id
stringlengths 27
33
| source
stringclasses 1
value | format
stringclasses 1
value | text
stringlengths 13
1.81M
|
|---|---|---|---|
warning/0001/math0001026.html | ar5iv | text | # Lattices and parameter reduction in division algebras
## 1. Introduction
Throughout this paper $`k`$ denotes a (fixed) algebraically closed base field of characteristic zero. Let $`K`$ be a field containing $`k`$ and let $`A`$ be a finite-dimensional $`K`$-algebra. We would like to write $`A`$ as $`A=A_0_{K_0}K`$ for some $`K_0`$-algebra $`A_0`$ over an intermediate field $`kK_0K`$ with $`\mathrm{trdeg}_k(K_0)`$ as low as possible; the minimal value of $`\mathrm{trdeg}_k(K_0)`$ will be denoted by $`\tau (A)`$. Note that if $`\mathrm{trdeg}_k(K_0)<\mathrm{trdeg}_k(K)`$ then passing from $`A`$ to $`A_0`$ may be viewed as โparameter reductionโ in $`A`$.
We shall be particularly interested in the case where $`A=\mathrm{UD}(n)`$ is the universal division algebra of degree $`n`$ and $`K`$ is the center of $`\mathrm{UD}(n)`$ which we shall denote by $`Z(n)`$. Recall that $`\mathrm{UD}(n)`$ is the subalgebra of $`\mathrm{M}_n(k(x_{ij},y_{ij}))`$ generated (as a division algebra) by two generic $`n\times n`$-matrices $`X=(x_{ij})`$ and $`Y=(y_{ij})`$, where $`x_{ij}`$ and $`y_{ij}`$ are $`2n^2`$ independent variables over $`k`$; see, e.g., \[Pr, Section II.1\] or \[Row<sub>1</sub>, Section 3.2\]. We will denote $`\tau (\mathrm{UD}(n))`$ by $`d(n)`$. It is easy to show that $`d(n)\tau (A)`$ for any central simple algebra $`A`$ of degree $`n`$ whose center contains $`k`$ (see, e.g., \[Re<sub>2</sub>, Lemma 9.2\]); in other words, every central simple algebra of degree $`n`$ can be โreduced to at most $`d(n)`$ parametersโ. In the language of \[Re<sub>2</sub>\], $`d(n)=\mathrm{ed}(\mathrm{PGL}_n)`$, where $`\mathrm{ed}`$ denotes the essential dimension; see \[Re<sub>2</sub>, Lemma 9.2\].
To the best of our knowledge, the earliest attempt to determine the value of $`d(n)`$ is due to Procesi, who showed that $`d(n)n^2`$; see \[Pr, Thm. 2.1\]. Note that if $`\mathrm{UD}(n)`$ is cyclic then $`d(n)=2`$, because we can take $`A_0`$ to be a symbol algebra; cf. \[Re<sub>2</sub>, Lemma 9.4\]. This is known to be the case for $`n=2`$, $`3`$ and $`6`$. For other $`n`$ the exact value of $`d(n)`$ is not known. However, the following inequalities hold:
$$d(n)n^22n\text{(}\text{[Re2, Proposition 4.5]}\text{)},$$
(1.1)
$$d(n)d(nm)d(n)+d(m)\text{if }(n,m)=1\text{(}\text{[Re2, Section 9.4]}\text{)},$$
(1.2)
$$d(n^r)2r\text{(}\text{[Re1, Theorem 16.1]}\text{)},$$
(1.3)
$$d(n)\frac{1}{2}(n1)(n2)+n\text{if }n\text{ is odd}\text{(}\text{[Row2]}\text{; cf. }\text{[Re2, Section 9.3]}\text{)}.$$
(1.4)
The last inequality is due to Rowen. In this paper we will sharpen it by showing that, in fact, $`d(n)\frac{1}{2}(n1)(n2)`$ for every odd $`n5`$. Moreover, in $`\mathrm{UD}(n)`$, reduction to this number of parameters can be arranged in a particularly nice fashion:
###### Theorem 1.1.
Let $`n5`$ be an odd integer, $`\mathrm{UD}(n)`$ be the universal division algebra of degree $`n`$ and $`Z(n)`$ be its center. Then there exists a subfield $`F`$ of $`Z(n)`$ and a division algebra $`D`$ of degree $`n`$ with center $`F`$ such that
1. $`\mathrm{UD}(n)=D_FZ(n)`$,
2. $`\mathrm{trdeg}_k(F)=\frac{1}{2}(n1)(n2)`$ and
3. $`Z(n)`$ is a rational extension of $`F`$.
In particular, $`d(n)=\mathrm{ed}(\mathrm{PGL}_n)\frac{1}{2}(n1)(n2)`$.
In the course of the proof of Theorem 1.1 we will obtain an explicit description of the center $`F`$ of $`D`$: $`Fk(^2A_{n1})^{๐ฎ_n}`$, where $`๐ฎ_n`$ denotes the symmetric group on $`n`$ symbols and
$$A_{n1}=\{(a_1,\mathrm{},a_n)^n|a_1+\mathrm{}+a_n=0\},$$
(1.5)
with the natural $`๐ฎ_n`$-action. Our argument relies on the results of \[LL\], where the symmetric square $`\mathrm{๐ฒ๐๐}^2A_{n1}`$ is shown to be stably permutation for $`n`$ odd; see Proposition 4.5 below.
For our next result, recall that if $`A`$ is a central simple algebra of degree $`n`$ with center $`F`$ and $`L`$ is a subfield of $`A`$ then $`L`$ is called strictly maximal if $`FL`$ and $`[L:F]=n`$.
###### Theorem 1.2.
Let $`A`$ be a finite-dimensional central simple algebra of degree $`n`$ with center $`F`$, $`L`$ be a strictly maximal subfield of $`A`$, $`L^{norm}`$ be the normal closure of $`L`$ over $`F`$, and $`๐ข=\mathrm{Gal}(L^{norm}/F)`$. Suppose $`๐ข`$ is generated by $`r`$ elements together with $`=\mathrm{Gal}(L^{norm}/L)`$. If either $`r2`$ or $`\{1\}`$ then $`\tau (A)r|๐ข|n+1`$.
If $`A`$ be a central simple algebra of degree $`n`$ then the upper bounds we have for $`\tau (A)`$ (or, equivalently, for $`d(n)`$), are all quadratic in $`n`$; see (1.1), (1.4) and Theorem 1.1. However, if we assume that $`A`$ is a crossed product, Theorem 1.2 yields an asymptotically better bound:
###### Corollary 1.3.
Suppose a group $`๐ข`$ of order $`n`$ can be generated by $`r2`$ elements. Then $`\tau (A)(r1)n+1`$ for any $`๐ข`$-crossed product central simple algebra $`A`$. In particular, $`\tau (A)(\mathrm{log}_2(n)1)n+1`$, for any crossed product central simple algebra of degree $`n4`$. Here, as usual, $`x`$ denotes the largest integer $`x`$. โ
Recall that $`A`$ is called a $`๐ข`$-crossed product if it contains a strictly maximal subfield $`L`$, such that $`L/F`$ is a Galois extension and $`\mathrm{Gal}(L/F)=๐ข`$; cf. \[Row<sub>1</sub>, Definition 3.1.23\]. Thus the first assertion of the corollary is an immediate consequence of Theorem 1.2. The second assertion follows from the first, because any group of order $`n`$ can be generated by $`r\mathrm{log}_2(n)`$ elements. (Indeed, $`|<๐ข_0,g>|2|๐ข_0|`$ for any subgroup $`๐ข_0`$ of $`๐ข`$ and any $`g๐ข๐ข_0`$.) Note also that $`\mathrm{log}_2(n)2`$ for any $`n4`$.
The case of central simple algebras of degree 4 is of special interest since, by a theorem of Albert, every such algebra is a $`/2\times /2`$-crossed product; see e.g., \[Row<sub>1</sub>, Theorem 3.2.28\]. Thus Corollary 1.3 says that $`d(4)=\tau (\mathrm{UD}(4))5`$. On the other hand, $`d(4)4`$ by (1.3). This proves:
###### Corollary 1.4.
$`d(4)=\mathrm{ed}(\mathrm{PGL}_4)=4`$ or $`5`$. โ
The rest of this paper is structured as follows. In Section 2 we discuss preliminary material from invariant theory and the theory of $`๐ข`$-lattices. In Section 3 we explain how $`๐ข`$-lattices can be used to give an upper bound on essential dimensions of certain groups. We prove Theorem 1.1 in Section 4 and Theorem 1.2 in Section 5. In Section 6 we show that the methods of this paper cannot be used to decide whether the exact value of $`d(4)`$ equals $`4`$ or $`5`$.
## 2. Preliminaries
### 2.1. $`G`$-varieties
A $`G`$-variety $`X`$ is an algebraic variety with a (regular) action of an algebraic group $`G`$. If $`G`$ acts freely (i.e., with trivial stabilizers) on a dense open subset of $`X`$, then $`X`$ is called a generically free $`G`$-variety.
A dominant rational map $`\pi :XY`$ is called the rational quotient map if $`k(Y)=k(X)^G`$ and $`\pi ^{}:k(Y)k(X)`$ is the natural inclusion $`k(X)^Gk(X)`$. We will usually denote the rational quotient $`Y`$ by $`X/G`$; note that $`X/G`$ is only defined up to birational equivalence. By a theorem of Rosenlicht a rational quotient map separates points in general position in $`X`$; see \[Ros<sub>1</sub>, Theorem 2\] and \[Ros<sub>2</sub>\] (also cf. \[PV, Theorem 2.3\]). In other words, there exists a dense open subset $`U`$ of $`X`$ such that $`f`$ is regular on $`U`$ and $`x,yU`$ lie in the same $`G`$-orbit iff $`f(x)=f(y)`$.
We will not generally assume that $`X`$ is irreducible; however, we will always require $`X`$ to be primitive. This means that $`G`$ transitively permutes the irreducible components of $`X`$; equivalently, $`X/G`$ is irreducible, i.e., $`k(X)^G`$ is a field; cf. \[Re<sub>2</sub>, Section 2.2\]. Note that an irreducible $`G`$-variety is always primitive, and, if $`G`$ is connected, a primitive variety is necessarily irreducible. Thus the notion of a primitive variety is only of interest if the group $`G`$ is disconnected.
If $`N`$ is a normal subgroup of $`G`$ then the $`G`$-action on $`X`$ induces a (rational) $`G/N`$-action on $`X/N`$; moreover, one can choose a model $`Y`$ of $`X/N`$ such that the $`G/N`$-action on $`Y`$ is regular; see \[PV, Proposition 2.6\]\[Re<sub>2</sub>, Remark 2.6\].
###### Lemma 2.1.
Let $`N`$ be a normal subgroup of $`G`$ and let $`X`$ be a $`G`$-variety. Then $`X`$ is generically free as a $`G`$-variety if and only if
1. $`X`$ is generically free as an $`N`$-variety and
2. $`X/N`$ is generically free as a $`G/N`$-variety.
###### Proof.
Assume (a) and (b) hold. Choose $`x`$ in $`X`$ in general position and suppose $`g\mathrm{Stab}(x)`$. Then (b) implies that $`gN`$ and (a) says that $`g=1`$. This shows that the $`G`$-action on $`X`$ is generically free. The converse in proved in a similar manner. โ
### 2.2. $`(G,H)`$-sections and compressions
Let $`X`$ be a $`G`$-variety and let $`\pi :XX/G`$ be the rational quotient map. Furthermore, let $`H`$ be a closed subgroup of $`G`$. An $`H`$-invariant subvariety $`S`$ of $`X`$ is called a $`(G,H)`$-section if the following conditions are satisfied.
(i) $`\pi (S)`$ contains an open dense subset of $`X/G`$ and
(ii) if $`x`$ is a point in general position in $`S`$ then $`gxS`$ if and only if $`gH`$.
Recall that a $`G`$-compression $`XY`$ is a dominant rational map of generically free $`G`$-varieties.
###### Lemma 2.2.
Suppose $`S`$ is a $`(G,H)`$-section of $`X`$. Then
1. $`k(S)^H=k(X)^G`$ and
2. any $`H`$-compression $`SS^{}`$ lifts to a $`G`$-compression $`XX^{}`$, where $`S^{}`$ is a $`(G,H)`$-section of $`X^{}`$.
###### Proof.
(a) See \[PV, Section 2.8\]\[Po, 1.7.2\] or \[Re<sub>2</sub>, Lemma 2.11\].
(b) Note that $`X`$ and $`G_HS`$ are birationally equivalent as $`G`$-varieties, where $`G_HS`$ is defined as $`G\times S/H`$ for the $`H`$-action given by $`h(g,s)=(gh^1,hs)`$; see \[Po, Theorem 1.7.5\] or \[Re<sub>2</sub>, Lemma 2.14\]. Now we set $`X^{}=G_HS^{}`$ and extend $`f`$ to a rational map $`XX^{}`$ by $`f(g,s)(g,f(s))`$. It is easy to see that this map has the desired properties; cf. the proof of \[Re<sub>2</sub>, Lemma 4.1\]. โ
### 2.3. $`๐ข`$-lattices
Let $`๐ข`$ be a finite group. A *$`๐ข`$-lattice* is a (left) module over the integral group ring $`[๐ข]`$ that is free of finite rank as a $``$-module. A $`๐ข`$-lattice $`M`$ is called
* *faithful* if the structure map $`๐ข\mathrm{Aut}_{}(M)`$ is injective, and
* a *permutation lattice* if $`M`$ has a $``$-basis that is permuted by $`๐ข`$.
For any $`๐ข`$-lattice $`M`$, the $`๐ข`$-action on $`M`$ extends canonically to actions of $`๐ข`$ on the group algebra $`k[M]`$ and on the field of fractions $`k(M)`$ of $`k[M]`$. The operative fact concerning the $`๐ข`$-fields (i.e., fields with $`๐ข`$-action) of the form $`k(M)`$ for our purposes is the following result of Masuda \[M\]; cf. also \[L, proof of Prop. (1.5)\].
###### Proposition 2.3.
Let $`0MEP0`$ be an exact sequence of $`๐ข`$-lattices with $`M`$ faithful and $`P`$ permutation. Then, as $`๐ข`$-fields, $`k(E)k(M)(t_1,\mathrm{},t_r)`$, where the elements $`t_i`$ are $`๐ข`$-invariant and transcendental over $`k(M)`$ and $`r=\mathrm{rank}P`$.
### 2.4. The symmetric and exterior squares
Let $`M`$ be a $`๐ข`$-lattice. By definition, the symmetric square $`\mathrm{๐ฒ๐๐}^2M`$ is the quotient of $`M^2=MM`$ modulo the subgroup generated by the elements $`mm^{}m^{}m`$ for $`m,m^{}M`$. Similarly, the exterior square $`^2M`$ is the quotient of $`M^2`$ modulo the subgroup generated by the elements $`mm`$, as $`m`$ ranges over $`M`$. The action of $`๐ข`$ on $`M^2`$ restricts down to $`\mathrm{๐ฒ๐๐}^2M`$ and $`^2M`$, making each a $`๐ข`$-lattice. The $`๐ข`$-lattice $`^2M`$ can be identified with the sublattice of *antisymmetric tensors* in $`M^2`$, that is,
$$\stackrel{2}{}M๐ _2^{}(M)=\{xM^2x^\tau =x\}$$
where $`\tau :M^2M^2`$ is the switch $`(mm^{})^\tau =m^{}m`$; see \[Bou, Exerc. 8 on p. A III.190\]. Furthermore, $`๐ _2^{}(M)`$ is exactly the kernel of the canonical map $`M^2\mathrm{๐ฒ๐๐}^2M`$. Hence, we have an exact sequence of $`๐ข`$-lattices
$$0\stackrel{2}{}MM^2\mathrm{๐ฒ๐๐}^2M0.$$
(2.1)
## 3. Groups of the form $`T_{n1}>๐ข`$ and lattices
### 3.1. Notations
In this section we shall focus on the following situation. Let $`T_{n1}=(๐พ_m)^n/\mathrm{\Delta }`$ be the (diagonal) maximal torus of $`\mathrm{PGL}_n`$; here $`\mathrm{\Delta }๐พ_m`$ diagonally embedded in $`(๐พ_m)^n`$. Recall that $`๐ฎ_n`$ acts on $`T_{n1}`$ by permuting the $`n`$ copies of $`๐พ_m`$ and that the normalizer $`N(T_{n1})`$ of $`T_{n1}`$ in $`\mathrm{PGL}_n`$ is isomorphic to $`T_{n1}>๐ฎ_n`$. We shall be interested in subgroups of $`N(T_{n1})`$ of the form $`T_{n1}>๐ข`$, where $`๐ข`$ is a subgroup of $`๐ฎ_n`$. These groups have two properties that will be important to us in the sequel: (i) $`T_{n1}>๐ข`$-varieties and their compressions can be constructed from $`๐ข`$-lattices and (ii) certain $`(\mathrm{PGL}_n,T_{n1}>๐ข)`$-sections will naturally come up in the proofs of Theorems 1.1 and 1.2. In this section we will focus on the relationship between $`T_{n1}>๐ข`$-varieties and $`๐ข`$-lattices.
### 3.2. $`T_{n1}>๐ข`$-varieties and $`๐ข`$-lattices
Suppose we are given a morphism
$$f:MA_{n1}$$
of $`๐ข`$-lattices, where $`A_{n1}`$ is the root lattice defined in (1.5). Note that $`A_{n1}X_{}(T_{n1})`$ as an $`๐ฎ_n`$-lattice (and hence as $`๐ข`$-lattice), where $`X_{}(T_{n1})`$ is the lattice of characters of $`T_{n1}`$. We will always identify $`A_{n1}`$ with $`X_{}(T_{n1})`$.
We will now associate to $`f`$ a $`T_{n1}>๐ข`$-variety $`X_f`$ as follows. Let $`๐ข`$ act on $`k[M]`$ as usual and define a $`T_{n1}`$-action on $`k[M]`$ by putting
$$t(m)=f(m)(t)m(tT_{n1},mM).$$
One easily checks that, by $`k`$-linear extension of this rule, one obtains a well-defined action of $`T_{n1}`$ by automorphism on $`k[M]`$. Moreover, for $`tT_{n1}`$, $`g๐ข`$ and $`mM`$, one calculates
$$t(gm)=f(gm)(t)gm=[gf(m)](t)gm=f(m)(t^g)gm=[gt^g](m);$$
so the actions of $`๐ข`$ and $`T_{n1}`$ combine to yield a locally finite action of $`T_{n1}>๐ข`$ on $`k[M]`$ and thus an algebraic action on $`X_f=\mathrm{Spec}k[M]`$.
###### Lemma 3.1.
The $`T_{n1}>๐ข`$-variety $`X_f`$ is a generically free if and only if
1. $`f`$ is surjective and
2. $`\mathrm{Ker}(f)`$ is a faithful $`๐ข`$-lattice.
###### Proof.
Condition (a) is equivalent to saying that the $`T_{n1}`$-action on $`X_f`$ is generically free; cf., e.g., \[OV, Theorem 3.2.5\]. To interpret condition (b) geometrically, note that $`k(X_f)=k(M)`$ and $`k(X_f/T_{n1})=k(M)^{T_{n1}}=k(\mathrm{Ker}(f))`$. Thus condition (b) holds iff $`๐ข`$ acts faithfully on $`X_f/T_{n1}`$ or, equivalently, iff the $`๐ข`$-action on $`X_f/T_{n1}`$ is and generically free (the two notions coincide for finite groups). The desired conclusion now follows from Lemma 2.1. โ
### 3.3. Compressions and $`๐ข`$-lattices
In the sequel we will only be interested in generically free $`T_{n1}>๐ข`$-varieties. In particular, we will assume that $`f`$ is surjective and expand it into an exact sequence of $`๐ข`$-lattices:
$$0K=\mathrm{Ker}(f)M\stackrel{f}{}A_{n1}0,$$
(3.1)
where $`K`$ is a faithful $`๐ข`$-lattice. For future reference, we extract the following equality from the proof of Lemma 3.1:
$$k(X_f/T_{n1}>๐ข)=k(X_f)^{T_{n1}>๐ข}=[k(M)^{T_{n1}}]^๐ข=k(K)^๐ข.$$
(3.2)
We can now obtain information about $`T_{n1}>๐ข`$-compressions of $`X_f`$ by studying this sequence more closely.
###### Lemma 3.2.
Suppose that the exact sequence (3.1) extends to a commutative diagram
of $`๐ข`$-lattices, where $`K_0`$ is faithful, and the vertical map $`M_0M`$ is injective. Then there exists a $`T_{n1}>๐ข`$-compression $`X_fX_{f_0}`$.
###### Proof.
Since $`k(M_0)=k(X_{f_0})`$ is an $`T_{n1}>๐ข`$-invariant subfield of $`k(M)=k(X_f)`$, it defines a dominant $`T_{n1}>๐ข`$-equivariant map $`X_fX_{f_0}`$. Furthermore, by Lemma 3.1, the $`T_{n1}>๐ข`$-action on both $`X_f`$ and $`X_{f_0}`$ is generically free. Thus, the rational map $`X_fX_{f_0}`$ we have constructed is a $`T_{n1}>๐ข`$-compression. โ
### 3.4. Linearization and essential dimension
If the $`๐ข`$-lattice $`M`$ in Section 3.2 is a permutation lattice then the $`T_{n1}>๐ข`$-variety $`X_f`$ is birationally linearizable. To see this, fix a $``$-basis $`m_1,\mathrm{},m_r`$ of $`M`$ that is permuted by $`๐ข`$. Clearly, $`k(X_f)=k(M)=k(m_1,\mathrm{},m_r)`$. Thus, putting $`V_f=_ikm_i`$ we obtain a $`T_{n1}>๐ข`$-invariant $`k`$-subspace of $`k(X_f)`$ with $`k(X_f)=k(V_f)`$.
A similar argument goes through if $`M`$ to be *permutation projective*, i.e., $`M`$ is a direct summand of a permutation $`๐ข`$-lattice.
###### Lemma 3.3.
If $`M`$ is permutation projective in (3.1) then there is a $`T_{n1}>๐ข`$-compression $`VX_f`$ with $`V`$ a generically free linear $`T_{n1}>๐ข`$-variety. In particular, $`\mathrm{ed}(T_{n1}>๐ข)\mathrm{rank}K`$.
###### Proof.
Suppose $`MN=P`$, where $`P`$ is a permutation $`๐ข`$-lattice. Then the sequence (3.1) embeds in the obvious fashion in an exact sequence $`0KNP=MNA_{n1}0`$. In view of the foregoing and Lemma 3.2, this proves the first assertion.
To complete the proof, recall that the essential dimension of an algebraic group $`G`$ is defined as the smallest possible value $`dim(X/G)`$, where $`X`$ is a generically free $`G`$-variety so that there is a $`G`$-compression $`VX`$ with $`V`$ a generically free linear $`G`$-variety; see \[Re<sub>2</sub>, Definition 3.5\]. For $`G=T_{n1}>๐ข`$ and $`X=X_f`$, we have $`dim(X/G)=\mathrm{rank}K`$ by (3.2). โ
## 4. Proof of Theorem 1.1
### 4.1. Reduction to a lattice-theoretic problem
The universal division algebra $`\mathrm{UD}(n)`$ is represented by a class $`cH^1(Z(n),\mathrm{PGL}_n)`$. We can write $`\mathrm{UD}(n)=D_{Z(D)}Z(n)`$ if and only if $`c`$ lies in the image of the natural map
$$H^1(Z(D),\mathrm{PGL}_n)H^1(Z(n),\mathrm{PGL}_n)$$
(4.1)
Recall that for any finitely generated field extension $`L/k`$, an element of $`\alpha H^1(L,\mathrm{PGL}_n)`$ may also be interpreted as a $`\mathrm{PGL}_n`$-torsor, i.e., a generically free $`\mathrm{PGL}_n`$-variety $`X_\alpha `$ such that $`k(X_\alpha )^{\mathrm{PGL}_n}=L`$. Moreover, $`X_\alpha `$ is uniquely determined (up to birational isomorphism of $`\mathrm{PGL}_n`$-varieties), and the central simple algebra corresponding to $`\alpha `$ can be recovered as the algebra of $`\mathrm{PGL}_n`$-equivariant rational maps $`X_\alpha \mathrm{M}_n`$. In particular, $`X_c=\mathrm{M}_n\times \mathrm{M}_n`$, where $`\mathrm{PGL}_n`$ acts on $`\mathrm{M}_n\times \mathrm{M}_n`$ by simultaneous conjugation; to say that $`c`$ lies in the image of the map (4.1) is equivalent to saying that there exists a $`\mathrm{PGL}_n`$-compression $`\mathrm{M}_n\times \mathrm{M}_nX^{}`$ such that $`k(X^{})^{\mathrm{PGL}_n}=Z(D)`$. For a more detailed discussion of these facts and further references, see \[RY, Section 3\].
Denote the linear subspace of $`\mathrm{M}_n`$ consisting of diagonal matrices by $`D_n`$. It is easy to see that $`D_n\times \mathrm{M}_n`$ is a $`(\mathrm{PGL}_n,N(T_{n1}))`$-section of $`\mathrm{M}_n\times \mathrm{M}_n`$, where $`N(T_{n1})=T_{n1}>๐ฎ_n`$ is the normalizer of the maximal torus $`T_{n1}=(๐พ_m)^n/\mathrm{\Delta }`$ in $`\mathrm{PGL}_n`$, as in the previous section. (See Lemma 5.2 below for a more general fact.) Thus, in view of Lemma 2.2, we have the following
###### Reduction 4.1.
In order to prove Theorem 1.1 it is enough to show that there exists an $`N(T_{n1})`$-compression
$$D_n\times \mathrm{M}_nX$$
(4.2)
such that
1. $`dim(X)=\frac{1}{2}(n1)(n2)+n1`$ or equivalently, $`dim(X/N(T_{n1}))=\mathrm{trdeg}_kk(X)^{N(T_{n1})}=\frac{1}{2}(n1)(n2)`$.
2. $`Z(n)=k(\mathrm{M}_n\times \mathrm{M}_n)^{\mathrm{PGL}_n}=k(D_n\times \mathrm{M}_n)^{N(T_{n1})}`$ is purely transcendental over $`k(X)^{N(T_{n1})}`$.
Our construction of the compression (4.2) will be based on Lemma 3.2. In order to apply this lemma, we need to write the linear $`N(T_{n1})`$-variety $`D_n\times \mathrm{M}_n`$ birationally in the form $`X_f`$, where $`f`$ is as in (3.1). Let $`x_i`$ and $`y_{rs}`$ be the standard coordinates on $`D_n`$ and $`\mathrm{M}_n`$ respectively. In this coordinate system, the $`๐ฎ_n`$-action on $`D_n\times \mathrm{M}_n`$ is given by
$$\sigma (x_i)=x_{\sigma (i)}\text{and}\sigma (y_{rs})=y_{\sigma (r)\sigma (s)}.$$
Thus, monomials in these coordinates and their inverses form an $`๐ฎ_n`$-lattice isomorphic to $`M=U_nU_n^2`$, where $`U_n=^n`$ be the standard permutation $`๐ฎ_n`$-lattice. Moreover, an element $`t=(t_1,\mathrm{},t_n)`$ of $`T_{n1}=(๐พ_m)^n/\mathrm{\Delta }`$ acts on monomials in $`x_i`$, $`y_{rs}`$ by characters determined (multiplicatively) by
$$t(x_i)=x_i\text{and}t(y_{rs})=t_rt_s^1y_{rs}.$$
Denoting the standard basis of $`U_n`$ by $`b_1,\mathrm{},b_n`$ and defining $`f:M=U_nU_n^2A_{n1}`$ by $`f(b_i,b_rb_s)=b_rb_s`$, the above formulas give exactly the action of $`N(T_{n1})=T_{n1}>๐ฎ_n`$ on $`X_f`$ as described in (3.2). The exact sequence (3.1) for this $`f`$ is the Formanek โ Procesi exact sequence
$$0K=\mathrm{Ker}(f)U_nU_n^2\stackrel{f}{}A_{n1}0;$$
(4.3)
see \[F\]. Note that
$$KU_nU_nA_{n1}^2.$$
Here, the first copy of $`U_n`$ is mapped identically onto the first summand of $`U_nU_n^2`$, the second $`U_n`$ corresponds to the sublattice of $`U_n^2`$ consisting of the monomials in $`y_{ii}K`$, and $`A_{n1}^2`$ describes the kernel of $`f`$, restricted to the sublattice $`y_{rs}rsU_nA_{n1}U_n^2`$ ; cf. \[B, p. 3573\].
We now want to apply Lemma 3.2 to the above sequence, with $`K_0=^2A_{n1}`$. Recall that $`^2A_{n1}`$ may be viewed as a sublattice $`A_{n1}^2`$; see (2.1). Let
$$\phi :\stackrel{2}{}A_{n1}U_nU_nA_{n1}^2$$
(4.4)
be the natural embedding of $`^2A_{n1}`$ into the third component of $`U_nU_nA_{n1}^2`$. We remark that for $`n>3`$, $`๐ฎ_n`$ acts faithfully on $`^2A_{n1}`$; in fact, $`^2A_{n1}_{}`$ is the irreducible $`S_n`$-representation corresponding to the partition $`(n2,1^2)`$ of $`n`$; see, e.g., \[FH, Exerc. 4.6\].
Combining Reduction 4.1 with Lemma 3.2, we obtain:
###### Reduction 4.2.
Theorem 1.1 follows from Propositions 4.3 and 4.4 stated below.
###### Proposition 4.3.
For odd $`n5`$, $`k(U_nU_nA_{n1}^2)k(^2A_{n1})(y_1,\mathrm{},y_r)`$ as $`๐ฎ_n`$-fields, where the elements $`y_i`$ are $`๐ฎ_n`$-invariant and transcendental over $`k(^2A_{n1})`$. In particular, $`k(U_nU_nA_{n1}^2)^{๐ฎ_n}`$ is rational over $`k(^2A_{n1})^{๐ฎ_n}`$.
###### Proposition 4.4.
For odd $`n`$, there exists a commutative diagram of $`๐ฎ_n`$-lattices:
Here the first row is the Formanek โ Procesi sequence (4.3).
Indeed, Proposition 4.4 in conjunction with Lemma 3.2 yields an $`N(T_{n1})`$-compression
$$D_n\times \mathrm{M}_n\stackrel{}{}X_fX=X_{f_0},$$
and formula (3.2) further implies that $`k(D_n\times M_n)^{N(T_{n1})}=k(X_f)^{N(T_{n1})}=k(U_nU_nA_{n1}^2)^{๐ฎ_n}`$ and $`k(X)^{N(T_{n1})}=k(^2A_{n1})^{๐ฎ_n}`$. Thus, condition (i) in Reduction 4.1 is clearly satisfied and Proposition 4.3 ensures that (ii) holds as well.
### 4.2. Solution of the lattice-theoretic problem
Our proofs of Propositions 4.3 and 4.4 will be based on the following result from \[LL, Section 3.5\]. If $`๐ข`$ is a finite group, $``$ is a subgroup of $`๐ข`$ and $`M`$ a $`[]`$-module then $`M_{}^๐ข=[๐ข]_{[]}M`$ will denote the induced $`[๐ข]`$-module.
###### Proposition 4.5.
For odd $`n`$, there is an isomorphism of $`๐ฎ_n`$-lattices
$$\mathrm{๐ฒ๐๐}^2A_{n1}U_n_{๐ฎ_{n2}\times ๐ฎ_2}^{๐ฎ_n}U_n,$$
where $``$ has the trivial $`๐ฎ_n`$-action. In particular, $`\mathrm{๐ฒ๐๐}^2A_{n1}U_n`$ is a permutation lattice.
#### Proof of Proposition 4.3
First, applying Proposition 2.3 to the obvious sequence $`0U_nA_{n1}^2U_nU_nA_{n1}^2U_n0`$, we see that
$$k(U_nU_nA_{n1}^2)k(U_nA_{n1}^2)(t_1,\mathrm{},t_n)$$
as $`๐ฎ_n`$-fields.
Next, sequence (2.1) for $`M=A_{n1}`$ combined with Proposition 4.5 gives rise to an exact sequence of $`๐ฎ_n`$-lattices
$$0\stackrel{2}{}A_{n1}A_{n1}^2U_nP0,$$
where $`P=\mathrm{๐ฒ๐๐}^2A_{n1}U_n`$ is permutation. Applying Proposition 2.3 to this sequence, we deduce that
$$k(A_{n1}^2U_n)k(\stackrel{2}{}A_{n1})(x_1,\mathrm{},x_m)$$
as $`๐ฎ_n`$-fields.
Finally,
$`k(U_nU_nA_{n1}^2)`$ $``$ $`k(U_nA_{n1}^2)(t_1,\mathrm{},t_n)`$
$`=`$ $`k(A_{n1}^2U_n)(t_1,\mathrm{},t_{n1})`$
$``$ $`k({\displaystyle \stackrel{2}{}}A_{n1})(x_1,\mathrm{},x_m,t_1,\mathrm{},t_{n1})`$
as $`๐ฎ_n`$-fields, which proves the first assertion of Proposition 4.3. The second assertion is an immediate consequence of the first. โ
#### Proof of Proposition 4.4
Recall that the embedding $`\phi `$ of (4.4) is defined as the composition
$$\phi :\stackrel{2}{}A_{n1}\stackrel{\psi }{}A_{n1}^2U_nU_nA_{n1}^2,$$
where $`\psi `$ is the injection from (2.1) (with $`M=A_{n1}`$) and the second map identifies $`A_{n1}^2`$ with the third component of $`U_nU_nA_{n1}^2`$. We aim to show that $`\phi `$ together with sequence (4.3) will give rise to a commutative diagram as in the statement of Proposition 4.4. In other words, our goal is to show that the class in $`\mathrm{Ext}_{[๐ฎ_n]}(A_{n1},U_nU_nA_{n1}^2)`$ corresponding to the extension (4.3) belongs to the image of the map
$$\phi _{}:\mathrm{Ext}_{[๐ฎ_n]}(A_{n1},\stackrel{2}{}A_{n1})\mathrm{Ext}_{[๐ฎ_n]}(A_{n1},U_nU_nA_{n1}^2).$$
In fact, we will prove:
###### Lemma 4.6.
For odd $`n`$, the map $`\phi _{}`$ is surjective.
