48)
where in three dimensions the symbols $`𝑭_i`$ and $`𝒙_i`$ denote vectors in $`^3`$ and $`𝕆`$, $`𝜸_{ij}`$, $`𝜸_i`$, and $`𝚪_i`$ denote $`3\times 3`$ matrices. This system is evidently symmetric, sparsely populated and weakly diagonally dominated, since $`\gamma _i^{\alpha \alpha }+\mathrm{\Gamma }_i^{\alpha \beta }>0i,\alpha `$. In addition, all friction coefficients are positive. Gershgorin’s circle theorem then suffices to guarantee positive definiteness of the dampening matrix. The number of next neighbours in contact corresponds to the number of off-diagonal blocks in the dampening matrix, such that the system becomes extremely sparse for large matrices.
Such systems can for efficiency be supplemented with the weighted Delaunay triangulation of a set of spheres for adjacency detection schaller2004 . Since the dampening matrix is positive definite, the method of conjugate gradients shewchuk1994 is well suited to the problem. However, since realistic systems will contain much more than 5 cells, the matrices would not fit into main memory, if stored completely. Fortunately, the matrices are only sparsely populated and the method of conjugate gradients can efficiently be combined with a row-indexed sparse storage scheme press1994 to compute a solution $`\dot{x}_i^\alpha `$. Note that the solution of the full system is an improvement over existing models: For example, in dallon2004 the tangential projector $`𝒫_{ij}^{\alpha \beta }`$ had for simplicity been approximated with the identity operator and in schaller2005a , the system was assumed to be diagonal.
The reaction-diffusion equation for the molecules
$`{\displaystyle \frac{u}{t}}=\left[D(𝒓;t)u\right]+Q(𝒓;t)`$ (50)
could in principle be solved using the method of Green functions along the lines of newman2004 . However, since spatiotemporally varying diffusion coefficients as well as nontrivial boundary conditions are considered here, the implementation of this method would require an enormous amount of work. Therefore, we have discretized above equation using the discrete element method on a cubic lattice. The arising coupled system of ordinary differential equations was then solved using the Crank-Nicolson scheme gershenfeld2000 and the algorithm of biconjugate gradients press1994 . Note that due to the strongly varying diffusion coefficients in cornified and non-cornified tissue (see table 1), the steady-state approximation is not applicable. The lattice constants for the rectangular reaction-diffusion grid discretisation have been chosen larger than the cellular diameters to ensure for validity of the discretisation approximation schaller2005a . Note that the reaction diffusion equation for the nutrients is made positive definite by cells entering necrosis below critical nutrient concentrations – necrotic cells do not consume nutrients. While the timestep of the simulation is determined by a maximum spatial cellular stepsize (fixed at $`0.5\mu \mathrm{m}`$), the timestep of the reaction-diffusion grids has been divided into several substeps such that the Courant factor press1994 is smaller than 1 in order to increase numerical accuracy. To connect the discrete reaction-diffusion grid with the spatially-continuous cellular positions, tri-linear interpolation has been used. In addition, the concentration has been fixed to vanish at grid nodes whose elementary cell did not contain cells. At grid nodes in the vicinity of cells the diffusive properties have been smoothly adapted in the range of the values given in table 1. |
warning/0507/cond-mat0507240.html | ar5iv | text | # Exact diagrammatic approach for dimer-dimer scattering and bound states of three and four resonantly interacting particles.
## I Introduction
Following the experimental realization of the Bose-Einstein condensation in ultracold bosonic gases, together with its intensive study, the physics of ultracold Fermi gases has taken off recently with a strong development of experimental and theoretical investigations within the last few years Lev (2004). In particular, much advantage has been taken of various Feshbach resonances which offer the possibility observing experimentally the so called BEC-BCS crossover. This has been done in particular in <sup>6</sup>Li and in <sup>40</sup>K. In the weak coupling limit of small negative scattering length, which is realized far away on one side of the resonance, the corresponding weak attractive interaction between fermions leads to a BCS type condensate of Cooper pairs. On the other side of the resonance, where the scattering length is positive, weakly bound dimers, or molecules, consisting of two different fermions are formed. When one goes far enough of the resonance on this positive side, one obtains a weakly interacting gas of these dimers, which may in particular form a Bose-Einstein condensate, as it has been recently observed experimentally Greiner et al. (2003); Jochim et al. (2003); Zwierlein et al. (2003); Bourdel et al. (2004).
In the present paper, motivated by the problem raised by the physics of this dilute gas of composite bosons, we will deal with the dimer-dimer elastic scattering and present an exact diagrammatic approach to its solution. This will be done by staying in the so-called resonance approximation which is quite suited to the physical situation found with a Fesbach resonance. It this case the (positive) scattering length greatly exceeds the characteristic radius $`r_0`$ for the attractive interaction between fermionic atoms. A problem of this kind was first investigated by Skorniakov and Ter-Martirosian Skorniakov and Ter-Martirosian (1956) in the case of the 3-body fermionic problem. They showed that the scattering length of a fermion on a weakly bound dimer is determined by a single parameter, namely the two-body scattering length $`a_F`$ between fermions, and it is equal to $`1.18a_F`$ in the zero-range limit for the interatomic potential. A similar situation is found in the case of four fermions, where the dimer-dimer scattering length is fully determined by this same scattering length $`a_F`$.
In a study of the crossover problem Haussmann Haussmann (1993) calculated this scattering length of composite bosons $`a_B`$ at the level of the Born approximation and found it equal to $`2a_F`$. This result was later on much improved by Pieri and Strinati Pieri and Strinati (2000), who took into account the repeated scattering of these composite bosons in the ladder approximation. This diagrammatic approach led them to a scattering length approximately equal to $`a_B0.75a_F`$. However, this ladder approximation is not exact, because it misses an infinite number of other diagrams which in principle lead to a contribution of the same order of magnitude as those taken into account. Very recently this problem has been solved exactly by Petrov, Salomon, and Shlyapnikov Petrov et al. (2004, 2005) who found for the scattering length of these composite bosons $`a_B=0.6a_F`$. This has been achieved by solving directly the Schrödinger equation for four fermions, using the well-known method of pseudopotentials. Here we will give an exact solution of this scattering problem of two weakly bound dimers, using a diagrammatic approach in the resonance approximation, which can be seen as a bridge between the approach of Pieri and Strinati Pieri and Strinati (2000) and the exact result of Petrov, Salomon, and Shlyapnikov Petrov et al. (2004, 2005).
In order to show the strength and the versatility of our approach, we make use of it to obtain new results for various systems, in the two-dimensional (2D) case which is of interest not only for cold gases, but also for high $`T_c`$ superconductivity. Specifically we consider first a system of resonantly interacting bosons. We calculate exactly the three bosons $`bbb`$ and four bosons $`bbbb`$ bound state energies in this case. We also make use of our approach for the study of 2D bosons interacting resonantly with fermions. In this case we calculate exactly the bound state energies of two bosons plus one fermion $`bbf`$, two bosons plus two fermions $`bf_{}bf_{}`$, and three bosons plus one fermion $`bbbf`$. In this respect the present paper is in the line of previous results obtained by of some of us. Indeed the possibility of two fermions Kagan and Rice (1994); Kagan et al. (1998) $`ff`$ and two bosons Kagan and Efremov (2002) $`bb`$ pairing was predicted, as well as the creation Kagan et al. (2004) of a composite fermion $`bf`$ in resonantly interacting $`(ar_0)`$ 2D Fermi-Bose mixtures.
## II Three particles scattering
As a preliminary exercise we will rederive the result of Skorniakov and Ter-Martirosian for the dimer-fermion scattering length $`a_3`$ using the diagrammatic method Bedaque and van Kolck (1998). Following Skorniakov and Ter-Martirosian, in the presence of the weakly bound resonance level $`E_b`$ (with $`E_b>0`$), we can limit ourselves to the zero-range interaction potential between fermions in the scattering of these two particles . The two-fermion vertex can be approximated by a simple one-pole structure, which reflects the presence of the s-wave resonance level in the spin-singlet state, and is essentially given by the scattering amplitude, namely:
$`T_{2\alpha \beta ;\gamma \delta }(P)=T_2(P)\times (\delta _{\alpha ,\gamma }\delta _{\beta ,\delta }\delta _{\alpha ,\delta }\delta _{\beta ,\gamma }).(\delta _{\alpha ,}\delta _{\beta ,}+\delta _{\alpha ,}\delta _{\beta ,})=T_2(P)\chi (\alpha ,\beta )\chi (\gamma ,\delta ),`$ (1)
$`T_2(P)={\displaystyle \frac{4\pi }{m^{3/2}}}{\displaystyle \frac{\sqrt{E_b}+\sqrt{𝐏^2/4mE}}{E𝐏^2/4m+E_b}},\chi (\alpha ,\beta )=\delta _{\alpha ,}\delta _{\beta ,}\delta _{\alpha ,}\delta _{\beta ,},`$ (2)
where $`P=\{\text{P},E\}`$, $`E`$ is the total frequency and P is the total momentum of incoming particles, $`m`$ is the fermionic mass, $`E_b=1/ma_F^2`$. Indices $`\alpha ,\beta `$ and $`\gamma ,\delta `$ denote the spin states of incoming and outgoing particles. The function $`\chi (\alpha ,\beta )`$ stands for the spin singlet state. We will draw this vertex in the way, shown on Fig. 1, where the double line can be regarded as a propagating dimer.
The simplest process that contributes to dimer-fermion interaction is the exchange of a fermion. We denote the corresponding vertex as $`\mathrm{\Delta }_3`$ and it is described by the diagram on Fig. 2. Its analytical expression reads
$$\mathrm{\Delta }_{3\alpha ,\beta }(p_1,p_2;P)=\delta _{\alpha ,\beta }G(Pp_1p_2),$$
(3)
where $`G(p)=1/\left(\omega 𝐩^2/2m+i0_+\right)`$ is the bare fermion Green’s function. The minus sign in the right hand side of Eq.(3) comes from the permutation of the two fermions. In order to obtain the full dimer-fermion scattering vertex $`T_3`$ we need to sum up all possible diagrams with indefinite number of $`\mathrm{\Delta }_3`$ blocks. In the present case these diagrams have a ladder structure. It is obvious that the spin projection is conserved in every order in $`\mathrm{\Delta }_3`$ and thus $`T_{3\alpha ,\beta }=\delta _{\alpha ,\beta }T_3`$. An equation for $`T_3`$ will have the diagrammatic representation shown in Fig. 3. It is obtained by writing that either the simplest exchange process occurs alone, or it is followed by any other process. In analytical form it reads
$$T_3(p_1,p_2;P)=G(Pp_1p_2)\underset{q}{}G(Pp_1q)G(q)T_2(Pq)T_3(q,p_2;P),$$
(4)
where $`\underset{q}{}id^3q𝑑\mathrm{\Omega }/(2\pi )^4`$.
We can integrate out the frequency $`\mathrm{\Omega }`$ in Eq.(4) by closing the integration contour in the lower half-plane, since both $`T_2(Pq)`$ and $`T_3(q,p_2;P)`$ are analytical functions of $`\mathrm{\Omega }`$ in this region (this property for $`T_3(p_1,p_2;P)`$ results from Eq.(4) itself). Hence only the ”on the shell” value $`T_3(\{𝐪,q^2/2m\},p_2;P)`$ comes in the right- hand side of Eq.(4). Moreover, if we are interested in the low-energy s-wave dimer-fermion scattering length $`a_3`$, we have to put $`P=\{𝐏,E\}=\{\mathrm{𝟎},E_b\}`$ and $`p_2=0`$. Hence Eq.(4) reduces to an equation for the ”on the shell” value of $`T_3(p_1,p_2;P)`$. Taking into account the standard relation between $`T`$-matrix and scattering amplitude (with reduced mass) and the fact that, from Eq.(1), $`T_2`$ has an additional factor $`8\pi /(m^2a_F)`$ compared to a standard boson propagator, we find that the full vertex $`T_3`$ is connected with $`a_3`$ by the following relation:
$$\left(\frac{8\pi }{m^2a_F}\right)T_3(0,0;\{\mathrm{𝟎},E_b\})=\frac{3\pi }{m}a_3.$$
(5)
This leads to introduce a new function $`a_3(𝐤)`$ defined by
$$a_3(𝐤)=\frac{4}{3m}\left(\sqrt{mE_b}+\sqrt{3k^2/4+mE_b}\right)T_3(\{𝐤,k^2/2m\},0;\{\mathrm{𝟎},E_b\}).$$
(6)
and substituting it in Eq.(4), we obtain Skorniakov - Ter-Martirosian equation for the scattering amplitude:
$$\frac{(3/4)a_3(𝐤)}{\sqrt{mE_b}+\sqrt{3k^2/4+mE_b}}=\frac{1}{k^2+mE_b}4\pi \frac{d𝐪}{(2\pi )^3}\frac{a_3(𝐪)}{q^2(k^2+q^2+𝐤.𝐪+mE_b)}.$$
(7)
Solving this equation one obtains the well known result Skorniakov and Ter-Martirosian (1956) for the dimer-fermion scattering length $`a_3=a_3(0)=1.18a_F`$.
## III Dimer - dimer scattering
By now we can proceed to the problem of the dimer-dimer scattering. This problem was previously solved by Petrov et al. Petrov et al. (2004, 2005) via studying Schrödinger equation for a 4-fermions wave function. Our diagrammatic approach is conceptually close to Petrov’s one. Its basic point is that it requires the introduction of a special vertex which describes an interaction of one dimer as a single object with the two fermions constituting the other dimer.
Let us investigate all the possible types of diagrams that contribute to the dimer-dimer scattering vertex $`T_4`$. In this process both dimers are temporarily ”broken” in their fermionic components, which means that the fermions of one dimer exchange and/or interact with the fermions of the other dimer. The simplest process is an exchange of fermions by two dimers shown on Fig. 4a. More complicated diagrams are composed by introducing intermediate interactions between exchanging fermions (see Fig. 4b,c). As long as one of the fermions does not interact or exchange with the other ones, all these complications can be summed up in the $`T_3`$ block (see Fig. 4d) which describes, as we have seen in the preceding section, the scattering of a fermion on a dimer. Furthermore we may exchange bachelor fermions participating in the $`T_3`$ scattering. The resulting series has the diagrammatic structure shown on Fig. 4e. This series describes a ”bare” interaction between dimers. The last obvious step is to compose ladder type diagrams from this ”bare” interaction. A typical ladder diagram is shown on Fig. 4f. These general ladder diagrams describe all possible processes which contribute to the dimer-dimer scattering.
The fact that the $`T_4`$ vertex should be expressed in terms of $`T_3`$ was first noticed by Weinberg in his work on multiparticle scattering problems Weinberg (1964). Note that a calculation of the diagrams shown on Fig. 4e, f requires information about an off-shell matrix $`T_3`$, that is about a matrix with arbitrary relation between frequencies and momenta of incoming and outgoing particles. On the other hand, for the calculation of the dimer-fermion scattering length $`a_3`$ in Eq.(7), only the simpler on-shell structure of $`T_3`$ is required as we have seen in the preceding section. Luckily, as we will see now, we can exclude $`T_3`$ from our considerations and express $`T_4`$ only in terms of $`T_2`$. By doing this we reduce the number of integral equations required for the calculation of the dimer-dimer scattering length $`a_4`$.
Since, as we have just seen, it is impossible to construct a closed equation for the dimer-dimer scattering vertex $`T_4`$, we wish to find an alternative way for taking into account in one equation all the diagrams contributing to dimer-dimer scattering. Inspired by the work of Petrov et al. Petrov et al. (2004, 2005) and looking at the diagrams we have considered above, we are naturally lead to look for a special vertex that describes the interaction of two fermions, constituting the first dimer, with the second dimer taken as a single object. This vertex would be the sum of all diagrams with two fermions and one dimer as incoming lines. It would be natural to suppose that these diagrams should have the same set of outgoing – two fermionic and one dimer – lines. However in this case there will be a whole set of disconnected diagrams contributing to our sum that describe interaction of a dimer with only one fermion. As it was pointed out by Weinberg Weinberg (1964), one can construct a good integral equation of Lippmann-Schwinger type only for connected class of diagrams. Thus we are forced to pay our attention to the vertex $`\mathrm{\Phi }_{\alpha \beta }(q_1,q_2;p_2,P)`$ corresponding to the sum of all diagrams with one incoming dimer, two incoming fermionic lines and two outgoing dimer lines (see Fig. 5). This is also quite natural from our view point since, in our scattering problem we are interested in a final state with two outcoming dimers. Indeed once this vertex $`\mathrm{\Phi }_{\alpha \beta }(q_1,q_2;p_2,P)`$ is known, it is straightforward to calculate the dimer-dimer scattering vertex $`T_4(p_1,p_2;P)`$ which is given by:
$$T_4(p_1,p_2;P)=\frac{1}{2}\underset{k;\alpha ,\beta }{}\chi (\alpha ,\beta )G(P+p_1k)G(k)\mathrm{\Phi }_{\alpha \beta }(P+p_1k,k;p_2,P).$$
(8)
The corresponding diagrammatic representation is given in Fig. 5. One can readily verify that, in any order of interaction, $`\mathrm{\Phi }`$ contains only connected diagrams.
The spin part of the vertex $`\mathrm{\Phi }_{\alpha ,\beta }`$ has the simple form $`\mathrm{\Phi }_{\alpha ,\beta }(q_1,q_2;P,p_2)=\chi (\alpha ,\beta )\mathrm{\Phi }(q_1,q_2;P,p_2)`$. The diagrammatic representation of the equation for $`\mathrm{\Phi }`$ is given in Fig. 6. One can assign some ”physical meaning” to the processes described by these diagrams. The diagram of Fig. 6a represents the simplest exchange process in a dimer-dimer interaction. The diagram of Fig. 6b accounts for a more complicated nature of a ”bare” dimer-dimer interaction. Finally the diagram of Fig. 6c allows for a multiple dimer-dimer scattering via a ”bare” interaction (it generates ladder-type diagrams analogous to those of Fig. 4f). The last term in Fig. 6 means that we should add another set of three diagrams analogous to those of Fig. 6a, b, c but with the two incoming fermions ($`q_1`$ and $`q_2`$) exchanged. The diagrammatic representation translates into the following analytical equation for the vertex $`\mathrm{\Phi }`$:
$$\begin{array}{c}\mathrm{\Phi }(q_1,q_2;p_2,P)=G(Pq_1+p_2)G(Pq_2p_2)\underset{k}{}G(k)G(2Pq_1q_2k)T_2(2Pq_1k)\mathrm{\Phi }(q_1,k;p_2,P)\hfill \\ \hfill \frac{1}{2}\underset{Q,k}{}G(Qq_1)G(2PQq_2)T_2(2PQ)T_2(Q)G(k)G(Qk)\mathrm{\Phi }(k,Qk;p_2,P)+(q_1q_2).\end{array}$$
(9)
Finally let us also indicate that it is possible to rederive the same set of equations, purely algebraically, by taking a complementary point of view. Instead of focusing, as we have done, on the free fermions lines as soon as a dimer is ”broken”, we can rather keep track of the fermions which make up a dimer. This leads again automatically to introduce the vertex $`\mathrm{\Phi }(q_1,q_2;p_2,P)`$. Then Eq.(9) is recovered when one keeps in mind that, after breaking dimers, one may have propagation of a single dimer and two free fermions before another break (this corresponds to the second term in the right hand side of Eq.(9)). Alternatively one may also have the propagation of two dimers, which leads to the third term in Eq.(9).
Coming back now more specifically to our problem, we can put $`p_2=0`$ and $`P=\{\mathrm{𝟎},E_b\}`$ since we are looking for an s-wave scattering length. At this point we have a single closed equation for the vertex $`\mathrm{\Phi }`$ in momentum representation, which we believe is analogous to Petrov et al equation in coordinate representation. To make this analogy more prominent we have to exclude frequencies from the equation by integrating them out. However this exclusion requires some more technical mathematics and we leave it out for Appendix A.
The dimer-dimer scattering length is directly related to the full symmetrized vertex $`T_4(p_1,p_2;P)`$. Just as in the preceding section, taking also statistics into account, we have:
$$\left(\frac{8\pi }{m^2a_F}\right)^2T_4(0,0;\{\mathrm{𝟎}E_b\},0)=\frac{2\pi (2a_B)}{m}.$$
(10)
If one skips the second term in Eq. (9), i.e. one omits diagram Fig.6b, one will arrive at the ladder approximation of Pieri and Strinati Pieri and Strinati (2000). The exact equation (9) corresponds to the summation of all diagrams. We have calculated the scattering length in the ladder approximation and the scattering length derived from the exact equation and obtained $`0.78a_F`$ and $`0.60a_F`$ respectively. Some details on our actual procedure are given in the next section. Thus our results in the ladder approximation are in agreement with the results Pieri and Strinati (2000) of Pieri and Strinati and, in the general form, with the results of Petrov et. al Petrov et al. (2004, 2005). Note also that our approach allows one to find the dimer-dimer scattering length in the 2D case (this problem was previously solved by Petrov et. al Petrov et al. (2003)).
Finally we would like to mention that our results allow one to find a fermionic Green’s function, chemical potential and sound velocity as a function of $`a_F`$ in the case of dilute superfluid bose gas of dimers at low temperatures. The problem of dilute superfluid bose gas of di-fermionic molecules was solved by Popov Popov (1966), and later deeply investigated by Keldysh and Kozlov Keldysh and Kozlov (1968). Those authors managed to reduce the gas problem to a dimer-dimer scattering problem in vacuum, but were unable to express the dimer-dimer scattering amplitude in a single two-fermion parameter. A direct combination of our results with those ones of Popov, Keldysh and Kozlov allows one to get all the thermodynamical values of a dilute superfluid resonance gas of composite bosons. Another interesting subject for the application of our results will be a high-temperature expansion for the thermodynamical potential and sound velocity in the temperature region $`TT_{}E_b`$, where the composite bosons begin to appear.
## IV Practical implementation
Let us give now some details on the way in which we have solved effectively the above equations. Actually we have dealed with two problems, the scattering length calculation discussed above and the bound states problem to be discussed below. Our two problems are quite closely related since, for the scattering length problem, we look for the scattering amplitude at zero outgoing wavevectors and energy for two dimers, while for the bound states we look for divergences of this same scattering amplitude at negative energy. As already indicated, in both cases the situation is somewhat simplified with respect to the variables we have to consider, due to the specific problem we handle. First with respect to $`P=\{𝐏,E\}`$, we have $`𝐏=\mathrm{𝟎}`$ since we work naturally in the rest frame of the four particles. Moreover, with respect to the total energy, $`E=ϵE_b`$ is negative. Specifically $`ϵ=1`$ when we look for the scattering length. Or when we consider bound states $`ϵ`$ gives the energy of the bound states we are looking for. Next, with respect to parameter $`p_2\{𝐩_2,\overline{p}_2\}`$ which characterizes the outgoing dimers, we will have naturally $`𝐩_2=\mathrm{𝟎}`$ as we have said since we consider zero outgoing wavevectors. Since we will evaluate $`\overline{p}_2`$ on the shell, we have merely $`\overline{p}_2=0`$, and this parameter drops out. Hence in the following we do not write anymore explicitely the value of parameter $`P`$. Both for the scattering length problem and the bound states problem, we have followed two main routes.
In our first route, we have written a specific integral equation for $`T_4(p_1,p_2)`$, which is then solved numerically. The details of our derivation for this integral equation are given in Appendix B. The kernel for this equation is itself obtained from a vertex $`\mathrm{\Gamma }`$. The defining integral equation Eq.(35) for this vertex has been inverted numerically, by calculating the inverse matrix, to obtain the vertex $`\mathrm{\Gamma }(q_1,q_2;p_2)`$. We have used Press et al. (1996) LU factorization and Gauss quadrature. The result has then been substituted in Eq.(34) which gives the kernel $`\mathrm{\Delta }_4(p_1,p_2)`$ coming in the integral equation Eq.(36). The solution of this last equation is naturally also handled numerically, for example by finding the eigenvalues of the kernel for the bound states problem.
In our second route we have kept both functions $`T_4`$ and $`\mathrm{\Phi }`$. In the following we do not write anymore the parameter $`p_2`$ which takes always the trivial value $`p_2=0`$, as explained above. Hence we are left with $`T_4(p_1)`$ which, because of rotational invariance, depends only on the energy $`\overline{p}_1`$ and the modulus $`|𝐩_1|`$ of the momentum. For brevity we denote this quantity $`t_4(|𝐩_1|,\overline{p}_1)`$. On the other hand it is shown in Appendix A that, in order to evaluate the second term in the right-hand side of Eq.(9), we need only the evaluation of $`\mathrm{\Phi }(q_1,q_2)`$ on the shell, which we denote as $`\varphi (𝐪_1,𝐪_2)`$. It depends only on the three variables $`|𝐪_1|`$, $`|𝐪_2|`$ and the angle between these two vectors. Hence it is enough to write Eq.(9) only for $`q_1`$ and $`q_2`$ taking on the shell values. From Eq.(9) this leads for $`\varphi (𝐪_1,𝐪_2)`$ to the following more convenient equation:
$`\varphi (𝐪_1,𝐪_2)={\displaystyle \frac{1}{(|E|+𝐪_1^2/m)(|E|+𝐪_2^2/m)}}+{\displaystyle \frac{d^D𝐤}{(2\pi )^D}\frac{2mt_2(2|E|+[2𝐤^2+2𝐪_1^2+(𝐤+𝐪_1)^2]/4m)\varphi (𝐪_1,𝐤)}{4m|E|+𝐤^2+𝐪_1^2+𝐪_2^2+(𝐤+𝐪_1+𝐪_2)^2}}`$ (11)
$`{\displaystyle \frac{1}{2}}{\displaystyle \underset{Q}{}}{\displaystyle \frac{(2m)^2t_2(|E|+\overline{Q}+𝐐^2/4m)t_2(|E|\overline{Q}+𝐐^2/4m)t_4(|𝐐|,\overline{Q})}{(2m(|E|\overline{Q})+𝐪_1^2+(𝐐+𝐪_1)^2)(2m(|E|+\overline{Q})+𝐪_2^2+(𝐐+𝐪_2)^2)}}+(𝐪_1𝐪_2)`$
where the dimer propagator $`t_2(x)`$ depends on the space dimension $`D`$. For $`D=3`$ it is given from Eq.(1) by $`t_2(x)4\pi /[m^{3/2}(\sqrt{x}\sqrt{E_b)}]`$, while for $`D=2`$ according to Eq.(21) we have $`t_2(x)4\pi /[m\mathrm{ln}(x/E_b)]`$. In the third term the angular integration can be performed analytically, and one is left with double integrals for the last two terms, for the $`3D`$ as well as for the $`2D`$ case. It is actually quite convenient, in the last term, to deform the $`\overline{Q}`$ contour from $`]\mathrm{},\mathrm{}[`$ to $`]i\mathrm{},i\mathrm{}[`$ by rotating it by $`\pi /2`$. No singularity is met in this deformation, and one is left to deal only with real quantities.
The above equation has to be supplemented by a corresponding equation for $`t_4(|𝐪|,\overline{q})`$ obtained from the definition Eq.(8). The important point is that the additional integrations can be performed analytically, owing to the various invariances under rotations found in the resulting terms. We just give here as an intermediate step the structure of the resulting equation:
$`t_4(k,iz)=S(k,z)+{\displaystyle _0^{\mathrm{}}}dp_1{\displaystyle _0^{\mathrm{}}}dp_2{\displaystyle _0^{2\pi }}d\alpha I(k,z,p_1,p_2,\alpha )t_2(2|E|+[3𝐩_1^2+3𝐩_2^2+2𝐩_1.𝐩_2)^2]/4m)\varphi (𝐩_1,𝐩_2)`$ (12)
$`+{\displaystyle _0^{\mathrm{}}}𝑑K{\displaystyle _0^{\mathrm{}}}𝑑ZJ(k,z,K,Z)|t_2(|E|+iZ+K^2/4m)|^2t_4(K,iZ)`$
where $`\alpha `$ is the angle between $`𝐩_1`$ and $`𝐩_2`$. Here $`S(k,z)`$, $`I(k,z,p_1,p_2,\alpha )`$ and $`J(k,z,K,Z)`$ are analytically known functions of the variables (except that $`J`$ requires to perform numerically a simple integration to be obtained, see below). In this equation and in particular in its last term, we have already gone to the purely imaginary frequency variable for $`t_4`$. The resulting $`t_4(x,iz)`$ turns out to be real and even with respect to $`z`$.
To be fully specific let us now give the actual self-contained integral equations which we have solved. We restrict ourselves to the $`3D`$ case and to the *bfbf* case (implying $`\alpha =1`$), corresponding to the dimer scattering problem treated Petrov et al. (2004, 2005) by Petrov et al. The only generalization is that we keep $`E=ϵ|E_b|`$, instead of setting $`ϵ=1`$ as we should if we considered only the scattering length problem. For clarity we write the resulting equations with dimensionless quantities, where $`1/a`$ has been taken as unit wavevector, and $`|E_b|=1/ma^2`$ as energy unit. For simplicity we keep basically the same notations for the various variables. We just indicate by a bar over the function name that they are expressed in reduced units, with reduced variables (actually we write $`\overline{t}_4(k,z)`$ instead of $`\overline{t}_4(k,iz)`$, and there is a change of sign between $`\varphi (𝐪_1,𝐪_2)`$ and $`\overline{\varphi }(𝐩_1,𝐩_2)`$). Equations for other cases and dimensions are completely similar with only few changes in coefficients, signs (for the particle statistics), for the expression of $`\overline{t}_2(x)`$ and for the explicit functions coming from analytical angular integrations.
We obtain:
$$\begin{array}{c}\overline{\varphi }(𝐩_1,𝐩_2)=\frac{1}{(ϵ+p_1^2)(ϵ+p_2^2)}+\frac{1}{\pi }_0^{\mathrm{}}k^2dk_0^\pi \mathrm{sin}\theta d\theta \frac{\overline{\varphi }(𝐩_1,𝐤)\overline{t}_2(2ϵ+(3p_1^2+3k^2+2kp_1\mathrm{cos}\theta )/4)}{\sqrt{A_+A_{}}}+(𝐩_1𝐩_2)\hfill \\ \hfill +\frac{8}{\pi p_1p_2}_0^{\mathrm{}}𝑑z_0^{\mathrm{}}𝑑k\overline{t}_4(k,z)|\overline{t}_2(ϵ+k^2/4+iz)|^2I(B_1,B_2,\alpha )\end{array}$$
(13)
with $`A_\pm =2ϵ+p_1^2+p_2^2+k^2+p_1p_2\mathrm{cos}\alpha +kp_1\mathrm{cos}\theta +kp_2\mathrm{cos}(\alpha \pm \theta )`$, and $`\alpha `$ is the angle between $`𝐩_\mathrm{𝟏}`$ and $`𝐩_\mathrm{𝟐}`$, while $`\theta `$ is the polar angle of $`𝐤`$ with $`𝐩_\mathrm{𝟏}`$. We have simply set now $`\overline{t}_2(x)=[1\sqrt{x}]^1`$. Here we have also defined the function:
$$I(B_1,B_2,\alpha )=𝔢\frac{1}{2\sqrt{E}}\mathrm{ln}\frac{B_1B_2^{}+\mathrm{cos}\alpha +\sqrt{E}}{B_1B_2^{}+\mathrm{cos}\alpha \sqrt{E}}$$
(14)
$$E=B_1^2+B_2^2+2B_1B_2^{}\mathrm{cos}\alpha \mathrm{sin}^2\alpha $$
(15)
$$B(p,k,z)=\frac{1}{kp}[ϵ+p^2+\frac{k^2}{2}iz]$$
(16)
$$B_iB(p_i,k,z)$$
(17)
The corresponding equation for $`\overline{t}_4(k,z)`$ is:
$$\begin{array}{c}\overline{t}_4(k,z)=\frac{1}{4\pi kz}\mathrm{ln}\frac{1+\mathrm{cos}\gamma +2\sqrt{\mathrm{cos}\gamma }\mathrm{cos}(\phi \gamma /2)}{1+\mathrm{cos}\gamma +2\sqrt{\mathrm{cos}\gamma }\mathrm{cos}(\phi +\gamma /2)}\hfill \\ \hfill \frac{1}{\pi ^3k^2}_0^{\mathrm{}}p_1𝑑p_1_0^{\mathrm{}}p_2𝑑p_2_0^\pi \mathrm{sin}\alpha d\alpha \overline{\varphi }(𝐩_1,𝐩_2)\overline{t}_2(2ϵ+(3p_1^2+3p_2^2+2p_1p_2\mathrm{cos}\alpha )/4)I(B_1,B_2,\alpha )\\ \hfill \frac{1}{2\pi ^3k}_0^{\mathrm{}}K𝑑K_0^{\mathrm{}}𝑑Z\overline{t}_4(K,Z)|\overline{t}_2(ϵ+K^2/4+iZ)|^2\overline{J}(k,z,K,Z)\end{array}$$
(18)
with $`\phi =\mathrm{arctan}(k/2)`$ and $`\gamma =\mathrm{arctan}[4z/(4+k^2)]`$ and we have defined the function:
$$\overline{J}(k,z,K,Z)=_0^{\mathrm{}}𝑑x\frac{1}{ϵ+x^2+\frac{k^2+K^2}{4}}\mathrm{ln}\frac{C(x,k,K,Z)}{C(x,k,K,Z)}\mathrm{ln}\frac{C(x,K,k,z)}{C(x,K,k,z)}$$
(19)
$$C(x,k,K,Z)=[ϵ+(x+\frac{k}{2})^2+\frac{K^2}{4}]^2+Z^2$$
(20)
It is seen on these integral equations for our two unknown functions $`\overline{t}_4(x,z)`$ and $`\overline{\varphi }(𝐩_1,𝐩_2)`$ that they require only at most a triple integrals to be performed numerically. In this sense they are not numerically more complicated than the work involved in solving directly for the corresponding Schrödinger equation, as it has been done Petrov et al. (2004, 2005) by Petrov et al. Indeed these integrals require only a few appropriate change of variables to take care of singular behaviours occuring on some boundaries. Otherwise they have been performed with unsophisticated integration routine.
In the case of the scattering length a mere iteration algorithm has been found to lead rapidly to the solution (provided an appropriate exact algebraic manipulation is made to make the iteration convergent). In this way we have been able to handle $`45\times 45\times 45`$ matrices (for the three variables entering $`\overline{\varphi }(𝐩_1,𝐩_2))`$. This size is large enough to allow improved precision by extrapolation to infinite size, although we have not done it in the present case, but rather for the ground state of the $`bbbb`$ complex discussed below. This leads to the result $`a_B=0.60a_F`$ in full agreement with Petrov Petrov et al. (2004, 2005) et al, within a quite reasonable computing time on (nowadays) unsophisticated computer. We have not tried to improve on the accuracy of the result, since there is no basic interest. In the case of the bound states, to be described below, we have proceeded to a straight diagonalization of the matrix equivalent to the right hand sides of Eq.(13) and Eq.(18) with the Lapack library algorithm. In the 2D case, it is worth noticing that, because of the logarithmic dependence of $`\overline{t}_2(x)`$ on $`x`$, it is quite an improvement to make the change of variables $`K=ϵ^{1/2}K^{}`$ and $`Z=ϵZ^{}`$, and so on, since the more appropriate variable turns out to be $`\mathrm{ln}ϵ`$ rather than $`ϵ`$ itself.
## V New results in a 2D case
We will now apply the diagrammatic approach developed in the previous sections (see also Appendix A) to get new results for the systems of resonantly interacting particles in a 2D case.
As it was first shown by Danilov Danilov (1961) (see also a paper by Minlos and Fadeev Minlos and Fadeev (1961)) in the 3D case, the problem of three resonantly interacting bosons could not be solved in the resonance approximation. This statement stems from the fact that in the case of identical bosons the homogeneous part of Skorniakov-Ter-Martirosian equation (7) has a non-zero solution at any energies. The physical meaning of this mathematical feature was elucidated by Efimov, who showed that a two-particle interaction leads to the appearance of an attractive $`1/r^2`$ interaction in a three-body system. Since in the attractive $`1/r^2`$ potential a particle can fall into the center, the short range physics is important and one can not replace the exact pair interaction by its resonance approximation.
On the contrary in the case of the 2D problem the phenomena of the particle fall into the center is absent and one can utilize the resonant approximation Bruch and Tjon (1979); Jensen et al. (2004). Therefore it is possible to describe three- and four-particle processes in terms of the two-particle binding energy $`E_b=1/ma^2`$ only (below, for simplicity we will assume that all particles under consideration have the same mass $`m`$). We will leave aside the problem of composite particles scattering and will concentrate on the problem of binding energies of complexes of three and four particles.
As well as in the case of the 3D problem, the cornerstone in the diagrammatic technique is the two-particle resonance scattering vertex $`T_2`$ (see Fig.1). For two resonantly interacting particles with total mass $`2m`$ it reads in 2D:
$$T_2(P)=\frac{4\pi }{m}\frac{\alpha }{\mathrm{ln}\left(\{𝐏^2/4mE\}/|E_B|\right)},$$
(21)
where we introduce a factor $`\alpha =\{1,2\}`$ in order to take into account whether two particles are indistinguishable or not. That is $`\alpha =2`$ for the case of a resonance interaction between identical bosons, while $`\alpha =1`$ for the case of a resonance interaction between fermion and boson, or for the case of two distinguishable fermions.
### V.1 Three particles in 2D
We start with a system of three resonantly interacting identical bosons - $`bbb`$ \- in 2D. An equation for the dimer-boson scattering vertex $`T_3`$ which describes interaction of three bosons has the same diagrammatic form as the one shown on the Fig.3, however there are small changes in the rules for its analytical evaluation. The resulting equation reads:
$$T_3(p_1,p_2;P)=G(Pp_1p_2)+\underset{q}{}G(Pp_1q)G(q)T_2(Pq)T_3(q,p_2;P),$$
(22)
where we have now $`\underset{q}{}id^2q𝑑\mathrm{\Omega }/(2\pi )^3`$, $`P=\{\mathrm{𝟎},E\}`$, and one should put $`\alpha =2`$ for the two-particle vertex $`T_2`$ in Eq.(21). The opposite signs in Eq. (4) for fermions and Eq. (22) for bosons are due to the permutational properties of the involved particles : an exchange of fermions (see Fig.2) results in a minus sign, while an analogous exchange of bosons brings no extra minus. Finally, as we mentionned above, we note that three-particle s-wave (s-wave channel of a boson-dimer scattering) binding energies $`E_3`$ correspond to the poles of $`T_3(0,0;\{\mathrm{𝟎},|E_3|\})`$ and, consequently, at energies $`E=E_3`$ the homogeneous part of Eq. (22) has a non-zero solution. Solving Eq. (22) we find that a complex of three identical bosons has two s-wave bound states $`E_3=16.5E_b`$ and $`E_3=1.27E_b`$ in accordance with the previous results of Bruch and Tjon Bruch and Tjon (1979); Jensen et al. (2004).
Let us now consider a complex - $`fbb`$ \- consisting of one fermion and two bosons. As noted above we take bosons and fermions with equal masses $`m_b=m_f=m`$. We assume that a fermion-boson interaction $`U_{fb}`$, characterized by the length $`r_{fb}`$, yields a resonant two-body bound state with an energy $`E=E_b`$. In the same time a boson-boson interaction $`U_{bb}`$, characterized by the interaction length $`r_{bb}`$, does not yield a resonance. Hence if we are interested in the low-energy physics the only relevant interaction is $`U_{fb}`$ and we can ignore the boson-boson interaction $`U_{bb}`$, the latter would give small corrections of the order $`|E_B|mr_{bb}^21`$ at low energies. In order to determine three-particle bound states one has to find poles in the dimer-boson scattering vertex $`T_3`$. Since we neglect the boson-boson interaction $`U_{bb}`$ the vertex $`T_3`$ is described by the same diagrammatic equation of Fig. 3 as for the problems of three bosons. The analytical form of this equation also coincides with Eq. (22) with the minor difference that the resonance scattering vertex $`T_2`$ now corresponds to the interaction between a boson and a fermion, and therefore we should put $`\alpha =1`$ in Eq. (21) for $`T_2`$. Solving the equation for $`T_3`$ we find that $`fbb`$ complex has only one s-wave bound state with the energy $`E_3=2.39E_b`$. Note that a complex - $`bff`$ \- consisting of a boson and two spinless identical fermions with resonance interaction $`U_{fb}`$ does not have any three-particle bound states.
### V.2 Four particles in 2D
After solving the above three-particle problems we may proceed to the complexes consisting of four particles. At first we will consider four identical resonantly interacting bosons $`bbbb`$ Platter et al. (2004). Any two bosons would form a stable dimer with binding energy $`E=E_b`$. We are going to find a four-particle binding energy as an energy of an s-wave bound state of two dimers. Generally speaking a bound state could emerge in channels with larger orbital moments, however this question will be a subject of further investigations. Just as in the preceding subsection, in order to find a binding energy we should examine the analytical structure of the dimer-dimer scattering vertex $`T_4`$ and find its poles. The set of equations for $`T_4`$ has the same diagrammatic structure as those shown on Fig. 5 and Fig. 6. The analytical expression for the first equation reads:
$$T_4(p_1,p_2;P)=\frac{1}{\alpha }\underset{k}{}G(P+p_1k)G(k)\mathrm{\Phi }(P+p_1k,k;p_2,P),$$
(23)
and the equation for the vertex $`\mathrm{\Phi }`$ is:
$$\begin{array}{c}\mathrm{\Phi }(q_1,q_2;p_2,P)=G(Pq_1+p_2)G(Pq_2p_2)+\underset{k}{}G(k)G(2Pq_1q_2k)T_2(2Pq_1k)\mathrm{\Phi }(q_1,k;p_2,P)\hfill \\ \hfill +\frac{1}{2\alpha }\underset{Q,k}{}G(Qq_1)G(2PQq_2)T_2(2PQ)T_2(Q)G(k)G(Qk)\mathrm{\Phi }(k,Qk;p_2,P)+(q_1q_2).\end{array}$$
(24)
where $`T_2`$ should be taken from Eq.(21) and one should put $`\alpha =2`$ for the case of identical resonantly interacting bosons. When we look for the poles of $`T_4`$ as a function of the variable $`E`$, with $`P=\{\mathrm{𝟎},E\}`$, we have naturally to consider only the homogeneous part of this equation. We have found 2 bound states for the $`bbbb`$ complex. The values of the total binding energy $`|E_4|=2|E|`$ are given in Table 1 below. Certainly for the validity of our approximation we should have $`|E_4|1/mr_0^2`$. For the case of four bosons $`bbbb`$ it means that $`197E_b1/mr_0^2`$ and hence $`a/r_0\sqrt{197}`$. This case can still be considered as quite realistic for the Feshbach resonance situation.
The case of a four-particle complex - $`bf_{}bf_{}`$ \- consisting of resonantly interacting bosons and fermions is still described by the same equations (23,24) but with parameter $`\alpha =1`$. In this case we found 2 bound states and they are also listed in Table 1.
In order to obtain bound states of the $`fbbb`$ complex one has to find energies $`P=\{\mathrm{𝟎},E\}`$ corresponding to nontrivial solutions of the following homogeneous equation
$$\mathrm{\Phi }(q_1,q_2;p_2,P)=\underset{k}{}G(k)G(2Pq_1q_2k)T_2(2Pq_1k)\mathrm{\Phi }(q_1,k;p_2,P)+(q_1q_2).$$
(25)
This equation corresponds to the diagram of Fig. 6b. We have found a single bound state for this $`fbbb`$ complex. Finally we summarize the results concerning binding energies of three and four resonantly interacting particles in 2D in Table 1.
For the $`bbbb`$ complex we find the beginning of a continuum of states at $`|E_4|/E_b=16.5`$, as it should be since this is, within our numerical precision, the binding energy of $`bbb`$. Similarly we find the beginning of a continuum at $`|E_4|/E_b=2.4`$ for the $`fbbb`$ and the $`bf_{}bf_{}`$ complex, in agreement with the binding energy of $`fbb`$. We display our corresponding results in Fig. 7 and Fig. 8. In all our calculations we find numerically, as a function of $`|E_4|`$, the eigenvalues $`\lambda `$ corresponding to the matrix on the right-hand side of our equations, for example Eq. (25). When one of these eigenvalues is equal to 1, this means that the corresponding $`E_4`$ is the energy of an eigenstate of our complex. In Fig. 7, we display the first highest eigenvalues for $`|E_4|=2.4`$, both for the $`bf_{}bf_{}`$ case and the $`bbbf`$ case. One sees clearly that a fair number of eigenvalues are essentially equal to 1. One could tune them exactly to 1 by changing very slightly $`|E_4|`$. Hence this corresponds to the beginning of the continuum. By contrast one sees also clearly two isolated eigenvalues larger than 1, for the $`bf_{}bf_{}`$ case, and one eigenvalue larger than 1 for the $`bbbf`$ case. One can bring them to $`\lambda =1`$ by increasing $`|E_4|`$, and therefore they correspond to the bound states that we have found. Similarly we display in Fig. 8 the eigenvalues for the $`bbbb`$ case, for the value $`|E_4|=16.5`$ corresponding essentially to the threshold for the continuum. Here again one sees many eigenvalues quite close to 1. On the same figure we also show the results of the same calculations for $`|E_4|=22.`$ in order to display the way in which this whole spectrum evolves with $`|E_4|`$. In particular one sees clearly the two isolated eigenvalues, corresponding to the two bound states found in this case. In particular since one of them is equal to 1, this means that the binding energy of one of the bound states is equal to $`22E_b`$, within our numerical precision.
Note finally that all our calculations correspond to the case of particles with equal masses $`m_f=m_b=m`$, although they can be quite easily generalized to the case of different masses.
## VI Conclusions and discussion
For the problem of resonantly interacting fermions in 3D we have developed an exact diagrammatic approach that allows to find the dimer-dimer scattering length $`a_B=0.60a_F`$ in exact agreement with known results. This exact diagrammatic solution of the dimer-dimer scattering length problem in 3D opens new horizons for the extension of the self-consistent mean-field schemes of Leggett and Nozières-Schmitt-Rink to the inclusion of the quite essential three and four-particle physics in the two-particle variational wave-functions of the BCS-type. This in turn will help us to get diagrammatically exact results for $`T_c`$, pseudogap and sound velocity in the dilute BEC-limit and to develop a more sophisticated interpolation scheme for these quantities toward the unitarity limit. The work on this very exciting project is now in progress.
We have applied the developed approach to get new results in the 2D case. Namely, we have calculated exactly the binding energies of the following complexes: three bosons $`bbb`$, two bosons plus one fermion $`bbf`$, three bosons plus one fermion $`bbbf`$, two bosons plus two fermions $`bf_{}bf_{}`$, and four bosons $`bbbb`$.
Our investigations enrich the phase-diagram for ultracold Fermi-Bose gases with resonant interaction. They serve as an important step for future calculations of the thermodynamical properties and the spectrum of collective excitations in different temperature and density regimes, in particular in the superfluid domain. Note that in purely bosonic models in 2D or in the Fermi-Bose mixtures in the case of prevailing density of bosons $`n_B>n_F`$ a creation of larger complexes consisting of 5, 6 and so on particles is also possible. In fact here we are dealing with the macroscopic phase separation (with the creation of large droplets). The radius of this droplet $`R_N`$ for $`N`$ bosons in 2D is estimated in Hammer and Son (2004) on the basis of a variational approach. Note that already for $`N=5`$ the exact calculation of the bound state requires huge computational capability, but it would be interesting to see precisely how this would appear with our approach.
## Appendix A Dimer-dimer scattering equation. Frequency integration
In this Appendix we will show how one can integrate explicitely over the frequency dependence in the dimer-dimer scattering equation (9) (we consider only this case, the other ones considered in Section IV would require trivial modifications). To simplify further computations we slightly change the notation and introduce a chemical potential $`\mu =E_b/2`$ and the single fermion energy $`\xi _𝐩=𝐩^2/2m\mu =𝐩^2/2m+E_b/2`$, with the modified fermion Green’s function $`𝒢(p)=1/(\omega \xi _𝐩)`$. In the expression Eq.(1) for $`𝒯_2(Q)`$ we have similarly to replace $`E`$ by $`EE_b`$.
The integral equation (9) reads more explicitely (with $`k=\{\text{k},\omega \}`$ and $`Q=\{\text{Q},\mathrm{\Omega }\}`$):
$$\begin{array}{c}\Phi (q_1,q_2)=𝒢(q_1)𝒢(q_2)i\underset{\mathrm{}}{\overset{\mathrm{}}{}}\frac{d\omega }{2\pi }\frac{d^3𝐤}{(2\pi )^3}𝒢(k)𝒢(q_1q_2k)𝒯_2(q_1k)\Phi (q_1,k)+\hfill \\ \hfill +\frac{1}{2}\frac{d^4Q}{(2\pi )^4}\frac{d^4k}{(2\pi )^4}𝒢(Qq_1)𝒢(Qq_2)𝒯_2(Q)𝒯_2(Q)𝒢(k)𝒢(Qk)\Phi (k,Qk)+(q_1q_2).\end{array}$$
(26)
From this equation $`\Phi (q_1,q_2)=\Phi (q_2,q_1)`$, as it is obvious physically. Note also that the third term is already explicitely symmetrical in $`q_1q_2`$.
First we note that, from Eq.(26) itself, $`\Phi (q_1,q_2)`$ is analytical with respect to the frequency variables $`\omega _1`$ and $`\omega _2`$ of the four-vectors $`q_1`$ and $`q_2`$ in the lower half-planes $`𝔪\omega _1<0`$ and $`𝔪\omega _2<0`$. This can be seen by assuming this property self-consistently in the right-hand side, and checking that the three terms are then indeed analytical, or equivalently one can proceed to a perturbative expansion. Then, if we make the ”on the shell” calculation of $`\Phi (q_1,q_2)`$ from Eq.(26), that is for $`\omega _1=\xi _{𝐪_1}`$ and $`\omega _2=\xi _{𝐪_2}`$, we see that, for second term in the right-hand side, the only singularity in the lower complex plane $`𝔪\omega <0`$ is the pole of $`𝒢(k)`$ at $`\omega =\xi _𝐤`$. Hence the integration contour can be closed in the lower half-plane, leading to:
$$i\underset{\mathrm{}}{\overset{\mathrm{}}{}}\frac{d\omega }{2\pi }𝒢(k)𝒢(q_1q_2k)𝒯_2(q_1k)\Phi (q_1,k)=\frac{𝒯_2(\xi _{𝐪_1}\xi _𝐤,𝐪_1+𝐤)}{\xi _{𝐪_1}+\xi _{𝐪_2}+\xi _𝐤+\xi _{𝐪_1+𝐪_2+𝐤}}\Phi (𝐪_1,𝐤).$$
(27)
Here we denote $`\Phi (𝐪_1,𝐪_2)=\Phi (\{𝐪_1,\xi _{𝐪_1}\},\{𝐪_2,\xi _{𝐪_2}\})`$.
The frequency integration of the third term in Eq.(26) over the frequencies $`\mathrm{\Omega }`$ and $`\omega `$ is more difficult because singularities are not essentially located in one half of the complex plane, as it was the case for the second term. For example $`\Phi (k,Qk)`$ has singularities in both half planes, with respect to $`\omega `$, and similarly for $`𝒯_2(Q)𝒯_2(Q)`$ with respect to $`\mathrm{\Omega }`$. We solve this problem by splitting the involved functions as the sum of two parts, one analytical in the upper complex plane, and the other one in the lower complex plane.
First we write $`F(\mathrm{\Omega },\text{Q},\text{q}_1,\text{q}_2)𝒢(Qq_1)𝒢(Qq_2)𝒯_2(Q)𝒯_2(Q)+(q_1q_2)`$ (we take into account that we want to calculate $`\Phi (q_1,q_2)`$”on the shell”) as:
$`F(\mathrm{\Omega },\text{Q},\text{q}_1,\text{q}_2)=U_+(\mathrm{\Omega },\text{Q},\text{q}_1,\text{q}_2)+U_{}(\mathrm{\Omega },\text{Q},\text{q}_1,\text{q}_2)`$ (28)
where $`U_+`$ and $`U_{}`$ are respectively analytical in the upper and lower complex planes of $`\mathrm{\Omega }`$. This is done by making use of the Cauchy formula $`f(\mathrm{\Omega })=(1/2i\pi )_C𝑑zf(z)/(z\mathrm{\Omega })`$ for a contour $`C`$ which encircles the real axis (on which $`F`$ has no singularity) and is infinitesimally near of it. This gives:
$`U_+(\mathrm{\Omega },\text{Q},\text{q}_1,\text{q}_2)={\displaystyle \frac{1}{2i\pi }}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑z{\displaystyle \frac{F(z,\text{Q},\text{q}_1,\text{q}_2)}{ziϵ\mathrm{\Omega }}}`$ (29)
with $`ϵ=0_+`$. Making use of $`F(\mathrm{\Omega })=F(\mathrm{\Omega })`$, we find $`U_{}(\mathrm{\Omega },\text{Q},\text{q}_1,\text{q}_2)=U_+(\mathrm{\Omega },\text{Q},\text{q}_1,\text{q}_2)`$.
On the other hand the last part of the third term $`\overline{T}_4(Q^{})d^4k^{}𝒢(k^{})𝒢(Q^{}k^{})\Phi (k^{},Q^{}k^{})=d^4k^{}𝒢(Q^{}/2+k^{})𝒢(Q^{}/2k^{})\Phi (Q^{}/2+k^{},Q^{}/2k^{})`$ satisfies $`\overline{T}_4(Q^{})=\overline{T}_4(Q^{})`$. This can be seen by substituting Eq.(26) for $`\Phi (Q^{}/2+k^{},Q^{}/2k^{})`$ in this last expression for $`\overline{T}_4(Q^{})`$. For the first term contribution, the result is trivial. For the second term, one has to make the shift $`kkQ^{}/2`$, and then $`kk^{}`$. In the third term one has to make the shift $`k^{}k^{}+Q/2`$ and then $`k^{}k^{}`$. Then, when we make the change $`QQ`$ in the third term of Eq.(26) and use $`\overline{T}_4(Q)=\overline{T}_4(Q)`$, we see that the $`U_{}`$ contribution is exactly identical to the $`U_+`$ contribution and we are left with a single contribution from $`U_{}`$ to evaluate.
In order to perform the $`\omega `$ integration in $`\overline{T}_4(Q)=d^4k𝒢(Q/2+k)𝒢(Q/2k)\Phi (Q/2+k,Q/2k)`$, we split:
$`\Phi (Q/2+k,Q/2k)=\Phi _+(Q/2+k,Q/2k)+\Phi _{}(Q/2+k,Q/2k)`$ (30)
into the sum of two functions, with $`\Phi _+`$ analytical in the upper complex plane with respect to $`\omega `$, and $`\Phi _{}`$ analytical in the lower complex plane. That this can be done is immediately seen from Eq.(26) itself. For the first term we just have to write the product of Green’s functions as $`𝒢(kQ/2)𝒢(kQ/2)=(𝒢(kQ/2)+𝒢(kQ/2))/(\mathrm{\Omega }+\xi _{𝐤+𝐐/2+\xi (𝐤𝐐/2}iϵ)`$, which has explicitely the required property. In the third term we can handle the product of the first two Green’s functions in the same way. Finally, in the second term, after performing the $`\omega `$ integration as indicated above (but without taking the ”on the shell” values for the frequencies), one sees that the result for the term written explicitely above in Eq.(26) is analytical in the lower complex plane with respect to $`\omega `$. The corresponding term obtained by $`(q_1q_2)`$ is analytic in the upper complex plane. In each case one checks that the functions analytical in the upper and lower complex plane are related by $`kk`$, so that $`\Phi _{}(Q/2+k,Q/2k)=\Phi _+(Q/2k,Q/2+k)`$. Hence by the change of variable $`kk`$, the contributions of $`\Phi _+`$ and $`\Phi _{}`$ are equal.
Then we have arrived, for the calculation of $`\overline{T}_4(Q)`$, to a situation which is similar to the one we met for three particles. Since $`\Phi _+(Q/2k,Q/2+k)`$ and $`𝒢(Q/2k)`$ are analytical in the lower complex plane, we can close the integration contour at infinity in this lower half plane and the only contribution comes from the pole of $`𝒢(Q/2+k)`$. This leads to:
$`\overline{T}_4(Q)=2i{\displaystyle \frac{d𝐤}{(2\pi )^3}\frac{(\mathrm{\Omega },𝐤,𝐐)}{\mathrm{\Omega }\xi _{𝐤+𝐐/2}\xi _{𝐤𝐐/2}+iϵ}}`$ (31)
where $`(\mathrm{\Omega },𝐤,𝐐)`$ is $`\Phi _+(Q/2k,Q/2+k)`$ evaluated for $`\omega =\xi _{𝐤+𝐐/2}\mathrm{\Omega }/2`$. An important property, which can be checked on each term contributing to $`\Phi _+(Q/2k,Q/2+k)`$ is that $`(\mathrm{\Omega },𝐤,𝐐)`$ is analytical in the lower complex plane with respect to $`\mathrm{\Omega }`$. Hence the integration of $`U_{}(\mathrm{\Omega },\text{Q},\text{q}_1,\text{q}_2)\overline{T}_4(Q)`$ over $`\mathrm{\Omega }`$ can also be performed by closing the contour in the lower half plane, since the only singularity in this half plane is the pole due to the denominator in Eq.(31). The contribution of this pole leads to the evaluation of $`(\mathrm{\Omega },𝐤,𝐐)`$ for $`\mathrm{\Omega }=\xi _{𝐤+𝐐/2}+\xi _{𝐤𝐐/2}`$. Taken with the above definition of $``$ this means that we have calculated $`\Phi _+(Q/2k,Q/2+k)`$ for $`\mathrm{\Omega }/2\omega =\xi _{𝐤𝐐/2}`$ and $`\mathrm{\Omega }/2+\omega =\xi _{𝐤+𝐐/2}`$, which is just an evaluation ”on the shell”. Because of the simple relation between $`\Phi _+`$ and $`\Phi _{}`$ the result can be expressed in terms of $`\Phi (𝐤+𝐐/2,𝐤+𝐐/2)`$ itself.
Gathering all the above results we end up with the following complete equation for $`\Phi (𝐪_1,𝐪_2)`$:
$$\begin{array}{c}\Phi (𝐪_\mathrm{𝟏},𝐪_\mathrm{𝟐})=\frac{1}{4\xi _{𝐪_1}\xi _{𝐪_2}}+\frac{d^3𝐤}{(2\pi )^3}\frac{𝒯_2(\xi _{𝐪_1}\xi _𝐤,𝐪_1+𝐤)}{\xi _{𝐪_1}+\xi _{𝐪_2}+\xi _𝐤+\xi _{𝐪_1+𝐪_2+𝐤}}\Phi (𝐪_1,𝐤)\hfill \\ \hfill \frac{d^3𝐐}{(2\pi )^3}\frac{d^3𝐤}{(2\pi )^3}U(\xi _{𝐤+𝐐/2}+\xi _{𝐤𝐐/2},\text{Q},\text{q}_1,\text{q}_2)\Phi (𝐤+𝐐/2,𝐤+𝐐/2)+(q_1q_2).\end{array}$$
(32)
In this equation we have modified the integration contour in the definition of $`U_{}`$ to have it running on the imaginary axis rather than on the real axis, and we have used the symmetry property of $`F(z,\text{Q},\text{q}_1,\text{q}_2)`$ with respect to $`z`$, together with symmetry properties of $`\Phi (𝐪_1,𝐪_2)`$, to rewrite the result in terms of the real function:
$`U(\mathrm{\Omega },\text{Q},\text{q}_1,\text{q}_2)={\displaystyle \frac{\mathrm{\Omega }}{\pi }}{\displaystyle _0^{\mathrm{}}}𝑑y{\displaystyle \frac{F(iy,\text{Q},\text{q}_1,\text{q}_2)}{y^2+\mathrm{\Omega }^2}}`$ (33)
which shows that $`\Phi (𝐪_\mathrm{𝟏},𝐪_\mathrm{𝟐})`$ itself is real.
We have made practical numerical use of Eq. (32) to find for example the ground state energy. Although this turned out to be quite feasable, this equation appears finally less convenient than what we have described in section IV. This was expected since the solution implies quadruple integrals, instead of the triple integrals we had only to deal with in section IV.
## Appendix B Modified dimer-dimer scattering equation
This appendix is devoted to an alternative description of the dimer-dimer scattering process. The purpose is to obtain a direct integral equation for $`T_4(p_1,p_2;P)`$, in a way convenient for numerical calculations. Below we derive such a set of equations, that were used for practical computations as indicated in section IV.
The first step is to construct for two dimers a ”bare” interaction potential, or vertex, $`\mathrm{\Delta }_4`$, which is the sum of all irreducible diagrams, and then to build ladder diagrams from this vertex, in order to obtain an integral equation (see Fig. 9). These irreducible diagrams are those ones which cannot be divided by a vertical line into two parts connected by two dimer lines. As it was pointed above the vertex $`\mathrm{\Delta }_4`$ is given by the series shown on Fig.4e, since the diagrams on Fig.4f are by contrast reducible. Again we can eliminate $`T_3`$ from our considerations and express $`\mathrm{\Delta }_4`$ only in terms of $`T_2`$. For this purpose we have to introduce a special vertex with two fermionic and one dimer incoming lines and two dimer outgoing lines $`\mathrm{\Gamma }_{\alpha \beta }(q_1,q_2;p_2,P)`$ (see Fig. 10). This vertex $`\mathrm{\Gamma }_{\alpha \beta }(q_1,q_2;p_2,P)`$ corresponds to the vertex $`\mathrm{\Delta }_4`$ with one incoming dimer line being removed, in much the same way as $`\mathrm{\Phi }(q_1,q_2;p_2,P)`$ and $`T_4(p_1,p_2;P)`$ are related in Eq. (8). The difference is that $`\mathrm{\Gamma }_{\alpha \beta }(q_1,q_2;p_2,P)`$ is irreducible with respect to two dimer lines while $`\mathrm{\Phi }(q_1,q_2;p_2,P)`$ is not, just in the same way as $`T_4(p_1,p_2;P)`$ and $`\mathrm{\Delta }_4(p_1,p_2;P)`$ are related. The corresponding equation relating $`\mathrm{\Gamma }_{\alpha \beta }(q_1,q_2;p_2,P)`$ and $`\mathrm{\Delta }_4(p_1,p_2;P)`$ is:
$$\mathrm{\Delta }_4(p_1,p_2;P)=\frac{1}{2}\underset{Q;\alpha ,\beta }{}\chi (\alpha ,\beta )G(P+p_1Q)G(Q)\mathrm{\Gamma }_{\alpha \beta }(P+p_1Q,Q;p_2,P).$$
(34)
One can readily verify that the diagrammatic expansion for $`\mathrm{\Gamma }`$ shown on Fig. 11 yields the same series as the one shown on Fig. 4e for the vertex $`\mathrm{\Delta }_4`$. The spin part of $`\mathrm{\Gamma }_{\alpha ,\beta }`$ has again the simple form $`\mathrm{\Gamma }_{\alpha ,\beta }(q_1,q_2;P,p_2)=\chi (\alpha ,\beta )\mathrm{\Gamma }(q_1,q_2;p_2,P)`$ and the function $`\mathrm{\Gamma }(q_1,q_2;p_2,P)`$ obeys the following equation:
$$\begin{array}{c}\mathrm{\Gamma }(q_1,q_2;p_2,P)=G(Pq_1+p_2)G(Pq_2p_2)G(Pq_2+p_2)G(Pq_1p_2)\hfill \\ \hfill \underset{Q}{}G(Q)G(2Pq_1q_2Q)\left[T_2(2Pq_1Q)\mathrm{\Gamma }(q_1,Q;p_2,P)+T_2(2Pq_2Q)\mathrm{\Gamma }(Q,q_2;p_2,P)\right].\end{array}$$
(35)
The sign minus in (35) is a consequence of the anticommutativity of Fermi operators. It is clear that Eqs. (34) and (35) can be analytically integrated over the variable $`\mathrm{\Omega }`$. Thus the $`s`$-wave component of the vertex $`\mathrm{\Gamma }(q_1,q_2;p_2,P)`$ is a function of the absolute values of vectors $`|𝐪_1|`$ and $`|𝐪_2|`$, the angle between them, the absolute value of vector $`|𝐩_2|`$, and the frequency $`\omega _2`$. The $`s`$-wave component of the sum of all irreducible diagrams $`\mathrm{\Delta }_4(p_1,p_2;P)`$ is a function of the absolute values of the vectors $`|𝐩_1|`$ and $`|𝐩_2|`$ and the frequencies $`\omega _1`$ and $`\omega _2`$.
The fully symmetrized vertex $`T_4(p_1,p_2;P)`$ of two-dimer scattering can be found from the solution of the following equation (see Fig. 9):
$$T_4(p_1,p_2;P)=\mathrm{\Delta }_4(p_1,p_2;P)+\frac{1}{2}\underset{q}{}\mathrm{\Delta }_4(p_1,q;P)T_2(P+q)T_2(Pq)T_4(q,p_2;P),$$
(36)
where $`\mathrm{\Delta }_4(p_1,p_2;P)`$ is the sum of all irreducible diagrams, $`P\pm p_{1,2}=\{E_b\pm \omega _{1,2},\pm 𝐩_{1,2}\}`$ are 4-vectors of incoming (1) and outgoing (2) dimers in the center-of-mass system.
Let us finally note that, equivalently to our above derivation, Eq. (34-36) can be also related to Eq. (8) and Eq. (9) algebraically by simple formal operator manipulations.
###### Acknowledgements.
This work was supported by Russian Foundation for Basic Research (Grant No. 04-02-16050), CRDF (Grant No. RP2-2355-MO-02) and the grant of Russian Ministry of Science and Education. We are grateful to A.F. Andreev, I.A. Fomin, P. Fulde, Yu. Kagan, L.V. Keldysh, Yu. Lozovik, S.V. Maleev, B.E. Meierovich, A.Ya. Parshin, P. Pieri T.M. Rice, V.N. Ryzhov, G.V. Shlyapnikov, G.C. Strinati, V.B. Timofeev, D. Vollhardt and P. Wölfle for fruitful discussions. M.Yu.K is grateful to the University Pierre and Marie Curie for the hospitality on the first stage of this work. Laboratoire de Physique Statistique is ” Laboratoire associé au Centre National de la Recherche Scientifique et aux Universités Paris 6 et Paris 7 ”. |
warning/0507/math0507318.html | ar5iv | text | # On Cheng’s Eigenvalue Comparison Theorems
## 1 Introduction
Let $`M`$ be a complete $`n`$-dimensional Riemannian manifold and denote by $`B_M(p,r)`$ the geodesic ball with center $`p`$ and radius $`r`$ and by $`\lambda _1(B_M(p,r))`$ the first Dirichlet eigenvalue of $`B_M(p,r)`$. Cheng in , using a result of Barta , proved that if the sectional curvature of $`M`$ is bounded above $`K_M\kappa `$ and $`r<\mathrm{min}\{\mathrm{inj}(p),\pi /\sqrt{\kappa }\}`$, ($`\pi /\sqrt{\kappa }=\mathrm{}`$ if $`\kappa 0`$), then $`\lambda _1(B_M(p,r))\lambda _1(B_{𝕄(\kappa )}(r))`$, where $`𝕄(\kappa )`$ denote the simply connected space form of constant sectional curvature $`\kappa `$. Cheng also in proved that if the Ricci curvature of $`M`$ is bounded below $`Ric_M(n1)\kappa `$ then the reverse inequality $`\lambda _1(B_M(p,r))\lambda _1(B_{𝕄(\kappa )}(r))`$ holds for $`r<inj(p)`$. In , choosing a suitable test function for the Rayleigh quotient, Cheng improved this later inequality proving that if $`Ric_M(n1)\kappa `$ then $`\lambda _1(B_M(p,r))\lambda _1(B_{𝕄(\kappa )}(r))`$ for every $`r>0`$, with equality holding (for some $`r`$) if and only if the geodesic balls $`B_M(p,r)`$ and $`B_{𝕄(\kappa )}(r)`$ are isometric and $`r<inj(p)`$. That raises the questions of whether it is possible to prove Cheng’s lower eigenvalue inequality beyond the cut locus and show that the geodesic balls are isometric if they have the same first eigenvalue. These questions were addressed in and proven to be true, (under upper sectional curvature bounds), provided that the ($`n1`$)-Hausdorff measure $`^{n1}(Cut(p)B_M(p,r))=0`$, where $`Cut(p)`$ is the cut locus of $`p`$. In this paper we apply our version of Barta’s theorem (Theorem 2.3) to prove an extension of Cheng’s lower and upper eigenvalues inequalities for geodesic balls within the cut locus (of its center) without sectional or Ricci curvature bounds. These inequalities have a weaker form of geometric rigidity in the equality case and we show with family of examples that this rigidity is all we can expect for. To state our result, consider $`B_M(p,r)M`$ and $`B_{𝕄(\kappa )}(r)𝕄(\kappa )`$ geodesic balls within the cut locus and let $`(t,\theta )(0,r]\times 𝕊^{n1}`$ be geodesic coordinates for $`B_M(p,r)`$ and $`B_{𝕄(\kappa )}(r)`$. Let $`H_M(t,\theta )`$ and $`H_{𝕄(\kappa )}(t,\theta )=H_{𝕄(\kappa )}(t)`$ be respectively the mean curvatures of the distance spheres $`B_M(p,t)`$ and $`B_{𝕄(\kappa )}(t)`$ at the point $`(t,\theta )`$ with respect to the unit vector field $`/t`$. Our first result is the following theorem.
###### Theorem 1.1
If $`H_M(s,\theta )H_{𝕄(\kappa )}(s)`$, for all $`s(0,r]`$ and all $`\theta 𝕊^{n1}`$ then
$$\lambda _1(B_M(p,r))\lambda _1(B_{𝕄(\kappa )}(r)).$$
(1)
If $`H_M(s,\theta )H_{𝕄(\kappa )}(s)`$, for all $`s(0,r]`$ and all $`\theta 𝕊^{n1}`$ then
$$\lambda _1(B_M(p,r))\lambda _1(B_{𝕄(\kappa )}(r)).$$
(2)
Equality in (1) or (2) holds if and only if $`H_M(s,\theta )=H_{𝕄(\kappa )}(s)`$, $`s(0,r]`$ and $`\theta 𝕊^{n1}`$.
Observe that the hypotheses of Theorem (1.1) are implied by an upper sectional curvature bound $`K_M\kappa `$ and a lower Ricci curvature bound $`Ric_M(n1)\kappa `$ respectively. On the other hand we construct examples of smooth metrics on $`^n=[0,\mathrm{})\times 𝕊^{n1}`$ such that the radial sectional curvatures is bounded below $`K(x)(t,v)>\kappa `$ outside a compact set $`(x^nB_^n(1))`$ but $`H_M(s,\theta )H_{𝕄(\kappa )}(s)`$, for all $`s(0,\mathrm{})`$ and all $`\theta 𝕊^{n1}`$, see example (4.1). This shows that Theorem (1.1) is a true extension of Cheng s eigenvalue comparison theorem (within the cut locus). The rigidity in case of equality of the eigenvalues, ($`H_M(s,\theta )=H_{𝕄(\kappa )}(s)`$, $`s(0,r]`$ and $`\theta 𝕊^{n1}`$), implies that the balls $`B_M(p,r)`$ and $`B_{𝕄(\kappa )}(r)`$ are isometric if we have that $`K_M\kappa `$ or $`Ric_M(n1)\kappa `$. Moreover, if the metric of $`B_M(p,r)`$ is expressed in geodesic coordinates by $`dt^2+f^2(t)d\theta ^2`$, $`f(0)=0`$, $`f^{}(0)=1`$, $`f(t)>0`$ for $`t>0`$ then the rigidity (even without curvature bounds) also implies that the balls $`B_M(p,r)`$ and $`B_{𝕄(\kappa )}(r)`$ are isometric, see Remark (4.2). This is the case if the the dimension of $`M`$ is two. On the other hand we also construct a family of complete smooth metrics $`g(\kappa )`$ on $`^n`$, $`\kappa <0`$ such that $`g(\kappa )`$ is non isometric to the constant sectional curvature metric of $`𝕄(\kappa )`$ but the geodesic balls $`B_{g(\kappa )}(r)`$, and $`B_{𝕄(\kappa )}(r)`$ have the same first eigenvalue $`\lambda _1(B_{𝕄(\kappa )}(r))`$ and their geodesic spheres of same radius have the same mean curvatures, see examples(4.3). These examples show that the rigidity stated in Theorem (1.1) in general is all we can expect without curvature bounds. The proof we present for Theorem (1.1) in fact proves more, we have few generalizations in section 3, (see Theorems 3.1, 3.2). We also generalize Veeravalli’ s examples , see Theorem (3.3).
## 2 Preliminaries
A powerful tool to obtain lower bounds for the first Dirichlet eigenvalue of smooth bounded domains in Riemannian manifolds is the following theorem proved by J. Barta in .
###### Theorem 2.1 (Barta)
Let $`\mathrm{\Omega }M`$ be a domain with compact closure and nonempty smooth boundary $`\mathrm{\Omega }`$. Let $`\lambda _1(\mathrm{\Omega })`$ be the first Dirichlet eigenvalue of $`\mathrm{\Omega }`$. Let $`fC^2(\mathrm{\Omega })C^0(\overline{\mathrm{\Omega }})`$ with $`f>0`$ in $`\mathrm{\Omega }`$ and $`f|\mathrm{\Omega }=0`$. Then
$$\underset{\mathrm{\Omega }}{sup}(\frac{\mathrm{}f}{f})\lambda _1(\mathrm{\Omega })\underset{\mathrm{\Omega }}{inf}(\frac{\mathrm{}f}{f}).$$
(3)
###### Remark 2.2
The first observation is that to prove the lower inequality in (2) it is necessary only to have that $`f>0`$ in $`\mathrm{\Omega }`$. A second observation is that each of the inequalities (3) is strict unless $`f`$ is a first eigenfunction of $`\mathrm{\Omega }`$. This observation although trivial is essencial in the proof of the rigidity statement in Theorem (1.1) and its seems to have passed unobserved by Cheng.
For arbitrary open sets $`\mathrm{\Omega }`$, we proved in the following extension of Barta’s Theorem that gives lower bounds for fundamental tone $`\lambda ^{}(\mathrm{\Omega })`$. Recall that the fundamental tone $`\lambda ^{}(\mathrm{\Omega })`$ of an open set $`\mathrm{\Omega }`$ is given by
$$\lambda ^{}(\mathrm{\Omega })=inf\{\frac{_\mathrm{\Omega }|f|^2}{_\mathrm{\Omega }f^2},fL_{1,0}^2(\mathrm{\Omega }),f0\},$$
where $`L_{1,0}^2(\mathrm{\Omega })`$ is the completion of $`C_0^{\mathrm{}}(\mathrm{\Omega })`$ with respect to the norm $`\phi _\mathrm{\Omega }^2=_\mathrm{\Omega }\phi ^2+_\mathrm{\Omega }|\phi |^2.`$
###### Theorem 2.3
Let $`\mathrm{\Omega }M`$ be an open subset of Riemannian manifold. Then
$$\lambda ^{}(\mathrm{\Omega })\underset{𝒳(\mathrm{\Omega })}{sup}\{\underset{\mathrm{\Omega }}{inf}(\mathrm{div}X|X|^2)\},$$
(4)
where $`𝒳(\mathrm{\Omega })`$ is the set of all vector fields $`X`$ in $`\mathrm{\Omega }`$ such that $`_\mathrm{\Omega }\mathrm{div}(fX)=0`$ for all $`fC_0^{\mathrm{}}(\mathrm{\Omega })`$. If $`\mathrm{\Omega }`$ is a relatively compact open set with smooth boundary then
$$\lambda _1(\mathrm{\Omega })=\underset{𝒳(\mathrm{\Omega })}{sup}\{\underset{\mathrm{\Omega }}{inf}(\mathrm{div}X|X|^2)\}.$$
(5)
Both results (Barta’s Theorem and Theorem (2.3)) coincides in bounded domains with smooth boundaries, but the vector field aspect of this version reveal the role of the mean curvatures of the distance spheres in the comparisons of eigenvalues.
### 2.1 Proof of Theorem 1.1
Let $`(t,\theta )(0,r]\times 𝕊^{n1}`$ be geodesic coordinates for $`B_M(p,r)`$ and $`B_{𝕄(\kappa )}(r)`$ and $`u:B_{𝕄(\kappa )}(r)`$ be a positive first Dirichlet eigenfunction. It is well known $`u`$ is radial function, i.e. $`u(t,\theta )=u(t)`$ and $`u^{}(t)0`$. Observe that $`u(t,\theta )=u(t)`$ also defines a smooth function on $`B_M(p,r)`$. Now, consider vector fields $`X_1`$ on $`B_M(p,r)`$ and $`X_2`$ on $`B_{𝕄(\kappa )}(r)`$ given by
$$\begin{array}{ccc}X_1(t,\theta )\hfill & =\hfill & \frac{u^{}(t)}{u(t)}\frac{_{\mathrm{\hspace{0.17em}1}}}{t}(t,\theta ),\hfill \\ & & \\ X_2(t,\theta )\hfill & =\hfill & \frac{u^{}(t)}{u(t)}\frac{_{\mathrm{\hspace{0.17em}2}}}{t}(t,\theta ).\hfill \end{array}$$
(6)
Here $`{\displaystyle \frac{_{\mathrm{\hspace{0.17em}1}}}{t}}`$ and $`{\displaystyle \frac{_{\mathrm{\hspace{0.17em}2}}}{t}}`$ are the radial vector fields in $`B_M(p,r)`$ and $`B_{𝕄(\kappa )}(r)`$ respectively. From now on let us write $`B_M(r)`$ instead $`B_M(p,r)`$ for simplicity of notation. Now we have that
$`{\displaystyle \frac{\mathrm{}_Mu}{u}}=\mathrm{div}_MX_1|X_1|^2`$ $`=`$ $`\mathrm{div}_MX_1\mathrm{div}_{𝕄(\kappa )}X_2+|X_2|^2|X_1|^2+\mathrm{div}_{𝕄(\kappa )}X_2|X_2|^2`$
$`=`$ $`\mathrm{div}_MX_1\mathrm{div}_{𝕄(\kappa )}X_2{\displaystyle \frac{\mathrm{}_{𝕄(\kappa )}u}{u}}`$
$`=`$ $`\mathrm{div}_MX_1\mathrm{div}_{𝕄(\kappa )}X_2+\lambda _1(B_{𝕄(\kappa )}(r)),`$
since $`\mathrm{div}_{𝕄(\kappa )}X_2|X_2|^2={\displaystyle \frac{\mathrm{}_{𝕄(\kappa )}u}{u}}=\lambda _1(B_{𝕄(\kappa )}(r))`$ and $`|X_1|^2=|X_2|^2`$.
By Theorem (2.1) or (2.3) and by identity (2.1) we have that
$$\lambda _1(B_M(r))\underset{(t,\theta )}{inf}(\mathrm{div}_MX_1|X_1|^2)\underset{(t,\theta )}{inf}[\mathrm{div}_MX_1\mathrm{div}_{𝕄(\kappa )}X_2]+\lambda _1(B_{𝕄(\kappa )}(r))$$
(8)
Since $`B_M(r)`$ is a smooth domain we can apply Barta’s Theorem and using identity (2.1) we have that
$$\lambda _1(B_M(r))\underset{(t,\theta )}{sup}[\mathrm{div}_MX_1|X_1|^2]\underset{(t,\theta )}{sup}[\mathrm{div}_MX_1\mathrm{div}_{𝕄(\kappa )}X_2]+\lambda _1(B_{𝕄(\kappa )}(r))$$
(9)
We will associate the difference $`\mathrm{div}_MX_1\mathrm{div}_{𝕄(\kappa )}X_2`$ to the mean curvature of the distance spheres through the following well known lemma.
###### Lemma 2.4
Let $`M\overline{M}`$ be a smooth hypersurface. Let $`X`$ be a smooth vector field on $`\overline{M}`$. Then at $`xM`$ we have that
$$\mathrm{div}_{\overline{M}}X(x)=\mathrm{div}_MX^t(x)X,\stackrel{}{H}(x)+\overline{}_\eta X,\eta (x),$$
(10)
where $`X^t`$ is the orthogonal projection of $`X`$ onto the tangent space $`T_xM`$, $`\stackrel{}{H}`$ is the mean curvature vector of $`M`$ at $`x`$, $`\overline{}`$ is the Levi-Civita connection of $`\overline{M}`$ and $`\eta T_xM^{}`$.
Using this lemma we can compute $`\mathrm{div}_MX_1\mathrm{div}_{𝕄(\kappa )}X_2`$ at points of $`B_M(r)`$ and of $`B_{𝕄(\kappa )}(r)`$ with the same coordinates $`(t,\theta )`$.
$`\mathrm{div}_MX_1\mathrm{div}_{𝕄(\kappa )}X_2`$ $`=`$ $`X_1,\underset{M}{\overset{}{H}}_M+X_2,\underset{𝕄(\kappa )}{\overset{}{H}}_{𝕄(\kappa )}`$ (11)
$`+\overline{}_{_1/t}^MX_1,{\displaystyle \frac{_1}{t}}_M\overline{}_{_2/t}^{𝕄(\kappa )}X_2,{\displaystyle \frac{_2}{t}}_{𝕄(\kappa )}`$
$`=`$ $`(u^{}/u)(H_MH_{𝕄(\kappa )})+(u^{}/u)^{}(u^{}/u)^{}`$
Since
$$\overline{}_{_1/t}^MX_1,\frac{_1}{t}_M=\overline{}_{_2/t}^{𝕄(\kappa )}X_2,\frac{_2}{t}_{𝕄(\kappa )}=(u^{}/u)^{}$$
and $`\underset{M}{\overset{}{H}}=H_M_1/t`$ and $`\underset{𝕄(\kappa )}{\overset{}{H}}=H_{𝕄(\kappa )}_2/t`$. Hence
$$\mathrm{div}_MX_1\mathrm{div}_{𝕄(\kappa )}X_2=(u^{}/u)(H_MH_{𝕄(\kappa )}).$$
(12)
Now recall that $`(u^{}/u)0`$. If $`(H_MH_{𝕄(\kappa )})0`$ then (8) and (12) implies (1). Likewise, if $`(H_MH_{𝕄(\kappa )})0`$ then (9) and (12) implies (2). To treat the equality case observe that the proof we presented was nothing but giving a suitable positive function $`u`$ on $`B_M(r)`$ then applying Barta’s Theorem to find the lower bound for $`inf_{B_M(r)}(\mathrm{}_Mu/u)\lambda _1(B_{𝕄(\kappa )}(r))`$. Now, suppose that $`\lambda _1(B_M(r))=\lambda _1(B_{𝕄(\kappa )}(r))`$ then (8) implies that $`\lambda _1(B_M(r))=inf_{(t,\theta )}(\mathrm{div}_MX_1|X_1|^2)`$ and $`inf_{(t,\theta )}[\mathrm{div}_MX_1\mathrm{div}_{𝕄(\kappa )}X_2]=0`$. The Remark (2.2) says that the infimum (supremum) in (3) is achieved by a positive function $`f`$ if and only if the function $`f`$ is an eigenfunction. Thus $`\lambda _1(B_M(r))=inf_{(t,\theta )}(\mathrm{div}_MX_1|X_1|^2)`$ is saying that the function $`u:B_M(r)`$ is a positive first eigenfunction of $`B_M(r)`$, in particular that $`\lambda _1(B_M(r))=\mathrm{div}_MX_1|X_1|^2`$. From (2.1) we have that $`\mathrm{div}_MX_1\mathrm{div}_{𝕄(\kappa )}X_2=\lambda _1(B_M(r))\lambda _1(B_{𝕄(\kappa )}(r))=0`$. On the other hand, $`\mathrm{div}_MX_1\mathrm{div}_{𝕄(\kappa )}X_2=(u^{}/u)(H_MH_{𝕄(\kappa )})`$ and $`u^{}(t)=0`$ if and only if $`t=0`$. Therefore we have that $`H_M(t,\theta )=H_{𝕄(\kappa )}(t,\theta )`$ for all $`t>0`$ and all $`\theta `$. The equality in (2) is treated in the same way.
## 3 Generalizations of Theorem 1.1
The first generalization we are going to consider is the following. Let $`M`$ be a $`n`$-dimensional complete Riemannian manifold and let $`B_M(r)M`$ be a geodesic ball within the cut locus. Consider $`^m=[0,\mathrm{})\times 𝕊^m`$ with metric $`ds^2=dt^2+g^2(t)d\xi ^2`$, where $`g:[0,\mathrm{})`$ is a smooth function satisfying $`g(0)=0`$, $`g^{}(0)=1`$, $`g(t)>0`$ for $`t(0,\mathrm{})`$. Let $`B_^m(r)`$ be a geodesic ball of radius $`r`$. Let $`(t,\theta )(0,r]\times 𝕊^{n1}`$ be geodesic coordinates for $`B_M(r)`$ and $`(t,\xi )(0,r]\times 𝕊^{m1}`$ be geodesic coordinates for $`B_^m(r)`$. Let $`H_M(t,\theta )`$ and $`H_^m(t,\xi )=H_^m(t)`$ be respectively the mean curvatures of the distance spheres $`B_M(t)`$ and $`B_^m(t)`$ at the points $`(t,\theta )`$ and $`(t,\xi )`$ with respect to the unit vector field $`/t`$.
###### Theorem 3.1
If $`H_M(s,\theta )H_^m(s)=(m1)(g^{}/g)(s)`$, $``$ $`s(0,r]`$ and $`\theta 𝕊^{n1}`$ then
$$\lambda _1(B_M(r))\lambda _1(B_^m(r)).$$
(13)
If $`H_M(s,\theta )H_^m(s)=(m1)(g^{}/g)(s)`$, $``$ $`s(0,r]`$ and $`\theta 𝕊^{n1}`$ then
$$\lambda _1(B_M(r))\lambda _1(B_^m(r)).$$
(14)
Equality in (13) or (14) holds if and only if $`n=m`$ and $`H_M(s,\theta )=H_^n(s)`$, $`s(0,r]`$ and $`\theta 𝕊^{n1}`$.
A positive first eigenfunction $`u`$ of a geodesic ball $`B_^m(r)`$ within the cut locus is radial ($`u(t,\xi )=u(t)`$) and $`u^{}(t)0`$ with $`u^{}(t)=0t=0`$. See a proof of that in , pages 40-44. Define $`v:B_M(r)`$ by $`v(t,\theta )=u(t)`$ and take vector fields $`X_1`$ in $`B_M(r)`$ and $`X_2`$ in $`B_^m(r)`$ by
$$\begin{array}{ccc}X_1(t,\theta )\hfill & =\hfill & \frac{u^{}(t)}{u(t)}\frac{_{\mathrm{\hspace{0.17em}1}}}{t}(t,\theta ),\hfill \\ & & \\ X_2(t,\xi )\hfill & =\hfill & \frac{u^{}(t)}{u(t)}\frac{_{\mathrm{\hspace{0.17em}2}}}{t}(t,\xi ).\hfill \end{array}$$
(15)
Proceeding as in the proof of Theorem (1.1)
$`{\displaystyle \frac{\mathrm{}_Mv}{v}}(t,\theta )=(\mathrm{div}_MX_1|X_1|^2)(t,\theta )`$ $`=`$ $`\mathrm{div}_MX_1(t,\theta )\mathrm{div}_^mX_2(t,\xi )`$
$`+`$ $`\mathrm{div}_^mX_2(t,\xi )|X_2|^2(t,\xi )`$
$`+`$ $`|X_2|^2(t,\xi )|X_1|^2(t,\theta ).`$
Since we have that $`(\mathrm{div}_^mX_2|X_2|^2)(t,\xi )={\displaystyle \frac{\mathrm{}_^mu}{u}}=\lambda _1(B_^m(r))`$, $`|X_2|^2(t,\xi )|X_1|^2(t,\theta )=0`$ and $`\mathrm{div}X_2(t,\xi )=\mathrm{div}X_2(t)`$. Thus we derive that
$$\lambda _1(B_M(r))\underset{(t,\theta )}{inf}(\mathrm{div}_MX_1|X_1|^2)\underset{(t,\theta )}{inf}[\mathrm{div}_MX_1\mathrm{div}_^mX_2(t)]+\lambda _1(B_^m(r)).$$
(17)
Likewise, we can derive
$$\lambda _1(B_M(r))\underset{(t,\theta )}{sup}(\mathrm{div}_MX_1|X_1|^2)\underset{(t,\theta )}{sup}[\mathrm{div}_MX_1\mathrm{div}_^mX_2(t)]+\lambda _1(B_^m(r)).$$
(18)
Then applying Lemma (2.4) we have that
$$\mathrm{div}_MX_1(t,\theta )\mathrm{div}_^mX_2(t)=\frac{u^{}(t)}{u(t)}(H_M(t,\theta )H_^m(t))$$
If $`H_M(s,\theta )H_^m(s)`$, $``$ $`s(0,r]`$ and $`\theta 𝕊^{n1}`$ then $`\lambda _1(B_M(r))\lambda _1(B_^m(r)).`$ On the other hand if $`H_M(s,\theta )H_^m(s)`$, $``$ $`s(0,r]`$ and $`\theta 𝕊^{n1}`$ then $`\lambda _1(B_M(r))\lambda _1(B_^m(r)).`$ In case that $`\lambda _1(B_M(r))=\lambda _1(B_^m(r))`$ we have by (17) that $`\lambda _1(B_M(r))=\mathrm{div}_MX_1|X_1|^2`$ and $`\mathrm{div}_MX_1(s,\theta )\mathrm{div}_^mX_2(s)=0`$ for all $`s(0,r]`$ and $`\theta 𝕊^{n1}`$. Thus by Remark (2.2) the function $`v`$ is a positive eigenfunction of $`B_M(r)`$ and $`H_M(s,\theta )=H_^m(s)`$ for all $`sr`$, $`\theta 𝕊^{n1}`$. To prove that $`m=n`$ we proceed as follows. Let $`p`$ be the center of the ball $`B_M(r)`$. For fixed $`\theta 𝕊^{n1}T_pM`$, let $`\tau _t`$ denote parallel translation by $`t`$ units along the unique minimal geodesic $`\gamma _\theta `$ satisfying $`\gamma _\theta (0)=p`$ and $`\gamma _\theta ^{}(0)=\theta `$. For $`\eta \theta ^{}T_pM`$ set $`\eta =\tau _t\{R(\gamma _\theta ^{}(t),\tau _t\eta )\gamma _\theta ^{}(t)\}`$, where $`R`$ is the Riemannian curvature tensor and set $`𝒜(t,\theta )`$ the path of linear transformations of $`\theta ^{}`$ satisfying $`𝒜^{\prime \prime }+𝒜=0`$ with initial conditions $`𝒜(0,\theta )=0`$, $`𝒜^{}(0,\theta )=I`$. The Riemannian metric of $`M`$ on the geodesic ball $`B_M(r)`$ is expressed by $`ds^2(\mathrm{exp}t\theta )=dt^2+|𝒜(t,\theta )d\theta |^2`$. Set $`\sqrt{G}(t,\theta )=det𝒜(t,\theta )`$. The mean curvature $`H_M(t,\theta )`$ of the geodesic sphere $`B_M(t)`$ at a point $`(t,\theta )`$ (with respect to $`/t`$) is given by $`{\displaystyle \frac{\sqrt{G}^{}(t,\theta )}{\sqrt{G}(t,\theta )}}`$. Moreover for small $`t`$ we have the Taylor expansions $`\sqrt{G}(t,\theta )=t^{n1}(1t^2Ric(\theta ,\theta )/6+O(t^3)).`$ See , pages 316-317. Thus,
$$\frac{\sqrt{G}^{}(t,\theta )}{\sqrt{G}(t,\theta )}=\frac{(n1)(n+1)t^2Ric(\theta ,\theta )/6+O(t^3)}{t(1t^2Ric(\theta ,\theta )/6+O(t^3))}$$
(19)
On the other hand the metric of $`B_^m(r)`$ is given by $`dt^2+g^2(t)d\xi ^2`$, where $`g(0)=0`$, $`g^{}(0)=1`$. The mean curvature $`H_^m(t,\xi )`$ of the geodesic sphere $`B_^m(t)`$ at a point $`(t,\xi )`$ is given by $`(m1){\displaystyle \frac{g^{}(t)}{g(t)}}`$. The Taylor expansion of $`g`$ is given by $`g(t)=t+g^{\prime \prime }(0)t^2/2+O(t^3)`$. Therefore,
$$(m1)\frac{g^{}(t)}{g(t)}=(m1)\frac{1+g^{\prime \prime }(0)t+O(t^2)}{t(1+g^{\prime \prime }(0)t/2+O(t^2))}$$
(20)
Now, we have that $`H_M(t,\theta )=H_^m(t)`$ for all $`t(0,r]`$. Then
$$\frac{(n1)(n+1)t^2Ric(\theta ,\theta )/6+O(t^3)}{(1t^2Ric(\theta ,\theta )/6+O(t^3))}=(m1)\frac{1+g^{\prime \prime }(0)t+O(t^2)}{(1+g^{\prime \prime }(0)t/2+O(t^2))}$$
(21)
Letting $`t0`$ we have that $`n=m`$.
Another generalization of Theorem (1.1) is obtained considering the incomplete cone over an $`(n1)`$-dimensional compact Riemannian manifold $`(N,dh^2)`$. The incomplete cone $`C_f(N)`$ over $`N`$ is the Riemannian space $`C(N)=(0,\mathrm{})\times N`$ with metric $`ds_f^2=dt^2+f^2(t,x)dh^2`$, where $`f:[0,\mathrm{})\times N`$ is a smooth function satisfying $`f(0,x)=0,f^{}(0,x)=1`$, $`f(t,x)>0`$ for all $`t>0`$. The completed cone $`\overline{C_f(N)}=C_f(N)\{p\}`$, $`p=\{0\}\times N`$. The Euclidean space $`^m`$ with metric $`ds^2=dt^2+g^2(t)d\theta ^2`$ is the completed cone $`\overline{C_g(𝕊^{m1})}`$. The next theorem compares the fundamental tone $`\lambda ^{}(C_f(N)(r))`$ of the the trunked cone $`C_f(N)(r)=(0,r)\times N`$ with the lowest Dirichlet eigenvalue $`\lambda _1(B_^m(r))`$ of the geodesic ball $`B_^m(r)`$.
###### Theorem 3.2
Let $`C_f(N)`$ be a incomplete cone over a compact $`(n1)`$-dimensional Riemannian manifold $`(N,dh^2)`$ and $`^m`$ with metric $`ds^2=dt^2+g^2(t)d\theta ^2`$. If
$$(n1)(f^{}/f)(t,x)(m1)(g^{}/g)(t),$$
(22)
for all $`xN`$ and all $`t(0,r)`$ where means the derivative with respect to the variable $`t`$. Then
$$\lambda ^{}(C_f(N)(r))\lambda _1(B_^n(r))$$
(23)
If (22) holds for all $`t>0`$ then letting $`r\mathrm{}`$ we have that
$$\lambda ^{}(C_f(N))\lambda ^{}(^m)$$
The proof of Theorem (3.2) is similar to the proof of Theorem (1.1). We take $`u`$ to be a positive first Dirichlet eigenfunction of $`B_^n(r)`$ and consider the vector fields $`X_1(t,x)={\displaystyle \frac{u^{}}{u}}(t){\displaystyle \frac{_{\mathrm{\hspace{0.17em}1}}}{t}}(t,x)`$ and $`X_2(t,\theta )={\displaystyle \frac{u^{}}{u}}(t){\displaystyle \frac{_{\mathrm{\hspace{0.17em}2}}}{t}}(t,\theta )`$. Thus we have by Theorem (2.3) that
$$\lambda ^{}(C_f(N)(r))inf[\mathrm{div}X_1|X_1|^2]=inf[\mathrm{div}X_1\mathrm{div}X_2]+\lambda _1(B_^n(r)).$$
(24)
Observe that the slice $`t\times N`$, $`\{t\}(0,r)`$ is a smooth hypersurface of $`C_f(N)(r)`$ thus we may apply Lemma (2.4) to obtain that $`\mathrm{div}X_1(t,x)\mathrm{div}X_2(t,\theta )=(n1){\displaystyle \frac{f^{}}{f}}(t,x)(m1){\displaystyle \frac{g^{}}{g}}(t)0`$. This together with (24) proves (23).
These ideas used in the proofs of theorems (1.1, 3.1, 3.2) can be used to obtain examples of Riemannian manifolds $`M`$ with arbitrary fundamental groups and variable sectional curvatures and with positive fundamental tone $`\lambda ^{}(M)>0`$. For instance, let $`M=^m\times N`$ with the metric $`ds^2=dt^2+f^2(t,\theta )d\theta ^2+g^2(t,\theta )dh^2`$ where $`(N,dh^2)`$ is a complete $`n`$-dimensional Riemannian manifold and $`f,g:^n[0,\mathrm{})`$ are smooth functions, $`f`$ satisfying $`f(0,\theta )=0,f^{}(0,\theta )=1`$ and $`f(t,\theta )>0`$ for $`t>0`$ and $`\theta 𝕊^{m1}`$, $`g(t,\theta )>0`$ for all $`(t,\theta )[0,\mathrm{})\times 𝕊^{m1}`$. Let $`\mathrm{\Omega }=B_^m(r)\times WM`$ where $`B_^m(r)^m`$ is a ball with radius $`r`$ nd $`WN`$ is a domain with compact closure and smooth boundary $`W`$ (possibly empty). Let $`B_{𝕄^l(\kappa )}(r)𝕄^l(\kappa )`$ be a geodesic ball of radius $`r`$ in the simply connected $`l`$-dimensional space form of constant sectional curvature $`\kappa `$ with metric $`dt^2+S_\kappa ^{\mathrm{\hspace{0.17em}2}}(t)d\theta ^2`$.
###### Theorem 3.3
If $`(m1){\displaystyle \frac{f^{}}{f}}(t,\theta )+n{\displaystyle \frac{g^{}}{g}}(t,\theta )(l1){\displaystyle \frac{S_\kappa ^{}}{S_\kappa }}(t)`$ for all $`t[0,r]`$ and $`\theta 𝕊^{m1}`$, then
$$\lambda _1(\mathrm{\Omega })\lambda _1(B_{𝕄^l(\kappa )}(r))+\underset{(t,\theta )\mathrm{\Omega }}{inf}[\frac{1}{g^2}]\lambda _1(W).$$
(25)
If $`r=\mathrm{}`$ and letting $`W=N`$ we have that
$$\lambda ^{}(M)(l1)^2\kappa ^2/4+\underset{t,\theta }{inf}[\frac{1}{g^2}]\lambda ^{}(N).$$
If $`(m1){\displaystyle \frac{f^{}}{f}}(t,\theta )+n{\displaystyle \frac{g^{}}{g}}(t,\theta )(l1){\displaystyle \frac{S_\kappa ^{}}{S_\kappa }}(t)`$ for all $`t[0,r]`$ and $`\theta 𝕊^{m1}`$, then
$$\lambda _1(\mathrm{\Omega })\lambda _1(B_{𝕄^l(\kappa )}(r))+\underset{(t,\theta )\mathrm{\Omega }}{sup}[\frac{1}{g^2}]\lambda _1(W).$$
(26)
If $`r=\mathrm{}`$, and letting $`W=N`$ we have that
$$\lambda ^{}(M)(l1)^2\kappa ^2/4+\underset{(t,\theta )}{sup}[\frac{1}{g^2}]\lambda ^{}(N)$$
Choose a positive function $`\psi :\mathrm{\Omega }`$ given by $`\psi (t,\theta ,x)=u(t)\xi (x)`$ where $`u`$ and $`\xi `$ are positive eigenfunctions of $`B_{𝕄^l(\kappa )}(r)`$ and $`W`$ respectively, i.e. $`u`$ satisfies the differential equation
$$\frac{^{\mathrm{\hspace{0.17em}2}}u}{t^2}(t)+(l1)\frac{S_k^{}}{S_k}(t)\frac{u}{t}(t)+\lambda _1(B_{𝕄^l(\kappa )}(r))u(t)=0$$
(27)
with $`u(0)=1`$, $`u^{}(0)=0`$ and $`\xi :W`$ satisfies $`\mathrm{}_{dh^2}\xi +\lambda _1(W)\xi =0`$ in $`W`$ and $`\xi |W=0`$. It is clear that $`\psi C^2(\mathrm{\Omega })C^0(\overline{\mathrm{\Omega }_2})`$ with $`\psi >0`$ in $`\mathrm{\Omega }`$ and $`\psi |\mathrm{\Omega }=0`$. The Laplace operator of $`ds^2`$ is written in geodesic coordinates is given by
$`\mathrm{}_{ds^2}`$ $`=`$ $`{\displaystyle \frac{^2}{t^2}}+[(m1){\displaystyle \frac{1}{f}}{\displaystyle \frac{f}{t}}+n{\displaystyle \frac{1}{g}}{\displaystyle \frac{g}{t}}]{\displaystyle \frac{}{t}}+{\displaystyle \frac{m3}{f^3}}d\theta ^2(_{d\theta ^2}f,_{d\theta ^2})`$ (28)
$`+`$ $`{\displaystyle \frac{n}{gf^2}}d\theta ^2(_{d\theta ^2}g,_{d\theta ^2})+{\displaystyle \frac{1}{f^2}}\mathrm{}_{d\theta ^2}+{\displaystyle \frac{1}{g^2}}\mathrm{}_{dh^2}`$
where $`_{d\theta ^2}`$ and $`\mathrm{}_{d\theta ^2}`$ are respectively the gradients and the Laplacian of $`𝕊^{m1}`$ and $`_{dh^2}`$ and $`\mathrm{}_{dh^2}`$ are respectively the gradients and the Laplacian of $`N`$. Computing $`\mathrm{}_{ds^2}\psi /\psi `$ we have,
$`{\displaystyle \frac{\mathrm{}_{ds^2}\psi }{\psi }}`$ $`=`$ $`{\displaystyle \frac{u^{\prime \prime }}{u}}+(m1){\displaystyle \frac{f^{}}{f}}{\displaystyle \frac{u^{}}{u}}+n{\displaystyle \frac{g^{}}{g}}{\displaystyle \frac{u^{}}{u}}{\displaystyle \frac{1}{g^2}}{\displaystyle \frac{\mathrm{}_{dh^2}\xi }{\xi }}`$ (29)
$`=`$ $`\lambda _1(B_{𝕄^s(\kappa )}(r)){\displaystyle \frac{u^{}}{u}}\left((m1){\displaystyle \frac{f^{}}{f}}+n{\displaystyle \frac{g^{}}{g}}(s1){\displaystyle \frac{S_k^{}}{S_k}}\right)+{\displaystyle \frac{1}{g^2}}\lambda _1(W)`$
If $`(m1){\displaystyle \frac{f^{}}{f}}+n{\displaystyle \frac{g^{}}{g}}(s1){\displaystyle \frac{S_k^{}}{S_k}}`$ then from (29)
$$inf(\frac{\mathrm{}\psi }{\psi })\lambda _1(B_{𝕄^s(\kappa )}(r))+inf\frac{1}{g^2}\lambda _1(W).$$
If $`(m1){\displaystyle \frac{f^{}}{f}}+n{\displaystyle \frac{g^{}}{g}}(s1){\displaystyle \frac{S_k^{}}{S_k}}0`$ then
$$sup(\frac{\mathrm{}\psi }{\psi })\lambda _1(B_{𝕄^s(\kappa )}(r))+sup\frac{1}{g^2}\lambda _1(W).$$
Since $`(u^{}/u)0`$.
## 4 Examples
In this section we construct examples of metrics showing certain aspects of Cheng’s eigenvalue comparison theorem. In this first example we construct a family of metrics on $`^n`$ with radial sectional $`K_^n>\kappa `$ outside a compact set and such that the mean curvatures of the distance spheres satisfy $`H_^n(t,\theta )H_{𝕄^n(\kappa )}(t,\theta )=(n1)(S_\kappa ^{}/S_\kappa )(t)`$.
###### Example 4.1
Let $`^n=[0,\mathrm{})\times 𝕊^{n1}`$ with the metric $`ds^2=dt^2+f^2(t)d\theta ^2`$, $`f(0)=0`$, $`f^{}(0)=1`$. Set $`\psi _\kappa (t)=(f^{}S_\kappa +fS_\kappa ^{})(t)`$, where means differentiation with respect to $`t`$ and $`S_\kappa `$ is given by
$$\begin{array}{cc}S_\kappa (t)=\{\begin{array}{ccc}\mathrm{sinh}(\sqrt{}\kappa t)/\sqrt{}\kappa \hfill & if\hfill & \kappa =k^2\hfill \\ t\hfill & if\hfill & \kappa =0\hfill \\ \mathrm{sin}(\sqrt{\kappa }t)/\sqrt{\kappa }\hfill & if\hfill & \kappa =k^2\hfill \end{array},\hfill & C_\kappa (t)=S_\kappa ^{}(t)\hfill \end{array}$$
(30)
The radial sectional curvature of $`(^n,ds^2)`$ is bounded above by $`\kappa `$ if and only if $`\psi _\kappa ^{}(t)0`$. The mean curvatures of $`B_^n(t)`$ and $`B_{𝕄^n(\kappa )}(t)`$ satisfies $`H_^n(t,\theta )H_{𝕄^n(\kappa )}(t)`$ if and only if $`\psi _\kappa (t)0`$. From $`\psi _\kappa (t)=(f^{}S_\kappa +fS_\kappa ^{})(t)`$ we have that $`\psi _\kappa (0)=\psi _\kappa ^{}(0)=0`$. Solving the differential equation we have
$$f(t)=S_\kappa (t)+S_\kappa (t)_0^t\psi _\kappa (s)/S_\kappa (s)𝑑s$$
Let $`\psi _\kappa :[0,\mathrm{})`$ be a smooth function satisfying $`\psi _\kappa (0)=\psi _\kappa ^{}(0)=0`$, $`\psi (t)0`$, $`\psi _\kappa ^{}(t)>0`$ for $`t>1`$ and $`|_0^t\psi _\kappa (s)/S_\kappa (s)𝑑s|<\mathrm{}`$. This yields a metric $`ds_f^2=dt^2+f^2(t)d\theta `$ with sectional curvature $`K_^n>\kappa `$ outside a compact set and such that the mean curvatures of the distance spheres satisfy $`(n1)(f^{}/f)(t)=H_^n(t,\theta )H_{(𝕄^n(\kappa )}(t,\theta )=(n1)(S_\kappa ^{}/S_\kappa )(t)`$.
###### Remark 4.2
If the metric of $`M`$ is expressed by $`dt^2+f^2(t)d\theta ^2`$ then $`H_M(s,\theta )=H_{𝕄(\kappa )}(s)`$ for all $`s(0,r]`$ and all $`\theta 𝕊^{n1}`$ implies that $`B_M(r)`$ is isometric to $`B_{𝕄(\kappa )}(r)`$. Because the equality $`H_M(s,\theta )=H_{𝕄(\kappa )}(s)`$ for all $`s(0,r]`$ and all $`\theta `$ is equivalent to have $`\psi _\kappa (s)=0`$, $`s[0,r]`$ but this would imply that $`f(s)=S_\kappa (s)`$, $`s[0,r]`$.
The next example shows that the rigidity in Theorem (1.1) is all we can expect without curvature bounds.
###### Example 4.3
For every $`\kappa `$, consider the metric $`g=g(\kappa )`$ on $`M=[0,a]\times 𝕊^{n1}`$, where $`a=\mathrm{}`$ if $`k0`$ and $`a=\pi /\sqrt{\kappa }`$ if $`\kappa >0`$, given in geodesic coordinates by the matrix $`g_{11}(t,\theta )=1`$, $`g_{22}(t,\theta )=(S_\kappa ^4(t)/t^2)\theta _{22}`$, $`g_{33}(t,\theta )=t^2\theta _{33}`$, $`g_{ii}(t,\theta )=S_\kappa ^2(t)\theta _{ii}`$, $`i4`$, $`g_{ij}(t,\theta )=0`$ if $`ij`$, where $`d\theta _{ij}^2=(\theta _{ij})`$ is the canonical metric of $`𝕊^{n1}(1)`$. This metric $`g(\kappa )`$ is smooth if $`\kappa 0`$. If $`\kappa >0`$ the metric $`g(\kappa )`$ is smooth except at $`(\pi ,\theta )`$. Let $`h=h(\kappa )`$ be the metric of constant sectional curvature of $`𝕄(\kappa )`$ given by the matrix $`h_{11}=1`$, $`h_{ii}=S_\kappa ^2(t)\theta _{ii}`$, $`i2`$. For $`\kappa 0`$, $`g(\kappa )`$ is not isometric to $`h(\kappa )`$. Let $`\mathrm{}_g`$ and $`\mathrm{}_h`$ denote the Laplace operator of these two metrics written in geodesic coordinates. They are given by
$`\mathrm{}_g`$ $`=`$ $`{\displaystyle \frac{^2}{t^2}}+(n1){\displaystyle \frac{C_\kappa }{S_\kappa }}{\displaystyle \frac{}{t}}+{\displaystyle \frac{t^2}{S_\kappa ^4}}{\displaystyle \frac{}{\theta _2}}+{\displaystyle \frac{1}{t^2}}{\displaystyle \frac{}{\theta _3}}+{\displaystyle \underset{i=4}{\overset{n}{}}}{\displaystyle \frac{1}{S_\kappa ^2}}{\displaystyle \frac{}{\theta _i}}`$
$`\mathrm{}_h`$ $`=`$ $`{\displaystyle \frac{^2}{t^2}}+(n1){\displaystyle \frac{C_\kappa }{S_\kappa }}{\displaystyle \frac{}{t}}+{\displaystyle \underset{i=2}{\overset{n}{}}}{\displaystyle \frac{1}{S_\kappa ^2}}{\displaystyle \frac{}{\theta _i}}.`$
We have that the geodesic spheres $`B_M(s)`$ and $`B_{𝕄(\kappa )}(s)`$ have the same mean curvature $`H_M(s)=H_{𝕄(\kappa )}(s)=(n1)(C_\kappa /S_\kappa )(s)`$, $`s(0,r]`$. And the geodesic balls $`B_M(r)`$ and $`B_{𝕄(\kappa )}(r)`$ have the same first eigenvalue. For if $`u`$ be a first Dirichlet eigenfunction of the geodesic ball $`B_{𝕄(\kappa )}(r)`$, if $`\kappa >0`$ suppose that that $`r<\pi /\sqrt{\kappa }`$. Thus $`\mathrm{}_hu+\lambda _1(B_{𝕄(\kappa )}(r))u=0`$ in $`B_{𝕄(\kappa )}(r)`$ and $`u=0`$ on $`B_{𝕄(\kappa )}(r)`$. Since $`u`$ is radial we have that $`\mathrm{}_hu(t)=\mathrm{}_gu(t)=\lambda _1(B_{𝕄(\kappa )}(r))u(t)`$. This shows that $`u(t)`$ is a first Dirichlet eigenfunction of the geodesic ball $`B_M(r)`$ with same eigenvalue $`\lambda _1(B_{𝕄(\kappa )}(r)).`$ |
warning/0507/math0507499.html | ar5iv | text | # Symplectic forms on six dimensional real solvable Lie algebras I
## 1 Introduction
Symplectic forms on Lie groups and algebras naturally appear in the context of Poisson geometry, the study of Hamiltonians in Mechanics and various other physical and geometrical problems, like the cotangent bundle of differentiable manifolds. Symplectic forms are also useful to construct other geometrical structures, like Kähler manifolds, in the case of combination of symplectic and compatible complex structures . Many works have been devoted to obtain conditions for constructing and classifying symplectic structures on manifolds, groups or algebras, and although no universal characterization has been obtained yet, various general procedures of interest have been obtained . Models for Lie algebras admitting symplectic forms have been developed in , which is related to the problem of determining the structure of Lie groups having symplectic forms. Various reductions have been obtained in this sense, which simplify the problem to the analysis of solvable Lie algebras and algebras with nontrivial Levi decomposition and a solvable non-nilpotent radical . For fixed dimensions there only exist complete results up to dimension four, as well as some special cases in higher dimension . Symplectic structures on four dimensional real Lie algebras have been classified in different contexts (see e.g. or for a recent review), and interesting applications to the Monge-Ampère equations were developed in . In dimension six, only the nilpotent Lie algebras have been systematically analized for symplectic forms , in combination with additional geometrical structures (see e.g. and references therein). Results for nilpotent Lie algebras of maximal nilindex have been obtained in , while Lie algebras in dimension $`n8`$ endowed with exact symplectic forms were determined in . This special case of symplectic forms is closely related to the problem of determining the invariant functions for the coadjoint representation of Lie groups.
In this work we begin with the systematic computation of symplectic structures on real solvable Lie algebras of dimension six. We focus only on solvable non-nilpotent Lie algebras, basing on the classification obtained by various authors. More precisely, we obtain the possible symplectic forms on six dimensional solvable Lie algebras that either decompose as the direct sum of two lower dimensional ideals or are indecomposable with a four dimensional nilradical. This covers all but one case, corresponding to indecomposable algebras with five dimensional nilradical.
Unless otherwise stated, any Lie algebra $`𝔤`$ considered in this work is defined over the field $``$ of real numbers. We convene that nonwritten brackets are either zero or obtained by antisymmetry. Abelian Lie algebras of dimension $`n`$ are denoted by $`nL_1`$.
## 2 Symplectic structures on Lie groups
Given a Lie algebra $`𝔤`$ with structure tensor $`\left\{C_{ij}^k\right\}`$ over a basis $`\{X_1,..,X_n\}`$, the identification of the dual space $`𝔤^{}`$ with the left-invariant Pfaffian forms on a Lie group whose algebra is isomorphic to $`𝔤`$ allows to define an exterior differential $`d`$ on $`𝔤^{}`$ by
$$d\omega (X_i,X_j)=C_{ij}^k\omega \left(X_k\right),\omega 𝔤^{}.$$
(1)
Therefore we can rewrite any Lie algebra $`𝔤`$ as a closed system of $`2`$-forms
$$d\omega _k=C_{ij}^k\omega _i\omega _j,\mathrm{\hspace{0.33em}1}i<jdim\left(𝔤\right),$$
(2)
called the Maurer-Cartan equations of $`𝔤`$. The closure condition $`d^2\omega _i=0`$ for all $`i`$ is equivalent to the Jacobi condition. Let $`(𝔤)=\left\{d\omega _i\right\}_{1idim𝔤}`$ be the linear subspace of $`^2𝔤^{}`$ generated by the $`2`$-forms $`d\omega _i`$. It follows at once that $`dim(𝔤)=dim\left(𝔤\right)`$ if and only if $`d\omega _i0`$ for all $`i`$, that is, if $`dim\left(𝔤\right)=dim[𝔤,𝔤]`$ holds. If $`\omega =a^id\omega _i\left(a^i\right)`$ is an element of $`(𝔤)`$, there exists a positive integer $`j_0\left(\omega \right)`$ such that
$$\stackrel{j_0\left(\omega \right)}{}\omega 0,\stackrel{j_0\left(\omega \right)+1}{}\omega 0.$$
(3)
This equation shows that $`r\left(\omega \right)=2j_0\left(\omega \right)`$ is the rank of the 2-form $`\omega `$. Define
$$j_0\left(𝔤\right)=\mathrm{max}\left\{j_0\left(\omega \right)|\omega (𝔤)\right\}.$$
(4)
This quantity $`j_0\left(𝔤\right)`$ depends only on the structure of $`𝔤`$ and is a numerical invariant of $`𝔤`$ .
An even dimensional Lie group $`G`$ is said to carry a left invariant symplectic structure if it possesses a left invariant closed $`2`$-form $`\omega `$ of maximal rank. At the Lie algebra level, this implies the existence of the form $`\omega ^2𝔤^{}`$ such that
$`d\omega =0`$ (5)
$`{\displaystyle \stackrel{n}{}}\omega 0,`$ (6)
where $`2n=dim(𝔤)`$. We say that $`𝔤`$ is endowed with a symplectic structure. For example, any Lie algebra in dimension 2 has a symplectic structure. For the abelian algebra the assertion is trivial, while for the affine Lie algebra $`𝔯_2=\{X_1,X_2|[X_1,X_2]=X_2\}`$ we have $`\omega 𝔯_2^{}𝔯_2^{}`$ defined by
$$\omega =\omega _1\omega _2.$$
(7)
This form is closed and of maximal rank. In particular, $`\omega (𝔯_2)`$. Symplectic form of this special kind are called exact symplectic forms, and they are of interest in the analysis of invariants of Lie algebras . That is, a Lie algebra is exact symplectic if it is symplectic and the form is moreover exact.
The structure of a Lie algebra plays an essential role in the existence of such forms. We briefly recall some results which will be used in this work.
###### Proposition 1
Let $`𝔤`$ be a Lie algebra. Then following conditions hold:
1. If $`Trace[ad(X)]=0X𝔤`$) and $`𝔤`$ is symplectic, then $`𝔤`$ is solvable.
2. No semisimple Lie algebra carries a symplectic form.
3. Direct sums of semisimple and solvable Lie algebras cannot be symplectic.
This implies that an indecomposable symplectic Lie algebra is either solvable or the semidirect product of a semisimple Lie algebra and a solvable non-nilpotent Lie algebra. In particular, any symplectic Lie algebra in dimension four is solvable. We remark that in this dimension any nilpotent Lie algebra is endowed with a symplectic form .
### 2.1 Six dimensional Lie algebras
Real Lie algebras of dimension six have been fully classified by Morozov (nilpotent real Lie algebras), Mubarakzyanov (decomposition conditions and solvable algebras with five dimensional nilradical) and Turkowski (solvable algebras with four dimensional nilradical and algebras with nontrivial Levi subalgebra) (see e.g. and references therein). According to the list in , there are four indecomposable Lie algebras with nonzero Levi subalgebra, corresponding to the semidirect product of the abelian Lie algebra $`3L_1`$ with $`𝔰𝔩(2,)`$ and $`𝔰𝔬(3)`$, the semidirect product of $`𝔰𝔩(2,)`$ and the Heisenberg algebra in dimension 3 and the Lie algebra $`𝔰𝔩(2,)\stackrel{}{}_{D_{\frac{1}{2}}D_0}A_{3,3}`$, which is the only to possess a (exact) symplectic form. The decomposable case follows from . It remains to analyze the solvable case. Nilpotent algebras endowed with symplectic forms can be found in , for which reason we do not reproduce them here. As known, the maximal nilpotent ideal (nilradical) $`NR`$ of a solvable Lie algebra $`𝔯`$ satisfies the inequality
$$dim(NR)\frac{𝔯}{2}.$$
(8)
Therefore a solvable algebra in dimension six has a nilradical of dimensions four or five if it is indecomposable, or it is the direct sum of ideals.
## 3 Decomposable solvable Lie algebras
In this section we analize the symplectic solvable Lie algebras $`𝔯`$ that decompose as the direct sum $`𝔯=𝔯_1𝔯_2`$ of lower dimensional ideals. Since any solvable Lie algebra is supposed to be non-nilpotent, at least one of the ideals $`𝔯_i`$ must be non-nilpotent.
Let $`𝔯_1`$ and $`𝔯_2`$ be solvable Lie algebras of odd dimension and let $`\{\omega _1,..,\omega _{2n+1}\}`$ and $`\{\omega _1^{},..,\omega _{2m+1}^{}\}`$ be bases of $`𝔯_1^{}`$, respectively $`𝔯_2^{}`$. Solvability implies that $`𝔯_i[𝔯_i,𝔯_i]`$ for $`i=1,2`$, so that without loss of generality we can suppose that
$$d\omega _1=d\omega _1^{}=0.$$
(9)
In these conditions, we give a sufficiency criterion for the existence of symplectic forms on the direct sum algebra $`𝔯_1𝔯_2`$:
###### Proposition 2
Suppose that there exist 2-forms $`\theta =_{i<j}\alpha ^{ij}\omega _i\omega _j^2𝔯_1^{}`$ and $`\theta ^{}=_{k<l}\beta ^{kl}\omega _k^{}\omega _l^{}^2𝔯_2^{}`$ satisfying
1. $`\alpha ^{1j}=\beta ^{1l}=0,j,l2.`$
2. $`d\theta =d\theta ^{}=0,`$
3. $`^n\theta 0`$ and $`^m\theta ^{}0.`$
Then $`𝔯_1𝔯_2`$ is endowed with a sympletic form $`\eta =\theta +\theta ^{}+\omega _1\omega _1^{}`$
Proof. Since $`\alpha ^{1j}=0`$ for any $`j`$, the 2-form $`\theta `$ can be written as
$$\theta =\underset{2i<j2n+1}{}\alpha ^{ij}\omega _i\omega _j.$$
(10)
By assumption, the $`n^{th}`$-exterior product $`^n\theta `$ is not zero, so that reordering the basis if necessary, we can suppose that
$$\underset{k=1}{\overset{n}{}}\alpha ^{2k,2k+1}0.$$
(11)
In consequence we obtain
$$\stackrel{n}{}\theta =\left(n!\underset{k=1}{\overset{n}{}}\alpha ^{2k,2k+1}\right)\omega _2\omega _3\mathrm{}\omega _{2n}\omega _{2n+1}.$$
(12)
A similar expression holds for $`\theta ^{}`$. Clearly the 2-form
$$\eta =\theta +\theta ^{}+\omega _1\omega _1^{}$$
(13)
belongs to $`\left(𝔯_1𝔯_2\right)^{}`$. Further
$$d\eta =d\theta +d\theta ^{}+d\omega _1\omega _1^{}\omega _1d\omega _1^{}=0$$
(14)
by condition 2, showing that $`\eta `$ is closed. Finally,
$$\stackrel{n+m+1}{}\eta =\left(\left(n+m+1\right)!\underset{k=1}{\overset{n}{}}\alpha ^{2k,2k+1}\underset{l=1}{\overset{m}{}}\right)\omega _1\mathrm{}\omega _{2n+1}\omega _1^{}\mathrm{}\omega _{2m+1}0,$$
(15)
showing that $`\eta `$ is of maximal rank.
The result has an interesting consequence concerning odd-dimensional solvable Lie algebras. A 1-form $`\omega 𝔯^{}`$ is called a linear contact form if
$$\omega \left(d\omega _\mu \right)^n0.$$
(16)
In particular, the left invariant Pfaff form induced by $`\omega `$ over the Lie group having $`𝔯`$ as Lie algebra is a contact form in the classical sense.
###### Corollary 1
Let $`𝔯`$ be solvable Lie algebra and $`\{\omega _1,..,\omega _{2n+1}\}`$ be a basis of $`𝔯^{}`$. Suppose that $`d\omega _1=0.`$ and that there exists a 2-form $`\theta =_{i<j}\alpha ^{ij}\omega _i\omega _j^2𝔯^{}`$ such that
1. $`\alpha ^{1j}=0,j2.`$
2. $`d\theta =0,`$
3. $`^n\theta 0`$
If $`\theta \left(𝔯\right)`$, then $`𝔯`$ is endowed with a linear contact form.
Proof. By assumption the 2-form $`\theta `$ can be written as
$$\theta =\underset{2i<j2n+1}{}\alpha ^{ij}\omega _i\omega _j.$$
(17)
Without loss of generality we can suppose that
$$\stackrel{n}{}\theta =\left(n!\underset{k=1}{\overset{n}{}}\alpha ^{2k,2k+1}\right)\omega _2\omega _3\mathrm{}\omega _{2n}\omega _{2n+1}.$$
(18)
If $`\theta \left(𝔯\right)`$, then there exist scalars $`a_{i_1},..,a_{i_k}`$ such that
$$\theta =a_{i_1}d\omega _{i_1}+..+a_{i_k}d\omega _{i_k}.$$
(19)
Further $`i_j1`$ for $`j\{1,..,k\}`$ since $`d\omega _1=0`$. Define the linear form
$$\eta =\omega _1+a_{i_1}\omega _{i_1}+..+a_{i_k}\omega _{i_k}$$
(20)
Then
$`\eta \left({\displaystyle \stackrel{n}{}}d\eta \right)=(\omega _1+a_{i_1}\omega _{i_1}+..+a_{i_k}\omega _{i_k})\left({\displaystyle \stackrel{n}{}}\theta \right)=`$
$`\left(n!{\displaystyle \underset{k=1}{\overset{n}{}}}\alpha ^{2k,2k+1}\right)\omega _1\omega _2\omega _3\mathrm{}\omega _{2n}\omega _{2n+1}0,`$ (21)
thus $`\eta `$ is a contact form.
### 3.1 $`dim𝔯_1=dim𝔯_2=3`$
There are five isomorphism classes (two of them depending on parameters) of indecomposable solvable Lie algebras in dimension 3, whose Maurer-Cartan equations are given in Table 4 of the appendix. From these algebras, only one is nilpotent (the Heisenberg algebra). We have therefore 14 solvable non-nilpotent algebras $`𝔯`$ which decompose as the direct sum of two three dimensional ideals. To determine the possible symplectic forms on these algebras, we use proposition 2. The resulting algebras admitting such structures are given in table 1.
### 3.2 $`dim𝔯_4=dim𝔯_2=2`$
The case of direct sums $`𝔯=𝔯_1𝔯_2`$ with $`dim𝔯_4=dim𝔯_2=2`$ follows at once from the classification of symplectic structures on four dimensional real Lie algebras carried out in . Indeed, since any two dimensional Lie algebra has a symplectic form, the sum with any four dimensional algebra having also a symplectic form gives a six dimensional algebra. On the other hand, it is immediate that no direct sum of a four dimensional Lie algebra with no symplectic form and a two dimensional algebra can result in a six dimensional symplectic Lie algebra. Therefore the result is obtained combining the results of with those of dimension two.
### 3.3 $`dim𝔯_5=dim𝔯_2=1`$
The Maurer-Cartan equations of the indecomposable real solvable Lie algebras in dimension five are given in tables 5 and 6 of the appendix. There are 33 cases to be analyzed. By (8), the nilradical of such algebras $`𝔯`$ have dimension three or four. It is not difficult to see that a direct sum $`𝔯L_1`$ endowed with a symplectic form must satisfy the requirements of proposition 2 (otherwise the closure of the 2-form would be violated). In particular we have $`\theta ^{}=0`$, since $`dimL_1=1`$. The algebras admitting a symplectic form are listed in table 2.
## 4 Solvable Lie algebras with four dimensional nilradical
As follows from the classification in , there are 33 indecomposable solvable non-nilpotent real Lie algebras in dimension six with a four dimensional nilradical. The corresponding Maurer-Cartan equations for these algebras are listed in tables 7 and 8 of the appendix. The search for symplectic structures in this case must be developed case by case, since the algebras are indecomposable.
###### Proposition 3
Let $`𝔯`$ be an indecomposable solvable non-nilpotent real Lie algebra of dimension six. Then $`𝔯`$ is endowed with a symplectic form $`\omega `$ if and only if it is isomorphic to one of the Lie algebras in table 3.
Proof. We give the detailed proof for $`N_{6,1}^{\alpha \beta \gamma \delta }`$, all the remaining cases are treated in a similar way.
The brackets over the basis $`\{N_1,..,N_4,X_1,X_2\}`$ are given by
$`[X_1,N_1]`$ $`=\alpha N_1,[X_1,N_2]=\gamma N_2,[X_1,N_4]=N_4`$ (22)
$`[X_2,N_1]`$ $`=\beta N_1,[X_2,N_2]=\delta N_2,[X_2,N_3]=N_3,`$ (23)
where $`\alpha ,\beta ,\gamma ,\delta `$ satisfy the restrictions $`\alpha \beta 0`$ and $`\gamma ^2+\delta ^20`$. Taking the dual basis $`\{\eta _1,..,\eta _4,\omega _1,\omega _2\}`$, the Maurer-Cartan equations of the algebra are
| $`d\eta _1=\alpha \omega _1\eta _1+\beta \omega _2\eta _1,`$ | $`d\eta _2=\gamma \omega _1\eta _2+\delta \omega _2\eta _2,`$ |
| --- | --- |
| $`d\eta _3=\omega _2\eta _3,`$ | $`d\eta _4=\omega _1\eta _4,`$ |
| $`d\omega _1=0,`$ | $`d\omega _2=0.`$ |
Now define an element $`\omega ^2\left(N_{6,1}^{\alpha \beta \gamma \delta }\right)^{}`$ in general position by
$$\omega =a^{ij}\eta _i\eta _j+b^{ik}\eta _i\omega _k+c^{12}\omega _1\omega _2,$$
(24)
where $`1i<j4,k=1,2`$ and $`a^{ij},b^{ik},c^{12}`$. If we impose the closure
$$d\omega =a^{ij}d\eta _i\eta _j+b^{ik}d\eta _i\omega _ka^{ij}\eta _id\eta _j=0$$
(25)
and take into account the relations satisfied by the parameters, then the following coefficients vanish independently of their value $`\alpha ,\beta ,\gamma `$ and $`\delta :`$
$$a^{13}=a^{14}=a^{34}=b^{31}=b^{42}=0,$$
(26)
and since $`\alpha \beta 0`$,
$$b^{12}=\frac{\beta }{\alpha }b^{11}.$$
(27)
We further obtain the following expression for the differential:
$`d\omega =a^{24}\delta \omega _2\eta _2\eta _4+a^{23}\left(1+\delta \right)\omega _2\eta _2\eta _3+a^{12}\left(\alpha +\gamma \right)\omega _1\eta _1\eta _2+`$
$`\left(\gamma b^{22}\delta b^{21}\right)\omega _1\eta _2\omega _2+a^{24}\left(1+\gamma \right)\omega _1\eta _2\eta _4+a^{23}\gamma \omega _1\eta _2\eta _3+`$
$`+a^{12}\left(\beta +\delta \right)\omega _2\eta _1\eta _2=0.`$ (28)
At this stage, the analysis must be divided into several steps, according to the different possibilities depending on the four parameters:
1. Let $`\gamma 0`$. Then the closure (28) implies
| $`a^{23}=0,`$ | $`b^{22}=\frac{\delta }{\gamma }b^{21},`$ |
| --- | --- |
| $`\delta a^{24}=0,`$ | $`a^{12}\left(\alpha +\gamma \right)=0,`$ |
| $`a^{24}\left(1+\gamma \right)=0,`$ | $`a^{12}\left(\beta +\delta \right)=0.`$ |
(29)
and we obtain the wedge product
$$\stackrel{3}{}\omega =6b^{32}(a^{12}b^{41}+b^{11}a^{24})\eta _1..\eta _4\omega _1\omega _2$$
(30)
1. If $`\delta =0`$, then $`a^{12}=0`$ since $`\beta `$ is nonzero, and the wedge product $`^3\omega `$ is nonzero if and only if
$$1+\gamma =0.$$
(31)
Therefore only the Lie algebra $`N_{6,1}^{\alpha ,\beta ,1,0}`$ has a symplectic form, given by
$`\omega =b^{11}\eta _1\omega _1+{\displaystyle \frac{\beta }{\alpha }}b^{11}\eta _1\omega _2+a^{24}\eta _2\eta _4+b^{21}\eta _2\omega _1+b^{32}\eta _3\omega _2+b^{41}\eta _4\omega _1+`$
$`+c^{12}\omega _1\omega _2={\displaystyle \frac{b^{11}}{\alpha }}d\eta _1b^{21}d\eta _2b^{32}d\eta _3b^{41}d\eta _4+a^{24}\eta _2\eta _4+c^{12}\omega _1\omega _2.`$ (32)
2. If $`\delta 0`$, then by (28) we have $`a^{24}=0`$. We obtain that
$$\stackrel{3}{}\omega =6b^{32}\left(a^{12}b^{41}\right)\eta _1..\eta _4\omega _1\omega _20.$$
(33)
This product is different from zero if and only if
$$\alpha +\gamma =\beta +\delta =0.$$
(34)
In this case the Lie algebra $`N_{6,1}^{\alpha ,\beta ,\alpha ,\beta }`$ has the symplectic form
$`\omega =a^{13}\eta _1\eta _2+b^{11}\left(\eta _1\omega _1+{\displaystyle \frac{\beta }{\alpha }}\eta _1\omega _2\right)+b^{21}\left(\eta _2\omega _1+{\displaystyle \frac{\beta }{\alpha }}\eta _2\omega _2\right)+b^{32}\eta _3\omega _2+`$
$`+b^{41}\eta _4\omega _1+c^{12}\omega _1\omega _2=a^{12}\eta _1\eta _2{\displaystyle \frac{b^{11}}{\alpha }}d\eta _1{\displaystyle \frac{b^{21}}{\alpha }}d\eta _2b^{32}d\eta _3b^{41}d\eta _4+c^{12}\omega _1\omega _2.`$
2. Let $`\gamma =0:`$ then $`\delta 0`$ and the closure implies
$`a^{12}`$ $`=0`$
$`a^{24}`$ $`=0,`$
$`a^{23}\left(1+\delta \right)`$ $`=0.`$
In addition
$$\stackrel{3}{}\omega =6\frac{\beta }{\alpha }b^{11}a^{23}b^{41}\eta _1..\eta _4\omega _1\omega _2.$$
(35)
It is nonzero if and only if $`1+\delta =0`$, and in this case the Lie algebra $`N_{6,1}^{\alpha ,\beta ,0,1}`$ has the symplectic form
$`\omega `$ $`=b^{11}\eta _1\omega _1+{\displaystyle \frac{\beta }{\alpha }}b^{11}\eta _1\omega _2+a^{23}\eta _2\eta _3+b^{22}\eta _2\omega _2+b^{32}\eta _3\omega _2+b^{41}\eta _4\omega _1+c^{12}\omega _1\omega _2`$
$`={\displaystyle \frac{b^{11}}{\alpha }}d\eta _1b^{22}d\eta _2b^{32}d\eta _3b^{41}d\eta _4+a^{23}\eta _2\eta _3+c^{12}\omega _1\omega _2.`$
Resuming, the Lie algebras $`N_{6,1}^{\alpha ,\beta ,\gamma ,\delta }`$ admit symplectic (and non-exact) forms if
$$(\gamma ,\delta )\{(0,,1),(1,0),(\alpha ,\beta )\}.$$
The parametrs $`\alpha `$ and $`\beta `$ are not subjected to further constraints.
### Acknowledgment
The author expresses his gratitude to M. Goze and A. Medina for useful comments. During the preparation of this work, the author was supported by a research project PR1/05-13283 of the U.C.M..
## Appendix.
In this appendix we give the Maurer-Cartan equations of the indecomposable solvable non-nilpotent Lie algebras in dimensions three and five, and those of dimension six having a four dimensional nilradical.
The notation and indices for the three dimensional Lie algebras correspond to those given in . In particular, the nilpotent Heisenberg Lie algebra $`A_{3,1}`$ has been included (see section 3.1). The notation for the five dimensional solvable Lie algebras has been taken from , while the list of six dimensional solvable Lie algebras with four dimensional nilradical has been adapted from .
1. For the three and five dimensional Lie algebras $`\{\omega _1,\omega _2,\omega _3\}`$, respectively $`\{\omega _1,..,\omega _5\}`$ denote the dual bases of the algebra. In particular $`A_{3,1}`$ is nilpotent, and has been included for technical purposes.
2. For the six dimensional Lie algebras with four dimensional nilradical, $`\{\eta _1,..,\eta _4\}`$ denotes the dual basis of the nilradical, while $`\{\omega _1,\omega _2\}`$ is a dual basis of the space of nil-independent elements (i.e., linearly independent non-nilpotent derivations of the nilradical). For all these Lie algebras $`d\omega _i=0`$.
3. The restrictions on the parameters of the Lie algebras have been indicated after the Maurer-Cartan equations. We remark that some authors alter the numbering of the isomorphism classes in references for special values of the parameters. |
warning/0507/math0507130.html | ar5iv | text | # A Relative Laplacian spectral recursion
## 1. Introduction
There are two good reasons to extend the Laplacian spectral recursion from simplicial complexes to relative simplicial pairs.
The spectral recursion for simplicial complexes expresses the eigenvalues of the combinatorial Laplacian $`^{}+^{}`$ of a simplicial complex $`\mathrm{\Delta }`$ in terms of the eigenvalues of its deletion $`\mathrm{\Delta }e`$, contraction $`\mathrm{\Delta }/e`$, and an “error term” $`(\mathrm{\Delta }e,\mathrm{\Delta }/e)`$. This recursion does not hold for all simplicial complexes, but does hold for independence complexes of matroids and shifted simplicial complexes . In each case, the deletion and contraction are again matroids or shifted complexes, respectively, but the error term is only a relative simplicial pair of the appropriate kind of complexes. Being able to apply the recursion to relative simplicial pairs, such as the error term, would make the spectral recursion truly recursive.
A more compelling reason comes from duality, the idea that a Boolean algebra looks the same upside-down as it does right-side-up. Many operations preserve the property of satisfying the spectral recursion , but the dual $`\mathrm{\Delta }^{}`$ (see equation (1)) of simplicial complex $`\mathrm{\Delta }`$, which is an order filter instead of a simplicial complex, satisfies only a slightly modified version of the spectral recursion when $`\mathrm{\Delta }`$ satisfies the spectral recursion \[2, Theorem 6.3\]. Relative simplicial pairs include both simplicial complexes and order filters as special cases, and so suggest a way to unify the two versions of the spectral recursion.
Furthermore, the Laplacian itself is self-dual (Section 3), and so we will state and prove most of our results in self-dual form. The first step is to think of relative simplicial pairs as intervals in the Boolean algebra of subsets of the set of vertices, since the dual of an interval is again an interval, in a very natural way. To further emphasize this symmetry, we represent these intervals by vertically symmetric capital Greek letters, such as $`\mathrm{\Phi }`$ and $`\mathrm{\Theta }`$. When we extend the spectral recursion from simplicial complexes to intervals, the ideas of deletion and contraction generalize easily and naturally. But, even with duality as a guide, it is not as clear what should replace $`(\mathrm{\Delta }e,\mathrm{\Delta }/e)`$ as the error term.
The answer turns out to be to remove from $`\mathrm{\Phi }`$ all the pairs $`\{F,F\dot{}e\}`$ in $`\mathrm{\Phi }`$. This simple operation, which we will call the reduction of interval $`\mathrm{\Phi }`$ with respect to $`e`$, and denote by $`\mathrm{\Phi }||e`$, has a few remarkable (but easy to prove) properties that will allow us to show that it is the correct error term. To start, it is clear that this operation is self-dual, which goes nicely with deletion and contraction being more or less duals of one another. Somewhat more surprising is that $`\mathrm{\Phi }||e`$ is still an interval, albeit in two separate components (Lemma 2.4 and Proposition 2.5). Finally, it is necessary for the error term to have the same homology as $`\mathrm{\Phi }`$ itself (see Lemma 3.3), and $`\mathrm{\Phi }||e`$ satisfies this as well (equation (2)). Perhaps reduction deserves further investigation, beyond Laplacians, since it is easy to compute, preserves homology, and produces a smaller interval. (Reduction is a special case of collapsing induced by a discrete Morse function coming from an acyclic, or Morse, matching, $`FF\dot{}e`$, for all possible $`F`$; see .)
Of course, the most important evidence that reduction is the right answer is that the spectral recursion for intervals, with $`\mathrm{\Phi }||e`$ as the error term (equation (3)), holds for a variety of intervals. We are able to prove (Theorem 5.12) that it does hold for shifted intervals, that is, relative simplicial pairs of complexes, each of which is shifted on the same ordered vertex set. The analogue for matroids would be relative simplicial pairs of matroids connected by a strong map, and here our success is more limited. Although experimental evidence supports the conjecture that the spectral recursion holds for all such pairs (Conjecture 6.3), we are only able to prove it in the case where the difference in ranks between the matroids is 1 (Theorem 6.2). This does at least provide strong evidence that $`\mathrm{\Phi }||e`$ is the correct error term. Further evidence is that the property of satisfying the spectral recursion is closed under many operations on intervals (Section 3), including duality (Proposition 3.7).
We formally define intervals and their operations, including reduction, in Section 2. We review Laplacians and introduce the spectral recursion for intervals in Section 3. Our main results, that skeleta preserve the property of satisfying the spectral recursion (Theorem 4.7), and that shifted intervals and certain matroid pairs satisfy the spectral recursion (Theorems 5.12 and 6.2), are the foci of Sections 4, 5, and 6, respectively.
## 2. Intervals
In this section, we formally define intervals, and extend many simplicial complex operations to intervals. We also introduce the reduction operation ($`\mathrm{\Phi }||e`$), and establish some of its properties.
###### Definition.
Let $`2^E`$ denote the Boolean algebra of subsets of finite set $`E`$. We will say $`\mathrm{\Phi }2^E`$ is an interval if $`FGH`$ and $`F,H\mathrm{\Phi }`$ together imply $`G\mathrm{\Phi }`$. We will call the set $`E`$ the ground set of $`\mathrm{\Phi }`$, individual members of $`E`$ the vertices of $`\mathrm{\Phi }`$, and members of $`\mathrm{\Phi }`$ the faces of $`\mathrm{\Phi }`$. Note that $`v`$ may be a vertex of interval $`\mathrm{\Phi }`$ without being in any face of $`\mathrm{\Phi }`$. In this case we call $`v`$ a loop of $`\mathrm{\Phi }`$. (This is in analogy to a loop of a matroid.)
An “interval” could be similarly defined on any partially ordered set, not just $`(2^E,)`$. Indeed, later on (Section 5), we will consider “intervals” on $`2^E`$ with respect to a different partial order. But what makes Laplacians work so well on intervals of $`(2^E,)`$ is that $`(2^E,)`$ forms a chain complex (Lemma 2.7, and the preceding disucssion). Hereinafter, the word “interval” will only refer to intervals on $`(2^E,)`$.
An important special case of an interval is a simplicial complex. As usual, $`\mathrm{\Delta }2^E`$ is a simplicial complex if $`GH`$ and $`H\mathrm{\Delta }`$ together imply $`G\mathrm{\Delta }`$. It is obvious that simplicial complexes may be defined as intervals containing the empty face $`\mathrm{}`$. Of course, our motivation runs in the oppposite direction; intervals are usually presented as pairs of simplicial complexes. If $`\mathrm{\Delta }^{}\mathrm{\Delta }`$ are a pair of simplicial complexes on the same vertex set, then the relative simplicial pair $`(\mathrm{\Delta },\mathrm{\Delta }^{})`$ is simply the set difference $`\mathrm{\Delta }\mathrm{\Delta }^{}`$. We now formally check that intervals and relative simplicial pairs represent the same objects.
###### Lemma 2.1.
Let $`\mathrm{\Phi }2^E`$. Then $`\mathrm{\Phi }`$ is an interval iff $`\mathrm{\Phi }=(\mathrm{\Delta },\mathrm{\Delta }^{})`$ for some simplicial complexes $`\mathrm{\Delta },\mathrm{\Delta }^{}`$.
###### Proof.
To prove the backwards implication, assume $`FGH`$ and $`F,H(\mathrm{\Delta },\mathrm{\Delta }^{})`$, so $`F,H\mathrm{\Delta }`$, but $`F,H\mathrm{\Delta }^{}`$. From $`GH\mathrm{\Delta }`$, we conclude $`G\mathrm{\Delta }`$, but from $`FG`$ and $`F\mathrm{\Delta }^{}`$ we conclude $`G\mathrm{\Delta }^{}`$. Thus $`G(\mathrm{\Delta },\mathrm{\Delta }^{})`$ as desired.
To prove the forwards implication, let $`\mathrm{\Delta }=\{GE:GH\mathrm{for}\mathrm{some}H\mathrm{\Phi }\}`$, and let $`\mathrm{\Delta }^{}=\mathrm{\Delta }\mathrm{\Phi }`$. Now, if $`FG\mathrm{\Delta }`$, then $`FGH`$ for some $`H\mathrm{\Phi }`$, so $`F\mathrm{\Delta }`$, and thus $`\mathrm{\Delta }`$ is a simplicial complex. If $`FG\mathrm{\Delta }^{}`$, then $`F\mathrm{\Delta }`$, since $`FG\mathrm{\Delta }`$; but $`F\mathrm{\Phi }`$, since $`G\mathrm{\Phi }`$ and $`FGH`$ for some $`H\mathrm{\Phi }`$. Thus $`F\mathrm{\Delta }^{}`$, and so $`\mathrm{\Delta }^{}`$ is a simplicial complex. ∎
###### Example 2.2.
Let $`\mathrm{\Phi }`$ be the interval on vertex set $`\{1,\mathrm{},6\}`$ whose faces are $`\{12456,1245,1246,1356,124,135,136\}`$. (Here, we are omitting brackets and commas on individual faces, for clarity.) It is easy to check that $`\mathrm{\Phi }`$ is an interval (see also Example 2.3). The formulas in Lemma 2.1 set $`\mathrm{\Delta }`$ to be the simplicial complex with facets (maximal faces) $`\{12456,1356\}`$, and $`\mathrm{\Delta }^{}`$ the simplicial complex with facets $`\{1256,1456,2456,356,13\}`$. But we could add the face 34 to both $`\mathrm{\Delta }`$ and $`\mathrm{\Delta }^{}`$, and they would still be simplicial complexes such that $`\mathrm{\Phi }=(\mathrm{\Delta },\mathrm{\Delta }^{})`$.
Although Lemma 2.1 shows that intervals are the same as relative simplicial pairs, we will strive to put all of our results in the language of intervals rather than relative simplicial pairs. One reason is the potential difficulty in describing properties of the interval in terms of the pair of simplicial complexes which are not necessarily unique, as demonstrated in Example 2.2. Another, as alluded to in the Introduction, is to better take advantage of duality. The dual of interval $`\mathrm{\Phi }`$ on ground set $`E`$ is
(1)
$$\mathrm{\Phi }^{}=\{EF:F\mathrm{\Phi }\}.$$
It is easy to see that the dual of an interval is again an interval, and that $`\mathrm{\Phi }^{}=\mathrm{\Phi }`$.
It is also easy to see the intersection of two intervals is again an interval, but we have to be more careful with union, even with disjoint union. If $`\mathrm{\Phi }`$ and $`\mathrm{\Theta }`$ are disjoint intervals with faces $`F\mathrm{\Phi }`$ and $`G\mathrm{\Theta }`$ such that $`FG`$, then $`\mathrm{\Phi }\dot{}\mathrm{\Theta }`$, the disjoint union of $`\mathrm{\Phi }`$ and $`\mathrm{\Theta }`$, might not be an interval. We thus define two intervals $`\mathrm{\Phi }`$ and $`\mathrm{\Theta }`$ to be totally unrelated if $`FG`$ and $`GF`$ whenever $`F\mathrm{\Phi }`$ and $`G\mathrm{\Theta }`$, and, in this case, define the direct sum of $`\mathrm{\Phi }`$ and $`\mathrm{\Theta }`$ to be $`\mathrm{\Phi }\mathrm{\Theta }=\mathrm{\Phi }\dot{}\mathrm{\Theta }`$. It is easy to check that the direct sum of two intervals is again an interval.
###### Example 2.3.
It is easy to see that the interval $`\mathrm{\Phi }`$ of Example 2.2 is a direct sum $`\{12456,1245,1246,124\}\{1356,135,136\}`$. The components of the direct sum are indeed totally unrelated, even though they share many vertices.
The join of two intervals $`\mathrm{\Phi }`$ and $`\mathrm{\Theta }`$ on disjoint vertex sets is
$$\mathrm{\Phi }\mathrm{\Theta }=\{F\dot{}G:F\mathrm{\Phi },G\mathrm{\Theta }\}.$$
When $`\mathrm{\Phi }`$ and $`\mathrm{\Theta }`$ are simplicial complexes, this matches the usual definition of join. It is easy to see that the join of two intervals is again an interval. Some special cases of the join deserve particular attention. If $`\mathrm{\Phi }`$ is an interval and $`R`$ is a set disjoint from the vertices of $`\mathrm{\Phi }`$, then define
$$R\mathrm{\Phi }=\{R\}\mathrm{\Phi }=\{R\dot{}F:F\mathrm{\Phi }\},$$
the join of $`\mathrm{\Phi }`$ with the interval whose only face is $`R`$. If $`v`$ a vertex not in $`\mathrm{\Phi }`$, then the cone of $`\mathrm{\Phi }`$ is
$$v\mathrm{\Phi }=\{v,\mathrm{}\}\mathrm{\Phi },$$
the join of $`\mathrm{\Phi }`$ with the interval whose two faces are $`v`$ and the empty face. The open star of $`\mathrm{\Phi }`$ is $`v\mathrm{\Phi }`$. Note that
$$v\mathrm{\Phi }=\mathrm{\Phi }\dot{}(v\mathrm{\Phi }).$$
Deletion and contraction are well-known concepts from matroid theory, and were easily extended to simplicial complexes in . Now we further extend to intervals. If $`\mathrm{\Phi }`$ is an interval and $`e`$ is a vertex of $`\mathrm{\Phi }`$, then the deletion and contraction of $`\mathrm{\Phi }`$ by $`e`$ are, respectively,
$`\mathrm{\Phi }e`$ $`=\{F\mathrm{\Phi }:eF\};`$
$`\mathrm{\Phi }/e`$ $`=\{Fe:F\mathrm{\Phi },eF\}.`$
As opposed to the simplicial complex case, $`\mathrm{\Phi }/e`$ is not necessarily a subset of $`\mathrm{\Phi }e`$. As with simplicial complexes, neither $`\mathrm{\Phi }/e`$ nor $`\mathrm{\Phi }e`$ contains $`e`$ in any of their faces, though we stil consider $`e`$ to a vertex, albeit a loop, in each case. It is also easy to check that $`\mathrm{\Phi }e`$ and $`\mathrm{\Phi }/e`$ are intervals when $`\mathrm{\Phi }`$ is an interval. Note that
$`(\mathrm{\Phi }e)^{}`$ $`=\{EF:F\mathrm{\Phi },eF\}=\{EF:F\mathrm{\Phi },eEF\}`$
$`=e(\mathrm{\Phi }^{}/e)`$
and, similarly,
$`(\mathrm{\Phi }/e)^{}`$ $`=\{E(Fe):F\mathrm{\Phi },eF\}=\{(EF)\dot{}e:F\mathrm{\Phi },eEF\}`$
$`=e(\mathrm{\Phi }^{}e).`$
We are now ready to define reduction, which will be a focal point for most of the rest of our work.
###### Definition.
If $`\mathrm{\Phi }`$ is an interval and $`e`$ is a vertex of $`\mathrm{\Phi }`$, then the star of $`e`$ in $`\mathrm{\Phi }`$ is
$$\mathrm{st}_\mathrm{\Phi }e=\underset{F,F\dot{}e\mathrm{\Phi }}{}\{F,F\dot{}e\}=e((\mathrm{\Phi }e)(\mathrm{\Phi }/e)),$$
and the reduction of $`\mathrm{\Phi }`$ by $`e`$ is
$$\mathrm{\Phi }||e=\mathrm{\Phi }\mathrm{st}_\mathrm{\Phi }e.$$
When $`\mathrm{\Phi }`$ is a simplicial complex, $`\mathrm{st}_\mathrm{\Phi }e`$ matches the usual definition. It is easy to check that $`\mathrm{st}_\mathrm{\Phi }e`$ is an interval when $`\mathrm{\Phi }`$ is an interval, but $`\mathrm{\Phi }||e`$ takes a little more work.
###### Lemma 2.4.
If $`\mathrm{\Phi }`$ is an interval with vertex $`e`$, then $`\mathrm{\Phi }||e`$ is again an interval.
###### Proof.
Assume otherwise, so $`FGH`$, and $`F,H\mathrm{\Phi }||e`$, but $`G\mathrm{\Phi }||e`$. Thus $`F,H\mathrm{\Phi }`$, and, since $`\mathrm{\Phi }`$ is an interval, $`G\mathrm{\Phi }`$.
If $`eG`$, then $`eF`$, and then $`FF\dot{}eG\dot{}e`$. But also $`G\mathrm{\Phi }||e`$ implies $`G\dot{}e\mathrm{\Phi }`$. Then, since $`\mathrm{\Phi }`$ is an interval, $`F\dot{}e\mathrm{\Phi }`$, which contradicts $`F\mathrm{\Phi }||e`$.
Similarly, if instead $`eG`$, then $`eH`$, and then $`GeHeH`$. But also $`G\mathrm{\Phi }||e`$ implies $`Ge\mathrm{\Phi }`$. Then since $`\mathrm{\Phi }`$ is an interval, $`He\mathrm{\Phi }`$, which contradicts $`H\mathrm{\Phi }||e`$. ∎
###### Proposition 2.5.
If $`\mathrm{\Phi }`$ is an interval with vertex $`e`$, then $`\mathrm{\Phi }||e`$ is the direct sum of $`\{F\mathrm{\Phi }||e:eF\}=\{F\mathrm{\Phi }:eF,F\dot{}e\mathrm{\Phi }\}`$ and $`\{G\mathrm{\Phi }||e:eG\}=\{G\mathrm{\Phi }:eG,Ge\mathrm{\Phi }\}`$.
###### Proof.
To show $`\mathrm{\Phi }||e`$ is the desired direct sum, let $`F,G\mathrm{\Phi }||e`$ such that $`eF`$, and $`eG`$; we must show $`F`$ and $`G`$ are unrelated. Since $`eG\backslash F`$, we know $`GF`$, so assume $`FG`$. Then $`FF\dot{}eG`$. Since $`F,G\mathrm{\Phi }`$, then also $`F\dot{}e\mathrm{\Phi }`$, which contradicts $`F\mathrm{\Phi }||e`$. ∎
###### Example 2.6.
Let $`\mathrm{\Theta }`$ be the interval of all faces $`F\{1,\mathrm{},6\}`$ such that $`F`$ is a subset of $`12356`$ or $`12456`$, but also a superset of $`12`$, $`135`$, or $`136`$. It is not hard to check that $`\mathrm{\Theta }||3`$ is the interval $`\mathrm{\Phi }`$ of Examples 2.2 and 2.3. The direct sum decomposition of $`\mathrm{\Phi }=\mathrm{\Theta }||3`$ given in Example 2.3 is the one guaranteed by Proposition 2.5.
In the special case where $`\mathrm{\Phi }`$ is a simplicial complex, $`\{G\mathrm{\Phi }||e:eG\}`$ is empty and $`\mathrm{\Phi }||e=(\mathrm{\Phi }e,\mathrm{\Phi }/e)`$. It is easy to check that $`(\mathrm{st}_\mathrm{\Phi }e)^{}=\mathrm{st}_{(\mathrm{\Phi }^{})}e`$, and so $`(\mathrm{\Phi }||e)^{}=\mathrm{\Phi }^{}||e`$.
We review our notation for boundary maps and homology groups of simplicial complexes (as in e.g., \[12, Chapter 1\]). As usual, let $`\mathrm{\Phi }_i`$ denote the set of $`i`$-dimensional faces of $`\mathrm{\Phi }`$, and let $`C_i=C_i(\mathrm{\Phi };):=C_i(\mathrm{\Delta };)/C_i(\mathrm{\Delta }^{};)`$ denote the $`i`$-dimensional oriented $``$-chains of $`\mathrm{\Phi }=(\mathrm{\Delta },\mathrm{\Delta }^{})`$, i.e., the formal $``$-linear sums of oriented $`i`$-dimensional faces $`[F]`$ such that $`F\mathrm{\Phi }_i`$. Let $`_{\mathrm{\Phi };i}=_i:C_iC_{i1}`$ denote the usual (signed) boundary operator. Via the natural bases $`\mathrm{\Phi }_i`$ and $`\mathrm{\Phi }_{i1}`$ for $`C_i(\mathrm{\Phi };)`$ and $`C_{i1}(\mathrm{\Phi };)`$, respectively, the boundary operator $`_i`$ has an adjoint map called the coboundary operator, $`_i^{}:C_{i1}(\mathrm{\Phi };)C_i(\mathrm{\Phi };)`$; i.e., the matrices representing $``$ and $`^{}`$ in the natural bases are transposes of one another.
As long as $`\mathrm{\Phi }`$ is an interval, $`C_{}(\mathrm{\Phi };)`$ forms a chain complex, i.e., $`_{i1}_i=0`$. This simple observation is the key step to several results that follow. To start with, the usual homology groups $`\stackrel{~}{H}_i(\mathrm{\Phi };)=\mathrm{ker}_i/\mathrm{im}_{i+1}`$ are well-defined. Recall $`\stackrel{~}{\beta }_i(\mathrm{\Phi })=dim_{}\stackrel{~}{H}_i(\mathrm{\Phi };)`$.
###### Lemma 2.7.
If $`\mathrm{\Phi }`$ is an interval with vertex $`e`$, then
$$\stackrel{~}{\beta }_i(\mathrm{\Phi }||e)=\stackrel{~}{\beta }_i(\mathrm{\Phi })$$
for all $`i`$.
###### Proof.
First note that $`\mathrm{st}_\mathrm{\Phi }e=e((\mathrm{\Phi }e)(\mathrm{\Phi }/e))=e(\mathrm{\Gamma },\mathrm{\Gamma }^{})=(e\mathrm{\Gamma },e\mathrm{\Gamma }^{})`$ for some simplicial complexes $`\mathrm{\Gamma }`$ and $`\mathrm{\Gamma }^{}`$, and so is acyclic. Now, $`\mathrm{\Phi }`$, $`\mathrm{\Phi }||e`$, and $`\mathrm{st}_\mathrm{\Phi }e`$ are all intervals, and thus chain complexes; furthermore, by definition of $`\mathrm{\Phi }||e`$,
$$0\mathrm{st}_\mathrm{\Phi }e\mathrm{\Phi }\mathrm{\Phi }||e0$$
is a short exact sequence of chain complexes. The resulting long exact sequence in reduced homology (e.g., \[12, Section 24\]),
$$\mathrm{}\stackrel{~}{H}_i(\mathrm{st}_\mathrm{\Phi }e)\stackrel{~}{H}_i(\mathrm{\Phi })\stackrel{~}{H}_i(\mathrm{\Phi }||e)\stackrel{~}{H}_{i1}(\mathrm{st}_\mathrm{\Phi }e)\mathrm{},$$
becomes
$$\mathrm{}0\stackrel{~}{H}_i(\mathrm{\Phi })\stackrel{~}{H}_i(\mathrm{\Phi }||e)0\mathrm{},$$
and the result follows immediately. ∎
We collect here the easy facts we need about how interval direct sums and joins (and thus cones and open stars) interact with deletion, contraction, stars, and reduction. Each fact is either immediate from the relevant definitions, or a routine calculation. For the identities with the join, we assume $`e`$ is a vertex of $`\mathrm{\Phi }`$.
$`(\mathrm{\Phi }\mathrm{\Theta })e`$ $`=(\mathrm{\Phi }e)(\mathrm{\Theta }e)`$ $`(\mathrm{\Phi }\mathrm{\Theta })e`$ $`=(\mathrm{\Phi }e)\mathrm{\Theta }`$
$`(\mathrm{\Phi }\mathrm{\Theta })/e`$ $`=(\mathrm{\Phi }/e)(\mathrm{\Theta }/e)`$ $`(\mathrm{\Phi }\mathrm{\Theta })/e`$ $`=(\mathrm{\Phi }/e)\mathrm{\Theta })`$
$`\mathrm{st}_{(\mathrm{\Phi }\mathrm{\Theta })}e`$ $`=\mathrm{st}_\mathrm{\Phi }e\mathrm{st}_\mathrm{\Theta }e`$ $`\mathrm{st}_{(\mathrm{\Phi }\mathrm{\Theta })}e`$ $`=\mathrm{st}_\mathrm{\Phi }e\mathrm{\Theta }`$
$`(\mathrm{\Phi }\mathrm{\Theta })||e`$ $`=(\mathrm{\Phi }||e)(\mathrm{\Theta }||e)`$ $`(\mathrm{\Phi }\mathrm{\Theta })||e`$ $`=(\mathrm{\Phi }||e)\mathrm{\Theta }`$
## 3. Laplacians
In this section, we define the Laplacian operators and the spectral recursion, develop the tools we will need later to work with them, and show that several operations on intervals, including duality (Proposition 3.7), preserve the property of satisfying the spectral recursion.
###### Definition.
The ($`i`$-dimensional ) Laplacian of $`\mathrm{\Phi }`$ is the map $`L_i(\mathrm{\Phi }):C_i(\mathrm{\Phi };)C_i(\mathrm{\Phi };)`$ defined by
$$L_i=L_i(\mathrm{\Phi }):=_{i+1}_{i+1}^{}+_i^{}_i.$$
It is not hard to see that $`L_i(\mathrm{\Phi })`$ maps each face $`[F]`$ to a linear combination of faces in $`\mathrm{\Phi }`$ adjacent to $`F`$, that is, faces in $`\mathrm{\Phi }`$ of the form $`Fv\dot{}w`$ for some (not necessarily distinct) vertices $`v,w`$, and such that $`Fv\mathrm{\Phi }`$ or $`F\dot{}w\mathrm{\Phi }`$. For details on the coefficients of these linear combinations (in the simplicial complex case, though the ideas are similar for intervals), see \[3, equations (3.2)–(3.4)\], but we will not need that level of detail here. For more information on Laplacians, also see, e.g., .
Each of $`_{i+1}_{i+1}^{}`$ and $`_i^{}_i`$ is positive semidefinite, since each is the composition of a linear map and its adjoint. Therefore, their sum $`L_i`$ is also positive semidefinite, and so has only non-negative real eigenvalues. (See also \[6, Proposition 2.1\].) These eigenvalues do not depend on the arbitrary ordering of the vertices of $`\mathrm{\Phi }`$, and are thus invariants of $`\mathrm{\Phi }`$; see, e.g., \[3, Remark 3.2\]. Define $`𝐬_i(\mathrm{\Phi })`$ to be the multiset of eigenvalues of $`L_i(\mathrm{\Phi })`$, and define $`m_\lambda (L_i(\mathrm{\Phi }))`$ to be the multiplicity of $`\lambda `$ in $`𝐬_i(\mathrm{\Phi })`$.
The first result of combinatorial Hodge theory, which goes back to Eckmann , is that
(2)
$$m_0(L_i(\mathrm{\Phi }))=\stackrel{~}{\beta }_i(\mathrm{\Phi }).$$
Though initially stated only for the case where $`\mathrm{\Phi }`$ is a simplicial complex, there is a simple proof that only relies upon $`\mathrm{\Phi }`$ being a chain complex, and so applies to all intervals $`\mathrm{\Phi }`$; see \[6, Proposition 2.1\].
A natural generating function for the Laplacian eigenvalues of an interval $`\mathrm{\Phi }`$ is
$$S_\mathrm{\Phi }(t,q):=\underset{i0}{}t^i\underset{\lambda 𝐬_{i1}(\mathrm{\Phi })}{}q^\lambda =\underset{i,\lambda }{}m_\lambda (L_{i1}(\mathrm{\Phi }))t^iq^\lambda .$$
We call $`S_\mathrm{\Phi }`$ the spectrum polynomial of $`\mathrm{\Phi }`$. It was introduced (with slightly different indexing) for matroids in , and extended to relative simplicial pairs in . Although $`S_\mathrm{\Phi }`$ is defined for any interval $`\mathrm{\Phi }`$, it is only truly a polynomial when the Laplacian eigenvalues are not only non-negative, but integral as well. This will be true for the cases we are concerned with, primarily shifted intervals , matroids , and matroid pairs $`(Me,M/e)`$ .
Let $`F`$ be a face in interval $`\mathrm{\Phi }`$. As usual, the boundary of $`F`$ in $`\mathrm{\Phi }`$ is the collection of faces $`\{Fv\mathrm{\Phi }:vF\}`$. Similarly, the coboundary of $`F`$ in $`\mathrm{\Phi }`$ is the collection of faces $`\{F\dot{}w\mathrm{\Phi }:wF\}`$. It is not hard to see that $`_{(\mathrm{\Phi }^{})}`$ and $`(_\mathrm{\Phi })^{}`$ each map $`[F]`$ to a linear combination of faces in the coboundary of $`F`$ in $`\mathrm{\Phi }`$. In fact, \[2, Lemma 6.1\] states that $`_{(\mathrm{\Phi }^{})}`$ and $`(_\mathrm{\Phi })^{}`$ are isomorphic, up to an easy change of basis (multiplying some basis elements by $`1`$). The easy corollary \[2, Corollary 6.2\] is that $`L_i(\mathrm{\Phi })`$ is, modulo that same change of basis, isomorphic to $`L_{ni2}(\mathrm{\Phi }^{})`$. Therefore \[2, equation (28)\],
$$S_\mathrm{\Phi }^{}(t,q)=t^{|E|}S_\mathrm{\Phi }(t^1,q).$$
By \[2, Corollary 4.3\],
$$S_{\mathrm{\Phi }\mathrm{\Theta }}=S_\mathrm{\Phi }S_\mathrm{\Theta };$$
it follows then that
$$S_{R\mathrm{\Phi }}=t^{|R|}S_\mathrm{\Phi }.$$
The following is the analogue for direct sums. It is simpler than the formula for disjoint union of simplicial complexes \[2, Lemma 6.9\], because even disjoint simplicial complexes share the empty face.
###### Lemma 3.1.
If $`\mathrm{\Phi }`$ and $`\mathrm{\Theta }`$ are intervals such that $`\mathrm{\Phi }\mathrm{\Theta }`$ is well-defined, then $`𝐬_i(\mathrm{\Phi }\mathrm{\Theta })=𝐬_i(\mathrm{\Phi })𝐬_i(\mathrm{\Theta })`$, the multiset union of $`𝐬_i(\mathrm{\Phi })`$ and $`𝐬_i(\mathrm{\Theta })`$, and $`S_{\mathrm{\Phi }\mathrm{\Theta }}=S_\mathrm{\Phi }+S_\mathrm{\Theta }`$.
###### Proof.
Since no face in $`\mathrm{\Theta }`$ is related to any face in $`\mathrm{\Phi }`$, there are no adjacencies between faces in $`\mathrm{\Phi }`$ and faces in $`\mathrm{\Theta }`$, nor do any of the faces in $`\mathrm{\Theta }`$ change any adjacencies in $`\mathrm{\Phi }`$. Similarly, no faces in $`\mathrm{\Phi }`$ change any adjacencies in $`\mathrm{\Theta }`$, and we conclude $`L_i(\mathrm{\Phi }\mathrm{\Theta })=L_i(\mathrm{\Phi })L_i(\mathrm{\Theta })`$. Thus $`𝐬_i(\mathrm{\Phi }\mathrm{\Theta })=𝐬_i(\mathrm{\Phi })𝐬_i(\mathrm{\Theta })`$, and so $`S_{\mathrm{\Phi }\mathrm{\Theta }}=S_\mathrm{\Phi }+S_\mathrm{\Theta }`$. ∎
Following , let the equivalence relation $`𝝀𝝁`$ on multisets $`𝝀`$ and $`𝝁`$ denote that $`𝝀`$ and $`𝝁`$ agree in the multiplicities of all of their non-zero parts, i.e., that they coincide except for possibly their number of zeros.
###### Lemma 3.2.
If $`\mathrm{\Phi }`$ and $`\mathrm{\Theta }`$ are two intervals such that $`\mathrm{\Phi }=\mathrm{\Theta }\dot{}𝒩`$, where $`𝒩`$ is a collections of faces with neither boundary nor coboundary in $`\mathrm{\Phi }`$, then $`𝐬_i(\mathrm{\Phi })𝐬_i(\mathrm{\Theta })`$.
###### Proof.
Since $`\mathrm{\Phi }`$ is an interval, the faces in $`𝒩`$ are not related to any other face in $`\mathrm{\Phi }`$. Thus $`\mathrm{\Phi }=\mathrm{\Theta }𝒩`$. Furthermore, since the faces in $`𝒩`$ are not related to each other, $`L_i(𝒩)`$ is the zero matrix for all $`i`$, and so $`𝐬_i(𝒩)`$ consists of all $`0`$’s. Now apply Lemma 3.1. ∎
###### Definition.
We will say that an interval $`\mathrm{\Phi }`$ satisfies the spectral recursion with respect to $`e`$ if $`e`$ is a vertex of $`\mathrm{\Phi }`$ and
(3)
$$S_\mathrm{\Phi }(t,q)=qS_{\mathrm{\Phi }e}(t,q)+qtS_{\mathrm{\Phi }/e}(t,q)+(1q)S_{\mathrm{\Phi }||e}(t,q).$$
We will say $`\mathrm{\Phi }`$ satisfies the spectral recursion if $`\mathrm{\Phi }`$ satisfies the spectral recursion with respect to every vertex in its vertex set. (Note that Lemma 3.5 below means we need not be too particular about the vertex set of $`\mathrm{\Phi }`$.)
When $`\mathrm{\Phi }`$ is a simplicial complex, $`\mathrm{\Phi }||e`$ becomes $`(\mathrm{\Phi }e,\mathrm{\Phi }/e)`$, and equation (3) immediately reduces to the spectral recursion for simplicial complexes in .
The statement and proof of the following lemma strongly resemble their simplicial complex counterparts \[2, Theorem 2.4 and Corollary 4.8\]. Here as there, specializations of the spectrum polynomial reduce it to nice invariants of the interval, and reduce the spectral recursion to basic recursions for those invariants. We sketch the proof in order to state what the spectrum polynomial and spectral recursion reduce to in each case.
###### Lemma 3.3.
The spectral recursion holds for all intervals when $`q=0`$, $`q=1`$, $`t=0`$, or $`t=1`$
###### Proof.
If $`q=0`$, then by equation (2), $`S_\mathrm{\Phi }`$ becomes $`_it^i\stackrel{~}{\beta }_{i1}(\mathrm{\Phi })`$, as in \[2, Theorem 2.4\]. The spectral recursion then reduces to the identity $`\stackrel{~}{\beta }_i(\mathrm{\Phi })=\stackrel{~}{\beta }_i(\mathrm{\Phi }||e)`$, which we established in Lemma 2.7.
If $`q=1`$, then $`S_\mathrm{\Phi }`$ becomes $`_it^if_{i1}(\mathrm{\Phi })`$, as in \[2, Theorem 2.4\], where $`f_i(\mathrm{\Phi })=|\mathrm{\Phi }_i|`$. The spectral recursion then reduces to the easy identity
(4)
$$f_i(\mathrm{\Phi })=f_i(\mathrm{\Phi }e)+f_{i1}(\mathrm{\Phi }/e).$$
If $`t=0`$, then $`S_\mathrm{\Phi }`$ becomes $`q^{f_0(\mathrm{\Phi })}`$ if $`\mathrm{}\mathrm{\Phi }`$ (as in \[2, Theorem 2.4\]), but becomes $`0`$ otherwise. If $`\mathrm{}\mathrm{\Phi }`$, then every term in the spectral recursion becomes $`0`$; if, on the other hand, $`\mathrm{}\mathrm{\Phi }`$, then, as in \[2, Theorem 2.4\], the spectral recursion reduces to the trivial observation that $`f_0(\mathrm{\Phi })=f_0(\mathrm{\Phi }e)`$ if $`e`$ is not a face of $`\mathrm{\Phi }`$, but $`f_0(\mathrm{\Phi })=1+f_0(\mathrm{\Phi }e)`$ if $`e`$ is a face of $`\mathrm{\Phi }`$.
If $`t=1`$, then $`S_\mathrm{\Phi }`$ becomes $`\chi (\mathrm{\Phi })=_i(1)^if_i(\mathrm{\Phi })=_i(1)^i\stackrel{~}{\beta }_i(\mathrm{\Phi })`$, the Euler characteristic of $`\mathrm{\Phi }`$, by \[2, Corollary 4.8\]. The spectral recursion now reduces to two easy identities about Euler characteristic: that $`\chi (\mathrm{\Phi })=\chi (\mathrm{\Phi }||e)`$, which follows from Lemma 2.7; and that $`\chi (\mathrm{\Phi })=\chi (\mathrm{\Phi }e)\chi (\mathrm{\Phi }/e)`$, which follows from the identity (4) above. ∎
If $`\mathrm{\Phi }`$ is an interval and $`e`$ is a vertex of $`\mathrm{\Phi }`$, define
$$𝒮_i(\mathrm{\Phi },e)=[t^i](S_\mathrm{\Phi }qS_{\mathrm{\Phi }e}qtS_{\mathrm{\Phi }/e}(1q)S_{\mathrm{\Phi }||e}),$$
where $`[t^i]p`$ denotes the coefficient of $`t^i`$ in polynomial $`p`$. Clearly, $`\mathrm{\Phi }`$ satisfies the spectral recursion with respect to $`e`$ precisely when $`𝒮_i(\mathrm{\Phi },e)=0`$ for all $`i`$.
###### Lemma 3.4.
Let $`\mathrm{\Phi }`$ and $`\mathrm{\Theta }`$ be intervals, each with vertex $`e`$, such that $`s_i(\mathrm{\Phi })𝐬_j(\mathrm{\Theta })`$, $`s_i(\mathrm{\Phi }e)𝐬_j(\mathrm{\Theta }e)`$, $`s_{i1}(\mathrm{\Phi }/e)𝐬_{j1}(\mathrm{\Theta }/e)`$, and $`s_i(\mathrm{\Phi }||e)𝐬_j(\mathrm{\Theta }||e)`$. Then $`𝒮_i(\mathrm{\Phi },e)=𝒮_j(\mathrm{\Theta },e)`$.
###### Proof.
Translating the $``$ assumptions to generating functions,
$`[t^i]S_\mathrm{\Phi }`$ $`=[t^j]S_\mathrm{\Theta }+C_1`$
$`[t^i]S_{\mathrm{\Phi }e}`$ $`=[t^j]S_{\mathrm{\Theta }e}+C_2`$
$`[t^{i1}]S_{\mathrm{\Phi }/e}`$ $`=[t^{j1}]S_{\mathrm{\Theta }/e}+C_3`$
$`[t^i]S_{\mathrm{\Phi }||e}`$ $`=[t^j]S_{\mathrm{\Theta }||e}+C_4,`$
where $`C_1,C_2,C_3`$, and $`C_4`$ are constants. It is then easy to compute
$$𝒮_i(\mathrm{\Phi },e)𝒮_j(\mathrm{\Theta },e)=(C_1C_4)+q(C_4C_2C_3).$$
This makes $`𝒮_i(\mathrm{\Phi },e)𝒮_j(\mathrm{\Theta },e)`$ a linear polynomial in $`q`$. But by Lemma 3.3, $`𝒮_i(\mathrm{\Phi },e)𝒮_j(\mathrm{\Theta },e)=0`$ when $`q=0`$ and when $`q=1`$. Therefore $`𝒮_i(\mathrm{\Phi },e)𝒮_j(\mathrm{\Theta },e)`$ must be identically $`0`$, as desired. ∎
The following two results are easy to verify directly; the third is not much harder.
###### Lemma 3.5.
If $`\mathrm{\Phi }`$ is an interval and $`e`$ is a loop, then $`\mathrm{\Phi }`$ satisfies the spectral recursion with respect to $`e`$.
###### Lemma 3.6.
The interval with only a single face, and the interval whose only two faces are a single vertex and the empty face, each satisfy the spectral recursion.
###### Proposition 3.7.
Let $`\mathrm{\Phi }`$ be an interval with vertex $`e`$. If $`\mathrm{\Phi }`$ satisfies the spectral recursion with respect to $`e`$, then so does $`\mathrm{\Phi }^{}`$.
###### Proof.
Calculate
$`S_\mathrm{\Phi }^{}(t,q)`$ $`=t^nS_\mathrm{\Phi }(t^1,q)`$
$`=t^n(qS_{\mathrm{\Phi }e}(t^1,q)+qt^1S_{\mathrm{\Phi }/e}(t^1,q)+(1q)S_{\mathrm{\Phi }||e}(t^1,q))`$
$`=qS_{(\mathrm{\Phi }e)^{}}(t,q)+qt^1S_{(\mathrm{\Phi }/e)^{}}(t,q)+(1q)S_{(\mathrm{\Phi }||e)^{}}(t,q)`$
$`=qS_{e(\mathrm{\Phi }^{}/e)}(t,q)+qt^1S_{e(\mathrm{\Phi }^{}e)}(t,q)+(1q)S_{\mathrm{\Phi }^{}||e}(t,q)`$
$`=qtS_{\mathrm{\Phi }^{}/e}(t,q)+qS_{\mathrm{\Phi }^{}e}(t,q)+(1q)S_{\mathrm{\Phi }^{}||e}(t,q).`$
Similar routine calculations establish the following two lemmas.
###### Lemma 3.8.
If $`\mathrm{\Phi }`$ and $`\mathrm{\Theta }`$ are intervals that satisfy the spectral recursion with respect to $`e`$, and such that $`\mathrm{\Phi }\mathrm{\Theta }`$ is well-defined, then $`\mathrm{\Phi }\mathrm{\Theta }`$ satisfies the spectral recursion with respect to $`e`$.
###### Lemma 3.9.
If $`\mathrm{\Phi }`$ is an interval that satisfies the spectral recursion with respect to $`e`$, and $`\mathrm{\Theta }`$ is another interval such that $`\mathrm{\Phi }\mathrm{\Theta }`$ is well-defined, then $`\mathrm{\Phi }\mathrm{\Theta }`$ satisfies the spectral recursion with respect to $`e`$.
###### Corollary 3.10.
Let $`\mathrm{\Phi }`$ be an interval. If $`\mathrm{\Phi }`$ satisfies the spectral recursion, then so do $`v\mathrm{\Phi }`$ and $`R\mathrm{\Phi }`$.
###### Proof.
Combine Lemmas 3.6 and 3.9
## 4. Skeleta
The main goal of this section is to show that taking skeleta preserves the property of satisfying the spectral recursion (Theorem 4.7). A key step is to show that skeleta and reduction interact reasonably well (Corollary 4.3).
###### Definition.
We will say interval $`\mathrm{\Phi }`$ is $`(i,j)`$-dimensional when $`idimFj`$ for all $`F\mathrm{\Phi }`$. Note that it is not necessary for there to be a face of every dimension between $`i`$ and $`j`$. If $`\mathrm{\Phi }`$ is an interval, we define the $`(i,j)`$-skeleton to be
$$\mathrm{\Phi }^{(i,j)}=\{F\mathrm{\Phi }:idimFj\}$$
It is immediate that
$`\mathrm{\Phi }^{(i,j)}e`$ $`=(\mathrm{\Phi }e)^{(i,j)},`$
$`\mathrm{\Phi }^{(i,j)}/e`$ $`=(\mathrm{\Phi }/e)^{(i1,j1)}.`$
The corresponding statement with reduction instead of deletion or contraction is not true. For instance, in Example 2.6, $`1256(\mathrm{\Theta }||3)^{(1,3)}`$ (since $`12356\mathrm{\Theta }`$), but $`1256\mathrm{\Theta }^{(1,3)}||3`$ (since $`12356`$ is 4-dimensional, and so is not in $`\mathrm{\Theta }^{(1,3)}`$). On the other hand, it will not be hard to show that at least the non-zero eigenvalues of $`\mathrm{\Phi }^{(i,j)}||e`$ and $`(\mathrm{\Phi }||e)^{(i,j)}`$ coincide. We first need two easy technical lemmas.
###### Lemma 4.1.
Let $`\mathrm{\Phi }`$ be an interval with vertices $`e`$ and $`v`$. If $`F,F\dot{}v\mathrm{\Phi }^{(i,j)}||e`$ for some $`i<j`$, then $`F,F\dot{}v\mathrm{\Phi }||e`$.
###### Proof.
First note that $`ve`$, since, otherwise, $`F,F\dot{}v\mathrm{\Phi }^{(i,j)}||e`$ would be impossible. Thus, either $`e`$ is a vertex of both $`F`$ and $`F\dot{}v`$, or $`e`$ is a vertex of neither.
First assume $`eF,F\dot{}v`$. Then $`F\mathrm{\Phi }^{(i,j)}||e`$ implies $`F\dot{}e\mathrm{\Phi }^{(i,j)}`$, and so $`F\dot{}e\mathrm{\Phi }`$ (note $`dimF<j`$). But then $`F\dot{}\{v,e\}\mathrm{\Phi }`$, since $`\mathrm{\Phi }`$ is an interval and $`F\mathrm{\Phi }`$. Now, with $`F\dot{}e,F\dot{}\{v,e\}\mathrm{\Phi }`$, we conclude $`F,F\dot{}v\mathrm{\Phi }||e`$.
Next assume $`eF,F\dot{}v`$. Then $`F\dot{}v\mathrm{\Phi }^{(i,j)}||e`$ implies $`F\dot{}ve\mathrm{\Phi }^{(i,j)}`$, and so $`F\dot{}ve\mathrm{\Phi }`$ (note $`dimF\dot{}v>i`$). But then $`Fe\mathrm{\Phi }`$, since $`\mathrm{\Phi }`$ is an interval and $`F\dot{}v\mathrm{\Phi }`$. Now, with $`Fe,F\dot{}ve\mathrm{\Phi }`$, we conclude $`F,F\dot{}v\mathrm{\Phi }||e`$. ∎
###### Lemma 4.2.
Let $`\mathrm{\Phi }`$ be an interval with vertex $`e`$. Then
$$\mathrm{\Phi }^{(i,j)}||e=(\mathrm{\Phi }||e)^{(i,j)}\dot{}𝒩,$$
where $`𝒩`$ is a set of faces with neither boundary nor coboundary in $`\mathrm{\Phi }^{(i,j)}||e`$.
###### Proof.
First we show $`(\mathrm{\Phi }||e)^{(i,j)}\mathrm{\Phi }^{(i,j)}||e`$. Let $`F(\mathrm{\Phi }||e)^{(i,j)}`$, so $`F\mathrm{\Phi }||e`$ and $`F\mathrm{\Phi }^{(i,j)}`$. If $`eF`$, then $`F\dot{}e\mathrm{\Phi }`$, so $`F\dot{}e\mathrm{\Phi }^{(i,j)}`$, and so $`F\mathrm{\Phi }^{(i,j)}||e`$. If, on the other hand, $`eF`$, then $`Fe\mathrm{\Phi }`$, so $`Fe\mathrm{\Phi }^{(i,j)}`$, and so $`F\mathrm{\Phi }^{(i,j)}||e`$.
Now let $`G\mathrm{\Phi }^{(i,j)}||e`$, $`G(\mathrm{\Phi }||e)^{(i,j)}`$. By Lemma 4.1, for every $`vG`$, we have $`Gv\mathrm{\Phi }^{(i,j)}||e`$, and, for every $`wG`$, we have $`G\dot{}w\mathrm{\Phi }^{(i,j)}||e`$. Therefore, $`G`$ has neither boundary nor coboundary in $`\mathrm{\Phi }^{(i,j)}||e`$, as desired. ∎
###### Corollary 4.3.
Let $`\mathrm{\Phi }`$ be an interval with vertex $`e`$, and let $`i<j`$. Then
$$𝐬_k(\mathrm{\Phi }^{(i,j)}||e)s_k((\mathrm{\Phi }||e)^{(i,j)}),$$
for all $`k`$.
###### Proof.
Apply Lemma 3.2 to Lemma 4.2. ∎
The following two equations are from \[3, equation (3.6)\], where they are established for simplicial complexes, but they are just easy consequences of $`\mathrm{\Phi }`$ being a chain complex.
(5)
$$𝐬_i(\mathrm{\Phi })𝐬_i(\mathrm{\Phi }^{(i1,i)})𝐬_i(\mathrm{\Phi }^{(i,i+1)}),$$
(6)
$$𝐬_{i1}(\mathrm{\Phi }^{(i1,i)})𝐬_i(\mathrm{\Phi }^{(i1,i)}).$$
As a result of this second equation, if $`\mathrm{\Phi }`$ is $`(i1,i)`$-dimensional, we will let $`𝐬(\mathrm{\Phi })`$ refer to the $``$ equivalence class of $`𝐬_{i1}(\mathrm{\Phi })𝐬_i(\mathrm{\Phi })`$.
###### Lemma 4.4.
If $`\mathrm{\Phi }`$ is an $`(i1,i)`$-dimensional interval with vertex $`e`$, then $`𝒮_i(\mathrm{\Phi },e)=𝒮_{i1}(\mathrm{\Phi },e)`$.
###### Proof.
By equation (6), since $`\mathrm{\Phi }`$ is $`(i1,i)`$-dimensional, $`𝐬_{i1}𝐬_i`$ for $`\mathrm{\Phi }`$, $`\mathrm{\Phi }e`$, and $`\mathrm{\Phi }||e`$. Similarly, $`𝐬_{i2}𝐬_{i1}`$ for $`\mathrm{\Phi }/e`$. Now apply Lemma 3.4. ∎
###### Lemma 4.5.
If $`\mathrm{\Phi }`$ is an interval with vertex $`e`$, then
$$𝒮_i(\mathrm{\Phi },e)=𝒮_i(\mathrm{\Phi }^{(i1,i)},e)+𝒮_i(\mathrm{\Phi }^{(i,i+1)},e).$$
###### Proof.
Let $`b`$ and $`t`$ be two new vertices not in $`\mathrm{\Phi }`$, and let
$$\mathrm{\Theta }=(b\mathrm{\Phi }^{(i1,i)})(t\mathrm{\Phi }^{(i,i+1)}).$$
It is immediate that $`\mathrm{\Theta }`$ is well-defined, since $`bt`$. (Indeed, $`b`$ and $`t`$ are introduced precisely to make a direct sum of out $`\mathrm{\Phi }^{(i1,i)}`$ and $`\mathrm{\Phi }^{(i,i+1)}`$.) It is easy to verify that
$`𝐬_{i+1}(\mathrm{\Theta })`$ $`=𝐬_i(\mathrm{\Phi }^{(i1,i)})𝐬_i(\mathrm{\Phi }^{(i,i+1)})𝐬_i(\mathrm{\Phi }),`$
$`𝐬_{i+1}(\mathrm{\Theta }e)`$ $`=𝐬_i((\mathrm{\Phi }e)^{(i1,i)})𝐬_i((\mathrm{\Phi }e)^{(i,i+1)})𝐬_i(\mathrm{\Phi }e),`$
$`𝐬_i(\mathrm{\Theta }/e)`$ $`=𝐬_{i1}((\mathrm{\Phi }/e)^{(i2,i1)})𝐬_{i1}((\mathrm{\Phi }/e)^{(i1,i)})𝐬_{i1}(\mathrm{\Phi }/e),`$
$`𝐬_{i+1}(\mathrm{\Theta }||e)`$ $`=𝐬_i(\mathrm{\Phi }^{(i1,i)}||e)𝐬_i(\mathrm{\Phi }^{(i,i+1)}||e)`$
$`𝐬_i((\mathrm{\Phi }||e)^{(i1,i)})𝐬_i((\mathrm{\Phi }||e)^{(i,i+1)})𝐬_i(\mathrm{\Phi }||e);`$
in each case, the last $``$-equivalence is by equation (5). Then, by Lemma 3.4, $`𝒮_{i+1}(\mathrm{\Theta },e)=𝒮_i(\mathrm{\Phi },e)`$, and so now it is easy to verify
$`𝒮_i(\mathrm{\Phi },e)`$ $`=𝒮_{i+1}(\mathrm{\Theta },e)`$
$`=𝒮_{i+1}(b\mathrm{\Phi }^{(i1,i)}t\mathrm{\Phi }^{(i,i+1)},e)`$
$`=𝒮_{i+1}(b\mathrm{\Phi }^{(i1,i)},e)+𝒮_{i+1}(t\mathrm{\Phi }^{(i,i+1)},e)`$
$`=𝒮_i(\mathrm{\Phi }^{(i1,i)},e)+𝒮_i(\mathrm{\Phi }^{(i,i+1)},e).`$
###### Lemma 4.6.
Let $`\mathrm{\Phi }`$ be an interval with vertex $`e`$. If every skeleton $`\mathrm{\Phi }^{(i1,i)}`$ satisfies the spectral recursion with respect to $`e`$ then so does $`\mathrm{\Phi }`$.
###### Proof.
This is an immediate corollary to Lemma 4.5
###### Theorem 4.7.
Let $`\mathrm{\Phi }`$ be an interval with vertex $`e`$. If $`\mathrm{\Phi }`$ satisfies the spectral recursion with respect to $`e`$, then so does every skeleton $`\mathrm{\Phi }^{(i,j)}`$.
###### Proof.
By Lemma 4.6, it suffices to prove that every $`\mathrm{\Phi }^{(i,i+1)}`$ satisfies the spectral recursion with respect to $`e`$, which we now do by induction on $`i`$.
If $`i2`$, then $`\mathrm{\Phi }^{(i,i+1)}`$ is either the interval whose only face is the empty face, or the empty interval with no faces whatsoever. Either way, $`\mathrm{\Phi }^{(i,i+1)}`$ trivially satisfies the spectral recursion.
If $`i>2`$, then, by induction, $`𝒮_i(\mathrm{\Phi }^{(i1,i)},e)=0`$, and by hypothesis, $`𝒮_i(\mathrm{\Phi },e)=0`$. Then by Lemma 4.5, $`𝒮_i(\mathrm{\Phi }^{(i,i+1)},e)=0`$, and so $`\mathrm{\Phi }^{(i,i+1)}`$ satisfies the spectral recursion with respect to $`e`$, by Lemma 4.4. ∎
## 5. Shifted Intervals
Our main goal of this section is to show that relative simplicial pairs that are shifted (on the same vertex order) satisfy the spectral recursion (Theorem 5.12). The key step is the construction of another interval $`\mathrm{\Phi }^{}`$ that satisfies the spectral recursion when $`\mathrm{\Phi }`$ does; this resembles, but is more involved than, a construction in the proof of the simplicial complex case \[2, Lemma 4.22\]. We first translate shifted relative simplicial pairs to shifted intervals, and show that the dual of a shifted interval is again a shifted interval (Proposition 5.6).
###### Definition.
If $`F=\{f_1<\mathrm{}<f_k\}`$ and $`G=\{g_1<\mathrm{}<g_k\}`$ are $`k`$-subsets of integers, then $`F_CG`$ under the componentwise partial order if $`f_pg_p`$ for all $`p`$. A simplicial complex $`\mathrm{\Delta }`$ on a vertex set of integers is shifted if $`G_CH`$ and $`H\mathrm{\Delta }`$ together imply $`G\mathrm{\Delta }`$. An interval $`\mathrm{\Phi }`$ is shifted when $`\mathrm{\Phi }=(\mathrm{\Delta },\mathrm{\Delta }^{})`$, for some shifted simplicial complexes $`\mathrm{\Delta }`$ and $`\mathrm{\Delta }^{}`$.
We would like to replace this definition of shifted interval, which depends on the simplicial complexes involved, to one that depends only on the interval itself. In order to do this, we will need a single partial order that combines the (separate) conditions of $`\mathrm{\Delta }`$ being shifted, and $`\mathrm{\Delta }`$ being a simplicial complex, an idea implicit in the work of Klivans (see e.g., \[8, Figure 1\] or \[7, Figure 1\]). If $`F=\{f_1<\mathrm{}<f_k\}`$ and $`G=\{g_1<\mathrm{}<g_m\}`$, then $`F_SG`$ under the shifted partial order when $`km`$ and $`f_{p+mk}g_p`$ for all $`1pk`$. In particular, it is easy to see that if $`FG`$ or $`F_CG`$, then $`F_SG`$.
###### Lemma 5.1.
If $`\mathrm{\Delta }2^E`$, then the following are equivalent:
1. $`\mathrm{\Delta }`$ is a shifted simplicial complex; and
2. $`F_SH`$ and $`H\mathrm{\Delta }`$ together imply $`F\mathrm{\Delta }`$.
###### Proof.
That (2) implies (1) is an easy exercise. To see that (1) implies (2), assume $`\mathrm{\Delta }`$ is a shifted simplicial complex and that $`F_SH\mathrm{\Delta }`$, and let $`G`$ consist of the last $`|F|`$ elements of $`H`$. Then it is easy to see that $`F_CGH`$. Therefore $`G\mathrm{\Delta }`$, and, consequently, $`F\mathrm{\Delta }`$. ∎
###### Lemma 5.2.
If $`\mathrm{\Phi }2^E`$, then the following are equivalent:
1. $`\mathrm{\Phi }`$ is a shifted interval; and
2. $`F_SG_SH`$ and $`F,H\mathrm{\Phi }`$ together imply $`G\mathrm{\Phi }`$.
###### Proof.
Thanks to Lemma 5.1, the proof is entirely analogous to that of Lemma 2.1, but with $`_S`$ instead of $``$. ∎
The following lemma, whose easy proof is omitted, means that the partial order $`_S`$ is admissible.
###### Lemma 5.3.
If $`vF,G`$, then $`F_SG`$ iff $`F\dot{}v_SG\dot{}v`$.
###### Corollary 5.4.
If $`AF=AG=\mathrm{}`$, then $`F_SG`$ iff $`F\dot{}A_SG\dot{}A`$.
###### Lemma 5.5.
If $`F,GE`$, then $`F_SG`$ iff $`EG_SEF`$.
###### Proof.
Let $`A=FG`$ and $`B=(EF)(EG)`$, and let $`F^{}=FA`$ and $`G^{}=GA`$. Thus $`F=F^{}\dot{}A`$, $`G=G^{}\dot{}A`$, $`EF=G^{}\dot{}B`$, and $`EG=F^{}\dot{}B`$. Then by Corollary 5.4 twice, $`F_SG`$ iff $`F^{}_SG^{}`$ iff $`EG_SEF`$. ∎
###### Proposition 5.6.
If $`\mathrm{\Phi }`$ is a shifted interval, then so is $`\mathrm{\Phi }^{}`$.
###### Proof.
Assume $`F_SG_SH`$ and $`F,H\mathrm{\Phi }^{}`$. Then $`EH_SEG_SEF`$, by Lemma 5.5, and $`EF,EH\mathrm{\Phi }`$. Therefore $`EG\mathrm{\Phi }`$, and so $`G\mathrm{\Phi }`$. ∎
We have one final lemma about $`_S`$ whose easy proof is omitted.
###### Lemma 5.7.
If $`F_SG`$ and $`dimF<dimG`$, then $`F\dot{}1_SG`$.
We now turn our attention to proving that shifted intervals satisfy the spectral recursion. We start with a definition that does not rely upon $`\mathrm{\Phi }`$ being shifted, but which will be very useful when $`\mathrm{\Phi }`$ is shifted. If $`\mathrm{\Phi }`$ is an $`(i1,i)`$-dimensional interval with vertex $`1`$, then define
$$\mathrm{\Phi }^{}=\mathrm{\Phi }𝒩_\mathrm{\Phi },$$
where
$$𝒩_\mathrm{\Phi }=\{F\mathrm{\Phi }_i:1F,F1\mathrm{\Phi }\}\dot{}\{F\mathrm{\Phi }_{i1}:1F,F\dot{}1\mathrm{\Phi }\}.$$
Computing $`\mathrm{\Phi }^{}`$ dimension by dimension, we see that, equivalently,
$`\mathrm{\Phi }^{}`$ $`=\{F\mathrm{\Phi }_i:1F\}\dot{}\{F\mathrm{\Phi }_i:1F,F1\mathrm{\Phi }\}`$
$`\dot{}\{F\mathrm{\Phi }_{i1}:1F,F\dot{}1\mathrm{\Phi }\}\dot{}\{F\mathrm{\Phi }_{i1}:1F\}`$
$`=(\mathrm{\Phi }_i1)\dot{}(1((\mathrm{\Phi }_i/1)(\mathrm{\Phi }_{i1}1)))`$
$`\dot{}((\mathrm{\Phi }_i/1)(\mathrm{\Phi }_{i1}1))\dot{}(1(\mathrm{\Phi }_{i1}/1))`$
(7) $`=(\mathrm{\Phi }_i1)\dot{}(1((\mathrm{\Phi }_i/1)(\mathrm{\Phi }_{i1}1)))\dot{}(1(\mathrm{\Phi }_{i1}/1)).`$
###### Lemma 5.8.
If $`\mathrm{\Phi }`$ is a shifted $`(i1,i)`$-dimensional interval on vertex set $`\{1,\mathrm{},n\}`$, then the faces of $`𝒩_\mathrm{\Phi }`$ have neither boundary nor coboundary in $`\mathrm{\Phi }`$.
###### Proof.
Let $`F𝒩_\mathrm{\Phi }`$. We split the proof into two cases, depending on the dimension of $`F`$.
First assume $`dimF=i1`$. Then $`F\mathrm{\Phi }`$ and $`F\dot{}1\mathrm{\Phi }`$, which imply $`F\dot{}v\overline{)}\mathrm{\Phi }`$ for any $`v`$, since $`FF\dot{}1_CF\dot{}v`$. Thus, $`F`$ has no coboundary in $`\mathrm{\Phi }`$; $`F`$ has no boundary in $`\mathrm{\Phi }`$ simply becasue it has minimal dimension in $`\mathrm{\Phi }`$.
Now assume, on the other hand, $`dimF=i`$. Then $`F\mathrm{\Phi }`$ and $`F1\mathrm{\Phi }`$, which imply $`Fv\mathrm{\Phi }`$ for any $`v`$, since $`Fv_CF1F`$. Thus, $`F`$ has no boundary in $`\mathrm{\Phi }`$; $`F`$ has no coboundary in $`\mathrm{\Phi }`$ simply because it has maximal dimension in $`\mathrm{\Phi }`$. ∎
###### Lemma 5.9.
Let $`\mathrm{\Phi }`$ be a shifted $`(i1,i)`$-dimensional interval on vertices $`\{1,\mathrm{},n\}`$, and let $`1en`$. Then $`\mathrm{\Phi }`$ satisfies the spectral recursion with respect to $`e`$ iff $`\mathrm{\Phi }^{}`$ does.
###### Proof.
By Lemma 3.4, it suffices to show $`𝐬(\mathrm{\Phi }^{})𝐬(\mathrm{\Phi })`$, $`𝐬(\mathrm{\Phi }^{}e)𝐬(\mathrm{\Phi }e)`$, $`𝐬(\mathrm{\Phi }^{}/e)𝐬(\mathrm{\Phi }/e)`$, and $`𝐬(\mathrm{\Phi }^{}||e)𝐬(\mathrm{\Phi }||e)`$. The main tools are Lemmas 3.2 and 5.8, which immediately show $`𝐬(\mathrm{\Phi }^{})𝐬(\mathrm{\Phi })`$.
In order to show $`𝐬(\mathrm{\Phi }^{}||e)𝐬(\mathrm{\Phi }||e)`$, we first claim that $`\mathrm{st}_\mathrm{\Phi }^{}e=\mathrm{st}_\mathrm{\Phi }e`$. Indeed, $`\mathrm{st}_{(\mathrm{\Phi }𝒩)}e=\mathrm{st}_\mathrm{\Phi }e`$ for any set $`𝒩`$ of faces in $`\mathrm{\Phi }`$ with neither boundary nor coboundary in $`\mathrm{\Phi }`$. Then
$`𝐬(\mathrm{\Phi }^{}||e)`$ $`=𝐬((\mathrm{\Phi }𝒩_\mathrm{\Phi })\mathrm{st}_{(\mathrm{\Phi }𝒩_\mathrm{\Phi })}e)=𝐬((\mathrm{\Phi }\mathrm{st}_\mathrm{\Phi }e)𝒩_\mathrm{\Phi })=𝐬((\mathrm{\Phi }||e)𝒩_\mathrm{\Phi })`$
$`𝐬(\mathrm{\Phi }||e),`$
by Lemmas 3.2 and 5.8, since the faces of $`𝒩_\mathrm{\Phi }`$ have neither boundary nor coboundary in $`\mathrm{\Phi }`$, nor in any subset of $`\mathrm{\Phi }`$, such as $`\mathrm{\Phi }||e`$.
To show $`𝐬(\mathrm{\Phi }^{}e)𝐬(\mathrm{\Phi }e)`$ and $`𝐬(\mathrm{\Phi }^{}/e)𝐬(\mathrm{\Phi }/e)`$, we split into two cases: $`e=1`$; and $`e1`$. If $`e1`$, then equation (7) makes it easy to show that $`\mathrm{\Phi }^{}e=(\mathrm{\Phi }e)^{}`$ and $`\mathrm{\Phi }^{}/e=(\mathrm{\Phi }/e)^{}`$. Then Lemmas 3.2 and 5.8 show $`𝐬(\mathrm{\Phi }^{}e)𝐬((\mathrm{\Phi }e)^{})𝐬(\mathrm{\Phi }e)`$ and $`𝐬(\mathrm{\Phi }^{}/e)𝐬((\mathrm{\Phi }/e)^{})𝐬(\mathrm{\Phi }/e)`$.
To address the $`e=1`$ case, first note that $`\mathrm{\Phi }^{}1=(\mathrm{\Phi }𝒩_\mathrm{\Phi })1=(\mathrm{\Phi }1)(𝒩_\mathrm{\Phi }(\mathrm{\Phi }1))`$. Let $`𝒩^{}=𝒩_\mathrm{\Phi }(\mathrm{\Phi }1)`$. Since $`𝒩^{}𝒩_\mathrm{\Phi }`$, every face in $`𝒩^{}`$ has neither boundary nor coboundary in $`\mathrm{\Phi }`$, nor in any subset of $`\mathrm{\Phi }`$, such as $`\mathrm{\Phi }1`$. Now apply Lemma 3.2 to see $`𝐬(\mathrm{\Phi }^{}1)𝐬(\mathrm{\Phi }1)`$. The proof that $`𝐬(\mathrm{\Phi }^{}/1)𝐬(\mathrm{\Phi }/1)`$ proceeds similarly. ∎
###### Definition.
Let $`\mathrm{\Phi }`$ be an $`(i1,i)`$-dimensional interval with vertex $`1`$. Define
$`\mathrm{\Phi }^+`$ $`=\mathrm{\Phi }^{}\dot{}\{F\dot{}1:1F,F\mathrm{\Phi }_i\}\dot{}\{F1:1F,F\mathrm{\Phi }_{i1}\}.`$
###### Lemma 5.10.
If $`\mathrm{\Phi }`$ is a shifted $`(i1,i)`$-dimensional interval on vertices $`\{1,\mathrm{},n\}`$, then $`\mathrm{\Phi }^+=1\mathrm{\Phi }^{}`$ for some shifted interval $`\mathrm{\Phi }^{}`$ on vertex set $`\{2,\mathrm{},n\}`$.
###### Proof.
First, by equation (7),
$`\mathrm{\Phi }^+`$ $`=(\mathrm{\Phi }_i1)\dot{}(1((\mathrm{\Phi }_i/1)(\mathrm{\Phi }_{i1}1)))\dot{}(1(\mathrm{\Phi }_{i1}/1))`$
$`\dot{}(1(\mathrm{\Phi }_i1))\dot{}(\mathrm{\Phi }_{i1}/1)`$
(8) $`=1((\mathrm{\Phi }_i1)\dot{}((\mathrm{\Phi }_i/1)(\mathrm{\Phi }_{i1}1))\dot{}(\mathrm{\Phi }_{i1}/1)).`$
Now, coning preserves shiftedness of intervals, since $`1(\mathrm{\Delta },\mathrm{\Delta }^{})=(1\mathrm{\Delta },1\mathrm{\Delta }^{})`$ and, as is well-known and easy to prove, coning preserves shiftedness of simplicial complexes. Equation (8) thus reduces the proof of this lemma to showing that
(9)
$$\mathrm{\Phi }^{}=(\mathrm{\Phi }_i1)\dot{}((\mathrm{\Phi }_i/1)(\mathrm{\Phi }_{i1}1))\dot{}(\mathrm{\Phi }_{i1}/1)$$
is a shifted interval.
Equation (9) means $`\mathrm{\Phi }_i^{}=(\mathrm{\Phi }_i1)`$, $`\mathrm{\Phi }_{i1}^{}=\dot{}((\mathrm{\Phi }_i/1)(\mathrm{\Phi }_{i1}1))`$, and $`\mathrm{\Phi }_{i2}^{}=(\mathrm{\Phi }_{i1}/1)`$, and so $`G\mathrm{\Phi }^{}`$ precisely when the following conditions are met:
1. $`i2dimGi`$;
2. if $`dimGi1`$, then $`G\dot{}1\mathrm{\Phi }`$; and
3. if $`dimGi1`$, then $`G\mathrm{\Phi }`$.
We will use the characterization of shifted intervals given in Lemma 5.2 to show that $`\mathrm{\Phi }^{}`$ is a shifted interval. So assume $`G\{2,\mathrm{},n\}`$; $`F,H\mathrm{\Phi }^{}`$; and $`F_SG_SH`$. We need to show $`G\mathrm{\Phi }^{}`$. Condition (1) follows directly from the hypotheses on $`G`$.
Next we establish condition (2); so assume $`dimGi1`$. First note that $`dimFdimGi1`$, so $`F\dot{}1\mathrm{\Phi }`$, and $`F\dot{}1_SG\dot{}1`$, by Lemma 5.3. Now, if $`dimH=dimGi1`$, then $`H\dot{}1\mathrm{\Phi }`$ and $`G\dot{}1_SH\dot{}1`$, by Lemma 5.3, but if $`dimH>dimG`$, then $`dimHi1`$, and so $`H\mathrm{\Phi }`$ and, by Lemma 5.7, $`G\dot{}1_SH`$. Either way, for some $`\stackrel{~}{H}`$ (either $`H`$ or $`H\dot{}1`$), $`F\dot{}1_SG\dot{}1_S\stackrel{~}{H}`$ and $`F\dot{}1,\stackrel{~}{H}\mathrm{\Phi }`$. Thus $`G\dot{}1\mathrm{\Phi }`$, as desired.
The proof that $`G`$ satisfies condition (3) is similar; we start by assuming $`dimGi1`$. First note that $`dimHdimGi1`$, so $`H\mathrm{\Phi }`$ while $`G_SH`$. Now if $`dimF=dimGi1`$, then $`F\mathrm{\Phi }`$ while $`F_SG`$, but if $`dimF<dimG`$, then $`dimFi`$, and so $`F\dot{}1\mathrm{\Phi }`$ and, by Lemma 5.7, $`F\dot{}1_SG`$. Either way, for some $`\stackrel{~}{F}`$ (either $`F`$ or $`F\dot{}1`$), $`\stackrel{~}{F}_SG_SH`$ and $`\stackrel{~}{F},H\mathrm{\Phi }`$. Thus $`G\mathrm{\Phi }`$, as desired. ∎
###### Lemma 5.11.
If $`\mathrm{\Phi }`$ is a shifted $`(i1,i)`$-dimensional interval, then $`\mathrm{\Phi }`$ satisfies the spectral recursion.
###### Proof.
By induction on the number of non-loop vertices. If $`\mathrm{\Phi }`$ has no non-loop vertices, the result is trivially true. So assume $`\mathrm{\Phi }`$ has vertex set $`\{1,\mathrm{},n\}`$ with $`n1`$.
By Lemma 5.9, it suffices to show $`\mathrm{\Phi }^{}`$ satisfies the spectral recursion. Note that, by Lemma 5.10, $`\mathrm{\Phi }^{}=(\mathrm{\Phi }^+)^{(i1,i)}=(1\mathrm{\Phi }^{})^{(i1,i)}`$ and that $`\mathrm{\Phi }^{}`$ is a shifted $`(i1,i)`$-dimensional interval with one less non-loop vertex (namely, vertex 1) than $`\mathrm{\Phi }^{}`$, and hence fewer non-loop vertices than $`\mathrm{\Phi }`$. By induction, then, $`\mathrm{\Phi }^{}`$ satisfies the spectral recursion. But since taking skeleta (Theorem 4.7) and coning (Corollary 3.10) preserve the property of satisfying the spectral recursion, $`\mathrm{\Phi }^{}`$ also satisfies the spectral recursion. ∎
###### Theorem 5.12.
If $`\mathrm{\Phi }`$ is a shifted interval, then $`\mathrm{\Phi }`$ satisfies the spectral recursion.
###### Proof.
It is immediate that, since $`\mathrm{\Phi }`$ is shifted, so is $`\mathrm{\Phi }^{(i1,i)}`$ for all $`i`$. By Lemma 5.11, each $`\mathrm{\Phi }^{(i1,i)}`$ satisfies the spectral recursion. By Lemma 4.6, then, $`\mathrm{\Phi }`$ satisfies the spectral recursion. ∎
###### Remark 5.13.
It is an easy exercise to verify that, if $`\mathrm{\Phi }`$ is shifted, then so are $`\mathrm{\Phi }e`$, $`\mathrm{\Phi }/e`$, and the two direct summands of $`\mathrm{\Phi }||e`$ from Proposition 2.5.
## 6. Matroid pairs
In this section, we show that some matroid pairs satisfy the spectral recursion, and conjecture that many more do as well. We first set our notation for matroids. For more details, see, e.g., . We let $`𝒞(M)`$ denote the set of circuits of matroid $`M`$, and $`IN(M)`$ denote the independence complex, which is the simplicial complex consisting of the independent sets of $`M`$, and whose Laplacian was first studied in . Our notation for deletion and contraction of intervals and simplicial complexes is consistent with the notation for deletion and contraction of matroids, e.g., $`IN(Me)=IN(M)e`$ and $`IN(M/e)=IN(M)/e`$. Similarly, $`e`$ is a loop of $`M`$ precisely when it is a loop of $`IN(M)`$.
The existence of a strong map $`NN^{}`$ is the natural condition on matroids $`N`$ and $`N^{}`$ to yield nice results about the interval $`(IN(N),IN(N^{}))`$; see, e.g., . Roughly speaking, it means that the matroid structures of $`N`$ and $`N^{}`$ are compatible, comparable to demanding that $`\mathrm{\Delta }`$ and $`\mathrm{\Delta }^{}`$ are shifted on the same ordered vertex set in order for $`(\mathrm{\Delta },\mathrm{\Delta }^{})`$ to be shifted pair. The factorization theorem (e.g., \[10, Theorem 8.2.8\]) says that one characterization of the existence of such a strong map is that $`N=MA`$ and $`N^{}=M/A`$ for some matroid $`M`$ with ground set $`E\dot{}A`$. The main result of this section is that, in the special case where $`|A|=1`$, i.e., $`\mathrm{rank}N\mathrm{rank}N^{}=dimIN(N)dimIN(N^{})=1`$, the interval $`(IN(N),IN(N^{}))`$ satisfies the spectral recursion. We need first one lemma.
###### Lemma 6.1.
If $`M`$ is a matroid with ground element $`e`$, and $`e`$ is not a loop, then
$$(IN(Me),IN(M/e))=\underset{\begin{array}{c}C𝒞(M)\\ eC\end{array}}{}(Ce)IN(M/C).$$
###### Proof.
This is essentially proved in \[2, Lemmas 3.3 and 3.4\]. We sketch the proof here, both for completeness, and to let the language of intervals, not found in the original, simplify some of the steps.
Let $`\mathrm{\Phi }=(IN(Me),IN(M/e))`$. If $`I\mathrm{\Phi }`$, then $`I`$ is independent in $`M`$, but $`I\dot{}e`$ is dependent in $`M`$, and so there is a unique circuit of $`M`$, which we denote by $`\mathrm{ci}_M(e,I)`$, contained in $`I\dot{}e`$. For each circuit $`C𝒞(M)`$, let $`M_C=\{I\mathrm{\Phi }:\mathrm{ci}_M(e,I)=C\}`$. Since each $`I\mathrm{\Phi }`$ has a unique $`\mathrm{ci}_M(e,I)`$, the $`M_C`$’s partition $`\mathrm{\Phi }`$.
In order to show that this partition is an interval direct sum, first note that, if $`I_1M_{C_1}`$ and $`I_2M_{C_2}`$, then $`I_1\dot{}e`$ cannot contain $`C_2`$, since $`\mathrm{ci}_M(e,I)`$ is the unique circuit of $`M`$ contained in $`I\dot{}e`$. Then, since $`C_2I_2\dot{}e`$, it follows that $`I_2I_1`$; similarly $`I_1I_2`$. We conclude that all the $`M_C`$’s are totally unrelated, as desired.
Finally, as in ,
$`M_C`$ $`=\{IIN(Me):CeI\}`$
$`=(Ce)IN((Me)/(Ce))`$
$`=(Ce)IN(M/C).`$
###### Theorem 6.2.
If $`M`$ is a matroid with ground element $`e`$, then the matroid pair $`(IN(Me),IN(M/e))`$ satisfies the spectral recursion.
###### Proof.
If $`e`$ is not a loop of $`M`$, then this is an immediate corollary to Lemmas 3.8 and 6.1, Corollary 3.10, and the fact \[2, Theorem 3.18\]) that matroids satisfy the spectral recursion. If $`e`$ is a loop of $`M`$, then it is a loop of $`IN(M)`$, and so $`(IN(Me),IN(M/e))=(IN(M)e,IN(M)/e)=(IN(M),\mathrm{})=IN(M)`$, which satisfies the spectral recursion. ∎
We are unable to prove anything about $`(IN(N),IN(N^{}))`$ if $`\mathrm{rank}N\mathrm{rank}N^{}>1`$, because we don’t have the analogue of Lemma 6.1 above. Still, experimental evidence on randomly chosen matroids supports the following natural conjecture.
###### Conjecture 6.3.
If there is a strong map $`NN^{}`$ between matroids $`N`$ and $`N^{}`$, then the interval $`(IN(N),IN(N^{}))`$ has integral Laplacian eigenvalues, and satisfies the spectral recursion.
## 7. Acknowledgements
I am grateful to Vic Reiner for suggesting strong maps on matroids to me. |
warning/0507/nucl-th0507068.html | ar5iv | text | # Probing the density dependence of the symmetry potential in intermediate energy heavy ion collisions
## I Introduction
The study of isospin effects in nuclear matter or finite nuclei is a well established topic in nuclear physics, as well as in nuclear astrophysics Pra97 ; Lat01 . On the mean field level, this means the contribution of the symmetry energy of particles to the total energy (see, e.g., the recent Refs. Cha97 ; LiB02 ; LiB05 ; Gai04 ; Gai042 ; Riz04 ; LiQ01 ; LiQ05 ), and on the two-body collision level, detailed experimental data on the free proton-proton and neutron-proton scattering cross sections have been obtained PDG96 as a function of energy ranging from the Coulomb barrier to ultra high energy interactions. At low energies (up to several hundreds of MeV/nucleon), large differences are visible between the proton-proton and the neutron-proton cross sections. In heavy ion collisions (HICs), it is also necessary to study the effect of density on the isospin-dependent nucleon-nucleon cross sections.
Recently, we have explored the density- and temperature-dependence of the nucleon-nucleon elastic scattering cross sections in collisions between neutron-rich nuclei at intermediate energies LiQ04 . We found a density dependence of the elastic scattering cross sections. This work is based on the theory of quantum hydrodynamics (QHD), in which the interaction between nucleons is described by the exchange of $`\sigma `$, $`\omega `$, $`\rho `$, and $`\delta `$ mesons. Since, in the QHD theory, both the mean field and the collision term originate from the same Lagrangian density, the density-dependence of the symmetry energy is our next subject to investigate. Currently the exploration of the isospin dependence of the nuclear interaction ( especially in exotic and very isospin-asymmetric systems) has gained high interest Akm97 ; Hei00 ; LiB02 ; LiB05 ; Gai04 ; Gai042 ; LiQ05 . In particular, the density dependence of the symmetry energy in dense nuclear matter has to be studied in detail, because it suffers from large uncertainties in the predictions of various theoretical models Bro00 ; Mar01 . Based on the QHD theory, with a similar effective Lagrangian density Hof01 ; LiuB02 ; Gai04 , the main uncertainty of the density dependence of the symmetry energy results from the uncertainties of the density-dependent coupling strengths of various meson-baryon interactions, especially that of $`\rho N`$ and $`\delta N`$ interactions Gai04 . Also, in forthcoming experiments at the Rare Isotope Accelerator (RIA) laboratory (USA) and at the new international accelerator facility FAIR (Facility for Antiproton and Ion Research) at the Gesellschaft für Schwerionenforschung (GSI, Germany), more neutron-rich beams are planned to be adopted, and hence call for further theoretical studies on isospin-dependent HICs in the intermediate energy region.
In order to obtain detailed information on the density dependence of the symmetry energy in dense nuclear matter, quite a few sensitive probes were brought up for experiments in recent years: e.g., the $`\pi ^{}/\pi ^+`$ yield ratio, the pion flow, the transverse momentum distribution of $`\pi ^{}/\pi ^+`$ ratios, the neutron-proton differential flow LiB02 , and the isospin equilibration and stopping Bass98 ; Rami00 ; Gai042 . More recently, we have also investigated LiQ05 the sensitivity of the $`\mathrm{\Sigma }^{}/\mathrm{\Sigma }^+`$ ratio on the density-dependent symmetry potentials. Further investigations seem to be worthwhile to shed light on the rather unclear isospin-independent and -dependent mean-field potentials of the hyperons. Since the contribution of the symmetry energy to the whole dynamics of HICs is rather small at SchwerIonen Synchrotron (SIS) energies, it is all the more difficult to explore the density dependence of the symmetry energy based on the current experimental situation. Hence, in theory, more sensitive probes should be searched for.
Also, such a detailed analysis is quite necessary and important because of the following arising questions: Is the effect of the density-dependent symmetry energy on the $`\pi ^{}/\pi ^+`$ ratio influenced by the beam energy? In which momentum, rapidity, or phase-space, region are the pion yields more suitable to probe the density-dependent symmetry energy? And, how does the Coulomb potential of mesons affect the emission of pions? Gaitanos et al. Gai04 ; Gai042 found that for beam energies higher than about $`2A`$ GeV, the sensitivity of the $`\pi ^{}/\pi ^+`$ ratio to the form of the symmetry energy at high densities is strongly reduced. Furthermore, they found that the $`\pi ^{}/\pi ^+`$ ratio at high transverse momenta $`p_t`$ could offer sound information on the density dependence of the symmetry energy. B.-A. Li et al. LiB05 revisited the $`\pi ^{}/\pi ^+`$ ratio by using an up-dated, new version of the isospin-dependent Boltzmann-Uehling-Uhlenbeck (IBUU04) model, in which an isospin- and momentum-dependent single nucleon potential was adopted. They found that the sensitivity of the $`\pi ^{}/\pi ^+`$ ratio on the density-dependence of the symmetry energy becomes obvious after considering the momentum-dependent single nucleon potential. They also noticed the effect of the Coulomb potential on the $`\pi `$ production, especially on the transverse momentum spectra. Their work implied that it is sufficient to measure accurately the low-energy (or the low-transverse-momentum $`p_t`$) pions, instead of the whole spectrum. In this paper, we address these aspects of the problem on the basis of the ultrarelativistic quantum molecular dynamics (UrQMD) model.
In this work, we also investigate the $`K^0/K^+`$ production ratio as a probe of the density-dependent symmetry potential in dense nuclear matter. Based on the constituent quark picture the $`K^0`$ meson contains one $`d`$-quark while the $`K^+`$ contains one $`u`$-quark, as well as one $`\overline{s}`$-quark in each meson. This is closely related to the isospin asymmetry of the nuclear medium at SIS energies. Furthermore, experimentally, the effects of system parameters, such as beam energy or impact parameter, on the sensitive probes also have to be kept in mind. Theoretically, the uncertainty of the isospin-independent nuclear equation of state (EoS), as well as the contribution of the Coulomb potential of mesons, might alter the calculated results of these probes, hence, they could even modify the conclusions and should be checked carefully.
In the UrQMD model (version 1.3) Bass98 ; Bleicher99 ; Web03 ; Bra04 , adopted here, we find that most of the calculations can simultaneously reproduce many experimental measurements, which offers a good starting point for studying the isospin effects at SIS energies. In this work, a ’hard’ and a ’soft’ Skyrme-type EoS, without momentum dependence, are adopted for central ($`b=02\mathrm{fm}`$) or near-peripheral ($`b=58\mathrm{fm}`$) $`{}_{}{}^{132}Sn+^{132}Sn`$ reactions at beam energies $`E_\mathrm{b}=0.5A`$ and $`1.5A`$ GeV. We consider phenomenological density-dependent symmetry potentials, which will be discussed in more detail in the next section. The Coulomb potentials of the mesons are switched on or off in order to analyze more conveniently the contributions of the various symmetry potentials to the dynamical evolution of the hadrons in the nuclear medium.
The paper is arranged as follows: In Section II, we clarify the inclusion of the isospin-dependent part of the mean field in the UrQMD transport model. In Section III, some basic isospin effects on the dynamics of the baryons (here the nucleons and $`\mathrm{\Delta }(1232)`$) in SIS energy HICs are shown. In Section IV, firstly the effects of the isospin-independent EoS, the beam energy, and the impact parameter on the emitted pion yields and the $`\pi ^{}/\pi ^+`$ production ratios are discussed. Then, the various phase-space distributions of pion yields and the $`\pi ^{}/\pi ^+`$ ratios, with the different density dependences of the symmetry potentials, are investigated. At the end of this section, the sensitivity of the $`K^0/K^+`$ ratio to the density dependence of the symmetry potential is discussed. Finally, the conclusions and outlook are given in Section V.
## II The treatment of the potential update of hadrons in the UrQMD model
The initialization of neutrons and protons, the corresponding Pauli blocking, the Coulomb potential of baryons, and the isospin dependence of the nucleon-nucleon cross sections have been introduced explicitly in the standard UrQMD model. In order to study the isospin effects in intermediate energy HICs, it is required to further introduce the symmetry potential of baryons.
The isospin-dependent EoS for asymmetric nuclear matter can be expressed as (see, e.g., Bom91 )
$$e(u,\delta )=\frac{ϵ(u,\delta )}{\rho }=e_0(u)+e_{\mathrm{sym}}(u)\delta ^2,$$
(1)
where $`u=\rho /\rho _0`$ is the reduced nuclear density and $`\delta =(\rho _n\rho _p)/\rho `$ is the isospin asymmetry in terms of neutron ($`\rho _n`$) and proton ($`\rho _p`$) densities. The $`e_0(u)`$ term is the isospin-independent part, which includes the Skyrme and the Yukawa potentials in the UrQMD model Bass98 . The Yukawa parameter is related to the Skyrme parameters since in infinite nuclear matter the contribution of the Yukawa potential to the total energy acts like a two-body Skyrme contribution. For comparison, both a soft ($`K=200`$ MeV) and a hard ($`K=300`$ MeV, the default parametrization in this work) EoS are adopted in this work. $`e_{\mathrm{sym}}`$ is the symmetry energy per nucleon, in which the kinetic ($`v_{\mathrm{sym}}^{\mathrm{kin}}`$) and potential ($`v_{\mathrm{sym}}^{\mathrm{pot}}`$) contributions are included. The average symmetry potential energy can be expressed as
$$v_{\mathrm{sym}}^{\mathrm{pot}}=e_\mathrm{a}F(u),$$
(2)
where $`e_\mathrm{a}`$ is the symmetry-potential strength and $`F(u)`$ is the density-dependent part. If the Fermi-gas model is adopted, the symmetry-potential strength $`e_a`$ is related to the symmetry energy at normal density, $`S_0`$, by
$$S_0e_\mathrm{a}+\frac{ϵ_F}{3}.$$
(3)
Here, $`ϵ_F`$ is the Fermi kinetic energy at normal nuclear density.
The relativistic mean field calculations Gai04 show a small variation of the symmetry kinetic energy extracted from various models, and the uncertainty of the symmetry energy with nuclear density results mainly from the symmetry potential energy. Concerning the symmetry potential energy, both the symmetry energy coefficient $`S_0`$ and the density dependence of the symmetry energy are quite uncertain. For $`S_0`$, deduced from the isovector GDR in $`{}_{}{}^{208}Pb`$ and from the available data of differences between neutron and proton radii for $`{}_{}{}^{208}Pb`$ and several $`Sn`$ isotopes, a rather small range of values is $`32\mathrm{MeV}S_036\mathrm{MeV}`$ Vre03 . More recently, a new value of $`S_031`$ MeV was obtained, based on the consistent folding analysis of the $`p(^6He,^6He)p`$ elastic scattering and $`p(^6He,^6Li^{})n`$ charge exchange reaction data measured at $`E_{\mathrm{Lab}}=41.6`$ MeV Kho05 , while $`S_0=34`$ MeV was obtained in a calculation made within the relativistic Brueckner framework, using the Bonn A potential Dal04 . Thus, it is necessary to further investigate the uncertainty of the $`S_0`$ value LiQ04a . In this paper, however, the value $`S_0=34`$ MeV is adopted, since we endeavor to investigate the density dependence of the symmetry potential energy.
In order to mimic the strong density dependence of the symmetry potential at high densities, we adopt the form of $`F(u)`$, used in LiB02 , as
$$F(u)=\{\begin{array}{c}F_1=u^\gamma \gamma >0\hfill \\ F_2=u\frac{au}{a1}a>1\hfill \end{array}.$$
(4)
Here, $`\gamma `$ is the strength of the density dependence of the symmetry potential. We choose $`\gamma =0.5`$ and $`1.5`$, denoted as the symmetry potentials F05 and F15, respectively. $`a`$ (in F<sub>2</sub>) is the reduced critical density; for $`u>a`$, the symmetry potential energy is negative. We adopt $`a=3`$ in this paper and the respective symmetry potential is named as Fa3. The symmetry potentials F05, F15, and Fa3 are shown in Fig. 1 as a function of the reduced nuclear density $`u`$, compared with a linear density-dependent symmetry potential.
From Fig. 1, it is apparent that for $`u<1`$, F05$`>`$Fa3$`>`$F15, whereas for $`u>1`$, F15$`>`$F05$`>`$Fa3. F05 is larger than Fa3 for all densities. This means that for neutron-rich HICs at intermediate energies, less neutrons are pushed into the low density region ($`u<1`$) as well as into the high density region ($`u>1`$) for density dependence Fa3, as compared to F05. In other words, the symmetry-potential parametrization F05 is always stiffer than Fa3 (at both subnormal and higher densities), except at normal nuclear density.
Besides nucleons (N), the resonances N(1440), $`\mathrm{\Delta }`$(1232), and the hyperons $`\mathrm{\Lambda }`$ and $`\mathrm{\Sigma }`$ should be considered for SIS energy HICs, where the $`\mathrm{\Delta }`$(1232) is dominating, denoted as $`\mathrm{\Delta }`$ in short. For simplicity, the isospin-independent part of the EoS, $`e_0(u)`$, for all other baryons is taken to be the same as for the nucleons. The symmetry potentials for the resonances are obtained through the constants of isospin coupling (the Clebsch-Gordan coefficients) in the process of $`\mathrm{\Delta }`$ \[or $`N^{}(1440)`$\] $`\pi N`$. For hyperons, based on the analysis of the Lane potential Lan69 ; Dab99 , we simply take the symmetry potential of hyperons to be nucleon-like. However, the symmetry potential of excited states of hyperons is not considered, for lack of information.
Combining both, resonances and hyperons, we express the symmetry potential in a unified form, which reads as
$$v_{\mathrm{sym}}^\mathrm{B}=\alpha v_{\mathrm{sym}}^n+\beta v_{\mathrm{sym}}^\mathrm{p},$$
(5)
where the values of $`\alpha `$ and $`\beta `$ for different baryons (B) are listed in Table 1. From this table we can see that the symmetry potentials of $`\mathrm{\Delta }^{}`$ and $`\mathrm{\Sigma }^{}`$ are neutron-like and those of $`\mathrm{\Delta }^{++}`$ and $`\mathrm{\Sigma }^+`$ are proton-like. On the other hand, the symmetry potentials of $`\mathrm{\Delta }^0`$, $`\mathrm{\Delta }^+`$, $`N^{}`$(1440), $`\mathrm{\Sigma }^0`$, and $`\mathrm{\Lambda }`$ are a mixture of the neutron and proton symmetry potentials.
In the standard UrQMD model (version 1.3), the cascade mode is usually adopted for produced mesons. In order to investigate the phase-space distributions of pions, especially the transverse momentum distributions in intermediate energy HICs, the Coulomb potential of mesons should be considered explicitly Pel97 ; LiB05 ; Gai04 . Thus, in this paper, we also investigate the contribution of the Coulomb potentials between mesons and baryons. Other mean-field potentials of mesons are not yet considered.
## III Effects of the symmetry potential on the dynamics of nucleons and $`\mathrm{\Delta }`$’s
First of all, let us discuss the basic consequences of the density dependence of the symmetry potentials in neutron-rich HICs at SIS energies. Fig. 2 (top) shows the time evolution of the neutron and proton numbers for the symmetry potentials F15, F05, and Fa3. Here, we do not distinguish whether the nucleons are free or bound in heavier fragments. The beam energies are $`E_\mathrm{b}=0.5A`$ and $`1.5A`$ GeV and central collisions ($`b=02\mathrm{fm}`$) are chosen. From this plot, we notice that independent of the choice of the density dependence of the symmetry potential, the neutron numbers reach their minima at $`15\mathrm{fm}/c`$ and $`10\mathrm{fm}/c`$, for the beam energies $`E_\mathrm{b}=0.5A`$ and $`1.5A`$ GeV, respectively. At this time the central nuclear density reaches its maximum and some of the neutrons are excited (to N or $`\mathrm{\Delta }`$) or converted to other baryons (e.g., hyperons) through various baryon-baryon or meson-baryon collisions. We also notice that the higher the beam energy, the more excited baryon states are produced. There are obviously less excited protons than neutrons at the compression stage. In other words, at the compression stage, the effect on the neutron number is stronger than on the proton number, which is due to the neutron-rich nuclear environment. After the decay of these unstable baryons, the emitted mesons (especially, the pions) alter the neutron/proton ratio (the neutron number decreases, and, correspondingly, the proton number increases), and hence the ratio tends to become unity with an increase of the beam energy. In addition the effect of the density-dependent symmetry potentials on the dynamical process is smaller at higher energies LiQ05 . With a softer symmetry potential Fa3, more neutrons are excited at the compression stage, which is clearly due to more neutrons being kept by the softer symmetry potential in the high density region, as implied from Fig. 1.
In Fig. 2 (bottom), the neutron/proton ratio of the all (free and bound) nucleons at freeze-out time ($`t_\mathrm{f}=50\mathrm{fm}/c`$, the maximum time used in the top plot) is shown as a function of the normalized rapidity ($`y_c^{(0)}=y_c/y_{\mathrm{beam}}`$, where $`y_c`$ is the rapidity of the particle in the center-of-mass system and $`y_{\mathrm{beam}}`$ is the beam rapidity). At $`E_\mathrm{b}=0.5A`$ GeV, the density dependence of the symmetry potential influences the neutron/proton ratio strongly, whereas at $`E_\mathrm{b}=1.5A`$ GeV the sensitivity is reduced, especially at midrapidity. This happens not only due to the relatively weak effect of the EoS, as compared to the two-body collision dynamics, but also due to the nucleon-nucleon cross sections at higher energies that depend less on isospin. Secondly, in the projectile and target rapidity regions, the $`n/p`$ ratio is more sensitive to the density dependence of the symmetry potential as compared to the midrapidity region. Less collisions take place in the projectile and target rapidity regions than in the midrapidity region (see also, Fig. 3). Thirdly, the $`n/p`$ ratio is larger with a softer symmetry potential Fa3 than that with F15, and vice versa in the projectile-target regions. The nucleons in the midrapidity region represent mainly the behavior of symmetry potential at high densities, where more neutrons are kept with a soft symmetry potential. We should also notice that at midrapidity, the difference of the ratios with the F05 and Fa3 symmetry potentials is almost negligible. From Fig. 1 we can see that F05 is always stiffer than Fa3 at both low and high densities, thus the $`n/p`$ ratio is also influenced by the density dependence of the symmetry potential at subnormal densities. Therefore, in order to investigate any isospin dependences, it is very important to know the correct density-dependent form of the symmetry potential at both, low and high densities.
Fig. 3 shows the calculated rapidity distribution of the collision number of neutrons and protons (upper plot), and the corresponding ratios (lower plot). Evidently, at both energies, the collision number of nucleons has a strong maximum at midrapidity. The collision number increases with beam energy, especially for neutrons because of the neutron-rich system. The average proton collision number is always larger than the average neutron collision number because of the differences in the nucleon-nucleon cross sections at intermediate energies PDG96 . When the beam energy increases from $`0.5A`$ to $`1.5A`$ GeV, this isospin effect of the nucleon-nucleon cross section is largely reduced, so that the ratio between the neutron and proton collision numbers approach unity, which is shown in the lower part of Fig. 3. In the midrapidity region, the collision ratio of neutrons and protons is $`82\%`$ at $`E_\mathrm{b}=0.5A`$ GeV, which increases to $`95\%`$ at $`E_\mathrm{b}=1.5A`$ GeV.
Since, in the SIS energy region, $`\pi `$-mesons are mainly produced from decaying $`\mathrm{\Delta }`$’s, it is also important to investigate the dynamics of the $`\mathrm{\Delta }`$’s with respect to the symmetry potential. Fig. 4 shows the time evolution of the various components of $`\mathrm{\Delta }`$ (upper plot) and the ratio $`\mathrm{\Delta }^{}/\mathrm{\Delta }^{++}`$ (lower plot) for the symmetry potentials F15 and Fa3 and for central <sup>132</sup>Sn+<sup>132</sup>Sn collisions at the two energies $`E_\mathrm{b}=0.5A`$ and $`1.5A`$ GeV. In the upper plot, only the case $`E_\mathrm{b}=0.5A`$ GeV is shown since, at $`E_\mathrm{b}=1.5A`$ GeV, the effect of the density dependence of the symmetry potentials on the $`\mathrm{\Delta }`$ production is almost negligible. This is supported by the time evolution of the ratio $`\mathrm{\Delta }^{}/\mathrm{\Delta }^{++}`$ at $`E_\mathrm{b}=1.5A`$ GeV in the lower plot.
From the upper part of Fig. 4, it is clear that the effect of the density-dependent symmetry potentials on the time evolution is strongest for the $`\mathrm{\Delta }^{}`$ production, like for neutrons in Fig. 2. From the time evolution of the $`\mathrm{\Delta }^{}/\mathrm{\Delta }^{++}`$ ratio in the lower plot of Fig. 4, we further notice that, similar to the case of nucleons, the effect of the density-dependent symmetry potentials on the $`\mathrm{\Delta }^{}/\mathrm{\Delta }^{++}`$ ratio is largely reduced at the higher beam energy $`1.5A`$ GeV. And the time evolution of this ratio is nearly flat since, at this energy, the effect of the Coulomb and the symmetry potentials becomes small. At the lower beam energy $`0.5A`$ GeV and during the compression stage ($`t12\mathrm{fm}/c`$), the $`\mathrm{\Delta }^{}/\mathrm{\Delta }^{++}`$ ratio decreases with time while it increases at the later stage, the expansion stage. This is apparently due to the dynamically mutual interaction between the Coulomb and the symmetry potentials: along the time evolution at the compression stage, more neutrons are pushed out because the relative strength of the symmetry potential as compared to the Coulomb potential increases. This can be understood from the study on the ratio of preequilibrium neutron number to proton number in the intermediate-energy neutron-rich HICs in Ref. Liu:2001ud : this ratio is larger than the initial neutron/proton ratio of the colliding system; at the expansion stage, more protons are pushed out due to the stronger Coulomb potential as compared to the symmetry potential. Note that the increase of the $`\mathrm{\Delta }^{}/\mathrm{\Delta }^{++}`$ ratio at the late stage is stronger for the symmetry potential Fa3 than for F15. The time evolution of the $`\mathrm{\Delta }`$ abundancies, depending on the symmetry potential, should also affect the production of pions.
Besides the $`\pi `$-yields, the pion spectra should also be influenced by the decay of $`\mathrm{\Delta }`$’s. Fig. 5 shows the transverse momentum spectrum of $`\mathrm{\Delta }`$’s (upper plot) and the $`\mathrm{\Delta }^{}/\mathrm{\Delta }^{++}`$ ratios (lower plot) at $`t=20\mathrm{fm}/c`$ and $`E_\mathrm{b}=0.5A\mathrm{GeV}`$, for the symmetry potentials F15, F05, and Fa3. We notice from the upper part of Fig. 5 that, similar to the neutrons in Fig. 2, the effect of different density-dependent symmetry potentials is strongest for the $`\mathrm{\Delta }^{}`$. From the lower plot of Fig. 5 we can see that, at lower $`p_t^{\mathrm{cm}}`$ ($`p_t^{\mathrm{cm}}0.8\mathrm{GeV}/c`$), the $`\mathrm{\Delta }^{}/\mathrm{\Delta }^{++}`$ ratios are quite different for the symmetry potentials F15 and F05 (or Fa3), and the results of the F05 and Fa3 symmetry potentials are nearly the same. These results are similar to the rapidity distribution of the neutron/proton ratio, shown in Fig. 2. $`\mathrm{\Delta }`$’s with low transverse momenta are not produced from very high densities, namely around normal density, where the difference between F05 and Fa3 is quite small and the effect of F05 and Fa3 on the $`\mathrm{\Delta }^{}/\mathrm{\Delta }^{++}`$ ratio are cancelled strongly between low and high densities. For $`p_t^{\mathrm{cm}}0.8\mathrm{GeV}/c`$, the differences between the results of F05 and Fa3 become distinguishable because $`\mathrm{\Delta }`$’s with the high transverse momenta are mainly produced from the central high-density region. Thus the $`\mathrm{\Delta }^{}/\mathrm{\Delta }^{++}`$ ratio at high transverse momenta can provide clearer information on the density dependence of the symmetry potential. In the next section, we investigate the question to what extent this property could be transferred to $`\pi `$’s, especially when the contribution of the Coulomb potential of mesons is also considered.
## IV Effects of the symmetry potential on the $`\pi `$ and $`K`$ production
Fig. 6 shows the time evolution of the $`\pi `$ abundancies (upper plot) and their ratios (lower plot) for the density-dependent symmetry potentials F15, F05, and Fa3. Here, the Coulomb potential of mesons is switched off. However, for comparison, we also provide the results (for the case of F05 and at $`E_\mathrm{b}=0.5A\mathrm{GeV}`$) when the Coulomb potential of mesons is taken into account (solid points in Fig. 6 at time $`t=45\mathrm{fm}/c`$). Apparently, the effect of the Coulomb potential of mesons on the $`\pi `$-yields is quite small. We also find that the effect of the different symmetry potentials on the $`\pi ^{}/\pi ^+`$ ratio is hardly affected by the Coulomb potential of mesons although it alters the absolute value of the ratio. The number of $`\pi ^{}`$ grows much stronger than the number of $`\pi ^+`$, almost independent of the choice of the symmetry potentials F05 and Fa3. The $`\pi ^{}`$-yields are, however, smaller for F15. So is the $`\pi ^{}/\pi ^+`$ ratio at $`E_\mathrm{b}=0.5A\mathrm{GeV}`$. On the other hand, for the higher beam energy $`E_\mathrm{b}=1.5A`$ GeV, the ratios are almost independent of the symmetry potentials. (These results are clearly due to the properties of $`\mathrm{\Delta }`$’s, shown in Figs. 4 and 5). Therefore, it is necessary to pay attention to the energy dependence of the isospin effect on the $`\pi ^{}/\pi ^+`$ ratios as well.
Next, we show the dependence of the $`\pi ^{}/\pi ^+`$ ratio on various isospin-independent EoS and impact parameter $`b`$. In view of the above noted beam-energy dependence, we only consider the lower beam energy $`E_\mathrm{b}=0.5A`$ GeV in the following. Fig. 7 shows the results for two EoS and two centralities. The effect of the density-dependent symmetry potentials on the $`\pi ^{}/\pi ^+`$ ratios is almost not affected by the uncertainty of the isospin-independent EoS. That is, the absolute shift in the $`\pi ^{}/\pi ^+`$ ratio due to different symmetry potentials remains almost the same, independent of the stiffness of the isospin-independent EoS. On the other hand, at larger impact parameters ($`b=58\mathrm{fm}`$), the $`\pi ^{}/\pi ^+`$ ratio is even more sensitive to the density dependence of the symmetry potential.
The rapidity $`y_c`$ and the polar angle $`\theta _c`$ (in the center-of-mass system) distributions of $`\pi ^{}`$ and $`\pi ^+`$ for various density-dependent symmetry potentials are shown in Fig. 8. The influence of the density-dependent symmetry potentials on the $`\pi `$-multiplicity distributions is strongest at midrapidity and at polar angles around $`90^0`$. This holds in particular for the $`\pi ^{}`$ distribution, which was also seen in LiB05 for the kinetic energy spectrum of pions in central <sup>132</sup>Sn+<sup>124</sup>Sn reactions at $`E_\mathrm{b}=0.4A\mathrm{GeV}`$.
The upper two plots of Fig. 9 show the transverse momentum distributions of $`\pi ^{}`$ and $`\pi ^+`$, with and without the contribution of the Coulomb potentials of mesons, as well as with a rapidity-cut ($`|y_c|<0.2`$) (left plots) and without any kinematical cut (right plots), for the symmetry potential F05. This is important because, in recent years, several experiments on charged pion production at intermediate energies were performed by the FOPI Collaboration at GSI where the possible contribution of the Coulomb potentials of pions to the $`\pi `$-transverse momentum distribution was also discussed Pel97 ; Hon05 . Although, as shown in Fig. 6, the effect of the Coulomb potential of mesons on the total $`\pi `$ yield is small, Fig. 9 demonstrates that it strongly influences the momentum distribution; the Coulomb potential of mesons always shifts the $`\pi ^{}`$ to lower, and the $`\pi ^+`$ to higher $`p_t^{\mathrm{cm}}`$ for the neutron-rich systems, in line with Ref. LiB05 . With the rapidity-cut $`|y_c|<0.2`$, this effect becomes even more pronounced; as a result, the transverse momentum distribution of the $`\pi ^{}/\pi ^+`$ ratio at low $`p_t^{\mathrm{cm}}`$ becomes steeper than the one without the rapidity cut (lower plots in Fig. 9 show the corresponding $`\pi ^{}/\pi ^+`$ ratios for the symmetry potentials F15, F05, and Fa3). If the Coulomb potential of mesons is switched off, the $`\pi ^{}/\pi ^+`$ ratio without rapidity-cut is nearly constant at low transverse momenta ($`p_t^{\mathrm{cm}}<0.3\mathrm{GeV}/c`$). With the rapidity-cut ($`|y_c|<0.2`$) it increases weakly with $`p_t^{\mathrm{cm}}`$ because of the contribution of the Coulomb potential of $`\mathrm{\Delta }`$’s. At higher transverse momenta ($`p_t>0.4\mathrm{GeV}/c`$), and with rapidity-cut (lower-left plot of Fig. 9), the effect of the density-dependent symmetry potentials on the $`\pi ^{}/\pi ^+`$ ratios is noticeable; the difference is also obvious for F05 and Fa3. It is recalled that a similar phenomenon can be seen in Fig. 5 for the $`\mathrm{\Delta }^{}/\mathrm{\Delta }^{++}`$ ratios. However, when the Coulomb potentials of mesons are taken into account, the effect of the density-dependent symmetry potentials on the $`\pi ^{}/\pi ^+`$ ratios at high transverse momenta is largely reduced, regardless of a rapidity cut. It was pointed out in LiB05 that the Coulomb potential is stronger than the symmetry potential, and the Coulomb potential acts directly on the charged pions but the symmetry potential does not, such that the $`\pi `$’s emitted from the dense region, which have higher $`p_t^{\mathrm{cm}}`$, may be influenced much more by their Coulomb potential than by the symmetry potential. Considering the limit of experiments so far, we claim that the $`\pi ^{}/\pi ^+`$ production ratio at $`p_t^{\mathrm{cm}}=0.1\mathrm{GeV}/c`$ is a suitable candidate for probing the density dependence of the symmetry potential, which is discussed in more detail in LiQ052 .
Fig. 10 shows the density distribution of pion emission, with and without Coulomb potential of mesons. In the upper plot, the percentages of each component of $`\pi `$ and of the total $`\pi `$ yield are shown as a function of the reduced density $`u`$, in steps of $`\mathrm{\Delta }u=0.2`$, with the Coulomb potential of mesons included. Most pions are emitted from densities higher than normal nuclear density (especially from the density region $`u12`$); a few pions ($`16\%`$) are emitted from subnormal densities. This follows from Fig. 5 where quite a few $`\mathrm{\Delta }`$’s are transported close to normal nuclear density. As these $`\mathrm{\Delta }`$’s decay, some of them also appear at subnormal densities.
In the lower part of Fig. 10, the ratios of the percentages of $`\pi ^{}`$ and $`\pi ^+`$ are shown as a function of the reduced density $`u`$, with and without the Coulomb potential of mesons. The ratios are always larger at lower densities than at higher densities since at the late stage, where the nuclear density decreases, the $`\mathrm{\Delta }^{}/\mathrm{\Delta }^{++}`$ ratio (shown in Fig. 4) increases with time. When the Coulomb potential of mesons is taken into account, more $`\pi ^{}`$ mesons are decelerated because of the larger amount of the negatively charged particles in the low density region, and are thus emitted with low momenta which leads to the rise of the $`\pi ^{}/\pi ^+`$ ratios at low $`p_t^{\mathrm{cm}}`$, as shown in Fig. 9.
Besides the $`\pi ^{}/\pi ^+`$ ratios, the $`K^0/K^+`$ ratios are also thought to be a suitable candidate to probe the density dependence of the symmetry potential in dense nuclear matter, as already discussed in the Introduction. Fig. 11 shows the time evolution of the $`K^0`$ and $`K^+`$ abundancies, as well as their ratios $`K^0/K^+`$, for central $`{}_{}{}^{132}Sn+^{132}Sn`$ collisions at beam energy $`1.5A`$ GeV and for the symmetry potentials F15 and Fa3. Similar to the $`\pi `$ production, the effect of different density dependences of the symmetry potential on the kaon yields is quite reduced at higher energies.
In SIS energy HICs, kaons are emitted mainly from the high density region and at the early stage of the reaction. The dominant production channels are $`BBBB^{}BKY`$ and $`B\pi B^{}KY`$ (here $`Y`$ represents a hyperon). With a softer symmetry potential, in the neutron-rich nuclear medium, more neutrons (protons) are shifted to the high (low) density region and hence more negatively (less positively) charged $`B^{}`$ are produced, and the $`K^0/K^+`$ ratio increases. Thinking of the energy dependence, the kaon production at energies much lower than its threshold ($`1.58`$ GeV from the nucleon-nucleon interaction in free space) might be more sensitive to the density dependence of the symmetry potential. As an example, we have also calculated the kaon yields from the reaction $`{}_{}{}^{208}Pb+^{208}Pb`$ at $`E_\mathrm{b}=0.8A`$ GeV and $`b=79\mathrm{fm}`$ with the symmetry potentials F15 and Fa3: the $`K^0/K^+`$ ratio for F15 is about $`1.25`$, whereas it is about $`1.4`$ for Fa3. We should point out that these results on kaons might be relatively rough since we do not consider any kaon-nucleon mean-field potential LiG95 ; Fuchs02 ; Nek02 . As far as we know, there is no explicit calculation about the difference of $`K^+`$\- and $`K^0`$-nucleon potentials in the nuclear medium.
## V Summary
In this paper, we have investigated the role of the density-dependent symmetry potential in heavy ion collisions (HICs) at SIS energies, based on the UrQMD model (v1.3). The contribution of the Coulomb potential of mesons has been studied, in order to better understand the effects of the density-dependent symmetry potential on the dynamics of pions produced in neutron-rich HICs. The calculated results show a strong beam-energy dependence of the effect of the density-dependent symmetry potentials on the production ratios $`\mathrm{\Delta }^{}/\mathrm{\Delta }^{++}`$ as well as $`\pi ^{}/\pi ^+`$. The uncertainty in the isospin-independent EoS alters the values of both the pion yields and the $`\pi ^{}/\pi ^+`$ ratios, but has almost no effect on the role of the density-dependent symmetry potential on the $`\pi ^{}/\pi ^+`$ ratios. The impact parameter is found to be important for the $`\pi ^{}/\pi ^+`$ ratios; for a larger impact parameter the effect of different density-dependent symmetry potentials on the $`\pi ^{}/\pi ^+`$ ratios becomes stronger.
The Coulomb potential of mesons changes the transverse momentum distribution of the $`\pi ^{}/\pi ^+`$ ratios significantly, though it leaves the total $`\pi ^{}`$ and $`\pi ^+`$ yields almost unchanged. We find that the negatively charged pion yields, especially at midrapidity and low transverse momenta, as well as the $`\pi ^{}/\pi ^+`$ ratios at low transverse momenta, could be sensitive to the density-dependent symmetry potential in a dense nuclear matter. Quite a few pions are still produced at subnormal densities and thus are affected by the density-dependent symmetry potential at subnormal densities. These studies are of interest for forthcoming experiments at RIA (USA) and FAIR/GSI (Germany).
In this work, we have also investigated the yields of K<sup>0</sup> and K<sup>+</sup> mesons and the ratios $`K^0/K^+`$ from neutron-rich HICs at $`E_\mathrm{b}=1.5A`$ GeV. It is shown that they do not seem to be suitable to investigate the density dependence of the symmetry potential in dense nuclear matter. However, kaon production at energies much lower than threshold might improve the situation and is worth investigating further. Meanwhile, it is necessary to consider the difference of the $`K^+`$\- and $`K^0`$-nucleon potentials in the nuclear medium.
## Acknowledgments
We would like to acknowledge valuable discussions with S. Schramm and A. Mishra. Q. Li thanks the Alexander von Humboldt-Stiftung for a fellowship. RK Gupta thanks Deutsche Forschungsgemeinschaft (DFG) for a Mercator Guest Professorship. This work is partly supported by the National Natural Science Foundation of China under Grant No. 10255030, by GSI, BMBF, DFG, and Volkswagen Stiftung. |
warning/0507/gr-qc0507082.html | ar5iv | text | # Weyl electrovacuum solutions and gauge invariance
## Abstract
It is argued that in Weyl electrovacuum solutions the linear term in the metric cannot be eliminated just on grounds of gauge invariance. Its importance is stressed.
In general relativity static electric fields alter the metric of spacetime through their energy-momentum tensor one
$$T_\nu ^\mu =\frac{\epsilon }{4\pi }\left(F^{\mu \alpha }F_{\nu \alpha }\frac{1}{4}\delta _\nu ^\mu F^{\alpha \beta }F_{\alpha \beta }\right),$$
(1)
where
$$F_{\mu \nu }=_\mu A_\nu _\nu A_\mu $$
(2)
is the electromagnetic tensor, $`A_\mu =(\overline{\varphi },0,0,0)`$ is the four-potential and $`\epsilon `$ is the dielectric constant of the medium. $`T_\nu ^\mu `$ enters the r.h.s. of the Einstein equations
$$R_\nu ^\mu =\kappa T_\nu ^\mu ,$$
(3)
where $`\kappa `$ is the Einstein constant. We have taken into account that $`T_\mu ^\mu =0`$. In addition, the Maxwell equations are coupled to gravity through the covariant derivatives of $`F_{\mu \nu }`$
$$F_{;\nu }^{\mu \nu }=\frac{1}{\sqrt{g}}\left(\sqrt{g}F^{\mu \nu }\right)_\nu =0,$$
(4)
where $`g`$ is the metric’s determinant and usual derivatives are denoted by subscripts. The electric field is $`E_\mu =F_{0\mu }=\overline{\varphi }_\mu `$ . Obviously, $`T_\nu ^\mu `$ from Eq. (1) contains only quadratic terms in $`\overline{\varphi }_\mu `$. This allows to hide $`\kappa `$ and $`\epsilon `$ by normalizing the electric potential to a dimensionless quantity
$$\varphi =\sqrt{\frac{\kappa \epsilon }{8\pi }}\overline{\varphi }.$$
(5)
The factor $`8\pi `$ is chosen for future convenience and we use CGS units. This is a much more efficient way to get rid of the constants in the Einstein-Maxwell equations than the choice of relativistic units.
Let us confine ourselves to the axially-symmetric static metric two
$$ds^2=f\left(dx^0\right)^2f^1\left[e^{2k}\left(dr^2+dz^2\right)+r^2d\phi ^2\right],$$
(6)
where $`x^0=ct`$, $`x^1=\phi ,`$ $`x^2=r,`$ $`x^3=z`$ are cylindrical coordinates, $`f=e^{2u}`$ and $`u`$ is the first, while $`k`$ is the second gravitational potential. Both of them depend only on $`r`$ and $`z`$. For the electric field one has
$$E_r=\overline{\varphi }_r,E_z=\overline{\varphi }_z.$$
(7)
The field equations read
$$\mathrm{\Delta }u=e^{2u}\left(\varphi _r^2+\varphi _z^2\right),$$
(8)
$$\mathrm{\Delta }\varphi =2\left(u_r\varphi _r+u_z\varphi _z\right),$$
(9)
$$\frac{k_r}{r}=u_r^2u_z^2e^{2u}\left(\varphi _r^2\varphi _z^2\right),$$
(10)
$$\frac{k_z}{r}=2u_ru_z2e^{2u}\varphi _r\varphi _z,$$
(11)
where $`\mathrm{\Delta }=_{rr}+_{zz}+_r/r`$ is the Laplacian. We have used the definition given in Eq. (5). The first two equations determine $`\varphi `$ and $`f`$. After that $`k`$ is determined by integration.
Weyl electrovacuum solutions three are obtained when the gravitational and the electric potential have the same equipotential surfaces, $`f=f\left(\varphi \right)`$. Eqs. (8-9) yield
$$\left(f_{\varphi \varphi }2\right)\left(\varphi _r^2+\varphi _z^2\right)=0,$$
(12)
which gives
$$f=A+B\varphi +\varphi ^2,$$
(13)
where $`A`$ and $`B`$ are arbitrary constants. Replacing it in Eqs. (8-9) one comes to an equation for $`\varphi `$
$$\mathrm{\Delta }\varphi =\frac{B+2\varphi }{A+B\varphi +\varphi ^2}\left(\varphi _r^2+\varphi _z^2\right).$$
(14)
Let us make one more assumption, that $`\varphi `$ depends on $`r,z`$ through some function $`\psi (r,z)`$ which satisfies the Laplace equation $`\mathrm{\Delta }\psi =0`$. Then $`\varphi \left(\psi \right)`$ is determined implicitly from
$$\psi =\frac{d\varphi }{A+B\varphi +\varphi ^2}.$$
(15)
An important equality follows
$$\varphi _i=f\psi _i,\overline{\varphi }_i=f\left(\varphi \right)\overline{\psi }_i,$$
(16)
where $`i=r,z`$. Eqs. (10-11) become
$$k_r=\frac{D}{4}r\left(\psi _r^2\psi _z^2\right),k_z=\frac{D}{2}r\psi _r\psi _z,$$
(17)
where $`D=B^24A`$. Thus in Weyl electrovac solutions the harmonic master potential $`\psi `$ determines the electric and the gravitational fields.
The theory should be invariant under gauge transformations, which in this case are simply translations: $`\varphi ^{}=\varphi +a`$ with $`a`$ being an arbitrary constant. Eqs.(1-4) and (7-11) are gauge invariant, but Eq.(13) is not because $`A,B`$ change into
$$A^{}=A+Ba+a^2,B^{}=B+2a.$$
(18)
This happens because $`f`$ depends directly on the electric potential and not on its derivatives. In some papers this is used to set $`B^{}`$ to zero and eliminate the linear term.
In this paper we shall show that this is not correct. In fact, the general solution (13) stays gauge invariant because $`A^{},B^{}`$ are also arbitrary constants. In a particular solution $`A^{},B^{}`$ should be fixed and should not change under a gauge transformation. This is possible when after $`a`$ is selected one compensates its effect by choosing $`A,B`$ in such a way that $`A^{},B^{}`$ stay fixed at any particular value. Eq.(18) shows that this always can be done and in this way the gauge invariance of $`f`$ is restored. For example, due to Eq.(5), the electric potential is very small everywhere for realistic fields and it is natural that it should go to zero at infinity or when the field is turned off. Then asymptotic flatness requires to set $`A^{}=1`$ and this condition can be kept in spite of possible gauge transformations. The coefficient $`B^{}`$ is not determined by the system of equations (8-9) and the Weyl conditions. One can not just put it to zero by a gauge transformation. In fact, arguments were given in four ; five that its value is $`2`$. Then $`f`$ becomes a perfect square, while $`k`$ vanishes and the space part of the metric is conformally flat. It should be noticed that $`D^{}=D`$ so that the vanishing of $`k`$ is gauge invariant.
The presence of the linear term in $`f`$ with a coefficient of order unity is not just of academic interest. Because of the gravitational potential a particle at rest feels an acceleration one
$$g_i=\frac{c^2}{2}\left(\mathrm{ln}g_{00}\right)_i=c^2f^1\left(\frac{B^{}}{2}\sqrt{\frac{\kappa \epsilon }{8\pi }}\overline{\varphi }_i+\frac{\kappa \epsilon }{8\pi }\overline{\varphi }\overline{\varphi }_i\right).$$
(19)
Covariant and contravariant components coincide in practice because for realistic electric fields the metric is almost flat. Eq.(7) shows that the first term is proportional to the electric field, which due to Eq.(16) may be derived also from the master potential because $`f`$ is extremely close to one. Let us note that
$$c^2\sqrt{\frac{\kappa }{8\pi }}=\sqrt{G}=2.58\times 10^4,c^2\frac{\kappa }{8\pi }=\frac{G}{c^2}=7.37\times 10^{27},$$
(20)
where $`G=6.674\times 10^8cm^3/g.s^2`$ is the Newton constant and $`c=2.998\times 10^{10}cm/s`$ is the speed of light. Due to the square root, the first coefficient is $`10^{23}`$ times bigger than the second and for realistic fields and media this cannot be compensated by the squares of potentials and the additional $`\sqrt{\epsilon }`$ factor in the second term. The latter is typical for linear perturbation theory. In relativistic units $`G=c=1`$ the difference does not show up. Thus, provided that $`B^{}=2`$, the linear term is essential and the coupling of electromagnetism to gravity appears to be much stronger than it is usually thought. It causes a number of effects, the most prominent being the movement of a usual capacitor towards one of its poles. In this case there is plane symmetry in the bulk, $`f`$ and $`\varphi `$ depend only on $`z`$, which means they are functionally related and the general solution belongs to the Weyl class. However, Eqs.(10,17) show that $`k`$ depends on $`r`$ and breaks the symmetry unless $`D=0`$, which gives $`B^{}=\pm 2`$. Putting the usual formula for the electric field inside a capacitor into Eq.(19) gives for the acceleration which acts on the dielectric inside it
$$g_z=\pm \sqrt{G\epsilon f}\frac{\overline{\psi }_0}{d}\pm 2.58\times 10^4\frac{\sqrt{\epsilon }}{d}\overline{\psi }_0,$$
(21)
where $`\psi _0`$ is the potential difference between the plates and $`d`$ is the distance between them. A more detailed derivation can be found in Refs.four ; five . If the capacitor is hanging freely, this effect may be tested experimentally. To increase the acceleration it is advantageous to make $`d`$ small (typically $`0.1cmd1cm`$), to raise $`\psi _0`$ up to $`2\times 10^4CGS`$ (six million volts, which is possible) and to take a ferroelectric material with $`\epsilon `$ in the range of $`10^4`$, like barium titanate ($`BaTiO_3`$) or many others. Thus $`\sqrt{\epsilon }/d`$ may reach in principle $`10^3`$ and the maximum acceleration $`g_{z,\mathrm{max}}=5.2g_{earth}`$ is more than enough to counter Earth’s gravity.
This effect has been discovered by the prominent electrical engineer Thomas Townsend Brown (1905-1985) already in 1923 together with Prof. P. A. Biefeld and called the Biefeld-Brown effect six . Brown worked on his own on it up to the sixties with high voltage equipment in the range $`70300kV.`$ He didn’t give a formula like Eq.(21) but stressed that the effect is bigger the closer the condenser plates, the higher the voltage and the greater the $`\epsilon ,`$ which is in accord with Eq.(21). He also found that the capacitor moves towards its positive pole, resolving experimentally the sign ambiguity in the above formula. There have been speculations that the effect might follow from some of the Einstein’s unified theories. Today one would mention string theory or some other alternative gravitational theory. However, it appears that the effect is a part of usual General Relativity due to its strong nonlinearity. It is worth to repeat Brown’s experiments in different laboratories and check formula (21). |
warning/0507/quant-ph0507042.html | ar5iv | text | # Casimir Effects: an Optical Approach II. Local Observables and Thermal Corrections
## I Introduction
The Casimir effectCasimir ; Lifshitz ; MT ; MPnew ; expt1 is a manifestation of the quantum fluctuations of a quantum field at a macroscopic level. Experiments on Casimir forces are precise tests of one of the less intuitive predictions of field theory. For a theoretician, predicting the outcome of these experiments is a worthy challenge. Hence it seems somewhat astonishing that an exact solution exists only for infinite, parallel plates caseCasimir . Other formal solutions for geometries not made of distinct rigid bodies free to move (like the wedge, the interior of a sphere or of a rectangular box MT ) are irrelevant for an experimental setup. Moreover in many such solutions divergences have been discarded in a way that leaves the result unrelated to practical materials and configurations Graham:2002xq . Interesting theoretical developments include the method developed in Kardar where the solution for infinite periodic geometries is obtained as a series expansion in the corrugation, and the numerical Montecarlo analysis in Gies03 .
Amongst the various effects that an experimentalist must take in account to interpret the data (e.g. finite conductivity, temperature and roughness corrections) probably the most challenging, interesting and full of connections with other branches of physics and mathematics is the dependence of the force on the geometry of the bodies. Calculating the Casimir force for perfectly reflecting bodies in the end reduces to finding the density of states (DOS) of the Scrhödinger Hamiltonian for the equivalent billiard problem *including the oscillatory ripple on the averaged DOS*. This is an incredibly difficult problem in spectral theory that still challenges mathematicians and physicists today Uribe and in essence is not solved beyond the semiclassical approximation.
In this context we have introduced in Refs. pap1 ; opt1 a method based on classical optics which has several virtues: accuracy, uniform validity when a symmetry is born, straightforward extension to higher spin fields, to non-zero temperatures, to include finite reflectivity and, the main topic of this paper, *it can provide an approximation to local observables.*
This paper is structured as follows: in Section II we show how to cast the energy momentum tensor into a sum over optical paths contributions and how to regulate and analyze the divergences, ubiquitous in Casimir energy calculations. Section III is dedicated to the analysis of the three examples already studied in opt1 with pedagogical intent. We study parallel plates, the Casimir torsion pendulum and a sphere opposite a plate. In Section IV we show how to calculate the same local observables and the free energy for a thermal state and we prove (within the limits of our approximation) the ‘classical limit’ theorem MPnew ; Feinberg , which states that at high $`T`$, Casimir forces become independent of $`\mathrm{}`$ and proportional to $`T`$. As far as we know this is the first time this assertion can be generalized to geometries other than parallel plates. We also study the example of parallel plates (finding the known results) and of a sphere opposite a plate at non-zero temperature. We find evidence, again within the framework of the optical approximation, that the low $`T`$ behavior of the Casimir force is a difficult problem, qualitatively different from the $`T=0`$ and high temperature cases.
## II Local Observables
Local properties of the quantum vacuum induced by the presence of boundaries are of broad interest in quantum field theory local . For example gravity couples locally to the energy-momentum tensor. Vacuum polarization induces local charge densities near boundaries, provided the symmetries of the theory allow it. Also, local densities are free from some of the cutoff dependencies that plague many other Casimir effects. Any local observable that can be expressed in terms of the Greens function can be estimated using the optical approach. In this section we study the energy, momentum and stress densities for a scalar field.
Some local observables are not unambiguously defined Weinberg . For example the charge density (in a theory with a conserved charge) is unambiguously defined while the energy density, in general, is not (while its integral over the volume, the total energy, is). In this paper we use the Noether definition of the energy-momentum tensor, similar results would be obtained with other interesting definitions.
### II.1 Energy-momentum tensor
We study the Noether energy-momentum tensor of a free, real scalar field $`\varphi `$ in a domain $`𝒟`$ with Dirichlet boundary conditions (BC) on $`𝒮=𝒟`$ made of (in general disconnected) surfaces. Other BC (Neumann, Robin) can be discussed but for simplicity we restrict ourselves to Dirichlet BC here.
The lagrangian is (we use $`\mathrm{}=c=1`$)
$$=\frac{1}{2}_\mu \varphi ^\mu \varphi \frac{1}{2}m^2\varphi ^2,$$
(II.1)
where Greek letters are used for 4-dimensional indices while the vector notation will be used for spatial vectors.
The Noether energy-momentum tensor for this real scalar field is
$$T_{\mu \nu }=\frac{}{(^\mu \varphi )}_\nu \varphi g_{\mu \nu }$$
(II.2)
$$T_{\mu \nu }=_\mu \varphi _\nu \varphi g_{\mu \nu }\frac{1}{2}\left(_\alpha \varphi ^\alpha \varphi m^2\varphi ^2\right)$$
(II.3)
from which we identify the energy density $`T_{00}`$, the momentum density $`T_{0i}`$, and the stress tensor $`T_{ij}`$. The definition of these quadratic operators involves divergences that we will regulate by point splitting. We hence replace quadratic operators like $`\varphi (x)^2`$ by $`lim_{x^{}x}\varphi (x^{})\varphi (x)`$. The energy density operator, for example, is
$`T_{00}(x,t)`$ $`=`$ $`\underset{x^{}x}{lim}\left[{\displaystyle \frac{1}{2}}_0\varphi (x^{},t)_0\varphi (x,t)+{\displaystyle \frac{1}{2}}\stackrel{}{}^{}\stackrel{}{}\varphi (x^{},t)\varphi (x,t)+{\displaystyle \frac{1}{2}}m^2\varphi (x^{},t)\varphi (x,t)\right]`$ (II.4)
$`=`$ $`\underset{x^{}x}{lim}[{\displaystyle \frac{1}{2}}_0\varphi (x^{},t)_0\varphi (x,t){\displaystyle \frac{1}{2}}\varphi (x^{},t)\stackrel{}{}^2\varphi (x,t)+`$
$`+{\displaystyle \frac{1}{2}}m^2\varphi (x^{},t)\varphi (x,t)+{\displaystyle \frac{1}{2}}(\stackrel{}{}^{}+\stackrel{}{})\varphi (x^{},t)\stackrel{}{}\varphi (x,t)]`$
The field $`\varphi `$ satisfies the free wave equation in $`𝒟`$
$$^2\varphi +m^2\varphi =0$$
(II.5)
and hence it can be decomposed into normal modes
$$\varphi (x,t)=\underset{j}{}\frac{1}{\sqrt{2E_j}}\left(\psi _j(x)e^{iE_jt}a_j+\psi _j^{}(x)e^{iE_jt}a_j^{}\right),$$
(II.6)
where $`\psi _j`$ and $`E_j`$ are the eigenfunctions and eigenvalues of the problem
$`(\stackrel{}{}^2+m^2)\psi _j`$ $`=`$ $`E_j^2\psi _j\text{for }x𝒟;\psi _j(x)=0\text{for }x𝒮.`$ (II.7)
We also use the definition $`E(k)=\sqrt{k^2+m^2}`$, and $`E_j=\sqrt{k_j^2+m^2}`$ so that the eigenvalue equation reads
$$\stackrel{}{}^2\psi _j=k_j^2\psi _j,$$
(II.8)
and because of the positivity of the operator $`\stackrel{}{}^2`$, the spectrum $`\{E_j\}`$ is contained in the half-line $`\{Em\}`$.
We now introduce the propagator $`G(x^{},x,k)`$, defined as in Ref. opt1 to be the Green’s function of the problem (II.7) or (II.8):
$`(\stackrel{}{}^2k^2)G(x^{},x,k)`$ $`=`$ $`\delta (x^{}x)`$
$`G(x^{},x)`$ $`=`$ $`0\text{for }x^{}\text{ or }x𝒮,`$ (II.9)
which can be written using the spectral decomposition as
$$G(x^{},x,k)=\underset{n}{}\frac{\psi _n(x^{})\psi _n(x)}{k_n^2k^2iϵ}$$
(II.10)
In Ref. pap1 we have developed an approximation for the propagator $`G(x^{},x,k)`$ in terms of optical paths (closed, in the limit $`x^{}x`$). The derivation can be found in Ref. opt1 , the general result valid for $`N`$ spatial dimensions being
$`G_{\mathrm{opt}}(x^{},x,k)`$ $`=`$ $`{\displaystyle \underset{r}{}}{\displaystyle \frac{(1)^{n_r}}{2^{N/2+1}\pi ^{N/21}}}\left(\mathrm{}_r\mathrm{\Delta }_r\right)^{1/2}k^{N/21}H_{\frac{N}{2}1}^{(1)}\left(k\mathrm{}_r\right),`$ (II.11)
$``$ $`{\displaystyle \underset{r}{}}G_r(x^{},x,k),`$
where $`H`$ is a Hankel function, $`r`$ labels the paths from $`x`$ to $`x^{}`$, $`n_r`$ is the number of reflections of the path $`r`$, $`\mathrm{}_r(x^{},x)`$ is its length and $`\mathrm{\Delta }_r(x^{},x)`$ is the *enlargement factor* familiar from classical optics,
$$\mathrm{\Delta }_r(x^{},x)=\frac{d\mathrm{\Omega }_x}{dA_x^{}}.$$
(II.12)
$`\mathrm{\Delta }_r(x^{},x)`$ is the ratio between the angular opening of a pencil of rays at the point $`x`$ and the area spanned at the final point $`x^{}`$ following the path $`r`$. For $`N=3`$ we have
$$G_r(x^{},x,k)=(1)^{n_r}\frac{\mathrm{\Delta }_r^{1/2}(x^{},x)}{4\pi }e^{ik\mathrm{}_r(x^{},x)}.$$
(II.13)
With this explicit form for the propagator $`G`$, we now have to rewrite the elements of the quadratic operator $`T_{\mu \nu }`$ as functions of $`G`$ and its derivatives. It is useful to pass from the point-splitting to a frequency cutoff by inserting the latter in the normal modes decomposition (II.6) as
$$e^{k_j/\mathrm{\Lambda }}=_0^{\mathrm{}}𝑑ke^{k/\mathrm{\Lambda }}2k\delta (k^2k_j^2).$$
(II.14)
The limit $`x^{}x`$ can then be exchanged with the $`dk`$ integral and we get for the energy density,
$$0|T_{00}(x,t)|0=_0^{\mathrm{}}𝑑ke^{k/\mathrm{\Lambda }}\frac{1}{2}E(k)\rho (x,k)+_0^{\mathrm{}}𝑑ke^{k/\mathrm{\Lambda }}\frac{k}{2E(k)}\stackrel{}{}\stackrel{}{j}(x,k).$$
(II.15)
The density $`\rho `$ and the vector $`\stackrel{}{j}`$ are defined as
$`\rho (x,k)`$ $`=`$ $`{\displaystyle \frac{2k}{\pi }}\mathrm{Im}G(x,x,k)`$ (II.16)
$`\stackrel{}{j}(x,k)`$ $`=`$ $`\underset{x^{}x}{lim}{\displaystyle \frac{1}{\pi }}\mathrm{Im}\stackrel{}{}G(x^{},x,k)={\displaystyle \frac{1}{2\pi }}\mathrm{Im}\stackrel{}{}G(x,x,k).`$ (II.17)
$``$ is obtained by integrating $`T_{00}`$ over the whole volume between the bodies:
$$=_𝒟d^3x_0^{\mathrm{}}𝑑ke^{k/\mathrm{\Lambda }}\frac{1}{2}E(k)\rho (x,k)+_0^{\mathrm{}}𝑑ke^{k/\mathrm{\Lambda }}\frac{k}{2E(k)}_𝒮𝑑\stackrel{}{S}\stackrel{}{j}(x,k).$$
(II.18)
We have turned the integral over the divergence of $`\stackrel{}{j}`$ into a surface integral using Gauss’s theorem. In the case of Dirichlet or Neumann boundary conditions, since $`d\stackrel{}{S}\stackrel{}{n}`$ we have (here $`_\stackrel{}{n}\stackrel{}{n}\stackrel{}{}`$ and $`j_\stackrel{}{n}=\stackrel{}{n}\stackrel{}{j}`$)
$$j_\stackrel{}{n}(x,k)=\frac{1}{\pi }\mathrm{Im}_\stackrel{}{n}G(x,x,k)=0,x𝒮$$
(II.19)
and the surface integral term disappears. It should be noted that the vanishing of the $`\stackrel{}{j}`$ contribution to the total energy relies on the continuity of the propagator for $`x^{},x𝒟`$. In some approximations, including the optical one, this continuity is lost. Hence spurious surface terms arise on the boundary of certain domains $`𝒟^{}𝒟`$. This region is what in wave optics is called the ‘penumbra’ region. Diffractive contributions are also not negligible in this region and they cancel the discontinuities in $`G`$, hence eliminating the surface terms.<sup>1</sup><sup>1</sup>1As an example see Kirchoff’s treatment of the diffraction from a hole in Ref. BornWolf . The surface terms in the energy are hence of the same order of the diffractive contributions which define the error in our approximation.
The divergence $`\stackrel{}{}\stackrel{}{j}`$ could also be eliminated from $`T_{00}`$ by changing the energy-momentum tensor according to
$$\stackrel{~}{T}_{\mu \nu }=T_{\mu \nu }+^\alpha \psi _{\alpha \mu \nu },$$
(II.20)
with
$$\psi _{\alpha \mu \nu }=\frac{1}{2}\varphi \left(g_{\mu \nu }_\alpha g_{\alpha \nu }_\mu \right)\varphi .$$
(II.21)
The total energy $``$ and momentum are not affected by this redefinition however the new tensor $`T_{\mu \nu }`$ is not symmetric.
It can be seen that the stress tensor $`T_{ij}`$ is normal on the surface $`𝒮`$ (for both Dirichlet and Neumann BC) so locally the force on the surface is given by the pressure alone
$$\frac{d\stackrel{}{F}}{dS}=\stackrel{}{n}P=0|T_{\stackrel{}{n},\stackrel{}{n}}|0.$$
(II.22)
The operator $`T_{\stackrel{}{n}\stackrel{}{n}}`$ regulated by point splitting is
$`T_{\stackrel{}{n},\stackrel{}{n}}(x,t)`$ $`=`$ $`\underset{x^{}x}{lim}\left[_\stackrel{}{n}^{}\varphi ^{}_\stackrel{}{n}\varphi {\displaystyle \frac{1}{2}}g_{\stackrel{}{n},\stackrel{}{n}}\left(_0^{}\varphi ^{}_0\varphi \stackrel{}{}^{}\varphi ^{}\stackrel{}{}\varphi m^2\varphi ^2\right)\right]`$ (II.23)
$`=`$ $`\underset{x^{}x}{lim}[_\stackrel{}{n}^{}\varphi ^{}_\stackrel{}{n}\varphi +{\displaystyle \frac{1}{2}}(_0^{}\varphi ^{}_0\varphi +\varphi ^{}\stackrel{}{}^2\varphi m^2\varphi ^2)`$
$``$ $`{\displaystyle \frac{1}{2}}(\stackrel{}{}^{}+\stackrel{}{})\varphi ^{}\stackrel{}{}\varphi ]`$
where $`\varphi ^{}`$ is shorthand for $`\varphi (x^{},t)`$. The second term in brackets is zero when averaged over an eigenstate of the number operator $`|\{n_j\}`$, by virtue of the equations of motion. For Dirichlet BC the term $`\varphi \stackrel{}{}^2\varphi =0`$ on the boundaries, so we have ($`\stackrel{}{}=\stackrel{}{n}_\stackrel{}{n}+\stackrel{}{}_t`$)
$$0|T_{\stackrel{}{n},\stackrel{}{n}}|0=\underset{x^{}x}{lim}\underset{j}{}\frac{1}{4E_j}\left(_\stackrel{}{n}^{}_\stackrel{}{n}\stackrel{}{}_t^{}\stackrel{}{}_t+k_j^2\right)\psi _j(x^{})\psi _j(x).$$
(II.24)
Since also $`\stackrel{}{}_t\psi _j(x)=0`$ on the boundaries this expression simplifies to
$$P(x)=\underset{x^{}x}{lim}\underset{j}{}\frac{1}{4E_j}_\stackrel{}{n}^{}_\stackrel{}{n}\psi _j(x^{})\psi _j(x).$$
(II.25)
This expression can be rewritten, in terms of the propagator $`G`$, regulated by a frequency cutoff as we did for $`T_{00}`$,
$$P(x)=\underset{x^{}x}{lim}_\stackrel{}{n}^{}_\stackrel{}{n}_0^{\mathrm{}}𝑑ke^{k/\mathrm{\Lambda }}\frac{k}{2\pi E(k)}\mathrm{Im}G(x^{},x,k).$$
(II.26)
In this regulated expression we can exchange the derivatives, limit and integral safely. Below we discuss what the divergences are when $`\mathrm{\Lambda }\mathrm{}`$ and how to interpret and dispose them.
All the above expressions are exact. Once the propagator $`G`$ is known, we can calculate the energy-momentum tensor components from them. However as discussed above in the interesting cases it is difficult to find an exact expression for $`G`$ and some approximations must be used.
For smooth impenetrable bodies we use the optical approximation to the propagator developed in Ref. pap1 ; opt1 and recalled in eq. (II.11). This gives $`G`$ as a series of optical paths and hence the pressure $`P`$ as a sum of optical paths contributions
$`P`$ $``$ $`{\displaystyle \underset{r}{}}P_r`$ (II.27)
$`P_r`$ $`=`$ $`(1)^{n_r}\underset{x^{}x}{lim}_\stackrel{}{n}^{}_\stackrel{}{n}{\displaystyle _0^{\mathrm{}}}𝑑ke^{k/\mathrm{\Lambda }}{\displaystyle \frac{k}{2\pi E(k)}}{\displaystyle \frac{\mathrm{\Delta }_r^{1/2}(x^{},x)}{4\pi }}\mathrm{sin}\left(k\mathrm{}_r(x^{},x)\right),`$ (II.28)
An important feature of the optical approximation is that all divergences are isolated in the low reflection terms whose classical path length can vanish as $`x^{},x𝒮`$. In practice only the zeroth and first reflection are potentially divergent. Before performing the integral in $`k`$ and taking $`\mathrm{\Lambda }\mathrm{}`$ then we have to put aside the divergent zero and one reflection terms $`P_0`$ and $`P_1`$ for a moment (in the next section we will show how their contributions are to be interpreted).
For the remaining families of paths (that we will denote as $`r`$) the integral over $`k`$ can be done and the limit $`\mathrm{\Lambda }\mathrm{}`$ taken safely. The result is finite and reads
$$P(x)=\underset{r}{}\underset{x^{}x}{lim}_\stackrel{}{n}^{}_\stackrel{}{n}(1)^{n_r}\frac{\mathrm{\Delta }_r^{1/2}(x^{},x)}{8\pi ^2\mathrm{}_r(x^{},x)}.$$
(II.29)
We can further simplify this expression. For simplicity let us call $`z`$ the normal direction. Notice that for any sufficiently smooth function $`f(z^{},z)`$ vanishing for either $`z^{}`$ or $`z`$ on the surface $`z=0`$
$$_z^{}_zf(z^{},z)|_{z^{}=z=0}=\frac{1}{2}_z^2f(z,z)|_{z=0}.$$
(II.30)
The proof is trivial: consider that the lowest order term in the expansion of $`f(z^{},z)`$ near $`z^{},z=0`$ is $`z^{}z`$. The propagator $`G(x^{},x,k)`$ satisfies all these properties and hence we can use this result to get rid of the limit $`x^{}x`$ and assume $`x^{}=x`$ from the beginning. We can therefore rewrite Eq. (II.29) as,
$$P(x)=\underset{r}{}(1)^{n_r}_z^2\frac{\mathrm{\Delta }_r^{1/2}(x,x)}{16\pi ^2\mathrm{}_r(x,x)}.$$
(II.31)
Equation (II.31) is one of the main results of this paper. In Ref. opt1 we reduced the computation of Casimir energy to a volume integral. The force is then found by taking the derivative with respect to the distance between the bodies. Calculating the pressure instead gives the force by means of just a double integral of a local function. The problem is then computationally lighter and sometimes (as we will see in the examples) can even lead to analytic results.
Essentially the problem has been reduced to finding the lengths and enlargement factors associated with the optical paths *for points close to the boundary*. In the case of the pressure (eq. (II.29) or (II.31)) it is necessary to know their derivatives in the direction transverse to the surfaces. We will see that this problem can be easily tackled numerically when it cannot be solved analytically.
### II.2 Regulate and eliminate divergences
As in the energy calculations opt1 , the only divergences occurring in the pressure come from by paths whose lengths $`\mathrm{}1/\mathrm{\Lambda }`$, where $`\mathrm{\Lambda }`$ is the plasma frequency of the material. There are only two such families of paths: the zero and one reflection paths. In this section we show that these divergent contributions are independent of the distances between the bodies. This fact is easily understood: in order for a path to have arbitrarily small length all of its points must be on the same body. So in order to study these terms we need only consider a single, isolated body (and a massless field). We are also careful in maintaining the double derivative $`_{z^{},z}^2`$ since we are calculating the terms $`P_0`$ and $`P_1`$ separately.
For $`r=0`$, the zero reflection term, introducing an exponential cutoff $`\mathrm{\Lambda }`$ on the material reflection coefficient we obtain
$$P_0=_0^{\mathrm{}}e^{k/\mathrm{\Lambda }}𝑑k\frac{k}{2\pi E(k)}\underset{x^{}x𝒮}{lim}_z^{}_z\left(\frac{\mathrm{sin}k|z^{}z|}{4\pi |z^{}z|}\right)=\frac{\mathrm{\Lambda }^4}{4\pi ^2}.$$
(II.32)
The same calculation for the $`r=1`$ or one reflection term gives:
$$P_1=_0^{\mathrm{}}e^{k/\mathrm{\Lambda }}𝑑k\frac{k}{2\pi E(k)}\underset{x^{}x𝒮}{lim}_z^{}_z\left(\frac{\mathrm{sin}k|z^{}+z|}{4\pi |z^{}+z|}\right)=\frac{\mathrm{\Lambda }^4}{4\pi ^2}.$$
(II.33)
Notice that these two terms are equal, so we could have substituted $`_{z^{},z}\frac{1}{2}_z^2`$ for their sum, after having properly regulated the divergence.
This positive, cut-off dependent pressure, $`P_\mathrm{\Lambda }P_0+P_1`$, must be dynamically balanced locally by a pressure generated by the material, lest it collapse. Moreover the total force obtained by integrating this quantity over the (closed) surface $`𝒮`$ of the whole body gives zero. However, if the space around the body were inhomogeneous, as in the presence of a gravitational field, a finite term survives the surface integration, giving rise to a “vacuum Archimedes effect” in which the pressure on one side is, due to gravitational effects, larger than on the other side, so the body feels a net force. We have analyzed this effect in detail in Ref. buoy and called it “Casimir buoyancy”.
Finally note that another important element of this class of quadratic operators is the Feynman propagator. In studying a field theory in a cavity or in between impenetrable bodies (for example hadrons as bags, photons in cavities or Bose-Einstein condensates in traps), we can consider expanding the Feynman propagator in a series of classical optical paths reflecting off the boundaries. The first term, related to the direct path is the familiar free propagator, the others give the finite volume corrections.
## III Examples
In this section we calculate the Casimir force from the pressure, using the formalism developed in the previous section, for three examples that were already addressed in Ref. opt1 using the energy method.
### III.1 Parallel Plates
The parallel plates calculation is a classic example, whose result is well known and constitutes the basis the widely used proximity force approximation (PFA) Derjagin . We use this standard example to establish the rank among contributions to the total pressure and show the similarity and differences with the energy method opt1 .
We calculate the force acting on the lower plate, denoted by $`d`$ or *down*, by calculating the pressure on its surface. We discard the zero and $`1d`$ (one reflection on the lower plate itself) reflection terms. The first term to be considered is the path that bounces once on the upper plate ($`u`$ or *up*) $`1u`$. For parallel plates $`\mathrm{\Delta }=1/\mathrm{}^2`$ and we have
$$P(x)=\underset{r1u}{}(1)^{n_r}_z^2\frac{1}{16\pi ^2\mathrm{}_r^2(x,x)}.$$
(III.1)
The length $`\mathrm{}_r(x,x)`$ for the paths that bounce an even number of times is a constant in $`z`$ and hence the derivatives vanish: they do not contribute to the pressure. This seemingly innocuous observation simplifies the calculations considerably and it is a test for any other geometry which reduces to parallel plates in some limit: in this limit the even reflections contributions must vanish. Generically their contributions are small. This parallels the role of the odd reflection paths in the energy method opt1 .
Figure 1 shows the odd reflection paths labelled with our conventions.
For the path $`1u`$ we have
$$P_{1u}(x)=\underset{z0}{lim}_z^2\frac{1}{16\pi ^2(2a2z)^2}=\frac{3}{32\pi ^2a^4}.$$
(III.2)
The next path to be considered is the path that bounces 3 times, first on $`d`$, then on $`u`$ and again on $`d`$, $`dud=3u`$ (3 stands for 3 reflections and $`u`$ for the plate where the middle reflection occurs) which gives a contribution
$$P_{3u}(x)=\underset{z0}{lim}_z^2\frac{1}{16\pi ^2(2a+2z)^2}=\frac{3}{32\pi ^2a^4}.$$
(III.3)
The two contributions Eq. (III.2) and Eq. (III.3) are equal. The reason is easily uncovered. One can recover Eq. (III.3) from Eq. (III.2) sending $`zz`$ but for the purpose of taking the second derivative at $`z=0`$ this is irrelevant. In the same fashion $`P_{3d}=P_{5d}`$, $`P_{5u}=P_{7u}`$ etc. and hence we find
$$P(x)=2\frac{3}{32\pi ^2a^4}2\frac{3}{32\pi ^2(2a)^4}2\frac{3}{32\pi ^2(3a)^4}+\mathrm{}=\frac{3}{16\pi ^2a^4}\frac{\pi ^4}{90},$$
(III.4)
which is the well-known result. Notice also that the rate of convergence is the same as in the calculation making use of the Casimir energy in Ref. opt1 ($`n`$-th term contributes $`1/n^4`$ of the first term, in this case $`1u+3u`$). These observations that allow us to determine the rank of the contributions are fundamental, and they apply as well to the other examples in this section.
### III.2 The Casimir Torsion Pendulum
In this section we study a geometry already considered in Ref. opt1 : a plate inclined at an angle $`\theta `$ above another infinite plate. We have called this configuration a ‘Casimir torsion pendulum’ because the Casimir force will generate a torque which can be experimentally measured. The configuration is analogous to the parallel plates case but the upper plate must be considered tilted at an angle $`\theta `$ from the horizontal. The length of the upper plate must be taken finite, we denote it by $`w`$, while the length of the lower plate can be infinite which we choose for simplicity. There is only one substantial difference with the parallel plates case: the even reflection paths do contribute in the pendulum, since their length varies as we move the final points $`x^{},x`$.
We calculate the force exerted on the lower, infinite plate for simplicity. We then obtain the energy $``$, by integrating over the distance along the normal to the lower plate and from this we can calculate the torque as
$$𝒯=\frac{}{\theta }.$$
(III.5)
The lower plate is taken infinite, the upper plate width is $`w`$, and the distance between the height at the midpoint of the upper plate is $`a`$. We will choose as the origin of the coordinates one point on the intersection line between the lower plate and the line obtained by prolonging the upper plate. This defines a fictitious wedge of opening angle $`\theta `$. We call $`x`$ the horizontal and $`z`$ the vertical coordinate, the third direction, along which one has translational symmetry, being $`y`$.
Since the surfaces are locally flat we have $`\mathrm{\Delta }=1/\mathrm{}^2`$ as in the case of the parallel plates, and again the odd reflections are exactly as in the case of the parallel plates. However now the even reflections contribute (the notation is the same as in the parallel plates case, in the even reflections $`2u`$ means the first reflection is on the upper plate etc.):
$$P=P_{1u+3u}+P_{2u+2d}+P_{3d+5d}+\mathrm{}$$
(III.6)
where we have grouped the terms with the symbolic notation $`P_{a+b}=P_a+P_b`$ when $`P_a=P_b`$. It is useful to recapitulate what we have learned about the rank of these contributions: $`P_{1u+3u}`$ dominates, $`P_{3d+5d}`$ is smaller by $`1/16`$, $`P_{5u+7u}`$ is smaller by $`1/81`$, etc. The even reflections are generically much smaller than the odd reflections, and vanish as $`\theta 0`$.
The first term in (III.6) is
$$P_{1u+3u}=2\frac{1}{16\pi ^2}_z^2\frac{1}{\mathrm{}_1^2(z,x)},$$
(III.7)
with $`\mathrm{}_1=2(x\mathrm{sin}\theta z/\mathrm{cos}\theta )`$, and an overall factor of $`2`$ takes into account the identity $`P_{1u}=P_{3u}`$. Taking the derivative and then setting $`z=0`$ we find
$$P_{1u+3u}=\frac{3}{16\pi ^2}\frac{1}{x^4\mathrm{sin}^4\theta \mathrm{cos}^2\theta },$$
(III.8)
and integrating from $`x_m=(a/\mathrm{sin}\theta w/2)/\mathrm{cos}\theta `$ to $`x_M=(a/\mathrm{sin}\theta +w/2)/\mathrm{cos}\theta `$ we find the force per unit length in the $`y`$ direction
$$F_{1u+3u}=\frac{\mathrm{cos}\theta }{32\pi ^2\mathrm{sin}^4\theta }\left(\frac{1}{(a/\mathrm{sin}\theta w/2)^2}\frac{1}{(a/\mathrm{sin}\theta +w/2)^2}\right).$$
(III.9)
Since term by term $`F=/a`$ we find the first term in optical expansion of the Casimir energy $``$ (the arbitrary constant is chosen so that $`0`$ when $`a\mathrm{}`$) as
$$_{1u+3u}=\frac{aw\mathrm{cos}^4\theta }{2\pi ^2\left(4a^2w^2\mathrm{sin}^2\theta \right)^2}$$
(III.10)
and from this one obtains the torque
$$𝒯_{1u+3u}=\frac{2aw\left(w^24a^2\right)\mathrm{cos}^3\theta \mathrm{sin}\theta }{\pi ^2\left(4a^2w^2\mathrm{sin}^2\theta \right)^3}$$
(III.11)
Analogously we can calculate the contribution to the pressure $`P`$ of the two reflections paths $`2u`$ and $`2d`$. Again the contributions of the two paths are identical and the result simplifies to
$$P_{2u+2d}=\frac{2}{8\pi ^2}\frac{1}{2}_z^2\frac{1}{\mathrm{}_2^2(z,x)},$$
(III.12)
and using $`\mathrm{}_2=2\sqrt{x^2+z^2}\mathrm{sin}\theta `$ we find
$$P_{2u+2d}=\frac{1}{16\pi ^2\mathrm{sin}^2\theta x^4}$$
(III.13)
which integrated from $`x_m=(a/\mathrm{sin}\theta w/2)/\mathrm{cos}\theta `$ and $`x_M=(a/\mathrm{sin}\theta +w/2)/\mathrm{cos}\theta `$ gives the force along the $`z`$ axis due to these paths:
$$F_{2u+2d}=\frac{\mathrm{cos}^3\theta }{48\pi ^2\mathrm{sin}^2\theta }\left(\frac{1}{(a/\mathrm{sin}\theta w/2)^3}\frac{1}{(a/\mathrm{sin}\theta +w/2)^3}\right).$$
(III.14)
This expression can now be expanded for $`\theta 1`$ (quasi-parallel plates)
$$F_{2u+2d}\frac{1}{16\pi ^2}\left(\frac{w}{a^4}\theta ^2+\frac{5w^311wa^2}{6a^6}\theta ^4+\mathrm{}\right).$$
(III.15)
Notice that this expression vanishes when $`\theta 0`$, as it should since for parallel plates all the contributions of even reflections paths vanish.
The next term in the series is $`F_{3d+5d}`$, whose calculation is performed in the same fashion. The result is:
$`F_{3d+5d}`$ $`=`$ $`{\displaystyle \frac{3}{16\pi ^2}}{\displaystyle \frac{\mathrm{cos}^52\theta }{\mathrm{sin}^42\theta }}\left({\displaystyle \frac{1}{(a/\mathrm{sin}\theta w/2)^3}}{\displaystyle \frac{1}{(a/\mathrm{sin}\theta +w/2)^3}}\right),`$ (III.16)
$``$ $`{\displaystyle \frac{1}{16\pi ^2}}\left({\displaystyle \frac{3w}{16a^4}}+{\displaystyle \frac{5w^348a^2w}{32a^6}}\theta ^2+\mathrm{}\right).`$ (III.17)
We can also present the term given by the 4 reflections paths,
$$F_{4u+4d}=\frac{\mathrm{cos}^32\theta }{48\pi ^2\mathrm{sin}^22\theta }\left(\frac{1}{(a/\mathrm{sin}\theta w/2)^3}\frac{1}{(a/\mathrm{sin}\theta +w/2)^3}\right)\frac{1}{16\pi ^2}\left(\frac{w}{4a^4}\theta ^2+\mathrm{}\right).$$
(III.18)
The terms independent of $`\theta `$ can be seen to reconstruct the parallel limit case $`F=(1+1/16+1/81+\mathrm{})3/16\pi ^2a^4`$.
Term by term, this series for the force reproduces the series in Ref. opt1 . The series for the energy and the torque agree as well. The results of the pressure method then coincide with those of the energy method (as for all the examples analyzed in this paper). In Ref. opt1 we discussed at some length the predictions of the optical method for the Casimir torsion pendulum. We will not repeat them here, referring the reader to that paper for further details.
### III.3 Sphere and Plane
The sphere facing a plane is an important example for several reasons: it has been analyzed theoretically with various exact or approximate numerical techniques SandS ; Gies03 ; it is an experimentally relevant configuration; the exact solution is unknown and probably will escape analytical methods for a long time to come. We have already calculated the optical approximation to the Casimir energy in Ref. opt1 up to $`5`$ reflections. In this paper we study this problem for mainly pedagogical purposes, leaving a more accurate and complete numerical analysis for the future. We believe it is worth studying this example because, contrary to the previous two examples, the enlargement factor plays an important role and moreover we will reanalyze this example with finite temperature in Section IV B 2.
We calculate the pressure (and by integrating, the force) exerted on the plate by the sphere which, of course, equals the force exerted by the plate on the sphere. We start from the qualitative observation that the rank of the contributions is the same as in the parallel plates case in the limit $`a/R0`$. In all the examples we have analyzed this rank is preserved for any value of $`a/R`$. Moreover the ratios of the contributions to the force $`F_{3+5}(a,R)/F_{1+3}(a,R)`$, $`F_4(a,R)/F_2(a,R)`$ etc. decrease quickly as $`a/R`$ increases, we believe due to the growing importance of the enlargement factor.
In this paper we calculate analytically the $`1s`$ term (here $`s`$ stands for ‘sphere’ and $`p`$ for ‘plate’) and by using the relation $`P_{1s+3s}P_{1s}+P_{3s}=2P_{1s}`$ proved in Section II (the notation is the same as in that section) we are able to include the $`3s`$ term as well.
Using the expressions for the length and enlargement factor for the $`1s`$ path obtained in Ref. opt1 we get
$`P_{1s+3s}`$ $`=`$ $`2{\displaystyle \frac{R}{16\pi ^2}}{\displaystyle \frac{^2}{z^2}}{\displaystyle \frac{\mathrm{\Delta }_{1s}^{1/2}}{\mathrm{}_{1s}}}`$ (III.19)
$`=`$ $`{\displaystyle \frac{R}{32\pi ^2}}{\displaystyle \frac{^2}{z^2}}\left(R\sqrt{\left(a+Rz\right)^2+\rho ^2}\right)^2\left(\left(a+Rz\right)^2+\rho ^2\right)^{1/2}|_{z=0}.`$
The final expression for the pressure $`P_{1s+3s}`$ obtained after the derivatives are taken is rather long, however the contribution to the force on the plate, $`F_{1s+3s}`$ (obtained by integration of $`P_{1s+3s}`$ over the infinite plate) is quite simple:
$$F_{1s+3s}=2\pi _0^{\mathrm{}}𝑑\rho \rho P_{1s+3s}=\frac{\mathrm{}cR}{8\pi a^3}.$$
(III.20)
This is the largest of the contributions and increasing $`a/R`$ improves the convergence of the series due to the presence of the enlargement factor, so the asymptotic behavior at large $`a/R`$ predicted by the optical approximation is that given by this formula, *i.e.* $`FR/a^3`$ or $`ER/a^2`$. This asymptotic law is in accordance with the numerics of Ref. pap1 and the predictions of other semiclassical methods SandS . However, eq. (III.20) is in disagreement with the Casimir-Polder law CasimirPolder which predicts $`ER^3/a^4`$ for $`aR`$. This is no great surprise, since our method is not valid for $`a/R1`$, the semiclassical reflections being corrected and eventually overshadowed by diffractive contributions Keller ; SandSdiffr .
We have calculated the contribution of the two reflections paths analytically as well. The calculation is more involved than the one reflection term but a big simplification occurs if one notices that, for the purpose of taking the second derivative with respect to $`z`$ at $`z=0`$, one can leave the reflection point on the sphere fixed. We could not prove a similar result for any other reflection. It is certainly not true for *odd* reflections but one can conjecture it to be true for *even* reflections. In this paper we have not calculated the 4 reflection terms and hence we could not check this conjecture for more than 2 reflections.
And finally, we have calculated the $`3p`$ (or $`sps`$) and hence obtained the $`5p`$, or $`pspsp`$, paths contribution $`P_{3p+5p}`$; $`P_{3p+5p}`$ in the parallel plates limit should account for $`1/16`$ of the total force. This contribution, unlike the previous ones, must be calculated partly numerically, mainly because finding the reflection point on the sphere requires the (unique) solution of a transcendental equation. This task is achieved much more quickly by a numerical algorithm than by patching together the several branches of the analytic solution.
The total pressure is plotted in Fig. in 2 while the various contributions (keeping in mind that $`P_{1s+3s}`$ and $`P_{3p+5p}`$ are negative and $`P_{2+2}`$ is mainly positive) are shown in Fig. 3. Figure 2 reveals some interesting features of the pressure in this geometry: the total pressure decays very quickly with the distance as $`P\rho ^\alpha `$: the exponent $`\alpha `$ seems to depend upon the distance $`a/R`$, but for $`a/R0.1`$ a good fit is obtained with $`\alpha =6`$, in accordance with the asymptotic expansion of the $`1+3`$ reflection term Eq. (III.19); by decreasing the distance between the sphere and the plate, the pressure becomes more and more concentrated near the tip, giving us reasons to trust our approximation and supporting the use of the PFA as a first approximation in the limit $`a/R0`$. Figure 3 shows the relative importance of the contributions due to the different paths. As expected the contribution to the total pressure decreases quite fast by increasing the number of reflections. In Fig. 4 one can also see that the sign of the pressure is not determined simply by the number of reflections of the underlying optical path — as for the contribution to the energy density.
By integrating the pressure over the whole plate we obtain the force $`F`$. It is useful to factor out the most divergent term of the force, as predicted by the PFA, so we define the quantity $`f(a/R)`$ as
$$F(a)=\frac{\pi ^3R}{720a^3}f(a/R).$$
(III.21)
Since we include only a finite number of reflections it is convenient to factor out the constant $`\zeta (4)/(1+1/16)`$ such that $`f`$ is normalized with $`f(0)=1`$. The function $`f(a/R)`$, calculated including paths $`1s,3s,2,3p`$ and $`5p`$, is plotted in Figure 5. When $`a/R0`$ $`f`$ is fitted by
$$f(a/R)=10.10a/R+𝒪\left((a/R)^2\right).$$
(III.22)
By comparing to the results of pap1
$$f_{\mathrm{energy}}(a/R)=1+0.05a/R+𝒪\left((a/R)^2\right)$$
(III.23)
there is the difference in the sub-leading term.
By neglecting the $`5s+7p`$ reflection paths (which in the parallel plates case contribute $`2\%`$ of the total force) we can only assert that the functions $`f`$ in (III.22) and (III.23) represent the optical approximation with an error of $`2\%`$. When plotted on the whole range of $`a/R`$ where the optical approximation is to be trusted the pressure and energy method curves never differ more than $`2\%`$. However there is no such a bound on the sub-leading term which, on the contary, depends on the higher reflections contributions which have not been included in this calculations.<sup>2</sup><sup>2</sup>2For example consider that including only $`1s,3p`$, and 2 and reflections would have given a sub-leading term $`0.16a/R`$ instead of $`0.10a/R`$ in Eq. (III.22). The sub-leading term then changes of $`50\%`$ by adding the $`3s+5p`$ reflection terms which contributes only up to $`8\%`$ of the total. With the terms calculated at this point, we cannot make a precise statement about the sub-leading term. We can however safely say that the subleading term $`a/R`$ coefficient is quite small and our method disagrees with the PFA prediction $`0.5a/R`$. The sphere opposite plate is such an experimentally relevant geometry that further, more accurate studies need to be performed to compare with experimental data.
In conclusion, the lessons to be learned from this example are two: 1) The calculations with the pressure method are even quicker and simpler than the energy method and sometimes can give analytic results for non-trivial geometries and 2) the sub-leading terms must be compared only between calculations performed with the same accuracy.<sup>3</sup><sup>3</sup>3AS would like to thank M. Schaden and S. Fulling for conversations on this point during the workshop ‘Semiclassical Approximations to Vacuum Energy’ held at Texas A & M, College Station, TX, January 2005. The concerns about the errors to be associated with the optical, semiclassical or proximity force approximation is still open to debate and is strictly connected to one of the most challenging open problems in spectral theory *i.e.* how to go beyond the semiclassical approximation to the density of states of a positive Hermitian operator.
## IV Casimir Thermodynamics.
As measurements of Casimir forces increase in accuracy they become sensitive to thermal effects. The natural scale for Casimir thermodynamics is a distance, $`\stackrel{~}{\beta }=\mathrm{}c/\pi T`$, which at room temperature is about 2.5 microns. \[To avoid confusion with the wave number $`k`$, we set Boltzmann’s constant equal to unity and measure temperature in units of energy. We continue to keep $`\mathrm{}`$ and $`c`$ explicit.\] So, assuming the corrections are of $`𝒪\left((a/\stackrel{~}{\beta })^\alpha \right)`$, depending on the value of $`\alpha `$ thermal effects might be expected between the $`10\%`$ (for $`\alpha =1`$) and $`0.3\%`$ (for $`\alpha =4`$, the standard parallel plates result) level for Casimir force measurements on the micron scale. In open geometries, like the sphere and plane, even longer distance scales are probed by Casimir effects, and this gives rise to interesting changes in the temperature dependence of the Casimir free energy in comparison with the case of parallel platesMT . The optical approximation is well suited for discussion of thermodynamics since the thermodynamic observables, like the Casimir energy, can be expressed in terms of the propagator. Here we consider again a non-interacting, scalar field outside rigid bodies on which it obeys Dirichlet boundary conditions.
Before entering into a technical discussion of temperature effects, it is useful to anticipate one of our central results which follows from qualitative observations alone. As $`T0`$ the temperature effects probe ever longer distances. Even at room temperature the natural thermal scale is an order of magnitude larger than the separation between the surfaces in present experiments (see Ref. expt1 ). Since long paths contribute little to the Casimir force, we can be confident that thermal effects vanish quickly at low temperature. However, the leading $`T`$-dependence at small $`T`$ comes from regions beyond the range of validity of the optical (or any other) approximation, so we are unable to say definitively how they vanish for geometries where no exact solution is possible (i.e. other than infinite parallel plates).
This section is organized as follows: First we discuss the free energy and check our methods on the parallel plates geometry; then we discuss the temperature dependence of the pressure, which we apply to the sphere and plate case. Finally we discuss the difficulties associated with the $`T0`$ limit.
### IV.1 Free Energy
The free energy is all one needs to calculate both thermodynamic corrections to the Casimir force and Casimir contributions to thermodynamic properties like the specific heat and pressure. However like the Casimir *energy*, Casimir contributions to the specific heat, pressure, etc., are cutoff dependent and cannot be defined (or measured) independent of the materials which make up the full system. So we confine ourselves here to the thermal corrections to the Casimir force. The problem of parallel plates has been addressed before and our results agree with thoseMT .
#### IV.1.1 Derivation
We start from the expression of the free energy for the scalar field as a sum over modes
$`_{\mathrm{tot}}`$ $`=`$ $`\beta ^1{\displaystyle \underset{n}{}}\mathrm{ln}\left({\displaystyle \frac{e^{\beta \frac{1}{2}\mathrm{}\omega _n}}{1e^{\beta (\omega _n\mu )}}}\right),`$ (IV.1)
$`=`$ $`\beta ^1{\displaystyle \underset{n}{}}\mathrm{ln}\left(1e^{\beta (\mathrm{}\omega _n\mu )}\right)+{\displaystyle \underset{n}{}}{\displaystyle \frac{1}{2}}\mathrm{}\omega _n,`$
$``$ $`+,`$
where $`\mu `$ is the chemical potential, and the last term is the Casimir energy, or the free energy at zero temperature, since $`=0`$ for $`T=0`$. The Casimir energy $``$, being independent of the temperature, does not contribute to the thermodynamic properties of the system. It however does contribute to the pressures and forces between two bodies. The force between two bodies, say $`a`$ and $`b`$, is obtained by taking the gradient of the free energy with respect to the relative distance $`\stackrel{}{r}_{ab}`$
$$\stackrel{}{f}_{ab}=\stackrel{}{}_{ab}.$$
(IV.2)
At $`T=0`$ we recover the familiar result $`\stackrel{}{f}=\stackrel{}{}`$.
Next we turn the sum over modes into a sum over optical paths. Following the same steps that led from Eq. (II.4) to Eq. (II.15) we obtain
$$=\beta ^1d^Nx_0^{\mathrm{}}𝑑k\rho (x,k)\mathrm{ln}\left(1e^{\beta (\mathrm{}\omega (k)\mu )}\right).$$
(IV.3)
where $`\rho (x,k)`$ is given by Eq. (II.16). By specializing to a massless field in $`3`$ dimensions with zero chemical potential (to mimic the photon field), and substituting the optical approximation for the propagator Eq. (II.13), we obtain the sum over paths
$$\underset{r=0}{}_r=\underset{r}{}(1)^r\frac{1}{2\pi ^2\beta }_{𝒟_r}d^3x\mathrm{\Delta }_r^{1/2}_0^{\mathrm{}}𝑑kk\mathrm{sin}(k\mathrm{}_r)\mathrm{ln}\left(1e^{\beta \mathrm{}ck}\right).$$
(IV.4)
Here the term $`_0`$, the direct path, gives the usual free energy for scalar black body radiation. Using the values for the direct path, we have $`\mathrm{\Delta }_0=1/\mathrm{}_0^2`$ and $`\mathrm{}_0=|x^{}x|0`$ when taking $`x^{}x`$. We get the familiar textbook expression
$$_0=V_0^{\mathrm{}}𝑑k\frac{k^2}{2\pi ^2}\beta ^1\mathrm{ln}\left(1e^{\beta \mathrm{}ck}\right)=\frac{\pi ^2}{90}\frac{VT^4}{(\mathrm{}c)^3},$$
(IV.5)
where $`V`$ is the (possibly infinite) volume outside the bodies.
The general term $`_r`$ associated with the path $`r`$ is calculated by performing the $`k`$ integral in Eq. (IV.4):
$$_r=(1)^{r+1}\frac{\mathrm{}c}{2\pi ^2}_{𝒟_r}d^3x\mathrm{\Delta }_r^{1/2}\frac{1}{2\mathrm{}_r^3}\left[2+\stackrel{~}{\mathrm{}}_r\left(\mathrm{coth}\stackrel{~}{\mathrm{}}_r+\stackrel{~}{\mathrm{}}_r\mathrm{csch}^2\stackrel{~}{\mathrm{}}_r\right)\right]$$
(IV.6)
where $`\stackrel{~}{\mathrm{}}_r=\mathrm{}_r\pi T/\mathrm{}c=\mathrm{}_r/\stackrel{~}{\beta }`$ measures the path length relative to the thermal length scale.
Eq. (IV.6) is the fundamental result of this section and gives a simple, approximate description of thermal Casimir effects for geometries where diffraction is not too important. There are no divergences in any of the $`_r`$, ultraviolet or otherwise, even for the direct path (as we saw in eq. (IV.5)) and the first reflection path. All the ultraviolet divergences are contained in the Casimir energy $``$. Indeed, by expanding the integrand of equation (IV.6) at short distances, *i.e.* $`\stackrel{~}{\mathrm{}}_r1`$, we obtain
$$\mathrm{\Delta }_r^{1/2}\frac{1}{2\mathrm{}_r^3}\left[2+\stackrel{~}{\mathrm{}}_r\left(\mathrm{coth}\stackrel{~}{\mathrm{}}_r+\stackrel{~}{\mathrm{}}_r\mathrm{csch}^2\stackrel{~}{\mathrm{}}_r\right)\right]\mathrm{\Delta }_r^{1/2}\frac{1}{\stackrel{~}{\beta }^3}\left[\frac{1}{45\stackrel{~}{\beta }}\mathrm{}_r\frac{4}{945\stackrel{~}{\beta }^3}\mathrm{}_r^3+\mathrm{}\right].$$
(IV.7)
Only the 1-reflection path length can go to zero to generate a divergence. For this contribution $`\mathrm{\Delta }_r`$ diverges like $`1/\mathrm{}_r^2`$ as $`\mathrm{}_r0`$, however this is compensated by the $`\mathrm{}_r`$ term in (IV.7) so the expression is finite and then integrable.
To check for infrared divergences notice that at large distances, $`\stackrel{~}{\mathrm{}}_r1`$, the integrand of (IV.6) goes to $`\mathrm{\Delta }_r^{1/2}/\mathrm{}_r^2`$. For an infinite flat plate the $`\mathrm{\Delta }_r1/z^2`$, where $`z`$ is the normal coordinate to the plate, and the integral is hence $`dz/z^3`$ at large $`z`$. For finite plates the domain of integration is finite and for curved plates the enlargement factor falls even faster than $`1/\mathrm{}^2`$, and the integral remains convergent.
Since the integral converges in both the infrared and ultraviolet, it is safe to estimate the important regions of integration by naive dimensional analysis. This leads to the conclusion that *The paths that dominate the temperature dependence of the Casimir force have lengths of order the thermal length $`\stackrel{~}{\beta }`$*. High temperature implies short paths. Very low temperatures are sensitive to very long paths. Long paths involve both paths experiencing many reflections, which are sensitive to the actual dynamics at and inside the metallic surface, or paths making long excursions in an open geometry, which are sensitive to diffraction. Either way, low temperatures will present a challenge.
#### IV.1.2 Parallel Plates
We know that in the limit of infinite, parallel plates the optical approximation to the propagator becomes exact. Hence our method gives another way to calculate the free energy of this configuration of conductors. It is convenient to study this example to check against known results and to prepare the way for a study of the $`T0`$ limit.
We recall that for this configuration the expression for the enlargement factor is $`\mathrm{\Delta }=1/\mathrm{}^2`$ and the lengths are given by $`\mathrm{}_{2n}=2na`$ (where $`a`$ is the distance between the plates) and $`\mathrm{}_{2n+1,u}=2(az)+2na`$, $`\mathrm{}_{2n+1,d}=2z+2na`$, the notation being the same as in Section III.1, should at this point be familiar to the reader.
As in the zero temperature case it is useful consider even and odd reflection contributions separately and as for the zero temperature case, the sum over odd reflections turns into an integral over $`z`$ from $`0`$ to $`\mathrm{}`$
$$_{\mathrm{odd}}=\underset{n=0}{\overset{\mathrm{}}{}}_{2n+1,d}+_{2n+1,u}=\frac{\mathrm{}c}{2\pi ^2\stackrel{~}{\beta }^3}S_0^{\mathrm{}}𝑑x\frac{1}{2x^4}\left[2+x(\mathrm{coth}x+x\mathrm{csch}^2x)\right],$$
(IV.8)
where $`x=2z/\stackrel{~}{\beta }`$ and $`S`$ is the area of the plate. The definite integral can be easily performed numerically and its value is $`\nu =0.06089\mathrm{}`$,
$$_{\mathrm{odd}}=2\frac{\mathrm{}c}{4\pi ^2\stackrel{~}{\beta }^3}S\nu =\frac{\pi T^3}{2(\mathrm{}c)^2}S\nu $$
(IV.9)
which is independent of the separation, $`a`$, and therefore does not contribute to the force.
Let us turn now to the even reflection paths. They have constant length $`2na`$, so the volume integral simply yields the volume between the surfaces $`v=Sa`$. We already calculated the zero-reflection term $`_0`$ in Eq. (IV.5). The remaining even reflection contributions (2,4,6,… reflections) $`_{\mathrm{even},r2}`$ can be written as an infinite sum
$$_{\mathrm{even},r2}=2\frac{\mathrm{}c}{2\pi ^2}Sa\frac{1}{\stackrel{~}{\beta }^4}\underset{n=1}{\overset{\mathrm{}}{}}\frac{1}{2x_n^4}\left[2+x_n\left(\mathrm{coth}x_n+x_n\mathrm{csch}^2x_n\right)\right]$$
(IV.10)
where $`x_n=2na/\stackrel{~}{\beta }n\tau `$ (this defines the dimensionless temperature $`\tau `$) and we have introduced an overall factor of two to take into account the multiplicity of the paths. Thus the total free energy for parallel plates is the sum of $`_0`$ (eq. (IV.5) and the results of eqs. (IV.9) and (IV.10)),
$$_{}=\frac{\pi ^4}{90}\frac{VT^4}{(\mathrm{}c)^3}+\frac{\pi T^3}{2(\mathrm{}c)^2}S\nu \frac{\mathrm{}c}{\pi ^2\stackrel{~}{\beta }^4}Sa\underset{n=1}{\overset{\mathrm{}}{}}\frac{1}{2x_n^4}\left[2+x_n\left(\mathrm{coth}x_n+x_n\mathrm{csch}^2x_n\right)\right]$$
(IV.11)
It is not possible to rewrite $`_{}`$ in a closed form, but the sum is easy to compute numerically and the high and low temperature expansions are easy to obtain analytically. At high temperatures (and fixed $`a`$) $`\tau \mathrm{}`$, and the summand $`g(n)`$ in eq. (IV.11) falls rapidly enough with $`n`$
$$g(n)=\frac{1}{2(\tau n)^4}\left[2+(\tau n)\left(\mathrm{coth}(\tau n)+(\tau n)\mathrm{csch}^2(\tau n)\right)\right]=\frac{1}{2(\tau n)^4}\left[2+\tau n\right]+𝒪\left(e^{\tau n}\right),$$
(IV.12)
that the limit may be taken under the summation, with the result,
$$_{\mathrm{even},r2}\frac{\mathrm{}c}{\pi ^2\stackrel{~}{\beta }^4}Sa\underset{n=1}{\overset{\mathrm{}}{}}\left[\frac{1}{2n^3\tau ^3}\frac{1}{\tau ^4n^4}\right]=\frac{\zeta (3)}{16\pi a^2}ST+\frac{\pi ^2\mathrm{}c}{1440a^3}S.$$
(IV.13)
Notice that the second term cancels the even paths contribution to the Casimir energy. Hence the final expression for the high $`T`$ expansion of the free energy is particularly simple,
$$_{\mathrm{tot}}=+=\frac{\pi ^2}{90\mathrm{}c}VT^4+\nu \frac{\pi }{2(\mathrm{}c)^2}ST^3\frac{\zeta (3)}{16\pi a^2}ST+𝒪\left(e^{\pi Ta/\mathrm{}c}\right).$$
(IV.14)
The first term is usual black body contribution to the bulk free energy. It does not contribute to the force. The second term is also independent of $`a`$ and does not give rise to any force. The third term instead gives the thermal Casimir force. Notice that $`\mathrm{}c`$ has disappeared from this expression. Called the “classical limit”, this high temperature behavior has been noted before and some early results are even due to Einstein (in Milonni pg. 2; see also Feinberg ). In the next section, after the thermal corrections to the pressure are calculated, we show how to extend this result to other geometries.
Note some interesting features of the $`T\mathrm{}`$ limit: First, the sum over paths converges like the sum of $`(1/n)^3`$ as indicated by the appearance of $`\zeta (3)`$. While slower than the $`T=0`$ convergence, it is still rapid enough to obtain a good approximation from low reflections. Second, note that the $`T\mathrm{}`$ problem in 3-dimensions corresponds exactly to a $`T=0`$ problem in 2-dimensions. This is an example of the familiar dimensional reduction expected as $`T\mathrm{}`$. We can give a short proof of this result. Let us first write:
$$F=\frac{1}{\beta }\mathrm{log}Z$$
(IV.15)
where $`Z`$ is the partition function. We need to evaluate $`Z`$ to the lowest order in $`\beta `$ when $`\beta 0`$. The thermal scalar field theory can be written as a free theory on the cylinder $`^3\times [0,\beta )`$. For $`\beta 0`$ the dynamics along the thermal coordinate is frozen in the ground state, with energy $`E_0=0`$, where $`\varphi `$ does not depend on the thermal coordinate. The partition function $`Z`$ is now $`Z=Z_3+𝒪\left(e^{\beta E_1}\right)`$ where $`E_1`$ is the first excited state $`E_11/\beta ^2`$ and $`Z_3`$ is the partition function of the remaining three-dimensional problem in $`^3`$. If the conductors geometry is symmetric along one spatial coordinate, say $`x`$ (in the parallel plates problem we have two of these directions, $`x`$ and $`y`$) this can now be interpreted as an Euclideanized time variable extending from $`0`$ to $`L_x/c`$. So we will write $`Z_3=Z_{2+1}=e^{\frac{1}{\mathrm{}}_2L_x/c}`$ where $`_2`$ is the Casimir energy of the 2 dimensional problem of two lines of length $`L_y`$, distant $`a`$. The free energy $`F`$ is then:
$$F=\frac{1}{\beta }\mathrm{log}Z\frac{1}{\beta }\mathrm{log}Z_{2+1}=T\frac{1}{\mathrm{}}\frac{L_x}{c}_2=TL_xL_y\frac{\zeta (3)}{16\pi ^2a^2}.$$
(IV.16)
Since $`S=L_xL_y`$ This is exactly the $`a`$-dependent term in eq. (IV.14). If the geometry is not translational invariant then we can only say from eq. (IV.16) that the free energy is linear in $`T`$ (since $`Z_{2+1}`$ is independent of $`\beta `$). Later, by using the optical approximation we will find an explicit analytic expression valid also for non-symmetric, smooth geometries.
For low temperatures, $`\tau 0`$, the terms in the $`n`$-sum in eq. (IV.11) differ very little from each other so we can use the Euler-McLaurin formulaAS ,
$$\underset{n=1}{\overset{\mathrm{}}{}}g(n)=_0^{\mathrm{}}𝑑xg(x)\frac{1}{2}g(0)\frac{1}{12}g^{}(0)+\mathrm{}=\frac{\nu }{\tau }\frac{1}{90}+𝒪\left(\tau \right).$$
(IV.17)
Substituting into eq. (IV.11) we find that the first term in eq. (IV.17) cancels the sum over odd reflections (the second term in eq. (IV.11)) and that the second term in eq. (IV.17) combines with $`_0`$ to give a very simple result,
$$_{\mathrm{tot}}=\frac{(VSa)\pi ^2T^4}{90(\mathrm{}c)^3}.$$
(IV.18)
at low temperatures. This has a simple physical interpretation: the typical thermal excitations of the field at low temperature have very long wavelengths, it is hence energetically inconvenient for them to live between the two plates. As a result the only modification of the $`T=0`$ result is to exclude from the standard black body free energy the contribution from the volume between the plates. One could imagine measuring this effect as a diminished heat capacity for a stack of conducting plates inside a cavity.
The low temperature result, eq. (IV.18), is deceptively simple. Its simplicity obscures an underlying problem with the $`T0`$ limit. We postpone further discussion until we have explored the temperature dependence of the pressure. Suffice it to say for the moment, that eq. (IV.18) probably does not apply to realistic conductor with finite absorption, surface roughness, and other non-ideal characteristics.
### IV.2 Temperature dependence of the pressure
In this section we will obtain the temperature dependence of the pressure within our approximation and apply it to a preliminary study of the sphere and plate case. To begin, we calculate the thermal average of an operator $`𝒪`$ quadratic in the real scalar field $`\varphi `$. The average of a generic operator $`𝒪`$ is given by the trace over a complete set of eigenstates $`|\mathrm{\Psi }_\alpha `$ of the Hamiltonian weighted by a Boltzmann factor:
$$𝒪_T=\underset{\alpha }{}e^{\beta _\alpha }\mathrm{\Psi }_\alpha |𝒪|\mathrm{\Psi }_\alpha .$$
(IV.19)
After some algebra we find
$`𝒪_T`$ $`=`$ $`{\displaystyle \underset{j}{}}𝒪_j2n_j+1_T`$ (IV.20)
$`=`$ $`{\displaystyle \underset{j}{}}𝒪_j{\displaystyle \frac{1+e^{\beta E_j}}{1e^{\beta E_j}}}`$
where $`_T`$ denotes the thermal average, $`j`$ labels the normal modes $`\psi _j`$ (cf. Section II), $`n_j`$ is the occupation number of the mode $`j`$ and $`E_j`$ its energy. The quantities $`𝒪_j`$ are read from the decomposition of the diagonal part of the operator $`𝒪`$ written as $`𝒪_{\mathrm{diag}}=_j𝒪_j(a_j^{}a_j+a_ja_j^{})`$ where $`a_j`$ is the annihilation operator of the mode $`j`$.
The $`𝒪_j`$ for the pressure can be read easily from the analysis in Section II:
$$P_j=\underset{x^{}x𝒮}{lim}\frac{1}{4E_j}_\stackrel{}{n}^{}_\stackrel{}{n}\psi _j(x^{})\psi _j(x)$$
(IV.21)
So we can write the pressure on the plate at non-zero temperature as
$`P(x𝒮)`$ $`=`$ $`\underset{x^{}x}{lim}{\displaystyle \underset{j}{}}{\displaystyle \frac{1}{4E_j}}_n^{}_n\psi _j(x^{})\psi _j(x)\left({\displaystyle \frac{1+e^{\beta E_j}}{1e^{\beta E_j}}}\right)`$ (IV.22)
$`=`$ $`\underset{x^{}x}{lim}_\stackrel{}{n}^{}_\stackrel{}{n}{\displaystyle _0^{\mathrm{}}}𝑑ke^{k/\mathrm{\Lambda }}{\displaystyle \frac{k}{2\pi E(k)}}\mathrm{Im}G(x^{},x,k)\left({\displaystyle \frac{1+e^{\beta E(k)}}{1e^{\beta E(k)}}}\right)`$
$`=`$ $`\mathrm{Im}{\displaystyle _0^{\mathrm{}}}𝑑ke^{k/\mathrm{\Lambda }}{\displaystyle \frac{k}{2\pi E(k)}}{\displaystyle \frac{1}{2}}_\stackrel{}{n}^2G(x,x,k)\left({\displaystyle \frac{1+e^{\beta E(k)}}{1e^{\beta E(k)}}}\right)`$
where we have used Eq. (II.30).
Next we introduce the optical approximation for the propagator and limit ourselves to massless scalars $`E(k)=\mathrm{}ck`$. The discussion of the divergences parallels that of Section II and needs not be repeated here. We remove $`P_0`$ and $`P_1`$ and leave all the finite contributions $`r`$. The optical approximation for the pressure exerted by a massless scalar field reads
$`P(x)`$ $`=`$ $`{\displaystyle \underset{r}{}}(1)^{n_r}_\stackrel{}{n}^2{\displaystyle \frac{\mathrm{\Delta }_r^{1/2}}{16\pi ^2}}\left[{\displaystyle \frac{1}{\stackrel{~}{\beta }}}\mathrm{coth}\left(\mathrm{}_r/\stackrel{~}{\beta }\right)\right].`$ (IV.23)
where it is understood that the zeroth and first reflection terms, which contribute to the pressure on each surface individually, but not to the force between surfaces, have been dropped.
Before applying this to the sphere and plate problem, let us again look at the limiting behavior as $`T\mathrm{}`$ and $`T0`$, and draw some conclusions independent of the detailed geometry. First consider $`T\mathrm{}`$. The shortest paths in the sum in eq. (IV.23) are of order $`a`$, the intersurface separation. \[Remember that the optical approximation is accurate as long as the important paths are short compared to $`R`$, a typical radius of curvature of the surfaces.\] At high $`T`$ we can take the $`\stackrel{~}{\beta }0`$ limit under the sum over reflections since the resulting sum still converges. Therefore low reflections dominate, and we can see, retrospectively, that the high temperature approximation applies when $`\stackrel{~}{\beta }/a0`$. So as $`T\mathrm{}`$,
$$P=\underset{r}{}(1)^{n_r}_\stackrel{}{n}^2\frac{\mathrm{\Delta }_r^{1/2}}{16\pi ^2}\left[\frac{1}{\stackrel{~}{\beta }}+𝒪\left(\frac{1}{\stackrel{~}{\beta }}e^{\mathrm{}_r/\stackrel{~}{\beta }}\right)\right].$$
(IV.24)
This limit has been called (it has been previously found for the parallel plates case) the “classical limit” Milonni ; Feinberg ; MT , since the final expression for high temperatures, reinserting $`\mathrm{}`$ and $`c`$,
$$P\underset{r}{}(1)^{n_r}_\stackrel{}{n}^2\frac{\mathrm{\Delta }_r^{1/2}}{16\pi }T$$
(IV.25)
is independent of $`\mathrm{}`$ and $`c`$ apart from exponentially small terms. This expression amounts in neglecting the 1 in the expression $`2n_j+1_T`$, corresponding to normal ordering or neglecting the contribution of the vacuum state.
At low temperatures, $`\stackrel{~}{\beta }\mathrm{}`$, it is not possible to interchange the limit with the sum. The relevant quantity is $`\frac{1}{\stackrel{~}{\beta }}\mathrm{coth}(\mathrm{}_r/\stackrel{~}{\beta })`$, which goes like
$$\frac{1}{\stackrel{~}{\beta }}\mathrm{coth}\left(\frac{\mathrm{}_r}{\stackrel{~}{\beta }}\right)=\frac{1}{\mathrm{}_r}+\frac{\mathrm{}_r}{3\stackrel{~}{\beta }^2}\frac{\mathrm{}_r^3}{45\stackrel{~}{\beta }^4}+𝒪\left(\frac{\mathrm{}_r^5}{\stackrel{~}{\beta }^4}\right)$$
(IV.26)
as $`\stackrel{~}{\beta }\mathrm{}`$. The first term yields the familiar $`T=0`$ expression. The others would give divergent contributions because of the factors of $`\mathrm{}_r`$ in the numerators (even after the inclusion of the enlargement factor $`\mathrm{\Delta }_r`$). Of course the sum over reflections of the *difference*, $`\frac{1}{\stackrel{~}{\beta }}\mathrm{coth}(\mathrm{}_r/\stackrel{~}{\beta })\frac{1}{\mathrm{}_r}`$, converges to zero as $`\stackrel{~}{\beta }\mathrm{}`$, so thermal corrections definitely vanish for any geometry as $`T0`$ as expected. Once again we relegate more detailed consideration of the $`T0`$ limit to a later subsection.