###### Proof.
We will tacitly use the following standard facts from homological algebra, valid for any finite group $`๐ข`$:
* If $`V`$ is a $`๐ข`$-module and $`M`$ an $``$-module for some subgroup $`๐ข`$ then $`\mathrm{Ext}_{[๐ข]}^{}(V,M_{}^๐ข)\mathrm{Ext}_{[]}^{}(V|_{},M)`$; see \[HS, Prop. IV.12.3\] and \[Br, Prop. III.5.9\]. For $`V=`$, the trivial $`๐ข`$-module, this isomorphism is the โShapiro isomorphismโ $`H^{}(๐ข,M_{}^๐ข)H^{}(,M)`$. In case, $`M`$ is actually a $`๐ข`$-module, the restriction map $`\mathrm{Res}_{}^๐ข:H^{}(๐ข,M)H^{}(,M)`$ factors through the Shapiro isomorphism:
$$\mathrm{Res}_{}^๐ข:H^{}(๐ข,M)\stackrel{\mu _{}}{}H^{}(๐ข,M_{}^๐ข)\stackrel{}{}H^{}(,M),$$
where $`\mu :MM_{}^๐ข`$ sends $`m_{g๐ข/}gg^1m`$; see \[Br, p. 81\].
* For any $`๐ข`$-lattices $`V`$ and $`W`$, $`\mathrm{Ext}_{[๐ข]}^{}(V,W)H^{}(๐ข,V^{}W)`$, where $`=_{}`$ and $`V^{}=\mathrm{Hom}_{}(V,)`$ is the dual $`๐ข`$-lattice; see \[Br, Prop. III.2.2\].
* If $`V`$ and $`W`$ are both permutation $`๐ข`$-lattices then $`\mathrm{Ext}_{[๐ข]}(V,W)=0`$; cf. \[L, Propositions 1.1, 1.2\].
Armed with these facts, we proceed as follows. First, $`\mathrm{Ext}_{[๐ฎ_n]}(A_{n1},U_n)=0`$, because $`U_n_{๐ฎ_{n1}}^{๐ฎ_n}`$ and $`A_{n1}|_{๐ฎ_{n1}}U_{n1}`$. Therefore, it suffices to show that the map
$$\psi _{}:\mathrm{Ext}_{[๐ฎ_n]}(A_{n1},\stackrel{2}{}A_{n1})\mathrm{Ext}_{[๐ฎ_n]}(A_{n1},A_{n1}^2)$$
is surjective. But the extension (2.1) (for $`M=A_{n1}`$) gives rise to an exact sequence
$$\mathrm{Ext}_{[๐ฎ_n]}(A_{n1},\stackrel{2}{}A_{n1})\stackrel{\psi _{}}{}\mathrm{Ext}_{[๐ฎ_n]}(A_{n1},A_{n1}^2)\mathrm{Ext}_{[๐ฎ_n]}(A_{n1},\mathrm{๐ฒ๐๐}^2A_{n1}).$$
Therefore, it suffices to prove:
$$\mathrm{Ext}_{[๐ฎ_n]}(A_{n1},\mathrm{๐ฒ๐๐}^2A_{n1})=0\text{for odd }n\text{.}$$
For this, we use the isomorphism $`\mathrm{๐ฒ๐๐}^2A_{n1}U_n_๐ข^{๐ฎ_n}U_n`$ of Proposition 4.5, where we have put $`๐ข=๐ฎ_{n2}\times ๐ฎ_2`$ for simplicity. This isomorphism entails
$`\mathrm{Ext}_{[๐ฎ_n]}(A_{n1},\mathrm{๐ฒ๐๐}^2A_{n1})`$ $``$ $`\mathrm{Ext}_{[๐ฎ_n]}(A_{n1},_๐ข^{๐ฎ_n})`$
$``$ $`\mathrm{Ext}_{[๐ข]}(A_{n1}|_๐ข,)`$
$``$ $`H^1(๐ข,A_{n1}^{}).`$
Dualizing the augmentation sequence $`0A_{n1}U_n\stackrel{ฯต}{}0`$ we obtain an exact sequence $`0\stackrel{ฯต^{}}{}U_nA_{n1}^{}0`$, where $`ฯต^{}(1)=_ie_i`$ , the sum of the natural basis elements of $`U_n`$. This sequence, viewed as exact sequence of $`๐ข`$-lattices, in turn yields an exact sequence
$$H^1(๐ข,U_n)=0H^1(๐ข,A_{n1}^{})H^2(๐ข,)H^2(๐ข,U_n).$$
Thus, it suffices to show that $`H^2(๐ข,)H^2(๐ข,U_n)`$ is injective. As a $`๐ข`$-module, $`U_n=VW`$, where $`V=_{i=1}^{n2}e_i_{๐ฎ_{n3}\times ๐ฎ_2}^๐ข`$ and $`W=e_{n1}e_n_{๐ฎ_{n2}}^๐ข`$. Therefore, the Shapiro isomorphism gives
$$H^2(๐ข,U_n)H^2(๐ฎ_{n3}\times ๐ฎ_2,)H^2(๐ฎ_{n2},)$$
and the map $`H^2(๐ข,)H^2(๐ข,U_n)`$ becomes the restriction map
$$\mathrm{Res}_{๐ฎ_{n3}\times ๐ฎ_2}^๐ข\times \mathrm{Res}_{๐ฎ_{n2}}^๐ข:H^2(๐ข,)H^2(๐ฎ_{n3}\times ๐ฎ_2,)H^2(๐ฎ_{n2},).$$
This map is indeed injective, as is easily seen by identifying $`H^2(๐ข,)`$ in the usual fashion with $`\mathrm{Hom}(๐ข,/)`$, and similarly for the subgroups $`๐ฎ_{n3}\times ๐ฎ_2`$ and $`๐ฎ_{n2}`$. This finishes the proof of the Lemma, and hence, of Proposition 4.4 and of Theorem 1.1. โ
## 5. Proof of Theorem 1.2
### 5.1. General observations
The following notations will be used throughout this section:
| $`F`$ | will be a field containing $`k`$; |
| --- | --- |
| $`A`$ | will be a finite-dimensional central simple algebra with center $`F`$; |
| $`L`$ | will be a strictly maximal commutative subfield of $`A`$; |
| $`n`$ | denotes the degree of $`A`$, so $`dim_FA=n^2`$ and $`[L:F]=n`$; |
| $`๐ข`$ | is the Galois group of the normal closure $`L^{norm}`$ of $`L`$ over $`F`$; |
| $`T_{n1}`$ | denotes the diagonal maximal torus of $`\mathrm{PGL}_n`$. |
###### Reduction 5.1.
In the course of proving Theorem 1.2 we may assume without loss of generality that $`F`$ is a finitely generated field extension of $`k`$.
###### Proof.
Indeed, choose a primitive element $`x`$ for the extension $`L/F`$ and complete $`1,x,\mathrm{},x^{n1}`$ to an $`F`$-basis $`e_1,\mathrm{},e_{n^2}`$ of $`A`$, where $`e_i=x^{i1}`$ for $`i=1,\mathrm{},n`$. Let $`c_{rs}^t`$ be the structure constants for $`A`$ in this basis, i.e.,
$$e_re_s=\underset{t=1}{\overset{n^2}{}}c_{rs}^te_t$$
(5.1)
for every $`r,s=1,\mathrm{},n^2`$. Then $`A=A_0_{F_0}F`$, where $`F_0=k(c_{rs}^t)`$ and $`A_0`$ is the $`n^2`$-dimensional $`F_0`$-algebra spanned by $`e_1,\mathrm{},e_{n^2}`$ with multiplication given by (5.1). Moreover, $`A_0`$ is a central simple algebra of degree $`n`$ with center $`F_0`$ and $`A_0`$ contains the strictly maximal subfield $`L_0=F_0(x)`$ whose normal closure $`L_0^{norm}`$ has Galois group $`๐ข`$ over $`F_0`$. Clearly, $`\tau (A)\tau (A_0)`$. Thus, after replacing $`A`$ by $`A_0`$ we may assume that $`F`$ is finitely generated extension of $`k`$. โ
Next we pass from central simple algebras to generically free $`\mathrm{PGL}_n`$-varieties, as we did at the beginning of Section 4. Recall that a central simple algebra $`A`$ of degree $`n`$ with center $`F`$ defines a class $`H^1(F,\mathrm{PGL}_n)`$. This class, in turn, gives rise to a $`\mathrm{PGL}_n`$-torsor $`X_A`$ over $`F`$. Since we are assuming $`F`$ is a finitely generated extension of $`k`$, $`X_A`$ is a generically free $`\mathrm{PGL}_n`$-variety; moreover, $`F=k(X_A/\mathrm{PGL}_n)`$ and $`A`$ can be recovered from $`X_A`$ as the algebra of $`\mathrm{PGL}_n`$-equivariant rational maps $`X_A\mathrm{M}_n`$; cf. \[RY, Section 3\].
The algebras $`A`$ we are concerned with in the context Theorem 1.2 are of a special form: they have a maximal subfield $`L/F`$ such that $`\mathrm{Gal}(L^{norm}/F)`$ is the given group $`๐ข`$. We would like to know how this extra structure is reflected in the geometry of the $`\mathrm{PGL}_n`$-variety $`X_A`$. The following lemma gives a partial answer. We continue to let $`T_{n1}`$ denote the diagonal maximal torus of $`\mathrm{PGL}_n`$, as in Section 3.1.
###### Lemma 5.2.
$`X_A`$ has a $`(\mathrm{PGL}_n,T_{n1}>๐ข)`$-section.
Note that the group $`๐ข`$ comes with a natural permutation representation of $`๐ข`$ on the $`n`$ embeddings $`LL^{norm}`$ over $`F`$. This permutation representation gives an embedding $`\alpha :๐ข๐ฎ_n`$ that we use to define the semidirect product $`T_{n1}>๐ข`$. Note that $`\alpha `$ is only defined up to an inner automorphism of $`๐ฎ_n`$, since the $`n`$ embedding $`LL^{norm}`$ are not naturally in a 1-1 correspondence with $`\{1,\mathrm{},n\}`$. Relabeling these embeddings (or, equivalently, reordering the roots of a defining polynomial for $`L/F`$) will cause $`T_{n1}>๐ข`$ to be replaced by a conjugate subgroup of $`\mathrm{PGL}_n`$; the corresponding section will be translates of each other by elements of $`๐ฎ_n\mathrm{PGL}_n`$. Thus it is sufficient to verify that such a section exists for a particular numbering of the embeddings $`LL^{norm}`$.
###### Proof.
Suppose $`L=F(r)`$ for some $`rL`$ and $`r=r_1,r_2,\mathrm{},r_n`$ are the conjugates of $`r`$ in $`L^{norm}`$. As we remarked above, the order of the roots is not intrinsic; however, we choose it at this point, and it will not be changed in the sequel. The group $`๐ข`$ permutes $`r_1,\mathrm{},r_n`$ transitively via a permutation representation $`\alpha :๐ข๐ฎ_n`$.
Now let $`x_1,\mathrm{},x_n`$ be commuting independent variables over $`k`$. The symmetric group acts on the polynomial ring $`k[x_1,\mathrm{},x_n]`$ by permuting these variables; composing this action with $`\alpha `$, we obtain a (permutation) action of $`๐ข`$ on $`k[x_1,\mathrm{},x_n]`$. Note that for every $`f(x_1,\mathrm{},x_n)k[x_1,\mathrm{},x_n]^๐ข`$, we have $`f(r_1,\mathrm{},r_n)F`$; we shall write this element of $`F`$ as $`f_r`$.
Recall that $`A`$ is the algebra of $`\mathrm{PGL}_n`$-equivariant rational maps $`X_A\mathrm{M}_n`$. We claim that
$$S=\{xX_A:\begin{array}{c}r(x)=\mathrm{diag}(\lambda _1,\mathrm{},\lambda _n)\text{is a diagonal matrix}\\ \text{and}\\ f(\lambda _1,\mathrm{},\lambda _n)=f_r(x)\text{for every }fk[x_1,\mathrm{},x_n]^๐ข\end{array}\}$$
is an $`(\mathrm{PGL}_n,T_{n1}>๐ข)`$-section $`S`$ of $`X_A`$. Note that $`S`$ is a $`T_{n1}>๐ข`$-invariant subvariety of $`X_A`$: indeed, $`T_{n1}`$ acts trivially on the set of diagonal matrices, and $`f_rF`$ is a $`\mathrm{PGL}_n`$-invariant rational function on $`X_A`$. Thus we need to show that
(i) $`\mathrm{PGL}_nx`$ intersects $`S`$ for $`x`$ in general position in $`X_A`$ and
(ii) $`gsSgT_{n1}>๐ข`$ for $`s`$ in general position in $`S`$;
cf. Section 2.2.
Let $`\pi :X_AX_A/\mathrm{PGL}_n`$ be the rational quotient map. Recall that $`k(X_A/\mathrm{PGL}_n)=k(X_A)^{\mathrm{PGL}_n}=F`$. Suppose $`p(t)=t^n+a_1t^{n1}+\mathrm{}+a_n`$ is the minimal polynomial of $`rL`$ over $`F`$. Note $`a_1,\mathrm{},a_nF`$ are $`\mathrm{PGL}_n`$-invariant rational functions on $`X_A`$; in particular, for $`xX_A`$ in general position, the matrix $`r(x)\mathrm{M}_n`$ satisfies the polynomial $`p_x(t)=t^n+a_1(x)t^{n1}+\mathrm{}+a_n(x)k[t]`$. Since $`p(t)`$ is an irreducible polynomial over $`F`$, its discriminant $`\delta `$ is a non-zero element of $`F`$, i.e., a non-zero $`\mathrm{PGL}_n`$-invariant rational function $`X_A`$. This means that for $`xX_A`$ in general position (i.e., away from the zero locus of $`\delta `$ and the indeterminacy locus of $`r`$), the $`n\times n`$-matrix $`r(x)`$ has distinct eigenvalues. We conclude that in this case $`p_x(t)`$ is the characteristic polynomial for $`r(x)`$; in particular, the eigenvalues of $`r(x)`$ are precisely the roots of $`p_x(t)`$.
To see what these eigenvalues are more explicitly, let $`YX_A/\mathrm{PGL}_n`$ be a rational map of varieties induced by the field extension $`L^{norm}/F`$. Then $`r_1,\mathrm{},r_nL^{norm}`$ are rational functions on $`Y`$. Thus we have the following diagram of rational maps
Suppose $`x`$ be a point of $`X_A`$ in general position and $`y`$ is a point of $`Y`$ lying above $`\pi (x)`$. Since $`r_1,\mathrm{},r_n`$ are distinct elements of $`L^{norm}=k(Y)`$, $`\lambda _1=r_1(y),\mathrm{},\lambda _n=r_n(y)`$ are the $`n`$ distinct roots of $`p_x(t)`$, i.e., the $`n`$ distinct eigenvalues of the $`n\times n`$-matrix $`r(x)`$.
Proof of (i): In view of the above discussion, we may assume without loss of generality that $`r(x)`$ is diagonalizable. In other words, the $`\mathrm{PGL}_n`$-orbit of $`r(x)`$ in $`\mathrm{M}_n`$ contains the diagonal matrix $`\mathrm{diag}(\lambda _1,\mathrm{},\lambda _n)`$ or equivalently, $`r(x^{})=\mathrm{diag}(\lambda _1,\mathrm{},\lambda _n)`$ for some $`x^{}\mathrm{PGL}_nx`$. It remains to show that $`x^{}S`$. Indeed, for any $`fk[x_1,\mathrm{},x_n]^๐ข`$, we have
$$f_r(x^{})=f(r_1(y),\mathrm{},r_n(y))=f(\lambda _1,\mathrm{},\lambda _n),$$
as desired. This completes the proof of (i).
Proof of (ii): Let $`x`$ be a point of $`S`$ in general position. We may assume without loss of generality that the eigenvalues $`\lambda _i`$ of the diagonal matrix $`r(x)=\mathrm{diag}(\lambda _1,\mathrm{},\lambda _n)`$ are distinct. Suppose $`gxS`$ for some $`g\mathrm{PGL}_n`$. Then $`gr(x)g^1`$ is again diagonal; hence, $`gT_{n1}>๐ฎ_n`$. In other words, $`g=t\sigma `$, where $`t`$ is a diagonal matrix and $`\sigma `$ is a permutation matrix; our goal is to show that $`\sigma ๐ข`$. Indeed, by the definition of $`S`$, $`\sigma `$ has the property that $`f(\lambda _{\sigma (1)},\mathrm{},\lambda _{\sigma (n)})=f(\lambda _1,\mathrm{},\lambda _n)`$ for every $`fk[x_1,\mathrm{},x_n]^๐ข`$. Since $`๐ข`$-invariant regular functions separate the orbits of the permutation $`๐ข`$-action on affine $`n`$-space (this is true for any finite group action on an affine variety; see, e.g., \[PV, Section 0.4\]), the points $`(\lambda _{\sigma (1)},\mathrm{},\lambda _{\sigma (n)})`$ and $`(\lambda _1,\mathrm{},\lambda _n)`$ are in the same $`๐ข`$-orbit for this action. On the other hand, since $`\lambda _1,\mathrm{},\lambda _n`$ are distinct, this is only possible if $`\sigma ๐ข`$. This completes the proof of (ii) and thus of Lemma 5.2. โ
###### Remark 5.3.
One can show that the converse of Lemma 5.2 is also true: if $`X_A`$ has a $`(\mathrm{PGL}_n,T_{n1}>๐ข)`$-section then $`A`$ contains a strictly maximal subfield $`L`$ such that $`\mathrm{Gal}(L^{norm}/F)=๐ข`$. Since this result is not needed in the sequel, we omit the proof.
### 5.2. Conclusion of the proof
We are now ready to finish the proof of Theorem 1.2. Let $`A`$ be a central simple algebra of degree $`n`$ and let $`X_A`$ denote the $`\mathrm{PGL}_n`$-variety associated to $`A`$. Recall that $`\tau (A)=\mathrm{ed}(X_A,\mathrm{PGL}_n)`$; see \[Re<sub>2</sub>, Theorem 8.8 and Lemma 9.1\]. Moreover, if $`X`$ has a $`(\mathrm{PGL}_n,H)`$-section $`S`$ then $`\mathrm{ed}(X,\mathrm{PGL}_n)\mathrm{ed}(S,H)\mathrm{ed}(H)`$; see \[Re<sub>2</sub>, Lemma 4.1 and Definition 3.5\]. Applying these inequalities to the situation described by Lemma 5.2, with $`H=T_{n1}>๐ข`$, we see that
$$\tau (A)\mathrm{ed}(T_{n1}>๐ข).$$
(5.2)
Thus Theorem 1.2 is a consequence of the following:
###### Lemma 5.4.
Suppose $`๐ข`$ is a transitive subgroup of $`๐ฎ_n`$ generated by the subgroup $`=๐ข๐ฎ_{n1}`$ together with elements $`g_1,\mathrm{},g_r`$. Assume that either $`r2`$ or $`\{1\}`$. Then $`\mathrm{ed}(T_{n1}>๐ข)r|๐ข|n+1`$.
###### Proof.
By Lemma 3.3, it suffices to construct an exact sequence (3.1) with $`M`$ permutation projective and $`K`$ faithful having $`\mathrm{rank}K=r|๐ข|n+1`$. To this end, note that $`U_n[๐ข/]`$ as $`๐ข`$-lattices. Let $`\overline{}:[๐ข][๐ข/]=U_n`$ denote the canonical epimorphism; the kernel of $`\overline{}`$ is $`[๐ข]\omega `$, where $`\omega `$ denotes the augmentation ideal of $`[๐ข]`$; cf. \[Pa\]. Then $`_i[๐ข]\overline{(g_i1)}=A_{n1}`$; see \[Pa, Lemma 3.1.1\]. Therefore, we obtain an epimorphism of $`๐ข`$-lattices
$$f:M=[๐ข]^rA_{n1},(\alpha _1,\mathrm{},\alpha _r)\underset{i}{\overset{r}{}}\alpha _i\overline{(g_i1)}.$$
Put $`K=\mathrm{Ker}f`$; so $`K`$ certainly has the required rank. For faithfulness, we may consider $`K`$ instead of $`K`$ and work over the semisimple algebra $`[๐ข]`$. Since $`f`$ and $`\overline{}`$ are split, we have $`[๐ข]`$-isomorphisms $`(A_{n1})(K)[๐ข]^r`$ and $`(A_{n1})[๐ข]\omega [๐ข]`$. Therefore,
$$K[๐ข]^{r1}[๐ข]\omega .$$
If $`r2`$ then $`[๐ข]^{r1}`$ is $`๐ข`$-faithful, and if $`\{1\}`$ then $`\omega `$ is $``$-faithful and so $`[๐ข]\omega (\omega )_{}^๐ข`$ is $`๐ข`$-faithful. In either case, $`K`$ is faithful, and hence so is $`K`$, as desired. โ
## 6. Algebras of degree four
Recall that Corollary 1.4 asserts that $`d(4)`$ equals $`4`$ or $`5`$. Whether the true value of $`d(4)`$ is four or five is an open question. The purpose of this section is to show that this question cannot be resolved by the methods of this paper.
For the rest of this section we will identify the Klein 4-group $`๐ฑ=(/2)\times (/2)`$ with the subgroup of $`๐ฎ_4`$ generated by $`(12)(34)`$ and $`(13)(24)`$. Let $`A_3`$ be the augmentation (or root) lattice of $`๐ฎ_4`$, restricted to $`๐ฑ`$; see (1.5). In other words,
$$A_3\omega ๐ฑ,$$
(6.1)
where $`\omega ๐ฑ`$ is the augmentation ideal of the group ring $`[๐ฑ]`$.
We now briefly recall how we arrived at the bound $`d(4)5`$. First of all, since $`\mathrm{UD}(4)`$ is a $`๐ฑ`$-crossed product, $`d(4)=\tau (\mathrm{UD}(4))\mathrm{ed}(T_3>๐ฑ)`$; see (5.2). Secondly, Lemma 3.2 tells us that $`\mathrm{ed}(T_3>๐ฑ)\mathrm{rank}(K_0)`$, for any commutative diagram
(6.2)
of $`๐ฑ`$-lattices with $`M`$ permutation projective, $`K_0`$ faithful, and $`\phi `$ is injective. Finally, in the course of the proof of Lemma 5.4 (with $`๐ข=๐ฑ`$ and $`r=2`$) we constructed a particular diagram (6.2) with $`M=M_0=[๐ฑ]^2`$ and $`\mathrm{rank}(K_0)=5`$. This gave us the bound $`d(4)5`$. The question we will now address is whether or not one can sharpen this bound by choosing a different diagram (6.2). The following proposition shows that the answer is โnoโ.
###### Proposition 6.1.
Let (6.2) be a commutative diagram of $`๐ฑ=(/2)\times (/2)`$-lattices with $`M`$ permutation projective. Then $`K_0`$ is faithful and $`\mathrm{rank}K_05`$.
Note that in the setting of Lemma 3.2 we assumed that (i) $`K_0`$ is faithful and (ii) $`\phi `$ is injective. Here we see that (i) is automatic and (ii) is irrelevant for the rank estimate.
###### Proof.
Since $`M`$ is permutation projective, we have $`H^1(,M)=0=H^1(,M)`$ for every subgroup $``$ of $`๐ฑ`$. (This condition is actually equivalent to $`M`$ being permutation projective; see \[C-TS, Proposition 4\].) Consequently, (6.2) yields a commutative diagram
Thus, $`\delta _0`$ is mono. Since $`H^1(๐ฑ,A_3)/4`$ and $`H^1(,A_3)/2`$ for any nonidentity cyclic subgroup $``$ of $`๐ฑ`$ (see \[LL, Lemma 4.3\]), we obtain
$$/4H^2(๐ฑ,K_0)\text{and}H^2(,K_0)0\text{ for all }1๐ฑ\text{.}$$
(6.3)
Similarly, $`0=H^1(,M)`$ implies $`H^1(,A_3)\widehat{H}^0(,K_0)`$. Using the identification (6.1), we have $`H^1(๐ฑ,A_3)\omega ๐ฑ/(\omega ๐ฑ)^2/2/2`$. Thus:
$$/2/2\widehat{H}^0(,K_0).$$
(6.4)
We will show that (6.3) forces $`K_0`$ to be faithful, and (6.3) and (6.4) together imply that $`\mathrm{rank}K_05`$. The discussion below could be shortened somewhat by a reference to \[N\]; however, for the sake of completeness, we will give a self-contained argument.
###### Lemma 6.2.
Let $`L`$ be a $`๐ฑ`$-lattice, $`1x๐ฑ`$ and $`L_\pm =\{lLxl=\pm l\}`$. If $`L|_x=L_+L_{}`$ then $`2H^2(๐ฑ,L)=0`$.
###### Proof.
Since $`L_+`$ and $`L_{}`$ are $`๐ฑ`$-sublattices of $`L`$, we may assume $`L=L_+`$ or $`L=L_{}`$. Write $`๐ฑ=x,y`$. Then $`y`$-sublattices of $`L`$ are stable under $`๐ฑ`$. Therefore, we may assume that $`L`$ is indecomposable as a $`y`$-lattice. This leaves the following possibilities for $`L`$: $`_\pm _x^๐ฑ`$ or $`_\lambda `$ for some $`\lambda \mathrm{Hom}(๐ฑ,)`$. In each case, $`2H^2(๐ฑ,L)=0`$ is easy to verify. โ
Lemma 6.2, in combination with the first condition in (6.3), implies that $`K_0`$ is faithful. Consequently, $`\overline{K_0}=K_0/K_0^๐ฑ`$ is faithful as well, and so $`\mathrm{rank}\overline{K_0}2`$. In addition, we know by (6.4) that $`\widehat{H}^0(๐ฑ,K_0)=K_0^๐ฑ/(_๐ฑv)K_0`$ is not cyclic. Hence, neither is $`K_0^๐ฑ`$, which forces $`\mathrm{rank}K_0^๐ฑ2`$ and thus $`\mathrm{rank}K_04`$. Suppose, by way of contradiction, that equality holds here, i.e., $`\overline{K_0}=K_0/K_0^๐ฑ`$ and $`K_0^๐ฑ`$ both have rank 2. By the well-known classification of finite subgroups of $`\mathrm{GL}_2()`$, the action of $`๐ฑ`$ on $`\overline{K_0}`$ is given by either
the matrices $`\left(\begin{array}{cc}1& \\ & 1\end{array}\right)`$ and $`\left(\begin{array}{cc}1& \\ & 1\end{array}\right)`$; so $`\overline{K_0}_{+,}_{,+}`$, or
the matrices $`\left(\begin{array}{cc}& 1\\ 1& \end{array}\right)`$ and $`\left(\begin{array}{cc}& 1\\ 1& \end{array}\right)`$; so $`\overline{K_0}_{}_{}^๐ฑ`$ for some cyclic $`๐ฑ`$.
In case non-diag, $`H^2(๐ฑ,\overline{K_0})H^2(,_{})\widehat{H}^0(,_{})=0`$, and hence $`H^2(๐ฑ,K_0^๐ฑ)`$ maps onto $`H^2(๐ฑ,K_0)`$. But $`H^2(๐ฑ,K_0^๐ฑ)H^2(๐ฑ,)^2\mathrm{Hom}(๐ฑ,/)^2(/2)^4`$. Thus $`H^2(๐ฑ,K_0^๐ฑ)`$ is annihilated by $`2`$, and hence so is $`H^2(๐ฑ,K_0)`$, contradicting (6.3). Therefore, diag must hold:
$$\overline{K_0}=K_0/K_0^๐ฑ_{+,}_{,+}.$$
The action of $`๐ฑ`$ on $`K_0`$ is given by matrices
$$c=\left(\begin{array}{cc}\mathrm{๐}_{2\times 2}& \begin{array}{cc}\mathrm{๐}& ๐ธ\end{array}\\ & \\ & \begin{array}{cc}1& \\ & 1\end{array}\end{array}\right)\text{and}d=\left(\begin{array}{cc}\mathrm{๐}_{2\times 2}& \begin{array}{cc}๐น& \mathrm{๐}\end{array}\\ & \\ & \begin{array}{cc}1& \\ & 1\end{array}\end{array}\right)$$
with $`๐ธ,๐นM_{2\times 1}()`$ and $`\mathrm{๐}=\left(\begin{array}{c}0\\ 0\end{array}\right)`$. By Lemma 6.2, $`๐ธ\mathrm{๐}`$ and $`๐น\mathrm{๐}`$. Conjugating by a suitable matrix of the form $`\left(\begin{array}{cc}\mathrm{๐}_{2\times 2}& \begin{array}{cc}\mathrm{๐}& ๐\end{array}\\ & \mathrm{๐}_{2\times 2}\end{array}\right)`$ we can ensure that the entries of $`๐ธ`$ are $`0`$ or $`1`$, and similarly for $`๐น`$. If $`๐ธ=\left(\begin{array}{c}1\\ 1\end{array}\right)`$ or $`\left(\begin{array}{c}1\\ 0\end{array}\right)`$ then conjugating, respectively, by $`\left(\begin{array}{cc}\begin{array}{cc}1& 1\\ & 1\end{array}& \\ & \mathrm{๐}_{2\times 2}\end{array}\right)`$ or $`\left(\begin{array}{cc}\begin{array}{cc}& 1\\ 1& \end{array}& \\ & \mathrm{๐}_{2\times 2}\end{array}\right)`$, we can replace $`๐ธ`$ by $`๐ธ=\left(\begin{array}{c}0\\ 1\end{array}\right)`$. Thus we may assume that $`๐ธ=\left(\begin{array}{c}0\\ 1\end{array}\right)`$. If $`๐น=\left(\begin{array}{c}1\\ 1\end{array}\right)`$, then conjugating by $`\left(\begin{array}{cc}\begin{array}{cc}1& \\ 1& 1\end{array}& \\ & \mathrm{๐}_{2\times 2}\end{array}\right)`$, we replace $`๐น`$ by $`\left(\begin{array}{c}1\\ 0\end{array}\right)`$ without changing $`c`$. This leaves us with two cases to consider:
$`๐น=\left(\begin{array}{c}0\\ 1\end{array}\right)`$: Then $`c=\left(\begin{array}{cc}1& \\ & c^{}\end{array}\right)`$ and $`d=\left(\begin{array}{cc}1& \\ & d^{}\end{array}\right)`$ with $`c^{}=\left(\begin{array}{ccc}1& 0& 1\\ 0& 1& 0\\ 0& 0& 1\end{array}\right)`$ and $`d^{}=\left(\begin{array}{ccc}1& 1& 0\\ 0& 1& 0\\ 0& 0& 1\end{array}\right)`$. Therefore, $`K_0(A_3_{}_\lambda )`$, where $`\lambda :๐ฑ`$ is the map sending the elements of $`๐ฑ`$ acting via $`c`$ and $`d`$ both to $`1`$. Tensoring the augmentation sequence $`0A_3=\omega ๐ฑ[๐ฑ]0`$ with $`_\lambda `$ we obtain an exact sequence $`0A_3_\lambda [๐ฑ]_\lambda =[๐ฑ]_\lambda 0`$. This sequence in turn implies that $`H^2(๐ฑ,A_3_\lambda )H^1(๐ฑ,_\lambda )`$, and the inflation-restriction sequence easily gives $`H^1(๐ฑ,_\lambda )=/2`$. Thus, $`H^2(๐ฑ,K_0)=H^2(๐ฑ,)H^2(๐ฑ,A_3_\lambda )(/2)^3`$, contradicting (6.3).
$`๐น=\left(\begin{array}{c}1\\ 0\end{array}\right)`$: In this case, $`cd=\left(\begin{array}{cc}\mathrm{๐}_{2\times 2}& \mathrm{๐}_{2\times 2}\\ & \mathrm{๐}_{2\times 2}\end{array}\right)`$. Letting $``$ denote the cyclic subgroup of $`๐ฑ`$ acting via $`cd`$, we have $`K_0|_{}[]^2`$. Thus, $`H^2(,K_0)=0`$, contradicting (6.3).
This completes the proof of the proposition. โ
###### Acknowledgment.
The work on this article was started while the authors were attending the Noncommutative Algebra program at MSRI in the Fall of 1999. The authors would like to thank the organizers of this program and MSRI staff for their hospitality and support. |
warning/0001/hep-th0001116.html | ar5iv | text | # References
Gauge theory models in $`(2+1)`$-space-time are useful in developing ideas for four-dimensional models, such as their high temprature behavior, boundaries of the spatial region occupied by gauge field etc. In this paper, we discuss topologically massive $`SU(2)`$ gauge theory coupled with fermions and compute one-loop correction to quark energy.
The total Lagrangian of three-dimensional topologically massive $`SU(2)`$-gluodynamics of potentials $`๐_\mu \tau ^a๐_\mu ^a/2`$, where $`\tau ^a`$ are Pauli matrices in color space, and quarks is described as follows
$$=\frac{1}{4}_{\mu \nu }^a^{a\mu \nu }\frac{\theta }{4}\epsilon ^{\mu \nu \alpha }\left(_{\mu \nu }^a๐_\alpha ^a\frac{g}{3}\epsilon ^{abc}๐_\mu ^a๐_\nu ^b๐_\alpha ^c\right)\frac{1}{2\xi }\left(_\mu ^{ab}a^{b\mu }\right)^2+\overline{\psi }(D_\mu \gamma ^\mu m)\psi ,$$
(1)
where $`\mu ,\nu =0,1,2`$; $`a=1,2,3`$, the color field tensor is given by $`_{\mu \nu }^a=_\mu ๐_\nu ^a_\nu ๐_\mu ^a+g\epsilon ^{abc}๐_\mu ^b๐_\nu ^c`$. The total gauge field $`๐^{a\mu }`$ is represented as the sum of the classical background field and quantum fluctuations $`a^{a\mu }`$, i.e.
$$๐^{a\mu }=A^{a\mu }+a^{a\mu },$$
so that $`_\mu ^{ab}=\delta ^{ab}_\mu +g\epsilon ^{abc}A_\mu ^c`$ is the background covariant derivative and $`D_\mu ^{ab}=\delta ^{ab}_\mu +g\epsilon ^{abc}๐_\mu ^c`$ is the total covariant derivative. The third term in (1) is the gauge-fixing term. The coefficient $`\theta `$ in front of the second term in (1) (Chern-Simons term) is the Chern-Simons (CS) mass of the gauge field. We use the following representation for the $`\gamma `$-matrices in $`2+1`$ dimensions: $`\gamma ^0=\sigma ^3`$, $`\gamma ^{1,2}=i\sigma ^{1,2}`$, with $`\sigma ^i`$ as Pauli matrices. The $`\gamma `$-matrices obey the following relation: $`\gamma ^\mu \gamma ^\nu =g^{\mu \nu }i\epsilon ^{\mu \nu \alpha }\gamma _\alpha `$.
It was shown in that the classical constant nonabelian potentials
$$A^{a\mu }=\frac{\theta }{2g}\delta ^{a\mu }\chi _{\lambda \omega }^{(a)},$$
(2)
with normalized constant vector $`\chi _{\lambda \omega }^{(a)}=(\lambda i,\lambda \omega i,\omega )`$ satisfy the field equations with the Chern-Simons term without external currents. In (2) $`\lambda =\pm 1`$ and $`\omega =\pm 1`$ take its values independently. The Kronecker delta $`\delta ^{a\mu }`$ in (2) implies that directions 1, 2, 3 in the color space correspond to directions 1, 2, 0 in the Minkowski $`2+1`$ space-time, respectively. In what follows, like in , where corrections to the gluon energy were considered, these solutions are chosen as the background.
Considering the one-loop corrections, it is sufficient to retain only the terms in the Lagrangian (1) quadratic in the quantum fields. They determine the quark energy spectrum in the gauge field (2)
$$\epsilon _1^2=๐ฉ^2+m_{\mathrm{eff}\mathrm{\hspace{0.17em}1}}^2,\epsilon _2^2=๐ฉ^2+m_{\mathrm{eff}\mathrm{\hspace{0.17em}2}}^2,$$
where
$$m_{\mathrm{eff}\mathrm{\hspace{0.17em}1}}^2=(m\stackrel{~}{\theta })^2,m_{\mathrm{eff}\mathrm{\hspace{0.17em}2}}^2=(m\stackrel{~}{\theta })(m+3\stackrel{~}{\theta })$$
(3)
and $`\stackrel{~}{\theta }=\theta /4`$. These branches of the energy spectrum correspond to the plane-wave solutions $`\psi _s(x)=\mathrm{exp}(i\epsilon _st+i\stackrel{}{px})\mathrm{\Psi }_s,`$ with $`s=1,2,`$ and $`\mathrm{\Psi }_s`$ as constant spinors, of the Dirac equation
$$[\gamma ^\mu (p_\mu +gA_\mu )m]\psi =0,$$
and are related to two opposite projections of the particle color spin operator
$$๐=J^a\tau ^a/2=\stackrel{~}{\theta }^1gA^\mu p_\mu \omega \gamma ^0\tau ^3(p_\mu \gamma ^\mu +\stackrel{~}{\theta }m),$$
(4)
defined as eigenvalues of the equation
$$๐\mathrm{\Psi }_s=(1)^s(\stackrel{~}{\theta }m)\mathrm{\Psi }_s.$$
We note, that the r.h.s. of the last equation vanishes, when $`m=\stackrel{~}{\theta }`$. In this special case, the quark effective masses in this gauge field (3) are also equal to zero, and the two branches of the energy spectrum coincide. For values of the mass lying in the interval $`\stackrel{~}{\theta }2|\stackrel{~}{\theta }|<m<\stackrel{~}{\theta }+2|\stackrel{~}{\theta }|`$, the energy squared, $`\epsilon _2^2`$, becomes negative for certain values of the quark momentum $`๐ฉ`$, and tachyonic modes arise in the quark spectrum. Moreover, it appears that $`\mathrm{\Psi }_1=\mathrm{\Psi }_2`$ under this condition.
We now consider the one-loop radiative shift of the quark energy. According to , the quark energy radiative correction $`\mathrm{\Delta }\epsilon `$ is obtained by averaging the mass operator over the quark state in the external field $`A^\mu `$, specified by equation (2). The one-loop radiative correction to the quark energy is given by formula
$$\mathrm{\Delta }\epsilon =\frac{i}{T}d^3xd^3x^{}\overline{\psi }_k(x)M_{kl}(x,x^{})\psi _l(x^{}),$$
where
$$iM_{kl}(x,y)=ig^2\gamma ^\mu (\tau ^a/2)_{kn}S_{nm}(x,y)\gamma ^\nu (\tau ^b/2)_{ml}D_{\mu \nu }^{ab}(x,y)$$
is the one-loop mass operator. The expression for $`\mathrm{\Delta }\epsilon `$ has to be applied with due regard to the fact that the quark Hamiltonian in the background (2) becomes non-Hermitian. In the momentum representation we have
$$\begin{array}{c}\mathrm{\Delta }\epsilon =\overline{\mathrm{\Psi }}_k(p)iM_{kl}(p)\mathrm{\Psi }_l(p),\\ iM_{kl}(p)=ig^2\gamma ^\mu (\tau ^a/2)_{kn}d^3kS_{nm}(pk)\gamma ^\nu (\tau ^b/2)_{ml}D_{\mu \nu }^{ab}(k).\end{array}$$
In order to calculate the quark mass operator in the external field the quark and gluon propagators have to be found. The quark Greenโs function satisfies the equation
$$[\gamma ^\mu (i_\mu +gA_\mu )m]S(x,y)=\delta (xy).$$
In the background gauge field $`A^\mu `$ considered above, the quark Greenโs function in the momentum representation has the form
$$\begin{array}{c}S(p)=[(p^2m_{\mathrm{eff}\mathrm{\hspace{0.17em}1}}^2)(p^2m_{\mathrm{eff}\mathrm{\hspace{0.17em}2}}^2)]^1\{(p^2(m\stackrel{~}{\theta })^2)[\gamma ^\mu (p_\mu +gA_\mu )+m]\\ 2(\gamma ^\mu p_\mu +m\stackrel{~}{\theta })[gA^\nu p_\nu +\stackrel{~}{\theta }(m\stackrel{~}{\theta })]\},\end{array}$$
with the quark effective masses $`m_{\mathrm{eff}\mathrm{\hspace{0.17em}1}}`$ and $`m_{\mathrm{eff}\mathrm{\hspace{0.17em}2}}`$ defined above.
The gluon propagator in the external field in the gauge $`\xi =1`$ takes the form
$$\begin{array}{c}D_{\mu \nu }^{ab}=\delta ^{ab}g_{\mu \nu }\left[\frac{1}{2}+\frac{}{2\alpha }\right]\frac{4g^2}{\theta ^2}A_\mu ^aA_\nu ^b\left[\frac{1}{3}\frac{}{3\beta }\right]+\frac{4g^2}{\theta ^2}A_\nu ^aA_\mu ^b\left[\frac{1}{2}\frac{}{2\alpha }\right]+\\ \frac{16g^2}{\theta ^2\alpha \beta }F_{\mu \alpha }^aF_{\nu \beta }^bp^\alpha p^\beta +\frac{8ig^2}{\theta ^2\beta }(F_{\mu \alpha }^aA_\nu ^bF_{\nu \alpha }^bA_\mu ^a)p^\alpha +\frac{16ig^2}{\theta ^3\alpha }\epsilon ^{\alpha \beta \gamma }F_{\mu \alpha }^aF_{\nu \beta }^bp_\gamma ,\end{array}$$
where we used the notations $`=p^2+\frac{1}{2}\theta ^2`$, $`\alpha =^24\theta ^2p^2`$ and $`\beta =^26\theta ^2p^2`$.
In the case of an arbitrary quark mass value $`m\stackrel{~}{\theta }`$, the quark has two different color states. It is interesting to consider the special case, when $`m=\stackrel{~}{\theta }`$, and, as it has been mentioned above, the quark effective masses vanish. We remind that under this condition, $`m_{\mathrm{eff}\mathrm{\hspace{0.17em}1}}=m_{\mathrm{eff}\mathrm{\hspace{0.17em}2}}=0`$, only one quark state survives. The corresponding plane wave solution of the Dirac equation $`\psi (x)=(2\pi )^1\mathrm{exp}(i\epsilon t+i\stackrel{}{px})\mathrm{\Psi }(p)`$ is determined by the constant spinor
$$\mathrm{\Psi }(p)=\frac{1}{8}\left(\begin{array}{c}[\lambda (\kappa 1)+(\kappa +1)][(\omega +1)e^{i\varphi }+(\omega 1)]\\ [\lambda (\kappa 1)(\kappa +1)][(\omega 1)e^{i\varphi }+(\omega +1)]\\ [\lambda (\kappa +1)(\kappa 1)][(\omega 1)e^{i\varphi }+(\omega +1)]\\ [\lambda (\kappa +1)+(\kappa 1)][(\omega +1)e^{i\varphi }+(\omega 1)]\end{array}\right).$$
It should be emphasized that in this formula the quark momentum $`๐ฉ`$ is assumed to be nonzero. Here $`\kappa =\pm 1`$ is the sign of the energy ($`\epsilon p_0=\kappa |๐ฉ|`$) and the phase $`\varphi `$ is defined by the relation $`p^2\pm ip^1=|๐ฉ|e^{\pm i\varphi }`$. We used representation for the Dirac operator and the Hamiltonian, where $`\gamma `$-matrices are inserted into isospin color Pauli matrices $`\tau ^a`$. It should be also emphasized that, since the fermion Hamiltonian is non-Hermitian, it provides only one solution of the Dirac equation for one value of the energy sign.
After integration over intermediate momentum $`k`$ the mass operator is expressed in the form
$$iM(p)=ig^2\frac{\pi ^2\sqrt{2}}{|\theta |}\left[b_1\theta +\frac{5}{9}\left(\frac{2}{\theta }gA^\mu +\frac{1}{2}\gamma ^\mu \right)p_\mu gA^\mu \gamma ^\nu (b_2g_{\mu \nu }b_3\theta ^2p_\mu p_\nu )\right],$$
where
$$\begin{array}{c}b_1=\frac{5}{6}\frac{i}{12}(3\sqrt{2}+4\sqrt{3}),\\ b_2=\frac{10}{9}+\frac{i}{9}(8\sqrt{3}+9\sqrt{2}),\\ b_3=\frac{4}{45}\frac{4\sqrt{3}i}{15}.\end{array}$$
Now we can find the one-loop radiative correction to the quark energy in the external field, which turns out to be vanishing in this particular state:
$$\mathrm{\Delta }\epsilon =\frac{5\pi ^2\sqrt{2}}{18}ig^2|\theta |^1\left(p_0\kappa \sqrt{(p^1)^2+(p^2)^2}\right)0.$$
It should be emphasized that this result is valid only for finite values of the quark momenta $`๐ฉ=(p^1,p^2)`$. For the case of vanishing $`๐ฉ`$, a special consideration is needed.
The fact that a fermion remains effectively massless even in the one-loop approximation signals the presence of a certain symmetry of the problem in question. In fact, the results obtained demonstrate the supersymmetry property of the state considered.
As is well known (see, e.g., ), the minimal representation of SUSYQ (SUSY Quantum Mechanics) is provided with the supercharges $`Q_1`$, $`Q_2`$ and the Hamiltonian $`H_S`$:
$$H_S=Q_1^2=Q_2^2,[H_S,Q_i]=0,\{Q_i,Q_j\}=2\delta _{ij}H_S,i,j=1,2.$$
The quark Hamiltonian in the external field (2) can be written in the form
$$H=ip^1\gamma ^2ip^2\gamma ^1+\stackrel{~}{\theta }(\gamma ^0\omega \tau ^3)\lambda \stackrel{~}{\theta }(\tau ^1\gamma ^2\omega \tau ^2\gamma ^1).$$
The SUSYQ Hamiltonian $`H_S=H^2`$ is found to be
$$H_S=๐ฉ^2+2\stackrel{~}{\theta }[i\lambda (p^1\tau ^1+\omega p^2\tau ^2)i\omega \tau ^3(p^1\gamma ^2p^2\gamma ^1)].$$
It is interesting to note that $`H_S`$ can be written in the form
$$H_S=๐ฉ^22\stackrel{~}{\theta }๐,$$
where $`๐`$ is the operator defined in (4). Under the condition $`m=\stackrel{~}{\theta }`$ we have $`๐^2=0`$. The supercharges, corresponding to the above Hamiltonian $`H_S`$, can be found:
$$\begin{array}{c}Q_1=\left[\sqrt{2}\gamma ^0\frac{i\lambda \omega \stackrel{~}{\theta }}{|๐ฉ|}\right](p^1\gamma ^1+p^2\gamma ^2)\frac{\omega \stackrel{~}{\theta }}{|๐ฉ|}\gamma ^0\tau ^3(\omega p^1\tau ^1+p^2\tau ^2)+\\ \frac{1}{|๐ฉ|}(p^1\gamma ^1+p^2\gamma ^2)(\omega p^1\tau ^1+p^2\tau ^2),\\ Q_2=\left[\frac{i\omega \sqrt{2}}{|๐ฉ|}+\frac{\lambda \stackrel{~}{\theta }}{๐ฉ^2}\gamma ^0\right](p^1\gamma ^1+p^2\gamma ^2)(\omega p^1\tau ^1+p^2\tau ^2)+i\stackrel{~}{\theta }\tau ^3+i\omega \gamma ^0(p^1\gamma ^1+p^2\gamma ^2).\end{array}$$
It is interesting to consider the quark ground state, when its momentum is equal to zero. Solution of the Dirac equation is easy to find in this case. It has the form
$$\mathrm{\Psi }(p=0)=\frac{c_1}{\sqrt{2}}\left(\begin{array}{c}\omega +1\\ \omega 1\\ 0\\ 0\end{array}\right)+\frac{c_2}{\sqrt{2}}\left(\begin{array}{c}0\\ 0\\ \omega 1\\ \omega +1\end{array}\right)+\frac{c_3}{\sqrt{8}}\left(\begin{array}{c}\omega 1\\ \omega +1\\ \lambda (\omega +1)\\ \lambda (\omega 1)\end{array}\right),$$
(5)
where constants $`c_i`$ are arbitrary up to a normalization condition.
In order to demonstrate the supersymmetry property of the Dirac equation $`D\psi =0`$, where
$$D=\gamma ^\mu (p_\mu +gA_\mu )m$$
is the Dirac operator, we consider its zero-mode solutions with $`p^0=0`$ under the condition $`m=\theta /4`$ and with vanishing fermion mass in virtue of the condition $`gA^0\gamma ^0\psi =m\psi `$, which means $`\sigma ^3\tau ^3\psi =\omega \psi `$. In this case the Dirac equation takes the form
$$(\gamma ^1^1+\gamma ^2^2)\psi _0=0.$$
Then we introduce combinations
$$b^\pm =(2i)^1(\gamma ^1\pm i\gamma ^2),$$
which play the role of fermionic anticommuting operators (creation and annihilation operators). The Dirac operator has the form
$$D=Q_++Q_{},$$
where $`Q_+=2b^+_u`$, $`Q_{}=2b^{}\overline{}_u`$ while $`_u=_uigA_u`$, $`\overline{}_u=\overline{}_uig\overline{A}_u`$, and here $`_u=\frac{1}{2}(_1i_2)`$, $`\overline{}_u=\frac{1}{2}(_1+i_2)`$, etc. The standard SUSY hamiltonian has the form
$$H_S=(D^2)=\{Q_+,Q_{}\},Q_\pm ^2=0,[H_S,Q_\pm ]=0.$$
We set $`\psi _0=\psi _\tau \psi _s`$, where $`\psi _\tau `$ describes the color (boson) state and $`\psi _s`$ specifies the spin (fermion) state. For the ground (zero-mode) state we have: $`H_S\psi =0`$. This means
$$Q_+\psi _0=0,Q_{}\psi _0=0.$$
Consider $`Q_+\psi _0=0`$, or $`b^+D_u\psi _0=0`$. Let $`\psi _s=|0_s`$, then $`b^+\psi _s=|1_s`$ and then $`D_u\psi _\tau =0`$ ($`|0`$ and $`|1`$ are fermionic states). For $`_{1,2}\psi _0=0`$ we have
$$(\tau ^1i\omega \tau ^2)\psi _\tau =0.$$
Then if $`\omega =+1`$, we recall that $`\tau ^3\sigma ^3=\omega =+1`$, and for $`\sigma ^3\psi _s=\psi _s`$, we have $`\tau ^3\psi _\tau =\psi _\tau `$ and $`(\tau _1i\tau _2)\psi _\tau =0`$. Hence $`\psi _\tau =|0_\tau `$. Excited solutions are constructed in a standard way. Now it is clear that the first two terms in (5) describe the states with the SUSY property.
One of the authors (N.A.P.) gratefully acknowledges the hospitality of Prof. Mรผller-Preuรker, Prof. Ebert and their colleagues at the particle theory group of the Institut fรผr Theoretische Physik, Humboldt Universitรคt zu Berlin extended to him during his stay there. |
warning/0001/quant-ph0001075.html | ar5iv | text | # Qudit EntanglementWork at UNM was supported in part by the U.S. Office of Naval Research (Grant No. N00014-93-1-0116), WJM acknowledges the support of the Australian Research Council, and KN thanks the Australian International Education Foundation (AIEF) for financial support.
## 1 Introduction
One of the distinguishing features of quantum mechanics, not found in classical physics, is the possibility of entanglement between subsystems. It lies at the core of many applications in the emerging field of quantum information science , such as quantum teleportation and quantum error correction . Entanglement is a distinctly quantum-mechanical correlation between subsystems, which cannot be created by actions on each subsystem separately; moreover, correlations between subsystem measurements on an entangled composite system cannot be explained in terms of correlations between local classical properties inherent in the subsystems. Thus one often says that an entangled composite system is nonseparable. Formally, the state of a composite system, pure or mixed, is separable if the state has an ensemble decomposition in terms of product states. A separable state has no quantum entanglement, and a nonseparable state is entangled. Though the nonclassical nature of quantum entanglement has been recognized for many years , only recently has considerable attention been focused on trying to understand and characterize its properties precisely.
This paper focuses on the question of whether various joint quantum states of $`D`$-dimensional quantum systems are entangled. For convenience, we call a $`D`$-dimensional quantum system a โqudit,โ by analogy with the name โqubitโ for $`D=2`$ and โqutritโ for $`D=3`$. We now have a general method for quantifying the degree of entanglement of a pair of qubits , and we have a criterion, the partial transposition condition of Peres , which determines whether a general state of two qubits is entangled and whether a general state of a qubit and a qutrit is entangled . The partial-transposition condition fails, however, to provide a criterion for entanglement in other cases, where the constituents have higher Hilbert-space dimensions or where there are more than two constituents. Indeed, at present there is no general criterion for determining whether the joint state of $`N`$ qudits is entangled, nor is there any general way to quantify the degree of entanglement if such a state is known to be entangled.
In Sect. 2 we review an operator representation of qudit states, which is applied in Sect. 3, where we consider states of two qudits that are a mixture of the maximally mixed state and a maximally entangled state. We show that such states are separable if and only if the probability for the maximally entangled state in the mixture does not exceed $`1/(1+D)`$. This result was obtained by Horodecki and Horodecki , and a more general result, of which this is a special case, was obtained by Vidal and Tarrach . In Sect. 4 we consider the separability of mixed states of $`N`$ qudits near the maximally mixed state. We find both lower and upper bounds on the size of the neighborhood of separable states around the maximally mixed state. Our results generalize and extend the results obtained by Braunstein et al. for qubits and by Caves and Milburn for qutrits . Before tackling the upper and lower bounds, we present, in Sect. 4.1, various mathematical results which are used to obtain the lower bound, but which might prove useful in other contexts as well.
## 2 Operator representation of qudit states
In this section we review an operator representation of qudit states, analogous to the Pauli, or Bloch-sphere, representation for qubits. We begin with the set of Hermitian generators of SU($`D`$); the generators, denoted by $`\lambda _j`$, are labeled by a Roman index taken from the middle of the alphabet, which takes on values $`j=1,\mathrm{},D^21`$. We represent the generators in an orthonormal basis $`|a`$, labeled by a Roman letter taken from the beginning of the alphabet, which takes on values $`a=1,\mathrm{},D`$. With these conventions the generators are given by
$`j=1,\mathrm{},D1:`$
$`\lambda _j=\mathrm{\Gamma }_a{\displaystyle \frac{1}{\sqrt{a(a1)}}}\left({\displaystyle \underset{b=1}{\overset{a1}{}}}|bb|(a1)|aa|\right),2aD,`$ (1)
$`j=D,\mathrm{},(D+2)(D1)/2:`$
$`\lambda _j=\mathrm{\Gamma }_{ab}^{(+)}{\displaystyle \frac{1}{\sqrt{2}}}\left(|ab|+|ba|\right),1a<bD,`$ (2)
$`j=D(D+2)/2,\mathrm{},D^21:`$
$`\lambda _j=\mathrm{\Gamma }_{ab}^{()}{\displaystyle \frac{i}{\sqrt{2}}}\left(|ab||ba|\right),1a<bD.`$ (3)
In Eqs. (2) and (3), the Roman index $`j`$ stands for the pair of Roman indices, $`ab`$, whereas in Eq. (1), it stands for a single Roman index $`a`$. The generators are traceless and satisfy
$$\lambda _j\lambda _k=\frac{1}{D}\delta _{jk}+d_{jkl}\lambda _l+if_{jkl}\lambda _l.$$
(4)
Here and wherever it is convenient throughout this paper, we use the summation convention to indicate a sum on repeated indices. The coefficients $`f_{jkl}`$, the structure constants of the Lie group SU($`D`$), are given by the commutators of the generators and are completely antisymmetric in the three indices. The coefficients $`d_{jkl}`$ are given by the anti-commutators of the generators and are completely symmetric.
By supplementing the $`D^21`$ generators with the operator
$$\lambda _0\frac{1}{\sqrt{D}}I,$$
(5)
where $`I`$ is the unit operator, we obtain a Hermitian operator basis for the space of linear operators in the qudit Hilbert space. This is an orthonormal basis, satisfying
$$\mathrm{tr}(\lambda _\alpha \lambda _\beta )=\delta _{\alpha \beta }.$$
(6)
Here the Greek indices take on the values $`0,\mathrm{},D^21`$; throughout this paper, Greek indices take on $`D^2`$ or more values. Using this orthonormality relation, we can invert Eqs. (1)โ(3) to give
$`|aa|`$ $`=`$ $`{\displaystyle \frac{I}{D}}+{\displaystyle \frac{1}{\sqrt{a(a1)}}}\left((a1)\mathrm{\Gamma }_a+{\displaystyle \underset{b=a+1}{\overset{D}{}}}\mathrm{\Gamma }_b\right),`$ (7)
$`|ab|`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}(\mathrm{\Gamma }_{ab}^{(+)}+i\mathrm{\Gamma }_{ab}^{()}),1a<bD,`$ (8)
$`|ba|`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}(\mathrm{\Gamma }_{ab}^{(+)}i\mathrm{\Gamma }_{ab}^{()}),1a<bD.`$ (9)
Any qudit density operator can be expanded uniquely as
$$\rho =\frac{1}{D}c_\alpha \lambda _\alpha ,$$
(10)
where the (real) expansion coefficients are given by
$$c_\alpha =D\mathrm{tr}(\rho \lambda _\alpha ).$$
(11)
Normalization implies that $`c_0=\sqrt{D}`$, so the density operator takes the form
$$\rho =\frac{1}{D}\left(I+c_j\lambda _j\right)=\frac{1}{D}(I+\stackrel{}{c}\stackrel{}{\lambda }).$$
(12)
Here $`\stackrel{}{c}=c_j\stackrel{}{e}_j`$ can be regarded as a vector in a $`(D^21)`$-dimensional real vector space, spanned by the orthonormal basis $`\stackrel{}{e}_j`$, and $`\stackrel{}{\lambda }=\lambda _j\stackrel{}{e}_j`$ is an operator-valued vector. If $`\rho =|\psi \psi |`$ is a pure qudit state, then $`\mathrm{tr}(\rho ^2)=1`$, from which it follows that
$$|\stackrel{}{c}|^2=\stackrel{}{c}\stackrel{}{c}=D(D1).$$
(13)
We could represent a pure state by a unit vector $`\stackrel{}{n}=\stackrel{}{c}/\sqrt{D(D1)}`$ on the unit sphere in $`D^21`$ dimensions, but in contrast to the situation with the Bloch sphere ($`D=2`$), most vectors on this unit sphere do not represent a pure state or, indeed, any state at all.
## 3 Mixtures of maximally mixed and <br>maximally entangled states
In this section we deal with two qudits, labeled $`A`$ and $`B`$. We consider a class of two-qudit states, specifically mixtures of the maximally mixed state, $`M_{D^2}=II/D^2`$, with a maximally entangled state, which we can choose to be
$$|\mathrm{\Psi }=\frac{1}{\sqrt{D}}\underset{a=1}{\overset{D}{}}|a|a.$$
(14)
Such mixtures have the form
$$\rho _ฯต=(1ฯต)M_{D^2}+ฯต|\mathrm{\Psi }\mathrm{\Psi }|,$$
(15)
where $`0ฯต1`$.
In analogy to Eq. (10), any state $`\rho `$ of two qudits can be expanded uniquely as
$$\rho =\frac{1}{D^2}c_{\alpha \beta }\lambda _\alpha \lambda _\beta ,$$
(16)
where the expansion coefficients are given by
$$c_{\alpha \beta }=D^2\mathrm{tr}(\rho \lambda _\alpha \lambda _\beta ),$$
(17)
with $`c_{00}=D`$ determined by normalization. Using Eq. (17) or Eqs. (7)โ(9), we can find the operator expansion for the maximally entangled state (14):
$$|\mathrm{\Psi }\mathrm{\Psi }|=\frac{1}{D^2}\left(II+D\underset{a}{}\mathrm{\Gamma }_a\mathrm{\Gamma }_a+D\underset{a<b}{}\left(\mathrm{\Gamma }_{ab}^{(+)}\mathrm{\Gamma }_{ab}^{(+)}\mathrm{\Gamma }_{ab}^{()}\mathrm{\Gamma }_{ab}^{()}\right)\right),$$
(18)
from which we can read off the expansion coefficients for the state $`\rho _ฯต`$ of Eq. (15):
$`c_{0j}`$ $`=`$ $`c_{j0}=0,`$ (19)
$`c_{jk}`$ $`=`$ $`\{\begin{array}{cc}0,\hfill & jk\text{,}\hfill \\ Dฯต,\hfill & j=k=1,\mathrm{},(D+2)(D1)/2\text{.}\hfill \\ Dฯต,\hfill & j=k=D(D+2)/2,\mathrm{},D^21\text{.}\hfill \end{array}`$ (20)
A state of the two qudits is separable if it can be written as an ensemble of product states. In this section we show that the mixed state state (15) is separable if and only if
$$ฯต\frac{1}{1+D}.$$
(21)
Our method is to prove the necessity of the condition (21) by considering the restrictions that separability places on the correlation coefficients (20) and then to construct an explicit product ensemble when $`ฯต1/(1+D)`$. Vidal and Tarrach found the separability boundary for a mixture of $`M_{D^2}`$ with any pure state by using the partial transpose condition to show that any state with $`ฯต`$ outside the boundary is nonseparable and by constructing an explicit product ensemble for states with $`ฯต`$ within the separability boundary. Horodecki and Horodecki found the separability boundary for the state (15) using other techniques. The reason for presenting in this section a more limited result than that of Vidal and Tarrach is, first, that our proof of necessity has a nice physical interpretation in terms of the correlation coefficients (20) and, second, that the product ensemble we use is different from the one used by Vidal and Tarrach.
The product pure states for two qudits, $`|\psi _A\psi _A||\psi _B\psi _B|`$, constitute an overcomplete operator basis. Thus we can expand any two-qudit density operator in terms of them,
$$\rho =๐๐ฑ_A๐๐ฑ_Bw(\psi _A,\psi _B)|\psi _A\psi _A||\psi _B\psi _B|.$$
(22)
Here the integral for each system runs over all of projective Hilbert space, i.e., the space of Hilbert-space rays, and the volume elements $`d๐ฑ_A`$ and $`d๐ฑ_B`$ are the unitarily invariant integration measures on projective Hilbert space. Because of overcompleteness of the pure-state projectors, the expansion function $`w(\psi _A,\psi _B)`$ is not unique. Notice that the expansion coefficients $`c_{\alpha \beta }`$ of Eq. (17) can be written as integrals over the expansion function,
$$c_{\alpha \beta }=๐๐ฑ_A๐๐ฑ_Bw(\psi _A,\psi _B)(c_A)_\alpha (c_B)_\beta ,$$
(23)
where $`(c_A)_\alpha =D\psi _A|\lambda _\alpha |\psi _A`$ and $`(c_B)_\alpha =D\psi _B|\lambda _\alpha |\psi _B`$ are the expansion coefficients for the pure states $`|\psi _A`$ and $`|\psi _B`$, satisfying $`\stackrel{}{c}_A\stackrel{}{c}_A=D(D1)=\stackrel{}{c}_B\stackrel{}{c}_B`$.
A two-qudit state is separable if and only if there exists an expansion function $`w(\psi _A,\psi _B)`$ that is everywhere nonnegative. In this case $`w(\psi _A,\psi _B)`$ can be thought of as a normalized classical probability distribution for the pure states $`\psi _A`$ and $`\psi _B`$, and the integral for $`c_{\alpha \beta }`$ in Eq. (23) can be interpreted as a classical expectation value of the product of the random variables $`(c_A)_\alpha `$ and $`(c_B)_\beta `$, i.e.,
$$c_{\alpha \beta }=E\left[(c_A)_\alpha (c_B)_\beta \right].$$
(24)
If the state $`\rho _ฯต`$ is separable, we have from Eq. (20) that for each value of $`j`$,
$$Dฯต=|c_{jj}|=\left|E\left[(c_A)_j(c_B)_j\right]\right|\frac{1}{2}\left(E[(c_A)_j^2]+E[(c_B)_j^2]\right).$$
(25)
Adding over the $`D^21`$ value of $`j`$ gives
$$D(D^21)ฯต\frac{1}{2}\left(E[\stackrel{}{c}_A\stackrel{}{c}_A]+E[\stackrel{}{c}_B\stackrel{}{c}_B]\right)=D(D1).$$
(26)
We conclude that if $`\rho _ฯต`$ is separable, then $`ฯต1/(1+D)`$.
To prove the converse, we construct an explicit product ensemble for the state $`\rho _ฯต`$ with $`ฯต=1/(1+D)`$. We define a vector $`\stackrel{}{z}=(z_1,\mathrm{},z_D)`$ whose components $`z_a`$ take on the values $`\pm 1`$ and $`\pm i`$, so that
$$\underset{z_j}{}z_j=\underset{z_j}{}z_j^2=0,\underset{z_j}{}|z_j|^2=4.$$
(27)
Associated with each vector $`\stackrel{}{z}`$ is a pure state
$$|\mathrm{\Phi }_\stackrel{}{z}=\frac{1}{\sqrt{D}}\underset{a=1}{\overset{D}{}}z_a|a.$$
(28)
There are $`4^D`$ vectors and thus that many states $`\mathrm{\Phi }_\stackrel{}{z}`$, although only $`4^{D1}`$ of these states are distinct in that they differ by more than a global phase. Now we define a product state for the two-qudit system:
$$\rho _\stackrel{}{z}=|\mathrm{\Phi }_\stackrel{}{z}\mathrm{\Phi }_\stackrel{}{z}||\mathrm{\Phi }_\stackrel{}{z}^{}\mathrm{\Phi }_\stackrel{}{z}^{}|.$$
(29)
The ensemble consisting of all $`4^D`$ of these states, each contributing with the same probability, produces the density operator
$$\frac{1}{4^D}\underset{\stackrel{}{z}}{}\rho _\stackrel{}{z}=\frac{1}{4^DD^2}\underset{a,b,c,d}{}\left(\underset{\stackrel{}{z}}{}z_az_b^{}z_c^{}z_d\right)|ab||cd|.$$
(30)
Since
$$\underset{\stackrel{}{z}}{}z_az_b^{}z_c^{}z_d=4^D(\delta _{ab}\delta _{cd}+\delta _{ac}\delta _{bd}\delta _{ab}\delta _{cd}\delta _{ac}),$$
(31)
it follows that
$$\frac{1}{4^D}\underset{\stackrel{}{z}}{}\rho _\stackrel{}{z}=\frac{II}{D^2}+\frac{1}{D}|\mathrm{\Psi }\mathrm{\Psi }|\frac{1}{D^2}\underset{a=1}{\overset{D}{}}|aa||aa|.$$
(32)
Multiplying by $`D/(D+1)`$ and rearranging yields
$$\frac{D}{1+D}\frac{II}{D^2}+\frac{1}{1+D}|\mathrm{\Psi }\mathrm{\Psi }|=\frac{D}{1+D}\frac{1}{4^D}\underset{\stackrel{}{z}}{}\rho _\stackrel{}{z}+\frac{1}{1+D}\frac{1}{D}\underset{a=1}{\overset{D}{}}|aa||aa|.$$
(33)
The left-hand side of Eq. (33) is the state (15) with $`ฯต=1/(1+D)`$, and the right-hand side is an explicit product ensemble for the state. This concludes the proof that $`\rho _ฯต`$ is separable if and only if $`ฯต1/(1+D)`$.
## 4 Separability of states near the <br>maximally mixed state
This section deals with $`N`$-qudit states of the form
$$\rho _ฯต=(1ฯต)M_{D^N}+ฯต\rho _1,$$
(34)
where $`M_{D^N}=I\mathrm{}I/D^N`$ is the maximally mixed state for $`N`$ qudits and $`\rho _1`$ is any $`N`$-qudit density operator. We establish lower and upper bounds on the size of the neighborhood of separable states surrounding the maximally mixed state. In particular, we show, first, that for
$$ฯต\frac{1}{1+D^{2N1}},$$
(35)
all states of the form (34) are separable and, second, that for
$$ฯต>\frac{1}{1+D^{N1}},$$
(36)
there are states of the form (34) that are not separable (i.e., they are entangled). These results generalize and extend the work of Braunstein et al. for qubits and of Caves and Milburn for qutrits .
### 4.1 Mathematical preliminaries
Before turning to the lower and upper bounds, it is useful to develop some mathematical apparatus that will be used in deriving the bounds.
#### 4.1.1 Superoperator formalism
We begin by reviewing a formalism for handling superoperators, introduced by Caves and used by Schack and Caves to generate product ensembles for separable $`N`$-qubit states.
The space of linear operators acting on a $`D`$-dimensional complex vector space is a $`D^2`$-dimensional complex vector space. In this space we introduce operator โketsโ $`|A)=A`$ and โbrasโ $`(A|=A^{}`$, distinguished from vector kets and bras by the use of smooth brackets. The natural operator inner product can be written as $`(A|B)=\mathrm{tr}(A^{}B)`$. An orthonormal basis $`|a`$ induces an orthonormal operator basis,
$$|ca|=\tau _{ca}=\tau _\alpha ,$$
(37)
where the Greek index $`\alpha `$ is an abbreviation for the pair of Roman indices, $`ca`$. Not all orthonormal operator bases are of this outer-product form.