#### IV.2.1 Sphere and plate
In this section we calculate the pressure and total force for the configuration of a sphere facing a plane at non-zero temperature within $`5p`$ reflections. The optical approximation should be accurate if the important paths are short compared to $`R`$, the radius of the sphere. On the other hand the thermal corrections to the force are sensitive to paths with lengths of order $`\stackrel{~}{\beta }`$. So we must have $`R\stackrel{~}{\beta }`$ and $`Ra`$ in order to obtain reliable results from the optical approximation. Fortunately this is a region of experimental interest: present experiments use, for example, $`a0.5\mu m`$, $`R100\mu m`$, and at room temperature, $`\stackrel{~}{\beta }2.5\mu m`$. In this regime the optical approximation should give a good description of the thermal corrections to the force between perfectly reflective, perfectly smooth conductors.
The expression for the pressure is given by Eq. (IV.23), the enlargement factors and lengths are the same as in the $`T=0`$ case. By applying Eq. (IV.23) to the $`1s+3s`$ paths we find the results in Figure 6. Notice that at high temperatures increasing the temperature essentially scales the whole plot proportionally to $`T`$. The force is then linearly dependent on the temperature (this is the ‘classical limit’ already discussed in Section IV.2). More details are given in the caption of Figure 6.
A dimensionless function $`f(a/R,\stackrel{~}{\beta }/R)`$ can again be defined by rescaling the total force $`F`$ to extract the leading term as $`a0`$. The limiting behavior $`a0`$ is not affected by temperature effects so we stick to the old definition for $`f`$:
$$F(a,\stackrel{~}{\beta },R)=\frac{\mathrm{}c\pi ^3R}{720a^3}f(a/R,\stackrel{~}{\beta }/R).$$
(IV.27)
In Figure 7 we present $`f`$ (up to 5 reflections) for 5 different values of $`\stackrel{~}{\beta }/R`$ (we choose 1, 1/2, 1/4, 1/8 and 1/16 recognizing that $`\stackrel{~}{\beta }1`$ strains the limits of our approximations) and varying $`a`$. Notice that in a neighborhood of $`a/R=0`$, shrinking as $`\stackrel{~}{\beta }/R`$ increases, the function $`f`$ is very well approximated by the $`T=0`$ form, already discussed in Section III.3, $`f(a/R)10.1a/R`$. It is not useful to study the derivative $`A(\stackrel{~}{\beta }/R)=f(x,\stackrel{~}{\beta })/x`$ as $`x=a/R0`$ since this will take the constant value predicted by the zero temperature analysis, or $`0.1`$ in this approximation, for any value of the temperature we choose.
It is also clear from the previous discussions leading to equation (IV.24) that in the opposite regime, for $`a/\stackrel{~}{\beta }1`$, we must have $`FR/a^2\stackrel{~}{\beta }=RT/a^2`$ (the ‘classical limit’). In fact, the first term in the high temperature expansion (IV.24) integrated over $`\rho `$ converges and gives a finite force linear in $`T`$. For this problem, the first term in the reflection expansion for high temperatures can even be calculated analytically:
$$F_{1s+3s}=\mathrm{}c\frac{R}{8\pi a^2\stackrel{~}{\beta }}+𝒪\left(e^{R/\stackrel{~}{\beta }}\right)\mathrm{}c\frac{R}{8a^2}T.$$
(IV.28)
Unfortunately there is no such simple closed expression for higher reflection terms (nor for this first term at arbitrary $`T`$). However, if one believes that the rank of contributions is similar to the parallel plates case one should feel safe to say that this truncation captures the optical approximation within a $`\zeta (3)120\%`$. Hence our statements are at least *qualitatively* correct.
This expression for the force gives a prediction for the function $`f`$, defined in Eq. (IV.27). At this level of accuracy ($`1s+3s`$ reflection) and for $`a/\stackrel{~}{\beta }1`$, apart for exponentially small terms in the temperature expansion we have
$$f_{1s+3s}\frac{90}{\pi ^4}\frac{a}{R}\frac{R}{\stackrel{~}{\beta }}$$
(IV.29)
which grows linearly in $`a/R`$ and is (interestingly enough) independent of $`R`$. This is evident in Fig. 7 for the curves with $`\stackrel{~}{\beta }=1/8,1/16`$. For higher $`\stackrel{~}{\beta }`$ the linear growth starts at higher values of $`a`$ not shown in Fig. 7. Moreover the exponential accuracy manifests itself in the sudden change of behavior from $`f10.1a/R`$ to $`fa/\stackrel{~}{\beta }`$.
It is quite easy to extract a universal prediction from this data, whatever the definitive numbers are, after the sum over optical paths is carried to sufficiently high order: *for any non-zero temperature the function $`f(a/R)`$ will deviate from his zero-temperature behavior at $`a\stackrel{~}{\beta }\mathrm{}c/T`$. The deviation will be in the upward direction, increasing the attractive force between the bodies.* Eventually, for sufficiently large distances, the high temperature behavior given by eq. (IV.25) (or (IV.29) for the sphere-plane problem) will be recovered.
### IV.3 Thermal corrections at low temperatures
The preceding examples have made it clear that in the language of the optical approximation, thermal corrections at low temperature arise from very long paths, $`\mathrm{}_r\stackrel{~}{\beta }`$. This can be seen from the general form of the free energy, eq. (IV.6), or in the attempt to take the $`\stackrel{~}{\beta }\mathrm{}`$ limit under the summation in eq. (IV.23), which fails because of the expansion, eq. (IV.26). Here we examine this non-uniformity more carefully in general and in particular for the parallel plate case, where all the expressions are available. We then attempt to draw some conclusions about the magnitude of corrections at low temperature and the possibility of calculating them reliably in an model that idealizes the behavior of materials.
We return to eq. (IV.22), which gives the exact expression for the pressure, and separate out the thermal contribution,
$$P(T)P(0)\delta P=\mathrm{Im}_0^{\mathrm{}}𝑑k\frac{1}{2\pi }_{n^{}n}^2𝒢(x^{},x,k)2\frac{e^{\beta \mathrm{}ck}}{1e^{\beta \mathrm{}ck}},$$
(IV.30)
still exact. Expanding the denominator in a geometric series, we find
$$\delta P=\frac{1}{\pi }\mathrm{Im}\underset{m=1}{\overset{\mathrm{}}{}}_0^{\mathrm{}}𝑑k_{n^{}n}^2𝒢(x^{},x,k)e^{m\beta \mathrm{}ck}.$$
(IV.31)
Each term in the sum is a Laplace transform of the Greens function. Clearly, as $`\beta \mathrm{}`$ the frequencies that dominate this integral are $`1/\beta T`$.
What are the low frequency contributions to $`𝒢(x^{},x,k)`$? In the ideal case of infinite, perfectly conducting, parallel plates, there is a gap in the spectrum at low $`k`$: $`k\frac{\pi }{a}`$. However *in realistic situations* the plates are finite and/or curved, the geometry is open, and there is no gap in the spectrum. The low-$`k`$ part of the spectrum is sensitive to the global geometry, including edges and curvature, and to the low frequency properties of the material. If the conditions are close to the ideal, the contributions to $`\delta P`$ from small $`k`$ may be small. However as $`T0`$, they dominate. We conclude that the $`T0`$ behavior of $`\delta P`$ cannot be calculated for realistic situations.
The optical approximation does not take account of diffraction, and cannot accurately describe the $`T0`$ limit. Nevertheless it is interesting to see how it fails, since this sheds light on the problem in general. Substituting the optical expansion for the Greens function (replacing $`_{n^{}n}^2\frac{1}{2}_z^2`$ and setting $`\mathrm{}=c=1`$) we find
$`\delta P`$ $`=`$ $`{\displaystyle \underset{m=1}{\overset{\mathrm{}}{}}}{\displaystyle \underset{r1}{}}{\displaystyle \frac{1}{8\pi ^2}}_z^2{\displaystyle _0^{\mathrm{}}}𝑑k\mathrm{\Delta }_r^{1/2}\mathrm{sin}(k\mathrm{}_r)e^{m\beta k}`$ (IV.32)
$`=`$ $`{\displaystyle \underset{m=1}{\overset{\mathrm{}}{}}}{\displaystyle \underset{r1}{}}{\displaystyle \frac{1}{8\pi ^2}}_z^2\mathrm{\Delta }_r^{1/2}{\displaystyle \frac{\mathrm{}_r}{m^2\beta ^2+\mathrm{}_r^2}}.`$
The problems with $`T0`$ are quite apparent: as $`\beta \mathrm{}`$ all paths become important.
Next we specialize to parallel plates where $`\mathrm{}_r=(2ar\pm 2z)`$. The derivative can be carried out explicitly. For simplicity we focus on $`m=1`$ ($`\delta P=_{m=1}^{\mathrm{}}\delta P_m`$),
$$\delta P_1=\frac{2}{\pi ^2}\underset{r=1}{\overset{\mathrm{}}{}}\frac{12(ar)^2\beta ^2}{\left(4(ar)^2+\beta ^2\right)^3},$$
(IV.33)
which can be rewritten using the variable $`\tau =2\pi a/\beta `$ introduced earlier,
$$\delta P_1=\frac{2\pi ^2}{\beta ^4}\underset{r=1}{\overset{\mathrm{}}{}}\frac{3\tau ^2r^2\pi ^2}{(\tau ^2r^2+\pi ^2)^3}.$$
(IV.34)
The sum can be performed, giving
$$\delta P_1=\frac{1}{\pi ^2}\left(\frac{1}{\beta ^4}\frac{\pi ^3}{8a^3\beta }\mathrm{coth}\left(\frac{\pi \beta }{2a}\right)\mathrm{csch}^2\left(\frac{\pi \beta }{2a}\right)\right).$$
(IV.35)
The second term in brackets is exponentially small as $`\beta \mathrm{}`$. If we ignore it, restore the $`m`$-dependence, and sum over $`m`$, we obtain
$$\delta P=\frac{\pi ^2}{90\beta ^4}.$$
(IV.36)
which agrees with our earlier calculation, as it must.
However eq. (IV.34) allows us to study the convergence of the sum over reflections as $`\beta \mathrm{}`$. Instead of performing the sum analytically, we sum up to some $`r_{\mathrm{max}}X`$. Since $`\tau 0`$, we can once again use Euler-Maclaurin, to rewrite the sum over $`r`$ as
$$\delta P_1=\frac{2}{\pi ^2}\frac{1}{\beta ^4}\left[\frac{1}{2}\frac{X}{(1+\tau ^2X^2/\pi ^2)^2}+\frac{1}{2}\frac{3\tau ^2X^2/\pi ^21}{(1+\tau ^2X^2/\pi ^2)^3}+\mathrm{}\right].$$
(IV.37)
where the omitted terms are higher Euler-Maclaurin contributions that are unimportant as $`\beta \mathrm{}`$ (*i.e.* $`\tau 0`$).
If the upper limit on the sum, $`X`$, is taken to $`\mathrm{}`$, only the first term, $`1/2`$, survives and gives the expected result. The question is: How large must $`X`$ be before the limiting behavior set in? Dropping the third term in eq. (IV.37), which is subdominant, we can rewrite $`\delta P_1`$ as
$$\delta P_1=\frac{2}{\pi ^2\beta ^4}\left[\frac{1}{2}\frac{X}{(1+X^2\tau ^2/\pi ^2)^2}\right]=\frac{2}{\pi ^2\beta ^4}\left[\frac{1}{2}+\frac{1}{\tau }f(\tau X)\right].$$
(IV.38)
The function $`f(z)`$ is negative definite and has a minimum at $`z=1/2\sqrt{3}0.29`$ where it takes the value $`3^{3/2}/320.16`$. So in order the result Eq. (IV.36) to be valid we must include $`XX_c=\pi /\sqrt{3}\tau `$ terms in the sum. For example in a typical experimental situation we have $`a=0.5\mu m`$ and $`T=300K`$ so $`\beta =8\mu m`$, $`\tau =8/\pi `$ and $`X_c=8/\sqrt{3}=4.6`$. In this case it is necessary to go to $`X20`$ before the contribution of $`|f(\tau X)/\tau |`$ is smaller than 1/2. This means paths with $`40`$ reflections and path lengths of order $`20\mu m`$. With 40 chances to sample the surface dynamics of the material and paths of $`20\mu m`$ available to wander away from the parallel plate regime, the idealizations behind the standard parallel plates calculation must be called into question.
It must be said however that in the modern experiments the temperature corrections are at most of the order of a few percent at $`a1\mu m`$ and vanish when $`a0`$. Nonetheless we want to point out that there is a conceptual difference between formulations based on the infinite parallel plates approximation, extended to curved geometries by means of the PFA, and a derivation (like ours) in which the curvature is inserted *ab initio*. The thermal and curvature scales interplay in a way that the usual derivations MT ; MPnew could not possibly capture, giving rise to different power law corrections in $`a/\beta `$. It is worth reminding the reader that the usual numerical estimates of thermal corrections are based on the infinite parallel plates power law $`(a/\beta )^4`$. A smaller power like $`(a/\beta )^2`$ would give a much bigger upper bound.
To summarize: temperature corrections are small at small $`T`$, but the existing methods of calculating them, including both our optical approximation and the traditional parallel plates idealization, cannot be trusted to give a reliable estimate of the $`T`$-dependence at small $`T`$.
## V Conclusions
In this paper we have shown how to adapt the optical approximation to the study of local observables. We have illustrated the method by studying the pressure, but the method applies as well to other components of the stress tensor, to charge densities, or any quantity that can be written in terms of the single particle Greens function. The advantage of the optical approximation is to extend the study of these local observables to novel geometries. In particular we developed an expression for the Casimir pressure on the bodies and applied our main result Eq. (II.31) to the study of three important examples: parallel plates, the Casimir pendulum and a sphere opposite a plate.
We have also shown how to calculate, within this approximation scheme, thermodynamic quantities and thermal corrections to the pressure in the general case and applied our results to the example of parallel plates (retrieving the known results) and to the case of a sphere opposite a plate. Along the way we have given a proof of the “classical limit” of Casimir force for any geometry (within our approximation), *i.e.* the fact that Casimir forces at high temperatures are proportional to the temperature and independent of $`\mathrm{}`$, a fact that previously was known only for parallel plates.
Finally, we argued that all known methods of computing the temperature dependence of the Casimir effect are suspect as $`T0`$.
## VI Acknowledgments
We would like to thank S. Fulling and M. Schaden for comments. AS would like to thank M. V. Berry for useful discussions. RLJ would like to thank the Rockefeller Foundation for a residency at the Bellagio Study and Conference Center on Lake Como, Italy, where much of this work was performed. This work is also supported in part by funds provided by the U.S. Department of Energy (D.O.E.) under cooperative research agreement DE-FC02-94ER40818. |
warning/0507/cond-mat0507704.html | ar5iv | text | # Frustrated two-level impurities in two-dimensional antiferromagnets
## I Introduction
Defects (or impurities) with intrinsic degrees of freedom exist in many condensed matter systems. Isolated spin in metal is the most known example (Kondo impurity). There is a number of other types of such impurities including two-level systems in glasses, crystal-field states of the rare-earth ions, degenerate or slightly split Jahn-Teller defects, quantum dots etc. The interaction of these impurities with propagating excitations of the host system (electrons in metals, phonons, spin-waves etc.) governs the impurity dynamics and the low-temperature thermal and transport properties of the host system.
The widely used model for investigation of two-level defect dynamics is the spin-boson model which Hamiltonian has the form: Leggett et al. (1987); Weiss (1999)
$$_{sbm}=_d+_0+_{int},$$
(1)
where the first, the second and the third terms describe, respectively, the isolated defect, the host system and their interaction. Then, $`_d=\frac{1}{2}\mathrm{\Delta }\sigma _x+\frac{1}{2}\epsilon \sigma _z`$, where $`𝝈`$ is the Pauli vector describing the defect, $`_0`$ is modeled by a set of harmonic oscillators: $`_0=_\alpha [\frac{1}{2}m_\alpha \epsilon _\alpha Q_\alpha ^2+P_\alpha ^2/(2m_\alpha )]`$ and the interaction term has the form $`_{int}=\sigma _z_\alpha C_\alpha Q_\alpha `$, where $`C_\alpha `$ are some constants. Essentially, dynamics of the defect is determined by the spectral function characterizing the system. Commonly a power-law dependence $`\omega ^v`$ of this function is discussed, where $`v0`$. In the most investigated Ohmic case $`v=1`$. Despite its simplicity the spin-boson model has found numerous applications ranging from electron transfer to quantum information processing. Leggett et al. (1987); Weiss (1999) Meanwhile its modifications are needed in some cases.
In Ref. Maleyev (1979) one of us (S.V.M.) has extensively studied the problem of interaction of a defect with intrinsic degrees of freedom with 3D acoustic phonons in dielectrics. A specific approach has been proposed in which degeneracy of the impurity is assumed to be arbitrary. This approach is based on Abrikosov’s pseudofermion technique Abrikosov (1965) and diagrammatic expansion. In the case of the two-level defect the Hamiltonian of the model considered in Ref. Maleyev (1979) differs from Eq. (1) by the absence of $`_d`$ (the defect is assumed to be degenerate) and by another type of interaction which has the more general form
$$_{int}=g\underset{\mu }{}S^\mu ϵ^\mu (𝐑_0),$$
(2)
where $`𝐒=\frac{1}{2}𝝈`$, $`𝐑_0`$ determines the position of the impurity in the crystal, $`g`$ is the interaction strength, index $`\mu `$ labels Cartesian components, and $`ϵ^\mu (𝐑_0)`$ are some operators of the host system. It was found in Ref. Maleyev (1979) that similar to the spin-boson model the effect of the host system on the defect is completely described by the spectral function given by the imaginary part of the retarded Green’s function of operators $`ϵ^\mu (𝐑_0)`$:
$$\mathrm{\Delta }_{\mu \nu }(\omega )=i_0^{\mathrm{}}𝑑te^{i\omega t}[ϵ^\mu (𝐑_0,t),ϵ^\nu (𝐑_0,0)],$$
(3)
where $`\mathrm{}`$ denotes the thermal average. In the case of 3D acoustic phonons $`\mathrm{Im}\mathrm{\Delta }_{\mu \nu }(\omega )`$ is proportional to $`\omega ^3`$. The approach proposed in Ref. Maleyev (1979) allows to obtain all the results in a general form independent of the particular view of the operators $`ϵ^\mu (𝐑_0)`$ and the value of $`S`$. The only restriction is that the spectral function is proportional to $`\omega ^3`$. The results for various systems would differ only by some constants. Then, one can discuss the effect of the new terms in the interaction in comparison with the spin-boson model. In the case of interaction (2) all the components of the susceptibility have $`T`$-independent non-resonant term and a Lorenz peak with the width $`\mathrm{\Gamma }f^4(T/\mathrm{\Theta })^5`$, where $`f`$ is the dimensionless coupling constant and $`\mathrm{\Theta }`$ is a characteristic energy. The real part of the non-resonant term is a constant at $`|\omega |\mathrm{\Theta }`$ and the imaginary one is proportional to $`\omega `$. At the same time in the spin-boson model the transverse susceptibility has only the non-resonant term. It was also shown Maleyev (1979); Kokshenev (1980) that the scattering on the impurities leads to the anomalous 3D acoustic phonon damping proportional to $`nf^4\omega ^2`$, where $`n`$ is concentration of the impurities (damping caused by scattering on static defects is proportional to $`\omega ^4`$). Afterward the suggested approach has been successfully applied to investigation of defects in glasses Maleyev (1983) and in cubic metals. Maleyev (1994)
In the present paper we apply the approach discussed in Ref. Maleyev (1979) with the spectral function proportional to $`\omega ^2`$ to the problem of two-level degenerate defect. As mentioned above, the nature of the defect and the host system is not essential. The results will depend on the special form of $`ϵ^\mu (𝐑_0)`$ in Eq. (2) via some constants. Thus, our discussion are applicable to all systems with degenerate defect and the spectral function proportional to $`\omega ^2`$. We demonstrate below that an example of such a system is 2D Heisenberg antiferromagnet (AF) with the impurity spin-$`\frac{1}{2}`$ coupled symmetrically to two neighboring host spins (see Fig. 1). This is the particular subject of the present investigation.
Below we show that the spectral function is proportional to $`\omega ^2`$ if the interaction of the defect with 2D AF is determined by spin waves. It is well known that there is no long range order in Heisenberg 2D AF at $`T>0`$. Mermin and Wagner (1966) Nevertheless, as it has been shown theoretically Kopietz (1990); Chakravarty et al. (1989); Tyč and Halperin (1990) and confirmed experimentally Thurber et al. (1997), the spin waves are well defined in paramagnetic phase of 2D AF if their wavelength is much smaller than the correlation length $`\xi \mathrm{exp}(\mathrm{const}/T)`$. It is found below that the interaction is determined by spin waves and the spectral function is proportional to $`\omega ^2`$ if $`\omega Ja/\xi `$, where $`J`$ is the coupling constant between the host spins and $`a`$ is the lattice constant. Then, a small interaction (for definiteness interplane interaction) of the value of $`\eta J`$ can stabilize the long range order at finite $`T`$. It is obtained below that the spectral function is proportional to $`\omega ^2`$ at $`\omega \eta `$ for the ordered quasi-2D AF. We assume that the interaction of the defect with AF has the form (2) with $`ϵ^\mu (𝐑_0)=s_1^\mu +s_2^\mu `$:
$$_{int}^{AF}=g𝐒(𝐬_1+𝐬_2),$$
(4)
where $`𝐬_{1,2}`$ denote the host spins from different sublattices. For the following consideration the sign of $`g`$ is insignificant.
It should be stressed that one must distinguish symmetrically and asymmetrically coupled impurities (see Fig. 1). Symmetrically coupled impurity is located in the zero molecular field. It remains degenerate and the spectral function is proportional to $`\omega ^2`$. In the case of asymmetrically coupled impurity, where the molecular field is nonzero, there is splitting of the impurity levels and the spectral function has terms with weaker $`\omega `$-dependence. For instance, we demonstrate below that the spectral function for defect coupled to one host spin is proportional to a constant. In this paper we consider only the symmetric case. Our results are also valid with certain additional restrictions for slightly split nearly symmetrically coupled impurities (see below).
Previously, different types of impurities in 2D Heisenberg AF have been extensively studied. It is believed that this problem has a relevance to the physics of some high-$`T_c`$ materials. In such compounds as La<sub>2-x</sub>Sr<sub>x</sub>CuO<sub>4</sub> and YBa<sub>2</sub>Cu<sub>3</sub>O<sub>6+x</sub> the mobility of the holes is very small at low level of doping before the onset of superconductivity. This finding has generated particular interest to problems of an extra spin coupled to one spin of 2D AF, Hoglund and Sandvik (2003, 2004); Sushkov (2000, 2003); Sachdev and Vojta (2003); Vojta et al. (2000); Nagaosa et al. (1989); Igarashi et al. (1995); Murayama and Igarashi (1996); Clarke et al. (1993) and an extra spin coupled to sublattices symmetrically. Oitmaa et al. (1995); Kotov et al. (1998); Clarke et al. (1993) It was proposed in Ref. Aharony et al. (1988) that the appearance of the hole could lead to a ferromagnetic interaction between corresponding two spins of the lattice. This work has stimulated studies of a missing or a ferromagnetic bond between one pair of spins in the lattice. Hoglund and Sandvik (2004); Aristov and Maleyev (1990) As inserting the static non-magnetic defects into the planes by means of replacing of Cu atoms with non-magnetic ones (e.g., Zn) is a common method to investigate the properties of CuO<sub>2</sub> planes of high-$`T_c`$ compounds, a missing spin in the lattice (vacancy) has been also discussed. Hoglund and Sandvik (2004); Chernyshov et al. (2002); Wan et al. (1993); Mucciolo et al. (2004); Sachdev and Vojta (2003); Vojta et al. (2000)
The problem of an added spin in 2D AF at $`T=0`$ has been studied theoretically in Ref. Nagaosa et al. (1989). One of the most remarkable findings of that paper is a singular logarithmic frequency behavior of the defect dynamical susceptibility. Impurity static magnetic susceptibility $`\chi (\omega =0)`$ for 2D AF has been evaluated in Ref. Clarke et al. (1993) in the case of symmetrically coupled impurity. It was demonstrated that $`\chi (0)`$ has a Curie-like term and a singular logarithmic correction proportional to $`g^2\mathrm{ln}(J/T)`$.
The defect static magnetic susceptibility in 2D AF being near quantum critical point (QCP) has been discussed recently in Refs. Sachdev and Vojta (2003); Vojta et al. (2000) using nonlinear sigma model. A classical-like behavior of the form $`\chi (0)=S^2/(3T)`$ with a logarithmic correction proportional to $`\mathrm{ln}(1/T)`$ was obtained. The constants before and under the logarithm were found to be universal near QCP, being the same for all types of defects and independent of the strength of the defect coupling. It is argued in Ref. Sachdev and Vojta (2003) that this behavior of static susceptibility also holds far from QCP for vacancy and for impurity spin coupled to one host spin when $`T|g|`$. Meanwhile the constants are no longer universal far from QCP. This finding is in agreement with results of numerical simulations. Hoglund and Sandvik (2003, 2004) They have been also confirmed by some other theoretical approaches. Sushkov (2003) Both the classical-like form of $`1/T`$-term and the logarithmic correction are related to the nontrivial long-range dynamics in the system. Sushkov (2003); Sachdev and Vojta (2003)
In the present paper we study Heisenberg 2D AF at $`T0`$ far from QCP. Our aim is to find the dynamical susceptibility of the impurity $`\chi (\omega )`$ at $`T0`$ and to discuss the influence of such impurities on the low-$`T`$ properties of 2D AF. These problems have not been addressed yet for symmetrically coupled defects: only the ground state properties at $`T=0`$ Kotov et al. (1998); Oitmaa et al. (1995) and the static susceptibility Clarke et al. (1993) have been discussed. The calculations are performed within the order of $`f^4`$, where $`fg/J`$ is the dimensionless coupling parameter. We show that the transverse impurity susceptibility $`\chi _{}(\omega )`$ has a Lorenz peak with the width $`\mathrm{\Gamma }f^4J(T/J)^3`$ that disappears at $`T=0`$, and a non-resonant term. The imaginary part of the non-resonant term is a constant independent of $`T`$ at $`|\omega |\mathrm{\Gamma }`$ and the real part has a logarithmic divergence as $`\omega ,T0`$. Similar logarithmic singularity was found in Ref. Nagaosa et al. (1989) at $`T=0`$. The longitudinal susceptibility $`\chi _{}(\omega )`$ has the non-resonant term which differs from that of $`\chi _{}(\omega )`$ by a constant and a Lorenz peak. We demonstrate that within the order of $`f^4`$ the width of the peak is zero. Its calculation is out of the scope of this paper.
The static susceptibility has the free-spin-like term $`S(S+1)/(3T)`$ and a correction proportional to $`f^2\mathrm{ln}(J/T)`$. We point out here the sharp difference between symmetrically and asymmetrically coupled impurities that takes place in the regime $`T|g|`$ (by asymmetrically coupled impurities we mean here either the added spin coupled to one host spin or the vacancy which is the particular case of the added spin with $`g\mathrm{}`$). The leading $`1/T`$-term has the free-spin-like form in the symmetric case and the classical-like form in the asymmetric one. Moreover, the logarithmic correction is proportional to $`g^2`$ in the symmetric case and it does not depend on $`g`$ in the asymmetric one. Sushkov (2003); Sachdev and Vojta (2003) The difference is related to the fact that the impurity spin coupled asymmetrically aligns with the local Neel order, Sushkov (2003); Sachdev and Vojta (2003) whereas the symmetrically coupled impurity is located in the zero molecular field.
The fact that the spectral function in 2D AF is proportional to $`\omega ^2`$ only at $`\omega \{\eta \text{ or }Ja/\xi \}`$ leads to the following restriction on the range of validity of the results obtained: $`\mathrm{max}\{\mathrm{\Gamma },|\omega |\}\{\eta \text{ or }Ja/\xi \}`$. If the defect is slightly split (for definiteness by magnetic field $`𝐇`$) this condition turns into $`\mathrm{max}\{\mathrm{\Gamma },|\omega |\}\mathrm{max}\{\{\eta \text{ or }Ja/\xi \},g\mu _BHS\}`$. For nearly symmetrically coupled impurity one has: $`\mathrm{max}\{\mathrm{\Gamma },|\omega |\}\mathrm{max}\{\{\eta \text{ or }Ja/\xi \},|g_1g_2|\}`$, where $`g_{1,2}`$ are values of coupling with the corresponding sublattices (see Fig. 1).
The results described above are valid for isotropic interaction (4). We also consider interaction containing only one term: $`_{int}=gS^x(s_1^x+s_2^x)`$. In this case the $`xx`$-component of the impurity susceptibility is zero whereas $`yy`$\- and $`zz`$\- ones have only the non-resonant term. This model is identical to the spin-boson model (1) without $`_d`$. The Hamiltonian can be diagonalized exactly and an exact expression for $`\chi (\omega )`$ can be obtained. Pirc and Dick (1974) Below we perform the corresponding calculations for the spectral function proportional to $`\omega ^2`$ and confirm the results obtained within our approach. One of the most interesting features of the exact result is that the static susceptibility has the form $`\chi (0)T^{1\zeta }`$, where $`\zeta f^2T/J`$. Within the first order of $`f^2`$ one has $`1/(4T)`$-term and the logarithmic correction. Thus, we see that in the modified spin-boson model taking into account the higher order logarithmic corrections leads to the non-trivial power-law $`T`$-dependence of $`\chi (0)`$.
The influence of the finite concentration $`n`$ of the defects on the low-temperature properties of 2D AF is also considered. For not too small $`\omega `$ we find the logarithmic correction to the spin-wave velocity of the form $`nf^4\mathrm{ln}|J/\omega |`$ and an anomalous damping of the spin-waves proportional to $`nf^4|\omega |`$. Similar logarithmic correction to the velocity and damping were obtained in Ref. Chernyshov et al. (2002), where vacancies in 2D AF were studied. It is demonstrated that interaction of the spin waves with defects modifies the spectral function which acquires new terms proportional to $`n`$ exhibiting weaker $`\omega `$-dependence. These terms should be taken into account at small enough $`\omega `$ and the problem should be solved self-consistently. The corresponding consideration is out of the scope of this paper. Within the range of validity of our study we do not obtain a renormalization of the magnetic specific heat which is proportional to $`T^2`$ in 2D AF without impurities. At the same time it was obtained Chernyshov et al. (2002); Mucciolo et al. (2004) that vacancies give rise to a constant contribution to the density of states that in turn leads to a large correction to the specific heat proportional to $`nT`$.
The rest of the paper is organized as follows. The model, Abrikosov’s pseudofermion and diagrammatic techniques employed for the calculations are discussed in Sec. II. The pseudofermion Green’s function, the pseudofermion vertex and the impurity dynamical susceptibility are derived in Secs. III.1III.3. Another type of interaction of the defect with the host system are discussed in Sec. III.4. The exactly solvable spin-boson model (without $`_d`$) which is a special case of our model is also studied in Sec. III.4 and a comparison with our results is made. Influence of the defects on low-temperature properties of 2D AF is considered in Sec. IV. The spin-wave spectrum and the specific heat are studied in detail in Sec. IV. Section V contains our conclusions. A few appendices are included with details of the calculations.
## II Model and technique
### II.1 Model
Let us formulate in somewhat detail the model discussed in this paper. We consider systems which Hamiltonian can be represented in the following form:
$$=\underset{𝐤}{}ϵ_𝐤\left(\alpha _𝐤^{}\alpha _𝐤+\frac{1}{2}\right)+_{int},$$
(5)
where the first term describes non-interacting low-energy propagating modes of the host system (e.g., phonons or magnons) and the second one is the coupling of the degenerate impurity with the system for which we have the general expression (2). It is also supposed in this paper that the imaginary part of the function $`\mathrm{\Delta }_{\mu \mu ^{}}(\omega )`$ given by Eq. (3) has the form
$$\mathrm{Im}\mathrm{\Delta }_{\mu \nu }(\omega )=A\left(\frac{\omega }{\mathrm{\Theta }}\right)^2\mathrm{sgn}(\omega )\mathrm{\Lambda }(\omega )d_{\mu \nu },$$
(6)
where $`A`$ is a positive constant which dimensionality is inverse energy, $`\mathrm{\Theta }`$ is the characteristic energy, $`\mathrm{\Lambda }(\omega )`$ is a cut-off function which is equal to unity at $`|\omega |<\mathrm{\Theta }`$ and decreases rapidly to zero outside this interval and $`d_{\mu \nu }`$ is a tensor. As was also mentioned above, the particular nature of the defect and the host system are not essential in our study. We will use the general expressions (2) and (6) in all calculations. Therefore results for different systems would differ only by some constants. Nevertheless we discuss now the specific system, Heisenberg 2D AF, which can be described by this model.
It is well known that 2D AF at $`T0`$ has no long range order. Mermin and Wagner (1966) The average $`z`$-component of the spin in AF is given by
$$s^z=s\frac{1}{N}\underset{𝐤}{}\frac{4sJϵ_𝐤}{2ϵ_𝐤}\frac{4sJ}{N}\underset{𝐤}{}\frac{N(ϵ_𝐤)}{ϵ_𝐤},$$
(7)
where $`N`$ is the number of spins in the lattice, $`s`$ and $`J`$ are values of the spin and the exchange, respectively, $`ϵ_𝐤`$ is the spin-wave energy which is equal to $`\sqrt{8}sJk`$ at small $`k`$ and $`N(ϵ_𝐤)=(e^{ϵ_𝐤/T}1)^1`$. The first term in Eq. (7) gives the well known correction to the average spin at $`T=0`$ which is approximately equal to 0.2. The last term describes the spin reduction due to thermal fluctuations. At $`ϵ_𝐤T`$ we have $`N(ϵ_𝐤)T/ϵ_𝐤`$. Thus the last term diverges logarithmically at small $`𝐤`$ in 2D AF. A weak interaction of the value of $`\eta J`$ such as anisotropy or an interplane interaction can screen this divergence and stabilize the long range order. For definiteness we consider interplane interaction. It leads the momentum to become a 3D vector. As a result an additional term appears in the spin-wave energy proportional to $`\eta k_{}`$ at small $`k_{}`$, where $`𝐤_{}`$ is the component of the momentum perpendicular to the plane of the lattice. As a result the last term in Eq. (7) is small and the spin waves are well defined if
$$\frac{T}{sJ}\mathrm{ln}\left(\frac{T}{s\eta }\right)1.$$
(8)
We will assume below that this condition holds.
It is shown in Appendix A that within the spin-wave approximation the function $`\mathrm{Im}\mathrm{\Delta }_{\mu \nu }(\omega )`$ has the form (6) for 2D Heisenberg AF when $`|\omega |\{\eta \text{ or }Ja/\xi \}`$. The particular expressions for $`\mathrm{\Theta }`$, $`A`$, $`d_{\mu \nu }`$ are also established in Appendix A. Within the spin-wave approximation the only nonzero components of $`d_{\mu \nu }`$ are $`xx`$\- and $`yy`$\- ones provided that $`z`$-axis is directed along magnetization of the sublattices.
Abrikosov’s pseudofermion representation of the impurity spin $`𝐒`$ is used below. The value of $`S`$ is assumed to be arbitrary in this approach. Nevertheless we restrict ourself in this paper by $`S=1/2`$. It is demonstrated below that the matrix structure of the pseudofermion Green’s function and the vertex is much simpler in this case. Consideration of larger impurity spins is out of the scope of the present paper.
We point out that a new approach has been suggested recently in Refs. Mao et al. (2003); Shnirman and Makhlin (2003) for spin-$`\frac{1}{2}`$ impurity problem. This approach bases on Majorana-fermion representation of the impurity spin. It was demonstrated that this representation simplifies significantly analysis of the model if one can restrict ourselves by first terms in the expansion by coupling parameter. It will be clear soon that in our case the question of possibility of such restriction requires analysis of the diagrams of the third order within this approach. Carrying out of such analysis is out of the scope of the present paper.
### II.2 Abrikosov’s pseudofermion technique
We use below Abrikosov’s pseudofermion technique for the calculation of the impurity dynamical susceptibility. It was suggested in Ref. Abrikosov (1965) for Kondo effect investigation (see also Refs. Zawadovski and Fazekas (1969); Larsen (1972) for discussions). The same approach has been applied for the problem of impurity in other systems by one of us (S.V.M.) in Refs. Maleyev (1979, 1983, 1994). Let us formulate this technique briefly in the convenient for our purpose form.
The impurity spin $`𝐒`$ is represented as
$$𝐒=\underset{mm^{}}{}a_m^{}𝐒_{mm^{}}a_m^{},$$
(9)
where $`m`$ is the spin projections, $`a_m^{}`$ and $`a_m`$ are operator of creation and annihilation of some particles (fermions for definiteness). It is easy to verify that the spin commutation rules are satisfied in this representation. A wave function of the impurity is characterized now by the occupation numbers of $`2S+1`$ states: $`|n_S,n_{S1},\mathrm{}n_S`$. Obviously, the states with zero or more than one particles are not physical ones and we have to eliminate them carrying out the thermodynamic average. As a result the thermodynamic average of some operator $`Y`$ has the form:
$$\overline{Y}=\frac{\mathrm{Tr}^{phys}(\rho Y)}{\mathrm{Tr}^{phys}(\rho )},$$
(10)
where the traces are limited to physical states and $`\rho =\mathrm{exp}(/T)`$ is the statistical operator with the Hamiltonian $``$ given by Eq. (5). This representation is not convenient not allowing to use the standard diagrammatic technique. To overcome this obstacle an additional term in the Hamiltonian is added: Abrikosov (1965)
$$_\lambda =\lambda N_{pf}=\lambda \underset{m}{}a_m^{}a_m,$$
(11)
where $`N_{pf}`$ is the number of pseudofermions. We show now that the average of $`Y`$ has the following form equivalent to (10):
$$\overline{Y}=\underset{\lambda \mathrm{}}{lim}\frac{\mathrm{Tr}(\stackrel{~}{\rho }Y)}{\mathrm{Tr}(\stackrel{~}{\rho }N_{pf})},$$
(12)
where $`\stackrel{~}{\rho }=\mathrm{exp}\{\stackrel{~}{}/T\}`$ and $`\stackrel{~}{}=+_\lambda `$. We consider in this paper such $`Y`$ that do not contain terms without pseudofermion operators. As a result there are no contributions both to numerator and denominator in Eq. (12) from states with no particles. As $``$ does not change the number of particles in state, we have for the matrix elements: $`\stackrel{~}{\rho }_{kl}=\rho _{kl}\mathrm{exp}(N_l\lambda /T)`$, where $`N_l`$ is the number of pseudofermions in states $`|k`$ and $`|l`$. Therefore, contributions from the states with more than one particles are exponentially small compared to those from the physical states which are proportional to $`e^{\lambda /T}`$. The common factors $`e^{\lambda /T}`$ in the numerator and the denominator of Eq. (12) cancel each other. Then contributions from states with one pseudofermion survive only in the limit of $`\lambda \mathrm{}`$ and Eqs. (10) and (12) appear to be equivalent.
Quantities in the right part of Eq. (12) can be calculated using the conventional diagrammatic technique. The Hamiltonian $`\stackrel{~}{}`$ in the pseudofermion representation has the form: Maleyev (1979)
$$\stackrel{~}{}=\left(\underset{𝐤}{}ϵ_𝐤\left(\alpha _𝐤^{}\alpha _𝐤+\frac{1}{2}\right)+\lambda \underset{m}{}a_m^{}a_m\right)+g\underset{m,m^{},\mu }{}a_m^{}^{}S_{m^{}m}^\mu a_mϵ^\mu (𝐑_0)=_0+_{int}.$$
(13)
The dynamical susceptibility of the impurity in the representation of interaction can be written using Eqs. (12) and (13) in the following form: Maleyev (1979)
$`\chi _P(i\omega _n)`$ $`=`$ $`\underset{\lambda \mathrm{}}{lim}𝒩^1{\displaystyle _0^{1/T}}𝑑\tau e^{i\omega _n\tau }\mathrm{Tr}\left[e^{_0/T}T_\tau \left\{P(\tau )P(0)𝔖\left({\displaystyle \frac{1}{T}}\right)\right\}\right],`$ (14)
$`𝒩`$ $`=`$ $`\mathrm{Tr}\left[e^{\stackrel{~}{}/T}{\displaystyle \underset{m}{}}a_m^{}a_m\right],`$ (15)
$`P`$ $`=`$ $`{\displaystyle \underset{mm^{}}{}}a_m^{}P_{mm^{}}a_m^{},`$ (16)
where $`P`$ is a spin projection. In the zeroth order of the interaction $`_{int}`$ we have $`𝒩=(2S+1)e^{\lambda /T}`$.
### II.3 Diagrammatic technique
First diagrams for $`\chi _P(\omega )`$ and a graphical representation of the result of all diagrams summation are shown in Fig. 2, where thin lines with arrows represent the bare particle Green’s functions:
$$G_{mm^{}}^{(0)}(i\omega _n)=\frac{\delta _{mm^{}}}{i\omega _n\lambda }$$
(17)
and wavy lines denote bosons Green’s functions $`\mathrm{\Delta }_{\mu \mu ^{}}(i\omega _n)`$. Notice that the diagrams with only one pseudofermion loop should be taken into account because, as is seen from Eq. (14), each loop is proportional to the small factor of $`e^{\lambda /T}`$. The contribution from diagrams with one loop is finite because their factor of $`e^{\lambda /T}`$ is canceled by that from $`𝒩`$.
For the calculation of $`\chi _P(\omega )`$ we use below the diagrammatic technique employed in Refs. Maleyev (1979, 1983, 1994). Let us discuss it briefly. Firstly, we have to make an analytical continuation of diagrams for the dressed pseudofermion Green’s function and for the vertex $`\mathrm{\Gamma }_P(\omega _1+\omega ,\omega _1)`$ from imaginary frequencies to real ones. Then, we have to express $`\chi _P(\omega )`$ via these quantities.
To make the first step of this program let us choose frequencies of wavy lines to be independent variables over which the summations are taken. It can be done in such a way that these frequencies are contained in arguments of $`G^{(0)}`$-functions with positive sign (see Fig. 2). Then, each sum over a discrete frequency can be replaced by an integral over a contour enveloped the imaginary axis with an additional factor of $`(2\pi i)^1N(\omega )`$:
$$T\underset{\omega _n}{}\frac{1}{2\pi i}𝑑\omega N(\omega ),$$
(18)
where $`N(\omega )=(e^{\omega /T}1)^1`$ is the Plank function. Maleyev (1970, 1979, 1983, 1994) The contour can be deformed so as to embrace the real axis. In evaluation of the resultant integral one should not take into account poles of $`G^{(0)}`$-functions because residues in these poles are proportional to $`N(\lambda )\mathrm{exp}(\lambda /T)`$. At the same time functions $`\mathrm{\Delta }_{\mu \mu ^{}}(\omega )`$ has a discontinuity on the real axis equal to $`2i\mathrm{Im}\mathrm{\Delta }_{\mu \mu ^{}}(\omega )`$. As a result all contour integrals can be easily transformed to those over the real axis and we lead to the following diagrammatic technique: each wavy line corresponds to $`\pi ^1N(\omega )\mathrm{Im}\mathrm{\Delta }_{\mu \mu ^{}}(\omega )`$; frequencies of wavy lines should be taken so as they are contained in arguments of $`G^{(0)}`$-functions with positive sign; integration over all frequencies of wavy lines is taken in the interval $`(\mathrm{},\mathrm{})`$.
One can conclude from analysis of all concrete diagrams for the vertex $`\mathrm{\Gamma }_P(i\omega _1+i\omega ,i\omega _1)`$ that it is an analytical function of two independent variables $`i\omega _1+i\omega `$ and $`i\omega _1`$ with cuts along the real axis. A general proof of this statement has been also given. Maleyev (1970)
As a result we have for the dynamical susceptibility after the analytical continuation from the discrete frequencies to the real axis: Maleyev (1979, 1983, 1994); Ginzburg (1974)
$`\chi _P(\omega )`$ $`=`$ $`(2\pi i𝒩)^1e^{\lambda /T}{\displaystyle _{\mathrm{}}^{\mathrm{}}}dxe^{x/T}\mathrm{Tr}\{P[G(x+\omega )\mathrm{\Gamma }_P^{++}(x+\omega ,x)G(x)G^{}(x)\mathrm{\Gamma }_P^{}(x,x\omega )G^{}(x\omega )`$ (19)
$`G(x+\omega )\mathrm{\Gamma }_P^+(x+\omega ,x)G^{}(x)+G(x)\mathrm{\Gamma }_P^+(x,x\omega )G^{}(x\omega )]\},`$
where $`G(\omega )`$ is the retarded Green’s function, the trace is taken over projections of the impurity spin and signs at superscript of $`\mathrm{\Gamma }_P`$ denote those of imaginary parts of the corresponding arguments (e.g., $`\mathrm{\Gamma }_P^+(x,y)=\mathrm{\Gamma }_P(x+i\delta ,yi\delta )`$). An energy shift by $`\lambda `$ has been performed during the derivation of Eq. (19). As a result the Fermi function $`(e^{(x+\lambda )/T}+1)^1`$ has been replaced by $`\mathrm{exp}((x+\lambda )/T)`$ and the functions $`G`$ and $`\mathrm{\Gamma }_P`$ no longer depend on $`\lambda `$. These are those functions we calculate in the next section by the diagrammatic technique. It is clear that the bare pseudofermion Green’s functions in this case are $`G_{mm^{}}^{(0)}(\omega )=\delta _{mm^{}}/\omega `$.
## III Dynamical susceptibility of the impurity
We derive analytical expressions for the dynamical susceptibility of the impurity in this section. Perturbation theory is used for this purpose. It can be done if the dimensionless constant
$$f^2=\frac{g^2A}{\mathrm{\Theta }}$$
(20)
is small. Meanwhile we have to take into account also terms of the order of $`f^4`$ because the finite width of the Lorenz peak in the dynamical susceptibility arises in this order. In 2D AF we have from Eqs. (91): $`f^2=(\sqrt{\pi }/4s)(g/J)^2`$.
### III.1 Pseudofermion Green’s function
We turn to the calculation of the pseudofermion Green’s function $`G_{mm^{}}(\omega )`$. The Dyson equation for it has the following form:
$$\omega G_{mm^{}}(\omega )=\delta _{mm^{}}+\underset{n}{}\mathrm{\Sigma }_{mn}(\omega )G_{nm^{}}(\omega ).$$
(21)
The first diagrams for $`\mathrm{\Sigma }_{mn}(\omega )`$ are presented in Fig. 3. Let us discuss its matrix structure first. It is determined by corresponding products of operators $`S^\mu `$ and tensors $`d_{\mu \nu }`$. For example, this combination for the second digram in Fig. 3 has the form $`S^\mu S^\nu S^\mu ^{}S^\nu ^{}d_{\mu \mu ^{}}d_{\nu \nu ^{}}`$. As is shown in Appendix B, all such combinations are proportional to the unit matrix for the two-level impurity. It should be pointed out that there is no such simplification in the case of the impurity with the value of spin greater than 1/2. As a result the equations for the Green’s function and the vertex become more complicated. The corresponding consideration of the large-spin impurities is out of the scope of the present paper.
Taking into account its matrix structure we have for the Green’s function:
$$G_{mm^{}}(\omega )=\delta _{mm^{}}G(\omega )=\frac{\delta _{mm^{}}}{\omega \mathrm{\Sigma }(\omega )}.$$
(22)
The diagram of the first order shown in Fig. 3 gives the following contribution to $`\mathrm{\Sigma }(\omega )`$:
$`\mathrm{\Sigma }^{(1)}(\omega )`$ $`=`$ $`R_\mathrm{\Sigma }{\displaystyle \frac{f^2}{\pi \mathrm{\Theta }}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑x{\displaystyle \frac{x|x|}{x+\omega +i\delta }}N(x)\mathrm{\Lambda }(x),`$ (23)
$`R_\mathrm{\Sigma }`$ $`=`$ $`S^\mu S^\nu d_{\mu \nu }`$ (24)
where the constant $`f^2`$ is given by Eq. (20) and $`R_\mathrm{\Sigma }=1/2`$ for 2D AF. It is convenient to extract from this expression a term proportional to $`\omega `$ as follows:
$$\mathrm{\Sigma }^{(1)}(\omega )=\omega R_\mathrm{\Sigma }\frac{f^2}{\pi \mathrm{\Theta }}_{\mathrm{}}^{\mathrm{}}𝑑x\frac{|x|}{x+\omega +i\delta }N(x)\mathrm{\Lambda }(x)R_\mathrm{\Sigma }\frac{f^2}{\pi \mathrm{\Theta }}_0^{\mathrm{}}𝑑xx\mathrm{\Lambda }(x),$$
(25)
where we have used that $`\mathrm{\Lambda }(\omega )`$ is an even function and $`N(x)=1N(x)`$. Notice that the second term in Eq. (25) is the $`T`$-independent constant. Then it can be included in the renormalization of $`\lambda `$ and omitted. The first term in Eq. (25) is proportional to $`f^2\omega T\mathrm{ln}(T/\omega )`$ at small $`\omega `$. It would seem that a great renormalization of the Green’s function takes place if $`f^2T\mathrm{ln}|T/\omega |\mathrm{\Theta }`$. Meanwhile we show now that the logarithmic singularities at real $`\omega `$ are screened by a finite damping which is of the order of $`f^4`$. Let us represent the Green’s function in the form
$$G(\omega )=\frac{1Z(\omega )}{\omega +i\gamma (\omega )},$$
(26)
where $`Z(\omega )`$ and $`\gamma (\omega )`$ are some functions, $`\gamma (\omega )`$ is a real one, and a constant term in the denominator has been attributed to the renormalization of $`\lambda `$ and discarded. Evaluating contribution from the first diagram in Fig. 3 using Eq. (26) we have in the first order a correction to the constant,
$$Z^{(1)}(\omega )=R_\mathrm{\Sigma }\frac{f^2}{\pi \mathrm{\Theta }}_{\mathrm{}}^{\mathrm{}}𝑑x\frac{|x|}{x+\omega +i\gamma (x+\omega )}N(x)\mathrm{\Lambda }(x)$$
(27)
and $`\gamma ^{(1)}(\omega )=0`$. The logarithmic divergence in expression (27) at real $`\omega `$ is screened by the term $`i\gamma (\omega +x)`$ in the denominator. It is shown below that $`\gamma (\omega )`$ is proportional to $`f^4T^3`$ at $`|\omega |T`$. Contributions to $`Z(\omega )`$ from the higher order diagrams are also small by the same reason and we can restrict ourself by the first correction to it.
To obtain $`\gamma (\omega )`$ one has to take into account $`f^4`$-terms from the first and the second diagrams shown in Fig. 3. Along with the small corrections to the constant and to $`Z(\omega )`$ we have:
$`\gamma (\omega )`$ $`=`$ $`R_\gamma {\displaystyle \frac{f^4}{\pi \mathrm{\Theta }^2}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑x|x(x+\omega )|N(x)N(x\omega )\mathrm{\Lambda }(x)\mathrm{\Lambda }(x+\omega ),`$ (28)
$`R_\gamma `$ $`=`$ $`S^\mu S^\nu [S^\nu ^{},S^\mu ^{}]d_{\mu \mu ^{}}d_{\nu \nu ^{}}.`$ (29)
It is important to note that $`\gamma (\omega )`$ is the constant at $`|\omega |T`$:
$$\gamma (\omega )\mathrm{\Gamma }_0=2\pi R_\gamma \left(\frac{f^2}{\pi \mathrm{\Theta }}\right)^2_0^{\mathrm{}}𝑑xx^2N(x)[1+N(x)]\mathrm{\Lambda }^2(x)=R_\gamma \frac{2\pi f^4}{3}\mathrm{\Theta }\left(\frac{T}{\mathrm{\Theta }}\right)^3.$$
(30)
Notice that from the physical reason $`R_\gamma `$ and $`\mathrm{\Gamma }_0`$ should be positive. For instance, $`R_\gamma =1/4`$ for 2D AF.
It is significant to note that $`\mathrm{Im}Z(\omega )`$ and $`\gamma (\omega )`$ have the following property at $`|\omega |\mathrm{\Gamma }_0`$ which will be useful in the following:
$$\begin{array}{c}\mathrm{Im}Z(\omega )=e^{\omega /T}\mathrm{Im}Z(\omega ),\hfill \\ \gamma (\omega )=e^{\omega /T}\gamma (\omega ).\hfill \end{array}$$
(31)
In fact these functions are exponentially small at negative $`\omega `$ if $`|\omega |T`$.
### III.2 Pseudofermion vertex
Let us turn to the consideration of the pseudofermion vertex $`\mathrm{\Gamma }_P(x+\omega ,x)`$. First diagrams for this quantity are presented in Fig. 4. It is shown in Appendix B that $`\mathrm{\Gamma }_{Pmm^{}}(x+\omega ,x)`$ is proportional to $`P_{mm^{}}`$. Thus, it is convenient to introduce a new quantity:
$$\mathrm{\Gamma }(x+\omega ,x)=\frac{\overline{P\mathrm{\Gamma }_P(x+\omega ,x)}}{\overline{P^2}},$$
(32)
where we use the following notification: $`\overline{Y}=\mathrm{Tr}(Y)`$. As is seen from Eq. (19), we need four different branches of $`\mathrm{\Gamma }(x+\omega ,x)`$. It is clear that $`\mathrm{\Gamma }^{++}=(\mathrm{\Gamma }^{})^{}`$ and within the first order of $`f^2`$ one has:
$`\mathrm{\Gamma }^{++}(x+\omega ,x)`$ $`=`$ $`1+R_1{\displaystyle \frac{f^2}{\pi \mathrm{\Theta }}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑yy|y|N(y)\mathrm{\Lambda }(y)G(x+y+\omega )G(x+y),`$ (33)
$`R_1`$ $`=`$ $`{\displaystyle \frac{\overline{PS^\mu PS^\nu }d_{\mu \nu }}{\overline{P^2}}}.`$ (34)
It is seen from Eq. (33) that the poles of $`G`$-functions in the integrand are on the one hand from the real axis. Hence, the second term in Eq. (33) is much smaller than unity and we can restrict ourself by this precision.
The situation is different in the case of $`\mathrm{\Gamma }^+=(\mathrm{\Gamma }^+)^{}`$. The first correction to it is given by Eq. (33) with $`G^{}(x+y)`$ put instead of $`G(x+y)`$. Therefore, poles of the Green’s functions are on the opposite sides of the real axis. As a result at $`\omega =0`$ the integral diverges at finite $`x`$ as $`\mathrm{\Gamma }_00`$ and one has to sum all series to determine $`\mathrm{\Gamma }^+`$. We write now an equation for $`\mathrm{\Gamma }^+`$ in which the most singular diagrams in each order of $`f^2`$ are taken into account. As a result of analysis of the diagrams up to the fourth order of $`f^2`$ we have obtained that in the most singular diagrams each wavy line connects points from different sides of the vertex (like in the second, the third and the fourth diagrams in Fig. 4 and not like in the last one) and crosses no more than one another wavy line. Thus, the most singular diagrams are taken into account in the following equation:
$`\mathrm{\Gamma }^+(x+\omega ,x)`$ $`=`$ $`1+R_1{\displaystyle \frac{f^2}{\pi \mathrm{\Theta }}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑yy|y|N(y)\mathrm{\Lambda }(y)\mathrm{\Gamma }^+(x+y+\omega ,x+y)G(x+y+\omega )G^{}(x+y)`$ (35)
$`+R_2\left({\displaystyle \frac{f^2}{\pi \mathrm{\Theta }}}\right)^2{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑y_1𝑑y_2y_1y_2|y_1y_2|N(y_1)N(y_2)\mathrm{\Lambda }(y_1)\mathrm{\Lambda }(y_2)\mathrm{\Gamma }^+(x+y_1+y_2+\omega ,x+y_1+y_2)`$
$`\times G(x+y_1+\omega )G(x+y_1+y_2+\omega )G^{}(x+y_1+y_2)G^{}(x+y_2),`$
$`R_2`$ $`=`$ $`{\displaystyle \frac{\overline{PS^\mu S^\nu PS^\mu ^{}S^\nu ^{}}d_{\mu \mu ^{}}d_{\nu \nu ^{}}}{\overline{P^2}}}.`$ (36)
The second and the third terms in Eq. (35) take into account diagrams with a rung and with crossing of two neighboring rungs, respectively. Let us try to solve this equation by iterations. It is easy to verify that at $`|\omega |\mathrm{\Gamma }_0`$ and $`\mathrm{\Gamma }_00`$ the divergence in the second term occurs in the second iteration only. Moreover it is of the same order as the divergence of the third term in the first iteration. As a result the equation (35) can be rewritten as follows:
$`\mathrm{\Gamma }^+(x+\omega ,x)`$ $`=`$ $`1+R_1{\displaystyle \frac{f^2}{\pi \mathrm{\Theta }}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑yy|y|N(y)\mathrm{\Lambda }(y)G(x+y+\omega )G^{}(x+y)`$ (37)
$`+\left({\displaystyle \frac{f^2}{\pi \mathrm{\Theta }}}\right)^2{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑y_1𝑑y_2y_1y_2|y_1y_2|N(y_1)N(y_2)\mathrm{\Lambda }(y_1)\mathrm{\Lambda }(y_2)\mathrm{\Gamma }^+(x+y_1+y_2+\omega ,x+y_1+y_2)`$
$`\times [R_1^2G^{}(x+y_1)+R_2G^{}(x+y_2)]G(x+y_1+\omega )G(x+y_1+y_2+\omega )G^{}(x+y_1+y_2).`$
It can be solved if one notes that at $`|\omega |T,\mathrm{\Theta }`$ in the third term the area of integration near poles of two last Green’s functions is essential. As the rest factors of the integrand changes slightly at such $`y_2`$ one can set in them $`y_2=xy_1`$ and neglect their dependence on $`\omega `$. As a result we obtain the following equation:
$`\mathrm{\Gamma }^+(x+\omega ,x)`$ $`=`$ $`1+R_1{\displaystyle \frac{f^2}{\pi \mathrm{\Theta }}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑yy|y|N(y)\mathrm{\Lambda }(y)G(x+y+\omega )G^{}(x+y)+K(x)\mathrm{\Gamma }^+(\omega ,0){\displaystyle \frac{2\pi i}{\omega +2i\mathrm{\Gamma }_0}},`$ (38)
$`K(x)`$ $`=`$ $`\left({\displaystyle \frac{f^2}{\pi \mathrm{\Theta }}}\right)^2{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑y{\displaystyle \frac{|y(x+y)|(R_2(x+y)R_1^2y)}{x+y+i\gamma (x+y)}}N(y)N(xy)\mathrm{\Lambda }(y)\mathrm{\Lambda }(x+y).`$ (39)
It can be easily solved with the result:
$`\mathrm{\Gamma }^+(x+\omega ,x)`$ $`=`$ $`1+R_1{\displaystyle \frac{f^2}{\pi \mathrm{\Theta }}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑yy|y|N(y)\mathrm{\Lambda }(y)G(x+y+\omega )G^{}(x+y)+K(x)Z_\mathrm{\Gamma }{\displaystyle \frac{2\pi i}{\omega +2i\mathrm{\Gamma }}},`$ (40)
$`\mathrm{\Gamma }`$ $`=`$ $`\mathrm{\Gamma }_0\pi K(0)=2\pi R_\mathrm{\Gamma }\left({\displaystyle \frac{f^2}{\pi \mathrm{\Theta }}}\right)^2{\displaystyle _0^{\mathrm{}}}𝑑yy^2N(y)[1+N(y)]\mathrm{\Lambda }^2(y),`$ (41)
$`Z_\mathrm{\Gamma }`$ $`=`$ $`1+R_1{\displaystyle \frac{f^2}{\pi \mathrm{\Theta }}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑yy|y|N(y)\mathrm{\Lambda }(y)|G(y)|^2,`$ (42)
$`R_\mathrm{\Gamma }`$ $`=`$ $`R_\gamma R_1^2+R_2={\displaystyle \frac{\overline{PS^\mu S^\nu [[S^\nu ^{},S^\mu ^{}],P]}d_{\nu \nu ^{}}d_{\mu \mu ^{}}}{\overline{P^2}}}.`$ (43)
In 2D AF $`R_\mathrm{\Gamma }=1/2`$ and 0 for $`P=S^x,S^y`$ and $`P=S^z`$, respectively. Evidently, the temperature dependence of $`\mathrm{\Gamma }`$ for $`P=S^x,S^y`$ is the same as that of $`\mathrm{\Gamma }_0`$ given by Eq. (30). Note, the third term in Eq. (40) is much smaller than the second one when $`|\omega |\mathrm{\Gamma }_0/f^2`$.
### III.3 Properties of the impurity susceptibility
We can derive now the impurity susceptibility using the general expression (19), Eqs. (26), (27), (28) and (30) for the Green’s function and Eqs. (33) and (40) for the branches of the vertex. As a result of tedious but simple calculations presented in Appendix C we have for the dynamical susceptibility of the impurity up to terms of the order of $`f^2`$:
$`\chi _P(\omega )`$ $`=`$ $`{\displaystyle \frac{\overline{P^2}}{2}}\{{\displaystyle \frac{2i\mathrm{\Gamma }}{T(\omega +2i\mathrm{\Gamma })}}(1+(R_12R_\mathrm{\Sigma }){\displaystyle \frac{f^2}{\pi \mathrm{\Theta }}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}dx{\displaystyle \frac{|x|N(x)\mathrm{\Lambda }(x)}{x+2i\mathrm{\Gamma }_0}})+{\displaystyle \frac{2i\mathrm{\Gamma }_0}{T(\omega +2i\mathrm{\Gamma }_0)}}R_1{\displaystyle \frac{f^2}{\pi \mathrm{\Theta }}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}dx{\displaystyle \frac{|x|N(x)\mathrm{\Lambda }(x)}{x+2i\mathrm{\Gamma }_0}}`$ (44)
$`+2R_\chi {\displaystyle \frac{f^2}{\pi \mathrm{\Theta }}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}dx{\displaystyle \frac{\mathrm{sgn}(x)\mathrm{\Lambda }(x)}{x+\omega +2i\mathrm{\Gamma }_0}}\},`$
$`R_\chi `$ $`=`$ $`R_\mathrm{\Sigma }R_1={\displaystyle \frac{\overline{PS^\mu [S^\nu ,P]}d_{\mu \nu }}{\overline{P^2}}}.`$ (45)
The first term in Eq. (44) is the Lorenz peak with the width $`\mathrm{\Gamma }`$. The second one is a Lorenz peak with the width $`\mathrm{\Gamma }_0`$ and small, proportional to $`f^2`$, amplitude. The last term is the non-resonant part of the susceptibility. The imaginary part of the non-resonant term in Eq. (44) at $`|\omega |\mathrm{\Gamma }_0`$ is proportional to $`\mathrm{sgn}(\omega )`$ and the real one contains the logarithmic singularity of the form $`\mathrm{ln}(\omega ^2+\mathrm{\Gamma }_0^2)`$. At $`T=0`$ and $`\omega 0`$ the nonresonant contribution survives only and the susceptibility has the logarithmic singularity. Such a singularity has been obtained for the two-level impurity at $`T=0`$ in Ref. Nagaosa et al. (1989). The first and the second terms in Eq. (44) are calculated assuming that $`|\omega |T`$. At $`|\omega |T`$ these terms are of the order of $`f^4`$ and their taking into account exceeds the range of accuracy.
It should be noted once more that the particular nature of the defect and the host system is not essential in the above consideration. We use the general expression (2) for the interaction and assume that the function $`\mathrm{Im}\mathrm{\Delta }_{\mu \nu }(\omega )`$ has the form (6). Apart from $`f^2`$ only coefficients $`R`$ in the resultant expression (44) depend on the nature of the defect and the host system. We demonstrate now that systems with different symmetries of the interaction show different behavior of the impurity. In the case of isotropic interaction all the components of tensor $`d`$ in Eq. (6) are nonzero and all components of the impurity susceptibility have the same structure: the Lorenz peaks and the nonresonant term. When one of the component of $`d`$ is zero, say $`zz`$\- one like in 2D AF, the behavior of transverse components $`\chi _x(\omega )`$ and $`\chi _y(\omega )`$ differs from that of longitudinal one $`\chi _z(\omega )`$. The transverse components have the Lorenz peak $`\mathrm{\Gamma }f^4`$ and the nonresonant term. The Lorenz peak with the width $`\mathrm{\Gamma }_0`$ disappears because $`R_1=0`$. The transverse component contains the nonresonant term but $`\mathrm{\Gamma }=0`$ (see Eqs. (41) and (43)) and our precision is insufficient to determine the resonance terms in $`\chi _z(\omega )`$. The corresponding calculations of $`\mathrm{\Gamma }_0`$ and $`\mathrm{\Gamma }`$ with higher precision is out of the scope of the present paper. If only $`d_{xx}`$ is nonzero, then $`\chi _x(\omega )0`$ whereas $`\chi _y(\omega )`$ and $`\chi _z(\omega )`$ have only the nonresonant term. This particular situation is considered in detail in the next section.
For static susceptibility $`\chi _P(0)`$ we have from Eq. (44):
$`\chi _P(0)`$ $`=`$ $`{\displaystyle \frac{\overline{P^2}}{2T}}\left(1+W(T)\right),`$ (46)
$`W(T)`$ $`=`$ $`2R_\chi {\displaystyle \frac{f^2}{\pi \mathrm{\Theta }}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑x{\displaystyle \frac{\mathrm{sgn}(x)\mathrm{\Lambda }(x)}{x+2i\mathrm{\Gamma }_0}}\left(TxN(x)\right),`$ (47)
where the first and the second terms in $`W(T)`$ stem from the non-resonant and the resonant parts in Eq. (44), respectively. It is easy to verify that $`W(T)`$ is a real value up to the order of $`f^2`$ (remember, the susceptibility has been calculated up to terms of this order). Integrations in Eq. (47) can be simply carried out and we have at $`T\mathrm{\Theta }`$:
$$\chi _P(0)=\frac{\overline{P^2}}{2T}\left(12R_\chi \frac{f^2}{\pi }\right)+\overline{P^2}2R_\chi \frac{f^2}{\pi \mathrm{\Theta }}\mathrm{ln}\left(\frac{\mathrm{\Theta }}{T}\right).$$
(48)
Thus the uniform susceptibility has the free-spin-like term $`\overline{P^2}(2T)^1`$ which amplitude is slightly reduced by the interaction and the correction proportional to $`f^2\mathrm{ln}(\mathrm{\Theta }/T)`$. Expression (48) is in accordance with that of Ref. Clarke et al. (1993). Similar result for the static susceptibility, $`1/T`$-term and a logarithmic singular correction to it, has been obtained in 2D AF near QCP. Sachdev and Vojta (2003) It was found that the static susceptibility exhibits the classical-like Curie behavior of the form $`S^2/(3T)`$ and the coefficients before and under the logarithm are universal values independent of the particular type of the impurity and the strength of its coupling to the host system. Remarkably, this behavior remains also far from QCP for asymmetrically coupled impurities (vacancy and added spin) at $`Tg`$. Hoglund and Sandvik (2003, 2004); Sachdev and Vojta (2003); Sushkov (2003) but the constant under the logarithm becomes non-universal. These findings are related to the nontrivial long-range dynamics of the 2D AF. Then we point out the significant difference between dynamical properties of symmetrically and asymmetrically coupled impurities in the regime $`Tg`$: the leading $`1/T`$-terms have the free-spin-like behavior and the classical-like one, respectively. Moreover, the logarithmic corrections is proportional to $`g^2`$ in the symmetric case and it does not depend on $`g`$ in asymmetric one. Sushkov (2003); Sachdev and Vojta (2003) As was also pointed out in Introduction, the difference can be explained by the fact that the impurity spin coupled asymmetrically aligns with the local Neel order Sushkov (2003); Sachdev and Vojta (2003) at $`Tg`$ whereas the symmetrically coupled impurity is located in the zero molecular field.
It is convenient from this point on to neglect in Eq. (44) the small corrections and use the following simple expression for the transverse susceptibility $`\chi _{}(\omega )=\chi _x(\omega )=\chi _y(\omega )`$:
$`\chi _{}(\omega )`$ $`=`$ $`\overline{P^2}{\displaystyle \frac{i\mathrm{\Gamma }}{T(\omega +2i\mathrm{\Gamma })}}+\overline{P^2}R_\chi {\displaystyle \frac{f^2}{\pi \mathrm{\Theta }}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑x{\displaystyle \frac{\mathrm{sgn}(x)\mathrm{\Lambda }(x)}{x+\omega +i\mathrm{\Gamma }_0}},`$ (49)
$`\mathrm{Im}\chi _{}(\omega )`$ $`=`$ $`\overline{P^2}{\displaystyle \frac{\omega \mathrm{\Gamma }}{T(\omega ^2+4\mathrm{\Gamma }^2)}}+\overline{P^2}R_\chi {\displaystyle \frac{f^2}{\mathrm{\Theta }}}\mathrm{\Lambda }(\omega )\mathrm{sgn}(\omega ).`$ (50)
The imaginary part of the non-resonant term is calculated at $`|\omega |\mathrm{\Gamma }_0`$ in Eq. (50). We see that it does not depend on the temperature at such $`\omega `$. It is seen from Eq. (49) that the non-resonant term gives the main contribution to the susceptibility when
$$|\omega |\omega _0=f^2\mathrm{\Theta }\left(\frac{T}{\mathrm{\Theta }}\right)^2.$$
(51)
As the sign of $`\mathrm{Im}\chi _P(\omega )`$ should coincide with that of $`\omega `$, the constant $`R_\chi `$ given by Eq. (45) should be positive. For example, we have for 2D AF: $`R_\chi =1/2`$.
It should be stressed that there is a restriction on the range of validity of the resultant expressions (44) and (49) for $`\chi _P(\omega )`$ in the case of 2D AF. It is the consequence of the fact that the function $`\mathrm{Im}\mathrm{\Delta }(\omega )`$ has the form (6) if $`\omega \{\eta \text{ or }Ja/\xi \}`$ only. It is easy to see that in all calculations performed above one can use the function of the form (6) if the following condition on $`\omega `$ and $`\mathrm{\Gamma }_0`$ holds: $`\mathrm{max}\{\mathrm{\Gamma }_0,|\omega |\}\{\eta \text{ or }Ja/\xi \}`$.
### III.4 Comparison with the exactly solvable model
Let us consider the special case when the interaction contains only one term: $`_{int}=gS^xϵ^x(𝐑_0)`$. It is seen from Eqs. (29) and (43) that $`\mathrm{\Gamma }=\mathrm{\Gamma }_0=0`$ and the resonant terms are zero at $`\omega 0`$ in Eq. (44). One can conclude from Eqs. (45) and (49) that $`\chi _x(\omega )=0`$ and
$$\chi _y(\omega )=\chi _z(\omega )=\chi ^{()}(\omega )=\frac{f^2}{2\pi \mathrm{\Theta }}\mathrm{ln}\left|\frac{\mathrm{\Theta }}{\omega }\right|+i\frac{f^2}{4\mathrm{\Theta }}\mathrm{\Lambda }(\omega )\mathrm{sgn}(\omega ).$$
(52)
As it was pointed out above, our model in this case is equivalent to the spin-boson model (1) without $`_d`$. The corresponding Hamiltonian can be diagonalized exactly and an exact expression for the impurity susceptibility can be derived. We perform in this subsection the corresponding calculations of $`\chi ^{()}(\omega )`$ and confirm our results obtained above.
The detailed discussion of the exactly solvable spin-boson model is also necessary by the following reason. It was important for our consideration that $`\gamma (\omega )`$ in expression (26) for the Green’s function is nonzero, i.e., the coefficient $`R_\gamma `$ given by Eq. (29) is finite. As is demonstrated in Sec. III.1, the imaginary term $`i\gamma (\omega )`$ in the denominator of the Green’s function $`G(\omega )`$ screens the logarithmic singularity of the self-energy part allowing us to restrict ourself by the second order of $`f^2`$. In the opposite case, when imaginary part in the denominator of $`G(\omega )`$ is zero, one has to sum all the series to calculate the Green’s function in the region of $`\omega `$ determined by the condition
$$f^2\frac{T}{\mathrm{\Theta }}\mathrm{ln}\left|\frac{T}{\omega }\right|1.$$
(53)
It should be stressed that Eq. (52) for $`\chi ^{()}(\omega )`$ is valid if the condition (53) on the frequency $`\omega `$ is not fulfilled. If the condition (53) is fulfilled the results of the exact diagonalization should be discussed.
We represent Hamiltonian (1) of the spin-boson model describing degenerate defect in the following form:
$$=\frac{1}{2}\underset{𝐤}{}(ϵ_𝐤^2Q_𝐤Q_𝐤+P_𝐤P_𝐤)+gS^x\underset{𝐤}{}F_𝐤Q_𝐤,$$
(54)
where the symbols $`ϵ_𝐤`$, $`Q_𝐤`$ and $`P_𝐤`$ stand for the frequency, normal coordinate and momentum of the system propagating modes (bosons) with momentum $`𝐤`$, where $`[Q_𝐤,P_𝐤^{}]=i\delta _{\mathrm{𝐤𝐤}^{}}`$, $`Q_𝐤=\alpha _𝐤+\alpha _𝐤^{}`$, $`P_𝐤=iϵ_𝐤(\alpha _𝐤\alpha _𝐤^{})`$. The last term in Eq. (54) describes coupling of the impurity with the system, where $`F_𝐤`$ is a coupling parameter. In this problem definition the spectral function given by Eq. (3) has the form
$$\mathrm{Im}\mathrm{\Delta }(\omega )=\frac{\pi }{2}\underset{𝐤}{}\frac{|F_𝐤|^2}{ϵ_𝐤}[\delta (\omega ϵ_𝐤)\delta (\omega +ϵ_𝐤)].$$
(55)
We treat the elementary excitations within Debye approximation and assume that the spectrum is linear in $`k`$: $`ϵ_𝐤=ck`$. We make also one more assumption which do not effect the results: $`𝐤`$ is a two-dimensional vector. In this case the spectral function is proportional to $`\omega ^2`$ when the coupling parameter is linear in $`k`$:
$$F_𝐤=\frac{1}{\mathrm{\Theta }\sqrt{V}}ϵ_𝐤,$$
(56)
where $`V`$ is the volume of the crystal and $`\mathrm{\Theta }`$ is the cut-off frequency. In this case the spectral function (55) has the form (6) with $`A=(4c^2)^1`$.
The Hamiltonian (54) can be diagonalized exactly. It is convenient for this purpose to apply the following canonical transformation: Pirc and Dick (1974)
$$e^Re^R=\frac{1}{2}\underset{𝐤}{}[ϵ_𝐤^2Q_𝐤Q_𝐤+P_𝐤P_𝐤]\frac{g^2}{8}\underset{𝐤}{}\frac{|F_𝐤|^2}{ϵ_𝐤^2},R=ig\underset{𝐤}{}\frac{F_𝐤}{ϵ_𝐤^2}P_𝐤S^x.$$
(57)
It can be shown that the correlation function can be brought to the form: Pirc and Dick (1974)
$`S^y(t)S^y(0)={\displaystyle \frac{1}{4}}e^{I(t)},`$ (58)
$`I(t)=g^2{\displaystyle \underset{𝐤}{}}{\displaystyle \frac{|F_𝐤|^2}{2ϵ_𝐤^3}}\{N(ϵ_𝐤)[e^{iϵ_𝐤t}1]+[N(ϵ_𝐤)+1][e^{iϵ_𝐤t}1]\}.`$ (59)
As a result we have for the transverse dynamical susceptibility using the representation $`I(t)=X(t)iU(t)`$:
$$\chi ^{()}(\omega )=\frac{1}{2}_0^{\mathrm{}}𝑑te^{i\omega t}e^{X(t)}\mathrm{sin}[U(t)].$$
(60)
One obtains for $`X(t)`$ and $`U(t)`$ from Eq. (59):
$`X(t)`$ $`=`$ $`{\displaystyle \frac{f^2}{\pi \mathrm{\Theta }}}{\displaystyle _0^\mathrm{\Theta }}𝑑x[\mathrm{cos}(xt)1][1+2N(x)],`$ (61)
$`U(t)`$ $`=`$ $`{\displaystyle \frac{f^2}{\pi \mathrm{\Theta }}}{\displaystyle _0^\mathrm{\Theta }}𝑑x\mathrm{sin}(xt)={\displaystyle \frac{f^2}{\pi }}{\displaystyle \frac{1\mathrm{cos}(\mathrm{\Theta }t)}{\mathrm{\Theta }t}},`$ (62)
where $`f^2`$ is given by Eq. (20). It is seen from Eq. (62) that $`U(t)`$ is a bounded function and we have in Eq. (60): $`\mathrm{sin}[U(t)]U(t)`$. The function $`X(t)`$ along with negligibly small terms of the order of $`f^2`$ has another one which is large at $`tT\mathrm{exp}\{\mathrm{\Theta }/(f^2T)\}`$ and for which we have with the logarithmic precision: $`X(t)2f^2T(\pi \mathrm{\Theta })^1\mathrm{ln}(tT)`$. <sup>1</sup><sup>1</sup>1It is seen from Eq. (58) that the correlation function tends to zero as $`t\mathrm{}`$. This signifies that the system is ergodic. Pirc and Dick (1974) It means, in particular, that the isolated static susceptibility (14) coincides with isothermal static one $`\chi ^T=(M/H)_T`$, where $`M`$ is the magnetization of the sample. Pirc and Dick (1974) As a result we obtain:
$`\chi ^{()}(\omega )={\displaystyle \frac{f^2}{2\pi }}\left({\displaystyle \frac{\mathrm{\Theta }}{T}}\right)^\zeta {\displaystyle _0^{\mathrm{}}}𝑑te^{i\omega t}{\displaystyle \frac{1\mathrm{cos}(t\mathrm{\Theta })}{(t\mathrm{\Theta })^{1+\zeta }}}`$ $`=`$ $`{\displaystyle \frac{1}{4T}}\left({\displaystyle \frac{\mathrm{\Theta }}{T}}\right)^\zeta \left[{\displaystyle \frac{1}{2}}\left|1+{\displaystyle \frac{\omega }{\mathrm{\Theta }}}\right|^\zeta +{\displaystyle \frac{1}{2}}\left|1{\displaystyle \frac{\omega }{\mathrm{\Theta }}}\right|^\zeta \left|{\displaystyle \frac{\omega }{\mathrm{\Theta }}}\right|^\zeta \right]`$
$`i{\displaystyle \frac{f^2}{4\mathrm{\Theta }}}\left({\displaystyle \frac{\mathrm{\Theta }}{T}}\right)^\zeta \left[{\displaystyle \frac{1}{2}}\left|1+{\displaystyle \frac{\omega }{\mathrm{\Theta }}}\right|^\zeta {\displaystyle \frac{1}{2}}\left|1{\displaystyle \frac{\omega }{\mathrm{\Theta }}}\right|^\zeta \mathrm{sgn}(\mathrm{\Theta }\omega )\left|{\displaystyle \frac{\omega }{\mathrm{\Theta }}}\right|^\zeta \right]\mathrm{sgn}(\omega ),`$
where $`\zeta =2f^2T(\pi \mathrm{\Theta })^1`$. We recover Eq. (52) from Eq. (III.4) if the condition (53) is not fulfilled and $`|\omega |\mathrm{\Theta }`$ or $`|\omega |\mathrm{\Theta }`$. At $`\omega \mathrm{\Theta }`$ Eq. (III.4) can be represented in the simple form:
$$\chi ^{()}(\omega )=\frac{1}{4T}\left(\frac{\mathrm{\Theta }}{T}\right)^\zeta \left[1\left(\frac{\omega }{i\mathrm{\Theta }}\right)^\zeta \right]$$
(64)
At small enough $`\omega `$, when (53) holds, equation (52) is incorrect and we see from Eq. (64) that the susceptibility shows the nontrivial $`\omega `$\- and $`T`$\- dependences. In particular, the static susceptibility is proportional to $`T^{1\zeta }`$. In the order of $`f^2`$ the static susceptibility has the same structure as that obtained above (see Eq. (48)): it has the conventional term $`1/(4T)`$ and the logarithmic correction to this term. Thus, we see that in the modified spin-boson model taking into account higher order logarithmic corrections results in the non-trivial power-law $`T`$-dependence of $`\chi (0)`$.
## IV Influence of the defects on the host system
We discuss in this section the influence of the defects with finite concentration $`n`$ on the low-temperature properties of 2D AF. The spin-wave spectrum and the specific heat of AF are considered below in detail. It will be assumed that $`n1`$ in order to neglect interaction between impurities.
### IV.1 Spin-wave spectrum
The spin-wave spectrum is determined by the poles of the spin Green’s functions. The Green’s functions of 2D AF with impurities are investigated in Appendix D. It is demonstrated there that their denominator has the form
$$𝒟(\omega ,𝐤)=\omega ^2ϵ_𝐤^2+4s^2g^2n\chi _{}(\omega )\left[J_0J_𝐤\mathrm{cos}(\mathrm{𝐤𝐑}_{12})\right],$$
(65)
where $`𝐑_{12}`$ is the vector connected host spins coupled to impurity (it is assumed for beginning that this vector is the same for all defects), $`J_𝐤=2J(\mathrm{cos}k_x+\mathrm{cos}k_z)`$, where $`J`$ is the exchange constant, and $`ϵ_𝐤=s\sqrt{J_0^2J_𝐤^2}`$.
Let us discuss the spin-wave spectrum near points $`k=0`$ and $`k=k_0`$, where $`𝐤_0`$ is the antiferromagnetic vector. It is seen that expression (65) is symmetric under replacement of $`𝐤`$ by $`𝐤\pm 𝐤_0`$. Thus, we consider below only the vicinity of the point $`k=0`$. We have from Eq. (65):
$`𝒟(\omega ,𝐤)`$ $`=`$ $`\omega ^2ϵ_𝐤^2\left[1{\displaystyle \frac{nf^2}{2\pi }}\mathrm{\Theta }u(𝐤)\chi _{}(\omega )\right],`$ (66)
$`u(𝐤)`$ $`=`$ $`{\displaystyle \frac{1}{2}}+{\displaystyle \frac{(\mathrm{𝐤𝐑}_{12})^2}{k^2}},`$ (67)
where it is used that the unperturbed spectrum is linear at $`kk_0`$: $`ϵ_𝐤=ck=\sqrt{8}sJk`$. It is seen from Eqs. (66) and (67) that the spectrum appears to be dependent on the direction of the momentum $`𝐤`$ as a result of interaction of magnons with the defects. This circumstance is a consequence of our assumption that the vector $`𝐑_{12}`$ is the same for all impurities. In fact, it can have four directions and the value $`(𝐑_{12}𝐤)^2/k^2`$ can have two different values: $`\mathrm{cos}^2\varphi _𝐤`$ and $`\mathrm{sin}^2\varphi _𝐤`$, where $`\varphi _𝐤`$ is the azimuthal angle of $`𝐤`$. It easy to realize that $`u(𝐤)=1`$ if all four ways of coupling of the impurity with AF are equally possible.
It is convenient for the following to consider separately the cases of $`|\omega |\omega _0`$ and $`|\omega |\omega _0`$, where $`\omega _0`$ is given by Eq. (51). In these cases the non-resonant and the resonant parts, respectively, are dominant in the impurity susceptibility (49).
$`|\omega |\omega _0`$. One obtains from Eqs. (49), (50) and (66) for the magnon damping $`\gamma _𝐤`$ and the renormalized spin-wave velocity $`\stackrel{~}{c}_𝐤`$:
$`\gamma _𝐤`$ $`=`$ $`|\omega |{\displaystyle \frac{nf^4}{16\pi }}u(𝐤),`$ (68)
$`\stackrel{~}{c}_𝐤^2`$ $`=`$ $`c^2\left(1\mathrm{ln}\left|{\displaystyle \frac{\mathrm{\Theta }}{\omega }}\right|{\displaystyle \frac{nf^4}{4\pi ^2}}u(𝐤)\right),`$ (69)
where we take into account that $`R_\chi =1/2`$. It is seen that the interaction with the defects leads to strong damping which is proportional to $`\omega `$ and to the logarithmic correction to the spin-wave velocity. It would seem that at small enough $`k`$ the spin-wave velocity becomes imaginary signifying a phase transition in the system. Meanwhile our theory is not applicable at such small $`k`$. The interaction with the defects changes the function $`\mathrm{Im}\mathrm{\Delta }_{\mu \nu }(\omega )`$ as well and this renormalization is strong at small enough $`\omega `$. Indeed, one has to use renormalized spin Green’s function derived in Appendix D to evaluate $`\mathrm{Im}\mathrm{\Delta }_{\mu \nu }(\omega )`$. As a result of simple calculations similar to those presented in Appendix A we obtain that at $`|\omega |\mathrm{\Theta }`$ in addition to the term proportional to $`\omega ^2\mathrm{sgn}(\omega )`$ there is another one proportional to $`\mathrm{sgn}(\omega )`$:
$`\mathrm{Im}\mathrm{\Delta }_{\mu \nu }(\omega )`$ $`=`$ $`A\left({\displaystyle \frac{\omega }{\mathrm{\Theta }}}\right)^2\mathrm{sgn}(\omega )d_{\mu \nu }\mathrm{sgn}(\omega ){\displaystyle \frac{snf^4B}{\mathrm{\Theta }}}d_{\mu \nu },`$ (70)
$`B`$ $`=`$ $`{\displaystyle \frac{s^4J^2}{\pi ^{5/2}}}{\displaystyle 𝑑𝐤\frac{(1+\mathrm{cos}(\mathrm{𝐤𝐑}_{12}))^2(J_\mathrm{𝟎}J_𝐤)^2}{ϵ_𝐤^4}}0.2,`$ (71)
where the integral is taken over the chemical Brillouin zone. The first term in Eq. (70) is greater than the second one when
$$|\omega |0.02\mathrm{\Theta }\sqrt{n}f^2.$$
(72)
This condition determines the range of validity of our theory at $`|\omega |\omega _0`$. At $`\omega `$ given by (72) the logarithmic correction to the spin-wave velocity in Eq. (69) is small.
$`|\omega |\omega _0`$. In this case the impurity susceptibility (49) is determined by the resonant term. We have for the spin-wave damping and the spin-wave velocity from Eqs. (49), (50) and (66):
$`\gamma _𝐤`$ $`=`$ $`{\displaystyle \frac{\mathrm{\Theta }}{T}}{\displaystyle \frac{\omega ^2\mathrm{\Gamma }}{\omega ^2+4\mathrm{\Gamma }^2}}{\displaystyle \frac{nf^2}{8\pi }}u(𝐤),`$ (73)
$`\stackrel{~}{c}_𝐤^2`$ $`=`$ $`c^2\left(1{\displaystyle \frac{\mathrm{\Theta }}{T}}{\displaystyle \frac{\mathrm{\Gamma }^2}{\omega ^2+4\mathrm{\Gamma }^2}}{\displaystyle \frac{nf^2}{2\pi }}u(𝐤)\right).`$ (74)
As in the case of $`|\omega |\omega _0`$, one has to take into account the renormalization of the function $`\mathrm{Im}\mathrm{\Delta }_{\mu \nu }(\omega )`$. After simple calculations we obtain at $`|\omega |\mathrm{\Theta }`$:
$$\mathrm{Im}\mathrm{\Delta }_{\mu \nu }(\omega )=A\left(\frac{\omega }{\mathrm{\Theta }}\right)^2\mathrm{sgn}(\omega )d_{\mu \nu }2snf^2B\frac{\omega \mathrm{\Gamma }}{T(\omega ^2+4\mathrm{\Gamma }^2)}d_{\mu \nu },$$
(75)
where the constant $`B`$ is given by Eq. (71). As a result the range of validity of our consideration is determined by
$$\frac{|\omega |(\omega ^2+4\mathrm{\Gamma }^2)}{\mathrm{\Theta }^3}0.004nf^2\frac{\mathrm{\Gamma }}{T}.$$
(76)
It is easy to show that the spin-wave damping (73) and the correction to the spin-wave velocity in Eq. (74) is small at such $`\omega `$.
It should be noted once more that if one of the conditions (72) or (76) is violated the interaction between defects becomes important and our approach is wrong.
### IV.2 Specific heat
We proceed with the discussion of the magnetic part of the specific heat $`C(T)`$. It is convenient to use the following formula for its evaluation:
$$C(T)=\frac{dE}{dT}=\frac{d}{dT}\underset{𝐤}{}ϵ_𝐤\left(\alpha _𝐤^{}\alpha _𝐤+\frac{1}{2}\right)+\underset{i}{}_{int}^{(i)},$$
(77)
where the first term describes magnetic excitations and index $`i`$ in the second term labels impurities. To calculate the first term in Eq. (77) we use bilinear part of the Hamiltonian (86) and spin Green’s functions (D) renormalized by interaction with impurities. As a result of simple calculations one obtains up to inessential terms not depending on the temperature:
$`{\displaystyle \underset{𝐤}{}}ϵ_𝐤\left(\alpha _𝐤^{}\alpha _𝐤+{\displaystyle \frac{1}{2}}\right)`$ $`=`$ $`{\displaystyle \underset{𝐤}{}}ϵ_𝐤N(ϵ_𝐤)+2Nnf^2\mathrm{\Theta }X{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑\omega N(\omega )\mathrm{Im}\chi _{}(\omega ),`$ (78)
$`X`$ $`=`$ $`{\displaystyle \frac{s^2J}{2\pi ^4}}{\displaystyle 𝑑𝐤\frac{J_\mathrm{𝟎}J_𝐤\mathrm{cos}(\mathrm{𝐤𝐑}_{12})}{ϵ_𝐤^2}}0.05,`$ (79)
where the integration in Eq. (79) is over the chemical Brillouin zone. The diagram for the second term in Eq. (77) is shown in Fig. 5. It contains the impurity susceptibility calculated above and $`\mathrm{\Delta }_{\mu \nu }(\omega )`$:
$$\underset{i}{}_{int}^{(i)}=N\frac{ng^2}{\pi }\underset{\nu }{}_{\mathrm{}}^{\mathrm{}}𝑑\omega N(\omega )[\mathrm{Im}\chi _\nu (\omega )\mathrm{Re}\mathrm{\Delta }_{\nu \nu }(\omega )+\mathrm{Re}\chi _\nu (\omega )\mathrm{Im}\mathrm{\Delta }_{\nu \nu }(\omega )].$$
(80)
Evaluating expression (80) and summing it with Eq. (78) one leads to the following expression:
$`{\displaystyle \frac{E}{N}}`$ $`=`$ $`{\displaystyle \frac{1}{N}}{\displaystyle \underset{𝐤}{}}ϵ_𝐤N(ϵ_𝐤){\displaystyle \frac{2nf^2}{\pi \mathrm{\Theta }}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑\omega N(\omega )\omega |\omega |\mathrm{\Lambda }(\omega )\mathrm{Re}\chi _{}(\omega )+nf^2\mathrm{\Theta }X{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑\omega N(\omega )\mathrm{Im}\chi _{}(\omega ),`$ (81)
where the constant $`X`$ is given by Eq. (79). The first term in Eq. (81) describes the energy of 2D AF without impurities. It is equal to $`E_0NT^3/\mathrm{\Theta }^2`$ at $`T\mathrm{\Theta }`$ and we come to the well known result: $`C(T)T^2`$ for pure 2D AF. Evaluation of corrections to $`E_0`$ from the second and the third terms in Eq. (81) is a tedious but straightforward work. Unfortunately Eq. (49) for $`\chi _{}(\omega )`$ has the limited range of validity which is discussed in the previous section. Therefore one can not carry out the integration in the second and the third terms of Eq. (81) in the whole range of $`\omega `$. We have evaluated these terms after the integration over $`\omega `$ at which Eq. (49) is valid. There are corrections having weaker $`T`$-dependence than $`E_0`$ stemming from both the resonant and the nonresonant parts of $`\chi _{}(\omega )`$. Meanwhile restrictions (72) and (76) result in the considered corrections to be bounded below on $`T`$ and to be smaller than $`E_0`$.
Thus, we do not obtain a renormalization of the specific heat within our precision.
## V Conclusion
We discuss the dynamical properties of the impurity spin-$`\frac{1}{2}`$ in 2D and quasi-2D Heisenberg antiferromagnets (AFs) at $`T0`$. The specific case of the impurity that is coupled symmetrically to two neighboring host spins is considered. It is shown that this problem is a generalization of the spin-boson model without the tunneling term and with a more complex interaction. It is demonstrated that the effect of the host system on the defect is completely described by the spectral function $`𝒥(\omega )`$ which is proportional to $`\omega ^2`$. It is found within the spin-wave approximation that the spectral function has this $`\omega `$-dependence in 2D AF for not too small $`\omega `$. In isotropic 2D AF at $`T>0`$ $`𝒥(\omega )\omega ^2`$ when $`\omega Ja/\xi `$, where $`J`$ is the exchange constant between the host spins, $`\xi `$ is the correlation length and $`a`$ is the lattice constant. For ordered 2D AF $`𝒥(\omega )\omega ^2`$ at $`\omega \eta `$, where $`\eta J`$ is the value of interaction (for definiteness interplane interaction) stabilizing the long range order at finite $`T`$.
We stress that one must distinguish symmetrically and asymmetrically coupled impurities (see Fig. 1). Symmetrically coupled impurity is located in the zero molecular field. It remains degenerate and the spectral function is proportional to $`\omega ^2`$. In the case of asymmetrically coupled impurity, where the molecular field is nonzero, there is splitting of the impurity levels and the spectral function has terms with weaker $`\omega `$-dependence. For instance, we demonstrate that the spectral function for impurity coupled to one host spin is proportional to a constant. In this paper we consider only the symmetric case. Our results are also valid with certain additional restrictions for slightly split nearly symmetrically coupled impurities (see below).
The defect dynamical susceptibility $`\chi (\omega )`$ is derived using Abrikosov’s pseudofermion technique and diagrammatic expansion. The calculations are performed within the order of $`f^4`$, where $`fg/J`$ is the dimensionless coupling parameter. For our study the sign of $`f`$ is insignificant. We show that the transverse impurity susceptibility $`\chi _{}(\omega )`$ has a Lorenz peak with the widths $`\mathrm{\Gamma }f^4J(T/J)^3`$ that disappears at $`T=0`$, and a non-resonant term. The longitudinal susceptibility $`\chi _{}(\omega )`$ has the non-resonant term which differs from that of $`\chi _{}(\omega )`$ by a constant and a Lorenz peak. The width of the peak is zero within the order of $`f^4`$. Its calculation is out of the scope of this paper. The imaginary part of non-resonant term is a constant independent of $`T`$ at $`|\omega |\mathrm{\Gamma }`$ and the real part has a logarithmic divergence as $`\omega ,T0`$. Similar logarithmic singularity was found in Ref. Nagaosa et al. (1989) at $`T=0`$.
The static susceptibility has the free-spin-like term $`S(S+1)/(3T)`$ and a correction proportional to $`f^2\mathrm{ln}(J/T)`$. We point out here the sharp difference between symmetrically and asymmetrically coupled impurities that takes place in the regime $`T|g|`$ (by asymmetrically coupled impurities we mean here either the added spin coupled to one host spin or the vacancy which is the particular case of the added spin with $`g\mathrm{}`$). The leading $`1/T`$-term has the free-spin-like form in the symmetric case and the classical-like form in the asymmetric one. Moreover, the logarithmic correction is proportional to $`g^2`$ in the symmetric case and it does not depend on $`g`$ in the asymmetric one. Sushkov (2003); Sachdev and Vojta (2003) The difference is related to the fact that the impurity spin coupled asymmetrically aligns with the local Neel order, Sushkov (2003); Sachdev and Vojta (2003) whereas the symmetrically coupled impurity is located in the zero molecular field.
The fact that the spectral function in 2D AF is proportional to $`\omega ^2`$ only at $`\omega \{\eta \text{ or }Ja/\xi \}`$ leads to the following restriction on the results obtained: $`\mathrm{max}\{\mathrm{\Gamma }_0,|\omega |\}\{\eta \text{ or }Ja/\xi \}`$. If the defect is slightly split (for definiteness by magnetic field $`𝐇`$) this condition turns into $`\mathrm{max}\{\mathrm{\Gamma }_0,|\omega |\}\mathrm{max}\{\{\eta \text{ or }Ja/\xi \},g\mu _BHS\}`$. For nearly symmetrically coupled impurity one has: $`\mathrm{max}\{\mathrm{\Gamma }_0,|\omega |\}\mathrm{max}\{\{\eta \text{ or }Ja/\xi \},|g_1g_2|\}`$, where $`g_{1,2}`$ are values of coupling with the corresponding sublattices (see Fig. 1).
The findings discussed above are valid for isotropic interaction of the impurity spin $`𝐒`$ with AF: $`_{int}=g𝐒(𝐬_1+𝐬_2)`$, where $`𝐬_{1,2}`$ are host spins from different sublattices. We consider also interaction containing only one component of $`𝐒`$: $`_{int}=gS^x(s_1^x+s_2^x)`$. The results in this case are quite specific. We show that $`xx`$\- component of the impurity susceptibility is zero whereas $`yy`$\- and $`zz`$\- ones have only the non-resonant term. Our model is identical to the spin-boson one (1) without $`_d`$ if the interaction contains a term with only one component of $`𝐒`$. The Hamiltonian can be diagonalized exactly Pirc and Dick (1974) and an exact expression for $`\chi (\omega )`$ can be obtained. We perform the corresponding calculations and confirm the results obtained by our approach. One of the most interesting features of the exact result is that the static susceptibility has the form $`\chi (0)T^{1\zeta }`$, where $`\zeta f^2T/J`$. Within the first order of $`f^2`$ one has $`1/(4T)`$-term and the logarithmic correction. Thus, we see that in the modified spin-boson model taking into account the higher order logarithmic corrections results in the non-trivial power-law $`T`$-dependence of $`\chi (0)`$.
The influence of the finite concentration $`n`$ of the defects on the low-temperature properties of 2D AF is also considered. For not too small $`\omega `$ we find the logarithmic correction to the spin-wave velocity of the form $`nf^4\mathrm{ln}|J/\omega |`$ and an anomalous damping of the spin-waves proportional to $`nf^4|\omega |`$. Similar logarithmic correction to the velocity and damping were obtained in Ref. Chernyshov et al. (2002), where vacancies in 2D AF were studied. It is demonstrated that interaction of the spin waves with defects modifies the spectral function which acquires new terms proportional to $`n`$ exhibiting weaker $`\omega `$-dependence. These terms should be taken into account at small enough $`\omega `$ and the problem should be solved self-consistently. The corresponding consideration is out of the scope of this paper. Within the range of validity of our study we do not obtain a renormalization of the magnetic specific heat which is proportional to $`T^2`$ in 2D AF without impurities.
The results of the present paper can be applied to other systems with a degenerate defect in which the spectral function is proportional to $`\omega ^2`$.
###### Acknowledgements.
We are thankful to A. V. Lazuta for stimulating discussions. This work was supported by Russian Science Support Foundation (A.V.S.), RFBR (Grant Nos. SS-1671.2003.2, 03-02-17340, 06-02-16702 and 00-15-96814), and Russian Programs ”Quantum Macrophysics”, ”Strongly correlated electrons in semiconductors, metals, superconductors and magnetic materials” and ”Neutron Research of Solids”.
## Appendix A Calculation of $`\mathrm{Im}\mathrm{\Delta }_{\mu \nu }(\omega )`$ in 2D AF
In this appendix we discuss properties of the imaginary part of the function $`\mathrm{\Delta }_{\mu \nu }(\omega )`$ general expression for which is given by Eq. (3). It is shown below that within the spin-wave approximation $`\mathrm{Im}\mathrm{\Delta }_{\mu \nu }(\omega )`$ has the form (6) and expressions for the constant $`A`$, the characteristic energy $`\mathrm{\Theta }`$ and the tensor $`d_{\mu \nu }`$ are obtained.
We have from Eqs. (3) and (4):
$$\mathrm{\Delta }_{\mu \nu }(\omega )=\frac{2}{N}\underset{𝐤}{}[1+\mathrm{cos}(\mathrm{𝐤𝐑}_{12})]s_𝐤^\mu s_𝐤^\nu _\omega ,$$
(82)
where $`\mathrm{}_\omega `$ denote retarded Green’s function, $`N`$ is the number of spins in the lattice and $`𝐑_{12}`$ is the vector connecting two host spins coupled to the defect. Thus we have to calculate the spin Green’s functions $`s_𝐤^\mu s_𝐤^\nu _\omega `$ of 2D AF. The Hamiltonian of 2D Heisenberg AF on the square lattice has the well-known form:
$$H=J\underset{ij}{}𝐬_i𝐬_j.$$
(83)
We perform for beginning all calculations for isotropic Heisenberg 2D AF at $`T=0`$ and then consider the effect of finite $`T`$ and of the additional small interaction stabilizing the long range order at finite $`T`$.
Instead of dividing of the lattice onto two sublattices it is convenient to represent operators $`s_𝐤^\mu `$ as follows (see, e.g., Refs. Maleyev (2000); Petitgrand et al. (1999)):
$$𝐬_𝐤=\widehat{x}s_𝐤^x+\widehat{y}s_{𝐤+𝐤_0}^y+\widehat{z}s_{𝐤+𝐤_0}^z,$$
(84)
$$s_𝐤^x=\sqrt{\frac{s}{2}}\left(a_𝐤+a_𝐤^{}\frac{(a^2a^{})_𝐤}{2s}\right),s_𝐤^y=i\sqrt{\frac{s}{2}}\left(a_𝐤a_𝐤^{}\frac{(a^2a^{})_𝐤}{2s}\right),s_𝐤^z=s(a^{}a)_𝐤,$$
(85)
where $`z`$ axis is parallel to the magnetization of sublattices, $`\widehat{x},\widehat{y},\widehat{z}`$ denote unit vectors directed along corresponding axes, $`𝐤_0=(\pi ,0,\pi )`$ is the antiferromagnetic vector and $`s`$ is the spin value. Substitution of Eqs. (84) and (85) to Eq. (83) leads to the following expression for the Hamiltonian: $`H=E_0+_{i=2}^6H_i`$, where $`E_0`$ is the ground state energy and $`H_i`$ denote terms containing products of $`i`$ operators $`a`$ and $`a^{}`$. We consider in this paper the spin-wave approximation, i.e., we restrict ourself by the bilinear part of the Hamiltonian which has the form
$$H_2=\underset{𝐤}{}\left[E_𝐤a_𝐤^{}a_𝐤+\frac{B_𝐤}{2}\left(a_𝐤^{}a_𝐤^{}+a_𝐤a_𝐤\right)\right],$$
(86)
where $`E_𝐤=sJ_0`$, $`B_𝐤=sJ_𝐤`$ and $`J_𝐤=2J(\mathrm{cos}k_x+\mathrm{cos}k_z)`$. As is seen from Eqs. (82), (84) and (85), the only nonzero components of the spin Green’s function are $`xx`$ and $`yy`$ and tensor $`d_{\mu \nu }`$ has the form:
$$d_{\mu \nu }=\{\begin{array}{cc}\delta _{\mu \nu },\hfill & \text{ if }\mu ,\nu =x,y,\hfill \\ 0,\hfill & \text{ if }\mu =z\text{ or }\nu =z.\hfill \end{array}$$
(87)
The corresponding components of $`\mathrm{\Delta }_{\mu \nu }(\omega )`$ can be derived using Green’s functions $`g(\omega ,𝐤)=a_𝐤,a_𝐤^{}_\omega `$, $`f(\omega ,𝐤)=a_𝐤,a_𝐤_\omega `$, $`\overline{g}(\omega ,𝐤)=a_𝐤^{},a_𝐤_\omega =g^{}(\omega ,𝐤)`$ and $`f^{}(\omega ,𝐤)=a_𝐤^{},a_𝐤^{}_\omega =f^{}(\omega ,𝐤)`$. For two of them we have the Dyson equation:
$$\begin{array}{c}g(\omega ,𝐤)=g^{(0)}(\omega ,𝐤)+g^{(0)}(\omega ,𝐤)B_𝐤f^{}(\omega ,𝐤),\hfill \\ f^{}(\omega ,𝐤)=\overline{g}^{(0)}(\omega ,𝐤)B_𝐤g(\omega ,𝐤),\hfill \end{array}$$
(88)
where $`g^{(0)}(\omega ,𝐤)=(\omega E_𝐤+i\delta )^1`$ is the bare Green’s function. Solving Eq. (88) one obtains:
$$g(\omega ,𝐤)=\frac{\omega +sJ_\mathrm{𝟎}}{(\omega +i\delta )^2ϵ_𝐤^2},f(\omega ,𝐤)=\frac{sJ_𝐤}{(\omega +i\delta )^2ϵ_𝐤^2},$$
(89)
where $`ϵ_𝐤=s\sqrt{J_0^2J_𝐤^2}`$ is the spin-wave energy. As a result of direct calculations we have:
$$\mathrm{Im}\mathrm{\Delta }_{\mu \nu }(\omega )=d_{\mu \nu }\frac{s^2}{4\pi }𝑑𝐤\left[\{1+\mathrm{cos}(\mathrm{𝐤𝐑}_{12})\}(J_\mathrm{𝟎}J_𝐤)+\{1+\mathrm{cos}((𝐤+𝐤_0)𝐑_{12})\}(J_\mathrm{𝟎}+J_𝐤)\right]\frac{1}{ϵ_𝐤}\left[\delta (\omega ϵ_𝐤)\delta (\omega +ϵ_𝐤)\right],$$
(90)
where the lattice constant is taken to be equal to unity and the integral is over the magnetic Brillouin zone. If $`|\omega |sJ`$ we have $`J_𝐤J_\mathrm{𝟎}(1k^2/4)`$, $`ϵ_𝐤=ck=\sqrt{8}sJk`$ and $`\mathrm{cos}(\mathrm{𝐤𝐑}_{12})1(\mathrm{𝐤𝐑}_{12})^2/2`$. Notice that $`(𝐤_0𝐑_{12})=\pi mod2\pi `$ if the impurity is coupled to spins from different sublattices and both terms in the first square brackets in Eq. (90) are proportional to $`k^2`$. Then integration in Eq. (90) can be easily carried out if one takes advantage of the approximation for magnons similar to Debye one for phonons: the spectrum is assumed to be linear, $`ϵ_𝐤=ck`$, up to cut-off momentum $`k_\mathrm{\Theta }`$ defined from the equation $`2N=V(2\pi )^1_0^{k_\mathrm{\Theta }}𝑑kk`$, where $`V`$ is the area of the lattice. As a result we lead to expression (6) for $`\mathrm{Im}\mathrm{\Delta }_{\mu \nu }(\omega )`$, where
$$\mathrm{\Theta }=ck_\mathrm{\Theta }=8\sqrt{\pi }sJ,A=\frac{2\pi }{J}.$$
(91)
The factor $`A`$ should be multiplied by 2 if the defect is coupled to four host spins (two by two from each sublattice).
It is well known that there is no long range order in Heisenberg 2D AF at $`T>0`$. Mermin and Wagner (1966) Nevertheless it is shown theoretically Kopietz (1990); Chakravarty et al. (1989); Tyč and Halperin (1990) and confirmed experimentally Thurber et al. (1997) that the spin waves are well defined in paramagnetic phase of 2D AF if their wavelength is much smaller than the correlation length $`\xi \mathrm{exp}(\mathrm{const}/T)`$. Thus, the above result for $`\mathrm{Im}\mathrm{\Delta }_{\mu \nu }(\omega )`$ is valid when $`|\omega |Ja/\xi `$, where $`a`$ is the lattice spacing.
It is easy to conclude that if a small interplane interaction of the value of $`\eta J`$ is taken into account the above result for $`\mathrm{Im}\mathrm{\Delta }_{\mu \nu }(\omega )`$ is valid when $`|\omega |\eta `$ (see discussion in Sec. II.1). At the same time $`\mathrm{Im}\mathrm{\Delta }_{\mu \nu }(\omega )`$ has another $`\omega `$-dependence if $`|\omega |\eta `$.
Finally, we note that when the impurity is coupled to one host spin we have $`\mathrm{\Delta }_{\mu \nu }(\omega )=N^1_𝐤s_𝐤^\mu s_𝐤^\nu _\omega `$ instead of Eq. (82). Comparing this equation with (82) and (90) one infers that the spectral function is proportional to a constant in this case.
## Appendix B Matrix structure of pseudofermion Green’s function and the vertex for 2D AF
In this appendix we discuss the matrix structure of the dressed pseudofermion Green’s functions $`G_{mm^{}}(x)`$ and the pseudofermion vertex $`\mathrm{\Gamma }_{Pmm^{}}(x+\omega ,x)`$ for 2D AF. Some lower-order diagrams for the self-energy $`\mathrm{\Sigma }_{mm^{}}(\omega )`$ and for $`\mathrm{\Gamma }_{Pmm^{}}`$ are shown in Figs. 3 and 4, respectively. Let us discuss firstly the self-energy. Its matrix structure is determined by the corresponding products of operators $`S^\mu `$ and tensors $`d_{\mu \nu }`$. For instance, this product for the second diagram in Fig. 3 has the form $`_{\mu \mu ^{}\nu \nu ^{}}S^\mu S^\nu S^\mu ^{}S^\nu ^{}d_{\nu \nu ^{}}d_{\mu \mu ^{}}`$. One can make sure that such combinations are proportional to the unit matrix using the following evident representation of an arbitrary matrix $`A`$ of the size $`2\times 2`$ via Pauli matrices:
$$A=a_0+(𝝈𝐚),$$
(92)
where $`a_0=\mathrm{Tr}(A)/2`$ and $`𝐚=\mathrm{Tr}(𝝈A)/2`$. It is seen from the view of the tensor $`d_{\mu \nu }`$ given by Eq. (87) that the combinations of $`S^\mu `$ contain products of even number of matrices $`\sigma _x`$ and $`\sigma _y`$. According to (92) such combinations are proportional to the unit matrix.
The similar consideration can be carried out for the vertex $`\mathrm{\Gamma }_{Pmm^{}}(x+\omega ,x)`$, where $`P`$ is one of Pauli matrices. The vertex matrix structure is determined by the products of operators $`S^\mu `$, tensors $`d_{\mu \nu }`$ and $`P`$. For example, the corresponding product for the fourth diagram in Fig. 4 has the form $`_{\mu \mu ^{}\nu \nu ^{}}S^\mu S^\nu PS^\mu ^{}S^\nu ^{}d_{\nu \nu ^{}}d_{\mu \mu ^{}}`$. It can be easily shown using (92) that such combinations are proportional to $`P`$.
## Appendix C Calculation of the impurity susceptibility
We present in this appendix some details of the impurity dynamical susceptibility calculation. We use for this the general expression (19) and Eqs. (26), (33) and (37) for the Green’s function and the branches of the vertex. The expression for $`\chi _P(\omega )`$ is derived up to the order of $`f^2`$. In this order the interaction does not change the average number of pseudofermions $`𝒩`$ given by Eq. (15), i.e. $`𝒩=2e^{\lambda /T}`$. To show this let us express $`𝒩`$ as an integral of the Green’s function: Abrikosov et al. (1963)
$$𝒩=\frac{1}{\pi }\underset{m}{}_{\mathrm{}}^{\mathrm{}}𝑑xn(x)\mathrm{Im}G_{mm}(x)=\frac{2e^{\lambda /T}}{\pi }_{\mathrm{}}^{\mathrm{}}𝑑xe^{x/T}\mathrm{Im}G(x)$$
(93)
where $`n(x)=(e^{x/T}+1)^1`$ is the Fermi function and $`G(x)`$ is given by Eq. (26). We make a shift by $`\lambda `$ in the last part of Eq. (93) and replace $`n(x+\lambda )`$ by $`e^{(x+\lambda )/T}`$ as it was done in Eq. (19) for $`\chi _P(\omega )`$. Because $`\gamma (x)`$ and $`Z(x)`$ are exponentially small at negative $`x`$ if $`|x|T`$ (see Eq. (31)) the integrand in Eq. (93) does not increase exponentially as $`x\mathrm{}`$. It is easy to make sure that the terms proportional to $`f^2`$ cancel each other in Eq. (93).
According to Eqs. (19), (33) and (40) the dynamical susceptibility can be represented as a sum of three components. The first one, $`\chi _1(\omega )`$, originates from Eq. (19) as a result of replacement of the vertex by unity. The second, $`\chi _2(\omega )`$, appears from $`f^2`$-terms in Eqs. (33) and (40). The third, $`\chi _3(\omega )`$, is a result of replacement of the vertex by the third term from Eq. (40).
The expression for $`\chi _1(\omega )`$ can be brought to the form
$$\chi _1(\omega )=\frac{\overline{P^2}}{2\pi }_{\mathrm{}}^{\mathrm{}}𝑑xe^{x/T}[G(x+\omega )+G^{}(x\omega )]\mathrm{Im}G(x).$$
(94)
The integrand in Eq. (94) does not increase exponentially as $`x\mathrm{}`$ because $`\gamma (x)`$ and $`Z(x)`$ obey the property (31). Using Eq. (26) for Green’s functions it is convenient to represent Eq. (94) in the following form:
$`\chi _1(\omega )`$ $`=`$ $`[J_0(\omega )+J_1(\omega )+J_2(\omega )]+[J_0^{}(\omega )+J_1^{}(\omega )+J_2^{}(\omega )],`$ (95)
$`J_0(\omega )`$ $`=`$ $`{\displaystyle \frac{\overline{P^2}}{2\pi }}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑xe^{x/T}{\displaystyle \frac{1}{x+\omega +i\mathrm{\Gamma }_0}}{\displaystyle \frac{\gamma (x)}{x^2+\mathrm{\Gamma }_0^2}},`$ (96)
$`J_1(\omega )`$ $`=`$ $`{\displaystyle \frac{\overline{P^2}}{4\pi i}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑xe^{x/T}{\displaystyle \frac{1}{x+\omega +i\mathrm{\Gamma }_0}}\left[{\displaystyle \frac{i\gamma (x)[Z(x)+Z^{}(x)]+x[Z(x)Z^{}(x)]}{x^2+\mathrm{\Gamma }_0^2}}\right],`$ (97)
$`J_2(\omega )`$ $`=`$ $`{\displaystyle \frac{\overline{P^2}}{2\pi }}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑xe^{x/T}{\displaystyle \frac{Z(x+\omega )}{x+\omega +i\mathrm{\Gamma }_0}}{\displaystyle \frac{\gamma (x)}{x^2+\mathrm{\Gamma }_0^2}},`$ (98)
where in all denominators we replace $`\gamma (x)`$ and $`\gamma (x+\omega )`$ by $`\mathrm{\Gamma }_0`$, their values at $`|x|T`$ and $`|x+\omega |T`$, respectively (see Eq. (30)). It can be done because $`|\gamma (x)||x|`$ at $`|x|T`$.
The integration in Eq. (96) can be easily carried out if one notes that the main contribution arises from the area of $`x\omega ,\mathrm{\Gamma }`$, where $`e^{x/T}1x/T`$ (it will be clear soon that the second term in this expansion is essential). As a result we have:
$$J_0(\omega )=\frac{\overline{P^2}}{2}\frac{1}{\omega +2i\mathrm{\Gamma }_0}\left(1\frac{i\mathrm{\Gamma }_0}{T}\right).$$
(99)
To take the integral in Eq. (97) for $`J_1(\omega )`$ we consider a contour integral with the same integrand. The contour is presented in Fig. 6. It consists of four lines which are parallel to the real axis. They pass through points $`x=0`$, $`x=i\mathrm{\Gamma }_0i\delta `$, $`x=i\mathrm{\Gamma }_0+i\delta `$ and $`x=i\pi T`$. It can be shown using definitions of $`\gamma (x)`$ and $`Z(x)`$ that the integrand is an analytical function inside the contour. Note, the contour envelops the cut of $`Z(x)^{}`$ passing through the point $`x=i\mathrm{\Gamma }_0`$ along the real axis. As a result we have:
$`J_1(\omega )`$ $`=`$ $`{\displaystyle \frac{\overline{P^2}}{2\pi }}{\displaystyle \frac{f^2R_\mathrm{\Sigma }}{\pi \mathrm{\Theta }}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑x𝑑x_1e^{x/T}|x_1|N(x_1)\mathrm{\Lambda }(x_1){\displaystyle \frac{\gamma (x+i\pi T)(x_1+x+i\pi T)+\gamma (x_1+x+i\pi T)(x+i\pi T)}{(x+i\pi T)^3(x_1+x+i\pi T)^2}}`$ (100)
$`+{\displaystyle \frac{\overline{P^2}}{4\pi i}}{\displaystyle \frac{f^2R_\mathrm{\Sigma }}{\pi \mathrm{\Theta }}}e^{i\mathrm{\Gamma }_0/T}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑x𝑑x_1{\displaystyle \frac{e^{x/T}|x_1|N(x_1)\mathrm{\Lambda }(x_1)}{x+\omega +2i\mathrm{\Gamma }_0}}\left[{\displaystyle \frac{1}{x_1+xi\delta }}{\displaystyle \frac{1}{xi\delta }}{\displaystyle \frac{1}{x_1+x+i\delta }}{\displaystyle \frac{1}{x+i\delta }}\right],`$
where the first term is the result of integration over the line of the contour passing through $`x=i\pi T`$ and the second term describes the sum of integration over lines passing through $`x=i\mathrm{\Gamma }_0i\delta `$ and $`x=i\mathrm{\Gamma }_0+i\delta `$. It is seen from Eq. (100) that the first term is of the order of $`f^6`$ and can be discarded. The second term in Eq. (100) can be simply calculated using the equation $`(x_1+xi\delta )^1(xi\delta )^1(x_1+x+i\delta )^1(x+i\delta )^1=2\pi i[\delta (x_1+x)(xi\delta )^1+\delta (x)(x_1+i\delta )^1]`$. As a result we have:
$$J_1(\omega )=\frac{\overline{P^2}}{2}\frac{e^{i\mathrm{\Gamma }_0/T}}{\omega +2i\mathrm{\Gamma }_0}\frac{f^2R_\mathrm{\Sigma }}{\pi \mathrm{\Theta }}_{\mathrm{}}^{\mathrm{}}𝑑x\frac{|x|N(x)\mathrm{\Lambda }(x)}{x+\omega +2i\mathrm{\Gamma }_0}.$$
(101)
One can carry out the integration in Eq. (98) for $`J_2(\omega )`$ in the similar way. As a result we obtain that $`J_2(\omega )=J_1(\omega )`$ and $`\chi _1(\omega )`$ has the form:
$$\chi _1(\omega )=\frac{\overline{P^2}}{2}\left(\frac{2i\mathrm{\Gamma }_0}{T(\omega +2i\mathrm{\Gamma }_0)}\left[12\frac{f^2R_\mathrm{\Sigma }}{\pi \mathrm{\Theta }}_{\mathrm{}}^{\mathrm{}}𝑑x\frac{|x|N(x)\mathrm{\Lambda }(x)}{x+2i\mathrm{\Gamma }_0}\right]+2R_\mathrm{\Sigma }\frac{f^2}{\pi \mathrm{\Theta }}_{\mathrm{}}^{\mathrm{}}𝑑x\frac{\mathrm{sgn}(x)\mathrm{\Lambda }(x)}{x+\omega +2i\mathrm{\Gamma }_0}\right),$$
(102)
where we omit $`\omega `$ in the denominator of the integrand in the first term.
The quantity $`\chi _2(\omega )`$ can be expressed as follows:
$`\chi _2(\omega )`$ $`=`$ $`J(\omega )+J^{}(\omega ),`$ (103)
$`J(\omega )`$ $`=`$ $`{\displaystyle \frac{\overline{P^2}}{2\pi }}R_1{\displaystyle \frac{f^2}{\pi \mathrm{\Theta }}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑x𝑑x_1e^{x/T}|x_1|x_1N(x_1)\mathrm{\Lambda }(x_1)G(x_1+x+\omega )G(x+\omega )\mathrm{Im}\{G(x)G(x_1+x)\}.`$ (104)
The integral in Eq. (104) can be taken similar to those of $`J_1(\omega )`$ and $`J_2(\omega )`$, the result being
$$\chi _2(\omega )=\frac{\overline{P^2}}{2}\left[\frac{2i\mathrm{\Gamma }_0}{T(\omega +2i\mathrm{\Gamma }_0)}2R_1\frac{f^2}{\pi \mathrm{\Theta }}_{\mathrm{}}^{\mathrm{}}𝑑x\frac{|x|N(x)\mathrm{\Lambda }(x)}{x+2i\mathrm{\Gamma }_0}2R_1\frac{f^2}{\pi \mathrm{\Theta }}_{\mathrm{}}^{\mathrm{}}𝑑x\frac{\mathrm{sgn}(x)\mathrm{\Lambda }(x)}{x+\omega +2i\mathrm{\Gamma }_0}\right],$$
(105)
where we omit $`\omega `$ in the denominator of the integrand in the first term.
The expression for $`\chi _3(\omega )`$ can be brought to the form:
$$\chi _3(\omega )=\frac{\overline{P^2}}{2}\frac{\omega }{T(\omega +2i\mathrm{\Gamma })}Z_\mathrm{\Gamma }_{\mathrm{}}^{\mathrm{}}𝑑xe^{x/T}G(x+\omega )G^{}(x)K(x).$$
(106)
The area near poles of the Green’s functions is essential in this integral. Therefore, we can replace $`e^{x/T}`$ by unity and $`K(x)`$ by $`K(0)=(\mathrm{\Gamma }_0\mathrm{\Gamma })/\pi `$. As a result we have:
$$\chi _3(\omega )=\frac{\overline{P^2}}{2}\left[\frac{2i\mathrm{\Gamma }}{T(\omega +2i\mathrm{\Gamma })}\frac{2i\mathrm{\Gamma }_0}{T(\omega +2i\mathrm{\Gamma }_0)}\right]\left[12R_\mathrm{\Sigma }\frac{f^2}{\pi \mathrm{\Theta }}_{\mathrm{}}^{\mathrm{}}𝑑x\frac{|x|N(x)\mathrm{\Lambda }(x)}{x+2i\mathrm{\Gamma }_0}+R_1\frac{f^2}{\pi \mathrm{\Theta }}_{\mathrm{}}^{\mathrm{}}𝑑x\frac{|x|N(x)\mathrm{\Lambda }(x)}{x+2i\mathrm{\Gamma }_0}\right].$$
(107)
Summing Eqs. (102), (105) and (107) we lead to Eq. (44) for the impurity susceptibility.
## Appendix D Green’s functions of 2D AF with impurities
We derive in this appendix Green’s functions of operators $`a`$ and $`a^{}`$ considered in Appendix A in the case of 2D AF with impurities. To investigate the influence of the impurities we have to take into account the corresponding interaction (4) in Dyson equations. It is assumed that $`N,N_i\mathrm{}`$ so as $`N_i/N=n=\mathrm{const}`$, where $`N`$ and $`N_i`$ are the number of spins in the lattice and the number of impurities, respectively. Within the linear spin-wave approximation the equations for $`g`$ and $`f^{}`$ have the form
$`g(\omega ,𝐤)`$ $`=`$ $`g^{(0)}(\omega ,𝐤)g^2ns[1+\mathrm{cos}(\mathrm{𝐤𝐑}_{12})][f^{(0)}(\omega ,𝐤)+g^{(0)}(\omega ,𝐤)]\chi _x(\omega )[g(\omega ,𝐤)+f^{}(\omega ,𝐤)]`$
$`+g^2ns[1\mathrm{cos}(\mathrm{𝐤𝐑}_{12})][f^{(0)}(\omega ,𝐤)g^{(0)}(\omega ,𝐤)]\chi _y(\omega )[g(\omega ,𝐤)f^{}(\omega ,𝐤)],`$
$`f^{}(\omega ,𝐤)`$ $`=`$ $`f^{(0)}(\omega ,𝐤)g^2ns[1+\mathrm{cos}(\mathrm{𝐤𝐑}_{12})][\overline{g}^{(0)}(\omega ,𝐤)+f^{(0)}(\omega ,𝐤)]\chi _x(\omega )[g(\omega ,𝐤)+f^{}(\omega ,𝐤)]`$ (108)
$`+g^2ns[1\mathrm{cos}(\mathrm{𝐤𝐑}_{12})][\overline{g}^{(0)}(\omega ,𝐤)f^{(0)}(\omega ,𝐤)]\chi _y(\omega )[g(\omega ,𝐤)f^{}(\omega ,𝐤)],`$
where superscript $`(0)`$ denotes Green’s functions without impurities which are given by Eq. (89) and $`\chi _x(\omega )`$ and $`\chi _y(\omega )`$ are the impurity susceptibility given by Eq. (49) with $`P=S^x`$ and $`S^y`$, respectively. Equations (D) can be easily solved with the result
$`g(\omega ,𝐤)`$ $`=`$ $`\overline{g}(\omega ,𝐤)^{}={\displaystyle \frac{sJ_0+\omega 2g^2ns\chi _{}(\omega )}{𝒟(\omega ,𝐤)}},`$
$`f^{}(\omega ,𝐤)`$ $`=`$ $`f(\omega ,𝐤)^{}={\displaystyle \frac{sJ_𝐤+2g^2ns\chi _{}(\omega )\mathrm{cos}(\mathrm{𝐤𝐑}_{12})}{𝒟(\omega ,𝐤)}},`$ (109)
where $`\chi _{}(\omega )=\chi _x(\omega )=\chi _y(\omega )`$ and the denominator $`𝒟(\omega ,𝐤)`$ is given by Eq. (65). |
warning/0507/nucl-th0507027.html | ar5iv | text | # Baseline Cold Matter Effects on 𝐽/𝜓 Production in 𝐴𝐴 Collisions at RHIC
## Acknowledgments
We thank Mike Leitch for suggesting this work and Olivier Drapier for discussions. This work was supported in part by the Director, Office of Energy Research, Division of Nuclear Physics of the Office of High Energy and Nuclear Physics of the U. S. Department of Energy under Contract Number DE-AC02-05CH11231. |
warning/0507/hep-ex0507053.html | ar5iv | text | # 2005 International Linear Collider Workshop – Stanford, U.S.A. Selectron Mass Reconstruction and the Resolution of the Linear Collider Detector
## I INTRODUCTION
For some time, the Santa Cruz Linear Collider R&D group has been interested in exploring the requirements that measuring forward selectron production places on the Linear Collider Detector. With right-handed selectron and lightest neutralino masses of 143 and 95 GeV, respectively, Snowmass Point SPS1a produces a substantial forward ($`|\mathrm{cos}\theta |0.8|`$) population of electrons from selectron decay. Over the past two years, the Santa Cruz group has developed techniques to isolate the forward selectron-decay signal from Standard-Model backgrounds, as well as techniques to extract the selectron mass from the observed selectron-decay electron spectrum. This paper reports on the results of those studies.
## II SIGNAL SELECTION
Backgrounds to the selectron-decay electron signal come from two primary sources: four-electron production for which two of the final-state fermions are close in angle to the beam trajectory, and thus escape detection, and the $`Z`$-boson fusion process $`e^+e^{}e^+e^{}\nu \overline{\nu }`$. During the course of searches for supersymmetry at LEP, several cuts were developed to isolate potential SUSY signals from Standard Model backgrounds.
* Fiducial Region Cut: Exactly one final-state positron and one final-state electron pair must be detected in the fiducial region. For this study, the fiducial region lies above 5 GeV in momentum, and within $`|\mathrm{cos}\theta |0.8`$ for central-region studies, or within $`|\mathrm{cos}\theta |0.994`$ for studies including the forward region.
* Tagging Cut: No observable electron or positron in the low-angle ‘tagging calorimetry’ (with assumed coverage of 20mrad $`<\theta <`$ 110mrad).
* Transverse Momentum Cut: Cuts events for which the vector sum of the observed $`e^+e^{}`$ pair transverse momenta is less than 500 Gev $`\times \mathrm{sin}`$(20 mrad).
For a signal region of $`|\mathrm{cos}\theta |0.8`$, these cuts completely eliminate the four-electron background up to diagrams with additional radiation, and sufficiently select against the $`ee\nu \nu `$ process. However, as one attempts to reconstruct the selectron decay signal beyond $`|\mathrm{cos}\theta |=0.8`$, both the radiative four-electron background and the $`ee\nu \nu `$ backgrounds approach poles in their respective production cross sections, and begin to dominate the selectron signal. To suppress these backgrounds to a degree appropriate for use of the forward selectron decay signal, two additional cuts were applied.
* Photon Cut: Remove event if there is an observed photon of 20 GeV or greater in either the fiducial or tagging region.
* High-Momentum Cut: Remove even if the vector sum of the $`e^+e^{}`$ pair momenta has a magnitude of greater than 225 GeV.
With the application of these cuts to events produced at $`E_{cm}=1`$ TeV with 80% right-handed-polarized electron beam and unpolarized positrons, it was found that Standard Model processes provided a negligible background to the right-handed selectron signal. In addition, the application of these selection criteria did not significantly alter the shape of the selectron-decay electron spectrum in either the central or forward region. Nevertheless, in the analysis that follows all of the selection cuts will be applied to the selectron signal; however, Standard Model backgrounds will be ignored from this point on. Finally, running with an 80% right-polarized electron beam, electrons from the decay of the more massive left-handed selectron provide a diffuse background to the right-handed selectron decay electron spectrum. This contribution was also ignored in the right-handed selectron analysis.
## III SIMULATION OF THE SELECTRON SIGNAL
The selectron signal was generated at $`E_{cm}=1`$ TeV using the Snowmass point SPS1a sps parameters $`m_0=100`$ GeV, $`m_{1/2}=250`$ GeV, $`A_0=100`$, $`\mathrm{tan}\beta =10`$, and sgn($`\mu `$) = $`+`$. The SPS1a specifications were implemented within the ISAJET isajet package, and included the effects of initial state radiation, beamstrahlung ($`Y=0.29`$), and three different values of the fractional beam energy spread (assuming a gaussian distribution): 1.0%, 0.16%, and 0.0%.
For this point in SUSY parameter space, the endpoint of the selectron-decay electron energy distribution lies above 270 GeV, making the precise measurement of the endpoint energy by the tracking system somewhat challenging. Nonetheless, the projected $`\frac{1}{2}\%`$ resolution of the tracking system at this energy is somewhat better than that expected for the electromagnetic calorimetry.
The $`\mathrm{cos}\theta `$ distribution of final-state electrons and positrons from the decay of right-handed selectrons is peaked towards $`|\mathrm{cos}\theta |=1.0`$, due to the sizeable t-channel contribution admitted by the relatively light selectron and neutralino. Approximately half of the signal lies in the forward region, beyond $`|\mathrm{cos}\theta |=0.8`$.
## IV DETERMINATION OF THE RIGHT-HANDED SELECTRON MASS
To focus on the accuracy of the selectron mass reconstruction, it was assumed for now that the lightest neutralino mass was known precisely from other measurements. The Santa Cruz group plans to relax this assumption in further studies.
The selectron mass was determined by finding the best fit of the selectron energy spectrum to that of a series of ‘template’ distributions generated with slightly varying right-handed selectron masses, but with all other aspects of the signal generation, including beam energy spread and detector smearing effects, the same as for the data simulation. Templates were produced for 15 right-handed selectron masses in a range of approximately $`\pm 1`$ GeV about the nominal SPS1a mass of 143.11 GeV.
A total of 120 independent data sets, each corresponding to an integrated luminosity of 115 fb<sup>-1</sup>, were generated at the nominal SPS1a mass. For each of these data sets, a $`\chi ^2`$ was formed against each of the templates points, according to
$$\chi ^2=\underset{i}{}\frac{(wn_iwm_i)^2}{(wn_im_i)^2}$$
(1)
where the sum is over energy bins in the selectron-decay electron spectrum, $`n_i`$ and $`m_i`$ are the data and template bin contents, respectively, and
$$w=\frac{\underset{i}{}n_i}{_im_i}$$
(2)
is the relative sample-size weighting factor. To minimize the statistical contributions of the template files, they were generated with an integrated luminosity of approximately 1000 fb<sup>-1</sup> each.
For each data set, the $`\chi ^2`$ contour was fit with a quartic polynomial, which was then minimized to find the best-fit selectron mass. This procedure was repeated for the 120 data sets, and the mean and root-mean-square deviation was calculated from the distribution of best-fit right-handed selectron masses.
In forming the $`\chi ^2`$ for each of the 120 independent data sets, it was found that the inclusion of energy-spectrum bins not sufficiently close to the upper and lower endpoints to provide useful information on the endpoint location introduced unacceptable scatter in the contour. Thus, for each scenario that was studied (beam energy spread, detector model, polar angle reach), only bins near the upper and lower endpoint were used in forming the $`\chi ^2`$. The exact energy ranges used in the $`\chi ^2`$ calculation depended on the scenario under study, and was determined in a data-driven manner by examining the spectrum in the region of the endpoints in the generated data samples. Table I shows the regions used in the forming the $`\chi ^2`$ for the different scenarios.
### IV.1 Scenarios Explored
To explore the effects of detector resolution, data sets and mass templates were generated assuming perfect detector resolution as well as that of the SDMAR01 detector. Detector resolution effects were incorporated via gaussian smearing based on error matrices from Billior-based tracking error calculations provided by the LCDTRK program lcdtrk . Two different ranges in $`|\mathrm{cos}\theta |`$ were explored: between 0.0 and 0.8 and between 0.0 and 0.994. Finally, three different values of the beam energy spread were explored: $`\pm 1.0\%`$ (to allow benchmark comparisons to previous studies uriel ; yang ), $`\pm 0.16\%`$ (approximately that expected for the superconducting RF design), and, for comparison, 0.0%.
## V RESULTS
Figure 1 shows the root-mean-square deviation observed for the 120 independent data sets for the different scenarios that were explored in the study (in all cases, the mean value of the fit mass was reasonably consistent with the input vale of the right-handed selectron mass of 143.1 GeV).
For a perfect detector (no measurement uncertainty), the inclusion of the forward region provides substantial improvement in the selectron mass measurement, independent of beam energy spread. In fact, the improvement is better than one would expect from the factor-of-two increase in the number of signal electrons that is associated with the inclusion of the forward region. A study of the two-dimensional frequency distribution of signal electrons as a function of energy and the cosine of the polar angle (Figure 2) reveals that the spectrum has a greater contribution at higher energy, where the sensitivity to the selectron mass is greatest, at high values of $`\mathrm{cos}\theta `$. Thus, in this case, most of the information on the selectron mass (and on slepton masses in general) is in the forward region.
For large beam energy spread, it is seen that the sensitivity to the selectron mass has little dependence on the detector resolution. For this case, the SiD design seems to be adequate to take advantage of the physics capabilities of the accelerator. However, for smaller beam energy spread, this is not the case. For the energy spread anticipated for the selected machine design, substantial improvement in the selectron mass determination can be achieved by improving the detector resolution, particularly in the forward region.
## VI CONCLUSIONS
To examine the issue of measuring selectron masses in the forward region of the Linear Collider Detector, we have simulated SPS1a selectron production at $`E_{cm}=1`$ TeV. By developing two new selection criteria, we have demonstrated that the selectron signal can be separated from Standard Model backgrounds through the entire forward tracking region $`|\mathrm{cos}\theta |<0.994`$.
Due to the stiffening of the selectron-decay electron spectrum at higher values of $`|\mathrm{cos}\theta |`$, for a light selectron most of the information on selectron mass (and slepton mass in general) comes from the forward region. For large beam energy spread ($`\pm 1.0\%`$), we find that the SiD detector design is adequate to exploit the potential of the Linear Collider; however, for the expected beam energy spread of the chosen cold RF technology, gains in sensitivity to the selectron mass can be made by further improvements in momentum resolution, in both the central and forward regions.
###### Acknowledgements.
Although they appear as authors on this paper, one of us (Schumm) would like to acknowledge the dedication and creativity of the other authors, who performed these studies as undergraduate physics majors at the University of California at Santa Cruz. |
warning/0507/cs0507033.html | ar5iv | text | # Multiresolution Kernels
## 1 Introduction
There is strong evidence that kernel methods can deliver state-of-the-art performance on most classification tasks when the input data lies in a vector space. Arguably, two factors contribute to this success. First, the good ability of kernel algorithms, such as the SVM, to generalize and provide a sparse formulation for the underlying learning problem; Second, the capacity of nonlinear kernels, such as the polynomial and RBF kernels, to quantify meaningful similarities between vectors, notably non-linear correlations between their components. Using kernel machines with non-vectorial data (e.g., in bioinformatics, pattern recognition or signal processing tasks) requires more arbitrary choices, both to represent the objects and to chose suitable kernels on those representations. The challenge of using kernel methods on real-world data has thus recently fostered many proposals for kernels on complex objects, notably for strings, trees, images or graphs to cite a few.
A strategy often quoted as the generative approach to this problem takes advantage of a generative model, that is an adequate statistical model for the objects, to derive feature representations for the objects. In practice this often yields kernels to be used on the histograms of smaller components sampled in the objects, where the kernels take into account the geometry of the underlying model in their similarity measures . The previous approaches coupled with SVM’s combine both the advantages of using discriminative methods with generative ones, and produced convincing results on many tasks.
One of the drawbacks of such representations is however that they implicitly assume that each component has been generated independently and in a stationary way, where the empirical histogram of components is seen as a sample from an underlying stationary measure. While this viewpoint may translate into adequate properties for some learning tasks (such as translation or rotation invariance when using histograms of colors to manipulate images ), it might prove too restrictive and hence inadequate for other types of problems. Namely, tasks which involve a more subtle mix of detecting *both* conditional (with respect to the location of the components for instance) and global similarities between the objects. Such problems are likely to arise for instance in speech, language, time series or image processing. In the first three tasks, this consideration is notably treated by most state-of-the-art methods through dynamic programming algorithms capable of detecting and penalizing accordingly local matches between the objects. Using dynamic programming to produce a kernel yielded fruitful results in different applications , with the limitation that the kernels obtained in practice are not always positive definite, as reviewed in . Other kernels proposed for sequences directly incorporate a localization information into each component, augmenting considerably the size of the component space, and then introduce some smoothing (such as mismatches) to avoid representations that would be too sparse.
We propose in this work a different approach grounded on the generative approach previously quoted, managing however to combine both conditional and global similarities when comparing two objects. The motivation behind this approach is both intuitive and computational: intuitively, the global histogram of components, that is the simple bag of components representation of Figure 1, may seem inadequate if the components’ appearance seem to be clearly conditioned by some external events. This phenomenon can be taken into account by considering collections (indexed on the same set of events, to be defined) of nested bags or histograms to describe the object. Kernels that would only rely on these detailed resolutions might however miss the bigger picture that is provided by the global histogram. We propose a trade-off between both viewpoints through a combination that aims at giving a balanced account of both fine and coarse perspectives, hence the name of multiresolution kernels, which we introduce formally in Section 2. On the computational side, we show how such a theoretical framework can translate into an efficient factorization detailed in Section 3. We then provide experimental results in Section 4 on an image retrieval task which shows that the methodology improves the performance of kernel based state-of-the art techniques in this field.
## 2 Multiresolution Kernels
In most applications, complex objects can be represented as histograms of components, such as texts as bags of words or images and sequences as histograms of colors and letters. Through this representation, objects are cast as probability laws or measures on the space $`𝒳`$ of components, typically multinomials if $`𝒳`$ is finite , and compared as such through kernels on measures. An obvious drawback of this representation is that all contextual information on how the components have been sampled is lost, notably any general sense of position in the objects, but also more complex conditional information that may be induced from neighboring components, such as transitions or long range interactions.
In the case of images for instance, one may be tempted to consider not only the overall histogram of colors, but also more specialized histograms which may be relevant for the task. If some local color-overlapping in the images is an interesting or decisive feature of the learning problem, these specialized histograms may be generated arbitrarily following a grid, dividing for instance the image into 4 equal parts, and computing histograms for each corner before comparing them pairwise between two images (see Figure 2 for an illustration). If sequences are at stake, these may also be sliced into predefined regions to yield local histograms of letters. If the strings are on the contrary assumed to follow some Markovian behaviour (namely that the appearance of letters in the string is independent of their exact location but only depends on the few letters that precede them), an interesting index would translate into a set of contexts, typically a complete suffix dictionary as detailed in . While the two previous examples may seem opposed in the way the histograms are generated, both methodologies stress a particular class of events (location or transitions) that give an additional knowledge on how the components were sampled in the objects. Since both these two approaches, and possibly other ones, can be applied within the framework of this paper using a unified formalism, we present our methodology using a general notation for the index of events. Namely, we note $`𝒯`$ for an arbitrary set of conditioning events, assuming these events can be directly observed on the object itself, by contrast with the latent variables approach of . Considering still, following the generative approach, that an object can be mapped onto a probability measure $`\mu `$ on $`𝒳`$, we have that the realization of an event $`t\mathrm{}𝒯`$ can be interpreted under the light of a joint probability $`\mu (x,t)`$, with $`x\mathrm{}𝒳`$, factorized through Bayes’ law as $`\mu (x|t)\mu (t)`$ to yield the following decomposition of $`\mu `$ as
$$\mu =\underset{t\mathrm{}𝒯}{}\mu _t,$$
where each $`\mu _t\stackrel{def}{=}\mu (|t)\mu (t)`$ is an element of the set of sub-probability measures $`M_+^s(𝒳)`$, that is the set of positive measures $`\rho `$ on $`𝒳`$ such that their total mass $`\rho (𝒳)`$ denoted as $`|\rho |`$ is *less than* or equal to $`1`$. To take into account the information brought by the events in $`𝒯`$, objects can hence be represented as families of measures of $`M_+^s(𝒳)`$ indexed by $`𝒯`$, namely elements $`\mu `$ contained in $`M_𝒯(𝒳)\stackrel{def}{=}M_+^s(𝒳)^𝒯.`$
### 2.1 Local Similarities Between Measures Conditioned by Sets of Events
To compare two objects under the light of their respective decompositions as sub-probability measures $`\mu _t`$ and $`\mu _t^{}`$, we make use of an arbitrary positive definite kernel $`k`$ on $`M_+^s(𝒳)`$ to which we will refer to as the base kernel throughout the paper. For interpretation purposes only, we may assume in the following sections that $`k`$ can be written as $`e^{d^2}`$ where $`d`$ is an Euclidian distance in $`M_+^s(𝒳)`$. Note also that the kernel is defined not only on probability measures, but also on sub-probabilities. For two elements $`\mu ,\mu ^{}`$ of $`M_𝒯(𝒳)`$ and a given element $`t\mathrm{}𝒯`$, the kernel
$$k_t(\mu ,\mu ^{})\stackrel{def}{=}k(\mu _t,\mu _t^{})$$
measures the similarity of $`\mu `$ and $`\mu ^{}`$ by quantifying how similarly their components were generated conditionally to event $`t`$. For two different events $`s`$ and $`t`$ of $`𝒯`$, $`k_s`$ and $`k_t`$ can be associated through polynomial combinations with positive factors to result in new kernels, notably their sum $`k_s+k_t`$ or their product $`k_sk_t`$. This is particularly adequate if some complementarity is assumed between $`s`$ and $`t`$, so that their combination can provide new insights for a given learning task. If on the contrary the events are assumed to be similar, then they can be regarded as a unique event $`\{s\}\{t\}`$ and result in the kernel
$$k_{\{s\}\{t\}}(\mu ,\mu ^{})\stackrel{def}{=}k(\mu _s+\mu _t,\mu _s^{}+\mu _t^{}),$$
which will measure the similarity of $`m`$ and $`m^{}`$ when *either* $`s`$ or $`t`$ occurs. The previous formula can be extended to model kernels indexed on a set $`T𝒯`$ of similar events, through
$$k_T(m,m^{})\stackrel{def}{=}k(\mu _T,\mu _T^{}),\text{where }\mu _T\stackrel{def}{=}\underset{t\mathrm{}T}{}\mu _t\text{ and }\mu _T^{}\stackrel{def}{=}\underset{t\mathrm{}T}{}\mu _t^{}.$$
Note that this equivalent to defining a distance between elements $`\mu `$ and $`\mu ^{}`$ conditionned by $`T`$ as $`d_T^2(\mu ,\mu ^{})\stackrel{def}{=}d^2(\mu _T,\mu _T^{})`$.
### 2.2 Resolution Specific Kernels
Let $`P`$ be a finite partition of $`𝒯`$, that is a finite family $`P=(T_1,\mathrm{},T_n)`$ of sets of $`𝒯`$, such that $`T_iT_j=\mathrm{}`$ if $`1i<jn`$ and $`_{i=1}^nT_i=𝒯`$. We write $`𝒫(𝒯)`$ for the set of all partitions of $`𝒯`$. Consider now the kernel defined by a partition $`P`$ as
$$k_P(\mu ,\mu ^{})\stackrel{def}{=}\underset{i=1}{\overset{n}{}}k_{T_i}(\mu ,\mu ^{}).$$
(1)
The kernel $`k_P`$ quantifies the similarity between two objects by detecting their joint similarity under all possible events of $`𝒯`$, given an a priori similarity assumed on the events which is expressed as a partition of $`𝒯`$. Note that there is some arbitrary in this definition since, following the convolution kernels approach for instance, a simple multiplication of base kernels $`k_{T_i}`$ to define $`k_P`$ is used, rather than any other polynomial combination. More precisely, the multiplicative structure of Equation (1) quantifies how two objects are similar given a partition $`P`$ in a way that imposes for the objects to be similar according to all subsets $`T_i`$. If $`k`$ can be expressed as a function of a distance $`d`$, $`k_P`$ can be expressed as the exponential of
$$d_P^2(\mu ,\mu ^{})\stackrel{def}{=}\underset{i=1}{\overset{n}{}}d_{T_i}^2(\mu ,\mu ^{}),$$
a quantity which penalizes local differences between the decompositions of $`\mu `$ and $`\mu ^{}`$ over $`𝒯`$, as opposed to the coarsest approach where $`P=\{𝒯\}`$ and only $`d^2(\mu ,\mu ^{})`$ is considered.
As illustrated in Figure 2 in the case of images expressed as histograms indexed over locations, a partition of $`𝒯`$ reflects a given belief on how events should be associated to belong to the same set or dissociated to highlight interesting dissimilarities. Hence, all partitions contained in the set $`𝒫(𝒯)`$ of all possible partitions<sup>1</sup><sup>1</sup>1which is quite a big space, since if $`𝒯`$ is a finite set of cardinal $`r`$, the cardinal of the set of partitions is known as the Bell Number of order $`r`$ with $`B_r=\frac{1}{e}_{u=1}^{\mathrm{}}\frac{u^r}{u!}\underset{r\mathrm{}}{}e^{r\mathrm{ln}r}`$. are not likely to be equally meaningful given that some events may look more similar than others. If the index is based on location, one would naturally favor mergers between neighboring indexes. For contexts, a useful topology might also be derived by grouping contexts with similar suffixes.
Such meaningful partitions can be obtained in a general case if we assume the existence of a prior hierarchical information on the elements of $`𝒯`$, translated into a series
$$P_0=\{𝒯\},..,P_D=\{\{t\},t\mathrm{}𝒯\}$$
of partitions of $`𝒯`$, namely a hierarchy on $`𝒯`$. To provide a hierarchical content, the family $`(P_d)_{d=1}^D`$ is such that any subset present in a partition $`P_d`$ is included in a (unique by definition of a partition) subset included in the coarser partition $`P_{d1}`$, and further assume this inclusion to be strict. This is equivalent to stating that each set $`T`$ of a partition $`P_d`$ is divided in $`P_{d+1}`$ through a partition of $`T`$ which is not $`T`$ itself. We note this partition $`s(T)`$ and name its elements the siblings of $`T`$. Consider now the subset $`𝒫_D𝒫(𝒯)`$ of all partitions of $`𝒯`$ obtained by using only sets in
$$P_0^D\stackrel{def}{=}\underset{d=1}{\overset{D}{}}P_d,$$
namely $`𝒫_D\stackrel{def}{=}\{P\mathrm{}𝒫(𝒯)\text{ s.t. }T\mathrm{}P,T\mathrm{}P_0^D\}.`$. The set $`𝒫_D`$ contains both the coarsest and the finest resolutions, respectively $`P_0`$ and $`P_D`$, but also all variable resolutions for sets enumerated in $`P_0^D`$, as can be seen for instance in the third image of Figure 2.
### 2.3 Averaging Resolution Specific Kernels
Each partition $`P`$ contained in $`𝒫_D`$ provides a resolution to compare two objects, and generates consequently a very large family of kernels $`k_P`$ when $`P`$ spans $`𝒫_D`$. Some partitions are probably better suited for certain tasks than others, which may call for an efficient estimation of an optimal partition given a task. We take in this section a different direction by considering an averaging of such kernels based on a Bayesian prior on the set of partitions. In practice, this averaging favours objects which share similarities under a large collection of resolutions.
###### Definition 1.
Let $`𝒯`$ be an index set endowed with a hierarchy $`(P_d)_{d=0}^D`$, $`\pi `$ be a prior measure on the corresponding set of partitions $`𝒫_D`$ and $`k`$ a base kernel on $`M_+^s(𝒳)\times M_+^s(𝒳)`$. The multiresolution kernel $`k_\pi `$ on $`M_𝒯(𝒳)\times M_𝒯(𝒳)`$ is defined as
$$k_\pi (\mu ,\mu ^{})=\underset{P\mathrm{}𝒫_D}{}\pi (P)k_P(\mu ,\mu ^{}).$$
(2)
Note that in Equation (2), each resolution specific kernel contributes to the final kernel value and may be regarded as a weighted feature extractor.
## 3 Kernel Computation
This section aims at characterizing hierarchies $`(P_d)_{d=0}^D`$ and priors $`\pi `$ for which the computation of $`k_\pi `$ is both tractable and meaningful. We first propose a type of hierarchy generated by trees, which is then coupled with a branching process prior to fully specify $`\pi `$. These settings yield a computational time for expressing $`k_\pi `$ which is loosely upperbounded by $`D\times \mathrm{card}𝒯\times c(k)`$ where $`c(k)`$ is the time required to compute the base kernel.
### 3.1 Partitions Generated by Branching Processes
All partitions $`P`$ of $`𝒫_D`$ can be generated iteratively through the following rule, starting from the initial root partition $`P:=P_0=\{𝒯\}`$. For each set $`T`$ of $`P`$:
1. either leave the set as it is in $`P`$,
2. either replace it by its siblings enumerated in $`s(T)`$, and reapply this rule to each sibling unless they belong to the finest partition $`P_D`$.
By giving a probabilistic content to the previous rule through a binomial parameter (i.e. for each treated set assign probability $`1\epsilon `$ of applying rule 1 and probability $`\epsilon `$ of applying rule 2) a candidate prior for $`𝒫_D`$ can be derived, depending on the overall coarseness of the considered partition. For all elements $`T`$ of $`P_D`$ this binomial parameter is equal to $`0`$, whereas it can be individually defined for any element $`T`$ of the $`D1`$ coarsest partitions as $`\epsilon _T\mathrm{}[0,1]`$, yielding for a partition $`P\mathrm{}𝒫_D`$ the weight
$$\pi (P)=\underset{T\mathrm{}P}{}(1\epsilon _T)\underset{T\mathrm{}\stackrel{}{P}}{}(\epsilon _T),$$
where the set $`\stackrel{}{P}=\{T\mathrm{}P_0^D\text{ s.t. }V\mathrm{}P,VT\}`$ gathers all coarser sets belonging to coarser resolutions than $`P`$, and can be regarded as all ancestors in $`P_0^D`$ of sets enumerated in $`P`$.
### 3.2 Factorization
The prior proposed in Section 3.1 can be used to factorize the formula in (2), which is summarized in this theorem, using notations used in Definition 1
###### Theorem 1.
For two elements $`m,m^{}`$ of $`M_𝒯(𝒳)`$, define for $`T`$ spanning recursively $`P_D,P_{D1},\mathrm{},P_0`$ the quantity
$$K_T=(1\epsilon _T)k_T(\mu ,\mu ^{})+\epsilon _T\underset{U\mathrm{}s(T)}{}K_U.$$
Then $`k_\pi (\mu ,\mu ^{})=K_𝒯`$.
###### Proof.
The proof follows from the prior structure used for the tree generation, and can be found in either or . Figure 3 underlines the importance of incorporating to each node $`K_T`$ a weighted product of the kernels $`K_U`$ computed by its siblings. ∎
If the hierarchy of $`𝒯`$ is such that the cardinality of $`s(T)`$ is fixed to a constant $`\alpha `$ for any set $`T`$, typically $`\alpha =4`$ for images as seen in Figure 2, then the computation of $`k_\pi `$ is upperbounded by $`(\alpha ^{D+1}1)c(k)`$. This computational complexity may even become lower in cases where the histograms become sparse at fine resolutions, yielding complexities in linear time with respect to the size of the compared objects, quantified by the length of the sequences in for instance.
## 4 Experiments
We present in this section experiments inspired by the image retrieval task first considered in and also used in , although the images used here are not exactly the same. The dataset was also extracted from the Corel Stock database and includes 12 families of labelled images, each class containing 100 color images, each image being coded as $`256\times 384`$ pixels with colors coded in 24 bits (16M colors). The families depict *bears, African specialty animals, monkeys, cougars, fireworks, mountains, office interiors, bonsais, sunsets, clouds, apes* and *rocks and gems*. The database is randomly split into balanced sets of 900 training images and 300 test images. The task consists in classifying the test images with the rule learned by training 12 one-vs-all SVM’s on the learning fold. The object are then classified according to the SVM performing the highest score, namely with a “winner-takes-all” strategy. The results presented in this section are averaged over 4 different random splits. We used the CImg package to generate histograms and the Spider toolbox for the SVM experiments<sup>2</sup><sup>2</sup>2http://cimg.sourceforge.net/ and http://www.kyb.tuebingen.mpg.de/bs/people/spider/.
We adopted a coarser representation of 9 bits per color for the $`98,304`$ pixels of each image, rather than the 24 available ones to reduce the size of the RGB color space to $`8^3=512`$ from the original set of $`256^3=16,777,216`$ colors. In this image retrieval experiment, we used localization as the conditioning index set, dividing the images into $`1,4,4^2=16,9`$ and $`9^2=81`$ local histograms (in Figure 2 the image was for instance divided into $`4^3=64`$ windows). To define the branching process prior, we simply set an uniform value over all the grid of $`\epsilon `$ of $`1/\alpha `$, an usage motivated by previous experiments led in a similar context . Finally, we used kernels described in both and to define the base kernel $`k`$. These kernels can be directly applied on sub-probability measures, which is not the case for all kernels on multinomials, notably the Information Diffusion Kernel . We report results for two families of kernels, namely the Radial Basis Function expressed for multinomials and the entropy kernel based on the Jensen divergence :
$$k_{a,b,\rho }(\theta ,\theta ^{})=e^{\rho {\scriptscriptstyle |\theta _i^a\theta _{i}^{}{}_{}{}^{a}|^b}},k_h(\theta ,\theta ^{})=e^{h\left(\frac{\theta +\theta ^{}}{2}\right)+\frac{1}{2}\left(h(\theta )+h(\theta ^{})\right)}.$$
For most kernels not presented here, the multiresolution approach usually improved the performance in a similar way than the results presented in Table 1. Finally, we also report that using only the finest resolution available in each $`(\alpha ,D)`$ setting, that is a branching process prior uniformly set to $`1`$, yielded better results than the use of the coarsest histogram without achieving however the same performance of the multiresolution averaging framework, which highlights the interest of taking both coarse and fine perspectives into account. When $`a=.25`$ for instance, this setting produced 16.5% and 16.2% error rates for $`\alpha =4`$ and $`D=1,2`$, and 15.8% for $`\alpha =9`$ and $`D=1`$.
## Acknowledgments
MC would like to thank Jean-Philippe Vert and Arnaud Doucet for fruitful discussions, as well as Xavier Dupré for his help with the CImg toolbox. |
warning/0507/math0507314.html | ar5iv | text | # Link complexes of subspace arrangements
## 1. Introduction
In , Steingrímsson introduced the coloring complex $`\mathrm{\Delta }_G`$. This is a simplicial complex associated with a graph $`G`$. The Hilbert polynomial of its Stanley-Reisner ring $`k[\mathrm{\Delta }_G]`$ is closely related to the chromatic polynomial $`P_G(x)`$ in a way that is made precise in Section 5.
Answering a question of Steingrímsson, Jonsson proved that $`\mathrm{\Delta }_G`$ is a Cohen-Macaulay complex by showing that it is constructible. In particular, $`\mathrm{\Delta }_G`$ being Cohen-Macaulay imposes restrictions on the Hilbert polynomial of $`k[\mathrm{\Delta }_G]`$, hence on $`P_G(x)`$.
Since $`\mathrm{\Delta }_G`$ is a Cohen-Macaulay complex, a natural question, asked already in , is whether it is shellable — a stronger property than constructibility.
In , $`\mathrm{\Delta }_G`$ was defined in a combinatorially very explicit way. Another way to view $`\mathrm{\Delta }_G`$ is, however, as a simplicial decomposition of the link (i.e. intersection with the unit sphere) of the graphical hyperplane arrangement associated with $`G`$. In this guise, $`\mathrm{\Delta }_G`$ appeared in work of Herzog, Reiner and Welker . Adopting this point of view, one may define a similar complex $`\mathrm{\Delta }_{𝒜,}`$ for any subspace arrangement $`𝒜`$, as long as it has an embedding in a simplicial hyperplane arrangement $``$.
This paper has two goals. The first is addressed in Section 4 where we show that $`\mathrm{\Delta }_{𝒜,}`$ is shellable whenever $`𝒜`$ consists of hyperplanes. In particular, this proves that the coloring complexes are shellable.
The chromatic polynomial of $`G`$ is essentially the characteristic polynomial of the corresponding graphical hyperplane arrangement. Bearing this in mind, one may hope to extend the aforementioned connection between the Hilbert polynomial of $`k[\mathrm{\Delta }_G]`$ and $`P_G(x)`$ to more general complexes $`\mathrm{\Delta }_{𝒜,}`$. Achieved in Section 5, our second goal is to carry out this extension whenever $``$ is a Coxeter arrangement of type $`A`$ or $`B`$. When $`𝒜`$ consists of hyperplanes and $``$ is of type $`A`$, Steingrímsson’s result is recovered.
We define the complexes $`\mathrm{\Delta }_{𝒜,}`$ in Section 3 after reviewing some necessary background in the next section.
## 2. Preliminaries
### 2.1. Subspace arrangements and characteristic polynomials
By the term subspace arrangement we mean a finite collection $`𝒜=\{A_1,\mathrm{},A_t\}`$ of linear subspaces, none of which contains another, of some ambient vector space. In our case, the ambient space will always be $`^n`$ for some $`n`$. To $`𝒜`$ we associate the intersection lattice $`L_𝒜`$ which consists of all intersections of subspaces in $`𝒜`$ ordered by reverse inclusion. (We emphasize the fact that $`𝒜`$ contains no strictly affine subspaces; in particular this implies that $`L_𝒜`$ is indeed a lattice.)
An important invariant of the arrangement $`𝒜`$ is its characteristic polynomial
$$\chi (𝒜;x)=\underset{YL_𝒜}{}\mu (\widehat{0},Y)x^{dim(Y)},$$
where $`\mu `$ is the Möbius function of $`L_𝒜`$ and $`\widehat{0}=^n`$ is the smallest element in $`L_𝒜`$.
Given a subspace $`A𝒜`$, we define two new arrangements, namely the deletion
$$𝒜A=𝒜\{A\}$$
and the restriction
$$𝒜/A=\mathrm{max}\{AB|B𝒜A\},$$
where $`\mathrm{max}𝒮`$ denotes the collection of inclusion-maximal members of a set family $`𝒮`$. Another way to think of $`𝒜/A`$ is as the set of elements covering $`A`$ in $`L_𝒜`$. In this way, we may extend the definition of $`𝒜/A`$ to arbitrary $`AL_𝒜`$. We consider $`𝒜A`$ to be an arrangement in $`^n`$, whereas $`𝒜/A`$ is an arrangement in $`A`$.
When $`𝒜`$ is a hyperplane arrangement, the next result is standard. We expect the general case to be known, too, although we have been unable to find it in the literature.
###### Theorem 2.1 (Deletion-Restriction).
For a subspace arrangement $`𝒜`$ and any subspace $`A𝒜`$, we have
$$\chi (𝒜;x)=\chi (𝒜A;x)\chi (𝒜/A;x).$$
###### Proof.
Choose $`YL_𝒜`$. We claim that
$$\mu _𝒜(\widehat{0},Y)=\{\begin{array}{cc}\mu _{𝒜A}(\widehat{0},Y)\mu _𝒜(A,Y)\hfill & \text{ if }YL_{𝒜A}\text{,}\hfill \\ \mu _𝒜(A,Y)\hfill & \text{ otherwise,}\hfill \end{array}$$
where $`\mu _𝒜`$ denotes the Möbius function of $`L_𝒜`$ which we think of as a function $`L_𝒜\times L_𝒜`$ with $`ST\mu _𝒜(S,T)=0`$ (and similarly for $`𝒜A`$).
The claim is true if $`Y=\widehat{0}=^n`$, so assume it has been verified for all $`Z<Y`$ in $`L_𝒜`$. If $`YL_{𝒜A}`$ we obtain
$$\begin{array}{cc}\hfill \mu _𝒜(\widehat{0},Y)& =\underset{\widehat{0}Z<Y}{}\mu _𝒜(\widehat{0},Z)=\underset{\stackrel{\widehat{0}Z<Y}{ZL_{𝒜A}}}{}\mu _{𝒜A}(\widehat{0},Z)+\underset{AZ<Y}{}\mu _𝒜(A,Z)\hfill \\ & =\mu _{𝒜A}(\widehat{0},Y)\mu _𝒜(A,Y),\hfill \end{array}$$
as desired. If, on the other hand, $`YL_{𝒜A}`$, then there is a unique largest element in $`L_{𝒜A}`$ which is below $`Y`$ in $`L_𝒜`$, namely the join of all atoms (weakly) below $`Y`$ except $`A`$; call this element $`W`$. If $`W=\widehat{0}`$, then $`Y=A`$ and we are done. Otherwise,
$$\begin{array}{cc}\hfill \mu _𝒜(\widehat{0},Y)& =\underset{\widehat{0}Z<Y}{}\mu _𝒜(\widehat{0},Z)=\underset{\stackrel{\widehat{0}ZW}{ZL_{𝒜A}}}{}\mu _{𝒜A}(\widehat{0},Z)+\underset{AZ<Y}{}\mu _𝒜(A,Z)\hfill \\ & =\underset{AZ<Y}{}\mu _𝒜(A,Z)=\mu _𝒜(A,Y),\hfill \end{array}$$
establishing the claim.
We conclude that
$$\chi (𝒜;x)=\underset{YL_{𝒜A}}{}\mu _{𝒜A}(\widehat{0},Y)x^{dim(Y)}\underset{YA}{}\mu _𝒜(A,Y)x^{dim(Y)}.$$
Not every $`Y`$ in the last sum belongs to $`L_{𝒜/A}`$ in general; the latter is join-generated by the elements covering $`A`$ in $`L_𝒜`$. However, it follows from Rota’s Crosscut theorem that for every $`YA`$ in $`L_A`$,
$$\mu _𝒜(A,Y)=\{\begin{array}{cc}\mu _{𝒜/A}(A,Y)\hfill & \text{ if }YL_{𝒜/A}\text{,}\hfill \\ 0\hfill & \text{ otherwise.}\hfill \end{array}$$
Thus,
$$\underset{YA}{}\mu _𝒜(A,Y)x^{dim(Y)}=\chi (𝒜/A;x),$$
and the theorem follows. ∎
Two (families of) hyperplane arrangements are of particular importance to us. The first is the braid arrangement $`𝒮_n`$. This is an arrangement whose ambient space is $`\{(x_1,\mathrm{},x_n)^nx_1+\mathrm{}+x_n=0\}^{n1}`$. The $`\left(\genfrac{}{}{0pt}{}{n}{2}\right)`$ hyperplanes in $`𝒮_n`$ are given by the equations $`x_i=x_j`$ for all $`1i<jn`$.
The braid arrangement is the set of reflecting hyperplanes of a Weyl group of type $`A`$. Considering type $`B`$ instead, we find our second important family of arrangements. Explicitly, $`_n`$ is the arrangement of the $`n^2`$ hyperplanes in $`^n`$ that are given by the equations $`x_i=\tau x_j`$ for all $`1i<jn`$, $`\tau \{1,1\}`$, and $`x_i=0`$ for all $`i[n]=\{1,\mathrm{},n\}`$.
### 2.2. Stanley-Reisner rings and $`h`$-polynomials
Let $`\mathrm{\Delta }`$ be a simplicial complex on the vertex set $`[n]`$. Regarding the vertices as variables, we want to consider the ring of polynomials that live on $`\mathrm{\Delta }`$. To this end, for a field $`k`$, we define the Stanley-Reisner ideal $`I_\mathrm{\Delta }k[x_1,\mathrm{},x_n]`$ by
$$I_\mathrm{\Delta }=\{x_{i_1}\mathrm{}x_{i_t}|\{i_1,\mathrm{},i_t\}\mathrm{\Delta }\}.$$
The quotient ring
$$k[\mathrm{\Delta }]=k[x_1,\mathrm{},x_n]/I_\mathrm{\Delta }$$
is the Stanley-Reisner ring of $`\mathrm{\Delta }`$, which is a graded algebra with the standard grading by degree. When speaking of algebraic properties, such as Cohen-Macaulayness, of $`\mathrm{\Delta }`$ we have the corresponding properties of $`k[\mathrm{\Delta }]`$ in mind.
Given a simplicial complex $`\mathrm{\Delta }`$ of dimension $`d1`$, its $`h`$-polynomial is
$$h(\mathrm{\Delta };x)=\underset{i=0}{\overset{d}{}}f_{i1}(x1)^{di},$$
where $`f_i`$ is the number of $`i`$-dimensional simplices in $`\mathrm{\Delta }`$ (including $`f_1=1`$ if $`\mathrm{\Delta }`$ is nonempty). One important feature of the $`h`$-polynomial is that it carries all information needed to compute the Hilbert series of $`k[\mathrm{\Delta }]`$. Specifically,
$$\text{Hilb}(k[\mathrm{\Delta }];x)=\frac{\overline{h}(\mathrm{\Delta };x)}{(1x)^d},$$
where $`\overline{h}`$ denotes the reverse $`h`$-polynomial:
$$\overline{h}(\mathrm{\Delta };x)=x^dh(\mathrm{\Delta };\frac{1}{x}).$$
### 2.3. Shellable complexes
Suppose $`\mathrm{\Delta }`$ is a pure simplicial complex, meaning that all facets (maximal simplices) have the same dimension $`d1`$. A shelling order for $`\mathrm{\Delta }`$ is a total ordering $`F_1,\mathrm{},F_t`$ of the facets of $`\mathrm{\Delta }`$ such that $`F_j\left(_{i<j}F_i\right)`$ is pure of dimension $`d2`$ for all $`j=2,\mathrm{},t`$. We say that $`\mathrm{\Delta }`$ is shellable if a shelling order for $`\mathrm{\Delta }`$ exists.
One good reason to care about shellability is that it implies Cohen-Macaulayness.
## 3. The objects of study
Suppose $``$ is a hyperplane arrangement in $`^n`$ such that $`=\{0\}`$. Then, $``$ determines a regular cell decomposition $`\mathrm{\Delta }_{}`$ of the unit sphere $`S^{n1}`$. In short, each point $`p`$ on $`S^{n1}`$ has an associated sign vector in $`\{0,,+\}^{||}`$ recording for each hyperplane $`h`$ whether $`p`$ is on, or on the negative, or on the positive side of $`h`$ (for some choice of orientations of the hyperplanes). A cell in $`\mathrm{\Delta }_{}`$ consists of the set of points with a common sign vector. The face poset of $`\mathrm{\Delta }_{}`$ is the big face lattice of the corresponding oriented matroid, see .
If $`\mathrm{\Delta }_{}`$ is a simplicial complex, then $``$ is called simplicial. A prime example of a simplicial hyperplane arrangement is the collection of reflecting hyperplanes of a finite Coxeter group. In this case, $`\mathrm{\Delta }_{}`$ coincides with the Coxeter complex.
From now on, let $``$ be a simplicial hyperplane arrangement.
Consider an antichain $`𝒜`$ in $`L_{}`$. We say that the subspace arrangement $`𝒜`$ is embedded in $``$. Observe that $`𝒜S^{n1}`$, which is known as the link of $`𝒜`$, has the structure of a simplicial subcomplex of $`\mathrm{\Delta }_{}`$. This subcomplex is the principal object of study in this paper. We denote it $`\mathrm{\Delta }_{𝒜,}`$.
###### Example 3.1.
A graph $`G=([n],E)`$ determines a graphical hyperplane arrangement $`\widehat{G}`$ in the $`(n1)`$-dimensional subspace of $`^n`$ given by the equation $`x_1+\mathrm{}+x_n=0`$. There is one hyperplane in $`\widehat{G}`$ for each edge in $`E`$; the hyperplane corresponding to the edge $`\{i,j\}`$ has the equation $`x_i=x_j`$.
The arrangement $`\widehat{K_n}`$ corresponding to the complete graph is nothing but the braid arrangement $`𝒮_n`$ which is simplicial. Any graph $`G`$ thus determines a simplicial complex $`\mathrm{\Delta }_{\widehat{G},𝒮_n}`$. It coincides with Steingrímsson’s coloring complex of $`G`$ which was denoted $`\mathrm{\Delta }_G`$ in the Introduction. The complex $`\mathrm{\Delta }_{\widehat{G},𝒮_n}`$ also appeared under the name $`\mathrm{\Delta }_{m,J}`$ in .
We remark that the homotopy type of the link of $`𝒜`$, hence of $`\mathrm{\Delta }_{𝒜,}`$, can be computed in terms of the order complexes of lower intervals in $`L_𝒜`$ by a formula of Ziegler and Živaljević . When $`𝒜`$ consists of hyperplanes we may simply note that $`\mathrm{\Delta }_{𝒜,}`$ is homotopy equivalent to the $`(n1)`$-sphere with one point removed for each connected region in the complement $`^n𝒜`$. Denoting by $`R(𝒜)`$ the number of such regions, $`\mathrm{\Delta }_{𝒜,}`$ is thus homotopy equivalent to a wedge of $`R(𝒜)1`$ spheres of dimension $`n2`$ in this case. For the arrangements $`\widehat{G}`$ of Example 3.1 it is not difficult to see that $`R(\widehat{G})`$ equals the number $`\text{AO}(G)`$ of acyclic orientations of $`G`$. Thus, $`\mathrm{\Delta }_{\widehat{G},𝒮_n}`$ has the homotopy type of a wedge of $`AO(G)1`$ $`(n3)`$-spheres (). In particular, the reduced Euler characteristic of $`\mathrm{\Delta }_{\widehat{G},𝒮_n}`$ is $`\pm (\text{AO}(G)1)`$ (\[10, Theorem 17\]).
## 4. Shellability in the hyperplane case
Our goal in this section is to show that $`\mathrm{\Delta }_{𝒜,}`$ is shellable whenever $`𝒜`$ consists of hyperplanes. Applied to the complexes $`\mathrm{\Delta }_{\widehat{G},𝒮_n}`$ of Example 3.1 this answers affirmatively a question of Steingrímsson which was restated in . The key tool is a particular class of shellings of $`\mathrm{\Delta }_{}`$ determined by the poset of regions of $``$ which we now define.
The complement $`^n`$ is cut into disjoint open regions by $``$. Restricting to the unit sphere, their closures are the facets of $`\mathrm{\Delta }_{}`$. Let $`=()`$ be the set of such facets. For $`R,R^{}`$, say that $`h`$ separates $`R`$ and $`R^{}`$ if their respective interiors are on different sides of $`h`$.
Choose a base region $`B`$ arbitrarily. We have a distance function $`\mathrm{}:`$ which maps a region $`R`$ to the number of hyperplanes in $``$ which separate $`R`$ and $`B`$. Now, for two regions $`R,R^{}`$, write $`RR^{}`$ iff $`R`$ and $`R^{}`$ are separated by exactly one hyperplane in $``$ and $`\mathrm{}(R)=\mathrm{}(R^{})1`$. The poset of regions $`P_{}`$ is the partial order on $``$ whose covering relation is $``$. It was first studied by Edelman .
From the point of view of this paper, the most important property of $`P_{}`$ is the following.
###### Theorem 4.1 (Theorem 4.3.3 in ).
Any linear extension of $`P_{}`$ is a shelling order for $`\mathrm{\Delta }_{}`$.
We are now ready to state and prove the main result of this section.
###### Theorem 4.2.
If $`𝒜`$ consists of hyperplanes, then $`\mathrm{\Delta }_{𝒜,}`$ is shellable.
###### Proof.
We proceed by induction over $`|𝒜|`$. When $`𝒜=\{A\}`$, we may apply Theorem 4.1 since $`\mathrm{\Delta }_{𝒜,}=\mathrm{\Delta }_{/A}`$ in this case.
Now suppose $`|𝒜|2`$ and that we have a shelling order for $`\mathrm{\Delta }_{𝒜A,}`$ for some $`A𝒜`$. We will append the remaining facets to this order.
The remaining facets are the facets of $`\mathrm{\Delta }_{\{𝒜\},}=\mathrm{\Delta }_{/A}`$. They are divided into equivalence classes in the following way: $`F`$ and $`G`$ belong to the same class iff their interiors belong to the same connected component of $`^n(𝒜A)`$ (or, equivalently, to the same connected component of $`A(𝒜/A)`$). Observe that if $`F`$ and $`G`$ belong to different classes, then $`FG\mathrm{\Delta }_{𝒜A,}`$. Thus, it is enough to show that the facets in any equivalence class can be appended to the shelling order for $`\mathrm{\Delta }_{𝒜A,}`$.
Without loss of generality, consider the class which contains the maximal element in $`P_{/A}`$, i.e. the region opposite to the base region. Call this class $`C`$. If $`FC`$ and $`GC`$ for $`F,GP_{/A}`$, then some hyperplane in $`𝒜/A/A`$ separates $`F`$ from $`G`$, and $`G`$ is on the positive side of this hyperplane. Thus, $`FG`$. This shows that $`C`$ is an order filter in $`P_{/A}`$. According to Theorem 4.1, $`\mathrm{\Delta }_{/A}`$ has a shelling order which ends with the facets in $`C`$. Now observe that $`(C)((P_{/A}C))=(C)\mathrm{\Delta }_{𝒜A,}`$. The facets in $`C`$ may therefore be appended in this order to the shelling order for $`\mathrm{\Delta }_{𝒜A,}`$. ∎
## 5. The $`h`$-polynomial of $`\mathrm{\Delta }_{𝒜,}`$
For brevity we write $`h(𝒜,;x)`$ meaning $`h(\mathrm{\Delta }_{𝒜,};x)`$ and similarly for $`\overline{h}`$. The following result of Steingrímsson serves as a motivating example for this section:
###### Theorem 5.1 (Theorem 13 in ).
Recall the complex $`\mathrm{\Delta }_{\widehat{G},𝒮_n}`$ defined in Example 3.1. We have
$$\frac{x\overline{h}(\widehat{G},𝒮_n;x)}{(1x)^n}=\underset{m0}{}\left(m^nP_G(m)\right)x^m,$$
where $`P_G`$ is the chromatic polynomial of $`G`$.
This theorem is interesting because of the connection between the left hand side and the Hilbert series of the Stanley-Reisner ring $`k[\mathrm{\Delta }_{\widehat{G},𝒮_n}]`$. In , Brenti began a systematic study of which polynomials arise as Hilbert polynomials of standard graded algebras. A question left open in , and later answered affirmatively by Almkvist , was whether chromatic polynomials of graphs have this property. Theorem 5.1 implies something similar, namely that $`(m+1)^nP_G(m+1)`$ is the Hilbert polynomial (in $`m`$) of a standard graded algebra; for details, see Corollary 5.7 below.
It is well-known that $`P_G(x)=x\chi (\widehat{G};x)`$; one way to prove it is to compare Theorem 2.1 with the standard deletion-contraction recurrence for $`P_G`$. The identity suggests the possibility of extending Theorem 5.1 to other complexes $`\mathrm{\Delta }_{𝒜,}`$. This turns out to be possible at least if $`\{𝒮_n,_n\}`$ and is the topic of this section.
Given a subspace $`T`$ of $`^n`$, let $`d(T)`$ denote its dimension. For a subspace arrangement $`𝒯`$, we also write
$$d(𝒯)=\underset{T𝒯}{\mathrm{max}}d(T).$$
###### Lemma 5.2.
Let $`A𝒜`$. Then,
$$\begin{array}{cc}\hfill h(𝒜,;x)& =(x1)^{d(𝒜)d(𝒜A)}h(𝒜A,;x)\hfill \\ & +(x1)^{d(𝒜)d(A)}h(\{A\},;x)\hfill \\ & (x1)^{d(𝒜)d(𝒜/A)}h(𝒜/A,/A;x).\hfill \end{array}$$
###### Proof.
Each simplex in $`\mathrm{\Delta }_{𝒜,}`$ belongs to $`\mathrm{\Delta }_{𝒜A,}`$ or to $`\mathrm{\Delta }_{\{A\},}`$ or to both. Also, observe that $`\mathrm{\Delta }_{𝒜A,}\mathrm{\Delta }_{\{A\},}=\mathrm{\Delta }_{𝒜/A,/A}`$. Denoting by $`f_i(𝒮,𝒯)`$ the number of $`i`$-dimensional simplices in $`\mathrm{\Delta }_{𝒮,𝒯}`$, we thus obtain for all $`i`$
$$f_i(𝒜,)=f_i(𝒜A,)+f_i(\{A\},)f_i(𝒜/A,/A).$$
The lemma now follows from the fact that $`dim(\mathrm{\Delta }_{𝒮,𝒯})=d(𝒮)1`$. ∎
We may use Lemma 5.2 to recursively compute $`h(𝒜,;x)`$. As it turns out, this recursion is particularly useful when $`\{𝒮_n,_n\}`$. The reason is given by the following two lemmata.
###### Lemma 5.3.
We have
$$\frac{x\overline{h}(\mathrm{\Delta }_{𝒮_n};x)}{(1x)^{n+1}}=\underset{m0}{}m^nx^m$$
and
$$\frac{\overline{h}(\mathrm{\Delta }__n;x)}{(1x)^{n+1}}=\underset{m0}{}(2m+1)^nx^m.$$
###### Proof.
The complexes $`\mathrm{\Delta }_{𝒮_n}`$ and $`\mathrm{\Delta }__n`$ coincide with the Coxeter complexes of types $`A_{n1}`$ and $`B_n`$, respectively. For the $`h`$-polynomials this implies that $`x\overline{h}(\mathrm{\Delta }_{𝒮_n};x)=A_n(x)`$ and $`\overline{h}(\mathrm{\Delta }__n;x)=B_n(x)`$, where $`A_n`$ is the $`n`$th Eulerian polynomial and $`B_n`$ is the $`n`$th $`B`$-Eulerian polynomial, see . The assertions are well-known properties of these polynomials \[4, Theorem 3.4.ii\]. ∎
###### Lemma 5.4.
1. For any subspace $`AL_{𝒮_n}`$, we have
$$\frac{x\overline{h}(\{A\},𝒮_n;x)}{(1x)^{d(A)+2}}=\underset{m0}{}m^{d(A)+1}x^m.$$
2. For any subspace $`𝒜L__n`$, we have
$$\frac{\overline{h}(\{A\},_n;x)}{(1x)^{d(A)+1}}=\underset{m0}{}(2m+1)^{d(A)}x^m.$$
###### Proof.
A key property of $`𝒮_n`$ ($`_n`$), which is readily checked, is that its restriction to any subspace in the intersection lattice is again a type $`A`$ ($`B`$) hyperplane arrangement. Thus, $`\mathrm{\Delta }_{\{A\},𝒮_n}=\mathrm{\Delta }_{𝒮_n/A}\mathrm{\Delta }_{𝒮_{d(A)+1}}`$ ($`\mathrm{\Delta }_{\{A\},_n}=\mathrm{\Delta }_{_n/A}\mathrm{\Delta }_{_{d(A)}}`$). The assertions now follow from Lemma 5.3
The leading term of $`\chi (𝒜;x)`$ is always $`x^n`$, where $`n`$ is the dimension of the ambient space. It is convenient to introduce the tail $`T(𝒜;x)=x^n\chi (𝒜;x)`$.
When $`𝒜`$ consists of hyperplanes, the following result coincides with Theorem 5.1.
###### Theorem 5.5.
Suppose $`𝒜`$ is a subspace arrangement embedded in $`𝒮_n`$. Then,
$$\frac{x\overline{h}(𝒜,𝒮_n;x)}{(1x)^{d(𝒜)+2}}=\underset{m0}{}mT(𝒜;m)x^m.$$
###### Proof.
We proceed by induction over $`|𝒜|`$, noting that $`|𝒜A|<|𝒜|`$ and $`|𝒜/A|<|𝒜|`$ for every $`A𝒜`$. If $`|𝒜|=1`$, we have $`\chi (𝒜;m)=m^{n1}m^{d(𝒜)}`$, so that $`T(𝒜;m)=m^{d(𝒜)}`$, and the theorem follows from part (i) of Lemma 5.4.
Now suppose $`|𝒜|2`$ and pick a subspace $`A𝒜`$. Using Lemma 5.2 and the induction hypothesis, we obtain
$$\begin{array}{cc}\hfill \frac{x^{d(𝒜)+1}h(𝒜,𝒮_n;\frac{1}{x})}{(1x)^{d(𝒜)+2}}& =\left(\frac{1x}{x}\right)^{d(𝒜)d(𝒜A)}\frac{x^{d(𝒜)+1}h(𝒜A,𝒮_n;\frac{1}{x})}{(1x)^{d(𝒜)+2}}\hfill \\ & +\left(\frac{1x}{x}\right)^{d(𝒜)d(A)}\frac{x^{d(𝒜)+1}h(\{A\},𝒮_n;\frac{1}{x})}{(1x)^{d(𝒜)+2}}\hfill \\ & \left(\frac{1x}{x}\right)^{d(𝒜)d(𝒜/A)}\frac{x^{d(𝒜)+1}h(𝒜/A,𝒮_n/A;\frac{1}{x})}{(1x)^{d(𝒜)+2}}\hfill \\ & =\underset{m0}{}m(m^{n1}\chi (𝒜A;m))x^m\hfill \\ & +\underset{m0}{}m(m^{n1}(m^{n1}m^{d(A)}))x^m\hfill \\ & \underset{m0}{}m(m^{d(A)}\chi (𝒜/A;m))x^m\hfill \\ & =\underset{m0}{}m(m^{n1}\chi (𝒜;m))x^m,\hfill \end{array}$$
where the last equality follows from Deletion-Restriction.
For completeness, we should also check the uninteresting case $`|𝒜|=0`$ which is not covered by the above arguments. Here, $`\overline{h}(\mathrm{},𝒮_n;x)=0`$ and $`T(\mathrm{};x)=0`$, and the assertion holds. ∎
Employing part (ii) of Lemma 5.4 instead of part (i), and keeping track of the fact that $`_n`$ is an arrangement in $`^n`$, whereas $`𝒮_n`$ sits in $`^{n1}`$, the proof of Theorem 5.5 is easily adjusted to a proof of the next result.
###### Theorem 5.6.
Suppose $`𝒜`$ is a subspace arrangement embedded in $`_n`$. Then,
$$\frac{\overline{h}(𝒜,_n;x)}{(1x)^{d(𝒜)+1}}=\underset{m0}{}T(𝒜;2m+1)x^m.$$
For subspace arrangements covered by Theorem 5.5 or Theorem 5.6, we may now draw the promised algebraic conclusions. To this end, for a simplicial complex $`\mathrm{\Gamma }`$ and a subcomplex $`\mathrm{\Gamma }^{}\mathrm{\Gamma }`$, let $`𝒥_{\mathrm{\Gamma }^{},\mathrm{\Gamma }}`$ be the ideal in the Stanley-Reisner ring $`k[\mathrm{\Gamma }]`$ generated by the (equivalence classes of) monomials corresponding to simplices in $`\mathrm{\Gamma }`$ that do not belong to $`\mathrm{\Gamma }^{}`$.
###### Corollary 5.7.
Suppose $`𝒜`$ is a subspace arrangement embedded in $`𝒮_n`$. Let $`\mathrm{\Gamma }`$ denote the double cone over $`\mathrm{\Delta }_{𝒮_n}`$, and write $`\mathrm{\Gamma }^{}`$ for the double cone over $`\mathrm{\Delta }_{𝒜,𝒮_n}`$ with the same cone points. The following holds:
1. The Hilbert polynomial of $`k[\mathrm{\Gamma }^{}]`$ is $`F(k[\mathrm{\Gamma }^{}];m)=(m+1)T(𝒜;m+1)`$.
2. The Hilbert polynomial of $`𝒥_{\mathrm{\Gamma }^{},\mathrm{\Gamma }}`$ is $`F(𝒥_{\mathrm{\Gamma }^{},\mathrm{\Gamma }};m)=(m+1)\chi (𝒜;m+1)`$.
###### Proof.
The dimension of $`\mathrm{\Gamma }^{}`$ is $`d(𝒜)+1`$. Taking a cone over a simplicial complex does not affect the $`\overline{h}`$-polynomial. Thus,
$$\text{Hilb}(k[\mathrm{\Gamma }^{}];x)=\frac{\overline{h}(𝒜,𝒮_n;x)}{(1x)^{d(𝒜)+2}}=\frac{1}{x}\underset{m0}{}mT(𝒜;m)x^m,$$
where the second equality follows from Theorem 5.5. This proves (i).
For (ii), we use that
$$k[\mathrm{\Gamma }^{}]k[\mathrm{\Gamma }]/𝒥_{\mathrm{\Gamma }^{},\mathrm{\Gamma }}.$$
For the Hilbert series, this implies
$$\text{Hilb}(k[\mathrm{\Gamma }^{}];x)=\text{Hilb}(k[\mathrm{\Gamma }];x)\text{Hilb}(𝒥_{\mathrm{\Gamma }^{},\mathrm{\Gamma }};x).$$
From part (i) and the fact that
$$\text{Hilb}(k[\mathrm{\Gamma }])=\frac{\overline{h}(\mathrm{\Delta }_{𝒮_n};x)}{(1x)^{n+1}}=\frac{1}{x}\underset{m0}{}m^nx^m,$$
we conclude
$$\text{Hilb}(𝒥_{\mathrm{\Gamma }^{},\mathrm{\Gamma }};x)=\frac{1}{x}\underset{m0}{}m^nx^m\frac{1}{x}\underset{m0}{}mT(𝒜;m)x^m=\frac{1}{x}\underset{m0}{}m\chi (𝒜;m)x^m.$$
The situation for $`_n`$ is analogous, although we use cones instead of double cones. This is a manifestation of the fact that $`_n`$ and $`𝒮_n`$ differ by one in dimension.
###### Corollary 5.8.
Suppose $`𝒜`$ is a subspace arrangement embedded in $`_n`$. Let $`\mathrm{\Gamma }`$ denote the cone over $`\mathrm{\Delta }__n`$, and write $`\mathrm{\Gamma }^{}`$ for the cone over $`\mathrm{\Delta }_{𝒜,_n}`$ with the same cone point. Then, the following holds:
1. The Hilbert polynomial of $`k[\mathrm{\Gamma }^{}]`$ is $`F(k[\mathrm{\Gamma }^{}];m)=T(𝒜;2m+1)`$.
2. The Hilbert polynomial of $`𝒥_{\mathrm{\Gamma }^{},\mathrm{\Gamma }}`$ is $`F(𝒥_{\mathrm{\Gamma }^{},\mathrm{\Gamma }};m)=\chi (𝒜;2m+1)`$.
###### Proof.
Proceeding as in the proof of Corollary 5.7, using Theorem 5.6 instead of Theorem 5.5, we prove (i) by observing
$$\text{Hilb}(k[\mathrm{\Gamma }^{}];x)=\frac{\overline{h}(𝒜,_n;x)}{(1x)^{d(𝒜)+1}}=\underset{m0}{}T(𝒜;2m+1)x^m.$$
For (ii), note that
$$\text{Hilb}(k[\mathrm{\Gamma }];x)=\frac{\overline{h}(\mathrm{\Delta }__n;x)}{(1x)^{n+1}}=\underset{m0}{}(2m+1)^nx^m.$$
Thus,
$$\text{Hilb}(𝒥_{\mathrm{\Gamma }^{},\mathrm{\Gamma }};x)=\underset{m0}{}(2m+1)^nx^m\underset{m0}{}T(𝒜;2m+1)x^m=\underset{m0}{}\chi (𝒜;2m+1)x^m.$$
Any hypergraph (without inclusions among edges) $`G`$ on $`n`$ vertices corresponds to a subspace arrangement $`\widehat{G}`$ embeddable in $`𝒮_n`$. The construction is virtually the same as in Example 3.1; with the hyperedge $`\{i_1,\mathrm{},i_t\}`$ is associated the subspace given by $`x_{i_1}=\mathrm{}=x_{i_t}`$. As for ordinary graphs (the hyperplane case), we have $`x\chi (\widehat{G};x)=P_G(x)`$, cf. \[9, Theorem 3.4\]. In this way, Corollary 5.7 allows us to interpret chromatic polynomials of hypergraphs in terms of Hilbert polynomials. For ordinary graphs, this is the content of Steingrímsson’s \[10, Corollary 10\].
Corollary 5.8, too, has an impact on chromatic polynomials. Any signed graph (in the sense of Zaslavsky ) $`G`$ on $`n`$ vertices corresponds to a hyperplane arrangement $`\widehat{G}_n`$, and vice versa. A signed graph $`G`$ has a chromatic polynomial $`P_G(x)`$, and $`P_G(x)=\chi (\widehat{G};x)`$ . |
warning/0507/astro-ph0507547.html | ar5iv | text | # Bivariate Galaxy Luminosity Functions in the Sloan Digital Sky Survey
## 1 Introduction
The galaxy luminosity function (LF) is well-known as a fundamental measurement of the properties of galaxies. It constrains theories of their formation and evolution, such as biased galaxy formation (e.g. Benson et al., 2000) and is needed for many other measurements. Examples are the luminosity density in the universe (e.g. Cross et al., 2001), the selection function in magnitude-limited galaxy surveys (e.g. Norberg et al., 2002), the distances to various types of object, the number of absorbing objects subdivided by redshift and deprojecting the angular correlation function from two dimensions to three (e.g. Limber, 1953; Binggeli et al., 1988).
The general or universal luminosity function, i.e. that for all galaxies, is defined as the comoving number density of galaxies from luminosity $`L`$ to $`L+dL`$
$$dN=\varphi (L)dLdV.$$
(1)
Measurements of the universal form began with Hubble (1936a, b, c), who claimed a Gaussian form. Various studies followed, which included data on dwarf galaxies in clusters (e.g. Abell, 1962) and the study of Holmberg (1974) of galaxies in the field. Both of these argued for a power-law faint-end slope to the overall LF, as opposed to a Gaussian form.
In the mid 1970s it was suggested (Schechter, 1976) that the optical LF can be approximated by the function
$$\varphi (L)dL=\varphi ^{}\left(\frac{L}{L^{}}\right)^\alpha \mathrm{exp}\left(\frac{L}{L^{}}\right)d\left(\frac{L}{L^{}}\right),$$
(2)
where $`\varphi ^{}`$ is the normalisation, $`L^{}`$ is the characteristic luminosity above which the function has an exponential cutoff and $`\alpha `$ is the value of the power-law slope below $`L^{}`$. The Schechter function, as it is now known, has been used for most subsequent characterisations of the universal LF. The function is motivated by self-similar gravitational condensation of structures (Press & Schechter, 1974), except that $`\alpha `$ is not a fixed parameter.
Until recently the study of the LF, particularly for field galaxies as opposed to those in clusters, has been hampered by the lack of large samples. Few galaxies had redshifts and the photometry was from photographic plates, resulting in a large variation in the measured LF between different surveys. In the past decade data from large redshift surveys, such as the 2dF Galaxy Redshift Survey (2dFGRS, Colless et al., 2001, 2003) and the Sloan Digital Sky Survey (SDSS, York et al., 2000) have become available, greatly improving the situation, although the LFs are still not in perfect agreement — for example, Liske et al. (2003) gives revisions to the normalisations of the LF in various surveys based on the deeper wide-field imaging obtained by the Millennium Galaxy Catalogue (MGC), which they also describe, and recent comparisons of LFs such as Driver (2004) conclude that the variations between surveys are dominated by the systematic errors, particularly at the faint end. The differences are most likely due to surface brightness selection effects.
Whilst the universal LF is reasonably well constrained, it is well-known that the LF varies as a function of intrinsic galaxy properties. Thus the universal LF needs to be augmented by a description of this variation. As above, the lack of data until recently has hampered such studies, but with the large sample sizes (2dFGRS and SDSS) and five band CCD photometry (SDSS) now available, detailed divisions of the galaxies into samples which are still statistically significant is now possible. One can also apply fairly strict sample cuts to minimize biases and still retain large samples.
Previous studies of non-universal LFs have included 1) those focusing on specific environments, particularly clusters as the galaxies are all at approximately the same distance and in a narrow field of view, 2) those focussing on specific types of galaxies, the large number of studies which divide the general LF according to some criterion and fit functions to each bin, and 3) those which study the LF bivariate with another parameter, either binned or as an analytical function. Recent examples, some of which are bivariate, are Popesso et al. (2005) for clusters, Mercurio et al. (2006) for superclusters, Hoyle et al. (2005) for voids, de Jong & Lacey (2000) and de Lapparent et al. (2004) for spirals, and Reda et al. (2004), Stocke et al. (2004) and de Jong et al. (2004) for ellipticals.
Various theories and simulations exist pertaining to the origin of the LF, tied in with the physics of galaxy formation. Examples are Benson et al. (2003) who begin with the mass function of dark matter haloes in the $`\mathrm{\Lambda }`$CDM cosmology and add gas cooling, photoionization, feedback (e.g. from supernovae), galaxy merging and thermal conduction. Mo et al. (2004) assume that the segregation of the galaxy population by environment is due to that of halo properties and use the halo occupation distribution to give predicted LFs by environment. Cooray & Milosavljević (2005) reconstructs the Schechter form of the LF using empirical relations between the central galaxy luminosity and halo mass, and the total galaxy luminosity and halo mass.
Further details of the LF are in various reviews, for example Binggeli, Sandage & Tammann (1988) and subdivided by morphological type in de Lapparent (2003).
In this paper the bivariate LF is studied in bins using the nonparametric stepwise maximum likelihood method of Efstathiou, Ellis & Peterson (1988) and two-dimensional analytical functions are fitted to the results. Previous studies that have done this include Chołoniewski (1985) subdivided by radius, Sodré & Lahav (1993) subdivided by galaxy diameter and Cross et al. (2001), Cross & Driver (2002) and Driver et al. (2005) subdivided by absolute effective surface brightness. The latter quantity is also known as the bivariate brightness distribution.
Ball et al. (2004) showed that morphological types can be reliably assigned to galaxies in the SDSS using artificial neural networks (ANNs) provided a representative training set is available. Here types are assigned to the resolution E, S0, Sa, Sb, Sc, Sd and Im for 37,047 galaxies. In Ball et al. (2006, in preparation, hereafter B06) we use these types to study the variation of galaxy morphology with environment and compare the results to those for colour.
The bivariate LFs are computed for galaxies in the New York Value-Added Galaxy Catalogue (VAGC, Blanton et al., 2005a), based on the Sloan Digital Sky Survey Data Release 4 (DR4, Adelman-McCarthy et al., 2006, http://www.sdss.org/dr4) for a range of galaxy properties, including the morphological type described, inverse concentration index, Sérsic index, absolute effective surface brightness, eClass spectral type, reference frame colours, absolute 90% radius, stellar mass and galaxy environment. The galaxy samples are flux limited to $`r<17.77`$, with the exception of the morphological type, which is limited to $`r<15.9`$.
Here the LF is only considered in the wavebands of the SDSS, but there are many studies in other wavebands, including the ultraviolet (e.g. Baldry et al., 2005; Budavári et al., 2005; Wyder et al., 2005), near-infrared K band (e.g. Loveday, 2000; Cole et al., 2001; Kochanek et al., 2001; Eke et al., 2004; Jones et al., 2006), radio (e.g. Sadler et al., 2002) and x-ray (e.g. Miyaji et al., 2000; Ranalli et al., 2005).
Blanton et al. (2003c) and Loveday (2004) show that the LFs in the SDSS DR1 (Abazajian et al., 2003) to $`z\mathrm{}<0.3`$ are consistent with evolution in the galaxy population over this redshift range. Loveday (2004) shows that even to a redshift of 0.15 evolution in the number density occurs, but that it is only marginally significant in DR1. Here, because we are measuring and fitting many different bivariate functions, we do not attempt to account for evolution, but restrict the sample to $`z0.15`$ to minimize its effect.
Throughout, the standard spatial geometry is assumed, with Euclidean space, $`\mathrm{\Omega }_{\mathrm{matter}}=0.3`$ and $`\mathrm{\Omega }_\mathrm{\Lambda }=0.7`$. For compatibility with previous studies, the dimensionless Hubble constant, $`h=H_0/100\mathrm{km}\mathrm{s}^1\mathrm{Mpc}^1`$, is set to 1.
## 2 Data
The SDSS is a project to map $`\pi `$ steradians of the northern galactic cap in five bands ($`u`$, $`g`$, $`r`$, $`i`$ and $`z`$) from 3,500–8,900 Å. This will provide photometry for of order $`5\times 10^7`$ galaxies. A multifibre spectrograph will provide redshifts and spectra for approximately $`10^6`$ of these. A technical summary of the survey is given in York et al. (2000). The telescope is described in Gunn et al. (2006). The imaging camera is described in Gunn et al. (1998). The photometric system and calibration are described in Fukugita et al. (1996), Hogg et al. (2001), Smith et al. (2002), Ivezić et al. (2004) and Tucker et al. (2006). The astrometric calibration is in Pier et al. (2003) and the data pipelines are in Lupton et al. (2001), Lupton (2006) for the deblender, Frieman et al. and Schlegel et al. (in preparation).
The targeting pipeline chooses targets for spectroscopy from the imaging. A tiling algorithm (Blanton et al., 2003a) then assigns the spectroscopic fibres to the targets, the main source of incompleteness being the minimum distance of 55 arcsec between the fibres. This causes about 6% of galaxies to be missed, and these will be biased towards regions with a high surface density of galaxies. The algorithm gives a more uniform completeness on the sky than a uniform tiling by taking into account large scale structure, but some bias is still present.
The SDSS galaxies with spectra consist of a ‘Main’, flux-limited sample, with a median redshift of 0.104 (Strauss et al., 2002), a luminous red galaxy sample (LRG), approximately volume-limited to $`z0.4`$ (Eisenstein et al., 2001) and a quasar sample (Richards et al., 2002). The limiting magnitude for the Main spectra is $`r<17.77`$, which is substantially brighter than that for the imaging so the redshift completeness is almost 100%. A typical signal-to-noise value is $`>4`$ per pixel and the spectral resolution is 1800. The redshifts have an RMS accuracy of $`\pm 30\mathrm{km}\mathrm{s}^1`$.
This paper uses galaxies from the DR4 version of the VAGC, which is described at http://wassup.physics.nyu.edu/vagc/. Blanton et al. (2005a) describes the DR2 (Abazajian et al., 2004) version. The catalogue is a publicly-available set of FITS files containing the Princeton reductions (http://photo.astro.princeton.edu; http://spectro.astro.princeton.edu) of the raw SDSS data with numerous additional derived quantities and matches to other surveys. The Princeton reductions are designed to improve on the original SDSS pipeline reductions in the publicly available DR4 Catalogue Archive Server (CAS; http://cas.sdss.org/astro/en) and are used throughout unless otherwise stated. The set of objects contained is designed to match the SDSS DR4 and contains galaxies which either match a slightly more inclusive version of the Main sample criteria, are within 2 arcsec of a Main, LRG or quasar target from the version of the photometry used for the targeting, or are within 2 arcsec of a hole drilled in an SDSS spectroscopic plate. The catalogue provides 6,851 square degrees of imaging coverage and 4,681 square degrees of spectroscopic coverage. The raw catalogue contains 1,223,536 objects, 722,866 of which have spectra.
The VAGC units differ from the CAS data in some quantities: the object brightnesses are given as fluxes in nanomaggies (where $`1\mathrm{maggie}=0\mathrm{mag}\mathrm{arcsec}^2`$); the errors are inverse variance and the Petrosian radii are in pixels. These are converted to the units used in which the Main Galaxy Sample is defined, i.e. magnitudes, $`1\sigma `$ errors and arcsec. The pixels are converted to arcsec using the pixscale values given in the full VAGC calibObj photometry outputs. These are all close to 0.396 arcsec pixel<sup>-1</sup>.
From the catalogue, the main file containing the imaging data, object\_sdss\_imaging.fits, was chosen as a base, and the rest of the required parameters for each object were obtained by matching the other datafiles to this one using a method appropriate for each parameter and file in question. The matches are described further below.
The magnitudes are corrected for galactic reddening using the usual corrections derived from the maps of Schlegel, Finkbeiner & Davis (1998). These are of order $`0.1\mathrm{mag}`$, with a mean value of $`0.14\mathrm{mag}`$ and $`1\sigma `$ variation of $`0.11\mathrm{mag}`$. The minimum and maximum values are $`0.02\mathrm{mag}`$ and $`1.04\mathrm{mag}`$ respectively, both in the $`u`$ band.
K-corrections are applied using version 4.1.14 of the code described in Blanton et al. (2003b). This provides a separate K-correction for each individual galaxy, but does not take into account galaxy evolution or dust. The K-corrections were made to a band-shift corresponding to a redshift of 0.1. Most of the datasets have approximately this mean. One could attempt to K-correct to the band-shift for the mean of each dataset, however, this would affect the dataset chosen and so this is not attempted here.
### 2.1 Galaxy Samples and Properties
The SDSS DR4 imaging outputs were used directly as supplied in the VAGC file object\_sdss\_imaging.fits and cross-matched from the full outputs available in the calibObj files. The DR4 spectroscopic outputs were used similarly from object\_sdss\_spectro.fits.
The absolute magnitude range used for each bivariate LF was $`24<M_r5\mathrm{l}\mathrm{o}\mathrm{g}h<15`$ in 18 bins of $`0.5\mathrm{mag}`$, with two exceptions: the JPG morphological type (see below) was binned in 12 bins of $`0.5\mathrm{mag}`$ for $`24<M_r5\mathrm{l}\mathrm{o}\mathrm{g}h<18`$, and the environmental density was in 22 bins for $`24.03<M_r5\mathrm{l}\mathrm{o}\mathrm{g}h<19.63`$, due to its restricted redshift range (see below). Details of each bivariate LF are given in Table 1.
The upper limit of approximately 40 evenly spaced bins in the second parameter was chosen to give a reasonable execution time. The extent of the bins is based on visual inspection of the distributions of the input parameters, but the value of the LF in each bin is independent of the others.
Our analysis allows for the sampling rate of the dataset, which is the ratio of the number of galaxies which have spectra taken to those meeting the criteria for spectra to be taken. The completeness maps as a function of imaging-only parameter pairs are given in §4. The overall completeness is estimated to be 93%. 6% of galaxies are missed due to the minimum 55 arcsec separation between adjacent spectroscopic fibres on a plate, and the Main Galaxy Sample is about 99% complete overall (Strauss et al., 2002). However, this completeness is in terms of those objects that are targeted for spectroscopy in the first place, and in particular is subject to an explicit Petrosian 50% light surface brightness cut of $`\mu _{\mathrm{app}}<24.5\mathrm{mag}\mathrm{arcsec}^2`$ for all objects. Blanton et al. (2005c) show that the faint end is incomplete. We therefore calculate the sampling rate for each bin in the bivariate LF and correct for this in the estimation of the LF (see §3).
Because we calculate the sampling rate, the galaxies used in the process of obtaining our results do not necessarily have spectra. We therefore apply the set of recommended photometric flags described at http://www.sdss.org/dr4 to generate a clean sample. We also require the resolve\_status flag to be survey\_primary and the vagc\_select flag to be main.
For spectra, as mentioned, the VAGC Main-like sample cuts are slightly more inclusive than the SDSS Main sample, in particular allowing in some binary stars by the removal of the cut for small bright objects. Here we reapply the Main criteria as given in Strauss et al. (2002). For objects with spectra we also require the specprimary flag to be set, the primtarget flag to be set to either galaxy, galaxy\_big, or galaxy\_bright\_core, the progname flag to be main, the zwarning flag to be 0 and the platequality flag to be good. The VAGC does not contain a redshift confidence parameter equivalent to the zConf in the CAS, but bad spectra and redshifts are excluded by zwarning.
In the optical, many spiral galaxies that are edge-on or near to this suffer internal extinction and reddening due to dust. Here we attempt to minimize this effect by restricting the sample to galaxies with an axis ratio of less than that for an E7 elliptical galaxy. Galaxies with larger axis ratios than this are likely to be edge-on or close to edge-on spirals. Because the radial light profiles available from the SDSS and VAGC are either axisymmetric or specific to certain types of galaxies (de Vaucouleurs or exponential), we use the $`25\mathrm{mag}\mathrm{arcsec}^2`$ isophotal major and minor axes isoA and isoB in $`r`$ and restrict the sample so that $`\mathrm{isoA}/\mathrm{isoB}<10/3`$. This excludes 30,407 (2.5%) of the galaxies. Fig. 4 of Vincent & Ryden (2005) shows that the vast majority of the bright galaxies removed are those with an exponential or mixed exponential and de Vaucouleurs profile. Very few galaxies with a pure de Vaucouleurs profile are removed. At faint magnitudes similar numbers of galaxies are removed for both profile types, so many more faint ellipticals are removed than bright. Here ‘bright’ depends on the profile type, varying from $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h<21.1`$ for de Vaucouleurs to $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h<19.7`$ for exponential.
A potential alternative measure is the axis ratio derived from the adaptive moments (Stoughton et al., 2002). However, Kuehn & Ryden (2005) show that this would be affected by seeing for most of the galaxies in our sample due to their small sizes and consequent poor resolution. As our purpose is simply to exclude very elongated objects we use the isophotal cut. A similar example is Zibetti et al. (2004), who require $`\mathrm{isoB}/\mathrm{isoA}<0.25`$ in $`g`$, $`r`$ and $`i`$, and $`a>10\mathrm{arcsec}`$ for their sample of edge-on galaxies.
The VAGC also contains eyeball quality checks of spectra for 17,422 objects, mostly bright ($`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h<15`$) or of low redshift ($`z<0.01`$). The sample is not complete in any particular sense, but it does exclude numerous spurious objects such as bad deblends of large galaxies, HII regions, clearly incorrect classifications and so on. We require that the quality flag be set to done with no other bit set, unless use\_anyway is set. The requirement noticeably reduces the numbers of objects in some of the sparsely populated outlying bins of the LF.
Besides the VAGC, we use data from five other sources: 1) the Japan Participation Group (JPG) catalogue of eyeball-classified galaxy morphologies, based on the SDSS Early Data Release (EDR, Stoughton et al., 2002), 2) the neural network morphological types of Ball et al. (2004), 3) the public SDSS DR4 CAS, 4) the Max Planck/Johns Hopkins (MPA/JHU) catalogues for galaxies and AGN at http://www.mpa-garching.mpg.de/SDSS and 5) the Pittsburgh-Carnegie Mellon Value-Added Catalogue (VAC; http://nvogre.phyast.pitt.edu/dr\_value\_added) for galaxies. The CAS gives the eClass spectral type, the MPA/JHU the stellar mass and the VAC the environmental density.
### 2.2 Petrosian and Sérsic Magnitudes
Petrosian magnitudes measure a constant fraction of the total light, in a model-independent manner. They are available in the SDSS in a modified form from that introduced by Petrosian (1976). The Petrosian flux is given by
$$F_\mathrm{P}_0^{N_\mathrm{P}r_\mathrm{P}}2\pi r^{}𝑑r^{}I(r^{})$$
(3)
where $`r_\mathrm{P}`$ is the Petrosian radius, which is the value at which the Petrosian ratio of surface brightnesses
$$R_\mathrm{P}(r)\frac{_{0.8r}^{1.25r}2\pi r^{}𝑑r^{}I(r^{})/[\pi (1.25^20.8^2)]}{_0^r2\pi r^{}𝑑r^{}I(r^{})/(\pi r^2)}$$
(4)
has a certain value, chosen in the SDSS to be 0.2. The number $`N_\mathrm{P}`$ of Petrosian radii within which the flux is measured is equal to 2 in the SDSS. Further details are given in Lupton, Gunn & Szalay (1999) and Stoughton et al. (2002).
The Petrosian radii are not seeing corrected, which causes the surface brightness and concentration to be underestimated for objects of size comparable to the PSF. However the seeing effect is not yet quantified, and there are other approximations such as using a circular as opposed to elliptical aperture and not correcting for dust obscuration. As described in Stoughton et al. (2002) the Petrosian aperture is not missing much flux compared to an ideal galaxy light profile, the amount missing being about 20% for a de Vaucouleurs profile and only 1% for an exponential profile. The effect of seeing, which would make the profiles tend towards a PSF, for which 5% of the light is lost, is also small for galaxies in the main sample. Blanton et al. (2003c) show that the resulting luminosity density in $`r`$ is very similar to that from the Sérsic profile, the difference being $`j_{0.1}(\mathrm{Sersic})=j_{0.1}(\mathrm{Petrosian})0.03`$. They suggest that the similarity shows that the true luminosity density is also of a similar value.
The VAGC also contains Sérsic fits to the galaxy light profiles. The Sérsic profile (Sérsic, 1968; Graham & Driver, 2005) is obtained by generalising the de Vaucouleurs profile (de Vaucouleurs, 1948) to have index $`\frac{1}{n}`$ instead of $`\frac{1}{4}`$, giving
$$I(r)=I_0\mathrm{exp}\{b_n[(r/r_\mathrm{e})^{1/n}]\},$$
(5)
where $`b_n`$ is such that half the total luminosity is within $`r_e`$. The fitting procedure used in the VAGC is described in Blanton et al. (2005a) and in more detail in the Appendix of Blanton et al. (2005b). The axisymmetric Sérsic profile is fitted to the azimuthally averaged galaxy light profile available from the SDSS database. The profiles are corrected for seeing, which is modelled using three axisymmetric Gaussians. The de Vaucouleurs and exponential profiles correspond to $`n=4`$ and $`n=1`$ respectively.
Graham et al. (2005) show the discrepancy between the Petrosian and Sérsic magnitudes. In particular, the difference depends on the shape of the galaxy light profile. Their table 1 for the SDSS Petrosian aperture shows that for inverse concentration index ($`CI_{\mathrm{inv}}`$; §2.4), the difference is negligible above $`CI_{\mathrm{inv}}=0.35`$, but below this increases to, where $`\mu _e`$ is the total effective half light surface brightness, $`\mu _e\mu _{\mathrm{app}}=0.52`$, 1.09 and 1.74 $`\mathrm{mag}\mathrm{arcsec}^2`$ for $`CI_{\mathrm{inv}}=0.30`$, 0.28 and $`0.27`$ respectively for mean half-light surface brightness, and $`m_{\mathrm{Petrosian}}m_{\mathrm{S}\stackrel{´}{\mathrm{e}}\mathrm{rsic}}=0.22\mathrm{mag}`$, $`0.38\mathrm{mag}`$ and $`0.54\mathrm{mag}`$ in magnitude over the same range. Thus only a few percent of our galaxies are affected significantly. Their fig. 6 shows that the VAGC Sérsic fits are consistent with the expected differences. The $`CI_{\mathrm{inv}}`$ of 0.28 corresponds to a Sérsic index of 6.
Although the Sérsic index is an improvement in that it fits a more accurate radial light profile to a galaxy and is seeing-corrected, there are still some biases present. In their fig. 9, Blanton et al. (2005a) show the residuals of their Sérsic fits for 1,200 simulated galaxies. As a function of $`n`$, the value of $`n_{out}n_{in}`$, where $`n_{out}`$ is the fitted value of $`n`$, decreases monotonically from less than -0.05 at $`n=1`$ to -0.1, -0.25, -0.5 and -0.65 at $`n=2,3,4`$ and 4.5 respectively. Thus a de Vaucouleurs galaxy ($`n=4`$) is assigned an index of $`n3.5`$ and the indices are all systematically underestimated. The error of 0.5 is comparable to the bin size in the same area of our bivariate luminosity-Sérsic index distribution. The range of the error as shown by the quartiles also broadens from around 0.1 to 0.5 over the same range. The fitted 50% light radius and flux decrease similarly over their range to around 0.8 and 0.9 of their true values. Also, the fitted radius and flux decrease with $`n`$, the fitted flux with radius and the fitted radius with flux. However, the fitted $`n`$ does not decrease with radius and flux. The overall performance for $`n`$ is characterized as good because the bias is similar to the uncertainty and comparable but opposite to that from the assumption of axial symmetry. The overall fits are thought to be approximately correct for Sérsic-shaped galaxies and supply a seeing-corrected estimate of size and concentration for the others.
Here we investigated the LFs for both Petrosian and Sérsic profiles. The LFs are generally similar and because the samples used are based on the SDSS Main Galaxy Sample, which is based on Petrosian magnitudes, we present the results based on those magnitudes. The derivation of a fully Sérsic-based sample is beyond the scope of this paper.
### 2.3 Neural Network Morphological Types
1,875 SDSS galaxies to $`r<15.9`$ in the SDSS EDR have been classified into morphological types by Nakamura et al. (2003), forming the JPG catalogue. The system used was a modified version of the T-type system (de Vaucouleurs, 1959), with the types being assigned in steps of 0.5. The corresponding Hubble types are E=0, S0=1, Sa=2, Sb=3, Sc=4, Sd=5 and Im=6.
We have previously shown (Ball et al., 2004) that artificial neural networks (e.g. Bishop, 1995; Lahav et al., 1996) are able to assign types to galaxies in the SDSS with an RMS accuracy of 0.5 on this scale, using the JPG catalogue as a training set. This is the same as the spread between the human-classified types by the members of the JPG team.
We used the same procedures here, updated for DR4 over DR1 and with some modifications. The DR4 CAS was matched to the JPG catalogue using a tolerance of 2 arcsec and a flux limit of $`r<15.9`$. Unassigned types ($`1`$) and galaxies flagged as being likely to have bad photometry were removed as before. The clean photometry flags described above were also applied here. After these cuts 1,290 galaxies remained, 1,131 with spectroscopic redshifts. The training and test sets were 645 each with no overlap, from which, as before, galaxies with severely outlying parameters and targets ($`>10\sigma `$ from the mean value for the parameter, generally indicative of a measurement error) were iteratively removed for each parameter in turn. The network trained on the 645 was applied to the DR4 sample using the CAS photometry. One could also retrain on all 1,290 for the final network to be applied to the sample, but it would make very little difference.
It was also found that when trained on the full set of 29 parameters in Ball et al. (2004) the types were biased towards early type with increasing redshift to a greater extent than in the JPG catalogue. However, when one trains on the purely morphological subset of the parameters (all those except the magnitudes and colours), the bias is similar to the JPG, reflecting the flux-limited nature of the sample, but the RMS variation between network type and target type is not significantly larger. Hence the latter set is used as the training set here.
The network also has a tendency to ‘avoid the ends of the scale’, resulting in few very late type galaxies and a slight bias away from early type galaxies. This is shown in Fig. 1 of Ball et al. (2004). This is likely due to the severe lack of galaxies in our training set of type $`T=5`$ or later compared to the earlier types, and the importance of the concentration index in the morphological training set, which shows a similar deviation away from the ends of the scale for both early and late types.
We restrict the absolute magnitude range of the morphologies presented to $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h<18`$. This is due to a lack of galaxies in the training data at fainter magnitudes than this. Thus the neural network types are not extrapolated from the training set in apparent or absolute magnitude.
### 2.4 SDSS DR4 VAGC Data
The bivariate LF parameters described in this section are the inverse concentration index, Sérsic index, absolute effective surface brightness, reference-frame colours and absolute radius. The morphological parameters are measured in the $`r`$ band, since this band is used to define the aperture through which Petrosian flux is measured for all five bands.
The inverse concentration index $`CI_{\mathrm{inv}}`$ is $`R_{50}/R_{90}`$ where $`R_{50}`$ and $`R_{90}`$ are the radii within which 50 and 90 per cent of the Petrosian flux is received. The inverse is used because it has the range 0–1.
The Sérsic index $`n`$ is monotonically related to the concentration (e.g. Graham et al., 2005), if the latter is measured using Sérsic radii. Here we use Petrosian circular radii and for the bivariate luminosity-Sérsic index distribution we use $`\mathrm{log}n`$, as the index is in the exponent in the equation defining the Sérsic profile (equation 5).
The absolute effective surface brightness used here is given by
$$\mu _{\mathrm{R}_{50}}=m_r+2.5\mathrm{log}(2\pi R_{50}^2)10\mathrm{l}\mathrm{o}\mathrm{g}(1+z)K,$$
(6)
where $`m_r`$ is the $`r`$ band magnitude and $`K`$ is the K-correction. The evolutionary correction term is set to zero.
The reference frame colours are taken from the K-corrected values described above. All ten colours from $`u`$, $`g`$, $`r`$, $`i`$ and $`z`$ were investigated.
The absolute radius of a galaxy at redshift $`z_{\mathrm{gal}}`$ is calculated from the apparent Petrosian radius using the $`\mathrm{\Omega }_{\mathrm{matter}}=0.3`$, $`\mathrm{\Omega }_\mathrm{\Lambda }=0.7`$, $`h=1`$ cosmology. It is also shown logarithmically and is given by
$$R_{\mathrm{abs}}=\frac{R_{90}d(z)}{1+z},$$
(7)
where $`d`$ is the relativistic coordinate distance
$$d=\frac{c}{H_0}_0^{z_{\mathrm{gal}}}\left[(1\mathrm{\Omega }_\mathrm{\Lambda })(1+z)^3+\mathrm{\Omega }_\mathrm{\Lambda }\right]^{0.5}.$$
(8)
### 2.5 Outputs from additional SDSS Catalogues
The additional catalogues used are all based on SDSS DR4 and use the same Euclidean space, $`\mathrm{\Omega }_{\mathrm{matter}}=0.3`$, $`\mathrm{\Omega }_\mathrm{\Lambda }=0.7`$, as used here. The quantities all require spectra to be calculated and are therefore matched using the unique spectroscopic identification, both in object and time of observation, provided by the quantities plate, mjd and fiberid.
The eClass (Connolly et al., 1995; Connolly & Szalay, 1999; Yip et al., 2004) is a continuous one-parameter type assigned from the projection of the first three principal components (PCs) of the ensemble of SDSS galaxy spectra. The locus of points forms an approximately one dimensional curve in the volume of PC1, PC2 and PC3. This is a generalization of the mixing angle $`\varphi `$ in PC1 and PC2
$$\varphi =\mathrm{tan}^1\left(\frac{a_2}{a_1}\right),$$
(9)
where $`a_1`$ and $`a_2`$ are the eigencoefficients of PC1 and PC2. The range is from approximately $`0.6`$, corresponding to early type galaxies, to 1, late type. The eClass is also robust to missing data in the spectra used for its derivation, and is almost independent of redshift. The quantity is not given in the VAGC and so was matched to the public SDSS DR4 data from the CAS.
The MPA/JHU value-added catalogues at http://www.mpa-garching.mpg.de/SDSS include the quantity rml50, the logarithm of the median dust-corrected stellar mass in solar units. It is calculated from the stellar mass to light ratio predicted by a large library of models of star formation history and is described further in Kauffmann et al. (2003).
The VAC contains a measure of galaxy environment as the parameter density\_z005095. This is the distance to the $`N`$th nearest neighbour within $`\pm 1000\mathrm{km}\mathrm{s}^1`$ in redshift for $`0.053<z<0.093`$. The $`\pm 1000\mathrm{km}\mathrm{s}^1`$ is used to minimize contamination from interlopers. Galaxies for which the survey edge is reached before the $`N`$th nearest neighbour are excluded to avoid downward bias in the estimated densities near the survey edges. Here the value of $`N`$ used is 5, following Balogh et al. (2004) who choose this value to approximate Dressler (1980) who uses $`N=10`$ before correction for superimposed galaxies. The surface density of galaxies is then given by $`\mathrm{\Sigma }_N=N/\pi d_N^2`$. The nearest neighbour must be of magnitude $`r<17.7`$ at $`z=0.093`$, which corresponds to $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h<19.63`$. This cut ensures a uniform density measurement over the redshift range of the sample rather than the density decreasing with redshift due to missing faint galaxies. The value of $`h`$ used in preparing this catalogue was $`h=0.7`$ so the densities are given here in units of $`h_{70}^2\mathrm{Mpc}^2`$.
## 3 Bivariate Luminosity Functions
Unlike the monovariate LF in which the Schechter function is often used, there is no common functional form for the bivariate LF, so a nonparametric estimator is chosen. We use the popular stepwise maximum likelihood (SWML) method of Efstathiou et al. (1988), following the extension to a bivariate distribution by Sodré & Lahav (1993). The maximum likelihood method has well defined error properties (Kendall & Stuart, 1961) and in the comparison between LF estimators of Takeuchi et al. (2000) the SWML was shown to be a good estimator.
The SWML gains independence of inhomogeneities in the galaxy distribution by assuming that this is the case via the universal form: $`n(L,𝐱)=\varphi (L)\rho (𝐱)`$. This means that the $`\varphi (L)`$ shape is independent of its normalisation, which then has to be found separately. The data must also be binned over the ranges of the parameters considered.
Here the bivariate LF is given by $`\psi (L,X)`$ or $`\psi (M,X)`$ where $`X`$ is the second parameter.
For a galaxy which is observable if its luminosity $`L`$ lies in the range $`L_{\mathrm{min}}`$$`L_{\mathrm{max}}`$ and $`X`$ lies in the range $`X_{\mathrm{min}}`$$`X_{\mathrm{max}}`$ (these limits in general being redshift dependent), the probability of seeing that galaxy with luminosity $`L_i`$ and $`X=X_i`$ at redshift $`z_i`$ is
$$\begin{array}{c}p_i\psi (L_i,X_i)f(L_i,X_i)\hfill \\ \hfill /_{L_{\mathrm{min}}(z_i)}^{L_{\mathrm{max}}(z_i)}_{X_{\mathrm{min}}(z_i)}^{X_{\mathrm{max}}(z_i)}\psi (L,X)f(L,X)dLdX,\end{array}$$
(10)
where $`f`$ is the completeness function, calculated for each bin in the bivariate LF. This automatically takes into account both the 6% sampling incompleteness resulting from galaxies more closely spaced than the spectroscopic fibres, and the further 1% from objects missed in the Main Galaxy Sample (see §2.1). It does not take into account any incompleteness in the imaging itself, which is discussed further in Appendix A below. The $`X`$ limits are a function of $`z`$ if the sample is explicitly or implicitly selected on $`X`$. The samples are not explicitly selected on $`X`$ if the quantity requires a spectrum.
The likelihood $`=p_i`$ is then maximised with respect to $`\psi (L,X)`$. In practice, one maximises the log-likelihood $`\mathrm{ln}=\mathrm{ln}p_i`$.
For the SWML method, the discrete version of this is used. $`\psi (M,X)`$ is parametrized as the number density of galaxies
$$\psi (M,X)=\psi _{jk}(j=1\mathrm{}N_M;k=1\mathrm{}N_X)$$
(11)
in $`N_M`$ and $`N_X`$ evenly spaced bins in absolute magnitude $`M_j^{}<M_j<M_j^+`$ and $`X_k^{}<X_k<X_k^+`$, where
$$M_j^\pm =M_j\pm \frac{\mathrm{\Delta }M}{2}$$
(12)
and
$$X_k^\pm =X_k\pm \frac{\mathrm{\Delta }X}{2}.$$
(13)
Magnitude bins are preferred to luminosity bins as the distribution is more evenly spread.
The likelihood is then given using the discrete equivalent of equation 10
$$\begin{array}{c}\mathrm{ln}=\underset{i=1}{\overset{N_g}{}}\underset{j=1}{\overset{N_M}{}}\underset{k=1}{\overset{N_X}{}}W_{ijk}\mathrm{ln}[\psi _{jk}f(M_i,X_i)]\hfill \\ \hfill \underset{i=1}{\overset{N_g}{}}\mathrm{ln}\left(\underset{j=1}{\overset{N_M}{}}\underset{k=1}{\overset{N_X}{}}H_{ijk}\psi _{jk}\right)+\mathrm{constant},\end{array}$$
(14)
where
$$W_{ijk}=\{\begin{array}{cc}1\hfill & \mathrm{if}\mathrm{\Delta }M/2M_iM_j<\mathrm{\Delta }M/2\hfill \\ & \mathrm{and}\mathrm{\Delta }X/2X_iX_k<\mathrm{\Delta }X/2,\hfill \\ 0\hfill & \mathrm{otherwise}\hfill \end{array}$$
(15)
and
$$H_{ijk}=\frac{1}{\mathrm{\Delta }M\mathrm{\Delta }X}\times _M^{}^M^{}_X^{}^X^{}f(M,X)𝑑M𝑑X,$$
(16)
where $`M^{}=\mathrm{max}[M^{},\mathrm{min}(M^+,M_{\mathrm{lim}}^i)]`$ and $`X^{}=\mathrm{max}[X^{},\mathrm{min}(X^+,X_{\mathrm{lim}}^i)]`$.
The constant in (14) is fixed by using a Lagrangian multiplier $`\lambda `$: the constraint
$$g=\underset{j}{}\underset{k}{}\psi _{jk}\mathrm{\Delta }M\mathrm{\Delta }X1=0$$
(17)
is applied and the new likelihood $`\mathrm{ln}^{^{}}=\mathrm{ln}+\lambda g(\psi _{jk})`$ is maximised with respect to $`\psi _{jk}`$ and $`\lambda `$, requiring $`\lambda =0`$.
The maximum likelihood estimate $`\mathrm{ln}^{^{}}/\psi _{jk}=0`$ is then
$$\psi _{jk}=\frac{_{i=1}^{N_g}W_{ijk}}{_{i=1}^{N_g}[H_{ijk}/_{l=1}^{N_M}_{m=1}^{N_X}\psi _{lm}H_{ilm}]},$$
(18)
where $`\psi _{lm}`$ is from the previous iteration.
Following Efstathiou et al., the errors on the parameters are estimated using the fact that the MLM estimates $`\psi _{jk}`$ are asymptotically normally distributed with the covariance matrix
$$\mathrm{cov}(\psi _{jk})=𝖨^1(\psi _{jk}),$$
(19)
where $`𝖨`$ is the information matrix (their equation 2.13b).
For the normalisation, the estimator of the space density $`\overline{n}`$ of galaxies is
$$f\overline{n}=\underset{i=1}{\overset{N_{gal}}{}}w(z_i)/_{z_{min}}^{z_{max}}S(z)w(z)𝑑V.$$
(20)
The selection function $`S`$ from $`L_1`$ to $`L_2`$ is
$$S(z)=\underset{L_1^{}}{\overset{L_2^{}}{}}\underset{X_1^{}}{\overset{X_2^{}}{}}\psi (L,X)/\underset{L_1}{\overset{L_2}{}}\underset{X_1}{\overset{X_2}{}}\psi (L,X),$$
(21)
where $`L_1^{}=\mathrm{max}(L_{\mathrm{min}}(z),L_1)`$, $`L_2^{}=\mathrm{min}(L_{\mathrm{max}}(z),L_2)`$ and similarly for $`X`$. The limits on the sums in the numerator depend on the K-correction, which is determined using the average SED for the sample. These limits will in general include partial bins with appropriate weighting due to the summation limits not falling at the edge of a bin.
We use the weighting function
$$w(z)=\frac{1}{[1+4\pi f\overline{n}J_3(r_c)S(z)]},$$
(22)
where
$$J_3(r_c)=_0^{r_c}r^2\xi (r)𝑑r,$$
(23)
$`r`$ is the distance and $`\xi (r)`$ the real space galaxy two point correlation function.
The variance in the correlation function can be estimated using a complex formula which minimizes the variance in the function and involves the three and four point correlation functions. These are more difficult to measure that the two point and require a large sample such as the SDSS. The weight function in equation 22 was found to be a good approximation by Hamilton (1993) and is widely used. The function minimizes the variance in the estimate of the number density so long as $`r_c`$ is much less than the survey depth. Here $`4\pi J_3=32000h^3\mathrm{Mpc}^3`$, using the $`\xi (r)`$ from Zehavi et al. (2002) which is given as $`\xi (r)=(r/6.1\pm 0.2h^1\mathrm{Mpc})^{1.75\pm 0.03}`$ for $`0.1h^1\mathrm{Mpc}r16h^1\mathrm{Mpc}`$. The effect of using an incorrect value would be to change the normalisation but not the shape of the LF. For example, Loveday (2004) uses the same values of $`J_3`$ and $`\xi `$ as here and finds that halving $`J_3`$ reduces the estimated density by 7%. An update to Zehavi et al. (2002), Zehavi et al. (2004b), has a similar $`\xi `$ value to that used here.
Equation 20 is solved iteratively as it contains $`w`$, which contains $`\overline{n}`$. The minimum-variance $`w(z)`$ is described further in Davis & Huchra (1982) and Hamilton (1993).
## 4 Results
### 4.1 Morphological Type
The LF bivariate with the JPG catalogue morphological type $`T_{\mathrm{JPG}}`$ obtained here is shown in Fig. 1 and that with the ANN morphological type $`T_{\mathrm{ANN}}`$ in Fig. 2. The corresponding completeness maps (number of galaxies with spectra / number of galaxies, in each bin) are shown in Figs. 16 and 17. There are no obvious biases within the regime for which galaxies are present, the completeness being consistent with 90% for the full range of apparent magnitude for $`T_{\mathrm{JPG}}`$ and 60% for $`T_{\mathrm{ANN}}`$, the latter being expected from the use of the whole survey area which is only about 70% complete in spectroscopy (Appendix A). Fig. 17 again shows the lack of types later than 4.5 assigned by the ANN (see §2.3). If the training set without the clean photometry flags were used, hence leaving in more late type galaxies, the $`\psi `$ greyscale would extend further right by approximately the width of one of the bins, so the LF for very late types would still not seen.
The bivariate luminosity-morphology distribution was computed for the same JPG catalogue and classification scheme by Nakamura et al. (2003). However, whilst being drawn from the same catalogue there are some differences in the method, leading to an altered final sample for the LF. A different subset of the EDR galaxies is used here, with 1,150 instead of 1,482; the sample is divided into smaller bins and the normalisation is without the explicit redshift cut used in Nakamura et al. (2003). The smaller value is 1,150 due to the masking applied to the datasets here. Of these, 1,019 have redshifts and 998 are selected for the LF.
Nakamura et al. show that the overall LF is well fit by a Schechter function with parameters $`\alpha =1.10`$, $`M^{}=20.65`$ and $`\varphi =0.0143`$. We obtain $`\alpha =1.30`$, $`M^{}=20.58`$ and $`\varphi =0.0098`$. Their LF is divided into the ranges $`0<T_{\mathrm{JPG}}<1.5`$, $`1.5<T_{\mathrm{JPG}}<3`$, $`3.5<T_{\mathrm{JPG}}<5`$ and $`5.5<T_{\mathrm{JPG}}<6`$. For 0–1 and 3.5–5 the faint-end slope is shallower, at $`\alpha =0.83`$ and $`\alpha =0.71`$. 1.5–3 is steeper, at $`\alpha =1.15`$. The early types are also brighter, as expected, being $`M^{}=20.75`$ compared to $`M^{}=20.30`$ for types later than 1. In our LFs there is no clear trend between the types, although the binning is narrower and the earliest type bin appears brighter and with a shallower faint-end slope as expected.
In the LF split by $`M_r`$, the main trend is for higher normalisation at fainter magnitudes, as expected for increased numbers of fainter galaxies. For bright galaxies, the normalisation with $`T_{\mathrm{JPG}}`$ increases from $`\varphi 10^4`$ to $`\varphi 5\times 10^3`$ between types 1 and 0 but otherwise is consistent with a flat distribution across the range in $`T_{\mathrm{JPG}}`$ at all magnitudes, with a possible decrease at types later than $`T_{\mathrm{JPG}}=4`$.
The neural network types enable the LF sample to be enlarged from 998 to 20,891 galaxies. This is the first time the LF with morphology assigned in this way to the resolution of individual Hubble types has been calculated for a sample of galaxies of this size. The trends hinted at in Fig. 1 become clearer, with the types earlier than $`T_{\mathrm{ANN}}=0.5`$ now showing a declining faint-end slope. The $`0.5<T_{\mathrm{ANN}}<1.5`$ are as bright but more abundant at fainter magnitudes. The spirals from $`T_{\mathrm{ANN}}=1.5`$ to $`T_{\mathrm{ANN}}=3.5`$, corresponding to S0a to Sbc, then show no difference in their LFs. The late types to 4.5 are then slightly fainter again, with a slightly steeper faint-end slope. Each slice appears consistent with a Schechter function. For the LF split by $`M_r`$, the galaxies brighter than $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h=21`$ are now clearly more abundant at $`T_{\mathrm{ANN}}<1`$, with a weaker trend seen for $`21<M_r5\mathrm{l}\mathrm{o}\mathrm{g}h<20`$. The rest to $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h=18`$ are still consistent with a flat distribution, except at $`T_{\mathrm{ANN}}<0.5`$ where they decline. The changes at the early type end are at least as strong as those indicated, as the ANN bias here is, as mentioned, away from early types and not towards. There are very few galaxies later than $`T_{\mathrm{ANN}}=2.5`$ and brighter than $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h=22`$.
In de Lapparent (2003), it is suggested that cross-contamination between Hubble types will make the LFs tend towards Schechter functions even if the underlying LFs for the giant (as opposed to dwarf) galaxies are Gaussian. Here the $`0.5<T_{\mathrm{ANN}}<0.5`$ bin in particular could be suffering from this effect if the true distribution were Gaussian brighter than $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h=20`$ and the fainter magnitudes were contaminated by dwarfs. Nakamura et al. (2003) do not find this to be a significant effect in their results as their fainter galaxies do not have the softer cores of later type dwarfs. However, the Gaussian LF is supported by fig. 1 of Bernardi et al. (2003), who show LFs for ellipticals with a turnover at $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h20.5`$. The LFs for the later types will similarly contain dwarf and giant galaxies and will therefore also appear Schechter-like.
Several recent papers discuss blue dwarf spheroidal galaxies. These have the same concentrations and smooth appearance as dwarf ellipticals, but are much bluer in colour. Whilst much more prevalent at higher redshifts (e.g. Ilbert et al., 2006), for local galaxies these objects are discussed by Driver et al. (2006). In their fig. 6, they show that when the blue spheroids are removed, the E/S0 LF gains a much more rapidly declining faint-end slope, with $`\alpha `$ changing from -0.9 to -0.4. This could be addressed in our work by adding colour as a classification criterion, moving towards a spectro-morphological classification.
It may also be the case that the morphological types are not well defined below a resolution of about 0.5 T types and will therefore inevitably be cross-contaminated. However with the large sample here the LF differences are significant, as indicated by the sizes of the error bars.
Another recent determination of the LF bivariate with Hubble type is presented in Read & Trentham (2005). This shows LFs following the same trends as seen here, and with only those for types Sa and Sb being similar.
### 4.2 Inverse Concentration Index
Various papers (e.g. Shimasaku et al., 2001) have shown that the inverse concentration index $`R_{50}/R_{90}`$ ($`CI_{\mathrm{inv}}`$) is correlated with the morphological type, and indeed can be used as a simple classifier. Here we show the LF bivariate with concentration index in Petrosian radii, using the same $`R_{50}/R_{90}`$ ratio.
Shimasaku et al. show that, whereas the $`CI_{\mathrm{inv}}`$ does correlate with morphology, the correlation is not perfect. In particular, for types E and S0, the index varies from around 0.3 to 0.35, whereas it increases to almost 0.5 to type Sc. It then stays in this range for even later types.
The bivariate luminosity-$`CI_{\mathrm{inv}}`$ distribution is shown in Fig. 3 and the completeness map in Fig. 18. The expected broad trend of fainter galaxies being less concentrated is seen. Subdivided by $`CI_{\mathrm{inv}}`$ the LF resembles a Schechter function with a steeper faint-end slope for less concentrated galaxies. For the bins outside $`0.3<CI_{\mathrm{inv}}<0.6`$ the normalisation is lowered due to lack of galaxies. The completeness similarly becomes very low and noisy outside this range. For $`0.3<CI_{\mathrm{inv}}<0.4`$ the LF shows an upturn at magnitudes fainter than $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h=18`$. This may be related to the prevalence of dwarf ellipticals in clusters and is discussed further below.
Subdivided by $`M_r`$, at the bright end, $`24<M_r5\mathrm{l}\mathrm{o}\mathrm{g}h<21`$, the $`CI_{\mathrm{inv}}`$ shows a peak at $`CI_{\mathrm{inv}}0.31`$. A pure de Vaucouleurs profile corresponds to $`CI_{\mathrm{inv}}=0.3`$, therefore this peak corresponds to the population of ellipticals. The slight offset above 0.31 may be due to the bias towards the $`CI_{\mathrm{inv}}`$ for a PSF, which is 0.5 (Gaussian), due to many of the galaxies in the sample being poorly resolved, in the sense that their isophotal axis ratio rather than their shape measure is used (§2.1). However, this region is also populated by S0 galaxies. At $`21<M_r5\mathrm{l}\mathrm{o}\mathrm{g}h<19.5`$, the peak at low $`CI_{\mathrm{inv}}`$ becomes broader and less pronounced, increasing to $`CI_{\mathrm{inv}}0.33`$. At $`19.5<M_r5\mathrm{l}\mathrm{o}\mathrm{g}h<18`$ the distribution becomes more symmetrical, peaking around $`CI_{\mathrm{inv}}0.38`$. At $`18<M_r5\mathrm{l}\mathrm{o}\mathrm{g}h<16.5`$ the distribution is approximately symmetrical around $`CI_{\mathrm{inv}}=0.43`$, which corresponds to an exponential profile. Besides the bright peak, the distributions are not otherwise obviously bimodal, just broad.
The completeness increases from $``$ 40% to 50% over the range $`0.3<CI_{\mathrm{inv}}<0.4`$ and remains at this level until $`CI_{\mathrm{inv}}0.55`$. This may reflect the bias due to the closest possible spacing of 55 arcsec for SDSS spectroscopic fibres (§2): more galaxies in environmentally dense regions, which tend to be early type and therefore of $`CI_{\mathrm{inv}}`$ in this range, will be missed. If this is the case then the peak in low $`CI_{\mathrm{inv}}`$ seen should be slightly more pronounced than shown.
Blanton et al. (2001) present the LF bivariate with the Petrosian $`CI_{\mathrm{inv}}`$ for the SDSS EDR. The same broads trends as seen here are present: a peak near $`CI_{\mathrm{inv}}=0.3`$ corresponding to galaxies with a de Vaucouleurs profile, particularly evident for the brightest galaxies at $`23.5<M_r5\mathrm{l}\mathrm{o}\mathrm{g}h<22`$, then becoming less prominent as the distribution broadens to spread as far as $`CI_{\mathrm{inv}}0.5`$ at $`20.5<M_r5\mathrm{l}\mathrm{o}\mathrm{g}h<19.0`$ and becoming symmetrical about $`CI_{\mathrm{inv}}0.43`$, the exponential profile, at $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h<17.5`$. The de Vaucouleurs peak is offset slightly above 0.3 in a similar way to our results. Unlike our LF, the normalisation split by $`M_r`$ increases then decreases, whereas ours increases monotonically from bright to faint, presumably due to a different method of normalisation. Subdivided by $`CI_{\mathrm{inv}}`$, the LFs show the usual trend of bright and shallow faint-end to dim and steep faint-end. Compared to our results which cut in axis ratio, there is no obvious difference in the widths of the distributions in $`CI_{\mathrm{inv}}`$.
Nakamura et al. (2003) also present the bivariate luminosity-$`CI_{\mathrm{inv}}`$ distribution for their JPG sample described in §4.1, dividing the sample at $`CI_{\mathrm{inv}}=0.35`$ into early and late types. As with our results above $`CI_{\mathrm{inv}}=0.3`$, the LF does not show a marked decline at the faint end. They show that their sample is not dominated at the faint end ($`M_r\mathrm{}>19`$) by dwarf ellipticals, which have softer cores.
### 4.3 Sérsic Index
The seeing-corrected Sérsic index is monotonically related to the Sérsic-radius-based concentration index (Graham et al., 2005) and as such should show similar trends to those for the Petrosian-based $`CI_{\mathrm{inv}}`$ presented in §4.2. However, unlike our concentration index, the Sérsic index is seeing-corrected. We plot the index logarithmically as explained in §2.4. The bivariate LF is shown in Fig. 4 and the completeness map in Fig. 19. The indices again correspond to profile types, with $`\mathrm{log}n=0`$ ($`n=1`$) being exponential and $`\mathrm{log}n=0.6`$ ($`n=4`$) de Vaucouleurs. The completeness decreases from around 55% at $`\mathrm{log}n0.2`$ to 40% at $`\mathrm{log}n=0.65`$. Again this could be due to bias against early types due to fibre spacing.
Subdivided by $`M_r`$, the distribution is clearly bimodal to fainter magnitudes than seen for $`CI_{\mathrm{inv}}`$, down to $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h=18`$. The high-$`n`$ peak is around $`\mathrm{log}n=0.65`$ for $`24<M_r5\mathrm{l}\mathrm{o}\mathrm{g}h<21`$, higher than de Vaucouleurs galaxies. For $`21<M_r5\mathrm{l}\mathrm{o}\mathrm{g}h<19.5`$ the peak shifts lower, to $`\mathrm{log}n0.575`$. Below $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h=19.5`$, the low-$`n`$ peak becomes prominent and is centred at $`\mathrm{log}n0.15`$. Again this is higher than the exponential profile of $`\mathrm{log}n=0`$. At this value, the number of galaxies has dropped sharply, although the completeness remains flat until $`\mathrm{log}n0.2`$.
Subdivided by $`n`$, the LFs are again Schechter-like and go from bright and shallow faint-end to dim and steep faint-end. As with $`CI_{\mathrm{inv}}`$, the faint end of the LF for high $`n`$, whilst decreasing, does not resemble a Gaussian. Thus if contamination by dwarfs affects the faint-end LF for morphology and $`CI_{\mathrm{inv}}`$, it is also at work for $`n`$. The LF turnup at $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h\mathrm{}<18`$ seen for $`CI_{\mathrm{inv}}`$ is less prominent here, merely hinted at for $`0.3<\mathrm{log}n<0.8`$.
As described in §2.4, the indices are systematically offset, so that a true de Vaucouleurs galaxy with $`\mathrm{log}n=0.6`$ is here assigned $`\mathrm{log}n0.55`$. However, the value of $`\mathrm{log}n`$ for a PSF is -0.3 ($`n=0.5`$), so this effect may again be at work, although the PSF is convolved in the Sérsic model fitting so this seems unlikely.
The bivariate luminosity-$`\mathrm{log}n`$ distribution is also presented by Driver et al. (2006) for the MGC. There it is subdivided into $`n2`$ and $`n<2`$ ($`\mathrm{log}n=0.3`$) and fitted with a Schechter function. The respective parameters change from $`\alpha =1.25`$ to $`\alpha =0.66`$, $`M^{}=19.48`$ to $`M^{}=19.35`$ and $`\varphi =0.0129`$ to $`\varphi =0.0087`$. The high-$`n`$ LF again is Schechter-like. The less obvious distinction in their two distributions than those seen in their LFs for the MGC continuum measure and colour is ascribed at least in part to dwarf ellipticals, which spread across the $`\mathrm{log}n=0.3`$ boundary. As with our results from §4.1, colours are needed, which they go on to consider, finding a clear bimodality in the bivariate distribution of core $`ur`$ colour and $`n`$. This and their spread in Sabc morphologies between the two peaks is argued as evidence for bulges and discs being the fundamental components and is discussed further in the context of our results below.
### 4.4 Surface Brightness
The bivariate luminosity-surface brightness (SB) distribution has been investigated by several previous authors and is also known as the bivariate brightness distribution (BBD). It is important as it allows one to better quantify selection effects in a survey. Here we compare our results to those of Blanton et al. (2001), Shen et al. (2003) and Blanton et al. (2005c) from the SDSS and Driver et al. (2005) from the Millennium Galaxy Catalogue. Further recent studies at low redshift in the optical are de Jong & Lacey (2000), Cross et al. (2001) and Cross & Driver (2002). The former studies a more local sample of spirals in more detail and the latter two contain slightly earlier results than the MGC, from the 2dFGRS. The SDSS studies use similar data and the same passbands as our data. The MGC represents the current state-of-the-art in the optical by probing to a deeper limiting surface brightness of $`26\mathrm{mag}\mathrm{arcsec}^2`$ in the $`B`$ band over smaller but still substantial ($`37.5\mathrm{deg}^2`$) area of sky. The broad trends here and in other recent work are in agreement with earlier work (e.g. Freeman, 1970; Phillipps & Disney, 1986, etc.), but are seen at higher signal to noise.
In the SDSS, the Petrosian and Sérsic radii are derived from axisymmetric profiles. As neither the Petrosian 2.4, Blanton et al., 2003c) nor the Sérsic apertures are missing much flux for most galaxies, the SBs will therefore be systematically overestimated as the semimajor axis of a half-light elliptical aperture is always at least as large. The MGC study uses elliptical apertures.
The bivariate LF is shown in Fig. 5 and the completeness map for apparent surface brightness is shown in Fig. 20. Also shown are contours at $`10^2`$$`10^7h^3\mathrm{Mpc}^3`$ which show the regions inside which these volumes were probed by the survey. Note that the axes of the top panel are extended relative to the LF slices shown in the lower two panels. This shows that normalisation of LF drops before the volume becomes small due to limiting surface brightness and that the volume at $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h<15`$, for which the LF is quoted, is at least of order $`10^5h^3\mathrm{Mpc}^3`$. As a function of $`\mu _{\mathrm{app}}`$, the completeness varies between 45% and 60% for $`19\mathrm{}<\mu _{\mathrm{app}}\mathrm{}<23`$ and outside this range becomes noisy and drops to being consistent with zero at $`\mu _{\mathrm{app}}17`$ and $`\mu _{\mathrm{app}}24`$. As a function of apparent magnitude, the completeness is between 55% and 60% in the range $`12\mathrm{}<r\mathrm{}<16.5`$ and drops outside this to 40% at $`r=17.77`$, and low values at the bright end, consistent with zero brighter than $`r11.5`$. There is a clear lack of faint high surface brightness or bright low surface brightness galaxies. Fig. 20 shows the incompleteness for spectra and, as described in Appendix A, Blanton et al. (2005c) estimate the incompleteness for the photometry. Their ranges at which the completeness dropped below 90% were $`\mu _{\mathrm{app}}>22`$, when both $`\mu _{\mathrm{app}}<19`$ and $`r>17`$, and $`\mu _{\mathrm{app}}\mathrm{}>18.7`$. They show that the SDSS galaxy distribution drops off well inside these completeness limits. Therefore, our distribution will similarly and the overall LF is unaffected away from the bright and faint end.
In the LF subdivided by $`M_r`$, we see the same overall distribution and trends as the previous studies. The $`\mu _{\mathrm{R}_{50}}`$ is a lognormal distribution with a roughly constant peak at bright magnitudes. This then broadens and dims at fainter magnitudes. The peaks are $`\mu _{\mathrm{R}_{50}}`$ 20.1, 20.5, 21.0, 21.5 and $`21.75`$ in steps of 1.5 magnitudes for $`22.5<M_r5\mathrm{l}\mathrm{o}\mathrm{g}h<15`$. No bimodality is evident, meaning that if the galaxies are divided into one or more populations, as suggested by some of the other measures here such as colour, the surface brightness LF is insensitive to it, or the distributions overlap to such an extent that there is no central dip, with perhaps just a broadening.
Subdivided by $`\mu _{\mathrm{R}_{50}}`$, the LFs appear Schechter-like, with a fainter $`M^{}`$ and steeper faint-end slope towards fainter surface brightness, as expected, although the brightest bins lack galaxies. This is also consistent with previous results.
Blanton et al. (2001) present the BBD for the SDSS EDR. Although the overall luminosity density found by them has been superseded by Blanton et al. (2003c), their BBD is consistent with ours. Petrosian magnitudes are used. They also use the EDR calibrations of the passbands $`u^{}`$, $`g^{}`$, etc. but the difference here is small. They find for $`23.5<M_r^{}<20.5`$ that the peak is at $`\mu _{\mathrm{R}_{50}}20`$, the same as ours for $`22.5<M_r5\mathrm{l}\mathrm{o}\mathrm{g}h<19.5`$. The distribution then broadens and the peak dims to $`\mu _{\mathrm{R}_{50}}20.5,21.5`$ and 22 over the steps of $`1.5\mathrm{mag}`$ to $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h<16.0`$. Given the slightly different binning and the broadness of the peaks these are consistent with our values. At the bright end ($`M_r^{}<19`$) the distribution drops off well above the EDR surface brightness cut of $`\mu _{\mathrm{R}_{50}}<23.5`$ but at the faint end this may be an issue, as they are not able to show the upturn in the faint-end LF later reported by Blanton et al. (2005c). We see similar results relative to the $`\mu _{\mathrm{R}_{50}}<24.5`$ cut. Blanton et al. also plot the LF versus SB, finding as we do that the LF appears Schechter-like, with a fainter $`M^{}`$ and steeper faint-end slope towards fainter surface brightness. We see a clearer downturn in the LF in the range $`18.2<\mu _{\mathrm{R}_{50}}<19.5`$.
Shen et al. (2003) use data from a sample similar to the SDSS DR1. In their fig. 14, they show the surface brightness distributions for early and late type galaxies, defined by concentration index less than and greater than 2.86 respectively. The measures use Sérsic magnitudes and $`h=0.7`$, which we convert to $`h=1`$ here. Late types show mean SBs of around $`20.2\mathrm{mag}\mathrm{arcsec}^2`$ for $`21.2<M_r5\mathrm{l}\mathrm{o}\mathrm{g}h<20.2`$. This then drops to $`20.5\mathrm{mag}\mathrm{arcsec}^2`$ for $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h=18.2`$ and more rapidly to $`21.5\mathrm{mag}\mathrm{arcsec}^2`$ for $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h=15.2`$, the faintest level. The early types are around $`20\mathrm{mag}\mathrm{arcsec}^2`$ for $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h=21.2`$, $`19.5\mathrm{mag}\mathrm{arcsec}^2`$ for $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h=20.2`$ to $`19\mathrm{mag}\mathrm{arcsec}^2`$ for $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h=18.2`$, the faintest. The distributions again broaden at fainter magnitudes. These are approximately consistent with our results.
Blanton et al. (2005c) use the SDSS DR2 and show the overall Petrosian SB distributions for $`20.5<M_r<12.5`$. Over the range $`20.5<M_r<15`$ their peak varies from $`20.6\mathrm{mag}\mathrm{arcsec}^2`$ to $`22.6\mathrm{mag}\mathrm{arcsec}^2`$. This compares to our $`20.5\mathrm{mag}\mathrm{arcsec}^2`$ and around $`21.75\mathrm{mag}\mathrm{arcsec}^2`$ over the same range.
Driver et al. (2005) study the BBD in the MGC, using an improved SB definition involving elliptical isophotes and the deeper $`\mu _{\mathrm{app}}`$ limit mentioned of $`26\mathrm{mag}\mathrm{arcsec}^2`$ in $`B`$. Due to the overestimation in the SDSS mentioned above, they find a systematic offset of $`0.4\mathrm{mag}`$ to fainter values of the SB compared to Shen et al. (2003). They also see the more constant SB values at bright $`M_r`$ but are otherwise consistent with Blanton et al. (2005c).
An improvement to the study here would be to correct for the effects of internal extinction due to dust. This is particularly strong in highly inclined spiral discs. In this paper we attempted to reduce this effect by excluding galaxies with an axis ratio greater than an E7 elliptical using the axis ratio cut described in §2.1. A recent study which does apply a dust correction is de Jong & Lacey (2000). They use a more local sample of galaxies from Mathewson et al. (1992) and Mathewson & Ford (1996), consisting of approximately 1,000 spirals of types Sb–Sdm. A detailed set of criteria that might be followed by future analysis are given by Boyce & Phillipps (1995).
### 4.5 Colours
The colours are computed using the Sérsic magnitudes K-corrected to $`z=0.1`$, the approximate mean of the sample. The $`r`$ band shifted to $`z=0.1`$ has $`\lambda _{\mathrm{eff}}5600`$ Å, similar to the V band. They are preferable to the observed frame colours as the galaxies over the whole redshift range are then more directly comparable. The LF fit implicitly assumes no evolution in the sample. One could take this into account by dividing the sample into redshift bins. All ten colours from $`ugriz`$ were investigated and Figs. 69 show the LF bivariate with $`ur`$, $`gr`$, $`ri`$ and $`rz`$, a representative sample of the results.
For the colours, the completeness (not shown) varies between 60% and 70% for $`13\mathrm{}<r\mathrm{}<16`$, dropping to 50% by $`r=17.77`$ at the faint end and to consistent with zero at $`r11.5`$.
In the LF split by colour, the expected trends of a fainter $`M^{}`$ and steeper $`\alpha `$ for bluer colours are seen. There is a clear lack of faint galaxies with redder colours in all four plots. The panels split by colour show that the LFs are Schechter-like, with a steepening faint-end slope with bluer colour and possible upturns fainter than $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h=18`$, apart from in the bluest bins where the slope is already steep.
In $`ur`$ we see a constant peak around $`ur=1.5`$ for $`18<M_r5\mathrm{l}\mathrm{o}\mathrm{g}h<15`$. This then brightens to $`ur1.75`$ by $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h=21`$ and disappears at brighter magnitudes. A red peak around $`ur=2.5`$ becomes evident at about the same magnitude and brightens to $`ur2.75`$ by $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h=22.5`$.
The bimodality is clearer in $`gr`$, with the blue peak showing similar behaviour to that in $`ur`$, moving from $`gr0.35`$ to $`gr0.6`$ from $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h=15`$ to $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h=21`$. The red peak is visible much fainter than in $`ur`$, moving from $`gr0.8`$ through 0.9 to $`gr1`$ by $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h=24`$.
In $`ri`$, bimodality is no longer evident, with a single peak monotonically moving from $`ri0.3`$ to $`ri0.4`$ from $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h=15`$ to $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h=21`$. The peak is then approximately constant at $`ri0.4`$ at brighter magnitudes. There is therefore either no bimodality in this colour, or the peaks due to the populations causing the peaks in $`ur`$ and $`gr`$ are indistinguishable, either intrinsically or at the colour resolution probed. The constant colour brighter than $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h=19.5`$ suggests that this could still correspond to the early type bulge-dominated population. The lack of an obvious blue peak is consistent with those peaks from $`ur`$ and $`gr`$ being due to star formation.
$`rz`$ shows similar behaviour to $`ri`$, with the peak moving from $`rz0.1`$ to $`rz0.3`$, again constant brighter than $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h=19.5`$.
The plots for $`ug`$, $`ui`$ and $`uz`$ are qualitatively similar to $`ur`$, those for $`gi`$ and $`gz`$ to $`gr`$ and $`iz`$ to $`rz`$, which suggests no obvious physical process apparent in these colours which has been missed by the four shown.
The bivariate luminosity-colour distribution is also presented by Blanton et al. (2001) and Driver et al. (2006). Blanton et al. subdivide the SDSS EDR by Petrosian $`g^{}r^{}`$ and see red and blue peaks at $`g^{}r^{}0.3`$ and $`g^{}r^{}0.75`$ respectively, with the same trends in the LF by colour. Our blue peak is the same but the red peak is at $`gr0.9`$ in both Petrosian and Sérsic magnitudes.
Driver et al. (2006) show the $`ur`$ colour, divided at $`ur=2.1`$. They show the Petrosian value and that from the DR1 PSF magnitudes, which are superior at separating the bimodal population. For the core colours the Schechter parameters are $`M^{}=19.15,19.25`$, $`\varphi ^{}=0.0111,0.0136`$ and $`\alpha =0.15,1.28`$ for the two populations respectively, thus showing a clear downward-turning but not Gaussian LF for the red galaxies.
The prominence of the bluer population in $`u`$ and $`g`$ suggests that this peak in the LF is associated with star formation. This is supported by the GALEX LFs of Wyder et al. (2005), Treyer et al. (2005) and Budavári et al. (2005).
The $`ur`$ colour is used in other investigations such as Balogh et al. (2004) and Baldry et al. (2004) where the galaxies are clearly divided into two populations using counts in colour and morphology bins, which are fit by Gaussians. This is also done in B06 where $`ur`$ is compared to the neural network morphological type.
Overall, the LFs subdivided by colour are consistent with a bimodal population of red bulges and star forming discs.
### 4.6 Absolute Petrosian 90% Radius
The bivariate LF is shown in Fig. 10. The completeness is 50–60% for $`13\mathrm{}<r\mathrm{}<16`$, zero at $`r11`$ and 40% at $`r=17.77`$.
The expected pattern of fainter magnitude and higher normalisation due to larger numbers of galaxies with decreasing radius is seen. There is a clear cutoff with magnitude for a given radius, giving a declining faint-end slope at large galaxy sizes, which turns into a rising faint-end slope at small sizes. This suggests that the rising faint-ends in the Schechter LFs seen in other plots here are due to the smaller galaxies in a sample. As with surface brightness, the regions in which very small volumes are probed do not cut off the LF contours.
### 4.7 eClass Spectral Type
Fig. 11 shows the bivariate luminosity-eClass distribution. The eClass is not measured in the absence of a spectrum so there is no completeness map (although Ball et al. (2004) show that the eClass can be predicted to a RMS accuracy of $`\pm 0.06`$ using ANNs). The range $`0.6\mathrm{}<\mathrm{eClass}\mathrm{}<1.0`$ corresponds to early–late type galaxies.
Subdivided by eClass, the LF shows the expected bright–shallow, dim–steep faint-end slope corresponding to early–late type galaxies. For $`0.2<\mathrm{eClass}<0.0`$ the faint end LF is decreasing but not towards being consistent with a Gaussian LF.
Subdivided by $`M_r`$, there is a clear peak at $`\mathrm{eClass}0.15`$ for galaxies brighter than $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h=18`$ which becomes less distinct at fainter magnitudes. There is a second much broader peak for $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h=19.5`$ and fainter, moving from $`\mathrm{eClass}0.15`$ to $`\mathrm{eClass}0.25`$.
The two peaks are thought, as with the colours above, to correspond to the passive bulges of bright ellipticals and star-forming discs.
The LF has not been presented bivariate with eClass previously in the literature. However, two recent studies use the similar PCA-based 2dFGRS $`\eta `$ class. This covers the range $`5\mathrm{}<\eta \mathrm{}<10`$, with low–high values corresponding to early–late types, in the same way as the eClass. Madgwick et al. (2002) show LFs for $`\eta <1.4`$, $`1.4\eta <1.1`$, $`1.1\eta <3.5`$ and $`\eta 3.5`$. For the first bin, the LF shows a pronounced increase in normalisation fainter than $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h=16`$, and appears Schechter-like in the other bins, with the usual dimming and steepening of the faint-end slope, although the Schechter function is always a poor fit. The $`\eta <1.4`$ class is better fit by a Schechter + power law function.
Driver et al. (2006) shows the LF in the MGC for the classes $`\eta =1`$, $`\eta =2`$ and $`\eta =3,4`$, showing a similar pattern to Madgwick et al. but without the faint-end upturns as the LFs are shown to $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h<16`$. Their MGC continuum type separates their LFs more clearly.
Kochanek et al. (2000) discusses possible biases in spectral LFs due to the use of spectroscopic fibres for the latter. The spectroscopic fibres will result in earlier spectral types for large galaxies due to bias toward their central regions. This bias is not quantified here, but their fig. 4 suggests that it is likely to be present even though most of our galaxies are poorly resolved (§2.1). However, it is unlikely to change the broad behaviour of the LF split by $`M_r`$, i.e. a roughly constant low eClass peak and a higher peak which increases in value for fainter galaxies. Also, various studies (e.g. Glazebrook et al., 2003) find that the effect is not large for star formation rate, which is a dominant constituent of the eClass and the $`\eta `$ type via emission lines.
### 4.8 Other measures
Two further parameters were investigated: the galaxy stellar mass and the $`\mathrm{\Sigma }_N`$ surface density of galaxies. These are shown in Figs. 12 and 13, again with no completeness maps as these measures require spectra.
The stellar mass is useful as a physical parameter which could be directly compared with simulations. A tight correlation is seen between the LF and the stellar mass, with the LF in each bin showing a symmetrical shape of similar Gaussian form to that in absolute effective surface brightness. However, these results are expected because the luminosity is used in the calculation of the stellar mass in the first place, as part of the mass to light ratio calculated from the models of star formation history (see Kauffmann et al., 2003). For consistency with the other LFs, the $`r`$ band is shown, although the correlation would be expected to be slightly tighter in $`i`$ and $`z`$, due to these being less dominated by short-lived bright stars than the bluer bands. In the LF subdivided by stellar mass, the galaxies with a stellar mass of $`M_{\mathrm{stellar}}<10^9\mathrm{M}_{\mathrm{}}`$ have distributions that are clearly truncated by the faint cut-off in absolute magnitude. This is evidence that there are further less massive galaxies not seen in this study.
The $`\mathrm{\Sigma }_N`$ surface density of galaxies from the VAC is used in the morphology-density relation in B06. The bivariate LF appears to be of very similar form across all surface densities, with a power-law form, differing only in normalisation, which increases for both fainter magnitudes and lower densities. The faint luminosity cutoff due to the $`z>0.053`$ limit on the density in the VAC is $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h=19.63`$, the top edge of the figure. The $`z<0.093`$ limit does not have a strong effect on the bright luminosity galaxies.
The variation in the LF with density and galaxy spectral type is investigated by Croton et al. (2005), who find that the LF is fitted by a Schechter function over all densities, that the $`\varphi ^{}`$ and $`M^{}`$ vary smoothly over the density range but that the faint-end slope does not change much. Their normalisation takes into account the volume in which galaxies are present in the density bin used. Our normalisation here is just the overall value for the bivariate LF, so the relative normalisations of the bins are essentially arbitrary. This is also seen in the LF divided by $`M_r`$, in which one expects fewer galaxies at low density with normalisation weighted by volume.
It is also possible that the broad lack of change in the shape of the LF with density is related to the similar lack of change in slope in the correlation function divided by luminosity seen by e.g. Connolly et al. (2002) and Zehavi et al. (2002), but this would require further investigation, especially now that Zehavi et al. (2004a) describes departures from the power law.
## 5 Function Fitting to the Bivariate LF
The Schechter function (equation 2) can, by taking $`L/L^{}=10^{0.4(M^{}M)}`$, be written as
$$\begin{array}{c}\varphi (M)dM=0.4\mathrm{ln}(10)\varphi ^{}\hfill \\ \hfill \times \mathrm{exp}[10^{0.4(M^{}M)}][10^{0.4(M^{}M)}]^{\alpha +1}dM.\end{array}$$
(24)
If one then applies Bayes’ theorem to the bivariate function then it can be written as
$$\psi (M,X)dMdX=\varphi (M)\phi (X|M)dMdX,$$
(25)
which gives two-dimensional functions to fit to the SWML estimate. In general any well-behaved function can be fitted, expressed using either side of equation 25.
One choice is to use a Schechter function for $`\varphi (M)`$ and a Gaussian with mean $`A(MM^{})+B`$ and standard deviation $`\sigma _X`$, where $`A`$ and $`B`$ are constants for $`\phi (X|M)`$. This was first used by Chołoniewski (1985) and is known as the Chołoniewski function:
$$\begin{array}{c}\phi (X|M)dMdX=\frac{1}{\sqrt{2\pi \sigma _X^2}}\hfill \\ \hfill \times \mathrm{exp}\left[\frac{(XA(MM^{})B)^2}{2\sigma _X^2}\right]dMdX.\end{array}$$
(26)
This gives a 6-parameter fit. Chołoniewski (1985) actually used $`M`$ rather than $`MM^{}`$ but the latter scales better with our data.
Due to this greater dimensionality than the monovariate LF, a simple maximum likelihood fit is impractical, as the parameter space to explore is large. The method adopted here is the simplex search, which, while not of the greatest accuracy near the minimum, is robust to starting conditions and finds the minimum rapidly.
Here the Matlab implementation, known as fminsearch, is used. It implements the Nelder-Mead simplex algorithm (Nelder & Mead, 1965) using the method of Lagarias et al. (1998) as described at http://www.mathworks.com. Options used were the Matlab default tolerance of $`10^4`$ and a maximum number of iterations of 10000. The search can sometimes become stuck, for example, if the simplex decays to a lower number of dimensions due to the new point being too close to the centroid. However if the search converged it did so in less than this number of iterations.
The measure of goodness of fit, the error function, is the standard $`\chi ^2`$, given by
$$\chi ^2=\left(\frac{\psi _{obs}\psi _{fit}}{\overline{\psi }_{err}}\right)^2.$$
(27)
The fit is done to $`\mathrm{log}_{10}\psi `$ rather than $`\psi `$ so the error is estimated using
$$\overline{\psi }_{err}=\frac{1}{2}\left[\mathrm{log}_{10}(\psi _{obs}+\psi _{err})\mathrm{log}_{10}(\psi _{obs}\psi _{err})\right],$$
(28)
with $`\psi _{err}`$ from the information matrix determined as described above.
## 6 Function Fitting Results
It was generally found that the functions are a poor fit to the data, as the deviation of the data from a smooth function is large compared to the size of the error bars, giving a large $`\chi ^2`$. Fig. 14 shows the fit of the Chołoniewski function to the BBD and Fig. 15 the $`\chi ^2`$ values. The fit is particularly poor for bright galaxies around $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h21.25`$ and $`\mu _{\mathrm{R}_{50}}19.25`$ as the distribution narrows. The best-fitting parameters are $`\varphi ^{}=0.0018,M^{}=20.6,\alpha =1.19,\sigma _{\mu _{\mathrm{R}_{50}}}=0.71,A=0.30,B=26.5`$.
Driver et al. (2005) also find a poor fit, but for faint galaxies. They suggest that this is due to the assumption of a constant width for the Gaussian in absolute effective surface brightness whereas in fact the distribution broadens at fainter values. The assumption of a linear relation between the mean of the Gaussian and $`M`$ may also be a factor. It is unlikely that the simplex is always getting stuck in a local minimum far from the global minimum as the fit is poor from a variety of starting points but the minimum reached is the same. One could add further parameters but six is already a fair number. The results here are consistent with the constant Gaussian width approximation also being the problem. We confirm the suggestion of Driver et al. (2005) that bulge to disc decomposition is needed.
It is possible that no simple function will provide a good fit, as the superposition of underlying processes generating the bivariate LF may be too complex at the level of precision now available from the data. Or it may be the case that a single function will not fit, but the sum of two or more, each of which describes a different population, for example the red and blue or bulge and disc indicated above, may provide a better fit. The population might be subdivided using an optimal colour-separation criterion, for example that used by Baldry et al. (2004), then fitting, for example, a Chołoniewski function to each half. The difficulty is that the two distributions overlap, which would distort the two halves, and the sum of two Chołoniewski functions involves a large number of parameters, especially if the relation is non-linear between the mean of the Gaussian and $`M`$, the Gaussians vary in width, or skew-Gaussians are required.
Other bivariate LFs such as that with eClass were investigated with Schechter-Gaussian and Gaussian-Gaussian for earlier SDSS dat releases but the fit was also found to be poor. A skew-Gaussian was also tried using an available Matlab port of the software described at http://azzalini.stat.unipd.it/SN. The skew-Gaussian is given by
$$f(x)=2\varphi (x)\mathrm{\Phi }(\alpha x),$$
(29)
where $`\varphi (x)`$ is the ordinary Gaussian and $`\mathrm{\Phi }(\alpha x)`$ adds the skewness $`\alpha `$
$$\varphi (x)=\mathrm{exp}(x^2/2)/\sqrt{2\pi }\mathrm{\Phi }(\alpha x)=_{\mathrm{}}^{\alpha x}\varphi (t)𝑑t.$$
(30)
A skewness of $`\alpha =0`$ corresponds to the ordinary Gaussian, and a negative $`\alpha `$ gives the same shape as the positive but reflected about the vertical at $`x=0`$. The fits obtained were very similar to those for the ordinary Gaussian, with the skewness tending to low values of order 0.01.
## 7 Discussion
A major question raised by these results and other bivariate LF studies is whether a theory or simulation can be constructed which will correctly predict the distributions over the different second LF parameters presented. Two major themes in the LFs are the bimodality of the galaxy population and whether the underlying LFs per galaxy type are Schechter, Gaussian or some other form.
Here we fitted a Chołoniewski function to the bivariate luminosity-surface brightness distribution, and for earlier SDSS data releases experimented with it on other bivariate distributions and fitted Schechter, Gaussian, dual power-law Schechter and a sum of two Schechters to some of the individual LF slices. In all cases, the formal fits were found to be poor, with $`\chi ^2/\nu `$ values of 10 or more. It is thus difficult to tell when a fit is the best that one can do and quote the numbers or if the fit really is poor and should be improved. We therefore have not quoted function fit parameters for individual LF slices in this paper, although with more work one could attempt to do so.
Concerning a bimodal population, the bivariate distributions are consistent with an early type, bright, concentrated, red population and a late type, faint, less concentrated, blue, star forming population. This idea of bimodality has been explored by others as it suggests two major underlying physical processes. These processes are thought to be connected to the formation of the bulge of the galaxy via mergers and the disc by accretion. Thus simulations of galaxy formation could be compared to the bivariate LFs presented. This is discussed further below.
Several results support the idea that the underlying LFs of the giant galaxies, i.e. the usual E, S0 and spirals are Gaussian, whereas those for the dwarfs are Schechter. Sandage, Binggeli & Tammann (1985) and Jerjen & Tammann (1997) study the cores of nearby clusters; de Lapparent (2003) reviews the LFs subdivided by morphology in various surveys and de Lapparent et al. (2003) focusses on the ESO-Sculptor survey. A physical model supporting the idea is given in Schaeffer & Silk (1988) in which the maximum galaxy mass is limited by baryonic cooling within a dynamical time-scale and a minimum mass is given by the minimum virial temperature required for cooling to occur. This gives the Gaussian LF for giants and gas stripping then gives the Schechter function for dwarfs.
Driver et al. (2006) find that the morphological classes E/S0, Sabc and Sd/Irr have distinct distributions relative to the bimodal population defined in the space of core $`ur`$ colour and Sérsic index $`n`$. The E/S0 and Sd/Irr correspond to the red and blue peaks respectively, whereas the Sabc spread across the two peaks. This is argued to be strong evidence that the bulges and discs are the fundamental two components to the bimodal population, not two separate galaxy types. Our ANN types contain few Sd/Irr due to the lack of these types in the training set, but our morphological LF supports this idea in the sense that the LFs for T types 1.5 through 3.5 (Sa through Sc corresponds to types 2–4) are essentially indistinguishable, in contrast to the E/S0 LFs which are distinctly brighter.
The bimodality in colour compared to that in morphology is also considered by B06 in the context of galaxy colour and ANN T-type morphology. There we find that whilst the distributions in colour are clearly bimodal and well-fit by a sum of two Gaussians, those in morphology are less so, with galaxies bridging the gap. This supports the non-fundamental nature of the Sabc types in this context.
Also discussed by, most recently, Driver et al. (2006) and Ilbert et al. (2006) are blue spheroidal galaxies. These mimic faint ellipticals in morphology but are indistinguishable from blue galaxies by spectra. These galaxies could cause type contamination in the LFs, for example causing a steeper faint-end slope in the LF subdivided by $`ur`$, as shown by Driver et al.. Again this agrees with the trends seen in our LFs and therefore requires spectroscopic and morphological information to resolve.
Numerous studies (e.g. Popesso et al., 2005; González et al., 2006) find that cluster LFs are better fit by a sum of a Gaussian at bright magnitudes and a Schechter at faint magnitudes, rather than a single Schechter function. The functions have a dip at $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h17`$, caused by the flattening of the Gaussian before the power-law faint end kicks in. Field galaxies are consistent with a Schechter function. A hint of a similar dip is seen here, as might be expected as the sample encompasses all galaxy environments. The faint end slope is thought to be caused by numerous faint red galaxies which are present in clusters but not in the field. Our LFs do not probe deep enough to say anything definitive.
For the more Schechter-like results such as ours, the question is whether the LFs are as they appear, or whether the underlying types are Gaussian, appearing Schechter because type mixing makes the functions tend towards the overall LF, which is Schechter, or a superposition of Gaussian at the bright end and Schechter or power law at the faint end. The idea has previously been discussed in terms of the Hubble morphological types and the effects can be seen here, particularly in Fig. 2. The LF for the early types could either be Schechter or Gaussian with a faint-end slope contaminated by dwarfs. In their sample which is essentially a subset of the one used here, Nakamura et al. (2003) use the concentration index to show that the sample is not contaminated with faint galaxies with soft cores, which will be dwarfs. Their concentration index LF is similar to their Hubble type LF. The same is seen here.
The bimodality supports the idea that concentrated, early spectral type, red, high Sérsic index galaxies correspond to the early morphological types. If this is the case then the Gaussian LF idea is also consistent with the LF plots bivariate with concentration index, eClass, reference frame colours and Sérsic index. These all show LFs for the early types which could be Gaussian with a rise at faint magnitudes, rising to a Schechter function for later types. The idea is also supported by the LF bivariate with absolute Petrosian radius, which shows steeply declining faint-end slopes which become less steep for galaxies smaller than $`2.5h^1\mathrm{kpc}`$.
Whatever processes cause the bimodality do not, however, show up in the LF subdivided by absolute effective surface brightness. However, it may well be the case that the sum of two bivariate distributions is still required and that, unlike the Chołoniewski function, the relation between the mean of the distribution on the surface brightness axis must be a non-linear function of absolute magnitude to fit the lack of variation in the peak surface brightness at bright magnitudes followed by dimming at fainter levels. At fainter levels the two distributions could diverge and thus reproduce the broadening seen. It may even be the case though that to get a formally good fit in the $`\chi ^2`$ sense requires the distribution to be able to change width in the surface brightness axis as a function of absolute magnitude. If the Schechter function is also required to be a dual power law or Schechter plus Gaussian the number of parameters to fit rises to around 20. This may be impractical, at which point one would want to consider summing perhaps a greater number of simpler physically motivated functions, becoming more akin to semi-analytic models. These functions might be based on the Kormendy relation for bulges (Kormendy, 1977), Freeman’s law for discs (Freeman, 1970), or the more recent relation of de Jong & Lacey (2000), and a relation for dwarf spheroids such as that in Mateo (1998). Another possibility is that central surface brightness may better distinguish between spirals and ellipticals than the absolute effective surface brightness used here.
A useful way to connect observations to simulations is the conditional luminosity function, i.e. the luminosity distribution of galaxies in a dark matter halo. Cooray & Milosavljević (2005) construct the Schechter form of the galaxy LF using a model in which the central massive galaxies in haloes are distributed with a lognormal scatter and the satellite galaxies, where the total galaxy luminosity is greater than the central galaxy, are a power law. This predicts that the faint end of the LF is determined by the galaxies in low mass haloes and should have a slope independent of colour selection or band. If this was combined with a prescription for predicting surface brightness then one might be able to predict the bivariate LFs presented here. This could be done by relating the absolute half-light radius to the dimensionless spin parameter of the dark matter halo (Peebles, 1969), as done by several studies (e.g. Fall & Efstathiou, 1980; Dalcanton et al., 1997). If the same dark matter haloes are used for both models then one could predict the bivariate brightness distribution, at least for discs. Relations of the haloes to other galaxy properties could similarly predict the other bivariate LFs, perhaps along the lines of e.g. Mo et al. (2004). However, our results show an LF faint end that does depend on the colour, so further comparison of observations and theory is clearly important.
The conclusion from various studies is that the two physical processes are likely to be merger to form spheroids and accretion on to discs to give the star formation. In the sense that we see the bimodal early-red versus late-blue population, our results support this conclusion. To test the ideas further requires bulge to disc decomposition and quantitative comparison with simulations incorporating the ideas discussed above, as well as the future work detailed below.
However, there is likely to be more to the story than simply mergers and accretion. For example, pseudobulges (e.g. Kormendy & Kennicutt, 2004) resemble bulges in disc galaxies but are formed due to internal (secular) evolutionary processes. Thus the relative importance of secular evolution on the observables discussed here is a further avenue of investigation. Also, for example, none of the morphologies here take into account the presence of bars, which are important in a secular evolution sense, if not in larger scale environment (e.g. van den Bergh, 2002). There is then the question of more complex light profiles such as the core-Sérsic model of Graham et al. (2003), types that don’t fit the classifications used here such as cD galaxies, higher redshifts, and so on.
Important followup work to the work presented in this paper is to more rigourously separate the dwarf and giant galaxies in the LF, to apply reliable bulge to disc decomposition and to fit functional forms to the two underlying distributions. More refined structural parameters than those used to generate the Main Galaxy Sample such as fully Sérsic-based measurements using elliptical isophotes should be used. Here dust-obscuration is addressed by simply requiring that the axis ratio is less than that for an E7 elliptical galaxy, therefore excluding edge-on spiral discs that are highly reddened by internal extinction. Multiwavelength data could improve this, for example matches to SWIRE (Lonsdale et al., 2003) data would be less affected. One could also match to GALEX (Martin et al., 2005) data to extend to the UV for the star formation.
## 8 Conclusions
The conclusions of this paper are as follows.
The luminosity function is measured bivariate with various galaxy properties from the SDSS and its associated value-added catalogues. The properties are eyeball morphological type, neural network morphological type, inverse concentration index $`CI_{\mathrm{inv}}`$, the Sérsic index of the light profile, absolute effective surface brightness, the reference frame colours $`ur`$, $`gr`$, $`ri`$ and $`rz`$, the absolute Petrosian 90% radius of the galaxy, eClass PCA spectral type, the stellar mass and the $`\mathrm{\Sigma }_5`$ surface density of galaxies. Some of the parameters plotted here are new in the context of a bivariate LF, and for those that have previously been studied, the trends seen are confirmed at high signal-to-noise. This is the largest number of parameters studied in one framework for a bivariate LF.
The SDSS is one of the best datasets currently available for this work, and the large sample combined with CCD photometry is ideal for precision bivariate LFs, enabling division of the galaxy population into numerous smaller samples, each of which is still highly statistically significant. In addition, the stepwise maximum likelihood estimator used is well-known and has been shown to be a good estimator. The magnitudes are K-corrected for each individual galaxy using the well-tested K-correct code written for the SDSS by Blanton et al. (2003b).
The K-corrections do not take into account galaxy evolution or the effects of reddening due to dust, however we restrict the sample to galaxies at $`z<0.15`$ and those with an isophotal axis ratio less than 10/3 to minimize these effects. Near-infrared datasets of comparable size to the SDSS should improve the situation.
The variation of the LF with all of the parameters is highly significant, both in shape and normalisation, except for $`\mathrm{\Sigma }_5`$ where the LF does not change significantly in shape and the normalisation is arbitrary.
The bivariate LFs are consistent with an early type, bright, concentrated, red population and a late type, faint, less concentrated, blue, star forming population. This idea of bimodality has been explored by others as it suggests two major underlying physical processes, which in agreement with other studies we conclude are likely to be mergers leading to bulges and accretion on to discs. The bimodality is considered further in B06 in the context of galaxy colour and morphology as a function of environment.
The various morphological measures are consistent, with the Sérsic index showing the same broad trends as the $`CI_{\mathrm{inv}}`$, eyeball and neural network morphological type.
The overall patterns of the LFs in the four colours $`ur`$, $`gr`$, $`ri`$ and $`rz`$ are similar, as are those for the other six from the SDSS $`ugriz`$ bands. The main change is the increased star formation in the bluer bands and the corresponding lack of bimodality in the redder bands.
The LF bivariate with absolute galactic radius shows a steeply declining faint-end slope which becomes flat then rising for the smallest galaxies, supporting the idea that it is the small galaxies which cause the LF to be shaped like a Schechter function.
The LF bivariate with galaxy stellar mass shows a tight correlation of higher mass for brighter magnitude, with Gaussian LFs, as expected as the luminosity is used in the calculation of the stellar mass.
The LF bivariate with $`\mathrm{\Sigma }_5`$ surface density of galaxies shows little variation in shape over a large range in density from that in a cluster core to the field, being of power-law form throughout. Further work would be needed to quantify the variation that is seen.
No definitive conclusion is reached as to whether the underlying LFs for individual types are Gaussian or Schechter, due to the possibility of type mixing. This is particularly prominent in the LFs bivariate with eyeball morphological type assigned by the neural network, as those for the normal galaxies E, S0, and spirals might be expected to be Gaussian but if type mixing is present this causes them to tend towards the overall LF, which is well fit by a Schechter function. The idea is supported by the LFs bivariate with absolute galactic radius and to a lesser extent by Sérsic index, concentration index, spectral type and colours (the bimodality suggests that these correspond to similar galaxy populations). However, the plots are also consistent with Schechter functions with more or less steeply declining faint-end slopes. Further work would be required to investigate the effect of blue spheroidals, which mimic faint red galaxies in morphology but have bluer colours.
It may be that the morphology of whole galaxies, particularly Sa–Sc spirals, is an intrinsically ‘fuzzy’ measure, so that the Gaussian LFs will not be seen. However, the contamination may be from dwarf galaxies so an important step is to separate these from the normal galaxies and determine their LF separately. The LF by absolute galactic radius is a step towards this but the next step should be using the azimuthally averaged light profiles available for all galaxies in the SDSS. Similarly, further work would be required to see to what extent the faint-end upturns seen in some of the LF slices are real.
Some of the LFs are fitted with analytical functions, which are generally poor fits, probably due to this detail being beyond that reproducible by a simple function. An example is the fit of the Chołoniewski function to the bivariate brightness distribution in which the fit may need to also allow for a variation in the width of the Gaussian with surface brightness, and a possibly non-linear relation between mean absolute magnitude and mean surface brightness, which results in the addition of two parameters to the six already present. The poor fit is also seen in Driver et al. (2005), where it is interpreted in terms of the diverging trends of bulges and discs with decreasing surface brightness. The fitting method used is easily extended to any of the other bivariate LFs presented here.
## Acknowledgments
We thank the referee for a useful report which greatly improved the paper.
Nick Ball thanks Osamu Nakamura for useful discussion on the JPG catalogue, Chris Miller and Michael Balogh for help with the VAC catalogue and Mike Blanton for help with the VAGC catalogue.
Nick Ball was funded by a PPARC studentship. NMB and RJB would like to acknowledge support from NASA through grants NAG5-12578 and NAG5-12580 as well as support through the NSF PACI Project.
Funding for the SDSS and SDSS-II has been provided by the Alfred P. Sloan Foundation, the Participating Institutions, the National Science Foundation, the U.S. Department of Energy, the National Aeronautics and Space Administration, the Japanese Monbukagakusho, the Max Planck Society, and the Higher Education Funding Council for England. The SDSS Web Site is http://www.sdss.org.
The SDSS is managed by the Astrophysical Research Consortium for the Participating Institutions. The Participating Institutions are the American Museum of Natural History, Astrophysical Institute Potsdam, University of Basel, Cambridge University, Case Western Reserve University, University of Chicago, Drexel University, Fermilab, the Institute for Advanced Study, the Japan Participation Group, Johns Hopkins University, the Joint Institute for Nuclear Astrophysics, the Kavli Institute for Particle Astrophysics and Cosmology, the Korean Scientist Group, the Chinese Academy of Sciences (LAMOST), Los Alamos National Laboratory, the Max-Planck-Institute for Astronomy (MPA), the Max-Planck-Institute for Astrophysics (MPIA), New Mexico State University, Ohio State University, University of Pittsburgh, University of Portsmouth, Princeton University, the United States Naval Observatory, and the University of Washington.
This research has made use of NASA’s Astrophysics Data System.
## Appendix A Completeness Maps
For each result in which the second LF parameter does not require spectroscopy, i.e. $`T_{\mathrm{JPG}}`$, $`T_{\mathrm{ANN}}`$, $`CI_{\mathrm{inv}}`$, and $`n`$, the completeness maps for the bins used in the LF SWML estimate are shown. The parameters are therefore selected with the explicit cuts implied by the extent of their axes on these maps. For $`\mu _{\mathrm{R}_{50}}`$ we show the completeness as a function of apparent SB. Maps for the other parameters are shown as a function of apparent magnitude. With the exception of $`T_{\mathrm{JPG}}`$, which is selected from a smaller area with much higher spectroscopic coverage, the overall completeness will generally not exceed the ratio of the spectroscopic areal coverage to that of the imaging over the whole survey. This is about 70%, from 4,783 or 4,681 $`\mathrm{deg}^2`$ versus 6,851 or 6,670 $`\mathrm{deg}^2`$ for the VAGC and the CAS respectively. In general, the completeness will be lower, due to further objects being masked out for the various reasons described above. The maps give an indication of the possible extent of biases in the LFs due to incompleteness in spectroscopy.
Whilst these maps show the completeness in spectroscopy compared to selected targets, they do not show incompletenesses in the imaging compared to the true galaxy distribution. These will arise if either the object is not detected, or it is masked out from the photometry that is compared to the spectroscopy in the maps.
The photometric incompleteness due to surface brightness selection effects is investigated by Blanton et al. (2005c) for the ranges in apparent surface brightness and magnitude $`18<\mu _{\mathrm{app}}<24.5`$ and $`14<r<17.5`$. They process simulated galaxies with the same software used to process the real DR2 imaging data. The effects of seeing, noise and sky subtraction are included. At low surface brightness, they find that the overall completeness drops below 90% around $`\mu _{\mathrm{app}}>22`$, reaching 50% at $`\mu _{\mathrm{app}}=23.4`$ and 0 at $`\mu _{\mathrm{app}}`$ = 24.3. This drop is fairly uniform over the whole range in $`r`$. At the bright end, there is slight incompleteness (90%) for dim high surface brightness objects ($`\mu _{\mathrm{app}}<19`$ and $`r>17`$), ascribed to the star/galaxy separator misclassifying the objects as stars in marginal seeing. There is more substantial incompleteness for bright high surface brightness objects, the overall completeness dropping to 80% for $`\mu _{\mathrm{app}}>18.5`$ and 64% for $`\mu _{\mathrm{app}}=18.2`$. This is ascribed to the explicit surface brightness cut in the Main Galaxy Sample to eliminate binary stars (§2.1). The overall fraction of galaxies affected by incompleteness at the faint end is small because the numbers observed drop well before the completeness drops.
At the bright end, objects will be missed either through being shredded into smaller objects by the deblender or, as mentioned, by being misclassified as an object other than a galaxy. Besides the marginal seeing example mentioned, in recent years evidence has also emerged for a class of galaxies, ultra-compact dwarfs (UCDs, e.g. Phillipps et al., 2001), which are affected by this. Again any effect on the overall LF for the apparent magnitude range investigated in this paper is unlikely to be significant because, as shown by Blanton et al. (2005c), few objects are observed in this regime.
The VAGC quality flag rejects objects which are bad deblends in the ranges $`z<0.01`$ and $`M_r5\mathrm{l}\mathrm{o}\mathrm{g}h>15`$. Most bad deblends are likely to be fragments of large galaxies and would therefore be within this range, being both nearby due to the apparent size of the galaxy in which they are contained and intrinsically faint due to their small mass and therefore light compared to a galaxy. Objects without a spectrum will not be in the range but will also not be shown in the LF.
Strauss et al. (2002) show that in the range $`23<\mu _{\mathrm{app}}<24.5`$, 35% of the objects are bad deblends after applying their local versus global sky value cut. Therefore objects that are in this 35% but not in the range rejected by the quality flag may remain in the LFs presented here. Accurately quantifying the number of objects for which this is the case would require detailed inspection of images in large numbers of LF bins, which is beyond the scope of this paper. Here we inspect the images for some of the outlying bins and plot the LF slices (but not the greyscale) only for bins containing more than 20 galaxies, for which any contamination is small. Bad deblends are therefore unlikely to cause any significant biases in the overall LFs. An exception is again made for $`T_{\mathrm{JPG}}`$, where the cut is 5 galaxies per bin, because these were manually inspected for outliers by the JPG and so the shot noise becomes the limiting factor.
Objects missed altogether by the survey will mean that we are not measuring the true LF. However, studies (e.g. Hayward et al., 2005) have shown that low-surface brightness galaxies do not contribute significantly to the luminosity density of the universe, particularly in the range of brightness we study here. Similarly, UCDs have absolute magnitudes of $`13\mathrm{}<M_B\mathrm{}<11`$ (Drinkwater et al., 2000) and so would not enter into our LFs. |
warning/0507/cond-mat0507285.html | ar5iv | text | # Path-integral approach to the dynamics in sparse random network
## I Introduction
Many systems in nature, such as food webs, metabolic systems, coauthorship of papers, the worldwide web, and so on, can be represented as complex networksWatts ; Barabashi-Review ; Newman ; Dorogovstev . Investigations of real networks have shown that these networks have topologies different from random networks. In particular, we have recognized that many networks have scale-free degree distribution, $`P(k)k^\gamma `$, where $`k`$ is the degree of nodes.
The dynamics involved in complex networks has become an important aspect of the complex network studies in recent times. This problem includes, for example, the spreading of virus in the internet, synchronization of neurons in a brain, change of populations in a food web. Recently, Pastor-Satorras and Vespignani obtained unexpected results in this regard SIS0 . They studied the spread of viruses in complex networks and found that no threshold of infection rate exists for the susceptible-infected-susceptible model in the random scale-free network with $`\gamma 3`$, if the size of the network $`N`$ is infinite. Though real networks such as the internet are finite-size network, this result implies that a virus with a small infection rate can spread over the whole network. We had previously presented another remarkable example of the unusual dynamics involved in complex networksichinomiya ; ichinomiya2 . We studied the Kuramoto synchronization in a random network of oscillators and found that the critical coupling for synchronization becomes zero in scale-free network with $`\gamma 3`$.
In these studies, the mean-field approximation plays an essential role. For the mean-field approximation, we consider a model in which a node $`i`$ couples to another node $`j`$ with a strength proportional to “mean coupling probability” $`k_ik_j/k_{tot.}`$, where $`k_i`$ and $`k_{tot.}`$ are the degree of node $`i`$ and total number of edges, respectively. The dynamics in complex network is much simplified by this approximation, and we can obtain analytical results. However, this model differs from the original network model, in which each node couples to a finite number of nodes. It is remarkable that the mean-field approximation is in good agreement with the numerical simulation result of a random network model.
One of the objectives of this paper is to provide a sound explanation for the mean-field approximation. It is unclear why the mean-field approximation performs well in the random network model. The validity of the mean-field approximation, particularly with regard to the Kuramoto transition, is debatable. Moreno, Pacheko and Vazquez-Prada carried out numerical simulations on the Kuramoto transition in the Barabáshi-Albert network.Moreno1 ; Moreno2 . They concluded that the critical coupling $`K_c`$ is not 0 even if $`N\mathrm{}`$. Their conclusion seems to contradict the result of mean-field theory $`K_c=0`$, though this discrepancy is possibly due to the difference of the order parameter used in these papers. Restrepo, Hunt, and Ott suggested that the the argument based on the largest eigenvalue of the network matrix is superior to that based on the mean-field theoryRestrepo . They demonstrated that the mean-field approximation is valid for Erdös-Rényi networks and random scale-free networks with $`\gamma =3`$, while this approximation does not hold in the case of scale-free networks with $`\gamma =2`$. However, they did not provide any explanation as to why the mean-field theory works well in some random network models. An appropriate explanation to this question is a matter of great interest and significance.
The second objective of this study is to extend the mean-field theory. Although the mean-field theory displays good qualitative coincidence with numerical simulation, it is impossible to examine the fluctuation of the variables by the mean-field approximation. Moereover, as noticed, we cannot apply the mean-field approximation in some network models. Therefore it is meaningful to make an approximation that covers a wider range of complex networks. In this paper, we demonstrate that the distribution of variables in the sparse random network model can be approximated by that obtained from a globally coupled network, in which the distribution of the interaction between the nodes is given by a Gaussian random number. This result indicates that the dynamics in random network can be approximated more precisely by appropriate methods such as dynamical mean-field theoryHertz .
In order to realize the above-mentioned objectives, we utilize the path-integral approach. The path integral, which was originally developed for application in quantum mechanicsFeynman , has also been applied to random impurity problemsMartin ; Dominicis , random spin glassesSommers ; Sompolinsky ; Crisanti , neural networksKree ; Balak , and oscillator systemsStiller . One of the advantages of this approach is that the average over an ensemble of networks can be calculated easily. Limitations of the path-integral include an infinite number of integrals and obtaining a precise average over the ensemble, which is not usually possible. However, this method enables approximation of the distribution of the variables in a systematic manner. In particular, the mean-field approximation can be derived as the lowest order approximation of the path integral. The methods used by us are similar to that used by Theumann for the Hopfield network modelTheumann .
The outline of this paper is as follows. In Sec. II, we present the general description of the dynamics of a network model based on the path-integral approach. We derive a formula that is general and can be applied to any network model in this section. In Sec. III we present two approximations of the path-integral formula, mean-field approximation and perturbation. We also prove that the dynamics of a random network is essentially identical to that of a random Gaussian network. In Sec. IV, we apply the analysis to the Kuramoto transition in a random sparse network. We present the results of numerical simulation, which are consistent with that obtained from the analysis. To conclude, we discuss our obtained results.
## II Path-integral approach to the dynamics of a network model
In this section, we introduce the formalism to study the dynamics of a network model using the path-integral approach. We consider the following differential equations for the network model:
$$\dot{x_i}=f_i(x_i)+\underset{j=1}{\overset{N}{}}a_{i,j}g(x_i,x_j)+\xi _i(t),$$
(1)
where $`\xi _i(t)`$ is a random force that satisfies $`\xi _i(t)=0`$, $`\xi _i(t)\xi _j(t^{^{}})=\delta _{i,j}\delta (tt^{^{}})\sigma ^2`$. We assume $`x_i=x_{i,0}`$ at $`t=0`$. In order to discuss the dynamics of this system, it is useful to introduce Matrin-Siggia-Rose(MSR) generating functional $`Z`$, which is defined asMartin ; Dominicis
$$Z[\{l_{i,k}\},\{\overline{l}_{i,k}\}]=\left(\frac{1}{\pi }\right)^{NN_t}\underset{i=1}{\overset{N}{}}\underset{k=0}{\overset{N_t}{}}dx_{i,k}d\overline{x}_{i,k}e^S\mathrm{exp}(l_{i,k}x_{i,k}+\overline{l}_{i,k}\overline{x}_{i,k})J,$$
(2)
where the action $`S`$ is given by
$$S=\underset{i,k}{}\left[\frac{\sigma ^2\mathrm{\Delta }t}{2}\overline{x}_{i,k}^2+i\overline{x}_{i,k}\left\{x_{i,k}x_{i,k1}\mathrm{\Delta }t(f_i(x_{i,k1})+\underset{j}{}a_{i,j}g(x_{i,k1},x_{j,k1}))\right\}\right],$$
(3)
and $`\mathrm{}`$ represents the average over the ensemble of networks. $`J`$ is the functional Jacobian term,
$$J=\mathrm{exp}\left(\frac{\mathrm{\Delta }t}{2}\underset{i,j,k}{}\frac{(f_i(x_{i,k})+a_{i,j}g(x_{i,k},x_{j,k}))}{x_{i,k}}\right).$$
(4)
Though this term is necessary for the renormalization $`Z(0)=1`$, it is a little cumbersome to treat it in a practical calculation, such as the mean-field approximation or a perturbation. Here we note that, as De Dominicis showedDominicis , the only effect of this Jacobian term is to subtract the nonretarded correlation function $`\overline{x}_{i,k}x_{j,k+k^{}}`$, where $`k^{}1`$. In the following discussion, we omit this Jacobian term, remembering that we only consider the retarded correlation function. Maintaining $`\mathrm{\Delta }tN_t`$ constant at the limit $`\mathrm{\Delta }t0`$, we obtain the MSR generating functional.
We consider the network described by
$$a_{i,j}=\{\begin{array}{cc}1\hfill & \text{ with probability }p_{i,j}\text{,}\hfill \\ 0\hfill & \text{with probability }1p_{i,j}.\hfill \end{array}$$
(5)
We note that $`p_{i,j}`$ can be a function of variables such as $`i`$ or $`j`$. For example, in the one-dimensional chain model, $`p_{i,j}`$ is 1 if $`|ij|=1`$, else it is 0. The average over all networks can be expressed as
$$\mathrm{exp}\left[\underset{i,k}{}i\mathrm{\Delta }t\overline{x}_{i,k}\underset{j}{}a_{i,j}g(x_{i,k1},x_{j,k1})\right]=\underset{i,j}{}\left[p_{i,j}\mathrm{exp}\left\{\underset{k}{}i\mathrm{\Delta }t\overline{x}_{i,k}g(x_{i,k1},x_{j,k1})\right\}+1p_{i,j}\right],$$
(6)
and we obtain
$$e^S=\mathrm{exp}(S_0)\underset{i,j}{}\left[p_{i,j}\mathrm{exp}\left\{\underset{k}{}i\mathrm{\Delta }t\overline{x}_{i,k}g(x_{i,k1},x_{j,k1})\right\}+1p_{i,j}\right],$$
(7)
where
$$S_0=\underset{i,k}{}\frac{\sigma ^2\mathrm{\Delta }t}{2}\overline{x}_{i,k}^2+i\overline{x}_{i,k}\{x_{i,k}x_{i,k1}\mathrm{\Delta }tf_i(x_{i,k1})\}.$$
(8)
The above-mentioned expression is a general one and can be applied to the dynamics of any network model. However, it is often impossible to calculate the precise value of $`e^S`$, particularly in the case of nonlinear dynamics. We need an approximation to obtain the value of $`e^S`$. In the next section, we approximate Eq. (6) by assuming $`p_{i,k}1`$ and $`p_{i,j}p_{k,l}p_{i,j}`$ for any $`i,j,k,`$ and $`l`$.
## III Approximation of the MSR generating functional in a sparse random network model
In this section, we develop an approximation for the MSR generating functional $`Z`$ in a sparse random network. For this, we assume $`p_{i,j}1`$ and $`p_{i,j}p_{k,l}p_{i,j}`$ for any $`i,j,k,`$ and $`l`$. In the case of the Erdös-Rényi model, $`p_{i,j}`$ is independent of $`i`$ and $`j`$; $`p_{i,j}=k/N`$. Therefore this assumption is valid for a sparse Erdös-Rényi model, because $`p_{i,j}p_{k,l}=k^2/N^2p_{i,j}`$. In the case of random network with distribution $`P(k)`$, we construct the network as follows. First, we define the “degee” of node $`i`$ as $`k_i^{}`$, whose distribution concides with $`P(k)`$. Second, we connect the nodes $`i`$ and $`j`$ with probability $`p_{i,j}=k_i^{}k_j^{}/_ik_i^{}`$. Using this procedure, we obtain the random network whose degree distribution is approximately given by $`P(k)`$. In this case, if the maximum degree of a node $`k_{max}`$ is much smaller than $`N`$, $`k_{max}N`$, the assumption is satisfied. On the other hand, this assumption is not satisfied in the Watts-Strogatz model, because $`p_{i,i+1}1`$.
Since $`p_{i,j}1`$, the approximate value of the logarithm of the right-hand side of Eq. (6) is expressed as follows:
$`\mathrm{ln}\left({\displaystyle \underset{i,j}{}}\left[p_{i,j}\mathrm{exp}\left\{{\displaystyle \underset{k}{}}i\mathrm{\Delta }t\overline{x}_{i,k}g(x_{i,k1},x_{j,k1})\right\}+1p_{i,j}\right]\right)`$ $``$ $`{\displaystyle \underset{i,j}{}}p_{i,j}+p_{i,j}\mathrm{exp}\left\{{\displaystyle \underset{k}{}}i\mathrm{\Delta }t\overline{x}_{i,k}g(x_{i,k1},x_{j,k1})\right\}`$ (9)
$`=`$ $`{\displaystyle \underset{i,j}{}}p_{i,j}{\displaystyle \underset{l=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{l!}}\left({\displaystyle \underset{k}{}}i\mathrm{\Delta }t\overline{x}_{i,k}g(x_{i,k1},x_{j,k1})\right)^l.`$
Therefore, we obtain
$`e^S`$ $``$ $`\mathrm{exp}\left[{\displaystyle \underset{i,k}{}}\left\{{\displaystyle \frac{\sigma ^2\mathrm{\Delta }t}{2}}\overline{x}_{i,k}^2i\overline{x}_{i,k}\{x_{i,k}x_{i,k1}\mathrm{\Delta }tf_i(x_{i,k1})\}\right\}\right]`$ (10)
$`\times `$ $`\mathrm{exp}\left[{\displaystyle \underset{i,j}{}}p_{i,j}{\displaystyle \underset{l=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{l!}}\left({\displaystyle \underset{k}{}}i\mathrm{\Delta }t\overline{x}_{i,k}g(x_{i,k1},x_{j,k1})\right)^l\right].`$
To calculate $`Z`$ from this equation, we need an infinite number of integrals, and we cannot carry out this integration practically. However, this formula gives us much information about the averaged dynamics of networks. In the following subsections, we consider two simple approximation schemes, the mean-field approximation and perturbation.
### III.1 Mean-field approximation and beyond
To begin with, we consider an approximation that ignores the $`l2`$ part of Eq.(10) and obtain
$$e^S\mathrm{exp}\left(\underset{i,k}{}\left\{\frac{\sigma ^2\mathrm{\Delta }t}{2}\overline{x}_{i,k}^2i\overline{x}_{i,k}\{x_{i,k}x_{i,k1}\mathrm{\Delta }t(f_i(x_{i,k1})+\underset{j}{}p_{i,j}g(x_{i,k1},x_{j,k1}))\}\right\}\right).$$
(11)
This result demonstrates that the MSR generating functional for Eq.(1) can be approximated as that for the system described by
$$\dot{x_i}=f_i(x_i)+\underset{j}{\overset{N}{}}p_{i,j}g(x_i,x_j)+\xi _i(t).$$
(12)
This equation implies that the mean-field approximation neglects the contribution of the term $`l2`$ in Eq.(10). The mean-field approximation method is based on two assumptions. First, the higher-order term in $`p_{i,j}`$ in Eq.(9) is neglected, and, second, the higher-order term in Eq. (10) is neglected. The former assumption is valid if $`p_{i,j}1`$ for all values of $`i`$ and $`j`$. However, neglecting the higher order term is not always valid. In order to examine this argument, we study the effect of the term $`l=2`$. From the Stratnovich-Hubbard transformation, we obtain the following:
$$\mathrm{exp}\left[p_{i,j}\frac{1}{2}\left(\underset{k}{}i\mathrm{\Delta }t\overline{x}_{i,k}g(x_{i,k1},x_{j,k1})\right)^2\right]=\sqrt{\frac{1}{2\pi p_{i,j}}}𝑑r_{i,j}\mathrm{exp}\left[\frac{r_{i,j}^2}{2p_{i,j}}+i\left(r_{i,j}\underset{k}{}\mathrm{\Delta }t\overline{x}_{i,k}g(x_{i,k1},x_{j,k1})\right)\right].$$
(13)
By comparing Eqs. (11), (12), and (13), we observe that the MSR generating functional is identical to that of the system described by
$$\dot{x_i}=f_i(x_i)+\underset{j}{\overset{N}{}}(p_{i,j}+r_{i,j})g(x_i,x_j)+\xi _i(t),$$
(14)
where $`r_{i,j}`$ is a random number and its distribution is given by a Gaussian distribution, with a mean value of 0 and a dispersion $`r_{i,j}^2=p_{i,j}`$.
Sequential application of the Stratonovich-Hubbard transformation yields the contribution from the term $`l=2^n`$. For example, if we consider the term $`l=4`$, then because
$$\mathrm{exp}\left[p_{i,j}\frac{1}{4!}\left(\underset{k}{}i\mathrm{\Delta }t\overline{x}_{i,k}g(x_{i,k1},x_{j,k1})\right)^4\right]=\sqrt{\frac{3}{2\pi p_{i,j}}}𝑑r_{i,j}^{}\mathrm{exp}\left[\frac{3r_{i,j}^2}{2p_{i,j}}+\frac{r_{i,j}^{}}{2}\left(\underset{k}{}i\mathrm{\Delta }t\overline{x}_{i,k}g(x_{i,k1},x_{j,k1})\right)^2\right],$$
(15)
the Gaussian fluctuation with dispersion $`\sqrt{p_{i,j}/3}`$ is added to $`p_{i,j}`$ in Eq.(13). Therefore, $`P`$ is given by the solution of
$$\dot{x_i}=f_i(x_i)+\underset{j}{\overset{N}{}}(p_{i,j}+r_{i,j})g(x_i,x_j)+\xi _i(t),$$
(16)
where the distribution of $`r_{i,j}`$ is given by a Gaussian with dispersion $`\sigma _r=\sqrt{p_{i,j}+r_{i,j}^{}}`$, and the distribution of $`r_{i,j}^{}`$ is also Gaussian with dispersion $`\sigma _r^{}=\sqrt{p_{i,j}/3}`$. By sequential application of this transformation, we can obtain the effect of the term $`l=2^n`$. However, as these correction terms are small for large $`l`$, the Gaussian random network is a good approximation of the random sparse network.
The estimation that utilizes the Stratonovich-Hubbard transformation is very effective in mapping the dynamics of the sparse network ensemble onto the dynamics of globally coupled networks. This method is very useful, especially in the case where the dynamics of Gaussian random networks is well known. However, the effectiveness of this transformation is limited, because it can only consider the terms $`l=2^n`$. This method does not elaborate on the effect of the terms $`l=3,5,6,\mathrm{}`$. In the next subsection we realize that the correction in $`Z`$ resulting from the term $`l=2m+1`$ is of the order $`p_{i,j}^2`$ in the random network.
### III.2 Perturbation
As shown in the preceding section, though the mean-field approximation and the Stratonovich-Hubbard transformation are very effective methods, they only consider a limited number of terms. To examine the effect of other tems, we use the perturbation technique for a network model in this section. The perturbation gives a formal estimate of the value of $`e^S`$. Furthermore, we note that this method is highly effective because it allows us to estimate the accuracy of the approximation by an order of $`p_{i,j}`$. However, this method has two drawbacks. First, it is typically impossible to obtain the value of $`e^S`$ in nonlinear physics. It is often difficult to obtain the value of $`e^S`$ even if there is no interaction, and the perturbation can therefore only be applied to limited systems. Second, bifurcation or phase transition cannot be obtained without including the infinite order of perturbation in $`p_{i,j}`$. In general, the MSR generating functional $`Z`$ becomes singular at the bifurcation point. However, the finite order perturbation yields $`Z`$, which is a nonsingular function of $`p_{i,j}`$. Therefore, it is impossible to examine a phase transition by perturbation. However, perturbation often yields important information. In this section, we demonstrate that the correction from the odd $`l`$ term is of the order $`p_{i,j}^2`$.
We begin with Eq. (10). On expanding $`\mathrm{exp}(p_{i,j}\mathrm{})`$, we obtain
$`e^S`$ $`=`$ $`\mathrm{exp}\left[{\displaystyle \underset{i,k}{}}\left\{{\displaystyle \frac{\sigma ^2\mathrm{\Delta }t}{2}}\overline{x}_{i,k}^2i\overline{x}_{i,k}\{x_{i,k}x_{i,k1}\mathrm{\Delta }tf_i(x_{i,k1})\}\right\}\right]`$ (17)
$`\times `$ $`{\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{m!}}\left[{\displaystyle \underset{i,j}{}}p_{i,j}{\displaystyle \underset{l=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{l!}}\left({\displaystyle \underset{k}{}}i\mathrm{\Delta }t\overline{x}_{i,k}g(x_{i,k1},x_{j,k1})\right)^l\right]^m.`$
We first consider the correction from the term $`l=3`$. Since the effect of the term $`l=2`$ can be expressed as the fluctuation in $`p_{i,j}`$, we consider
$`e^S`$ $``$ $`\mathrm{exp}({\displaystyle \underset{i,k}{}}\{{\displaystyle \frac{\sigma ^2\mathrm{\Delta }t}{2}}\overline{x}_{i,k}^2i\overline{x}_{i,k}[x_{i,k}x_{i,k1}\mathrm{\Delta }t(f_i(x_{i,k1})+{\displaystyle \underset{j}{}}p_{i,j}^{^{}}g(x_{i,k1},x_{j,k1}))]\}`$ (18)
$`+`$ $`{\displaystyle \underset{i,j}{}}{\displaystyle \frac{p_{i,j}}{3!}}\left({\displaystyle \underset{k}{}}i\mathrm{\Delta }t\overline{x}_{i,k}g(x_{i,k1},x_{j,k1})\right)^3)`$
To account for the effect of the term $`l=3`$, we treat this term using perturbation. We expand the term $`l=3`$ as
$`\mathrm{exp}\left[{\displaystyle \underset{i,j}{}}{\displaystyle \frac{p_{i,j}}{3!}}\left({\displaystyle \underset{k}{}}i\mathrm{\Delta }t\overline{x}_{i,k}g(x_{i,k1},x_{j,k})\right)^3\right]`$ $`=`$ $`1+{\displaystyle \underset{i,j}{}}{\displaystyle \frac{p_{i,j}}{3!}}\left({\displaystyle \underset{k}{}}i\mathrm{\Delta }t\overline{x}_{i,k}g(x_{i,k1},x_{j,k1})\right)^3`$ (19)
$`+`$ $`{\displaystyle \frac{1}{2}}\left\{{\displaystyle \underset{i,j}{}}{\displaystyle \frac{p_{i,j}}{3!}}\left({\displaystyle \underset{k}{}}i\mathrm{\Delta }t\overline{x}_{i,k}g(x_{i,k1},x_{j,k1})\right)^3\right\}^2+\mathrm{}.`$
Here we assume $`p_{i,j}p_{k,l}p_{i,j}`$ for any value of $`i,j,k,`$ and $`l`$ again.
Based on this assumption, $`e^S`$ can be approximated as
$$e^S=\left\{1+\underset{i,j}{}\frac{p_{i,j}}{3!}\left(\underset{k}{}i\mathrm{\Delta }t\overline{x}_{i,k}g(x_{i,k1},x_{j,k1})\right)^3\right\}e^{S_1},$$
(20)
where
$$S_1=\underset{i,k}{}\left\{\frac{\sigma ^2\mathrm{\Delta }t}{2}\overline{x}_{i,k}^2+i\overline{x}_{i,k}[x_{i,k}x_{i,k1}\mathrm{\Delta }t(f_i(x_{i,k1})+\underset{j}{}p_{i,j}^{^{}}g(x_{i,k1},x_{j,k1}))]\right\}$$
(21)
To calculate the contributions to $`Z`$ from these terms, it is convenient to define
$`P(\{ϵ_{i,k}\})`$ $`=`$ $`\left({\displaystyle \frac{1}{\pi }}\right)^{\frac{NN_t}{2}}{\displaystyle }{\displaystyle \underset{i,k}{}}d\overline{x}_{i,k}\mathrm{exp}({\displaystyle \underset{i,k}{}}\{{\displaystyle \frac{\sigma ^2\mathrm{\Delta }t}{2}}\overline{x}_{i,k}^2i\overline{x}_{i,k}[x_{i,k}x_{i,k1}\mathrm{\Delta }t(f_i(x_{i,k1})`$ (22)
$`+{\displaystyle \underset{j}{}}p_{i,j}^{}g(x_{i,k1},x_{j,k1})ϵ_{i,k})]\}).`$
Since
$$\underset{i,k}{}d\overline{x}_{i,k}(i\mathrm{\Delta }t)^3\overline{x}_{i_1,k_1}\overline{x}_{i_1,k_2}\overline{x}_{i_1,k_3}e^{S_1}=\frac{^3}{ϵ_{i_1,k_1}ϵ_{i_1,k_2}ϵ_{i_1,k_3}}P(\{ϵ_{i,k}\})|_{ϵ_{i,k}0},$$
(23)
the MSR generating functional $`Z`$ can be calculated from Eq. (20) if the differential of $`P`$ is known. We first consider the differential of $`P`$ in the case where $`k_1,k_2`$, and $`k_3`$ are distinct. On integrating Eq. (22), we obtain
$$P(\{ϵ_{i,k}\})=C\mathrm{exp}\left[\underset{i,k}{}\frac{1}{2\sigma ^2\mathrm{\Delta }t}\left\{x_{i,k}x_{i,k1}\mathrm{\Delta }t\left(f_i(x_{i,k1})+\underset{j}{}p_{i,j}^{}g(x_{i,k1},x_{j,k1})ϵ_{i,k}\right)\right\}^2\right],$$
(24)
where $`C=(1/\sigma ^2\mathrm{\Delta }t)^{NN_t/2}`$. We note that the integration of $`P(\{ϵ_{i,k}\})`$ over $`x_{i,k}`$ is $`O(1)`$, while $`P(\{ϵ_{i,k}\})`$ is $`O((\mathrm{\Delta }t)^{3/2})`$. On differentiating Eq. (24) with respect to $`ϵ_{i_1,k_1}`$, we obtain
$$\frac{P(\{ϵ_{i,k}\})}{ϵ_{i_1,k_1}}|_{ϵ=0}=\frac{C}{\sigma ^2}\left\{x_{i_1,k_1}x_{i_1,k_11}\mathrm{\Delta }t\left(f_{i_1}(x_{i_1,k_11})+\underset{j}{}p_{i_1,j}^{}g(x_{i_1,k_11},x_{j,k_11})\right)\right\}e^{S_1^{}},$$
(25)
where
$$S_1^{}=\underset{i,k}{}\frac{1}{2\sigma ^2\mathrm{\Delta }t}\left\{x_{i,k}x_{i,k1}\mathrm{\Delta }t\left(f_i(x_{i,k1})+\underset{j}{}p_{i,j}^{}g(x_{i,k1},x_{j,k1})\right)\right\}^2.$$
(26)
On differentiating Eq. (25) with respect to $`ϵ_{i_1,k_2}`$ and $`ϵ_{i_1,k_3}`$ we obtain $`^3P/ϵ_{i_1,k_1}ϵ_{i_1,k_2}ϵ_{i_1,k_3}`$. However, we temporarily consider the term $`P/ϵ_{i,k}`$, because further differentiations with respect to $`ϵ_{i_2,k_2}`$ and $`ϵ_{i_3,k_3}`$ do not modify the following discussion.
In the limit $`\mathrm{\Delta }t0`$, $`\mathrm{exp}(S_1^{})`$ approaches to the $`\delta `$ function $`\delta (x_{i,k}x_{i,k1}\mathrm{\Delta }t[f_i(x_{i,k1})+_jp_{i,j}^{}g(x_{i,k1},x_{j,k1})])`$. Therefore, in the limit $`\mathrm{\Delta }t0`$, Eq. (25) always attains the value 0. However, $`\mathrm{\Delta }tN_t`$ is maintained constant when the limit $`\mathrm{\Delta }t0`$ is taken. Therefore, the sum of the integrals of $`P/ϵ_{i,k}`$ over $`x_{i,k}`$ may have a finite value at the limit $`\mathrm{\Delta }t0`$, if Eq.(25) has a magnitude $`O(\mathrm{\Delta }t)`$. In order to obtain a more accurate estimate of Eq.(25), we express $`S_1^{}`$ as
$`S_1^{}`$ $`=`$ $`{\displaystyle \underset{i,k}{}}{\displaystyle \frac{1}{2\sigma ^2\mathrm{\Delta }t}}\left\{x_{i,k}x_{i,k1}\mathrm{\Delta }t\left(f_i(x_{i,k1})+{\displaystyle \underset{j}{}}p_{i,j}^{}g(x_{i,k1},x_{j,k1})\right)\right\}^2`$ (27)
$`+`$ $`{\displaystyle \frac{1}{\sigma ^2}}\left(f_i(x_{i,k})+{\displaystyle \underset{j}{}}p_{i,j}^{}g(x_{i,k},x_{j,k})\right)(x_{i,k+1}x_{i,k})+O(\mathrm{\Delta }t)+(\text{the terms that do not include }x_{i,k}).`$
We consider the integral $`𝑑x_{i,k}h(x_{i,k})P/ϵ_{i,k}`$, where $`h(x_{i,k})`$ is an arbitrary nonsingular function. Since the value of $`h(x_{i,k})[x_{i,k}x_{i,k1}\mathrm{\Delta }t\{f_i(x_{i,k1})+p_{i,j}^{}g(x_{i,k1},x_{j,k1})\}]`$ is 0 at $`x_{i,k}=x_{i,k1}+\mathrm{\Delta }t(f_i(x_{i,k1})+p_{i,j}^{}g(x_{i,k1},x_{j,k1}))`$, we introduce $`y_{i,k}=x_{i,k}x_{i,k1}\mathrm{\Delta }t\{f_i(x_{i,k1})+p_{i,j}^{}g(x_{i,k1},x_{j,k1})\}`$, and expand using the Taylor expansion.
$$h(x_{i,k})\mathrm{exp}\left\{\frac{1}{\sigma ^2}\left(f_i(x_{i,k})+\underset{j}{}p_{i,j}^{}g(x_{i,k},x_{j,k})\right)(x_{i,k+1}x_{i,k})+O(\mathrm{\Delta }t)+\mathrm{}\right\}=\underset{m=1}{\overset{\mathrm{}}{}}h_my_{i,k}^m.$$
(28)
Therefore, we obtain
$`{\displaystyle 𝑑x_{i,k}h(x_{i,k})\frac{P}{ϵ_{i,k}}}|_{ϵ=0}`$ $`=`$ $`C{\displaystyle 𝑑y_{i,k}\underset{m}{}h_my_{i,k}^m\mathrm{exp}(y_{i,k}^2/2\sigma ^2\mathrm{\Delta }t)}`$ (29)
$`=`$ $`C{\displaystyle 𝑑y_{i,k}^{}\underset{m}{}(2\sigma ^2\mathrm{\Delta }t)^{m+1/2}h_my_{i,k}^m\mathrm{exp}(y_{i,k}^2)}.`$
Since the value of this integral becomes 0 if $`m`$ is odd, the leading order of this integral is obtained from the term $`m=2`$, and has a magnitude of $`O(\mathrm{\Delta }t^{3/2})`$. Since the magnitude of this contribution is smaller than $`O(\mathrm{\Delta }t)`$, we can neglect this term at the limit $`\mathrm{\Delta }t0`$.
Therefore, if $`k_1`$, $`k_2`$, and $`k_3`$ are unequal, then the value of the integral of the second term in Eq. (20) becomes 0. In the case where $`k_1=k_2=k_3`$, $`^3P(ϵ)/ϵ^3`$ is negligible as $`\mathrm{\Delta }t0`$ from a similar argument. In general, the contribution from the term $`^mP(ϵ)/ϵ^m`$ is negligible for odd $`m`$, because it is expressed in the form $`e^{S_1^{}}(x_{i,k}x_{i,k1}\mathrm{\Delta }t(f_i(x_{i,k1})+_jp_{i,j}^{}g(x_{i,k1},x_{j,k1})))\times (\text{nonsingular function})`$.
Similarly, we can prove that $`[^m/ϵ_{i_1,k_1}\mathrm{}ϵ_{i_m,k_m}]P`$ is nonzero if and only if each $`ϵ_{i,k}`$ appears $`2n`$ times in the delimiter. From these results, we conclude that the contribution of the term $`l=3`$ to $`Z`$ is of the order $`p_{i,j}^2`$. In addition, we conclude that the contribution from the term $`l=2m+1`$ is of the order $`p_{i,j}^2`$. The contribution from the term $`l=2^m(2m^{}+1)`$ is also estimated by the Stratonovich-Hubbard transformation. In the case of the random network model, the correction due to these terms is small. For example, the contribution from the term $`l=6`$ to the MSR generating functional term is estimated using
$$\mathrm{exp}\left[p_{i,j}\frac{1}{6!}\left(\underset{k}{}i\mathrm{\Delta }t\overline{x}_{i,k}g(x_{i,k1},x_{j,k1})\right)^6\right]=\sqrt{\frac{180}{\pi p_{i,j}}}𝑑r_{i,j}\mathrm{exp}\left[\frac{r_{i,j}^2}{p_{i,j}}+ir_{i,j}\left(\underset{k}{}i\mathrm{\Delta }t\overline{x}_{i,k}g(x_{i,k1},x_{j,k1})\right)^3\right].$$
(30)
The dispersion of $`r_{i,j}`$ is $`\sqrt{p_{i,j}/90}`$ and the contribution of this term is much smaller compared to that from the Gaussian fluctuation obtained from the term $`l=2`$.
From the discussion based on the Stratonovich-Hubbard transformation and perturbation, we demonstrated that the MSR generating functional for the dynamics of a random sparse network model is almost identical to that for the dynamics of a random Gaussian network. In the next section, we demonstrate that the above analysis is consistent with the result of a numerical simulation of the Kuramoto transition in a network model.
## IV Example: the Kuramoto transition
In the previous section, we developed a general scheme to approximate the dynamics of the random sparse network and found its dynamics can be described by
$$\dot{x_i}=f_i(x_i)+\underset{j}{\overset{N}{}}(p_{i,j}+r_{i,j})g(x_i,x_j)+\xi _i(t)$$
(31)
when $`p_{i,j}1`$. In this case, the distribution of $`r_{i,j}`$ is provided by a Gaussian with dispersion $`\sigma =\sqrt{p_{i,j}}`$. In this section, we apply this approximation to the dynamics of oscillators in random networks.
We consider a random network of oscillators
$$d\theta _i/dt=\omega _i+K\underset{j}{}a_{i,j}\mathrm{sin}(\theta _j\theta _i)$$
(32)
where $`\theta _i`$ and $`\omega _i`$ represent the phase and velocity of the oscillator $`i`$, respectively. The value of $`a_{i,j}`$ is 1 if nodes $`i`$ and $`j`$ are connected, otherwise it is 0. We note that the random Gaussian matrix needs to be considered as symmetric. We consider the case where the distribution of $`\omega _i`$ is given by $`g(\omega )=(1/\sqrt{2\pi \sigma _\omega })\mathrm{exp}((\omega \omega _0)^2/2\sigma _\omega ^2)`$, and $`a_{i,j}`$ represents a random network with its mean degree $`k_0`$. The above discussion suggests that the dynamics of this network can be approximated using the following equation:
$$d\theta _i/dt=\omega _i+\underset{j}{}\left(\frac{k_0}{N}+r_{i,j}\right)K\mathrm{sin}(\theta _j\theta _i),$$
(33)
where the distribution of $`r_{i,j}`$ is given by $`P(r_{i,j})=\sqrt{N2\pi k_0}\mathrm{exp}(Nr_{i,j}^2/(2k_0))`$. This model is similar to the dynamic glass model proposed by DaidoDaido . However, the mean interaction between oscillators is positive in our model, while it is 0 in Daido’s model.
It is difficult to calculate analytically the dynamics of this globally coupled model. In this section, we present the numerical results for the random sparse and random Gaussian networks. For this simulation, we set $`N=1000`$, $`2\sigma _\omega ^2=1.0`$, $`\omega _0=0`$, and $`k_0=10`$. The result obtained is averaged over 50 different networks.
First we examine the coupling dependence of the order parameter $`r`$. In our previous paper, we defined the order parameter $`r`$ as $`r=_ik_ie^{i\theta _i}/k_i`$ for a random sparse network model. However, it is difficult to define such an order parameter for a random Gaussian model. In this paper, we therefore use $`e^{i\theta }`$ as the order parameter for a random Gaussian network, and $`r=_ik_ie^{i\theta _i}/k_i`$ for a random sparse network model. Although these two order parameters are distinct, the difference between them is small, because the distribution of degree has a strong peak at $`k=k_0`$ for a random sparse network. The values of $`r`$ are plotted in Fig. 1 for both the networks for the range $`K=0.02`$-0.20. In both these models, the order parameter remains almost constant for $`K`$ values less than 0.1. There is a rapid increase in $`r`$ for $`K`$ values greater than 0.1. The values of $`r`$ coincide qualitatively for these two models. When $`K=K_c`$, a sharp transition, given by $`r\sqrt{KK_c}`$, is observed in the mean-field approximation. This sharp transition gets smeared out in a random sparse network. The Gaussian model approximates this smearing well.
The order parameters being identical is not unusual, because their obtained values were close to those obtained using the mean-field theory. We now explain the distribution of velocity $`d\theta _i/dt`$. In the mean-field approximation, $`d\theta /dt`$ has a $`\delta `$-function-like peak at $`d\theta /dt=0`$. However, if the coupling between the oscillators is random, the strong peak at $`d\theta /dt`$ will get smeared. In Fig. 2, the distribution of $`d\theta /dt`$ for sparse random and random Gaussian networks is plotted. At $`K=0.02`$, there is an absence of synchronization and the distribution of $`d\theta /dt`$ is Gaussian-like. On the other hand, the oscillators are well synchronized and the distribution has a strong peak at $`d\theta /dt=0`$ when $`K=0.16`$. For the present study, we focus on the distribution at $`K=0.10`$. This value of $`K`$ is close to the critical point, and we suggest that the large fluctuation appears at this point. In the case of sparse networks, the peak at $`d\theta /dt=0`$ is sharper at $`k=0.10`$ than at $`k=0.02`$. The same tendency is observed in the case of a Gaussian network. For example, we observe that $`P(0.1<d\theta /dt<0.1)=0.137`$ for a sparse random network. This value is close to $`P(0.1<d\theta /dt<0.1)=0.131`$ obtained from a random Gaussian network. This consistency in the observed value suggests that a random sparse network can be approximated by a Gaussian random network.
Finally, we present the distribution of the phase $`\theta `$ for both networks. The phase distribution in the $`(\omega ,\theta )`$ plane at $`K=0.16`$ is shown in Fig. 3. Although the coupling strength is sufficiently large for synchronization, the phase does not entirely lie on a single line obtaind from the mean-field approximation method, $`\theta =\mathrm{arcsin}(\omega /Kk_0r)`$. In order to observe the dispersion around the mean-field line, we present the phase distribution of oscillators with $`|\omega |0.05`$ in Fig.4. In this region, $`|\mathrm{arcsin}(\omega /Kk_0r)|`$ is less than 0.05 and the $`\omega `$ dependence of the phase distribution can be neglected. In both the models, the phase distribution lies in a wide range of $`\theta `$. The dispersion $`\sigma `$ for these two figures is $`\sigma ^2=0.105`$ and $`0.122`$ for the random sparse and the random Gaussian network, respectively. Since these two values coincide qualitatively, random Gaussian network is a good approximation of the random sparse network.
## V Conclusion and Discussion
In this paper, we studied the dynamics of a random network model using the path-integral approach. We identified that the mean-field approximation is the lowest-order approximation of $`p_{i,j}`$ and $`l=1`$, as shown in Eq. (11). We also demonstrated that the contribution of the term $`l=2^n`$ can be described by the fluctuation of coupling in the globally coupled approximation method. The contribution of the odd $`l`$ terms is difficult to estimate, though it is of the order $`p_{i,j}^2`$. We applied these general results to the Kuramoto transition, and observed a good agreement with numerical simulations.
The path-integral approach developed through this study is a general one and is applicable to dynamics of any random network. In particular, if the precise result for a randomly coupled model is known, a good approximation can be obtained for random sparse network models. There are several models, such as the replicator modelDiederich , for which the exact results are known for a Gaussian random network . Our analysis proves that the dynamics of random sparse networks can be easily obtained for such models.
The analysis presented in this study is limited to the dynamics in a random network model. In the case of another network model, we need to include the higher-order terms to evaluate the MSR generating functional. It is usually difficult to carry out such a calculation. However, our result provide much informations regarding the validity of the mean-field approximation. For example, the mean-field approximation is applicable if $`p_{i,j}p_{k,l}p_{i,j}1`$. On the other hand, such an approximation is not applicable to the dynamics of a highly clustered network. In such a network, $`p_{i,j}p_{j,k}p_{k,i}O(p_{i,j}p_{j,k})`$, $`p_{i,j}p_{k,l}p_{i,j}`$ cannot be assumed and the contribution from the neglected terms needs to be calculated. It is usually believed that the dynamics of networks with high clustering coefficients cannot be approximated using the mean-field approximation method because of the high clustering coefficient. However, our analysis reveals that the validity of the mean-field approximation methods depends on the value of $`p_{i,j}p_{k,l}`$ and $`p_{i,j}`$. For example, the mean-field approximation method cannot be applied to the square-lattice model even if the clustering coefficient is zero, because the value of $`p_{i,j}p_{k,l}`$ can be as large as that of $`p_{i,j}`$.
We also discuss other studies conducted on the Kuramoto transition in random network models. Restrepo et al. examined the mean-field theory and studied the Kuramoto transition Restrepo . They concluded that synchronization occurs when $`K`$ satisfies the relation $`K>2/\pi g(0)\lambda `$, where $`\lambda `$ is the largest eigenvalue of the network matrix $`a_{i,j}`$ and $`g(0)`$ is the density of the oscillators at $`\omega =0`$. They stated that the mean-field approximation, which was developed by us in previous papers, functions only when $`r_ik_i`$, where $`r_i`$ is the local field defined as $`r_i=_ja_{i,j}e^{i(\theta _i\theta _j)}_t`$, where $`\mathrm{}_t`$ means the average over a long time interval. However, they did not explain the reason why this assumption is valid some random network models, though they stated that there exists some relationship between the eigenvectors of $`a_{i,j}`$ and degree of each node. In this paper, we demonstrated that the mean-field theory is an approximation that considers only the term $`l=1`$ in the MSR generating functional. In this case, the mean-field approximation coincides with the discussion obtained from the largest eigenvalues, because the largest eigenvalue of the matrix $`p_{i,j}=x_ix_j`$ is $`_ix_i^2`$ and its eigenvector $`v`$ is given by $`v=(x_1,x_2,\mathrm{}x_n)`$. In the random network model, $`p_{i,j}=k_ik_j/Nk`$, where $`k_i`$ and $`k_j`$ are the degrees of the nodes $`i`$ and $`j`$. Therefore, the largest eigenvalue of this matrix is $`k^2/k`$, and the critical condition for synchronization in the mean-field approximation becomes identical to that in the discussion based on eigenvalues. In order to examine the applicability of the mean-field approximation, the term $`l=2`$ should be considered. In the case of a random matrix, the largest eigenvalue with a dispersion $`p`$ is expressed by $`2\sqrt{Np}=2\sqrt{k}`$ based on Wigner’s semicircle lawWigner . This result suggests that the mean-field approximation can be applied if $`\sqrt{k}k^2/k`$. In order to examine this, we consider the matrix $`M+G`$, where $`M`$ is the matrix obtained from the mean-field approximation and $`G`$ the Gaussian random matrix, i.e., the distribution of each element of the matrix is Gaussian with dispersion $`\sqrt{k}`$. As observed earlier, the largest eigenvector $`v`$ of matrix $`M`$ satisfies the condition $`Mv=\lambda v`$, where $`\lambda =k^2/k`$. On the other hand, $`|Gv|`$ is of the order $`\sqrt{2k}|v|`$, because all eigenvalues of $`G`$ lie between $`\sqrt{2k}`$ and $`\sqrt{2k}`$. Therefore, $`|(M+G)v|`$ equals approximately $`\lambda |v|`$ based on the assumption that $`\lambda \sqrt{k}`$, and the direction of $`(M+G)v`$ is approximately identical to $`v`$. It should be noted that all vectors $`u`$ that are perpendicular to $`v`$, i.e., $`(u,v)=0`$, satisfy the condition $`Mu=0`$. This implies that $`|(M+G)u|\sqrt{2k}|u|\lambda |u|`$. Therefore, the largest eigenvalue and corresponding eigenvector of $`M+G`$ can be approximated as $`k^2/k`$ and $`v=(k_1,\mathrm{},k_n)`$, respectively. Therefore, the mean-field approximation is a suitable approximation if $`k`$ is sufficiently large. In the case of a scale-free network, the spectrum density differs from that suggested by Wigner’s law and the above-mentioned conclusion should be modified. However, this discussion suggests that the validity of the mean-field approximation is determined by the largest eigenvalues of the mean-field matrix $`M`$ and random matrix $`G`$. If the largest eigenvalue of matrix $`G`$ is as large as $`\lambda `$, the mean-field approximation is not valid. Based on the idea presented in this paper, the claim made by Restrepo et al. implies that the term $`l=2`$ must be included in order to discuss the critical behavior more accurately, especially in the case of a scale-free network with $`\gamma =2`$. Therefore, their work was not a denial of the mean-field theory, but an extension of it.
###### Acknowledgements.
We would like to acknowledge Y. Nishiura, Y. Kuramoto, M. Iima, and T. Yanagita for fruitful discussion. |
warning/0507/astro-ph0507546.html | ar5iv | text | # Inhomogeneous systematic signals in cosmic shear observations
## I Motivation
The power spectrum of the weak gravitational lensing distortions of background galaxies is quite directly related to the power spectrum of intervening matter (Miralda-Escude, 1991; Kaiser, 1992). The weak lensing (WL) power spectrum depends upon the linear and non-linear rates of growth of structure since recombination, and upon the redshift-distance relation produced by the expansion. These dependences, plus the straightforward theoretical framework, make WL a very attractive tool for the constraint of the post-recombination Universe, e.g. dark energy. Current 5–10% measurements of the WL power spectrum have already begun to place interesting constraints (Jarvis et al., 2005), and very much larger-scale projects are planned to reduce the statistical errors on the WL signal to 1 part in $`10^3`$ or lower.
To reap the benefits of these large surveys, systematic errors must be well below the small expected statistical errors. WL measurements are subtle and difficult compared to most astronomical data analyses. There are no “standard lenses” on the sky, so calibration of the WL shear data is a significant worry. The finite point spread function (PSF) width tends to circularize the appearance of background galaxies, squelching the WL shear signal. This must be corrected analytically, and any errors in this process, or inaccuracies in the estimate of the PSF size, will lead to calibration errors. Huterer et al. (2005) investigate the effect of overall mean shear calibration errors on cosmological parameter estimation. It is also likely, however, that there will be spatially varying calibration errors that are larger than the error in the mean calibration. For example, as the PSF size $`\sigma _{}`$ varies during a ground-based survey, the resolution parameter $`R1\frac{\sigma _{}^2}{\sigma _g^2}`$ of the galaxies will vary (here $`\sigma _g`$ is the angular size of a target galaxy). If we fail to track this variation properly, the inferred shear will be modulated by a factor $`(1\delta R(𝜽)/R(𝜽))`$ (Bernstein and Jarvis, 2002). In this paper we calculate the effect of spatially varying calibration errors upon the measured power spectrum, and determine criteria on these systematic errors which will have to be met if they are to be made negligible in future surveys.
Other spatially varying errors could arise from photometric errors or Galactic extinction which will modulate the effective depth of the survey. Both effects may lead to errors in the (photometric) redshift estimation and source galaxy distribution, see e.g. (Jarvis et al., 2003), which would in turn lead to local modulation of the observed shear. While a modulation of redshift depth is not strictly equivalent to a multiplicative modulation of shear, our calculations will still permit an estimate of the level at which depth modulations become significant.
The lensing shear of sources at some redshift $`z_s`$ is a 2-component tensor function of the angular variable $`𝜽`$. As reviewed briefly below, the shear field can be divided into “E” and “B” modes corresponding to curl-free and divergence-free deflections, with corresponding power spectra $`P_E(l)`$ and $`P_B(l)`$ for stationary isotropic fields. Gravitational deflections, being derived from the scalar potential, will produce only E-mode power in the weak limit. It is therefore $`P_E(l)`$ that will be used for cosmological constraints, with the B mode serving as the “canary in the coal mine” to alert us to potential non-gravitational sources of systematic error. A spatially varying scalar calibration factor will alter the amplitude of the E-mode power and convert some into B-mode power. In §III we quantify this effect in terms of the 2-point statistics of the calibration systematic. In §IV we present solutions using several models for the systematic error, and show how the deleterious effects are in general determined just by the rms amplitude and characteristic angular scale of the calibration errors. §V summarizes the results and the requirements upon future surveys that can be derived from these results.
Some related calculations exist in the literature. Schneider et al. (2002) investigate the B-mode signal that is created by inhomogeneous source distributions, using a formalism similar to that employed here. We note that the inhomogeneous-source effect can be avoided by considering only cross-correlations between source bins that are disjoint in redshift. The calibration inhomogeneity that we analyze here will likely not be so easily avoided. Vale et al. (2004) conduct a numerical test of calibration inhomogeneity by modulating the shear seen in a ray-tracing simulation, and then calculating the resultant power spectra. We will test our analytic results against their simulated data.
A rough target for calibration systematics is that their effect on $`P_E(l)`$ be smaller than the expected statistical errors. For a single-screen lens analysis, the uncertainty in $`P_E(l)`$ averaged over an interval $`\mathrm{\Delta }\mathrm{ln}l=1`$ is $`l^1f_{\mathrm{sky}}^{1/2}P_E(l)`$ in the sample-variance limit. So for ambitious surveys with $`f_{\mathrm{sky}}0.5`$, the power spectrum statistical errors are $`1`$ part in $`10^3`$ at $`l=1000`$. At higher $`l`$, the uncertainties due to shape noise and inaccuracies in the non-linear clustering theory will become important. The tolerances may be tighter when one examines the impact of power-spectrum tomography rather than just a single power spectrum. So a good goal is to have the calibration-induced error $`\mathrm{\Delta }P_E(l)`$ be $`10^4P_E(l)`$.
## II Shear field decomposition into E and B modes
Decomposition of a spin-2 field, such as shear or the Stokes parameters, into curl-free and divergence-free part was suggested to be useful for weak gravitational lensing studies by Stebbins (1996), and for cosmic microwave background (CMB) polarization by Kamionkowski et al. (1997) and Zaldarriaga and Seljak (1997). Crittenden et al. (2002) and Schneider et al. (2002) study its use in revealing non-gravitational signals in weak lensing surveys. We briefly review the decomposition of the shear field into independent E and B modes, following the notation of Schneider et al. (2002).
The gravitational lens equation in the one-screen approximation relates the detected direction $`𝜽`$ of photons on the sky to the (unobservable) direction $`𝜷`$ of photons emitted by a source: $`𝚫=𝜽𝜷`$, where $`𝚫`$ is the deflection angle scaled by a factor depending on the angular diameter distances in the observer-lens-source system (Bartelmann and Schneider, 2001). The gradient of the deflection field, being a tensor of rank two, is usually decomposed locally into the trace, symmetric traceless part, and antisymmetric part as follows $`\mathrm{\Delta }_{i,j}=\kappa \delta _{ij}+\gamma _{ij}+\omega ϵ_{ij}`$, where the shear tensor $`\gamma _{ij}`$ is symmetric and $`ϵ_{ij}`$ is the Levi-Civita symbol in two dimensions. We denote partial derivatives with respect to directions in the tangent plane on the sky in a standard fashion by a comma. Thus we may express the convergence $`\kappa `$ and the rotation $`\omega `$ as linear combinations of derivatives of the deflection angle: $`2\kappa =\mathrm{\Delta }_{1,1}+\mathrm{\Delta }_{2,2}`$, $`2\omega =\mathrm{\Delta }_{1,2}\mathrm{\Delta }_{2,1}`$. Also, the shear components $`(\gamma _1,\gamma _2)`$, defined as $`\gamma _1\gamma _{11}=\gamma _{22}`$, $`\gamma _2\gamma _{12}=\gamma _{21}`$, may be written as $`2\gamma _1=\mathrm{\Delta }_{1,1}\mathrm{\Delta }_{2,2}`$, $`2\gamma _2=\mathrm{\Delta }_{1,2}+\mathrm{\Delta }_{2,1}`$. Moreover, we can write the deflection field as a sum of curl free and divergence free parts $`𝚫=𝚫_++𝚫_\times `$ which can be expressed as the gradient of a scalar potential $`\varphi _+`$ and the curl of a pseudoscalar potential $`\varphi _\times `$ respectively (Stebbins, 1996). We designate as “E-mode” the curl-free deflection $`𝚫_+`$, which resembles an electric field pattern, and can be due to the mass distribution. It produces the tangential shear pattern $`\gamma _+`$ (Stebbins, 1996). On the other hand, the divergence-free “B-mode” deflection $`𝚫_\times `$ resembles a magnetic field pattern. This mode reveals in measurements as a “radial” shear $`\gamma _\times `$ (i.e., $`\gamma _+`$ rotated by $`45^{}`$) and it cannot be generated by lensing in the single-screen approximation. The potentials $`\varphi _+`$ and $`\varphi _\times `$ are closely related to the convergence $`\kappa `$ and rotation $`\omega `$ via the Poisson equation, $`^2\varphi _+=2\kappa `$ and $`^2\varphi _\times =2\omega `$. Since the single-screen approximation is thought to be valid in cosmological situations (Van Waerbeke and Mellier, 2003), gravitational lensing information is confined to the E mode while the B mode should be zero. Thus the presence of non-zero B mode would be due to breaking of the single-screen approximation or, more importantly, to a variety of processes not related directly to lensing, such as measurement calibration errors (Hirata and Seljak, 2003; Van Waerbeke et al., 2005; Jarvis and Jain, 2004), clustering of source galaxies (Schneider et al., 2002), or their intrinsic alignments (Heymans and Heavens, 2003). For a more thorough discussion of E/B-mode decomposition see Crittenden et al. (2002).
In order to quantify the contribution of systematic uncertainties to the E and B mode power spectra we introduce a pair of two-point correlation functions $`\xi _+^\gamma (\theta )`$ and $`\xi _{}^\gamma (\theta )`$, following Schneider et al. (2002). They are linear combinations of correlation functions of E and B components of the shear, defined for each pair of galaxies with respect to the preferred coordinate system in which their positions are $`𝜽_1=(0,0)`$ and $`𝜽_2=(\theta ,0)`$:
$$\xi _\pm ^\gamma (\theta )=\gamma _1(𝜽_1)\gamma _1(𝜽_2)\pm \gamma _2(𝜽_1)\gamma _2(𝜽_2).$$
(1)
Moreover, the correlation functions (1) can be expressed as follows in terms of E and B-mode power spectra, $`P_E(l)`$ and $`P_B(l)`$, defined as $`\kappa (𝒍)\kappa (𝒍^{\mathbf{}})\left(2\pi \right)^2\delta _\mathrm{D}(𝒍+𝒍^{\mathbf{}})P_E(l)`$ and $`\omega (𝒍)\omega (𝒍^{\mathbf{}})\left(2\pi \right)^2\delta _\mathrm{D}(𝒍+𝒍^{\mathbf{}})P_B(l)`$:
$$\xi _\pm ^\gamma (\theta )=\frac{1}{2\pi }_0^{\mathrm{}}𝑑ll\left(P_E(l)\pm P_B(l)\right)J_{0,4}(l\theta ).$$
(2)
We can invert those relations and obtain power spectra expressed in terms of the correlation functions \[in what follows we use a convention that upper sign in the sum on the right hand side refers to E-mode power spectrum, lower to B-mode\]:
$$P_{E,B}^\gamma (l)=\pi _0^{\mathrm{}}𝑑\theta \theta \left[\xi _+^\gamma (\theta )J_0(l\theta )\pm \xi _{}^\gamma (\theta )J_4(l\theta )\right].$$
(3)
We do not consider cross-power spectrum of E and B modes as it will vanish due to parity conservation (Schneider et al., 2002).
## III Effect of systematics on E/B power spectra
Ideally, we would like to measure the shear field $`𝜸(𝜽)`$ directly. What we observe, however, is the coherent ellipticity induced on an ensemble of galaxies, which (a) is defined by the distortion $`𝒈=𝜸/(1\kappa )`$; (b) is imparted on galaxies that are not intrinsically circular, and (c) are viewed through a finite point-spread function (PSF). The measured shear field $`𝒅(𝜽)`$ will in practice be modulated or contaminated by various observational effects (Van Waerbeke and Mellier, 2003). Although techniques for shear extraction from galaxy images have been extensively developed and tested (Kaiser et al., 1995; Bernstein and Jarvis, 2002; Hirata and Seljak, 2003), there remain imperfections which can be detrimental to precision cosmology.
Throughout the paper we assume that the observed field is related to the true shear field by a position-dependent multiplicative scalar factor $`1+ϵ(𝜽)`$ such as will result from a misestimation of the “resolution” (Bernstein and Jarvis, 2002) or “shear polarizability” (Kaiser et al., 1995). The systematic field $`ϵ(𝜽)`$ is a random field assumed to have zero mean and described to the lowest interesting order by the two-point correlation function. Thus we express the observed field in terms of the shear and the systematics fields as
$$𝒅(𝜽)=(1+ϵ(𝜽))𝜸(𝜽).$$
(4)
This relation is local in real space, so it is going to couple modes of the shear field in the Fourier space, i.e. have some non-local effect on the relevant power spectra. We assume that the shear field $`𝜸`$ due to massive structures in the Universe is uncorrelated with systematics field $`ϵ(𝜽)`$, which is a Galactic or instrumental foreground. The observed two-point correlation function $`\xi ^d(\theta )`$ can in this case be written as
$`\xi ^d(\theta )`$ $``$ $`𝒅(\mathit{\varphi })𝒅(\mathit{\varphi }+𝜽)`$ (5)
$`=`$ $`(1+ϵ(\mathit{\varphi })ϵ(\mathit{\varphi }+𝜽))𝜸(\mathit{\varphi })𝜸(\mathit{\varphi }+𝜽)`$ (6)
$`=`$ $`\left(1+\xi ^ϵ(\theta )\right)\xi ^\gamma (\theta ).`$ (7)
We have introduced two-point correlation functions $`\xi ^\gamma (\theta )`$ for the shear field and $`\xi ^ϵ(\theta )`$ for the systematics field. For simplicity we will assume that the systematics field $`ϵ(𝜽)`$ is homogeneous and isotropic. In practice the assumption of isoptropy is not restrictive, as the effects of an anisotropic systematic could be approximated to first order by considering the azimuthally averaged correlation function.
Correlation functions for the distortion field, $`\xi _+^d(\theta )`$ and $`\xi _{}^d(\theta )`$, may be expressed as products of the correlation functions for the shear and systematics $`\xi _\pm ^d(\theta )=\left(1+\xi ^ϵ(\theta )\right)\xi _\pm ^\gamma (\theta )`$ which follow from eqs. (1) and (7). We can rewrite eq. (3) in terms of the distortion instead of the shear and then account for systematic signals $`ϵ(𝜽)`$. We split the observed E and B mode power spectra $`P_{E,B}^d(l)`$ into two contributions as follows
$$P_{E,B}^d(l)=P_{E,B}^\gamma (l)+\mathrm{\Delta }P_{E,B}^ϵ(l),$$
(8)
where the term $`P_{E,B}^\gamma (l)`$ is E mode (B mode) power spectrum of the shear and $`\mathrm{\Delta }P_{E,B}^ϵ(l)`$ represents the contributions to the E mode (B mode) power due to systematic signals. We focus on these error terms in the remainder of the paper. Using eqs. (3) and (8) they can be written as
$$\mathrm{\Delta }P_{E,B}^ϵ(l)=\pi _0^{\mathrm{}}𝑑\theta \theta \xi ^ϵ(\theta )\left[\xi _+^\gamma (\theta )J_0(l\theta )\pm \xi _{}^\gamma (\theta )J_4(l\theta )\right].$$
(9)
We assume that the shear correlation functions $`\xi _\pm ^\gamma `$ receive contribution from E mode only, i.e. $`P_E^\gamma (l)=P_\kappa (l)`$, $`P_B^\gamma (l)0`$, since B-mode cosmological contributions are expected to be a few orders of magnitude smaller on scales $`>1^{}`$ (Schneider et al., 2002). The systematic errors $`\mathrm{\Delta }P_{E,B}(l)`$ to E and B mode power spectra can be written as integrals over the convergence power spectrum $`P_\kappa (l)`$ with a window function $`W_{E,B}(l,q)`$:
$$\mathrm{\Delta }P_{E,B}(l)=_0^{\mathrm{}}𝑑qqP_\kappa (q)W_{E,B}(l,q),$$
(10)
where the window function depends solely on the correlation function $`\xi ^ϵ`$ of the systematic modulation, and is given by
$$W_{E,B}=\frac{1}{2}_0^{\mathrm{}}𝑑\theta \theta \xi ^ϵ(\theta )\left[J_0(l\theta )J_0(q\theta )\pm J_4(l\theta )J_4(q\theta )\right].$$
(11)
In the limit of a systematic that is completely correlated across the entire observation, i.e. a constant calibration error, we have $`\xi ^ϵ(\theta )=\mathrm{\Sigma }^2`$, where $`\mathrm{\Sigma }^2`$ is the variance of the calibration error. In this limit we obtain $`W_E(l,q)=\mathrm{\Sigma }^2q^1\delta _\mathrm{D}(lq)`$ and $`W_B(l,q)0`$, where we have used an integral relation for the Bessel functions $`_0^{\mathrm{}}𝑑\theta \theta J_n(l\theta )J_n(q\theta )=q^1\delta _\mathrm{D}(ql)`$ (Abramowitz and Stegun, 1965). Thus the error contributions to E/B power spectra are $`\mathrm{\Delta }P_E(l)=\mathrm{\Sigma }^2P_\kappa (l)`$ and $`\mathrm{\Delta }P_B(l)=0`$ in this case, and there is no conversion of E power to B power, as expected.
Numerical simulations of calibration inhomogeneity in (Vale et al., 2004) are presented in terms of the aperture mass statistics $`M_{\mathrm{ap}}(R)`$ and $`M_\times (R)`$ with compensated filter defined in (Schneider et al., 1998; Bartelmann and Schneider, 1999). We produce analytic predictions for inhomogeneous calibration errors for comparison with the numerical results of (Vale et al., 2004) using the same filter as they did.
## IV Modeling of systematics
We consider several potentially useful models of the correlation function of the systematic signal $`\xi ^ϵ(\theta )`$ and we examine the dependence of E and B mode power spectra (10) on the characteristics of $`\xi ^ϵ(\theta )`$. The correlation functions considered here are analytically tractable and able to describe a wide variety of random processes leading to systematic signals. Each correlation function considered here is assumed to describe a stationary, isotropic random field. We assume that the systematic field has a finite variance $`\mathrm{\Sigma }^2`$. Equation (11) shows that $`\mathrm{\Delta }P_{E,B}(l)\mathrm{\Sigma }^2`$ if $`\mathrm{\Sigma }^2`$ is a prefactor to some otherwise fixed functional form for $`\xi ^ϵ`$. Moreover, a correlation function in 2-D has to be bounded from below by a global minimum value of the Bessel function $`J_0(x)`$. This condition is met by our models because they are assumed to be non-negative (Ripley, 1981).
We also introduce for each correlation function a characteristic scale $`R_{1/2}`$ where the correlation function drops to 50% of its zero-lag value $`\mathrm{\Sigma }^2`$.
### IV.1 Gaussian family
As a first model let us consider correlation function having a Gaussian shape with characteristic scale $`\theta _0`$
$$\xi ^ϵ(\theta )=\mathrm{\Sigma }^2e^{\frac{\theta ^2}{2\theta _0^2}}.$$
(12)
We have $`R_{1/2}^2=2\theta _0^2\mathrm{ln}2`$ for the Gaussian. The Gaussian is chosen because they are usually easy to handle analytically. In this case an integral over scale $`\theta `$ in eq. (11) can be done analytically (Gradshteyn and Ryzhik, 2000) and we obtain the following window function
$$W_{E,B}(l,q)=\frac{\mathrm{\Sigma }^2\theta _0^2}{2}e^{\frac{1}{2}\theta _0^2(l^2+q^2)}\left[I_0(\theta _0^2lq)\pm I_4(\theta _0^2lq)\right],$$
(13)
where $`I_0(x)`$ and $`I_4(x)`$ are the modified Bessel functions of the first kind of zeroth and fourth order respectively (Abramowitz and Stegun, 1965). Because of the exponential growth of $`I_0(x)`$ and $`I_4(x)`$ with $`x`$ it is useful to rearrage terms in (13) and rewrite this equation as
$$W_{E,B}(l,q)=\frac{\mathrm{\Sigma }^2\theta _0^2}{2}e^{\frac{1}{2}\theta _0^2(lq)^2}\left[\widehat{I}_0(\theta _0^2lq)\pm \widehat{I}_4(\theta _0^2lq)\right],$$
(14)
where we have introduced functions $`\widehat{I}_n(x)=e^xI_n(x)`$.
The large-scale amplitude of $`W_{E,B}(l,q)`$ can be derived by noting the asymptotic behavior of the modified Bessel function for small arguments: $`I_0(x)1`$ and $`I_4(x)x^4/384`$ if $`x1`$. Thus for scales large compared to $`R_{1/2}`$ when $`\theta _0l1`$, we obtain $`W_{E,B}(l,q)\frac{1}{2}\mathrm{\Sigma }^2\theta _0^2e^{\frac{1}{2}\theta _0^2l^2}`$. When we consider power spectra $`\mathrm{\Delta }P_{E,B}(l)`$ in this regime we get the following expression
$`\mathrm{\Delta }P_{E,B}(l)`$ $``$ $`{\displaystyle \frac{1}{2}}\mathrm{\Sigma }^2\theta _0{\displaystyle _0^{\mathrm{}}}𝑑qqP_\kappa (q)e^{\frac{1}{2}\theta _0^2q^2}`$ (15)
$``$ $`{\displaystyle \frac{1}{2}}\mathrm{\Sigma }^2P_\kappa \left(l={\displaystyle \frac{1}{\theta _0}}\right).`$ (16)
In the above we used the fact that the function $`q\theta _0e^{1/2\theta _0^2q^2}`$ has a maximum at $`q=1/\theta _0`$ and can be regarded narrow around its maximum. The error we make using this approximation is less than $`10\%`$ for $`R_{1/2}=1^{}`$ and $`40\%`$ for $`R_{1/2}=1^{}`$.
For small scales where $`\theta _0l1`$ and $`\theta _0q1`$, we may use another asymptotic formula for modified Bessel functions which leads to $`\widehat{I}_n(x)\sqrt{2\pi }x^{1/2}`$ (Abramowitz and Stegun, 1965). This limit is safely taken when the argument of $`\widehat{I}_0`$ or $`\widehat{I}_4`$ is greater than $`1`$ or $`100`$, respectively. Thus we obtain from (14) for small scales the following
$$\begin{array}{c}W_E(l,q)\hfill \\ W_B(l,q)\hfill \end{array}\}\frac{\mathrm{\Sigma }^2\theta _0^2}{\sqrt{2\pi \theta _0^2lq}}e^{\frac{1}{2}\theta _0^2(lq)^2}\{\begin{array}{c}1\frac{4}{\theta _0^2lq},\hfill \\ \frac{31}{8\theta _0^2lq}.\hfill \end{array}$$
(17)
The asymptotic expression (17) is useful when computing the small-scale systematics contribution to E and B-mode power spectra (10). Due to the exponential term in the window function (17) it is effectively a Dirac delta function $`\delta _\mathrm{D}(lq)`$. Thus we can write eq. (10) as
$$\begin{array}{c}\mathrm{\Delta }P_E(l)\hfill \\ \mathrm{\Delta }P_B(l)\hfill \end{array}\}\mathrm{\Sigma }^2P_\kappa (l)\{\begin{array}{c}1,\hfill \\ \frac{31}{8}\theta _0^2l^2.\hfill \end{array}$$
(18)
if we consider small scales compared to $`R_{1/2}`$. The asymptotic behavior of E and B-mode systematics power spectra is seen in fig. 1. It is notable that, for sufficiently large $`l`$, the systematic E-mode contribution $`\mathrm{\Delta }P_E`$ is simply a factor $`\mathrm{\Sigma }^2`$ of the convergence power spectrum $`P_\kappa (l)`$. Moreover, $`\mathrm{\Delta }P_E`$ follows the shape of $`P_\kappa (l)`$, whereas the B-mode contribution $`\mathrm{\Delta }P_B`$ drops rapidly.
### IV.2 Patchy
Let us consider a systematic signal $`ϵ(𝜽)`$ which is perfectly correlated within circular patches of diameter $`\theta _0`$ on the sky. This kind of systematics could arise if the survey is a mosaic of (circular) telescope pointings, and each pointing has a constant calibration error that is statistically independent of all other pointings. For example, the impact of time-variable atmospheric seeing on galaxy shape measurements could produce such a pattern. Assuming such a model, we can compute the correlation function $`\xi ^ϵ(\theta )`$ (independent of the specific distribution (pdf) of $`ϵ(𝜽)`$ amplitude)
$$\xi ^ϵ\left(\theta \right)=\mathrm{\Sigma }^2\left[1\frac{2}{\pi }\left(\mathrm{arcsin}\frac{\theta }{\theta _0}+\frac{\theta }{\theta _0}\sqrt{1\frac{\theta ^2}{\theta _0^2}}\right)\right]\mathrm{H}(\theta _0\theta )$$
(19)
where $`\mathrm{H}`$ is the Heaviside step function and $`R_{1/2}0.404\theta _0`$. This type of correlation and its effect on E mode signal degradation and B mode generation were studied by Vale et al. (2004) using ray-tracing simulations (their “sharp modulation” model). Although they assumed square areas of correlation (the shape of CCD detectors), our analytic model should match this well if we perform angular averaging over the square pattern to get an isotropic correlation function.
In this case a closed form for the window function is not attained, so we have to rely on numerical integration. We can deal analytically with the window function (11) in the limit of small scales $`l\theta _01`$ and $`q\theta _01`$. For this purpose we can approximate (19) by $`\xi ^ϵ\left(\theta \right)\mathrm{\Sigma }^2\left[1\theta /\theta _0\right]`$ where $`\theta _0=2R_{1/2}`$. Let us use asymptotic formulae for the Bessel functions for large arguments (Abramowitz and Stegun, 1965) and write the E mode window function as follows:
$`W_E(l,q)`$ $``$ $`{\displaystyle \frac{\mathrm{\Sigma }^2}{\pi \sqrt{lq}}}`$ (20)
$`\times {\displaystyle _0^{\mathrm{}}}d\theta (1{\displaystyle \frac{\theta }{\theta _0}})\mathrm{cos}(l\theta {\displaystyle \frac{\pi }{4}})\mathrm{cos}(q\theta {\displaystyle \frac{\pi }{4}}).`$
Using the formulae for addition of cosines and subsequently perform elementary integration leads us to
$`W_E(l,q)`$ $``$ $`{\displaystyle \frac{\mathrm{\Sigma }^2\theta _0}{\pi \sqrt{lq}}}`$ (21)
$`\times \left[{\displaystyle \frac{1\mathrm{cos}\theta _0\left(lq\right)}{\theta _0^2(lq)^2}}+{\displaystyle \frac{1\mathrm{sin}\theta _0\left(l+q\right)}{\theta _0^2(l+q)^2}}\right].`$
Thus in the interesting case of small scales main contribution to the window function comes from $`lq`$ which leads to
$$W_E(l,q)\frac{\mathrm{\Sigma }^2\theta _0}{2\pi l}\frac{\mathrm{sin}^2\frac{\theta _0\left(lq\right)}{2}}{\frac{\theta _0^2\left(lq\right)^2}{4}}\mathrm{\Sigma }^2l^1\delta _\mathrm{D}(lq).$$
(22)
The B-mode window tends to zero because of the identical asymptotic behavior of $`J_0`$ and $`J_4`$. Thus in the small scales limit we have $`\mathrm{\Delta }P_E(l)\mathrm{\Sigma }^2P_\kappa (l)`$, $`\mathrm{\Delta }P_B(l)0`$ which was the case for a Gaussian correlations as well.
### IV.3 Generalized exponential family
A broad class of correlation functions can be described by a generalized exponential family (Ripley, 1981) as follows
$$\xi ^ϵ(\theta )=\frac{\mathrm{\Sigma }^2}{2^{\nu 1}\mathrm{\Gamma }(\nu )}\left(\frac{\theta }{\theta _0}\right)^\nu K_\nu \left(\frac{\theta }{\theta _0}\right),$$
(23)
where $`K_\nu (x)`$ is the modified Bessel function of the second kind, $`\theta _0`$ is a characteristic scale and $`0<\nu <1`$. For $`\nu =1/2`$ we obtain an exponential correlation function $`\xi ^ϵ(\theta )=\mathrm{\Sigma }^2e^{\theta /\theta _0}`$ with $`R_{1/2}=\theta _0\mathrm{ln}2`$. When $`\nu <1/2`$ the correlation function depends on $`\theta `$ sub-exponentially on small scales and super-exponentially on large scales. For $`\nu >1/2`$ the above behavior is reversed. Exponential-type correlations decay more slowly than do Gaussians with the same $`R_{1/2}`$, so offer a test of the generality of the behavior of $`\mathrm{\Delta }P_{E,B}(l)`$ for a given characteristic scale $`R_{1/2}`$. The window functions $`W_{E,B}`$ also decay slowly compared to the Gaussian case (§IV.1).
For a generalized exponential family (23) we can compute analytically the respective power spectra (Gradshteyn and Ryzhik, 2000)
$$P^ϵ(l)=\frac{4\pi \mathrm{\Sigma }^2\nu }{\theta _0^2}\left(1+\left(l\theta _0\right)^2\right)^{(\nu +1)}.$$
(24)
The power spectrum has power-law scaling at small scales: $`P(l)l^n`$ with $`n=2(\nu +1)`$ for $`l\theta _01`$. The allowed range of spectral indices is $`4<n<2`$ (a power-law correlation function must have $`n>2`$). An example of a systematic signal of this type could be dust extinction in our Galaxy. Schlegel et al. (1998) show that the extinction pattern on the sky can roughly be described by the power-law power spectrum $`P(l)l^{5/2}`$ for scales larger than $`15^{}`$, corresponding to $`\nu =1/4`$.
## V Results
The cosmological background model we assume is $`\mathrm{\Lambda }`$CDM with $`\mathrm{\Omega }_m=0.3`$, $`\mathrm{\Omega }_\mathrm{\Lambda }=1\mathrm{\Omega }_m`$, $`H_0=70\mathrm{km}\mathrm{s}^1\mathrm{Mpc}^1`$, $`\sigma _8=0.93`$. The distribution of source galaxies in redshift is assumed to be $`dN/dzz^2\mathrm{exp}\left[\left(z/z_0\right)^{3/2}\right]`$ with $`z_0=2/3`$ and mean redshift $`1`$. These are as assumed by Vale et al. (2004), so that we may test our results against their ray-tracing simulation results. We compute the convergence power spectrum $`P_\kappa (l)`$ using fitting formula for 3-D dark matter power spectrum given by Smith et al. (2003).
In fig. 1 we show the power spectrum contributions due to an inhomogeneous calibration field described by the Gaussian correlation function (12) for three characteristic scales $`R_{1/2}`$: 1, 10, and 1. We take the rms of the systematics field to be $`\mathrm{\Sigma }=4\%`$. Recall that $`\mathrm{\Delta }P_{E,B}(l)\mathrm{\Sigma }^2`$. We notice that $`\mathrm{\Delta }P_E`$ spectra are featureless and have maxima near the maximum $`P_\kappa (l)`$ (except for very small $`R_{1/2}`$). On the other hand, the B-mode power spectra $`\mathrm{\Delta }P_B`$ have maxima near the characteristic scale of the correlation function.
In order to assess whether the signal due to systematics can be potentially harmful for weak lensing results, let us compare the contaminating power spectra $`\mathrm{\Delta }P_{E,B}(l)`$ to the statistical errors on the convergence power spectrum $`\delta P_\kappa (l)`$ (Kaiser, 1998). Assuming gaussianity of the convergence field we have with sufficient accuracy for our purpose that
$$\delta P_\kappa (l)=\frac{1}{\left(l\mathrm{\Delta }lf_{\mathrm{sky}}\right)^{1/2}}P_\kappa (l)\left(1+\frac{\sigma _\gamma ^2}{n_gP_\kappa (l)}\right),$$
(25)
where $`f_{\mathrm{sky}}`$ is the fraction of the sky covered by a survey, $`n_g`$ is the density of source galaxies with measured shapes, and $`\sigma _\gamma 0.3`$ is galaxy shape noise. The $`P_\kappa (l)`$ data will have to be binned over some interval $`\mathrm{\Delta }l`$ for a meaningful comparison with the systematic error $`\mathrm{\Delta }P_{E,B}(l)`$. Because $`P_\kappa (l)`$ is virtually featureless and there is no cosmological information in its detailed structure, we choose broad bins of width $`\mathrm{\Delta }l=l`$. An even broader binning scheme would lower the $`\delta P_\kappa (l)`$ line in Figure 2 and our derived requirements on $`\mathrm{\Sigma }`$ would scale as $`(\mathrm{\Delta }l)^{1/2}`$. Future, ground based, wide-field surveys like *LSST* <sup>1</sup><sup>1</sup>1http://www.lsst.org are expected to cover $`f_{\mathrm{sky}}50\%`$ of the sky and obtain good shape measurements for about $`30`$ galaxies per $`\mathrm{arcmin}^2`$. Figure 2 shows the convergence power spectrum $`P_\kappa (l)`$ and its statistical errors (25) for these values of $`f_{\mathrm{sky}}`$ and $`n_g`$. The encouraging implication of fig. 2 is that keeping systematics (e.g, shear calibration errors) below $`3\%`$ rms ($`\mathrm{\Sigma }3\%`$) should avoid significant contamination of the observed $`P_\kappa (l)`$, even for future surveys.
Figure 2 plots systematic-error power spectra $`\mathrm{\Delta }P_{E,B}(l)`$ for a gaussian (12), “patchy” (19), and exponential (23) correlation functions, each with $`R_{1/2}=10^{}`$ and $`\mathrm{\Sigma }=4\%`$. We notice that the shape of $`\mathrm{\Delta }P_{E,B}(l)`$ is nearly independent of the specific shape of the correlation function $`ϵ(𝜽)`$ of the systematic field. Thus from a practical point of view the important features of the systematic field are the characteristic scale of correlations and the rms of the field. The latter affects the overall amplitude of the systematic errors as $`\mathrm{\Delta }P_{E,B}(l)\mathrm{\Sigma }^2`$. The former fixes the amplitude of the $`\mathrm{\Delta }P_E(l)=\mathrm{\Delta }P_B(l)`$ at large scales, and gives the scale where the B mode starts decaying.
We can compare our analytic results for the “patchy” correlation function to the numerical tests of Vale et al. (2004). We set $`\theta _0=25^{}`$ and $`\mathrm{\Sigma }=10\%`$ to match the calibration-error pattern they superpose on their ray-tracing data. Our analytic estimates of the errors induced in the aperture mass variances $`M_{\mathrm{ap},\times }^2(R)`$ are shown in fig. 3. These errors can be directly compared to those shown in Figure 2 of Vale et al. (2004), which we reproduce in our Figure. Our estimates closely reproduce the results of the numerical simulation, except that we do not produce trough of $`\mathrm{\Delta }M_{\mathrm{ap}}^2`$ at the characteristic scale around $`25\mathrm{arcmin}`$. The trough might be attributable to sample variance from the finite number (64) of patches used in the simulated images.
## VI Conclusions
We consider the effect of spatially varying multiplicative systematic errors (assumed uncorrelated with the cosmological signal) on the measured power spectra $`P_E(l)`$ and $`P_B(l)`$ in the case of the 2D lensing. The prime example of this type of systematic would be shear calibration errors which vary across the survey area due to changing observing conditions. As shown by Hirata and Seljak (2003) overall shear calibration errors of existing methods of shear measurement can reach $`10\%`$ for galaxies of size comparable to the PSF. Such errors grow larger if one uses more poorly-resolved galaxies, as would be the case for deep ground-based surveys like *LSST*. Uncorrected Galactic extinction could also introduce spatially correlated systematics in survey depth, altering the observed shear correlation functions. When we examine a variety of functional forms for the correlation function $`\xi ^ϵ`$ of the inhomogeneous systematic, we find that all salient effects on the measured power spectrum can be characterized by the variance $`\mathrm{\Sigma }^2`$ of the correlation and its characteristic scale $`R_{1/2}`$. A wide variety of functional forms for $`\xi ^ϵ`$ induced very similar effects on measurements of the convergence power spectrum. Only the small-scale B-mode spectrum is sensitive to the detailed shape of $`\xi ^ϵ`$.
Comparison of the systematics errors $`\mathrm{\Delta }P_E(l)`$ on the power spectrum to the statistical errors expected for future weak lensing surveys indicates that we should not be afraid of systematic contamination if we keep calibration errors below $`\mathrm{\Sigma }3\%`$.
The absence of B-mode contamination in the most recent cosmic-shear measurements (Van Waerbeke et al., 2005; Jarvis and Jain, 2004; Mandelbaum et al., 2005) suggests that systematic errors of the type considered here are below current statistical errors (5–10%) and hence do not bias the conclusions. Note, however, that B-mode power $`\mathrm{\Delta }P_B(l)`$ consistent with zero on scales $`l1/\theta _0`$ does not necessarily imply the absence of significant calibration error $`\mathrm{\Delta }P_E(l)`$ (see Figure 1). Hence the present cosmic shear results could be significantly affected by calibration errors, if they have a correlation length $`\theta _0`$ that is larger than scales considered in the B-mode measurement ($`\theta _01/l`$). Future surveys will beat down statistical errors, so we will have to understand and beat down systematic errors as well. This work suggests that spatially-varying calibration errors will have to be reduced to 3%. This is well below the levels that have been demonstrated to date, but is probably achievable for well-behaved data with careful shape-measurement techniques (Heymans et al., 2005; Nakajima and Bernstein, 2005).
###### Acknowledgements.
We would like to thank Bhuvnesh Jain for frequent discussions. J.G. would like to thank Laura Marian for help with Mathematica. This work is supported by grants AST-0236702 from the National Science Foundation, Department of Energy grant DOE-DE-FG02-95ER40893, and Polish State Committee for Scientific Research grant 1P03D01226. |
warning/0507/cond-mat0507460.html | ar5iv | text | # Stability of Bosonic atomic and molecular condensates near a Feshbach resonance
## .1 Phase diagram
We model the Hamiltonian for a mixture of atoms and molecules near a one-channel Feshbach resonance as
$`={\displaystyle \left[F_m(x)+F_a(x)+F_{am}(x)\right]d^3x}`$ (1)
$`F_a={\displaystyle \frac{\psi _a^{}\psi _a}{2m}}\mu \psi _a^{}\psi _a+{\displaystyle \frac{\lambda _a}{2}}\psi _a^{}\psi _a^{}\psi _a\psi _a`$
$`F_{am}=g\left[\psi _m^{}\psi _a\psi _a+\psi _a^{}\psi _a^{}\psi _m\right]+\lambda _{am}\psi _m^{}\psi _a^{}\psi _a\psi _m`$
$`F_m={\displaystyle \frac{\psi _m^{}\psi _m}{4m}}+(ϵ2\mu )\psi _m^{}\psi _m+{\displaystyle \frac{\lambda _m}{2}}\psi _m^{}\psi _m^{}\psi _m\psi _m`$
where $`F_a`$ and $`F_m`$ represent the pure atomic and molecular contributions, and $`F_{am}`$ the coupling between them. Field operators $`\psi _a(x)`$ and $`\psi _m(x)`$ respectively annihilate atoms and molecules at position $`x`$ (which is suppressed in these equations). Parameters $`\lambda `$ represent the strengths of elastic scattering, while $`g`$ represents the strength of conversion between atoms and molecules, $`\mu `$ is the chemical potential, and $`ϵ<0`$ is the binding energy of a molecule, which can be controlled by tuning an external magnetic field. To treat this Hamiltonian within mean field theory one must renormalize the coupling constants from their bare values. For example, Duine and StoofDuine and Stoof (2003) have derived a simple renormalization scheme which connects these quantities with their bare values, providing their magnetic field dependence.
In this letter, we find the stationary points of (1), and analyze their stability. We discuss two types of stability: dynamic, where small fluctuations do not grow in time; and thermodynamic, where small fluctuations cannot reduce the free energy. Although a thermodynamic instability implies that the system will eventually decay, the timescale, which is governed by kinetics and dissipation, may be long enough that the system appears stable. (In fact, since the ground state of alkali atoms at nano-Kelvin temperatures is a solid, all experiments on ultracold atoms involve states which are thermodynamically unstable.) Following convention, we take a thermodynamically unstable (but dynamically stable) phase to be metastable.
As shown by previous authorsRomans et al. (2004); Radzihovsky et al. (2004), for $`g0`$, there are two possible superfluid orders: (a) a pure molecular condensate $`\varphi _m=\psi _m0`$, $`\varphi _a=\varphi _a=0`$; and (b) a mixed atomic/molecular condensate $`\varphi _a0`$, $`\varphi _m0`$. States with these respective orders will be called a molecular superfluid (MSF) and an atomic superfluid (ASF).
Generically there are two classes of modes which can destabilize these states: density fluctuations, and pairing fluctuations. The latter modes change the relative population of atomic and molecular states without changing the total density. Mueller and Baym Mueller and Baym (2000) characterized both types of modes within a random phase approximation, showing that in the absence of a molecular bound state there is no phase transition between an atomic and paired superfluid. Our current calculation extends this result to the case where a true molecular bound state exists.
Our primary result is the phase diagram in figure 1, shown for $`\lambda _{am}=0`$ and $`\lambda _m/\lambda _a=2`$. Due to the presence of metastability in these experiments we do not limit our discussion to the thermodynamic ground state in each region, but also analyze the stability of other stationary points of the energy, which can have either ASF or MSF character. These stationary points, hereafter called solutions, can be found by working at either fixed density or fixed chemical potential.
Fixing the density, the “forbidden” region<sup>1</sup><sup>1</sup>1Such forbidden regions, corresponding to coexistence of two bulk phases, are generic features of first order phase transitions. contains three or four solutions: A<sub>1</sub>, an ASF thermodynamically unstable to pairing; A<sub>2</sub>, an ASF dynamically unstable to density fluctuations; M, an MSF unstable to pairing; and optionally A<sub>3</sub>, an ASF dynamically unstable to relative phase fluctuations. The “MSF” region contains a stable MSF, and either one (A<sub>1</sub>) or three (A<sub>0</sub>, A<sub>1</sub>, A<sub>2</sub>) ASF solutions, two of which (A<sub>0</sub>, A<sub>1</sub>) are thermodynamically unstable to pairing while the other (A<sub>2</sub>) is dynamically unstable to density fluctuations. A<sub>1</sub>, however, is *dynamically* stable against all fluctuations if $`ϵ\lambda _a<2g^2`$. That is, under these conditions, A<sub>1</sub> is metastable. The “ASF” region contains two (A<sub>1</sub>, A<sub>2</sub>) or three (A<sub>1</sub>, A<sub>2</sub>, A<sub>3</sub>) ASF solutions, one (A<sub>2</sub>) of which is stable. In this region the MSF (M) is unstable to pairing.
Fixing the chemical potential, the “vacuum”, where the ground state contains no particles, has an unphysical ASF solution<sup>2</sup><sup>2</sup>2It corresponds to negative atomic density.. The “MSF” region contains a stable MSF solution, an unphysical ASF solutionendnote21, and possibly two more ASF solutions, one unstable and the other is metastable, possessing a higher free energy than the MSF. The “ASF” region contains three ASF solutions, one of which is unphysicalendnote21, one unstable<sup>3</sup><sup>3</sup>3This solution is stabilized for $`\mu >0`$ (not shown)., and one stable. The MSF is either unstable to pairing or has a higher free energy than this ASF solution.
In the remainder of this paper, we derive these results; we find the stationary states of the Hamiltonian (1) and analyze their dynamic and thermodynamic stability against the two forms of fluctuation at $`\lambda _m=\lambda _{am}=0`$. We then explore the role of finite $`\lambda _m`$ and $`\lambda _{am}`$. We give full details for the calculation at fixed density, and briefly sketch the procedure for fixed chemical potential.
## .2 Stationary States (fixed density)
Assuming a uniform condensate exists, we replace the field operators in Eq. (1) by their expectation values, $`\varphi _m=\psi _m=\sqrt{n_m}e^{i\theta _m}`$ and $`\varphi _a=\psi _a=\sqrt{n_a}e^{i\theta _a}`$, where $`n_{a/m}`$ and $`\theta _{a/m}`$ are the number of condensed atoms/molecules and their phase. The energy only depends upon the phase difference $`4\xi =\theta _m2\theta _a`$, so without any loss of generality we will take $`\varphi _a`$ to be real and positive. Setting $`/\xi =0`$ shows that $`\varphi _m`$ must also be real, but not necessarily positive. We work at fixed density, $`n=n_a+2n_m`$, writing $`\varphi _m=\sqrt{n/2}x`$, and $`\varphi _a=\sqrt{n}\sqrt{1x^2}`$ with $`1x1`$. The points $`x=\pm 1`$ represent the same state. The shifted energy $`=+(\mu ϵ/2)n`$ is then
$$=\frac{\lambda _an^2}{2}(1x^2)^2+\frac{ϵn}{2}(x^21)+\sqrt{2n^3}gx(1x^2)$$
(2)
We define the dimensionless parameters $`\alpha =\lambda _an^{1/2}/2g\sqrt{2}`$ and $`\beta =ϵ/2g\sqrt{2n}0`$. For $`\beta <1`$, as long as $`\alpha `$ is not too negative, there are two extrema as a function of $`x`$: the boundaries $`x=\pm 1`$ are local minima ($`M`$) and a maximum ($`A_1`$) lies between $`x=0`$ and $`x=1`$. However, if $`\alpha `$ is reduced until $`(3+16\alpha ^28\alpha \beta )^3=27(14\alpha \beta )^2`$, we find two additional local extrema; a minimum at $`A_2`$ and a maximum at $`A_0`$. At $`\beta =1`$, the $`x=1`$ point bifurcates, and for $`\beta >1`$ it is a local maximum and the local minimum ($`A_2`$) is found in the region $`1<x<0`$. Illustrative plots are shown in figure 2(a).
Previous analyses Romans et al. (2004); Radzihovsky et al. (2004) show that the MC state $`M`$ is always stable against density fluctuation, and is (thermodynamically and dynamically) stable against pairing fluctuations if and only if $`\beta <1`$.
Thermodynamic stability of the ASF is explored by calculating the Hessian $`H_{ij}=^2/ij`$, where $`i,j=x,n`$. Using the condition $`/x=0`$, these derivatives can be written as $`H_{xx}=n[ϵ+2\lambda (4n_mn_a)12g\varphi _m]`$, $`H_{xn}=H_{nx}=[g(4n_mn_a)2ϵ\varphi _m]/\sqrt{2n}`$, $`H_{nn}=n_a(2n_a\lambda +3g\varphi _m)/(2n^2)`$. The determinant of the Hessian (the discriminant) is related to the compressibility, $`\mu /n=(H_{nn}H_{xx}H_{nx}^2)/H_{xx}`$. For $`A_2`$ the discriminant is always negative, while for $`A_1`$ and $`A_0`$ it is negative for $`\lambda ϵ>2g^2`$ and otherwise positive. Thus $`A_2`$, which is always stable against pairing fluctuations ($`H_{xx}>0`$), is always thermodynamically unstable towards density fluctuations (i.e. has a negative compressibility). Similarly $`A_1`$ and $`A_0`$ are always thermodynamically unstable against pairing fluctuations ($`H_{xx}<0`$), and are thermodynamically unstable against density fluctuations if and only if $`\lambda ϵ>2g^2`$.
Dynamical stability is explored by calculating the equations of motion for the fluctuations. We write the field operators in terms of density fluctuation $`\widehat{\rho }(r)`$, pairing fluctuation $`\widehat{y}(r)`$, relative phase fluctuation $`\widehat{\chi }(r)`$, and total phase fluctuation $`\widehat{\theta }(r)`$.
$`\widehat{\psi }_m(r)`$ $`=\sqrt{{\displaystyle \frac{n+\widehat{\rho }(r)}{2}}}[x+\widehat{y}(r)]e^{2i[\widehat{\theta }(r)+\xi +\widehat{\chi }(r)]}`$ (3)
$`\widehat{\psi }_a(r)`$ $`=\sqrt{n+\widehat{\rho }(r)}\sqrt{1(x+\widehat{y}(r))^2}e^{i[\widehat{\theta }(r)\xi \widehat{\chi }(r)]}`$
The equations of motion are found by making stationary the action
$$S=i\widehat{\psi }_a^{}_t\widehat{\psi }_a+i\widehat{\psi }_m^{}_t\widehat{\psi }_m$$
(4)
Working to quadratic order in the fluctuations, we find
$`\dot{\rho }_k={\displaystyle \frac{n}{m}}k^2\theta _k{\displaystyle \frac{nu}{m}}k^2\chi _k`$ (5)
$`u\dot{\rho }_k4nxy_k={\displaystyle \frac{nu}{m}}k^2\theta _k\left[{\displaystyle \frac{n}{m}}k^2+H_{\xi \xi }\right]\chi _k`$
$`nx\dot{\chi }_k=\left[vk^2H_{nx}\right]{\displaystyle \frac{\rho _k}{4}}\left[{\displaystyle \frac{n(3x^2+1)}{4m(1x^2)}}k^2+H_{xx}\right]{\displaystyle \frac{y_k}{4}}`$
$`\dot{\theta }_ku\dot{\chi }_k=\left[{\displaystyle \frac{3x^24}{16mn}}k^2H_{nn}\right]\rho _k+\left[vk^2H_{nx}\right]y_k`$
where $`H_{\xi \xi }=16\sqrt{2n^3}gx(1x^2)`$, $`u=12x^2`$, $`v=3x/8m`$, $`\dot{a}_ta`$ and the Fourier components of the fluctuation operators are defined by $`O(r)=_kO_ke^{ikr}`$. As $`k0`$ the density and pairing modes decouple, and their frequencies are
$$\begin{array}{cc}\hfill \omega _{\mathrm{density}}^2& =c_s^2k^2+𝒪(k^4)\hfill \\ \hfill \omega _{\mathrm{pair}}^2& =\mathrm{\Delta }^2+𝒪(k^2)\hfill \end{array}$$
(6)
where the speed of sound is related to the compressibility by the standard expression $`c_s^2=(n/m)\mu /n`$, and the gap to pairing excitation is $`\mathrm{\Delta }^2=H_{xx}H_{\xi \xi }/16n^2x^2`$. Since $`H_{\xi \xi }x`$, $`A_1`$ and $`A_2`$ are dynamically stable against pairing fluctuations, while $`A_0`$ is unstable. Conversely, we see a dynamic instability towards density fluctuations if and only if a thermodynamic instability exists.
## .3 Effect of non-zero $`\lambda _m`$ and $`\lambda _{am}`$
We have seen that in the absence of $`\lambda _{am}`$ and $`\lambda _m`$ there is no stable ASF, and the metastable ASF always has larger energy than the MSF. Hence there is no MSF$``$ASF phase transition. We now show the existence of a continuous MSF$``$ASF phase transition when $`\lambda _m>0`$. To produce such a continuous phase transition it is necessary and sufficient to show that there exists a stable ASF, with arbitrarily small atomic fraction, at the point the MSF becomes destabilized. In the presence of a non-zero $`\lambda _m`$ and $`\lambda _{am}`$ figure 2(b) represents the generic structure of $``$; two minima at $`A_2`$ and $`A_3`$ and a maximum at $`A_1`$. In terms of dimensionless parameters $`\gamma =\lambda _m\sqrt{n/2}/8g`$ and $`\eta =\lambda _{am}\sqrt{n/2}/2g`$, $`A_{2,3}`$ appears at $`x=1`$ when $`\beta +2\gamma \eta =1`$, where the upper signs correspond to $`A_2`$ and the lower signs correspond to $`A_3`$.
The compressibility at $`x=1`$ when $`A_{2,3}`$ first appears is proportional to $`16\alpha \gamma (12\eta )^2`$. So neither ASF is stable if $`\gamma =0`$, i.e., even when $`\lambda _{am}0`$, a continuous MSF$``$ASF phase transition cannot exist if $`\lambda _m=0`$.
The curvature ($`H_{xx}`$) at $`x=1`$ when $`A_{2,3}`$ first appears is proportional to $`\pm 1+2(\alpha +\gamma \eta )`$. At $`A_3`$, $`H_{\xi \xi }x`$ is negative, and therefore $`\mathrm{\Delta }^2H_{xx}H_{\xi \xi }`$ is negative whenever $`H_{xx}>0`$; i.e., $`A_3`$ is always either dynamically or thermodynamically unstable against pairing fluctuations. The dynamical instability of $`A_3`$ even when $`H_{xx}>0`$ can be understood as instability against fluctuations in $`\xi `$, i.e., in the $`x\xi `$ plane, the energy has a saddle-point at $`A_3`$ (recall that $`4\xi `$ is the relative phase between the atomic and molecular components). At $`A_2`$, however, $`H_{xx}>0`$ is equivalent to $`\mathrm{\Delta }^2H_{xx}H_{\xi \xi }>0`$.
When the atom-molecule scattering vanishes ($`\eta =0`$), the stability conditions at $`A_2`$, viz. $`H_{xx}>0`$ and $`\mu /n>0`$ are simultaneously satisfied if and only if $`\gamma >0`$ and $`16\alpha \gamma >1`$. Thus there exists an MSF$``$ASF continuous phase transition when $`\lambda _{am}=0`$ iff $`n\lambda _a\lambda _m>2g^2`$ and $`\lambda _m>0`$ <sup>4</sup><sup>4</sup>4The general criterion for the existence of a continuous MSF$``$ASF phase transition when $`\lambda _m0`$, $`\lambda _{am}0`$ can be worked out in the $`\lambda _m\lambda _{am}`$ space from the conditions $`16\alpha \gamma >(12\eta )^2`$ and $`1+2(\alpha +\gamma \eta )>0`$..
## .4 Stationary States (fixed chemical potential)
Working at fixed chemical potential (and taking $`\lambda _{am}=0`$), there are two type of stationary points of Eq. (1); an MSF: $`\varphi _a=0`$, $`\varphi _m^2=(2\mu ϵ)/\lambda _m`$, and an ASF: $`\lambda _a\lambda _m\varphi _m^3+(\lambda _a(ϵ2\mu )2g^2)\varphi _m+\mu g=0`$, $`\varphi _a^2=(\mu 2g\varphi _m)/\lambda _a`$. The ASF equation has three solutions, one of which can be ruled outendnote21. Stability analysis is done for both ASF and MSF states by considering fluctuations in $`\varphi _m`$, $`\varphi _m^{}`$, $`\varphi _a`$ and $`\varphi _a^{}`$, analogous to Eq. (5). In terms of dimensionless quantities $`\phi =\varphi _m\lambda _m/g`$, $`r=\lambda _m/\lambda _a`$, $`\epsilon =ϵ\lambda _a/g^2`$ and $`\nu =\mu \lambda _a/g^2`$, the MSF solution first appears at $`2\nu =\epsilon `$ and is stable for $`\nu <(42\sqrt{4\epsilon r})/r`$. The two physical ASF solutions exist for $`\nu >\nu _c`$ where $`4(\epsilon 2\nu _c2)^3+27\nu _c^2r=0`$; one of them is always stable, the other is stable for $`\nu >0`$. The ASF$``$MSF tricritical point is obtained by demanding that the two physical ASF solutions appear exactly when the MSF destabilizes. Mathematically,
$`4(\epsilon 2\nu 2)^3+27\nu ^2r`$ $`=0`$
$`\phi ^3+\phi r(\epsilon 2\nu 2)+\nu r^2`$ $`=0`$ (7)
$`\phi ^2r(2\nu \epsilon )`$ $`=0`$
Solving these three simultaneously gives the tricritical point $`\nu _{tc}=2/\sqrt{r}`$ and $`\epsilon _{tc}=14/\sqrt{r}`$; $`r`$ therefore uniquely determines the phase diagram. Coupled with $`n=2\varphi _m^2=2(2\mu ϵ)/\lambda _m`$, this yields the familiar result $`n\lambda _a\lambda _m=2g^2`$.
## .5 Discussion
We have shown that a continuous ASF$``$MSF phase transition can occur at sufficiently high density in a Bose gas near a Feshbach resonance with repulsive molecule-molecule interaction. This ASF does not, however, correspond to the phase currently studied in cold atom experiments. The experimental “phase” is a saddle point of the free energy, and always has a higher energy than the MSF.
The most obvious route to studying this transition would involve first creating an MSF (for instance, using the technique of Xu et al Xu et al. (2003)), then slowly ramping toward the resonance (making $`\left|ϵ\right|`$ smaller). As pointed out by previous authors Romans et al. (2004); Radzihovsky et al. (2004) the transition could be detected by observing the behavior of vortices.
We caution that this transition does not occur at arbitrarily low densities, nor in the absence of molecule-molecule scattering<sup>5</sup><sup>5</sup>5Estimating $`\lambda _m4\pi \mathrm{}^2a_s/2m`$ far from resonance, we see that in current experiments Xu et al. (2003); Claussen et al. (2002); Donley et al. (2002) $`n2g^2/\lambda _a\lambda _m`$, making observation of this phase transition impossible ($`g^2(4\pi \mathrm{}^2/m)a_{\text{bg}}\mathrm{\Delta }\mu \mathrm{\Delta }B`$ Mukaiyama et al. (2004)).. Therefore inelastic 3-body processes will inevitably limit the lifetime of the cloud Petrov (2004), perhaps making these experiments impractical. In fact, estimating the time scale of three-body recombinationsBraaten and Hammer (2001) to be $`\tau _{\text{3-body}}m/\mathrm{}a_s^4n^2`$ with $`a_s=\mathrm{}/\sqrt{mϵ_c}`$ ($`ϵ_c`$ being binding energy for the transition) and $`n=2g^2/\lambda _a\lambda _m`$ already gives $`\tau _{\text{3-body}}10^4\text{s}`$. Quantum interference effects can drastically reduce this decay rate, but only at particular binding energiesBraaten and Hammer (2001). Using a photoassociation transition in lieu of Feshbach resonance may provide sufficient control over the parameters of the system to avoid these difficultiesHeinzen et al. (2000).
This work was partially performed at the Aspen Center for Physics and was supported by the National Science Foundation (NSF) under grant PHY-0456261 and the Cornell Center for Material Research (DMR-0079992). |
warning/0507/cond-mat0507288.html | ar5iv | text | # Sodium Bose-Einstein Condensates in an Optical Lattice
## Introduction
Optical lattices have become a powerful tool to enhance the effects of interaction in ultracold atomic systems to create strong correlations and probe many-body physics beyond the mean-field theory catalotti01 ; orzel01 ; greiner02 ; greiner02colrev ; paredes04 ; schori04 ; kohl05 . Simply through varying the depth of the lattice potential, one changes the tunneling rate as well as the on-site interaction energy by changing the confinement of the atoms. The strength of the atomic interaction can be directly tuned with a magnetic Feshbach resonance inou98 . In comparison to <sup>87</sup>Rb, which has been used in almost all experiments on optical lattices, <sup>23</sup>Na has stronger and wider Feshbach resonances that are experimentally accessible sten98stro ; marte02 . One such resonance has been used to produce quantum degenerate Na<sub>2</sub> molecules xu03 . Therefore, a sodium condensate loaded into an optical lattice would be a rich and flexible system for studying strong correlations.
So far, most optical lattice experiments have been performed with relatively heavy atomic species (e.g. rubidium and potassium) for which the recoil frequencies are lower and lasers are readily available to achieve trap depths of several tens of recoil frequencies at a few tens of milliwatts. For <sup>23</sup>Na, high power single-mode lasers are necessary for similar experiments. In this work, we chose to use a dye laser red-detuned by $`5`$ nanometers from the D lines of sodium (589 nm). The spontaneous scattering rate limited the time window of the experiment to less than 50 ms, but was still sufficient to satisfy the adiabaticity condition to explore the quantum phase transition from a superfluid to a Mott insulator. We also observed strong atom losses at various lattice laser detunings, which were interpreted as photoassociation transitions. The particular molecular states responsible for these transitions were identified through theoretical calculations and previous experimental data.
## Experiment Setup
A <sup>23</sup>Na Bose-Einstein condensate containing up to $`10^6`$ atoms in the $`|F=1,m_F=1`$ state was first produced in a magnetic trap and subsequently loaded into a crossed optical dipole trap. The optical trap was derived from a single-mode 1064 nm infrared laser, with the horizontal and vertical beams detuned by 60 MHz through acousto-optic modulators. The number of condensed atoms was varied through three-body decay in a tight trap ($`\omega _{x,y,z}=2\pi \times 200,328,260`$ Hz) , after which the trap was decompressed ($`\omega _{x,y,z}=2\pi \times 110,155,110`$ Hz) to allow further evaporation and re-thermalization. A vertical magnetic field gradient was applied to compensate for gravity and avoid sagging in the weaker trap.
A dye laser operated at 594.710 nm was used to set up a three dimensional optical lattice. The three beams were focused to $`1/e^2`$-waist of $`82`$ $`\mu `$m at the condensate, and retro-reflected to form standing waves. The two horizontal beams were orthogonal to each other, while the third beam was slanted at $`20^{}`$ with respect to the vertical axis due to limited optical access. The three beams were frequency-shifted by $`\pm 30`$ MHz and 80 MHz to eliminate cross interference between different beams.
The gaussian profile of the lattice beams added an additional harmonic trapping potential, while the localization of atoms at the lattice sites increased the repulsive mean field interaction. At the maximum lattice depth, the trap frequencies due to the combined potentials of the optical dipole trap and the lattice beams were $`510`$ Hz for all three dimensions. The trap parameters were chosen such that during the ramping of the optical lattice potential, the overall size of the cloud (parametrized by Thomas-Fermi radii) remained approximately constant in order to minimize intra-band excitations (the mean Thomas-Fermi radius is $`14\mu `$m for $`10^6`$ atoms). The peak per-lattice-site occupancy numbers achieved in our experiment were between 3 to 5.
## Quantum Phase Transition
Atoms held in a shallow optical lattice can tunnel freely from site to site and form a superfluid phase. As the lattice is made deeper, the atomic interaction is increased while the tunneling rate between lattice sites is exponentially suppressed. The system then undergoes a phase transition to an insulating phase – the Mott-insulator – in which each lattice site contains a definite fixed number of atoms. According to the mean-field theory for the homogenous systems of atoms in the lowest band of an optical lattice, the critical point for the phase transition from a superfluid to a Mott-insulator state with $`n`$ atoms per lattice site is determined by fisher89 ; krauth92 ; jaksch98 :
$$U=z(2n+1+2\sqrt{n(n+1)})J$$
(1)
where:
$$U=\frac{4\pi \mathrm{}^2a_s}{m}d^3x|w(x)|^4$$
(2)
is the on-site interaction energy;
$$J=d^3xw^{}(x\lambda _{latt}/2)(\frac{\mathrm{}^2}{2m}^2+V_{latt}(x))w(x)$$
(3)
is the tunneling rate between adjacent lattice sites; $`z`$ is the number of nearest neighbors in the lattice (6 for a cubic lattice); $`m`$ is the atomic mass; $`a_s`$ is the $`s`$-wave scattering length (2.75 nm for <sup>23</sup>Na)); $`w(x)`$ is the Wannier function; $`\lambda _{latt}`$ is the lattice wavelength; $`V_{latt}(x)`$ is the lattice potential.
Figure 1 shows $`U`$ and $`J_n=z(2n+1+2\sqrt{n(n+1)})J`$ for a cubic lattice as a function of the lattice depth, obtained through a band-structure calculation. All energies are expressed in units of the recoil energy $`E_{recoil}=\mathrm{}^2k_{latt}^2/2m`$, where $`k_{latt}=2\pi /\lambda _{latt}`$ is the lattice wavenumber. With this scaling $`J`$ is independent of $`\lambda _{latt}`$. The peak occupancy number in our experiment was $`5`$. From Fig. 1, we find that the the critical points are at a lattice depth of 14.2, 16.2, 17.6, 18.7, and 19.5 (all in units of $`E_{recoil}`$) for $`n=1,2,3,4,`$ and 5 respectively. The inset of Fig. 1 shows that the ratio of $`U/J`$ increases rapidly with increasing lattice depth.
When a weak harmonic trap is present in addition to the lattice potential, as is the case for the experiment, the atomic density is not uniform. Nevertheless, Eqs. (13) can be used to estimate the lattice depth needed to observe the Mott-insulator phase transition at any point in the harmonic trap. Given the local density of the initial condensate, a local value of $`n`$ can be estimated and thus the local critical lattice depth can be read off from Fig. 1. Since the critical depth increases with $`n`$, one expects that as the lattice depth is increased, shells of different occupancies will undergo the transition to the Mott-insulator phase starting from the edge of the density profile and moving in towards the center.
In our experiment, the optical lattice was linearly ramped up to a maximum potential of $`20E_{recoil}`$ in a variable time $`\tau _{ramp}`$ ($`E_{recoil}=h\mathrm{\hspace{0.17em}24.4}`$ kHz for our system). The lattice depth was calibrated by probing the energy difference between the first and the third band at zero quasi-momentum with small amplitude modulation of the lattice beams (see, e.g., schori04 ). After reaching the peak value, the lattice was ramped back down again in $`\tau _{ramp}`$. The ramp sequence was stopped at different times when both the trap and the lattice were suddenly switched off (in $`1\mu `$s). Absorption images were then taken after some time-of-flight as shown in Figure 2. The disappearance of the interference pattern as the lattice depth was increased indicated the loss of phase coherence and a transition from the superfluid state to the Mott insulator state greiner02 . The subsequent revival of the interference patterns as the lattice depth was reduced ensured that the system remained essentially in the ground state during the ramping process. Different $`\tau _{ramp}`$’s were used to check the adiabaticity condition. The peak spontaneous light scattering rate was about 21 s<sup>-1</sup> at the maximum intensity. Therefore for $`\tau _{ramp}10`$ ms, less than 20% of the atoms spontaneously scattered a photon.
After the lattice was fully ramped down, most of the atoms ($`>80\%`$) remained in the condensed fraction while the rest were heated and distributed across the first Brillouin zone. Based on the number of atoms that remained in the condensate after the lattice was fully ramped down, we conclude that $`\tau _{ramp}1`$ ms satisfies the intra-band adiabaticity condition. In the following discussion, all measurements were performed for $`\tau _{ramp}=1,5,10`$ ms, but only the data for $`\tau _{ramp}=5`$ ms are shown as representative of similar results unless otherwise noted.
To characterize the lifetime of the Mott-insulator phase, we held the lattice depth at the maximum level for various amounts of time before ramping the lattice down to $`8E_{recoil}`$ (below the Mott-insulator transition point) and taking the time-of-flight image. If the system remains in the ground state, the contrast of the interference pattern should be recovered, whereas additional heating populates the Brillouin zone and reduces the interference contrast. A cross-section of the density profile was taken along the horizontal direction showing the interference peaks on top of a broad background (see Figure 3). The 5 interference peaks and the broad background were fit by 6 gaussians. The ratio between the total integrated area of the peaks and the background was used as the contrast to quantify the heating of the system. The contrast gives a more sensitive measure of the heating compared to simply counting the recovered condensate atoms. As the atoms in the interference peaks quickly move apart, they are not as broadened by the mean-field expansion as a single condensate.
We performed the same measurement for two different peak occupancy numbers $`n3`$ and 5. Figure 3 shows the decay of the contrast and the lifetime $`\tau `$ was determined using an exponential fit. The fitting error on the lifetime was less than 17%. The lifetime was about 50% longer for $`n=3`$, implying that inelastic collision processes significantly contributed to the heating of the system. The three-body decay rate at the maximum lattice potential for the peak on-site atomic density ($`10^{16}`$ cm<sup>-3</sup> for $`n=5`$) is about 100 s<sup>-1</sup> stamp98odt , consistent with the observed lifetimes of $`10`$ ms. The peak density of a condensate in a harmonic trap and therefore the peak occupancy number scales with $`2/5`$ power of the total number of atoms, and our method for varying the number of atoms (through three body decay) was unable to consistently produce low enough atom numbers for peak occupancy $`2`$. The signal-to-noise ratio of our current imaging system also became marginal for such low atom numbers.
## Photoassociation Resonances
In this experiment, in addition to losses due to three body recombination, we observed large losses of atoms for certain specific tunings of the lattice laser in the range 592 nm to 595 nm. A sample of such a loss feature is shown in Fig. 4. For this measurement, the same ramp sequence was used as before with $`\tau _{ramp}=1`$ ms. The peak intensity is about 280 W/cm<sup>2</sup> in each lattice beam. Due to the intentional frequency shifts between the three lattice beams the effective bandwidth of the lattice light as seen by the atoms is $`100`$ MHz. For the narrow frequency scan range of Fig. 4, the relative frequency scale was determined to better than 25 MHz using a Fabry-Perot cavity with a 2 GHz free spectral range.
Single-photon photoassociations proved to have caused these losses. The lattice laser is tuned by $`160`$ cm<sup>-1</sup> to the red of the atomic $`3^2`$S$`3^2`$P<sub>3/2</sub> transition and thus is in a spectral region where it might drive photoassociation transitions Weiner1999 ; Jones2006 ; Stwalley1999 to rovibrational levels in molecular states dissociating to either the $`3^2`$S$`+3^2`$P<sub>1/2</sub> or $`3^2`$S$`+3^2`$P<sub>3/2</sub> limits. Such a photoassociation transition, followed by the spontaneous radiative decay of the excited molecule into either a bound ground electronic-state molecule or into “hot” atoms, results in significant losses of atoms from the lattice. It is therefore important to identify the locations and strengths of these resonances and choose the appropriate lattice wavelength to avoid such losses.
There is an extensive body of knowledge on the photoassociation of ultracold alkali-metal atoms and the behavior of the molecular potentials dissociating to the $`3^2`$S$`+3^2`$P<sub>1/2</sub> or $`3^2`$S$`+3^2`$P<sub>3/2</sub> limits Weiner1999 ; Jones2006 ; Stwalley1999 . Figure 5 shows the relevant excited molecular potentials as a function of internuclear separation $`R`$. The ground electronic states of Na<sub>2</sub> are the X$`{}_{}{}^{1}\mathrm{\Sigma }_{g}^{+}`$ and a$`{}_{}{}^{3}\mathrm{\Sigma }_{u}^{+}`$ states and two colliding ground state atoms will be some mixture of these symmetries. To the extent that the excited states are well described as $`\mathrm{\Sigma }`$ or $`\mathrm{\Pi }`$ states, the $`gu`$ and $`\mathrm{\Delta }S=0`$ selection rules imply that photoassociation transitions are allowed only to the two $`\mathrm{\Sigma }`$ states (A$`{}_{}{}^{1}\mathrm{\Sigma }_{u}^{+}`$ and (1)$`{}_{}{}^{3}\mathrm{\Sigma }_{g}^{+}`$).
Previous experiments have identified the locations of the strong transitions to the A$`{}_{}{}^{1}\mathrm{\Sigma }_{u}^{+}`$ state Tiemann1996 and the weak transitions to the $`(1)^1\mathrm{\Pi }_g(1_g)`$ state ratliff94 . These are shown in Fig. 6. We looked for but failed to find any significant losses attributable to the weak $`(1)^1\mathrm{\Pi }_g(1_g)`$ state resonances. We were able to confirm one of the A-state resonances, indicated by the dot in Fig. 6. Since the primary goal of the present work was to avoid photoassociation losses we did not investigate known A-state locations further. Our search focused on the strong resonances due to the (1)$`{}_{}{}^{3}\mathrm{\Sigma }_{g}^{+}`$ state which had not been previously observed in this spectral region. In 0.5 GHz steps, we scanned through a 30 GHz range around the theoretically predicted locations based on the model and auxiliary experimental data discussed below. In all but one such scans, we were able to observe between 1 to 3 dips in the remaining atom numbers within a range of $`15`$ GHz, including the loss feature shown in Fig. 4. As shown in Fig. 6 the agreement of the observed locations with the predictions and auxiliary measurements confirms that these loss features are due to photoassociation to the (1)$`{}_{}{}^{3}\mathrm{\Sigma }_{g}^{+}`$ state. The locations of the rovibrational levels of the b$`{}_{}{}^{3}\mathrm{\Pi }_{u}^{}`$ state are not known in the current tuning range. Their spacing, however, should be equal to that of the $`(1)^1\mathrm{\Pi }_g`$ levels and our observations are not consistent with such spacings.
The potential curves used in the calculation of the (1)$`{}_{}{}^{3}\mathrm{\Sigma }_{g}^{+}`$ vibrational levels were generated from an extended version of the model developed by Movre and Pichler for calculating the combined effects of the $`1/R^3`$ resonant dipole interaction and the atomic spin-orbit interaction. Such models have been extensively used to interpret photoassociation experiments Weiner1999 ; Jones2006 ; Stwalley1999 . To the long range potentials generated by the Movre-Pichler type calculation we append the results of ab initio calculations on the short range molecular potentials (in the chemical bonding region). These short range potentials are not sufficiently accurate to allow predictions of absolute vibrational positions. It is necessary to make slight adjustments to these potentials to match experimentally measured vibrational positions, which were obtained through a separate photoassociation spectroscopy experiment in a dark-spot magnetooptical trap (MOT) containing Na $`3^2`$S$`(f=1)`$ atoms at about 300 $`\mu `$K, using a two color ionization scheme and an ion detector jones97 ; Amelink2000 . The photoassociation spectra taken in the MOT have higher resolution than the loss features in the lattice experiment.
In the spectral region of interest for the lattice experiment, three vibrational levels of the (1)$`{}_{}{}^{3}\mathrm{\Sigma }_{g}^{+}`$ were identified by measurements in the MOT. The spectra show a $`1_g`$ component with a complicated hyperfine/rotational pattern, and, slightly higher ($`0.3`$ cm<sup>-1</sup>) in energy, a $`0_g^{}`$ component with a simpler nearly rotational pattern. This ordering of the $`1_g`$ and $`0_g`$ components is in agreement with the Movre-Pichler model. For these levels, the $`J=0`$ feature of the $`0_g^{}`$ component were found to be at 16845.155 cm<sup>-1</sup>, 16852.585 cm<sup>-1</sup>, and 168908.201 cm<sup>-1</sup>, calibrated to iodine lines with an estimated uncertainty of about 0.004 cm<sup>-1</sup> and are shown in Fig. 6. This new data on the (1)$`{}_{}{}^{3}\mathrm{\Sigma }_{g}^{+}`$ was sufficient to calibrate our extended Movre-Pichler model without any further adjustable parameters. The predicted positions shown in Fig. 6 agree well with those of the loss features observed in the lattice experiment, thus identifying the loss features as photoassociation transitions to levels in the (1)$`{}_{}{}^{3}\mathrm{\Sigma }_{g}^{+}`$ state.
Based on this insight, we chose a wavelength of 594.710 nm for our lattice experiment (corresponding to -158.5 cm<sup>-1</sup> in Fig. 6). This tuning lies $`>45`$ GHz from the closest molecular resonance. Given the observed on-resonance photoassociation rate of 1 ms<sup>-1</sup> in the lattice, photoassociative decay can be ignored at such detunings as the rate scales as the square of the ratio of the natural linewidth to the detuning.
## Conclusions
In this paper, we explored the possibility of using a dye laser detuned $`5`$ nm from the Na D lines to study many-body physics of a sodium BEC in an optical lattice, which could allow for an independent control of interaction using magnetic Feshbach resonances. The superfluid to Mott-insulator transition was observed in a Na Bose-Einstein condensate for the first time. The main technical difficulties are due to the heating from the spontaneous light scattering and three-body decay processes. In addition, several photoassociation resonances were observed and identified by means of auxiliary spectroscopy measurements combined with theoretical modeling. These resonances were avoided by choosing an appropriate lattice wavelength. In future experiments, we plan to use a high-power single-mode infrared laser at 1064 nm to eliminate atomic light scattering (Figure 1 also shows the transition points for a 1064 nm lattice). Moreover, heating from three body recombination can be avoided by using occupancy numbers less than three.
## Acknowledgement
The authors would like to thank Fredrik Fatemi and Paul Lett for their help in obtaining the MOT photoassociation data used here, and Widagdo Setiawan for experimental assistance. We also thank Carl Williams for initiating the communications regarding the photoassociation resonances. This research is supported by NSF, ONR, ARO, and NASA. |
warning/0507/quant-ph0507096.html | ar5iv | text | # Non-destructive Orthonormal State Discrimination
## I Introduction
Entangled states play a key role in the transmission and processing of quantum information Niel ; suter . Using an entanglement channel, an unknown state can be teleported bouw with local unitary operations, appropriate measurement and classical communication; one can achieve entanglement swapping through joint measurement on two entangled pairs pan1 . Entanglement leads to increase in the capacity of the quantum information channel, known as quantum dense coding Mattle . The bipartite, maximally entangled Bell states provide the most transparent illustration of these aspects, although three particle entangled states like GHZ and W states are beginning to be employed for various purposes carvalho ; hein .
Making use of single qubit operations and the C-NOT gates, one can produce various entangled states in a quantum network Niel . It may be of interest to know the type of entangled state that is present in a quantum network, at various stages of quantum computation and cryptographic operations, without disturbing these states. Nonorthogonal states cannot be discriminated with certainty wootters , while the discrimination of orthogonal states are possible. A large number of results regarding distinguishing various orthogonal states, have recently been established walgate ; gosh1 ; vermani ; chen . If two copies belonging to the four orthogonal Bell states are provided, local operations and classical communication (LOCC) can be used to distinguish them with certainty. It is not possible to discriminate using only LOCC, either deterministically or probabilistically among the four Bell states, if only a single copy is provided gosh1 . It is also not possible to discriminate multipartite orthogonal states by using LOCC only gosh2 . However, any two multipartite orthogonal states can be unequivocally distinguished through LOCC walgate .
A number of theoretical and experimental results already exist in this area of unambiguous state discrimination cola ; pan2 ; kim . Appropriate unitary transforms and measurements, which transfer the Bell states into disentangled basis states, can unambiguously identify all the four Bell states pan2 ; kim ; boschi . However, in the process of measurement the entangled state is destroyed. Of course, the above is satisfactory when the Bell state is not required further in the quantum network.
We consider in this work the problem of discriminating a complete set of orthogonal basis states in $`C^{d^n}`$– of which the conventional Bell states form a special case– where the $`n`$ qudits ($`d`$-level systems) are distributed among $`n`$ players. We present a scheme which deterministically discriminates between these states without vandalizing them, such that these are preserved for further use. This article is divided as follows. In Section II, we present circuits for the non-destructive Bell state discrimination for $`n`$ qubits shared among $`n`$ players, beginning with the case of conventional Bell states. In Section III, this result is generalized to construct circuits for Bell state discrimination among qudits. In Section IV, we point out the underlying mathematical structure that clarifies how our proposed circuits work. In principle, this can be used to further generalize our results of Section III to discrimination of any set of orthogonal states. In Section V, we examine specific situations where such non-destructive measurements can be useful in computing and cryptography. An appendix is attached at the end, which shows closure property of generalized Bell states, used in the text under Hadamard operations.
## II Bell state discrimination in $`C^{2^n}`$ Hilbert space
In principle, any set of orthogonal states can be discriminated in quantum mechanics, but LOCC may not be sufficient if the state is distributed among two or more players. Here we start with a $`C^{2^n}`$ Hilbert space. To describe any state in this Hilbert space we need $`2^n`$ orthonormal basis vectors. The choice of the basis is not unique, but one choice of particular importance is the set of maximally entangled $`n`$-qubit generalization of Bell states given by:
$`|\psi _x^+`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}(|x+|\overline{x}),`$ (1a)
$`|\psi _x^{}`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}(|x|\overline{x})`$ (1b)
where $`x`$ varies from 0 to $`2^{n1}1`$ and $`\overline{x}1^nx`$ in modulo 2 arithmetic. The set of complete basis vectors (1) reduces to Bell basis for $`n=2`$ and to GHZ states for $`n=3`$. As an example, setting $`n=2`$ in (1) we get the usual Bell states
$`|\psi _{00}=|\psi ^+`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}(|00+|11),`$
$`|\psi _{01}=|\varphi ^+`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}(|01+|10),`$
$`|\psi _{10}=|\psi ^{}`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}(|00|11),`$
$`|\psi _{11}=|\varphi ^{}`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}(|01|10).`$ (2)
A circuit to non-destructively discriminate the generalized orthonormal entangled basis states (1) employing ancilla is shown in Fig. 1. To discriminate the members of the entangled, orthonormal basis set in $`C^{2^n}`$, we have to communicate and carry out measurements on $`n`$ ancillary qubits in the computational basis. The first measurement is done on the state $`|R_{nA_1}`$, as shown in Eq. (3a). This measurement determines the relative phase between $`|x`$ and $`|\overline{x}`$. It will give $`0`$ for $`\frac{1}{\sqrt{2}}(|x+|\overline{x})`$ and $`1`$ for $`\frac{1}{\sqrt{2}}(|x|\overline{x})`$. The next measurements compare the parity between two consecutive bits and yield zero if the bits coincide and one, otherwise. This follows from Eq. (3b), which shows the state for the complex of the system and the $`i`$th ancilla, where $`2in`$. Each ancilla $`A_i`$ is sequentially interacted with the system and then measured. It can be shown (Section III) that this action leaves the states $`|\psi _x^\pm `$ undisturbed. This means that the corresponding measurements, $`M_i`$, represent commuting observables. In general, $`M_1`$ gives the phase bit, and $`M_i`$ gives the parity of the string comprising of the $`i`$th and $`i+1`$th qubits.
In a way clarified in Section IV, $`M_1`$ may be regarded as the non-destructive equivalent of measuring $`X^n`$ and $`M_i`$ ($`2in`$) that of measuring $`ZZ`$, so that the simultaneous measurability of any pair of $`M_i`$’s follows from the fact that $`[X^n,Z(j)Z(k)]=0`$ and $`[Z(j)Z(k),Z(j^{})Z(k^{})]=0`$ where $`Z(j)`$ is the Pauli $`Z`$ operator acting on the $`j`$th qubit.
A note on notation: the sign $`Q(jk)`$ signifies a C-NOT gate, with $`k`$ being (ancilla) control index number, and $`j`$ being (system) target index number. Conversely, $`Q(jk)`$ signifies a C-NOT gate with $`j`$ being (system) control index number and $`k`$ being (ancilla) target index number.
$`|R_{(n\times 2)A1}`$ $`=`$ $`\left[I_2^nH_2\right]\times \left[{\displaystyle \underset{j=1}{\overset{n}{}}}Q(j1)\right]\times \left[I_2^nH_2\right](|\mathrm{\Psi }_{1\mathrm{}n}|0_{A1}),`$ (3a)
$`|R_{(n\times 2)Ai}`$ $`=`$ $`[Q([i1]i)Q\left([ii)\right](|\mathrm{\Psi }_{1\mathrm{}n}|0_{Ai}),`$ (3b)
where $`2in1`$. Therefore, all together we need $`n`$ measurements on $`n`$ ancillary qubits to discriminate $`2^n`$ orthonormal, entangled basis states of the form (1). Furthermore, we require $`3n2`$ applications of CNOT gates. The question of quantity of quantum communication required, which depends on the topology of the quantum communication network, is discussed in Section V in detail.
A proof that the circuit described in Eq. (3), and depicted in Fig. 1 achieves the required Bell state discrimination is deferred to Section III. Here we simply illustrate it using the specific example of the usual Bell states (2). Since (1) reduces to (2) for $`n=2`$, our generalized circuit reduces to that shown in Fig. 2, where one needs only two ancillary qubits, four CNOT gates, two measurements and two qubits of quantum communication.
In Table 1, we have shown the results of the measurements on both the ancillas when different Bell states are present in the given circuit (Fig. 2). Just before measurement, the states can be explicitly written as,
$`|R_{(2\times 2)A1}`$ $`=`$ $`\left[I_2I_2H_2\right]\times [Q(11)Q(21)]\times \left[I_2I_2H_2\right](|\mathrm{\Psi }_{12}|0_{A1})`$ (4a)
$`|R_{(2\times 2)A2}`$ $`=`$ $`\left[Q(12)Q(22)\right]\left(|\mathrm{\Psi }_{12}|0_{A2}\right).`$ (4b)
Thus we have provided a circuit for orthonormal qubit Bell state discrimination shared between two or more parties. These results can be straightforwardly generalized, as shown in the following Section.
## III Generalized Bell state discrimination in $`^{d^n}`$
The results of the preceding Section can be generalized to entangled states of $`n`$ qudits. To this end, we replace the regular Pauli matrices with their $`d`$-dimensional analogs kni96 . We generalize $`X`$ and $`Z`$ gates; these denoted by $`X_d`$ and $`Z_d`$, respectively, have the action:
$`Z_d|j`$ $``$ $`e^{2\pi \iota j/d}|j`$ (5a)
$`X_d|j`$ $``$ $`|j1,`$ (5b)
where the increment in the ket is in mod $`d`$ arithmetic. The operators $`X_d`$ and $`Z_d`$ are related by a Fourier transform $`X_d=H_dZ_dH_d^{}`$, where $`H_d`$ is the generalized Hadamard transformation given by:
$$(H_d)_{jk}=\frac{1}{\sqrt{d}}e^{2\pi \iota jk/d}.$$
(6)
Unlike the qubit case, $`Z_d,X_d`$ and $`H_d`$ are not Hermitian.
The $`d`$ generalized Bell states are
$$|\mathrm{\Psi }_{pq}=\frac{1}{\sqrt{d}}\underset{j}{}e^{2\pi \iota jp/d}|j|j+q,(0p,qd1)$$
(7)
which form an orthogonal, complete basis of maximally entangled vectors for the $`d^2`$ dimensional ”qudit” space ben93 . The parameter $`p`$ denotes phase and $`q`$ the generalized parity. The states $`|\mathrm{\Psi }_{pq}`$ are $`d`$-dimensional analogs of Bell states (2) in that they are eigenstates of the operator $`X_dX_d`$, which is equivalent to the phase observable, whose eigenvalues are $`p`$ or some function $`f(p)`$, and $`Z_d^{}Z_d`$, which is equivalent to the parity observable, whose eigenvalues are $`q`$ or some real-valued function $`f(q)`$. Therefore, measurements equivalent to these operators guarantee a complete characterization of the generalized Bell states. Furthermore, the set of generalized Bell states remains closed under the action $`H_d^{}H`$ or $`H_dH_d^{}`$ or (cf. Appendix A).
The generalization of the CNOT that we require is the one, whose action we define by,
$$𝒞_X:|j|k|j|jk.$$
(8)
The reason for this choice is clarified in Section IV. We use the following notation: the sign $`𝒞_X(jk`$) signifies a C-SUM gate with $`k`$ being (ancilla) control index number, and $`j`$ being (system) target index number; $`𝒞_X(jk`$) signifies a C-SUM gate with the control-target order reversed. A similar terminology extends to the two-qudit gate $`𝒞_X^{}`$, whose action is given by either $`|j|k|j|kj`$ or $`|j|k|jk|j`$, depending on whether the system or ancilla is the control register.
A direct generalization to $`d`$-dimension of Eq. (4) is
$`|R_{(2\times d)A1}`$ $`=`$ $`[I_dI_dH_d]\times \left[𝒞_X(11)𝒞_X(21)\right]\times [I_dI_dH_d^{}](|\mathrm{\Psi }_{12}|0_{A1}).`$ (9a)
$`|R_{(2\times d)A2}`$ $`=`$ $`\left[𝒞_X(12)𝒞_X^{}(22)\right](|\mathrm{\Psi }_{12}|0_{A2}).`$ (9b)
We will denote the observables corresponding to circuits (9a) and (9b) as $`M_1`$ and $`M_2`$, respectively. $`M_1`$ will yield the ‘phase value’ $`p`$, and $`M_2`$ the generalized parity, $`q`$. In a way clarified in Section IV, $`M_1`$ and $`M_2`$ correspond, respectively, to the unitary operations $`XX`$ and $`Z^{}Z`$, so that the simultaneous measurability of $`M_1`$ and $`M_2`$ can be shown as a consequence of the fact that $`[XX,Z^{}Z]=0`$. More directly, we will show that both measurements leave the state $`|\mathrm{\Psi }_{pq}`$ undisturbed.
Let us now consider the more general system of $`n`$ qudits. The elements of the $`d^n`$ dimensional vector space over the modulo $`d`$ field is given by the set $`V_d^{\times n}\{𝐱_j=(x_1,x_2,\mathrm{},x_n)\}`$. Consider the equivalence relation given by $`𝐱_j𝐱_k`$ if and only if $`𝐱_j𝐲_k`$ is a uniform vector, i.e., one of the form $`(r,r,r,\mathrm{},r)`$, where $`r\{0,1,2,\mathrm{},d1\}`$. There are $`d^{n1}`$ equivalence classes, uniquely labeled by the coordinates $`(q_1,q_2,\mathrm{},q_{n1})V_d^{\times (n1)}`$. A complete, maximally entangled Bell basis for the Hilbert space $`^{d^n}`$ can be given by:
$$|\mathrm{\Psi }_{pq_1q_2\mathrm{}q_{n1}}=\underset{j=0}{\overset{d1}{}}e^{2\pi \iota jp/d}|j,q_1+j,q_2+j,\mathrm{},q_{n1}+j.$$
(10)
We call them Bell states in the sense that any state $`|\mathrm{\Psi }_{pq_1q_2\mathrm{}q_{n1}}`$ is an eigenstate of $`X_d^n`$ and $`Z_d(j)Z_d^{}(j+1)`$ ($`1j(n1)`$), which, in a way clarified in Section IV, correspond to observables with eigenvalues $`p`$ and $`q_{j+1}q_j`$ respectively, the latter being called the relative parity.
A generalization of Eq. (9) to $`n`$ qudits is Eq. (11), which describes a circuit to measure phase information $`p`$ and generalized parity information $`q_1,q_2,\mathrm{},q_{n1}`$ of such states. The circuit is depicted in Fig. 3.
The required ancilla are $`n`$ qudits. The corresponding equation is obtained by generalizing Eqs. (3).
$`|R_{(n\times d)A1}`$ $`=`$ $`\left[I_d^nH_d^{}\right]\times \left[\mathrm{\Pi }_{j=1}^n𝒞_X(j1)\right]\times \left[I_d^nH_d\right](|\mathrm{\Psi }_{1\mathrm{}n}|0_{A1}),`$ (11a)
$`|R_{(n\times d)Ai}`$ $`=`$ $`\left[𝒞_{X_d}([i1]i)𝒞_X^{}(ii)\right]\left(|\mathrm{\Psi }_{1\mathrm{}n}|0_{Ai}\right).`$ (11b)
We will denote the measurements realized by these circuits, via ancilla $`A_i`$, by $`M_i`$ ($`1in`$). To see that the $`M_i`$’s are compatible, and that therefore their actions are non-destructive, it turns out to be sufficient to note that $`[X_d^n,Z_d(j)Z_d^{}(k)]=0`$ ($`jk`$) and $`[Z_d(j)Z_d^{}(k),Z_d(j^{})Z_d^{}(k^{})]=0`$ ($`jk`$, $`j^{}k^{}`$), which indeed follows from the fact the states $`|\mathrm{\Psi }_{pq_1q_2\mathrm{}q_{n1}}`$ are eigenstates of $`X_d^n`$ and $`Z_d^{}(j)Z_d(k)`$. We show below explicitly that the $`M_i`$’s measure $`|\mathrm{\Psi }_{pq_1q_2\mathrm{}q_{n1}}`$ non-destructively.
To see this, we note that the action of the first two (boxed) operations in Eq. 11a on a state $`|\mathrm{\Psi }_{pq_1q_2\mathrm{}q_{n1}}|k`$ is
$`|\mathrm{\Psi }_{pq_1,q_2,\mathrm{},q_{n1}}|k`$ $`=`$ $`\left[{\displaystyle \underset{j=0}{\overset{d1}{}}}e^{2\pi \iota jp/d}|j,q_1+j,q_2+j,\mathrm{},q_n+j\right]|k`$ (12)
$``$ $`\left[{\displaystyle \underset{j=0}{\overset{d1}{}}}e^{2\pi \iota jp/d}|j,q_1+jk,q_2+jk,\mathrm{},q_n+jk\right]|k`$
$`=`$ $`\left[{\displaystyle \underset{j^{}=0}{\overset{d1}{}}}e^{2\pi \iota j^{}p/d}|j^{},q_1+j^{},q_2+j^{},\mathrm{},q_n+j^{}\right]|k`$
$`=`$ $`e^{2\pi \iota kp/d}|\mathrm{\Psi }_{pq_1,q_2,\mathrm{},q_{n1}}|k,`$
from which it follows that full effect of the operation described in Eq. (11a) produces the state:
$`|\mathrm{\Psi }_{pq_1,q_2,\mathrm{},q_{n1}}H_d|k`$ $`=`$ $`|\mathrm{\Psi }_{pq_1,q_2,\mathrm{},q_{n1}}\left({\displaystyle \frac{1}{\sqrt{d}}}{\displaystyle \underset{j=0}{\overset{d1}{}}}|j\right)`$ (13)
$``$ $`|\mathrm{\Psi }_{pq_1,q_2,\mathrm{},q_{n1}}\left({\displaystyle \frac{1}{\sqrt{d}}}{\displaystyle \underset{j=0}{\overset{d1}{}}}e^{2\pi \iota pj/d}|j\right)`$
$``$ $`|\mathrm{\Psi }_{pq_1,q_2,\mathrm{},q_{n1}}|p.`$
This yields the phase bit upon the ancilla being measured.
It is easily seen that the action (11b) non-destructively extracts the relative parity information. For,
$`\left[𝒞_{X_d}([i1]i)𝒞_{X_d}^{}(ii)\right]|\mathrm{\Psi }_{pq_1,q_2,\mathrm{},q_{n1}}|0_i`$ (14)
$`=`$ $`𝒞_{X_d}([i1]i){\displaystyle \underset{j=0}{\overset{d1}{}}}e^{2\pi \iota jp/d}|j,q_1+j,q_2+j,\mathrm{},q_{n1}+j|q_{i+1}+j_i`$
$`=`$ $`{\displaystyle \underset{j=0}{\overset{d1}{}}}e^{2\pi \iota jp}|j,q_1+j,q_2+j,\mathrm{},q_{n1}+j|q_{i+1}q_i_i`$
$`=`$ $`|\mathrm{\Psi }_{pq_1,q_2,\mathrm{},q_{n1}}|q_{i+1}q_i_i.`$
The operation $`\left[𝒞_{X_d}([i1]i)𝒞_{X_d}^{}(ii)\right]`$ serves to entangle and then disentangle the input Bell state and the ancilla, such that the relative parity of the two concerned qudits can be read off the latter in the computational basis. This also proves that the circuits given in Eqs. (3), (4) and (9) perform non-destructive Bell state discrimination in dimensions $`2^n`$, $`2\times 2`$ and $`d\times d`$, respectively, for they are all special cases of the circuit described in Eq. (11).
Note that although the circuit for qubits in Fig. 1 and for qudits in Fig.3 use relative parity measurements on consecutive pairs of qudits, they need not do so. Given any set of $`n1`$ relative parity values $`q_jq_k`$ that suffice to fully determine the $`q_j`$’s in a state $`|\mathrm{\Psi }_{pq_1q_2\mathrm{}q_{n1}}`$, our non-destructive measurements are such that the generalized Bell states are eigenstates of such operators, and hence form a complete set of compatible observables. In Section IV, we show that such relative parity measurements correspond to an observable compatible with $`Z_d^{}(j)Z_d(k)`$ (in the $`d=2`$ case, the observable is identical with $`Z(j)Z(k)`$). Depending on the topology of the quantum communication network available, the choice of relative parity measurements can vary. For example, if the communication network has a star topology, as in Fig. 4(a), then the set of observables can correspond to $`Z_d^{}(1)Z_d(j)`$, where 1 is the hub index (marked $`A`$ in the figure), and $`j`$ runs through the remaining vertices. Since any of the operators $`X_d^n`$ and $`Z_d^{}(j)Z_d(k)`$ commute, by corollary 1 (in Section IV) the non-destructive versions of measurements compatible with them can be simultaneously determined.
## IV General circuits for non-destructive orthonormal state discrimination
In this Section, we will examine the basic mathematical structure underlying our circuits. In so doing, we will be able to adapt the ideas of the preceding Sections to the case of any orthonormal state discrimination. As pointed out earlier, the generalized Bell states are eigenstates of the unitary operators $`X_d^n`$ and $`Z_d^{}Z_d`$, where $`d,n2`$. We mentioned that the non-destructive measurement $`M_1`$, effected through the ancilla $`A_1`$ was equivalent to measuring an observable compatible with the unitary operators $`X_d^n`$, while the non-destructive measurement $`M_i`$ ($`2in`$), effected through the ancilla $`A_i`$, was equivalent to measuring an observable compatible with the unitary operators $`Z_d^{}Z_d`$. That is to say, the ancillary measurements are such that $`X_d^n=\mathrm{exp}(2\pi \iota M_1/d)`$ and $`Z_d^{}(i1)Z_d(i)=\mathrm{exp}(2\pi \iota M_i/d)`$.
In the case of $`d=2`$, of course, the observable and the unitary operator, given by the $`XX`$ and $`ZZ`$, are identical though in general this need not be the case. In the context of distributed computing, the separable form of $`X_d^n`$ and $`Z_d^{}(j)Z_d(k)`$ means that observables compatible with them can be evaluated by local measurements and classical communication, but in so doing, the states will of course be destroyed and thus not be available beyond the first measurement, so that multiple copies of the state would be necessary for full discrimination. Our circuits overcome this problem by employing quantum communication, consisting in the movement of the ancillary qubits between players. Note that such quantum communication is necessary, since Bell states, being entangled, possess nonlocal correlations that cannot be accessed locally. Further we note that to ‘outsource’ the measurement of an observable from the system to an ancilla, the system and ancilla are brought into interaction by means of a control operation (CNOT when $`d=2`$) built from the corresponding unitary operation. If this is not entirely clear so far, it is because, as is clarified below, the nature of this interaction can be modified in various ways. In this Section, we will find it convenient to use the notation where the ancilla appears to the left of the system qudit(s).
The above arguments suggest the following generalization that allow us to go beyond Bell state discrimination: that for a Hilbert space of any finite dimension $`d2`$, an observable $`W`$ compatible with a given unitary operator $`U`$ can be effectively measured by ‘outsourcing’ the measurement to an ancilla by means of a suitably generalized control-$`U`$ operation. This is the object of the Theorem 1.
###### Theorem 1
Given unitary operator $`U`$ and an observable $`W`$ compatible with it, measurement of $`W`$ can be outsourced to an ancilla using the controlled operation given by $`𝒞_U_j|jj|U^j`$, where $`\{|j\}`$ is the possibly degenerate, simultaneous eigenbasis of $`U`$ and $`W`$.
Proof. The unitary operator can in general be written in its diagonal basis by $`U=_{j,k}e^{2\pi \iota j/d}|j;kj;k|`$ ($`0jd1`$), where $`k`$ accounts for degeneracy. The observable compatible with it is designated to be $`W=_{j,k}f(j)|j;kj;k|`$, where $`f()`$ is any real-valued function. The state to be measured is some $`|\mathrm{\Psi }=_{k,l}\alpha _{k,l}|k;l`$ entering the upper wire in Fig. (5). At stage 1, the state of the ancilla-system complex is $`d^{1/2}_{j,k,l}\alpha _{k,l}|j|k;l`$. Via action of controlled-$`U`$ gate, in stage 2, the state of the complex is $`d^{1/2}_{j,k,l}\alpha _ke^{2\pi \iota jk/d}|j|k;l`$. At stage 3, by the action of $`H_d^{}`$, the above state is transformed to $`d^{1/2}_{j,k,l,m}\alpha _ke^{(2\pi \iota j/d)(km)}|m|k;l=_{k,l}\alpha _{k;l}|k|k;l`$ since the summation over $`j`$ is non-vanishing only when $`k=m`$. Therefore, a measurement on the ancilla in the computational basis $`\{|j\}`$ is equivalent to a measurement of any observable $`W`$ on the system. $`\mathrm{}`$
It follows from the above that if $`|j;k`$ is an eigenstate of $`U`$, then the outsourced measurement of $`W`$ on $`|j;k`$ will be non-destructive but return the value $`j`$. This gives us the following corollary.
###### Corollary 1
If $`U_1`$ and $`U_2`$ are commuting unitary operators, then the corresponding outsourced observables $`W_1`$ and $`W_2`$ can be simultaneously measured.
If the operator $`U`$ is a product of operations on subsystems, then the control-operation can be done pair-wise on each subsystem and a common ancilla, before the ancilla is finally measured. This is proved in Theorem 2.
###### Theorem 2
The outsourced measurement of observable $`W`$ compatible with unitary operator $`U=_mU_m`$, where $`m`$ ($`=1,2,\mathrm{},n`$) labels the subsystems, can be performed by separate control-operations on the individual subsystems $`j`$ from the same ancilla. The control-operations may be performed in any order.
Proof. Note that $`𝒞_U=_j|jj|_m(U_m)^j=\left(_j|jj|(U_1)^j𝕀^{(m1)}\right)\left(_j^{}|j^{}j^{}|𝕀(U_2)^j^{}𝕀^{(m2)}\right)\mathrm{}`$ $`\left(_{j^{\prime \prime }}|j^{\prime \prime }j^{}|𝕀^{(m1)}(U_2)^{j^{\prime \prime }}\right)`$. Therefore $`𝒞_U=𝒞_{U_1}\times 𝒞_{U_2}\times \mathrm{}𝒞_{U_m}`$, where $`𝒞_{U_k}|jj|(U_k)^j`$. Since the $`𝒞_{U_j}`$’s commute with each other, they may be performed in any order. $`\mathrm{}`$
However, note that though the control operations are separable, there is a quantum communication of the ancilla along the chain formed by the players. The measurement of $`M_1`$ in the preceding Section can be seen as a special case of Theorems 1 and 2. To see this, we set $`UX^n`$, where each $`U_i=X_d`$. Since $`X^n|\mathrm{\Psi }_{pq_1,\mathrm{},q_{n1}}=e^{2\pi \iota p/d}|\mathrm{\Psi }_{pq_1,\mathrm{},q_{n1}}`$, by Theorem 1, the observable $`M_1_{p,q_1,\mathrm{}}f(p)|\mathrm{\Psi }_{pq_1,\mathrm{},q_{n1}}\mathrm{\Psi }_{pq_1,\mathrm{},q_{n1}}|`$ can be outsourced using the control operation $`𝒞_U|jj|U^j`$. In view of Eq. (5b), this has the effect: $`𝒞_U:|j|j_1\mathrm{}|j_n|j|j_1+j\mathrm{}|j_n+j`$. It then follows from Theorem 2 that $`𝒞_U`$ can be broken into $`n`$ applications of $`𝒞_X`$ operations on an ancilla-qudit pair, for each qudit of the system and a fixed ancilla, where $`𝒞_X`$ is precisely the operation defined in Eq. (8). In a distributed computing scenario, this ancilla must be sequentially interacted with each system qudit. This clarifies our use of the Eq. (8) as the generalization of the CNOT gate. We also obtain the general Bell state discrimination circuit described in Eq. (11a) as a special case of Theorems 1 and 2.
In general, given any set of orthonormal states that form a complete basis to an observable $`W`$, Theorem 1 allows us to ‘outsource’ their measurement to an ancilla. To do so, we first construct a unitary operator $`U`$ with respect to which these states are ‘dark’, i.e., of which these states are eigenstates, and using this to construct a control-$`U`$ operation $`𝒞_U`$. If $`U`$ is separable, as is the case in our problem, then Theorem 2 allows $`𝒞_U`$ to be broken up into a sequence of pair-wise control gates.
Consider measurement of the relative parity observable $`Z_d(i1)Z_d^{}(i)`$. Following Theorems 1 and 2, the measurement here can be outsourced using control-$`Z_d^{}`$ ($`𝒞_{Z_d^{}}`$) and control-$`Z_d`$ ($`𝒞_{Z_d}`$) operations from the ancilla sequentially to the two qudits. According to Eq. (5a), these require controlled-phase operations. However, by means of applying Hadamards, it is possible to turn them into $`𝒞_X`$ operations. To see this, we note that for any integer $`j`$,
$`(Z^{}Z_d)^j`$ $`=`$ $`(Z_d^{})^j(Z_d)^j`$ (15)
$`=`$ $`(H_dX_d^{}H_d^{})^j(H_dX_dH_d^{})^j`$
$`=`$ $`(H_d(X_d^{})^jH_d^{})(H_d(X_d)^jH_d^{})`$
$`=`$ $`(H_dH_d)\times (X_d^{}X_d)^j\times (H_d^{}H_d^{}).`$
This means that the outsourcing of measurement of $`Z_d^{}Z_d`$ is equivalent to the circuit in Fig. 6(a), where only $`𝒞_X`$ and $`𝒞_X^{}`$ are used.
The last result we require says that, by dropping the Hadamards in Fig. 6(a), we can reverse the control direction. This is shown in Theorem 3. Two advantages of such a step is that for each outsourced measurement of $`Z_d^{}Z_d`$, the number of Hadamards is reduced by a factor of six and furthermore instances of only one nonlinear gate (namely, $`𝒞_{X_d}`$ or $`𝒞_{X_d^{}}`$) need to be used.
###### Theorem 3
The two measurement circuits depicted in Fig. 6 are equivalent.
Proof. Let the incoming state of the two system wires be the pure state $`|\mathrm{\Psi }=_{jk}\alpha _{jk}|j|k`$ (we ignore the fact that the summation can run on a single index on account of Schmidt decomposability). At stage 1, the state of the ancilla-system complex is: $`(1/\sqrt{d})\left(_l|l\right)\left(_{j,k,j^{},k^{}}\alpha _{jk}\mathrm{exp}[(2\pi \iota /d)(jj^{}+kk^{})]|j^{}|k^{}\right)`$. By the action of the two control-gates, the state in stage 2 is $`(1/\sqrt{d})\left(_{l,j,k,j^{},k^{}}\alpha _{jk}\mathrm{exp}[(2\pi \iota /d)(jj^{}+kk^{})]|l|j^{}l|k^{}l\right)`$. In stage 3, by the action of the three Hadamards, the state $`|\mathrm{\Psi }^{}`$ of the complex is
$`|\mathrm{\Psi }^{}`$ $`=`$ $`(1/\sqrt{d})\left({\displaystyle \underset{l,j,k,j^{},k^{},j^{\prime \prime },k^{\prime \prime }}{}}\alpha _{jk}\mathrm{exp}\left[(2\pi \iota /d)(jj^{}+kk^{}ll^{}+j^{\prime \prime }[j^{}l]k^{\prime \prime }[k^{}k])\right]|l^{}|j^{\prime \prime }|k^{\prime \prime }\right)`$ (16)
$`=`$ $`(1/\sqrt{d})\left({\displaystyle \underset{l,j,k,j^{},k^{},j^{\prime \prime },k^{\prime \prime }}{}}\alpha _{jk}\mathrm{exp}[(2\pi \iota /d)(l(l^{}j^{\prime \prime }+k^{})+j^{}[j^{\prime \prime }j]+k^{}[kk^{\prime \prime }])]|l^{}|j^{\prime \prime }|k^{\prime \prime }\right)`$
$`=`$ $`(1/\sqrt{d})({\displaystyle \underset{l,l^{},j,k}{}}\mathrm{exp}[(2\pi \iota /d)(l(l^{}j+k)]|l^{}|j|k)`$
$`=`$ $`{\displaystyle \underset{j,k}{}}\alpha _{jk}|kj|j|k,`$
which is the situation described by the circuit in Fig. 6(b). In general, the two wires, being part of a larger system, are in a mixed state. Since a mixed state can be regarded as an ensemble of pure states, Eq. (16) implies the equivalence of the circuits in the Fig. 6(a) and 6(b) even for mixed states. $`\mathrm{}`$
From Theorems 1, 2 and 3, it follows that the circuitry described by Eq. (11b), or equivalently, depicted in the second bounded box of Fig. 3, indeed outsources measurement of $`Z_dZ_d^{}`$. More generally, Theorem 3 can be used to reverse the direction of control in the outsourcing of two-qudit observables, by replacing $`U`$ with $`H_dUH_d^{}`$ as the unitary operator on which the control gate is based.
## V Some applications
Such non-destructive state discrimination can be useful in distributed quantum computing, especially when there are restrictions coming from the topology of the quantum communication network. Unlike their classical counterparts, quantum channels are expected to be expensive and not amenable to change to suit a problem at hand. Rather, it is worthwhile to use protocols that minimize quantum communication complexity, that is, the quantity of quantum information that must be communicated between different parties to perform a computation or process some information, in a given network.
A simple way to perform Bell state discrimination is for all other members to communicate their qudits to single station, whose member (called, say Alice) performs a joint measurement on all $`n`$ qubits or qudits to determine the state. She then re-creates the measured state and transmits them for further use. Actually, in the present situation, instead of a joint measurement on all qubits, Alice can apply a string of $`n1`$ $`𝒞_X^{}`$ operations on each consecutive pair of qudits in the Bell state $`|\mathrm{\Psi }_{pq_1q_2\mathrm{}q_{n1}}`$ and $`H_d^{}`$ finally on the first qudit. It is easily seen that each application of $`𝒞_X^{}`$ will disentangle the controlled qudit from the rest. For the Bell states, this procedure effects the transformation:
$$|\mathrm{\Psi }_{pq_1q_2\mathrm{}q_{n1}}|p|q_2q_1\mathrm{}|q_{n1}q_{n2}.$$
(17)
Subsequent measurement of each qudit in the computational basis completely characterizes the Bell state. The Bell state thus being discriminated, the above procedure can be reversed to re-create the state $`|\mathrm{\Psi }_{pq_1q_2\mathrm{}q_{n1}}`$ and transmit it back to the remaining players.
Irrespective of network topology, such a disentangle-and-reentangle strategy requires in all $`2(n1)`$ two-qudit gates to be implemented. In our method, the number of two-qudit gates is the sum of $`n`$ two-qudit gates for determining phase parameter $`p`$ and $`2(n1)`$ for determining the (relative) parities, giving $`3n2`$ two-qudit gates. From this viewpoint of consumption of nonlinear resources, our method does not offer any advantage. However, this turns out not to be the case from the viewpoint of quantum communication complexity.
Suppose a quantum communication network with a star topology and $`n`$ members is given, as for example in Fig. 4(a). For all members to transmit their qudits to Alice (at $`A`$), and for her to transmit them back would require $`2(n1)`$ qudits to be communicated, where the factor 2 comes from the two-way requirement. In our protocol, one way quantum communication suffices. For measuring the ‘phase observable’ $`M_1`$, the number of qudits communicated is seen to be $`2(n1)`$, since the ancilla must pass through the hub to reach each member on a single-edge vertex; and if measured edgewise, the communication complexity for relative parity measurement is $`n`$ qudits. In all, this requires $`3n2`$ qudits to be communicated, which is larger than that required for a plain disentangle-reentangle method.
However consider a linear configuration of the communication network, as in Fig. 4(b), where members are linked up in a single series. In the disentangle-reentangle method, if Alice is located at one end, the communication complexity is seen to be $`n(n1)`$ qudits; it is $`(n^21)/2`$ if she is in the middle. In either case, it is of order $`O(n^2)`$. In contrast, our non-destructive method can be implemented using $`n1`$ qudits communicated both for phase and relative parity measurement, requiring in all only $`2(n1)`$ qudits to be communicated, so that the required communication is only of order $`O(n)`$. Thus our method gives a quadratic saving in quantum communication complexity.
A further advantage, that may be of some importance in certain situations, is that our method divides the required resources in terms of applying nonlinear gates and of measurements equally among the various members. In a real life situation, this may facilitate the distribution of quantum information processing resources among the various members.
###### Acknowledgements.
We are thankful to Prof. J. Pasupathy, V. Aravindan and H. Harshavardhan, Dr. Ashok Vudayagiri, Dr. Ashoka Biswas, Dr. Shubhrangshu Dasgupta for useful discussions.
## Appendix A Closure of generalized Bell states under Hadamards
The action of $`HH^{}`$ on $`|\mathrm{\Psi }_{pq}`$ on the states in Eq. (9) produces the effect of effectively interchanging the indices $`pq`$ of $`|\mathrm{\Psi }_{pq}`$:
$`(HH^{})|\mathrm{\Psi }_{pq}`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{d}}}{\displaystyle \underset{j,k,l}{}}e^{(2\pi \iota /d)(j[p+kl]ql)}|k|l`$ (18)
$`=`$ $`{\displaystyle \frac{1}{\sqrt{d}}}{\displaystyle \underset{j,l}{}}e^{(2\pi \iota /d)(ql)}|lp|l`$
$`=`$ $`{\displaystyle \frac{1}{\sqrt{d}}}{\displaystyle \underset{j}{}}e^{(2\pi \iota /d)([dq]l)}|j|j+p,`$
$`=`$ $`|\mathrm{\Psi }_{q^{}p},`$
where $`q^{}=(dq)modd`$ and the second step follows from noting that the only non-zero contributions come for the case $`p+kl=0`$, and an overall phase factor has been dropped in the third step. Similarly, one finds $`(HH^{})|\mathrm{\Psi }_{pq}=|\mathrm{\Psi }_{qp^{}}`$, where $`p^{}=dp`$ mod $`d`$. |
warning/0507/q-bio0507012.html | ar5iv | text | # Activation and protonation of dinitrogen at the FeMo cofactor of nitrogenase
## I Introduction
Although dinitrogen is the main part of the atmosphere, it is, in its molecular form, inaccessible to biological organisms. Its inactivity is caused by the triple bond—one of the strongest covalent bonds in nature. While high pressure and high temperature are required to convert N<sub>2</sub> to NH<sub>3</sub> in the industrial Haber-Bosch process, biological nitrogen fixation achieves the same goal at ambient conditions. For this purpose nature employs the enzyme nitrogenase, which is one of the most complex bioinorganic catalysts in nature.
Nitrogenase converts N<sub>2</sub> to biologically accessible ammonia.Burges (1990); Burges and Lowe (1996); Eady (1996); Christiansen et al. (2001) During the reaction, non-stoichiometricHadfield and Bulen (1969) amounts of hydrogen are produced.
$`\text{N}_2+(6+2x)\text{H}^++(6+2x)\text{e}^{}2\text{NH}_3+x\text{H}_2.`$
Values for $`x`$ range from $`<`$1 (Ref. Rivera-Ortiz and Burris, 1975) upwards and there is an ongoing discussion about the question whether hydrogen is produced in stoichiometric amounts.Rees and Howard (2000) In the gas phase, the first protonation is by far the most difficult step in the reaction. The catalyst has to find a way to dramatically reduce this barrier.
The active center of the enzyme, shown in Fig. 1, is the FeMo cofactor, MoFe<sub>7</sub>S<sub>9</sub>N$``$homocitrate. The FeMo cofactor is linked to the protein via two amino acid residues. The question arises why nature employs such a large multi-center cluster.
The reaction mechanism has being studied for over 40 years, but the atomistic mechanism of substrate conversion at the FeMo cofactor still remains an open issue. In a previous paper, we presented the most salient features of the reaction cycle as it emerged from our simulations.Kästner and Blöchl (2005a) In the present paper, we present the results of state-of-the-art first principles calculations of the most difficult step in the reaction pathway, the first protonation of N<sub>2</sub> bound to the FeMo cofactor.
Nitrogenase consists of two proteins: (1) the molybdenum-iron protein, which holds the FeMo cofactor (FeMoco) and (2) the iron protein, which hydrolyzes MgATP and uses the energy obtained to provide the molybdenum-iron protein with electrons.
The structures of the two proteins were resolved already in 1992.Kim and Rees (1992a, b); Georgiadis et al. (1992) Since then increasingly more refined crystallographic data became available M.K.Chan et al. (1993); Peters et al. (1997); Mayer et al. (1999) until, in 2002, a central ligand of the FeMoco, shown in Fig. 1 was found,Einsle et al. (2002) which was undetected in previous studies. While crystallographic studies determined the central ligand to be either a C, N, or O atom, theoretical studiesHinnemann and Nørskov (2003); Dance (2003); Lovell et al. (2003); Huniar et al. (2004) provide strong support for it to be a nitrogen atom. It turns out that the central ligand (N<sub>x</sub>) plays a critical role in the mechanism.Kästner and Blöchl (2005a)
Kinetic studies of the mechanism of biological nitrogen fixation Lowe and Thorneley (1984a); Thorneley and Lowe (1984a); Lowe and Thorneley (1984b); Thorneley and Lowe (1984b) indicate that the rate-limiting step of the reaction is the dissociation of the two proteins. In each of these association-dissociation cycles one electron is transferred to FeMoco. The so-called Thorneley-Lowe scheme provides insight into the first reduction steps, stating that N<sub>2</sub> binds after three or four electrons have been transferred to the MoFe protein. Theoretical modelsKurnikov et al. (2001); Liao and Beratan (2004) indicate that geometrical changes of the backbone of the Fe-protein are responsible for using the energy from MgATP to transfer electrons to the MoFe protein.
Since the appearance of the first crystal structures, nitrogenase has been subject to numerous theoretical investigations.Deng and Hoffmann (1993); Plass (1994); Dance (1994, 1996, 1997, 1998); Machado and Davidson (1995); Stavrev and Zerner (1996, 1997, 1998); Zhong and Liu (1997); Siegbahn et al. (1998); Rod et al. (1999); Rod and Nørskov (2000); Rod et al. (2000); Szilagyi et al. (2000, 2001); Durrant (2001a, b, 2002a, 2002b)
Barrière et al. (2001); Lovell et al. (2001, 2002a, 2002b, 2002c); Cui et al. (2003) Unfortunately, the central ligand was unknown at that time, so that the conclusions have to be reconsidered. However, several more recent theoretical studies Hinnemann and Nørskov (2003); Dance (2003); Lovell et al. (2003); Vrajmasu et al. (2003); Huniar et al. (2004); Schimpl et al. (2003); Kästner and Blöchl (2005a) have been performed taking the central ligand into account.
A critical parameter for theoretical calculations is the oxidation state of the cofactor. By comparing theoretical resultsDance (2003); Lovell et al. (2003); Schimpl et al. (2003) with experimental observations, a consensus has been reached that the cofactor in the resting state is charge neutral, i.e., \[MoFe<sub>7</sub>S<sub>9</sub>N\]<sup>0</sup>.
The mechanisms for nitrogen fixation proposed up to now may be divided in two main groups: (1) conversion at Mo and (2) conversion at Fe.
* The Mo atom has been the target of numerous theoreticalPickett (1996); Grönberg et al. (1998); Szilagyi et al. (2001); Durrant (2002a, b) studies. A strong indication in favor of Mo comes from experiment: an Mo-based model complexYandulov and Schrock (2003) has been found to catalytically reduce N<sub>2</sub>. However, there are also active nitrogenasesEady (1996) with Mo replaced by V or Fe, which indicates that Mo is not essential.
* Numerous proposals have been made for the reduction involving the Fe atoms. (1) The direct way for N<sub>2</sub> to bind at Fe atoms is head-on binding.Dance (1997, 1998); Rod et al. (1999); Rod and Nørskov (2000); Rod et al. (2000); Hinnemann and Nørskov (2004) In this mechanism N<sub>2</sub> binds to one Fe atom of the central cluster, where it is protonated until one and then the second ammonia dissociate. The intermediates of the complete cycle have been identified, albeit still without the central ligand by Rod and co-workersRod et al. (1999); Rod and Nørskov (2000); Rod et al. (2000) Recently, Hinnemann and NørskovHinnemann and Nørskov (2004) extended this proposal to the cofactor with central ligand. The theoretical predictions are, however, at variance with this mechanism, because the calculations show that N<sub>2</sub> binding is highly endothermic.Hinnemann and Nørskov (2004) (2) Sellmann and co-workersThorneley et al. (1996); Sellmann et al. (1999, 2000); Kirchner et al. (2005) suggested an opening of the cage in analogy to smaller Fe complexes. In his model, two octahedrally coordinated low-spin Fe atoms positioned in close proximity bind dinitrogen between them, where it is reduced. The mechanism has been lined out up to the first four protonations. (3) Our own recent calculationsSchimpl et al. (2003); Kästner and Blöchl (2005a) support the view of cage opening even though, in our model, the Fe atoms are in a high-spin tetrahedral coordination, which points towards a quite different chemistry than expected for low-spin octahedral Fe atoms. The opening and closing of an SH bridge at the cofactor is critical for the activation of N<sub>2</sub> and for the dissociation of the second ammonia. A critical role is attributed to the central ligand, which enables required bond rearrangements. In a similar approach, we were able to explain many experimental findings in the conversion of acetylene by nitrogenase.Kästner and Blöchl (2005b) (4) Another proposalHuniar et al. (2004) suggests the opening of a sulfur bridge upon coordination of water to an Fe atom, complete protonation of the central ligand, and dissociation of ammonia. Then N<sub>2</sub> inserts into the central cavity of the cofactor, where one nitrogen atom is fully protonated and dissociated, which closes the catalytic cycle. This intriguing proposal seems to be in conflict with isotope exchange (ESEEM/ENDOR) experimentsLee et al. (2003) that exclude an exchange of a central nitrogen ligand.
## II Calculational Details
We considered the complete FeMo cofactor with truncated ligands as in the previous study.Schimpl et al. (2003) The histidine was replaced by imidazole, the homocitrate ligand by glycolate and the cysteine, bound to the terminal iron atom by an SH group.
We performed density-functional theoryHohenberg and Kohn (1964); Kohn and Sham (1965) (DFT) calculations based on the projector-augmented waveBlöchl (1994); Blöchl et al. (2003) (PAW) method. The gradient-corrected PBE (Ref. Perdew et al., 1996) functional was used for exchange and correlation. The PAW method is a frozen-core all-electron method. Like other plane-wave-based methods, the PAW method leads to the occurrence of artificial periodic images of the structures. This effect was avoided by explicit subtraction of the electrostatic interaction between them.Blöchl (1986) Wave-function overlap was avoided by choosing the unit cell large enough to keep a distance of more than 6 Å between atoms belonging to different periodic images. We used a plane wave cutoff of 30 Ry for the auxiliary wave functions of the PAW method. The following shells were treated in the frozen-core approximation: Fe \[Ne\], Mo \[Ar3d<sup>10</sup>\], S \[Ne\], O \[He\], N \[He\], and C \[He\]. The following sets of projector functions were employed, Fe $`2s2p2d`$, Mo $`2s2p2d`$, S $`2s2p2d`$, O $`2s2p1d`$, N $`2s2p1d`$, C $`2s2p1d`$, and H $`2s1p`$, which provides the number of projector functions per angular momentum magnetic and spin quantum numbers $`m`$, $`s`$ in each main angular momentum channel $`\mathrm{}`$.
Atomic structures were optimized by damped Car-ParrinelloCar and Parrinello (1985) molecular dynamics with all degrees of freedom relaxed. The convergence was tested by monitoring if the kinetic temperature remains below 5 K during a simulation of 0.05 ps (200 time steps). During that simulation, no friction was applied to the atomic motion and the friction on the wave function dynamics was chosen sufficiently low to avoid a noticeable effect on the atomic motion.
Transition states were determined by applying a one-dimensional constraint on the atomic positions. In the present application, bond-length, angle, and torsion constraints were used. To get a first upper bound for the barrier, the specific constraint was varied within 1000 molecular-dynamics (MD) steps. If this upper bound is less than 20 kJ/mol, the barrier will be easily overcome and it has not been calculated more accurately. In case of a higher estimate, the bond length was fixed to discrete values around the transition state to maximize the energy while all unconstrained degrees of freedom were allowed to relax to minimize the energy. Proof that this approach, when converged, determines exactly first-order transition states is given elsewhere.Blöchl and Togni (1996)
The reaction rates $`\mathrm{\Gamma }`$ can be estimated using $`\mathrm{\Gamma }=\mathrm{\Gamma }_0e^{E_\mathrm{A}/(k_\mathrm{B}T)}`$ from the calculated activation energy $`E_\mathrm{A}`$ and a typical vibrational frequency $`\mathrm{\Gamma }_0=3\times 10^{13}\text{s}^1`$ corresponding to about 1000 cm$`^1.`$
The FeMoco exhibits seven high-spin iron atoms antiferromagnetically coupled to each other. Many different spin configurations may easily lead to metastable states in conventional collinear spin-polarized calculations. Therefore, we used a noncollinear description of the spin density for our calculations. In a noncollinear description each one-electron wave function is a two-component spinor wave function.Sandratskii and Guletskii (1986); Kübler et al. (1988); Oda et al. (1998); Hobbs et al. (2000) This method not only correctly describes real noncollinear spin states, which occur within the reaction mechanism, but also avoids artificial barriers between different spin configurations occurring in collinear calculations. Our resulting spin distribution is therefore independent of the random starting conditions. Such a dependence is a common problem of conventional (collinear) spin-polarized calculations for this system. We found that the spin ordering depends on subtle changes in the atomic structure. It also changes between different states of the reaction mechanism. The spin orderings encountered in our calculations are given in Fig. 2, where we follow the notation introduced by Lovell et al.Lovell et al. (2001) noncollinear spin arrangements have been found, in this study, only for energetically unfavorable states, which is why we do not specify them further.
The spin quantum number $`S`$ is specified alongside with the corresponding structures in Figs. 3 and 4. A spin with $`S=1`$, for example, corresponds to a triplet. Noncollinear states are indicated by a spin quantum number that differs from half-integer values. This corresponds, in analogy with the collinear expression, to the absolute value of the integrated spin density divided by $`\mathrm{}`$.
During the reaction, protons and electrons are transferred to the cofactor and the substrate. In this work we made the assumption that electrons and protons are transferred in a correlated way, i.e., that one proton is transferred with each electron transfer. This assumption implies one of the two scenarios: either a reduction of the cofactor increases the proton affinity such that a proton transfer is induced, or, if the proton transfer precedes the electron transfer, it implies that the electron affinity is sufficiently enhanced by the positive charge next to the cofactor to induce an electron transfer to the cofactor. This is the main assumption in our work besides the accuracy of the density functionals and the neglect of the protein environment. This assumption has been shown to be valid for the cofactor before the binding of the substrate.Schimpl et al. (2003)
The energies of protons and electrons, which are consumed during the reaction, affect the overall reaction energy. Thus we need to define a value that reflects their energies in the environment. For protons, the relevant environment is the proton transfer channel, while for electrons it is expected to be the P-cluster. The exact energies cannot be determined by theory alone.
As a consequence of our assumption that the reduction and protonation occur in a correlated manner, only the sum $`\mu _\mathrm{H}`$ of the energies of protons and electrons is relevant for the relative energies of the intermediates. A range of possible values can be derived by comparing experimental x-ray and extended x-ray-absorption fine structure (EXAFS) data with our calculated geometries: we found indirect evidence for the cofactor being unprotonated in the resting state, while being protonated in the reduced state.Schimpl et al. (2003) Therefore, $`\mu _\mathrm{H}`$ is sufficiently high to drive protonation, that is, $`\mu _\mathrm{H}>E(`$MH$`)E(`$M$`)`$. On the other hand, no protonation occurs under the same conditions, but in the absence of MgATP. Thus the chemical potential in the absence of MgATP, denoted by $`\mu _\mathrm{H}^{}`$, must be sufficiently low not to drive protonation, that is, $`\mu _\mathrm{H}^{}<E(`$MH$`)E(`$M$`)`$. As two MgATP are hydrolyzed in each electron transfer, the difference between the chemical potentials with and without MgATP is smaller than twice the energy of hydrolysis of MgATP, that is, $`\mu _\mathrm{H}\mu _\mathrm{H}^{}<64.4`$ kJ/mol.Voet et al. (2002) It is smaller because a fraction of the energy supplied by MgATP will be dissipated. Therefore, we assume the lower bound for $`\mu _\mathrm{H}`$, that is, $`\mu _\mathrm{H}=E(`$MH$`)E(`$M$`)`$, in our calculations. This is the most conservative assumption possible. A less conservative value would make the reactions including protonation more exothermic.
Previous studiesRod et al. (1999); Rod and Nørskov (2000); Rod et al. (2000); Hinnemann and Nørskov (2004); Huniar et al. (2004) made a different choice for $`\mu _\mathrm{H}`$, namely, $`\mu _\mathrm{H}=\frac{1}{2}E(`$H$`{}_{2}{}^{})`$. This would be the appropriate choice if the protons and electrons would be obtained from gaseous hydrogen. While the production of gaseous hydrogen 2H$`{}_{}{}^{+}+2e^{}`$H<sub>2</sub> is energetically neutral with this choice, this reaction is exothermic by 71 kJ/mol when using our choice of $`\mu _\mathrm{H}=\frac{1}{2}E(`$H$`{}_{2}{}^{})+35`$ kJ/mol. Thus our reaction energies can be translated to the values for H<sub>2</sub> reference by adding 35 kJ/mol per added hydrogen atom. We additionally list the reaction energies with H<sub>2</sub> as the reference energy in parentheses after the values we obtain with our $`\mu _\mathrm{H}`$ for all those energies were influenced by the choice of $`\mu _\mathrm{H}`$.
In our work we evaluate not only the energetics of the intermediates, but also the barriers for the transitions. To estimate the barriers for protonation, we need, however, to specify a proton donor, which models the proton channel. We used the ammonium ion to mimic the proton donor. Note, however, that this choice only affects the barriers but not the relative energies of the intermediates. Our finding that the barriers for protonations are small and will therefore be overcome easily indicates that the barriers for protonation are not relevant for the overall picture.
The notation for the structures is chosen as follows. The complexes of dinitrogen with the cofactors are given in letters in alphabetic order according to the number of proton transfers and in numerals for their energetic order. A numeral $`0`$ denotes the ground state for the selected composition.
## III Results
The most difficult step of the reaction from N<sub>2</sub> to NH<sub>3</sub> is the first protonation of dinitrogen. In the reaction in the gas phase we find that the reaction step N<sub>2</sub> \+ H<sup>+</sup> \+ e$`{}_{}{}^{}`$ N<sub>2</sub>H is endothermic with 164 kJ/mol using the choice for $`\mu _\mathrm{H}`$ described under calculational details (and 199 kJ/mol with H<sub>2</sub> as reference). The main goal for nitrogenase must be to lower this barrier. At the FeMo cofactor, the reaction energy is dramatically reduced to only 41 kJ/mol (76 kJ/mol). Nevertheless, in the catalyzed reaction this remains the most endothermic step.
We explored eight different isomers of the N<sub>2</sub>H adduct bound to the MH<sub>2</sub> state of the cofactor. MH<sub>2</sub> represents the resting state (M) of the cofactor reduced by two electrons and protonated at two of the sulfur bridges. The isomers are shown in Fig. 4 and will be discussed in the following.
### III.1 Protonation of the central ligand
Nitrogen in the bridged binding mode A1 opens the structure of the cofactor and leaves space for the coordination at the central ligand. Protonation of the central ligand leads to the most stable isomer with one protonated nitrogen, namely B0, which is lower in energy by 37 kJ/mol than the second most stable configuration, the protonated dinitrogen bridge B1.
There is only limited space for a donor to access the central ligand in A1. Correspondingly, we find a large barrier for protonation of 79 kJ/mol from NH$`{}_{}{}^{+}{}_{4}{}^{}`$, which was used as the model for the proton donor. This barrier corresponds to a reaction rate of the order of 0.1–1 s<sup>-1</sup>, which is substantially lower than the electron transfer rate. Furthermore, as discussed below, A1, the common starting configuration for both B0 and B1 can be protonated with a negligible barrier at dinitrogen leading to B1. Since the pathways to B0 and B1 are in direct competition, protonation of the central ligand is kinetically hindered. Therefore, we consider a reaction path via protonation of the central ligand unlikely.
In B0, one iron atom is in a planar threefold coordination shell, which appears to be an unusual configuration. To explore if this structure is an artifact of our structural model, we investigated if this iron atom could form a complex bond to a nearby water molecule, which would restore a tetrahedral coordination of the Fe atom. However, in our calculations including an additional water molecule, no significant complex bonds between the three-coordinate Fe atom and a water are formed: for B0 an Fe–O distance of 2.7 Å was calculated, which is substantially larger than typical complex bonds to Fe.
One may ask if the solvent effects affect the barrier to access B0. We find that the interaction between an additional H<sub>2</sub>O molecule and the triangular Fe atom in the initial state A1 and in the transition state is even weaker than in the final state B0, expressed in even longer Fe–O distances. From the absence of strong interactions we conclude that the large barrier is not appreciably influenced by the solvent.
A slow rate of formation does not rule out that the cofactor is accidentally trapped in this low-energy intermediate. Therefore, we investigated the subsequent steps starting from B0. We find that after one reduction the next proton attaches to the bridged dinitrogen, which is similar to the second most stable intermediate discussed below. However, upon protonation of dinitrogen, the proton at the central cluster is destabilized and is transferred to the unprotonated nitrogen atom of dinitrogen. The reaction energy for the internal proton transfer from the central ligand to the protonated dinitrogen is $`69`$ kJ/mol. The reaction has a barrier of 20 kJ/mol. Thus even if the central ligand is protonated, the reaction mechanism quickly leads back to what we consider the most likely pathway. Intermediate structures of this side branch of the reaction cycle are shown in Fig. 5.
### III.2 Protonation of bridged dinitrogen
The second most stable isomer with one protonated nitrogen is the bridged binding mode B1 shown in Fig. 4.
As shown previously,Schimpl et al. (2003) dinitrogen can dock at the FeMo cofactor MH<sub>2</sub> in two configurations with similar energies. They are shown in Fig. 3. The axial binding mode, denoted by A0, is slightly more stable than the bridged one, denoted by A1.
One of the most interesting questions of biological nitrogen fixation is how dinitrogen is activated for the first protonation. Therefore, we investigated the two binding modes in some detail.
The most relevant geometric parameters and the N–N stretching vibration frequencies of dinitrogen bound to the cofactor are given in Table 1 and compared to the experimental data obtained from a model complex with a N<sub>2</sub> bridging two low-coordinated iron atoms.Smith et al. (2001) The elongation of the N–N bond as well as the reduction of the vibration frequency with respect to gaseous N<sub>2</sub> are indications of the activation of the triple bond for the following protonation. The elongation of the dinitrogen bond and the reduction of the stretch frequency in A1 compares well with the model complex, while A0 appears to be much less activated.
The activation of A1 can be traced back to occupied N<sub>2</sub>-$`\pi ^{}`$ orbitals as shown in Fig. 6. These orbitals can be seen as an antisymmetric combination of the Fe–N bonds. Interestingly they are only dominant in the minority spin direction of the two neighboring Fe sites Fe3 and Fe7. The origin is that the interaction with the unoccupied $`d`$ states split the $`\pi ^{}`$ orbital in a bonding and an antibonding orbital. The bonding orbital, still having $`\pi ^{}`$ character but containing the Fe–N bonding contribution, becomes occupied and is located about 1 eV below the highest occupied molecular orbital (HOMO), while the antibonding orbital, having the Fe–N antibonding contribution, remains unoccupied. The former, bonding orbital is the relevant frontier orbital for the protonation.
The energy to add a hydrogen atom to dinitrogen in the bridged binding mode A1 and to obtain B1 is 41 kJ/mol (76 kJ/mol), substantially less than the 164 kJ/mol (199 kJ/mol) of the gas-phase reaction. B1 is energetically slightly more favorable by 8 kJ/mol than protonating the other nitrogen atom of N<sub>2</sub>.
The protonation of the reduced complex from an ammonium, which mimics the proton donor in our calculations, proceeds with a negligible barrier of only 4 kJ/mol and is exothermic by 63 kJ/mol. Note, however, that the calculated protonation energies taken individually are not as reliable as the reaction energy, as the former depend on the choice of NH$`{}_{}{}^{+}{}_{4}{}^{}`$ as the proton donor. Nevertheless, this calculation indicates that protonation of the reduced complex at dinitrogen proceeds without difficulty and much more readily than protonation at the central ligand.
Interestingly, the spin ordering of the cofactor, as obtained from DFT, changed from BS6 in A1 to BS7 in B1 during the reduction and protonation. (See Fig. 2 for a definition of the spin labels.) This corresponds to a spin flip of the iron atom Fe3, to which N<sub>2</sub> as well as the SH group are bound. In comparison, the FeMo cofactor in the resting state has the same spin arrangement BS7 as B1. After two protonations and reductions, BS7 of the resting state is transformed to a noncollinear spin distribution, which changes to BS7 during dinitrogen binding in the axial mode A0 and to BS6 during the conversion to the bridged binding mode A1. These transitions show the importance of allowing spin flips and noncollinear spins when simulating the reaction. Previous calculationsRod et al. (1999); Rod and Nørskov (2000); Rod et al. (2000); Hinnemann and Nørskov (2004) assumed a fixed spin ordering during the entire reaction, which may cause errors in the energy profile.
### III.3 Protonation of axial dinitrogen
If N<sub>2</sub> is axially bound as in A0 during the first protonation, the lowest-energy protonation site is the terminal nitrogen atom. The resulting structure is B2, shown in Fig. 4. The protonation does not induce major structural changes. The energy of B2 is 19 kJ/mol higher than the energy of B1, i.e., the bridged mode. Nevertheless, even though the energy for protonation of the reduced complex from an ammonium is, compared to the one of the bridged mode, 30 kJ/mol smaller, the reaction is still exothermic with 33 kJ/mol and proceeds with a negligible barrier of less than 10 kJ/mol.
Also in this case, the spin ordering changed during the reduction and protonation from BS7 in A0 to BS6 in B2.
### III.4 Other binding modes
Only 5 kJ/mol above the energy of B2, we find the complex B3 bridging two Fe atoms with a single nitrogen atom. This structural principle is found again later in the reaction cycle.Kästner and Blöchl (2005a) However, like the axial binding mode, it lies substantially, that is, by 24 kJ/mol, higher in energy than the bridged mode B1.
The protonated dinitrogen bound to molybdenum, i.e., B4, lies 29 kJ/mol above B1. Together with B7, it will be discussed in Sec. III.5, where we discuss the potential role of molybdenum.
As in B5, dinitrogen can also bridge the two Fe atoms with its axis perpendicular to the Fe–Fe direction, so that both nitrogen atoms are connected to both Fe atoms. A similar binding mode has recently been found experimentally for nitrogen bridging two zirconium centers.Pool et al. (2004) The energy of this structure, i.e., B5, lies 32 kJ/mol above that of B1.
In the relevant intermediate B6 of the mechanism proposed by Hinnemann and Nørskov,Hinnemann and Nørskov (2004) dinitrogen binds axially to one Fe atom like our structure B2. In contrast to B2, however, the sulfur bridge is still intact and the bond between this Fe atom and the central ligand is broken. This structure is 48 kJ/mol above B1 and it is 29 kJ/mol less stable than the axial binding mode B2 discussed previously.
### III.5 Molybdenum as coordination site
Numerous theoretical Grönberg et al. (1998); Szilagyi et al. (2001); Durrant (2001a, b, 2002a, 2002b) and experimental Barrière (2003); Yandulov and Schrock (2003) studies have been performed to investigate the Mo atom as a possible adsorption site. For this reason, we discuss nitrogen coordination to molybdenum here in some detail. As shown previously,Schimpl et al. (2003) the molybdenum atom is, according to our simulations, not a favorable adsorption site for nitrogen: nitrogen bound to molybdenum is higher in energy by 50 kJ/mol than the one bound to the iron atoms. Nevertheless we investigate the most favorable pathway for the first protonation via dinitrogen bound to the Mo site.
As shown previously, nitrogen adsorption at pentacoordinate Mo is endothermic by 30 kJ/mol.Schimpl et al. (2003) This indicates that, even if the coordination site is vacant, dinitrogen binds to Mo only for fleetingly short periods of time. If protonation proceeds sufficiently easy so that the proton is transferred during these short periods, dinitrogen may be stabilized bound to Mo. However, protonation leading to B4, shown in Fig. 4 is energetically unfavorable. B4 is 29 kJ/mol higher in energy than B1, with N<sub>2</sub>H bound to Fe atoms.
DurrantDurrant (2002a) proposed a transition of the protonated dinitrogen B4 into a bridging position between Mo and Fe as in B7 shown in Fig. 4. We find B7 to be 79 kJ/mol higher in energy than B4.
## IV Discussion
The following picture emerges from our calculations: dinitrogen exists in two competing docking modes at the cofactor, the axial mode A0 and the bridged mode A1.Schimpl et al. (2003) They are separated by a large barrier of 66 kJ/mol, which, however, is still sufficiently small to be overcome.Schimpl et al. (2003) Before the reduction and protonation take place, the axially bound dinitrogen is even slightly more stable by 6 kJ/mol than the bridged configuration.Schimpl et al. (2003) We assume that axial and bridged modes are in equilibrium until the proton is transferred. However, the electron transfer and the protonation reverse the energetic order between them so that the axially bound dinitrogen B2 ends up 19 kJ/mol higher in energy than the bridged dinitrogen B1.
In the bridged binding mode A1, N<sub>2</sub> is activated for accepting a proton by forming bonds to the Fe sites. As a result the $`\pi ^{}`$ orbitals are occupied, which implies that the triple bond is effectively broken. The occupied $`\pi ^{}`$ orbital exposes two nucleophilic lobes to which a proton can bind.
These occupied $`\pi ^{}`$ orbitals are binding combinations of the dinitrogen $`\pi ^{}`$ orbital with the $`d`$ orbitals of the Fe sites. They are only occupied in the minority spin direction of the Fe atoms binding to dinitrogen. This is because only the high-lying $`d`$ orbitals of the minority spin direction can lead to a stabilization of the $`\pi ^{}`$ orbital; hybridization with the low-lying Fe-$`d`$ states of the majority spin direction would shift the $`\pi ^{}`$ orbitals further up, while the corresponding bonding orbital is mostly of Fe character. Note here that both Fe sites binding to dinitrogen have the same spin orientation in A1.
In the axial binding mode A0, the corresponding $`\pi ^{}`$ orbital lies close to the Fermi level and its weight on dinitrogen is substantially smaller than in the corresponding orbital of the bridging mode. These differences between the bridged and axial binding modes are also reflected in the bond-length expansion and the reduction of the stretch frequency of the dinitrogen bond, and suggest a smaller activation of the axial binding mode.
Nevertheless, dinitrogen is readily protonated both in the axial and in the bridged binding mode after the transfer of one electron. Compared to the gas phase, the reaction energy for the first protonation is dramatically reduced from 164 kJ/mol (199 kJ/mol) for the gas phase to 41 kJ/mol (76 kJ/mol) at the FeMo cofactor in the bridged mode.
Interestingly, the most stable site for protonation is the central ligand. The resulting isomer is more stable by 37 kJ/mol than the one with protonated dinitrogen. However, this isomer is not considered relevant because protonation of the central ligand directly competes with protonation of dinitrogen. While protonation of the central ligand requires a large barrier of 79 kJ/mol to be overcome, protonation of dinitrogen proceeds nearly barrier-less. We attribute the large barrier to the breaking of the bond between the central ligand and one of the Fe atoms, which changes its tetrahedral coordination to a trigonal planar one.
Even if the central ligand is protonated, the second proton is added to dinitrogen and induces an internal proton transfer from the central ligand to dinitrogen.Kästner and Blöchl (2005a) Thus even if the first proton binds to the central ligand, this step neither poisons the catalyst nor does it lead to an entirely different branch of the reaction cycle. This side branch may potentially be relevant at small turnover frequencies.
The branch via protonation of the central ligand is reminiscent of the proposal by Huniar et al.,Huniar et al. (2004) who propose complete protonation of the central ligand and its cleavage as ammonia, before dinitrogen binds to the cofactor. In their study, the central cage is opened by a water molecule coordinating to one Fe atom. Additional calculations would be necessary to directly compare our results with their proposal. However, the latter seems to be in conflict with isotope exchange (ESEEM/ENDOR) experimentsLee et al. (2003) that exclude an exchange of a central nitrogen ligand.
A mechanism via dinitrogen bound to the Mo site is inconsistent with our calculations. Binding to the Mo atom is substantially more endothermic than binding to the Fe sites.
Dinitrogen bridging two Fe atoms is part of the proposal by Sellmann and co-workersThorneley et al. (1996); Sellmann et al. (1999, 2000); Kirchner et al. (2005). Our model differs from theirs, in that all Fe atoms remain tetrahedrally coordinated and in the high-spin state. We are not aware of any theoretical investigations of this system including the cofactor with its central ligand.
The presence of the central ligand is crucial for the reaction mechanism of nitrogen fixation at the FeMo cofactor, and may explain why the mechanism has remained elusive for a long time. The main feature of the central ligand is its ability to form a variable number of bonds to the six Fe atoms. The central ligand changes its coordination from sixfold to fivefold and fourfold. This allows other ligands such as nitrogen and sulfur to form and cleave bonds to the Fe atoms without deviations from the preferred tetrahedral coordination of the latter. Tetrahedral coordination seems to be a common structural principle to nearly all relevant intermediates of the reaction cycle. This is particularly apparent when dinitrogen binds:Schimpl et al. (2003) while the Fe atom to which dinitrogen binds in the axial binding mode A0 maintains its coordination shell by giving up its sulfur bridge, the bridged binding mode A1 would result in an unfavorable fivefold coordination of one Fe atom, if the latter would not give up its bond to the central ligand.
Let us mention here some variants of the reaction steps discussed: our calculations do not allow to distinguish between Fe3 and Fe7 as potential docking sites of dinitrogen. We find that the axial binding mode A0 is more stable, that is, by 30 kJ/mol, on Fe7 than on Fe3. The bridged mode accessed via the axial binding at Fe7 is more stable by 14 kJ/mol than the other variant. We expect that these differences are sensitive to the environment of the cofactor: due to the large motion of the sulfur atom, the opening of the SH bridge may be strongly affected by the shape and the specific interactions of the cavity. Thus, no conclusive answer regarding the initial binding site can be given at this point.
The cofactor has an approximate threefold symmetry, which is broken by the ligands and the protein environment. As long as the protein environment is not taken into account, as in the present study, the energetics will proceed similar for all three equivalent orientations. Nevertheless, the position of the proton channel in the protein indicates that nitrogen fixation proceeds near the iron atoms Fe3 and Fe7. Furthermore the cavity in this region provides sufficient space to accommodate nitrogen bound to an Fe atom.Schimpl et al. (2003)
## V Summary
We have studied the first protonation of dinitrogen at the cofactor on the basis of density-functional calculations. We made an effort to explore the phase space for the reaction without prejudice for one particular model of the mechanism. A large number of intermediates and the barriers between them have been explored and placed into perspective. While an unambiguous determination of this reaction step is not yet possible, three possible branches could be identified. One proceeds via dinitrogen in the bridged configuration, a second one proceeds via axially bound dinitrogen, and the third proceeds via protonation of the central ligand. The latter, however, is unlikely to play a role in the reaction cycle due to kinetic competition with direct protonation of the bridging dinitrogen. The activation of dinitrogen is discussed in detail.
## ACKNOWLEDGMENTS
We acknowledge support by the HLRN for granting access to their IBM pSeries 690 supercomputers. This work has benefited from the collaboration within the ESF program on “Electronic Structure Calculations for Elucidating the Complex Atomistic Behaviour of Solids and Surfaces.” |
warning/0507/astro-ph0507399.html | ar5iv | text | # Dark Matter Haloes and Rotation Curves via Brans-Dicke Theory
## Abstract
In the present work, the Brans-Dicke (BD) theory of gravity is taken as a possible theory of k-essence. Then starting with the (already known) Brans-Dicke-Schwarzschild solution which can represent the gravitationally bound static configurations of the BD scalar k-essence, issues like whether these configurations can reproduce the observed properties of galactic dark matter haloes have been addressed. It has been realized that indeed the BD scalar k-essence can cluster into dark matter halo-like objects with flattened rotation curves while exhibiting a dark energy-like negative pressure on larger scales.
I. Introduction
Currently, perhaps the most fashionable candidates for the unified model of dark matter and dark energy wmap with non-trivial dynamics are quintessence and k-essence. The main difference between the two models is that the quintessence models quintessence involve canonical kinetic terms and the sound speed of $`c_s^2=1`$ while the k-essence models kessence employ rather exotic scalar fields with non-canonical (non-linear) kinetic terms which typically lead to the negative pressure. And the most remarkable property of these k-essence models is that the typical k-essence field can overtake the matter energy density and induce cosmic acceleration only at the onset of the matter-dominated era and particularly at about the present epoch. These models are also expected to provide a successful explanation of the phenomena associated with the dark matter. In the present work, we take the Brans-Dicke (BD) theory of gravity bd as a possible k-essence theory since it involves probably the simplest form of such non-linear kinetic term for the (BD) scalar field. Besides, the BD scalar field (and the BD theory itself) is not of quantum origin. Rather it is classical in nature and hence can be expected to serve as a very relevant candidate to play some role in the late-time evolution of the universe such as the present epoch. Indeed, the BD theory is the most studied and hence the best-known of all the alternative theories of classical gravity to Einstein’s general relativity will . This theory can be thought of as a minimal extension of general relativity designed to properly accomodate both Mach’s principle will ; weinberg and Dirac’s large number hypothesis will ; weinberg . Namely, the theory employs the viewpoint in which the Newton’s constant $`G`$ is allowed to vary with space and time and can be written in terms of a scalar (“BD scalar”) field as $`G=1/\mathrm{\Phi }`$. As a scalar-tensor theory of gravity, it involves an adjustable but undetermined “BD-parameter” $`\omega `$ and as is well-known, the larger the value of $`\omega `$, the more dominant the tensor (curvature) degree and the smaller the value of $`\omega `$, the larger the effect of the BD scalar. And as long as we select sufficiently large value of $`\omega `$, the predictions of the theory agree perfectly with all the observations/experiments to date will . For this reason, the BD theory has remained a viable theory of classical gravity. However, no particularly overriding reason thus far has ever emerged to take it seriously over the general relativity. As shall be presented shortly in this work, here we emphasize that it is the existence of dark matter (and dark energy as well, see hongsu2 ) that puts the BD theory over the general relativity as a more relevant theory of classical gravity consistent with observations that have so far been unexplained within the context of general relativity.
II. Haloes of BD scalar k-essence
In general, the Brans-Dicke theory of gravity is described, in the absence of ordinary matter, by the action
$`S={\displaystyle d^4x\sqrt{g}\frac{1}{16\pi }\left[\mathrm{\Phi }R\omega \frac{_\alpha \mathrm{\Phi }^\alpha \mathrm{\Phi }}{\mathrm{\Phi }}\right]}`$ (1)
where $`\mathrm{\Phi }`$ is the BD scalar field representing the inverse of Newton’s constant which is allowed to vary with space and time and $`\omega `$ is the generic dimensionless parameter of the theory. Extremizing this action then with respect to the metric $`g_{\mu \nu }`$ and the BD scalar field $`\mathrm{\Phi }`$ yields the classical field equations given respectively by
$`G_{\mu \nu }`$ $`=`$ $`R_{\mu \nu }{\displaystyle \frac{1}{2}}g_{\mu \nu }R=8\pi T_{\mu \nu }^{BD},_\alpha ^\alpha \mathrm{\Phi }=0\mathrm{where}`$
$`T_{\mu \nu }^{BD}`$ $`=`$ $`{\displaystyle \frac{1}{8\pi }}\left[{\displaystyle \frac{\omega }{\mathrm{\Phi }^2}}(_\mu \mathrm{\Phi }_\nu \mathrm{\Phi }{\displaystyle \frac{1}{2}}g_{\mu \nu }_\alpha \mathrm{\Phi }^\alpha \mathrm{\Phi })+{\displaystyle \frac{1}{\mathrm{\Phi }}}(_\mu _\nu \mathrm{\Phi }g_{\mu \nu }_\alpha ^\alpha \mathrm{\Phi })\right].`$ (2)
Note that here in the present work, we are interested in the role played by the BD scalar field (i.e., a k-essence) as a dark matter particularly in forming galactic dark matter haloes inside of which the well-known rotation curves have been observed. Since the galactic dark matter haloes are roughly static and spherically-symmetric, we first should look for such dark matter halo-like solution of these BD field equations. Interestingly enough, the Brans-Dicke-Schwarzschild (BDS) spacetime solution to these vacuum BD field equations that happens to meet our above-mentioned needs has been found some time ago hongsu1 in rather a theoretical attempt to construct non-trivial black hole spacetime solutions in BD theory. To summarize, the BDS spacetime solution that can be obtained by setting $`a=e=0`$ in the Brans-Dicke-Kerr-Newman (BDKN) solution in eq.(11) of hongsu1 takes the form
$`ds^2`$ $`=`$ $`\mathrm{\Delta }^{2/(2\omega +3)}\mathrm{sin}^{4/(2\omega +3)}\theta \left[\left(1{\displaystyle \frac{2M}{r}}\right)dt^2+r^2\mathrm{sin}^2\theta d\varphi ^2\right]`$
$`+`$ $`\mathrm{\Delta }^{2/(2\omega +3)}\mathrm{sin}^{4/(2\omega +3)}\theta \left[\left(1{\displaystyle \frac{2M}{r}}\right)^1dr^2+r^2d\theta ^2\right],`$
$`\mathrm{\Phi }`$ $`(r,\theta )=\mathrm{\Delta }^{2/(2\omega +3)}\mathrm{sin}^{4/(2\omega +3)}\theta `$
where $`\mathrm{\Delta }=r(r2M)`$. A remarkable feature of this BDS solution is the fact that, unlike the Schwarzschild solution in general relativity, the spacetime it describes is static (i.e., non-rotating) but not spherically-symmetric. Of great interest in this earlier construction was the realization that non-trivial black hole solutions different from general relativistic solutions could occur in this BD theory for the generic BD-parameter values in the range $`5/2\omega <3/2`$ hongsu1 . In the present study, however, since we are interested in the galactic halo-like configuration, we do not want this BDS solution to become “black” and this amounts to considering the BDS solution having the value of $`\omega `$-parameter well outside this range. Besides, we are only interested in whether the self-gravitating k-essence, i.e., the BD scalar field can generally cluster into dark matter halo-like objects which would be the gravitationally bound static solution configurations of super-galactic scale (i.e., the large but finite-$`r`$ behavior). Therefore, the peculiar microscopic geometrical nature of this BDS solution such as the issue of regularity of the potential Killing horizon (i.e., the finiteness of the invariant curvature polynomials there) addressed in hongsu1 or that of seemingly failure of asymptotic flatness and internal infinity nature of the symmetry axis discussed in hongsu3 are all irrelevant for the present purposes.
Therefore first, it appears that the BD scalar k-essence can indeed cluster into halo-like configurations as it can be represented by the BDS solution. Our natural next mission is then to ask whether these configurations really can reproduce the properties of dark matter haloes, namely if our BD scalar k-essence model for dark matter can reproduce the flattening of the rotation velocity curves inside these halo configurations consistent with the observations. Thus we now attempt to obtain the rotation curves in our BD scalar k-essence halo. Since henceforth we need concrete “numbers”, we now restore both Newton’s constant $`G_0`$ and the speed of light $`c`$ in order to come from the geometrical unit ($`G_0=c=1`$) back to the CGS (or MKS) unit. Then the energy-momentum tensor of the BD scalar field given earlier in eq.(2) now takes the form
$`T_{\mu \nu }^{BD}`$ $`=`$ $`{\displaystyle \frac{c^4}{8\pi G_0}}\left[{\displaystyle \frac{\omega }{\mathrm{\Phi }^2}}(_\mu \mathrm{\Phi }_\nu \mathrm{\Phi }{\displaystyle \frac{1}{2}}g_{\mu \nu }_\alpha \mathrm{\Phi }^\alpha \mathrm{\Phi })+{\displaystyle \frac{1}{\mathrm{\Phi }}}(_\mu _\nu \mathrm{\Phi }g_{\mu \nu }_\alpha ^\alpha \mathrm{\Phi })\right]`$ (4)
and in the BDS solution in eq.(3) above, we should replace $`MG_0M/c^2\stackrel{~}{M}`$ where $`G_0`$ denotes the present value of the Newton’s constant. Apparently, (4) has the dimension of the energy-momentum density in the CGS unit, $`(erg/cm^3)`$. We now turn to the computation of energy density profile and (anisotropic) pressure components of the k-essence playing the role of the dark matter by treating the BD scalar field as a (dark matter) fluid. The BD scalar field fluid, however, would fail to be a “perfect” fluid as can readily be envisaged from the fact that the associated BDS solution configuration is not spherically-symmetric. Namely, its pressure cannot be “isotropic”, i.e., $`P_rP_\theta P_\varphi `$. Such fluid may be called imperfect fluid due to the anisotropic pressure components and as such its stress tensor can be written as
$`T_\nu ^{BD\mu }=\left(\begin{array}{cccc}c{}_{}{}^{2}\rho & 0& 0& 0\\ 0& P_r& T_\theta ^r& 0\\ 0& T_r^\theta & P_\theta & 0\\ 0& 0& 0& P_\varphi \end{array}\right).`$ (5)
And it is to be contrasted to its counterpart of the usual perfect fluid with isotropic pressure given by the well-known form, $`T_\nu ^\mu =P\delta _\nu ^\mu +(c^2\rho +P)U^\mu U_\nu =diag(c^2\rho ,P,P,P)`$ where $`U^\alpha =dX^\alpha /d\tau `$ (with $`\tau `$ being the proper time) denotes the 4-velocity of the fluid element normalized such that $`U^\alpha U_\alpha =1`$. Note that in addition to the diagonal entries representing the (anisotropic) pressure components $`T_i^i=P_i`$ (where no sum over $`i`$), there are off-diagonal entries $`T_\theta ^r`$, $`T_r^\theta `$ representing a shear stress which also results from the failure of spherical symmetry. Thus by substituting the BDS solution given in eq.(3) into the BD energy-momentumm tensor in eq.(4) and then setting (4) equal to (5), we can eventually read off the energy density and the pressure components of the BD scalar field imperfect fluid to be
$`\rho `$ $`=`$ $`{\displaystyle \frac{c^2}{8\pi G_0}}{\displaystyle \frac{4}{(2\omega +3)^2}}{\displaystyle \frac{1}{r^2\mathrm{\Delta }}}\mathrm{\Delta }^{2/(2\omega +3)}\mathrm{sin}^{4/(2\omega +3)}\theta `$
$`\times `$ $`\left[2(\omega +1)\left\{(r\stackrel{~}{M})^2+\mathrm{\Delta }\mathrm{cot}^2\theta \right\}(2\omega +3)\stackrel{~}{M}(r\stackrel{~}{M})\right],`$
$`P_r`$ $`=`$ $`{\displaystyle \frac{c^4}{8\pi G_0}}{\displaystyle \frac{4}{(2\omega +3)^2}}{\displaystyle \frac{1}{r^2\mathrm{\Delta }}}\mathrm{\Delta }^{2/(2\omega +3)}\mathrm{sin}^{4/(2\omega +3)}\theta `$
$`\times `$ $`\left[2(\omega +2)(r\stackrel{~}{M})^2+2(\omega 1)\mathrm{\Delta }\mathrm{cot}^2\theta (2\omega +3)\left\{\mathrm{\Delta }+\stackrel{~}{M}(r\stackrel{~}{M})\right\}\right],`$
$`P_\theta `$ $`=`$ $`{\displaystyle \frac{c^4}{8\pi G_0}}{\displaystyle \frac{4}{(2\omega +3)^2}}{\displaystyle \frac{1}{r^2\mathrm{\Delta }}}\mathrm{\Delta }^{2/(2\omega +3)}\mathrm{sin}^{4/(2\omega +3)}\theta [2(\omega 1)\{\mathrm{\Delta }\mathrm{cot}^2\theta (r\stackrel{~}{M})^2\}`$ (6)
$`+`$ $`(2\omega +3)(r\stackrel{~}{M})(r2\stackrel{~}{M})+\{4\mathrm{cos}^2\theta (2\omega +3)\}{\displaystyle \frac{\mathrm{\Delta }}{\mathrm{sin}^2\theta }}],`$
$`P_\varphi `$ $`=`$ $`{\displaystyle \frac{c^4}{8\pi G_0}}{\displaystyle \frac{4}{(2\omega +3)^2}}{\displaystyle \frac{1}{r^2\mathrm{\Delta }}}\mathrm{\Delta }^{2/(2\omega +3)}\mathrm{sin}^{4/(2\omega +3)}\theta `$
$`\times `$ $`\left[2(\omega +1)(r\stackrel{~}{M})^2\mathrm{\Delta }\mathrm{cot}^2\theta (2\omega +3)(r\stackrel{~}{M})(r2\stackrel{~}{M})\right]`$
$`T_\theta ^r`$ $`=`$ $`\mathrm{\Delta }T_r^\theta ={\displaystyle \frac{c^4}{8\pi G_0}}{\displaystyle \frac{4}{(2\omega +3)^2}}{\displaystyle \frac{1}{r^2}}\mathrm{cot}\theta \mathrm{\Delta }^{2/(2\omega +3)}\mathrm{sin}^{4/(2\omega +3)}\theta `$
$`\times `$ $`\left[4\omega (r\stackrel{~}{M})(2\omega +3)(r2\stackrel{~}{M})\right].`$
Note that the off-diagonal components $`T_\theta ^r`$, $`T_r^\theta `$ are odd functions of $`\theta `$ while the diagonal components $`(\rho ,P_r,P_\theta ,P_\varphi )`$ are even functions of the polar angle under $`\theta (\pi \theta )`$. As a result, the off-diagonal components vanish (i.e., no shear stress survives) if we average over this polar angle to get a net stress. Thus, first the equation of state of this BD scalar k-essence fluid forming a galactic halo is given by
$`w={\displaystyle \frac{P}{c^2\rho }}={\displaystyle \frac{\left[2(\omega +2)(r\stackrel{~}{M})^2+2(\omega 1)\mathrm{\Delta }\mathrm{cot}^2\theta (2\omega +3)\left\{\mathrm{\Delta }+\stackrel{~}{M}(r\stackrel{~}{M})\right\}\right]}{2(\omega +1)\left\{(r\stackrel{~}{M})^2+\mathrm{\Delta }\mathrm{cot}^2\theta \right\}(2\omega +3)\stackrel{~}{M}(r\stackrel{~}{M})}}`$ (7)
where $`P=P_r`$. Namely, $`P=w(r,\theta )c^2\rho `$ with $`w(r,\theta )O(1)`$ meaning that this k-essence fluid is essentially a barotropic fluid but with “position-dependent” coefficient $`w(r,\theta )`$. Note that although the BD scalar k-essence is a candidate for dark matter, it is not quite a dust. In principle, the speed of sound in this BD scalar field fluid can also be evaluated via $`c_s^2=dP/d\rho `$ but we shall not discuss it in any more detail in this work. We are now ready to compute the behavior of rotation curves in the outer region (i.e., at large but finite-$`r`$, say, $`r>>G_0M/c^2`$ ) of our BD scalar k-essence halo. To be more precise, for a galaxy of typical (total) mass $`M10^{11}M_{}`$, the outer region of its dark matter halo, say, $`r10(kpc)10^{23}(cm)`$ is much greater than $`G_0M/c^210^{16}(cm)`$ by a factor of $`10^7`$. Thus to this end, we first approximate the expressions for the energy density and the (radial) pressure of the k-essence given in eq.(6) for large-$`r`$. They are
$`\rho `$ $``$ $`{\displaystyle \frac{c^2}{8\pi G_0}}{\displaystyle \frac{8(\omega +1)}{(2\omega +3)^2}}{\displaystyle \frac{1}{r^2\mathrm{sin}^2\theta }}\mathrm{\Delta }^{2/(2\omega +3)}\mathrm{sin}^{4/(2\omega +3)}\theta ,`$ (8)
$`P`$ $``$ $`{\displaystyle \frac{c^4}{2\pi G_0}}{\displaystyle \frac{1}{(2\omega +3)^2}}{\displaystyle \frac{1}{r^2}}\left[2(\omega 1)\mathrm{cot}^2\theta +1\right]\mathrm{\Delta }^{2/(2\omega +3)}\mathrm{sin}^{4/(2\omega +3)}\theta .`$
Note that in the above approximations and in the discussions below, it was and it shall be assumed that the metric function $`\mathrm{\Delta }=r(r2\stackrel{~}{M})r^2`$ for large-$`r`$. It is interesting to note that as a “k-essence” constituting a dark matter halo, the energy density $`\rho `$ of the BD scalar field is almost certainly positive everywhere (i.e., for both small and large-$`r`$). In the mean time, its (radial) pressure $`P`$ particularly at larger scale (i.e., for large-$`r`$) turns out to be negative although its sign appears unclear at small scale (i.e., for small-$`r`$).
Finally, we are ready to determine the rotation curve inside our BD scalar k-essence halo. First in the most naive sense, the apparent rotation velocity of an object at radius $`r`$ from the galactic center is given by the Kepler’s third law, $`v^2=G_0M(r)/r`$. Thus for our case, using the BD scalar k-essence energy density profile given earlier, we have $`M(r)=_0^{2\pi }𝑑\varphi _ϵ^\pi 𝑑\theta _0^r𝑑r\sqrt{g_{rr}g_{\theta \theta }g_{\varphi \varphi }}\rho (r,\theta )=\left(2c^2/G_0\right)\left[(\omega +1)/(2\omega +1)(2\omega +3)\right]f(\omega )rr^{2/(2\omega +3)}`$ and hence
$`v^2(r)={\displaystyle \frac{G_0M(r)}{r}}=c^2{\displaystyle \frac{2(\omega +1)}{(2\omega +1)(2\omega +3)}}f(\omega )r^{\frac{2}{(2\omega +3)}}`$ (9)
where $`f(\omega )_ϵ^\pi 𝑑\theta \mathrm{sin}^{[1+2/(2\omega +3)]}\theta =2_0^{1\delta }𝑑x[1x^2]^{(2\omega +4)/(2\omega +3)}`$ with $`ϵ,\delta <<1`$. (Note here that the integration over the polar angle $`\theta `$ starts not from $`0`$ but from $`ϵ<<1`$ as the symmetry axis $`\theta =0`$ of the BDS solution in eq.(3) possesses an internal infinity nature, namely, the symmetry axis is infinite proper distance away as discussed carefully in hongsu3 .) It has been known for some time that in order for the BD theory to remain a viable theory of classical gravity passing all the observational/experimental tests to date, the BD $`\omega `$-parameter has to have a large value, say, $`|\omega |500`$ will . In our previous study hongsu1 , in the mean time, it has been realized that the static solution to the vacuum BD field equations given in eq.(3) above can turn into a black hole spacetime for $`5/2\omega <3/2`$. Thus now for $`|\omega |500`$, the same static solution eq.(3) we are considering represents just a halo-like configuration with regular geometry everywhere (i.e., having no horizon) which is static but not exactly spherically-symmetric (note that the galactic haloes are also believed to be nearly spherically-symmetric but not exactly). Thus if we substitute a large-$`\omega `$ value, say, $`\omega 10^6`$ into eq.(9) above, evidently $`M(r)r`$ and hence we get
$`v(r)100(km/s)\times r^{(1/10^6)}`$ (10)
since for $`\omega 10^6`$, $`f(\omega )O(1)`$. Namely for this large-$`\omega `$ value, the rotation curve gets flattened out as $`r^{(10^6)}constant`$ and its magnitude becomes several hundred $`(km/s)`$. Indeed, this is in impressive agreement with the data for rotation curves observed in spiral/elliptic galaxies with $`M/L(1020)M_{}/L_{}`$ and in low-surface-brightness (LSB)/dwarf galaxies with $`M/L(200600)M_{}/L_{}`$ (where $`M/L`$ denotes the so-called “mass-to-light” ratio given in the unit of solar mass-to-luminosity ratio exhibiting the large excess of dark matter over the luminous matter) rc . Rotation curves are observed usually via the measurements of the Doppler shift of the $`21cm`$ emission line from neutral hydrogen (HI) for distant galaxies and of the light emitted by stars for nearby galaxies halo1 ; halo2 . It is also interesting to note that this behavior of the rotation curve in our BD theory k-essence dark matter halo model is independent of the mass of the host galaxy as it should be. Namely, this behavior of the rotation curve comes exclusively from the nature of the dark matter, i.e., the BD scalar field k-essence. We also point out that even if we employ more careful expression for the rotation velocity curve involving the Doppler shift of light emitted by the orbiting objects (assuming that the k-essence halo is almost spherically-symmetric), namely $`v^2(r)=G_0M(r)/r+4\pi r^2G_0P/c^2`$ khalo (with $`P`$ being the radial pressure given in eq.(8) above), the conclusions above remain the same.
Next, the equation of state in eq.(7) of this BD scalar k-essence becomes, in the outer region of the galactic dark matter halo (i.e., at large-$`r`$),
$`w{\displaystyle \frac{\left[2(\omega 1)\mathrm{cos}^2\theta +\mathrm{sin}^2\theta \right]}{2(\omega +1)}}`$ (11)
which is obviously negative due to the negative pressure (and still positive energy density) in this outer region. Moreover, for the large-$`\omega `$ value, i.e., $`\omega 10^6`$ for which the rotation curve gets flattened out that we just have realized, this equation of state at large-$`r`$ further approaches $`w\mathrm{cos}^2\theta O(1)`$. (Incidentally, it is interesting to note that in the vicinity of the equatorial plane $`\theta =\pi /2`$, $`w=0`$, namely, the BD scalar k-essence behaves like nearly a dust.) This observation is particularly interesting as it appears to indicate that the BD scalar k-essence we are considering possesses dark energy-like negative pressure on larger scales. And this observation is indeed consistent with our previous study hongsu2 that on the cosmological scale, the BD scalar field does exhibit the nature of dark energy possessing the negative pressure.
III. Concluding remarks
In the present work, starting with the (already known) BDS solution which can represent the gravitationally bound static configurations of the BD scalar k-essence, issues like whether these configurations can reproduce the observed properties of galactic dark matter haloes have been investigated. It has been realized that indeed the BD scalar k-essence can cluster into dark matter halo-like objects with flattened rotation curves while exhibiting a dark energy-like negative pressure on larger scales. Thus to conclude, from this success of “BD scalar field as a k-essence” to account for the asymptotic flattening of galaxy rotation curves while forming galactic dark matter haloes plus the original spirit of the BD theory in which the BD scalar field is prescribed not to have direct interaction with ordinary matter fields (in order not to interfere with the great success of equivalence principle), we suggest that the Brans-Dicke theory of gravity is a very promising theory of dark matter. And this implies, among others, that dark matter (and dark energy as well, see hongsu2 ) might not be some kind of unknown exotic “matter”, but the effect resulting from the space-time varying nature of the Newton’s constant represented by a (k-essence) scalar field. Even further, this successful account of the phenomena associated with the dark matter of the present universe via the BD gravity theory might be an indication that the truly relevant theory of classical gravity at the present epoch is not general relativity but its simplest extension, the Brans-Dicke theory with its generic parameter value $`\omega 10^6`$ fixed by the dark matter observation !
This work was financially supported by the BK21 Project of the Korean Government.
References |
warning/0507/quant-ph0507280.html | ar5iv | text | # A General Setting for Geometric Phase of Mixed States Under an Arbitrary Nonunitary Evolution
## Abstract
The problem of geometric phase for an open quantum system is reinvestigated in a unifying approach. Two of existing methods to define geometric phase, one by Uhlmann’s approach and the other by kinematic approach, which have been considered to be distinct, are shown to be related in this framework. The method is based upon purification of a density matrix by its uniform decomposition and a generalization of the parallel transport condition obtained from this decomposition. It is shown that the generalized parallel transport condition can be satisfied when Uhlmann’s condition holds. However, it does not mean that all solutions of the generalized parallel transport condition are compatible with those of Uhlmann’s one. It is also shown how to recover the earlier known definitions of geometric phase as well as how to generalize them when degeneracy exists and varies in time.
The concept of geometric phase was originally introduced by Pancharatnam in the classical context of comparing two polarized light beams through their interference panch . Later, Berry pointed out to its importance even in quantum systems undergoing a cyclic adiabatic evolution berry . After that, this important notion has been subject of interest in many different aspects, which has led to many different generalizations and applications anandan ; shapere . Of course in general cases to retain purely geometrical nature of the phase one has to put some constrains, namely parallel transport (PT) conditions. In this manner, geometric phase is a feature which only depends on geometry of the path traversed by the system in its motion during evolution.
It is also worth noting that an important source of the renewed interest in geometric phases is their relevance to geometric quantum computation and holonomic quantum computation zanardi . Indeed, it is known that quantum logic gates can be implemented only by using the concept of geometric phases. It is believed that the purely geometric nature of this phase makes such computations intrinsically fault tolerant and robust against noise rob .
A pure state is merely an idealization and in real experiments a description of the system in terms of mixed states is usually required. This point accounts for attempts towards extending the concept of geometric phase to mixed states. In fact, Uhlmann was the first in tackling the problem through the mathematical approach of purification of mixed states uhl1 . This method is rather general in that it is independent of the type of evolution of the system. Next, Sjöqvist et al. put forward a quantum interferometric based definition for geometric phase of nondegenerate density matrices undergoing a unitary evolution sjoq1 . Later, Singh et al. proposed a kinematic description and extended the results to the case of degenerate mixed states singh . It must be mentioned that there also exists another, differential geometric, approach to define geometric phase for mixed states undergoing a unitary evolution chaturvedi . In this approach, mixed state geometric phase appears as an immediate and direct generalization of the pure state case.
Indeed, in the case of environmental effects such as decoherence, one has to consider nonunitary evolutions of mixed states. Some generalizations in this direction have been addressed in Refs. uhl1 ; ericsson2 ; gam ; pix ; pati ; carollo1 ; carollo2 ; nonunitary ; lidar . The proposition in Ref. ericsson2 for completely positive maps in spite of being operationally well-defined depends on the specific Kraus representation for the map. In Refs. carollo1 ; carollo2 , the problem of geometric phase of an initially pure open quantum system, based on the standard definition of pure state geometric phase, has been addressed through the quantum jump method. A more recent effort is based on a kinematic approach, with no a priori assumption about dynamics of the system nonunitary . However, most of these different definitions do not agree with each other. In fact, Uhlmann’s method even in the case of unitary evolution does not agree with the interferometric definition ericsson ; slater . The source of such disagreement is known to be the use of different types of PT conditions. Hence, it has been argued ericsson ; slater ; levay that these approaches are not generally equivalent and one cannot obtain one from the other. Therefore, it could be desirable to find a more unified approach which can bring together the previous general ideas. Recently, in the unitary evolution case it has been argued that using (nonorthogonal) decompositions different from spectral decomposition can make it possible to unify the kinematic and Uhlmann’s approaches gen . In this framework, a suitable notion of PT condition of the mixed state is based on the PT condition of the vectors constituting this decomposition.
In this paper, we shall use a rather similar mechanism plus uniform decomposition of density matrices, and propose a generalized kinematic approach for geometric phase of mixed states under an arbitrary nonunitary evolution. This approach vividly shows how it is possible to merge Uhlmann’s approach and kinematic approach. It is also shown how to recover the earlier definitions of geometric phase from this more general approach. In addition, it is shown that the approach can be easily modified to include the more general case of degenerate mixed states. This investigation may as well be useful in the study of robustness of geometric phases against decoherence tidstrom .
Let us suppose that the density matrix of our system of interest (with the Hilbert space $`_s`$) is $`\varrho (t)=_{k=1}^Np_k(t)|w_k(t)w_k(t)|`$, in which $`p_k(t)`$’s ($`|w_k(t)`$’s) are considered to be its eigenvalues (normalized eigenvectors). In a general evolution both $`p_k`$ and $`|w_k`$ are subject to change in time. For simplicity of our discussion, we in the sequel assume that the rank of this matrix is constant at all instants, and even more the matrix is nondegenerate. In the case of unitary evolution, we have $`p_k(t)=p_k(0)`$ and $`|w_k(t)=U(t)|w_k(0)`$, where $`U(t)`$ is the unitary evolution operator. However, when evolution is nonunitary, the eigenvalues, $`p_k`$, can also vary in time. Thus, generally $`U(t)=_k|w_k(t)w_k(0)|`$ does not encompass the whole dynamical information. In fact, in such cases, to obtain $`\varrho (t)`$ one often has to resort to some approximative methods in the theory of open quantum systems open , such as the Lindblad equation lindblad .
Since in our construction we use Uhlmann’s PT condition uhl1 we need to recall it briefly. Uhlmann’s approach is based upon the standard purification $`\text{w}(t)`$, where $`\varrho (t)=\text{w}(t)\text{w}^{}(t)`$, for density matrices. In other words, w can be considered as a purification of $`\varrho `$ in the larger Hilbert space of Hilbert-Schmidt operators with scalar product $`\text{w}(t),\text{w}(t^{})=\text{tr}(\text{w}^{}(t)\text{w}(t^{}))`$ such that $`\text{w}\text{w}^{}=\varrho `$. It is clear that $`\text{w}(t)=\sqrt{\varrho (t)}V(t)`$ is an acceptable purification of $`\varrho `$ for any unitary $`V(t)`$. For a special purification where each $`|\text{w}(t),\text{w}(t^{})|`$ is constrained to its maximum value Uhlmann has defined the geometric phase associated to the evolution from $`\varrho (0)`$ to $`\varrho (\tau )`$ as $`\gamma _g(\tau )=\text{arg}(\text{w}(0),\text{w}(\tau )),`$ where the PT condition $`\text{w}^{}(t)\dot{\text{w}}(t)=\dot{\text{w}}^{}(t)\text{w}(t)`$ has to be satisfied.
Let us also briefly review the construction of the geometric phase in Ref. nonunitary . Consider a purification for the density matrix $`\varrho (t)`$ as
$`|\mathrm{\Psi }(t)_{sa}={\displaystyle \underset{k}{}}\sqrt{p_k(t)}|w_k(t)_s|a_k_a,t[0,\tau ].`$ (1)
Now after imposing the PT condition; $`w_k(t)|\frac{\text{d}}{\text{d}t}|w_k(t)=0`$, the geometric phase, defined a la Pancharatnam panch ; $`\gamma (\tau )=\text{arg}(\mathrm{\Psi }(0)|\mathrm{\Psi }(\tau ))`$, reads as $`\gamma _g(\tau )=\text{arg}(_{k=1}^N\sqrt{p_k(0)p_k(\tau )}w_k(0)|w_k(\tau )e^{_0^\tau w_k(t)|\dot{w}_k(t)\text{d}t})`$. Indeed, by using the PT condition one fixes the general form of the unitary operators which like $`U(t)`$ can run system’s dynamics. As is clear, in this method purification of mixed state of the system is done based on its spectral decomposition and the PT condition is considered to be the PT condition of all the vectors constituting this (spectral) decomposition. We know that a purification as in Eq. (1), is only one of the possible purifications that can give rise to the correct mixed state of the system. So, one has the freedom to choose other decompositions and study the problem of geometric phase with respect to them. In the sequel, we follow such a strategy and look for a specific purification in which all normalized terms can be treated in a naturally uniform manner, unlike Eq. (1) where the contribution of the $`k`$-th normalized term is the time dependent variable $`\sqrt{p_k(t)}`$. In other words, instead of starting from the spectral decomposition of a density matrix which is the usual starting point of purification based approaches, we start with another decomposition which can result to the mentioned uniformity. In order to do so, we need the next two important theorems on different decompositions of a density matrix $`\varrho `$.
Theorem 1 hugh : Let $`\varrho `$ has the spectral ensemble $`\{p_k,|w_k\}`$. Then $`\{q_l,|x_l\}`$ is another ensemble for it iff there exists a unitary matrix $`𝒰=(𝒰_{kl})`$ such that
$`\sqrt{q_l}|x_l={\displaystyle \underset{k}{}}\sqrt{p_k}𝒰_{lk}|w_k.`$ (2)
Theorem 2 prob : Let $`\{q_l\}`$ be a probability distribution. Then there exist normalized quantum states $`\{|x_l\}`$ such that $`\varrho =_lq_l|x_lx_l|`$, iff $`\stackrel{}{q}`$ is majorized by $`\stackrel{}{p}`$.
An immediate corollary of Theorem 2 is the existence of a uniform ensemble for any density matrix. Therefore, there exist normalized pure states $`|x_1,\mathrm{},|x_𝒩`$ such that $`\varrho `$ is an equal mixture of these states with probability $`1/𝒩`$ ($`𝒩N`$), i.e. $`\varrho =\frac{1}{𝒩}_{l=1}^𝒩|x_lx_l|`$. For the rest of discussion we assume that $`𝒩=N`$. Now, let us see how this uniform decomposition is related to the spectral decomposition. By using Theorem 1, we have $`\frac{1}{\sqrt{N}}|x_k=_{l=1}^N\sqrt{p_l}𝒰_{kl}|w_l`$. It is easy to see that if one chooses an $`N\times N`$ Fourier matrix (corresponding to discrete Fourier transformations fourier ) $`𝒰_{kl}=\frac{1}{\sqrt{N}}e^{2\pi i\frac{kl}{N}}`$ ($`k,l=0,\mathrm{},N1`$), and momentarily run all indices from 0 to $`N1`$ (rather than 1 to $`N`$) this equation is satisfied. Then, by using a Fourier matrix one can find a uniform ensemble for any density matrix. If we define $`C(t)=_k\sqrt{p_k(t)}|w_k(0)w_k(0)|`$ and use the definition of $`U(t)`$, we can rewrite $`|x_k(t)`$ in the following matrix form
$`|x_k(t)=\sqrt{N}U(t)C(t)𝒰|w_k(0).`$ (3)
Now we show that the above mentioned uniform decomposition is useful in our discussion of geometric phase. Consider the following pure state of the combined system $`sa`$
$`|\mathrm{\Phi }(t)_{sa}={\displaystyle \frac{1}{\sqrt{N}}}{\displaystyle \underset{k}{}}|x_k(t)_sV(t)|a_k_a,`$ (4)
where $`V(t)`$ is the unitary evolution of the $`|a_k`$’s. This state is a legitimate purification of the density matrix $`\varrho (t)`$ of the system; $`\varrho (t)=\text{tr}_a(|\mathrm{\Phi }(t)_{sa}\mathrm{\Phi }(t)|)`$. If $`V(t)=I`$, since $`x_k|x_k=1`$ and all $`|x_k(t)`$ vectors enter with equal and constant probability of $`\frac{1}{N}`$ in the decomposition of the density matrix, it seems natural to consider our (generalized) PT conditions in the form of
$`x_k(t)|{\displaystyle \frac{\text{d}}{\text{d}t}}|x_k(t)=0,k=1,\mathrm{},N,`$ (5)
that is, a density matrix undergoes a PT condition when all of the vectors in its uniform decomposition do so. Here a point is in order. It must be mentioned that, except the pure state case, this PT condition is generally different from the one considered in earlier literature sjoq1 ; nonunitary <sup>1</sup><sup>1</sup>1 Tong et al.’s PT condition nonunitary imposes a constraint on the form of parallel transported evolution operator, $`U^{}(t)`$, as $`w_k(0)|U^{}\dot{U}^{}|w_k(0)=0,k=1,\mathrm{},N,`$ whereas Eq. (5) results into $`w_k(0)|𝒰^{}CU^{}\dot{U}^{}C𝒰|w_k(0)=0,k=1,\mathrm{},N.`$.
In general, in the purification (4) ancillary vectors could also vary in time, and we have to find a natural picture for geometric phase in this case. Let us first remind a simple and useful property of Schmidt decomposition of bipartite pure states gen . If $`|\mathrm{\Phi }_{ab}=_kc_k|a_k_a|b_k_b`$, then $`(UV)|\mathrm{\Phi }_{ab}=(UC𝒱^TI)_k|a_k_a|b_k_b`$, where $`C`$ is a diagonal matrix in the $`\{|a_k\}`$ basis defined as $`C=_kc_k|a_ka_k|`$ and $`𝒱=_{kk^{}}b_k|V|b_k^{}|a_ka_k^{}|`$. Here, for notational purposes, we omit $`T`$ sign of $`𝒱^T`$. Now, noting this property and assuming that the basis vectors of the ancillary Hilbert space are $`\{|w_k(0)\}`$, one can rewrite Eq. (4) as
$`|\mathrm{\Phi }(t)_{sa}={\displaystyle \underset{k}{}}|\stackrel{~}{x}_k(t)_s|w_k(0)_a,`$ (6)
where $`|\stackrel{~}{x}_k(t)=U(t)C(t)𝒰𝒱(t)|w_k(0)`$. This purification now results into the nonorthogonal decomposition $`\varrho (t)=_k|\stackrel{~}{x}_k(t)\stackrel{~}{x}_k(t)|`$ for the density matrix. Unlike the $`\{|x_k(t)\}`$ decomposition, now for a general $`𝒱`$, $`\stackrel{~}{x}_k(t)|\stackrel{~}{x}_k(t)`$ is not time independent and, as well, is not equal for all $`k`$’s. However, if we consider the normalized vectors $`|\widehat{\stackrel{~}{x}}(t)=\frac{|\stackrel{~}{x}_k(t)}{\stackrel{~}{x}_k(t)}`$ it still looks natural to consider our generalized PT condition to be in the following form
$`\widehat{\stackrel{~}{x}}_k(t)|{\displaystyle \frac{\text{d}}{\text{d}t}}|\widehat{\stackrel{~}{x}}_k(t)=0.`$ (7)
In terms of $`|\stackrel{~}{x}_k(t)`$ vectors this is equal to $`\stackrel{~}{x}_k(t)|\frac{\text{d}}{\text{d}t}|\stackrel{~}{x}_k(t)=\frac{1}{2}\frac{\text{d}}{\text{d}t}(\stackrel{~}{x}_k(t)|\stackrel{~}{x}_k(t))`$, or equivalently in more detail it is
$`\begin{array}{cc}\hfill & w_k(0)|𝒱^{}𝒰^{}CU^{}\dot{U}C𝒰𝒱+𝒱^{}𝒰^{}C\dot{C}𝒰𝒱+𝒱^{}𝒰^{}C^2𝒰\hfill \\ & \times \dot{𝒱}|w_k(0)={\displaystyle \frac{1}{2}}{\displaystyle \frac{\text{d}}{\text{d}t}}(w_k(0)|𝒱^{}𝒰^{}C^2𝒰𝒱|w_k(0)).\hfill \end{array}`$ (8)
Now let us see what is the form of Uhlmann’s PT condition. We note that $`\text{w}(t)`$ operator reads as $`\text{w}(t)=U(t)C(t)𝒰𝒱(t)`$. Hence, the explicit form of Uhlmann’s PT condition is
$`\begin{array}{cc}& 𝒱^{}𝒰^{}CU^{}\dot{U}C𝒰𝒱+𝒱^{}𝒰^{}C\dot{C}𝒰𝒱+𝒱^{}𝒰^{}C^2𝒰\dot{𝒱}\hfill \\ & =𝒱^{}𝒰^{}C\dot{U}^{}UC𝒰𝒱+𝒱^{}𝒰^{}\dot{C}C𝒰𝒱+\dot{𝒱}^{}𝒰^{}C^2𝒰𝒱.\hfill \end{array}`$ (9)
As is seen lhs of this equation is exactly the expression within bra-ket of the PT condition (8). If sandwiched between $`w_k(0)|`$ and $`|w_k(0)`$, Eq. (LABEL:explicitul) gives rise to
$`\begin{array}{cc}\hfill \text{lhs}\text{ of (}\text{8}\text{)}& ={\displaystyle \frac{1}{2}}w_k(0)|\text{lhs}+\text{rhs}\text{ of (}\text{LABEL:explicitul}\text{)}|w_k(0)\hfill \\ & ={\displaystyle \frac{1}{2}}{\displaystyle \frac{\text{d}}{\text{d}t}}(w_k(0)|𝒱^{}𝒰^{}C^2𝒰𝒱|w_k(0)).\hfill \end{array}`$ (10)
This is what we wanted to show; by using Uhlmann’s PT condition the generalized PT conditions (8) are also satisfied. However, it must be noted that generally number of equations of the two PT conditions are not equal. In other words, Eq. (LABEL:explicitul) is a matrix equation which constitutes $`N^2`$ different equations (for $`𝒱`$) though Eq. (7) is just a set of $`N`$ equations. This simply means that there might be solutions of Eq. (8) that are not solutions of Eq. (LABEL:explicitul). If it is assumed that $`𝒱(t)=e^{i\stackrel{~}{H}(t)}`$, then solution of Eq. (LABEL:explicitul) is as follows uhl1
$`\begin{array}{cc}\hfill i\stackrel{~}{H}(t)=& 2{\displaystyle \underset{kk^{}}{}}𝒰^{}|w_k^{}(0)w_k(0)|𝒰\hfill \\ & \times {\displaystyle _0^t}\text{d}t^{}w_k^{}(t^{})|\dot{w}_k(t^{}){\displaystyle \frac{\sqrt{p_k^{}(t^{})p_k(t^{})}}{p_k^{}(t^{})+p_k(t^{})}}.\hfill \end{array}`$ (11)
Now it is easy to show that Eq. (8) can have solutions other than (11). For example, if we suppose that $`[𝒰𝒱,C]=0`$, and $`𝒰𝒱=_ke^{il_k(t)}|w_k(0)w_k(0)|`$, then Eq. (8) gives
$`l_k(t)=i{\displaystyle _0^t}\text{d}t^{}w_k(t^{})|\dot{w}_k(t^{}),`$ (12)
which does not generate a $`𝒱(t)`$ compatible with (11). This comes from the fact that to satisfy Eq. (8) we only need to have the diagonal terms of Uhlmann’s PT condition, whereas off-diagonal terms of this equation may put extra constraints that are redundant for validity of Eq. (8).
Now geometric phase can be simply defined a la Pancharatnam as
$`\gamma _g(t)`$ $`=\text{arg}(\mathrm{\Phi }(0)|\mathrm{\Phi }(t))=\text{arg}({\displaystyle \underset{k}{}}\nu _k(t)e^{i\gamma _k(t)}),`$ (13)
where $`\stackrel{~}{x}_k(0)|\stackrel{~}{x}_k(t)=\nu _k(t)e^{i\gamma _k(t)}`$, i.e. $`\nu _k(t)`$ ($`\gamma _k(t)`$) is the visibility (geometric phase) of the $`k`$-th component of $`|\mathrm{\Phi }(t)`$. The explicit form is obtained by insertion of the definition of $`|\stackrel{~}{x}_k(t)`$ in this equation, which gives
$`\begin{array}{cc}\hfill \gamma _g(t)=& \text{arg}({\displaystyle \underset{kk^{}}{}}\sqrt{p_k^{}(0)p_k(t)}w_k^{}(0)|w_k(t)\hfill \\ & \times w_k(0)|𝒰𝒱(t)𝒰^{}|w_k^{}(0)).\hfill \end{array}`$ (14)
This equation shows that geometric phase, as described here to be combined with Uhlmann’s definition, generally retains a memory of the evolution of both system and the ancilla, that is, it is a general property of the whole system which depends on the history of the system as well as the history of the ancilla entangled with it ericsson .
In the remainder, we investigate how the earlier definitions of geometric phase sjoq1 ; nonunitary can be obtained from the present framework as special cases. If we confine ourselves to a restriction of the solution of (11) for $`\stackrel{~}{𝒱}(t)`$, such that $`\stackrel{~}{𝒱}_{kk^{}}(t)=𝒱_{kk^{}}(t)\delta _{kk^{}}`$ and has the property
$`[\stackrel{~}{𝒱},𝒰^{}C^2𝒰]=0,`$ (15)
or equivalently $`\stackrel{~}{𝒱}(t)=_ke^{il_k(t)}𝒰^{}|w_k(0)w_k(0)|𝒰`$, where $`l(t)`$ is defined as in Eq. (12), then explicit form of $`\gamma _g`$ becomes
$`\gamma _g(t)=\text{arg}({\displaystyle \underset{k}{}}\sqrt{p_k(0)p_k(t)}w_k(0)|w_k(t)e^{il_k(t)}),`$ (16)
as in Ref. nonunitary . Thus, in the context of the discussion of Ref. ericsson2 , it can be said that the physical role of the commutation relation (15) appears like removing memory effects of ancilla’s evolution from geometric phase.
Let us end by mentioning some remarks on the initial assumptions of the approach while our stress is still on derivation of earlier results and their possible generalizations. Based upon Theorem 2, it is seen that one can always choose $`𝒩`$, number of the vectors in uniform decomposition, such that $`𝒩N`$. For example, we can assume that $`𝒩=\text{dim}(_s)`$. Now we show how the whole framework can be modified in the degenerate case. Consider the evolution for the density matrix of the system from $`\varrho (0)`$ to $`\varrho (t)=_{k=1}^N_{\mu =1}^{n_k}p_k(t)|w_k^\mu (t)w_k^\mu (t)|`$, where $`p_k(t)`$, $`k=1,\mathrm{},N`$, are the $`n_k`$-fold degenerate eigenvalues of $`\varrho (t)`$, and $`|w_k^\mu (t)`$, $`\mu =1,\mathrm{},n_k`$, are considered the corresponding eigenvectors. In this case, the pure state of the total system is $`|\mathrm{\Phi }(t)_{sa}=_{k=1}^N_{\mu =1}^{n_k}|\stackrel{~}{x}_k^\mu (t)_s|w_k^\mu (0)_a`$, where $`|\stackrel{~}{x}_k^\mu (t)`$ is defined as in Eq. (6) in which $`|w_k(0)`$ is replaced by $`|w_k^\mu (0)`$. Then, one notes that
$`\begin{array}{cc}\hfill \mathrm{\Phi }(0)|\mathrm{\Phi }(t)=& {\displaystyle \underset{kk^{}\mu \mu ^{}}{}}\sqrt{p_k(0)p_k^{}(t)}w_k^\mu (0)|w_k^{}^\mu ^{}(t)\hfill \\ & \times w_k^\mu (0)|𝒰𝒱(t)𝒰^{}|w_k^{}^\mu ^{}(0),\hfill \end{array}`$ (17)
which is determined when all elements of $`𝒱(t)`$ are known. Now we choose our PT condition in this general case as
$`\widehat{\stackrel{~}{x}}_k^\mu (t)|{\displaystyle \frac{\text{d}}{\text{d}t}}|\widehat{\stackrel{~}{x}}_k^\mu ^{}(t)=0,\mu ,\mu ^{}=1,\mathrm{},n_k.`$ (18)
It can be checked that this PT condition can also be satisfied by assuming Uhlmann’s PT condition, Eq. (LABEL:explicitul). In this case, it is easily seen that the most general form for $`\stackrel{~}{𝒱}`$ which satisfies Eq. (15) is as follows
$`\stackrel{~}{𝒱}(t)={\displaystyle \underset{k\mu \mu ^{}}{}}\alpha _k^{\mu \mu ^{}}(t)𝒰^{}|w_k^\mu (0)w_k^\mu ^{}(0)|𝒰.`$ (19)
After some algebra and using the commutation relation (15), it is obtained that $`\alpha _k^{\mu \mu ^{}}(t)=w_k^\mu (0)|\text{P}e^{_0^tU^{}(t^{})\dot{U}(t^{})\text{d}t^{}}|w_k^\mu ^{}(0)`$, where P denotes path ordering. After inserting this relation back into Eq. (17), non-Abelian factors show up in the geometric phase.
In general, when degeneracies vary in time, a level–crossing like behavior can occur. In this situation, in the discussion of differentiability of the eigenvalues (and eigenvectors) the notion of ordering of the eigenvalues becomes important. For example, it can happen that the natural ordering as $`p_1(t)\mathrm{}p_N(t)`$ (for all $`t`$) destroys differentiability, thus, one has to seek for some ordering which respects it Bhatia . If such an ordering can be found, then the operator $`U(t)`$, eigenvalues, and eigenvectors are still well–defined differentiable functions and our approach may be generalized as well.
In summary, the notion of geometric phase of a mixed state undergoing nonunitary evolution has been investigated in a unifying picture in which two of the previous general definitions, Uhlmann’s definition and kinematic approach, have been related to each other. In this formalism, we have used the idea of purification of state of a system by uniform decomposition of its density matrix rather than the spectral one, and by attaching a time varying ancilla to it. Then, as a natural choice for parallel transport condition, we have considered that a mixed state is undergoing parallel transport condition when all the (normalized) vectors of its corresponding purification are subject to this condition. This generalized parallel transport condition is different from the ones defined previously in the literature. It has been shown that the new conditions are satisfied when Uhlmann’s condition holds. However, because of different numbers of equations in the two parallel transport conditions, the generalized parallel conditions are only diagonal equations of Uhlmann’s condition. Finally, it has been shown how to recover earlier definitions of geometric phase of a mixed state. Extension of the method to the more general cases of degenerate density matrices with time varying degeneracies have also been discussed.
We thank N. Paunkovic for useful discussions. This work was supported by EU project TOPQIP under Contract No. IST-2001-39215. |
warning/0507/quant-ph0507199.html | ar5iv | text | # Quasi-exactly solvable periodic potentials with three known eigenstates
## 1 Introduction
Description of the electron’s motion on a lattice has been investigated for a long time as a central problem of the condensed matter physics. Such quantum problem is reduced to the solving of the Schrödinger equation with some model potential which is periodic often. Therefore, the periodic quantum mechanics problems remain at the investigation’s focus up to now.
The general properties of the solutions of Schrödinger equation with periodic potential energy are described by the oscillation theorem . Energy spectrum of the periodic potential has band structure, i.e. eigenvalues belong to the allowed bands (energy bands) $`[E_0,E_1],[E_1^{^{}},E_2],\mathrm{}`$. The wave functions are the Bloch functions, which are bounded and extended on the full real axe
$$\psi (x+L)=\mathrm{exp}^{ikL}\psi (x),$$
(1)
where $`L`$ is potential period and $`k`$ is a so-called quasi-momentum. The limits of the energy bands are given by the equation $`kL=\{0,\pi \}`$, and the wave functions, which belong to the limiting energy values, satisfy the condition $`\psi (x+L)=\pm \psi (x)`$. These energy values and wave functions are often called the eigenvalues and the eigenfunctions of the described above problem.
Oscillation theorem claims that in the case of periodic potentials the eigenfunctions, which belong to the limits of the energy bands and are arranged in the energy of the increasing order $`E_0E_1E_1^{}E_2E_2^{}E_3\mathrm{}`$, are the periodic functions with the period $`L,2L,2L,L,L,2L,2L,\mathrm{}`$ and have $`0,1,1,2,2,3,3,\mathrm{}`$ nodes in the interval $`L`$ respectively.
Despite long term investigations, there is rather a limited number of exactly solvable periodic potentials even in one dimension. The classical examples are the Kronig-Penney model potential or Lamé’s potentials .
Because of limited number of the exactly solvable potentials, recently much attention has been given to the quasi exactly solvable (QES) potentials for which a finite number of the energy levels and the corresponding wave functions are known explicitly. A general treatment of the quasi exact solvability has been introduced by Turbiner and Ushveridze . The class of QES trigonometric potentials was presented in . In it was shown that the Lamé equation is a peculiar example of QES systems. The authors of the paper considered a family of spectral equation which extends those of . Authors of has applied quantum Hamilton-Jacobi formalism to the QES periodic potentials. In the latest paper an unified treatment of quasi exactly solvable potentials was proposed.
The powerful tool for studying the problem of exact solvability of the Schrödinger equation is the supersymmetric (SUSY) quantum mechanic introduced by Witten (for a review of SUSY quantum mechanics see ). The SUSY method for constructing QES potentials was used for the first time in \- . The idea of this method starts from some initial QES potential with $`n+1`$ known eigenstates and using the properties of the unbroken supersymmetry to obtain the SUSY partner potential, which is a new QES one with $`n`$ know eigenstates.
In - using the formalism of SUSY quantum mechanics a large number of new solvable and QES periodic potential was proposed. It is worth mentioning recent paper , where the highest order SUSY transformations was applied for studying periodic potential.
In recent Tkachuk’s papers - a new SUSY method for constructing of the QES potentials with two and three known eigenstates has been proposed. This method does not require knowledge of the initial QES potential in order to generate a new QES one. Within the frame of this method QES potentials has been obtained for which the explicit form of the energy levels and the wave functions of the ground and the excited states can be found. After the paper by Dolya and Zaslavskii, where they showed how to generate QES potentials with arbitrary two known eigenstates without resorting to the SUSY quantum mechanics, SUSY method has been extended for constructing QES potentials with arbitrary two known eigenstates. In our recent works using the SUSY method periodic and disordered QES potentials were obtained.
In the present paper using the results of previous study - we extend Tkachuk’s SUSY method for constructing QES periodic potentials with three known eigenstates.
## 2 The Witten model of SUSY quantum mechanics
Witten model of supersymmetric quantum mechanics is a quantum mechanics of the matrix Hamiltonian
$$H=\left(\begin{array}{ccc}H_+\hfill & 0\hfill & \\ 0\hfill & H_{}\hfill & \end{array}\right),$$
(2)
where Hamiltonians
$$H_\pm =\frac{1}{2}\frac{d^2}{dx^2}+V_\pm (x)=B^{}B^\pm $$
(3)
are supersymmetric partners and
$$B^\pm =\frac{1}{\sqrt{2}}\left(\frac{d}{dx}+W(x)\right).$$
(4)
Here $`\mathrm{}=m=1`$ units are used. Function $`W(x)`$ is referred to as superpotential, $`V_\pm (x)`$ are so-called supersymmetric partner potentials
$$2V_\pm (x)=W^2(x)\pm W^{}(x).$$
(5)
Energy spectrum of the supersymmetric partners $`H_+`$ and $`H_{}`$ is identical except for zero-energy ground state which exists in the case of the unbroken supersymmetry. This leads to twofold degeneracy of the energy spectrum of $`H`$, except for the unique zero-energy ground state. Only one of the Hamiltonians $`H_\pm `$ has zero-energy eigenvalue. We shall use the convention that the zero-energy eigenstate belongs to $`H_{}`$
$$\{\begin{array}{ccc}E_{n+1}^{}\hfill & =\hfill & E_n^+\hfill \\ E_0^{}\hfill & =\hfill & 0\hfill \end{array},$$
(6)
where $`n=0,1,2,\mathrm{}`$. The wave functions of the supersymmetric partners $`H_\pm `$ are related by the supersymmetric transformations
$$\{\begin{array}{ccc}\psi _{n+1}^{}(x)\hfill & =\hfill & \frac{1}{\sqrt{E_n^+}}B^+\psi _n^+(x)\hfill \\ \psi _n^+(x)\hfill & =\hfill & \frac{1}{\sqrt{E_{n+1}^{}}}B^{}\psi _{n+1}^{}(x)\hfill \end{array}.$$
(7)
Due to the factorization $`H_{}=B^+B^{}`$, we can find solution of the Schrödinger problem for the eigenstate with zero energy
$$H_{}\psi _0^{}(x)=E_0\psi _0^{}(x)=0.$$
(8)
It is easy to see that
$$\psi _0^{}(x)=C_0^{}\mathrm{exp}(W(x)𝑑x),$$
(9)
where $`C_0^{}`$ is an arbitrary constant.
In the present paper we shall consider the systems on the full real axe $`\mathrm{}<x<\mathrm{}`$ with periodic superpotential. The periodic superpotential $`W(x+L)=W(x)`$ leads to the periodic potential energy $`V_\pm (x+L)=V_\pm `$, which results in the bounded and extended eigenfunction. A satisfactory condition for the existence of periodic eigenfunctions, written in the terms of the SUSY quantum mechanics, is
$$_0^LW(x)=0.$$
(10)
In a detailed analysis of the SUSY quantum mechanics was made for this case.
## 3 SUSY constructing QES potentials
We shall study the Hamiltonian $`H_{}`$ with the potential energy
$$V_{}(x)=W_0^2(x)W_0^{}(x),$$
(11)
the ground state of which is given by (9).
Let us consider Hamiltonian $`H_+`$ which is the SUSY partner of Hamiltonian $`H_{}`$ . If we calculate the ground state of $`H_+`$ we immediately find the first excited state of $`H_{}`$ using the degeneracy of the spectrum of SUSY Hamiltonian and SUSY transformations (7). In order to calculate the ground state of $`H_+`$ let us rewrite Hamiltonian in the following form
$$H_+=H_{}^{(1)}+ϵ,ϵ>0$$
(12)
where
$`H_{}^{(1)}=B_1^+B_1^{},`$ (13)
$`B_1^\pm ={\displaystyle \frac{1}{\sqrt{2}}}\left({\displaystyle \frac{d}{dx}}+W_1(x)\right),`$ (14)
and $`W_1(x)`$ is a some new function. Note that $`ϵ`$ is the energy of the ground state of $`H_+`$ since $`H_{}^{(1)}`$ has zero-energy ground state.
The ground state wave function of $`H_+`$ with the energy $`E=ϵ`$ is also zero energy wavefunction of $`H_{}^{(1)}`$ and it satisfies the equation
$$B_1^{}\psi _0^+(x)=0.$$
(15)
The solution of this equation is
$$\psi _0^+(x)=C_0^+\mathrm{exp}\left(W_1(x)𝑑x\right),$$
(16)
where $`C_0^+`$ is an arbitrary constant.
Using the SUSY transformation (7) we can calculate the wavefunction of the first excited state of $`H_{}`$. Repeating the described procedure for $`H_{}^{(1)}`$ we can obtain the second excited state for $`H_{}`$ and so on. This procedure is well known in the SUSY quantum mechanics (see review for example). The wavefunctions and corresponding energy levels read
$$\{\begin{array}{ccc}\psi _n^{}(x)\hfill & =\hfill & C_n^{}B_0^+\mathrm{}B_{n2}^+B_{n1}^+\mathrm{exp}\left(W_n(x)𝑑x\right)\hfill \\ E_n^{}\hfill & =\hfill & _{i=0}^{n1}ϵ_i\hfill \end{array},$$
(17)
where $`n=1,2,\mathrm{},N`$; $`ϵ_0=ϵ`$, $`B_0^\pm =B^\pm `$, $`W_0(x)=W(x)`$, $`C_n^{}`$ are an arbitrary constants. Operators $`B_n^\pm `$ are given by (4) with the superpotentials $`W_n(x)`$.
Equation (12) rewritten for N steps
$$H_+^{(n)}=H_{}^{(n+1)}+ϵ_n,$$
(18)
where $`n=0,1,\mathrm{},N1`$, leads to the set of equations for superpotentials
$$W_n^2(x)+W_n^{}(x)=W_{n+1}^2(x)W_{n+1}^{}(x)+2ϵ_n,$$
(19)
where $`n=0,1,\mathrm{},N1`$.
Unfortunately, each of the equations in (19) are the Rikatti equation, which can not be solved in the general case. Previously this set of equations was solved in special cases of shape-invariant potentials and self-similar potentials for arbitrary $`N`$ (see review ). For $`N=1`$ in the context of parasupersymmetric quantum mechanics one can obtain a general solution of (19) without restricting ourselves to shape-invariant and self-similar potentials . In recent papers - a solution of (19) for $`N=1`$ and $`N=2`$ in order to obtain non-singular QES potentials with two and three known eigenstates respectively has been constructed.
Let us write set of equations (19) for the case $`N=2`$ in the explicit form
$$\{\begin{array}{c}W_0^2(x)+W_0^{}(x)=W_1^2(x)W_1^{}(x)+2ϵ_0\hfill \\ W_1^2(x)+W_1^{}(x)=W_2^2(x)W_2^{}(x)+2ϵ_1\hfill \end{array}.$$
(20)
It is convenient to introduce new functions
$$\begin{array}{cc}\{\begin{array}{c}W_+(x)=W_1(x)+W_0(x)\hfill \\ W_{}(x)=W_1(x)W_0(x)\hfill \end{array}\{\begin{array}{c}\stackrel{~}{W}_+(x)=W_2(x)+W_1(x)\hfill \\ \stackrel{~}{W}_{}(x)=W_2(x)W_1(x)\hfill \end{array},\hfill & \end{array}$$
(21)
then superpotentials can be rewritten in the following form
$$\begin{array}{cc}\{\begin{array}{c}2W_0(x)=W_+(x)W_{}(x)\hfill \\ 2W_1(x)=W_+(x)+W_{}(x)\hfill \end{array}\{\begin{array}{c}2W_1(x)=\stackrel{~}{W}_+(x)\stackrel{~}{W}_{}(x)\hfill \\ 2W_2(x)=\stackrel{~}{W}_+(x)+\stackrel{~}{W}_{}(x)\hfill \end{array}.\hfill & \end{array}$$
(22)
In the terms of new functions (21) the set of equations (20) read as follows
$$\{\begin{array}{ccc}W_+^{}(x)=W_{}(x)W_+(x)+2ϵ_0\hfill & & \\ \stackrel{~}{W}_+^{}(x)=\stackrel{~}{W}_{}(x)\stackrel{~}{W}_+(x)+2ϵ_1\hfill & & \end{array}.$$
(23)
Note, that there are two terms for the $`W_1(x)`$ in the equations (22) with respect to $`W_{}(x)`$ and with respect to $`\stackrel{~}{W}_+(x)`$. This gives us a possibility to obtain relation between $`W_+(x)`$ and $`\stackrel{~}{W}_+(x)`$
$$W_+(x)+\frac{W_+^{}(x)2ϵ_0}{W_+(x)}=\stackrel{~}{W}_+(x)\frac{\stackrel{~}{W}_+^{}(x)2ϵ_1}{\stackrel{~}{W}_+(x)},$$
(24)
here (23) are used. It is easy to rewrite this equation as follows
$`W_+(x)\stackrel{~}{W}_+(x)[\stackrel{~}{W}_+(x)W_+(x)][W_+(x)\stackrel{~}{W}_+(x)]^{}+`$ (25)
$`+2[ϵ_1W_+(x)+ϵ_0\stackrel{~}{W}_+(x)]=0,`$
or
$$U(x)\left(\frac{U(x)}{W_+(x)}W_+(x)\right)U^{}(x)+2\left(ϵ_1W_+(x)+ϵ_0\frac{U(x)}{W_+(x)}\right)=0,$$
(26)
where we have introduced a new function
$$U(x)=W_+(x)\stackrel{~}{W}_+(x).$$
(27)
We arrive again to the Riccati equation with respect to $`U(x)`$. On the other hand, this is an algebraic equation with respect to $`W_+(x)`$, which can be solved explicitly
$$\{\begin{array}{ccc}W_+(x)\hfill & =\hfill & \frac{2U(x)(U(x)+2ϵ_0)}{U^{}(x)(1+R(x))}\hfill \\ \stackrel{~}{W}_+(x)\hfill & =\hfill & \frac{U^{}(x)(1+R(x))}{2(U(x)+2ϵ_0)}\hfill \end{array},$$
(28)
where
$`(x)=1+4{\displaystyle \frac{U(x)(U(x)+2ϵ_0)(U(x)2ϵ_1)}{U^{}(x)^2}},R(x)=\pm \sqrt{(x)}.`$ (29)
The square root $`(x)`$ is a positively defined value, while the function $`R(x)`$ can be chosen in the form of $`(x)`$ or $`(x)`$ within different intervals separated by zeros of the function $`(x)`$.
Thus, we can start from an arbitrary function $`U(x)`$ to construct the functions $`W_+(x)`$ and $`\stackrel{~}{W}_+(x)`$ given by (28). Using (22) we obtain three consequent superpotentials
$$\{\begin{array}{ccc}W_0(x)\hfill & =\hfill & \frac{1}{2}\left(W_+(x)\frac{W_+^{}(x)2ϵ_0}{W_+(x)}\right)\hfill \\ W_1(x)\hfill & =\hfill & \frac{1}{2}\left(W_+(x)+\frac{W_+^{}(x)2ϵ_0}{W_+(x)}\right)\hfill \\ W_2(x)\hfill & =\hfill & \frac{1}{2}\left(\stackrel{~}{W}_+(x)+\frac{\stackrel{~}{W}_+^{}(x)2ϵ_1}{\stackrel{~}{W}_+(x)}\right)\hfill \end{array}.$$
(30)
Then because of (17), we can find the wavefunctions of three explicitly known eigenstates of the Hamiltonian $`H_{}`$
$$\{\begin{array}{ccc}\psi _0^{}(x)\hfill & =\hfill & C_0^{}e^{{\scriptscriptstyle W(x)𝑑x}}\hfill \\ \psi _1^{}(x)\hfill & =\hfill & C_1^{}W_+(x)e^{{\scriptscriptstyle W_1(x)𝑑x}}\hfill \\ \psi _2^{}(x)\hfill & =\hfill & C_2^{}\left((W_0(x)+W_2(x))\stackrel{~}{W}_+(x)\stackrel{~}{W}_+^{}(x)\right)e^{{\scriptscriptstyle W_2(x)𝑑x}}\hfill \end{array}$$
(31)
where energy values are $`E_0^{}=0`$, $`E_1^{}=ϵ_0`$, $`E_2^{}=ϵ_0+ϵ_1`$ and potential energy
$$V_{}(x)=\frac{1}{2}(W_0(x)^2W_0^{}(x)).$$
(32)
Simultaneously we can find the wave function of two explicitly known eigenstates of the Hamiltonian $`H_+`$
$$\{\begin{array}{ccc}\psi _1^+(x)\hfill & =\hfill & B_0^{}\psi _1^{}(x)\hfill \\ \psi _2^+(x)\hfill & =\hfill & B_0^{}\psi _2^{}(x)\hfill \end{array}$$
(33)
with energy values $`E_1^+=ϵ_0`$, $`E_2^+=ϵ_0+ϵ_1`$ and potential energy
$$V_{}(x)=\frac{1}{2}(W_0(x)^2+W_0^{}(x)).$$
(34)
Note that obtained terms for the superpotentials, potentials and wave functions allow existence of two different solutions depending on the selected sign before the square root $`\pm \sqrt{R(x)}`$ in the $`W_+(x)`$ and $`\stackrel{~}{W}_+(x)`$ definitions. Here and later we shall distinguish solutions which were obtained for different signs, by superscript in the parenthesis after the function designation, for example $`Y(x)^{(+)}`$. We will denote as $`Y(x)`$ solutions which are identical for different signs $`Y(x)^{(+)}=Y(x)^{()}`$.
Choosing different generating functions $`U(x)`$ we will obtain different QES potentials (32) with three explicitly known eigenstates (31) and QES potentials (34) with two explicitly known eigenstates (33). Of course, function $`U(x)`$ must satisfy some conditions to provide physical solutions of the Schrödinger equation.
The main obvious condition imposed on the function $`U(x)`$ is a positivity of the expression under the square root (29)
$$1+\frac{4U(x)(U(x)+2ϵ_0)(U(x)2ϵ_1)}{U^{}(x)^2}0$$
(35)
on the all periodicity interval.
Another set of restrictions imposed on function $`U(x)`$ appears due to the requirement of the non-singularity of resulting potential $`V_{}(x)`$. The full analysis of the properties of superpotential $`W_0(x)`$ which provides non-singular potential $`V_{}(x)`$ and wave functions $`\psi _0^{}(x)`$, $`\psi _1^{}(x)`$ in the terms of function $`W_+(x)`$ was done in - for the case of the quasi exactly solvable potentials with two exactly known eigenstates. Below we extend this analysis to the case of the quasi exactly solvable potential with three exactly known eigenstates.
As we can see from the superpotentials $`W_0(x)`$, $`W_1(x)`$ and $`W_2(x)`$ definitions, potential $`V_{}(x)`$ can have poles at the points $`x_0`$ where $`W_+(x_0)=0`$ or $`\stackrel{~}{W}_+(x_0)=0`$. Fortunately, such poles can be removed when
$$\{\begin{array}{ccc}W_+^{}(x_0)\hfill & =\hfill & \pm 2ϵ_0,\hfill \\ \stackrel{~}{W}_+^{}(x_0)\hfill & =\hfill & \pm 2ϵ_1.\hfill \end{array}$$
(36)
Besides, potential energy $`V_{}(x)`$ can have poles at the points of singularity $`x_{\mathrm{}}`$ of the function $`W_+(x)`$. As it was shown in , if function $`W_+(x)`$ at the singularity points $`x_{\mathrm{}}`$ has the behavior
$$W_+(x)=const+\frac{1}{xx_{\mathrm{}}}+o(xx_{\mathrm{}}),$$
(37)
or
$$W_+(x)=\frac{3}{xx_{\mathrm{}}}+o(xx_{\mathrm{}}),$$
(38)
obtained potential energy and wave functions will be continuous functions at the points $`x_{\mathrm{}}`$.
To provide bounded and extended wave functions $`\psi _0^{}(x)`$, $`\psi _1^{}(x)`$, $`\psi _2^{}(x)`$ the conditions (10) should be satisfied
$$\{\begin{array}{ccc}_0^LW_0(x)𝑑x\hfill & =\hfill & 0\hfill \\ _0^LW_1(x)𝑑x\hfill & =\hfill & 0\hfill \\ _0^LW_2(x)𝑑x\hfill & =\hfill & 0\hfill \end{array}.$$
(39)
These conditions are satisfied in the simplest way if the corresponding superpotenials $`W_0(x)`$, $`W_1(x)`$, $`W_2(x)`$ are odd function with regard to the middle of the periodicity interval $`x_m`$. To obtain odd superpotentials it is enough to expect the odd behavior of the function $`W_+(x)`$.
Let us choose the $`U(x)`$ as even function with regard to the middle of the periodicity interval $`x_m`$. Then, if we apply solutions with the different signs before square root to the parts of the periodicity interval from the left and from the right of $`x_m`$, $`W_+(x)`$ will be odd function. It is easy to see if we rewrite term (28) for the $`W_+(x)`$ as follows
$$W_+(x)=\frac{2U(x)(U+2ϵ_0)}{U^{}(x)\pm \sqrt{U^{}(x)^2+4U(x)(U(x)+2ϵ_0)(U(x)2ϵ_1)}}.$$
(40)
Application of the solutions with the different signs leads to the finite breaks of the function $`W_+(x)`$. These breaks can be removed if the value of the function $`W_+(x)`$ will tend to zero both from the left and right direction.
Thus, to provide existence of the bounded extended wave functions, $`U(x)`$ should be even function with regard to the middle of the periodicity interval $`x_m`$, and obtained function $`W_+(x)`$ should have zero at the point $`x_m`$. Function $`W_+(x)`$ can have zeros at the points, where $`U(x)=0`$, and, since $`U(x)`$ should be even function with regard to $`x_m`$, function $`U(x)`$ can have at the point $`x_m`$ zero of the even-order only.
Of course, function $`U(x)`$ can have zeros at the other points of the periodicity interval too. Let us analyze in details the behavior of the superpotentials, potential energy and the wave functions in the vicinity of the $`U(x)`$ zeros.
Let the function $`U(x)`$ have the first-order zeros at the points $`x_0^a`$
$$U(x)=U^{}(x_o^a)(xx_o^a)+\frac{1}{2}U^{\prime \prime }(x_o^a)(xx_o^a)^2+o(xx_o^a)^3.$$
(41)
Then behavior of the functions $`W_+(x)`$, $`\stackrel{~}{W}_+(x)`$ in the vicinity of the points $`x_0^a`$ will be as follows
$$\{\begin{array}{ccc}W_+(x)^{(+)}\hfill & =\hfill & 2ϵ_0(xx_0^a)+o(xx_0^a)^2\hfill \\ W_+(x)^{()}\hfill & =\hfill & \frac{U^{}(x_0^a)}{2ϵ_1}+o(xx_0^a)\hfill \\ \stackrel{~}{W}_+(x)^{(+)}\hfill & =\hfill & \frac{U^{}(x_0^a)}{2ϵ_0}+o(xx_0^a)\hfill \\ \stackrel{~}{W}_+(x)^{()}\hfill & =\hfill & 2ϵ_1(xx_0^a)+o(xx_0^a)^2\hfill \end{array}.$$
(42)
It is easy to see that functions $`W_+(x)`$ and $`\stackrel{~}{W}_+(x)`$ at the points $`x_0^a`$ will have non-zero values or will have zeros which satisfy (36).
Superpotentials $`W_0(x)`$, $`W_1(x)`$, $`W_2(x)`$ will be the following
$$\{\begin{array}{ccc}W_0(x)^{(\pm )}\hfill & =\hfill & A_0^{(\pm )}+o(xx_0^a)\hfill \\ W_1(x)^{(\pm )}\hfill & =\hfill & A_1^{(\pm )}+o(xx_0^a)\hfill \\ W_2(x)^{(\pm )}\hfill & =\hfill & A_2^{(\pm )}+o(xx_0^a)\hfill \end{array},$$
(43)
where
$$\{\begin{array}{ccc}A_0^{(+)}\hfill & =\hfill & A_1^{(+)}=\frac{8ϵ_0^2ϵ_1+U^{}(x_0^a)^2ϵ_0U^{\prime \prime }(x_0^a)}{2ϵ_0U^{}(x_0^a)}\hfill \\ A_1^{()}\hfill & =\hfill & A_2^{()}=\frac{U^{}(x_0^a)^2+ϵ_1(U^{\prime \prime }(x_0^a)8ϵ_0ϵ_1)}{2ϵ_1U^{}(x_0^a)}\hfill \\ A_0^{()}\hfill & =\hfill & A_2^{(+)}=\frac{U^{\prime \prime }(x_0^a)8ϵ_0ϵ_1}{2U^{}(x_0^a)}\hfill \end{array}.$$
(44)
Obtained potential will be regular function too
$$V_{}(x)^{(\pm )}=\alpha _{}^{(\pm )}+o(xx_0^a),$$
(45)
where
$$\{\begin{array}{ccc}\alpha _{}^{()}\hfill & =\hfill & \frac{64ϵ_0^2ϵ_1^2+8ϵ_1U^{}(x_0^a)^2+U^{\prime \prime }(x_0^a)^216ϵ_0(U^{}(x_0^a)^2+ϵ_1U^{\prime \prime }(x_0^a))2U^{}(x_0^a)U^{(3)}(x_0^a)}{8U^{}(x_0^a)^2}\hfill \\ \alpha _{}^{(+)}\hfill & =\hfill & 3\alpha _{}^{()}+4ϵ_0+\frac{U^{(3)}(x_0^a)}{2U^{}(x_0^a)}\hfill \end{array}.$$
(46)
The wave functions $`\psi _0^{}(x)`$, $`\psi _1^{}(x)`$, $`\psi _2^{}(x)`$ will read as follows
$$\{\begin{array}{ccc}\psi _0^{}(x)=1+o(xx_0^a)\hfill & & \\ \psi _1^{}(x)^{(+)}=2ϵ_0(xx_0^a)+o(xx_0^a)^2\hfill & & \\ \psi _1^{}(x)^{()}=\frac{U^{}(x_0^a)}{2ϵ_1}+o(xx_0^a)\hfill & & \\ \psi _2^{}(x)=2ϵ_1+o(xx_0^a)\hfill & & \end{array}.$$
(47)
Thus, at the points $`x_0^a`$, where function $`U(x)`$ has first-order zeros, potential $`V_{}(x)`$ and wave functions $`\psi _0^{}(x)`$, $`\psi _1^{}(x)`$, $`\psi _2^{}(x)`$ will be continuous functions, and wave function $`\psi _1^{}(x)`$ can have simple zeros at the points $`x_0^a`$ depending on the selected sign before the square root.
Let us consider potential $`V_+(x)`$, which is the supersymmetric partner of the obtained potential $`V_{}(x)`$. Potential $`V_+(x)`$ will be regular function in the vicinity of the points $`x_0^a`$ too
$$\{\begin{array}{ccc}V_+(x)^{(\pm )}=\alpha _+^{(\pm )}+o(xx_0^a)\hfill & & \\ \alpha _+^{(+)}=\alpha _{}^{()}+2ϵ_1+\frac{U^{}(x_0^a)^22ϵ_0U^{\prime \prime }(x_0^a)}{4ϵ_0^2}\hfill & & \\ \alpha _+^{()}=\alpha _{}^{(+)}2ϵ_1\hfill & & \end{array}$$
(48)
with continuous wave functions
$$\{\begin{array}{ccc}\psi _1^+(x)=\sqrt{2}ϵ_0+o(xx_0^a)\hfill & & \\ \psi _2^+(x)^{(+)}=2\sqrt{2}ϵ_1(ϵ_0+ϵ_1)(xx_0^a)+o(xx_0^a)^2\hfill & & \\ \psi _2^+(x)^{()}=\frac{(ϵ_0+ϵ_1)U^{}(x_0^a)}{\sqrt{2}ϵ_0}+o(xx_0^a)\hfill & & \end{array}.$$
(49)
Depending on the selected sign before the square root wave function $`\psi _2^+(x)`$ can have nodes at the points $`x_0^a`$.
Now let the function $`U(x)`$ have second-order zeros at the points $`x_0^b`$
$$U(x)=\frac{1}{2}U^{\prime \prime }(x_0^b)(xx_o^b)^2+\frac{1}{6}U^{(3)}(x_o^b)(xx_o^b)^3+o(xx_o^b)^4,$$
(50)
then behavior of the functions $`W_+(x)`$ and $`\stackrel{~}{W}_+(x)`$ will be as follows
$$\{\begin{array}{ccc}W_+(x)=2ϵ_0(xx_0^b)+o(xx_0^b)^{3/2}\hfill & & \\ \stackrel{~}{W}_+(x)=2ϵ_1(xx_0^b)+o(xx_0^b)^{3/2}\hfill & & \end{array}.$$
(51)
i.e. at the points $`x_0^b`$ functions $`W_+(x)`$ and $`\stackrel{~}{W}_+(x)`$ will have zeros. Keeping in mind (36), it is easy to obtain the following coefficient restriction
$$\begin{array}{ccc}U^{\prime \prime }(x_0^b)\hfill & =\hfill & (W_+(x_0^b)\stackrel{~}{W}_+(x_0^b))^{\prime \prime }\hfill \\ & =\hfill & W_+^{\prime \prime }(x_0^b)\stackrel{~}{W}_+(x_0^b)+2W_+^{}(x_0^b)\stackrel{~}{W}_+^{}(x_0^b)+W_+(x_0^b)\stackrel{~}{W}_+^{\prime \prime }(x_0^b)\hfill \\ & =\hfill & 2W_+^{}(x_0^b)\stackrel{~}{W}_+^{}(x_0^b)=8ϵ_0ϵ_1.\hfill \end{array}$$
(52)
Note, that existence of the fractional powers in the series expansion leads to undesired poles of $`W_0(x)`$ at the points $`x_0^b`$
$$W_0(x)^{(\pm )}=\pm \frac{1}{8}\sqrt{\frac{3U^{(3)}(x_0^b)}{ϵ_0ϵ_1(xx_0^b)}}+o(xx_0^b)^{1/2},$$
(53)
which can bring the singularity to the potential energy $`V_{}(x)`$. It is easy to see, that in the case of $`U^{(3)}(x_0^b)=0`$ the fractional powers in the series expansions disappear
$$\{\begin{array}{ccc}W_+(x)\hfill & =\hfill & 2ϵ_0(xx_0^b)+o(xx_0^b)^2\hfill \\ \stackrel{~}{W}_+(x)\hfill & =\hfill & 2ϵ_1(xx_0^b)+o(xx_0^b)^2\hfill \end{array}.$$
(54)
The condition $`U^{(3)}(x_0^b)=0`$ is satisfied in the simplest way if the point $`x_0^b`$ is a middle of the periodicity interval and $`U(x)`$ is even function with regard to the $`x_0^b`$, which at the same time provides the fulfilment of (39).
Then superpotentials read
$$\{\begin{array}{ccc}W_0(x)^{(\pm })\hfill & =\hfill & B_0^{(\pm )}+o(xx_0^b)\hfill \\ W_1(x)^{(\pm })\hfill & =\hfill & B_1^{(\pm )}+o(xx_0^b)\hfill \\ W_2(x)^{(\pm })\hfill & =\hfill & B_2^{(\pm )}+o(xx_0^b)\hfill \end{array},$$
(55)
where
$$\{\begin{array}{ccc}B_0^{(+)}\hfill & =\hfill & B_1^{()}=B_2^{(+)}=B\hfill \\ B_0^{()}\hfill & =\hfill & B_1^{(+)}=B_2^{()}=B\hfill \\ B\hfill & =\hfill & 1/4\sqrt{32(ϵ_0ϵ_1)+U^{(4)}(x_0^b)/(2ϵ_0ϵ_1)}\hfill \end{array},$$
(56)
Obtained potential $`V_{}(x)`$ will be continuous function
$$\{\begin{array}{ccc}V_{}(x)^{(\pm )}\hfill & =\hfill & \beta _{}^{(\pm )}+o(xx_0^a)\hfill \\ \beta _{}^{(\pm )}\hfill & =\hfill & ϵ_0+\frac{U^{(4)}(x_0^b)}{64ϵ_0ϵ_1}\frac{U^{(5)}(x_o^b)}{320ϵ_0ϵ_1B}\hfill \end{array}.$$
(57)
Wave functions $`\psi _0^{}(x)`$, $`\psi _1^{}(x)`$, $`\psi _2^{}(x)`$ will read as follows
$$\{\begin{array}{ccc}\psi _0^{}(x)\hfill & =\hfill & 1+o(xx_0^b)\hfill \\ \psi _1^{}(x)\hfill & =\hfill & 2ϵ_0(xx_0^b)+o(xx_0^b)^2\hfill \\ \psi _2^{}(x)\hfill & =\hfill & 2ϵ_1+o(xx_0^b)\hfill \end{array}.$$
(58)
Thus, in the vicinity of the second-order zero of the $`U(x)`$ potential energy $`V_{}(x)`$ and the wave functions $`\psi _0^{}(x)`$, $`\psi _1^{}(x)`$, $`\psi _2^{}(x)`$ will be continuous functions, if $`x_0^b=x_m`$ is the middle of the periodicity interval and $`U(x)`$ is even function with respect to $`x_0^b`$. Wave function $`\psi _1^{}(x)`$ will have node at the points $`x_0^b`$.
The supersymmetric partner $`V_+(x)`$ of the $`V_{}(x)`$ potential in the vicinity of the $`x_0^b`$ will have the following behavior
$$\{\begin{array}{ccc}V_+(x)^{(\pm )}\hfill & =\hfill & \beta _+^{(\pm )}+o(xx_0^b)\hfill \\ \beta _+^{(\pm )}\hfill & =\hfill & ϵ_02ϵ_1+\frac{U^{(4)}(x_0^b)}{64ϵ_0ϵ_1}\pm \frac{U^{(5)}(x_0^b)}{320ϵ_0ϵ_1B}\hfill \end{array},$$
(59)
with the following wave functions
$$\{\begin{array}{ccc}\psi _1^+(x)\hfill & =\hfill & \sqrt{2}ϵ_0+o(xx_0^b)\hfill \\ \psi _2^+(x)\hfill & =\hfill & 2\sqrt{2}ϵ_1(ϵ_0+ϵ_1)(xx_0^b)+o(xx_0^b)^2\hfill \end{array}.$$
(60)
Thus, in the vicinity of the second-order zeros $`x_0^b`$ of the function $`U(x)`$ potential energy $`V_+(x)`$ and the corresponding wave functions $`\psi _1^+(x)`$, $`\psi _2^+(x)`$ will be continuous function and wave function $`\psi _2^+(x)`$ will have nodes at the points $`x_0^b`$.
Let us analyze the case when the function $`U(x)`$ has the highest order of zeros at the points $`x_0^c`$ using the particular case of the third-order zeros
$$U(x)=\frac{1}{6}U^{(3)}(x_o^c)(xx_o^c)^3+\frac{1}{24}U^{(4)}(x_o^c)(xx_o^c)^4+o(xx_o^c)^5.$$
(61)
Then the series expansion for the function $`W_+(x)`$ in the vicinity of the points $`x_0^c`$ will start from the terms which will be proportional to the $`(xx_0^c)^{3/2}`$, thus, condition (36) will not be satisfied, and then obtained potential energy $`V_{}(x)`$ will have poles at the points $`x_0^c`$. Consequently, function $`U(x)`$ should not have zeros of the highest then second orders.
Singularities at the potential energy, except the zeros of $`U(x)`$, can appear at the points where $`U^{}(x)=0`$ or $`1\sqrt{R(x)}=0`$, that is
$$[\begin{array}{ccc}U^{}(x)\hfill & =\hfill & 0,\hfill \\ U(x)\hfill & =\hfill & 0,\hfill \\ U(x)\hfill & =\hfill & 2ϵ_0,\hfill \\ U(x)\hfill & =\hfill & 2ϵ_1.\hfill \end{array}$$
(62)
Case of $`U(x)=0`$ was considered in the details above. In the vicinity of the points $`a_0`$, where the derivative of $`U(x)`$ is equal to zero, i.e. $`U^{}(a_0)=0`$ and $`U(a_0)0`$, generating function $`U(x)`$ can be written as
$$U(x)=U(a_0)+\frac{1}{2}U^{\prime \prime }(a_0)(xa_0)^2+o(xa_0)^3.$$
(63)
Then behavior of function $`W_+(x)`$ in the vicinity of $`a_0`$ will be the following
$$W_+(x)^{(\pm )}=\pm \sqrt{\frac{U(a_0)(2ϵ_0+U(a_0))}{U(a_0)2ϵ_1}}+\frac{U^{\prime \prime }(a_0)}{4ϵ_12U(a_0)}(xa_0)+o(xa_0)^2,$$
(64)
in other words, in the vicinity of zeros of $`U^{}(x)`$, which do not coincide with zeros of $`U(x)`$, obtained solutions will be continuous functions.
In the vicinity of $`b_0`$, where $`U(b_0)=2ϵ_1`$, function $`W_+(x)`$ will behave as follows
$$\{\begin{array}{ccc}W_+(x)^{(+)}\hfill & =\hfill & \frac{4ϵ_1(ϵ_0+ϵ_1)}{U^{}(b_0)}+o(xb_0),\hfill \\ W_+(x)^{()}\hfill & =\hfill & \frac{1}{xb_0}+const+o(xb_0).\hfill \end{array}$$
(65)
Despite singularity of the function $`W_+(x)^{()}`$, potential energy and wave functions will be continuous functions, because pole of $`W_+(x)`$ satisfies the condition (37).
In the vicinity of $`c_0`$, where $`U(c_0)=2ϵ_0`$, function $`W_+(x)`$ can be figured out as follows
$$\{\begin{array}{ccc}W_+(x)^{(+)}\hfill & =\hfill & 2ϵ_0(xc_0)+o(xc_0)^2,\hfill \\ W_+(x)^{()}\hfill & =\hfill & \frac{U^{}(c_0)}{2(ϵ_0+ϵ_1)}+o(xc_0),\hfill \end{array}$$
(66)
thus, potential energy and wave functions will be continuous functions again. Consequently, at the all points, where denominator of $`W_+(x)`$ can turn into zero, potential energy $`V_{}(x)`$ and wave functions $`\psi _0^{}(x)`$, $`\psi _1^{}(x)`$, $`\psi _2^{}(x)`$ will be free of singularities.
Similar analysis, which we shall omit due to its inconvenience, with respect to the potential $`V_+(x)`$ shows, that potential $`V_+(x)`$ and corresponding wave functions will be free of singularities at the all considered points except the points $`c_0`$, where potential $`V_+(x)^{(+)}`$ will have pole with the following behavior
$$V_+(x)^{(+)}=\frac{1}{(xc_0)^2}+const+o(xc_0).$$
(67)
Fortunately, this singularity can be avoided if within the parts of periodicity interval which contains $`c_0`$ we apply solution $`V_+(x)^{()}`$ instead of $`V_+(x)^{(+)}`$.
Another way to avoid singularities in the potential $`V_+(x)`$ is to exclude zeros in the denominator of $`W_+(x)`$ by picking up the amplitude of the function $`U(x)`$ in such a manner that equations $`U(x)+2ϵ_0=0`$ and $`U(x)2ϵ_1=0`$ not be fulfilled. Indeed, since energy levels $`ϵ_0`$, $`ϵ_1`$ are positively defined values and $`U(x)`$ is a periodic bounded function, we can always fit the amplitude of generating function $`U(x)`$ using the following rule
$$\{\begin{array}{ccc}ϵ_0<1/2\mathrm{min}U(x),\hfill & & \\ ϵ_1>1/2\mathrm{max}U(x),\hfill & & \end{array}$$
(68)
where $`\mathrm{min}U(x)`$ and $`\mathrm{max}U(x)`$ \- minimal and maximal values of the $`U(x)`$ at the periodicity interval respectively.
Thus, periodic function $`U(x)`$ generates quasi exactly solvable potential $`V_{}(x)`$ with three known eigenfunctions $`\psi _0^{}(x)`$, $`\psi _1^{}(x)`$, $`\psi _2^{}(x)`$ for the energy values $`ϵ_0>0`$ and $`ϵ_1>0`$, if $`R(x)0`$ for all periodicity interval. Simultaneously, function $`U(x)`$ generates quasi exactly solvable potential $`V_+(x)`$ with two known eigenfunctions $`\psi _1^+(x)`$, $`\psi _2^+(x)`$ in the case of $`U(x)(2ϵ_0;2ϵ_1)`$ and $`R(x)0`$ for all periodicity interval.
To provide free of singularities potential energy and extended bounded wave functions, $`U(x)`$ must be even function with respect to the middle of the periodicity interval $`x_m`$ and must have second order zero at this point. Generating function $`U(x)`$ may have first-order zeros at the other points of the periodicity interval and should not have zeros of the highest order. The derivative of the $`U^{\prime \prime }(x)`$ at the point $`x_m`$ should satisfy the condition $`U^{\prime \prime }(x_m)=8ϵ_0ϵ_1`$. It is necessary to use solutions with opposite signs from the left and right sides with regard to point $`x_m`$.
To illustrate the above described method we give a short example.
Trigonometric extension of the Razavy potential. Let us start from the generating function
$$U(x)=4ϵ_0ϵ_1\mathrm{sin}^2x.$$
(69)
Similar generating function $`U(x)=4ϵ_0ϵ_1\mathrm{sinh}^2x`$ at $`ϵ_1=ϵ_0+1/2`$ gives well known quasi exactly solvable Razavy potential . Than $`(x)`$ can be rewritten in the following form
$$(x)=(1+2ϵ_02ϵ_1+4ϵ_0ϵ_1\mathrm{sin}^2x)\mathrm{tan}^2x.$$
(70)
We shall omit the general expression for the superpotentials and potential energy as it is huge and rather useless. There are at least three sets of $`ϵ_0`$, $`ϵ_1`$, which allow us to resolve the root in the function $`R(x)`$ and therefore to significantly simplify the final results.
The first set is
$$\{\begin{array}{ccc}\hfill 4ϵ_0ϵ_1& =\hfill & 0\hfill \\ \hfill 1+2ϵ_02ϵ_1& \hfill & 0\hfill \end{array},$$
(71)
for which we obtain trivial solution $`ϵ_0=0`$ or $`ϵ_1=0`$, what leads to the $`U(x)=0`$.
In the case of the second set
$$\{\begin{array}{ccc}\hfill 1+2ϵ_02ϵ_1& =\hfill & 4ϵ_0ϵ_1\hfill \\ \hfill 1+2ϵ_02ϵ_1& \hfill & 0\hfill \end{array}$$
(72)
we obtain $`ϵ_1=1/2`$. Then
$$W_+(x)=\frac{ϵ_0\mathrm{sin}2x}{1+\sqrt{2ϵ_0}\mathrm{sin}x}.$$
(73)
Function $`W_+(x)`$ has zeros at the points $`x_k=\pi n/2,n=0,\pm 1,\mathrm{}`$. The derivations $`W_+^{}(x)`$ at this points are $`2ϵ_0/(1+\sqrt{2ϵ_0})`$ or $`2ϵ_0`$ and condition (36) is not fulfilled.
The last set
$$\{\begin{array}{ccc}\hfill 1+2ϵ_02ϵ_1& =\hfill & 0\hfill \\ \hfill 4ϵ_0ϵ_1& \hfill & 0\hfill \end{array}$$
(74)
gives $`ϵ_1=ϵ_01/2`$, then square root can be rewritten in the following form
$$\{\begin{array}{ccc}\hfill R(x)& =\hfill & 2\sqrt{ϵ_0ϵ_1}\mathrm{sin}x\mathrm{tan}x\hfill \\ \hfill ϵ_0& \hfill & 1/2\hfill \end{array}$$
(75)
Function $`W_+(x)`$ reads as follows
$$W_+(x)=\frac{2ϵ_0(\mathrm{cos}^2x+2ϵ_0\mathrm{sin}^2x)\mathrm{tan}x}{1+2\sqrt{ϵ_0ϵ_1}\mathrm{sin}x\mathrm{tan}x}.$$
(76)
Function $`W_+(x)`$ has singularities at the points $`x_k^{(1)}=\pm \mathrm{arccos}\sqrt{ϵ_0/ϵ_1}+2\pi n,n=0,\pm 1,\mathrm{}`$ and $`x_k^{(2)}=\pm \mathrm{arccos}(\sqrt{ϵ_1/ϵ_0})+2\pi n,n=0,\pm 1,\mathrm{}`$. Due to the limitation $`ϵ_01/2`$, solutions $`x_k^{(1)}`$ belong to the complex space and thus, can be dismissed. At the points $`x_k^{(2)}`$ function $`W_+(x)`$ has simple poles with the pole coefficient $`1`$, thus potential energy $`V_{}(x)`$ will be regular function at points $`x_k^{(2)}`$ for any $`ϵ_0`$. Additionally, function $`W_+(x)`$ has simple zeros at the points $`x_k=\pi n,n=0,\pm 1,\mathrm{}`$. The derivations $`W_+(x)`$ at all these points are equal to $`2ϵ_0`$, so all conditions imposed on generating function $`U(x)`$ to provide non-singular real potential energy $`V_{}(x)`$ are satisfied for any $`ϵ_0>1/2`$.
Then, using the definition of function $`\stackrel{~}{W}_+(x)`$ (28), solution for superpotentials $`W_0(x)`$, $`W_1(x)`$, $`W_2(x)`$ (22) and relation between superpotential $`W_0(x)`$ and potential energy $`V_{}(x)`$ (11), we can find three eigenstates of the potential
$$V_{}(x)=ϵ_0\frac{1}{2}+\frac{1}{4}\left(ϵ_0ϵ_16\sqrt{ϵ_0ϵ_1}\mathrm{cos}xϵ_0ϵ_1\mathrm{cos}2x\right),$$
(77)
where $`ϵ_1=ϵ_01/2`$. The energy values of this eigenstates are $`E_0^{}=0`$, $`E_1^{}=ϵ_0`$, $`E_2^{}=ϵ_0+ϵ_1`$ and wave functions are given by (31)
$$\{\begin{array}{ccc}\psi _0^{}(x)\hfill & =\hfill & C_0^{}e^{\sqrt{4ϵ_0ϵ_1}\mathrm{cos}^2\frac{x}{2}}\left(1+4(\sqrt{ϵ_0ϵ_1}+ϵ_1)\mathrm{cos}^2\frac{x}{2}\right)\hfill \\ \psi _1^{}(x)\hfill & =\hfill & C_1^{}e^{\sqrt{4ϵ_0ϵ_1}\mathrm{cos}^2\frac{x}{2}}ϵ_0\mathrm{sin}x\hfill \\ \psi _2^{}(x)\hfill & =\hfill & C_2^{}e^{\sqrt{4ϵ_0ϵ_1}\mathrm{cos}^2\frac{x}{2}}2ϵ_1\left(1+4(\sqrt{ϵ_0ϵ_1}ϵ_0)\mathrm{cos}^2\frac{x}{2}\right)\hfill \end{array}.$$
(78)
Potential $`V_{}(x)`$ and the wave functions $`\psi _0^{}(x)`$, $`\psi _1^{}(x)`$, $`\psi _2^{}(x)`$ are presented at the figure 1.
Because wave function $`\psi _0^{}(x)`$ does not have nodes, eigenstate with energy $`E_0^{}=0`$ is a ground state of this potential. The wave functions $`\psi _1^{}(x)`$ and $`\psi _2^{}(x)`$ have two nodes per interval of periodicity, then eigenstates with energies $`E_1^{}`$ and $`E_2^{}`$ describe the limits of the second forbidden energy band.
This quasi exactly solvable potential belongs to the class of QES potentials presented by Turbiner in his paper in the following form
$$V(x)=\frac{1}{2}(a^2\mathrm{cos}^2(2\alpha x)2\alpha a(2n+1)\mathrm{cos}(2\alpha x)),$$
(79)
in the case of $`n=1`$, $`\alpha =1/2`$; $`a`$ is a free parameter of quantum mechanics problem.
Now let us consider supersymmetric partner of the potential $`V_{}(x)`$:
$$V_+(x)=\frac{1}{2}\left[ϵ_0^2+\frac{3}{2}ϵ_01\sqrt{ϵ_0ϵ_1}\mathrm{cos}xϵ_0ϵ_1\mathrm{cos}^2x\right]+\frac{\underset{i=0}{\overset{7}{}}a_i\mathrm{cos}^ix}{2\underset{i=0}{\overset{8}{}}b_i\mathrm{cos}^ix},$$
(80)
$$\{\begin{array}{c}a_0=16ϵ_0^2ϵ_1^2\hfill \\ a_1=8\sqrt{ϵ_0ϵ_1}ϵ_0(25ϵ_0+2ϵ_0^2)\hfill \\ a_2=12ϵ_0(12ϵ_02ϵ_0^2+4ϵ_0^3)\hfill \\ a_3=8\sqrt{ϵ_0ϵ_1}(1+3ϵ_012ϵ_0^2+6ϵ_0^3)\hfill \\ a_4=1+16ϵ_048ϵ_0^2(1ϵ_0^2)\hfill \\ a_5=6\sqrt{ϵ_0ϵ_1}(1+2ϵ_012ϵ_0^2+8ϵ_0^3)\hfill \\ a_6=8ϵ_1^2ϵ_0(3+2ϵ_0)\hfill \\ a_7=16ϵ_1^2ϵ_0\sqrt{ϵ_0ϵ_1}\hfill \end{array},$$
(81)
$$\{\begin{array}{c}b_0=8ϵ_1ϵ_0^3\hfill \\ b_1=8ϵ_0^2\sqrt{ϵ_0ϵ_1}\hfill \\ b_2=2ϵ_0^2(112ϵ_0+16ϵ_0^2)\hfill \\ b_3=8ϵ_0\sqrt{ϵ_0ϵ_1}(3ϵ_01)\hfill \\ b_4=ϵ_0(1+10ϵ_048ϵ_0^2(1ϵ_0))\hfill \\ b_5=2\sqrt{ϵ_0ϵ_1}(18ϵ_0+12ϵ_0^2)\hfill \\ b_6=2ϵ_1^2(14ϵ_0+16ϵ_0^2)\hfill \\ b_7=8ϵ_1^2\sqrt{ϵ_0ϵ_1}\hfill \\ b_8=8ϵ_0ϵ_1^3.\hfill \end{array}.$$
(82)
Since we know eigenfunctions $`\psi _1^{}(x)`$ and $`\psi _2^{}(x)`$ of Hamiltonian $`H_{}`$, using supersymmetric relations (7) we can find the wave functions $`\psi _1^+(x)`$ and $`\psi _2^+(x)`$, which are eigenfunctions of the Hamiltonian $`H_+`$ with the corresponding energy values $`E_1^+=ϵ_0`$ and $`E_2^+=ϵ_0+ϵ_1`$:
$$\{\begin{array}{c}\psi _1^+(x)=C_1^+ϵ_0e^{\sqrt{4ϵ_0ϵ_1}\mathrm{cos}^2\frac{x}{2}}\frac{\underset{i=0}{\overset{4}{}}k_i\mathrm{cos}^ix}{2\underset{i=0}{\overset{4}{}}l_i\mathrm{cos}^ix}\hfill \\ \psi _2^+(x)=C_2^+e^{\sqrt{4ϵ_0ϵ_1}\mathrm{cos}^2\frac{x}{2}}\mathrm{sin}x\frac{\underset{i=0}{\overset{3}{}}m_i\mathrm{cos}^ix}{2\underset{i=0}{\overset{4}{}}n_i\mathrm{cos}^ix}\hfill \end{array},$$
(83)
where
$$\begin{array}{cc}\{\begin{array}{c}k_0=4\sqrt{2}ϵ_0ϵ_1\hfill \\ k_1=4\sqrt{2ϵ_0ϵ_1}\hfill \\ k_2=\sqrt{2}(8ϵ_0ϵ_11)\hfill \\ k_3=4\sqrt{2ϵ_0ϵ_1}\hfill \\ k_4=4\sqrt{2}ϵ_0ϵ_1\hfill \end{array}\{\begin{array}{c}l_0=4ϵ_0\sqrt{ϵ_0ϵ_1}\hfill \\ l_1=2ϵ_0\hfill \\ l_2=2(14ϵ_0)\sqrt{ϵ_0ϵ_1}\hfill \\ l_3=2ϵ_1\hfill \\ l_4=4ϵ_1\sqrt{ϵ_0ϵ_1},\hfill \end{array}\hfill & \end{array},$$
(84)
$$\{\begin{array}{c}m_0=4\sqrt{2}ϵ_0ϵ_1(4ϵ_01)(ϵ_1\sqrt{ϵ_0ϵ_1})\hfill \\ m_1=2\sqrt{2ϵ_0}(\sqrt{ϵ_1}\sqrt{ϵ_0})(8ϵ_0^314ϵ_0^2+7ϵ_01)\hfill \\ m_2=\sqrt{2}(\sqrt{ϵ_0ϵ_1}ϵ_1)(14ϵ_04ϵ_0^2+16ϵ_0^3)\hfill \\ m_3=4ϵ_1^2\sqrt{2ϵ_0}(\sqrt{ϵ_1}\sqrt{ϵ_0})(4ϵ_01)\hfill \end{array},$$
(85)
$$\{\begin{array}{c}n_0=2ϵ_0\sqrt{ϵ_0ϵ_1}\hfill \\ n_1=ϵ_0\hfill \\ n_2=2(ϵ_0+ϵ_1)\sqrt{ϵ_0ϵ_1}\hfill \\ n_3=ϵ_1\hfill \\ n_4=2ϵ_1\sqrt{ϵ_0ϵ_1.}\hfill \end{array}$$
(86)
Thus we obtain QES potential $`V_+(x)`$ (80) with two exactly know eigenstates $`E_1^+=ϵ_0`$, $`\psi _1^+(x)`$ and $`E_2^+=ϵ_0+ϵ_2`$, $`\psi _2^+(x)`$ given by (83). Potential $`V_+(x)`$ and the wave functions $`\psi _1^+(x)`$, $`\psi _2^+(x)`$ are presented at the figure 2.
Because the wave functions $`\psi _1^+(x)`$ and $`\psi _2^+(x)`$ have two nodes per periodicity interval, the eigenstates with energy values $`E_1^+`$ and $`E_2^+`$ describe limits of the second forbidden energy band. Note, that QES potential (80) does not belong to the general Turbiner’s case and is completely new.
## 4 Conclusions
In the present paper we have extended the SUSY method of constructing well-like QES potentials with three known eigenstates potentials for the case of periodic potentials.
Thus, periodic function $`U(x)`$ generates quasi exactly solvable potential $`V_{}(x)`$ with three known eigenstates $`\psi _0^{}(x)`$, $`\psi _1^{}(x)`$, $`\psi _2^{}(x)`$ and quasi exactly solvable potential $`V_+(x)`$ with two known eigenstates $`\psi _1^+(x)`$, $`\psi _2^+(x)`$.
Since we are interested in the real potential energy, condition $`R(x)0`$ should be satisfied. To provide free of singularities potential energy and extended bounded wave functions, generating function $`U(x)`$ must have second order zero at the middle of the periodicity interval $`x_m`$ and must be even function with respect to this point. $`U(x)`$ may have first-order zeros at the other points of the periodicity interval and should not have zeros of the highest order. The derivative of the $`U^{\prime \prime }(x)`$ at the point $`x_m`$ should satisfy the condition $`U^{\prime \prime }(x_m)=8ϵ_0ϵ_1`$. It is necessary to use solutions for the superpotentials, potentials and wave functions with opposite signs from the left and right sides with regard to point $`x_m`$ to obtain continuous extended wave functions.
As an example of the above described method starting from the generating functions $`U(x)=4ϵ_0ϵ_1\mathrm{sin}^2x`$ we have obtained QES periodic potential $`V_{}(x)=ϵ_01/2+1/4(ϵ_0ϵ_16\sqrt{ϵ_0ϵ_1}\mathrm{cos}xϵ_0ϵ_1\mathrm{cos}2x)`$, which is trigonometric extension of the well known Razavy QES potential, with three known eigenstates $`E_0^{}=0`$, $`E_1^{}=ϵ_0`$ and $`E_2^{}=ϵ_0+ϵ_1`$, where $`ϵ_1=ϵ_01/2`$ and $`ϵ_0`$ is a free parameter. This potential belongs to the class of QES potentials presented by Turbiner at . Eigenstate with energy $`E_0^{}=0`$ is the ground state of this potential. Eigenstates with energies $`E_1^{}`$ and $`E_2^{}`$ describes the limits of the second forbidden energy band.
The supersymmetric partner $`V_+(x)`$ of potential $`V_{}(x)`$ gives us a new QES periodic potential for which we know two eigenstates $`E_1^+=ϵ_0`$ and $`E_2^+=ϵ_0+ϵ_2`$ in the explicit form, where $`ϵ_1=ϵ_01/2`$ and $`ϵ_0`$ is a free parameter. This eigenstates describe the limits of the second forbidden energy band.
Author is grateful to V. M. Tkachuk for enlightening suggestions, helpful comments and discussions. |
warning/0507/cond-mat0507631.html | ar5iv | text | # Quantum dynamics of localized excitations in a symmetric trimer molecule
## I Introduction
The study of the classical and quantum dynamics of excitations in non-linear systems with few degrees of freedom has been used for decades to understand the processes of energy redistribution after an initial local bond excitation in polyatomic molecules Uzer ; Jaffe ; keshavamurthy ; Carney ; Comp ; Farantos . Equally, interest in these issues evolved from the more mathematical perspective of nonlinear dynamics, localization of energy and solitons EilbeckPhysicaD16 . This second path was boosted by the observation of discrete breathers (DB) - time-periodic and spatially localized excitations - in a huge variety of spatially discrete lattice systems FlachPhysRep295 ; physicstoday ; Sievers ; AubryPhysicaD103 . The flood of recent experimental observations of DBs in various systems includes such different systems as bond excitations in molecules, lattice vibrations and spin excitations in solids, electronic currents in coupled Josephson junctions, light propagation in interacting optical waveguides, cantilever vibrations in micromechanical arrays, cold atom dynamics in Bose-Einstein condensates loaded on optical lattices, among others SchwarzPRL83 ; SatoNature ; SwansonPRL82 ; TriasPRL84 ; BinderPRL84 ; EisenbergPRL81 ; FleischerNature422 ; SatoPRL90 ; TrombettoniPRL86 ; OstrovskayaPRL90 ; EiermannPRL92 . In a substantial part of these cases quantum dynamics of excitations is either unavoidable (molecules, solids) or reachable by corresponding parameter tuning (Josephson junctions, Bose-Einstein condensates).
Progress in classical theory was achieved using a synergy of analytical results and computational approaches. The computational aspect is vital because we deal generically with non-integrable systems, which can not be completely solved analytically. Computational studies of classical systems of many interacting degrees of freedom (say $`N`$ oscillators) are straightforward since we have to integrate $`2N`$ coupled first-order ordinary differential equations, so $`N10^4`$ is no obstacle to do even long-time simulations in order to study statistical properties. The quantum case is much less accessible by computational studies. This is because in general each degree of freedom (e.g. an oscillator) is now embedded in an infinite-dimensional Hilbert space. Even after restricting to only $`s`$ states per oscillator, the dimension of the Hilbert space of the interacting system is now $`s^N`$, making it nearly impossible to treat both large values of $`s`$ and $`N`$ \- independent of whether we aim at integrating the time-dependent Schroedinger equation or diagonalizing the corresponding Hamiltonian. However it is possible to treat small systems with $`N=2,3`$, which adds to the above mentioned studies of bond excitations in molecules, perspective cases of few coupled Josephson junctions, and Bose-Einstein condensates in optical traps with just a few wells SmerziPRL79 ; RaghavanPRA59 . Remarkably, in the last case there is already an experimental realization AlbiezPRL95 .
Extensive studies of a dimer model $`N=2`$ with additional conservation of the number of excitations (bosons) have been accomplished ScottPhysLettA119 ; BernsteinPhysicaD68 ; Aubry ; BernsteinNonlin3 ; kalosakas1 ; kalosakas2 ; largespin1 ; largespin2 . The conservation of energy and boson number makes this system integrable. Due to the nonlinearity of the model the invariance under permutation of the two sites (bonds, spin flips etc) is not preventing from having classical trajectories which are not invariant under permutations. These trajectories correspond to a majority of bosons (and thus energy) being concentrated on one of the sites. Quantum mechanics reinforces the symmetry of the eigenstates via dynamical tunneling in phase space (without obvious potential energy barriers being present) Davis ; Kesha2 . The tunneling time is inversely proportional to the energy splitting of the corresponding tunneling pairs of eigenstates. Notice that while most of the quantum computations concerned diagonalization of the Hamiltonian, a few results show consistency with numerical integration of the Schroedinger equation kalosakas2 ; Flach1 . In AlbiezPRL95 the first experimental observation of non-linearity-induced localization of Bose-Einstein condensates in a double-well system was obtained, in agreement with results discussed above.
The extension of the dimer to a trimer $`N=3`$ allows to study the fate of the tunneling pairs in the presence of nonintegrability WrightPhysicaD69 ; Flach1 ; Fillippo ; Cruzeiro ; Chefles ; FlachPRB63 and in an effective presence of the fluctuation of the number of bosons (on the dimer). Diagonalization showed that tunneling pairs survive up to a critical strength of nonintegrability Flach1 , while the pair splittings showed characteristic resonances due to interactions with other eigenstates FlachPRB63 .
Trimer models have been also extensively studied in order to describe spectral properties and then energy transfer in ABA molecules like water Lawton ; Child ; Sibert ; Schmid ; Kellman , which is connected with the appearance of quantum local modes (discrete breathers). In these studies the presence of local modes was already identified as nearly degenerate eigenstates (tunneling pairs) in the eigenvalue spectra of the considered systems.
In this work we study the time evolution of localized excitations in the trimer, and compare with the spectral properties of the system. We compute the eigenvalues and eigenstates of the quantum system and then the expectation values of the number of bosons at every site on the trimer and the survival probability of different initial excitations as a function of time. We also compute the spectral intensity of the initial excitations to see how many eigenstates overlap are involved. That allows to draw conclusions about the correspondence between the time evolution of a localized initial quantum state (not an eigenstate) and the presence or absence of quantum breathers, i.e. dynamical tunneling eigenstates. We identify novel degeneracies in the trimer spectrum due to avoided crossings, and relate these events to unusual classical-like behaviour of quantum localized excitations.
## II Local bond excitations in the classical case
The classical trimer is described by the Hamiltonian Flach1
$`H=H_d+{\displaystyle \frac{1}{2}}(P_3^2+X_3^2)+{\displaystyle \frac{\delta }{2}}(X_1X_3+P_1P_3`$
$`+X_2X_3+P_2P_3),`$ (1)
$`H_d={\displaystyle \frac{1}{2}}(P_1^2+P_2^2+X_1^2+X_2^2)+{\displaystyle \frac{1}{8}}[(P_1^2+X_1^2)^2`$
$`+(P_2^2+X_2^2)^2]+{\displaystyle \frac{C}{2}}(X_1X_2+P_1P_2),`$ (2)
where $`H_d`$ is the dimer part. In all of this work we use dimensionless quantities. $`C`$ is the coupling inside the dimer, and $`\delta `$ is the coupling between site 3 and the dimer which also destroys the integrability of the system. Using the transformation $`\mathrm{\Psi }_i=(1/\sqrt{2})(X_i+\mathrm{i}P_i)`$ the Hamiltonian becomes
$$H=H_d+\mathrm{\Psi }_3^{}\mathrm{\Psi }_3+\delta (\mathrm{\Psi }_1^{}\mathrm{\Psi }_3+\mathrm{\Psi }_2^{}\mathrm{\Psi }_3+cc)$$
(3)
$`H_d=\mathrm{\Psi }_1^{}\mathrm{\Psi }_1+\mathrm{\Psi }_2^{}\mathrm{\Psi }_2+{\displaystyle \frac{1}{2}}[(\mathrm{\Psi }_1^{}\mathrm{\Psi }_1)^2+(\mathrm{\Psi }_2^{}\mathrm{\Psi }_2)^2]`$
$`+C(\mathrm{\Psi }_1^{}\mathrm{\Psi }_2+cc),`$ (4)
and the equations of motion transform to $`\mathrm{i}\dot{\mathrm{\Psi }}_i=H/\mathrm{\Psi }_i^{}`$. Note that the total norm $`B=\mathrm{\Psi }_1^{}\mathrm{\Psi }_1+\mathrm{\Psi }_2^{}\mathrm{\Psi }_2+\mathrm{\Psi }_3^{}\mathrm{\Psi }_3`$ is conserved, and hence the problem is effectively two-dimensional. Also the trimer (and the dimer) is invariant under permutation of sites 1 and 2.
We are interested in the fate of localized excitations, where some energy is excited e.g. on site 1, and none on site 2 (inside the dimer). The third site may have some nonzero energy as well (like an environment). For different initial conditions
$$\mathrm{\Psi }_1(0)=\sqrt{\frac{B}{2}+\nu },\mathrm{\Psi }_2(0)=0,\mathrm{\Psi }_3(0)=\sqrt{\frac{B}{2}\nu }$$
(5)
we computed the time evolution of the quantities $`|\mathrm{\Psi }_i|^2=\mathrm{\Psi }_i^{}\mathrm{\Psi }_i`$ by numerically solving the equations of motion. In all computations we used $`B=40,C=2`$, and $`\delta =1`$. We also generate a Poincare map (Fig.1) using the condition $`\mathrm{\Delta }_{13}=0`$ ($`\mathrm{\Psi }_i=A_ie^{\mathrm{i}\phi _i},\mathrm{\Delta }_{ij}=\phi _i\phi _j`$) and the plane $`X=|\mathrm{\Psi }_1|^2,Y=|\mathrm{\Psi }_2|^2`$.
We observe that for positive $`\nu `$ the evolution is regular and not invariant under permutation, so most of the energy initially placed on site 1 stays there, with site 2 becoming only little excited. Negative values of $`\nu `$ yield chaotic motion which is permutation invariant.
This transition from localization to delocalization of energy is also nicely observed in the temporal evolution in Fig.2. Increasing $`\nu `$ from negative to positive values the energy exchange between sites 1 and 2 of the dimer is stopped.
## III Local bond excitations in the quantum trimer
The quantum trimer is obtained after replacing the complex functions $`\mathrm{\Psi },\mathrm{\Psi }^{}`$ by the bosonic operators $`a`$ and $`a^{}`$ (rewriting $`\mathrm{\Psi }^{}\mathrm{\Psi }=(1/2)(\mathrm{\Psi }^{}\mathrm{\Psi }+\mathrm{\Psi }\mathrm{\Psi }^{})`$ previously to insure the invariance under exchange $`\mathrm{\Psi }\mathrm{\Psi }^{}`$):
$$\widehat{H}=\widehat{H}_d+\frac{3}{2}\widehat{a}_3^{}\widehat{a}_3+\delta (\widehat{a}_1^{}\widehat{a}_3+\widehat{a}_2^{}\widehat{a}_3+c.c.),$$
(6)
$`\widehat{H}_d={\displaystyle \frac{15}{8}}+{\displaystyle \frac{3}{2}}(\widehat{a}_1^{}\widehat{a}_1+\widehat{a}_2^{}\widehat{a}_2)+{\displaystyle \frac{1}{2}}[(\widehat{a}_1^{}\widehat{a}_1)^2+(\widehat{a}_2^{}\widehat{a}_2)^2]`$
$`+C(\widehat{a}_1^{}\widehat{a}_2+c.c.),`$ (7)
where we take $`\mathrm{}=1`$. The boson number operator $`\widehat{B}=\widehat{a}_1^{}\widehat{a}_1+\widehat{a}_2^{}\widehat{a}_2+\widehat{a}_3^{}\widehat{a}_3`$ commutes with the Hamiltonian, so we may diagonalize (6) in the basis of eigenfunctions of $`\widehat{B}`$, $`\{|n_1,n_2,n_3\}`$, where $`n_1,n_2,n_3`$ respectively are the number of bosons at site 1, 2, and 3. There are $`(b+1)(b+2)/2`$ eigenstates in the subspace corresponding to a fixed value of the eigenvalue $`b`$ of $`\widehat{B}`$. Since the Hamiltonian is invariant under permutation between sites 1 and 2 we expanded the wave function in the basis of symmetric and antisymmetric eigenstates of $`\widehat{B}`$, $`\{|n_1,n_2,n_3_{S,A}\}`$, where
$$|n_1,n_2,n_3_{S,A}=\frac{1}{\sqrt{2}}(|n_1,n_2,n_3\pm |n_2,n_1,n_3).$$
(8)
Then the initial state $`|\mathrm{\Psi }_0=|n_0,m_0,l_0`$ writes as
$`|\mathrm{\Psi }_0`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}(|n_0,m_0,l_0_S+|n_0,m_0,l_0_A),`$ (9)
$``$ $`{\displaystyle \frac{1}{\sqrt{2}}}(|\mathrm{\Psi }_0_S+|\mathrm{\Psi }_0_A),`$
In this representation diagonalization of the Hamiltonian reduces to diagonalize two smaller matrices—symmetric and antisymmetric decompositions of $`\widehat{H}`$—whose eigenvalues are $`E_\mu ^{(S,A)}`$, with less computing cost than diagonalization of the full Hamiltonian. All computations were done using this representation.
We computed the time evolution of expectation values of the number of bosons at every site on the trimer $`n_i(t)=\mathrm{\Psi }_t|\widehat{n}_i|\mathrm{\Psi }_t`$ and the survival probability $`P_t=|\mathrm{\Psi }_0|\mathrm{\Psi }_t|^2`$ (see appendix for explicit expressions), starting with various boson number distributions among site 1 and site 3 controlled by the number $`\nu `$: $`|\mathrm{\Psi }_0=|b/2+\nu ,0,b/2\nu `$, with $`b=40,C=2`$, and $`\delta =1`$. In computations we dropped the two first terms of the Hamiltonian, which are diagonal and just shift the spectrum.
### Tunneling pairs and localization
In Fig.3 we show the time evolution of expectation values of the number of bosons in the trimer. When the initial excitation is mainly localized at the third site in the trimer there is a fast redistribution of bosons between the two sites in the non-linear dimer until the dimer sites are equally occupied ( Fig.3-a). As we place more bosons on the dimer (site 1) the tunneling time of the excitation increases rapidly until the time of computation becomes too short to observe slow tunneling. On these timescales we thus observe localization of bosons on one site in the dimer ( Fig.3-b and 3-c), in analogy to the classical case. The reason for this behavior is the appearance of tunneling pairs of symmetric and antisymmetric eigenstates with very close eigenenergies in comparison to the mean energy separation between eigenstates ( Fig.5). These pairs strongly overlap with the initial state, as observable from the spectral intensity $`I_\mu ^0=|\varphi _\mu ^0|^2`$ in the inset of the Fig.4.
The results in Fig.4 show an enhancement of the survival probability with increasing boson number at site 1, which is consistent with the results discussed above. The dominant tunneling pairs in the spectral intensity (inset of Fig.4) give the main contribution to the time dependence of the survival probability Flach1 .
### Avoided crossings and degenerate eigenstates
Energy levels exhibit avoided crossings when we vary the parameter $`\delta `$ which regulates the strength of nonintegrability of the system Flach1 , as shown in the Fig.5. Of particular interest is the outcome of the interaction of a single eigenstate and a tunneling pair. The principal difference between these states is that the member of a tunneling pair has exponentially small weight in the dynamical barrier region, which is roughly defined by $`n_1=n_2`$ in the $`n_1n_2`$ plane. A single eigenstate will in general have much larger weight in this region. Since each eigenstate is either symmetric or antisymmetric, and a tunneling pair consists always of states with both symmetries, the interaction with a third eigenstate will in general allow for an exact degeneracy of two states with different symmetries. While this is in principle possible for any two states of different symmetry, the exponentially small weight of the tunneling pair states in the dynamical barrier region makes a difference. Indeed, a linear combination of two states with large (not exponentially small) weight in a barrier region yields again, though an assymetric state, but one with large weight in the barrier region. Contrary, for the case of a tunneling pair and a single state, we may expect an asymmetric eigenstate which has much less weight in the barrier region, leading to a much stronger localization of the state similar to a classical one.
We analyze three particular avoided crossings identified by numbered circles in Fig.5 by computing the energy separation $`\mathrm{\Delta }E(\delta )`$ between such a single state and a quantum breather tunneling pair. We identify three different situations. The first one shows that the energy levels intersect once in some degeneracy point (Fig.6). At some value of the parameter $`\delta `$ the energy separation between one of the members of the tunneling pair and the single state vanishes. Tunneling is suppressed completely, and then an asymmetric linear combination of the degenerate eigenstates will constitute a non-decaying localized state. This situation has been well described by perturbation theory FlachPRB63 , where effects of other eigenstates have been neglected. We computed the density $`\rho (n_1,n_2)=|n_1,n_2,n_3|\varphi |^2`$ of the asymmetric eigenstate $`|\varphi =(|\varphi _d^{(S)}+|\varphi _d^{(A)})/\sqrt{2}`$, where $`|\varphi _d^{(S)}`$ and $`|\varphi _d^{(A)}`$ are the degenerate eigenstates. The result is shown in the Fig.7 where we can see that there is only a partial localization of the excitation, since the wave function has visible contributions around the diagonal $`X=Y`$ ($`n_1=n_2`$). Note that in addition it also shows sizable contribution on the other side of the barrier ($`n_12,n_226`$ in Fig.7). In fact the expectation values of the number of bosons for this state are $`n_1=14.99,n_2=14.89,n_3=10.12`$. Thus in terms of averages practically no localization occurs since $`n_1n_2`$ despite the observable asymmetry in Fig.7.
The other two cases appear as a consequence of the influence of other states in the spectrum. In one case the energy levels do not intersect at all ( Fig.8), due to the presence of another avoided crossing located nearby. In the third case surprisingly we observe that the energy levels intersect twice. The situation is shown in Fig.9, and is sketched in the inset of the figure. Due to the interaction with other states of the system we observe an intersection of the two states of the tunneling pair at some distance from the actual avoided crossing with the third state. Consequently both states have exponentially small weight in the barrier region, and we may expect a very strong localization. In Fig.10 we can see that the asymmetric quantum breather in the degeneracy point $`\delta 1.462`$ in Fig.9 (see arrow) is strongly localized and the tunneling is suppressed for all times. Note that in both cases two and three the order of the participating levels before and after the avoided crossing is not conserved, at variance to the first case we discussed above and which was described also in reference FlachPRB63 . The abovementioned strong localization of this exact asymmetric eigenstate is reflected in the fact that the wave function has practically zero weight around the barrier region $`X=Y`$ ($`n_1=n_2`$). Note that at variance with Fig.7, here the wave function has no sizable contribution on the other side of the barrier as well. This state is thus very close to its classical discrete breather counterpart (Fig.1). Indeed, for this state $`n_1=25.62,n_2=2.38,n_3=12.00`$. Consequently we find a very strong localization for the expectation values, in addition to the observed asymmetry in Fig.10.
It is interesting to test whether initial states with some distribution of bosons at every site of the trimer ($`|\mathrm{\Psi }_0=|n_0,m_0,l_0`$) can significantly overlap with the above described asymmetric eigenstates. This distribution is given by the maxima in the density for every asymmetric eigenstate (around $`n_1=26`$, and $`n_2=2`$). In figures 11 and 12 we show the time evolution of the expectation value of the number of bosons at every site on the trimer and the survival probability of such an initial excitation. A detailed analysis of the spectral intensity of the initial state $`|\mathrm{\Psi }_0=|26,2,12`$ (inset in the Fig.12) shows that this initial excitation overlaps strongly with the degenerate eigenstates corresponding to the strong localization shown in Fig.10. It implies that this excitation ( Fig.11-b) will never distribute its quanta evenly over both sites of the dimer. For the case shown in Fig.11-a the initial excitation has a smaller overlap with the degenerate eigenstates which gives the partial localization shown in Fig.7. Since the overlap is not zero the excitation will also stay localized in the sense that the crossing of curves corresponding to $`n_1`$ and $`n_2`$ as in Fig.3-a and 3-b will never occur. Note that despite the difference between the analyzed cases one and three, the evolution of the expectation values and the survival probabilities do not differ drastically. It needs more sensitive details in the preparation of an initial state to observe a practically total localization of bosons on one of the dimer sites for case three as compared to case one.
## IV Conclusions
In this work we observed how spectral properties of the Hamiltonian are reflected in the time evolution of different localized excitations in a trimer molecule model by monitoring the spectrum, the time evolution of expectation values of the number of bosons at every site on the trimer and survival probabilities of different localized excitations.
The tunneling pair splitting determines the lifetime of localized excitations. The survival probability and the time evolution of the expectation values of the number of bosons are clear indicators for a localized excitation being close or far from a quantum breather tunneling pair, while the spectral intensity of localized excitations is typically broad and does not show the peculiarities of the tunneling dynamics. Probing the time evolution of initially localized excited states thus allows to conclude about the presence or absence of tunneling pair eigenstates.
We report on the existence of degenerate levels in the spectrum due to the presence of both avoided crossings and tunneling pairs. In these degenerate points tunneling is suppressed for all times. While in general the asymmetric exact eigenstates will have quite large weight in the dynamical barrier region (contributed by the single level), we observe specific parameter cases where the weight is very small and the corresponding asymmetric eigenstate very strongly localized. Full or partial localization of bosons appears for all time scales for some specific states and some specific values of the parameters. This effect could be studied in experimental situations of Bose-Einstein condensates in few traps which weakly interact, as well as in systems of few coupled Josephson junctions which operate in the quantum regime. Tuning experimental control parameters will allow to lock localized excitations for specific values and both prevent the excitation from tunneling, as well as allowing for a fine tuning of the tunneling frequency from a small value down to zero in the vicinity of these specific control parameter values.
###### Acknowledgements.
We thank G. Kalosakas, S. Keshavamurthy, M. Johansson, and L. Schulman for useful discussions.
*
## Appendix A Expectation values and survival probability
Expanding the wave function in the basis of symmetric and antisymmetric eigenstates of the Hamiltonian
$`|\mathrm{\Psi }_t={\displaystyle \underset{\mu }{}}\varphi _0^{\mu (S)}e^{iE_\mu ^{(S)}t}|\varphi ^{\mu (S)}+`$
$`{\displaystyle \underset{\nu }{}}\varphi _0^{\nu (A)}e^{iE_\nu ^{(A)}t}|\varphi ^{\nu (A)},`$ (10)
where $`\varphi _0^{\mu (S,A)}=\varphi ^{\mu (S,A)}|\mathrm{\Psi }_0_{S,A}`$ and $`\varphi _{n_1,n_2,n_3}^{\mu (S,A)}=\varphi ^\mu |n_1,n_2,n_3_{S,A}`$, the expectation value of the number of bosons at site $`i`$ writes as
$$n_i(t)=n_i^{(S)}(t)+n_i^{(A)}(t)+n_i^{(M)}(t),$$
(11)
where
$`n_1^{(S,A)}(t)`$ $`=`$ $`{\displaystyle \frac{1}{4}}{\displaystyle \underset{\mu ,\mu ^{}}{}}\varphi _0^{\mu (S,A)}\overline{\varphi }_0^{\mu ^{}(S,A)}e^{i(E_\mu ^{(S,A)}E_\mu ^{}^{(S,A)})t}`$ (12)
$`\times F_{\mu ,\mu ^{}}^{(S,A)},`$
$`n_2^{(S,A)}(t)`$ $`=`$ $`n_1^{(S,A)}(t),`$ (13)
$`F_{\mu ,\mu ^{}}^{(S,A)}`$ $`=`$ $`{\displaystyle \underset{\{n_i\}_{S,A}}{}}\overline{\varphi }_{n_1,n_2,n_3}^{\mu (S,A)}(n_1+n_2)\varphi _{n_1,n_2,n_3}^{\mu ^{}(S,A)},`$ (14)
$`n_1^{(M)}(t)`$ $`=`$ $`\mathrm{}\{{\displaystyle \frac{1}{2}}{\displaystyle \underset{\mu ,\nu }{}}\varphi _0^{\mu (S)}\overline{\varphi }_0^{\nu (A)}e^{i(E_\mu ^{(S)}E_\nu ^{(A)})t}`$ (15)
$`\times F_{\mu ,\nu }^{(M)}\},`$
$`n_2^{(M)}(t)`$ $`=`$ $`n_1^{(M)}(t),`$ (16)
$`F_{\mu ,\nu }^{(M)}`$ $`=`$ $`{\displaystyle \underset{\{n_i\}_A}{}}\overline{\varphi }_{n_1,n_2,n_3}^{\mu (S)}(n_1n_2)\varphi _{n_1,n_2,n_3}^{\nu (A)},`$ (17)
$`n_3^{(S,A)}(t)`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \underset{\mu ,\mu ^{}}{}}\varphi _0^{\mu (S,A)}\overline{\varphi }_0^{\mu ^{}(S,A)}e^{i(E_\mu ^{(S,A)}E_\mu ^{}^{(S,A)})t}`$ (18)
$`\times G_{\mu ,\mu ^{}}^{(S,A)},`$
$`G_{\mu ,\mu ^{}}^{(S,A)}`$ $`=`$ $`{\displaystyle \underset{\{n_i\}_{S,A}}{}}\overline{\varphi }_{n_1,n_2,n_3}^{\mu (S,A)}(n_1+n_2)`$ (19)
$`\times \varphi _{n_1,n_2,n_3}^{\mu ^{}(S,A)},`$
$`n_3^{(M)}(t)`$ $`=`$ $`0,`$ (20)
where bars mean complex conjugation. |
warning/0507/physics0507154.html | ar5iv | text | # N-body decomposition of bipartite networks
## I Introduction
It is well-known in statistical physics that N-body correlations have to be carefully described in order to characterize statistical properties of complex systems. For instance, in the case of the Liouville equation for Hamiltonian dynamics, this problem is at the heart of the derivation of the reduced BBGKY hierarchy, thereby leading to the Boltzmann and Enskog theories for fluids balescu . In this line of though, it is primordial to discriminate N-body correlations that are due to intrinsic N-body interactions, from those that merely develop from lower order interactions. This issue is directly related to a well-known problem in complex network theory, i.e. the ”projection” of bipartite networks onto simplified structures. As a paradigm for such systems, people usually consider co-authorship networks newman , namely networks composed by two kinds of nodes, e.g. the scientists and the articles, with links running between scientists and the papers they wrote. In that case, the usual projection method newman2 consists in focusing e.g. on the scientist nodes, and in drawing a link between them if they co-authored a common paper (see Fig.1). As a result, the projected system is a unipartite network of scientists, that characterizes the community structure of science collaborations. Such studies have been very active recently, due to their complex social structure newman3 , to the ubiquity of such bipartite networks in complex systems bara ramasco , and to the large databases available.
A standard quantity of interest in order to characterize the structure of the projected network is the clustering coefficient watts , which measures network ”transitivity”, namely the probability that two scientist’s co-authors have themselves coauthored a paper. In topological terms, it is a measure of the density of triangles in a network, a triangle being formed every time two of one’s collaborators collaborate with each other. This coefficient is usually very high in systems where sociological cliques develop eurovision . However, part of the clustering in co-authorship network is due to papers with three or more coauthors. Such papers introduce trivial triangles of collaborating authors, thereby increasing the clustering coefficient. This problem, that was raised by Newman et al. newman2 , was circumvented by studying directly the bipartite network, in order to infer the authors community structure. Newman et al. showed on some examples that these high order interactions may account for one half of the clustering coefficient. One should note, however, that if this approach offers a well-defined theoretical framework for bipartite networks, it suffers a lack of transparency as compared to the original projection method, i.e. it does not allow a clear visualisation of the unipartite structure.
In this article, we propose an alternative approach that is based on a more refine unipartite projection, and follows Statistical Mechanics usual expansion methods. To do so, we focus on a small dataset, retrieved from the arXiv database and composed of articles dedicated to complex network theory. This choice is motivated by their relatively few co-authors per article, a property typical to theoretical physics papers grossman . Our method consists in discriminating the different kinds of scientists collaborations, based upon the number of co-authors per article. This discrimination leads to a diagram representation feynman ; mayer of co-authorship (see also berg for the applicability of Feynman diagrams in complex networks). The resulting N-body projection reconciles the visual features of the usual projection, and the exact description of Newman’s theoretical approach. Empirical results confirm the importance of high order collaborations in the network structure. Therefore, we introduce in the last section a simple network model, that is based on random triangular connections between the nodes. We study numerically percolation for the model.
## II N-body projection method
The data set contains all articles from arXiv in the time interval $`[1995:2005]`$, that contain the word ”network” in their abstract and are classified as ”cond-mat”. In order to discriminate the authors and avoid spurious data, we checkeed the names and the first names of the authors. Moreover, in order to avoid multiple ways for an author to cosign a paper, we also took into account the initial notation of the prenames. For instance, Marcel Ausloos and M. Ausloos are the same person, while Marcel Ausloos and Mike Ausloos are considered to be different. Let us stress that this method may lead to ambiguities if an initial refers to two different first names, e.g. M. Ausloos might be Marcel or Mike Ausloos. Nonetheless, we have verified that this case occurs only once in the data set (Hawoong, Hyeong-Chai and H. Jeong), so that its effects are negligible. In that sole case, we attributed the papers of H. Jeong to the most prolific author (Hawoong Jeong in the dataset). Given this identification method, we find $`n_P=2533`$ persons and $`n_A=1611`$ articles. The distribution of the number of co-authors per article (Fig.2) shows clearly a rapid exponential decrease, associated to a clear predominance of small collaborations, as expected.
Formally, the bipartite structure authors-papers may be mapped exactly on the vector of matrices $``$ defined by:
$$=[𝐌^{(1)},𝐌^{(2)},\mathrm{},𝐌^{(j)},\mathrm{}.,𝐌^{(n_P)}]$$
(1)
where $`𝐌^{(j)}`$ is a square $`n_P^j`$ matrix that accounts for all articles co-authored by $`j`$ scientists. By definition, the element $`M_{a_1\mathrm{}a_j}^{(j)}`$ are equal to the number of collaborations between the $`j`$ authors $`a_1\mathrm{}a_j`$. In the following, we assume that co-authorship is not a directed relation, thereby neglecting the position of the authors in the collaboration, e.g. whether or not the author is the first author. This implies that the matrices are symmetric under permutations of indices. Moreover, as people can not collaborate with themselves, the diagonal elements $`M_{aa\mathrm{}a}^{(j)}`$ vanish by construction. For example, $`M_{a_1}^{(1)}`$ and $`M_{a_1a_2}^{(2)}`$ represent respectively the total number of papers written by $`a_1`$ alone, and the total number of papers written by the pair ($`a_1`$, $`a_2`$).
A way to visualize $``$ consists in a network whose nodes are the scientists, and whose links are discriminated by their shape. The intrinsic co-authorship interactions form loops (order 1), lines (order 2), triangles (order 3) (see Fig.3)… To represent the intensity of the multiplet interaction, the width of the lines is taken to be proportional to the number of collaborations of this multiplet. Altogether, these rules lead to a graphical representation of $``$, that is much more refine than the usual projection method (Fig.4).
It is important to point out that the vector of matrices $``$ describes without approximation the bipartite network, and that it reminds the Liouville distribution in phase space of a Hamiltonian system. Accordingly, a relevant macroscopic description of the system relies on a coarse-grained reduction of its internal variables. The simplest reduced matrix is the one-scientist matrix $`𝐑^{(1)}`$ that is obtained by summing over the N-body connections, $`N2`$:
$`R_{a_1}^{(1)}=M_{a1}^{(1)}+{\displaystyle \underset{a_2}{}}M_{a_1a_2}^{(2)}+{\displaystyle \underset{a_2}{}}{\displaystyle \underset{a_3<a_2}{}}M_{a_1a_2a_3}^{(3)}+\mathrm{}.`$ (2)
$`+{\displaystyle \underset{a_2}{}}\mathrm{}.{\displaystyle \underset{a_j<a_{j1}}{}}M_{a_1\mathrm{}a_j}^{(j)}+\mathrm{}`$ (3)
It is straightforward to show that the elements $`R_{a_j}^{(1)}`$ denote the total number of articles written by the scientist $`a_j`$. The second order matrix:
$`R_{a_1a_2}^{(2)}=M_{a_1a_2}^{(2)}+{\displaystyle \underset{a_3}{}}M_{a_1\mathrm{}a_3}^{(3)}+\mathrm{}.`$ (4)
$`+{\displaystyle \underset{a_3}{}}\mathrm{}.{\displaystyle \underset{a_j<a_{j1}}{}}M_{a_1\mathrm{}a_j}^{(j)}+\mathrm{}`$ (5)
Its elements represent the total number of articles written by the pair of scientists ($`a_1`$, $`a_2`$). Remarkably, this matrix reproduces the usual projection method (Fig. 1), and obviously simplifies the structure of the bipartite structure by hiding the effect of high order connections. The three-scientist matrix read similarly:
$`R_{a_1a_2a_3}^{(3)}=M_{a_1a_2a_3}^{(3)}+{\displaystyle \underset{a_4}{}}M_{a_1\mathrm{}a_4}^{(4)}+\mathrm{}.`$ (6)
$`+{\displaystyle \underset{a_4}{}}\mathrm{}.{\displaystyle \underset{a_j<a_{j1}}{}}M_{a_1\mathrm{}a_j}^{(j)}+\mathrm{}`$ (7)
This new matrix counts the number of papers co-written by the triplet ($`a_1`$, $`a_2`$, $`a_3`$), and may be represented by a network whose links are triangles relating three authors. The generalization to higher order matrices $`𝐑^{(j)}`$ is straightforward, but, as in the case of the BBGKY hierarchy, a truncature of the vector $``$ must be fixed at some level in order to describe usefully and compactly the system. It is therefore important to point that the knowledge of $`𝐌^{(2)}`$ together with $`𝐑^{(3)}`$ is completely sufficient in order to characterize the triangular structure of $``$. Consequently, in this paper, we stop the reduction procedure at the 3-body level, and define the triangular projection of $``$ by the application:
$`[M_{a1}^{(1)},M_{a_1a_2}^{(2)},M_{a_1a_2a_3}^{(3)},\mathrm{}.,M_{a_1\mathrm{}a_{n_P}}^{(n_P)}]`$ (8)
$`[M_{a1}^{(1)},M_{a_1a_2}^{(2)},R_{a_1a_2a_3}^{(3)}]`$ (9)
The triangular projection is depicted in Fig. 5, and compared to the usual projection method. In order to test the relevance of this description, we have measured in the data set the total number of triangles generated by edges. We discriminate two kinds of triangles: those which arise from one 3-body interaction of $`𝐑^{(3)}`$, and those which arise only from an interplay of different interactions. There are respectively 5550 and 30 such triangles, namely $`99.5\%`$ of triangles are of the first kind. This observation by itself therefore justifies the detailed projection method introduced in this section, and shows the importance of co-authorship links geometry in the characterization of network structures, precisely the clustering coefficient in the present case.
## III Triangular Erdös-Renyi networks
The empirical results of the previous section have shown the significance of N-body connections in social networks. A more complete framework for networks is therefore required in order to describe correctly the system complexity. In this article, we focus on the most simple generalization, namely a network whose links relate triplets of nodes. To so, we base our modeling on the Erdös-Renyi uncorrelated random graph renyi , i.e. the usual prototype to be compared with more complex random graphs. The usual Erdös-Renyi network (ERN) is composed by $`N_n`$ labeled nodes connected by $`N_e^{(2)}`$ edges, which are chosen randomly from the $`N_n(N_n1)/2`$ possible edges. In this paper, we define the triangular ER network ($`\text{ERN}^3`$) to be composed by $`N_n`$ labeled nodes, connected by $`N_e^{(3)}`$ triangles, which are chosen randomly from the $`N_n(N_n1)(N_n2)/6`$ possible triangles. As a result, connections in the system relate triplets of nodes $`(a_1,a_2,a_3)`$, and the matrix vector $``$ reduces to the matrix $`𝐌^{(3)}`$. Before going further, let us point that the clustering coefficient of triangular ER networks is very high by construction, but, contrary to intuition, it is different from 1 in general. For instance, for the two triplets $`(a_1,a_2,a_3)`$ and $`(a_1,a_4,a_5)`$, the local clustering coefficient of $`a_1`$ is equal to $`\frac{1}{3}`$.
In this paper, we focus numerically on the percolation transition vicsek in $`\text{ERN}^3`$, i.e. on the appearance of a giant component by increasing the number of nodes in the system (Fig.6). This transition is usually associated to dramatic changes in the topological structure, that are crucial to ensure communicability between network nodes, e.g. the spreading of scientific knowledge in the case under study. In the following, we work at fixed number of nodes, and focus on the proportion of nodes in the main cluster as a function of the number of binary links in the system. Moreover, in order to compare results with the usual ERN, we do not count twice redundant links, i.e. couples of authors who interact in different triplets. For instance, the triplet $`(a_1,a_2,a_3)`$ accounts for 3 binary links, but $`(a_1,a_2,a_3)`$ and $`(a_1,a_2,a_4)`$ account together for 5 links, so that $`N_e^{(3)}3N_e^{(2)}`$ in general. Whatever, this detailed counting has small effects on the location of the percolation transition. Numerical results are depicted in figure 7, where we consider networks with $`N_n=1000`$. Obviously, the triangular structure of interactions displaces the bifurcation point, by requiring more links in order to observe the percolation transition. This feature comes from the triangular structure of connections that restrains the network exploration as compared to random structures. Indeed, 3 links relate only 3 nodes in $`\text{ERN}^3`$, while 3 links typically relate 4 nodes in ERN. Finally, let us stress that the same mechanism takes place in systems with high clustering coefficients clustering ; preparation .
## IV Conclusion
In this paper, we show the importance of N-body interactions in co-authorships networks. By focusing on data sets extracted from the arXiv database, we introduce a way to project bipartite networks onto unipartite networks. This approach generalizes usual projection methods by accounting for the complex geometrical figures connecting authors. To do so, we present a simple theoretical framework, and define N-body reduced and projected networks. The graphical representation of these simplified networks rests on a ”shape-based” discrimination of the different co-authorship interactions (for a ”color-based” version, see the author’s website website ), and allows a clear visualization of the different mechanisms occurring in the system. Finally, we apply the method to some arXiv data subset, thereby showing the importance of such ”high order corrections” in order to characterize the community structure of scientists. The empirical results motivate therefore a better study of networks with complex weighted geometrical links. In the last section, we focus on the simplest case by introducing a triangular random model, $`\text{ERN}^3`$. Moreover, we restrict the scope by analyzing the effect of the 3-body connection on percolation. A complete study of the topological of $`\text{ERN}^3`$ as well as its generalization to higher order connections is let for a forthcoming work.
Acknowledgements
Figures 3, 4, 5 and 6 were plotted thanks to the visone graphical tools. This work has been supported by European Commission Project CREEN FP6-2003-NEST-Path-012864. |
warning/0507/physics0507120.html | ar5iv | text | # Deterministic approach to microscopic three-phase traffic theory
## 1 Introduction
Theoretical studies of freeway traffic flow dynamics is one of the rapid developing fields of statistical and nonlinear physics (see the reviews , the book , and the conference proceedings ). For a mathematical description of freeway traffic flow, a huge number of different microscopic and macroscopic traffic flow models have been introduced. In macroscopic models, individual dynamic vehicle behaviour is averaged, i.e., these models describe dynamics of average traffic flow characteristics like average vehicle speed and density (see e.g., )<sup>1</sup><sup>1</sup>1It should be noted that transferring the information delivered from one vehicle interacting with the neighbour ones requires to deal carefully with a complex averaging process by derivation of a macroscopic traffic flow model. The related mathematical theory is developed in Ref. .. Microscopic traffic flow models describe individual dynamic vehicle behaviour, which should simulate empirical spatiotemporal features of phase transitions and congested patterns in freeway traffic. In this article, we restrict a consideration of microscopic traffic flow models only.
There are two types of microscopic traffic flow models: Deterministic models and stochastic models . In deterministic models, some dynamic rules of vehicle motion in traffic flow are responsible for spatiotemporal features of traffic patterns that the models exhibit. Contrastingly, stochastic models, in addition to dynamic rules of vehicle motion, exhibit model fluctuations, which play a fundamental role for traffic pattern features.
There are at least two classes of deterministic traffic flow models . In the first class, driver time delays in vehicle acceleration (deceleration) $`a`$ are explicitly taken into account. An example is the classic model of Herman, Montroll, Potts, and Rothery : If the vehicle speed $`v`$, or the speed difference between the vehicle speed and the speed of the preceding vehicle $`v_{\mathrm{}}`$, or else the net distance $`g`$ (space gap) between vehicles changes, then the driver accelerates (decelerates) with a time delay $`\tau `$ :
$$a(t+\tau )=f(v(t),v_{\mathrm{}}(t),g(t)).$$
(1)
Based on (1), Gazis, Herman, and Rothery have developed a microscopic traffic flow model, which is capable of describing traffic beyond of instabilities; steady state solutions of this model lie on a one-dimensional curve in the flow–density plane (the fundamental diagram) (see the review by Nagel et al. for more detail). Recall that steady state solutions are hypothetical model solutions in which all vehicles move at the same time-independent speed and the same space gap between vehicles. One of the mathematical descriptions of this model class first proposed by Nagatani and Nakanishi and further developed by Lubashevky et al. reads as follows
$$\frac{da}{dt}=\frac{f(v(t),v_{\mathrm{}}(t),g(t))a(t)}{\tau }.$$
(2)
In both models , steady state model solutions in the flow–density plane lie on the fundamental diagram.
There is also another class of deterministic microscopic models in which the vehicle speed satisfies the equation :
$$\frac{dv}{dt}=\varphi (v(t),v_{\mathrm{}}(t),g(t)).$$
(3)
Examples are optimal velocity (OV) models of Newell , Whitham , Bando, Sugiyama et al. , and the intelligent driver model (IDM) of Treiber and Helbing . Steady state solutions of this model class that obviously satisfy the conditions $`\varphi (v,v_{\mathrm{}},g)=0`$ and $`v=v_{\mathrm{}}`$ lie on the fundamental diagram in the flow–density plane.
If functions and model parameters in the models (2) and (3) are chosen in an appropriated way, then there is a range of vehicle density in which steady state model solutions for free flow are unstable. This instability, which should explain the onset of congestion, leads to wide moving jam emergence in free flow (F$``$J transition) .
However, as explained in the book , the above models that are in the context of the fundamental diagram approach, as well as all other traffic flow models reviewed in cannot explain the fundamental empirical feature of traffic breakdown, i.e., that the onset of congestion in free flow at a bottleneck is associated with a local first-order phase transition from free flow to synchronized flow (F$``$S transition) rather than with an F$``$J transition. For this reason, Kerner introduced a three-phase traffic theory. In this theory, there are three traffic phases: (i) free flow, (ii) synchronized flow, and (iii) wide moving jam.
The first microscopic models in the context of three-phase traffic theory introduced in 2002 are stochastic models . As in empirical observations , in these models wide moving jams emerge spontaneously only in synchronized flow (S$``$J transition), i.e., the models exhibit the sequence of F$``$S$``$J transitions leading to wide moving jam emergence in free flow; in addition, the models show all types of congested patterns found in empirical observations . Recently, some new microscopic models based on three-phase traffic theory have been developed . However, there are no deterministic models in the context of three-phase traffic theory, which can exhibit the F$``$S$``$J transitions found in empirical observations and the diagram of congested patterns of three-phase traffic theory . In stochastic models , driver time delays in acceleration (deceleration) are simulated mainly through the use of model fluctuations. Therefore, a development of deterministic models based on three-phase traffic theory is important for a more realistic theory of car following behaviour.
In this paper, two deterministic microscopic three-phase traffic models are presented. In an acceleration time delay model (ATD-model for short; Sect. 2), an explicit description of driver time delays in vehicle acceleration (deceleration) is used. In a speed adaptation model (SA-model for short; Sect. 3), vehicle speed adaptation occurs in synchronized flow depending on driving conditions. In Sects. 4 and 5, we show that these models exhibit the F$``$S$``$J transitions and congested patterns associated with results of empirical observations. In addition, a stochastic SA-model is introduced and compared with the deterministic SA-model of Sect. 3. In Sect. 6, the deterministic microscopic three-phase traffic models of Sects. 2 and 3 are compared with earlier deterministic models and a critical discussion of models in the context of the fundamental diagram approach is performed.
## 2 Acceleration Time Delay Model
### 2.1 Driver Behavioural Assumptions and Empirical Basis of ATD-Model
A deterministic three-phase traffic flow model with driver time delays (ATD-model) is based on the following empirical features of phase transitions and congested patterns as well as driver behavioural assumptions of three-phase traffic theory (Sects. 2.3, 2.4, and 8.6 of the book ):
(i) In synchronized flow, a driver accepts a range of different hypothetical steady states with various space gaps $`g`$ at the same vehicle speed $`v`$, i.e., steady states of synchronized flow cover a two-dimensional region in the flow–density plane.
(ii) To avoid collisions, in the steady states a driver does not accept the vehicle speed that is higher than some safe speed (denoted by $`v_\mathrm{s}(g,v_{\mathrm{}})`$) that depends on the speed of the preceding vehicle $`v_{\mathrm{}}`$. In contrast with earlier models in which a safe speed determines a multitude of steady states on the fundamental diagram , in the ATD-model the safe speed determines the upper boundary of the two-dimensional region for the steady states in the flow–density plane .
(iii) If a driver cannot pass the preceding vehicle, then the driver tends to adjust the speed to the preceding vehicle within a synchronization gap $`G(v,v_{\mathrm{}})`$, i.e., at
$$gG(v,v_{\mathrm{}})$$
(4)
a speed adaptation effect occurs. The synchronization gap determines the lower boundary of the two-dimensional region for the steady states in the flow–density plane. In the ATD-model, the speed adaptation effect is modelled through a driver acceleration $`K(v,v_{\mathrm{}})(vv_{\mathrm{}})`$ adjusting the speed to the preceding vehicle under the conditions (4); $`K(v,v_{\mathrm{}})`$ is a sensitivity.
(iv) In traffic flow with greater space gaps, a driver searches for the opportunity to accelerate and to pass. This leads to driver over-acceleration, which is modelled through a driver acceleration $`A(V^{(\mathrm{free})}(g)v)`$ adjusting the vehicle speed at
$$g>G(v,v_{\mathrm{}})$$
(5)
to a gap-dependent optimal speed in free flow $`V^{(\mathrm{free})}(g)`$, where $`A`$ is a sensitivity of this effect.
A competition between the speed adaptation effect and driver over-acceleration simulates a first-order F$``$S transition leading to the onset of congestion in real traffic flow (see explanations in Sect. 2.4 in ).
(v) In empirical observations, due to an F$``$S transition there is a maximum point of free flow associated with the maximum density $`\rho _{\mathrm{max}}^{(\mathrm{free})}`$, maximum flow rate $`q_{\mathrm{max}}^{(\mathrm{free})}`$, and maximum speed $`v_{\mathrm{min}}^{(\mathrm{free})}`$ given by the formula $`v_{\mathrm{min}}^{(\mathrm{free})}=q_{\mathrm{max}}^{(\mathrm{free})}/\rho _{\mathrm{max}}^{(\mathrm{free})}`$ (Sect. 2.3 in ). This maximum point is modelled through F$``$S transition, which occurs already due to infinitesimal local perturbations in steady states of free flow associated with the optimal speed in free flow $`V^{(\mathrm{free})}(g)`$ at the density $`\rho _{\mathrm{max}}^{(\mathrm{free})}`$.
(vi) In high density flow, a driver decelerates stronger than it is required to avoid collisions if the preceding vehicle begins to decelerate unexpectedly (driver over-deceleration). In the ATD-model, the over-deceleration effect, which explains and simulates moving jam emergence in synchronized flow, is modelled by a driver time delay $`\tau `$ in reduction of a current driver deceleration (denoted by $`\tau =\tau _1^{(\mathrm{dec})}(v)`$). The longer $`\tau _1^{(\mathrm{dec})}`$, the stronger the over-deceleration effect. In empirical observations, the lower the synchronized flow speed, the greater the probability for moving jams emergence (Sect. 2.4 in ). For this reason, $`\tau _1^{(\mathrm{dec})}(v)`$ is chosen to be longer at lower speeds than at higher ones.
(vii) At the downstream front of a wide moving jam or a synchronized flow region, a driver within the jam or the synchronized flow region does not accelerate before the preceding vehicle has begun to accelerate. In the ATD-model, this effect is modelled through the use of a mean driver time delay in acceleration at the downstream front of the synchronized flow region, which depends on a time delay in driver acceleration (denoted by $`\tau =\tau _0^{(\mathrm{acc})}`$) and on the sensitivity $`K(v,v_{\mathrm{}})`$ at $`v<v_{\mathrm{}}`$. At the downstream front of a wide moving jam, a mean time delay in acceleration from a standstill $`v=0`$ within the jam should be longer than the mean driver time delay in synchronized flow . To simulate this longer mean time delay in vehicle acceleration, in addition with two mentioned above model effects, a vehicle within the jam does not accelerate before the condition
$$gg_{\mathrm{max}}^{(\mathrm{jam})}$$
(6)
is satisfied, in which $`g_{\mathrm{max}}^{(\mathrm{jam})}`$ is the maximum space gap within the wide moving jam phase.
(viii) Moving in synchronized flow of lower speeds, a driver comes closer to the preceding vehicle than the synchronization gap $`G`$. In empirical observations, this self-compression of synchronized flow is called the pinch effect (Sect. 12.2 in ). In the ATD-model, the pinch effect is simulated through the use of two model assumptions. Firstly, a time delay in reduction of a current driver acceleration (denoted by $`\tau =\tau _1^{(\mathrm{acc})}`$) increases if the speed decreases. Secondly, the sensitivity $`K(v,v_{\mathrm{}})`$, which describes the speed adaptation effect (item (iii)), is chosen at $`vv_{\mathrm{}}`$ different from $`K(v,v_{\mathrm{}})`$ at $`v<v_{\mathrm{}}`$ (item (vii)). Specifically, $`K(v,v_{\mathrm{}})`$ at $`vv_{\mathrm{}}`$ is chosen to be smaller at low speeds than at higher ones. As a result, at lower speeds vehicles choose smaller space gaps than the synchronization gap $`G`$.
(ix) At the upstream front of a wide moving jam or a synchronized flow region, a driver begins to decelerate after a time delay denoted by $`\tau =\tau _0^{(\mathrm{dec})}`$. This delay time should describe realistic velocities of deceleration fronts in congested traffic patterns.
### 2.2 Main Equations
An ATD-model reads as follows:
$`{\displaystyle \frac{dx}{dt}}=v,`$ (7)
$`{\displaystyle \frac{dv}{dt}}=a,`$ (8)
$`{\displaystyle \frac{da}{dt}}=\{\begin{array}{cc}(a^{(\mathrm{free})}a)/\tau \hfill & \text{at }g>G\text{ and }g>g_{\mathrm{max}}^{(\mathrm{jam})},\hfill \\ (a^{(\mathrm{syn})}a)/\tau \hfill & \text{at }gG\text{ and }g>g_{\mathrm{max}}^{(\mathrm{jam})},\hfill \\ (a^{(\mathrm{jam})}a)/\tau \hfill & \text{at }0gg_{\mathrm{max}}^{(\mathrm{jam})},\hfill \end{array}`$ (12)
where $`x`$ is the vehicle space co-ordinate; $`g=x_{\mathrm{}}xd`$; the lower index $`\mathrm{}`$ marks variables related to the preceding vehicle; all vehicles have the same length $`d`$, which includes the minimum space gap between vehicles within a wide moving jam; $`a^{(\mathrm{free})}`$, $`a^{(\mathrm{syn})}`$, and $`a^{(\mathrm{jam})}`$ are vehicle accelerations (deceleration) in the free flow, synchronized flow, and wide moving jam phases, respectively. If the condition (5) is satisfied, then a vehicle moves in accordance with the rules for free flow. Within synchronized flow associated with the condition $`g_{\mathrm{max}}^{(\mathrm{jam})}<gG(v,v_{\mathrm{}})`$, the vehicle tends to adapt the speed to the preceding vehicle. Within a wide moving jam, the space gap is small, specifically $`gg_{\mathrm{max}}^{(\mathrm{jam})}`$, and the vehicle decelerates.<sup>2</sup><sup>2</sup>2Since the vehicle speed $`v`$ cannot be negative, the following condition is also used for Eqs. (8), (12): $`a(t)0\text{at }v(t)=0.`$ (13) To satisfy this condition in numerical simulation, the acceleration $`a(t)`$ is replaced by the value $`\mathrm{max}(a(t),0)`$ if $`v(t)=0`$ at time $`t`$.
### 2.3 Driver Acceleration
The accelerations (decelerations) $`a^{(\mathrm{free})}`$, $`a^{(\mathrm{syn})}`$, and $`a^{(\mathrm{jam})}`$ are found from the condition
$`a^{(\mathrm{phase})}=\mathrm{min}(\mathrm{max}(\stackrel{~}{a}^{(\mathrm{phase})},a_{\mathrm{min}}),a_{\mathrm{max}},a_\mathrm{s}),`$ (14)
the superscript $`\mathrm{`}\mathrm{`}`$phase” in (14) means either $`\mathrm{`}\mathrm{`}`$free”, or $`\mathrm{`}\mathrm{`}`$syn”, or else $`\mathrm{`}\mathrm{`}`$jam” for the related traffic phase; $`a_\mathrm{s}`$ is a deceleration related to safety requirements; $`a_{\mathrm{min}}`$ and $`a_{\mathrm{max}}`$ ($`a_{\mathrm{min}}<0`$, $`a_{\mathrm{max}}0`$) are respectively the minimum and maximum accelerations for cases in which there are no safety restrictions. In (14), functions $`\stackrel{~}{a}^{(\mathrm{free})}`$, $`\stackrel{~}{a}^{(\mathrm{syn})}`$, and $`\stackrel{~}{a}^{(\mathrm{jam})}`$ associated with driver acceleration within the related traffic phase – free flow, or synchronized flow, or else wide moving jam – are determined as follows:<sup>3</sup><sup>3</sup>3In the article, large enough flow rates on the main road are considered at which congested patterns can occur at a bottleneck. For this reason, in (2.3) $`K`$ is chosen to be independent on $`g`$ in the free flow phase. At considerably smaller flow rates in free flow, specifically, if $`g`$ increases, $`K`$ in (2.3) should tend towards zero when $`gG`$.
$`\stackrel{~}{a}^{(\mathrm{free})}(g,v,v_{\mathrm{}})=A(V^{(\mathrm{free})}(g)v)+`$
$`K(v,v_{\mathrm{}})(v_{\mathrm{}}v),`$ (15)
$`\stackrel{~}{a}^{(\mathrm{syn})}(g,v,v_{\mathrm{}})=A\mathrm{min}(V_{\mathrm{max}}^{(\mathrm{syn})}(g)v,0)+`$
$`K(v,v_{\mathrm{}})(v_{\mathrm{}}v),`$ (16)
$$\stackrel{~}{a}^{(\mathrm{jam})}(v)=K^{(\mathrm{jam})}v.$$
(17)
Here $`V_{\mathrm{max}}^{(\mathrm{syn})}(g)`$ is a gap-dependent maximum vehicle speed in synchronized flow; $`K^{(\mathrm{jam})}`$ is a sensitivity.
### 2.4 Safety Conditions
Safety deceleration with a deceleration $`a_\mathrm{s}`$ can be applied, if the vehicle speed becomes higher than the safe speed $`v_\mathrm{s}(g,v_{\mathrm{}})`$. We use safety deceleration found from the condition:
$`a_\mathrm{s}(g,v,v_{\mathrm{}})=A_\mathrm{s}(v_\mathrm{s}(g,v_{\mathrm{}})v),`$ (18)
where $`A_\mathrm{s}`$ is the sensitivity related to safety requirements.
The speed $`v_\mathrm{s}(g,v_{\mathrm{}})`$ in (18) is found based on the safety condition of Gipps :
$`v_\mathrm{s}T_\mathrm{s}+v_\mathrm{s}^2/(2b_\mathrm{s})g+v_{\mathrm{}}^2/(2b_\mathrm{s}),`$ (19)
where $`T_\mathrm{s}`$ is a safety time gap, $`b_\mathrm{s}`$ is a constant deceleration.<sup>4</sup><sup>4</sup>4Note that Eqs. (12) of the ATD-model can also be written without the term $`a^{(\mathrm{jam})}`$ as follows $`{\displaystyle \frac{da}{dt}}=\{\begin{array}{cc}(a^{(\mathrm{free})}a)/\tau \hfill & \text{at }g>G,\hfill \\ (a^{(\mathrm{syn})}a)/\tau \hfill & \text{at }gG.\hfill \end{array}`$ (22) In (22), the speed $`v_\mathrm{s}(g,v_{\mathrm{}})`$ in (18) is found based on the Gipps-condition (19) when $`gg_{\mathrm{max}}^{(\mathrm{jam})}`$ and the speed $`v_\mathrm{s}(g,v_{\mathrm{}})=0`$ when $`g<g_{\mathrm{max}}^{(\mathrm{jam})}`$. In the latter case, the formula (18) with $`v_\mathrm{s}(g,v_{\mathrm{}})=0`$ plays the role of vehicle deceleration within the wide moving jam phase. We use an approximated formula for $`v_\mathrm{s}(g,v_{\mathrm{}})`$ derived from (19) in A, which enables us to write $`a_\mathrm{s}(g,v,v_{\mathrm{}})`$ (18) as follows
$`a_\mathrm{s}(g,v,v_{\mathrm{}})=A_\mathrm{s}^{(\mathrm{g})}(v_{\mathrm{}})(g/T_\mathrm{s}v)+K_\mathrm{s}(v_{\mathrm{}})(v_{\mathrm{}}v),`$ (23)
where
$`A_\mathrm{s}^{(\mathrm{g})}(v_{\mathrm{}})=A_\mathrm{s}T_\mathrm{s}(T_\mathrm{s}+v_{\mathrm{}}/(2b_\mathrm{s}))^1,`$ (24)
$`K_\mathrm{s}(v_{\mathrm{}})=A_\mathrm{s}(T_0+v_{\mathrm{}}/(2b_\mathrm{s}))(T_\mathrm{s}+v_{\mathrm{}}/(2b_\mathrm{s}))^1,`$ (25)
$`T_0`$ is a constant.
### 2.5 Physics of Driver Time Delays
In Eqs. (12), the time delay $`\tau `$ is chosen as
$`\tau =\{\begin{array}{cc}\tau _\mathrm{s}\hfill & \text{at }a_\mathrm{s}<\mathrm{min}(0,\mathrm{max}(\stackrel{~}{a}^{(\mathrm{phase})},a_{\mathrm{min}}),a),\hfill \\ \stackrel{~}{\tau }\hfill & \text{otherwise}.\hfill \end{array}`$ (28)
Here, $`\tau _\mathrm{s}`$ is a short driver time delay associated with a finite driver reaction time that must be taken into account in the cases when the driver should decelerate unexpectedly to avoid collisions; $`\stackrel{~}{\tau }`$ is a time delay in other traffic situations, which is chosen different depending on whether the vehicle accelerates or decelerates:
$`\stackrel{~}{\tau }=\{\begin{array}{cc}\tau ^{(\mathrm{acc})}\hfill & \text{at }a>0,\hfill \\ \tau ^{(\mathrm{dec})}\hfill & \text{at }a0.\hfill \end{array}`$ (31)
In turn, $`\tau ^{(\mathrm{acc})}`$ and $`\tau ^{(\mathrm{dec})}`$ in (31) depend on the acceleration $`a`$:
$`\tau ^{(\mathrm{acc})}=\{\begin{array}{cc}\tau _0^{(\mathrm{acc})}\hfill & \text{at }a<a^{(\mathrm{phase})},\hfill \\ \tau _1^{(\mathrm{acc})}\hfill & \text{ otherwise},\hfill \end{array}`$ (34)
$`\tau ^{(\mathrm{dec})}=\{\begin{array}{cc}\tau _0^{(\mathrm{dec})}\hfill & \text{at }aa^{(\mathrm{phase})},\hfill \\ \tau _1^{(\mathrm{dec})}\hfill & \text{otherwise}.\hfill \end{array}`$ (37)
The driver time delays $`\tau _0^{(\mathrm{acc})}`$, $`\tau _0^{(\mathrm{dec})}`$, $`\tau _1^{(\mathrm{dec})}`$, and $`\tau _1^{(\mathrm{acc})}`$ in (34), (37) are associated with human expectation of local driving conditions, in particular, with spatial and temporal anticipation of a driver in accordance with local adaptation to those traffic situations in which the driver takes into account both the current and expected future behaviour of many vehicles ahead (see also Sect. 2.1).
$`\tau _0^{(\mathrm{acc})}`$ is the mean time delay when a driver starts to accelerate or wants to increase the acceleration. This can often occur at the downstream front of a wide moving jam or a synchronized flow region, i.e., when the speed in traffic flow downstream of the vehicle is higher than the current vehicle speed. In these cases, after the preceding vehicle has begun to accelerate, the driver also begins to accelerate, however, after a time delay to have a desired time gap to the preceding vehicle.
$`\tau _0^{(\mathrm{dec})}`$ is the mean time delay when the driver starts to decelerate or wants to decelerate harder in cases in which the driver approaches a region of a lower speed downstream.
$`\tau _1^{(\mathrm{acc})}`$ corresponds to situations in which the driver accelerates currently but wants either to stop the acceleration or to reduce it. Thus, $`\tau _1^{(\mathrm{acc})}`$ is the mean driver time delay in interruption or reduction of driver acceleration in cases in which the driver recognizes that current acceleration is greater than a desired acceleration in the current driving situation.
$`\tau _1^{(\mathrm{dec})}`$ corresponds to situations in which the driver decelerates currently but wants either to stop the deceleration or to reduce it. Thus, $`\tau _1^{(\mathrm{dec})}`$ is the mean time delay in interruption or reduction of driver deceleration in cases in which the driver recognizes that current deceleration is more negative than a desired deceleration in the current driving situation.
### 2.6 Model of Road with On-Ramp Bottleneck
$`\mathrm{`}\mathrm{`}`$Open” boundary conditions are applied on the main road of the length $`L_0`$. At the beginning of the road free flow conditions are generated for each vehicle one after another at equal time intervals $`\tau _{\mathrm{in}}=1/q_{\mathrm{in}}`$ where $`q_{\mathrm{in}}`$ is the flow rate in the incoming boundary flow. To satisfy safety conditions, a new vehicle appears only if the distance from the beginning of the road ($`x=x_\mathrm{b}`$) to the position $`x=x_{\mathrm{}}`$ of the farthest upstream vehicle in the lane exceeds the distance $`v_{\mathrm{}}T_\mathrm{s}+d`$. The speed $`v`$ and coordinate $`x`$ of a new vehicle are $`v=v_{\mathrm{}}`$ and $`x=x_\mathrm{b}`$, respectively. After a vehicle has reached the end of the road, it is removed; before this, the farthest downstream vehicle maintains its speed. In the initial state ($`t=0`$), all vehicles have the same initial speed $`v=V^{(\mathrm{free})}(g)`$ and space gap $`g`$, and $`q_{\mathrm{in}}=v/(g+d)`$.
An on-ramp bottleneck on the main road is considered. The on-ramp consists of two parts: (i) The merging region of the length $`L_\mathrm{m}`$ that begins at $`x=x_{\mathrm{on}}`$. Within this region, vehicles can merge onto the main road from the on-ramp. (ii) The part of the on-ramp lane of length $`L_\mathrm{r}`$ upstream of the merging region at which vehicles move according to the model equations for a homogeneous road with the maximum speed $`v_{\mathrm{free},\mathrm{on}}=90`$ km/h. At the beginning of the on-ramp lane the flow rate to the on-ramp $`q_{\mathrm{on}}`$ is given as the flow rate on the main road $`q_{\mathrm{in}}`$.
The following rules are applied for vehicle merging within the merging region. A speed $`\widehat{v}`$ is calculated corresponding to formula
$$\widehat{v}=\mathrm{min}(v^+,v+\mathrm{\Delta }v_\mathrm{r}^{(1)})$$
(38)
and then it is used in the merging rules
$$g^+>g_{\mathrm{min}}^{(\mathrm{on})}+\gamma \widehat{v}T_\mathrm{s},g^{}>g_{\mathrm{min}}^{(\mathrm{on})}+\gamma v^{}T_\mathrm{s}.$$
(39)
Here superscripts $`g^+`$ and $`g^{}`$ are space gaps to the preceding vehicle and the trailing vehicle on the main road, respectively; $`v^+`$ and $`v^{}`$ are speeds of the preceding vehicle and the trailing vehicle, respectively; $`\gamma `$, $`g_{\mathrm{min}}^{(\mathrm{on})}`$ and $`\mathrm{\Delta }v_\mathrm{r}^{(1)}`$ are constants, where $`g_{\mathrm{min}}^{(\mathrm{on})}`$ is the minimum gap at which vehicle merging is possible, $`\mathrm{\Delta }v_\mathrm{r}^{(1)}`$ describes the maximum possible increase in speed after vehicle merging. Note that the finite increase $`\mathrm{\Delta }v_\mathrm{r}^{(1)}`$ in the vehicle speed (38) is used to simulate a complex driver behaviour during merging onto the main road, especially in synchronized flow: In some cases, before merging the driver has to accelerate abruptly, to adjust the speed to the speed of the preceding vehicle.
If the conditions (39) are satisfied, then the vehicle merges onto the main road. After merging the vehicle speed $`v`$ is set to $`\widehat{v}`$ (38) and the vehicle coordinate does not change. If the conditions (39) are not satisfied, the vehicle does not merge onto the main road. In this case, the vehicle moves in the on-ramp lane until it comes to a stop at the end of the merging region.
### 2.7 Model Functions and Parameters
Model functions and parameters are shown in Tables 1 and 2, respectively. As explained in Sect. 2.1, driver time delays $`\tau _1^{(\mathrm{dec})}`$ and $`\tau _1^{(\mathrm{acc})}`$ are chosen to be functions of the vehicle speed; additionally, the synchronization gap $`G(v,v_{\mathrm{}})`$ and sensitivity $`K(v,v_{\mathrm{}})`$ are chosen to be asymmetric speed functions depending on whether the vehicle speed $`v`$ is higher or lower than the speed $`v_{\mathrm{}}`$. Explanations of the function $`K(v,v_{\mathrm{}})`$ have been made in item (vii) and (viii) of Sect. 2.1.
Speed dependence and an asymmetric function for the synchronization gap $`G(v,v_{\mathrm{}})`$ are explained by driver behaviour as follows. The synchronization gap is the space gap at which a driver adapts its speed to the speed of the preceding vehicle. Firstly, the synchronization gap is an increasing function of speed: The lower the speed, the smaller the maximum gap at which the driver can comfortably move in synchronized flow. Secondly, if $`v<v_{\mathrm{}}`$, the driver accelerates and he/she can start speed adaptation at a smaller space gap than in the opposite case $`v>v_{\mathrm{}}`$. The function $`T^{(\mathrm{syn})}(v)`$ is used to have a difference in vehicle space gap in steady states of free flow and synchronized flow at a given flow rate. This space gap difference, which is used for simulation of a first-order F$``$S transition, tends towards zero when the density in free flow approaches the maximum point for free flow $`\rho _{\mathrm{max}}^{(\mathrm{free})}`$ (figures 1 (a) and (c)); see also item (v) of Sect. 2.1).
### 2.8 Steady States
In steady states, all vehicles have the same speed $`v=v_{\mathrm{}}`$ and the same space gap $`g`$, and all accelerations and their time derivatives are zero, and the density $`\rho `$ and the flow rate $`q`$ are related to the space gap $`g`$ and to the speed $`v`$ by the obvious conditions
$$\rho =1/(x_{\mathrm{}}x)=1/(g+d),q=\rho v=v/(g+d).$$
(40)
According to (8)–(18) and formulae for $`V^{(\mathrm{free})}(g)`$, $`V_{\mathrm{max}}^{(\mathrm{syn})}(g)`$ (Table 1) for steady states, we get
$$v=V(g)\mathrm{at}g>G(v)\mathrm{and}g>g_{\mathrm{max}}^{(\mathrm{jam})},$$
(41)
$$vV(g)\mathrm{at}gG(v)\mathrm{and}g>g_{\mathrm{max}}^{(\mathrm{jam})},$$
(42)
$$v=0\mathrm{at}gg_{\mathrm{max}}^{(\mathrm{jam})},$$
(43)
$$vv_\mathrm{s}(g,v).$$
(44)
According to (41)–(44), the model steady states consist of the curve $`v=V(g)`$ (41) at $`gg_{\mathrm{min}}^{(\mathrm{free})}`$ (curve $`F`$ in figure 1 (a)) for free flow, a two-dimensional region in the space-gap–speed plane for synchronized flow determined by inequalities in (42), (44), and the line $`v=0`$ at $`gg_{\mathrm{max}}^{(\mathrm{jam})}`$ (43) for wide moving jams (figure 1 (a)). $`g_{\mathrm{min}}^{(\mathrm{free})}`$ is the minimum space gap in free flow found as a solution of the set of the equations $`v=V(g)\mathrm{and}g=G(v,v_{\mathrm{}})`$ at $`v=v_{\mathrm{}}`$.
The two-dimensional region for steady states of synchronized flow is limited by the following boundaries: the boundary $`U`$, the curve $`L`$, the curve $`v=V(g)`$ at $`g<g_{\mathrm{min}}^{(\mathrm{free})}`$, and the horizontal line $`g=g_{\mathrm{max}}^{(\mathrm{jam})}`$. The boundary $`U`$ is associated with the safe speed, i.e., this is determined by the condition (44) when it is an equality. This leads to the condition for the boundary $`U`$
$$g=vT_\mathrm{s}.$$
(45)
The boundary $`L`$ is found from the condition that the vehicle space gap is equal to the synchronization gap
$$g=G(v).$$
(46)
In the flow–density plane, free flow (curve $`F`$ in figure 1 (b)) is found from
$$q=\rho V_\mathrm{F}(\rho )$$
(47)
at $`\rho \rho _{\mathrm{max}}^{(\mathrm{free})}`$ where $`V_\mathrm{F}(\rho )=V(g)_{g=\rho ^1d}`$, $`\rho _{\mathrm{max}}^{(\mathrm{free})}=(g_{\mathrm{min}}^{(\mathrm{free})}+d)^1`$. A wide moving jam is associated with the horizontal line $`q=0`$ at $`\rho _{\mathrm{min}}^{(\mathrm{jam})}\rho \rho _{\mathrm{max}}`$ (figure 1 (b)), where $`\rho _{\mathrm{min}}^{(\mathrm{jam})}=(g_{\mathrm{max}}^{(\mathrm{jam})}+d)^1`$, $`\rho _{\mathrm{max}}=d^1`$. The boundaries of a two-dimensional region for steady states of synchronized flow are: the upper line $`U`$ determined by the condition $`q=(1\rho d)/T_\mathrm{s}`$, the lower curve $`L`$ determined by the condition $`\rho G(q/\rho )=1\rho d`$, the curve (47) at $`\rho >\rho _{\mathrm{max}}^{(\mathrm{free})}`$, and the vertical line $`\rho =\rho _{\mathrm{min}}^{(\mathrm{jam})}`$.<sup>5</sup><sup>5</sup>5We have also studied another version of the ATD-model in which there is a separation of steady states in free flow and synchronized flow in the flow–density plane, i.e., the maximum point for free flow $`\rho _{\mathrm{max}}^{(\mathrm{free})}`$ is related to the intersection point of the line $`U`$ and the curve $`F`$ (figure 1 (d)). Simulations of this version of the ATD-model show qualitatively the same features of phase transitions and congested patterns as those discussed in Sect. 4.