The space of superoperators, i.e., linear maps on operators, is a $`D^4`$-dimensional complex vector space. Any superoperator $`๐ฎ`$ is specified by its โmatrix elementsโ
$$๐ฎ_{ca,db}=c|๐ฎ(|ab|)|d,$$
(38)
for the superoperator can be written in terms of its matrix elements as
$$๐ฎ=\underset{c,a,d,b}{}๐ฎ_{ca,db}|ca||bd|=\underset{c,a,d,b}{}๐ฎ_{ca,db}\tau _{ca}\tau _{db}^{}=\underset{\alpha ,\beta }{}๐ฎ_{\alpha \beta }|\tau _\alpha )(\tau _\beta |.$$
(39)
The tensor product here is an ordinary operator tensor product, but we use the symbol $``$ to distinguish it from a tensor product between objects associated with different systems, which is denoted by $``$. In the final form of Eq. (39), the tensor product is written as an operator outer product, with $`\alpha =ca`$ and $`\beta =db`$.
The ordinary action of $`๐ฎ`$ on an operator $`A`$, used to generate the matrix elements, is obtained by dropping an operator $`A`$ into the center of the representation of $`๐ฎ`$, in place of the tensor-product sign,
$$๐ฎ(A)=\underset{\alpha ,\beta }{}๐ฎ_{\alpha \beta }\tau _\alpha A\tau _\beta ^{}.$$
(40)
There is clearly another way that $`๐ฎ`$ can act on $`A`$, the left-right action,
$$๐ฎ|A)=\underset{\alpha ,\beta }{}๐ฎ_{\alpha \beta }|\tau _\alpha )(\tau _\beta |A),$$
(41)
in terms of which the matrix elements are
$$๐ฎ_{\alpha \beta }=(\tau _\alpha |๐ฎ|\tau _\beta )=(\tau _{ca}|๐ฎ|\tau _{db})=c|๐ฎ(|ab|)|d.$$
(42)
This expression provides the fundamental connection between the two actions of a superoperator. We can define an operation, called sharp, that exchanges the ordinary and left-right actions:
$$๐ฎ^\mathrm{\#}(A)=๐ฎ|A).$$
(43)
Equation (42) implies that
$$๐ฎ_{ca,db}^\mathrm{\#}=c|๐ฎ^\mathrm{\#}(|ab|)|d=(\tau _{cd}|๐ฎ|\tau _{ab})=๐ฎ_{cd,ab}$$
(44)
or, equivalently, that
$$๐ฎ^\mathrm{\#}=\underset{c,a,d,b}{}๐ฎ_{ca,db}|cd||ba|.$$
(45)
With respect to the left-right action, a superoperator works just like an operator. Multiplication of superoperators $``$ and $`๐ฎ`$ is given by
$$๐ฎ=\underset{\alpha ,\beta ,\gamma }{}_{\alpha \gamma }๐ฎ_{\gamma \beta }|\tau _\alpha )(\tau _\beta |,$$
(46)
and the adjoint is defined by
$$(A|๐ฎ^{}|B)=(B|๐ฎ|A)^{}๐ฎ^{}=\underset{\alpha ,\beta }{}๐ฎ_{\beta \alpha }^{}|\tau _\alpha )(\tau _\beta |.$$
(47)
With respect to the ordinary action, superoperator multiplication, denoted as a composition $`๐ฎ`$, is given by
$$๐ฎ=\underset{\alpha ,\beta ,\gamma ,\delta }{}_{\gamma \delta }๐ฎ_{\alpha \beta }\tau _\gamma \tau _\alpha \tau _\beta ^{}\tau _\delta ^{}.$$
(48)
The adjoint with respect to the ordinary action, denoted by $`๐ฎ^\mathrm{\times }`$, is defined by
$$\mathrm{tr}\left([๐ฎ^\mathrm{\times }(B)]^{}A\right)=\mathrm{tr}\left(B^{}๐ฎ(A)\right)๐ฎ^\mathrm{\times }=\underset{\alpha ,\beta }{}๐ฎ_{\alpha \beta }^{}\tau _\alpha ^{}\tau _\beta .$$
(49)
The identity superoperator with respect to the left-right action can be written as
$$๐=\underset{\alpha }{}|\tau _\alpha )(\tau _\alpha |=\underset{c,a}{}|ca||ac|.$$
(50)
When sharped, $`๐`$ becomes the identity superoperator with respect to the ordinary action, denoted by $``$:
$$๐^\mathrm{\#}=\underset{c,a}{}|cc||aa|=II.$$
(51)
The final ingredient we need is the superoperator trace relative to the left-right action, defined by
$$\mathrm{Tr}(๐ฎ)=\underset{\alpha }{}(\tau _\alpha |๐ฎ|\tau _\alpha )=\underset{c,a}{}c|๐ฎ(|aa|)|c=\mathrm{tr}(๐ฎ(I)).$$
(52)
Notice that $`๐(I)=DI`$ and $`(I)=I`$, which give $`\mathrm{Tr}(๐)=D^2`$ and $`\mathrm{Tr}()=D`$.
Now suppose the operators $`|N_\alpha )`$ constitute a complete or overcomplete operator basis; i.e., let the operator kets $`|N_\alpha )`$ span the vector space of operators. It follows that the superoperator $`๐ข`$ defined by
$$๐ข=\underset{\alpha }{}|N_\alpha )(N_\alpha |=๐ข^{}$$
(53)
is invertible with respect to the left-right action. The operators
$$|Q_\alpha )=๐ข^1|N_\alpha )$$
(54)
form a dual basis, which gives rise to the following expressions for the identity superoperator:
$$๐=\underset{\alpha }{}|Q_\alpha )(N_\alpha |=\underset{\alpha }{}|N_\alpha )(Q_\alpha |.$$
(55)
An arbitrary operator $`A`$ can be expanded in terms of the original basis or the dual basis:
$`A`$ $`=`$ $`{\displaystyle \underset{\alpha }{}}|N_\alpha )(Q_\alpha |A)={\displaystyle }_\alpha N_\alpha \mathrm{tr}(Q_\alpha ^{}A),`$ (56)
$`A`$ $`=`$ $`{\displaystyle \underset{\alpha }{}}|Q_\alpha )(N_\alpha |A)={\displaystyle }_\alpha Q_\alpha \mathrm{tr}(N_\alpha ^{}A).`$ (57)
These expansions are unique if and only if the operators $`|N_\alpha )`$ are linearly independent. Later in this section we apply expansions of this sort to density operators.
#### 4.1.2 Pure states and their dual basis
The set of all pure-state projectors in a $`D`$-dimensional Hilbert space,
$$P_\psi =|\psi \psi |,$$
(58)
forms an overcomplete operator basis. To develop operator expansions in terms of the pure-state projectors, we follow the discussion in the preceding subsection and consider the superoperator
$$๐ข=d๐ฑ|P_\psi )(P_\psi |=d๐ฑ|\psi \psi ||\psi \psi |,$$
(59)
where $`d๐ฑ`$ is the unitarily invariant integration measure on projective Hilbert space.
The only Hilbert-space integrals we need to calculate explicitly are those for which the integrand is a function only of an angle $`\theta `$ defined by $`\mathrm{cos}\theta =|e|\psi |`$, where $`|e`$ is some particular unit vector (pure state). The angle $`\theta `$, which runs over the range $`0\theta \pi /2`$, can be thought of as a โpolar angleโ relative to the โpolar axisโ defined by $`|e`$. For integrals of this sort, a convenient form of the integration measure is
$$d๐ฑ=(\mathrm{sin}\theta )^{2D3}\mathrm{cos}\theta d\theta d๐ฎ_{2D3},$$
(60)
where $`d๐ฎ_{2D3}`$ is the standard integration measure on a $`(2D3)`$-dimensional unit sphere. Thus the total volume of $`D`$-dimensional projective Hilbert space is
$$๐ฑ=๐ฎ_{2D3}_0^{\pi /2}๐\theta (\mathrm{sin}\theta )^{2D3}\mathrm{cos}\theta d\theta =\frac{๐ฎ_{2D3}}{2(D1)}=\frac{\pi ^{D1}}{(D1)!},$$
(61)
where $`๐ฎ_{2D3}=2\pi ^{D1}/(D2)!`$ is the volume of a $`(2D3)`$-dimensional unit sphere.
To use the expansions (56) and (57), we need the dual basis $`|Q_\psi )`$, and for that purpose, we need to invert $`๐ข`$. Since $`๐ข`$ is Hermitian relative to the left-right action, we can invert it by diagonalizing it with respect to the left-right action. Given an orthonormal basis $`|a`$, we can write $`๐ข`$ as in Eq. (39),
$$๐ข=\underset{c,a,d,b}{}๐ข_{ca,db}|ca||bd|=\underset{c,a,d,b}{}๐ข_{ca,db}|\tau _{ca})(\tau _{db}|,$$
(62)
where the matrix elements are given by Eq. (38):
$$๐ข_{ca,db}=c|๐ข(|ab|)|d=๐๐ฑc|\psi \psi |ab|\psi \psi |d.$$
(63)
The unitary invariance of the integration measure places stringent constraints on the matrix elements (63). Since the integral in Eq. (63) remains unchanged under a change in the sign of the amplitude $`a|\psi `$ corresponding to a particular basis vector $`|a`$, the matrix elements vanish except when (i) $`a=bc=d`$ or $`a=cb=d`$ or (ii) $`a=b=c=d`$. Furthermore, unitary invariance implies that for each of these cases, all the matrix elements have the same value. Gathering these conclusions together, we have
$$๐ข_{ca,db}=\{\begin{array}{cc}\alpha ,\hfill & a=bc=d\text{ or }a=cb=d\text{,}\hfill \\ \gamma ,\hfill & a=b=c=d\text{,}\hfill \\ 0,\hfill & \text{otherwise.}\hfill \end{array}$$
(64)
We get a relation between $`\alpha `$ and $`\gamma `$ by noting that
$$D(D1)\alpha +D\gamma =\underset{c,a=1}{\overset{D}{}}๐ข_{ca,ca}=๐ฑ,$$
(65)
where the second equality follows from doing the sum within the integral in Eq. (63). We need one more relation, which we get by evaluating explicitly the integral for $`\gamma `$:
$$\gamma =๐๐ฑ|a|\psi |^4=๐ฎ_{2D3}_0^{\pi /2}๐\theta (\mathrm{sin}\theta )^{2D3}(\mathrm{cos}\theta )^5=\frac{2๐ฑ}{D(D+1)}2K.$$
(66)
It follows that $`\alpha =K`$. As a result, we have
$$๐ข=K(2\underset{a}{}|\tau _{aa})(\tau _{aa}|+\underset{\genfrac{}{}{0pt}{}{a,b}{ab}}{}(|\tau _{ab})(\tau _{ab}|+|\tau _{aa})(\tau _{bb}|\left)\right)=K(๐+).$$
(67)
This result gives us immediately that
$$๐๐ฑ|\psi \psi |=๐ข(I)=\frac{๐ฑ}{D}I.$$
(68)
The operators $`\lambda _\alpha `$ introduced in Sect. 2 constitute a complete, orthonormal operator basis, so we can write $`๐`$ as
$$๐=\underset{\alpha }{}|\lambda _\alpha )(\lambda _\alpha |=\frac{|I)(I|}{D}+๐ฏ,$$
(69)
where
$$๐ฏ=\underset{j}{}|\lambda _j)(\lambda _j|$$
(70)
is the superoperator that projects onto the subspace of traceless operators. Plugging Eq. (69) into Eq. (67) gives the diagonal form of $`๐ข`$:
$$๐ข=K\left((D+1)\frac{|I)(I|}{D}+๐ฏ\right).$$
(71)
Orthonormal eigenoperators of $`๐ข`$ are $`\lambda _0=I/\sqrt{D}`$, with eigenvalue $`K(D+1)=๐ฑ/D`$ and the traceless operators $`\lambda _j`$, which are degenerate with eigenvalue $`K=๐ฑ/D(D+1)`$.
We are now prepared to write the inverse of $`๐ข`$ with respect to the left-right action as
$$๐ข^1=\frac{1}{K}\left(\frac{1}{D+1}\frac{|I)(I|}{D}+๐ฏ\right)=\frac{1}{K}\left(๐\frac{}{D+1}\right).$$
(72)
Thus the dual operators of Eq. (54) are given by
$$|Q_\psi )=๐ข^1|P_\psi )=\frac{1}{K}(|P_\psi )\frac{|I)}{D+1})=\frac{D}{๐ฑ}((D+1)P_\psi I).$$
(73)
#### 4.1.3 Alternative diagonalization of $`๐ข`$
In this subsection we rederive Eq. (71) using the special properties of the superoperator $`๐ข`$. These properties are evident from the integral form of $`๐ข`$ in Eq. (59).
* The superoperator $`๐ข`$ is Hermitian relative to the left-right action, which implies that it has a complete, orthonormal set of eigenoperators $`\eta _\alpha `$, $`\alpha =1,\mathrm{},D^2`$, with real eigenvalues $`q_\alpha `$:
$$๐ข=๐ข^{}๐ข=\underset{\alpha }{}q_\alpha |\eta _\alpha )(\eta _\alpha |=\underset{\alpha }{}q_\alpha \eta _\alpha \eta _\alpha ^{}.$$
(74)
* The superoperator $`๐ข`$ is Hermitian relative to the ordinary action,
$$๐ข=๐ข^\mathrm{\times }=\underset{\alpha }{}q_\alpha \eta _\alpha ^{}\eta _\alpha =\underset{\alpha }{}q_\alpha |\eta _\alpha ^{})(\eta _\alpha ^{}|,$$
(75)
which implies that if $`\eta _\alpha `$ is an eigenoperator of $`๐ข`$, then $`\eta _\alpha ^{}`$ is also an eigenoperator with the same eigenvalue. This means that we can choose all the eigenoperators to be Hermitian.
* The superoperator $`๐ข`$ is unitarily invariant, i.e.,
$$๐ข=UU^{}๐ขU^{}U=\underset{\alpha }{}q_\alpha U\eta _\alpha U^{}U\eta _\alpha ^{}U^{},$$
(76)
for any unitary operator $`U`$, which implies that if $`\eta _\alpha `$ is an eigenoperator of $`๐ข`$, then $`U\eta _\alpha U^{}`$ is also an eigenoperator with the same eigenvalue.
The upshot of these three properties is that the eigensubspaces of $`๐ข`$ are invariant under Hermitian conjugation and under all unitary transformations. It is not hard to show that the only such operator subspaces are the subspace of traceless operators and its orthocomplement, the one-dimensional subspace spanned by the unit operator. The result is that $`๐ข`$ must have the form
$$๐ข=K\left(\mu \frac{II}{D}+๐ฏ\right)=K\left(๐+\frac{\mu 1}{D}\right),$$
(77)
where $`K`$ is the eigenvalue of any traceless operator and $`K\mu `$ is the eigenvalue of $`\lambda _0=I/\sqrt{D}`$. Now we use the final property to evaluate $`\mu `$.
* The superoperator $`๐ข`$ is invariant under exchange of the two kinds of action:
$$๐ข=๐ข^\mathrm{\#}=K\left(+\frac{\mu 1}{D}๐\right).$$
(78)
This implies that $`\mu =D+1`$, thus bringing $`๐ข`$ into the form (67), but with $`K`$ not yet determined.
We find the value of $`K`$ by evaluating the superoperator trace, first using Eq. (59),
$$\mathrm{Tr}(๐ข)=\mathrm{tr}(๐ข(I))=๐ฑ,$$
(79)
and then using Eq. (67),
$$\mathrm{Tr}(๐ข)=K\left(\mathrm{Tr}(๐)+\mathrm{Tr}()\right)=KD(D+1).$$
(80)
This gives $`K=๐ฑ/D(D+1)`$, in agreement with Eq. (66).
### 4.2 Separability bounds
We turn now to demonstrating the lower and upper bounds, Eqs. (35) and (36), on the size of the neighborhood of separable states surrounding the maximally state.
To establish the lower bound, we use the results of Sect. 4.1 to formulate operator expansions in terms of product pure states. For a single qudit, any density operator can be expanded as
$$\rho =d๐ฑ|P_\psi )(Q_\psi |\rho )=d๐ฑw_\rho (\psi )P_\psi ,$$
(81)
where
$$w_\rho (\psi )=\mathrm{tr}(\rho Q_\psi )=\frac{D}{๐ฑ}\left((D+1)\psi |\rho |\psi 1\right)$$
(82)
is a quasi-probability distribution, normalized to unity, but possibly having negative values. The analogous product representation for an $`N`$-qudit density operator is
$$\rho =๐๐ฑ_1\mathrm{}๐๐ฑ_Nw_\rho (\psi _1,\mathrm{},\psi _N)P_{\psi _1}\mathrm{}P_{\psi _N},$$
(83)
where
$$w_\rho (\psi _1,\mathrm{},\psi _N)=\mathrm{tr}(\rho Q_{\psi _1}\mathrm{}Q_{\psi _N}).$$
(84)
The $`N`$-qudit quasi-distribution obeys the bound
$$w_\rho (\psi _1,\mathrm{},\psi _N)\left(\begin{array}{c}\text{smallest eigenvalue of}\\ Q_{\psi _1}\mathrm{}Q_{\psi _N}\end{array}\right)=\frac{D^{2N1}}{๐ฑ^N}$$
(85)
This follows from the fact that $`Q_\psi `$ has a nondegenerate eigenvalue, $`D^2/๐ฑ`$, and a $`(D1)`$-fold degenerate eigenvalue, $`D/๐ฑ`$. Thus the most negative eigenvalue of the product operator $`Q_{\psi _1}\mathrm{}Q_{\psi _N}`$ is $`(D/๐ฑ)(D^2/๐ฑ)^{N1}=D^{2N1}/๐ฑ^N`$.
We can use the lower bound (85) to place a similar lower bound on the quasi-distribution for the mixed state $`\rho _ฯต`$ of Eq. (34). Since the quasi-distribution for the maximally mixed state, $`M_{D^N}`$, is the uniform distribution $`1/๐ฑ^N`$, we have
$$w_{\rho _ฯต}(\psi _1,\mathrm{},\psi _N)=\frac{1ฯต}{๐ฑ^N}+ฯตw_{\rho _1}\frac{1ฯต(1+D^{2N1})}{๐ฑ^N}.$$
(86)
We conclude that if $`ฯต1/(1+D^{2N1})`$, then $`w_{\rho _ฯต}`$ is nonnegative and the qudit state $`\rho _ฯต`$ is separable. This establishes the lower bound (35) on the size of the neighborhood of separable states surrounding the maximally mixed state.
The upper bound (36) on the size of the separable neighborhood can be established with the help of an exact separability condition for a particular $`N`$-qubit state, obtained by Dรผr, Cirac, and Tarrach and also by Pittenger and Rubin . We consider the $`N`$-qudit state,
$$\rho _ฯต=(1ฯต)M_{D^N}+ฯต|\mathrm{\Psi }_{\mathrm{cat}}\mathrm{\Psi }_{\mathrm{cat}}|,$$
(87)
where
$$|\mathrm{\Psi }_{\mathrm{cat}}=\frac{1}{\sqrt{D}}\underset{a=1}{\overset{D}{}}|a\mathrm{}|a,$$
(88)
is an $`N`$-qudit โcat state.โ We call the mixed state (87) an $`ฯต`$-cat state.
Now project each qudit onto the two-dimensional (qubit) subspace spanned by $`|1`$ and $`|2`$. The local projection operator on each qudit is $`\mathrm{\Pi }=|11|+|22|`$, and the normalized $`N`$-qubit state after projection is
$$\rho _ฯต^{}=\frac{\mathrm{\Pi }^N\rho _ฯต\mathrm{\Pi }^N}{\mathrm{tr}(\mathrm{\Pi }^N\rho _ฯต)}=(1ฯต^{})M_{2^N}+ฯต^{}|\mathrm{\Phi }_{\mathrm{cat}}\mathrm{\Phi }_{\mathrm{cat}}|,$$
(89)
where
$$|\mathrm{\Phi }_{\mathrm{cat}}\frac{1}{\sqrt{2}}\left(|1\mathrm{}|1+|2\mathrm{}|2\right)$$
(90)
is the cat state for $`N`$ qubits and
$$ฯต^{}=\frac{2ฯต/D}{(2/D)^N(1ฯต)+2ฯต/D}.$$
(91)
Dรผr, Cirac, and Tarrach and also Pittenger and Rubin have shown that the $`N`$-qubit $`ฯต`$-cat state (89) is nonseparable (entangled) if and only if $`ฯต^{}>1/(1+2^{N1})`$, a condition equivalent to $`ฯต>1/(1+D^{N1})`$. Since local projections on each qudit cannot create entanglement, we can conclude that the $`N`$-qudit $`ฯต`$-cat state (87) is nonseparable under the same condition. This establishes the upper bound (36) on the size of the separable neighborhood around the maximally mixed state.
Pittenger and Rubin have recently extended the result of Dรผr, Cirac, and Tarrach for the $`N`$-qubit $`ฯต`$-cat state. They have shown directly that the $`N`$-qudit $`ฯต`$-cat state (87) is nonseparable if $`ฯต>1/(1+D^{N1})`$, and they have also shown that the same condition is a necessary and sufficient condition for entanglement when $`D`$ is prime. Their argument is akin to the correlation-coefficient argument we give in Sect. 3. |
warning/0001/nlin0001003.html | ar5iv | text | # On Soliton-type Solutions of Equations Associated with N-component Systems 11footnote 1PACS numbers 03.40.Gc, 11.10.Ef, 68.10.-m, AMS Subject Classification 58F07, 70H99, 76B15
## 1 Introduction
The solution of nonlinear evolution equations using techniques from algebraic geometry was initially developed to handle $`N`$-phase wave trains. With this approach, parameterized families of quasi-periodic and soliton solutions are associated with Hamiltonian flows on level sets of finite-dimensional phase spaces. In Section 2, these flows are described using the $`\mu `$ variable representation on symmetric products of Riemann surfaces. The first integrals in the quasi-periodic case have the form $`P_j^2=C(\mu _j)`$ where $`C(\mu _j)`$ is a polynomial with constant coefficients and $`P_j`$ is the conjugate variable for $`\mu _j`$. The polynomial $`C(E)`$ is called the spectral polynomial and determines the form of the first integrals.
The algebraic geometric approach provides a way to construct solutions by analytical or numerical integration of a system of Hamiltonian equations for these $`\mu `$ variables. Solutions of the nonlinear PDEโs are expressed in terms of these variables by using the trace formulas. The $`\mu `$ variable representation yields an action angle representation on a Jacobi variety (invariant variety in the phase space) that linearizes the Hamiltonian flow, and in the quasi-periodic case the solution of the $`\mu `$ equations is reduced to a Jacobi inversion problem. Solutions are then expressed using Riemann theta functions. For details see, for example, Mumford(1983) and Ercolani and McKean (1990).
As pairs of roots of the polynomial $`C(E)`$ coalesce, soliton solutions begin to appear. Applying this limit to the first integrals and the equations of motion in terms of the $`\mu `$ variables for quasi-periodic solutions, Hamiltonian systems of ODEโs that describe soliton solutions are obtained. Soliton solutions are computed by solving these equations and using the trace formula to connect them to the associated nonlinear PDEโs. In the soliton limit the Jacobi inversion problem for the system often reduces to a system of algebraic equations. As will be shown, these algebraic equations are exactly solvable in the case of genus one and two. For details about the connection between soliton and quasi-periodic solutions see for example Ablowitz and Ma (1981) and Alber and Alber(1985). The exact representations obtained for soliton solutions are shown to be related to the Hirota $`\tau `$-functions. The case of genus $`n`$ solutions is analogous to the lower genus case and can be described using similar formulas. Using the action-angle representation, the algebraic geometric approach also introduces a powerful way to compute the phase shifts due to soliton interactions (see Alber and Marsden(1992) for details.)
In this paper, soliton solutions of multicomponent systems of equations are studied using the algebraic geometric approach. The soliton fission effect, kink to anti-kink transitions, and multi-peaked solitons are demonstrated using a class of commuting Hamiltonian systems on Riemann surfaces. These first two effects manifest themselves in the soliton limit of the genus two quasi-periodic solution when six roots of the spectral polynomial $`C(E)`$ coalesce in a pairwise fashion. Exact formulas for these solutions are obtained and asymptotic and numerical analysis of them is performed. The technique used to obtain these limiting solutions is demonstrated explicitly in the case of the coupled Korteweg-de Vries (cKdV) and the coupled Dym (cDym) equations and physically relevant equations associated with them.
The modified cKdV system is
$`u_t`$ $`=`$ $`v_x\frac{3}{2}uu_x+K_1u_x,`$ (1.1)
$`v_t`$ $`=`$ $`\frac{1}{4}u_{xxx}vu_x\frac{1}{2}uv_x+K_1v_x,`$ (1.2)
which reduces to the cKdV system for $`K_1=0`$. Otherwise $`K_1`$ is viewed as a small constant. The coupled Dym equations are given by
$`u_t`$ $`=`$ $`\frac{1}{4}u_{xxx}\frac{3}{2}uu_x+v_x+K_1u_x,`$ (1.3)
$`v_t`$ $`=`$ $`u_xv\frac{1}{2}uv_x+K_1v_x.`$ (1.4)
The general method is demonstrated by describing quasi-periodic and soliton solutions for the cKdV and cDym systems for genus two and less.
The cKdV and cDym equations are both generic examples of $`N`$-component systems. Energy dependent Schrลdinger operators and $`bi`$-Hamiltonian structures for multicomponent systems were investigated in Antonowicz and Fordy(1987). Quasi-periodic and soliton solutions were studied in connection with Hamiltonian systems on Riemann surfaces in Alber et al.(1997). In Alber et al. (1994) it was also shown that the presence of a pole in the associated Schrลdinger operator yields a special class of weak billiard solutions for nonlinear PDEโs.
The soliton fission effect, kink to anti-kink transitions, and multi-peaked solitons extend to equations that model physical phenomena. The generalized Kaup equation, the classical Boussinesq system, and the equations governing second harmonic generation (SHG) are each connected to the cKdV system through nonsingular transformations. Direct application of these transformations enables solutions of the cKdV system to be interpreted in the context of these related equations. Such transformations are given explicitly in Appendix A. Both the Boussinesq System,
$`U_t+W_x+UU_x`$ $`=`$ $`\gamma U_x,`$ (1.5)
$`W_t+U_{xxx}+(WU)_x`$ $`=`$ $`\gamma W_x,`$ (1.6)
and the generalized Kaup equations
$`\pi _t`$ $`=`$ $`\varphi _{xx}+\frac{1}{3}(13\sigma )\delta ^2\varphi _{xxxx}ฯต(\varphi _x\pi )_x+\alpha \pi _x,`$ (1.7)
$`\varphi _t`$ $`=`$ $`\alpha \varphi _x+{\displaystyle \frac{ฯต}{2}}\varphi _x^2+\pi ,`$ (1.8)
arise from the theory of shallow water waves (see Whitham(1974)). Here $`\gamma `$ and $`\alpha `$ are small parameters.
In optics, the interaction of a wave envelope at frequency and wavenumber $`(w,k)`$ with a second wave at twice the frequency is modeled by the system of equations
$`(q_1)_x`$ $`=`$ $`2q_2q_1^{},`$ (1.9)
$`(q_2)_\tau `$ $`=`$ $`q_1^2.`$ (1.10)
This process is called second harmonic generation in nonlinear optics and is used to convert laser light to its second harmonic frequency.
The scattering problem for the energy dependent Schลdinger operators was studied by Jaulent(1972) and Jaulent and Jean(1976). The completely integrable variant of the Boussinesq system (1.5)-(1.6) was first introduced by Kaup(1972). In Matveev and Yavor(1979) $`\theta `$-functions were used to describe quasi-periodic solutions of the Boussinesq System. They also described a particular type of $`N`$-soliton solution using singular classes of $`\theta `$-functions. Rational solutions were studied in Sachs(1998) in connection with a pair of Calogero-Moser equations coupled through the constraints. Martinez Alonso and Medina Reus(1992) and Estevez et al.(1994) described some of the soliton solutions using Hirotaโs $`\tau `$-functions. Using asymptotics of these soliton solutions they also demonstrated soliton fission. A connection between the SHG system and the cKdV system was recently discussed by Khusnutdinova and Steudel(1998).
## 2 Generating Equations for the Coupled KdV and Dym Equations
We begin by describing the general approach of generating equations and applying it to the cKdV and cDym equations. Details for the general case of $`N`$-component systems are discussed in Alber et al.(1997).
### 2.1 Dynamical Generating Equations.
The hierarchy of the coupled KdV and Dym equations is obtained as the compatibility condition for the eigenfunction of the linear system of equations
$`L\psi `$ $`=`$ $`0,`$ (2.1)
$`\psi _t`$ $`=`$ $`A\psi .`$ (2.2)
The time flow is produced by the linear differential operator
$$A=B\frac{d}{dx}\frac{B_x}{2},$$
(2.3)
where $`B(x,t,E)`$ is a specified rational function. The operator $`L`$ is assumed to be of the energy dependent Schrลdinger type,
$$L=\frac{d^2}{dx^2}+V(x,t,E),$$
(2.4)
with a rational potential having the form
$$V(x,t,E)=\frac{_{j=0}^Nv_j(x,t)E^j}{_{i=0}^Mr_iE^i},$$
(2.5)
where $`r_i`$ are constants and $`v_j(x,t)`$ are functions of the variable $`x`$ and the parameter $`t`$. $`E`$ is a complex spectral parameter. In particular, the potential is chosen as
$$V(E)=\kappa E^2+u(x,t)E+v(x,t),$$
(2.6)
for the cKdV system or
$$V(E)=u(x,t)+E+\frac{v(x,t)}{E},$$
(2.7)
to recover the cDym system. Here $`\kappa =\pm 1`$. One chooses $`\kappa =1`$ to establish the transformation from the cKdV system to the SHG system, and $`\kappa =1`$ to establish the transformation from the cKdV system to the Boussinesq System. Notice that the main difference between the cKdV and cDym systems is the presence of a pole in the Schrลdinger operator (2.4) associated with the cDym equations. The pole in the potential for the cDym case was shown in Alber et al.(1994) to be a necessary feature for systems with weak billiard solutions.
The compatibility condition of (2.1)-(2.2) can be found by taking the $`t`$ derivative of (2.1), acting on (2.2) by $`L`$, and forcing the fact that these two operators commute. This leads to the following system of equations
$$(L_t+[L,A])\psi =0,L\psi =0,$$
(2.8)
where $`[L,A]=LAAL`$ is the commutator of $`L`$ and $`A`$. Using the definition of the differential operator $`A`$ in (2.3) and $`L`$ in (2.4), this Lax equation yields
$$\frac{V}{t}=\frac{1}{2}\frac{^3B}{x^3}+2\frac{B}{x}V+B\frac{V}{x},$$
(2.9)
which is a generating equation for the coefficients of the differential operator $`A`$. By taking $`B`$ to be the rational function,
$$B(x,t,E)=\underset{k=r}{\overset{m}{}}b_{mk}(x,t)E^k=E^r\underset{k=1}{\overset{n}{}}(E\mu _k(x,t)),$$
(2.10)
substituting it into the generating equation (2.9) and equating like powers of $`E`$, a recurrence chain of equations for the coefficients $`b_j`$ is obtained. Evaluating these equations one by one, a PDE for the coefficient $`b_n`$ is obtained, where $`n=m+r`$. By considering all possible values of $`n`$ and $`m`$, a hierarchy of systems generated by the Lax equation with a given potential (generating equation) is obtained.
Assuming that $`{\displaystyle \frac{V}{t}}=0`$ in (2.9) and integrating gives the stationary generating equation which has the form
$$B^{\prime \prime }B+\frac{1}{2}(B^{})^2+2B^2V=C(E),$$
(2.11)
where the choice of $`B(E)`$ from (2.10) ensures that $`C(E)`$ is a rational function with constant coefficients. These coefficients are the first integrals and parameters of the coupled system of equations. (For details about the general method see Alber and Alber(1985).) This equation gives rise to level sets in the phase space $`๐^n`$ corresponding to the Riemann surface
$$W^2=C(E).$$
(2.12)
The dynamics of genus $`n`$ quasi-periodic solutions for $`u`$ and $`v`$ with respect to the $`x`$ and $`t`$ coordinates are captured as flows on the level set produced by a symmetric product of $`n`$ copies of the Riemann surface (2.12) (see Alber et al.(1997) for details).
The method of generating equations yields the cKdV and cDym equations by using $`B(x,t,E)=b_0(x,t)E+b_1(x,t)`$. In Appendix A solutions of the generalized Boussinesq and generalized Kaup equations are linked to these systems. Further, the second harmonic generation equations are obtained when $`B(x,t,E)=b_2(x,t)E^1`$.
## 3 Solutions of the cKdV and cDym Systems from Dynamical Systems on Riemann Surfaces
In this section we obtain finite-dimensional Hamiltonian systems on Riemann surfaces for the $`\mu `$ variables defined in (2.10) as the roots of the function $`B(x,t,E)`$. These systems capture the essential dynamics of quasi-periodic and soliton solutions of the cKdV and cDym systems. The quasi-periodic solutions are often called $`n`$-gap solutions in physics literature and quasi-periodic solutions of genus $`n`$ in mathematics literature. In Appendix B the solutions of these finite-dimensional Hamiltonian systems are linked to the solutions of the integrable nonlinear PDEโs through trace formulas of the general form $`u=\alpha _{j=1}^n\mu _j+\beta `$, where $`\alpha `$ and $`\beta `$ are constants. This general construction also provides links between the solutions of the cKdV and cDym systems and other nonlinear PDEโs.
Choose $`B(x,t,E)`$ as in (2.10) where $`n=r+m`$ is the genus of the desired solution. For the cKdV case, we substitute (2.6) into (2.9) and (2.11) to obtain
$$u_tE+v_t=\frac{1}{2}B^{\prime \prime \prime }+2\kappa B^{}E^2+2B^{}uE+2B^{}v+Bu^{}E+Bv^{},$$
(3.1)
and
$$B^{\prime \prime }B+\frac{1}{2}(B^{})^2+2\kappa B^2E^2+2B^2uE+2B^2v=C(E).$$
(3.2)
For the cDym system we substitute the potential (2.7) into the same equations to obtain
$$u_tE+v_t=\frac{1}{2}B^{\prime \prime \prime }E+2B^{}Eu+2B^{}E^2+2B^{}v+Bu^{}E+Bv^{},$$
(3.3)
and
$$B^{\prime \prime }B+\frac{1}{2}(B^{})^2+2B^2u+2B^2E+\frac{2B^2v}{E}=C(E).$$
(3.4)
By equating like powers of $`E`$ on the left and right hand side of the equations (3.2) and (3.4) we obtain the necessary forms for $`C(E)`$. Therefore we write
$$C(E)=2\kappa E^{2r}\underset{i=1}{\overset{2(n+1)}{}}(Em_i),$$
(3.5)
for cKdV equations and
$$C(E)=2E^{(2r+1)}\underset{i=1}{\overset{2(n+1)}{}}(Em_i),$$
(3.6)
for the cDym equations, where the real numbers $`m_i`$ are the roots of the polynomials $`C(E)`$.