## 3 Speed Adaptation Model
### 3.1 Empirical F$``$S$``$J Transitions as Physical Basis of Speed Adaptation Model
The fundamental hypothesis of three-phase traffic theory, which postulates that hypothetical steady states of synchronized flow cover a two-dimensional region in the flow–density plane, is also one of the basic hypotheses of the ATD-model presented above (figures 1 (a) and (b)). In contrast with the ATD-model, in a speed adaptation model (speed adaptation model, SA-model for short) hypothetical steady states of synchronized flow are associated with a curve (curve $`S`$ in figures 2 (a) and (b)), i.e., they cover a one-dimensional region in the flow–density plane. The curve $`S`$ is associated with an averaging of an infinite number of steady states of synchronized flow to one synchronized flow speed for each vehicle space gap. A gap dependence of the average speed in synchronized flow steady states on the curve $`S`$ is denoted by $`V_{\mathrm{av}}^{(\mathrm{syn})}(g)`$. The basis hypothesis of the SA-model is associated with the sequence of F$``$S$``$J transitions, which determine moving jam emergence in empirical observations .
Note that as in the models and theories in the context of the fundamental diagram approach , in the SA-model steady state model solutions cover a one-dimension region(s) in the flow–density plane. However, in the models and theories reviewed , which claim to show spontaneous moving jam emergence, the F$``$J transition governs the onset of congestion. This is inconsequent with empirical results . In contrast, in the SA-model the onset of congestion is associated with an F$``$S transition, whereas moving jams occur spontaneously only in synchronized flow, in accordance with empirical results.
The SA-model is simpler than the ATD-model. However, due to this simplification the SA-model cannot show some features of congested patterns of the ATD-model (Sect. 5.2), which are observed in empirical observations. The purpose of the SA-model is to simulate an F$``$S transition and features of the sequence of F$``$S$``$J transitions, as observed in empirical observations , in a simple way. This confirms an assumption of three-phase traffic theory that if rather than the fundamental hypothesis the hypothesis about the F$``$S$``$J transitions is the basis of a mathematical model, then the model can show and predict some important empirical features of the phase transitions (see footnote 4 of Sect. 4.3.4 in ).
In the SA-models, an F$``$S transition is modelled through two effects: (i) Discontinuouty of steady speed solutions (figures 2 (a), (c), and (e)) or their instability (curve $`FS`$ in figure 2 (f)) in the vicinity of the maximum point of free flow $`v_{\mathrm{min}}^{(\mathrm{free})}`$, $`\rho _{\mathrm{max}}^{(\mathrm{free})}`$. (ii) The speed adaptation effect is modelled through the term $`K(v,v_{\mathrm{}})(vv_{\mathrm{}})`$ that adjusts the speed to the preceding vehicle in synchronized flow.
Moving jam emergence is simulated through an instability of some of the synchronized flow model steady states associated with the curve $`V_{\mathrm{av}}^{(\mathrm{syn})}(g)`$. This instability occurs in synchronized flow at lower speeds and greater densities (i.e., smaller space gaps). The associated critical density and speed of the synchronized flow steady states are denoted by $`\rho _{\mathrm{cr}}^{(\mathrm{SJ})}`$ and $`v_{\mathrm{cr}}^{(\mathrm{SJ})}`$, respectively (figure 2). To simulate this instability, as in the ATD-model (item (viii) of Sect. 2.1), in the SA-models the sensitivity $`K(v,v_{\mathrm{}})`$ at $`vv_{\mathrm{}}`$ is a decreasing speed function. Similarly with the ATD-model, to simulate the mean time delay in acceleration at the downstream jam front in the SA-model, a vehicle within the jam does not accelerate before (6) is satisfied (item (vii) of Sect. 2.1).
### 3.2 Basic Equations
There can be different possibilities for a separation of steady states of free flow and synchronized flow in SA-models, which all exhibit qualitatively the same features of the F$``$S$``$J transitions. To illustrate this, here we consider two variants of SA-models; in B other possible variants of SA-models are discussed. All these variants of the SA-models exhibit very similar features of phase transitions and spatiotemporal congested traffic patterns that are associated with the same physics of these SA-models.
A formulation for the SA-model reads as follows
$`{\displaystyle \frac{dx}{dt}}=v,`$ (48)
$`{\displaystyle \frac{dv}{dt}}=\{\begin{array}{cc}a^{(\mathrm{free})}\hfill & \text{at }vv_{\mathrm{min}}^{(\mathrm{free})}\text{ and }g>g_{\mathrm{max}}^{(\mathrm{jam})},\hfill \\ a^{(\mathrm{syn})}\hfill & \text{at }v<v_{\mathrm{min}}^{(\mathrm{free})}\text{ and }g>g_{\mathrm{max}}^{(\mathrm{jam})},\hfill \\ a^{(\mathrm{jam})}\hfill & \text{at }0gg_{\mathrm{max}}^{(\mathrm{jam})}.\hfill \end{array}`$ (52)
The vehicle acceleration $`a=dv/dt`$ in (52) is supposed to be limited by the maximum acceleration $`a_{\mathrm{max}}`$, i.e., in (52)
$`a^{(\mathrm{phase})}=\mathrm{min}(\stackrel{~}{a}^{(\mathrm{phase})},a_{\mathrm{max}}).`$ (53)
Here and below the associated designations of functions and parameters have the same meaning as those in the ATD-model (Sect. 2).
### 3.3 Vehicle Acceleration
Functions $`\stackrel{~}{a}^{(\mathrm{free})}(g,v,v_{\mathrm{}})`$, $`\stackrel{~}{a}^{(\mathrm{syn})}(g,v,v_{\mathrm{}})`$, and $`\stackrel{~}{a}^{(\mathrm{jam})}(v)`$ in (53) are chosen as follows
$`\stackrel{~}{a}^{(\mathrm{free})}(g,v,v_{\mathrm{}})=A^{(\mathrm{free})}(V^{(\mathrm{free})}(g)v)+`$
$`K(v,v_{\mathrm{}})(v_{\mathrm{}}v),`$ (54)
$`\stackrel{~}{a}^{(\mathrm{syn})}(g,v,v_{\mathrm{}})=A^{(\mathrm{syn})}\left(V_{\mathrm{av}}^{(\mathrm{syn})}(g)v\right)+`$
$`K(v,v_{\mathrm{}})(v_{\mathrm{}}v),`$ (55)
$$\stackrel{~}{a}^{(\mathrm{jam})}(v)=K^{(\mathrm{jam})}v.$$
(56)
Two versions of functions $`V_{\mathrm{av}}^{(\mathrm{syn})}(g)`$ in (3.3) that lead to two different versions of the SA-models are considered:
$$V_{\mathrm{av}}^{(\mathrm{syn})}(g)=\stackrel{~}{g}(g)/T_{\mathrm{av}}^{(\mathrm{syn})},$$
(57)
and
$$V_{\mathrm{av}}^{(\mathrm{syn})}(g)=V_1\left[\mathrm{tanh}\left(\frac{\stackrel{~}{g}(g)}{T_{\mathrm{av}}^{(\mathrm{syn})}V_1}\right)+c\stackrel{~}{g}(g)\right],$$
(58)
where $`\stackrel{~}{g}(g)=gg_{\mathrm{max}}^{(\mathrm{jam})}`$; $`T_{\mathrm{av}}^{(\mathrm{syn})}`$, $`V_1`$ and $`c`$ are constants.
### 3.4 Steady States and Model Parameters
In the SA-models, in accordance with (52) there are three isolated curves for steady states of the SA-models associated with the three traffic phases: free flow, synchronized flow, and wide moving jam (figures 2 (a) and (b)).
Steady states of free flow are related to a curve $`v=V_\mathrm{F}(\rho )`$ and formula (47) (the curve $`F`$ in figures 2 (a)–(d)) associated with the condition
$$v=V^{(\mathrm{free})}(g)\mathrm{at}vv_{\mathrm{min}}^{(\mathrm{free})}.$$
(59)
Steady states of synchronized flow are related to a curve $`S`$ in the space-gap–speed plane (figures 2 (a) and (c)) given by the condition
$$v=V_{\mathrm{av}}^{(\mathrm{syn})}(g)\mathrm{at}v<v_{\mathrm{min}}^{(\mathrm{free})}\mathrm{and}g>g_{\mathrm{max}}^{(\mathrm{jam})}.$$
(60)
In terms of the flow rate $`q`$ and density $`\rho `$, the formula (60) reads
$$q=\rho V_\mathrm{S}(\rho )\mathrm{at}\rho _{\mathrm{min}}^{(\mathrm{syn})}<\rho <\rho _{\mathrm{min}}^{(\mathrm{jam})},$$
(61)
where $`V_\mathrm{S}(\rho )=V_{\mathrm{av}}^{(\mathrm{syn})}(g)_{g=\rho ^1d}`$, $`\rho _{\mathrm{min}}^{(\mathrm{syn})}=(g_{\mathrm{max}}^{(\mathrm{syn})}+d)^1`$, $`g_{\mathrm{max}}^{(\mathrm{syn})}`$ is found from the equation $`V_{\mathrm{av}}^{(\mathrm{syn})}(g_{\mathrm{max}}^{(\mathrm{syn})})=v_{\mathrm{min}}^{(\mathrm{free})}`$.
In the case of the function $`V_{\mathrm{av}}^{(\mathrm{syn})}(g)`$ given by (57), the formula (61) yields the equation for a curve $`S`$ with a negative slope in the flow–density plane (figure 2 (b))
$$q=(1\rho /\rho _{\mathrm{min}}^{(\mathrm{jam})})/T_{\mathrm{av}}^{(\mathrm{syn})}\mathrm{at}\rho _{\mathrm{min}}^{(\mathrm{syn})}<\rho <\rho _{\mathrm{min}}^{(\mathrm{jam})}.$$
(62)
When the function $`V_{\mathrm{av}}^{(\mathrm{syn})}(g)`$ is given by formula (58), the curve $`S`$ has a maximum in the flow–density plane (figure 2 (d)).
Steady states for a wide moving jam are the same as those in the ATD-model, i.e., they are given by a horizontal line
$$q=0\mathrm{at}\rho _{\mathrm{min}}^{(\mathrm{jam})}\rho \rho _{\mathrm{max}}$$
(63)
in the flow–density plane (figures 2 (b) and (d)).
Parameters of the SA-models are shown in Table 3.
## 4 Diagram of Congested Traffic Patterns at On-Ramp Bottleneck in ATD-Model
Numerical simulations of the ATD-model show that congested patterns (figure 3), which appear on the main road upstream of the bottleneck, are qualitatively the same as those for the stochastic models of Ref. reviewed in the book . However, dynamics of phase transitions leading to congested pattern formation and a diagram of these patterns in the flow–flow plane with co-ordinates are $`q_{\mathrm{on}}`$ and $`q_{\mathrm{in}}`$ (figure 3 (a)) exhibit some important peculiarities in comparison with the stochastic models . These peculiarities are associated with a deterministic character of the ATD-model. To understand this, firstly features of an F$``$S transition at the bottleneck in the deterministic ATD-model should be considered.
### 4.1 Local Perturbation and F$``$S Transition in Free Flow at Bottleneck
Vehicle merging results in a abrupt local space gap reduction on the main road. This can lead to abrupt local vehicle deceleration. For this reason, a dynamic decrease in speed (figure 4 (a)) and the associated increase in density in the on-ramp merging region appear. This local disturbance in the speed and density localized at the bottleneck can be considered a time-dependent dynamic perturbation in free flow. The dynamic nature of this perturbation (there are no random fluctuations in the deterministic ATD-model) is explained by dynamic rules of vehicle motion and by a spatial non-homogeneity localized in the on-ramp merging region within which on-ramp inflow and flow on the main road merge. If the flow rate $`q_{\mathrm{in}}`$ is great enough, then due to dynamic merging rules of Sect. 2.6 vehicles can merge onto the main road at different locations within the merging region. This complex dynamic vehicle merging behaviour causes the associated complex dynamic spatiotemporal dependence of the speed and, respectively, density within the dynamic perturbation (figure 4 (a)).
If the speed and density within the perturbation are averaged over time with an averaging time interval that is considerably longer than time intervals between merging of vehicles, then spatial distributions of the speed and density within the associated average perturbation (figures 4 (b) and (c)) can be considered a $`\mathrm{`}\mathrm{`}`$deterministic” perturbation localized at on-ramp bottleneck. At this time scale the deterministic perturbation is motionless, the total flow rate (across the main road and on-ramp lane) within the perturbation does not depend on spatial co-ordinate. This total flow rate in free flow is $`q_{\mathrm{sum}}=q_{\mathrm{in}}+q_{\mathrm{on}}`$. In contrast, the average speed and density spatially vary in free flow at the bottleneck. In particular, $`q_{\mathrm{sum}}=v_{\mathrm{free}}^{(\mathrm{B})}\rho _{\mathrm{free}}^{(\mathrm{B})}=v^{(\mathrm{free})}\rho ^{(\mathrm{free})}`$, where $`v_{\mathrm{free}}^{(\mathrm{B})}`$ and $`\rho _{\mathrm{free}}^{(\mathrm{B})}`$ are the minimum speed and maximum density within the deterministic perturbation, respectively; $`v^{(\mathrm{free})}`$, $`\rho ^{(\mathrm{free})}`$ are the speed and density downstream of the perturbation, respectively (figures 4 (b) and (c)); $`v_{\mathrm{free}}^{(\mathrm{B})}<v^{(\mathrm{free})}`$, $`\rho _{\mathrm{free}}^{(\mathrm{B})}>\rho ^{(\mathrm{free})}`$.
At a given $`q_{\mathrm{in}}`$, the greater $`q_{\mathrm{on}}`$, the lower the speed $`v_{\mathrm{free}}^{(\mathrm{B})}`$ and the greater the density $`\rho _{\mathrm{free}}^{(\mathrm{B})}`$ within the perturbation, i.e., the greater the amplitude of the deterministic perturbation (figures 4 (b) and (c)). This growth in the perturbation amplitude has a limit associated with an F$``$S transition that occurs spontaneously at the bottleneck when $`q_{\mathrm{on}}`$ gradually increases. The multitude of the flow rates $`q_{\mathrm{in}}`$ and $`q_{\mathrm{on}}`$, at which the F$``$S transition occurs, determines the boundary $`F_\mathrm{S}^{(\mathrm{B})}`$ in the pattern diagram (figure 3 (a)). At the boundary $`F_\mathrm{S}^{(\mathrm{B})}`$ a first-order F$``$S transition (see Sect. 5.1) occurs spontaneously during a chosen time interval $`T_{\mathrm{ob}}`$ that is considerably longer than a time interval $`\tau _{\mathrm{determ}}^{(\mathrm{grow}\mathrm{B})}`$ (about 60 s) required for the average speed to decrease from the speed within a dynamic perturbation in free flow at the bottleneck to a synchronized flow speed (see explanations in Sect. 5.3.7 of ). The necessity of the time interval $`T_{\mathrm{ob}}`$ is associated with a time delay $`T_{\mathrm{FS}}^{(\mathrm{B})}`$ for an F$``$S transition found in the ATD-model: After the time delay $`T_{\mathrm{FS}}^{(\mathrm{B})}`$, a time-dependent (dynamic) perturbation (figure 4 (a)), which can cause a short-time decrease in the speed within the perturbation markedly lower than $`v_{\mathrm{free}}^{(\mathrm{B})}`$, can occur. This perturbation occurrence leads to the F$``$S transition. The boundary $`F_\mathrm{S}^{(\mathrm{B})}`$ is determined from the condition $`T_{\mathrm{ob}}T_{\mathrm{FS}}^{(\mathrm{B})}`$.
In stochastic models , the boundary $`F_\mathrm{S}^{(\mathrm{B})}`$ is also determined by the considition that an F$``$S transition occurs at given $`q_{\mathrm{in}}`$ and $`q_{\mathrm{on}}`$ after a time delay $`T_{\mathrm{FS}}^{(\mathrm{B})}`$ during a chosen time interval $`T_{\mathrm{ob}}`$. However, in the stochastic models $`T_{\mathrm{FS}}^{(\mathrm{B})}`$ is a random value: In different realizations made at the same $`q_{\mathrm{in}}`$ and $`q_{\mathrm{on}}`$ various $`T_{\mathrm{FS}}^{(\mathrm{B})}`$ are found. This stochastic model nature enables us also to calculate the probability for F$``$S transition occurrence .
In contrast with the stochastic models , in the deterministic ATD-model there are no random fluctuations. Time-dependent perturbations in free flow localized at the bottleneck (figure 4 (a)) have dynamic nature explained above. For this reason, in the ATD-model $`T_{\mathrm{FS}}^{(\mathrm{B})}`$ is a fixed value at given $`q_{\mathrm{in}}`$ and $`q_{\mathrm{on}}`$; consequently, the probability for F$``$S transition occurrence cannot be found.
In addition, numerical simulations of the ATD-model show that a duration of a dynamic speed decrease within the perturbation below the speed $`v_{\mathrm{free}}^{(\mathrm{B})}`$ is considerably shorter (1–3 s) than $`\tau _{\mathrm{determ}}^{(\mathrm{grow}\mathrm{B})}`$. As a result, it is found that at a given $`q_{\mathrm{in}}`$ the time delay $`T_{\mathrm{FS}}^{(\mathrm{B})}`$ is a strong decreasing function of $`q_{\mathrm{on}}`$ in a neightborhood of the boundary $`F_\mathrm{S}^{(\mathrm{B})}`$: Already a small increase in $`q_{\mathrm{on}}`$ behind the boundary $`F_\mathrm{S}^{(\mathrm{B})}`$ leads to a decrease in $`T_{\mathrm{FS}}^{(\mathrm{B})}`$ down to $`\tau _{\mathrm{determ}}^{(\mathrm{grow}\mathrm{B})}`$. Thus, we can suggest that in the ATD-model the boundary $`F_\mathrm{S}^{(\mathrm{B})}`$ is very close to the boundary for the deterministic F$``$S transition (see explanation of the deterministic F$``$S transition in Sect. 5.3.7 of Ref. ).
The dynamic character of perturbations at the bottleneck, which is responsible for the above mentioned physics of the boundary $`F_\mathrm{S}^{(\mathrm{B})}`$ for an F$``$S transition in the ATD-model, can clear be seen, if smaller disturbances in speed and density occur due to vehicle merging. Smaller disturbances can be simulated by an increase in the parameter $`\mathrm{\Delta }v_\mathrm{r}^{(1)}`$ of vehicle merging (Sect. 2.6). As a result, at the same $`q_{\mathrm{in}}`$ and $`q_{\mathrm{on}}`$ as those in figures 4 (a)–(c) both time-dependent (figure 4 (d)) and deterministic perturbations (figures 4 (e) and (f)) become smaller. This leads to a shift of the boundary $`F_\mathrm{S}^{(\mathrm{B})}`$ in the diagram of congested patterns to greater $`q_{\mathrm{on}}`$ (curve 2 in figure 4 (g)):
### 4.2 Perculiarities of S$``$J Transitions and Congested Patterns
In the ATD-model, moving jam formation in synchronized flow (S$``$J transition), which occurs at the boundary $`S_\mathrm{J}^{(\mathrm{B})}`$ in the congested pattern diagram (figure 3 (a)), exhibits also some qualitative different features in comparison with the stochastic models .
As in the stochastic models , in the ATD-model after a synchronized flow pattern (SP) occurs upstream of the bottleneck due to F$``$S transition at the bottleneck, a further increase in $`q_{\mathrm{on}}`$ leads to a subsequent decrease in the speed within the SP. This can cause an S$``$J transition with the following general pattern (GP) formation. In the stochastic models, a self-growth of random model fluctuations is mostly responsible for the S$``$J transition. In contrast, in the ATD-model there are no random model fluctuations.
In the ATD-model, dynamic merging of vehicles from the on-ramp lane onto the main road can cause dynamic speed and density waves that propagate upstream in synchronized flow of the SP (figure 5). It turns out that if the flow rate $`q_{\mathrm{on}}`$ is related to a point $`(q_{\mathrm{on}},q_{\mathrm{in}})`$ between the boundaries $`F_\mathrm{S}^{(\mathrm{B})}`$ and $`S_\mathrm{J}^{(\mathrm{B})}`$ (figure 3 (a)), then these dynamic waves decay during their upstream propagation within synchronized flow of the SP (figures 5 (a) and (b)). In contrast, at the boundary $`S_\mathrm{J}^{(\mathrm{B})}`$ the waves begin to self-growth in their amplitude leading wide moving jam formation, i.e., one of GPs appears upstream of the bottleneck (figures 5 (c) and (d)).
As in the KKW cellular automata (CA) model , in the ATD-model the maximum flow rate in free flow downstream of the bottleneck $`q_{\mathrm{max}}^{(\mathrm{free}\mathrm{B})}`$ is a decreasing function of $`q_{\mathrm{on}}`$ (figure 3 (b)). Recall, that the flow rate $`q_{\mathrm{max}}^{(\mathrm{free}\mathrm{B})}(q_{\mathrm{on}})`$ is the flow rate in free flow downstream of the bottleneck associated with the boundary $`F_\mathrm{S}^{(\mathrm{B})}`$. After a congested pattern is formed at the bottleneck, the flow rate downstream of the congested bottleneck called discharge flow rate $`q_{\mathrm{out}}^{(\mathrm{bottle})}`$ (figure 3 (b)) is usually smaller than the initial flow rate $`q_{\mathrm{max}}^{(\mathrm{free}\mathrm{B})}`$. The difference $`\delta q(q_{\mathrm{on}})=q_{\mathrm{max}}^{(\mathrm{free}\mathrm{B})}(q_{\mathrm{on}})q_{\mathrm{out}}^{(\mathrm{bottle})}(q_{\mathrm{on}})`$ called $`\mathrm{`}\mathrm{`}`$capacity drop” is an increasing function of $`q_{\mathrm{on}}`$ at the boundary $`F_\mathrm{S}^{(\mathrm{B})}`$ in the diagram of congested patterns.
In accordance with empirical results , in the ATD-model moving jams do not emerge spontaneously in free flow. This is because in all states of free flow critical perturbations required for an F$``$S transition are considerably smaller than those for F$``$J transition. In the model, all synchronized flow states that are above the line $`J`$ in the flow–density plane (figure 1 (c)) are metastable ones against wide moving jam emergence.
## 5 Phase Transitions and Congested Patterns in SA-Models
### 5.1 Nucleation and Metastability Effects of Pattern Formation
As the ATD-model, the SA-models exhibit a first-order F$``$S transition at the bottleneck, which is accompanied by nucleation and metastability effects, as well as by a hysteresis in SP emergence and dissolution. To illustrate these effects found for both the ATD- and SA-models, we restrict a consideration to the SA-model (48)–(57) (figures 6 and 7). When an initial state at the bottleneck is free flow in which $`q_{\mathrm{in}}`$ is given and $`q_{\mathrm{on}}`$ increases gradually, then, as in the ATD-model (figure 4 (a)), a dynamic disturbance in free flow localized at the bottleneck appears spontaneously. A time averaging of spatial speed and density distributions within the perturbation leads to the associated deterministic perturbation (figures 6 (a) and (b)). Deterministic perturbation features are the same as those for the ATD-model (Sect. 4.1).
The speed $`v_{\mathrm{free}}^{(\mathrm{B})}`$ within the deterministic perturbation decreases when $`q_{\mathrm{on}}`$ increases (from point 1 to 5 in figure 6 (c)). Consequently, $`\rho _{\mathrm{free}}^{(\mathrm{B})}`$ increases. In the flow–density plane, the flow rate on the main road associated with this density increases too (from point 1 to 5 in figure 6 (d)), whereas the flow rate upstream of the perturbation is equal to $`q_{\mathrm{in}}`$, i.e., it does not change (from point 1 to 5 in figure 6 (e)). This increase in the deterministic perturbation amplitude with $`q_{\mathrm{on}}`$ has a limit $`q_{\mathrm{on}}=q_{\mathrm{on}}^{(\mathrm{determ},\mathrm{FS})}`$ associated with the deterministic F$``$S transition (dotted down-arrow in figure 6 (c)). However, non-homogeneous free flow dynamics (Sect. 4.1), which is caused by vehicle merging, results in an F$``$S transition at a smaller $`q_{\mathrm{on}}`$ (point 5 in figure 6 (c)) related to a point on the boundary $`F_S^{(\mathrm{B})}`$ in the diagram of congested patterns (figure 8 (a)). The speed decreases and density increases abruptly within the initial perturbation (arrows F$``$S from point 5 to $`5^{}`$ in figures 6 (c)–(e)) and a congested pattern emerges at the bottleneck. In the example, a widening SP (WSP) occurs upstream of the bottleneck due to the F$``$S transition (figure 8 (c)).
If now $`q_{\mathrm{on}}`$ decreases, the speed within the WSP increases (from point $`5^{}`$ to $`3^{}`$ in figures 6 (c)–(e)). This synchronized flow speed increase has a limit: The speed increases and density decreases abruptly within the synchronized flow (arrows S$``$F from the point $`3^{}`$ to 3 in figures 6 (c)–(e)) and free flow returns at the bottleneck.
Note that $`q_{\mathrm{in}}`$ in figure 6 is chosen to be greater than the threshold flow rate $`q_{\mathrm{th}}`$ for moving SP (MSP) existence. As a result, the initial motionless downstream front of synchronized flow at the bottleneck begins to move away upstream. Consequently, an MSP emerges (figure 8 (d)) (range of $`q_{\mathrm{on}}`$ within which MSPs occur is shown by a dashed part of the synchronized flow states $`v_{\mathrm{syn}}^{(\mathrm{B})}`$ in figure 6 (c)). At greater $`q_{\mathrm{on}}`$ on the dashed part of the synchronized flow states $`v_{\mathrm{syn}}^{(\mathrm{B})}`$ in figure 6 (c) this free flow at the bottleneck can persist for a short time only: A new F$``$S transition occurs spontaneously and a new MSP emerges at the bottleneck, and so on. Due to this effect, a sequence of MSPs appears.
At $`q_{\mathrm{in}}>q_{\mathrm{th}}`$ an MSP can also be induced by application a short-time local perturbation in free flow. The speed within this external perturbation should be lower than the critical speed $`v_{\mathrm{cr},\mathrm{FS}}^{(\mathrm{B})}`$ associated with the critical branch on the Z-characteristic for the F$``$S and reverse S$``$F transitions at the bottleneck. As in the stochastic models , this Z-characteristic consists of the states for free flow associated with the deterministic perturbation at the bottleneck $`v_{\mathrm{free}}^{(\mathrm{B})}`$, the critical branch $`v_{\mathrm{cr},\mathrm{FS}}^{(\mathrm{B})}`$, and synchronized flow states $`v_{\mathrm{syn}}^{(\mathrm{B})}`$ (figure 6 (c)). In accordance with this Z-characteristic, we get the associated hysteresis effects on the fundamental diagram (arrows F$``$S and S$``$F in figures 6 (d) and (e)).
If in contrast $`q_{\mathrm{on}}`$ increases, the speed within the WSP decreases (figure 6 (c)). This speed decrease has a limit associated with the flow rate $`q_{\mathrm{on}}=q_{\mathrm{on}}^{(\mathrm{cr},\mathrm{SJ})}`$ at which an S$``$J transition must occur (dotted down-arrow S$``$J in figure 6 (c)). However, because there are speed and density waves of a finite amplitude in synchronized flow, an S$``$J transition occurs already for $`q_{\mathrm{on}}<q_{\mathrm{on}}^{(\mathrm{cr},\mathrm{SJ})}`$ (point 6 and solid down-arrow in figure 6 (c)). As a result, an GP emerges. This is because in the SA-models, synchronized flow steady states with the speed $`v>v_{\mathrm{cr}}^{(\mathrm{SJ})}`$, which are above the line $`J`$ in the flow–density plane, are metastable ones against wide moving jam emergence. This metastability can be seen from another Z-characteristic in the speed–flow plane associated with an S$``$J transition in synchronized flow. The Z-characteristic consists of the states for synchronized flow $`v_{\mathrm{syn}}^{(\mathrm{B})}`$, the critical branch for critical perturbations in synchronized flow $`v_{\mathrm{cr},\mathrm{SJ}}^{(\mathrm{B})}`$, and the line $`v=0`$ for wide moving jams (figure 6 (c)).
From the resulting double Z-characteristic (figure 6 (c)), it can concluded that in a metastable free flow at the bottleneck (left of the boundary $`F_\mathrm{S}^{(\mathrm{B})}`$ in the diagram in figure 8 (a)) depending on amplitude of a time-limited perturbation caused, for example, by an increase in $`q_{\mathrm{on}}`$ (curves 1 and 2 in figure 7 (a)), either an WSP (figure 7 (b)) or an GP (figure 7 (c)) can be induced. At smaller $`q_{\mathrm{on}}`$ (curve 3 in figure 7 (a)), an MSP (figure 7 (d)) can be excited in free flow. All results presented in figures 6 and 7 for the SA-model remain qualitatively equal for the ATD-model.
### 5.2 Comparison of Congested Patterns in ATD- and SA-Models
The SA-model (52)–(57) (figure 8) exhibits the following shortcoming in comparison with the ATD-model (figure 3):
(i) If the flow rate $`q_{\mathrm{on}}`$ is within a flow rate range $`q_{\mathrm{on}}^{(\mathrm{FSJ})}<q_{\mathrm{on}}<q_{\mathrm{on}}^{(\mathrm{LSP})}`$, then no SP can be formed at the boundary $`F_\mathrm{S}^{(\mathrm{B})}`$ in the diagram (figure 8 (a)): The sequence of F$``$S$``$J transitions occurs spontaneously at this boundary, leading to GP emergence. For this reason, the related part of the boundary at which GPs emerge spontaneously in free flow at the bottleneck is labelled $`F_\mathrm{S}^{(\mathrm{B})}\&S_\mathrm{J}^{(\mathrm{B})}`$.
(ii) If the flow rate $`q_{\mathrm{in}}`$ at this boundary decreases, another characteristic flow rate $`q_{\mathrm{in}}=q_{\mathrm{in}}^{(\mathrm{LSP})}`$ associated with the flow rate $`q_{\mathrm{on}}=q_{\mathrm{on}}^{(\mathrm{LSP})}`$ at this boundary is reached: At $`q_{\mathrm{in}}<q_{\mathrm{in}}^{(\mathrm{LSP})}`$ moving jams do not emerge in synchronized flow upstream of the bottleneck. As a result, at $`q_{\mathrm{in}}<q_{\mathrm{in}}^{(\mathrm{LSP})}`$ and right of the boundary $`F_\mathrm{S}^{(\mathrm{B})}`$ only an LSP remains at the bottleneck. Within this LSP the speed is very low. This LSP has a qualitative different nature in comparison with an LSP of higher synchronized flow speed in the ATD-model that occurs at considerably greater $`q_{\mathrm{in}}`$ (figure 3).
In the SA-model (48)–(56), (58), the branch for average synchronized flow states $`V_{\mathrm{av}}^{(\mathrm{syn})}`$ has a part with a positive slope (figure 2 (d)). Then LSPs of higher speeds appear in the diagram of congested patterns (figure 9). However, these LSPs are not related to LSPs observed in empirical observations. To explain this, note that these model LSPs are very narrow ones (figure 9 (e)). They are localized within the merging region of the on-ramp and consist of two narrow fronts only (figure 10 (a)): There is no region of synchronized flow between the fronts within these LSPs. This is regardless of the flow rates $`q_{\mathrm{in}}`$ and $`q_{\mathrm{on}}`$. Conflictingly, in empirical observations rather than such narrow LSPs, an extended region of synchronized flow is usually observed within an empirical LSP. The LSP width (in the longitudinal direction) changes over time considerably. These empirical features of LSPs shown by the ATD-model (figure 10 (b)) are not found in the SA-model.
(iii) In the ATD-model (figures 10 (e) and (f)) as in empirical observations, both free and synchronized flows can be formed between wide moving jams within an GP. In contrast, in the SA-models only free flow can be formed between wide moving jams within the GP (figures 10 (c) and (d)). The reason for this is as follows. The average branch for synchronized flow lies for speeds $`v>v_{\mathrm{cr}}^{(\mathrm{SJ})}`$ above the line $`J`$ (figure 2 (g)). Flow states in the jam outflow should be related to points on the line $`J`$. Thus, there are no synchronized flow states between the jams. This explains why only free flow can be formed between the jams in the SA-models.<sup>6</sup><sup>6</sup>6The only exclusion is the SA-model with the average branch for synchronized flow (58), if parameters for the curve $`S`$ and/or the line $`J`$, i.e., for wide moving jam propagation are chosen different as those shown in figures 2 (c), (d), and (h): These different parameters should lead to an intersection of the line $`J`$ with the average branch for synchronized flow with a positive slope in the flow–density plane. However, in this specific case only one state of the synchronized flow, which is associated with the point of the latter intersection, is possible. This model effect is not agreed with empirical results, in which the flow rate and speed between wide moving jams within GPs can change over time considerably .
The mentioned shortcoming of the SA-model result from the averaging of a 2D-region of steady states for synchronized flow in the flow–density plane of the ATD-model (figure 1 (a)) to the branch for average synchronized flow states (curve $`S`$ in figure 2).
At chosen SA-model parameters the condition
$$q_{\mathrm{out}}>q^{(\mathrm{pinch})}$$
(64)
is satisfied, where $`q^{(\mathrm{pinch})}`$ is the flow rate within the pinch region of an GP in which narrow moving jams emerge. Under the condition (64), no DGPs appear in the SA-models (figures 8 and 9). At other parameters of the SA-models, an opposite condition
$$q_{\mathrm{out}}<q^{(\mathrm{pinch})}$$
(65)
can be satisfied. Then DGPs appear in the SA-models.
The maximum flow rate in free flow downstream of the bottleneck $`q_{\mathrm{max}}^{(\mathrm{free}\mathrm{B})}(q_{\mathrm{on}})`$, the discharge flow rate $`q_{\mathrm{out}}^{(\mathrm{bottle})}`$, and the $`\mathrm{`}\mathrm{`}`$capacity drop” $`\delta q`$ can sometimes exhibit different features as those in the ATD-model (figure 3 (b)) when the flow rate $`q_{\mathrm{on}}`$ changes (figures 8 (b) and 9 (b)). Particularly, in contrast with the ATD-model, in the SA-model (48)–(57) $`q_{\mathrm{max}}^{(\mathrm{free}\mathrm{B})}`$ does not depend on $`q_{\mathrm{on}}`$, whereas in the SA-model (48)–(56), (58) $`q_{\mathrm{max}}^{(\mathrm{free}\mathrm{B})}`$ depends on $`q_{\mathrm{on}}`$ but at considerably greater $`q_{\mathrm{on}}`$ than for the ATD-model.
Simulations show that the SA models presented in B show qualitatively the same features of the phase transitions and spatiotemporal congested patterns as those in the SA-model (48)–(57).
### 5.3 Comparison with Stochastic SA-Models
It is interesting to compare the deterministic SA-models with possible stochastic SA-models. Such models can be derived from the stochastic model of Ref. , if 2D region of synchronized flow steady states is averaged to synchronized flow states related to a 1D region in the flow–density plane.
A stochastic SA-model can easily be derived from the stochastic model of Ref. based on the physics and ideas for the SA-model approach discussed in Sect. 3. To reach this goal, in the part of the stochastic model of Ref.
$`v_{\mathrm{n}+1}=\mathrm{max}(0,\mathrm{min}(v_{\mathrm{free}},v_{\mathrm{c},\mathrm{n}},v_{\mathrm{s},\mathrm{n}})),`$ (66)
$`x_{\mathrm{n}+1}=x_\mathrm{n}+v_{\mathrm{n}+1}\tau `$ (67)
for a desired speed in synchronized flow $`v_{\mathrm{c},\mathrm{n}}`$, rather than the formula (3) of Ref. leading to a 2D region of synchronized flow steady states in the flow–density plane, the following equations associated with the physics of the SA-models of Sect. 3 are used:
$`v_{\mathrm{c},\mathrm{n}}=v_\mathrm{n}+\mathrm{max}(b_\mathrm{n}\tau ,\mathrm{min}(a_\mathrm{n}\tau ,\mathrm{\Delta }_\mathrm{n})),`$ (68)
$$\mathrm{\Delta }_\mathrm{n}=\{\begin{array}{cc}A^{(\mathrm{free})}(g_\mathrm{n})(v_{\mathrm{free}}v_\mathrm{n})+\hfill & \\ K(v_{\mathrm{},\mathrm{n}}v_\mathrm{n})\hfill & \text{at }gg_{\mathrm{min}}^{(\mathrm{free})}\text{ },\hfill \\ A^{(\mathrm{syn})}(V_{\mathrm{av}}^{(\mathrm{syn})}(g_\mathrm{n})v_\mathrm{n})\hfill & \\ +K(v_{\mathrm{},\mathrm{n}}v_\mathrm{n})\hfill & \text{at }g<g_{\mathrm{min}}^{(\mathrm{free})}.\hfill \end{array}$$
In (66)–(5.3), $`v_\mathrm{n}`$ and $`x_\mathrm{n}`$ are the speed and space co-ordinate of a vehicle; the index $`n`$ corresponds to the discrete time $`t=n\tau `$, $`n=0,1,2,..`$; $`\tau `$ is the time step; $`v_{\mathrm{free}}`$ is the maximum speed in free flow, which is a constant; $`v_{\mathrm{s},\mathrm{n}}`$ is the save speed of Ref. ; $`a_\mathrm{n}0`$ is acceleration, $`b_\mathrm{n}0`$ is deceleration, which are taken as the same stochastic functions used in the model of Ref. ; the space gap $`g_\mathrm{n}=x_{\mathrm{},\mathrm{n}}x_\mathrm{n}d`$; the average speed in synchronized flow steady states $`V_{\mathrm{av}}^{(\mathrm{syn})}`$ is given by the formula (57) at $`g_{\mathrm{max}}^{(\mathrm{jam})}=0`$. Of course, other formulations for the average synchronized flow steady states $`V_{\mathrm{av}}^{(\mathrm{syn})}`$, for example used in the deterministic SA-models (figures 2 (b), (d), and (f)) can also be applied.
In general, descriptions of random vehicle acceleration and deceleration are the same as those in the stochastic model of Ref. : At the first step, the preliminary speed $`\stackrel{~}{v}_{\mathrm{n}+1}`$ is set to $`\stackrel{~}{v}_{\mathrm{n}+1}=v_{\mathrm{n}+1}`$ where the speed $`v_{\mathrm{n}+1}`$ is calculated from the equations (66)–(5.3). At the second step, a noise component $`\xi _\mathrm{n}`$ is added to the calculated speed $`\stackrel{~}{v}_{\mathrm{n}+1}`$. Then the final speed is found from the condition
$$v_{\mathrm{n}+1}=\mathrm{max}(0,\mathrm{min}(v_{\mathrm{free}},\stackrel{~}{v}_{\mathrm{n}+1}+\xi _\mathrm{n},v_\mathrm{n}+a_{\mathrm{max}}\tau ,v_{\mathrm{s},\mathrm{n}})),$$
(69)
where $`a_{\mathrm{max}}`$ is the maximum acceleration.
However, in contrast with the stochastic model of Ref. , the noise component $`\xi _\mathrm{n}`$ in (69) is chosen to be different from zero only if the vehicle decelerates, specifically
$$\xi _\mathrm{n}=\{\begin{array}{cc}b_{\mathrm{max}}\tau \theta (p_\mathrm{b}r)\hfill & \text{if }\stackrel{~}{v}_{\mathrm{n}+1}<v_\mathrm{n}\delta \hfill \\ 0\hfill & \text{otherwise},\hfill \end{array}$$
(70)
where $`r=\mathrm{rand}(0,1)`$, $`\theta (z)=0`$ at $`z<0`$ and $`\theta (z)=1`$ at $`z0`$, $`\delta \tau a_{\mathrm{max}}`$, $`b_{\mathrm{max}}`$, $`p_\mathrm{b}`$ are constants.
Simulations show that the stochastic SA-model (66)–(70) exhibits qualitatively similar spatiotemporal congested patterns at the on-ramp bottleneck (figure 11) as those in the associated deterministic SA-models (figure 8). However, there are qualitative differences in the dynamics of first-order F$``$S and S$``$J transitions leading to pattern formation explained in Sect. 5.1: In the stochastic SA-model, random model fluctuations are important for phase transition nucleation, whereas in the deterministic SA-models the F$``$S and S$``$J transitions are nucleated by dynamic perturbations emerging within the on-ramp merging region.
Note that under the chosen model parameters in the stochastic SA-model (66)–(70) the condition (65) can be satisfied at smaller $`q_{\mathrm{on}}`$. As a result, there is a region in the diagram of congested patterns in which DGPs occur (region labelled $`DGP`$ in figure 11 (a)). After the wide moving jam of the DGP is upstream of the bottleneck as well as in the ATD-model, an LSP remains at the bottleneck (figure 11 (h)). In contrast with the ATD-model, this LSP exists for a finite time interval only and free flow returns at the bottleneck.
However, in a small neighbourhood of the boundary labelled $`D`$ in the diagram, which separates DGPs and GPs, there is a peculiarity in pattern formation under the condition (65): If the flow rate $`q_{\mathrm{on}}`$ increases in the neighbourhood of the boundary $`D`$ in the diagram, then the lifetime of an LSP, which occurs within an DGP increases and it tends towards infinity at the boundary $`D`$ (figure 12 (a)). This quasi-steady LSP is explained by a very long interval between wide moving jam emergence in the synchronized flow at the bottleneck (figure 12 (b)). This interval tends towards the infinity at the boundary $`D`$.
## 6 Discussion
### 6.1 Comparison of ATD- and SA-models with OV-models and other Deterministic Models
The first term in the formula for vehicle acceleration $`\stackrel{~}{a}^{(\mathrm{free})}`$ (2.3), $`A(V^{(\mathrm{free})}(g)v)`$, describes the dynamics of the speed $`v`$ in the vicinity of the optimal speed $`V^{(\mathrm{free})}(g)`$ in free flow. At a time scale that is considerably greater than the time delay $`\tau `$, this dynamic behaviour is the same as those in different OV-models , which can be written as follows
$$\frac{dv}{dt}=A(g,v)(V(g)v).$$
(71)
However, in (71) the vehicle acceleration $`A(g,v)(V(g)v)`$ is valid for the whole possible space gap range
$`g0.`$ (72)
In contrast with the OV-models, in the ATD-model this vehicle acceleration is applied for large space gaps (5) associated with free flow only.
The crucial difference of the ATD-model with the OV-models and all other deterministic microscopic traffic flow models (see references in the reviews ) is that the vehicle acceleration behaviour qualitatively changes when the vehicle is within the synchronization gap, i.e., if the condition (4), which is opposite to the condition (5), is satisfied.
The condition (4) is associated with the synchronized flow phase in which there is no optimal speed in the ATD-model. This conclusion follows from (2.3) and its analysis made in Sect. 2.8 in which it has been shown that for a given steady space gap in synchronized flow there are an infinity of steady vehicle speeds within a finite speed range (figure 1 (a)).
The concept of safe speed $`v_\mathrm{s}(g,v_{\mathrm{}})`$ for vehicle collision prevention used in the ATD-model is qualitatively different from the concept of optimal speed that is the basis of the deterministic approaches (2) and (3): The optimal speed is a desired one (this explains the term $`\mathrm{`}\mathrm{`}`$optimal” speed) for a driver to be reached (the driver moves comfortable with the optimal speed during a long time), whereas the safe speed is not an optimal one but a limiting speed that is still permitted (the driver should not move with this speed during a long time because this is strain for the driver and, therefore, non-comfortable). The qualitative difference of these two concepts is mathematically reflected in the dynamic model behaviour. In the ATD-model, when the vehicle speed is higher than the safe speed and safe deceleration is applied, then a driver time delay is equal to a small driver reaction time: $`\tau =\tau _\mathrm{s}`$ (28). In all other driving situations, which are not associated with safe speed, driver time delays are different from $`\tau _\mathrm{s}`$. This is because these driver time delays are associated mostly with qualitatively different expected events occurring within different traffic phases (Sect. 2.5). As a result, in the ATD-model driver deceleration to the safe speed occurs considerably quicker, then in other driving situations. In contrast, in accordance with the concept of optimal speed, in OV-models there is a driver time delay in deceleration that characterizes speed relaxation to the optimal speed .
The crucial differences between the SA-models and all other traffic flow models in which steady states covering a one-dimensional region(s) in the flow–density plane (see references in the reviews ) are as follows. In contrast with the models of Ref. , in the SA-models at each density of free flow states the critical amplitude of a local perturbation required for an F$``$S transition is considerably smaller than the critical amplitude of a local perturbation required for an F$``$J transition.
In the SA-models, there are two ranges of model steady states separated one from another by a model discontinuity in vehicle space gap or in speed (figures 2 (a)–(e)) or else due to instability of model steady states against infinitesimal non-homogeneous fluctuations (figure 2 (f)) This simulates the hypothesis of three-phase traffic theory about a competition between over-acceleration and speed adaptation effect that is responsible for F$``$S and S$``$F transitions: The first range of steady states simulates free flow, whereas the second simulates synchronized flow. To simulate S$``$J transitions within synchronized flow, steady states associated with synchronized flow of higher speeds are metastable with respect to moving jam emergence, i.e., moving jams emerge in these synchronized flow states only if large enough amplitude local perturbations appear; synchronized flow states of lower speeds are unstable with respect to moving jam emergence. These requirements to the SA-models lead to F$``$S$``$J transitions that are responsible for moving jam emergence found in empirical data .
### 6.2 Critical Discussion of Theories and Models based on the Fundamental Diagram Approach
In the OV model (71), as in other deterministic (and stochastic) traffic flow models in the context of the fundamental diagram approach reviewed in , which claim to show spontaneous jam emergence, there is a range of the density on the fundamental diagram in which steady states on this diagram are unstable against infinitesimal perturbations.<sup>7</sup><sup>7</sup>7It should be noted that some of the models based on the fundamental diagram approach are not valid far from equilibrium. It is not simply a matter of a phase transition type that a model exhibits in steady conditions, but of the difficulty of closing equations, which should work in unsteady conditions by relations valid only in steady uniform conditions. This instability leads to wide moving jam emergence in these models both on homogeneous road and at a bottleneck. We denote the minimum density of this density range, in which infinitesimal fluctuations grow, by $`\rho _{\mathrm{cr}}^{(\mathrm{J})}`$ (figures 13 (a) and (b)).
There are two possibilities for the arrangement of the point of this instability $`(\rho _{\mathrm{cr}}^{(\mathrm{J})},q_{\mathrm{cr}}^{(\mathrm{J})})`$ on the fundamental diagram in the OV model and other models in the context of the fundamental diagram approach:
(i) The point $`(\rho _{\mathrm{cr}}^{(\mathrm{J})},q_{\mathrm{cr}}^{(\mathrm{J})})`$ lies left of the maximum point of the fundamental diagram $`(\rho _0,q_0)`$, i.e., on the branch of the diagram with a positive slope (figure 13 (a)):
$$\rho =\rho _{\mathrm{cr}}^{(\mathrm{J})}<\rho _0$$
(73)
(ii) The point $`(\rho _{\mathrm{cr}}^{(\mathrm{J})},q_{\mathrm{cr}}^{(\mathrm{J})})`$ lies right of the maximum point of the fundamental diagram $`(\rho _0,q_0)`$, i.e., on the branch of the diagram with a negative slope (figure 13 (b)):
$$\rho =\rho _{\mathrm{cr}}^{(\mathrm{J})}>\rho _0.$$
(74)
Note that in both cases (i) and (ii) all states on the fundamental diagram, in which the density satisfies the condition
$$\rho _{\mathrm{min}}\rho <\rho _{\mathrm{cr}}^{(\mathrm{J})},$$
(75)
where $`\rho _{\mathrm{min}}`$ is the density in the wide moving jam outflow associated with the flow rate $`q_{\mathrm{out}}`$, are metastable states with respect to moving jam emergence . The case (i) has intensively been considered in the literature and criticized in Sect. 3.3.2 of the book .
In the case (ii) (figure 13 (b)), the flow rate in free flow downstream of an on-ramp bottleneck $`q_{\mathrm{sum}}`$ cannot exceed the maximum flow rate on the fundamental diagram $`q_0`$. If $`q_{\mathrm{in}}`$ is a given large enough value and the flow rate $`q_{\mathrm{on}}`$ begins to increase, then a localized perturbation as that in the ATD- and SA-models (Sects. 4.1 and 5.1) appears at the bottleneck (figures 14 (a) and (b)). The minimum speed within the time-averaged (deterministic) perturbation on the main road decreases when $`q_{\mathrm{on}}`$ increases (points 1–3 in figure 14 (c)). In the OV-model, when $`q_{\mathrm{on}}`$ increases, the location of the minimum speed within the deterministic perturbation on the main road exhibits firstly a slight shift downstream and then upstream within the merging region of the on-ramp (figure 14 (a, b)). For this reason, the flow rate on the main road at the location of the minimum speed within the deterministic perturbation on the main road firstly slightly increases and then decreases (points 1–3 in figure 14 (d)), whereas the flow rate on the main road upstream of the on-ramp merging region is equal to $`q_{\mathrm{in}}`$ (points 1–3 in figure 14 (e)). When $`q_{\mathrm{on}}`$ increases beginning from zero, the flow rate $`q_{\mathrm{sum}}`$ downstream of the bottleneck increases beginning from $`q_{\mathrm{sum}}=q_{\mathrm{in}}`$ (points 1–3 in figures 14 (f) and (g)).
At a given large enough flow rate $`q_{\mathrm{in}}`$ this growth of the local perturbation in free flow at the bottleneck with $`q_{\mathrm{on}}`$ has a limit. This limit is reached when the flow rate $`q_{\mathrm{on}}`$ reaches some critical value $`q_{\mathrm{on}}=q_{\mathrm{on}}^{(\mathrm{d})}`$ at which the flow rate $`q_{\mathrm{sum}}`$ is equal to the maximum flow rate on the fundamental diagram:
$$q_{\mathrm{sum}}=q_{\mathrm{in}}+q_{\mathrm{on}}^{(\mathrm{d})}=q_0.$$
(76)
When the flow rate $`q_{\mathrm{on}}`$ increases further, i.e.,
$$\mathrm{\Delta }q=q_{\mathrm{in}}+q_{\mathrm{on}}q_0>0,$$
(77)
then the upstream front of the initial perturbation, which is motionless at the condition $`q_{\mathrm{sum}}=q_{\mathrm{in}}+q_{\mathrm{on}}q_0`$ (curve II in figure 14 (h)), begins to move upstream of the bottleneck, i.e., a wave of lower speed and greater density propagating upstream appears (spatial speed distributions related to the times $`t_1`$$`t_4`$ in figure 14 (h)). As a result, a dense flow associated with the branch of the diagram with a negative slope occurs upstream of the bottleneck (figures 13 (c) and (d) and points 4–6 in figures 14 (c)–(e)). At the critical point (76), the derivative of the minimum average speed on the main road on the flow rate $`q_{\mathrm{on}}`$ is discontinuous, whereas this speed is a continuous decreasing function of $`q_{\mathrm{on}}`$ (figure 14 (c)). The greater the flow rate $`q_{\mathrm{on}}`$, specifically, the greater $`\mathrm{\Delta }q`$ (77), the greater absolute velocity of the wave of dense flow propagation $``$$`v_\mathrm{d}`$$``$ (figures 13 (c) and (d)). In addition, the flow rate downstream of the bottleneck, which is equal to $`q_0`$ under the condition (76), remains approximately to be equal to $`q_0`$, when $`q_{\mathrm{on}}`$ increases (points 4–6 in figures 14 (f) and (g)).
It must be noted that the above mentioned behaviour of the upstream front of the perturbation at the bottleneck in the OV model (71), (74) is qualitatively different from those for the upstream front of the perturbation at the bottleneck in the ATD- and SA-models. In the latter case, when the flow rate $`q_{\mathrm{on}}`$ reaches the critical value for an F$``$S transition, a wave of synchronized flow occurs abruptly and propagates upstream with a finite velocity. This is associated with a first-order F$``$S transition. In contrast, in the OV-model there is no discontinuous change in the velocity $`v_\mathrm{d}`$ when due to an increase in $`q_{\mathrm{on}}`$ the condition (77) is satisfied: $``$$`v_\mathrm{d}`$$``$ increases continuously beginning from zero, when $`q_{\mathrm{on}}`$ first reaches and then exceeds the critical flow rate $`q_{\mathrm{on}}^{(\mathrm{d})}`$ associated with the condition (76). Specifically, we find that if $`\mathrm{\Delta }q0`$, then $``$$`v_\mathrm{d}`$$``$$`0`$. Thus, in the OV-model there is no first-order phase transition from free flow to dense flow.
The widening dense flow upstream of the bottleneck (figures 13 (c) and (d) and 14 (h)) can exist only, when the density $`\rho _\mathrm{d}`$ in the dense flow satisfies the condition
$$\rho _0<\rho _\mathrm{d}<\rho _{\mathrm{cr}}^{(\mathrm{J})}.$$
(78)
This is because at the density $`\rho _\mathrm{d}=\rho _{\mathrm{cr}}^{(\mathrm{J})}`$ the dense flow loses its stability against wide moving jam emergence (point 7 and dotted down-arrow in figure 14 (c)). However, dynamic waves that emerge due to vehicle merging at the bottleneck propagate through the dense flow. For this reason, in numerical simulations this moving jam emergence occurs already at the density $`\rho _\mathrm{d}<\rho _{\mathrm{cr}}^{(\mathrm{J})}`$ (point 6 and solid down-arrow in figure 14 (c)).
The congested patterns in figures 13 (d) and (e) at the first glance resemble a widening SP and an GP, respectively. Indeed, in both cases a dense flow occurs upstream of the bottleneck whose downstream front is fixed at the bottleneck. Thus, this dense flow should satisfy the macroscopic spatiotemporal objective criteria for the synchronized flow phase (Sect. 1). This conclusion is, however, incorrect. To explain this, note that in empirical observations application of the objective criteria, which define the traffic phases in congested traffic, leads to clear distinction of the synchronized flow phase. This synchronized flow exhibits the following fundamental empirical feature: An F$``$S transition leading to synchronized flow emergence is a first-order phase transition. In contrast, in a traffic flow model an application of the objective criteria does not guarantee that dense flow occurrence in free flow is associated with a first-order phase transition, which is one of the requirements for the synchronized flow phase.
This conclusion concerns the OV model (71), (74) (figure 13 (b)) as well as other models in the context of the fundamental diagram approach under condition (74). Whereas for the SA-model there is a Z-shaped speed–flow characteristic associated with a first-order F$``$S transition in free flow at the bottleneck (figures 6 (c)–(e)), for the OV model the on-ramp flow rate dependence of the speed at the bottleneck is a monotonous decreasing function (figure 14 (c)): There is no first-order phase transition, when a dense flow related to the fundamental diagram with a negative slope is formed upstream of the bottleneck. Thus, the dense traffic flow in the case of the OV model and other models in the context of the fundamental diagram approach under condition (74) does not exhibit the important empirical feature of synchronized flow and, therefore, the dense flow is not associated with the synchronized flow phase.
There are also traffic flow models in the context of the fundamental diagram approach, in which there is no instability of steady model states on the fundamental diagram regardless of the vehicle density. Examples of this model class are as follows: (i) An OV model (71) in which the sensitivity $`A(g,v)`$ is great enough regardless of $`v`$ and $`g`$. (ii) The Nagel-Schreckenberg cellular automata model in the deterministic model limit, i.e., when probability of model fluctuations in this model is equal zero ($`p=0`$. (iii) The Lighthill-Whitham-Richards model and the associated cell-transmission models . In this model class, traffic patterns at a freeway bottleneck are qualitatively similar as those found in the OV model (71), (74) at the density considerably smaller than the critical density $`\rho _{\mathrm{cr}}^{(\mathrm{J})}`$ (the patterns associated with points 1–5 in figures 14 (c)–(g)). These common model features are as follows: 1) the local perturbation at the bottlenecks at $`\mathrm{\Delta }q<0`$ (figures 14 (a) and (b)); 2) widening dense flow upstream of the bottleneck at $`\mathrm{\Delta }q>0`$ (figures 13(c) and (d) and 14 (h)); 3) there is no discontinuous change in speed (no speed breakdown) at the bottleneck when widening dense flow occurs; 4) with an increase in traffic demand at $`\mathrm{\Delta }q0`$, the upstream front velocity of widening dense flow increases continuously beginning from zero. Thus, in this model class, there is no first-order F$``$S transition observed during the onset of congestion at the bottleneck, i.e., this dense flow has no relation to real freeway traffic.
### 6.3 Conclusions
(i) Two different deterministic microscopic traffic flow model classes in the context of three-phase traffic theory, the ATD- and SA-models, have been introduced in the article.
(ii) The ATD- and SA-models reproduce important empirical spatiotemporal features of phase transitions in traffic flow and congested traffic patterns.
(iii) In contrast with all other known deterministic microscopic traffic flow models, in the ATD- and SA-models vehicles moving in free flow and vehicles moving in synchronized flow exhibit qualitatively different dynamic behaviour. This is a result of the introduction of two separated regions of steady state model solutions for free flow and synchronized flow in the ATD- and SA-models as well as different dynamic rules of vehicle motion in free flow and synchronized flow implemented in the models.
(iv) As in empirical observations, there is a first-order phase transition in the ATD- and SA-models from free flow to synchronized flow that explained the onset of congestion at bottlenecks in these models.
(v) The nature of the onset of congestion as a first-order F$``$S transition in free flow at the bottleneck, which the ATD- and SA-models show, is also associated with metastability of free flow at the bottleneck against external short-time disturbances in this flow in a neighbourhood of the bottleneck. As a result, there is multiple congested pattern emergence in an initial free flow at the bottleneck in the ATD- and SA-models: Depending on an amplitude (or duration) of an external disturbance, one of the SPs or else an GP can be induced in free flow at the bottleneck at the same chosen model parameters.
(vi) In accordance with empirical results, in the ATD- and SA-models moving jams can emerge spontaneously in synchronized flow only, i.e., as a result of F$``$S$``$J transitions.
(vii) In addition to the above common behaviour of the ATD- and SA-models, these models exhibit also some qualitatively different features. This is because in the ATD-model synchronized flow model steady states are related to a 2D-region in the flow–density plane, whereas synchronized flow model steady states in the SA-models belong to an 1D-region (a curve) in the flow–density plane. In particular, the following differences of model features have been found:
(1) The ATD-model can show all types of spatiotemporal congested patterns at an on-ramp bottleneck observed in empirical observations.
(2) In contrast, SA-models cannot show LSPs associated with empirical results as well as some of empirical features of synchronized flow between wide moving jams within GPs.
(viii) Models in the context of the fundamental diagram approach reviewed in cannot explain the onset of congestion in free flow, which in empirical observations is associated with a first-order F$``$S transition. Depending on the model type and model parameters, in these models either wide moving jam emergence is responsible for the onset of congestion at an on-ramp bottleneck rather than an empirically observed F$``$S transition or a widening dense traffic flow occurs upstream of the bottleneck when the density in free flow at the bottleneck exceeds the density associated with the maximum point on the fundamental diagram. In the latter case, in contrast with empirical observations there is no first-order phase transition from an initial free flow to this dense flow at the bottleneck: The dense flow results from non-homogeneity of a freeway in a neighbourhood of the bottleneck. Thus, these models cannot show a first-order F$``$S transition observed during the onset of congestion at the bottleneck in real freeway traffic, i.e., this dense flow has no relation to real freeway traffic. Indeed, the first-order F$``$S transition is a fundamental empirical feature of the onset of congestion in free flow with the subsequent synchronized flow phase emergence at the bottleneck.
## Appendix A
To derive formula (23, let us consider a solution of (19) when it is an equality:
$`v_\mathrm{s}(g,v_{\mathrm{}})={\displaystyle \frac{2b_\mathrm{s}g+v_{\mathrm{}}^2}{b_\mathrm{s}T_\mathrm{s}+\sqrt{b_\mathrm{s}^2T_\mathrm{s}^2+2b_\mathrm{s}g+v_{\mathrm{}}^2}}}.`$ (79)
From (19), (79), it can be seen that if $`g=v_{\mathrm{}}T_\mathrm{s}`$, then the safe speed $`v_\mathrm{s}=v_{\mathrm{}}`$; if in contrast $`g<v_{\mathrm{}}T_\mathrm{s}`$, then the speed $`v_\mathrm{s}<v_{\mathrm{}}`$. In particular, this ensures collision less vehicle motion. To simplify the formula (79), let us replace the space gap $`g`$ in denominator of (79) by the value $`v_{\mathrm{}}T_\mathrm{s}`$. This reduces the safe speed $`v_\mathrm{s}`$ at $`g<v_{\mathrm{}}T_\mathrm{s}`$, therefore, the safety condition (19) remains to be valid. Then from formula (79), we get
$`v_\mathrm{s}(g,v_{\mathrm{}})={\displaystyle \frac{g+v_{\mathrm{}}^2/(2b_\mathrm{s})}{T_\mathrm{s}+v_{\mathrm{}}/(2b_\mathrm{s})}}.`$ (80)
To provide more comfortable vehicle deceleration, an anticipated gap $`g^{(\mathrm{a})}=g+(v_{\mathrm{}}v)T_0`$ is used in formula (80) rather than the gap $`g`$. As a result, (80) takes the form
$`v_\mathrm{s}(g,v_{\mathrm{}})={\displaystyle \frac{g+(v_{\mathrm{}}v)T_0+v_{\mathrm{}}^2/(2b_\mathrm{s})}{T_\mathrm{s}+v_{\mathrm{}}/(2b_\mathrm{s})}}.`$ (81)
Substituting (81) into (18), we find formula (23) with coefficiens (24), (25).
Note that we have also tested another formulation for the safe speed in the ATD-model when the speed $`v_\mathrm{s}(g,v_{\mathrm{}})`$ in (18) is given by formula (79). Simulations of the ATD-model show that both formulations (23) and (79) ensure collision less vehicle motion at an appropriate choice of model parameters and lead to qualitatively the same features of phase transitions and congested patterns.
## Appendix B
In this Appendix, two further variants of the SA-models are presented. In the first of these variants, the formula (52) reads as follows
$`{\displaystyle \frac{dv}{dt}}=\{\begin{array}{cc}a^{(\mathrm{free})}\hfill & \text{at }gg_{\mathrm{min}}^{(\mathrm{free})}\text{ },\hfill \\ a^{(\mathrm{syn})}\hfill & \text{at }g_{\mathrm{max}}^{(\mathrm{jam})}<g<g_{\mathrm{min}}^{(\mathrm{free})},\hfill \\ a^{(\mathrm{jam})}\hfill & \text{at }0gg_{\mathrm{max}}^{(\mathrm{jam})},\hfill \end{array}`$ (85)
where $`a^{(\mathrm{free})}`$, $`a^{(\mathrm{syn})}`$, $`a^{(\mathrm{jam})}`$ are given by (53)–(57). In this SA-model, steady states of free flow (the curve $`F`$ in figure 2 (e)) correspond to the condition (59), averaged steady states of synchronized flow are related to a line $`S`$ given by the condition
$$q=(1\rho /\rho _{\mathrm{min}}^{(\mathrm{jam})})/T_{\mathrm{av}}^{(\mathrm{syn})}\mathrm{at}\rho _{\mathrm{max}}^{(\mathrm{free})}<\rho <\rho _{\mathrm{min}}^{(\mathrm{jam})},$$
(86)
steady states for a wide moving jam are associated with the condition (63) (figure 2 (e)).
In another variant of SA-model, formula (52) reads as follows
$`{\displaystyle \frac{dv}{dt}}=\{\begin{array}{cc}a^{(\mathrm{free})}\hfill & \text{at }gg_{\mathrm{min}}^{(\mathrm{free})},\hfill \\ a^{(\mathrm{FS})}\hfill & \text{at }g_{\mathrm{max}}^{(\mathrm{syn})}<g<g_{\mathrm{min}}^{(\mathrm{free})},\hfill \\ a^{(\mathrm{syn})}\hfill & \text{at }g_{\mathrm{max}}^{(\mathrm{jam})}<gg_{\mathrm{max}}^{(\mathrm{syn})},\hfill \\ a^{(\mathrm{jam})}\hfill & \text{at }0gg_{\mathrm{max}}^{(\mathrm{jam})}.\hfill \end{array}`$ (91)
In (91), $`g_{\mathrm{max}}^{(\mathrm{syn})}`$ is the maximum space gap in synchronized flow; $`a^{(\mathrm{jam})}`$, $`a^{(\mathrm{free})}`$, $`a^{(\mathrm{syn})}`$ are given by (53) in which $`\stackrel{~}{a}^{(\mathrm{jam})}`$ is taken from (56),
$`\stackrel{~}{a}^{(\mathrm{free})}(g,v,v_{\mathrm{}})=A^{(\mathrm{free})}(V^{(\mathrm{free})}(g)v)+`$
$`K^{(\mathrm{free})}(v_{\mathrm{}}v),`$ (92)
$`\stackrel{~}{a}^{(\mathrm{syn})}(g,v,v_{\mathrm{}})=A^{(\mathrm{syn})}\left(V_{\mathrm{av}}^{(\mathrm{syn})}(g)v\right)+`$
$`K^{(\mathrm{syn})}(v,v_{\mathrm{}})(v_{\mathrm{}}v),`$ (93)
where the sensitivity
$`K^{(\mathrm{syn})}(v,v_{\mathrm{}})=\{\begin{array}{cc}K^{(\mathrm{acc})}\hfill & \text{at }v<v_{\mathrm{}},\hfill \\ K^{(\mathrm{dec})}\hfill & \text{at }vv_{\mathrm{}},\hfill \end{array}`$ (96)
$`K^{(\mathrm{free})}`$ is a sensitivity, $`V_{\mathrm{av}}^{(\mathrm{syn})}(g)`$ is given by (57). A function $`a^{(\mathrm{FS})}(g,v,v_{\mathrm{}})`$ in (91) is taken as follows
$`a^{(\mathrm{FS})}(g,v,v_{\mathrm{}})=\mathrm{min}(a_{\mathrm{max}},A^{(\mathrm{FS})}(V^{(\mathrm{FS})}(g)v)+`$
$`K^{(\mathrm{free})}(v_{\mathrm{}}v)),`$ (97)
where the function $`V^{(\mathrm{FS})}(g)=V(g)`$ at $`g_{\mathrm{max}}^{(\mathrm{syn})}<g<g_{\mathrm{min}}^{(\mathrm{free})}`$, $`A^{(\mathrm{FS})}`$ is a sensitivity that in a general case can be different from the sensitivity $`A^{(\mathrm{free})}`$ in free flow.
In the SA-model (91)–(B), steady states of free flow (the curve $`F`$ in figure 2 (f)) are associated with the condition (59), averaged steady states of synchronized flow are related to a line $`S`$ given by the condition
$$q=(1\rho /\rho _{\mathrm{min}}^{(\mathrm{jam})})/T_{\mathrm{av}}^{(\mathrm{syn})}\mathrm{at}\rho _{\mathrm{min}}^{(\mathrm{syn})}\rho <\rho _{\mathrm{min}}^{(\mathrm{jam})},$$
(98)
where $`\rho _{\mathrm{min}}^{(\mathrm{syn})}=1/(g_{\mathrm{max}}^{(\mathrm{syn})}+d)`$, and steady states for a wide moving jam are found from the condition (63) (figure 2 (f)).
In contrast with the other SA-models, the SA-model (91)–(B) has a limited density range of steady states between steady states for free flow and synchronized flow, which are found from the condition
$`a^{(\mathrm{FS})}=0\text{at }g_{\mathrm{max}}^{(\mathrm{syn})}<g<g_{\mathrm{min}}^{(\mathrm{free})}\text{ }.`$ (99)
Eq. (99) yields the following condition for these steady states in the flow–density plane (curve $`FS`$ in figure 2 (f))
$`q=\rho V_\mathrm{F}(\rho )\mathrm{at}\rho _{\mathrm{max}}^{(\mathrm{free})}<\rho <\rho _{\mathrm{min}}^{(\mathrm{syn})},`$ (100)
where the density $`\rho _{\mathrm{max}}^{(\mathrm{free})}`$ at the maximum point for free flow is not greater than the density $`\rho _0`$ associated with the maximum point on the curve $`FS`$ (figure 2 (f)). To simulate a first-order F$``$S transition, the steady state model solutions (100) should be unstable against infinitesimal non-homogeneous perturbations. This requirement to the SA-model (91)–(B) is easy satisfied through an appropriated choice of the function $`V^{(\mathrm{FS})}(g)`$ and the sensitivities $`A^{(\mathrm{FS})}`$, $`K^{(\mathrm{free})}`$, $`K^{(\mathrm{syn})}`$. In this case, numerical simulations of the SA-model (91)–(B) made show that this model exhibits F$``$S$``$J transitions in accordance with empirical results (we used the following parameters for the SA-model (91)–(B): $`V(g)=V_0\left(\mathrm{tanh}((gg_0)/g_1)+\mathrm{tanh}(g_0/g_1)\right)`$ with $`V_0=`$ 14 m/s, $`g_0=`$ 21 m, $`g_1=`$ 7 m; $`g_{\mathrm{max}}^{(\mathrm{syn})}=`$ 24 m; $`A^{(\mathrm{free})}=A^{(\mathrm{FS})}=0.1s^1`$; $`K^{(\mathrm{free})}=0.6s^1`$; $`K^{(\mathrm{acc})}=0.4s^1`$; $`K^{(\mathrm{dec})}`$ is taken from Table 1 with $`K_1^{(\mathrm{dec})}=1s^1`$, $`v_\mathrm{c}=`$ 9 m/s, $`ϵ=`$ 0.05; other parameters are the same as those in the SA-model (48)–(57)). |
warning/0507/cond-mat0507282.html | ar5iv | text | # Decoherence of Einstein-Podolsky-Rosen pairs in a noisy Andreev entangler
## I Model of the noisy entangler
### I.1 Hamiltonian of the noisy Andreev entangler
Similar to Ref. Recher, , we exploit a tunneling Hamiltonian description of the system, $`H=H_0+H_T+H_D^{bath}`$, where
$$H_0=H_S+\underset{l}{}(H_{D_l}+H_{L_l})+\underset{l}{}H_l^{bath}.$$
(3)
The superconductor is embodied by the usual BCS theorySchrieffer which for convenience has been summarized in Appendix A. In brief, $`H_S=_{k\sigma }E_𝐤\gamma _{𝐤\sigma }^{}\gamma _{𝐤\sigma }`$ where $`\sigma =,`$ represents the spin index, $`\gamma _{𝐤\sigma }`$ describe excitations out of the BCS ground state $`|0_S`$ defined by $`\gamma _{𝐤\sigma }|0_S=0`$, $`E_𝐤=\sqrt{\mathrm{\Delta }^2+\xi _𝐤^2}`$ is the quasiparticle energy, and $`\xi _𝐤=ϵ_𝐤\mu _S`$ is the normal state single-electron energy counted from the level $`\mu _S`$ where live the Cooper spin-singlet particles. Both dots are embodied by a single level with energy $`ϵ_l`$ very close to $`\mu _S`$ and are typically governed by an Anderson model $`H_{D_l}=ϵ_l_\sigma d_{l\sigma }^{}d_{l\sigma }+Un_{}n_{}`$ and $`l=1,2`$. The resonant dot level $`ϵ_l`$ can be adjusted by the related gate voltage (or by $`V_l`$ on Fig. 1). Other levels do not participate in the transport when the level spacing of the dots is sufficiently large implying $`\delta ϵ>\mu >k_BT`$. Again, $`\mu =\mu _S\mu _l`$ is the difference of electrochemical potentials between the superconductor and the leads. Moreover, through the on-site Coulomb $`U`$ repulsion a double occupied state is rather hindered to form on each dot; $`U`$ is equal to $`2E_c`$ where $`E_c=e^2/(2C)`$ is typically the charging energy on each dot and $`C`$ denotes the total dot’s capacitance. Keep in mind that this Coulomb blockade argument stands for a key point in the efficiency of this Andreev entangler of EPR pairsRecher . The leads are normal and embodied by a non-interacting theory $`H_{L_l}=_{𝐤\sigma }ϵ_𝐤a_{l𝐤\sigma }^{}a_{l𝐤\sigma }`$. We have to consider final two-particle states of the form $`|f=(1/\sqrt{2})[a_{1𝐩}^{}a_{2𝐪}^{}a_{1𝐩}^{}a_{2𝐪}^{}]|i`$ with energy $`ϵ_f=ϵ_𝐩+ϵ_𝐪`$. The preserved spin singlet state is formed out of two electrons,
one being in the $`𝐩`$ state in lead 1 while the other one is in the $`𝐪`$ state in lead 2. Since the total spin is conserved, the singlet state of the initial Cooper pair will be conserved in the transport process and the final state must satisfy $`S_z=0`$. The $`S_z=0`$ configuration of the triplet state is excluded as long as the distance between the two dots is smaller than $`\xi `$. Tunneling events from the dot $`l`$ to the lead $`l`$ or to the point $`𝐫_l`$ in the s-wave superconductor is described by the tunnel Hamiltonian $`H_T=H_{SD}+(H_{DL}+H_{DL}^{})`$, with
$`H_{SD}`$ $`=`$ $`{\displaystyle \underset{l\sigma }{}}T_{SD}d_{l\sigma }^{}\psi _\sigma (𝐫_l)+h.c.={\displaystyle \underset{l}{}}H_{SD_l},`$ (4)
$`H_{DL}`$ $`=`$ $`{\displaystyle \underset{l𝐤\sigma }{}}T_{DL}a_{l𝐤\sigma }^{}d_{l\sigma }={\displaystyle \underset{l}{}}H_{D_lL_l}.`$
Here, $`\psi _\sigma (𝐫_l)`$ annihilates an electron in the superconductor at the site $`𝐫_l`$ and $`d_{l\sigma }^{}`$ creates it again on dot $`l`$ with amplitude $`T_{SD}`$. Tunneling from the dot (lead) $`l`$ to the lead (dot) $`l`$ is described by the tunneling amplitude $`T_{DL}`$ $`(T_{DL}^{})`$. The $`𝐤`$ dependence of $`T_{DL}`$ can be safely neglected. Moreover, like in Ref. Recher, , we require the dot-lead coupling to be much stronger than the superconductor-dot coupling, i.e., $`|T_{SD}|<|T_{DL}|`$, so that electrons that enter the dots from the superconductor will leave the quantum dots to the leads much faster that new electrons can be provided to the dots from the superconductor. Additionally, a stationary occupation due to the couplings to the leads is exponentially small if $`\mu >k_BT`$. Thus, in the asymmetric barrier case, the resonant dot levels $`ϵ_l`$ are occupied only during a virtual process. Other important parameters are the tunneling rates $`\gamma _l=2\pi \nu _l|T_{DL}|^2`$ and $`\gamma _S=2\pi \nu _S|T_{DS}|^2`$ where $`\nu _l`$ is the density of states per spin of the leads at the Fermi energy and $`\nu _S`$ for the superconductor will be defined as $`1/\mathrm{\Delta }`$. Recall that we will work in the regime where $`\gamma _l>\gamma _S`$ and $`\mathrm{\Delta },U,\delta ϵ\mu >\gamma _l,k_BT`$ and close to the resonant condition for the dots $`ϵ_l\mu _S`$.
We model the impedance $`Z_l(\omega )`$ in a microscopic fashion through a long dissipative transmission line composed of an infinite collection of $`L_{tl}C_{tl}`$ oscillators (Fig. 2). Our environments are modeled in a usual way akin to Refs. Markus, and karyn, : the charge operator $`\widehat{Q}_{nl}`$ on the capacitor between two inductances $`L_{tl}`$ and the conjugate flux $`(\mathrm{}/e)\widehat{\varphi }_{nl}`$ are mapped onto the operatorsnote3 $`\widehat{Q}_l(x)`$ and $`\widehat{\varphi }_l(x)`$ which are precisely described by the diagonalized Hamiltonian
$`H_l^{bath}={\displaystyle _0^1}𝑑x\left\{{\displaystyle \frac{\widehat{Q}_l^2(x)}{2C_{tl}}}+{\displaystyle \frac{\mathrm{}^2}{e^2}}{\displaystyle \frac{2}{L_{tl}}}\mathrm{sin}^2\left({\displaystyle \frac{\pi x}{2}}\right)\widehat{\varphi }_l^2(x)\right\}.`$ (5)
The charge (fluctuation) operator $`\widehat{Q}_l(x)`$ and the phase operator $`\widehat{\varphi }_l(x)`$ obey the commutation relation $`[\widehat{\varphi }_l(x),\widehat{Q}_l(y)/e]=i\delta (xy)`$. The Hamiltonian containing the couplings with the dots reads
$$H_D^{bath}=\underset{l\sigma }{}\frac{e}{C_l}\widehat{Q}_{0l}d_{l\sigma }^{}d_{l\sigma }=\underset{l\sigma }{}e\delta V_ld_{l\sigma }^{}d_{l\sigma }.$$
(6)
This term may arise from the extra capacitive coupling between each dot and the voltage fluctuations (the quantum noise) $`\delta V_l(t)=\widehat{Q}_{0l}/C_l`$ with $`\widehat{Q}_{0l}`$ denoting the charge fluctuation operator on the given capacitor $`C_l`$, emerging from the finite impedance $`Z_l(\omega )`$Markus ; karyn . According to Ref. Markus, , one can thoroughly identify $`\widehat{Q}_{0l}=\sqrt{2}_0^1𝑑x\mathrm{cos}(\pi x/2)\widehat{Q}_l(x)`$. At low frequency $`\omega <\omega _{cl}=1/(R_lC_{tl})`$ where $`R_l=\sqrt{L_{tl}/C_{tl}}`$, the transmission line provides an impedance $`Z_l(\omega )=R_l/(1+i\omega /\omega _{cl})R_l`$.
Below, we will absorb the $`H_D^{bath}`$ coupling into the tunneling terms through the unitary transformationNazarov
$$U=\mathrm{exp}\left[i\underset{l\sigma }{}\delta \varphi _l(t)d_{l\sigma }^{}d_{l\sigma }\right],$$
(7)
where we have defined $`\delta \varphi _l(t)=(e/\mathrm{})^t𝑑t^{}\delta V_l(t^{})=\sqrt{2}_0^1𝑑x\mathrm{cos}(\pi x/2)\widehat{\varphi }_l(x)`$. We get $`Ud_{l\sigma }U^{}=e^{i\delta \varphi _l(t)}d_{l\sigma }`$ and $`Ud_{l\sigma }^{}U^{}=e^{i\delta \varphi _l(t)}d_{l\sigma }^{}`$. Moreover, exploiting the correspondance $`H^{}=UHU^{}+i\mathrm{}\frac{dU}{dt}U^{}`$ we realize that the couplings of the dots to the electrical baths can be completely absorbed in a redefinition of the tunneling Hamiltonian as $`H_T^{}=H_{SD}^{}+H_{DL}^{}+H_{DL}^{}`$ where
$`H_{SD}^{}={\displaystyle \underset{l\sigma }{}}T_{SD}d_{l\sigma }^{}e^{i\delta \varphi _l(t)}\psi _\sigma (𝐫_l)+h.c.={\displaystyle \underset{l}{}}H_{SD_l}^{}`$ (8)
$`H_{DL}^{}={\displaystyle \underset{l𝐤\sigma }{}}T_{DL}a_{l𝐤\sigma }^{}d_{l\sigma }e^{i\delta \varphi _l(t)}={\displaystyle \underset{l}{}}H_{D_lL_l}^{}.`$
At a very general level, the effect of the environments can be embodied by a fluctuating phase bound to the dot’s electron creation and annihilation operators such that the total Hamiltonian turns into $`H^{}=H_0+H_T^{}`$.
### I.2 T-matrix and general current formula
In the quantum (zero-temperature) regime the current of two electrons passing from the superconductor via virtual dot states to the leads is formally given by
$`I`$ $`=`$ $`{\displaystyle \frac{2e}{\mathrm{}}}{\displaystyle \underset{p,q}{}}\rho _iW_{fi}`$
$`=`$ $`{\displaystyle \frac{2e}{\mathrm{}}}{\displaystyle \underset{p,q}{}}2\pi \rho _i|f_B|f|T(ϵ_i+E_B^i)|i|i_B|^2`$
$`\times `$ $`\delta (ϵ_fϵ_iE_B^i+E_B^f)`$
$`=`$ $`{\displaystyle \frac{2e}{\mathrm{}}}{\displaystyle \underset{p,q}{}}2\pi \rho _iW_{i,DD}W_{DD,f},`$
where $`W_{fi}`$ embodies the transition rate from the superconductor to the leads taking into account transitions into the electrical environments; $`f_B|`$ denotes the final (excited) state of the baths when the injected electrons arrive in the leads, $`(E_B^fE_B^i)`$ represents the energy supplied to the environments during the EPR transportation process, $`W_{DD,f}`$ and $`W_{i,DD}`$ stand for the transition rates from the dots to the leads and from the superconductor to the dots respectively in the presence of the fluctuations in the gate voltages, and $`\rho _i`$ is the stationary occupation probability for the entire system to be in the initial ground state $`|i|i_B`$ where as introduced in the introduction $`|i=|0_S|0_D|\mu _l`$ and $`|i_B`$ depicts the initial state for the environments. Along the lines of Ref. Recher, , to calculate the transition rates $`W_{DD,f}`$ and $`W_{i,DD}`$ we resort to the T-matrix approach. The on-shell transmission or T-matrix at the energy $`E`$ is precisely defined as
$$T(E)=H_T^{}G(E)\frac{1}{G_0(E)}=H_T^{}\frac{1}{E+i\eta H^{}}(EH_0);$$
(10)
we have introduced the Lipmann-Schwinger operators $`G(E)=1/(EH^{})`$ and $`G_0=1/(EH_0)`$ as well as the small positive real number $`\eta `$ that we take to zero at the end of the calculation. We rewrite the T-matrix as
$$T(E)=H_T^{}\underset{n=0}{\overset{\mathrm{}}{}}\left(\frac{1}{EH_0+i\eta }H_T^{}\right)^n.$$
(11)
It is appropriate to decompose $`T(E=ϵ_i+E_B^i)`$ into the partial T matrices $`𝒯^{}`$ and $`𝒯^{\prime \prime }`$. When two spin-entangled electrons from the superconductor leave to the dots
$`𝒯^{\prime \prime }`$ $`=`$ $`{\displaystyle \frac{1}{i\eta +EH_0}}H_{SD_1}^{}{\displaystyle \frac{1}{i\eta +EH_0}}H_{SD_2}^{}`$
$`+`$ $`(12)`$
$`=`$ $`{\displaystyle \frac{1}{i\eta +EH_0}}H_{SD_1}e^{i\delta \varphi _1}{\displaystyle \frac{1}{i\eta +EH_0}}H_{SD_2}e^{i\delta \varphi _2}`$
$`+`$ $`(12),`$
where $`(12)`$ refers to the same term but exchanging the roles of the labels $`1`$ and $`2`$. $`𝒯^{\prime \prime }`$ refers to the (dissipative) crossed Andreev process. For the resonant dot $``$ lead tunneling we must keep all the terms of the series
$`𝒯^{}`$ $`=`$ $`H_{D_1L_1}^{}`$
$`\times `$ $`{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}\left({\displaystyle \frac{1}{EH_0+i\eta }}H_{D_1L_1}^{}{\displaystyle \frac{1}{EH_0+i\eta }}H_{D_1L_1}^{}\right)^n`$
$`\times `$ $`{\displaystyle \frac{1}{EH_0+i\eta }}H_{D_2L_2}^{}`$
$`\times `$ $`{\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}\left({\displaystyle \frac{1}{i\eta +EH_0}}H_{DL}^{}{\displaystyle \frac{1}{i\eta +EH_0}}H_{DL}^{}\right)^m`$
$`+`$ $`H_{D_2L_2}^{}{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}\left({\displaystyle \frac{1}{EH_0+i\eta }}H_{D_2L_2}^{}{\displaystyle \frac{1}{EH_0+i\eta }}H_{D_2L_2}^{}\right)^n`$
$`\times `$ $`{\displaystyle \frac{1}{EH_0+i\eta }}H_{D_1L_1}^{}`$
$`\times `$ $`{\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}\left({\displaystyle \frac{1}{Ei\eta H_0}}H_{DL}^{}{\displaystyle \frac{1}{Ei\eta H_0}}H_{DL}^{}\right)^m.`$
To compute the resonant dot $``$ lead tunneling, we have explicitly taken into account virtual $`|DD|DD`$ transitions via the sequences $`|DD|pD\mathrm{}|DD`$ or $`|DD|Dq\mathrm{}|DD`$. Again the baths are insensitive to those virtual transitions which take place in a very short time and let the state of the electron system unchanged. Here, $`|pD`$ stands for $`a_{1𝐩\sigma }^{}d_{2\sigma }^{}|i`$ and implies that the electron 1 is in lead 1 (L1) whereas the other electron resides on dot 2. We have also introduced the state $`|Dq=d_{1\sigma }^{}a_{1𝐪\sigma }^{}|i`$ representing the state with one electron on dot 1 and the other one in lead 2 (L2). At this point we emphasize that virtual states with both electrons in the leads leading to $`|DD|pD|pq|Dq|DD`$ are suppressed by a factor $`\gamma _l/\mu <1`$ compared to that with only one electron in the leads and therefore can be neglectedRecher . Recall that $`|pq`$ stands for $`a_{1𝐩\sigma }^{}a_{2𝐪\sigma }^{}|i`$, where $`𝐩`$ is the momentum from lead 1 and $`𝐪`$ from lead 2; this is the final state $`|f`$ of the electrons on the leads with energy $`ϵ_f=ϵ_𝐪+ϵ_𝐩`$.
### I.3 “1-photon” approximation
In Eq. (13), we have assumed that the tunneling events are almost instantaneous implying that only one “photon” is emitted in each bath during the two-particle Breit-Wigner resonance between dots and leads. In all the products $`H_{D_lL_l}^{}(t)H_{D_lL_l}^{}(0)`$ appearing in Eq. (13) we have identified $`e^{i\delta \varphi _l(t)}e^{i\delta \varphi _l(0)}(t\omega _{cl})^{2R_l/R_K}1`$, that is ensured when the time $`t`$ for an electron on dot $`l`$ to virtually jump in lead $`l`$ and then to go back to dot $`l`$ is much shorter than $`1/\omega _{cl}R_lC_{tl}`$. Assuming that $`C_{tl}`$ is large enough this should be well satisfied even for weak resistances $`R_l`$. Note that configurations with emission of multiple “photons” in the same bath would only result in small corrections in the final current. Keeping only the “1-photon” contribution, $`𝒯^{}`$ may be summarized asnoteC
$$𝒯^{}=e^{i(\delta \varphi _1+\delta \varphi _2)}T^{},$$
(14)
where $`T^{}`$ yields the same form as $`𝒯^{}`$ if one replaces $`H_{D_lL_l}^{}`$ by $`H_{D_lL_l}`$, i.e., without dissipation $`(R_1=R_2=0)`$; this part is at the origin of the two-particle Breit-Wigner resonance between the dots and the leadsRecher . Since $`|T_{SD}|<|T_{DL}|`$ we may always rewrite $`𝒯^{\prime \prime }=e^{i(\delta \varphi _1+\delta \varphi _2)}T^{\prime \prime }`$.
Below, we will distinguish between the case of a single dissipative bath implying $`R_2=0`$ and therefore $`\delta \varphi _2=0`$ and that of two (independent) baths. When $`R_2=0`$, the transition rates $`W_{DD,f}`$ and $`W_{i,DD}`$ are given by
$`W_{DD,f}`$ $`=`$ $`\left|pq|T^{}(ϵ_i)|DDf_B|e^{i\delta \varphi _1}|D_B\right|^2`$
$`\times `$ $`\delta (ϵ_fϵ_1ϵ_2+E_B^fE_B^D),`$
$`W_{i,DD}`$ $`=`$ $`\left|DD|T^{\prime \prime }(ϵ_i)|iD_B|e^{i\delta \varphi _1}|i_B\right|^2`$
$`\times `$ $`\delta (ϵ_1+ϵ_2+E_B^DE_B^iϵ_i).`$
Note that we have replaced $`pq|T^{}(E)|DD`$ and $`DD|T^{\prime \prime }(E)|i`$ by $`pq|T^{}(ϵ_i)|DD`$ and $`DD|T^{\prime \prime }(ϵ_i)|i`$ respectively; this will be thoroughly justified in Sec. II B. Furthermore, we have introduced the energy of the intermediate state of the bath $`E_B^D`$. The energy-conserving $`\delta `$ functions traduce the fact that for each tunneling process the energy of the full system including the bath is conserved or that the energy supplied to the bath is equal to the energy lost by the electrons. The product of those two $`\delta `$ functions is equivalent to the $`\delta `$ function in Eq. (9) (dimension of current will be implicitly respected below).
When $`R_1=R_2=0`$, we recover the formulas of Ref. Recher,
$`W_{DD,f}`$ $`=`$ $`\left|pq|T^{}|DD\right|^2\delta (ϵ_fϵ_1ϵ_2),`$ (16)
$`W_{i,DD}`$ $`=`$ $`\left|DD|T^{\prime \prime }|i\right|^2\delta (ϵ_1+ϵ_2ϵ_i),`$
or more precisely we obtain the following expressionRecher ,
$`W_{fi}=2\pi \left|pq|T^{}|DD\right|^2\left|DD|T^{\prime \prime }|i\right|^2\delta (ϵ_fϵ_i).`$ (17)
The chemical potentials $`ϵ_1`$ and $`ϵ_2`$ of the quantum dots can be tuned by external gate voltages such that the coherent tunneling of two electrons into different leads is at resonance, described by a two-particle Breit-Wigner resonance peaked at $`ϵ_1+ϵ_2=2\mu _S=ϵ_𝐪+ϵ_𝐩`$ (Fig. 3); the situation we consider is when the two dots are close to the resonance condition $`ϵ_1=ϵ_2=0`$. We will choose energies such that $`ϵ_i=2\mu _S=0`$. It is then an interesting question to understand how the quantum dissipation affects the elastic tunneling of those EPR pairs. In the case of two independent baths, we straightforwardly generalize
$`W_{i,DD}=\left|DD|T^{\prime \prime }|i|^2\right|D_{B_1}|e^{i\delta \varphi _1}|i_{B_1}|^2\times `$ (18)
$`\left|D_{B_2}|e^{i\delta \varphi _2}|i_{B_2}\right|^2\delta (ϵ_1+ϵ_2+{\displaystyle \underset{l}{}}E_{B_l}^D{\displaystyle \underset{l}{}}E_{B_l}^iϵ_i),`$
and similarly
$`W_{DD,f}=\left|pq|T^{}|DD|^2\right|f_{B_1}|e^{i\delta \varphi _1}|D_{B_1}|^2\times `$ (19)
$`\left|f_{B_2}|e^{i\delta \varphi _2}|D_{B_2}\right|^2\times \delta (ϵ_fϵ_1ϵ_2+{\displaystyle \underset{l}{}}E_{B_l}^f{\displaystyle \underset{l}{}}E_{B_l}^D).`$
We have decomposed $`|D_B=|D_{B_1}|D_{B_2}`$,… .
Before to pursue, we shall discuss what is the value of $`\rho _i`$ from Eq. (I.2) in the presence of the environments. Similar to the noiseless case, we estimate $`\rho _i=1𝒪(\gamma )1`$ where $`\gamma =\gamma _1+\gamma _2`$. More precisely, the initial (gound) state $`|i=|0_S|0_D|\mu _l`$ is such that the highest level of the dots is unoccupied, there is no quasiparticle on the superconductor which is immediately fulfilled when $`k_BT\mathrm{\Delta }`$, and the Fermi level of the leads remain fixed to $`\mu _l`$. Since we consider asymmetric barriers $`|T_{DL}|>|T_{SD}|`$ the most prominent transfer of electrons is between the leads and the dots. Therefore, the probability for the system to remain in the state $`|i|i_{B_1}|i_{B_2}`$ after a time $`t`$ can be estimated as $`1𝒪(\gamma )`$. We argue that the environments will not affect this equality because as long there is no electron on the dots the environments will unambiguously remain in their ground states $`|i_{B_l}`$.
## II Dissipation on dot 1
Let us assume that $`R_2=0`$. To calculate the rate for electron tunneling from the superconductor to the dots we have to evaluate $`W_{i,DD}`$ from Eq. (15). Similar to Refs. Nazarov, ; Marquardt2, we trace out environmental states leading to
$`W_{i,DD}=\left|DD|T^{\prime \prime }|i\right|^2{\displaystyle _{\mathrm{}}^+\mathrm{}}{\displaystyle \frac{dt}{2\pi \mathrm{}}}e^{i(ϵ_iϵ_1ϵ_2)t/\mathrm{}}`$ (20)
$`\times e^{i\delta \varphi _1(t)}e^{i\delta \varphi _1(0)}.`$
The brackets denote an average over the initial bath ground state $`|i_B`$. For later convenience we like to introduce the abbreviation $`J_1(t)=[\delta \varphi _1(t)\delta \varphi _1(0)]\delta \varphi _1(0)`$ as well as the Fourier transform
$$P_1(E)=\frac{1}{2\pi \mathrm{}}_{\mathrm{}}^+\mathrm{}𝑑t\mathrm{exp}\left[J_1(t)+\frac{i}{\mathrm{}}Et\right].$$
(21)
This already permits us to write down
$`W_{i,DD}=\left|DD|T^{\prime \prime }|i\right|^2P_1(ϵ_iϵ_1ϵ_2).`$ (22)
In a similar way, we extractnote2
$`W_{DD,f}=\left|pq|T^{}|DD\right|^2P_1(ϵ_1+ϵ_2ϵ_f).`$ (23)
### II.1 General discussion on $`P_1(E)`$
We may interpret $`P_1(E)`$ as the probability to emit the energy $`E`$ to the electrical circuit when transferring an electron from the superconductor to the dot 1 or from
the latter to the corresponding lead 1. It is certainly useful to know more about the function $`P_1(E)`$ as well as $`J_1(t)`$. We will first envision the case of a very large resistance $`R_1`$. In that case the dissipative lead yields an effective impedance $`Z_{eff}(\omega )=1/(R_1^1+i\omega C_1)`$ which tends to $`(\pi /C_1)\delta (\omega )`$, $`C_1`$ being the capacitance between the dissipative lead and the dot 1. For the correlation function $`J_1(t)`$ this concentration of environmental modes at low frequency means that the short-time expansionNazarov
$$J_1(t)=_{\mathrm{}}^+\mathrm{}\frac{d\omega }{\omega }\frac{\mathrm{}eZ_{eff}(\omega )}{R_K}e^{i\omega t}=\frac{\pi }{C_1R_K}it,$$
(24)
works for all times. This results in
$$P_1(E)=\delta (EE_{c1}),$$
(25)
so that in order to hop onto the dot 1 an electron must transfer to the environment an amount of energy corresponding to the charging energy $`E_{c1}=e^2/(2C_1)`$ of the capacitor $`C_1`$; this will fatally lead to a Coulomb gapNazarov in the current for low applied bias voltage $`\mu E_{c1}`$ between the superconductor and the leads (L1 and L2). For small resistances $`\alpha _1=R_K/R_11`$ the gate lead may be described by the frequency-independent impedance $`Z_{eff}=Z_1=R_1`$. Based on the transmission line representation for the environment then we may identifyDevoret
$$P_1(E)=\frac{\mathrm{exp}(2\gamma _e/\alpha _1)}{\mathrm{\Gamma }(2/\alpha _1)}\frac{1}{E}\left[\frac{\pi }{\alpha _1}\frac{E}{E_{c1}}\right]^{2/\alpha _1},$$
(26)
where $`\gamma _e=0.577\mathrm{}`$ is the Euler constant. The factor appearing may be motivated by the behavior of the correlation function $`J_1(t)`$ for large timesGrabert
$$J_1(t)=\frac{2}{\alpha _1}\left[\mathrm{ln}(\alpha _1E_{c1}t/\pi \mathrm{})+i\frac{\pi }{2}+\gamma _e\right].$$
(27)
The function $`P_1(E)`$ has been summarized through Fig. 4 with the two distinct behaviors at low and large $`R_1`$.
### II.2 Discussion on tunneling matrix elements
Now we want to properly justify the fact that the baths “cancel out” in the computation of the tunneling matrix elements in Eq. (15). In the limit of a weak resistance $`R_1`$ this is straightforward since the bath 1 only absorbs a small amount of energy during the tunneling events ($`P_1(E)`$ is strongly diverging at $`E=0`$) and therefore in all the Lipmann-Schwinger operators appearing in Eqs. (12) and (13), for a given bath state $`|\alpha _B`$, one can always formally replace $`\alpha _B|_lH_l^{bath}|\alpha _BE_B^i`$. Interestingly, we like to emphasize that for large resistances this argument still holds. In the large resistance limit, one must typically satisfy $`E_B^DE_B^i+E_{c1}`$. Thus, in order to get a finite current, one must thoroughly re-adjust the chemical potential of the SC lead such as $`ϵ_iϵ_i+E_{c1}`$ with $`ϵ_i=0`$; see Eq. (25). When focussing on the tunneling of a Cooper pair from the SC to the dots, hence one requires to evaluate $`D_B|DD|\frac{1}{E_B^i+E_{c1}H_0+i\eta }`$. Applying $`D_B|H_0=E_B^D`$ we check that $`D_B|DD|\frac{1}{E_B^i+E_{c1}H_0+i\eta }`$ is equivalent to $`DD|\frac{1}{H_0+i\eta }`$ by setting $`H_l^{bath}=0`$ in $`H_0`$. We can thus substitute $`G_0`$ by its expression in the absence of the bath $`1`$. This procedure can be extended to two baths. For sake of clarity, calculations of $`pq|T^{}(ϵ_i)|DD`$ and $`DD|T^{\prime \prime }(ϵ_i)|i`$ (with the substitution $`H_l^{bath}=0`$ in $`H_0`$) are derived in Appendices B and C.
### II.3 Large resistance limit
If we maintain the electrochemical potentials of the leads L1 and L2 and of the superconductor so that $`ϵ_1+ϵ_2=2\mu _S=ϵ_𝐩+ϵ_𝐪=0`$ due to Eq. (25) we immediately infer that the current will inevitably go to zero. Indeed
$`W_{i,DD}=\left|DD|T^{\prime \prime }|i\right|^2\delta (2\mu _Sϵ_1ϵ_2E_{c1})`$ , (28)
$`W_{DD,f}=\left|pq|T^{}|DD\right|^2\delta (ϵ_1+ϵ_2ϵ_fE_{c1})`$ ,
reflecting the dynamical Coulomb blockade phenomenon resulting from large impedancesNazarov . On the other hand, one could envision to symmetrically modify the electrochemical potentials of the leads L1 and L2 so that $`\mu _S\mu _S+E_{c1}/2`$ and $`\mu _l\mu _lE_{c1}/2`$ (Fig. 5). In that case, the two $`\delta `$ functions above would be satisfied. Tunneling becomes possible only if the energy at disposal is equal to $`E_{c1}`$.
We assume for this circumstance that the superconducting gap is large enough $`\mathrm{\Delta }E_{c1}`$ so that the superconductor is not subject to quasiparticle poisoning. Using Appendix B and mostly Eq. (66), we value
$`|pq|T^{}|DD|^2|T_{DL}|^4\left(ϵ_1+ϵ_2i\eta \right)^2{\displaystyle \frac{16}{E_{c1}^2}}{\displaystyle \frac{\pi \delta (ϵ_𝐩+E_{c1}/2)}{\gamma _1}}.`$ (29)
We have used the energy conservation $`ϵ_𝐩+ϵ_𝐪=E_{c1}`$ together with $`ϵ_1+ϵ_2=0`$ as well as $`ϵ_𝐩+E_{c1}/2\gamma _l`$. Resorting to Appendix C, we obtain the following current
$$I[\delta \mu E_{c1}]=\frac{4e\gamma \gamma _S^2}{\mathrm{}E_{c1}^2}\left(\frac{\mathrm{sin}(k_F\delta r)}{k_F\delta r}\right)^2e^{2\delta r/\pi \xi }.$$
(30)
Exploiting Eq. (90), we realize that compared to the noiseless case where $`R_1=R_2=0`$ the current becomes suppressed by a factor $`(\gamma /E_{c1})^2`$. This stems from the physical fact that shifting the electrochemical potentials of the leads such that $`\mu _l\mu _lE_{c1}/2`$ hampers the two-particle Breit-Wigner resonance between the dots at resonance $`(ϵ_1+ϵ_2=0)`$ and the leads. In brief, the application of a prominent bias voltage between the dots and the leads somehow produces the “decoherence” of the EPR pair hence affecting the crossed Andreev current.
### II.4 Small resistance
We now turn our attention to the realistic situation of a small resistance so that $`\alpha _1=R_K/R_11`$. Of great interest to us is to understand how the quantum noise affects the long-time coherence of the EPR pair during the two-particle Breit-Wigner process involving the dots and the leads. An explicit calculation has been performed in Appendix D and we find an EPR current of the form
$`II[R_1=0]{\displaystyle \frac{\mathrm{exp}(2\gamma _e/\alpha _1)}{\mathrm{\Gamma }(1+2/\alpha _1)}}\left({\displaystyle \frac{\pi }{\alpha _1}}\right)^{\frac{2}{\alpha _1}}\left({\displaystyle \frac{2\mu }{E_{c1}}}\right)^{\frac{2}{\alpha _1}},`$ (31)
where the crossed Andreev current at $`R_1=0`$ readsRecher
$`I[R_1=0]={\displaystyle \frac{4e\gamma _S^2}{\mathrm{}\gamma }}\left({\displaystyle \frac{\mathrm{sin}(k_F\delta r)}{k_F\delta r}}\right)^2e^{2\delta r/\pi \xi }.`$ (32)
Note that the Breit-Wigner resonance still occurs when $`ϵ_1+ϵ_22\mu _Sϵ_𝐩+ϵ_𝐪`$ reflecting the fact that for weak resistances the function $`P_1(E)`$ diverges at $`E=0`$ and therefore the bath 1 only absorbs a tiny amount of energy during the EPR transportation process. The suppression factor $`\left(2\mu /E_{c1}\right)^{2R_1/R_K}`$ traduces the orthogonality catastrophe arising in dissipative tunneling problemsNazarov ; Devoret ; Girvin ; karyn ; Recher4 . Now, let us return to the discussion on the decoherence (decay) of a noisy EPR Cooper pair that is described, after averaging over the bath degrees of freedom, by the dot state $`e^{i\delta \varphi _1}[d_1^{}d_2^{}d_1^{}d_2^{}]|i`$. In the weak resistance realmNazarov , we can evaluate
$$e^{i\delta \varphi _1(t)}=\mathrm{exp}\left[\frac{1}{\alpha _1}\mathrm{ln}\left(\frac{E_{c1}\alpha _1t}{\pi \mathrm{}}\right)\right].$$
(33)
At very long-times $`t\mathrm{}\pi /(\alpha _1E_{c1})`$ or very low energies $`\mu E_{c1}`$, by coupling to the environment the noisy EPR pair looses its phase coherence affecting drastically the efficiency of the two-particle Breit-Wigner resonance between the dots and the leads L1 and L2. It is worth to note the similitude with the power-law suppression in the Andreev entangler with Luttinger leadsRecher2 ; Smitha . For larger resistances, at low $`\mu `$, we observe a precursor effect of the dynamical Coulomb gap mentioned above.
## III Two independent baths
We can straightforwardly generalize the previous analysis to the case of two (independent) environments assuming $`R_20`$. The two-particle Breit-Wigner transport between the leads and the dots gets modified as
$`W_{DD,f}=\left|pq|T^{}|DD\right|^2P_{12}(ϵ_1+ϵ_2ϵ_𝐩ϵ_𝐪),`$ (34)
where we have introduced
$`P_{12}(E)={\displaystyle _{\mathrm{}}^+\mathrm{}}{\displaystyle \frac{dt}{2\pi \mathrm{}}}e^{iEt/\mathrm{}}e^{J_1(t)+J_2(t)},`$ (35)
and $`J_2(t)=[\delta \varphi _2(t)\delta \varphi _2(0)]\delta \varphi _2(0)`$. When the two resistances are larger than $`R_K`$, $`P_{12}(E)=\delta (EE_{c1}E_{c2})`$ where $`E_{c2}=e^2/(2C_2)E_{c1}`$ is the charging energy of the capacitor $`C_2`$. To get a finite current between the superconductor and the leads 1 and 2 one needs to re-adjust $`2\mu _S2\mu _S+E_{c1}+E_{c2}`$, $`\mu _1\mu _1E_{c1}`$, and $`\mu _2\mu _2E_{c2}`$ (we assume $`E_{c1}E_{c2}`$ so that L1 and L2 are kept at the same electrochemical potential). We find
$$I[\delta \mu 2E_{c1}]=\frac{I[R_1=0]}{4}\gamma ^2\left(\frac{1}{E_{c1}}+\frac{1}{E_{c2}}\right)^2.$$
(36)
We can observe a huge suppression factor $`(\gamma /E_{c1}+\gamma /E_{c2})^2`$. When the two resistances are much smaller than $`R_K`$, which is the situation of most interest, we get
$$P_{12}(E)=\frac{\mathrm{exp}(2\gamma _e/\alpha )}{\mathrm{\Gamma }(2/\alpha )}\frac{1}{E}\left[\frac{\pi }{\alpha }\frac{E}{E_{c1}}\right]^{2/\alpha },$$
(37)
where
$$\alpha ^1=\alpha _1^1+\alpha _2^1=\frac{R_1}{R_K}+\frac{R_2}{R_K}.$$
(38)
We obtain a result identical to that of a unique weakly-resistive bath with $`\alpha _1\alpha `$ (see Appendix D). Recall that the orthogonality catastrophe becomes more pronounced and the suppression factor in the crossed Andreev current now follows $`(2\mu /E_{c1})^{2/\alpha }=(2\mu /E_{c1})^{2/\alpha _1+2/\alpha _2}`$; consult Eq. (D15). Finally one could envision to investigate the asymmetric case where one resistance is well prominent, e.g., $`R_1`$, and the other is weak but nonzero resulting in
$$P_{12}(E)=\frac{\mathrm{exp}(2\gamma _e/\alpha _2)}{\mathrm{\Gamma }(2/\alpha _2)}\frac{1}{E}\left[\frac{\pi }{\alpha _2}\frac{EE_{c1}}{E_{c1}}\right]^{2/\alpha _2}.$$
(39)
It is important to visualize that in that case $`P_{12}(E)`$ yields a visible singularity at $`E=E_{c1}`$ (which is reminiscent of the situation where $`R_1`$ is large and $`R_2=0`$) and therefore to get a current through the structure again one must re-adjust $`\mu _S\mu _S+E_{c1}/2`$ and $`\mu _l\mu _lE_{c1}/2`$. We replace $`\left|pq|T^{}|DD\right|^2`$ by its value in Eq. (29) leading to
$$I=I[\delta \mu E_{c1}]_{\mu _l}^{ϵ_c}𝑑ϵ_𝐪P_2(ϵ_1+ϵ_2\mu _lϵ_𝐪),$$
(40)
and therefore
$`I[\delta \mu E_{c1},\alpha _21]I[\delta \mu E_{c1}]{\displaystyle \frac{\mathrm{exp}(2\gamma _e/\alpha _2)}{\mathrm{\Gamma }(1+2/\alpha _2)}}\left({\displaystyle \frac{\mu }{E_{c1}}}\right)^{\frac{2}{\alpha _2}}.`$ (41)
The crossed Andreev current becomes markedly suppressed by a factor proportional to $`(\gamma /E_{c1})^2(2\mu /E_{c1})^{2/\alpha _2}`$ and $`\mu E_{c1}`$ is the applied bias voltage before the re-adjustment $`\mu _S\mu _S+E_{c1}/2`$ and $`\mu _l\mu _lE_{c1}/2`$.
## IV Parasitic direct Andreev processes
Thus far, we have completely omitted processes allowing the two electrons forming the Cooper pair to jump onto the same dot. In absence of thermal effects, the latter are somehow reduced due to the traditional Coulomb blockade phenomenon on the dots as well as the superconducting gap. Below, we will precisely discuss the quantum noise effect on those processes as well as the efficiency condition(s) of the noisy Andreev entangler.
In the absence of quantum noise, direct Andreev processes (where the two electrons take the same dot) are suppressed by a factor $`(\gamma _l/U)^2`$ and/or $`(\gamma _l/\mathrm{\Delta })^2`$ compared to the crossed Andreev process as a consequence of the Coulomb blockade and the superconducting gapRecher . Nevertheless, those direct Andreev processes do not suffer from a suppression resulting from the spatial separation of the quantum dots. We will use the terminology of Recher et al.Recher by identifying two distinct direct Andreev processes: (I) In the first step, one electron tunnels from the superconductor to, say, dot 2, and in a second step the second electron also tunnels to dot 2. There are now two electrons on dot 2 which costs the Coulomb repulsion energy $`U`$; this virtual state is suppressed by $`1/U`$. Hence the two electrons leave dot 2 and tunnel to lead 2 (L2) one after the other. (II) There is a competing process that avoids double occupancy on dots but leaves an excitation on the superconductor that costs $`1/\mathrm{\Delta }`$. Here, one electron tunnels to, say, dot 2 and then the same electron tunnels further into lead 2. Finally, the second electron tunnels from the superconductor via dot 2 into lead 2.
We first concentrate on the tunneling process (II) including the quantum noise. The current $`I^{(II)}`$ from the superconductor to the final lead state takes the form
$$I^{(II)}=\frac{2e}{\mathrm{}}\underset{𝐩,𝐩^{}}{}\underset{l}{}2\pi W_{i,Dp^{\prime \prime }}^{(II)}W_{Dp^{\prime \prime },f}^{(II)}.$$
(42)
We define the transition rate from $`|Dp^{\prime \prime }`$ to $`|f`$ as $`W_{Dp^{\prime \prime },f}^{(II)}=|w_{Dp^{\prime \prime },f}^{(II)}|^2\delta (ϵ_f+E_B^fϵ_lϵ_{𝐩^{\prime \prime }}E_B^{Dp^{\prime \prime }})`$ and
$`w_{Dp^{\prime \prime },f}^{(II)}={\displaystyle \underset{𝐩^{\prime \prime }\sigma }{}}f_B|f|H_{DL}^{}|Dp^{\prime \prime }\sigma |D_Bp^{\prime \prime }D_Bp^{\prime \prime }|Dp^{\prime \prime }\sigma |`$ (43)
$`\times {\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}\left({\displaystyle \frac{1}{i\eta H_0}}H_{DL}^{}{\displaystyle \frac{1}{i\eta H_0}}H_{DL}^{}\right)^n|Dp^{\prime \prime }\sigma |D_Bp^{\prime \prime }.`$
The final state $`|f`$ with two electrons in the same lead in the singlet state obeys $`|f=(1/\sqrt{2})(a_𝐩^{}a_𝐩^{}^{}a_𝐩^{}a_𝐩^{}^{})|i`$ and we have introduced the intermediate state $`|Dp^{\prime \prime }=d_\sigma ^{}a_{𝐩^{\prime \prime }\sigma }^{}|i`$ with one electron on, say, dot 2, with spin $`\sigma `$ and the other one already in lead 2 with spin $`\sigma `$ and momentum $`𝐩^{\prime \prime }`$. The process where the first electron leaves the superconductor and tunnels to the lead 2 and where the second electron tunnels onto the dot 2 has to be accomplished in a very short-time $`\mathrm{}/\mathrm{\Delta }`$ and thus we assume that this process is instantaneous implying that the bath only reacts when the second electron resides on the dot 2. The intermediate state of the bath with one electron on the dot and the other in the lead has been denoted $`|D_Bp^{\prime \prime }`$ and its energy $`E_B^{Dp^{\prime \prime }}`$. We also decompose $`W_{i,Dp^{\prime \prime }}^{(II)}=|w_{i,Dp^{\prime \prime }}^{(II)}|^2\delta (ϵ_{𝐩^{\prime \prime }}+ϵ_l+E_B^{Dp^{\prime \prime }}ϵ_iE_B^i)`$ where
$`w_{i,Dp^{\prime \prime }}^{(II)}=D_Bp^{\prime \prime }|Dp^{\prime \prime }\sigma |{\displaystyle \frac{1}{i\eta H_0}}H_{SD}^{}{\displaystyle \frac{1}{i\eta H_0}}H_{DL}^{}`$ (44)
$`\times {\displaystyle \frac{1}{i\eta H_0}}H_{SD}^{}|i|i_B.`$
In the Lipmann-Schwinger operators, again we have taken into account that the bath’s energies cancel out (in $`H_0`$ now one must equate $`H_l^{bath}=0`$). From Sec. II B, we know that this is always justified in the weak-resistance limit which will be the situation of interest below. Note that a tunnel process from the state $`|i`$ to the state $`|Dp^{\prime \prime }`$ does not have to be resummed further since this would lead either to a double occupancy of the dot that is suppressed by $`1/U`$ or to a state with two electrons simultaneously in the lead that is suppressed by a factorRecher $`\gamma _l/\mu <1`$. For convenience, we have suppressed the label $`l=1,2`$ in the Hamiltonians $`H_{D_lL_l}^{}`$ and $`H_{SD_l}^{}`$ above. In the presence of voltage noise, we can yet decompose
$`W_{Dp^{\prime \prime },f}^{(II)}`$ $`=`$ $`W_{Dp^{\prime \prime },f}^{(II)o}\left|f_B|e^{i\delta \varphi _l}|D_Bp^{\prime \prime }\right|^2`$
$`\times `$ $`\delta (ϵ_f+E_B^fϵ_lϵ_{𝐩^{\prime \prime }}E_B^{Dp^{\prime \prime }})`$
$`=`$ $`W_{Dp^{\prime \prime },f}^{(II)o}P_l(ϵ_{𝐩^{\prime \prime }}+ϵ_lϵ_f),`$
as well as
$`W_{i,Dp^{\prime \prime }}^{(II)}`$ $`=`$ $`W_{i,Dp^{\prime \prime }}^{(II)o}\left|D_Bp^{\prime \prime }|e^{i\delta \varphi _l}|i_B\right|^2`$
$`\times `$ $`\delta (ϵ_{𝐩^{\prime \prime }}+ϵ_l+E_B^{Dp^{\prime \prime }}ϵ_iE_B^i)`$
$`=`$ $`W_{i,Dp^{\prime \prime }}^{(II)o}P_l(ϵ_iϵ_{𝐩^{\prime \prime }}ϵ_l),`$
and $`(\mathrm{})^{(II)o}`$ are exactly the transition rates occurring in the absence of dissipative effects. Note that the product of Hamiltonians $`H_{SD}^{}H_{DL}^{}H_{SD}^{}`$ is equivalent to $`e^{i\delta \varphi _l}H_{SD}H_{DL}H_{SD}`$. Moreover we identify
$`W_{Dp^{\prime \prime },f}^{(II)o}W_{i,Dp^{\prime \prime }}^{(II)o}`$ $`=`$ $`{\displaystyle \frac{2^{3/2}\nu _S(T_{SD}T_{DL})^2}{\mathrm{\Delta }(ϵ_𝐩+ϵ_li\gamma _l/2)}}`$
$`\times `$ $`{\displaystyle \frac{(ϵ_𝐩+ϵ_𝐩^{})/2+ϵ_li\gamma _l/2}{ϵ_𝐩^{}+ϵ_li\gamma _l/2}}.`$
The momentum $`𝐩^{\prime \prime }`$ does not appear in the final result because one must satisfy either $`\delta _{𝐩,𝐩^{\prime \prime }}`$ or $`\delta _{𝐩,𝐩^{}}`$. In the absence of noise where $`ϵ_𝐩+ϵ_𝐩^{}=2\mu _S=0=ϵ_{𝐩^{\prime \prime }}+ϵ_l`$ we explicitly recover the formula of Ref. Recher, . Now, repeating the same calculation for the process (I) we find that $`W_{Dp^{\prime \prime },f}^{(I)}=W_{Dp^{\prime \prime },f}^{(II)}`$ and furthermore $`w_{i,Dp^{\prime \prime }}^{(I)}`$ obeys
$`w_{i,Dp^{\prime \prime }}^{(I)}=D_Bp^{\prime \prime }|Dp^{\prime \prime }\sigma |{\displaystyle \frac{1}{i\eta H_0}}H_{DL}^{}{\displaystyle \frac{1}{i\eta H_0}}H_{SD}^{}`$ (48)
$`\times {\displaystyle \frac{1}{i\eta H_0}}H_{SD}^{}|i|i_B.`$
Compared to $`w_{i,Dp^{\prime \prime }}^{(II)}`$ the order of tunneling events has changed. $`W_{i,Dp^{\prime \prime }}^{(I)}`$ has a form similar to Eq. (IV) and the amplitude product $`W_{Dp^{\prime \prime },f}^{(I)o}W_{i,Dp^{\prime \prime }}^{(I)o}`$ is given by Eq. (IV) but with $`\mathrm{\Delta }`$ being replaced by the Coulomb gap $`U/\pi `$.
In Appendix E, we have summarized the calculations of the direct Andreev current $`I^{}=I^{(I)}+I^{(II)}`$ and the main result is that for relatively weak resistances
$`I^{}[\alpha _l1]={\displaystyle \frac{I^{}[R_l=0]}{2}}{\displaystyle \underset{l}{}}{\displaystyle \frac{\mathrm{exp}(2\gamma _e/\alpha _l)}{\mathrm{\Gamma }(1+2/\alpha _l)}}\left({\displaystyle \frac{2\mu \pi }{\alpha _lE_{cl}}}\right)^{2/\alpha _l},`$ (49)
and for the noiseless case we again agree with Ref. Recher,
$`I^{}[R_l=0]={\displaystyle \frac{2e\gamma _S^2\gamma }{\mathrm{}^2}};`$ (50)
$`^1=1/(\pi \mathrm{\Delta })+1/U`$ has been already mentioned in the introduction. It is important to stress that even though the direct Andreev current is suppressed by a factor $`(\gamma _l/)^2`$ compared to the crossed Andreev current $`I`$, the former is less sensitive to voltage noise in the sense that this is only affected by an extra factor $`(\mu /E_{c1})^{2/\alpha _l}`$ as opposed to $`(\mu /E_{c1})^{2/\alpha _1+2/\alpha _2}`$ for the crossed Andreev current $`I`$ given explicitly in Eq. (D15) assuming two environments. This stems from the fact that the crossed Andreev current involves a two-particle Breit-Wigner resonance and thus is more sensitive to voltage noise than the direct Andreev processes which demand that one electron instantaneously leaves to the lead $`l`$ hence producing a single-particle Breit-Wigner type transport through, e.g., $`W_{Dp^{\prime \prime },f}^{(II)}`$. This is the uppermost issue of our paper.
Note, in passing, that for the asymmetric case where $`R_2=0`$ strictly whereas $`R_1`$ would be finite (but much smaller than $`R_K`$), which means that dissipation only concerns electrons residing on dot 1, then we easily extract
$`{\displaystyle \frac{I^{}[\alpha _11]}{I^{}[R_l=0]}}={\displaystyle \frac{1}{2}}\left[1+{\displaystyle \frac{\mathrm{exp}(2\gamma _e/\alpha _1)}{\mathrm{\Gamma }(1+2/\alpha _1)}}\left({\displaystyle \frac{2\mu \pi }{\alpha _1E_{c1}}}\right)^{2/\alpha _1}\right].`$ (51)
When the two electrons forming the injected Cooper pair take the dot 2, the direct Andreev current is equivalent to that of the noiseless case. Moreover the direct Andreev current stemming from the passage of the two spin-entangled electrons through dot 1 is affected by the noise in the same manner as the crossed Andreev current.
## V Conclusion
In brief, in the setup of Fig. 1 we have thoroughly investigated the effect of voltage noise produced by the electrical circuits in the vicinity of the quantum dotsMarkus ; karyn on the transportation of nonlocal charged-2e Cooper pairs (spin-based EPR pairs). We emphasize that even though electron spins in a semiconductor environment show unusually long dephasing times approaching microseconds and can be transported phase-coherently over distances exceeding $`100\mu m`$Asc , in the realistic dot-based Andreev entangler introduced in Ref. Recher, the voltage noise may affect the transportation of those EPR pairs through the charge degrees of freedom. Although the spin entanglement is preserved at long times, as a result of the entanglement of the charge 2e with the electromagnetic noise the (noisy) Cooper pair object inevitably decays at long times. Assuming almost instantaneous tunneling events, we have been able to build a “P(E) theory” for this problem along the lines of Ref. Marquardt2, . More precisely, when investigating the Breit-Wigner resonances between the dots and the leads, we have kept only the dominant one “photon” contribution from each bath. For moderate and symmetric resistances $`R_1R_2R_K`$, the condition for the noisy EPR entangler to be efficient ($`I/I^{}>1`$) reads:
$$\left(\frac{}{\gamma }\right)\left(\frac{\mu }{E_{c1}}\right)^{R_l/R_K}>(k_F\delta r).$$
(52)
Here, we have considered that $`\delta r<\pi \xi `$ and two independent baths. It is important to recall that the crossed Andreev current is triggered by an “EPR-pair” Breit-Wigner resonance between the two leads and the two dots and is therefore (slightly) more sensitive to voltage noise than the parasitic direct Andreev processes which only involve a single-particle Breit-Wigner resonance between, say, dot $`1`$ and lead $`1`$ (one electron has been instantaneously transmitted to the lead, e.g., due to the Coulomb blockade effect). Assuming that the Coulomb gap is large enough to satisfy this renormalized efficiency condition, it would be interesting to probe experimentally the long-time decoherence of the EPR pair through the orthogonality catastrophe factor $`(2\mu /E_{c1})^{4R_l/R_K}`$ appearing in the crossed Andreev current. We like to emphasize that one could envision to exploit the dissipative GaAs heterostructures of Ref. Rimberg, to build a quantum dot in the proximity of a two-dimensional electron gas in low density which then serves as a tunable source of dissipation; interestingly, the resistance of the envionment can reach few $`k\mathrm{\Omega }`$. At this step, it is certainly important to also give our opinion on the case where the two dots would be subject to the voltage fluctuations of the same environment possessing a resistance $`R`$. We find that Eq. (19) would turn into $`W_{DD,f}=|pq|T^{}|DD|^2|f_B|e^{2i\delta \varphi }|D_B|^2\delta (ϵ_fϵ_1ϵ_2+E_B^fE_B^D)`$. It follows that the crossed Andreev current would be subject to a more dramatic suppression $`(2\mu /E_{c1})^{8R/R_K}`$ whereas the direct Andreev current would exhibit the same power-law suppression as the two independent bath case. The efficiency condition of the entangler turns into
$$\left(\frac{}{\gamma }\right)^2\left(\frac{\mu }{E_{c1}}\right)^{6R/R_K}>(k_F\delta r)^2.$$
(53)
It is relevant to note that the setup of interest to us is quite different from that of a BCS-superconductor directly coupled to two highly-resistive normal leads being described by two electromagnetic environmentsRecher4 . In that case, similar to Luttinger liquid leadsRecher3 ; Smitha , tunneling of two spin-entangled electrons into the same lead is diminished compared to the crossed Andreev process where the pair splits and each electron tunnels into different leads. The reason is that when a charge $`2e`$ tunnels into the same lead $`l`$ the tunneling process is accompanied by a phase $`e^{i2\varphi _l}`$; this doubling of the phase leads to a more prominent suppression of current compared to the case where the two electrons take different leads. We insist on the fact that in the setup based on quantum dots, when the two electrons tunnel into the same lead the current cannot be triggered by charges $`2e`$ due to the Coulomb blockade; the main process between the dot $`l`$ and the lead $`l`$ is a single-particle Breit-Wigner resonance.
Note in passing that similar conclusions would typically arise when including the noise in the normal leads L1 and L2 in the setup of Fig. 1; more precisely, in that case we should carefully replace $`a_{l𝐤\sigma }a_{l𝐤\sigma }e^{i\delta \varphi _l(t)}`$ in Eq. (8) and hence the same conclusions could be derived.
Since the spin entanglement is not really affected by the electrical noise, one could ask whether it would be possible to detect the spin entanglement of a noisy EPR pair despite the suppressed transmission probability of the latter at low voltage or long time. For example, Ref. Samuelsson, envisions to introduce a beam-splitter and focus on zero-frequency current correlations. Those quantities will be affected in a similar way as the currents (e.g., the total noise of the current flowing out of the superconductor is related to currents through the Schottky’s resultSamuelsson ) and thus the detection of current correlations becomes highly dependent on the bias voltage. However, since the currents and the zero-frequency current correlations are affected in a similar way, we must admit that the Fano factors given in Samuelsson, should not be modified that might give some hope to detect the nonlocal spin entanglement.
Finally, the question whether the EPR pair can survive if the dots are pushed away from resonance and are singly-occupied, is an interesting question that would be worthwhile to investigate further. The (related) structure with a double dot in the Coulomb blockade regime coupled to two superconducting leadsChoi is well known to induce an antiferromagnetic coupling between the dots.
Acknowledgments: The authors are very grateful to M. Büttiker, D. Feinberg, Mei-Rong Li, D. Loss, and E. Sukhorukov, for useful comments on the paper. K.L.H. is also grateful to P. Recher for a very careful reading of the manuscript as well as very constructive discussions. K.L.H. was supported by CIAR, FQRNT, and NSERC.
## Appendix A BCS notations
The s-wave superconductor is described by the BCS theory. The BCS Hamiltonian takes the form
$$H_S=\underset{𝐤\sigma }{}E_𝐤\gamma _{𝐤\sigma }^{}\gamma _{𝐤\sigma },$$
(54)
$`E_𝐤=\sqrt{\xi _𝐤^2+\mathrm{\Delta }^2}`$ being the quasiparticle energy and $`\mathrm{\Delta }`$ the superconducting gap. The quasiparticle operator $`\gamma _{𝐤\sigma }`$ is related to the electron annihilation and creation operators $`c_{𝐤\sigma }`$ and $`c_{𝐤\sigma }^{}`$ through the Bogoliubov transformation
$`c_𝐤=u_𝐤\gamma _𝐤+v_𝐤\gamma _𝐤^{},`$ (55)
$`c_𝐤=u_𝐤\gamma _𝐤v_𝐤\gamma _𝐤^{},`$
where the coherence factors $`u_𝐤=\sqrt{1+(\xi _𝐤/E_𝐤)}/\sqrt{2}`$ and $`v_𝐤=\sqrt{1(\xi _𝐤/E_𝐤)}/\sqrt{2}`$ have been introduced and $`\xi _𝐤=ϵ_𝐤\mu _S`$ is the normal state single-electron energy counted from the Fermi level $`\mu _S`$. We choose energies such that $`\mu _S=0`$. $`\psi _\sigma (𝐫_l)`$ annihilates an electron in the superconductor at the site $`𝐫_l`$ and $`\psi _\sigma (𝐫_l)`$ is related to $`c_{𝐤\sigma }`$ by the Fourier transform $`\psi _\sigma (𝐫_l)=_𝐤e^{i\mathrm{𝐤𝐫}_l}c_{𝐤\sigma }`$. In our calculations, we will have to compute quantities like $`i|\gamma _{𝐤\sigma }\psi _\sigma (𝐫_l)=_𝐤^{}e^{i𝐤^{}𝐫_l}i|\gamma _{𝐤\sigma }c_{𝐤^{}\sigma }`$. The only terms which are non-zero should be proportional to $`i|\gamma _{𝐤\sigma }\gamma _{𝐤\sigma }^{}=i|`$. Hence we infer that we shall select $`c_{𝐤^{}\sigma }=c_{𝐤\sigma }ϵ_\sigma v_𝐤\gamma _{𝐤\sigma }^{}`$ in the equations (A2) above resulting in $`i|\gamma _{𝐤\sigma }c_{𝐤^{}\sigma }=i|v_𝐤\delta (𝐤+𝐤^{})ϵ_\sigma `$ $`(ϵ_\sigma =\pm `$ for $`\sigma =,`$). We evaluate $`i|\psi _\sigma (𝐫_l)\gamma _{𝐤\sigma }^{}=_{𝐤^{\prime \prime }}i|e^{i𝐤^{\prime \prime }𝐫_l}c_{𝐤^{\prime \prime }\sigma }\gamma _{𝐤\sigma }^{}`$ in a similar way and we easily extract $`i|c_{𝐤^{\prime \prime }\sigma }\gamma _{𝐤\sigma }^{}=i|u_𝐤\delta (𝐤𝐤^{\prime \prime })`$.
## Appendix B Calculation of $`pq|T^{}|DD`$
Our aim is now to present a detailed calculation of $`pq|T^{}|DD`$ without imposing $`ϵ_𝐩+ϵ_𝐪=2\mu _S=0`$ (equality stemming from the energy conservation in the absence of dissipation). First, we can rigorously simplify
$`DD|{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}\left({\displaystyle \frac{1}{i\eta H_0}}H_{DL}^{}{\displaystyle \frac{1}{i\eta H_0}}H_{DL}\right)^n|DD`$ (56)
$`={\displaystyle \frac{1}{1DD|\frac{1}{i\eta H_0}H_{DL}^{}\frac{1}{i\eta H_0}H_{DL}|DD}}.`$
Hence, we exploit $`DD|\frac{1}{i\eta H_0}=DD|\frac{1}{i\eta ϵ_1ϵ_2}`$ as well as $`DD|(H_{D_1L_1}^{}\frac{1}{i\eta H_0}H_{D_1L_1}+H_{D_2L_2}^{}\frac{1}{i\eta H_0}H_{D_2L_2})|DD=|T_{DL}|^2_{l,𝐤}\frac{1}{i\eta ϵ_𝐤ϵ_l}`$ resulting in
$`DD|{\displaystyle \frac{1}{i\eta H_0}}H_{DL}^{}{\displaystyle \frac{1}{i\eta H_0}}H_{DL}|DD={\displaystyle \frac{\mathrm{\Sigma }}{i\eta ϵ_1ϵ_2}},`$ (57)
where we have introduced the self-energy
$$\mathrm{\Sigma }=|T_{DL}|^2\underset{l,𝐤}{}(i\eta ϵ_lϵ_𝐤)^1.$$
(58)
Akin to Ref. Recher, , we can straightforwardly decompose $`\mathrm{\Sigma }=\mathrm{}e\mathrm{\Sigma }i\gamma /2`$ where $`\gamma =\gamma _1+\gamma _2`$ and $`\mathrm{}e\mathrm{\Sigma }\gamma _l\mathrm{ln}(ϵ_c/\mu _l)`$ can be neglected assuming that the renormalization of the energy level is small. This leads to the expression
$`DD|{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}\left({\displaystyle \frac{1}{i\eta H_0}}H_{DL}^{}{\displaystyle \frac{1}{i\eta H_0}}H_{DL}\right)^n|DD`$ (59)
$`={\displaystyle \frac{1}{1\frac{i\gamma /2}{i\eta ϵ_1ϵ_2}}}={\displaystyle \frac{ϵ_1+ϵ_2i\eta }{ϵ_1+ϵ_2i\gamma /2}}.`$
Similar results hold for the one-particle resummation
$$pD|\underset{n=0}{\overset{\mathrm{}}{}}(\frac{1}{i\eta H_0}H_{D_2L_2}^{})\frac{1}{i\eta H_0}H_{D_2L_2})^n|pD$$
(60)
then providing
$`pD|{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}\left({\displaystyle \frac{1}{i\eta H_0}}H_{D_2L_2}^{}{\displaystyle \frac{1}{i\eta H_0}}H_{D_2L_2}\right)^n|pD`$ (61)
$`={\displaystyle \frac{1}{1\frac{i\gamma _2/2}{i\eta ϵ_𝐩ϵ_2}}}={\displaystyle \frac{ϵ_𝐩+ϵ_2i\eta }{ϵ_𝐩+ϵ_2i\gamma _2/2}}.`$
Again, $`|pD`$ stands for $`a_{1𝐩\sigma }^{}d_{2\sigma }^{}|i`$ and implies that the electron 1 is in lead 1 (L1) whereas the other electron resides on dot 2. We can also introduce the state $`|Dq=d_{1\sigma }^{}a_{2𝐪\sigma }^{}|i`$ representing the state with one electron on dot 1 and the other one in lead 2 (L2) leading to
$`Dq|{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}\left({\displaystyle \frac{1}{i\eta H_0}}H_{D_1L_1}^{}{\displaystyle \frac{1}{i\eta H_0}}H_{D_1L_1}\right)^n|Dq`$ (62)
$`={\displaystyle \frac{1}{1\frac{i\gamma _1/2}{i\eta ϵ_𝐪ϵ_1}}}={\displaystyle \frac{ϵ_𝐪+ϵ_1i\eta }{ϵ_𝐪+ϵ_1i\gamma _1/2}}.`$
We can now proceed and compute
$`pq|T^{}|DD=pq|H_{D_1L_1}|Dq`$
$`\times `$ $`Dq|{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}\left({\displaystyle \frac{1}{i\eta H_0}}H_{D_1L_1}^{}{\displaystyle \frac{1}{i\eta H_0}}H_{D_1L_1}\right)^n|Dq`$
$`\times `$ $`Dq|{\displaystyle \frac{1}{i\eta H_0}}H_{D_2L_2}|DD`$
$`\times `$ $`DD|{\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}\left({\displaystyle \frac{1}{i\eta H_0}}H_{DL}^{}{\displaystyle \frac{1}{i\eta H_0}}H_{DL}\right)^m|DD`$
$`+`$ $`pq|H_{D_2L_2}|pD`$
$`\times `$ $`pD|{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}\left({\displaystyle \frac{1}{i\eta H_0}}H_{D_2L_2}^{}{\displaystyle \frac{1}{i\eta H_0}}H_{D_2L_2}\right)^n|pD`$
$`\times `$ $`pD|{\displaystyle \frac{1}{i\eta H_0}}H_{D_1L_1}|DD`$
$`\times `$ $`DD|{\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}\left({\displaystyle \frac{1}{i\eta H_0}}H_{DL}^{}{\displaystyle \frac{1}{i\eta H_0}}H_{DL}\right)^m|DD.`$
We obtain
$`pq|T^{}|DD=T_{DL}{\displaystyle \frac{ϵ_𝐪+ϵ_1i\eta }{ϵ_𝐪+ϵ_1i\gamma _1/2}}`$
$`\times `$ $`\left({\displaystyle \frac{T_{DL}}{i\eta ϵ_1ϵ_𝐪}}\right){\displaystyle \frac{ϵ_1+ϵ_2i\eta }{ϵ_1+ϵ_2i\gamma /2}}`$
$`+`$ $`T_{DL}{\displaystyle \frac{ϵ_𝐩+ϵ_2i\eta }{ϵ_𝐩+ϵ_2i\gamma _2/2}}\left({\displaystyle \frac{T_{DL}}{i\eta ϵ_2ϵ_𝐩}}\right){\displaystyle \frac{ϵ_1+ϵ_2i\eta }{ϵ_1+ϵ_2i\gamma /2}},`$
and therefore
$`pq|T^{}|DD=T_{DL}^2{\displaystyle \frac{1}{ϵ_𝐩+ϵ_2i\gamma _2/2}}{\displaystyle \frac{ϵ_1+ϵ_2i\eta }{ϵ_1+ϵ_2i\gamma /2}}`$ (65)
$`T_{DL}^2{\displaystyle \frac{1}{ϵ_𝐪+ϵ_1i\gamma _1/2}}{\displaystyle \frac{ϵ_1+ϵ_2i\eta }{ϵ_1+ϵ_2i\gamma /2}}.`$
This finally leads to
$`pq|T^{}|DD=T_{DL}^2{\displaystyle \frac{ϵ_1+ϵ_2i\eta }{ϵ_1+ϵ_2i\gamma /2}}\times `$ (66)
$`{\displaystyle \frac{ϵ_𝐩+ϵ_𝐪+ϵ_1+ϵ_2i\gamma /2}{(ϵ_𝐩+ϵ_2i\gamma _2/2)(ϵ_𝐪+ϵ_1i\gamma _1/2)}}.`$
When neglecting the resistances of the leads that contain the capacitors $`C_1`$ and $`C_2`$, one can exploit that $`ϵ_𝐪+ϵ_𝐩=2\mu _S=0`$ and recover the result from Ref. Recher,
$`pq|T^{}|DD=T_{DL}^2(ϵ_1+ϵ_2i\eta )\times `$ (67)
$`{\displaystyle \frac{1}{(ϵ_𝐩+ϵ_2i\gamma _2/2)(ϵ_𝐪+ϵ_1i\gamma _1/2)}}.`$
## Appendix C Calculation of $`[d_2d_1\pm d_2d_1]T\mathrm{"}`$
Here, we would like to compute $`(1/\sqrt{2})[d_2d_1\pm d_2d_1]T\mathrm{"}`$ where the abbreviation $`\mathrm{}`$ stands for $`i|\mathrm{}|i`$. This part is formally equivalent to
$`DD|T^{\prime \prime }|i={\displaystyle \frac{1}{\sqrt{2}}}i|[d_2d_1d_2d_1]\times `$ (68)
$`{\displaystyle \frac{1}{i\eta H_0}}H_{SD_1}{\displaystyle \frac{1}{i\eta H_0}}H_{SD_2}|i.`$
We can already evaluate $`i|[d_2d_1d_2d_1]\frac{1}{i\eta H_0}=i|[d_2d_1d_2d_1]\frac{1}{i\eta ϵ_1ϵ_2}`$ leading to
$`DD|T^{\prime \prime }|i={\displaystyle \frac{1}{\sqrt{2}}}{\displaystyle \frac{1}{i\eta ϵ_1ϵ_2}}\times `$ (69)
$`i|[d_2d_1d_2d_1]H_{SD_1}{\displaystyle \frac{1}{i\eta H_0}}H_{SD_2}|i.`$
Now, following Ref. Recher, , we insert a complete set of single-particle (virtual) states
$$1=\underset{l𝐤\sigma }{}\gamma _{𝐤\sigma }^{}d_{l\sigma }^{}|ii|d_{l\sigma }\gamma _{𝐤\sigma },$$
(70)
such that
$`DD|T^{\prime \prime }|i={\displaystyle \frac{1}{\sqrt{2}}}{\displaystyle \frac{1}{i\eta ϵ_1ϵ_2}}i|[d_2d_1d_2d_1]H_{SD_1}`$ (71)
$`\times {\displaystyle \underset{l𝐤\sigma }{}}\gamma _{𝐤\sigma }^{}d_{l\sigma }^{}|ii|d_{l\sigma }\gamma _{𝐤\sigma }{\displaystyle \frac{1}{i\eta H_0}}H_{SD_2}|i.`$
Now we can use $`i|d_{l\sigma }\gamma _{𝐤\sigma }\frac{1}{i\eta H_0}=i|d_{l\sigma }\gamma _{𝐤\sigma }\frac{1}{i\eta E_𝐤ϵ_l}`$; $`E_𝐤`$ being the energy of a BCS quasiparticle. Moreover, considering that the dots are at resonance we can approximate $`i\eta E_𝐤ϵ_lE_𝐤`$ and obtain
$`DD|T^{\prime \prime }|i={\displaystyle \frac{1}{\sqrt{2}}}{\displaystyle \frac{1}{i\eta ϵ_1ϵ_2}}i|[d_2d_1d_2d_1]H_{SD_1}`$ (72)
$`\times {\displaystyle \underset{l𝐤\sigma }{}}{\displaystyle \frac{1}{E_𝐤}}\gamma _{𝐤\sigma }^{}d_{l\sigma }^{}|ii|d_{l\sigma }\gamma _{𝐤\sigma }H_{SD_2}|i`$
$`={\displaystyle \frac{1}{\sqrt{2}}}{\displaystyle \frac{T_{SD}^2}{i\eta +ϵ_1+ϵ_2}}i|[d_2d_1d_2d_1]{\displaystyle \underset{l𝐤\sigma }{}}{\displaystyle \underset{\sigma ^{}}{}}{\displaystyle \underset{\sigma ^{\prime \prime }}{}}\times `$
$`{\displaystyle \frac{1}{E_𝐤}}d_{1\sigma ^{}}^{}\mathrm{\Psi }_\sigma ^{}(𝐫_1)\gamma _{𝐤\sigma }^{}d_{l\sigma }^{}|ii|d_{l\sigma }\gamma _{𝐤\sigma }d_{2\sigma ^{\prime \prime }}^{}\mathrm{\Psi }_{\sigma ^{\prime \prime }}(𝐫_2)|i.`$
The terms which survive are those with $`l=2`$ and $`\sigma ^{}=\sigma ^{\prime \prime }=\sigma `$ ($`\sigma `$ is the spin polarization opposite to $`\sigma `$):
$`|DD|T^{\prime \prime }|i|={\displaystyle \frac{1}{\sqrt{2}}}{\displaystyle \frac{T_{SD}^2}{i\eta +ϵ_1+ϵ_2}}i|[d_2d_1d_2d_1]`$ (73)
$`\times {\displaystyle \underset{𝐤\sigma }{}}{\displaystyle \frac{1}{E_𝐤}}d_{1\sigma }^{}\mathrm{\Psi }_\sigma (𝐫_1)\gamma _{𝐤\sigma }^{}d_{2\sigma }^{}|ii|\gamma _{𝐤\sigma }\mathrm{\Psi }_\sigma (𝐫_2)d_{2\sigma }d_{2\sigma }^{}|i.`$
Now we can develop $`\mathrm{\Psi }_\sigma (r_2)=_𝐤^{}e^{i𝐤^{}r_1}c_{𝐤^{}\sigma }`$ as well as $`\mathrm{\Psi }_\sigma (r_1)=_{𝐤^{\prime \prime }}e^{i𝐤^{\prime \prime }r_1}c_{𝐤^{\prime \prime }\sigma }`$, and exploit the precious equalities $`i|\gamma _{𝐤\sigma }c_{𝐤^{}\sigma }=i|v_𝐤\delta (𝐤+𝐤^{})ϵ_\sigma `$ and $`i|c_{𝐤^{\prime \prime }\sigma }\gamma _{𝐤\sigma }^{}=i|u_𝐤\delta (𝐤𝐤^{\prime \prime })`$ which have been demonstrated in Appendix A. Note that in agreement with the BCS theory we satisfy $`𝐤^{}=𝐤^{\prime \prime }`$. Hence we converge to
$`|DD|T^{\prime \prime }|i|={\displaystyle \frac{1}{\sqrt{2}}}{\displaystyle \frac{T_{SD}^2}{i\eta +ϵ_1+ϵ_2}}i|[d_2d_1d_2d_1]`$ (74)
$`\times {\displaystyle \underset{𝐤\sigma }{}}{\displaystyle \frac{u_𝐤v_𝐤}{E_k}}e^{i𝐤(𝐫_1𝐫_2)}ϵ_\sigma d_{1\sigma }^{}d_{2\sigma }^{}|i.`$
Now we can resort to the important equality
$$\underset{\sigma }{}i|[d_2d_1d_2d_1]ϵ_\sigma d_{1\sigma }^{}d_{2\sigma }^{}|i=2,$$
(75)
then leading to
$$|DD|T^{\prime \prime }|i|=\frac{\sqrt{2}T_{SD}^2}{i\eta +ϵ_1+ϵ_2}\underset{𝐤}{}\frac{u_𝐤v_𝐤}{E_𝐤}e^{i𝐤(𝐫_1𝐫_2)}.$$
(76)
In fact, another term which consists of exchanging the roles of $`H_{SD_1}`$ and $`H_{SD_2}`$ or $`𝐫_1`$ and $`𝐫_2`$ in Eq. (C1) should be also included (see Eq. (12)). This means that
$`|DD|T^{\prime \prime }|i|={\displaystyle \frac{\sqrt{2}T_{SD}^2}{i\eta +ϵ_1+ϵ_2}}{\displaystyle \underset{𝐤}{}}`$ (77)
$`\times {\displaystyle \frac{u_𝐤v_𝐤}{E_𝐤}}\left(e^{i𝐤(𝐫_1𝐫_2)}+e^{i𝐤(𝐫_1𝐫_2)}\right),`$ (78)
and finally
$$|DD|T^{\prime \prime }|i|=\frac{2\sqrt{2}T_{SD}^2}{i\eta +ϵ_1+ϵ_2}\underset{𝐤}{}\frac{u_𝐤v_𝐤}{E_𝐤}\mathrm{cos}(𝐤\delta 𝐫).$$
(79)
Now, we can exploit $`u_𝐤v_𝐤=\frac{1}{2E_𝐤}\sqrt{E_𝐤^2\xi _𝐤^2}=\mathrm{\Delta }/(2E_𝐤)`$:
$$|DD|T^{\prime \prime }|i|=\frac{2\sqrt{2}T_{SD}^2}{i\eta +ϵ_1+ϵ_2}\underset{𝐤}{}\frac{\mathrm{\Delta }}{2E_𝐤^2}\mathrm{cos}(𝐤.\delta 𝐫).$$
(80)
Here, we can assume that $`|𝐤|k_F`$ and we note $`𝐤_F.\delta 𝐫=k_F\delta r\mathrm{sin}\theta `$ where $`\delta r=|\delta 𝐫|`$ such that
$$|DD|T^{\prime \prime }|i|\frac{2\sqrt{2}T_{SD}^2}{i\eta +ϵ_1+ϵ_2}\frac{1}{2\mathrm{\Delta }}_0^\pi 𝑑\theta \mathrm{cos}(k_F\delta r\mathrm{sin}\theta ).$$
(81)
We recover the crossed Andreev contribution of Ref. Recher,
$$|DD|T^{\prime \prime }|i|\frac{1}{\sqrt{2}}\frac{\gamma _S}{i\eta +ϵ_1+ϵ_2}\frac{\mathrm{sin}(k_F\delta r)}{k_F\delta r},$$
(82)
which is still valid in the presence of voltage noise.
## Appendix D Current calculations from crossed Andreev reflection
### D.1 No dissipation
Without dissipation, from Appendices B and C we get
$`I`$ $`=`$ $`{\displaystyle \frac{e}{\mathrm{}}}{\displaystyle \underset{𝐩,𝐪}{}}|T_{DL}|^42\pi \gamma _S^2\left({\displaystyle \frac{\mathrm{sin}(k_F\delta r)}{k_F\delta r}}\right)^2`$
$`\times `$ $`\left|{\displaystyle \frac{1}{(ϵ_𝐩+ϵ_2i\gamma _2/2)(ϵ_𝐪+ϵ_1i\gamma _1/2)}}\right|^2\delta (ϵ_𝐩+ϵ_𝐪).`$
Now, when $`ϵ_𝐩+ϵ_𝐪=0`$ we can rewrite
$`{\displaystyle \frac{1}{(ϵ_𝐩+ϵ_2i\gamma _2/2)(ϵ_𝐪+ϵ_1i\gamma _1/2)}}=`$ (84)
$`{\displaystyle \frac{1}{ϵ_1+ϵ_2i\gamma /2}}\left({\displaystyle \frac{1}{ϵ_𝐩+ϵ_2i\gamma _2/2}}+{\displaystyle \frac{1}{ϵ_𝐩+ϵ_1i\gamma _1/2}}\right),`$
and hence the current turns into
$`I={\displaystyle \underset{𝐩}{}}{\displaystyle \frac{e}{2\mathrm{}}}|T_{DL}|^2{\displaystyle \frac{\gamma \gamma _S^2}{(ϵ_1+ϵ_2)^2+\gamma ^2/4}}\left({\displaystyle \frac{\mathrm{sin}(k_F\delta r)}{k_F\delta r}}\right)^2`$ (85)
$`\times \left|{\displaystyle \frac{1}{ϵ_𝐩+ϵ_2i\gamma _2/2}}+{\displaystyle \frac{1}{ϵ_𝐩+ϵ_1i\gamma _1/2}}\right|^2.`$
Owing to the Breit-Wigner resonance, we can eventually simplify $`ϵ_𝐩ϵ_cϵ_1`$ and thus this results in
$`\left|{\displaystyle \frac{1}{ϵ_𝐩+ϵ_2i\gamma _2/2}}+{\displaystyle \frac{1}{ϵ_𝐩+ϵ_1i\gamma _1/2}}\right|^2={\displaystyle \frac{4\pi }{\gamma _2}}\delta (ϵ_𝐩ϵ_1),`$ (86)
which results in:
$`I={\displaystyle \frac{e}{\mathrm{}}}{\displaystyle \frac{\gamma \gamma _S^2}{(ϵ_1+ϵ_2)^2+\gamma ^2/4}}\left({\displaystyle \frac{\mathrm{sin}(k_F\delta r)}{k_F\delta r}}\right)^2.`$ (87)
This is the formula in Ref. Recher, . Exactly at the resonance condition for the dots where $`ϵ_1=ϵ_2=0`$ the crossed Andreev reflection gives
$`I={\displaystyle \frac{4e\gamma _S^2}{\mathrm{}\gamma }}\left({\displaystyle \frac{\mathrm{sin}(k_F\delta r)}{k_F\delta r}}\right)^2.`$ (88)
It is important to keep in mind that one requires the electrons residing simultaneously on the dots to be in the singlet state configuration so that results of Appendix C can be safely appliedRecher . This is well satisfied when the distance between the dots $`<\xi `$. Indeed, the current carried by the $`S_z=0`$ configuration of the triplet state on the dots would be zero because
$$\underset{\sigma }{}i|[d_2d_1+d_2d_1]ϵ_\sigma d_{1\sigma }^{}d_{2\sigma }^{}|i=0.$$
(89)
In the same spirit, to be non-zero Eq. (C7) assumes that the two injected electrons have anti-parallel spin configurations $`\sigma ^{\prime \prime }=\sigma ^{}`$ and opposite momenta $`𝐤^{\prime \prime }=𝐤^{}`$ which is well-satisfied for $`\delta r<\xi `$. Therefore we can rewrite
$`I[R_1=0]={\displaystyle \frac{4e\gamma _S^2}{\mathrm{}\gamma }}\left({\displaystyle \frac{\mathrm{sin}(k_F\delta r)}{k_F\delta r}}\right)^2e^{2\delta r/\pi \xi }.`$ (90)
### D.2 Weak dissipation on dot 1
For a weak resistance $`R_1`$ so that $`\alpha _1=R_K/R_11`$ the EPR current can be approximated as
$`I`$ $``$ $`{\displaystyle \frac{e}{\mathrm{}}}{\displaystyle \underset{𝐩,𝐪}{}}|T_{DL}|^42\pi \gamma _S^2\left({\displaystyle \frac{\mathrm{sin}(k_F\delta r)}{k_F\delta r}}\right)^2P_1(2\mu _Sϵ_𝐩ϵ_𝐪)`$
$`\times `$ $`\left|{\displaystyle \frac{1}{(ϵ_𝐩+ϵ_2i\gamma _2/2)(ϵ_𝐪+ϵ_1i\gamma _1/2)}}\right|^2.`$
Since we consider that the dots are close to resonance, we have replaced $`P_1(2\mu _Sϵ_1ϵ_2)\delta (2\mu _Sϵ_1ϵ_2)`$ where the function $`P_1(E)`$ (which yields a blatant singularity at $`E=0`$) has been precisely defined in Eq. (26). Now, we concentrate mainly on the effect of the quantum noise on the two particle Breit-Wigner resonance between the dots and the leads. For large $`\alpha _1`$ again the function $`P_1(E)`$ has unambiguously a pronounced singularity at $`E=0`$ implying that $`ϵ_𝐩+ϵ_𝐪0`$ in the tunneling process between the dots and the leads. Using Eq. (D4) this allows us to approximate
$`I`$ $``$ $`{\displaystyle \frac{e}{\mathrm{}}}{\displaystyle \underset{𝐩,𝐪}{}}{\displaystyle \frac{2\pi |T_{DL}|^4\gamma _S^2}{(ϵ_1+ϵ_2)^2+\gamma ^2/4}}\left({\displaystyle \frac{\mathrm{sin}(k_F\delta r)}{k_F\delta r}}\right)^2`$
$`\times `$ $`{\displaystyle \frac{4\pi }{\gamma _2}}\delta (ϵ_𝐩ϵ_1)P_1(2\mu _Sϵ_𝐩ϵ_𝐪).`$
We then converge to
$`I{\displaystyle \frac{e}{\mathrm{}}}{\displaystyle \frac{\gamma \gamma _S^2}{(ϵ_1+ϵ_2)^2+\gamma ^2/4}}\left({\displaystyle \frac{\mathrm{sin}(k_F\delta r)}{k_F\delta r}}\right)^2`$ (93)
$`\times {\displaystyle _{\mu _l}^{ϵ_cϵ_2}}dϵ_𝐪P_1(2\mu _Sϵ_1ϵ_𝐪).`$
We have used the fact that the dots are close to the resonance condition $`ϵ_1+ϵ_2=0`$. Using Eq. (26) we obtain
$`I{\displaystyle \frac{e}{\mathrm{}}}{\displaystyle \frac{\gamma \gamma _S^2}{(ϵ_1+ϵ_2)^2+\gamma ^2/4}}\left({\displaystyle \frac{\mathrm{sin}(k_F\delta r)}{k_F\delta r}}\right)^2`$ (94)
$`{\displaystyle \frac{\mathrm{exp}(2\gamma _e/\alpha _1)}{\mathrm{\Gamma }(1+2/\alpha _1)}}\left({\displaystyle \frac{\pi }{\alpha _1}}\right)^{2/\alpha _1}\left({\displaystyle \frac{2\mu }{E_{c1}}}\right)^{2/\alpha _1}.`$
Hence we can summarize
$`I[\alpha _11]I[R_1=0]{\displaystyle \frac{\mathrm{exp}(2\gamma _e/\alpha _1)}{\mathrm{\Gamma }(1+2/\alpha _1)}}\left({\displaystyle \frac{2\mu \pi }{\alpha _1E_{c1}}}\right)^{2/\alpha _1}.`$ (95)
### D.3 Two independent environments
In the case of two symmetric and moderate environments such that $`R_1,R_2R_K`$, we get
$`I{\displaystyle \frac{e}{\mathrm{}}}{\displaystyle \frac{\gamma \gamma _S^2}{(ϵ_1+ϵ_2)^2+\gamma ^2/4}}\left({\displaystyle \frac{\mathrm{sin}(k_F\delta r)}{k_F\delta r}}\right)^2`$ (96)
$`\times {\displaystyle _{\mu _l}^{ϵ_cϵ_2}}dϵ_𝐪P_{12}(2\mu _Sϵ_1ϵ_𝐪),`$
where the function $`P_{12}(E)`$ has been defined in Eq. (37). Assuming that $`E_{c1}E_{c2}`$ this leads to
$`I[\alpha _l1]I[R_1=0]{\displaystyle \frac{\mathrm{exp}(2\gamma _e/\alpha )}{\mathrm{\Gamma }(1+2/\alpha )}}\left({\displaystyle \frac{2\mu \pi }{\alpha E_{c1}}}\right)^{2/\alpha }.`$ (97)
## Appendix E Current due to direct Andreev processes
Using Eq. (IV), for weak resistances $`\alpha _l1`$ we find that the current $`I^{}=I^{(I)}+I^{(II)}`$ due to direct Andreev processes (when the two electrons tunnel to the same dot) can be easily valued leading to
$`I^{}`$ $``$ $`{\displaystyle \frac{2e}{\mathrm{}}}{\displaystyle \underset{𝐩,𝐩^{}}{}}{\displaystyle \underset{l}{}}\pi {\displaystyle \frac{\gamma _S^2|T_{DL}|^4}{^2}}P_l(2\mu _Sϵ_𝐩ϵ_𝐩^{})`$
$`\times `$ $`\left|{\displaystyle \frac{1}{ϵ_𝐩+ϵ_li\gamma _l/2}}+{\displaystyle \frac{1}{ϵ_𝐩^{}+ϵ_li\gamma _l/2}}\right|^2.`$
We have approximated $`P_l(2\mu _Sϵ_{𝐩^{\prime \prime }}ϵ_l)\delta (2\mu _Sϵ_{𝐩^{\prime \prime }}ϵ_l)`$ and we have explicitly introduced $`^1=1/(\pi \mathrm{\Delta })+1/U`$. In the absence of noise, the energy conservation implies $`ϵ_𝐩+ϵ_𝐩^{}=0ϵ_l`$ and therefore one can exploit
$`\left|{\displaystyle \frac{1}{ϵ_𝐩+ϵ_li\gamma _l/2}}+{\displaystyle \frac{1}{ϵ_𝐩^{}+ϵ_li\gamma _l/2}}\right|^2={\displaystyle \frac{4\pi }{\gamma _l}}\delta (ϵ_𝐩ϵ_l),`$ (99)
resulting in
$`I^{}[R_l=0]`$ $`=`$ $`{\displaystyle \frac{4e}{\mathrm{}}}{\displaystyle \underset{𝐩,𝐩^{}}{}}4\pi ^2{\displaystyle \frac{\gamma _S^2|T_{DL}|^4}{\gamma _l^2}}\delta (ϵ_𝐩+ϵ_𝐩^{})\delta (ϵ_𝐩ϵ_l)`$
$`=`$ $`{\displaystyle \frac{4e}{\mathrm{}}}4\pi ^2\nu _l^2{\displaystyle \frac{\gamma _S^2|T_{DL}|^4}{\gamma _l^2}}={\displaystyle \frac{2e\gamma _S^2\gamma }{\mathrm{}^2}}.`$
Again, we reproduce the result announced in Ref. Recher, . In the limit of weak ohmic resistors $`R_1`$ and $`R_2`$ we estimate
$`I^{}[\alpha _l1]{\displaystyle \frac{I^{}[R_l=0]}{2}}{\displaystyle \underset{l}{}}{\displaystyle _{\mu _l}^{ϵ_cϵ_l}}𝑑ϵ_𝐪P_l(2\mu _Sϵ_lϵ_𝐪)`$ (101)
$`={\displaystyle \frac{I^{}[R_l=0]}{2}}{\displaystyle \underset{l}{}}{\displaystyle \frac{\mathrm{exp}(2\gamma _e/\alpha _l)}{\mathrm{\Gamma }(1+2/\alpha _l)}}\left({\displaystyle \frac{2\mu \pi }{\alpha _lE_{cl}}}\right)^{2/\alpha _l}.`$
Note that for the most probable situation of symmetric environments, the suppression factor $`(2\mu /E_{cl})^{2/\alpha _l}`$ is less considerable than that for the EPR-pair current $`I`$. We argue that this is well justified in this setup because when two spin-entangled electrons tunnel onto the same dot the Coulomb blockade forbids charge-2e transport; this results in a single-particle Breit-Wigner resonance between, say, dot $`l`$, and lead $`l`$ which is less affected by the baths than the two-particle Breit-Wigner resonance. |
warning/0507/gr-qc0507097.html | ar5iv | text | # Frequency and surface dependence of the mechanical loss in fused silica
## I Introduction
As part of the research and development for the LIGO LIGO and TAMA TAMA gravitational wave detectors, we have conducted investigations into the internal friction of fused silica. Displacement of the interferometer’s mirror faces arising from thermal motion of the fused silica test mass mirrors sets a fundamental limit to the detector sensitivity. The frequency distribution of this noise is directly related to the internal friction of the mirror material.
An Advanced LIGO detector has recently been proposed AdvLIGO with better sensitivity than initial LIGO. The Advanced LIGO mirror thermal noise must be as low as possible. Two materials have been under consideration for the mirror substrate: fused silica and single crystal sapphire. To its advantage, sapphire has the higher Young’s modulus and a low bulk mechanical loss ($`\varphi 3\times 10^9`$sapphireQ . However, sapphire also has high thermoelastic noise SapphireThermo .
In the advanced detectors, thermal noise in the mirror coatings makes a significant contribution to the total noise budget in the central frequency region of 30-500 Hz. Discussion on the mechanical loss in the mirror coatings can be found elsewhere GreggCoating ; PennCoating ; Crooks .
Recent measurements on the mechanical loss in fused silica have revealed a dependence on frequency Numata ; AndriThesis and on surface-to-volume ratio Ageev ; Penn ; GreggAndri . This paper combines data from several of these research groups in order to model both of these effects. The frequency dependence of the loss agrees well with results from Weidersich et al. Weidersich . In that work, loss data spanning six decades in frequency is modeled by an asymmetric double-well potential in the bond angle. Together these results provide a more complete picture of the loss in ultrapure glasses and a more physically motivated prediction for the thermal noise in advanced interferometric detectors. It was previously predicted that fused silica’s loss dependence would make it suitable for low frequency detectors (10 – 100 Hz) Riccardo . Indeed, this model’s prediction of a very low mechanical loss in the LIGO frequency regime has motivated the recent selection of fused silica as the Advanced LIGO test mass substrate DownSelect .
## II Theory of Loss in Fused Silica
The thermal noise motion of the mirror surface is related to the internal friction of the substrate by the fluctuation-dissipation theorem Callen52 . The internal friction of very pure fused silica is associated with strained Si-O-Si bonds, where the energy of the bond has minima at two different bond angles, forming an asymmetric double-well potential. Redistribution of the bond angles in response to an applied strain leads to mechanical dissipation, which at audio frequencies has a peak in the cryogenic range 20-60K. Because fused silica is an amorphous material, there is a distribution of potentials which must be inferred from measurements of the dissipation. It can be shown Weidersich that the frequency dependence of the loss should exhibit a power law spectrum with exponent $`k_BT/V_0`$ at low temperatures. Both this power law, with $`V_0/k_B=319K`$, and the distribution of potentials have been measured Weidersich . The power law exponent of a relaxation process cannot exceed 1, and is expected to saturate near 300 K. At room temperature the exponent is 0.76.
At elevated temperatures there is another loss peak arising from a double-well potential associated with the Si-O-Si bond angles. For this peak the bond angle shift and potential barrier are much larger; the double-well of the cryogenic loss peak is a small feature at the minima of this larger potential well. At room temperature, thermal fluctuations allow the bonds to span the cryogenic double-well but not to cross the larger potential barrier, where $`V_0/k_B=3.54\times 10^4K`$Bartenev . The calculated internal friction for this loss peak at audio frequencies and room temperature is utterly negligible compared to other loss mechanisms cited herein.
A separate loss mechanism exists in the surface of the glass. The contribution from the surface loss depends on the mode of the sample. The total energy lost per oscillation in an isotropic sample undergoing slowly decaying vibration, can be described by the integral of the local loss angle,$`\varphi (\stackrel{}{r})`$ with the energy density $`\rho _E(\stackrel{}{r})`$
$$\begin{array}{c}\mathrm{\Delta }E=2\pi _𝒱\rho _E(\stackrel{}{r})\varphi (\stackrel{}{r})d^3r\hfill \end{array}$$
(1)
where $`𝒱`$ is the sample volume. Assuming that the local loss angle is constant and equal to $`\varphi _{\mathrm{bulk}}`$ everywhere except within a distance $`h`$ of the surface, and that the energy density in that surface layer is approximately the energy density at the surface, then the loss can be expressed as AndriThesis
$$\begin{array}{c}\varphi =\varphi _{\mathrm{bulk}}+\mu \alpha _\mathrm{s}\frac{S}{V}\hfill \end{array}$$
(2)
where $`S`$ is the surface area of the sample and $`\mu `$ is a factor of order unity that depends on the mode shape. The surface loss parameter, $`\alpha _\mathrm{s}`$, is typically several picometers for flame polished or flame drawn fused silica but much higher for abrasively polished surfaces.
## III Experimental Method
The measurements at Syracuse University (SU) GreggAndri ; AndriThesis ; Penn ; Ageev were performed on fiber/rod samples with diameters ranging from 0.1 – 8 mm over resonant frequencies less than 5 kHz. The samples were drawn from and left attached to a massive bob of Suprasil heraeus , thus forming a cantilever beam. This bob was welded to a vibration isolating suspension formed by similar silica bobs connected by thin silica fibers. In a vacuum of $`10^6`$ torr, the samples were made resonant by an electrostatic comb exciter, and their position was measured using a shadow sensor.
The measurements at University of Tokyo Numata were performed on cylindrical samples with optically polished surfaces. The diameters and heights were 70 mm and 60 mm, respectively. The samples were annealed in a vacuum furnace. To exclude the support loss, the samples were supported at nodal points of their vibrational modes during the $`Q`$ measurements.
The Caltech measurements were performed on a spare input test mass for the initial LIGO interferometers, a superpolished right cylinder made from Suprasil 312 with a diameter of 25.4 cm and a thickness of 10 cm. It was suspended in a $`10^6`$ Torr vacuum by a loop of polished stainless steel wire greased with lard. The elastic modes of the mass were excited with an electrostatic actuator and the mode amplitude was monitored using a birefringence sensor. Since friction at the wire could reduce $`Q`$, only modes with small motion at the point of wire contact were used in the fit.
## IV Modeling Method
Resonant $`Q`$ measurements from each of the labs were submitted for generating this model of the loss. The measurements spanned several types of fused silica, V/S ratios from 0.03 – 28 mm, and frequency up to $`10^5`$ Hz. The data was first separated by silica type since the loss is known to vary significantly between varieties of fused silica Startin ; Ageev ; Numata . Only Suprasil 2 and 312 had sufficient data to warrant a fit over both frequency and V/S ratio. Characteristics of these samples are listed in Table 1.
We chose a model for the mechanical loss that included terms describing the frequency dependence, the surface loss, and the thermoelastic loss. The loss function took the form:
$`\varphi (f,{\displaystyle \frac{V}{S}})`$ $`=`$ $`\varphi _{\mathrm{surf}}+\varphi _{\mathrm{bulk}}+\varphi _{\mathrm{th}}`$ (3)
$`=`$ $`C_1({\displaystyle \frac{V}{S}})^1+C_2(f/1\mathrm{Hz})^{C_3}+C_4\varphi _{\mathrm{th}}`$ (4)
where $`C_1=\mu \alpha _\mathrm{s}`$ from Eqn. 2. Given that the surface loss term only contributes significantly to the rod (fiber) samples, we have assumed for all samples that $`\mu 2`$ which is appropriate for cylindrical rods. We have also not distinguished the loss angle arising from the Young’s modulus from that due to the shear modulus.
The thermoelastic loss, $`\varphi _{\mathrm{th}}`$, which is negligible in all but the thinnest fiber samples, is described for fibers by:
$$\begin{array}{c}\varphi _{\mathrm{th}}=\frac{Y\alpha ^2T}{\rho C_\mathrm{m}}\frac{2\pi f\tau }{1+(2\pi f\tau )^2}\hfill \\ \\ \tau =(d^2\rho C_\mathrm{m})/(13.55\kappa )\hfill \end{array}$$
(5)
where $`Y`$ is the Young’s modulus, $`\alpha `$ is the coefficient of thermal expansion, $`T`$ is temperature, $`\rho `$ is the density, $`C_\mathrm{m}`$ is the mass specific heat capacity, $`d`$ is the diameter, and $`\kappa `$ is the thermal conductivity. We fit the amplitude of $`\varphi _{\mathrm{th}}`$ to account for small changes in the coefficient of thermal expansion among samples. Variations in the fiber diameter can also slightly alter the shape of the thermoelastic peak. Neither of these effects significantly affect the frequency or surface loss terms.
Measurements of large resonant $`Q`$’s are subject to numerous mechanisms that can greatly reduce the $`Q`$ and few processes that can increase it. These effects produce a distribution in the systematic error that is asymmetric, heavily skewed toward lower $`Q`$, and unique for each experiment. Standard data analysis techniques based on normally distributed error, such as linear least squares (LLS) fitting, are therefore inappropriate for analyzing our full data set. We circumvent this problem by first limiting our data to the best measurement at each $`(f,V/S)`$ point for each sample. A LLS fitting routine is applied with the sample variance approximating the actual variance of the data. This method is commonly used in analyzing mechanical loss measurements where the lowest loss measurement closely approximates actual mechanical loss for a sufficiently large set of measurements GreggAndri ; Penn . The results of the method are displayed in Figure 1 for Suprasil 2 and in Figure 2 for Suprasil 312. The fit coefficients are listed in Table 2. The frequency dependence, $`C_3`$, agrees well with results from Weidersich et al. Weidersich . The thermoelastic amplitude, $`C_4`$, is similar to earlier measurements AndriThesis . Assuming no unforeseen loss mechanisms, the Advanced LIGO test masses ($`V/S40`$ mm) have a predicted loss ($`\varphi (100\mathrm{Hz})4\times 10^{10}`$) that is a several-fold improvement over previous estimates.
## V Implications for Advanced Detectors
The low mechanical loss of silica in the 10 – 1000 Hz bandwidth, coupled with its optical and thermal properties, makes it an attractive material for the optics of next generation interferometric gravitational wave detectors. Fused silica has recently been chosen as the test mass substrate for Advanced LIGO AdvLIGO , which has been approved and recommended for funding by the US National Science Foundation.
If the bulk and surface loss predicted herein can be achieved, the mirror thermal noise in Advanced LIGO with fused silica mirrors will likely be dominated by the coating Levin ; GreggCoating ; PennCoating . The mirror thermal noise contributions to the total Advanced LIGO noise budget are shown in Figure 3. Table 3 shows the predicted sensitivity of Advanced LIGO with silica optics to two possible sources of gravitational waves: binary neutron star inspirals (BNSI) and binary 10 $`M_{}`$ black hole inspirals. Two different scenarios of coating thermal noise are shown: the best measurements to date GreggOttawa and the research goal. The sensitivity goal for a single Advanced LIGO interferometer is to observe BNSI, averaged over sky position and polarization, to a distance of $``$ 200 Mpc. (See Harry HarrySphere for a description of a LIGO range calculation.)
## VI Conclusions
We have shown that the mechanical loss of fused silica can be described by a model that includes surface loss and a frequency dependent bulk loss. The frequency dependent loss, thought to arise from an asymmetric double-well potential of the bond angle, agrees well with earlier measurements Weidersich that spanned six decades in frequency. This improved understanding of the loss indicates that at large geometries and low frequency, fused silica is an excellent material for test masses in advanced interferometric gravitational wave detectors.
## VII Acknowlegments
The authors would like to thank the LIGO laboratory and LIGO Science Collaboration for their support and review of this work. This research was supported by the National Science Foundation under cooperative agreements PHY-9210038 & PHY-0107417 (LIGO laboratory) and awards PHY-9801158 & PHY-0098715 (Caltech), PHY-0245118 & PHY-0355118 (HWS), and PHY-0140335 (SU). |
warning/0507/quant-ph0507158.html | ar5iv | text | # Non-Holonomic Control III : Coherence Protection by the Quantum Zeno Effect and Non-Holonomic Control
## 1 Introduction
The uncontrollable interaction of a quantum system with its environment is responsible for ”quantum errors”which lead to a partial or complete loss of the information initially stored in its quantum state. After Shor’s demonstration that error-correcting schemes exist in quantum computation, a general framework of error-correction has been built upon the formalism of quantum operations. The main contributions concern quantum codes, and particularly the class of stabilizer codes ; other strategies developed suggest the use of ”noiseless quantum codes” or ”decoherence-free subspaces”. All these methods usually demand that errors act independently on different qubits (the independent error model), and make use of the symmetry properties associated with these requirements. This implies that the set of errors to be corrected is restricted to a special subgroup, called the Clifford group. In this paper, we present a protection method which does not make so drastic assumptions.
For low dimensional systems, one can take advantage of the Quantum Zeno Effect allied with basic ideas of coding theory in order to protect the information encoded on a subspace of the total state space. To be more explicit, one frequently repeats a three step sequence comprising coding, decoding and projection, which prevents errors from developing in the system : coding and decoding consist in the application of a unitary matrix and its inverse, which can be achieved through non-holonomic control, and act in such a way that erroneous infinitesimal components are orthogonal to the information subspace ; projection is performed through an irreversible process such as spontaneous emission which clears the unwanted orthogonal increments out.
Promising for small quantum systems, this method becomes however exponentially complex when the number of the qubits involved increases. For large quantum systems, we suggest to employ random coding to reduce the influence of errors consisting in binary interactions. In this context, high dimensionality does not appear as an impediment, but rather as an advantage, since it ’dilutes’ the influence of the errors.
This paper is organized as follows. In the second section, we show that the Quantum Zeno Effect allied with basic ideas of coding theory allows one to protect the quantum information contained in low-dimensional quantum systems. In the third section, we present the random coding technique which can protect the information stored in large systems against the errors resulting from binary interactions.
## 2 Coherence Protection in low-dimensional systems : Quantum Zeno Effect and Non-Holonomic Control
The Quantum Zeno Effect (QZE) appears in a system which is frequently measured in its (necessarily known) initial state: if the time interval between two projective measurements is small enough, the evolution of the system is nearly ”frozen” ; in other words the one-dimensional subspace spanned by the (necessarily known) initial state is protected against the influence of the natural Hamiltonian of the system. We suggest to generalize this effect in order to protect any (unknown) vector of a given multidimensional subspace of the whole Hilbert space. To this end, we propose an information protection scheme, described in the first paragraph of this section , as well as the algorithmic tools which allow its implementation, and which are presented in the second paragraph.
### 2.1 Multidimensional Zeno Effect and Coherence Protection
In this paragraph, we shall first present the multidimensional QZE which allows us to protect an arbitrary subspace of the Hilbert space against the action of a set of given interaction Hamiltonians. Then, we shall take advantage of this phenomenon to protect an information-carrying subsystem of a compound quantum system from the influence of some uncontrolled error-inducing external fields.
Consider a quantum system $`𝒮`$, whose $`N`$-dimensional Hilbert space is denoted by $``$ and whose time-dependent Hamiltonian has the form
$$\widehat{H}(\tau )=\underset{m=1}{\overset{M}{}}f_m(\tau )\widehat{E}_m,$$
(1)
where $`\left\{\widehat{E}_m\right\}_{m=1,\mathrm{},M}`$ are $`M`$ given independent Hermitian matrices on $``$ and $`\left\{f_m(\tau )\right\}_{m=1,\mathrm{},M}`$ are $`M`$ unknown functions of time. The Hamiltonian $`\widehat{H}(\tau )`$ accounts for the errors we want to get rid of. Note that the unperturbed part of the Hamiltonian (1) is assumed to be zero (or proportional to the identity so that one can set it to zero). The standard QZE allows us to nearly ”freeze” the evolution of the system by measuring it frequently enough in its (known) initial state ; in other words, through this effect we can protect the one-dimensional subspace spanned by the initial state of the system from the influence of the error-inducing Hamiltonian (1). In what follows, we generalize this effect so as to protect an arbitrary multidimensional subspace $`𝒞`$ from $`\widehat{H}(\tau )`$.
Any vector $`|\psi `$ of $`𝒞`$ evolves according to $`\widehat{U}(t,t_0)=𝒯\left\{\mathrm{exp}\left[i_{t_0}^t\widehat{H}(\tau )𝑑\tau \right]\right\},`$ where $`𝒯`$ denotes time-ordering, and where we set $`\mathrm{}=1`$. For the QZE to hold, we shall only consider evolution in short time periods, the duration $`\tau _Z`$ of which is so short that the corresponding action of the $`M`$ components of the Hamiltonian (1) is small, *i.e.* $`\left|\widehat{E}_m_t^{t+T}f_m(\tau )𝑑\tau \right|1.`$ We can thus expand
$$\widehat{U}(t+\tau _Z,t)=\widehat{U}_{inf}\widehat{I}i\underset{m=1}{\overset{M}{}}\left(_t^{t+\tau _Z}f_m(\tau )𝑑\tau \right)\widehat{E}_m.$$
(2)
After a Zeno interval $`\tau _Z`$, the initial state $`|\psi `$ is thus transformed into $`|\psi _e=|\psi +|\delta \psi _e`$ where $`|\delta \psi _ei_{m=1}^M\epsilon _m\widehat{E}_m|\psi `$ with $`\epsilon _m=\left(f_m(\tau )𝑑\tau \right).`$
Let us assume that we are physically able to perform the measurement-induced projection onto $`𝒞`$ in the system $`𝒮`$ (see below the discussion of such projections for compound systems comprising an information subsystem and an ancilla). If we straightforwardly apply the standard QZE procedure by merely projecting the state vector $`|\psi _e,`$ resulting from the infinitesimal evolution of the initial state $`|\psi ,`$ onto $`𝒞`$, we get the vector $`|\psi _p`$, which, a priori, differs from $`|\psi `$ (see Fig.1a), since, usually, the vectors $`\widehat{E}_m|\psi `$ and thus the increment vector $`|\delta \psi _e`$ itself are not orthogonal to $`𝒞`$. It is thus clear that we have to adapt the standard Zeno strategy.
To this end, we assume a unitary matrix $`\widehat{C}`$ acting on $``$, which we call the coding matrix, such that the Hermitian operators $`\left\{\widehat{E}_m\right\}_{m=1,\mathrm{},M}`$ act orthogonally on the subspace $`\stackrel{~}{𝒞}=\widehat{C}𝒞`$, called the code space. Let us denote by $`I1`$ the dimension of $`𝒞`$ and by $`\left\{|\gamma _i\right\}_{i=1,\mathrm{},I}`$ one of its orthonormal bases ; $`\left\{|\stackrel{~}{\gamma }_i=\widehat{C}|\gamma _i\right\}_{i=1,\mathrm{},I}`$ will denote one of the orthonormal bases of $`\stackrel{~}{𝒞}`$, the state vectors $`|\stackrel{~}{\gamma }_i`$ being called the codewords. For any pair $`(|\stackrel{~}{\gamma }_s,|\stackrel{~}{\gamma }_t)`$ of codewords and any operator $`\widehat{E}_m\left\{\widehat{E}_m\right\}_{m=1,\mathrm{},M}`$ we have, by the definitions of $`\widehat{C}`$ and $`\stackrel{~}{𝒞}`$
$`\stackrel{~}{\gamma }_t|\stackrel{~}{\gamma }_s`$ $`=\delta _{st}\text{ (orthonormality conditions)}`$ (3)
$`\stackrel{~}{\gamma }_t\left|\widehat{E}_m\right|\stackrel{~}{\gamma }_s`$ $`=0\text{ (orthogonality of the errors)}`$ (4)
Equivalently, for any pair $`(|\psi ,|\chi )`$ of vectors of $`𝒞`$ and for any operator $`\widehat{E}_m\left\{\widehat{E}_m\right\}_{m=1,\mathrm{},M}`$
$$\chi \left|\widehat{C}^{}\widehat{E}_m\widehat{C}\right|\psi =0.$$
(5)
In particular, for any pair $`(|\gamma _s,|\gamma _t)`$ of basis vectors of $`𝒞`$ and for any operator $`\widehat{E}_m\left\{\widehat{E}_m\right\}_{m=1,\mathrm{},M}`$
$$\gamma _t\left|\widehat{C}^{}\widehat{E}_m\widehat{C}\right|\gamma _s=0.$$
(6)
If we apply the coding matrix $`\widehat{C}`$ to the initial state vector $`|\psi `$, before exposing it to the action of the Hamiltonian (1), we obtain the new vector $`|\stackrel{~}{\psi }=\widehat{C}|\psi \stackrel{~}{𝒞}`$ (Fig.1b1,2) which is transformed after a Zeno interval $`\tau _Z`$ into $`|\stackrel{~}{\psi }_e=\widehat{U}_{inf}|\stackrel{~}{\psi }=|\stackrel{~}{\psi }+|\delta \stackrel{~}{\psi }_e,`$ where $`|\delta \stackrel{~}{\psi }_ei_{m=1}^M\epsilon _m\widehat{E}_m|\stackrel{~}{\psi }=i_{m=1}^M\epsilon _m\widehat{E}_m\widehat{C}|\psi `$ (Fig.1b3). Decoding $`|\stackrel{~}{\psi }_e`$ yields the vector $`|\psi _e^{}=\widehat{C}^1|\stackrel{~}{\psi }_e=|\psi +|\delta \psi _e^{}`$ where $`|\delta \psi _e^{}i_{m=1}^M\epsilon _m\widehat{C}^{}\widehat{E}_m\widehat{C}|\psi `$. From Eq.(5) it can be seen that for any vector $`|\chi 𝒞`$, $`\chi |\delta \psi _e^{}=i_{m=1}^M\epsilon _m\chi \left|\widehat{C}^{}\widehat{E}_m\widehat{C}\right|\psi =0`$ which means that $`|\delta \psi _e^{}`$ is orthogonal to $`𝒞`$ (Fig.1b4). A measurement-induced projection onto $`𝒞`$ finally recovers the initial vector $`|\psi `$ with a probability very close to $`1`$ (the error probability is proportional to $`\tau _Z^2`$). If the coding-decoding-projection sequence is frequently repeated, any vector $`|\psi `$ of the subspace $`𝒞`$ can thus be protected from the Hamiltonian (1) for as long as needed.
The multidimensional generalization of the QZE we have just presented allows one to protect any subspace $`𝒞`$ of a Hilbert space $``$ against Hamiltonians of the form (1), and is thus very useful in the context of information protection as we shall see in the following.
Indeed, let us consider an information system $``$ of Hilbert space $`_I`$ and dimensionality $`I`$. This system is subject to a set of $`M`$ error-inducing Hamiltonians $`\left\{\widehat{E}_m\right\}_{m=1,\mathrm{},M}`$ which, for instance, represent the interactions of the system with $`M`$ uncontrolled external classical fields $`f_m(t)`$: we want to get rid of this external influence which is likely to result in the loss of the information stored in the initial state vector $`|\psi _I=_{i=1}^Ic_i|\nu _i`$, where $`\left\{|\nu _i\right\}_{i=1,\mathrm{},I}`$ denotes an orthonormal basis of $`_I`$. To this end, we shall use the multidimensional Zeno Effect. As the multidimensional QZE can only protect a subspace of the whole Hilbert space, we first have to add an $`A`$-dimensional auxiliary system $`𝒜`$ (called ancilla) to our system $``$, so that the information is transferred from $`_I`$ into an $`I`$-dimensional subspace $`𝒞`$ of the $`\left(N=I\times A\right)`$-dimensional Hilbert space $`=_I_A`$ of the compound system $`𝒮=𝒜`$. Furthermore, we shall suppose that all the state vectors of the different Hilbert spaces $`_I`$, $`_A`$ and hence $``$ are degenerate in energy so that the unperturbed part $`\widehat{H}_0`$ of the Hamiltonian can be set to zero as in the first part of this section: the subspace $`𝒞`$ and the information it carries can thus be protected through the multidimensional QZE. Note that $`𝒜`$ and $``$ need not be ”physically separate” systems, but only have to possess independent Hilbert spaces $`_A`$ and $`_I`$.
Let us now return to our problem and first consider the simple case when the ancilla is initially in the pure state $`|\alpha `$. The information initially stored by $`|\psi _I_I`$ is transferred into the factorized state $`|\psi =|\psi _I|\alpha =`$ $`_{i=1}^Ic_i|\nu _i|\alpha =`$ $`_{i=1}^Ic_i|\gamma _i`$ of $`𝒞=_ISpan\left[|\alpha \right]=Span\left[\left\{|\gamma _i=|\nu _i|\alpha \right\}_{i=1,\mathrm{},I}\right]`$. Equivalently, the initial density matrix of the compound system $`𝒮`$ is $`\widehat{\rho }=\left(|\psi _I\psi _I|\right)\left(|\alpha \alpha |\right)`$, which is transformed after the coding step into $`\widehat{\stackrel{~}{\rho }}=\widehat{C}^{}\widehat{\rho }\widehat{C}`$ ; at the end of the action of the errors it is transformed into $`\widehat{\stackrel{~}{\rho }}_e=\widehat{U}_{inf}^{}\widehat{C}^{}\widehat{\rho }\widehat{C}\widehat{U}_{inf}`$ ; finally, after decoding, it takes the form $`\widehat{\rho }_e=\widehat{C}\widehat{U}_{inf}^{}\widehat{C}^{}\widehat{\rho }\widehat{C}\widehat{U}_{inf}\widehat{C}^{}`$. In this setting, the projection onto $`𝒞`$ can be simply achieved by measuring the ancilla in its initial state $`|\alpha `$. As $`\tau _Z`$ is very short, the state of the ancilla evolves just a little within a Zeno interval : the probability of detecting it in its initial state $`|\alpha `$, and thus of projecting the state of the compound system onto $`𝒞`$ is thus very close to $`1`$. After projection, we trace out the ancilla to obtain the final reduced density matrix $`\widehat{\rho }_I^{}=\alpha \left|\widehat{C}\widehat{U}_{inf}^{}\widehat{C}^{}\widehat{\rho }\widehat{C}\widehat{U}_{inf}\widehat{C}^{}\right|\alpha `$ for the information system $``$; in the same way, one can calculate the initial reduced density matrix is $`\widehat{\rho }_I=|\psi _I\psi _I|.`$ The variation $`\delta \widehat{\rho }_I=`$ $`\widehat{\rho }_I^{}\widehat{\rho }_I`$ of the information-space density matrix during the whole process can then be expressed as the commutator
$$\delta \widehat{\rho }_I=i[\underset{m=1}{\overset{M}{}}f_m(\tau )𝑑\tau \alpha \left|\widehat{C}^{}\widehat{E}_m\widehat{C}\right|\alpha ,\widehat{\rho }_I],$$
from which we infer that $`\widehat{\rho }_I`$ satisfies the equation $`i\frac{d\widehat{\rho }_I}{dt}=[\widehat{h}_e,\widehat{\rho }_I],`$ where $`\widehat{h}_e=_{m=1}^Mf_m\alpha \left|\widehat{C}^{}\widehat{E}_m\widehat{C}\right|\alpha `$ is an effective Hamiltonian which is determined by the error-inducing Hamiltonians transformed by the coding and decoding and projected onto the initial state of the ancilla. From Eq.(5) one can see that $`\widehat{h}_e=0`$ and hence $`\widehat{\rho }_I`$ remains constant in time: as long as we repeat the coding-decoding-ancilla resetting sequence, the information initially stored in $``$ is protected.
It is not always feasible to directly measure the ancilla independently from the information system ; in other words, it is sometimes impossible to perform a projection onto disentangled subspaces of $``$ of the form $`_ISpan\left[|\alpha \right]`$ : in some cases, one can only project onto entangled subspaces of the total Hilbert space $``$. In such a case the information initially stored in the vector $`|\psi _I=_{i=1}^Ic_i|\nu _i_I`$ is transferred into an entangled state of $``$ and $`𝒜`$ of the form $`|\psi =`$ $`_{i=1}^Ic_i|\gamma _i`$ where the $`I`$ vectors $`|\gamma _i`$ ($`i=1,\mathrm{},I`$) which form an orthonormal basis of the information-carrying subspace $`𝒞`$, are not factorized as earlier but are in general entangled states. Nevertheless the same method as before can be used in that case to protect information, albeit in a different subspace $`𝒞`$.
To conclude this description of our method, let us now return to conditions (3) and (4) imposed on the codewords $`\left\{|\stackrel{~}{\gamma }_i,i=1,\mathrm{},I\right\}`$ and make two remarks about them:
A. We can establish a useful relation between the dimension $`A`$ of the ancilla and the number $`M`$ of correctable error Hamiltonians. The set of the $`I`$ codewords can indeed be seen as a collection of $`2I\times N=2I^2A`$ real numbers on which $`2I^2+2MI^2=2I^2(1+M)`$ constraints, directly derived from Eqs.(3,4), are imposed. As the number of free parameters must be larger than the number of constraints, we necessarily have $`2I^2A2I^2(1+M)`$, or equivalently
$$A1M.$$
(7)
This condition, called the ”Hamming bound”, gives an upper-bound on the number of independent error-inducing Hamiltonians that our method can correct simultaneously.
B. We may compare our correctability conditions (4) with the more general conditions of standard quantum error-correction
$`(|\stackrel{~}{\gamma }_s,|\stackrel{~}{\gamma }_t)`$ $`\stackrel{~}{𝒞}^2,\text{ }(\widehat{𝐄}_k,\widehat{𝐄}_l)\left\{\widehat{𝐄}_j\left(\left\{\widehat{E}_m\right\}\right)\right\},`$
$`\stackrel{~}{\gamma }_t\left|\widehat{𝐄}_k^{}\widehat{𝐄}_l\right|\stackrel{~}{\gamma }_s`$ $`=\alpha _{kl}\stackrel{~}{\gamma }_t|\stackrel{~}{\gamma }_s`$ (8)
which ensure the existence of a code space that is completely protected against the error-inducing Hamiltonians $`\widehat{E}_m`$. Here $`\alpha _{kl}`$ are complex numbers, and the set $`\left\{\widehat{E}_m\right\}`$ of Hermitian operators $`\widehat{E}_m`$ generates a group $`𝒢\left(\left\{\widehat{E}_m\right\}\right)`$ of all possible error-induced evolutions (2). By $`\left\{\widehat{𝐄}_j\left(\left\{\widehat{E}_m\right\}\right)\right\}`$ we denote a complete basis set of operators which spans the space of evolution operators $`\widehat{U}`$ and allows one to represent any $`\widehat{U}`$ as a linear combination of the basis operators $`\widehat{𝐄}_j`$. In addition to all the $`\widehat{E}_m`$, the variety of all linear combinations of $`\widehat{𝐄}_j`$ includes also many other operators given by commutators of all orders in $`\widehat{E}_m`$ entering the expansion of $`\widehat{U}`$ for long times. The condition (8) is therefore much more restrictive than Eq.(4). Moreover, even for two generic matrices $`\widehat{E}_m`$, the basis $`\left\{\widehat{𝐄}_j\right\}`$ spans the entire Hilbert space $``$, yielding $`\stackrel{~}{𝒞}=\mathrm{}`$. Only if the set $`\left\{\widehat{E}_m\right\}`$ belongs to an extraspecial algebra restricting the error evolution operators $`\widehat{U}`$ to a subgroup $`𝒢\left(\left\{\widehat{E}_m\right\}\right)𝒢_U\left(\right)`$ of the full unitary group in $``$, a non-trivial code space $`\stackrel{~}{𝒞}`$ may exist. The Zeno effect is the only way to suppress loss of coherence if it is not the case.
### 2.2 The code space and the coding matrix
It is sometimes possible to build the code space $`\stackrel{~}{𝒞}`$ explicitly from physical considerations. However, in general, we need an algorithm to calculate the code basis $`\left\{|\stackrel{~}{\gamma }_i\right\}_{i=1,\mathrm{},I}`$ or, equivalently, the coding matrix $`\widehat{C}`$. In this paragraph, we shall first describe this algorithm, then, we shall show that the non-holonomic control technique can be employed to implement the coding matrix physically. We will also provide an algorithm which achieves the appropriate control.
Let us first make a remark which will be useful. Consider a vector $`|𝖢`$ of some Hilbert space and a matrix $`\widehat{𝖤}`$ on this space. From the vector $`|𝖢`$ we want to calculate a vector $`|\stackrel{~}{𝖢}`$ such that $`\stackrel{~}{𝖢}\left|\widehat{𝖤}\right|\stackrel{~}{𝖢}=0`$. If $`𝖢\left|\widehat{𝖤}\right|𝖢=0`$, then $`|𝖢=|\stackrel{~}{𝖢}`$ and the function $`f_{\stackrel{~}{𝖢}}(\lambda )=|\stackrel{~}{𝖢}+\lambda \widehat{𝖤}|\stackrel{~}{𝖢}^2,`$ depending on the c-number $`\lambda `$, is minimal for $`\lambda =0`$: indeed
$`|\stackrel{~}{𝖢}+\lambda \widehat{𝖤}|\stackrel{~}{𝖢}^2`$ $`=\stackrel{~}{𝖢}|\stackrel{~}{𝖢}+\lambda \stackrel{~}{𝖢}\left|\widehat{𝖤}\right|\stackrel{~}{𝖢}+\lambda ^{}\stackrel{~}{𝖢}\left|\widehat{𝖤}^{}\right|\stackrel{~}{𝖢}+\left|\lambda \right|^2\stackrel{~}{𝖢}\left|\widehat{𝖤}^{}\widehat{𝖤}\right|\stackrel{~}{𝖢}`$
$`=1+\left|\lambda \right|^2\stackrel{~}{𝖢}\left|\widehat{𝖤}^{}\widehat{𝖤}\right|\stackrel{~}{𝖢},`$
and as $`\stackrel{~}{𝖢}\left|\widehat{𝖤}^{}\widehat{𝖤}\right|\stackrel{~}{𝖢}0`$, $`f_{\stackrel{~}{𝖢}}(\lambda )`$ is minimal for $`\left|\lambda \right|=0`$, that is $`\lambda =0`$. But, if $`𝖢\left|\widehat{𝖤}\right|𝖢0`$, we can apply the following iterative method: we minimize $`f_𝖢(\lambda )`$ with respect to $`\lambda `$, then we set $`|𝖢^{}=|𝖢+\frac{\lambda }{2}\widehat{𝖤}|𝖢`$ and take $`\frac{|𝖢^{}}{\sqrt{𝖢^{}|𝖢^{}}}`$ as our new $`|𝖢`$ ; repeating this sequence finally leads $`|\stackrel{~}{𝖢}`$, such that $`\stackrel{~}{𝖢}\left|\widehat{𝖤}\right|\stackrel{~}{𝖢}=0`$.
Let us now return to our problem and show how to use the previous remark. We want to find $`I`$ vectors $`|\stackrel{~}{\gamma }_i`$ which meet the conditions (3) and (4) ; equivalently, we look for an orthonormal basis in which all the matrices $`\widehat{𝖤}_k`$ have their $`I\times I`$ upper left blocks equal to zero. To solve this problem, we propose to transform our initial problem in such a way that it can be dealt with by the iterative algorithm presented in the previous paragraph. Let us combine the $`I`$ vectors $`|\stackrel{~}{\gamma }_i`$ into a $`\left(N\times I\right)`$ ”supervector”
$$|\stackrel{~}{𝖢}=\left(\begin{array}{c}|\stackrel{~}{\gamma }_1\\ \mathrm{}\\ |\stackrel{~}{\gamma }_I\end{array}\right).$$
Then let us build $`E=\left(\frac{I(I1)}{2}+M\frac{I(I+1)}{2}\right)`$ different $`\left(N\times I\right)\times \left(N\times I\right)`$-dimensional super-matrices $`\widehat{𝖤}_k`$ in the following way: we consider them as made of $`I^2`$ blocks of dimension $`N\times N`$ and we successively fill each of these blocks with the different Hamiltonians $`\widehat{E}_m`$ or the identity matrix $`\widehat{I}`$ or $`0`$. To be more explicit, the first $`\frac{I(I1)}{2}`$ matrices are built by simply placing the $`N\times N`$ identity matrix in each of the $`\frac{I(I1)}{2}`$ blocks situated above the diagonal. In the last $`\frac{MI(I+1)}{2}`$ ones, the $`M`$ operators $`\widehat{E}_m`$ are successively placed in each of the $`\frac{I(I+1)}{2}`$ blocks on and above the diagonal. One can thus reformulate the conditions (3) as follows: for $`1k\frac{I(I1)}{2}`$, $`\stackrel{~}{𝖢}\left|\widehat{𝖤}_k\right|\stackrel{~}{𝖢}=0.`$ This form does not take the normalization condition into account, which will be imposed in a different manner. Similarly, the conditions (4) are translated into the following form: for $`\frac{I(I1)}{2}+1k\frac{I(I1)}{2}+\frac{MI(I+1)}{2}`$, $`\stackrel{~}{𝖢}\left|\widehat{𝖤}_k\right|\stackrel{~}{𝖢}=0.`$ This new problem can be handled by the same kind of iterative algorithm as in our preliminary remark.
First, we randomly pick a supervector $`|𝖢_0`$ which will be the starting point of the first step: we normalize this vector by imposing to each of its $`I`$ components to have norm = $`\frac{1}{I}`$. If one of the components of $`|𝖢_0`$ is non normalizable, that is equals zero, we pick up a new random supervector $`|𝖢_0`$ as a starting point.
Then, we minimize $`F_{𝖢_0}(\lambda _1^{(0)},\lambda _2^{(0)},\mathrm{},\lambda _E^{(0)})=_{k=1}^E|𝖢_0+\lambda _k^{(0)}\widehat{𝖤}_k|𝖢_0^2`$ with respect to the $`E`$ c-numbers $`\lambda _k^{(0)}`$, and we calculate $`|\mathrm{\Delta }𝖢_0=_k\lambda _k^{(0)}\widehat{𝖤}_k|𝖢_0`$ and $`|𝖢_0^{}=|𝖢_0+\frac{1}{2}|\mathrm{\Delta }𝖢_0`$. We normalize $`|𝖢_0^{}`$ by requiring each of its $`I`$ components to have the norm = $`\frac{1}{I}`$, and take the result of this operation as our new starting point $`|𝖢_1`$. If one of the components of $`|𝖢_0^{}`$ is non normalizable, that is equals zero, we pick up a new random supervector $`|𝖢_0`$ as a starting point.
We repeat this sequence of operations as long as needed and obtain the desired vector $`|\stackrel{~}{𝖢}`$ asymptotically. Practically, as our algorithm converges quickly, the number of iterations needed is small.
The coding matrix $`\widehat{C}`$ is a complex unitary operator on the Hilbert space of the compound system $`𝒮=𝒜`$. We have just shown how to calculate the codewords, which actually form the first $`I`$ columns of $`\widehat{C}`$, but one can wonder how to implement it physically. This question can be solved by the non-holonomic control technique.
Indeed, we can directly apply the results of the first of our articles to our coding problem in the following way: first, we find the codewords $`\left\{|\stackrel{~}{\gamma }_i,i=1,\mathrm{},I\right\}`$ by the iterative algorithm we have previously presented, then we complete the set of $`I`$ vectors $`\left\{|\stackrel{~}{\gamma }_i,i=1,\mathrm{},I\right\}`$ with $`\left(NI\right)`$ vectors $`\left\{|\stackrel{~}{\gamma }_j,j=I+1,\mathrm{},N\right\}`$ to form an orthonormal basis of $``$, we build the coding matrix by taking the vectors $`\left\{|\stackrel{~}{\gamma }_i,i=1,\mathrm{},N\right\}`$ as columns of $`\widehat{C}`$, and finally we calculate the $`N^2`$ appropriate timings $`\left\{\tau _i\right\}`$ such that
$$\widehat{U}(\tau _1,\mathrm{},\tau _{N^2})=\mathrm{exp}\left(i\widehat{H}_a\tau _{N^2}\right)\mathrm{}\mathrm{exp}\left(i\widehat{H}_b\tau _1\right)=\widehat{C}$$
through the complete control algorithm we have previously presented (we suppose we have two distinct perturbations $`\widehat{P}_a`$ and $`\widehat{P}_b`$ such that the system is completely controllable). Note that we assume $`\widehat{H}_0=0`$, hence $`\widehat{H}_a=\widehat{P}_a`$ and $`\widehat{H}_b=\widehat{P}_b`$.
Actually, this straightforward procedure provides a lot of useless work. Indeed, most of the information contained in the coding matrix is irrelevant and the $`N^2`$ real parameters of $`\widehat{C}`$ do not all have to be controlled exactly: the number $`n_C`$ of necessary control parameters $`\left\{\tau _i\right\}`$ is actually much less than $`N^2`$, as we shall see now.
The coding matrix is characterized by the relations (6). The problem of control thus reduces to finding $`n_C`$ timings $`\tau _i`$, forming the time-vector $`\stackrel{}{\tau }=\left(\begin{array}{c}\tau _1\\ \mathrm{}\\ \tau _{n_C}\end{array}\right)`$, such that the non-holonomic evolution matrix
$$\widehat{U}\left(\stackrel{}{\tau }\right)=\mathrm{exp}\left(i\widehat{H}_a\tau _{n_C}\right)\mathrm{}\mathrm{exp}\left(i\widehat{H}_a\tau _1\right)$$
checks (6). The number $`n_C`$ of control parameters must exceed the number of independent constraints which is clearly $`MI^2`$, that is $`n_CMI^2`$. Thus the number of necessary control parameters appears to be much smaller than $`N^2`$. So we need a new algorithm which achieves a partial and less expensive control of the evolution operator of the system.
The algorithm we shall use to calculate the appropriate control timings $`\tau _i`$ mixes the iterative algorithm presented at the beginning of this paragraph and the non-holonomic control technique. If we introduce the $`\left(N\times I\right)\times \left(N\times I\right)`$-dimensional block-diagonal matrix
$$\widehat{𝖴}\left(\stackrel{}{\tau }\right)=\left(\begin{array}{cccc}\widehat{U}\left(\stackrel{}{\tau }\right)& 0& \mathrm{}& 0\\ 0& \widehat{U}\left(\stackrel{}{\tau }\right)& \mathrm{}& 0\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ 0& 0& \mathrm{}& \widehat{U}\left(\stackrel{}{\tau }\right)\end{array}\right)$$
and the $`\left(N\times I\right)`$-dimensional supervector
$$|𝖢=\left(\begin{array}{c}|\gamma _1\\ \mathrm{}\\ |\gamma _I\end{array}\right)$$
composed of the coordinates of the $`I`$ basis vectors of $`𝒞`$, we can set our problem of control into the following equivalent form: we look for a time-vector $`\stackrel{}{\tau }`$ such that
$$k,\text{ }𝖢\left|\widehat{𝖴}^{}\left(\stackrel{}{\tau }\right)\widehat{𝖤}_k\widehat{𝖴}\left(\stackrel{}{\tau }\right)\right|𝖢=0$$
(9)
where the matrices $`\left\{\widehat{𝖤}_k\right\}_{k=1,\mathrm{},E}`$ denote $`E`$ different matrices of dimension $`\left(N\times I\right)\times \left(N\times I\right)`$ which have been introduced in the beginning of this section. The idea of our algorithm is to take the super vector $`|𝖢_0=\widehat{𝖴}\left(\stackrel{}{\tau }_0\right)|𝖢`$, where $`\stackrel{}{\tau }_0`$ is a random time-vector, as the starting point for an elementary step of the iterative algorithm and look for the small time increment $`\stackrel{}{d\tau }_0`$ such that $`\widehat{𝖴}\left(\stackrel{}{\tau }_0+\stackrel{}{d\tau }_0\right)|𝖢`$ follows the direction provided by the result $`|𝖢_0+|\mathrm{\Delta }𝖢_0`$ of the iterative algorithm. The repetition of this sequence finally yields $`\stackrel{}{\tau }=\stackrel{}{\tau }_0+\stackrel{}{d\tau }_0+\stackrel{}{d\tau }_1+\mathrm{}`$ which meets Eq.(9).
Let us now describe the algorithm in more detail. First, we randomly pick a set of timings $`\tau _{0,i}`$ in a ”realistic range”, dictated by the system under consideration: in particular, control-pulse timings have to be much shorter than the typical lifetime of the system but much longer than the typical response delay required by the experiment. Then we calculate $`|\mathrm{\Delta }𝖢_0=_k\lambda _k\widehat{𝖤}_k|𝖢_0`$ by minimizing the same function $`F_{𝖢_0}(\lambda _1^{\left(0\right)},\lambda _2^{\left(0\right)},\mathrm{},\lambda _E^{\left(0\right)})`$ as in the algorithm presented at the beginning of this section. At that point, we look for the small increment $`\stackrel{}{d\tau }_0`$ of the time-vector $`\stackrel{}{\tau }_0`$ such that
$`k,𝖢|({\displaystyle \frac{\widehat{𝖴}^{}}{\stackrel{}{\tau }}}\left(\stackrel{}{\tau }_0\right).\stackrel{}{d\tau }_0)\widehat{𝖤}_k\widehat{𝖴}\left(\stackrel{}{\tau }_0\right)+\widehat{𝖴}^{}\left(\stackrel{}{\tau }_0\right)\widehat{𝖤}_k({\displaystyle \frac{\widehat{𝖴}}{\stackrel{}{\tau }}}\left(\stackrel{}{\tau }_0\right)..\stackrel{}{d\tau }_0)|𝖢`$
$`={\displaystyle \frac{𝖢_0+\frac{1}{2}\mathrm{\Delta }𝖢_0\left|\widehat{𝖤}_k\right|𝖢_0+\frac{1}{2}\mathrm{\Delta }𝖢_0𝖢_0\left|\widehat{𝖤}_k\right|𝖢_0}{𝖢_0+\frac{1}{2}\mathrm{\Delta }𝖢_0|𝖢_0+\frac{1}{2}\mathrm{\Delta }𝖢_0}}.`$ (10)
It should be noticed that we do not consider the error super-matrices $`\widehat{𝖤}_k`$ corresponding to orthonormality conditions: in other words, we just take matrices $`\left\{\widehat{𝖤}_k\right\}_{k[\frac{I(I1)}{2}+1,\frac{I(I1)}{2}+\frac{MI(I+1)}{2}]}`$ into account. Thus we deal with $`\frac{MI(I+1)}{2}`$ complex equations. This set of equations can be reduced to the real linear system
$$\widehat{S}\left(\stackrel{}{\tau }_0\right)\stackrel{}{d\tau }_0=\stackrel{}{W}\left(|\mathrm{\Delta }𝖢_0\right)$$
(11)
where $`\widehat{S}\left(\stackrel{}{\tau }_0\right)`$ and $`\stackrel{}{W}\left(|\mathrm{\Delta }𝖢_0\right)`$ are respectively an $`MI^2\times n_C`$ real matrix and a $`MI^2`$-dimensional real vector. We obtained Eq.(11) by splitting the set of $`\frac{MI(I+1)}{2}`$ complex equations (10) into two sets of $`\frac{MI(I+1)}{2}`$ real equations, and rejecting those which are trivial ($`0=0`$) or redundant. Though straightforward, the explicit expressions of the different elements of $`\widehat{S}`$ and $`\stackrel{}{W}`$ involve many indices and are so unpleasant that we prefer not to reproduce them here.
The linear system we have just found is, a priori, rectangular $`(MI^2\times n_C)`$, but actually the number $`n_C`$ has not been fixed yet. Previously, we stated that $`n_CMI^2`$: we could be tempted to set $`n_C=MI^2`$ so as to obtain a square system, easily solvable by standard techniques of linear algebra. Yet we will proceed in a slightly different way. We set $`n_C=MI^2+\delta n>MI^2`$, where $`\delta n`$ is an integer of order $`1`$, then we randomly pick $`MI^2`$ timings $`t_i`$ among the $`n_C`$ which will be considered as free parameters, whereas the other $`\delta n`$ ones will be regarded as frozen. The new version of Eqs(11) is now clearly a square system, which yields the $`MI^2`$-dimensional increment $`\stackrel{}{d\tau }_0`$, corresponding to the $`MI^2`$ free varying timings, which we complete with $`\delta n`$ zeros, corresponding to the frozen timings, into a $`n_C`$-dimensional vector $`\stackrel{}{d\tau }_0`$. Then we set $`\stackrel{}{\tau }_1^\alpha `$ $`=\stackrel{}{\tau }_0+`$ $`\alpha `$ $`\stackrel{}{d\tau }_0`$ where $`\alpha `$ is a convergence coefficient and calculate the test function $`G\left(\stackrel{}{t}\right)=_k\left|𝖢\left|\widehat{𝖴}^{}\left(\stackrel{}{\tau }\right)\widehat{𝖤}_k\widehat{𝖴}\left(\stackrel{}{\tau }\right)\right|𝖢\right|^2`$ in $`\stackrel{}{\tau }=\stackrel{}{\tau }_1^\alpha `$ for different values of $`\alpha [0,1]`$. If we find an $`\alpha _1`$ such that $`G\left(\stackrel{}{\tau }_1^\alpha \right)<G\left(\stackrel{}{\tau }_0\right)`$, we take $`\stackrel{}{\tau }_1\stackrel{}{\tau }_1^{\alpha _1}`$ as our new time-vector, and keep the same free-varying timings. If we cannot find such an $`\alpha _1`$, this means we are situated in a local minimum of $`G`$ ; then we set $`\stackrel{}{\tau }_1\stackrel{}{\tau }_0`$ and pick a new set of free varying parameters. This rotation procedure among control parameters allows us to avoid possible local minima of the test function $`G`$ we want to cancel.
We repeat this sequence of operations as long as needed and obtain the desired vector $`\stackrel{}{\tau }`$ asymptotically. Practically, as our algorithm converges quickly, the number of iterations needed is small.
We have not said anything about decoding so far. If the signs of the two Hamiltonians $`\widehat{H}_a=\widehat{P}_a`$ and $`\widehat{H}_b=\widehat{P}_b`$ can be reversed by altering the control field parameters, decoding amounts to reversing $`\widehat{P}_a`$ and $`\widehat{P}_b`$ and applying the same control timing sequence backwards. Otherwise, one must use the general non-holonomic control technique, involving $`N^2`$ control parameters, to find timings which realize $`\widehat{C}^1`$.
## 3 Zeno Coherence Protection by Random Coding
The protection method we presented in the previous section seems promising for relatively low-dimensional systems. However, for large systems, it is likely to lead to very heavy computations and long control sequences. To deal with such systems, we therefore propose to employ an approach inspired by classical random coding : in this method, linear codes $`[n,k,d]`$ are produced, in which $`k`$-bit words are encoded as randomly chosen $`n`$-bit sequences. The minimal Hamming distance $`d`$ between any two codewords approaches the Hamming bound $`dnk`$ as $`n\mathrm{}`$. In this section, we show how to extend the idea of random coding to the quantum case.
Strong mixing or entanglement occur in the phase or Hilbert spaces of complex classical or quantum systems, respectively, and can, in principle, be used for random coding. However, in practice, in the classical case, the dynamics of such systems is not reversible, which makes subsequent decoding hardly possible. By contrast, the dynamics of multi-dimensional quantum systems can be reversed, when the underlying physical mechanism is simple enough : the spin-echo phenomenon is a typical example of this topic. High dimensionality of simple quantum systems is responsible for the massive parallel computing capacity of quantum computers. Therefore, we have to find an operation which produces strong mixing in the multidimensional Hilbert space, and which can be inverted in a simple way : the non-holonomic control suits perfectly this purpose. The essential requirement for the protection scheme we propose to apply is that error-inducing interactions are relatively simple, resulting, for instance, either from a binary qubit interaction or, generally speaking, from a few-particle coupling.
To combine strong mixing with irreversibility, we assume that we have a quantum system with a large number of separate energy levels and with two simple interactions which satisfy the bracket generation condition and can therefore be employed for the non-holonomic control. In such a system, one can encode quantum data into strongly mixed states by straightforwardly applying a unitary transformation $`\widehat{C}`$, the decoding procedure being achieved by the inverse transformation $`\widehat{C}^1`$. Encoding the data into many levels allows us to strongly reduce the error-rate, as will be shown. In turn, by applying the Zeno effect we can restore the slightly corrupted data back to its original value with high probability.
To encode the data in a high dimensional Hilbert space of $`n`$ qubits, we introduce $`nk`$ ancillary qubits in addition to the required number $`k`$ of data qubits. This results in an increase in the number of possible errors, which depends polynomially on $`n`$, but the error rate decreases at will because the infinitesimal errors are semi-orthogonal to the encoded data to a degree exponential in $`nk`$. The degree of semi-orthogonality reflects the error correction efficiency of the coding. Efficiency requires a precise and careful choice of the code for coding in minimal dimensions, but in high dimensions it is naturally achieved by random coding. In mathematical terms, the method relies on the fact, that in a multidimensional space, a pair of randomly chosen vectors are almost orthogonal with high probability. In physical terms, random coding is equivalent to strong mixing, or full population of all energy levels, which can be reached by the non-holonomic control with a number of interaction switchings depending only polynomially on $`n`$. Thus the random coding approach of the present section complements our earlier non-holonomic Zeno coherence loss suppression scheme, which requires exponential effort to find and achieve an exact code, and is thus efficient only for low-dimensional systems.
The essence of random coding can be elucidated as follows. Consider an $`n`$-qubit system, comprising a $`k`$-qubit information carrying subsystem and an $`(nk)`$-qubit ancilla. In the $`2^n`$-dimensional Hilbert space of the system, the error-inducing Hamiltonians $`\widehat{E}_m`$ corresponding to a few-particle interaction are represented by sparse matrices in the computational basis, which is composed of all the possible tensor products of individual qubits eigenstates. In this basis, the number of non-zero matrix elements is indeed polynomial in $`n`$ : for instance, for binary interactions, this number scales as $`n^2`$ . The coding-decoding transformation $`\widehat{C}^{}\widehat{E}_m\widehat{C}`$, where $`\widehat{C}`$ stands for a generic unitary matrix, ’smoothes’ all the matrix elements by mixing them : finally, all these elements are of the same order of magnitude, which is, up to a polynomial factor, $`2^n`$-times smaller than the typical value of non-zero matrix elements in the computational basis before the coding-decoding sequence. Error matrices elements are thus exponentially reduced : the error suppression condition Eq.(6) for the projection of these matrices onto the subspace of the initial state of the ancilla is not fulfilled exactly any longer : the projection differs from zero, but its norm $`\gamma _t\left|\widehat{C}^{}\widehat{E}_m\widehat{C}\right|\gamma _s2^{kn}`$ remains small, and decreases exponentially with the size of the ancilla. The error-accumulation rate is thus inhibited by a factor of the order of the ancilla Hilbert space dimension.
Note, that this mechanism is efficient only when the generic coding matrix $`\widehat{C}`$ can be achieved by a small number of switchings, such that the coding procedure does not take exponentially long time. Fortunately, the coding matrix takes a generic form after a relatively small number of switchings, which scales linearly with the number of qubits $`n`$ (see ). Moreover, if the signs of the two interactions employed for the non-holonomic control can be inverted, the decoding operation can be performed at the same level of complexity as the coding procedure, by straightforwardly changing the signs of the interactions and inverting the timing sequence in which these interactions are applied. The main restriction to practical implementation of the random coding protection scheme arises from that one necessarily has to remain in the Zeno regime: the measurement time, which does not depend on the size of the system, has to be much shorter than the coherence loss timescale, which decreases, although polynomially, with the size of the system.
## 4 Conclusion
The non-holonomic control allied with the Quantum Zeno Effect can be employed to overcome the influence of the environment on the quantum system considered. On the one hand, in the case of low-dimensional systems, we showed that quantum information can be protected by frequently repeating the cycle coding-infinitesimal errors-decoding-projection : coding and decoding correspond to a unitary transformation of the Hilbert space and its inverse, respectively, which are determined in such a way that the projection onto the initial information carrying subspace of the state resulting from coding-infinitesimal errors-decoding yields the initial state vector. All the needed algorithmic tools have been presented. On the other hand, for high-dimensional systems, one can adapt the classical idea of random coding to the quantum case : the basic principle is to use non-holonomic control to impose generic and easily reversible unitary evolutions to the system in order to encode/decode the information ; this procedure ”dilutes” the influence of the errors in the large Hilbert space and then decreases their influence. |
warning/0507/astro-ph0507535.html | ar5iv | text | # New light on the Initial Mass Function of the Galactic Halo Globular Clusters
## 1 Introduction
Globular clusters are the oldest bound stellar systems in our Galaxy. Their study provides therefore valuable information about early Galactic evolution. In this respect, a major problem is that we do not know whether what we presently observe is still representative of the initial conditions and, thus, a fossil imprint of the formation process, or whether the initial conditions have been wiped out by a 13 Gyr long evolution within the tidal fields of what is now the Milky Way. Modelling the dynamical evolution of the Galactic globular cluster system is thus of great interest as it helps us to go back in time to the earliest stages of the cluster system and to disentangle the formation and evolutionary fingerprints (see, e.g., Okazaki & Tosa 1995, Baumgardt 1998, Vesperini 1998, Fall & Zhang 2001). In spite of numerous efforts however, the shape of the initial distribution in mass of the halo globular clusters has remained ill-determined so far.
In our Galaxy, the mass function of the halo clusters <sup>1</sup><sup>1</sup>1In what follows, we adopt the nomenclature of McLaughlin & Pudritz (1996). We call luminosity/mass spectrum the number of objects per linear luminosity/mass interval, $`\mathrm{d}N/\mathrm{d}L`$ or $`\mathrm{d}N/\mathrm{d}m`$, while we refer to the luminosity/mass function to describe the number of objects per logarithmic luminosity/mass interval, $`\mathrm{d}N/\mathrm{dlog}L`$ or $`\mathrm{d}N/\mathrm{dlog}m`$. (the number of clusters per logarithmic mass interval, which is proportional to the number of objects per magnitude unit) is bell-shaped and usually fitted with a gaussian (e.g., Ashman & Zepf 1998). However, the underlying mass spectrum (i.e., the number of objects per linear mass interval) is well described by a two-index power-law, with exponents $`2`$ and $`0.2`$ above and below $`1.5\times 10^5`$ M, respectively (McLaughlin 1994). The peak of the gaussian function in fact coincides with the cluster mass at which the slope of the mass spectrum changes. The slope of the high mass regime is reminiscent of what is observed in interacting and merging galaxies where systems of young massive clusters show well defined power-law luminosity spectra (i.e., $`\mathrm{d}NL^\alpha \mathrm{d}L`$) with $`\alpha `$ in the range 1.8 to 2 (see, e.g., Whitmore & Schweizer 1995, Whitmore et al. 2002). This observational fact gave rise to the popular idea that the initial mass spectrum of the halo clusters may have been a power-law with a similar slope.
However, the luminosity spectrum $`\mathrm{d}N/\mathrm{d}L`$ constitutes a faithfull mirror of the underlying mass spectrum $`\mathrm{d}N/\mathrm{d}m`$ only if any variations of the mass-to-light ratio from cluster to cluster are small. This is true for any cluster system whose stellar initial mass function is invariant and whose cluster age range is small. This is roughly the case for the Galactic halo globular clusters. Their visual mass-to-light ratio ranges from $``$1 to $``$4 (see Pryor & Meylan 1993 and Parmentier & Gilmore 2001, their Fig. 1), partly reflecting different cluster dynamical evolutions. This may not be true for cluster systems formed in ongoing or recent starbursts. Their formation duration may be a significant fraction of the system’s median age and, thus, age spread effects among the young star cluster population may not be negligible. Being an age related quantity, the cluster integrated mass-to-light ratio can no longer be considered as a constant even for an invariant stellar initial mass function. For instance, a system of young clusters with an age range of 3 to 300 million years shows variations in the cluster mass-to-light ratio as large as a factor of 20 (see, e.g., Bruzual & Charlot 2003). As a result, the shape of the luminosity spectrum may differ substantially from the shape of the mass spectrum (Meurer et al. 1995, Fritze v. Alvensleben 1998, 1999). To unveil the mass spectrum of young clusters therefore requires estimates of their individual mass-to-light ratio with, e.g., spectral synthesis models. However, the results have remained unconclusive so far: the young cluster system of the nearby starburst galaxy NGC 3310 displays a gaussian mass function (and, thus, a two-index power-law mass spectrum; de Grijs et al. 2003) while the young massive clusters located in the Antenna merger (NGC 4038/39) are distributed in mass according to a pure power-law with slope $`2`$ (Zhang & Fall 1999). These contrasting results may be interpreted as evidence against the universality of the globular cluster initial mass function.
Whether the Galactic halo cluster system actually started with a power-law mass spectrum or with a gaussian mass function similar to the current one has remained a very puzzling issue since both evolve into the presently observed lognormal mass function. As for the initial power-law mass spectrum, this form gets severely depleted below a turnover of $`1.5\times 10^5`$M due to the preferential removal of low-mass clusters through evaporation and tidal disruption, leading after a Hubble-time of evolution to the current two-index power-law mass spectrum (e.g., Okazaki & Tosa 1995, Baumgardt 1998). On the other hand, Vesperini (1998) demonstrated that the presently observed gaussian mass function represents a state of quasi-equilibrium, that is, the gaussian shape and its associated parameters (mean and standard deviation) are preserved during the entire evolution through a subtle balance between disruption of clusters and evolution of the masses of those which survive, even though a significant fraction of the clusters is destroyed. The globular cluster initial mass distribution may thus have been a gaussian mass function or it may have been a power-law mass spectrum. As a result, the low-mass regime cannot be recovered by studying the temporal evolution of the mass spectrum only.
Along with the issue of the mass spectrum, several studies dedicated to the dynamical evolution of the Galactic globular cluster system also address the evolution with time of the cluster radial number density profile $`n(D)`$ (i.e., the number of clusters per unit volume in space as a function of Galactocentric distance $`D`$). Yet, the temporal evolution of the radial mass density profile $`\rho (D)`$ (i.e., the spatial distribution around the Galactic centre of the halo cluster system mass) is not discussed (e.g., Baumgardt 1998, Vesperini 1998). Considering the case of a power-law mass spectrum with slope $`2`$ and extending down to 500 M, McLaughlin (1999) emphasized the relative robustness of mass-related quantities with respect to number-related quantities (see his equations 4-7). The fact that the total mass and the mass density profile of a cluster system are better indicators of the initial conditions than are the number of clusters and the number density profile will be further illustrated in our Section 3.
In contrast to the robustness of the mass density profile, the disparity in the results derived by Vesperini (1998) and Baumgardt (1998) regarding the temporal evolution of the number density profile is more puzzling. The presently observed spatial distribution of the Galactic halo clusters is centrally concentrated with the density varying as $`D^{3.5}`$ (Zinn 1985), except in the inner 3-4 kpc where the distribution flattens to something closer to an $`D^2`$ dependence. As this flattening is likely to result (at least partly) from the shorter time-scale for cluster disruption at smaller galactocentric distance, Vesperini (1998) assumes that the initial number density profile scales as $`D^{3.5}`$ through the whole halo extent. Evolving such a system up to an age of 15 Gyr, he concludes that the initial steepness of the distribution is preserved, except in the inner Galactic regions where the greater efficiency of cluster destruction processes flattens the profile, in agreement with the presently observed one. On the other hand, Baumgardt ’s (1998) results suggest that an initial slope of $`3.5`$ (i.e., a steepness similar to what it is now) is ruled out as this one leads to a final spatial distribution significantly flatter than what is observed. Accordingly, the initial distribution must have been steeper and an initial slope of $`4.5`$ is required to match the present spatial distribution. It is worth pointing out however that, while Baumgardt (1998) builds on a power-law mass spectrum with a slope $`\alpha =2`$ and probing down to 1000 M (i.e., a choice inspired by the luminosity spectrum of young massive clusters in starbursts and mergers), Vesperini (1998) investigates the case of the equilibrium gaussian mass function (i.e., an initial mass function similar to the present one because its shape remains well-preserved). Thus, their divergence about the initial steepness of the number density profile is likely to arise from a different choice for the initial cluster mass function.
This also suggests that the mass and number density profiles evolve differently with time. While the former remains a reliable mirror of what it initially was, irrespective of the initial distribution in mass of the clusters (as we will confirm in Section 3), the steepness of the latter after a Hubble-time of evolution may depend sensitively on the initial mass spectrum. This leads us to consider the possibility that a comparison between evolved and observed profiles, in terms of mass and in terms of number, may help shed light on the initial mass range and/or the initial distribution in mass of the Galactic halo clusters. Actually, if we assume that, soon after their formation, globular clusters show the same mass range and the same mass spectrum, irrespective of their galactocentric distance, then the mass and number density profiles are initially identical in shape. On the other hand, the presently observed mass and number density profiles of the halo cluster system also show shapes that match each other (see, e.g., McLaughlin 1999). Therefore, the robustness of the mass density profile combined with (1) the uniformity in shape for the presently observed mass and number density profiles and (2) the assumed uniformity in shape for the initial ones implies that the initial mass spectrum of globular clusters was such that the number density profile has been preserved in spite of evolution in the radially-dependent tidal fields of the Milky Way.
The outline of the paper is as follows. In Section 2, we build the radial mass and number density profiles of the Old Halo cluster system and we compare their shape. In Section 3, we briefly describe the analytic model as obtained by Vesperini & Heggie (1997) which enables us to evolve the number and mass density profiles of a cluster system. We evolve various globular cluster systems with different initial mass functions and different initial spatial distributions. In Section 4, we then compare the model outcomes with the observations in order to derive new constraints on the initial mass function of the Old Halo clusters, as well as on their initial spatial distribution. Finally, we present our conclusions in Section 5.
## 2 Radial Profiles of the Old Halo
Comparing the slopes of the mass density profile and of the number density profile, which is the core of this paper, is equivalent to studying the radial variation of the mean cluster mass (and, in fact, gives the physical reason for such a variation, namely, the possibly different responses to the dynamical evolution of the number-related and mass-related quantities). The observed radial variations of the mean cluster mass in the Milky Way, M31 and M87 have already been tackled by Gnedin (1997). Sorting the clusters in two equal size parts with respect to their galactocentric distance, he detected the existence of “statistically significant differences between the inner and outer populations, inner clusters being on average brighter than the outer clusters, as would be expected if the inner population had been depleted by tidal shocks.” The dynamical interpretation of the results is presented in a complementary study by Ostriker & Gnedin (1997). However, his Milky Way sample does not discriminate among the different cluster populations, that is, disc/bulge, Old Halo and Younger Halo (see below). Due to obvious differences with respect to their age, their metallicity and their evolution history (i.e., accreted clusters have not been constantly subjected to the Galactic potential since the time of their formation), the interpretation of differences in the mean luminosity of the inner and outer populations may not be that straightforward. As for M87, the conclusions of Gnedin (1997) and Ostriker & Gnedin (1997) have been revisited by Kundu et al. (1999) who detected no significant variations of the cluster luminosity function turnover with respect to the projected galactocentric distance. Kundu et al. (1999) claimed that “the apparent brightening observed by Gnedin is probably due to undercompensation of completeness corrections in the inner regions, where the dimmer clusters are harder to detect against the strong galaxy background.” In contrast to Kundu et al.’s (1999) results, Barmby et al. (2001) found that the mean luminosity of the inner clusters of M31 is significantly brighter than that of the outer ones. However, they cautionned that variations driven by differences in cluster metallicity, age and stellar initial mass function may also be important and must be accounted for properly before using any luminosity function variation as a probe to differences in the globular cluster mass function.
In the present study, we restrict our attention to the Galactic globular cluster system. In the Galaxy only can we clearly discriminate among the different cluster populations regarding their age and evolutionary history. Specifically, we aim at constraining the initial mass spectrum of the first generation globular clusters which formed within the gravitational potential well of the Galaxy. Hence, we do not consider the more metal-rich, presumably second-generation, bulge/disc globular clusters (Zinn 1985). The halo cluster system itself hosts two distinct populations of clusters, the so-called Old Halo and Younger Halo (Van den Bergh 1993, Zinn 1993). The Old Halo globular clusters might have been formed ’in situ’. In contrast, younger halo globular clusters are suspected of having been accreted. Regardless of their formation history, the Old Halo globular clusters form a coherent and well-defined group, well-suited to an analysis of their properties: thus, we consider the Old Halo subgroup only. Lists were initially compiled by Lee et al. (1994) and Da Costa & Armandroff (1995) and have recently been updated by Mackey & Gilmore (2004). We note in passing that, although coeval with the inner halo (Harris et al., 1997), we have not included NGC 2419 in our Old Halo sample. This cluster is located at a galactocentric distance of order 90 kpc and is thus unlikely to belong to the main body of the Galaxy. Moreover, van den Bergh & Mackey (2004) show that NGC 2419 and $`\omega `$ Cen on the one hand, and the other halo clusters on the other hand, are at different locii in a half-light radius vs absolute visual magnitude diagram. This thus suggests that, as for $`\omega `$ Cen (which we also exclude from our sample), NGC 2419 might be the tidally stripped core of a former dwarf spheroidal galaxy.
To build the mass and number density profiles, our source for the galactocentric distances $`D`$ and the absolute visual magnitudes M<sub>v</sub> is the McMaster database compiled and maintained by Harris (1996, updated February 2003). Cluster absolute visual magnitudes have been turned into luminous mass estimates by assuming a constant mass-to-light ratio M/L<sub>v</sub>=2.35 (i.e., the average of the mass-to-light ratios of the halo clusters for which Pryor & Meylan (1993) obtained dynamical mass estimates). The Old Halo mass and number profiles are derived by binning the data with two different bin sizes: $`\mathrm{\Delta }\mathrm{log}D=0.1\mathrm{and}0.2`$ ($`D`$ is in kpc), corresponding to 16 and 8 points, respectively. As for the size of the error bars, a Poissonian error on the number of clusters in each bin is combined with a fixed error on the mass-to-light ratio. In fact, not all globular clusters show the same mass-to-light ratio, the standard deviation in the Pryor & Meylan (1993) compilation being of order $`\sigma _{\mathrm{Log}(M/L_v)}=0.17`$.
As already mentioned, the observed radial distribution of halo clusters obeys $`D^{3.5}`$ except in the innner 3-4 kpc where the distribution gets shallower. As a result, it is often parametrized by a power law with a core (see equation 1). Previous fits having been obtained for either the whole Galactic globular cluster system (e.g., Djorgovski & Meylan 1994) or the whole halo system (e.g., McLaughlin 1999), we now consider the Old Halo subsystem only. Using a Levenberg-Marquardt algorithm (Press et al. 1992), we fit the Old Halo number density profile with:
$$\mathrm{Log}n(D)=\mathrm{Log}n_0\gamma \mathrm{Log}\left(1+\frac{D}{D_c}\right).$$
(1)
The values obtained for the slope $`\gamma `$ and the core $`D_c`$ are presented in the left part of Table 1. For each fit, we also give the $`\chi ^2`$ and the incomplete gamma function $`Q(\nu /2,\chi ^2/2)`$ ($`\nu `$ is the number of degrees of freedom) which provides a quantitative measure for the goodness-of-fit of the model <sup>2</sup><sup>2</sup>2We remind the reader that: a $`Q`$ value of 0.1 or larger indicates a satisfactory agreement between the model and the data; if $`Q0.001`$, the fit may be acceptable if, e.g., the errors have been moderately underestimated; if $`Q<0.001`$, the model can be called into question (Press et al. 1992).. Imposing a slope $`\gamma `$ of –3.5 or –4, as found by Djorgovski & Meylan (1994) for the whole globular cluster system, provides a good fit to the number density profile of the Old Halo cluster system as well. Keeping all three parameters free, we obtain a steeper slope ($`\gamma 4.5`$) coupled with a larger core. The core reflects the flattening of the spatial distribution at small galactocentric distances, presumably owing to the greater efficiency of disruptive processes. Ignoring this core region and focusing on the Old Halo clusters located at galactocentric distances $`3`$ kpc, that is, where memory of the initial conditions has perhaps been better preserved, we find that both the mass and the number density profiles of the Old Halo are well-approximated by pure power-laws with slope $`3.5`$ (see Table 2). The steepness of the Old Halo spatial distribution is thus similar to that of the whole halo (Zinn 1985). While the mass and number density profiles show very similar steepness at distances larger than 3 kpc, their overall shapes are also very similar. Fitting the Old Halo mass density profile with the same functions as used for the number density profile (i.e., equation 1 and the ($`\gamma `$, $`D_c`$) couples listed in Table 1) provides equally good values of the incomplete gamma function (see the last column of Table 1). Therefore, the number and the mass density profiles of the Old Halo are indistinguishable through the whole extent of the halo.
## 3 Evolved Radial Density Profiles of Globular Cluster Systems
The previous section shows that the currently observed mass density and number density profiles of the Old Halo are identical in shape. If we assume that the globular cluster formation mechanism produced the same mass range and the same mass spectrum for the clusters, irrespective of their galactocentric distance, then the initial mass and number density profiles were identical in shape as well. If the mass density profile has been preserved (and we show in this section that it is actually the case), all together, these facts imply that the number density profile itself has remained fairly unaltered during evolution in the tidal field of the Milky Way. In what follows, we evolve various putative globular cluster systems, considering different combinations of initial mass spectra and initial spatial distributions. We then investigate in which case(s) has the number density profile been reasonably preserved. We also compare in a least-squares sense the evolved spatial distributions to the observed ones that we have derived in Section 2.
To evolve the radial mass and number density profiles of a cluster system from the time of its formation up to an age of 15 Gyr, we adopt the analytic formula of Vesperini & Heggie (1997) which supplies at any time $`t`$ the mass $`m`$ of a star cluster with initial mass $`m_i`$ which is moving along a circular orbit perpendicular to the galactic disc at a galactocentric distance $`D`$. The assumption of circular orbits is clearly a simplifying one since it implies that the time variations of the tidal field for clusters on elliptical orbits are not allowed for in our calculations. Yet, the system of relevance here is the Old Halo. This shows less extreme kinematics than the Younger Halo group of clusters, making this assumption less critical than if we have dealt with the whole halo system. As for the influence of the cluster orbit inclination with respect to the Galactic disc, Murali & Weinberg (1997) found that, although low-inclination halo clusters evolve more rapidly than high-inclination ones, the differences are not extreme. Furthermore, our sample excluding disc clusters, the assumption of high inclination orbits is a reasonable one. The simulations of Vesperini & Heggie (1997) were designed in the frame of a host galaxy modelled as a simple isothermal sphere with a constant circular velocity. This actually constitutes a reasonable assumption for the Old Halo system whose radial extent is 1-40 kpc, that is, where the mass profile of the Milky Way (i.e, the total Galactic mass enclosed within a radius $`D`$) grows linearly with the galactocentric distance (Harris 2001). We consider the effect of non-circular orbits in more detail in Section 4, below.
The relations describing the temporal evolution of the mass of a globular cluster have been obtained by fitting the results of a large set of N-body simulations in which Vesperini & Heggie (1997) take into account the effects of stellar evolution as well as two-body relaxation, which leads to evaporation through the cluster tidal boundary. Disc shocking can also be included (see below). In order to take into account dynamical friction, globular clusters whose time-scale of orbital decay (see, e.g., Binney & Tremaine 1987) is smaller than $`t`$ are removed from the cluster system at that time (see Vesperini 1998, his Section 2, for further details). It is important to note a specific assumption underlying the validity of this analysis. The large-scale Galactic gravitational potential is assumed constant, that is, this model considers the evolution of a globular cluster system only after it has been assembled in a time-independent Galaxy. That is the physical basis for a restriction to the system of Old Halo globular clusters.
The temporal evolution of the mass of a cluster orbiting at constant galactocentric distance $`D`$ is found to follow:
$$\frac{m(t)}{m_i}=1\frac{\mathrm{\Delta }m_{st,ev}}{m_i}\frac{0.828}{F_{cw}}t.$$
(2)
$`\mathrm{\Delta }m_{st,ev}/m_i`$ is the fraction of cluster mass lost due to stellar evolution (18 per cent in this particular model). The time $`t`$ is expressed in units of 1 Myr and $`F_{cw}`$, a quantity proportional to the initial relaxation time, is defined as:
$$F_{cw}=\frac{m_i\times D}{\mathrm{ln}N},$$
(3)
where $`m_i`$ and $`D`$ are in units of 1 M and 1 kpc, respectively, and $`N`$ is the initial number of stars in the cluster. To take into account disc shocking, the factor 0.828/$`F_{cw}`$ is merely replaced by $`\lambda `$ as defined by equation (3) of Vesperini (1998).
We have distributed 20,000 clusters following various radial and mass distributions. Four different initial distributions in cluster mass have been considered:
* a gaussian mass function $`\mathrm{d}N/\mathrm{dlog}m`$ with parameters equal to those of the equilibrium mass function of Vesperini (1998), that is, a mean $`\mathrm{log}m_0=5.03`$ and a standard deviation $`\sigma =0.66`$;
* three power-law mass spectra $`\mathrm{d}N/\mathrm{d}m`$, each with a slope of –1.9 and different lower mass limits, namely, 1E3, 1E4 and 1E5 M. The value of the slope agrees with what is obtained for the high mass regime of the Galactic halo cluster system, that is, a slope of around –1.8 to –2 (see, e.g., Ashman & Zepf 1998).
As for the last low-mass cut-off, Fall & Zhang (2001) indeed show that the cluster mass spectrum may have started with a truncation at mass of order 1E5 M, the low-mass tail of the present mass distribution resulting from the evaporation of initially more massive clusters located close to the Galactic centre. This illustrates one more time the difficulty of deducing the initial distribution in mass of the globular clusters on the sole ground of evolving it with time. Three different initial mass spectra manage to evolve into the presently observed bell-shaped mass function: (1) a power-law probing down to 1000 M or (2) truncated at $`10^5`$ M as well as (3) a two-index power-law with a turnover around $`10^5`$ M. We thus are ignorant of the contribution of the low-mass objects to the initial population of globular clusters. Did they constitute the overwhelming contribution by number (case 1), were they missing (case 2), or were they present in limited number (case 3) ?
Regarding the initial number density profile (equivalent in shape to the initial mass density profile following our assumption of a unique mass range and a unique mass spectrum through the Old Halo extent), we assume two different functional forms:
* $`n(D)\rho (D)D^{3.5}`$, as suggested by our fits to the observed spatial distributions for globular clusters located beyond 3 kpc (see Table 2);
* $`n(D)\rho (D)D^{4.5}`$ (Baumgardt 1998).
We have thus considered a total of 16 different cases, combining the 4 different initial mass spectra/functions with the 2 different initial spatial-densities and including or not disc shocking.
Table 3 lists the fraction of surviving clusters ($`F_N`$) and the ratio of the final to the initial mass in clusters ($`F_M`$) for each case. For a given initial space-density, we note the relative homogeneity of the mass fractions $`F_M`$ in spite of widely different initial mass spectra, the extreme values differing by a factor of 2 at most. Also, as noted by previous studies (Vesperini 1998, Baumgardt 1998, McLaughlin 1999), the evaporation and the disruption of globular clusters cannot account for the overwhelming contribution of field stars to the luminous Galactic halo. The mass of the Old Halo cluster subsystem is $`2\times 10^7`$ M, that is, about two per cent only of the stellar halo mass ($`10^9`$ M, Freeman & Bland-Hawthorn 2002). On the other hand, the largest destruction rates in Table 3 (F$`{}_{N}{}^{}=`$ 0.01-0.02) correspond to a total mass in survivors of order 15 to 20 per cent of the initial cluster system mass. Hence, even in this extreme case, disrupted and evaporated clusters account for 10 to 13 per cent of the stellar halo mass only.
In contrast to the rather limited dispersion shown by the mass fraction, $`F_M`$, the fraction of survivors in the number density distribution $`F_N`$ is characterized by a scatter as large as an order of magnitude for both initial spatial distributions. Larger destruction rates are of course achieved in case of a power-law initial mass spectrum probing down to 1000 M as this one favours low-mass easily disrupted clusters. The fraction of survivors is also smaller in case of a steeper initial spatial distribution (i.e., D<sup>-4.5</sup> instead of D<sup>-3.5</sup>) since a larger fraction of globulars are then located at smaller galactocentric distances where destruction processes proceed on a shorter time-scale.
The initial (dotted curves) and evolved (solid curves: no disc shocking, dashed-dotted curves: with disc shocking) mass and number density profiles are displayed in the left and right panels, respectively, of Figs. 2 ($`n_{init}D^{3.5}`$) and 3 ($`n_{init}D^{4.5}`$). Examination of the left panels shows that the mass density profile $`\rho (D)`$ has remained fairly well preserved during evolution for a Hubble-time <sup>3</sup><sup>3</sup>3It should be kept in mind that Vesperini & Heggie ’s (1997) model includes a gaseous mass loss of 18 per cent fed by stellar winds. This decrease in the cluster mass is independent of galactocentric distance and, therefore, does not alter the shape of the mass density profile even though this one is reduced by the same amount.. The strongest change takes place for a power-law down to 1000 M. Even in that case however, the slope of the initial distribution is reduced by $`0.3`$ only. The initial steepness of the cluster system mass density profile is thus robust, irrespective of the initial mass spectrum. In contrast, the evolution with time of the number density profile is much dependent on initial conditions, as revealed by the right panels of Figs. 2 and 3. While the initial and evolved number density profiles are similar in case of a gaussian mass function (i.e., an initial mass spectrum depleted in low-mass clusters with respect to a power-law extending down to low-mass objects) or in case of a power-law mass spectrum truncated at 10<sup>5</sup> M, the slope of the profile is reduced by $``$ 1 if the initial power-law mass spectrum goes down to 1000 M. Equation 2 shows that no globular clusters with masses lower than 4000 M are able to survive within the outer bound of the Old Halo (say, $`D40`$ kpc). Although these clusters represent a small fraction of the cluster system mass ($`15`$ per cent), by number, they account for $`70`$ per cent of the clusters. Hence, the destruction of these low-mass objects leads to a significantly altered number density profile while leaving the mass profile almost unaffected. Obviously, mass-related quantities are more reliable indicators of the initial conditions than are number-related quantities, especially in the case of an initial mass spectrum favouring low-mass clusters. In addition to possible slope alterations, in all cases, the spatial distribution develops a core, that is, it gets flattened in the galactocentric range 1 to 2 kpc, illustrating the larger efficiency of the disruptive processes closer to the Galactic centre.
## 4 What density profiles tell us about the globular cluster initial mass function
In Figs. 2 and 3, the evolved spatial distributions (solid and dashed-dotted curves) have been vertically shifted in order to provide the best least-squares agreement with the presently observed distributions. The initial space-densities (dotted curves) have then been shifted accordingly. Table 4 shows the corresponding $`\chi ^2`$ and incomplete gamma functions $`Q(\nu /2,\chi ^2/2)`$ (i.e., the goodness-of-fit).
Considering the case of an initial spatial distribution scaling as D<sup>-3.5</sup>, we note the excellent agreement between the evolved and observed mass density profiles, irrespective of the initial cluster mass spectrum. This is an expected result since the Old Halo mass density profile also scales as D<sup>-3.5</sup> (see Table 2) and since mass profiles are fairly well-preserved in spite of the destruction of a significant fraction of the initial cluster population. Including disc shocking leads to a larger decrease in mass density at small galactocentric distance and to an even better modelling of the Old Halo profile. We note in passing that some values of the significance as measured by the incomplete gamma function are puzzlingly high, that is, larger than 0.8. It is worth keeping in mind however that the initial spatial distribution of relevance here (i.e., $`n_{init}D^{3.5}`$) is not that much a model but, instead, derives from a fit to the observed mass density profile for D$`3`$ kpc (see Table 2). These high $`Q`$ values thus illustrate furthermore that the evolved mass density profile mirrors faithfully the initial one.
As for the evolved number density profile, this agrees with the data only if the globular cluster system started with either a gaussian mass function or a power-law mass spectrum truncated at 10<sup>5</sup> M. In case of a power-law mass spectrum probing down to low cluster mass, the $`Q`$ values get extremely low, disproving such initial mass spectra. This discrepancy results from the sharp change experienced by the slope of the number density profile (panel \[h\] in Fig. 2), leading to a present steepness significantly shallower ($`2.5`$) than the observed one ($`3.5`$, see Table 2). As a result, examination of Fig. 2 and Table 4 shows that the comparison of the evolved (i.e., modelled) and observed radial density profiles, in terms of mass as well as in terms of number, enables us to constrain the initial mass spectrum of globular clusters. The $`Q`$ values strongly favour either an initial gaussian mass function or a power-law with slope of order $`1.9`$ and truncated at large mass, around 10<sup>5</sup> M, that is, an initial mass distribution which is somehow depleted in low-mass objects. In this respect, our results confirm those achieved by Vesperini (1998).
We now consider the steeper initial spatial distribution, namely $`n_{init}D^{4.5}`$. As already suggested by Baumgardt (1998), there is an excellent agreement between the halo number density profile and its modelled counterpart in the case of a power-law extending down to 1000 M, especially if disc shocking is taken into account. Owing to the large contribution of low-mass clusters, the initially steep profile is turned into a shallower one, thus matching the halo $`3.5`$ slope. However, we caution that the evolved mass density profile does not fit its Old Halo counterpart convincingly in any case. The best match is obtained for a globular cluster system with a power-law initial mass spectrum probing down to low-mass (see Table 4). Even though such a possibility cannot be ruled out firmly, the goodness-of-fit is very marginal ($`Q0.001`$). This case is therefore much less likely than the one we have previously discussed, namely, a globular cluster system whose initial spatial steepness is similar to the present one.
Baumgardt (1998) himself noted the oddity of this result as it implies a discrepancy between the initial slope of the globular cluster spatial distribution on the one hand and the steepness of the stellar halo density profile on the other hand. Indeed, the space-density of halo RR Lyrae (Suntzeff, Kinman & Kraft 1991) as well as of halo blue horizontal branch stars (Kinman, Suntzeff & Kraft 1994) falls off as $`D^{3.5}`$. Baumgardt (1998) suggests that this discrepancy results from a varying star cluster to field star formation efficiency. Considering the $`Q`$ values listed in Table 4, a much safer conclusion may be that the globular cluster system started with a space-density scaling as $`D^{3.5}`$ coupled with either a gaussian mass function or a power-law truncated at 10<sup>5</sup> M.
While our simulations start with many thousands of clusters, the survival rates $`F_N`$ quoted in Table 3 indicate that the initial number of clusters is on the order of that today in case of a gaussian initial mass function or in case of a power-law mass spectrum truncated at $`10^5`$ M. As for the gaussian initial mass function, we have checked that an initial total number of clusters of 200 only does not introduce a significant scatter in the evolved mass density profile with respect to the size of the error bars. Using Vesperini & Heggie ’s (1997) model with disc shocking and considering a slope $`\gamma =3.5`$ for the initial radial distribution, the incomplete gamma function for the evolved mass density profile ($`\mathrm{\Delta }logD=0.2`$) ranges from 0.3 up to 0.9 (10 random samplings of the gaussian cluster IMF). We note that these results are obtained in case of a gaussian truncated at 1.5E6 M in order to avoid the presence of clusters significantly more massive than is observed today. Actually, inspection of the luminous mass estimates of the halo globular clusters shows $`10^6`$ M to be an upper limit to the present-day globular cluster mass. More massive clusters do actually exist, e.g., $`\omega `$ Cen, M54 (=Sagittarius core), NGC 2419 and a few disc clusters, but none of them are relevant to the present study. Running the same simulations in case of a non-truncated gaussian, the incomplete gamma function tends to get smaller (i.e., down to 0.002 in one case). This is due to the occasional sampling of very massive (i.e.,$`>`$ 1.5E6 M) clusters, giving rise to an upwards scatter in the outermost least-populated bins of the mass density profile.
As for the case of a power-law mass spectrum truncated at $`10^5`$ M, we have evolved a cluster system initially comprising 150 clusters only. The incomplete gamma function for the evolved mass density profile ranges from 0.01 to 0.7 (10 random realisations, of which 8 give $`Q>0.1`$). This robustness, despite a limited number of clusters, is due to the narrow mass range associated to the truncation at large cluster mass. <sup>4</sup><sup>4</sup>4In case of a power-law initial mass spectrum extending down to 1000 M, the globular cluster system initially hosted several thousands of clusters and the small-number sampling thus does not constitute an issue. Should the initial number of clusters be on the order of that today (say, 100), the evolved population would contain just a very few survivors, possibly none at all since the selection of low-mass clusters would be favoured owing to the small-number sampling. Actually, following a Hubble-time of evolution, the number of survivors is in the range () if the slope of the initial radial distribution is $`3.5`$ ($`4.5`$) (10 random realisations), leading to a discrepancy with the present-day cluster number.
In fact, our results are reminiscent of those obtained by Vesperini (2000, 2001) in the case of globular cluster systems hosted by elliptical galaxies. Investigating the case of a power-law initial mass spectrum extending down to low-mass combined with a coreless $`D^{3.5}`$ initial number density profile, Vesperini (2001) noted that the evolutionary processes produce a significant dependence of the average cluster mass on the galactocentric distance in the sense that clusters located in the inner galactic regions are more massive. This dependence is equivalent to the difference between the slopes of the evolved mass and number density profiles highlighted in the bottom panels of Fig. 2. Vesperini ’s (2001) result contrasts with several observational studies that fail to find a significant radial gradient of the average cluster mass within elliptical galaxies (e.g., M87, Kundu et al. 1999). Conversely, Vesperini (2000) emphasized that, in the case of a gaussian initial mass function, the radial gradient of the average cluster mass is weak and consistent with the observations. Equivalently, the evolved mass and number density profiles match each other, as shown by the top panels of Fig. 2.
The marked evolution of the number density profile with respect to the mass density profile in the case of a power-law mass spectrum arises because, as Vesperini (1998, 2000, 2001), we have assumed that clusters are orbiting at constant galactocentric distance. Should a substantial radial mixing take place, a more radially uniform mass function may emerge. We have thus tested whether our results are significantly affected if the orbital eccentricity is $`e=0.5`$. According to Baumgardt & Makino (2003), the lifetime of a cluster on an orbit with eccentricity $`e`$ is decreased by a factor $`(1e)`$ with respect to a cluster on a circular orbit with a radius similar to the apogalactic radius of the eccentric orbit. Considering the two extreme (and key) cases, namely, the gaussian \[G\] and power-law \[PL3\] initial cluster mass distributions, combined with a coreless $`D^{3.5}`$ initial number density profile, we have run additional simulations in which the quantity $`F_{cw}`$ (equation 3) is halved (i.e., multiplied by $`1e`$). The corresponding evolved mass and number density profiles are in remarkable agreement with those derived under the assumption of circular orbits. The only significant difference is the destruction of all the clusters confined within 1.2 kpc from the Galactic centre if $`e=0.5`$. The comparison between the predicted profiles and the Old Halo distributions leads to (considering $`\mathrm{\Delta }logD=0.2`$):
1. for a Gaussian initial mass function \[G\]: Q=0.03 \[$`\rho (D)`$\] and Q=0.003 \[$`n(D)`$\] for the mass and number density profiles, respectively;
2. for a power-law initial mass spectrum \[PL3\]: Q=0.005 \[$`\rho (D)`$\] and Q=10<sup>-16</sup> \[$`n(D)`$\].
As for the gaussian mass function, the incomplete gamma function is smaller than that derived for circular orbits (Table 4). Yet, this effect is mostly driven by the first bin, this being located on the edge of the region of complete cluster destruction (i.e., D $``$ 1.2kpc). We note that the power-law initial mass spectrum \[PL3\] is again rejected by the poor agreement between the predicted and observed number density profiles (that is, the increase in the mean cluster mass with decreasing galactocentric distance is much larger than is observed). This result is not unexpected. The evaporation rate of a cluster with a given mass depends on its orbit, especially its pericentre (see e.g., Baumgardt 1998). Considering a system of clusters extending up to 40 kpc from the Galactic centre, if e=0.5, the range of perigalactic distances is $``$ 1-13.3 kpc. The lack of consistency between the slopes of the evolved mass and number density profiles in the \[PL3\] case illustrates that such a range of perigalactic distances is still too large to erase radial variations in the mean cluster mass. As demonstrated by Fall & Zhang (2001), a narrow distribution of pericentres and, thus, disruption rates almost independent of the cluster galactocentric distance, can be achieved if the initial velocity distribution shows some radial anisotropy and the radial anisotropy is increasing outward. Should such conditions be satisfied, the mean cluster masses in the inner ($`D<5`$ kpc) and outer ($`D>5`$ kpc) groups of clusters are similar. However, that result is obtained in the case of an initial Schechter mass spectrum (i.e., $`dN/dmm^2e^{m/5\times 10^6M_{}}`$), i.e., a mass distribution steeper than a power-law $`\alpha =2`$. In the case of an initial power-law mass spectrum, the evolved mass distribution fails to reproduce the high mass-regime of the present-day distribution (Fall & Zhang 2001, their Fig. 3). A radially uniform cluster mass function can thus be achieved even if the initial mass distribution is steadily rising toward low-mass, although such a solution requires appropriate tuning.
## 5 Summary and conclusions
The initial distribution in mass of the Galactic halo globular clusters has so far remained a poorly-known function. This is due to the fact that both a gaussian initial mass function (i.e., a two-index power-law mass spectrum) and a power-law initial mass spectrum evolve into the presently observed bell-shaped cluster mass function. As a result, the study of the temporal evolution of the mass function/spectrum only does not enable one to assess how large was the contribution of low-mass (say a few thousand solar masses) objects to the initial population of clusters.
In this paper, we have proposed a new method for shedding light on this issue which consists in comparing globular cluster system density profiles with their modelled counterparts, as a function of mass density as well as a function of number density. We assume, as in previous studies of globular cluster system dynamical evolution, that the cluster mass range and the cluster mass spectrum are initially independent of their galactocentric distance. This is equivalent to assuming that the initial mass and number density profiles are identical in shape. On the other hand, the present mass and number density profiles of the Old Halo are alike as well (see Section 2). The new constraints on the initial globular cluster mass function we have derived arise from combining this widespread assumption and this observational fact.
The clusters most vulnerable to evaporation and disruption are the low-mass ones as well as those located at small galactocentric distance. In other words, the disruption rate of globular clusters depends on their initial distribution in space as well as on their initial distribution in mass. The key point is that the mass and number density profiles show contrasting temporal evolutions. While the evolution of the latter is heavily driven by the initial cluster mass spectrum, the former remains almost unchanged during evolution for a Hubble-time, providing always that the Galactic potential remains smooth and was slowly varying. Hence, the robustness of the mass density profile provides us with an immediate estimate of the initial steepness of the spatial distribution. This can then be evolved for various initial cluster mass spectra and the resulting number density profiles compared to the observed one in order to discriminate which cluster mass spectrum provides the best match between the data and the model.
In order to test this idea, we have evolved various putative globular cluster systems characterized by different combinations of initial number density profiles (i.e., how clusters are distributed in space) and initial mass spectra (i.e., how clusters are distributed in mass). Results of these simulations are displayed in Figs. 2 and 3. We have shown that, irrespective of the initial globular cluster mass spectrum, the damage performed to the initial mass content in clusters is limited to one effective radius, that is, D$`3`$ kpc (see also McLaughlin 1999). While in this range, the spatial distribution of the cluster system mass flattens owing to the greater efficiency of cluster destruction processes, the overall slope remains close to its initial value. In sharp contrast, the temporal evolution of the number density profile depends sensitively on the initial mass spectrum. The steepness of the space-density is stationary in case of a gaussian mass function or of a power-law truncated at 10<sup>5</sup> M. On the other hand, it gets significantly shallower in the case of a mass spectrum favouring low-mass clusters, e.g., a power-law extending down to 10<sup>3</sup> M.
For each simulation, we have compared in a least-squares sense the presently observed spatial distributions with the modelled ones, obtaining the $`\chi ^2`$ and the incomplete gamma function measure of probability (see Table 4). The most likely initial conditions of course correspond to the cases for which the evolved mass and number density profiles are in good agreement with their Old Halo counterparts. The best match is achieved when an initial spatial distribution with a slope of $`3.5`$ is combined with an initial mass spectrum depleted in low-mass clusters, that is, either a gaussian mass function or a power-law mass spectrum truncated at 10<sup>5</sup> M. In this case, the cluster destruction rate is limited, as also is the corresponding temporal evolution of the number density profile, thus preserving its initial $`3.5`$ steepness, in agreement with what is observed for the Old Halo (see Table 2) . If the Galactic halo globular cluster system had actually started with this initial spatial distribution, it is very unlikely that the initial mass spectrum was a pure power-law extending down to 1000 M. The abundance of low-mass objects in such a globular cluster system would make the number density profile shallow with time, making it unable to fit the present $`3.5`$ slope.
We confirm Baumgardt ’s (1998) finding following which a power-law probing down to 1000 M combined with a steep (i.e., $`\gamma =4.5`$) spatial distribution leads to good agreement with the observed present-day number density profile. However, we caution that the observed mass density profile is then not well fitted by its evolved counterpart. In fact, owing to their robustness, all the evolved mass density profiles, irrespective of the initial globular cluster mass spectrum, are locked close to their initial $`4.5`$ slope. As a result, they remain significantly steeper, even after a 15 Gyr long evolution, than the $`3.5`$ slope shown by the present spatial distribution of the Old Halo cluster system mass. Even though such a possibility cannot be firmly ruled out ($`Q0.001`$ if disc shocking is included in the simulations), it remains much less likely than the initial $`D^{3.5}`$ space-density which we have just discussed. As a result, although the number density profile alone indicates that the Galactic globular cluster system may have started with a very steep initial spatial distribution and a power-law mass spectrum covering three orders of magnitude in mass, the mass density profile tends to dismiss this possibility.
All together, our results support the hypothesis following which the Galactic halo globular cluster system started with an initial space-density scaling as D<sup>-3.5</sup> and an initial mass spectrum somehow depleted in low-mass clusters, that is, a bell-shaped mass function similar to the current one, or a power-law mass spectrum truncated near 10<sup>5</sup> M.
## Acknowledgments
This research was supported by a Marie Curie Intra-European Fellowships within the $`6^{th}`$ European Community Framework Programme. |
warning/0507/cond-mat0507290.html | ar5iv | text | # Dephasing of a flux-qubit coupled to a harmonic oscillator
## I Introduction
A series of recent experiments have made it clear that it is possible to manipulate the quantum state of macroscopic electrical circuits based on Josephson junctions qubits ; Vion02 ; Chiorescu03 . This breakthrough opens the way to the realization of fundamental tests of quantum mechanics, up to now confined to atomic physics and quantum optics, in a solid-state physics context. An additional interest comes from the eventual possibility of using these circuits as building blocks for a quantum computer Makhlin02 . In view of this latest application, it is highly desirable to understand better how they become dephased by environmental noise.
To estimate the dephasing rates, the Bloch-Redfield theory assumes that the qubit is weakly coupled to a bath at temperature $`T`$ with a memory short compared to all relevant timescales (white noise). In that limit, it is well known that the dephasing rate is proportional only to the low-frequency part of the environment spectral density. It becomes increasingly clear however that this description is inadequate in a number of cases. For example, the Bloch-Redfield assumptions are obviously unjustified when dephasing is due to the fluctuations of slow environmental degrees of freedom which typically have a $`1/f`$ spectrum decoherence . This is also the case when a resonance of large quality factor ($`Q>>1`$) occurs in the environment at a frequency comparable to the qubit, since the memory of the environment can not be neglected then Thorwart04 . Both processes are relevant in our experiments Chiorescu03 ; Chiorescu04 . We study the quantum coherence of a circuit called the flux-qubit, measured by a DC-SQUID. The flux-qubit is sensitive to a number of microscopic degrees of freedom : motion of nearby vortices trapped in superconducting thin-films, fluctuations of the junctions critical current, and charge noise. In addition, it is strongly coupled to the harmonic oscillator (called HO in the remaining of this work) constituted by the underdamped DC-SQUID and a shunt capacitor to which it is connected to improve its resolution as a detector. In recent experiments we observed clear signatures of the strong coupling between the two systems, manifested by the appearance of sideband resonances in the spectrum Chiorescu04 . In the present article we investigate theoretically the effects of this coupling on the qubit decoherence. Thermal fluctuations of the photon number $`n`$ stored in the oscillator shift the qubit frequency by an amount $`n\delta \nu _0`$ and lead to dephasing. We note that a similar effect has been recently observed in the case of a Cooper-pair box coupled to a waveguide resonator Wallraff ; Wallraffac . When the resonator was driven to perform the measurement, the qubit line was shifted and broadened due to ac-Stark shift and photon shot noise.
In the first part of this article, we propose a simple analytical formula giving the pure dephasing time as a function of the system parameters. In the remaining we apply this model to the specific case of our circuit. We start by deriving the coupling hamiltonian between a superconducting flux-qubit and the HO. In addition to the linear coupling term Thorwart04 ; Guido04 , we find it necessary to consider the next order term which is quadratic in the oscillator variables. We finally investigate the dependence of the shift per photon $`\delta \nu _0`$ on the system bias parameters, namely the magnetic flux enclosed by the qubit loop $`\mathrm{\Phi }_x`$ and the SQUID bias current $`I_b`$. In particular, we find that it is possible to cancel the dephasing per photon $`\delta \nu _0`$ for specific bias conditions, so that the influence of thermal fluctuations on the qubit should be suppressed.
## II Derivation of an approximate formula for the dephasing time
Let us consider the situation where a qubit of frequency $`\nu _q`$ is linearly coupled to an underdamped HO of frequency $`\nu _p`$ with a coupling strength $`g`$. The qubit is supposed to be an ideal undamped two-level system, whereas the HO is coupled to a bath at temperature $`T`$ which damps its dynamics with a rate $`\kappa `$. We are interested in the limit $`\kappa <<\nu _p`$ where the oscillator is underdamped. We can write the total hamiltonian as $`H=h[(1/2)\nu _q\sigma _z+\nu _pa^{}a+g\sigma _x(a+a^{})]`$, where we introduced the Pauli matrices $`\sigma _{x,y,z}`$ in the qubit Hilbert space and the usual annihilation (creation) operator $`a`$ ($`a^{}`$) for the HO. This is the well-studied Jaynes-Cummings hamiltonian Jaynes\_Cummings ; Haroche . Let us recall a few results useful in the following. In the limit where $`|\delta ||\nu _q\nu _p|>>g`$ (called the dispersive regime), the energy eigenstates of the coupled system can be written as a function of the uncoupled energy states $`|i,n`$, where $`i=0,1`$ refers to the qubit state and $`n`$ to the photon state of the HO, as Blais
$`|+_n`$ $``$ $`|1,n+{\displaystyle \frac{g\sqrt{n+1}}{\delta }}|0,n+1`$
$`|_n`$ $``$ $`{\displaystyle \frac{g\sqrt{n+1}}{\delta }}|1,n+|0,n+1`$ (1)
their energies being
$$E_{\pm _n}=(n+1)h\nu _p\pm \frac{h}{2}\delta \pm h\frac{g^2(n+1)}{\delta }$$
(2)
One sees that because of the coupling, the energy eigenstates are shifted by a quantity $`\pm h\frac{g^2(n+1)}{\delta }`$. In the presence of $`n`$ photons, the dressed qubit excited state is $`|+,n`$ and the ground state $`|,n1`$ so the qubit transition is $`E_{+_n}E_{_{n1}}=h(\nu _q+2g^2(n+1)/\delta )`$. This means that the qubit frequency is shifted by an amount $`\nu _{q,n}\nu _{q,0}=n(2g^2/\delta )=n\delta \nu _0`$. Thus, any temporal fluctuation of the photon number will lead to dephasing Wallraffac . Let us introduce the mean photon number in the HO assumed to be at thermal equilibrium $`\overline{n}=1/(\mathrm{exp}(h\nu _p/kT)1)`$. The stationary photon number distribution is given by a Boltzmann law $`p(n)=(1/(\overline{n}+1))(\overline{n}/(\overline{n}+1))^n`$. The temporal fluctuations are characterized by the two-time correlation function $`C(\tau )=<n(0)n(\tau )>`$. It is possible to estimate $`C(\tau )`$ using a quantum Langevin equation approach as in Mandel\_Wolff . We show in this way in the annex A that
$$C(\tau )=\overline{n}(\overline{n}+1)\mathrm{exp}(\kappa |\tau |)+\overline{n}^2$$
(3)
In order to quantify the effect of these fluctuations on the qubit coherence, we follow the analysis of Blais et al. Blais . The total phase accumulated by the qubit during a free evolution is $`\varphi (t)=2\pi _0^t\nu _q(t^{})𝑑t^{}=\overline{\varphi (t)}+\delta \varphi (t)`$ where we isolated the deterministic quantity $`\overline{\varphi (t)}=2\pi (\nu _{q,0}+\overline{n}\delta \nu _0)t`$ from the fluctuating $`\delta \varphi (t)=2\pi \delta \nu _0_0^t(n(t^{})\overline{n})𝑑t^{}`$. Dephasing is described by the quantity $`f_\varphi (t)=<\mathrm{exp}(i\delta \varphi (t))>`$ called the dephasing factor. In the limit where $`t>>\kappa ^1`$, the variable $`\delta \varphi (t)`$ should have gaussian statistics so that
$$\begin{array}{ccc}\hfill f_\varphi (t)& =& \mathrm{exp}(<\delta \varphi (t)^2>/2)\hfill \\ & =& \mathrm{exp}[(2\pi \delta \nu _0)^2/2_0^t_0^t<(n(t^{})\overline{n})(n(t^{\prime \prime })\overline{n})>𝑑t^{}𝑑t^{\prime \prime }]\hfill \end{array}$$
(4)
Combining equations 3 and 4, we obtain that
$$f_\varphi (t)=\mathrm{exp}[((2\pi \delta \nu _0)^2\overline{n}(\overline{n}+1)/2)_0^t_0^t\mathrm{exp}(\kappa |t^{}t^{\prime \prime }|)𝑑t^{}𝑑t^{\prime \prime }]$$
(5)
Since $`_0^t_0^texp(\kappa |t^{}t^{\prime \prime }|)=(2/\kappa )t+(2/\kappa ^2)[\mathrm{exp}(\kappa t)1]`$, we find that the long-time decay of the dephasing factor is given by
$$<\mathrm{exp}(i\delta \varphi (t)>=\mathrm{exp}[\frac{1}{2}(2\pi \delta \nu _0)^2\overline{n}(\overline{n}+1)\frac{2}{\kappa }t]$$
(6)
This describes an exponential decay of time constant
$$\tau _\varphi =\frac{\kappa }{(2\pi \delta \nu _0)^2\overline{n}(\overline{n}+1)}$$
(7)
It is interesting to compare this formula with the one derived in Blais for the case when the HO is driven by a coherent field of mean photon $`\overline{n}`$. The photon-photon correlator is then $`C(\tau )=\overline{n}\mathrm{exp}(\kappa |t|/2)`$. Compared to equation 3, we notice a factor $`1/2`$ in the exponent which is due to the presence of the external drive, and a replacement of the $`\overline{n}(\overline{n}+1)`$ by $`\overline{n}`$. This reflects the fact that a coherent field has a poissonian distribution of photon numbers of variance $`\overline{n}`$ whereas a thermal field has a superpoissonian distribution of variance $`\overline{n}(\overline{n}+1)`$.
We will now show that in the limit where formula 7 applies (namely $`\tau _\varphi >>\kappa ^1`$) the same formula also gives the decay of the spin-echo time $`\tau _E`$. The reason is that the fluctuations of the photon number occur on a much shorter timescale than dephasing so that they can not be compensated by a refocusing pulse. More quantitatively, the phase accumulated during the echo sequence is
$$\delta \varphi _E(t)=2\pi \delta \nu _0\left[_0^{t/2}(n(t^{})\overline{n})𝑑t^{}_{t/2}^t(n(t^{})\overline{n})𝑑t^{}\right]$$
(8)
so that the fluctuations are
$`<\delta \varphi _E(t)^2>`$ $`=`$ $`(2\pi \delta \nu _0)^2[{\displaystyle _0^{t/2}}{\displaystyle _0^{t/2}}(n(t^{})\overline{n})(n(t^{\prime \prime })\overline{n})dt^{}dt^{\prime \prime }`$ (9)
$`+`$ $`{\displaystyle _{t/2}^t}{\displaystyle _{t/2}^t}(n(t^{})\overline{n})(n(t^{\prime \prime })\overline{n})𝑑t^{}𝑑t^{\prime \prime }`$
$``$ $`2{\displaystyle _0^{t/2}}{\displaystyle _{t/2}^t}(n(t^{})\overline{n})(n(t^{\prime \prime })\overline{n})dt^{}dt^{\prime \prime }]`$
Obviously we need to calculate only the last term. Using the expression for the correlation function given earlier we find that
$$_0^{t/2}_{t/2}^t(n(t^{})\overline{n})(n(t^{\prime \prime })\overline{n})𝑑t^{}𝑑t^{\prime \prime }=\frac{1}{\kappa ^2}\overline{n}(\overline{n}+1)(1\mathrm{exp}(\kappa t/2))^2$$
Combining with previous results we obtain
$$<\delta \varphi _E(t)^2>=(2\pi \delta \nu _0)^2\overline{n}(\overline{n}+1)\left[\frac{2}{\kappa }t+\frac{4}{\kappa ^2}(\mathrm{exp}(\kappa t/2)1)\frac{2}{\kappa ^2}(1\mathrm{exp}(\kappa t/2))^2\right]$$
(10)
In the limit where $`t>>\kappa ^1`$, the long-time decay is still dominated by the term linear in $`t`$ and we obtain $`\tau _E=\tau _\varphi `$. If on the other hand we were in the opposite limit in which the dephasing time is shorter than the photon correlation time, the decay of the Ramsey signal would be gaussian and the echo would decay with a $`\mathrm{exp}(\kappa t^3/12)`$ law much slower than the Ramsey decay. We note that this crossover between a lorentzian and a gaussian lineshape when the dephasing time becomes shorter than $`\kappa ^1`$ has been observed in Wallraffac .
Whereas in this reasoning we only considered the case where the qubit-HO coupling is linear, it can also be applied evidently for a more complex interaction hamiltonian whenever the qubit frequency shift is proportional to the photon number. As we will see in the next paragraph, this is the case for our circuit in which a flux-qubit is coupled to the plasma mode of its measuring DC-SQUID.
## III Qubit-plasma mode coupling Hamiltonian
### III.1 Description of the system
The flux-qubit is a superconducting loop containing three Josephson junctions threaded by an external flux $`\mathrm{\Phi }_xf(\mathrm{\Phi }_0/2\pi )`$ Mooij99 ; Caspar00 ; Chiorescu03 . It is coupled to a DC-SQUID detector shunted by an external capacitor $`C_{sh}`$ whose role is to limit phase fluctuations across the SQUID as well as to filter high-frequency noise from the dissipative impedance. The SQUID is threaded by a flux $`\mathrm{\Phi }_{Sq}f^{}(\mathrm{\Phi }_0/2\pi )`$. The circuit diagram is shown in figure 1a. There, the flux-qubit is the loop in red containing the three junctions of phases $`\varphi _i`$ and capacitances $`C_i`$ ($`i=1,2,3`$). It also includes an inductance $`L_1`$ which models the branch inductance and eventually the inductance of a fourth larger junction Bertet04 . The two inductances $`K_1`$ and $`K_2`$ model the kinetic inductance of the line shared by the SQUID and the qubit. The SQUID is the larger loop in blue. The junction phases are called $`\varphi _4`$ and $`\varphi _5`$ and their capacitances $`C_4`$ and $`C_5`$. The critical current of the circuit junctions is written $`I_{Ci}`$ ($`i=1`$ to $`5`$). The SQUID loop also contains two inductances $`K_3`$ and $`L_2`$ which model its self-inductance. The SQUID is connected to the capacitor $`C_{sh}`$ through superconducting lines of parasitic inductance $`L_s`$. The phase across the stray inductance and the SQUID is denoted $`\varphi _A`$. The whole circuit is biased by a current source $`I_b`$ in parallel with a dissipative admittance $`Y(\omega )`$. Since our goal is primarily to determine the qubit-plasma mode coupling hamiltonian, we will neglect the admittance $`Y(\omega )`$.
We start writing the total hamiltonian of the circuit shown in figure 1 using the circuit theory presented in Guido03 . We first choose a spanning tree containing all the capacitors as shown in figure 1a. We then write the loop submatrices
$$𝐅_{CL}=\left(\begin{array}{ccc}1& 0& 0\\ 1& 0& 0\\ 1& 0& 0\\ 0& 1& 0\\ 0& 1& 1\\ 0& 0& 1\end{array}\right)\text{ , }𝐅_{CB}=\left(\begin{array}{c}0\\ 0\\ 0\\ 0\\ 0\\ 1\end{array}\right)$$
(11)
and
$$𝐅_{KL}=\left(\begin{array}{ccc}1& 1& 1\\ 1& 1& 0\\ 0& 1& 1\end{array}\right)\text{ , }𝐅_{KB}=\left(\begin{array}{c}0\\ 0\\ 0\end{array}\right)$$
(12)
We note $`M`$ the mutual inductance between the qubit and SQUID loops. In the notations of Guido03 the inductance matrices are
$$𝐋=\left(\begin{array}{ccc}L_1& M& 0\\ M& L_2& 0\\ 0& 0& L_3\end{array}\right)\text{ , }𝐋_K=\left(\begin{array}{ccc}K_1& 0& 0\\ 0& K_2& 0\\ 0& 0& K_3\end{array}\right)\text{ , }𝐋_{LK}=\left(\begin{array}{ccc}0& 0& 0\\ 0& 0& 0\\ 0& 0& 0\end{array}\right)$$
(13)
and
$$𝐋_{LL}=\left(\begin{array}{ccc}L_1+K_1+K_2& MK_1K_2& K_1\\ MK_1K_2& L_2+K_1+K_2& K_1K_3\\ K_1& K_1K_2& K_1+K_3+L_3\end{array}\right)$$
(14)
We note $`l_{ij}^1`$ the matrix elements of $`L_{LL}^1`$ whose expressions can be easily computed. We finally obtain the total hamiltonian as
$$H_S=H_{kin}+(\mathrm{\Phi }_0/2\pi )^2U(\varphi )$$
(15)
where
$$\begin{array}{ccc}\hfill H_{kin}& =& (\frac{\mathrm{\Phi }_0}{2\pi })^2(\frac{1}{2}_{i=1}^5Q_i^2/C_i+Q_A^2/C_{sh})\hfill \\ \hfill U(\varphi )& =& _{i=1}^5\frac{1}{L_{J,i}}\mathrm{cos}\varphi _i+\frac{1}{2l_{11}}(_{i=1}^3\varphi _if)^2\hfill \\ & & +(_{i=1}^3\varphi _if)[l_{21}^1(\varphi _4+\varphi _5f^{})l_{31}^1(\varphi _5+\varphi _A)]+u(\varphi _4,\varphi _5,\varphi _A)\hfill \end{array}$$
(16)
The $`Q_i`$ ($`Q_A`$) are the charges stored on the capacitors $`C_i`$ ($`C_{sh}`$) and $`u(\varphi _4,\varphi _5,\varphi _A)`$ is defined by
$$\begin{array}{ccc}\hfill u& =& +\frac{1}{2l_{22}}(\varphi _4+\varphi _5f^{})^2\frac{1}{2l_{32}}[(\varphi _5+\varphi _Af^{})^2)+(\varphi _4+\varphi _5)^2(\varphi _4\varphi _A)^2]\hfill \\ & & +\frac{1}{2l_{33}}(\varphi _5+\varphi _A)^2+\frac{2\pi }{\mathrm{\Phi }_0}I_b\varphi _A\hfill \end{array}$$
(17)
Our first goal will be here to simplify this hamiltonian so that the coupling of the relevant degrees of freedom is made clear. We will consider here that they are only two : the qubit, in the two-level approximation, and the plasma mode considered, if uncoupled to the qubit, as a harmonic oscillator. In particular, we will neglect the SQUID junctions capacitance which bring additional resonances at higher frequencies, and only consider the shunt capacitance $`C_{sh}`$. Our approach is justified by the fact that only the plasma mode and the qubit have comparable energy scales, that the plasma mode is strongly coupled to the environment and therefore relevant for studying dephasing and relaxation, and that it undergoes thermal fluctuations because of its relatively low frequency. We also observed experimental evidence for the qubit-plasma mode strong coupling Chiorescu04 . These results are a clear indication that a quantum-mechanical description of the coupled “qubit-plasma mode” is indeed needed. We will start by doing the two-level approximation on the qubit variables.
### III.2 Qubit hamiltonian and two-level approximation
The hamiltonian for the qubit alone is
$$H_q(f,I_b)=H_{kin}(\mathrm{\Phi }_0/2\pi )^2[\underset{i}{}\frac{1}{L_{J,i}}\mathrm{cos}\varphi _i+\frac{1}{2l_{11}}(\varphi _if)^2]$$
(18)
We first write this hamiltonian in a two-level approximation in terms of the Pauli matrices, which is valid here around $`\mathrm{\Phi }_0/2`$ because of the specific properties of the circuit eigenstates. We define the states $`|0`$ and $`|1`$ as the eigenstates of $`H_q(\pi ,0)H_q^{(0)}`$. Then when restraining ourselves to the $`0,1`$ states we have by definition $`H_q^{(0)}=(h\mathrm{\Delta }/2)\sigma _z`$ We define $`H_q^{(1)}(f,I_b)=H_q(f,I_b)H_q^{(0)}`$. We have
$$\begin{array}{ccc}\hfill H_q^{(1)}(f,I_b)& =& (\mathrm{\Phi }_0/2\pi )^2[(1/2l_{11})[(\varphi _if)^2(\varphi _i\pi )^2]]\hfill \\ & =& (\mathrm{\Phi }_0/2\pi )^2[(1/2l_{11})(2\varphi _i2\pi (f\pi ))(f\pi )]\hfill \\ & =& (\mathrm{\Phi }_0/2\pi )^2(\varphi _i\pi )(f\pi )/l_{11}\hfill \end{array}$$
(19)
forgetting constant terms in the last equation. We now want to decompose $`H_q^{(1)}(f,I_b)`$ on the $`|0,|1`$ subspace. We start by writing that $`H_q^{(1)}(f,I_b)=(1/2)[(h_{00}+h_{11})I+(h_{00}h_{11})\sigma _z+(h_{01}+h_{10})\sigma _x+(ih_{01}ih_{10})\sigma _y]`$, where $`h_{ij}=<i|H_q^{(1)}|j>`$. For symmetry reasons $`<0|(\varphi _i\pi )|0>=<1|(\varphi _i\pi )|1>=0`$ so that $`h_{00}=h_{11}=0`$. Indeed, the hamiltonian $`H_q^{(0)}`$ is invariant under the transformation $`T`$ $`\varphi _1\varphi _1`$, $`\varphi _2\varphi _2`$ and $`\varphi _32\pi \varphi _3`$. This means that the eigenstates of $`H_q^{(0)}`$ have to also be eigenstates of $`T`$. Since $`T^2=I`$, these eigenstates should be either $`1`$ or $`1`$ so that $`\psi _i(\varphi _1,\varphi _2,2\pi \varphi _3)=\pm \psi _i(\varphi _1,\varphi _2,\varphi _3)`$ and $`|\psi _i(\varphi _1,\varphi _2,2\pi \varphi _3)|^2=|\psi _i(\varphi _1,\varphi _2,\varphi _3)|^2`$. This implies that the matrix elements $`h_{00}=h_{11}=0`$.
Since we can always chose the global phases of $`|0`$ and $`|1`$ so that $`h_{01}`$ and $`h_{10}`$ are real, we obtain that $`H_q^{(1)}(f,I_b)=h_{01}\sigma _x`$ where $`h_{01}=(\mathrm{\Phi }_0/2\pi )^2(0|\varphi _i|1)[(f\pi )/l_{11}]`$. We next define $`I_p(\mathrm{\Phi }_0/2\pi )(1/l_{11})<0|\varphi _i|1>`$, $`ϵ=2I_p(\mathrm{\Phi }_x\mathrm{\Phi }_0/2)/h`$. In the end we can write the total hamiltonian as
$$H_q(f)=\frac{h}{2}(\mathrm{\Delta }\sigma _z+ϵ\sigma _x)$$
(20)
We note that this derivation generalizes the analysis of Orlando99 which was made under the assumption that the qubit loop self-inductance is negligible. Here we retrieve the result of Nakamura\_theory which showed numerically that the form of the qubit hamiltonian was little affected by taking into account this inductance.
The hamiltonian 20 yields a qubit transition frequency $`\nu _q=\sqrt{\mathrm{\Delta }^2+ϵ^2}`$. The corresponding dependence is plotted in figure 2 for realistic parameters. An interesting property is that when the qubit is biased at $`ϵ=0`$ (dashed line in figure 2), it is insensitive to first order to noise in the bias variable $`ϵ`$.
### III.3 Plasma mode hamiltonian
Next we turn to the “SQUID+shunt capacitor” variables $`\varphi _4,\varphi _5,\varphi _A`$. As already explained, we will here make a crude approximation and completely neglect the SQUID junctions capacitance. This is justified by the fact that at the bias current that we use the modes to which they correspond are at frequencies much higher than the qubit and plasma mode. Moreover, we will keep only the terms of second order in the SQUID potential, which is equivalent to considering the SQUID as one inductance $`L_J(I_b,f^{})`$. On the other hand we will keep in the analysis the dependence on flux of $`L_J`$ which has important effects.
In this approximation, the only dynamical variable in the system is $`\varphi _A`$. Its hamiltonian is very simply given by the hamiltonian of a harmonic oscillator of capacitance $`C_{sh}`$ and inductance $`L=L_s+L_J(I_b,f^{})`$
$$H_p=Q_A^2/2C_{sh}+(\mathrm{\Phi }_0/2\pi )^2(\varphi _A\overline{\varphi _A})^2/2L$$
(21)
where $`\overline{\varphi _A}=(2\pi /\mathrm{\Phi }_0)LI_b`$ is the mean value of $`\varphi _A`$. We call $`a`$ and $`a^{}`$ the creation and annihilation operators corresponding to this harmonic oscillator :
$$a=\frac{\mathrm{\Phi }_0}{2\pi }\sqrt{\frac{\pi C_{sh}\nu _0}{\mathrm{}}}(\varphi _A\overline{\varphi _A})+\frac{i}{\sqrt{2C_{sh}h\nu _0}}Q_A$$
(22)
and $`a^{}`$ is the conjugate operator. Then the hamiltonian is simply $`H_p=\mathrm{}\omega _0(a^{}a+1/2)`$, and the phase $`\varphi _A`$ can be written : $`\varphi _A=\overline{\varphi _A}+\delta \varphi _0(a+a^{})`$ ($`\delta \varphi _0`$ is the rms amplitude of the vacuum fluctuations of $`\varphi _A`$).
In our model where the SQUID junctions have no capacitance, for a given value of $`\varphi _A`$ and $`f^{}`$ all the phases of the SQUID are well-defined functions $`\varphi _{4,5}(\varphi _A,f^{})`$. So the quantum fluctuations of $`\varphi _A`$ translate directly into fluctuations of $`\varphi _4`$ and $`\varphi _5`$ as follows :
$$\varphi _{4,5}=\overline{\varphi _{4,5}}+(d\varphi _{4,5}/d\varphi _A)\delta \varphi _0(a+a^{})+(1/2)(d^2\varphi _{4,5}/d\varphi _A^2)(\delta \varphi _0)^2(a+a^{})^2$$
(23)
where $`\overline{\varphi _{4,5}}=\varphi _{4,5}(\overline{\varphi _A})`$. We develop the functions to second order in $`a`$ and $`a^{}`$ for consistency. Again, the sensitivity coefficients $`d\varphi _{4,5}/d\varphi _A`$ and $`d^2\varphi _{4,5}/d\varphi _A^2`$ depend on $`I_b`$ and $`f^{}`$ and can be easily calculated. We also note that $`d\varphi _A=(2\pi /\mathrm{\Phi }_0)LdI_b`$ so that
$$\varphi _{4,5}=\overline{\varphi _{4,5}}+\frac{\mathrm{\Phi }_0}{2\pi L}\frac{d\varphi _{4,5}}{dI_b}\delta \varphi _0(a+a^{})+\frac{1}{2}(\frac{\mathrm{\Phi }_0}{2\pi L})^2\frac{d^2\varphi _{4,5}}{dI_b^2}(\delta \varphi _0)^2(a+a^{})^2$$
(24)
We will finally show how to evaluate the sensitivity coefficients for any SQUID parameters. Following Lefevre ; Balestrothesis we introduce the parameters $`x=(\varphi _4+\varphi _5)/2`$, $`y=(\varphi _4\varphi _5)/2`$, $`s=I_b/(I_{C4}+I_{C5})`$, $`b=\mathrm{\Phi }_0/\pi L_{Sq}(I_{C4}+I_{C5})`$, $`U_0=\mathrm{\Phi }_0(I_{C4}+I_{C5})/2\pi `$, $`\alpha =(I_{C4}I_{C5})/(I_{C4}+I_{C5})`$, and $`\delta =(K_3+K_1L_2K_2)/L_{Sq}`$. The stationary solutions for the SQUID phases are obtained by minimizing the $`2`$-dimensional potential
$$U(x,y)=U_0[sx\mathrm{cos}x\mathrm{cos}y\alpha \mathrm{sin}x\mathrm{sin}y\delta sy+b(yf^{}/2)^2]$$
(25)
that is to solve the equations
$$\begin{array}{ccc}\hfill U/x& =& U_0[s+\mathrm{sin}x\mathrm{cos}y\alpha \mathrm{cos}x\mathrm{sin}y]=0\hfill \\ \hfill U/y& =& U_0[\mathrm{cos}x\mathrm{sin}y\alpha \mathrm{sin}x\mathrm{cos}y\delta s+2b(yf^{}/2)]=0\hfill \end{array}$$
(26)
From this it is possible to obtain numerically the functions $`\mathrm{\Phi }_{4,5}(\varphi _A,f^{})`$ and thus all the sensitivity coefficients needed in the model.
### III.4 Qubit-plasma mode coupling hamiltonian
The coupling hamiltonian is due to two contributions. First, the explicit coupling term in the equation 36. But we also need to rewrite the plasma mode hamiltonian since the parameters of this hamiltonian (notably the SQUID Josephson inductance) now depends on the qubit state. We therefore need to reconsider the following variables in the plasma mode hamiltonian :
$$\begin{array}{ccc}\hfill f^{}& & f^{}+(2\pi /\mathrm{\Phi }_0)MI_p\sigma _x\hfill \\ \hfill L& & L+\delta L\sigma _x\hfill \\ \hfill \overline{\varphi _A}& & \overline{\varphi _A}+(2\pi /\mathrm{\Phi }_0)\delta LI_b\sigma _x\hfill \end{array}$$
(27)
where $`\delta L(2\pi /\mathrm{\Phi }_0)(L_J/f^{})MI_p`$. The SQUID-qubit coupling term writes
$$V=(\mathrm{\Phi }_0/2\pi )^2(\underset{i=1}{\overset{3}{}}\varphi _if)[l_{21}^1(\varphi _4+\varphi _5f^{})l_{31}^1(\varphi _5+\varphi _A)]$$
(28)
Since $`\sigma _x=(\varphi _i\pi )/<0|\varphi _i|1>`$, we can rewrite $`V`$ in the form
$$V=(\mathrm{\Phi }_0/2\pi )l_{11}I_p\sigma _x[l_{21}^1(\varphi _4+\varphi _5f^{})l_{31}^1(\varphi _5+\varphi _A)]$$
(29)
Keeping only the terms which contain explicit couplings we obtain that
$$\begin{array}{ccc}\hfill V& =& (\mathrm{\Phi }_0/2\pi )l_{11}I_p\sigma _x[(\mathrm{\Phi }_0/2\pi L)(l_{21}^1(d\varphi _4/dI_b)+(l_{21}^1l_{31}^1)(d\varphi _5/dI_b))\delta \varphi _0(a+a^{})\hfill \\ & +& (1/2)(\mathrm{\Phi }_0/2\pi L)^2(l_{21}^1(d^2\varphi _4/dI_b^2)+(l_{21}^1l_{31}^1)(d^2\varphi _5/dI_b^2))\delta \varphi _0^2(a+a^{})^2]\hfill \end{array}$$
(30)
On the other hand, the plasma mode hamiltonian now writes
$`H_{plq}`$ $`=`$ $`Q_A^2/2C_{sh}+(\mathrm{\Phi }_0/2\pi )^2[\varphi _A\overline{\varphi _A}(2\pi /\mathrm{\Phi }_0)\delta LI_b\sigma _x]^2/2(L+\delta L\sigma _x)`$ (31)
$`=`$ $`Q_A^2/2C_{sh}+\left({\displaystyle \frac{\mathrm{\Phi }_0}{2\pi }}\right)^2\left[{\displaystyle \frac{(\varphi _A\overline{\varphi _A})^2}{2L}}{\displaystyle \frac{2\pi \delta LI_b}{\mathrm{\Phi }_0L}}(\varphi _A\overline{\varphi _A})\sigma _x{\displaystyle \frac{\delta L}{2L^2}}(\varphi _A\overline{\varphi _A})^2\sigma _x\right]`$
so that
$$H_{plq}=h\nu _0(a^{}a+1/2)\frac{\mathrm{\Phi }_0\delta LI_b}{2\pi L}\delta \varphi _0(a+a^{})\left(\frac{\mathrm{\Phi }_0}{2\pi }\right)^2\delta \varphi _0^2(\delta L/2L^2)(a+a^{})^2\sigma _x$$
(32)
Finally the total interaction hamiltonian $`H_I=V+H_{plq}`$ can be written as
$$H_I=h[g_1(a+a^{})+g_2(a+a^{})^2]\sigma _x$$
(33)
The coupling constants $`g_1`$ and $`g_2`$ could be deduced from the above expressions. Nevertheless we propose a way to determine them experimentally. We first note that this coupling hamiltonian can be rewritten
$$H_I=h[\lambda _1\delta \varphi _A+\lambda _2\delta \varphi _A^2]\sigma _x$$
(34)
where $`\delta \varphi _A=\varphi _A\overline{\varphi _A}`$. This gives us a very direct way of evaluating the coupling constants $`g_1`$ and $`g_2`$ : indeed varying the bias current $`I_b`$ through the SQUID by an amount $`\delta I_b`$ is equivalent to a variation $`\delta \varphi _A=2\pi L\delta I_b/\mathrm{\Phi }_0`$. Since we can experimentally measure the dependence of the qubit frequency on the bias current $`ϵ(I_b)`$ Bertet04 and thus measure $`ϵ/I_b`$ and $`^2ϵ/I_b^2`$, we can obtain the coupling constants from the experiment by the following expressions :
$$\begin{array}{ccc}\hfill g_1& =& (1/2)(ϵ/I_b)(\mathrm{\Phi }_0/2\pi L)\delta \varphi _0\hfill \\ \hfill g_2& =& (1/4)(^2ϵ/I_b^2)(\mathrm{\Phi }_0/2\pi L)^2\delta \varphi _0^2\hfill \end{array}$$
(35)
Finally the total qubit-plasma mode hamiltonian can be written as
$$H/h=(1/2)(\mathrm{\Delta }\sigma _zϵ\sigma _x)+\nu _0(a^{}a+1/2)+[g_1(I_b)(a+a^{})+g_2(I_b)(a+a^{})^2]\sigma _x$$
(36)
We note that a more rigorous derivation for the coupling between the plasma mode and the qubit after elimination of the internal dynamics of the SQUID thanks to the Feynman-Vernon influence functional has been performed in Nakano and gives ultimately the same form if we develop their interaction hamiltonian to the second order in the oscillator variables. It is also instructive to compare this hamiltonian to the simpler situation studied in Wallraff . There a Cooper-pair box is capacitively coupled to a coplanar waveguide resonator. The interaction hamiltonian contains only one term, linear in the oscillator variables, and with a fixed coupling constant depending on the geometry of the circuit. In our case the coupling is mediated by the SQUID flux-dependent and current-dependent inductance ; therefore the coupling constants are tunable and higher-order terms are of importance. This made possible to induce transitions in which both the HO and the qubit state are modified Chiorescu04 .
### III.5 Coupling constants
We now want to give analytical formulae for the coupling constants $`g_1`$ and $`g_2`$ in the simplest case where a number of assumptions are made : 1) the SQUID-qubit coupling is supposed to be symmetric ($`K_1=K_2`$) so that the qubit bias is only coupled to the current $`J`$ circulating in the SQUID loop $`ϵ=(2I_p/h)(\mathrm{\Phi }_x+MJ(I_b))`$ 2) the SQUID loop self-inductance and the stray inductance $`L_s`$ are negligible so that the total inductance of the plasma mode is the SQUID Josephson inductance $`L=L_J(f,I_b)`$. Within these assumptions the equations 26 are easily solved and yield that
$`x`$ $`=`$ $`\mathrm{arcsin}\left({\displaystyle \frac{I_b}{2I_C\mathrm{cos}(f^{}/2)}}\right)`$
$`y`$ $`=`$ $`f^{}/2`$ (37)
which implies that the current circulating in the SQUID loop is
$$J(I_1I_2)/2=I_C\sqrt{1\left(\frac{I_b}{2I_C\mathrm{cos}(f^{}/2)}\right)^2}\mathrm{sin}(f^{}/2)$$
(38)
The rms phase fluctuations of the plasma mode are $`\delta \varphi _0=(2\pi /\mathrm{\Phi }_0)\sqrt{h\nu _pL/2}`$ so that we obtain
$`g_1`$ $`=`$ $`{\displaystyle \frac{MI_p}{h}}{\displaystyle \frac{\mathrm{sin}f^{}/2}{4I_C\mathrm{cos}^2f^{}/2}}\sqrt{{\displaystyle \frac{h\nu _p}{2L}}}I_b\left[1\left({\displaystyle \frac{I_b}{2I_C\mathrm{cos}f^{}/2}}\right)^2\right]^{1/2}`$
$`g_2`$ $`=`$ $`{\displaystyle \frac{MI_p}{h}}{\displaystyle \frac{\mathrm{sin}f^{}/2}{8I_C\mathrm{cos}^2f^{}/2}}{\displaystyle \frac{h\nu _p}{2L}}`$ (39)
$`\left[\left({\displaystyle \frac{I_b}{2I_C\mathrm{cos}f^{}/2}}\right)^2\left(1\left({\displaystyle \frac{I_b}{2I_C\mathrm{cos}f^{}/2}}\right)^2\right)^{3/2}+\left(1\left({\displaystyle \frac{I_b}{2I_C\mathrm{cos}f^{}/2}}\right)^2\right)^{1/2}\right]`$
Around $`I_b=0`$ these expressions can be simplified by keeping only the lowest order in $`I_b`$ :
$`g_1`$ $``$ $`{\displaystyle \frac{MI_p}{h}}{\displaystyle \frac{\mathrm{sin}f^{}/2}{4I_C\mathrm{cos}^2f^{}/2}}\sqrt{{\displaystyle \frac{h\nu _p}{2L_J}}}I_b`$
$`g_2`$ $``$ $`{\displaystyle \frac{1}{16}}{\displaystyle \frac{\mathrm{sin}f^{}/2}{\mathrm{cos}^2f^{}/2}}{\displaystyle \frac{MI_p}{LI_C}}\nu _p`$ (40)
We will now discuss quantitatively the behaviour of $`g_1`$ and $`g_2`$ for actual sample parameters Bertet04 : $`I_C=3.4\mu A`$, $`M=6.5pH`$, $`I_p=240nA`$, $`\mathrm{\Delta }=5.5GHz`$, $`\nu _p=3.1GHz`$, $`L_J=300pH`$, $`f^{}/2=1.45\pi `$. We will restrict ourselves to a range of bias conditions relevant for our conditions, supposing that $`I_b`$ varies between $`\pm 300nA`$ and that $`f^{}/2`$ varies by $`df^{}=\pm 410^3\pi `$ around $`1.45\pi `$. We chose such an interval for $`f^{}`$ because it corresponds to changing the qubit bias point $`ϵ`$ by $`\pm 2GHz`$ around $`0`$. The constants $`g_1`$ and $`g_2`$ are plotted in figure 3 as a function of $`I_b`$ for two different values of $`f^{}`$ ($`g_1`$ is shown as a full line, $`g_2`$ as a dashed line, and the two different values of $`f^{}`$ are symbolized by gray for $`df^{}=2\pi 410^3`$ and black for $`df^{}=0`$). It can be seen that the coupling constants only weakly depend on the value of the flux in this range, so that we will neglect this dependence in the following and consider that $`g_1`$ and $`g_2`$ only depend on the bias current $`I_b`$. Moreover we see from figure 3 that the approximations made in equation III.5 are justified in this range of parameters since $`g_1`$ is closely linear in $`I_b`$ and $`g_2`$ nearly constant. We also note that $`g_1=0`$ for $`I_b=0`$. This fact can be generalized to the case where the SQUID-qubit coupling is not symmetric and the junctions critical current are dissimilar : in certain conditions these asymmetries can be compensated for by applying a bias current $`I_b^{}`$ for which $`g_1(I_b^{})=0`$ Guido04 . At the current $`I_b^{}`$, the qubit is effectively decoupled from the measuring circuit fluctuations to first order.
## IV Energy levels and dephasing
In this paragraph we investigate the discuss the energy levels of the hamiltonian 36 and we estimate the frequency shift per photon $`\delta \nu _0`$. We will only consider the case where the qubit is detuned from the plasma mode $`\nu _q\nu _p>>g_1`$ and also $`|\nu _q2\nu _p|>>g_2`$.
### IV.1 Energy levels
If the qubit and the plasma mode were uncoupled (case $`g_1=g_2=0`$), the system energy eigenstates would be $`|i|n`$, where $`i`$ refers to the qubit state and can be eother $`0`$ or $`1`$, and $`n`$ to the plasma mode occupation number. The energy levels would simply be $`E_{i,n}^{(u)}=h(i\sqrt{\mathrm{\Delta }^2+ϵ^2}+n\nu _0)`$. When either $`g_1`$ or $`g_2`$ are non zero, these eigenstates are modified, but for convenience we will still label them thanks to the uncoupled state from which they are the closest $`|i,n`$. However, the energies are now modified : $`E_{i,n}=E_{i,n}^{(u)}+\delta \nu _{i,n}`$. The aim of this paragraph is to estimate the quantity $`\delta \nu _{i,n}`$.
We first rewrite the hamiltonian in a more convenient way for our purpose, by introducing the rotated axes $`X`$ and $`Z`$ defined as $`\sigma _Z=(\mathrm{\Delta }\sigma _z+ϵ\sigma _x)/\sqrt{\mathrm{\Delta }^2+ϵ^2}`$ and $`\sigma _X=(ϵ\sigma _z+\mathrm{\Delta }\sigma _x)/\sqrt{\mathrm{\Delta }^2+ϵ^2}`$. The angle $`\theta `$ is defined so that $`\mathrm{cos}\theta ϵ/\sqrt{\mathrm{\Delta }^2+ϵ^2}`$ and $`\mathrm{sin}\theta \mathrm{\Delta }/\sqrt{\mathrm{\Delta }^2+ϵ^2}`$. The system hamiltonian now writes
$$H/h=(1/2)\sqrt{\mathrm{\Delta }^2+ϵ^2}\sigma _Z+\nu _p(a^{}a+1/2)+[g_1(a+a^{})+g_2(a+a^{})^2](\mathrm{cos}\theta \sigma _Z+\mathrm{sin}\theta \sigma _X)$$
(41)
#### IV.1.1 Linear term
Let us first suppose that $`g_2=0`$. Then the coupling is linear in the oscillator variables, with a longitudinal component proportional to $`\mathrm{cos}\theta `$ and a transverse component proportional to $`\mathrm{sin}\theta `$.
We first notice that the longitudinal component has no effect on the energy states. Indeed, the term $`\sigma _Z(a+a^{})`$ does not shift the energy levels to first order of the perturbation theory since $`<i,n|\sigma _Z(a+a^{})|i,n>=0`$ for all states. To second order, all energy levels are shifted by the same quantity $`(g_1\mathrm{cos}\theta )^2/\nu _p`$ which implies that all the transition frequencies stay unchanged. This conclusion stays true to all orders of perturbation theory.
On the other hand, the transverse coupling term $`g_1\mathrm{sin}\theta \sigma _X(a+a^{})`$ produces the well-known dipersive shift in cavity QED Raimond ; Wallraff to second-order in perturbation theory, as calculated in the first section of this article. In the rotating wave approximation, we recall that $`\delta \nu _{i,n}^1=i(g_1\mathrm{sin}\theta )^2/\delta +(2i1)n(g_1\mathrm{sin}\theta )^2/\delta `$ where $`\delta =\sqrt{\mathrm{\Delta }^2+ϵ^2}\nu _0`$ is the qubit-plasma mdoe detuning. However, it is necessary here to go beyond the rotating wave approximation since $`\delta `$ is of the same order of magnitude as the qubit and the oscillator frequency. It is easily seen that to second order of perturbation theory, one obtains
$$\delta \nu _{i,n}^1=2i(g_1\mathrm{sin}\theta )^2\frac{\sqrt{\mathrm{\Delta }^2+ϵ^2}}{\mathrm{\Delta }^2+ϵ^2\nu _0^2}+(2i1)2n(g_1\mathrm{sin}\theta )^2\frac{\sqrt{\mathrm{\Delta }^2+ϵ^2}}{\mathrm{\Delta }^2+ϵ^2\nu _0^2}$$
(42)
The first term of this equation describes the Lamb shift, which simply renormalizes the bare qubit frequency and has no influence on dephasing. We will thus neglect it in the following. The frequency shift per photon is
$$\delta \nu _0^1=4(g_1\mathrm{sin}\theta )^2\frac{\sqrt{\mathrm{\Delta }^2+ϵ^2}}{\mathrm{\Delta }^2+ϵ^2\nu _0^2}$$
(43)
From the previous expression it is clear that the sign of $`\delta \nu _0^1`$ is fully determined by the sign of $`\delta `$. In particular, in our experiments where $`\nu _q>\nu _p`$, $`\delta \nu _0^1>0`$.
#### IV.1.2 Quadratic term
Next, we consider the case $`g_1=0`$ but $`g_2>0`$ (which is the case notably at the decoupled current $`I_b=I_b^{}`$ Guido04 ). The quadratic coupling term produces effects which are sensibly different from the cavity QED case. Indeed, it generates a frequency shift to first order in perturbation theory via the term $`2g_2\mathrm{cos}\theta \sigma _Za^{}a`$. Considering that the $`g_2`$ coupling is already second order, we only keep the first order of perturbation theory. We therefore obtain that $`\delta \nu _{i,n}^2=(2i1)2g_2n\mathrm{cos}\theta `$ so that the shift per photon is
$$\delta \nu _0^2=4g_2\mathrm{cos}\theta $$
(44)
Contrary to the shift produced by the linear coupling term, the sign of this frequency shift now depends on $`ϵ`$. Since $`g_2`$ is negative (see figure 3), $`\delta \nu _0^2`$ actually has the same sign as $`ϵ`$. We also note that the quadratic term has no effect on the qubit when $`ϵ=0`$, since at that point the average flux generated by both qubit states $`|0`$ and $`|1`$ averages out to zero so that the SQUID Josephson inductance is unchanged.
#### IV.1.3 Total frequency shift and dependence on the bias parameters
The total frequency shift per photon is the sum of the two contributions identified above :
$`\delta \nu _0(ϵ,I_b)`$ $``$ $`\delta \nu _0^1+\delta \nu _0^2`$ (45)
$`=`$ $`4\left[(g_1\mathrm{sin}\theta )^2{\displaystyle \frac{\sqrt{\mathrm{\Delta }^2+ϵ^2}}{\mathrm{\Delta }^2+ϵ^2\nu _p^2}}g_2\mathrm{cos}\theta \right]`$
Because of the different dependence on $`ϵ`$ of the two contributions discussed above, we expect a cancellation of the AC-Zeeman term (due to $`g_1`$) by the quadratic term (due to $`g_2`$) for some bias parameters corresponding to a negative value of $`ϵ`$. This is shown in figure 4 where we plotted $`\delta \nu _0(ϵ,I_b)`$ as calculated with the formula above for the sample parameters considered in the previous paragraph. The curved full line corresponds to the points $`ϵ_m(I_b)`$ for which $`\delta \nu _0=0`$. For these bias conditions, it is expected that the qubit is insensitive to the thermal fluctuations of the plasma mode (see formula 7). Therefore we predict an increase of the dephasing time whenever $`ϵ=ϵ_m(I_b)`$.
We stress that these biasing conditions are non-trivial in the sense that they do not satisfy an obvious symmetry in the circuit. This point is emphasized in figure 4 where we plotted as a dashed line the bias conditions $`ϵ=0`$ for which the qubit is insensitive to phase noise (due to flux or bias current noise) ; and as a dotted line the decoupling current conditions $`I_b=I_b^{}`$ for which the qubit is effectively decoupled from its measuring circuit. The $`ϵ_m(I_b)`$ line shares only one point with these two curves : the point $`(I_b^{},ϵ)`$ which is optimal with respect to flux, bias current, and photon noise. For the rest, the three lines are obviously distinct. This makes it possible to experimentally discriminate between the various noise sources limiting the qubit coherence by studying the dependence of $`\tau _\varphi `$ on bias parameters.
## V Conclusion
Superconducting qubits are often measured by circuits behaving as underdamped oscillators to prevent energy relaxation of the qubit. If these oscillators have a frequency comparable to $`kT`$, their photon number undergoes thermal fluctuations. This induces frequency dispersive frequency shifts of the qubit frequency $`n\delta \nu _0`$ and leads to dephasing. In this article we derive a simple fomrula to account for this process. We apply our model to the specific case of a flux-qubit coupled to the plasma mode of its DC-SQUID. Because of the SQUID internal degrees of freedom (circulating current), the interaction hamiltonian contains two terms, one linear in the oscillator variables which describes an effective inductive coupling between the two circuit, but also a quadratic term due to the flux-dependence of the SQUID Josephson inductance. Moreover, the coupling constants can be tuned over a wide range by changing the SQUID bias current. We study the qubit frequency shift per photon $`\delta \nu _0`$ and find that $`\delta \nu _0=0`$ for non-trivial biasing conditions. When they are fulfilled, the effect of thermal fluctuations on the qubit should be suppressed.
## VI Annex A
Here we show how we evaluate the correlation function $`C(t)=<a^{}(0)a(0)a^{}(t)a(t)>`$. In order to do so, we follow Mandel\_Wolff . We model the damping of the HO by a linear coupling to a bath of harmonic oscillators
$$\begin{array}{ccc}\hfill H& =& H_{HO}+H_{bath}+H_{int}\hfill \\ & =& h\nu _p(a^{}a+\frac{1}{2})+_\omega \mathrm{}\omega (A^{}(\omega )A(\omega )+\frac{1}{2})+_\omega \mathrm{}[g(\omega )a^{}A(\omega )+g^{}(\omega )A^{}(\omega )a]\hfill \end{array}$$
(46)
Under the assumption that the bath has a short memory, it can be shown that the evolution of the HO variables in the Heisenberg representation is given by
$$\dot{a}=(i2\pi \nu _p\kappa /2)a(t)F(t)$$
(47)
where $`F(t)=i_\omega g(\omega )A(\omega ,0)\mathrm{exp}(i\omega t)`$ is a quantum-mechanical operator describing the random force acting on the HO. The damping rate of the field, which we write $`\kappa /2`$ since in our notations $`\kappa `$ is the damping rate of the intensity stored in the HO, can be related to the bath parameters. Introducing the density of states $`\eta (\omega )`$ we have $`\kappa /2=\pi \eta (2\pi \nu _p)|g(2\pi \nu _p)|^2`$. Integrating equation 47 we obtain
$$a(t)=a(0)\mathrm{exp}(i2\pi \nu _p\kappa /2)t\mathrm{exp}(i2\pi \nu _p\kappa /2)t_O^tF(t^{})\mathrm{exp}(i2\pi \nu _p+\kappa /2)t^{}𝑑t^{}$$
(48)
This allows us to calculate $`C(t)=<a^{}(0)a(0)a^{}(t)a(t)>`$
$$\begin{array}{ccc}\hfill C(t)& =& <a^{}(0)a(0)a^{}(t)a(t)>\hfill \\ & =& <n(0)^2>\mathrm{exp}(\kappa t)\hfill \\ & & \mathrm{exp}(\kappa t)_0^t<a^{}(0)a(0)F^{}(t^{})a(0)>\mathrm{exp}(i2\pi \nu _p+(\kappa /2))t^{}\hfill \\ & +& \mathrm{exp}(\kappa t)<a^{}(0)a(0)a^{}(0)F(t^{})>\mathrm{exp}(i2\pi \nu _p+(\kappa /2))t^{}\hfill \\ & +& \mathrm{exp}(\kappa t)_0^t_0^t<a^{}(0)a(0)F^{}(t^{})F(t^{\prime \prime })>\mathrm{exp}[i2\pi \nu _p(t^{}t^{\prime \prime })]\mathrm{exp}[(\kappa /2)(t^{}+t^{\prime \prime })]dt^{}dt^{\prime \prime })\hfill \end{array}$$
(49)
In this equation, the second and third term vanish. Indeed,
$$\begin{array}{ccc}\hfill <a^{}(0)a(0)F^{}(t^{})a(0)>& =& i_\omega g(\omega )\mathrm{exp}^{i\omega t}<a^{}(0)a(0)a^{}(0)A(\omega ,0)>\hfill \\ & =& i_\omega g(\omega )\mathrm{exp}^{i\omega t}<a^{}(0)a(0)a^{}(0)><A(\omega ,0)>\hfill \end{array}$$
(50)
since it is assumed that at time $`t=0`$ the bath and the HO are uncorrelated. For a bath at thermal equilibrium, $`<A(\omega ,0)>=0`$ so that $`<a^{}(0)a(0)F^{}(t^{})a(0)>=0`$. The same reasoning holds of course to show that $`<a^{}(0)a(0)a^{}(0)F(t^{})>=0`$ as well. Using the fact that $`<F^{}(t)F(t^{\prime \prime })>=\kappa N(2\pi \nu _p)\delta (t^{}t^{\prime \prime })`$ Mandel\_Wolff , where $`N(\omega )=<A^{}(\omega ,0)A(\omega ,0)>`$, we can calculate the last term :
$$\begin{array}{ccc}& & \mathrm{exp}(\kappa t)_0^t_0^t<a^{}(0)a(0)F^{}(t^{})F(t^{\prime \prime })>\mathrm{exp}[i2\pi \nu _p(t^{}t^{\prime \prime })]\mathrm{exp}[(\kappa /2)(t^{}+t^{\prime \prime })]𝑑t^{}𝑑t^{\prime \prime }\hfill \\ & =& <n(0)>\kappa N(2\pi \nu _p)\mathrm{exp}(\kappa t)_0^t_0^t\delta (t^{}t^{\prime \prime })\mathrm{exp}[i2\pi \nu _p(t^{}t^{\prime \prime })]\mathrm{exp}[\kappa (t^{}+t^{\prime \prime })/2]𝑑t^{}𝑑t^{\prime \prime }\hfill \\ & =& <n(0)>\kappa N(2\pi \nu _p)\mathrm{exp}(\kappa t)_0^t\mathrm{exp}(\kappa t^{})𝑑t^{}=N(2\pi \nu _p)<n(0)>(1\mathrm{exp}(\kappa t))\hfill \end{array}$$
(51)
Since we assume that the HO and the bath are permanently in thermal equilibrium, $`N(2\pi \nu _p)=<n(0)>=\overline{n}`$, whereas $`<n^2(0)><n(0)>^2=\overline{n}(\overline{n}+1)`$ (non-poissonian photon statistics of a thermal field). Therefore we obtain
$`C(\tau )`$ $`=`$ $`[<n^2(0)><n(0)>^2]\mathrm{exp}(\kappa \tau )+<n(0)>^2`$ (52)
$`=`$ $`\overline{n}(\overline{n}+1)\mathrm{exp}(\kappa \tau )+\overline{n}^2`$ |
warning/0507/math-ph0507068.html | ar5iv | text | # Riemann–Finsler and Lagrange Gerbes and the Atiyah–Singer Theorems
## 1 Introduction
Connections and curving on gerbes (abelian and, more recently, nonabelian ones) play an important role in modern differential geometry and mathematical physics. Gerbes enabled with connection structure were introduced as a natural higher-order generalization of abelian bundles with connection provided a new possible framework to generalized gauge theories. They appeared in algebraic geometry , and were subsequently developed by Brylinski , see a review of results in . Bundles and gerbes and their higher generalizations ($`n`$–gerbes) can be understood both in two equivalent terms of local geometry (local functions and forms) and of non-local geometry (holonomies and parallel transports) .
The first applications of gerbes formalism were considered for higher Yang–Mills fields and gravity and for a special case of a topological quantum field theory . The approaches were renewed following Hitchin with further applications in physics, for instance, in investigating anomalies , new geometrical structures in string theory and Chern–Simons theory .
A motivation for noncommutative gerbes , related to deformation quantization , follows from the noncommutative description of D–branes in the presence of topologically non–trivial background fields. In a more general context, the geometry of commutative and noncommutative gerbes may be connected to the nonholonomic frame method in (non) commutative gauge realizations and generalizations of the Einstein gravity , nonholonomic deformations with noncommutative and/or algebroid symmetry and to the geometry of Lagrange–Fedosov nonholonomic manifolds . Here, we note that a manifold is nonholonomic (equivalently, anholonomic) if it is provided with a nonintegrable global distribution. In our works, we restrict the constructions to a subclass of such nonholonomic manifolds, or bundle spaces, when their nonholonomic distribution defines a nonlinear connection (in brief, N–connection) structure. We use the term of N–anholonomic manifold for such spaces. The geometry of N–connections came from the Finsler and Lagrange geometry (see, for instance, details in Ref. , and Refs. related to the Ehresmann connection and geometrization of classical mechanics and field theory). Nevertheless, the N–connection structures have to be introduced in general relativity, string theory and on Riemann–Cartan and/or noncommutative spaces and various types of non–Riemannian spaces if generic off–diagonal metrics, nonholonomic frames and genalized connections are introduced into consideration, see discussion and references in .
The concept of N–anholonomic spaces unifies a large class of nonholonomic manifolds and bundle spaces, nonholonomic Einstein, non–Riemannian and Lagrange–Finsler geometries which are present in modern gravity and string theory, geometric mechanics and classical field theory and geometric quantization formalism. N–anholonomic spaces are naturally provided with certain canonical N–connection and linear connection structures, Sasaki type metric, almost complex and/or almost sympletic structures induced correspondingly by the Lagrange, or Finsler, fundamental functions, and for gravitational models by the generic off–diagonal metric terms. The N–connection curvatures and Riemannian curvatures are very useful to study the topology of such manifolds.
The study of bundles of spinors on N–anholonomic spaces provides a number of geometric and physical results. For instance, it was possible to give a definition of nonholonomic spinor structures for Finsler spaces , to get a spinor interpretation of Lagrange and Hamilton spaces (and their higher order extensions), to define N–anholonomic Dirac operators in connection to noncommutative extensions of Finsler–Lagrange geometry and to construct a number of exact solutions with nonholonomic solitons and spinor interactions . But every compact manifold (being an holonomic or anhlolonomic one) is not spin. The obstruction to the existence of spin structure, in general, of any nonholonomic spin structure of a nonholonomic space, is the second Stiefel–Whitney class. This class is also the classifying cocycle associated to a $`/2`$ gerbe, which with respect to nonholonomic manifolds is a nonholonomic gerbe. This way one can be constructed a number of new examples of gerbes (Finsler and/or Lagrange ones, parametrizing some higher symmetries of generic off–diagonal solutions in Einstein and string/ brane gravity, with additional noncommutative and/or algebroid symmetries).
The aim of this paper is to study the main geometric properties of nonholonomic gerbes. We shall generalize the Lichnerowicz theorem and prove Atiyah–Singer type theorems for nonholonomic gerbes. For trivial holonomic manifolds, our results will transform into certain similar ones from Refs. but not completely if there are considered ’non-perturbative’ and nonlinear configurations as exact solutions in gravity models, see details in .
We note that the problem of formulating and proof of Atiyah–Singer type theorems for nonholonomic manifolds is not a trivial one. For instance, in Ref. , it is advocated the point that it is not possible to define the concept of curvature for general nonholonomic manifolds. Without curvatures, one can not be formulated any types of Atiyah–Singer theorems. In Refs. one proposed such definitions for supermanifolds when supersymmetric structure is treated as nonholonomic distribution. There is a long term history on defining torsions and curvatures for various classes of nonholonomic manifolds (see, for instance, Refs. ). More recently, one concluded that such definitions can be given by using the concept of N–connection structure at least for Lagrange–Finsler and Hamilton–Cartan spaces . We note that the problem of definition of curvatures was discussed and solved (also by using the N–connection formalism) in modern approaches to the geometry of noncommutative Riemann–Finsler or Einstein spaces and generalizations , Fedosov N–anholonomic manifolds , as well in the works on nonholonomic Clifford structures and spinors and generalized Finsler superspaces . The Lichnerowicz type formula and Atiyah–Singer theorems to be proved here for N–anholonomic spaces have strong relations to the mentioned classes of manifolds and supermanifolds.
The article is organized as follows. In section 2 we recall the main results on nonholonomic manifolds provided with N–connection structure, consider some examples of such N–anholonomic spaces (generalized Lagrange/Finsler spaces and Riemann–Cartan manifolds provided with N–connections) and define the concept of nonholonomic gerbe. Section 3 is a study of nonhlonomic Clifford gerbes: we consider lifts and nonholonomic vector gerbes and study pre–Hilbertian and scalar structures and define distinguished (by the N–connection structure) linear connections on N–anholonomic gerbes and construct the characteristic classes. Section 4 is devoted to operators and symbols on nonholnomic gerbes. In section 5 we present a $`K`$–theory framework for N–anholonomic manifolds and gerbes. We define elliptic operators and index formulas adapted to the N–connection structure. There are proven the Atiyah–Singer theorems for the canonical d–connection and N–connection structures. The results are applied for a topological study of N–anholonomic spinors and related Dirac operators. Appendix A outlines the basic results on distinguished connections, torsions and curvatures for N–anholonomic manifolds. Appendix B is an introduction into the geometry of nonholonomic spinor structures and N–adapted spin connections.
Acknowledgments: This paper contains a series of results elaborated during long term visits at CSIC, Madrid, Spain and Fields Institute, Toronto, Canada. Some important ideas and theorems were communicated in a review lecture at the VII-th International Conference Finsler Extensions of Relativity Theory - FERT 2011, 29 Aug - 4 Sep 2011, Brasov - Romania, September 2, 2011. Author is grateful to M. Anastasiei, G. Munteanu and D. Pavlov, N. Voicu and M. Neagu for kind support of his participation, hospitality and important discussions. The research in this paper is partially supported by the Program IDEI, PN-II-ID-PCE-2011-3-0256.
## 2 Nonholonomic Manifolds and Gerbes
The aim of this section is to outline some results from the geometry of nonholonomic manifolds provided with N–connection structure and to elaborate the notion of nonholonomic gerbes.
### 2.1 The geometry of N–anholonomic spaces
We consider $`(n+m)`$–dimensional manifold of necessary smoothly class $`𝐕`$ with locally fibred structure. A particular case is that of a vector bundle, when we shall write $`𝐕=𝐄`$ (where $`𝐄`$ is the total space of a vector bundle $`\pi :`$ $`𝐄M`$ with the base space $`M).`$ We denote by $`\pi ^{}:T𝐕TM`$ the differential of a map $`\pi :V^{n+m}V^n`$ defined by fiber preserving morphisms of the tangent bundles $`T𝐕`$ and $`TM.`$ The kernel of $`\pi ^{}`$ is just the vertical subspace $`v𝐕`$ with a related inclusion mapping $`i:v𝐕T𝐕.`$
###### Definition 2.1
A nonlinear connection (N–connection) $`𝐍`$ on a manifold $`𝐕`$ is defined by the splitting on the left of an exact sequence
$$0v𝐕\stackrel{𝑖}{}T𝐕T𝐕/v𝐕0,$$
i. e. by a morphism of submanifolds $`𝐍:T𝐕v𝐕`$ such that $`𝐍𝐢`$ is the unity in $`v𝐕.`$
In an equivalent form, we can say that a N–connection is defined by a splitting to subspaces with a Whitney sum of conventional horizontal (h) subspace, $`\left(h𝐕\right),`$ and vertical (v) subspace, $`\left(v𝐕\right),`$
$$T𝐕=h𝐕v𝐕$$
(1)
where $`h𝐕`$ is isomorphic to $`M.`$ In general, a distribution (1) in nonintegrabe, i.e. nonholonomic (equivalently, anholonomic). In this case, we deal with nonholonomic manfiolds/ spaces.
###### Definition 2.2
A manifold $`𝐕`$ is called N–anholonomic if on the tangent space $`T𝐕`$ it is defined a local (nonintegrable) distribution (1), i.e. $`𝐕`$ is N–anholonomic if it is enabled with a N–connection structure.
Locally, a N–connection is defined by its coefficients $`N_i^a(u),`$
$$𝐍=N_i^a(u)dx^i_a$$
where the local coordinates (in general, abstract ones both for holonomic and nonholonomic variables) are split in the form $`u=(x,y),`$ or $`u^\alpha =(x^i,y^a),`$ where $`i,j,k,\mathrm{}=1,2,\mathrm{},n`$ and $`a,b,c,\mathrm{}=n+1,n+2,\mathrm{},n+m`$ when $`_i=/x^i`$ and $`_a=/y^a.`$ The well known class of linear connections consists on a particular subclass with the coefficients being linear on $`y^a,`$ i.e., $`N_i^a(u)=\mathrm{\Gamma }_{bj}^a(x)y^b.`$
A N–connection is characterized by its N–connection curvature (the Nijenhuis tensor)
$$𝛀=\frac{1}{2}\mathrm{\Omega }_{ij}^adx^idx^j_a,$$
with the N–connection curvature coefficients
$$\mathrm{\Omega }_{ij}^a=\delta _{[j}N_{i]}^a=\delta _jN_i^a\delta _iN_j^a=_jN_i^a_iN_j^a+N_i^b_bN_j^aN_j^b_bN_i^a.$$
(2)
Any N–connection $`𝐍=N_i^a(u)`$ induces a N–adapted frame (vielbein) structure
$$𝐞_\nu =\left(e_i=_iN_i^a(u)_a,e_a=_a\right),$$
(3)
and the dual frame (coframe) structure
$$𝐞^\mu =\left(e^i=dx^i,e^a=dy^a+N_i^a(u)dx^i\right).$$
(4)
The vielbeins (4) satisfy the nonholonomy (equivalently, anholonomy) relations
$$[𝐞_\alpha ,𝐞_\beta ]=𝐞_\alpha 𝐞_\beta 𝐞_\beta 𝐞_\alpha =W_{\alpha \beta }^\gamma 𝐞_\gamma $$
(5)
with (antisymmetric) nontrivial anholonomy coefficients $`W_{ia}^b=_aN_i^b`$ and $`W_{ji}^a=\mathrm{\Omega }_{ij}^a.`$<sup>1</sup><sup>1</sup>1One preserves a relation to our previous denotations if we consider that $`𝐞_\nu =(e_i,e_a)`$ and $`𝐞^\mu =(e^i,e^a)`$ are, respectively, the former $`\delta _\nu =\delta /u^\nu =(\delta _i,_a)`$ and $`\delta ^\mu =\delta u^\mu =(d^i,\delta ^a)`$ when emphasize that operators (3) and (4) define, correspondingly, the “N–elongated” partial derivatives and differentials which are convenient for calculations on N–anholonomic manifolds.
The geometric constructions can be adapted to the N–connection structure:
###### Definition 2.3
A distinguished connection (d–connection) $`𝐃`$ on a N–anholonomic manifold $`𝐕`$ is a linear connection conserving under parallelism the Whitney sum (1).
In this work we use boldfaced symbols for the spaces and geometric objects provided/adapted to a N–connection structure. For instance, a vector field $`𝐗T𝐕`$ is expressed $`𝐗=(X,^{}X),`$ or $`𝐗=X^\alpha 𝐞_\alpha =X^ie_i+X^ae_a,`$ where $`X=X^ie_i`$ and $`{}_{}{}^{}X=X^ae_a`$ state, respectively, the irreducible (adapted to the N–connection structure) horizontal (h) and vertical (v) components of the vector (which following Refs. is called a distinguished vectors, in brief, d–vector). In a similar fashion, the geometric objects on $`𝐕`$ like tensors, spinors, connections, … are called respectively d–tensors, d–spinors, d–connections if they are adapted to the N–connection splitting.
One can introduce the d–connection 1–form
$$𝚪_\beta ^\alpha =𝚪_{\beta \gamma }^\alpha 𝐞^\gamma ,$$
when the N–adapted components of d-connection $`𝐃_\alpha =(𝐞_\alpha 𝐃)`$ are computed following formulas
$$𝚪_{\alpha \beta }^\gamma \left(u\right)=\left(𝐃_\alpha 𝐞_\beta \right)𝐞^\gamma ,$$
(6)
where ”$`\mathrm{"}`$ denotes the interior product. This allows us to define in standard form the torsion
$$𝒯^\alpha \mathrm{𝐃𝐞}^\alpha =d𝐞^\alpha +\mathrm{\Gamma }_\beta ^\alpha 𝐞^\beta $$
(7)
and curvature
$$_\beta ^\alpha 𝐃𝚪_\beta ^\alpha =d𝚪_\beta ^\alpha \mathrm{\Gamma }_\beta ^\gamma 𝚪_\gamma ^\alpha .$$
(8)
There are certain preferred d–connection structures on N–anholonomic manifolds (see local formulas in Appendix and Refs. , for details on computation the components of torsion and curvatures for various classes of d–connections).
### 2.2 Examples of N–anholonomic spaces:
We show how the N–connection geometries can be naturally derived from Lagrange–Finsler geometry and in gravity theories.
#### 2.2.1 Lagrange–Finsler geometry
Such geometries are usually modelled on tangent bundles but it is possible to define such structures on general N–anholonomic manifolds, in particular in (pseudo) Riemannian and Riemann–Cartan geometry if nonholonomic frames are introduced into consideration . In the first approach the N–anholonomic manifold $`𝐕`$ is just a tangent bundle $`(TM,\pi ,M),`$ where $`M`$ is a $`n`$–dimensional base manifold, $`\pi `$ is a surjective projection and $`TM`$ is the total space. One denotes by $`\stackrel{~}{TM}=TM\backslash \{0\}`$ where $`\{0\}`$ means the null section of map $`\pi .`$
A differentiable Lagrangian $`L(x,y),`$ i. e. a fundamental Lagrange function, is defined by a map $`L:(x,y)TML(x,y)`$ of class $`𝒞^{\mathrm{}}`$ on $`\stackrel{~}{TM}`$ and continuous on the null section $`0:MTM`$ of $`\pi .`$ For simplicity, we consider any regular Lagrangian with nondegenerated Hessian
$${}_{}{}^{L}g_{ij}^{}(x,y)=\frac{1}{2}\frac{^2L(x,y)}{y^iy^j}$$
(9)
when $`rank\left|g_{ij}\right|=n`$ on $`\stackrel{~}{TM}`$ and the left up ”L” is an abstract label pointing that the values are defined by the Lagrangian $`L.`$
###### Definition 2.4
A Lagrange space is a pair $`L^n=[M,L(x,y)]`$ with $`{}_{}{}^{L}g_{ij}^{}(x,y)`$ being of constant signature over $`\stackrel{~}{TM}.`$
The notion of Lagrange space was introduced by J. Kern and elaborated in details in Ref. as a natural extension of Finsler geometry.
By straightforward calculations, one can be proved the fundamental results:
1. The Euler–Lagrange equations
$$\frac{d}{d\tau }\left(\frac{L}{y^i}\right)\frac{L}{x^i}=0$$
where $`y^i=\frac{dx^i}{d\tau }`$ for $`x^i(\tau )`$ depending on parameter $`\tau ,`$ are equivalent to the “nonlinear” geodesic equations
$$\frac{d^2x^i}{d\tau ^2}+2G^i(x^k,\frac{dx^j}{d\tau })=0$$
defining paths of the canonical semispray
$$S=y^i\frac{}{x^i}2G^i(x,y)\frac{}{y^i}$$
where
$$2G^i(x,y)=\frac{1}{2}^Lg^{ij}\left(\frac{^2L}{y^ix^k}y^k\frac{L}{x^i}\right)$$
with $`{}_{}{}^{L}g_{}^{ij}`$ being inverse to (9).
2. There exists on $`\stackrel{~}{TM}`$ a canonical N–connection
$${}_{}{}^{L}N_{j}^{i}=\frac{G^i(x,y)}{y^i},$$
(10)
defined by the fundamental Lagrange function $`L(x,y),`$ prescribing nonholonomic frame structures of type (3) and (4), $`{}_{}{}^{L}𝐞_{\nu }^{}=(e_i,^{}e_k)`$ and $`{}_{}{}^{L}𝐞_{}^{\mu }=(e^i,^{}e^k).`$ <sup>2</sup><sup>2</sup>2On the tangent bundle the indices related to the base space run the same values as those related to fibers: we can use the same symbols but have to distinguish like $`{}_{}{}^{}e_{k}^{}`$ certain irreducible v–components with respect to, (or for) N–adapted bases and co–bases.
3. The canonical N–connection (10), defining $`{}_{}{}^{}e_{i}^{},`$ induces naturally an almost complex structure $`𝐅:\chi (\stackrel{~}{TM})\chi (\stackrel{~}{TM}),`$ where $`\chi (\stackrel{~}{TM})`$ denotes the module of vector fields on $`\stackrel{~}{TM},`$
$$𝐅(e_i)=^{}e_i\text{ and }𝐅(^{}e_i)=e_i,$$
when
$$𝐅=^{}e_ie^ie_i^{}e^i$$
(11)
satisfies the condition $`𝐅𝐅=𝐈,`$ i. e. $`F_\beta ^\alpha F_\gamma ^\beta =\delta _\gamma ^\alpha ,`$ where $`\delta _\gamma ^\alpha `$ is the Kronecker symbol and “$``$” denotes the interior product.
4. On $`\stackrel{~}{TM},`$ there is a canonical metric structure
$${}_{}{}^{L}𝐠=^Lg_{ij}(x,y)e^ie^j+^Lg_{ij}(x,y)^{}e^i^{}e^j$$
(12)
constructed as a Sasaki type lift from $`M.`$
5. There is also a canonical d–connection structure $`\widehat{𝚪}_{\alpha \beta }^\gamma `$ defined only by the components of $`{}_{}{}^{L}N_{j}^{i}`$ and $`{}_{}{}^{L}g_{ij}^{},`$ i.e. by the coefficients of metric (12) which in its turn is induced by a regular Lagrangian. The d–connection $`\widehat{𝚪}_{\alpha \beta }^\gamma `$ is metric compatible and with vanishing $`h`$\- and $`v`$–torsions. Such a d–connection contains also nontrivial torsion components induced by the nonholonomic frame structure, see Proposition 5.2 and formulas (A.3) in Appendix. The canonical d–connection is the ”simplest” N–adapted linear connection related by the ”non N–adapted” Levi–Civita connection by formulas (A.2).
We can conclude that any regular Lagrange mechanics can be geometrized as an almost Kähler space with N–connection distribution, see . For the Lagrange–Kähler (nonholonomic) spaces, the fundamental geometric structures (semispray, N–connection, almost complex structure and canonical metric on $`\stackrel{~}{TM})`$ are defined by the fundamental Lagrange function $`L(x,y).`$
For applications in optics of nonhomogeneous media and gravity (see, for instance, Refs. ) one considers metrics of type $`g_{ij}e^{\lambda (x,y)}{}_{}{}^{L}g_{ij}^{}(x,y)`$ which can not be derived from a mechanical Lagrangian but from an effective ”energy” function. In the so–called generalized Lagrange geometry one considers Sasaki type metrics (12) with any general coefficients both for the metric and N–connection.
###### Remark 2.1
A Finsler space is defined by a fundamental Finsler function $`F(x,y),`$ being homogeneous of type $`F(x,\lambda y)=\lambda F(x,y),`$ for nonzero $`\lambda ,`$ may be considered as a particular case of Lagrange geometry when $`L=F^2.`$
Now we show how N–anholonomic configurations can defined in gravity theories. In this case, it is convenient to work on a general manifold $`𝐕,dim𝐕=n+m`$ with global splitting, instead of the tangent bundle $`\stackrel{~}{TM}.`$
#### 2.2.2 N–connections and gravity
Let us consider a metric structure on $`𝐕`$ with the coefficients defined with respect to a local coordinate basis $`du^\alpha =(dx^i,dy^a),`$
$$𝐠=\underset{¯}{g}_{\alpha \beta }(u)du^\alpha du^\beta $$
with
$$\underset{¯}{g}_{\alpha \beta }=\left[\begin{array}{cc}g_{ij}+N_i^aN_j^bh_{ab}& N_j^eh_{ae}\\ N_i^eh_{be}& h_{ab}\end{array}\right].$$
(13)
In general, such a metric (13) is generic off–diagonal, i.e it can not be diagonalized by any coordinate transforms. We not that $`N_i^a(u)`$ in our approach are any general functions. They my be identified with some gauge potentials in Kaluza–Klein models if the corresponding symmetries and compactifications of coordinates $`y^a`$ are considered, see review . Performing a frame transform
$$𝐞_\alpha =𝐞_\alpha ^{\underset{¯}{\alpha }}_{\underset{¯}{\alpha }}\text{ and }𝐞_{}^\beta =𝐞_{\underset{¯}{\beta }}^\beta du^{\underset{¯}{\beta }}.$$
with coefficients
$`𝐞_\alpha ^{\underset{¯}{\alpha }}(u)`$ $`=`$ $`\left[\begin{array}{cc}e_i^{\underset{¯}{i}}(u)& N_i^b(u)e_b^{\underset{¯}{a}}(u)\\ 0& e_a^{\underset{¯}{a}}(u)\end{array}\right],`$ (16)
$`𝐞_{\underset{¯}{\beta }}^\beta (u)`$ $`=`$ $`\left[\begin{array}{cc}e_{\underset{¯}{i}}^i(u)& N_k^b(u)e_{\underset{¯}{i}}^k(u)\\ 0& e_{\underset{¯}{a}}^a(u)\end{array}\right],`$ (19)
we write equivalently the metric in the form
$$𝐠=𝐠_{\alpha \beta }\left(u\right)𝐞^\alpha 𝐞^\beta =g_{ij}\left(u\right)e^ie^j+h_{ab}\left(u\right)^{}e^a^{}e^b,$$
(20)
where $`g_{ij}𝐠(e_i,e_j)`$ and $`h_{ab}𝐠(e_a,e_b)`$ and the vielbeins $`𝐞_\alpha `$ and $`𝐞^\alpha `$ are respectively of type (3) and (4). We can consider a special class of manifolds provided with a global splitting into conventional “horizontal” and “vertical” subspaces (1) induced by the “off–diagonal” terms $`N_i^b(u)`$ and prescribed type of nonholonomic frame structure.
If the manifold $`𝐕`$ is (pseudo) Riemannian, there is a unique linear connection (the Levi–Civita connection) $`,`$ which is metric, $`𝐠=\mathrm{𝟎},`$ and torsionless, $`{}_{}{}^{}T=0.`$ Nevertheless, the connection $``$ is not adapted to the nonintegrable distribution induced by $`N_i^b(u).`$ In this case, it is more convenient to work with more general classes of linear connections (for instance, with the canonical d–connection (A.3)) which are N–adapted but contain nontrivial torsion coefficients because of nontrivial nonholonomy coefficients $`W_{\alpha \beta }^\gamma `$ (5).
For a splitting of a (pseudo) Riemannian–Cartan space of dimension $`(n+m)`$ ( we considered also certain (pseudo) Riemannian configurations), the Lagrange and Finsler type geometries were modelled by N–anholonomic structures as exact solutions of gravitational field equations .
### 2.3 The notion of nonholonomic gerbes
Let denote by $`𝐒`$ a sheaf of categories on a N–anholonomic manifold $`𝐕,`$ defined by a map of $`𝐔𝐒(𝐔),`$ where $`𝐔`$ is a open subset of $`𝐕,`$ with $`𝐒(𝐔).`$
###### Definition 2.5
A sheaf of categories $`𝐒`$ is called a nonholonomic gerbe if there are satisfied the conditions:
1. There exists a map $`r_{𝐔_{\widehat{1}}𝐔_{\widehat{2}}}:𝐒(𝐔_{\widehat{1}})𝐒(𝐔_{\widehat{2}})`$ such that for superpositions of two such maps $`r_{𝐔_{\widehat{1}}𝐔_{\widehat{2}}}r_{𝐔_{\widehat{2}}𝐔_{\widehat{3}}}=r_{𝐔_{\widehat{1}}𝐔_{\widehat{3}}}`$ for any inclusion $`𝐔_{\widehat{1}}𝐔_{\widehat{2}}.`$
2. It is satisfied the gluing condition for objects, i.e. for a covering family $`_{\widehat{i}}𝐔_{\widehat{i}}`$ of $`𝐔`$ and objects $`𝐮_{\widehat{i}}`$ of $`𝐒(𝐔_{\widehat{i}})`$ for each $`\widehat{i},`$ when there are maps of type
$$q_{\widehat{i}\widehat{j}}:r_{𝐔_{\widehat{i}}𝐔_{\widehat{j}},𝐔_{\widehat{j}}}\left(𝐮_{\widehat{j}}\right)r_{𝐔_{\widehat{i}}𝐔_{\widehat{j}},𝐔_{\widehat{i}}}\left(𝐮_{\widehat{i}}\right)$$
such that $`q_{\widehat{i}\widehat{j}}q_{\widehat{j}\widehat{k}}=q_{\widehat{i}\widehat{k}},`$ then there exists and object $`𝐮𝐒(𝐔)`$ such that $`r_{𝐔_{\widehat{i}},𝐔}\left(𝐮\right)𝐮_{\widehat{i}}.`$
3. It is satisfied the gluing condition for arrows, i.e. for any two objects $`𝐏,𝐐𝐒(𝐕)`$ the map
$$𝐔Hom(r_{\mathrm{𝐔𝐕}}(𝐏),r_{\mathrm{𝐔𝐕}}(𝐐))$$
is a sheaf.
This Definition is adapted to the N–connection structure (1) and define similar objects and maps for h– and v–subspaces of a N–anholonomic manifold $`𝐕.`$ That why we use ”boldfaced” symbols.
For certain applications is convenient to work with another sheaf $`𝐀`$ called the N–bund of the nonholonomic gerbe $`𝐒.`$ It is constructed to satisfy the conditions:
* There is a covering N–adapted family $`\left(𝐔_{\widehat{i}}\right)_{\widehat{i}I}`$ of $`𝐕`$ such that the category of $`𝐒(𝐔_{\widehat{i}})`$ is not empty for each $`\widehat{i}.`$
* For any $`𝐮_{(1)},𝐮_{(2)}𝐔𝐕,`$ there is a covering family $`\left(𝐔_{\widehat{i}}\right)_{\widehat{i}I}`$ of $`𝐔`$ such that $`r_{𝐔_{\widehat{i}}𝐔}(𝐮_{(1)})`$ and $`r_{𝐔_{\widehat{i}}𝐔}(𝐮_{(2)})`$ are isomorphic.
* The N–bund is introduced as a family of isomorphisms $`𝐀(𝐕)Hom(𝐮,`$ $`𝐮),`$ for each object $`𝐮𝐒(𝐔),`$ defined by a sheaf $`𝐀`$ in groups, for which every arrow of $`𝐒(𝐔)`$ is invertible and such isomorphisms commute with the restriction maps.
For given families $`\left(𝐔_{\widehat{i}}\right)_{\widehat{i}I}`$ of $`𝐕`$ and objects $`𝐮_{\widehat{i}}`$ of $`𝐒(𝐔_{\widehat{i}}),`$ we denote by $`𝐮_{\widehat{i}_1\mathrm{}\widehat{i}_k}^{\widehat{i}}`$ the element $`r_{(𝐔_{\widehat{i}_1}\mathrm{}𝐔_{\widehat{i}_k},𝐔_{\widehat{j}})}\left(𝐮_{\widehat{i}}\right)`$ and by $`𝐔_{\widehat{i}_1\mathrm{}\widehat{i}_k}`$ the elements of the intersection $`𝐔_{\widehat{i}_1}\mathrm{}𝐔_{\widehat{i}_k}.`$ The N–connection structure distinguishes (d) $`𝐕`$ into h– and v–components, i.e. defines a local fiber structure when the geometric objects transform into d–objects, for instance, d–vectors, d–tensors,….. There are two possibilities for further constructions: a) to consider the category of vector bundles over an open set $`𝐔`$ of N–anholonomic manifold $`𝐕,`$ being the base space or b) to consider such N–anholonomic vector bundles modelled as $`𝐕=𝐄`$ with a base $`M,`$ where $`dimM=n`$ and $`dim𝐄=n+m.`$
###### Definition 2.6
a) A N–anholonomic vector gerbe $`𝐂_{NQ}`$ is defined by the category of vector bundles $`𝐒(𝐔)`$ over $`𝐔𝐕`$ with typical fiber the vector space $`Q.`$
b) A nonholonomic gerbe $`𝐂_{Nd}`$ is a d–vectorial gerbe $`𝐒(U)`$ if and only if for the each open $`UM`$ on the h–subspace $`M`$ of $`𝐕`$ the set $`𝐒(U)`$ is a category of N–anholonomic manifolds with h–base $`M.`$
In both cases of nonholonomic gerbes a) and b) the maps between d–objects are isomorphisms of N–anholonomic bundles/ manifolds adapted to the N–connection structures.
Let us consider more precisely the case a) (the constructions for the case b) being similar by substituting $`𝐔U`$ and $`𝐕M).`$ There is a covering family $`\left(𝐔_{\widehat{\alpha }}\right)_{\widehat{\alpha }I}`$ of $`𝐕`$ and a commutative subgroup $`H`$ of the set of linear transforms $`Gl(Q),`$ such that there exit maps
$`q_{\widehat{\alpha }\widehat{\beta }}^{}`$ $`:`$ $`𝐔_{\widehat{\alpha }}𝐔_{\widehat{\beta }}\times Q𝐔_{\widehat{\alpha }}𝐔_{\widehat{\beta }}\times Q,`$
$`q_{\widehat{\alpha }\widehat{\beta }}^{}`$ $`:`$ $`(𝐮_{(1)},𝐮_{(2)})(𝐮_{(1)},q_{\widehat{\alpha }\widehat{\beta }}^{}(𝐮_{(1)})𝐮_{(2)})`$
when $`c_{\widehat{\alpha }\widehat{\beta }\widehat{\gamma }}=q_{\widehat{\alpha }\widehat{\beta }}^{}q_{\widehat{\beta }\widehat{\gamma }}^{}q_{\widehat{\gamma }\widehat{\alpha }}^{}`$ is an $`H`$ 2–Cech cocycle. Locally, such maps are parametrized by non–explicit functions because of nonholonomic character of manifolds and subspaces under consideration.
## 3 Nonholonomic Clifford Gerbes
Let $`𝐕`$ be an N–anholnomic manifold of dimension $`dim𝐕=n+m.`$ We denote by $`O(𝐕),`$ see applications and references in , the reduction of linear N–adapted frames which defines the d–metric structure (20) of the $`𝐕.`$ The typical fiber of $`O(𝐕)`$ is $`O(n+m)`$ which with respect to N–adapted frames splits into $`O(n)O(m).`$ There is the exact sequence
$$1/2Spin(n+m)O(n+m)1$$
with two N–distinguished, respectively, h– and v–components
$`1`$ $``$ $`/2Spin(n)O(n)1,`$
$`1`$ $``$ $`/2Spin(m)O(m)1`$
where $`Spin(n+m)`$ is the universal covering of $`O(n+m)`$ splitting into $`Spin(n)Spin(m)`$ distinguished as the universal covering of $`O(n)O(m).`$ To such sequences, one can be associated a nonholnomic gerbe with band $`/2`$ and such that for each open set $`𝐔𝐕`$ it defined $`Spin_N(𝐔)`$ as the category of $`Spin`$ $`N`$–anholonomic bundles over $`𝐔,`$ such spaces were studied in details in Refs. , see also Appendix 6. The classified cocycle of this N–anholonomic gerbe is defined by the second Stiefel–Whitney class.
In a more general context, the N–anholonomic gerbe and $`Spin_N(𝐔)`$ are associated to a vectorial N–ahnolonomic gerbe called the Clifford N–gerbe (in a similar form we can consider associated Clifford d–gerbe, for the case b) of Definition 2.6). This way one defines the category $`Cl_N(𝐕)`$ which for any open set $`𝐔𝐕,`$ one have the category of objects being Clifford bundles provided with N–connection structure associated to the objects of $`Spin_N(𝐔).`$ We can consider such gerbes in terms of transition functions. Let $`q_{\widehat{\alpha }\widehat{\beta }}^{}`$ be the transitions functions of the bundle $`O(𝐕).`$ The N–connection distinguish them to couples of h- and v–transition functions, i.e. $`q_{\widehat{\alpha }\widehat{\beta }}^{}=(q_{\widehat{i}\widehat{j}}^{},q_{\widehat{a}\widehat{b}}^{}).`$ For such d–functions one can be considered elements $`q_{\widehat{\alpha }\widehat{\beta }}=(q_{\widehat{i}\widehat{j}},q_{\widehat{a}\widehat{b}})`$ acting correspondingly in $`Spin(n+m)=(Spin(n),Spin(m)).`$ Such elements act, by left multiplication, correspondingly on $`Cl(^{n+m})`$ distinguished into $`(Cl(^n),Cl(^m)).`$ We denote by $`s_{\widehat{\alpha }\widehat{\beta }}(𝐮)=(s_{\widehat{i}\widehat{j}}(𝐮),s_{\widehat{a}\widehat{b}}(𝐮))`$ the resulting automorphisms on Clifford spaces. We conclude that the Clifford N–gerbe is defined by maps
$$s_{\widehat{\alpha }\widehat{\beta }}:𝐔_{\widehat{\alpha }}𝐔_{\widehat{\beta }}Spin(n+m)$$
distinguished with respect to N–adapted frames by couples
$$s_{\widehat{i}\widehat{j}}:𝐔_{\widehat{i}}𝐔_{\widehat{j}}Spin(n)\text{ and }s_{\widehat{a}\widehat{b}}:𝐔_{\widehat{a}}𝐔_{\widehat{b}}Spin(m).$$
For trivial N–connections, such Clifford N–gerbes transform into the usual Clifford gerbes defined in Ref. . <sup>3</sup><sup>3</sup>3We apply the ideas and results developed in that paper in order to investigate N–anholonomic manifolds and gerbes.
### 3.1 Gerbes and lifts associated to d–vector bundles
There are two classes of nonholonomic gerbes defined by lifting problems, respectively, associated to a vector bundle $`𝐄`$ on a N–anholonomic manifold $`𝐕`$ and/or associated just to $`𝐕`$ considering that locally a such space posses a fibered structure distinguished by the N–connection, see corresponding cases a) and b) in Definition 2.6.
#### 3.1.1 Lifts and N–anholonomic vector gerbes
Let us denote by $`Q`$ the typical fiber of a vector bundle $`𝐄`$ on $`𝐕`$ with associated principal bundle $`Gl(Q).`$ We suppose that this bundle has a reduction $`𝐄_K`$ for a subgroup $`KGl(Q)`$ and consider a central extension for a group $`G`$ when
$$1HGK1.$$
(21)
Such an extension defines a N–anholonomic gerbe $`𝐂_H`$ on $`𝐕`$ when for each open set $`𝐔𝐕`$ the objects of $`𝐂_H(𝐔)`$ are $`G`$–principal bundles over $`𝐔`$ when the quotient by $`H`$ is the restriction of $`𝐄_K`$ to $`𝐔.`$
We consider the projection $`\pi :GK`$ and suppose that for (existing) a representation $`r:GGl(Q^{})`$ and surjection $`f:Q^{}Q`$ one can be defined a commutative Diagram 1,
For such cases it is defined a N–anholonomic vectorial gerbe $`𝐂_{H,Q^{}}`$ on $`𝐕`$ when an object of $`𝐂(𝐔)`$ is parametrized $`𝐞_Ur`$ where $`𝐞_U`$ is an object of $`𝐂_H(𝐔).`$ Such constructions are adapted to the N–connection structure on $`𝐔.`$ If $`\left(𝐔_{\stackrel{~}{\alpha }}\right)_{\stackrel{~}{\alpha }I}`$ is a trivialization of $`𝐄,`$ with transition functions $`q_{\stackrel{~}{\alpha }\stackrel{~}{\beta }}^{}=(q_{\stackrel{~}{i}\stackrel{~}{j}}^{},q_{\stackrel{~}{a}\stackrel{~}{b}}^{}),`$ we can define the maps $`q_{\stackrel{~}{\alpha }\stackrel{~}{\beta }}:𝐔_{\stackrel{~}{\alpha }}𝐔_{\stackrel{~}{\beta }}G`$ over $`q_{\stackrel{~}{\alpha }\stackrel{~}{\beta }}^{}.`$ This states that $`𝐂_{H,Q^{}}`$ is defined by $`r(q_{\stackrel{~}{\alpha }\stackrel{~}{\beta }}).`$
There is a natural scalar product defined on such N–anholonomic gerbes. Its existence follows from the construction of Clifford N–gerbe $`Cl(𝐕)`$ because the group $`Spin`$ is compact and its action on $`Cl(^{n+m})`$ preserves a scalar product which is distinguished by the N–connection structure . We can consider this scalar product on each fiber of an object $`𝐞_U`$ of $`Cl(𝐔)`$ and define a Riemannian d–metric
$$<,>_{e_U}=<,>_{he_U}+<,>_{ve_U}$$
distinguished by the N–splitting into h- and v–components. The family of such Riemannian d–metrics defines the Riemannian d–metric on the N–anholonomic gerbe $`Cl(𝐕).`$ There is a canonical such structure defined by the N–connection when $`(g_{ij},h_{ab})`$ in (20) are taken to be some Euclidean ones but $`N_i^a(𝐮)`$ are the coefficients for a nontrivial N–connection.
It should be emphasized that the band has to be contained in a compact group in order to preserve the Riemannian d–metric. In result, we can give the
###### Definition 3.1
A Riemannian d–metric on a N–anholonomic vector gerbe $`𝐂_{NQ}`$ is given by a distinguished scalar product $`<,>_{e_U}=(<,>_{he_U},<,>_{ve_U})`$ on the vector bundle $`𝐞_U,`$ defined for every object of $`𝐂_{NQ}`$ and preserved by morphisms of such objects.
We can define a global section of a N–anholonomic vector gerbe associated to a 1–Cech N–adapted chain $`q_{\widehat{\alpha }\widehat{\beta }}=(q_{\widehat{i}\widehat{j}},q_{\widehat{a}\widehat{b}})`$ by considering a covering space $`\left(𝐔_{\widehat{\alpha }}\right)_{\widehat{\alpha }I}`$ of $`𝐕`$ when for each element $`\widehat{\alpha }`$ of $`I,`$ an object $`𝐞_{\widehat{\alpha }}𝐂(𝐔_{\widehat{\alpha }}),`$ a section $`z_{\widehat{\alpha }}`$ of $`𝐞_{\widehat{\alpha }}`$ and a family of morphisms $`q_{\widehat{\alpha }\widehat{\beta }}:𝐞_{\widehat{\beta }}^{\widehat{\alpha }}𝐞_{\widehat{\alpha }}^{\widehat{\beta }}`$ one has $`z_{\widehat{\alpha }}=q_{\widehat{\alpha }\widehat{\beta }}(z_{\widehat{\beta }}).`$ The family of global sections $`𝐙\left(q_{\widehat{\alpha }\widehat{\beta }}\right)`$ associated to $`\left(q_{\widehat{\alpha }\widehat{\beta }}\right)_{\widehat{\alpha },\widehat{\beta }I}`$ defining a d–vector space. If $`𝐕`$ is compact and $`I`$ is finite, one can prove that $`𝐙\left(q_{\widehat{\alpha }\widehat{\beta }}\right)`$ is not empty. For such conditions, we can generalize for N–anholonomic vector gerbes the Proposition 5 from Ref. ,
###### Proposition 3.1
Let the N–anholonomic vector gerbe $`𝐂_{NQ}`$ is a nonholonomic gerbe associated to the lifting problem defined by the extension (21) and the vector bundle $`E`$ and for a reprezentation $`r:GGl(Q^{})`$ the conditions of the Diagram 1 are satisfied. Then for each $`G`$–chain $`q_{\widehat{\alpha }\widehat{\beta }}`$ and each element $`\left(z_{\widehat{\alpha }}\right)_{\widehat{\alpha }I}`$ of the d–vector space of global sections $`𝐙\left(q_{\widehat{\alpha }\widehat{\beta }}\right)`$ it is satisfied the condition that there is a section $`𝐳`$ of $`𝐄`$ such that $`𝐳_{|U_{\widehat{\alpha }}}=f`$ $`𝐳_{\widehat{\alpha }}.`$
Proof. It is similar to that for the usual vector bundles given in but it should be considered for both h– and v–subspaces of $`𝐕`$ and $`𝐄.`$ In ”non–distinguished” form, we can consider a global section $`\left(z_{\widehat{\alpha }}\right)_{\widehat{\alpha }I}`$ associate to $`𝐙\left(q_{\widehat{\alpha }\widehat{\beta }}\right).`$ One has $`z_{\widehat{\alpha }}=q_{\widehat{\alpha }\widehat{\beta }}(z_{\widehat{\beta }})`$ implying that $`f(z_{\widehat{\alpha }})=f(z_{\widehat{\beta }})`$ on $`𝐔_{\widehat{\alpha }\widehat{\beta }}.`$ We conclude that the family $`\left[f(z_{\widehat{\alpha }})\right]_{\widehat{\alpha }I}`$ of local sections $`𝐳`$ of $`𝐄,`$ such that $`𝐳_{|U_{\widehat{\alpha }}}=f`$ $`𝐳_{\widehat{\alpha }}`$ being distinguished in N–adapted sections $`𝐳_{|U_{\widehat{i}}}=f`$ $`𝐳_{\widehat{i}}`$ and $`𝐳_{|U_{\widehat{a}}}=f`$ $`𝐳_{\widehat{a}}.\mathrm{}`$
For a chain $`z_{\widehat{\alpha }\widehat{\beta }}=`$ $`z_{\widehat{\alpha }}q_{\widehat{\alpha }\widehat{\beta }}(z_{\widehat{\beta }}),`$ we can construct a 2–cocycle
$$z_{\widehat{\beta }\widehat{\gamma }}z_{\widehat{\alpha }\widehat{\gamma }}+z_{\widehat{\alpha }\widehat{\beta }}=z_{\widehat{\alpha }\widehat{\beta }\widehat{\gamma }}.$$
Nevertheless, even there are a chain $`q_{\widehat{\alpha }\widehat{\beta }}`$ and a global section $`𝐳=\left(𝐳_{\widehat{\alpha }}\right)_{\widehat{\alpha }I}`$ such that $`z_{\widehat{\alpha }}=q_{\widehat{\alpha }\widehat{\beta }}(z_{\widehat{\beta }})`$ and $`f(z_{\widehat{\alpha }})=𝐳_{|U_{\widehat{\alpha }}}`$ it may does not exist a global section for another N–adapted chain. One has to work with the d–vector space $`𝐙`$ of formal global sections of the N–anholonomic vector gerbe $`𝐂_{NQ}.`$ The d–vector space is defined by generators $`\left[𝐳\right]`$ where $`𝐳`$ is an element of a set of global sections
$$𝐙\left(q_{\widehat{\alpha }\widehat{\beta }}\right)=[Z\left(q_{\widehat{i}\widehat{j}}\right),Z\left(q_{\widehat{a}\widehat{b}}\right)].$$
We can consider that any element of the space $`𝐙`$ is defined by a formal finite sum of global sections.
#### 3.1.2 Lifts and d–vector gerbes
The constructions from the previous section were derived for vector bundles on N–anholonomic manifolds. But such a nonholomic manifold in its turn has a local fibered structure resulting in definition of a nonholonomic gerbe $`𝐂_{Nd}`$ as a d–vectorial gerbe. We denote by $`Q^m`$ the typical fiber which can be associated to a N–anholonomic manifold $`𝐕`$ of dimension $`n+m,`$ from the map $`\pi :V^{n+m}V^n,`$ see Definition 2.1. We can also associate a principal bundle $`Gl(Q^m)`$ supposing that this bundle has a reduction $`𝐕_K`$ for a subgroup $`KGl(Q^m)`$ with a central extension of type (21). For such an extension, we define a d–vector gerbe $`𝐂_{dH}`$ on $`h𝐕`$ when for each open set $`Uh𝐕`$ the objects of $`𝐂_{dH}(U)`$ are $`G`$–principal bundles over $`U`$ when the quotient by $`H`$ is the restriction of $`𝐕_K`$ to $`U.`$
For the projection $`\pi :GK`$ and (supposed to exist) representation $`r:GGl(Q^m)`$ and surjection $`f:Q^mQ^m`$ one can be defined a commutative Diagram 2,
By such a Diagram, it is defined a d–vectorial gerbe $`𝐂_{gH,Q^{}}`$ on $`h𝐕`$ when an object of $`𝐂(U)`$ is parametrized $`e_Ur`$ where $`e_U`$ is an object of $`𝐂_{gH}(U).`$ The constructions are adapted to the N–connection structure on $`U.`$ We note that the objects and regions defined with respect to h–subspaces are not boldfaced as those considered for N–anholonomic vector bundles. Stating $`\left(U_{\stackrel{~}{i}}\right)_{\stackrel{~}{i}I}`$ as a trivialization of $`h𝐕,`$ with transition functions $`q_{\stackrel{~}{i}\stackrel{~}{j}}^{},`$ we can define the maps $`q_{\stackrel{~}{i}\stackrel{~}{i}}:U_{\stackrel{~}{i}}U_{\stackrel{~}{i}}G`$ over $`q_{\stackrel{~}{i}\stackrel{~}{j}}^{}.`$ This means that $`𝐂_{gH,Q^{}}`$ is defined by $`r(q_{\stackrel{~}{i}\stackrel{~}{j}}).`$
There is also is a natural scalar product (a particular case of that for N–anholonomic gerbers) in our case defined by d–vector gerbes. We can consider such a scalar product just for the Clifford N–gerbe $`Cl(h𝐕)`$ following the fact that the group $`Spin`$ is compact and its action on $`Cl(^n)`$ preserves a scalar product. We conclude that this scalar product exists for any object $`e_U`$ of $`Cl(U)`$ and that a d–metric (20) states a splitting $`<,>_{e_U}=<,>_{he_U}+<,>_{ve_U}.`$ There are some alternatives: There is a family of Riemannian d–metrics on the d–vector gerbe $`Cl(h𝐕)`$ but this is not adapted to the N–connection structure. One has to apply the concept of d–connection in order to define N–adapted objects. If the d–metric structure is not prescribed, we can introduce a scalar product structure defined by the N–connection when $`(g_{ij},h_{ab})`$ in (20) are taken to be some Euclidean ones but $`N_i^a(𝐮)`$ are the coefficients for a nontrivial N–connection.
###### Definition 3.2
A d–metric (it is connected to a N–anholonomic Riemann–Cartan structure) on a d–vector gerbe $`𝐂_{Nd}`$ is given by a distinguished scalar product $`<,>_{e_U}=(<,>_{he_U},<,>_{ve_U})`$ on the d–vector bundle $`𝐞_U,`$ defined for every object of $`𝐂_{Nd}`$ and preserved by morphisms of such objects.
Following this Definition, for the d–vector gerbes, one holds the Proposition 3.1 and related results.
### 3.2 Pre–Hibertian and d–connection structures
For simplicity, hereafter we shall work only with N–anhlonomic manifolds. We emphasize that the constructions can be extended to vector bundles $`𝐄`$ on such a nonhonomic manifold $`𝐕.`$ The proofs will be omitted if they are similar to those given for holonomic manifolds and vector spaces but (in our case) adapted to the splitting defined by the N–connection structure. We shall point out the nonholonomic character of the constructions. Such computations may be performed directly by applying ”boldfaced” objects.
#### 3.2.1 Distinguished pre–Hilbertian and scalar structures
Let us consider the two elements $`{}_{1}{}^{}z`$ and $`{}_{2}{}^{}z`$ of the d–vector space $`𝐙\left(q_{\widehat{\alpha }\widehat{\beta }}\right)=[Z\left(q_{\widehat{i}\widehat{j}}\right),Z\left(q_{\widehat{a}\widehat{b}}\right)]`$ (we use left low labels which are not indices running values). For a partition of unity $`(𝐔_{\widehat{\alpha }^{}}^{},f_{\widehat{\alpha }^{}})_{\widehat{\alpha }^{}I^{}}`$ subordinated to $`\left(𝐔_{\widehat{\alpha }}\right)_{\widehat{\alpha }I}.`$ <sup>4</sup><sup>4</sup>4this means that for each $`\widehat{\alpha }^{}`$ there is an $`\widehat{\alpha }(\widehat{\alpha }^{})`$ such that $`𝐔_{\widehat{\alpha }^{}}^{}`$ is a subset of $`𝐔_{\widehat{\alpha }(\widehat{\alpha }^{})}`$ Since the support of $`f_{\widehat{\alpha }^{}}`$ is a couple of compact subsets of $`𝐔_{\widehat{\alpha }^{}}^{}`$ distinguished by the N–connection structure, we can consider restrictions of $`{}_{1}{}^{}z_{\widehat{\alpha }(\widehat{\alpha }^{})}^{}`$ and $`{}_{2}{}^{}z_{\widehat{\alpha }(\widehat{\alpha }^{})}^{}`$ to $`𝐔_{\widehat{\alpha }^{}}^{}`$ denoted respectively $`{}_{1}{}^{}z_{\widehat{\alpha }\widehat{\alpha }(\widehat{\alpha }^{})}^{}`$ and $`{}_{2}{}^{}z_{\widehat{\alpha }\widehat{\alpha }(\widehat{\alpha }^{})}^{}.`$ We can calculate the value $`<_1z_{\widehat{\alpha }\widehat{\alpha }(\widehat{\alpha }^{})},_2z_{\widehat{\alpha }\widehat{\alpha }(\widehat{\alpha }^{})}>`$ which is invariant for any partition. This proofs
###### Proposition 3.2
The scalar product
$`<`$ $`{}_{1}{}^{}z,_2z>={\displaystyle \underset{\widehat{\alpha }^{}}{}}{\displaystyle }<f_{\widehat{\alpha }^{}}{}_{1}{}^{}z_{\widehat{\alpha }\widehat{\alpha }(\widehat{\alpha }^{})}^{},f_{\widehat{\alpha }^{}}{}_{2}{}^{}z_{\widehat{\alpha }\widehat{\alpha }(\widehat{\alpha }^{})}^{}>`$
$`=`$ $`{\displaystyle \underset{\widehat{i}^{}}{}}{\displaystyle }<f_{\widehat{i}^{}}{}_{1}{}^{}z_{\widehat{i}\widehat{i}(\widehat{i}^{})}^{},f_{\widehat{i}^{}}{}_{2}{}^{}z_{\widehat{i}\widehat{i}(\widehat{i}^{})}^{}>+{\displaystyle \underset{\widehat{a}^{}}{}}{\displaystyle }<f_{\widehat{a}^{}}{}_{1}{}^{}z_{\widehat{a}\widehat{a}(\widehat{a}^{})}^{},f_{\widehat{a}^{}}{}_{2}{}^{}z_{\widehat{a}\widehat{a}(\widehat{a}^{})}^{}>`$
defines a pre–Hilbert d–structure of $`(𝐙\left(q_{\widehat{\alpha }\widehat{\beta }}\right),<,>).`$
For two formal global sections of d–vector gerbe $`𝐂_{Nd},`$ we can write
$${}_{1}{}^{}z=\left[z_{\beta _1}\right]+\mathrm{}+\left[z_{\beta _p}\right]\text{ and }_2z=\left[z_{\gamma _1}\right]+\mathrm{}+\left[z_{\gamma _p}\right]$$
where $`z_{\beta _p}`$ and $`z_{\gamma _p}`$ are global sections. The scalar product on the space $`𝐙,`$ of formal global sections of the d–vector gerbe $`𝐂_{Nd},`$ can be defined by the rule
$$<\left[{}_{1}{}^{}z\right],\left[{}_{2}{}^{}z\right]>=<_1z,_2z>_{𝐙\left(q_{\widehat{\alpha }\widehat{\beta }}\right)}$$
if $`{}_{1}{}^{}z`$ and $`{}_{2}{}^{}z`$ are elements of the same set of global sections $`𝐙\left(q_{\widehat{\alpha }\widehat{\beta }}\right)`$ and
$$<\left[{}_{1}{}^{}z\right],\left[{}_{2}{}^{}z\right]>=0$$
for the elements belonging to different sets of such global sections.
###### Proposition 3.3
Any element $`𝐳_{\widehat{\alpha }}`$ of the Hilbert completion $`L^2(𝐙\left(q_{\widehat{\alpha }\widehat{\beta }}\right))`$ of the pre–Hilbert d–structure (3.2) is a N–adapted family of $`L^2`$ sections $`𝐳_{\widehat{\alpha }}`$ of $`𝐞_{\widehat{\alpha }}`$ such that $`𝐳_{\widehat{\alpha }}=q_{\widehat{\alpha }\widehat{\beta }}(z_{\widehat{\beta }}).`$
Proof. The proof is similar to that for the Proposition 7 in Ref. and follows defining a corresponding Cauchy sequence $`\left(z^{\widehat{\alpha }}\right)_{\widehat{\alpha }}`$ of
$`(𝐙\left(q_{\widehat{\alpha }\widehat{\beta }}\right),<,>).`$ For nonholonomic configurations, one uses d–metric structures which can be Riemannian or Riemann–Cartan ones depending on the type of linear connection is considered, a not N–adapted, or N–adapted one.$`\mathrm{}`$
We can consider morphisms between d–objects commuting with Laplacian $`^s`$ and define a pre–Hilbertian structure defined by
$$<_1z,_2z>=<^s(_1z),_2z>,$$
where $`^s(_1z)_{\widehat{\alpha }}=^s(_1z_{\widehat{\alpha }}).`$ There is a canonical N–adapted Laplacian structure defined on nonholonomic spaces by using the canonical d–connection structure, see Proposition 5.2 in Appendix. We denote by $`H_s(𝐙\left(q_{\widehat{\alpha }\widehat{\beta }}\right))`$ the Hilbert completion of the pre–Hilbert space constructed by using $`^s.`$
#### 3.2.2 D–connections on N–anholonomic gerbes and characteristic classes
The canonical d–connection structure gives rise to a such connection on each d–object $`𝐞_U`$ of $`𝐂(𝐔)`$ and defined a family of d–connections inducing such a structure on the N–anholonomic gerbe $`𝐂.`$ We consider an open covering $`\left(𝐔_{\widehat{\alpha }}\right)_{\widehat{\alpha }I}`$ of $`𝐕`$ and d–objects $`𝐂(𝐔_{\widehat{\alpha }})`$ as trivial bundles with $`m`$–dimensional fibers. The d–connection $`𝚪_{\widehat{\alpha }}`$ of a d–object $`𝐞_{\widehat{\alpha }}`$ of $`𝐂(𝐞_{\widehat{\alpha }})`$ is defined by a 1–form with coefficients (A.3) on $`T`$ $`𝐔_{\widehat{\alpha }}.`$ The curvature of this d–connection is $`_{\widehat{\alpha }},`$ see (8).
Having defined the curvature of the N–anholonomic gerbe, it is possible to compute the $`2k`$ Chern class of $`𝐞_{\widehat{\alpha }},`$
$$c_{2k}^{\widehat{\alpha }}Tr\left[\left(\frac{i}{2\pi }_{\widehat{\alpha }}\right)^k\right],$$
where $`Tr`$ denotes the trace operation, which is invariant for transforms $`𝐞_{\widehat{\alpha }}𝐞_{\widehat{\alpha }}^{}.`$ In a more general case, we can compute the sum
$$c(𝐂)=c_1(𝐕)+\mathrm{}+c_2(𝐕)$$
for the total Chern form of an N–anholonomic gerbe. This form define locally the total Chern character
$$ch(𝐂)_{𝐔_{\widehat{\alpha }}}=Tr\left[\mathrm{exp}\left(\frac{i}{2\pi }\right)\right].$$
(23)
It should be noted that the formula (23) is defined by a d–metric (20) and its canonical d–connection (A.2) which correspond respectively to the Riamannian metric and the Levi–Civita connection. The notion of connection is not well–defined for general vector gerbes but the existence of Riemannian structures gives a such possibility. In the case of N–anholonomic frame, even a d–metric structure is not stated, we can derive a canonical d–connection configuration by considering a formal d–metric with $`g_{ij}`$ and $`h_{ab}`$ taking diagonal Euclidean values and computing a curvature tensor $`^{[N]}`$ by contracting the N–connection coefficients $`N_i^a`$ and theirs derivatives. As a matter of principle, we can take the N–connection curvature $`\mathrm{\Omega }`$ (2) instead of $`^{[N]}`$ but in this case we shall deal with metric noncommpatible d–connections. Finally, we not that we need at leas to Chern characters, one for the d–connection structure and another one for the N–connection structure in order to give a topological characteristic of N–anholonomic gerbes.
## 4 Operators and Symbols on Nonholonomic Gerbes
On N–anholonomic manifolds we deal with geometrical objects distinguished by a N–connection structure. The aim of this section is to analyze pseudo–differential operators $`𝐃_{\overline{\alpha }}`$ on such spaces.
### 4.1 Operators on N–anholonomic spaces
In local form, the geometric constructions adapted to a N–connection are for open sets of couples $`(^n,^m),`$ or for $`^{n+m}.`$ Let us consider an open set $`𝐔^{n+m}`$ and denote by $`Z^r(𝐔)`$ the set of smooth functions $`p(𝐯,𝐮)`$ defined on $`𝐔\times ^{n+m}`$ satisfying the conditions that for any compact $`𝐔^{}𝐔`$ and every multi–indices $`\alpha `$ and $`\beta `$ one has
$$𝐃^{\overline{\alpha }}𝐃^{\overline{\beta }}p(𝐯,𝐮)<C_{\overline{\alpha },\overline{\beta },𝐔^{}}\left(1+u\right)^{r|\overline{\alpha }|},$$
for $`C_{\overline{\alpha },\overline{\beta },𝐔^{}}=const.`$
###### Definition 4.1
A map $`\widehat{p}`$ of two smooth functions $`k(𝐔)`$ and $`k^{}(𝐔)`$ with compact support defined on $`𝐔,`$ $`\widehat{p}:`$ $`k(𝐔)`$ $``$ $`k^{}(𝐔),`$ such that locally
$$\widehat{p}(f)=p(𝐯,𝐮)\widehat{f}(𝐮)e^{i<𝐯,𝐮>}\delta 𝐮,$$
where $`\widehat{f}`$ is the Fourier transform of function $`f,`$ is called to be a pseudo–differential distinguished operator, in brief pdd–operator.
Now we extend the concept of pdd–operator for a N–anholonomic manifold $`𝐕`$ endowed with d–metric structure (20). In this case, $`k(𝐔)`$ and $`k^{}(𝐔)`$ are smooth sections, with compact support, of $`𝐕`$ provided with local fibered structure. A map $`\widehat{p}`$ is defined for a covering family $`\left(𝐔_{\widehat{\alpha }}\right)_{\widehat{\alpha }I}`$ satisfying the conditions:
1. Any restriction of $`𝐕`$ to $`𝐔_{\widehat{\alpha }}`$ is trivial.
2. The map $`\widehat{p}_{\widehat{\alpha }}:k(𝐔_{\widehat{\alpha }}\times Q^m)`$ $``$ $`k^{}(𝐔_{\widehat{\alpha }}^{}\times Q^m),`$ where the vector space $`Q^m`$ is isomorphic to $`v𝐕,`$ $`dim(v𝐕)=m,`$ defines the restriction of $`\widehat{p}`$ to $`𝐔_{\widehat{\alpha }}.`$ There is a horizontal component of the map, $`\widehat{p}_{\widehat{i}}:k(U_{\widehat{i}}\times Q^m)`$ $``$ $`k^{}(U_{\widehat{i}}^{}\times Q^m)`$
3. For any section $`z^{}`$ over $`h𝐔_{\widehat{\alpha }}=U_{\widehat{i}}`$ and $`z=(z_1,\mathrm{}z_m)=\varphi _a(z^{}),`$ we can define
$$t_b=\underset{a=1}{\overset{a=m}{}}p_{ab}(x^i,v^k)\widehat{z}_a(v^l)e^{i<x,v>}\delta v^j$$
and $`\widehat{p}_a(z^{})=\psi _a^1(t_b),`$ where the carts $`\varphi _a`$ and $`\psi _a`$ are such that
$$\varphi _a(U_{\widehat{i}}\times Q^m)=\psi _a(U_{\widehat{i}}\times Q^m)h𝐔\times ^m$$
and the map $`\widehat{p}_a`$ is defined by a matrix $`p_{ab}`$ defining an operator of degree $`r.`$
4. The values $`t_b,z_a,\varphi _a,\psi _a`$ and $`p_{ab}`$ can be extended to corresponding distinguished objects
$`t_b`$ $``$ $`t_\alpha =(t_i,t_b),z_az_\alpha =(z_i,z_a),`$
$`\varphi _a`$ $``$ $`\varphi _\alpha =(\varphi _i,\varphi _a),\psi _a\psi _\alpha =(\psi _i,\psi _a),p_{ab}p_{\alpha \beta }.`$
###### Definition 4.2
A map $`\widehat{p}`$ satisfying the conditions 1-4 defines a pdd–operator on N–anholonomic manifold $`𝐕`$ provided with d–metric structure (20). For Euclidean values for h- and v–components of d–metric, with respect to N–adapted frames, one gets a pdd–operator generated by the N–connection structure.
The Definition 4.2 can be similarly formulated for N–anholonomic vector bundles $`𝐄𝐕.`$ We denote by $`{}_{}{}^{loc}H_{s}^{}(𝐕,𝐄),`$ with $`s`$ being a positive integer, the space of distributions sections $`\underset{¯}{𝐮}`$ of $`𝐄`$ such that $`𝐃(\underset{¯}{𝐮})`$ is a $`{}_{}{}^{loc}L_{}^{2}`$ section, where $`𝐃`$ is any differential d–operator of order less than $`s.`$ The subset of elements of $`{}_{}{}^{loc}H_{s}^{}(𝐕,𝐄)`$ with compact support is written $`{}_{}{}^{comp}H_{s}^{}(𝐕,𝐄).`$ The space $`{}_{}{}^{loc}H_{s}^{}(𝐕,𝐄)`$ is defined to be the dual space of $`{}_{}{}^{comp}H_{s}^{}(𝐕,𝐄)`$ and the space $`{}_{}{}^{comp}H_{s}^{}(𝐕,𝐄)`$ is defined to be the dual space of $`{}_{}{}^{loc}H_{s}^{}(𝐕,𝐄).`$
###### Definition 4.3
The Sobolev canonical d–space $`H_s`$ is an Hilbert space provided with the norm
$$<\widehat{}^s\underset{¯}{𝐮},\underset{¯}{𝐮}>^{1/2}$$
defined by the Laplace operator $`\widehat{}^s\widehat{𝐃}_\alpha \widehat{𝐃}^\alpha `$ of the canonical d–connection structure (A.2). Every d–operator $`\widehat{p}`$ of order less than $`r`$ can be extended to a continuous morphism $`H_sH_{sr}.`$
We can generalize the last two Definitions for N–anholonomic gerbes:
###### Definition 4.4
A d–operator $`𝐃`$ of degree $`r`$ on N–anholonomic gerbe $`𝐂`$ provided with d–metric and canonical d–connection structures is defined by a family of operators $`𝐃_e`$ of degree $`r`$ defined on an object $`𝐞`$ of the category $`𝐂(𝐔)`$ when for each morphism $`\phi :𝐞f`$ one holds $`𝐃_f\phi ^\mathrm{\#}=\phi ^\mathrm{\#}𝐃_e.`$
In this Definition the map $`\phi ^\mathrm{\#}`$ transforms a section $`z`$ to $`\phi (z)`$ and it is supposed that $`𝐃_e`$ is invariant under N–adapted authomorphisms of $`𝐞.`$ It is also assumed to be a continuous operator as a map
$$𝐃_e:^{comp}H_s(𝐔,𝐞)^{loc}H_{sr}(𝐔,𝐞).$$
For a global distributional section $`z`$ as an element of $`H_s(𝐙_{q_{\alpha \beta }}),`$ we can write $`q_{\widehat{\alpha }\widehat{\beta }}\left(𝐃_{e_{\widehat{\beta }}^{\widehat{\alpha }}}(e_{\widehat{\beta }}^{\widehat{\alpha }})\right)=\left(𝐃_{e_{\widehat{\alpha }}^{\widehat{\beta }}}(e_{\widehat{\alpha }}^{\widehat{\beta }})\right).`$ This proves
###### Proposition 4.1
Any d–operator $`𝐃`$ of degree $`r`$ defined on an N–anholonomic gerbe $`𝐂`$ provided with d–metric and canonical d–connection structures induces two maps
$$𝐃_{𝐙\left(q_{\widehat{\alpha }\widehat{\beta }}\right)}:H_s\left(𝐙\left(q_{\widehat{\alpha }\widehat{\beta }}\right)\right)H_{sr}\left(𝐙\left(q_{\widehat{\alpha }\widehat{\beta }}\right)\right)\text{ and }𝐃_𝐙:H_s\left(𝐙\right)H_{sr}\left(𝐙\right).$$
In this paper, we shall consider only pdd–operators preserving $`C^{\mathrm{}}`$ sections.
### 4.2 The symbols of nonholonomic operators
Let us consider a pdd–operator $`\widehat{p}_{\widehat{\alpha }}(f)=p_{\widehat{\alpha }}(𝐯,𝐮)\widehat{f}(𝐮)e^{i<𝐯,𝐮>}\delta 𝐮`$ defined for a restriction of $`\widehat{p}`$ to $`𝐔_{\widehat{\alpha }}`$ from a covering $`\left(𝐔_{\widehat{\alpha }}\right)_{\widehat{\alpha }I}`$ of an open $`𝐔𝐕.`$
###### Definition 4.5
The operator $`\widehat{p}_{\widehat{\alpha }}`$ is of degree $`r`$ with the symbol $`\sigma (𝐩)`$ if there exist the limit $`\sigma (𝐩_{𝐔_{\widehat{\alpha }}})=lim_\lambda \mathrm{}\left(p_{\widehat{\alpha }}(𝐯,\lambda 𝐮)/\lambda ^r\right).`$
This definition can be extended for a sphere bundle $`S𝐕`$ of the cotangent bundle $`T^{}𝐕`$ of $`𝐕`$ and for $`\pi ^{}𝐄`$ being the pull–back of the vector bundle $`𝐄`$ on $`𝐕`$ to $`T^{}𝐕.`$ The symbols defined by the matrix $`p_{\alpha \beta }`$ define a map $`\sigma :\pi ^{}𝐄\pi ^{}𝐄.`$ This map also induces a map $`\sigma _S:\pi _S^{}𝐄\pi _S^{}𝐄`$ if we consider the projection $`\pi _S:S𝐕𝐕.`$
Now we analyze the symbols of operators on a N–anholonomic vector gerbe $`𝐂_{NQ}`$ defined on $`𝐕`$ endowed with the operator $`D`$ of degree $`r.`$ For each object $`𝐞`$ of $`𝐂(𝐔),`$ it is possible to pull back the bundle $`𝐞`$ by the projection map $`\pi _{SU}:S𝐔𝐔`$ to a bundle $`\pi _{S𝐔}^{}e`$ over $`S𝐔.`$ This nonholonomic bundle is the restriction of the co–sphere bundle defined by a fixed d–metric on $`T^{}𝐕.`$ We can define a category consisting from the family $`𝐂_S(𝐔)`$ with elements $`\pi _{SU}^{}e`$ and baps of such elements induced by maps between elements of $`𝐂(𝐔).`$ In result, we can consider that the distinguished by N–connection map $`𝐔C_S(𝐔)`$ is a N–anholonomic gerbe with the same band as for $`𝐂.`$ For an object $`e,`$ it is possible to define the symbol $`\sigma _{D_e}:\pi _{S𝐔}^{}e\pi _{S𝐔}^{}e.`$
###### Proposition 4.2
For any sequence $`f_k`$ of elements of $`H_s\left(𝐙\left(q_{\widehat{\alpha }\widehat{\beta }}\right)\right)`$ and a constant $`f_{[0]}`$ such that $`f_k_s<f_{[0]},`$ there is a subsequence $`f_k^{}`$ converging in $`H_s^{}`$ for any $`s>s^{}.`$
The proof follows from the so–called Relich Lemma (Proposition 5) in Ref. : we have only to consider it both for the so–called h– and v–subspaces. For simplicity, in this subsection, we outline only some basic properties of d–operators for N–anholonomic gerbes which are distinguished by the N–connection structure: First, the space $`Op(𝐂)`$ of continuous linear N–adapted maps of $`H_s\left(𝐙\left(q_{\widehat{\alpha }\widehat{\beta }}\right)\right)`$ is a Banach space. Secondly, the last Preposition states the possibility to define $`O^r`$ the completion of the pseudo–differential operators in $`OP(𝐂)`$ of order $`r`$ and to extend the symbol $`\sigma `$ to this completion. Finally, the kernel of such an extension of the symbol to $`O^r`$ contains only compact operators.
## 5 $`K`$–Theory and the N–Adapted Index
This section is devoted to $`K`$–theory groups $`K_0`$ and $`K_1`$ associated to symbols of d–operators on N–anholonomic gerbes as elements of $`K_0(T^{}𝐕).`$
### 5.1 $`K`$–theory groups $`K_0`$ and $`K_1`$ and N–anholonomic spaces
We show how some basic results from $`K`$–theory (see, for instance, Ref. ) can be applied for nonholonomic manifolds.
#### 5.1.1 Basic definitions
Let us denote by $`A_n`$ the vector space of $`n\times n`$ complex matrices, consider natural injections $`A_nA_n^{}`$ for $`nn^{}`$ and denote by $`A_{\mathrm{}}`$ the inductive limit of the vector space $`A_n,n.`$ For a ring $`B`$ and two idempotents $`a^{}`$ and $`b^{}`$ of $`B_{\mathrm{}}=BA_{\mathrm{}},`$ one says that $`ab`$ if and only if there exists elements $`a^{},b^{}A_{\mathrm{}}`$ such that $`a=a^{}b^{}`$ and $`b=b^{}a^{}.`$ Let us denote by $`[a]`$ the class of $`a`$ and by $`Idem(B_{\mathrm{}})`$ the set of equivalence classes. Representing, respectively, $`[a]`$ and $`[b]`$ by elements of $`B`$ $`A_n`$ and $`B`$ $`A_n^{},`$ we can define an idempotent of $`B`$ $`A_{n+n^{}}`$ represented by $`[a+b]=\left(\begin{array}{cc}a& 0\\ 0& b\end{array}\right).`$ The semi–group $`Idem(B_{\mathrm{}})`$ provided with the operation $`[a]+[b]=[a+b]`$ is denoted by $`K_0(B).`$
One can extend the construction for a compact N–manifold $`𝐕`$ and a the set of complex valued functions $`(𝐕)`$ on $`𝐕.`$ For compact manifolds, it is possible to consider a trivial bundle isomorphic to $`𝐕\times ^r`$ and identify a vector bundle over $`𝐕`$ to an idempotent of $`(𝐔)A_r,`$ for $`𝐔𝐕,`$ which is also an idempotent of $`(𝐔)_{\mathrm{}}.`$ This allows us to identify $`K_0(𝐕)`$ to $`K_0((𝐔)).`$ Such a group for N–anholonomc manifold is a distinguished one, i.e. d–groups, into two different components, respectively for the h–subspace and the v–subspaces of $`𝐕.`$
Now we define the $`K_1`$ group: Let $`Al_r(B)`$ is the group of invertible elements contained in the matrix group $`A_r(B)`$. For $`r^{}r,`$ there is a canonical inclusion map $`Al_r^{}(B)Al_r(B).`$ The group $`Al_{\mathrm{}}(B)`$ denotes the inductive limit of the groups $`Al_r(B)`$ and $`Al_{\mathrm{}}(B)_{con}`$ is the respective connected component. In result, the group $`K_1(B)`$ is the quotient $`Al_{\mathrm{}}(B)/Al_{\mathrm{}}(B)_{con}.`$ For a compact N–anholonomic manifold $`𝐕,`$ one defines $`K_1(𝐕)`$ by $`K_1((𝐔)).`$
#### 5.1.2 N–anholonomic elliptic operators and indices
We consider a N–anholonomic gerbe $`𝐂`$ on manifold $`𝐕`$ and an elliptic operator $`D`$ of degree $`r`$ on $`𝐂,`$ inducing a distinguished morphism
$$D:L^2\left(𝐙\left(q_{\widehat{\alpha }\widehat{\beta }}\right)\right)H_{2r}\left(𝐙\left(q_{\widehat{\alpha }\widehat{\beta }}\right)\right).$$
Let $`(1)^r^{}`$ be an operator of degree $`r,`$ for instance, we can take $``$ to be the Laplace operator $`\widehat{}^s\widehat{𝐃}_\alpha \widehat{𝐃}^\alpha `$ of the canonical d–connection structure (A.2). It is possible to define the symbol $`\sigma (D)=(1)^r^{}D`$ which is an operator with image in $`𝐙\left(q_{\widehat{\alpha }\widehat{\beta }}\right)`$ being an invertible morphism.
For an exact sequence
$$0B_1B_2B_30$$
of $`C^{}`$–algebras, one has the following exact sequence in $`K`$–theory
$$K_1\left(B_1\right)K_1\left(B_2\right)K_1\left(B_3\right)K_0\left(B_1\right)K_0\left(B_2\right)K_0\left(B_3\right).$$
Let us consider $`O(),`$ the space of continuous operators on an Hilbert space $`,`$ and denote by $`𝒦`$ the subspace of compact continuous operators. There is the following exact sequence
$$0𝒦O()O()/𝒦=\mathrm{𝐂𝐚}0$$
when $`K_0\left(𝒦\right)=.`$ Now, it is possible to introduce the class $`\left[\sigma (D^{})\right],`$ for $`D^{}=(1)^r^{}D,`$ of $`K_1(\mathrm{𝐂𝐚}).`$
###### Definition 5.1
The image of $`\left[\sigma (D^{})\right]`$ in $`K_0\left(𝒦\right)`$ depends only on the symbol of operator $`D`$ and define the index of this operator.
We consider finite covering families $`𝐔_{\widehat{\alpha }}`$ of $`𝐕`$ when $`𝐂(𝐔_{\widehat{\alpha }})`$ are trivial bundles. One holds
###### Proposition 5.1
One exists a class $`\left[\sigma _D\right]`$ in $`K_1(T^{}𝐕)`$ associated to the symbol of an elliptic operator $`D`$ of degree $`r`$ on N–anholonomic gerbe $`𝐂`$ on $`𝐕.`$
Proof. A similar result is proven in for the Riemannian gerbes. We do not repeat those constructions in distinguished form for h- and v–components but note two important differences: In the N–anholonomic case there are N–connections, d–metrics and d–connections. In result, one can follow two ways: to define the class for the canonical d–connection and/or to derive the class from the N–connection structure and related curvature of N–connection.
We denote by $`X^{}𝐕`$ (with the fibers isomorphic to the unit ball defined by the N–connection) the compactification of $`T^{}𝐕.`$ The sphere N–anholonomic bundle $`S^{}𝐕`$ is identified to $`X^{}𝐕/T^{}𝐕.`$ In result, one can define the exact sequence
$$0𝐂(T{}_{}{}^{}𝐕)𝐂(X{}_{}{}^{}𝐕)𝐂(S{}_{}{}^{}𝐕)0$$
resulting to the following exact sequence
$`K_1\left[𝐂(S{}_{}{}^{}𝐕)A_r\right]`$ $``$ $`K_0\left[𝐂(T{}_{}{}^{}𝐕)A_r\right]`$
$``$ $`K_0\left[𝐂(X{}_{}{}^{}𝐕)A_r\right]K_1\left[𝐂(S{}_{}{}^{}𝐕)A_r\right]0.`$
The last sequence allows us to consider the boundary operator $`\delta (\left[\sigma _D^{}\right])`$ as an element of $`K_0\left[𝐂(T{}_{}{}^{}𝐕)A_r\right]`$ being isomorphic to $`K_0\left[T{}_{}{}^{}𝐕\right]`$ where
$$\left[\sigma _D^{}\right]K_1\left[𝐂(S{}_{}{}^{}𝐕)A_r\right]K_1\left[𝐂(S{}_{}{}^{}𝐕)\right]$$
for any N–anholonomic component of the sequence (5.1.2). This concludes that the index of a d–operator $`D`$ depends only on the class of boundary operator $`\delta (\left[\sigma _D^{}\right]).`$ For N–anholonomic configurations, such constructions are possible both the canonical d–connection and if it is not defined by a d–metric, one can re–define the constructions just for the N–connection and related metric compatible and N–adapted linear connection and resulting curvatures.
It should be noted that the class $`\left[\sigma _D\right]`$ is not unique. For N–anholonomic spaces, we can define such classes, for instance, by using the canonical d–connection or following d–metrics and d–connections derived from the N–connection structure.
### 5.2 The index formulas and applications
The results stated in previous subsections allow us to deduce Atiyah–Singer type theorems for N–anholonomic gerbes (in general form, for any their explicit realizations like Lagrange, or Finsler, gerbes and Riemann–Cartan gerbes provided with N–connection structure). Such theorems my have a number of applications in modern noncommutative geometry and physics. We shall consider the topic related to Dirac d–operators and N–anholonomic gerbes.
#### 5.2.1 The index formulas for d–operators and gerbes
The Chern character of the cotangent bundle $`T^{}M`$ induces a well known isomorphism $`K_0(T^{}M)^{ev}H_c(M,)`$ when for elements $`u^{}T^{}M`$ and $`t`$ one has the map $`u^{}ttch(u^{}).`$ The constructions may be generalized for N–anholonomic gerbes, see (23), with additional possibilities related to the N–connection and d–connection structures.
We denote by $`d^{}Vect(Ind)`$ the subspace of $`K_0(𝐕^{}),`$ with N–anholonomic manifold $`𝐕^{}`$ constructed to have local coordinates $`{}_{}{}^{}u_{}^{\alpha }=(x^i,^{}y_a),`$ with $`{}_{}{}^{}y_{a}^{}`$ being dual to $`y^a,`$ where $`u^\alpha =(x^i,y^a)`$ are local coordinates on $`𝐕.`$ We consider the symbol $`\sigma _p`$ of a d–operator $`\widehat{p}`$ on corresponding to $`𝐕^{}`$ N–anholonomic gerbe. The mentioned subspace is also a subspace of $`{}_{}{}^{ev}H_{c}^{}(𝐕,).`$ In result, the map $`d^{}Vect(\sigma _p)`$ given by $`ch(\left[\sigma _p^{}\right])ind(\widehat{p})`$ can be extended to a linear map $`{}_{}{}^{ev}H_{c}^{}(𝐕,).`$
One can be performed similar constructions starting from the N–anholonomic space $`T^{}𝐕`$ and using the distinguished isomorphism
$$K_0(T^{}𝐕)^{ev}H_c(𝐕,).$$
In this case, we introduce $`dVect(Ind)`$ as the subspace of $`K_0(T^{}𝐕)`$ generated by $`\sigma _p`$ related to $`T^{}𝐕.`$ Here we also note that the symbol operator $`\sigma _p`$ and related maps can be introduced for d–metric and canonical d–connection structures or for a ”pure” N–connection structure. So, there are two classes of symbols $`\sigma _p`$ (both in the case related to the constructions with $`𝐕^{}`$ and to the case for constructions with $`T^{}𝐕),`$ i. e. four variants of relations Chern character – index of d–operator and corresponding extensions to linear maps.
The above presented considerations consist in a proof (of four Atiyah–Singer type theorems):
###### Theorem 5.1
The Poincare duality of N–anholonomic gerbes implies the existence of classes $`t_d^{}(𝐕),t_V^{}(𝐕)`$ and $`t_d(𝐕),t_V(𝐕)`$ for which, respectively,
$`Ind_d^{}(\widehat{p})`$ $`=`$ $`{\displaystyle _𝐕^{}}ch\left(\left[\sigma _p^{}\right]\right)t_d^{}(𝐕),ch\left(\left[\sigma _p^{}\right]\right)\text{ related to d–metric},`$
$`Ind_N^{}(\widehat{p})`$ $`=`$ $`{\displaystyle _𝐕^{}}ch\left(\left[\sigma _p^{}\right]\right)t_N^{}(𝐕),ch\left(\left[\sigma _p^{}\right]\right)\text{ related to N–connection}`$
and
$`Ind_d(\widehat{p})`$ $`=`$ $`{\displaystyle _{T^{}𝐕}}ch\left(\left[\sigma _p^{}\right]\right)t_d(𝐕),ch\left(\left[\sigma _p^{}\right]\right)\text{ related to d–metric},`$
$`Ind_N(\widehat{p})`$ $`=`$ $`{\displaystyle _{T^{}𝐕}}ch\left(\left[\sigma _p^{}\right]\right)t_N(𝐕),ch\left(\left[\sigma _p^{}\right]\right)\text{ related to N–connection}.`$
The index formulas from this Theorem present topological characteristics for Lagrange (in particular, Finsler) spaces and gerbes, see (10) and (12), of nonholonomic Riemann–Cartan spaces, see ansatz (20), considered for constructing exact solutions in modern gravity .
#### 5.2.2 N–anholonomic spinors and the Dirac operator
The theory and methods developed in this paper have a number of motivations following from applications to the theory of nonholonomic Clifford structures and Dirac operators on N–anholonomic manifolds . In Appendix 6, there are given the necessary results on N–anholonomic spinor structurs and spin d–connections.
##### The Dirac d–operator: <br>
We consider a vector bundle $`𝐄`$ on an N–anholonomic manifold $`𝐕`$ (with two compatible N–connections defined as h– and v–splittings of $`T𝐄`$ and $`T𝐕`$)). A d–connection
$$𝒟:\mathrm{\Gamma }^{\mathrm{}}(𝐄)\mathrm{\Gamma }^{\mathrm{}}(𝐄)\mathrm{\Omega }^1(𝐕)$$
preserves by parallelism splitting of the tangent total and base spaces and satisfy the Leibniz condition
$$𝒟(f\sigma )=f(𝒟\sigma )+\delta f\sigma $$
for any $`fC^{\mathrm{}}(𝐕),`$ and $`\sigma \mathrm{\Gamma }^{\mathrm{}}(𝐄)`$ and $`\delta `$ defining an N–adapted exterior calculus by using N–elongated operators (3) and (4) which emphasize d–forms instead of usual forms on $`𝐕,`$ with the coefficients taking values in $`𝐄.`$
The metricity and Leibniz conditions for $`𝒟`$ are written respectively
$$𝐠(𝒟𝐗,𝐘)+𝐠(𝐗,𝒟𝐘)=\delta [𝐠(𝐗,𝐘)],$$
(25)
for any $`𝐗,𝐘\chi (𝐕),`$ and
$$𝒟(\sigma \beta )𝒟(\sigma )\beta +\sigma 𝒟(\beta ),$$
(26)
for any $`\sigma ,\beta \mathrm{\Gamma }^{\mathrm{}}(𝐄).`$
For local computations, we may define the corresponding coefficients of the geometric d–objects and write
$$𝒟\sigma _{\stackrel{´}{\beta }}𝚪_{\stackrel{´}{\beta }\mu }^{\stackrel{´}{\alpha }}\sigma _{\stackrel{´}{\alpha }}\delta u^\mu =𝚪_{\stackrel{´}{\beta }i}^{\stackrel{´}{\alpha }}\sigma _{\stackrel{´}{\alpha }}dx^i+𝚪_{\stackrel{´}{\beta }a}^{\stackrel{´}{\alpha }}\sigma _{\stackrel{´}{\alpha }}\delta y^a,$$
where fiber ”acute” indices, in their turn, may split $`\stackrel{´}{\alpha }(\stackrel{´}{ı},\stackrel{´}{a})`$ if any N–connection structure is defined on $`T𝐄.`$ For some particular constructions of particular interest, we can take $`𝐄=T^{}𝐕`$ and/or any Clifford d–algebra $`𝐄=\text{IC}l(𝐕)`$ with a corresponding treating of ”acute” indices to of d–tensor and/or d–spinor type as well when the d–operator $`𝒟`$ transforms into respective d–connection $`𝐃`$ and spin d–connections $`\widehat{}^𝐒`$ (A.18), $`\widehat{}^{SL}`$ (A.19)…. All such, adapted to the N–connections, computations are similar for both N–anholonomic (co) vector and spinor bundles.
The respective actions of the Clifford d–algebra and Clifford–Lagrange algebra can be transformed into maps $`\mathrm{\Gamma }^{\mathrm{}}(\mathrm{𝐒𝐩})\mathrm{\Gamma }^(\text{IC}l(𝐕))`$ to $`\mathrm{\Gamma }^{\mathrm{}}(\mathrm{𝐒𝐩})`$ by considering maps of type (A.8) and (A.14)
$$\widehat{𝐜}(\stackrel{˘}{\psi }𝐚)𝐜(𝐚)\stackrel{˘}{\psi }\text{ and }\widehat{c}(\psi a)c(a)\psi .$$
###### Definition 5.2
The Dirac d–operator (Dirac–Lagrange operator) on a spin N–anholonomic manifold $`(𝐕,\mathrm{𝐒𝐩},J)`$ where $`J:`$ $`\mathrm{𝐒𝐩}\mathrm{𝐒𝐩}`$ is the antilinear bijection, is defined
ID $``$ $`i(\widehat{𝐜}^𝐒)`$
$`=`$ $`(\text{I}\text{D}=i(\widehat{c}^𝐒),^{}\text{I}\text{D}=i(^{}\widehat{c}^{}^𝐒))`$
Such N–adapted Dirac d–operators are called canonical and denoted $`\widehat{\text{I}\text{D}}=(\widehat{\text{I}\text{D}},^{}\widehat{\text{I}\text{D}})`$ if they are defined for the canonical d–connection (A.3) and respective spin d–connection (A.18) ((A.19)).
Now we can formulate the (see Proof of Theorem 6.1 )
###### Theorem 5.2
Let $`(𝐕,\mathrm{𝐒𝐩},J)`$ be a spin N–anholonomic manifold ( spin Lagrange space). There is the canonical Dirac d–operator (Dirac–Lagrange operator) defined by the almost Hermitian spin d–operator
$$\widehat{}^𝐒:\mathrm{\Gamma }^{\mathrm{}}(\mathrm{𝐒𝐩})\mathrm{\Gamma }^{\mathrm{}}(\mathrm{𝐒𝐩})\mathrm{\Omega }^1(𝐕)$$
commuting with $`J`$ and satisfying the conditions
$$(\widehat{}^𝐒\stackrel{˘}{\psi }|\stackrel{˘}{\varphi })+(\stackrel{˘}{\psi }|\widehat{}^𝐒\stackrel{˘}{\varphi })=\delta (\stackrel{˘}{\psi }|\stackrel{˘}{\varphi })$$
and
$$\widehat{}^𝐒(𝐜(𝐚)\stackrel{˘}{\psi })=𝐜(\widehat{𝐃}𝐚)\stackrel{˘}{\psi }+𝐜(𝐚)\widehat{}^𝐒\stackrel{˘}{\psi }$$
for $`𝐚\text{IC}l(𝐕)`$ and $`\stackrel{˘}{\psi }\mathrm{\Gamma }^{\mathrm{}}(\mathrm{𝐒𝐩})`$ determined by the metricity (25) and Leibnitz (26) conditions.
##### The Clifford N–gerbe and the Dirac operator: <br>
We consider the Clifford N–gerbe $`Cl_N(𝐕)`$ on a N–anholonomic manifold $`𝐕,`$ see section 3, provided with d–connection structure (20). For an opening covering $`\left(𝐔_{\widehat{\alpha }}\right)_{\widehat{\alpha }I}`$ of $`𝐕,`$ the canonical d–connection (A.2) is extended on all family of $`𝐔_{\widehat{\alpha }}`$ as a corresponding family of N–adapted $`so(n+m)`$ forms $`\widehat{\mathrm{\Gamma }}_{\widehat{\alpha }}`$ on $`𝐔_{\widehat{\alpha }}`$ satisfying
$$\widehat{\mathrm{\Gamma }}_{\widehat{\alpha }}=ad(q_{\widehat{\beta }\widehat{\alpha }})^1\widehat{\mathrm{\Gamma }}_{\widehat{\beta }}+(q_{\widehat{\beta }\widehat{\alpha }})^1\delta \left(q_{\widehat{\beta }\widehat{\alpha }}\right)$$
where $`\delta \left(q_{\widehat{\beta }\widehat{\alpha }}\right)`$ is computed by using the N–elongated partial derivatives (3) and (4). This d–connection induces the covariant derivative $`\delta +\widehat{\mathrm{\Gamma }}_{\widehat{\alpha }}.`$ Fixing an orthogonal basis $`𝐞_{\widehat{\alpha }}`$, to which (3) are transformed in general, on $`T𝐔_{\widehat{\alpha }},`$ we can write
$$\widehat{\mathrm{\Gamma }}_{\widehat{\alpha }}=\left(\widehat{\mathrm{\Gamma }}_{\widehat{i}}=\underset{\widehat{k}=1}{\overset{\widehat{k}=n}{}}\widehat{\mathrm{\Gamma }}_{\widehat{i}\widehat{k}}e_{\widehat{k}},\widehat{\mathrm{\Gamma }}_{\widehat{a}}=\underset{\widehat{b}=n+1}{\overset{\widehat{b}=n+m}{}}\widehat{\mathrm{\Gamma }}_{\widehat{a}\widehat{b}}e_{\widehat{b}}\right).$$
In local form, the d–spinor covariant derivative was investigated in Refs. . We can extend it to a covering family $`\left(𝐔_{\widehat{\alpha }}\right)_{\widehat{\alpha }I}`$ of $`𝐕`$ by introducing the d–object $`\varrho _{\widehat{\beta }\widehat{\alpha }}=\frac{1}{4}`$ $`\widehat{\mathrm{\Gamma }}_{\widehat{\beta }\widehat{\alpha }}=\varrho _{\widehat{\alpha }\widehat{\beta }},`$ see formula (A.18) in Appendix.
Let $`𝐞_U`$ be an object of $`Cl_N(𝐔)`$. For a trivialization $`\left(𝐔_{\widehat{\alpha }}\right)_{\widehat{\alpha }I},`$ we can generalize for N–anholonomic gerbes the Definition 5.2 and write
$$\text{I}\text{D}_{𝐞_U}\underset{\widehat{\alpha }=1}{\overset{\widehat{\alpha }=n+m}{}}𝐞_{\widehat{\alpha }}\text{I}\text{D}=(\underset{\widehat{i}=1}{\overset{\widehat{i}=n}{}}e_{\widehat{i}}\text{I}\text{D},\underset{\widehat{a}=n+1}{\overset{\widehat{a}=n+m}{}}e_{\widehat{a}}^{}\text{I}\text{D})$$
where ID is defined locally by (5.2). On any such object one holds a distinguished variant of the Lichnerowicz–Weitzenbock formula (defined by the d–connection and/or N–connection structure)
$$D^2=\text{I}\text{D}^{}\text{I}\text{D}+\frac{1}{4}\stackrel{}{𝐑}$$
where $`\text{I}\text{D}^{}\text{I}\text{D}`$ is the Laplacian and $`\stackrel{}{R}`$ is the scalar curvature (A.6) of the corresponding d–connection (the definition of curvature of a general nonholonomic manifold is not a trivial task but for N–anholonomic manifolds this follows from a usual N–adapted tensor and differential calculus (see the suppersymmetric variant in ) like that presented in Appendix The Canonical d–Connection.
Such global d–spinors are canonically d–harmonic if $`\text{I}\text{D}_{𝐞_{\widehat{\alpha }}}(\stackrel{}{𝐑}_{\widehat{\alpha }})=0`$ for each $`\stackrel{}{𝐑}_{\widehat{\alpha }}.`$ If a d–metric (20) is not provided, we can define a formal canonical d–connection $`{}_{}{}^{N}\widehat{𝚪}_{\beta \gamma }^{\alpha }`$ computed by formulas (A.3) with $`g_{ij}`$ and $`h_{ab}`$ taken for Euclidean spaces (this metric compatible canonical d–connection is defined only by the N–connection coefficients). Introducing $`{}_{}{}^{N}\widehat{𝚪}_{\beta \gamma }^{\alpha }`$ into (A.4) and (A.6), one computes respectively the curvature $`{}_{}{}^{N}_{\beta }^{\alpha }`$ and scalar curvature $`{}_{}{}^{N}\stackrel{}{𝐑}.`$
For the Riemannian gerbes derived for compact Riemannian manifolds $`V`$ with strictly positive curvature, it is known the result that the topological class $`\tau (V)`$ associated to the index formula for operators on the $`Cl(V)`$ gerbe is zero . One holds a similar result for N–anholonomic manifolds and gerbes $`Cl(𝐕)`$ but in terms of d–metrics and canonical d–connections defining $`\widehat{\tau }(𝐕)=0.`$ We can compute the class $`{}_{}{}^{N}\widehat{\tau }(𝐕)=0`$ even a d–metric is not given but its $`{}_{}{}^{N}\stackrel{}{𝐑}`$ is strictly positive and the N–anholonomic manifold is compact.
## The Canonical d–Connection
A d–connection splits into h– and v–covariant derivatives, $`𝐃=D+^{}D,`$ where $`D_k=(L_{jk}^i,L_{bk}^a)`$ and $`{}_{}{}^{}D_{c}^{}=(C_{jk}^i,C_{bc}^a)`$ are correspondingly introduced as h- and v–parametrizations of (6),
$$L_{jk}^i=\left(𝐃_ke_j\right)e^i,L_{bk}^a=\left(𝐃_ke_b\right)e^a,C_{jc}^i=\left(𝐃_ce_j\right)e^i,C_{bc}^a=\left(𝐃_ce_b\right)e^a.$$
The components $`𝚪_{\alpha \beta }^\gamma =(L_{jk}^i,L_{bk}^a,C_{jc}^i,C_{bc}^a)`$ completely define a d–connection $`𝐃`$ on a N–anholonomic manifold $`𝐕.`$
The simplest way to perform a covariant calculus by applying d–connections is to use N–adapted differential forms like $`𝚪_\beta ^\alpha =𝚪_{\beta \gamma }^\alpha 𝐞^\gamma `$ with the coefficients defined with respect to (4) and (3).
###### Theorem 5.3
The torsion $`𝒯^\alpha `$ (7) of a d–connection has the irreducible h- v– components (d–torsions) with N–adapted coefficients
$`T_{jk}^i`$ $`=`$ $`L_{jk}^iL_{kj}^i,T_{ja}^i=T_{aj}^i=C_{ja}^i,T_{ji}^a=\mathrm{\Omega }_{ji}^a,`$
$`T_{bi}^a`$ $`=`$ $`T_{ib}^a={\displaystyle \frac{N_i^a}{y^b}}L_{bi}^a,T_{bc}^a=C_{bc}^aC_{cb}^a.`$ (A.1)
Proof. By a straightforward calculation, we can verify the formulas.$`\mathrm{}`$
The Levi–Civita linear connection $`=\{^{}𝚪_{\beta \gamma }^\alpha \},`$ with vanishing both torsion and nonmetricity, is not adapted to the global splitting (1).
One holds:
###### Proposition 5.2
There is a preferred, canonical d–connection structure,$`\widehat{𝐃},`$ on N–aholonomic manifold $`𝐕`$ constructed only from the metric and N–connection coefficients $`[g_{ij},h_{ab},N_i^a]`$ and satisfying the conditions $`\widehat{𝐃}𝐠=0`$ and $`\widehat{T}_{jk}^i=0`$ and $`\widehat{T}_{bc}^a=0.`$
Proof. By straightforward calculations with respect to the N–adapted bases (4) and (3), we can verify that the connection
$$\widehat{𝚪}_{\beta \gamma }^\alpha =^{}𝚪_{\beta \gamma }^\alpha +\widehat{𝐏}_{\beta \gamma }^\alpha $$
(A.2)
with the deformation d–tensor <sup>5</sup><sup>5</sup>5$`\widehat{𝐏}_{\beta \gamma }^\alpha `$ is a tensor field of type (1,2). As is well known, the sum of a linear connection and a tensor field of type (1,2) is a new linear connection.
$$\widehat{𝐏}_{\beta \gamma }^\alpha =(P_{jk}^i=0,P_{bk}^a=e_b(N_k^a),P_{jc}^i=\frac{1}{2}g^{ik}\mathrm{\Omega }_{kj}^ah_{ca},P_{bc}^a=0)$$
satisfies the conditions of this Proposition. It should be noted that, in general, the components $`\widehat{T}_{ja}^i,\widehat{T}_{ji}^a`$ and $`\widehat{T}_{bi}^a`$ are not zero. This is an anholonomic frame (or, equivalently, off–diagonal metric) effect.$`\mathrm{}`$
With respect to the N–adapted frames, the coefficients
$`\widehat{𝚪}_{\alpha \beta }^\gamma =(\widehat{L}_{jk}^i,\widehat{L}_{bk}^a,\widehat{C}_{jc}^i,\widehat{C}_{bc}^a)`$ are computed:
$`\widehat{L}_{jk}^i`$ $`=`$ $`{\displaystyle \frac{1}{2}}g^{ir}\left(e_kg_{jr}+e_jg_{kr}e_rg_{jk}\right),`$ (A.3)
$`\widehat{L}_{bk}^a`$ $`=`$ $`e_b(N_k^a)+{\displaystyle \frac{1}{2}}h^{ac}\left(e_kh_{bc}h_{dc}e_bN_k^dh_{db}e_cN_k^d\right),`$
$`\widehat{C}_{jc}^i`$ $`=`$ $`{\displaystyle \frac{1}{2}}g^{ik}e_cg_{jk},\widehat{C}_{bc}^a={\displaystyle \frac{1}{2}}h^{ad}\left(e_ch_{bd}+e_ch_{cd}e_dh_{bc}\right).`$
For the canonical d–connection, there are satisfied the conditions of vanishing of torsion on the h–subspace and v–subspace, i.e., $`\widehat{T}_{jk}^i=\widehat{T}_{bc}^a=0.`$
The curvature of a d–connection $`𝐃`$ on an N–anholonomic manifold is defined by the usual formula
$$𝐑(𝐗,𝐘)𝐙𝐃_X𝐃_Y𝐙𝐃_Y𝐃_X𝐙𝐃_{[X,X]}𝐙.$$
By straightforward calculations, we can prove:
###### Theorem 5.4
The curvature $`_\beta ^\alpha 𝐃𝚪_\beta ^\alpha =d𝚪_\beta ^\alpha 𝚪_\beta ^\gamma 𝚪_\gamma ^\alpha `$ of a d–connection has the irreducible h- v– components (d–curvatures) of $`𝐑_{\beta \gamma \delta }^\alpha `$,
$`R_{hjk}^i`$ $`=`$ $`e_kL_{hj}^ie_jL_{hk}^i+L_{hj}^mL_{mk}^iL_{hk}^mL_{mj}^iC_{ha}^i\mathrm{\Omega }_{kj}^a,`$
$`R_{bjk}^a`$ $`=`$ $`e_kL_{bj}^ae_jL_{bk}^a+L_{bj}^cL_{ck}^aL_{bk}^cL_{cj}^aC_{bc}^a\mathrm{\Omega }_{kj}^c,`$
$`R_{jka}^i`$ $`=`$ $`e_aL_{jk}^iD_kC_{ja}^i+C_{jb}^iT_{ka}^b,`$ (A.4)
$`R_{bka}^c`$ $`=`$ $`e_aL_{bk}^cD_kC_{ba}^c+C_{bd}^cT_{ka}^c,`$
$`R_{jbc}^i`$ $`=`$ $`e_cC_{jb}^ie_bC_{jc}^i+C_{jb}^hC_{hc}^iC_{jc}^hC_{hb}^i,`$
$`R_{bcd}^a`$ $`=`$ $`e_dC_{bc}^ae_cC_{bd}^a+C_{bc}^eC_{ed}^aC_{bd}^eC_{ec}^a.`$
Contracting respectively the components of (A.4), one proves
###### Corollary 5.1
The Ricci d–tensor $`𝐑_{\alpha \beta }𝐑_{\alpha \beta \tau }^\tau `$ has the irreducible h- v–components
$$R_{ij}R_{ijk}^k,R_{ia}R_{ika}^k,R_{ai}R_{aib}^b,R_{ab}R_{abc}^c,$$
(A.5)
for a N–holonomic manifold $`𝐕.`$
###### Corollary 5.2
The scalar curvature of a d–connection is
$$\stackrel{}{𝐑}𝐠^{\alpha \beta }𝐑_{\alpha \beta }=g^{ij}R_{ij}+h^{ab}R_{ab},$$
(A.6)
defined by the ”pure” h– and v–components of (A.5).
## 6 Nonholonomic Spinors and Spin Connections
We outline the necessary results on spinor structures and N–connections .
Let us consider a manifold $`M`$ of dimension $`n.`$ We define the the algebra of Dirac’s gamma matrices (in brief, h–gamma matrices defined by self–adjoints matrices $`A_k()`$ where $`k=2^{n/2}`$ is the dimension of the irreducible representation of the set gamma matrices, defining the Clifford structure $`\text{IC}l(M),`$ for even dimensions, or of $`\text{IC}l(M)^+`$ for odd dimensions) from the relation
$$\gamma ^{\widehat{ı}}\gamma ^{\widehat{ȷ}}+\gamma ^{\widehat{ȷ}}\gamma ^{\widehat{ı}}=2\delta ^{\widehat{ı}\widehat{ȷ}}\text{II}.$$
(A.7)
We can consider the action of $`dx^i\text{IC}l(M)`$ on a spinor $`\psi Sp`$ via representations
$$c(dx^{\widehat{ı}})\gamma ^{\widehat{ı}}\text{ and }c(dx^i)\psi \gamma ^i\psi e_{\widehat{ı}}^i\gamma ^{\widehat{ı}}\psi .$$
(A.8)
For any tangent bundle $`TM`$ and/or N–anholonomic manifold $`𝐕`$ possessing a local (in any point) or global fibered structure $`F`$ (being isomorphic to a real vector space of dimension $`m`$) and, in general, enabled with a N–connection structure, we can introduce similar definitions of the gamma matrices following algebraic relations and metric structures on fiber subspaces,
$$e^{\widehat{a}}e_{\underset{¯}{a}}^{\widehat{a}}(x,y)e^{\underset{¯}{a}}\text{ and }e^ae_{\underset{¯}{a}}^a(x,y)e^{\underset{¯}{a}},$$
(A.9)
where
$$g^{\underset{¯}{a}\underset{¯}{b}}(x,y)e_{\underset{¯}{a}}^{\widehat{a}}(x,y)e_{\underset{¯}{b}}^{\widehat{b}}(x,y)=\delta ^{\widehat{a}\widehat{b}}\text{ and }g^{\underset{¯}{a}\underset{¯}{b}}(x,y)e_{\underset{¯}{a}}^a(x,y)e_{\underset{¯}{b}}^b(x,y)=h^{ab}(x,y).$$
In a similar form, we define the algebra of Dirac’s matrices related to typical fibers (in brief, v–gamma matrices defined by self–adjoint matrices $`M_k^{}()`$ where $`k^{}=2^{m/2}`$ is the dimension of the irreducible representation of $`\text{IC}l(F)`$ for even dimensions, or of $`\text{IC}l(F)^+`$ for odd dimensions, of the typical fiber) from the relation
$$\gamma ^{\widehat{a}}\gamma ^{\widehat{b}}+\gamma ^{\widehat{b}}\gamma ^{\widehat{a}}=2\delta ^{\widehat{a}\widehat{b}}\text{II}.$$
(A.10)
The action of $`dy^a\text{IC}l(F)`$ on a spinor $`{}_{}{}^{}\psi ^{}Sp`$ is considered via representations
$${}_{}{}^{}c(dy^{\widehat{a}})\gamma ^{\widehat{a}}\text{ and }^{}c(dy^a)^{}\psi \gamma ^a{}_{}{}^{}\psi e_{\widehat{a}}^a\gamma ^{\widehat{a}}{}_{}{}^{}\psi .$$
(A.11)
A more general gamma matrix calculus with distinguished gamma matrices (in brief, d–gamma matrices) can be elaborated for N–anholonomic manifolds $`𝐕`$ provided with d–metric structure $`𝐠=[g,^{}g]`$ and for d–spinors $`\stackrel{˘}{\psi }(\psi ,^{}\psi )\mathrm{𝐒𝐩}(Sp,^{}Sp),`$ which are usual spinors but adapted locally to the N–connection structure, i. e. they are defined with respect to N–elongated bases (3) and (4). Firstly, we should write in a unified form, related to a d–metric (20), the formulas (A.9),
$$e^{\widehat{\alpha }}e_{\underset{¯}{a}}^{\widehat{\alpha }}(u)e^{\underset{¯}{\alpha }}\text{ and }e^\alpha e_{\underset{¯}{\alpha }}^\alpha (u)e^{\underset{¯}{\alpha }},$$
(A.12)
where
$$g^{\underset{¯}{\alpha }\underset{¯}{\beta }}(u)e_{\underset{¯}{\alpha }}^{\widehat{\alpha }}(u)e_{\underset{¯}{\beta }}^{\widehat{\beta }}(u)=\delta ^{\widehat{\alpha }\widehat{\beta }}\text{ and }g^{\underset{¯}{\alpha }\underset{¯}{\beta }}(u)e_{\underset{¯}{\alpha }}^\alpha (u)e_{\underset{¯}{\beta }}^\beta (u)=g^{\alpha \beta }(u).$$
The second step, is to consider d–gamma matrix relations (unifying (A.7) and (A.10))
$$\gamma ^{\widehat{\alpha }}\gamma ^{\widehat{\beta }}+\gamma ^{\widehat{\beta }}\gamma ^{\widehat{\alpha }}=2\delta ^{\widehat{\alpha }\widehat{\beta }}\text{II},$$
(A.13)
with the action of $`du^\alpha \text{IC}l(𝐕)`$ on a d–spinor $`\stackrel{˘}{\psi }\mathrm{𝐒𝐩}`$ resulting in distinguished irreducible representations (unifying (A.8) and (A.11))
$$𝐜(du^{\widehat{\alpha }})\gamma ^{\widehat{\alpha }}\text{ and }𝐜=(du^\alpha )\stackrel{˘}{\psi }\gamma ^\alpha \stackrel{˘}{\psi }e_{\widehat{\alpha }}^\alpha \gamma ^{\widehat{\alpha }}\stackrel{˘}{\psi }$$
(A.14)
which allows us to write
$$\gamma ^\alpha (u)\gamma ^\beta (u)+\gamma ^\beta (u)\gamma ^\alpha (u)=2g^{\alpha \beta }(u)\text{II}.$$
(A.15)
In the canonical representation we can write in irreducible form $`\stackrel{˘}{\gamma }\gamma ^{}\gamma `$ and $`\stackrel{˘}{\psi }\psi ^{}\psi ,`$ for instance, by using block type of h– and v–matrices, or, writing alternatively as couples of gamma and/or h– and v–spinor objects written in N–adapted form,
$$\gamma ^\alpha (\gamma ^i,\gamma ^a)\text{ and }\stackrel{˘}{\psi }(\psi ,^{}\psi ).$$
(A.16)
The decomposition (A.15) holds with respect to a N–adapted vielbein (3). We also note that for a spinor calculus, the indices of spinor objects should be treated as abstract spinorial ones possessing certain reducible, or irreducible, properties depending on the space dimension. For simplicity, we shall consider that spinors like $`\stackrel{˘}{\psi },\psi ,^{}\psi `$ and all type of gamma objects can be enabled with corresponding spinor indices running certain values which are different from the usual coordinate space indices. In a ”rough” but brief form we can use the same indices $`i,j,\mathrm{},a,b\mathrm{},\alpha ,\beta ,\mathrm{}`$ both for d–spinor and d–tensor objects.
The spin connection $`{}_{}{}^{S}`$ for the Riemannian manifolds is induced by the Levi–Civita connection $`{}_{}{}^{}\mathrm{\Gamma },`$
$${}_{}{}^{S}d\frac{1}{4}^{}\mathrm{\Gamma }_{jk}^i\gamma _i\gamma ^jdx^k.$$
(A.17)
On N–anholonomic spaces, it is possible to define spin connections which are N–adapted by replacing the Levi–Civita connection by any d–connection.
###### Definition 6.1
The canonical spin d–connection is defined by the canonical d–connection (A.2) as
$${}_{}{}^{𝐒}\widehat{}\delta \frac{1}{4}\widehat{𝚪}_{\beta \mu }^\alpha \gamma _\alpha \gamma ^\beta \delta u^\mu ,$$
(A.18)
where the absolute differential $`\delta `$ acts in N–adapted form resulting in 1–forms decomposed with respect to N–elongated differentials like $`\delta u^\mu =(dx^i,\delta y^a)`$ (4).
We note that the canonical spin d–connection $`{}_{}{}^{𝐒}\widehat{}`$ is metric compatible and contains nontrivial d–torsion coefficients induced by the N–anholonomy relations (see the formulas (A.1) proved for arbitrary d–connection). It is possible to introduce more general spin d–connections $`{}_{}{}^{𝐒}𝐃`$ by using the same formula (A.18) but for arbitrary metric compatible d–connection $`𝚪_{\beta \mu }^\alpha .`$
###### Proposition 6.1
On Lagrange spaces, there is a canonical spin d–connection (the canonical spin–Lagrange connection),
$${}_{}{}^{SL}\widehat{}\delta \frac{1}{4}^L𝚪_{\beta \mu }^\alpha \gamma _\alpha \gamma ^\beta \delta u^\mu ,$$
(A.19)
where $`\delta u^\mu =(dx^i,\delta y^k=dy^k+^LN_i^kdx^i).`$
We emphasize that even regular Lagrangians of classical mechanics without spin particles induce in a canonical (but nonholonomic) form certain classes of spin d–connections like (A.19).
For the spaces provided with generic off–diagonal metric structure (13) (in particular, for such Riemannian manifolds) resulting in equivalent N–anholonomic manifolds, it is possible to prove a result being similar to Proposition 6.1:
###### Remark 6.1
There is a canonical spin d–connection (A.18) induced by the off–diagonal metric coefficients with nontrivial $`N_i^a`$ and associated nonholonomic frames in gravity theories.
The N–connection structure also states a global h– and v–splitting of spin d–connection operators, for instance,
$${}_{}{}^{SL}\widehat{}\delta \frac{1}{4}^L\widehat{L}_{jk}^i\gamma _i\gamma ^jdx^k\frac{1}{4}^L\widehat{C}_{bc}^a\gamma _a\gamma ^b\delta y^c.$$
(A.20)
So, any spin d–connection is a d–operator with conventional splitting of action like $`^𝐒(^𝐒,{}_{}{}^{}_{}^{(𝐒)}),`$ or $`^{(SL)}(^{SL},{}_{}{}^{}_{}^{SL}).`$ For instance, for $`\widehat{}^{SL}(\widehat{}^{SL},{}_{}{}^{}\widehat{}_{}^{SL}),`$ the operators $`\widehat{}^{SL}`$ and $`{}_{}{}^{}\widehat{}_{}^{SL}`$ act respectively on a h–spinor $`\psi `$ as
$$\widehat{}^{SL}\psi dx^i\frac{\delta \psi }{dx^i}dx^k\frac{1}{4}^L\widehat{L}_{jk}^i\gamma _i\gamma ^j\psi $$
(A.21)
and
$${}_{}{}^{}\widehat{}_{}^{SL}\psi \delta y^a\frac{\psi }{dy^a}\delta y^c\frac{1}{4}^L\widehat{C}_{bc}^a\gamma _a\gamma ^b\psi $$
being defined by the canonical d–connection (A.2). |
warning/0507/math0507088.html | ar5iv | text | # Gradient bounds for minimizers of free discontinuity problems related to cohesive zone models in fracture mechanics
## 1. Introduction
The study of cohesive zone models in fracture mechanics in the one dimensional case (see, e.g., and ) leads to functionals of the form
$`{\displaystyle _0^l}F(|\dot{u}|)𝑑x+{\displaystyle \underset{S(u)}{}}G(|[u]|)uSBV(0,l),`$ (1)
where $`F:[0,+\mathrm{})[0,+\mathrm{})`$ is $`C^1`$, strictly convex, increasing, superlinear at infinity, and satisfies $`F(0)=F^{}(0)=0`$, and $`G:[0,+\mathrm{})[0,+\mathrm{})`$ is $`C^1`$, concave, and satisfies $`G(0)=0`$ and $`G^{}(0)>0`$. Here and in the rest of the paper $`SBV`$ is the space of special functions with bounded variation, for which we refer to , $`S(u)`$ denotes the jump set of $`u`$, and $`[u]`$ denotes the jump of $`u`$.
To prove the existence of a minimizer of (1) with appropriate boundary conditions we can consider the corresponding relaxed functional in $`L^1(0,l)`$, which for every $`uBV(0,l)`$ can be written as
$`{\displaystyle _0^l}\overline{F}(|\dot{u}|)𝑑x+{\displaystyle \underset{S(u)}{}}G(|[u]|)+G^{}(0)|u_c^{}|(0,l),`$ (2)
where $`\dot{u}`$ is the density of the absolutely continuous part of the distributional derivative $`u^{}`$ and $`u_c^{}`$ is its Cantor part. In (2) $`\overline{F}(\xi )=F(\xi )`$ for $`\xi e_M`$ and $`\overline{F}(\xi )=F(e_M)+F^{}(e_M)(\xi e_M)`$ if $`\xi >e_M`$, where $`e_M`$ is the unique constant such that $`F^{}(e_M)=G^{}(0)`$. It is possible to prove that the minimum problem for the relaxed functional (2) with appropriate boundary conditions has a solution. Moreover in it was proved, by using one dimensional arguments, that if $`G`$ is strictly concave, then every local minimizer $`u`$ of (2) satisfies
$$|\dot{u}|e_M\text{a.e. on }(0,l),|u_c^{}|(0,l)=0.$$
In particular this implies that $`\overline{F}(\dot{u})=F(\dot{u})`$ a.e. on $`(0,l)`$, so that $`u`$ is a local minimizer of (1). Moreover
$$F^{}(|\dot{u}|)G^{}(0).$$
This justifies the interpretation of $`G^{}(0)`$ as the ultimate stress for the problem (see, e.g., ).
In this note we study the same problem in dimension $`n1`$. We consider functionals of the form
$`{\displaystyle _\mathrm{\Omega }}F(|u|)𝑑x+{\displaystyle _{S(u)}}G(|[u]|)𝑑^{n1}uSBV(\mathrm{\Omega }),`$ (3)
where $`u`$ is the density of the absolutely continuous part of the distributional gradient $`Du`$, and $`F`$ and $`G`$ satisfy the same properties considered for (1).
Also in this case the functional is not lower semicontinuous, so in order to prove existence results we consider its relaxed functional in $`L^1(\mathrm{\Omega })`$ (see ), which is represented on $`BV(\mathrm{\Omega })`$ by
$`(u)={\displaystyle _\mathrm{\Omega }}\overline{F}(|u|)𝑑x+{\displaystyle _{S(u)}}G(|[u]|)𝑑^{n1}+G^{}(0)|D^cu|(\mathrm{\Omega }),`$ (4)
where $`\overline{F}`$ is defined as for (2) and $`D^cu`$ denote the Cantor part of $`Du`$. Under appropriate boundary conditions the minimum problems for (4) have a solution. A local minimizer of $``$ in $`\mathrm{\Omega }`$ is a function $`uBV(\mathrm{\Omega })`$, with $`(u)<+\mathrm{}`$, for which there exists $`\eta >0`$ such that $`(u)(v)`$ for every $`vBV(\mathrm{\Omega })`$ with $`\mathrm{supp}(vu)\mathrm{\Omega }`$ and $`vu_{BV(\mathrm{\Omega })}<\eta `$.
Also in this case it is reasonable to expect that any local minimizer $`u`$ satisfies
$`|u|e_M\text{a.e. on }\mathrm{\Omega },|D^cu|(\mathrm{\Omega })=0,`$ (5)
where $`e_M`$ is defined as for (2). In fracture mechanics the functionals (3) and (4) are used to study cohesive zone models in the antiplane case. In this context the first inequality in (5) says that the norm of the deformation gradient of the solution cannot exceed the constant $`e_M`$, which is interpreted as the yield strain of the problem. Since (5) implies $`F^{}(|u|)G^{}(0)`$ a.e. on $`\mathrm{\Omega }`$, the constant $`G^{}(0)`$ plays the role of the ultimate stress for the crack problem.
The aim of this note is to present a partial result in this direction. Namely, we prove that, if
$$\underset{t0+}{lim}\frac{G(t)G^{}(0)t}{t^2}<0$$
and $`u`$ is a local minimizer of (4) in $`\mathrm{\Omega }`$, then
$$|u|e_M$$
in every open subset of $`\mathrm{\Omega }`$ where $`u`$ is of class $`C^1`$. As a consequence we have that, if $`u`$ is a $`C^1`$ local minimizer for (4) in $`\mathrm{\Omega }`$, then it is also a local minimizer for (3).
## 2. Statement and proof of the result
Let $`\mathrm{\Omega }`$ be an open subset of $`^n`$, $`n1`$. We assume that the functions $`F`$ and $`G`$ satisfy the following properties:
* $`F`$ is $`C^1`$, strictly convex, increasing, and superlinear at infinity, and satisfies $`F(0)=F^{}(0)=0`$;
* $`G`$ is $`C^1`$, nonnegative, concave, and satisfies $`G(0)=0`$, $`G^{}(0)>0`$, and
$`\underset{t0+}{lim}{\displaystyle \frac{G(t)G^{}(0)t}{t^2}}<0.`$ (6)
The function $`\overline{F}`$ is defined as follows
$`\overline{F}(\xi )=\{\begin{array}{cc}F(\xi )\hfill & \text{if }\xi e_M,\hfill \\ F(e_M)+F^{}(e_M)(\xi e_M)\hfill & \text{if }\xi >e_M,\hfill \end{array}`$ (7)
where $`e_M`$ is the unique solution of the equation $`F^{}(e_M)=G^{}(0)`$.
###### Theorem 1.
Assume that $`F`$ and $`G`$ satisfy conditions (a) and (b) and let $`u`$ be a local minimizer of the functional $``$ defined by (4). Suppose that $`u`$ is of class $`C^1`$ on an open subset $`U`$ of $`\mathrm{\Omega }`$. Then $`|u|e_M`$ in $`U`$.
The result stated in Theorem 1 implies that, if $`u`$ is a local minimizer of (4) satisfying $`uC^1(\mathrm{\Omega }K)`$, with $`K`$ closed and $`^{n1}(K)<+\mathrm{}`$, then $`u`$ is also a local minimizer of (3). Indeed in this case $`D^cu=0`$, hence $`uSBV(\mathrm{\Omega })`$, and $`\overline{F}(|u|)=F(|u|)`$ a.e. in $`\mathrm{\Omega }`$ by Theorem 1.
###### Proof of Theorem 1.
Without loss of generality we consider only the case $`e_M=F^{}(e_M)=G^{}(0)=1`$ and $`U=\mathrm{\Omega }`$. We argue by contradiction and we assume that there exists a point $`x_0\mathrm{\Omega }`$ such that $`|u(x_0)|=\lambda `$, with $`\lambda >1`$. By changing the coordinate system, it is not restrictive to assume that $`x_0=0`$, $`u(0)=0`$, and $`u(0)=\lambda e_n`$, where $`e_n:=(0,\mathrm{},0,1)`$ is the last vector of the canonical basis of $`^n`$.
We want to construct a competitor $`w`$ by modifying $`u`$ in a small set $`V\mathrm{\Omega }`$ with piecewise $`C^1`$ boundary in such a way that $`w`$ is close to $`u`$ in the $`BV`$ norm and the energy of $`w`$ is strictly below the energy of $`u`$, contradicting the local minimality. In all cases we will take $`w`$ of the form
$`w=\{\begin{array}{cc}\alpha u\hfill & \text{in }V,\hfill \\ u\hfill & \text{otherwise,}\hfill \end{array}`$ (8)
for a suitable constant $`\alpha <1`$. The problem is reduced to choose $`\alpha `$ and $`V`$ such that
$`uw_{BV(\mathrm{\Omega })}<\eta \text{and}(u)(w)>0,`$ (9)
where $`\eta `$ is the constant in the definition of local minimality for $`u`$.
We consider three cases corresponding to different hypotheses on $`G`$ and $`u`$ with increasing level of difficulty.
Case 1: $`G^{\prime \prime }(0)=\mathrm{}`$.
Let us first consider the case where $`G`$ satisfies the following condition
$`\underset{t0+}{lim}{\displaystyle \frac{G(t)t}{t^2}}=\mathrm{}.`$ (10)
Let us fix $`\epsilon (0,\frac{1}{2})`$, with $`\lambda \epsilon >1`$. By the continuity of $`u`$ we can find $`R>0`$ small enough so that
$`|u\lambda e_n|<\epsilon \text{in }B_R,`$ (11)
where $`B_R`$ is the closed ball with center $`0`$ and radius $`R`$. As a consequence we can show that $`|u|>\lambda \epsilon `$ in $`B_R`$ and that there exists $`\delta >0`$ such that
$`u(x)>\delta \text{for every }xB_R\text{ with }x_n=\epsilon R,`$
$`u(x)<\delta \text{for every }xB_R\text{ with }x_n=\epsilon R.`$
This implies that for $`0<\sigma <\delta `$ the projection of the set $`\{xB_R:u(x)=\sigma \}`$ onto the hyperplane $`\{x_n=0\}`$ contains the projection of the set $`\{xB_R:x_n=\epsilon R\}`$, and therefore
$`^{n1}(B_R\{u=\sigma \})K_{\epsilon ,R}:=\omega _{n1}R^{n1}(1\epsilon ^2)^{(n1)/2},`$ (12)
where $`\omega _{n1}`$ is the $`(n1)`$-dimensional measure of the unit ball in $`^{n1}`$. Moreover (11) implies that there exists a constant $`L<+\mathrm{}`$ such that
$`^{n1}(\{xB_R:0<u(x)<\sigma \})L\sigma `$ (13)
for every $`\sigma >0`$.
For $`0<\sigma <\delta `$ we define
$$V_\sigma :=\{xB_R:0<u(x)<\sigma \}.$$
Since $`u`$ is $`C^1`$, there exists a constant $`M`$ such that
$$^{n1}(V_\sigma )M$$
for $`0<\sigma <\delta `$.
We now fix $`\alpha <1`$ such that $`\alpha (\lambda \epsilon )>1`$ and $`(1\alpha )(u_{BV(\mathrm{\Omega })}+\delta M)<\eta `$, and define $`w`$ as in (8) with $`V:=V_\sigma `$ for some $`\sigma (0,\delta )`$ to be chosen later. Since
$$wu_{BV(\mathrm{\Omega })}(1\alpha )u_{BV(\mathrm{\Omega })}+(1\alpha )\sigma ^{n1}(V_\sigma ),$$
we have $`wu_{BV(\mathrm{\Omega })}\eta `$ for $`0<\sigma <\delta `$, so that the first inequality in (9) is satisfied.
Using the definition of $``$ and $`\overline{F}`$, we get
$`(u)(w)`$ $`=`$ $`(1\alpha ){\displaystyle _{V_\sigma }}|u|𝑑x{\displaystyle _{B_R\{u=\sigma \}}}G((1\alpha )u)𝑑^{n1}`$ (14)
$`{\displaystyle _{B_RV_\sigma }}G((1\alpha )u)𝑑^{n1}.`$
Since $`u`$ is a $`C^1`$ local minimum of $``$ and $`|u|>1`$ in $`B_R`$, in particular $`u`$ is a $`C^1`$ local minimum of
$$_{B_R}|u|𝑑x$$
and then, it satisfies the Euler equation
$`\mathrm{div}\left({\displaystyle \frac{u}{|u|}}\right)=0\text{in the sense of distributions on }B_R.`$ (15)
Thus, by the divergence theorem, we have
$`{\displaystyle _{V_\sigma }}|u|𝑑x`$ $`=`$ $`{\displaystyle _{B_R\{u=\sigma \}}}u𝑑^{n1}+{\displaystyle _{B_RV_\sigma }}{\displaystyle \frac{u}{|u|}}{\displaystyle \frac{x}{|x|}}u𝑑^{n1}`$ (16)
$``$ $`{\displaystyle _{B_R\{u=\sigma \}}}u𝑑^{n1}{\displaystyle _{B_RV_\sigma }}u𝑑^{n1}.`$
Moreover, by condition (10), for any given $`c>0`$ we can choose $`\sigma `$ small enough so that
$$G((1\alpha )u)<(1\alpha )uc(1\alpha )^2u^2\text{on }V_\sigma .$$
This, together with (16) and (14), implies
$`(u)(w)`$ $``$ $`(1\alpha )\sigma ^{n1}(B_R\{u=\sigma \})(1\alpha ){\displaystyle _{B_RV_\sigma }}u𝑑^{n1}`$
$`(1\alpha )^{n1}(B_R\{u=\sigma \})[\sigma c(1\alpha )\sigma ^2]`$
$`(1\alpha ){\displaystyle _{B_RV_\sigma }}[uc(1\alpha )u^2]𝑑^{n1}`$
$``$ $`(1\alpha )\sigma [c(1\alpha )\sigma ^{n1}(B_R\{u=\sigma \})2^{n1}(B_RV_\sigma )].`$
From (12) and (13) we get
$$(u)(w)(1\alpha )\sigma ^2[c(1\alpha )K_{\epsilon ,R}2L],$$
which gives the second inequality in (9) when $`c`$ is big enough.
Next we consider the general case where $`G`$ does not necessarily satisfy (10). In this case we must choose the set $`V`$ more carefully. In order to explain the new ideas of the proof without technicalities, we prove first the result in two dimensions in the simplest case: when $`u`$ is an affine function.
Case 2: $`\mathrm{}<G^{\prime \prime }(0)<0`$, $`u`$ affine, and $`n=2`$.
We now consider the case $`n=2`$ with $`u`$ affine. We assume that $`G`$ satisfies the following condition
$`\mathrm{}<\underset{t0+}{lim}{\displaystyle \frac{G(t)t}{t^2}}<0.`$ (17)
Then there exist two constants $`c_2>c_1>0`$ such that
$`tc_2t^2<G(t)<tc_1t^2`$ (18)
for $`t>0`$ small enough.
It is not restrictive to take $`u(x)=\lambda x_2`$ for every $`x=(x_1,x_2)\mathrm{\Omega }^2`$. We assume by contradiction that $`\lambda >1`$. It is easy to check that in general we may not choose $`V`$ to be a rectangle. Indeed, if $`V=\{(x_1,x_2)\mathrm{\Omega }:0<x_1<S,0<x_2<\delta \}`$, following the computation of Case 1 we get for $`\delta >0`$ small enough
$`(u)(w)`$ $``$ $`(1\alpha )\lambda S\delta {\displaystyle _0^\delta }[(1\alpha )\lambda x_2c_2(1\alpha )^2\lambda ^2x_2^2]𝑑x_2`$
$`(1\alpha )\lambda \delta S+c_2S(1\alpha )^2\lambda ^2\delta ^2`$
$`=`$ $`(1\alpha )\lambda {\displaystyle \frac{\delta ^2}{2}}+c_2(1\alpha )^2\lambda ^2{\displaystyle \frac{\delta ^3}{3}}+c_2S(1\alpha )^2\lambda ^2\delta ^2,`$
and the right-hand side is positive for every $`\delta >0`$ only if $`S[2(1\alpha )\lambda c_2]^1`$. This condition may be incompatible with the inclusion $`V\mathrm{\Omega }`$. For the same reason we can not define $`V`$ as in Case 1.
Since the previous computation shows that the problem is given by the short sides of the rectangle, we are led to overcome this difficulty by defining a special profile for the boundary of $`V`$. Let us fix $`r`$ and $`R`$, with $`r<R`$, and let $`\phi :[0,R][0,+\mathrm{})`$ be a nonincreasing function, to be chosen later, satisfying $`\phi (\rho )=1`$ in $`0\rho r`$ and $`\phi (R)=0`$. We take $`V`$ of the form
$$V:=\{(x_1,x_2):|x_1|<R,0<x_2<\sigma \phi (|x_1|)\},$$
with $`0<\sigma <1`$, and we consider the function $`w`$ defined by (8). Let us compute the energy of $`w`$ and show that (9) holds for a suitable choice of $`r`$, $`R`$, $`\phi `$, $`\sigma `$, and $`\alpha `$.
If $`\alpha <1`$ and $`\alpha \lambda >1`$, using the definition of $`w`$ we get
$`(u)(w)`$ $`=`$ $`(1\alpha )\lambda ^2(V){\displaystyle _{V\{x_2=0\}}}G((1\alpha )\lambda x_2)𝑑^1(x)`$
$`=`$ $`2(1\alpha )\lambda r\sigma +2(1\alpha )\lambda {\displaystyle _r^R}\sigma \phi (\rho )𝑑\rho 2rG((1\alpha )\lambda \sigma )`$
$`2{\displaystyle _r^R}G((1\alpha )\lambda \sigma \phi (\rho ))\sqrt{1+(\sigma \phi ^{}(\rho ))^2}𝑑\rho .`$
Using the fact that $`\sqrt{1+t^2}1+\frac{1}{2}t^2`$ and $`0G(t)tc_1t^2`$ for small $`t>0`$ we obtain
$`{\displaystyle \frac{1}{(1\alpha )\lambda }}\left((u)(w)\right)`$ $``$ $`2c_1r(1\alpha )\lambda \sigma ^2{\displaystyle _r^R}\sigma ^3\phi (\rho )(\phi ^{}(\rho ))^2𝑑\rho `$
$`+2c_1{\displaystyle _r^R}(1\alpha )\lambda \sigma ^2(\phi (\rho ))^2𝑑\rho `$
$``$ $`{\displaystyle _r^R}[2c_1(1\alpha )\lambda \sigma ^2(\phi (\rho ))^2\sigma ^3\phi (\rho )(\phi ^{}(\rho ))^2]𝑑\rho .`$
The inequality $`(u)(w)>0`$ can be obtained easily for $`\sigma `$ small enough if
$$\phi ^{}(\rho )^2\phi (\rho )=k(\phi (\rho ))^2$$
for a suitable constant $`k`$ independent of $`\sigma `$. It is easy to check that a solution of this equation on $`[r,R]`$, with $`\phi (r)=1`$ and $`\phi (R)=0`$, is given by
$`\phi (\rho )={\displaystyle \frac{(\rho R)^2}{(rR)^2}},`$ (19)
with $`k=4(Rr)^2`$. With this choice of the profile $`\phi `$ we get
$`(u)(w)(1\alpha )\lambda {\displaystyle _r^R}[2c_1(1\alpha )\lambda \sigma ^24\sigma ^3(Rr)^2](\phi (\rho ))^2𝑑\rho .`$ (20)
Now we choose $`\alpha <1`$ such that $`\alpha \lambda >1`$ and
$$(1\alpha )[u_{BV(\mathrm{\Omega })}+2r\lambda +2\lambda _r^R\phi (\rho ))\sqrt{1+(\phi ^{}(\rho ))^2}d\rho ]<\eta .$$
Since
$$wu_{BV(\mathrm{\Omega })}(1\alpha )[u_{BV(\mathrm{\Omega })}+2\sigma r\lambda +2\lambda _r^R\sigma \phi (\rho ))\sqrt{1+(\sigma \phi ^{}(\rho ))^2}d\rho ],$$
the first inequality in (9) is satisfied for $`0<\sigma <1`$. By (20) the second inequality in (9) is satisfied for $`0<\sigma <c_1(1\alpha )\lambda (Rr)^2/2`$. This concludes the proof of Case 2.
Case 3: General case.
We finally prove the result in the general case. As in Case 1, for a given $`\epsilon (0,\frac{1}{2})`$ such that $`\lambda \epsilon >1`$ we may select $`R>0`$ so small that $`|u\lambda e_n|<\epsilon `$ and $`|u|>\lambda \epsilon `$ in $`B_R`$. Now, inspired by the calculation of Case 2, we fix $`r>0`$, with $`r<R`$, and we consider the function $`a(x)`$ defined in $`B_R`$ by $`a(x)=\phi (|x|)`$; i.e.,
$$a(x)=\{\begin{array}{cc}\frac{(|x|R)^2}{(rR)^2}\hfill & \text{if }r<|x|<R,\hfill \\ 1\hfill & \text{if }|x|r.\hfill \end{array}$$
Let $`v:=u/a`$ and $`S_\sigma =:\{xB_R:v(x)=\sigma \}=\{xB_R:u(x)=\sigma a(x)\}`$. Since $`u`$ is $`C^1`$, there exist $`\delta >0`$ and $`M>0`$ such that
$`^{n1}(S_\sigma )M`$ (21)
for $`0<\sigma <\delta `$.
We now fix $`\alpha <1`$ such that $`\alpha (\lambda \epsilon )>1`$ and $`(1\alpha )[u_{BV(\mathrm{\Omega })}+\delta M]<\eta `$, and define $`w`$ as in (8) with $`V:=\{xB_R:0v(x)\sigma \}`$. Since
$$wu_{BV(\mathrm{\Omega })}(1\alpha )u_{BV(\mathrm{\Omega })}+(1\alpha )\sigma ^{n1}(S_\sigma ),$$
we have $`wu_{BV(\mathrm{\Omega })}\eta `$ for $`0<\sigma <\delta `$, so that the first inequality in (9) is satisfied.
To conclude the proof we have to show that $`\sigma `$ can be chosen in $`(0,\delta )`$ so that the second inequality in (9) holds, contradicting the local minimality of $`u`$. If $`\delta `$ is small enough, we may assume that $`G`$ satisfies the second inequality of (18) for $`0<t<\delta `$. Let $`C_R^r:=B_RB_r`$. By the definition of $`w`$ we have $`|w|=\alpha |u|>1`$ a.e. on $`V`$ and thus
$`(u)(w)=(1\alpha ){\displaystyle _V}|u|𝑑x{\displaystyle _{B_RS_\sigma }}G((1\alpha )u)𝑑^{n1}`$ (22)
$``$ $`(1\alpha ){\displaystyle _V}|u|𝑑x(1\alpha ){\displaystyle _{B_rS_\sigma }}u𝑑^{n1}(1\alpha ){\displaystyle _{C_R^rS_\sigma }}u𝑑^{n1}`$
$`+c_1(1\alpha )^2{\displaystyle _{C_R^rS_\sigma }}u^2𝑑^{n1}+c_1(1\alpha )^2{\displaystyle _{B_rS_\sigma }}u^2𝑑^{n1}.`$
As in Case 1 we use the fact that $`u`$ satisfies (15). Since $`\frac{v}{|v|}`$ is the outer unit normal to $`S_\sigma `$ and $`v=u`$ on $`B_rS_\sigma `$, by the divergence theorem we get
$`{\displaystyle _V}|u|𝑑x={\displaystyle _{C_R^rS_\sigma }}{\displaystyle \frac{u}{|u|}}{\displaystyle \frac{v}{|v|}}u𝑑^{n1}+{\displaystyle _{B_rS_\sigma }}u𝑑^{n1}.`$ (23)
Since $`v=u/aua/a^2=(1/a)(u\sigma a)`$ on $`C_R^rS_\sigma `$, we have
$$\frac{u}{|u|}\frac{v}{|v|}=\frac{|u|\sigma {\displaystyle \frac{au}{|u|}}}{|u\sigma a|}$$
on $`C_R^rS_\sigma `$. Using Taylor’s expansion of the right-hand side with respect to $`\sigma `$ we obtain
$$\frac{u}{|u|}\frac{v}{|v|}=1+\frac{(au)^2|a|^2|u|^2}{2|u|^4}\sigma ^2+O(\sigma ^3),$$
and hence
$`{\displaystyle \frac{u}{|u|}}{\displaystyle \frac{v}{|v|}}1{\displaystyle \frac{|a|^2}{2(\lambda \epsilon )^2}}\sigma ^2+O(\sigma ^3)`$ (24)
on $`C_R^rS_\sigma `$. Since $`|a|^2a=4(Rr)^2a^2`$ on $`C_R^rS_\sigma `$, by (22), (23), and (24) we have
$$(u)(w)\sigma ^2(1\alpha )[c_1(1\alpha )K_{\epsilon ,r,R}\sigma ]_{C_R^rS_\sigma }a^2𝑑^{n1}+O(\sigma ^4),$$
with $`K_{\epsilon ,r,R}:=2(Rr)^2(\lambda \epsilon )^2`$. Taking now $`\sigma >0`$ small enough we obtain $`(u)(w)>0`$, which concludes the proof. ∎ |
warning/0507/hep-ph0507114.html | ar5iv | text | # Nonspectator effects in 𝐵→𝐾^∗𝛾 within the vector quark model.
## I Introduction
The precision measurement of the radiative decay mode $`BK^{}\gamma `$ has provided an exciting opportunity to test the Standard Model(SM) and beyond. Besides the branching ratio, the isospin asymmetry in this process which is defined as:
$$\mathrm{\Delta }_0=\frac{\mathrm{\Gamma }(\overline{B}^0\overline{K}^0\gamma )\mathrm{\Gamma }(B^{}K^{}\gamma )}{\mathrm{\Gamma }(\overline{B}^0\overline{K}^0\gamma )+\mathrm{\Gamma }(B^{}K^{}\gamma )},$$
(1)
could prove to be an important observable for examining the SM as well as discriminating between various new physics scenarios. The data from Bellebelle and Babarbabar point to isospin asymmetries of at most a few percent and consistent with zero within the experimental error:
$`\mathrm{\Delta }_0`$ $`=`$ $`+0.051\pm 0.044(\mathrm{stat}.)\pm 0.023(\mathrm{sys}.)\pm 0.024(R^{+/0})(Babar),`$ (2)
$`\mathrm{\Delta }_{0+}`$ $`=`$ $`+0.012\pm 0.044(\mathrm{stat}.)\pm 0.026(\mathrm{sys}.)(Belle),`$ (3)
where $`\mathrm{\Delta }_{0+}`$ is defined as in eq.(1) but using the charge conjugate modes. The last error in eq.(2) is due to the uncertainty in the ratio of the branching fractions of the neutral and charged B meson production in $`\mathrm{{\rm Y}}(4S)`$ decays. This asymmetry is due to the non-spectator contributions and has been estimated to be around a few percent in the SM within the QCD factorization approach in Refs. kn and bb , Brodsky-Lepage formalism petrov and the perturbative QCD method in Ref. kms . The more accurate measurement of the isospin asymmetry in the near future and a better understanding of the SM prediction for this observable should provide a sensitive testing venue for possible models of new physics. One such model is the extension of the SM with an extra generation of iso-singlet quarksahmadynagashima . Unlike the three generations of ordinary quarks in the SM, both the left- and the right-handed components of the quarks of this additional generation are invariant under $`SU(2)_L`$ gauge group. Therefore, the flavor changing weak interactions of these exotic quarks proceeds only through mixing with ordinary quarks and this results in the non-unitarity of the extended $`4\times 4`$ quark mixing matrix and thus non-vanishing flavor changing neutral currents (FCNC) at the tree level. Isospin asymmetry in $`BK^{}\gamma `$ transitions offers an excellent physical observable for constraining the parameters of this so-called vector quark model (VQM). As is shown in our result, nonspectator effects like the isospin and direct CP asymmetry in $`BK^{}\gamma `$ offer the advantage of being sensitive to only one model parameter, namely the non-unitarity parameter $`U^{sb}`$ and therefore, can provide a good constraint on the size of the FCNC in the context of VQM irrespective of the masses of the additional quarks.
## II Isospin symmetry breaking in $`BK^{}\gamma `$
The non-vanishing FCNC at the tree level leads to an additional contributing Feynmann diagram which is illustrated in Fig. 1. The amplitude for $`b\overline{q}s\overline{q}`$ transition via $`Z^0`$ exchange in the VQM can be written asahmadynagashima :
$`A^{VQM}`$ $`=`$ $`{\displaystyle \frac{ig}{2\mathrm{cos}(\theta )}}\left({\displaystyle \frac{1}{2}}U^{sb}\right)\overline{s}\gamma ^\mu (1\gamma _5)b\times {\displaystyle \frac{1}{M_Z^2}}`$ (4)
$`{\displaystyle \frac{ig}{2\mathrm{cos}(\theta )}}\left[(I_W^qQ_q\mathrm{sin}^2\theta )\overline{q}\gamma _\mu (1\gamma _5)qQ_q\mathrm{sin}^2\theta \overline{q}\gamma _\mu (1+\gamma _5)q\right],`$
where $`U^{sb}=(V^{}V)^{sb}`$ is a measure of the non-unitarity of the extended quark mixing matrix and $`I_W^q`$ is the third component of the weak isospin of quark flavor $`q`$. One can then write (4) in terms of the effective operators $`O_3`$ and $`O_5`$ which are defined as:
$`O_3`$ $`=`$ $`\overline{s}_\alpha \gamma ^\mu (1\gamma _5)b_\alpha \overline{q}_\beta \gamma _\mu (1\gamma _5)q_\beta ,`$
$`O_5`$ $`=`$ $`\overline{s}_\alpha \gamma ^\mu (1\gamma _5)b_\alpha \overline{q}_\beta \gamma _\mu (1+\gamma _5)q_\beta .`$ (5)
and therefore the contribution of the extra vector quarks results in additional terms in the Wilson coefficients $`C_3`$ and $`C_5`$ to the leading order in the strong coupling $`\alpha _s`$.
$`C_3^{VQM}`$ $`=`$ $`{\displaystyle \frac{U^{sb}}{V_{tb}V_{ts}^{}}}(I_W^qQ_q\mathrm{sin}^2\theta )={\displaystyle \frac{U^{sb}}{V_{tb}V_{ts}^{}}}\{{}_{1/2+1/3\mathrm{sin}^2\theta =0.42\mathrm{}q=down}{}^{1/22/3\mathrm{sin}^2\theta =0.35\mathrm{}q=up},`$
$`C_5^{VQM}`$ $`=`$ $`{\displaystyle \frac{U^{sb}}{V_{tb}V_{ts}^{}}}Q_q\mathrm{sin}^2\theta ={\displaystyle \frac{U^{sb}}{V_{tb}V_{ts}^{}}}\{{}_{1/3\mathrm{sin}^2\theta =0.08\mathrm{}q=down}{}^{2/3\mathrm{sin}^2\theta =0.15\mathrm{}q=up}.`$ (6)
With the upper bound $`|U^{sb}|10^3`$ coming from the rare B decays ahmadynagashima , the additional contribution due to the tree level FCNC could be comparable to the SM value of these coefficients at $`\mu =m_b`$, i.e. $`C_3=0.014`$ and $`C_5=0.041`$.
Here an explanation is in order. Strictly speaking, one should include the extra terms given in eq. (6), which are proportional to the electric charge of the light quark, in the electroweak penguin operators $`O_{7\mathrm{}10}`$hiller . However, since, as far as nonspectator effects to the leading order of $`\alpha _s`$ are concerned, one can ignore these operators within SM, we prefer to write the additional VQM-generated contributions in terms of the dominant QCD penguin operators. In any case, our results do not change had we followed the strict formulation of the problem.
Following the method of Ref. kn , one can write the nonspectator contributions to $`BK^{}\gamma `$ amplitude as $`A_q=b_qA_{lead}`$, where $`q`$ is the flavor of the light anti-quark in the B meson and $`A_{lead}`$ is the leading spectator amplitude. To leading order in the strong coupling constant $`\alpha _s`$, the main contribution to $`BK^{}\gamma `$ is from the electromagnetic penguin operator $`O_7`$ and the factorizable amplitude $`A_{lead}`$ is proportional to the form factor $`T_1^{BK^{}}`$ which parameterizes the hadronic matrix element of this operator to the leading order in $`\mathrm{\Lambda }_{QCD}/m_b`$. $`b_q`$ is the parameter that depends on the flavor of the spectator and, in fact, this parameterization leads to a simple expression for the isospin asymmetry (as defined by eq. (1)) in terms of $`b_q`$:
$$\mathrm{\Delta }_0=\mathrm{}(b_{\overline{d}}b_{\overline{u}}),$$
(7)
Using the expression for $`b_q`$ which is derived within the QCD factorization method in Ref kn , we obtain the contribution of vector quarks to the isospin asymmetry as follows:
$$\mathrm{\Delta }_0^{VQM}=\mathrm{}\left(\frac{4\pi ^2f_B}{N_cm_bT_1^{BK^{}}a_7^c}\frac{U^{sb}}{V_{tb}V_{ts}^{}}\left[0.22\frac{f_K^{}^{}F_{}}{m_b}0.28\frac{f_K^{}m_K^{}}{6\lambda _Bm_B}\right]\right).$$
(8)
The numerical input for the parameters of eq. (8) are tabulated in Table 1, which results in an isospin asymmetry due to the extra generation of quarks of the form:
$$\mathrm{\Delta }_0^{VQM}=0.08\left|\frac{U^{sb}}{V_{tb}V_{ts}^{}}\right|\mathrm{cos}(\theta +\varphi _s).$$
(9)
In the above formula, $`\theta `$ is the weak phase (CP odd) of the ratio $`\frac{U^{sb}}{V_{tb}V_{ts}^{}}`$ in the extended $`4\times 4`$ quark mixing matrix. In a particular parametrization of the mixing matrix where $`V_{tb}`$ and $`V_{ts}`$ are taken to be real as in SM, $`\theta `$ is the phase of the nonunitarity parameter $`U^{sb}`$. On the other hand, $`\varphi _s`$ is the strong phase (CP even) entering in eq. (8). For example, the imaginary part of the effective Wilson coefficient $`a_7^c`$ can be one possible source of this latter phase. We would like to point out that the extra contribution due to vector-like quarks to $`\mathrm{\Delta }_{0+}`$, i.e.
$$\mathrm{\Delta }_{0+}^{VQM}=0.08\left|\frac{U^{sb}}{V_{tb}V_{ts}^{}}\right|\mathrm{cos}(\theta \varphi _s),$$
(10)
is expected to be different from eq. (9) if $`\varphi _s`$ is appreciable. It is interesting to see if the difference between $`\mathrm{\Delta }_0`$ and $`\mathrm{\Delta }_{0+}`$, as reflected in eqs. (2) and (3), will persist in the future measurements of the isospin asymmetry. One could explain this within the vector quark model if $`|U^{sb}|`$ happens to be around its upper allowed limit, i.e. a few times $`10^3`$, with $`\theta \varphi _s\pi /4`$. In case that the strong phase $`\varphi _s`$ is negligible (clearly this is the case if the phase of $`a_7^c`$ is the only CP even phase in this transition), the isospin asymmetry due to an extra generation of vector-like quarks is sensitive only to the magnitude and phase of $`U^{sb}`$, and therefore, with more precise experimental data becoming available in the future, this observable could serve to impose a stringent constrain on the important nonunitarity parameter of the vector quark model.
## III Direct CP violation within the VQM
In case that the strong phase $`\varphi _s`$ in eq. (9) is significant, one could look into another important observable in the $`BK^{}\gamma `$ transition, i.e. direct CP violation , which in combination with isospin asymmetry help to constrain $`U^{sb}`$. Direct CP asymmetry, which is defined as:
$$A_{CP}=\frac{\mathrm{\Gamma }(B^+K^+\gamma )\mathrm{\Gamma }(B^{}K^{}\gamma )}{\mathrm{\Gamma }(B^+K^+\gamma )+\mathrm{\Gamma }(B^{}K^{}\gamma )},$$
(11)
is nonzero if at least two different diagrams with non-identical weak and strong phases contribute to the decay process. In other words, for $`BK^{}\gamma `$ transition, we should expect non-vanishing $`A_{CP}`$ from nonspectator processes if $`b_q`$ happens to include both strong as well as weak phases. In this case, eq. (11) can be written as:
$$A_{CP}=\mathrm{}(b_ub_{\overline{u}}).$$
(12)
The SM prediction for direct CP violation in $`BK^{}\gamma `$ is vanishingly small within the theoretical error. For example, using the perturbative QCD method, it is calculated to be $`A_{CP}=(0.62\pm 0.13)\times 10^2`$kms . Therefore, any significant CP asymmetry is an indication of new physics. In our case, the contribution of extra vector-like quarks to eq. (12) within the QCD factorization method is obtained as follows:
$$A_{CP}^{VQM}=\left(\frac{16\pi ^2f_B}{N_cm_bT_1^{BK^{}}}\left|\frac{U^{sb}}{a_7^cV_{tb}V_{ts}^{}}\right|\left[0.15\frac{f_K^{}^{}F_{}}{m_b}+0.35\frac{f_K^{}m_K^{}}{6\lambda _Bm_B}\right]\right)\mathrm{sin}\theta \mathrm{sin}\varphi _s.$$
(13)
Inserting the values given in Table 1 for the parameters in the above formula leads to a simple expression of the additional direct CP violation in terms of the nonunitarity parameter, its CP odd phase and the strong phase:
$$A_{CP}^{VQM}=0.27\left|\frac{U^{sb}}{V_{tb}V_{ts}^{}}\right|\mathrm{sin}\theta \mathrm{sin}\varphi _s.$$
(14)
The available experimental data on this asymmetry, i.e. $`A_{CP}=0.007\pm 0.074\pm 0.017`$ babar has large errors and is consistent with zero. The combination of isospin (eqns. (9) and 10)) and direct CP (eq. (14)) asymmetries due to vector quarks leads to the following prediction: if the difference between $`\mathrm{\Delta }_0`$ and $`\mathrm{\Delta }_{0+}`$ is mainly due to extra vector quarks then one expects a direct CP asymmetry in $`BK^{}\gamma `$ of around $`67\%`$. It will be exciting to see if the more accurate experimental measurements in the future will result in a significant shift of the central value of $`A_{CP}`$.
###### Acknowledgements.
M.A.’s research is partially funded by a discovery grant from NSERC. |
warning/0507/math-ph0507023.html | ar5iv | text | # Universality at the edge of the spectrum for unitary, orthogonal and symplectic ensembles of random matrices
## 1. Introduction
This paper is a continuation of \[DG\]. In \[DG\], the authors proved universality in the bulk for orthogonal and symplectic ensembles: here we prove universality at the edge for orthogonal and symplectic ensembles, and also for unitary ensembles. For the convenience of the reader, and to fix notation, we now summarize some of the basic theory of invariant ensembles ($`\beta =1`$, $`2`$ or $`4`$), borrowing freely and extensively from the introduction in \[DG\]. We are concerned with ensembles of matrices $`\{M\}`$ with probability distributions
(1.1)
$$𝒫_{N,\beta }(M)dM=\frac{1}{𝒵_{N,\beta }}e^{trV_\beta (M)}dM,$$
for $`\beta =1`$, $`2`$ and $`4`$, the so-called Orthogonal, Unitary and Symplectic ensembles, respectively (see \[M1\]). For $`\beta =1`$, $`2`$, $`4`$, the ensemble consists of $`N\times N`$ real symmetric matrices, $`N\times N`$ Hermitian matrices, and $`2N\times 2N`$ Hermitian self-dual matrices, respectively. In general the potential $`V_\beta (x)`$ is a real-valued function growing sufficiently rapidly as $`|x|\mathrm{}`$, but we will restrict our attention henceforth to $`V_\beta `$’s which are polynomials,
(1.2)
$$V_\beta (x)=\kappa _{2m,\beta }x^{2m}+\mathrm{},\kappa _{2m,\beta }>0.$$
In (1.1), $`dM`$ denotes Lebesgue measure on the algebraically independent entries of $`M`$, and $`𝒵_{N,\beta }`$ is a normalization constant. The above terminology for $`\beta =1`$, $`2`$ and $`4`$ reflects the fact that (1.1) is invariant under conjugation of $`M`$, $`MUMU^1`$, by orthogonal, unitary and unitary-symplectic matrices $`U`$. It follows from (1.1) that the distribution of the eigenvalues $`x_1,\mathrm{},x_N`$ of $`M`$ is given (see \[M1\]) by
(1.3)
$$P_{N,\beta }(x_1,\mathrm{},x_N)=\frac{1}{Z_{N,\beta }}\underset{1j<kN}{}|x_jx_k|^\beta \underset{j=1}{\overset{N}{}}w_\beta (x_j)$$
where again $`Z_{N,\beta }`$ is a normalization constant (partition function). Here
(1.4)
$$w_\beta (x)=\{\begin{array}{cc}e^{V_\beta (x)},\hfill & \beta =1,2\hfill \\ e^{2V_\beta (x)},\hfill & \beta =4.\hfill \end{array}$$
(The factor $`2`$ in $`w_{\beta =4}`$ reflects the fact that the eigenvalues of self-dual Hermitian matrices come in pairs.) Let $`\{p_j\}_{j0}`$ be the normalized orthogonal polynomials (OP’s) on $``$ with respect to the weight $`ww_{\beta =2}`$, and define $`\varphi _jp_jw^{1/2}`$. Note that $`(\varphi _j,\varphi _k)=\delta _{jk}`$ where $`(,)`$ denotes the standard inner product in $`L^2()`$.
For the unitary matrix ensembles an important role is played by the Christoffel–Darboux (CD) kernel
(1.5)
$$K_N(x,y)K_{N,2}(x,y)=\underset{k=0}{\overset{N1}{}}\varphi _k(x)\varphi _k(y).$$
In particular the probability density (1.3), the $`l`$-point correlation function $`R_{N,l,2}`$ and also the gap probability $`E_2(0;J)`$ that a set $`J`$ contains no eigenvalues, can all be expressed in terms of $`K_N`$, see e.g. \[M1\]. For example
(1.6)
$$R_{N,l,2}(x_1,\mathrm{},x_l)=det(K_N(x_j,x_k))_{1j,kl}.$$
The Universality Conjecture, in our situation, states that the limiting statistical behavior of the eigenvalues $`x_1,\mathrm{},x_N`$ distributed according to the law (1.3), in the appropriate scale as $`N\mathrm{}`$, should be independent of the weight $`w_\beta `$, and should depend only on the invariance properties of $`𝒫_{N,\beta }`$, $`\beta =1`$, $`2`$ or $`4`$, mentioned above. Universality has been considered extensively in the physics literature, see e.g. \[BrZ, Be, HWe, SeVe\].
The kernel $`K_N(x,y)`$ can also be expressed via the Christoffel–Darboux formula
(1.7)
$$K_N(x,y)=b_{N1}\frac{\varphi _N(x)\varphi _{N1}(y)\varphi _{N1}(x)\varphi _N(y)}{xy},$$
where $`b_{N1}`$ is a coefficient in the three-term recurrence relation for OP’s, see \[Sz\]. In view of the preceding remarks it follows that in the case $`\beta =2`$, the study of the large $`N`$ behavior of $`P_{N,2}`$, and in particular the proof of universality, reduces to the asymptotic analysis of $`b_{N1}`$ and the OP’s $`p_{N+j}`$ with $`j=0`$ or $`1`$. By a fundamental observation of Fokas, Its and Kitaev \[FoIKi\] the OP’s solve a Riemann–Hilbert problem (RHP) of a type that is amenable to the steepest descent method introduced by Deift and Zhou in \[DZ\] and further developed in \[DVZ\]. In \[DKMVZ1, DKMVZ2\] the authors analyzed the asymptotics of OP’s for very general classes of weights. In particular they proved the Universality Conjecture in the bulk in the case $`\beta =2`$ for weights $`w(x)=e^{V(x)}`$ where $`V(x)`$ is a polynomial as above, and also for $`w(x)=e^{NV(x)}`$ where $`V(x)`$ is real analytic and $`V(x)/\mathrm{log}|x|+\mathrm{}`$, as $`|x|\mathrm{}`$. The bulk scaling limit as $`N\mathrm{}`$ is described in terms of the so-called sine kernel $`K_{\mathrm{}}(xy)`$ where
(1.8)
$$K_{\mathrm{}}(t)\frac{\mathrm{sin}\pi t}{\pi t}.$$
For example \[DKMVZ2, Theorem 1.4\], for $`w(x)=e^{V(x)}`$, $`V(x)`$ polynomial, and for any $`l=2,3,\mathrm{}`$ and $`r,y_1,\mathrm{},y_l`$ in a compact set, one has as $`N\mathrm{}`$
(1.9)
$$\frac{1}{(K_N(0,0))^l}R_{N,l,2}(r+\frac{y_1}{K_N(0,0)},\mathrm{},r+\frac{y_l}{K_N(0,0)})det(K_{\mathrm{}}(y_jy_k))_{1j,kl}.$$
The scale $`x=y/K_N(0,0)`$ is chosen so that the expected number of eigenvalues per unit $`y`$-interval is one. This scaling in the bulk is standard in Random Matrix Theory. Indeed for any Borel set $`B`$,
(1.10)
$$_BR_{N,l=1,2}(x)𝑑x=𝔼\{\text{number of eigenvalues in }B\}.$$
Thus by (1.6) $`K_N(0,0)=R_{N,1,2}(0)`$ gives the density of the expected number of eigenvalues near zero. From (1.9), we see that, in the appropriate scale, the large $`N`$ behavior of the eigenvalues is universal (i.e. independent of $`V`$). Pioneering mathematical work on the Universality Conjecture in the bulk was done in \[PS\] and for the case of quartic two-interval potential $`V(x)=N(x^4tx^2)`$, $`t>0`$ (sufficiently) large, in \[BI\]. We note again that all these results apply only in the case $`\beta =2`$.
In the case $`\beta =1`$ and $`4`$ the situation is more complicated. In place of (1.5) one must use $`2\times 2`$ matrix kernels (see e.g. \[M1, TW2\])
(1.11)
$$K_{N,1}(x,y)=\left(\begin{array}{cc}S_{N,1}(x,y)& (S_{N,1}D)(x,y)\\ & \\ (ϵS_{N,1})(x,y)\frac{1}{2}sgn(xy)& S_{N,1}(y,x)\end{array}\right),N\text{ even},$$
and
(1.12)
$$K_{N,4}(x,y)=\frac{1}{2}\left(\begin{array}{cc}S_{N,4}(x,y)& (S_{N,4}D)(x,y)\\ & \\ (ϵS_{N,4})(x,y)& S_{N,4}(y,x)\end{array}\right).$$
Here $`S_{N,\beta }(x,y)`$, $`\beta =1,4`$, are certain scalar kernels (see (1.17), (1.18) below), $`D`$ denotes the differentiation operator, and $`ϵ`$ is the operator with kernel $`ϵ(x,y)=\frac{1}{2}sgn(xy)`$<sup>1</sup><sup>1</sup>1We use the standard notation $`sgnx=1,`$ $`0`$, $`1`$ for $`x>0`$, $`x=0`$, $`x<0`$, respectively.. Such matrix kernels were first introduced by Dyson \[Dy\] in the context of circular ensembles with a view to computing correlation functions. Dyson’s approach was extended to Hermitian ensembles, first by Mehta \[M2\] for $`V(x)=x^2`$, and then for more general weights by Mahoux and Mehta in \[MaM\]. A more direct and unifying approach to the results of Dyson–Mahoux–Mehta was given by Tracy and Widom in \[TW2\], where formulae (1.17), (1.18) below were derived. We see that once the kernels $`S_{N,\beta }(x,y)`$ are known, then so are the other kernels in $`K_{N,\beta }`$. As in the case $`\beta =2`$, the kernels $`K_{N,\beta }`$ give rise to explicit formulae for $`R_{N,l,\beta }`$ and $`E_\beta (0;J)`$. For example for $`\beta =1,4`$
(1.13)
$$R_{N,1,\beta }(x)R_{1,\beta }(x)=\frac{1}{2}trK_{N,\beta }(x,x)$$
and
$$R_{N,2,\beta }(x,y)=\frac{1}{4}\left(trK_{N,\beta }(x,x)\right)\left(trK_{N,\beta }(y,y)\right)\frac{1}{2}tr\left(K_{N,\beta }(x,y)K_{N,\beta }(y,x)\right),$$
and so on, see \[TW2\]. We will discuss some of the literature on edge scaling after the statement of our results, Theorem 1.1, Corollary 1.2 and 1.3 below. As indicated above, formula (1.11) only applies to the case when $`N`$ is even. When $`N`$ is odd, there is a similar, but slightly more complicated, formula (see \[AFNvM\]). As in \[DG\], throughout this paper, for $`\beta =1`$, we will restrict our attention to the case when $`N`$ is even. We expect that the methods in this paper also extend to the case $`\beta =1`$, $`N`$ odd, and we plan to consider this situation in a later publication. Of course, in situations where the asymptotics of (1.11) has been analyzed (e.g. $`V(x)=x^2`$) for all $`N`$ as $`N\mathrm{}`$, the limiting behavior of $`R_{N,l,\beta =1}`$ is indeed seen to be independent of the parity of $`N`$ (see e.g. \[M1, NW\]).
Let $`\{q_j(x)\}_{j0}`$ be any sequence of polynomials of exact degree $`j`$, $`q_j(x)=q_{j,j}x^j+\mathrm{}`$, $`q_{j,j}0`$. For $`j=0,1,2,\mathrm{}`$, set
(1.14)
$$\psi _{j,\beta }(x)=\{\begin{array}{cc}q_j(x)w_1(x),\hfill & \beta =1\hfill \\ q_j(x)(w_4(x))^{1/2},\hfill & \beta =4.\hfill \end{array}$$
Let $`M_{N,1}`$ denote the $`N\times N`$ matrix with entries
(1.15)
$$(M_{N,1})_{jk}=(\psi _{j,1},ϵ\psi _{k,1}),0j,kN1,$$
and let $`M_{N,4}`$ denote the $`2N\times 2N`$ matrix with entries
(1.16)
$$(M_{N,4})_{jk}=(\psi _{j,4},D\psi _{k,4}),0j,k2N1,$$
where again $`(,)`$ denotes the standard real inner product on $``$. The matrices $`M_{N,1}`$ and $`M_{N,4}`$ are invertible (see e.g. \[AvM, (4.17), (4.20)\]). Let $`\mu _{N,1}`$, $`\mu _{N,4}`$ denote the inverses of $`M_{N,1}`$, $`M_{N,4}`$ respectively. With these notations we have \[TW2\] the following formulae for $`S_{N,\beta }`$ in (1.11), (1.12)
(1.17)
$$S_{N,1}(x,y)=\underset{j,k=0}{\overset{N1}{}}\psi _{j,1}(x)(\mu _{N,1})_{jk}(ϵ\psi _{k,1})(y)$$
(1.18)
$$S_{N,4}(x,y)=\underset{j,k=0}{\overset{2N1}{}}\psi _{j,4}^{}(x)(\mu _{N,4})_{jk}\psi _{k,4}(y).$$
An essential feature of the above formulae is that the polynomials $`\{q_j\}`$ are arbitrary and we are free to choose them conveniently to facilitate the asymptotic analysis of (1.11), (1.12) as $`N\mathrm{}`$ (see discussion in \[DG\] and (1.21) below).
In order to state our main result we need more notation. For any $`m`$ let $`V(x)`$ be a polynomial of degree $`2m`$
(1.19)
$$V(x)=\kappa _{2m}x^{2m}+\mathrm{},\kappa _{2m}>0$$
and let $`w(x)w_{\beta =2}(x)=e^{V(x)}`$ as before. Let $`p_j(x)`$, $`j0`$, denote the OP’s with respect to $`w`$, and set $`\varphi _j(x)p_j(x)(w(x))^{1/2}`$, $`j0`$, as above. For $`\beta =1,4`$ set
(1.20)
$$V_\beta (x)\frac{1}{2}V(x)$$
and let $`N`$ be even. Then by (1.4), $`w_4=e^{2V_4}=e^V`$ and $`w_1=e^{V_1}=e^{V/2}`$. This ensures that for the choice $`q_j=p_j`$ in (1.14)
(1.21)
$$\psi _{j,\beta =1}(x)=\psi _{j,\beta =4}(x)=\varphi _j(x),$$
which enables us in turn to handle $`S_{N,1}`$ and $`S_{N/2,4}`$ in (1.17), (1.18) simultaneously (see \[DG, Remark 1.3\]). Henceforth and throughout the paper, $`K_N`$ denotes the Christoffel–Darboux (CD) kernel (1.5), (1.7) constructed out of these functions $`\varphi _j`$.
For the bulk scaling limit in \[DKMVZ1\] ($`\beta =2`$) and \[DG\] ($`\beta =1,4`$), the authors used the standard scale of one (expected) eigenvalue per unit interval. At the edge it is standard (see e.g. \[TW3\]) to use a slightly different scaling which ensures that the kernel $`K_{\text{Airy}}(\xi ,\eta )`$ (see (1.25) below) appears in the limiting forms (1.26), (1.27), (1.28) below, without any additional factors. Note that formula (1.10) also holds for $`\beta =1,4`$ and so $`R_{N,l=1,\beta }(x)`$ gives the density of the expected number of (simple) eigenvalues near $`x`$ for $`\beta =1,2,4`$. In view of (1.10), and also in view of (1.13) and (1.11), (1.12)
(1.22) $`R_{N,1,2}(x)=K_N(x,x),R_{N,1,1}(x)=S_{N,1}(x,x),R_{N/2,1,4}(x)={\displaystyle \frac{1}{2}}S_{N/2,4}(x,x).`$
To leading order, the right edge of the spectrum is located at the point $`c_N+d_N`$ where $`c_N,d_N`$ are the Mhaskar–Rakhmanov–Saff numbers in (3.1), (3.2) below. For all three cases, in the neighborhood of $`c_N+d_N`$, we use the scale
(1.23)
$$\xi \xi ^{(N)}c_N\left(1+\frac{\xi }{\alpha _NN^{2/3}}\right)+d_N$$
where $`\alpha _N`$ is given in (3.10)(2) below. As we will see (cf. Remark 1.3 below) this scaling differs slightly from a scale of one (expected) eigenvalue per unit interval.
It turns out that the off-diagonal elements in $`K_{N,\beta }`$ scale differently as $`N\mathrm{}`$. On the other hand, the statistics of the ensembles are clearly invariant (cf. discussion following (2.8) below) under the conjugation
$$K_{N,\beta }K_{N,\beta }^{(\lambda )}\left(\begin{array}{cc}\lambda ^1& 0\\ 0& \lambda \end{array}\right)K_{N,\beta }\left(\begin{array}{cc}\lambda & 0\\ 0& \lambda ^1\end{array}\right)=\left(\begin{array}{cc}(K_{N,\beta })_{11}& \lambda ^2(K_{N,\beta })_{12}\\ \lambda ^2(K_{N,\beta })_{21}& (K_{N,\beta })_{22}\end{array}\right)$$
for any scalar $`\lambda `$. For example, this is obviously true for the cluster functions $`T_{N,l,\beta }`$, $`\beta =1`$ or $`4`$, which have the form
(1.24)
$$T_{N,l,\beta }(y_1,\mathrm{},y_l)=\frac{1}{2l}\underset{\sigma }{}tr\left(K_{N,\beta }(y_{\sigma _1},y_{\sigma _2})K_{N,\beta }(y_{\sigma _2},y_{\sigma _3})\mathrm{}K_{N,\beta }(y_{\sigma _l},y_{\sigma _1})\right)$$
where the sum is taken over all permutations of $`\{1,\mathrm{},l\}`$ (see \[TW2, p. 816\]), etc.
Denote
(1.25) $`K_{\text{Airy}}(\xi ,\eta )`$ $`{\displaystyle \frac{Ai(\xi )Ai^{}(\eta )Ai^{}(\xi )Ai(\eta )}{\xi \eta }}`$
$`={\displaystyle _0^{\mathrm{}}}Ai(z+\xi )Ai(z+\eta )dz.`$
Set
$$\lambda _{(N)}\left(\frac{c_N}{\alpha _NN^{2/3}}\right)^{1/2}.$$
Theorem 1.1, and Corollary 1.2 and 1.3 below are the main results in this paper.
###### Theorem 1.1.
Let $`\beta =2`$, $`1`$ or $`4`$. For any $`V(x)`$ of degree $`2m`$ as in (1.19) define $`V_\beta (x)`$ and $`w_\beta (x)`$ as in (1.20), (1.4). Fix a number $`L_0`$. Then there exists $`c=c(L_0)>0`$ such that as $`N\mathrm{}`$<sup>2</sup><sup>2</sup>2For $`\beta =1,4`$, $`N`$ is even. the following holds uniformly for $`\xi ,\eta [L_0,+\mathrm{})`$.
In the case $`\beta =2`$:
(1.26) $`_{N,2}`$ $`{\displaystyle \frac{1}{\lambda _{(N)}^2}}K_N(\xi ^{(N)},\eta ^{(N)})K_{\text{Airy}}(\xi ,\eta )0.`$
In the case $`\beta =1`$:
(1.27) $`_{N,1}`$ $`{\displaystyle \frac{1}{\lambda _{(N)}^2}}K_{N,1}^{(\lambda _{(N)})}(\xi ^{(N)},\eta ^{(N)})K^{(1)}(\xi ,\eta )0`$
where
$`(K^{(1)})_{11}(\xi ,\eta )`$ $`=(K^{(1)})_{22}(\eta ,\xi )K_{\text{Airy}}(\xi ,\eta )+{\displaystyle \frac{1}{2}}Ai(\xi ){\displaystyle _{\mathrm{}}^\eta }Ai(t)dt`$
$`(K^{(1)})_{12}(\xi ,\eta )`$ $`_\eta K_{\text{Airy}}(\xi ,\eta ){\displaystyle \frac{1}{2}}Ai(\xi )Ai(\eta )`$
$`(K^{(1)})_{21}(\xi ,\eta )`$ $`{\displaystyle _\xi ^{\mathrm{}}}K_{\text{Airy}}(t,\eta )𝑑t`$
$`{\displaystyle \frac{1}{2}}{\displaystyle _\xi ^\eta }Ai(t)dt+{\displaystyle \frac{1}{2}}{\displaystyle _\xi ^{\mathrm{}}}Ai(t)dt{\displaystyle _\eta ^{\mathrm{}}}Ai(t)dt{\displaystyle \frac{1}{2}}sgn(\xi \eta ).`$
In the case $`\beta =4`$:
(1.28) $`_{N,4}`$ $`{\displaystyle \frac{1}{\lambda _{(N)}^2}}K_{N/2,4}^{(\lambda _{(N)})}(\xi ^{(N)},\eta ^{(N)})K^{(4)}(\xi ,\eta )0`$
where
$`2(K^{(4)})_{11}(\xi ,\eta )`$ $`=2(K^{(4)})_{22}(\eta ,\xi )K_{\text{Airy}}(\xi ,\eta ){\displaystyle \frac{1}{2}}Ai(\xi ){\displaystyle _\eta ^{\mathrm{}}}Ai(t)dt`$
$`2(K^{(4)})_{12}(\xi ,\eta )`$ $`_\eta K_{\text{Airy}}(\xi ,\eta ){\displaystyle \frac{1}{2}}Ai(\xi )Ai(\eta )`$
$`2(K^{(4)})_{21}(\xi ,\eta )`$ $`{\displaystyle _\xi ^{\mathrm{}}}K_{\text{Airy}}(t,\eta )𝑑t+{\displaystyle \frac{1}{2}}{\displaystyle _\xi ^{\mathrm{}}}Ai(t)dt{\displaystyle _\eta ^{\mathrm{}}}Ai(t)dt.`$
For the error term we have as $`N\mathrm{}`$
(1.29) $`_{N,2}`$ $`=O(N^{2/3})e^{c\xi }e^{c\eta }`$
$`_{N,1}`$ $`=o(1)\left(\begin{array}{cc}e^{c\xi }& e^{c\xi }e^{c\eta }\\ & \\ 1& e^{c\eta }\end{array}\right)`$
$`_{N,4}`$ $`=o(1)e^{c\xi }e^{c\eta }`$
uniformly for $`\xi ,\eta [L_0,+\mathrm{})`$.
###### Remark 1.1.
For $`\beta =4`$, but not for $`\beta =1`$, our methods actually prove that $`_{N,4}=O(N^{1/(2m)})e^{c\xi }e^{c\eta }`$. In order to obtain power law decay for $`_{N,1}`$, it would be sufficient to obtain power law decay in the error term in \[DG, Theorem 2.2\]: such power law decay can be obtained using more sophisticated estimates as in \[DGKV\].
We immediately have the following result. Recall formula (1.24) for the cluster functions for $`\beta =1,4`$; for $`\beta =2`$, the cluster functions have the form \[TW2, p. 815\]
$$T_{N,l,2}(y_1,\mathrm{},y_l)=\frac{1}{l}\underset{\sigma }{}K_N(y_{\sigma _1},y_{\sigma _2})K_N(y_{\sigma _2},y_{\sigma _3})\mathrm{}K_N(y_{\sigma _l},y_{\sigma _1}).$$
###### Corollary 1.2.
Let $`\beta =2`$, $`1`$ or $`4`$. Let $`V`$ be a polynomial of degree $`2m`$ and let $`K^{(\beta )}`$, $`\beta =1,4`$ be as in Theorem 1.1. Fix a number $`L_0`$. Then for $`\beta =1`$ and $`l=2,3,\mathrm{}`$ we have uniformly for $`\xi _1,\mathrm{},\xi _lL_0`$
(1.30) $`\underset{N\mathrm{}}{lim}{\displaystyle \frac{1}{(\lambda _{(N)}^2)^l}}`$ $`T_{N,l,1}((\xi _1)^{(N)},\mathrm{},(\xi _1)^{(N)})`$
$`={\displaystyle \frac{1}{2l}}{\displaystyle \underset{\sigma }{}}tr\left(K^{(1)}(\xi _{\sigma _1},\xi _{\sigma _2})K^{(1)}(\xi _{\sigma _2},\xi _{\sigma _3})\mathrm{}K^{(1)}(\xi _{\sigma _l},\xi _{\sigma _1})\right).`$
For $`\beta =4`$, the same result is true provided we replace $`T_{N,l,1}T_{N/2,l,4}`$ and $`K^{(1)}K^{(4)}`$. For $`\beta =2`$, the same result is true provided we replace $`T_{N,l,1}T_{N,l,2}`$, $`K^{(1)}K_{\text{Airy}}`$, $`\frac{1}{2l}\frac{1}{l}`$, and remove the trace.
Together with some additional estimates (see Section 2), Theorem 1.1 also yields the following universality result for the gap probabilities. Recall that for a $`2\times 2`$ block operator $`A=(A_{ij})_{i,j=1,2}`$ with $`A_{11},A_{22}`$ in trace class and $`A_{12},A_{21}`$ Hilbert–Schmidt, the regularized $`2`$-determinant (see e.g. \[Si\]) is defined by $`det_2(I+A)det((I+A)e^A)e^{tr(A_{11}+A_{22})}`$.
Let $`\lambda _1`$ denote the largest eigenvalue of a random matrix $`M`$.
###### Corollary 1.3.
Let $`\beta =2`$, $`1`$ or $`4`$. Let $`V`$ be a polynomial of degree $`2m`$ and let $`K^{(\beta )}`$, $`\beta =1,4`$ be as in Theorem 1.1. Fix a number $`L_0`$. Then the following holds.
In the case $`\beta =2`$:
(1.31) $`\underset{N\mathrm{}}{lim}Prob\left\{\lambda _1(L_0)^{(N)}\right\}=det\left(IK_{\text{Airy}}|_{L^2([L_0,+\mathrm{}))}\right)F^{(2)}(L_0).`$
In the case $`\beta =4`$:
(1.32) $`\underset{N\mathrm{}}{lim}Prob\left\{\lambda _1(L_0)^{(N)}\right\}=\sqrt{det(IK^{(4)})|_{L^2([L_0,+\mathrm{}))})}F^{(4)}(L_0).`$
In the case $`\beta =1`$, let $`g(\xi )\sqrt{1+\xi ^2}`$, $`G=diag(g,g^1)`$. Then
(1.33) $`\underset{N\mathrm{}}{lim}Prob\left\{\lambda _1(L_0)^{(N)}\right\}=\sqrt{\underset{2}{det}\left(IGK^{(1)}G^1|_{L^2([L_0,+\mathrm{}))}\right)}F^{(1)}(L_0).`$
###### Remark 1.2.
The regularized $`2`$-determinant is needed for $`\beta =1`$ because the operator with kernel $`\frac{1}{2}sgn(\xi \eta )`$ is Hilbert–Schmidt but not trace class in $`L^2([L_0,+\mathrm{}))`$. The auxiliary function $`g`$ is needed to ensure that $`GK^{(1)}G^1`$ indeed has a $`2`$-determinant: there is considerable freedom in the choice of the function $`g`$, see Remark 2.2 below.
###### Remark 1.3.
From Theorem 1.1 and (1.22) we have as $`N\mathrm{}`$,
(1.34) $`{\displaystyle \frac{c_N}{\alpha _NN^{2/3}}}R_{N,1,2}\left(t^{(N)}\right)`$ $`=K_{\text{Airy}}(t,t)+o(1)`$
$`{\displaystyle \frac{c_N}{\alpha _NN^{2/3}}}R_{N,1,1}\left(t^{(N)}\right)`$ $`=K_{\text{Airy}}(t,t)+{\displaystyle \frac{1}{2}}Ai(t){\displaystyle _{\mathrm{}}^t}Ai(u)du+o(1)`$
$`{\displaystyle \frac{c_N}{\alpha _NN^{2/3}}}R_{N/2,1,4}\left(t^{(N)}\right)`$ $`={\displaystyle \frac{1}{4}}K_{\text{Airy}}(t,t){\displaystyle \frac{1}{8}}Ai(t){\displaystyle _t^{\mathrm{}}}Ai(u)du+o(1)`$
uniformly for $`t`$ in any fixed half-line $`[L_0,+\mathrm{})`$. In particular the density of the expected number of eigenvalues at the edge of the spectrum $`c_N+d_N`$ is given by
(1.35) $`\gamma _2`$ $`(Ai^{}(0))^20.066987484`$
$`\gamma _1`$ $`(Ai^{}(0))^2+{\displaystyle \frac{1}{3}}Ai(0)0.185330168`$
$`\gamma _4`$ $`{\displaystyle \frac{1}{4}}(Ai^{}(0))^2{\displaystyle \frac{1}{24}}Ai(0)0.001954035`$
for the indicated values of $`\beta =2,1,4`$, where we have used the formula $`K_{\text{Airy}}(t,t)=(Ai^{}(t))^2t(Ai(t))^2`$ and $`_{\mathrm{}}^0Ai(u)du=\frac{2}{3}`$, $`_0^{\mathrm{}}Ai(u)du=\frac{1}{3}`$ (see \[AbSt\]). Thus setting $`t\widehat{t}/\gamma _\beta `$, $`\beta =2,1,4`$, rescales the axis so that the density of the expected number of eigenvalues per unit $`\widehat{t}`$-interval is one.
The distributions $`F^{(\beta )}(L_0)`$, $`\beta =1,2,4`$, are the celebrated Tracy–Widom distributions which turn out to have applications in an extraordinary variety of different areas of pure and applied mathematics (see for example the recent review \[TW6\]). The distributions $`F^{(\beta )}(L_0)`$ can all be expressed in terms of a certain solution of the Painlevé II equation (\[TW4, TW5\]).
The literature on edge scaling, in particular in the physics community, is vast, and we make no attempt to present an exhaustive survey. Rather we will focus on aspects of the literature which are particularly relevant to this paper. In the physics literature, early work on edge scaling for $`\beta =2`$ is due to Moore \[Mo\] and Bowick and Brézin \[BoBr\]. In the mathematical literature for $`\beta =2`$ with Gaussian weight $`V(x)=x^2`$, early work can be found in Forrester \[F\] and in the seminal work of Tracy and Widom \[TW4\], where the authors derived the Painlevé II representation mentioned above for $`F^{(2)}`$. For $`\beta =1`$ and $`4`$ in the Gaussian case $`V(x)=x^2`$, the Painlevé expressions for $`F^{(\beta )}`$ were obtained by Tracy and Widom in \[TW5\], but without computing directly the edge scaling limit of the Fredholm determinants. The edge scaling limits of matrix kernels $`K_{N,\beta }`$, $`\beta =1,4`$, in the Gaussian case were obtained by Forrester, Nagao and Honner in \[FNH\]. The convergence of the Fredholm determinants in the Gaussian case for $`\beta =1,4`$ (and also for $`\beta =2`$) was first proved only recently by Tracy and Widom in \[TW3\].
Universality at the edge for $`\beta =2`$ was considered by many authors in the physics literature (see e.g. \[KaFr\]), and for the cases $`\beta =1,4`$ see e.g. \[SeVe\]. The proof of universality at the edge for $`\beta =2`$ in Theorem 1.1 above is based on the estimates in \[DKMVZ2\] and does not use any results from \[W, TW2, DG\]. Many researchers have noted that universality at the edge for $`\beta =2`$ is true (see e.g. \[CKu\]), but we believe that the details of the proof (Theorem 1.1, $`\beta =2`$) have not been written down previously. In \[St1, St2, St3\], for $`\beta =2,1,4`$, Stojanovic proves universality at the edge (and also in the bulk) in the special case of an even quartic (two-interval) potential considered previously by Bleher and Its \[BI\] for $`\beta =2`$. Stojanovic uses a variant of the formulae in \[W\] together with the asymptotics for OP’s obtained in \[BI\]. Universality for the distribution of the largest eigenvalue for a wide class of real and complex Wigner ensembles (see \[M1\]) was proven by Soshnikov in \[So\]: the methods in \[So\] are completely different from those in the present paper and are based on the method of moments. Laguerre ensembles have been considered by many authors, see e.g. \[F, FNH\]. Various universality issues at the soft edge, and also at the hard edge and in the bulk, for generalized Laguerre ensembles for $`\beta =2`$ were analyzed recently in \[V\]. The authors are currently completing an analysis of universality questions for such ensembles in the cases $`\beta =1`$ and $`4`$, together with Kriecherbauer and Vanlessen, see \[DGKV\].
We complete this introduction with a description of Widom’s result \[W\] which is basic for our approach in this paper. Widom’s method applies to general weights $`w_\beta `$ with the property that $`w_\beta ^{}/w_\beta `$ is a rational function. This property certainly holds for our weights as in (1.4), (1.2), and also for general Laguerre type weights which we consider in the forthcoming paper \[DGKV\]. Introduce the matrices
(1.36)
$$D_N((D\varphi _j,\varphi _k))_{0j,kN1},ϵ_N((ϵ\varphi _j,\varphi _k))_{0j,kN1}.$$
It follows from \[TW1, Section 6\] that the matrix $`D_N`$ is banded with bandwidth $`2n+1`$ where
(1.37)
$$n2m1.$$
Thus $`(D_N)_{jk}=0`$ if $`|jk|>n`$. Next, let $`N`$ be greater than $`n`$, and introduce the following $`N`$-dependent $`n`$-column vectors
(1.38) $`\mathrm{\Phi }_1(x)`$ $`(\varphi _{Nn}(x),\mathrm{},\varphi _{N1}(x))^T`$
$`\mathrm{\Phi }_2(x)`$ $`(\varphi _N(x),\mathrm{},\varphi _{N+n1}(x))^T`$
$`ϵ\mathrm{\Phi }_1(x)`$ $`(ϵ\varphi _{Nn}(x),\mathrm{},ϵ\varphi _{N1}(x))^T`$
$`ϵ\mathrm{\Phi }_2(x)`$ $`(ϵ\varphi _N(x),\mathrm{},ϵ\varphi _{N+n1}(x))^T`$
and the following $`2n\times 2n`$ matrices consisting of four $`n\times n`$ blocks
(1.39)
$$B\left(\begin{array}{cc}B_{11}& B_{12}\\ & \\ B_{21}& B_{22}\end{array}\right)=((ϵ\varphi _j,\varphi _k))_{Nnj,kN+n1}.$$
and
(1.40)
$$A\left(\begin{array}{cc}0& A_{12}\\ & \\ A_{21}& 0\end{array}\right)=\left(\begin{array}{cc}0& D_{12}\\ & \\ D_{21}& 0\end{array}\right)$$
where $`\left(\begin{array}{cc}D_{11}& D_{12}\\ & \\ D_{21}& D_{22}\end{array}\right)((D\varphi _j,\varphi _k))_{Nnj,kN+n1}.`$ Finally, set
$$C=\left(\begin{array}{cc}C_{11}& C_{12}\\ & \\ C_{21}& C_{22}\end{array}\right)\left(\begin{array}{cc}I_n+(BA)_{11}& (BA)_{12}\\ & \\ (BA)_{21}& (BA)_{22}\end{array}\right).$$
Note that
(1.41)
$$C_{11}=I_n+B_{12}A_{21}=I_nB_{12}D_{21}.$$
The main result in \[W\] is the following pair of formulae for $`S_{N,1}`$ and $`S_{N/2,4}`$
(1.42) $`S_{N,1}(x,y)=K_N(x,y)(\mathrm{\Phi }_1(x)^T,0^T)`$ $`(AC(I_{2n}BAC)^1)^T`$
$`(ϵ\mathrm{\Phi }_1(y)^T,ϵ\mathrm{\Phi }_2(y)^T)^T`$
and
(1.43) $`S_{N/2,4}(x,y)=K_N(x,y)`$ $`+\mathrm{\Phi }_2(x)^TD_{21}ϵ\mathrm{\Phi }_1(y)`$
$`+\mathrm{\Phi }_2(x)^TD_{21}C_{11}^1B_{11}D_{12}ϵ\mathrm{\Phi }_2(y).`$
Observe that $`S_{N,1}`$ and $`S_{N/2,4}`$ are sums of the $`\beta =2`$ kernel $`K_N(x,y)`$ together with correction terms that depend only on $`\varphi _{N+j}`$ for $`j\{n,\mathrm{},n1\}`$. The $`\beta =4`$ case is different from the case $`\beta =1`$ since, by (1.18), for any $`x`$,
(1.44)
$$S_{N/2,4}(x,+\mathrm{})=0,K_N(x,+\mathrm{})=0.$$
Therefore in (1.43) for any (even) $`N`$ and for all $`x`$
(1.45)
$$\mathrm{\Phi }_2(x)^TD_{21}ϵ\mathrm{\Phi }_1(+\mathrm{})+\mathrm{\Phi }_2(x)^TD_{21}C_{11}^1B_{11}D_{12}ϵ\mathrm{\Phi }_2(+\mathrm{})=0.$$
As the entries of $`\mathrm{\Phi }_2(x)`$ are functionally independent, and as $`D_{12}`$ is invertible for large $`N`$ (see \[DG, (2.13)\]), it follows that
(1.46)
$$ϵ\mathrm{\Phi }_1(+\mathrm{})+C_{11}^1B_{11}D_{12}ϵ\mathrm{\Phi }_2(+\mathrm{})=0$$
for large $`N`$. From the definition of $`ϵ`$ for any integrable $`\psi `$
(1.47)
$$ϵ\psi (y)=\frac{1}{2}_{\mathrm{}}^{\mathrm{}}\psi (t)𝑑t_y^{\mathrm{}}\psi (t)𝑑t=ϵ\psi (+\mathrm{})_y^{\mathrm{}}\psi (t)𝑑t.$$
Hence (1.43), (1.45) imply
(1.48) $`S_{N/2,4}(x,y)=K_N(x,y)`$ $`+\mathrm{\Phi }_2(x)^TD_{21}\left({\displaystyle _y^{\mathrm{}}}\mathrm{\Phi }_1(t)𝑑t\right)`$
$`+\mathrm{\Phi }_2(x)^TD_{21}C_{11}^1B_{11}D_{12}\left({\displaystyle _y^{\mathrm{}}}\mathrm{\Phi }_2(t)𝑑t\right).`$
Formula (1.48) makes clear the decay properties of $`S_{N/2,4}(x,y)`$ as $`x,y+\mathrm{}`$. Note that $`S_{N,1}`$ does not satisfy (1.44): this is the reason why we introduce auxiliary functions (cf. $`G=diag(g,g^1))`$ when proving convergence of the determinant in Corollary 1.3. As noted earlier, the question of convergence of the determinants for $`\beta =1,4`$ in the Gaussian case was first treated in \[TW3\].
The following observations apply to the $`21`$ entries in the matrix kernels in the $`\beta =1`$ and $`4`$ cases. Note that by (1.17), $`(ϵS_{N,1})(x,y)`$ is skew symmetric. Thus
(1.49)
$$(ϵS_{N,1})(x,y)=(ϵS_{N,1})(x,y)(ϵS_{N,1})(y,y)=_x^yS_{N,1}(t,y)𝑑t.$$
Also, from (1.18), we see that $`(ϵS_{N/2,4})(+\mathrm{},y)=0`$ for all $`y`$. Together with (1.47), this implies that
(1.50)
$$(ϵS_{N/2,4})(x,y)=_x^{\mathrm{}}S_{N/2,4}(t,y)𝑑t.$$
These observations simplify evaluation of integrals of the CD kernel, and also integrals of the functions $`\varphi _{N+j}`$ in Sections 3 and 4 below.
###### Remark 1.4.
We note that (1.49) is also true for $`S_{N/2,4}`$, but (1.50) is more relevant for the calculations that follow.
In Section 2, we prove Theorem 1.1 and Corollary 1.3 using results on the edge scaling limits of the CD terms and the correction terms in $`K_{N,1}`$ and $`K_{N,4}`$. These scaling limits are proved in turn in Section 3 for the CD terms, and in Section 4 for the correction terms. Note that Corollary 1.2 is an immediate consequence of Theorem 1.1.
Notational remark: Throughout this paper $`c,c^{},C,C(m),c_1,c_2,\mathrm{}`$ refer to constants independent of $`N,\xi ,\eta `$. The symbols $`c,c^{},C,\mathrm{}`$ refer to generic constants, whose precise value may change from one inequality to another. The symbol $`c_N`$ however always refers to the $`N`$-dependent constant (3.1) below.
Acknowledgments. The work of the first author was supported in part by NSF grants DMS–0296084 and DMS–0500923. The second author would like to thank the Courant Institute, New York University, where he has spent a part of the academic year 2004–05, for hospitality and financial support. The second author also would like to thank Caltech for hospitality and financial support. Finally, the second author would like to thank the Swedish foundation STINT for providing basic support to visit Caltech.
## 2. Proofs of Theorem 1.1 and Corollary 1.3
The key estimates for the proofs of Theorem 1.1 and Corollary 1.3 are obtained below in Section 3 for the CD terms and in Section 4 for the correction terms.
### 2.1. Proof of Theorem 1.1
Inequality (3.8) proves the result for the $`\beta =2`$ case.
In the case $`\beta =4`$, we use (1.48) and consider the CD part and the correction term separately. The properly scaled $`11`$, $`22`$ and $`12`$ entries of $`K_{N/2,4}^{(\lambda _{(N)})}(\xi ^{(N)},\eta ^{(N)})`$ converge to the corresponding entries in (1.28) et seq. with the error estimate $`o(1)e^{c\xi }e^{c\eta }`$, uniformly for $`\xi ,\eta [L_0,+\mathrm{})`$: this follows from (3.8) for the CD kernel part, and from (4.22) and (4.17), respectively, for the correction term. By (1.50), (3.56) and (4.26), the (unscaled) $`21`$ entry $`(ϵS_{N/2,4})(\xi ^{(N)},\eta ^{(N)})`$ of $`K_{N/2,4}`$ satisfies
(2.1) $`|2(ϵS_{N/2,4})(\xi ^{(N)},\eta ^{(N)})`$ $`[({\displaystyle _\xi ^{\mathrm{}}}K_{\text{Airy}}(t,\eta )dt`$
$`+{\displaystyle \frac{1}{2}}\left({\displaystyle _\xi ^{\mathrm{}}}Ai(t)dt\right)\left({\displaystyle _\eta ^{\mathrm{}}}Ai(t)dt\right)\left]\right|o(1)e^{c\xi }e^{c\eta }`$
uniformly for $`\xi ,\eta [L_0,+\mathrm{})`$. This completes the proof of Theorem 1.1 for $`\beta =4`$.
In the case $`\beta =1`$, we use (1.42) and again consider the CD part and the correction term separately. The properly scaled $`11`$ and $`22`$ entries of $`K_{N,1}^{(\lambda _{(N)})}(\xi ^{(N)},\eta ^{(N)})`$ converge to the corresponding entries in (1.27) et seq. with the error estimates $`o(1)e^{c\xi }`$ and $`o(1)e^{c\eta }`$, respectively, uniformly for $`\xi ,\eta [L_0,+\mathrm{})`$: this follows from (3.8) for the CD kernel part (giving rise to a smaller error $`o(1)e^{c\xi }e^{c\eta }`$) and from (4.49) for the correction term. The properly scaled $`12`$ entry converges to the corresponding entry in (1.27) et seq. with error $`o(1)e^{c\xi }e^{c\eta }`$, uniformly for $`\xi ,\eta [L_0,+\mathrm{})`$: this follows from (3.8) for the CD kernel part and from (4.39) for the correction term. Finally, in view of (1.49), (3.56) and (4.50), the (unscaled) $`21`$ entry of $`K_{N,1}^{(\lambda _{(N)})}(\xi ^{(N)},\eta ^{(N)})`$ satisfies
(2.2) $`|(ϵS_{N,1})(`$ $`\xi ^{(N)},\eta ^{(N)})[{\displaystyle _\xi ^\eta }K_{\text{Airy}}(t\eta )dt{\displaystyle \frac{1}{2}}{\displaystyle _\xi ^\eta }Ai(s)ds`$
$`+{\displaystyle \frac{1}{2}}\left({\displaystyle _\xi ^\eta }Ai(s)ds\right)\left({\displaystyle _\eta ^{\mathrm{}}}Ai(t)dt\right)\left]\right|o(1)e^{c\mathrm{min}(\xi ,\eta )}=o(1)`$
with the uniform estimate $`o(1)`$ for $`\xi ,\eta L_0`$. In order to obtain the same form for the limit as claimed in Theorem 1.1, we note that for all $`\xi ,\eta `$
(2.3) $`{\displaystyle _\xi ^\eta }`$ $`K_{\text{Airy}}(t,\eta )dt+{\displaystyle \frac{1}{2}}\left({\displaystyle _\xi ^\eta }Ai(t)dt\right)\left({\displaystyle _\eta ^{\mathrm{}}}Ai(t)dt\right)`$
$`={\displaystyle _\xi ^{\mathrm{}}}K_{\text{Airy}}(t,\eta )𝑑t+{\displaystyle \frac{1}{2}}\left({\displaystyle _\xi ^{\mathrm{}}}Ai(t)dt\right)\left({\displaystyle _\eta ^{\mathrm{}}}Ai(t)dt\right).`$
Indeed, a direct calculation using the representation (1.25) for $`K_{\text{Airy}}`$ shows that the RHS of (2.3) is skew symmetric in $`\xi `$ and $`\eta `$. In particular, the RHS vanishes for $`\xi =\eta `$, as is also evident for the LHS. But the $`\xi `$ derivatives of both sides are equal and hence the identity follows. This finishes the proof of Theorem 1.1.
### 2.2. Proof of Corollary 1.3
The following basic fact is well-known (see e.g. \[ReSi\]). Let $`D=d/dx`$ denote differentiation and let $`\rho (x)`$ be any positive function such that $`\rho ^1L^2()`$. Then the operator
(2.4)
$$A=\frac{1}{\rho }\frac{1}{D+I}$$
is Hilbert–Schmidt in $`L^2()`$. Indeed, by the Fourier transform, $`A`$ is unitarily equivalent to an operator with square integrable kernel $`\widehat{(\rho ^1)}(kk^{})\frac{1}{ik^{}+1}`$, $`k,k^{}`$.
#### 2.2.1. The case $`\beta =2`$
Let $`\lambda _1`$ denote the largest eigenvalue of the matrix $`M`$ in the unitary ensemble. It is well-known (see e.g. \[TW2\]) that for finite $`N`$
$`Prob\{`$ $`\lambda _1c_N(1+{\displaystyle \frac{L_0}{\alpha _NN^{2/3}}})+d_N\}`$
$`=det\left(1{\displaystyle \frac{c_N}{\alpha _NN^{2/3}}}K_N(\xi ^{(N)},\eta ^{(N)})|_{L^2([L_0,+\mathrm{}))}\right).`$
Since $`K_N`$ is finite rank, it is indeed trace class. As the trace class determinant is continuous under the trace class convergence, we only have to prove that
(2.5)
$$\mathrm{\Delta }_N(\xi ,\eta )\frac{c_N}{\alpha _NN^{2/3}}K_N(\xi ^{(N)},\eta ^{(N)})K_{\text{Airy}}(\xi ,\eta )0,\text{as }N\mathrm{},$$
in the trace norm in $`L^2([L_0,+\mathrm{}))`$, in order to prove Corollary 1.3 for $`\beta =2`$. Let $`\chi _{L_0}^\mathrm{\#}(\xi )`$ be a $`C^{\mathrm{}}`$ function such that $`\chi _{L_0}^\mathrm{\#}(\xi )=1`$ for $`\xi L_0`$ and $`\chi _{L_0}^\mathrm{\#}(\xi )=0`$ for $`\xi L_01`$. We will show that
(2.6)
$$\chi _{L_0}^\mathrm{\#}\mathrm{\Delta }_N\chi _{L_0}^\mathrm{\#}0,N\mathrm{},$$
in the trace norm in $`L^2()`$. But then $`\chi _{L_0}\mathrm{\Delta }_N\chi _{L_0}=\chi _{L_0}\left(\chi _{L_0}^\mathrm{\#}\mathrm{\Delta }_N\chi _{L_0}^\mathrm{\#}\right)\chi _{L_0}`$ also converges to zero in trace norm in $`L^2()`$, where $`\chi _{L_0}`$ is the characteristic function of $`[L_0,+\mathrm{})`$, and this clearly proves (2.5).
Let $`\rho (\xi )=(1+\xi ^2)^{1/2}`$ and write
$$\chi _{L_0}^\mathrm{\#}\mathrm{\Delta }_N\chi _{L_0}^\mathrm{\#}=\left[\frac{1}{\rho }\frac{1}{D+I}\right]\left[(D+I)\rho \chi _{L_0}^\mathrm{\#}\mathrm{\Delta }_N\chi _{L_0}^\mathrm{\#}\right].$$
The first operator is Hilbert–Schmidt (see (2.4)) and the second operator is of order $`O(N^{2/3})`$ in Hilbert–Schmidt norm by (3.8), with $`L_0`$ replaced with $`L_01`$. This proves (2.6).
#### 2.2.2. The case $`\beta =4`$
Let $`\lambda _1`$ denote the largest eigenvalue of the matrix $`M`$ in the symplectic ensemble. Then in \[TW2\] the authors prove
$`Prob\{`$ $`\lambda _1c_N(1+{\displaystyle \frac{L_0}{\alpha _NN^{2/3}}})+d_N\}`$
$`=\sqrt{det\left(1{\displaystyle \frac{c_N}{\alpha _NN^{2/3}}}K_{N/2,4}^{(\lambda _{(N)})}(\xi ^{(N)},\eta ^{(N)})|_{L^2([L_0,+\mathrm{}))}\right)}.`$
The proof will therefore be complete if we could prove that all the four entries of $`K_{N/2,4}^{(\lambda _{(N)})}(\xi ^{(N)},\eta ^{(N)})`$ converge to the corresponding entries of $`K^{(4)}(\xi ,\eta )`$ in trace class norm in $`L^2([L_0,\mathrm{}))`$. Again we use (1.48) and prove the trace class convergence of the CD part and of the correction term separately. The trace class convergence of the CD parts of all the four entries of $`K_{N/2,4}^{(\lambda _{(N)})}`$ follows by using (3.8) and (3.56) together with the trace class convergence method in Subsection 2.2.1.
To prove the convergence in trace class for the $`11`$ and $`22`$ correction terms, we must show that
$`\mathrm{\Delta }_N(\xi ,\eta ){\displaystyle \frac{c_N}{\alpha _NN^{2/3}}}[\mathrm{\Phi }_2`$ $`(\xi ^{(N)})^TD_{21}\left({\displaystyle _{\eta ^{(N)}}^{\mathrm{}}}\mathrm{\Phi }_1(t)𝑑t\right)`$
$`+\mathrm{\Phi }_2(\xi ^{(N)})^TD_{21}C_{11}^1B_{11}D_{12}\left({\displaystyle _{\eta ^{(N)}}^{\mathrm{}}}\mathrm{\Phi }_2(t)𝑑t\right)`$
$`({\displaystyle \frac{1}{2}}Ai(\xi ){\displaystyle _\eta ^{\mathrm{}}}Ai(t)dt)]`$
(cf. (4.18), (4.22)) converges to zero in trace class in $`L^2([L_0,\mathrm{}))`$. But $`\mathrm{\Delta }_N`$ is an operator with finite rank at most $`n+1=2m=\mathrm{deg}V`$, independent of $`N`$. For such operators we have the following inequality
(2.7)
$$\mathrm{\Delta }_N_1\sqrt{2m}\mathrm{\Delta }_N_{HS}$$
where $`_1`$, $`_{HS}`$ denote the trace norm, Hilbert–Schmidt norm in $`L^2([L_0,\mathrm{}))`$, respectively. Indeed, $`|\mathrm{\Delta }_N|=\sqrt{\mathrm{\Delta }_N^{}\mathrm{\Delta }_N}`$ is also an operator of rank at most $`2m`$, and hence it has at most $`2m`$ nonzero eigenvalues, $`\sigma _1\sigma _2\mathrm{}\sigma _j>0`$, $`0j2m`$. Thus
$`\mathrm{\Delta }_N_1=tr|\mathrm{\Delta }_N|={\displaystyle \underset{i=1}{\overset{j}{}}}\sigma _i\sqrt{j}\left({\displaystyle \underset{i=1}{\overset{j}{}}}\sigma _i^2\right)^{1/2}\sqrt{2m}\mathrm{\Delta }_N_{HS}.`$
But from (4.22), $`\mathrm{\Delta }_N_{HS}=o(1)\left(_{L_0}^{\mathrm{}}_{L_0}^{\mathrm{}}e^{c\xi }e^{c\eta }𝑑\xi 𝑑\eta \right)^{1/2}=o(1)`$, $`N\mathrm{}`$, and we conclude that $`\mathrm{\Delta }_N_10`$, $`N\mathrm{}`$, as desired. A similar argument using (4.17) for the $`12`$ entry and (4.26) for the $`21`$ entry, completes the proof of Corollary 1.3 for $`\beta =4`$.
#### 2.2.3. The case $`\beta =1`$
Let $`\lambda _1`$ denote the largest eigenvalue of the matrix $`M`$ in the orthogonal ensemble. Let $`g(\xi )=\sqrt{1+\xi ^2}`$ and set $`G(\xi )=\left(\begin{array}{cc}g(\xi )& 0\\ & \\ 0& g^1(\xi )\end{array}\right)`$. Note that $`g^1(\xi )L^2()`$. Let $`g_{(N)}(t)=\sqrt{1+[\frac{\alpha _NN^{2/3}}{c_N}(tc_Nd_N)]^2}`$ and $`G_{(N)}(\xi )=\left(\begin{array}{cc}g_{(N)}(\xi )& 0\\ & \\ 0& g_{(N)}^1(\xi )\end{array}\right)`$. Note that $`g_{(N)}(\xi ^{(N)})=g(\xi )`$. Recall the definition of $`det_2`$ in the Introduction. A slight modification of the calculations in \[TW2, Section 9\] shows that
(2.8) $`Prob\{`$ $`\lambda _1c_N(1+{\displaystyle \frac{L_0}{\alpha _NN^{2/3}}})+d_N\}`$
$`=\sqrt{\underset{2}{det}\left(1{\displaystyle \frac{c_N}{\alpha _NN^{2/3}}}\left(G_{(N)}K_{N/2,4}^{(\lambda _{(N)})}G_{(N)}^1\right)(\xi ^{(N)},\eta ^{(N)})|_{L^2([L_0,+\mathrm{}))}\right)}.`$
In \[TW2, Section 9\] the authors use the fact that $`det(1+AB)=det(1+BA)`$ for appropriate operators $`A`$ and $`B`$. But one clearly has the freedom to write $`AB=AG_{(N)}^1G_{(N)}B`$, and so we also have $`det(1+AB)=det(1+AG_{(N)}^1G_{(N)}B)=det(1+G_{(N)}BAG_{(N)}^1)`$ and this leads to (2.8). We have chosen $`G_{(N)}`$ as above in such a way as to ensure that $`1+G_{(N)}BAG_{(N)}^1`$ has a $`2`$-determinant, but there is clearly great freedom in the choice of $`g_{(N)}`$, and hence of $`G_{(N)}`$. From (2.8) we see that in order to prove (1.33) it is enough to show \[Si\] that the diagonal (respectively the off-diagonal) entries of $`\frac{c_N}{\alpha _NN^{2/3}}\left(G_{(N)}K_{N/2,4}^{(\lambda _{(N)})}G_{(N)}^1\right)(\xi ^{(N)},\eta ^{(N)})`$ converge to the respective entries of $`(GK^{(1)}G^1)(\xi ,\eta )`$ in trace (respectively Hilbert–Schmidt) norm in $`L^2([L_0,\mathrm{}))`$.
We consider first the $`11`$ entry (again the $`22`$ entry can be considered similarly). This entry has the form
$`{\displaystyle \frac{c_N}{\alpha _NN^{2/3}}}g_{(N)}(\xi ^{(N)})`$ $`S_{N,1}(\xi ^{(N)},\eta ^{(N)})g_{(N)}^1(\eta ^{(N)})={\displaystyle \frac{c_N}{\alpha _NN^{2/3}}}g(\xi )S_{N,1}(\xi ^{(N)},\eta ^{(N)})g^1(\eta )`$
where $`S_{N,1}`$ is given by the CD part and the correction term as in (1.42). The proof that $`g(\xi )\left[\frac{c_N}{\alpha _NN^{2/3}}K_N(\xi ^{(N)},\eta ^{(N)})K_{\text{Airy}}(\xi ,\eta )\right]g^1(\eta )0`$, $`N\mathrm{}`$, in trace norm in $`L^2([L_0,\mathrm{}))`$ is completely analogous to the $`\beta =2`$ case in Subsection 2.2.1 (note that $`g`$ and its derivative are polynomially bounded) and the details are left to the reader.
As in the $`\beta =4`$ case above, the fact that the correction term in the $`11`$ entry has a fixed maximal rank independent of $`N`$ implies that the trace norm convergence follows from the Hilbert–Schmidt convergence. But by (4.40), (4.49)
$`|g(\xi )[\mathrm{\Phi }_1(\xi ^{(N)})^T`$ $`G_{11}\left({\displaystyle _{\eta ^{(N)}}^{\mathrm{}}}\mathrm{\Phi }_1(t)𝑑t\right)\mathrm{\Phi }_1(\xi ^{(N)})^TG_{12}\left({\displaystyle _{\eta ^{(N)}}^{\mathrm{}}}\mathrm{\Phi }_2(t)𝑑t\right)`$
$`\mathrm{\Phi }_1(\xi ^{(N)})^TG_{11}ϵ\mathrm{\Phi }_1(+\mathrm{})\mathrm{\Phi }_1(\xi ^{(N)})^TG_{12}ϵ\mathrm{\Phi }_2(+\mathrm{})`$
$`{\displaystyle \frac{1}{2}}Ai(\xi ){\displaystyle _{\mathrm{}}^\eta }Ai(t)dt]g^1(\eta )|o(1)g(\xi )e^{c\xi }g^1(\eta )`$
which is $`o(1)`$ in Hilbert–Schmidt norm in $`L^2([L_0,\mathrm{}))`$. This proves the trace class convergence of the $`11`$ (and similarly of the $`22`$) entry.
Finally, we note from the uniform pointwise bounds in (1.29) that the error terms in the $`12`$ and $`21`$ entries are bounded by $`o(1)g(\xi )e^{c\xi }e^{c\eta }g(\eta )`$ and $`o(1)g^1(\xi )g^1(\eta )`$, respectively, uniformly for $`\xi ,\eta L_0`$. This immediately implies the Hilbert–Schmidt convergence of the off-diagonal entries to their appropriate limits. This completes the proof of Corollary 1.3.
###### Remark 2.1.
With a little more work one can show that in the $`\beta =1`$ case the off-diagonal entries (apart from the term $`g^1(\xi )sgn(\xi \eta )g^1(\eta )`$) in fact converge in trace class norm, and not just in Hilbert–Schmidt norm.
###### Remark 2.2.
As noted earlier, there is considerable freedom in the choice of the auxiliary function $`g`$. We see that all we need is that $`g,g^{}`$ are polynomially bounded and $`g^1L^2()`$.
## 3. The edge scaling limits of the Christoffel–Darboux ($`\beta =2`$) kernel, and of its derivatives and integrals
### 3.1. Auxiliary facts from \[DKMVZ2\]
We now recall some notation from \[ibid.\]. Let $`d\mu _N^{(\text{eq})}(x)`$ denote the equilibrium measure (see e.g. \[SaTo\]) for OP’s corresponding to the rescaled weight $`e^{NV_N(x)}`$, $`V_N=\frac{1}{N}V(c_Nx+d_N)`$, where $`c_N`$, $`d_N`$ are the so-called Mhaskar–Rakhmanov–Saff (MRS) numbers (see \[MhSa, Ra\]). For $`V(x)=\kappa _{2m}x^{2m}+\kappa _{2m1}x^{2m1}+\mathrm{}`$ as in (1.19), we have \[ibid., Thm. 2.1\] to any order $`q`$ as $`N\mathrm{}`$
(3.1)
$$c_N=\left(\frac{1}{\kappa _{2m}}\frac{(2m)!!}{m(2m1)!!}\right)^{1/(2m)}N^{1/(2m)}+\underset{j=0}{\overset{q}{}}c_{(j)}N^{j/(2m)}+O(N^{(q+1)/(2m)})$$
and
(3.2)
$$d_N=\frac{\kappa _{2m1}}{2m\kappa _{2m}}+\underset{j=1}{\overset{q}{}}d_{(j)}N^{j/(2m)}+O(N^{(q+1)/(2m)}).$$
As $`N\mathrm{}`$, the equilibribum measure is absolutely continuous with respect to Lebesgue measure, $`d\mu _N^{(\text{eq})}(x)=\psi _N^{(\text{eq})}(x)dx`$, and is supported on the (single) interval $`[1,1]`$,
(3.3)
$$\psi _N^{(\text{eq})}(x)\psi _N(x)=\frac{1}{2\pi }|1x^2|^{1/2}\chi _{[1,1]}(x)h_N(x)$$
(see \[ibid., (2.4)\]) where $`h_N(x)`$ is a real polynomial of degree $`2m2`$ satisfying \[ibid., Prop. 5.3\]
(3.4)
$$h_N(x)h_{\text{min}}>0,x,NN_1(V).$$
Set \[ibid., (5.33)\]
$$g(z)g_N(z)=_1^1\mathrm{log}(zx)𝑑\mu _N^{\text{(eq)}}(x)=_1^1\mathrm{log}(zx)\frac{1}{2\pi }|1x^2|^{1/2}h_N(x)𝑑x,$$
$`z(\mathrm{},1]`$, and for $`z(1,1)`$ \[ibid., (5.34)\]
(3.5)
$$\mathrm{\Xi }_N(z)g_+(z)g_{}(z)=i_z^1|1x^2|^{1/2}h_N(x)𝑑x.$$
We also use the same symbol for the analytic continuation of $`\mathrm{\Xi }_N`$ to $`((\mathrm{},1][1,+\mathrm{}))`$.
Notational remark: Here we denote by $`\mathrm{\Xi }_N`$ what was denoted by $`\xi _N`$ in \[ibid.\].
For a fixed $`\delta >0`$ sufficiently small (cf. \[DG, Rem. 4.3\]), let $`R`$ denote the matrix function defined in \[DKMVZ2, (7.47)\]. The function $`R`$ is analytic in the complement of the contour $`\widehat{\mathrm{\Sigma }}_R`$ as in \[ibid., Fig. 7.6\] and is continuous up to the boundary. Furthermore by \[ibid., Thm. 7.10\], it has an asymptotic expansion
(3.6)
$$R(z)I+N^1\underset{k=0}{\overset{\mathrm{}}{}}r_k(z)N^{k/(2m)}$$
where $`\{r_k(z)\}`$ are bounded functions that are analytic in the complement of $`\{|z1|=\delta \}\{|z+1|=\delta \}`$. The expansion (3.6) is uniform for $`z\widehat{\mathrm{\Sigma }}_R`$. Moreover, by the proof of \[ibid., Thm 7.10\] and Cauchy’s theorem, it follows that (3.6) can be differentiated term by term,
(3.7)
$$\frac{d^j}{dz^j}R(z)N^1\underset{k=0}{\overset{\mathrm{}}{}}\frac{d^j}{dz^j}r_k(z)N^{k/(2m)},j=1,2,\mathrm{},$$
where again the expansion is uniform for $`z\widehat{\mathrm{\Sigma }}_R`$. Also, each $`\frac{d^j}{dz^j}r_k(z)`$ is bounded (and analytic) in the complement of $`\{|z1|=\delta \}\{|z+1|=\delta \}`$.
### 3.2. Estimates on the CD kernel and its derivatives
We will only consider the end point $`1`$ (the end point $`1`$ can be considered similarly). Let $`L_0`$ be fixed. Recall the notation in (1.25), (1.23) and (1.5). Our goal in this Subsection is to prove that for $`j,k=0,1`$, and some $`C=C(L_0)>0`$, $`c=c(L_0)>0`$, one has uniformly for $`\xi ,\eta [L_0,+\mathrm{})`$
(3.8)
$$\left|_\xi ^j_\eta ^k\left[\frac{c_N}{\alpha _NN^{2/3}}K_N(\xi ^{(N)},\eta ^{(N)})K_{\text{Airy}}(\xi ,\eta )\right]\right|CN^{2/3}e^{c\xi }e^{c\eta }.$$
Note that \[AbSt\]
(3.9) $`|`$ $`Ai(\xi )|C(1+|\xi |)^{1/4},|Ai^{}(\xi )|C(1+|\xi |)^{1/4},\xi ,`$
$`|`$ $`d^qAi(\xi )/d\xi ^q|C_1e^\xi C_2,\xi [L_0,+\mathrm{}),q=0,1,2,\mathrm{}.`$
#### 3.2.1. Auxiliary notation
Set (see \[DKMVZ2, (2.15)\] and also \[DG, (4.10)\])
(3.10)
$$f_N(x)=\alpha _NN^{2/3}(x1)\widehat{f}_N(x)$$
which satisfies the following (see (the proof of) \[DKMVZ2, Proposition 7.3\])
1. $`\widehat{f}_N(x)`$ is real analytic on $`(12\delta ,1+2\delta )`$, and to any order $`q=0,1,2,\mathrm{}`$
$$\widehat{f}_N(x)=\underset{j=0}{\overset{q}{}}N^{j/(2m)}\widehat{f}_{(j)}(x)+O(N^{(q+1)/(2m)})$$
uniformly for $`x`$ in the interval. Moreover, the functions $`\widehat{f}_{(j)}(x)`$ are also real analytic on $`12\delta <x<1+2\delta `$
2. to any order $`q=1,2,\mathrm{}`$
$$\alpha _N\left(h_N^2(1)/2\right)^{1/3}=2m^{2/3}+\underset{j=1}{\overset{q}{}}N^{j/(2m)}\alpha _{(j)}+O(N^{(q+1)/(2m)})$$
3. $`f_N^{}(x)=\alpha _NN^{2/3}W_N(x)`$, where $`W_N(x)=\widehat{f}_N(x)+(x1)\widehat{f}_N^{}(x)`$ also has an expansion uniform in $`x`$ to any order $`q=0,1,2,\mathrm{}`$ as above
$$W_N(x)=\underset{j=0}{\overset{q}{}}N^{j/(2m)}W_{(j)}(x)+O(N^{(q+1)/(2m)}).$$
The terms $`W_{(j)}(x)`$ are real analytic on $`12\delta <x<1+2\delta `$
4. $`\mathrm{max}_{k=0,1,2}\mathrm{max}_{12\delta x1+2\delta }|d^k\widehat{f}_N(x)/dx^k|M<\mathrm{}`$ for $`NN_2(V)`$
5. $`\widehat{f}_N(1)=1=W_N(1)`$ and $`\mathrm{min}_{12\delta x1+2\delta }\widehat{f}_N(x)\frac{1}{2}`$ for $`NN_2(V)`$. Also $`\widehat{f}_{(0)}(1)=1=W_{(0)}(1)`$.
Denote
(3.11) $`\xi _N`$ $`\xi /(\alpha _NN^{2/3}),\eta _N\eta /(\alpha _NN^{2/3})`$
$`I_N`$ $`[L_0,\alpha _NN^{2/3}\delta ],II_N[\alpha _NN^{2/3}\delta ,+\mathrm{}).`$
Thus, recalling (1.23), $`\xi ^{(N)}=c_N(1+\xi _N)+d_N`$ and similarly $`\eta ^{(N)}=c_N(1+\eta _N)+d_N`$. As above, let $`\delta >0`$ be fixed and sufficiently small. Consider first $`\xi _N,\eta _N`$ in a neighborhood of $`0`$. Set
(3.12) $`g_N(\xi )`$ $`\xi \widehat{f}_N\left(1+\xi _N\right)`$
$`\widehat{F}_N\left(1+\xi _N\right)`$ $`\left(2+\xi _N\right)^{1/4}\left(\widehat{f}_N\left(1+\xi _N\right)\right)^{1/4}`$
$`F_N\left(1+\xi _N\right)`$ $`N^{1/6}\alpha _N^{1/4}\widehat{F}_N\left(1+\xi _N\right)`$
and also
(3.13) $`A_0(\xi )`$ $`N^{1/6}\alpha _N^{1/4}\widehat{F}_N\left(1+\xi _N\right)Ai(g_N(\xi ))`$
$`A_1(\xi )`$ $`N^{1/6}\alpha _N^{1/4}\left(\widehat{F}_N\left(1+\xi _N\right)\right)^1Ai^{}(g_N(\xi )).`$
Note that in view of (3.10)(1)(5) and the formula
(3.14) $`g_N^{}(\xi )=\widehat{f}_N(1+\xi _N)+\xi _N\widehat{f}_N^{}(1+\xi _N)`$
there exist $`c_2>c_1>0`$ such that
(3.15) $`c_1{\displaystyle \frac{g_N(\xi )}{\xi }}c_2,\xi I_N`$
and
(3.16)
$$c_1g_N^{}(\xi )c_2,|g_N^{\prime \prime }(\xi )|c_2N^{2/3},\xi I_N.$$
Similarly one has uniformly for $`\xi I_N`$
(3.17)
$$c_1\widehat{F}_N(1+\xi _N)c_2,\left|\frac{d^k}{dz^k}\widehat{F}_N(z)|_{z=1+\xi _N}\right|C(k)$$
for some $`C(k)`$, $`k=1,2,\mathrm{}`$.
#### 3.2.2. Estimates for $`(\xi ,\eta )I_N\times I_N`$
With the above notation the following holds.
###### Proposition 3.1.
For $`(\xi ,\eta )I_N\times I_N`$
(3.18)
$$\frac{c_N}{\alpha _NN^{2/3}}K_N(\xi ^{(N)},\eta ^{(N)})=K_{\text{Airy}}(\xi ,\eta )+\frac{1}{\alpha _NN^{2/3}}\underset{j=1}{\overset{4}{}}Q_{1,j}(\xi ,\eta )$$
where
(3.19) $`Q_{1,1}(\xi ,\eta )`$ $`\left(\begin{array}{cc}A_0(\eta )& A_1(\eta )\end{array}\right)\left(\begin{array}{cc}1& i\\ & \\ 1& i\end{array}\right)`$
$`{\displaystyle _0^1}(R^T)^{}(1+\xi _N+t(\eta _N\xi _N))dt`$
$`\left(\begin{array}{cc}1& i\\ & \\ 1& i\end{array}\right)^1\left(\begin{array}{c}A_1(\xi )\\ \\ A_0(\xi )\end{array}\right)`$
and
(3.20) $`Q_{1,2}(\xi ,\eta )`$ $`Ai(g_N(\xi ))Ai^{}(g_N(\eta ))T_N(\xi ,\eta )Ai(g_N(\eta ))Ai^{}(g_N(\xi ))T_N(\eta ,\xi )`$
$`T_N(\xi ,\eta )`$ $`{\displaystyle \frac{_0^1\widehat{F}_N^{}(1+\eta _N+\tau (\xi _N\eta _N))𝑑\tau }{\widehat{F}_N(1+\eta _N)}}`$
and
(3.21) $`Q_{1,3}(\xi ,\eta )`$ $`E_N(\xi ,\eta ){\displaystyle _0^{\mathrm{}}}Ai(z+g_N(\xi ))Ai(z+g_N(\eta ))dz`$
$`E_N(\xi ,\eta )`$ $`{\displaystyle _0^1}[\eta +\tau (\xi \eta )][\widehat{f}_N^{}(1+\eta _N+\tau (\xi _N\eta _N))`$
$`+{\displaystyle _0^1}\widehat{f}_N^{}(1+\sigma (\eta _N+\tau (\xi _N\eta _N)))d\sigma ]d\tau `$
and
(3.22) $`Q_{1,4}(\xi ,\eta )`$ $`\xi ^2L_N(\xi ){\displaystyle _0^{\mathrm{}}}U_N(\xi ,z)Ai(z+g_N(\eta ))dz`$
$`+\eta ^2L_N(\eta ){\displaystyle _0^{\mathrm{}}}Ai(z+\xi )U_N(\eta ,z)dz`$
$`L_N(\xi )`$ $`{\displaystyle _0^1}\widehat{f}_N^{}(1+\sigma \xi _N)𝑑\sigma `$
$`U_N(\xi ,z)`$ $`{\displaystyle _0^1}Ai^{}\left(z+\xi +\tau (g_N(\xi )\xi )\right)d\tau .`$
###### Proof.
First, some algebra: let $`Y`$ solve the Fokas–Its–Kitaev Riemann–Hilbert problem for the polynomials orthogonal with respect to the weight $`e^{V(x)}dx`$ (see \[DKMVZ2, Thm. 3.1\]). Then as in \[DKMVZ1, (6.3)\] we find for any $`x,y`$
(3.23) $`K_N(x,y)`$ $`=e^{(V(x)+V(y))/2}{\displaystyle \frac{Y_{11}(y)Y_{21}(x)Y_{11}(x)Y_{21}(y)}{2\pi i(xy)}}`$
$`=e^{(V(x)+V(y))/2}{\displaystyle \frac{\left(\begin{array}{cc}1& 0\end{array}\right)Y_+^T(y)Y_+^T(x)\left(\begin{array}{cc}0& 1\end{array}\right)^T}{2\pi i(xy)}}.`$
Here and below $`+/`$ refer to the boundary values taken from above/below the real axis. (The choice $`Y_+`$ is made only for definiteness. Formula (3.23) clearly remains true if $`Y_+`$ is replaced with $`Y_{}`$.) Consider first $`z=1+\xi _N(1\delta ,1]`$ for $`\xi (\delta \alpha _NN^{2/3},0]`$. By \[DKMVZ2, (4.2), (4.6), (4.22)\] we have for $`S`$, the solution of the Riemann–Hilbert problem \[ibid., (4.24)–(4.26)\], (cf. \[ibid., (7.46),(7.47)\])
(3.24) $`S_+(z)=`$ $`c_N^{N\sigma _3}e^{\frac{Nl}{2}\sigma _3}Y_+(c_Nz+d_N)`$
$`\times e^{N(g_+(z)\frac{l}{2})\sigma _3}\left(\begin{array}{cc}1& 0\\ & \\ e^{N(g_+(z)g_{}(z))}& 1\end{array}\right),`$
where $`\sigma _3=\left(\begin{array}{cc}1& 0\\ & \\ 0& 1\end{array}\right)`$ and the constant $`ll_N`$ is given by \[ibid., (5.35)\]. Solving for $`Y_+`$ and substituting in (3.23) we find for $`\xi ,\eta (\delta \alpha _NN^{2/3},0]`$
(3.25) $`{\displaystyle \frac{c_N}{\alpha _NN^{2/3}}}`$ $`K_N(\xi ^{(N)},\eta ^{(N)})={\displaystyle \frac{e^{\frac{N}{2}(V_N(1+\xi _N)+V_N(1+\eta _N))}}{2\pi i(\xi \eta )}}`$
$`\times \left(\begin{array}{cc}e^{N(g_+(1+\eta _N)\frac{l}{2})}& e^{N(g_{}(1+\eta _N)\frac{l}{2})}\end{array}\right)S_+^T(1+\eta _N)`$
$`\times S_+^T(1+\xi _N)\left(\begin{array}{c}e^{N(g_{}(1+\xi _N)\frac{l}{2})}\\ \\ e^{N(g_+(1+\xi _N)\frac{l}{2})}\end{array}\right).`$
Now note that for $`z(1\delta ,1]`$, by \[ibid., (7.46), (7.47)\], $`S(z)=R(z)P_N(z)`$. By \[ibid., (7.24), (7.9), (7.23), (7.4)\],
$`P_{N,+}(z)=\sqrt{\pi }e^{i\pi /6}`$ $`\left(\begin{array}{cc}1& 1\\ & \\ i& i\end{array}\right)\left(\begin{array}{cc}F_N(z)& 0\\ & \\ 0& 1/F_N(z)\end{array}\right)`$
$`\times AI_+(f_N(z))e^{i\pi \sigma _3/6}\left(\begin{array}{cc}1& 0\\ & \\ 1& 1\end{array}\right)e^{N\mathrm{\Xi }_N(z)\sigma _3/2}`$
where
$$AI(f_N(z))\left(\begin{array}{cc}Ai(f_N(z))& Ai(\omega ^2f_N(z))\\ & \\ Ai^{}(f_N(z))& \omega ^2Ai^{}(\omega ^2f_N(z))\end{array}\right),\omega =e^{2\pi i/3}.$$
For $`z(1,1)`$, in view of \[ibid., (5.38)\]
$$V_N(z)+g_+(z)+g_{}(z)l=0$$
and we find from (3.25)
(3.26) $`{\displaystyle \frac{c_N}{\alpha _NN^{2/3}}}`$ $`K_N(\xi ^{(N)},\eta ^{(N)})={\displaystyle \frac{e^{\pi i/3}}{2\pi i(\xi \eta )}}\left(\begin{array}{cc}1& 0\end{array}\right)AI_+^T(f_N(1+\eta _N))`$
$`\times \left(\begin{array}{cc}F_N(1+\eta _N)& 0\\ & \\ 0& 1/F_N(1+\eta _N)\end{array}\right)\left(\begin{array}{cc}1& i\\ & \\ 1& i\end{array}\right)R_+^T(1+\eta _N)`$
$`\times R_+^T(1+\xi _N)\left(\begin{array}{cc}1& i\\ & \\ 1& i\end{array}\right)^1\left(\begin{array}{cc}1/F_N(1+\xi _N)& 0\\ & \\ 0& F_N(1+\xi _N)\end{array}\right)`$
$`\times AI_+^T(f_N(1+\xi _N))\left(\begin{array}{c}0\\ \\ 1\end{array}\right),\xi ,\eta (\delta \alpha _NN^{2/3},0].`$
Similar calculations for $`z[1,1+\delta )`$ lead to the same formula for all other cases $`\xi <0,\eta >0`$, etc., $`|\xi |,|\eta |\delta \alpha _NN^{2/3}`$.
Now writing
(3.27)
$$R^T(1+\eta _N)=R^T(1+\xi _N)+(\eta _N\xi _N)_0^1(R^T)^{}\left(1+\xi _N+t(\eta _N\xi _N)\right)𝑑t$$
and taking into account $`detAI_+(f_N(1+\xi _N))=1/(2\pi ie^{i\pi /3})`$ (use \[ibid., (8.38)\]) we obtain from (3.26) that
(3.28) $`{\displaystyle \frac{c_N}{\alpha _NN^{2/3}}}`$ $`K_N(\xi ^{(N)},\eta ^{(N)})={\displaystyle \frac{1}{\alpha _NN^{2/3}}}Q_{1,1}(\xi ,\eta )`$
$`+{\displaystyle \frac{1}{\xi \eta }}\{Ai(g_N(\xi ))Ai^{}(g_N(\eta )){\displaystyle \frac{F_N(1+\xi _N)}{F_N(1+\eta _N)}}(\xi \eta )\}`$
where $`Q_{1,1}`$ is as in (3.19). Now
(3.29) $`{\displaystyle \frac{F_N(1+\xi _N)}{F_N(1+\eta _N)}}`$ $`={\displaystyle \frac{\widehat{F}_N(1+\xi _N)}{\widehat{F}_N(1+\eta _N)}}`$
$`=1+(\xi _N\eta _N){\displaystyle \frac{_0^1\widehat{F}_N^{}(1+\eta _N+t(\xi _N\eta _N))𝑑t}{\widehat{F}_N(1+\eta _N)}}`$
and hence using (1.25) we rewrite (3.28) as
(3.30) $`{\displaystyle \frac{c_N}{\alpha _NN^{2/3}}}K_N(\xi ^{(N)},\eta ^{(N)})=`$ $`{\displaystyle \frac{g_N(\xi )g_N(\eta )}{\xi \eta }}K_{\text{Airy}}(g_N(\xi ),g_N(\eta ))`$
$`+{\displaystyle \frac{1}{\alpha _NN^{2/3}}}\left(Q_{1,1}(\xi ,\eta )+Q_{1,2}(\xi ,\eta )\right)`$
where $`Q_{1,2}`$ is as in (3.20). Next we write $`\frac{g_N(\xi )g_N(\eta )}{\xi \eta }=_0^1g_N^{}(\eta +\tau (\xi \eta ))𝑑\tau ,`$ and use (3.14) and
$$\widehat{f}_N\left(1+\eta _N+\tau (\xi _N\eta _N)\right)=\widehat{f}_N(1)+(\eta _N+\tau (\xi _N\eta _N))_0^1\widehat{f}_N^{}\left(1+\sigma (\eta _N+\tau (\xi _N\eta _N))\right)𝑑\sigma $$
to conclude that $`\frac{g_N(\xi )g_N(\eta )}{\xi \eta }=1+E_N(\xi ,\eta )`$ from (3.21). Hence recalling (1.25) we obtain from (3.30)
(3.31) $`{\displaystyle \frac{c_N}{\alpha _NN^{2/3}}}K_N(\xi ^{(N)},\eta ^{(N)})=K_{\text{Airy}}(g_N(\xi ),g_N(\eta ))+{\displaystyle \frac{1}{\alpha _NN^{2/3}}}{\displaystyle \underset{j=1}{\overset{3}{}}}Q_{1,j}(\xi ,\eta )`$
where $`Q_{1,3}`$ is as in (3.21). Finally again using (1.25) we find
(3.32) $`K_{\text{Airy}}(g_N(\xi ),g_N(\eta ))`$ $`={\displaystyle _0^{\mathrm{}}}Ai(z+g_N(\xi ))Ai(z+g_N(\eta ))dz`$
$`={\displaystyle _0^{\mathrm{}}}Ai(z+\xi )Ai(z+\eta )dz`$
$`+{\displaystyle _0^{\mathrm{}}}Ai(z+\xi )[Ai(z+g_N(\eta ))Ai(z+\eta )dz`$
$`+{\displaystyle _0^{\mathrm{}}}\left[Ai(z+g_N(\xi ))Ai(z+\xi )\right]Ai(z+g_N(\eta ))dz.`$
The first integral equals $`K_{\text{Airy}}(\xi ,\eta )`$. To evaluate the third integral we recall (3.12), (3.22) and note that
(3.33) $`g_N(\xi )\xi `$ $`=\xi \left[\widehat{f}_N(1+\xi _N)\widehat{f}_N(1)\right]`$
$`={\displaystyle \frac{\xi ^2}{\alpha _NN^{2/3}}}{\displaystyle _0^1}\widehat{f}_N^{}(1+\sigma \xi _N)𝑑\sigma ={\displaystyle \frac{\xi ^2}{\alpha _NN^{2/3}}}L_N(\xi )`$
which implies
(3.34) $`Ai(z+g_N(\xi ))Ai(z+\xi )`$ $`={\displaystyle \frac{\xi ^2}{\alpha _NN^{2/3}}}{\displaystyle _0^1}\widehat{f}_N^{}(1+\sigma \xi _N)𝑑\sigma `$
$`\times {\displaystyle _0^1}Ai^{}(z+\xi +\tau (g_N(\xi )\xi ))d\tau `$
$`={\displaystyle \frac{\xi ^2}{\alpha _NN^{2/3}}}L_N(\xi )U_N(\xi ,z).`$
The second integral in (3.32) is treated analogously. We conclude from (3.31), (3.32) that (3.18) holds where $`Q_{1,4}`$ is as in (3.22). The proof of Proposition 3.1 is complete. ∎
Now we prove the estimate (3.8) for $`\xi ,\eta I_N`$. Note that by (3.15) it follows that $`g_N(\xi ),g_N(\eta )`$ are bounded below by some constant $`M_0`$, and hence in the region $`(\xi ,\eta )I_N\times I_N`$, both variables are bounded below by the constant $`L_0`$. Using in addition (3.16) we conclude that we can always use the exponenial bounds on $`Ai`$ and its derivatives in (3.9), and hence for any $`m`$ and $`k=0,1,2`$, as $`N\mathrm{}`$
(3.35)
$$\left|\xi ^m\left(\frac{d}{d\xi }\right)^kAi(g_N(\xi ))\right|C(m)e^{c(m)\xi },\xi I_N.$$
Consider $`Q_{1,1}(\xi ,\eta )`$ first. Recall from (3.7) that, in particular, $`\frac{d^j}{dz^j}R(z)=O(N^1)`$, $`j=1,2,3`$, uniformly for $`|z1|\delta `$. It follows then by (3.19) using (3.13), (3.17), (3.35) that for $`j,k=0,1`$
(3.36)
$$\left|_\xi ^j_\eta ^kQ_{1,1}(\xi ,\eta )\right|\mathrm{const}N^{4/3}e^{c\xi }e^{c\eta }$$
uniformly for $`\xi ,\eta I_N`$. In the same way we find that for $`j,k=0,1`$ and $`l=2,3,4`$
(3.37)
$$\left|_\xi ^j_\eta ^kQ_{1,l}(\xi ,\eta )\right|\mathrm{const}N^{2/3}e^{c\xi }e^{c\eta }$$
uniformly for $`\xi ,\eta I_N`$. In estimating $`Q_{1,4}`$, we use the estimate
$$|g_N(\xi )\xi |C\delta |\xi |,|\xi |\delta \alpha _NN^{2/3},$$
which follows from (3.33), together with the uniform boundedness of $`L_N(\xi )`$ (see (3.10)(4)): for $`\delta `$ sufficiently small this implies that
(3.38)
$$|U_N(\xi ,z)|Ce^{cz}e^{c\xi },\xi I_N,z0,$$
with similar estimates for the $`\xi `$\- (and $`z`$-) derivatives. This proves (3.8) for $`(\xi ,\eta )I_N\times I_N`$.
#### 3.2.3. Estimates for $`(\xi ,\eta )II_N\times II_N`$
Recall from \[ibid., (4.30), (4.31), (6.16)\]
(3.39)
$$S^{(\mathrm{})}(z)N(z)=\frac{1}{2}\left(\begin{array}{cc}a(z)+a(z)^1& i(a(z)^1a(z))\\ & \\ i(a(z)a(z)^1)& a(z)+a(z)^1\end{array}\right)$$
where $`a(z)\left(\frac{z1}{z+1}\right)^{1/4}1`$ as $`z\mathrm{}`$.
###### Proposition 3.2.
For $`j,k=0,1`$ and some $`C,c>0`$
(3.40)
$$\left|_\xi ^j_\eta ^k\left(\frac{c_N}{\alpha _NN^{2/3}}K_N(\xi ^{(N)},\eta ^{(N)})\right)\right|Ce^{cN}e^{cN(\xi _N\delta )}e^{cN(\eta _N\delta )}$$
uniformly for $`\xi ,\eta II_N`$.
###### Proof.
Note first of all that (3.23) still holds. For $`z=1+\xi _N[1+\delta ,+\mathrm{})`$ we now have in place of (3.24)
(3.41) $`S_+(z)=c_N^{N\sigma _3}e^{\frac{Nl}{2}\sigma _3}Y_+(c_Nz+d_N)e^{N(g_+(z)\frac{l}{2})\sigma _3}`$
where again the constant $`ll_N`$ is given by \[ibid., (5.35)\]. Solving for $`Y_+`$ and substituting in (3.23) we find for $`\xi ,\eta II_N`$
(3.42) $`{\displaystyle \frac{c_N}{\alpha _NN^{2/3}}}K_N(\xi ^{(N)},\eta ^{(N)})=`$ $`e^{\frac{N}{2}(V_N(1+\xi _N)2g_+(1+\xi _N)+l)}e^{\frac{N}{2}(V_N(1+\eta _N)2g_+(1+\eta _N)+l)}`$
$`\times {\displaystyle \frac{\left(\begin{array}{cc}1& 0\end{array}\right)S_+^T(1+\eta _N)S_+^T(1+\xi _N)\left(\begin{array}{cc}0& 1\end{array}\right)^T}{2\pi i(\xi \eta )}}.`$
In view of \[ibid., (5.38)\]
$$V_N(1+\xi _N)+2g_+(1+\xi _N)l=\mathrm{\Xi }_{N,+}(1+\xi _N),\xi II_N.$$
Now by \[ibid., (2.14), (5.34)\] for some $`C_1(\delta ),C_2(\delta )>0`$ and $`c>0`$ for $`N`$ large enough
(3.43) $`\mathrm{\Xi }_{N,+}(1+\xi _N)`$ $`=\left({\displaystyle _1^{1+\delta }}+{\displaystyle _{1+\delta }^{1+\xi _N}}\right)\sqrt{t^21}h_N(t)dt`$
$`{\displaystyle _1^{1+\delta }}\sqrt{t^21}h_{\text{min}}𝑑t{\displaystyle _{1+\delta }^{1+\xi _N}}ch_{\text{min}}𝑑t`$
$`C_1C_2(\xi _N\delta ),\xi (\delta \alpha _NN^{2/3},+\mathrm{}).`$
By \[ibid., (7.46), (7.47)\] for $`z1+\delta `$, $`S(z)=R(z)S^{(\mathrm{})}(z)`$. Using (3.27), which is still valid for $`\xi ,\eta II_N`$, we obtain
(3.44) $`S_+^T`$ $`(1+\eta _N)S_+^T(1+\xi _N)=S_+^{(\mathrm{})T}(1+\eta _N)S_+^{(\mathrm{})T}(1+\xi _N)`$
$`+(\eta _N\xi _N)S_+^{(\mathrm{})T}(1+\eta _N)\left({\displaystyle _0^1}(R_+^T)^{}\left(1+\xi _N+t(\eta _N\xi _N)\right)𝑑t\right)S_+^{(\mathrm{})T}(1+\xi _N).`$
Substituting
$`S_+^{(\mathrm{})T}`$ $`(1+\eta _N)=S_+^{(\mathrm{})T}(1+\xi _N)`$
$`+(\eta _N\xi _N)({\displaystyle _0^1}(S_+^{(\mathrm{})T})^{}(1+\xi _N+t(\eta _N\xi _N))dt`$
in the first term in the RHS of (3.44) and noting that $`\left(\begin{array}{cc}1& 0\end{array}\right)I\left(\begin{array}{cc}0& 1\end{array}\right)^T=0`$, we obtain an expression for $`\left(\begin{array}{cc}1& 0\end{array}\right)S_+^{(\mathrm{})T}(1+\eta _N)S_+^{(\mathrm{})T}(1+\xi _N)\left(\begin{array}{cc}0& 1\end{array}\right)^T`$ which is proportional to $`(\xi \eta )`$. The exponential bounds (3.40) then follow from (3.43) and the properties of $`S^{(\mathrm{})}`$ and $`R`$ (see (3.39) and (3.6), respectively). ∎
Now we prove (3.8) for $`\xi ,\eta II_N`$ by showing that both of the two terms on the LHS of (3.8) satisfy the exponential bound. More precisely, let $`\xi II_N`$. Then either $`\xi _N2\delta `$ or $`\xi _N[\delta ,2\delta ]`$. In the former case
(3.45)
$$e^{cN(\xi _N\delta )}=e^{cN((\xi _N/2)\delta )}e^{(cN/(2\alpha _NN^{2/3}))\xi }e^\xi ,N\mathrm{},$$
since $`\alpha _N(2m)^{2/3}0`$ as $`N\mathrm{}`$. In the latter case
$$e^\xi =e^{\alpha _NN^{2/3}\xi _N}e^{\alpha _NN^{2/3}2\delta }e^{(c/2)N},N\mathrm{}$$
and hence
(3.46)
$$e^{cN}e^{cN(\xi _N\delta )}e^{(c/2)N}e^{(c/2)N}e^{(c/2)N}e^\xi ,\xi _N[\delta ,2\delta ].$$
Combining (3.45) and (3.46) we conclude that Proposition 3.2 implies
(3.47)
$$\left|_\xi ^j_\eta ^k\left(\frac{c_N}{\alpha _NN^{2/3}}K_N(\xi ^{(N)},\eta ^{(N)})\right)\right|Ce^{c^{}N}e^\xi e^\eta ,\xi ,\eta II_N.$$
Now we consider $`K_{\text{Airy}}(\xi ,\eta )`$ for $`\xi ,\eta II_N`$. It follows from \[AbSt\] that
(3.48)
$$|Ai(x)|,|Ai^{}(x)|C(L_0)e^{c(L_0)|x|^{3/2}},xL_0.$$
Using the integral representation (1.25) we estimate for $`\xi ,\eta II_N`$
(3.49)
$$|K_{\text{Airy}}(\xi ,\eta )|C_0^{\mathrm{}}e^{c(z+\xi )^{3/2}}e^{c(z+\eta )^{3/2}}𝑑z.$$
Let $`\xi II_N`$. Then $`\xi 1`$ for large $`N`$. It is elementary to verify that
(3.50)
$$(z+\xi )^{3/2}z^{3/2}+\xi ^{3/2},z0,\xi 1.$$
Next, $`\xi ^{3/2}(\delta \alpha _N)^{3/2}N\stackrel{~}{c}N`$, $`N\mathrm{}`$ and hence
(3.51)
$$\xi ^{3/2}\xi =\xi ^{3/2}(1\xi ^{1/2})c^{\prime \prime }N,N\mathrm{}.$$
Inserting (3.50), (3.51) and their analogs for $`\eta `$ in (3.49) we find
(3.52)
$$|K_{\text{Airy}}(\xi ,\eta )|Ce^{cN}e^{c\xi }e^{c\eta },\xi ,\eta II_N.$$
A similar argument using (3.48) also shows that the derivatives of $`K_{\text{Airy}}`$ satisfy the same bound. Combining (3.47) and (3.52) completes the proof of (3.8) for $`\xi ,\eta II_N`$.
#### 3.2.4. The “mixed” neighborhoods of the end point $`1`$: $`(\xi ,\eta )(I_N\times II_N)(II_N\times I_N)`$
Let us consider the case $`(\xi ,\eta )I_N\times II_N`$ (the other case is treated analogously). For $`K_{\text{Airy}}`$, we use the bound in (3.9) for $`\xi `$,
$$|Ai(z+\xi )|,|Ai^{}(z+\xi )|C(L_0)e^ze^\xi ,z0,\xi L_0,$$
together with the bound (3.48) for $`\eta `$. Inserting these bounds in (1.25) we obtain for $`j,k=0,1`$
(3.53)
$$|_\xi ^j_\eta ^kK_{\text{Airy}}(\xi ,\eta )|Ce^{cN}e^{c\xi }e^{c\eta },(\xi ,\eta )I_N\times II_N$$
as before. For $`K_N(\xi ^{(N)},\eta ^{(N)})`$, there are two cases: $`|\xi _N\eta _N|\delta /2`$ and $`|\xi _N\eta _N|>\delta /2`$. In the first case we can treat both points as lying in a $`I_N\times I_N`$ region corresponding to a larger (fixed) value of $`\delta `$ (more precisely, set $`\delta 3\delta /2`$) and hence (3.8) follows by the arguments in Subsection 3.2.2.
It remains to consider the case $`(\xi ,\eta )I_N\times II_N`$, $`|\xi _N\eta _N|\delta /2`$. For such $`\xi ,\eta `$, we have
(3.54)
$$|\xi \eta |^1N^{2/3}\alpha _N^12\delta ^1.$$
The computations that led to (3.26) and (3.42) now imply for $`\xi I_N,\eta II_N`$
(3.55) $`{\displaystyle \frac{c_N}{\alpha _NN^{2/3}}}`$ $`K_N(\xi ^{(N)},\eta ^{(N)})={\displaystyle \frac{e^{\pi i/6}e^{N\mathrm{\Xi }_N(1+\xi _N)/2}}{2\pi i(\xi \eta )}}`$
$`\times \left(\begin{array}{cc}1& 0\end{array}\right)S^{(\mathrm{})T}(1+\eta _N)R_+^T(1+\eta _N)R_+^T(1+\xi _N)`$
$`\times \left(\begin{array}{cc}1& i\\ & \\ 1& i\end{array}\right)^1\left(\begin{array}{cc}1/F_N(1+\xi _N)& 0\\ & \\ 0& F_N(1+\xi _N)\end{array}\right)`$
$`\times AI_+^T(f_N(1+\xi _N))\left(\begin{array}{c}0\\ \\ 1\end{array}\right),`$
and using the preceding estimates we find for $`j,k=0,1`$
$$\left|_\xi ^j_\eta ^k\frac{c_N}{\alpha _NN^{2/3}}K_N(\xi ^{(N)},\eta ^{(N)})\right|CN^{2/3}N^{1/6}e^{cN}e^{c\xi }e^\eta Ce^{c^{}N}e^{c\xi }e^\eta $$
uniformly for $`(\xi ,\eta )I_N\times II_N`$, $`|\xi _N\eta _N|>\delta /2`$.
There is a similar estimate for $`(\xi ,\eta )II_N\times I_N`$ which, together with (3.53), then proves (3.8) for $`(\xi ,\eta )(I_N\times II_N)(II_N\times I_N)`$.
### 3.3. Estimates on integrals of the CD kernel
For $`\xi ,\eta [L_0,+\mathrm{})`$, making a change of variables $`s=c_N(1+t_N)+d_N`$, and using (3.8) with $`j=k=0`$, we readily find
(3.56) $`|`$ $`{\displaystyle _{\xi ^{(N)}}^{\mathrm{}}}K_N(s,\eta ^{(N)})ds({\displaystyle _\xi ^{\mathrm{}}}K_{\text{Airy}}(t,\eta )dt)|CN^{2/3}e^{c\xi }e^{c\eta }`$
$`|`$ $`{\displaystyle _{\xi ^{(N)}}^{\eta ^{(N)}}}K_N(s,\eta ^{(N)})ds({\displaystyle _\xi ^\eta }K_{\text{Airy}}(t,\eta )dt)|CN^{2/3}e^{c\mathrm{min}(\xi ,\eta )}e^{c\eta }`$
uniformly for $`\xi ,\eta [L_0,+\mathrm{})`$.
## 4. The contribution of the correction term for $`\beta =1`$ and $`4`$
### 4.1. Auxiliary facts concerning integrals of the orthogonal functions $`\varphi _j`$
It was shown in \[DG, (4.14)\] that for a fixed $`j`$ the following holds as $`N\mathrm{}`$ (see (1.21))
(4.1) $`{\displaystyle _{\mathrm{}}^+\mathrm{}}\varphi _{N+j}(y)𝑑y=c_N^{1/2}N^{1/2}(2m)^{1/2}(1+(1)^{N+j}+O(N^{1/(2m)}))`$
where $`2m=\mathrm{deg}V`$. Introduce the following column vectors of size $`2m1`$
(4.2) $`𝐚`$ $`(1,0,1,0,\mathrm{},1)^T,𝐛(0,1,0,1,\mathrm{},0)^T`$
$`𝐞`$ $`𝐚+𝐛=(1,1,1,\mathrm{},1)^T.`$
By (1.47) and (4.1) as (even) $`N\mathrm{}`$
(4.3) $`ϵ\mathrm{\Phi }_1(+\mathrm{})={\displaystyle \frac{1}{2}}{\displaystyle _{\mathrm{}}^+\mathrm{}}\mathrm{\Phi }_1(y)𝑑y`$ $`=c_N^{1/2}N^{1/2}(2m)^{1/2}(𝐛+o(1))`$
$`ϵ\mathrm{\Phi }_2(+\mathrm{})={\displaystyle \frac{1}{2}}{\displaystyle _{\mathrm{}}^+\mathrm{}}\mathrm{\Phi }_2(y)𝑑y`$ $`=c_N^{1/2}N^{1/2}(2m)^{1/2}(𝐚+o(1)).`$
We need also the following result. Recall the notation (1.23), (3.11).
###### Proposition 4.1.
For any fixed $`j`$ there exist $`C,c>0`$ such that the following holds as $`N\mathrm{}`$,
(4.4)
$$\left|\varphi _{N+j}\left(t^{(N)}\right)\frac{\alpha _N^{1/4}N^{1/6}2^{1/4}}{c_N^{1/2}}Ai(t)\right|Cc_N^{1/2}N^{1/6}e^{ct},tI_NII_N.$$
This estimate implies that for a fixed $`j`$ there exist $`C,c>0`$ such that
(4.5)
$$\left|_{\xi ^{(N)}}^{\mathrm{}}\varphi _{N+j}(s)𝑑s\frac{c_N^{1/2}}{N^{1/2}}\frac{2^{1/4}}{\alpha _N^{3/4}}_\xi ^{\mathrm{}}Ai(t)\right|Cc_N^{1/2}N^{5/6}e^{c\xi },\xi I_NII_N.$$
###### Proof.
Assume first that $`j=0`$. It was shown in \[DKMVZ2, Thm. 2.2\] that (in our notation)
(4.6) $`\varphi _N(t^{(N)})=c_N^{1/2}[`$ $`\alpha _N^{1/4}N^{1/6}\widehat{F}_N(1+t_N)Ai(g_N(t))\left(1+O(N^1)\right)`$
$`\alpha _N^{1/4}N^{1/6}(\widehat{F}_N(1+t_N))^1Ai^{}(g_N(t))(1+O(N^1))]`$
where the error terms are uniform for $`tI_N`$. Using (3.35) we immediately estimate the second term above by $`Cc_N^{1/2}N^{1/6}e^{ct}`$ uniformly for $`tI_N`$. The part of the first term that corresponds to $`O(N^1)`$ is estimated similarly by $`Cc_N^{1/2}N^{5/6}e^{ct}`$, $`tI_N`$. To estimate the leading part of the first term we write
$`\widehat{F}_N`$ $`(1+t_N)Ai(g_N(t))=\widehat{F}_N(1)Ai(t)`$
$`+\widehat{F}_N(1)\left[Ai(g_N(t))Ai(t)\right]+Ai(g_N(t))\left[\widehat{F}_N(1+t_N)\widehat{F}_N(1)\right].`$
By (3.10)(5) and (3.12), $`\widehat{F}_N(1)=2^{1/4}`$. By formula (3.34) and (3.38)
$$\left|Ai(g_N(t))Ai(t)\right|CN^{2/3}t^2e^{ct}C^{}N^{2/3}e^{c^{}t},tI_N.$$
Also using (3.35) and (3.17) we obtain
$$\left|Ai(g_N(t))\right|\left|\widehat{F}_N(1+t_N)\widehat{F}_N(1)\right|CN^{2/3}|t|e^{ct}C^{}N^{2/3}e^{c^{}t},tI_N.$$
Combining the above estimates we find that
$$\left|\widehat{F}_N(1+t_N)Ai(g_N(t))2^{1/4}Ai(t)\right|CN^{2/3}e^{ct},tI_N,$$
which completes the proof of (4.4) for $`j=0`$ and $`tI_N`$. We now consider (4.4) for $`j=0`$ and $`tII_N=[\delta \alpha _NN^{2/3},\mathrm{})`$. For such $`t`$, by (3.9),
$$|Ai(t)|Ce^tCe^{cN^{2/3}}e^{t/2}$$
and hence $`\left|\frac{\alpha _N^{1/4}N^{1/6}2^{1/4}}{c_N^{1/2}}Ai(t)\right|Cc_N^{1/2}N^{1/6}e^{ct}`$. Also from \[DG, (4.8)\], we find $`\left|\varphi _N(t^{(N)})\right|Cc_N^{1/2}e^{cN}e^{ct}`$. These two estimates for $`tII_N`$, together with the previous estimate for $`tI_N`$, yield (4.4) in the case $`j=0`$ for all $`t[L_0,\mathrm{})`$.
Now fix any $`j`$ and write
(4.7)
$$\varphi _{N+j}\left(c_N\left(1+\frac{t}{\alpha _NN^{2/3}}\right)+d_N\right)=\varphi _{N+j}\left(c_{N+j}\left(1+\frac{t_{N,j}}{\alpha _{N+j}(N+j)^{2/3}}\right)+d_{N+j}\right)$$
where
(4.8) $`t_{N,j}`$ $`=t{\displaystyle \frac{c_N}{c_{N+j}}}{\displaystyle \frac{\alpha _{N+j}}{\alpha _N}}{\displaystyle \frac{(N+j)^{2/3}}{N^{2/3}}}+\left({\displaystyle \frac{c_N}{c_{N+j}}}1\right)\alpha _{N+j}(N+j)^{2/3}`$
$`+{\displaystyle \frac{d_{N+j}}{c_{N+j}}}\left({\displaystyle \frac{d_N}{d_{N+j}}}1\right)\alpha _{N+j}(N+j)^{2/3}`$
$`=(1+O(N^1))t+O(N^{1/3})`$
by (3.1), (3.2), (3.10)(2). In particular, as $`N\mathrm{}`$, $`t_{N,j}(1\frac{1}{2}sgnL_0)L_0`$. Now the RHS of (4.7) can be written as $`\varphi _N^{}((t_{N,j})^{(N^{})})`$ where $`N^{}=N+j`$. Applying the estimate (4.4) just derived for $`j=0`$, with $`L_0`$ replaced by $`(1\frac{1}{2}sgnL_0)L_0`$, we obtain
(4.9)
$$\left|\varphi _N^{}\left((t_{N,j})^{(N^{})}\right)\frac{\alpha _{N+j}^{1/4}(N+j)^{1/6}2^{1/4}}{c_{N+j}^{1/2}}Ai(t_{N,j})\right|Cc_{N+j}^{1/2}(N+j)^{1/6}e^{ct_{N,j}}$$
for all $`tL_0`$. Using (4.8), and also (3.1), (3.10)(2), together with the elementary estimate
$$\left|Ai(t_{N,j})Ai(t)\right|C^{}N^{1/3}e^{c^{}t}$$
(use (3.9)), we obtain (4.4) from (4.9) for any fixed $`j`$.
Finally (4.5) follows readily by integrating (4.4). ∎
Recall the notation (4.2). Proposition 4.1 implies that for $`j=1,2`$ one has uniformly for $`t,\xi ,\eta L_0`$
(4.10) $`|`$ $`\mathrm{\Phi }_j\left(t^{(N)}\right){\displaystyle \frac{\alpha _N^{1/4}N^{1/6}2^{1/4}}{c_N^{1/2}}}Ai(t)𝐞|Cc_N^{1/2}N^{1/6}e^{ct}`$
$`|`$ $`{\displaystyle _{\xi ^{(N)}}^{\mathrm{}}}\mathrm{\Phi }_j(s)ds\left({\displaystyle \frac{c_N^{1/2}}{N^{1/2}}}{\displaystyle \frac{2^{1/4}}{\alpha _N^{3/4}}}{\displaystyle _\xi ^{\mathrm{}}}Ai(t)dt\right)𝐞|Cc_N^{1/2}N^{5/6}e^{c\xi }`$
$`|`$ $`{\displaystyle _{\xi ^{(N)}}^{\eta ^{(N)}}}\mathrm{\Phi }_j(s)ds\left({\displaystyle \frac{c_N^{1/2}}{N^{1/2}}}{\displaystyle \frac{2^{1/4}}{\alpha _N^{3/4}}}{\displaystyle _\xi ^\eta }Ai(t)dt\right)𝐞|Cc_N^{1/2}N^{5/6}e^{c\mathrm{min}(\xi ,\eta )}.`$
### 4.2. The case $`\beta =4`$
#### 4.2.1. The contribution of the correction term to the $`12`$ entry of $`K_{N,4}`$
Since $`(SD)(x,y)=_yS(x,y)`$, the correction term in (1.43) has the form
(4.11)
$$\mathrm{\Phi }_2(x)^TD_{21}\mathrm{\Phi }_1(y)\mathrm{\Phi }_2(x)^TD_{21}C_{11}^1B_{11}D_{12}\mathrm{\Phi }_2(y).$$
Set $`x=\xi ^{(N)}`$, $`y=\eta ^{(N)}`$. Recall $`n=2m1`$, $`2m=\mathrm{deg}V`$. Note that by \[DG, (2.13)\]
(4.12)
$$D_{21}=\frac{m\kappa _{2m}}{2^{2m1}}c_N^{2m1}\left[\left(\begin{array}{cccccc}& \left(\genfrac{}{}{0pt}{}{n}{0}\right)& 0& \left(\genfrac{}{}{0pt}{}{n}{1}\right)& \mathrm{}& \left(\genfrac{}{}{0pt}{}{n}{(n1)/2}\right)\\ & 0& 1& 0& \mathrm{}& 0\\ & \mathrm{}\\ & 0& 0& 0& \mathrm{}& 1\end{array}\right)+o(1)\right]$$
and $`D_{21}=O(c_N^{2m1})=O(N^{11/(2m)})`$ as $`N\mathrm{}`$ (see (3.1)). Also since $`C_{11}=IB_{12}D_{21}`$, we see from \[ibid., (2.19)\] that $`C_{11}^1B_{11}`$ is skew symmetric, being the lower right corner of the skew symmetric matrix $`D_N^1`$. (Note that $`B_{11}`$ is the lower right $`n\times n`$ corner of $`ϵ_N`$.) Hence $`D_{21}C_{11}^1B_{11}D_{12}`$ in (4.11) is also skew, and using \[ibid., (2.13)\] and the fact that $`C_{11}^1`$ is bounded as $`N\mathrm{}`$ \[ibid., Thm. 2.4, 2.6\], we see that $`D_{21}C_{11}^1B_{11}D_{12}=O(N^{11/(2m)})`$ as $`N\mathrm{}`$. Recall that the $`12`$ entry in $`K_{N,4}(\xi ^{(N)},\eta ^{(N)})`$ has an overall scaling factor $`\left(\frac{c_N}{\alpha _NN^{2/3}}\right)^2`$. Substituting the leading term in the representation of $`\mathrm{\Phi }_j`$ in (4.10) into the first term in (4.11), and using (4.12), we obtain
(4.13) $`{\displaystyle \frac{c_N^2}{\alpha _N^2N^{4/3}}}{\displaystyle \frac{m\kappa _{2m}}{2^{2m1}}}c_N^{2m1}{\displaystyle \frac{\alpha _N^{1/2}N^{1/3}2^{1/2}}{c_N}}(\mathrm{\Sigma }_n+o(1))Ai(\xi )Ai(\eta )`$
where $`o(1)`$ is independent of $`\xi ,\eta `$ and $`\mathrm{\Sigma }_n`$ denotes the sum of all elements of the first (binomial) matrix on the RHS in (4.12). Using the formula preceding \[ibid., (6.7)\] one finds
$$\mathrm{\Sigma }_n=\frac{1}{2}\frac{m(2m)!}{(m!)^2}.$$
Recall that by (3.10)(2), \[ibid., (2.14)\] and (3.9),
(4.14) $`\alpha _N=2m^{2/3}+o(1),{\displaystyle \frac{c_N^{2m}}{N}}{\displaystyle \frac{m\kappa _{2m}}{2^{2m1}}}={\displaystyle \frac{2(m!)^2}{(2m)!}}+o(1),N\mathrm{}`$
$`|Ai(\xi )|Ce^\xi ,\xi L_0.`$
Inserting these estimates, (4.13) becomes
$$\frac{1}{2}Ai(\xi )Ai(\eta )+o(1)e^\xi e^\eta $$
uniformly for $`\xi ,\eta L_0`$ and $`o(1)`$ is independent of $`\xi ,\eta `$. The error that was made by substituting only the leading term in (4.10) in the first term in (4.13), is estimated as follows:
(4.15) $`O(c_N^2N^{4/3})O(c_N^{2m1})\{`$ $`O(N^{1/6}c_N^{1/2})c_N^{1/2}N^{1/6}\left(|Ai(\xi )|e^{c\eta }+|Ai(\eta )|e^{c\xi }\right)`$
$`+O(N^{1/3}c_N^1)e^{c\xi }e^{c\eta }\}`$
$`O(N^{1/3})e^{c\xi }e^{c\eta }`$
uniformly for $`\xi ,\eta L_0`$, and independent of the degree $`2m`$ of $`V`$.
Next we substitute the leading terms in the representation of $`\mathrm{\Phi }_2`$ in (4.10) in the second term in (4.11). By the skew symmetry of $`D_{21}C_{11}^1B_{11}D_{12}`$ noted above, the result is precisely zero. The error that is made by such a substitution is estimated in exactly the same way as in (4.15) and is also of order
(4.16) $`O(N^{1/3})e^{c\xi }e^{c\eta }`$
uniformly for $`\xi ,\eta L_0`$. We conclude that the contribution of the correction term to the $`12`$ entry is given by
(4.17)
$$\frac{1}{2}Ai(\xi )Ai(\eta )+o(1)e^{c\xi }e^{c\eta }$$
uniformly for $`\xi ,\eta L_0`$.
#### 4.2.2. The contribution of the correction term to the $`11`$ and $`22`$ entries of $`K_{N,4}`$
We consider the $`11`$ entry of $`K_{N,4}`$ (the $`22`$ entry is analyzed in the same way). The correstion term in (1.48) has the form
(4.18) $`\mathrm{\Phi }_2(x)^T`$ $`D_{21}\left({\displaystyle _y^{\mathrm{}}}\mathrm{\Phi }_1(t)𝑑t\right)`$
$`+\mathrm{\Phi }_2(x)^TD_{21}C_{11}^1B_{11}D_{12}\left({\displaystyle _y^{\mathrm{}}}\mathrm{\Phi }_2(t)𝑑t\right).`$
We set $`x=\xi ^{(N)}`$, $`y=\eta ^{(N)}`$ in (4.18). The $`11`$ (and $`22`$) entry in $`K_{N,4}(\xi ^{(N)},\eta ^{(N)})`$ has an overall scaling factor $`\frac{c_N}{\alpha _NN^{2/3}}`$. Hence, substituting the leading terms in the representation of $`\mathrm{\Phi }_2`$, $`\mathrm{\Phi }_1`$ in (4.10) into the first term in (4.18) and using (4.12), we obtain
(4.19) $`{\displaystyle \frac{c_N}{\alpha _NN^{2/3}}}{\displaystyle \frac{m\kappa _{2m}}{2^{2m1}}}c_N^{2m1}{\displaystyle \frac{\alpha _N^{1/4}N^{1/6}2^{1/4}}{c_N^{1/2}}}{\displaystyle \frac{c_N^{1/2}2^{1/4}}{\alpha _N^{3/4}N^{1/2}}}(\mathrm{\Sigma }_n+o(1))Ai(\xi )\left({\displaystyle _\eta ^{\mathrm{}}}Ai(t)dt\right)`$
where $`o(1)`$ is independent of $`\xi ,\eta `$. Computing the factor and using (4.14) as above we see that (4.19) becomes
$$\frac{1}{2}Ai(\xi )_\eta ^{\mathrm{}}Ai(t)dt+o(1)e^\xi e^\eta $$
uniformly for $`\xi ,\eta L_0`$ and $`o(1)`$ is independent of $`\xi ,\eta `$. The error that was made by substituting only the leading terms for $`\mathrm{\Phi }_2`$, $`\mathrm{\Phi }_1`$ in (4.10) into the first term in (4.18), is estimated as follows:
(4.20) $`O(c_NN^{2/3})O(c_N^{2m1})(`$ $`{\displaystyle \frac{N^{1/6}}{c_N^{1/2}}}{\displaystyle \frac{c_N^{1/2}}{N^{5/6}}}+{\displaystyle \frac{1}{N^{1/6}c_N^{1/2}}}{\displaystyle \frac{c_N^{1/2}}{N^{1/2}}}+{\displaystyle \frac{c_N^{1/2}}{N^{5/6}}}{\displaystyle \frac{1}{N^{1/6}c_N^{1/2}}})e^{c\xi }e^{c\eta }`$
$`=O(N^{1/3})\left(N^{2/3}+N^{2/3}+N^1\right)e^{c\xi }e^{c\eta }`$
$`=O(N^{1/3})e^{c\xi }e^{c\eta }`$
uniformly for $`\xi ,\eta L_0`$, and again independent of the degree $`2m`$ of $`V`$.
Next we substitute the leading terms in the representation of $`\mathrm{\Phi }_2`$, $`\mathrm{\Phi }_2`$ into (4.10) in the second term in (4.18). Again by skew symmetry, the result is precisely zero. The error that is made by such a substitution is estimated in exactly the same way as in (4.20) and also has order
(4.21) $`O(N^{1/3})e^{c\xi }e^{c\eta }`$
uniformly for $`\xi ,\eta L_0`$. We conclude that the contribution of the correction term to the $`11`$ entry is given by
(4.22)
$$\frac{1}{2}Ai(\xi )_\eta ^{\mathrm{}}Ai(t)dt+o(1)e^{c\xi }e^{c\eta }$$
uniformly for $`\xi ,\eta L_0`$ (for the $`22`$ entry $`\xi `$ and $`\eta `$ should be interchanged).
#### 4.2.3. The contribution of the correction term to the $`21`$ entry of $`K_{N,4}`$
By (1.48), (1.50) the correction term in the $`21`$ entry of $`K_{N,4}`$ is given by
(4.23) $`{\displaystyle _x^{\mathrm{}}}`$ $`\mathrm{\Phi }_2^T(s)dsD_{21}{\displaystyle _y^{\mathrm{}}}\mathrm{\Phi }_1(t)𝑑t`$
$`+{\displaystyle _x^{\mathrm{}}}\mathrm{\Phi }_2^T(s)𝑑sD_{21}C_{11}^1B_{11}D_{12}{\displaystyle _y^{\mathrm{}}}\mathrm{\Phi }_2(t)𝑑t.`$
Again we replace $`x=\xi ^{(N)}`$, $`y=\eta ^{(N)}`$. Recall that the $`21`$ entry in $`K_{N,4}(\xi ^{(N)},\eta ^{(N)})`$ has no overall scaling factor. Substituting the leading terms in the representation of $`\mathrm{\Phi }_j`$, $`j=1,2`$, in (4.10) into the first term in (4.23) in the same way as before, we obtain
$`{\displaystyle \frac{m\kappa _{2m}}{2^{2m1}}}c_N^{2m1}`$ $`{\displaystyle \frac{c_N2^{1/2}}{\alpha _N^{3/2}N}}(\mathrm{\Sigma }_n+o(1)){\displaystyle _\xi ^{\mathrm{}}}Ai(s)ds{\displaystyle _\eta ^{\mathrm{}}}Ai(t)dt`$
$`={\displaystyle \frac{1}{2}}{\displaystyle _\xi ^{\mathrm{}}}Ai(s)ds{\displaystyle _\eta ^{\mathrm{}}}Ai(t)dt+o(1)e^\xi e^\eta `$
uniformly for $`\xi ,\eta L_0`$ and $`o(1)`$ is independent of $`\xi ,\eta `$. The error just made is estimated as follows:
(4.24) $`O(c_N^{2m1})(`$ $`2{\displaystyle \frac{c_N^{1/2}}{N^{1/2}}}{\displaystyle \frac{c_N^{1/2}}{N^{5/6}}}+{\displaystyle \frac{c_N}{N^{5/3}}})e^{c\xi }e^{c\eta }`$
$`=O(N^{1/3})\left(N^{4/3}+N^{5/3}\right)e^{c\xi }e^{c\eta }`$
$`=O(N^{1/3})e^{c\xi }e^{c\eta }`$
uniformly for $`\xi ,\eta L_0`$, here all order factors are independent of $`\xi ,\eta `$. (Here we have used $`|_\xi ^{\mathrm{}}Ai(t)dt|Ce^\xi `$ uniformly for $`\xi L_0`$.)
Finally, we substitute the leading terms in the representation of $`\mathrm{\Phi }_j`$, $`j=1,2`$, in (4.10) into the second term in (4.23). By the skew symmetry the result is again precislely zero. The error that is made by such a substitution is estimated in exactly the same way as in (4.24) and is also of order
(4.25) $`O(N^{1/3})e^{c\xi }e^{c\eta }`$
uniformly for $`\xi ,\eta L_0`$. We conclude that the contribution of the correction term to the $`21`$ entry is given by
(4.26)
$$\frac{1}{2}_\xi ^{\mathrm{}}Ai(s)ds_\eta ^{\mathrm{}}Ai(t)dt+o(1)e^{c\xi }e^{c\eta }$$
uniformly for $`\xi ,\eta L_0`$.
### 4.3. The case $`\beta =1`$
As we will see, this case is more involved than the case $`\beta =4`$. Consider the $`2n\times 2n`$ ($`n=2m1`$, $`2m=\mathrm{deg}V`$) matrix $`(AC(I_{2n}BAC)^1)^T`$ in the $`\beta =1`$ correction term in (1.42) as a two by two block matrix with blocks of size $`n\times n`$. Denote the upper left and the upper right blocks by $`G_{11}`$ and $`G_{12}`$, respectively. With this notation the correction term has the form
(4.27) $`\mathrm{\Phi }_1(x)^TG_{11}ϵ\mathrm{\Phi }_1(y)\mathrm{\Phi }_1(x)^TG_{12}ϵ\mathrm{\Phi }_2(y).`$
As in \[DG\] let $`RR_n`$ denote the $`n\times n`$ matrix with all entries zero apart from ones on the anti-diagonal (thus $`R_{i,j}=1`$ if $`j=ni+1`$, $`1in`$, and $`R_{i,j}=0`$ otherwise). Note that $`R^2=I_n`$. Define
(4.28)
$$\stackrel{~}{G}_{11}RD_{21}C_{11}^1B_{11}D_{12}R.$$
Note from Subsection 4.2.1 that $`D_{21}C_{11}^1B_{11}D_{12}`$ is skew and of order $`O(N^{11/(2m)})`$ as $`N\mathrm{}`$. Hence $`\stackrel{~}{G}_{11}`$ is also skew and has the same order as $`N\mathrm{}`$. We need the following result.
###### Proposition 4.2.
As (even) $`N\mathrm{}`$ we have $`G_{11},G_{12}=O(N^{11/(2m)})`$, more precisely
(4.29)
$$G_{11}=\stackrel{~}{G}_{11}+o(N^{11/(2m)}),N\mathrm{},$$
and also
(4.30)
$$G_{12}=D_{12}+o(N^{11/(2m)}),N\mathrm{}.$$
###### Proof.
It was shown in \[DG, Theorem 2.3\] that, as $`N\mathrm{}`$,
(4.31) $`(BA)_{22}`$ $`=R(BA)_{11}R+o(1)`$
$`BAC`$ $`=\left(\begin{array}{cc}0& 0\\ & \\ (BA)_{21}+o(1)& (BA)_{22}+o(1)\end{array}\right).`$
Denote
(4.32)
$$TI_n(BAC)_{22}=I_n(BA)_{22}+o(1)=I_n+R(BA)_{11}R+o(1)=RC_{11}R+o(1).$$
It was shown in \[DG, Theorem 2.6\] that, as $`N\mathrm{}`$, $`T`$ approaches a constant nondegenerate matrix. Thus
$$(I_{2n}BAC)^1=\left(\begin{array}{cc}I_n& 0\\ & \\ T^1((BA)_{21}+o(1))& T^1\end{array}\right)$$
and simple algebra using (1.39), (1.40) now shows that in the product $`AC(I_{2n}BAC)^1=\left(\begin{array}{cc}G_{11}^T& \\ & \\ G_{12}^T& \end{array}\right)`$ we have by (4.31)
(4.33) $`G_{11}^T`$ $`=A_{12}\left[(BA)_{21}+(BA)_{22}T^1((BA)_{21}+o(1))\right]`$
$`G_{12}^T`$ $`=A_{21}\left[I_n+(BA)_{11}+(BA)_{12}T^1((BA)_{21}+o(1))\right].`$
Using (4.32), this implies
(4.34) $`N^{1+1/(2m)}G_{11}^T`$ $`=N^{1+1/(2m)}A_{12}\left[I_n+(BA)_{22}T^1\right](BA)_{21}+o(1)`$
$`=N^{1+1/(2m)}A_{12}\left[T+(BA)_{22}\right]T^1(BA)_{21}+o(1)`$
$`=N^{1+1/(2m)}A_{12}T^1(BA)_{21}+o(1).`$
Now from
(4.35)
$$BA=\left(\begin{array}{cc}B_{12}A_{21}& B_{11}A_{12}\\ & \\ B_{22}A_{21}& B_{21}A_{12}\end{array}\right)$$
and (4.34), (4.32) we obtain
$`N^{1+1/(2m)}G_{11}^T`$ $`=N^{1+1/(2m)}A_{12}(RR+R(BA)_{11}R)^1B_{22}A_{21}+o(1)`$
$`=N^{1+1/(2m)}A_{12}R(I_n+(BA)_{11})^1RB_{22}A_{21}+o(1)`$
$`=N^{1+1/(2m)}A_{12}RC_{11}^1RB_{22}RRA_{21}+o(1).`$
Using the asymptotic relations
(4.36) $`N^{1+1/(2m)}RA_{12}R`$ $`=N^{1+1/(2m)}A_{21}+o(1)`$
$`N^{11/(2m)}RB_{22}R`$ $`=N^{11/(2m)}B_{11}+o(1)`$
from \[DG, Subsec. 5.2\] we see that
$`N^{1+1/(2m)}RG_{11}^TR`$ $`=N^{1+1/(2m)}(RA_{12}R)C_{11}^1(RB_{22}R)(RA_{21}R)+o(1)`$
$`=N^{1+1/(2m)}A_{21}C_{11}^1B_{11}A_{12}+o(1)`$
$`=N^{1+1/(2m)}D_{21}C_{11}^1B_{11}D_{12}+o(1).`$
As noted above, the matrix $`D_{21}C_{11}^1B_{11}D_{12}`$ is skew symmetric and hence
$$N^{1+1/(2m)}G_{11}=N^{1+1/(2m)}RD_{21}C_{11}^1B_{11}D_{12}R+o(1)$$
which proves (4.29).
Now let us prove (4.30). From (4.33) we derive
$`N^{1+1/(2m)}G_{12}^T`$ $`=N^{1+1/(2m)}A_{21}\left[I_n+(BA)_{11}+(BA)_{12}T^1(BA)_{21}\right]+o(1)`$
and hence, because $`A_{21}^T=A_{12}=D_{12}`$, we note that we just have to prove
$$N^{1+1/(2m)}A_{21}\left[(BA)_{11}+(BA)_{12}T^1(BA)_{21}\right]=o(1).$$
Since $`N^{1+1/(2m)}A_{21}=O(1)`$, it is sufficient to prove
$$(BA)_{11}+(BA)_{12}T^1(BA)_{21}=o(1).$$
By (4.35) the LHS is $`B_{12}A_{21}+B_{11}A_{12}T^1B_{22}A_{21}`$ and so we see that it is sufficient to show that
$$B_{12}+B_{11}A_{12}T^1B_{22}=o(N^{1+1/(2m)}).$$
Using (4.32) this reduces to showing that
$$B_{12}+B_{11}A_{12}RC_{11}^1RB_{22}=o(N^{1+1/(2m)})$$
or
$$RB_{12}R+(RB_{11}R)(RA_{12}R)C_{11}^1(RB_{22}R)=o(N^{1+1/(2m)}).$$
Using $`N^{1+1/(2m)}RB_{12}R=N^{1+1/(2m)}B_{21}+o(1)`$ which follows as in (4.36), we are reduced to proving finally
(4.37)
$$B_{21}+B_{22}A_{21}C_{11}^1B_{11}=o(N^{1+1/(2m)}).$$
But
$$B_{21}+B_{22}A_{21}C_{11}^1B_{11}=0$$
by (taking the transposes of) \[DG, (5.12)\]. The proof of Proposition 4.2 is complete. ∎
###### Remark 4.1.
The second relation in (4.31) was sharpened recently by Kriecherbauer and Vanlessen \[KV\] who showed that the $`o(1)`$ terms are in fact identically zero. One might hope that this improved result could be used to strengthen the estimates in (4.29), (4.30). This is indeed the case for (4.30): one can show that $`G_{12}=D_{12}`$ identically. However we have not been able to use \[KV\] to improve the estimate in (4.29).
#### 4.3.1. The contribution of the correction term to the $`12`$ entry of $`K_{N,1}`$
In view of (4.27), since $`(SD)(x,y)=_yS(x,y)`$, the correction term has the form
(4.38)
$$\mathrm{\Phi }_1(x)^TG_{11}\mathrm{\Phi }_1(y)+\mathrm{\Phi }_1(x)^TG_{12}\mathrm{\Phi }_2(y).$$
Again set $`x=\xi ^{(N)}`$, $`y=\eta ^{(N)}`$. Using Proposition 4.2 and proceeding in the same way as in Subsection 4.2.1 we find that as $`N\mathrm{}`$, the term (4.38), multiplied as before by $`(\frac{c_N}{\alpha _NN^{2/3}})^2`$, becomes
(4.39)
$$\frac{1}{2}Ai(\xi )Ai(\eta )+o(1)e^{c\xi }e^{c\eta }$$
uniformly for $`\xi ,\eta L_0`$. Note that the sum of all elements of (the binomial matrix in the limiting form of) $`D_{12}`$ is, up to a sign, the same as for $`D_{21}`$.
Remark: Note also that the only new element in the above proof as compared with the case $`\beta =4`$ in Subsection 4.2.1, is that the matrix $`G_{11}`$ is only asymptotically (and not identically) skew symmetric. This leads to the estimate $`o(1)e^{c\xi }e^{c\eta }`$ in place of (4.16).
#### 4.3.2. The contribution of the correction term to the $`11`$ and $`22`$ entries of $`K_{N,1}`$
We consider the $`11`$ entry of $`K_{N,1}`$ (the $`22`$ entry is considered in the same way). Using (1.47), (4.27) we rewrite the correction term as
(4.40) $`\mathrm{\Phi }_1(x)^T`$ $`G_{11}\left({\displaystyle _y^{\mathrm{}}}\mathrm{\Phi }_1(t)𝑑t\right)\mathrm{\Phi }_1(x)^TG_{12}\left({\displaystyle _y^{\mathrm{}}}\mathrm{\Phi }_2(t)𝑑t\right)`$
$`\mathrm{\Phi }_1(x)^TG_{11}ϵ\mathrm{\Phi }_1(+\mathrm{})\mathrm{\Phi }_1(x)^TG_{12}ϵ\mathrm{\Phi }_2(+\mathrm{}).`$
Again set $`x=\xi ^{(N)}`$, $`y=\eta ^{(N)}`$. The first two terms can be treated in the same way as in Subsections 4.2.2 and 4.3.1. More precisely we find that the first two terms in (4.40), multiplied by $`\frac{c_N}{\alpha _NN^{2/3}}`$, become, as $`N\mathrm{}`$
(4.41)
$$\frac{1}{2}Ai(\xi )_\eta ^{\mathrm{}}Ai(t)dt+o(1)e^{c\xi }e^{c\eta }$$
uniformly for $`\xi ,\eta L_0`$. Now consider the (scaled) sum of the last two terms in (4.40)
(4.42)
$$\frac{c_N}{\alpha _NN^{2/3}}\left(\mathrm{\Phi }_1(\xi ^{(N)})^TG_{11}ϵ\mathrm{\Phi }_1(+\mathrm{})+\mathrm{\Phi }_1(\xi ^{(N)})^TG_{12}ϵ\mathrm{\Phi }_2(+\mathrm{})\right).$$
By (4.3), (4.10), Proposition 4.2, this becomes as $`N\mathrm{}`$
(4.43) $`{\displaystyle \frac{c_N}{\alpha _NN^{2/3}}}`$ $`{\displaystyle \frac{\alpha _N^{1/4}N^{1/6}2^{1/4}}{c_N^{1/2}}}{\displaystyle \frac{c_N^{1/2}}{(2m)^{1/2}N^{1/2}}}`$
$`\times \left\{𝐞^TG_{11}(𝐛+o(1))+𝐞^TG_{12}(𝐚+o(1))\right\}Ai(\xi )+o(1)e^{c\xi }.`$
Setting $`𝐚=𝐞𝐛`$, we find that (4.43) reduces to
(4.44) $`{\displaystyle \frac{1}{2}}Ai(\xi )`$ $`+o(1)e^{c\xi }+\left(𝐞^TG_{11}𝐛𝐞^TG_{12}𝐛\right)Ai(\xi )O(N^{1+1/(2m)}).`$
So if we could prove
(4.45)
$$𝐞^TG_{11}𝐛𝐞^TG_{12}𝐛=o(N^{11/(2m)}),N\mathrm{},$$
then we would find that (4.44) equals
(4.46)
$$\frac{1}{2}Ai(\xi )+o(1)e^{c\xi }$$
uniformly for $`\xi L_0`$.
We prove (4.45). We will, perhaps surprisingly, use a property of the $`\beta =4`$ correlation kernel $`S_{N/2,4}`$: it is not clear how to prove (4.45) directly using the asymptotic properties of $`(D\varphi _{N+j},\varphi _{N+k})`$ and $`(ϵ\varphi _{N+j},\varphi _{N+k})`$ given in \[DG\]. More precisely, (4.45) follows from (4.29), (4.30) and the relation
(4.47)
$$𝐛+C_{11}^1B_{11}D_{12}𝐚=o(1),N\mathrm{},$$
which is proved by dividing (1.46) by $`(\frac{c_N}{2mN})^{1/2}`$ and using (4.3) as $`N\mathrm{}`$. Multiplying (4.47) from the left by $`𝐞^TRD_{21}`$ and noting $`𝐛=R𝐛`$, $`𝐚=R𝐚`$, we find
$$𝐞^TRD_{21}R𝐛+𝐞^TRD_{21}C_{11}^1B_{11}D_{12}R(𝐞𝐛)=o(N^{11/(2m)}).$$
But the second matrix is skew symmetric (see (4.28) et seq.). By (4.29) the above relation becomes
(4.48)
$$𝐞^TRD_{21}R𝐛+𝐞^TG_{11}𝐛=o(N^{11/(2m)}).$$
But from (4.36)
$$RD_{21}R=D_{12}+o(N^{11/(2m)})$$
and hence (4.48), (4.30) imply (4.45).
Collecting the estimates (4.41), (4.46) we see that since $`_{\mathrm{}}^{\mathrm{}}Ai(t)dt=1`$, the correction term in the $`11`$ entry has the form
(4.49) $`{\displaystyle \frac{1}{2}}Ai(\xi )`$ $`\left(1{\displaystyle _\eta ^{\mathrm{}}}Ai(t)dt\right)+o(1)e^{c\xi }`$
$`={\displaystyle \frac{1}{2}}Ai(\xi ){\displaystyle _{\mathrm{}}^\eta }Ai(t)dt+o(1)e^{c\xi }`$
uniformly for $`\xi ,\eta L_0`$. The correction term in the $`22`$ entry has the same asymptotic form with $`\xi `$ and $`\eta `$ interchanged.
#### 4.3.3. The contribution of the correction term to the $`21`$ entry of $`K_{N,1}`$
By (1.49), (4.27) the correction term in this case has the form
$$\left(_x^y\mathrm{\Phi }_1(t)^T𝑑t\right)G_{11}ϵ\mathrm{\Phi }_1(y)+\left(_x^y\mathrm{\Phi }_1(t)^T𝑑t\right)G_{12}ϵ\mathrm{\Phi }_2(y)$$
which equals
$`({\displaystyle _x^y}`$ $`\mathrm{\Phi }_1(t)^Tdt)G_{11}({\displaystyle _y^{\mathrm{}}}\mathrm{\Phi }_1(t)dt)`$
$`+\left({\displaystyle _x^y}\mathrm{\Phi }_1(t)^T𝑑t\right)G_{12}\left({\displaystyle _y^{\mathrm{}}}\mathrm{\Phi }_2(t)𝑑t\right)`$
$`+\left({\displaystyle _x^y}\mathrm{\Phi }_1(t)^T𝑑t\right)G_{11}ϵ\mathrm{\Phi }_1(+\mathrm{})+\left({\displaystyle _x^y}\mathrm{\Phi }_1(t)^T𝑑t\right)G_{12}ϵ\mathrm{\Phi }_2(+\mathrm{})`$
by (1.47). Again set $`x=\xi ^{(N)}`$, $`y=\eta ^{(N)}`$. A calculation very similar to the one in Subsection 4.3.2, using the last estimate in (4.10) in place of the first, leads to the following asymptotic form for the $`21`$ correction as $`N\mathrm{}`$
(4.50)
$$\frac{1}{2}_\xi ^\eta Ai(s)ds+\frac{1}{2}\left(_\xi ^\eta Ai(s)ds\right)\left(_\eta ^{\mathrm{}}Ai(t)dt\right)+o(1)e^{c\mathrm{min}(\xi ,\eta )}$$
uniformly for $`\xi ,\eta L_0`$. (Recall that there is no overall scaling factor for the $`21`$ entry.) This completes the analysis of the contribution of the correction term to the $`21`$ entry of $`K_{N,1}`$. |
warning/0507/hep-ph0507155.html | ar5iv | text | # 1 Introduction
## 1 Introduction
The $`\text{SU}(2)_L\text{U}(1)_Y`$ symmetry of the standard model (SM) is a partial unification of the weak and electromagnetic interactions. It leaves many striking features of the physics of our world unexplained. Some of these are the quantization of electric charge (ECQ) and the generation number problem (GNP). The hydrogen atom is known to be electrically neutral to extraordinary accuracy. This implies that there is a relation between the charges of the quarks and of the electron. However, the electric charge operator in the SM owned with the form $`Q=T_3+Y`$, while it can describe the observed charges, does not explain them. The problem is the $`Y`$ generator. The values of $`T_3`$ are quantized because of the non-Abelian nature of the $`\text{SU}(2)`$ algebra. However, the values of Y are completely arbitrary . They are chosen to describe the discrete charges. However, as in the grand unified theory (GUT) , both $`T_3`$ and $`Y`$ are embedded into the $`\text{SU}(5)`$ simple group, thus the values of $`Y`$, like those of $`T_3`$ are constrained by the structure of the algebra, hence the ECQ has been derived. However, like the SM, the GUT also cannot explain the GNP; moreover, this simplest version of the GUT is fairly convincingly ruled out by the experiments on proton decay.
A very interesting alternative to explain the origin of the generations comes from the cancelation of chiral anomalies . In particular, the models based on the $`\text{G}_{331}=\text{SU}(3)_C\text{SU}(3)_L\text{U}(1)_X`$ gauge group, also called 3-3-1 models , arise as a possible solution to this puzzle, since some of such models require three generations in order to cancel completely chiral anomalies. In addition, in the literature on the ECQ in some 3-3-1 models, some solutions have been explored ; thus, it is hoped that the ECQ will be solved. However, the status is still opened with many problems which are not solved or not cleared, namely, the ECQ is obviously not dependent on the condition to generate mass for leptons and quarks. The usual theoretical grounds are completely unrestricted such as the definition of the fermion content and the electric charge operator, the relation between the ECQ and the GNP. They are still kept as the open questions!
In this paper, we will prove that the ECQ exists in the minimal 3-3-1 model and in the 3-3-1 model with right-handed (RH) neutrinos. We see that alternative to the GUT in which the problem is solved on the algebra structure of the simple group; here, it is a direct consequence from the usual fermion content in those models. We argue that the solution for the ECQ should be based on general laws, such as the conservation of the electric charge, the parity invariance of the electromagnetic interaction and the anomaly cancelation.
The rest of this paper is organized as follows: In Sec.2 a brief review of the 3-3-1 models is presented. It is emphasized that the photon eigenstate is dependent only on form of the electric charge operator from which the ECQ is derived. Next, in Sec.3, the fermion content, the electric charge operator and the classification of the models are represented. In Sec.4, from parity invariance of the electromagnetic vertices and anomaly cancelation, the ECQ is obtained. Our conclusions are summarized in the last section - Sec.5.
## 2 Some remarks on the gauge sector
To proceed further, in this section we review some essential consequences on the gauge sector of any 3-3-1 model which has been verified in . Basing on these and two general properties of the electromagnetic interaction-the conservation of the electric charge and the parity invariance, we get equations for the usual $`\text{U}(1)_X`$ charges. Then, the ECQ in the 3-3-1 models is derived.
Suppose that, under the $`\text{G}_{331}`$ symmetry, there are a fermion triplet $`\mathrm{𝟑}=(f_u,f_d,f_s)_L^T`$ which is composed of a doublet $`(f_u,f_d)_L^T`$ and a singlet $`f_{sL}`$ of the $`\text{SU}(2)_L`$ group of the SM, and an electric charge operator (eco) in this basic owning the form
$$Q=T_3+\beta T_8+X.$$
(2.1)
To beak symmetry spontaneously, in general, three Higgs triplets are introduced
$$\chi (1,3,X_\chi ),\eta (1,3,X_\eta ),\rho (1,3,X_\rho ),$$
(2.2)
which must acquire the vacuum expectation values (VEVs) as follows
$`\chi ^T`$ $`=`$ $`(0,0,{\displaystyle \frac{v_s}{\sqrt{2}}}),`$
$`\eta ^T`$ $`=`$ $`({\displaystyle \frac{v_u}{\sqrt{2}}},0,0),`$ (2.3)
$`\rho ^T`$ $`=`$ $`(0,{\displaystyle \frac{v_d}{\sqrt{2}}},0).`$
The $`\text{G}_{331}`$ group is decomposed into the gauge group of the SM by the Higgs triplet $`\chi `$. Next, the gauge group of the SM is decomposed into the $`\text{SU}(3)_C\text{U}(1)_Q`$ by the two remaining Higgs triplets $`\eta ,\rho `$. To keep the conservation of the electric charge, the operator Q must annihilate the vacuums: $`Q\chi =0`$, $`Q\rho =0`$ and $`Q\eta =0`$, then we get
$`X_\eta `$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \frac{\beta }{2\sqrt{3}}},`$
$`X_\rho `$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \frac{\beta }{2\sqrt{3}}},`$ (2.4)
$`X_\chi `$ $`=`$ $`{\displaystyle \frac{\beta }{\sqrt{3}}},`$
which are the fixing conditions for the $`\text{U}(1)_X`$ charges of the Higgs scalars. They yield
$$X_\eta +X_\rho +X_\chi =0.$$
(2.5)
The mass Lagrangian for the neutral gauge bosons is given by
$$_{mass}=\frac{1}{2}V^TM^2V,$$
(2.6)
where $`V^T=(W^3,W^8,B)`$ and
$$M^2=\frac{1}{4}g^2\left(\begin{array}{ccc}m_{11}& m_{12}& m_{13}\\ m_{12}& m_{22}& m_{23}\\ m_{13}& m_{23}& m_{33}\end{array}\right),$$
(2.7)
with
$`m_{11}`$ $`=`$ $`v_u^2+v_d^2,`$
$`m_{12}`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{3}}}\left(v_u^2v_d^2\right),`$
$`m_{13}`$ $`=`$ $`{\displaystyle \frac{t}{\sqrt{6}}}\left[v_u^2\left(1{\displaystyle \frac{\beta }{\sqrt{3}}}\right)v_d^2\left(1{\displaystyle \frac{\beta }{\sqrt{3}}}\right)\right],`$
$`m_{22}`$ $`=`$ $`{\displaystyle \frac{1}{3}}\left(v_u^2+v_d^2+4v_s^2\right),`$
$`m_{23}`$ $`=`$ $`{\displaystyle \frac{t}{3\sqrt{2}}}\left[v_u^2\left(1{\displaystyle \frac{\beta }{\sqrt{3}}}\right)+v_d^2\left(1{\displaystyle \frac{\beta }{\sqrt{3}}}\right)v_s^2{\displaystyle \frac{4\beta }{\sqrt{3}}}\right],`$
$`m_{33}`$ $`=`$ $`{\displaystyle \frac{t^2}{6}}\left[v_u^2\left(1{\displaystyle \frac{\beta }{\sqrt{3}}}\right)^2+v_d^2\left(1{\displaystyle \frac{\beta }{\sqrt{3}}}\right)^2+v_s^2\left({\displaystyle \frac{2\beta }{\sqrt{3}}}\right)^2\right].`$
Here $`tg_X/g`$ with $`g_X,`$ $`g`$ are the gauge coupling constants of the $`\text{U}(1)_X`$ and $`\text{SU}(3)_L`$ groups, respectively. We have shown , for any 3-3-1 model (containing Higgs triplets, antitriplets as well as sextets or any necessary Higgs scalar), the mass matrix of the neutral gauge bosons always has the above form. In addition, if additional Higgs scalars have non-zero VEVs, one just makes the following appropriate replaces
$`v_u^2`$ $``$ $`v_u^2+v_{u1}^2+v_{u2}^2+\mathrm{},`$
$`v_d^2`$ $``$ $`v_d^2+v_{d1}^2+v_{d2}^2+\mathrm{},`$
$`v_s^2`$ $``$ $`v_s^2+v_{s1}^2+v_{s2}^2+\mathrm{},`$
where $`v_{ui},v_{dj},v_{sk}`$ are the VEVs of the neutral members in the additional Higgs, respectively. Thus, this is the general form of the mass matrix for the neutral gauge boson sector.
It can be checked that the matrix $`M^2`$ has a non-degenerate zero eigenvalue. Therefore, the zero eigenvalue is identified with the photon mass, $`M_\gamma ^2=0`$. The physical photon field $`A_\mu `$ is directly defined from the equation $`M^2A_\mu =0`$:
$$A_\mu =\frac{t}{\sqrt{6+(1+\beta ^2)t^2}}W_\mu ^3+\frac{\beta t}{\sqrt{6+(1+\beta ^2)t^2}}W_\mu ^8+\frac{\sqrt{6}}{\sqrt{6+(1+\beta ^2)t^2}}B_\mu .$$
(2.8)
Hence, for any 3-3-1 model, the photon eigenstate and mass are independent on the VEVs structure. These are a natural consequence of the $`\text{U}(1)_Q`$ invariance - the conservation of the electric charge. Moreover, to be consistent with the QED based on the unbroken $`\text{U}(1)_Q`$ gauge group, the photon field has to keep the general properties of the electromagnetic interaction in the framework of the 3-3-1 model, such as the parity invariant nature (for more discussions, see ). These would help us to obtain some consequences related to quantities which are independent on VEVs structure such as the matching of gauge coupling constants <sup>3</sup><sup>3</sup>3As a result, the condition for matching of the gauge coupling constants in any 3-3-1 model is very natural as done in the SM. and the ECQ.
Next, using (2.8) we write the coupling of the up member with the photon, $`\overline{f}_uf_u\gamma `$ (also see ):
$`_{\overline{f}_uf_u\gamma }^{em}`$ $`=`$ $`\overline{f}_{uL}i\gamma ^\mu [{\displaystyle \frac{ig}{2}}{\displaystyle \frac{t}{\sqrt{6+(1+\beta ^2)t^2}}}+{\displaystyle \frac{ig}{2\sqrt{3}}}{\displaystyle \frac{\beta t}{\sqrt{6+(1+\beta ^2)t^2}}}`$
$`+`$ $`{\displaystyle \frac{ig_X}{\sqrt{6}}}X_3{\displaystyle \frac{\sqrt{6}}{\sqrt{6+(1+\beta ^2)t^2}}}]A_\mu f_{uL}`$
$`+`$ $`\overline{f}_{uR}i\gamma ^\mu \left[{\displaystyle \frac{ig_X}{\sqrt{6}}}X_{f_u}{\displaystyle \frac{\sqrt{6}}{\sqrt{6+(1+\beta ^2)t^2}}}\right]A_\mu f_{uR}`$
$`=`$ $`{\displaystyle \frac{(X_3X_\eta )g_X}{\sqrt{6+(1+\beta ^2)t^2}}}\overline{f}_{uL}\gamma ^\mu f_{uL}A_\mu {\displaystyle \frac{X_{f_u}g_X}{\sqrt{6+(1+\beta ^2)t^2}}}\overline{f}_{uR}\gamma ^\mu f_{uR}A_\mu ,`$
where $`X_3`$ and $`X_{f_u}`$ are the $`\text{U}(1)_X`$ charges of the 3 triplet and of the $`f_{uR}`$ singlet, respectively. Since the electromagnetic interaction is invariant under the parity transformation , then we get
$$X_{f_u}=X_3X_\eta .$$
(2.9)
Similarly for the vertices $`\overline{f}_df_d\gamma `$ and $`\overline{f}_sf_s\gamma `$, we get
$`X_{f_d}`$ $`=`$ $`X_3X_\rho ,`$ (2.10)
$`X_{f_s}`$ $`=`$ $`X_3X_\chi ,`$ (2.11)
where $`X_{f_d}`$, $`X_{f_s}`$ are the $`\text{U}(1)_X`$ charges of the $`\text{SU}(3)_L`$ singlets $`f_{dR}`$ and $`f_{sR}`$, respectively. Note that $`f_{uR}`$, $`f_{dR}`$ and $`f_{sR}`$ are the right-handed counterparts of the triplet $`(f_u,f_d,f_s)_L^T`$.
For a fermion antitriplet $`\mathrm{𝟑}^{}=(f_d^{},f_u^{},f_s^{})_L^T`$ which is composed of an antidoublet $`(f_d^{},f_u^{})_L^T`$ and a singlet $`f_{sL}^{}`$ of the $`\text{SU}(2)_L`$ group of the SM with $`f_{uR}^{}`$, $`f_{dR}^{}`$ and $`f_{sR}^{}`$ are its right-handed counterparts, we also have
$`X_{f_d^{}}`$ $`=`$ $`X_3^{}+X_\eta ,`$ (2.12)
$`X_{f_u^{}}`$ $`=`$ $`X_3^{}+X_\rho ,`$ (2.13)
$`X_{f_s^{}}`$ $`=`$ $`X_3^{}+X_\chi .`$ (2.14)
Here $`X_3^{}`$, $`X_{f_u^{}}`$, $`X_{f_d^{}}`$ and $`X_{f_s^{}}`$ stand for the $`\text{U}(1)_X`$ charges of the antitriplet and singlets $`f_{uR}^{}`$, $`f_{dR}^{}`$ and $`f_{sR}^{}`$, respectively.
As we will see, the lepton sector owns the triplets either $`\mathrm{𝟑}_𝐥=(\nu _l,l,\nu _l^C)_L^T`$ or $`\mathrm{𝟑}_𝐥=(\nu _l,l,l^C)_L^T`$, where $`\nu _{lL}^C=(\nu _{lR})^C`$, $`l_L^C=(l_R)^C`$ with $`C`$ is the conjugate operator and $`l`$ stands for the lepton $`e`$, $`\mu `$, $`\tau `$. Then, the two equations either (2.9) and (2.11) or (2.10) and (2.11) are replaced by equation one either
$$X_{3_l}=\frac{X_\eta +X_\chi }{2},$$
(2.15)
or
$$X_{3_l}=\frac{X_\rho +X_\chi }{2},l=e,\mu ,\tau ,$$
(2.16)
respectively, which is directly obtained from the vertex $`\overline{\nu _l}\nu _l\gamma `$ or $`\overline{l}l\gamma `$. It is worth to mention on significance of the equation (2.15) or (2.16) which is the fixing condition for the $`\text{U}(1)_X`$ charge of the lepton triplet as a natural consequence of the lepton content. Further, with anomaly cancelation, the charges for all remaining chiral fermions are also fixed. Hence, these give the constraints on the hypercharge values $`Y=\beta T_8+X`$ from which the ECQ in the 3-3-1 models will explicitly be explained. Otherwise, if (2.15), (2.16) do not exist in some lepton content, there is not the ECQ unless add auxiliary conditions such as on Majorana neutrino mass, non-RH neutrino singlets, etc. Thus, this means that (2.15) or (2.16) is quantized condition.
It is to be emphasized that if the ECQ in the GUT has been found by the mean of the algebra structure of the simple group, here for the 3-3-1 models, by their fermion content (the fermion structure under the anomaly free conditions).
## 3 Fermion content
In the framework of the 3-3-1 models, the essential basic concepts for building the models such as the fermion representations (reps), the electric charge operator, the anomaly cancelation and the fermion content will be explained. However, for our purpose in studying the ECQ, it is necessary to note that the electric charges of the particles will be kept as parameters.
The SM is well done with the $`\text{SU}(2)_L`$ doublets for the left-handed chiral spinors and the $`\text{SU}(2)_L`$ singlets for the right-handed chiral spinors. Each generation of the SM consists of
the doublets: $`(\nu _l,l)_L^T,\text{ }(u^\alpha ,d^\alpha )_L^T`$
and the singlets: $`\nu _{lR},\text{ }l_R,\text{ }u_R^\alpha ,\text{ }d_R^\alpha ,`$ (3.1)
where $`L,R`$ stand for the left-handed and the right-handed counterparts, respectively, $`\alpha `$ is the color index. Here $`l=e,\text{ }\mu ,\mathrm{}`$; $`\nu _l=\nu _e,\text{ }\nu _\mu ,\mathrm{}`$; $`u=u,\text{ }c,\mathrm{}`$ and $`d=d,\text{ }s,\mathrm{}`$ are the lepton and the quark particles in each generation, respectively.
Under the gauge symmetry $`\text{G}_{331}`$, the fermions transform like triplets 3, antitriplets 3 or singlets 1 of the $`\text{SU}(3)_L`$ group. Requiring the models at low energy to be fitly with the SM, the $`\text{G}_{331}`$ symmetry must be spontaneously broken down that of the SM. Thus, the triplets or antitriplets are composed of the doublets 2 or antitriplets 2 and singlets 1 of the $`\text{SU}(2)_L`$ group of the SM. The decomposional rule into the SM for the triplets yields
$`(\nu _l,l,S^l)_L^T`$ $`=`$ $`(\nu _l,l)_L^TS_L^l,`$ (3.2)
$`(u,d,S^q)_L^T`$ $`=`$ $`(u,d)_L^TS_L^q.`$ (3.3)
Similarly for the antitriplets
$`(l,\nu _l,S^l)_L^T`$ $`=`$ $`(l,\nu _l)_L^TS_L^l,`$ (3.4)
$`(d,u,S^q)_L^T`$ $`=`$ $`(d,u)_L^TS_L^q,`$ (3.5)
where $`S^l`$, $`S^l`$, $`S^q`$, $`S^q`$ stand for the lepton and quark singlets, respectively. Note that if $`(f_u,f_d)_L^T`$ is a doublet of the $`\text{SU}(2)_L`$, then $`(f_d,f_u)_L^T`$ is its antidoublet, they are equivalent and real reps.
Since the right-handed leptons in (3.1) are color singlets, they are put in the singlet $`S^l`$ or $`S^l`$ by two ways :
$`(\nu _{lR})^C=S_L^l,\text{ or }S_L^l`$ (3.6)
and
$`(l_R)^C=S_L^l,\text{ or }S_L^l.`$ (3.7)
However, under the $`\text{G}_{331}`$ and the Lorentz invariance, we cannot put the left-handed antiquarks in the bottom (singlet) of the triplets, so the existence of the exotic quarks is not able to avoid in all 3-3-1 models. In addition, the exotic leptons can be also in the singlets (in bottom of the lepton triplets or antitriplets), however, they are not considered here.
The fermion content under the $`\text{G}_{331}`$ symmetry must be satisfied with the following criteria:
1. All singlets of the lepton triplets and antitriplets are either $`(\nu _{lR})^C`$ or $`(l_R)^C`$.
2. Both $`(\nu _l,l,\nu _l^C)_L^T`$, $`(l^{},\nu _l^{},\nu _l^{}^C)_L^T`$ and $`(\nu _l,l,l^C)_L^T`$, $`(l^{},\nu _l^{},l^C)_L^T`$ are not conjugate pairs, namely triplet and antitriplet.
Without loss of generality due to the second criterion, we can put the left-handed leptons in the triplets. Hence, on the first criterion there are two models: with $`(l_R)^C`$ called the minimal 3-3-1 model , and also for $`(\nu _{lR})^C`$ called the 3-3-1 model with RH neutrinos . In this paper, they are called usual 3-3-1 models.
Due to the conservation and additive nature of the electric charge, the eco must be embedded in the neutral generators of the $`\text{SU}(3)_L\text{U}(1)_X`$ group:
$$Q=\alpha T_3+\beta T_8+\gamma X.$$
(3.8)
Here the $`\text{SU}(3)_L`$ charges $`T_3=\lambda _3/2`$, $`T_8=\lambda _8/2`$ with $`\lambda _3`$, $`\lambda _8`$ are the two diagonal Gell-Mann matrices, and $`X`$ is the $`\text{U}(1)_X`$ charge. Without loss of generality, the $`\gamma `$ coefficient can be normalized to $`1`$ due to a scaling symmetry, $`g_X\gamma g_X`$, $`XX/\gamma `$, where $`g_X`$ is the $`\text{U}(1)_X`$ coupling constant . Finally, the two remaining coefficients $`\alpha `$ and $`\beta `$ get the same dimension of the electric charge. At the breaking point, the $`\text{SU}(3)_L`$ group is embedded properly in the $`\text{SU}(2)_L`$ group of the SM, therefore, the gauge boson $`W`$ takes an electric charged value equal $`+\alpha `$ or $`\alpha `$. To see this, we should apply the eco (3.8) on a $`\text{SU}(3)_L`$ triplet, $`\mathrm{𝟑}=(f_u,f_d,f_s)_L^T`$,
$`{\displaystyle \frac{1}{2}}\alpha +{\displaystyle \frac{1}{2\sqrt{3}}}\beta +\gamma X_\mathrm{𝟑}`$ $`=`$ $`q_{f_u},`$
$`{\displaystyle \frac{1}{2}}\alpha +{\displaystyle \frac{1}{2\sqrt{3}}}\beta +\gamma X_\mathrm{𝟑}`$ $`=`$ $`q_{f_d},`$ (3.9)
$`{\displaystyle \frac{1}{\sqrt{3}}}\beta +\gamma X_\mathrm{𝟑}`$ $`=`$ $`q_{f_s},`$
with $`q_{f_u}`$, $`q_{f_d}`$ and $`q_{f_s}`$ are the electric charges of the members. Hence, the electric charge of W is $`q_{f_u}q_{f_d}=\alpha `$. The normalization of the eco is undetermined, however we can always use the freedom in assigning the scale of the electric charge by putting the charged $`W`$ in unit, $`\alpha =1`$ . The eco is given by
$$Q=T_3+\beta T_8+X.$$
(3.10)
When the $`Y`$ hypercharge is embedded into the $`\text{SU}(3)_L\text{U}(1)_X`$ group, it is a linear combination of two terms, $`Y=\beta T_8+X`$. The first term in the $`\text{SU}(3)_L`$ is constrained by the structure of the algebra therefore quantized ; and, the second term in the $`\text{U}(1)_X`$ with the values are kept as undetermined parameters. This differs the GUT from enlarging to the simple group, hence the ECQ is a direct consequence. However, as in the previous section, since all X charges are fixed, the present ECQ signs a different structure which refers to the particle reps.
Noting that the presence of the coefficient $`\beta `$ signifies that the first term of the $`Y`$ hypercharge is not properly normalized to be one of the $`\text{SU}(3)_L`$ generators which have their scale fixed by the non-linear commutation relations ,
$`[T_a,T_b]=if_{abc}T_c,`$
with
$`\text{T}r[T_aT_b]=\delta _{ab}/2.`$
The value of $`\beta `$ is obtained by comparing in the fundamental rep the values of $`T_8`$ and the hypercharge values of the particles in some multiplet.
The action of the eco on an antitriplet $`\mathrm{𝟑}^{}=(f_d^{},f_u^{},f_s^{})^T`$ is thanked to the usual rule
$`Q\mathrm{𝟑}=q_3\mathrm{𝟑},`$ (3.11)
which yields
$$Q\mathrm{𝟑}^{}=q_3\mathrm{𝟑}^{}=q_3^{}\mathrm{𝟑}^{}.$$
(3.12)
Noting on the minus sign in the r.h.s of (3.12), we get $`X_3^{}=X_3=\text{Tr}Q`$.
Demanding for the fermion $`\text{SU}(3)_C`$ reps to be vector-like and the color number $`N_C=3`$, we get the non-trivial triangular anomaly cancelation conditions as follows
$`\left[\text{SU}(3)_C\right]^2\text{U}(1)_X`$ $`:`$ $`3X_q^L{\displaystyle \underset{singlet}{}}X_q^R=0,`$ (3.13)
$`\left[\text{SU}(3)_L\right]^3`$ $`:`$ $`{\displaystyle \frac{1}{2}}A_{\alpha \beta \gamma }=0,`$ (3.14)
$`\left[\text{SU}(3)_L\right]^2\text{U}(1)_X`$ $`:`$ $`{\displaystyle \underset{family}{}}X_l^L+3{\displaystyle \underset{family}{}}X_q^L=0,`$ (3.15)
$`\left[Grav\right]^2\text{U}(1)_X`$ $`:`$ $`3{\displaystyle \underset{family}{}}X_l^L+9{\displaystyle \underset{family}{}}X_q^L`$ (3.16)
$``$ $`3{\displaystyle \underset{family}{}}{\displaystyle \underset{singlet}{}}X_q^R{\displaystyle \underset{family}{}}{\displaystyle \underset{singlet}{}}X_l^R=0,`$
$`\left[\text{U}(1)_X\right]^3`$ $`:`$ $`3{\displaystyle \underset{family}{}}(X_l^L)^3+9{\displaystyle \underset{family}{}}(X_q^L)^3`$ (3.17)
$``$ $`3{\displaystyle \underset{family}{}}{\displaystyle \underset{singlet}{}}(X_q^R)^3{\displaystyle \underset{family}{}}{\displaystyle \underset{singlet}{}}(X_l^R)^3=0.`$
Here $`X_l^L`$, $`X_q^L`$, $`X_l^R`$ and $`X_q^R`$ refer to the $`\text{U}(1)_X`$ charges of the left-handed lepton, quark triplets or antitriplets and the right-handed lepton, quark singlets, respectively.
The cancelation of the $`[\text{SU}(3)_L]^3`$ anomaly (3.14) demands for the number of fermion triplets to be the same as that of antitriplets. As mentioned above the $`\text{SU}(2)_L`$ doublet and antidoublet are equivalent and real, hence, all the left-handed leptons and quarks are always ordered in the doublets. Moreover, using two conditions such as some known fermion generations are completely free from anomaly and the anomaly over all the quark and lepton generations must be canceled, we deduce that the number of the quark generations must be equal to that of the leptons. So, if $`N_f`$ is the number of the fermion generations; thus, it is also the number of the lepton triplets as mentioned above. And, k is the number of the quark generations which are ordered in the triplets; then, there are the remaining $`N_fk`$ quark generations therefore in the antitriplets, satisfying
$`N_f+3k=3(N_fk)`$ $``$ $`N_f=3k.`$ (3.18)
Hence, the generation number $`N_f`$ is a multiple of three. If further, one adds the condition of the QCD asymptotic freedom, which is valid only when the quark generation number is to be less than five, then it follows that $`N_f`$ is equal to $`3`$, and hence $`k=1`$.
Therefore, the classification of the 3-3-1 models is given as follows,
1. The 3-3-1 model with RH neutrinos which the fermion reps are ordered by
$`(\nu _l,l_,\nu _l^C)_L^T`$ $``$ $`(1,3,X_l^L),l=e,\mu ,\tau ,`$ (3.19)
$`l_R`$ $``$ $`(1,1,X_l^R),`$ (3.20)
$`(u_{3L},d_{3L},s_{3L})`$ $``$ $`(3,3,X_{q_3}^L),`$ (3.21)
$`(u_{iL},d_{iL},s_{iL})`$ $``$ $`(3,3^{},X_{q_i}^L),i=1,2,`$ (3.22)
$`u_{iR},u_{3R}`$ $``$ $`(3,1,X_{u_i}^R),(3,1,X_{u_3}^R),\text{ also for }d\text{ and }s.`$ (3.23)
2. The minimal 3-3-1 model with the fermion reps read
$`(\nu _l,l,l^C)_L^T`$ $``$ $`(1,3,X_l^L),l=e,\mu ,\tau ,`$ (3.24)
$`\nu _{lR}`$ $``$ $`(1,1,X_{\nu _l}^R),`$ (3.25)
$`(u_{3L},d_{3L},s_{3L})`$ $``$ $`(3,3,X_{q_3}^L),`$ (3.26)
$`(u_{iL},d_{iL},s_{iL})`$ $``$ $`(3,3^{},X_{q_i}^L),i=1,2,`$ (3.27)
$`u_{iR},u_{3R}`$ $``$ $`(3,1,X_{u_i}^R),(3,1,X_{u_3}^R),\text{ also for }d\text{ and }s.`$ (3.28)
Here, the (c,f,X) denotes the respective quantum numbers to the color, the flavor and the X-charge, and $`s_a`$, $`a=1,2,3`$ are the added exotic quarks.
## 4 The ECQ
Now we turn on the ECQ in the 3-3-1 models. We first deal with the minimal version.
### 4.1 The ECQ in the minimal model
It is known that, the electromagnetic interaction is invariant under parity transformation. Using this property and anomaly cancelation we will get the needed ECQ. Let us deal with the lepton sector.
#### 4.1.1 The ECQ in the lepton sector
For the minimal model with the given lepton triplets, using Eq. (2.16), we get
$`X_l^L`$ $`=`$ $`{\displaystyle \frac{X_\rho +X_\chi }{2}}`$ (4.1)
$`=`$ $`{\displaystyle \frac{X_\eta }{2}},\text{ }l=e,\mu ,\tau .`$
Applying Eq. (2.9) for the neutrinos, we have
$$X_{\nu _l}^R=X_l^LX_\eta =\frac{3}{2}X_\eta ,\text{ }l=e,\mu ,\tau .$$
(4.2)
Therefore, the application of the eco on the lepton triplets and the neutrino singlets yields the electric charges for the leptons as follows
$`q_{\nu _l}`$ $`=`$ $`{\displaystyle \frac{3+\sqrt{3}\beta }{4}},`$ (4.3)
$`q_l`$ $`=`$ $`{\displaystyle \frac{1+\sqrt{3}\beta }{4}},\text{ }l=e,\mu ,\tau .`$ (4.4)
So, with the help of the parity invariance of the electromagnetic vertices for all leptons, the ECQ of the lepton sector is derived. The electric charges of all leptons in the model are defined in terms of the $`\beta `$.
#### 4.1.2 The ECQ in the quark sector
Applying the equations (2.9), (2.10) and (2.11) for the quark triplets; and, (2.12), (2.13) and (2.14) for the quark antitriplets, we get
$`X_{u_3}^R`$ $`=`$ $`X_{q_3}^LX_\eta ,\text{ }X_{u_i}^R=X_{q_i}^L+X_\rho ,`$ (4.5)
$`X_{d_3}^R`$ $`=`$ $`X_{q_3}^LX_\rho ,\text{ }X_{d_i}^R=X_{q_i}^L+X_\eta ,`$ (4.6)
$`X_{s_3}^R`$ $`=`$ $`X_{q_3}^LX_\chi ,\text{ }X_{s_i}^R=X_{q_i}^L+X_\chi .`$ (4.7)
With these equations, we can write all Yukawa couplings to generate mass to all quarks
$`_Y`$ $`=`$ $`h_{33}^s\overline{q}_3s_3\chi +h_{ii}^s\overline{q_i}s_i\chi ^{}`$ (4.8)
$`+`$ $`h_{33}^u\overline{q}_3u_3\eta +h_{ii}^u\overline{q_i}u_i\rho ^{}`$
$`+`$ $`h_{33}^d\overline{q}_3d_3\rho +h_{ii}^d\overline{q_i}d_i\eta ^{}+h.c..`$
Since the CKM matrix is non-diagonal, there are the flavor mixing terms in the Lagrangian (4.8). Therefore, the some terms in (4.8) must be changed as follows
$`\overline{q}_3u_3\eta `$ $``$ $`\overline{q}_3u_a\eta ,\text{ }a=1,\text{ }2,\text{ }3,`$
$`\overline{q}_3d_3\rho `$ $``$ $`\overline{q}_3d_a\rho ,`$
$`\overline{q_i}u_i\rho ^{}`$ $``$ $`\overline{q_i}u_a\rho ^{},\text{ }i=1,\text{ }2,`$
$`\overline{q_i}d_i\eta ^{}`$ $``$ $`\overline{q_i}d_a\eta ^{}.`$ (4.9)
Under the $`\text{U}(1)_Q`$ invariance, we have
$`X_{u_1}^R`$ $`=`$ $`X_{u_2}^R=X_{u_3}^RX_u^R,`$
$`X_{d_1}^R`$ $`=`$ $`X_{d_2}^R=X_{d_3}^RX_d^R.`$ (4.10)
Thus, it is easy to get
$`X_{q_1}^L`$ $`=`$ $`X_{q_2}^LX_q^L,`$
$`X_{s_1}^R`$ $`=`$ $`X_{s_2}^RX_s^R.`$
Using the anomaly cancelation (3.15), we have
$`2X_q^L+X_{q_3}^L`$ $`=`$ $`{\displaystyle \frac{1}{2}}X_\eta .`$ (4.11)
Combination of (4.6) and (4.10) yields
$$X_q^L+X_\eta =X_{q_3}^LX_\rho .$$
(4.12)
From (4.11) and (4.12) it follows
$`X_q^L`$ $`=`$ $`{\displaystyle \frac{1}{6}}(X_\eta +2X_\chi ),`$ (4.13)
$`X_{q_3}^L`$ $`=`$ $`{\displaystyle \frac{1}{6}}(X_\eta 4X_\chi ).`$ (4.14)
With the help of (4.13) and (4.14), we can express all charges for all right-handed quark counterparts in terms of $`X_\eta `$ and $`X_\chi `$, namely
$`X_u^R`$ $`=`$ $`{\displaystyle \frac{1}{6}}(5X_\eta +4X_\chi ),`$ (4.15)
$`X_d^R`$ $`=`$ $`{\displaystyle \frac{1}{6}}(7X_\eta 4X_\chi ),`$ (4.16)
$`X_s^R`$ $`=`$ $`{\displaystyle \frac{1}{6}}(X_\eta +8X_\chi ),`$ (4.17)
$`X_{s_3}^R`$ $`=`$ $`{\displaystyle \frac{1}{6}}(X_\eta 10X_\chi ).`$ (4.18)
The equations (4.13)-(4.18) are the fixing conditions for the X-charges of the quark reps, therefore the charges of all fermion reps are indeed fixed.
Knowing the $`X`$-charges of the multiplets, we get the electric charges of their members as follows
$`q_u`$ $`=`$ $`{\displaystyle \frac{5\sqrt{3}\beta }{12}},\text{ }u=u,\text{ }c,\text{ }t,`$ (4.19)
$`q_d`$ $`=`$ $`{\displaystyle \frac{7+\sqrt{3}\beta }{12}},\text{ }d=d,\text{ }s,\text{ }b,`$ (4.20)
$`q_{s_3}`$ $`=`$ $`{\displaystyle \frac{1+7\sqrt{3}\beta }{12}},`$ (4.21)
$`q_s`$ $`=`$ $`{\displaystyle \frac{15\sqrt{3}\beta }{12}},\text{ }s=s_1,\text{ }s_2.`$ (4.22)
In addition, it is easy to check that all the remaining anomaly cancelation conditions (3.13), (3.16) and (3.17) are satisfied. So, the ECQ of the quark sector is also given with the help of the anomaly cancelation.
We have the following remarks:
1. We can check the electric charge of the proton composed of three quarks $`uud`$
$`q_p`$ $`=`$ $`2q_u+q_d`$ (4.23)
$`=`$ $`{\displaystyle \frac{1\sqrt{3}\beta }{4}},`$
which yields
$$q_p=q_e.$$
(4.24)
2. The neutron is composed of three quarks $`ddu`$; therefore, its electric charge is given by
$`q_n`$ $`=`$ $`q_u+2q_d`$ (4.25)
$`=`$ $`{\displaystyle \frac{3+\sqrt{3}\beta }{4}},`$
which yields the following interesting consequence
$$q_n=q_\nu .$$
(4.26)
As mentioned above, the coefficient $`\beta `$ should be fixed from the known hypercharge values in the SM . Hence, for a lepton triplet, we have
$`Y(3_l)`$ $`=`$ $`({\displaystyle \frac{1}{2}},{\displaystyle \frac{1}{2}},+1)^T=\beta T_8+X_l^L.`$ (4.27)
From (4.1) and (4.27), it follows that $`\beta =\sqrt{3}`$. Then the electric charges get the correct values as follows
$`q_{\nu _e}`$ $`=`$ $`0,\text{ }e=e,\text{ }\mu ,\text{ }\tau ,`$ (4.28)
$`q_e`$ $`=`$ $`1,\text{ }e=e,\text{ }\mu ,\text{ }\tau ,`$ (4.29)
$`q_u`$ $`=`$ $`+{\displaystyle \frac{2}{3}},\text{ }u=u,\text{ }c,\text{ }t,`$ (4.30)
$`q_d`$ $`=`$ $`{\displaystyle \frac{1}{3}},\text{ }d=d,\text{ }s,\text{ }b,`$ (4.31)
$`q_{s_3}`$ $`=`$ $`+{\displaystyle \frac{5}{3}},`$ (4.32)
$`q_s`$ $`=`$ $`{\displaystyle \frac{4}{3}},\text{ }s=s_1,\text{ }s_2.`$ (4.33)
These relations have also been found in the literature , but they are based on the two principal conditions such as the classical constraints (to generate mass for the all fermions) and the anomaly cancelation.
### 4.2 The ECQ in the 3-3-1 model with RH neutrinos
For the 3-3-1 model with RH neutrinos, the $`\beta `$ takes a value of $`\frac{1}{\sqrt{3}}`$. Therefore, the exotic quarks get the electric charges different from those in the minimal model as follows
$`q_{s_3}`$ $`=`$ $`+{\displaystyle \frac{2}{3}},`$ (4.34)
$`q_s`$ $`=`$ $`{\displaystyle \frac{1}{3}},\text{ }s=s_1,\text{ }s_2.`$ (4.35)
This means that this model does not contain the exotic charges.
The electric charges of the usual leptons and quarks are the same as in the minimal model.
Thus, basing on the parity invariance of the electromagnetic interaction and the anomaly cancelation, we have shown that the usual 3-3-1 models contain in their framework the quantization of the electric charge.
## 5 Conclusions
Analyzing the photon eigenstate structure, we have shown that the general properties of the electromagnetic interaction such as the parity invariance is properly kept in the framework of the 3-3-1 models. This is a natural consequence of the conservation of the electric charge. As a result, the electric charge quantization in the usual 3-3-1 models has been derived.
Examining the fermion contents, we have found that the 3-3-1 models contain themselves two solutions such as the electric charge quantization and the generation number problem. Moreover, theoretically, the electric charges of the neutron and of the neutrino as well as of the proton and of the electron are opposite.
We pointed out that the electric charge quantization is independent on generating mass to the fermions. This is the main difference between our approach and that in , which was based on.
We have also shown that if in the GUT, the electric charge quantization results from the algebra structure, here in the 3-3-1 models based on the semi-simple group, it is a direct consequence of the fermion contents.
This conclusion adds one more nice feature to the 3-3-1 models.
## Acknowledgments
This work was supported in part by National Council for Natural Sciences of Vietnam contract No: KT - 41064. |
warning/0507/hep-ph0507304.html | ar5iv | text | # Topics in quantum field theory: Renormalization groups in Hamiltonian framework and baryon structure in a non-local QCD model
##
”We are at the very beginning of time for the human race. It is not unreasonable that we grapple with problems. But there are tens of thousands of years in the future. Our responsibility is to do what we can, learn what we can, improve the solutions, and pass them on.”
Richard Feynman
## Acknowledgement
I must thank my wife, Shiva, for all her patience and love. Without her company and support this work would never have been completed.
I would like to thank my supervisor Niels Walet for his wise advice, encouragement, support, and his friendship over the duration of my PhD. Niels has taught me how to become an independent researcher as I should. His democratic attitude toward me allowed me to learn many different aspects of physics during my PhD, thanks for all your help Niels.
I would also like to thank Raymond F. Bishop, the head of our theoretical physics group, for all his continuous support and encouragement which made my PhD possible.
It has been my great pleasure to work with Mike Birse, his warm character and sense of humour makes it very easy to learn a lot, I am very grateful to him for all his inspiration and help.
My special thanks go to Hans-Juergen Pirner and Franz Wegner for their hospitality and fruitful discussions and remarks during my visit in Heidelberg.
I would like to thank Bob Plant for useful correspondence and Boris Krippa for interesting discussions.
I thank my fellow colleagues at the department, Theodoros, Bernardo, Toby, Neill, Daniel for their friendship.
Finally, I would like to acknowledge support from British Government Oversea Research Award and UMIST scholarship.
## Declaration
No portion of the work referred to in this thesis has been submitted in support of an application for another degree or qualification of this or any other university, or other institution of learning.
## Abstract
In this thesis I investigate aspects of two problems. In the first part of this thesis, I will investigate how an effective field theory can be constructed. One of the most fundamental questions in physics is how new degrees of freedom emerge from a fundamental theory. In the Hamiltonian framework this can be rephrased as finding the correct representation for the Hamiltonian matrix. The similarity (not essentially unitary) renormalization group provides us with an intuitive framework, where a transition from a perturbative region to a non-perturbative one can be realised and physical properties can be computed in a unified way. In this context, we have shown that the well-known coupled-cluster many-body theory techniques can be incorporated in the Wilsonian renormalization group to provide a very powerful framework for construction of effective Hamiltonian field theories. Eventhough the formulation is intrinsically non-perturbative, we have shown that a loop-expansion can be implemented.
The second part of my thesis is rather phenomenologically orientated. In this part, I will employ an effective field-theoretical model as can be constructed by means of the techniques of the first part of my thesis, a quark-confining non-local Nambu-Jona-Lasinio model and study the nucleon and diquarks in this model. For certain parameters the model exhibits quark confinement, in the form of a propagator without real poles. After truncation of the two-body channels to the scalar and axial-vector diquarks, a relativistic Faddeev equation for nucleon bound states is solved in the covariant diquark-quark picture. The dependence of the nucleon mass on diquark masses is studied in detail. We find parameters that lead to a simultaneous reasonable description of pions and nucleons. Both the diquarks contribute attractively to the nucleon mass. Axial-vector diquark correlations are seen to be important, especially in the confining phase of the model. We study the possible implications of quark confinement for the description of the diquarks and the nucleon. In particular, we find that it leads to a more compact nucleon.
## General remarks
This thesis is organized as follows: in the first three chapters, we concentrate on renormalization group methods in Hamiltonian framework. In chapter 1, we introduce the coupled-cluster theory. In chapter 2, we show how renormalization group can be employed in the context of the coupled-cluster theory. In order to highlight the merits and the shortcomings of our approach over previous ones, in Sec. 2.2, we review different RG methods in Hamiltonian framework. Different aspects of our approach is introduced in sections 2.4-2.8. In sections 2.9 and 2.10, as illustrative examples, we apply our formulation on the $`\mathrm{\Phi }^4`$ theory and an extended Lee model. In chapter 3, we pursue a different approach for the renormalization of the many-body problem. We show that a combination of the coupled-cluster theory and the Feshbach projection techniques leads to a renormalized generalized Brueckner theory.
In the second part of this thesis, we investigate the baryon structure in a chiral quantum chromodynamics model based on the relativistic Faddeev approach. In sections 4.1 and 4.2, we introduce the most important properties of QCD which are needed for the modelling of hadrons. In sections 4.3 and 4.4, we show how an effective low-energy field theory can be constructed from the underlying QCD theory. In chapter 5, we introduce alternative field theoretical approaches for describing baryons, such as Skyrme models, bag models and diquark-quark models in the context of the relativistic Faddeev approach. Finally, in chapter 6, we study baryons based on the diquark-quark picture in a quark-confining non-local NJL model. The non-local NJL model is introduced in Sec. 6.2. In Sec. 6.3, we discuss the pionic sector of the model. In Sec. 6.4 the diquark problem is solved and discussed. In Sec. 6.5 the three-body problem of baryons is investigated. The numerical technique involved in solving the effective Bethe-Salpeter equation is given and the results for three-body sector are presented.
## Chapter 1 Basic structure of the coupled-cluster formalism
### 1.1 Introduction
In order to understand fully the properties of quantum many-body systems, various methods have been developed which aim to go beyond perturbation theory. One of the simplest approaches has been the so-called configuration-interaction method which diagonalises the Hamiltonian in a finite subspace of the full many-body Hilbert space. An extension of this method has been introduced via various versions of coupled-cluster methods . The coupled cluster method (CCM) in its simplest form originated in nuclear physics around forty years ago in the work of Coester and Kümmel . The configuration-interaction method (CIM), and various version of coupled-cluster methods: normal coupled cluster method (NCCM) and the extended coupled cluster method (ECCM) form a hierarchy of many-body formulations for describing quantum systems of interacting particles or fields . They are denoted generically as independent-cluster (IC) parametrizations, in the sense that they incorporate the many-body correlations via sets of amplitudes that describe the various correlated clusters within the interacting system as mutually independent entities. The intrinsic non-perturbative nature of the methods is considered to be one of their advantages which make them almost universally applicable in many-body physics. The IC methods differ from each other in the way they incorporate the locality and separability properties; in a diagrammatic language they differ in their linking properties. Each of the IC methods has been shown to provide an exact mapping of the original quantum mechanical problem to a corresponding classical mechanics in terms of a set of multiconfigurational canonical field amplitudes. The merit of IC has been outlined in Ref. and literature cited therein.
### 1.2 Formalism
In this section we concentrate on the NCCM and the ECCM from a formal viewpoint.
Exponential structures arise frequently in physics for similar underlying fundamental reasons. For example, in the Ursell-Mayer theory in statistical mechanics, in the Goldstone linked-cluster theorem and the Gell-Mann and Low theorem . The complexity of the vacuum (the ground state) of an arbitrary many-body system in the NCCM parametrization is expressed by an infinite set of correlation amplitudes $`\{s_I,\stackrel{~}{s}_I\}`$ which have to be determined by the dynamics,
$`|\psi =K(t)e^S|\psi _0,\stackrel{~}{\psi }|={\displaystyle \frac{1}{K(t)}}\psi _0|\stackrel{~}{S}e^S,`$
$`S={\displaystyle \underset{I0}{}}s_IC_I^{},\stackrel{~}{S}=1+{\displaystyle \underset{I0}{}}\stackrel{~}{s_I}C_I.`$ (1.1)
Here $`K(t)`$ is a time-dependent scale factor. The coefficients $`s_I`$ and $`\stackrel{~}{s_I}`$ are time dependent. The intermediate normalization condition $`\stackrel{~}{\psi }|\psi =1`$ is explicit for all times $`t`$. We restrict ourselves to the non-degenerate system, so that the exact states of the system may sensibly be refereed to some suitably chosen single reference state denoted as $`|\psi _0`$. The state $`|\psi _0`$ is a ground state, e.g. a special (Hartree) bare vacuum in quantum field theory (QFT). The function $`|\psi _0`$ can be chosen rather generally, but is tied to the choice of generalized creation operators $`\{C_I^{}\}`$; the state $`|\psi _0`$ is annihilated by $`\{C_I\}`$ $`I0`$ (where $`C_01`$, the identity operator) and is a cyclic vector in the sense that the algebra of all possible operators in the many-body Hilbert space $``$ is spanned by the two Abelian subalgebras of creation and annihilation operators defined with respect to it. In this way we can define proper complete orthonormal sets of mutually commuting configuration creations operators $`\{C_I^{}\}`$ and their Hermitian adjoint counterparts $`\{C_I\}`$, defined in terms of a complete set of many-body configuration $`\{I\}`$. These are, in turn, defined by a set-index $`I`$, which labels the cluster configuration created by $`C_I^{}`$ with respect to the reference state $`|\psi _0`$. Therefore, $`\{I\}`$ defines a subsystem or cluster within the full system of a given configuration and the actual choice of these clusters depends upon the particular system under consideration. We assume that the creation and annihilation subalgebras and the state $`|\psi _0`$ are cyclic, so that all ket states in the Hilbert space $``$ can be constructed from linear combinations of states $`\{C_I^{}|\psi _0\}`$; and for the bra states with respect to states $`\{\psi _0|C_I\}`$. It is well-known in the many-body application that the above parametrization Eq. (1.2), guarantees automatically proper size-extensivity (see section 1.3) and conformity with the Goldstone linked-cluster theorem to all levels of truncation. In contrast, the configuration-interaction method is size extensive and linked only in the full, infinite-dimensional space . The NCCM parametrization of bra- and ket-states Eq. (1.2), in its asymmetrical (independent) form, does not manifestly preserve their Hermitian congugacy, hence we have here a biorthogonal formulation of the many-body problem. However, this is the most reasonable parametrization if one is to preserve the canonical form of the equations of motion with respect to phase space $`\{s_I,\stackrel{~}{s}_I\}`$ and the Hellmann-Feynman theorem <sup>1</sup><sup>1</sup>1According to Hellmann-Feynman theorem, if we perturb the Hamiltonian, $`HH^{}=H+\lambda A`$ (where $`\lambda `$ is infinitesimally small quantity) then the ground state energy changes as $`E_0E_0+\lambda dE_0/d\lambda +O(\lambda ^2)`$ with $`dE_0/d\lambda =\psi |dH/d\lambda |\psi `$. . Nevertheless, non-hermiticity is negligible if the reference state and its complement are not strongly correlated . We may hope that $`S`$ and $`\stackrel{~}{S}`$ are small (in a somewhat ill-defined non-perturbative sense), in other words, we may require that some of the coherence has already been obtained by optimizing the reference state. (This can be done, for example, by a Hartree-Bogolubov transformation). Then the remaining correlations can be added via the CCM<sup>2</sup><sup>2</sup>2 The application of this procedure to two-dimensional $`\varphi ^4`$ theory has been shown to give a stable result outside the critical region .. Therefore, defining a good reference state can in principle control the accuracy of CCM.
In the CIM, one defines the ket and bra states as follows,
$`|\psi =F|\psi _0;\stackrel{~}{\psi }|=\psi _0|\stackrel{~}{F},`$
$`F={\displaystyle \underset{I}{}}f_IC_I^{};\stackrel{~}{F}={\displaystyle \underset{I}{}}\stackrel{~}{f}_IC_I,`$ (1.2)
where the normalization condition $`\stackrel{~}{\psi }|\psi =1`$ can not be imposed trivially. Although the CIM has a simpler parametrization than the CCM, it does not satisfy the size-extensivity after truncation and contains unlinked pieces emerging from the products of noninteracting subclusters. A formal relation between the CIM and coupled cluster theory will be demonstrated in section 1.3.
While all ground-state expectation values $`\stackrel{~}{\psi }|A|\psi =\overline{A}(s_I,\stackrel{~}{s}_I)`$ and amplitudes $`\{s_I\}`$ are linked in NCCM, the amplitudes $`\{\stackrel{~}{s_I}\}`$ contain unlinked terms. This is resolved in the ECCM where we reparametrize the Hilbert space such that all basic amplitudes are linked,
$`|\psi =K(t)e^S|\psi _0;\stackrel{~}{\psi }|={\displaystyle \frac{1}{K(t)}}\psi _0|e^{\stackrel{~}{\mathrm{\Sigma }}}e^S,`$
$`\mathrm{\Sigma }|\psi _0=Qe^{\stackrel{~}{\mathrm{\Sigma }}}e^SS|\psi _0;Q=1|\psi _0\psi _0|,`$
$`\mathrm{\Sigma }={\displaystyle \underset{I0}{}}\sigma _IC_I^{};\stackrel{~}{\mathrm{\Sigma }}={\displaystyle \underset{I0}{}}\stackrel{~}{\sigma }_IC_I.`$ (1.3)
The inverse relationships between the ECCM and the NCCM counterparts are given by
$$\sigma _I\psi _0|C_I\stackrel{~}{S}S|\psi _0,s_I\psi _0|C_Ie^{\stackrel{~}{\mathrm{\Sigma }}}\mathrm{\Sigma }|\psi _0.$$
(1.4)
The ECCM amplitudes $`\{\sigma _I,\stackrel{~}{\sigma }_I\}`$ are canonically conjugate (which comes from time dependent variation)<sup>3</sup><sup>3</sup>3The equation of motion for the ECCM amplitudes can be obtained from a variational principle by requiring the action-like functional
$$𝒜=𝑑t\stackrel{~}{\psi }|i/tH(t)|\psi ,$$
(1.5) to be stationary against small variations of amplitudes. After some straightforward algebra, one can obtain
$$𝒜=𝑑t\left[i\underset{I0}{}\dot{\stackrel{~}{\sigma }}_I\sigma _IH\right].$$
(1.6) The stationary conditions lead us to the pair of equations of motion,
$$i\dot{\sigma }_I=\frac{\delta H}{\delta \stackrel{~}{\sigma }_I}i\dot{\stackrel{~}{\sigma }}_I=\frac{\delta H}{\delta \sigma _I}.$$
(1.7) Therefore, $`\{\sigma _I,\stackrel{~}{\sigma }_I\}`$ are obviously canonically conjugate to each other in the usual terminology of classical Hamiltonian mechanics. We will elaborate more on this property of the coupled-cluster theory in the context of the renormalization group in the chapter 2.7.. It must be clear that this parametrization for the ECCM is not unique, however it is complete and sufficient to specify the ECCM phase space. It has been pointed out earlier that this choice of parametrization is the most convenient one . The ECCM is believed to be the unique formulation of quantum many-body with full locality and separability at all levels of approximation. The individual amplitudes $`\{s_I,\stackrel{~}{s}_I\}`$ (or $`\{\sigma _I,\stackrel{~}{\sigma }_I\}`$) in both parametrizations are determined independently by solving an infinite set of non-linear equations which emerge from the dynamics of the quantum system. In practice, one needs to truncate both sets of coefficients. A consistent truncation scheme is the so called SUB($`n`$) scheme, where the $`n`$-body partition of the operator $`\{s_I,\stackrel{~}{s}_I\}`$ (or $`\{\sigma _I,\stackrel{~}{\sigma }_I\}`$) is truncated so that the general set-index $`\{I\}`$ contains up to $`n`$-tuple excitation (e.g., of single particle for bosonic systems with a reference state). Determination of the amplitudes corresponds to summing infinite sets of diagrams which in perturbation language take into account arbitrary high-order contributions in the coupling constant, therefore the NCCM and ECCM are not an expansion in this coupling constant. This property demands a precise truncation scheme for hierarchies without losing the renormalizability of the theory. We will consider this problem in the next chapter.
In the following, we confine our consideration to a real Klein-Gordon field, as an example. The configuration operators are specified by $`\{C_Ia_{k_1}\mathrm{}a_{k_I},C_I^{}a_{k_1}^{}\mathrm{}a_{k_I}^{}\}`$ with subalgebra
$$[a_k,a_k^{}^{}]=\delta _{kk^{}},[a_k,a_k^{}]=0.$$
(1.8)
The corresponding $`S`$ and $`\stackrel{~}{S}`$ operators are
$`S={\displaystyle \underset{n=1}{}}S_n,S_n={\displaystyle \underset{q_1,..q_n}{}}{\displaystyle \frac{1}{n!}}s_n(q_1,..q_n)a_{q_1}^{}..a_{q_n}^{},`$
$`\stackrel{~}{S}=1+{\displaystyle \underset{n=1}{}}\stackrel{~}{S}_n,\stackrel{~}{S}_n={\displaystyle \underset{q_1,..q_n}{}}{\displaystyle \frac{1}{n!}}\stackrel{~}{s}_n(q_1,..q_n)a_{q_1}..a_{q_n},`$ (1.9)
and $`|\psi _0`$ is the Fock vacuum. The individual amplitudes $`\{s_n,\stackrel{~}{s}_n\}`$ which describe excitations of $`n`$ Fock particles have to be fixed by the dynamics of the quantum system. Using Fock states in QFT has often been ambiguous due to problems connected with Haag’s theorem (Haag’s theorem says that there can be no interaction picture - that we cannot use the Fock space of noninteracting particles as a Hilbert space - in the sense that we would identify Hilbert spaces via field polynomials acting on a vacuum at a certain time). It is well-known that the algebraic structure, does not, in general, fix the Hilbert space representation and therefore dynamical considerations are required. The IC formalism invokes dynamics rigorously. The dynamical principle to fix the physical vacuum is Poincaré invariance,
$$H|\psi =P|\psi =L|\psi =K|\psi =0,$$
(1.10)
where $`H,P,L`$ and $`K`$ are generators of the Poincaré group (Hamiltonian, momentum, angular momentum and boost operators, respectively). The excited states are no longer invariant under these symmetry operations and the spectrum can be obtained in an extended version of IC . By putting the Ansatz Eqs. (1.2,1.2) into the conditions Eq. (1.10) one can obtain:
$`q_1,\mathrm{},q_n|e^SPe^S|\psi _0=q_1,\mathrm{},q_n|(P+[P,S])|\psi _0=\left[{\displaystyle \underset{i=1}{\overset{n}{}}}q_i\right]s_n(q_1,\mathrm{},q_n)=0,`$
(1.11)
$`\psi _0|\stackrel{~}{S}e^SPe^S|q_1,\mathrm{},q_n=\psi _0|\stackrel{~}{S}\left(P+[P,S]\right)|q_1,\mathrm{},q_n=\left[{\displaystyle \underset{i=1}{\overset{n}{}}}q_i\right]\stackrel{~}{s}_n(q_1,\mathrm{},q_n)`$
$`+{\displaystyle \underset{m}{}}{\displaystyle \frac{\stackrel{~}{s}_{n+m}(q_1,\mathrm{},q_{n+m})}{m!}}\left[[{\displaystyle \underset{i=1}{\overset{m}{}}}q_i]s_m(q_1,\mathrm{},q_m)\right]=0,`$ (1.12)
$`q_1,\mathrm{},q_n|e^SLe^S|\psi _0={\displaystyle \underset{\alpha }{\overset{n}{}}}ϵ_{ijl}(q_\alpha )_j{\displaystyle \frac{}{(q_\alpha )_l}}s_n(q_1,\mathrm{},q_n)=0.`$ (1.13)
In the same fashion one can impose the condition $`\psi _0|\stackrel{~}{S}e^SLe^S|q_1,\mathrm{},q_n=0`$, which leads to the following condition, having made used of the equation (1.13):
$$\underset{\alpha =1}{\overset{n}{}}ϵ_{ijl}(q_\alpha )_j\frac{}{(q_\alpha )_l}\stackrel{~}{s}_n(q_1,\mathrm{},q_n)=0$$
(1.14)
Similarly for Hamiltonian and boost operators we have
$`q_1,\mathrm{},q_n|e^SHe^S|\psi _0=\psi _0|\stackrel{~}{S}e^SHe^S|q_1,\mathrm{},q_n=0,`$ (1.15)
$`q_1,\mathrm{},q_n|e^SKe^S|\psi _0=\psi _0|\stackrel{~}{S}e^SKe^S|q_1,\mathrm{},q_n=0.`$ (1.16)
The equations (1.2,1.12,1.13,1.14) lead us to
$$s_n(q_1,\mathrm{},q_n)=\delta \left[\underset{i=1}{\overset{n}{}}q_i\right]s_n(\{q_i.q_j\}),\stackrel{~}{s}_n(q_1,\mathrm{},q_n)=\delta \left[\underset{i=1}{\overset{n}{}}q_i\right]\stackrel{~}{s}_n(\{q_i.q_j\}),$$
(1.17)
which means that $`\{s_I,\stackrel{~}{s}_I\}`$ depend on scalar quantities only and momentum is preserved. The same result can be obtained for the ECCM parametrization. To complete the determination of phase space, we use the energy hierarchy Eq. (1.15), where $`H`$ is a normal ordered Hamiltonian and the vacuum energy vanishes. Conceptually, it is evident that Eq. (1.15) suffices to fix all amplitudes without invoking Eq. (1.16), but explicit verification to all orders seems to be impossible.