### 3.1 Finite-Dimensional Hamiltonian Systems.
From the generating equation above it follows that the $`\mu _i`$โs, which are the roots of the function $`B(x,t,E)`$, are solutions of finite-dimensional Hamiltonian systems. By solving these Hamiltonian systems and using the trace formula, the dynamics of the roots $`\mu _i`$ are connected to the functions $`u`$ and $`v`$ to obtain solutions of the PDEโs.
Namely, Hamiltonian equations for $`\mu _i`$โs are obtained by substituting $`E=\mu _i`$ into the generating equations (3.2), (3.4). This yields the following systems of equations for cKdV:
$$\mu _i^{}=\pm \frac{2\sqrt{\kappa _{j=1}^{2n+2}(\mu _im_j)}}{_{ji}(\mu _i\mu _j)}i=1,\mathrm{},n,$$
(3.7)
for the flow in space, and
$$\dot{\mu }_i=\frac{2(_{ji}\mu _j)\sqrt{\kappa _{j=1}^{2n+2}(\mu _im_j)}}{_{ji}(\mu _i\mu _j)}i=1,\mathrm{},n,$$
(3.8)
for the flow in time. Here $`n=m+r`$ from the definition of $`B`$ in (2.10) and $`m_i`$ are fixed real roots of $`C(E)`$ in (3.5). The plus/minus refers to which branch of the Riemann surface (2.12) the solution is on. Systems (3.7)-(3.8) are commuting Hamiltonian systems with Hamiltonians
$$H=\underset{j=1}{\overset{n}{}}\frac{D(\mu _j)(P_j^2C(\mu _j))}{_{rj}^n(\mu _j\mu _r)}$$
(3.9)
where $`D(\mu _i)=1`$ and $`D(\mu _i)=_{ji}\mu _j`$ in the stationary and dynamical cases respectively. These systems share the same complete set of first integrals: $`P_j^2=C(\mu _j),j=1,\mathrm{},n`$ where the polynomial $`C(E)`$ is defined by (3.5) (for details see Alber et al.(1997)).
For the cDym equation we see that
$$\mu _i^{}=\pm \frac{2\sqrt{_{j=1}^{2n+2}(\mu _im_j)}}{\sqrt{\mu _i}_{ji}(\mu _i\mu _j)}i=1,\mathrm{},n,$$
(3.10)
for the spatial flow, and
$$\dot{\mu }_i=\frac{2(_{ji}\mu _j)\sqrt{_{j=1}^{2n+2}(\mu _im_j)}}{\sqrt{\mu _i}_{ji}(\mu _i\mu _j)}i=1,\mathrm{},n,$$
(3.11)
for the time flow. Systems (3.10)-(3.11) are also Hamiltonian systems and they share the same complete set of first integrals: $`P_j^2=C(\mu _j),j=1,\mathrm{},n`$ where polynomial $`C(E)`$ is defined by (3.6).
Systems (3.7)-(3.8) and (3.10)-(3.11) can be solved analytically by reducing them to Jacobi inversion problems. We will demonstrate the general method in the next section. These equations are also easily integrated numerically. Perhaps the best method for accomplishing numerical integration with the use of symplectic integrators. Such integrators preserve the Poincare invariants and are stable over a long period of time, see for example Channell and Scovel(1990).
### 3.2 The Trace Formulas.
The connection between the solutions $`u`$ and $`v`$ of the cKdV equation and the $`\mu _i`$โs from (3.7),(3.8) is derived in Appendix B and is given by
$`u`$ $`=`$ $`2\kappa {\displaystyle \underset{i=1}{\overset{n}{}}}\mu _i+2\kappa K_1,`$ (3.12)
$`v`$ $`=`$ $`2\kappa {\displaystyle \underset{i<jn}{}}\mu _i\mu _j+{\displaystyle \frac{3}{4\kappa }}u^2K_1u+K_2,`$ (3.13)
where
$`K_1`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \underset{i=1}{\overset{2n+2}{}}}m_i,`$ (3.14)
$`K_2`$ $`=`$ $`\kappa {\displaystyle \underset{i<j2n+2}{}}m_im_j\kappa K_1^2.`$ (3.15)
Observe that the small parameter $`K_1`$ from (1.1) is zero if $`_{i=1}^{2n+2}m_i=0`$. The trace formulas for the cDym system are
$`u`$ $`=`$ $`2{\displaystyle \underset{i=1}{\overset{n}{}}}\mu _i+2K_1,`$ (3.16)
$`v`$ $`=`$ $`{\displaystyle \frac{1}{4}}u^{\prime \prime }2{\displaystyle \underset{1i<jn}{}}\mu _i\mu _j+{\displaystyle \frac{3}{4}}u^2K_1u+K_2.`$ (3.17)
## 4 Classification of Limits of Periodic Solutions
For the next three sections we seek various periodic (genus-one) solutions of the Boussinesq System. Therefore unless otherwise stated we assume $`r=0`$ from (2.10) and $`\kappa =1`$ from (2.6). We first obtain the periodic traveling-wave solutions and then show that they are equivalent to solutions obtained in terms of a $`\mu `$ variable in the case $`n=1`$. This provides a natural introduction to the algebraic geometric method. A one-soliton solution is then obtained by deforming the Riemann surface of the genus-one periodic solution. This method of first finding periodic/quasi-periodic solutions and then deforming the level set (Riemann surface) in the phase space to obtain soliton solutions will be utilized throughout this paper. (For details about general approach see amongst others Ablowitz and Ma(1981) and Alber and Alber(1985).)
### 4.1 Periodic Traveling-Wave Solutions
Let $`U=U(\zeta )`$ and $`W=W(\zeta )`$ where $`\zeta =xct`$ so that the Boussinesq System becomes
$`cU^{}+W^{}+UU^{}`$ $`=`$ $`\gamma U^{},`$ (4.1)
$`cW^{}+U^{\prime \prime \prime }+(WU)^{}`$ $`=`$ $`\gamma W^{},`$ (4.2)
where $`W^{}`$ and $`U^{}`$ denote differentiation with respect to $`\zeta `$. Then (4.1) gives
$$W=\eta U\frac{1}{2}U^2+\tau _0,$$
(4.3)
where $`\eta =c+\gamma `$ and $`\tau _0`$ is a constant of integration. Plugging this into (4.2) and integrating twice gives
$$\frac{\tau _0\eta ^2}{2}U^2+\frac{\eta }{2}U^3+\frac{1}{2}(U^{})^2\frac{1}{8}U^4=\tau _1U+\tau _2,$$
(4.4)
where $`\tau _1,\tau _2`$ are constants of integration. Writing this as an integral equation and taking a square root we obtain
$$d\zeta =\pm \frac{dU}{\sqrt{C_4(U)}}$$
(4.5)
where
$$C_4(U)=\underset{l=1}{\overset{4}{}}(Um_j)=U^44\eta U^34(\tau _0\eta ^2)U^28\tau _1U8\tau _2.$$
Notice that the right hand side of this differential equation is multi valued since it involves a square root. This is uniquely defined on a Riemann surface of genus $`1`$ parametrized by a pair $`(W,E)`$ where
$$W^2=C_4(E)=\underset{l=1}{\overset{4}{}}(Em_j).$$
(4.6)
One indicates one of two sheets of the Riemann surface by choosing a particular sign in front of the square root: $`W=\pm \sqrt{C_4(E)}`$. Therefore $`U`$ in (4.5) is considered on a particular sheet of the Riemann surface (4.6) and so meaningful integration can take place.
The equation (4.5) can also be obtained from the $`\mu `$-equations (3.7) and (3.8) for $`n=1`$. Notice that the trace formula shows that $`U`$ and $`\mu `$ are linearly related so that one can substitute $`\mu `$ instead of $`U`$ in (4.5). Using $`\mu `$ in this case will make it consistent with the formulas in the case when $`n=2`$ to be described in the next section.
After integrating (4.5), the following Jacobi inversion problem is obtained
$$\theta =xct+\theta _0=\frac{1}{2}_{\mu _0}^\mu \frac{d\mu }{\sqrt{_{i=1}^4(\mu m_i)}},$$
(4.7)
As stated before, this is a typical Jacobi inversion problem (Mumford (1983)). This integral is inverted using Jacobiโs elliptic functions,
$$\frac{\sqrt{(m_1m_3)(m_2m_4)}}{2}_{\mu _0}^\mu \frac{d\mu }{\sqrt{_{i=1}^4(\mu m_i)}}=_{z_0}^z\frac{dz}{\sqrt{(1z^2)(1k^2z^2)}},$$
(4.8)
where
$$z^2=\frac{(m_2m_4)(\mu m_1)}{(m_1m_4)(\mu m_2)},k^2=\frac{(m_1m_4)(m_2m_3)}{(m_2m_4)(m_1m_3)}.$$
(4.9)
Notice that (4.8) is an elliptic integral of the first kind (see Mumford (1983)). Therefore, $`\mu `$ is obtained explicitly and
$$\mu (x,t)=\frac{m_2(m_4m_1)\mathrm{sn}^2(k,\omega )+m_1(m_2m_4)}{m_2m_4+(m_4m_1)\mathrm{sn}^2(k,\omega )},$$
(4.10)
where $`\mathrm{sn}(k,\omega )`$ is the Jacobi $`\mathrm{sine}`$ function and $`\omega =(xct+\theta _0)/\sqrt{(m_1m_3)(m_2m_4)}`$. This function is plotted in Figure B.1.
Periodic solutions correspond to the case when all $`m_i`$ are distinct. In this case, each fixed point in the phase space repels the trajectories so that for any initial value, $`\mu `$ oscillates periodically between the two nearest points. Notice that $`\mu (0)`$ must be chosen so that the right hand side of (4.7) is real. Then the fixed points repel and $`\mu `$ remains real valued. The solution $`\mu `$ leads to solutions for $`U`$ and $`W`$ through the trace formulas. The shape of $`U`$ is essentially the same as $`\mu `$. The shape of $`W`$ is aperiodic and is discussed further in the next section. Equation (4.7) can be interpreted as defining an angle variable $`\theta `$ where $`c`$ is then the action variable. From (4.7) it follows that in terms of these variables the initial Hamiltonian flow linearizes. It also can be viewed as an Abel -Jacobi map from a hyperelliptic curve
$$W^2=C(E)=2(\mu m_1)(\mu m_2)(\mu m_3)(\mu m_4),$$
(4.11)
or in other words, a Riemann surface of genus one, onto the Jacobi variety: $`J=[๐|wZ]`$ where $`๐`$ is a complex plane and $`w`$ is the period lattice of the holomorphic differential from (4.7).
### 4.2 One-Soliton Solution of Kaup Type
To examine soliton solutions, one deforms the Riemann surface $`W^2=C_4(E)`$. (For details about soliton deformations see amongst others Ablowitz and Ma(1981) and Alber and Alber(1985).) Here we consider the limit $`m_1m_2a`$, i.e. where the 2 roots $`m_1`$ and $`m_2`$ coalesce into one point. As this limit is approached, a soliton solution is obtained as the period of the periodic solution increases to infinity. In this case $`a`$ is called a double point. This double point is an attractor in the phase space. Without loss of generality, we assume $`a<m_3<m_4`$. This assumption leads to the realization that the only pertinent solution is obtained when $`a<\mu (0)<m_3`$.
On the principle branch of the square root,
$`xct+\theta _0`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle _{\mu _0}^\mu }{\displaystyle \frac{d\mu }{(\mu a)\sqrt{(\mu m_3)(\mu m_4)}}},`$ (4.12)
$`=`$ $`\left[(am_4)(am_3)\right]^{1/2}\mathrm{arctanh}\left(\sqrt{{\displaystyle \frac{(am_4)(\mu m_3)}{(am_3)(\mu m_4)}}}\right),`$ (4.13)
so that
$$\mu =\frac{m_3(m_4a)+m_4(am_3)\mathrm{tanh}^2(\omega )}{(m_4a)+(am_3)\mathrm{tanh}^2(\omega )},$$
(4.14)
where $`\omega =(xct+\theta _0)\sqrt{(am_4)(am_3)}`$ and $`a,m_3,m_4`$ are functions of $`c`$. Using the trace formulas (3.12)-(3.13) and the transformations (A.3)-(A.4), the exact formulas for $`U`$ and $`W`$ from the Boussinesq equation are obtained. For example,
$$U=\frac{4a^2+(m_3m_4)^2+2a(m_3+m_4)+(m_4^2m_3^2)\mathrm{cosh}(2\omega )}{2a(m_3+m_4)+(m_3m_4)\mathrm{cosh}(2\omega )}.$$
(4.15)
Notice that $`\mu `$ has a shape similar to a KdV solitary wave. It has a peak at $`m_3`$ and approaches $`m_1`$ for large $`|x|`$. From this we conclude that $`U`$ is also shaped like a KdV soliton with a peak of height $`4a+2m_42m_3`$ and for large $`|x|`$ approaches $`2m_3+2m_4`$. One might be surprised that the soliton does not approach 0 as $`x\pm \mathrm{}`$. But remember that this is the modified cKdV equation with $`K_10`$. Howver we may choose the parameters so that the solution does approach zero. The solution is plotted in Figure B.2. This solution was first found by Kaup(1972) using the inverse scattering transform in case when $`m_3=m_4`$ and $`a=0`$, hence $`K_1=0`$. One advantage of the approach used here is that $`W`$ is easily found using the trace formulas, and the relationship between $`U`$ and $`W`$ is seen explicitly.
### 4.3 Solutions with Two Peaks.
Observe next that since $`n=1`$, $`W`$ is only a quadratic in $`U`$. The polynomial
$$W=\frac{1}{2}U^2+(2a+m_3+m_4)U4am_34am_4+(m_3m_4)^2,$$
(4.16)
obtained from the trace formula is a parabola in $`U`$ with vertex at $`U=2a+m_3+m_4`$. Since $`U`$ has the shape of a solitary wave, $`W`$ has either one or two peaks depending on the parameters defining the vertex of $`U`$. (See Figure B.2). If $`U`$ is concave down, then $`W`$ is a double peaked soliton if $`m_3+m_4<2a`$ and $`3m_3<2a+m_4`$. In the unperturbed case ($`\gamma =0`$), these conditions reduce to $`m_3<0<a`$. Otherwise $`W`$ has a single peak. The concave up case is similar.
### 4.4 One Kink Solution
The one-kink solutions were initially found by Alonso and Rues(1992) as the simplest solutions obtained using the bilinear formalism of the Kyoto school. Our method introduces an alternate description of soliton fusion and fission for $`n=2`$ and simplifies the computations.
To obtain the one-kink solutions, we deform the Riemann surface further by taking the limit $`m_3m_4`$, so that $`m_1=m_2=a_1`$ and $`m_3=m_4=a_2`$. Note that if $`\kappa =1`$ this limit is not permitted for real valued $`\mu `$ in (3.7). However, for $`\kappa =1`$ the system is readily solved for real $`\mu `$ producing a special case of equation (4.5) where $`\tau _1=0=\tau _2=\gamma `$. After inverting, the angle variable is
$`\theta =xct+\theta _0`$ $`=`$ $`\pm {\displaystyle \frac{2dU}{U(U2c)}}={\displaystyle \frac{2}{c}}\mathrm{arctanh}\left({\displaystyle \frac{Uc}{c}}\right),`$ (4.17)
and we find using the trace formulas and the transformations obtained in Appendix A that
$`U`$ $`=`$ $`\pm c\mathrm{tanh}({\displaystyle \frac{c}{2}}(xct+\theta _0))+c_2,`$ (4.18)
$`W`$ $`=`$ $`\pm {\displaystyle \frac{c^2}{2}}\mathrm{sech}^2({\displaystyle \frac{c}{2}}(xct+\theta _0)).`$ (4.19)
These functions are plotted in Figure B.3. Here $`U`$ is always a kink or anti-kink while $`W`$ is always similar to a typical KdV soliton. Observe that the speed of the soliton, $`c^2/2`$ is exactly the amplitude of $`W`$ and is proportional to the square of the amplitude of the $`U`$ soliton. This connection between wave speed and amplitude is reminiscent of that found in the KdV equation.
## 5 Classification of Limits of Genus-Two Solutions
In this section genus-two quasi-periodic solutions are constructed. By deforming the Riemann surface (spectral polynomial), two-soliton solutions are obtained and several types of soliton-soliton interactions are described.
### 5.1 Quasi-periodic Genus-Two Solutions
In the genus-two case $`U`$ is given by the trace formulas to be the sum of two periodic functions, $`\mu _1`$ and $`\mu _2`$, and a constant. In general the functions, $`\mu _1`$ and $`\mu _2`$, have noncommensurate periods, so that the solution $`U`$ is generically quasi-periodic and systems (3.7) and (3.8) are defined on the symmetric product of two copies of the Riemann surface (hyperelliptic curve) of genus two given by
$$W^2=C_6(E),$$
(5.1)
where
$$C_6(E)=\kappa \underset{l=1}{\overset{6}{}}(Em_j).$$
(5.2)
After reordering the equations, summing them up, and integrating, the following Jacobi inversion problem is obtained:
$`\theta _1`$ $`=`$ $`{\displaystyle _{\mu _1^0}^{\mu _1}}{\displaystyle \frac{d\mu _1}{\sqrt{C_6(\mu _1)}}}+{\displaystyle _{\mu _2^0}^{\mu _2}}{\displaystyle \frac{d\mu _2}{\sqrt{C_6(\mu _2)}}}=2x2a_1t+\theta _1^0,`$ (5.3)
$`\theta _2`$ $`=`$ $`{\displaystyle _{\mu _1^0}^{\mu _1}}{\displaystyle \frac{\mu _1d\mu _1}{\sqrt{C_6(\mu _1)}}}+{\displaystyle _{\mu _2^0}^{\mu _2}}{\displaystyle \frac{\mu _2d\mu _2}{\sqrt{C_6(\mu _2)}}}=2x2a_2t+\theta _2^0.`$ (5.4)
Inverting the Abel-Jacobi map defined by (5.3)-(5.4) results in expressions for $`\mu _1`$ and $`\mu _2`$ in terms of Riemann $`\theta `$-functions. (For details about the Abel-Jacobi map see Mumford(1983), Matveev and Yavor(1979), and Ercolani and McKean (1990).)
Having derived the genus-two quasi-periodic solutions, several limiting cases will now be explored to introduce solitons. Below each distinct case is considered.
### 5.2 One-Soliton Solution on a Quasi-periodic Background
A one-soliton solution on a quasi-periodic background is obtained in the limit $`m_1m_2a`$. Just as in the one-soliton case, a soliton is created as the Riemann surface is manipulated by pinching two elements of the spectrum. However, in this case there are two $`\mu `$ variables and the orbit for only one $`\mu `$-variable is changed. The other $`\mu `$ variable remains periodic. The result is a solution with a Kaup type soliton on a quasi-periodic background, and it is plotted in Figure B.4. The problem of inversion may be written in the following way,
$`\theta _1`$ $`=`$ $`{\displaystyle _{\mu _1^0}^{\mu _1}}{\displaystyle \frac{d\mu _1}{(\mu _1a)\sqrt{P_4(\mu _1)}}}+{\displaystyle _{\mu _2^0}^{\mu _2}}{\displaystyle \frac{d\mu _2}{(\mu _2a)\sqrt{P_4(\mu _2)}}}=2t+\theta _1^0,`$ (5.5)
$`\theta _2`$ $`=`$ $`{\displaystyle _{\mu _1^0}^{\mu _1}}{\displaystyle \frac{d\mu _1}{\sqrt{P_4(\mu _1)}}}+{\displaystyle _{\mu _2^0}^{\mu _2}}{\displaystyle \frac{d\mu _2}{\sqrt{P_4(\mu _2)}}}=2x2at+\theta _2^0,`$ (5.6)
where
$$P_4(E)=(Em_3)(Em_4)(Em_5)(Em_6).$$
(5.7)
(For details about inverting problems of this type see Alber and Fedorov(1999).)
### 5.3 Two-Soliton Solutions of Kaup Type
The two-soliton solution is obtained by piecewise pinching together two pairs of elements of the spectrum so that $`m_1m_2a_1`$ and $`m_3m_4a_2`$. Here there are two double points $`a_1`$ and $`a_2`$, as well as two remaining hyperelliptic points at $`m_5`$ and $`m_6`$. Choosing the initial data on the positive branch of the Riemann surface for both $`\mu `$ variables leads to the following problem of inversion
$`\theta _1`$ $`=`$ $`{\displaystyle _{\mu _1^0}^{\mu _1}}{\displaystyle \frac{d\mu _1}{(\mu _1a_2)\sqrt{(\mu _1m_5)(\mu _1m_6)}}}`$ (5.8)
$`+{\displaystyle _{\mu _2^0}^{\mu _2}}{\displaystyle \frac{d\mu _2}{(\mu _2a_2)\sqrt{(\mu _2m_5)(\mu _2m_6)}}}=2x2a_1t+\theta _1^0,`$
$`\theta _2`$ $`=`$ $`{\displaystyle _{\mu _1^0}^{\mu _1}}{\displaystyle \frac{d\mu _1}{(\mu _1a_1)\sqrt{(\mu _1m_5)(\mu _1m_6)}}}`$ (5.9)
$`+{\displaystyle _{\mu _2^0}^{\mu _2}}{\displaystyle \frac{d\mu _2}{(\mu _2a_1)\sqrt{(\mu _2m_5)(\mu _2m_6)}}}=2x2a_2t+\theta _2^0.`$
This angle representation is similar to the one found in the case of the defocusing NLS equation. (See Alber and Marsden(1994).)
Notice that the $`\theta _i`$โs are essentially the sum of two Kaup type solitons as $`t\pm \mathrm{}`$. The integrals in system (5.8),(5.9) may be evaluated to obtain the following nonlinear algebraic system of equations for the $`\mu _i`$:
$`(s_{1625}+s_{1526})(s_{2625}+s_{2526})`$ $`=`$ $`A_1(s_{1625}s_{1526})(s_{2625}s_{2526}),`$ (5.10)
$`(s_{1615}+s_{1516})(s_{2615}+s_{2516})`$ $`=`$ $`A_2(s_{1615}s_{1516})(s_{2615}s_{1516}),`$ (5.11)
where $`s_{ijkl}=\sqrt{(\mu _im_j)(a_km_l)}`$ and
$`A_1`$ $`=`$ $`\mathrm{exp}[\sqrt{(a_2m_5)(a_2m_6)}(2x2a_1t+\theta _1^0)],`$ (5.12)
$`A_2`$ $`=`$ $`\mathrm{exp}[\sqrt{(a_1m_5)(a_1m_6)}(2x2a_2t+\theta _2^0)].`$ (5.13)
### 5.4 Phase Shift Formulas
When two solitons interact, they normally re-emerge with their initial profile and velocity. However, they have shifted ahead or behind where they would have been had there been no interaction at all. The amount a soliton shifts is called its phase shift, and the integral equations (5.8),(5.9) can be used to compute it. Assume that $`a_1>a_2`$ and define
$$M(x)=\sqrt{\frac{xm_5}{xm_6}}.$$
(5.14)
The phase shifts for the systems (5.8)-(5.9) are calculated by using the following procedure. First, consider the reference frame where $`\theta _1`$ is a constant, that is $`2x2a_1t+\theta _1^0=\alpha _1`$ for some constant $`\alpha _1`$. Then observe that $`\theta _2`$ can be written only as a function of $`t`$ and $`\alpha _1`$. Namely, $`\theta _2=\alpha _1+2(a_1a_2)t+\theta _2^0\theta _1^0`$. Notice that the integrals on the left hand side of (5.8) are exactly the expression for single Kaup type solitons. Integrals of this form may be integrated as
$$_{\mu _2^0}^{\mu _2}\frac{d\mu _2}{2(\mu _2a_1)\sqrt{(\mu _2m_5)(\mu _2m_6)}}=\frac{1}{2\sqrt{(a_1m_5)(a_1m_6)}}\mathrm{log}\left|\frac{\psi M(a_1)}{\psi +M(a_1)}\right|,$$
(5.15)
where for clarity we have defined $`\psi ^2=M(\mu _2)^2`$. Then we see that in this frame when $`t\mathrm{}`$ the right side of (5.9) goes to infinity so that the left hand side must also grow without bound. On evaluation of the integrals we see that this can only happen when $`\psi M(a_1)`$. Similarly when $`t\mathrm{}`$, $`\psi `$ must approach $`M(a_1)`$. Substituting these values for $`\mu _2`$ in (5.9) gives asymptotics for $`\theta _1`$ as $`t\mathrm{}`$ and as $`t\mathrm{}`$, respectively. The phase shift for one of the solitons will be the difference between the behavior of $`\theta _1`$ at minus infinity and its behavior at plus infinity in the frame given by $`\theta _1=\alpha _1`$. In this way we obtain the shift between phases before and after interaction for $`\theta _1`$ to be
$$\mathrm{\Delta }_1=\frac{1}{\sqrt{(a_2m_5)(a_2m_6)}}\mathrm{log}\left|\frac{M(a_2)+M(a_1)}{M(a_2)M(a_1)}\right|.$$
(5.16)
Similarly the phase shift for $`\mu _2`$ along $`\theta _2=\alpha _2`$ is computed to be
$$\mathrm{\Delta }_2=\frac{1}{\sqrt{(a_1m_5)(a_1m_6)}}\mathrm{log}\left|\frac{M(a_1)+M(a_2)}{M(a_1)M(a_2)}\right|.$$
(5.17)
These formulas were initially obtained by Matveev and Yavor(1979) by studying asymptotics of singular $`\theta `$-functions.
### 5.5 Kink-Anitkink Interaction Solutions
The interaction of kink and antikink solutions will now be considered. The two-kink solutions are constructed by taking the limit of the spectral parameters so that $`m_1m_2a_1`$, $`m_3m_4a_2`$, and $`m_5m_6a_3`$. We examine both the case when the $`\mu _i`$โs are both initially on the positive branch of the Riemann surface and the case when only one of the $`\mu _i`$โs is initially on the positive branch and the other is on the negative branch. In contrast to the KdV equation, this difference in the initial conditions produces qualitatively different solutions. This difference arises because the KdV and cKdV equations contain at least one hyperelliptic branch point, while all of the fixed points are double points for the two-kink solutions.
### 5.6 Initial values of $`\mu _1`$ and $`\mu _2`$ are chosen on the positive branches of the Riemann surface.
In this situation, the angle variables are computed to be
$`\theta _1`$ $`=`$ $`+{\displaystyle _{\mu _1^0}^{\mu _1}}{\displaystyle \frac{d\mu _1}{(\mu _1a_2)(\mu _1a_3)}}`$ (5.18)
$`+{\displaystyle _{\mu _2^0}^{\mu _2}}{\displaystyle \frac{d\mu _2}{(\mu _2a_2)(\mu _2a_3)}}=2x2a_1t+\theta _1^0,`$
$`\theta _2`$ $`=`$ $`+{\displaystyle _{\mu _1^0}^{\mu _1}}{\displaystyle \frac{d\mu _1}{(\mu _1a_1)(\mu _1a_3)}}`$ (5.19)
$`+{\displaystyle _{\mu _2^0}^{\mu _2}}{\displaystyle \frac{d\mu _2}{(\mu _2a_1)(\mu _2a_3)}}=2x2a_2t+\theta _2^0.`$
These integrals are tractable, and the symmetric polynomials $`\mu _1\mu _2`$ and $`\mu _1+\mu _2`$ which appear in the trace formulas, can be calculated explicitly from the resulting system
$`(1A_1)\mu _1\mu _2+(a_3A_1a_2)(\mu _1+\mu _2)`$ $`=`$ $`A_1a_3^2a_2^2,`$ (5.20)
$`(1A_2)\mu _1\mu _2+(a_3A_2a_1)(\mu _1+\mu _2)`$ $`=`$ $`A_2a_3^2a_1^2,`$ (5.21)
with
$`A_1={\displaystyle \frac{(\mu _1^0a_2)(\mu _2^0a_2)}{(\mu _1^0a_3)(\mu _2^0a_3)}}e^{2(a_2a_3)(xa_1t+\theta _1^0)},`$ (5.22)
$`A_2={\displaystyle \frac{(\mu _1^0a_1)(\mu _2^0a_1)}{(\mu _1^0a_3)(\mu _2^0a_3)}}e^{2(a_1a_3)(xa_2t+\theta _1^0)}.`$ (5.23)
These calculations are carried out for the unperturbed Boussinesq equations ($`K_1=0`$) in the next section and soliton fusion and fission are discussed.
### 5.7 Soliton Fusion and Fission.
Given this particular deformation of the Riemann surface, solitons can undergo fission or fusion. This interesting phenomenon occurs when two separate solitons enter an interaction but only one single soliton emerges from this interaction. Since all of the equations under study are invariant under the space-time inversion $`(xx,tt)`$, the reverse process of soliton fission may also occur where a single soliton breaks into two distinct solitons at some critical time. Observe that soliton fission can be interpreted as an infinite phase shift. This is because the solitons change their speed since when they fuse together. Therefore they will be infinitely far from where they would have been had there been no interaction. In this sense formulas (5.16)-(5.17) are still correct since in the limit $`m_5=m_6=a_3`$, $`M(x)=1`$ and so the formulas become singular.
For this discussion we assume that $`\gamma =2_{i=1}^{2n+2}m_i=0`$ although the general case is similar. Hence the system (5.20)-(5.21) can be solved for $`\mu _1+\mu _2`$ and is found to be
$$U=4(\mu _1+\mu _2)=4\frac{a_1f_1+a_2f_2+a_3f_3}{f_1+f_2+f_3},$$
(5.24)
where
$`f_1`$ $`=`$ $`\mathrm{exp}(2a_1x+2a_1^2t+\stackrel{~}{\theta _1}),`$ (5.25)
$`f_2`$ $`=`$ $`\mathrm{exp}(2a_2x+2a_2^2t+\stackrel{~}{\theta _2}),`$ (5.26)
$`f_3`$ $`=`$ $`\mathrm{exp}(2a_3x+2a_3^2t+\stackrel{~}{\theta _3}),`$ (5.27)
and
$`\stackrel{~}{\theta _1}`$ $`=`$ $`a_1\theta _1^0+\mathrm{log}[(a_2a_3)(\mu _1^0a_1)(\mu _2^0a_1)],`$ (5.28)
$`\stackrel{~}{\theta _2}`$ $`=`$ $`a_2\theta _1^0+\mathrm{log}[(a_3a_1)(\mu _1^0a_2)(\mu _2^0a_2)],`$ (5.29)
$`\stackrel{~}{\theta _3}`$ $`=`$ $`a_3\theta _2^0+\mathrm{log}[(a_1a_2)(\mu _1^0a_3)(\mu _2^0a_3)].`$ (5.30)
The expression (5.24) coincides with a solution to Burgerโs equation which describes a confluence of shocks (see Whitham(1974)) and has the same form as a solution obtained by using the Hirota method. This expression is now analyzed to see why it represents soliton fusion. Suppose $`a_1<0<a_2<a_3`$ and consider $`t=0`$. We claim that at this instant (5.24) is a two-tiered kink (sum of two kinks), see Figure B.8. To see why this is the case, consider $`2x<(\stackrel{~}{\theta _2}\stackrel{~}{\theta _1})/(a_1a_2)`$. For these values of $`x`$, $`f_1`$ is the largest of the terms $`f_1`$, $`f_2`$, and $`f_3`$. In fact as $`x\mathrm{}`$, the terms $`f_2`$,$`f_3`$ are negligible in comparison to $`f_1`$. So for these values of $`x`$, $`\mu _1+\mu _2`$ is nearly constant and is approximately $`a_1`$. Similarly for $`(\stackrel{~}{\theta _2}\stackrel{~}{\theta _1})/(a_1a_2)<2x<(\stackrel{~}{\theta _3}\stackrel{~}{\theta _2})/(a_2a_3)`$, $`f_2`$ is the dominant term and $`\mu _1+\mu _2a_2`$. For the remaining values of $`x`$, $`f_3`$ dominates so $`\mu _1+\mu _2a_3`$.
We now apply this analysis for arbitrary $`t`$. In this manner we get that $`U`$ is essentially constant in three regions, $`D_1`$, $`D_2`$ and $`D_3`$, of the $`(x,t)`$ plane as seen in Figure B.7. When $`t`$ is sufficiently small, these regions are bounded by the points where the functions $`f_1=f_2`$ and where $`f_2=f_3`$. At
$$t^{}=\frac{(a_2a_3)(\stackrel{~}{\theta _2}\stackrel{~}{\theta _1})(a_1a_2)(\stackrel{~}{\theta _3}\stackrel{~}{\theta _2})}{2(a_1a_2)(a_2a_3)(a_1a_3)},$$
(5.31)
the functions $`f_1=f_2=f_3`$ for some $`x^{}`$. From this instant on, $`f_2`$ is never the largest term and so for $`t>t^{}`$, the plane is divided into two regions bounded by the points where $`f_1=f_3`$ as seen in Figure B.7. For more details on this analysis see Whithamโs(1974) chapter on Burgerโs equation. As explained above, the regions $`D_i`$, which contain all the information regarding fission of the two-kink solution (times, speeds etc.), are derived from the parameterization of the lines $`f_1=f_2`$, $`f_2=f_3`$, and $`f_1=f_3`$. These lines are computed explicitly giving that $`f_1=f_2`$ along the line where $`2x=2a_3t+(\stackrel{~}{\theta _2}\stackrel{~}{\theta _1})/(a_1a_2)`$, $`f_2=f_3`$ along the line where $`2x=2a_1t+(\stackrel{~}{\theta _3}\stackrel{~}{\theta _2})/(a_2a_3)`$ and $`f_1=f_3`$ along the lines where $`2x=2a_2t+(\stackrel{~}{\theta _3}\stackrel{~}{\theta _1})/(a_1a_3)`$. From this it is possible to predict the times at which two solitons experience fission or fusion based solely on the initial phase and the values of the three spectrum points $`a_i`$. The speeds of the solitons are given by the slopes of the lines, and in this way a complete explanation of fission is given. Notice that the arguments are quite general and that this approach can be applied directly to the entire class of $`N`$-component systems.