The fully linked feature of the $`e^SHe^S`$ term in Eq. (1.15) can be made explicit by denoting it as $`\{He^S\}_{}`$, which can be written as a set of nested commutators,
$$e^SHe^S=\{He^S\}_{}=H+[H,S]+\frac{1}{2!}[[H,S],S]+\mathrm{}.$$
(1.18)
This procedure is still rigorous. Lorentz symmetry, stability and causality are examples of features normally expected to hold in physical quantum field theories. In renormalized QFT stability and causality are closely intertwined with Lorentz invariance. For example, stability includes the need for energy positivity of Fock states of ordinary momenta, while causality is implemented microscopically by the requirement that observables commute at spacelike separation , so-called microcausality. In addition, both are expected to hold in all inertial frames. A stable and causal theory without Lorentz symmetry could in principle still be acceptable . In the framework of many-body theory, medium contributions always affect the local and global properties of hadrons especially at high density. In dense matter, the Pauli principle and cluster properties can affect causality since they restrict the permissible process in a scattering reaction. To understand these effects we need firstly to consider if our formalism itself can in principle preserve the causal structure of given physical system. Let us introduce a new set $`\{b_k,b_k^{}\}`$ which are connected to previous $`\{a_k,a_k^{}^{}\}`$ defined in Eq. (1.8) via a generalized Bogolubov transformation,
$$b_k^{}=A_{ki}a_i^{}+B_{ki}a_i+D_i,$$
(1.19)
This is the most general linear transformation, which preserves commutator relations ($`b_k`$ is an annihilation operator and $`b_k^{}`$ is the Hermitian conjugate of $`b_k`$),
$$[b_k,b_k^{}^{}]=\delta _{kk^{}},[b_k,b_k^{}]=0,$$
(1.20)
provided that $`AA^{}BB^{}=1`$. This obviously preserves the commutator relations for boson fields and preserves microcausality explicitly. The bare “$`a`$ vacuum” defined by $`a_k|\psi _0_a=0`$, is replaced by a bare “$`b`$ vacuum” which satisfies $`b_k|\psi _0_b=0`$. Using Thouless theorem , $`|\psi _b`$ can be written as
$$|\psi _b=N^{1/2}e^{S_1+S_2}|\psi _a,$$
(1.21)
with
$$S_1=\underset{k}{}S_k^1(A,B,D)a_k^{},S_2=\underset{kk^{}}{}\frac{1}{2!}S_{kk^{}}^2(A,B,D)a_k^{}a_k^{}^{},$$
(1.22)
where $`S^1`$ and $`S^2`$ are known functions of the matrices $`A`$ and $`B`$ and the vector $`D`$ . It is obvious that $`|\psi _b`$ is a low-order approximation to the CCM wave function Eq. (1.2) or (1.2). The above-mentioned parametrization of vacuum wave function can be generalized to the CCM wave function, however it may require a nonlinear transformation from which it can be constructed. The inspiration for this transformation can be taken from the extension IC formulation for excited states
$$b_k^{}=e^SF(k),F(k)=a_k^{}+\underset{n=3}{\overset{\mathrm{}}{}}F_n(p_1,\mathrm{},p_{n1},k)a_{p_1}^{}\mathrm{}a_{p_{n1}}^{},$$
(1.23)
where the correlation operator $`S`$ is known from Eq. (1.2) and $`F`$ is a new amplitude which includes momentum conservation. This new amplitudes has to be determined and it changes the momentum of the bosons before creation. It is not hard to derive equations for the energy spectrum which lead us to $`N`$-body effective Hamiltonians and yields the folded diagrams of degenerate many-body perturbation theory . This nonlinear transformation manifestly invalidates the commutator relation Eq. (1.20) and accordingly the microcausal commutation relations of the boson field. It should be noted that causality of the underlying theory can not be fully determined at this level and one needs to take into account the dynamics of the underlying quantum system. Causality in the context of IC formulation can be ensured by requiring
$$\stackrel{~}{\psi }|[\varphi (X),\varphi (Y)]|\psi =0,(XY)^2<0,$$
(1.24)
where $`Y`$ denotes space-time coordinates $`(y,y_0)`$ and $`\varphi (Y)`$ is Klein-Gordon field operator which is expressed in terms of Fock space operators by
$$\varphi (Y)=\underset{k}{}\xi (Y)_ka_k+\xi _k^{}(Y)a_k^{},$$
(1.25)
where $`\xi _k(Y)`$ form a complete orthonormal set of states. Eq. (1.24) can be evaluated explicitly by using the following identity
$$\stackrel{~}{\psi }|[A,B]|\psi =\overline{[A,B]}=\underset{I}{}\frac{\overline{A}}{x_I}\frac{\overline{B}}{\stackrel{~}{x}_I}\frac{\overline{A}}{\stackrel{~}{x}_I}\frac{\overline{B}}{x_I},$$
(1.26)
where $`\{x_I,\stackrel{~}{x}_I\}`$ are the canonical coordinate and momenta of the NCCM or the ECCM parametrization. By making use of Eqs. (1.2,1.18) and (1.25) one can find
$$\stackrel{~}{\psi }|\varphi (Y)|\psi =\overline{\varphi (Y)}=\underset{n=1}{}\underset{k}{}\xi _k(Y)\stackrel{~}{x}_{n1}(q_1,\mathrm{},q_{n1})x_n(q_1,\mathrm{},q_{n1},k)+\xi _k^{}(Y)\stackrel{~}{x}_1(k),$$
(1.27)
where $`\{x_ns_n,\sigma _n,\stackrel{~}{x}_n\stackrel{~}{s}_n,\stackrel{~}{\sigma }_n\}`$ and we define $`\stackrel{~}{x}_0=1`$. By exploiting Eq. (1.26) one can show
$$\overline{[\varphi (X),\varphi (Y)]}=\underset{k,k^{}}{}[\xi _k(X)\xi _k^{}^{}(Y)\xi _k^{}(X)\xi _k^{}(Y)]=\mathrm{\Delta }(XY)\mathrm{\Delta }(YX).$$
(1.28)
The function $`\mathrm{\Delta }(XY)`$ is the Pauli-Jordan function defined for field operators expanded in plane-wave basis. When $`(XY)^2<0`$, we can perform a Lorentz transformation on the second term, taking $`(XY)(XY)`$. The two terms are therefore equal and cancel to give zero, hence, Eq. (1.24) is satisfied. It should be noted that for a general quantum system by considering just the dynamics of the underlying system, determining the amplitudes $`\{x_I,\stackrel{~}{x}_I\}`$ and verifying Eq. (1.24) at every level of truncation one can ensure causality. Obviously this might introduce a lower limit of truncation in a consistent SUB(n) scheme which contains causality.
In relativistic quantum mechanics there is another distinct type of causality, the fact that there is a well-posed initial value problem, so-called Cauchy causality. In the local field theory the Poincaré invariance implies the existence of a unitary representation of the Poincaré group that acts on the Hilbert space. In other words, the transformed final state is uniquely determined by time evolving the transformed initial state. Therefore, Cauchy causality is a consequence of Poincaré invariance and is independent of any consideration concerning microcausality.
One of the standard approaches in the nuclear many-body theory is to introduce an effective interaction and reduce the full many-body problem to a problem in a small model-space. As an example for the meson-nucleon system below threshold, it was shown that one has either hermitian effective operators with all desirable transformation properties under the Poincaré group or non-hermitian ones obtained by the CCM with desirable simplicity . Thus effective operators introduced by the CCM can not manifestly preserve Cauchy causality, at any level of truncation. However, it economically disentangles and reduces the complexities of the many-body problem. It is notable that in some cases, the violation of relativistic invariance due to truncation is consistent with errors introduced via the approximation . In practice one should not expect that Poincaré invariance holds for an approximation, and forcing the approximated wave function to obey Lorentz invariance might lead to inconsistent results. Having said that, we will show that in the context of coupled-cluster renormalization group framework, one can in principle define a truncation scheme where Poincaré invariance is preserved.
### 1.3 Linked-cluster theorem and Wightman functionals
In this section we proceed from a formal view to show the relation between the CIM and the coupled cluster method. We will show that the CCM is a natural reparametrization of the CIM which incorporates the full size-extensivity. For this purpose we exploit the reconstruction theorem , well known in axiomatic Quantum field theory :
Denote the state vector of the vacuum by $`|\mathrm{\Omega }`$ (in the language of the CCM $`|\mathrm{\Omega }`$ is the full interacting ground state $`|\psi `$). The physical vacuum expectation value of products of local fields
$$w^n(x_1\mathrm{}x_n)=\mathrm{\Omega }|\varphi (x_1)\mathrm{}\varphi (x_n)|\mathrm{\Omega },$$
(1.29)
are tempered distributions<sup>4</sup><sup>4</sup>4Tempered distributions generalize the bounded (or slow-growing) locally integrable functions; all distributions with compact support and all square-integrable functions can be viewed as tempered distributions. over $`^{4n}`$. These $`w^n`$ are called Wightman distributions. If the hierarchy of distributions $`w^n(n=0,\mathrm{})`$ is known then the Hilbert space can be constructed. This is the so-called reconstruction theorem .
Now consider a configuration $`(x_1,\mathrm{},x_n)`$ consisting of several clusters. A cluster here is a subset of points so as all the points in one cluster have a large space-like separation from all the points in any other clusters. One may expect that in infinite separation between the clusters, one has
$$w^n(x_1,\mathrm{},x_n)\underset{r}{}w^{n_r}(y_{r,1},\mathrm{}y_{r,n_r}),$$
(1.30)
the index $`r`$ and $`n_r`$ denote the cluster and number of points in the $`r`$-th cluster, respectively. $`y_{r,n_k}`$ indicates a subset of the $`x_k`$ belonging to this cluster. Thus we require that in the vacuum state the correlation of quantities relating to different regions decreases to zero as the space-like separation of the regions increases to infinity. Therefore it makes sense to introduce another hierarchy of functions $`w_T^n`$,
$$w^n(x_1,\mathrm{},x_n)=\underset{p}{}\underset{r}{}w_T^{n_r}(y_{r,k}).$$
(1.31)
Here $`p`$ denotes a partition of the set of points $`x_i`$ into subsets labeled by the index $`r`$. The sum is over all possible partitions, The objects $`w_T^n`$ are called truncated functions or correlated functions<sup>5</sup><sup>5</sup>5For free fields all truncated functions with $`n2`$ vanish, thus it suffices to know the two-point functions.. Within the hierarchy $`\{w^n\}`$ the truncated functions can be computed recursively:
$`w_T^1(x)=w^1(x),`$
$`w_T^2(x_1,x_2)=w^2(x_1,x_2)w^1(x_1)w^1(x_2),`$ (1.32)
$`\mathrm{}`$
The asymptotic property Eq. (1.30) of the $`w^n`$ is converted to the simpler property
$$w_T^n(x_1,\mathrm{}x_n)0,$$
(1.33)
if any space like separation $`x_ix_j`$ goes to infinity.
Let us now consider a hierarchy of functions $`P^n(k_1,\mathrm{}k_n)`$ (for simplicity totally symmetric under permutation of their arguments). We define a generating functional $`𝒫\{f\}`$ for an arbitrary function $`f(k)`$:
$$𝒫\{f\}=\underset{n}{}\frac{1}{n!}P^n(k_1,\mathrm{}k_n)f(k_1)\mathrm{}f(k_n).$$
(1.34)
The $`P^n`$ can be found from
$$P^n(k_1,\mathrm{}k_n)=\frac{\delta ^n}{\delta f(x_1)..\delta f(x_n)}𝒫\{f\}|_{f=0}.$$
(1.35)
Now one can simply use the definition Eq. (1.31) to obtain a relation between the hierarchy $`\{P^n\}`$ and the hierarchy of the truncated functions $`\{P_T^n\}`$ in terms of respective generating function, namely
$$𝒫\{f\}=e^{𝒫_T\{f\}}.$$
(1.36)
This relation is called the Linked cluster theorem . Having in mind the relations Eqs. (1.34) and (1.36), one observes that the CCM parametrization introduced in Eqs. (1.2) and (1.2) is a natural reparametrization of the CIM defined in Eq. (1.2) which incorporates the size-extensivity , at finite levels of truncation. This leads to a cluster decomposition and linked decomposition property of CCM at any level of truncation. Thus for extensive variables as the energy, the linked terms lead to contributions which obey the proper linear scaling in the particle number. The CIM contains unlinked diagrams for the ground-state energy expectation value, and thereby suffers from the so-called size-extensivity problems.
The cluster decomposition condition Eq. (1.30) is fundamental to a quantum field theory. It may break down partly when, in a statistical-mechanical sense, the theory is in a mixed phase . This implies that there is more than one possible vacuum state, and therefore the cluster decomposition should be restored in principle if one builds the Hilbert space on one of the vacua.
Notice that in QCD, the cluster decomposition for colour singlet objects can break down due to the confinement. In axiomatic local quantum field theory with an indefinite metric space $`\nu `$ (e.g., Minkowski space) the following theorem holds; for the vacuum expectation values of two (smeared local) operators $`A`$ and $`B`$ with spacelike distance $`R`$ from each other :
$`|\mathrm{\Omega }|A(x)B(0)|\mathrm{\Omega }`$ $``$ $`\mathrm{\Omega }|A(x)|\mathrm{\Omega }\mathrm{\Omega }|B(0)|\mathrm{\Omega }|`$ (1.39)
$``$ $`\{\begin{array}{cc}\text{const}\times R^{3/2+2N}e^{MR}\hfill & \text{if there is a mass gap M},\hfill \\ \text{const}\times R^{2+2N}\hfill & \text{if there is no mass gap},\hfill \end{array}`$
where $`M`$ is the mass gap. (Herein, we assume $`N=0`$, however, a positive $`N`$ is possible for the indefinite inner product structure in $`\nu `$.) In order to avoid the decomposition property for product of unobservable operators $`A`$ and $`B`$ which together with Kugo-Ojima criterion<sup>6</sup><sup>6</sup>6The Kugo-Ojima confinement scenario describes a mechanism by which the physical state space contains only colourless states, the colored states are not BRS-singlets and therefore do not appear in $`S`$-matrix elements, since they are confined . for the confinement is equivalent to failure of the decomposition property for coloured clusters, there can not be a mass gap in the indefinite space $`\nu `$. This would thus eliminate the possibility of scattering a physical state into color singlet states consisting of widely separated colored clusters (the “behind-the-moon” problem). However, this has no information for the physical spectrum of the mass operator in $``$ which indeed does have a mass gap. In another words, there is a mass gap (but not in the full indefinite space $`\nu `$) in the semi-definite physical subspace induced by the BRST charge operator $`Q_B`$, $`\nu _{\text{phy}}=\text{Ker}Q_B`$, where states within are annihilated by $`Q_B`$,
$`v_{\text{phys}}`$ $`=`$ $`\{|\psi v:Q_B|\psi =0\}=\text{Ker}Q_B,`$
$`(Q_B,v)`$ $`=`$ $`\text{Ker}Q_B/\text{Im}Q_B,`$ (1.40)
where states in $`\text{Im}Q_B=(\text{Ker}Q_B)^{}`$ are called BRST-coboundaries in the terminology of de Rham cohomology and do not contribute in $`\nu _{\text{phys}}`$ . It has been shown that there is a connection between Kugo-Ojima criterion and an infrared enhancement of the ghost propagator. In Landau gauge, the gluon-ghost vertex function offers a convenient possibility to define a non-perturbative running coupling. The infrared fixed point obtained from this running coupling determines the two-point color-octet interactions and leads to the existence of unphysical massless states which are necessary to escape the cluster decomposition of colored clusters . Notice that a dynamical mechanism, responsible for breaking of cluster decomposition for colored objects, is yet to be discovered, since our knowledge about color confinement is still preliminary.
## Chapter 2 Hamiltonian renormalization Groups
### 2.1 Introduction
The main goal of traditional renormalization theory is to determine when and how the cancellation of divergences originating from the locality of quantum field theory may occur. This is essential if one wants to have meaningful quantitative results. However, it is by no means obvious how quantum fluctuations associated with short distance scales can be incorporated and controlled through the choice of only a few parameters, typically the bare masses and coupling constants, or by the counterterms in renormalized perturbation theory.
The development of Wilson’s renormalization group (RG) formalism allowed physicists to produce a logically consistent picture of renormalization in which perturbation theory at arbitrary high energy scale can be matched with the perturbation expansion at another scale, without invoking the details of intermediate scales. In the Wilsonian approach, all of the parameters of a renormalizable field theory can be thought of as scale-dependent objects and their flows are governed by the so-called RG differential equations.
Motivated by Wilson’s picture, effective field theory (EFT) approaches have been introduced to replace complicated fundamental theories with simpler theories based only on the relevant degrees of freedom at the physical scale of interest . The basis of the EFT concept is the recognition of the importance of different typical energy scales in nature, where each scale has its own characteristic degrees of freedom. In strong interactions the transition from the fundamental to the effective level is induced by a phase transition that takes place around $`\mathrm{\Lambda }_{\text{QCD}}1\text{GeV}`$ via the spontaneous breaking of chiral symmetry, which generates pseudoscalar Goldstone bosons. This coincides, of course, with the emergence of nuclei and nuclear mather, as opposed to the quark-gluon plasma and quark matter expected to occur at high temperature and high density. Therefore, at low energies ($`E<\mathrm{\Lambda }_{\text{QCD}}`$), the relevant degrees of freedom are not quarks and gluons, but pseudoscalar mesons and other hadrons. The resulting description is a chiral EFT, which has much in common with traditional potential models. In particular, there might be an intermediate regime where a non-relativistic model is inadequate but where relatively few hadronic degrees of freedom can be used to faithfully describe both nuclear structure and response. If this is the case, one should be able to describe strongly interacting hadronic systems in this effective model.
The power of Hamiltonian methods is well known from the study of non-relativistic many-body systems and from strongly-interacting few particle systems, even though a Lagrangian approach is usually chosen for relativistic theories. One might prefer to obtain the effective interaction using the covariant Lagrangian formalism and then consider the ground state and collectively excited states in the non-covariant Hamiltonian formalism by exploiting many-body techniques. For the description of physical states and in particular bound states, the Hamiltonian formalism is preferable over the Lagrangian one. This is due to the fact that such problems are not naturally defined covariantly. For a bound state the interaction time scale is infinite, and thus a time-independent approach is better suited. However, the Lagrangian can not generally be converted to a Hamiltonian if the effective Lagrangian contains higher-order time-derivatives, since no Legendre transformation exits for such a case. Therefore one may wish to obtain the effective interaction in a unified self-consistent way within the Hamiltonian formalism. The renormalization group transformation is used to derive the physical Hamiltonian that can describe experiment. To implement a RG calculation, one should define a space of Hamiltonians and find a certain RG transformation which maps this space into itself. Then one should study the topology of the Hamiltonian space, by searching for the fixed points and studying the trajectories of the Hamiltonian with respect to these fixed points. At the fixed points we have a scale-invariant quantum field theory. Near the fixed point, the irrelevant operators in the original Hamiltonian have small coefficients, and we are only left with the relevant and marginal operators. The coefficients of these operators correspond to the parameters of the renormalized field theory. In this way, in a renormalizable field theory, we disregard information about the evolution of irrelevant terms. The fact that irrelevant terms can be dropped near the fixed point does not necessarily imply that they are unimportant at low-energy scales of experimental interest. This depends on how sensitive the physical observables are to physics near the scale of the cutoff. Therefore, it is of interest to develop a RG method in which the irrelevant variables are treated on an equal footing with the relevant and marginal variables (the schemes presented here have this advantage).
Hamiltonian methods for strongly-interacting systems are intrinsically non-perturbative and usually contain a Tamm-Dancoff type approximation, in the sense that one expands the bound state in states containing a small number of particles. This truncation of the Fock space gives rise to a new class of non-perturbative divergences, since the truncation does not allow us to take into account all diagrams for any given order in perturbation theory. Therefore renormalization issues have to be considered carefully. Two very different remedies for this issue are the use of light-front Tamm-Dancoff field theory (LFFT) and the application of the coupled cluster method (CCM) . However, both methods are too complicated to attack the issue in a self-consistent way. In the last decade extensive attempts have been made to give a workable prescription for renormalization within the Hamiltonian formalism . Commonly unitary transformations are used to decouple the high- and low-energy modes aiming at the partial diagonalization of the Hamiltonian. One of the most elegant approaches in this context is the similarity renormalization group (SRG) proposed by Glazek and Wilson (and by Wegner independently). The SRG is designed to be free of small energy denominators and to eliminate interactions which change the energies of the unperturbed states by a large amount. However, there are several problems with this approach: it is hard to incorporate loop expansions within the method, the SRG can not systematically remove interactions which change the number of particles (i.e, when the Hamiltonian is not diagonal in particle number space), and most importantly, the computations are complex and there is no efficient non-perturbative calculation scheme.
In this chapter we introduce a new method for obtaining the low-energy effective operators in the framework of a CCM approach. The transformation constructed avoids the small denominators that plague old-fashioned perturbation theory. Neither perturbation theory nor unitarity of the transformation are essential for this method. The method is non-perturbative, since there is no expansion in the coupling constant; nonetheless, the CCM can be conceived as a topological expansion in the number of correlated excitations. We show that introducing a double similarity transformation using linked-cluster amplitudes will simplify the partial diagonalization underlying renormalization in Hamiltonian approaches. However, a price must be paid: due to the truncation the similarity transformations are not unitary, and accordingly the hermiticity of the resultant effective Hamiltonian is not manifest. This is related to the fact that we have a biorthogonal representation of the given many-body problem. There is a long tradition of such approaches. The first we are aware of are Dyson-type bosonization schemes . \[Here one chooses to map the generators of a Lie algebra, such that the raising generators have a particularly simple representation.\] The space of states is mapped onto a larger space where the physically realizable states are obtained by constrained dynamics. This is closely related to CCM formalism, where the extended phase space is a complex manifold, the physical subspace constraint function is of second class and the physical shell itself is a Kähler manifold . The second is the Suzuki-Lee method in the nuclear many-body (NMT) problem , which reduces the full many-body problem to a problem in a small configuration space and introduces a related effective interaction. The effective interaction is naturally understood as the result of certain transformations which decouple a model space from its complement. As is well known in the theory of effective interactions, unitarity of the transformation used for decoupling or diagonalization is not necessary. Actually, the advantage of a non-unitarity approach is that it can give a very simple description for both diagonalization and ground state. This has been discussed by many authors and, although it might lead to a non-hermitian effective Hamiltonian, it has been shown that hermiticity can be recovered . Notice that defining a good model space can in principle control the accuracy of CCM .
To solve the relativistic bound state problem one needs to systematically and simultaneously decouple 1) the high-energy from low-energy modes and 2) the many- from the few-particle states. We emphasize in this chapter that CCM can in principle be an adequate method to attack both these requirements. Our hope is to fully utilize Wilsonian Exact renormalization group within the CCM formalism. Here the high energy modes will be integrated out leading to a modified low-energy Hamiltonian in an effective many-body space. Notice that our formulation does not depend on the form of dynamics and can be used for any quantization scheme, e.g., equal time or light-cone.
### 2.2 Traditional approaches and their problems
The Tamm-Dancoff approximation was developed in the 1950’s to describe a relativistic bound state in terms of a small number of particles. It was soon revealed that the Tamm-Dancoff truncation gives rise to a new class of non-perturbative divergences, since the truncation does not allow us to take into account all diagrams at a given order in perturbation theory. On the other hand, any naive renormalization violates Poincaré symmetry and the cluster decomposition property (the cluster decomposition means that if two subsystems at very large space-like separation cease to interact then the wave function becomes multiplicatively separable). One of the simplest example of the Tamm-Dancoff approximation is the constituent picture of QCD where one describes a QCD bound state $`|\psi `$ within a truncated Fock-space,
$$|\psi =\varphi _1|q\overline{q}+\varphi _2|q\overline{q}g+\varphi _3|q\overline{q}q\overline{q}+\mathrm{}$$
we use a shorthand notation for the Fock-space where $`q`$ is a quark, $`\overline{q}`$ an antiquark, and $`g`$ stands for a gluon. Now the bound-state problem is solved via the Schrödinger equation $`H^{QCD}|\psi =E|\psi `$. It is well known that in the complicated equal-time vacuum bound states contain an infinite number of particles that are part of the physical vacuum on which hadrons are built. On the other hand, interactions in a field theory couple states with arbitrarily large difference in both free energy and numbers of particles. Thus any Fock-space expansion can hardly be justified without being supplemented with a prescription for decoupling of the high-energy from the low-energy modes and the many-body from few-particle states. In the context of the Hamiltonian formulation this problem can be expresed by asking how the Hamiltonian matrix is diagonalized in particle- and momentum-space.
In his earliest work Wilson exploited a Bloch type transformation to reduce the Hamiltonian matrix by lowering a cutoff which was initially imposed on the individual states. Later Wilson abandoned this formulation in favour of a Lagrangian one. The most important reason was that the Bloch transformation is ill-defined and produces unphysical divergences. These divergences emerge from denominators which contain a small energy difference between states retained and states removed by the transformation, and appear across the boundary line at $`\lambda `$ .
Two remedies for this issue are the use of light front coordinates and application of the CCM . In the light-front Tamm-Dancoff field theory (LFFT) the quantization plane is chosen to coincide with the light front, therefore the divergences that plagued the original theories seem to disappear since here vacuum remains trivial. Furthermore, not having to include interactions in boost operators allows a renormalizable truncation scheme . One of the most important difficulties in LFFT is the complicated structure of the renormalization process . In principle, ad-hoc counterterms can not be prevented if one is to preserve the underlying symmetry. In the standard form of CCM, on the other hand, the amplitudes obey a system of coupled non-linear equations which contain some ill-defined terms because of ultraviolet divergences. It has been shown that the ill-defined amplitudes, which are also called critical topologies, can be systematically removed, by exploiting the linked-cluster property of the ground state. This can be done by introducing a mapping which transfers them into a finite representations without making any approximation such as a coupling expansion. Thus far this resummation method has been restricted to superrenormalizable theories due to its complexity.
Recently Wilson and Glazek and independently Wegner have re-investigated this issue and introduced a new scheme, the similarity renormalization group (SRG). The SRG resembles the original Wilsonian renormalization group formulation , since a transformation that explicitly runs the cutoff is developed. However, here one runs a cutoff on energy differences rather than on individual states. In the SRG framework one has to calculate a narrow matrix instead of a small matrix, and the cutoff can be conceived as a band width (see Fig. \[2.1\]). Therefore by construction the perturbation expansion for transformed Hamiltonians contains no small-energy denominators. Here we review the general formulation of the SRG.
#### 2.2.1 Glazek-Wilson formulation
The detailed description of Glazek-Wilson RG method can be found in Ref. . Here we only concentrate on the key elements of their method. We introduce a unitary transformation aiming at partially diagonalizing the Hamiltonian so that no couplings between states with energy differences larger than $`\lambda `$ are present. The unitary transformation defines a set of Hamiltonians $`H_\lambda `$ which interpolate between the initial Hamiltonian ($`\lambda =\mathrm{\Delta }`$ ) and the effective Hamiltonian $`H_\lambda `$. This $`\lambda `$ plays the role of a flow parameter. We will assume that $`H_\lambda `$ is dominated by its diagonal part which we will denote $`H_{0\lambda }`$ with eigenvalues $`E_{i\lambda }`$ and eigenstates $`|i`$:
$$f|H_{0\lambda }|i=E_{i\lambda }f|i.$$
(2.1)
Therefore the Hamiltonian at scale $`\lambda `$ can be written $`H_\lambda =H_{0\lambda }+H_{I\lambda }`$, where $`H_{I\lambda }`$ is non-diagonal part. Notice that $`H_{0\lambda }`$ is not necessarily the bare free Hamiltonian which is independent of $`\lambda `$. In order to introduce the infinitesimal unitary transformations which induce the infinitesimal changes in $`H_\lambda `$ when $`\lambda `$ changes by an infinitesimal amount, we need to define various zones of the operators We introduce an auxiliary function $`x_{ij\lambda }=i|x_\lambda |j`$ of the states with labels $`i`$ and $`j`$ for a given $`\lambda `$. If we denote the eigenvalue of $`H_{0\lambda }`$ with $`E_{i\lambda }`$, $`x`$ is defined as
$$x_{ij\lambda }=\frac{|E_{i\lambda }E_{j\lambda }|}{E_{i\lambda }+E_{j\lambda }+\lambda }.$$
(2.2)
The modulus of the function $`x_{ij\lambda }`$ is close to $`1`$ when one of the energies is much larger than the other and also much larger than the cutoff $`\lambda `$ and $`x`$ approaches $`0`$ when the energies are similar or small in comparison to the cutoff. One then introduces smooth projectors
$$u_{ij\lambda }=i|u_\lambda |j=u(x_{ij\lambda }),$$
(2.3)
and
$$r_{ij\lambda }=1u_{ij\lambda }=r(x_{ij\lambda }).$$
(2.4)
By means of $`u`$ and $`r`$ one can separate each matrix into two parts $`M=D(M)+R(M)`$, where we define
$$D(M)_{ij}=u_{ij\lambda }M_{ij},$$
(2.5)
and corresponding
$$R(M)_{ij}=r_{ij\lambda }M_{ij}.$$
(2.6)
The functions $`u`$ and $`r`$ are needed in order to ensure smoothness and differentiability, and consequently a continuous transition between different parts of the Hamiltonian matrix. They are the key elements in making the diagonalization free of small-energy denominators. These functions are implemented as a “form factor“ in every vertex of the interaction. We construct an infinitesimal unitary transformation eliminating the part of Hamiltonian which has only non-zero elements far away from the diagonal, thus they can not produce small-energy denominators. This continuous unitary transformation satisfies,
$$\frac{dH_\lambda }{d\lambda }=[T_\lambda ,H_\lambda ].$$
(2.7)
The generator $`T_\lambda `$ is anti-Hermitian and is chosen in such a way that
$$H_\lambda =D(Q_\lambda )=Q_\lambda u_\lambda ,$$
(2.8)
where $`Q`$ is arbitrary in the far-off-diagonal region. In terms of $`Q`$, the matrix elements of Eq. (2.7) satisfy
$$\frac{du_{ij\lambda }}{d\lambda }Q_{ij\lambda }+u_{ij\lambda }\frac{dQ_{ij\lambda }}{d\lambda }=T_{ij\lambda }(E_{j\lambda }E_{i\lambda })+[T_\lambda ,H_{I\lambda }]_{ij}.$$
(2.9)
Notice that $`Q_\lambda `$ can be arbitrary in the far-off diagonal region where $`u_\lambda =\frac{du_\lambda }{d\lambda }=0`$ if it is finite and its derivative is finite. In the above equation we have two unknowns, $`\frac{dQ_\lambda }{d\lambda }`$ and $`T_\lambda `$. As additional input we use the fact that, for a given $`\lambda `$, the Hamiltonian $`H_\lambda `$ can be additionally unitary transformed without violating the relation $`D(Q_\lambda )=H_\lambda `$. We regroup the above equation in the form,
$$u_{ij\lambda }\frac{dQ_{ij\lambda }}{d\lambda }T_{ij\lambda }(E_{j\lambda }E_{i\lambda })=[T_\lambda ,H_{I\lambda }]_{ij}\frac{du_{ij\lambda }}{d\lambda }Q_{ij\lambda }=G_{ij\lambda }.$$
(2.10)
The unknowns on the left hand-side are determined by first solving Eq. (2.10) neglecting the commutator on the right-hand side; the result is then substituted into the right-hand side, and solved iteratively until convergence is reached. The $`D`$ and $`R`$ parts of the operator $`G`$ can be defined,
$`u_{ij\lambda }{\displaystyle \frac{dQ_{ij\lambda }}{d\lambda }}`$ $``$ $`D(G_\lambda )_{ij},`$
$`T_{ij\lambda }(E_{j\lambda }E_{i\lambda })`$ $``$ $`R(G_\lambda )_{ij}.`$ (2.11)
By evaluating the matrix elements of both sides of the above equations in different zones of the operators, one obtains differential equations for matrix elements of $`Q_\lambda `$. Therefore, from Eqs. (2.10, 2.11) one can immediately obtain the generator $`T`$ and the Hamiltonian flow equation in terms of the matrix elements $`f|H_\lambda |i=H_{\lambda fi}`$ and $`f|T_\lambda |i=T_{\lambda fi}`$,
$`T_{ij\lambda }`$ $`=`$ $`{\displaystyle \frac{r_{ij\lambda }}{E_{i\lambda }E_{j\lambda }}}\left([T_\lambda ,H_{I\lambda }]_{ij}{\displaystyle \frac{d\mathrm{ln}u_{ij\lambda }}{d\lambda }}H_{ij\lambda }\right),`$ (2.12)
$`{\displaystyle \frac{dH_{ij\lambda }}{d\lambda }}`$ $`=`$ $`u_{ij\lambda }[T_\lambda ,H_{I\lambda }]_{ij}+r_{ij\lambda }{\displaystyle \frac{d\mathrm{ln}u_{ij\lambda }}{d\lambda }}H_{ij\lambda }.`$ (2.13)
It is obvious from the above equations that no small-energy denominators $`E_{i\lambda }E_{j\lambda }`$ arises out-side of the band (off-diagonal region), and within the band zone, we have $`T=0`$. The equations (2.12,2.13) can be solved iteratively.
#### 2.2.2 Wegner Formulation
Wegner’s formulation of the SRG is defined in a very elegant way aiming at diagonalization of the Hamiltonian in a block-diagonal form with the number of particles conserved in each block. Again, a unitary transformation is used with flow parameter $`s`$ that range from $`0`$ to $`\mathrm{}`$,
$$\frac{dH(s)}{ds}=[T(s),H(s)].$$
(2.14)
We separate the Hamiltonian in a diagonal part $`H_d`$ and the remainder $`H_r`$. We use the fact that $`\mathrm{tr}H^2`$ is invariant under the unitary transformation, therefore we have
$$\mathrm{tr}H_d^2+\mathrm{tr}H_r^2=\mathrm{tr}H^2=\text{const}.$$
(2.15)
This means that $`\mathrm{tr}H_r^2`$ falls monotonically if $`\mathrm{tr}H_d^2`$ increases. One can use Eq. (2.14) to obtain
$$\frac{d\mathrm{tr}H_d^2}{ds}=\frac{d}{ds}\underset{i}{}H_{ii}^2=2\underset{i}{}H_{ii}\underset{j}{}\left(T_{ij}H_{ji}H_{ij}T_{ji}\right)=2\underset{ij}{}T_{ji}H_{ji}(H_{jj}H_{ii}).$$
(2.16)
In order to ensure that $`_{ij}H_{ij}^2`$ falls monotonically, one can simply choose the generator as $`T_{ji}=H_{ji}(H_{jj}H_{ii})`$ or
$$T(s)=[H_d(s),H_r(s)].$$
(2.17)
One can show by making use of this definition that,
$`{\displaystyle \frac{dH_{ij}(s)}{ds}}`$ $`=`$ $`{\displaystyle \underset{j}{}}\left(H_{ii}(s)+H_{jj}(s)2H_{kk}(s)\right)H_{ik}(s)H_{kj}(s),`$
$`{\displaystyle \frac{d}{ds}}{\displaystyle \underset{ij}{}}H_{ij}^2`$ $`=`$ $`{\displaystyle \frac{d}{ds}}{\displaystyle \underset{k}{}}H_{kk}^2=2{\displaystyle \underset{ij}{}}T_{ij}^2.`$ (2.18)
Because $`_{ij}H_{ij}^2`$ falls monotonously and is restricted from below, When $`s\mathrm{}`$ the derivative vanishes and we have $`T_s0`$, at this limit the procedure of block-diagonalization is completed. At this point, the unitary transformation Eqs (2.14,2.17) is completely defined. The only freedom is in the choice of separation of the Hamiltonian into a “diagonal” and a “rest” part. Of course this depends on the given physical problem. As a illustration of the method, we show how perturbation theory can be applied in this formalism. For a given values of $`s`$ we have,
$`H(s)`$ $`=`$ $`H_d^{(0)}+H_r^{(1)}+H_r^{(2)}+\mathrm{},`$ (2.19)
$`T(s)`$ $`=`$ $`[H_d(s),H_r(s)]=[H_d^{(0)},H_r^{(1)}]+[H_d^{(0)},H_r^{(2)}]+\mathrm{}.`$ (2.20)
$`=`$ $`T^{(1)}+T^{(2)}+\mathrm{},`$
where the superscript denotes the order in the bare coupling constant. The part $`H_d^{(0)}`$ is the free Hamiltonian. The index $`r`$ denotes the rest of the Hamiltonian. Note that generally the diagonal part in the flow equation is the full particle number conserving part of the effective Hamiltonian. The choice of only $`H_d^{(0)}`$ as the diagonal part leads to the simplest band-diagonal structure where the particle number is conserved. We use the basis of the eigenfunctions of the free Hamiltonian $`H_d^{(0)}|i=E_i|i`$ to obtain the matrix elements of Eqs (2.14,2.17),
$`{\displaystyle \frac{dH_{ij}}{ds}}`$ $`=`$ $`(E_iE_j)^2H_{rij}^{(1)}+[T^{(1)},H_{rij}^{(1)}](E_iE_j)^2H_{rij}^{(2)}+\mathrm{},`$ (2.21)
$`T_{ij}`$ $`=`$ $`(E_iE_j)H_{rij}^{(1)}+(E_iE_j)H_{rij}^{(2)}+\mathrm{},`$ (2.22)
where the energy differences are given by
$$E_iE_j=\underset{k=1}{\overset{n_1}{}}E_{ik}\underset{k=1}{\overset{n_2}{}}E_{jk},$$
(2.23)
and $`E_{ik}`$ and $`E_{jk}`$ are the energies of the creation and annihilation particles, respectively. To leading order in perturbation theory, one finds
$`{\displaystyle \frac{dH_{rij}^{(1)}}{ds}}`$ $`=`$ $`(E_iE_j)^2H_{rij}^{(1)},`$
$`H_{rij}^{(1)}(s)`$ $`=`$ $`H_{rij}^{(1)}(s=0)u_{ij}(s),`$
$`u_{ij}(s)`$ $`=`$ $`e^{(E_iE_j)^2s}.`$ (2.24)
The flow-parameter $`s`$ has dimension $`1/(\text{energy})^2`$ and is related to the similarity width $`\lambda `$ (ultraviolet cutoff) by $`s=1/\lambda ^2`$. This implies that matrix elements of the interaction which change the number of particles are strongly suppressed, since we have $`|E_iE_j|>\lambda `$. The similarity generator in Wegner’s formulation corresponds to the choice of a gaussian similarity function with uniform width. In the next leading order, one has to deal separately with the diagonal and rest parts. In analogy to the Glazek-Wilson method we introduce $`H_r^{(2)}=u(s)\overline{H}_r^{(2)}(s)`$, and the solution reads,
$`\overline{H}_{rij}^{(2)}(s)`$ $`=`$ $`\overline{H}_{rij}^{(2)}(s=0)+{\displaystyle _0^s}𝑑s^{}u(s^{})[T^{(1)},H^{(1)}]_{rij}(s^{}),`$
$`H_{dij}^{(2)}(s)`$ $`=`$ $`H_{dij}^{(2)}(s=0)+{\displaystyle _0^s}𝑑s^{}[T^{(1)},H^{(1)}]_{dij}(s^{}).`$ (2.25)
It is obvious that for the non-diagonal term a smooth form factor appears in a natural way to suppress the off-diagonal interaction. In other words, the particle number changing interactions are eliminated while a new terms are produced. The procedure can be extended to arbitrarily high orders. The counterterms can be determined order-by-order using the idea of coupling coherence, namely that under similarity transformation Hamiltonian remains form invariant (see next section).
### 2.3 Coupling coherence condition
One of the most severe problems for the traditional light-front RG is that an infinite number of relevant and marginal operators are required . This is due to the fact that light-front cutoff violates the underlying symmetry, e.g., Lorentz invariance and gauge symmetries. Since these are continuous symmetries, their violation in principle leads to infinite number of symmetry violating counterterms (with new couplings), in order to maintain the symmetry of the effective Hamiltonian. In terms of the effective field theory approach, some sort of fine tuning is required to fix the strength of the new couplings so that the underlying symmetry is restored. One should note that in order to reduce the number of momentum degrees of freedom, one must introduce a real cutoff, such as a momentum cutoff or a lattice cutoff. However, dealing with divergences do not require necessarily a decrease in degrees of freedom. In fact, one may even increase the degrees of freedom, e.g., the Pauli-Villars or the dimensional regularization methods . The main idea behind the coupling coherence renormalization condition is that the Hamiltonian is form-invariant on the RG trajectory. This condition isolates and repairs the hidden symmetries .
$$H(\mathrm{\Lambda })(\mu ),$$
(2.26)
In order words, rewriting the Hamiltonian in different degrees of freedom does not change the operator itself. One may think of $`(\mu )`$ as QCD written in terms of constituent quarks and gluons and $`H(\mathrm{\Lambda })`$ as the same QCD Hamiltonian written in terms of canonical quarks and gluons, associated with partons and current quarks. The SRG and the coupled-cluster RG with coupling coherence allows one to construct effective theories with the same number of couplings as the underlying fundamental theory, even when the cutoff violate symmetries of the theory. This does not preclude the emergence of the new couplings, however, they depend on the original coupling and will vanish if the fundamental couplings are turned off. This boundary condition together with the RG equations determines their dependence on the fundamental coupling.
As an illustrative example , we consider following interaction
$$V(\varphi )=\frac{\lambda _1}{4!}\varphi _1^4+\frac{\lambda _2}{4!}\varphi _2^2+\frac{\lambda _3}{4!}\varphi _1^2\varphi _2^2,$$
(2.27)
where $`\varphi _1`$ and $`\varphi _2`$ are scalar fields. We want to investigate that under what conditions the couplings are independent of each other. Consider the Gell-Mann-Low equations up to one-loop in perturbation theory, ignoring the masses,
$`{\displaystyle \frac{\lambda _1}{t}}`$ $`=`$ $`3\xi \lambda _1^2+{\displaystyle \frac{1}{12}}\xi \lambda _3^2,`$
$`{\displaystyle \frac{\lambda _2}{t}}`$ $`=`$ $`3\xi \lambda _2^2+{\displaystyle \frac{1}{12}}\xi \lambda _3^2,`$
$`{\displaystyle \frac{\lambda _3}{t}}`$ $`=`$ $`{\displaystyle \frac{2}{3}}\xi \lambda _2^3+\xi \lambda _3(\lambda _1+\lambda _2),`$ (2.28)
where $`t=\mathrm{log}(\mathrm{\Lambda }/\mu )`$ and $`\xi =\mathrm{}/(16\pi ^2)`$. Assume that there is only one independent coupling $`\overline{\lambda }=\lambda _1`$, and $`\lambda _2`$, $`\lambda _3`$ are functions of $`\overline{\lambda }`$. Now one can simplify Eq. (2.28),
$`\left(3\overline{\lambda }^2+{\displaystyle \frac{1}{12}}\lambda _3^2\right){\displaystyle \frac{\lambda _2}{\overline{\lambda }}}`$ $`=`$ $`3\lambda _2^2+{\displaystyle \frac{1}{12}}\lambda _3^2,`$
$`\left(3\overline{\lambda }^2+{\displaystyle \frac{1}{12}}\lambda _3^2\right){\displaystyle \frac{\lambda _3}{\overline{\lambda }}}`$ $`=`$ $`{\displaystyle \frac{2}{3}}\lambda _3^2+\overline{\lambda }\lambda _3+\lambda _2\lambda _3.`$ (2.29)
In the leading order, the above equations have two distinct solutions, one is when $`\lambda _2=\overline{\lambda }`$ and $`\lambda _3=2\overline{\lambda }`$. In this case we have
$$V(\varphi )=\frac{\overline{\lambda }}{4!}\left(\varphi _1^2+\varphi _2^2\right)^2.$$
(2.30)
Therefore we recover the $`O(2)`$ symmetric theory. The other solution is $`\lambda _2=\overline{\lambda }`$ and $`\lambda _3=6\overline{\lambda }`$ which leads to two decoupled scalar fields,
$$V(\varphi )=\frac{\overline{\lambda }}{2.4!}\left((\varphi _1+\varphi _2)^4+(\varphi _1\varphi _2)^4\right),$$
(2.31)
One can conclude that $`\lambda _2`$ and $`\lambda _3`$ do not run independently with the cutoff if and only if there is a symmetry which connects their strength to $`\lambda _1`$. The condition that a limited number of couplings run with cutoff independently reveals the symmetries broken by the regulator and repairs them. More interesting, it may be used as well to uncover symmetries that are broken by the vacuum. This may reconcile the trivial vacuum in light-front field theory and vacuum symmetry breaking problem.
### 2.4 General formulation of the similarity renormalization group
In this section we pave the way for an introduction of the coupled-cluster RG, and consider the similarity renormalization group in a more general framework without requiring the unitarity. The discussion in this section is partially based on the work of Suzuki and Okamoto . Let us consider a system described by a Hamiltonian $`H(\mathrm{\Lambda })`$ which has, at the very beginning, a large cut-off $`\mathrm{\Lambda }`$. We assume that the renormalized Hamiltonian $`H^{\text{eff}}(\mathrm{\Lambda })`$ up to scale $`\mathrm{\Lambda }`$ can be written as the sum of the canonical Hamiltonian and a “counterterm” $`H_C(\mathrm{\Lambda })`$,
$$H^{\text{eff}}(\mathrm{\Lambda })=H(\mathrm{\Lambda })+H_C(\mathrm{\Lambda }).$$
(2.32)
Our aim is to construct the renormalized Hamiltonian by obtaining this counterterm. Now imagine that we restrict the Hamiltonian to a lower energy scale $`(\mu )`$, where we want to find an effective Hamiltonian $`H^{\text{eff}}(\mu )`$ which has the same energy spectrum as the original Hamiltonian in the smaller space. The cut-off $`\mu `$ can be conceived as a flow parameter. The value of $`\mu =\mathrm{\Lambda }`$ corresponds to the initial bare regulated Hamiltonian. Formally, we wish to transform the Hamiltonian to a new basis, where the medium-energy modes $`\mu <k<\mathrm{\Lambda }`$, decouple from the low-energy ones, while the low-energy spectrum remains unchanged. We split the Hilbert space by means of flow-parameter $`\mu `$ into two subspaces, the intermediate-energy space $`Q`$ containing modes with $`\mu <k<\mathrm{\Lambda }`$ and a low-energy space $`P`$ with $`k\mu `$. Our renormalization approach is based on decoupling of the complement space $`Q`$ from the model space $`P`$. Thereby the decoupling transformation generates a new effective interaction $`\delta H(\mu ,\mathrm{\Lambda })`$ containing the effects of physics between the scales $`\mathrm{\Lambda }`$ and $`\mu `$. One can then determine the counterterm by requiring coupling coherence , namely that the transformed Hamiltonian has the same form as the original one but with $`\mathrm{\Lambda }`$ replaced by $`\mu `$ everywhere. (This is in contrast to the popular Effective Field Theory approach, where one includes all permissible couplings of a given order and fixes them by requiring observable computed be both cutoff-independent and Lorentz covariant.)
We define two projector operators, also called $`P`$ and $`Q`$, which project a state onto the model space and its complement, satisfy $`P^2=P`$, $`Q^2=Q`$, $`PQ=0`$ and $`P+Q=1`$. We introduce an isometry operator $`G`$ which maps states in the $`P`$\- onto the $`Q`$\- space,
$$|q=G|p(|qQ,|pP).$$
(2.33)
The operator $`G`$ is the basic ingredient in a family of “integrating-out operators”, and passes information about the correlations of the high energy modes to the low-energy space. The operator $`G`$ obeys $`G=QGP`$, $`GQ=0`$, $`PG=0`$ and $`G^n=0`$ for $`n2`$. The counterintuitive choice that $`G`$ maps from model to complement space is due to the definition Eq. (2.36) below (c.f. the relation between the active and passive view of rotations). In order to give a general form of the effective low-energy Hamiltonian, we define another operator $`X(n,\mu ,\mathrm{\Lambda })`$,
$$X(n,\mu ,\mathrm{\Lambda })=(1+G)(1+G^{}G+GG^{})^n.$$
(2.34)
($`n`$ is a real number.) The inverse of $`X(n,\mu ,\mathrm{\Lambda })`$ can be obtained explicitly,
$$X^1(n,\mu ,\mathrm{\Lambda })=(1+G^{}G+GG^{})^n(1G).$$
(2.35)
The special case $`n=0`$ is equivalent to the transformation introduced in ref. to relate the hermitian and non-hermitian effective operators in the energy-independent Suzuki-Lee approach. We now consider the transformation of $`H(\mathrm{\Lambda })`$ defined as
$$\overline{H}(n,\mu ,\mathrm{\Lambda })=X^1(n,\mu ,\mathrm{\Lambda })H(\mathrm{\Lambda })X(n,\mu ,\mathrm{\Lambda }),$$
(2.36)
where we have
$$H(\mathrm{\Lambda })\overline{H}(n,\mu ,\mathrm{\Lambda })H(\mu )+\delta H(\mu ,\mathrm{\Lambda }).$$
(2.37)
One can prove that if $`\overline{H}(n,\mu ,\mathrm{\Lambda })`$ satisfies the desirable decoupling property,
$$Q\overline{H}(n,\mu ,\mathrm{\Lambda })P=0,$$
(2.38)
or more explicitly, by substituting the definition of $`X(n,\mu ,\mathrm{\Lambda })`$ and $`X^1(n,\mu ,\mathrm{\Lambda })`$ from Eqs. (2.34)–(2.35),
$$QH(\mathrm{\Lambda })P+QH(\mathrm{\Lambda })QGGPH(\mathrm{\Lambda })PGPH(\mathrm{\Lambda })QG=0,$$
(2.39)
that $`H^{\text{eff}}(\mu )(n,\mu )=P\overline{H}(n,\mu ,\mathrm{\Lambda })P`$ is an effective Hamiltonian for the low energy degrees of freedom. In other words, it should have the same low-energy eigenvalues as the original Hamiltonian. The proof is as follows:
Consider an eigenvalue equation in the $`P`$ space for a state $`|\varphi (k)P`$,
$$P\overline{H}(n,\mu ,\mathrm{\Lambda })P|\varphi (k)=E_kPX^1(n,\mu ,\mathrm{\Lambda })X(n,\mu ,\mathrm{\Lambda })P|\varphi (k).$$
(2.40)
By multiplying both sides by $`X(n,\mu ,\mathrm{\Lambda })`$ and making use of the decoupling property Eq. (2.38), we obtain
$$H(\mathrm{\Lambda })X(n,\mu ,\mathrm{\Lambda })P|\varphi (k)=E_kX(n,\mu ,\mathrm{\Lambda })P|\varphi (k).$$
(2.41)
This equation means that $`E_k`$ in Eq. (2.40) agrees with one of the eigenvalue of $`H(\mathrm{\Lambda })`$ and $`X(n,\mu ,\mathrm{\Lambda })P|\varphi (k)`$ is the corresponding eigenstate. Now we demand that
$$H^{\text{eff}}(\mu )H(\mu )+H_C(\mu ).$$
(2.42)
This requirement uniquely determines the counterterm $`H_C`$.
In the same way we can also obtain the $`Q`$-space effective Hamiltonian, from the definition of $`\overline{H}(n,\mu ,\mathrm{\Lambda })`$. It can be seen that if $`G`$ satisfies the requirement in Eq. (2.39), then we have additional decoupling condition
$$P\overline{H}(n,\mu ,\mathrm{\Lambda })Q=0.$$
(2.43)
Although the above condition is not independent of the decoupling condition previously introduced in Eq. (2.38) in the exact form, but this is not the case after involvement of some approximation (truncation). We will argue later that *both* of the decoupling conditions Eqs. (2.38) and (2.43) are necessary in order to have a sector-independent renormalization scheme. The word “sector” here means the given truncated Fock space. Let us now clarify the meaning of this concept. To maintain the generality of the previous discussion, we use here the well known Bloch-Feshbach formalism . The Bloch-Feshbach method exploits projection operators in the Hilbert space in order to determine effective operators in some restricted model space. This technique seems to be more universal than Wilson’s renormalization formulated in a Lagrangian framework. This is due to the fact that in the Bloch-Feshbach formalism, other irrelevant degrees of freedom (such as high angular momentum, spin degrees of freedom, number of particles, etc.) can be systematically eliminated in the same fashion.
Assume that the full space Schrödinger equation is $`H|\psi =E_\psi |\psi `$ and for simplicity $`|\psi `$ has been normalized to one. The similarity transformed Schrödinger equation now reads $`\overline{H}|\psi _{X^1}=E_\psi |\psi _{X^1}`$, where we defined $`|\psi _{X^1}=X^1|\psi `$ and $`\overline{H}`$ is a similarity transformed Hamiltonian. This equation is now separated into two coupled equations for $`P`$\- and $`Q`$-space.
$`(E_\psi P\overline{H}P)P|\psi _{X^1}=P\overline{H}Q|\psi _{X^1},`$ (2.44)
$`(E_\psi Q\overline{H}Q)Q|\psi _{X^1}=Q\overline{H}P|\psi _{X^1}.`$ (2.45)
We may formally solve Eq. (2.45) as
$$Q|\psi _{X^1}=\frac{Q\overline{H}P}{E_\psi Q\overline{H}Q}P|\psi _{X^1},$$
(2.46)
and substitute this equation into Eq. (2.44) to obtain a formally exact uncoupled equation in $`P`$-space,
$$H^{\text{eff}}P|\psi _{X^1}=E_\psi P|\psi _{X^1},$$
(2.47)
where we have
$$H^{\text{eff}}=P\overline{H}P+P\overline{H}Q\frac{1}{E_\psi Q\overline{H}Q}Q\overline{H}P.$$
(2.48)
The effective Hamiltonian $`H^{\text{eff}}`$ constructed in this fashion is explicitly energy dependent. This equation resembles the one for Brueckner’s reaction matrix (or “G”-matrix) equation in nuclear many-body theory (NMT). In the same way for arbitrary operator $`O`$ (after a potential similarity transformation), we construct the effective operator. Let us define a similarity transformed operator $`\overline{O}`$,
$`\overline{O}={\displaystyle |\varphi _{x^1}\varphi _{x^1}|\overline{O}|\psi _{x^1}\psi _{x^1}|},`$
$`\overline{O}={\displaystyle |\varphi _{x^1}\varphi _{x^1}|(P+Q)\overline{O}(P+Q)|\psi _{x^1}\psi _{x^1}|},`$ (2.49)
where $`|\psi `$ and $`|\varphi `$ are eigen functions of Hamiltonian and we have $`|\varphi _{x^1}\varphi _{x^1}|=|\psi _{x^1}\psi _{x^1}|=1`$ and $`P+Q=1`$. We now write $`\varphi _{x^1}|Q`$ and $`Q|\psi _{x^1}`$ in terms of their solution in the $`P`$-space obtained in Eq. (2.46). One can immediately show
$`\varphi _{x^1}|P\overline{O}Q|\psi _{x^1}=\varphi _{x^1}|P\overline{O}Q{\displaystyle \frac{Q\overline{H}P}{E_\psi Q\overline{H}Q}}P|\psi _{x^1},`$
$`\varphi _{x^1}|Q\overline{O}P|\psi _{x^1}=\varphi _{x^1}|P{\displaystyle \frac{P\overline{H}Q}{E_\varphi Q\overline{H}Q}}Q\overline{O}P|\psi _{x^1},`$
$`\varphi _{x^1}|Q\overline{O}Q|\psi _{x^1}=\varphi _{x^1}|P{\displaystyle \frac{P\overline{H}Q}{E_\varphi Q\overline{H}Q}}Q\overline{O}Q{\displaystyle \frac{Q\overline{H}P}{E_\psi Q\overline{H}Q}}P|\psi _{x^1}.`$ (2.50)
By plugging the above equations into Eq. (2.49), one can obtain the effective operator in the $`P`$-space
$$\overline{O}=|\varphi _{x^1}\varphi _{x^1}|PO^{\text{eff}}P|\psi _{x^1}\psi _{x^1}|=PO^{\text{eff}}P,$$
(2.51)
where we have
$`O^{\text{eff}}`$ $`=`$ $`P\overline{O}P+P\overline{H}Q{\displaystyle \frac{1}{E_\varphi Q\overline{H}Q}}Q\overline{O}P+P\overline{O}Q{\displaystyle \frac{1}{E_\psi Q\overline{H}Q}}Q\overline{H}P`$
$`+`$ $`P\overline{H}Q{\displaystyle \frac{1}{E_\varphi Q\overline{H}Q}}Q\overline{O}Q{\displaystyle \frac{1}{E_\psi Q\overline{H}Q}}Q\overline{H}P.`$
Notice that Eq. (2.4) can be converted into the form of Eq. (2.48) with the effective Schrödinger equation Eq. (2.47) when $`\overline{O}\overline{H}`$ <sup>1</sup><sup>1</sup>1To this end, we organize Eq. (2.4) when $`\overline{O}\overline{H}`$, $`O^{\text{eff}}`$ $`=`$ $`P\overline{O}P+P\overline{H}Q\left({\displaystyle \frac{1}{E_\varphi Q\overline{H}Q}}+{\displaystyle \frac{1}{E_\psi Q\overline{H}Q}}+{\displaystyle \frac{1}{E_\varphi Q\overline{H}Q}}Q\overline{H}Q{\displaystyle \frac{1}{E_\psi Q\overline{H}Q}}\right)Q\overline{H}P,`$ (2.53) $`=`$ $`P\overline{O}P+P\overline{H}Q\left({\displaystyle \frac{1}{E_\psi Q\overline{H}Q}}+{\displaystyle \frac{E_\psi }{(E_\varphi Q\overline{H}Q)(E_\psi Q\overline{H}Q)}}\right)Q\overline{H}P.`$ One now can use Eq. (2.46) to rewrite the last part of the above equation into the form of $`\varphi _{x^1}|P\overline{H}Q\left({\displaystyle \frac{E_\psi }{(E_\varphi Q\overline{H}Q)(E_\psi Q\overline{H}Q)}}\right)Q\overline{H}P|\psi _{x^1}`$ $`=`$ $`E_\psi \varphi _{x^1}|Q^2|\psi _{x^1}`$ $`=`$ $`E_\psi \left(\varphi _{x^1}|\psi _{x^1}\varphi _{x^1}|P|\psi _{x^1}\right),`$ where we used $`Q+P=1`$ and $`Q^2=Q`$. Having made used of Eqs. (2.51,2.53,1), one can immediately obtain Eq. (2.47) with the effective Hamiltonian in the form of Eq. (2.48)..
The $`E`$-dependence in Eqs. (2.48) and (2.4) emerges from the fact that the effective interaction in the reduced space is not assumed to be decoupled from the excluded space. However, by using the decoupling conditions introduced in Eqs. (2.38) and (2.43), we observe that energy dependence can be removed, and the effective operators become
$`H^{\text{eff}}`$ $`=`$ $`P\overline{H}P=(n,\mu ),`$
$`O^{\text{eff}}`$ $`=`$ $`P\overline{O}P=𝒪(n,\mu ).`$ (2.55)
The decoupling property makes the operators in one sector independent of the other sector. The effects of the excluded sector is taken into account by imposing the decoupling conditions. This is closely related to the folded diagram method in NMT for removing energy-dependence . (It is well-known in NMT that $`E`$-dependence in the $`G`$-matrix emerges from non-folded diagrams which can be systematically eliminated using the effective interaction approach). The above argument was given without assuming an explicit form for $`X`$ and thus the decoupling conditions are more fundamental than the prescription used to derive these conditions. We will show later that one can choose a transformation $`X`$, together with the model space and its complement, which avoids the occurrence of “small energy denominators”. We now show that Lorentz covariance in a given sector does not hinge on a special form of similarity transformation. Assume the existence of ten Poincaré generators $`L_i`$ satisfying
$$[L_i,L_j]=a_{ij}^kL_k,$$
(2.56)
where $`a_{ij}^k`$ are the structure coefficients. One can show that if the operators $`L_i`$ satisfy the decoupling conditions $`Q\overline{L}_iP=0`$ and $`P\overline{L}_iQ=0`$ it follows that
$$[L_i^{\text{eff}},L_j^{\text{eff}}]=a_{ij}^kL_k^{\text{eff}}.$$
(2.57)
This leads to a relativistic description even after simultaneously integrating out the high-frequency modes and reducing the number of particles. Indeed we conjecture that requiring the decoupling conditions makes the effective Hamiltonian free of Lorentz-noninvariant operators for a given truncated sector regardless of the regularization scheme. However, one may still be faced with an effective Hamiltonian which violates gauge invariance (for e.g., when sharp cutoff is employed).
Note that the solution to Eq. (2.39) is independent of the number $`n`$. One can make use of Eq. (2.39) and its complex conjugate to show that for any real number $`n`$, the following relation for the effective low-energy Hamiltonian
$$(n,\mu )=^{}(n1,\mu ).$$
(2.58)
The case $`n=1/2`$ is special since the effective Hamiltonian is hermitian,
$$(1/2,\mu )=(P+G^{}G)^{1/2}H(\mathrm{\Lambda })(P+G)(P+G^{}G)^{1/2}.$$
(2.59)
Hermiticity can be verified from the relation
$$e^TP=(1+G)(P+G^{}G)^{1/2},$$
(2.60)
where,
$$T=\mathrm{arctan}(GG^{})=\underset{n=0}{\overset{\mathrm{}}{}}\frac{(1)^n}{2n+1}(G(G^{}G)^n\text{h.c}.).$$
(2.61)
Since the operator $`T`$ is anti-hermitian, $`e^T`$ is a unitary operator. From the above expression Eq. (2.59) can be written in the explicitly hermitian form
$$(\frac{1}{2},\mu )=Pe^TH(\mathrm{\Lambda })e^TP.$$
(2.62)
As was already emphasized, renormalization based on unitary transformations is more complicated and non-economical. Thus we will explore a non-unitary approach. An interesting non-hermitian effective low-energy Hamiltonian can be obtained for $`n=0`$,
$$(0,\mu ,\mathrm{\Lambda })=PH(\mathrm{\Lambda })(P+QG).$$
(2.63)
This form resembles the Bloch and Horowitz type of effective Hamiltonian as used in NMT , and was the one used by Wilson in his original work on quantum field theory (see section II). In the context of the CCM, this form leads to the folded diagram expansion well known in many-body theory . It is of interest that various effective low-energy Hamiltonians can be constructed according to Eq. (2.55) by the use of the mapping operator $`G`$ which all obey the decoupling property Eq. (2.38) and Eq. (2.43). Neither perturbation theory nor hermiticity is essential for this large class of effective Hamiltonians.
### 2.5 The coupled-cluster renormalization group
The CCM approach is, of course, just one of the ways of describing the relevant spectrum by means of non-unitary transformations. According to our prescription the model space is $`P:\{|L|0,b_h,L\mu \}`$, where $`|0,b_h`$ is a bare high energy vacuum (the ground state of high-momentum of free-Hamiltonian) which is annihilated by all the high frequency annihilation operators $`\{C_I\}`$ (for a given quantization scheme, e.g., equal time or light-cone) , the set of indices $`\{I\}`$ therefore defines a subsystem, or cluster, within the full system of a given configuration. The actual choice depends upon the particular system under consideration. In general, the operators $`\{C_I\}`$ will be products (or sums of products) of suitable single-particle operators. We assume that the annihilation and its corresponding creation $`\{C_I^{}\}`$ subalgebras and the state $`|0,b_h`$ are cyclic, so that the linear combination of state $`\{C_I^{}|0,b_h\}`$ and $`\{{}_{h}{}^{}b,0|C_I\}`$ span the decoupled Hilbert space of the high-momentum modes, $`\{|H\}`$, where $`\mu <H<\mathrm{\Lambda }`$. It is also convenient, but not necessary, to impose the orthogonality condition, $`0|C_IC_J^{}|0=\delta (I,J)`$, where $`\delta (I,J)`$ is a Kronecker delta symbol. The complement space is $`Q:\{|L\left(|H|0,b_h\right)\}`$. Our main goal is to decouple the $`P`$-space from the $`Q`$-space. This gives sense to the partial diagonalization of the high-energy part of the Hamiltonian. The states in full Hilbert space are constructed by adding correlated clusters of high-energy modes onto the $`P`$-space, or equivalently integrating out the high-energy modes from the Hamiltonian,
$`|f`$ $`=`$ $`X(\mu ,\mathrm{\Lambda })|p=e^{\widehat{S}}e^{\widehat{S}^{}}|0,b_h|L=e^{\widehat{S}}|0,b_h|L,`$ (2.64)
$`\stackrel{~}{f}|`$ $`=`$ $`L|{}_{h}{}^{}b,0|X^1(\mu ,\mathrm{\Lambda })=L|{}_{h}{}^{}0|e^{\widehat{S}^{}}e^{\widehat{S}},`$ (2.65)
where the operators $`X(\mu ,\mathrm{\Lambda })`$ and $`X^1(\mu ,\mathrm{\Lambda })`$ have been expanded in terms of independent coupled cluster excitations $`I`$,
$$\begin{array}{cccccc}\hfill \widehat{S}& =& \underset{m=0}{}\widehat{S}_m\left(\frac{\mu }{\mathrm{\Lambda }}\right)^m,\hfill & \hfill \widehat{S}_m& =& \underset{I}{^{}}\widehat{s}_I^mC_I^{},\hfill \\ \hfill \widehat{S}^{}& =& \underset{m=0}{}\widehat{S}_m^{}\left(\frac{\mu }{\mathrm{\Lambda }}\right)^m,\hfill & \hfill \widehat{S}_m^{}& =& \underset{I}{^{}}\widehat{s}_I^mC_I.\hfill \end{array}$$
(2.66)
Here the primed sum means that at least one fast particle is created or destroyed $`(I0)`$, and momentum conservation in $`PQ`$ is included in $`\widehat{s}_I`$ and $`\widehat{s}_I^{}`$. $`\widehat{S}_m(\widehat{S}_m^{})`$ are not generally commutable in the low-energy Fock space, whereas they are by construction commutable in the high-energy Fock space. One of the crucial difference between our approach and the traditional CCM is that, here $`\widehat{s}_I^m`$ and $`\widehat{s}_I^m`$ are only $`c`$-number in the high-energy Fock space, but they are operators in the low-energy Fock space (unless when $`\mu 0`$). We have in the Wilsonian high-energy shell $`[\widehat{S},C_I^{}]=[\widehat{S}^{},C_I]=0`$ (see Fig. 2.2). Therefore, one can guarantee the proper size-extensivity and consequently conformity with the linked-cluster theorem at any level of approximation in the Wilsonian high-energy shell. Nevertheless one can still apply the standard CCM to the induced effective low-energy Hamiltonian to extend the proper size-extensivity to the entire Fock space.
It is immediately clear that states in the interacting Hilbert space are normalized, $`\stackrel{~}{f}|f={}_{h}{}^{}b,0|0,b_{h}^{}=1`$. We have two types of parameters in this procedure: One is the coupling constant of the theory ($`\lambda `$), and the other is the ratio of cutoffs ($`\mu /\mathrm{\Lambda }`$). The explicit power counting makes the degree of divergence of each order smaller than the previous one. According to our logic, Eq. (2.63) can be written as
$$(\mu )=_hb,0|X^1(\mu ,\mathrm{\Lambda })H(\mathrm{\Lambda })X(\mu ,\mathrm{\Lambda })|0,b_h,$$
(2.67)
with $`X(\mu ,\mathrm{\Lambda })`$ and $`X^1(\mu ,\mathrm{\Lambda })`$ defined in Eq. (2.64) and Eq. (2.65). We require that effective Hamiltonian $`(\mu )`$ obtained in this way remains form invariant or coherent . This requirement satisfies on an infinitely long renormalization group trajectory and thus does constitutes a renormalized Hamiltonian. Thereby one can readily identify the counter terms produced from expansion of Eq. (2.67).
The individual amplitudes for a given $`m`$, $`\{\widehat{s}_I^m,\widehat{s}^{}{}_{I}{}^{m}\}\{\widehat{s}_I,\widehat{s}_I^{}\}_m`$, have to be fixed by the dynamics of quantum system. This is a complicated set of requirements. However, we require less than that. Suppose that after a similar transformation of Hamiltonian, $`\overline{H}`$, we obtain an effective Hamiltonian of the form
$$\overline{H}=H(\text{low})+H_{\text{free}}(\text{high})+C_I^{}V_{IJ}C_J,$$
(2.68)
where $`V_{IJ}`$ is an arbitrary operator in the low frequency space. The $`I`$ and $`J`$ indices should be chosen such that the last term in Eq. (2.68) contains at least one creation- operator and one annihilation-operator of high frequency. By using Rayleigh-Schrödinger perturbation theory, it can be shown that the free high-energy vacuum state of $`H_{\text{free}}(\text{high})`$ is annihilated by Eq. (2.68) and remains without correction at any order of perturbation theory. Having said that, we will now consider how to find the individual amplitudes $`\{\widehat{s}_I,\widehat{s}_I^{}\}_m`$ that transfer the Hamiltonian into the form Eq. (2.68). We split the Hamiltonian in five parts:
$$H=H_1+H_2^{\text{free}}(\text{high})+V_C(C_I^{})+V_A(C_I)+V_B,$$
(2.69)
where $`H_1`$ contains only the low frequency modes with $`k\mu `$, $`H_2`$ is the free Hamiltonian for all modes with $`\mu <k<\mathrm{\Lambda }`$, $`V_C`$ contains low frequency operators and products of the high frequency creation operators $`C_I^{}`$ and $`V_A`$ is the hermitian conjugate of $`V_C`$. The remaining terms are contained in $`V_B`$, these terms contain at least one annihilation and creation operators of the high energy modes. Our goal is to eliminate $`V_C`$ and $`V_A`$ since $`V_B`$ annihilates the vacuum. The ket-state coefficients $`\{\widehat{s}_I\}_m`$ are worked out via the ket-state Schrödinger equation $`H(\mathrm{\Lambda })|f=E|f`$ written in the form
$$0|C_Ie^{\widehat{S}}He^{\widehat{S}}|0=0,I0.$$
(2.70)
The bra-state coefficients $`\{\widehat{s}_I,\widehat{s}_I^{}\}_m`$ are obtained by making use of the Schrödinger equation defined for the bra-state, $`\stackrel{~}{f}|H(\mathrm{\Lambda })=\stackrel{~}{f}|E`$. First we project both sides on $`C_I^{}|0`$, then we eliminate $`E`$ by making use of the ket-state equation projection with the state $`0|e^{\widehat{S}^{}}C_I^{}`$ to yield the equations
$$0|e^{\widehat{S}^{}}e^{\widehat{S}}[H,C_I^{}]e^{\widehat{S}}e^{\widehat{S}^{}}|0=0,I0.$$
(2.71)
Alternatively one can in a unified way apply $`e^{\widehat{S}}e^{\widehat{S}^{}}C_I^{}|0`$ on the Schrödinger equation for the bra-state and obtain
$$0|e^{\widehat{S}^{}}e^{\widehat{S}}He^{\widehat{S}}e^{\widehat{S}^{}}C_I^{}|0=0,I0.$$
(2.72)
Equation (2.70) and Eqs. (2.71) or (2.72) provide two sets of formally exact, microscopic, operatorial coupled non-linear equations for the ket and bra. One can solve the coupled equations in Eq. (2.70) to work out $`\{\widehat{s}_I\}_m`$ and then use them as an input in Eqs. (2.71) or (2.72).
It is important to notice that Eqs. (2.70) and (2.71) can be also derived by requiring that the effective low-energy Hamiltonian defined in Eq. (2.67), be stationary (i.e. $`\delta (\mu )=0`$) with respect to all variations in each of the independent functional $`\{\widehat{s}_I,\widehat{s}_I^{}\}_m`$. One can easily verify that the requirements $`\delta (\mu )/\delta \widehat{s}_I^m=0`$ and $`\delta (\mu )/\delta \widehat{s}_I^m=0`$ yield Eqs. (2.71) and (2.70). The combination of Eqs. (2.70) and (2.71) does not manifestly satisfy the decoupling property as set out in Eqs. (2.38) and (2.43). On the other hand Eqs. (2.70) and (2.72) satisfy these conditions. Equations (2.70) and (2.72) imply that all interactions including creation and annihilation of fast particles (“$`I`$”) are eliminated from the transformed Hamiltonian $`(\mu )`$ in Eq. (2.67). In other words, these are decoupling conditions leading to the elimination of $`V_C`$ and $`V_A`$ from Eq. (2.69), which is, in essence, a block-diagonalization. Therefore it makes sense for our purpose to use Eqs. (2.70) and (2.72) for obtaining the unknown coefficients. We postpone the discussion of the connection between decoupling and variational equations in next section.
So far everything has been introduced rigorously without invoking any approximation. In practice one needs to truncate both sets of coefficients $`\{\widehat{s}_I,\widehat{s}_I^{}\}_m`$ at a given order of $`m`$. A consistent truncation scheme is the so-called SUB($`𝒩,`$) scheme, where the $`n`$-body partition of the operator $`\{\widehat{S},\widehat{S}^{}\}`$ is truncated so that one sets the higher partition with $`I>𝒩`$ to zero up to a given accuracy $`m=`$. Notice that, Eqs. (2.71) and (2.72) provide two equivalent sets of equations in the exact form, however after the truncation they can in principle be different. Eqs. (2.70) and (2.72) are compatible with the decoupling property at any level of the truncation, whereas Eqs. (2.70) combined with (2.71) are fully consistent with HFT at any level of truncation. Thus the low-energy effective form of an arbitrary operator can be computed according to Eq. (2.55) in the same truncation scheme used for the renormalization of the Hamiltonian. In particular, we will show that only in the lowest order ($`m=0`$), equations (2.71) and (2.72) are equivalent, independent of the physical system and the truncation scheme.
Although our method is non-perturbative, perturbation theory can be recovered from it. In this way, its simple structure for loop expansion will be obvious and we will observe that at lower order hermiticity is preserved. Now we illustrate how this is realizable in our approach. Assume that $`V_C`$ and $`V_A`$ are of order $`\lambda `$, we will diagonalize the Hamiltonian, at leading order in $`\lambda `$ up to the desired accuracy in $`\mu /\mathrm{\Lambda }`$. We use the commutator-expansion Eq. (1.18) to organize Eq. (2.70) perturbatively in order of $`m`$, aiming at elimination of the high momenta degree of freedom up to the first order in the coupling constant, thus yields
$`m=0:`$ $`0|C_I(V_C+[H_2,\widehat{S}_0])|0=0,`$
$`m=1:`$ $`0|C_I([H_1,\widehat{S}_0]+[H_2,\widehat{S}_1]+[V_A,\widehat{S}_1]+[V_C,\widehat{S}_1])|0=0,`$
$`\mathrm{}`$
$`m=n:`$ $`0|C_I([H_1,\widehat{S}_{n1}]+[H_2,\widehat{S}_n]+[V_A,\widehat{S}_n]+[V_C,\widehat{S}_n])|0=0,`$ (2.73)
where $`I0`$. Notice that $`\widehat{S}_0`$ is chosen to cancel $`V_C`$ in the effective Hamiltonian, hence it is at least of order of $`\lambda `$, consequently it generates a new term $`[H_1,\widehat{S}_0]`$ which is of higher order in $`\mu /\mathrm{\Lambda }`$ and can be canceled out on the next orders by $`\widehat{S}_1`$. The logic for obtaining the equations above is based on the fact that $`\widehat{S}_n`$ should be smaller than $`\widehat{S}_{n1}`$ (for sake of convergence) and that the equations should be consistent with each other. Since $`H_2,V_A,V_C\mathrm{\Lambda }`$ and $`H_1\mu `$, from Eq. (2.73) we have the desired relation $`\widehat{S}_n\frac{\mu }{\mathrm{\Lambda }}\widehat{S}_{n1}`$. Notice that if $`\mu \mathrm{\Lambda }`$ then one needs to keep all coupled equations in Eq. (2.73) up to $`m=n`$. In other words, there is no perturbative expansion in ratio of cutoffs. This may introduce small-energy denominator i. e, $`1/(\mathrm{\Lambda }\mu )`$, since one may obtain the solution of these equations, e. g., $`S_0`$ in form of a geometrical series which can be resummed and leads to a small-energy denominator <sup>2</sup><sup>2</sup>2Thanks to Prof. F. Wegner, for bringing this point to my attention.. Having said that as far as the renormalization is concern we are always interested on condition that $`\mathrm{\Lambda }>>\mu `$. Furthermore, for $`\mu \mathrm{\Lambda }`$ one should resort to the non-perturbative decoupling equations which are by construction free of small-energy denominator. The same procedure can be applied for Eq. (2.72) which leads to the introduction of a new series of equations in order of $`m`$,
$`m=0:`$ $`0|(V_A[H_2,\widehat{S}_0^{}])C_I^{}|0=0,`$
$`m=1:`$ $`0|([H_1,\widehat{S}_0^{}]+[H_2,\widehat{S}_1^{}]+[V_C,\widehat{S}_1^{}]+[V_A,\widehat{S}_1^{}][V_A,\widehat{S}_1])C_I^{}|0=0,`$
$`\mathrm{}`$
$`m=n:`$ $`0|([H_1,\widehat{S}_{n1}^{}]+[H_2,\widehat{S}_n^{}]+[V_C,\widehat{S}_n^{}]+[V_A,\widehat{S}_n^{}][V_A,\widehat{S}_n])C_I^{}|0=0.`$ (2.74)
Alternatively, we can use Eq. (2.71) to yield the equations
$`m=0:`$ $`0|\left([V_A,C_I^{}][[H_2,C_I^{}],\widehat{S}_0^{}]\right)|0=0,`$
$`m=1:`$ $`0|\left([[V_A,C_I^{}],\widehat{S}_1][[H_2,C_I^{}],\widehat{S}_1^{}][[V_A,C_I^{}],\widehat{S}_1^{}]\right)|0=0,`$
$`\mathrm{}`$
$`m=n:`$ $`0|([V_A,C_I^{}],\widehat{S}_n][[H_2,C_I^{}],\widehat{S}_n^{}][[V_A,C_I^{}],\widehat{S}_n^{}])|0=0.`$ (2.75)
It is obvious that at order $`m=0`$, Eqs. (2.74) and (2.75) are the same and $`\widehat{S}_0^{}=\widehat{S}_0^{}`$, which indicates that the similarity transformation at this level remains unitary. It should be noted that diagonalization at first order in the coupling constant introduces a low-energy effective Hamiltonian in Eq. (2.67) which is valid up to the order $`\lambda ^3`$. In the same way, diagonalization at second order in $`\lambda `$ modifies the Hamiltonian at order $`\lambda ^4`$ and leads generally to a non-unitarity transformation. In this way one can proceed to diagonalize the Hamiltonian at a given order in $`\lambda `$ with desired accuracy in $`\mu /\mathrm{\Lambda }`$ . Finally, the renormalization process is completed by introducing the correct $`Z(\mathrm{\Lambda })`$ factors which redefine the divergences emerging from Eq. (2.67).
### 2.6 The decoupling conditions versus the variational principle
In section 2.3 we introduced the effective low-energy phase space $`\{\widehat{s}_I,\widehat{s}_I^{}\}_m`$ induced by integrating out the high-energy modes. We argued later that the induced low-energy phase space can be obtained either from Eqs. (2.70, 2.72) which are manifestly compatible with the decoupling conditions or from Eqs. (2.70, 2.71) compatible with the variational principle and the Hellmann-Feynman theorem. Here we will show that in fact this two formulations are related through an isomorphic transformation (or in a strict sense by a symplectomorphism). In what follows, for simplicity, we absorb the factor $`(\frac{\mu }{\mathrm{\Lambda }})^m`$ in Eq. (2.66) into the definition of $`\widehat{S}_m`$ and $`\widehat{S}_m^{}`$, hence we assume $`\widehat{S}_m(\widehat{S}_m^{})(\frac{\mu }{\mathrm{\Lambda }})^m`$, therefore we have desirable convergence relation $`\widehat{S}_{m+1}(\widehat{S}_{m+1}^{})\frac{\mu }{\mathrm{\Lambda }}\widehat{S}_m(\widehat{S}_m^{})`$. In the spirit of Schrödinger representation in quantum field theory, one may now define a generalized many-body ket and bra wave functional at a given scale $`\mu `$,
$$|\widehat{\psi _\mu }=e^{\widehat{S}(\mu ,\mathrm{\Lambda })}e^{\widehat{S}^{}(\mu ,\mathrm{\Lambda })}|0=e^{\widehat{S}(\mu ,\mathrm{\Lambda })}|0,\widehat{\stackrel{~}{\psi }}_\mu |=0|e^{\widehat{S}^{}(\mu ,\mathrm{\Lambda })}e^{\widehat{S}(\mu ,\mathrm{\Lambda })},$$
(2.76)
It is clear that by construction, we have $`\widehat{\stackrel{~}{\psi }}_\mu |\widehat{\psi _\mu }=1`$ at any level of approximation. Notice that the bra and ket are parametrised independently, since they are not hermitian-adjoint of each other. Therefore, we have a biorthogonal representation of the many-body system. We will illuminate below the main underlying reasons behind this parametrization. With this definition the effective low-energy Hamiltonian Eq. (2.67) is rewritten as,
$$\widehat{}(\mu )=\widehat{\stackrel{~}{\psi }}_\mu |H|\widehat{\psi _\mu }.$$
(2.77)
Here, the bra and ket wave functional are built by adding independent clusters of high-energy correlation (where $`\mu <k<\mathrm{\Lambda }`$) in the vacuum of the free high-energy Hamiltonian, or by integrating out the high-energy modes from the Hamiltonian. The unknown low-energy operators $`\{\widehat{s}_I^m,\widehat{s}_I^m\}`$ are obtained by solving the Schrödinger equation in the high-energy shell ($`\mu <k<\mathrm{\Lambda }`$) and are decoupling equations,
$`D_1`$ $`=`$ $`Q\overline{H}P=0|C_Ie^{\widehat{S}}He^{\widehat{S}}|0=0,I0`$ (2.78)
$`D_2`$ $`=`$ $`P\overline{H}Q=0|e^{\widehat{S}^{}}e^{\widehat{S}}He^{\widehat{S}}e^{\widehat{S}^{}}C_I^{}|0=0,I0.`$ (2.79)
The decoupling equations Eqs. (2.78,2.79) imply that all interactions containing creation and annihilation of “$`I`$” high-momentum particles are eliminated from the transformed Hamiltonian while it generates new low-momentum interactions through $`\{\widehat{s}_I^m,\widehat{s}_I^m\}`$. These equations show the changes of the generalized wave functional and accordingly the effective Hamiltonian with the flow-parameter $`\mu `$. Alternatively one can obtain $`\{\widehat{s}_I^m,\widehat{s}_I^m\}`$ by requiring that the effective low-energy be stationary $`\delta \widehat{}(\mu )=0`$ with respect to all variations in each of the independent functional $`\{\widehat{s}_I^m,\widehat{s}_I^m\}`$,
$`{\displaystyle \frac{\delta \widehat{}(\mu )}{\delta \widehat{s}_I^m}}=00|e^{\widehat{S}^{}}e^{\widehat{S}}[H,C_I^{}]e^{\widehat{S}}e^{\widehat{S}^{}}|0=0,I0`$ (2.80)
$`{\displaystyle \frac{\delta \widehat{}(\mu )}{\delta \widehat{s}_I^m}}=0D_1=0`$ (2.81)
where $`D_1`$ is introduced in Eq. (2.78). Notice that we have already shown that Eq. (2.80) is as well derivable from dynamics . The decoupling conditions Eqs. (2.78, 2.79) seems generally to be in conflict with variational equations (2.80, 2.81). Here we will show that in fact this two formulations are related via an isomorphic transformation (or in a strict sense by a symplectomorphism). Let us introduce a new set of variables $`\{\widehat{\sigma }_I^m,\widehat{\overline{\sigma }}_I^m\}`$ within the low-energy phase space, by transforming the induced low-energy phase space operators set $`\{\widehat{s}_I^m,\widehat{s}_I^m\}`$ into a new set $`\{\widehat{\sigma }_I^m,\widehat{\overline{\sigma }}_I^m\}`$,
$`\{\begin{array}{cc}\widehat{\sigma }_I^m=0|C_Ie^{\widehat{S}^{}}S|0=\underset{J}{{\displaystyle ^{}}}\widehat{s}_J^m\widehat{\omega }_{JI},\hfill & \\ \widehat{\overline{\sigma }}_I^m=\widehat{s}_I^m,\hfill & \end{array}`$ (2.84)
where the low-energy operator $`\widehat{\omega }_{JI}`$ is defined as
$$\widehat{\omega }_{JI}=0|C_Ie^{\widehat{S}^{}}C_J^{}|0.$$
(2.85)
In the same way, one may conversely write $`\widehat{s}`$ in terms of $`\widehat{\sigma }`$,
$$\widehat{s}_I^m(\widehat{\sigma },\widehat{\overline{\sigma }})=\underset{J}{^{}}\widehat{\sigma }_J^m\widehat{\overline{\omega }}_{JI},$$
(2.86)
where we have used the following orthogonality property,
$`\widehat{\overline{\omega }}_{JI}=0|C_Ie^{\widehat{S}^{}}C_J^{}|0,`$
$`{\displaystyle \underset{K}{}}\widehat{\omega }_{JK}\widehat{\overline{\omega }}_{KI}={\displaystyle \underset{K}{}}\widehat{\overline{\omega }}_{JK}\widehat{\omega }_{KI}=\delta (J,I).`$ (2.87)
We use the canonical transformation Eq. (2.84) to rewrite the decoupling equations $`D_1`$ and $`D_2`$ in terms of derivative with respect to new variables, one can immediately show,
$$D_1=0|C_Ie^{\widehat{S}^{}}e^{\widehat{S}}H(\mathrm{\Lambda })e^{\widehat{S}}e^{\widehat{S}^{}}|0=\frac{\delta \widehat{}(\mu )}{\delta \widehat{s}_I^m}=\frac{\delta \widehat{}}{\delta \widehat{\overline{\sigma }}_I^m}+\underset{J}{^{}}\widehat{\sigma }_{I+J}^m\frac{\delta \widehat{}}{\delta \widehat{\sigma }_J^m}.$$
(2.88)
In the same fashion, having made use of the closure identity $`I`$ in the high-energy Fock space $`𝒢`$, and Eq. (2.87), one can rewrite the other decoupling equation $`D_2`$ in the form of
$`D_2`$ $`=`$ $`0|e^{\widehat{S}^{}}e^{\widehat{S}}H(\mathrm{\Lambda })e^{\widehat{S}}(I)e^{\widehat{S}^{}}C_I^{}|0,`$ (2.89)
$`=`$ $`\widehat{}(\mu )0|e^{\widehat{S}^{}}C_I^{}|0+\underset{J}{{\displaystyle ^{}}}\widehat{\overline{\omega }}_{IJ}\{0|e^{\widehat{S}^{}}e^{\widehat{S}}[H(\mathrm{\Lambda }),C_I^{}]e^{\widehat{S}}|0,`$
$`+`$ $`0|e^{\widehat{S}^{}}C_I^{}e^{\widehat{S}^{}}(I)e^{\widehat{S}^{}}e^{\widehat{S}}H(\mathrm{\Lambda })e^{\widehat{S}}e^{\widehat{S}^{}}|0\},`$
$`=`$ $`{\displaystyle \frac{\delta \widehat{}(\mu )}{\delta \widehat{\sigma }_I^m}}+\underset{J}{{\displaystyle ^{}}}{\displaystyle \frac{\delta \widehat{}(\mu )}{\delta \widehat{\overline{\sigma }}_J^m}}\widehat{L}_{JI}+\underset{J,K}{{\displaystyle ^{}}}{\displaystyle \frac{\delta \widehat{}(\mu )}{\delta \widehat{\sigma }_J^m}}\widehat{\sigma }_{J+K}^m\widehat{L}_{KI},`$
where in final step we have made use of the orthogonality relation Eq. (2.87) and the relation in Eq. (2.88). The operator $`\widehat{L}_{IJ}`$ is defined as
$$\widehat{L}_{IJ}=\underset{K}{}0|e^{\widehat{S}^{}}C_K^{}e^{\widehat{S}^{}}C_J^{}|00|C_Ke^{\widehat{S}^{}}C_I^{}|0.$$
(2.90)
The operator $`\widehat{L}_{IJ}`$ is symmetric and doubly linked. It is obvious from Eqs. (2.88, 2.89) that the requirement of $`\delta \widehat{}(\mu )/\delta \widehat{\sigma }_I^m=0`$ and $`\delta \widehat{}(\mu )/\delta \widehat{\overline{\sigma }}_I^m=0`$ lead directly to the decoupling conditions $`D_1=0`$ and $`D_2=0`$. However the reverse is not always correct. Therefore, the decoupling conditions are weaker than the variational equations. A set of similar canonical variables as in Eq. (2.84) (where the variables are $`c`$-numbers), was introduced by Arponen, Bishop and Pajanne in the context of the traditional CCM and turned out to be quite practical.
We have already proved (in section 2.4) that the effective low-energy operators Eq. (2.67) or Eq. (2.77) equipped with decoupling conditions Eqs. (2.78,2.79) have the same low-energy eigenvalues as the original Hamiltonian. The decoupling conditions are thus sufficient requirements to ensure partial diagonalization of the Hamiltonian in the particle and momentum space.
Having obtained the effective bra and ket Eq. (2.76) at a given scale $`\mu `$, one can compute the effective low-energy of an arbitrary operator $`A`$,
$$\widehat{𝒜}(\widehat{s}_I,\widehat{s}_I^{})=\widehat{\stackrel{~}{\psi }}_\mu |A|\widehat{\psi _\mu },$$
(2.91)
One may now pose the question if the approximation (truncation) used to obtain the induced bra- and ket-state, and accordingly the effective Hamiltonian, is sufficient to obtain the effective low-energy of a given operator by Eq. (2.91). On the other hand, one might be curious about the necessity of the double similarity transformation and as well the introduction of two independent sets of variables $`\{\widehat{s}_I^m,\widehat{s}_I^m\}`$, since one could have defined a single transformation $`He^{\widehat{S}}He^{\widehat{S}}`$ where $`\widehat{S}`$ is defined in Eq. (2.66) (without introducing a new set of variables $`\{\widehat{s}_I^m\}`$, and invoking the definition of the bar state). In this case, the variables set $`\{\widehat{s}_I^m\}`$ would be determined by Eq. (2.78) alone. Let us pursue this idea and apply the Hellmann-Feynman theorem<sup>3</sup><sup>3</sup>3The Hellmann-Feynman theorem is originally introduced for the ground state, however its proof is more general and can be applied here. to such a parametrization. In our framework, the Hellmann-Feynman theorem states that if we perturb the Hamiltonian $`HH^{}=H+JA`$, where $`J`$ is an infinitesimally small (a source term) and $`A`$ is an arbitrary operator, such that the effective Hamiltonian changes as $`\widehat{}\widehat{}^{}=\widehat{}+Jd\widehat{}/dJ+O(J^2)`$ then we have,
$$d\widehat{}/dJ=\widehat{𝒜}=\psi _\mu |dH/dJ|\psi _\mu ,$$
(2.92)
where we define an effective low-energy operator $`\widehat{𝒜}=\psi _\mu |A|\psi _\mu `$. We can find the effective low-energy operator $`\widehat{𝒜}`$ by using the Hellmann-Feynman theorem,
$$\widehat{𝒜}=\frac{d}{dJ}0|e^{\widehat{S}}(H+JA)e^{\widehat{S}}|0=0|e^{\widehat{S}}Ae^{\widehat{S}}|0+0|e^{\widehat{S}}He^{\widehat{S}}C_I^{}|0\frac{\delta \widehat{s}_I^m}{\delta J},$$
(2.93)
where we used $`[\widehat{S},C_I^{}]=0`$. If we now calculate $`\delta \widehat{s}_I^m`$ from the first decoupling equation (2.78) which involves only $`\widehat{S}`$, having retained only $`O(J,\delta S)`$ terms, we find
$$0|C_Ie^{(\widehat{S}+\delta \widehat{S})}(H+JA)e^{(\widehat{S}+\delta \widehat{S})}|0=0|C_Ie^{\widehat{S}}JAe^{\widehat{S}}|0+0|C_Ie^{\widehat{S}}[H,C_J^{}]e^{\widehat{S}}|0\delta \widehat{s}_J^m=0.$$
(2.94)
Therefore, one can show that,
$$C_I^{}|0\frac{\delta \widehat{s}_I^m}{\delta J}=𝒬(\widehat{}𝒬e^{\widehat{S}}He^{\widehat{S}}𝒬)^1𝒬e^{\widehat{S}}Ae^{\widehat{S}}|0,$$
(2.95)
where the operator $`𝒬=1|00|=\underset{J}{{\displaystyle ^{}}}C_J^{}|00|C_J`$ is introduced. Now one can make use of Eq. (2.95) to show that the right-hand side of Eq. (2.93) can be recast in the following form,
$$\widehat{𝒜}=0|e^{\widehat{S}^{}}e^{\widehat{S}}Ae^{\widehat{S}}|0,$$
(2.96)
where we introduced the notation $`e^{\widehat{S}^{}}`$ by,
$$0|e^{\widehat{S}^{}}=0|+0|e^{\widehat{S}}He^{\widehat{S}}𝒬(\widehat{}𝒬e^{\widehat{S}}He^S𝒬)^1𝒬.$$
(2.97)
Interestingly, Eq. (2.97) satisfies the second decoupling condition Eq. (2.79). Thereby Eq. (2.96) becomes exactly equivalent to Eq. (2.91). Therefore the use of the the Hellmann-Feynman theorem and assuming the coupled-cluster parametrization leads naturally to the definition of the bra-state in Eq. (2.76) and emergence of a new set of variable $`\widehat{S}^{}`$. In other words, no other bra-state parametrization (including hermitian-adjoint of ket-state) is compatible with the Hellmann-Feynman theorem.
It is well known that the traditional multiplicative renormalization is not sufficient for the renormalization of more than one composite operator inserted into the renormalized Green functions. To avoid ad hoc subtractions, one can introduce the composite operators into the Lagrangian with a space-time dependent source (coupling). It has been shown that the renormalization of the source produces requires counter terms to render all Green functions containing the insertion of composite operators renormalized . These new counterterms do not affect the renormalization of the original theory.
The main advantage of the compatibility of our parametrization with the Hellmann-Feynman theorem underlies that here the effective low-energy of an arbitrary operator, can be obtained in the same truncation scheme SUB$`(𝒩,)`$ used for the Hamiltonian matrix <sup>4</sup><sup>4</sup>4As Thouless pointed out, the Hellmann-Feynman theorem immediately implies that an expectation value $`A`$ of an arbitrary operator is computed diagrammatically from the same set of Goldstone diagrams as for the energy $`H`$, where all interaction is replaced by the operator $`A`$.. Moreover, the above-mentioned technique used in Lagrangian formalism, can be employed in our framework by means of the Hellmann-Feynman theorem. Therefore, the renormalization of an arbitrary operator can be calculated in a unified way. This is indeed, one of advantages of the non-unitary parametrization of the similarity renormalization group.
### 2.7 The symplectic structure
Despite extensive progress in development of various RG techniques, little is still known about the geometrical interpretation of the RG . Here, we introduce the geometrical structure emerging from our approach. We define a low-energy action-like functional $`\widehat{𝒜}`$, having integrated out fast modes and making use of the reparametrization Eq. (2.84),
$`\widehat{𝒜}(\mu )`$ $`=`$ $`{\displaystyle 𝑑t0|e^{\widehat{S}^{}(t)}e^{\widehat{S}(t)}(i\frac{\delta }{\delta t}\widehat{H}(t,\mathrm{\Lambda }))e^{\widehat{S}(t)}e^{\widehat{S}^{}(t)}|0},`$ (2.98)
$`=`$ $`{\displaystyle 𝑑t(i\underset{I}{^{}}\widehat{\overline{\sigma }}_I\widehat{\dot{\sigma }}_I\widehat{}(\mu ,\sigma ,\overline{\sigma }))},`$
$`=`$ $`{\displaystyle 𝑑t(i\underset{I}{^{}}\widehat{\dot{\overline{\sigma }}}_I\widehat{\sigma }_I\widehat{}(\mu ,\sigma ,\overline{\sigma }))},`$
where in the final step we employed integration by parts. The operator $`(\mu ,\sigma ,\overline{\sigma })`$ is the effective low-energy Hamiltonian defined in Eq. (2.67) and $`\widehat{\sigma }_I(\widehat{\overline{\sigma }}_I)`$ are defined,
$$\widehat{\sigma }_I=\underset{m=0}{\overset{}{}}\widehat{\sigma }_I^m\widehat{\overline{\sigma }}_I=\underset{m=0}{\overset{}{}}\widehat{\overline{\sigma }}_I^m$$
(2.99)
The stationary of $`\widehat{𝒜}`$ with respect to the complete set of variables $`\{\widehat{\sigma }_I,\widehat{\overline{\sigma }}_I\}`$ for a given truncation SUB$`(𝒩,)`$ yields
$$\delta \widehat{𝒜}(\mu )=0\left(i\frac{\delta \widehat{\sigma }_I}{\delta t}=\frac{\delta \widehat{}(\mu )}{\delta \widehat{\overline{\sigma }}_I};i\frac{\delta \widehat{\overline{\sigma }}_I}{\delta t}=\frac{\delta \widehat{}(\mu )}{\delta \widehat{\sigma }_I}\right).$$
(2.100)
These equations can be obtained for the set $`\{\widehat{s}_I,\widehat{s}_I^{}\}`$ as well, without invoking the canonical transformation Eq. (2.84). However, if one wants to obtain the above equations in a straightforward manner from time-dependent Schrödinger equation, then the canonical transformation Eq. (2.84) is necessary.
We note that although our transformation is not unitary and the parametrization of $`\widehat{S}`$ and $`\widehat{S}^{}`$ have been introduced in a very asymmetric fashion, their fundamental dynamics in the low-energy phase space follows the canonical equation of motion. Therefore the induced low-energy operators $`\widehat{\sigma }_I`$ and $`\widehat{\overline{\sigma }}_I`$ are canonically conjugate in the terminology of the classical Hamiltonian mechanics. This can be made even more suggestive by defining a new set of variables, the generalized field $`\widehat{\varphi }_I`$ and their canonically conjugate generalized momentum densities $`\widehat{\pi }_I`$,
$$\widehat{\varphi }_I=\frac{1}{2}(\widehat{\sigma }_I+\widehat{\overline{\sigma }}_I),\widehat{\pi }_I=\frac{i}{2}(\widehat{\overline{\sigma }}_I\widehat{\sigma }_I).$$
(2.101)
In terms of the new operators the equation of motion Eq. (2.100) can be rewritten into the form
$`{\displaystyle \frac{d\widehat{\varphi }_I}{dt}}=\{\widehat{\varphi }_I,\widehat{}(\mu )\}={\displaystyle \frac{\delta \widehat{}(\mu )}{\delta \widehat{\pi }_I}},`$
$`{\displaystyle \frac{d\widehat{\pi }_I}{dt}}=\{\widehat{\pi }_I,\widehat{}(\mu )\}={\displaystyle \frac{\delta \widehat{}(\mu )}{\delta \widehat{\varphi }_I}},`$ (2.102)
where a generalized Poisson bracket $`\{A,B\}`$ for two arbitrary operators is defined as
$$\{A,B\}=\underset{I}{^{}}\left(\frac{\delta A}{\delta \widehat{\varphi }_I}\frac{\delta B}{\delta \widehat{\pi }_I}\frac{\delta B}{\delta \widehat{\varphi }_I}\frac{\delta A}{\delta \widehat{\pi }_I}\right).$$
(2.103)
For the nonzero mutual Poisson brackets of canonical coordinates and momentums we have $`\{\widehat{\varphi }_I,\widehat{\pi }_J\}=\delta (I,J)`$. We can also look at the behaviour of the low-energy effective operator for a product of operators. One can write the product of operators after a double similarity transformation as a product of transformed operators,
$`\widehat{\stackrel{~}{\psi }}_\mu |AB|\widehat{\psi _\mu }`$ $`=`$ $`0|e^{\widehat{S}^{}}e^{\widehat{S}}ABe^{\widehat{S}}e^{\widehat{S}^{}}|0=0|\overline{A}\overline{B}|0,`$
$`=`$ $`\widehat{𝒜}\widehat{}+\underset{I}{{\displaystyle ^{}}}0|e^{\widehat{S}^{}}e^{\widehat{S}}Ae^{\widehat{S}}e^{\widehat{S}^{}}C_I^{}|00|C_Ie^{\widehat{S}^{}}e^{\widehat{S}}Be^{\widehat{S}}e^{\widehat{S}^{}}|0,`$
where we have used the closure identity in $`Q`$-space. The effective operator $`\widehat{𝒜}`$ and $`\widehat{}`$ are defined by Eq. (2.91). Now we express the last part of Eq. (2.7) in terms of a derivatives of the effective operators $`\widehat{𝒜}(\widehat{})`$ with respect to low-energy phase space $`\{\widehat{\sigma }_I,\widehat{\overline{\sigma }}_I\}`$. In fact, we have already shown such relations in Eqs. (2.88, 2.89). The same relationships are valid for any arbitrary operator for new variables $`\{\widehat{\sigma }_I,\widehat{\overline{\sigma }}_I\}`$. Therefore with a replacement of $`HA,B`$ and $`𝒜,`$ in Eqs. (2.88, 2.89) we have the desired equations. The same procedure can be carried out for evaluating of $`\widehat{\stackrel{~}{\psi }}_\mu |BA|\widehat{\psi _\mu }`$. After some straightforward algebra, one finds that,
$$\widehat{\stackrel{~}{\psi }}_\mu |ABBA|\widehat{\psi _\mu }=\widehat{\stackrel{~}{\psi }}_\mu |[A,B]|\widehat{\psi _\mu }=i\{\widehat{𝒜},\widehat{}\},$$
(2.105)
where the Poisson bracket is as defined in Eq. (2.103).
Therefore our parametrization is clearly suggestive that the effective low-energy phase space induced by integrating out the high-energy particles can be described in terms of highly non-linear classical dynamics with the canonical coordinates $`(\widehat{\varphi }_I,\widehat{\pi }_I)`$, without losing any quantum-field-theoretical information. The canonical equation of motions Eq. (2.102) are valid at any level of approximation SUB$`(𝒩,)`$ on the renormalization group trajectory. In this way, we are led to a manifestation of the correspondence principle in a more generalized form, that is the induced low-energy Hamiltonian obtained by integrating short-distance modes are governed by a hierarchy of non-linear classical mechanical equations for quasi-local fields $`\widehat{\varphi }_I`$ and $`\widehat{\pi }_I`$. One may hope to recover the full classical mechanics (with c-numbers variables as coordinates) at the other extreme of the RG trajectory.
The general form for the action completely determines the symplectic structure of our low-energy phase space $`(\widehat{\varphi }_I,\widehat{\pi }_I)`$. The phase space is the cotangent bundle<sup>5</sup><sup>5</sup>5In differential geometry, the cotangent bundle is the union of all cotangent spaces of a manifold . of the configuration space $`𝒞`$, $`\mathrm{\Gamma }=T^{}(𝒞)`$ (the coordinates $`\widehat{\varphi }_I`$ label the points of configuration space $`𝒞`$) . The Poisson bracket Eq. (2.103) induces a symplectic structure
$$\omega =\frac{1}{2}\omega _{ab}dx^adx^b,$$
(2.106)
where the coordinates on the symplectic manifold $`\mathrm{\Gamma }`$ are denoted by $`x^a\{\widehat{\varphi }_I\}`$ and $`\omega _{ab}`$ is the inverse of the symplectic matrix
$$\omega ^{ab}=\{x^a,x^b\}\{A,B\}=\omega ^{ab}A_{,a}B_{,b}.$$
(2.107)
Moreover, a symplectic form defines an isomorphism between the tangent and cotangent spaces of $`\mathrm{\Gamma }`$. One may associate a vector field $`X_f`$ to every function $`f𝒞^{\mathrm{}}(\mathrm{\Gamma })`$ by
$$i(X_f)\omega =df,$$
(2.108)
where $`i(X)`$ and $`X_f`$ denote the interior product and the symplectic gradient of $`f`$, respectively. $`X_f`$ is the so-called the Hamiltonian vector field of $`f`$, and it generates a flow on $`\mathrm{\Gamma }`$ which leaves $`\omega `$ invariant, since the Lie derivative of $`\omega `$ along $`X_f`$ is zero. In this way, one can rewrite the Poisson bracket in the form
$$\{f,g\}=i(X_f)i(X_g)\omega =i(X_f)dg=\omega (X_f,X_g)𝒞^{\mathrm{}}(\mathrm{\Gamma }),$$
(2.109)
which shows the change of $`g`$ along $`X_f`$. We require that $`\omega `$ to be closed ($`d\omega =0`$)<sup>6</sup><sup>6</sup>6A symplectic manifold ($`\mathrm{\Gamma },\omega `$) is a smooth real $`N`$-dimensional manifold without boundary, equipped with a closed non-degenerate two-form $`\omega `$, i. e., $`d\omega =0`$ where $`d`$ is the exterior differential ., which implies the Jacobi identities $`(d\omega )(X_f,X_g,X_h)=0`$. Therefore, the existence of Poisson bracket in our definition of the coordinates of the phase-space introduces a symplectic manifold for the phase-space.
### 2.8 The constrained induced phase space
In this section we closely follow Ref. (one should pay some extra care here, since the phase space here is operatorial rather than $`c`$-number). Let us assume that the renormalized form of an arbitrary operator $`A`$, Eq. (2.91) can be obtained by,
$$𝒜=0|e^{\widehat{S}^{}}e^{\widehat{S}}Ae^{\widehat{S}}e^{\widehat{S}^{}}|0\frac{0|e^{\widehat{S}^{}}Ae^{\widehat{S}}|0}{0|e^{\widehat{S}^{}}e^{\widehat{S}}|0}.$$
(2.110)
Generally the left-hand side does not agree with right-hand side (it is by no means clear that this relation will be held after a truncation). In order to ensure the unitarity of the similarity transformation we require,
$$0|e^{\widehat{S}^{}}\frac{0|e^{\widehat{S}^{}}e^{\widehat{S}}}{0|e^{\widehat{S}^{}}e^{\widehat{S}}|0},e^{\widehat{S}^{}}|0\frac{e^{\widehat{S}^{}}e^{\widehat{S}}|0}{0|e^{\widehat{S}^{}}e^{\widehat{S}}|0}.$$
(2.111)
Thereby, the ket-state and bra-state defined in Eq. (2.76) become hermitian-adjoint of each another. We assume the induced low-energy phase space to be a complex manifold $`\{(\widehat{s}_I,\widehat{s}_I^{}),(\widehat{s}_I^{},\widehat{s}_I^{})\}`$. The hermiticity conditions are introduced by constraint functions $`\widehat{\chi }_I(\mu )`$ and $`\widehat{\chi }_I^{}(\mu )`$
$`\widehat{\chi }_I(\mu )(0|e^{\widehat{S}^{}}e^{\widehat{S}}|0)^10|e^{\widehat{S}^{}}C_I^{}e^{\widehat{S}}|00|\widehat{S}^{}C_I^{}|0,`$
$`\widehat{\chi }_I^{}(\mu )(0|e^{\widehat{S}^{}}e^{\widehat{S}}|0)^10|e^{\widehat{S}^{}}C_Ie^{\widehat{S}}|00|C_I\widehat{S}^{}|0.`$ (2.112)
Therefore the physical submanifold shell is defined through:
$`\widehat{\chi }_I=0PX^1(\mu ,\mathrm{\Lambda })C_I^{}X(\mu ,\mathrm{\Lambda })P={\displaystyle \underset{m=0}{\overset{}{}}}\widehat{s}_I^m\widehat{s}_I^{},`$
$`\widehat{\chi }_I^{}=0PX^1(\mu ,\mathrm{\Lambda })C_IX(\mu ,\mathrm{\Lambda })P={\displaystyle \underset{m=0}{\overset{}{}}}\widehat{s}_I^m\widehat{s}_I^{},`$ (2.113)
where $`X(\mu ,\mathrm{\Lambda })`$ is defined in Eq. (2.64, 2.65). This implies that in the physical submanifold where we have exact hermiticity, and there is an isomorphic mapping between a cluster of high-energy creation $`C_I^{}`$ (annihilation $`C_I`$) operators and a low-energy operators $`\widehat{s}_I^{}(\widehat{s}_I^{})`$. This isomorphism is invariant under the renormalization group transformation. We introduce a Poisson bracket for the complex representation of the phase space,
$$\{A,B\}=\underset{I}{^{}}\left(\frac{\delta A}{\delta \widehat{s}_I}\frac{\delta B}{\delta \widehat{s}_I^{}}\frac{\delta A}{\delta \widehat{s}_I^{}}\frac{\delta B}{\delta \widehat{s}_I}+\frac{\delta A}{\delta \widehat{s}_I^{}}\frac{\delta B}{\delta \widehat{s}_I^{}}\frac{\delta B}{\delta \widehat{s}_I^{}}\frac{\delta B}{\delta \widehat{s}_I^{}}\right).$$
(2.114)
The non-zero commutators of the canonical coordinates follow the canonical symplectic structure $`\{\widehat{s}_I,\widehat{s}_I^{}\}=\{\widehat{s}_I^{},\widehat{s}_J^{}\}=\delta (I,J)`$. The nature of the constraint can be revealed by considering the commutators between the constraints functional $`\widehat{\chi }_I`$ and $`\widehat{\chi }_I^{}`$. After some tedious but straightforward algebra, one obtains,
$`\{\widehat{\chi }_I,\widehat{\chi }_J\}=\{\widehat{\chi }_I^{},\widehat{\chi }_J^{}\}=0,`$
$`\{\widehat{\chi }_I,\widehat{\chi }_J^{}\}=2\left(0|e^{\widehat{S}^{}}e^SC_IC_J^{}e^{\widehat{S}}e^{\widehat{S}^{}}|0\widehat{\overline{\sigma }}_I\widehat{\overline{\sigma }}_J^{}\right).`$
This implies that the constraints are of second class and do not correspond to any gauge symmetry degrees of freedom. However, these superfluous degrees of freedom can be eliminated by Dirac bracket technique . Of course, since the constrained manifold $`N`$ has dimension less than the full manifold $`M`$, one may define a pullback map $`f^{}:T^{}(M)T^{}(N)`$ to obtain the induced symplectic structure of the constraint surface $`\widehat{\omega }^0`$, $`\widehat{\omega }^0=f^{}\widehat{\omega }`$. In analogy to ordinary CCM , we define a symplectic two-form in the full manifold,
$$\widehat{\omega }=\underset{I}{^{}}(d\widehat{s}_Id\widehat{s}_I^{}+d\widehat{s}_I^{}d\widehat{s}_I^{}).$$
(2.115)
The induced symplectic two-form on the physical shell can be found by substituting the value of $`d\widehat{s}_I^{}`$ and $`d\widehat{s}_I^{}`$ in terms of $`d\widehat{s}_I`$ and $`d\widehat{s}_I^{}`$ by using on-shell condition $`d\widehat{\chi }_I=d\widehat{\chi }_I^{}=0`$, hence we find
$$\widehat{\omega }^0=2\underset{I,J}{^{}}\widehat{w}^{IJ}d\widehat{\sigma }_Id\widehat{\sigma }_J^{},$$
(2.116)
where $`\widehat{w}^{IJ}=\frac{\delta \widehat{\sigma }_I^{}}{\delta \widehat{\overline{\sigma }}_J}`$, consequently $`\widehat{\omega }^0`$ is closed and $`\widehat{\omega }`$ is positive matrix. Therefore the induced physical low-energy phase space is a Kähler manifold <sup>7</sup><sup>7</sup>7A Kähler manifold is a Hermitian manifold (M,g) whose Kähler form $`\mathrm{\Omega }`$ is closed: $`d\mathrm{\Omega }=0`$. The metric $`g`$ is called the Kähler metric of $`M`$ .. This defines a positive hermitian metric in the physical shell <sup>8</sup><sup>8</sup>8 Notice that it is well known that the geometrical quantization can be applied on Kähler manifold since it has a natural polarization..
In the following next two sections, we apply our RG formalism to compute the effective Hamiltonian for $`\varphi ^4`$ and extended Lee theory up to two- and one-loop order, respectively.
### 2.9 Example I: $`\mathrm{\Phi }^4`$ theory
In this section, we obtain the effective Hamiltonian of $`\varphi ^4`$ theory up to two-loop order in equal-time quantization. In following we will quote from Ref. . The bare $`\varphi ^4`$ theory Hamiltonian is
$$H=d^3x\left(\frac{1}{2}\pi ^2(x)+\frac{1}{2}\varphi (x)\left(^2+m^2\right)\varphi (x)+g\varphi ^4(x)\right).$$
(2.117)
According to our logic the ultraviolet-finite Hamiltonian is obtained by introducing counterterms, which depend on the UV cutoff $`\mathrm{\Lambda }`$ and some arbitrary renormalization scale. This redefines the parameters of the theory and defines the effective low-energy Hamiltonian. The renormalized Hamiltonian has the form
$$H=d^3x\left(\frac{Z_\pi }{2}\pi ^2(x)+\frac{1}{2}\sqrt{Z_\varphi }\varphi (x)\left(^2+Z_mm^2\right)\sqrt{Z_\varphi }\varphi (x)+Z_gZ_\varphi ^2g\varphi ^4(x)+\mathrm{}\right).$$
(2.118)
Each of the $`Z`$-factors has an expansion of the form.
$$Z=1+f_1(\mathrm{\Lambda })\lambda +f_2(\mathrm{\Lambda })\lambda ^2+\mathrm{},$$
(2.119)
where $`\lambda `$ is a generic coupling constant of theory and has been defined at a given renormalization scale $`M`$. The functions $`f_n`$ will be obtained order-by-order, by summing up contributions of the fast modes between $`\mu `$ and $`\mathrm{\Lambda }`$, in the sense that $`Z(\mathrm{\Lambda })Z(\mu )`$ and $`f_n(\mathrm{\Lambda })f_n(\mu )`$. This means that the low-energy correlation functions are invariants of the renormalization group flow. One can therefore assume that the $`Z`$’s are initially $`1`$ and choose the corresponding $`f`$’s from the condition that the cut-off dependence be cancelled out after computing the effective Hamiltonian in the desired loop order. Even though the newly generated interactions are sensitive to the regularization scheme (as is well known , a sharp cutoff may lead to new non-local interaction terms), nevertheless one can ignore these if they are finite and do not produce any divergence as the cutoff $`\mathrm{\Lambda }`$ approaches to infinity. We now split field operators into high- and low-momentum modes; $`\varphi (x)=\varphi _L(x)+\varphi _H(x)`$, where $`\varphi _L(x)`$ denotes modes of low-frequency with momentum $`k\mu `$ and $`\varphi _H(x)`$ denotes modes of high-frequency with momentum constrained to a shell $`\mu <k\mathrm{\Lambda }`$. The field $`\varphi _L(x)`$ can be conceived as a background to which the $`\varphi _H(x)`$-modes are coupled. Therefore, in the standard diagrammatic language, integrating out the high-frequency modes $`\varphi _H(x)`$ implies that only high-frequency modes appear in internal lines. The field $`\varphi _H(x)`$ is represented in Fock space as
$$\varphi _H(x)=\underset{\mu <k\mathrm{\Lambda }}{}\frac{1}{\sqrt{2\omega _k}}(a_ke^{ikx}+a_k^{}e^{ikx}),$$
(2.120)
where $`\omega _k=\sqrt{k^2+m^2}`$ and the operators $`a_k`$ and $`a_k^{}`$ satisfy the standard boson commutation rules. From now on all summations are implicitly over the high-frequency modes $`\mu <k\mathrm{\Lambda }`$. The Hamiltonian in terms of high- and low-frequency modes can be written as, after normal ordering with respect to high-frequency modes,
$$H=H_1+H_2+V_B+V_C+V_A,$$
(2.121)
where we define,
$`H_1`$ $`=`$ $`{\displaystyle \left(\frac{1}{2}\pi _L^2(x)+\frac{1}{2}\varphi _L(x)\left(^2+m^2\right)\varphi _L(x)+g\varphi _L^4(x)\right)},`$
$`H_2`$ $`=`$ $`{\displaystyle \omega _ka_k^{}a_k},`$
$`V_B`$ $`=`$ $`g{\displaystyle \frac{e^{i(p+q+rk)x}}{\sqrt{\omega _k\omega _p\omega q\omega _r}}a_k^{}a_pa_qa_r}+{\displaystyle \frac{3e^{i(p+qrk)x}}{4\sqrt{\omega _k\omega _p\omega q\omega _r}}}a_k^{}a_p^{}a_qa_r`$
$`+6\varphi _L(x){\displaystyle \frac{e^{i(p+qk)x}}{\sqrt{2\omega _k\omega _p\omega _q}}}a_k^{}a_pa_q+3\left(\varphi _L^2(x)+{\displaystyle \frac{1}{2\omega _r}}\right){\displaystyle \frac{e^{i(kp)x}}{\sqrt{\omega _k\omega _p}}}a_p^{}a_k`$
$`+{\displaystyle \frac{3\varphi _L^2(x)}{2\omega _r}}+\text{h.c.},`$
$`V_C`$ $`=`$ $`g{\displaystyle V_C^4a_k^{}a_p^{}a_q^{}a_r^{}}+V_C^3a_k^{}a_p^{}a_q^{}+V_C^2a_k^{}a_p^{}+V_C^1a_k^{},`$
$`V_A`$ $`=`$ $`V_C^{},`$
$`V_C^1`$ $`=`$ $`\left({\displaystyle \frac{6\varphi _L(x)}{\omega _p}}+4\varphi _L^3(x)\right){\displaystyle \frac{e^{ikx}}{\sqrt{2\omega _k}}},V_C^2=3\left(\varphi _L^2(x)+{\displaystyle \frac{1}{2\omega _r}}\right){\displaystyle \frac{e^{i(k+p)x}}{\sqrt{\omega _k\omega _p}}},`$
$`V_C^3`$ $`=`$ $`2\varphi _L(x){\displaystyle \frac{e^{i(k+p+q)x}}{\sqrt{2\omega _k\omega _p\omega _k}}},V_C^4={\displaystyle \frac{e^{i(k+p+q+r)x}}{4\sqrt{\omega _k\omega _p\omega _q\omega _r}}}.`$ (2.122)
The high-energy configurations in the Fock space are specified by $`\{C_I_{i=1}a_{k_i}\}`$ and $`\{C_I^{}_{i=1}a_{k_i}^{})\}`$. Up to two-loop expansion, our renormalization scheme requires to keep $`S(S^{})`$ at least to order $`n=4`$, which allows us to eliminate the pure terms $`V_C`$ and $`V_A`$ at a lower level of expansion. The $`\widehat{S}(\widehat{S}^{})`$ operators consistent with a $`SUB(4,m)`$ truncation scheme are,
$`\widehat{S}_m`$ $`=`$ $`{\displaystyle \left(\widehat{S}_m^1a_k^{}+\widehat{S}_m^2a_k^{}a_p^{}+\widehat{S}_m^3a_k^{}a_p^{}a_q^{}+\widehat{S}_m^4a_k^{}a_p^{}a_q^{}a_r^{}\right)},`$
$`\widehat{S}_m^{}`$ $`=`$ $`{\displaystyle \left(\widehat{S}_m^1a_k+\widehat{S}_m^2a_ka_p+\widehat{S}_m^3a_ka_pa_q+\widehat{S}_m^4a_ka_pa_qa_r\right)}.`$ (2.123)
We split the diagonalization of the Hamiltonian matrix in an upper and lower triangle part, by using the double similarity transformation. One may notice that the “most non-diagonal” terms in the Hamiltonian are $`V_C`$ and $`V_A`$ (in the light-front Hamiltonian such terms do not exist because modes with longitudinal momentum identically zero are not allowed). The potential $`V_B`$ is already partially diagonalized and does not change the vacuum of the high-energy states. Therefore, here we employ a minimal scheme, aiming at removal of $`V_A`$ and $`V_C`$ only.
We restrict ourselves to the elimination of the high-energy degrees of freedom up to the first order in the coupling constant $`g`$ and second order in the ratio of cutoffs $`\mu /\mathrm{\Lambda }`$. Therefore, our truncation scheme is called $`SUB(4,2)`$. For $`m=0`$ one finds,
$`S_0^1`$ $`=`$ $`g{\displaystyle \frac{V_C^1}{\omega _k}},S_0^2=g{\displaystyle \frac{V_C^2}{\omega _k+\omega _p}},`$
$`S_0^3`$ $`=`$ $`g{\displaystyle \frac{V_C^3}{\omega _k+\omega _p+\omega _q}},S_0^4=g{\displaystyle \frac{V_C^4}{\omega _k+\omega _p+\omega _q+\omega _r}},`$ (2.124)
where the $`V_C^{14}`$ are defined in Eq. (2.9). Here, one has $`S_0^{}=S_0^{}`$. At this stage the results for the one-loop renormalization can be computed. We evaluate the effective Hamiltonian by substituting $`S(S^{})`$ from Eqs. (2.123) and (2.124) into Eq. (2.67). In order to achieve renormalization, one should identify the potentially divergent terms ( when $`\mathrm{\Lambda }\mathrm{}`$) in the expansion of $`H^{\text{eff}}(\mu )`$. Such a process generally can be done by inventing a power-counting rule, using the property $`S_n\frac{\mu }{\mathrm{\Lambda }}S_{n1}`$. Here we take $`\omega _k|k|`$ for $`\mu m`$ and replace $`_k`$ by $`\frac{d^3k}{(2\pi )^3}`$. The standard tadpole one-loop mass renormalization arises from $`V_B`$ due to normal-ordering. We add this divergent term to $`H_1`$ and renormalize the bare mass
$`\delta H^{\text{1-loop}}`$ $`=`$ $`0|V_B|0=6g{\displaystyle \frac{\varphi ^2(x)}{2\omega _k}}={\displaystyle \frac{3g}{4\pi ^2}}(\mathrm{\Lambda }^2\mu ^2){\displaystyle d^3x\varphi ^2(x)},`$
$`Z_m`$ $`=`$ $`1{\displaystyle \frac{3g}{2\pi ^2}}(\mathrm{\Lambda }^2\mu ^2).`$ (2.125)
In this order the contribution of the terms $`[V_C,S],[V_A,S^{}]`$ and $`[H_1,S(S^{})]`$ are zero, after projection on to the high-energy vacuum. The only divergent contributions come from $`[V_A^{2(3)},S_0^{2(3)}]`$ due to a double and third contraction of the high-frequency fields respectively. There are two other divergent terms, $`([V_C^{2(3)},S_0^{2(3)}]`$, however they are harmless and are cancelled out by the divergence of $`[[H_2,S_0],S_0^{2(3)}]`$. One thus obtains,
$`\delta H`$ $`=`$ $`{\displaystyle \frac{18g^2}{(2\pi )^6}}{\displaystyle \frac{\varphi ^2(x)\varphi ^2(y)}{\omega _k\omega _p(\omega _k+\omega _p)}e^{i(k+p)(xy)}}`$ (2.126)
$``$ $`{\displaystyle \frac{12g^2}{(2\pi )^9}}{\displaystyle \frac{\varphi (x)\varphi (y)}{\omega _k\omega _p\omega _q(\omega _k+\omega _p+\omega _q)}e^{i(k+p+q)(xy)}}.`$
In general evaluation of integrals like Eq. (2.126) may produce non-localities. This is due to the fact that the total momentum in integrands of Eq. (2.126), namely $`r_1=p+q`$ and $`r_2=k+p+q`$ are in the low-momentum space. To evaluate such integrations, one can firstly reduce the potential divergent integrals by a change of variable, for example for the first integrand we use $`p,qp,r_1`$, and then expand the integrand in $`r_1/p`$. Therefore, after expansion and evaluating the momentum integrals, one may be faced with non-analytic terms in the low-momentum space. However here these are non-divergent and will thus be ignored. We find
$`\delta H^{\text{1-loop}}`$ $`=`$ $`{\displaystyle \frac{9g^2}{2\pi ^2}}\mathrm{ln}\left({\displaystyle \frac{\mathrm{\Lambda }}{\mu }}\right){\displaystyle d^3x\varphi ^4(x)}{\displaystyle \frac{3g^2}{2\pi ^4}}(2\mathrm{ln}21)\mathrm{\Lambda }^2{\displaystyle d^3x\varphi ^2(x)}`$ (2.127)
$`+`$ $`{\displaystyle \frac{3g^2}{16\pi ^4}}\mathrm{ln}\left({\displaystyle \frac{\mathrm{\Lambda }}{\mu }}\right){\displaystyle d^3x(\varphi (x))^2}+\text{finite terms}.`$
One can immediately deduce the renormalization factors $`Z_g`$ and $`Z_\varphi `$ from above expression
$`Z_g`$ $`=`$ $`1+{\displaystyle \frac{9g^2}{2\pi ^2}}\mathrm{ln}\left({\displaystyle \frac{\mathrm{\Lambda }}{\mu }}\right),`$ (2.128)
$`Z_\varphi `$ $`=`$ $`1{\displaystyle \frac{3g^2}{8\pi ^4}}\mathrm{ln}\left({\displaystyle \frac{\mathrm{\Lambda }}{\mu }}\right).`$ (2.129)
The unknown coefficients in expression $`S_1`$ is computed by making use of Eq. (2.124) and solving coupled equations (2.70), therefore one may yield,
$`S_1^1`$ $`=`$ $`{\displaystyle \frac{6ge^{ikx}}{\omega _k^2\sqrt{2\omega _k}}}\left(2\varphi _L(x)2i\pi _L(x)\varphi _L^2(x){\displaystyle \frac{i\pi _L(x)}{\omega _p}}\right){\displaystyle \frac{g}{\omega _k}}{\displaystyle \underset{\nu =1}{\overset{3}{}}}{\displaystyle \frac{1}{\nu !}}V_A^\nu S_1^{\nu +1},`$
$`S_1^2`$ $`=`$ $`{\displaystyle \frac{3ge^{i(k+p)x}}{(\omega _k+\omega _p)^2\sqrt{\omega _k\omega _p}}}\left(1i2\pi _L(x)\varphi _L(x)\right){\displaystyle \frac{g}{\omega _k+\omega _p}}\left([V_C^1,S_1^1]+{\displaystyle \underset{\nu =1}{\overset{2}{}}}{\displaystyle \frac{1}{\nu !}}V_A^\nu S_1^{\nu +2}\right)`$
$`S_1^3`$ $`=`$ $`{\displaystyle \frac{2ige^{i(k+p+q)x}}{(\omega _k+\omega _p+\omega _q)^2\sqrt{2\omega _k\omega _p\omega _q}}}\pi _L(x){\displaystyle \frac{g}{(\omega _k+\omega _p+\omega _q)}}\left(V_A^1S_1^4+{\displaystyle \underset{\nu =1}{\overset{2}{}}}[V_C^\nu ,S_1^{3\nu }]\right),`$
$`S_1^4`$ $`=`$ $`{\displaystyle \frac{g}{(\omega _k+\omega _p+\omega _q+\omega _r)}}{\displaystyle \underset{\nu =1}{\overset{3}{}}}[V_C^{4\nu },S_1^\nu ].`$ (2.130)
In the above expression summation over dummy momentum indices is assumed. One can find $`\widehat{S}_1^{}`$ in the same manner by exploiting Eq. (2.72) and using Eq. (2.130) as an input, which leads to
$$S_1^\nu =(S_1^\nu )^{}+S_1^{\nu a}\nu =1,\mathrm{},4,$$
(2.131)
with the notations,
$`S_1^{1a}`$ $`=`$ $`{\displaystyle \frac{g}{\omega _k}}\left({\displaystyle \underset{\nu =1}{\overset{3}{}}}{\displaystyle \frac{1}{\nu !}}S_1^{(\nu +1)a}V_C^\nu {\displaystyle \underset{\nu =1}{\overset{3}{}}}{\displaystyle \frac{1}{\nu !}}V_A^{\nu +1}S_1^\nu \right),`$
$`S_1^{2a}`$ $`=`$ $`{\displaystyle \frac{g}{\omega _k+\omega _p}}\left({\displaystyle \underset{\nu =1}{\overset{2}{}}}{\displaystyle \frac{1}{\nu !}}S_1^{(\nu +2)a}V_C^\nu (q){\displaystyle \underset{\nu =1}{\overset{2}{}}}{\displaystyle \frac{1}{\nu !}}V_A^{\nu +2}S_1^\nu +[V_A^1,S_1^{1a}]\right),`$
$`S_1^{3a}`$ $`=`$ $`{\displaystyle \frac{g}{\omega _k+\omega _p+\omega _q}}\left(S_1^{4a}V_C^1V_A^4S_1^1+{\displaystyle \underset{\nu =1}{\overset{2}{}}}[V_A^\nu ,S_1^{(3\nu )a}]\right),`$
$`S_1^{4a}`$ $`=`$ $`{\displaystyle \frac{g}{(\omega _k+\omega _p+\omega _q+\omega _r)}}{\displaystyle \underset{\nu =1}{\overset{3}{}}}[V_A^\nu ,S_1^{(4\nu )a}].`$ (2.132)
The only divergent contribution up to order $`g^2`$ arises from,
$$\delta H=0|[H_1,S_1],S_0^{}]|0,$$
(2.133)
After the evaluation of the leading divergent part, we find that
$$\delta H=\frac{3g^2}{16\pi ^4}\mathrm{ln}\left(\frac{\mathrm{\Lambda }}{\mu }\right)d^3x\pi ^2(x),$$
(2.134)
which contributes to the two-loop wave-function renormalization $`Z_\pi `$. By comparing Eqs. (2.129) and (2.134), one may conclude that $`Z_\pi =Z_\varphi ^1`$, as it should be. To finish the renormalization up to two-loop order, one should also take into account the contribution at order $`g^3`$. The divergent terms at this level originate from
$$\delta H=0|[[\left(V_A+1/2V_C+V_B\right),S_0],S_0^{}]|0.$$
(2.135)
After a straightforward but lengthy computation one can obtain the leading divergent parts,
$$\delta H=\frac{27g^3}{2\pi ^4}\left[\left[\mathrm{ln}\left(\frac{\mathrm{\Lambda }}{\mu }\right)\right]^2+\mathrm{ln}\left(\frac{\mathrm{\Lambda }}{\mu }\right)\right]d^3x\varphi ^4(x),$$
(2.136)
this term should be added to Eq. (2.126), therefore one can immediately deduce the correct total renormalization factor $`Z_g`$ up to two-loop order,
$$Z_g=1+\frac{9g^2}{2\pi ^2}\mathrm{ln}\left(\frac{\mathrm{\Lambda }}{\mu }\right)+\frac{g^3}{4\pi ^4}\left(81\left(\mathrm{ln}\left(\frac{\mathrm{\Lambda }}{\mu }\right)\right)^251\mathrm{ln}\left(\frac{\mathrm{\Lambda }}{\mu }\right)\right).$$
(2.137)
One can now immediately obtain the well-known two-loop $`\beta `$-function and anomalous dimension by making use of Eqs. (2.129, 2.137).
$`\beta (g)={\displaystyle \frac{g}{\mathrm{log}\mu }}|_\mathrm{\Lambda }={\displaystyle \frac{9}{2\pi ^2}}g^2{\displaystyle \frac{51}{4\pi ^4}}g^3,`$ (2.138)
$`\gamma (g)={\displaystyle \frac{1}{2}}{\displaystyle \frac{\mathrm{log}Z_\varphi }{\mathrm{log}\mu }}|_\mathrm{\Lambda }={\displaystyle \frac{3}{16\pi ^4}}g^2.`$ (2.139)
It is important to point out that the diagonalization at first order in the coupling constant defines a correct low-energy effective Hamiltonian which is valid up to order $`g^3`$. Having said that, from Eq. (2.131) one can observe that the non-hermiticity of the $`\widehat{S}`$ operator appears at order $`g^2`$ and in a lower order of $`\mu /\mathrm{\Lambda }`$. As we have shown, non-hermiticity is negligible up to two-loop order (asymmetric terms appear in irrelevant contributions (which are non-divergent and vanish as $`\mathrm{\Lambda }`$ goes to infinity). We conjecture that, for the present model, non-hermitian terms only appear in irrelevant contributions, whatever the order of truncation.
### 2.10 Example II: Extended Lee Model
As another illustrative example, we will now apply coupled-cluster RG to determine the effective Hamiltonian for an extended Lee model (ELM) up to the one-loop order.
We define four kinds of particles, the $`V`$-particle and $`N`$-particle as two different fermions and the $`\theta `$ and $`\overline{\theta }`$ as a scalar boson and anti-boson respectively. Here $`a(k)`$, $`a^{}(k)`$ and $`b(k)`$, $`b^{}(k)`$ are the annihilation and creation operators which satisfy boson commutator rules. The $`V(p)`$, $`V^{}(p)`$ and $`N(p)`$, $`N^{}(p)`$ define the fermion sector and obey the usual anticommutator rules. The bare ELM Hamiltonian then reads
$`H`$ $`=`$ $`H_0+H_I,`$
$`H_0`$ $`=`$ $`{\displaystyle d^3p\omega _V(p)V^{}(p)V(p)}+{\displaystyle d^3p\omega _N(p)N^{}(p)N(p)}`$
$`+`$ $`{\displaystyle d^3k\omega _\theta (k)a^{}(k)a(k)}+{\displaystyle d^3k\omega _{\overline{\theta }}(k)b^{}(k)b(k)},`$
$`H_I`$ $`=`$ $`\lambda _1(2\pi )^{3/2}{\displaystyle \frac{d^3kd^3p}{(2\omega _\theta (k))^{1/2}}V^{}(p)N(pk)a(k)}`$ (2.140)
$`+`$ $`\lambda _2(2\pi )^{3/2}{\displaystyle \frac{d^3kd^3p}{(2\omega _{\overline{\theta }}(k))^{1/2}}N^{}(p)V(pk)b(k)}+\text{h.c.}.`$
The kinetic energy generically is defined $`\omega _O(k)=\sqrt{k^2+m_O^2}`$ where the indice $`O`$ can be either $`(V,N,\theta ,\overline{\theta })`$. The interaction term in $`H_I`$ describes the processes;
$`VN+\theta ,`$ (2.141)
$`NV+\overline{\theta }.`$ (2.142)
The crossing symmetry become manifest if we take $`\lambda _1=\lambda _2`$ and equal masses for boson and anti-boson. For sake of generality we will ignore crossing symmetry at the moment. The Lee model can be recovered if we decouple the anti-boson $`\overline{\theta }`$, $`\lambda _20`$. In the Lee model the virtual process Eq. (2.142) is not included and thus the $`N`$-particle state remains unrenormalized and the model become exactly solvable.
It is believed that the Lee model is asymptotically free for space-time dimension $`D`$ less than four . With on-shell renormalization one can show that the Lee model for $`D>4`$ (odd $`D`$) is ultraviolet stable and not asymptotically free . It is well known that such a model in four dimension exhibit a ghost state as the cutoff is removed. The Hamiltonian Eq. (2.10) exhibits two symmetries; it is straightforward to verify that following operators commute with $`H`$
$`B`$ $`=`$ $`{\displaystyle d^3pV^{}(p)V(p)}+{\displaystyle d^3pN^{}(p)N(p)},`$
$`Q`$ $`=`$ $`{\displaystyle d^3pN^{}(p)N(p)}+{\displaystyle d^3kb^{}(p)b(p)}{\displaystyle d^3ka^{}(p)a(p)}.`$ (2.143)
Clearly $`B`$ is a baryon number operator and $`Q`$ is a charge operator. We assign the charges $`1,0,1`$ and $`1`$ to the $`N,V,\theta `$ and $`\overline{\theta }`$, respectively. The sectors of the ELM are labeled by the eigenvalue $`(b,q)`$ of the operators $`(B,Q)`$. According to our formulation the ultraviolet-finite Hamiltonian is obtained by introducing $`Z`$-factors, which depend on the UV cutoff $`\mathrm{\Lambda }`$ and some arbitrary renormalization scale $`M`$ in such way that effective Hamiltonian does not depend on $`\mathrm{\Lambda }`$. The bare Hamiltonian can be rewritten
$`H`$ $`=`$ $`{\displaystyle d^3pZ_V^2Z_{M_V}\omega _V(p)V^{}(p)V(p)}+{\displaystyle d^3pZ_N^2Z_{M_N}\omega _N(p)N^{}(p)N(p)}`$ (2.144)
$`+`$ $`{\displaystyle d^3kZ_\theta ^2Z_{M_\theta }\omega _\theta (k)a^{}(k)a(k)}+{\displaystyle d^3kZ_{\overline{\theta }}^2Z_{M_{\overline{\theta }}}\omega _{\overline{\theta }}(k)b^{}(k)b(k)}`$
$`+`$ $`{\displaystyle \frac{\lambda _1d^3kd^3p}{(2(2\pi )^3\omega _\theta (k))^{1/2}}Z_{\lambda _1}Z_VZ_NZ_\theta V^{}(p)N(pk)a(k)}`$
$`+`$ $`{\displaystyle \frac{\lambda _2d^3kd^3p}{(2(2\pi )^3\omega _{\overline{\theta }}(k))^{1/2}}Z_{\lambda _2}Z_VZ_NZ_{\overline{\theta }}N^{}(p)V(pk)b(k)}`$
$`+`$ $`\text{h.c.}.`$
We split the original Hamiltonian in the form of Eq. (2.69);
$`H_1`$ $`=`$ $`H({\displaystyle _0^\mu }),`$
$`H_2`$ $`=`$ $`H_0({\displaystyle _\mu ^\mathrm{\Lambda }}),`$
$`V_C`$ $`=`$ $`{\displaystyle _0^\mu }{\displaystyle _\mu ^\mathrm{\Lambda }}{\displaystyle \frac{d^3p^{}d^3k}{(2(2\pi )^3\omega _\theta (k))^{1/2}}}\lambda _1N^{}(p^{}k)V(p^{})a^{}(k)`$
$`+`$ $`{\displaystyle _0^\mu }{\displaystyle _\mu ^\mathrm{\Lambda }}{\displaystyle \frac{d^3p^{}d^3k}{(2(2\pi )^3\omega _{\overline{\theta }}(k))^{1/2}}}\lambda _2V^{}(p^{}k)N(p^{})b^{}(k),`$
$`V_A`$ $`=`$ $`V_C^{},`$
$`V_B`$ $`=`$ $`{\displaystyle _0^\mu }{\displaystyle _\mu ^\mathrm{\Lambda }}{\displaystyle \frac{d^3pd^3k^{}}{(2(2\pi )^3\omega _\theta (k^{}))^{1/2}}}\lambda _1V^{}(p)N(pk^{})a(k^{})`$ (2.145)
$`+`$ $`{\displaystyle _0^\mu }{\displaystyle _\mu ^\mathrm{\Lambda }}{\displaystyle \frac{d^3pd^3k^{}}{(2(2\pi )^3\omega _{\overline{\theta }}(k^{}))^{1/2}}}\lambda _2N^{}(p)V(pk^{})b(k^{})`$
$`+`$ $`{\displaystyle _\mu ^\mathrm{\Lambda }}{\displaystyle _\mu ^\mathrm{\Lambda }}{\displaystyle \frac{d^3pd^3k}{(2(2\pi )^3\omega _\theta (k))^{1/2}}}\lambda _1V^{}(p)N(pk)a(k)`$
$`+`$ $`{\displaystyle _\mu ^\mathrm{\Lambda }}{\displaystyle _\mu ^\mathrm{\Lambda }}{\displaystyle \frac{d^3pd^3k}{(2(2\pi )^3\omega _{\overline{\theta }}(k))^{1/2}}}\lambda _2N^{}(p)V(pk)b(k)+\text{h.c.}.`$
Here $`p^{}`$ and $`k^{}`$ stand for low momenta ($`p^{},k^{}<\mu `$). If the arguments of an operator are all low momenta ($`p^{}`$ or $`k^{}`$), this indicates low momentum operators. The arguments in $`H(_0^\mu )`$ and $`H_0(_\mu ^\mathrm{\Lambda })`$ means that all the momentum integrations involved in Eq. (2.10) are running between $`0<p^{}<\mu `$ for the former and $`\mu <p<\mathrm{\Lambda }`$ for the latter, respectively. The configuration space of the high momentum operators are specified by $`\{C_IV^{n_1}N^{n_2}a^{n_3}b^{n_4},C_I^{}C_I^{}(V^{})^{n_1}(N^{})^{n_2}(a^{})^{n_3}(b^{})^{n_4}\}`$ with $`n_1+n_2+n_3+n_4=I`$. Aiming at a one-loop expansion the corresponding $`S`$ and $`S^{}`$ operators which preserve the symmetry property Eq. (2.10), can be chosen as
$`S_m`$ $`=`$ $`{\displaystyle d^3p^{}d^3kS_m^1(p^{})V^{}(p^{}k)b^{}(k)}+{\displaystyle d^3p^{}d^3kS_m^2(p^{})N^{}(p^{}k)a^{}(k)},`$
$`S_m^1`$ $`=`$ $`S_m^NN(p^{})+S_m^{Vb}V(p^{}k^{})b(k^{})+S_m^{Va}V(p^{}+k^{})a^{}(k^{}),`$
$`S_m^2`$ $`=`$ $`S_m^VV(p^{})+S_m^{Na}N(p^{}k^{})a(k^{})+S_m^{Nb}N(p^{}+k^{})b^{}(k^{}).`$ (2.146)
We have ignored the $`I=1`$ configuration, since there are no tadpole type diagrams.( The truncation of $`S_I`$ in configuration space should be consistent with our loop expansion.) Here we confine our attention to the elimination of the high-momentum degrees of freedom up to the first order in coupling constant and second order in $`\mu /\mathrm{\Lambda }`$. The unknown coefficients in Eq. (2.146) can be obtained by making use of Eq. (2.73),
$`S_0^V`$ $`=`$ $`{\displaystyle \frac{\lambda _1}{(2(2\pi )^3\omega _\theta (k))^{1/2}(\omega _N(p^{}k)+\omega _\theta (k))}},`$
$`S_0^N`$ $`=`$ $`{\displaystyle \frac{\lambda _2}{(2(2\pi )^3\omega _{\overline{\theta }}(k))^{1/2}(\omega _V(p^{}k)+\omega _{\overline{\theta }}(k))}},`$
$`S_0^{Vb}`$ $`=`$ $`S_0^{Va}=0,`$
$`S_1^V`$ $`=`$ $`{\displaystyle \frac{\lambda _1\omega _V(p^{})}{(2(2\pi )^3\omega _\theta (k))^{1/2}(\omega _N(p^{}k)+\omega _\theta (k))^2}},`$
$`S_1^N`$ $`=`$ $`{\displaystyle \frac{\lambda _2\omega _N(p^{})}{(2(2\pi )^3\omega _{\overline{\theta }}(k))^{1/2}(\omega _V(p^{}k)+\omega _{\overline{\theta }}(k))^2}},`$
$`S_1^{Va}`$ $`=`$ $`{\displaystyle \frac{\lambda _1\lambda _2}{2(2\pi )^3(\omega _{\overline{\theta }}(k)\omega _\theta (k^{}))^{1/2}(\omega _V(p^{}k)+\omega _{\overline{\theta }}(k))^2}},`$
$`S_1^{Nb}`$ $`=`$ $`{\displaystyle \frac{\lambda _1\lambda _2}{2(2\pi )^3(\omega _{\overline{\theta }}(k^{})\omega _\theta (k))^{1/2}(\omega _N(p^{}k)+\omega _\theta (k))^2}},`$
$`S_1^{Na}`$ $`=`$ $`{\displaystyle \frac{\lambda _1^2}{2(2\pi )^3(\omega _\theta (k^{})\omega _\theta (k))^{1/2}(\omega _N(p^{}k)+\omega _\theta (k))^2}},`$
$`S_1^{Vb}`$ $`=`$ $`{\displaystyle \frac{\lambda _2^2}{2(2\pi )^3(\omega _{\overline{\theta }}(k^{})\omega _{\overline{\theta }}(k))^{1/2}(\omega _V(p^{}k)+\omega _{\overline{\theta }}(k))^2}}.`$ (2.147)
It is easy to observe that Eq. (2.74) will be satisfied if we require $`S_m^{}=S_m^{}`$, since, up to the first in $`\lambda `$ the similarity transformation introduced in Eq. (2.67) remains unitary. Equally one could use Eq. (2.75) to obtain $`S^{}`$, it is obtained that $`S_0^{}=S_0^{}`$ and $`S_1^{}=0`$, we will show that the renormalization feature of our model up to this order will remain unchanged, however the effective low-energy Hamiltonian will be different. As was already pointed out, this is because Eq. (2.75) requires a different truncation scheme. The effective Hamiltonian is now produced by plugging the $`S`$ and $`S^{}`$ defined in Eq. (2.146) into Eq. (2.67). With naive power-counting one can identify the potentially divergent terms. At the lower order of expansion, the divergent term is $`0|[V_A,S_0]|0`$, the divergence in this term arises from a double contraction of high-energy fields.
At this step the contributions of the terms $`[V_B,S_0(S_0^{})]`$ and $`[H_1,S_0(S_0^{})]`$ are zero, after projection on to the high-frequency vacuum. There is one other divergent term, $`0|[V_C,S_0^{}]|0`$, but this is harmless and will be cancelled out by $`0|[[H_2,S_0],S_0^{}]|0`$. We thus obtain
$`\delta H(\lambda )`$ $`=`$ $`{\displaystyle \frac{\lambda _1^2}{2(2\pi )^3}}{\displaystyle _\mu ^\mathrm{\Lambda }}{\displaystyle \frac{d^3k}{\omega _\theta (k)(\omega _N(p^{}k)+\omega _\theta (k))}}\left[{\displaystyle d^3p^{}N^{}(p^{})N(p^{})}\right]`$ (2.148)
$``$ $`{\displaystyle \frac{\lambda _2^2}{2(2\pi )^3}}{\displaystyle _\mu ^\mathrm{\Lambda }}{\displaystyle \frac{d^3k}{\omega _{\overline{\theta }}(k)(\omega _V(p^{}k)+\omega _{\overline{\theta }}(k))}}\left[{\displaystyle d^3p^{}V^{}(p^{})V(p^{})}\right].`$
From this expression one can immediately deduce the renormalization factors $`Z_{m_V}`$ and $`Z_{m_N}`$, we take $`\omega _O|k|`$ for $`\mu m_O`$, therefore
$`Z_{M_V}`$ $`=`$ $`1+{\displaystyle \frac{\lambda _1^2}{8\pi ^2}}(\mathrm{\Lambda }\mu ),`$
$`Z_{M_N}`$ $`=`$ $`1+{\displaystyle \frac{\lambda _2^2}{8\pi ^2}}(\mathrm{\Lambda }\mu ).`$ (2.149)
There is no mass renormalization for $`\theta `$ and $`\overline{\theta }`$ and accordingly there are no vacuum polarization type diagrams. Thus $`\theta `$ and $`\overline{\theta }`$ remain unrenormalized, $`Z_\theta =Z_{m_\theta }=Z_{\overline{\theta }}=Z_{m_{\overline{\theta }}}=1`$. The other contribution of $`H^{\text{eff}}`$ at one-loop which are not zero after projecting on to vacuum come from
$$\delta H(\lambda )=0|[[H_1,S_1],S_0^{}]+0|[[H_1,S_1],S_1^{}]|0.$$
(2.150)
The divergent contribution emerges from the first terms, the leading divergence of this expression is logarithmic which means that we can neglect the difference between $`p`$ and $`kp^{}`$ (for the divergent contribution only). After evaluating a momentum integral we finally get,
$`\delta H`$ $`=`$ $`{\displaystyle \frac{\lambda _1^2}{16\pi ^2}}\mathrm{ln}\left[{\displaystyle \frac{\mathrm{\Lambda }}{\mu }}\right]{\displaystyle \omega _V(p^{})V^{}(p^{})V(p^{})}{\displaystyle \frac{\lambda _2^2}{16\pi ^2}}\mathrm{ln}\left[{\displaystyle \frac{\mathrm{\Lambda }}{\mu }}\right]{\displaystyle \omega _N(p^{})N^{}(p^{})N(p^{})}`$ (2.151)
$``$ $`{\displaystyle \frac{\lambda _1^2+\lambda _2^2}{32\pi ^2}}\mathrm{ln}\left[{\displaystyle \frac{\mathrm{\Lambda }}{\mu }}\right]{\displaystyle \frac{\lambda _1}{((2\pi )^3\omega _\theta (k^{}))}V^{}(p^{})N(p^{}k^{})a(k^{})}`$
$``$ $`{\displaystyle \frac{\lambda _1^2+\lambda _2^2}{32\pi ^2}}\mathrm{ln}\left[{\displaystyle \frac{\mathrm{\Lambda }}{\mu }}\right]{\displaystyle \frac{\lambda _2}{((2\pi )^3\omega _{\overline{\theta }}(k^{}))}N^{}(p^{})V(p^{}k^{})b(k^{})}.`$
From this expression we deduce the renormalization factor $`Z_{\lambda _1},Z_{\lambda _2},Z_V`$ and $`Z_N`$:
$`Z_V^2=1+{\displaystyle \frac{\lambda _1^2}{16\pi ^2}}\mathrm{ln}{\displaystyle \frac{\mathrm{\Lambda }}{\mu }},`$
$`Z_N^2=1+{\displaystyle \frac{\lambda _2^2}{16\pi ^2}}\mathrm{ln}{\displaystyle \frac{\mathrm{\Lambda }}{\mu }},`$
$`Z_{\lambda _1}=Z_{\lambda _2}=1+{\displaystyle \frac{\lambda _1^2+\lambda _2^2}{32\pi ^2}}\mathrm{ln}{\displaystyle \frac{\mathrm{\Lambda }}{\mu }}.`$ (2.152)
It is obvious from the equations above that one can define renormalized coupling constants in terms of the bare couplings and wave function renormalization
$$\lambda _i=\lambda _i^0/Z_VZ_N,i=1,2.$$
(2.153)
This definition corresponds for $`\lambda _1=\lambda _2`$ with the renormalization introduced to compute the $`T`$-matrix for the $`N\theta `$ interaction . The one-loop $`\beta `$-function and anomalous dimension $`\gamma `$ are
$`\beta _{\lambda _i}(\lambda _1,\lambda _2)={\displaystyle \frac{\lambda _i}{\mathrm{ln}M}}|_\mathrm{\Lambda }={\displaystyle \frac{\lambda _i}{32\pi ^2}}(\lambda _1^2+\lambda _2^2),i=1,2.`$
$`\gamma _V=1/2{\displaystyle \frac{\mathrm{ln}Z_V}{\mathrm{ln}M}}|_\mathrm{\Lambda }={\displaystyle \frac{\lambda _1^2}{32\pi ^2}},`$
$`\gamma _N=1/2{\displaystyle \frac{\mathrm{ln}Z_N}{\mathrm{ln}M}}|_\mathrm{\Lambda }={\displaystyle \frac{\lambda _2^2}{32\pi ^2}}.`$ (2.154)
Since the fixed points of the theory are the zero solutions of the $`\beta `$-function, one immediately identifies the trivial solution $`\lambda _1=\lambda _2=0`$ (we ignore the nonphysical imaginary solution). It is now obvious from Eq. (2.154) that $`\gamma _V=\gamma _N=0`$ only at trivial solutions of the $`\beta `$-functions. This result is in correspondence with property that for real field theories the $`\gamma `$-function is not zero unless at trivial fixed point of the theory . Now to investigate the behaviour of the theory at high momentum, we must compute the momentum-dependent effective coupling constant $`\lambda _1(k)`$ and $`\lambda _2(k)`$ by
$`\{\begin{array}{cc}& k\frac{d\lambda _1(k)}{dk}=\beta _{\lambda _1}(\lambda _1(k)\lambda _2(k)),\lambda _1(k)|_{k=1}=\lambda _1^{ph}\hfill \\ & ,\hfill \\ & k\frac{d\lambda _2(k)}{dk}=\beta _{\lambda _2}(\lambda _1(k),\lambda _2(k)),\lambda _2(k)|_{k=1}=\lambda _2^{ph}\hfill \end{array}`$ (2.158)
where $`\lambda _1^{ph}`$ and $`\lambda _2^{ph}`$ are dimensionless physical renormalized coupling constants defined at the renormalization scale $`k=1`$. The coupled equations (2.158) can be solved by going to polar coordinates $`r^2(k)=\lambda _1^2(k)+\lambda _2^2(k)`$ and $`\theta (k)=\mathrm{tan}^1\frac{\lambda _1(k)}{\lambda _2(k)}`$,
$`\lambda _1(k)`$ $`=`$ $`{\displaystyle \frac{\overline{r}}{\sqrt{1(16\pi ^2)^1\overline{r}^2\mathrm{ln}k}}}\mathrm{sin}\overline{\theta },`$
$`\lambda _2(k)`$ $`=`$ $`{\displaystyle \frac{\overline{r}}{\sqrt{1(16\pi ^2)^1\overline{r}^2\mathrm{ln}k}}}\mathrm{cos}\overline{\theta },`$ (2.159)
where $`\overline{r}`$ and $`\overline{\theta }`$ denote the value at the renormalization scale. The behaviour of the ELM in the deep-Euclidean region is obtained by allowing $`k\mathrm{}`$. From Eq. (2.159) one observes that $`\lambda _1(k)`$ and $`\lambda _2(k)`$ in this region are imaginary. This means that the effective Hamiltonian is non-hermitian and the theory generates ghost states when the cut-off is removed. The ghost state appears as a pole in $`V`$ and $`N`$-propagators. Since a theory is said to exhibit asymptotic freedom if (i) $`\frac{d\beta }{d\lambda }|_{\lambda (\mathrm{})}<0`$ (ultraviolet stability at the fixed point $`\lambda (\mathrm{})`$ and (ii) $`\lambda (\mathrm{})=lim_k\mathrm{}\lambda (k)=0`$, Eqs. (2.159) indicate that the ELM can not exhibit asymptotic freedom at $`D=4`$.
### 2.11 Conclusion
In this chapter we have reviewed the merits and shortcomings of several approaches for the construction of the effective field theories in the Hamiltonian framework.
We have outlined a new strategy to derive effective renormalized operators. The formulation is not restricted to any quantization scheme (e.g., equal time or light cone). The effective low-frequency operator is obtained by the condition that it should exhibit decoupling between the low- and high-frequency degrees of freedom. All other irrelevant degrees of freedom like many-body states can be systematically eliminated in the same way. We have shown that the similarity transformation approach to renormalization can be systematically classified. The non-hermitian formulation gives a very simple description of decoupling, leading to a partial diagonalization of the high-energy part.
The techniques proposed are known from the coupled cluster many-body theory. We fully utilized Wilsonian Exact renormalization group within the CCM formalism. Our approach invoke neither perturbation nor unitarity transformation. It can be conceived as a topological expansion in number of correlated excitation of the high-energy modes. We showed that our formalism can be solved perturbatively. In this way, it was revealed that diagonalization at first order in coupling constant defines a correct low-energy effective Hamiltonian which is valid up to the order $`\lambda ^3`$.
We showed that non-unitarity representation inherent in our formulation is in favour of economic computation and does not produce any non-hermiticity in the relevant terms. One can show that the non-hermiticity of the effective Hamiltonian is controllable and might appear in higher order which is beyond our approximation or in irrelevant terms which can be ignored in renormalization group sense. We argued that our formulation is free of any small-energy denominator plaguing old-fashion perturbation theory. We showed that the non-hermiticity of the coupled-cluster parametrization leads to the compatibility of the formulation with the Hellmann-Feynman theorem and it also induces a symplectic structure. More importantly, it provides a simple framework for the renormalization of an arbitrary operators. Notice that all these features are connected with each other and one can not give up any of them without spoiling the others. One may conceive that the non-hermiticity adds a auxiliary (non-physical) sector to the physical phase space, thereby, it makes the whole phase space geometrically meaningful and moreover it gives enough room to keep the formulation to be conformed with the cluster decomposition property and Poincaré invariance regardless of a regularization method. There is a long tradition behind such approaches, of course with different motivation, e.g., in the BRST formulation, the phase space is enlarged by anti-commuting canonical coordinates, another example is the bosonization of spin algebraic or fermionic system, where one maps the original Hilbert space of the system into a boson Hilbert space $`^B`$ which turns out to be larger than original Hilbert space, in the sense that physically realizable states in the original space map into a subspace of $`^B`$. Interestingly, in this approach as well, the boson Hamiltonian can be either Hermitian or non-Hermitian .
Notice that our RG method is non-perturbative although we have already shown that perturbation expansion in coupling can be easily implemented . We successfully applied our RG formalism to compute perturbatively the effective Hamiltonian for $`\varphi ^4`$ and extended Lee theory up to two- and one-loop order, respectively. We have employed a sharp cutoff, however this idealization should be removed since generally it may lead to pathologies in renormalization, since it induces non-locality and moreover potentially violates the gauge symmetry.
One of the key features which has not yet been exploited is the non-perturbative aspect of the method; it may well be able to obtain effective degrees of freedom that are very different from the ones occurs at the high-energy scale. This is a promising avenue for future work. Another interesting question is that the connection between our non-perturbative truncation scheme $`SUB(𝒩,)`$ and other non-perturbative scheme e. g., the large $`N_c`$ truncation. A systematic scheme which relates the large $`N_c`$ limit with an approximate RG equation remains yet to be discovered.
## Chapter 3 Renormalization problem in many-body system
### 3.1 Introduction
In this chapter we shall concentrate on the renormalization problem in many-body theory of nuclear matter. As we already pointed out, renormalization of many-body system in a truncated (in number of particles) space is problematic, the so-called Tamm-Dancoff problem. The question is how can one renormalize the many-body problem equations obtained by non-perturbative approaches (such as the CCM, Brueckner’s reaction matrix (or $`G`$-matrix) theory and etc.,) in a truncated Fock space.
In the last chapter we showed that a renormalized effective interaction in small number of particles can be obtained by imposing certain decoupling conditions between the model- and excluded-spaces. In this sense, Feshbach formalism is in contrast with the effective interaction theory since it is not derivable from such decoupling conditions. Notice that the energy-dependence of the Feshbach formalism (and any Green function type formulations, e.g., Schwinger-Dyson resummation, Faddeev approach) emerges from the fact that the effects of the excluded Hilbert space is taken into account by a “quasi-potential”, while in the effective interaction approach the latter is taken into account by imposing a certain decoupling conditions. Therefore a given truncated Hilbert space becomes independent of the remaining sectors and accordingly it can in principle be described by an energy-independent prescription.
In order to clarify the differences between an energy-dependent and an energy-independent formulation, here we investigate how can one resolve the renormalization problem by fully utilizing the Feshbach projection operator technique in the framework of the CCM. Therefore, we pursue an inverse of the EI approach. With a field theoretical consideration, we show that the coupled-cluster formalism by means of Feshbach projection technique leads to a renormalized generalized Brueckner ($`E`$-dependent) theory.
The Feshbach projection technique was introduced to treat nuclear reactions with many channels present. It was originally formulated under the assumption that the number of elementary particles involved is conserved, however clearly this is not that case for field theory. An extension of the Feshbach formalism has been developed for the pion-deuteron system and general pion-nucleus reactions . This technique bears some resemblance to Okubo’s methods , which is consistent with meson field theory, but is developed in terms of an effective Schrödinger equation so as to remain in close contact with conventional nuclear physics. This approach was already pursued by Schütte and Providencia in the framework of the CCM for the Lee model. However, in the Lee model, because of an inherent Tamm-Dancoff approximation (which limits the number of mesons present at any instant), and exact solvability of the model, the issue of renormalizibility of the nuclear matter properties is unclear. Here, we follow their approach in an extended version of Lee model which is not exactly solvable and does not display these trivialities. The renormalization of the extended Lee model in the few-body sector was already investigated in section 2.10.
Notice that the CCM as introduced by Schütte and Providencia based on Rayleigh-Ritz-Variational principle (e.g., see ) is different from the CCM introduced by Arponen and Bishop from the standpoint of variational principle and the Hellman-Feynman theorem . As we have already illustrated in the first two chapters, we believe that the latter has more advantages and is more suitable for a field theoretical application since it is naturally embedded in the modern effective field theoretical framework. Having said that our main goal in this chapter is to introduce other quantum many-body theory techniques and challenge if they are adoptable for a field theoretical application.
### 3.2 The extended Lee Model
The extended Lee model (ELM) is a simple model connecting elementary and composite particles . Although this model is not a chiral model but it exhibits many field-theoretical features. We define four kinds of particles. The $`V`$\- and $`N`$\- particles, two different types of fermions, and the $`\theta `$ scalar boson and the $`\overline{\theta }`$ anti-boson. We take for the Hamiltonian the expression:
$`H^0`$ $`=`$ $`H_0^0+H_I^0,`$
$`H_0^0`$ $`=`$ $`{\displaystyle \underset{\alpha }{}}E_\alpha ^0V_\alpha ^{}V_\alpha +{\displaystyle \underset{\beta }{}}E_\beta ^0N_\beta ^{}N_\beta +{\displaystyle \underset{k}{}}\omega _k(a_k^{}a_k+b_k^{}b_k),`$
$`H_I^0`$ $`=`$ $`{\displaystyle \underset{\alpha \beta k}{}}W_{\alpha \beta k}^0V_\alpha ^{}N_\beta a_k+{\displaystyle \underset{\alpha \beta k^{}}{}}W_{\alpha \beta k^{}}^0N_\beta ^{}V_\alpha b_k^{}+h.c..`$ (3.1)
Here $`a_k,a_k^{}`$ and $`b_k,b_k^{}`$ are the annihilation and creation operators which satisfy boson commutation rules. The $`V_\alpha ^{},V_\alpha `$ and $`N_\beta ^{},N_\beta `$ define the fermion sector and obey the usual anticommutator rules. The $`\alpha ,\beta ,k`$ and $`k^{}`$ are abbreviations for all quantum numbers (such as momentum, spin and isospin, etc). Within our formal investigation, we leave open the specification of $`E_\alpha ^0,E_\beta ^0`$ and $`\omega _k`$. It can be taken as either a relativistic or non-relativistic expression. The bare kinetic energies $`E_\alpha ^0`$ and $`E_\beta ^0`$ are renormalized to $`E_\alpha `$ and $`E_\beta `$ by the interaction. The matrices $`W_{\alpha \beta k}^0`$ and $`W_{\alpha \beta k^{}}^0`$ are the bare interaction strength renormalizing to $`W_{\alpha \beta k}`$ and $`W_{\alpha \beta k^{}}`$, respectively. The interaction strength is defined by the kind of bosons exchanged (the scalar, pseudoscalar or vector bosons). The exchange of higher spin $`(J2)`$ bosons, such as the f(1260), $`\text{A}_2`$(1310), $`\text{f}^{}`$(1514) and g(1680) seems to have little influence on the low-energy $`NN`$ data. The reason is that their poles are located far away from the physical region. In other words, they give rise to the contributions of the very short range which are masked by the very strong repulsion coming from vector meson-exchange. These contributions, however, are essentially masked or parametrized by form factors necessary to regularize the one-boson-exchange diagrams. The interaction term in $`H`$ describes the process
$`VN+\theta ,`$ (3.2)
$`NV+\overline{\theta }.`$ (3.3)
The Lee model can be recovered if we decouple the anti-boson $`\overline{\theta }`$, $`W_{\beta \alpha k^{}}^00`$ (i. e. to remove crossing symmetry) . In the Lee model the virtual process Eq. (3.3) is not included and thus the $`N`$-particle state remains unrenormalized and the model becomes exactly solvable . We have already shown that the ELM can not exhibit asymptotic freedom at space-time dimension $`D`$=4. It is well known that such a model in four dimension exhibit a ghost state as the cutoff is removed. It has been shown that the composite-particle theory, the meson pair theory of Wentzel , can be obtained as a strong-coupling limit of the ELM, in which limit the wave function renormalization constant of the $`V`$-particle vanish . One can find two operators commuting with the Hamiltonian,
$`B`$ $`=`$ $`{\displaystyle \underset{\alpha }{}}V_\alpha ^{}V_\alpha +{\displaystyle \underset{\beta }{}}N_\beta ^{}N_\beta ,`$ (3.4)
$`Q`$ $`=`$ $`{\displaystyle \underset{\beta }{}}N_\beta ^{}N_\beta +{\displaystyle \underset{k^{}}{}}b_k^{}^{}b_k^{}{\displaystyle \underset{k}{}}a_k^{}a_k.`$
Clearly $`B`$ is a baryon number operator and $`Q`$ is a charge operator. In setting up the charge operator, we have assigned the charges 1,0,-1 and 1 to the particles $`N,V,\theta `$ and $`\overline{\theta }`$, respectively. The sectors of the model are labeled by the eigenvalues $`(b,q)`$ of the operators $`(B,Q)`$. The most trivial sectors are: $`(0,0)=|o`$, the physical vacuum is thus the same as the free particle vacuum for this model, $`(0,1)=\{b_k^{}^{}|0=|k^{}\}`$ and $`(0,1)=\{a_k^{}|0=|k\}`$ which define respectively anti-meson and meson states, these states stay unrenormalized. The sectors $`(1,0)`$ and $`(1,1)`$ contain:
$`(1,0)=\{V_\alpha ^{}|0=|\alpha ,N_\beta ^{}a_k^{}|0=|\beta ,k,|\alpha kk^{},\mathrm{}\},`$
$`(1,1)=\{|\beta ,|\alpha k^{},|\beta kk^{},\mathrm{}\}.`$ (3.5)
In the Lee model $`|\beta `$ stays unrenormalized. This makes the model exactly solvable, but this is not the case in ELM. The renormalization of such a model was already studied in section 2.5. The one-loop renormalization of the interaction strength is defined as
$`W_{\alpha \beta k}(W_{\alpha \beta k^{}})=W_{\alpha \beta k}^0(W_{\alpha \beta k^{}}^0)/Z_\alpha Z_\beta ,`$
$`Z_\alpha ^2=1+{\displaystyle \underset{\beta k}{}}{\displaystyle \frac{|W_{\alpha \beta k}^0|^2}{(E_\alpha E_\beta ^0\omega _k)^2}},`$
$`Z_\beta ^2=1+{\displaystyle \underset{\alpha k^{}}{}}{\displaystyle \frac{|W_{\alpha \beta k^{}}^0|^2}{(E_\beta E_\alpha ^0\omega _k^{})^2}}.`$ (3.6)
In one-loop order, the physical neutron state $`|\psi _\alpha `$ is a bound state in the $`(1,0)`$ sector with $`H|\psi _\alpha =E_\alpha |\psi _\alpha `$, where $`E_\alpha `$ is renormalized $`V`$-particle energy. We note that in the limit that the coupling constant vanishes, the $`V`$-particle state $`|\psi _\alpha `$ goes over into $`|\alpha `$. Therefore, the simplest form of a bound state in the $`(1,0)`$ sector at one-loop order can be written as
$$|\psi _\alpha =|\alpha +\underset{\beta ,k}{}\varphi (\beta ,k)|\beta k,$$
(3.7)
where the unknown coefficient $`\varphi (\beta ,k)`$ is determined by requiring that $`|\psi _\alpha `$ be an eigen-function of Hamiltonian with eigenvalue $`E_\alpha `$. After straightforward calculations, one obtains
$`|\psi _\alpha =|\alpha +{\displaystyle \underset{\beta k}{}}{\displaystyle \frac{W_{\alpha \beta k}^0}{E_\alpha E_\beta ^0\omega _k}}|\beta k,`$
$`E_\alpha =E_\alpha ^0+h_\alpha (E_\alpha ),`$
$`h_\alpha (z)={\displaystyle \underset{\beta k}{}}{\displaystyle \frac{|W_{\alpha \beta k}^0|^2}{zE_\beta ^0\omega _k}}.`$ (3.8)
The same argument can be applied for the sector $`(1,1)`$ and one can obtain in lowest order, the physical proton state $`|\psi _\beta `$ as a bound state in this sector which can be shown to obey
$`|\psi _\beta =|\beta +{\displaystyle \underset{\alpha k^{}}{}}{\displaystyle \frac{W_{\alpha \beta k^{}}^0}{E_\beta E_\alpha ^0\omega _k^{}}}|\alpha k^{},`$
$`E_\beta =E_\beta ^0+h_\beta (E_\beta ),`$
$`h_\beta (z)={\displaystyle \underset{\alpha k^{}}{}}{\displaystyle \frac{|W_{\alpha \beta k^{}}^0|^2}{zE_\alpha ^0\omega _k^{}}}.`$ (3.9)
The $`h_\alpha (z)`$ and $`h_\beta (z)`$ are the mass operators, and show the off-shell contribution to the self-energy, see Fig. 3.1. The mass renormalization in the model is now performed by adding corresponding terms as counter terms to the Hamiltonian (these terms will not change the conservation of $`B`$ and $`Q`$ operators). We assume that the parameters of the model are chosen in such a way that there exists one bound state for each $`\alpha `$ and $`\beta `$ . The form factors contained in $`W_{\alpha \beta k}(W_{\beta \alpha k^{}})`$ are assumed to make $`Z_\alpha >0`$, $`Z_\beta >0`$ and consequently the coupling constant renormalization finite.
### 3.3 Nuclear matter equations
We now wish to consider the binding energy problem of $`H`$ for the sector $`(2n,n)`$ with $`n\mathrm{}`$ so-called $`NV`$ matter. We assume that the non-interacting ground-state wave function of this system be the Slater determinant built up by an equal number of $`V`$\- and $`N`$\- particle up to a Fermi momentum $`P_F`$,
$$|\varphi =\underset{\alpha ,\beta <P_F}{}V_\alpha ^{}N_\beta ^{}|0.$$
(3.10)
We denote by $`a(b)`$ occupied $`V`$-particle ($`N`$-particle) states, by $`A(B)`$ the unoccupied states and by $`\alpha (\beta )`$ either states. We write the exact correlated ket ground state, in a coupled cluster formulation, as
$$|\psi =e^{S+R}|\varphi .$$
(3.11)
Here the cluster operator is separated into two parts, one, $`S`$, without mesons which is obviously the same as in the CCM with a phenomenological potential and $`R`$ with mesons contributions which contains the additional parts originating from quantum field theory. Here we define bra ground state as a hermitian conjugate of ket ground state in every level of truncation. Using the symmetry properties Eq. (3.4) and momentum conservation the general form of $`R`$ and $`S`$ are
$`S`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{n/2}{}}}{\displaystyle \underset{j=0}{\overset{n/2}{}}}\widehat{S}_{ij},`$
$`S_{ij}`$ $`=`$ $`{\displaystyle \frac{1}{i!^2j!^2}}{\displaystyle \underset{ABab}{}}A_1..A_iB_1..B_j|S_{ij}|a_1..a_ib_1..b_j_A`$
$`\times `$ $`V_{A_1}^{}..V_{A_i}^{}N_{B_1}^{}..N_{B_j}^{}N_{b_j}..N_{b_1}V_{a_i}..V_{a_1},`$
$`R`$ $`=`$ $`{\displaystyle \underset{k=1}{\overset{n/2}{}}}{\displaystyle \underset{l=0}{\overset{n/2}{}}}{\displaystyle \underset{m=1}{\overset{k}{}}}{\displaystyle \underset{n=1}{\overset{k^{}}{}}}\widehat{R}_{klmn},`$
$`R_{klmn}`$ $`=`$ $`{\displaystyle \frac{1}{k!^2l!^2}}{\displaystyle \underset{ABabKK^{}}{}}A_1..A_{km+n}B_1..B_{l+mn}K_1..K_mK_n^{}..K_1^{}|R_{klmn}`$ (3.12)
$`\times `$ $`|a_1..a_kb_1..b_l_AV_{A_1}^{}..V_{A_{km+n}}^{}N_{B_1}^{}..N_{B_{l+mn}}^{}..N_{b_l}..N_{b_1}`$
$`\times `$ $`V_{a_k}..V_{a_1}a_{k_1}^{}..a_{k_m}^{}b_{k_1^{}}^{}..b_{k_n^{}}^{}.`$
The subscript $`A`$ of the $`R`$\- and $`S`$-amplitudes stands for antisymmetrization of the states. For obtaining the binding energy of $`N`$-$`V`$ matter one should solve the Schrödinger equation by making use of the definitions in Eqs. (3.10)-(3.3), therefore
$$e^Se^RHe^Re^S|\varphi =E|\varphi .$$
(3.13)
Projecting this equation onto the complete orthonormal set of states
$$|\varphi ,N_B^{}V_aa_k^{}|\varphi ,V_A^{}N_bb_k^{}^{}|\varphi ,..,$$
(3.14)
one obtains a system of coupled integral equations of which the first determines the binding energy and others fix the corresponding amplitudes $`R`$ and $`S`$. These coupled integral equations from effective interaction view point are the decoupling conditions and leads to energy-independent prescription. In order to derive the standard Brueckner theory we apply the Rayleigh-Ritz-Variational principle. This type of the CCM formulation was originally proposed by Providencia and Shakin . This is of course different from coupled-cluster theory introduced by Arponen and Bishop from the standpoint of variational principle and the Hellman-Feynman theorem (the so-called NCCM scheme, see chapter 1.2).
One of our goals here is to show that in the consistent truncation scheme the many-body problem does not require further renormalization and all the bare parameters introduced in definition of the wave function can be fixed in the nucleon-nucleon scattering. It is clear that Eq. (3.3) needs to be truncated. Aiming at a two-hole line and one-meson-exchange expansion of the ground-state energy, we choose the following terms for $`S`$ and $`R`$ from Eq. (3.3).
$`S`$ $`=`$ $`{\displaystyle R_{ABab}N_B^{}V_A^{}V_aN_b},`$
$`R`$ $`=`$ $`{\displaystyle R_{aBk}N_B^{}V_aa_k^{}}+{\displaystyle R_{Abk^{}}V_A^{}N_bb_k^{}^{}}+{\displaystyle 1/2R_{BB^{}abk}N_B^{}N_B^{}^{}N_bV_aa_k^{}}`$ (3.15)
$`+{\displaystyle 1/2R_{AA^{}abk^{}}V_A^{}V_A^{}^{}V_aN_bb_k^{}^{}}=R_1+R_2+R_3+R_4.`$
All other term vanish up to this order because of the symmetries of the ELM Hamiltonian. Now we define cluster expansions of expectation value of the Hamiltonian with respect to $`|\psi `$, which is a straightforward generalization of the standard expansion . We neglect three-body and higher-order cluster. The one-body, two-body, etc., correlated wavefunction are defined as
$`|\psi _a=e^{S+R}V_a^{}|0=(1+R_1)V_a^{}|0,`$
$`|\psi _b=e^{S+R}N_b^{}|0=(1+R_2)N_b^{}|0,`$
$`|\psi _{ab}=e^{S+R}N_b^{}V_a^{}|0=(1+S+R)N_b^{}V_a^{}|0.`$ (3.16)
The expectation value of the Hamiltonian is then obtained by and shown in Fig. 3.2
$$E=\frac{\psi |H|\psi }{\psi |\psi }=\underset{a}{}h_a\rho _a+\underset{b}{}h_b\rho _b+\underset{ab}{}h_{ab}\rho _a\rho _b,$$
(3.17)
where the Hamiltonian cluster integrals have been introduced as
$`h_a=\psi _a|H|\psi _a,`$
$`h_b=\psi _b|H|\psi _b,`$
$`h_{ab}=\psi _{ab}|H|\psi _{ab}h_an_bh_bn_a,`$ (3.18)
here $`n_a`$ and $`n_b`$ are the norm of $`|\psi _a`$ and $`|\psi _b`$ respectively. We disregard the other cluster integrals like $`h_{aa^{}}`$ because they would only get contributions from two-mesons intermediate states. This makes sense in consistent one-meson exchange approximation. According to the selective summation made in the expression Eq. (3.17), the occupation numbers $`\rho _a`$ and $`\rho _b`$ should satisfy the following algebraic equations:
$`n_a\rho _a+{\displaystyle \underset{b}{}}n_{ab}\rho _a\rho _b=1,`$
$`n_b\rho _b+{\displaystyle \underset{a}{}}n_{ab}\rho _b\rho _a=1,`$ (3.19)
with notation,
$$n_{ab}=\psi _{ab}|\psi _{ab}n_an_b.$$
(3.20)
The basic idea for the treatment of $`\psi |H|\psi /\psi |\psi `$ is to expand it term by term using the Wick-rule to keep track of all possible contribution. The expansion of $`\psi |H|\psi `$ and $`\psi |\psi `$ can therefore be characterized by a set of suitable diagrams and one can check that the division by $`\psi |\psi `$ cancels out all contribution from disconnected diagrams (the linked-cluster theorem). Now we minimize the quantity $`E`$ with respect to the constrains given by Eq. (3.3) and denote it $`\overline{E}`$, thus we multiply Eq. (3.3) by the Lagrange multipliers $`ϵ_a`$ and $`ϵ_b`$ and subtract them from Eq. (3.17), then by using the definition Eq. (3.3) we find
$`\overline{E}`$ $`=`$ $`{\displaystyle \underset{a}{}}\rho _a(E_a^0ϵ_a)+{\displaystyle \underset{b}{}}\rho _b(E_b^0ϵ_b)+{\displaystyle \underset{a}{}}\rho _aa|R_1^{}(H_0^0ϵ_a)R_1`$ (3.21)
$`+R_1^{}H_I^0+H_I^0R_1|a+{\displaystyle \underset{b}{}}\rho _bb|R_2^{}(H_0^0ϵ_b)R_2+R_2^{}H_I^0+H_I^0R_2|b`$
$`+{\displaystyle \underset{ab}{}}\rho _b\rho _aba|S^{}(H_0^0ϵ_aϵ_b)S+R_3^{}(H_0^0ϵ_aϵ_b)R_3`$
$`+R_4^{}(H_0^0ϵ_aϵ_b)R_4+S^{}H_I^0R+R^{}H_I^0S|ab.`$
It is convenient to treat the $`\rho _a`$ and $`\rho _b`$ as independent variables, therefore we minimize $`\overline{E}`$ respect to $`R_{aBk}^{},R_{Abk^{}}^{},S_{ABab}^{},R_{BB^{}abk}^{},R_{AA^{}abk}^{},\rho _a`$,$`\rho _b`$ which respectively yields:
$`R_{aBk}(E_B^0+\omega _kϵ_a)+W_{aBk}^0+{\displaystyle \underset{Ab}{}}\rho _bS_{ABab}W_{Abk}^0=0,`$ (3.22)
$`R_{Abk^{}}(E_A^0+\omega _k^{}ϵ_b)+W_{Abk^{}}^0+{\displaystyle \underset{Ba}{}}\rho _aS_{ABab}W_{aBk^{}}^0=0,`$ (3.23)
$`AB|(H_0^0ϵ_aϵ_b)S+H_I^0R|ab=0,`$ (3.24)
$`BB^{}k|(H_0^0ϵ_aϵ_b)R_3+H_I^0S|ab=0,`$ (3.25)
$`AA^{}k^{}|(H_0^0ϵ_aϵ_b)R_4+H_I^0S|ab=0,`$ (3.26)
$`ϵ_a=E_a^0+a|R_1^{}(H_0^0ϵ_a)R_1+R_1^{}H_I^0+H_I^0R_1|a+{\displaystyle \underset{b}{}}\rho _bab|(R_1^{}+R_2^{})H_I^0S|ab,`$
$`ϵ_b=E_b^0+b|R_2^{}(H_0^0ϵ_b)R_2+R_2^{}H_I^0+H_I^0R_2|b+{\displaystyle \underset{a}{}}\rho _aab|(R_1^{}+R_2^{})H_I^0S|ab.`$
Equations (3.22)-(3.3) represent a set of unrenormalized Variational coupled integral equations for $`NV`$ matter, consistent with two-hole-line truncation.
### 3.4 The renormalized nuclear matter equations
The method that we will use to reformulate Eqs. (3.22)-(3.3) in order to obtain renormalized equations for $`NV`$ matter is based on the Feshbach projection operator formalism . Projection operator techniques have also been used to analyze several sectors of the Lee model .
The total energy of $`NV`$ matter can be obtained by making use of Eq. (3.21) and Eqs. (3.24)-(3.3) and exploiting the constrains Eq. (3.3);
$$E=\underset{a}{}ϵ_a+\underset{b}{}ϵ_b\underset{ab}{}\rho _a\rho _bab|(R_1^{}+R_2^{})H_I^0S|ab.$$
(3.28)
We can now directly add mass counter terms to the bare Hamiltonian in Eqs. (3.22) and (3.23) to dress masses. We expand the last term in the binding energy Eq. (3.28) in terms of the coupling constants and $`S`$. This form will be used later to achieve full renormalization for the binding energy since there is a subtle relation between $`S`$ and the renormalized coupling constant,
$`E`$ $`=`$ $`{\displaystyle \underset{a}{}}ϵ_a+{\displaystyle \underset{b}{}}ϵ_b{\displaystyle \underset{abk^{}AB}{}}\rho _a\rho _b{\displaystyle \frac{W_{aBk^{}}^0W_{Abk^{}}^0}{ϵ_bE_A\omega _k^{}}}S_{ABab}`$
$``$ $`{\displaystyle \underset{abkAB}{}}\rho _a\rho _b{\displaystyle \frac{W_{Abk}^0W_{aBk}^0}{ϵ_aE_B\omega _k}}S_{ABab}{\displaystyle \underset{aa^{}bk^{}ABB^{}}{}}\rho _a\rho _b\rho _a^{}{\displaystyle \frac{W_{aBk^{}}^0W_{a^{}B^{}k^{}}^0}{ϵ_bE_A\omega _k^{}}}S_{AB^{}a^{}b}^{}S_{ABab}`$
$``$ $`{\displaystyle \underset{abb^{}kAA^{}B}{}}\rho _a\rho _b\rho _b^{}{\displaystyle \frac{W_{Abk}^0W_{A^{}b^{}k}^0}{ϵ_aE_B\omega _k}}S_{A^{}Bab^{}}^{}S_{ABab}.`$
The last two terms in Eq. (3.4) are a three-hole-line contribution which we ignore, consistent with our aim for a two-hole-line expansion. We introduce the projection operators
$`Q={\displaystyle |ABAB|},Q^{}=1/2{\displaystyle |BB^{}kBB^{}k|},`$ (3.30)
$`Q^{\prime \prime }=1/2{\displaystyle |AA^{}k^{}AA^{}k^{}|}.`$
Inspired by the projection technique of feshbach, one can use the above definition to combine Eqs. (3.24)-(3.26),
$$\left[H_0^0z+QH_I^0\left(Q^{}\frac{1}{zH_0^0}Q^{}+Q^{\prime \prime }\frac{1}{zH_0^0}Q^{\prime \prime }\right)H_I^0Q\right]S|ab=QH_I^0(R_1+R_2)|ab,$$
(3.31)
where $`z=ϵ_a+ϵ_b`$. We can show that the right hand side of Eq. (3.31) can be reduced by using Eqs. (3.22) and (3.23)
$`AB|H_I^0R_1|ab={\displaystyle \underset{k}{}}W_{Abk}^0R_{aBk}`$
$`={\displaystyle \underset{k}{}}{\displaystyle \frac{W_{Abk}^0W_{aBk}^0}{ϵ_aE_B\omega _k}}{\displaystyle \underset{b^{}A^{}k}{}}\rho _b^{}{\displaystyle \frac{S_{A^{}Bab}W_{Abk}^0W_{A^{}b^{}k}^0}{ϵ_aE_B\omega _k}},`$ (3.32)
$`AB|H_I^0R_2|ab={\displaystyle \underset{k^{}}{}}W_{aBk^{}}^0R_{Abk^{}}`$
$`={\displaystyle \underset{k^{}}{}}{\displaystyle \frac{W_{aBk^{}}^0W_{Abk^{}}^0}{ϵ_bE_A\omega _k^{}}}{\displaystyle \underset{a^{}B^{}k^{}}{}}\rho _a^{}{\displaystyle \frac{S_{AB^{}ab}W_{aBk^{}}^0W_{a^{}B^{}k^{}}^0}{ϵ_bE_A\omega _k^{}}}.`$ (3.33)
The second term in Eqs. (3.4) and (3.4) are a three-hole-line contribution which are again ignoreable in our approximation. Let us introduce the unrenormalized effective interaction $`\overline{U}(z)`$
$`\alpha \beta |\overline{U}(z)|\alpha ^{^{}}\beta ^{^{}}=\alpha \beta |H_I^0{\displaystyle \frac{1}{z\overline{H}_0}}H_I^0|\alpha ^{^{}}\beta ^{^{}}_{\text{linked}}`$
$`={\displaystyle \underset{k}{}}{\displaystyle \frac{W_{\alpha \beta ^{^{}}k}^0W_{\alpha ^{^{}}\beta k}^0}{zϵ_\beta ϵ_\beta ^{^{}}\omega _k}}{\displaystyle \underset{k^{}}{}}{\displaystyle \frac{W_{\alpha \beta ^{^{}}k^{}}^0W_{\alpha ^{^{}}\beta k^{}}^0}{zϵ_\alpha ϵ_\alpha ^{^{}}\omega _k^{}}},`$
$`\overline{H}_0={\displaystyle \underset{\alpha }{}}ϵ_\alpha V_\alpha ^{}V_\alpha +{\displaystyle \underset{\beta }{}}ϵ_\beta V_\beta ^{}V_\beta +{\displaystyle \underset{k}{}}\omega _k(a_k^{}a_k+b_k^{}b_k).`$ (3.34)
The mass renormalization terms has been taken into account by using renormalized masses in the propagators. (Notice that $`ϵ_A=E_A`$ and $`ϵ_B=E_B`$). This definition correspondences to the quasi-potential in a Lippmann-Schwinger type equation if one replaces $`H_0\overline{H}_0`$. Having made use of defined effective interaction $`\overline{U}(z)`$ one can show
$$AB|H_I^0(R_1+R_2)|abAB|\overline{U}(z)|ab.$$
(3.35)
This expression is consistent with the two-hole-line approximation. Now we decompose the “p-p” and “n-n” interaction involved in Eq. (3.31) after adding mass counter terms in bare Hamiltonian,
$$QH_I^0\left(Q^{}\frac{1}{zH_0}Q^{}+Q^{\prime \prime }\frac{1}{zH_0}Q^{\prime \prime }\right)H_I^0Q=Q\overline{U}(z)Q+h(z)+\overline{q}(z),$$
(3.36)
where the operator $`h(z)`$ and $`\overline{q}(z)`$ are defined via Eqs. (3.2,3.2) as
$`h(z)|AB=[h_A(zE_B)+h_B(zE_A)]|AB,`$ (3.37)
$`\overline{q}(z)|AB=\left({\displaystyle \underset{bk}{}}{\displaystyle \frac{|W_{Abk}^0|^2}{zE_bE_B\omega _k}}+{\displaystyle \underset{ak^{}}{}}{\displaystyle \frac{|W_{aBk^{}}^0|^2}{zE_aE_A\omega _k^{}}}\right)|AB.`$
In diagrammatic language Eq. (3.36) contains self-energy diagrams which contain unoccupied intermediate states because of the operators $`Q^{},Q^{\prime \prime }`$, This is achieved by adding $`\overline{q}(z)`$ to $`h(z)`$. We now introduce the operator $`B(z)`$ to accomplish coupling constant renormalization. Its structure emerges from renormalization of finite sectors of ELM,
$$B(z)|\alpha \beta =Z_\alpha ^2(zE_\beta )Z_\beta ^2(zE_\alpha )|\alpha \beta ,$$
(3.38)
with notations,
$`Z_\alpha ^2(z)=1+{\displaystyle \underset{\beta k}{}}{\displaystyle \frac{|W_{\alpha \beta k}^0|^2}{(E_\alpha E_\beta ^0\omega _k)(zE_\beta ^0\omega _k^{})}},`$
$`Z_\beta ^2(z)=1+{\displaystyle \underset{\alpha k^{}}{}}{\displaystyle \frac{|W_{\alpha \beta k^{}}^0|^2}{(E_\beta E_\alpha ^0\omega _k^{})(zE_\alpha ^0\omega _k^{})}},`$ (3.39)
here $`z=E_\alpha +E_\beta `$. The benefit of $`B(z)`$ can be manifested by following factorization property
$$Q(zH_0^0h(z))Q=Q(zH_0)B(z)Q.$$
(3.40)
One may use $`B(z)`$ to show that $`\overline{U}(z)`$ is related to renormalized effective $`NV`$ potential
$$U(z)=B(z)^{1/2}\overline{U}(z)B(z)^{1/2}.$$
(3.41)
Therefore, the renormalized effective two-body $`NV`$ potential can be readily found ( see Fig. 3.2)
$`\alpha \beta |U(z)|\alpha ^{^{}}\beta ^{^{}}={\displaystyle \underset{k}{}}{\displaystyle \frac{r_\alpha (zE_\beta )r_\beta (zE_\alpha )W_{\alpha \beta ^{^{}}k}^{}W_{\alpha ^{^{}}\beta k}r_\alpha ^{^{}}(zE_\beta ^{^{}})r_\beta ^{^{}}(zE_\alpha ^{^{}})}{zϵ_\beta ϵ_\beta ^{^{}}\omega _k}}`$
$`{\displaystyle \underset{k^{}}{}}{\displaystyle \frac{r_\alpha (zE_\beta )r_\beta (zE_\alpha )W_{\alpha \beta ^{^{}}k^{}}W_{\alpha ^{^{}}\beta k^{}}^{}r_\alpha ^{^{}}(zE_\beta ^{^{}})r_\beta ^{^{}}(zE_\alpha ^{^{}})}{zϵ_\alpha ϵ_\alpha ^{^{}}\omega _k^{}}},`$ (3.42)
where we have introduced twice-subtracted “dressing factor” defining as
$$r_\alpha (z)=\frac{Z_\alpha (E_\alpha )}{Z_\alpha (z)},r_\beta (z)=\frac{Z_\beta (E_\beta )}{Z_\beta (z)}.$$
(3.43)
Dressing factor appears independent of truncation made and describes off shell correction of a self-energy of $`V`$ and $`N`$ particle. In Lee model $`r_\beta (z)=1`$, since $`N`$-particle stay unrenormalized. Twice-subtracted dressing factors remain finite even without form factor, due to the occurrence of a cubic energy denominator, this can be verified by expanding Eq. (3.43) using relativistic expression for $`E_\alpha ,E_\beta ,\omega _k,\omega _k^{}`$ and interaction strength. Now one can rewrite Eq. (3.31), by making use of Eqs. (3.35, 3.36) and exploiting the factorization property Eq. (3.40)
$$\left(H_0z+q(z)+U(z)\right)B(z)^{1/2}S|ab=QU(z)B(z)^{1/2}|ab,$$
(3.44)
where $`q(z)`$ is renormalized two-body operators obtained from Eq. (3.37)
$`q(z)|AB`$ $`=({\displaystyle \underset{bk}{}}{\displaystyle \frac{|W_{Abk}|^2r_A^2(zE_B)r_B^2(zE_A)}{zE_bE_B\omega _k}}`$ (3.45)
$`+{\displaystyle \underset{ak^{}}{}}{\displaystyle \frac{|W_{aBk^{}}|^2r_A^2(zE_B)r_B^2(zE_A)}{zE_aE_A\omega _k^{}}})|AB.`$
It is observed the operator $`B(z)`$, transfers unrenormalized correlation function to renormalized one. This is due to the simple structure of vertices renormalization in this model. We now turn to Eq. (3.3) for $`ϵ_a`$ and $`ϵ_b`$, one may use Eqs. (3.22,3.23) to obtain
$`ϵ_a`$ $`=`$ $`E_a^0+{\displaystyle \underset{Bk}{}}R_{aBk}W_{aBk}^0+{\displaystyle \underset{bk^{}AB}{}}\rho _bR_{Abk^{}}^{}W_{aBk^{}}^0S_{ABab}`$
$`=`$ $`E_a^0+{\displaystyle \underset{Bk}{}}{\displaystyle \frac{|W_{aBk}^0|^2}{ϵ_aE_B^0\omega _k}}+{\displaystyle \underset{bkAB}{}}\rho _b{\displaystyle \frac{W_{aBk}^0W_{Abk}^0}{ϵ_aE_B^0\omega _k}}S_{ABab}`$
$`+`$ $`{\displaystyle \underset{bk^{}AB}{}}\rho _b{\displaystyle \frac{W_{aBk^{}}^0W_{Abk^{}}^0}{ϵ_bE_A^0\omega _k^{}}}S_{ABab}+{\displaystyle \underset{a^{}bk^{}ABB^{}}{}}\rho _b\rho _a^{}{\displaystyle \frac{W_{aBk^{}}^0W_{a^{}B^{}k^{}}^0}{ϵ_bE_A^0\omega _k^{}}}S_{AB^{}a^{}b}^{}S_{ABab},`$
$`ϵ_b`$ $`=`$ $`E_b^0+{\displaystyle \underset{Ak^{}}{}}R_{Abk^{}}W_{Abk^{}}^0+{\displaystyle \underset{akAB}{}}\rho _aR_{aBk}^{}W_{bAk}^0S_{ABab}`$
$`=`$ $`E_b^0+{\displaystyle \underset{Ak^{}}{}}{\displaystyle \frac{|W_{Abk^{}}^0|^2}{ϵ_bE_A^0\omega _k^{}}}+{\displaystyle \underset{ak^{}AB}{}}\rho _a{\displaystyle \frac{W_{Abk^{}}^0W_{aBk^{}}^0}{ϵ_bE_A^0\omega _k^{}}}S_{ABab}`$
$`+`$ $`{\displaystyle \underset{bkAB}{}}\rho _a{\displaystyle \frac{W_{bAk}^0W_{aBk}^0}{ϵ_aE_B^0\omega _k}}S_{ABab}+{\displaystyle \underset{ab^{}kA^{}AB}{}}\rho _a\rho _b^{}{\displaystyle \frac{W_{Abk}^0W_{A^{}b^{}k}^0}{ϵ_aE_B^0\omega _k}}S_{A^{}Bab^{}}^{}S_{ABab}.`$
The last term in $`ϵ_a(ϵ_b)`$ in Eq. (3.4) are a three-hole-line contribution. From Eqs. (3.2) and (3.2) it is obvious that the first two terms in $`ϵ_a(ϵ_b)`$ in Eq. (3.4) are the renormalized energy of $`N(V)`$ particles if we add the contribution of intermediate occupied $`a(b)`$ states in self-energy. The corresponded terms should be subtracted which contribute to the definition of $`\overline{ϵ_a}(\overline{ϵ_b})`$. Having introduced mass counter terms, one can eliminate $`S`$ by using Eq. (3.44) and obtain the renormalized equation for $`E`$, $`ϵ_a`$ and $`ϵ_a`$,
$`E={\displaystyle \underset{a}{}}ϵ_a+{\displaystyle \underset{b}{}}ϵ_b{\displaystyle \underset{ab}{}}\rho _b\rho _aab|G(ϵ_a+ϵ_b)|ab+{\displaystyle \underset{ab}{}}\rho _a\rho _bab|U(ϵ_a+ϵ_b)|ab,`$
(3.47)
$`ϵ_a=E_a+\overline{ϵ_a}+{\displaystyle \underset{b}{}}\rho _bab|G(ϵ_a+ϵ_b)|ab,`$ (3.48)
$`ϵ_b=E_b+\overline{ϵ_b}+{\displaystyle \underset{a}{}}\rho _aab|G(ϵ_a+ϵ_b)|ab,`$ (3.49)
$`G(z)=U(z)+U(z){\displaystyle \frac{Q}{zH_0q(z)}}G(z),`$ (3.50)
$`\overline{ϵ_a}={\displaystyle \underset{bk}{}}{\displaystyle \frac{|W_{abk}|^2r_a^2(ϵ_a)r_b^2(ϵ_b)}{ϵ_aE_b\omega _k}}{\displaystyle \underset{b}{}}\rho _bab|U(ϵ_a+ϵ_b)|ab,`$ (3.51)
$`\overline{ϵ_b}={\displaystyle \underset{ak^{}}{}}{\displaystyle \frac{|W_{abk^{}}|^2r_a^2(ϵ_a)r_b^2(ϵ_b)}{ϵ_bE_a\omega _k^{}}}{\displaystyle \underset{a}{}}\rho _aab|U(ϵ_a+ϵ_b)|ab,`$ (3.52)
where $`ϵ_a`$ and $`ϵ_b`$ are given by self-consistent equation of the Brueckner type and $`G(z)`$ is the solution of a Brueckner-Bethe-Goldstone equation .
### 3.5 Conclusion
The correction to the standard many-body scheme emerges in this model in different steps which are connected with the following physical effects. We have to replace the $`V^{OBE}`$ by $`U(z)`$, which accounts for the renormalization effects of the boson propagating in the $`NV`$ matter system. This can be understood since fermions feel the average potential given by $`ϵ_a`$ and $`ϵ_b`$ during the time when a boson is exchanged in $`NV`$ matter . In scattering formalism ($`T`$-matrix) we have $`z=E_\alpha +E_\beta `$, the total energy of $`NV`$ system, whereas there is a medium effect in $`NV`$ matter bringing the $`z`$-value to $`z=ϵ_a+ϵ_b`$. The energy denominator of the Brueckner Eq. (3.50) contains the correction $`q(z)`$ which is not present for standard $`NV`$ scattering. This correction is due to the effect of the Pauli principle on the self-energy of the fermions. The comparison between many-body solution in the Lee model and the ELM reveals following distinctions:
The simplest correction in the ELM is due to the self-energy diagrams of $`N`$-particle and Pauli principle on this diagrams which are manifested through the occurrence of the corresponding dressing factor $`r_\beta (z)`$ and $`q(z)`$ respectively, whereas these corrections in the Lee model do not appear because $`N`$-particle remains unrenormalized . In the Lee model the lowest order of $`ϵ_b`$ would appear in a three-hole-line expansion while in the ELM this term arises in two-hole-line contribution on an equal footing with lowest order of $`ϵ_a`$.
The $`\rho _b`$ renormalization correction of the Lee model (see Fig. 3.4) goes to contribution of mass renormalization of $`N`$-particle in the ELM. Therefore the renormalization correction of occupation numbers gets contribution from higher order which are not generated in our approximation. It is notable that the ELM due to possessing the crossing symmetry preserves the symmetry between $`V`$ and $`N`$ part of the renormalization correction, whereas this is not the case for many-body problem within Lee model.
This presentation is conclusive that coupled-cluster theory without the decoupling property in systematic truncation scheme leads to derivation of generalized Brueckner ($`E`$-dependent) theory which includes renormalization correction originating from medium effect. We showed as well that a combination of the coupled-cluster theory and Feshbach projection technique provides a powerful method to renormalize quantum many-body problem in a truncated Fock space.
### 3.6 Some final remarks
We have employed the coupled-cluster theory in various versions and investigated the merits and short-comings of such techniques for field theoretical applications. We showed that the CCM version introduced by Arponen and Bishop can be easily adopted with Wilsonian renormalization group and provides very strong tools for describing (non)-perturbative phenomena (please see the conclusion of the last chapter 2.11). On the other hand, the CCM version introduced by Schütte and Providencia can be equipped with the Feshbach projection formalism and produces a renormalized generalized Brueckner theory. One should bear in mind that in the latter, the famous “small-energy denominator” problem is not avoidable and is not at all clear how systematically high-energy modes can be integrated out for a very complicated system.
In the rest of this work, we employ an effective QCD model as can be constructed from the techniques of the first part of this thesis and address with greater details, non-perturbative phenomenological phenomena such as nucleon and diquark solutions, confinement and spontaneous chiral symmetry breaking.
## Chapter 4 QCD properties and effective quark chiral models
It is believed that the strong interactions are described by a quantum field theory known as quantum chromodynamics (QCD) . QCD from many aspects is an unique theory. Quantum electrodynamics (QED), and its expansion to the electroweak standard model of particle physics, is also a quantum field theory. However, QED breaks down at short distances and is not well-defined. QED is renormalizable theory but it loses all his credibility as we approach to Landau pole. On the other hand if the cutoff goes to infinity, QED becomes trivial. QED is not the only theory with a Landau pole problem, every theory which is not asymptotically free suffers from this problem. The quantum field theory of gravity obtained from general relativity suffers from nonrenormalizability. QCD is the only known theory which is free from such problems. QCD only needs few parameters to be defined completely, one universal coupling strength and one mass for each kind of quark.
In following chapter we introduce part of QCD which is relevant to nuclear physics. We refrain from discussing all details since it can be found in many quantum-field theory textbooks, see e.g., Refs. .
### 4.1 QCD Symmetries
In this section we will introduce the underlying symmetries of QCD. The Lagrangian of QCD is given by ,
$$=\overline{q}(i\gamma ^\mu _\mu m^0)q\frac{1}{4}(F_{\mu \nu }^a)^2+g\overline{q}\gamma ^\mu A_\mu q,$$
(4.1)
where $`q`$ is the quark field which is defined in the fundamental representation of the color and flavor group, and the conjugate Dirac field is defined as $`\overline{q}=q^{}\gamma ^0`$. The gluon field matrix $`A^\mu =A_\mu ^a\lambda ^a/2`$ is defined in the fundamental $`SU(N_c=3)`$ representation, $`\lambda ^a`$ being the generators of the gauge group which satisfies $`[\lambda ^a/2,\lambda ^b/2]=if^{abc}\lambda ^c/2`$. We define $`g`$ as strong coupling constant. The field strength $`F_{\mu \nu }^a`$ is given by
$$F_{\mu \nu }^a=_\mu A_\nu ^a_\nu A_\mu ^a+gf^{abc}A_\mu ^bA_\mu ^c.$$
(4.2)
The non-Abelian nature of QCD is manifested by the quadratic term in the gauge field strength. The color and flavour indices of the quark field are suppressed. $`m^0`$ is the current quark mass which is not directly observable if QCD confines quarks. The current quark mass is color-independent and can be brought diagonal in flavour space. There are six flavours of quarks, each of which has a different mass. The three light quarks are called up (u), down (d) and strange (s), while the three heavy quarks are called charm (c), bottom (b) and top (t). The following values for the light current quark masses are found from the Particle Data group ,
$$m_u^0=2\text{to}8\text{MeV},m_d^0=5\text{to}15\text{MeV},m_s^0=100\text{to}300\text{MeV}.$$
(4.3)
Notice that the quark masses are renormalization-scheme dependent. The above values are obtained in a subtraction scheme at a renormalization scale $`𝒪(1\text{GeV})`$. In addition to flavour, quarks carry another quantum number known as colour. Each quark comes in three colours, red, green and blue.
The Lagrangian Eq. (4.1) has a large classical symmetry: we have the local gauge symmetry $`SU(N_c)`$ by construction,
$`qU_cq,\overline{q}\overline{q}U_c^{},U_c(x)=\mathrm{exp}(i\theta ^a(x)({\displaystyle \frac{\lambda ^a}{2}})_c),`$
$`A_\mu U_cA_\mu U_c^{}{\displaystyle \frac{1}{g}}U_ci_\mu U_c^{}.`$
We have also global flavour symmetry which does not affect the gluon fields,
$$qU_Vq,\overline{q}\overline{q}U_V^{},U_V=\mathrm{exp}(i\theta _V^a(\frac{\lambda ^a}{2})_F).$$
(4.4)
where $`(\frac{\lambda ^a}{2})_F`$ denotes the generators of the flavor group $`U(N_f)`$ and $`N_f`$ denotes the number of flavors. The above symmetry is referred to as vector flavor symmetry $`U_V(N_f)`$. When the generator matrix is taken unit matrix we have $`U_V(1)`$ symmetry associated with conservation of baryon number. There is another global symmetry which is exact at $`m^0=0`$, namely chiral symmetry. This symmetry is very similar to vector flavor symmetry, apart from an extra factor of $`\gamma _5`$ in the generator of the transformation.
$$qU_Aq,\overline{q}\overline{q}U_A,U_A=\mathrm{exp}\left(i\gamma _5\theta _A^a(\frac{\lambda ^a}{2})_F\right).$$
(4.5)
Notice that due to the factor $`\gamma _5`$ the quark field and its conjugate partner are transformed by the same matrix in contrast to vector transformation Eq. (4.4). This transformation is called the axial-vector transformation and can be combined with the vector transformation to define a bigger symmetry at chiral $`m^0=0`$ which is then called chiral symmetry $`U_V(N_f)\times U_A(N_f)`$. One may alternatively define right- and left-handed quark fields by following transformation
$$q_L=\frac{1\gamma _5}{2}q,q_R=\frac{1+\gamma _5}{2}q,$$
(4.6)
The right- and left-handed massless fermions are eigenvalues of the helicity or chirality (with eigenvalue $`\pm 1`$) and are not mixed together. The chiral symmetry can be equivalently written as $`U_L(N_f)\times U_R(N_f)`$.
It is believed that intermediate-energy hadronic physics, say over range of energy MeV-GeV is adequately described by the dynamics of the lowest-mass quarks, $`u`$ and $`d`$. The overall classical symmetry of the Lagrangian with $`N_f=2`$ becomes
$$SU(N_c)_{\text{local}}\times (SU(2)_L\times SU(2)_R\times U(1)_V\times U(1)_A)_{\text{global}},$$
(4.7)
Not all above-mentioned symmetries survive quantization. Particles with opposite helicity are related by a parity transformation, therefore in a chirally symmetric world the hadrons should come in parity doublets. However, in a real life we do not observe such degeneracy. Therefore, one can conclude that chiral symmetry is not realized in the ground state and chiral symmetry is spontaneously broken. The Goldstone theorem tell us that the spontaneous breaking of a continuous global symmetry implies the existence of associated massless spinless particles. This indeed confirmed due to the existence of the light pseudoscalar mesons in nature (pions, kaons and etas) as the corresponding Goldstone bosons . Moreover, the existence of quark condensate $`\overline{q}q`$ implies that the $`SU(N_f)_L\times SU(N_f)_R`$ symmetry is spontaneously broken down to $`SU(N_f)_V`$. Therefore one may conceive QCD quark condensate as an order parameter for chiral symmetry breaking. The concept of spontaneous broken chiral symmetry is the cornerstone in the understanding of the low-energy hadronic spectrum.
The $`U(1)_A`$ symmetry implies that all hadrons should come with opposite parity partners. However, this is not the case, therefore this symmetry must be broken somehow. If the spontaneous symmetry breaking mechanism works here, then one should observe Goldstone boson associated with $`U(1)_A`$, namely an $`I=0`$ pseudoscalar meson having roughly the same mass as the pion. Surprisingly there is no such Goldstone boson. This problem is sometime called $`U(1)_A`$ puzzle. It turned out that the $`U(1)_A`$ symmetry is explicitly broken by quantum effects. This effect is known as the axial anomaly . It was shown by ’t Hoof that due to instanton effects, the $`U(1)_A`$ symmetry is not manifested in nature .
Finally, at $`m^0=0`$, the QCD Lagrangian is invariant under a scale transformation which is called dilatational symmetry:
$$q(x)ϵ^{3/2}q(ϵ^1x),A_\mu ^a(x)ϵA_\mu ^a(ϵ^1x),x_\mu ϵ^1x_\mu .$$
(4.8)
This symmetry is again broken at quantum level due to the trace anomaly .
### 4.2 Non-perturbative features of QCD
In this section we shall recapitulate the most important features of QCD which are not accessible perturbatively. Let us firstly elaborate why perturbation theory in terms of coupling $`g`$ can not be used in the low-energy regime of the theory. Having introduced the gauge fixing term and an associated ghost term by means of the Faddeev-Popv procedure , one can carry out perturbation theory in terms of coupling. Due to the renormalization process, a renormalization scale $`\mu `$ enters the algebra . Therefore the running coupling is described by the RG equation
$$\frac{dg(\mu )}{d\mathrm{ln}\mu }=\beta (g).$$
(4.9)
Where the coupling is small, the $`\beta `$ function can be computed perturbatively,
$$\beta (g)=\frac{\beta _0}{(4\pi )^2}g^3\frac{\beta _1}{(4\pi )^4}g^5+\mathrm{},$$
(4.10)
with
$$\beta _0=11\frac{2}{3}N_f,\beta _1=102\frac{38}{3}N_f.$$
(4.11)
Therefore, one can readily calculate the effective running coupling
$$\alpha _s(\mu )=\frac{g^2(\mu )}{4\pi }=\frac{12\pi }{(332N_f)\mathrm{ln}(\mu ^2/\mathrm{\Lambda }^2)}\times [1\frac{(918114N_f)\mathrm{ln}(\mathrm{ln}(\mu ^2/\mathrm{\Lambda }^2))}{(332N_f)^2\mathrm{ln}(\mu ^2/\mathrm{\Lambda }^2)}],$$
(4.12)
where $`\mathrm{\Lambda }`$ is a scale parameter of QCD and depends on the subtraction scheme and the number of active flavours,
$$\mathrm{\Lambda }_{\overline{MS}}^{(5)}=(208_{23}^{+25})\text{MeV},$$
(4.13)
where the symbol $`\overline{MS}`$ stands for minimal subtraction scheme and the superscripts indicate the number of active flavours. This value is taken from an analysis of the various high energy processes, see Ref. . The most striking feature of the running coupling is that it decreases logarithmically as $`\mu `$ increases. Therefore perturbation theory works very well for large $`\mu `$. This phenomenon is called asymptotic freedom . However, if $`\mu `$ is near $`\mathrm{\Lambda }_{\overline{MS}}`$, perturbation theory does not work anymore and non-perturbative phenomena enter the stage. Admittedly, there is no unambiguous method available to connect small and large distances in QCD.
One of the most important non-perturbative features of QCD is dynamical chiral symmetry breaking which is responsible for generation of a quark mass from nothing<sup>1</sup><sup>1</sup>1There is another very different way to generate mass from vacuum, the so-called Casimir effect which originates from the response of vacuum in the presence of non-perturbative boundary condition. The existence of boundary conditions in quantum field theory is not always free of problems (see e. g., ).. In order to show that this phenomenon is purely non-perturbative, we employ the QCD gap equation ,
$$S(p)^1=(i\gamma .p+m^0)+\frac{d^4q}{(2\pi )^4}g^2D_{\mu \nu }(pq)\frac{\lambda ^a}{2}\gamma _\mu S(q)\mathrm{\Gamma }_\nu ^a(p,q),$$
(4.14)
where $`m^0`$ and $`g`$ are the current-quark bare mass and the coupling constant, respectively. $`D_{\mu \nu }(pq)`$ is the dressed-gluon propagator and $`\mathrm{\Gamma }_\nu ^a(p,q)`$ is the dressed-quark-gluon vertex. The general solution of the gap equation is a dressed-quark propagator of the form
$$S(p)=\frac{1}{i\gamma .pA(p^2)+B(p^2)}=\frac{Z(p^2)}{i\gamma .p+M(p^2)}.$$
(4.15)
One may now use the gap equation to work out the fermion self-energy perturbatively . One obtains,
$$B(p^2)=m^0\left(1\frac{\alpha }{\pi }\mathrm{ln}(p^2/m^2)+\mathrm{}\right).$$
(4.16)
It is observed that at all orders of loop expansion terms are proportional to the current-quark mass and consequently vanish as $`m^00`$. Therefore, no mass (the quark mass is defined as a pole of the dressed-quark propagator) is generated at current-quark mass equal to zero, i.e. the dynamical chiral symmetry breaking is impossible in perturbation theory and there is no mixing between left- and right-handed quarks “perturbatively”. Notice that apart from the trivial solution $`B(p^2)=0`$ at $`m=0`$, a non-trivial solution $`B(p^2)0`$ can indeed be found at the chiral point, albeit accessible non-perturbatively. The renormalization effect is not included in Eq. (4.14), but it does not change the above argument . As we already mentioned, there is a close relationship between the generation of the quark mass, $`B(p^2)0`$, and the fact that $`\overline{q}q0`$. The quark condensate in QCD is given by the trace of the full quark propagator Eq. (4.15),
$$\overline{q}q=i\underset{yx}{lim}\mathrm{Tr}S(x,y).$$
(4.17)
Notice that since $`\overline{q}q`$ is a gauge invariant object, one may take any gauge to obtain the dressed quark propagator which has a general form as equation (4.15). It is obvious when we have $`B(p^2)=0`$, never does the quark condensate take place, simply because of the identity $`\mathrm{Tr}\gamma _\mu =0`$. It has been shown in Landau gauge that the dynamical quark mass, $`M(p^2)`$ is large in infrared, $`M(0)0.5`$ GeV, but is power-law suppressed in the ultraviolet ,
$$M(p^2)^{\text{large}p^2}=\frac{2\pi ^2\gamma _m}{3}\frac{\overline{q}q^0}{p^2\left(\frac{1}{2}\mathrm{ln}[\frac{p^2}{\mathrm{\Lambda }_{QCD}}]\right)^{1\gamma _m}},$$
(4.18)
where $`\gamma _m=12/(332N_f)`$ is the mass anomalous dimension and $`\overline{q}q^0`$ is the renormalization group invariant vacuum quark condensate. The dressed-quark mass-function Eq. (4.18) is a longstanding prediction of the Dyson-Schwinger equation and has been recently confirmed by quenched lattice QCD . It has been shown in many non-perturbative approaches that the emergence of a dynamical quark mass leads to the non-vanishing of quark condensate and vice versa, e.g., see chapter 6.
Another important non-perturbative feature of QCD is color confinement . Loosely speaking, confinement is defined as the absence of any colored object in nature. But it is possible that there exists a composite colored particle which can form colorless bound states with another colored particle like quarks. The color confinement is still not properly understood, and a clear and indisputable mechanism responsible for this effect yet remains to be discovered <sup>2</sup><sup>2</sup>2The Clay Foundation is offering $`\$1`$ million prize to anyone who can provide a mathematical proof of confinement.. Confinement originates non-perturbatively, since it is associated with a linear potential with a string tension $`\sigma \mathrm{\Lambda }^2e^{{\scriptscriptstyle {\scriptscriptstyle \frac{dg}{\beta (g)}}}}`$ which is obviously non-perturbative in the coupling<sup>3</sup><sup>3</sup>3Note that the string picture of quark confinement is not free of flaws, since string breaking will occur once the potential energy approaches the quark pair creation threshold..
One may wonder if there is a non-trivial solution for gap equation $`B(p^2)0`$ which gives rise to a pole of the quark propagator, which would contradict QCD confinement since the quark is colored <sup>4</sup><sup>4</sup>4It is well-known that for confinement it is sufficient that no colored Schwinger function possesses a spectral representation. It is equivalent to say that all colored Schwinger functions violate reflection positivity . This is one way of realization of QCD confinement. There are in fact many different ways that the confinement can be realized such as monopole condensation effect, infrared enhancement of the ghost propagator etc. For a review of this subject see Ref... Indeed this is one of the subtle point in every QCD model and can not be easily resolved. In principle, there will be a long-range force between massive quarks to confine them and also a short range spin-spin interaction between massive dressed quarks. The former will modify the low momentum part of the propagator to remove the quark from being on-shell. Actually, this describes a phenomenologically motivated picture of a constituent quark model based on the dynamical symmetry breaking. Having said that it is very hard to incorporate the dynamical symmetry breaking and the confinement into a QCD model. In fact, many models constructed to describe the low-energy properties of hadrons are assumed to be only dominated by the quark flavor dynamics and dynamical symmetry breaking and are indeed reliable only at intermediate scales, between confinement scale few hundred MeV up to a scale about 1 GeV.
### 4.3 Effective low-energy quark interaction
Physical hadrons are colorless objects and their properties seem to be determined by the flavor dynamics which is induced by an effective QCD interaction. The first step toward a construction of such effective theory is to integrate out gluonic degrees of freedom, then by standard bozonization and hadronization methods derive the desired effective low-energy theories based on relevant degrees of freedom. There have been many attempts to attack this difficult problem. Two very well established methods are the global color model approach introduced by Cahill and Roberts , and the so-called field strength approach introduced by Reinhardt and collaborators . Our main goal in this section is to give a taste of such approaches and focus only on the main themes rather than details. First we rewrite the quark-gluon interaction term in QCD Lagrangian Eq (4.1) by rewriting
$$\overline{q}\gamma ^\mu A_\mu q=A_\mu ^aJ_a^\mu ,J_a^\mu =\overline{q}\frac{\lambda ^a}{2}\gamma ^\mu q.$$
(4.19)
The full quantum theory of QCD is solved by computing the functional integral describing the vacuum-to-vacuum transition amplitude,
$$Z_{QCD}=𝒟q𝒟\overline{q}𝒟A_\mu ^ae^{i{\scriptscriptstyle d^4x_{QCD}}}.$$
(4.20)
In order to handle the gluonic part of the QCD functional integral, one first has to define gauge inequivalent orbits by using a standard gauge fixing procedure and the Faddeev-Popov method (in what follows we assume that this procedure has already been carried out). We now split the above generating functional integral as
$$Z_{QCD}=𝒟q𝒟\overline{q}\mathrm{exp}\left(id^4x\overline{q}(i\gamma ^\mu _\mu m^0)q+\mathrm{\Gamma }[J]\right),$$
(4.21)
where the gluon part of action is defined in
$$\mathrm{\Gamma }[J]=\mathrm{log}𝒟A_\mu ^a\mathrm{exp}\left(\frac{1}{4}F^2+gA_\mu ^aJ_a^\mu \right).$$
(4.22)
If we could evaluate exactly the gluonic part of the functional integral then we would be done. But unfortunately this integration can not be handled unless we resort to some sort of systematic approximation (this is due to the presence of cubic and quartic terms of $`A_\mu ^a`$ in the Lagrangian). One possibility to proceed is to expand the effective action in powers of the quark current $`J_\mu ^a`$ as suggested by Cahill and Roberts ,
$`\mathrm{\Gamma }[J]`$ $`=`$ $`\mathrm{\Gamma }[J=0]+g{\displaystyle \mathrm{\Gamma }_\mu ^{(1)a}J_a^\mu 𝑑x_1}+{\displaystyle \frac{g^2}{2}}{\displaystyle \mathrm{\Gamma }^2(x_1,x_2)_{\mu _1\mu _2}^{a_1a_2}J_{a_1}^{\mu _1}J_{a_2}^{\mu _2}𝑑x_1𝑑x_2}+\mathrm{}`$ (4.23)
$`+`$ $`{\displaystyle \frac{g^n}{n!}}{\displaystyle }\mathrm{\Gamma }^{(n)}(x_1,x_2,..x_n)_{\mu _1\mathrm{}\mu _n}^{a_1..a_n}J_{a_1}^{\mu _1}\mathrm{}J_{a_n}^{\mu _n}dx_1\mathrm{}dx_n,`$
where the coefficients
$$\mathrm{\Gamma }^{(n)}(x_1,x_2,\mathrm{},x_n)_{\mu _1\mathrm{}\mu _n}^{a_1\mathrm{}a_n}=\left(\frac{^n\mathrm{\Gamma }[J]}{J_{\mu _1}^{a_1}..J_{\mu _n}^{a_n}}\right)_{J=0},$$
(4.24)
are defined as one-particle irreducible gluon correlation functions in the absence of quarks,
$`\mathrm{\Gamma }^{(1)}(x_1)_{\mu _1}^{a_1}`$ $`=`$ $`A_{\mu _1}^{a_1}(x_1),`$
$`\mathrm{\Gamma }^{(2)}(x_1,x_2)_{\mu _1,\mu _2}^{a_1,a_2}`$ $`=`$ $`A_{\mu _1}^{a_1}(x_1)A_{\mu _2}^{a_2}(x_2)A_{\mu _1}^{a_1}(x_1)A_{\mu _2}^{a_2}(x_2).`$
In the above expression, the brackets denotes the functional average over the gluon field
$$\mathrm{}=\frac{𝒟A\mathrm{}\mathrm{exp}\left(\frac{1}{4}F^2\right)}{𝒟A\mathrm{exp}\left(\frac{1}{4}F^2\right)}.$$
(4.25)
The zeroth order term $`\mathrm{\Gamma }[J=0]`$ does not depend on the quark field and is therefore an irrelevant constant. The first order term $`\mathrm{\Gamma }^{(1)}`$ gives the expectation value of the gluon field and is zero in the absence of external fields. The leading non-trivial term is $`\mathrm{\Gamma }^2(x_1,x_2)_{\mu _1\mu _2}^{a_1a_2}=D_{\mu \nu }^{ab}(xy)`$ which is the exact gluon propagator and includes all gluon self-interactions and gluon-ghost interactions. Notice that the quark loops are incorporated through the quark current attached as legs to the exact gluon propagator. The main approximation in this approach is to ignore all terms $`n3`$. Note that none of the gluon correlation functions is gauge or Lorentz invariant. While each term in the expansion is separately invariant under Lorentz and “global color symmetry”, nevertheless the whole expansion in Eq. (4.23) is invariant under local gauge symmetry. We truncate Eq. (4.23) up to $`n=2`$, and with this simplification the QCD generating functional is approximated by
$$Z_{QCD}𝒟q𝒟\overline{q}\mathrm{exp}(iS_{QFD}),$$
(4.26)
where the $`S_{QFD}`$ is the induced non-local QCD action which describes the quark flavour dynamics,
$$S_{QFD}=\overline{q}(i\gamma ^\mu _\mu m^0)q+\frac{g^2}{2}\overline{q}(x)\gamma _\mu \frac{\lambda ^a}{2}q(x)D(xy)\overline{q}(y)\gamma _\mu \frac{\lambda ^a}{2}q(y).$$
(4.27)
The exact form of $`D(x)`$ is not available at the moment. Hence, the main phenomenology task is to simulate the simplest form of gluon propagator in order to reproduce the confinement and asymptotic freedom of quarks. Although there is no a priori reason to believe that the effective interaction of quarks propagating in the QCD vacuum should retain the form of a one gluon exchange interaction, it has been proved that already this simple approximation reproduces most phenomenological models such as the instanton liquid model , Nambu-Jona-Lasinio (NJL) model , various chiral bag or topological-soliton models and the Dyson-Schwinger equation approximation , etc. As an example, a useful starting point for low energy effective action is to employ a gluon propagator which is reduced to its crudest form,
$$D_{\mu \nu }^{ab}(xy)=\frac{1}{\mathrm{\Lambda }_\chi ^2}\delta (xy)g_{\mu \nu }\delta ^{ab},$$
(4.28)
where $`\mathrm{\Lambda }_\chi `$ is a constant with dimension of $`(\text{energy})^2`$. This choice leads to a local NJL type model with interaction:
$$_I=\frac{g^2}{2\mathrm{\Lambda }_\chi ^2}[\overline{q}(x)\gamma _\mu \frac{\lambda ^a}{2}q(x)][\overline{q}(x)\gamma ^\mu \frac{\lambda ^a}{2}q(x)].$$
(4.29)
This interaction describes a system of quarks interacting via a two-body-force. The local form of the interaction of course causes ultraviolet divergences, which introduces an energy scale $`\mathrm{\Lambda }_\chi `$, breaking the scale invariance of the classical Yang-Mills Lagrangian (at zero current quark mass), in an anomalous fashion. We have sketched how an effective quark theory can be approximately obtained from QCD by eliminating gluonic degrees of freedom. This is slightly different from the RG approach discussed in the first part of the thesis in which our main goal was to eliminate the high-energy degrees of freedom. Having said that, both have the same foundation, namely eliminating the irrelevant degrees of freedom and as we already discussed can be implemented at the same time.
Unfortunately, there is no economic way to derive effective theories from QCD for a given system. Therefore, one may write down the most general Lagrangian based on relevant degrees of freedom, having imposed some general constraints such as symmetry properties. This approach was introduced by Weinberg and later by Gasser and Leutwyler . For an example, an effective chiral quark theory can be presented as
$$_{\text{eff}}(x)=\underset{n}{}c_n𝒪_n(x)\left(\frac{1}{\mathrm{\Lambda }}\right)^{\text{dim}𝒪_n4},$$
(4.30)
where $`𝒪_n`$ are the local chiral-invariant operators consisting of quark fields and $`c_n`$ are dimensionless coupling constants. The theory is only valid below the scale $`\mathrm{\Lambda }`$, and the momenta of the loop integrals are cut off at $`\mathrm{\Lambda }`$. One may now truncate the above series and by first obtaining the coupling constant through experimental input, calculate other quantities. This EFT approach has been discussed in detail in Ref. .
### 4.4 Fierz-transformation and the effective quark interaction
The NJL model Lagrangian Eq. (4.29) contains the color octet flavor singlet currents of quark. Since hadrons are color singlets, it is desirable to recast the Lagrangian in such to act in color singlet channels. This can be accomplished by a Fierz transformation, see e.g., Ref. , using the relation
$$\left(\frac{\lambda ^a}{2}\right)_{ij}\left(\frac{\lambda ^a}{2}\right)_{kl}=\frac{1}{2}\left(1\frac{1}{N_c^2}\right)\delta _{il}\delta _{kj}\frac{1}{N_c}\left(\frac{\lambda ^a}{2}\right)_{il}\left(\frac{\lambda ^a}{2}\right)_{kj},$$
(4.31)
If we take the interaction in $`\overline{q}q`$ channel, then the first part of the above expression creates color singlet mesons, while the second part is color octet. However, the signs behind the color singlet and the color octet are opposite, therefore if we choose the negative sign for color singlet (in the Lagrangian level), i.e. an attractive interaction, then the interaction for color octet will be repulsive and consequently no bound state exist for this channel consistent with nature. Moreover, at large $`N_c`$ the color octet $`\overline{q}q`$ can be neglected. In the same fashion one can recast the interaction in the $`qq`$ channel. In order to make a color-singlet baryon out of three quarks, we first couple two quarks in the fundamental triplet representation $`3_c`$ which leads to either a sextet $`6_c`$ or an antitriplet $`\overline{3}_c`$. But only an antitriplet can be combined with another quark in $`3_c`$ to make a color singlet state. Notice as well that $`\overline{3}_c`$ is antisymmetric and $`6_c`$ is symmetric in color quantum numbers. We can now use the following Fierz transformation,
$$\left(\frac{\lambda ^a}{2}\right)_{ij}\left(\frac{\lambda ^a}{2}\right)_{kl}=\frac{1}{2}\left(1\frac{1}{N_c^2}\right)\delta _{il}\delta _{kj}+\frac{1}{2N_c}ϵ_{mik}ϵ_{mlj}.$$
(4.32)
It is obvious that the interaction in the $`qq`$ $`\overline{3}_c`$ channel becomes attractive, therefore diquark formation in $`\overline{3}_c`$ is in principle possible. In the same way, one can “Fierz” the flavor and Dirac quantum numbers. For the meson channel we use
$$\delta _{ij}\delta _{kl}=2\left(\frac{\lambda ^0}{2}\right)_{il}\left(\frac{\lambda ^0}{2}\right)_{kj}+2\underset{a=1}{\overset{N_f^21}{}}\left(\frac{\lambda ^a}{2}\right)_{il}\left(\frac{\lambda ^a}{2}\right)_{kj}.$$
(4.33)
Notice that in contrast to color group $`SU(N_c)`$ the flavor group is $`U(N_f)`$, hence the flavor generators includes $`\lambda ^0/2=\text{1}\text{1}/\sqrt{2N_f}`$. One may immediately read off from the above decomposition that for $`N_f=3`$, mesons occur as nonets under flavor transformation. For the diquark channel we use
$$\delta _{ij}\delta _{kl}=2\underset{m=1}{\overset{3}{}}\left(\frac{\lambda _s^m}{2}\right)_{il}\left(\frac{\lambda _s^m}{2}\right)_{kj}+2\underset{n=1}{\overset{6}{}}\left(\frac{\lambda _a^n}{2}\right)_{il}\left(\frac{\lambda _a^n}{2}\right)_{kj},$$
(4.34)
where $`\lambda _s^m`$ denotes the symmetric generators of $`U(N_f)`$, i. e. $`\lambda ^7,\lambda ^5`$, and $`\lambda ^2`$. The antisymmetric part $`\lambda _a^n`$ stands for $`\lambda ^{0,1,3,4,6,8}`$ . Finally the Dirac indices can be rearranged for meson channels by making use of
$$(\gamma _\mu )_{ij}(\gamma ^\mu )_{kl}=\delta _{il}\delta _{kj}+(i\gamma _5)_{il}(i\gamma _5)_{kj}\frac{1}{2}\left((\gamma _\mu )_{il}(\gamma ^\mu )_{kj}+(\gamma _\mu \gamma _5)_{il}(\gamma ^\mu \gamma _5)_{kj}\right).$$
(4.35)
For the diquark channel we employ
$$(\gamma _\mu )_{ij}(\gamma ^\mu )_{kl}=C_{ik}C_{lj}+(i\gamma _5C)_{ik}(Ci\gamma _5)_{lj}\frac{1}{2}\left((\gamma _\mu C)_{ik}(C\gamma ^\mu )_{lj}+(\gamma _\mu \gamma _5C)_{ik}(C\gamma ^\mu \gamma _5)_{lj}\right),$$
(4.36)
where $`C=i\gamma ^2\gamma ^0`$ is the charge conjugate matrix. Using all the above transformations, one can now rewrite the gluon exchange Lagrangian Eq. (4.29) (for $`N_f=N_c=3`$) in the following compact form
$`_I`$ $`=`$ $`{\displaystyle \frac{g^2}{2\mathrm{\Lambda }_\chi ^2}}\left[\overline{q}\gamma _\mu {\displaystyle \frac{\lambda ^a}{2}}q\right]\left[\overline{q}\gamma ^\mu {\displaystyle \frac{\lambda ^a}{2}}q\right]=_I^{\overline{q}q}+_I^{qq},`$ (4.37)
$`=`$ $`{\displaystyle \frac{g^2}{3\mathrm{\Lambda }_\chi ^2}}\{\left(\overline{q}_\alpha q\right)\left(\overline{q}^\alpha q\right)+\left(\overline{q}\mathrm{\Sigma }_\alpha q^c\right)\left(\overline{q}^c\mathrm{\Sigma }^\alpha q\right)\},`$
where the charge conjugate spinor spinor is given by $`q^c=C\overline{q}^T`$ and the other notations are defined as follows
$`_\alpha `$ $`=`$ $`\text{1}\text{1}_c\left({\displaystyle \frac{\lambda ^A}{2}}\right)_F\mathrm{\Gamma }_\alpha ,A=0,\mathrm{}8,`$
$`\mathrm{\Sigma }_\alpha `$ $`=`$ $`\left({\displaystyle \frac{iϵ^a}{\sqrt{2}}}\right)_ct_F^A\mathrm{\Gamma }_\alpha a=1,2,3;t_F^A\{\lambda _a,\lambda _s\},`$
$`\mathrm{\Gamma }_\alpha `$ $``$ $`\{\text{1}\text{1},i\gamma _5,{\displaystyle \frac{i}{\sqrt{2}}}\gamma _\mu ,{\displaystyle \frac{i}{\sqrt{2}}}\gamma _\mu \gamma _5\}.`$ (4.38)
We will study with full detail the mesons, diquarks and the baryons structures within a non-local version of the above Lagrangian in the chapter 6.
Here some remarks are in order. First of all, the Fierz transformation does not spoil chiral symmetry. The meson channel $`_I^{\overline{q}q}`$ and diquark $`_I^{qq}`$ are separately chirally invariant. The main difference between mesonic and diquark interaction stems from the Pauli principle. The diquark $`qq`$ vertices are antisymmetric in color indices, hence the Pauli principle requires a certain combinations of Dirac and flavor vertices. Notice that for $`N_c3`$, one has to replace the factor $`1/3`$ with $`(N_c1)/2N_c`$ and $`1/N_c`$ for meson and diquark channels, respectively. This indicates that in large $`N_c`$ limit, the colored diquark channels are suppressed by a $`1/N_c`$ factor and only color singlet mesons and baryons survive. This is in accordance with Witten’s conjecture that in the large-$`N_c`$ limit, QCD transforms into a theory of weakly interacting mesons and baryons emerging as solitons of this theory . We will elaborate on this conjecture in the next chapter.
## Chapter 5 QCD inspired pictures for baryons
### 5.1 Introduction
Despite all efforts to describe hadron physics in terms of its underlying QCD theory, a unified and unambiguous description of hadrons is still missing. This is due to the complexities and the non-perturbative features of low-energy sector of QCD which prohibits a straightforward computation of hadronic properties in this regime. One well-established method toward understanding the hadronic physics is the use of QCD sum rules , which aim to interpolate between the calculable high-energy behaviour of the QCD and low-energy phenomenology. Although this approach has been reasonably successful phenomenologically, there are many uncertainties induced by the choice and formulation of the phenomenological part, and the result can exhibit a significant dependence on the mass scale at which the matching is performed. More importantly, the confinement phenomenon and instanton effects are not incorporated in this formalism. Another method is to simulate QCD on a lattice of spacetime points. Some appreciable progress has been made in ab initio calculations of low-lying baryon resonances using lattice QCD simulation on a computer. This is limited by computational resources. This leads to pion masses of usually more than $`400`$ MeV (thus, the chiral point is not accessible in lattice QCD) or to simulations within the quenched approximation, where sea quark effects are neglected. Moreover, there is yet incomplete understanding of systematic errors, e.g., finite-size effects <sup>1</sup><sup>1</sup>1For understanding QCD phase structure, one should describe the important non-perturbative nature of QCD and the hot/dense QCD in a unified way. Thus far, lattice regularization have not be able to overcome the issue associated with the fact that at non-zero chemical potential the fermionic determinant is complex. On the other hand due to computational difficulties, a single lattice of equally spaced points is forced to span all distance scale which does not allow a sequence of descriptions intermediate between the constituent quark model and QCD. Full consideration of QCD phase structure is not possible. Nevertheless, the effective field theories approach based on simple models embodied essential ingredients of QCD at low-energy to give a qualitative (and even sometimes quantitative) prescription..
Therefore, any models which simplify the QCD dynamics and is capable of describing part of the hadronic physics (which might not be accessible to ab initio methods, like lattice QCD) is greatly appreciated.
Our main goal in the following is to study baryon pictures in a relativistic framework by employing only quark degrees of freedom. There are two distinct possibilities to build a model for baryons, one is in the limit of a large number of colors based on the picture of baryons as soliton, the second one is to describe baryons in term of bound state of a finite number of colors. The key ingredient of the former will be introduced in section 5.2, and the latter will be described in sections 5.3,4.
### 5.2 Baryons as chiral solitons; Skyrme model
The idea to describe baryons as solitons was first introduced by Skyrme before the advent of the quarks and gluons. However, this was not fully appreciated until the conjecture of Witten . It was ’t Hooft’s proposal that the inverse of the number of colors $`1/N_c`$, can be treated as an effective expansion parameter . Later, Witten pursued this idea and showed that QCD is reduced to an effective theory of weakly interacting mesons (with an effective four-meson vertex scaling like $`1/N_c`$), and that baryons emerge as soliton solutions of this meson field without any further reference to their quark content . Since then, this subject has been well studied. Here, we intended to explore this approach in a selective way, following Ref. closely .
At low-energy one expects that the effective meson theory is a type of non-linear $`\sigma `$ model with pions as the lighter mesons. We shall focus on the massless two flavor model. One can combine the four fields $`(𝝅,\sigma )`$ into one unitary $`2\times 2`$ matrix $`U`$ (chiral field) with only one isovector field $`\phi (x,t)`$,
$$U(x)=\mathrm{exp}(\frac{i}{f_\pi }𝝉.\phi (x)),$$
(5.1)
where the isovector $`𝝉`$ contains the pauli matrices. In terms of $`U(x)`$ the non-linear $`\sigma `$ model is defined by the Lagrangian
$$^{(2)}=\frac{c^2}{4}\mathrm{tr}\left(_\mu U^\mu U^{}\right),$$
(5.2)
where $`c`$ is a constant. The elements of chiral $`SU(2)\times SU(2)`$ transform any chiral field as
$$U(x)LU(x)R^{},$$
(5.3)
where $`R`$ and $`L`$ are arbitrary $`SU(2)`$ matrices. The Lagrangian $`^{(2)}`$ is invariant under such transformations. The vacuum configuration ($`\phi =0`$ i.e. $`U=1`$) is only invariant under the coset $`L=R`$ which reflects the spontaneous breaking of chiral symmetry. One can readily construct the Noether current associated with the symmetry transformation Eq. (5.3). The vector and axial-vector current correspond to $`R=L`$ and $`R^{}=L`$, respectively. The matrix element of the axial-vector current between the vacuum and a one pion state relates the unknown coefficient $`c`$ in Eq. (5.2) to the pion decay constant. It turns out that $`c=f_\pi =93`$ MeV .
Now, we want to find a solitonic solution for our meson theory. A fundamental requirement for a solitonic solution of the equation of motion is finiteness of energy. The static soliton configuration $`U(r)`$ represents mappings $`U:^3SU(2)`$. A necessary condition is to require the boundary condition
$$\underset{r\mathrm{}}{lim}U(r)=1.$$
(5.4)
This means that spatial infinity is mapped to one point in flavor space, i.e. $`^3`$ is compactified to the hypersphere $`S^3`$.
$$U:S^3S^3,$$
(5.5)
which induces a topological invariant, the winding numbers. The associated topological current has been given by Skyrme
$$B^\mu =\frac{1}{24\pi ^2}ϵ^{\mu \nu \rho \lambda }\mathrm{Tr}[(U^{}_\nu U)(U^{}_\rho U)(U^{}_\lambda U)],$$
(5.6)
with conservation law $`^\mu B_\mu =0`$. The integral over the zero component defines the topological charge
$$B=B_0d^3x,$$
(5.7)
which as we will show, introduces the winding number $`n`$. A static configuration of the soliton is given by the spherically symmetric hedgehog Ansatz
$$U(r)=\mathrm{exp}(i\tau .\widehat{r}\mathrm{\Theta }(r)).$$
(5.8)
In order to understand the geometrical meaning of the above Ansatz, we use this expression to compute the topological charge from Eq. (5.7).
$`B`$ $`=`$ $`{\displaystyle \frac{1}{24\pi ^2}}ϵ^{ijk}{\displaystyle \mathrm{Tr}[(U^{}_iU)(U^{}_jU)(U^{}_kU)]d^3x},`$ (5.9)
$`=`$ $`{\displaystyle \frac{1}{2\pi ^2}}{\displaystyle \frac{\mathrm{sin}^2(\mathrm{\Theta })}{r^2}_r\mathrm{\Theta }d^3x}={\displaystyle \frac{2}{\pi }}{\displaystyle _{\mathrm{\Theta }(0)}^{\mathrm{\Theta }(\mathrm{})}}\mathrm{sin}^2(\mathrm{\Theta })𝑑\mathrm{\Theta }=n,`$
where we imposed the boundary conditions $`\mathrm{\Theta }(0)=n\pi `$ and $`\mathrm{\Theta }(\mathrm{})=0`$. Solitons with different winding numbers are topologically distinct, and there thus exists no continuous deformation connecting solitons of different winding number. It has been conjectured by Skyrme that the topological current $`B^\mu `$ can be related to the baryon current and the winding number with the baryon number.
One may scale the spatial coordinate $`r`$ in $`U(r)`$ by $`\lambda r`$ i.e. $`UU(\lambda r)`$, this leads to scaling the potential part of Lagrangian Eq. (5.2): $`^{(2)}\frac{1}{\lambda }^{(2)}`$. Therefore the minimal energy is only obtained for $`\lambda \mathrm{}`$, e. i., no stable solitons can be found and the soliton collapse to zero size. However, if the Lagrangian contains a term containing products of four spatial derivatives (but only quadratic in time derivative for sake of quantization), it will scale as $`^{(2)}\frac{1}{\lambda }^{(2)}+\lambda ^{(4)}`$ which stabilizes at $`\lambda =^{(2)}/^{(4)}`$. A possible form of $`^{(4)}`$ is given by
$$^{(4)}=\frac{1}{32e^2}\left(Tr[(U^{}_\mu U),(U^{}_\nu U)][(U^{}^\mu U),(U^{}^\nu U)]\right),$$
(5.10)
where $`e`$ is an extra parameter which determines the size of the particle.
A natural question is how baryons get their half integer spin and isospin within soliton picture since the pion field possesses spin zero and isospin one. An immediate answer is quantization. We refrain to go through details here and concentrate only on main points. The time dependent soliton solution should be obtained as a first step toward quantization. However, such solutions are hardly available and one needs to resort to an approximation. A reasonable Ansatz for such a time-dependent solution is given by
$$U(r,t)=A(t)U_0(r)A^{}(t),$$
(5.11)
where $`A(t)SU(2)`$ is referred to as the collective rotation and contains collective coordinates (it can be parametrized in terms of Euler angles). This Ansatz does not change the potential energy of the hedgehog. The time derivative in the Lagrangian, produces terms which represent the rotational energy of a rotating skyrmions. Having used hedgehog properties and a suitable definition of angular velocities as canonical variables, one can show that the absolute values of spin and isospin are equal, $`T^2=J^2`$, and the energy eigenvalues of the system forms a rotational spectrum $`E_J=\frac{1}{2\theta }J(J+1)+M`$, where $`M`$ is the static energy of the soliton and $`\theta `$ denotes the moment of inertia. One can immediately observe that the quantum numbers of the low energy baryons (e.g. the nucleon with $`J=T=1/2`$ and delta $`J=T=3/2`$ ) are consistent with this picture. One of the obvious shortcomings of this presentation is that it does not give any reason in favor or against half-integer or integer values of $`J`$ for the quantized skyrmions. We will argue later that the fermionic character of the quantized baryon can be revealed in a model with $`SU(3)\times SU(3)`$ symmetry by inclusion of the Wess-Zumino-Witten term.
An extension of the theory to three flavors with chiral $`SU(3)\times SU(3)`$ symmetry is essential if one wants to consider the baryon octet and decuplet, see for details Ref. . We assume the generic form of the underlying Lagrangian remains unchanged $`=^{(2)}+^{(4)}`$. However, the chiral field $`U`$ needs to be increased to a $`SU(3)`$ field with the mesons fields $`\varphi ^a,a=1,\mathrm{},8`$, which in addition to the pions, contains the kaons and the octet component of the $`\eta `$
$$U=\mathrm{exp}\left(i\underset{a=1}{\overset{8}{}}\frac{\varphi ^a}{f_a}\lambda ^a\right),$$
(5.12)
where $`\lambda ^a`$ denotes the Gell-Mann matrices. The decay constants $`f_a`$ are defined through the gradient expansion of the axial-vector current, analogous to the case with only two flavors. In a similar fashion to the two flavor case, one can obtain solitonic solutions of the theory and quantization can be done by introducing the collective rotations.
In order to link the effective meson theory to QCD, one should firstly consider if all symmetries of the Lagrangian $``$ are in accordance with QCD. Witten observed that the Lagrangian Eq. (5.2) possesses an extra discrete symmetry that is not a symmetry of QCD. Under parity transformation $`P`$, the pseudoscalar meson fields as described by QCD should obey $`P𝝅(x,t)=𝝅(x,t)`$. In our meson theory this means,
$$P:𝒙𝒙,tt,UU^{}.$$
(5.13)
But it is obvious that the Lagrangian Eq. (5.2) is invariant under $`𝒙𝒙`$ and $`UU^{}`$, separately. In order to break this unwanted symmetry, one needs to add some extra term to the meson action. Unfortunately, there exists no local term in four spacetime dimension which can be added to the Lagrangian, so as to get rid of this separate symmetry. However, it is very easy to look for such extra term by considering the equation of motion. Witten suggested that the simplest term (with lowest possible number of derivatives) which needs to be added are as follows
$$\frac{f_\pi ^2}{8}^\mu C_\mu +\lambda ϵ^{\mu \nu \rho \sigma }C_\mu C_\nu C_\rho C_\sigma =0,$$
(5.14)
where $`C_\mu =U^{}_\mu U`$. The new term is odd under $`𝒙𝒙`$ while the first and similar higher-order terms are even. However, the new term is even under transformation $`UU^{}`$ while the first term is odd. Therefore, Eq. (5.14) is invariant only under the combined action of $`P`$ Eq. (5.13). The problem now is that the four-dimensional action corresponding to the new extra term can not be written in a chirally invariant form. However, this action can be rewritten in such a form in five dimensions. Therefore, we extend the coordinate space to five-dimensional manifold $`M_5`$ in such a way that our conventional four-dimensional spacetime $`M_4`$ is the boundary of a five-dimensional volume i.e. $`M_5=M_4`$. Therefore, one can write
$$\mathrm{\Gamma }=\lambda _{M_5}ϵ^{ijklm}\mathrm{Tr}(C_iC_jC_kC_lC_m)d^5x.$$
(5.15)
This action indeed leads to the equations of motion Eq. (5.14) where written in four-dimensional spacetime by means of Stokes’theorem. Witten in his remarkable paper , argued that the coefficient $`\lambda `$ in above equation must be integer multiple of a normalization factor,
$$\lambda =n\frac{i}{240\pi ^2}.$$
(5.16)
This is comprehensible, since the path-integral formulation requires the action to be changed by a multiple of $`2\pi `$ when going from $`M_5`$ to its complement, which has the identical boundary with opposite orientation. This is indeed in the same spirit of Dirac’s quantization of a magnetic monopole. The physical meaning of the integer $`n`$ is fixed through a connection to the Wess-Zumino action <sup>2</sup><sup>2</sup>2 The Wess-Zumino action was introduced to account for anomalies which occur through the renormalization of fermion loops in quantum field theories where pseudoscalar mesons are coupled to fermions . Witten included the photon fields in a gauge invariant way which generates a vertex for the decay $`\pi ^0\gamma \gamma `$ with
$$\frac{n}{96\pi ^2f_\pi }\pi ^0ϵ_{\mu \nu \rho \sigma }F^{\mu \nu }F^{\rho \sigma },$$
(5.17)
where $`F^{\mu \nu }`$ is the field strength tensor of the photon. One can immediately compare this result with the well-known triangle anomaly in QCD and find that $`n=N_C`$. In this way, the effects of anomalies in QCD are correctly reproduced by the action Eq. (5.15) which is called Wess-Zumino-Witten (WZW) term. Notice that if one considers an adiabatic $`2\pi `$ rotation of the soliton, the WZW term produces a contribution $`N_c\pi `$ to the action while other terms do not contribute. Therefore, the soliton acquires a phase $`(1)^{N_c}`$ for such a rotation as required for fermions with $`N_c`$ odd or bosons with $`N_c`$ even. It is interesting that for two flavours, this conclusion can not be made since the WZW term is zero. Therefore, for three flavors, the WZW term provides some hint about the statistical difference between baryons and mesons (fermions or bosons). The fact that the Skyrme Lagrangian needs to be augmented by the WZW term indicates that the underlying physics of the Skyrme is a model of quarks and gluons which possess QCD properties.
As we discused in the last chapter, one of the simplest but viable quark model at the present is the NJL model. The solitonic solutions of NJL models have been extensively investigated . Notice that, in contrast to Skyrme-like models with infinite energy barriers separating sectors with different winding numbers, chiral quark models, such as the NJL models have finite energy barriers separating the different sectors, and they give rise to so-called non-topological solitons.
Despite all appealing features of soliton models, there are some shortcomings: It is well-known that solitonic models are not very accurate, e.g., in leading order of $`N_c`$, the quasi-classical soliton configuration with quantized collective variables can produce baryonic observables with errors of about $`20\%30\%`$, and corrections due to mesonic fluctuations seem to be very important. Another point of weakness is that exotic states are very controversial in these models<sup>3</sup><sup>3</sup>3As Cohen argued the main reason is that the rigid-rotor quantization is not valid for such states. In other words, the assumption that the collective motion is orthogonal to vibrational motion is only true for non-exotic motion, but the Wess-Zumino term induces mixing at leading order between collective and vibrational motion with exotic quantum numbers. Recent discovery of Pentaquark $`\theta ^+`$ which was already predicted based on soliton model have brought a lot of activity on this subject..
### 5.3 Bag models
In 1974 MIT group developed a new picture of hadrons based on the simple assumption that the physical vacuum prohibits free quarks and gluons, but instead creates bubbles of hadronic size in which quarks and gluons may propagate ordinarily. This idea has been employed in various models: a hybrid bag model where the nucleon consists of a quark bag surrounded by a meson cloud, the little bag model where in contrast to the hybrid bag, pions are not allowed to propagate inside the bag, a cloudy bag model where the mesons are constrained to the chiral circle and are allowed inside bag, and finally the chiral bag model where the constrained mesons outside bag are described by the Skyrme Lagrangian. In all these models, there is one extra parameter in the model, the bag radius, which provides some hint about the quark and pion distributions. A review of various bag models can be found in Ref. .
Chiral bag models seem to interpolate between two different aspects of QCD, the long range-low energy (non-perturbative) and the small distance (perturbative) behaviours. This idea was proposed by Rho et al . The very small volume $`V`$ represents the perturbative domain of QCD containing quarks and gluons only, as opposed to its complementary piece containing the confined phase of QCD with color-singlet objects such as mesons,
$``$ $`=`$ $`_{\text{bag}}\theta _V+_{\text{meson}}(1\theta _V)+_{\text{boundary}}\delta _V,`$
$`_{\text{bag}}`$ $`=`$ $`i\overline{q}\gamma _\mu ^\mu q,`$
$`_{\text{boundary}}`$ $`=`$ $`\overline{q}\mathrm{exp}(\gamma _5\widehat{x}.\widehat{\tau }\mathrm{\Theta }(r))q,`$
where inside the bag we have massless quark fields $`q`$ and outside we have chiral meson fields $`U`$ which obeys the Skyrme Lagrangian (for simplicity the hedgehog configurations $`U=\mathrm{exp}(i\tau .\widehat{x}\mathrm{\Theta }(r))`$ with spherical bags of radius $`R`$ are assumed ). The boundary term cause the full Lagrangian to be invariant under a combined chiral symmetry;
$$q^{}\mathrm{exp}(i\gamma _5\alpha .\tau )q,U^{}\mathrm{exp}(i\alpha .\tau )U\mathrm{exp}(i\alpha .\tau ).$$
(5.18)
We already identified the topological charge carried by the meson field as baryon number. On the other hand, the quarks inside the bag each carry one third amount of baryon charge, therefore, it seems puzzling that the total baryon number is not integral. Goldstone and Jaffe proved that a conjecture of Rho et al that baryon number in the hybrid model remains one is indeed right. The crucial observation they made is that the charge of the vacuum baryon number, inside the bag is changed due to boundary effects. This shows that the baryon number remains one regardless of the profile of the Skyrme fields and the size of bag. It was later proved that the total energy from bag, chiral fields and vacuum (Casimir energy) is insensitive to the bag size as well . Therefore, it is tantalising to assume that observables should not depend on the details of the bag (e.g., its size) also. This statement is called “The Cheshire Cat Principle (CCP)” . Topological quantities, like the baryon number satisfy an exact CCP in (3+1) dimensions while non-topological observables such as masses, static properties and also non-static properties satisfy it approximately well .
The main problem with various bag models is that they are not fully covariant and possible large modification of observables due to quantum fluctuations make model prediction less reliable.
### 5.4 Diquark-quark picture; Relativistic Faddeev approach
In the previous sections, we reviewed the basic foundations of two well established picture of baryons, motivated from QCD properties. Two main shortcomings within these approaches, namely the uncertainty of computed quantities and lack of covariance, make these not viable for phenomenological usage in the intermediate-energy regime where a fully covariant formulation is required. A new generation of continuous beam facilities such as CEBAF at TJNAF, ELSA, COSY, MAMI, etc, which are designed to explore the intermediate energy between non-perturbative and perturbative regime of QCD, needs accurate covariant formulation to describe the forthcoming data. The diquark-quark picture of baryons based on a relativistic Faddeev approach, is a framework for such a fully covariant approach. This approach has been extensively employed during the last decade, and is phenomenologically very successful . In this picture, a bound state of a baryon <sup>4</sup><sup>4</sup>4There is another very old-fashioned approach to obtain baryon bound state which was invented in the early sixties, the constituent quark models or quark potential models. In these models one starts with simple potential e.g, the hyperfine type interaction and employs 3-particle Schrödinger (or Dirac ) equation to obtain the spectrum. This approach is not covariant and does not incorporate the minimal field theoretical aspect of QCD, like quarks degree of freedom. We refrain from discussing this approach here. is obtained as a pole of the three-quark correlation by summing over infinitely many interaction graphs. This process is very similar to obtaining a two-body bound state which leads to the Bethe-Salpeter equation.
In the context of local quantum field theory, the few-body problem seems to be ill-defined, since any restriction on degrees of freedom (e.g., particle numbers) may spoil the Lorentz invariance (e.g., on the equal-time quantization, the boosts generators involve interactions and change the number of particles, therefore limiting the number of particles is against the Lorentz invariance). Nevertheless, we know from a phenomenological point of view that a fixed number of constituent quarks might be enough to describe baryons. Therefore, we introduce the notion of baryon wave functions as matrix elements of three quark operators between the physical vacuum $`|\mathrm{\Omega }`$ and a (nucleon) bound state $`|P_N`$; $`\mathrm{\Psi }\mathrm{\Omega }|T(qqq)|P_N`$. Having said that, QCD vacuum is non-perturbative and indeed possesses non-trivial condensates, and thus the wave function contains sea quark and gluonic parts. However, it may be reasonable to ignore all other operator matrix elements which involve an arbitrary number of quarks and gluons as irrelevant in comparison to the dominant amplitude $`\mathrm{\Psi }`$. One can now solve the three-body problem by means of the Green’s function formulation of quantum field theory. As we will show, the non-perturbative feature of the vacuum can then be effectively incorporated into the formalism in a systematic fashion.
In the following we introduce an approximation scheme based on the relativistic Faddeev approach to simplify the three-quark problem in form of a diquark and quark interacting via quark exchange. For simplicity, we use a formal presentation; all color, flavor and Dirac indices are implicit in the single particle labels. We use a Euclidean metric in momentum space. We denote a dressed single quark propagator by $`S_i`$, with
$$(2\pi )^4\delta ^4(k_ip_i)S_i(k_i;p_i)=d^4x_{k_i}d^4x_{p_i}e^{i(k_i.x_{k_i}p_i.y_{p_i})}0|Tq_i(x_i)q_j(x_j)|0.$$
(5.19)
In our definition of Green’s functions and bound state matrix elements, we always take out one $`\delta `$-function corresponding to conservation of energy-momentum. We define the full quark six-point function (or the three-quark correlation function) as
$`(2\pi )^4\delta ^4\left({\displaystyle \underset{i=1}{\overset{3}{}}}(k_ip_i)\right)G(k_i;p_i)=`$
$`{\displaystyle }\mathrm{\Pi }_{i=1}^3d^4x_{k_i}d^4y_{p_i}\mathrm{exp}\left(i{\displaystyle \underset{i=1}{\overset{3}{}}}(k_i.x_{k_i}p_i.y_{p_i})\right)0|T\mathrm{\Pi }_{i=1}^3q(x_{k_i})\overline{q}(y_{p_i})|0.`$ (5.20)
The three-quark correlation function satisfies the Dyson equation,
$$G=G_0+G_0KG,$$
(5.21)
where $`G_0`$ denotes the disconnected three dressed quark propagator and $`K`$ stands for the three-quark scattering kernel containing all two and three-body irreducible diagrams. The symbol ”$``$” denotes summation and integration over all internal and dummy indices. A bound state of mass $`M`$ with wave function $`\mathrm{\Psi }`$ emerges as a pole of the three-quark correlation function,
$$G(k_i,p_i)\frac{\mathrm{\Psi }(k_1,k_2,k_3)\mathrm{\Psi }(p_1,p_2,p_3)}{P^2+M^2},$$
(5.22)
where $`P=p_1+p_2+p_3`$ and we defined $`\mathrm{\Psi }`$ as a three-body wave function which represents the transition matrix element between the vacuum and a bound state with mass $`M`$,
$$(2\pi )^4\delta ^4\left(\underset{i=1}{\overset{3}{}}(p_iP)\right)\mathrm{\Psi }(p_1,p_2,p_3)=\mathrm{\Pi }_{i=1}^3d^4x_i\mathrm{exp}(i\underset{i=1}{\overset{3}{}}p_i.x_i)0|\mathrm{\Pi }_{i=1}^3q_i(x_i)|P.$$
(5.23)
We now substitute the bound state parametrization Eq. (5.22) into Eq. (5.21) and compare the residues. This leads to the homogeneous bound state equation,
$$\mathrm{\Psi }=G_0K\mathrm{\Psi },G^1\mathrm{\Psi }=0.$$
(5.24)
Solving this equation exactly is almost impossible since neither the detail of all two- and three-particle irreducible graphs appearing in $`K`$, nor the full dressed quark propagator contained in $`G_0`$ are known. It is well known that the problem becomes more tractable if one employs the Faddeev approximation, by discarding all three-body irreducible graphs from the interaction kernel $`K`$. In this way one can write the kernel as a sum of three two-body interaction kernels,
$$K=K_1+K_2+K_3,$$
(5.25)
where $`K_i`$ with $`i=1,2,3`$ refers to the interactions of quark pairs $`(jk)`$ with a spectator quark $`(i)`$. The two-quark propagators $`g_i`$ satisfy their own Dyson equation with kernel $`K_i`$,
$$g_i=G_0+G_0K_ig_i,$$
(5.26)
where $`g_i`$ and $`K_i`$ are defined in three-body space, since $`G_0`$ is defined in three-body space. Hence, the former contains a factor $`S_i`$ (the propagator of the spectator quark), and the latter contains a factor $`S_i^1`$ i.e., $`K_i=k_{qq}S_i^1`$. One may associate a disconnected scattering amplitude to every spectator quark $`(i)`$, i.e. $`T_i=t_iS_i^1`$, where $`t_i`$ describes the scattering between the quarks $`(j)`$ and $`(k)`$ in two-quark subspace. The matrix $`T_i`$ is obtained by amputating all incoming and outgoing quark legs from the connected part of $`g_i`$,
$$g_i=G_0+G_0T_iG_0.$$
(5.27)
By combining the two previous equations, the Dyson equation for $`T_i`$ can be found as
$$T_i=K_i+K_iG_0T_i.$$
(5.28)
Now we define Faddeev components $`\mathrm{\Psi }_i`$ via Eqs. (5.24,5.25)
$$\mathrm{\Psi }_i=G_0K_i\mathrm{\Psi },$$
(5.29)
where we have $`\mathrm{\Psi }=\mathrm{\Psi }_i`$. We rewrite the Eqs. (5.27) as $`g_iG_0^1=1+G_0T_i`$ and plug this expression into Eq. (5.29) and make use of Eq. (5.26). We then find the well-known Faddeev bound state equations,
$$\mathrm{\Psi }_i=G_0T_i(\psi _j+\mathrm{\Psi }_k)=S_jS_kt_i(\mathrm{\Psi }_j+\mathrm{\Psi }_k).$$
(5.30)
We have shown that the complicated three-quark problem can be systematically simplified by employing the full two-quark correlation function $`t_i`$, instead of the kernel $`K`$. In this way the eight dimensional Eq. (5.24) is reduced to a set of coupled four-dimensional equations (5.30). A further simplification of this equation can be achieved by approximating the full two-quark correlation $`t_i`$ as a sum of separable correlations,
$$t_i(k_1,k_2;p_1,p_2)=\underset{a}{}\chi _i^a(k_1,k_2)D_i^a(k_1+k_2)\overline{\chi }_i^a(p_1,p_2),$$
(5.31)
where the function $`\chi _i^a`$ is the vertex function of two-quark with a diquark and $`\overline{\chi }_j^a`$ denotes its complex conjugate. The index “$`a`$” denotes the different channels of the diquarks, and $`D_i^a`$ is the corresponding diquark propagator. Note that separability implies that $`t_i`$ does not depend on any of the scalar products $`k_i.p_j`$. The diquarks parametrization to some extend contains the unknown non-perturbative physics within the baryon structure. A natural Ansatz for the Faddeev component $`\mathrm{\Psi }_i`$ is given by
$$\mathrm{\Psi }_i^{\alpha \beta \gamma }(p_i,p_j,p_k)=\underset{a}{}S_i^{\alpha \alpha ^{}}S_j^{\beta \beta ^{}}S_k^{\gamma \gamma ^{}}\chi _{i,\beta ^{}\gamma ^{}}^a(p_i,p_k)D_i^a(p_j+p_k)\mathrm{\Phi }_{i\alpha ^{}}^a(p_i,p_j+p_k),$$
(5.32)
where summation over repeated indices is assumed. The Greek multi-indices $`\alpha ,\beta ,..`$ denote color, flavor and Dirac indices, and $`i,j,k`$ indicate the type of quarks. The quark $`(i)`$ and diquark labels $`(jk)`$ are fixed. The quantity $`\mathrm{\Phi }`$ is the baryon-quark-diquark vertex function and depends only on the relative momentum between the momentum of the spectator quark, $`p_i`$ and the momentum of the diquark quasi-particle, $`P_j+P_k`$. In a relativistic formulation of a few-body system there is no unique definition of the momentum transfered between individual particles. Therefore, we introduce a new parameter $`\eta `$ which parametrises this ambiguity and shows the distribution of the total momentum within the system (diquark and quark). We define a relative momentum between the quark $`(i)`$ and the diquark consisting of the quarks $`(jk)`$ by
$$p=(1\eta )p_i\eta (p_j+p_k)=p_i\eta P,$$
(5.33)
where $`P=p_1+p_2+p_3`$. The physical properties of the baryons will indeed not depend on $`\eta `$. One can employ the definition of (5.32) to rewrite the Faddeev equation (5.30) in terms of a vertex function $`\mathrm{\Phi }`$,
$$\mathrm{\Phi }_{i,\alpha }^a=\underset{b}{}[\overline{\chi }_{i,\beta \gamma }^aS_k^{\gamma \gamma ^{}}\chi _{j,\gamma ^{}\alpha }^b][D^bS_j^{\beta \beta ^{}}\mathrm{\Phi }_{j\beta ^{}}^b]+(jk).$$
(5.34)
In the above derivation we assumed that the quark-diquark vertex function is antisymmetric under the exchange of the quark labels ($`\chi _{i,\beta \gamma }^a=\chi _{i,\gamma \beta }^a`$) as a consequence of the Pauli exclusion principle. Eq. (5.34) resembles a Bethe-Salpeter equation; the first bracket can be conceived as a quark-diquark interaction kernel since it contains the exchange of a single quark between a quark and a diquark, the second term couples the interaction kernel to baryon-quark-diquark vertex via diquarks and quarks propagator.
For identical quarks the antisymmetrization of the vertex functions is essential, we can now sum over the type of particle and drop the index like $`(i)`$. Therefore, equation (5.34) can be rewritten as,
$`\mathrm{\Phi }_\alpha ^a(p,P)`$ $`=`$ $`{\displaystyle \underset{b}{}}{\displaystyle \frac{d^4k}{(2\pi )^4}K_{\alpha \gamma }(k,p,P)G_{0,\gamma \beta }(k,P)\mathrm{\Phi }_\beta ^b(k,P)},`$
$`G_{0,\gamma \beta }(k,P)`$ $`=`$ $`2S_{\gamma \beta }(k_q)D^b(k_d),`$
$`K_{\alpha \gamma }(k,p,P)`$ $`=`$ $`\overline{\chi }_{\beta ^{}\gamma }^a(k,q)S_{\beta ^{}\gamma ^{}}\chi _{\alpha \gamma ^{}}^b(q,p_q),`$ (5.35)
where we define the spectator quark and diquark momenta as $`k_q(p_q)=\eta P+k(p)`$, $`k_d(p_d)=(1\eta )Pk(p)`$, respectively. Momentum conservation fixes the momentum of exchanged quark $`q=kp+(12\eta )P`$, see fig. 5.1. The factor of two in the above presentation originates from the summation over the type of quarks and can be absorbed into the definition of diquark propagator.
In conclusion, we managed to recast the three-body Faddeev equation in the form of an effective two-body Bethe-Salpeter equation between the diquarks and the quarks, having summed over the ladder-type quark exchange diagrams between the quarks and the diquarks. The only assumptions we have made are; 1) we neglect all three-particle irreducible contributions 2) we model the connected two-body correlation as a sum of separable terms which are identified by diquark channels, see Eq. (5.31). In order to obtain an equation for physical baryons, we have to project Eq. (5.34) onto the baryon quantum numbers. This will be carried out for a model in the next section.
## Chapter 6 Baryons structure in a non-local quark confining NJL model
### 6.1 Introduction
The NJL model is a successful (low-energy) phenomenological field theory inspired by QCD . The model is constructed to obey the basic symmetries of QCD in the quark sector, but unlike the case of low-energy QCD, quarks are not confined. The basic ingredient of the model, apart from the standard bilinear Lagrangian in the quark fields, is a zero-range interaction containing four fermion fields. This means that the model is not renormalizable. If we make the standard one over the number of colours ($`1/N_C`$) loop expansion, already at one-loop level an ultraviolet subtraction (usually implemented by a cut-off) supplemented with a regularisation method is required. The value of the cut-off can be related to the scale of physical processes not included in the model, and thus determines its range of validity. Consequently, processes involving a large momentum transfer, such as anomalous decay, can not be described by the model. At higher orders in the loop expansion, which are necessary for calculating mesonic (baryonic) fluctuations , one needs extra cut-off parameters. It is hard to determine these parameters from independent physics, and thus to build a viable phenomenology. A similar problem appears in the diquark-quark picture of baryons where an additional cut-off parameter is required to regularise the diquark-quark loops . It has been shown that a renormalizable extension of the NJL model (at least to one-loop level) can be constructed by matching the NJL-contact interaction at low energy with a one-gluon exchange type interaction above the Landau pole. Another way to cope with the non-renormalisability of the model is to embed this model into a renormalizable theory such as the linear $`\sigma `$-model and apply a renormalisation group approach . However, all these approaches add extra complexity which make them difficult to apply for anything but very simple problems.
Another drawback of the model is the absence of confinement, which makes it questionable for the description of few-quark states and for quark matter. If energetically allowed, the mesons of the model can decay into free quark-antiquark pairs, and the presence unphysical channel is another limitation of the applicability of NJL model. At the same time, it is also known that the NJL model exhibits a zero-temperature phase transition at unrealistically low baryon density . This problem is caused by the formation of unphysical coloured diquark states. These may be explicitly excluded at zero density by a projection onto the physical channels, but dominate the behaviour at finite density. The model is not able to describe nuclear matter, even in the low-density regime .
We do not know how to implement colour confinement in the model and, anyway, the exact confining mechanism of QCD is still unknown. In the context of an effective quark theory, a slightly different mechanism of “quark confinement” can be described by a quark propagator which vanishes due to infra-red singularities or when it does not produce any poles corresponding to asymptotic quark states . Another realisation of quark confinement can be found in Ref. . It has been shown that a non-local covariant extension of the NJL model inspired by the instanton liquid model can lead to quark confinement for acceptable values of the parameters . Here the quark propagator has no real pole, and consequently quarks do not appear as asymptotic states. The quark propagator has infinitely many pairs of complex poles corresponding to quarks which have a finite lifetime. This phenomenon was also noticed in Schwinger-Dyson equation studies in QED and QCD .
We can simply accept the appearance of these poles as an artifact of the naive truncation scheme involved. However, it has been recently suggested that it might be a genuine feature of the full theory, and be connected with the underlying confinement mechanism . For example, it has been shown by Maris that if we remove the confining potential in QED in 2+1D the mass singularities are located almost on the time axis, and if there is a confining potential, the mass-like singularities move from the time axis to complex momenta . In this chapter, we study this kind of confinement from another viewpoint. We show that when we have quark confinement in the non-local NJL model, the baryons become more compact, compared to a situation where we have only real poles for quark propagator.
There are several other advantages of the non-local version of the model over the local NJL model: the dynamical quark mass is momentum-dependent and also found in lattice simulations of QCD . More importantly, the non-locality regularises the model preserving anomalies , and the regulator makes the theory finite to all orders in the $`1/N_c`$ expansion, and leads to small next-to-leading order corrections . As a result, the non-local version of the NJL model may have more predictive power.
The instanton-liquid model is only one way to motivate such a model . Many effective field theories constructed by the Wilsonian renormalisation group approach lead to a non-locality, at least as irrelevant terms in the renormalisation group sense. Non-locality also emerges naturally in the Schwinger-Dyson resummation . Considerable work has been done on these nonlocal NJL models including applications to the mesonic sector , phase transitions at finite temperature and densities , and the study of chiral solitons .
In this chapter we present our first results from a calculation of the relativistic Faddeev equation for a non-local NJL model , based on the covariant diquark-quark picture of baryons . Such an approach has been extensively employed to study baryons in the local NJL model, see, e.g., Refs. . We include both scalar and the axial-vector diquark correlations. We do not assume a special form for the interaction Lagrangian, but we rather treat the coupling in the diquark channels as free parameters and consider the range of coupling strengths which lead to a reasonable description of the nucleon. We construct diquark and nucleon solutions and study the possible implications of the quark confinement in the solutions. Due to the separability of the non-local interaction, the Faddeev equations can be reduced to a set of effective Bethe-Salpeter equations. This makes it possible to adopt the numerical method developed for such problems in Refs. .
### 6.2 A non-local NJL model
We consider a non-local NJL model Lagrangian with $`SU(2)_f\times SU(3)_c`$ symmetry.
$$=\overline{\psi }(i/m_c)\psi +_I,$$
(6.1)
where $`m_c`$ is the current quark mass of the $`u`$ and $`d`$ quarks and $`_I`$ is a chirally invariant non-local interaction Lagrangian. Here we restrict the interaction terms to four-quark interaction vertices.
There exist several versions of such non-local NJL models. Regardless of what version is chosen, by a Fierz transformation one can rewrite the interaction in either the $`q\overline{q}`$ or $`qq`$ channels, and we therefore use the interaction strengths in those channels as independent parameters. For simplicity we truncate the mesonic channels to the scalar ($`0^+,T=0`$) and pseudoscalar ($`0^{},T=1`$) ones. The $`qq`$ interaction is truncated to the scalar ($`0^+,T=0`$) and axial vector ($`1^+,T=1`$) colour $`\overline{3}`$ $`qq`$ channels (the colour $`6`$ channels do not contribute to the colourless three-quark state considered here). We parametrise the relevant part of interaction Lagrangian as
$`_I`$ $`=`$ $`{\displaystyle \frac{1}{2}}g_\pi j_\alpha (x)j_\alpha (x)+g_s\overline{J}_s(x)J_s(x)+g_a\overline{J}_a(x)J_a(x),`$
$`j_\alpha (x)`$ $`=`$ $`{\displaystyle d^4x_1d^4x_3f(xx_3)f(x_1x)\overline{\psi }(x_1)\mathrm{\Gamma }_\alpha \psi (x_3)},`$
$`\overline{J}_s(x)`$ $`=`$ $`{\displaystyle d^4x_1d^4x_3f(xx_3)f(x_1x)\overline{\psi }(x_1)\left[\gamma _5C\tau _2\beta ^A\right]\overline{\psi }^T(x_3)},`$
$`J_s(x)`$ $`=`$ $`{\displaystyle d^4x_2d^4x_4f(xx_4)f(x_2x)\psi ^T(x_2)\left[C^1\gamma _5\tau _2\beta ^A\right]\psi (x_4)}.`$
$`\overline{J}_a(x)`$ $`=`$ $`{\displaystyle d^4x_1d^4x_3f(xx_3)f(x_1x)\overline{\psi }(x_1)\left[\gamma _\mu C\tau _i\tau _2\beta ^A\right]\overline{\psi }^T(x_3)},`$
$`J_a(x)`$ $`=`$ $`{\displaystyle d^4x_2d^4x_4f(xx_4)f(x_2x)\psi ^T(x_2)\left[C^1\gamma ^\mu \tau _2\tau _i\beta ^A\right]\psi (x_4)},`$ (6.2)
where $`\mathrm{\Gamma }_\alpha =(1,i\gamma _5\tau )`$. The matrices $`\beta ^A=\sqrt{3/2}\lambda ^A(A=2,5,7)`$ project onto the colour $`\overline{3}`$ channel with normalisation $`\mathrm{tr}(\beta ^A\beta ^A^{})=3\delta ^{AA^{}}`$ and the $`\tau _i`$’s are flavour $`SU(2)`$ matrices with $`\mathrm{tr}(\tau _i\tau _j)=2\delta _{ij}`$. The object $`C=i\gamma _2\gamma _5`$ is the charge conjugation matrix.
Since we do not restrict ourselves to specific choice of interaction, we shall treat the couplings $`g_s`$, $`g_a`$ and $`g_\pi `$ as independent parameters. We assume $`g_{\pi ,s,a}>0`$, which leads to attraction in the given channels (and repulsion in the $`q\overline{q}`$ colour octet and $`qq`$ colour antisextet channels). The coupling parameter $`g_\pi `$ is responsible for the pions and their isoscalar partner $`\sigma `$. The coupling strengths $`g_s`$ and $`g_a`$ specify the behaviour in the scalar and axial-vector diquark channel, respectively.
For simplicity, we assume the form factor $`f(xx_i)`$ to be local in momentum space, since it leads to a separable interaction
$$f(xx_i)=\frac{d^4p}{(2\pi )^4}e^{i(xx_i)p}f(p).$$
(6.3)
It is exactly this separability that is also present in the instanton liquid model . The dressed quark propagator $`S(k)`$ is now constructed by means of a Schwinger-Dyson equation (SDE) in the rainbow-ladder approximation. Thus the dynamical constituent quark mass, arising from spontaneously broken chiral symmetry, is obtained in Hartree approximation<sup>1</sup><sup>1</sup>1Notice that as we demonstrated in section 4.4, the exchange diagrams (for four-fermion interactions) can always be cast in form of direct diagram via a Fierz transformation. This means, the Hartree-Fock approximation is equivalent to the Hartree approximation with properly redefining coupling constants. Therefore, Hartree approximation is as good as Hartree-Fock one as long as the interaction terms in Lagrangian are not fixed by some underlying theory. (the symbol $`\mathrm{Tr}`$ denotes a trace over flavour, colour and Dirac indices and $`\mathrm{tr}_D`$ denotes a trace over Dirac indices only)
$$M(p)=m_c+ig_\pi f^2(p)\frac{d^4k}{(2\pi )^4}\mathrm{Tr}[S(k)]f^2(k),$$
(6.4)
where
$$S^1(k)=k/M(k),$$
(6.5)
one can simplify this equation by writing $`M(p)`$ in the form
$$M(p)=m_c+(M(0)m_c)f^2(p).$$
(6.6)
The non-linear equation can then be solved iteratively for $`M(0)`$.
Following Ref. , we choose the form factor to be Gaussian in Euclidean space, $`f(p_E)=\mathrm{exp}(p_E^2/\mathrm{\Lambda }^2)`$, where $`\mathrm{\Lambda }`$ is a cutoff of the theory. If one assumes that $`\mathrm{\Lambda }`$ is related to the average inverse size of instantons $`1/\overline{\rho }`$, then its value parametrises aspects of the non-perturbative properties of the QCD vacuum . This choice respects Poincaré invariance and for certain values of the parameters it leads to quark, but not colour, confinement. For values of $`M(0)`$ satisfying
$$\frac{M(0)m_c}{\sqrt{m_c^2+\mathrm{\Lambda }^2}m_c}>\frac{1}{2}\mathrm{exp}\left(\frac{(\sqrt{m_c^2+\mathrm{\Lambda }^2}+m_c)^2}{2\mathrm{\Lambda }^2}\right)$$
(6.7)
the dressed quark propagator has no poles at real $`p^2`$ in Minkowski space ($`p^2+M^2(p^2)0`$). The propagator has infinitely many pairs of complex poles, both for confining and non-confining parameter sets. This is a feature of these models and due care should be taken in handling such poles, which can not be associated with asymptotic states if the theory is to satisfy unitarity. One should note that the positions of these poles depend on the details of the chosen form factor and the cut-off, hence one may regard them as a pathology of the regularisation scheme. Since the choice of the cut-off is closely related to the truncation of the mesonic channels, (for example, if one allows mixing of channels, the cut-off and the positions of poles will change. In Fig. 6.5 we have shown the positions of the first poles of the quark propagator for various cutoff. We have examined that in the presence of $`\pi a_1`$ mixing, these positions will change, but it follows very similar trend). Even though the confinement in this model has no direct connection to the special properties of the pion, there is an indirect connection through the determination of the parameters from the pionic properties.
From the gap equation Eq. (6.4) one can show that dynamical symmetry breaking occurs for $`1/g_\pi 1/g_\pi ^{cr}<0`$, where $`1/g_\pi ^{cr}=\frac{N_cN_f\mathrm{\Lambda }^2}{12\pi ^2}`$, $`N_c`$ and $`N_f`$ are the number of colours and flavours, respectively. For $`g_\pi >g_\pi ^{cr}`$ fermions become massive and the vacuum accommodates a non-vanishing condensate $`\overline{\psi }\psi `$, and consequently there exists a massless Nambu-Goldstone boson, see Fig 6.1.
### 6.3 Meson channel
The quark-antiquark $`T`$-matrix in the pseudoscalar channel can be solved by using the Bethe-Salpeter equation in the random phase approximation (RPA), as shown in Fig. 6.2, see Ref. .
$`T(p_1,p_2,p_3,p_4)`$ $`=`$ $`f(p_1)f(p_2)\left[i\gamma _5\tau _i\right]{\displaystyle \frac{g_\pi i}{1+g_\pi J_\pi (q^2)}}\left[i\gamma _5\tau _i\right]f(p_3)f(p_4)`$ (6.8)
$`\times \delta (p_1+p_2p_3p_4),`$
where
$`J_\pi (q^2)`$ $`=`$ $`i\mathrm{Tr}{\displaystyle \frac{d^4k}{(2\pi )^4}f^2(k)\gamma _5\tau _iS(k)\gamma _5\tau _iS(q+k)f^2(q+k)},`$ (6.9)
$`=`$ $`6i{\displaystyle \frac{d^4k}{(2\pi )^4}\mathrm{tr}_D[\gamma _5S(k)\gamma _5S(k+q)]f^2(k)f^2(q+k)},`$
where $`q`$ denotes the total momentum of the $`\overline{q}q`$ pair. The pion mass $`m_\pi `$ corresponds to the pole of $`T`$-matrix. One immediately finds that $`m_\pi =0`$ if the current quark mass $`m_c`$ is zero, in accordance with Goldstone’s theorem. The residue of the $`T`$-matrix at this pole has the form
$$V^\pi (p_1,p_2)=ig_{\pi qq}[\text{1}\text{1}_c\tau ^a\gamma _5]f(p_1)f(p_2),$$
(6.10)
where $`g_{\pi qq}`$ is the pion-quark-antiquark coupling constant and is related to the corresponding loop integral $`J_\pi `$ by
$$g_{\pi qq}^2=\frac{dJ_\pi }{dq^2}|_{q^2=m_\pi ^2}.$$
(6.11)
Notice that $`Z=g_{\pi qq}^2`$ can be regarded as a pion wavefunction renormalisation constant. For space-time dimension $`D=4`$, one can show $`Z^1\mathrm{\Lambda }^2`$ (for the local NJL model we have $`Z^1\mathrm{Ln}\mathrm{\Lambda }`$), therefore in the continuum limit $`\mathrm{\Lambda }\mathrm{}`$ we have $`Z=0`$ which is precisely the compositeness condition . In this extreme limit pions become pointlike. The cutoff for spacetime $`D=4`$ can be removed only at the expense of making the theory trivial in the continuum limit. It has been shown that for four-fermion theories the renormalisability, nontriviality and compositeness are intimately related .
The pion decay constant $`f_\pi `$ is obtained from the coupling of the pion to the axial-vector current. Notice that due to the non-locality the axial-vector current is modified and consequently the one-pion-to-vacuum matrix element gets the additional contribution shown in Fig. 6.3. This new term is essential in order to maintain Gell-Mann-Oakes-Renner relation and makes a significant contribution. The pion decay constant is given by
$`f_\pi `$ $`=`$ $`{\displaystyle \frac{ig_{\pi \overline{q}q}}{m_\pi ^2}}{\displaystyle \frac{d^4k}{(2\pi )^4}\mathrm{Tr}[q/\gamma _5\frac{\tau _a}{2}(S(p_{}))\gamma _5\tau _a(S(p_+))]f(p_{})f(p_+)}`$ (6.12)
$`+`$ $`{\displaystyle \frac{ig_\pi }{2m_\pi ^2}}{\displaystyle \frac{d^4k}{(2\pi )^4}\mathrm{Tr}[S(k)]\frac{d^4k}{(2\pi )^4}\mathrm{Tr}[V^\pi (p_{},p_+)S(p_{})\gamma _5\tau _aS(p_+)]}`$
$`\times [f^2(k)\left(f^2(p_+)+f^2(p_{})\right)f(p_+)f(p_{})f(k)\left(f(k+q)+f(kq)\right)],`$
where $`V_\pi (p_{},p_+)`$ is defined in Eq. (6.10), with notation $`p_\pm =p\pm \frac{1}{2}q`$.
Our model contains five parameters: the current quark mass $`m_c`$, the cutoff ($`\mathrm{\Lambda }`$), the coupling constants $`g_\pi `$ , $`g_s`$ and $`g_a`$. We first fix $`g_\pi `$ and the current quark mass $`m_c`$ for arbitrary values of $`\mathrm{\Lambda }`$ by fitting $`f_\pi `$ and $`m_\pi `$ to their empirical values. In this way, we can consider the entire parameter space of the model. The corresponding solution of gap equation are shown in Fig. 6.4. In the left panel the constituent mass $`M`$ at zero momentum is shown as a function of the cutoff. It is obvious that for very small cutoff, there is no solution for the gap equation. On the right panel of Fig. 6.4, we show the corresponding values of the quark condensate. The quark condensate $`\overline{\psi }\psi =i\mathrm{Tr}S(0)`$ is closely related to the gap equation Eq (6.4). In the latter there appears an extra form factor inside the loop integral. The quark condensate with non-zero current-quark mass is quadratically divergent and is regulated by a subtraction of its perturbative value. These values can fall within the limits extracted from QCD sum rules $`190\text{MeV}\overline{q}q^{1/3}260\text{MeV}`$ at a renormalisation scale of $`1`$ GeV and lattice calculation , having in mind that QCD condensate is a renormalised and scale-dependent quantity. In right-hand side of Fig. 6.1 we show the quark condensate with respect to current quark mass for various parameter sets, it is seen that for large coupling (or large $`M(0)`$), the magnitude of quark condensate decreases as the current quark mass increases. This feature is consistent with the behaviours of lattice and sum rule results.
We analyse two sets of parameters, see Table 6.1. The set $`A`$ is a non-confining parameter set, while set $`B`$ leads to quark confinement (i.e., it satisfies the condition Eq. (6.7). The quark condensate in the chiral limit is $`(207\text{ MeV})^3`$ and $`(186\text{ MeV})^3`$ for sets A and B, respectively. At non-zero current quark mass one obtains $`(215\text{MeV})^3`$ and $`(191\text{MeV})^3`$ for sets A and B respectively. The position of the quark poles are given in Table 6.2 for two sets of parameters. In Fig. 6.5 we show the position of the first pole of quark propagator with respect to various cutoff, which indicates that for large cutoff, we have only real poles and we have indeed complex poles for reasonable range of cutoff. The real part of the first pole of dressed quark propagator can be considered in much the same as the quark mass in the ordinary NJL model. Since we do not believe in on-shell quarks or quark resonances, this is also a measure for a limit on the validity of the theory. The real part of the first quark propagator pole $`m_R^\text{q}`$ is larger than the constituent quark mass at zero momentum $`M(0)`$, as can be seen in Table 6.1.
As we will see the mass $`m_R^\text{q}`$ appears as an important parameter in diquark and nucleon solution rather than the constituent quark mass. The same feature has been seen in the studies of the soliton in this model, where $`m_R^\text{q}`$ determines the stability of the soliton . In contrast to the local NJL model, here the dynamical quark mass Eq. (6.6) is momentum dependent and follows a trend similar to that estimated from lattice simulations . Although this is less fundamental since one is free to choose the form factor, nevertheless the quark mass is a gauge dependent object and is not directly observable.
The parameters $`g_s`$ and $`g_a`$ are yet to be determined, we shall treat them as free parameters, which allows us to analyse baryon solutions in terms of a complete set of couplings. The coupling-constant dependence appears through the ratios $`r_s=g_s/g_\pi `$ and $`r_a=g_a/g_\pi `$.
### 6.4 Diquark channels
In the rainbow-ladder approximation the scalar $`qq`$ $`T`$-matrix can be calculated from a very similar diagram to that shown in Fig. 6.2 (the only change is that the anti-quark must be replaced by a quark with opposite momentum). It can be written as
$`T(p_1,p_2,p_3,p_4)`$ $`=`$ $`f(p_1)f(p_2)\left[\gamma _5C\tau _2\beta ^A\right]\tau (q)\left[C^1\gamma _5\tau _2\beta ^A\right]f(p_3)f(p_4)`$ (6.13)
$`\times \delta (p_1+p_2p_3p_4),`$
with
$$\tau (q)=\frac{2g_si}{1+g_sJ_s(q^2)},$$
(6.14)
where $`q=p_1+p_2=p_3+p_4`$ is the total momentum of the $`qq`$ pair, and
$`J_s(q^2)`$ $`=`$ $`i\mathrm{Tr}{\displaystyle \frac{d^4k}{(2\pi )^4}f^2(k)\left[\gamma _5C\tau _2\beta ^A\right]S(k)^T\left[C^1\gamma _5\tau _2\beta ^A\right]S(q+k)f^2(q+k)},`$ (6.15)
$`=`$ $`6i{\displaystyle \frac{d^4k}{(2\pi )^4}\mathrm{tr}_D[\gamma _5S(k)\gamma _5S(k+q)]f^2(k)f^2(q+k)}.`$
In the above equation the quark propagators $`S(k)`$ are the solution of the rainbow SDE Eq. (6.5). The denominator of Eq. (6.14) is the same as in the expression for the pion channel, Eq. (6.8), if $`g_s=g_\pi `$. One may thus conclude that at $`r_s=1`$ the diquark and pion are degenerate. This puts an upper limit to the choice of $`r_s`$, since diquarks should not condense in vacuum.
One can approximate $`\tau (q)`$ by an effective diquark exchange between the external quarks, and parametrise $`\tau (q)`$ around the pole as
$$\tau (q)2ig_{dsqq}^2V^s(q)D(q),D^1(q)=q^2M_{ds}^2,$$
(6.16)
where $`M_{ds}`$ is the scalar diquark mass, defined as the position of the pole of $`\tau (q)`$. The strength of the on-shell coupling of scalar diquark to quarks, $`g_{dsqq}`$ is related to the polarisation operator $`J_s`$ by
$$g_{dsqq}^2=\frac{dJ_s}{dq^2}|_{q^2=M_{ds}^2},$$
(6.17)
and $`V^s(q)`$ is the ratio between the exact $`T`$-matrix and on-shell approximation. It is obvious that we should have $`V^s(q)|_{q^2=M_{ds}^2}=1`$.
Here, there is no mixing between the axial-vector diquark channel with others, therefore in the same way one can write the axial-vector diquark $`T`$-matrix in a similar form
$`T(p_1,p_2,p_3,p_4)`$ $`=`$ $`f(p_1)f(p_2)\left[\gamma _\mu C\tau _i\tau _2\beta ^A\right]\tau ^{\mu \nu }(q)\left[C^1\gamma _\nu \tau _2\tau _i\beta ^A\right]f(p_3)f(p_4)`$ (6.18)
$`\times \delta (p_1+p_2p_3p_4),`$
with
$$\tau ^{\mu \nu }(q)=2g_ai\left[\frac{g^{\mu \nu }q^\mu q^\nu /q^2}{1+g_aJ_a^T(q^2)}+\frac{q^\mu q^\nu /q^2}{1+g_aJ_a^L(q^2)}\right].$$
(6.19)
Here we prefer to decompose the axial polarisation tensor into longitudinal and transverse channels as well,
$`J_a^{\mu \nu }(q^2)`$ $`=`$ $`i\mathrm{Tr}{\displaystyle \frac{d^4k}{(2\pi )^4}f^2(k)\left[\gamma ^\mu C\tau _i\tau _2\beta ^A\right]S(k)^T\left[C^1\gamma ^\nu \tau _2\tau _i\beta ^A\right]S(q+k)f^2(q+k)},`$ (6.20)
$`=`$ $`6i{\displaystyle \frac{d^4k}{(2\pi )^4}\mathrm{tr}_D[\gamma ^\mu S(k)\gamma ^\nu S(k+q)]f^2(k)f^2(q+k)}`$
$`=`$ $`J_a^T(q^2)(g^{\mu \nu }q^\mu q^\nu /q^2)+J_a^L(q^2)q^\mu q^\nu /q^2.`$
We find that the longitudinal channel does not produce a pole (see Fig. 6.6), and thus the bound axial-vector diquark solution corresponds to a pole of the transverse $`T`$-matrix. The transverse component of $`\tau ^{\mu \nu }(q)`$ matrix is now approximated by $`M_{da}`$ as,
$$\tau ^{\mu \nu }(q)2ig_{daqq}^2V^a(q)D^{\mu \nu }(q),D^{\mu \nu }(q)=\frac{g^{\mu \nu }q^\mu q^\nu /q^2}{q^2M_{da}^2},$$
(6.21)
where $`V^a(q)`$ includes the off-shell contribution to the $`T`$-matrix. The coupling constant $`g_{daqq}`$ is related to the residue at the pole of the $`T`$-matrix,
$$g_{daqq}^2=\frac{dJ_a^T}{dq^2}|_{q^2=M_{da}^2}.$$
(6.22)
#### 6.4.1 Diquark Solution
The loop integrations in Eq. (6.15, 6.20) are evaluated in Euclidean space<sup>2</sup><sup>2</sup>2We work in Euclidean space with metric $`g^{\mu \nu }=\delta ^{\mu \nu }`$ and a hermitian basis of Dirac matrices $`\{\gamma _\mu ,\gamma _\nu \}=2\delta _{\mu \nu }`$, with a standard transcription rules from Minkowski to Euclidean momentum space: $`k^0ik_4`$, $`\stackrel{}{k}^M\stackrel{}{k}^E`$. For the current model, the usual analytic continuation of amplitudes from Euclidean to Minkowski space can not be used. This is due to the fact that quark propagators of the model contain many poles at complex energies leading to opening of a threshold for decay of a diquark (or meson) into other unphysical states. Any theory of this type need to be equipped with an alternative continuation prescription consistent with unitarity and macrocausality. Let us define a fictitious two-body threshold as twice $`m_R^\text{q}`$. For a confining parameter set, each quark propagator has a pair of complex-conjugate poles. Above the two-body pseudo-threshold $`q^2<4(m_R^\text{q})^2`$, where $`q`$ is meson (diquark) momentum, the first pair of complex poles of the quark propagator has a chance to cross the real axis. According to the Cutkosky prescription , if one is to preserve the unitarity and the microcausality, the integration contour should be pinched at that point. In this way, one can ensure that there is no spurious $`\overline{q}q`$ (or $`qq`$) production threshold, for energies below the next pseudo-threshold, i.e. twice the real part of the second pole of the quark propagator. Note that it has been shown that the removal of the $`\overline{q}q`$ pseudo-threshold is closely related to the existence of complex poles in the form of complex-conjugate pairs. Since there is no unique analytical continuation method available for such problems, any method must be regarded as a part of the model assumptions . Here, we follow the method used in Ref. .
We use the parameter sets determined in the mesonic sector shown in table 6.1. Our numerical computation is valid below the first $`qq`$ pseudo-threshold. Note that the longitudinal polarisability $`J_a^L(q)`$ defined in Eq. (6.20) does not vanish here. However this term can probably be ignored since it does produce any poles in the $`T`$-matrix, and moreover there is no conserved current associated to this channel. We find that for a wide range of $`r_s`$ and $`r_a`$, for all parameter sets, a bound scalar and axial-vector diquark exist (the results for additional sets can be found in ). This is in contrast to the normal NJL model where a bound axial-vector diquark exists only for very strong interaction . The diquark masses for various values of $`r_s`$ and $`r_a`$ are plotted in Fig. 6.7. As already pointed out, the scalar diquark mass is equal to the pion mass at $`r_s=1`$. It is obvious from Fig. 6.7 that for $`r_s=r_a`$ the axial-vector diquark is heavier than the scalar diquark, and consequently is rather loosely bound. For very small $`r_s`$ and $`r_a`$ one finds no bound state of either diquark. This kind of diquark confinement is due to the screening effect of the ultraviolet cutoff and can not be associated with confinement in QCD which originates from the infrared divergence of the gluon and ghost propagators. Having said that, it is possible that real diquark confinement may arise beyond the ladder approximation . There, in order to preserve Goldstone’s theorem at every order, we must include additional terms in the interaction. Although these new terms should have minimal impact on the solutions for the colour-singlet meson channels, they can provide a repulsive contribution to the colour-antitriplet diquark channels which removes the asymptotic-diquark solutions from the spectrum. This would indicate that diquark confinement is an independent phenomenon and is not related to the particular realisation of quark confinement.
One should note that the nucleon bound state in the diquark-quark picture does not require asymptotic-diquark states since the diquark state is merely an intermediate device which simplifies the three-body problem. Nevertheless, evidence for correlated diquark states in baryons is found in deep-inelastic lepton scatterings and in hyperon weak decays . At the same time, diquarks appear as bound states in many phenomenological models. It is puzzling that diquarks are even seen in lattice calculations . In contrast to our perception of QCD colour confinement, the corresponding spectral functions for these supposedly confined objects in the colour anti-triplet channel are very similar to mesonic ones .
In Fig. 6.8, we show the scalar diquark-quark-quark coupling defined in Eq. (6.17) with respect to various scalar diquark couplings. A pronounced change in behaviour around the quark-quark pseudo-threshold is observed in the confining set $`B`$, and this seems to justify our emphasis on the pseudo-threshold defined by twice the real part of the quark pole.
Next we study the off-shell behaviour of the diquark $`T`$-matrix. In Fig. 6.9 we show the discrepancy between the exact $`T`$-matrix and the on-shell approximation $`V^{s,a}(q)`$. At the pole we have by definition that $`V^{s,a}(q)|_{q^2=M_{s,a}^2}=1`$. We see that elsewhere the off-shell contribution is very important due to the non-locality of our model. We find that the bigger the diquark mass is, the bigger the off-shell contribution. The off-shell behaviour of the scalar and the axial-vector channel for both parameter sets $`A`$ and $`B`$ are rather similar.
### 6.5 Three-body sector
In order to make three-body problem tractable, we discard any three-particle irreducible graphs (this is sometimes called the Faddeev approximation). The relativistic Faddeev equation can be then written as an effective two-body BS equation for a quark and a diquark due to the locality of the form factor in momentum space (see Eq. (6.3)) and accordingly the separability of the two-body interaction in momentum-space. We adopt the formulation developed by the Tübingen group to solve the resulting BS equation. In the following we work in momentum space with Euclidean metric. The BS wave function for the octet baryons can be presented in terms of scalar and axialvector diquarks correlations,
$$\psi (p,P)u(P,s)=\left(\begin{array}{c}\psi ^5(p,P)\\ \psi ^\mu (p,P)\end{array}\right)u(P,s),$$
(6.23)
where $`u(P,s)`$ is a basis of positive-energy Dirac spinors of spin $`s`$ in the rest frame. The parameters $`p=(1\eta )p_i\eta (p_j+p_k)`$ and $`P=p_i+p_j+p_k`$ are the relative and total momenta in the quark-diquark pair, respectively. The Mandelstam parameter $`\eta `$ describes how the total momentum of the nucleon $`P`$ is distributed between quark and diquark.
One may alternatively define the vertex function associated with $`\psi (p,P)`$ by amputating the external quark and diquark propagators (the legs) from the wave function;
$$\varphi (p,P)=S^1(p_q)\stackrel{~}{D}^1(p_d)\left(\begin{array}{c}\psi ^5(p,P)\\ \psi ^\nu (p,P)\end{array}\right),$$
(6.24)
with
$$\stackrel{~}{D}^1(p_d)=\left(\begin{array}{cc}D^1(p_d)& 0\\ 0& (D^{\mu \nu }(p_d))^1,\end{array}\right)$$
(6.25)
where $`D(p),D^{\mu \nu }(p)`$ and $`S(p)`$ are Euclidean versions of the diquark and quark propagators which are obtained by the standard transcription rules from the expressions in Minkowski space, Eqs. (6.16,6.21) and Eq. (6.5), respectively. The spectator quark momentum $`p_q`$ and the diquark momentum $`p_d`$ are given by
$`p_q`$ $`=`$ $`\eta P+p,`$ (6.26)
$`p_d`$ $`=`$ $`(1\eta )Pp,`$ (6.27)
with similar expressions for $`k_{q,d}`$, where we replace $`p`$ by $`k`$ on the right-hand side. In the ladder approximation, the coupled system of BS equations for octet baryon wave functions and their vertex functions takes the compact form,
$$\varphi (p,P)=\frac{d^4k}{(2\pi )^4}K^{BS}(p,k;P)\psi (k,P),$$
(6.28)
where $`K^{BS}(p,k;P)`$ denotes the kernel of the nucleon BS equation representing the exchange quark within the diquark with the spectator quark (see Fig. 5.1), and in the colour singlet and isospin $`\frac{1}{2}`$ channel we find (see Ref. )
$$K^{BS}(p,k;P)=3\left(\begin{array}{cc}\chi ^5(p_1,k_d)S^T(q)\overline{\chi }^5(p_2,p_d)& \sqrt{3}\chi ^\alpha (p_1,k_d)S^T(q)\overline{\chi }^5(p_2,p_d)\\ \sqrt{3}\chi ^5(p_1,k_d)S^T(q)\overline{\chi }^\mu (p_2,p_d)& \chi ^\alpha (p_1,k_d)S^T(q)\overline{\chi }^\mu (p_2,p_d)\end{array}\right),$$
(6.29)
where $`\chi `$ and $`\chi ^\mu `$ (and their adjoint $`\overline{\chi }`$ and $`\overline{\chi }^\mu `$) stand for the Dirac structures of the scalar and the axial-vector diquark-quark-quark vertices and can be read off immediately from Eqs. (6.13, 6.16) and Eqs. (6.18, 6.21), respectively. Therefore we have
$`\chi ^5(p_1,k_d)`$ $`=`$ $`g_{dsqq}(\gamma ^5C)\sqrt{2V^s(k_d)}f(p_1+(1\sigma )k_d)f(p_1+\sigma k_d),`$
$`\chi ^\mu (p_1,k_d)`$ $`=`$ $`g_{daqq}(\gamma ^\mu C)\sqrt{2V^a(k_d)}f(p_1+(1\sigma )k_d)f(p_1+\sigma k_d).`$ (6.30)
We have used an improved on-shell approximation for the contribution of diquark $`T`$-matrix occurring in the Faddeev equations. Instead of the exact diquark $`T`$-matrices we use the on-shell approximation with a correction of their off-shell contribution through $`V^{s,a}(p)`$. What is neglected is then the contribution to the $`T`$-matrix beyond the pseudo-threshold. As we will see this approximation is sufficient to obtain a three-body bound state. In order to evaluate the structure of the diquark $`T`$-matrix completely, one normally employs the dispersion relation, however, this is not applicable here, due to non-analyticity of the diquark $`T`$ matrix. Notice, as we already pointed out for the confining set B, we do not have $`qq`$ continuum, however, there exists many complex poles beyond the pseudo-threshold which might be ignored, provided that they lie well above the energies of interest.
The relative momentum of quarks in the diquark vertices $`\chi `$ and $`\chi ^\mu `$ are defined as $`p_1=p+k/2(13\eta )P/2`$ and $`p_2=kp/2+(13\eta )P/2`$, respectively. The momentum $`k_d`$ of the incoming diquark and the momentum $`p_d`$ of the outgoing diquark are defined in Eq. (6.27) (see Fig. 5.1). The momentum of the exchanged quark is fixed by momentum conservation at $`q=pk+(12\eta )P`$.
It is interesting to note the non-locality of the diquark-quark-quark vertices naturally provides a sufficient regularisation of the ultraviolet divergence in the diquark-quark loop. In the expressions for the momenta we have introduced two independent Mandelstam parameters $`\eta ,\sigma `$, which can take any value in $`[0,1]`$. They parametrise different definitions of the relative momentum within the quark-diquark ($`\eta `$) or the quark-quark system ($`\sigma `$). Observables should not depend on these parameters if the formulation is Lorentz covariant. This means that for every BS solution $`\psi (p,P;\eta _1,\sigma _1)`$ there exists a equivalent family of solutions. This provides a stringent check on calculations, see the next section for details.
We now constrain the Faddeev amplitude to describe a state of positive energy, positive parity and spin $`s=1/2`$. The parity condition can be immediately reduced to a condition for the BS wave function:
$$𝒫\left(\begin{array}{c}\psi ^5(p,P)\\ \psi ^\mu (p,P)\end{array}\right)=\left(\begin{array}{c}\gamma ^4\psi ^5(\overline{p},\overline{P})\gamma ^4\\ \gamma ^4\mathrm{\Lambda }_𝒫^{\mu \nu }\psi ^\nu (\overline{p},\overline{P})\gamma ^4\end{array}\right)=\left(\begin{array}{c}\psi ^5(p,P)\\ \psi ^\mu (p,P)\end{array}\right),$$
(6.31)
where we define $`\overline{p}=\mathrm{\Lambda }_𝒫p`$ and $`\overline{P}=\mathrm{\Lambda }_𝒫P`$, with $`\mathrm{\Lambda }_𝒫^{\mu \nu }=\text{diag}(1,1,1,1)`$. In order to ensure the positive energy condition, we project the BS wave function with the positive-energy projector $`\mathrm{\Lambda }^+=(1+\widehat{P/})`$, where the hat denotes a unit four vector (in rest frame we have $`\widehat{P}=P/iM`$). Now we expand the BS wave function $`\psi (p,P)`$ in Dirac space $`\mathrm{\Gamma }\{\text{1},\gamma _5,\gamma ^\mu ,\gamma _5\gamma ^\mu ,\sigma ^{\mu \nu }\}`$. The above-mentioned conditions reduce the number of independent component from sixteen to eight, two for the scalar diquark channel, $`S_i,(i=1,2)`$ and six for the axial-diquark channel, $`A_i,(i=1,\mathrm{}6)`$. The most general form of the BS wave function is given by
$`\psi ^5(p,P)`$ $`=`$ $`\left(S_1i\widehat{p/}_TS_2\right)\mathrm{\Lambda }^+,`$
$`\psi ^\mu (p,P)`$ $`=`$ $`(i\widehat{P}^\mu \widehat{p/}_TA_1+\widehat{P}^\mu A_2\widehat{p}_T^\mu \widehat{p/}_TA_3+i\widehat{p}_T^\mu A_4+(\widehat{p}_T^\mu \widehat{p/}_T\gamma _T^\mu )A_5`$ (6.32)
$`(i\gamma _T^\mu \widehat{p/}_T+i\widehat{p}_T^\mu )A_6)\gamma _5\mathrm{\Lambda }^+.`$
Here we write $`\gamma _T^\mu =\gamma ^\mu \widehat{P/}\widehat{P}^\mu `$. The subscript $`T`$ denotes the component of a four-vector transverse to the nucleon momentum, $`p_T=p\widehat{P}(p.\widehat{P})`$. In the same way, one can expand the vertex function $`\varphi `$ in Dirac space, and since the same constraints apply to the vertex function, we obtain an expansion quite similar to Eq. (6.32), with new unknown coefficients $`𝕊_i`$ and $`𝔸_i`$ which are substituted the coefficients $`S_i`$ and $`A_i`$, respectively. The unknown scalar function $`S_i(𝕊_i)`$ and $`A_i(𝔸_i)`$ depend on the two scalars which can be built from the nucleon momentum $`P`$ and relative momentum $`p`$, $`z=\widehat{P}.\widehat{p}=\mathrm{cos}\omega `$ (the cosine of the four-dimensional azimuthal angle of $`p^\mu `$) and $`p^2`$.
In the nucleon rest frame, one can rewrite the Faddeev amplitude in terms of tri-spinors each possessing definite orbital angular momentum and spin . It turns out that these tri-spinors can be written as linear combinations of the eight components defined in Eq. (6.32). Thus from knowledge of $`S_i`$ and $`A_i`$, a full partial wave decomposition can be immediately obtained . Notice that although the diquarks are not pointlike objects here, they do not carry orbital angular momentum i. e. $`L^2\chi ^{5,\mu }(q)=0`$. This is due to the fact that the off-shell contribution $`V^{s,a}(q)`$ is a function of scalar $`q^2`$. Moreover, the form factor in our model Lagrangian is also scalar, hence the total momentum dependent part of the diquark-quark-quark vertices are scalar functions and carry no orbital angular momentum. Therefore, the partial wave decomposition obtained in Ref. for pointlike diquarks can be used here. Note that no such partial wave decomposition can be found if one uses the BS vertex function $`\varphi ^{5,\mu }`$ since the axial-vector diquark propagator mixes the space component of the vertex function and time component of the axial-vector diquark.
#### 6.5.1 Numerical method for the coupled BS equations
For solving the BS equations we use the algorithm introduced by Oettel et al . The efficiency of this algorithm has already been reported in several publications, see for example Refs. . We will focus here only on the key ingredients of this method. The momentum dependence of quark mass in our model increases the complexity of the computation significantly.
As usual, we work in the rest frame of the nucleon $`P=(0,iM_N)`$. In this frame we are free to chose the spatial part of the relative momentum $`p`$ parallel to the third axis. Thus the momenta $`p`$ and $`k`$ are given by
$`p^\mu `$ $`=`$ $`|p|(0,0,\sqrt{1z^2},z),`$
$`k^\mu `$ $`=`$ $`|k|(\mathrm{sin}\theta ^{}\mathrm{sin}\varphi ^{}\sqrt{1z^2},\mathrm{sin}\theta ^{}\mathrm{cos}\varphi ^{}\sqrt{1z^2},\mathrm{cos}\theta \sqrt{1z^2},z^{}),`$ (6.33)
where we write $`z=\mathrm{cos}\omega `$ and $`z^{}=\mathrm{cos}\omega ^{}`$. The wave function Eq. (6.32) consists of $`2\times 2`$-blocks in Dirac space can be simplified to
$`\psi ^5(p,P)`$ $`=`$ $`\left(\begin{array}{cc}S_1(p^2,z)& 0\\ \sigma _3\sqrt{1z^2}S_2(p^2,z)& 0\end{array}\right),\psi ^4(p,P)=\left(\begin{array}{cc}\sigma _3\sqrt{1z^2}A_1(p^2,z)& 0\\ A_2(p^2,z)& 0\end{array}\right),`$ (6.38)
$`\psi ^3(p,P)`$ $`=`$ $`\left(\begin{array}{cc}i\sigma _3A_3(p^2,z)& 0\\ i\sqrt{1z^2}A_4(p^2,z)& 0\end{array}\right),\psi ^2(p,P)=\left(\begin{array}{cc}i\sigma _2A_5(p^2,z)& 0\\ \sigma _1\sqrt{1z^2}A_6(p^2,z)& 0\end{array}\right),`$ (6.43)
$`\psi ^1(p,P)`$ $`=`$ $`\left(\begin{array}{cc}i\sigma _1A_5(p^2,z)& 0\\ \sigma _2\sqrt{1z^2}A_6(p^2,z)& 0\end{array}\right).`$ (6.46)
The great advantage of this representation is that the scalar and the axial-vector components are decoupled. Therefore the BS equation decomposes into two sets of coupled equations, two for the scalar diquark channel and six for the axial diquark channel. We expand the vertex (wave) functions in terms of Chebyshev polynomials of the first kind, which are closely related to the expansion into hyperspherical harmonics. This decomposition turns out to be very efficient for such problems . Explicitly,
$`F_i^\psi (p^2,z)`$ $`=`$ $`{\displaystyle \underset{n=0}{\overset{n_{max}}{}}}i^nF_i^{\psi (n)}(p^2)T_n(z),`$
$`F_i^\varphi (p^2,z)`$ $`=`$ $`{\displaystyle \underset{m=0}{\overset{m_{max}}{}}}i^nF_i^{\varphi (m)}(p^2)T_m(z),`$ (6.47)
where $`T_n(z)`$ is the Chebyshev polynomial of the first kind. We use a generic notation, the functions $`F_i^\psi `$( and $`F_i^\varphi `$) substituting the function $`S_i,A_i`$ (and $`𝕊_i,𝔸_i`$),
$`S_{1,2}F_{1,2}^\psi ,A_{1\mathrm{}6}F_{3\mathrm{}8}^\psi ,`$
$`𝕊_{1,2}F_{1,2}^\varphi ,𝔸_{1\mathrm{}6}F_{3\mathrm{}8}^\varphi .`$ (6.48)
We truncate the Chebyshev expansions involved in $`F_i^\psi `$ and $`F_i^\varphi `$ at different orders $`n_{max}`$ and $`m_{max}`$, respectively. We also expand the quark and diquark propagators into Chebyshev polynomials. In this way one can separate the $`\widehat{P}\widehat{p}`$ and $`\widehat{P}\widehat{k}`$ dependence in Eqs. (6.286.24). Using the orthogonality relation between the Chebyshev polynomials, one can reduce the four dimensional integral equation into a system of coupled one-dimensional equations. Therefore one can rewrite Eqs. (6.24, 6.28) in the matrix form
$`F_i^{\psi (n)}(p^2)`$ $`=`$ $`{\displaystyle \underset{j=1}{\overset{8}{}}}{\displaystyle \underset{m=0}{\overset{m_{max}}{}}}g_{ij}^{nm}(p^2)F_j^{\varphi (m)}(p^2),`$
$`F_i^{\varphi (m)}(p^2)`$ $`=`$ $`{\displaystyle \underset{j=1}{\overset{8}{}}}{\displaystyle \underset{n=0}{\overset{n_{max}}{}}}{\displaystyle _0^{\mathrm{}}}d|k||k|^3H_{ij}^{mn}(k^2,p^2)F_j^{\psi (n)}(k^2).`$ (6.49)
Here $`g_{ij}^{nm}`$ and $`H_{ij}^{mn}`$ are the matrix elements of the propagator and the quark exchange matrices, respectively. The indices $`n,m`$ give the Chebyshev moments and $`i,j`$ denote the individual channels. To solve Eq. (6.49), we first rewrite it in the form of linear eigenvalue problem. Schematically
$$\lambda (P^2)\phi =K(P^2)\phi ,$$
(6.50)
with the constraint that $`\lambda (P^2)=1`$ at $`P^2=M_N^2`$. This can be used to determine the nucleon mass $`M_N`$ iteratively.
As already pointed out the BS solution should be independent of the Mandelstam parameters $`\eta ,\sigma `$. As can be seen in Fig. 6.10, there is indeed a large plateau for the $`\eta `$ dependence if we use a high cut-off on the Chebyshev moments. The limitations on the size of this area of stability can be understood by considering where the calculation contains singularities due to quark and diquark poles,
$`\eta `$ $``$ $`[1{\displaystyle \frac{M_{ds}}{M_N}},{\displaystyle \frac{m_R^\text{q}}{M_N}}],\text{if}M_{ds}<M_{da},`$
$`\eta `$ $``$ $`[1{\displaystyle \frac{M_{da}}{M_N}},{\displaystyle \frac{m_R^\text{q}}{M_N}}],\text{if}M_{da}<M_{ds}.`$ (6.51)
A similar plateau has been found in other applications . \[Other complex poles lies out side the minimal region given in Eq. (6.51).\] The singularities in the quark-exchange propagator put another constraint on the acceptable range of $`\eta `$; $`\eta >\frac{1}{2}(1\frac{m_R^\text{q}}{M_N})`$. No such constraint exists for $`\sigma `$, which relates to the relative momentum between two quarks. To simplify the algebra we take $`\sigma =1/2`$.
In what follows we use a momentum mesh of $`60\times 60`$ for $`p,k`$, mapped in a non-linear way to a finite interval. In the non-singular regime of Mandelstam parameter $`\eta `$ Eq. (6.51), the Faddeev solution is almost independent of the upper limit on the Chebyshev expansion, and for $`m_{max}=10,n_{max}=12`$, see the Fig. 6.10, this seems to be satisfied. This limit is some higher than the reported values for simple models .
#### 6.5.2 Nucleon Solution
In order to understand the role of the axial diquark in nucleon solution, we first consider the choice $`r_a=0`$. For this case we find that the non-confining set A can not generate a three-body bound state in this model without the inclusion of the off-shell contribution. For the confining set B one also has to enhance the diquark-quark-quark coupling $`g_{dsqq}`$ by a factor of about $`1.73`$ over the value defined in Eq. (6.17) (as we will show, this extra factor is not necessary when the axial-vector is included). The situation is even more severe in the on-shell treatment of the local NJL model, since one needs to include the full $`qq`$ continuum contribution in order to find a three-body bound state when the axial-vector diquark channel is not taken into account .
As can be seen from Fig. 6.7 a decrease in $`r_s`$ leads to a larger diquark mass, and an increase in the off-shell contribution to the $`qq`$ $`T`$-matrix (see Fig. 6.9). It is this off-shell correction is need for a bound nucleon.
The nucleon result is shown in Fig. 6.11. We also show a fictitious diquark-quark threshold defined as $`M_{ds}+m_R^\text{q}`$. The nucleon mass can be seen to depend roughly linearly on the scalar diquark mass. A similar behaviour is also seen in the local NJL model . Increasing the diquark mass (or decreasing $`r_s`$) increases the nucleon mass, i.e. the scalar diquark channel is attractive. In order to obtain a nucleon mass of $`940`$ MeV, we need diquark mases of $`608`$ MeV and $`623`$ MeV for set A and B, respectively. The corresponding nucleon binding energy measured from the diquark-quark threshold are $`56`$ MeV and $`91`$ MeV for set A and B, respectively, compared to the binding of the diquarks (relative to the $`qq`$ pseudo threshold) of about $`174`$ and $`193`$ MeV for set A and B, respectively. Such diquark clustering within the nucleon is also observed in the local NJL model , and is qualitatively in agreement with a instanton model and lattice simulations . The nucleon solutions for sets A and B behave rather differently with respect to the diquark-quark threshold, see Fig. 6.12. This indicates that for non-confining set A, the nucleon solution is rather sensitive to the diquark-quark threshold and tends not avoid it. However, for confining set, since there is no well-defined threshold, this tendency is absent and as we approach to the diquark-quark threshold, the nucleon binding energy decreases and can approach to zero.
Next we investigate the effect of the axial-vector diquark channel on nucleon solution. We find that the axial-vector diquark channel contributes considerably to the nucleon mass<sup>3</sup><sup>3</sup>3Note however that neglecting $`\pi N`$-loops may lead to a quantitative overestimate of the axial-vector diquark role in the nucleon . and takes away the need for the artificial enhancement of the coupling strength for set B. In Figs. 6.13, 6.14 we show the nucleon mass as a function of the scalar and axial-vector diquark mass. Similar to the scalar diquark channel, we define the axial-vector diquark-quark threshold as $`M_{da}+m_R^\text{q}`$. We see that as one increase the axial-vector diquark (and scalar diquark) masses, the $`qq`$ interaction is weakened and consequently the nucleon mass is increases. Therefore the contribution of the axial-vector channel to the nucleon mass is also attractive.
In Fig. 6.15 we plot the parameter space of the interaction Lagrangian with variable $`r_s`$ and $`r_a`$ which leads to the nucleon mass $`M_N=0.940`$ GeV. The trend of this plot for the non-confining set A is very similar to the one obtained in the local NJL model (although we use a different parameter set) and roughly depends linearly on the ratios $`r_s`$ and $`r_a`$,
$$M_N[0.940\text{GeV}]=r_s0.94r_a+1.2.$$
(6.52)
Therefore, any interaction Lagrangian with $`r_s`$ and $`r_a`$ which satisfies the relation $`r_s+0.94r_a=0.30`$ gives a nucleon mass at about the experimental value. This relation shows how interaction is shared between scalar and axial-vector channels. If the scalar diquark interaction $`r_s`$ is less than $`0.14`$, we need the axial-vector interaction to be stronger than the scalar diquark channel $`r_a>r_s`$ in order to get the experimental value of nucleon mass. For set B, as we approach to $`r_a=0`$, the curve bends upward, reflecting the fact that we have no bound state with only scalar diquark channel. In Fig. 6.15 we see for the confining set B that the interaction is again shared between the scalar and the axial-vector diquark and for small $`r_s<0.19`$ one needs a dominant axial-vector diquark channel $`r_a>r_s`$. It is obvious that the axial-vector diquark channel is much more important in the confining than the non-confining phase of model.
In order to study the implications of the quark confinement for the description of the nucleon, we compare in Table 6.3 three representative cases for both the non-confining and confining parameter sets, which all give nucleon mass about $`940`$ MeV. The first three columns contain results for set $`A`$, and the last three columns for the confining set $`B`$.
Given the definition of diquark-quark thresholds, in the presence of both scalar and axial-vector diquark channels, the diquarks in the nucleon can be found much more loosely bound, although one obtains a very strongly bound nucleon solution near its experimental value, see table 6.3. Next we study the nucleon BS wave function for the various sets given in Table 6.3. The nucleon wave and vertex function are not physical observables, but rather they suggest how observables in this model will behave. In Figs. 6.16-21 we show the leading Chebyshev moments of the scalar functions of the nucleon BS wave function for various sets (A1-3 and B1-3) which describes the strengths of the quark-diquark partial waves with $`S`$ as a total quark-diquark spin and $`L`$ as a total orbital angular momentum. They are normalised to $`F_1^{\varphi (0)}(p_1)=1`$, where $`p_1`$ is the first point of the momentum mesh. It is seen that the contribution of higher moments are considerably small, indicating a rapid convergence of the wave function amplitudes in terms of Chebyshev polynomials. In the confining case Figs. 6.19-21 there is a clear interference which is not present in the non-confining Figs. 6.16-19. Therefore in the confining case, all wave function amplitudes are somehow shifted to higher relative four-momenta between diquark and quark. In order to understand the role of this interference we obtain a mass density for the various channels. This density is defined as
$$\rho (p_{},P)=𝑑p_4\psi ^{}(p_{},p_4,P)\stackrel{~}{D}^1(p_d)\psi (p_{},p_4,P)$$
(6.53)
where $`p_{}`$ stands for space component of relative momentum $`p`$ and $`\stackrel{~}{D}^1(p_d)`$ defined in Eq. (6.25). This definition corresponds to a diagram occurring for the calculation of the isoscalar quark condensate in the impulse approximation . In the above definition of the density function, we have integrated over the time component of the relative momentum. In this way the density function becomes very similar to its counterpart in Minkowski space. Although the above definition of density is not unique, it does provide a useful measure of the spatial extent of the wave function (we have examined the possibility of taking matrix elements of other operators between the BS wave function, since this does not lead to any significant effect, the results are not presented here). The results are plotted in Figs. 6.22 and 6.23.
It is noticeable that in various sets, the $`s`$-wave is the dominant contribution to the ground state. The relative importance of the scalar and the axial diquark amplitude in the nucleon changes with the strength of the diquark-quark couplings $`g_{dsqq}(g_{daqq})`$ and accordingly with $`r_s(r_a)`$. We see that in the confining sets, the nucleon density extends to higher relative momentum between the diquark and the quark. This indicates a more compact nucleon in the confining case. In order to find a qualitative estimation of the confinement effect in our model, we calculate $`p_{}^{\text{RMS}}=(p_{}^2p_{}^2)^{1/2}`$, the results can be found in table 6.3. This can be related to the mean-square radius of the nucleon, if we assume minimal uncertainty. We see in the both confining and non-confining cases a decrease in $`p_{}^{\text{RMS}}`$ with weakening axial-vector diquark interaction (and consequently increasing the scalar diquark interaction strength). If we compare $`p_{}^{\text{RMS}}`$ for the two sets $`A2,B2`$, which have very similar interaction parameters $`r_s(r_a)`$, an increase about $`25\%`$ is found. This effect can not only be associated with the non-locality of our interaction, since that is present in both confining and non-confining cases.
### 6.6 Summary and Outlook
In this chapter we investigated the two- and three-quark problems in a non-local NJL model. We have truncated the diquark sector to the bound scalar and the axial-vector channels. We have solved the relativistic Faddeev equation for this model and have studied the behaviour of the nucleon solutions with respect to various scalar and the axial-vector interactions. We have also investigated a possible implication of the quark confinement of our model in the diquark and the baryon sector.
Although the model is quark confining, it is not diquark confining (at least in the rainbow-ladder approximation). A bound diquark can be found in both scalar and the axial-vector channel for a wide range of couplings. We have found that the off-shell contribution to the diquark $`T`$-matrix is crucial for the calculation of the structure of the nucleon: without its inclusion the attraction in the diquark channels is too weak to form a three-body bound state. We have also found that both the scalar and the axial-vector contribute attractively to the nucleon mass. The role of axial-vector channel is much more important in the confining phase of model. The nucleon in this model is strongly bound although the diquarks within nucleon are loosely bound. The confining aspects of the model are more obvious in three-body, rather than the two-body sector. By investigating the nucleon wave function we showed that quark confinement leads to a more compact nucleon. The size of nucleon is reduced by about $`25\%`$ in confining phase.
For both confining and non-confining phases, an increase in the scalar diquark channel interaction $`r_s`$ leads to a lower nucleon mass, see Figs. 6.13 and 6.14, but the mass of the $`\mathrm{\Delta }`$ remains unchanged since it does not contain scalar diquarks. In the standard NJL model where the axial-vector diquark does not contribute significantly to the nucleon binding , the difference between the $`\mathrm{\Delta }`$ and nucleon mass is directly related to the scalar diquark interaction. In the current model where the axial-vector diquark makes a larger contribution to the nucleon mass, therefore a detailed calculation for the delta states is needed to understand the mechanism behind the $`\mathrm{\Delta }`$-$`N`$ mass difference. In the standard NJL model this leads to a contradiction, since for an axial-vector interaction which gives a reasonable description of nucleon properties, both the axial-vector diquark and more importantly the $`\mathrm{\Delta }`$ are unbound . The crucial role of the axial-vector diquark correlation in the non-local NJL model, especially in the confining phase of the model, indicate that this model might do better.
In order to understand the implications of this model in baryonic sector fully one should investigate properties of the nucleon such as the axial vector coupling constant, the magnetic moment, etc. On the other hand, the role of quark confinement in this model can be better clarified by investigating quark and nuclear matter in this model. One of the long standing problem in four-Fermi chiral quark models is the fact that quark/nuclear matter does not saturate , mainly due to the strongly attractive quark interactions responsible for the spontaneous chiral symmetry breaking. Recently, Bentz and Thomas have shown that a sufficient strong repulsive contribution can arise if confinement effects are incorporated, albeit in the cost of introducing a new parameter into model. It was shown that such repulsive contribution can lead to saturation of nuclear matter equation of state. It is indeed of interest to investigate the stability of nuclear matter within this quark confining model. Such problems can be studied based on the Faddeev approach, see e. g., Ref. . |
warning/0507/hep-ex0507007.html | ar5iv | text | # SEARCH FOR SOLAR AXIONS: THE CAST EXPERIMENT AT CERN
## 1 Introduction
QCD is the universally accepted theory for describing the strong interactions, but it has one serious blemish: the so-called “strong CP problem”. In the following we will give a brief review of it, a more general introduction on the subject can be found in $`^{\mathrm{?},\mathrm{?}}`$.
Because of the existence of non-trivial vacuum gauge configurations, QCD has a very rich vacuum structure. All these degenerate vacuum configurations of the theory are characterized by the topological winding number n associated with them
$$n=\frac{ig^3}{24\pi ^2}d^3xTr\epsilon _{ijk}A^i(x)A^j(x)A^k(x)$$
(1)
where g is the gauge coupling, $`A^i`$ is the gauge field, and the temporal gauge ($`A^0=0`$) has been used. Then, the correct vacuum state of the theory is a superposition of all these degenerate states $`|n`$,
$$|\mathrm{\Theta }=\underset{n}{}exp(in\mathrm{\Theta })|n$$
(2)
where, a priori, the angle $`\mathrm{\Theta }`$ is an arbitrary parameter of the theory. States of different $`\mathrm{\Theta }`$ are the physically distinct vacua for the theory, each with a distinct world of physics built upon it. By appropriate means the effects of this $`\mathrm{\Theta }`$-vacuum can be recast into a single, additional non-perturbative term in the QCD Lagrangian:
$$_{QCD}=_{pert}+\overline{\mathrm{\Theta }}\frac{g^2}{32\pi ^2}G^{a\mu \nu }\stackrel{~}{G}_{a\mu \nu },\overline{\mathrm{\Theta }}=\mathrm{\Theta }+Argdet$$
(3)
where $`G^{a\mu \nu }`$ is the field strength tensor, $`\stackrel{~}{G}_{a\mu \nu }`$ is its dual, and $``$ is the quark mass matrix. This extra term in the QCD Lagrangian arises due to two separate and independent effects: the $`\mathrm{\Theta }`$ structure of the pure QCD vacuum, and electroweak effects involving the quark masses.
However, such a term in the QCD Lagrangian clearly violates CP, T and P in the case of $`\overline{\mathrm{\Theta }}0`$, yet Nature has never exhibited this in any experiment. Moreover, the value of the neutron electric dipole moment depends on $`\overline{\mathrm{\Theta }}`$, and the present experimental bound $`^\mathrm{?}`$ $`d_N<6.3\times 10^{26}\mathrm{e}.\mathrm{cm}`$ constrains $`\overline{\mathrm{\Theta }}`$ to be less than (or of the order of) $`10^{10}`$. The mystery of why the arbitrary parameter $`\overline{\mathrm{\Theta }}`$ must be so small is the strong CP problem.
Various theoretical attempts to solve this strong CP problem have been postulated $`^{\mathrm{?},\mathrm{?}}`$, being the most elegant solution the one proposed by Peccei and Quinn in 1977 $`^{\mathrm{?},\mathrm{?}}`$. Their idea was to make $`\overline{\mathrm{\Theta }}`$ a dynamical variable with a classical potential that is minimized by $`\overline{\mathrm{\Theta }}=0`$. This is accomplished by introducing an additional global, chiral symmetry, known as PQ (Peccei-Quinn) symmetry $`U(1)_{PQ}`$, which is spontaneously broken at a scale $`f_{PQ}`$. Immediately and independently, Weinberg $`^\mathrm{?}`$ and Wilczek $`^\mathrm{?}`$ realized that, because $`U(1)_{PQ}`$ is spontaneously broken, there should be a pseudo-Goldstone boson, “the axion” (or as Weinberg originally referred to it, “the higglet”). Because $`U(1)_{PQ}`$ suffers from a chiral anomaly, the axion acquires a small mass of the order of $`m_a6\mu \mathrm{eV}\left(10^{12}\mathrm{GeV}/f_{PQ}\right)`$.