### 5.8 Initial values of $`\mu _1`$ and $`\mu _2`$ are chosen on different branches of the Riemann surface.
This case gives a similar problem of inversion as the fusion case, namely
$`\theta _1`$ $`=`$ $`+{\displaystyle _{\mu _1^0}^{\mu _1}}{\displaystyle \frac{d\mu _1}{(\mu _1a_2)(\mu _1a_3)}}`$ (5.32)
$`{\displaystyle _{\mu _2^0}^{\mu _2}}{\displaystyle \frac{d\mu _2}{(\mu _2a_2)(\mu _2a_3)}}=2x2a_1t+\theta _1^0,`$
$`\theta _2`$ $`=`$ $`+{\displaystyle _{\mu _1^0}^{\mu _1}}{\displaystyle \frac{d\mu _1}{(\mu _1a_1)(\mu _1a_3)}}`$ (5.33)
$`{\displaystyle _{\mu _2^0}^{\mu _2}}{\displaystyle \frac{d\mu _2}{(\mu _2a_1)(\mu _2a_3)}}=2x2a_2t+\theta _2^0.`$
These integrals may be evaluated in terms of the $`\mathrm{log}`$ function and give rise to the following algebraic equations for the $`\mu _i`$โs
$`{\displaystyle \frac{(\mu _1a_1)(\mu _2a_3)}{(\mu _1a_3)(\mu _2a_1)}}`$ $`=`$ $`{\displaystyle \frac{(\mu _1^0a_1)(\mu _2^0a_3)}{(\mu _1^0a_3)(\mu _2^0a_1)}}\mathrm{exp}[(a_1a_3)(2x2a_2t+\theta _2^0)],`$ (5.34)
$`{\displaystyle \frac{(\mu _1a_2)(\mu _2a_3)}{(\mu _1a_3)(\mu _2a_2)}}`$ $`=`$ $`{\displaystyle \frac{(\mu _1^0a_2)(\mu _2^0a_3)}{(\mu _1^0a_3)(\mu _2^0a_2)}}\mathrm{exp}[(a_2a_3)(2x2a_1t+\theta _1^0)].`$ (5.35)
This is a system from which the $`\mu _i`$โs can be found explicitly. The solutions are plotted in Figure B.9 and the corresponding $`U`$ and $`W`$ are graphed in Figure B.10. By initially choosing different branches of this particular Riemann surface we see that the solitons โchange formโ, i.e. a kink changes to an antikink and vice versa. An analytic explanation is given in the next section. This was first observed in Alonso and Rues(1992) who noticed this phenomenon asymptotically. Using the algebraic geometric construction the finite-time interactions can be analyzed as well.
### 5.9 Change of Form of Kinks to Antikinks.
The solutions obtained from (5.34) can be put in the following form
$`\mu _1`$ $`=`$ $`{\displaystyle \frac{a_1g_1a_2g_2a_3g_3}{g_1+g_2+g_3}},`$ (5.36)
$`\mu _2`$ $`=`$ $`{\displaystyle \frac{a_1h_1+a_2h_2+a_3h_3}{h_1+h_2+h_3}},`$ (5.37)
where
$`g_1`$ $`=`$ $`\mathrm{exp}\left[2a_1x2a_2a_3ta_1\theta _2^0+\mathrm{log}{\displaystyle \frac{(a_3a_2)(\mu _1^0a_1)}{\mu _2^0a_1}}\right],`$ (5.38)
$`g_2`$ $`=`$ $`\mathrm{exp}\left[2a_2x2a_1a_3ta_2\theta _1^0+\mathrm{log}{\displaystyle \frac{(a_1a_3)(\mu _1^0a_2)}{\mu _2^0a_2}}\right],`$ (5.39)
$`g_3`$ $`=`$ $`\mathrm{exp}\left[2a_3x2a_1a_2ta_3(\theta _1^0+\theta _2^0)+\mathrm{log}{\displaystyle \frac{(a_2a_1)(\mu _1^0a_3)}{\mu _2^0a_3}}\right],`$ (5.40)
$`h_1`$ $`=`$ $`\mathrm{exp}\left[2a_1x+2a_2a_3t+a_1\theta _2^0+\mathrm{log}{\displaystyle \frac{(a_3a_2)(\mu _1^0a_1)}{\mu _2^0a_1}}\right],`$ (5.41)
$`h_2`$ $`=`$ $`\mathrm{exp}\left[2a_2x+2a_1a_3t+a_2\theta _1^0+\mathrm{log}{\displaystyle \frac{(a_1a_3)(\mu _1^0a_2)}{\mu _2^0a_2}}\right],`$ (5.42)
$`h_3`$ $`=`$ $`\mathrm{exp}\left[2a_3x+2a_1a_2t+a_3(\theta _1^0+\theta _2^0)+\mathrm{log}{\displaystyle \frac{(a_2a_1)(\mu _1^0a_3)}{\mu _2^0a_3}}\right].`$ (5.43)
Notice the similarity between the form of the solutions (5.36),(5.37) and the solution (5.24) for $`U`$ in the previous case. Since they have an identical form as that in (5.24), all the analysis from the last section applies and we conclude that both $`\mu _1`$ and $`\mu _2`$ are kinks which experience fission or fusion respectively. We find that as $`t\mathrm{}`$, $`\mu _1`$ is an antikink and $`\mu _2`$ decomposes into two kinks. This implies that $`\frac{1}{4}U=\mu _1+\mu _2`$ consists of two kinks and one antikink. As $`t\mathrm{}`$, $`\mu _1`$ fissions into two antikinks and the 2 kinks comprising $`\mu _2`$ fuse into one kink. Therefore $`\mu _1+\mu _2`$ is the sum of one kink and two antikinks. This explains the transformation of kinks to antikinks and vice versa. See Figure B.11 to see how these $`\mu `$ variables combine to form $`U`$. Of course the same analysis can be performed as above to see this analytically. This is the first time two-soliton solutions of this equation have been derived in this simple form.
## 6 The SHG Equations
Solutions of the cKdV may also be transformed into solutions of the SHG equation if $`\kappa `$ is chosen to be $`1`$. This is shown explicitly in Appendix A. The results from the previous sections can be viewed in the context of the SHG equations. First, if $`\kappa =1`$, the initial conditions must be chosen differently or the $`\mu _i`$โs will not be real valued. That is, the cases $`\kappa =1`$ and $`\kappa =1`$ are dual to each other in the sense that in the process of the construction of the Riemann surface the cuts in the complex plane for the $`\kappa =1`$ case correspond to where real valued solutions lie in the $`\kappa =1`$ case and vice versa. This is because the kappa appears under the square root in (3.7) so that real values of $`\mu `$ for $`kappa=1`$ correspond exactly to imaginary values of $`\mu `$ when $`\kappa =1`$ and vice versa.
Another important difference is that no kink solutions exist for the SHG equations. When $`\kappa =1`$ there is no way to deform the Riemann surface so that all points coalesce piecewise as required for kink solitons. Therefore there are always at least two hyperelliptic points left. This means that SHG solutions do not have a possibility of either fusing or fissioning and no change of form can occur.
The phase shift formulas for this system are very similar to those in the cKdV hierarchy with $`\kappa =1`$, namely
$`\mathrm{\Delta }_1`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{(a_2m_5)(a_2m_6)}}}\mathrm{log}\left|{\displaystyle \frac{M(a_2)+M(a_1)}{M(a_2)M(a_1)}}\right|,`$ (6.1)
$`\mathrm{\Delta }_2`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{(a_1m_5)(a_1m_6)}}}\mathrm{log}\left|{\displaystyle \frac{M(a_1)+M(a_2)}{M(a_1)M(a_2)}}\right|.`$ (6.2)
This is the first time these formulas have been derived. It is remarkable that the formulas are so similar to those of the coupled KdV equation since they are not in the same hierarchy of equations - the two equations are derived from two different potentials. The phase shift formulas show another strength of the algebraic geometric method since the details of the inverse scattering transform have not been completed at this time, see Khusnutdinova(1998). $`u`$ and $`v`$ are plotted in Figure B.12. These functions can be transformed into $`q_1,q_2`$ of the SHG by transformations in Appendix A.
## 7 Modified Coupled Dym Equations
### 7.1 Genus-1 solutions for the cDym System
#### 7.1.1 Periodic solutions.
A periodic solution of the cDym equation is described by the following differential equation
$$\frac{\sqrt{\mu }d\mu }{\sqrt{_{i=1}^4(\mu m_i)}}=dX,$$
(7.1)
for particular choice of $`m_i`$โs. Here $`X=2x2ct+\theta _0`$ and integration is carried out on the Riemann surface $`W^2={\displaystyle \frac{C(E)}{E}}`$. This can be reduced to a standard form by introducing a new variable $`Y`$ by
$$dY=\frac{dX}{\sqrt{\mu }}.$$
(7.2)
After integration (7.1) becomes
$$_{\mu _0}^\mu \frac{d\mu }{\sqrt{_{i=1}^4(\mu m_i)}}=Y.$$
(7.3)
Notice that this holomorphic differential is defined on a genus two Riemann surface. To invert this integral one has first to consider the following problem of inversion
$`{\displaystyle _{\mu _1^0}^{\mu _1}}{\displaystyle \frac{d\mu _1}{\sqrt{\mu _1C_4(\mu _1)}}}+{\displaystyle _{\mu _2^0}^{\mu _2}}{\displaystyle \frac{d\mu _2}{\sqrt{\mu _2C_4(\mu _2)}}}`$ $`=`$ $`\theta _1^0,`$ (7.4)
$`{\displaystyle _{\mu _1^0}^{\mu _1}}{\displaystyle \frac{\mu _1d\mu _1}{\sqrt{\mu _1C_4(\mu _1)}}}+{\displaystyle _{\mu _2^0}^{\mu _2}}{\displaystyle \frac{\mu _2d\mu _2}{\sqrt{\mu _2C_4(\mu _2)}}}`$ $`=`$ $`X_1+\theta _2^0,`$ (7.5)
where $`C_4(\mu )=_{l=1}^4(\mu m_j)`$. One needs to rearrange integrals in such a way that to obtain $`X_1`$ on the right hand side of the first integral equation. This yields exact formulas for $`\mu _1`$ and $`\mu _2`$ in terms of Riemann $`\theta `$-functions. By fixing $`\mu _2=m_3`$ and writing $`\mu =\mu _1`$ we resolve the initial problem of inversion. (For details see Alber and Fedorov(1999).)
#### 7.1.2 Kink solutions.
Now consider a kink limit by setting $`m_1,m_2a_1`$ and $`m_3,m_4a_2`$ such that $`a_2>a_1>0`$. Integral (7.3) becomes
$$\frac{d\mu }{(\mu a_1)(\mu a_2)}=dY.$$
(7.6)
Observe that this is the same problem of inversion as in the cKdV case except that we have $`Y`$ instead of $`X`$ on the right hand side. After integrating we obtain
$$\mu (Y)=\frac{a_1(\mu _0a_2)a_2(\mu _0a_1)\mathrm{exp}((a_1a_2)Y)}{(\mu _0a_2)(\mu _0a_1)\mathrm{exp}((a_1a_2)Y)}.$$
(7.7)
This gives $`\mu `$ as a function of $`Y`$ and as in the cKdV case this is a kink. $`X`$ is defined in terms of $`Y`$ by (7.2) once we know $`\mu (Y)`$ from (7.7). The integration in (7.2) may be carried out explicitly. Note that $`dX/dY>0`$ and so $`X(Y)`$ is always increasing. By the definition of $`X`$ it is also clear that the range of $`X(Y)`$ is all real numbers. Therefore the Inverse Function Theorem implies that an inverse function $`Y=Y(X)`$ exists for all values of $`X`$ and is monotonically decreasing. Therefore the graph of $`\mu (X)=\mu (Y(X))`$ will have a similar appearance to that of $`\mu (Y)`$, that is, it is also a kink.
After combining numerics for $`Y(X)`$ with the expression for $`\mu `$, a description for the kink of the cDym system is obtained, see Figure B.13.
#### 7.1.3 Cusp solution
For this solution, the limit $`m_1,m_2a_1`$ and $`m_3,m_4a_2`$ such that $`a_2<0<a_1`$ is analyzed. The analysis is the same as in the kink case except that now $`dX/dY`$ changes sign exactly once when the branch point $`Y^{}`$ is crossed where $`\mu (Y^{})=0`$. In this case, $`Y(X)`$ has two branches begining at the hyperelliptic point $`X^{}`$ where $`Y(X^{})=Y^{}`$. Therefore $`\mu (X)`$ has two branches and reaches a cusp at the point $`X^{}`$. See Figure B.14.
#### 7.1.4 Peakon solution
If the Camassa-Holm shallow water equation is an indication(1994), a peakon may develop in the limit $`m_1,m_2a_1`$ and $`m_3,m_4a_2`$ with $`a_2=0<a_1`$. The analysis is similar to that of the kink case, except now the range of $`X(Y)`$ is bounded above by some number $`X^{}`$. This means that the inverse function $`Y(X)`$ is defined only for those $`X<X^{}`$. But, it can be defined symmetrically, as if the integration were carried out on the negative branch of the square root, and this gives rise to a peakon solution. The difference between this and a cusp solution is that in the cusp,
$$\frac{dY}{dX}=\frac{1}{\sqrt{\mu }},$$
(7.8)
is infinite at the branch point, while in the Peakon case $`\mu (Y)>0`$ for all $`Y`$ and so at the branch point the derivative is finite.
### 7.2 Genus 2 solutions
The algebraic geometric procedure outlined thus far in this paper can be used for other equations as well, even when other methods may fail. (For details see Alber and Fedorov (1999).) Using our experience with the cKdV system, the case when there are three double points will be considered. For the positive branch of the Riemann Surface $`W^2=C(E)/E`$, the problem of inversion can be written as
$`{\displaystyle \frac{\mu _1d\mu _1}{2(\mu _1a_1)(\mu _1a_2)(\mu _1a_3)\sqrt{\mu _1}}}+{\displaystyle \frac{\mu _2d\mu _2}{2(\mu _2a_1)(\mu _2a_2)(\mu _2a_3)\sqrt{\mu _2}}}`$ $`=`$ $`dt,`$ (7.9)
$`{\displaystyle \frac{\mu _1^2d\mu _1}{2(\mu _1a_1)(\mu _1a_2)(\mu _1a_3)\sqrt{\mu _1}}}+{\displaystyle \frac{\mu _2^2d\mu _2}{2(\mu _2a_1)(\mu _2a_2)(\mu _2a_3)\sqrt{\mu _2}}}`$ $`=`$ $`dx.`$ (7.10)
This inversion problem is very similar to that in the cKdV case, with the exception of the poles present in the left hand side of the equation. This case is complicated by the fact that only differentials of the third kind appear having simple poles at $`a_1,a_2,a_3`$. Following the procedure outlined in Alber and Fedorov(1999), a third variable $`y`$ is introduced such that
$$\frac{d\mu _1}{2(\mu _1a_1)(\mu _1a_2)(\mu _1a_3)\sqrt{\mu _1}}+\frac{d\mu _2}{2(\mu _2a_1)(\mu _2a_2)(\mu _2a_3)\sqrt{\mu _2}}=dy.$$
(7.11)
Also the normalized differentials of the third kind are introduced
$$\mathrm{\Omega }_i=\frac{\alpha _id\mu }{(\mu \alpha _i^2)\sqrt{\mu }},i=1,2,3,$$
(7.12)
where $`\alpha _i=\sqrt{a_i}`$. Next, consider three points $`z_i`$ given by
$$\underset{j=1}{\overset{3}{}}_{P_0}^{P_j}\mathrm{\Omega }_i=z_i,i=1,2,3.$$
(7.13)
The $`z_i`$ are dependent on $`x,t,y`$ by
$`z_1`$ $`=`$ $`2\sqrt{a_1}[x(a_2+a_3)t+a_2a_3y],`$ (7.14)
$`z_2`$ $`=`$ $`2\sqrt{a_2}[x(a_1+a_3)t+a_1a_3y],`$ (7.15)
$`z_3`$ $`=`$ $`2\sqrt{a_3}[x(a_1+a_2)t+a_1a_2y].`$ (7.16)
Integrating (7.13) and putting $`\xi =\sqrt{\mu }`$ gives
$$\frac{(\xi _1\alpha _i)(\xi _2\alpha _i)(\xi _3\alpha _i)}{(\xi _1+\alpha _i)(\xi _2+\alpha _i)(\xi _3+\alpha _i)}=e^{z_i},i=1,2,3.$$
(7.17)
This gives the following system for the symmetric polynomials in $`\xi `$.
$`\left(\begin{array}{ccc}1e^{z_1}& \alpha _1(1+e^{z_1})& \alpha _1^2(1e^{z_1})\\ 1e^{z_2}& \alpha _2(1+e^{z_2})& \alpha _2^2(1e^{z_2})\\ 1e^{z_3}& \alpha _3(1+e^{z_3})& \alpha _3^2(1e^{z_3})\end{array}\right)\left(\begin{array}{c}\xi _1\xi _2\xi _3\\ \xi _1\xi _2+\xi _2\xi _3+\xi _1\xi _3\\ \xi _1+\xi _2+\xi _3\end{array}\right)=\left(\begin{array}{c}\alpha _1^3(1+e^{z_1})\hfill \\ \alpha _2^3(1+e^{z_2})\hfill \\ \alpha _3^3(1+e^{z_3})\hfill \end{array}\right)`$ (7.27)
ยฟFrom this, the expression $`\mu _1+\mu _2+\mu _3=\xi _1^2+\xi _2^2+\xi _3^2`$ may be found as a function of $`z_1,z_2,z_3`$. The determinant of the matrix in the system must be zero to obtain nonzero solutions. This added equation gives $`\mu _3`$ in terms of $`\mu _1`$ and $`\mu _2`$. Then we can find $`\mu _1+\mu _2`$ as a function of $`z_1,z_2`$. Then this solution must be connected to the $`(x,t)`$ variables by using (7.14).
Numerics are provided in Figures B.15 and B.16. As expected, the phenomena of change of form and fission occurs. A more detailed analysis of fission/fusion for the cDym case will be described in a forth coming paper.
## 8 Acknowledgments
The research of Mark Alber and Gregory Luther was partially supported by NSF grant DMS 9626672. Mark Alber would like to thank Peter Miller for a helpful discussion and bringing to his attention references Estevez et al.(1994) and Martinez Alonso and Medina Reus(1992).
## Appendix A Transformations to Related Equations
In what follows we describe exact connections between solutions of the cKdV and the Kaup equations, Boussinesq systems and SHG system.
### A.1 Generalized Coupled KdV System
For this system, assume that $`\kappa =1`$ and $`r=0`$. Observe that if we take $`n=1`$, $`B`$ takes the form of $`B=E+b_1`$. Furthermore, as in (B.1) we see that $`b_1=u/2+K_1`$. Collecting coefficients of order 1 and 0 respectively in (2.9) and substituting the value of $`b_1`$ shows that $`u,v`$ satisfy the modified cKdV system
$`u_t`$ $`=`$ $`v^{}\frac{3}{2}uu^{}+K_1u^{},`$ (A.1)
$`v_t`$ $`=`$ $`\frac{1}{4}u^{\prime \prime \prime }vu^{}\frac{1}{2}uv^{}+K_1v^{}.`$ (A.2)
$`K_1`$ is a small parameter. When $`K_1=0`$ this is the cKdV.
### A.2 Generalized Boussinesq System
To connect $`u`$ and $`v`$ to the classical Boussinesq system the following change of variables must be made
$`u(x,t)`$ $`=`$ $`\frac{1}{2}U(X,T),`$ (A.3)
$`v(x,t)`$ $`=`$ $`\frac{1}{16}U(X,T)^2\frac{1}{4}W(X,T),`$ (A.4)
where $`X=x`$ and $`T=t/2`$. Plugging in $`u`$ and $`v`$ from (A.3)-(A.4) into (A.1)-(A.2) and defining $`\gamma =2K_1`$ we see that $`U`$ and $`W`$ satisfy
$`U_T+W_X+UU_X`$ $`=`$ $`\gamma U_X,`$ (A.5)
$`W_T+U_{XXX}+(WU)_X`$ $`=`$ $`\gamma W_X.`$ (A.6)
Again $`\gamma `$ is a small parameter. When $`\gamma =0`$, we have exactly the Boussinesq System. Otherwise the system gives rise to a generalization of the Boussinesq system. Such a transformation can be found in Sattinger(1995) for example.
### A.3 Generalized Kaup Equations
To connect $`U(X,T)`$ and $`W(X,T)`$ from above to $`\pi (x,t)`$ and $`\varphi (x,t)`$ from the Kaup equations (1.7),(1.8) the following change of variables was suggested by Kaup(1972):
$`U(X,T)`$ $`=`$ $`{\displaystyle \frac{ฯต}{\beta }}\varphi _x(x,t),`$ (A.7)
$`W(X,T)`$ $`=`$ $`\beta ^2[1ฯต\pi (x,t)],`$ (A.8)
where $`X=x`$, $`T=\beta t`$, $`\beta =\delta \sqrt{3}/\sqrt{13\sigma }`$, and $`\alpha =\beta \gamma `$. Then substituting this into (A.5)-(A.6) gives
$`\pi _t`$ $`=`$ $`\varphi _{xx}+\frac{1}{3}(13\sigma )\delta ^2\varphi _{xxxx}ฯต(\varphi _x\pi )_x+\alpha \pi _x,`$ (A.9)
$`\pi `$ $`=`$ $`\varphi _t+\frac{1}{2}ฯต\varphi _x^2\alpha \varphi _x,`$ (A.10)
which is a perturbed Kaup equations and reduces identically to it when $`\alpha =0`$. Summarizing we see that every solution of the coupled KdV system yields, using the transformations stated, a solution of the Boussinesq and Kaup systems.
### A.4 The SHG System
For the SHG system we choose $`\kappa =1`$ in the potential (2.6), and define $`w,\nu `$ by the equations $`u=w`$, and $`v=\nu _x/2+\nu ^2/4`$. Then the generating equation (3.1) becomes
$`Ew_t+\frac{1}{2}\nu _{xt}+\frac{1}{2}\nu \nu _t`$ $`=`$ $`\frac{1}{2}B_{xxx}+2B_xwE+B_x\nu _x+\frac{1}{2}B_x\nu ^2`$ (A.11)
$`2B_xE^2+Bw_xE+\frac{1}{2}B\nu _{xx}+\frac{1}{2}B\nu \nu _x.`$
Now we choose $`B=b/E`$, that is we choose $`m=1,n=1,b=b_2`$. Then (A.11) becomes
$`Ew_t+\frac{1}{2}\nu _{xt}+\frac{1}{2}\nu \nu _t`$ $`=`$ $`\frac{1}{2E}b_{xxx}+2b_xw+\frac{1}{E}b_x\nu _x+\frac{1}{2E}b_x\nu ^2`$ (A.12)
$`2b_xE+bw_x+\frac{1}{2E}b\nu _{xx}+\frac{1}{2E}b\nu \nu _x.`$
Collecting coefficients of $`E`$ gives
$$b=\frac{\eta }{2},$$
(A.13)
where we define $`\eta `$ by $`\eta _x=w_t`$. Then collecting coefficients of orders zero and one respectively and substituting in the value for $`b`$ gives
$`\frac{1}{2}\nu _{xt}+\frac{1}{2}\nu \nu _t`$ $`=`$ $`\eta _xw\frac{1}{2}\eta w_x`$ (A.14)
$`0`$ $`=`$ $`\frac{1}{4}\eta _{xxx}\frac{1}{2}\eta _x\nu _x\frac{1}{4}\eta _x\nu ^2\frac{1}{4}\eta \nu _{xx}\frac{1}{4}\eta \nu \nu _x.`$ (A.15)
(A.15) may be integrated to get
$$(\eta _x)_x=(\eta \nu )\nu +\eta \nu _x+\nu _x(\eta \nu \eta _x)๐x.$$
(A.16)
Notice that this equation is satisfied when $`\eta _x=\eta \nu `$. Next we define the new function $`s`$ by the relation $`\nu _t=s^1\eta w`$. Then plugging this into (A.14) gives
$$\frac{s_x}{s^2}\eta _xw\eta w_x+\frac{\nu }{s}\eta \nu w=2\eta _xw\eta w_x.$$
(A.17)
Substituting $`\eta _x=\eta \nu `$ and canceling like terms yields $`s_x=s\nu `$. These two equations, along with the two we defined will determine the SHG system. Summarizing, we have
$`\eta _x`$ $`=`$ $`\eta \nu =w_t,`$ (A.18)
$`\nu _t`$ $`=`$ $`s^1\eta w,`$ (A.19)
$`s_x`$ $`=`$ $`s\nu .`$ (A.20)
Next define $`Q=s^1`$ and $`\varphi _t=\eta `$. This and $`\eta _x=w_t`$ implies that $`w=\varphi _x`$. Therefore (A.18) becomes
$`(Q\varphi _t)_x`$ $`=`$ $`Q_x\varphi _t+Q\varphi _{xt},`$ (A.21)
$`=`$ $`{\displaystyle \frac{s_x}{s^2}}\eta +{\displaystyle \frac{\eta _x}{s}},`$
$`=`$ $`{\displaystyle \frac{1}{s}}(\eta _x\eta \nu )=0,`$
and
$`(\mathrm{ln}Q)_{xt}\varphi _x\varphi _t`$ $`=`$ $`(\mathrm{ln}s)_{xt}\eta w,`$ (A.22)
$`=`$ $`\left({\displaystyle \frac{s_x}{s}}\right)_t\eta w,`$
$`=`$ $`\nu _t\eta w,`$
$`=`$ $`(s^1\eta w)\eta w=Q.`$
Finally, substitute $`q_1=(\sqrt{Q}/2)\mathrm{exp}[i(\varphi /2)]`$ and the real and imaginary parts of the following equation correspond to (A.21) and (A.22), so that
$$q_{1xt}q_1^{}q_{1t}^{}q_{1x}=2(q_1q_1^{})^2.$$
(A.23)
If we define $`q_2=[(\mathrm{ln}Q)_x+i\varphi _x]\mathrm{exp}(i\varphi )/4`$. Then $`q_1`$ and $`q_2`$ satisfy
$`q_{1x}`$ $`=`$ $`2q_2q_1^{},`$ (A.24)
$`q_{2t}`$ $`=`$ $`q_1^2,`$ (A.25)
which is exactly the SHG equation, and $`q_1`$ and $`q_2`$ are obtained from the $`\mu `$ variables through the relation
$`\varphi _x`$ $`=`$ $`u,`$ (A.26)
$`Q`$ $`=`$ $`\mathrm{exp}\left({\displaystyle \frac{u_t}{u_t๐x}}\right).`$ (A.27)
The above transformations were inspired by Khusnutdinova(1998). Summarizing the above gives that solutions of the SHG system may be expressed as follows
$`q_1`$ $`=`$ $`{\displaystyle \frac{1}{2}}\mathrm{exp}\left({\displaystyle \frac{1}{2}}{\displaystyle u_t\left(u_t๐x\right)^1๐x}\right)\mathrm{exp}\left({\displaystyle \frac{i}{2}}{\displaystyle u๐x}\right),`$ (A.28)
$`q_2`$ $`=`$ $`{\displaystyle \frac{1}{4}}\left({\displaystyle \frac{d^2}{dx^2}}\mathrm{log}\left({\displaystyle u_t๐x}\right)iu\right)\mathrm{exp}\left(i{\displaystyle u๐x}\right).`$ (A.29)
Notice that the dependence on $`v`$ is implicit by the fact that $`\nu =u_t\left(u_t๐x\right)^1`$.
## Appendix B Trace Formulas
### B.1 cKdV System
In this section, the connection between $`u,v,`$ and $`\mu `$ for cKdV is derived. This connection is called the trace formula for this system. Collect coefficients of $`E^{nr+2}`$ in (3.1) to see that $`2\kappa b_0^{}=0`$ from which we assume that $`b_0=1`$. Now gather coefficients of order $`nr+1`$ to arrive at
$$b_1=\frac{u}{2\kappa }+K_1,$$
(B.1)
for some constant of integration $`K_1`$. Next collect coefficients of order $`2n2r+1`$ in (3.2) to see, along with (B.1), that
$$K_1=\frac{1}{2}\underset{i=1}{\overset{2n+2}{}}m_i.$$
(B.2)
(B.1) then yields the trace formula for $`u`$,
$$u=2\kappa \underset{i=1}{\overset{n}{}}\mu _i\kappa \underset{i=1}{\overset{2n+2}{}}m_i.$$
(B.3)
Now we just need the trace formula for $`v`$. For this we collect coefficients of order $`nr`$ in (3.1) to get
$$0=2\kappa b_2^{}+2b_1^{}u+b_1u^{}+v^{},$$
(B.4)
which gives upon substitution of $`b_1`$ from (B.1) that
$$v^{}=2\kappa b_2^{}+\frac{3uu^{}}{2\kappa }K_1u^{},$$
(B.5)
or simply
$$v=2\kappa b_2+\frac{3u^2}{4\kappa }K_1u+K_2,$$
(B.6)
where $`K_2`$ is some constant and $`K_1`$ is the same as above. From the definition of $`B(E)`$ we see that
$$b_2=\underset{1i<jn}{}\mu _i\mu _j.$$
(B.7)
All that remains is to find $`K_2`$. To derive this we collect coefficients of order $`2n2r`$ in (3.2). To simplify calculations let $`c_n`$ be the $`(2n2r)^{th}`$ coefficient of the polynomial $`C(E)`$. Then we see that
$$2\kappa (2b_2+b_1^2)+2u(2b_1)+2v=c_n,$$
(B.8)
or after substitution of the value of $`b_1`$ that
$$4\kappa b_2\frac{3u^2}{2\kappa }+2uK_1+2\kappa K_1^2+2v=c_n.$$
(B.9)
Now solving for $`K_2`$ in (B.6) and substituting in $`v`$ from (B.9),
$`K_2`$ $`=`$ $`v+2\kappa b_2{\displaystyle \frac{3u^2}{4\kappa }}+K_1u,`$ (B.10)
$`=`$ $`\left({\displaystyle \frac{c_2}{2}}2\kappa b_2+{\displaystyle \frac{3u^2}{4\kappa }}uK_1\kappa K_1^2\right)+2\kappa b_2{\displaystyle \frac{3u^2}{4\kappa }}+K_1u,`$ (B.11)
$`=`$ $`{\displaystyle \frac{c_2}{2}}\kappa K_1^2.`$ (B.12)
Now from the definition of $`C(E)`$ we see that
$$c_n=2\kappa \underset{1i<j2n+2}{}m_im_j.$$
(B.13)
### B.2 cDym System
In an analogous manner, we derive the trace formulas for the cDym system. The derivation of the trace formula for $`u`$ is identical to the cKdV case so that
$$u=2\underset{i=1}{\overset{n}{}}\mu _i\underset{i=1}{\overset{2n+2}{}}m_i.$$
(B.14)
Only the trace formula for $`v`$ is different. For this we collect coefficients of order $`nr`$ in (3.3) to get
$$0=\frac{b_1^{\prime \prime \prime }}{2}+2b_2^{}+2b_1^{}u+b_1u^{}+v^{},$$
(B.15)
which, upon substitution of $`b_1`$ from (B.1) and integrating, gives
$$v=\frac{u^{\prime \prime }}{4}2b_2+\frac{3u^2}{4}K_1u+K_2,$$
(B.16)
where $`K_2`$ is some constant and $`K_1`$ is the same as in cKdV. From the definition of $`B(E)`$ we see that
$$b_2=\underset{1i<jn}{}\mu _i\mu _j.$$
(B.17)
All that remains is to find $`K_2`$. To derive this we collect coefficients of order $`2n2r1`$ in (3.4). To simplify calculations let $`c_{n1}`$ be the $`(2n2r1)^{st}`$ coefficient of the polynomial $`C(E)`$. Then we see that
$$b_1^{\prime \prime }+2(2b_2+b_1^2)+2u(2b_1)+2v=c_{n1},$$
(B.18)
or after substitution of the value of $`b_1`$ that
$$\frac{u^{\prime \prime }}{2}+4b_2\frac{3u^2}{2}+2uK_1+2K_1^2+2v=c_{n1}.$$
(B.19)
Now solving for $`K_2`$ in (B.16) and substituting in $`v`$ from (B.19) gives
$`K_2`$ $`=`$ $`v+2b_2{\displaystyle \frac{3u^2}{4}}+K_1u+{\displaystyle \frac{u^{\prime \prime }}{4}},`$ (B.20)
$`=`$ $`({\displaystyle \frac{c_2}{2}}2b_2+{\displaystyle \frac{3u^2}{4}}uK_1K_1^2{\displaystyle \frac{u^{\prime \prime }}{4}})+2b_2{\displaystyle \frac{3u^2}{4}}+K_1u+{\displaystyle \frac{u^{\prime \prime }}{4}},`$ (B.21)
$`=`$ $`{\displaystyle \frac{c_{n1}}{2}}K_1^2.`$ (B.22)
ยฟFrom the definition of $`C(E)`$ we see that
$$c_{n1}=2\underset{1i<j2n+2}{}m_im_j.$$
(B.23)
## Bibliography
D. Mumford, Tata Lectures on Theta I and II, Progress in Math.28 and 43, (Birkhauser, Boston 1983).
Ercolani, N. and H. McKean, โGeometry of KdV(4). Abel sums, Jacobi variety, and theta function in the scattering case,โ Invent. Math 99, 483-544 (1990).
Ablowitz, M.J. and Y-C. Ma, โThe Periodic Cubic Schrลdinger Equation,โ Studies in Appl. Math. 65, 113โ158 (1981).
Alber, M.S. and S.J. Alber, โ Hamiltonian formalism for finite-zone solutions of integrable equations,โ C. R. Acad. Sci. Paris Sr. I Math. 301, 777โ781 (1985).
Alber, M.S. and J.E. Marsden, โOn Geometric Phases for Soliton Equations,โ Comm. Math. Phys. 149, 217โ240 (1992).
Antonowicz, M. and A.P. Fordy, โA family of completely integrable multi-Hamiltonian systemsโ, Phys. Lett. A 122, 95โ99 (1987).
Antonowicz, M. and A.P. Fordy, โCoupled KdV equations with multi-Hamiltonian structures,โ Physica D 28, 345โ357 (1987).
Antonowicz, M. and A.P. Fordy, โ Coupled Harry Dym equations with multi-Hamiltonian structures,โ J. Phys. A 21, L269โL275 (1988).
Antonowicz, M. and A.P. Fordy, โ Factorization of energy dependent Schrลdinger operators: Miura maps and modified systems,โ Comm. Math. Phys. 124, 465โ486 (1989).
Alber, M.S., G.G. Luther, and J.E. Marsden, โ Energy Dependent Schrลdinger Operators and Complex Hamiltonian Systems on Riemann Surfaces,โ Nonlinearity 10, 223โ242 (1997).
Channell, P.J. and C. Scovel, โSymplectic integration of Hamiltonian systems,โ Nonlinearity 3, 231โ259 (1990).
Channell, P.J. and C. Scovel, โAn introduction to symplectic integrators,โ Fields Institute Communications 10, 45โ58 (1996).
Alber, M.S., R. Camassa, D.D. Holm and J.E. Marsden, โThe geometry of peaked solitons and billiard solutions of a class of integrable PDEโs,โ Lett. Math. Phys. 32, 137โ151 (1994).
Alber, M.S., R. Camassa, D.D. Holm and J.E. Marsden, โOn Umbilic Geodesics and Soliton Solutions of Nonlinear PDEโs,โ Proc. Roy. Soc. London Ser. A 450, 677โ692 (1995).
Alber, M.S., R. Camassa, Yu.N. Fedorov, D.D. Holm and J.E. Marsden, โThe geometry of new classes of weak billiard solutions of nonlinear PDEโs,โ (preprint) (1999).
Alber, M.S. and Yu.N. Fedorov, โAlgebraic Geometric Solutions for Nonlinear Evolution Equations and Flows on the Nonlinear Subvarieties of Jacobians,โ (preprint) (1999)
Whitham, G.B., Linear and Nonlinear Waves, (Pure and applied mathematics, John Wiley & Sons, Inc. 1974).
Jaulent, M., โOn an inverse scattering problem with an energy dependent potential,โ Ann. Inst. H. Poincare A 17, 363โ372 (1972).
Jaulent, M. and C. Jean, โThe inverse problem for the one-dimensional Schลdinger operator with an energy dependent potential,โ Ann. Inst. H. Poincare A I, II25, 105โ118, 119โ137 (1976).
Kaup, D.J., โA Higher-Order Water-Wave Equation and the Method for Solving It,โ Prog. Theor. Phys. 54, 72โ78, 396โ408 (1975).
Matveev, V.B. and M.I. Yavor, โSolutions presque periodiques et a $`N`$-solitons de lรฉquation hydrodynamique non lineaire de Kaup,โ Ann. Inst. Henri Poincare: Sec. A 31, 25โ41 (1979).
Sachs, R.L., โOn the integrable variant of the Boussinesq system: Painlevรฉ property, rational solutions, a related many-body system, and equivalence with the AKNS hierarchy,โ Physica D 30, 1โ27 (1998).
Martinez Alonso, L. and E. Medina Reus, โSoliton interaction with change form in the classical Boussinesq system,โ Phys. Lett. A 167, 370โ376 (1992).
Estevez, P.G., P.R. Gordoa, L. Martinez Alonso, and E. Medina Reus , โOn the characterization of a new soliton sector in the classical Boussinesq system,โ Inverse Problems 10, L23โL27 (1994).
Khusnutdinova, K.R. and H. Steudel, โSecond harmonic generation: Hamiltonian structures and particular solutions,โ J. Math. Phys. 39, 3754โ3764 (1998).
Sattinger, David and Szmigielski, Jacek, โEnergy dependent scattering theory,โ Differential and Integral Equations 8, 945โ959 (1995).
Sattinger, David and Szmigielski, Jacek, โA Riemann Hilbert problem for an energy dependent Schrลdinger operator,โ Inverse Problems 12, 1003โ1025 (1996).
Ablowitz, M.J. and Segur, H. , Solitons and the Inverse Scattering Transform, (SIAM, Philadelphia, 1981).
Belokolos, E.D., A.I. Bobenko, V.Z. Enolโsii, A.R. Its, and V.B. Matveev, Algebro-Geometric Approach to Nonlinear Integrable Equations, (Springer-Verlag series in Nonlinear Dynamics, 1994).
Ercolani, N. โGeneralized Theta functions and homoclinic varieties,โ Proc. Symp. Pure Appl. Math. 49, 87โ100 (1989).
Kupershmidt, B.A., โMathematics of Dispersive Water Waves,โ Commun. Math. Phys. 99, 51โ73 (1985).
## Figures |
warning/0001/hep-ex0001061.html | ar5iv | text | # 1 Introduction
## 1 Introduction
This review covers a few selected topics from the searches performed at Tevatron, HERA and LEP2. Details on the data samples analysed at the time of the conference are given in Table 1.
New particle searches are dominated by LEP whose results come mostly from data up to 189 GeV except for some updates from โ99 data up to 196 GeV. Combined results from the LEP experiments also exist on some subjects and are denoted ADLO.
After a brief outline of the searches on exotic particles, results on supersymmetric particles and Higgs bosons are detailed. All exclusion limits are at the 95% Confidence Level.
## 2 Exotic particles
Searches for exotic particles encompass a great variety of topics, such as technicolor particles , new Zโ bosons , four fermion contact interactions , new, excited or exotic fermions , leptoquarks. Constraints are derived at the three colliders either from direct searches or from comparing precise measurements with Standard Model (SM) expectations.
As an illustration, the case of leptoquark (LQ) searches is detailed. The phenomenology of leptoquarks is describbed by three parameters: M<sub>LQ</sub>, the LQ mass, $`\lambda _{\mathrm{lq}}`$ and $`\beta _\mathrm{l}`$, the LQ coupling and branching ratio into a given pair of SM lepton and quark, (l,q). Results are interpreted either assuming leptoquarks to be coupled to a single SM generation, with fixed $`\beta _\mathrm{l}`$ (as in the BRW model ) or in more generic models, with $`\beta _\mathrm{l}`$ variable and possible mixed couplings.
The current constraints in the first approach are summarized in Table 2. The interpretation in generic models has been pioneered by H1 . As an example, Figure 1 shows the regions excluded by H1 and DO in the plane M<sub>LQ</sub> vs $`\beta _\mathrm{e}`$ for first generation leptoquarks. HERA sensitivity extends down to small $`\beta _\mathrm{e}`$ even for small couplings, e.g. for $`\beta _\mathrm{e}10\%`$ and $`\lambda _{\mathrm{eq}}0.05`$ leptoquark masses up to 200 GeV/$`c^2`$ are excluded. The interesting case of mixed couplings has also been studied . To cite one result, leptoquarks with couplings to both (e, q) and ($`\tau `$, q) have been found to be excluded with a sensitivity similar to that quoted for pure first generation leptoquarks.
## 3 Supersymmetric particles
All supersymmetric particle searches are conducted within the Minimal Supersymmetric Standard Model (MSSM) with additional assumptions to decrease the number of free parameters. Depending on those assumptions, the phenomenology differs and so do the experimental signatures. At present, three theoretical frameworks are studied.
### 3.1 Constrained MSSM
The most common framework assumes R-parity conservation and soft supersymmetry breaking mediated by gravity. Soft breaking terms are thus unified at high energy (the so-called GUT scale) and the number of free parameters is reduced to five: the common sfermion<sup>1</sup><sup>1</sup>1Tevatron experiments use a somewhat different and more constrained scheme, called minimal supergravity (mSUGRA), which defines $`\mathrm{m}_0`$ as a common scalar mass term at the GUT scale and assumes in addition radiative EW symmetry breaking, so that $`\mu `$ is fixed up to a sign. mass term at GUT scale, $`\mathrm{m}_0`$, the common gaugino mass term at GUT scale, $`\mathrm{m}_{1/2}`$, the common trilinear coupling at GUT scale, A<sub>0</sub>, the Higgs mixing parameter, $`\mu `$, and the ratio of the two Higgs doublet vacuum expectation values, $`\mathrm{tan}\beta `$.
The phenomenology at low energy is derived using renormalisation group equations. The lightest supersymmetric particle (LSP) is in most cases the lightest neutralino, $`\stackrel{~}{\chi }_1^0`$. Due to R-parity conservation, sparticles are produced in pairs and decay in their SM partner and a sparticle. At the end of the decay chain, the LSP appears and, as it is stable, gives rise to missing energy. Results on all types of sparticles have been reported at the conference.
#### 3.1.1 Sfermions
Charged sleptons and light squarks are searched for at LEP through the decays summarized in Table 3, which are the dominant ones if the researched sparticle is the next lightest sparticle. The experimental sensitivity depends on both the sfermion mass and the mass difference $`\mathrm{\Delta }M`$ between the sfermion and the LSP. The experimental sensitivity starts from $`\mathrm{\Delta }M`$ above a few GeV/$`c^2`$ and covers sfermion masses up to 70 to 90 GeV/$`c^2`$ depending on the sfermion, as can be seen in Table 3 for the LEP combination at 189 GeV . Similar sensitivities are reached by the individual LEP experiments at 196 GeV . All LEP results are derived for minimal production cross-sections and hence have a general validity.
Light squarks are also searched for at Tevatron . The experimental sensitivity is complementary from that of LEP since it covers higher squark masses and starts at higher $`\mathrm{\Delta }M`$, as illustrated in Table 3. Searches for heavy squarks and gluinos belong to Tevatron . The final states result from $`\stackrel{~}{\mathrm{q}}`$ and $`\stackrel{~}{\mathrm{g}}`$ cascade decays to the LSP and quarks, gluons, W or Z bosons. The present experimental reach is around 250 GeV/$`c^2`$ but it must be noted that most results are derived for specific values of some mSUGRA parameters (A$`{}_{0}{}^{}=0`$, $`\mu <0`$ and $`\mathrm{tan}\beta `$= 2) which restricts their range of validity.
#### 3.1.2 Charginos and neutralinos
Due to its excellent coverage of the various signatures of supersymmetry, LEP provides limits on the masses of the lightest chargino and neutralino which are practically absolute in the constrained MSSM framework.
Direct searches for the lightest chargino $`\stackrel{~}{\chi }_1^\pm `$ provide limits on $`m_{\stackrel{~}{\chi }_1^\pm }`$ close to the kinematical limit in most of the parameter space . As $`\stackrel{~}{\chi }_1^0`$ cannot be detected, direct neutralinos searches rely on the production of heavier neutralinos (e.g. $`\stackrel{~}{\chi }_1^0`$ $`\stackrel{~}{\chi }_2^0`$ or $`\stackrel{~}{\chi }_2^0`$ $`\stackrel{~}{\chi }_2^0`$ ) and thus bring little constraint on $`m_{\stackrel{~}{\chi }_1^0}`$ except when $`\mathrm{tan}\beta `$ is close to 1. But, combining the results from $`\stackrel{~}{\chi }_\mathrm{i}^0`$ and $`\stackrel{~}{\chi }_1^\pm `$ searches provides a limit on $`m_{\stackrel{~}{\chi }_1^0}`$ valid for large m<sub>0</sub> whatever the other parameters .
For low values of m<sub>0</sub> (that is light $`\stackrel{~}{\nu }`$), the $`\stackrel{~}{\chi }_1^\pm `$ production cross-section drops due to the negative interference between the s-channel production process and the t-channel $`\stackrel{~}{\nu }`$ exchange diagram. The $`\stackrel{~}{\chi }_1^\pm `$ decay into l $`\stackrel{~}{\nu }`$ becomes dominant and escapes detection when $`\stackrel{~}{\chi }_1^\pm `$ and $`\stackrel{~}{\nu }`$ are denegerated. Under the same conditions, the $`\stackrel{~}{\chi }_\mathrm{i}^0`$ production increases (because of a positive interference term) but the $`\stackrel{~}{\chi }_\mathrm{i}^0`$ decay into $`\nu `$ $`\stackrel{~}{\nu }`$ opens, leading to invisible final states. But, low values of m<sub>0</sub> also mean light sleptons, which are thus within experimental reach. Combining $`\stackrel{~}{\chi }_\mathrm{i}^0`$, $`\stackrel{~}{\chi }_1^\pm `$ and $`\stackrel{~}{\mathrm{l}}`$ searches, limits on $`m_{\stackrel{~}{\chi }_1^\pm }`$ and $`m_{\stackrel{~}{\chi }_1^0}`$ valid for A<sub>0</sub>=0 and for all values of the other parameters can be derived , as illustrated in Table 4 and Figure 2. Preliminary studies varying A<sub>0</sub> (that is allowing for $`\stackrel{~}{\tau }`$ mixing) show that the limit on $`m_{\stackrel{~}{\chi }_1^\pm }`$ may be affected but not that on $`\stackrel{~}{\chi }_1^0`$.
### 3.2 R-parity breaking
In the second theoretical framework, soft supersymmetry breaking is still mediated by gravity but R-parity ($`\mathrm{R}_\mathrm{p}`$) is assumed to be broken via an extra-term to the superpotential of the form:
$$๐ฒ=\lambda _{ijk}\mathrm{L}_i\mathrm{L}_j\overline{\mathrm{E}}_k+\lambda _{ijk}^{^{}}\mathrm{L}_i\mathrm{Q}_j\overline{\mathrm{D}}_k+\lambda _{ijk}^{^{\prime \prime }}\overline{\mathrm{U}}_i\overline{\mathrm{D}}_j\overline{\mathrm{D}}_k$$
(1)
where $`ijk`$ denote generation indices and the capital letters refer to superfields associated to left-handed doublets of leptons (L) and quarks (Q), and right-handed singlets of charged leptons (E), down-type quarks (D) and up-type quarks (U). $`๐ฒ`$ implies violation of the leptonic and baryonic numbers. In addition to the five parameters related to supersymmetry breaking ($`\mathrm{m}_0`$, $`\mathrm{m}_{1/2}`$, A<sub>0</sub>, $`\mu `$ and $`\mathrm{tan}\beta `$), $`\mathrm{R}_\mathrm{p}`$ breaking ($`\mathrm{R}_\mathrm{p}`$ / ) introduces 45 couplings (9 $`\lambda _{ijk}`$, 27 $`\lambda _{ijk}^{^{}}`$ and 9 $`\lambda _{ijk}^{^{\prime \prime }}`$). For sake of simplicity, searches are conducted assuming only one coupling to dominate at a time and all sparticles to decay close to the interaction vertex. This latter hypothesis corresponds to assuming the $`\mathrm{R}_\mathrm{p}`$ / couplings to be greater than values which are at least two orders of magnitude below the current experimental limits on most couplings.
Compared to $`\mathrm{R}_\mathrm{p}`$ conservation, $`\mathrm{R}_\mathrm{p}`$ / modifies the phenomenology at low energy. Single sparticle production is possible, the LSP (which is still $`\stackrel{~}{\chi }_1^0`$ in most cases) is no longer stable and the sparticle decay patterns change a lot. Sparticle can decay into SM particles through one $`\mathrm{R}_\mathrm{p}`$ / vertex (e.g. $`\stackrel{~}{\nu }`$ case in Figure 3a) or via an $`\mathrm{R}_\mathrm{p}`$ conserving vertex leading to an off-shell sparticle that decays through an $`\mathrm{R}_\mathrm{p}`$ / vertex (e.g. $`\stackrel{~}{\chi }_1^0`$ and $`\stackrel{~}{\chi }_1^\pm `$ cases in Figure 3a). These decays are referred to as direct decays. A second type of decays, called indirect decays, implies cascade decays to SM particles through several $`\mathrm{R}_\mathrm{p}`$ conserving and $`\mathrm{R}_\mathrm{p}`$ / vertices with some sparticles on-shell, such as in Figure 3b.
Thus, there are many final states to consider, with multileptons and/or multijets and possibly missing energy from neutrinos. As in the constrained MSSM case, all types of sparticles have been searched for.
#### 3.2.1 Charginos and neutralinos
As in the $`\mathrm{R}_\mathrm{p}`$ conserving scheme, LEP sets constraints on charginos and neutralinos which are valid over wide ranges of the underlying parameters (except A<sub>0</sub> which is set to 0). The $`\stackrel{~}{\chi }_1^\pm `$ and $`\stackrel{~}{\chi }_\mathrm{i}^0`$ production modes are as in the $`\mathrm{R}_\mathrm{p}`$ conserving case while their decay pattern is completely modified. The most notable change concerns the LSP whose production leads to observable final states whatever the dominant $`\mathrm{R}_\mathrm{p}`$ / coupling. All experimental signatures expected from the production and decays of $`\stackrel{~}{\chi }_1^+`$ $`\stackrel{~}{\chi }_1^{}`$, $`\stackrel{~}{\chi }_1^0`$ $`\stackrel{~}{\chi }_1^0`$ and $`\stackrel{~}{\chi }_2^0`$ $`\stackrel{~}{\chi }_1^0`$ have been investigated at LEP2. Due to $`\stackrel{~}{\chi }_1^0`$ being detectable, combining $`\stackrel{~}{\chi }_\mathrm{i}^0`$ and $`\stackrel{~}{\chi }_1^\pm `$ searches alone suffices to derive constraints on the sparticle masses valid for A$`{}_{0}{}^{}=0`$ and all values of the other parameters, as illustrated in Table 5.
#### 3.2.2 Sneutrinos
Another appreciable change due to $`\mathrm{R}_\mathrm{p}`$ / is the allowed decay of sneutrinos which make them directly observable e.g. at LEP through double and single productions.
All final states expected from pair production have been searched for. An example of result for a dominant $`\lambda _{133}`$ coupling is given in Figure 4. Using also the limit on $`m_{\stackrel{~}{\chi }_1^0}`$, exclusion limits can be derived for each of the three $`\mathrm{R}_\mathrm{p}`$ / coupling classes. The current limits are presented in Table 6. They are valid for all values of the underlying parameters, except A<sub>0</sub> which is set to 0.
Single production of $`\stackrel{~}{\nu }`$<sub>ฮผ</sub> or $`\stackrel{~}{\nu }`$<sub>ฯ</sub> sneutrinos is also possible at LEP, as illustrated in Figure 5. Sneutrino direct decays would lead to effects observable as deviations wrt SM expectations while indirect decays would manifest as specific final states that require dedicated searches. In both cases, constraints on sneutrino masses have been derived as a function of the $`\mathrm{R}_\mathrm{p}`$ / couplings. Currently, masses between 100 to 200 GeV/$`c^2`$ are probed and limits on couplings are of the order of a few 10<sup>-2</sup> .
#### 3.2.3 Charged sleptons and squarks
Charged sleptons and light squarks are searched for at LEP. The coverage of the final states expected in case of $`\mathrm{R}_\mathrm{p}`$ / (from direct and indirect decays and for any type of couplings) is not yet as complete as for the other sparticles. The results achieved so far are summarized in Table 7. As for other LEP results, they hold for A$`{}_{0}{}^{}=0`$, irrespective of the values of the other underlying parameters.
Finally, there are also constraints on heavy squarks and gluinos if $`\mathrm{R}_\mathrm{p}`$ is violated. LEP experiments reinterpret their results about leptoquark searches while dedicated searches are performed at HERA and Tevatron for particular $`\mathrm{R}_\mathrm{p}`$ / couplings. HERA experiments have access mostly to $`\lambda _{1j1}^{^{}}`$ couplings and derive constraints on heavy squark masses as a fuction of $`\lambda _{1j1}^{^{}}`$, varying the MSSM underlying parameters. Currently, masses around 240 GeV/$`c^2`$are excluded for $`\lambda _{1j1}^{^{}}0.3`$ . Using multilepton events in run I data, Tevatron experiments have probed $`\lambda _{1jk}^{^{}}`$ and $`\lambda _{121}`$ couplings, and constrained gluino and heavy squark masses. The limits achieved are around 250 (resp. 350) GeV/$`c^2`$ for the $`\lambda _{1jk}^{^{}}`$ (resp. $`\lambda _{121}`$) coupling analysis . These limits are valid for A$`{}_{0}{}^{}=0`$, $`\mu <0`$ and $`\mathrm{tan}\beta `$= 2. Most of the other choices (in particular, higher values of $`\mathrm{tan}\beta `$ or $`\mu >0`$) would lead to lower limits, due to a loss in sensitivity (resulting from reduced branching fractions into leptons, softer leptonsโฆ) .
### 3.3 Gauge mediated supersymmetry breaking
The third theoretical framework assumes $`\mathrm{R}_\mathrm{p}`$ conservation and soft supersymmetry breaking mediated by gauge interactions. Such models usually need six basic parameters: the supersymmetry breaking scale, $`\sqrt{\mathrm{F}}`$, the universal mass scale of supersymmetric particles, $`\mathrm{\Lambda }`$, the messenger mass scale, $`\mathrm{M}_\mathrm{s}`$, the number of messenger generations, $`\mathrm{n}_\mathrm{s}`$, the Higgs mixing parameter, $`\mu `$, and the ratio of the two Higgs doublet vacuum expectation values, $`\mathrm{tan}\beta `$. The breaking scale is expected to be much lower than in gravity-mediated models, down to about 10<sup>4</sup> GeV.
As far as phenomenology at low energy is concerned, $`\mathrm{R}_\mathrm{p}`$ conservation implies as usual sparticle pair production and a stable LSP. As a consequence of gauge-mediated breaking, the LSP is the gravitino, $`\stackrel{~}{\mathrm{G}}`$, whose mass depends on $`\sqrt{\mathrm{F}}`$ and thus is expected to be very small, in the range \[10<sup>-6</sup> eV, 1 keV\]. The next lightest sparticle (NSLP) is either $`\stackrel{~}{\chi }_1^0`$ or a charged slepton ($`\stackrel{~}{\tau }_1`$ or three degenerated $`\stackrel{~}{\mathrm{l}}`$). The NSLP lifetime is governed by the $`\stackrel{~}{\mathrm{G}}`$ mass and hence can be non negligible, giving rise to specific topologies, some being experimental challenges.
#### 3.3.1 $`\stackrel{~}{\chi }_1^0`$ NLSP
The main decay of a $`\stackrel{~}{\chi }_1^0`$ NLSP would be in $`\gamma `$$`\stackrel{~}{\mathrm{G}}`$. Thus $`\stackrel{~}{\chi }_1^0`$ searches rely on final states with photons, either single-photon or diphoton events. Such final states provide clean experimental signatures and have been used since long to chase new physics, whatever the underlying theoretical framework. Results are usually expressed as model-independent upper limits on the cross-section times branching fraction product, as a function of the mass of the unknown particle decaying to a photon plus missing energy. These cross-section limits are then compared with predictions from various models , such as the GMSB scenario explaining an ee$`\gamma \gamma `$ event reported by CDF some years ago. To quote but one result, even if not strictly a GSMB model, cross-section limits have been converted into a lower limit on the mass of a superlight $`\stackrel{~}{\mathrm{G}}`$, assuming all other sparticles to be above threshold. The current limit from single-photon events at LEP2 is 10<sup>-5</sup> eV/$`c^2`$, similar to the result reached by CDF using monojet events .
Topologies more specific to GMSB models have also been searched for. The first example is given by $`\stackrel{~}{\chi }_1^0`$ searches in case of long neutralino lifetimes. The decay photons would not be produced at the interaction point but could have large impact parameters. Searches for non-pointing single-photon events cover that case. Searches for sparticles heavier than $`\stackrel{~}{\chi }_1^0`$ have also been performed, so far for charginos and sleptons only. Compared with the topologies expected in the gravity-mediated scheme, the final states are identical except for additional photons that can be detected if the NLSP lifetime is not too long. As photons help to better discriminate against background, the exclusion limits for negligible or moderate NSLP lifetimes are usually tigther than those in the gravity-mediated framework.
#### 3.3.2 $`\stackrel{~}{\tau }_1`$ NLSP
The main decay of a $`\stackrel{~}{\tau }_1`$ NLSP would be in $`\tau `$ $`\stackrel{~}{\mathrm{G}}`$, thus giving the same experimental signature as in the gravity-mediated case only if the $`\stackrel{~}{\tau }_1`$ lifetime is negligible. In the contrary case, $`\stackrel{~}{\tau }_1`$ decays would lead to kinks, large impact parameters or decay vertices. Eventually, a $`\stackrel{~}{\tau }_1`$ decaying outside the detector would appear as a stable charged particle. All these experimental signatures have been used in $`\stackrel{~}{\tau }_1`$ searches . An example of result is given in Figure 6 which illustrates the interplay of the different signatures as a function of the $`\stackrel{~}{\mathrm{G}}`$ mass which defines the $`\stackrel{~}{\tau }_1`$ mean decay length. Irrespective of the $`\stackrel{~}{\mathrm{G}}`$ mass, these results exclude a $`\stackrel{~}{\tau }_1`$ NLSP up to 73 GeV/$`c^2`$. The limit is 6 GeV/$`c^2`$ higher if the results are reinterpreted in a scenario with three degenerated co-NLSP charged sleptons.
Searches for sparticles heavier than $`\stackrel{~}{\tau }_1`$ have also been performed, so far for charginos, neutralinos and sleptons and only for negligible NLSP lifetimes.
#### 3.3.3 Constraining underlying parameters
Even if not complete, the coverage of the final states expected from GMSB models is at present sufficient to exclude large fractions of the parameter space in order to set constraints on the key parameters of the model, like the NLSP mass or $`\mathrm{\Lambda }`$. A first attempt has been reported in for a minimal GMSB model.
### 3.4 Other searches, recent developments
To conclude about supersymmetry, it is worth mentioning a few complementary results from LEP, either recent ones or results aside the main stream of searches described in the previous sections.
There is an ongoing effort to combine the four experiment results (on $`\stackrel{~}{\chi }_1^\pm `$, $`\stackrel{~}{\mathrm{e}}_\mathrm{R}`$, $`\stackrel{~}{\tau }_1`$ searches, using also Higgs searches and the $`\mathrm{\Gamma }_Z`$ constraint) to derive an absolute limit on the $`\stackrel{~}{\chi }_1^0`$ mass in the mSUGRA framework. Preliminary results from data up to 189 GeV give a limit at 44 GeV/$`c^2`$ if the difference between the $`\stackrel{~}{\tau }_1`$ and $`\stackrel{~}{\chi }_1^0`$ masses is above 5 GeV/$`c^2`$ irrespective of the values of the underlying parameters, except A<sub>0</sub> which is set to 0. Recently, dedicated searches have been performed to cover mass differences below 5 GeV/$`c^2`$ and first results confirmed the validity of the above limit also in that case . The next step should be to check the impact of A<sub>0</sub>, which is expected to be large.
About $`\mathrm{R}_\mathrm{p}`$ / the superpotential quoted in equation (1) is not the most complete $`\mathrm{R}_\mathrm{p}`$ violating potential. Extra bilinear terms of the form $`ฯต_i\mathrm{L}_\mathrm{i}\mathrm{H}_2`$, where H<sub>2</sub> is the Higgs superfield with positive hypercharge, are also possible candidates to generate $`\mathrm{R}_\mathrm{p}`$ / . First results have been reported recently on a search for stops decaying through a bilinear term into a b$`\tau `$ pair .
Finally, if $`\mathrm{R}_\mathrm{p}`$ is conserved, the LSP being stable restricts the experimental sensitivity to other sparticles to mass differences between the sparticle and the LSP in excess of a few GeV/$`c^2`$. Searches have been conducted to explore nearly mass degeneracy cases for the lightest chargino and more recently for the lightest selectron . Note that sparticles degenerated with the LSP exist in restricted regions of the parameter space of the constrained MSSM as defined in section 3.1 but they are mostly predicted in other supersymmetric scenarios, such as a constrained MSSM without gaugino mass unification at the GUT scale or the recent anomaly mediated SUSY breaking models .
## 4 Search for extra dimensions
It was pointed out recently that extra spatial dimensions, which are present in any superstring theory, can also solve the hierarchy problem, independently of the underlying theoretical framework. Indeed, if n extra compact spatial dimensions of radius R exist, the quantum gravity scale in n+4 dimensions, M<sub>D</sub>, is related to the Planck scale, M<sub>Pl</sub>, by M$`{}_{\mathrm{Pl}}{}^{2}\mathrm{R}^n\mathrm{M}_\mathrm{D}^{2+\mathrm{n}}`$. If R and n are such that M<sub>D</sub> is of the order of the electroweak (EW) scale, the hierarchy vanishes. The case n=1 is ruled out since it would imply quantum gravity effects observable over solar system distances.
At low energy, extra spatial dimensions are expected to manifest through the production of gravitons, G, observable in both direct searches and precise measurements. Searching for the associated production of a pair ($`\gamma `$, G) in single photon final states at LEP2 leads to constraints on M<sub>D</sub> depending on n. As an example, M<sub>D</sub> has been found to be greater than 1.1 TeV, 0.7 TeV and 0.53 TeV for 2, 4 or 6 extra dimensions, respectively . Gravitons would also be responsible for deviations wrt the SM in precise measurements. Combining observables in several final states at LEP, the ultra-violet cut-off of the underlying quantum gravity theory has been constrained to be larger than 0.8 TeV (resp. 1.1 TeV) if the interference between the SM and G exchange amplitudes is negative (resp. positive) .
## 5 Higgs bosons
The phenomenology of Higgs bosons is little model-dependent which allows to cover several theoretical frameworks with a limited number of searches. Results encompass neutral Higgs bosons as expected in the SM, MSSM and beyond, as well as charged Higgs bosons in two Higgs doublet models.
### 5.1 The Standard Model Higgs boson
In the mass range currently under study, i.e. around 100 GeV/$`c^2`$, only LEP would be sensitive to the SM Higgs boson. The main production process leads to pairs of Higgs and Z bosons, with the Z boson on-shell. Due to the clean experimental environment, all Z final states are exploited. In addition, the Higgs boson is expected to decay mainly into a $`\mathrm{b}\overline{\mathrm{b}}`$ pair (the branching fraction is $``$ 82% for a 100 GeV/$`c^2`$ Higgs boson) so that excellent b-tagging capabilities of the LEP detectors help a lot in these searches.
The results obtained at 189 GeV by the four LEP experiments are detailed in Table 8 at the level of selections where data are actually compared with simulation to test the background and signal+background hypotheses and derive exclusion limits or discovery significances. To achieve the highest sensitivity to the signal, these derivations rely on a test-statistic which, besides the rates, take also into account the pattern of the selected events (the reconstructed Higgs boson mass, m<sub>H</sub>, or m<sub>H</sub> and another discriminant variable like the event b-quark content) . The more information in the comparison, the earlier the event selection procedure is stopped, as illustrated in Table 8 by the different selection levels in the four experiments. There is an excess in data in three experiments, which is partly signal-like in two of them, as revealed by the difference of a few GeV/$`c^2`$ between the observed and expected exclusion limits. After investigation, part of the excess was attributed to a systematic bias. When combining the four LEP experiments, the exclusion limit is 95.2 GeV/$`c^2`$ for an expectation of 97.2 GeV/$`c^2`$ .
A preliminary update at 196 GeV was reported at the conference , as shown in Table 9. The reconstructed Higgs boson mass distribution after tighter selections is given in Figure 7. The excess seen at 189 GeV has not been confirmed at higher energies and data agree with expectations. After the conference, these results were combined, giving an exclusion limit of 102.6 GeV/$`c^2`$, for an expected limit of 102.3 GeV/$`c^2`$ . At the end of the โ99 run, which reached $`\sqrt{s}`$=202 GeV, experiments reported preliminary limits up to 106 GeV/$`c^2`$, in good agreement with the expected ones . Prospects for the last run of LEP at $`\sqrt{s}`$ up to 206 GeV are 114 GeV/$`c^2`$ for the 95% exclusion or 3$`\sigma `$ discovery potentials and 111 GeV/$`c^2`$ for the 5$`\sigma `$ discovery sensitivity, when the four experiments are combined . Higher masses up to 180 GeV/$`c^2`$ should then be accessible in the future high luminosity run at the Tevatron .
### 5.2 MSSM neutral Higgs bosons
Most results about neutral Higgs bosons in the MSSM come also from LEP, which is sensitive to the two lightest bosons, h and A. There are two production processes, $`\mathrm{e}^+\mathrm{e}^{}`$hZ, like in the SM case, and $`\mathrm{e}^+\mathrm{e}^{}`$hA. The two processes are complementary in the parameter space. In the mass range between 80 and 110 GeV/$`c^2`$, the main decay mode of both bosons is again in $`\mathrm{b}\overline{\mathrm{b}}`$ in most of the parameter space, with branching fractions greater than in the SM ($``$ 91%). The dominant hZ final states are as in the SM case, while hA is expected to give mostly $`\mathrm{b}\overline{\mathrm{b}}\mathrm{b}\overline{\mathrm{b}}`$ and $`\tau ^+\tau ^{}\mathrm{b}\overline{\mathrm{b}}`$ final states. Here again, b-tagging plays a crucial rรดle.
The theoretical framework of these searches is the MSSM with $`\mathrm{R}_\mathrm{p}`$ conservation and soft breaking terms unified at the EW scale. In the MSSM the Higgs boson masses are connected to each other, so that at tree level, there are only two free parameters: $`\mathrm{tan}\beta `$ and one Higgs boson masses, or, alternatively, two Higgs boson masses, eg $`\mathrm{m}_\mathrm{A}`$ and $`\mathrm{m}_\mathrm{h}`$. The properties of the MSSM Higgs bosons, and in particular the mass relationships, are modified by radiative corrections which introduce five additional parameters: the mass of the top quark, the Higgs mixing parameter, $`\mu `$, the common sfermion mass term at the EW scale, M<sub>S</sub>, the common SU(2) gaugino mass term <sup>2</sup><sup>2</sup>2The U(1) gaugino mass term at the EW scale, M<sub>1</sub>, is related to M<sub>2</sub> through the GUT relation M$`{}_{1}{}^{}=(5/3)\mathrm{tan}^2\theta _\mathrm{w}\mathrm{M}_2`$, while the SU(3) gaugino mass term, M<sub>3</sub>, is set via the gluino mass, which is taken equal to M<sub>S</sub>. at the EW scale, M<sub>2</sub>, and the common squark tri-linear coupling at the EW scale, A. The interpretation of the experimental results depend on the values assumed for these parameters.
#### 5.2.1 Benchmark hypotheses
Using leading order two-loop calculations of the radiative corrections , benchmark values have been defined for the parameters beyond tree-level : 175 GeV/$`c^2`$ for the top mass, 1 TeV/c<sup>2</sup> for M<sub>S</sub> and 1.6 TeV/c<sup>2</sup> for M<sub>2</sub>. Two benchmark scenarios have been defined for the parameters A and $`\mu `$, which determine the mixing in the stop sector: no mixing (A$`=0`$, $`\mu =100`$ GeV) and maximal mixing (A$`=\sqrt{6}\mathrm{M}_\mathrm{S}`$, $`\mu =100`$ GeV). The no mixing hypothesis leads to minimal radiative corrections to $`\mathrm{m}_\mathrm{h}`$ while the maximal mixing induces the largest corrections. This defines the usual framework for the interpretation of the MSSM neutral Higgs boson searches.
Results obtained at LEP at 189 GeV in the hA channel are summarized in Table 10. These results, together with the hZ results reinterpreted in the MSSM framework, allow to set constraints on $`\mathrm{m}_\mathrm{h}`$, $`\mathrm{m}_\mathrm{A}`$ and $`\mathrm{tan}\beta `$. As an example, Figure 8 represents the region of the ($`\mathrm{tan}\beta `$, $`\mathrm{m}_\mathrm{h}`$) plane excluded by the combination of the LEP results up to 189 GeV, in the less favourable case of the maximal mixing . Whatever the mixing hypothesis, these combined results exclude Higgs bosons up to 80.7 GeV/$`c^2`$ for $`\mathrm{m}_\mathrm{h}`$ and 80.9 GeV/$`c^2`$ for $`\mathrm{m}_\mathrm{A}`$ for $`\mathrm{tan}\beta `$ greater than 0.4. At $`\mathrm{tan}\beta `$$`1`$, the experimental lower limit on $`\mathrm{m}_\mathrm{h}`$ is above the theoretical upper bound on $`\mathrm{m}_\mathrm{h}`$ so that $`\mathrm{tan}\beta `$ is excluded between 0.9 and 1.6 (0.6 and 2.6) in the maximal mixing (no mixing) hypothesis. The expected limits on both masses are 5 GeV/$`c^2`$ higher while the expected excluded ranges in $`\mathrm{tan}\beta `$ agree with the observed ones.