A priori the mass of the axion (or equivalently the $`f_{PQ}`$ scale) is arbitrary, but it can be constraint using the data from various experiments, astrophysical considerations (cooling rates of stars) and cosmological arguments (overclosure of the Universe) $`^{\mathrm{?},\mathrm{?}}`$. Nowadays it is believed to fall inside the so-called “axion mass window”: $`10^6\mathrm{eV}<m_a<10^3\mathrm{eV}`$. The upper limit depends on the axion-nucleon interaction that it is constrained in two different ways by the observed neutrino signal of supernova (SN)1987A $`^{\mathrm{?},\mathrm{?}}`$. However, these values rely on the model-dependent axion-nucleon coupling, they involve large statistical and systematical uncertainties, and perhaps unrecognized loop-holes. Therefore, it is prudent to consider other experimental or astrophysical methods to constraint axions in this range of parameters.
The interaction strength of axions with ordinary matter (photons, electrons and hadrons) scales $`^\mathrm{?}`$ as $`1/f_{PQ}`$ and so the larger this number, the more weakly the axion couples. The present constraints on its mass make the axion a weakly interacting particle, therefore a nice candidate for the Dark Matter of the Universe $`^\mathrm{?}`$.
One generic property of the axions is a two-photon interaction of the form:
$$_{a\gamma }=\frac{1}{4}g_{a\gamma }F_{\nu \mu }\stackrel{~}{F}^{\nu \mu }a=g_{a\gamma }𝐄𝐁a$$
(4)
where $`F`$ is the electromagnetic field-strength tensor, $`\stackrel{~}{F}`$ is its dual, and $`𝐄`$ and $`𝐁`$ the electric and magnetic fields. As a consequence axions can transform into photons in external electric or magnetic fields $`^\mathrm{?}`$, an effect that may lead to measurable consequences in laboratory or astrophysical observations. For example, stars could produce these particles by transforming thermal photons in the fluctuating electromagnetic field of the stellar plasma $`^{\mathrm{?},\mathrm{?}}`$, or axions could contribute to the magnetically induced vacuum birefringence, interfering with the corresponding QED effect $`^{\mathrm{?},\mathrm{?}}`$. The PVLAS $`^\mathrm{?}`$ experiment apparently observes this effect, although an interpretation in terms of axion-like particles requires a coupling strength far larger than existing limits.
The sun would be a strong axion source and thus offers a unique opportunity to actually detect such particles by taking advantage of their back-conversion into X-rays in laboratory magnetic fields $`^\mathrm{?}`$. The expected solar axion flux at the Earth due to the Primakoff process is: $`\mathrm{\Phi }_a=g_{10}^2\mathrm{\hspace{0.17em}3.67}\times 10^{11}\mathrm{cm}^2\mathrm{s}^1\mathrm{with}g_{10}g_{a\gamma }\mathrm{\hspace{0.17em}10}^{10}\mathrm{GeV}`$, with an approximately thermal spectral distribution given by (Fig. 1):
$$\frac{d\mathrm{\Phi }_a}{dE_a}=g_{10}^2\mathrm{\hspace{0.17em}3.821}\times 10^{10}\frac{(E_a/\mathrm{keV})^3}{(e^{E_a/1.103\mathrm{keV}}1)}\mathrm{cm}^2\mathrm{s}^1\mathrm{keV}^1$$
(5)
and an average energy of 4.2 keV <sup>1</sup><sup>1</sup>1The spectrum in $`^\mathrm{?}`$ has been changed to that proposed in $`^\mathrm{?}`$, however with a modified normalization constant to match the total axion flux used here, which is predicted by a more recent solar model. The possible flux variations due to solar-model uncertainties are negligible.
Axion interactions other than the two-photon vertex would provide for additional production channels, but in the most interesting scenarios these channels are severely constrained, leaving the Primakoff effect as the dominant one $`^\mathrm{?}`$. In any case, it is conservative to use the Primakoff effect alone when deriving limits on $`g_{a\gamma }`$.
## 2 Principle of detection
A particularly intriguing application of magnetically induced axion-photon conversions is to search for solar axions using an “axion helioscope” as proposed by Sikivie $`^\mathrm{?}`$. One looks at the sun through a “magnetic telescope” and places an X-ray detector at the far end. Inside the magnetic field, the axion couples to a virtual photon, producing a real photon via the Primakoff effect: $`a+\gamma _{virtual}\gamma `$. The energy of this photon is then equal to the axion’s total energy. The expected number of these photons that reach the X-ray detector is:
$$N_\gamma =\frac{d\mathrm{\Phi }_a}{dE_a}P_{a\gamma }ST𝑑E_a$$
(6)
where $`d\mathrm{\Phi }_a/dE_a`$ is the axion flux at the Earth as given by eq.(5), $`S`$ is the magnet bore area ($`\mathrm{cm}^2`$), $`T`$ is the measurement time (s) and $`P_{a\gamma }`$ is the conversion probability of an axion into a photon. If we take some realistic numbers ($`g_{a\gamma }=10^{10}\mathrm{GeV}^1,T=100\mathrm{h}`$ and $`S=15\mathrm{cm}^2`$) this number of photons would be nearly 30 events.
The conversion probability in vacuum is given by:
$$P_{a\gamma }=\left(\frac{Bg_{a\gamma }}{2}\right)^2\mathrm{\hspace{0.17em}2}L^2\frac{1\mathrm{cos}(qL)}{(qL)^2}$$
(7)
where $`B`$ and $`L`$ are the magnetic field and its length (given in natural units), and $`q=m_a^2/2E`$ is the longitudinal momentum difference between the axion and an X-ray of energy $`E`$. The conversion process is coherent when the axion and the photon fields remain in phase over the length of the magnetic field region. The coherence condition states that $`^{\mathrm{?},\mathrm{?}}`$ $`qL=<\pi `$ so that a coherence length of 10 m in vacuum requires $`m_a0.02\mathrm{eV}`$ for a photon energy $`4.2\mathrm{keV}`$. Coherence can be restored for a solar axion rest mass up to $`1\mathrm{eV}`$ by filling the magnetic conversion region with a buffer gas $`^\mathrm{?}`$ so that the photons inside the magnet pipe acquire an effective mass whose wavelength can match that of the axion. For an appropriate gas pressure, coherence will be preserved for a narrow axion mass window. Thus, with the proper pressure settings it is possible to scan for higher axion masses.
The first implementation of the axion helioscope concept was performed at BNL $`^\mathrm{?}`$. More recently, the Tokyo axion helioscope $`^\mathrm{?}`$ with $`L=2.3\mathrm{m}`$ and $`B=3.9\mathrm{T}`$ has provided the limit $`g_{10}<6.0`$ at 95% CL for $`m_a0.03\mathrm{eV}`$ (vacuum) and $`g_{10}<6.8`$–10.9 for $`m_a0.3\mathrm{eV}`$ (using a variable-pressure buffer gas) $`^\mathrm{?}`$. Limits from crystal detectors $`^{\mathrm{?},\mathrm{?},\mathrm{?}}`$ are much less restrictive.
## 3 CAST experiment
In order to detect solar axions or to improve the existing limits on $`g_{a\gamma }`$ an axion helioscope has been built at CERN by refurbishing a decommissioned LHC test magnet $`^\mathrm{?}`$ which produces a magnetic field of $`B=9.0\mathrm{T}`$ in the interior of two parallel pipes of length $`L=9.26\mathrm{m}`$ and a cross–sectional area $`S=2\times 14.5`$ cm<sup>2</sup>. The aperture of each of the bores fully covers the potentially axion-emitting solar core ($`1/10`$th of the solar radius). The magnet is mounted on a platform with $`\pm 8^{}`$ vertical movement, allowing for observation of the sun for 1.5 h at both sunrise and sunset. The horizontal range of $`\pm 40^{}`$ encompasses nearly the full azimuthal movement of the sun throughout the year. The time the sun is not reachable is devoted to background measurements. A full cryogenic station is used to cool the superconducting magnet down to 1.8 K needed for its superconducting operation $`^\mathrm{?}`$. The hardware and software of the tracking system have been precisely calibrated, by means of geometric survey measurements, in order to orient the magnet to any given celestial coordinates. The overall CAST pointing precision is better $`^\mathrm{?}`$ than 0.01 including all sources of inaccuracy such as astronomical calculations, as well as spatial position measurements. At both ends of the magnet, three different detectors have searched for excess X-rays from axion conversion in the magnet when it was pointing to the sun. Covering both bores of one of the magnet’s ends, a conventional Time Projection Chamber (TPC) is looking for X-rays from “sunset” axions. At the other end, facing “sunrise” axions, a second smaller gaseous chamber with novel MICROMEGAS (micromesh gaseous structure – MM) $`^\mathrm{?}`$ readout is placed behind one of the magnet bores, while in the other one, a X-ray mirror telescope is used with a Charge Coupled Device $`^\mathrm{?}`$ (pn-CCD) as the focal plane detector. Both the pn-CCD and the X-ray telescope are prototypes developed for X-ray astronomy $`^\mathrm{?}`$. The X-ray mirror telescope can produce an “axion image” of the sun by focusing the photons from axion conversion to a $`6\mathrm{mm}^2`$ spot on the pn-CCD. The enhanced signal-to-background ratio substantially improves the sensitivity of the experiment.
### 3.1 First phase of CAST
During the years 2003 and 2004 the CAST experiment has gone through the so-called first phase, where the data has been taken with vacuum inside the magnetic field area, so that we were sensitive to axion masses up to $`m_a0.02\mathrm{eV}`$ as explained in section 2.
### 3.2 Second phase of CAST
In order to extend the range of axion masses to which we are sensitive, the magnet pipes will be filled with Helium gas in phase II. As explained in section 2, a gas with a given pressure will provide a refractive photon mass so that the coherence of the photon and axion fields will be restored for a certain range of axion masses. The second phase of the experiment is very challenging because, for the first time, a laboratory experiment will search for axions in the theoretically motivated range of axion parameters (see Fig. 2).
Data taking for this second phase it is scheduled to begin at the end of 2005, with low pressure $`{}_{}{}^{4}\mathrm{He}`$ gas inside the pipes at 1.8 K, the magnet’s operating temperature. There is a limit in the pressure that we can reach with $`{}_{}{}^{4}\mathrm{He}`$ before it liquefies, so in order to be able to extend the mass axion searches up to $`0.82\mathrm{eV}`$ we will have to switch to $`{}_{}{}^{3}\mathrm{He}`$, which has a higher vapor pressure. These steps are scheduled to occur during 2006 and 2007.
Beyond these plans CAST could search for axions with still higher masses up to $`1.4\mathrm{eV}`$ with the actual set-up, by installing thermally isolated gas cells inside the magnet bores. This would allow us to work at higher temperatures ($``$ 5.4 K) so that we could reach higher pressures and densities of the $`{}_{}{}^{4}\mathrm{He}`$ buffer gas.
## 4 Data analysis and first results
### 4.1 2003 data tacking
CAST operated for about 6 months from May to November in 2003, during most of which time at least one detector was taking data. The results $`^\mathrm{?}`$ presented in this paper were obtained after the analysis of the data sets listed in Table 1.
An independent analysis was performed for each data set. Finally, the results from all data sets are combined.
An important feature of the CAST data treatment is that the detector backgrounds are measured with $``$10 times longer exposure during the non-alignment periods. The use of these data to estimate and subtract the true experimental background during sun tracking data is the most sensitive step in the CAST analysis. To assure the absence of systematic effects, the main strategy of CAST is the use of three independent detectors with complementary approaches. In the event of a positive signal, it should appear consistently in each of the three detectors when it is pointing at the sun. In addition, an exhaustive recording of experimental parameters was done, and a search for possible background dependencies on these parameters was performed. A dependence of the TPC background on the magnet position was found, caused by its relatively large spatial movements at the far end of the magnet, which resulted in appreciably different environmental radioactivity levels. Within statistics, no such effect was observed for the sunrise detectors which undergo a much more restricted movement. To correct for this systematic effect in the TPC data analysis, an effective background spectrum is constructed only from the background data taken in magnet positions where sun tracking has been performed and this is weighted accordingly with the relative exposure of the tracking data. Further checks have been performed in order to exclude any possible systematic effect. They were based on rebinning the data, varying the fitting window, splitting the data into subsets and verifying the null hypothesis test in energy windows or areas of the detectors where no signal is expected. In general, the systematic uncertainties are estimated to have an effect of less than $``$10% of the final upper limits obtained.
For a fixed $`m_a`$, the theoretically expected spectrum of axion-induced photons has been calculated and multiplied by the detector efficiency curves of the detectors, including all hardware and software efficiency losses, such as window transmissions (for TPC and MM), X-ray mirror reflectivity (for pn-CCD), detection efficiency and dead time effects. These spectra, which are proportional to $`g_{a\gamma }^4`$, are directly used as fit functions to the experimental subtracted spectra (tracking minus background) for the TPC and MM. For these data, the fitting is performed by standard $`\chi ^2`$ minimization. Regarding the pn-CCD data, the analysis is restricted to the small area on the pn-CCD where the axion signal is expected after the focusing of the X-ray telescope. During the data taking period of 2003 a continuous monitoring of the pointing stability of the X-ray telescope was not yet possible, therefore a signal area larger than the size of the sun spot had to be considered. Taking into account all uncertainties of the telescope alignment, the size of the area containing the signal was conservatively estimated to be $`34\times 71`$pixels ($`54.3\mathrm{mm}^2`$). As in the other detectors, the background is defined by the data taken from the same area during the non-tracking periods, but, in addition, the background in the signal area was also determined by extrapolating the background measured during tracking periods in the part of the pn-CCD not containing the sun spot. Both methods of background selection led to the same final upper limit on the coupling constant $`g_{a\gamma }`$. The resulting low counting statistics in the pn-CCD required the use of a likelihood function in the minimization procedure, rather than a $`\chi ^2`$-analysis. The best fit values of $`g_{a\gamma }^4`$ obtained for each of the data sets are shown in Table 1, together with their 1$`\sigma `$ error and the corresponding $`\chi _{\mathrm{min}}^2`$ values and degrees of freedom.
Each of the data sets is individually compatible with the absence of any signal as can be seen from the $`\chi _{\mathrm{null}}^2`$ values shown in Table 1. The excluded value of $`g_{a\gamma }^4`$ was conservatively calculated by taking the limit encompassing 95% of the physically allowed part (i.e. positive signals) of the Bayesian probability distribution with a flat prior in $`g_{a\gamma }^4`$. The described procedures were done using $`g_{a\gamma }^4`$ instead of $`g_{a\gamma }`$ as the minimization and integration parameter because the signal strength (i.e. number of counts) is proportional to $`g_{a\gamma }^4`$. The 95% CL limits on $`g_{a\gamma }`$ for each of the data sets are shown in the last column of Table 1. They can be statistically combined by multiplying the Bayesian probability functions and repeating the previous process to find the combined result for the 2003 CAST data:
$$g_{a\gamma }<1.16\times 10^{10}\mathrm{GeV}^1(95\%\mathrm{CL}).$$
(8)
Thus far, our analysis was limited to the mass range $`m_a0.02`$ eV where the expected signal is mass-independent because the axion-photon oscillation length far exceeds the length of the magnet. For higher $`m_a`$ the overall signal strength diminishes rapidly and the spectral shape differs. Our procedure was repeated for different values of $`m_a`$ to obtain the entire 95% CL exclusion line shown in Fig. 2.
### 4.2 2004 data tacking
The data taken from 2004 have not yet been fully analyzed. However, the stable operation of the experiment allowed the CAST collaboration to take enough high-quality data to anticipate that the final sensitivity will be close to the value presented in the CAST proposal (see Fig. 2)
## 5 Summary
The origin of the axion as a particle that solves the strong CP problem has been reviewed. Some properties of this pseudoescalar particle have been pointed out, among them the fact that it can transform into a photon in external electric or magnetic fields, this being the only property of the axion on which CAST relies. The CAST experiment and its first results $`^\mathrm{?}`$have been presented. Our limit improves the best previous laboratory constraints $`^\mathrm{?}`$ on $`g_{a\gamma }`$ by a factor 5 in our coherence region $`m_a0.02`$ eV. A higher sensitivity is expected from the 2004 data with improved conditions in all detectors, which should allow us to surpass the astrophysical limit. In addition, starting in 2005, CAST plans to take data with a varying-pressure buffer gas in the magnet pipes, in order to restore coherence for axion masses above 0.02 eV. The extended sensitivity to higher axion masses will allow us to enter into the region shown in Fig. 2 which is especially motivated by axion models $`^\mathrm{?}`$.
## References |
warning/0507/math0507295.html | ar5iv | text | # Schrödinger operator with a junction of two 1-dimensional periodic potentials
## 1 Introduction
We consider the Schrödinger operator $`T_ty=y^{\prime \prime }+q_t(x)y`$ acting on $`L^2()`$, where the real potential $`q_t`$ is given by
$$q_t(x)=\{\begin{array}{cc}p_1(x)\hfill & \text{if }x<0,p_1L^1(\tau 𝕋)\hfill \\ p(x+t)\hfill & \text{if }x>0,pL^1(𝕋)\hfill \end{array},t[0,1],$$
(1.1)
where $`\tau >0`$ is the period of $`p_1`$ and $`𝕋=/`$. We call such a potential $`q_t`$ biperiodic. If $`p_1=p`$ and $`t=0,`$ then we obtain the well known periodic case, that is the Hill operator $`H=\frac{d^2}{dx^2}+p`$ in $`L^2()`$. It is well known (see \[T\]) that the spectrum of $`H`$ is absolutely continuous and consists of intervals $`\sigma _n=[\alpha _{n1}^+,\alpha _n^{}],`$ where $`\alpha _{n1}^+<\alpha _n^{}\alpha _n^+,n1`$. These intervals are separated by gaps $`\gamma _n(H)=(\alpha _n^{},\alpha _n^+),n1`$. We set $`\gamma _0(H)=(\mathrm{},\alpha _0^+)`$ and $`\gamma (H)=\gamma _n(H)`$. If a gap degenerates, i.e. $`\gamma _n=\mathrm{},n1`$, then the corresponding segments $`\sigma _n,\sigma _{n+1}`$ merge. Define the Hill operator $`H(t)=\frac{d^2}{dx^2}+p(x+t)`$ in $`L^2()`$. It is clear that $`\sigma (H)=\sigma (H(t))`$ for any $`t[0,1]`$. Introduce the Hill operator $`H_m=\frac{d^2}{dx^2}+p_m(x)`$ in $`L^2(),m=1,2`$ where here and below $`p_2p`$. In the biperiodic case (see Theorem 2.1) we prove that the spectrum of each $`T_t,t`$ has the following form
$$\sigma (T_t)=\sigma _{ac}(T_t)\sigma _d(T_t),\sigma _{ac}(T_t)=\sigma (H_1)\sigma (H_2),\sigma _d(T_t)\gamma (H_1)\gamma (H_2).$$
(1.2)
Note that (1.2) implies $`\sigma _{sc}(T_t)=\mathrm{}`$ and there are no embedded eigenvalues. The absolutely continuous spectrum $`\sigma _{ac}(T_t)`$ consists of intervals $`\sigma _n(T_t),n1`$. These intervals are separated by the gaps $`\gamma _n(T_t),n1`$. In general, there exist eigenvalues in these gaps and we have (see Theorem 2.1)
$$\mathrm{\#}(T_t,\gamma _n(T_t))2,\mathrm{for}\mathrm{each}n0,$$
(1.3)
where $`\mathrm{\#}(T_t,\omega )`$ is the number of eigenvalues of $`T_t`$ on an interval $`\omega `$. The basic goal of this paper is to study eigenvalues in these gaps.
In this paper we study in more detail two cases of biperiodic potentials. Firstly, we consider the case of dislocation, that is the operator $`T_t^{di}=\frac{d^2}{dx^2}+p_{(t)}(x)`$ in $`L^2(),`$ with the potential $`p_{(t)}(x)=\chi _{}(x)p(x)+\chi _+(x)p(x+t),t`$, here and below $`\chi _\pm (x)=1,\pm x0`$ and $`\chi _\pm (x)=0,\pm x<0`$ and the potential $`p`$ as above. Changing $`t`$ we get different potentials $`p_{(t)}()`$. It can be shown that the eigenvalues of $`T_t^{di}`$ are periodic in $`t`$ (see (1.4)). Actually, they are 2-periodic, not 1-periodic. If $`t=0`$, then $`p_{(0)}(x)=p(x)`$ is periodic and $`T_0^{di}=H`$ and eigenvalues are absent. But if $`t0`$ what are we able to say about the eigenvalues in the gaps $`\gamma _n,n0,`$ of the operator $`T_t^{di}`$?
Secondly, we consider the half-solid, i. e. the operator $`T_t^sy=y^{\prime \prime }+q_t^s(x)y`$ in $`L^2()`$, where $`q_t^s(x)=s\chi _{}(x)+\chi _+(x)p(x+t)`$ and $`t,s`$. Changing constants $`t,s`$ we get different potentials $`q_t^s(x)`$. The eigenvalues of $`T_t^s`$ depend on $`t,s`$. If the first gap of $`H`$ is open and $`s>\alpha _1^{}`$, then there exists a gap in the spectrum of $`T_t^s`$. What are we able to say on the eigenvalues in these gaps $`\gamma _n(T_t^s),n0`$?
We describe the main results which are proved in the present paper:
i) relations (1.2-3),
ii) the description of the possible coexistence of eigenvalues and resonances for different classes of potentials,
iii) we prove that for any fixed numbers $`m(m=0,1,2)`$ and $`N1`$ there exists a biperiodic potential such that $`\mathrm{\#}(T_0,\gamma _n(T_0))=m`$ for all $`1nN`$,
iv) the discrete spectrum of $`T_t,T_t^{di},T_t^s`$ is studied. For example, if $`t`$ is small then there exist eigenvalues of $`T_t^d`$ in the gaps $`\gamma _n,n0`$, and their asymptotics are determined as $`t0`$.
v) The analytic continuation of $`(T_0\lambda )^1,Im\lambda >0`$, into the second sheet of the energy surface is obtained and the resonances are studied. For the dislocation, $`T_t^{di}=T_t`$, we have: $`\lambda _1`$ is an eigenvalue of $`T_t`$, iff the number $`\lambda _1`$ considered as a point on the second sheet is a resonance of the operator $`\stackrel{~}{T}_t`$ given by $`\stackrel{~}{T}_ty=y^{\prime \prime }+\stackrel{~}{q}_t(x)y`$, where $`\stackrel{~}{q}_t(x)=\chi _{}(x)p_2(x+t)+\chi _+(x)p_1(x)`$.
We briefly describe the proof. Using results of self-adjoint extensions \[Kr\] we obtain (1.2-3). An eigenvalue of $`T_t`$ is a zero of the corresponding Wronskian. Using techniques from inverse spectral theory \[L\], \[Tr\], \[K2\] we study how such a zero depends on $`t.`$ Then we prove (1.3) with $`\mathrm{\#}(T,\gamma _n(T))=m,m=0,1,`$ for all $`n1.`$ In order to obtain the case m=2 we use additional results from \[GT\] (see also \[KK1\]) on the inverse problem for the Hill operator. In order to determine the asymptotics of the eigenvalues as $`t0`$ we study the corresponding Wronskian. Here we use the implicit function Theorem, various properties of the fundamental solutions and the quasimomentum.
Tamm \[Ta\] was the first, who considered the Schrödinger operator with biperiodic potentials (for the Kronig-Penny model) and proved the existence of eigenvalues (the famous surface states). The spectral properties of the Schrödinger operator with biperiodic potentials were studied in various papers (see e.g. \[A\] \[A1\], \[DS\]). The spectral problem with biperiodic potentials arises in non-linear equations. For example, Bikbaev and Sharipov \[BS\] considered the KdV equation with biperiodic initial data. In this paper \[BS\] it is assumed that eigenvalues are absent. Anoshchenko \[A\] studied the inverse problem for the Schrödinger operator with a biperiodic potential plus a decreasing one. But the author does not study the discrete spectrum of the Schrödinger operator with a biperiodic potential. The author \[K3\] obtained the following result concerning the motion of eigenvalues for the case of the dislocation: in each gap $`\gamma _n\mathrm{},n1`$ there exist two unique ”states” (an eigenvalue and a resonance) $`\lambda _n^\pm (t)`$ of the dislocation operator, such that $`\lambda _n^\pm (0)=\alpha _n^\pm `$ and the point $`\lambda _n^\pm (t)`$ runs clockwise around the gap $`\gamma _n`$ changing the energy sheet whenever it hits $`\alpha _n^\pm `$, making $`n/2`$ complete revolutions in unit time. On the first sheet $`\lambda _n^\pm (t)`$ is an eigenvalue and on the second sheet $`\lambda _n^\pm (t)`$ is a resonance. Moreover, the following identities are fulfilled:
$$\lambda _{2n}^\pm (t+1)=\lambda _{2n}^\pm (t),and\lambda _{2n+1}^\pm (t+1)=\lambda _{2n+1}^\pm (t)+\alpha _{2n+1}^{}+\alpha _{2n+1}^+,anyn0,t.$$
(1.4)
We think that the results of the present paper are needed to study the KdV equation, the inverse problem and the spectral properties of $`T_t+V(x)`$, where $`V(x)0,`$ as $`x\pm \mathrm{}`$ or $`V=\epsilon x`$ ( Stark effect), and junctions of two d-dimensional periodic potentials, $`d>1`$.
## 2 Main results
We recall some properties of the Hill operator $`H=\frac{d^2}{dx^2}+p`$ acting in $`L^2()`$, where the potential $`pL^1(𝕋)`$ is real. The spectrum of $`H`$ is absolutely continuous and consists of intervals $`\sigma _n=[\alpha _{n1}^+,\alpha _n^{}],`$ where $`\alpha _{n1}^+<\alpha _n^{}\alpha _n^+,n1`$. We set $`\alpha _0^+=0.`$ These intervals are separated by gaps $`\gamma _n(H)=(\alpha _n^{},\alpha _n^+),n1`$. Let $`\phi (x,z),\vartheta (x,z)`$ be the solutions of the equation
$$y^{\prime \prime }+py=\lambda y,\lambda ,$$
(2.1)
satisfying $`\phi _x(0,\lambda )=\vartheta (0,\lambda )=1,`$ and $`\phi (0,\lambda )=\vartheta _x(0,\lambda )=0`$. We define the Lyapunov function $`\mathrm{\Delta }(\lambda )=\frac{1}{2}(\phi _x(1,\lambda )+\vartheta (1,\lambda ))`$ and the function $`a(\lambda )=\frac{1}{2}(\phi _x(1,\lambda )\vartheta (1,\lambda )).`$ Note that $`\mathrm{\Delta }(\alpha _n^\pm )=(1)^n,n1`$. The sequence $`\alpha _0^+<\alpha _1^{}\alpha _1^+<\mathrm{}`$ is the spectrum of Eq. (2.1) with 2-periodic boundary conditions, that is $`y(x+2)=y(x),x`$. Let $`\mathrm{\Psi }(x,\alpha _n^\pm )`$ be the corresponding real normalized eigenfunctions, i.e. $`_0^1\mathrm{\Psi }(x,\alpha _n^\pm )^2𝑑x=1`$. If $`\alpha _n^{}=\alpha _n^+`$, then $`\alpha _n^{}`$ is a double eigenvalue. The lowest eigenvalue $`\alpha _0^+`$ is simple, $`\mathrm{\Delta }(\alpha _0^+)=1,`$ and the corresponding eigenfunction has period 1. The eigenfunctions corresponding to $`\alpha _n^\pm `$ have period 1 when $`n`$ is even and they are anti-periodic, $`\mathrm{\Psi }(x+1,\alpha _n^\pm )=\mathrm{\Psi }(x,\alpha _n^\pm ),x`$, when $`n`$ is odd. Let $`\mu _n(p),n1,`$ be the Dirichlet spectrum of (2.1) with the boundary condition $`y(0)=y(1)=0`$. Let $`\nu _n(p),n0,`$ be the Neumann spectrum of (2.1) with the boundary condition $`y^{}(0)=y^{}(1)=0.`$ It is well known that $`\mu _n,\nu _n[\alpha _n^{},\alpha _n^+]`$ and $`\nu _0\alpha _0^+`$. The energy Riemann surface $`\mathrm{\Lambda }_E(H)`$ for the Hill operator consists of 2 sheets $`\mathrm{\Lambda }^{(1)}=\sigma _n`$ and $`\mathrm{\Lambda }^{(2)}`$. Each sheet is a copy of the complex plane slit along $`\sigma (H)`$. The first sheet $`\mathrm{\Lambda }^{(1)}`$ is glued to the sheet $`\mathrm{\Lambda }^{(2)}`$ by identifying ”crosswise” all slits in $`\sigma (H)`$ (if $`p0`$, then we have the Riemann surface of $`\sqrt{\lambda }`$). We define the quasimomentum $`k(\lambda )=\mathrm{arccos}\mathrm{\Delta }(\lambda ),\lambda _+`$, which is fixed by $`k(0)=0`$. The function $`\mathrm{sin}k(\lambda )`$ is analytic on the Riemann surface $`\mathrm{\Lambda }_E(H)`$ (see \[M\]). The function $`k(\lambda )`$ has an analytic extension from $`_+`$ into the Riemann surface $`\mathrm{\Lambda }_E(H)`$ without slits $`\gamma _n(H),n1`$. Recall that if a gap $`|\gamma _n|=0,`$ then $`k(\lambda )`$ is analytic at $`\lambda =\alpha _n^\pm `$; if $`|\gamma _n|0,`$ then $`k(\lambda )`$ has branch points $`\alpha _n^\pm `$ and
$$k(\lambda )=\pi n+i\sqrt{2M_n^\pm (\lambda \alpha _n^\pm )}(1+\mathrm{o}(1)),\mathrm{as}\lambda \alpha _n^\pm ,\lambda \gamma _n(H),$$
where $`\pm M_n^\pm >0`$ is the effective mass; here and below $`\sqrt{z}>0,z>0`$ (see \[KK2\]). We introduce the Bloch functions $`\psi _\pm (x,\lambda )`$ and the Weyl functions $`m^\pm (\lambda )`$ by
$$\psi _\pm (x,\lambda )=\vartheta (x,\lambda )+m_\pm (\lambda )\phi (x,\lambda ),m^\pm (\lambda )=\frac{a(\lambda )\pm i\mathrm{sin}k(\lambda )}{\phi (1,\lambda )}.$$
We define resonances of $`T_t`$. There are different kinds of the resonances. Due to (1.2) $`\sigma _{ac}(T_t)=\sigma _{ac}(T_0)`$ does not depend on $`t`$. It has the decomposition
$$\sigma _{ac}(T_t)=_2^4\sigma ^{(n)},\sigma ^{(2)}=\sigma (H_1)\sigma (H),\sigma ^{(3)}=\sigma (H_1)\sigma (H),\sigma ^{(4)}=\sigma (H)\sigma (H_1).$$
(2.2)
In order to describe the resonances we need the energy Riemann surface $`\mathrm{\Lambda }_E(T_0)`$ for the operator $`T_0`$. There are 2 cases. In the first case $`\sigma _{ac}(T_0)\sigma ^{(2)}`$, the Riemann surface $`\mathrm{\Lambda }_E(T_0)`$ consists of 4 sheets $`\mathrm{\Lambda }_0^{(1)},\mathrm{\Lambda }_0^{(2)},\mathrm{\Lambda }_0^{(3)},\mathrm{\Lambda }_0^{(4)}`$. Each sheet is a copy of the complex plane slit along $`\sigma _{ac}(T_0)`$. The first sheet $`\mathrm{\Lambda }_0^{(1)}`$ is glued to the sheet $`\mathrm{\Lambda }_0^{(n)}`$ by identifying the sides of all slits in $`\sigma ^{(n)},n=2,3,4`$. Similarly, the second sheet $`\mathrm{\Lambda }_0^{(2)}`$ is glued to the sheets $`\mathrm{\Lambda }_0^{(3)}`$ and $`\mathrm{\Lambda }_0^{(4)}`$ by identifying ”crosswise” the sides of all slits in $`\sigma ^{(4)}`$ and $`\sigma ^{(3)}`$ respectively. The third sheet $`\mathrm{\Lambda }_0^{(3)}`$ is glued to the sheet $`\mathrm{\Lambda }_0^{(4)}`$ along slits in $`\sigma ^{(2)}`$. For each complex number $`\lambda \mathrm{\Lambda }_0^{(1)}`$ the number $`\lambda ^{(n)},n=1,2,3,4`$ will denote the corresponding point on the sheet $`\mathrm{\Lambda }_0^{(n)}`$. In the second case $`\sigma _{ac}(T_0)=\sigma ^{(2)}`$ the energy Riemann surface $`\mathrm{\Lambda }_E(T_0)`$ consists of 2 sheets $`\mathrm{\Lambda }_0^{(1)},\mathrm{\Lambda }_0^{(2)}`$ and coincides with the Riemann surface $`\mathrm{\Lambda }_E(H)`$ for the Hill operator $`H`$.
Let $``$ denote the class of bounded operators in $`L^2()`$. For each $`\eta C_0^{\mathrm{}}()`$ we introduce the operator-valued function $`A:\mathrm{\Lambda }_0^{(1)}`$ by $`A(\lambda )=\eta R(\lambda )\eta `$, where $`R(\lambda )=(T_t\lambda )^1`$. Assume that $`A(\lambda ),\lambda \mathrm{\Lambda }_0^{(1)}`$, has a meromorphic continuation across the set $`\sigma ^{(m)}`$ on ”the sheet” $`\mathrm{\Lambda }_0^{(m)},m=2,3,4.`$ Suppose that $`A(\lambda )`$ has a pole $`\lambda _r\mathrm{\Lambda }_0^{(m)}`$ (this does not depend on the choice of $`\eta `$). We call $`\lambda _r`$ a resonance of $`T_t`$. Let $`\mathrm{\#}^{(m)}(T_t,\omega )`$ be the number of resonances of $`T_t`$ in an interval $`\omega \mathrm{\Lambda }_0^{(m)},m=2,3,4`$. We formulate our first result.
Theorem 2.1. Let $`q_t,t`$, be a biperiodic potential in the sense of (1.1). Then
i) Relations (1.2-3) are fulfilled.
ii) For each $`\eta C_0^{\mathrm{}}()`$ the operator-valued function $`A:\mathrm{\Lambda }_0^{(1)}`$ has a meromorphic continuation into the Riemann surface $`\mathrm{\Lambda }_E(T_0)`$ described above. The number $`\lambda _e\mathrm{\Lambda }_0^{(1)}`$ is an eigenvalue of the operator $`T_t`$ iff the same number considered as an element $`\lambda _e^{(2)}\mathrm{\Lambda }_0^{(2)}`$ on the second sheet is a resonance of $`\stackrel{~}{T}_t`$ given by $`\stackrel{~}{T}_ty=y^{\prime \prime }+\stackrel{~}{q}_t(x)y`$, where $`\stackrel{~}{q}_t=\chi _{}p_2(+t)+\chi _+p_1`$.
iii) The Riemann surface $`\mathrm{\Lambda }_E(T_t^{di})`$ for the dislocation operator coincides with the Riemann surface $`\mathrm{\Lambda }_E(H)`$ for the Hill operator. If $`ps`$, then the Riemann surface $`\mathrm{\Lambda }_E(T_t^s)`$ for the half-solid case consists of 4 sheets.
Remark. 1) To say anything about all complex resonances on the level of generality of Theorem 2.1 is highly non-trivial. It is not known, for precisely which subclass of potentials there exist non real resonances. Theorem 2.1, however, gives control on all resonances on the second sheet $`\mathrm{\Lambda }_0^{(2)}`$ (but not on the third and fourth), since they are eigenvalues of the self-adjoint operator $`\stackrel{~}{T}_t`$. In particular, they are real and belong to $`\gamma _n(T_t)\mathrm{\Lambda }_0^{(2)}`$. It is this result which makes the operator $`\stackrel{~}{T}_t`$ important in Theorem 2.1 (and later on in Lemma 4.1). 2) In the case of the dislocation operator ($`p_1=p_2`$) the energy Riemann surface $`\mathrm{\Lambda }_E(T_t^d)`$ consists of only 2 sheets $`\mathrm{\Lambda }_0^{(1)},\mathrm{\Lambda }_0^{(2)}`$. Thus, by Theorem 2.1, all resonances are real in this case. 3) In the case of the half-solid ($`p_1=const`$) there are (except for $`p_1=p_2`$) 4 sheets and it can be shown that in this case there exist complex resonances \[KP\].
Introduce the subspaces of even potentials $`L_{even}^r(𝕋)L^r(𝕋)`$ given by
$$L_{even}^r(𝕋)=\{pL^r(𝕋):p(x)=p(1x),0<x<1\},r1.$$
It is well known that a potential $`pL_{even}^1(𝕋)`$ iff $`|\gamma _n(H)|=|\mu _n(p)\nu _n(p)|`$ for all $`n1`$. Next we consider a biperiodic potential with even potentials $`p_1,p_2`$.
Theorem 2.2. Let $`q_0`$ be a biperiodic potential in the sense of (1.1), where $`p_1,pL_{even}^1(𝕋)`$ and $`t=0`$. Assume that some gap of $`T_0`$ is given by $`\stackrel{~}{\gamma }(T_0)=\stackrel{~}{\gamma }_1\stackrel{~}{\gamma }_2\mathrm{}`$ for some gap $`\stackrel{~}{\gamma }_n`$ of $`H_n,n=1,2`$. In the case $`|\stackrel{~}{\gamma }_n|=|\stackrel{~}{\mu }_n\stackrel{~}{\nu }_n|<\mathrm{},n=1,2`$ we denote by $`\stackrel{~}{\mu }_n,\stackrel{~}{\nu }_n\overline{\stackrel{~}{\gamma }_n}`$ the corresponding Dirichlet and Neumann eigenvalues. Then
$$\mathrm{\#}(T_0,\stackrel{~}{\gamma }(T_0))=\{\begin{array}{cc}0\hfill & \text{if }(\stackrel{~}{\mu }_1\stackrel{~}{\nu }_1)(\stackrel{~}{\mu }_2\nu _2)>0,\hfill \\ 1\hfill & \text{if }(\stackrel{~}{\mu }_1\stackrel{~}{\nu }_1)(\stackrel{~}{\mu }_2\stackrel{~}{\nu }_2)<0.\hfill \end{array},\mathrm{if}|\stackrel{~}{\gamma }_1|,|\stackrel{~}{\gamma }_2|<\mathrm{},$$
(2.3)
$$\mathrm{\#}(T_0,\stackrel{~}{\gamma }(T_0))=\{\begin{array}{cc}0,\hfill & \text{if }\stackrel{~}{\mu }_2<\stackrel{~}{\nu }_2,\hfill \\ 1,\hfill & \text{if }\stackrel{~}{\nu }_2<\stackrel{~}{\mu }_20.\hfill \end{array},\mathrm{if}|\stackrel{~}{\gamma }_2|<\mathrm{}=|\stackrel{~}{\gamma }_1|.$$
(2.4)
Moreover, if $`\stackrel{~}{\gamma }(T_0)=(\mathrm{},0)`$, then $`\mathrm{\#}(T_0,\stackrel{~}{\gamma }(T_0))=0`$.
We now consider the problem of the reconstruction of $`p_1`$: if we know $`p`$ plus some spectral data. In fact we consider the inverse problem including the problem of characterization for $`T_0=\frac{d^2}{dx^2}+q_0`$. We introduce the sequence $`d=\{d_n\}_1^{\mathrm{}}`$, where $`d_n\{0,1\}=0`$. Define the spaces
$$Q=\{q_0=\chi _{}p_1+\chi _+p:p_1,pL_{even}^2(𝕋),\gamma _n(H_1)=\gamma _n(H),\mathrm{for}\mathrm{all}n0\},$$
$$P=\{(p,d):pL_{even}^2(𝕋),d=\{d_n\}_1^{\mathrm{}},d_n\{0,1\},d_n=0\mathrm{if}|\gamma _n(H)|=0\}.$$
Define the mapping $`\rho :QP`$ given by $`\rho (q_0)=(p,d)`$, where $`d_n=\mathrm{\#}(T_0,\gamma _n(T_0)),n1`$. Note that (2.3) yields $`d_n\{0,1\},n1`$ and (1.2) implies $`d_n=0`$ if $`|\gamma _n(H)|=0`$. We shall show that any sequence $`d=\{d_n\}_1^{\mathrm{}}`$ of the number of eigenvalues in the gaps actually occurs. More precisely, we have the following inverse result:
Theorem 2.3. The mapping $`\rho :QP`$ is 1-to-1 and onto. Moreover, for $`p^2_𝕋|p(x)|^2𝑑x`$ the following estimates are fulfilled:
$$p_1=p2G(1+G^{1/3}),G2p(1+p^{1/3}),G=(|\gamma _n(T_0)|^2)^{1/2}.$$
(2.5)
Remark. i) It follows from assertion of Theorem, that $`p_1`$ in the decomposition $`q_0=\chi _{}p_1+\chi _+p`$ is uniquely determined by $`(p,d)`$. ii) Assume that $`q_0Q`$ and $`|\gamma _n(H)|>0,n1`$ for some $`pL_{even}^2(𝕋)`$. Then the definition of $`Q`$ and (1.2) yield $`\gamma _n(T_0)=\gamma _n(H)`$ for all $`n1`$. If $`d_n=1`$ for all $`n1`$, then Theorem 2.3 yields $`\mathrm{\#}(T_0,\gamma _n(T_0))=1,n1,`$ and $`\mathrm{\#}(T_0,)=\mathrm{}`$. If $`d_n=0,n1`$, then Theorem 2.3 yields $`\mathrm{\#}(T_0,\gamma _n(T_0))=0,n1,`$ and $`T_0=H`$.
We now consider the dislocation operator
$$T_t^{di}=\frac{d^2}{dx^2}+p_{(t)}(x)\mathrm{in}L^2(),p_{(t)}=\chi _{}p+\chi _+p(+t),pL^1(𝕋),t.$$
Using (1.2-3) we obtain the following relations
$$\sigma (T_t^{di})=\sigma _{ac}(T_t^{di})\sigma _d(T_t^{di}),\sigma _{ac}(T_t^{di})=\sigma (H),\sigma _d(T_t^{di})\gamma _n(H),\mathrm{\#}(T_t^{di},\gamma _n(T_t^{di}))2.$$
(2.6)
We emphasize that $`\gamma _n(T_t^{di})=\gamma _n(H)`$ for any $`t[0,1],n0`$, i.e., $`\gamma _n(T_t^{di})`$ are independent of $`t`$. This follows from Theorem 2.1.i. Then $`\sigma _{ac}(T_t^{di})=\sigma ^{(2)},`$ and $`\sigma ^{(3)}=\sigma ^{(4)}=\mathrm{}`$, and the Riemann surface $`\mathrm{\Lambda }_E(T_t^{di})`$ coincides with the Riemann surface $`\mathrm{\Lambda }_E(H)`$, and consists of 2 sheets $`\mathrm{\Lambda }_0^{(1)},\mathrm{\Lambda }_0^{(2)}`$. Introduce the function (here and below $`\dot{u}=\frac{}{t}u`$)
$$L(t,\lambda )=[\dot{\mathrm{\Psi }}(t,\lambda )^2(p(t)\lambda )\mathrm{\Psi }(t,\lambda )^2],\lambda =\alpha _n^\pm ,n0,t[0,1].$$
(2.7)
Let $`W_r^2(a,b)`$ be the Sobolev space of functions $`f`$ on the interval such that $`f^{(r)}L^2(a,b),r0`$. We have the following result about eigenvalues of the dislocation operator.
Theorem 2.4. Let the dislocation potential $`p_{(t)}=\chi _{}p+\chi _+p(+t)`$, where $`pL^2(𝕋),t`$. Then for each gap $`\gamma _n(H)=(\alpha _n^{},\alpha _n^+)\mathrm{},n1`$ there exists a function $`z_n^\pm ()W_1^2(\epsilon _n,\epsilon _n),`$ for some $`\epsilon _n>0`$, such that: $`z_n^\pm (0)=0`$ and $`\lambda _n^\pm (t)\alpha _n^\pm z_n^\pm (t)^2\gamma _n(H)=\gamma _n(T_t^{di})`$ satisfies
i) If $`z_n^\pm (t)>0`$ for some $`t(\epsilon _n,\epsilon _n)`$, then $`\lambda _n^\pm (t)\gamma _n(H)`$ is an eigenvalue of $`T_t^{di}`$.
If $`z_n^\pm (t)<0`$ for some $`t(\epsilon _n,\epsilon _n)`$, then $`\lambda _n^\pm (t)\gamma _n(H)\mathrm{\Lambda }_0^{(2)}`$ is a resonance of $`T_t^{di}`$.
Moreover, the following asymptotics are fulfilled:
$$z_n^\pm (t)=\sqrt{|M_n^\pm |/2}_0^tL(s,\alpha _n^\pm )𝑑s+O(t^{3/2}),t0,$$
(2.8)
$$z_n^\pm (t)=t\sqrt{|M_n^\pm /2|}\dot{\mathrm{\Psi }}(0,\alpha _n^\pm )^2+O(t^2),t0,\mathrm{if}\mu _n(p)=\alpha _n^\pm .$$
(2.9)
ii) If, in addition, $`p`$ is real analytic, then $`z_n^\pm ()`$ is real analytic on $`(\epsilon _n,\epsilon _n)`$.
Remark. i) If in (2.9) we have $`\mu _n(p)=\alpha _n^{},`$ then $`z_n^{}(t)>0`$ for $`t>0`$ and $`\lambda _n^{}(t)`$ is an eigenvalue; if $`\mu _n(p)=\alpha _n^+,`$ then $`z_n^{}(t)<0`$ for $`t>0`$ and $`\lambda _n^{}(t)`$ is a resonance. ii) Let $`p`$ be smooth potential and $`n1`$. Then using (2.8) we deduce that $`z_n^{}(t)>0,`$ as $`t>0`$ (we get an eigenvalue) and $`z_n^+(t)<0`$ (we get a resonance). Hence for fixed $`n1`$ and any small $`t>0`$ we have an eigenvalue near $`\alpha _n^{}`$ and a resonance near $`\alpha _n^+`$ on the second sheet. iii) It is possible to formulate Theorem 2.4 for potentials $`pL^1(𝕋)`$ and then the asymptotics in (2.8) has the form $`z_n^\pm (t)=\sqrt{|M_n^\pm /2|}_0^tL(s,\alpha _n^\pm )𝑑s+O(t),`$ as $`t0`$
Using Theorem 2.4 we construct the dislocation operator $`T_t^{di}`$, such that $`\mathrm{\#}(T_t^{di},\gamma _n(T_t^{di}))=2`$ for all $`n=1,..,N`$ for each fixed $`N1`$.
Theorem 2.5. For any finite sequences $`d=\{d_n\}_1^N,\{s_n\}_1^N,d_n\{0,1\},s_n>0,N1`$ there exists a potential $`pL_{even}^2(𝕋)`$ and $`\epsilon >0`$ such that the each dislocation operator $`T_t^{di},t(0,\epsilon )`$ has gaps with lengths $`|\gamma _n(T_t^{di})|=s_n,`$ and $`\mathrm{\#}(T_t^{di},\gamma _n(T_t^{di}))=2d_n`$ for any $`n=1,2,..,N`$.
We now consider a half-solid, that is the Schrödinger operator
$$T_t^s=\frac{d^2}{dx^2}+q_t^s(x)\mathrm{in}L^2(),q_t^s(x)=s\chi _{}(x)+\chi _+(x)p(x+t),$$
where $`s,t`$ and $`pL^1(𝕋)`$ is real. Theorem 2.1 yields :
$$\sigma (T_t^s)=\sigma _{ac}(T_t^s)\sigma _d(T_t^s),\sigma _{ac}(T_t^s)=\sigma (H)[s,\mathrm{}),$$
(2.10)
$$\gamma _n(T_t^s)=\gamma _n(H)(\mathrm{},s),\mathrm{\#}(T_t^s,\gamma _n(T_t^s))2,n0,t[0,1].$$
(2.11)
Hence Theorem 2.1 yields if $`ps`$, then the Riemann surface $`\mathrm{\Lambda }_E(T_t^s)`$ consists of 4 sheets. Note that if $`s\alpha _1^{}(H)`$, then $`\sigma _{ac}(T_t^s)=(s_{},\mathrm{})`$, where $`s_{}=\mathrm{min}(0,s)`$. If $`s>\alpha _1^{}(H)`$, then there exists a gap in the spectrum of $`T_t^s.`$ Our goal is to study the eigenvalues in the gaps $`\gamma _n(T_t^s),n0,`$ and to find how these eigenvalues depend on $`t,s`$. It is clear that they depend on $`t`$ periodically. Define sets $`Q_N=\{q_0^s=s\chi _{}+\chi _+p_2:`$ $`pL_{even}^2(𝕋),\alpha _N^+<s<\alpha _{N+1}^{}\}`$ and $`P_N=\{(r,d,\epsilon ):\{r_n\}_1^{\mathrm{}}\mathrm{}^2,d=\{d_n\}_1^N`$, where for any $`nN`$ the number $`r_n0`$ and $`d_n=0`$ if $`r_n=0,\epsilon (0,1)\}`$ for some integer $`N0`$ . Define the mapping $`\omega :Q_NP_N`$ by $`\omega (q_0^s)=(r,d,\epsilon )`$, where
$$r_n=|\gamma _n(T_0^s)|,d_n=\mathrm{\#}(T_0^s,\gamma _n(T_0^s)),nN,r_n=|\sigma _n^1(T_0^s)|sign(\mu _n(p)\nu _n(p)),n>N,$$
$`\epsilon =\frac{\alpha _{N+1}^{}s}{\alpha _{N+1}^{}\alpha _N^+}`$ and $`\sigma _n^1(T_0^s)=(s,\mathrm{})\gamma _n(H)`$ is a segment of the spectrum of $`T_0^s`$ with multiplicity one. We consider a half-solid with an even potential $`p`$.
Theorem 2.6 i) Let $`T_0^s=\frac{d^2}{dx^2}+q_0^s(x)`$ acting in $`L^2()`$, where $`q_0^s=s\chi _{}+\chi _+p`$ and $`pL_{even}^2(𝕋)`$. Then $`\mathrm{\#}(T_0^s,\gamma _0(T_0^s))=0`$. Let, in addition, $`s>a_m^+(p)`$ for some $`m0`$. Then $`\sigma ^{(4)}\mathrm{}`$, and for each $`\gamma _n(T_0^s)\mathrm{},n=1,2,..,m,`$ the following identities are fulfilled:
$$\mathrm{\#}^{(4)}(T_0^s,\gamma _n(T_0^s))+\mathrm{\#}(T_0^s,\gamma _n(T_0^s))=1,\mathrm{\#}(T_0^s,\gamma _n(T_0^s))=\{\begin{array}{cc}1\hfill & \text{if }\mu _n>\nu _n,\hfill \\ 0\hfill & \text{if }\nu _n>\mu _n.\hfill \end{array}$$
(2.12)
ii) Each mapping $`\omega :Q_NP_N,N0,`$ is 1-to-1 and onto.
In this Theorem we proved that for any $`\{r_n\}_1^{\mathrm{}}\mathrm{}^2,d=\{d_n\}_1^N`$, and $`\epsilon (0,1)`$ (such that $`r_n0`$ and $`d_n=0`$ if $`r_n=0,1nN`$), there exists a unique half-solid potential $`q_0^s=s\chi _{}+\chi _+p`$ with $`pL_{even}^2(𝕋)`$.
For fixed $`\gamma _n(T_t^s)\mathrm{},n0,s>\alpha _n^\pm `$ we introduce the equation
$$\mathrm{\Psi }_y(y,\alpha _n^\pm )=\sqrt{s\alpha _n^\pm }\mathrm{\Psi }(y,\alpha _n^\pm ),y[0,1].$$
(2.13)
For each $`n1`$ there exist $`Nn`$ roots $`y_1,..,y_N[0,1]`$ of Eq. (2.13) on the interval $`[0,1]`$, see below Lemma 3.4. Let $`m^\pm (\lambda ,t)`$ be the Weyl function for the potential $`p(x+t)`$ and let $`\alpha _0^{}\mathrm{}`$.
Theorem 2.7. Let $`T_t^sy=y^{\prime \prime }+q_t^sy`$ be an operator acting in $`L^2()`$, where $`q_t^s=s\chi _{}+\chi _+p(+t)`$ and $`pL^1(𝕋),t`$. Then
i) Suppose that $`\alpha _n^{}<\alpha _n^+<s`$ for some $`n0`$ and $`y[0,1]`$ is some root of (2.13). Then there exists a unique function $`z_n^\pm ()C(\epsilon ,\epsilon ),z_n^\pm (0)=0`$ for some $`\epsilon >0`$, such that :
if $`z_n^\pm (t)>0`$ for some $`t(\epsilon ,\epsilon )`$, then $`\lambda _n^\pm (t)\alpha _n^\pm z_n^\pm (t)^2\gamma _n`$ is an eigenvalue of $`T_{y+t}^s,`$
if $`z_n^\pm (t)<0`$ for some $`t(\epsilon ,\epsilon )`$, then $`\lambda _n^\pm (t)\gamma _n\mathrm{\Lambda }_0^{(4)}`$ is a resonance of $`T_{y+t}^s`$.
Moreover, the following asymptotics is fulfilled:
$$z_n^\pm (t)=\sqrt{|M_n^\pm /2|}\mathrm{\Psi }(y,\alpha _n^\pm )^2_0^t[p(y+\tau )\alpha _n^\pm ]𝑑\tau +O(t),t0.$$
(2.14)
ii) Suppose that $`\alpha _n^{}<s<\alpha _n^+,`$ for some $`n1,`$ or $`\nu _0^{}<s\alpha _0^+.`$ Let $`y`$ be a root of the equation $`m^+(s,y)=0,0y1`$. Then there exists a unique function $`z()W_1^2(\epsilon ,\epsilon ),z(0)=0`$ for some $`\epsilon >0`$ such that
if $`z(t)>0`$ for some $`t(\epsilon ,\epsilon )`$ then $`\lambda (t)s+z(t)^2\gamma _n`$ is an eigenvalue of $`T_{y+t}^s,`$
if $`z(t)<0`$ for some $`t(\epsilon ,\epsilon )`$ then $`\lambda (t)\gamma _n\mathrm{\Lambda }_0^{(3)}`$ is a resonance of $`T_{y+t}^s`$.
Moreover, the following asymptotic estimates are fulfilled:
$$z(t)=_0^t[p(y+\tau )s]𝑑\tau +O(t),t0.$$
(2.15)
## 3 Hill operator and translations
Let $`\phi (x,\lambda ,t),\vartheta (x,\lambda ,t)`$ be the solutions of the equation
$$y^{\prime \prime }+p(x+t)y=\lambda y,\lambda ,t,$$
(3.1)
satisfying $`\phi _x(0,\lambda ,t)=\vartheta (0,\lambda ,t)=1,`$ and $`\phi (0,\lambda ,t)=\vartheta _x(0,\lambda ,t)=0`$. Remark that the Lyapunov function $`\mathrm{\Delta }(\lambda )`$ for (3.1) coincides with the Lyapunov function for (2.1) (see \[L\]). Then we define the quasimomentun $`k`$ for (3.1) and again $`k(\lambda )`$ does not depend on $`t`$. Let $`\mu _n(p,t),n1,`$ be the Dirichlet spectrum of $`p(x+t),`$ i.e the spectrum of (3.1) with the boundary condition $`y(0)=y(1)=0`$ and let $`\nu _n(p,t),n0,`$ be the Neumann spectrum of $`p(x+t)`$, that is the spectrum of (3.1) with the boundary condition $`y^{}(0)=y^{}(1)=0.`$ We need some results on Eq. (3.1) (see \[L\], \[PTr\], \[T\],\[K2\] ). It is well known that $`\phi (1,\lambda ,t)`$ is an entire function of $`\lambda `$ of order $`\frac{1}{2}`$ for fixed $`t`$. The zeros of $`\phi (1,\lambda ,t)`$ coincide with the Dirichlet eigenvalues $`\mu _n(p,t),n1,`$ and the following asymptotics are fulfilled:
$$\mu _n(p,t)=(\pi n)^2+_0^1p(x)𝑑x_0^1p(x+t)\mathrm{cos}2\pi nxdx+O(1/n),\mathrm{as}n\mathrm{},$$
(3.2)
uniformly on bounded subsets of $`[0,1]\times L^1(0,1)`$. It is well known that $`\vartheta _x(1,\lambda ,t)`$ is an entire function of $`\lambda `$ of order $`1/2`$ for fixed $`t`$. The zeros of $`\vartheta _x(1,\lambda ,t)`$ coincide with the Neumann eigenvalues $`\nu _n(p,t),n0.`$ The functions $`\mu _n(p,t),\nu _n(p,t)`$ are 1-periodic. If the parameter $`t`$ runs through the interval $`[0,1]`$, then $`\mu _n(p,t),\nu _n(p,t)`$ run through the gap $`\gamma _n=(\alpha _n^{},\alpha _n^+),n1`$. If the gap $`\gamma _n=\mathrm{}`$, then $`\mu _n(p,t),\nu _n(p,t)`$ don’t move and $`\mu _n(p,t)=\nu _n(p,t)=\alpha _n^\pm `$. The eigenvalue $`\nu _0(p,t)\alpha _0^+,t[0,1].`$ In the book \[PTr\] there are the asymptotics of the solutions $`\phi (x,\lambda ),\vartheta (x,\lambda )`$ as $`|\lambda |\mathrm{}.`$ Repeating it for Eq. (3.1) we obtain the asymptotics for $`\phi (x,\lambda ,t),\vartheta (x,\lambda ,t)`$ as $`|\lambda |\mathrm{}`$. For example,
$$\phi (x,\lambda ,t)=\frac{\mathrm{sin}\sqrt{\lambda }x}{\sqrt{\lambda }}+O(\frac{\mathrm{exp}Im\sqrt{\lambda }x}{\lambda })\mathrm{as}x,t[0,1],|\lambda |\mathrm{}.$$
(3.3)
These asymptotics can be differentiated with respect to $`x,t`$ and /or $`\lambda `$ and are uniform on $`[0,1]\times [0,1]\times L^1(0,1)`$. We have the Trubowitz identity (see \[Tr\])
$$\phi (1,\lambda ,t)=\phi (1,\lambda )\psi _+(t,\lambda )\psi _{}(t,\lambda )=\underset{n1}{}\frac{(\mu _n(p,t)\lambda )}{(\pi n)^2},\lambda ,t,$$
(3.4)
and the equality $`\phi (1,\lambda )\psi _+(t,\lambda )\psi _{}(t,\lambda )=2\mathrm{\Delta }^{}(\lambda )\mathrm{\Psi }(t,\lambda )^2`$ at $`\lambda =\alpha _n^\pm `$, yields
$$\phi (1,\alpha _n^\pm ,t)=(1)^n2M_n^\pm \mathrm{\Psi }(t,\alpha _n^\pm )^2,t,n0.$$
(3.5)
where the eigenfunction $`\mathrm{\Psi }(t,\alpha _n^\pm )`$ is defined at the start of Section 2. Define a function $`\varphi _n(\lambda ,t)`$ by $`\phi (1,\lambda ,t)(1)^n(\lambda \mu _n(p,t))\varphi _n(\lambda ,t)`$. Then we have
$$\varphi _n(\lambda ,t)>0,\lambda \gamma _n,\mathrm{and}(1)^n\phi (1,\lambda ,t)>0,\lambda >\mu _n(t),\lambda \gamma _n,$$
(3.6)
Introduce a function $`b(\lambda )=i\mathrm{sin}k(\lambda )`$. It is known that
$$k(\lambda )\pi +iv(\lambda ),v(\lambda )>0,b(\lambda )=(1)^n\sqrt{\mathrm{\Delta }^2(\lambda )1},\lambda \gamma _n\mathrm{\Lambda },$$
(3.7)
where $`\sqrt{\mathrm{\Delta }(\lambda )^21}>0`$ as $`\lambda \gamma _n\mathrm{\Lambda }`$. Below we need the identity (see \[KK1\])
$$M_n^\pm =\mathrm{\Delta }(\alpha _n^\pm )\mathrm{\Delta }^{}(\alpha _n^\pm ),n0,$$
(3.8)
and using (3.7-8) we obtain
$$b(\lambda )=(1)^nz(\sqrt{|2M_n^\pm |}+O(z^2)),\mathrm{as}\lambda \gamma _n\mathrm{},\lambda =\alpha _n^\pm z^2,z0.$$
(3.9)
The Weyl function $`m^\pm `$ for the potential $`p(x+t)`$ has the form
$$m^\pm (\lambda ,t)=\frac{a(\lambda ,t)\pm i\mathrm{sin}k(\lambda )}{\phi (1,\lambda ,t)}=\frac{a(\lambda ,t)b(\lambda )}{\phi (1,\lambda ,t)},a(\lambda ,t)\frac{\phi _x(1,\lambda ,t)\vartheta (1,\lambda ,t)}{2}.$$
(3.10)
Below we need the identity
$$a^2(\lambda ,t)+1\mathrm{\Delta }^2(\lambda )=a^2(\lambda ,t)b^2(\lambda )=\phi (1,\lambda ,t)\vartheta _x(1,\lambda ,t).$$
(3.11)
Recall that $`\dot{u}=\frac{}{t}u`$. We have the equations
$`\dot{\vartheta }_x(1,\lambda ,t)=(\lambda p(t))\dot{\phi }(1,\lambda ,t),\dot{\phi }(1,\lambda ,t)=2a(\lambda ,t),`$ $`\dot{a}(\lambda ,t)=\vartheta _x(1,\lambda ,t)(\lambda p(t))\phi (1,\lambda ,t),`$ (3.12)
(see \[L\]) and the following identities
$$\phi (x,\lambda ,t)=\vartheta (t,\lambda )\phi (x+t,\lambda )\vartheta (x+t,\lambda )\phi (t,\lambda ),$$
(3.13)
$$\vartheta (x,\lambda ,t)=\vartheta (x+t,\lambda )\phi _x(t,\lambda )\vartheta _x(t,\lambda )\phi (x+t,\lambda ).$$
(3.14)
Let $`C^m(𝕋),m0,`$ be the space of m times continuously differentiable real-valued 1-periodic functions. Suppose that $`pC^1(𝕋),`$ then for any $`t[0,1]`$ the identity (the trace formula)
$$p(t)=\alpha _0^++\underset{n1}{}(\alpha _n^{}+\alpha _n^+2\mu _n(p,t)),$$
(3.15)
holds, where the series converges absolutely and uniformly (see \[L\]). We need the following result on the Dirichlet spectrum (see \[K2\]).
Theorem 3.1. Let a real potential $`pL^1(𝕋)`$ and $`\mu _n(t)=\mu _n(p,t),n1`$. Then
i) Each $`\mu _n()C^2(𝕋),n1,`$ and $`\mu _n^{\prime \prime \prime }L^1(𝕋).`$ Let in addition $`pL^2(𝕋)`$ (or $`pC^m(𝕋),m0`$). Then $`\mu _n^{\prime \prime \prime }L^2(𝕋)`$ (or $`\mu _n()C^{m+3}(𝕋),m0`$).
ii) There exists a function $`y_nC^1(),`$ such that $`\mu _n(t)=\alpha _n^{}+|\gamma _n|\mathrm{sin}^2y_n(t),`$ where $`\mu _n(0)=\alpha _n^{}+|\gamma _n|\mathrm{sin}^2y_n(0),`$ and uniformly on $`t[0,1]`$ the following asymptotics are fulfilled:
$$y_n(t)=y_n(0)+\pi nt+O(1/n),\dot{y}_n(t)=\pi n+o(1),n\mathrm{}.$$
(3.16)
iii) Suppose that $`\mu _n(t_0)=\alpha _n^{}`$ or $`\mu _n(t_0)=\alpha _n^+,`$ for some $`t_0[0,1],`$ and $`n1.`$ Then the following asymptotics is fulfilled:
$$\mu _n(t_0+t)=\mu _n(t_0)+t^2\ddot{\mu }_n(t_0)/2+o(t^2),\ddot{\mu }_n(t_0)=4M_n^\pm /\phi _\lambda ^{}(\mu _n(t_0),t_0)^2.$$
(3.17)
Remark. i) In other words we have the following result. Slit the n-th gap $`\gamma _n\mathrm{}`$ and place $`\mu _n`$ on the upper or lower lip according to the signature of $`\mathrm{sinh}q(\mu _n)`$, i.e., on the upper when positive and on the lower when negative. Then $`\mu _n(t)`$ runs clockwise around the ”circle ”, changing lips when it hits $`\alpha _n^\pm `$, making $`n`$ complete revolutions in unit time. Then (roughly speaking) $`\mu _n(t)=\alpha _n^{}+|\gamma _n|\mathrm{sin}^2\pi nt,`$ when $`t`$ runs through the interval $`[0,1].`$
In order to study the eigenvalues in gaps we need properties of the function $`\zeta (\lambda ,t)=\dot{\phi }(1,\lambda ,t)/(2\phi (1,\lambda ,t))`$.
Lemma 3.2. For each $`(t,p,\lambda )F=[0,1]\times L^1(𝕋)\times \{\mu _n(p,t),n1\}`$ the following identities are fulfilled:
$$\zeta (\lambda ,t)\frac{\dot{\phi }(1,\lambda ,t)}{2\phi (1,\lambda ,t)}=\frac{1}{2}\underset{n1}{}\frac{\dot{\mu }_n(p,t)}{\mu _n(p,t)\lambda },$$
(3.18)
$$\dot{m}^\pm (\lambda ,t)=(p(t)\lambda )m^\pm (\lambda ,t)^2,$$
(3.19)
$$\dot{\zeta }(\lambda ,t)=(p(t)\lambda )\zeta ^2(\lambda ,t)\frac{b^2(\lambda )}{\phi ^2(1,\lambda ,t)},$$
(3.20)
$$_0^1\zeta (\lambda ,t)𝑑t=0,\lambda <\alpha _1^{},_0^1\zeta (\alpha _0^+,t)^2𝑑t=_0^1p(t)𝑑t\alpha _0^+,$$
(3.21)
$$\phi (1,\lambda ,t)\dot{\zeta }(\lambda ,t)=\pm (1)^n2M_n^\pm L(t,\lambda ),if\lambda =\alpha _n^\pm \mu _n(p,t).$$
(3.22)
where the series converges absolutely and uniformly on compact sets in $`F`$ and $`L`$ is given by (2.7).
Proof. Let $`\mu _n(t)=\mu _n(p,t)`$. The functions $`\phi (1,\lambda ,t)`$ and $`\dot{\phi }(1,\lambda ,t)`$ are entire in $`\lambda `$. xThe zeros of $`\phi (1,\lambda ,t)`$ have the asymptotics (3.2). Using (3.2-3) we get the asymptotics $`\zeta (\lambda ,t)=O(1/\sqrt{\lambda })`$ as $`|\lambda |\mathrm{},|\sqrt{\lambda }\pi n|1/4`$. Then (3.4) yields
$$\zeta (\lambda ,t)\underset{n1}{\overset{N}{}}\frac{\dot{\mu }_n(t)}{\mu _n(t)\lambda }=\frac{1}{2\pi i}_{|\sqrt{z}|=\pi (2N+1)/2}\frac{\zeta (z,t)dz}{z\lambda },$$
since the residue of the function $`\zeta (z,t)`$ at the simple pole $`\mu _n(t)`$ has the form: $`\mathrm{Res}\zeta (\lambda ,t)=\dot{\phi }(1,\mu _n(t),t)/\phi ^{}(1,\mu _n(t),t)=\dot{\mu }_n(t)`$. Then as $`N\mathrm{}`$ we get (3.18). Remark that (3.16) yields $`\dot{\mu }_n(t)=O(n|\gamma _n|)`$ as $`n\mathrm{}`$ uniformly on $`t[0,1].`$ Then by (3.2), series (3.18) converges absolutely and uniformly on compact sets.
We will obtain the equations for the function $`m^+(\lambda ,t)`$, the proof for $`m^{}(\lambda ,t)`$ and (3.20) is similar. Using (3.12), (3.10-12) we deduce that
$$\dot{m}^+=\frac{\dot{a}}{\phi }\frac{(ab)\dot{\phi }}{\phi ^2}=(p\lambda )\frac{\vartheta _x}{\phi }\frac{2(ab)a}{\phi ^2}=(p\lambda )+\frac{a^2b^2}{\phi ^2}\frac{2am^+}{\phi },$$
which yields (3.19). Integrating $`\zeta =\dot{\phi }/2\phi `$ and since $`\phi (\lambda ,t)>0`$ if $`\lambda <\alpha _1^{},t[0,1],`$ we obtain the first identity in (3.21). Integrating (3.20) implies the second identity in (3.21). Substituting (3.5) into (3.20) we have (3.22).
Now we will obtain more exact estimates concerning $`L(t,\alpha _n^\pm )`$ defined in (2.7). These results are used to prove the existence of two eigenvalues in gaps (see Theorem 2.5). Recall that if $`pL_{even}^2(0,1)`$, then $`\nu _0<\mu _1`$ and $`\gamma _n(H)=(\mu _n,\nu _n)`$ or $`\gamma _n(H)=(\nu _n,\mu _n),n1`$ \[GT\].
Lemma 3.3. i) Let $`pC^2(𝕋)`$ be even. Assume that for some $`N1`$ and for all $`n>N`$ the gap $`\gamma _n(H)`$ are given by $`\gamma _n(H)=(\mu _n(p),\nu _n(p))`$. Then
$$p(0)\alpha _N^\pm (\pi N)^2+\underset{mN+1}{}|\gamma _m|2\underset{1}{\overset{N}{}}|\gamma _m|.$$
(3.23)
ii) For any finite sequences $`\{s_n\}_1^N,\{d_n\}_1^N,`$ where $`s_n>0,d_n\{0,1\}`$, there exists $`pL_{even}^2(𝕋)`$ with gap lengths $`|\gamma _n(H)|=s_n`$ and $`(1)^{d_n}L(0,\alpha _n^\pm )<0`$ for all $`n=1,\mathrm{},N.`$
Remark. Roughly speaking the result of ii) is the effect of a big gap $`\gamma _{N+1}`$ such that $`|\gamma _{N+1}|>(\pi N)^2+2(|\gamma _1|+\mathrm{}+|\gamma _N|)`$. It is important that we have 2 types of gaps: 1) the gap $`\gamma _n(H)=(\mu _n(p),\nu _n(p))`$ for all $`n>N`$, 2) $`\gamma _n(H)=(\mu _n,\nu _n)`$ or $`\gamma _n(H)=(\nu _n,\mu _n)`$, which depends on $`d_n`$ for $`1nN`$.
Proof. i) Using the trace formula (3.14) and the identity $`\alpha _n^+=_{m1}^n(|\sigma _m|+|\gamma _m|)`$ we obtain
$$p(0)\alpha _N^+=\underset{m1}{}(\alpha _m^++\alpha _m^{}2\mu _m(p))\underset{m=1}{\overset{N}{}}(|\sigma _m|+|\gamma _m|)\underset{mN+1}{}|\gamma _m|2\underset{m1}{\overset{N}{}}|\gamma _m|\underset{m=1}{\overset{n}{}}|\sigma _m|$$
and the estimate $`|\sigma _m|<\pi ^2(2m1)`$ (see \[Mos\]) yields (3.23).
ii) We need a result from \[GT\]. For an even periodic potential we define a signed gap length $`l_n=\mu _n\nu _n,n1,`$ and the corresponding sequence $`l=\{l_n\}_1^{\mathrm{}}`$. For any sequence $`\{t_n\}_1^{\mathrm{}}\mathrm{}^2`$ there exists a unique even periodic potential $`p`$ such that the signed gap length $`l_n=t_n,n1.`$ Using this result we fix the number $`N`$ and due to (3.23) we take a sequence of signed gap lengths $`\{l_n\}_1^{\mathrm{}}`$ such that $`p(0)\alpha _N^+>0`$. Hence for all $`n=1,..,N`$ we get $`p(0)\alpha _n^\pm >0`$. Thus for each $`1nN`$ we obtain:
If $`d_n=1`$ we take $`\mu _n=\alpha _n^{}`$ and $`\nu _n=\alpha _n^+`$, then $`L(0,\alpha _n^\pm )>0`$.
If $`d_n=0`$ we take $`\nu _n=\alpha _n^{}`$ and $`\mu _n=\alpha _n^+`$, then $`L(0,\alpha _n^\pm )<0.\text{ }`$
We need some results about the roots of the equation $`m^\pm (\lambda ,t)=\omega ,t[0,1],`$ and some formulas for even potentials.
Lemma 3.4. i) Let $`pL^1(𝕋)`$ and a gap $`\gamma _n(H)\mathrm{}`$ for some $`n1`$. Then for any fixed $`(\lambda ,\omega )[\alpha _n^{},\alpha _n^+]\times `$, there exist $`N_\pm n`$ roots of the equation $`m^\pm (\lambda ,t)=\omega ,t[0,1].`$
ii) For any even potential $`pL^1(𝕋)`$ the following identities are fulfilled:
$$a(\lambda ,0)0,\mathrm{\Delta }(\lambda )\vartheta (1,\lambda ),m^\pm (\lambda )\frac{\pm i\mathrm{sin}k(\lambda )}{\phi (1,\lambda ,0)}.$$
(3.24)
Proof. Let $`\mu _n(t)=\mu _n(p,t),n1`$. i) We consider $`m^+`$ and firstly let $`\lambda =\alpha _n^{},`$ the proof for $`m^+`$ and $`\lambda =\alpha _n^+`$ is similar. By Theorem 3.1, $`\mu _n()C^2(𝕋)`$ and there exist points $`\tau _r[0,1),r=1,..,n,`$ such that $`\mu _n(\tau _r)=\alpha _n^{}`$ and $`\dot{\mu }_n(\tau _r)=0,\ddot{\mu }_n(\tau _r)>0.`$ By (3.18),
$$m^+(\lambda ,t)=\zeta (\lambda ,t)=\frac{\dot{\mu }_n(t)}{\mu _n(t)\alpha _n^{}}+\underset{mn}{}\frac{\dot{\mu }_m(t)}{\mu _m(t)\alpha _n^{}}.$$
Then $`m^+(\lambda ,)`$ maps the interval $`I_r=(\tau _r,\tau _{r+1})`$ onto the real line $``$. Therefore, for any number $`\omega `$ there exist $`Nn`$ roots of the equation $`m^+(\lambda ,t)=\omega `$.
Secondly, let $`\lambda (\alpha _n^{},\alpha _n^+)`$ and $`b(\lambda )>0,`$ the proof for $`b(\lambda )<0`$ is similar. We get $`2a(\lambda ,t)=\dot{\phi }(1,\lambda ,t)=\dot{\mu }_n(t)(1)^n\varphi (\lambda ,t)`$ at $`\lambda =\mu _n(t).`$ Since $`\mu _n(t)`$ crosses the point $`\lambda `$ exactly $`2n`$ times there exist points $`\tau _r[0,1)`$ $`r=1,2,3,..,n,`$ such that $`\mu _n(\tau _r)=\lambda `$ and $`(1)^n\dot{\mu }_n(\tau _r)<0.`$ Then $`a(\lambda ,\tau _r)b(\lambda )0`$, and we have
$$m^+(\lambda ,t)=\frac{a(\lambda ,t)b(\lambda )}{\phi (1,\lambda ,t)}=\frac{a(\lambda ,t)b(\lambda )}{(1)^n(\lambda \mu _n(t))\varphi _n(\lambda ,t)}.$$
(3.25)
Recall that each function $`\varphi _n(\lambda ,),\lambda \gamma _n`$ has no zero on the interval $`[0,1]`$. By the properties of $`\mu _n(t)`$ from Theorem 3.1, the function $`m^+(\lambda ,)`$ maps the interval $`I_p=(\tau _p,\tau _{p+1})`$ onto the real line $``$ . Then for any $`\omega `$ there exist $`N_\pm n`$ roots of the equation $`m^+(\lambda ,t)=\omega `$.
ii) It is well known that for an even potentials the points $`\mu _n(0)`$ and $`\nu _n(0)`$ lie on the endpoints of the gap $`\gamma _n`$(see \[GT\]) . Then by Theorem 3.1, $`\dot{\mu }_n(0)=0`$, and using (3.18) we get $`\zeta (\lambda ,0)=0`$ for all $`\lambda `$. Hence relations (3.10), (3.12) imply (3.24).
We consider the equation $`\zeta (0,t)=r,t[0,1),`$ for fixed $`r.`$ Let $`\psi (t)`$ be any smooth 1-periodic positive function. Then we take $`\mathrm{\Psi }(t,0)=\psi (t)`$ and the potential $`p(t)=\ddot{\psi }(t)/\psi (t).`$ In this case the function $`\zeta (0,t)`$ is periodic smooth with $`_0^1\zeta (0,t)𝑑t=0`$ see (3.21). We are able to get a function $`\psi `$, when the equation $`\zeta (0,t)=r`$ has any number of roots (depending on $`\psi `$). In order to study the ground state we need some properties of the Weyl function.
Lemma 3.5. Let $`pL^1(0,1).`$ Then $`\pm m^\pm (\lambda )\mathrm{}`$ as $`\lambda \mathrm{},`$ and
$$\pm m^\pm (\lambda )<0\mathrm{if}\lambda <\nu _0,\mathrm{and}m^+(\lambda )m^{}(\lambda )>0,\mathrm{if}\nu _0<\lambda <0,$$
(3.26)
Moreover, if $`\pm m^+(0)>0`$, then $`\pm m^+(\lambda )>0`$ for $`\nu _0<\lambda <0`$.
Proof. (4.5) yields the asymptotics $`\pm m^\pm (\lambda )\mathrm{}`$ as $`\lambda \mathrm{}`$. We have the identity $`a^2b^2=\phi \vartheta _x`$. It is well known that: $`\phi (\lambda )>0,\lambda <\mu _1,`$ and $`\vartheta _x(\lambda )>0,\lambda <\nu _0`$ and $`\vartheta _x(\lambda )<0,\nu _0<\lambda <0`$. Hence we get (3.26) since $`m^\pm `$ has no zero on the intervals $`(\mathrm{},\nu _0)`$ and $`(\nu _0,0)`$.
Let $`\pm m^+(0)>0`$. Due to (3.11), the function $`m^+`$ has no zero on the interval $`(\nu _0,0)`$. Then for the case $`\lambda (\nu _0,0)`$ we have $`m^+(\lambda )>0`$, if $`m^+(0)>0`$, and $`m^+(\lambda )<0,`$ if $`m^+(0)<0.\text{ }`$
## 4 Biperiodic potentials
In this section we study the spectrum of $`T_t`$ at $`t=0`$, i.e., $`T_0=\frac{d^2}{dx^2}+q_0`$, where $`q_0=p_1\chi _{}+\chi _+p`$ and $`pL^1(𝕋),p_1L^1(\tau 𝕋)`$. We now determine the eigenfunctions of $`T_0`$. Let $`u(x,\lambda ),v(x,\lambda )`$ be the solutions of the equation
$$y^{\prime \prime }+q_0y=\lambda y,\lambda ,$$
(4.1)
satisfying $`u^{}(0,\lambda )=v(0,\lambda )=1,`$ and $`u(0,\lambda )=v^{}(0,\lambda )=0.`$ Let $`\phi _j(x,\lambda ),\vartheta _j(x,\lambda ),j=1,2,`$ be the solutions of the equation
$$y^{\prime \prime }+p_jy=\lambda y,\lambda ,$$
satisfying $`\phi _j^{}(0,\lambda )=\vartheta _j(0,\lambda )=1,`$ and $`\phi _j(0,\lambda )=\vartheta _j^{}(0,\lambda )=0.`$ Remark that $`\vartheta ,\phi ,\vartheta _j,\phi _j`$ are entire functions of $`\lambda ,`$ and are real on the real line. Then we obtain
$$v(x,\lambda )=\{\begin{array}{cc}\vartheta _1(x,\lambda )\mathrm{if}x<0,& \\ \vartheta _2(x,\lambda )\mathrm{if}x>0,& \end{array}u(x,\lambda )=\{\begin{array}{cc}\phi _1(x,\lambda )\mathrm{if}x<0,& \\ \phi _2(x,\lambda )\mathrm{if}x>0,& \end{array}$$
(4.2)
Let $`\mathrm{\Delta }_j(\lambda )`$ be the Lyapunov function for the potential $`p_j,j=1,2`$. We need also the quasimomentum $`k_j(\lambda )=\mathrm{arccos}\mathrm{\Delta }_j(\lambda )`$, which is analytic in the domain $`\mathrm{\Lambda }_j=\sigma (H_j)`$. Introduce the Bloch functions $`\psi _j^\pm (,\lambda )L^2(_\pm ),\lambda \mathrm{\Lambda }_j,`$ for the operator $`H_j=\frac{d^2}{dx^2}+p_j,j=1,2`$, by
$$\psi _j^\pm (x,\lambda )=\vartheta _j(x,\lambda )+m_j^\pm (\lambda )\phi _j(x,\lambda ),m_j^\pm (\lambda )=\frac{a_j(\lambda )\pm i\mathrm{sin}k_j(\lambda )}{\phi _j(\lambda )}.$$
(4.3)
Then for $`\lambda \mathrm{\Lambda }_j,j=1,2`$ we have
$$\phi _j(x,\lambda )=\frac{\psi _j^+(x,\lambda )\psi _j^{}(x,\lambda )}{w_j(\lambda )},\vartheta _j(x,\lambda )=\frac{m_j^+(\lambda )\psi _j^{}(x,\lambda )m_j^{}(\lambda )\psi _j^+(x,\lambda )}{w_j(\lambda )}.$$
(4.4)
The function $`e^{ik_j(\lambda )x}\psi _j^\pm (x,\lambda )`$ is $`\tau _j`$periodic in $`x`$ for any $`\lambda `$, where $`\tau _1=\tau ,\tau _2=1.`$ Introduce the domain $`\mathrm{\Lambda }_j(\delta )=\{\lambda \mathrm{\Lambda }_j,|\sqrt{\lambda }\pi n/\tau _j|\delta ,n1\},\delta >0`$. Below we need the asymptotics
$$\psi _j^\pm (x,\lambda )=e^{\pm i\sqrt{\lambda }x}(1+O(\lambda ^{1/2})),m_j^\pm (\lambda )=\pm i\sqrt{\lambda }(1+O(\lambda ^{1/2})),|\lambda |\mathrm{},$$
(4.5)
$`\lambda \mathrm{\Lambda }_j(\delta )`$, uniformly on $`[0,\tau _j]`$ (see \[T\]). We introduce $`\mathrm{\Psi }^\pm (,\lambda )L^2(_\pm ),\lambda \mathrm{\Lambda }_0^{(1)}=\sigma _c(T_0)`$, which are solutions of Eq. (4.1) and having the form
$$\mathrm{\Psi }^+(x,\lambda )=\psi _2^+(x,\lambda ),x>0;\mathrm{\Psi }^{}(x,\lambda )=\psi _1^{}(x,\lambda ),x<0,\lambda \mathrm{\Lambda }_0^{(1)}.$$
We determine these functions on the real line. First we find the Wronskians
$$w(\lambda )=\{\mathrm{\Psi }_{},\mathrm{\Psi }_+\}=m_2^+(\lambda )m_1^{}(\lambda ),\lambda \mathrm{\Lambda }_0^{(1)},$$
(4.6)
$$w_j(\lambda )=\{\psi _j^{},\psi _j^+\}=m_j^+(\lambda )m_j^{}(\lambda ),\lambda \mathrm{\Lambda }_0^{(1)},j=1,2.$$
(4.7)
Hence we get
$$ww_2=(m_2^+m_1^{})(m_2^+m_2^{})=m_2^{}m_1^{},$$
(4.8)
$$ww_1=(m_2^+m_1^{})(m_1^+m_1^{})=m_2^+m_1^+.$$
(4.9)
The definition of the Weyl functions yields
$$w(\lambda )=\left(\frac{a_2(\lambda )}{\phi _2(1,\lambda )}\frac{a_1(\lambda )}{\phi _1(1,\lambda )}\right)+i\left(\frac{\mathrm{sin}k_2(\lambda )}{\phi _2(1,\lambda )}+\frac{\mathrm{sin}k_1(\lambda )}{\phi _1(1,\lambda )}\right),\lambda \mathrm{\Lambda }_0^{(1)}.$$
(4.10)
Relations (4.6), (4.3) yield: if $`\lambda \sigma (H)(\sigma (H)\{\mu _n(p_j),j=1,2,n1\})`$, then we get $`\pm \mathrm{Im}w(\lambda \pm i0)>0`$. Using (4.1-2) we have
$$\mathrm{\Psi }^+(x,\lambda )=\psi _2^+(x,\lambda )=\vartheta _2(x,\lambda )+m_2^+(\lambda )\phi _2(x,\lambda )=v(x,\lambda )+m_2^+(\lambda )u(x,\lambda ),x0,\lambda \mathrm{\Lambda }_0^{(1)},$$
and (4.2) implies $`\mathrm{\Psi }^+(x,\lambda )=\vartheta _1(x,\lambda )+m_2^+(\lambda )\phi _1(x,\lambda ),x<0,\lambda \mathrm{\Lambda }_0.`$ Then identities (4.3-9) yield for $`x<0,\lambda \mathrm{\Lambda }_0^{(1)}`$:
$$\mathrm{\Psi }^+=\frac{m_1^+\psi _1^{}m_1^{}\psi _1^+}{w_1}+m_2^+\frac{\psi _1^+\psi _1^{}}{w_1}=\frac{m_1^+m_2^+}{w_1}\psi _1^{}+\frac{m_2^+m_1^{}}{w_1}\psi _1^+=\frac{w_1w}{w_1}\psi _1^{}+\frac{w}{w_1}\psi _1^+,$$
and therefore,
$$\mathrm{\Psi }^+(x,\lambda )=\{\begin{array}{cc}\psi _2^+(x,\lambda )\hfill & \text{if }x>0,\hfill \\ \frac{w_1(\lambda )w(\lambda )}{w_1(\lambda )}\psi _1^{}(x,\lambda )+\frac{w(\lambda )}{w_1(\lambda )}\psi _1^+(x,\lambda )\hfill & \text{if }x<0\hfill \end{array},\lambda \mathrm{\Lambda }_0^{(1)},$$
(4.11)
and similarly we have
$$\mathrm{\Psi }^{}(x,\lambda )=\{\begin{array}{cc}\frac{w_2(\lambda )w(\lambda )}{w_2(\lambda )}\psi _2^+(x,\lambda )+\frac{w(\lambda )}{w_2(\lambda )}\psi _2^{}(x,\lambda )\hfill & \text{if }x>0,\hfill \\ \psi _1^{}(x,\lambda )\hfill & \text{if }x<0\hfill \end{array},\lambda \mathrm{\Lambda }_0^{(1)}.$$
(4.12)
Substituting (4.5) into (4.11-12) we obtain the asymptotics
$$\mathrm{\Psi }^\pm (x,\lambda )=\mathrm{exp}(\pm i\sqrt{\lambda }x)(1+O(\lambda ^{1/2})),\mathrm{as}|\lambda |\mathrm{},\lambda \mathrm{\Lambda }_1(\delta )\mathrm{\Lambda }_2(\delta ),$$
(4.13)
uniformly on $`[\tau ,1]`$, for some $`\delta >0`$. Since the functions $`\mathrm{\Psi }_\pm (x,\lambda ),x`$, are real on the half-line $`\lambda (\mathrm{},0)`$ we have the identities
$$\overline{\mathrm{\Psi }}^\pm (x,\lambda )=\mathrm{\Psi }^\pm (x,\overline{\lambda }),\lambda \mathrm{\Lambda }_0^{(1)}.$$
(4.14)
The kernel $`R(x,x^{},\lambda )`$ of the operator $`(T_0\lambda )^1,\lambda \mathrm{\Lambda }`$, has the form
$$R(x,x^{},\lambda )=\frac{1}{w(\lambda )}\{\begin{array}{cc}\mathrm{\Psi }_+(x,\lambda )\mathrm{\Psi }_{}(x^{},\lambda )\mathrm{if}x>x^{}& \\ \mathrm{\Psi }_{}(x,\lambda )\mathrm{\Psi }_+(x^{},\lambda )\mathrm{if}x<x^{}& \end{array},\lambda \mathrm{\Lambda }_0^{(1)}.$$
(4.15)
We prove the first main theorem concerning $`T_t`$.
Proof of Theorem 2.1 i) We shall use standard tools from the theory of self-adjoint extensions and the Weyl Theorem. For the sake of the reader, we shall briefly mention all arguments. We need the following result of Krein \[Kr\]. Let $`A`$ be a closed symmetric operator in a separable Hilbert space $`.`$ Define the deficiency index $`n_\pm =dim[ker(A^{}\pm i)]`$. Remark that by the Neumann extension theory, every closed symmetric operator with $`n_+=n_{}`$ admits a self-adjoint extension. In accordance with Krein an open interval $`(\alpha ^{},\alpha ^+),`$ is called a gap of $`A`$ if
$$(A\frac{\alpha ^{}+\alpha ^+}{2})f\frac{\alpha ^+\alpha ^{}}{2}f,\mathrm{for}\mathrm{all}fD(A).$$
Assume that: a) $`n_{}=n_+=n0`$, b) the operator $`A`$ has a gap $`(\alpha ^{},\alpha ^+),`$ c) there exists a self-adjoint extension $`A_e`$ of $`A.`$ Then for any self-adjoint extension $`A_0`$ of $`A`$ we have:
$`\sigma _{ess}(A_0)=\sigma _{ess}(A_e),`$ and the spectrum of $`A_0`$ is discrete inside the gap and consists of at most $`n`$ eigenvalues counting multiplicities.
We need the simple fact (see Zheludev \[Z\]). Define the operator $`H_m^\pm y=y^{\prime \prime }+p_my`$ in $`L^2(_\pm )`$, with the Dirichlet boundary condition $`y(0)=0`$ for $`m=1,2,`$. Furthermore, $`\sigma (H_m^\pm )=\sigma _{ac}(H_m^\pm )\sigma _d(H_m^\pm ),\sigma _{ac}(H_m^\pm )=\sigma (H_m),`$ and the eigenvalues of $`H_m^\pm `$ coincide with some $`\mu _n(p_m),n1`$. Define the symmetric operator $`Af=T_0f,fC_0^{\mathrm{}}(\{0\})`$. Then $`n_{}(A)=n_+(A)=2`$ and there exists a special self-adjoint extension $`A_e=H_1^{}H_2^+`$ of $`A`$ in the space $`L^2(_{})L^2(_+)`$. Note that the operator $`(T_0i)^1(A_ei)^1`$ is compact with rank 2, since $`n_{}(A)=n_+(A)=2`$ (see \[Kr\]).
Hence using the results of Krein and Zheludev, (and the Weyl Theorem about the invariance of the essential spectrum) we obtain: a) $`\sigma _{ess}(T_0)=\sigma (H_1)\sigma (H_2),`$ b) there exist gaps $`\gamma _{nm}(T_0)=\gamma _n(H_1)\gamma _m(H_2),`$ for some $`n,m0,`$ in the essential spectrum of $`T_0`$, c) for each $`n,m0,`$ the spectrum of $`T_0`$ is discrete inside the gap $`\gamma _{nm}(T_0)\mathrm{}`$ and consists of at most two eigenvalues.
Moreover, using the properties of $`w,\mathrm{\Psi }^\pm `$ we deduce that the function $`R(x,x^{},\lambda \pm i0),\lambda ,`$ is real analytic in $`\lambda `$ away from the discrete set of the zeros of the functions $`w(\lambda ),\phi _m(1,\lambda ),m=1,2`$, which implies $`\sigma _{sc}(T_0)=\mathrm{}`$.
We prove the absence of eigenvalues in the continuous spectrum by contradiction. Let $`\lambda \sigma _c(T_0),f`$ be an eigenvalue and a corresponding eigenfunction of $`T_0`$. Hence $`\lambda \sigma (H_1)`$ or $`\lambda \sigma (H_2)`$. Let $`\lambda \sigma (H_2)`$, the proof for $`\lambda \sigma (H_1)`$ is similar. Then $`f^{\prime \prime }+p_2f=\lambda f,x>0.`$ But it is well known that in this case the function $`fL^2(_+)`$ (see \[T\] or \[CL\]). Hence we have a contradiction.
The results of ii) will be proved in Lemma 4.1-3.
iii) The Wronskian for the dislocation has the form $`w(\lambda ,t)=m^+(\lambda ,t)m^+(\lambda ,0)`$, which follows from (4.10). The Wronskian for the half-solid has the form $`w(\lambda ,t)=m^+(\lambda ,t)\sqrt{s\lambda }`$, which follows from (4.10). Using these identities and the definitions of $`m^+(\lambda ,t)`$ (see(3.10)) we obtain the statement iii).
Lemma 4.1. Let $`q_0`$ be a biperiodic potential in the sense of (1.1). Then
i) $`\sigma ^{(2)}=\sigma (H_1)\sigma (H)\mathrm{}`$ and for each $`x,x^{}`$ the functions $`R(x,x^{},),w,w_1,w_2,\mathrm{\Psi }^\pm (x,)`$ have a meromorphic continuation from $`\mathrm{\Lambda }_0^{(1)}`$ across the set $`\sigma ^{(2)}`$ to the second sheet $`\mathrm{\Lambda }_0^{(2)}`$, where the following identities are fulfilled:
$$w_n(\lambda ^{(2)})=w_n(\lambda ),m_n^\pm (\lambda ^{(2)})=m_n^{}(\lambda ),\psi _n^\pm (,\lambda ^{(2)})=\psi _n^{}(,\lambda ),n=1,2,$$
(4.16)
$$w(\lambda ^{(2)})=m_2^{}(\lambda )m_1^+(\lambda ),$$
(4.17)
$$\mathrm{\Psi }^+(x,\lambda ^{(2)})=\{\begin{array}{cc}\psi _2^{}(x,\lambda )\hfill & \text{if }x>0,\hfill \\ \mathrm{\Psi }^+(x,\lambda )+w_2(\lambda )\phi _1(x,\lambda )\hfill & \text{if }x<0.\hfill \end{array}$$
(4.18)
$$\mathrm{\Psi }^{}(x,\lambda ^{(2)})=\{\begin{array}{cc}\mathrm{\Psi }^{}(x,\lambda )+w_1(\lambda )\phi _2(v,\lambda )\hfill & \text{if }x>0\hfill \\ \psi _1^+(x,\lambda )\hfill & \text{if }x<0.\hfill \end{array}$$
(4.19)
ii) The number $`\lambda _e`$ is an eigenvalue of $`T_0`$ iff the number $`\lambda _e`$ considered as a point on the second sheet $`\lambda _e^{(2)}\mathrm{\Lambda }_0^{(2)}`$ is a resonance of $`\stackrel{~}{T}_0`$. All resonances of $`T_0`$ on the sheet $`\mathrm{\Lambda }_0^{(2)}`$ lie on the gaps $`\gamma _n(T_0),n0`$ and $`\mathrm{\#}^{(2)}(T_0,\gamma _n)2`$ for all $`n0`$.
Proof. i) In order to get an analytic continuation we have to study the function $`\mathrm{sin}k_j(\lambda ),j=1,2`$. The function $`\mathrm{sin}k_j(\lambda )`$ is real on $`\sigma (H_j)`$. Then $`k_j`$ has a meromorphic continuation across $`\sigma (H_j)`$ by the formula $`\overline{k}_j(\lambda )=k_j(\overline{\lambda })`$. Using (3.7) we have $`k_j(\lambda )=\pi n+iv(\lambda ),\lambda \gamma _n\mathrm{\Lambda }_0^{(1)}`$ and $`k_j(\lambda )=\pi niv(\lambda ),\lambda \gamma _n\mathrm{\Lambda }_0^{(2)}`$, where $`v=Imk`$. Then we get $`\mathrm{sin}k_j(\lambda )=\mathrm{sin}k_j(\lambda ^{(2)}),\lambda \gamma _n,`$ and using (4.3), (4.6-7) we obtain (4.16-17); (4.16-17) and (4.11-12) imply (4.18-19). Relations (4.16-19) yield a meromorphic continuation of $`R(x,x^{},\lambda )`$ into the second sheet $`\mathrm{\Lambda }_0^{(2)}`$.
ii) Let $`\stackrel{~}{w}`$ be the Wronskian for $`\stackrel{~}{T}_0`$. Identity (4.17) yields $`w(\lambda ^{(2)})=\stackrel{~}{w}(\lambda )`$. Then for any eigenvalue $`\lambda _e`$ of $`T_0`$ the number $`\lambda _e^{(2)}`$ is a resonance of $`\stackrel{~}{T}_0`$ and inversely. Hence all resonances on the sheet $`\mathrm{\Lambda }_0^{(2)}`$ of $`T_0`$ are real and lie on the gaps $`\gamma _n(T_0),n0,`$ of this sheet. Moreover, (1.3) yields $`\mathrm{\#}^{(2)}(T_0,\gamma _n)2.`$
We formulate the results about an analytic continuation into the sheet $`\mathrm{\Lambda }_0^{(3)}.`$
Lemma 4.2 Let $`q_0`$ be a biperiodic potential in the sense of (1.1) and let $`\sigma ^{(3)}=\sigma (H_1)\sigma (H_2)\mathrm{}`$. Then for each $`x,x^{}`$ the functions $`R(x,x^{},),w,w_1,w_2,\mathrm{\Psi }^\pm (x,)`$ have meromorphic continuations from $`\mathrm{\Lambda }_0^{(1)}`$ across the set $`\sigma ^{(3)}`$ to the sheet $`\mathrm{\Lambda }_0^{(3)}`$, where the following identities are fulfilled:
$$m_1^\pm (\lambda ^{(3)})=m_1^{}(\lambda ),\psi _1^\pm (,\lambda ^{(3)})=\psi _1^{}(,\lambda ),w(\lambda ^{(3)})=m_2^+(\lambda )m_1^+(\lambda ),$$
(4.20)
$$\mathrm{\Psi }^{}(x,\lambda ^{(3)})=\{\begin{array}{cc}\mathrm{\Psi }^{}(x,\lambda )w_1(\lambda )\phi _2^+(x,\lambda ),\hfill & \text{if }x>0,\hfill \\ \psi _1^+(x,\lambda )\hfill & \text{if }x<0,\hfill \end{array}$$
(4.21)
$`w_1(\lambda ^{(3)})=w_1(\lambda ),`$ and the functions $`\mathrm{\Psi }^+(x,),w_2,m_2^\pm ,\psi _2^\pm `$ don’t change.
We formulate the results about an analytical continuation into the sheet $`\mathrm{\Lambda }^{(4)}.`$
Lemma 4.3 Let $`q_0`$ be a biperiodic potential in the sense of (1.1) and let $`\sigma ^{(4)}=\sigma (H_2)\sigma (H_1)\mathrm{}`$. Then for each $`x,x^{}`$ the functions $`R(x,x^{},),w,w_1,w_2,\mathrm{\Psi }^\pm (x,)`$ have meromorphic continuation from $`\mathrm{\Lambda }_0^{(1)}`$ across the set $`\sigma ^{(4)}`$ to the sheet $`\mathrm{\Lambda }_0^{(4)}`$ and the following identities are fulfilled:
$$m_2^\pm (\lambda ^{(3)})=m_2^{}(\lambda ),\psi _2^\pm (x,\lambda ^{(3)})=\psi _2^{}(x,\lambda ),w(\lambda ^{(4)})=m_2^{}(\lambda )m_1^{}(\lambda ),$$
(4.22)
$$\mathrm{\Psi }^+(x,\lambda ^{(3)})=\{\begin{array}{cc}\psi _2^{}(x,\lambda )\hfill & \text{if }x>0,\hfill \\ \mathrm{\Psi }^+(x,\lambda )w_2(\lambda )\phi _1(x,\lambda )\hfill & \text{if }x<0\hfill \end{array}$$
(4.23)
$`w_2(\lambda ^{(4)})=w_2(\lambda ),`$ and the functions $`\mathrm{\Psi }^{}(x,),w_1,m_1^\pm ,\psi _1^\pm `$ don’t change.
The proof of Lemmas 4.2-3 repeats the proof of Lemma 4.1.
## 5 Eigenvalues and resonances
We consider the eigenvalues of the operator $`T_0`$ with even potentials $`p_1,p.`$ Recall that for an even potential the function $`a()0`$ and the Weyl function $`m^\pm =\pm i\mathrm{sin}k/\phi (1,)`$.
Proof of Theorem 2.2 i) If $`\stackrel{~}{\mu }_1<\stackrel{~}{\nu }_1`$ and $`\stackrel{~}{\mu }_2<\stackrel{~}{\nu }_2`$, then the gaps have the form $`\stackrel{~}{\gamma }_1=(\stackrel{~}{\mu }_1,\stackrel{~}{\nu }_1),\stackrel{~}{\gamma }_2=(\stackrel{~}{\mu }_2,\stackrel{~}{\nu }_2)`$. Using (3.6-7), (3.24) we have $`m_2^+<0`$ and $`m_1^{}>0`$ in the gap $`\stackrel{~}{\gamma }(T_0)`$. Then we deduce that the Wronskian $`w<0`$ in this gap and $`\mathrm{\#}(T_0,\stackrel{~}{\gamma }(T_0))=0.`$
Assume $`\stackrel{~}{\mu }_1>\stackrel{~}{\nu }_1`$ and $`\stackrel{~}{\mu }_2<\stackrel{~}{\nu }_2`$ and let $`\stackrel{~}{\gamma }(T_0)=(\alpha ^{},\alpha ^+)`$. The function $`w`$ is real analytic on $`(\alpha ^{},\alpha ^+)`$ and relations (3.6-7), (3.24) imply $`w(\alpha ^{})<0,w(\alpha ^+)>0`$. Then the estimate $`\mathrm{\#}(T_0,\stackrel{~}{\gamma }(T_0))2`$ gives $`\mathrm{\#}(T_0,\stackrel{~}{\gamma }(T_0))=1.`$
ii) Using (3.6-7), (3.25) we deduce that $`m_1^{}>0`$ on the interval $`\stackrel{~}{\gamma }(T_0).`$ Let $`\stackrel{~}{\mu }_2<\stackrel{~}{\nu }_2`$. Then (3.6-7), (3.25) yield $`m_2^+<0`$ on $`(\stackrel{~}{\mu }_2,\stackrel{~}{\nu }_2)`$, and the function $`w<0`$ on the interval $`\stackrel{~}{\gamma }(T_0).`$ Let $`\stackrel{~}{\nu }_2<\stackrel{~}{\mu }_20`$. Then $`w(\stackrel{~}{\nu }_2)<0`$ and $`w(\stackrel{~}{\mu }_2)>0`$. Then since $`\mathrm{\#}(T_0,\stackrel{~}{\gamma }(T_0))2`$, we get $`\mathrm{\#}(T_0,\stackrel{~}{\gamma }(T_0))=1.`$
iii) We consider the ground state in the gap $`\stackrel{~}{\gamma }_0(T_0)`$. Using (3.6-7), (3.24) we obtain $`m_2^+<0`$ and $`m_1^{}>0`$ in the gap $`\stackrel{~}{\gamma }_0(T_0)`$. Then the function $`w<0`$ on the interval $`\stackrel{~}{\gamma }_0(T_0)`$ and $`\mathrm{\#}(T_0,\stackrel{~}{\gamma }_0(T_0))=0.`$
Next we obtain biperiodic potentials with a prescribed number of eigenvalues.
Proof of Theorem 2.3. Let $`q_0Q`$ and let $`pL_{even}^2(𝕋)`$. Then due to Theorem 2.2 we have the sequence $`\{d_n\}_1^{\mathrm{}}`$, $`d_n=\mathrm{\#}(T_0,\gamma _n(T_0))\{0,1\},n1`$. Note that Theorem 2.1 yields $`d_n=0`$ if $`|\gamma _n(T_0)|=0`$. Hence we have a mapping $`q_0(p,d)P`$.
Let $`(p,d)P`$. For fixed $`pL_{even}^2(𝕋)`$ the gaps have the forms: $`\gamma _n(H)=(\mu _n(p),\nu _n(p))`$ or $`\gamma _n(H)=(\nu _n(p),\mu _n(p))`$ for all $`n1`$ (see \[GT\]). In order to construct $`p_1`$ we need the following result from \[GT\]: for any $`pL_{even}^2(𝕋)`$ with gaps $`\gamma _n(H)=(\alpha _n^{},\alpha _n^+)`$ there exists unique $`p_1L_{even}^2(𝕋)`$ such that $`\gamma _n(H)=\gamma _n(H_1)`$ and $`\mu _n(p_1)\{\alpha _n^{},\alpha _n^+\}`$ for all $`n1`$.
In order to determine $`p_1`$, firstly, we set $`\gamma _n(H)=\gamma _n(H_1)`$. Secondly, using the number $`d_n\{0,1\}`$, according to Theorem 2.2, we choose the position of $`\mu _n(p_1)`$ by the following: if $`d_n=1`$, then $`\mu _n(p_1)=\nu _n(p)`$ and if $`d_n=0`$, then $`\mu _n(p_1)=\mu _n(p)`$. Thus, we know the gaps $`\gamma _n(H_1)`$ and the position of $`\mu _n(p_1)`$ for all $`n1`$ and by the result of \[GT\], Theorem 2.2, there exists a unique isospectral $`p_1L_{even}^2(𝕋)`$ such that $`\mathrm{\#}(T_0,\gamma _n(T_0))=d_n`$ for all $`n1`$. Hence the mapping $`q_0(p,d)`$ from $`Q`$ into $`P`$ is 1-to-1 and onto.
For isospectral potentials $`p,p_1`$ the identity $`p_1=p`$ holds (see \[M\] or \[K1\]) and the estimates (2.5) are fulfilled (see \[K1\]).
We consider the eigenvalues for the general case. In order to prove Theorem 5.1 we need the following definitions. Suppose that a gap of $`T_t`$ has the form $`\stackrel{~}{\gamma }(T_t)=(\alpha ^{},\alpha ^+)=\gamma _n(H_2)\gamma _m(H_1)\mathrm{}`$ for some gaps $`\gamma _k(H_j)=(\alpha _{k,j}^{},\alpha _{k,j}^+),j=1,2,k=n,m`$. By Lemma 3.4, the equation $`m_2^+(\alpha ^{},y)=m_1^{}(\alpha ^{}),y[0,1]`$ has $`Nn`$ roots $`y_1,\mathrm{},y_N`$. Assume for simplicity $`y_1=0`$ and introduce the function
$$(t,\alpha ^{})=\{\begin{array}{cc}\sqrt{2|M_{n,2}^{}|}L_2(t,\alpha ^{}),\hfill & \text{if }\alpha _{m,1}^{}<\alpha _{n,2}^{},\hfill \\ \sqrt{2|M_{n,2}^{}|}L_2(t,\alpha ^{})/r(\alpha ^{},t),\hfill & \text{if }\alpha _{m,1}^{}=\alpha _{n,2}^{},\hfill \\ \sqrt{2|M_{m,1}^{}|}[\dot{\mathrm{\Psi }}_1(0,\alpha ^{})^2(p_2(t)\alpha ^{})\mathrm{\Psi }_1(0,\alpha ^{})^2],\hfill & \text{if }\alpha _{m,1}^{}>\alpha _{n,2}^{},\hfill \end{array}$$
(5.1)
where $`L_j(t,\lambda )=[\dot{\mathrm{\Psi }}_j(t,\lambda )^2(p_j(t)\lambda )\mathrm{\Psi }_j(t,\lambda )^2]`$ for $`\lambda =\alpha _{k,j}^\pm ,k0,t[0,1]`$. Here $`\mathrm{\Psi }_j(t,\alpha _{k,j}^\pm )`$ is the corresponding real normalized eigenfunction for 2-periodic problem $`y^{\prime \prime }+p_jy=\lambda y`$, and $`\pm M_{k,j}^\pm >0`$ are the corresponding effective masses for $`H_j,j=1,2`$ and
$$r(\lambda ,t)=1+\sqrt{|M_{n,2}^{}/M_{m,1}^{}|}\mathrm{\Psi }_2(t,\lambda )^2/\mathrm{\Psi }_1(t,\lambda )^2,\lambda =\alpha _{m,1}^{}=\alpha _{n,2}^{}.$$
Theorem 5.1. Let $`q_t`$ be a biperiodic potential given by (1.1), where $`p_1,p_2L_{loc}^2()`$. Suppose that a gap $`\stackrel{~}{\gamma }(T_t)=(\alpha ^{},\alpha ^+)=\gamma _n(H_2)\gamma _m(H_1)\mathrm{}`$ for some gaps $`\gamma _k(H_j)=(\alpha _{k,j}^{},\alpha _{k,j}^+),j=1,2,k=n,m`$. Assume that $`\mu _{m,1}(p_1)(\alpha ^{},\alpha _{m,1}^+]`$ and let $`m_2^+(\alpha ^{},0)=m_1^{}(\alpha ^{})`$. Then there exists a unique function $`z()W_1^2(\epsilon ,\epsilon ),z^{}(0)=0`$ for some $`\epsilon >0`$, such that
i) If $`z(t)>0`$, then $`\lambda (t)\alpha ^{}+z(t)^2\stackrel{~}{\gamma }(T_t)`$ is an eigenvalue of $`T_t`$.
If $`z(t)<0`$, then $`\lambda (t)\stackrel{~}{\gamma }(T_t)\mathrm{\Lambda }_0^{(m_0)}`$ is a resonance of $`T_t`$. Here $`m_0=2`$ if $`\alpha _{m,1}^{}=\alpha _{n,2}^{}`$; $`m_0=3`$ if $`\alpha _{m,1}^{}>\alpha _{n,2}^{}`$ and $`m_0=4`$ if $`\alpha _{m,1}^{}<\alpha _{n,2}^{}`$. Moreover, the following asymptotics is fulfilled:
$$z(t)=_0^t(t,\alpha ^{})𝑑t+O(t^{3/2})ast0.$$
(5.2)
ii)If $`p_1,p_2`$ are continuous then $`z()C^1(\epsilon ,\epsilon )`$ and $`z(t)=_0^t(t,\alpha ^{})𝑑t+O(t^2),t0`$.
iii) If $`p_1,p_2`$ are analytic functions, then $`z()`$ is real analytic on $`(\epsilon ,\epsilon )`$.
Remark. Result for an eigenvalue in some neighborhood of $`\alpha _n^+,n0`$, is similar.
Proof. i) There are 3 cases: $`\alpha _{n,2}^{}<\alpha _{m,1}^{},a_{n,2}^{}=a_{m,1}^{},a_{n,2}^{}>a_1^{}`$. Recall $`b_2(\lambda )=\sqrt{\mathrm{\Delta }_2^2(\lambda )1},`$ and
$$\zeta _2(\lambda ,t)=\frac{\dot{\phi }_2(1,\lambda ,t)}{2\phi _2(1,\lambda ,t)},m_2^+(\lambda ,t)=\zeta _2(\lambda ,t)\frac{b_2(\lambda )}{\phi _2(1,\lambda ,t)},\lambda \gamma _2,t[0,1).$$
(5.3)
Define a new parameter $`z`$ by the formula $`z=z(\lambda )=\sqrt{\lambda \alpha ^{}}`$, where $`z(\lambda )>0,\lambda \mathrm{\Lambda }`$. Firstly, let $`\alpha _{n,2}^{}=\alpha ^{}>\alpha _{m,1}^{}`$. By (3.9), the function $`b_{02}(z)b_2(\alpha ^{}+z^2)`$ is analytic in some neighborhood of zero and we get $`b_{02}(z)=z\beta _2^{}(1+O(z^2)),`$ as $`z0`$, where $`\beta _2^{}\mathrm{\Delta }_2(\alpha ^{})\sqrt{2|M_{n,2}^{}|}>0`$. The function $`1/\phi _2(1,\alpha ^{}+z^2,t)`$ is smooth in the parameter $`(z,t)`$ in some neighborhood of zero. Therefore, $`w(\alpha ^{}+z^2,t)`$ is a smooth function of the parameter $`(z,t)`$ in some neighborhood of zero and $`\dot{w}=\dot{m}_2^+`$ (see (3.19)). Moreover, $`w_z(\alpha ^{}+z^2,0)|_{z=0}=\beta _2^{}/\phi _2(1,\alpha ^{},0)0.`$ Then by the implicit function Theorem there exists a function $`z()W_1^2(\epsilon ,\epsilon )`$ for some $`\epsilon >0`$ such that $`w(\alpha ^{}+z^2(t),t)=0`$ and $`z(0)=0`$. We prove (5.2). Let $`\mathrm{\Omega }=1+O(|z|+|t|)`$. Relations (3.19-20), (3.9) yield
$$\dot{w}(\lambda ,t)=\dot{m}_2^+(\lambda ,t)=\dot{\zeta }_2(\lambda ,t)+O(z)=\dot{\zeta }_2(\alpha ^{},t)+O(z),w_z(\lambda ,t)=\frac{\beta _2^{}\mathrm{\Omega }}{\phi _2(1,\alpha ^{},t)},$$
(5.4)
as $`t0`$. Using (5.4), (3.22), we obtain
$$\dot{z}(t)=\frac{\dot{w}(\lambda (t),t)}{w_z(\lambda (t),t)}=\frac{\phi (1,\alpha ^{},t)\dot{\zeta }_2(\alpha ^{},t)+O(z)}{\beta _2^{}\mathrm{\Omega }}=((t,\alpha ^{})+O(z))\mathrm{\Omega }.$$
(5.5)
Let $`z_m=\mathrm{max}|z(y)|,0yt`$. Then integration implies
$$z(t)=_0^t\dot{z}(s)𝑑s=_0^t(s,\alpha ^{})𝑑s+\sqrt{t}p_2O(|z_m|+|t|)+O(t^2+|tz_m|),$$
(5.6)
hence $`z(t)=O(\sqrt{t})`$ and substituting this into (5.6) we get (5.2).
Secondly, let $`\alpha _{n,2}^{}=\alpha ^{}=\alpha _{m,1}^{}`$. By (3.9), the function $`b_{0m}(z)b_m(\alpha ^{}+z^2),m=1,2,`$ is analytic in some neighborhood of zero and we get $`b_{0m}(z)=z\beta _m^{}(1+O(z^2)),z0`$, where $`\beta _m^{}\mathrm{\Delta }_2(\alpha _{})\sqrt{|2M_{m,1}^{}|}0`$. The function $`1/\phi _2(1,\alpha ^{}+z^2,t)`$ is smooth in the parameter $`(z,t)`$ in some neighborhood of zero. Therefore, $`w(\alpha ^{}+z^2,t)`$ is a smooth function of the parameter $`(z,t)`$ in some neighborhood of zero. Moreover,
$$\frac{w(\alpha ^{}+z^2,0)}{z}|_{z=0}=\frac{\beta _2^{}}{\phi _2(1,\alpha ^{},0)}\frac{\beta _1^{}}{\phi _1(1,\alpha ^{})}=\frac{\beta _2^{}r(\alpha ^{},0)}{\phi _2(1,\alpha ^{},0)}0,$$
(5.7)
since $`\beta _1^{}\phi _2/(\beta _2^{}\phi _1)=(\beta _2^{}\mathrm{\Psi }_2)^2/(\beta _1^{}\mathrm{\Psi }_1)^2)`$. Then by the implicit function Theorem there exists a function $`z()W_1^2(\epsilon ,\epsilon )`$ for some $`\epsilon >0`$, such that $`w(\alpha ^{}+z^2(t),t)=0`$ and $`z(0)=0`$. We find the asymptotics of $`z(t)`$ as $`t0`$. Using (3.19-20) we deduce that
$$\dot{w}(\lambda ,t)=\dot{m}_2^+(\lambda ,t)=\dot{\zeta }_2(\lambda ,t)+O(z)=\dot{\zeta }_2(\alpha ^{},t)+O(z),t0,$$
(5.8)
and by (5.7),
$$w_z(\lambda _n(t),t)=\left[\frac{\beta _2^{}}{\phi _2(1,\alpha ^{},0)}+\frac{\beta _1^{}}{\phi _1(1,\alpha ^{})}\right]\mathrm{\Omega }=\frac{\beta _2^{}r(0,\alpha ^{})}{\phi _2(1,\alpha ^{},t)}\mathrm{\Omega }.$$
(5.9)
Then (5.8-9), (3.22) yield
$$\dot{z}(t)=\frac{\dot{w}(\lambda (t),t)}{w_z(\lambda (t),t)}=\frac{\phi _2(1,\alpha ^{},t)\dot{\zeta }_2(\lambda _n(t),t)+O(z)}{r(\alpha ^{},0)\beta _2^{}\mathrm{\Omega }}=((t)+O(z))\mathrm{\Omega }.$$
(5.10)
Then as above we get (5.2).
Next, we consider the last case $`\alpha _{n,2}^{}<\alpha ^{}=\alpha _{m,1}^{}`$. By (3.9), the function $`b_{01}(z)b_1(\alpha ^{}+z^2)`$ is analytic in some neighborhood of zero and we get $`b_{01}(z)=z\beta _1^{}(1+O(z^2)),`$ as $`z0.`$ The function $`1/\phi _2(\alpha ^{}+z^2,t)`$ is smooth in the parameter $`(z,t)`$ in some neighborhood of zero. Therefore, $`w(\alpha ^{}+z^2,t)`$ is a smooth function of the parameter $`(z,t)`$ in some neighborhood of zero. Moreover, $`w_z(\alpha ^{}+z^2,0)|_{z=0}=\beta _1^{}/\phi _1(1,\alpha ^{},0)0.`$ Then by the implicit function Theorem there exists a function $`z()W_1^2(\epsilon ,\epsilon )`$ for some $`\epsilon >0`$ such that $`w(\alpha ^{}+z^2(t),t)=0`$ and $`z(0)=0`$. We find the asymptotics of $`z(t)`$ as $`t0`$. By (3.19-20),
$$\dot{w}(\lambda ,t)=\dot{m}_2^+(\lambda ,t)=(p_2(t)\lambda )m_2^+(\lambda ,t)^2=(p_2(t)\lambda )\zeta _1^+(\alpha ^{})^2+O(z),t0,$$
(5.11)
and by (3.9),
$$w_z(\lambda ,t)=\beta _1^{}\phi _1(1,\alpha ^{},0)^1\mathrm{\Omega }.$$
(5.12)
Hence the asymptotics (5.11-12) and the identity (3.22) imply
$$\dot{z}(t)=\frac{\dot{w}(\lambda _n(t),t)}{w_z(\lambda _n(t),t)}=\frac{\phi _1(1,\alpha ^{})[(p_2(t)\lambda )\zeta _1(\alpha ^{})^2+O(z)]}{\beta _1^{}\mathrm{\Omega }}=((t,\alpha ^{})+O(z))\mathrm{\Omega }.$$
(5.13)
Then, as above we get (5.2).
ii) If $`p_,p_2`$ is continuous, then using (5.5), (5.10), (5.13) we obtain the needed estimates.
iii) If $`p_1,p_2`$ are analytic functions, then by the implicit function Theorem, $`z()`$ is real analytic on $`(\epsilon ,\epsilon )`$. If $`z(t)>0`$ there exists an eigenvalue $`\lambda (t)`$ with the asymptotics (5.2). But if $`z(t)<0`$, then there is no an eigenvalue of $`T_t`$ in the interval $`[\alpha ^{},\alpha ^{}\epsilon ]`$ and we have a resonance.
We study an eigenvalue $`\lambda _0`$ of $`T_0`$ in the 0-th gap $`\gamma _0(T_0)`$.
Proposition 5.2 Let $`q_0`$ be a biperiodic potential given by (1.1). Let $`\gamma _0(H_j)=(\mathrm{},\alpha _{0,j}^+)`$ be the infinite gaps of $`H_j,j=1,2`$ and $`\mathrm{min}\{\alpha _{0,1}^+,\alpha _{0,2}^+\}=0`$. Assume that $`\nu _{0,j}\gamma _0(H_j)`$ is the corresponding Neumann eigenvalue. Then $`\mathrm{\#}(T_0,(\mathrm{},\nu ^0])=0`$, where $`\nu ^0=\mathrm{min}\{\nu _{0,1},\nu _{0,2}\}`$. Let in addition $`w(0)>0,`$ then $`\mathrm{\#}(T_0,\gamma _0(T_0))=1`$.
Proof. If $`\lambda <\nu ^0`$, then by Lemma 3.5, $`w(\lambda )=m_2^+(\lambda )m_1^{}(\lambda )<0`$, which yields the identity $`\mathrm{\#}(T,(\mathrm{},\nu ^0])=0`$. Let $`w(0)>0.`$ Then by (4.5), $`w(\lambda )\mathrm{}`$ as $`\lambda \mathrm{}`$, and there exist a zero of $`w(\lambda ),\lambda <0.`$ But there is no other eigenvalue in the gap since if we have 2, then we get 3 eigenvalues in the gap. This is impossible.
## 6 Dislocations and half-solid
1. Dislocations. We consider the operator $`T_t^{di}=\frac{d^2}{dx^2}+p_{(t)}`$ acting in $`L^2()`$, where the dislocation potential $`p_{(t)}=\chi _{}p+\chi _+p(+t),t`$ and $`pL^1(𝕋)`$. Recall that due to (2.6-7) we have only two sheets $`\mathrm{\Lambda }_0^{(1)},\mathrm{\Lambda }_0^{(2)}`$ of the energy surface. Identities (3.10), (4.10) yield
$$w(\lambda ,t)=m^+(\lambda ,t)m^{}(\lambda ,0)=\frac{a(\lambda ,t)b(\lambda )}{\phi (1,\lambda ,t)}\frac{a(\lambda ,0)b(\lambda )}{\phi (1,\lambda ,0)},\lambda \mathrm{\Lambda }_E(T_0).$$
(6.1)
Proof of Theorem 2.4. We consider the case $`\alpha _n^{}`$ and let $`n`$ be even. The proof for $`\alpha _n^+`$ or for odd $`n`$ is similar. Let $`\mu _n(t)=\mu _n(p,t)`$. Firstly, we consider the simple case $`\alpha _n^{}\mu _n(0)`$. Define a new parameter $`z`$ by the formula $`z=z(\lambda )=\sqrt{\lambda \alpha ^{}}`$, where $`z(\lambda )>0,\lambda \mathrm{\Lambda }_0`$. Recall that the function $`b_0(z)b(\alpha _n^{}+z^2)`$ is analytic in some neighborhood of zero and has asymptotics (3.9). The function $`1/\phi (1,\alpha _n^{}+z^2,t)`$ is smooth in the parameter $`(z,t)`$ in some neighborhood of zero. Therefore, $`w(\alpha _n^{}+z^2,t)`$ is a smooth function of the parameter $`(z,t)`$ in some neighborhood of zero. Moreover, $`w_z(\alpha _n^{}+z^2,0)|_{z=0}=2\beta _n/\phi (1,\alpha _n^{},0)0.`$ Then by the implicit function Theorem there exists a function $`z()W_1^2(\epsilon ,\epsilon )`$ for some $`\epsilon >0`$ such that $`w(\alpha _n^{}+z^2(t),t)=0,t(\epsilon ,\epsilon )`$ and $`z(0)=0`$. We prove (2.8). By (3.19-20),
$$\dot{w}(\lambda _n(t),t)=\dot{m}^+(\lambda _n(t),t)=\dot{\zeta }(\lambda _n(t),t)+O(z)=\dot{\zeta }(\alpha _n^{},t)+O(z),$$
(6.2)
$$w_z(\lambda _n(t),t)=2\beta _n/\phi (1,\alpha _n^{},t)\mathrm{\Omega },\mathrm{\Omega }=1+O(|z|+|t|),$$
(6.3)
as $`t0`$. Recall $`L(t,\lambda )=\dot{\mathrm{\Psi }}(t,\lambda )^2(p(t)\lambda )\mathrm{\Psi }(t,\lambda )^2`$. Using (6.2-3), (3.22) we obtain
$$\dot{z}(t)=\frac{\dot{w}(\lambda _n(t),t)}{w_z(\lambda _n(t),t)}=\frac{\phi (1,\alpha _n^{},t)\dot{\zeta }(\lambda _n(t),t)+O(z)}{2\beta _n\mathrm{\Omega }}=\frac{\beta _n}{2}L(t,\alpha _n^{})\mathrm{\Omega }+O(z).$$
(6.4)
Let $`z_m=\mathrm{max}|z(y)|,0y|t|`$, then integration yields
$$z(t)=_0^t\dot{z}(s)𝑑s=(\beta _n/2)_0^tL(s,\alpha _n^{})𝑑s+p\sqrt{t}O(|z_m|+|t|)+O(t^2+|tz_m|).$$
(6.5)
Hence $`z(t)=O(|t|^{1/2})`$ and then $`z_m=O(|t|^{1/2})`$ and (6.6) yields (2.8).
Secondly we consider the more complicated case $`\mu _n(0)=\alpha _n^{}`$. Recall $`\mu _n(t)=\mu _n(p,t)`$. Let $`a^0=a(\lambda ,0),\vartheta _x^0=\vartheta _x(1,\lambda ,0)`$. If we multiply (6.1) by $`(a+b)`$ and $`(a^0b)`$ and use (3.11) we obtain $`\vartheta _x(a^0b)+\vartheta _x^0(a+b)=0`$. Therefore, we have the following equation for the eigenvalues:
$$\mathrm{\Phi }(\lambda ,t)b(\vartheta _x+\vartheta _x^0)+(a\vartheta _x^0a^0\vartheta _x)=0.$$
(6.6)
Recall that $`\phi (1,\lambda ,t)=(\lambda \mu _n(t))\varphi (\lambda ,t)`$. Using (3.17) we have
$$\dot{\phi }=\dot{\mu }_n\varphi +(\lambda \mu _n)\dot{\varphi },\ddot{\phi }=\ddot{\mu }_n\varphi 2\dot{\mu }_n\dot{\varphi }+(\lambda \mu _n)\ddot{\varphi },$$
(6.7)
$$\phi (1,\lambda ,t)=O(|z|^2+t^2),\dot{\phi }(1,\lambda ,t)=O(|z|^2+|t|),\mathrm{as}t0,$$
(6.8)
(6.8) yields $`\mathrm{\Phi }_z(\alpha _n^{},0)=2\beta _n\vartheta _x(1,\alpha _n^{},0)0`$ and the function $`\mathrm{\Phi }(z,t)`$ is analytic in $`z`$ and continuous in $`t`$. Then by the implicit function Theorem, there exists a function $`z()W_1^2(\epsilon ,\epsilon )`$ for some $`\epsilon >0`$ such that $`\mathrm{\Phi }(z(t),t)=0`$ and $`z(0)=0.`$ We will prove (2.9) and we have
$$\mathrm{\Phi }_z(z,t)=2\beta _n\vartheta _x(1,\alpha _n^{},0)+O(|z|+|t|),\mathrm{as}t0,$$
(6.9)
and due to (3.12) we get
$$\dot{\mathrm{\Phi }}(z,t)=(\beta a^0)\dot{\vartheta }_x+\dot{a}\vartheta _x^0=(\beta a^0)(\lambda p)\dot{\phi }\vartheta _x\vartheta _x^0(\lambda p)\phi \vartheta _x,$$
(6.10)
$$\dot{\mathrm{\Phi }}(z,t)=\vartheta _x(1,\alpha _n^{})^2+p(t)O(t^2+z^2)+O(|t|+|z|)$$
(6.11)
Note that $`\vartheta _x(\mu _n(0),0)0`$ and by (3.11), $`(b^2)_\lambda =\vartheta _x(1,\lambda ,0)\phi _\lambda (1,\lambda ,0),`$ at $`\lambda =\alpha _n^{}`$. Then (3.17) yields
$$\vartheta _x(1,\alpha _n^{},0)=2M_n^{}/\phi _\lambda (1,\alpha _n^{},0)=\ddot{\mu }_n(0)\varphi /2>0.$$
(6.12)
Using (3.12), we have $`\ddot{\phi }(1,\alpha _n^{},0)=2\beta _n^2\dot{\mathrm{\Psi }}(0,\alpha _n^{})^2`$ and (6.9) yields $`\ddot{\phi }(1,\alpha _n^{},0)=\ddot{\mu }_n(0)\varphi `$. Therefore, we get $`\vartheta _x^0/2\beta _n=(\beta _n/2)\dot{\mathrm{\Psi }}(0,\alpha _n^{})^2`$. Using (6.8-12) we obtain
$$z(t)=_0^t\frac{\dot{\mathrm{\Phi }}(\lambda _n(t),t)}{\mathrm{\Phi }_z(\lambda _n(t),t)}𝑑t=t\frac{\vartheta _x^0}{2\beta _n}+_0^t[|p(t)|O(t^2+z^2)+O(|t|+|z|)]𝑑t$$
(6.13)
and since $`pL^2(0,1)`$, (6.13) implies
$$z(t)=t(\beta _n/2)\dot{\mathrm{\Psi }}(0,\alpha _n^{})^2+\sqrt{|t|}(p)O(|z_m|^2+t^2))+O(|z_mt|+t^2).$$
(6.14)
Therefore, $`z(t)=t(\beta _n/2)\dot{\mathrm{\Psi }}(0,\alpha _n^{})^2+\sqrt{|t|}O(|z_m|),`$ and $`z_m(t)=O(|t|^{1/2})`$. Substituting these into (6.14) we have (2.9). Note that $`\lambda _n(t)=\alpha ^{}+z(t)^2`$ while $`z()`$ has the asymptotics (2.9) and the function $`a+b`$ has a zero $`\mu _n(t)`$ in the neighborhood of zero. It is important that $`\lambda _n(t)`$ and $`\mu _n(t)`$ have different asymptotics, see (2.9) and (3.17).
Assume that $`p`$ is an analytic function. Hence by the implicit function theorem, there exists a unique analytic function $`z_n^{}(t)`$ of the equation $`w(t,z)=0`$ for all $`t(\epsilon ,\epsilon )`$ for some small $`\epsilon >0.`$ Then $`z(t)=rt^m(1+O(t)),`$ for some $`r0,m1.\text{ }`$
We prove the existence of two eigenvalues in the gap for some potentials.
Proof of Theorem 2.5. In Lemma 3.4 we have proved that for any finite sequences $`\{d_n\}^N,\{s_n\}^N,d_n\{0,1\},s_n>0,N1`$ there exists an even potential $`pC^2(𝕋)`$ such that $`\gamma _n(p)=s_n,`$ and $`L(0,\alpha _n^\pm )>0,`$ if $`d_n=1`$ and $`L(0,\alpha _n^\pm )<0,`$ if $`d_n=0`$ for $`n=1,\mathrm{},N.`$ We take the dislocation potential in the form $`p_{(t)}=\chi _{}p+\chi _+p(+t)`$. Then by Theorem 2.4, there exists $`\epsilon >0`$ such that $`|\gamma _n(T_t^{di})|=s_n,`$ and $`\mathrm{\#}(T_t^{di},\gamma _n(T_t^{di}))=2d_n`$ for all $`n=1,2,..,N,`$ and each $`t(0,\epsilon )`$.
2. Half-solid. We consider the half-solid operator $`T_t^s=\frac{d^2}{dx^2}+q_t^s(x)`$ acting in $`L^2()`$, where the potential $`q_t^s(x)=s\chi _{}+\chi _+p(+t)`$ and $`s,t`$, $`pL^1(𝕋)`$ is real. Assume $`\alpha _0^+(H)=0`$. We have the following simple results concerning $`\sigma (T_t^s).`$ If $`s\alpha _1^{}(H)`$, then $`\sigma _{ac}(T_t^s)=(s_{},\mathrm{})`$, where $`s_{}=\mathrm{min}(0,s)`$. If $`s>\alpha _1^{}`$ then there is a gap in the spectrum of $`T_t^s.`$ Our goal is to study the eigenvalues in the gaps $`\gamma _n(T_t^s),n0,`$ and to find how these eigenvalues depend on $`t,s.`$ It is clear that they depend on $`t`$ periodically. Using (4.6) we rewrite the Wronskian $`w(z,t)`$ in the form
$$w(\lambda ,t)=m^+(\lambda ,t)\sqrt{s\lambda }=\frac{a(\lambda ,t)b(\lambda )}{\phi (1,\lambda ,t)}\sqrt{s\lambda },\lambda <s,$$
(6.15)
where $`\sqrt{s\lambda }>0`$ if $`\lambda <s,\lambda \mathrm{\Lambda }_0^{(1)}`$.
Proof of Theorem 2.6. i) Using Theorem 2.2 with $`p_1=s,p_2=p`$, we obtain (2.12) and the identity $`\mathrm{\#}(T_0^s,\gamma _0(T_0^s))=0.`$ By Lemma 4.3, the Wronskian $`w`$ on the second sheet $`\mathrm{\Lambda }^{(4)}`$ has the form $`w(\lambda ^{(4)})=b(\lambda )\phi (1,\lambda )^1\sqrt{s\lambda }`$. Repeating the proof of Theorem 2.2 we get $`\mathrm{\#}^{(4)}(T_0^s,\gamma _n(T_0^s))=1\mathrm{\#}(T_0^s,\gamma _n(T_0^s))`$. Moreover, (2.4) yields (2.12).
ii) Let $`q_0^sQ_N,N0`$ and $`pL^2(𝕋)`$ be even. Then due to Theorem 2.2 we have the sequence $`\{d_n\}_1^N`$, $`d_n=\mathrm{\#}(T_0^s,\gamma _n(T_0^s)),n1`$, and $`\epsilon (0,1)`$ and $`r\mathrm{}^2`$. Note that Theorem 2.1 yields $`d_n=0`$ if $`|\gamma _n(T_0)|=0`$. Hence we have a mapping $`q_0^s(r,d,\epsilon )P`$.
Let $`(r,d,\epsilon )P_N,N0`$. Define the signed gap length $`l_n(p)=(\mu _n(p),\nu _n(p)),n1`$ (see \[GT\]). For the sequence $`r\mathrm{}^2`$, there exists an even potential $`pL^2(𝕋)`$ such that $`|l_n(p)|=r_n0,n=1,..,N,l_n(p)=r_n,nN+1.`$ In order to get uniqueness of $`p`$ we need to determine $`signl_n(p),n=1,..,N`$, (see \[GT\]).
Fix a number $`n=1,..,N`$ and the gap $`\gamma _n(H)=(\alpha _n^{},\alpha _n^+)\mathrm{}`$. We have 2 cases. First, if $`d_n=1`$, then using (2.12) we obtain $`l_n(p)>0`$. Second, if $`d_n=0`$, then using (2.12) we obtain $`l_n(p)<0`$. Hence we have $`\{l_n\}\mathrm{}^2`$ and there exists a unique $`pL^2(𝕋)`$. Using $`\epsilon (0,1)`$ and the definition of $`\epsilon `$ we obtain $`s`$, which yields a unique $`q_0^sQ_N`$.
By Lemma 3.4, for $`n1`$, Eq. (2.13) has $`Nn`$ roots $`y_1,y_2,\mathrm{},y_N[0,1)`$. The case $`n1`$ arises for the gap $`\gamma _n(T_t(s)),n1,`$ and $`n=0`$ corresponds the basic gap $`\gamma _0(T_t^s)`$.
Proof of Theorem 2.7 i) We consider the case $`\alpha _n^{}<s`$. The proof for $`\alpha _n^+<s`$ is similar. The proof follows from i) of Theorem 5.1. In this case the Weyl function has the form (6.16) and $`(y,\lambda )=[\mathrm{\Psi }^{}(y,\lambda )^2+(\lambda p(y))\mathrm{\Psi }(y,\lambda )^2],\lambda =\alpha _n^{}`$. The identity (3.5) yields $`\zeta (y,\alpha _n^{})=\mathrm{\Psi }^{}(y,\alpha _n^{})/\mathrm{\Psi }(y,\alpha _n^{})`$. Together with (6.16) this implies $`w(\alpha _n^{},y)=0.`$ All conditions of Theorem 5.1, i) are fulfilled and we get $`z_n^\pm (t)`$ with the needed properties. By (2.13), $`\mathrm{\Psi }_y(y,\lambda )=\sqrt{s\lambda }\mathrm{\Psi }(y,\lambda )`$, then $`2(y,\alpha _n^{})=(p(y)s)\mathrm{\Psi }(y,\alpha _n^{})^2`$ and (5.2) yields (2.14).
ii) In this case we use i) of Theorem 5.1, when $`s`$ is the end of the gap $`(\alpha _n^{},s)`$ of the operator $`T_y(s)`$. We have $`\mathrm{\Psi }_11,\mathrm{\Psi }_1^{}0,M_1^+=1/2`$, then $`2(y,s)=(p(y)s)`$ and Theorem 5.1 (see the remark after this Theorem) implies the needed results including (2.15).
Now we consider eigenvalues (the ground state) in the gap $`\gamma _0(T_t^s)`$. The existence of eigenvalue is connected with some property of the function $`m^+(0,t),t[0,1]`$, when $`b>0,\phi >0`$ in the gap $`\gamma _0(T_t^s)`$. We have the following result.
Theorem 6.1 Let $`T_0^s=\frac{d^2}{dx^2}+s\chi _{}(x)+\chi _+(x)p(x)`$, where $`pL^1(𝕋)`$ is real and let $`\alpha _0^+=0`$. If $`m^+(0)<0`$, then $`\mathrm{\#}(T_0^s,\gamma _0(T_0^s))=0`$ and if $`m^+(0)>0`$, then
$$\mathrm{\#}(T_0^s,\gamma _0(T_0^s))=\{\begin{array}{cc}0,\hfill & \text{if }s\nu _0\mathrm{or}sm^+(0)^2\hfill \\ 1,\hfill & \text{if }\nu _0<s<m^+(0)^2\hfill \end{array}.$$
(6.16)
Proof. Let $`m^+(0)<0`$, then Lemma 3.5 yields $`w(\lambda )<0`$ for $`\lambda <\mathrm{min}\{s,0\}`$. Hence eigenvalues in $`\gamma _0(T_0^s)`$ are absent.
Let $`m^+(0)>0`$. Assume $`\lambda <s\nu _0`$. Again by Lemma 3.5, $`m^+(\lambda )<0`$ for $`\lambda <\nu _0`$. Then $`w(\lambda )<0`$ for $`\lambda <s\nu _0`$, and eigenvalues in $`\gamma _0(T_0^s)`$ are absent. Now, assume $`m^+(0)^2s.`$ Let the function $`w`$ have a simple zero $`\lambda _1(\nu _0,0)`$. Then for some $`s_0>s`$ the function $`m^+(\lambda )\sqrt{s_0\lambda }`$ has a multiple root $`\lambda _1(\nu _0,0)`$. But this is impossible. Then the first line in (6.17) is true. Consider the case $`\nu _0<s<m^+(0)^2.`$ Then by Lemma 3.5, $`w(\nu _0)<0`$ and $`w(\lambda _1)>0,\lambda _1=\mathrm{min}\{s,0\}`$. Hence Proposition 5.2 yields the second line in (6.17).
References
\[A\] Anoshchenko, O.: The inverse scattering problem for the Schrödinger equation with a potential that has periodic asymptotics. J. Soviet Math. 48 (1990), no. 6, 662–668
\[A1\] Anoshchenko, O.: Eigenfunction expansion of the Schrödinger equation with a potential that has periodic asymptotics. J. Soviet Math. 49 (1990), no. 6, 1237-1241
\[BS\] Bikbaev R., Sharipov R.: Asymptotics as $`t\mathrm{}`$ of the Cauchy problem for the Korteveg-de Vries equation in the class of potentials with finite-gap behavior at $`x\pm \mathrm{}`$, Theoret. and Math. Phys. 78, 1989.
\[CL\] Coddington, E., Levinson, N.: Theory of ordinary differential equations. McGraw-Hill Book Company, Inc., New York-Toronto-London, 1955.
\[DS\] Davies E., Simon B. Scattering theory for systems with different spatial asymptotics on the left and right, Commun. Math. Phys. 63, 277-301, 1978
\[GT\] Garnett J., Trubowitz E.: Gaps and bands of one dimensional periodic Schrödinger operator. Comment. Math. Helv. 59, 258-312 (1984).
\[KK1\] Kargaev P., Korotyaev E.: The Inverse Problem for the Hill Operator, the Direct Approach, Invent. Math., 129, no. 3, 567-593 (1997)
\[KK2\] Kargaev P., Korotyaev E.: Effective masses and conformal mappings. Commum. Math. Phys. 169, 597-625 (1995).
\[K1\] Korotyaev E.: Estimates for the Hill operator, I. J. Diff. Eq. 162, 2000, 1-26
\[K2\] Korotyaev E.: Inverse Problem and the trace formula for the Hill Operator, II, Math. Z., 231, 345-368 (1999)
\[K3\] Korotyaev E.: Lattice dislocations in 1-dimensional model. Commun. Math. Phys. 213, 471-489, 2000
\[KP\] Korotyaev E., Pokrovski, A.: One dimensional half - crystal (in preparation)
\[Kr\] Krein M.: Theory of selfadjoint extensions of semi- bounded Hermitian operators and its applications. I. Mat. Sbornik 20, No. 1, 1-95, 1947.(Russian)
\[L\] Levitan B.M.: The Inverse Sturm-Liouville Problems. Moscow, Nauka, 1984.(Russian)
\[M\] Marchenko V.: Sturm-Liouville operator and applications. Basel: Birkhäuser 1986.
\[Mos\] Moser J. : An Example of a Scrödinger operator with almost periodic potential and nowhere dense spectrum. Comment. Math. Helv. 56, 198-224 (1981).
\[Ta\] Tamm I. Phys.Z. Sowjet.1, 1932, 733.
\[T\] Titchmarsh E.: Eigenfunction expansions associated with second-order differential equations 2, Clerandon Press, Oxford, 1958.
\[Tr\] Trubowitz E.: The Inverse problem for Periodic Potentials . Commun. on Pure and Applied Math. V. 30, 321-337, 1977.
\[PTr\] Pöschel, P.; Trubowitz E.: Inverse Spectral Theory. Boston: Academic Press, 1987.
\[Z\] Zheludev V. On the spectrum of Schrödinger operator with periodic potentials on the half-line. Trudy kafedry mat. anal. Kaliningarad Univ. 1969. |
warning/0507/cs0507013.html | ar5iv | text | # An 𝑂(𝑛log𝑛)-Time Algorithm for the Restricted Scaffold Assignment
## 1 Introduction
Consider two finite sets of points $`S`$ and $`T`$ with total cardinality $`n`$. The objective of the assignment problem is to establish a correspondence between the points in $`S`$ and the points in $`T`$, such that each point in $`S`$ corresponds to exactly one point in $`T`$, and each point in $`T`$ corresponds to at least one point in $`S`$. This correspondence is measured by a cost function $`\delta `$ that assigns a cost $`\delta (s,t)`$ to each assigned pair $`(s,t)`$. The cost of an assignment is the sum of the costs of all assigned pairs. The goal of the assignment problem is to find an assignment of minimum cost.
The general assignment problem is also known as the many-to-one assignment problem. The one-to-one version of the assignment problem requires that each point in $`S`$ maps to exactly one point in $`T`$ and each point in $`T`$ gets mapped exactly one point in $`S`$. Throughout the paper, whenever we talk about the assignment problem, we refer to the many-to-one version of the problem.
The simplest version of the assignment problem assumes that the points in $`S`$ and $`T`$ lie on a line and the cost function is the $`L_1`$ metric. In this setting, the one-to-one assignment problem has a simple $`O(n\mathrm{log}n)`$ time solution when $`|S|=|T|`$: first sort the points in $`O(n\mathrm{log}n)`$ time, then map the $`k^{th}`$ point in $`S`$ to the $`k^{th}`$ point in $`T`$ in $`O(n)`$ time . However, the situation $`|S|<|T|`$ arises in many practical applications. This situation was first addressed by Karp and Li , who provided an $`O(n\mathrm{log}n)`$ time algorithm for the one-to-one assignment problem ($`O(n)`$ time, if $`S`$ and $`T`$ are given in sorted order). Simpler and equally efficient solutions have later been provided in .
Eiter and Mannila studied the assignment problem in the context of measuring the distance between two theories expressed in a logical language. They showed that for points in arbitrary dimensions, this problem has a polynomial time solution. When restricted to points on a line, a minimum cost assignment can be used in measuring the similarity between musical rhythms. In this context, Toussaint proposed the use of the directed swap distance as a similarity measure. If the onsets of a rhythm are represented as points on a line separated by “silence” intervals, the directed swap distance between two rhythms with onset sets $`S`$ and $`T`$ is precisely the cost of an optimal assignment between $`S`$ and $`T`$, with underlying cost function $`L_1`$.
The assignment problem also appears in the shape of the restriction scaffold assignment problem in computational biology . The goal here is to establish a correspondence between sparse experimental data and a restricted set of known structural building blocks. Ben-Dor et. al. model the restriction scaffold assignment as an assignment problem for points on a line, and provide an $`O(n\mathrm{log}n)`$ time algorithm to solve this problem. However, as later shown by Colannino and Toussaint , this algorithm fails to always produce a minimum cost assignment. Thus, the best existing solution to the assignment problem in one dimension is the $`O(n^2)`$ algorithm presented in .
In this paper, we show that the assignment problem with underlying cost function $`L_1`$ in one dimension can be solved in $`O(n\mathrm{log}n)`$ time ($`O(n)`$ if the points in $`S`$ and $`T`$ are given in sorted order). Our algorithm is a simple extension of the $`O(n\mathrm{log}n)`$ time algorithm of Karp and Li for finding the minimum cost one-to-one assignment over $`T`$ and all subsets $`S^{}S`$ of size $`|T|`$, assuming $`|S|>|T|`$. We present our algorithm in Section 4, after a few preliminary results (Section 2.1) and a close look at some properties of an optimal solution (Section 3).
## 2 Background
Let $`S=\{s_0,s_1,s_2,\mathrm{}\}`$ and $`T=\{t_0,t_1,t_2,\mathrm{}\}`$ be two finite sets of points that lie on a horizontal line, with $`|S|+|T|=n`$ and $`|S|>|T|`$. For any $`sS`$ and $`tT`$, the cost $`\delta (s,t)`$ of an assigned pair $`(s,t)`$ is the absolute value of the difference between the $`x`$-coordinates of $`s`$ and $`t`$. To avoid overloading the notation, we use the same symbol for a point and its $`x`$-coordinate. Thus, $`\delta (s,t)=|st|`$. We assume that $`s_i<s_{i+1},0i<|S|1`$ and $`t_j<t_{j+1},0j<|T|1`$.
An assignment $`𝒜`$ between $`S`$ and $`T`$ consists of pairs of points $`(s,t)`$ (henceforth edges), with $`sS`$ and $`tT`$, such that each point in $`S`$ belongs to exactly one edge in $`𝒜`$, and each point in $`T`$ belongs to at least one edge in $`𝒜`$. The cost of $`𝒜`$ is
$$cost(𝒜)=\underset{(s,t)𝒜}{}\delta (s,t)$$
Our goal is to find an assignment $`𝒜`$ of minimum cost. If two points in $`ST`$ have the same $`x`$-coordinate, we can slightly shift one of them to the left or right. If the minimum cost assignment is unique and the change is sufficiently small, this change will not affect the optimal assignment. If there are several assignments with the same optimal cost, at least one of them will be the optimal solution of the new point set. So we may assume without loss of generality that all points in $`ST`$ are distinct.
### 2.1 Preliminaries
For any $`sS`$ and $`tT`$, the value $`|st|`$ can be expressed in a different way as follows. Define a function $`f_{s,t}`$ to be $`1`$ in the interval between $`s`$ and $`t`$ and $`0`$ at any other point (see Figure 1). Then $`|st|=_{\mathrm{}}^+\mathrm{}f_{s,t}(x)𝑑x`$.
The cost of an assignment $`𝒜`$ is therefore
$$cost(𝒜)=\underset{(s,t)𝒜}{}_{\mathrm{}}^+\mathrm{}f_{s,t}(x)𝑑x=_{\mathrm{}}^+\mathrm{}\underset{(s,t)𝒜}{}f_{s,t}(x)dx$$
If we define
$$f_𝒜(x)=\underset{(s,t)𝒜}{}f_{s,t}(x)$$
then the value $`f_𝒜(a)`$ is simply the number of edges in $`𝒜`$ pierced by the vertical line $`x=a`$, and the cost of $`𝒜`$ is
$$cost(𝒜)=_{\mathrm{}}^+\mathrm{}f_𝒜(x)𝑑x$$
(1)
Our definition of $`f_𝒜`$ is similar in nature to the height function $`H:`$ introduced by Karp and Li . Informally, they define $`H(a)`$ at each point $`a`$ as the difference between the number of points in $`S`$ and the number of points in $`T`$ restricted to the interval $`(\mathrm{},a]`$ (or equivalently, to the left of the vertical line $`x=a`$). Thus $`H`$ remains constant throughout each interval that does not contain a point in $`ST`$. Figure 2 shows the stair-shaped curve of $`H`$ for a small example. Note that up transitions in the curve correspond to points in $`S`$ and down transitions correspond to points in $`T`$. We refer to the value $`H(x)`$ as the height of $`x`$. Note that $`H(\mathrm{})=|S||T|`$.
###### Lemma 1
If $`|S|=|T|`$, then $`_{\mathrm{}}^+\mathrm{}|H(x)|𝑑x`$ is the cost of the assignment that assigns the $`k^{th}`$ largest element of $`S`$ to the $`k^{th}`$ largest element of $`T`$.
Proof: Follows immediately from (1) and the fact that, for this particular assignment, $`f_𝒜(x)=|H(x)|`$ at each point $`x`$.
Figure 3a shows an assignment for two sets $`S`$ and $`T`$, with $`|S|=|T|`$. The cost of this assignment is equal to the area shaded in Figure 3b, which is precisely the value of the integral $`_{\mathrm{}}^+\mathrm{}|H(x)|𝑑x`$.
## 3 Properties of a Minimum Cost Assignment
Our algorithm for computing a minimum cost assignment $`𝒜`$ exploits several important properties of $`𝒜`$, which we discuss next. A crossing is defined by a pair of edges $`(a,d)`$ and $`(b,c)`$ such that $`a<b`$ in $`S`$ and $`c<d`$ in $`T`$.
###### Lemma 2
There exists a minimum cost assignment with no crossings.
Proof: Let $`𝒜`$ be a minimum cost assignment between $`S`$ and $`T`$ with a minimum number of crossings. If $`𝒜`$ has zero crossings, the proof is finished. Otherwise, pick two crossing edges $`(a,d)`$ and $`(b,c)`$ in $`𝒜`$, with $`a<b`$ in $`S`$ and $`c<d`$ in $`T`$. We show that $`𝒜^{}=𝒜\{(a,d),(b,c)\}\{(a,c),(b,d)\}`$ is an assignment with $`cost(𝒜^{})cost(𝒜)`$, a contradiction. In particular, we show that $`f_𝒜^{}(x)f_𝒜(x)`$ at each point $`x`$; then $`cost(𝒜^{})cost(𝒜)`$ follows immediately from (1).
First note that $`f_𝒜^{}(x)f_𝒜(x)`$ is true for any $`x`$ such that the vertical line $`L`$ at $`x`$ intersects neither of $`(a,d)`$ and $`(b,c)`$. Suppose now that $`L`$ intersects $`(a,c)`$. Then $`L`$ must also intersect either $`(a,d)`$ (see Figure 4a) or $`(b,c)`$ (see Figure 4b) or both (see Figure 4c). Similarly, if $`L`$ intersects $`(b,d)`$, then $`L`$ also intersects at least one of $`(a,d)`$ and $`(b,c)`$. Furthermore, if $`L`$ intersects both $`(a,c)`$ and $`(b,d)`$, then $`L`$ also intersects both $`(a,d)`$ and $`(b,c)`$ (see Figure 4c). It follows that $`f_𝒜^{}(x)f_𝒜(x)`$.
An assignment $`𝒜`$ can also be regarded as a function $`𝒜:ST`$ such that $`𝒜(s)=t`$ for each $`(s,t)𝒜`$. For any $`tT`$, let $`𝒜^1(t)`$ denote the set of elements $`sS`$ such that $`𝒜(s)=t`$. For each point $`sS`$, define the nearest neighbor $`N(s)`$ to be point in $`T`$ closest to $`s`$, i.e, $`|N(s)s||ts|`$ for any $`tT`$. In the case of a tie, $`N(s)`$ is arbitrarily picked from among the two candidate neighbors.
###### Lemma 3
Let $`𝒜`$ be optimal and let $`tT`$ be such that $`𝒜^1(t)`$ contains two or more elements. Then for each $`s𝒜^1(t)`$, $`t`$ is a nearest neighbor of $`s`$. Furthermore, $`T`$ contains no points in between $`s`$ and $`t`$.
Proof: Assume to the contrary that there is $`sS`$ with $`𝒜(s)=t,|𝒜^1(t)|>1,`$ and $`N(s)t`$. Refer to Figure 5. Define a new assignment $`𝒜^{}`$ with $`𝒜^{}(s)=N(s)`$ and $`𝒜^{}(x)=𝒜(x)`$ for $`xs`$. Note that $`𝒜^{}`$ is also an assignment: $`𝒜^1(t)`$ contains at least one point. Also $`cost(𝒜^{})=cost(𝒜)|st|+|sN(s)|`$ (see Figures 5a and 5b).
Since $`|sN(s)|<|st|`$, it follows that $`cost(𝒜^{})<cost(𝒜)`$, contradicting the fact that $`𝒜`$ is of minimum cost. Thus, $`t`$ is a nearest neighbor of $`s`$.
The claim that $`T`$ contains no points in between $`s`$ and $`t`$ is immediate: if such a point $`t_1T`$ existed, then $`|st_1|<|st|`$, contradicting the fact that $`N(s)=t`$.
Observe that for any subset $`RS`$ of size $`|R|=|S||T|`$, there is a unique minimum cost assignment (with no crossings) from $`SR`$ to $`T`$. Let $`𝒜_{SR}`$ denote the edges of such an assignment, and define a new assignment $`𝒜_R:ST`$ as follows:
$$𝒜_R(x)=\{\begin{array}{cc}N(x)\hfill & \text{if }xR\text{,}\hfill \\ y\hfill & \text{if }xSR\text{ and }(x,y)𝒜_{SR}\hfill \end{array}$$
(2)
Lemma 3 implies that there always exists a subset $`R`$ such that $`𝒜_R`$ defines a minimum cost assignment from $`S`$ to $`T`$. Furthermore, $`R`$ satisfy a special height condition, stated in the lemma below.
###### Lemma 4
There exists a subset $`RS`$ with $`|R|=|S||T|`$ such that $`𝒜_R`$ defines a minimum cost assignment from $`S`$ to $`T`$, and the $`k^{th}`$ smallest element of $`R`$ has height $`k`$.
Proof: Let $`𝒜:ST`$ define a minimum cost assignment. We prove the existence of $`𝒜_R`$ by constructing a set $`RS`$ with the properties stated in this lemma. Initially $`R`$ is empty. If $`|𝒜^1(t)|=1`$ for all $`tT`$, then $`R`$ is empty and the proof is finished. Otherwise, we process points $`tT`$ for which $`𝒜^1(t)`$ has two or more elements. For each such point we consider two cases, as depicted in Figure 6. If all points in $`𝒜^1(t)`$ are less than $`t`$, then we add to $`R`$ all but the largest (rightmost) point in $`𝒜^1(t)`$ (see Figure 6a). Otherwise, we add to $`R`$ all points in $`𝒜^1(t)`$ except for the smallest (leftmost) point greater than $`t`$ (see Figure 6b).
We now define $`A_R`$ as in (2). Since $`𝒜_R`$ is identical to $`𝒜`$, $`𝒜_R`$ is a minimum cost many-to-one assignment from $`S`$ to $`T`$.
It remains to show that the $`k^{th}`$ smallest element of $`R`$ has height $`k`$. To see this, first consider the smallest element of a nonempty set $`𝒜^1(t)R`$. Call this element $`r`$ and suppose it is the $`k^{th}`$ smallest element of $`R`$. It follows then that (i) $`R`$ contains $`k1`$ points less than $`r`$, and (ii) $`T`$ and $`SR`$ contain an equal number of elements less than $`r`$. This latter claim follows from Lemma 3, which tells us that $`T`$ contains no elements in between $`r`$ and $`t`$, and the following observation: the way in which we have selected $`R`$ ensures that if $`t`$ lies to the left of $`r`$ (i.e., $`t<r`$), the assigned item for $`t`$ in $`S/R`$ lies to the left of $`r`$, and if $`t`$ lies to the right of $`r`$ ($`t>r`$), the assigned item for $`t`$ in $`S/R`$ lies to the right of $`r`$. These together imply that $`H(r)=k`$.
We now show that the points in $`𝒜^1(t)\{r\}`$ have height values $`k+1,k+2,\mathrm{}`$, in order from smallest to largest. By Lemma 3, $`T`$ contains no points in between $`s`$ and $`t`$, for each $`s𝒜^1(t)`$. Then the points in $`R𝒜^1(t)`$ have incrementally increasing height values. It follows that the height of the $`k^{th}`$ smallest element of $`R`$ is $`k`$.
Let $`H_R`$ represent the height function restricted to sets $`SR`$ and $`T`$. This means that for each $`x`$, $`H_R(x)`$ is the difference between the number of points in $`SR`$ and the number of points in $`T`$ restricted to the interval $`(\mathrm{},x]`$.
###### Lemma 5
The cost of assignment $`𝒜_R`$ is
$$\underset{rR}{}|rN(r)|+_{\mathrm{}}^+\mathrm{}|H_R(x)|𝑑x$$
(3)
Proof: By Lemma 1 we have that the contribution of $`SR`$ to the cost of $`𝒜_R`$ is $`_{\mathrm{}}^+\mathrm{}|H_R(x)|𝑑x`$. Since each point in $`R`$ maps to its nearest neighbor, the contribution of $`R`$ to the cost of $`𝒜_R`$ is $`_{rR}|rN(r)|`$. These together conclude the lemma.
###### Theorem 6
Let $`RS`$ be a subset of size $`|R|=|S||T|`$ with two properties:
1. The $`k^{th}`$ smallest element of $`R`$ has height $`k`$.
2. $`R`$ minimizes the quantity from (3).
Then $`𝒜_R`$ defines a minimum cost assignment from $`S`$ to $`T`$.
Proof: By Lemma 4, we know that there exists a set $`R`$ that satisfies (i). By Lemma 3, $`R`$ satisfies (ii). It follows that $`𝒜_R`$ is a minimum cost assignment from $`S`$ to $`T`$.
## 4 Computing a Minimum Cost Assignment
Theorem 6 gives an exact description of the set $`R`$ that yields a minimum cost assignment $`𝒜_R`$. We now turn to the problem of efficiently determining this set. With this goal in mind, we introduce the following notation. For any point $`x`$ and any integer $`k`$, define the relative height of $`x`$ with respect to $`k`$ as
$$h^k(x)=\{\begin{array}{cc}1\text{,}\hfill & \text{if }H(x)k\hfill \\ 1\text{,}\hfill & \text{if }H(x)<k\hfill \end{array}$$
Observe that when a point $`s`$ is removed from $`S`$, $`H(x)`$ decreases by 1 for all $`x>s`$. Suppose that $`H(s)=k`$, and let $`m`$ be the largest point in $`ST`$. The removal of $`s`$ causes the area under the height function between $`s`$ and $`m`$ to decrease by the quantity $`_s^mh^k(x)𝑑x`$. We use this observation to define the profit of removing $`s`$ from $`S`$ and placing it in $`R`$ (recall that $`𝒜_R`$ assigns each item in $`R`$ to its nearest neighbor), as follows:
$$P(s)=_s^mh^k(x)𝑑x|sN(s)|$$
(4)
The profit function quantifies the benefit of placing $`s`$ in $`R`$, the goal being to minimize the cost of the assignment defined by $`𝒜_R`$. The integral term in (4) represents the effect of excluding $`s`$ from the one-to-one assignment from $`SR`$ to $`T`$, as depicted in Figure 7. The term $`|sN(s)|`$ in (4) represents the cost of assigning $`s`$ to its nearest neighbor. We minimize the cost of the assignment defined by $`𝒜_R`$ by choosing items $`s`$ that maximize $`P(s)`$. This is formalized in the following lemma.
###### Lemma 7
Let $`RS`$ be a set with elements $`r_1<r_2\mathrm{}<r_{|S||T|}`$ such that $`H(r_k)=k`$ and $`r_k`$ maximizes $`P(s)`$ among all points $`sS`$ of height $`k`$. Then $`R`$ minimizes
$$\underset{rR}{}|rN(r)|+_{\mathrm{}}^+\mathrm{}|H_R(x)|𝑑x$$
Proof: Karp and Li proved that any set $`R`$ of size $`|S||T|`$ whose $`k^{th}`$ smallest element has height $`k`$ satisfies the equality
$$_{\mathrm{}}^+\mathrm{}|H_R(x)|𝑑x=_0^m|H(x)|𝑑x\underset{rR}{}_r^mh^k(x)𝑑x$$
Summing up the cost contribution of $`R`$ to both sides of the equality yields
$$\underset{rR}{}|rN(r)|+_{\mathrm{}}^+\mathrm{}|H_R(x)|𝑑x=\underset{rR}{}|rN(r)|+_0^m|H(x)|𝑑x\underset{rR}{}_r^mh^k(x)𝑑x$$
This is equivalent to
$$\underset{rR}{}|rN(r)|+_{\mathrm{}}^+\mathrm{}|H_R(x)|𝑑x=_0^m|H(x)|𝑑x\underset{rR}{}P(r)$$
Since $`P(r_k)`$ is maximized at each height $`k`$ and there is only one element in $`R`$ at each height, we have that $`R`$ maximizes $`_{rR}P(r)`$, which in turn minimizes
$$\underset{rR}{}|rN(r)|+_{\mathrm{}}^+\mathrm{}|H_R(x)|𝑑x$$
as required (refer to Lemma 3).
The following algorithm uses the preceding lemma to determine the optimal set $`R`$, and then compute the minimum cost assignment.
### 4.1 The Assignment Algorithm
Initially $`R`$ is the empty set.
* Sort $`S`$ and $`T`$.
* Calculate $`H(x)`$ for each $`xST`$. In between consecutive points, $`H`$ is constant.
* Calculate $`P(s)`$ for each $`sS`$.
* For $`k=1,2,\mathrm{}|S||T|`$
+ Find the leftmost point $`r_k`$ of height $`k`$ that maximizes $`P(r_k)`$.
+ Add $`r_k`$ to $`R`$.
* Return $`𝒜_R`$.
###### Lemma 8
The assignment algorithm computes a minimum cost assignment from $`S`$ to $`T`$.
Proof: Let $`r_k`$ be the element of $`R`$ of height $`k`$ returned by the algorithm. If we show that $`r_1<r_2<\mathrm{}<r_{|S||T|}`$, then it follows by Lemma 7 that $`𝒜_R`$ is a minimum cost assignment. We prove below, by contradiction, that indeed $`r_1<r_2<\mathrm{}<r_{|S||T|}`$.
Let $`m`$ be the largest point in $`S`$. Assume that there exists some $`k(1k|S||T|1)`$ for which the algorithm returns $`r_k`$ and $`r_{k+1}`$, with $`r_k>r_{k+1}`$. Let $`s_k`$ be the maximal element at height $`k`$ in $`SR`$ which is less than $`r_{k+1}`$. By continuity, such an $`s_k`$ must exist. Similarly, let $`s_{k+1}`$ be the minimal element at height $`k+1`$ in $`SR`$ which is greater than $`r_k`$. Such an $`s_{k+1}`$ must exist since the height at $`\mathrm{}`$ is $`H(\mathrm{})=|S||T|`$. Refer to Figure 8.
Since $`H(r_{k+1})=H(s_{k+1})`$ and $`r_{k+1}<s_{k+1}`$, we have that
$$_{r_{k+1}}^mh^{k+1}(x)𝑑x=_{r_{k+1}}^{s_{k+1}}h^{k+1}(x)𝑑x+_{s_{k+1}}^mh^{k+1}(x)𝑑x$$
From this and equation (4), we can derive the following relation between the profit functions of $`r_{k+1}`$ and $`s_{k+1}`$:
$$P(r_{k+1})=P(s_{k+1})+_{r_{k+1}}^{s_{k+1}}h^{k+1}(x)𝑑x|r_{k+1}N(r_{k+1})|+|s_{k+1}N(s_{k+1})|$$
(5)
Note that equality (5) is the result of breaking up the integral corresponding to $`P(r_{k+1})`$ into two parts, and taking into account the distance from each element to its nearest neighbor. Similarly, we can derive the following relation between $`P(r_k)`$ and $`P(s_k)`$:
$$P(s_k)=P(r_k)+_{s_k}^{r_k}h^k(x)𝑑x|s_kN(s_k)|+|r_kN(r_k)|$$
(6)
The nearest neighbor of $`s_k`$ cannot be farther than $`N(r_{k+1})`$. This translates into:
$$|s_kN(s_k)||r_{k+1}N(r_{k+1})|+|s_kr_{k+1}|$$
Also note that $`h^k(x)`$ is positive on the interval $`(s_k,r_{k+1})`$, which allows us to rewrite the previous equation as:
$$|s_kN(s_k)||r_{k+1}N(r_{k+1})|+_{s_k}^{r_{k+1}}h^k(x)𝑑x$$
(7)
Similar arguments lead to the following relationship between nearest neighbors of $`r_k`$ and $`s_{k+1}`$:
$$|r_kN(r_k)||s_{k+1}N(s_{k+1})|+_{r_k}^{s_{k+1}}h^{k+1}(x)𝑑x$$
(8)
Finally, on the interval $`(r_{k+1},r_k)`$ note that
$$_{r_{k+1}}^{r_k}h^{k+1}(x)𝑑x_{r_{k+1}}^{r_k}h^k(x)𝑑x$$
(9)
Let $`M_k=|s_kN(s_k)||r_kN(r_k)|`$. Simple arithmetic that involves inequalities (7), (8) and (9) yields
$$_{s_k}^{r_k}h^k(x)𝑑xM_k_{r_{k+1}}^{s_{k+1}}h^{k+1}(x)𝑑x+M_{k+1}$$
This along with (5) and (6) implies that
$$P(s_k)P(r_k)P(r_{k+1})P(s_{k+1})$$
Since $`r_{k+1}`$ was picked by the assignment algorithm, we have that $`P(r_{k+1})P(s_{k+1})`$. This implies that $`P(s_k)P(r_k)`$, but since $`s_k`$ lies to the left of $`r_k`$, the assignment algorithm would have picked $`s_k`$ instead of $`r_k`$, a contradiction.
### 4.2 Complexity Analysis
Sorting in step 1 takes $`O(n\mathrm{log}n)`$ time. All other steps run in $`O(n)`$ time. The only steps where this is not obvious are steps 2 and 3 that involve computing $`H(x)`$ and $`P(x)`$ respectively. $`H(x)`$ can be computed for all $`sS`$ by conducting a sweep of the sorted points in $`ST`$, adding one when we encounter an element of $`S`$ and subtracting one when we encounter an element of $`T`$.
Since all nearest neighbors of the elements of $`S`$ can easily be computed in linear time, to show that we can compute the profit function for all elements of $`S`$ in linear time we concern ourselves only with computing the integral of relative height function $`h^k`$. This integral can be computed in linear time for all points in $`S`$ at height $`k`$ in a sweep from right to left. For the rightmost element $`s_r`$ of $`S`$ at height $`k`$ $`_{s_r}^mh^k(x)𝑑x=|s_rm|`$, where $`m`$ is the largest point in $`S`$. Suppose that we know $`_s^mh^k(x)𝑑x`$ for some item $`s`$ at height $`k`$. Let $`s^{}<s`$ be the largest element in $`S`$ also at height $`k`$, and let $`t<s`$ be the largest element in $`T`$ at height $`k`$. Note that by continuity, $`t`$ exists and must be greater than $`s^{}`$. Also note that $`h^k(x)`$ is positive for all $`s^{}xt`$, and $`h^k(x)`$ is negative for all $`t<x<s`$. Thus we can derive the following equation:
$$_s^{}^mh^k(x)𝑑x=_s^mh^k(x)𝑑x+|s^{}t||ts|$$
(10)
This value can be computed in constant time for each $`s^{}S`$. Thus we can compute $`P(s)`$ for all $`sS`$ in linear time.
It follows that the assignment algorithm runs in $`O(n\mathrm{log}n)`$ time. Furthermore, if $`S`$ and $`T`$ are given in sorted order, the assignment algorithm runs in $`O(n)`$ time.
## 5 Conclusion
We have shown that the one-to-one assignment algorithm in can be extended to produce a minimum cost many-to-one assignment. The algorithm runs in $`O(n\mathrm{log}n)`$ time, if the input points are given in arbitrary order, and in $`O(n)`$ time, if the input points are presorted. To our knowledge, this is the first solution to the assignment problem that achieves this time complexity. |
warning/0507/math-ph0507059.html | ar5iv | text | # The momentum map for nonholonomic field theories with symmetry
## 1. Introduction
In this note, we study nonholonomic field theories with symmetry. Our goal is to show that the results obtained in the context of mechanical systems, such as the nonholonomic momentum map and the associated Noether theorem, have a natural counterpart in covariant field theory. We will mainly be concerned with the so-called *multisymplectic* approach to field theories (see and the references therein).
In section 2 we review the multisymplectic treatment of first-order Lagrangian field theories, with special emphasis, in subsection 2.3, on the inclusion of nonholonomic constraints into this picture. The rest of the paper is then devoted to studying the action of a symmetry group: as a warming-up, we treat in section 3 the case where no constraints are present. We review the covariant Noether theorem in a way suitable for generalization to the constrained case. In section 4, we introduce constraints into the framework and we study the implications for the Noether theorem. Finally, in section 5 we break covariance to make the link with the geometric structures known from nonholonomic mechanical systems with symmetry.
## 2. Lagrangian first-order field theories
### 2.1. Notations
Let $`\pi :YX`$ be a fibre bundle of rank $`m`$, with $`(n+1)`$-dimensional orientable base space $`X`$. In addition, we will fix a volume form $`\mu `$ on $`X`$. Typically, $`X`$ will represent space-time and the sections of $`\pi `$ will be the field configurations that we wish to study. For example, in electromagnetism, $`Y`$ is the cotangent bundle $`T^{}X`$ and the fields are $`1`$-forms representing the electromagnetic potential. For other physically relevant examples, we refer to .
From time to time, it will be handy to consider coordinate expressions of the objects involved: to this end, we choose a coordinate system $`(x^1,\mathrm{},x^{n+1})`$ on $`X`$ such that $`\mu `$ is locally given by $`\mu :=d^{n+1}x=dx^1\mathrm{}dx^{n+1}`$. On $`Y`$, we will choose a coordinate system $`(x^\mu ,y^a)`$ adapted to the projection $`\pi `$ (where $`a=1,\mathrm{},m`$). On the first jet bundle $`J^1\pi `$ we then have the induced coordinate system $`(x^\mu ,y^a,y_\mu ^a)`$. We will denote the projection of $`J^1\pi `$ onto $`Y`$ by $`\pi _{1,0}`$, and that onto $`X`$ by $`\pi _1`$ (such that $`\pi _1=\pi \pi _{1,0}`$). The bundle of $`\pi _1`$-vertical (resp. $`\pi _{1,0}`$-vertical) vectors on $`J^1\pi `$ will be denoted by $`V\pi _1`$ (resp. $`V\pi _{1,0}`$).
For later use we also mention here a particular vector-valued $`(n+1)`$-form $`S_\mu `$ on $`J^1\pi `$, called the vertical endomorphism (see ). In coordinates, $`S_\mu `$ reads
$$S_\mu =(dy^ay_\nu ^adx^\nu )d^nx_\mu \frac{}{y_\mu ^a},$$
where $`d^nx_\mu `$ is a short-hand notation for $`\frac{}{x^\mu }\text{ }\text{ }d^{n+1}x`$.
### 2.2. The Cartan form
Given a regular first-order Lagrangian $`L`$, one can construct the associated Cartan $`(n+1)`$-form $`\mathrm{\Theta }_L`$ and the multisymplectic form $`\mathrm{\Omega }_L=d\mathrm{\Theta }_L`$. The coordinate expression of $`\mathrm{\Theta }_L`$ is given by
$$\mathrm{\Theta }_L=\frac{L}{y_\mu ^a}(dy^ay_\nu ^adx^\nu )d^nx_\mu +Ld^{n+1}x.$$
We will not dwell into the precise intrinsic definition of these objects any further, but instead we refer the reader to and the references therein.
In this note, we will mainly consider the so-called De Donder-Weyl equation (see ): a connection $`\mathrm{{\rm Y}}`$ on $`\pi _1`$ with horizontal projector $`𝐡`$ is said to be a solution of the De Donder-Weyl equation if
(1)
$$i_𝐡\mathrm{\Omega }_L=n\mathrm{\Omega }_L.$$
If $`𝐡`$ is a solution of (1) and $`L`$ a regular Lagrangian, then a section $`\psi `$ of $`\pi _1`$ is an integral section of $`𝐡`$ if $`\psi =j^1\varphi `$ for a section $`\varphi `$ of $`\pi `$ (implying that $`\mathrm{{\rm Y}}`$ is semi-holonomic) and, in addition, $`j^1\varphi `$ satisfies the Euler-Lagrange equations:
(2)
$$\frac{d}{dx^\mu }\left(\frac{L}{y_\mu ^a}(j^1\varphi )\right)\frac{L}{y^a}(j^1\varphi )=0.$$
See for a proof of this statement.
### 2.3. Nonholonomic constraints
In this section, we will briefly show how to derive the nonholonomic equations of motion for a system with Lagrangian $`L`$ and a set of constraints represented by a submanifold $`𝒞`$. For a more detailed treatment, we refer to .
Let $`𝒞`$ be a $`k`$-codimensional submanifold of $`J^1\pi `$, with $`\pi _{1,0}(𝒞)=Y`$ and such that $`(\pi _{1,0})_{|𝒞}:𝒞Y`$ is a subbundle of $`\pi _{1,0}`$. The submanifold $`𝒞`$ will represent some external (nonholonomic) constraints imposed on the system. Assume that $`𝒞`$ is locally given by the vanishing of $`k`$ independent functions $`\phi ^\alpha `$ and consider the subset $`F`$ of $`^{n+1}(T^{}J^1\pi )`$ spanned by $`\mathrm{\Phi }^\alpha =S_\mu ^{}(d\phi ^\alpha )`$, where $`S_\mu `$ is the vertical endomorphism on $`J^1\pi `$. In coordinates, we have
$$\mathrm{\Phi }^\alpha =\frac{\phi ^\alpha }{y_\mu ^a}(dy^ay_\nu ^adx^\nu )d^nx_\mu .$$
The $`(n+1)`$-forms $`\mathrm{\Phi }^\alpha `$ are linearly independent because of the initial assumption that $`(\pi _{1,0})_{|𝒞}`$ is a subbundle of $`\pi _{1,0}`$. Hence, $`F`$ is a subbundle of $`^{n+1}(T^{}J^1\pi )`$.
In the presence of nonholonomic constraints, the field equations become
(3)
$$\frac{d}{dx^\mu }\left(\frac{L}{y_\mu ^a}(j^1\varphi )\right)\frac{L}{y^a}=\lambda _{\alpha \mu }\frac{\phi ^\alpha }{y_\mu ^a},$$
together with the constraint that $`j^1\varphi 𝒞`$ (see ). Accordingly, the unconstrained De Donder-Weyl equations are replaced by the following conditions along $`𝒞`$:
(4)
$$i_𝐡\mathrm{\Omega }_Ln\mathrm{\Omega }_L(F)\text{and }\mathrm{Im}𝐡T𝒞,$$
where $`(F)`$ is the ideal generated by $`F`$. The terms on the right-hand side of (3) and (4) represent the constraint forces that keep the section $`j^1\varphi `$ constrained to $`𝒞`$. The unknown multipliers $`\lambda _{\alpha \mu }`$ should be determined from the condition that $`j^1\varphi 𝒞`$.
Remark: In general, the constraints represented by the submanifold $`𝒞`$ are nonlinear. Linear constraints can be treated as a special case of this formalism by considering a distribution $`D`$ on $`Y`$ and taking $`𝒞`$ to be
$$𝒞=\{j_x^1\varphi J^1\pi :\mathrm{Im}T_x\varphi D_{\varphi (x)}\}.$$
If $`D`$ is annihilated by the $`k`$ one-forms $`A_a^\alpha dy^a+B_\mu ^\alpha dx^\mu `$, then $`𝒞`$ is given by the vanishing of the $`kn`$ functions $`\phi _\mu ^\alpha =A_a^\alpha y_\mu ^a+B_\mu ^\alpha `$. Whenever $`D`$ is integrable, these constraint functions can be written as total derivatives with respect to $`x^\mu `$ of functions on $`Y`$, in which case the constraints can reasonably be said to be holonomic. This case is treated in far greater detail in .
### 2.4. Connections on $`\pi _1`$
In this section, we will prove a number of straightforward properties of connections on $`\pi _1`$ that will be useful later on. The reader is referred to for a more comprehensive treatment.
We recall that a connection $`\mathrm{{\rm Y}}`$ on $`\pi _1`$ is said to be *semi-holonomic* if the associated horizontal projector $`𝐡`$ satisfies $`i_𝐡\theta =0`$ for each contact one-form $`\theta `$. In coordinates, if
$$𝐡=dx^\mu \left(\frac{}{x^\mu }+\mathrm{\Gamma }_\mu ^a\frac{}{y^a}+\mathrm{\Gamma }_{\mu \nu }^a\frac{}{y_\nu ^a}\right),$$
semi-holonomicity implies that $`\mathrm{\Gamma }_\mu ^a=y_\mu ^a`$. This implies that any integral section of $`𝐡`$ is automatically the prolongation of a section of $`\pi `$.
###### Lemma 1.
For each semi-holonomic connection $`\mathrm{{\rm Y}}`$ with horizontal projector $`𝐡`$, the following holds:
$$i_𝐡\mathrm{\Theta }_L=n\mathrm{\Theta }_L+L\mu .$$
Proof: We give the proof in coordinates. For any connection $`𝐡`$, we have
$$i_𝐡d^{n+1}x=(n+1)d^{n+1}x\text{and}i_𝐡d^nx_\mu =nd^nx_\mu .$$
Therefore,
$$i_𝐡\mathrm{\Theta }_L=\frac{L}{y_\nu ^a}i_𝐡\theta ^ad^nx_\nu +n\frac{L}{y_\nu ^a}\theta ^ad^nx_\nu +(n+1)Ld^{n+1}x,$$
where we have introduced the contact forms $`\theta ^a=dy^ay_\mu ^adx^\mu `$. If $`𝐡`$ is semi-holonomic, the first term on the right-hand side is zero and we obtain the desired expression. $`\mathrm{}`$
This lemma can be seen as the jet-bundle analogue of the well-known fact in Lagrangian mechanics that $`i_X\theta _L=\mathrm{\Delta }(L)`$ for any second-order vector field $`X`$, where $`\theta _L`$ is the Cartan one-form corresponding to $`L`$, and $`\mathrm{\Delta }`$ the Liouville vector field.
###### Lemma 2.
Let $`X`$ be a vertical vector field on $`Y`$ and $`X^{(1)}`$ its prolongation to $`J^1\pi `$. If $`\mathrm{{\rm Y}}`$ is a semi-holonomic connection on $`\pi _1`$ with horizontal projector $`𝐡`$, then the Frölicher-Nijenhuis bracket $`[X^{(1)},𝐡]`$ is a vector-valued one-form taking values in $`V\pi _{1,0}`$.
Proof: If $`X=X^a\frac{}{y^a}`$, then
$$X^{(1)}=X^a\frac{}{y^a}+\left(\frac{X^a}{x^\mu }+\frac{X^a}{y^b}y_\mu ^b\right)\frac{}{y_\mu ^a}.$$
(see e.g. ) For the bracket, we have that $`[X^{(1)},𝐡]=_{X^{(1)}}𝐡`$ and a straightforward calculation then shows that this is a semi-basic vector-valued one-form taking values in $`V\pi _1`$. We now focus on the coefficient of $`dx^\mu \frac{}{y^a}`$, which is just
$$X^{(1)}(\mathrm{\Gamma }_\mu ^a)\left(\frac{X^a}{x^\mu }+\mathrm{\Gamma }_\mu ^b\frac{X^a}{y^b}\right).$$
This coefficient is easily seen to vanish when $`\mathrm{\Gamma }_\mu ^a=y_\mu ^a`$, i.e. when $`𝐡`$ is semi-holonomic, which completes the proof. $`\mathrm{}`$
As a corollary, we note that this lemma implies that the contraction of $`[X^{(1)},𝐡]`$ with a semi-basic form (in particular with $`\mathrm{\Theta }_L`$) vanishes.
## 3. Symmetry in the absence of nonholonomic constraints
Let $`G`$ be a Lie group acting on $`Y`$ by bundle automorphisms $`\mathrm{\Phi }_g`$ over the identity in $`X`$. The assumption that $`G`$ acts vertically is probably superfluous, but for the sake of clarity we will assume it nevertheless.
The Lie group $`G`$ acts on $`J^1\pi `$ by prolonged bundle automorphisms, i.e. $`j^1\mathrm{\Phi }_g(j_x^1\varphi )=j_x^1(\mathrm{\Phi }_g\varphi )`$. Now, let $`LC^{\mathrm{}}(J^1\pi )`$ be a $`G`$-invariant Lagrangian. The action of $`G`$ on $`J^1\pi `$ is called *Lagrangian* if, for each $`\xi 𝔤`$, there exists an $`n`$-form $`J_\xi `$ (depending linearly on $`\xi `$) such that $`i_{\xi _{J^1\pi }}\mathrm{\Omega }_L=dJ_\xi `$, where $`\xi _{J^1\pi }`$ denotes the infinitesimal generator corresponding to $`\xi `$. In this case, the map $`J:J^1\pi ^n(T^{}J^1\pi )𝔤^{}`$ defined by $`J,\xi :=J_\xi `$ is called the *covariant momentum map* for the action $`\mathrm{\Phi }`$.
In general, we can also consider actions of $`G`$ on $`J^1\pi `$ that are not prolonged actions of an action on $`Y`$, but in this note we will nevertheless restrict ourselves to this special case. It is easy to see that Lagrangian actions satisfy $`_{\xi _{J^1\pi }}\mathrm{\Omega }_L=0`$; for prolonged actions we have in addition that $`_{\xi _{J^1\pi }}\mathrm{\Theta }_L=0`$ (see lemma 3).
If $`G`$ acts on $`J^1\pi `$ by prolonged bundle automorphisms, then for each $`\xi 𝔤`$ the infinitesimal generator $`\xi _{J^1\pi }`$ on $`J^1\pi `$ is the prolongation of the infinitesimal generator $`\xi _Y`$ on $`Y`$. From now on, we will denote $`\xi _{J^1\pi }`$ by $`\xi ^{(1)}`$.
###### Lemma 3.
The Cartan $`(n+1)`$-form $`\mathrm{\Theta }_L`$ is invariant with respect to the action of $`G`$ lifted to $`J^1\pi `$:
$$_{\xi ^{(1)}}\mathrm{\Theta }_L=0.$$
Proof: See \[9, p. 45\]. $`\mathrm{}`$
For a prolonged action, there always exists a covariant momentum map which is explicitly given by
$$J_\xi =i_{\xi ^{(1)}}\mathrm{\Theta }_L.$$
(see \[9, p. 45\]). The importance of the covariant momentum map lies in the *covariant Noether theorem*, first proved in .
###### Proposition 4 (Covariant Noether theorem).
Let $`\mathrm{{\rm Y}}`$ be a connection on $`\pi _1`$ such that the associated horizontal projector $`𝐡`$ is a solution of the unconstrained De Donder-Weyl equation (1). For every $`\xi 𝔤`$, the momentum map $`J_\xi `$ is constant on integral sections of $`𝐡`$:
$$d_𝐡J_\xi =0.$$
Proof: In this proof, as well as in the remainder of this note, we make frequent use of some elementary properties of the Frölicher-Nijenhuis bracket. For the sake of completeness, we have summarized these properties in the appendix.
We have
$`d_𝐡J_\xi `$ $`=d_𝐡i_{\xi ^{(1)}}\mathrm{\Theta }_L`$
$`=(i_𝐡ddi_𝐡)i_{\xi ^{(1)}}\mathrm{\Theta }_L`$
(5) $`=i_𝐡_{\xi ^{(1)}}\mathrm{\Theta }_Li_𝐡i_{\xi ^{(1)}}d\mathrm{\Theta }_Ldi_𝐡i_{\xi ^{(1)}}\mathrm{\Theta }_L.`$
In the last expression, the first term vanishes because of lemma 3. The second term can be rewritten by using the field equations (note that $`𝐡(\xi ^{(1)})=0`$ as $`\xi ^{(1)}`$ is $`\pi _1`$-vertical):
$$i_𝐡i_{\xi ^{(1)}}d\mathrm{\Theta }_L=i_{\xi ^{(1)}}i_𝐡d\mathrm{\Theta }_L=ni_{\xi ^{(1)}}\mathrm{\Omega }_L,$$
whereas for the last term we have, using lemma 2,
$`di_𝐡i_{\xi ^{(1)}}\mathrm{\Theta }_L`$ $`=di_{\xi ^{(1)}}i_𝐡\mathrm{\Theta }_L`$
$`=di_{\xi ^{(1)}}\left(n\mathrm{\Theta }_L+L\mu \right).`$
Now, $`i_{\xi ^{(1)}}(L\mu )=0`$ and so we obtain
$$d_𝐡J_\xi =ni_{\xi ^{(1)}}\mathrm{\Omega }_Lndi_{\xi ^{(1)}}\mathrm{\Theta }_L=n_{\xi ^{(1)}}\mathrm{\Theta }_L=0,$$
again due to the invariance of $`\mathrm{\Theta }_L`$. $`\mathrm{}`$
Remark: In \[9, p. 45\], the authors prove a slightly different Noether theorem. They show that, if $`\varphi `$ is a solution of the field equations, then $`d(j^1\varphi )^{}J_\xi =0`$. It is not hard to prove that, for any $`k`$-form $`\alpha `$ on $`J^1\pi `$, $`(j^1\varphi )^{}d_𝐡\alpha =d(j^1\varphi )^{}\alpha `$ if and only if $`j^1\varphi `$ is an integral section of $`𝐡`$. Proposition 4 therefore implies that $`d(j^1\varphi )^{}J_\xi =0`$. The proof of the Noether theorem in is more straightforward; our proof has the advantage that it will be easily extendible to the case where nonholonomic constraints are present.
## 4. The constrained momentum map
In this section, we study the case of a constrained field theory, with regular Lagrangian $`L`$ and constraint submanifold $`𝒞`$ satisfying the assumptions of section 2.3. The constrained De Donder-Weyl equations are then given by (4).
Suppose now that in addition to these nonholonomic constraints, there is also a symmetry group $`G`$ acting on $`J^1\pi `$ by prolonged bundle automorphisms, such that both the Lagrangian $`L`$ and the constraint manifold $`𝒞`$ are $`G`$-invariant, i.e.
$$Lj^1\mathrm{\Phi }_g=L\text{as well as}j^1\mathrm{\Phi }_g(𝒞)𝒞$$
for all $`gG`$. In general, as in the case of nonholonomic mechanics (see ), it will no longer be true that these symmetries give rise to conserved quantities; the precise link will be made clear by the *nonholonomic momentum equation* or constrained Noether theorem (theorem 5). Our treatment extends the one in ; we refer to that paper, as well as to and the references therein, for more information about the nonholonomic momentum equation in mechanics.
We first introduce the following distribution:
$$(\gamma )=\{vT_\gamma J^1\pi :i_v(S_\mu ^{}d\phi _\alpha )=0\text{ for each }\alpha =1,\mathrm{},k\}\text{where }\gamma 𝒞.$$
For a given $`\gamma 𝒞`$ we consider all elements $`\xi `$ of the Lie algebra $`𝔤`$ such that $`\xi ^{(1)}(\gamma )(\gamma )`$. The set of all such $`\xi `$ we denote by $`𝔤^\gamma `$. We take $`𝔤^{}`$ to be the disjoint union of all these spaces $`𝔤^\gamma `$ and we assume that $`𝔤^{}`$ can be given the structure of a bundle over $`𝒞`$.
With these elements in mind, we define the *constrained momentum map* as a map $`J^{\mathrm{n}.\mathrm{h}.}:𝒞^n(J^1\pi )g^{}`$, constructed as follows. With every section $`\overline{\xi }`$ of $`𝔤^{}`$, one may associate a vector field $`\stackrel{~}{\xi }`$ on $`J^1\pi `$ by putting $`\stackrel{~}{\xi }(\gamma )=(\overline{\xi }(\gamma ))_{J^1\pi }(\gamma )`$. Remark that $`\stackrel{~}{\xi }`$ is a section of $``$. We then define $`J_{\overline{\xi }}^{\mathrm{n}.\mathrm{h}.}`$ along $`𝒞`$ as
$$J_{\overline{\xi }}^{\mathrm{n}.\mathrm{h}.}=i_{\stackrel{~}{\xi }}\mathrm{\Theta }_L.$$
The importance of the nonholonomic momentum map lies in the nonholonomic momentum equation:
###### Theorem 5 (Nonholonomic momentum equation).
Let $`\mathrm{{\rm Y}}`$ be a connection on $`\pi _1`$ such that the associated horizontal projector $`𝐡`$ is a solution of the constrained De Donder-Weyl equation. Assume furthermore that $`G`$ is a Lie group acting on $`J^1\pi `$ in the way described above. Then the nonholonomic momentum map satisfies the following equation:
$$d_𝐡J_{\overline{\xi }}^{\mathrm{n}.\mathrm{h}.}=_{\stackrel{~}{\xi }}(L\mu )\text{along }𝒞.$$
Proof: Equation (5) from the proof of proposition 4 can be used without modification:
$`d_𝐡J_{\overline{\xi }}^{\mathrm{n}.\mathrm{h}.}`$ $`=i_𝐡_{\stackrel{~}{\xi }}\mathrm{\Theta }_Li_𝐡i_{\stackrel{~}{\xi }}d\mathrm{\Theta }_Ldi_𝐡i_{\stackrel{~}{\xi }}\mathrm{\Theta }_L`$
$`=i_𝐡_{\stackrel{~}{\xi }}\mathrm{\Theta }_L+i_{\stackrel{~}{\xi }}(n\mathrm{\Omega }_L+\zeta )n_{\stackrel{~}{\xi }}\mathrm{\Theta }_L+ni_{\stackrel{~}{\xi }}d\mathrm{\Theta }_L.`$
In this expression, we have substituted the constrained De Donder-Weyl equation: $`\zeta `$ is an element of $`(F)`$. As $`\zeta `$ can be written as $`\zeta =\lambda _{\alpha \mu }dx^\mu f^\alpha `$ (see ), with $`f^\alpha `$ taking values in the bundle $`F`$, we may conclude that $`i_{\stackrel{~}{\xi }}\zeta =0`$. Therefore, we end up with
$`d_𝐡J_{\stackrel{~}{\xi }}^{\mathrm{n}.\mathrm{h}.}`$ $`=i_𝐡_{\stackrel{~}{\xi }}\mathrm{\Theta }_Ln_{\stackrel{~}{\xi }}\mathrm{\Theta }_L`$
$`=_{\stackrel{~}{\xi }}i_𝐡\mathrm{\Theta }_Li_{[\stackrel{~}{\xi },𝐡]}\mathrm{\Theta }_Ln_{\stackrel{~}{\xi }}\mathrm{\Theta }_L`$
$`=_{\stackrel{~}{\xi }}(L\mu ),`$
where we have used the remark following lemma 2 to conclude that $`i_{[\stackrel{~}{\xi },𝐡]}\mathrm{\Theta }_L=0`$. $`\mathrm{}`$
We finish by noting that in the case where $`\stackrel{~}{\xi }`$ can be written as $`\xi ^{(1)}`$ (for example, when $`\overline{\xi }`$ is a constant section), we may conclude from the $`G`$-invariance of $`L`$ that $`d_𝐡J_{\stackrel{~}{\xi }}^{\mathrm{n}.\mathrm{h}.}=0`$. In general, though, this will not be the case.
## 5. The Cauchy formalism
Up until now, all of our results have been derived in a purely covariant setting where all of the coordinates on the base space $`X`$ are treated on an equal footing. In particular, there is no distinguised time coordinate. We will now assume that the Euler-Lagrange equations associated to the Lagrangian $`L`$ describe an (hyperbolic) initial-value problem. In this case, it is meaningful to single out a global direction of time and break covariance by making the transition to the space of Cauchy data. We can then rephrase the field equations accordingly as a time-dependent mechanical system on an infinite-dimensional configuration space (see ).
This is done by fixing a particular diffeomorphism $`\mathrm{\Psi }:\times MX`$, where $`M`$ is an $`n`$-dimensional manifold (and where we tacitly assume that the topology of $`X`$ is such that $`\mathrm{\Psi }`$ can indeed be globally defined), thus singling out a “splitting” of $`X`$ into space and time. To avoid the technical matters arising when considering the behaviour of the field “at infinity”, we assume that $`M`$ is compact. We define the space $`\stackrel{~}{X}`$ to consist of all embeddings $`\tau `$ of $`M`$ into $`X`$ such that there exists a $`t`$ for which $`\tau =\mathrm{\Psi }(t,)`$. Hence, there is a one-to-one correspondence between $``$ and $`\stackrel{~}{X}`$. This correspondence, or the existence of the diffeomorphism $`\mathrm{\Psi }`$, induces a distinguished vector field $`𝐓`$ on $`X`$, defined at $`xX`$, by
$$𝐓(x)=\frac{d}{ds}\mathrm{\Psi }(s,u)|_{s=t},\text{where }x=\mathrm{\Psi }(t,u).$$
For the sake of convenience, we will assume that $`M`$ is equipped with a volume form $`\mu _M`$ such that $`\mu :=dt\mu _M`$ is a volume form for $`X`$, where $`t`$ is a global coordinate labelling $`\stackrel{~}{X}`$.
We define the *space of Cauchy data* (denoted by $`\stackrel{~}{Z}`$) as the space of embeddings $`\kappa :MJ^1\pi `$ for which there exists a section $`\varphi `$ of $`\pi `$ and an element $`\tau `$ of $`\stackrel{~}{X}`$ such that $`\kappa =j^1\varphi \tau `$. For more information on this space (which can be given the structure of a smooth manifold in some suitable sense), we refer the reader to . There exists a convenient way of viewing the tangent bundle of $`\stackrel{~}{Z}`$: a tangent vector $`vT_\kappa \stackrel{~}{Z}`$ can be seen as a section of $`\mathrm{\Gamma }(\kappa ^{}TJ^1\pi )`$ (a vector field along $`\kappa `$). There exist similar interpretations of $`T\stackrel{~}{X}`$ and $`T\stackrel{~}{Y}`$.
A vector field $`V`$ on $`J^1\pi `$ induces a vector field $`\stackrel{~}{V}`$ on $`\stackrel{~}{Z}`$ by composition: $`\stackrel{~}{V}(\kappa )=V\kappa `$. Similarly, an $`(n+k)`$-form $`\alpha `$ on $`J^1\pi `$ induces a $`k`$-form $`\stackrel{~}{\alpha }`$ on $`\stackrel{~}{Z}`$ by integration:
(6)
$$\stackrel{~}{\alpha }(\kappa )(\stackrel{~}{V}_1,\mathrm{},\stackrel{~}{V}_k)=_M\kappa ^{}i_{\stackrel{~}{V}_1\mathrm{}\stackrel{~}{V}_k}\alpha .$$
By use of this correspondence, the multisymplectic form $`\mathrm{\Omega }_L`$ and the volume form $`\mu `$ induce respectively a two-form $`\stackrel{~}{\mathrm{\Omega }}_L`$ and a one-form $`\stackrel{~}{\mu }`$ on $`\stackrel{~}{Z}`$, whereas the Lagrangian $`L`$ can be seen to induce a function on $`\stackrel{~}{Z}`$:
$$\stackrel{~}{L}(\kappa )=_M\kappa ^{}i_𝐓(L\mu ).$$
Strictly speaking, on the right-hand side of this expression one should replace $`𝐓`$ by an arbitrary vector field $`V`$ on $`J^1\pi `$ projecting down to $`𝐓`$, but since $`L\mu `$ is semi-basic, the contraction does not depend on $`V`$ but only on $`𝐓`$.
Remark: It has been shown that the covariant field equations induce a dynamical system $`\mathrm{\Gamma }`$ on $`\stackrel{~}{Z}`$ whose determining equations are formally identical to those of a time-dependent mechanical system with an infinite-dimensional configuration space (see ):
(7)
$$i_\mathrm{\Gamma }\stackrel{~}{\mathrm{\Omega }}_L=0\text{and}i_\mathrm{\Gamma }\stackrel{~}{\mu }=1.$$
In , we showed that in the case of nonholonomic field theory, the induced dynamical system on $`\stackrel{~}{Z}`$ is determined by
(8)
$$i_\mathrm{\Gamma }\stackrel{~}{\mathrm{\Omega }}_L|_{\stackrel{~}{𝒞}}\stackrel{~}{F}\text{and}\mathrm{\Gamma }T\stackrel{~}{𝒞},$$
where $`\stackrel{~}{F}`$ is a codistribution induced by $`F`$ and $`\stackrel{~}{𝒞}`$ is the subset of $`\stackrel{~}{Z}`$ induced by $`𝒞`$ and defined as
$$\stackrel{~}{𝒞}=\{\kappa \stackrel{~}{Z}:\mathrm{Im}\kappa 𝒞\}.$$
In both the constrained and the unconstrained case, a connection $`\mathrm{{\rm Y}}`$ solving the covariant field equations induces a vector field $`\mathrm{\Gamma }`$ on $`\stackrel{~}{Z}`$ which is a solution of the corresponding dynamical system on $`\stackrel{~}{Z}`$. In the unconstrained case, this dynamical system is given by (7), whereas in the constrained case the equations of motion are given by (8). The precise relation between $`𝐡`$ and $`\mathrm{\Gamma }`$ is
(9)
$$\mathrm{\Gamma }(\kappa )=𝐡(Tj^1\varphi (𝐓))\kappa ,$$
where we have decomposed $`\kappa `$ as $`\kappa =j^1\varphi \tau `$. With some abuse of notation, we will also write $`\mathrm{\Gamma }=𝐡(𝐓)`$.
In the next sections, we will exhibit the structures on $`\stackrel{~}{Z}`$ induced by the (nonholonomic) momentum map and we will show how the covariant momentum equation give rises to a momentum equation on $`\stackrel{~}{Z}`$ which is formally identical to the one encountered in nonholonomic mechanics (see for example ).
By (6), the component $`J_\xi :J^1\pi ^n(J^1\pi )`$ of the covariant momentum map induces a map $`\stackrel{~}{J}_\xi C^{\mathrm{}}(\stackrel{~}{Z})`$ on the space of Cauchy data:
$$\stackrel{~}{J}_\xi (\kappa )=_M\kappa ^{}J_\xi .$$
In the constrained case, there is a similar definition for the map $`\stackrel{~}{J}_\xi ^{\mathrm{n}.\mathrm{h}.}`$ in the Cauchy formalism, induced by the component $`J_\xi ^{\mathrm{n}.\mathrm{h}.}`$ of the constrained momentum map. Note that $`J_\xi ^{\mathrm{n}.\mathrm{h}.}`$ is defined along $`𝒞`$.
### 5.1. The unconstrained case
We now turn to proving the analogue of Noether’s theorem in the Cauchy framework. There are essentially two ways in which one could approach this problem: either by directly defining the action of $`G`$ on $`\stackrel{~}{Z}`$ and using the standard techniques known from mechanics, or by showing that the covariant Noether theorem leads in a straightforward way to the Noether theorem on the space of Cauchy data. We choose to follow the second approach, as it allows us to postpone to the very end all of the technical matters associated with the calculus on infinite-dimensional manifolds.
###### Proposition 6.
Let $`\mathrm{{\rm Y}}`$ be a connection in $`\pi _1`$ such that the associated horizontal projector $`𝐡`$ is a solution of the De Donder-Weyl equation (1). Let $`\stackrel{~}{J}`$ be the momentum map associated to the covariant momentum map $`J`$. Then Noether’s theorem holds: $`\mathrm{\Gamma }(\stackrel{~}{J}_\xi )=0`$ for all $`\xi 𝔤`$, where $`\mathrm{\Gamma }`$ is a solution to the equations of motion (7) in the Cauchy formalism.
Proof: We will use the following characterisation of the exterior derivative $`d\stackrel{~}{J}_\xi `$ in terms of $`dJ_\xi `$:
$$\stackrel{~}{V},d\stackrel{~}{J}_\xi (\kappa )=_M\kappa ^{}(i_{\stackrel{~}{V}}dJ_\xi ),$$
for an arbitrary vector field $`\stackrel{~}{V}`$ on $`\stackrel{~}{Z}`$. For a proof, we refer to \[14, prop. 3.3.9\] or to the expressions used in \[10, lemma 5.1\].
The embedding $`\kappa :MJ^1\pi `$ can be written as $`\kappa =j^1\varphi \tau `$. Without loss of generality, we may take $`\varphi `$ to be a solution of the field equations. This lies at the heart of the Cauchy analysis: $`\kappa `$ specifies the values of the fields and their derivatives on a hypersurface and due to the (supposed) hyperbolicity of the equations of motion, the subsequent evolution is then fixed (and given by $`j^1\varphi `$). Formally, let $`tc(t)`$ be an integral curve of $`\mathrm{\Gamma }`$ such that $`c(0)=\kappa `$. Then $`j_x^1\varphi =[c(t)](u)`$, where $`x=\mathrm{\Phi }(t,u)`$.
We then have, noting that $`𝐡(𝐓)=Tj^1\varphi (𝐓)`$,
$$\mathrm{\Gamma },d\stackrel{~}{J}_\xi (\kappa )=_M\kappa ^{}(i_\mathrm{\Gamma }dJ_\xi )=_M\tau ^{}(j^1\varphi )^{}(i_{𝐡(𝐓)}dJ)=_M\tau ^{}i_𝐓((j^1\varphi )^{}dJ).$$
As we pointed out in the remark following proposition 4, one can check that $`(j^1\varphi )^{}d_𝐡\alpha =d(j^1\varphi )^{}\alpha `$ if and only if $`j^1\varphi `$ is an integral section of $`𝐡`$. We conclude that
(10)
$$\mathrm{\Gamma },d\stackrel{~}{J}_\xi (\kappa )=_M\tau ^{}i_𝐓((j^1\varphi )^{}d_𝐡J_\xi ).$$
As the $`\xi `$-component $`J_\xi `$ of the covariant momentum map satisfies Noether’s theorem, i.e. $`d_𝐡J_\xi =0`$, we have that $`\mathrm{\Gamma }(\stackrel{~}{J}_\xi )=0`$. This establishes the theorem of Noether in the Cauchy framework. $`\mathrm{}`$
### 5.2. The constrained case
Quite surprisingly, much of the material developed in the preceding section carries over quite naturally to the constrained case. In particular, for the nonholonomic momentum map, equation (10) still holds:
$$\mathrm{\Gamma },d\stackrel{~}{J}_{\overline{\xi }}^{\mathrm{n}.\mathrm{h}.}(\kappa )=_M\tau ^{}i_𝐓((j^1\varphi )^{}d_𝐡J_{\overline{\xi }}^{\mathrm{n}.\mathrm{h}.}),\text{for }\kappa 𝒞,$$
where we attribute a similar meaning to all terms involved: $`𝐡`$ is a solution of the constrained De Donder-Weyl equation, $`j^1\varphi `$ is an integral section of the corresponding connection and $`\mathrm{\Gamma }=𝐡(𝐓)`$. Note that $`\mathrm{\Gamma }`$ is now a solution of (8).
Now, if $`J_{\overline{\xi }}^{\mathrm{n}.\mathrm{h}.}`$ satisfies the nonholonomic momentum equation, then
(11)
$$\mathrm{\Gamma },d\stackrel{~}{J}_{\overline{\xi }}^{\mathrm{n}.\mathrm{h}.}(\kappa )=_M\tau ^{}i_𝐓((j^1\varphi )^{}_{\overline{\xi }}(L\mu )).$$
In the following proposition, we further elaborate the right-hand side. We recall that the vector field $`\stackrel{~}{\xi }`$ on $`J^1\pi `$ naturally induces a vector field $`\widehat{\xi }`$ on $`\stackrel{~}{Z}`$ by putting $`\widehat{\xi }(\kappa )=\stackrel{~}{\xi }\kappa `$.
###### Proposition 7.
Let $`\mathrm{{\rm Y}}`$ be a connection on $`\pi _1`$ such that along the constraint submanifold $`𝒞`$ the associated horizontal projector $`𝐡`$ satisfies the constrained De Donder-Weyl equation. Assume a Lie group $`G`$ acts in the way described above and let $`\stackrel{~}{J}^{\mathrm{n}.\mathrm{h}.}`$ be the momentum map associated to the covariant momentum map $`J^{\mathrm{n}.\mathrm{h}.}`$. Then $`\stackrel{~}{J}^{\mathrm{n}.\mathrm{h}.}`$ satisfies the nonholonomic momentum equation: for all $`\overline{\xi }𝔤^{}`$,
$$\mathrm{\Gamma }(\stackrel{~}{J}_{\overline{\xi }}^{\mathrm{n}.\mathrm{h}.})=\widehat{\xi }(\stackrel{~}{L})\text{along }𝒞.$$
Proof: We rewrite the right-hand side of (11) by performing exactly the opposite manipulations as we did to obtain eq. (10). This leads to
$$\mathrm{\Gamma },d\stackrel{~}{J}_{\overline{\xi }}^{\mathrm{n}.\mathrm{h}.}(\kappa )=_M\kappa ^{}i_{𝐡(𝐓)}_{\stackrel{~}{\xi }}(L\mu )=_M\kappa ^{}_{\stackrel{~}{\xi }}(i_{𝐡(𝐓)}(L\mu ))+_M\kappa ^{}i_{[𝐡(𝐓),\stackrel{~}{\xi }]}(L\mu ).$$
The last term vanishes as $`L\mu `$ is semi-basic and $`[𝐡(𝐓),\stackrel{~}{\xi }]`$ is $`\pi _1`$-vertical ($`\stackrel{~}{\xi }`$ is $`\pi _1`$-vertical). By lemma 3.3.9 of , we see that the first term on the right-hand side equals
$$_M\kappa ^{}_{\stackrel{~}{\xi }}(i_{𝐡(𝐓)}(L\mu ))=_{\widehat{\xi }}(\stackrel{~}{L}),$$
and this proves the momentum equation in the Cauchy formalism. $`\mathrm{}`$
## Acknowledgements
Financial support of the Research Foundation–Flanders (FWO-Vlaanderen) is gratefully acknowledged. I would also like to thank Frans Cantrijn for useful discussions and a critical reading of this manuscript, as well as Manuel de León and David Martín de Diego for many fruitful discussions and their kind hospitality during several research visits to the CSIC (Madrid).
## Appendix: elementary properties of the Frölicher-Nijenhuis bracket
In this section, we review some properties of the Frölicher-Nijenhuis bracket and the various derivations associated to vector-valued forms on a manifold. For a detailed treatment of the Frölicher-Nijenhuis bracket, we refer the reader to .
Let $`M`$ be a manifold. A *vector-valued one-form $`𝐡`$* is a section of $`TMT^{}M`$. Associated to $`𝐡`$ is a derivation $`i_𝐡`$ (of type $`i_{}`$ and degree $`0`$), defined by
$$(i_𝐡\alpha )(v_0,\mathrm{},v_k)=\underset{i=0}{\overset{k}{}}(1)^i\alpha (𝐡(v_i),v_0,\mathrm{},\widehat{v_i},\mathrm{},v_k)\text{for }\alpha \mathrm{\Omega }^{k+1}(M).$$
We then define $`d_𝐡`$ as $`d_𝐡=i_𝐡ddi_𝐡`$; this is a derivation of type $`d_{}`$ and degree $`1`$.
Vector-valued forms of higher degree are defined accordingly as sections of the tensor product $`TM^k(T^{}M)`$. A vector-valued $`k`$-form $`R`$ can easily be seen to give rise to a derivation $`i_R`$ of degree $`k1`$ (by virtue of a generalization of eq. Appendix: elementary properties of the Frölicher-Nijenhuis bracket) as well as a derivation $`d_R`$ of degree $`k`$. A vector-valued form of degree zero is simply a vector field, and the associated derivations are in this case the contraction $`i_X`$ and the Lie derivative $`_X`$.
The Frölicher-Nijenhuis bracket of a vector-valued $`r`$-form $`R`$ and a vector-valued $`s`$-form $`S`$ is then defined as the unique vector-valued $`(r+s)`$-form $`[R,S]`$ for which
$$d_Rd_S(1)^{rs}d_Sd_R=d_{[R,S]}.$$
We have deliberately been vague about the nature of this bracket: most of the time we will only need the bracket of a vector field $`X`$ with a vector-valued one-form $`𝐡`$ (which will be the horizontal projector of a connection). In this case, it is not hard to prove that
$$[X,𝐡]=_X𝐡.$$
The following lemma collects the properties of the Frölicher-Nijenhuis bracket that we will be needing in the body of the text. They can be suitably generalized and form part of a well-investigated calculus, for which we refer to .
###### Lemma 8.
Let $`X`$ be a vector field on $`M`$ and $`𝐡`$ a vector-valued one-form. Then, for any $`k`$-form $`\alpha `$ on $`M`$, the following holds:
1. $`i_Xi_𝐡\alpha =i_𝐡i_X\alpha +i_{𝐡(X)}\alpha `$;
2. $`i_𝐡_X\alpha =_Xi_𝐡\alpha i_{[X,𝐡]}\alpha `$.
Proof: Let $`\alpha `$ be a $`2`$-form (the case of a $`k`$-form $`\alpha `$ is completely similar) and $`Y`$ a vector field on $`M`$. Then
$`(i_Xi_𝐡\alpha )(Y)`$ $`=\alpha (𝐡(X),Y)\alpha (𝐡(Y),X)`$
$`=(i_{𝐡(X)}\alpha )(Y)+(i_𝐡i_X\alpha )(Y),`$
which confirms the first property.
The second property (a special case of lemma 8.6 in ) can be proved directly by noting that a derivation is completely determined by its action on functions and one-forms. For a function $`f`$ both sides of the relation (2) vanish and for a one-form $`\alpha `$ we have for the left-hand side
$`(i_𝐡_X\alpha )(Y)`$ $`=(_X\alpha )(𝐡(Y))=_X(\alpha (𝐡(Y)))\alpha ([X,𝐡(Y)])`$
$`=_X(\alpha (𝐡(Y)))\alpha ((_X𝐡)(Y))\alpha (𝐡([X,Y])).`$
Taking together the first and third term, we obtain $`_X(i_𝐡\alpha )(Y)`$, whereas the second term is just $`i_{[X,𝐡]}\alpha (Y)`$. $`\mathrm{}`$ |
warning/0507/physics0507054.html | ar5iv | text | # Scaling and data collapse for the mean exit time of asset prices
## I Introduction
The Continuous Time Random Walk (CTRW) formalism introduced four decades ago by Montroll and Weiss montrollweiss has been successfully applied to a wide and diverse variety of physical phenomena over the years weissllibre but only recently to finance general1 ; general2 ; general3 ; general4 ; general5 ; general6 ; ivanov ; masoliver1 ; masoliver2 ; montero . In this latter context, the efforts have been mostly focused on the statistical properties of the waiting time between successive transactions and the asset return at each transaction. Different studies in different markets are conceiving the idea that the empirical distributions of both random variables are compatible with an asymptotic fat tail behavior general1 ; general2 ; general3 ; general4 ; general5 ; general6 ; masoliver1 ; masoliver2 ; ivanov .
Within the CTRW formalism some of us have recently investigated the mean exit time (MET) of asset prices out of a given interval of size $`L`$ for financial time series montero . This study shows that the MET follows a quadratic growth in terms of $`L`$ for small interval lengths $`L`$. In the same study, this functional form was observed for a specific time series of the foreign exchange (FX) market, which is the U.S. dollar/Deutsche mark futures time series montero .
In this paper we investigate both theoretically and empirically the MET of price returns traded in a stock exchange. In our empirical investigation we study the MET of high frequency return time series of 20 highly capitalized stocks traded in New York Stock Exchange. Empirical results about this market confirm that the MET follows a power law with a pre-factor that depends on the specific stock chosen. This observation motivates us to first verify and then release some of the assumptions used in Ref. montero therefore generalizing the model discussed in that paper. The theoretical generalization has been performed by introducing and solving a new two-state chain Markovian model able to both describe the quadratic scaling property of the MET and provide the data collapse of the MET stock pre-factor.
We show that a satisfactory data collapse of the MET is obtained when some degree of autocorrelation in the stock returns is introduced in the two-state chain Markovian model. We attempt to further improve the accuracy by extending the model to a three-state Markov chain for which we are still able to evaluate the MET. Nevertheless, empirical data show that the three-state model does not improve the quality of data collapse in the MET profiles of the different stocks although the theoretical curve shows a better agreement with the empirical data than that of the two-state Markov chain model.
The paper is organized as follows. In Sections II and III, we discuss the MET behavior under the CTRW formalism and a series of simplifying assumptions. In Sect. IV we empirically investigate the scaling and data collapse properties of highly capitalized stock data. Section V relates the time correlation of stock return with the absence of data collapse of MET observed in the previous section. In Sect. VI we introduce and solve a two-state and a three-state Markov chain model to describe the empirical MET observations. Conclusions are drawn in Sect. VII.
## II Mean exit time for i.i.d. processes
In the most common version of the CTRW formalism a given random process $`X(t)`$ shows a series of random increments or jumps at random times $`\mathrm{},t_1,t_0,t_1,t_2,\mathrm{},t_n,\mathrm{}`$ remaining constant between these jumps. Therefore, after a given time interval $`\tau _n=t_nt_{n1}`$, the process experiences a random increment $`\mathrm{\Delta }X_n(\tau _n)=X(t_n)X(t_{n1})`$ and the resulting trajectory consists of a series of steps as shown in Fig. 1. Both waiting times $`\tau _n`$ and random jumps $`\mathrm{\Delta }X_n(\tau _n)`$ are assumed to be independent and identically distributed (i.i.d.) random variables described by their probability density functions (pdfs) which we denote by $`\psi (\tau )`$ and $`h(x)`$ respectively.
However, in the most general representation of the formalism, another function is needed to describe the time evolution of $`X(t)`$. We denote this function by $`\rho (x,\tau )`$ which is the joint pdf of waiting times and jumps:
$$\rho (x,\tau )dxd\tau =\text{Prob}\{x<\mathrm{\Delta }X_nx+dx;\tau <\tau _n\tau +d\tau \}.$$
(1)
Note that the functions $`\psi (\tau )`$ and $`h(x)`$ are the marginal probability density functions of $`\rho (x,\tau )`$. We refer the reader to Ref. montrollweiss ; weissllibre ; general1 ; general2 ; general3 ; general4 ; general5 ; general6 ; masoliver1 ; masoliver2 for a more complete account of the CTRW formalism.
In this paper we will apply the CTRW to study some aspects of the exit problem of financial time series. We will take as underlying random process $`X(t)`$ the logarithmic price $`X(t_n)=\mathrm{ln}(S(t_n))`$, where $`S(t)`$ is the stock price at time $`t`$. We specifically consider the problem of obtaining the mean exit time of $`X(t)`$ out of a given interval $`[a,b]`$. We assume that at certain reference time $`t_0`$, right after an event, the price has a known value $`X(t_0)=x_0`$, $`x_0[a,b]`$. We focus our attention on a particular realization of the process and suppose that at certain time $`t_n>t_0`$ the process first leaves the interval (see Fig. 1). We call the interval $`t_nt_0`$ the exit time out of the region $`[a,b]`$ and denote it by $`t_{a,b}(x_0)`$. This quantity is a random variable since it depends on the particular trajectory of $`X(t)`$ chosen and the MET is simply the average $`T_{a,b}(x_0)=\text{E}[t_{a,b}(x_0)]`$.
The standard approach to exit time problems is based on the knowledge of the survival probability which is generally quite involved weissrubin . However, within the CTRW formalism one can assume that the events compose a series of independent and identically distributed two-dimensional random variables. Under such an assumption, some of us montero have recently shown that one can obtain the MET directly, without making use of the survival probability. In this framework the MET obeys the following integral equation montero ; footnote\_ab
$$T(x_0)=\text{E}[\tau ]+_a^bh(xx_0)T(x)𝑑x,$$
(2)
where
$$\text{E}[\tau ]=_0^{\mathrm{}}\psi (\tau ^{})\tau ^{}𝑑\tau ^{}$$
is the mean waiting time between jumps. It is worth noticing that Eq. (2) is still valid even when $`\tau _n`$ and $`\mathrm{\Delta }X_n`$ are cross-correlated. In fact, in the case of an i.i.d. process the MET only depends on the pdfs of waiting times $`\psi (\tau )`$ and jumps $`h(x)`$, but it does not depend on the particular form of the joint pdf $`\rho (x,\tau )`$. However if we would remove the i.i.d. hypothesis we should specify a functional form for $`\rho (x,\tau )`$.
We now assume that returns increments are distributed according to an even pdf, $`h(x)=h(x)`$, which also satisfies the following scaling condition,
$$h(x)=\frac{1}{\kappa }H\left(\frac{x}{\kappa }\right)$$
(3)
where $`\kappa `$ is the scale of the fluctuations given by the standard deviation $`\kappa `$ of jumps, where $`\kappa ^2=`$ $`\text{E}[\mathrm{\Delta }X_n^2E[\mathrm{\Delta }X_n]^2]`$. The parameter $`\kappa `$ corresponds to the transaction-to-transaction volatility. Under these assumptions some of us showed that the MET out of a small region of size $`Lba\kappa `$ is montero
$`T(a`$ $`+`$ $`L/2)=\text{E}[\tau ][1+2H(0)\left({\displaystyle \frac{L}{2\kappa }}\right)`$ (4)
$`+`$ $`(H^{}(0^+)+4H(0)^2)\left({\displaystyle \frac{L}{2\kappa }}\right)^2+O\left({\displaystyle \frac{L^3}{\kappa ^3}}\right)],`$
where it is assumed that the return process is initially in the middle of the interval $`[a,b]`$.
In Ref. montero some of us have applied the above result to the FX market of the U.S. dollar/Deutsche mark future price. One important conclusion there was that the quadratic growth of the MET is still a good approximation even for large intervals, i.e. $`L\kappa `$. It is less clear nonetheless that the coefficients of the polynomial which we have obtained in Eq. (4) are those that better reproduce the global behavior of the MET. We refer the reader to Ref. montero for a more detailed discussion on this particular aspect of the problem.
In this paper we analyze the scaling properties of the MET for $`20`$ highly capitalized stocks traded at the NYSE in the four year period 1995-1998 and spanning $`1,011`$ trading days. Table 1 shows the list of stocks and the relevant parameters. We have measured the MET for each stock and compared them in the scaled variables $`T(a+L/2)/\text{E}[\tau ]`$ and $`L/2\kappa `$. If all previous hypotheses of the model were correct, and the function $`H(u)`$ is of universal nature, one should observe the same curve for all stocks, i.e. a data collapse, as well as a quadratic growth in $`L`$ at least for small $`L`$. However, Fig. 2 shows that there is a considerable spread of the curves. although the parabolic shape is recovered in all cases not only for small intervals, as expected, but for the whole investigated range of $`L/2\kappa `$ .
## III A discrete state model
Our first objective is to understand why the quadratic term governs both the long and short range behavior of the MET, without a drastic change of the general model and its assumptions. In the present approach we develop a model for $`h(x)`$ based on the small-scale properties of the system.
It is worth noticing that this approach is also used in the context of option pricing, when the fair price of a derivative product is obtained by making use of the binomial trees methodology, where it is assumed that the stock price makes a jump up or down with some probability CRR . Here we introduce the following symmetrical three-state discrete model:
$$h(x)=Q\delta (x)+\frac{1Q}{2}\left[\delta (xc)+\delta (x+c)\right].$$
(5)
where $`Q`$ represents the probability that the price remains unchanged, and $`c`$ is the basic jump size footnote\_grid . By substituting this expression of $`h(x)`$ into Eq. (2) we obtain
$$T(x_0)=\text{E}[\tau ]+QT(x_0)+\frac{1Q}{2}\left[T(x_0+c)+T(x_0c)\right],$$
with the convention that the term $`T(x_0+c)`$ only counts if $`x_0bc`$, and similarly that $`T(x_0c)`$ only appears when $`x_0a+c`$. Let us analyze these two boundary conditions in greater detail. In general, the limits of our interval can be expressed in the following form:
$`a=x_0(l+\epsilon _a)c,`$
$`b=x_0+(m+\epsilon _b)c,`$
with $`l,m`$ and $`\epsilon _a,\epsilon _b[0,1)`$. However it is easy to conclude that $`T(x_0)`$ can depend neither on $`\epsilon _a`$ nor on $`\epsilon _b`$. The only way of leaving the interval is by reaching the points $`x=x_0(l+1)c`$ or $`x=x_0+(m+1)c`$, because $`x=x_0lc`$, and $`x=x_0+mc`$ lay always inside the interval. Therefore we will not loose generality by setting $`\epsilon _a=\epsilon _b=0`$. After that, the length of the interval in the natural scale units of the problem is $`NL/c=l+m`$, and the use of the following notation $`T_nT(a+nc)`$, with $`n\{0,1,\mathrm{},N1,N\}`$, arises in a natural way.
Summing up, for the discrete model given by Eq. (5) the MET out of the interval $`[a,b]`$ obeys the following set of difference equations
$$T_n=\frac{\text{E}[\tau ]}{1Q}+\frac{1}{2}\left(T_{n+1}+T_{n1}\right)$$
(6)
$`(n=0,1,2,\mathrm{},N)`$ with boundary conditions:
$$T_1=T_{N+1}=0.$$
(7)
The solution to Eqs. (6)-(7) is given by
$$T_n=\text{E}[\stackrel{~}{\tau }](n+1)(N+1n),$$
(8)
where the random variable $`\stackrel{~}{\tau }`$ is related to $`\tau `$ in such a way that
$$\text{E}[\stackrel{~}{\tau }]=\frac{\text{E}[\tau ]}{1Q}.$$
(9)
By repeating the above derivation leading to Eq. (6), one can show that the random variable $`\stackrel{~}{\tau }`$ represents the waiting time between jumps if one neglects zero-return trades, i.e. if one identifies the occurrence of a jump when $`X(t)`$ actually changes its value. Thus, for instance, if $`\mathrm{\Delta }X_{i1}0`$, $`\mathrm{\Delta }X_i=0`$, and $`\mathrm{\Delta }X_{i+1}0`$ with corresponding waiting times $`\tau _{i1}`$, $`\tau _i`$, and $`\tau _{i+1}`$, we can replace the pair of events $`(\mathrm{\Delta }X_i,\mathrm{\Delta }X_{i+1})`$ with a single transaction of size $`\mathrm{\Delta }\stackrel{~}{X}_j=\mathrm{\Delta }X_i+\mathrm{\Delta }X_{i+1}=\mathrm{\Delta }X_{i+1}`$, taking a waiting time $`\stackrel{~}{\tau }_j=\tau _i+\tau _{i+1}`$.
From Eq. (8) we see that, for even values of $`N`$, the MET starting from the middle of the interval reads
$$T_{N/2}=\text{E}[\stackrel{~}{\tau }]\left(1+\frac{N}{2}\right)^2.$$
(10)
Looking at the general solution given by Eq. (8) and also at Eq. (10) we clearly observe a quadratic behavior of the MET as a function of $`N`$, that is to say, as a function of the length of the interval. Indeed, from Eq. (10) we have:
$$\frac{T(a+L/2)}{\text{E}[\stackrel{~}{\tau }]}=\left(1+\frac{L}{2\stackrel{~}{\kappa }}\right)^2.$$
(11)
where $`N=L/c`$ and $`\stackrel{~}{\kappa }^2=`$ $`\text{E}[\mathrm{\Delta }\stackrel{~}{X}_n^2E[\mathrm{\Delta }\stackrel{~}{X}_n]^2]`$ $`=c^2`$. The same kind of scaling also holds in terms of the parameters of the three-state model. In fact, from Eq. (9) we get
$$\frac{T(a+L/2)}{\text{E}[\tau ]}=\frac{1}{1Q}\left(1+\frac{L\sqrt{1Q}}{2\kappa }\right)^2.$$
(12)
where $`\kappa ^2=`$ $`\text{E}[\mathrm{\Delta }X_n^2E[\mathrm{\Delta }X_n]^2]`$ $`=(1Q)c^2`$. Hence, for large values of $`L/\kappa `$,
$$\frac{T(a+L/2)}{\text{E}[\tau ]}\left(\frac{L}{2\kappa }\right)^2.$$
(13)
## IV Causes of absence of data collapse
In the previous section we have shown that a simple discrete model for the jump distribution results in a quadratic growth of the MET valid for arbitrary values of the length of the interval and not only for small values of $`L/\kappa `$, as was the case of Eq. (4).
Unfortunately the discrete model does not properly account for the spread of the MET curves observed in Fig. 2 when we consider different stocks. One could argue that this spread can be controlled through the parameter $`Q`$ appearing in Eq. (12) since $`Q`$ may distinguish one stock from another. However, as Eq. (13) shows the MET is practically independent of $`Q`$ for large values of $`L`$ and the difference between stocks would disappear in this range of lengths. Nevertheless, we clearly see in Fig. 2 that the spread between stocks does not tend to vanish but, even in some cases, it increases with $`L`$.
In this Section we try to identify the possible reasons of the failure of the data collapse of our previous model. We revisit some of our assumptions and derive consequences with the aim of finding the most important feature that we are leaving aside. The final goal is to improve our description in the simplest possible way. Let us first summarize some potential causes for the lack of data collapse in the previous models:
1. The probability density $`h(x)`$ is different for different stocks. This would imply different mean exit time curves.
2. There is some dependency on the cross-correlation between waiting times and price returns.
3. The time auto-correlation of waiting times should be included in the model.
4. The time auto-correlation of returns should be included in the model.
We analyze the impact of these hypotheses on the empirical outcomes by performing shuffling experiments. Thus, in order to test the first hypothesis we shuffle independently the time series of $`\mathrm{\Delta }X_n`$ and $`\tau _n`$ and we perform the mean exit time analysis on the shuffled time series. This shuffling destroys all the time- and cross- correlations but it preserves the shape of the pdfs $`h(x)`$ and $`\psi (\tau )`$ and therefore the values of the scaling parameter $`\kappa `$ and the average waiting time $`\text{E}[\tau ]`$. Figure 3 shows a very good collapse indicating that the assumption of a master pdf for the returns of all the stocks is good working hypothesis and it is not the reason for the lack of collapse of Fig. 2. Moreover the MET curves are well fitted by the functional form $`y=(1+x)^2`$.
The other three hypotheses can be similarly tested by performing three different shuffling experiments. Specifically, hypothesis 2) can be tested by shuffling simultaneously the two series and preserving the cross-correlation between $`\mathrm{\Delta }X_n`$ and $`\tau _n`$. Notice that even if Ref. montero has shown that in the absence of autocorrelation of waiting times and returns (i.i.d. model) the MET is independent of the cross-correlations $`\rho (x,\tau )`$, in the general case $`\rho (x,\tau )`$ may play a role. Hypothesis 3) can be tested by shuffling only the series of waiting times and preserving the order in the series of returns. Finally, hypothesis 4) can be tested by shuffling only the time series of returns and preserving the order of the series of waiting times. Figure 4 shows the results of these shuffling experiments for the General Electric (GE) stock. The Figure shows that neglecting the autocorrelation function of waiting times does not change the MET curve. On the other hand, when one destroys the auto correlation of returns the MET (star) changes dramatically and becomes close to the one predicted by the i.i.d. model.
By summarizing, in order to have a good model of the mean exit time one cannot neglect the correlation properties of price return, whereas the other correlations can be neglected as a first approximation.
## V The correlation of return increments
What is the origin of correlation of returns? There are, in principle, two possible answers. One contribution comes from the linear autocorrelation of the increments of returns given by $`\text{E}[\mathrm{\Delta }X_n\mathrm{\Delta }X_m]`$. A second contribution is related to nonlinear properties which can be exemplified by the nonlinear correlation $`\text{E}[|\mathrm{\Delta }X_n||\mathrm{\Delta }X_m|]`$.
The first contribution can be easily evaluated by taking the linear autocorrelation function of the increments of price returns. Figure 5 shows this quantity for GE (solid line). In the Figure it is clear that for a lag of one trade the linear autocorrelation function is negative and significantly different from zero. This is a known effect of transaction prices which is due to the presence of a spread between the best bid and the best ask (the “bid-ask bounce”, see for example campbell ). It is reasonable to assume that this short range correlation can be included in the model by using a Markov process.
The nonlinear correlation is related to the volatility and can be quantified by plotting the autocorrelation function of $`|\mathrm{\Delta }X_n|`$. The result (dashed line) in Fig. 5 is a slowly decaying function indicating a long range correlation, which is probably not compatible with a Markovian model.
We can test the relative importance of the two contributions to the correlation of returns by performing another shuffling experiment. We can obtain a surrogate time series with the same linear autocorrelation function but with an uncorrelated volatility (absolute value of return increments). The method (see for example Chapter 7 of Ref. kantz ) consists in taking the Fourier transform of the original time series and then randomize its phases. Because of the Wiener-Kinchine theorem the linear autocorrelation of the surrogate time series is the same as the original, but the nonlinear correlation will be zero. Therefore, we have a time series with the same bid-ask bounce properties but with an uncorrelated volatility. One difficulty of the method is the fact that the pdf of the surrogate series will be in general different from the original one (unless the time series is Gaussian). Since we know that, in general, the MET depends on the return pdf, we should control that the distortion of the pdf introduced by the phase randomization is not critical in changing the properties of the MET. To this end we first compare the MET curve of the shuffled original time series to the shuffled phase-randomized time series. These two series are both i.i.d. but with different pdfs. Figure 6 shows that the two METs (dashed line and squares) are very close, thus indicating that the distortion of the pdf introduced by the phase randomization is not critical for the MET properties.
We can now compare the METs of the original data to the ones for the phase-randomized time series. The two series have the same linear autocorrelation function, but the first one displays a clustered volatility whereas the randomized series does not. The two METs (see solid line and circles in Fig. 6) are again very close, thus showing that the most important time correlation contribution to the MET is the linear autocorrelation (bid-ask bounce), while the clustered volatility plays a minor role.
## VI Mean exit time for a Markov-chain model
In order to incorporate the linear correlations of returns, we derive an integral equation for the mean exit time out of a given interval when the driving process is a continuous time random walk with memory. In particular, we will consider Markov processes within the CTRW framework. Such models lead to the following joint conditional probability density function:
$$\rho (x,\tau |x^{},\tau ^{})dxd\tau =\text{Prob}\{x<\mathrm{\Delta }X_nx+dx;\tau <\tau _n\tau +d\tau |\mathrm{\Delta }X_{n1}=x^{};\tau _{n1}=\tau ^{}\}.$$
(14)
In this Markovian case the conditional MET, $`T(x_0|\mathrm{\Delta }X_0,\tau _0)`$, will also depend on both the magnitude of the previous jump $`\mathrm{\Delta }X_0=x_0x_1`$ and its sojourn time $`\tau _0=t_0t_1`$ (see Fig. 1). Now, and contrary to Eq. (2), the integral equation for the conditional MET depends on the complete joint probability density function. It reads:
$$T(x_0|\mathrm{\Delta }X_0,\tau _0)=\text{E}\left[\tau |\mathrm{\Delta }X_0,\tau _0\right]+_0^{\mathrm{}}𝑑\tau _a^b\rho (xx_0,\tau |\mathrm{\Delta }X_0,\tau _0)T(x|\mathrm{\Delta }X,\tau )𝑑x,$$
(15)
where $`\mathrm{\Delta }X=xx_0`$ and $`\tau =tt_0`$. The level of complexity of Eq. (15) can be considerably reduced by noting that, as we have shown in Sect. V, it is possible to remove the correlation between consecutive waiting times without affecting the MET. Therefore, we will assume that the correlation involving waiting times is negligible, and that all relevant information we have to consider when dealing with the $`n`$-th event is the magnitude of the previous change. In such a case instead of Eq. (14) we write
$$\rho (x,\tau |x^{})dxd\tau =\text{Prob}\{x<\mathrm{\Delta }X_nx+dx;\tau <\tau _n\tau +d\tau |\mathrm{\Delta }X_{n1}=x^{}\}.$$
(16)
Hence, the integral equation for the MET is simpler, because on the right hand side of Eq. (15) we can perform the integral over time. We thus obtain
$`T(x_0|\mathrm{\Delta }X_0)`$ $`=`$ $`\text{E}\left[\tau |\mathrm{\Delta }X_0\right]`$ (17)
$`+`$ $`{\displaystyle _a^b}h(xx_0|\mathrm{\Delta }X_0)T(x|\mathrm{\Delta }X)𝑑x.`$
In this case the MET only depends on the marginal probability density function of the return increments, $`h(x|\mathrm{\Delta }X_0)`$,
$$h(x|\mathrm{\Delta }X_0)=_0^{\mathrm{}}\rho (x,\tau |\mathrm{\Delta }X_0)𝑑\tau ,$$
and on the conditional expectation of waiting times $`\text{E}\left[\tau |\mathrm{\Delta }X_0\right]`$ which has to be evaluated through the marginal pdf, $`\psi (\tau |\mathrm{\Delta }X_0)`$,
$$\psi (\tau |\mathrm{\Delta }X_0)=_{\mathrm{}}^{\mathrm{}}\rho (x,\tau |\mathrm{\Delta }X_0)𝑑x.$$
We finally observe that although $`x_0[a,b]`$, we let $`\mathrm{\Delta }X_0`$ to be any real number.
### VI.1 A two-state Markov chain model
In order to solve Eq. (17) and obtain explicit expressions for the MET that can be compared with empirical data, we follow the same approach of Sect. III and choose a discrete model for $`h(x|\mathrm{\Delta }X_0)`$. At this point we can opt for a two-state model in which, at any time step, returns can only go up and down a fixed quantity $`c`$, or for a three-state model where in addition the return increment can be zero. We have shown in Sect. III that for an i.i.d. process both alternatives are equivalent. In the case of a Markovian process the equivalence is not complete. As we will see below, the final expressions obtained for the unconditional MET are slightly different although, for large values of $`L`$, the leading term is the same in both cases.
Let us start with a two-state Markov chain model. In the symmetrical case in which up and down movements are equally likely, the conditional pdf for return increments is
$$h(x|y)=\frac{c+ry}{2c}\delta (xc)+\frac{cry}{2c}\delta (x+c),$$
(18)
where $`r`$ is the correlation between the magnitude of two consecutive jumps:
$$r\frac{\text{Cov}[\mathrm{\Delta }\stackrel{~}{X}_n,\mathrm{\Delta }\stackrel{~}{X}_{n1}]}{\sqrt{\text{Var}[\mathrm{\Delta }\stackrel{~}{X}_n]\text{Var}[\mathrm{\Delta }\stackrel{~}{X}_{n1}]}}.$$
(19)
From Eq. (18) we see that the squared volatility,
$$\stackrel{~}{\kappa }^2(y)_{\mathrm{}}^{\mathrm{}}x^2h(x|y)𝑑x=c^2,$$
is independent of $`y`$. By substituting Eq. (18) into Eq. (17) we get the following difference equation for the MET:
$`T(x_0|\mathrm{\Delta }\stackrel{~}{X}_0)`$ $`=`$ $`\text{E}[\stackrel{~}{\tau }|\mathrm{\Delta }\stackrel{~}{X_0}]+{\displaystyle \frac{c+r\mathrm{\Delta }\stackrel{~}{X_0}}{2c}}T(x_0+c|c)`$ (20)
$`+`$ $`{\displaystyle \frac{cr\mathrm{\Delta }\stackrel{~}{X_0}}{2c}}T(x_0c|c),`$
where $`\mathrm{\Delta }\stackrel{~}{X_0}=\pm c`$, $`T(x_0+c|c)=0`$ if $`x_0>bc`$ and $`T(x_0c|c)=0`$ if $`x_0<a+c`$. We extend the notation introduced in Sect. III and define
$$T_{n,n1}T(x_0=a+nc|\mathrm{\Delta }\stackrel{~}{X}_0=\pm c).$$
(21)
Now, Eq. (20) is equivalent to the following set of recurrence equations:
$`T_{n,n1}`$ $`=`$ $`\text{E}[\stackrel{~}{\tau }]+{\displaystyle \frac{1+r}{2}}T_{n+1,n}+{\displaystyle \frac{1r}{2}}T_{n1,n},`$ (22)
$`T_{n,n+1}`$ $`=`$ $`\text{E}[\stackrel{~}{\tau }]+{\displaystyle \frac{1r}{2}}T_{n+1,n}+{\displaystyle \frac{1+r}{2}}T_{n1,n},`$ (23)
($`n=0,1,\mathrm{},N`$) with boundary conditions:
$$T_{1,0}=T_{N+1,N}=0.$$
(24)
Note that in writing Eqs. (22)-(23) we have set $`\text{E}[\stackrel{~}{\tau }|\pm c]=\text{E}[\stackrel{~}{\tau }]`$ which is consistent with the assumed symmetry between up and down movements (see also Eq. (25) below). The solution to problem (22)-(24) reads
$`T_{n,n1}`$ $`=`$ $`\text{E}[\stackrel{~}{\tau }](N+1n)\left[1+n{\displaystyle \frac{1r}{1+r}}\right],`$
$`T_{n,n+1}`$ $`=`$ $`\text{E}[\stackrel{~}{\tau }](n+1)\left[1+(Nn){\displaystyle \frac{1r}{1+r}}\right].`$
The quantity of interest for our analysis is the unconditional MET $`T_n`$, which is related to $`T_{n,n\pm 1}`$ by
$$T_n=\frac{1}{2}\left(T_{n,n1}+T_{n,n+1}\right),$$
(25)
that is,
$$T_n=\text{E}[\stackrel{~}{\tau }]\left[1+\frac{N}{1+r}+\frac{1r}{1+r}n(Nn)\right].$$
(26)
Thus the MET starting from the middle of the interval reads
$$T_{N/2}=\text{E}[\stackrel{~}{\tau }]\left[\frac{2r}{1+r}\left(1+\frac{N}{2}\right)+\frac{1r}{1+r}\left(1+\frac{N}{2}\right)^2\right].$$
(27)
In this case, and contrary to the i.i.d. case given in Eq. (10), the MET $`T_{N/2}`$ is not a perfect quadratic expression. However, taking into account that $`N=L/c`$ and $`\stackrel{~}{\kappa }=c`$ we have
$$\frac{T(a+L/2)}{\text{E}[\stackrel{~}{\tau }]}=\frac{2r}{1+r}\left(1+\frac{L}{2\stackrel{~}{\kappa }}\right)+\frac{1r}{1+r}\left(1+\frac{L}{2\stackrel{~}{\kappa }}\right)^2,$$
(28)
and for large values of $`L/\stackrel{~}{\kappa }`$ we recover the expected quadratic behavior in the leading term:
$$\frac{T(a+L/2)}{\text{E}[\stackrel{~}{\tau }]}\frac{1r}{1+r}\left(1+\frac{L}{2\stackrel{~}{\kappa }}\right)^2.$$
(29)
Note that the value of $`r`$ depends on each particular stock. Therefore, the scaled MET defined as:
$$T_{sc}(L)\left(\frac{1+r}{1r}\right)\frac{T(a+L/2)}{\text{E}[\stackrel{~}{\tau }]},$$
(30)
tends, for increasing values of $`L`$, to a quadratic function of the interval length which is independent of the particular stock chosen.
In Fig. 7 we compare the MET curves for the data scaled in the original way (inset), as given in Eq. (11), and for the data scaled according to Eq. (30). The inset of Figure 7 is essentially the same as Fig. 2. Figure 7 shows that the scaling of Eq. (29) gives a significant improvement with respect to the original one. We note that the scaled curves are systematically below the curve $`y=(1+x)^2`$, probably because some correlation is not taken correctly into account by the two-state model. We will now attempt to correct this bias by introducing a three-state model.
### VI.2 A three-state Markov chain model
One can argue that the two-state model just developed would need an improvement in order to include zero-return transactions, i.e. those with $`\mathrm{\Delta }X_n=0`$. In Sect. III we have shown that for the i.i.d. process the inclusion of a third possible state is completely equivalent to a two-state (up and down) model after redefining the mean waiting time and the volatility $`\kappa `$ by including the probability $`Q`$ of zero-return transactions. However, when memory is present, as is now the case, this equivalence is not complete. We thus outline a discrete three-state Markov chain model.
The Markov-chain model is now characterized by the following transition matrix:
$$𝐓=\left(\begin{array}{ccc}P(|)& P(|0)& P(|+)\\ P(0|)& P(0|0)& P(0|+)\\ P(+|)& P(+|0)& P(+|+)\end{array}\right),$$
(31)
where $`P(|)\text{Prob}\{\mathrm{\Delta }X_n=c|\mathrm{\Delta }X_{n1}=c\}`$ and similar definitions for the rest of the matrix elements. Since we are also assuming that the process is symmetrical for positive and negative returns, the transition matrix $`𝐓`$ can be written in the following form:
$$𝐓=\left(\begin{array}{ccc}\frac{1+r}{2}p& \frac{1q}{2}& \frac{1r}{2}p\\ 1p& q& 1p\\ \frac{1r}{2}p& \frac{1q}{2}& \frac{1+r}{2}p\end{array}\right),$$
(32)
where $`q=P(0|0)`$ is the probability for trapping,
$$p=P(|)+P(+|)=P(+|+)+P(|+)$$
and $`r`$ measures the strength of the persistence:
$$r=\frac{P(|)P(+|)}{P(|)+P(+|)}=\frac{P(+|+)P(|+)}{P(+|+)+P(|+)}.$$
Note that the first order autocorrelation coefficient, defined as on the right-hand side of Eq. (19) is now given by (cf. Eq. (19))
$`{\displaystyle \frac{\text{Cov}[\mathrm{\Delta }X_n,\mathrm{\Delta }X_{n1}]}{\sqrt{\text{Var}[\mathrm{\Delta }X_n]\text{Var}[\mathrm{\Delta }X_{n1}]}}}=pr.`$ (33)
We also need to specify the unconditional probabilities of each state:
$$𝐏=\left(\begin{array}{c}P()\\ P(0)\\ P(+)\end{array}\right)\left(\begin{array}{c}\frac{1Q}{2}\\ Q\\ \frac{1Q}{2}\end{array}\right).$$
(34)
However, in this case $`Q`$ is not an independent parameter as was the case of the i.i.d. model (cf Sect. III). Indeed, using the total probability formula and taking into account the values of the transition matrix $`𝐓`$ and the vector $`𝐏`$, we get
$$p=1\frac{1q}{1Q}Q.$$
Observe that when $`q=Q`$ we have $`p=1Q`$ and there is no trapping in the value of the random process.
Now the pdf of the returns reads
$`h(x|y)`$ $`=`$ $`a(y)\delta (x)`$
$`+`$ $`{\displaystyle \frac{(1a(y))}{2c}}\left[\left(c+ry\right)\delta (xc)+\left(cry\right)\delta (x+c)\right],`$
where
$$a(y)=\{\begin{array}{cc}q,\hfill & \text{if }y=0\text{,}\hfill \\ 1p,\hfill & \text{if }y0\text{.}\hfill \end{array}$$
In this case the integral equation (17) is equivalent to the following set of difference equations for $`T_{n,n}`$ and $`T_{n,n\pm 1}`$:
$`T_{n,n1}`$ $`=`$ $`\text{E}[\tau |c]+(1p)T_{n,n}`$ (36)
$`+`$ $`p\left[{\displaystyle \frac{1+r}{2}}T_{n+1,n}+{\displaystyle \frac{1r}{2}}T_{n1,n}\right]`$
$`T_{n,n}=\text{E}[\tau |0]`$ $`+`$ $`qT_{n,n}`$ (37)
$`+`$ $`{\displaystyle \frac{1q}{2}}\left[T_{n+1,n}+T_{n1,n}\right]`$
$`T_{n,n+1}`$ $`=`$ $`\text{E}[\tau |c]+(1p)T_{n,n}`$ (38)
$`+`$ $`p\left[{\displaystyle \frac{1r}{2}}T_{n+1,n}+{\displaystyle \frac{1+r}{2}}T_{n1,n}\right]`$
where $`T_{n,n\pm 1}`$ are defined as in Eq. (21) and
$$T_{n,n}=T(x_0=a+nc|\mathrm{\Delta }X_0=0).$$
In writing Eqs. (36) and (38) we have taken into account the symmetry
$$\text{E}[\tau |+c]=\text{E}[\tau |c]\text{E}[\tau |c].$$
Finally the solution to Eqs. (36)-(38) with boundary conditions
$$T_{1,0}=T_{N+1,N}=0$$
(cf Eq. (24)) reads
$$T_{n,n1}=\frac{\text{E}[\tau ]}{1Q}(N+1n)\left[1+n\frac{1pr}{1+pr}\right]$$
(39)
$`T_{n,n}={\displaystyle \frac{\text{E}[\tau |0]}{1q}}`$ $`+`$ $`{\displaystyle \frac{\text{E}[\tau ]}{2(1Q)}}[{\displaystyle \frac{1pr}{1+pr}}[(Nn)(n+1)`$ (40)
$`+`$ $`n(Nn+1)]+N]`$
$$T_{n,n+1}=\frac{\text{E}[\tau ]}{1Q}(n+1)\left[1+(Nn)\frac{1pr}{1+pr}\right],$$
(41)
where $`\text{E}[\tau ]`$ is the (unconditional) mean waiting time which is related to $`\text{E}[\tau |c]`$ and $`\text{E}[\tau |0]`$ by
$$\text{E}[\tau ]=Q\text{E}[\tau |0]+(1Q)\text{E}[\tau |c].$$
In terms of $`T_{n,n\pm 1}`$ and $`T_{n,n}`$ the unconditional MET $`T_n`$ is given by
$$T_n=QT_{n,n}+\left(\frac{1Q}{2}\right)(T_{n,n1}+T_{n,n+1}),$$
and starting from the center of the interval we explicitly have
$`T_{N/2}`$ $`=`$ $`{\displaystyle \frac{\text{E}[\tau ]}{1Q}}\left[{\displaystyle \frac{2pr}{1+pr}}\left(1+{\displaystyle \frac{N}{2}}\right)+{\displaystyle \frac{1pr}{1+pr}}\left(1+{\displaystyle \frac{N}{2}}\right)^2\right]`$ (42)
$`+`$ $`Q\left[{\displaystyle \frac{E[\tau |0]}{1q}}{\displaystyle \frac{E[\tau ]}{1Q}}\right].`$
The main difference between this expression and Eq. (27) is the final constant term that accounts for the two kind of trapping that the system may experience: the probabilistic one, $`qQ`$, and the temporal one, $`E[\tau |0]E[\tau ]`$. By using the fact that $`\kappa ^2=(1Q)c^2`$, the leading term is again of the form:
$`{\displaystyle \frac{T(a+L/2)}{\text{E}[\tau ]}}{\displaystyle \frac{1pr}{1+pr}}\left({\displaystyle \frac{L}{2\kappa }}\right)^2.`$ (43)
In Fig. 8 we show the MET curves for the data scaled according to Eq. (43). It is worth noting that the quantity $`pr`$ appearing in Eq. (43) can be estimated in two different ways. One could separately compute $`p`$, the probability of a change in the price return provided a previous change, and $`r`$ which is the strength of the persistence. Alternatively one can estimate directly the quantity $`pr`$ by using Eq. (33). This second approach has the advantage of reducing the dependence of the estimates from the specific details of the model. This is the reason why in the curves shown in Fig. 8 the quantity $`pr`$ has been estimated by using Eq. (33). The scaling shown in Figure 8 is not satisfactory, we would even say that it is less satisfactory than the scaling corresponding to the two-state model which has been shown in Fig. 7. However, it is worth noting that the rescaled curves are not systematically below the parabolic curve $`(1+x)^2`$ as it was in the case of the two-state model.
## VII Conclusions
This paper presents theoretical and empirical results about the MET of financial time series. Specifically the scaling property of the MET as a function of the size $`L`$ has been confirmed to follow a quadratic law for a number of stock price time series. We empirically verify that the quadratic scaling law has associated a pre-factor which is specific to the analyzed stock. We have performed a series of tests to determine which kind of correlation are responsible for this dependence. It turned out that the main contribution is associated with the linear autocorrelation property of stock returns.
We have therefore introduced and solved analytically both a two-state and a three-state Markov chain models. The analytical results obtained through the two-state model allow us to get a quite satisfactory data collapse of the 20 MET profiles into a single parabolic curve as predicted by the model. However, this parabolic curve appears to be systematically above real data, that is, the model overstimates the mean exit time.
We have been able to solve a three-state Markov chain model as well. Unfortunately this more detailed model does not provide an improvement on data collapse. The main advantage of this generalization is that the MET provided by the model lies close to the empirical curves. In other words, the three-state model does not overestimate the METs.
We do not have a convincing explanation for this observation but only some indications. Specifically, we have seen that the symmetries assumed in the three state model are not present in some empirical transition matrices transition . Perhaps this assumption prevents the data from a convincing data collapse and the system would perform a better collapse in a model taking into account a certain degree of asymmetry in the Markovian transition matrix. However, obtaining the analytical solution for this more general case seems to be very involved and it has been left for future research.
In conclusion, the MET and the search of a data collapse in the MET curves of stock prices provide a good occasion for testing the underlying hypothesis characterizing the return dynamics and for the improvement of the CTRW models describing this phenomenon. We believe we have detected the essential ingredients to be accounted for a feasible model within the CTRW. We hope that the results may provide a certain insight on the way to bridge the gap between the description of the stochastic dynamics at very short time horizons with that of longer time scales campbell ; ohara ; gabaix1 ; gabaix2 ; gabaix3 ; gabaix4 ; gabaix5 .
###### Acknowledgements.
The authors acknowledge support from the ESF project “Cost Action P10 Physics of Risk”. M.M., J.P. and J.M. acknowledge partial support from Dirección General de Investigación under contract No. BFM2003-04574 and by Generalitat de Catalunya under contract No. 2001 SGR-00061. F.L., S.M. and R.N.M acknowledge support from the research project MIUR-FIRB RBNE01CW3M “Cellular Self-Organizing nets and chaotic nonlinear dynamics to model and control complex system”, from the research project MIUR 449/97 “High frequency dynamics in financial markets” and from the European Union STREP project n. 012911 “Human behavior through dynamics of complex social networks: an interdisciplinary approach.”. |
warning/0507/math0507109.html | ar5iv | text | # Mathematical computability questions for some classes of linear and non-linear differential equations originated from Hilbert’s tenth problem
## Introduction
Hilbert’s tenth problem is concerned with the availability of a universal procedure to determine whether an arbitrarily given Diophantine equation
(1) $`D(x_1,\mathrm{},x_K)`$ $`=`$ $`0`$
has any positive integer solution or not. After more than 70 years since its inception, it has finally been shown that the problem is recursively noncomputable since no such universal procedure which is also classically recursive can exist. The problem is equivalent to the Turing halting problem and intimately links to the concept of effective computability as defined by the Church-Turing thesis in classical recursion theory.
Nevertheless, we have established elsewhere, through an inspiration provided by quantum mechanics , some surprising connections of the above problem in number theory to problems over the continuous variables. In particular, we have been able to reformulate Hilbert’s tenth problem in terms of a set of infinitely coupled non-linear differential equations for any given Diophantine equation . Also, through the framework of quantum adiabatic computation , we have also associated Hilbert’s tenth problem with a class of linear Schrödinger equations with appropriate time-dependent Hamiltonians. It has been proposed then that a physical implementation of quantum mechanical processes for these Schrödinger equations could provide the physical means to solve Hilbert’s tenth problem . Here, in this paper we will only concern ourselves with the differential equations as mathematical objects, however, and will not appeal to any real physical processes. The mathematical objects so derived are extremely valuable as they provide us a direct link between the well-established classical recursion theory and the infantile subject of computable analysis, through a known noncomputable problem in the former theory.
We especially want to raise in this paper the computability questions for these differential equations in the domain of computable analysis, outside and encompassing the classical recursion theory where Hilbert’s tenth problem was originally formulated.
From the recursive noncomputability of Hilbert’s tenth problem, one might conclude that these differential equations should also be noncomputable in the wider framework. Such hasty conclusion, however, is not warranted because of several reasons. Firstly, however likely the case one might expect, it would need to be established rigorously as a mathematical truth –because the unsolvability of Hilbert’s tenth problem is only established in the framework of Turing computability, not necessarily in mathematics in general.
Secondly, there are many computation models in computable analysis but they are not all equivalent. And it is known that computability in one model may not be the same in some other model, see , for example, for a brief discussion comparison of the various models. This situation is in stark contrast to the classical recursion theory of functions from $``$ to $``$. There, many different formulations have been given (notably by Kleene, Turing, Post, Herbrand/Gödel, Markov) but in the end these all lead to the same notion of computability with the same class of computable functions. Such an equivalence has led to the postulation and support of the Church-Turing thesis in that theory.
The computability definitions for a single real number in most, if not all, different computation models of computable analysis are equivalent. However, for sequences of reals the definitions diverge and are not all equivalent. (The discussions on sequences of real numbers are necessary because of the necessity of topological notions in analysis.) This divergence results in the dependence of the notion of computability on the different computation models,<sup>1</sup><sup>1</sup>1For example, the so-called real-RAM approach introduces computable real functions directly and borrows only the concept of control structure from Turing computation, without any further referencing to the latter. The computability notion in this model, as a result, is different from that of other real computation models that are based and built from the Turing computation. or even on different choices within a single model.<sup>2</sup><sup>2</sup>2One famous example is that the choice of different norms in a Banach space can lead to opposing conclusions about the computability of solutions of the same wave equation in three spatial dimensions with computable initial functions .
In view of such an inequivalence of computability in different approaches, it is not at all a forgone conclusion that the noncomputablity of the differential equations mentioned above is trivially the only, inevitable possibility.
In the next Section we briefly present the observation inspired by quantum mechanics that leads to the connections between Hilbert’s tenth problem with unbounded self-adjoint operators acting on some infinite-dimensional Hilbert space. From this we then reformulate Hilbert’s tenth problem in terms of a set of infinitely coupled non-linear differential equations (eqs. (15, 16) below). We also propose a so-called continuation procedure to approximate some relevant part of its solution; the computability of the continuation procedure is left as an unanswered question for the time being.
We then present a class of linear Schrödinger equations (eq. (18) below) with a special class of time-dependent Hamitonians that are intimately connected to Hilbert’s tenth problem. We call this the dynamical approach, because of its use of the Schrödinger equations, to distinguish it from the kinematic approach above from which we derive the set of non-linear differential equations. We conclude the paper with some remarks and a discussion on the possible implications, including those on the Church-Turing thesis, if the differential equations are indeed computable in some computation model of computable analysis.
## Hilbert’s tenth problem and unbounded operators in Hilbert spaces
Given a Diophantine equation with $`K`$ unknowns $`x`$’s as in eq. (1), we can make a connection, following , with the following self-adjoint operator acting on some appropriate Fock space (a special type of Hilbert space)
(2) $`H_P`$ $`=`$ $`\left(D(a_1^{}a_1,\mathrm{},a_K^{}a_K)\right)^2,`$
where
(3) $`[a_j,a_k^{}]=\delta _{jk},`$ $`[a_k,a_j]=0,`$
which are usually termed the creation and annihilation operators, and most commonly seen in text-book treatment of the quantum simple harmonic oscillators. The Fock space is built out of the “vacuum” $`_{j=1}^K|0_j`$ by repeating applications of the creation operators $`a_j^{}`$.
The operator (2) has non-negative and discrete eigenvalues $`(D(n_1,\mathrm{},n_K))^2`$, with natural numbers $`n_1,\mathrm{},n_K`$. There is an eigenstate $`|E_g`$ corresponding to the smallest eigenvalue $`E_g`$.<sup>3</sup><sup>3</sup>3Assuming that we could always, if we need to, eliminate any degeneracy of the eigenvalue by, for example, modifying the Diophantine equation or by adding some small perturbation terms in $`H_P`$. If the self-adjoint operator is considered as a Hamiltonian for some dynamical process then these are respectively the ground state and its energy.
It is then clear that the Diophantine equation (1) has at least one integer solution if and only if $`E_g=\left(D(n_1^{(0)},\mathrm{},n_K^{(0)})\right)^2=0`$, for some $`K`$-tuple of natural numbers $`(n_1^{(0)},\mathrm{},n_K^{(0)})`$.
To sort out this $`E_g`$ among the infinitely many eigenvalues is almost an impossible task. The strategy we will employ, as inspired by quantum adiabatic processes, is to tag the state $`|E_g`$ by some other known state $`|E_I`$ which is the ground state of some other self-adjoint operator $`H_I`$, which can be smoothly deformed to $`H_P`$ through some continuous parameter $`s[0,1]`$. To that end, we consider the interpolating operator
(4) $`(s)`$ $`=`$ $`H_I+f(s)(H_PH_I),`$
$``$ $`H_I+f(s)W,`$
which has an eigenproblem at each instant $`s`$,
(5) $`[(s)E_q(s)]|E_q(s)=0,`$ $`q=0,1,\mathrm{}`$
with the subscript ordering according to the sizes of the eigenvalues, and $`f(s)`$ some continuous and monotonically increasing function in $`[0,1]`$
(6) $`f(0)=0;`$ $`f(1)=1.`$
Clearly, $`E_0(0)=E_I`$ and $`E_0(1)=E_g`$.
A suitable $`H_I`$ is, where $`\alpha `$’s $``$,
(7) $`H_I={\displaystyle \underset{i=1}{\overset{K}{}}}\lambda _i(a_i^{}\alpha _i^{})(a_i\alpha _i),`$
which we will employ from now on. Here, $`E_I=0`$ and $`|E_I=|\alpha _1\mathrm{}\alpha _K`$ is the Cartesian product of the coherent states
(8) $`|\alpha _i`$ $`=`$ $`\mathrm{e}^{\frac{|\alpha _i|^2}{2}}{\displaystyle \underset{n_i=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{\alpha _i^n}{\sqrt{n_i!}}}|n_i.`$
where $`|n_i`$ are the eigenstates of $`a_i^{}a_i`$ with eigenvalues $`n_i`$. The $`\lambda _i`$ can be chosen to be rational or even irrational numbers such that the first order equation
(9) $`{\displaystyle \underset{i=1}{\overset{K}{}}}\lambda _ip_i`$ $`=`$ $`0,`$
has no integer solutions in $`p_i`$. This condition is to ensure that all the eigenvalues of $`H_I`$ are non-degenerate, since the eigenvalues of $`H_I`$, which are of the form $`_{i=1}^K\lambda _in_i`$, are then easily seen to be unique for different $`K`$-tuples of natural numbers $`n`$’s.
## The spectral flow – The “kinematic” approach
We now derive the differential equations for the tagging connection between the instantaneous eigenvalues and eigenvectors at different instant $`s`$ in (5).
Note firstly that, from the normalisation condition $`E_q|E_q=1`$, we can write
$`E_q|_s|E_q`$ $`=`$ $`i_s\varphi _q,`$
for some real $`\varphi _q`$. This can be absorbed away with the redefinition
(10) $`\mathrm{e}^{i\varphi _q(s)}|E_q(s)`$ $``$ $`|E_q(s),`$
upon which
(11) $`E_q|_s|E_q`$ $`=`$ $`0.`$
Differentiating (5) with respect to $`s`$ yields
(12) $`[f^{}(s)W_sE_q]|E_q+[E_q]_s|E_q`$ $`=`$ $`0.`$
We next insert the resolution of unity at each instant $`s`$,
$`\mathrm{𝟏}`$ $`=`$ $`{\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}|E_m(s)E_m(s)|,`$
just after $``$ in (12) to get, by virtue of (11),
(13) $`E_q_s|E_q`$ $`=`$ $`[f^{}(s)W_sE_q]|E_q+{\displaystyle \underset{mq}{\overset{\mathrm{}}{}}}E_mE_m|_s|E_q|E_m.`$
The inner product of the last equation with $`|E_l`$ gives
(14) $`(E_qE_l)E_l|_s|E_q`$ $`=`$ $`f^{}(s)E_l|W|E_q_sE_q\delta _{ql}.`$
Thus, for $`ql`$ this gives the components of $`_s|E_q`$ in $`|E_l`$, provided $`E_q(s)E_l(s)`$ at any $`s(0,1)`$, a condition whose proof has been outlined in . Consequently, together with (11),
(15) $`_s|E_q`$ $`=`$ $`f^{}(s){\displaystyle \underset{lq}{\overset{\mathrm{}}{}}}{\displaystyle \frac{E_l|W|E_q}{E_qE_l}}|E_l.`$
Also, putting $`q=l`$ in (14) we have
(16) $`_sE_q(s)`$ $`=`$ $`f^{}(s)E_q(s)|W|E_q(s).`$
Equations (15) and (16) form the set of infinitely coupled differential equations providing the tagging linkage we have been looking for.
In this reformulation, the Diophantine equation (1) has an integer solution if and only if
(17) $`\underset{s1}{lim}E_0(s)`$ $`=`$ $`0,`$
from the constructively known eigenvalues and eigenstates of $`H_I`$ as the initial conditions. The limiting process might be necessary since $`H_P`$, i.e. $`(1)`$, may have a degenerate ground state eigenvalue in general.
## “Kinematic” continuation procedure?
The non-linear differential equations of the last Section are infinitely coupled and may not be solved explicitly or computably in general. However, we are only interested in the ground state eigenvalue $`E_0(1)`$ being zero or not. And since the influence on the ground state by states having larger and larger indices diminishes more and more thanks to the denominators in (15), this information might be derived in some approximation scheme in which the number of states involved is truncated to a finite number. The size of the truncation cannot be universal and must of course depend on the particular Diophantine equation under consideration.
In the below we speculate on an analytic approximation under the name of continuation procedure, and we make no claim about its computability here but leave it as a challenge for the future.
* Starting from the initial condition comprising of the constructively known eigenvalues and eigenvectors of $`H_I`$ at $`s=0`$, the differential equations (15) and 16) give us the series expansions
$`|E_q(ϵ_1)`$ $`=`$ $`|E_q(0)+ϵ_1f^{}(0){\displaystyle \underset{lq}{\overset{N_1}{}}}{\displaystyle \frac{E_l(0)|W|E_q(0)}{E_q(0)E_l(0)}}|E_l(0)+_1,`$
$`E_q(ϵ_1)`$ $`=`$ $`E_q(0)+ϵ_1f^{}(0)E_q(0)|W|E_q(0)+𝔔_1.`$
* If the remainders $`_1`$ and $`𝔔_1`$ above are computable, we would be able to evaluate the radii of convergence in $`s`$, which contain $`s=ϵ_1`$, and also the truncation size $`N_1`$ which determines the accuracy of the expansions..
* We then proceed to evaluate new series approximations similar to the ones above but this time centred at $`s=ϵ_1`$. The new series have new radii of convergence and a new truncation of $`N_2`$ eigenvectors previously approximated at $`s=ϵ_1`$. After this step we have then covered a finite domain in $`s`$ away from zero, with some computable degree of accuracy.
* We keep reiterating this procedure, if possible, to obtain new remainders and radii of convergence and thus extend the covered domain in $`s`$, until we could evaluate the limit $`lim_{s0}E_0(s)`$.
This is reminiscent of the procedure of analytic continuation of functions in complex analysis.
Note that at each step of the above procedure we only require some finite truncation $`N_i`$ for a given accuracy of the series approximations. (That accuracy would also “recursively” determine the truncations $`N_i`$ at all previous steps, $`j<i`$.) Note also that, in general, the condition $`x=0`$ for a computable real number $`x`$ may not be effectively decidable in some computation model. But here we have the imposed condition that the eigenvalues at $`s=1`$ must be integer-valued. This additional condition might help making the equality condition effectively decidable at $`s=1`$. That is, we would only need to approximate $`E_0(1)`$ by the procedure above up to some accuracy, say 0.3, which sufficiently enables us to distinguish different integers, and we would not require infinite precision. The imposed condition of integer-valued eigenvalues for $`H_P`$, by construction, would provide us the built-in infinite precision at no extra cost!
## Hilbert’s tenth and the Schrödinger equation – The “dynamical” approach
The decision result for Hilbert’s tenth problem can also be encoded in yet another class of linear differential equations, apart from the class of nonlinear equations (15, 16) above. The linear equation is just the Schrödinger equation which captures the dynamics of a proposed quantum adiabatic algorithm .
Let $`|\psi (t)`$ be the quantum state at time $`t`$ (of some quantum system), its time evolution is given by the Schrödinger equation, for $`0<t<\tau `$,
(18) $`_t|\psi (t)`$ $`=`$ $`i(t/\tau )|\psi (t),`$
$`|\psi (0)`$ $`=`$ $`|\alpha _1\mathrm{}\alpha _K,`$
where we have chosen the initial state at time $`t=0`$ to be the non-degenerate ground state of $`H_I`$.
Once again we are only interested in the ground state of $`(1)=H_P`$. The quantum adiabatic theorem asserts that as $`\tau \mathrm{}`$ (that is, when the Hamiltonian $`(t/\tau )`$ in eq. (18) varies sufficiently slowly in the time $`t`$) the state of the above system at $`t=\tau `$ would be in the desired ground state, $`|\psi (\tau )|E_g`$, to any arbitrary degree of precision! For a given precision, various versions of the theorem dictate different conditions on $`\tau `$ in terms of some intrinsic properties of the eigenvalues and eigenfunctions of $`(s)`$. Those conditions thus are highly dependent on the individual Diophantine equation, and hence are not suitable to the spirit of a universal procedure required by Hilbert’s tenth problem.
We have proposed a different and universal criterion for the identification of the ground state $`|E_g`$ from $`|\psi (\tau )`$: the Fock state $`|n_1^{(0)},\mathrm{},n_K^{(0)}`$ is the ground state $`|E_g`$ if it has an occupation probability greater than one-half. That is,
(19) $`\left|\psi (\tau )|n_1^{(0)},\mathrm{},n_K^{(0)}\right|^2>1/2`$ $``$ $`|n_1^{(0)},\mathrm{},n_K^{(0)}=|E_g.`$
It should be emphasised here that such a criterion should be taken at the present time as a postulate as it has only been proved in some limited settings . With this criterion, we only need to repeatedly solve the Schrödinger (18) for larger and larger $`\tau `$ each time until the probability condition is satisfied so that the ground state can be identified. Such a time $`\tau `$ can be shown to exist and be finite by the quantum adiabatic theorem. More details of these can be found in .
Note also that while the criterion (19) may not the only one suitable for a physical implementation of the Schrödinger equation, it is the only one that we could yet find suitable for the mathematical discussion of this paper.
Unlike the case of nonlinear differential equations of a previous Section, here we could try to make use of a powerful computability result in a computation model which is known as the First Main Theorem by Pour-El & Richards . Essentially, the theorem asserts that a bounded linear operator from a Banach space to a Banach space which maps a computable sequence of spanning vectors into another computable sequence will also map any computable element into another computable element. For the case at hand, our Schrödinger equation defines a linear operator,
(20) $`U(\tau )`$ $`=`$ $`𝔗\mathrm{exp}\left\{i\tau {\displaystyle _0^1}(s)𝑑s\right\},`$
where $`𝔗`$ is the time-ordering symbol, which maps the initial state to the final state in the same separable Hilbert space. Now, our initial state $`|\alpha _1\mathrm{}\alpha _K`$ is computable by construction. On the other hand, the linear operator (20) coming out of the Schrödinger equation must be unitary and thus be bounded. Hence, the conditions of the theorem remained to be checked for fulfillment are: (i) which mathematical norm should be chosen to enable the identification criterion (19), or some other equivalent criterion, for the ground state at $`t=\tau `$; and (ii) whether the image of a particular computable basis is computable or not with this suitably chosen norm.
## Concluding remarks
Inspired by quantum mechanics, we have reformulated the question of solution existence of a Diophantine equation into the question of certain properties conceived in an infinitely coupled set of nonlinear differential equations. In words, we encode the answer of the former question into the smallest eigenvalue and corresponding eigenvector of a self-adjoint operator whose integer-valued eigenvalues are bounded from below. To find these eigen-properties we next deform the operator continuously to another self-adjoint operator whose spectrum is known. Once the deformation is also expressible in the form of a set of nonlinearly coupled differential equations, we could now start from the constructive knowns as a handle to study the desired unknowns.
In addition, we also explicitly present a class of linear Schrödinger equations whose solutions at some time $`\tau `$ from appropriate initial conditions contain the decision results for any given Diophantine equation.
These reformulations map a noncomputable problem in the domain of integer arithmetics into the wider framework of computable analysis. We have given the names, as explained in the Introduction, “dynamical” and “kinematic” respectively for the resulting linear and non-linear equations. In particular, our set of nonlinear equations would be an important topic for the largely untouched subject of nonlinear computable analysis.
For the various reasons given in the Introduction Section, the questions of computability for these differential equations are non-trivial and important. Towards some answers for these questions, we have advocated and speculated on an approach based on the First Main Theorem by Pour-El & Richards for the Schrödinger equations, and some other approximation procedure for the set of nonlinear differential equations. However, for now, these computability questions will have to be left as open problems and challenges.
If in the (admittedly unlikely) event that there exists computation model (with suitably chosen norm) in which the differential equations above are computable and that this model could be restricted and applied to functions from $``$ to $``$ then Hilbert’s tenth problem would seem to be solvable in integer arithmetics (as opposed to be solvable in, for example, some physical quantum adiabatic computation)! And this would entail a logical breakdown of the Church-Turing thesis. Because the restriction of such a computable analysis model to the domain of integers would provide a new notion of effective computability different from that of the class of Turing computation.
To be sure that this is not the case, further investigations are urgently needed to find out (i) whether the above equations are noncomputable in all computation models in computable analysis; or (ii) whether the model that admits such computability, if exists, cannot be restricted to integer arithmetics.
## Acknowledgements
I am indebted to Peter Hannaford and Alan Head for discussions and support. I would also like to thank Marian Pour-El for a discussion on the computability of quantum mechanics and for her gift of the out-of-print book . This work has been supported by the Swinburne University Strategic Initiatives. |
warning/0507/math0507573.html | ar5iv | text | # Densities in free groups and ℤ^𝑘, Visible Points and Test Elements
## 1. Introduction
The idea of genericity and generic-case behavior in finitely presented groups was introduced by Gromov and is currently the subject of much research. (See, for example .) Looking at the properties of random groups led Gromov to a probabilistic proof that there exists a finitely presented group that is not uniformly embeddable in a Hilbert space. It also turns out that random group-theoretic objects exhibit various kinds of algebraic rigidity properties. In particular, Kapovich, Schupp and Shpilrain proved that a random cyclically reduced element of a free group $`F=F(A)`$ is of minimal length in its $`Aut(F)`$-orbit and that such an element has a trivial stabilizer in $`Out(F)`$. Moreover, it turns out that random one-relator groups satisfy a strong Mostow-type rigidity. Specifically, two random one-relator groups $`G_r=a_1,\mathrm{},a_k|r`$ and $`G_s=a_1,\mathrm{},a_k|s`$ are isomorphic if and only if their Cayley graphs on the *given* set of generators $`\{a_1,\mathrm{},a_k\}`$ are isomorphic as labelled graphs where the graph isomorphism is only allowed to permute the label set $`\{a_1,\mathrm{},a_k\}^{\pm 1}`$.
The most straightforward definition of “genericity” is based on the notion of “asymptotic density”.
###### Definition 1.1 (Asymptotic density).
Suppose that $`T`$ is a countable set and that $`\mathrm{}:T`$ is a function (referred to as *length*) such that for every $`n`$ the set $`\{xT:\mathrm{}(x)n\}`$ is finite. If $`XT`$ and $`n0`$, we denote $`\rho _{\mathrm{}}(n,X):=\mathrm{\#}\{xX:\mathrm{}(x)n\}`$ and $`\gamma _{\mathrm{}}(n,S)=\mathrm{\#}\{xX:\mathrm{}(x)=n\}`$.
Let $`ST`$. The *asymptotic density* of $`S`$ in $`T`$ is
$$\overline{\rho }_{T,\mathrm{}}(S):=\underset{n\mathrm{}}{lim\; sup}\frac{\mathrm{\#}\{xS:\mathrm{}(x)n\}}{\mathrm{\#}\{xT:\mathrm{}(x)n\}}=\underset{n\mathrm{}}{lim\; sup}\frac{\rho _{\mathrm{}}(n,S)}{\rho _{\mathrm{}}(n,T)},$$
where we treat a fraction $`\frac{0}{0}`$, if it occurs, as $`0`$.
If the actual limit exists, we denote it by $`\rho _{T,\mathrm{}}(S)`$ and call this limit the *strict asymptotic density* of $`S`$ in $`T`$. We say that $`S`$ is *generic in $`T`$ with respect to $`\mathrm{}`$* if $`\rho _{T,\mathrm{}}(S)=1`$ and that $`S`$ is *negligible* in $`T`$ if $`\rho _{T,\mathrm{}}(S)=0`$.
If $`S`$ is $`T`$-generic then the probability that a uniformly chosen element of $`T`$ of length at most $`n`$ belongs to $`S`$ tends to $`1`$ as $`n`$ tends to infinity.
It turns out that a different density, recording the proportions of a set in two successive spheres, is sometimes more suitable for subsets of a free group.
###### Definition 1.2 (Annular Density).
Let $`T,S,\mathrm{}`$ be as in Definition 1.1. The *annular density* of $`S`$ in $`T`$ with respect to $`\mathrm{}`$ is:
$$\overline{\sigma }_{T,\mathrm{}}(S):=\underset{n\mathrm{}}{lim\; sup}\frac{1}{2}\left(\frac{\mathrm{\#}\{xS:\mathrm{}(x)=n1\}}{\mathrm{\#}\{xT:\mathrm{}(x)=n1\}}+\frac{\mathrm{\#}\{xS:\mathrm{}(x)=n\}}{\mathrm{\#}\{xT:\mathrm{}(x)=n\}}\right)=$$
$$\underset{n\mathrm{}}{lim\; sup}\frac{1}{2}\left(\frac{\gamma _{\mathrm{}}(n1,S)}{\gamma _{\mathrm{}}(n1,T)}+\frac{\gamma _{\mathrm{}}(n,S)}{\gamma _{\mathrm{}}(n,T)}\right),$$
where we treat a fraction $`\frac{0}{0}`$, if it occurs, as $`0`$. Again, if the actual limit exists, we denote this limit by $`\sigma _{T,\mathrm{}}(S)`$ and call it the *strict annular density* of $`S`$ in $`T`$ with respect to $`\mathrm{}`$.
###### Convention 1.3.
Throughout this paper $`F=F(A)`$ will be a free group of rank $`k2`$ with a fixed finite basis $`A=\{a_1,\mathrm{},a_k\}`$. If $`wF`$ then $`|w|_A`$ denotes the freely reduced length of $`w`$ with respect to the basis $`A`$. In discussing the density (asymptotic or annular) of subsets of $`F`$ using the notation above, we will assume that the ambient set is $`T=F`$ and that the length function $`\mathrm{}(w)`$ is $`|w|_A`$. If $`SF`$ we denote its asymptotic and annular densities by $`\overline{\rho }_A(S)`$ and $`\overline{\sigma }_A(S)`$ respectively, and if the strict asymptotic density or the strict annular density exist we denote them by $`\rho _A(S)`$ and $`\sigma _A(S)`$ respectively. Also, we denote $`\gamma _A(S):=\gamma _{\mathrm{}}(S)`$ and $`\rho _A(n,S):=\rho _{\mathrm{}}(n,S)`$ in this case.
For subsets of $`^k`$ a length function $`\mathrm{}:^k`$ will usually be the restriction to $`^k`$ of $`||.||_p`$-norm from $`^k`$ for some $`1p\mathrm{}`$. In this case for $`SZ^k`$ we denote the corresponding asymptotic density of $`S`$ in $`T=^k`$ by $`\overline{\rho }_p(S)`$ and if the strict asymptotic density exists, we denote it by $`\rho _p(S)`$.
It is not hard to see that if for a subset $`SF`$ the strict asymptotic density $`\rho _A(S)`$ exists then the strict annular density $`\sigma _A(S)`$ also exists and in fact $`\sigma _A(S)=\rho _A(S)`$. Namely, since the sizes of both the balls and the spheres in $`F(A)`$ grow as constant multiples of $`(2k1)^n`$, if the strict asymptotic density $`\rho _A(S)`$ exists, then the limit $`lim_n\mathrm{}\frac{\gamma _A(n,S)}{\gamma _A(n,F)}`$ exists and is equal to $`\rho _A(S)`$. Then the definition of $`\sigma _A(S)`$ implies that the strict annular density $`\sigma _A(S)`$ exists and is also equal to $`\rho _A(S)`$. However, as Example 1.5 below shows, it is possible that $`\sigma _A(S)`$ exists while $`\rho _A(S)`$ does not. Thus there are reasonable situations where the parity of the radius of a sphere or a ball affects the outcome when measuring the relative size of a subset of a free group, and annular density turns out to be a more suitable and relevant quantity. This is the case when we consider a subset of $`^k`$ and the full preimage of this subset in $`F`$ under the abelianization map. Moreover, annular density and its “close relatives” also make sense from the computational prospective. A typical experiment for generating a “random” element in a ball $`B(n)`$ of radius $`n`$ in $`F(A)`$ might proceed as follows. First choose a uniformly random integer $`m[0,n]`$ and then choose a uniformly random element $`x`$ from the $`m`$-sphere in $`F(A)`$ via a simple non-backtracking random walk of length $`m`$. It is easy to see that this experiment, while very natural, does not correspond to the uniform distribution on $`B(n)`$. For example, if $`F`$ has rank $`k=2`$, then for the uniform distribution on $`B(n)`$ the probability that the element $`x`$ has length $`n`$ is approximately $`\frac{2}{3}`$ for large $`n`$ while in our experiment described above this probability is $`1/(n+1)`$. In fact if $`wB(n)`$ then the probability of choosing the element $`w`$ in the above experiment is $`\frac{1}{(n+1)\mathrm{\#}S(m)}`$ where $`m`$ is the freely reduced length of $`w`$ and where $`S(m)`$ is the sphere of radius $`m`$ in $`F`$. Thus if $`XF`$ then the probability of choosing an element of $`X`$ in the above experiment is
$$\frac{1}{n+1}\underset{m=0}{\overset{n}{}}\frac{\mathrm{\#}(XS(m))}{\mathrm{\#}S(m)}.$$
If in our experiment we choose an element of $`S(n1)S(n)`$ by first randomly and uniformly choosing $`m\{n1,n\}`$ and then choosing a uniformly random element of $`S(m)`$, then for a subset $`X`$ of $`F`$ the probability of picking an element of $`X`$ is
$$\frac{1}{2}\left(\frac{\mathrm{\#}(XS(n1))}{\mathrm{\#}S(n1)}+\frac{\mathrm{\#}(XS(n))}{\mathrm{\#}S(n)}\right),$$
and the formulas from the definition of annular density appear.
For most of the cases where one can actually compute the asymptotic density of the set of elements in a free group having some natural algebraic property, this set turns out to be either generic or negligible. (Of course, a subset is negligible if and only if its complement is generic.) The following subsets are known to be negligible in a free group $`F=F(A)`$ of rank $`k2`$, both in the sense of asymptotic and annular densities: the set of all proper powers , a finite union of conjugacy classes, a subgroup of infinite index , a finite union of automorphic orbits (e.g. the set of all primitive elements) , the set of all elements whose cyclically reduced forms are not automorphically minimal , the union of all proper free factors of $`F`$ (this follows from results of and ). Examples of generic sets, again in the sense of both the asymptotic and the annular densities, include: the set of all words whose symmetrizations satisfy the $`C^{}(1/6)`$ small cancellation condition , the set of words with nontrivial images in the abelianization of $`F(A)`$ and the set of elements of $`F(A)`$ with cyclic stabilizers in $`Aut(F(A))`$ . It is therefore interesting to find examples of natural properties of elements of free groups which are “intermediate” in the sense that they have density different from either $`0`$ or $`1`$. In this article we show that being a test element in the free group of rank two is such an example.
###### Convention 1.4 (The abelianization map).
Recall that $`F`$ is a free group of rank $`k2`$ with free basis $`A=\{a_1,\mathrm{},a_k\}`$. We identify $`^k`$ with the abelianization of $`F`$ where the abelianization homomorphism $`\alpha :F^k`$ is given by $`a_ie_i`$, $`i=1,\mathrm{},k`$. We also denote $`\alpha (w)`$ by $`\overline{w}`$.
It is easy to construct an example of a subset $`H`$ of $`F`$ such that the annular density of $`H`$ in $`F`$ and the asymptotic density of $`\alpha (H)`$ in $`^k`$ are different. For instance, let $`F=F(a,b)`$ and consider the subgroup $`H=a,b[a,b]^3F`$. Then $`H`$ has infinite index in $`F`$ and hence has both asymptotic and annular density $`0`$ in $`F`$. On the other hand, $`\alpha (H)=\alpha (F)=^2`$ has asymptotic density $`1`$ in $`^k`$ with respect to any length function on $`^2`$.
###### Example 1.5.
Let $`F=F(a,b)`$, where $`A=\{a,b\}`$, be free of rank two. Let $`\alpha :F^2`$ be the abelianization map. Note that for any $`wF`$ the length $`|w|_A`$ and $`\alpha (w)_1`$ have the same parity, since $`\alpha (w)_1=|w_a|+|w_b|`$, where $`w_a,w_b`$ are the exponent sums on $`a`$ and $`b`$ in $`w`$. Let $`S=\{z^2:z_1\text{ is even }\}`$ and let $`\stackrel{~}{S}:=\alpha ^1(S)F`$. Then $`\stackrel{~}{S}=\{wF:|w|_A\text{ is even }\}`$. It is not hard to see that the strict asymptotic density of $`S`$ in $`^2`$, with respect to $`||.||_p`$ for any $`1p\mathrm{}`$, exists and is equal to $`1/2`$.
Since $`\stackrel{~}{S}`$ is exactly the union of all spheres of even radii in $`F`$, and the ratio of the sizes of spheres of radius $`n`$ and $`n1`$ is equal to $`3`$, it follows that the limits $`lim_n\mathrm{}\frac{\mathrm{\#}\{w\stackrel{~}{S}:|w|_A=n\}}{\mathrm{\#}\{wF:|w|=n\}}`$ and $`lim_n\mathrm{}\frac{\mathrm{\#}\{w\stackrel{~}{S}:|w|_An\}}{\mathrm{\#}\{wF:|w|n\}}`$ do not exist. However, it is easy to see that for every $`n1`$
$$\frac{\mathrm{\#}\{w\stackrel{~}{S}:|w|_A=n1\}}{\mathrm{\#}\{wF:|w|_A=n1\}}+\frac{\mathrm{\#}\{w\stackrel{~}{S}:|w|_A=n\}}{\mathrm{\#}\{wF:|w|_A=n\}}=1,$$
and therefore $`\sigma _A(\stackrel{~}{S})=\frac{1}{2}`$. Thus although the strict asymptotic density of $`\stackrel{~}{S}F`$ does not exist, the strict annular density does exist and is equal to the strict asymptotic density of $`S^2`$. More examples of a similar nature are discussed in Remark 1.8 of .
Example 1.5 demonstrates why the notion of annular density is suitable for working with subsets of free groups, while asymptotic density is more suitable for subsets of free abelian groups. Geometrically, this difference comes from the fact that free abelian groups are amenable with balls forming a Folner sequence, while free groups are non-amenable.
Although the counting occurs in very different places, it is interesting to ask how the asymptotic density of a subset $`S^k`$, with respect to some natural length function, and the annular density of its full preimage $`\alpha ^1(S)`$ in $`F`$ are related. We shall see that there is a reasonable assumption about the set $`S`$ which guarantees that the two densities are actually equal.
To do this we need to understand the image of the uniform distribution on the sphere of radius $`n`$ in $`F`$ under the abelianization map $`\alpha `$. There is an explicit formula for the size of the preimage of an element, and there is also a Central Limit Theorem saying that, when appropriately normalized, the distribution converges to a normal distribution. The methods of also give a *Local Limit Theorem* showing that, when working with width-two spherical shells in a free group, the densities of the image distributions in $`^k`$ converge to a normal density. Such a result was later also shown (by rather different methods, and in greater generality) by Richard Sharp in . Recently Petridis and Risager obtained a similar Local Limit Theorem for counting conjugacy classes rather than elements of $`F`$. On the face of it, studying the annular density of the set $`\alpha ^1(S)`$ in $`F`$ presents new challenges. The central limit theorem by itself seems too crude a tool and *a priori* it would appear that one would need very sharp error bounds in the local limit theorem. Nevertheless, we produce a short argument solving this problem where one of the key ingredients is the ergodicity of the $`SL(k,)`$-action on $`^k`$.
We can now state our main result:
###### Theorem A.
Let $`F=F(A)`$ be a free group of rank $`k2`$ with free basis $`A=\{a_1,\mathrm{},a_k\}`$ and let $`\alpha :F^k`$ be the abelianization homomorphism.
Let $`S^k`$ be an $`SL(k,)`$-invariant subset and put $`\stackrel{~}{S}=\alpha ^1(S)F`$.
Then
1. For every $`1p\mathrm{}`$ the strict asymptotic density $`\rho _p(S)`$ exists and, moreover, for every $`1p\mathrm{}`$ we have $`\rho _p(S)=\rho _{\mathrm{}}(S)`$.
2. The strict annular density $`\sigma _A(\stackrel{~}{S})`$ exist and, moreover, $`\sigma _A(\stackrel{~}{S})=\rho _{\mathrm{}}(S)`$.
That is,
$$\underset{n\mathrm{}}{lim}\frac{1}{2}\left(\frac{\gamma _A(n1,\{wF:\alpha (w)S\})}{\gamma _A(n1,F)}+\frac{\gamma _A(n,\{wF:\alpha (w)S\}}{\gamma _A(n,F)}\right)=$$
$$\underset{n\mathrm{}}{lim}\frac{\mathrm{\#}\{z:z^k,z_{\mathrm{}}n,\text{ and }zS\}}{\mathrm{\#}\{z:z^k,z_{\mathrm{}}n\}}.$$
The requirement that $`S`$ be $`SL(k,)`$-invariant essentially says that the subset $`S`$ of $`^k`$ is defined in “abstract” group-theoretic terms, not involving the specific choice of a free basis for $`^k`$. Note that Proposition 2.2 below gives an explicit formula for $`\rho _{\mathrm{}}(S)`$ in Theorem A.
Our main application of Theorem A concerns the case where $`S`$ is the set of all “visible” points in $`^k`$. A nonzero point $`z`$ of $`^k`$ is called *visible* if the greatest common divisor of the coordinates of $`z`$ is equal to $`1`$. This terminology is standard in number theory and reflects the fact that if $`z`$ is visible then the line segment between the origin and $`z`$ does not contain any other integer lattice points. For a nonzero point $`z^k`$ being visible is also equivalent to $`z`$ not being a proper power in $`^k`$, that is, to $`z`$ generating a maximal cyclic subgroup of $`^k`$. More generally, if $`t1`$ is an integer, we will say that $`z^k`$ is *$`t`$-visible* if $`z=z_1^t`$ for some visible $`z_1^k`$, that is, if the greatest common divisor of the coordinates of $`z`$ is equal to $`t`$.
We want to “lift” this terminology to free groups.
###### Definition 1.6 (Visible elements in free groups).
Let $`F=F(A)`$ be a free group of rank $`k2`$ with free basis $`A=\{a_1,\mathrm{},a_k\}`$ and let $`\alpha :F^k`$ be the abelianization homomorphism, that is, $`\alpha (a_i)=e_i^k`$. We say that an element $`wF`$ is *visible* if $`\alpha (w)`$ is visible in $`^k`$. Let $`V`$ be the set of visible elements of $`F`$. Similarly, for an integer $`t1`$ an element $`wF`$ is *$`t`$-visible* if $`\alpha (w)`$ is $`t`$-visible in $`^k`$. We use $`V_t`$ to denote the set of all $`t`$-visible elements of $`F`$ and we use $`U_t`$ to denote the set of all $`t`$-visible elements of $`^k`$.
Note that $`V=V_1`$ and that for every $`t1`$ the definition of $`V_t`$ does not depend on the choice of the free basis $`A`$ of $`F`$.
The following proposition giving the asymptotic density of the set of $`t`$-visible points in $`^k`$ in terms of the Riemann zeta-function is well-known in number theory .
###### Proposition 1.7.
For any integer $`t1`$ we have
$$\rho _{\mathrm{}}(U_t)=\frac{1}{t^k\zeta (k)}.$$
The case $`k=2`$ and $`t=1`$ of Proposition 1.7 was proved by Mertens in 1874 . (See also Theorem 332 of the classic book of Hardy and Wright .) Recall that $`\zeta (k)=_{n=1}^{\mathrm{}}\frac{1}{n^k}`$ and, in particular, $`\zeta (2)=\frac{\pi ^2}{6}`$.
It is therefore natural to investigate the asymptotic density of the set of visible elements in $`F`$. As a direct corollary of Theorem A, of Proposition 2.9 below and of Proposition 1.7 we obtain:
###### Theorem B.
Let $`F=F(A)`$ be a free group of rank $`k2`$ with free basis $`A=\{a_1,a_2,\mathrm{},a_k\}`$. Let $`t1`$ be an integer. Then the strict annular density $`\sigma _A(V_t)`$ exists and
$$\sigma _A(V_t)=\frac{1}{t^k\zeta (k)}.$$
Moreover, in this case
$$0<\frac{4k4}{(2k1)^2t^k\zeta (k)}\underset{n\mathrm{}}{lim\; inf}\frac{\rho _A(n,V_t)}{\rho _A(n,F)}$$
$$\underset{n\mathrm{}}{lim\; sup}\frac{\rho _A(n,V_t)}{\rho _A(n,F)}1\frac{4k4}{(2k1)^2}\left(1\frac{1}{t^k\zeta (k)}\right)<1.$$
A result similar to Theorem B for counting conjugacy classes in $`F_k`$ with primitive images in $`^k`$ has been recently independently obtained by Petridis and Risager .
For the case of the free group of rank two we compute the two “spherical densities” for the density of the set $`V_1`$, corresponding to even and odd $`n`$ tending to infinity:
###### Theorem C.
Let $`k=2`$. We have
$$\underset{m\mathrm{}}{lim}\frac{\gamma _A(2m,V_1)}{\gamma _A(2m,F)}=\frac{2}{3\zeta (2)}=\frac{4}{\pi ^2}$$
and
$$\underset{m\mathrm{}}{lim}\frac{\gamma _A(2m1,V_1)}{\gamma _A(2m1,F)}=\frac{8}{\pi ^2}.$$
Theorem B implies that $`\sigma _A(V_1)=\frac{6}{\pi ^2}`$ in this case. Since the two limits in Theorem C are different, the statements of Theorem A and Theorem B cannot be substantially improved. This fact underscores the conclusion that annular density is the right kind of notion for measuring the sizes of subsets of free groups, where the abelianization map is concerned.
We apply Theorem B to compute the annular density of test elements in a free group of rank two. Recall that an element $`gG`$ is called a *test element* if every endomorphism of $`G`$ fixing $`g`$ is actually an automorphism of $`G`$. It is easy to see that for two conjugate elements $`g_1,g_2G`$ the element $`g_1`$ is a test element if and only if $`g_2`$ is also a test element and thus the property of being a test element depends only on the conjugacy class of an element $`g`$. The notion of a test element was introduced by Shpilrain and has since become a subject of active research both in group theory and in the context of other algebraic structures such as polynomial algebras and Lie algebras. (See, for example, .) It turns out that studying test elements in a particular group $`G`$ produces interesting information about the automorphism group of $`G`$.
Here we prove:
###### Theorem D.
Let $`F=F(a,b)`$ be a free group of rank two with free basis $`A=\{a,b\}`$. Then for the set $`𝒯`$ of all test elements in $`F`$ the strict annular density exists and
$$\sigma _A(𝒯)=1\frac{6}{\pi ^2}.$$
Moreover,
$$0<\frac{4}{9}(1\frac{6}{\pi ^2})\underset{n\mathrm{}}{lim\; inf}\frac{\rho _A(n,𝒯)}{\rho _A(n,F)}$$
$$\underset{n\mathrm{}}{lim\; sup}\frac{\rho _A(n,𝒯)}{\rho _A(n,F)}1\frac{8}{3\pi ^2}<1.$$
By a result of Turner , an element of $`F`$ is a test elements if and only if it does not belong to a proper retract of $`F`$. Therefore Theorem D implies that the strict annular density of the union of all proper retracts of $`F(a,b)`$ is $`\frac{6}{\pi ^2}`$. In Theorem D above we have $`\frac{4}{9}(1\frac{6}{\pi ^2})0.1742`$, $`1\frac{6}{\pi ^2}.3920`$ and $`1\frac{8}{3\pi ^2}0.7298`$. Thus Theorem D shows that being a test element is an “intermediate” property in the free group of rank two. More generally, Theorem B implies that, for every $`k2`$ and for $`N`$ sufficiently large, the set $`V_N`$ of $`N`$-visible elements in $`F_k`$ has strictly positive annular density arbitrarily close to $`0`$ while the set $`S_N=_{t=1}^NV_t`$ has annular density less than but arbitrarily close to $`1`$. It is well-known that every positive rational number has an “Egyptian fraction” representation as a finite sum of distinct terms of the form $`\frac{1}{n}`$. It does not seem clear what values can be obtained as finite sums of distinct terms of the form $`\frac{1}{n^2}`$ and there are excluded intervals. In particular, if such a sum uses $`1`$, it is at least $`1`$ while if $`1`$ is not used then the sum is at most $`\frac{\pi ^2}{6}1`$. Multiplying by the scale factor $`\frac{1}{\zeta (2)}`$, we see that we cannot obtain a annular density in the open interval $`(1\frac{6}{\pi ^2},\frac{6}{\pi ^2})`$ by taking a finite union of the sets $`V_t`$ (Proposition 2.2 below shows that the same is true for infinite unions). It is interesting to note that the probabilities of being a test element or of not being a test element are the boundary points of this excluded interval.
Nathan Dunfield and Dylan Thurston recently proved that for a two-generator one-relator group being free-by-cyclic is an intermediate property. While they do not provide an exact value for the asymptotic density (nor do they prove that either the strict asymptotic density or the strict annular density exist), they show that it is strictly between $`0`$ and $`1`$. Computer experiments by Kapovich and Schupp, by Mark Sapir and by Dunfield and Thurston indicate that in the two-generator case this asymptotic density is greater than $`0.9`$.
The authors are grateful to Laurent Bartholdi, John D’Angelo,Iwan Duursma, Kevin Ford, Steve Lalley, Alexander Ol’shanskii, Yuval Peres, Yannis Petridis, Alexandru Zaharescu and Andrzej Zuk for very helpful conversations.
## 2. Comparing densities in $`^k`$ and in $`F_k`$
###### Convention 2.1.
Throughout this section let $`S^k`$ be as in Theorem A and let $`\delta :=\overline{\rho }_{\mathrm{}}(S)`$.
We can now prove that for every $`SL(k,)`$-invariant subset $`S`$ of $`^k`$ the strict asymptotic density $`\rho _{\mathrm{}}(S)`$ exists. Recall that Proposition 1.7 stated in the introduction gives the precise value of the strict asymptotic density of the set of all $`t`$-visible elements in $`^k`$. The crucial points of the proof below are that the complement of an $`SL(k,)`$-invariant is also $`SL(k,)`$-invariant and that $`_{t=1}^{\mathrm{}}\frac{1}{t^k\zeta (k)}=1`$.
###### Proposition 2.2.
Let $`YZ^k`$ be a nonempty $`SL(k,)`$-invariant subset that does not contain $`\mathrm{𝟎}^k`$. Let $`I`$ be the set of all integers $`t1`$ such that there exists a $`t`$-visible element in $`Y`$. Then
1. $`Y=_{tI}U_t`$.
2. The strict asymptotic density $`\rho _{\mathrm{}}(Y)`$ exists and
$$\rho _{\mathrm{}}(Y)=\underset{tI}{}\rho (U_t)=\underset{tI}{}\frac{1}{t^k\zeta (k)}.$$
###### Proof.
Observe first that
$$\underset{t=1}{\overset{\mathrm{}}{}}\frac{1}{t^k\zeta (k)}=\frac{1}{\zeta (k)}\underset{t=1}{\overset{\mathrm{}}{}}\frac{1}{t^k}=\frac{\zeta (k)}{\zeta (k)}=1.$$
Since $`k2`$, two nonzero elements $`z,z^{}^k`$ lie in the same $`SL(k,)`$-orbit if and only if the greatest common divisors of the coordinates of $`z`$ and of $`z^{}`$ are equal. Thus every $`SL(k,)`$-orbit of a nonzero element of $`^k`$ has the form $`U_t`$ for some $`t1`$. This implies part (1) of Proposition 2.2.
Let $`I^{}:=\{t:t1,tI\}`$. If either $`I`$ or $`I^{}`$ is finite, part (2) of Proposition 2.2 follows directly from proposition 1.7. Suppose now that both $`I`$ and $`I^{}`$ are infinite and let $`Y^{}:=^k(Y\{\mathrm{𝟎}\})=_{tI^{}}U_t`$.
For every finite subset $`JI`$ let $`Y_J:=_{tJ}U_t`$. Since $`Y_JY`$, it follows that
$$\underset{n\mathrm{}}{lim\; inf}\frac{\mathrm{\#}\{zY:z_{\mathrm{}}n\}}{\mathrm{\#}\{z^k:z_{\mathrm{}}n\}}\rho _{\mathrm{}}(Y_J)=\underset{tJ}{}\frac{1}{t^k\zeta (k)}.$$
Since this is true for every finite subset of $`I`$, we conclude that
$$\underset{n\mathrm{}}{lim\; inf}\frac{\mathrm{\#}\{zY:z_{\mathrm{}}n\}}{\mathrm{\#}\{z^k:z_{\mathrm{}}n\}}\underset{tI}{}\frac{1}{t^k\zeta (k)}.$$
The same argument applies to the $`SL(k,)`$-invariant set $`Y^{}`$ and therefore:
$$\underset{n\mathrm{}}{lim\; inf}\frac{\mathrm{\#}\{zY^{}:z_{\mathrm{}}n\}}{\mathrm{\#}\{z^k:z_{\mathrm{}}n\}}\underset{tI^{}}{}\frac{1}{t^k\zeta (k)}.$$
This implies
$$1\underset{n\mathrm{}}{lim\; inf}\frac{\mathrm{\#}\{zY^{}:z_{\mathrm{}}n\}}{\mathrm{\#}\{z^k:z_{\mathrm{}}n\}}1\underset{tI^{}}{}\frac{1}{t^k\zeta (k)}$$
$$\underset{n\mathrm{}}{lim\; sup}\left(1\frac{\mathrm{\#}\{zY^{}:z_{\mathrm{}}n\}}{\mathrm{\#}\{z^k:z_{\mathrm{}}n\}}\right)1\underset{tI^{}}{}\frac{1}{t^k\zeta (k)}$$
$$\underset{n\mathrm{}}{lim\; sup}\frac{\mathrm{\#}\{zY:z_{\mathrm{}}n\}}{\mathrm{\#}\{z^k:z_{\mathrm{}}n\}}1\underset{tI^{}}{}\frac{1}{t^k\zeta (k)}=\underset{tI}{}\frac{1}{t^k\zeta (k)}.$$
Hence
$$\underset{n\mathrm{}}{lim}\frac{\mathrm{\#}\{zY:z_{\mathrm{}}n\}}{\mathrm{\#}\{z^k:z_{\mathrm{}}n\}}=\underset{tI}{}\frac{1}{t^k\zeta (k)},$$
as required. ∎
Recall that $`S^k`$ is an $`SL(k,)`$-invariant subset and that $`\delta =\overline{\rho }_{\mathrm{}}(S)`$. Proposition 2.2 implies that in fact $`\delta =\rho _{\mathrm{}}(S)`$.
It is well known that if $`\mathrm{\Omega }^k`$ is a “nice” bounded open set then the Lebesgue measure $`\lambda (\mathrm{\Omega })`$ can be computed as
$$\lambda (\mathrm{\Omega })=\underset{r\mathrm{}}{lim}\frac{\mathrm{\#}(^kr\mathrm{\Omega })}{r^k}.$$
Here we say that a bounded open subset of $`^k`$ is “nice” if its boundary is piecewise smooth.
We need a similar formula for counting the points of $`S`$. For a real number $`r1`$ and a nice bounded open set $`\mathrm{\Omega }^k`$ let
$$\mu _{r,S}(\mathrm{\Omega }):=\frac{\mathrm{\#}(Sr\mathrm{\Omega })}{r^k}.$$
###### Proposition 2.3.
For any nice bounded open set $`\mathrm{\Omega }^k`$ we have
$$\underset{r\mathrm{}}{lim}\mu _{r,S}(\mathrm{\Omega })=\delta \lambda (\mathrm{\Omega }).$$
###### Proof.
Each $`\mu _{r,S}`$ can be regarded as a measure on $`^k`$. We prove the theorem by showing that the $`\mu _{r,S}`$ weakly converge to $`\delta \lambda `$ as $`r\mathrm{}`$, where $`\lambda `$ is the Lebesgue measure.
By Helly’s theorem there exists a sequence $`(r_i)_{i=1}^{\mathrm{}}`$ with $`\underset{i\mathrm{}}{lim}r_i=\mathrm{}`$ such that the sequence $`\mu _{r_1,S},\mu _{r_2,S},\mathrm{}`$ is weakly convergent to some limiting measure. We now show that for every such convergent subsequence of $`\mu _{r_i,S}`$ the limiting measure is indeed equal to $`\delta \lambda `$, where $`\lambda `$ is the Lebesgue measure.
Indeed, suppose that $`\sigma =(r_i)_{i=1}^{\mathrm{}}`$ is a sequence with $`\underset{i\mathrm{}}{lim}r_i=\mathrm{}`$ such that the sequence $`\mu _{r_i,S}`$ converges to the limiting measure $`\mu _\sigma =\underset{i\mathrm{}}{lim}\mu _{r_i,S}`$. Every $`\mu _{r_i,S}`$ is invariant with respect to the natural $`SL(k,)`$-action since this action preserves the set $`S`$ and also commutes with homotheties of $`^k`$ centered at the origin. Therefore the limiting measure $`\mu _\sigma `$ is also $`SL(k,)`$-invariant.
Moreover, the measures $`\mu _{r,S}`$ are dominated by the measures $`\lambda _r`$ defined as $`\lambda _r(\mathrm{\Omega })=\frac{\mathrm{\#}(^kr\mathrm{\Omega })}{r^k}`$. Since, as observed earlier, the measures $`\lambda _r`$ converge to the Lebesgue measure $`\lambda `$, it follows that $`\mu _\sigma `$ is absolutely continuous with respect to $`\lambda `$. It is known that the natural action of $`SL(k,)`$ on $`^k`$ is ergodic with respect to $`\lambda `$. (See, for example, Zimmer’s classic monograph .) Therefore $`\mu _\sigma `$ is a constant multiple $`c\lambda `$ of $`\lambda `$. The constant $`c`$ can be computed explicitly for a set such as the open unit ball $`B`$ in the $`||.||_{\mathrm{}}`$ norm on $`^k`$ defining the length function $`\mathrm{}`$ on $`^k`$.
By assumption we know that
$$\underset{r\mathrm{}}{lim}\frac{\mathrm{\#}\{z^k:zSrB\}}{\mathrm{\#}\{z^k:zrB\}}=\delta .$$
We also have
$$\underset{r\mathrm{}}{lim}\frac{\mathrm{\#}\{z^k:zrB\}}{r^k}=\lambda (B)$$
and hence
$$\underset{r\mathrm{}}{lim}\frac{\mathrm{\#}\{z^k:zSrB\}}{r^k}=\delta \lambda (B).$$
Therefore $`c=\delta `$ and $`\mu _\sigma =\delta \lambda `$. The above argument in fact shows that every convergent subsequence, with $`r\mathrm{}`$, of $`\mu _{r,S}`$ converges to $`\delta \lambda `$ and therefore $`\underset{r\mathrm{}}{lim}\mu _{r,S}=\delta \lambda `$.
###### Remark 2.4.
Let $`1p\mathrm{}`$. Then the open unit ball in $`^k`$ with respect to $`||.||_p`$ is ”nice”. Proposition 2.3, applied to $`\mathrm{\Omega }`$ being this ball, implies that $`\rho _p(S)=\rho _{\mathrm{}}(S)=\delta `$.
###### Convention 2.5.
As always, $`F=F(a_1,\mathrm{},a_k)`$ is the free group of rank $`k2`$ with free basis $`A=\{a_1,\mathrm{},a_k\}`$ and $`\alpha :F^k`$ is the abelianization homomorphism sending $`a_i`$ to $`e_i`$ in $`^k`$. We will denote $`\alpha (w)`$ by $`\overline{w}`$. For $`n1`$, $`B_F(n)`$ denotes the set of all $`wF`$ with $`|w|_An`$. Also, for a point $`x=(x_1,\mathrm{},x_k)^k`$ we denote by $`x`$ the $`||.||_2`$-norm of $`x`$, that is $`x=\sqrt{_{i=1}^kx_i^2}`$.
###### Notation 2.6.
For an integer $`n1`$ and a point $`x^k`$ let
$$p_n(x)=\frac{\gamma _A(n1,\{fF:\alpha (f)=x\sqrt{n}\})}{2\gamma _A(n1,F)}+\frac{\gamma _A(n,\{fF:\alpha (f)=x\sqrt{n}\})}{2\gamma _A(n,F)}.$$
Thus $`p_n`$ is a distribution supported on finitely many points of $`\frac{1}{\sqrt{n}}^k`$.
We need the following facts about the sequence of distributions $`p_n`$. Of these the most significant is part (2) which is a local limit theorem in our context. It was obtained by Rivin and, independently and via different methods, by Sharp (specifically, we use Theorem 1 of for part (2) of Proposition 2.7 below).
###### Proposition 2.7.
Let $`k2`$ and let $`p_n`$ be as above. Then:
1. The sequence of distributions $`p_n`$ converges weakly to a normal distribution $`𝔑`$, with density $`𝔫`$.
2. We have
$$\underset{x^k/\sqrt{n}}{sup}|p_n(x)n^{k/2}𝔫(x)|0\text{ as }n\mathrm{}.$$
3. We have
$$\underset{c\mathrm{}}{lim}\{p_n(x):x^k/\sqrt{n}\text{ and }xc\}=0.$$
###### Theorem 2.8.
Let $`\mathrm{\Omega }^k`$ be a nice bounded open set. Then
$$\underset{n\mathrm{}}{lim}\underset{xS\sqrt{n}\mathrm{\Omega }}{}p_n(x/\sqrt{n})=\delta 𝔑(\mathrm{\Omega }).$$
###### Proof.
We have
$$\underset{x_t^k\sqrt{n}\mathrm{\Omega }}{}p_n(x/\sqrt{n})=\underset{y\frac{1}{\sqrt{n}}S\mathrm{\Omega }}{}p_n(y)=$$
$$n^{k/2}\underset{y\frac{1}{\sqrt{n}}S\mathrm{\Omega }}{}𝔫(y)+n^{k/2}\underset{y\frac{1}{\sqrt{n}}S\mathrm{\Omega }}{}(n^{k/2}p_n(y)𝔫(y)).$$
The local limit theorem in part (2) of Proposition 2.7 tells us that, as $`n\mathrm{}`$, each summand $`n^{k/2}p_n(y)𝔫(y)`$ of the second sum in the last line of equation above converges to zero and hence so does their Cesaro mean. Proposition 2.3 implies that, as $`n\mathrm{}`$, the first summand $`n^{k/2}{\displaystyle \underset{y\frac{1}{\sqrt{n}}S\mathrm{\Omega }}{}}𝔫(y)`$ converges to
$$\delta _\mathrm{\Omega }𝔫𝑑\lambda =\delta 𝔑(\mathrm{\Omega }).$$
We can now compute the strict asymptotic density of $`\stackrel{~}{S}=\alpha ^1(S)`$ in $`F`$ and obtain Theorem A.
###### Proof of Theorem A.
Recall that $`S^k`$ is an $`SL(k,)`$-invariant set and that $`\delta =\overline{\rho }_{\mathrm{}}(S)`$. Proposition 2.2 implies that in fact $`\rho _{\mathrm{}}(S)`$ exists and $`\delta =\rho _{\mathrm{}}(S)`$. Moreover, as we have seen in Remark 2.4, for every $`1p\mathrm{}`$ the strict asymptotic density $`\rho _p(S)`$ exists and $`\rho _p(S)=\delta =\rho _{\mathrm{}}(S)`$. This proves part (1) of Theorem A.
To prove part (2) of Theorem A we need to establish that the strict annular density $`\sigma _A(S)`$ exists and that $`\sigma _A(S)=\delta `$.
For $`c>0`$ denote $`\mathrm{\Omega }_c:=\{x^k:x<c\}`$. Then $`\underset{c\mathrm{}}{lim}𝔑(\mathrm{\Omega }_c)=1`$. Let $`ϵ>0`$ be arbitrary. Choose $`c>0`$ such that
$$|𝔑(\mathrm{\Omega }_c)1|ϵ/3$$
and such that
$$\underset{n\mathrm{}}{lim}\{p_n(x):x^k/\sqrt{n}\text{ and }xc\}ϵ/6.$$
By Theorem 2.8 and the above formula there is some $`n_01`$ such that for all $`nn_0`$ we have
$$\left|\underset{xS\sqrt{n}\mathrm{\Omega }_c}{}p_n(x/\sqrt{n})\delta 𝔑(\mathrm{\Omega }_c)\right|ϵ/3$$
and
$$\{p_n(x):x^k/\sqrt{n}\text{ and }xc\}ϵ/3.$$
Let
$$Q(n):=\frac{\gamma _A(n1,\{wF:\overline{w}S\})}{2\gamma _A(n1,F)}+\frac{\gamma _A(n,\{wF:\overline{w}S\})}{2\gamma _A(n,F)}.$$
For $`nn_0`$ we have
$$Q(n)=$$
$$\frac{\mathrm{\#}\{wF:\overline{w}S,|w|_A=n1\text{ and }\overline{w}<c\sqrt{n}\}}{2\gamma _A(n1,F)}+$$
$$\frac{\mathrm{\#}\{wF:\overline{w}S,|w|_A=n\text{ and }\overline{w}<c\sqrt{n}\}}{2\gamma _A(n,F)}+$$
$$\frac{\mathrm{\#}\{wF:\overline{w}S,|w|_A=n1\text{ and }\overline{w}c\sqrt{n}\}}{2\gamma _A(n1,F)}+$$
$$\frac{\mathrm{\#}\{wF:\overline{w}S,|w|_A=n\text{ and }\overline{w}c\sqrt{n}\}}{2\gamma _A(n,F)}=$$
$$\underset{xS\sqrt{n}\mathrm{\Omega }_c}{}p_n(x/\sqrt{n})+\underset{xS(^k\sqrt{n}\mathrm{\Omega }_c)}{}p_n(x/\sqrt{n})$$
In the last line of the above equation, the first sum differs from $`\delta 𝔑(\mathrm{\Omega }_c)`$ by at most $`ϵ/3`$ since $`nn_0`$ and the second sum is $`ϵ/3`$ by the choice of $`c`$ and $`n_0`$. Therefore, again by the choice of $`c`$, we have $`|Q(n)\delta |ϵ`$. Since $`ϵ>0`$ was arbitrary, this implies that $`\underset{n\mathrm{}}{lim}Q(n)=\delta `$, as claimed. ∎
The following observation shows how to estimate the asymptotic density in terms of the annular density.
###### Proposition 2.9.
Let $`YF`$ be a subset such that the strict annular density $`\delta =\sigma _A(Y)`$ exists. Then
$$\frac{4k4}{(2k1)^2}\delta \underset{n\mathrm{}}{lim\; inf}\frac{\rho _A(n,S)}{\rho _A(n,F)}\underset{n\mathrm{}}{lim\; sup}\frac{\rho _A(n,S)}{\rho _A(n,F)}1\frac{4k4}{(2k1)^2}(1\delta ).$$
In particular, if $`0<\delta <1`$ then
$$0<\underset{n\mathrm{}}{lim\; inf}\frac{\rho _A(n,S)}{\rho _A(n,F)}\underset{n\mathrm{}}{lim\; sup}\frac{\rho _A(n,S)}{\rho _A(n,F)}<1.$$
###### Proof.
Note that for $`n1`$ we have $`\gamma _A(n,F)=2k(2k1)^n`$ and that, up to an additive constant, $`\rho _A(n,F)=\frac{k}{k1}(2k1)^n`$. Denote $`a_n=\gamma _A(n,Y)`$. We have
$$\delta =\underset{n\mathrm{}}{lim}\frac{1}{2}\left(\frac{a_{n1}}{2k(2k1)^{n2}}+\frac{a_n}{2k(2k1)^{n1}}\right)=\frac{1}{2}\underset{n\mathrm{}}{lim}\frac{a_{n1}\frac{2k1}{2k2}+a_n\frac{1}{2k2}}{\frac{k}{k1}(2k1)^{n1}}$$
Therefore
$$\underset{n\mathrm{}}{lim\; inf}\frac{\rho _A(n,Y)}{\rho _A(n,F)}=\underset{n\mathrm{}}{lim\; inf}\frac{a_1+\mathrm{}+a_n}{\frac{k}{k1}(2k1)^n}\underset{n\mathrm{}}{lim\; inf}\frac{a_{n1}+a_n}{\frac{k}{k1}(2k1)^n}=$$
$$\frac{2k2}{2k1}\underset{n\mathrm{}}{lim\; inf}\frac{a_{n1}\frac{2k1}{2k2}+a_n\frac{2k1}{2k2}}{\frac{k}{k1}(2k1)^n}\frac{2k2}{2k1}\underset{n\mathrm{}}{lim\; inf}\frac{a_{n1}\frac{2k1}{2k2}+a_n\frac{1}{2k2}}{\frac{k}{k1}(2k1)^n}=$$
$$\frac{4k4}{(2k1)^2}\underset{n\mathrm{}}{lim\; inf}\frac{1}{2}\frac{a_{n1}\frac{2k1}{2k2}+a_n\frac{1}{2k2}}{\frac{k}{k1}(2k1)^{n1}}=\frac{4k4}{(2k1)^2}\delta .$$
Applying the same argument to the set $`FY`$, we get
$$\underset{n\mathrm{}}{lim\; inf}\frac{\rho _A(n,FY)}{\rho _A(n,F)}\frac{4k4}{(2k1)^2}(1\delta ).$$
Therefore
$$\underset{n\mathrm{}}{lim\; sup}\frac{\rho _A(n,Y)}{\rho _A(n,F)}=1\underset{n\mathrm{}}{lim\; inf}\frac{\rho _A(n,FY)}{\rho _A(n,F)}1\frac{4k4}{(2k1)^2}(1\delta ).$$
## 3. Spherical densities
In this section we will prove Theorem C from the Introduction and, for the case of $`k=2`$, compute the “spherical densities”
$$\underset{m\mathrm{}}{lim}\frac{\gamma _A(2m,V_1)}{\gamma _A(2m,F)}$$
and
$$\underset{m\mathrm{}}{lim}\frac{\gamma _A(2m1,V_1)}{\gamma _A(2m1,F)}.$$
This is done by computing the “spherical densities” for the set $`V_1(e)F`$ consisting of all points of $`V_1`$ of even length and comparing it with the strict asymptotic density of the set $`U_1(e)^k`$ of all elements of $`^k`$ of even $`||.||_{\mathrm{}}`$-length. The key point is that for $`n=2m`$ we have $`\gamma _A(2m1,V_1(e))=0`$ and $`\gamma _A(2m,V_1)=\gamma _A(2m,V_1(e))`$. Therefore for the quantities from the definition of annular density we have
$$\frac{1}{2}\left(\frac{\gamma _A(2m,V_1(e))}{\gamma _A(2m,F)}+\frac{\gamma _A(2m1,V_1(e))}{\gamma _A(2m1,F)}\right)=\frac{\gamma _A(2m,V_1(e))}{2\gamma _A(2m,F)}=\frac{\gamma _A(2m,V_1)}{2\gamma _A(2m,F)}.$$
This allows us to essentially repeat the proof of Theorem A, applied to the sets $`U_1(e)`$ and $`V_1(e)=\alpha ^1(U_1(e))`$, except that instead of ergodicity of the action of $`SL(k,)`$ we use the ergodicity of the action on $`^k`$ of a congruence subgroup of $`SL(k,)`$ that leaves $`U_1(e)`$ invariant.
###### Convention 3.1.
We say that an element $`z=(z_1,\mathrm{},z_k)^k`$ is *even* if $`z_1=|z_1|+\mathrm{}+|z_k|`$ is even and that $`z`$ is *odd* if $`z_1`$ is odd. Similarly, $`wF`$ is *even* if $`|w|_A`$ is even and $`wF`$ is *odd* if $`|w|_A`$ is odd. Note that $`wF`$ is even if and only if $`\alpha (w)^k`$ is even.
Let $`G_k`$ be the set of all $`MSL(k,)`$ such that $`M=I_k`$ in $`SL(k,/2)`$. Thus $`G_k`$ is a finite index subgroup of $`SL(k,)`$ also known as the $`2`$-congruence subgroup. Denote by $`^k(e)`$ the set of all even elements in $`^k`$. Also denote $`U_1(e):=U_1^k(e)`$. Observe that $`^k(e)`$ and $`U_1(e)`$ are $`G_k`$-invariant and that $`V_1(e)=\alpha ^1(U_1(e))`$. (The actual set-wise stabilizer of $`^k(e)`$ in $`SL(k,)`$ contains $`G_k`$ as a subgroup of finite index.)
###### Proposition 3.2.
Let $`S^k`$ be a subset such that $`\delta =\rho _{\mathrm{}}(S)`$ exists and such that $`S`$ is $`G_k`$-invariant. Then for every bounded nice open set $`\mathrm{\Omega }^k`$ we have
$$\underset{r\mathrm{}}{lim}\mu _{r,S}(\mathrm{\Omega })=\delta \lambda (\mathrm{\Omega }).$$
###### Proof.
The proof is the same as for Proposition 2.3. The only difference is that instead of ergodicity of the $`SL(k,)`$-action on $`^k`$ we use ergodicity of the $`G_k`$-action on $`^k`$ with respect to the Lebesgue measure (see for the proof of this ergodicity). ∎
Let $`p_n(x)`$ be defined exactly as in Notation 2.6.
###### Theorem 3.3.
Let $`\mathrm{\Omega }^k`$ be a nice bounded open set. Let $`S^k`$ be a $`G_k`$-invariant subset such that $`\delta :=\rho _{\mathrm{}}(S)`$ exists.
Then
$$\underset{n\mathrm{}}{lim}\underset{xS\sqrt{n}\mathrm{\Omega }}{}p_n(x/\sqrt{n})=\delta 𝔑(\mathrm{\Omega }).$$
###### Proof.
The proof is exactly the same as that of Theorem 2.8, with the only change that instead of Proposition 2.3 we use Proposition 3.2. ∎
###### Convention 3.4.
From now and until the end of this section we assume that $`k=2`$.
###### Proposition 3.5.
We have
$$\rho _{\mathrm{}}(U_1(e))=\frac{1}{3}\rho _{\mathrm{}}(U_1)=\frac{1}{3\zeta (2)}.$$
###### Proof.
Let $`r,s1`$ be real numbers. For $`X,Y\{A,O,E\}`$ we denote by $`XY(r,s)`$ the number of all $`z=(z_1,z_2)U_1`$ such that $`0z_1<r`$, $`0z_2<s`$ and such that the parity of $`z_1`$ is $`X`$ and the parity of $`z_2`$ is $`Y`$. Here $`A`$ stands for “any”, $`E`$ stands for “even” and $`O`$ stands for “odd”.
Let $`n1,m1`$ be integers. Then $`AA(n,m)=nm`$. We will also use $`=^{}`$ to signify the equality up to an additive error term that is $`o(nm)`$. Note that $`EE(n,m)=0`$. Then we have
$$EO(n,m)=AO(n/2,m)=AA(n/2,m)AE(n/2,m)=$$
$$AA(n/2,m)OE(n/2,m)=^{}AA(n/2,m)EO(n/2,m)=^{}$$
$$\frac{1}{2}AA(n,m)\frac{1}{2}EO(n,m).$$
Therefore
$$\frac{3}{2}EO(n,m)=^{}\frac{1}{2}AA(n,m)EO(n,m)=^{}OE(n,m)=^{}\frac{1}{3}AA(n,m).$$
Hence $`EO(n,m)+OE(n,m)=^{}\frac{2}{3}AA(n,m)`$ which implies
$$OO(n,m)=OO(n,m)+EE(n,m)=^{}\frac{1}{3}AA(n,m).$$
Since $`\rho _{\mathrm{}}(U_1)=\frac{1}{\zeta (2)}`$, we have
$$\underset{n\mathrm{}}{lim}\frac{AA(n,n)}{n^2}=\frac{1}{\zeta (2)}$$
Therefore
$$\underset{n\mathrm{}}{lim}\frac{EE(n,n)+OO(n,n)}{n^2}=\frac{1}{3\zeta (2)},$$
which implies $`\rho _{\mathrm{}}(U_1(e))=\frac{1}{3\zeta (2)}`$, as required. ∎
We can now compute the limits for the spherical densities of the set of visible points for even and odd $`n`$ tending to infinity for the case $`k=2`$.
###### Theorem 3.6.
Let $`k=2`$. We have
$$\underset{m\mathrm{}}{lim}\frac{\gamma _A(2m,V_1)}{\gamma _A(2m,F)}=\frac{2}{3\zeta (2)}=\frac{4}{\pi ^2}$$
and
$$\underset{m\mathrm{}}{lim}\frac{\gamma _A(2m1,V_1)}{\gamma _A(2m1,F)}=\frac{8}{\pi ^2}.$$
###### Proof.
The proof is essentially the same as that of Theorem A. We present the details for completeness.
For $`c>0`$ denote $`\mathrm{\Omega }_c:=\{x^2:x<c\}`$. Then $`\underset{c\mathrm{}}{lim}𝔑(\mathrm{\Omega }_c)=1`$. Let $`ϵ>0`$ be arbitrary. Choose $`c>0`$ such that
$$|𝔑(\mathrm{\Omega }_c)1|ϵ/3$$
and such that
$$\underset{n\mathrm{}}{lim}\{p_n(x):x^2/\sqrt{n}\text{ and }xc\}ϵ/6.$$
By Theorem 3.3 and the above formula there is some $`n_01`$ such that for all $`nn_0`$ we have
$$\left|\underset{x\sqrt{n}\mathrm{\Omega }_c}{}p_n(x/\sqrt{n})\frac{1}{3\zeta (2)}𝔑(\mathrm{\Omega }_c)\right|ϵ/3$$
and
$$\{p_n(x):x^k/\sqrt{n}\text{ and }xc\}ϵ/3.$$
For an *even* $`n2`$ let
$$Q(n):=$$
$$\frac{\gamma _A(n1,\{wF:\overline{w}U_1(e)\})}{2\gamma _A(n1,F)}+\frac{\gamma _A(n,\{wF:\overline{w}U_1(e)\})}{2\gamma _A(n,F)}=$$
$$\frac{\gamma _A(n,\{wF:\overline{w}U_1(e)\})}{2\gamma _A(n,F)}.$$
In the above equality we use the fact that $`n`$ is even and all the points of $`U_1(e)`$ are even.
For an even $`nn_0`$ we have
$$Q(n)=$$
$$\frac{\mathrm{\#}\{wF:\overline{w}U_1(e),|w|_A=n1\text{ and }\overline{w}<c\sqrt{n}\}}{2\gamma _A(n1,F)}+$$
$$\frac{\mathrm{\#}\{wF:\overline{w}U_1(e),|w|_A=n\text{ and }\overline{w}<c\sqrt{n}\}}{2\gamma _A(n,F)}+$$
$$\frac{\mathrm{\#}\{wF:\overline{w}U_1(e),|w|_A=n1\text{ and }\overline{w}c\sqrt{n}\}}{2\gamma _A(n1,F)}+$$
$$\frac{\mathrm{\#}\{wF:\overline{w}U_1(e),|w|_A=n\text{ and }\overline{w}c\sqrt{n}\}}{2\gamma _A(n,F)}=$$
$$\underset{xU_1(e)\sqrt{n}\mathrm{\Omega }_c}{}p_n(x/\sqrt{n})+\underset{xU_1(e)(^k\sqrt{n}\mathrm{\Omega }_c)}{}p_n(x/\sqrt{n})$$
In the last line of the above equation, the first sum differs from $`\frac{1}{3\zeta (2)}𝔑(\mathrm{\Omega }_c)`$ by at most $`ϵ/3`$ since $`nn_0`$, and the second sum is $`ϵ/3`$ by the choice of $`c`$ and $`n_0`$. Therefore, again by the choice of $`c`$, we have $`|Q(n)\frac{1}{3\zeta (2)}|ϵ`$. Since $`ϵ>0`$ was arbitrary, this implies that $`\underset{m\mathrm{}}{lim}Q(2m)={\displaystyle \frac{1}{3\zeta (2)}}`$. Therefore
$$\underset{m\mathrm{}}{lim}\frac{\gamma _A(2m,V_1)}{\gamma _A(2m,F)}=2\underset{m\mathrm{}}{lim}Q(2m)=\frac{2}{3\zeta (2)}=\frac{4}{\pi ^2}.$$
Together with the conclusion of Theorem B this implies that
$$\underset{m\mathrm{}}{lim}\frac{\gamma _A(2m1,V_1)}{\gamma _A(2m1,F)}=\frac{8}{\pi ^2},$$
as claimed. ∎
## 4. Test elements in the free group of rank two
A subgroup $`H`$ of a group $`G`$ is called a *retract* of $`G`$ if there exists a *retraction* from $`G`$ to $`H`$, that is, an endomorphism $`\varphi :GG`$ such that $`H=\varphi (G)`$ and that $`\varphi |_H=Id_H`$. A retract $`HG`$ is *proper* if $`HG`$ and $`H1`$.
The following result is due to Turner :
###### Proposition 4.1.
Let $`F`$ be a free group of finite rank $`k2`$ and let $`wF`$. Then $`w`$ is a test element in $`F`$ if and only if $`w`$ does not belong to a proper retract of $`F`$.
If $`F`$ is a free group of rank two, then a proper retract of $`F`$ is necessarily cyclic. The following explicit characterization of retracts in this case is actually Exercise 25 on page 103 of Magnus, Karrass, Solitar . We present a proof here for completeness.
###### Lemma 4.2.
Let $`F`$ be a free group of rank two and let $`H=hF`$ be an infinite cyclic subgroup of $`F`$.
Then $`H`$ is a retract of $`F`$ if and only if there is a free basis $`\{a,b\}`$ of $`F`$ such that $`h`$ can be represented as $`h=ac`$, where $`c`$ belongs to the normal closure of $`b`$ in $`F`$. In particular, if $`H`$ is a retract of $`F`$ then $`H`$ is a maximal cyclic subgroup of $`F`$.
###### Proof.
Suppose first that $`H`$ is a retract of $`F`$ and that $`\varphi :FF`$ is a retraction with $`\varphi (F)=H`$. Choose a free basis $`x,y`$ of $`F`$. Since $`H=h=\varphi (x),\varphi (y)F`$ is infinite cyclic, the pair $`(x,y)`$ is Nielsen equivalent to the pair $`(h,1)`$. Applying the same sequence of Nielsen transformations to $`(x,y)`$ we obtain a free basis $`(a,b)`$ of $`F`$ such that $`\varphi (a)=h`$ and $`\varphi (b)=1`$. Then the kernel of $`\varphi `$ is the normal closure of $`b`$ in $`F`$.
Since $`\varphi `$ is a retraction onto $`H`$, we have $`\varphi (h)=h=\varphi (a)`$. Hence $`a^1hker(\varphi )`$ and therefore $`h=ac`$, where $`c`$ belongs to the kernel of $`\varphi `$, that is, to the normal closure of $`b`$, as required.
Suppose now that for some free basis $`a,b`$ of $`F`$ we have $`h=ac`$ where $`c`$ belongs to the normal closure of $`b`$ in $`F`$. Consider the endomorphism $`\psi :FF`$ defined by $`\psi (a)=h`$, $`\psi (b)=1`$. Then, clearly, $`\psi (h)=h`$ and $`\psi `$ is a retraction from $`F`$ to $`H`$. ∎
We can now obtain an explicit characterization of test elements in free group $`F`$ of rank with free basis $`A=\{a,b\}`$. We identify the abelianization of $`F`$ with $`^2`$ so that $`\overline{a}=(1,0)`$ and $`\overline{b}=(0,1)`$. If $`xA`$ and $`wF`$, then $`w_x`$ denotes the exponent sum on $`x`$ in $`w`$ when $`w`$ is written as a freely reduced word in $`A`$ and $`\overline{w}`$ denotes the image of $`w`$ in the abelianization of $`F`$. Thus $`\overline{w}=(w_a,w_b)`$.
###### Proposition 4.3.
Let $`F`$ be a free group of rank two. Let $`wF`$ be a nontrivial element that is not a proper power in $`F`$.
Then $`w`$ is a test element in $`F`$ if and only if there exists an integer $`n2`$ such that $`\overline{w}`$ is an $`n`$-th power in $`^2`$. That is, $`w`$ is not a test element if and only if $`w`$ is visible in $`F`$.
###### Proof.
Suppose first that $`w`$ is a test element but that $`\overline{w}`$ cannot be represented as an $`n`$-th power in $`^2`$ for $`n2`$. Then $`gcd(w_a,w_b)=1`$. Hence there exist integers $`p`$ and $`q`$ such that $`pw_a+qw_b=1`$. Consider an endomorphism $`\varphi :FF`$ defined by $`\varphi (a)=w^p`$ and $`\varphi (b)=w^q`$. Then $`\varphi (w)=w`$ and $`\varphi `$ is not an automorphism of $`F`$ since $`\varphi (F)`$ is cyclic. Hence, by definition, $`w`$ is not a test element in $`F`$, yielding a contradiction.
Suppose now that $`\overline{w}`$ is an $`n`$-th power in $`^2`$ for some $`n2`$ but that $`w`$ is not a test element. Then by Proposition 4.1 $`w`$ belongs to an infinite cyclic proper retract $`H`$ of $`F`$. Since by assumption $`w`$ is not a proper power in $`F`$, it follows that $`w`$ generates $`H`$. Lemma 4.2 implies that for some free basis $`(a_1,b_1)`$ of $`F`$ we have $`w=a_1c`$ where $`c`$ belongs to the normal closure of $`b_1`$ in $`F`$. Hence when $`w`$ is expressed as a word in $`a_1,b_1`$, the exponent sum on $`a_1`$ in $`w`$ is equal to $`1`$, which contradicts the assumption that $`\overline{w}`$ is an $`n`$-th power in the abelianization of $`F`$. ∎
Note that if $`wF=F(a,b)`$ then $`\overline{w}`$ is an $`n`$-th power in $`^2`$ for some $`n2`$ if and only if $`gcd(w_a,w_b)>1`$. By convention we set $`gcd(0,0)=\mathrm{}`$.
It is well-known and easy to prove that the set of proper powers in a free group is negligible :
###### Proposition 4.4.
Let $`F=F(A)`$ be a free group of finite rank $`k2`$ with free basis $`A=\{a_1,\mathrm{},a_k\}`$. Let $`P`$ be the set of all nontrivial elements of $`F`$ that are proper powers.
Then
$$\underset{n\mathrm{}}{lim}\frac{\gamma _A(n,P)}{\gamma _A(n,F)}=\underset{n\mathrm{}}{lim}\frac{\rho _A(n,P)}{\rho _A(n,F)}=0.$$
and the convergence in both limits is exponentially fast.
###### Proof of Theorem D.
Since $`\zeta (2)=\frac{\pi ^2}{6}`$, Theorem D now follows directly from Theorem B, Proposition 4.3 and Proposition 4.4. ∎
## 5. Open Problems
As before, let $`F=F(A)`$ be a free group of rank $`k2`$ with free basis $`A=\{a_1,\mathrm{},a_k\}`$ and let $`\alpha :F^k`$ be the abelianization homomorphism.
###### Problem 5.1.
Let $`k3`$. Is the set of test elements negligible in $`F`$? In view of Proposition 4.1, this is equivalent to asking if the union of all proper retracts of $`F`$ is generic in $`F`$.
The proof of Proposition 4.3 shows that a visible element in $`F`$ is never a test element and therefore by Theorem B the asymptotic density of the set of test elements in $`F`$ is at most $`1\frac{1}{\zeta (k)}`$. For $`k2`$ we have $`0<1\frac{1}{\zeta (k)}<1`$ and $`\underset{k\mathrm{}}{lim}1{\displaystyle \frac{1}{\zeta (k)}}=0`$. Thus the asymptotic density of the set of test elements of $`F`$ tends to zero as the rank $`k`$ of $`F`$ tends to infinity.
Note that every free factor of $`F`$ is a retract, but the converse is not true. As mentioned in the Introduction, the union of all proper free factors is negligible in $`F`$, whereas the union of all proper retracts is not since every visible element of $`F`$ belongs to a proper retract.
###### Problem 5.2.
For $`k2`$ find a subset $`S^k`$ such that $`\overline{\rho }_{||.||_{\mathrm{}}}(S)\overline{\sigma }_A(\alpha ^1(S))`$.
Note that if such a set $`S`$ exists then it is not invariant under the action of $`SL(k,)`$ in view of Theorem A,
###### Problem 5.3.
For $`wF`$ define $`T(w)=0`$ if $`\alpha (w)=0`$ and define $`T(w)`$ to be the greatest common divisor of the coordinates of $`\alpha (w)`$ if $`\alpha (w)0`$. Let $`T_n^{}`$ be the expected value of $`T`$ over the sphere of radius $`n`$ in $`F`$ with respect to the uniform distribution on that sphere and let $`T_n=(T_{n1}^{}+T_n^{})/2`$. What can one say about the behavior of $`T_n`$ as $`n\mathrm{}`$?
Using the results of this paper we can show that $`lim_n\mathrm{}T_n=\mathrm{}`$ for the case $`k=2`$. It also seems plausible that for each $`k3`$ we have $`lim\; sup_n\mathrm{}T_n<\mathrm{}`$ and heuristic considerations allow us to conjecture that in fact $`lim_n\mathrm{}T_n=\frac{\zeta (k1)}{\zeta (k)}`$. A similar question for $`^2`$ has been studied in detail by Diaconis and Erdös , who computed the precise asymptotics, as $`n\mathrm{}`$, of the expected value for the greatest common divisor of the coordinates, computed for the uniform distribution on the $`n\times n`$-square in $`^2`$.
###### Problem 5.4.
Let $`G`$ be a torsion-free *one-ended* word-hyperbolic group. Is it true that the set of test elements in $`G`$ is generic with respect to the word-length?
Although we do not know the asymptotic density of the set of test elements in a free group of rank $`k3`$, one may still expect a positive answer to Problem 5.4, especially if the hyperbolic group $`G`$ is not just one-ended but also does not admit essential $``$-splittings. In this case the structure of endomorphisms and automorphisms of $`G`$ is much more restricted than in free groups.
As we have seen, the set of proper powers is negligible in free groups of rank $`k2`$ but has positive asymptotic density in free abelian groups of finite rank. This raises the corresponding question about free groups in other varieties. It is possible to show that if $`G`$ is a finitely generated nilpotent group and $`t2`$ then the set of $`t`$-th powers has positive asymptotic density in $`G`$.
###### Problem 5.5.
Let $`G`$ be a finitely generated nonabelian free solvable group. What can be said about the asymptotic density of the set of all proper powers in $`G`$? |
warning/0507/astro-ph0507033.html | ar5iv | text | # Corrections of Errors in ‘The First Detections of the Extragalactic Background Light at 3000, 5500, and 8000 Å. I, II, and III’
## 1 Introduction
In Bernstein, Freedman & Madore (2002a,b,c; hereafter, BFM1, BFM2 and BFM3, respectively), we described a measurement of the mean flux of the extragalactic background light (EBL) in a $`5`$ arcmin<sup>2</sup> field of view. In that study, the EBL contribution was identified by measuring the total flux in the target field and subtracting from it the flux contributed by known foreground sources, namely diffuse Galactic light (DGL) and zodiacal light (ZL):
$$I_{\mathrm{EBL}}=I_{\mathrm{tot}}I_{\mathrm{ZL}}I_{\mathrm{DGL}}.$$
(1)
The total flux, $`I_{\mathrm{tot}}`$, was measured from space using HST/WFPC2 imaging in the $`U`$, $`V`$, and $`I`$bands and using HST/FOS spectroscopy covering roughly 4000–7000Å. The zodiacal light, $`I_{\mathrm{ZL}}`$, was measured using ground–based spectrophotometry obtained at the du Pont 2.5 m telescope at Las Campanas Observatory (LCO) in Chile. The diffuse Galactic light, $`I_{\mathrm{DGL}}`$, was estimated from a simple scattering model. The EBL values at 3000, 5500, and 8000Å were measured to be 4.0 ($`\pm `$2.5), 2.7 ($`\pm `$1.4), and 2.2 ($`\pm `$1.0) $`\times 10^9`$ ergs sec<sup>-1</sup> cm<sup>-2</sup> sr<sup>-1</sup> Å<sup>-1</sup>, respectively.
The zodiacal light measurement, described in BFM2, is the only part of the experiment involving ground–based observations. The Earth’s atmosphere influences these observations through absorption, scattering, and airglow emission. Absorption and scattering cause “extinction” of the light in the target field; scattering causes a fraction of the light from the full, visible hemisphere of the sky to be added to the line of sight; and airglow is an additive foreground source produced in the atmosphere. The resulting night–sky spectrum observed from the ground ($`I_{\mathrm{NS}}`$) was therefore described as a function of time($`t`$), airmass ($`\chi `$), atmospheric extinction ($`\tau _{\mathrm{obs}}`$), and wavelength ($`\lambda `$), as follows (Equation 3 of BFM2):
$$I_{\mathrm{NS}}(\lambda ,t,\chi )=I_{\mathrm{target}}e^{\tau _{\mathrm{obs}}(\lambda )\chi }+I_{\mathrm{scat}}(\lambda ,t,\chi )+I_{\mathrm{air}}(\lambda ,t,\chi ),$$
(2)
in which $`I_{\mathrm{scat}}`$ is the spectrum of light scattered into the line of sight and $`I_{\mathrm{air}}`$ is the airglow spectrum. To measure the zodiacal light in our experiment, we utilized the fact that the zodiacal light is known to have a slightly–reddened Solar spectrum in which the Solar–strength Fraunhofer lines are preserved. The zodiacal light contamination can therefore be expressed as the product of a fiducial Solar spectrum, $`I_{}(\lambda )`$, and a scaling function, $`C(\lambda )`$, that varies roughly linearly with wavelength. The airglow spectrum, on the other hand, is not known to contain Fraunhofer features. We therefore identified the ZL flux by iteratively determining the scaling factor, $`C(\lambda )`$, for which the resulting residual airglow spectrum has the minimum correlation with the Solar spectrum. We expressed the airglow spectrum as
$$I_{\mathrm{air}}(\lambda ,t,\chi )=I_{\mathrm{NS}}(\lambda ,t,\chi )I_{\mathrm{ZL}}\left[e^{\tau _{\mathrm{obs}}(\lambda )\chi }+\left(\frac{I_{\mathrm{EBL}}(\lambda )+I_{\mathrm{DGL}}(\lambda )}{I_{\mathrm{ZL}}}\right)e^{\tau _{\mathrm{obs}}(\lambda )\chi }+\frac{I_{\mathrm{scat}}(\lambda ,t,\chi )}{I_{\mathrm{ZL}}}\right],$$
(3)
in which the term $`e^{\tau _{\mathrm{obs}}(\lambda )\chi }`$ accounts for ZL flux lost from the beam due to extinction and the scattered light term, $`I_{\mathrm{scat}}`$, includes ZL, ISL (integrated star light), and DGL (diffuse Galactic light) as contaminating sources. As discussed in BFM2, we then needed a model for each scattering source over the visible spectrum in order to calculate the scattered light. To eliminate the absolute flux of the ZL from the models, we expressed the scattered light from all sources as a fraction of the ZL in the target field, as implied by Equation 3. We then combined the net ZL loss due to extinction with net ZL gain due to scattering to give an effective extinction, $`\tau _{\mathrm{eff}}(\lambda )`$. This let us express the effect of the atmosphere on the ZL as a relative (multiplicative) correction. The absolute value of the ISL, DGL, and EBL remain in the calculation. However, the EBL and DGL terms were then dropped because they were not expected to have Fraunhofer features and so were not expected to impact the spectral measurement based on the strength of these features.<sup>4</sup><sup>4</sup>4This is correct for the EBL, however the DGL can and does contribute Fraunhofer features and its spectrum should be included as a contribution to the target field and as a source of scattering in the spectral measurement of the ZL. The strength of zodiacal Fraunhofer lines in the DGL is weaker than in the solar spectrum by a factor of three, so that the impact on our measurement would be roughly $`0.01\times I_{\mathrm{ZL}}`$. We therefore identified the ZL flux (expressed as $`C(\lambda )I_{}`$) according to the equation:
$$I_{\mathrm{air}}(\lambda ,t,\chi )=I_{\mathrm{NS}}(\lambda ,t,\chi )C(\lambda )I_{}\left[e^{\tau _{\mathrm{eff}}(\lambda )\chi }+\frac{I_{\mathrm{scat}}^{\mathrm{ISL}}(\lambda ,t,\chi )}{I_{\mathrm{ZL}}}\right].$$
(4)
We correct errors in our (2002) papers regarding the dates of the Las Campanas Observatory (LCO) observations, a statement regarding the location of the Moon on those dates, and quantify the implications of these corrections. We also include a discussion of analysis errors which pertains to all unrefereed work prior to the (2002) papers (Bernstein, Freedman, & Madore 1996; Bernstein & Madore 1997; Bernstein 1998PhDT; Bernstein 1999a; Bernstein 1999b); these were corrected before publication of BFM2 and lead to no corrections here. Some of these errors were noted by Mattila (2003). We adopt nomenclature consistent with that of our earlier work to allow clear discussion of what was done there. Throughout this paper, we abbreviate $`10^9`$ ergs sec<sup>-1</sup> cm<sup>-2</sup> sr<sup>-1</sup> Å<sup>-1</sup> as cgs.
## 2 Errors in Early Analysis
Mattila (2003) has correctly noted that the analysis of the ground–based data as detailed in the unpublished thesis (Bernstein 1998) contained an incorrect treatment of atmospheric effects. In that early analysis, atmospheric scattering was not included as a contribution to the observed night sky spectrum. In addition, due to a programming error in a subroutine, an incorrect extinction correction was applied. The incorrect treatment of atmospheric scattering was identified by the referee and both errors were corrected before publication. In the unpublished thesis, prior to the correction of these two errors, the ZL was therefore calculated based on the following expression (compare to Equation 4 above):
$$I_{\mathrm{air}}(\lambda ,t,\chi )=I_{\mathrm{NS}}(\lambda ,t,\chi )C(\lambda )I_{}e^{[\tau _{\mathrm{obs}}(\lambda )\tau _{\mathrm{obs}}(4600)]\chi }.$$
(5)
Over the wavelength range used in that analysis (4200–5100Å) and at the mean airmass of our observations ($`\chi `$ 1.2), the exponential term in Equation 5 has values between 0.93 and 1.06, with an average value of 1.00. In effect, the data were analyzed with no scattering correction and an incorrect extinction correction.
In brief, the analysis in Bernstein (1998) involved preparing solar spectra appropriate to each observation using IRAF routines to resample in wavelength, and apply an extinction curve for the airmass of each observation. The zodiacal light solution is then a scaling value relative to these prepared fiducial solar spectra. The solar spectra, corrected for extinction, were compared with the corresponding LCO spectra to find the contribution of zodiacal light. These solar spectra were produced many times in the course of data reduction, because the extinction and sensitivity solutions were recalculated many times. A check was therefore included in a subroutine to confirm that the solar spectra were correctly prepared. That check involved multiplying the solar spectrum by $`\mathrm{exp}[\tau _{\mathrm{obs}}(4600)\chi ]`$ with $`\tau _{\mathrm{obs}}(4600)=0.2`$ mag/airmass, roughly removing the extinction correction. The error then occurred by passing the wrong vector back to the main program from the subroutine. The programming error was not identified until the anonymous referee pointed out the incorrect treatment of atmospheric scattering. For completeness, we note that Figure 4.4 in Bernstein (1998) does not show the final extinction solution used in the thesis. The correct extinction solution used in all versions is shown in BFM2. Table 2.9 in Bernstein (1998), which lists values for $`I_{\mathrm{tot}}`$, was also updated in the published papers.
When properly treated, the scattering and extinction are nearly equal in magnitude but opposite in sign, and so they cancel to a high degree (to about 0.5% averaged over wavelength and airmass), giving the same result as the original analysis to within the accuracy of the measurements. The cancellation of the scattering and extinction terms in the proper analysis can be seen in the following quantitative example. At 4600Å and at our mean airmass (as given above), the extinction coefficient (Figure 29 of BFM2) is $`\tau _{\mathrm{eff}}(4600)=0.042`$ mag/airmass and, accordingly, $`\mathrm{exp}[\tau _{\mathrm{eff}}(\lambda )\chi ]`$ = 0.955. Over the spectral range 3900–5100Å and at the same airmass, the net flux gained from scattered ISL is in the range 10–17 cgs. At 4600Å, the ISL scattered flux is 12.5 cgs. The ISL contribution impacts our ZL measurement to the degree that it contributes to the strength of the Fraunhofer lines in the observed night sky spectrum; however, those features are only 10% to 40% as strong in the scattered ISL as in the ZL over the range 3900–5100Å, and 30% to 40% as strong around 4600Å (see Figures 29 and 30 of BFM2). The scattered ISL therefore contributes +4.4 cgs (=12.5 cgs $`\times `$ 0.35) to the solution, or $`0.040\times I_{\mathrm{ZL}}`$ (given that $`I_{\mathrm{ZL}}`$ is roughly 110 cgs.) The term in square brackets in Equation 4 is therefore nearly unity (0.995 for this example). At higher airmasses and shorter wavelengths, scattered flux ($`I_{\mathrm{scat}}`$) and extinction ($`\tau _{\mathrm{eff}}`$) both increase. At smaller airmasses and longer wavelengths, they both decrease. In either case, the term in square brackets in Equation 4 is still nearly unity. In other words, the net loss due to extinction and the net gain due to scattering are synchronized and cancel to a level that is much smaller than the uncertainty in identifying the ZL flux contribution in the 16 spectra used in this analysis, which have an rms scatter of 2.3%. Because of this cancellation, statistically indistinguishable results were obtained in the early (Bernstein 1998) and published (BFM2) versions of the analysis. Note that the similarity between the net effects of atmospheric scattering and extinction alone are coincidental, and would likely not occur along lines of sight where the ZL in the target field is much brighter or fainter. They are also, of course, dependent on the parameters used to calculate the scattering model, which are documented explicitly in BFM2.
In the early analysis, no change in $`C(\lambda )`$ with wavelength was detected because the incorrect extinction correction masked the reddening of the ZL relative to the Sun. Because no color term was detected, several broad bandpasses were used. In the published version, an increase in $`C(\lambda )`$ with wavelength was identified, consistent with the reddened color of the ZL relative to the solar spectrum. Narrow bandpasses focused on the solar features were then used to help identify this trend.
## 3 Dates of Ground–Based Observations.
The dates of the ground–based observations were incorrectly recorded in the unpublished thesis and were subsequently transcribed by RAB from there into the published papers. The original observing logs and the records of the observatory show that the correct dates of the run were the local–time nights of 23/24 November 1995 through 27/28 November 1995 (5 nights total). The last night of the run was used for imaging. Data from the first and third nights were not used due to weather and mechanical problems, as described in Bernstein (1998) and BFM2. The spectra cited and analyzed in BFM2 were therefore taken on nights 2 and 4 of the run, having local–time dates 24/25 and 26/27 November 1995. The corresponding UT dates were 25 and 27 November 1995. The incorrect dates affect the application of the zodiacal light measurement to the HST observations at the level of 0.2% (although with significant uncertainty) and also affect the scattering calculations at the level of $`<0.1`$%. We describe and quantify these two effects below.
### 3.1 Relevance for HST Observations
The HST observations analyzed in BFM1 were executed on the UT nights of 29 November 1995 and 16–17 December 1995. Ground–based observations were assigned and scheduled by the time allocation committee about one year earlier. As stated in the abstract of BFM1, the observations were designed to occur contemporaneously with one of the sets of HST observations, but they were not executed simultaneously.
To get an idea what the change in the ZL value might be between our LCO observations on 25/27 November 1995 and the HST observations on 29 November 1995, we can look at data in the literature and our own HST data. From 29 November 1995 to 16/17 December 1995, the HST/WFPC2 and FOS data both showed a $`2`$% decrease in the mean surface brightness of the EBL target field. As discussed in BFM1 and BFM2, this difference is what would be expected in sign and magnitude as the heliocentric longitude ($`\lambda \lambda _{}`$) of the target field goes from about $`150^{}`$ on 29 November 1995 to $`130^{}`$ on 16/17 December 1995. One expects this small decrease in intensity because the ZL is slightly brighter in the anti-solar direction ($`\lambda \lambda _{}=180^{}`$) and has a broad minimum at $`\lambda \lambda _{}=130^{}`$.
For comparison, several data sets are available in the literature. The only all–sky measurements of the ZL surface brightness are from the ground. Of these, the most reliable are those tabulated by Levasseur–Regourd & Dumont (1980, hereafter LRD80) from their 1964–1975 observations at Tenerife Observatory. That data set is reproduced in Leinert et al. (1998), where it is updated with space–based values within 30 degrees of the sun. Although these data are ground–based and subject to scattering corrections, they are in good agreement with space–based results, as discussed in LRD80, Leinert et al. (1998), and BFM2. Between $`\lambda \lambda _{}=150^{}`$ and $`130^{}`$ and ecliptic latitude $`31^{}35^{}`$, the data tabulated in LRD80 show a –6% change in the ZL flux. At these latitudes, data are also available from several other sources. As compared and discussed in Leinert et al. (1981) and Leinert et al. 1998), Frey et al. (1970) find a change of about $`+2`$% over these same angles, and the Helios space probes (Leinert et al. 1981, Leinert et al. 1982) find a change of –1%. These three published results are in good agreement to within the errors of any of the measurements, which are of order 5–10%.
To be conservative in estimating the change in the ZL between 29 November 1995 and 16/17 December 1995, we simply average the values discussed above (–2%, –6%, $`+2`$%, –1%) to obtain –1.7% with a standard deviation of 3.3%. As the errors are probably systematic, the standard deviation may be more indicative of the uncertainty than the error in the mean. We then estimate that the change in the zodiacal light between 25/27 November LCO observations ($`\lambda \lambda _{}`$ = 153 and 151) and 29 November for Hubble Space Telescope (HST) observations ($`\lambda \lambda _{}`$ = 149) should be $`0.2(\pm 0.3)`$% (i.e., slightly fainter on the 29th). To conservatively allow for any systematic uncertainties between the data sets, we double this error bar to $`\pm 0.6`$%. We include this offset in the summary in Table 2.
### 3.2 Relevance for Scattering Calculations
Because of the transcription error in the dates of the observations, the scattering calculations in BFM2 were performed for the wrong date, namely the 29th rather than the 25th and 27th of November 1995. The sky visible at a particular UT time shifts by roughly one degree per day. However, the target field and all sources of scattering obviously move in consort, so that the scattering calculated for a given zenith angle of the target is correct on any date. The one source which does not move in consort is the moon; however, any spectrum affected by moonlight should not be included in the analysis (see §4), and so the moon is not included in the calculation of scattered light from extraterrestrial sources. The only change in the scattering calculations between November 25, 27, and 29 is therefore caused by the fact that the target field will rise 4 minutes earlier on each successive night. The scattering calculations for UT=2:00 on November 29 are therefore correct for UT=2:16 on November 25, and UT=2:08 on November 23. The net change in the scattered ZL and ISL for the largest difference in timing (16 minutes) is very small. Moreover, as illustrated in §2, the net effect of the atmosphere (scattering and extinction of ZL, and scattering of ISL) nearly cancels at every zenith angle. For that reason, the change over 8 or 16 minutes is not detectable.
To illustrate this quantitatively, we note that the mean change in the effective extinction (Figure 25, BFM2) is smaller by an average of 0.0013 mag/airmass between a given UT time and 16 minutes earlier. This translates into a fractional change of 0.03 in $`\tau _{\mathrm{eff}}`$. Using the mean airmass, $`\chi =1.2`$, this corresponds to an increase in the net ZL by a factor of 1.0015. As the scattered ISL gets brighter with increasing UT time, the scattered ISL would be correspondingly fainter by about 4% over that same time period (16 minutes earlier), and would decrease the strength of the ISL spectral features by the factor 0.998 (= $`0.04\times 0.35\times `$ 12.5 cgs/$`I_{\mathrm{ZL}}`$). The net change with time is then 0.998$`\times `$1.0015 $`=`$ 0.9995, which is not significant. Nevertheless, for completeness, we list this term, and all other corrections discussed here in the summary in Table 2.
## 4 Location of the Moon
The LCO data were obtained by RAB and took place several days after new moon. Each night, as the Moon was setting, the open–dome time was used to observe bright, standard stars for calibration. Observations of the EBL target field began as the Moon approached the horizon. The general, but quantitative, statement in BFM2 that the Moon was at least $`14`$ degrees below the horizon during all but one exposure is incorrect. The correct statement is that one exposure that was used in the analysis began as the Moon was still setting (UT = 03:10 on 27 November 1995); all other exposures used in the analysis were taken with the Moon below the horizon by several degrees. The times of the first few exposures and the corresponding position of the Moon during those exposures are given in Table 1 for all spectra taken until the Moon was more than 22 degrees below the horizon on November 25 and 27. The spectra from these exposures are plotted in Figure 1.
Because of its high mean flux level, it was clear to us that the exposure beginning at UT = 02:37 (open square in Figure 1) was affected by moonlight and for that reason it was not used in the analysis in BFM2. The moon contributes exponentially less light with time after passing below the horizon (like the sun at sunset). The remaining exposures were not obviously affected and were therefore included in the subsequent analysis. We now consider what the impact of the Moon might have been on the included spectra.
To obtain a theoretical estimate of the scattered moonlight which might influence each exposure, we can use the scattering model described in BFM2. These estimates are given in Table 1 as a fraction of the ZL flux in the target field. Simpler models for moonlight sky brightness, such as that implemented by Skycalc.V5 (Thorstensen 2001), give consistent values at zenith angles smaller than about $`85^{}`$, but yield higher values for the sky brightness very near the horizon. (These models do not predict the moonlight below the horizon. See Krisciunas & Schaefer (1991) for a discussion of the model implemented in Skycalc.V5 and its uncertainties.) For the lunar phase and angular distance of the target from the Moon ($`90^{}`$ on 27 Nov. 1995), the estimated moonlight flux is negligible ($`<<1`$%) by the time the Moon reaches a zenith angle of $`98^{}`$.
We can also obtain an empirical estimate of the scattered moonlight in each exposure by simply comparing their mean fluxes. For the exposure beginning at UT = 02:37 (which was clearly affected by moonlight and was not used in the analysis), the scattered moonlight is estimated to be about $`0.60\times I_{\mathrm{ZL}}`$ at 4600Å. The ZL accounts for roughly two thirds of the night sky flux (see Figure 9, BFM2), so that this spectrum is predicted to be about 40% brighter due to moonlight than spectra taken later. This is generally consistent with the empirical mean flux of the spectrum, which appears to be about 43% brighter than later spectra (open square, Figure 1). For the exposure beginning at UT = 03:10 (27 Nov. 1995), the scattered moonlight is estimated to be $`0.11\times I_{\mathrm{ZL}}`$, implying that the mean flux for this spectrum should be about 8% higher than later spectra. The flux of this spectrum appears to be about 5% higher than the mean (open circle, Figure 1), which is again generally consistent to within the accuracy of the scattering models at very high airmasses. Note also that there is about 5% peak–to–peak variation in the mean flux of spectra that are not influenced by moonlight. This is presumably due to changes in atmospheric effects (changes in airglow, changes in extinction with airmass, and changes in scattered starlight and diffuse galactic light). For this reason, the spectrum at UT=03:10 did not obviously appear to be problematic.
We conclude from the predicted and empirical fluxes of the spectra discussed above that the spectrum taken at UT=03:10 was probably affected by moonlight, and so we recalculate the final result without it. The ZL value derived from that exposure alone is $`113\pm 3`$ (1$`\sigma `$), which is about 3–4% higher than the mean (see Figures 12 & 13, BFM2). Excluding this data point, the final ZL result (based on the average of 16 observations) is lower by 0.3%, which is roughly 1/2 the quoted statistical uncertainty and 1/4 the systematic uncertainty. The effect of excluding this exposure from the analysis is included in the summary in Table 2.
## 5 Summary
We have presented corrections to the published results (BFM1,2,3), including the dates of the ground–based observations, the location of the Moon during each exposure, and the quantitative impact of these corrections. In addition, we have explicitly documented corrections made to the analysis between the unpublished thesis (Bernstein 1998) and published versions of this work (BFM2). The measured value of the ZL decreases by 0.5($`\pm `$0.6)%, or 0.5($`\pm `$0.6 cgs) as a result of these changes. For comparison, the quoted measurement uncertainties in BFM2 are 0.6%, statistical, and 1.1%, systematic. The inferred EBL increases correspondingly by $`0.1`$, $`0.5`$, and $`0.7`$ cgs in the $`U`$, $`V`$, and $`I`$bands. For comparison, the quoted $`1\sigma `$ uncertainties in each band were 2.5, 1.4, and 1.0 cgs, respectively. The corrections discussed here therefore yield a result that is consistent with the previously quoted result and errors; however, this is not intended to be a new or updated analysis.
We thank K. Mattila for his work and comments regarding these results. |
warning/0507/cs0507057.html | ar5iv | text | # A new sibling of 𝐵𝑄𝑃
## 1 Introduction
Quantum computing used to be a very popular discipline ranging from pure theoretical questions, concerning the complexity of the quantum polynomial-time class BQP, to practical concerns such as how to build a quantum computer. Nowadays, it seems a little bit that scientists are loosing their interest. Is it because all ”easy” questions have been answered and what reamins is too hard or unintresting? We can find many unanswered questions considering complexity classes. For example, we know that BPP$``$BQP$``$AWPP. But questions whether is BQP equal to its ”father” AWPP or its ”son” BPP have not been answered up to now. Here, we do not answer them either. Instead, we introduce a nontrivial ”brother”, which we call MQ<sup>2</sup>. Surprisingly, this brother can also factorize long integers in polynomial time. Moreover, it has a very compact mathematical definition, which does not explicitly involve any physics.
The paper is organized as follows. In section 2, we introduce the neccessary notation and definitions. In section 3, we briefly review classical polynomial-time classes definition. In the following section, we define the class MQ<sup>2</sup> itself. Fifth sections demontrates two quantum algorithms, Shor’s and Deutsch-Jozsa’s, in class MQ<sup>2</sup>.
## 2 Definitions and notation
In this text, we frequently encounter matrices. All matrices here will have square shape. We will denote the element of a matrix $`M`$ in $`i`$th row and $`j`$th column as $`i|M|j`$, in accordance with the usual notation used in quantum computing, because we think it makes the text more readable than writing $`M_{i,j}`$. The range of the indices will be $`0..n1`$ where $`n`$ is the number of columns or rows.
We will define our new complexity class with use of matrices. To make the class uniform, we will want that all the matrices are constructed by the same algorithm. We formalize this in the following definition.
###### Definition 1 (Poly-computable matrix family).
A sequence of matrices $`T_1,T_2,\mathrm{}`$ is called a poly-computable matrix family if there exists a function $`f:\times \times `$ computable in time polynomial in the length of all the arguments such that
$$i,j,n:f(i,j,n)=j|T_n|i.$$
Sometimes, we will drop the index $`n`$ when it will be clear from the context.
To show how the classical polynomial-time classes definitions correspond to our definition, we will start with a Turing machine and express its transition function as a transition matrix. A transition matrix is in fact a linear operator defined in a vector space spanned by configurations playing role of base vectors.
###### Definition 2 (Configuration of a Turing machine).
A configuration of a Turing machine is an ordered triple consisting of:
* -the contents of the tape
-the current state
-the position of the head <sup>2</sup><sup>2</sup>2We assume without loss of generality that the machine has only one tape.
We emphasize here that the configuration as defined above contains also the content of the tape, which is not true for configurations as defined elsewhere. Without loss of generality, we will further assume that configurations are indexed and denoted by their indices. A special position among the configurations has the family of initial configuration $`I(x)`$, which we allow to be dependent on $`x`$, but require to be computable in polynomial time. Again, we will drop the index $`n`$ when it will be clear which member of the family we mean. We will also need to be able to recognize accepting configurations. For this purpose, we will have a function $`a(x,c)`$, computable in polynomial time, where the first argument will be the input for the algorithm and $`c`$ is a configuration. The function $`a(x,c)`$ will return 1 iff the configuration $`c`$ is accepting (possibly depending on $`x`$) and 0 otherwise.
Now we are ready to jump to the notion of transition matrix. If a configuration $`c_1`$ leads to another configuration $`c_2`$ in the next step with probability $`p`$, there is $`p`$ on the position $`c_2|T|c_1`$, otherwise there is zero. Because the tapes are of unbounded size, so is the matrix. However, if we know that the time complexity of a Turing machine is $`T(n)`$, we may for fixed $`n`$ have a finite matrix cutting the tapes at the distance $`T(n)`$ from the initial position on both sides. The size of the matrix for inputs of length $`n`$ is then $`2^{O(T(n))}\times 2^{O(T(n))}`$. For a probabilistic Turing machine, the transition matrix is stochastic, e.g. every row sums up to 1. Transition matrices naturally form a poly-computable matrix family, since for each pair $`c_1`$, $`c_2`$, the probability of going from one to another can be read from the description of the underlying probabilistic Turing machine, which is a finite object<sup>3</sup><sup>3</sup>3It should be pointed out that for most poly-computable stochastic matrix families, no corresponding probabilistic Turing machine exists. The reason is that each probabilistic Turing machine has finite description of a bounded size, independent on the length of input, while in our definition 1, we allowed function $`f`$ to arbitrarily depend on $`n`$..
## 3 Traditional complexity classes
We will now briefly review classical complexity classes definitions. The common definitions of P, BPP, NP and PP involve a probabilistic Turing machine and look at the accepting probability for each input<sup>4</sup><sup>4</sup>4Obviously, the class P can be viewed as a special case of a probabilistic class, with probabilities either one or zero..
One step of a probabilistic Turing machine corresponds to multiplying the transition matrix with a vector representing the current configuration. Thus, instead of saying ”the probability of accepting on a configuration $`I(x)`$ after $`S`$ steps is $`p`$”, we may eqvivalently say ”$`_{c:a(c,x)=1}c|T^S|I(x)=p`$”. We will use this observation in the following definitions.
###### Definition 3 (Polynomial time classes in matrix notation).
A language $`L`$ is in class C if there exists a polynomial $`p(n)`$ and a probabilistic Turing machine $`M`$ with transition matrix family $`T_i`$ and functions $`I(x)`$, $`a(x,c)`$ computable in polynomial time, such that for all $`n`$ and for all $`x`$ of length $`n`$:
| complexity class C | P | NP | PP | BPP |
| --- | --- | --- | --- | --- |
| For $`xL`$, $`_{c:a(c,x)=1}c|T_n^{p(n)}|I(x)`$ | $`=1`$ | $`>0`$ | $`>\frac{1}{2}`$ | $`\frac{2}{3}`$ |
| For $`xL`$, $`_{c:a(c,x)=1}c|T_n^{p(n)}|I(x)`$ | $`=0`$ | $`=0`$ | $`\frac{1}{2}`$ | $`\frac{1}{3}`$ |
where exactly one of the columns applies.
In the same manner, we may define the quantum class BQP:
###### Definition 4 (BQP).
A language $`L`$ is in class BQP if there exists a polynomial $`p(n)`$ and a quantum Turing machine $`M`$ with transition matrix family $`T_i`$ of unitary matrices and functions $`I(x)`$, $`a(x,c)`$ computable in polynomial time, such that
* For $`xL`$: $`\left|_{c:a(c,x)=1}c|T^{p(n)}|I(x)\right|^2\frac{2}{3}`$
* For $`xL`$: $`\left|_{c:a(c,x)=1}c|T^{p(n)}|I(x)\right|^2\frac{1}{3}`$
We emphasize here that there are exactly two points in which Definition 4 and the Definition of BPP in Definition 3 differ: First, in Definition 3 we have stochastic matrices while in Definition 4 we have unitary matrices. Second, in Definition 3 we are looking at the value of $`_{c:a(c,x)=1}c|T^{p(n)}|I(x)`$, while in Definition 4 we look at the square norm of this value. However, the latter can be avoided, since we can eqvivalently use the square norm in the Definition 3 of BPP. Thus, the only remaining difference between the two classes is the type of matrices used.
Now we will define the class MQ<sup>2</sup> itself.
## 4 The definition
We alter the Definition 4 of class BQP to get a new class MQ<sup>2</sup>. At first, we will not use transition matrices, but instead a unitary, poly-computable matrix family. That is a weaker requirement. In turn, we will be more strict in the number of matrices allowed - we will only use two copies of the matrix.
###### Definition 5 (MQ<sup>2</sup>).
A language $`L`$ is in class MQ<sup>2</sup> iff there exists a unitary, poly-computable matrix family $`T`$, A poly-computable vector family $`I(x)`$ and function $`a(x,c)`$ computable in polynomial time such that <sup>5</sup><sup>5</sup>5 The numbers $`\frac{2}{3}`$ and $`\frac{1}{3}`$ in the definition can be amplified in the same way as in the case of BPP and BQP. Proven in
* For $`xL`$, $`_{c:a(c,x)=1}\left|c|T^2|I(x)\right|^2\frac{2}{3}`$
* For $`xL`$, $`_{c:a(c,x)=1}\left|c|T^2|I(x)\right|^2\frac{1}{3}`$
The resulting hierarchy is visualised in Figure 1. The inclusion MQ$`{}_{}{}^{2}`$AWPP was shown in .
## 5 Expressing quantum algorithms
In this section, we will demonstrate how two famous quantum algorithms fit into the class MQ<sup>2</sup>. Here we will only present the main ideas of the proofs. Full proofs can be found in .
At first, we show that class MQ<sup>2</sup> captures Deutsch-Jozsa’s problem. For this problem, see . In the Deutsch-Jozsa problem, a quantum oracle is used. That is a diagonal quantum gate, or in other words a diagonal unitary matrix, realizing the transformation $`x(1)^{f(x)}`$. Here, our matrix will be a product of a poly-computable matrix and this oracle. The result if then poly-computable too.
###### Theorem 6.
The class MQ<sup>2</sup> solves the Deutsch-Jozsa problem.
Proof sketch. We will mimic the circuit from Deutsch-Jozsa’s algorithm (see Figure 2 a)) by two copies of a poly-computable matrix family $`T`$ (see Figure 2 b)). For a fixed $`n`$, the matrix $`T`$ will be a product of the oracle and $`H^n`$. We may add another matrix for the $`f`$ function to the front, since it will only add the number $`(1)^{f(0)}`$ to the global phase and thus will not change the result. Formally, we define a matrix $`T`$ as
$`y|T|x\stackrel{\text{DEF}}{=}(1)^{f(x)}y|H^n|x={\displaystyle \frac{1}{\sqrt{2^n}}}(1)^{f(x)}(1)^{_ix_iy_imod2}`$
which is obviously computable in polytime and unitary. We define $`c_i(x)=0^n`$ and $`c_A(x)=0^n`$ for $`x`$ of length $`n`$. Then we have
$$\begin{array}{c}\left|0^n|T^2|0^n\right|^2=\left|\underset{k}{}0^n|T|kk|T|0^n\right|^2=\hfill \\ \hfill \left|\underset{k}{}(1)^{f(k)}\frac{1}{\sqrt{2^n}}(1)^{_ik_i0_{}^{n}{}_{i}{}^{}mod2}\frac{1}{\sqrt{2^n}}(1)^{f(0^n)}(1)^{_i0_{}^{n}{}_{i}{}^{}k_imod2}\right|^2=\\ \hfill =\left|\underset{k}{}\frac{1}{2^n}(1)^{f(k)}(1)^{2_ik_i0_{}^{n}{}_{i}{}^{}mod2}\right|^2=\frac{1}{2^{2n}}\left|\underset{k}{}(1)^{f(k)}\right|^2\end{array}$$
If the function is constant, then the sum $`_k(1)^{f(k)}`$ equals $`\pm 2^n`$ and the probability $`|0^n|T^2|0^n|^2`$ equals $`1`$. If the function is balanced, both the sum and the probability is 0.
$`\mathrm{}`$
Now we will show how to implement the famous Shor’s algorithm in class MQ<sup>2</sup>:
###### Theorem 7.
The class MQ<sup>2</sup> solves the factoring problem. More precisely, there exists a constant k such that given numbers x and N$``$k as in the Shor’s algorithm, the language
$$L=\{N,i|x^amodN\text{ has a period }r\text{ whose i-th bit is 1}\}$$
is in MQ<sup>2</sup>.
Proof sketch. We will use the same notation as in the original paper by Shor . $`N`$ is the number to factorize and $`N=p_1p_2`$ where both $`p_1`$ $`p_2`$ are primes and are different from each other. We then arbitrarily choose an $`x`$ coprime to $`N`$. The pair $`(x,N)`$ is the input of the algorithm. The goal is to find the smallest $`r0`$ such that $`x^rmodN=1`$. This $`r`$ is called a *period* of $`x`$. We choose a number $`q`$ such that $`q`$ is a power of $`2`$ and $`q2^{2log_2N}`$. This number will, together with the length of $`x`$ and $`N`$, determine the size of the matrix.
In the original setup, see Figure 3 a), we have three different transformations, namely: Hadamard transformation, a transformation computing power modulo $`N`$, and the $`DFT`$. One can notice that the effect on $`0^n`$ of the $`DFT`$ and of the Hadamard transform is the same, so we suffice with only two different transformations. Furthermore, the matrices realizing them are poly-computable:
$`x^{},N^{},a^{},i^{}|DFT|x,N,a,i`$ $`\stackrel{\text{DEF}}{=}{\displaystyle \frac{1}{\sqrt{q}}}\delta _{x,x^{}}\delta _{N,N^{}}\delta _{i,i^{}}e^{\frac{2i\pi }{q}aa^{}}`$
$`x^{},N^{},a^{},i^{}|MOD|x,N,a,i`$ $`\stackrel{\text{DEF}}{=}\delta _{x,x^{}}\delta _{N,N^{}}\delta _{a,a^{}}\delta _{i^{}+i,x^amodN}`$
One may also simply check that both the matrices are unitary. Their product $`T\stackrel{\text{DEF}}{=}DFTMOD`$ is thus unitary too. Its elements read
$$x^{},N^{},a^{},i^{}|T|x,N,a,i=\frac{1}{\sqrt{q}}\delta _{x,x^{}}\delta _{N,N^{}}e^{\frac{2i\pi }{q}a^{}a}\delta _{i+i^{},x^amodN}$$
and are again clearly functions computable in polytime. Applying two times the matrix $`T`$, we apply an extra $`MOD`$ transformation comparing to the original setup. Nevertheless, on $`0^n`$, this transform acts as identity. Any classical post-processing, as in the original algorithm, can be incorporated into the function $`a(x,c)`$.
$`\mathrm{}`$
## 6 Conclusion
We saw a complexity class which has a compact purely mathematical definition. In order to describe an algorithm, we suffice with three polynomial-time computable functions taking bit-strings as arguments: $`f(i,j,n)`$, $`I(x)`$, and $`a(x,c)`$. Even quite a complex algorithm, as the Shor’s certainly is, can be described on three lines. Further, the class MQ<sup>2</sup> shows that BQP is not the only possible class, lying in between BPP and AWPP, and not being trivially equal to either of them, which can do factorization and exponential speedup with oracles as in Deutsch-Jozsa’s algorithm.
## Acknowledgment
This material is partially based on the author’s master thesis , which was supervised by Harry Buhrman. I would like to use this opportunity to thank him for motivating discussions. |
warning/0507/hep-ph0507336.html | ar5iv | text | # Electromagnetic structure and weak decay of pseudoscalar mesons in a light-front QCD-inspired model
## I Introduction
The hadronic structure viewed through effective light-front theories inspired by Quantum Chromodynamics pauli0 ; pauli1 ; pauli2 ; pauli3 ; pauli4 ; tobpauli can shed light in the investigation of the interaction between the hadron constituents and in the study of the transition from effective to fundamental degrees of freedom that should be revealed at large momentum scales. The effective QCD-theory is expressed through a squared mass operator acting on the valence component of the hadron light-front wave function. The effective interaction embeds, in principle, all the complexity of QCD through the coupling of the valence state with higher Fock-states, reduced to the valence sector pauli4 . The effective squared mass operator acting depend on few physical parameters: the constituent quark masses, the effective quark-gluon coupling entering in the Coulomb-like interaction and the strength of the short range hyperfine interaction, fixed by the pion mass tobpauli . Hadronic observables are functions of these parameters, and one physically sensible is the dependence on the quark masses, which allows to get insight on the limit of heavy quarks. In particular, this limit is reflected in the weak decay constant of pseudoscalar heavy-light mesons, which in potential models were found to be $`1/\sqrt{m_Q}`$ cap90 ($`m_Q`$ is the heavy quark mass). Also, this scaling property has been shown in a light-front constituent quark model jaus96 . (We do not intend to be complete in our references.)
The weak decay constant is closely related to the physics at small distances contained in the valence component of the pseudoscalar meson light-front wave function. It constitutes and important source of information on the short-range part of interaction between the quark and antiquark. Experimental values of the weak decay constant are known for the pion, kaon, $`D^+`$ and $`D_s^+`$ pdg . The few parameter dependence of the effective theory can be translated into correlations between observables of a particular hadron or among different hadrons. Therefore, it is possible to indicate relevant relations between physical quantities that otherwise would have no simple reason to exhibit a close dependence, besides being properties of the same basic theory. For example, in the recent review cr of the application of Dyson-Schwinger equations to QCD, it was shown systematic correlations between different meson properties with mass scales, which was also useful to compare the pseudoscalar decay constants with results from Lattice QCD.
The experimental values of the mass splitting between the ground states of pseudoscalars and vector mesons presents a systematic dependence on the corresponding pseudoscalar mass, which is well described by the effective QCD-theory even without confinement tobpauli . The mass splitting is associated with the binding energy of the quark-antiquark pair in the meson, as that model lack a confining interaction. (It is worthwhile to note that the model account for the binding energy of the spin 1/2 ground state baryons containing two light and one heavy quarksuisso ).
In Ref. khoplov , its was pointed out that $`f_{ps}`$ should scale with the sum of the constituent quark masses, and more recently in the context of a light-front QCD-inspired model it was found that $`f_{ps}`$ scales with the vector meson mass luiz . In the light-front QCD-inspired model without a short-range regulator it was assumed the dominance of the asymptotic wave function luiz . Both frameworks assume that the log-type singularity in the matrix element of the axial current between the vacuum and the meson state is fixed by the pion decay constant $`f_\pi `$. In these approaches, the weak decay constant depends directly on the constituent masses and on the short distance component of the valence part of the light-front meson wave function parameterized by $`f_\pi `$. Although, the experimental results for light mesons up to $`D`$ pdg suggests such increase, relativistic constituent quark models in the heavy-quark limit (see Refs. cap90 ; jaus96 ) and numerical simulations with quenched lattice-QCD flynn indicate that $`f_D>f_B`$flynn , which is still maintained with two flavor sea quarks flynn ; c and in the most recent global average of lattice results wittig . This behavior of the weak decay constants of the heavy-light pseudoscalars is also found in a Dyson-Schwinger formalism applied to QCD, where general arguments says that in the heavy quark limit $`f_{ps}`$ should be $`1/\sqrt{M_{ps}}`$cr1 ($`M_{ps}`$ is the pseudoscalar mass). In order to study the mass dependence of the weak decay constant we can attempt to use a regulated form of the light-front constituent quark QCD-inspired model pauli2 ; pauli3 ; tobpauli . In this case, the systematical investigation of the mass dependence of meson observables, can be easily performed as the masses of constituent quarks acts as model parameters, which can varied while the effective quark-gluon coupling entering in the Coulomb-like interaction is flavor independent.
The short distance interaction between the constituent quarks in the squared mass operator equation of the effective light-front QCD-theory tobpauli if regulated allows a finite result for the decay constant and electromagnetic form factor. In this case, the eigenstate of the effective mass operator, i.e., the valence component of the light-front wave function, would decrease faster than $`p_{}^2`$ for large transverse momentum, which is enough to make finite the one-loop integration in the weak decay constant and form factor. One can get some information on the short-distance behavior of the valence component of the pseudoscalar meson from the electromagnetic form factor, which is experimentally well known for the pion (see Ref. pionexp ), while for the kaon data exists below 0.15 (GeV/c)<sup>2</sup> dally80 ; amen86a . In particular, when the asymptotic wave function is assumed for the soft-pion, its radius and decay constant are related by $`\sqrt{r_\pi ^2}=\sqrt{3}/(2\pi f_\pi )`$ tarr ; mill .
Our aim in this work, is to study systematically the mass dependence of the pseudoscalar weak decay constant, electromagnetic form factor and the mass splitting between the ground states of pseudoscalar and vector mesons, using a light-front QCD inspired model regulated at short distances. We choose the regulator in a separable form to simplify the formalism. The effective mass operator equation for the valence component of a constituent quark-antiquark bound system was derived in the effective one-gluon-exchange interaction approximation pauli0 and simplified in Refs. pauli2 ; pauli3 . The squared mass operator includes a Coulomb-like and a Dirac-delta hyperfine interactions acting on the spin singlet state responsible for the mass separation between pseudoscalar and the vector meson states. Here, we extend the model by introducing a regulator in a separable form in the singular part of the interaction. Then, the eigenvalue equation for the effective squared mass operator is written as:
$`M_{ps}^2\psi (x,\stackrel{}{k}_{})`$ $`=`$ $`M_0^2\psi (x,\stackrel{}{k}_{}){\displaystyle \frac{dx^{}d\stackrel{}{k}_{}^{}\theta (x^{})\theta (1x^{})}{\sqrt{x(1x)x^{}(1x^{})}}}`$ (1)
$`\times `$ $`\left({\displaystyle \frac{4m_1m_2}{3\pi ^2}}{\displaystyle \frac{\alpha }{Q^2}}\lambda g(M_0^2)g(M_0^2)\right)\psi (x^{},\stackrel{}{k}_{}^{}),`$
$`m_1`$ and $`m_2`$ are the constituent quark masses. The free squared mass operator in the meson rest frame is
$`M_0^2={\displaystyle \frac{\stackrel{}{k}_{}^{}{}_{}{}^{2}+m_1^2}{x}}+{\displaystyle \frac{\stackrel{}{k}_{}^{}{}_{}{}^{2}+m_2^2}{1x}};`$ (2)
and $`M_0^2`$ has primed momentum arguments. The form factor of the separable regulator function is $`g(M_0^2)`$. The projection of the light-front wave-function in the quark-antiquark Fock-state is given by $`\psi `$. The mean four-momentum transfer is $`Q^2`$. The strength of the Coulomb-like potential is proportional to $`\alpha `$ and the coupling constant of the regulated Dirac-delta hyperfine interaction is given by $`\lambda `$. Note that for $`g(M_0^2)1`$, the original unregulated form of the model presented in Refs. pauli2 ; pauli3 is retrieved. (In Refs. pauli2 and pauli5 were used a local Yukawa potential for the regularization of the contact interaction, here we use a separable form for simplicity.)
The dependence of the form factor in terms of $`M_0^2`$ appears to be natural in a light-front theory, in which the virtuality of an intermediate state is measured by the value of the corresponding free squared mass. In the rest frame of the quark-antiquark pair $`M_0^2=P_0^{}P^+`$ and therefore proportional to the free value of $`P_0^{}`$ \- the minus component of the free momentum ($`P_0^\pm =P_0^0\pm P_0^3`$).
Although, it is known a more developed form of the model which contains the explicit confinement conf , we will be content in solving Eq. (1) which is enough for our purpose of studying only the ground state. In practice from the solution of Eq. (1), the constituents quarks are bound and, in that sense, confined in the interior of the mesons tobpauli .
The present light-front model is a drastic approximation to a severe truncation of the Fock-space in the effective theory. In the initial truncation of QCD only one-gluon exchange was kept, which includes Fock states with up to $`q\overline{q}`$ plus one gluon, leaving out the complex nonlinear structure of QCD pauli0 ; pauli4 . The spin-dependence and momentum-dependence in the hyperfine interaction are greatly simplified to get Eq.(1) (with $`g(M^2))=1`$) and confinement is absent in the model. Therefore, the success of model should be understood as an useful guide in the investigation of mesonic properties which present a systematic behavior that depends only on few basic quantities, that are parameterized in the effective theory.
This work is organized as follows. In sec.II, the QCD-inspired model is transformed to the instant form representation and the eigenvalue equation for the squared mass operator is solved. The valence component of the meson wave function is derived. In sec. III, we give the formulae for the electromagnetic form factor and weak decay constant, derived from an effective pseudoscalar Lagrangian used to construct the spin part of the pseudoscalar meson wave-function. In sec. IV, we present and discuss the results obtained with the regularized model for the mass splittings between the pseudoscalar and vector mesons, the weak decay constants and the pion and kaon form factors. Also, in sec. IV we summarize our conclusions.
## II The QCD-inspired model in instant form representation
The effective mass operator equation for the lowest Light-Front Fock-state component of a bound system of a constituent quark and antiquark is rewritten in terms of the instant form momentum. Here we follow closely Ref. tobpauli . The general transformation from the light-cone momentum to three-momentum was derived in Ref. pauli1 . The form of Eq. (1) in the instant form momentum basis is particularly simple and convenient for the numerical solution, when the momentum carried by the effective gluon is approximated by a rotational invariant form. The momentum fraction is transformed to
$`x(k_z)={\displaystyle \frac{(E_1+k_z)}{E_1+E_2}},`$ (3)
with $`\stackrel{}{k}_{}`$ unchanged. The individual energies are $`E_i=\sqrt{m_i^2+k^2}`$ ($`i`$=1,2) and $`k|\stackrel{}{k}|`$. The Jacobian of the transformation $`(x,\stackrel{}{k}_{})`$ to $`\stackrel{}{k}`$ is:
$`dxd\stackrel{}{k}_{}={\displaystyle \frac{x(1x)}{m_rA(k)}}d\stackrel{}{k},`$ (4)
where the dimensionless phase-space function is
$`A(k)={\displaystyle \frac{1}{m_r}}{\displaystyle \frac{E_1E_2}{E_1+E_2}};`$ (5)
and the reduced mass is $`m_r=m_1m_2/(m_1+m_2)`$.
Using the momentum transformation defined above, the eigenvalue equation, (1), written in the instant form momentum basis is:
$`M_{ps}^2\phi (\stackrel{}{k})=M_0^2\phi (\stackrel{}{k}){\displaystyle 𝑑\stackrel{}{k}^{}\left(\frac{4m_s}{3\pi ^2}\frac{\alpha }{\sqrt{A(k)A(k^{})}Q^2}\frac{\lambda g(M_0^2)g(M_0^2)}{m_r\sqrt{A(k)A(k^{})}}\right)\phi (\stackrel{}{k}^{})},`$ (6)
where $`m_s=m_1+m_2`$, $`M_0=E_1+E_2`$ and $`M_0^{}`$ has the primed momentum arguments. The square momentum transfer is approximated by the rotationally invariant form $`Q^2=|\stackrel{}{k}\stackrel{}{k^{}}|^2`$. The phase-space factor is included in the factor $`1/\sqrt{A(k)A(k^{})}`$.
The valence component of the light-front wave function is
$`\psi (x,\stackrel{}{k}_{})=\sqrt{{\displaystyle \frac{A(k)}{x(1x)}}}\phi (\stackrel{}{k}).`$ (7)
The higher Fock-state components of the light-front wave function of the composite system can be expressed in terms of the lower ones, as shown by the method of the iterated resolvents pauli3 (presented in greater detail in pauli4 ) and by a quasi-potential expansion on the light-front of the Bethe-Salpeter equation sales . Therefore, it is possible to reconstruct recursively all the Fock-state components of the wave function from the valence component. In this way, the full complexity of a quantum field theory can in principle be described by a light-front effective Hamiltonian acting in the lowest Fock-state component of a composite system.
### II.1 Meson Valence Wave Function
To easily manipulate and solve Eq. (6), it is convenient to work with the operator representation:
$`\left(M_0^2+V+V^\delta \right)|\phi =M^2|\phi .`$ (8)
The matrix elements of the Coulomb-like potential $`V`$ are given by:
$`\stackrel{}{k}|V|\stackrel{}{k^{}}`$ $`=`$ $`{\displaystyle \frac{4m_s}{3\pi ^2}}{\displaystyle \frac{\alpha }{\sqrt{A(k)}Q^2\sqrt{A(k^{})}}},`$ (9)
and the for the short-range regularized singular interaction one has:
$`\stackrel{}{k}|V^\delta |\stackrel{}{k^{}}=\stackrel{}{k}|\chi {\displaystyle \frac{\lambda }{m_r}}\chi |\stackrel{}{k^{}}={\displaystyle \frac{\lambda }{m_r}}{\displaystyle \frac{g(M_0^2)}{\sqrt{A(k)}}}{\displaystyle \frac{g(M_0^2)}{\sqrt{A(k^{})}}}.`$ (10)
Just for convenience we kept the the same superscript $`\delta `$ in the short-range part of the interaction as in Ref. tobpauli , although it is regulated here. We introduce a form factor defined by $`\stackrel{}{k}|\chi =g(M_0^2)/\sqrt{A(k)}`$, which now includes the regulator.
The eigenstate of the squared mass operator, (8), is trivially given by:
$`|\phi =G^V(M_{ps}^2)|\chi ,`$ (11)
where $`G^V(M_{ps}^2)=\left[M_{ps}^2M_0^2V\right]^1`$ is the resolvent of the operator $`M_0^2+V`$. The characteristic equation for the eigenvalue of the squared mass operator is:
$`\lambda ^1={\displaystyle \frac{1}{m_r}}\chi |G^V(M_{ps}^2)|\chi ={\displaystyle \frac{1}{m_r}}{\displaystyle 𝑑\stackrel{}{k}𝑑\stackrel{}{k^{}}\frac{g(M_0^2)}{\sqrt{A(k)}}\stackrel{}{k}|G^V(M_{ps}^2)|\stackrel{}{k^{}}\frac{g(M_0^2)}{\sqrt{A(k^{})}}}.`$ (12)
We have not yet defined $`\lambda `$ in Eq. (12). To do that, we first remind the characteristic equation of the renormalized theory with the singular hyperfine interaction ($`g(M_0^2)=1`$). The bare coupling constant is obtained from the value of the pion mass and substituted in the characteristic equation which gives the mass of the pseudodscalars. Then, the characteristic equation appears in a subtracted form, in which the divergence in the momentum integration is removed tobpauli :
$`\left[{\displaystyle \frac{1}{m_r}}{\displaystyle 𝑑\stackrel{}{k}𝑑\stackrel{}{k^{}}\frac{1}{\sqrt{A(k)}}\stackrel{}{k}|G^V(M_\pi ^2)|\stackrel{}{k^{}}\frac{1}{\sqrt{A(k^{})}}}\right]_{(m_u,m_{\overline{u}})}`$
$`\left[{\displaystyle \frac{1}{m_r}}{\displaystyle 𝑑\stackrel{}{k}𝑑\stackrel{}{k^{}}\frac{1}{\sqrt{A(k)}}\stackrel{}{k}|G^V(M_{ps}^2)|\stackrel{}{k^{}}\frac{1}{\sqrt{A(k^{})}}}\right]_{(m_1,m_2)}=0,`$ (13)
where $`m_{u(\overline{u})}`$ is the mass of the light constituent quark. Observe that, the physical information contained in the pion wave function at short distances is carried to any other quark-antiquark system in Eq. (13) by the operator
$`𝒪_\pi (M_\pi ^2):=\left[{\displaystyle \frac{1}{m_r}}{\displaystyle \frac{1}{\sqrt{A(\widehat{k})}}}G^V(M_\pi ^2){\displaystyle \frac{1}{\sqrt{A(\widehat{k})}}}\right]_{(m_u,m_{\overline{u}})},`$ (14)
which has its matrix element evaluated at the origin in (13). The hat indicates the operator quality.
In the case of the present regulated model, we define for each meson a value of $`\lambda `$ assuming that the form-factor $`g(M_0^2)`$ selects the relevant momentum region of the interacting quarks, or the relevant region of virtuality of the quark-antiquark pair, within the particular meson. Thus, the matrix element of the operator $`𝒪(M_\pi ^2)`$ should be taken between states defined by the function $`g(M_0^2)`$. Introducing the operator $`𝒪_{ps}(M_{ps}^2)`$ for a general pseudoscalar meson, which has expression analogous to Eq. (14), one has:
$`𝒪_{ps}(M_{ps}^2):=\left[{\displaystyle \frac{1}{m_r}}{\displaystyle \frac{1}{\sqrt{A(\widehat{k})}}}G^V(M_{ps}^2){\displaystyle \frac{1}{\sqrt{A(\widehat{k})}}}\right]_{(m_1,m_2)}.`$ (15)
Then, using the operators defined in Eqs. (14) and (15), it is reasonable to generalize Eq. (13) to the following form:
$`{}_{ps}{}^{}g|𝒪_\pi (M_\pi ^2)𝒪_{ps}(M_{ps}^2)|g_{ps}^{}=0,`$ (16)
where $`\stackrel{}{k}|g_{ps}:=g(M_0^2)`$ with $`M_0=\sqrt{k^2+m_1^2}+\sqrt{k^2+m_2^2}`$. The strength of the short-range interaction for each pseudoscalar meson is determined by the pion mass and the regulator form-factor, according to:
$`\lambda _{ps}^1={}_{ps}{}^{}g|𝒪_\pi (M_\pi ^2)|g_{ps}^{}.`$ (17)
In the three-momentum basis Eq. (16) reads:
$`{\displaystyle }d\stackrel{}{k}{\displaystyle }d\stackrel{}{k^{}}g(M_0^2)(\left[{\displaystyle \frac{1}{m_r}}{\displaystyle \frac{1}{\sqrt{A(k)}}}\stackrel{}{k}|G^V(M_\pi ^2)|\stackrel{}{k^{}}{\displaystyle \frac{1}{\sqrt{A(k^{})}}}\right]_{(m_u,m_{\overline{u}})}`$
$`\left[{\displaystyle \frac{1}{m_r}}{\displaystyle \frac{1}{\sqrt{A(k)}}}\stackrel{}{k}|G^V(M_{ps}^2)|\stackrel{}{k^{}}{\displaystyle \frac{1}{\sqrt{A(k^{})}}}\right]_{(m_1,m_2)})g(M_0^2)=0,`$ (18)
where $`M_0^2`$ and $`M_0^2`$ are computed for the quarks with masses $`m_1`$ and $`m_2`$.
In our calculation procedure, the resolvent is numerically obtained from
$`G^V(M_{ps}^2)=G_0(M_{ps}^2)+G_0(M_{ps}^2)T^V(M_{ps}^2)G_0(M_{ps}^2),`$ (19)
where the T-matrix is the solution of the Lippman-Schwinger equation:
$`T^V(M_{ps}^2)=V+VG_0(M_{ps}^2)T^V(M_{ps}^2),`$ (20)
where the free resolvent is $`G_0(M_{ps}^2)=\left[M_{ps}^2M_0^2\right]^1`$. The detailed expressions can be found in Ref. tobpauli .
The valence component of the light-front wave function of the meson is the solution of Eq. (1) given by Eq. (11), which we write explicitly as, using Eq. (19):
$`\psi (x,\stackrel{}{k}_{})`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{x(1x)}}}{\displaystyle \frac{G_{ps}}{M_{ps}^2M_0^2}}[g(M_0^2)+{\displaystyle }d\stackrel{}{k^{}}\sqrt{{\displaystyle \frac{A(k)}{A(k^{})}}}\stackrel{}{k}|T^V(M_{ps}^2|\stackrel{}{k^{}}g(M_0^2)],`$ (21)
where the overall normalization factor of the $`q\overline{q}`$ Fock-component of the meson wave-function is $`G_{ps}`$. The three-momemtum is expressed in terms of the light-cone momentum with the transformation (3). The first term in Eq. (21) dominates for large momentum transfers if $`g(M_0^2)=1`$ (corresponding to the asymptotic form), differently from this situation when $`g(M_0^2)1`$ the two terms can compete even in the asymptotic region.
### II.2 Constituent Quark Masses and Mass Splittings
Within the present model, the low-lying vector mesons are weakly bound systems of constituent quarks while the pseudo-scalars are strongly bound. Therefore in this model, the masses of the constituent quarks are obtained directly from the vector meson masses, as suisso :
$`m_u`$ $`=`$ $`{\displaystyle \frac{1}{2}}M_\rho =384MeV,`$
$`m_s`$ $`=`$ $`M_K^{}{\displaystyle \frac{1}{2}}M_\rho =508MeV,`$
$`m_c`$ $`=`$ $`M_D^{}{\displaystyle \frac{1}{2}}M_\rho =1623MeV,`$
$`m_b`$ $`=`$ $`M_B^{}{\displaystyle \frac{1}{2}}M_\rho =4941MeV,`$ (22)
where the masses of the vector mesons are 768 MeV, 892 MeV, 2007 MeV and 5325 MeV for the $`\rho `$, $`K^{}`$, $`D^{}`$ and $`B^{}`$, respectively pdg . The constituent masses for the up and down quarks are considered equal (we disregarded the small few MeV difference in the current up and down masses pdg ). Using the value of the light-constituent quark mass of 384 MeV, and assuming that the effect of chiral symmetry breaking is about the same for each flavors one gets an estimate of the current quark mass as $`m_Q^{curr}=m_Qm_u`$ suisso . The current quark masses are estimated as $`m_s^{curr}=124`$ MeV, $`m_c^{curr}=1239`$ MeV and $`m_b^{curr}=4557`$ MeV consistent with Ref. pdg .
In our model, the binding energy of the constituent quarks in the pseudoscalar mesons, is interpreted as the $`{}_{}{}^{1}S_{0}^{}`$ \- $`{}_{}{}^{3}S_{1}^{}`$ meson mass splitting, and thus a quantity directly related to data. The binding energy is simply $`B_{ps}=M_vM_{ps}`$ defined to be positive. The experimental values for the ground state quantities show evidence for a strong correlation of $`B_{ps}`$ and $`M_{ps}`$ qualitatively reproduced by the renormalized model with singular interaction tobpauli . We will see in sec. IV, that Eq. (16) also provides a reasonable description of the mass splitting.
## III Electromagnetic Form Factor and Weak Decay Constant
To obtain the electromagnetic form pseudoscalar decay constants, we follow the suggestion of Refs. mill ; mill94 . To construct such observables, one describe the coupling of the pseudoscalar meson field to the quark field, by an effective Lagrangian density with a pseudo-scalar coupling between the quark ($`q_1(\stackrel{}{x})`$ and $`q_2(\stackrel{}{x})`$) and meson $`\left(\mathrm{\Phi }_{ps}(\stackrel{}{x})\right)`$ fields:
$`_{eff}(\stackrel{}{x})=i\mathrm{\Gamma }_{ps}\mathrm{\Phi }_{ps}(\stackrel{}{x})\overline{q}_1(\stackrel{}{x})\gamma ^5q_2(\stackrel{}{x})+h.c.,`$ (23)
where $`\mathrm{\Gamma }_{ps}`$ is a constant vertex. After the integration in the minus momentum component of the momentum integration of the one-loop amplitudes that define the electromagnetic form factor and weak decay constant, the asymptotic form of the wave function is substituted by the valence component of the model wave function. The integration in the minus momentum component eliminates the relative time between the quarks in the intermediate states sales .
### III.1 Form Factor of Pseudoscalar Mesons
The pseudoscalar meson electromagnetic form-factor is obtained from the impulse approximation of the plus component of the current $`(j^+=j^0+j^3)`$ in the Breit-frame with momentum transfer $`q^+=0`$ and $`q^2=\stackrel{}{q}^2`$ satisfying the Drell-Yan condition. The general structure of the $`q\overline{q}`$ bound state forming the meson comes from the pseudoscalar coupling (23). We use such spin structure in the computation of the photo-absorption amplitude in the impulse approximation (represented by a Feynman triangle diagram), which is written as:
$`(p_\pi ^\mu +p_\pi ^\mu )F_{ps}(q^2)`$ $`=`$ $`i\mathrm{\Gamma }_{ps}^2e_1N_c{\displaystyle }{\displaystyle \frac{d^4k}{(2\pi )^4}}tr[{\displaystyle \frac{\text{/}k+m_2}{k^2m_2^2+i\epsilon }}\gamma ^5{\displaystyle \frac{\text{/}k\text{/}p^{}+m_1}{(kp^{})^2m_1^2+i\epsilon }}`$ (24)
$`\times `$ $`\gamma ^\mu {\displaystyle \frac{\text{/}k\text{/}p+m_1}{(kp)^2m_1^2+i\epsilon }}\gamma ^5]+[12],`$
where $`F_{ps}(q^2)`$ is the electromagnetic form-factor and $`e_i`$ is the quark charge. The meson momentum in the initial and final states are defined by $`p^0=p^0`$ and $`\stackrel{}{p^{}}_{}=\stackrel{}{p}_{}=\stackrel{}{\frac{q_{}}{2}}`$. $`N_c=3`$ is the number of colors.
The choice of the plus-component of the current is adequate in the case of the pseudoscalars, because after the integration over $`k^{}=k^0k^3`$ the suppression of the pair diagram is maximal for this component in the frame where $`q^+=0`$ and just the wave-function components contribute to the form-factor mill ; mill94 ; pach ; pach1 . In our model only the valence component is considered. Although, we compute the integration in the minus momentum component assuming a constant vertex, one can identify in the expression how the valence component of the wave-function correspondent to the non-constant vertex of Eq.(21) should be introduced. As the details of this derivation is by now standard, we present directly the final result:
$`F_{ps}(q^2)`$ $`=`$ $`e_1{\displaystyle \frac{N_c}{(2\pi )^3}}{\displaystyle _0^1}{\displaystyle \frac{dx}{1x}}{\displaystyle }d^2k_{}[2m_1m_22m_1^2+k_{1on}^{}p^++`$ (25)
$`k^+(m_1m_2)^2k^+\stackrel{}{q}_{}^2)]\psi _{ps}(x,\stackrel{}{K}_{})\psi _{ps}(x,\stackrel{}{K_{}^{}})+[12],`$
where the momentum fraction is $`x=k^+/p^+`$ and $`k_{1on}^{}=(\stackrel{}{k}_{}^2+m_1^2)/k^+`$. The quark transverse momentum in the meson rest frame is given by:
$`\stackrel{}{K}_{}=\stackrel{}{k}_{}+x{\displaystyle \frac{\stackrel{}{q}_{}}{2}}`$ (26)
and $`\stackrel{}{K_{}^{}}=\stackrel{}{K}_{}x\stackrel{}{q}_{}`$. The expression for form factor gives the standard Drell-Yan formula once the bound-state wave-function of the constant vertex model (asymptotic form) is recognized
$`\psi ^{\mathrm{}}(x,\stackrel{}{K}_{})={\displaystyle \frac{G_{ps}}{\sqrt{x(1x}\left(m_\pi ^2M_0^2\right)}},`$ (27)
which is the first term in Eq. (21) for $`g(M_0^2)=1`$. The second term in (21) comes from the Coulomb-like interaction. The other factors in Eq.(25) compose the Melosh rotations of the individual spin wave function of the quarks.
The size of the meson is closely related to the square-root mean square charge radius which is calculated as
$$\sqrt{r_{ps}^2}=\left[6\frac{d}{dq^2}F_{ps}(q^2)|_{q^2=0}\right]^{\frac{1}{2}}.$$
In sec. IV we adjust the regularization parameter (see Eq.(34) by fitting the pion charge radius, $`\sqrt{r_\pi ^2}`$, which has the experimental value of $`0.67\pm 0.02`$ fm amen . The charge radius from Eq.(25) in the soft-pion limit ($`m_\pi =0`$) using the asymptotic wave-function (27), with $`\mathrm{\Gamma }_\pi =\sqrt{2}m_{u(d)}/f_\pi `$ from the Goldberger-Treiman izuber relation at the quark level, results in the well known expression $`\sqrt{r_\pi ^2}=\sqrt{3}/(2\pi f_\pi )`$ from Ref. tarr . In this case, also the form factor (25) for $`q^2=0`$, reduces to expression for $`f_\pi `$ as given in mill . We observe that our model, for $`\alpha =0`$, $`g(M_0^2)=1`$ and $`m_\pi =0`$ recovers the soft-pion result, i.e, $`\sqrt{r_\pi ^2}=`$ 0.58 fm.
### III.2 Weak Decay Constant
The leptonic weak decay constant of the pseudoscalar meson ($`f_{ps}`$) is a physical quantity that depends directly on the probability to find the quark-antiquark Fock-state component in the meson wave-function pauli0 . Also, $`f_{ps}`$ depends on the short-range physics carried by the wave function when the quark and antiquark are close.
The meson weak decay constant is calculated from the matrix element of the axial current $`A^\mu (0)`$ between the vacuum state $`|0`$, and the meson state $`|p_{ps}`$ with four momentum $`p`$ pdg :
$`0A^\mu (0)p_{ps}=ı\sqrt{2}f_{ps}p^\mu ,`$ (28)
where $`A^\mu (\stackrel{}{x})=ı\overline{q}(\stackrel{}{x})\gamma ^\mu \gamma ^5q(\stackrel{}{x})`$.
The matrix element of the plus component of the axial current is derived from the pseudoscalar Lagrangian, (23), and it is expressed by a one-loop diagram, which is given by:
$`ı\sqrt{2}M_{ps}f_{ps}=N_c\mathrm{\Gamma }_{ps}{\displaystyle \frac{d^4k}{(2\pi )^4}\frac{Tr\left[\gamma ^+\gamma ^5(\text{/}k\text{/}p+m_2)\gamma ^5(\text{/}k+m_1)\right]}{((kp)^2m_2^2+i\epsilon )(k^2m_1^2+i\epsilon )}},`$ (29)
the plus component is used to eliminate the instantaneous terms of the Dirac propagator.
By integration of Eq. (29) over $`k^{}`$, one obtains the expression of $`f_{ps}`$ suitable for the introduction of the meson light front wave-function. So, performing the Dirac algebra and separating the poles in the $`k^{}`$-plane and integrating, one gets:
$`f_{ps}=`$ $``$ $`{\displaystyle \frac{\sqrt{2}}{8\pi ^3}}N_c{\displaystyle _0^1}{\displaystyle \frac{dx}{x(1x)}}\left((1x)m_1+xm_2\right){\displaystyle d^2k_{}\frac{\mathrm{\Gamma }_{ps}}{M_{ps}^2M_0^2}},`$ (30)
where quark 1 has momentum fraction $`x`$. This expression is written in the meson rest-frame and we have used the momentum fraction $`x=k^+/p^+`$. The free square mass is defined in Eq. (2). Note that this expression has a log-type divergence in the transverse momentum integration which was discussed in Ref. luiz and parameterized in terms of $`f_\pi `$.
One can write Eq.(30) in terms of the valence component of the pseudoscalar meson wave function from Eq.(21), as:
$`f_{ps}=`$ $`{\displaystyle \frac{\sqrt{2}}{8\pi ^3}}N_c{\displaystyle _0^1}{\displaystyle \frac{dx}{\sqrt{x(1x)}}}\left((1x)m_1+xm_2\right){\displaystyle d^2k_{}\psi (x,\stackrel{}{k}_{})},`$ (31)
with $`x`$ being the momentum fraction of quark 1. If one chooses $`g(M_0^2)`$ decaying as $`M_0^\eta `$ for any $`\eta >0`$ is enough to make $`f_{ps}`$ finite.
In the particular case of $`g(M_0^2)=1`$ the meson wave function, Eq.(21), for large transverse momentum behaves as the asymptotic wave-function, which decreases slowly as $`p_{}^2`$ implying in logarithmic divergences in the transverse momentum integrations of the weak decay constant and form factor. In Refs. khoplov and luiz , the log-type divergent factors in the pseudoscalar decay constants were parameterized in terms of $`f_\pi `$ and the sum of the constituent quark masses, which in the QCD-inspired model tobpauli could be identified with the ground state vector meson mass. Therefore, one has:
$`f_{ps}=\text{const.}{\displaystyle _0^1}𝑑x\left((1x)m_2+xm_1\right),`$ (32)
and $`f_{ps}m_1+m_2`$ khoplov . In our model $`m_1+m_2`$ is the vector meson mass, and then as suggested by Ref. luiz , the decay constant scales as:
$`{\displaystyle \frac{f_{ps}}{f_\pi }}={\displaystyle \frac{M_v}{M_\rho }},`$ (33)
which approximates the existing data up to the kaon and $`D`$ mesons but is not supported by relativistic constituent quark potential models and Lattice-QCD calculations as we have discussed in the introduction. The use of the separable regulator in the model brings this consistency. We will return to this discussion in the next section.
## IV Discussion of the Numerical Results and Conclusion
The present QCD-inspired model for the effective squared mass operator acting only on the valence component of the meson wave function, Eq. 1 has the canonical number of parameters (the quark masses and $`\alpha `$) plus two, when we choose the regulator form factor as
$`g^{(a)}(M_0)={\displaystyle \frac{1}{\beta ^{(a)}+M_0^2}}\text{or}g^{(b)}(M_0)={\displaystyle \frac{1}{M_0^2}}+\left({\displaystyle \frac{\beta ^{(b)}}{M_0^2}}\right)^2.`$ (34)
The form factor (a) for equal mass quarks is the familiar Yukawa form in coordinate space. The other choice just contains the first two terms of a Taylor expansion of $`g^{(a)}`$ for large $`M_0^2`$. In the limit of $`\beta ^{(a)}\mathrm{}`$ the model with form factor $`g^{(a)}`$ reduces to the renormalized model of Ref. tobpauli , while by construction the form factor $`g^{(b)}`$ is qualitatively different as it does not allow to match that model. Nonetheless, for finite $`\beta `$’s as we will see from our numerical calculations both form factors produce practically the same results.
The parameters $`\beta `$ and the strength of the separable interaction $`\lambda _{ps}`$, Eq. (17), are adjusted to reproduce the experimental pion charge radius $`0.67\pm 0.02`$ fm amen and mass, $`M_\pi `$= 140 MeV, for each form factor $`g^{(a)}`$ and $`g^{(b)}`$ independently. We use the light quark constituent mass of 384 MeV (see Eq. (22)) and $`f_\pi `$ results 110 MeV for both separable interactions (see Table I) . The somewhat higher value for $`f_\pi `$ compared to the experimental value of $`92.4\pm .07\pm 0.25`$ pdg , is a common shortcoming of light-front models of the pion when the valence wave function is normalized to one mill94 ; pach1 . The valence component appears to have a probability of about 70% (see also mill94 ; pach1 ) which brings the model results for $`f_\pi `$ to 92 MeV. In the case of $`\alpha =0.5`$, the resulting parameters for the form factors are $`\beta ^{(a)}=`$-(634.5 MeV)<sup>2</sup> and $`\beta ^{(b)}=`$(1171 MeV)<sup>2</sup>.
In figure 1, we show the results for the $`{}_{}{}^{3}S_{1}^{}`$-$`{}_{}{}^{1}S_{0}^{}`$ meson mass splitting, the binding energy $`B_{ps}`$, as a function of the pseudoscalar meson mass. We choose $`\alpha =0.5`$ and the form factor regulator $`g^{(a)}`$ (The differences between the masses obtained with the two form of regulators are less than 1 MeV). In the figure, we see the dependence of the mass splittings of $`q\overline{Q}`$ mesons with the pseudoscalar mass, obtained by the the variation of $`m_Q`$, while $`m_q`$ is fixed at the values of 384 MeV (solid line), 508 MeV (dashed line) and 1623 MeV (dotted line). In this way, we simulate the families of mesons with an up or down, a strange and charm quarks and a distinct one which has mass $`m_Q`$. First, we compare the results of the present regularized model with the previous results of the renormalized model tobpauli found for $`m_q=`$ 384 MeV and $`\alpha =0.5`$. In this case, the regularization increases $`B_{ps}`$ as seen in the figure. The regularization procedure naturally softens at short-distances the attractive part of interaction, which should be compensated by an effective increase of the strength of the separable interaction to keep the pion still strongly bound at its physical mass. The increase of the strength is reflected in the increase of binding, as seen in the figure. Still the trend of the experimental values of the mass splitting for $`\rho \pi `$, $`K^{}K^\pm `$, $`D^0D^0`$ and $`B^{}B^\pm `$ pdg is found.
The results for the mass splitting for mesons containing at least one strange meson (dashed line in figure 1) exhibit the same qualitative behavior found for mesons with an up or down quarks, i.e., the mass splitting decreases with the rise of the mass of the heavy quark. This should be the case since the masses of the constituents up-down and strange are very much similar, with an expected increase in the mass splitting when the up-down quark is exchanged with one strange quark which is heavier. By rising the mass of one of the constituents for mesons with charm (dotted line of figure 1) the splitting increases, because the quarks become spatially closer and the binding is expected to rise as in nonrelativistic potential models. Also, as expected, the saturation of the binding energy appears for large masses.
In figure 2, we show the weak decay constant as a function of the intensity parameter $`\alpha `$ of the Coulomb-like interaction for different mesons. The calculations are performed with the regulator form-factor $`g^{(a)}(M_0)=(b+M_0^2)^1`$ with the parameter $`b`$ adjusted for each given $`\alpha `$ between 0.1 and 0.5 in order to reproduce $`f_\pi =110`$ MeV. The kaon weak decay constant varies less than one MeV in this interval keeping the value 126 MeV (see Table I). We show in the figure only results for $`D^+`$ (solid line), $`D_s^+`$ (dashed line), $`B^+`$ (solid line with dots) and $`B_c^+`$ (dashed line with dots). The decay constants rise with $`\alpha `$, as the $`q\overline{Q}`$ systems become more bound and compact due to increase in the Coulomb-like interaction. The effect is particularly dramatic for the heavier mesons $`B^+`$ and $`B_c^+`$ as could be anticipated thinking within a nonrelativistic potential model, where the probability to found the quarks at the origin should increase when the attractive force is strengthened.
The pseudoscalar meson weak decay constant as a function of the vector meson mass is shown in figure 3 as suggested by Eq. (33). The calculations were performed with the regulator form-factor (a) and $`\alpha =0.5`$. In the figure, we see dependence of $`f_{ps}`$ for $`q\overline{Q}`$ mesons with the vector meson mass, obtained through the variation of $`m_Q`$, while $`m_q`$ is fixed at the values of 384 MeV (solid line), 508 MeV (dashed line) and 1623 MeV (dotted line). The naive model of Eq. (33), which presents a linear increase of $`f_{ps}`$ with $`M_v`$, although it gives some qualitative insight into the data fails to describe the saturation and decrease of the results of the regulated model, which as expected has a $`f_{ps}`$ decreasing with the mass of the meson. The data for $`D_s^+`$ is indeed below the linear curve and consistent with the dashed curve calculated with the regulated model for $`s\overline{Q}`$ pseudoscalars. There are several experimental values for $`f_{D^+}`$ obtained by different collaborations as quoted in Table I. In figures 3 and 4 we just indicate the experimental result from cleo .
The values of $`f_{ps}`$ for mesons with one charm quark (dotted line in figure 3) increase with $`M_v`$, as the system becomes more compact up to the point that $`f_{ps}`$ saturates for $`M_v>>m_c`$=1623 MeV (the probability density at the origin does not change anymore) while the expected $`1/\sqrt{M_Q}`$ dependence dominates for large values of $`M_v`$.
In figure 4, the weak decay constant as a function of the pseudoscalar meson mass obtained in our regulated model with form factor (a) and $`\alpha =0.5`$ is compared to the recent global average of lattice-QCD results wittig . The short-dashed line gives a least-square fit to the experimental values of $`f_\pi `$ and $`f_K`$ together with the lattice estimates for $`D^+`$ and $`B^+`$ flynn given by $`f_{ps}^2=(0.0065+0.014M_{ps})/(1+0.055M_{ps}+0.15M_{ps}^2)`$ GeV<sup>2</sup> as given in Ref. cr , where the $`1/\sqrt{M_{ps}}`$ behavior for large masses is built in. The results for $`u\overline{Q}`$ pseudoscalars are qualitatively agreement with that fit. Our calculations for $`u\overline{Q}`$ and $`s\overline{Q}`$ pseudoscalar mesons are in a good consistency with the global lattice averages of the weak decay constants, as seen by comparing the solid line with the full circles for $`u\overline{Q}`$ mesons and the dashed line with the full stars for $`s\overline{Q}`$ mesons.
To close our study of the present regulated model in figures 5, 6 and 7 we show results for the pion and kaon electromagnetic form factors using $`\alpha =0.5`$ with the model regulated with form factor (a). The pion mean square radius is reasonable fitted and as well as the form factor up to about 4 \[GeV/c\]<sup>2</sup> as shown in figure 5. The experimental values for kaon form factor dally80 ; amen86a present large errors and do not allow a definite conclusion as seen in figure 6. For completeness, we present the kaon form factor calculation up to 10 \[GeV/c\]<sup>2</sup>. We also compare with the calculations with the form factor (b), and we do not observe a strong model dependence below 4 \[GeV/c\]<sup>2</sup>.
In Table I, we present the results for the pseudoscalar weak decay constants $`f_{ps}`$ for $`\pi `$, $`K`$, $`D^+`$, $`D_s^+`$, $`B^+`$, $`B_s^0`$ and $`B_c^+`$ compared to global estimates of lattice-QCD results and experimental data. The consistence with lattice results indicates that the regularized model is able to parameterize the QCD-physics at short-distances in the ground state of the pseudoscalar mesons quite reasonably. The pseudoscalar masses are underestimated for the heavy mesons, as seen already in figure 1, although the saturation behavior of the mass splitting that the data indicates is verified by the calculation. This problem can be overcome by the introduction of confinement in the model conf ; bjpgraca .
In summary, we have shown that the suggested separable form to regulate the singular interaction in the square mass operator provides a reasonable description of the mass splitting between $`{}_{}{}^{3}S_{1}^{}`$ and $`{}_{}{}^{1}S_{0}^{}`$ meson ground states, the weak decay constants as found in a recent global average of lattice results wittig and the pion form factor up to 4 \[GeV/c\]<sup>2</sup>. The main point here is that the model can describe the mass dependence of the weak decay constant, revealing that the physics in this observable is dominated by the mass of the meson itself, through the quark masses and binding. The effective squared mass operator acting on the valence component of the light-front meson wave function is again tested and proved to reasonably parameterize the dynamics of the constituents at short distances. The present version of the model does not have explicit confining interaction, therefore it is not able to account for the spectra. A more sophisticated version of the model that includes confinement was shown to describe the meson spectrum conf and the pion form-factor in the space and time-like regions plb04 , can also be used in the future in a regularized form to allow the calculation of the pseudoscalar decay constants.
Acknowledgments: We thank CNPq (Conselho Nacional de Desenvolvimento Científico e Tecnológico) and FAPESP (Fundação de Amparo a Pesquisa do Estado de São Paulo) of Brasil for financial support.
TABLES AND FIGURES |
warning/0507/hep-ph0507244.html | ar5iv | text | # Comparison of NNLO DIS scheme splitting functions with results from exact gluon kinematics at small 𝑥
## 1 $`P_{qg}^{DIS}`$ and $`C_{Lg}^{DIS}`$ with exact gluon kinematics
After solving the BFKL equation for $`f(N,k^2,Q_0^2)`$, the latter two factors in equation (1) combine to give a regularised unintegrated gluon distribution, and $`h_i`$ can naively be interpreted as the coefficient function linking the gluon density with the structure function. However, in the case of $`F_2`$, the impact factor diverges as $`k^2/Q^20`$. One can understand this given that at $`𝒪(\alpha _S^0)`$, $`F_2`$ is proportional to the quark singlet parton distribution with no gluon contribution. One does not expect to describe this nonperturbative dependence using perturbation theory, and thus the impact factor diverges. One must instead consider solving the evolution equation for $`F_2`$, via the quantity:
$$\frac{_2(Q^2,N)}{\mathrm{ln}Q^2}=\alpha _S_0^{\mathrm{}}\frac{dk^2}{k^2}h_2(k^2/Q^2)f(N,k^2,Q_0^2)g_B(N,Q_0^2),$$
(2)
which serves to define the impact factor $`h_2`$. In a general factorisation scheme, one loses the simple interpretation of $`h_2`$ as the coefficient function relating the gluon distribution to the structure function. Instead it represents a mixture of the coefficient $`C_{2g}`$ and the anomalous dimension $`\gamma _{qg}`$. If one chooses to work in the DIS scheme , where $`F_2`$ is given by the naive parton model expression to all orders, then $`h_2`$ can be interpreted directly as the quark gluon anomalous dimension $`\gamma _{qg}^{DIS}`$. From equation (1), the longitudinal impact factor $`h_L`$ is identified with the coefficient function $`C_{Lg}^{DIS}`$ and does not diverge due to the fact the the longitudinal structure function vanishes at $`𝒪(\alpha _S^0)`$. It is convenient to perform a second Mellin transformation on the factorisation formulae to unravel the convolution in $`k`$-space:
$$\stackrel{~}{}_L(\gamma ,N)=_0^{\mathrm{}}𝑑k^2(k^2)^{1\gamma }_L(k^2,N)=\stackrel{~}{h}_L(\gamma )\stackrel{~}{G}(\gamma ,N),$$
(3)
and similarly for equation (2).
Diagrams contributing to the impact factor are shown in figure 1.
One may introduce a Sudakov decomposition for the $`4`$-momenta $`k,r`$:
$`l`$ $`=\alpha q^{}+\beta p+l_{};`$ (4)
$`k`$ $`={\displaystyle \frac{q^2}{s}}q^{}+x_gp+k_{},`$ (5)
where p is the proton $`4`$-momentum (light-like if one ignores the proton mass), $`q^{}=q+xp`$ a second light-like vector involving the Bjorken variable $`x=Q^2/(2pq)`$, and $`s=(p+q)^2`$. The on-shell requirements for the intermediate quarks ($`(l+k)^2=(ql)^2=0`$) then lead to the relation:
$$x_g=x\left[\frac{\widehat{Q}^2\widehat{k}_{}^2l_{}^{}{}_{}{}^{2}}{\widehat{Q}^2}\right],$$
(6)
with $`\widehat{A}\alpha (1\alpha )A^2`$, $`Q^2=q^2`$, and $`l_{}^{}=l_{}+(1\alpha )k_{}`$. At LL order, the momentum fraction $`x_g`$ of the incident proton carried by the gluon is undetermined, as equation (6) implies the difference $`\mathrm{log}x_g\mathrm{log}x`$ is finite as $`x0`$. By imposing correct kinematics for the gluon, one includes in the impact factor significant higher order information.
The resulting $`N`$ dependent factors $`h_2(\gamma ,N)`$ and $`h_L(\gamma ,N)`$ can be found in , and from them one may derive estimates of $`\gamma _{qg}^{DIS}`$ and $`C_{Lg}^{DIS}`$ at fixed order in $`\alpha _S`$. One first expands the relevant impact factor as a Taylor series in $`\gamma `$ with coefficients $`h_i^{(n)}`$. In solving the BFKL equation, $`\gamma `$ is identified as the anomalous dimension $`\gamma (N)`$ given at NLL accuracy in (any further accuracy would require knowledge of the NNLL BFKL kernel). One thus has:
$$C_{Lg}^{DIS(e)}(\alpha _S,N)=\underset{n=0}{\overset{\mathrm{}}{}}h_L^{(n)}(N)[\gamma (\alpha _S,N)]^{(n)},$$
(7)
as the exact kinematics result for the coefficient function up to NLL order, and similarly for $`\gamma _{qg}^{DIS(e)}`$ in terms of $`h_2`$. The BFKL anomalous dimension has a perturbative expansion:
$$\gamma (\alpha _S,N)=\underset{n=1}{\overset{\mathrm{}}{}}\alpha _S^nf_n(N),$$
(8)
so that equation (7) is a power series in $`\alpha _S`$, beginning at $`𝒪(\alpha _S)`$. The corresponding $`x`$-space expressions are given in Appendix A. One may compare order by order with the complete results for $`C_{Lg}`$ and $`\gamma _{qg}`$. The corresponding $`\overline{\text{MS}}`$ functions have been computed up to $`𝒪(\alpha _S^3)`$ . However, for a direct comparison with the exact kinematics results one needs the corresponding results in the DIS scheme rather than the conventional $`\overline{\text{MS}}`$ scheme.
## 2 DIS scheme splitting and coefficient functions
The NNLO singlet and non-singlet splitting functions have recently been computed in the $`\overline{\text{MS}}`$ scheme , along with the $`𝒪(\alpha _S^3)`$ coefficient functions for neutral boson exchange . They are extremely lengthy - for example, the typeset NNNLO gluon coefficient $`C_{2g}`$ is around twelve pages long . The NNLO splitting functions are shorter, but like the coefficients are made more complicated by the nature of the algebraic functions involved. All of the NNLO coefficient and splitting functions involve combinations of harmonic sums in $`N`$-space, which after inverse Mellin transformation yield harmonic polylogarithms in $`x`$-space (up to weight five at this order). These are non-standard functions and thus must be generated numerically . The combination of length and numerical complexity makes it is infeasible that the complete results can be immediately used in phenomenological applications. Instead one may parameterise the results in $`x`$-space in terms of simple algebraic functions, with a precision that far exceeds that due to higher order corrections. The Mellin transforms of the parameterisations then give suitably accurate $`N`$-space representations. Parameterisations of the $`\overline{\text{MS}}`$ functions are given in . From these we have derived corresponding representations of the DIS scheme quantities, accurate to within a percent apart from near the zeros. Our results are presented in Appendix B.
In transforming between the $`\overline{\text{MS}}`$ and DIS schemes, we follow the argument presented in . The DIS scheme is characterised by the singlet structure function $`F_2`$ having the same form as the naive parton model to all orders :
$$F_{2s}(x,Q^2)=\mathrm{\Sigma }^{\text{DIS}}C_{2q}^{\overline{\text{MS}}}\mathrm{\Sigma }^{\overline{\text{MS}}}+C_{2g}g^{\overline{\text{MS}}},$$
(9)
where $`\mathrm{\Sigma }=_i(q_i+\overline{q}_i)`$ is the singlet quark density, and the factorisation and renormalisation scales have been chosen as $`Q^2`$. The factorisation scheme independence of $`F_2`$ then imposes a transformation between the DIS and $`\overline{\text{MS}}`$ scheme partons. There remains an ambiguity in the definition of the DIS gluon. However, the momentum sum rule fixes:
$$_0^1𝑑xx[\mathrm{\Sigma }(x,Q^2)+g(x,Q^2)]=1$$
(10)
in both schemes. In Mellin space, this becomes:
$$\mathrm{\Sigma }^{\text{DIS}}(N)+g^{\text{DIS}}(N)=\mathrm{\Sigma }^{\overline{\text{MS}}}(N)+g^{\overline{\text{MS}}}(N)$$
(11)
for $`N=1`$<sup>1</sup><sup>1</sup>1This corresponds to our choice of Mellin variable and that of . The alternative definition $`\stackrel{~}{f}(N)=_0^1x^{(N1)}f(x)`$ is also in common use, and in that case the second moment is constrained.. One may remove the ambiguity by extending equation (11) to all $`N`$, and one obtains:
$$𝒒^{\text{DIS}}\left(\begin{array}{c}\mathrm{\Sigma }^{\text{DIS}}\\ g^{\text{DIS}}\end{array}\right)=\left(\begin{array}{cc}C_{2q}^{\overline{\text{MS}}}& C_{2g}^{\overline{\text{MS}}}\\ C_{2q}^{\overline{\text{MS}}}& C_{2g}^{\overline{\text{MS}}}\end{array}\right)\left(\begin{array}{c}\mathrm{\Sigma }^{\overline{\text{MS}}}\\ g^{\overline{\text{MS}}}\end{array}\right)𝒁𝒒^{\overline{\text{MS}}}.$$
(12)
To obtain the splitting functions, one differentiates equation (12) with respect to $`Q^2`$ and rearranges yielding:
$$𝑷^{\text{DIS}}=\left(𝒁𝑷^{\overline{\text{MS}}}+\beta (\alpha _S)\frac{d𝒁}{d\alpha _S}\right)𝒁^1.$$
(13)
where $`𝑷=\left(\begin{array}{cc}P_{qq}& P_{qg}\\ P_{gq}& P_{gg}\end{array}\right)`$. Substituting the perturbative expansions of the $`\overline{\text{MS}}`$ scheme coefficient and splitting functions<sup>2</sup><sup>2</sup>2Conventionally, $`P_{ij}^{(n)}`$ is the coefficient of $`a^{n+1}`$, where $`a=\alpha _S/(4\pi )`$; $`C_{\{2,L\}i}^{(n)}`$ is the coefficient of $`a^n`$., along with the QCD $`\beta `$ function<sup>3</sup><sup>3</sup>3Here $`\beta _n`$ is the coefficient of $`a^{n+2}`$., one can derive the DIS scheme results order by order in $`\alpha _S`$. The explicit transformations at $`𝒪(\alpha _S^3)`$ are:
$`P_{qq}^{(2)\text{DIS}}`$ $`=P_{qq}^{(2)\overline{\text{MS}}}+C_{2q}^{(2)\overline{\text{MS}}}P_{qg}^{(0)\overline{\text{MS}}}+C_{2g}^{(2)\overline{\text{MS}}}P_{gq}^{(0)\overline{\text{MS}}}+C_{2g}^{(1)\overline{\text{MS}}}P_{gq}^{(1)\overline{\text{MS}}}+C_{2q}^{(1)\overline{\text{MS}}}P_{qg}^{(1)\overline{\text{MS}}}`$
$`C_{2g}^{(1)\overline{\text{MS}}}P_{gq}^{(0)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}+C_{2g}^{(1)\overline{\text{MS}}}P_{gg}^{(0)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}C_{2g}^{(1)\overline{\text{MS}}}P_{qq}^{(0)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}`$
$`+C_{2g}^{(1)\overline{\text{MS}}}P_{qg}^{(0)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}+\beta _0C_{2q}^{(1)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}\beta _0C_{2g}^{(1)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}2\beta _0C_{2q}^{(2)\overline{\text{MS}}}`$
$`\beta _1C_{2q}^{(1)\overline{\text{MS}}};`$ (14)
$`P_{gq}^{(2)\text{DIS}}`$ $`=P_{gq}^{(2)\overline{\text{MS}}}C_{2q}^{(2)\overline{\text{MS}}}P_{qq}^{(0)\overline{\text{MS}}}C_{2g}^{(2)\overline{\text{MS}}}P_{gq}^{(0)\overline{\text{MS}}}C_{2q}^{(2)\overline{\text{MS}}}P_{gq}^{(0)\overline{\text{MS}}}+C_{2q}^{(2)\overline{\text{MS}}}P_{gg}^{(0)\overline{\text{MS}}}`$
$`C_{2q}^{(1)\overline{\text{MS}}}P_{qq}^{(1)\overline{\text{MS}}}C_{2g}^{(1)\overline{\text{MS}}}P_{gq}^{(1)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}P_{gq}^{(1)\overline{\text{MS}}}+C_{2q}^{(1)\overline{\text{MS}}}P_{gg}^{(1)\overline{\text{MS}}}`$
$`+C_{2q}^{(1)\overline{\text{MS}}}P_{qq}^{(0)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}+C_{2q}^{(1)\overline{\text{MS}}}P_{gq}^{(0)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}P_{gg}^{(0)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}`$
$`C_{2q}^{(1)\overline{\text{MS}}}P_{qg}^{(1)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}\beta _0C_{2q}^{(1)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}+\beta _0C_{2g}^{(1)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}+2\beta _0C_{2q}^{(2)\overline{\text{MS}}}`$
$`+\beta _1C_{2q}^{(1)\overline{\text{MS}}};`$ (15)
$`P_{gg}^{(2)\text{DIS}}`$ $`=P_{gg}^{(2)\overline{\text{MS}}}C_{2q}^{(2)\overline{\text{MS}}}P_{qg}^{(0)\overline{\text{MS}}}C_{2g}^{(2)\overline{\text{MS}}}P_{gq}^{(0)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}P_{qg}^{(1)\overline{\text{MS}}}C_{2g}^{(1)\overline{\text{MS}}}P_{gq}^{(1)\overline{\text{MS}}}`$
$`+C_{2g}^{(1)\overline{\text{MS}}}P_{qq}^{(0)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}+C_{2g}^{(1)\overline{\text{MS}}}P_{gq}^{(0)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}C_{2g}^{(1)\overline{\text{MS}}}P_{gg}^{(0)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}`$
$`C_{2g}^{(1)\overline{\text{MS}}}P_{qg}^{(0)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}+\beta _0C_{2g}^{(1)\overline{\text{MS}}}C_{2g}^{(1)\overline{\text{MS}}}\beta _0C_{2q}^{(1)\overline{\text{MS}}}C_{2g}^{(1)\overline{\text{MS}}}+2\beta _0C_{2g}^{(2)\overline{\text{MS}}}`$
$`+\beta _1C_{2g}^{(1)\overline{\text{MS}}};`$ (16)
$`P_{qg}^{(2)\text{DIS}}`$ $`=P_{qg}^{(2)\overline{\text{MS}}}+C_{2q}^{(2)\overline{\text{MS}}}P_{qg}^{(0)\overline{\text{MS}}}+C_{2g}^{(2)\overline{\text{MS}}}P_{gg}^{(0)\overline{\text{MS}}}C_{2g}^{(2)\overline{\text{MS}}}P_{qq}^{(0)\overline{\text{MS}}}+C_{2g}^{(2)\overline{\text{MS}}}P_{qg}^{(0)\overline{\text{MS}}}`$
$`+C_{2q}^{(1)\overline{\text{MS}}}P_{qg}^{(1)\overline{\text{MS}}}+C_{2g}^{(1)\overline{\text{MS}}}P_{gg}^{(1)\overline{\text{MS}}}C_{2g}^{(1)\overline{\text{MS}}}P_{qq}^{(1)\overline{\text{MS}}}+C_{2g}^{(1)\overline{\text{MS}}}P_{qg}^{(1)\overline{\text{MS}}}`$
$`C_{2g}^{(1)\overline{\text{MS}}}P_{gq}^{(0)\overline{\text{MS}}}C_{2g}^{(1)\overline{\text{MS}}}+C_{2g}^{(1)\overline{\text{MS}}}P_{gg}^{(0)\overline{\text{MS}}}C_{2g}^{(1)\overline{\text{MS}}}C_{2g}^{(1)\overline{\text{MS}}}P_{qq}^{(0)\overline{\text{MS}}}C_{2g}^{(1)\overline{\text{MS}}}`$
$`+C_{2g}^{(1)\overline{\text{MS}}}P_{qg}^{(0)\overline{\text{MS}}}C_{2g}^{(1)\overline{\text{MS}}}+\beta _0C_{2q}^{(1)\overline{\text{MS}}}C_{2g}^{(1)\overline{\text{MS}}}\beta _0C_{2g}^{(1)\overline{\text{MS}}}C_{2g}^{(1)\overline{\text{MS}}}2\beta _0C_{2g}^{(2)\overline{\text{MS}}}`$
$`\beta _1C_{2g}^{(1)\overline{\text{MS}}}.`$ (17)
Non-singlet quark combinations transform according to:
$$q_{ns}^{\text{DIS}}=C_{2ns}^{\overline{\text{MS}}}q_{ns}^{\overline{\text{MS}}},$$
(18)
which has the form of equation (12) but with a trivial transformation matrix. Hence one obtains for the non-singlet splitting functions relevant to neutral and charged current scattering :
$$P_{ns}^{+,(2)\text{DIS}}=P_{ns}^{+,(2)\overline{\text{MS}}}+\beta _0C_{2q}^{(1)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}2\beta _0C_{2ns}^{+,(2)\overline{\text{MS}}}\beta _1C_{2q}^{(1)\overline{\text{MS}}};$$
(19)
The pure singlet splitting function is given by:
$$P_{ps}^{(2)\text{DIS}}=P_{qq}^{(2)\text{DIS}}P_{ns}^{+(2)\text{DIS}}.$$
(20)
The $`F_2`$ coefficient functions are simply defined to all orders in the DIS scheme. For the longitudinal coefficients, one considers:
$$F_L=\left(\begin{array}{ccc}C_{Lq}^{\overline{\text{MS}}}& C_{Lg}^{\overline{\text{MS}}}& C_{Lns}^{\overline{\text{MS}}}\end{array}\right)\left(\begin{array}{c}\mathrm{\Sigma }^{\overline{\text{MS}}}\\ g^{\overline{\text{MS}}}\\ q_{ns}^{\overline{\text{MS}}}\end{array}\right).$$
(21)
Using the transformation equations (12) and (18), one finds:
$$\left(\begin{array}{ccc}C_{Lq}^{\text{DIS}}& C_{Lg}^{\text{DIS}}& C_{Lns}^{\text{DIS}}\end{array}\right)=\left(\begin{array}{ccc}C_{Lq}^{\overline{\text{MS}}}& C_{Lns}^{\overline{\text{MS}}}& C_{Lg}^{\overline{\text{MS}}}\end{array}\right)\left(\begin{array}{cc}𝒁& 0\\ 0& C_{2ns}^{+\overline{\text{MS}}}\end{array}\right)^1.$$
(22)
Explicit results at $`𝒪(\alpha _S^3)`$ after substituting the expansions of the $`\overline{\text{MS}}`$ scheme coefficient functions are:
$`C_{Lg}^{(3)\text{DIS}}`$ $`=C_{Lg}^{(3)\overline{\text{MS}}}+C_{Lg}^{(1)\overline{\text{MS}}}C_{2g}^{(2)\overline{\text{MS}}}C_{Lq}^{(1)\overline{\text{MS}}}C_{2g}^{(2)\overline{\text{MS}}}+C_{Lg}^{(2)\overline{\text{MS}}}C_{2g}^{(1)\overline{\text{MS}}}C_{Lq}^{(2)\overline{\text{MS}}}C_{2g}^{(1)\overline{\text{MS}}}`$
$`+C_{Lg}^{(1)\overline{\text{MS}}}C_{2g}^{(1)\overline{\text{MS}}}C_{2g}^{(1)\overline{\text{MS}}}C_{Lq}^{(1)\overline{\text{MS}}}C_{2g}^{(1)\overline{\text{MS}}}C_{2g}^{(1)\overline{\text{MS}}}C_{Lg}^{(1)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}C_{2g}^{(1)\overline{\text{MS}}}`$
$`+C_{Lq}^{(1)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}C_{2g}^{(1)\overline{\text{MS}}};`$ (23)
$`C_{Lq}^{(3)\text{DIS}}`$ $`=C_{Lq}^{(3)\overline{\text{MS}}}+C_{Lg}^{(1)\overline{\text{MS}}}C_{2q}^{(2)\overline{\text{MS}}}C_{Lq}^{(1)\overline{\text{MS}}}C_{2q}^{(2)\overline{\text{MS}}}+C_{Lg}^{(2)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}C_{Lq}^{(2)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}`$
$`+C_{Lg}^{(1)\overline{\text{MS}}}C_{2g}^{(1)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}C_{Lq}^{(1)\overline{\text{MS}}}C_{2g}^{(1)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}C_{Lg}^{(1)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}`$
$`+C_{Lq}^{(1)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}};`$ (24)
$$C_{Lns}^{+(3)\text{DIS}}=C_{Lns}^{(3)\overline{\text{MS}}}C_{Lns}^{(2)\overline{\text{MS}}}C_{2ns}^{(1)\overline{\text{MS}}}C_{Lns}^{(1)\overline{\text{MS}}}C_{2ns}^{(2)\overline{\text{MS}}}+C_{Lns}^{(1)\overline{\text{MS}}}C_{2ns}^{(1)\overline{\text{MS}}}C_{2ns}^{(1)\overline{\text{MS}}}.$$
(25)
Then the pure singlet coefficient is given by:
$$C_{Lps}^{(3)\text{DIS}}=C_{Lqq}^{(3)\text{DIS}}C_{Lns}^{+(3)\text{DIS}}.$$
(26)
The transformation terms were evaluated in $`N`$-space, and divergent high and low $`N`$ limits were then extracted. Up to $`𝒪(1/N)`$, one has a choice in how to extract the high $`N`$ piece. We have chosen this in such a way as to lead to simple plus distributions and logarithms of $`(1x)`$ in the $`x`$-space functions. The remaining finite functions as $`N0,\mathrm{}`$ were parameterised in $`x`$-space by evaluating the inverse Mellin transform numerically. Finally the transformation terms were added to the existing $`\overline{\text{MS}}`$ parameterisations. Thus the plus distribution and small-$`x`$ divergent terms are exact up to truncation of the coefficients, as also are the parts of the $`\mathrm{log}(1x)`$ terms not involving $`(1x)\mathrm{log}(1x)`$.
The coefficients of $`\delta (1x)`$ in $`P_{ns}^+`$, $`P_{gq}`$ and $`P_{gg}`$ have been modified, and $`\delta (1x)`$ contributions added to $`P_{qg}`$ that should in principle be absent. This, following refs. , is to increase the $`N`$-space accuracy, such that the parameterised functions satisfy the momentum sum rules:
$`\gamma _{qg}^{(2)DIS}(N)+\gamma _{gg}^{(2)DIS}(N)`$ $`=0;`$ (27)
$`\gamma _{gq}^{(2)DIS}(N)+\gamma _{qq}^{(2)DIS}(N)`$ $`=0,`$ (28)
for $`N=1`$ <sup>4</sup><sup>4</sup>4This also implies that the $`n_f`$ independent part of $`P_{gg}^{(2)DIS}`$ should vanish, given that $`P_{qg}`$ has no term at $`𝒪(n_f^0)`$.. One can also introduce such terms into the longitudinal coefficient functions, by fitting to numerical values of the low integer moments. We choose not to introduce these, however, given the size of these effects (no more than a few parts permille) do not exceed the uncertainty of the parameterisations. We have checked all of our expressions against known numerical moments .
Particularly noteworthy is the singularity structure of the DIS scheme functions as $`x1`$. One sees that the singlet quark splitting functions contain plus distributions up to $`𝒟_2`$ <sup>5</sup><sup>5</sup>5See Appendix B for the definition of these functions., or $`\mathrm{log}^3(N)`$ in Mellin space. However, $`P_{gq}^{(2)\text{DIS}}`$ contains more singular terms up to $`𝒟_4\mathrm{log}^5(N)`$. One can understand this by considering what happens in the $`\overline{\text{MS}}`$ scheme. There $`\mathrm{log}(N)`$ terms arise in the coefficients $`C_{2q}`$ and $`C_{2ns}`$ as a result of soft gluon emission from the quark probed by the virtual photon. As $`x1`$, there is insufficient phase space for the emission of real gluons, and thus an incomplete cancellation between singularities arising from virtual and real emission. The leading logarithms in $`N`$ exponentiate , and the sub-leading logarithms can also be resummed . Combining the known resummation and fixed order results allows knowledge of the four leading towers of high $`N`$ logarithms in $`C_{2q}`$ to all orders in $`\alpha _S`$ <sup>6</sup><sup>6</sup>6This analysis has very recently been extended to include even higher order logarithmic corrections to DIS and Drell-Yan type processes .. In the DIS scheme there are no such logarithms in the coefficients, as $`C_{2q}`$ is defined trivially to all orders. Instead the soft gluon resummation effects enter the splitting functions. The leading $`\mathrm{log}(N)`$ terms in $`C_{2ns}^{\overline{\text{MS}}}`$ are produced by exponentiating those in $`\gamma _{ns}^{\text{DIS}}`$. This follows from the $`N`$-space evolution equation for the non-singlet quark density:
$$\frac{\stackrel{~}{q}_{ns}^{\text{DIS}}}{\mathrm{log}Q^2}=\gamma _{ns}^{\text{DIS}}(N)\stackrel{~}{q}_{ns}^{\text{DIS}},$$
(29)
which is easily solved to give:
$`\stackrel{~}{q}_{ns}^{\text{DIS}}(Q^2)`$ $`=\stackrel{~}{q}_{ns}^{\text{DIS}}(Q_0^2)\mathrm{exp}\left[{\displaystyle _{\alpha _S(Q_0^2)}^{\alpha _S(Q^2)}}\gamma _{ns}^{\text{DIS}}{\displaystyle \frac{d\alpha _S}{\beta (\alpha _S)}}\right]`$
$`=q_{ns}^{\text{DIS}}(Q_0^2)\mathrm{exp}\left[{\displaystyle _{\alpha _S(Q_0^2)}^{\alpha _S(Q^2)}}{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}{\displaystyle _{\alpha _S(Q_0^2)}^{\alpha _S(Q^2)}}c_{n,m}\beta _m\gamma _{ns}^{(n)\text{DIS}}\alpha _S^{m+n1}𝑑\alpha _S\right],`$ (30)
where the $`c_{n,m}`$ are coefficients obtained after substituting in the perturbative expansions of the $`\beta `$ function and anomalous dimension. Performing the integration in the exponent gives:
$$q_{ns}^{\text{DIS}}(Q^2)=q_{ns}^{\text{DIS}}(Q_0^2)\left[\frac{\alpha _S(Q^2)}{\alpha _S(Q_0^2)}\right]^{\frac{\gamma _{ns}^{(0)DIS}}{\beta _0}}\mathrm{exp}\left\{\left(\frac{\gamma _{ns}^{(1)DIS}}{\beta _0}+\frac{\beta _1\gamma _{ns}^{(0)DIS}}{\beta _0^2}\right)\left[\frac{\alpha _S(Q^2)}{4\pi }\frac{\alpha _S(Q_0^2)}{4\pi }\right]+\mathrm{}\right\},$$
(31)
where the ellipsis denotes terms giving rise to sub-leading logarithms. Given that $`\gamma _{ns}^{(0)DIS}\mathrm{log}N`$ and $`\gamma _{ns}^{(1)DIS}\mathrm{log}^2(N)`$ as $`N\mathrm{}`$, the leading logarithms in the exponent come from the term in $`\gamma _{ns}^{(1)}`$. The form of the non-singlet structure function in the DIS and $`\overline{\text{MS}}`$ schemes is:
$$F_{2ns}=q_{ns}^{\text{DIS}}C_{2ns}^{\overline{\text{MS}}}q_{ns}^{\overline{\text{MS}}}.$$
(32)
Thus from equation (30), ones sees that the leading powers of $`\mathrm{log}(N)`$ in the $`\overline{\text{MS}}`$ scheme non-singlet coefficient function are generated by exponentiation of those in the DIS scheme NLO anomalous dimension $`\gamma _{ns}^{(1)\text{DIS}}`$ (the LO anomalous dimension is independent of the factorisation scheme, and thus the prefactor in equation (31) is also found in the $`\overline{\text{MS}}`$ scheme). The next-to-leading $`\mathrm{log}(N)`$ terms in $`C_{ns}^{\overline{\text{MS}}}`$ are not so straightforward, but are determined by the exponentiation of a mixture of $`\gamma _{ns}^{(1)DIS}`$ and $`\gamma _{ns}^{(2)DIS}`$, and so on for the other sub-leading logarithms. A similar argument relates the leading $`\mathrm{log}(N)`$ terms in $`C_{2q}^{\overline{\text{MS}}}`$ with those in $`\gamma _{qq}^{(1)\text{DIS}}`$.
This explains the absence of more singular logarithms $`\mathrm{log}^4(N)`$ in the DIS scheme $`\gamma _{qq}^{(2)}`$, as the highest power of $`\mathrm{log}(N)`$ is limited by the fact that it cannot exceed the power obtained by exponentiation of the leading log term in $`\gamma _{qq}^{(1)\text{DIS}}`$. Looking at equation (14), the transformation terms in $`P_{qq}^{(2)\overline{\text{MS}}}P_{qq}^{(2)\text{DIS}}`$ involve the combination $`\beta _0C_{2q}^{(1)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}2\beta _0C_{2q}^{(2)\overline{\text{MS}}}`$. Thus $`\mathrm{log}^4(N)`$ terms in $`\stackrel{~}{C}_{2q}^{(2)\overline{\text{MS}}}`$ are cancelled by the combination $`[\stackrel{~}{C}_{2q}^{(1)\overline{\text{MS}}}]^2/2!`$ due to the exponential structure of the leading logs in the coefficient function.
The $`𝒟_4`$ term in $`P_{gq}^{(2)\text{DIS}}`$ corresponds to a next to leading high $`x`$ divergence in $`C_{2q}^{\overline{\text{MS}}}`$ ($`\alpha _S^3\mathrm{log}^5(N)`$ in Mellin space), arising from the terms in equation (15):
$`\left[P_{gq}^{(2)DIS}\right]_{𝒟_4}`$ $`=[C_{2q}^{(2)\overline{\text{MS}}}P_{qq}^{(0)\overline{\text{MS}}}+C_{2q}^{(0)\overline{\text{MS}}}P_{gg}^{(0)\overline{\text{MS}}}+C_{2q}^{(1)\overline{\text{MS}}}P_{qq}^{(0)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}`$
$`C_{2q}^{(1)\overline{\text{MS}}}P_{gg}^{(0)\overline{\text{MS}}}C_{2q}^{(1)\overline{\text{MS}}}]_{𝒟_4}.`$ (33)
The small $`x`$ limit of $`P_{qg}^{(2)\text{DIS}}`$ will be discussed in section 3 of this paper. Looking at the other splitting functions, one may verify the LL relations :
$`P_{gq}={\displaystyle \frac{C_F}{C_A}}P_{gg},`$ $`P_{qq}={\displaystyle \frac{C_F}{C_A}}\left[P_{qg}{\displaystyle \frac{\alpha _S}{2\pi }}T_R{\displaystyle \frac{2}{3}}\right],`$ (34)
where $`C_A=3`$, $`C_F=4/3`$ are the QCD Casimir invariants and $`T_R=1/2`$. These relations are also true at LL order in the $`\overline{\text{MS}}`$ scheme.
The non-singlet and singlet splitting functions are plotted in figures 2 and 3 respectively. The singlet functions have been multiplied by $`x`$ to alleviate the small $`x`$ divergence.
Aside from the differences at high $`x`$ discussed above, one sees that the DIS scheme functions are more divergent at small $`x`$. This is analogous to the high $`x`$ behaviour - in changing schemes one transfers divergences from the quark singlet coefficient to the splitting functions. Note also the qualitatively different structures at intermediate $`x`$ in the two schemes. Each of the singlet splitting functions develops an extra turning point in the DIS scheme.
The NNLO $`P_{qg}`$ develops a negative dip in the DIS scheme at intermediate $`x`$, before increasing again as $`x0`$. Together with the large negative dip at high $`x`$ this gives a negative result at intermediate $`x`$ when convolved with a model gluon distribution which is more singular than the splitting function. We return to this feature in section 3. In fact, the qualitative structure of the DIS scheme splitting function can be reproduced from the truncated transformation:
$$P_{qg}^{(2)\text{DIS}}P_{qg}^{(2)\overline{\text{MS}}}+C_{2g}^{(2)\overline{\text{MS}}}P_{gg}^{(0)\overline{\text{MS}}}2\beta _0C_{2g}^{(2)\overline{\text{MS}}}.$$
(35)
The longitudinal quark and gluon coefficient functions are shown in figures 4 and 5.
The two gluon coefficients are extremely similar. The pure singlet and gluon coefficients share the same small $`x`$ limit, as the LL coefficients are the same in both schemes. There is, however, an extra turning point in the DIS scheme pure singlet function at higher $`x`$. Also of note is the negativity of the non-singlet coefficient at small $`x`$, a property also shared by the NLO result such that the complete non-singlet coefficient is negative at small $`x`$. However, it is not divergent as $`x0`$ so that convolution with a suitable non-singlet test function does not give a negative non-singlet structure function (see figure 6).
We now compare the DIS scheme quark-gluon anomalous dimension and longitudinal gluon coefficient with the corresponding results derived from exact gluon kinematics.
## 3 Comparison of exact kinematics with NLO and NNLO results
At $`𝒪(\alpha _S)`$, the exact kinematics results correspond exactly with the complete results as at this order, all the relevant diagrams are included in the impact factor calculation. The imposition of exact kinematics then supplements the $`x`$ dependence that is missing when evaluating these diagrams in the LL limit. At higher orders, one can compare the complete $`N`$-space functions with the estimates obtained from the modified impact factors. Expanding in $`N`$ one finds:
$`\gamma _{qg}^{(1)DIS(e)}`$ $`={\displaystyle \frac{34.67n_f}{N}}(102.9n_f+.2140n_f^2)+(172.1n_f+.4391n_f^2)N`$
$`(246.0n_f+.6598n_f^2)N^2+𝒪(N^3);`$ (36)
$`\gamma _{qg}^{(2)DIS(e)}`$ $`={\displaystyle \frac{441.47n_f}{N^2}}{\displaystyle \frac{2635n_f+49.53n_f^2}{N}}+(7555n_f+118.0n_f^2+.01682n_f^3)`$
$`(15089n_f+204.0n_f^2+.06958n_f^3)N+(25166n_f+312.1n_f^2`$
$`+.1462n_f^3)N^2+𝒪(N^3),`$ (37)
The complete NLO and NNLO results give:
$`\gamma _{qg}^{(1)DIS}`$ $`={\displaystyle \frac{34.67n_f}{N}}+(109.3n_f+.8889n_f^2)+(233.6n_f5.072n_f^2)N`$
$`+(374.6n_f+11.70n_f^2)N^2+𝒪(N^3);`$ (38)
$`\gamma _{qg}^{(2)DIS}`$ $`={\displaystyle \frac{441.47n_f}{N^2}}+{\displaystyle \frac{3165n_f+30.19n_f^2}{N}}+(12945n_f399.5n_f^2+.5926n_f^3)`$
$`+(34493n_f+1589n_f^26.121n_f^3)N`$
$`+(73141n_f4389n_f^2+27.53n_f^3)N^2+𝒪(N^3).`$ (39)
The leading logarithms in $`x`$ (most divergent terms as $`N0`$) are correctly predicted from the resummation, and one sees that the next to leading terms in $`\gamma _{qg}^{(2)DIS}`$ are well estimated by the exact kinematics expression (within 2% at NLO and 7% at NNLO, for $`n_f=4`$). Accuracy falls off for higher order terms in $`N`$, although these are not associated with small $`x`$ divergence. The $`x`$-space functions are shown in figure 7.
The exact kinematics results qualitatively reproduce the structures of the complete results, even at high $`x`$. They clearly do much better than the LL terms at approximating the splitting functions. Note that the small-$`x`$ behaviour does not set in until rather low $`x`$, as can be seen by the splitting function only turning positive for $`x2\times 10^3`$ at NNLO (for $`n_f=4`$). The qualitative trend is that at higher order in $`\alpha _S`$, the splitting function turns positive at lower $`x`$. We have confirmed, for example, that the NNNLO exact kinematics splitting function does not turn positive until $`x10^4`$. A good estimate for these values is obtained by approximating the exact kinematics splitting functions by their asymptotic limits as $`x0`$:
$`xP_{qg}^{(2)DIS(e)}`$ $`441.47n_f\mathrm{log}{\displaystyle \frac{1}{x}}(2635n_f+49.53n_f^2)+\mathrm{};`$ (40)
$`xP_{qg}^{(3)DIS(e)}`$ $`11671n_f{\displaystyle \frac{1}{2!}}\mathrm{log}^2{\displaystyle \frac{1}{x}}+(78095n_f1410n_f^2)\mathrm{log}{\displaystyle \frac{1}{x}}+(248414n_f+6924n_f^2+8.265n_f^3)+\mathrm{},`$ (41)
where the ellipses represent terms vanishing in this limit. Setting $`xP_{qg}^{(n)DIS(e)}=0`$ gives the approximate value $`x=x_0`$ at which the LL terms begin to dominate over the sub-leading logarithms. For $`n_f=4`$, one finds $`x_03\times 10^3,1\times 10^4`$ at NNLO, NNNLO respectively. The lower value of $`x_0`$ with increasing order of $`\alpha _S`$ implies that the leading small-$`x`$ resummation effects become less important phenomenologically at higher orders, as sub-leading logarithms dominate until very small $`x`$.
From equations (38) and (39), we note that in the non-leading logarithmic terms, contributions involving higher powers of $`n_f`$ are estimated poorly - including being of the wrong sign. This is expected given that higher powers of $`n_f`$ in the perturbative contribution to the structure functions may arise from diagrams such as those shown in figure 8, with fermion bubbles in the vertical rungs of the gluon ladder and in the quark loop at the top of the diagram. The former are included in the NLL BFKL anomalous dimension<sup>7</sup><sup>7</sup>7Fermion bubbles in the bottom vertical rung of the ladder are not in the NLL anomalous dimension, but contribute to the scale of the coupling. See ., but the latter are missing in the exact kinematics calculation due the LL nature of the impact factor. However, one can see that the higher order $`n_f`$ terms do not constitute a very significant contribution relative to those at $`𝒪(n_f)`$.
Similar expansions for the longitudinal coefficient are:
$`\stackrel{~}{C}_{Lg}^{(2)DIS(e)}`$ $`={\displaystyle \frac{5.333n_f}{N}}+(18.22n_f+.03292n_f^2)+(62.52n_f+.1427n_f^2)N`$
$`(86.88n_f+.2551n_f^2)N^2+𝒪(N^3);`$ (42)
$`\stackrel{~}{C}_{Lg}^{(3)DIS(e)}`$ $`={\displaystyle \frac{409.5n_f}{N^2}}+{\displaystyle \frac{1246n_f+1.727n_f^2}{N}}+(2127n_f+40.14n_f^2+.01561n_f^3)`$
$`(3436n_f+68.95n_f^2+.01892n_f^3)N+(5345n_f+89.73n_f^2`$
$`+.03326n_f^3)N^2+𝒪(N^3).`$ (43)
The complete results give:
$`\stackrel{~}{C}_{Lg}^{(2)DIS}`$ $`={\displaystyle \frac{5.333n_f}{N}}+(6.229n_f+.8889n_f^2)+(80.69n_f4.850n_f^2)N`$
$`+(133.8n_f+10.04n_f^2)N^2+𝒪(N^3);`$ (44)
$`\stackrel{~}{C}_{Lg}^{(3)DIS}`$ $`={\displaystyle \frac{409.5n_f}{N^2}}+{\displaystyle \frac{2076n_f+102.4n_f^2}{N}}+(4730n_f340.8n_f^2.1139fl_{11}^gn_f^2`$
$`+.5926n_f^3)+(9211n_f+854.1n_f^2.4340fl_{11}^gn_f^25.973n_f^3)N`$
$`+(20054n_f2251n_f^2+.08264fl_{11}^gn_f^2+25.74n_f^3)N^2+𝒪(N^3),`$ (45)
where $`fl_{11}^g=<e>^2/<e^2>`$, taking averages over the active quark charges. The estimation of NLL terms is not as good as for $`\gamma _{qg}`$, even for the $`𝒪(n_f)`$ contribution. Again taking $`n_f=4`$, the NLL term in the NNLO coefficient is estimated to within $`35\%`$. Nevertheless, the exact kinematics results are in good qualitative agreement with the complete results. The term in $`fl_{11}^g`$ will not be estimated by the exact kinematics calculation due to missing diagrams of the type shown in figure 9. Also, this term is not associated with a small $`x`$ divergence at $`𝒪(\alpha _S^3)`$. Higher order terms in $`n_f`$ and $`fl_{11}^g`$ are not very significant contributions.
The $`x`$-space functions are shown in figure 10.
Again the exact kinematics results have the same qualitative behaviour as the complete results at both small and large $`x`$, whereas the LL approximations are comparatively poor.
The greater accuracy in $`\gamma _{qg}`$ can in part be attributed to the derivative of $`F_2`$ in $`\mathrm{log}(Q^2)`$. In Mellin space, this amounts to multiplication of the sum of the transverse and longitudinal impact factors by $`\gamma (N)`$, which suppresses the differences noted above by $`\alpha _S`$.
The $`x`$-space functions will ultimately be convolved with parton distribution functions. Hence it is necessary to check the behaviour of the $`x`$-space exact kinematics expressions when convolved with a suitable gluon distribution. Following , we convolve with the model gluon distributions:
$`xg(x)`$ $`=x^{0.3}(1x)^4;`$ (46)
$`xg(x)`$ $`=x^{0.5}(1x)^4,`$ (47)
where the former corresponds to a high $`Q^2`$ scale ($`30\text{GeV}^2`$), and the latter reflects the fact that the gluon can be valence-like (or even negative at low $`x`$) at low $`Q^21\text{GeV}^2`$ . One expects resummation of small $`x`$ terms to be more important at low $`Q^2`$, due to the higher value of $`\alpha _S`$. The results for $`P_{qg}^{(1)}`$ and $`P_{qg}^{(2)}`$ are shown in figures 11 and 12.
Results for $`C_{Lg}^{(2)DIS}g`$ and $`C_{Lg}^{(3)DIS}g`$ are shown in figure 13 and 14.
For both the splitting and coefficient functions, the exact kinematics results qualitatively approximate the structure of the complete results. Comparing them with the LL terms at small $`x`$, after convolution with the gluon, one sees that they are much closer to the complete results. The exception is $`C_{Lg}^{(2)}`$, where the LL terms convolved with the gluon distribution are closer to the complete result at low $`x`$, aided by the fact that at this order the coefficient function has no next-to-leading small $`x`$ divergence. At NNLO, where NLL terms are present, the exact kinematics results perform better at small $`x`$. The exact kinematics and complete results generally agree more at the lower momentum scale. This is due to the less singular gluon distribution at low $`Q^2`$, and the small $`x`$ part of the coefficient playing a more dominant role. However, at higher $`Q^2`$, the effect of a more singular gluon distribution will be compensated in part by a lower value of $`\alpha _S`$, and resummation becomes less important.
The NNLO exact kinematics splitting function gives a negative result when convolved with the more singular gluon, turning positive only at very low $`x10^7`$. This can be attributed to the large negative dip in the exact kinematics function (see figure 7) at intermediate $`x`$. Given that the gluon is more singular than the splitting function, the low $`x`$ limit of the convolution is dominated by both high and low $`x`$ information in the splitting function. To see how this works, consider the model splitting function:
$$P=\frac{A}{x}+B\delta (1x),$$
(48)
where the first and second terms give the dominant behaviour at small and high $`x`$ respectively. Consider convolving this with the following “gluon”:
$$xf=x^\alpha \theta (x_0x),$$
(49)
which is singular or valence-like at low $`x`$ depending on whether $`\alpha <0`$ or $`\alpha >0`$, and vanishes at high $`x`$. One then has:
$$\frac{(xf)}{\mathrm{log}(Q^2)}=xPf=\left[B\frac{A}{\alpha }\right]x^\alpha +\frac{Ax_0^\alpha }{\alpha }.$$
(50)
If $`\alpha >0`$, the small $`x`$ term in the splitting function dominates the convolution. If on the other hand $`\alpha <0`$, the bracketed term in equation (50) gives the leading small $`x`$ behaviour, which is a mixture of both the small and large $`x`$ terms of the splitting function. This also accounts for the lack of a common small $`x`$ limit in the left-hand plots of figures 11, 12, 13, 14, as each of the three splitting functions has a different high $`x`$ behaviour. Note that the complete NNLO $`P_{qg}`$ also has a negative dip at intermediate $`x`$. This leads to some negative behaviour after the convolution, but not for $`x0.05`$ in figure 12.
## 4 Conclusions
We have shown that the imposition of exact gluon kinematics in the LL virtual photon-gluon impact factor gives a good approximation to the NLL parts of the NLO and NNLO splitting and coefficient functions $`P_{qg}^{DIS}`$ and $`C_{Lg}^{DIS}`$ up to $`𝒪(\alpha _S^3)`$. The qualitative behaviour is also good over the whole $`x`$ range. We see this both by examining poles in $`N`$-space and also convolving the $`x`$-space functions with suitable model gluon distributions. Hence in the absence of the full NLL impact factor <sup>8</sup><sup>8</sup>8Calculation is, however, in progress ., we have confidence that the exact kinematics results can be used for an accurate NLL analysis of the proton structure functions.
It may also be possible to impose exact kinematics in the impact factors for heavy quark production . In this case, however, one needs to define a suitable factorisation scheme in order to interpret the impact factors in terms of splitting and coefficient functions.
The NNLO DIS scheme splitting and longitudinal coefficient functions (excluding the coefficients for charged current scattering) have been parameterised and presented here. There are significant qualitative differences with the $`\overline{\text{MS}}`$ scheme results, particularly in the appearance of divergent high $`x`$ terms from soft gluon resummations in the splitting functions. These functions are available on request and can easily be applied for parton analyses in the DIS scheme at NNLO.
## 5 Acknowledgements
CDW would like to thank Jeppe Andersen for useful discussions, and is grateful to PPARC for a research studentship. RST thanks the Royal Society for the award of a University Research Fellowship.
## Appendix A Appendix: The exact kinematics splitting and coefficient functions
The $`N`$-space splitting and coefficient functions derived from the exact kinematic impact factors involve the function $`\psi (N)=\mathrm{\Gamma }(N)/\mathrm{\Gamma }^{}(N)`$ and its derivatives, where $`\mathrm{\Gamma }(N)`$ is the Euler gamma function. The $`\psi `$ functions can be expressed as analytically continued harmonic sums , which one can then inverse Mellin transform to $`x`$-space. For brevity we define:
$`L_0=\mathrm{log}(x),`$ $`L_1=\mathrm{log}(1x).`$ (51)
Then the results are:
$`C_{Lg}^{(2)DIS(e)}(x)`$ $`=(240x^2+272x)L_01196/3x^29216/3x^1+496x)n_f`$
$`+\left(32/27x(1x)L_056/27x^28/27+64/27x^1\right)n_f^2;`$ (52)
$`C_{Lg}^{(3)DIS(e)}(x)`$ $`=([3468x2700x^2]L_0^2+[96x^1L_1^2+(2312x512x^11800x^2)L_12820`$
$`11320x^2+13432x+(192\text{Li}_2(x)+32\pi ^22176/3)x^1]L_0+[900x^2+256x^1`$
$`1156x]L_1^2+[1412+1800x+(32\pi ^2384192\text{Li}_2(x))x^14x^2]L_11032`$
$`40892/9x^2+6952x+[64\text{Li}_2(x)192\text{Li}_3(1x)384\text{Li}_3(x)12388/9`$
$`+384\zeta (3)32/3\pi ^2]x^1)n_f`$
$`+([272/9x80/3x^2]L_0^2+[(160/9x^264/27x^1+544/27x)L_1992/27x^2`$
$`+64/9x16]L_0+[272/27x+32/27x^1+80/9x^2]L_1^2+[160/9x64/9x^1`$
$`304/27+16/27x^2]L_1+1168/27+5024/81x^22800/27x+[128/81`$
$`+32/81\pi ^264/27\text{Li}_2(x)]x^1)n_f^2`$
$`+16/243(x(1x)[3L_0^2+2L_1L_0+14L_0L_1^2]L_0[x^2+12x]L_1`$
$`27x^2+10xx^1)n_f^3;`$ (53)
$`P_{qg}^{(1)DIS(e)}(x)`$ $`=\left([92+120x^2136x]L_0+1048/3x^2+44+104/3x^1384x\right)n_f`$
$`+\left([16/27x(x1)^2+8/27]L_0+8/27+16/9x(x1)\right)n_f^2;`$ (54)
$`P_{qg}^{(2)DIS(e)}(x)`$ $`=([1350x^21734x+1587]L_0^2+[96x^1L_1^2+(1156x+900x^2+1058560x^1)L_1`$
$`890+10160x^211340x+(192\text{Li}_2(x)+32\pi ^22272/3)x^1]L_0+[578x529`$
$`450x^2+280x^1]L_1^2+[706900x+(32\pi ^2288192\text{Li}_2(x))x^12x^2]L_1`$
$`+5992+86146/9x^212248x+[136/3\pi ^2+384\zeta (3)272\text{Li}_2(x)29842/9384\text{Li}_3(x)`$
$`192\text{Li}_3(1x)]x^1)n_f`$
$`+([40/3x^2+92/9136/9x]L_0^2+[(184/2764/27x^1272/27x+80/9x^2)L_1`$
$`+1696/27x^21448/271184/27x]L_0+[92/27+32/27x^1+136/27x40/9x^2]L_1^2`$
$`+[8/27x^2+152/2732/9x^180/9x]L_1+800/27472/81x^2+88/3x`$
$`+[32/81\pi ^24304/8164/27\text{Li}_2(x)]x^1)n_f^2`$
$`+4/729(2x(x1)[3L_0^2+2L_0L_1L_1^2]+3L_0^2+2L_1+2[322x+24x^2]L_0`$
$`L_1^2+2[12xx^2]L_1+50x^256x+82x^1)n_f^3,`$ (55)
where $`\text{Li}_n(x)`$ is the $`n^{\text{th}}`$ polylogarithm function.
## Appendix B Appendix: The DIS scheme splitting and coefficient functions
Here we present parameterisations of the coefficient and splitting functions at NNLO in the DIS scheme . For completeness, all singlet and non-singlet splitting functions are given. The longitudinal coefficient functions are given only for neutral current structure functions, as the $`\overline{\text{MS}}`$ scheme coefficient functions for charged current scattering have yet to be published. First we define:
$`𝒟_n=\left[{\displaystyle \frac{\mathrm{log}^n(1x)}{1x}}\right]_+,`$ $`L_0=\mathrm{log}(x),`$ $`L_1=\mathrm{log}(1x)`$ (56)
Then the results are:
$`P_{ns}^{+(2)DIS}`$ $`785.06𝒟_02974.4𝒟_1+645.33𝒟_2+14669.3758\delta (1x)+1868.3+6601.3x+243.6x^2`$
$`522.1x^3+77.391L_1^3+[2771.56x+3059.61]L_1^2+[2695.85x+13750]L_1`$
$`+[1.580215.818x]L_0^4+83.639L_0^3+[83.48L_1+915.18]L_0^2+[272.00x+2497.50`$
$`750.9L_1+8314.3L_1^2+544.00/(1x)]L_0`$
$`+n_f(325.18𝒟_0+403.89𝒟_178.222𝒟_22150.5868\delta (1x)+12.951+217.65x`$
$`+358.28x^2+44.79x^3+.95867xL_0^4+[2.573x5.6436]L_0^3+[118.68+10.503L_1]L_0^2`$
$`+[67.556/(1x)155.46L_1503.89L_1^2+33.778x327.76]L_04.6904L_1^3`$
$`+[152.43+167.97x]L_1^2+[180.68x1417.78]L_1)`$
$`+n_f^2(7.6750𝒟_011.457𝒟_1+2.3704𝒟_2+63.6358\delta (1x)4.883728.501x`$
$`17.293x^2.24667xL_0^3+[1.1852x/(1x).59259x+3.5556]L_0^2+[11.457`$
$`+3.9506x/(1x)4.3457x+10.817L_1]L_02.0000L_1^2+33.160L_1);`$ (57)
$`P_{ns}^{(2)DIS}`$ $`785.06𝒟_02974.4𝒟_1+645.33𝒟_2+14659\delta (1x)42.670+10704x+297.0x^2`$
$`433.2x^3+1.4321L_0^4+106.84L_0^3+[994.40860.64L_1]L_0^2+[272.00x630.82L_1`$
$`+2107.4+9310.9L_1^2+544.00/(1x)]L_0+65.291L_1^3+[3085.12771.5x]L_1^2`$
$`+[14503.+2695.9x]L_1`$
$`+n_f(325.18𝒟_0+403.89𝒟_178.222𝒟_22150.0\delta (1x)+75.786+413.96x+77.89x^2`$
$`+34.76x^3[1.136x+7.4805]L_0^3+[.59212L_1125.14]L_0^2+[381.08L_167.556/(1x)`$
$`+33.778x564.29L_1^2321.26]L_03.9570L_1^3+[167.97x149.42]L_1^2`$
$`[1421.1+180.68x]L_1)`$
$`+n_f^2(7.6750𝒟_011.457𝒟_1+2.3704𝒟_2+63.585\delta (1x)2.057255.288x+[3.5846`$
$`.59259x+4.0479L_1+1.1852x/(1x)]L_0^2+[23.959L_1+12.039+3.9506x/(1x)`$
$`4.3457x]L_02.2760L_1^2+30.600L_1);`$ (58)
$`P_{ps}^{(2)DIS}`$ $`n_f(193299672088x+104121x^2+964027x^3201675x^41327.61x^1820.836L_0^3`$
$`+[61102.6L_19741.91]L_0^2+[307888L_1190993L_1^275017.3196.207x^1]L_0+2332.86L_1`$
$`1876643L_1^2+10385.8L_1^31121.25xL_1+1876481xL_1^210509.5xL_1^3+24.88888L_1^4)`$
$`+n_f^2(530.035+7303.79x937.737x^27925.90x^3+1105.83x^4+9.36593x^11.65094L_0^4`$
$`+1.51450L_0^3+[251.096L_131.8095]L_0^2+[1502.08L_1+123.563+13313.6L_1^2]L_0+2649.16L_1`$
$`+13574.4L_1^22639.02xL_113583.3xL_1^2+1.18519L_1^3);`$ (59)
$`P_{gg}^{(2)DIS}`$ $`2643.5𝒟_0+4425.8739\delta (1x)20852+3968x3363x^2+4848x^3+14214x^1144L_0^4`$
$`+72L_0^3+[8757L_17471]L_0^2+[274.4+2675.8x^1+7305L_1]L_0+3589L_1`$
$`+n_f(412.172𝒟_0534.1666\delta (1x)+94680.9+423522x62541.01x^2569436x^3`$
$`+120946x^4+1149.99x^1+18.9631L_0^4+660.814L_0^3+[24297.9L_1+5133.55]L_0^2`$
$`+[1099250L_1^2175012L_1+220.737x^1+40461.3]L_024.8889L_1^4+[2062.11x1913.21]L_1^3`$
$`+[1093524x+1093454]L_1^2+[22404.4x+22442.5]L_1)`$
$`+n_f^2(1.77778𝒟_0+6.44153\delta (1x)19903.181663.3x+11472.2x^2+114322x^324596.2x^4`$
$`17.8171x^181.0657L_0^3+[5570.25L_11006.77]L_0^2+[7493.22215788L_1^2+(85.25x`$
$`+37019.4)L_1]L_0+[784.390787.057x]L_1^3+[212770x212747]L_1^2+[162.579283.399x]L_1)`$
$`+n_f^3(14.399+15.108x104.84x^2+41.797x^3+33.545x^4+.44376L_0^3+[22.307L_1`$
$`+2.6393]L_0^2+[139.31L_1^2112.57L_1+9.5276]L_0.26473(x1)L_1^3+140.68(x1)L_1^2`$
$`+112.93(x1)L_1);`$ (60)
$`P_{qg}^{(2)DIS}`$ $`n_f(5.2833\delta (1x)3963541679228x+400583x^2+2086958x^3413461x^4`$
$`3164.7x^1+19.852L_0^4+[1610.1252.5x]L_0^3+[130777L_118993]L_0^2`$
$`+[152378+612746L_1441.47x^13977765L_1^2]L_0+36.004L_1^4+[2923.93143.9x]L_1^3`$
$`+[4014417x4013999]L_1^2+[325972x323111]L_1)`$
$`+n_f^2(0.2404\delta (1x)2697.216699x+11510x^2+8064.9x^3+30.195x^110.237L_0^4`$
$`+[32.113+11.70x]L_0^3+[1088.9L_1427.6798.07x]L_0^2+[2782.9L_11477.3`$
$`28176L_1^2]L_0+2.6670L_1^3+[29629x29652]L_1^2+[7440.6x7360.6]L_1)`$
$`+n_f^3(0.0013\delta (1x)+143.623835.290x+540.973x^2+150.487x^3+.326927L_0^4+3.75075L_0^3`$
$`+[21.1805L_1+25.3458]L_0^2+[95.2390+196.232L_1865.734L_1^2]L_0+881.848(x1)L_1^2`$
$`+32.8575(x1)L_1);`$ (61)
$`P_{gq}^{(2)DIS}`$ $`10172.599𝒟_0+2619.956𝒟_1+3026.479𝒟_275.85220𝒟_3118.5187𝒟_4`$
$`17666.5673\delta (1x)63856.3226976x+5645.05x^2+370971x^380184.6x^4+6133.90x^1`$
$`52.9383L_0^4269.675L_0^3+[972.9x16883.4L_16887.42]L_0^2+[25459.0+1189.3x^1`$
$`692833L_1^2+122913L_1]L_0+89.4494L_1^4+[7443.027918.33x]L_1^3+[666832+658887x]L_1^2`$
$`+[16626.432060.8x]L_1`$
$`+n_f(935.1848𝒟_0550.4791𝒟_156.29637𝒟_2+25.28401𝒟_3+2589.9531\delta (1x)+35445+73884x`$
$`11203x^2127979x^3+27317x^4+350.55x^1+4.7407L_0^4+312.26L_0^3+[9521.1L_1+108.6x`$
$`+2357.5]L_0^2+[99.282x^145381L_1+16599+260063L_1^2]L_014.222L_1^4+[1762.4`$
$`+1847.8x]L_1^3+[254267x+254637]L_1^2+[4256.9+4324.8x]L_1)`$
$`+n_f^2(12.698𝒟_0+17.185𝒟_13.5556𝒟_293.6748\delta (1x)103.10+809.52x655.80x^2`$
$`5.0491x^18.0350L_0^3+[31.758L_120.430]L_0^2+[273.77L_1+317.30L_1^2144.58]L_0`$
$`3.5556L_1^3+[230.63x+241.15+3.5556x^1]L_1^2+[96.502x+11.852x^1110.71]L_1);`$ (62)
$`C_{Lns}^{(3)+DIS}(x)`$ $`3634.5+5025.2x614.77x^3996.21x^4+(1x)[8452.3L_1+4090.2L_1^2+175.59L_1^3`$
$`+225.30L_1^4]3280.3L_0L_11082.7L_0^2L_1911.45L_081.823L_0^2.72047L_0^31780.0L_0L_1^2`$
$`4059.2L_1+125.02L_1^2+21.113L_1^3+1.6059L_1^4`$
$`+n_f(617.051670.5x+(1x)[23.584L_1106.82L_1^2]+1717.8L_0L_1`$
$`+465.96L_0^2L_1+171.90L_0+6.9942L_0^2+370.25L_145.190L_1^2)`$
$`+n_f^2(17.038+35.968x+(1x)[22.215L_1+23.829L_1^2]66.179L_0L_112.884L_0^2L_1`$
$`5.1888L_0.025315L_0^2+24.794L_0L_1^215.012L_1+2.3704L_1^2)`$
$`+fl_{11}^{ns}n_f([107.0+321.05x54.62x^2](1x)26.717+9.773L_0`$
$`+[363.8+68.32L_0]xL_0320/81L_0^2[2+L_0])x;`$ (63)
$`C_{Lps}^{(3)DIS}(x)`$ $`n_f(1769.7441.62x[182.00L_0+899.64]x^1(1x)[23.584L_1106.82L_1^2]`$
$`+53648L_0L_1+11604L_0^2L_1894.81L_0+105.36L_0^2+(1x)[81652L_1+3880.7L_1^2]`$
$`76.310L_0^38700.9L_0L_1^2)`$
$`+n_f^2(4087.24143.1x+47.29x^1+(1x)[2293.9L_1+654.38L_1^2]+978.21L_0L_1`$
$`+2199.0L_0^2L_1+1484.3L_0+176.52L_0^2+18.327L_0^3+511.80L_0L_1^2`$
$`(1x)[22.215L_1+23.829L_1^2])`$
$`+fl_{11}^{ps}n_f([107.0+321.05x54.62x^2](1x)26.717+9.773L_0`$
$`+[363.8+68.32L_0]xL_0320/81L_0^2[2+L_0])x;`$ (64)
$`C_{Lg}^{(3)DIS}(x)`$ $`n_f(4573.1+77228x70637x^3[409.506L_0+2076.4]x^1`$
$`+(1x)[8666.9L_1+267612L_1^24500.1L_1^3]8146.1L_0L_1+4257.5L_0^2L_1`$
$`4277.2L_0241.08L_0^2246.51L_0^3+272818L_0L_1^2+.32800L_1)`$
$`+n_f^2(8878.114399x+5430.1x^3+102.40x^1+(1x)[5143.2L_183.489L_1^2]`$
$`+7051.6L_0L_1+5593.0L_0^2L_1+3258.9L_0+481.19L_0^2+68.034L_0^3+516.40L_0L_1^2)`$
$`+n_f^3(287.66494.46x+782.72x^3+(1x)[614.31L_11547.3L_1^2]32.680L_0L_1`$
$`112.24L_0^2L_1132.42L_026.899L_0^22.6004L_0^31490.4L_0L_1^2)`$
$`+fl_{11}^gn_f^2([0.0105L_1^3+1.550L_1^2+19.72xL_166.745x+0.615x^2](1x)`$
$`+20/27xL_0^4+[280/81+2.260x]xL_0^3[15.402.201x]xL_0^2`$
$`[71.660.121x]xL_0).`$ (65) |
warning/0507/physics0507101.html | ar5iv | text | # Production networks and failure avalanches
## 1 Networks of firms
Firms are not simply independent agents competing for customers on markets. Their activity involves many interactions, and some of them even involve some kind of cooperation. Interactions among firms might include:
* information exchange,;
* loans,;
* common endeavours;
* partial ownership;
* and of course economic transactions allowing production (the present paper).
Economic activity can be seen as occurring on an economic network (“le tissu économique”): firms are represented by vertices and their interactions by edges. The edges are most often asymmetric (think for instance of providers/customers interactions). The availability of empirical data has provoked research on the structure of these networks: many papers discuss their “small world properties” and frequently report scale free distribution of the connections among firms.
The long term interest of economic network research is rather the dynamics creating or occurring on these nets: how are connections evolving, what are the fluxes of information, decisions,, economic transactions etc … But dynamic studies lag behind statistical approaches because of conceptual difficulties and because time series of individual transactions are harder to obtain than time aggregated statistics.
The recent cascade of bankruptcies that occurred in Eastern Asia in 1997, provoked some research on the influence of the loans network structure on the propagation of “bad debts” and resulting avalanches of bankruptcies (,) . One of the most early papers on avalanche distribution in economic networks is due to Bak et al . It concerns production networks: edges represent suppliers/customers connections among firms engaged in batch production activity. The authors describe the distribution of production avalanches triggered by random independent demand events at the output boundary of the production network.
These papers (, and) are not based on any empirical description of the network structure, but assume a very simple interaction structure: star structure in the first case,, periodic lattice in Bak et al paper. They neither take into account price dynamics.
The present paper is along these lines: we start from a very simple lattice structure and we study the consequences of simple local processes of orders/production (with or without failure)/delivery/ profit/investment on the global dynamics: evolution of global production and wealth in connection to their distribution and local patterns. In the spirit of complex systems analysis, our aim is not to present specific economic prediction, but primarily to concentrate on the generic properties (dynamical regimes, transitions, scaling laws) common to a large class of models of production networks.
A minimal model of a production network will first be introduced in section 2. Simulation results are presented in section 3. Section 4 is a discussion of the genericity of the obtained results: reference is made to comparable soluble models. We also summarise the results of several variants of the simplest model. The conclusion is a discussion of possible applications to geographical economics.
## 2 A simple model of a production network
We can schematise the suppliers/customers interactions among firms by a production network, where firms are located at the vertices and directed edges represent the delivery of products from one firm to its customers (see figure 1).
Independent local failures to produce (or to deliver) by a firm might give rise to the propagation of shortage across the production network.
We have chosen a simple periodic lattice with three input connections of equal importance and three output per firm. The network is oriented from an input layer (say natural resources) towards an output layer (say the shelves of supermarkets). The transverse axis can be thought as representing either geographical position or some product space while the longitudinal axis relates to production. We here use a one dimensional transverse space to facilitate the representation of the dynamics by two-dimensional patterns, but there is no reason to suppose geographical or product space to be one-dimensional in the real world.
In real economies, the network structure is more heterogenous with firms of unequal importance and connectivity. Furthermore some delivery connections go backwards. Most often these backward connections concern equipment goods; neglecting them as we do here implies considering equipment goods dynamics as much slower than consumption goods dynamics. Anyway, since these backward connections enter positive feedback loops, we have no reason to suppose that they would qualitatively disrupt the dynamics that we further describe.
At each time step two opposite flows get across the lattice: orders are first transmitted upstream from the output layer; production is then transmitted downstream from the input layer to the output layer.
* Orders at the output layer
We suppose that orders are only limited by the production capacity<sup>2</sup><sup>2</sup>2A number of simplifying assumptions of our model are inspired from , especially the assumption that production is limited by production capacity, not by market. $`A_{0i}`$ of the firm in position $`0,i`$, where $`0`$ indicates the output layer, and $`i`$ the transverse position in the layer.
$`Y_{0i}`$ $`=`$ $`qA_{0i}`$ (1)
$`Y_{0i}`$ is the order in production units, and $`q`$ a technological proportionality coefficient relating the quantity of product $`Y`$ to the production capacity $`A`$, combining the effect of capital and labor. $`q`$ is further taken equal to 1 without loss of generality.
* Orders
Firms at each layer $`k`$, including the output layer, transfer orders upstream to get products from layer $`k+1`$ allowing them to produce. These orders are evenly distributed across their 3 suppliers upstream. But any firm can only produce according to its own production capacity $`A_{ki}`$. The planned production $`Y_{ki}`$ is then a minimum between production capacity and orders coming from downstream:
$`Y_{ki}`$ $`=`$ $`min(qA_{ki},{\displaystyle \underset{v}{}}{\displaystyle \frac{Y_{(k1)i}}{3}})`$ (2)
$`v`$ stands for the supplied neighborhood, here supposed to be the three firms served by firm $`k,i`$ (see figure 1).
We suppose that resources at the input layer are always in excess and here too, production is limited only by orders and production capacity.
* Production downstream
Starting from the input layer, each firm then starts producing according to inputs and to its production capacity; but production itself is random, depending upon alea. We suppose that at each time step some catastrophic event might occur with constant probability $`𝒫`$ and completely destroy production. Failures result in canceling production at the firm where they occur, but also reduce production downstream, since firms downstream have to reduce their own production by lack of input. These failures to produce are uncorrelated in time and location on the grid. Delivered production $`Y_{ki}^d`$ by firm $`k,i`$ then depends upon the production delivered upstream from its delivering neighborhood $`v_i^{}`$ at level $`k+1`$:
$`Y_{ki}^d`$ $`=`$ $`({\displaystyle \underset{i^{}v_i^{}}{}}Y_{(k+1)i^{}}^d{\displaystyle \frac{Y_{ki}}{_{i^{\prime \prime }v_i^{}}Y_{ki^{\prime \prime }}}})ϵ(t)`$ (3)
+ Whenever any of the firms $`i^{}v_i^{}`$ at level $`k+1`$ is not able to deliver according to the order it received, it delivers downstream at level $`k`$ to its delivery neighbourhood $`v_i^{}`$ in proportion of the initial orders it received, which corresponds to the fraction term;
+ $`ϵ(t)`$ is a random term equals to 0 with probability $`𝒫`$ and 1 with probability $`1𝒫`$.
The propagation of production deficit due to local independent catastrophic event is the collective phenomenon we are interested in.
* Profits and production capacity increase
Production delivery results into payments without failure. For each firm, profits are the difference between the valued quantity of delivered products and production costs, minus capital decay. Profits $`\mathrm{\Pi }_{ki}`$ are then written:
$`\mathrm{\Pi }_{ki}`$ $`=`$ $`pY_{ki}^dcY_{ki}^d\lambda A_{ki}`$ (4)
where $`p`$ is the unit sale price, $`c`$ is the unit cost of production, and $`\lambda `$ is the capital decay constant due to interest rates and material degradation. We suppose that all profits are re-invested into production. Production capacities of all firms are thus upgraded (or downgraded in case of negative profits) according to:
$`A_{ki}(t+1)=A_{ki}(t)+\mathrm{\Pi }_{ki}(t)`$ (5)
* Bankruptcy and re-birth.
We suppose that firms which capital becomes negative go into bankruptcy. Their production capacity goes to zero and they neither produce nor deliver. In fact we even destroy firms which capital is under a minimum fraction of the average firm (typically 1/500). A re-birth process occurs for the corresponding vertex after a latency period: re-birth is due to the creation of new firms which use the business opportunity to produce for the downstream neighborhood of the previously bankrupted firm. New firms are created at a unique capital, a small fraction of the average firm capital (typically 1/250).<sup>3</sup><sup>3</sup>3Adjusting these capital values relative to the average firm capital $`<A>`$ is a standard hypothesis in many economic growth models: one supposes that in evolving economies such processes depend upon the actual state of the economy and not upon fixed and predefined values..
The dynamical system that we defined here belongs to a large class of non linear systems called reaction-diffusion systems (see e.g. ) from chemical physics. The reaction part here is the autocatalytic loop of production and capital growth coupled with capital decay and death processes. The diffusion part is the diffusion of orders and production across the lattice. We can a priori expect a dynamical behaviour with spatio-temporal patterns, well characterised dynamical regimes separated in the parameter space by transitions or crossovers, and scale free distributions since the dynamics is essentially multiplicative and noisy. These expectations guided our choices of quantities to monitor during simulations.
## 3 Simulation results
### 3.1 Methods and parameter choice
Unless otherwise stated, the following results were obtained for a production network with 1200 nodes and ten layers between the input and the output.
Initial wealth is uniformly and randomly distributed among firms:
$$A_{ki}[1.0,1.1]$$
(6)
One time step correspond to the double sweep of orders and production across the network, plus updating capital according to profits. The simulations were run for typically 5000 time steps.
The figures further displayed correspond to:
* a capital threshold for bankruptcy of $`<A>/500`$;
* an initial capital level of new firms of $`<A>/250`$;
Production costs $`c`$ were 0.8 and capital decay rate $`\lambda =0.2`$. In the absence of failures, stability of the economy would be ensured by sales prices $`p=1.0`$. In fact, only the relative difference between these parameters influences stability. But their relative magnitude with respect to the inverse delay between bankruptcy and creation of new firm also qualitatively influence the dynamics.
In the limits of low probability of failures, when bankruptcies are absent, the linear relation between failure probability $`𝒫`$ and equilibrium price $`p`$ is written:
$`p=c+\lambda +{\displaystyle \frac{l}{2}}𝒫`$ (7)
where $`l`$ is the total number of layers. The $`\frac{l}{2}`$ comes from the fact that the integrated damage due to an isolated failure is proportional to the average number of downstream layers. The slopes at the origin of the breakeven lines of figure 2 verify this equation.
Most simulations were monitored online: we directly observed the evolution of the local patterns of wealth and production which our choice of a lattice topology made possible. Most of our understanding comes from these direct observations. But we can only display global dynamics or static patterns in this manuscript.
### 3.2 Monitoring global economic performance
The performance of the economic system under failures can be tested by checking which prices correspond to breakeven: the capital dynamics being essentially exponential, the parameter space is divided in two regions, where economic growth or collapse are observed. Drawing the breakeven manifolds for instance in the failure probability $`𝒫`$ and sale price $`p`$ plane allows to compare the influence of other parameters . The growth regime is observed in the low $`𝒫`$ and high $`p`$ region, the collapse regime in the high $`𝒫`$ and low $`p`$ region.
Figure 2 displays four breakeven manifolds corresponding to different lattice depths.
At low failure probability, the breakeven lines follow equation 7. At higher values of $`𝒫`$, interactions among firms failures are important, hence the non linear increase of compensating prices.
Breakeven manifold are a simple test of the economic performances of the network: when performances are poor, the compensating sales price has to be larger. We checked for instance that increasing the bankruptcy threshold and new firms initial capital increase global economic performance. On the other hand, increasing the time lag between bankruptcy and the apparition of new firms increase breakeven sale prices in the non-linear region.
Among other systematic tests, we checked parent models with more realistic representations of production costs such as:
* Influence of capital inertia; production costs don’t instantly readjust to orders: capital and labour have some inertia which we modeled by writing that productions costs are a maximum function of actual costs and costs at the previous period.
* Influence of the cost of credit: production failures increase credit rates.
Both variants of course yield higher breakeven sale prices; nevertheless these variants display the same generic properties that we will discuss in the next sections.
Most further results, dynamical and statistical, are based on runs close to the breakeven price in order to avoid systematic drifts and recalibrations.
### 3.3 Time evolution
The simplest way to monitor the evolution of the system is to display the time variations of some of its global performance. Figure 3 displays the time variations of total delivered production $`Y^d`$, total wealth $`A`$, total undelivered production due to failures and the fraction of active firms for a 1200x10 lattice, with a probability of failure of 0.05 and a compensation sale price of 1.185. Time lag between bankruptcy and and new firm creation is either 1 (for the left diagram) or 5 (for the right digram).
The features that we here report are generic to most simulation at breakeven prices. During the initial steps of the simulation, here say 1000, the wealth distribution widens due to the influences of failures. Bankruptcies cascades do not occur as observed by checking the number of active firms, until the lowest wealth values reach the bankruptcy threshold. All quantities have smooth variations. Later, for $`t>1000`$ one observes large production and wealth fluctuations characteristic of critical systems.
At larger time lag (5) between bankruptcy and firm re-birth, when bankruptcies become frequent, they can cascade across the lattice and propagate in both network directions as seen on the right diagram of figure 3. A surprising feature of the dynamics is that avalanches of bankruptcies are not correlated with production level. Even when only one tenth of the firms are active, the total production is still high. In fact, in this model, most of the total production is dominated by large firms, and avalanches which concern mostly small firms are of little consequence for the global economy.
Battiston etal study more thoroughly the time dynamics of a related model (large sale price fluctuations possibly inducing bankruptcies and lack of payment) in .
### 3.4 Wealth and production patterns
Like most reaction-diffusion systems, the dynamics is not uniform in space and display patterns. The wealth and production patterns displayed after 5000 time steps on figure 4 and 5 were obtained for $`𝒫=0.05`$ . They reflect wide distributions and spatial organisation. In these diagrams, production flows upward. The upper diagram displays wealth $`A`$ and the lower one production $`Y_d`$. The intermediate bar is the colour scale, black=0, violet is the maximum wealth or production. (We in fact displayed square roots of $`A`$ and $`Y_d`$ in order to increase the visual dynamics of the displays; otherwise large regions of the patterns would have been red because of the scale free distributions of $`A`$ and $`Y_d`$, see further).
The important result is that although production has random fluctuations and diffuses across the lattice, the inherent multiplicative (or autocatalytic) process of production + re-investment coupled with local diffusion results in a strong metastable local organisation: the dynamics clusters rich and productive firms in ”active regions” separated by ”poor regions” (in red or black).
These patterns are evolving in time, but are metastable on a long time scale, say of the order of several 100 time steps as seen on the succession of production patterns at different steps of the simulation as one can observe on figure 6: successive patterns at time 1250, 1750 and 2250.
The relative importance of active (and richer) regions can be checked by a Zipf plot. We first isolate active regions by ”clipping” the dowstream (along $`k`$ axis) integrated wealth at a level of one thousandth of the total production<sup>4</sup><sup>4</sup>4Clipping here means that when the production level is lower than the threshold it is set to zero.
We then transversally (along $`i`$ axis) integrate the wealth of active regions and order these regional wealths to get the Zipf plots.
All 3 Zipf plots display some resemblance with standard Zipf plots of individual wealth, firm size and city size. For the model discussed here, the size decrease following approximately a power law. The apparent<sup>5</sup><sup>5</sup>5the approximate algorithm that we use to isolate high productivity regions is responsible for the kinks in the Zipf plot exponent is one when the time lag is 1. It is much higher when the time lag is 5.
Zipf plots of output<sup>6</sup><sup>6</sup>6 rather than vertically integrating production, we applyed the clipping, horizontal integration and ordering algorithm to firms at the output layer ($`k=0`$) active regions (not shown here) display the same characteristics.
When the time lag is 5, the most productive region accounts for more than 50 perc. of total production. The figure is 18 perc. for the second peak. The distribution typically is ”winner takes all”. The equivalent figures when the time lag is 1 are 10 and 8.5 perc..
In conclusion, the patterns clearly display some intermediate scale organisation in active and less active zones: strongly correlated active regions are responsible for most part of the production. The relative importance of these regions obeys a Zipf distribution.
### 3.5 Wealth and production histograms
The multiplicative random dynamics of capital and the direct observation of wealth and production would lead us to predict a scale free distribution<sup>7</sup><sup>7</sup>7What we mean here by scale free is that no characteristic scale is readily apparent from the distribution as opposed for instance to gaussian distributions. Power law distributions are scale free. A first discussion of power law distributions generated by multiplicative processes appeared in . of wealth and production.
The cumulative distribution functions (cdf) of wealth and production observed on figure 8 are indeed wide range and do not display any characteristic scale: The data wealth and production were taken for the same conditions as the previous figures at the end of the simulation, i.e. after 5000 time steps. The medium range of the cdf when time lag is 1 (figure 8a) extends on one and a half decade with an apparent slope of $`1\pm 0.05`$ in log-log scale.
This observed dependence of the wealth cdf, log normal at lower $`A`$ values followed by power law at intermediate $`A`$ values, is consistent with expressions derived for pdf in the literature on coupled differential equations with multiplicative noise. Bouchaud and Mézard e.g. obtained:
$$P(w)=Z\frac{exp\frac{1\mu }{w}}{w^{1+\mu }}$$
(8)
(where $`w`$ stands for the wealth relative to average wealth $`\overline{A}`$), from the differential system:
$$\frac{dA_i}{dt}=\eta _i(t)A_i+J(\overline{A}A_i).$$
(9)
where $`\eta _i(t)`$ is a random multiplicative noise, with variance $`\sigma ^2`$; $`\mu =1+\frac{J}{\sigma ^2}`$.
At higher wealth, the straight line giggles and drops much faster: this is because of the underlying region structure. The last 80 perc. of the wealth is concentrated in two rich regions and its distribution is dominated by local diffusion phenomena in these regions.
The departure form the standard (equ.8) distribution is even more noticeable when avalanches are present. The large wealth shoulder is bigger (95 perc. of production) and the first point at zero wealth stands well above the rest of the distribution: it corresponds to those 50 perc. of the firms which are momentarily bankrupted. The fraction of bankrupted firms fluctuates in time and so does the slope of the linear segment<sup>8</sup><sup>8</sup>8 both fluctuations are correlated since the slope of the linear segment depends upon the number of firms in the distribution.
In conclusion, the observed statistics largely reflect the underlying region structure: at intermediate levels of wealth, the different wealth peaks overlap (in wealth, not in space!): we then observed a smooth cdf obeying equation 8. At the large wealth extreme the fine structure of peaks is revealed.
## 4 Conclusions
The simple model of production networks that we proposed presents some remarkable properties:
* Scale free distributions of wealth and production.
* Large spatial distribution of wealth and production.
* A few active regions are responsible for most production.
* Avalanches of bankruptcies occur for larger values of the time lag between bankruptcy and firm re-birth. But even when most firms are bankrupted, the global economy is little perturbed.
Are these properties generic to a large class of models? we will first briefly report on equations which display similar behaviour and then examine the results which we obtained with variants of the model.
### 4.1 Formal approaches of similar dynamics
A number of models which display equivalent phenomena have been proposed and formally solved. We kept our own notation to display similarities:
* Growth by deposition on surfaces, Edwards/Wilkinson:
$`{\displaystyle \frac{dA}{dt}}=D\mathrm{\Delta }A+\eta (x,t)`$ (10)
$`A`$ stands for the distance to the interface. $`D`$ is the surface diffusion constant of the deposited material and $`\eta _i(t)`$ is an addititive noise. Other models were proposed by Karkar/Parisi/Zhang, Derrida/Spohn, etc.
* Generalised Volterra-Lotka from econophysics: (Bouchaud, Cont, Mezard, Sornette, Solomon etc.)
$`{\displaystyle \frac{dA_i}{dt}}=A_i\eta _i(t)+{\displaystyle \underset{j}{}}J_{ij}A_j{\displaystyle \underset{j}{}}J_{ji}A_i`$ (11)
$`A`$ stands for individual wealth of agents and $`\eta _i(t)`$ is a multiplicative noise. Agents are involved in binary transactions of ”intensity” $`J_{ij}`$. Mean field formal solutions displays scale free distribution of wealth. Simulations display patterns on lattice structures (Souma etal).
* Solomon etal. Reaction-Diffusion AB models.
$`{\displaystyle \frac{dA}{dt}}=kA\eta (x,t)+D\mathrm{\Delta }A`$ (12)
$`A`$ is the chemical concentration of a product involved in an auto-catalytic chemical reaction, $`D`$ is its diffusion constant. Simulations and formal derivations yield spatio-temporal patterns similar to ours.
### 4.2 Variants of the original model
We started checking three variants, with for instance more realistic production costs taking into account:
* Influence of capital inertia: production costs don’t instantly readjust to orders; capital and labour have some inertia which we modeled by writing that productions costs are a maximum function of actual costs and costs at the previous period.
* Influence of the cost of credit: production failures increase credit rates.
The preliminary simulations confirm the genericity of our results.
The third variant is a model with ”adaptive firms”. The lattice connection structure supposes a passive reactive behaviour of firms. But if a firm is consistently delivering less than the orders it receives, its customers should order less from it and look for alternative suppliers. Such adaptive behaviour leading to an evolutive connection structure would be more realistic.
We then also checked an adaptive version of the model by writing that orders of firm $`i`$ are proportional to the production capacity $`A`$ of the upstream firms connected to firm $`i`$. Simulations gave qualitative results similar to those obtained with fixed structures.
We observe that adaptation strongly re-enforce the local structure of the economy. The general picture is the same scale free distribution of production and wealth with metastable patterns. Due to the strong local character of the economy:
* Avalanches of production are observed (see figure 9), even when time lag is short (time lag of 1).
* The spatial periodicity of the active zones is increased (see figure 9 with larger density of smaller zones). But again the activity distribution among zones is like ”winner takes all” (figure 7).
### 4.3 Checking stylised facts
Even though the present model is quite primitive<sup>9</sup><sup>9</sup>9 We e.g. discuss a ”Mickey Mouse” economy with fixed prices independent from supply and demand. Introducing price dynamics is not a major challenge: we would simply face an ”embarras de richesse” having to choose among either local or global prices. In fact both kind of adjustment have already been tested: global adjustment in the case of production cost connected to production failure through credit costs, or local adjustment in the case of adaptive behaviour. We have already shown that they don’t change the generic properties of the dynamics. it is still tempting to draw some conclusions that could apply to real economies. The most striking result to our mind is the strong and relatively stable spatial disparities that it yields. Let us compare this prediction to the economic situation of developing countries: large and persistent disparities in wealth and production level as compared to developed countries. We can even go further and raise questions about the influence of the depth of the production network or the kind of investment needed:
* One generally agrees that disparities between developing and developed countries increased since industrial revolution. This is also a period during which production became more specialised, which translates in our model as increasing the network depth: for instance a shoemaker would in the past make and sell a pair of shoes from leather obtained from a cattle breeder. Nowadays the shoe production and delivery process involve many more stages. Our simulations have shown that increasing depth increases the fragility of economies to failures and bankruptcies. The new industrial organisation may have detrimental effects on developing economies.
* Obviously investment policies in developing countries yield some coordination across the whole production chain. Bad economic results might be due to very local conditions but can also reflect the lack of suppliers/producers connections.
The above remarks are not counter-intuitive and these conclusions could have been reached by verbal analysis. What is brought by the model is the dramatic and persistent consequences of such apparently trivial details.
Acknowledgments: We thank Bernard Derrida and Sorin Solomon for illuminating discussions and the participants to CHIEF Ancona Thematic Institute, especially Mauro Gallegati. CHIEF was supported by EXYSTENCE network of excellence, EC grant FET IST-2001-32802. This research was also supported by COSIN FET IST-2001-33555, E2C2 NEST 012975 and CO3 NEST 012410 EC grants. |
warning/0507/cond-mat0507175.html | ar5iv | text | # Nonlinear magnetoelastic behavior of the metastable bcc phases Co and Ni: Importance of third-order contributions for bcc Ni
## Abstract
The first- and second-order magnetoelastic coefficients of the metastable bcc phases Co and Ni are calculated by using a combination of the phenomenological theory of nonlinear magnetoelasticity with the ab-initio density functional electron theory. The magnetoelastic behavior of the bcc phases is drastically different from that of the corresponding fcc phases. The recently synthesized bcc phase of Ni appears to be an example of a material for which third-order magnetoelastic effects are essential.
In recent years it became possible to stabilize metastable phases of materials by growth on appropriate substrates. For the transition metals Fe, Co and Ni this is especially interesting because in these systems magnetism and structure are closely related. Using molecular beam epitaxy, the fcc phases of Fe and Co could be stabilized on substrates at room temperature Pescia et al. (1987). It became even possible to synthesize the bcc phase of Co Prinz (1985) on various substrates (Ref. Valvidares et al., 2004 and references therein), and most recentlyTian et al. (2005) the bcc phase of Ni. Both of these materials turned out to be ferromagnetic at room temperature, with a magnetic moment per atom of $`1.53\mu _\mathrm{B}`$ (Co) and $`0.54\mu _\mathrm{B}`$ (Ni).
From the viewpoint of technological applications of ultrathin magnetic films the most important feature is the magnetic anisotropy. Because in general there will be a lattice mismatch between the substrate and the magnetic film, magnetoelastic contributions to the magnetic anisotropy may be important. For instance, it has been suggestedSubramanian et al. (1995) that the in-plane anisotropy of bcc Co on GaAs is dominated by the magnetoelastic contribution, although the epitaxial strains in this material are rather small, about 0.25%. For comparison, for bcc Co on Pt(001) the epitaxial strains are considerably largerValvidares et al. (2004) (-1.8% in plane and 5.1% out of plane). It is well known that for considerable epitaxial strains nonlinear contributions to the magnetoelastic energy become essential. This has been demonstrated experimentally by cantilever bending beam experiments (see, e.g., Refs. Weber et al., 1994; Sander, 1999; Wedler et al., 1999): When changing the direction of the magnetization in the epitaxial film by changing the direction of the external magnetic field, the magnetostrictive stress $`\mathrm{\Delta }\sigma ^\mathrm{m}`$ along the cantilever axis changes, resulting in a detectable change of the bending of the film-substrate composite. In the framework of linear magnetoelastic theory, this change should be independent of the magnitude of the epitaxial strain and should be determined by the first-order magnetoelastic coefficients, i.e., $`B_1`$ and $`B_2`$ for cubic materials. Experimentally, however, a linear dependence of $`\mathrm{\Delta }\sigma ^\mathrm{m}`$ on the strain was found, which was ascribed to nonlinear magnetoelastic effects.
For a proof of this conjecture a knowledge of the first- and second-order magnetoelastic coefficients of the respective bulk material is required. The standard method to determine them is the ultrasonic pulse echo experiment. Because the attainable strains in these experiments are very small, it is, however, nearly impossible to explore the second-order magnetoelastic coefficients by these experiments. The first confirmation of the conjecture therefore was supplied by theory. By a combination of the phenomenological theory of nonlinear magnetoelasticitydu Trémolet de Lacheisserie (1993) with the ab-initio density functional theory it has been shown (see, for example, Refs. Komelj and Fähnle, 2002a, b; Fähnle and Komelj, 2002; Komelj and Fähnle, 2001 and references therein) that the second-order magnetoelastic contribution indeed may be very large, especially for the case of Fe. The theory was also ableKomelj and Fähnle (2002b) to suggest a complete set of six cantilever experiments to determine the first-order ($`B_1`$ and $`B_2`$) and the second-order ($`m_1^{\gamma ,2}`$, $`m_2^{\gamma ,2}`$, $`m_1^{ϵ,2}`$, $`m_2^{ϵ,2}`$, $`m_3^{\gamma ,2}`$, $`m_3^{ϵ,2}`$) magnetoelastic coefficients of a cubic material. Thereby ($`m_1^{\gamma ,2}`$, $`m_2^{\gamma ,2}`$) is related to pure tensile strains, ($`m_1^{ϵ,2}`$, $`m_2^{ϵ,2}`$) to tensile and shear strains, and ($`m_3^{\gamma ,2}`$, $`m_3^{ϵ,2}`$) to pure shear strains. The first- and second-order coefficients have been calculatedFähnle and Komelj (2002) by the ab-initio electron theory for Fe, fcc Co, Ni, $`\mathrm{Ni}_3\mathrm{Fe}`$ and CoFe.
The determination of the magnetoelastic coefficients is especially difficult for metastable phases which can be synthesized only as epitaxial films on substrates, like fcc Co, bcc Co and bcc Ni, and in these cases the help of electron theory is very important. For the case of fcc Co, the theory has shownKomelj and Fähnle (2001) that the nonlinear magnetoelastic coupling coefficients are essential for the magnetostrictive strain but have only little influence on the strain-induced out-of-plane anisotropy. In the present paper we apply the theory to the case of bcc Co and bcc Ni. It will be shown that in these systems the nonlinear magnetoelastic coefficients are again very large. Furthermore, it will be shown that bcc Ni is the first example of a system for which third-order magnetoelastic effects become relevant.
According to Ref. Komelj and Fähnle, 2002b the magnetoelastic coefficients may be obtained by exposing the cubic material to certain strain modes $`ϵ_i`$. Then the difference $`\mathrm{\Delta }e_i`$ in the total energy per atom when changing the direction of the magnetization from $`\alpha _i^1`$ to $`\alpha _i^2`$ is calculated:
$`i=1:ϵ_1=ϵ_{xx}=ϵ_0,\alpha _i^1=100,\alpha _i^2=001`$
$$\mathrm{\Delta }e_1=B_1ϵ_0+\left(B_1+\frac{1}{2}m_1^{\gamma ,2}\right)ϵ_0^2$$
(1)
$`i=2:ϵ_2=ϵ_{yy}=ϵ_{zz}=ϵ_0,\alpha _i^1=100,\alpha _i^2=001`$
$$\mathrm{\Delta }e_2=B_1ϵ_0+\frac{1}{2}\left(B_1m_1^{\gamma ,2}+m_2^{\gamma ,2}\right)ϵ_0^2$$
(2)
$`i=3:ϵ_3=ϵ_{xx}=ϵ_{xy}=ϵ_0,\alpha _i^1=010,\alpha _i^2=110`$
$$\mathrm{\Delta }e_3=\left(\frac{B_1}{2}+B_2\right)ϵ_0+\frac{1}{2}\left(\frac{1}{2}\left(B_1+m_1^{\gamma ,2}\right)+B_2+m_2^{ϵ,2}\right)ϵ_0^2$$
(3)
$`i=4:ϵ_4=ϵ_{zz}=ϵ_{xy}=ϵ_0,\alpha _i^1=010,\alpha _i^2=110`$
$$\mathrm{\Delta }e_4=B_2ϵ_0+\frac{1}{2}m_1^{ϵ,2}ϵ_0^2$$
(4)
$`i=5:ϵ_5=ϵ_{xy}=ϵ_0,\alpha _i^1=110,\alpha _i^2=001`$
$$\mathrm{\Delta }e_5=B_2ϵ_0+\frac{1}{2}\left(m_3^{\gamma ,2}B_1\right)ϵ_0^2$$
(5)
$`i=6:ϵ_6=ϵ_{yz}=ϵ_{zx}=ϵ_0,\alpha _i^1=11\overline{2},\alpha _i^2=111`$
$$\mathrm{\Delta }e_6=\frac{8}{3}B_2ϵ_0+\frac{1}{12}\left(B_1+2B_2m_3^{\gamma ,2}+2m_3^{ϵ,2}\right)ϵ_0^2$$
(6)
The coefficient $`B_1`$ and the pair ($`m_1^{\gamma ,2}`$, $`m_2^{\gamma ,2}`$) of second-order coefficients are obtained from eqs.(1,2) by fitting parabola to the data points for $`\mathrm{\Delta }e_1(ϵ_0)`$ and $`\mathrm{\Delta }e_2(ϵ_0)`$. Similarly, the coefficients $`B_2`$, ($`m_3^{\gamma ,2}`$, $`m_3^{ϵ,2}`$) are obtained from eqs.(5,6) by parabolic fits. Finally, the pair ($`m_1^{ϵ,2}`$, $`m_2^{ϵ,2}`$) is obtained from eqs.(3,4) via parabolic fits using the already determined coefficient $`m_1^{\gamma ,2}`$. As long as the parabolic fits represent the calculated data points $`\mathrm{\Delta }e_i(ϵ_0)`$ well, we can conclude that third-order magnetoelastic effects can be neglected for the considered range of $`ϵ_0`$.
The calculations of $`\mathrm{\Delta }e_i(ϵ_0)`$ were performed by applying the ab-initio density functional theory taking into account the spin-orbit coupling which is responsible for magnetoelasticity in a perturbative manner using the second-variational methodSingh (1994). Furthermore, we use the WIEN97 codeBlaha et al. (1990) which adopts the full-potential linearized-augmented-plane-wave method (FLAPW)Wimmer et al. (1981) as well as the local-spin-density approximation (LSDA)Perdew and Wang (1992) and the generalized-gradient approximation (GGA)Perdew et al. (1992) for the exchange-correlation functional. The strains $`ϵ_i`$ were applied with respect to the theoretically determined equilibrium lattice parameters $`a=0.273(0.281)\mathrm{nm}`$ for bcc Co and $`a=0.273(0.279)\mathrm{nm}`$ for bcc Ni in LSDA (GGA). The resulting LSDA (GGA) magnetic moments per atom of $`1.63(1.74)\mu _\mathrm{B}`$ for bcc Co and of $`0.47(0.53)\mu _\mathrm{B}`$ for bcc Ni are in agreement with the experimental values of $`1.53\mu _\mathrm{B}`$ and $`0.54\mu _\mathrm{B}`$, respectively.
For the case of bcc Co all the data points $`\mathrm{\Delta }e_i(ϵ_0)`$ could be perfectly fitted by parabola in the range $`0.03ϵ_00.03`$, i.e., third-order effects can be neglected. Like for other materials Komelj and Fähnle (2002a, b); Fähnle and Komelj (2002); Komelj and Fähnle (2001) the discrepancy between LSDA and GGA may be quite large. Because for the experimentally well investigated $`B_1`$ of bcc Co, fcc Ni and fcc Co the agreement with the GGA values was better than the agreement with LSDA, we concentrate in the following on the GGA results. For bcc Co the values of $`B_1`$ and $`B_2`$ are quite large as compared to bcc Fe, fcc Ni and fcc Co. The second-order coefficients are also large. It is interesting that there is a very large difference between bcc Co and fcc Co. This holds even for the first-order coefficients $`B_1`$ and $`B_2`$ which magnitudes are considerably larger and of opposite sign for bcc Co as compared to fcc Co. In Ref. Subramanian et al., 1995 it has been assumed that for bcc Co the first-order magnetoelastic coefficients can be approximated by those of fcc Co, in contrast to the results of our calculation.
The case of bcc Ni is even much more interesting because, as shown in Fig. 1, the data points for $`\mathrm{\Delta }e_i(ϵ_0)`$ show a drastic deviation from a parabolic behavior in the range $`0.03ϵ_00.03`$. To the best of our knowledge, bcc Ni therefore represents the first known material for which third-order magnetoelastic effects become very important. Another surprising result is that for bcc Ni the magnitude of $`B_1`$ is very small ($`1.3\mathrm{MJ}/\mathrm{m}^3`$), much smaller than the one for fcc Ni ($`10.2\mathrm{MJ}/\mathrm{m}^3`$). As in the case of Co, the magnetoelastic properties of the bcc phase are drastically different from those of the fcc phase. This is in line with the experimental observationsTian et al. (2005) that the cubic magnetic anisotropy constant $`K_1`$ of bcc Ni is drastically different from the one of fcc Ni, and this was attributed to the different electronic band structures as found by angle-resolved photoemission.
We hope that our prediction of strong third-order contributions to the magnetoelastic properties of bcc Ni will initiate an experimental investigation by cantilever bending-beam experiments. To do this one has to grow epitaxial films of bcc Ni with various average epitaxial strains $`ϵ_0`$ which may be controlled with the film thickness Weber et al. (1994) and then the change $`\mathrm{\Delta }\sigma ^\mathrm{m}`$ of the magnetostrictive stress due to a change of the magnetization direction has to be measured. For the case that third-order effects are relevant we expect a parabolic dependence:
$$\mathrm{\Delta }\sigma ^\mathrm{m}=a+D_1ϵ_0+D_2ϵ_0^2$$
(7)
As discussed above, a linear dependence has been observed experimentally already for several materials. The observation of a quadratic contribution would mean that for the first time a material was found for which the third-order magnetoelastic contribution is relevant. |
warning/0507/math0507554.html | ar5iv | text | # Manifolds with Commuting Jacobi Operators
## 1. Introduction
Let $``$ be the Levi-Civita connection of a connected smooth $`m`$-dimensional Riemannian manifold $`(M,g)`$. Let $`(x,y):=_x_y_y_x_{[x,y]}`$ be the associated curvature operator. The Jacobi operator $`J(x):y(y,x)x`$ is intimately related with the underlying geometry of the manifold. One says that $`(M,g)`$ is Osserman if the eigenvalues of $`J(x)`$ are constant on the sphere bundle $`S(M,g)`$ of unit tangent vectors. One says that $`(M,g)`$ is a local $`2`$-point homogeneous space if the local isometries of $`(M,g)`$ act transitively on $`S(M,g)`$; this necessarily implies that $`(M,g)`$ is Osserman. Osserman wondered if the converse held; this has been called the Osserman conjecture by subsequent authors. Chi and Nikolayevsky established the Osserman conjecture if $`m16`$. This gives a very pretty characterization of local $`2`$-point homogeneous spaces in terms of the geometry of the Jacobi operator.
Tsankov began a similar characterization of manifolds of constant sectional curvature in terms of the Jacobi operator by showing:
###### Theorem 1.1.
Let $`(M,g)`$ be a hypersurface in $`^{m+1}`$. If $`J(x)J(y)=J(y)J(x)`$ provided $`xy`$ and if $`m3`$, then $`(M,g)`$ has constant sectional curvature.
In this paper, we remove the hypothesis that $`(M,g)`$ is a hypersurface to characterize flat manifolds and manifolds of constant sectional curvature in terms of commutation properties of the Jacobi operator:
###### Theorem 1.2.
Let $`(M,g)`$ be a Riemannian manifold of dimension $`m3`$.
1. If $`J(x)J(y)=J(y)J(x)`$ for all $`x,y`$, then $`(M,g)`$ is flat.
2. If $`J(x)J(y)=J(y)J(x)`$ provided $`xy`$, then $`(M,g)`$ has constant sectional curvature.
It is convenient to work in the purely algebraic context. Let $`V`$ be a real vector space of dimension $`m3`$ which is equipped with a positive definite inner product $`,`$. Let $`R^4V^{}`$ be an algebraic curvature tensor on $`V`$, i.e. $`R`$ has the usual symmetries of the curvature tensor:
(1.a)
$$\begin{array}{c}R(x,y,z,w)=R(z,w,x,y)=R(y,x,z,w),\hfill \\ R(x,y,z,w)+R(y,z,x,w)+R(z,x,y,w)=0.\hfill \end{array}$$
Let $`(x,y)`$ be the associated curvature operator; $``$ is characterized by the identity:
$$(x,y)z,w=R(x,y,z,w).$$
The Jacobi operator $`J(x)`$ is then defined by $`J(x):y(y,x)x`$. One says that $`R`$ is Osserman if the eigenvalues of $`R`$ are constant on the associated unit sphere $`S(V):=\{\xi V:\xi ,\xi =1\}`$. Motivated by Theorem 1.1, one says that $`R`$ is Jacobi-Tsankov if
$$XY\text{implies}J(X)J(Y)=J(Y)J(X).$$
The following two tensors will play a crucial role in our investigation. The tensor $`R_0`$ of constant sectional curvature $`+1`$ is defined by
$$R_0(x,y)z=y,zxx,zy.$$
Let $`\mathrm{\Theta }`$ be a linear endomorphism of $`V`$. If $`\mathrm{\Theta }^2=\text{id}`$ and if $`\mathrm{\Theta }^{}=\mathrm{\Theta }`$, then $`\mathrm{\Theta }`$ is said to be a Hermitian almost complex structure on $`V`$. Such a $`\mathrm{\Theta }`$ exists, of course, if and only if $`m`$ is even. If $`\mathrm{\Theta }`$ is a Hermitian almost complex structure on $`V`$, set:
$$R_\mathrm{\Theta }(x,y)z:=\mathrm{\Theta }y,z\mathrm{\Theta }x\mathrm{\Theta }x,z\mathrm{\Theta }y2\mathrm{\Theta }x,y\mathrm{\Theta }z.$$
The tensors $`R_0`$ and $`R_\mathrm{\Theta }`$ play a crucial role in the proof of the Osserman conjecture and also play a crucial role in our present analysis. We will derive Theorem 1.2 from the following purely algebraic assertion:
###### Theorem 1.3.
Let $`\mathrm{\Theta }`$ be a Hermitian almost complex structure on $`V`$.
1. If $`J(x)J(y)=J(y)J(x)`$ for all $`x,y`$, then $`R=0`$.
2. $`R_0`$ and $`R_\mathrm{\Theta }`$ are Jacobi-Tsankov curvature tensors.
3. If $`R`$ is Jacobi-Tsankov, then either $`R=cR_0`$ or $`R=cR_\mathrm{\Theta }`$ for some Hermitian almost complex structure $`\mathrm{\Theta }`$.
Here is a brief guide to this paper. In Section 2, we establish Theorem 1.3. In Section 3, we use Theorem 1.3 to prove Theorem 1.2.
## 2. Algebraic results
Throughout this section, let $`R`$ be an algebraic curvature tensor on a vector space $`V`$ of dimension $`m`$ equipped with a positive definite inner product $`,`$.
###### Proof of Theorem 1.3 (1).
Suppose that $`J(x)J(y)=J(y)J(x)`$ for all $`x,yV`$. The Jacobi operators $`\{J(x)\}|_{xV}`$ form a commuting family of self-adjoint operators. Such a family can be simultaneously diagonalized - i.e. we can decompose
$$V=_\lambda V_\lambda \text{where}J(x)\xi =\lambda (x)\xi \xi V_\lambda .$$
Choose $`\xi V`$ and decompose $`\xi =_\lambda \xi _\lambda `$ for $`\xi _\lambda V_\lambda `$. Let
$$𝒪:=\{\xi V:\xi _\lambda 0\lambda \}.$$
Since $`𝒪=_\lambda \{VV_\lambda ^{}\}`$, $`𝒪`$ is a dense open subset of $`V`$. We have
$$0=J(\xi )\xi =_\lambda \lambda (\xi )\xi _\lambda .$$
Since $`\{\xi _\lambda \}`$ is a linearly independent set, this implies $`\lambda (\xi )=0`$ for all $`\lambda `$. Fix unit vectors $`\eta _\lambda V_\lambda `$. Then $`\lambda (\xi )=J(\xi )\eta _\lambda ,\eta _\lambda `$, so $`\lambda `$ is a continuous function of $`\xi `$. As $`\lambda ()`$ vanishes on $`𝒪`$, which is a dense open subset of $`V`$, $`\lambda ()`$ vanishes identically. Thus $`J(x)=0`$ for all $`xV`$; it now follows that $`R=0`$; see, for example, the discussion in . ∎
###### Proof of Theorem 1.3 (2).
Suppose that $`\mathrm{\Theta }`$ defines $`R`$ where $`\mathrm{\Theta }`$ is a Hermitian almost complex structure on $`V`$. If $`xy`$, then $`\mathrm{\Theta }x,\mathrm{\Theta }y=x,y=0`$. Consequently:
$`J(x)\xi =R(\xi ,x)x=\mathrm{\Theta }x,x\mathrm{\Theta }\xi \mathrm{\Theta }\xi ,x\mathrm{\Theta }x2\mathrm{\Theta }\xi ,x\mathrm{\Theta }x=3\xi ,\mathrm{\Theta }x\mathrm{\Theta }x,`$
$`\{J(x)J(y)J(y)J(x)\}z=9z,\mathrm{\Theta }y\mathrm{\Theta }y,\mathrm{\Theta }x\mathrm{\Theta }x9z,\mathrm{\Theta }x\mathrm{\Theta }x,\mathrm{\Theta }y\mathrm{\Theta }y=0.`$
Thus $`R_\mathrm{\Theta }`$ is Tsankov. Similarly if $`R=R_0`$ is the algebraic curvature tensor of constant sectional curvature $`+1`$, then:
$`J(x)\xi =x,x\xi \xi ,xx,`$
$`J(x)J(y)z=J(x)\{y,yzz,yy\}`$
$`=y,yx,xzy,yz,xxz,yx,xy+z,yy,xx`$
$`=y,yx,xzy,yz,xxz,yx,xy=J(y)J(x).`$
Assertion (2) of Theorem 1.3 now follows. ∎
Before establishing Assertion (3) of Theorem 1.3, we need a number of technical results. It is convenient to polarize the Jacobi operator to define a bilinear and self-adjoint operator:
$$J(x,y):z\frac{1}{2}(z,x)y+\frac{1}{2}(z,y)x.$$
We note that
(2.a)
$$\begin{array}{c}J(x)=J(x,x),J(x,y)y=\frac{1}{2}J(y)x,\text{and}\hfill \\ J(\mathrm{cos}\theta x+\mathrm{sin}\theta y)=\mathrm{cos}^2\theta J(x)+2\mathrm{cos}\theta \mathrm{sin}\theta J(x,y)+\mathrm{sin}^2\theta J(y).\hfill \end{array}$$
Let $`r(x):=\mathrm{Rank}\{J(x)\}`$. Since $`J(x)x=0`$, $`r(x)m1`$ for all $`xV`$.
###### Lemma 2.1.
Let $`0R`$ be a Jacobi-Tsankov algebraic curvature tensor. Suppose that $`r(x)=m1`$ for some $`xS(V)`$.
1. $`J()`$ has maximal rank on an open dense subset of $`V`$.
2. $`R`$ has constant sectional curvature $`c0`$.
###### Proof.
We clear the previous notation and let $`𝒪:=\{xV:r(x)=m1\}`$. As $`𝒪`$ is non-empty, $`𝒪`$ is a dense open subset of $`V`$. Let $`x𝒪`$ and let $`yx^{}`$. Then
$$J(x)J(y)x=J(y)J(x)x=0.$$
Consequently $`g(J(y)x,J(x)z)=0`$ for all $`z`$. As $`\mathrm{range}(J(x))=x^{}`$, we have $`g(J(y)x,z)=0`$ if $`zx`$. Thus $`R(x,y,y,z)=0`$ if $`zx`$ and $`yx`$. This relation holds for $`x`$ belonging to an open dense subset of $`V`$. Consequently it holds for all $`xV`$. Thus if $`\{e_i\}`$ is an orthonormal basis for $`V`$ and if $`\{i,j,k\}`$ are distinct indices, we have $`R(e_i,e_j,e_j,e_k)=0`$. Thus if $`m=3`$, the only non-zero curvatures are $`R(e_i,e_j,e_j,e_i)`$.
Suppose $`m4`$. Let $`\mathrm{}`$ be a fourth distinct index. Polarization yields
(2.b)
$$R(e_i,e_j,e_{\mathrm{}},e_k)+R(e_i,e_{\mathrm{}},e_j,e_k)=0.$$
We use the relations of Equations (1.a) and (2.b) to see:
$`0`$ $`=`$ $`R(e_i,e_j,e_k,e_{\mathrm{}})+R(e_i,e_k,e_{\mathrm{}},e_j)+R(e_i,e_{\mathrm{}},e_j,e_k)`$
$`=`$ $`R(e_i,e_j,e_k,e_{\mathrm{}})R(e_i,e_k,e_j,e_{\mathrm{}})R(e_i,e_j,e_{\mathrm{}},e_k)`$
$`=`$ $`R(e_i,e_j,e_k,e_{\mathrm{}})+R(e_i,e_j,e_k,e_{\mathrm{}})+R(e_i,e_j,e_k,e_{\mathrm{}})`$
$`=`$ $`3R(e_i,e_j,e_k,e_{\mathrm{}}).`$
Thus the only non-zero curvatures are $`R(e_i,e_j,e_j,e_i)=c_{ij}`$. Consider the new basis
$$e_\nu (\theta ):=\{\begin{array}{ccc}\hfill \mathrm{cos}\theta e_i+\mathrm{sin}\theta e_j& \text{if}\hfill & \nu =i,\hfill \\ \hfill \mathrm{sin}\theta e_i+\mathrm{cos}\theta e_j& \text{if}\hfill & \nu =j,\hfill \\ \hfill e_k& \text{if}\hfill & \nu i,j.\hfill \end{array}$$
We compute $`0=R(e_i(\theta ),e_k,e_k,e_j(\theta ))=\mathrm{cos}\theta \mathrm{sin}\theta \{c_{ik}+c_{jk}\}.`$ It now follows that $`c_{ik}=c_{jk}`$ and consequently $`R`$ has constant sectional curvature.∎
We assume $`r(x)<m1`$ henceforth; thus there exists $`y`$ so $`xy`$ and $`J(x)y=0`$.
###### Lemma 2.2.
Let $`R`$ be a Jacobi-Tsankov algebraic curvature tensor. Assume that $`r(x)<m1`$ $`x`$. Let $`xS(V)`$. Choose $`yS(x^{})`$ so $`J(x)y=0`$. Then:
1. $`J(y)x=0`$ and $`J(x)J(y)=0`$.
2. $`0=J(y)^2+J(x)^24J(x,y)^2`$, $`J(x,y)J(x)=J(y)J(x,y)`$, and
$`J(x)J(x,y)=J(x,y)J(y)`$.
3. Let $`\{x,z_1,z_2\}`$ be an orthonormal set. Suppose that $`J(x)z_1=\lambda _1z_1`$ and $`J(x)z_2=\lambda _2z_2`$ where $`\lambda _1\lambda _2`$. Then $`J(z_1)z_2=0`$.
4. We can choose an orthonormal basis for $`V`$ so that
$$J(x)=\left(\begin{array}{c}A00\hfill \\ 000\hfill \\ 000\hfill \end{array}\right),J(y)=\left(\begin{array}{cc}000\hfill & \\ 0A0\hfill & \\ 000\hfill & \end{array}\right),J(x,y)=\frac{1}{2}\left(\begin{array}{c}0A0\hfill \\ A00\hfill \\ 000\hfill \end{array}\right).$$
###### Proof.
Suppose that $`J(x)y=0`$. We use the relations of Equation (2.a) to compute:
(2.c)
$$\begin{array}{c}J(\mathrm{cos}\theta x+\mathrm{sin}\theta y)J(\mathrm{sin}\theta x+\mathrm{cos}\theta y)y\hfill \\ =J(\mathrm{cos}\theta x+\mathrm{sin}\theta y)\{\mathrm{sin}^2\theta J(x)2\mathrm{sin}\theta \mathrm{cos}\theta J(x,y)+\mathrm{cos}^2\theta J(y)\}y\hfill \\ =J(\mathrm{cos}\theta x+\mathrm{sin}\theta y)\{\mathrm{sin}\theta \mathrm{cos}\theta J(y)x\}\hfill \\ =\{\mathrm{cos}^2\theta J(x)+2\mathrm{sin}\theta \mathrm{cos}\theta J(x,y)+\mathrm{sin}^2\theta J(y)\}\{\mathrm{sin}\theta \mathrm{cos}\theta J(y)x\}\hfill \\ =2\mathrm{sin}^2\theta \mathrm{cos}^2\theta J(x,y)J(y)x+\mathrm{sin}^3\theta \mathrm{cos}\theta J(y)J(y)x\hfill \end{array}$$
and
(2.d)
$$\begin{array}{c}J(\mathrm{sin}\theta x+\mathrm{cos}\theta y)J(\mathrm{cos}\theta x+\mathrm{sin}\theta y)y\hfill \\ =J(\mathrm{sin}\theta x+\mathrm{cos}\theta y)\{\mathrm{cos}^2\theta J(x)+2\mathrm{cos}\theta \mathrm{sin}\theta J(x,y)+\mathrm{sin}^2\theta J(y)\}y\hfill \\ =\{\mathrm{sin}^2\theta J(x)2\mathrm{cos}\theta \mathrm{sin}\theta J(x,y)+\mathrm{cos}^2\theta J(y)\}\{\mathrm{cos}\theta \mathrm{sin}\theta J(y)x\}\hfill \\ =2\mathrm{cos}^2\theta \mathrm{sin}^2\theta J(x,y)J(y)x\mathrm{cos}^3\theta \mathrm{sin}\theta J(y)J(y)x.\hfill \end{array}$$
Subtracting Equation (2.d) from Equation (2.c) yields $`\mathrm{sin}\theta \mathrm{cos}\theta J(y)^2x=0`$ and hence as $`\theta `$ was arbitrary, $`J(y)^2x=0`$. Since $`J(y)`$ is diagonalizable, $`J(y)x=0`$.
We have $`J(x)J(y)y=0`$ and $`J(x)J(y)x=0`$. Let $`z\{x,y\}`$. To complete the proof of Assertion (1), we must show $`J(x)J(y)z=0`$. We compute:
$`0`$ $`=`$ $`J(\mathrm{cos}\theta x+\mathrm{sin}\theta z)J(y)x`$
$`=`$ $`J(y)J(\mathrm{cos}\theta x+\mathrm{sin}\theta z)x`$
$`=`$ $`J(y)\{\mathrm{cos}^2\theta J(x)+2\mathrm{cos}\theta \mathrm{sin}\theta J(x,z)+\mathrm{sin}^2\theta J(z)\}x`$
$`=`$ $`2\mathrm{cos}\theta \mathrm{sin}\theta J(y)J(x,z)x+\mathrm{sin}^2\theta J(y)J(z)x`$
$`=`$ $`\mathrm{cos}\theta \mathrm{sin}\theta J(y)J(x)z+\mathrm{sin}^2\theta J(z)J(y)x`$
$`=`$ $`\mathrm{cos}\theta \mathrm{sin}\theta J(y)J(x)z.`$
We now prove Assertion (2). Because $`J(x,y)x=\frac{1}{2}J(x)y=0`$ and because $`J(x,y)y=\frac{1}{2}J(y)x=0`$, we have:
$`J(x,y)\{\mathrm{sin}\theta x+\mathrm{cos}\theta y\}=0,\text{so}`$
$`J(\mathrm{cos}\theta x+\mathrm{sin}\theta y)\{\mathrm{sin}\theta x+\mathrm{cos}\theta y\}=0.`$
Thus by applying Assertion (1) to the pair $`\{\mathrm{cos}\theta x+\mathrm{sin}\theta y,\mathrm{sin}\theta x+\mathrm{cos}\theta y\}`$,
$`0`$ $`=`$ $`J(\mathrm{cos}\theta x+\mathrm{sin}\theta y)J(\mathrm{sin}\theta x+\mathrm{cos}\theta y)`$
$`=`$ $`\{\mathrm{cos}^2\theta J(x)+2\mathrm{sin}\theta \mathrm{cos}\theta J(x,y)+\mathrm{sin}^2\theta J(y)\}`$
$`\{\mathrm{sin}^2\theta J(x)2\mathrm{sin}\theta \mathrm{cos}\theta J(x,y)+\mathrm{cos}^2\theta J(y)\}`$
$`=`$ $`\mathrm{cos}^2\theta \mathrm{sin}^2\theta \{J(y)^2+J(x)^24J(x,y)^2\}`$
$`+`$ $`2\mathrm{sin}^3\theta \mathrm{cos}\theta \{J(x,y)J(x)J(y)J(x,y)\}`$
$`+`$ $`2\mathrm{sin}\theta \mathrm{cos}^3\theta \{J(x,y)J(y)J(x)J(x,y)\}.`$
Assertion (2) now follows since this identity holds for all $`\theta `$.
Let $`\{x,z_1,z_2\}`$ be an orthonormal set with $`J(x)z_i=\lambda _iz_i`$ where $`\lambda _1\lambda _2`$. To prove Assertion (3), we compute
$`J(x)J(\mathrm{cos}\theta z_1+\mathrm{sin}\theta z_2)z_1=J(x)\{2\mathrm{cos}\theta \mathrm{sin}\theta J(z_1,z_2)+\mathrm{sin}^2\theta J(z_2)\}z_1`$
$`=J(x)\{\mathrm{cos}\theta \mathrm{sin}\theta J(z_1)z_2+\mathrm{sin}^2\theta J(z_2)z_1\}`$
$`=\lambda _2\mathrm{cos}\theta \mathrm{sin}\theta J(z_1)z_2+\lambda _1\mathrm{sin}^2\theta J(z_2)z_1,`$
$`J(\mathrm{cos}\theta z_1+\mathrm{sin}\theta z_2)J(x)z_1=\lambda _1\{2\mathrm{cos}\theta \mathrm{sin}\theta J(z_1,z_2)+\mathrm{sin}^2\theta J(z_2)\}z_1`$
$`=\lambda _1\mathrm{cos}\theta \mathrm{sin}\theta J(z_1)z_2+\lambda _1\mathrm{sin}^2\theta J(z_2)z_1.`$
Assertion (3) now follows since we have
$$\lambda _2J(z_1)z_2=\lambda _1J(z_1)z_2.$$
To prove the final assertion, choose a basis $`\{e_1,\mathrm{},e_r\}`$ for $`\mathrm{Range}(J(x))`$ so that
$$J(x)e_i=\lambda _ie_i\text{for}\lambda _i0.$$
We then have $`J(y)e_i=0`$ and thus $`4J(x,y)^2e_i=\lambda _i^2e_i`$. We set
$$f_i:=2\lambda _i^1J(x,y)e_i.$$
We then compute:
$`f_i,f_j=4\lambda _i^1\lambda _j^1J(x,y)e_i,J(x,y)e_j=4\lambda _i^1\lambda _j^1J(x,y)^2e_i,e_j=\delta _{ij},`$
$`J(x)f_i=2\lambda _i^1J(x)J(x,y)e_i=2\lambda _i^1J(x,y)J(y)e_i=0\text{so}`$
$`f_i\mathrm{ker}(J(x))=\mathrm{Range}(J(x))^{}=\mathrm{Span}\{e_1,\mathrm{},e_r\}^{}.`$
Thus $`\{e_1,\mathrm{},e_r,f_1,\mathrm{},f_r\}`$ is an orthonormal set. Set
$$A=\mathrm{diag}(\lambda _1,\mathrm{}.,\lambda _{\mathrm{}}).$$
Since $`J(y)J(x,y)=J(x,y)J(x)`$, we have $`J(y)f_i=\lambda _if_i`$. Since $`J(x,y)e_i=\frac{1}{2}\lambda _if_i`$ and $`J(x,y)f_i=2\lambda _i^1J(x,y)^2e_i=\frac{1}{2}\lambda _ie_i`$, on $`V_i:=\mathrm{Span}\{e_i,f_i\}`$ we then have on $`\mathrm{Span}\{e_1,\mathrm{},e_{\mathrm{}},f_1,\mathrm{},f_{\mathrm{}}\}`$ that
$$J(x)=\left(\begin{array}{ccc}A\hfill & 0\hfill & \\ 0\hfill & 0\hfill & \end{array}\right),J(y)=\left(\begin{array}{ccc}0\hfill & 0\hfill & \\ 0\hfill & A\hfill & \end{array}\right),J(x,y)=\left(\begin{array}{ccc}0\hfill & \frac{1}{2}A\hfill & \\ \frac{1}{2}A\hfill & 0\hfill & \end{array}\right).$$
Let $`W=\mathrm{Span}\{e_1,\mathrm{},e_{\mathrm{}},f_1,\mathrm{},f_{\mathrm{}}\}^{}`$. Since $`\mathrm{Span}\{e_1,\mathrm{},e_{\mathrm{}},f_1,\mathrm{},f_{\mathrm{}}\}`$ is preserved by $`\{J(x),J(y),J(x,y)\}`$, $`W`$ is preserved by these operators. As $`W\mathrm{Range}\{J(x)\}^{}`$, $`J(x)=0`$ on $`W`$. Thus $`J(y)^2=4J(x,y)^2`$ on $`W`$. If $`J(y)z=\mu z`$ for $`zW`$, then
$$\mu ^3z=J(y)^3z=\frac{1}{4}J(x,y)^2J(y)z=\frac{1}{4}J(x,y)J(x)J(x,y)z.$$
Since $`J(x,y)zW`$, $`J(x)J(x,y)z=0`$ and thus $`\mu =0`$. This shows $`J(y)`$ and hence $`J(x,y)`$ vanishes on $`W`$. Choosing an orthonormal basis for $`W`$ then leads to the desired decomposition. ∎
We continue our study. Let
$`W(x):=x\mathrm{Range}(J(x)).`$
###### Lemma 2.3.
Let $`0R`$ be a Jacobi-Tsankov algebraic curvature tensor. Assume that $`r(x)<m1`$ for all $`xV`$. Let $`xS(V)`$ with $`r(x)0`$. If $`wS(W(x))`$,
1. $`\mathrm{Range}(J(w))W(x)`$ and $`J(w)`$ vanishes on $`W(x)^{}`$.
2. $`J(w)`$ is similar to $`J(x)`$.
3. $`J(x)`$ has exactly two eigenvalues.
###### Proof.
Fix $`wS(W(x))`$. Expand $`w=a_0x+a_iw_i`$ where $`J(x)w_i=\lambda _iw_i`$ for $`\lambda _i0`$. Let $`yS(W(x)^{})`$. We apply Lemma 2.2. As $`y\mathrm{Range}(J(x))`$, $`J(x)y=0`$ so $`J(y)x=0`$. Furthermore since $`J(x)y=0`$, since $`J(x)w_i=\lambda _iw_i`$, and since $`\lambda _i0`$, $`J(y)w_i=0`$. Thus $`J(y)w=0`$ and consequently $`J(w)y=0`$ for all $`yW(x)^{}`$. Thus
$$\mathrm{Range}(J(w))W(x)\text{and}J(w)=0\text{on}W(x)^{}.$$
This proves Assertion (1). Furthermore $`J(x)y=0`$ and $`J(w)y=0`$ implies $`J(x)`$ is similar to $`J(y)`$ and $`J(w)`$ is similar to $`J(y)`$. This establishes Assertion (2).
To show that Assertion (3) is true, we apply Assertion (2) to see
$$\mathrm{Rank}(J(w))=\mathrm{Rank}(J(w)|_{W(x)})=dim(W(x))1=r(x).$$
Suppose $`J(x)`$ has two distinct non-zero eigenvalues $`\lambda _i\lambda _j`$. Then $`J(w_i)w_j=0`$. Since $`J(w_i)w_i=0`$, we would have $`\mathrm{Rank}\{J(w_i)\}r(x)1`$ which is false. Thus $`J(x)=\lambda \mathrm{id}`$ on $`\mathrm{Range}(J(x))`$. This shows that $`J(x)`$ has at most $`2`$ eigenvalues.
To complete the proof, we must show $`\lambda 0`$. Suppose to the contrary that $`\lambda =0`$. This means $`J(x)=0`$. Let $`zS(V)`$. Choose $`yS(x^{}z^{})`$. Then $`J(x)y=0`$ so $`J(x)`$ is similar to $`J(y)`$ and thus $`J(y)=0`$. Since $`J(y)z=0`$, $`J(y)`$ is similar to $`J(z)`$ and $`J(z)=0`$. This implies $`J0`$ and hence $`R=0`$ which is false. Thus $`J(x)0`$ and $`\lambda 0`$. ∎
###### Lemma 2.4.
Let $`0R`$ be a Jacobi-Tsankov algebraic curvature tensor. Assume that $`r(x)<m1`$ for all $`xV`$.
1. $`r(x)=1`$.
2. $`R`$ is Osserman.
###### Proof.
Choose $`yS(x^{})`$ with $`J(x)y=0`$. Let $`e_0=x`$ and $`f_0=y`$. Let $`\lambda `$ be the non-zero eigenvalue for $`J(x)`$ and let $`r=r(x)`$. Choose an orthonormal basis $`\{e_0,\mathrm{},e_r,f_0,\mathrm{},f_r,g_1,\mathrm{},g_{\mathrm{}}\}`$ for $`V`$ so that
$$\begin{array}{ccc}J(e_0)e_j=\lambda (1\delta _{0j})e_j\hfill & J(e_0)f_j=0,\hfill & J(e_0)g_k=0,\hfill \\ J(f_0)f_j=\lambda (1\delta _{0j})f_j\hfill & J(f_0)e_j=0,\hfill & J(f_0)g_k=0,\hfill \\ J(e_0,f_0)e_j=\frac{1}{2}\lambda (1\delta _{0j})f_j\hfill & J(e_0,f_0)f_j=\frac{1}{2}\lambda (1\delta _{0j})e_j\hfill & J(e_0,f_0)g_k=0.\hfill \end{array}$$
As $`J(e_1)`$ preserves $`\mathrm{Span}\{e_0,\mathrm{},e_r\}=W(e_0)`$, as $`\lambda `$ is an eigenvalue of multiplicity $`r`$ for $`J(e_1)`$ on $`W(e_0)`$, that $`J(e_1)`$ vanishes on $`W(e_0)^{}`$, and as $`J(e_1)e_1=0`$,
$$J(e_1)e_j=\lambda (1\delta _{1j})e_j,J(e_1)f_j=0,J(e_1)g_k=0.$$
Let $`\xi =\frac{1}{\sqrt{2}}(e_0+f_0)`$. Then $`J(\xi )=\frac{1}{2}\{J(e_0)+J(f_0)+2J(e_0,f_0)\}`$. We show $`r=1`$ and prove Assertion (1) by deriving the following contradiction:
$`J(\xi )e_2=\frac{1}{2}\lambda (e_2+f_2),J(\xi )f_2=\frac{1}{2}\lambda (e_2+f_2),`$
$`J(\xi )J(e_1)f_2=0,J(e_1)J(\xi )f_2=\frac{1}{2}\lambda ^2e_2.`$
Fix $`eS(V)`$. Consider the $`2`$-plane
$$\pi :=\mathrm{Span}\{e,\mathrm{Range}(J(e))\}.$$
Decompose $`xS(V)`$ in the form $`x=\mathrm{cos}\theta e_1+\mathrm{sin}\theta f_1`$ for $`\theta [0,\frac{\pi }{2}]`$, $`e_1S(\pi )`$ and $`f_1S(\pi ^{})`$; $`e_1`$ is not unique if $`\theta =\frac{\pi }{2}`$ and $`f_1`$ is not unique if $`\theta =0`$. As $`e_1\pi `$, $`\mathrm{Range}(J(e_1))\pi `$ so
$$\pi =\mathrm{Span}\{e_1,\mathrm{Range}(J(e_1))\}.$$
Since $`f_1\pi `$, Lemma 2.2 pertains. As $`J(f_1)`$ is similar to $`J(e_1)`$ and also to $`J(e)`$, $`\lambda (e_1)=\lambda (f_1)=\lambda (e)`$. By Lemma 2.2, we can extend $`\{e_1,f_1\}`$ to an orthonormal set $`\{e_1,e_2,f_1,f_2\}`$ so that
$$\begin{array}{ccc}J(e_1)e_2=\lambda (e)e_2,\hfill & J(f_1)e_2=0,\hfill & J(e_1,f_1)e_2=\frac{1}{2}\lambda (e)f_2,\hfill \\ J(e_1)f_2=0,\hfill & J(f_1)f_2=\lambda (e)f_2,\hfill & J(e_1,f_1)f_2=\frac{1}{2}\lambda (e)e_2.\hfill \end{array}$$
We may now compute
$`J(x)(\mathrm{cos}\theta e_2+\mathrm{sin}\theta f_2)`$
$`=`$ $`(\mathrm{cos}^2\theta J(e_1)+2\mathrm{cos}\theta \mathrm{sin}\theta J(e_1,f_1)+\mathrm{sin}^2\theta J(f_1))(\mathrm{cos}\theta e_2+\mathrm{sin}\theta f_2)`$
$`=`$ $`\lambda (e)(\mathrm{cos}^3\theta e_2+\mathrm{cos}^2\theta \mathrm{sin}\theta f_2+\mathrm{cos}\theta \mathrm{sin}^2\theta e_2+\mathrm{sin}^3\theta f_2)`$
$`=`$ $`\lambda (e)(\mathrm{cos}\theta e_2+\mathrm{sin}\theta f_2).`$
This implies that $`\lambda (x)=\lambda (e)`$ as desired. ∎
###### Proof of Theorem 1.3 (3).
Suppose $`0R`$ is Jacobi-Tsankov. If $`r(x)=m1`$ for any vector $`xS(V)`$, then $`R`$ has constant sectional curvature by Lemma 2.1. On the other hand, if $`r(x)<m1`$ for every point $`xS(V)`$, then $`R`$ is Osserman and $`\mathrm{Rank}\{J(x)\}=1`$ for every $`xS(V)`$ by Lemma 2.4. Work of Chi then shows $`R=R_\mathrm{\Theta }`$ for some Hermitian almost complex structure on $`V`$. ∎
## 3. The Geometric Setting
###### Proof of Theorem 1.2.
Let $`(M,g)`$ be a Riemannian manifold with $`m3`$. Suppose first that $`J(x)J(y)=J(y)J(x)`$ for $`x,yT_PM`$. Then $`R_P=0`$. Consequently $`(M,g)`$ is flat; this proves Assertion (1).
Suppose $`(M,g)`$ is Jacobi-Tsankov and $`m3`$. If $`m`$ is not even, then $`R_P`$ has constant sectional curvature at each point $`P`$ of $`M`$ and hence $`(M,g)`$ has constant sectional curvature globally. Let $`𝒪`$ be open subset of points $`PM`$ so that there exists a unit tangent vector $`x(P)`$ with $`r(x(P))=m1`$. Then $`g|_𝒪`$ has constant sectional curvature. Thus $`R=cR_0`$ on $`𝒪`$ and hence $`R=cR_0`$ on the closure of $`𝒪`$. Thus $`𝒪`$ is an open and closed subset of $`M`$ and hence all of $`M`$.
Thus if $`(M,g)`$ does not have constant sectional curvature, $`r(x)<m1`$ for all $`xS(M,g)`$. Suppose $`(M,g)`$ is not flat; there is then a point $`PM`$ and a tangent vector $`xS(T_PM)`$ so that $`r(x)=1`$. Let $`𝒰`$ be a small open contractable of $`P`$. Then $`r(x)=1`$ for $`xS(𝒰)`$. Thus there is an almost complex structure $`\mathrm{\Theta }(Q)`$ defined on $`T_Q`$ for every $`QU`$ so that $`R=\lambda R_\mathrm{\Theta }`$. Furthermore, the almost complex structure is uniquely determined up to sign. After a bit of technical fuss, one can see that $`\mathrm{\Theta }`$ can be chosen to vary smoothly with $`Q`$, at least locally; global questions are irrelevant to our argument. The metric in question is Einstein and thus $`\rho (x,x)=\lambda (x)`$ is constant. Thus $`(M,g)`$ is globally Osserman. The work of Chi then implies $`(M,g)`$ is a complex space form - i.e. isometric to $`^j`$ or its negative curvature dual. This is false as these manifolds have $`\mathrm{Rank}(J(x))2`$. This completes the proof.∎
## Acknowledgments
The research of M. Brozos-Vázquez was partially supported by project BFM 2003-02949 (Spain). The research of P. Gilkey was partially supported by the Max Planck Institute for the Mathematical Sciences (Leipzig, Germany). It is a pleasure for both authors to acknowledge helpful conversations with C. Dunn. |
warning/0507/hep-ex0507049.html | ar5iv | text | # Recent measurement of Δ𝐺/𝐺 at COMPASS
## Abstract
We present a preliminary measurement of the gluon polarization $`\mathrm{\Delta }G/G`$ in the nucleon, based on the spin asymmetry of quasi-real photoproduction events for which a pair of large transverse momentum hadrons is produced. The data were obtained by the COMPASS experiment at CERN using a 160 GeV polarized muon beam scattered on a large polarized <sup>6</sup>LiD target. The preliminary helicity asymmetry for the selected events is $`A_{}/D=0.002\pm 0.019(\mathrm{stat}.)\pm 0.003(\mathrm{exp}.\mathrm{syst}.)`$. From this value, a leading order analysis based on the model of PYTHIA leads to the gluon polarization in the nucleon $`\mathrm{\Delta }G/G(x_g=0.095,\mu ^2=3\mathrm{GeV}^2)=0.024\pm 0.089(\mathrm{stat}.)\pm 0.057(\mathrm{syst}.)`$. This value is consistent with parameterizations obtained from QCD fits to the $`g_1`$ data, with a first moment $`\mathrm{\Delta }G_0^1\mathrm{\Delta }G(x)𝑑x1`$, at the same scale.
###### Keywords:
gluons polarization nucleon
The decomposition of the nucleon spin in the contributions from its constituents has been a central topic of investigation in polarized lepton-nucleon scattering in the last 20 years. The EMC measurement of the proton spin structure Ashman et al. (1988) has shown that only 20 to 30 % of the proton spin could be attributed to the total quark spin $`\mathrm{\Delta }\mathrm{\Sigma }`$, in contrast to the 60 % expected in the naive quark-parton model. In inclusive lepton-nucleon scattering the contribution of the gluon spin $`\mathrm{\Delta }G`$ to the nucleon spin can only be measured indirectly by studying the $`Q^2`$ dependence of the polarized spin-structure functions in QCD. Fits at next-to-leading order (NLO) provide evaluations of $`\mathrm{\Delta }G`$ which are of the order of 0.50. The precision of these fits is strongly limited by the small $`Q^2`$ range covered by the data at any value of $`x`$, a situation resulting from the lack of a polarized lepton-nucleon collider. In addition, the shape of $`\mathrm{\Delta }G(x)`$ at a fixed $`Q^2`$ used as reference, has to be provided as an input parameterization. It varies considerably between the different analyses and is only poorly constrained by the results of the fits. A direct measurement of the gluon polarization $`\mathrm{\Delta }G(x)/G(x)`$ can be obtained from the helicity asymmetry of the photon-gluon fusion (PGF, $`\gamma ^{}gq\overline{q}`$) cross-section, which constitutes an important part of the experimental program of COMPASS
The COMPASS experiment Mallot (2004) is located at the M2 beam line of the CERN SPS, which provides a 160 GeV $`\mu ^+`$ beam, with a natural polarization of $`76\pm 5\%`$. The target consists in an upstream and a downstream cell, longitudinally polarized in opposite directions. Typical target polarizations of $`50.0\pm 2.5\%`$ are obtained. The forward spectrometer is divided in two stages allowing the reconstruction of the scattered muon and of the produced hadrons in broad momentum and angular ranges. The trigger system provides efficient tagging down to $`Q^2=0.002`$ GeV<sup>2</sup>.
The present analysis focuses on the data collected during the 2002 and 2003 runs. We only consider quasi-real photoproduction events ($`Q^2<1`$ GeV<sup>2</sup>), in which at least two charged hadrons are associated to the primary vertex in addition to the incident and the scattered muons. The fraction of PGF is enhanced by asking the two leading hadrons to have a large transverse momentum: $`p_T^{h1}>0.7`$ GeV, $`p_T^{h2}>0.7`$ GeV and $`(p_T^{h1})^2+(p_T^{h2})^2>2.5`$ GeV<sup>2</sup>. In total, around 350,000 events are selected. For this high $`p_T`$ sample, the measured helicity asymmetry (defined as in Ageev et al. (2005)) is
$$\frac{A_{}}{D}=0.002\pm 0.019(stat)\pm 0.003(exp.syst),$$
(1)
where the quoted systematic error accounts for the false asymmetries related to the apparatus. Other sources of systematic errors, including the error on the beam and target polarizations, are only a few percents of the (small) measured asymmetry, and are therefore negligible. The 2004 data are currently under analysis, and represent the same amount of data as 2002 and 2003 altogether.
A Monte-Carlo simulation is needed to extract the gluon polarization from the high $`p_T`$ asymmetry. The selected sample of high $`p_T`$ events covers the transition region ranging from photoproduction ($`Q^20`$) up to DIS ($`Q^21`$ GeV<sup>2</sup>). For this reason, we chose PYTHIA as an event generator because it provides a model for the lepton-nucleon interactions Friberg and Sjostrand (2000) at low $`Q^2`$. Two different kinds of processes are generated. In the so-called direct processes, the virtual photon takes part in the hard partonic interaction. In the resolved processes, it fluctuates to a hadronic state, from which a parton is extracted. This parton then interacts with a parton from the nucleon. At $`Q^2<1`$, the resolved processes constitute half of the high $`p_T`$ sample. Their contribution falls to about 10% for $`Q^2>1`$ GeV<sup>2</sup> and becomes negligible for $`Q^2>2`$ GeV<sup>2</sup>. Note that the analysis requires the factorization between the hard and soft parts of the reaction, hence the presence of a hard scale $`\mu ^2`$. As $`Q^2<1`$ GeV<sup>2</sup>, the scale is provided by the transverse momentum of the partons involved in the reaction. Events for which no hard scale can be found are classified as low $`p_T`$.
The generated events are tracked through a GEANT description of the COMPASS spectrometer, and processed using the same reconstruction program as for real data. Then, the Monte-Carlo sample of high $`p_T`$ events is selected through the same cuts.
Only one parameter of PYTHIA had to be changed to reach a good agreement with the data: the width of the Gaussian distribution of intrinsic transverse momentum of partons within the resolved virtual photon (PARP(99)) was decreased from 1 GeV/c to 0.5 GeV/c. Fig. 1 presents a comparison between the simulated and real data samples of high $`p_T`$ events. Fig. 2 shows how the Monte-Carlo sample of high $`p_T`$ events divides into the various PYTHIA subprocesses.
The high $`p_T`$ asymmetry can be written in terms of the contributions of the different processes:
$`{\displaystyle \frac{A_{}}{D}}=R_{PGF}\widehat{a}_{LL}^{PGF}/D{\displaystyle \frac{\mathrm{\Delta }G}{G}}`$ $`+R_{QCDC}\widehat{a}_{LL}^{QCDC}/DA_1`$
$`+{\displaystyle \underset{f,f^{}=u,d,s,\overline{u},\overline{d},\overline{s},g}{}}R_{ff^{}}\widehat{a}_{LL}^{ff^{}}\left({\displaystyle \frac{\mathrm{\Delta }f}{f}}\right)^d\left({\displaystyle \frac{\mathrm{\Delta }f^{}}{f^{}}}\right)^\gamma .`$ (2)
Note that we have neglected the small contributions of the leading process $`\gamma ^{}qq`$ and of the low $`p_T`$ scattering events because there is no hard scale allowing a perturbative treatment of these subprocesses (low transverse momentum, $`Q^2<1`$ GeV<sup>2</sup> events). As can be seen on Fig. 2, the fraction of photon-gluon fusion events is $`R_{PGF}=0.31`$. The analyzing power $`\widehat{a}_{LL}^{PGF}`$ is the helicity asymmetry of the $`\mu g\mu ^{}q\overline{q}`$ scattering cross-section, $`\widehat{a}_{LL}^{PGF}d\mathrm{\Delta }\sigma _{PGF}^{\mu g}/d\sigma _{PGF}^{\mu g}`$. It is calculated from the kinematic variables of the partonic reaction for each PGF event in the high $`p_T`$ Monte-Carlo sample. Averaging over the PGF events, we obtain $`\widehat{a}_{LL}^{PGF}/D=0.933`$. The contribution of the PGF process to the high $`p_T`$ asymmetry is thus $`0.292\times \frac{\mathrm{\Delta }G}{G}`$. The contribution of the QCD Compton events is calculated in the same way to be 0.0063, using a fit on the world data for the virtual-photon deuteron asymmetry $`A_1^d`$.
Resolved photon subprocesses involve either a quark or a gluon from the nucleon. In the latter case, they are sensitive to the gluon polarization $`\mathrm{\Delta }G/G`$, and contribute to the signal. The analyzing powers $`\widehat{a}_{LL}^{ff^{}}`$ are calculated in pQCD at leading order Bourrely et al. (1989), and are positive for all relevant channels. The polarizations $`(\mathrm{\Delta }f/f)^d`$ of the $`u`$, $`d`$ and $`s`$ quarks in the deuteron are calculated using the unpolarized parton distribution functions from GRV98, and the polarized parton distribution functions from GRSV2000 Gluck et al. (2001a), at leading order. The polarizations of quarks and gluons in the virtual photon $`(\mathrm{\Delta }f/f)^\gamma `$ are unknown as the polarized PDFs of the virtual photon have not yet been measured. Nevertheless, theoretical considerations provide a minimum and a maximum value for each $`\mathrm{\Delta }f^\gamma `$ Gluck et al. (2001b), called the minimal and maximal scenarios. The total contribution of the resolved photon processes to the high $`p_T`$ asymmetry, which ranges between $`0.000+0.012\times \mathrm{\Delta }G/G`$ and $`0.002+0.078\times \mathrm{\Delta }G/G`$, will be taken into account in the systematic error on $`\mathrm{\Delta }G/G`$.
The tuning of the PYTHIA parameters relevant to this analysis is an important source of systematic errors. This error was estimated by scanning these parameters independently over a range where the agreement between the simulation and real data remains reasonable. This resulted in several values for $`\mathrm{\Delta }G/G`$, all based on the same high $`p_T`$ asymmetry, Eq. (1). The value of $`\mathrm{\Delta }G/G`$ appears to depend predominantly on the width of the intrinsic transverse momentum distribution for the partons in the photon. For instance, varying this parameter between 0.1 and 1 GeV/c results in a 30% variation of the fraction of photon-gluon fusion $`R_{PGF}`$. NLO effects seem to be small: varying the scale and (de)activating parton showers does not affect the result.
Using Eq. (2) to extract the gluon polarization from the high $`p_T`$ asymmetry, Eq. (1), we finally get:
$$\frac{\mathrm{\Delta }G}{G}(x_g=0.095,\mu ^2=3\mathrm{GeV}^2)=0.024\pm 0.089(stat.)\pm 0.057(syst.).$$
(3)
This result was compared to the recent distributions of $`\mathrm{\Delta }G(x)/G(x)`$ from AAC Hirai et al. (2004) and LSS Leader et al. (2002). These two distributions, which strongly differ in shape, are almost equal at $`x_g=0.095`$ and are compatible with our result within $`1.5\sigma `$. The first moments of $`\mathrm{\Delta }G`$ are equal to $`0.8\pm 0.56`$ and $`1.1\pm 0.52`$ for the AAC and LSS fits, respectively. The fraction of PGF can also be enhanced by selecting events with charmed hadrons ($`D^0`$ and $`D^{}`$), instead of high $`p_T`$ hadrons. The results obtained with this method, which is much less model-dependent but suffers from low statistics, will be produced soon. |
warning/0507/math0507484.html | ar5iv | text | # A lower bound for average values of dynamical Green’s functions
## 1. Introduction
### 1.1. Background
Throughout this paper, $`K`$ will denote a field endowed with an absolute value, which can be either archimedean or non-archimedean. Let $`\phi K(T)`$ be a rational function of degree $`d2`$, and let $`g_\phi `$ be the normalized Arakelov-Green’s function associated to $`\phi `$ (see (1.6) below for a definition). In several different situations, one is led to consider lower bounds for $`g_\phi `$-discriminant sums of the form
$$D_\phi (z_1,\mathrm{},z_N)=\underset{\begin{array}{c}1i,jN\\ ij\end{array}}{}g_\phi (z_i,z_j).$$
Since $`g_\phi (z,w)`$ is bounded below, one knows that $`D_\phi (z_1,\mathrm{},z_N)CN^2`$ for some constant $`C>0`$. The main result of this paper is the following stronger estimate:
###### Theorem 1.1.
There is an effective constant $`C>0`$, depending on $`\phi `$ and $`K`$, such that if $`N2`$ and $`z_1,\mathrm{},z_N`$ are distinct points of $`^1(K)`$, then
(1.2)
$$D_\phi (z_1,\mathrm{},z_N)CN\mathrm{log}N.$$
We will first give some historical context for this result, and then we will explain some of its applications.
The first estimate of this kind appeared in the work of K. Mahler , who proved a useful inequality between the discriminant and Mahler measure of a polynomial. Specifically, suppose that $`f(x)=a_0x^N+a_1x^{N1}+\mathrm{}+a_N=a_0(x\alpha _1)\mathrm{}(x\alpha _N)`$ is a polynomial with complex coefficients and degree $`N2`$, let $`\mathrm{Disc}(f)=a_0^{2N2}_{i<j}(\alpha _i\alpha _j)^2`$ be the discriminant of $`f`$, and let $`M(f)=|a_0|_{j=1}^N\mathrm{max}(1,|\alpha _j|)`$ be the Mahler measure of $`f`$. Mahler’s inequality states that
(1.3)
$$|\mathrm{Disc}(f)|N^N\{M(f)\}^{2N2}.$$
The inequality (1.3) can be used, for example, to give a short proof of Bilu’s equidistribution theorem (see e.g. ). Mahler’s proof of (1.3) uses Hadamard’s inequality to estimate the determinant of a suitable Vandermonde matrix. Essentially the same argument gives rise to the following reformulation of (1.3). For $`z,w^1()`$ with $`zw`$, define
$$g(z,w)=\{\begin{array}{cc}\mathrm{log}|zw|+\mathrm{log}^+|z|+\mathrm{log}^+|w|\hfill & z,w\mathrm{}\hfill \\ \mathrm{log}^+|z|\hfill & w=\mathrm{}\hfill \\ \mathrm{log}^+|w|\hfill & z=\mathrm{}.\hfill \end{array}$$
Then if $`N2`$ and $`z_1,\mathrm{},z_N^1()`$ are distinct points, we have
(1.4)
$$\underset{ij}{}g(z_i,z_j)N\mathrm{log}N.$$
Note that $`g(z,w)\mathrm{log}2`$ for all $`zw`$, and thus (1.4) gives a significant improvement over the trivial lower bound of $`(\mathrm{log}2)N(N1)`$.
The reformulation (1.4) of inequality (1.3) is strongly reminiscent of the following result of Elkies (see \[12, §VI, Theorem 5.1\] and \[2, Appendix\]). Let $`E/`$ be an elliptic curve, and let $`g(z,w)`$ be the normalized Arakelov-Green’s function associated to the Haar measure $`\mu `$ on $`E()`$. Then there exists a constant $`C>0`$ such that if $`N2`$ and $`z_1,\mathrm{},z_N`$ are distinct points of $`E()`$, then
(1.5)
$$\underset{ij}{}g(z_i,z_j)CN\mathrm{log}N.$$
The result proved in is actually a generalization of (1.5) to an arbitrary compact Riemann surface $`X`$ of genus at least 1, where Haar measure is replaced by the canonical volume form on $`X`$. Lang remarks on page 152 that Elkies’ argument can be used to prove a similar theorem for an arbitrary compact Riemannian manifold.
Bounds of the form (1.5) are used in Arakelov theory to prove the existence of small sections of large powers of arithmetically ample metrized line bundles. In addition, the bound (1.5) is used in a series of papers by Hindry and Silverman to obtain lower bounds for the canonical height of a non-torsion $`k`$-rational point on an elliptic curve $`E`$ over a number field $`k`$, as well as upper bounds for the number of $`k`$-rational torsion points on $`E`$. Further applications of (1.5) to canonical heights on elliptic curves can be found in . The proof of (1.5) for a general Riemann surface $`X`$ uses the spectral theory of the Laplacian operator; when $`X=E`$ is an elliptic curve, this amounts to Fourier analysis on the complex torus $`/\mathrm{\Lambda }`$. An analogue of (1.5) for metrized graphs is proved in .
Theorem 1.1 is a simultaneous generalization of (1.4) and (1.5) which applies to Arakelov-Green’s functions attached to an arbitrary dynamical system on $`^1`$. It works equally well over archimedean and non-archimedean fields. The case where $`\phi (T)=T^2`$ and $`K=`$ corresponds to Mahler’s result, and the case where $`\phi (T)`$ is the degree 4 “Lattès map” corresponding to multiplication by 2 on the quotient of an elliptic curve over $``$ by the involution $`PP`$ corresponds to (1.5). Since the analytic tools employed by Elkies do not seem to be available in the context of the dynamical systems attached to arbitrary rational functions, our proof of (1.2) is based on a suitable modification of Mahler’s original approach. In particular, we obtain a new and much more elementary proof of (1.5), albeit with different constants.
The possibility of proving (1.2) by a Mahler-style argument was suggested to us by Lemma 4.1 of R. Benedetto’s paper , which gives a bound of the form (1.2) in the special case where $`\phi `$ is a polynomial and $`z_1,\mathrm{},z_N`$ belong to the filled Julia set of $`\phi `$. Note that the case where $`\phi `$ is a polynomial is simpler than the general case, due to the fact that polynomial maps have a superattracting fixed point at infinity.
When $`K`$ is non-archimedean, the $`CN\mathrm{log}N`$ term in (1.5) can be replaced by $`CN`$, see \[11, Lemma 2.1\]. For general rational maps over a non-archimedean field, however, the bound (1.2) is optimal. This can be seen using Lemma 3.4(a) and Remark 4.3 of .
### 1.2. Definition and properties of $`g_\phi `$
In order to define the Arakelov-Green’s function $`g_\phi `$ which appears in the statement of (1.2), we begin by recalling some definitions and results from .
Write $`\phi `$ in the form
$$\phi ([z_0:z_1])=[F_1(z_0,z_1):F_2(z_0,z_1)]$$
for some homogeneous polynomials $`F_1,F_2K[x,y]`$ of degree $`d2`$ with no common linear factor over $`\overline{K}`$. The polynomials $`F_1,F_2`$ are uniquely determined by $`\phi `$ up to multiplication by a common scalar $`cK^{}`$. The mapping
$$F=(F_1,F_2):𝔸^2(K)𝔸^2(K)$$
is a lifting of $`\phi `$ to $`𝔸^2`$.
Let $`\mathrm{Res}(F):=\mathrm{Res}(F_1,F_2)`$ denote the homogeneous resultant of the polynomials $`F_1`$ and $`F_2`$ (see \[16, §5.8\],\[7, §6\]). Since $`F_1`$ and $`F_2`$ have no common linear factor over $`\overline{K}`$, we have $`\mathrm{Res}(F)0`$.
Let $`(z_0,z_1):=\mathrm{max}\{|z_0|,|z_1|\}`$. For $`z=(z_0,z_1)K^2\backslash \{0\}`$, define the homogeneous local dynamical height $`\widehat{H}_F:K^2\backslash \{0\}`$ by
$$\widehat{H}_F(z):=\underset{n\mathrm{}}{lim}\frac{1}{d^n}\mathrm{log}F^{(n)}(z).$$
By convention, we define $`\widehat{H}_F(0,0):=\mathrm{}`$. It is proved in that the limit $`lim_n\mathrm{}\frac{1}{d^n}\mathrm{log}F^{(n)}(z)`$ exists for all $`zK^2\backslash \{0\}`$, and that $`\frac{1}{d^n}\mathrm{log}F^{(n)}(z)`$ converges uniformly on $`K^2\backslash \{0\}`$ to $`\widehat{H}_F(z)`$. (Note that the results in are stated in the special case $`K=_v`$, but the proofs extend without modification to the more general case considered here.) The definition of $`\widehat{H}_F`$ is independent of the norm used to define it.
For $`z=(z_0,z_1),w=(w_0,w_1)K^2`$, put
$$zw:=z_0w_1z_1w_0.$$
When $`z,wK^2`$ are linearly independent over $`K`$, we define
$$G_F(z,w):=\mathrm{log}|zw|+\widehat{H}_F(z)+\widehat{H}_F(w)r(F),$$
where $`r(F)=\frac{1}{d(d1)}\mathrm{log}|\mathrm{Res}(F)|`$.
An explanation for the appearance of the constant term $`r(F)`$ will be given shortly.
According to , the function $`G_F`$ is doubly scale-invariant, in the sense that if $`\alpha ,\beta K^{}`$, then
$$G_F(\alpha z,\beta w)=G_F(z,w).$$
In addition, for all $`\gamma K^{}`$, we have
$$G_{\gamma F}(z,w)=G_F(z,w).$$
In particular, $`G_F`$ descends to a well-defined function $`g_\phi (z,w)`$ on $`^1(K)`$: for $`z,w^1(K)`$ and any lifts $`\stackrel{~}{z},\stackrel{~}{w}K^2`$,
(1.6)
$$g_\phi (z,w)=\mathrm{log}|\stackrel{~}{z}\stackrel{~}{w}|+\widehat{H}_F(\stackrel{~}{z})+\widehat{H}_F(\stackrel{~}{w})r(F).$$
If $`zw`$ then the right-hand side of (1.6) is finite; if $`z=w`$ then we define $`g_\phi (z,z):=+\mathrm{}`$.
###### Remark 1.7.
In , it is proved in the case $`K=_v`$ that the function $`g_\phi (z,w)`$ is a normalized Arakelov-Green’s function for the canonical measure $`\mu _\phi `$ on the Berkovich analytic space $`_{\mathrm{Berk}}^1`$ over $`_v`$; we refer to for a precise explanation of these objects. Briefly, the fact that $`g_\phi (z,w)`$ is a normalized Arakelov-Green’s function means that it is the unique solution to the differential equation
(1.8)
$$\mathrm{\Delta }_zg_\phi (z,w)=\delta _w\mu _\phi $$
subject to the normalization condition
(1.9)
$$g_\phi (z,w)𝑑\mu _\phi (z)𝑑\mu _\phi (w)=0.$$
When $`v`$ is archimedean, $`_{\mathrm{Berk}}^1=^1()`$ and $`\mu _\phi `$ is the canonical probability measure on $`^1()`$ introduced by Lyubich and Freire-Lopes-Mañé whose support is equal to the Julia set of $`\phi `$. For our purposes, one can take (1.8) as the definition of $`\mu _\phi `$. The constant $`r(F)=\frac{1}{d(d1)}\mathrm{log}|\mathrm{Res}(F)|`$ which appears in the definition (1.6) of $`g_\phi `$ is chosen precisely so that (1.9) holds; see \[3, Corollary 4.13\].
###### Remark 1.10.
If $`K`$ is non-archimedean and $`\phi `$ has good reduction, then $`g_\phi `$ is nonnegative and Theorem 1.1 is trivial.
### 1.3. The homogeneous transfinite diameter of the filled Julia set
In order to explain some of the applications of (1.2), we recall some additional facts from .
Let $`EK^2`$ be a bounded set. By analogy with the classical transfinite diameter, define
$$d_n^0(E):=\underset{z_1,\mathrm{},z_nE}{sup}\left(\underset{ij}{}|z_iz_j|\right)^{\frac{1}{n(n1)}}.$$
By , the sequence of nonnegative real numbers $`d_n^0(E)`$ is non-increasing. In particular, the quantity $`d_{\mathrm{}}^0(E):=lim_n\mathrm{}d_n^0(E)`$ is well-defined. We call $`d_{\mathrm{}}^0(E)`$ the homogeneous transfinite diameter of $`E`$.
The *filled Julia set* $`K_F`$ of $`F`$ in $`K^2`$ is the set of all $`zK^2`$ for which the iterates $`F^{(n)}(z)`$ remain bounded. Clearly $`F^1(K_F)=K_F`$, and the same is true for each $`F^{(n)}`$. It is proved in that $`K_F`$ is a closed and bounded subset of $`K^2`$, and that
$$K_F=\{zK^2:\widehat{H}_F(z)0\}.$$
The following theorem from is a generalization from $`K=`$ to an arbitrary valued field of a formula of DeMarco :
###### Theorem 1.11.
If $`K`$ is algebraically closed, then
(1.12)
$$d_{\mathrm{}}^0(K_F)=|\mathrm{Res}(F)|^{1/d(d1)}.$$
The proof of this result given in is rather indirect, and requires the development of a lot of capacity-theoretic machinery, including a detailed analysis of the relationship between the homogeneous transfinite diameter and the homogeneous sectional capacity. The principal application of Theorem 1.11 given in is an adelic equidistribution theorem for points of small dynamical height with respect to a rational map defined over a number field. This application requires only the inequality
(1.13)
$$d_{\mathrm{}}^0(K_F)|\mathrm{Res}(F)|^{1/d(d1)}.$$
Theorem 1.1 furnishes a simpler and more direct proof of (1.13) (which does not require that the field $`K`$ be algebraically closed). Indeed, if $`z_1,\mathrm{},z_N`$ are chosen to belong to $`K_F`$, then $`\widehat{H}_F(z_i)0`$ and (1.2) yields
$$\frac{1}{N(N1)}\underset{ij}{}\mathrm{log}|z_iz_j|C\frac{\mathrm{log}N}{N1}r(F).$$
Passing to the limit as $`N`$ tends to infinity and exponentiating gives (1.13).
### 1.4. Application to canonical heights
We now consider a global application of (1.2). Let $`k`$ be a number field, let $`\phi k(T)`$ be a rational map with coefficients in $`k`$, and let $`\widehat{h}_\phi `$ denote the Call-Silverman canonical height function attached to $`\phi `$; this can be defined for $`P=[z_0:z_1]^1(k)`$ by
$$\widehat{h}_\phi (P)=\underset{vM_k}{}\frac{[k_v:_v]}{[k:]}\widehat{H}_{F,v}(z_0,z_1).$$
Here $`\widehat{H}_{F,v}`$ denotes the function $`\widehat{H}_F`$ over the completion $`k_v`$ (or over $`_v`$).
The proof of the following result will be given in §3.
###### Theorem 1.14.
There exist constants $`A,B>0`$, depending only on $`\phi `$ and $`k`$, such that for any finite extension $`k^{}/k`$ with $`D=[k^{}:]`$, we have
(1.15)
$$\mathrm{\#}\{P^1(k^{}):\widehat{h}_\phi (P)\frac{A}{D}\}BD\mathrm{log}D.$$
###### Remark 1.16.
In the special case $`\phi (T)=T^2`$, the function $`\widehat{h}_\phi `$ is the standard logarithmic Weil height, and (1.15) implies that there is a constant $`C>0`$ such that if $`\alpha (k^{})^{}`$ has $`h(\alpha )>0`$ (i.e., $`\alpha `$ is not a root of unity), then
(1.17)
$$h(\alpha )\frac{C}{D^2\mathrm{log}D}.$$
Indeed, if $`M`$ is the largest integer such that $`(M1)h(\alpha )=h(\alpha ^{M1})\frac{A}{D}`$, then $`\{1,\alpha ,\alpha ^2,\mathrm{},\alpha ^{M1}\}\{P^1(k^{}):h(P)\frac{A}{D}\}`$, and thus $`MBD\mathrm{log}D`$. By the maximality of $`M`$, we have $`Mh(\alpha )>\frac{A}{D}`$, and therefore $`h(\alpha )>\frac{A}{B}\frac{1}{D^2\mathrm{log}D}`$. Of course, much stronger results are known in connection with Lehmer’s problem (see e.g. ). Similarly, applying (1.15) to a Lattès map (for which $`\widehat{h}_\phi `$ is the Néron-Tate canonical height on the elliptic curve $`E/k`$) and using the quadraticity of the canonical height gives the bound
(1.18)
$$\widehat{h}_E(P)\frac{C}{D^3\mathrm{log}^2D}$$
for all non-torsion points $`PE(k^{})`$. This is precisely the same bound for the elliptic Lehmer problem obtained by Masser . Unfortunately, for a general rational map, Theorem 1.14 does not seem to imply a Lehmer-type estimate such as (1.17) or (1.18) which is polynomial in $`1/D`$.
Finally, we note that one can deduce from Theorem 1.14 the following result concerning fields of definition for preperiodic points of $`\phi `$:
###### Corollary 1.19.
There exists a constant $`C`$ such that if $`P_1,\mathrm{},P_N`$ are distinct preperiodic points of $`\phi `$ defined over $`k^{}`$, then
$$[k^{}:k]C\frac{N}{\mathrm{log}N}.$$
In particular, if $`k_n`$ denotes the extension of $`k`$ obtained by adjoining to $`k`$ the $`N_n`$ points of exact period $`n`$, then
(1.20)
$$[k_n:k]\frac{N_n}{\mathrm{log}N_n}.$$
###### Remark 1.21.
For a generic rational map $`\phi `$, we would expect that something stronger than Corollary 1.19 should hold. On the other hand, there are rational maps for which Corollary 1.19 is nearly sharp. For example, if $`\phi (T)=T^2`$ and $`k=`$, then the $`N`$th roots of unity are distinct preperiodic points of $`\phi `$ defined over $`(\zeta _N)`$ and $`[(\zeta _N):]=\varphi (N)`$. Since $`N/\mathrm{log}\mathrm{log}N\varphi (N)N`$, Corollary 1.19 is not far from the truth in this case.
## 2. Discussion and proof of the main result
### 2.1. A sharpening of Theorem 1.1
Let $`(x_1,y_1),\mathrm{},(x_N,y_N)`$ be nonzero points in the filled Julia set $`K_F`$ whose images in $`^1(K)`$ are distinct. Our proof of (1.2) comes from a lower bound for the sum
$$\underset{ij}{}\mathrm{log}|(x_i,y_i)(x_j,y_j)|$$
when $`N`$ belongs to the subset
$$\mathrm{\Sigma }=\{N|N=td^k\text{ for some integers }2t2d1\text{ and }k1\}$$
of the natural numbers.
Specifically, we will prove the following technical result. For notational convenience, let $`ϵ_K`$ be zero if the absolute value on $`K`$ is non-archimedean, and $`1`$ if it is archimedean.
###### Theorem 2.1.
Let $`N=td^k\mathrm{\Sigma }`$, and let $`z_1,\mathrm{},z_N`$ be nonzero elements of the filled Julia set $`K_F`$ whose images in $`^1(K)`$ are all distinct. Let $`R(F)`$ denote the smallest radius so that $`K_F\{zK^2:zR(F)\}`$. Then
$$\begin{array}{ccc}_{ij}\mathrm{log}|z_iz_j|\hfill & \hfill & r(F)N^2ϵ_KN\mathrm{log}N\hfill \\ & & 2\left(\mathrm{log}R(F)\right)\alpha Nr(F)(1+\alpha )N,\hfill \end{array}$$
where $`\alpha =t1+(d1)k>0`$ satisfies $`2\alpha (d1)(\mathrm{log}_dN+2)`$.
###### Remark 2.2.
Using the fact that $`|Tz_iTz_j|=|z_iz_j|`$ for all $`T\mathrm{GL}_2(K)`$ with $`det(T)=\pm 1`$, it can be shown that the quantity $`R(F)`$ appearing in the statement of Theorem 2.1 may be replaced by the smallest radius $`R`$ such that $`K_F\{z:TzR\}`$ for some $`T`$ with $`det(T)=\pm 1`$.
As a corollary of Theorem 2.1, we obtain:
###### Corollary 2.3.
Keeping the notation of Theorem 2.1, let $`N\mathrm{\Sigma }`$, let $`z_1,\mathrm{},z_N`$ be distinct points in $`^1(K)`$, and set
$$D_\phi (z_1,\mathrm{},z_N):=\underset{ij}{}g_\phi (z_i,z_j).$$
Then
(2.4)
$$D_\phi (z_1,\mathrm{},z_N)ϵ_KN\mathrm{log}N\left(2\mathrm{log}R(F)+r(F)\right)\alpha N.$$
In particular,
$$\underset{N\mathrm{}}{lim\; inf}\underset{z_1,\mathrm{},z_N^1(K)}{inf}\frac{1}{N(N1)}\underset{ij}{}g_\phi (z_i,z_j)0.$$
###### Remark 2.5.
If $`K`$ is non-archimedean and $`\phi `$ has good reduction, then $`ϵ_K=\mathrm{log}R(F)=r(F)=0`$, and (2.4) just says that $`D_\phi (z_1,\mathrm{},z_N)0`$, which of course already follows from Remark 1.10.
### 2.2. Proof of Theorem 2.1
An outline of the proof of Theorem 2.1 is as follows. First, we express $`_{ij}|(x_i,y_i)(x_j,y_j)|`$ as the determinant of a Vandermonde matrix $`S`$. We then replace this matrix with a new matrix $`H`$ whose entries involve $`F_1^{(k)}(x_i,y_i)`$ and $`F_2^{(k)}(x_i,y_i)`$ for various $`k0`$ rather than just the standard monomials $`x_i^ay_i^b`$. Replacing $`S`$ by $`H`$ amounts to choosing a different basis for the space of homogeneous polynomials in $`x`$ and $`y`$ of degree $`N1`$, and to calculate the determinant of the change of basis matrix we use a generalization of Lemma 6.5 of . This is the key step in the argument, and is the place where we use the hypotheses that $`N\mathrm{\Sigma }`$. Finally, we use Hadamard’s inequality to estimate the determinant of $`H`$, using the fact that $`F^{(k)}(x_i,y_i)R(F)`$ for all $`k0`$.
Let $`\mathrm{\Gamma }^0(m)`$ denote the vector space of homogeneous polynomials of degree $`m`$ in $`K[x,y]`$, which has dimension $`N=m+1`$ over $`K`$. If $`N\mathrm{\Sigma }`$, i.e., if $`m=td^k1`$ with $`2t2d1`$ and $`k1`$, we consider the collection $`H(m)`$ of polynomials
$$\begin{array}{ccc}H(m)\hfill & =\hfill & \{x^{a_0}y^{b_0}F_1(x,y)^{a_1}F_2(x,y)^{b_1}\mathrm{}F_1^{(k)}(x,y)^{a_k}F_2^{(k)}(x,y)^{b_k}|\hfill \\ & & a_i+b_i=d1\text{ for }0ik1\text{ and }a_k+b_k=t1\}\hfill \\ & \hfill & \mathrm{\Gamma }^0(m).\hfill \end{array}$$
The cardinality of $`H(m)`$ is easily seen to be $`N=dim\mathrm{\Gamma }^0(m)`$. The following proposition shows that $`H(m)`$ forms a basis for $`\mathrm{\Gamma }^0(m)`$, and explicitly calculates the determinant of the change of basis matrix between $`H(m)`$ and the standard monomial basis $`S(m)`$ given by
$$S(m)=\{x^ay^b|a+b=m\}.$$
Its proof is modeled after Lemma 6.5 of .
###### Proposition 2.6.
Let $`A`$ be the matrix expressing the polynomials $`H(m)`$ (in some order) in terms of some ordering of the standard basis $`S(m)`$. Then $`det(A)=\pm \mathrm{Res}(F)^r`$, where
$$r=\frac{N^2}{2d(d1)}\frac{N}{2d(d1)}\left(t+k(d1)\right).$$
In particular, since $`\mathrm{Res}(F)0`$, $`H(m)`$ is a basis for $`\mathrm{\Gamma }^0(m)`$.
###### Proof.
Let $`H_1,\mathrm{},H_N`$ be an ordering of the elements of $`H(m)`$, and let $`S_1,\mathrm{},S_N`$ be an ordering of the elements of $`S(m)`$, so that $`A`$ is the $`N\times N`$ matrix whose $`(i,j)`$th entry is the coefficient of the monomial $`S_i`$ which appears in the expansion of $`H_j`$ as a polynomial in $`x`$ and $`y`$. We have $`det(A)=0`$ if and only if some nontrivial linear combination of the elements of $`H(m)`$ is zero.
Suppose $`det(A)=0`$. Then there exist homogeneous polynomials $`h_1H(d^k1)`$ and $`h_2H((t1)d^k1)`$, not both zero, such that
$$h_1\left(F_1^{(k)}(x,y)\right)^{t1}+h_2F_2^{(k)}(x,y)=0.$$
We may assume that neither of $`h_1,h_2`$ is the zero polynomial. Thus $`F_2^{(k)}`$ divides $`h_1\left(F_1^{(k)}\right)^{t1}`$. Since $`\mathrm{deg}(h_1)<\mathrm{deg}(F_2^{(k)})`$, it follows that $`F_1^{(k)}`$ and $`F_2^{(k)}`$ have a common irreducible factor. Thus $`\mathrm{Res}(F^{(k)})=0`$. But $`\mathrm{Res}(F^{(k)})`$ is a power of $`\mathrm{Res}(F)`$ by \[7, Corollary 6.4\], so $`\mathrm{Res}(F)=0`$ as well.
Conversely, suppose $`\mathrm{Res}(F)=0`$. Then by a standard fact about resultants (see \[16, §5.8\]), there is a nontrivial relation of the form
(2.7)
$$h_1F_1+h_2F_2=0$$
with $`h_1,h_2K[x,y]`$ homogeneous of degree $`d1`$. If $`k=1`$, this already implies that $`det(A)=0`$. If $`k2`$, then multiplying both sides of (2.7) by $`G(x,y)`$, where
$$G(x,y)=F_1^{d2}(F_1^{(2)})^{d1}\mathrm{}(F_1^{(k)})^{t1},$$
gives a linear relation which shows that $`det(A)=0`$.
Now expand both $`det(A)`$ and $`\mathrm{Res}(F)`$ as polynomials in the coefficients of $`F_1`$ and $`F_2`$. Since $`\mathrm{Res}(F)`$ is irreducible (see \[16, §5.9\]), we find that
$$det(A)=C\mathrm{Res}(F)^r$$
for some $`CK^{}`$ and some natural number $`r`$. Now $`\mathrm{Res}(F)`$ is homogeneous of degree $`2d`$ in the coefficients of $`F_1`$ and $`F_2`$, and a straightforward calculation shows that $`F_1^{(j)}`$ and $`F_2^{(j)}`$ are each homogeneous of degree $`(d^j1)/(d1)`$ in the coefficients of $`F_1`$ and $`F_2`$. It follows that $`det(A)`$ is homogeneous of degree
$$r^{}=N\left(\underset{j=1}{\overset{k1}{}}(d^j1)+(t1)\frac{d^k1}{d1}\right)$$
in the coefficients of $`F_1`$ and $`F_2`$. Comparing degrees and performing some straightforward algebraic manipulations, we find that
$$r=\frac{r^{}}{2d}=\frac{N^2}{2d(d1)}\frac{N}{2d}\left(\frac{t}{d1}+k\right).$$
Finally, to compute $`C`$ we set $`F_1=x^d`$ and $`F_2=y^d`$, in which case $`H(m)`$ is just a permutation of the standard monomial basis $`S(m)`$. It follows that $`C=\pm 1`$. ∎
Before turning to the proof of Theorem 2.1, we first recall Hadamard’s inequality. If $`v=(v_1,\mathrm{},v_N)^\mathrm{T}K^N`$, define $`v`$ to be the $`L^2`$-norm $`v=(|v_i|^2)^{1/2}`$ if $`K`$ is archimedean, and to be the sup-norm $`v=sup|v_i|`$ if $`K`$ is non-archimedean. If $`H`$ is a matrix with columns $`h_1,\mathrm{},h_NK^N`$, then Hadamard’s inequality states that
(2.8)
$$|det(H)|\underset{i=1}{\overset{N}{}}h_i.$$
If $`K`$ is non-archimedean, (2.8) is immediate from the definition of the determinant and the ultrametric inequality. If $`K=`$ or $``$ with the usual absolute value, (2.8) is a well-known result from linear algebra. Finally, (2.8) for general archimedean $`K`$ can be deduced from the cases $`K=,`$ using a theorem of Ostrowski (see \[15, Chapter II, Theorem 4.2\]) which says that a complete archimedean valued field $`K`$ is isometric to either $`(,||^s)`$ or $`(,||^s)`$ for some $`0<s1`$.
We can now give the proof of Theorem 2.1.
###### Proof.
Let $`\{S_1,\mathrm{},S_N\}`$ and $`\{H_1,\mathrm{},H_N\}`$ be as in the proof of Proposition 2.6. Using the homogeneous version of the standard formula for the determinant of a Vandermonde matrix, we have
(2.9)
$$\underset{ij}{}|x_iy_jx_jy_i|=|det(S)|^2,$$
where $`S`$ is the matrix whose $`(i,j)`$th entry is $`S_j(x_i,y_i)`$. If $`H`$ is the matrix whose $`(i,j)`$th entry is $`H_j(x_i,y_i)`$, then $`H=SA`$, so that by Proposition 2.6, we have
(2.10)
$$det(S)^2=det(H)^2(det(A))^2=det(H)^2\mathrm{Res}(F)^{2r}.$$
On the other hand, we can estimate $`|det(H)|^2`$ using (2.8). Letting $`h_i`$ be the $`i`$th column of $`H`$, we obtain
(2.11)
$$\begin{array}{ccc}|det(H)|^2\hfill & \hfill & _{i=1}^Nh_i\hfill \\ & \hfill & N^{ϵ_KN}_iR(F)^{2\left(k(d1)+(t1)\right)}\hfill \\ & =\hfill & N^{ϵ_KN}R(F)^{\left(2t2+k(2d2)\right)N}.\hfill \end{array}$$
Putting together (2.9), (2.10), and (2.11) gives
(2.12)
$$\begin{array}{ccc}_{ij}|x_iy_jx_jy_i|\hfill & =\hfill & |det(S)|^2\hfill \\ & =\hfill & |det(H)|^2|\mathrm{Res}(F)|^{2r}\hfill \\ & \hfill & N^{ϵ_KN}R(F)^{2\alpha N}|\mathrm{Res}(F)|^{2r},\hfill \end{array}$$
where $`2r=\frac{N^2}{d(d1)}+\frac{N}{d(d1)}\left(t+k(d1)\right)`$.
Taking the negative logarithm of both sides of (2.12) gives the desired lower bound
$$\begin{array}{ccc}_{ij}\mathrm{log}|x_iy_jx_jy_i|\hfill & \hfill & ϵ_KN\mathrm{log}N2N\alpha \mathrm{log}R(F)\hfill \\ & & +\frac{N^2}{d(d1)}\mathrm{log}|\mathrm{Res}(F)|\hfill \\ & & \frac{N}{d(d1)}\left(t+k(d1)\right)\mathrm{log}|\mathrm{Res}(F)|.\hfill \end{array}$$
Corollary 2.3 is deduced from Theorem 2.1 as follows.
###### Proof.
Without loss of generality, we may assume that the valuation on $`K`$ is nontrivial, and we may replace $`K`$ by $`\overline{K}`$, We may therefore assume that the value group $`|K^{}|`$ is dense in the group $`_{>0}`$ of nonnegative reals. Since $`\widehat{H}_F(cz)=\widehat{H}_F(z)+\mathrm{log}|c|`$ for all $`zK^2\{0\}`$ and all $`cK^{}`$, given $`\epsilon >0`$ we can choose coordinates $`(x_i,y_i)`$ for $`z_i`$ so that $`\epsilon \widehat{H}_F(x_i,y_i)0`$. Then $`(x_i,y_i)K_F`$ and we can apply Theorem 2.1 to $`(x_1,y_1),\mathrm{},(x_N,y_N)`$. Simplifying the resulting expression and letting $`\epsilon 0`$ gives the desired inequality. ∎
Finally, we explain how to deduce Theorem 1.1 from Corollary 2.3. We first recall the following lemma from \[3, Lemma 3.48\]:
###### Lemma 2.13.
For $`N2`$, define
$$D_N:=\underset{z_1,\mathrm{},z_N^1(K)}{inf}\frac{1}{N(N1)}D_\phi (z_1,\mathrm{},z_N).$$
Then the sequence $`D_N`$ is non-decreasing.
We now give the proof of Theorem 1.1:
###### Proof.
Without loss of generality, we may assume that $`N2d`$. Let $`N^{}`$ be the smallest integer less than or equal to $`N`$ which belongs to $`\mathrm{\Sigma }`$. One deduces easily from the definition of $`\mathrm{\Sigma }`$ that
(2.14)
$$\frac{N1}{2}N^{}1N1.$$
Since $`N^{}\mathrm{\Sigma }`$, it follows from Corollary 2.3 that there is a constant $`C^{}>0`$ (independent of $`N`$ and $`N^{}`$) such that
$$D_N^{}C^{}\frac{\mathrm{log}N^{}}{N^{}1}.$$
From this, Lemma 2.13, and (2.14), it follows that
$$\frac{1}{N(N1)}D_\phi (z_1,\mathrm{},z_N)D_N^{}C\frac{\mathrm{log}N}{N1}$$
where $`C=2C^{}`$. This immediately yields the desired result. ∎
## 3. Global applications
In this section, we give a proof of Theorem 1.14 and Corollary 1.19. The idea is to use a pigeonhole principle argument as in ,, . Throughout this section, $`k`$ denotes a number field and $`\phi k(T)`$ is a rational function of degree at least 2. We will denote by $`M_k`$ the set of places of $`k`$, and by $`g_{\phi ,v}`$ the normalized Arakelov-Green’s function over $`k_v`$ (or $`_v`$) attached to $`\phi `$. We begin with the following lemma:
###### Lemma 3.1.
Suppose $`vM_k`$ is archimedean. Then:
* The function
$$g_{\phi ,v}(z,w):^1()\times ^1()\{\mathrm{}\}$$
is lower semicontinuous.
* Given $`\delta >0`$, there exists a finite covering $`U_1,\mathrm{},U_s`$ of $`^1()`$ by open subsets so that for each $`1is`$, $`g_{\phi ,v}(z,w)>\delta `$ for all $`z,wU_i`$.
###### Proof.
(a) It suffices to note that $`g_{\phi ,v}`$ is an increasing limit of the continuous functions
$$g_{\phi ,v}^{(M)}=\mathrm{max}(g_{\phi ,v},M)$$
for $`M>0`$ (see \[3, §3.5\]).
(b) By part (a), the fact that $`g_{\phi ,v}=+\mathrm{}`$ along the diagonal, and the definition of the product topology on $`^1()\times ^1()`$, for each $`x^1()`$ there is an open neighborhood $`U_x`$ of $`x`$ in $`^1()`$ such that $`g_{\phi ,v}(z,w)>\delta `$ for $`z,wU_x`$. By compactness of $`^1()`$, the covering $`\{U_x|x^1()\}`$ has a finite subcover $`U_1,\mathrm{},U_s`$. ∎
We can now prove Theorem 1.14.
###### Proof.
Fix a finite extension $`k^{}/k`$ and let $`D=[k^{}:]`$. By the product formula, for all $`z,w^1(k^{})`$ with $`zw`$ we have
(3.2)
$$\underset{vM_k^{}}{}\frac{[k_v^{}:_v]}{[k^{}:]}g_{\phi ,v}(z,w)=\widehat{h}_\phi (z)+\widehat{h}_\phi (w).$$
Let $`v_0`$ be a fixed archimedean place of $`k^{}`$; then by Lemma 3.1, there exists an open covering $`U_1,\mathrm{},U_s`$ of $`^1()`$ and a constant $`C_1>0`$ such that $`g_{\phi ,v_0}(z,w)C_1`$ whenever $`z,wU_i`$ ($`i=1,\mathrm{},s`$). If $`z_1,\mathrm{},z_N^1(k^{})`$, then by Theorem 1.1 and the fact that $`g_{\phi ,v}0`$ whenever $`\phi `$ has good reduction at $`v`$, there is a constant $`C_2>0`$ depending only on $`\phi `$ and $`k`$ such that
$$g(z_1,\mathrm{},z_N):=\underset{vM_k^{}}{}\underset{ij}{}\frac{[k_v^{}:_v]}{[k^{}:]}g_v(z_i,z_j)C_2N\mathrm{log}N.$$
Moreover, if $`z_1,\mathrm{},z_NU_i`$ for some $`i`$, then
(3.3)
$$g(z_1\mathrm{},z_N)\frac{C_1}{D}N^2C_2N\mathrm{log}N.$$
Let $`M=[\frac{N1}{s}]+1`$. By the pigeonhole principle, in any subset of $`^1()`$ of cardinality $`N`$, there is an $`M`$-element subset contained in some $`U_i`$. Without loss of generality, order the $`z_j`$’s so that this subset is $`\{z_1,\mathrm{},z_M\}`$. Applying (3.2) and (3.3), we obtain
$$\frac{C_1}{D}M^2C_2M\mathrm{log}Mg(z_1,\mathrm{},z_M)2M^2\underset{j}{\mathrm{max}}\widehat{h}_\phi (z_j).$$
If $`\widehat{h}_\phi (z_j)\frac{C_1}{4D}`$ for all $`j=1,\mathrm{},N`$, then we obtain
$$\frac{C_1}{2D}MC_2\mathrm{log}M,$$
which implies that $`MB^{}D\mathrm{log}D`$ for some constant $`B^{}>0`$ depending only on $`\phi `$ and $`k`$. Thus $`NMs+1BD\mathrm{log}D`$ for some $`B>0`$ depending only on $`\phi `$ and $`k`$. Setting $`A=C_1/4`$ now gives the desired result. ∎
Corollary 1.19 follows easily:
###### Proof.
Let $`\{P_1,\mathrm{},P_N\}`$ be a set of preperiodic points defined over a number field $`k^{}`$ with $`[k^{}:k]=D`$. Then by Theorem 1.14, we have $`NBD\mathrm{log}D`$, which implies that $`DN/\mathrm{log}N`$ as claimed. ∎ |
warning/0507/hep-th0507139.html | ar5iv | text | # 1 Introduction
## 1 Introduction
The famous Baxter $`TQ`$ relation , which schematically has the form
$`t(u)Q(u)=Q(u^{})+Q(u^{\prime \prime }),`$ (1.1)
holds for many integrable models associated with the $`sl_2`$ Lie algebra and its deformations, such as the closed XXZ quantum spin chain. This relation provides one of the most direct routes to the Bethe Ansatz expression for the eigenvalues of the transfer matrix $`t(u)`$.
We propose here a generalization of this relation which involves more than one $`Q(u)`$,
$`t(u)Q_1(u)`$ $`=`$ $`Q_2(u^{})+Q_2(u^{\prime \prime }),`$
$`t(u)Q_2(u)`$ $`=`$ $`Q_1(u^{\prime \prime \prime })+Q_1(u^{\prime \prime \prime \prime }).`$ (1.2)
This structure arises naturally in the open XXZ quantum spin chain for special values of the bulk and boundary parameters. We expect that such generalized $`TQ`$ relations, involving two or more independent $`Q(u)`$’s, may also appear in other integrable models.
The open XXZ chain with general integrable boundary terms has a Hamiltonian which can be written as
$``$ $`=`$ $`{\displaystyle \underset{n=1}{\overset{N1}{}}}H_{n,n+1}+{\displaystyle \frac{1}{2}}\mathrm{sinh}\eta [\mathrm{coth}\alpha _{}\mathrm{tanh}\beta _{}\sigma _1^z+csch\alpha _{}sech\beta _{}(\mathrm{cosh}\theta _{}\sigma _1^x+i\mathrm{sinh}\theta _{}\sigma _1^y)`$ (1.3)
$``$ $`\mathrm{coth}\alpha _+\mathrm{tanh}\beta _+\sigma _N^z+csch\alpha _+sech\beta _+(\mathrm{cosh}\theta _+\sigma _N^x+i\mathrm{sinh}\theta _+\sigma _N^y)],`$
where $`H_{n,n+1}`$ is given by
$`H_{n,n+1}={\displaystyle \frac{1}{2}}\left(\sigma _n^x\sigma _{n+1}^x+\sigma _n^y\sigma _{n+1}^y+\mathrm{cosh}\eta \sigma _n^z\sigma _{n+1}^z\right),`$ (1.4)
$`\sigma ^x,\sigma ^y,\sigma ^z`$ are the usual Pauli matrices, $`\eta `$ is the bulk anisotropy parameter, $`\alpha _\pm ,\beta _\pm ,\theta _\pm `$ are boundary parameters, and $`N`$ is the number of spins.
Although this model remains unsolved, the case of diagonal boundary terms ($`\alpha _\pm `$ or $`\beta _\pm \pm \mathrm{}`$) was solved long ago . Moreover, the case of nondiagonal boundary terms when the boundary parameters obey the constraint
$`\alpha _{}+\beta _{}+\alpha _++\beta _+=\pm (\theta _{}\theta _+)+\eta k,`$ (1.5)
(where $`k[(N+1),N+1]`$ is an even/odd integer if $`N`$ is odd/even, respectively) has recently been solved by two approaches: generalized algebraic Bethe Ansatz , and functional relations at roots of unity .
It would clearly be desirable to overcome this constraint, and find the solution for further values of the boundary parameters. Some progress was achieved recently using the functional relations approach . Namely, Bethe Ansätze were proposed for the special cases that at most one of the boundary parameters is nonzero, and the bulk anisotropy has values $`\eta =\frac{i\pi }{3},\frac{i\pi }{5},\mathrm{}`$.
We find here (again by means of the functional relations approach) that by allowing the possibility of generalized $`TQ`$ relations, we can obtain Bethe-Ansatz-type expressions for the transfer matrix eigenvalues for the cases that at most two of the boundary parameters $`\{\alpha _{},\alpha _+,\beta _{},\beta _+\}`$ are nonzero, and the bulk anisotropy has values $`\eta =\frac{i\pi }{2},\frac{i\pi }{4},\mathrm{}`$.
In order to make the paper self-contained, we summarize in Section 2 the construction of the transfer matrix and the functional relations which it satisfies at roots of unity. In order to derive the generalized $`TQ`$ relation, it is instructive to first understand why we are unable to obtain a conventional relation with a single $`Q(u)`$. We present this analysis in Section 3. Finally, in Section 4 we derive the generalized $`TQ`$ relations. We conclude in Section 5 with a brief discussion of our results and of some remaining open problems.
## 2 Transfer matrix and functional relations at roots of unity
The transfer matrix $`t(u)`$ of the open XXZ chain with general integrable boundary terms, which satisfies the fundamental commutativity property
$`[t(u),t(v)]=0,`$ (2.1)
is given by
$`t(u)=tr_0K_0^+(u)T_0(u)K_0^{}(u)\widehat{T}_0(u),`$ (2.2)
where $`T_0(u)`$ and $`\widehat{T}_0(u)`$ are the monodromy matrices
$`T_0(u)=R_{0N}(u)\mathrm{}R_{01}(u),\widehat{T}_0(u)=R_{01}(u)\mathrm{}R_{0N}(u),`$ (2.3)
and $`tr_0`$ denotes trace over the “auxiliary space” 0. The $`R`$ matrix is given by
$`R(u)=\left(\begin{array}{cccc}\mathrm{sinh}(u+\eta )& 0& 0& 0\\ 0& \mathrm{sinh}u& \mathrm{sinh}\eta & 0\\ 0& \mathrm{sinh}\eta & \mathrm{sinh}u& 0\\ 0& 0& 0& \mathrm{sinh}(u+\eta )\end{array}\right),`$ (2.8)
where $`\eta `$ is the bulk anisotropy parameter; and $`K^{}(u)`$ are $`2\times 2`$ matrices whose components are given by
$`K_{11}^{}(u)`$ $`=`$ $`2\left(\mathrm{sinh}\alpha _{}\mathrm{cosh}\beta _{}\mathrm{cosh}u+\mathrm{cosh}\alpha _{}\mathrm{sinh}\beta _{}\mathrm{sinh}u\right)`$
$`K_{22}^{}(u)`$ $`=`$ $`2\left(\mathrm{sinh}\alpha _{}\mathrm{cosh}\beta _{}\mathrm{cosh}u\mathrm{cosh}\alpha _{}\mathrm{sinh}\beta _{}\mathrm{sinh}u\right)`$
$`K_{12}^{}(u)`$ $`=`$ $`e^\theta _{}\mathrm{sinh}2u,K_{21}^{}(u)=e^\theta _{}\mathrm{sinh}2u,`$ (2.9)
and
$`K_{11}^+(u)`$ $`=`$ $`2\left(\mathrm{sinh}\alpha _+\mathrm{cosh}\beta _+\mathrm{cosh}(u+\eta )\mathrm{cosh}\alpha _+\mathrm{sinh}\beta _+\mathrm{sinh}(u+\eta )\right)`$
$`K_{22}^+(u)`$ $`=`$ $`2\left(\mathrm{sinh}\alpha _+\mathrm{cosh}\beta _+\mathrm{cosh}(u+\eta )+\mathrm{cosh}\alpha _+\mathrm{sinh}\beta _+\mathrm{sinh}(u+\eta )\right)`$
$`K_{12}^+(u)`$ $`=`$ $`e^{\theta _+}\mathrm{sinh}2(u+\eta ),K_{21}^+(u)=e^{\theta _+}\mathrm{sinh}2(u+\eta ),`$ (2.10)
where $`\alpha _{},\beta _{},\theta _{}`$ are the boundary parameters. The transfer matrix “contains” the Hamiltonian (1.3),
$`=c_1{\displaystyle \frac{}{u}}t(u)|_{u=0}+c_2𝕀,`$ (2.11)
where
$`c_1`$ $`=`$ $`{\displaystyle \frac{1}{16\mathrm{sinh}\alpha _{}\mathrm{cosh}\beta _{}\mathrm{sinh}\alpha _+\mathrm{cosh}\beta _+\mathrm{sinh}^{2N1}\eta \mathrm{cosh}\eta }},`$
$`c_2`$ $`=`$ $`{\displaystyle \frac{\mathrm{sinh}^2\eta +N\mathrm{cosh}^2\eta }{2\mathrm{cosh}\eta }},`$ (2.12)
and $`𝕀`$ is the identity matrix.
The transfer matrix eigenvalues $`\mathrm{\Lambda }(u)`$ have $`i\pi `$ periodicity
$`\mathrm{\Lambda }(u+i\pi )=\mathrm{\Lambda }(u),`$ (2.13)
crossing symmetry
$`\mathrm{\Lambda }(u\eta )=\mathrm{\Lambda }(u),`$ (2.14)
and the asymptotic behavior
$`\mathrm{\Lambda }(u)\mathrm{cosh}(\theta _{}\theta _+){\displaystyle \frac{e^{u(2N+4)+\eta (N+2)}}{2^{2N+1}}}+\mathrm{}\text{for}u\mathrm{}.`$ (2.15)
For bulk anisotropy values
$`\eta ={\displaystyle \frac{i\pi }{p+1}},p=1,2,\mathrm{},`$ (2.16)
so that $`qe^\eta `$ is a root of unity, the eigenvalues obey functional relations of order $`p+1`$
$`\mathrm{\Lambda }(u)\mathrm{\Lambda }(u+\eta )\mathrm{}\mathrm{\Lambda }(u+p\eta )`$ (2.17)
$``$ $`\delta (u)\mathrm{\Lambda }(u+2\eta )\mathrm{\Lambda }(u+3\eta )\mathrm{}\mathrm{\Lambda }(u+p\eta )`$
$``$ $`\delta (u+\eta )\mathrm{\Lambda }(u)\mathrm{\Lambda }(u+3\eta )\mathrm{\Lambda }(u+4\eta )\mathrm{}\mathrm{\Lambda }(u+p\eta )`$
$``$ $`\delta (u+2\eta )\mathrm{\Lambda }(u)\mathrm{\Lambda }(u+\eta )\mathrm{\Lambda }(u+4\eta )\mathrm{}\mathrm{\Lambda }(u+p\eta )\mathrm{}`$
$``$ $`\delta (u+p\eta )\mathrm{\Lambda }(u+\eta )\mathrm{\Lambda }(u+2\eta )\mathrm{}\mathrm{\Lambda }(u+(p1)\eta )`$
$`+`$ $`\mathrm{}=f(u).`$
For example, for the case $`p=3`$, the functional relation is
$`\mathrm{\Lambda }(u)\mathrm{\Lambda }(u+\eta )\mathrm{\Lambda }(u+2\eta )\mathrm{\Lambda }(u+3\eta )\delta (u)\mathrm{\Lambda }(u+2\eta )\mathrm{\Lambda }(u+3\eta )\delta (u+\eta )\mathrm{\Lambda }(u)\mathrm{\Lambda }(u+3\eta )`$
$`\delta (u+2\eta )\mathrm{\Lambda }(u)\mathrm{\Lambda }(u+\eta )\delta (u+3\eta )\mathrm{\Lambda }(u+\eta )\mathrm{\Lambda }(u+2\eta )`$
$`+\delta (u)\delta (u+2\eta )+\delta (u+\eta )\delta (u+3\eta )=f(u).`$ (2.18)
The functions $`\delta (u)`$ and $`f(u)`$ are given in terms of the boundary parameters $`\alpha _{},\beta _{},\theta _{}`$ by
$`\delta (u)=\delta _0(u)\delta _1(u),f(u)=f_0(u)f_1(u),`$ (2.19)
where
$`\delta _0(u)`$ $`=`$ $`\left(\mathrm{sinh}u\mathrm{sinh}(u+2\eta )\right)^{2N}{\displaystyle \frac{\mathrm{sinh}2u\mathrm{sinh}(2u+4\eta )}{\mathrm{sinh}(2u+\eta )\mathrm{sinh}(2u+3\eta )}},`$ (2.20)
$`\delta _1(u)`$ $`=`$ $`2^4\mathrm{sinh}(u+\eta +\alpha _{})\mathrm{sinh}(u+\eta \alpha _{})\mathrm{cosh}(u+\eta +\beta _{})\mathrm{cosh}(u+\eta \beta _{})`$ (2.21)
$`\times `$ $`\mathrm{sinh}(u+\eta +\alpha _+)\mathrm{sinh}(u+\eta \alpha _+)\mathrm{cosh}(u+\eta +\beta _+)\mathrm{cosh}(u+\eta \beta _+),`$
and therefore,
$`\delta (u+i\pi )=\delta (u),\delta (u2\eta )=\delta (u).`$ (2.22)
For $`p`$ odd,
$`f_0(u)`$ $`=`$ $`(1)^{N+1}2^{2pN}\mathrm{sinh}^{2N}\left((p+1)u\right)\mathrm{tanh}^2\left((p+1)u\right),`$ (2.23)
$`f_1(u)`$ $`=`$ $`2^{32p}(`$ (2.24)
$`\mathrm{cosh}\left((p+1)\alpha _{}\right)\mathrm{cosh}\left((p+1)\beta _{}\right)\mathrm{cosh}\left((p+1)\alpha _+\right)\mathrm{cosh}\left((p+1)\beta _+\right)\mathrm{sinh}^2\left((p+1)u\right)`$
$``$ $`\mathrm{sinh}\left((p+1)\alpha _{}\right)\mathrm{sinh}\left((p+1)\beta _{}\right)\mathrm{sinh}\left((p+1)\alpha _+\right)\mathrm{sinh}\left((p+1)\beta _+\right)\mathrm{cosh}^2\left((p+1)u\right)`$
$`+`$ $`(1)^N\mathrm{cosh}\left((p+1)(\theta _{}\theta _+)\right)\mathrm{sinh}^2\left((p+1)u\right)\mathrm{cosh}^2\left((p+1)u\right)).`$
For $`p`$ even,
$`f_0(u)`$ $`=`$ $`(1)^{N+1}2^{2pN}\mathrm{sinh}^{2N}\left((p+1)u\right),`$ (2.25)
$`f_1(u)`$ $`=`$ $`(1)^{N+1}2^{32p}(`$ (2.26)
$`\mathrm{sinh}\left((p+1)\alpha _{}\right)\mathrm{cosh}\left((p+1)\beta _{}\right)\mathrm{sinh}\left((p+1)\alpha _+\right)\mathrm{cosh}\left((p+1)\beta _+\right)\mathrm{cosh}^2\left((p+1)u\right)`$
$``$ $`\mathrm{cosh}\left((p+1)\alpha _{}\right)\mathrm{sinh}\left((p+1)\beta _{}\right)\mathrm{cosh}\left((p+1)\alpha _+\right)\mathrm{sinh}\left((p+1)\beta _+\right)\mathrm{sinh}^2\left((p+1)u\right)`$
$``$ $`(1)^N\mathrm{cosh}\left((p+1)(\theta _{}\theta _+)\right)\mathrm{sinh}^2\left((p+1)u\right)\mathrm{cosh}^2\left((p+1)u\right)).`$
Hence, $`f(u)`$ satisfies
$`f(u+\eta )=f(u),f(u)=f(u).`$ (2.27)
We also note the identity
$`f_0(u)^2={\displaystyle \underset{j=0}{\overset{p}{}}}\delta _0(u+j\eta ).`$ (2.28)
## 3 An attempt to obtain a conventional $`TQ`$ relation
In order to obtain Bethe Ansatz expressions for the transfer matrix eigenvalues, we try (following ) to recast the functional relations as the condition that the determinant of a certain matrix vanishes. To this end, let us consider again the $`(p+1)\times (p+1)`$ matrix given by
$`(u)=\left(\begin{array}{cccccccc}\mathrm{\Lambda }(u)& \frac{\delta (u)}{h(u+\eta )}& 0& \mathrm{}& 0& h(u)& & \\ h(u+\eta )& \mathrm{\Lambda }(u+\eta )& \frac{\delta (u+\eta )}{h(u+2\eta )}& \mathrm{}& 0& 0& & \\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& & \\ \frac{\delta (u\eta )}{h(u)}& 0& 0& \mathrm{}& h(u+p\eta )& \mathrm{\Lambda }(u+p\eta )& & \end{array}\right),`$ (3.5)
where $`h(u)`$ is a function which is $`i\pi `$-periodic, but otherwise not yet specified. Evidently, successive rows of this matrix are obtained by simultaneously shifting $`uu+\eta `$ and cyclically permuting the columns to the right. Hence, this matrix has the symmetry property
$`S(u)S^1=(u+\eta ),`$ (3.6)
where $`S`$ is the $`(p+1)\times (p+1)`$ matrix given by
$`S=\left(\begin{array}{cccccccc}0& 1& 0& \mathrm{}& 0& 0& & \\ 0& 0& 1& \mathrm{}& 0& 0& & \\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& & \\ 0& 0& 0& \mathrm{}& 0& 1& & \\ 1& 0& 0& \mathrm{}& 0& 0& & \end{array}\right),S^{p+1}=1.`$ (3.12)
This symmetry implies that the corresponding $`TQ`$ relation would involve only one $`Q(u)`$. Indeed, if we assume $`det(u)=0`$ (which, as we discuss below, turns out to be false for the cases which we consider here), then $`(u)`$ has a null eigenvector,
$`(u)v(u)=0.`$ (3.13)
The symmetry (3.6) is consistent with
$`Sv(u)=v(u+\eta ),`$ (3.14)
which in turn implies that $`v(u)`$ has the form
$`v(u)=(Q(u),Q(u+\eta ),\mathrm{},Q(u+p\eta )),Q(u+i\pi )=Q(u).`$ (3.15)
That is, all the components of $`v(u)`$ are determined by a single function $`Q(u)`$. The null eigenvector condition (3.13) together with the explicit forms (3.5), (3.15) of $`(u)`$ and $`v(u)`$ would then lead to a conventional $`TQ`$ relation.
One can verify that the condition $`det(u)=0`$ indeed implies the functional relations (2.17), if $`h(u)`$ satisfies
$`f(u)={\displaystyle \underset{j=0}{\overset{p}{}}}h(u+j\eta )+{\displaystyle \underset{j=0}{\overset{p}{}}}{\displaystyle \frac{\delta (u+j\eta )}{h(u+j\eta )}}.`$ (3.16)
Setting
$`z(u){\displaystyle \underset{j=0}{\overset{p}{}}}h(u+j\eta ),`$ (3.17)
it immediately follows from (3.16) that $`z(u)`$ is given by
$`z(u)={\displaystyle \frac{1}{2}}\left(f(u)\pm \sqrt{\mathrm{\Delta }(u)}\right),`$ (3.18)
where $`\mathrm{\Delta }(u)`$ is defined by
$`\mathrm{\Delta }(u)f(u)^24{\displaystyle \underset{j=0}{\overset{p}{}}}\delta (u+j\eta ).`$ (3.19)
We wish to focus here on new special cases that $`\mathrm{\Delta }(u)`$ is a perfect square. <sup>1</sup><sup>1</sup>1When the constraint (1.5) is satisfied, $`\mathrm{\Delta }(u)`$ is a perfect square; these are the cases studied in . For even values of $`p`$, $`\mathrm{\Delta }(u)`$ is also a perfect square if at most one of the boundary parameters is nonzero; these are the cases studied in . For odd values of $`p`$, $`\mathrm{\Delta }(u)`$ is also a perfect square if at most two of the boundary parameters $`\{\alpha _{},\alpha _+,\beta _{},\beta _+\}`$ are nonzero. We henceforth restrict to such parameter values. In particular, we assume that $`\eta `$ is given by (2.16), with $`p`$ odd (i.e., bulk anisotropy values $`\eta =\frac{i\pi }{2},\frac{i\pi }{4},\mathrm{}`$). For definiteness, here we present results for the case that $`\alpha _{},\beta _{}0`$ and $`\alpha _+=\beta _+=\theta _\pm =0`$. (In the Appendix, we present results for the case that $`\alpha _\pm 0`$ and $`\beta _\pm =\theta _\pm =0`$; and similar results hold for the other cases.) Moreover, we also restrict to even values of $`N`$. (We expect similar results to hold for odd $`N`$.)
For such parameter values, it is easy to arrive at a contradiction. Indeed, on one hand, the definition (3.17) together with the assumed $`i\pi `$-periodicity of $`h(u)`$ (which is required for the symmetry (3.6)) imply the result $`z(u)=z(u+\eta )`$. On the other hand, (3.19) together with (2.19)-(2.24) and (2.28) imply
$`\sqrt{\mathrm{\Delta }(u)}=2^{32p}f_0(u)\left(\mathrm{cosh}((p+1)\alpha _{})+\mathrm{cosh}((p+1)\beta _{})\right)\mathrm{sinh}^2((p+1)u)\mathrm{cosh}((p+1)u).`$ (3.20)
Hence, it follows from (3.18) that $`z(u)z(u+\eta )`$, which contradicts the earlier result. We conclude that for such parameter values, it is not possible to find a function $`h(u)`$ which is $`i\pi `$-periodic and satisfies the condition (3.16). Hence, for such parameter values, the matrix $`(u)`$ given by (3.5) does not lead to the solution of the model, and we fail to obtain a conventional $`TQ`$ relation.
We remark that if either $`\alpha _+`$ or $`\alpha _{}`$ is zero, then the Hamiltonian is no longer given by (1.3), since the coefficient $`c_1`$ (2.12) is singular. Indeed, as noted in , $`t^{}(0)`$ is then proportional to $`\sigma _N^x`$. Hence, in order to obtain a nontrivial integrable Hamiltonian, one must consider the second derivative of the transfer matrix. For the case $`\alpha _{},\beta _{}0`$,
$`t^{\prime \prime }(0)`$ $`=`$ $`16\mathrm{sinh}^{2N1}\eta \mathrm{cosh}\eta (\mathrm{sinh}\alpha _{}\mathrm{cosh}\beta _{}\{\sigma _N^x,{\displaystyle \underset{n=1}{\overset{N1}{}}}H_{n,n+1}\}`$ (3.21)
$`+`$ $`\mathrm{sinh}\alpha _{}\mathrm{cosh}\beta _{}(N\mathrm{cosh}\eta +\mathrm{sinh}\eta \mathrm{tanh}\eta )\sigma _N^x`$
$`+`$ $`\mathrm{sinh}\eta (\sigma _1^x+\mathrm{sinh}\beta _{}\mathrm{cosh}\alpha _{}\sigma _1^z)\sigma _N^x),`$
where $`H_{n,n+1}`$ is given by (1.4). The case $`\alpha _\pm 0`$, for which the Hamiltonian instead has a conventional local form, is discussed in the Appendix.
## 4 The generalized $`TQ`$ relations
Instead of demanding the symmetry (3.6), let us now demand only the weaker symmetry
$`T(u)T^1=(u+2\eta ),TS^2,`$ (4.1)
where $`S`$ is given by (3.12). Indeed, (3.6) implies (4.1), but the converse is not true. A matrix $`(u)`$ with such symmetry is given by
$`(u)=\left(\begin{array}{cccccccc}\mathrm{\Lambda }(u)& \frac{\delta (u)}{h^{(1)}(u)}& 0& \mathrm{}& 0& \frac{\delta (u\eta )}{h^{(2)}(u\eta )}& & \\ h^{(1)}(u)& \mathrm{\Lambda }(u+\eta )& h^{(2)}(u+\eta )& \mathrm{}& 0& 0& & \\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& & \\ h^{(2)}(u\eta )& 0& 0& \mathrm{}& h^{(1)}(u+(p1)\eta )& \mathrm{\Lambda }(u+p\eta )& & \end{array}\right),`$ (4.6)
where $`h^{(1)}(u)`$ and $`h^{(2)}(u)`$ are functions which are $`i\pi `$-periodic, but otherwise not yet specified.
This symmetry implies that the corresponding $`TQ`$ relations will involve two $`Q(u)`$’s. Indeed, assuming again that
$`det(u)=0,`$ (4.7)
then $`(u)`$ has a null eigenvector $`v(u)`$,
$`(u)v(u)=0.`$ (4.8)
The symmetry (4.1) is consistent with
$`Tv(u)=v(u+2\eta ),`$ (4.9)
which implies that $`v(u)`$ has the form
$`v(u)=(Q_1(u),Q_2(u),\mathrm{},Q_1(u2\eta ),Q_2(u2\eta )),`$ (4.10)
with
$`Q_1(u)=Q_1(u+i\pi ),Q_2(u)=Q_2(u+i\pi ).`$ (4.11)
That is, the components of $`v(u)`$ are determined by two independent functions, $`Q_1(u)`$ and $`Q_2(u)`$. The null eigenvector condition (4.8) together with the explicit forms (4.6), (4.10) of $`(u)`$ and $`v(u)`$ now lead to generalized $`TQ`$ relations,
$`\mathrm{\Lambda }(u)`$ $`=`$ $`{\displaystyle \frac{\delta (u)}{h^{(1)}(u)}}{\displaystyle \frac{Q_2(u)}{Q_1(u)}}+{\displaystyle \frac{\delta (u\eta )}{h^{(2)}(u\eta )}}{\displaystyle \frac{Q_2(u2\eta )}{Q_1(u)}},`$ (4.12)
$`=`$ $`h^{(1)}(u\eta ){\displaystyle \frac{Q_1(u\eta )}{Q_2(u\eta )}}+h^{(2)}(u){\displaystyle \frac{Q_1(u+\eta )}{Q_2(u\eta )}}.`$ (4.13)
Since $`\mathrm{\Lambda }(u)`$ has the crossing symmetry (2.14) and $`\delta (u)`$ has the crossing property (2.22), it is natural to have the two terms in (4.12) transform into each other under crossing. Hence, we set
$`h^{(2)}(u)=h^{(1)}(u2\eta ),`$ (4.14)
and we make the Ansatz
$`Q_1(u)`$ $`=`$ $`{\displaystyle \underset{j=1}{\overset{M_1}{}}}\mathrm{sinh}(uu_j^{(1)})\mathrm{sinh}(u+u_j^{(1)}+\eta ),`$
$`Q_2(u)`$ $`=`$ $`{\displaystyle \underset{j=1}{\overset{M_2}{}}}\mathrm{sinh}(uu_j^{(2)})\mathrm{sinh}(u+u_j^{(2)}+3\eta ),`$ (4.15)
which is consistent with the required periodicity (4.11) and crossing properties
$`Q_1(u)`$ $`=`$ $`Q_1(u\eta ),Q_2(u)=Q_2(u3\eta ).`$ (4.16)
Analyticity of $`\mathrm{\Lambda }(u)`$ (4.12), (4.13) implies Bethe-Ansatz-type equations for the zeros $`\{u_j^{(1)},u_j^{(2)}\}`$ of $`Q_1(u),Q_2(u)`$, respectively,
$`{\displaystyle \frac{\delta (u_j^{(1)})h^{(2)}(u_j^{(1)}\eta )}{\delta (u_j^{(1)}\eta )h^{(1)}(u_j^{(1)})}}`$ $`=`$ $`{\displaystyle \frac{Q_2(u_j^{(1)}2\eta )}{Q_2(u_j^{(1)})}},j=1,2,\mathrm{},M_1,`$
$`{\displaystyle \frac{h^{(1)}(u_j^{(2)})}{h^{(2)}(u_j^{(2)}+\eta )}}`$ $`=`$ $`{\displaystyle \frac{Q_1(u_j^{(2)}+2\eta )}{Q_1(u_j^{(2)})}},j=1,2,\mathrm{},M_2.`$ (4.17)
Note that the function $`h^{(1)}(u)`$ has not yet been specified, nor has the important assumption that $`(u)`$ has a vanishing determinant (4.7) yet been verified. These problems are closely related, and we now address them both.
One can verify that the condition $`det(u)=0`$ indeed implies the functional relations (2.17), if $`h^{(1)}(u)`$ satisfies
$`f(u)=w(u){\displaystyle \underset{j=0,2,\mathrm{}}{\overset{p1}{}}}\delta (u+j\eta )+{\displaystyle \frac{1}{w(u)}}{\displaystyle \underset{j=1,3,\mathrm{}}{\overset{p}{}}}\delta (u+j\eta ),`$ (4.18)
where
$`w(u){\displaystyle \frac{\underset{j=1,3,\mathrm{}}{\overset{p}{}}h^{(2)}(u+j\eta )}{_{j=0,2,\mathrm{}}^{p1}h^{(1)}(u+j\eta )}}.`$ (4.19)
It immediately follows from (4.18) that $`w(u)`$ is given by
$`w(u)={\displaystyle \frac{f(u)\pm \sqrt{\mathrm{\Delta }(u)}}{2_{j=0,2,\mathrm{}}^{p1}\delta (u+j\eta )}},`$ (4.20)
where $`\mathrm{\Delta }(u)`$ is the same quantity defined in (3.19).
Let us recall that we are considering the case that $`p`$ is odd, and that at most $`\alpha _{}`$ and $`\beta _{}`$ are nonzero. For this case, $`\sqrt{\mathrm{\Delta }(u)}`$ is given by (3.20). It follows from (4.20) that for $`p=3,7,11,\mathrm{}`$ the two solutions for $`w(u)`$ are given by
$`w(u)`$ $`=`$ $`\mathrm{coth}^{2N}\left({\displaystyle \frac{1}{2}}(p+1)u\right),`$
$`w(u)`$ $`=`$ $`\left({\displaystyle \frac{\mathrm{cosh}((p+1)u)\mathrm{cosh}((p+1)\alpha _{})}{\mathrm{cosh}((p+1)u)+\mathrm{cosh}((p+1)\alpha _{})}}\right)\left({\displaystyle \frac{\mathrm{cosh}((p+1)u)\mathrm{cosh}((p+1)\beta _{})}{\mathrm{cosh}((p+1)u)+\mathrm{cosh}((p+1)\beta _{})}}\right)`$ (4.21)
$`\times \mathrm{coth}^{2N}\left({\displaystyle \frac{1}{2}}(p+1)u\right),p=3,7,11,\mathrm{};`$
and for $`p=1,5,9,\mathrm{}`$ the two solutions for $`w(u)`$ are given by
$`w(u)`$ $`=`$ $`\left({\displaystyle \frac{\mathrm{cosh}((p+1)u)\mathrm{cosh}((p+1)\alpha _{})}{\mathrm{cosh}((p+1)u)+\mathrm{cosh}((p+1)\alpha _{})}}\right)\mathrm{coth}^{2N}\left({\displaystyle \frac{1}{2}}(p+1)u\right),`$
$`w(u)`$ $`=`$ $`\left({\displaystyle \frac{\mathrm{cosh}((p+1)u)+\mathrm{cosh}((p+1)\beta _{})}{\mathrm{cosh}((p+1)u)\mathrm{cosh}((p+1)\beta _{})}}\right)\mathrm{coth}^{2N}\left({\displaystyle \frac{1}{2}}(p+1)u\right),`$ (4.22)
$`p=1,5,9,\mathrm{}.`$
There are many solutions of (4.19) for $`h^{(1)}(u)`$ (with $`h^{(2)}(u)`$ given by (4.14)) corresponding to the above expressions for $`w(u)`$, which also have the required $`i\pi `$ periodicity. We consider here the solutions
$`h^{(1)}(u)=4\mathrm{sinh}^{2N}(u+2\eta ),M_2={\displaystyle \frac{1}{2}}N+p1,M_1=M_2+2,p=3,7,11,\mathrm{}`$ (4.23)
and
$`h^{(1)}(u)=\{\begin{array}{cc}2\mathrm{cosh}(u+\alpha _{})\mathrm{cosh}(u\alpha _{})\mathrm{cosh}(2u)\mathrm{sinh}^{2N}(u+2\eta ),\hfill & M_1=M_2=\frac{1}{2}N+2p1,\hfill \\ p=9,17,25,\mathrm{}\hfill & \\ 2\mathrm{cosh}(u+\alpha _{})\mathrm{cosh}(u\alpha _{})\mathrm{cosh}(2u)\mathrm{sinh}^{2N}(u+2\eta ),\hfill & M_1=M_2=\frac{1}{2}N+\frac{3}{2}(p1),\hfill \\ p=5,13,21,\mathrm{}\hfill & \\ 2\mathrm{cosh}(u+\alpha _{})\mathrm{cosh}(u\alpha _{})\mathrm{cosh}(2u)\mathrm{sinh}^{2N}(u+2\eta ),\hfill & M_1=M_2=\frac{1}{2}N+2,\hfill \\ p=1,\hfill & \end{array}`$ (4.30)
corresponding to the first solutions for $`w(u)`$ given in (4.21), (4.22), respectively. We have searched for solutions largely by trial and error, verifying numerically (along the lines explained in ) for small values of $`N`$ that the eigenvalues can indeed be expressed as (4.12), (4.13) with $`Q(u)`$’s of the form (4.15).
Note that the values of $`M_1`$ and $`M_2`$ (i.e., the number of zeros of $`Q_1(u)`$ and $`Q_2(u)`$, respectively) depend on the particular choice for the function $`h^{(1)}(u)`$. Our reason for choosing (4.23), (4.30) over the other solutions which we found is that the former solutions gave the lowest values of $`M_1`$ and $`M_2`$, for given values of $`N`$ and $`p`$. (It would be interesting to know whether there exist other solutions for $`h^{(1)}(u)`$ which give even lower values of $`M_1`$ and $`M_2`$.) Our conjectured values of $`M_1`$ and $`M_2`$ given in (4.23), (4.30) are consistent with the asymptotic behavior (2.15). Moreover, these values have been checked numerically for small values of $`N`$ (up to $`N=6`$) and $`p`$ (up to $`p=21`$). That is, we have verified numerically that, with the above choice of $`h^{(1)}(u)`$, the generalized $`TQ`$ relations (4.12), (4.13) correctly give all $`2^N`$ eigenvalues, with $`Q_1(u)`$ and $`Q_2(u)`$ of the form (4.15) and with $`M_1`$ and $`M_2`$ given in (4.23), (4.30). We expect that similar results can be obtained corresponding to the second solutions for $`w(u)`$.
To summarize, we propose that for the case that $`p`$ is odd and that at most $`\alpha _{}`$, $`\beta _{}`$ are nonzero, the eigenvalues $`\mathrm{\Lambda }(u)`$ of the transfer matrix $`t(u)`$ (2.2) are given by the generalized $`TQ`$ relations (4.12), (4.13), with $`Q_1(u)`$ and $`Q_2(u)`$ given by (4.15), $`h^{(2)}(u)`$ given by (4.14), and $`h^{(1)}(u)`$ given by (4.23), (4.30). The zeros $`\{u_j^{(1)},u_j^{(2)}\}`$ of $`Q_1(u)`$ and $`Q_2(u)`$ are solutions of the Bethe Ansatz equations (4.17). We expect that there are sufficiently many such equations to determine all the zeros. As already mentioned, similar results hold for the case that at most two of the boundary parameters $`\{\alpha _{},\alpha _+,\beta _{},\beta _+\}`$ are nonzero.
## 5 Discussion
We have argued that the eigenvalues of the transfer matrix of the open XXZ chain, for the special case that $`p`$ is odd and that at most two of the boundary parameters $`\{\alpha _{},\alpha _+,\beta _{},\beta _+\}`$ are nonzero, can be given by generalized $`TQ`$ relations (4.12), (4.13) involving more than one $`Q(u)`$. Although we have not ruled out the possibility of expressing these eigenvalues in terms of a conventional $`TQ`$ relation, the analysis in Section 3 suggests to us that this is unlikely.
Many interesting problems remain to be explored. It should be possible to explicitly construct operators $`Q_1(u)`$, $`Q_2(u)`$ which commute with each other and with the transfer matrix $`t(u)`$, and whose eigenvalues are given by (4.15). There may be further special cases for which the quantity $`\mathrm{\Delta }(u)`$ (3.19) is a perfect square, in which case it should not be difficult to find the corresponding Bethe Ansatz solution. The general case that $`\mathrm{\Delta }(u)`$ is not a perfect square and/or that $`\eta i\pi /(p+1)`$ remains to be understood.
Generalized $`TQ`$ relations are novel structures, which merit further investigation. The corresponding Bethe Ansatz equations (e.g., (4.17)) have some resemblance to the “nested” equations which are characteristic of higher-rank models. Such generalized $`TQ`$ relations, involving two or even more $`Q(u)`$’s, may also lead to further solutions of integrable open chains of higher rank and/or higher-dimensional representations. (For recent progress on such models, see e.g. .)
## Acknowledgments
One of us (RN) is grateful to S. Ruijsenaars for helpful correspondence. This work was supported in part by the National Science Foundation under Grant PHY-0244261.
## Appendix A Appendix: the case $`\alpha _\pm 0`$ and $`\beta _\pm =\theta _\pm =0`$
Here we consider the case that $`\alpha _\pm 0`$ and $`\beta _\pm =\theta _\pm =0`$, for which the Hamiltonian is local,
$``$ $`=`$ $`{\displaystyle \underset{n=1}{\overset{N1}{}}}H_{n,n+1}+{\displaystyle \frac{1}{2}}\mathrm{sinh}\eta \left(csch\alpha _{}\sigma _1^x+csch\alpha _+\sigma _N^x\right),`$ (A.1)
as follows from (1.3). For this case, the quantity $`\sqrt{\mathrm{\Delta }(u)}`$ is given by (3.20) with $`\beta _{}`$ replaced by $`\alpha _+`$, namely,
$`\sqrt{\mathrm{\Delta }(u)}=2^{32p}f_0(u)\left(\mathrm{cosh}((p+1)\alpha _{})+\mathrm{cosh}((p+1)\alpha _+)\right)\mathrm{sinh}^2((p+1)u)\mathrm{cosh}((p+1)u).`$ (A.2)
It follows that the two solutions for $`w(u)`$ (4.20) are given by
$`w(u)`$ $`=`$ $`\mathrm{coth}^{2N}\left({\displaystyle \frac{1}{2}}(p+1)u\right),`$
$`w(u)`$ $`=`$ $`\left({\displaystyle \frac{\mathrm{cosh}((p+1)u)\mathrm{cosh}((p+1)\alpha _{})}{\mathrm{cosh}((p+1)u)+\mathrm{cosh}((p+1)\alpha _{})}}\right)\left({\displaystyle \frac{\mathrm{cosh}((p+1)u)\mathrm{cosh}((p+1)\alpha _+)}{\mathrm{cosh}((p+1)u)+\mathrm{cosh}((p+1)\alpha _+)}}\right)`$ (A.3)
$`\times \mathrm{coth}^{2N}\left({\displaystyle \frac{1}{2}}(p+1)u\right),p=3,7,11,\mathrm{},`$
and
$`w(u)`$ $`=`$ $`\mathrm{coth}^{2N+2}\left({\displaystyle \frac{1}{2}}(p+1)u\right),`$
$`w(u)`$ $`=`$ $`\left({\displaystyle \frac{\mathrm{cosh}((p+1)u)\mathrm{cosh}((p+1)\alpha _{})}{\mathrm{cosh}((p+1)u)+\mathrm{cosh}((p+1)\alpha _{})}}\right)\left({\displaystyle \frac{\mathrm{cosh}((p+1)u)\mathrm{cosh}((p+1)\alpha _+)}{\mathrm{cosh}((p+1)u)+\mathrm{cosh}((p+1)\alpha _+)}}\right)`$ (A.4)
$`\times \mathrm{coth}^{2N+2}\left({\displaystyle \frac{1}{2}}(p+1)u\right),p=1,5,9,\mathrm{}.`$
For simplicity, let us once again consider just the first solutions for $`w(u)`$ given in (A.3) and (A.4), which are independent of $`\alpha _\pm `$. Corresponding solutions of (4.19) for $`h^{(1)}(u)`$ (with $`h^{(2)}(u)`$ given by (4.14)) are
$`h^{(1)}(u)=4\mathrm{sinh}^{2N}(u+2\eta ),M_2={\displaystyle \frac{1}{2}}N+{\displaystyle \frac{1}{2}}(3p1),M_1=M_2+2,p=3,7,11,\mathrm{}`$ (A.5)
and
$`h^{(1)}(u)=\{\begin{array}{cc}2\mathrm{cosh}(2u)\mathrm{sinh}^2u\mathrm{sinh}^{2N}(u+2\eta ),\hfill & M_1=M_2=\frac{1}{2}N+2p1,\hfill \\ p=9,17,25,\mathrm{}\hfill & \\ 2\mathrm{cosh}(2u)\mathrm{sinh}^2u\mathrm{sinh}^{2N}(u+2\eta ),\hfill & M_1=M_2=\frac{1}{2}N+\frac{3}{2}(p1),\hfill \\ p=5,13,21,\mathrm{}\hfill & \\ 2\mathrm{cosh}(2u)\mathrm{sinh}^2u\mathrm{sinh}^{2N}(u+2\eta ),\hfill & M_1=M_2=\frac{1}{2}N+2,\hfill \\ p=1.\hfill & \end{array}`$ (A.12)
That is, the eigenvalues $`\mathrm{\Lambda }(u)`$ of the transfer matrix $`t(u)`$ (2.2), for $`\eta `$ values (2.16) with $`p`$ odd and for $`\alpha _\pm 0`$ and $`\beta _\pm =\theta _\pm =0`$, are given by the generalized $`TQ`$ relations (4.12), (4.13), with $`Q_1(u)`$ and $`Q_2(u)`$ given by (4.15), $`h^{(2)}(u)`$ given by (4.14), and $`h^{(1)}(u)`$ given by (A.5), (A.12). The zeros $`\{u_j^{(1)},u_j^{(2)}\}`$ of $`Q_1(u)`$ and $`Q_2(u)`$ are solutions of the Bethe Ansatz equations (4.17). |
warning/0507/quant-ph0507216.html | ar5iv | text | # 1 The general scheme for converting one mixed state of zero and one photon to another.
Interconvertibility of single-rail optical qubits
Dominic W. Berry
Department of Physics, The University of Queensland, Queensland 4072, Australia
Alexander I. Lvovsky and Barry C. Sanders
Institute for Quantum Information Science, University of Calgary, Alberta T2N 1N4, Canada
## Abstract
We show how to convert between partially coherent superpositions of a single photon with the vacuum using linear optics and postselection based on homodyne measurements. We introduce a generalized quantum efficiency for such states and show that any conversion that decreases this quantity is possible. We also prove that our scheme is optimal by showing that no linear optical scheme with generalized conditional measurements, and with one single-rail qubit input can improve the generalized efficiency.
OCIS codes: 270.5290, 270.5570
The *single-rail optical qubit* is a coherent superposition of the single-photon and vacuum states of light: $`|\psi =\alpha |0+\beta |1`$. Such qubits, along with their dual-rail siblings, are basic units of information in quantum-optical information processing. Recently, several experiments implemented preparation of arbitrary single-rail qubits from the single-photon state using linear optics and conditional measurements. The quantum catalysis scheme used an ancillary coherent state input and conditional single-photon measurements. The quantum scissors setup employed the Bennett quantum teleportation protocol with a delocalized single photon as an entangled resource and a coherent state as an input. Most recently, a similar resource was used for remote preparation of single-rail qubits via field quadrature (homodyne) measurements by one of the entangled parties.
All these schemes can be reversed to generate a single-photon state, and, more generally, another single-rail qubit from a single-rail qubit input. In this paper, we concentrate on, and generalize, the setup of Babichev et al., which is shown in Fig. 1. The initial qubit, $`\widehat{\rho }`$, is combined with vacuum at a beam splitter, generating a two-mode state $`\widehat{\rho }_{\mathrm{BS}}`$. A measurement described by a positive operator-valued measure (POVM) is then performed on mode 1. Conditioned on measurement result $`k`$, the output in mode 2 is $`\widehat{\rho }^{}\mathrm{Tr}_1(\widehat{M}_k\widehat{\rho }_{\mathrm{BS}})`$.
For a pure input $`\widehat{\rho }=|\psi \psi |`$, and a projective measurement in mode 1, the output $`\widehat{\rho }^{}`$ is also a single-rail qubit. However, with an imperfect input or generalized measurement, the output state may not be pure. In a realistic experiment, the single most significant imperfection of the input state is the admixture of the vacuum:
$$\widehat{\rho }=E|\psi \psi |+(1E)|00|$$
(1)
with $`E1`$ being the quantum efficiency (we take the convention that $`E=\beta =0`$ for the vacuum state). The output state is then also of the form (1), and we use primed symbols for the notation related to the output state. In this Letter, we answer the following question: under which circumstances can an imperfect single-rail qubit characterized by parameters $`(\alpha ,\beta ,E)`$ be converted to another qubit with parameters $`(\alpha ^{},\beta ^{},E^{})`$?
The initial state is transformed by the beam splitter of transmissivity $`t^2`$ and reflectivity $`r^2`$ to $`\widehat{\rho }_{\mathrm{BS}}=E|\varphi \varphi |+(1E)|00`$, with $`|\varphi =\alpha |00+\beta r|10+\beta t|01`$. We begin by considering a projective measurement $`|QQ|`$. Projecting mode 1 onto a state $`|Q`$ will yield the unnormalized output state
$`\widehat{\rho }^{}`$ $`=E[(\alpha \theta _0+\beta r\theta _1)|0+\beta t\theta _0|1]\mathrm{H}.\mathrm{c}.`$
$`+(1E)|\theta _0|^2|00|,`$ (2)
where $`\theta _j=Q|j`$, from which we find
$$\beta ^{}(\alpha \theta _0+\beta r\theta _1)=\alpha ^{}\beta t\theta _0.$$
(3)
The trace
$$\mathrm{Tr}[\widehat{\rho }^{}]=E(|\alpha \theta _0+\beta r\theta _1|^2+|\beta t\theta _0|^2)+(1E)|\theta _0|^2$$
(4)
is equal to (proportional to in the case of $`Q`$ being a continuous observable) the probability for the desired measurement result to occur, and the efficiency of the output qubit is
$$E^{}=E(|\alpha \theta _0+\beta r\theta _1|^2+|\beta t\theta _0|^2)/\mathrm{Tr}[\widehat{\rho }^{}].$$
(5)
Using Eq. (3) as well as $`\alpha ^2+\beta ^2=\alpha ^2+\beta ^2=1`$, we simplify Eqs. (4) and (5) as follows:
$`t\beta \sqrt{E(1E^{})}=\beta ^{}\sqrt{E^{}(1E)},`$ (6)
$`p_{\mathrm{success}}\mathrm{Tr}[\widehat{\rho }^{}]=|\theta _0|^2(1E)/(1E^{}).`$ (7)
Eq. (3) determines the efficiency-independent parameters of the output qubit that can be controlled by choosing the beam splitter and the measurement. An example is a field quadrature measurement using a homodyne detector. With local oscillator phase $`\varphi `$, result $`Q`$ gives the projection $`Q_\varphi |`$ satisfying (scaling convention is $`[\widehat{Q},\widehat{P}]=i/2`$)
$`\theta _0=Q_\varphi |0`$ $`=\left(2/\pi \right)^{1/4}e^{|Q|^2},`$
$`\theta _1=Q_\varphi |1`$ $`=2Q\left(2/\pi \right)^{1/4}e^{|Q|^2}e^{i\varphi },`$ (8)
so $`\theta _1/\theta _0=2Qe^{i\varphi }`$. With any beam splitter, by choosing appropriate $`Q`$ and $`\varphi `$, one can obtain any desired qubit transformation except the transformation to vacuum. (The trivial transformation to vacuum may be obtained by taking $`t=0`$.)
We now turn to restrictions on the efficiency of the output qubit. We define the generalized efficiency by
$$(\widehat{\rho })\frac{\rho _{11}}{1|\rho _{01}|^2/\rho _{11}}=\frac{|\beta |^2E}{1|\alpha |^2E},$$
(9)
where $`\rho _{ij}=i|\widehat{\rho }|j`$, and the $`\rho _{ij}`$ are given by
$$(\rho _{ij})=\left[\begin{array}{cccccccccc}1E|\beta |^2& E\alpha ^{}\beta & & & & & & & & \\ E\alpha \beta ^{}& E|\beta |^2& & & & & & & & \end{array}\right].$$
(10)
This efficiency has a number of useful properties: it is convex (see Appendix A); it reduces to $`E`$ for imperfect single photon sources ($`\alpha =0`$); $`(\widehat{\rho })=1`$ corresponds to pure states, and $`(\widehat{\rho })=0`$ corresponds to the vacuum state. Most importantly, since the beam splitter transmission cannot exceed one, Eq. (6) entails $`(\widehat{\rho })(\widehat{\rho }^{})`$. Hence *only those transformations that do not increase the generalized efficiency are possible*.
Furthermore (see Appendix B), *our scheme with homodyne measurements permits any transformation that decreases the generalized efficiency*. There is a complication for the case $`(\widehat{\rho })=1`$ and $`(\widehat{\rho })>(\widehat{\rho }^{})`$: a mixed state cannot be generated from a pure one using a projective measurement. In this case we can slightly attenuate the input state so its generalized efficiency reduces to a value between 1 and $`(\widehat{\rho }^{})`$. We can then transform the state to $`\widehat{\rho }^{}`$ using a second beam splitter and projective measurement. Mathematically, this procedure is equivalent to that of Fig. 1 with a POVM.
There are also two cases where it is possible to transform between states of equal generalized efficiency. These cases are:
Case 1: $`(\widehat{\rho })=(\widehat{\rho }^{})=1`$
Case 2: $`E=E^{}`$ and $`|\beta |=|\beta ^{}|`$
Case 1 corresponds to transforming between different pure states. Case 2 corresponds to a trivial phase shift, and also includes the trivial vacuum to vacuum transform. One can show (Appendix B) that these cases are the only ones where a transform between states of equal efficiency is possible.
We now generalize the obtained efficiency enhancement restriction to any scheme involving linear optics and conditional measurements (Fig. 2(a)). We assume that, first, there is only one nonclassical input: the single-rail qubit $`\widehat{\rho }`$ and, second, the output state $`\widehat{\rho }^{}`$ has only zero- and one-photon terms in its Fock decomposition. Consider a processing scheme where the state is combined with any number of vacuum and coherent states at the input, then a POVM measurement is performed on all the modes except 1. A $`U(N)`$ interferometer can be decomposed into a line of beam splitters followed by a $`U(N1)`$ interferometer. The $`U(N1)`$ interferometer can then be absorbed into the measurement, so the resulting interferometer is as in Fig. 2(b).
The beam splitters leave the coherent modes unentangled. One of these is an input to the last beam splitter, whereas a measurement is performed on the others. As these modes are not entangled, they may be omitted entirely, so there is only one coherent state input. In order to eliminate the multiphoton components in $`\widehat{\rho }^{}`$, the amplitude of this coherent state must be zero. The simplified configuration is then just as in Fig. 1.
We now recall that the measurement on mode 2 is generally not a projective measurement, but a POVM measurement. However, by a singular value decomposition, $`\widehat{M}_k=p_i|\sigma _iQ_i|`$, where the $`|Q_i`$’s are orthogonal. The state $`\widehat{\rho }^{}`$ is thus a statistical mixture of outputs $`\widehat{\rho }_i^{}`$ associated with projections onto $`|Q_i`$. Because the generalized efficiency is convex (see Appendix A), we find $`i`$ $`(\widehat{\rho }^{})(\widehat{\rho }_i^{})(\widehat{\rho })`$, which completes the proof.
One application of our general scheme is to transform an incoherent mixture of the single-photon and vacuum states to a partially coherent state; this corresponds to the state preparation achieved by Refs. 3, 5, 7. A reverse transformation is also possible. In other words, the state $`\widehat{\rho }`$ is equivalent to a photon source with efficiency $`(\widehat{\rho })`$, in the sense that we may interconvert between these two states arbitrarily accurately. Therefore, it is reasonable to state that the generalized efficiency we have defined for mixed states with partial coherence is equivalent to the usual efficiency for single photon sources.
Whereas the generalized efficiency cannot be increased, it is possible to enhance the specific efficiency $`E`$ of a state, as demonstrated experimentally. However, this will occur at the expense of the single-photon fraction $`\beta `$ in the qubit part of the state. It is not possible, for example, to enhance the efficiency of a single-photon source by first converting it into a single-rail qubit and then back into a single-photon state: while the efficiency may increase in the first step, it will reduce in the second step to no more than its original value.
Recently, Berry et al. proved for some special cases that it is impossible to enhance the efficiency of a single-photon source with any interferometric scheme such as in Fig. 2(a) but with an arbitrary number of inefficient single-photon inputs. If this limitation is correct in general, the impossibility proof made in this paper would also be valid to the same degree of generality. Indeed, if there existed a scheme allowing one to conditionally convert an inefficient single-rail qubit $`\widehat{\rho }`$ to another $`\widehat{\rho }^{}`$ so that $`(\widehat{\rho }^{})>(\widehat{\rho })`$, one could use it to enhance the efficiency of a single-photon source. One would first set up a circuit as in Fig. 1 that converted the single photon to $`\widehat{\rho }`$ with small loss of generalized efficiency, then transform $`\widehat{\rho }`$ to $`\widehat{\rho }^{}`$ enhancing the generalized efficiency, then convert $`\widehat{\rho }^{}`$ back to a single-photon state.
In summary, we have presented a general state transformation scheme for single-rail qubits. This scheme uses only a beam splitter and conditional measurement, and includes existing experimental arrangements as special cases. We have also introduced a generalized measure of the efficiency of partially coherent mixed states of zero and one photon. This efficiency corresponds to the efficiency of imperfect single photon sources, because it is possible to reversibly convert between this state and a single photon source with arbitrarily low efficiency loss. Our scheme allows any transformation that decreases this generalized efficiency. Transformations that increase the efficiency are not possible, and transformations that keep the generalized efficiency constant are not possible, except for some trivial cases.
Appendix A: Convexity Proof. – Here we show that
$$(p\widehat{\rho }_1+(1p)\widehat{\rho }_2)p(\widehat{\rho }_1)+(1p)(\widehat{\rho }_2),$$
(11)
for $`p[0,1]`$. If both $`\widehat{\rho }_1`$ and $`\widehat{\rho }_2`$ are the vacuum state, then clearly equality is achieved. If one of $`\widehat{\rho }_1`$ and $`\widehat{\rho }_2`$ is the vacuum state, then we can take that state to be $`\widehat{\rho }_2`$ without loss of generality. Then, according to Eq. (10),
$`(p\widehat{\rho }_1+(1p)\widehat{\rho }_2)`$ $`={\displaystyle \frac{p|\beta |^2E}{1p|\alpha |^2E}}{\displaystyle \frac{p|\beta _1|^2E_1}{1|\alpha _1|^2E_1}}`$
$`=p(\widehat{\rho }_1)+(1p)(\widehat{\rho }_2).`$ (12)
For the case where neither state is the vacuum we define the function $`f(p)=[p\widehat{\rho }_1+(1p)\widehat{\rho }_2]`$. Taking the second derivative of this function, we obtain
$`f^{\prime \prime }(p)`$ $`={\displaystyle \frac{2\rho _{11}^2|\mathrm{\Delta }\rho _{10}|^2}{(\rho _{11}|\rho _{10}|^2)^2}}`$
$`+{\displaystyle \frac{2[\mathrm{\Delta }\rho _{11}|\rho _{10}|^2\rho _{11}2\mathrm{R}\mathrm{e}(\rho _{01}\mathrm{\Delta }\rho _{10})]^2}{(\rho _{11}|\rho _{10}|^2)^3}},`$ (13)
where $`\rho _{ij}=i|[p\widehat{\rho }_1+(1p)\widehat{\rho }_2]|j`$ and $`\mathrm{\Delta }\rho _{ij}=i|(\widehat{\rho }_1\widehat{\rho }_2)|j`$. As neither state is the vacuum, this second derivative exists and is non-negative for $`p[0,1]`$. Hence $`f(p)`$ is convex, which implies Eq. (11). Thus, in each case we find that Eq. (11) holds, so the generalized efficiency is convex.
Appendix B: Optimality Proof. – We first show that transforms that decrease the generalized entropy are possible. If $`(\widehat{\rho })>(\widehat{\rho }^{})`$, then $`|\beta |^2E(1E^{})>|\beta ^{}|^2E^{}(1E)`$, so there is a solution of Eq. (6) with $`|t|<1`$ and $`r0`$. If $`\beta ^{}=0`$, the transform can be obtained with $`t=0`$. If $`\beta ^{}0`$, then from Eq. (3) $`\theta _0`$ is also nonzero. The probability for success is then nonzero for $`E<1`$.
Now we show that the transformation that preserves the generalized efficiency is possible for Case 1 (Case 2 is trivial). For Case 1, $`E=E^{}=1`$, so Eq. (6) is satisfied regardless of the value of $`t`$. Taking $`t=r=1/\sqrt{2}`$, $`r\beta \beta ^{}0`$, so $`\theta _00`$. The probability of success is then $`|\theta _0|^2(|t\alpha ^{}\beta /\beta ^{}|^2+|\beta t|^2)`$, which is nonzero, so the transformation is possible.
If $`(\widehat{\rho })=(\widehat{\rho }^{})`$ but Cases 1 and 2 do not hold, then $`|\beta |^2E(1E^{})=|\beta ^{}|^2E^{}(1E)`$. For a projective measurement we obtain $`|t|=1`$, so $`r=0`$ and $`t\alpha ^{}\beta \alpha \beta ^{}`$. Then Eq. (3) gives $`\theta _0=0`$, so the probability for success is zero. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.