It must be noted that, contrary to the limits on masses, the limit on $`\mathrm{tan}\beta `$ is very sensitive to the values of the underlying parameters, which have a large impact on the theoretical upper bound on $`\mathrm{m}_\mathrm{h}`$. As an example, this upper bound increases with increasing top quark masses and no limit on $`\mathrm{tan}\beta `$ is obtained if the top mass is moved by two standard deviations. The pure MSSM parameters or the order of the radiative correction calculations have also a non negligible effect. So, the excluded ranges in $`\mathrm{tan}\beta `$ cannot be taken as an absolute result, even if the maximal mixing hypothesis is a pessimistic scenario.
Also displayed in Figure 8 is the recent CDF result at large $`\mathrm{tan}\beta `$. In this region, the production cross-section is large enough to make Tevatron sensitive to the production process $`\mathrm{p}\overline{\mathrm{p}}`$$`\mathrm{b}\overline{\mathrm{b}}`$ h,H,A where one Higgs boson is emited off a bottom quark, leading to four b-tagged jets in the final state.
Results from LEP experiments with data up to 196 GeV were also reported at the conference , as shown in Table 11. These results were combined later on, giving exclusion limits of 84.3 GeV/$`c^2`$ on $`\mathrm{m}_\mathrm{h}`$ and 84.5 GeV/$`c^2`$ on $`\mathrm{m}_\mathrm{A}`$, independent of the mixing hypothesis, and excluded ranges in $`\mathrm{tan}\beta `$ between 0.8 and 1.9 (0.5 and 3.2) in the maximal mixing (no mixing) hypothesis . At the end of the โ99 data taking, LEP experiments reported mass limits up to 90 GeV/$`c^2`$ and excluded ranges in $`\mathrm{tan}\beta `$ somewhat larger than the combined result at 196 GeV . The excess in data observed by OPAL in the hA channel at 196 GeV was not confirmed at higher energies nor by the other experiments.
#### 5.2.2 General scans
More general interpretations of the LEP Higgs boson searches have been performed, scanning over the MSSM underlying parameters. The parameter space is however usually restricted by imposing additional constraints, e.g. the experimental results on supersymmetric particles or the $`\mathrm{\Gamma }_Z`$ constraint. It was shown that the benchmark limits hold in more than 99.99% of the parameter sets and that the mass limits from general scans are a few GeV/$`c^2`$ weaker than the benchmark ones . As general scans usually vary also the top mass quark within two standard deviations, the limit on $`\mathrm{tan}\beta `$ vanishes.
#### 5.2.3 Recent developments
Recent theoretical work led to two-loop calculations of the radiative corrections at the next-to-leading order and to a redefinition of the benchmark values for the underlying parameters . In particular, the theoretical upper bound on $`\mathrm{m}_\mathrm{h}`$ was found to be underestimated by $``$GeV/$`c^2`$ in the maximal mixing scenario previously used. The new scenario now proposed (called $`\mathrm{m}_\mathrm{h}`$<sup>max</sup> scenario) should lead to more realistic bounds on $`\mathrm{tan}\beta `$. Another new scenario (called large $`\mu `$ scenario) with the h boson within kinematical reach at LEP but with vanishing branching fraction into $`\mathrm{b}\overline{\mathrm{b}}`$ was proposed to check the sensitivity of LEP to Higgs bosons with non-dominant b decays. Future LEP results will be derived in these new benchmark schemes.
### 5.3 Neutral Higgs bosons beyond MSSM
Searches for neutral Higgs bosons as expected beyond the MSSM have also been performed, mostly at LEP. Three lines of searches have been followed. First, the existing LEP analyses on MSSM h and A bosons have been used, either as such or with some modifications (e.g. relaxing the b-tagging requirements) to cover the final states expected in more general models. Thus, a first study showed that LEP is sensitive to neutral Higgs bosons of two Higgs doublet models (2HDM), even in a scenario with dominant decays into $`\mathrm{c}\overline{\mathrm{c}}`$ or in a model with CP violation . Recently, the 2HDM parameter space (with CP conservation) has been explored in a detailed scan . Finally, for the first time, a non minimal supersymmetric model containing one gauge-singlet Higgs field in addition to the MSSM has also been investigated .
As a second research line, the case of a Higgs boson h decaying invisibly has been studied. Dedicated searches in the hZ channel translate into upper limits on the production cross-section times branching ratio, which are compared with expectations from specific models . As an example, assuming a SM production rate and a 100% branching ratio into invisible products, a lower limit on $`\mathrm{m}_\mathrm{h}`$ at 95.4 GeV/$`c^2`$ is obtained.
The third topic deals with a Higgs boson h with anomalous couplings to photons. From dedicated searches in the hZ, h$`\gamma `$ and hA channels, general constraints are set on the production cross-section times branching ratio or directly on the anomalous couplings. They are again compared with expectations from specific models . As an example, assuming a SM production rate and a fermiophobic Higgs boson, a lower limit on $`\mathrm{m}_\mathrm{h}`$ at 97.5 GeV/$`c^2`$ is achieved.
### 5.4 Charged Higgs bosons
Recent results on charged Higgs bosons, $`\mathrm{H}^\pm `$, have been reported by the LEP experiments. The framework of these searches is the general 2HDM scheme with as sole free parameters the $`\mathrm{H}^\pm `$ mass and its leptonic decay branching fraction, assuming that the hadronic (into cs) and leptonic (into $`\tau \nu _\nu `$) decays saturate the width of the particle, which is the case in the mass range below $`m_W`$ that is presently tested.
The results obtained by the LEP experiments at 189 GeV are shown in Table 12. Once combined, these results exclude an $`\mathrm{H}^\pm `$ boson up to 77.3 GeV/$`c^2`$ whatever its leptonic decay branching ratio, while the expected limit is 74.9 GeV/$`c^2`$ . Results from data up to 196 GeV were combined after the conference, with no improvement in the mass limit independent of the branching ratio, due to the large background from WW pairs which is penalizing for the analyses in the hadronic mode. On the other hand, $`\mathrm{H}^\pm `$ bosons with pure (50%) leptonic decays have been excluded up to 84.9 (78.4) GeV/$`c^2`$ by the same results . Individual limits reported at the end of the โ99 data taking were below the combined results at 196 GeV .
Higher masses were tested at the Tevatron runI, searching for $`\mathrm{H}^\pm `$ in decays of pair-produced top quarks. The mass limits are $`\mathrm{tan}\beta `$ dependent and restricted to those values of $`\mathrm{tan}\beta `$ for which the top quark branching fraction into $`\mathrm{H}^+`$b is large enough. In the most favourable case ($`\mathrm{tan}\beta `$=150) masses up to 153 GeV/$`c^2`$ have been excluded .
## 6 Conclusions
New particle searches cover an impressive variety of topics and topologies. The way results are interpreted has undergone substantial changes during the past few years. To get higher sensitivity to the researched signals, different channels and/or experiments are combined and more information is put in the statistical analysis of the results, like in the Higgs boson searches. There is also an effort to go to more model-independent results by relaxing theoretical assumptions, scanning parameter values or testing more general models. Supersymmetric particle searches are examples of that kind.
To give but a few results, in gravity-mediated SUSY breaking models, the lightest neutralino has been excluded up to 32 GeV/$`c^2`$ whether $`\mathrm{R}_\mathrm{p}`$ is conserved or not, while the lightest chargino has been excluded up to 68 (94) GeV/$`c^2`$ if $`\mathrm{R}_\mathrm{p}`$ is conserved (broken). A SM Higgs boson has been excluded up to 106 GeV/$`c^2`$, MSSM neutral Higgs bosons up to 90 GeV/$`c^2`$and 2HDM charged Higgs bosons up to 77 GeV/$`c^2`$.
I am grateful to all my colleagues who provided me with information when preparing this talk. I would like to thank M.Besancon, R.Nikolaidou, E.Perez and D.Treille for very helpful discussions.
Discussion
Lee Roberts (Boston University): Can you comment again on the limits on tan$`\beta `$ from the Higgs searches?
Ruhlmann-Kleider: The limits on tan$`\beta `$ have been derived with two-loop leading order calculations of the radiative corrections and for specific values of the underlying parameters so they cannot be regarded as absolute, even if the maximal mixing scenario represents a difficult case. Going to next-to-leading order two-loop calculations, varying the underlying parameters or increasing the top quark mass will make the excluded range in tan$`\beta `$ decrease if not vanish. |
warning/0001/cond-mat0001451.html | ar5iv | text | # Iterative algorithm versus analytic solutions of the parametrically driven dissipative quantum harmonic oscillator
## I Introduction
Exactly solvable systems have a special status among physical models. Although oversimplified in many cases, they may serve as starting point for testing the reliability of methods which can then be transferred to more realistic, but only numerically solvable models. An important class of such models are quantum systems coupled to a dissipative environment and being driven by a time-dependent external field . A wide variety of physical phenomena have been described by this kind of models, e.g. electron and proton transfer, tunneling processes of a macroscopic spin , hydrogen tunneling in condensed phases , single defect tunneling in mesoscopic quantum wires , or tunneling of the magnetic flux in a SQUID , to name but a few. Usually such dissipative quantum systems consist of a model Hamiltonian bilinearly coupled to a bath of harmonic oscillators. Additional external time dependent driving fields render the mathematical solution even more difficult or even impossible.
One of the few analytically tractable time-dependent dissipative quantum systems is the parametrically driven harmonic oscillator whose analytic solution was found by Zerbe and Hรคnggi in Ref. . A physical realization of this model is the Paul trap , which provides an oscillating quadrupole potential for the enclosed ion. Furthermore, the parametrically driven dissipative harmonic oscillator may serve as a benchmark for approximation schemes which were developed for more general dissipative systems . The interesting feature is that the parametric driving induces a non-trivial quasienergy spectrum , in contrast to additive driving where the quasienergy spectrum coincides with the spectrum of the undriven system apart from a constant shift. This is further corroborated by the fact that the solution of the parametrically driven linear oscillator can be utilized to obtain solutions of certain nonlinear dynamical systems .
Powerful approximative numerical procedures for simulating dissipative and possibly time-dependent quantum systems are the Quantum Monte Carlo method and the quasiadiabatic propagator path integral algorithm (QUAPI) developed by Makri and Makarov . The former method works very well for problems involving path integrals in imaginary time, however, the calculation of real time path integrals is afflicted by the so called sign-problem due to the rapidly oscillating integrand. The QUAPI algorithm has been applied to low dimensional dissipative systems such as the driven spin-1/2-particle coupled to a harmonic oscillator bath (driven spin-Boson-system) and to the driven double-well potential in order to study quantum hysteresis and quantum stochastic resonance . Moreover, the QUAPI algorithm has recently been used as a basis for a very efficient memory equation algorithm for spin-boson-models .
The purpose of this paper is to apply the QUAPI algorithm to the parametrically driven harmonic oscillator and to compare the results with the analytic solution from Ref. . Whilst harmonic oscillator systems are known to exhibit some untypical features this is not the case with respect to the QUAPI algorithm. Our results thus show that not only intrinsically discrete models like the spin-boson-system but also spatially continuous systems can be accurately described by few energy eigenstates if the temperature is restricted to a moderate regime. Most importantly, this is the first work in which the numerical approximative QUAPI results are compared against analytic solutions of a spatially continuous driven dissipative quantum system.
The paper is organized as follows: In section II, we introduce our model of the parametrically driven dissipative quantum harmonic oscillator and briefly review the analytic solution given in Ref. . Section III is devoted to a short review of the QUAPI method. The comparative main results are presented in section IV, before we give the conclusions in section V.
## II The model and its analytic solution
In this section we briefly review the analytic solution of the parametrically driven dissipative harmonic oscillator from .
A quantum particle with mass $`M`$, position operator $`๐ฑ`$ and momentum operator $`๐ฉ`$ moving in a one-dimensional harmonic potential with periodically modulated curvature is described by the Hamiltonian
$$๐_\mathrm{S}(t)=\frac{๐ฉ^2}{2M}+\frac{M}{2}[\omega _0^2+ฯต\mathrm{cos}\mathrm{\Omega }t]๐ฑ^2.$$
(1)
Following the common approach to model the influence of the environment by an ensemble of harmonic oscillators, the bath Hamiltonian $`๐_\mathrm{B}`$ (including the interaction with the system) is given by
$$๐_\mathrm{B}=\underset{j}{}๐_j(๐ฑ)=\underset{j}{}\frac{1}{2}\left[\frac{๐ฉ_j^2}{m_j}+m_j\omega _j^2\left(๐ช_j\frac{c_j}{m_j\omega _j^2}๐ฑ\right)^2\right],$$
(2)
and the whole system is described by the Hamiltonian $`๐(t)=๐_\mathrm{S}(t)+๐_\mathrm{B}`$. In the case of a thermal equilibrium bath, it turns out that its influence on the system is fully characterized by the spectral density
$$J(\omega )=\frac{\pi }{2}\underset{j}{}\frac{c_j^2}{m_j\omega _j}\delta (\omega \omega _j).$$
(3)
With the number of harmonic oscillators going to infinity, we arrive at a continuous spectral density. In the following, we choose for the sake of definiteness a truncated Ohmic spectral density, i.e.
$$J(\omega )=M\gamma \omega f_c(\omega ,\omega _c).$$
(4)
Here, $`\gamma `$ is the coupling strength to the heat bath and $`f_c(\omega ,\omega _c)`$ denotes a cut-off function which avoids unphysical divergences due to high-frequency bath modes. For our calculations, we consider two examples for the cut-off function: (i) a smooth exponential cut-off
$$f_c(\omega ,\omega _c)=\mathrm{exp}(\omega /\omega _c)$$
(5)
and (ii) a step-function
$$f_c(\omega ,\omega _c)=\mathrm{\Theta }(\omega _c\omega )$$
(6)
with cut-off frequency $`\omega _c\omega _0,\mathrm{\Omega }`$ (see discussion given below).
We choose a factorizing initial condition of Feynman-Vernon form which means that at time $`t=t_0`$, the full density operator $`๐(t_0)`$ is given as a product of the initial system density operator $`๐_\mathrm{S}(t_0)`$ and the canonical bath density operator at temperature $`T=1/k_B\beta `$, i.e.,
$$๐(t_0)=๐_\mathrm{S}(t_0)Z_\mathrm{B}^1\mathrm{exp}(\beta ๐_\mathrm{B}^0),$$
(7)
where $`Z_\mathrm{B}^1=Tr\mathrm{exp}(\beta ๐_\mathrm{B}^0)`$ and
$$๐_\mathrm{B}^0=\underset{j}{}\frac{1}{2}\left[\frac{๐ฉ_j^2}{m_j}+m_j\omega _j^2๐ช_j^2\right].$$
(8)
By way of integrating out the bath degrees of freedom in Eq. (2) one obtains the following one-dimensional Heisenberg equation for the position operator $`๐ฑ`$, i.e.
$$\ddot{๐ฑ}(t)+_{t_0}^t\widehat{\gamma }(tt^{})\dot{๐ฑ}(t^{})๐t^{}+(\omega _0^2+ฯต\mathrm{cos}\mathrm{\Omega }t)๐ฑ(t)=\frac{1}{M}๐ช(t)\widehat{\gamma }(tt_0)๐ฑ(t_0),$$
(9)
with the friction kernel given by
$$\widehat{\gamma }(t)=\frac{2}{M\pi }_0^{\mathrm{}}๐\omega \frac{J(\omega )}{\omega }\mathrm{cos}(\omega t).$$
(10)
$`๐ช(t)`$ is a time-dependent fluctuating (operator) force
$$๐ช(t)=\underset{j}{}c_j\left(\frac{๐ฉ_j(t_0)}{m_j\omega _j}\mathrm{sin}(\omega _j(tt_0))+๐ช_j(t_0)\mathrm{cos}(\omega _j(tt_0))\right)$$
(11)
which contains the initial conditions of the bath and of the particleโs position at time $`t_0`$. The last term on the r.h.s. (proportional to $`๐ฑ(t_0)`$) in Eq. (9) is the so-called initial slip, caused by the specific choice (7) of the initial conditions.
Exploiting the thermal distribution of the bath one recovers the usual connection (via $`J(\omega )`$) between the random and the frictional forces of the bath in Eq. (11) in the form of the fluctuation-dissipation-relation, reading, $`tt^{}`$,
$$๐ช(t)๐ช(t^{})_\beta =Tr\left[Z_\mathrm{B}^1\mathrm{exp}(\beta ๐_\mathrm{B}^0)๐ช(t)๐ช(t^{})\right]=\mathrm{}L(tt^{}),$$
(12)
$$L(t)=\frac{1}{\pi }_0^{\mathrm{}}๐\omega J(\omega )\left[\text{coth}\left(\frac{\mathrm{}\omega \beta }{2}\right)\mathrm{cos}(\omega t)i\mathrm{sin}(\omega t)\right],$$
(13)
where the subscript $`\beta `$ indicates thermal averaging performed with the canonical density operator for $`๐_\mathrm{B}^0`$ defined in Eq. (8). The response function $`L(t)`$ will play an important role in the numerical QUAPI algorithm.
It turns out that for the description of the parametric dissipative quantum oscillator the solution of the classical deterministic limit ($`\mathrm{}0,T0`$) with $`\omega _c\mathrm{}`$ plays a prominent role. Thus, in Eq. (9) the position operator $`๐ฑ`$ is replaced by the classical coordinate $`x`$ and $`_{t_0}^t\widehat{\gamma }(tt^{})\dot{๐ฑ}(t^{})๐t^{}`$ goes over into $`\gamma \dot{x}(t)`$. Moreover, on the right hand side of Eq. (9), the fluctuations $`๐ช(t)`$ are zero and the initial slip is also omitted, which can be achieved by either a somewhat different choice of the initial conditions than in Eq. (7) or by replacing the coupling coefficients $`c_j`$ in Eq. (2) by $`c_j\mathrm{\Theta }(tt_0^+)`$ so that $`๐_\mathrm{B}`$ and $`๐_\mathrm{B}^0`$ from Eq. (8) coincide at $`t=t_0`$. For convenience we furthermore introduce scaled quantities
$`\begin{array}{ccccccccccc}\stackrel{~}{t}& =& \frac{\mathrm{\Omega }}{2}t,\hfill & & \stackrel{~}{x}(\stackrel{~}{t})& =& \sqrt{M\mathrm{\Omega }/2\mathrm{}}x(t=\frac{2\stackrel{~}{t}}{\mathrm{\Omega }}),\hfill & & \stackrel{~}{\omega }_0& =& \frac{2}{\mathrm{\Omega }}\omega _0,\hfill \end{array}`$
$$\begin{array}{ccccccccccccccc}\stackrel{~}{ฯต}& =& \frac{2}{\mathrm{\Omega }^2}ฯต,\hfill & & \stackrel{~}{\gamma }& =& \frac{2}{\mathrm{\Omega }}\gamma ,\hfill & & \stackrel{~}{T}& =& \frac{2k_B}{\mathrm{}\mathrm{\Omega }}T,\hfill & & \stackrel{~}{\omega }_c& =& \frac{2}{\mathrm{\Omega }}\omega _c.\hfill \end{array}$$
(14)
In the remainder of this paper, we exclusively use dimensionless quantities but omit all the tildes for the sake of better readability. In order to recover the dimensionful quantities, one has to re-introduce tildes wherever it makes sense and then exploit Eq. (14). By substituting $`x(t)=y(t)\mathrm{exp}[\gamma (tt_0)/2]`$ we arrive at an undamped oscillator equation for $`y`$ which is the well-known Mathieu equation
$$\ddot{y}(t)+(\omega _0^2\frac{\gamma ^2}{4}+2ฯต\mathrm{cos}2t)y(t)=0.$$
(15)
Its mathematical properties like stability and instability regions in the parameter space are well known . Nevertheless, there exists no closed analytic expression for the solution and the equation has to be integrated numerically. In the following, we will need two linear independent solutions $`\mathrm{\Phi }_i(t),i=1,2`$, of Eq. (15) belonging to two different sets of initial conditions
$$\begin{array}{ccccccc}\mathrm{\Phi }_1(t_0)& =& 0,\hfill & & \dot{\mathrm{\Phi }}_1(t_0)& =& 1,\hfill \\ \mathrm{\Phi }_2(t_0)& =& 1,\hfill & & \dot{\mathrm{\Phi }}_2(t_0)& =& 0.\hfill \end{array}$$
(16)
They can be determined numerically, e.g. by means of a regular fourth-order Runge-Kutta integration of the Mathieu equation (15).
Let us return to the dissipative quantum parametric oscillator. The quantities of interest are the variances of the position and the momentum operator, i.e.
$`\sigma _{xx}(t)`$ $``$ $`๐ฑ^2(t)๐ฑ(t)^2,`$ (17)
$`\sigma _{xp}(t)`$ $``$ $`{\displaystyle \frac{1}{2}}๐ฑ(t)๐ฉ(t)+๐ฉ(t)๐ฑ(t)๐ฑ(t)๐ฉ(t),`$ (18)
$`\sigma _{pp}(t)`$ $``$ $`๐ฉ^2(t)๐ฉ(t)^2.`$ (19)
Here, the quantum mechanical expectation value is understood as usual as $`=Tr[\rho (t)]`$. By determining the propagator $`๐(t,t_0)=๐ฏ\mathrm{exp}(i_{t_0}^t๐t^{}๐(t^{})/\mathrm{})`$ ($`๐ฏ`$ is the time ordering operator) for the driven dissipative system according to , the reduced density matrix $`๐(t)=Tr_{Bath}(๐(t,t_0)๐(t_0)๐^1(t,t_0))`$ can be calculated analytically. Here, $`๐(t_0)`$ denotes the full density operator at time $`t_0`$ and $`Tr_{Bath}`$ the trace over the bath degrees of freedom. Having obtained the reduced density operator $`๐(t)`$, the quantum mechanical expectation values in Eq. 19) can be evaluated. After some algebra, we find for the dimensionless variances the expressions
$`\sigma _{xx}(t)`$ $`=`$ $`e^{\gamma (tt_0)}\{[\mathrm{\Phi }_2(t){\displaystyle \frac{\gamma }{2}}\mathrm{\Phi }_1(t)]^2\sigma _{xx}^0+2\mathrm{\Phi }_1(t)[\mathrm{\Phi }_2(t){\displaystyle \frac{\gamma }{2}}\mathrm{\Phi }_1(t)]\sigma _{xp}^0+\mathrm{\Phi }_1^2(t)\sigma _{pp}^0\}+\mathrm{\Sigma }_{xx}(t),`$ (21)
$`\sigma _{xp}(t)`$ $`=`$ $`{\displaystyle \frac{1}{2}}\dot{\sigma }_{xx}(t),`$ (22)
$`\sigma _{pp}(t)`$ $`=`$ $`\dot{\sigma }_{xp}(t)+\gamma \sigma _{xp}(t)+[\omega _0^2+2ฯต\mathrm{cos}(2t)]\sigma _{xx}(t)\mathrm{\Sigma }_{pp}(t).`$ (23)
Thereby, we have rectified some minor misprints in and simplified the equations in for $`\sigma _{xp}`$ and $`\sigma _{pp}`$. Here, $`\sigma _{xx}^0,\sigma _{xp}^0`$ and $`\sigma _{pp}^0`$ denote the initial variances of the uncoupled system at time $`t=t_0`$ which depend on the choice of the initial state for the bare system $`๐_\mathrm{S}(t_0)`$. The initial conditions for Eqs. (II) at time $`t=t_0^+`$ are given by
$`\sigma _{xx}(t_0^+)`$ $`=`$ $`\sigma _{xx}^0,`$ (24)
$`\sigma _{xp}(t_0^+)`$ $`=`$ $`\gamma \sigma _{xx}^0+\sigma _{xp}^0,`$ (25)
$`\sigma _{pp}(t_0^+)`$ $`=`$ $`\gamma ^2\sigma _{xx}^02\gamma \sigma _{xp}^0+\sigma _{pp}^0.`$ (26)
The discontinuity of the variances at time $`t_0`$ is a well-known consequence of the initial slip term in Eq. (9); it is due to the factorizing initial condition (7). The first terms in the three equations (II) posses the same form as in the classical case. The specific quantum mechanical features enter via the functions $`\mathrm{\Sigma }_{xx}(t)`$ and $`\mathrm{\Sigma }_{pp}(t)`$, which read
$`\mathrm{\Sigma }_{xx}(t)`$ $`=`$ $`{\displaystyle \frac{\gamma }{\pi }}{\displaystyle _0^{\mathrm{}}}d\omega \omega f_c(\omega ,\omega _c)\mathrm{coth}({\displaystyle \frac{\omega }{2T}})\{\left[{\displaystyle _{t_0}^t}dsG(t,s)\mathrm{exp}({\displaystyle \frac{\gamma }{2}}(ts))\mathrm{cos}(\omega s)\right]^2`$ (29)
$`+\left[{\displaystyle _{t_0}^t}dsG(t,s)\mathrm{exp}({\displaystyle \frac{\gamma }{2}}(ts))\mathrm{sin}(\omega s)\right]^2\},`$
$`\mathrm{\Sigma }_{pp}(t)`$ $`=`$ $`{\displaystyle \frac{\gamma }{\pi }}{\displaystyle _0^{\mathrm{}}}๐\omega \omega f_c(\omega ,\omega _c)\mathrm{coth}({\displaystyle \frac{\omega }{2T}}){\displaystyle _{t_0}^t}๐sG(t,s)\mathrm{exp}({\displaystyle \frac{\gamma }{2}}(ts))\mathrm{cos}(\omega (ts)),`$ (30)
where $`G(t,s)=\mathrm{\Phi }_1(t)\mathrm{\Phi }_2(s)\mathrm{\Phi }_1(s)\mathrm{\Phi }_2(t)`$. While in Eq. (II) a general form of the cut-off function $`f_c(\omega ,\omega _c)`$ with $`\omega _c\omega _0`$ is kept, the analytic solution (II) is based on the assumption of a strictly Ohmic classical dynamics ($`\omega _c\mathrm{}`$) in Eq. (10). The consequence of this assumption is the discontinuity at $`t=t_0`$ in Eq. (26) when the system-bath-interaction is switched on instantaneously. A finite cut-off in the spectral density $`J(\omega )`$ in the damping kernel (10) would induce a smoothend time evolution of the variances (II) close to $`t=t_0`$ on a time scale $`\omega _c^1`$.
The relations (II, II) are evaluated by standard numerical methods. The efficiency is improved if one applies Floquetโs theorem for the fundamental solutions $`\mathrm{\Phi }_j(s)`$. Then, the periodic part of the Floquet solutions can be expanded in a Fourier series and the integrations over the intermediate times $`s`$ in Eq. (II) can be performed analytically. Finally, the remaining $`\omega `$-integrations and the sum over the Fourier modes can be readily carried out.
## III Numerical solution with real-time path integrals
In the following section, we recapitulate the essentials of the QUAPI algorithm. Further details can be found in the original works by Makri and Makarov . In order to describe the dynamics of the system of interest it is sufficient to consider the time evolution of the elements of the reduced density matrix which reads in position representation
$`\rho (x_f,x_f^{};t_f)`$ $`=`$ $`Tr_{Bath}x_f\mathrm{\Pi }_jq_j|๐(t_f,t_0)๐(t_0)๐^1(t_f,t_0)|x_f^{}\mathrm{\Pi }_jq_j^{},`$ (31)
$`๐(t_f,t_0)`$ $`=`$ $`๐ฏ\mathrm{exp}\left\{i/\mathrm{}{\displaystyle _{t_0}^{t_f}}๐(t^{})๐t^{}\right\}.`$ (32)
Here, $`๐ฏ`$ denotes the chronological operator, $`๐(t_0)`$ the full density operator at the initial time $`t_0`$ and $`Tr_{Bath}`$ the partial trace over the harmonic bath oscillators $`q_j`$. Due to our assumption that the bath is initially at thermal equilibrium and decoupled from the system, $`๐(t_0)`$ becomes the product of the initial system density operator $`๐_\mathrm{S}(t_0)`$ and the canonical bath density operator at temperature $`T`$, see Eq. (7). Then, the partial trace over the bath can be performed and the reduced density operator be rewritten according to Feynman and Vernon as
$$\rho (x_f,x_f^{},t_f)=dx_0dx_0^{}G((x_f,x_f^{},t_f;x_0,x_0^{},t_0)\rho (x_0,x_0^{},t_0),$$
(33)
with the propagator $`G`$ given by
$$G(x_f,x_f^{},t_f;x_0,x_0^{},t_0)=๐x๐x^{}\mathrm{exp}\left\{\frac{i}{\mathrm{}}\left(S_\mathrm{S}[x]S_\mathrm{S}[x^{}]\right)\right\}_{FV}[x,x^{}].$$
(34)
$`S_\mathrm{S}[x]`$ is the classical action functional of the system-variable $`x`$ along a path $`x(t)`$ and $`_{FV}[x,x^{}]`$ denotes the Feynman-Vernon influence functional
$$_{FV}[x,x^{}]=\mathrm{exp}\left\{\frac{1}{\mathrm{}}_{t_0}^{t_f}๐t_t^{t_f}๐t^{}\left[x(t^{})x^{}(t^{})\right]\left[\eta (t^{}t)x(t)\eta ^{}(t^{}t)x^{}(t)\right]\right\},$$
(35)
with the integral kernel
$$\eta (t)=L(t)+i\delta (t)\frac{2}{\pi }_0^{\mathrm{}}\frac{d\omega }{\omega }J(\omega )$$
(36)
and the autocorrelation function $`L(t)`$ given in Eq. (13). As usual, the restriction to paths that satisfy the boundary conditions $`x_0(t_0)=x_0`$, $`x_f(t_f)=x_f`$ and similarly for $`x^{}(t)`$ is understood implicitly in Eq. (34). Likewise, the dependence of the density operator $`๐`$ in Eq. (33) on the initial time $`t_0`$ and on $`๐_\mathrm{S}(t_0)`$ has been dropped.
To make the equations numerically tractable, we discretize $`t_ft_0`$ into $`N`$ steps $`\mathrm{\Delta }t`$, such that $`t_k=t_0+k\mathrm{\Delta }t`$ and split the full propagator over one time step $`๐(t_{k+1},t_k)`$ in Eq. (32) according to the Trotter formula symmetrically into a system and an environmental part:
$`๐(t_{k+1},t_k)`$ $``$ $`\mathrm{exp}(i๐_\mathrm{B}\mathrm{\Delta }t/2\mathrm{})๐_\mathrm{S}(t_{k+1},t_k)\mathrm{exp}(i๐_\mathrm{B}\mathrm{\Delta }t/2\mathrm{}),`$ (37)
$`๐_\mathrm{S}(t_{k+1},t_k)`$ $`=`$ $`๐ฏ\mathrm{exp}\left\{{\displaystyle \frac{i}{\mathrm{}}}{\displaystyle _{t_k}^{t_{k+1}}}๐t^{}๐_\mathrm{S}(t^{})\right\}.`$ (38)
The symmetric splitting of the propagator in Eq. (38) causes an error proportional to $`\mathrm{\Delta }t^3`$. This error will be studied in detail in section IV below. The short-time propagator $`๐_\mathrm{S}`$ of the bare system is numerically evaluated by means of a RungeโKutta scheme with adaptive step-size control. Exploiting the approximation (38), the propagator in the position representation now factorizes as
$$x\mathrm{\Pi }_jq_j|๐(t_{k+1},t_k)|x^{}\mathrm{\Pi }_jq_j^{}x|๐_\mathrm{S}(t_{k+1},t_k)|x^{}\underset{j}{}q_j|e^{i๐_j(x)\mathrm{\Delta }t/2\mathrm{}}e^{i๐_j(x^{})\mathrm{\Delta }t/2\mathrm{}}|q_j^{},$$
(39)
where the $`๐_j(x)`$ are defined in Eq. (2). By exploiting this approximation and performing the partial trace over the bath modes in Eq. (32), one recovers Eq. (33), but now with a discretized version of the propagating function (34), i.e.,
$`\rho (x_f,x_f^{};t_f)`$ $`=`$ $`{\displaystyle ๐x_0\mathrm{}๐x_N๐x_0^{}\mathrm{}๐x_N^{}\delta (x_f^{}x_N^{})\delta (x_fx_N)}`$ (43)
$`\times x_N|๐_\mathrm{S}(t_f,t_f\mathrm{\Delta }t)|x_{N1}\mathrm{}x_1|๐_\mathrm{S}(t_0+\mathrm{\Delta }t,t_0)|x_0`$
$`\times x_0|๐_\mathrm{S}(t_0)|x_0^{}x_0^{}|๐_\mathrm{S}^1(t_0+\mathrm{\Delta }t,t_0)|x_1^{}\mathrm{}x_{N1}^{}|๐_\mathrm{S}^1(t_f,t_f\mathrm{\Delta }t)|x_N^{}`$
$`\times _{FV}^{(N)}(x_0,x_0^{},\mathrm{},x_N,x_N^{}).`$
Here, $`_{FV}^{(N)}(x_0,\mathrm{},x_N^{})`$ is the discrete FeynmanโVernon influence functional (35) where the paths $`x(t)`$ and $`x^{}(t)`$ consist of constant segments $`x_k`$ and $`x_k^{}`$, respectively, within each time interval $`t_k\frac{1}{2}\mathrm{\Delta }t<t_k<t_k+\frac{1}{2}\mathrm{\Delta }t`$ and can be rewritten in the form
$$_{FV}^{(N)}(x_0,\mathrm{},x_N^{})=\mathrm{exp}\left\{\frac{1}{\mathrm{}}\underset{k=0}{\overset{N}{}}\underset{k^{}=k}{\overset{N}{}}[x_k^{}x_k^{}^{}][\eta _{k^{}k}x_k\eta _{k^{}k}^{}x_k^{}]\right\}.$$
(44)
The coefficients $`\{\eta _{k^{}k}\}`$ are closely related to their continuous time counterpart $`\eta (t)`$ in Eq. (36). Their explicit form is lengthy and not very illuminating for our purposes; their detailed form can be looked up in Ref. .
To make further progress, it is necessary to approximately break the influence kernel $`_{FV}^{(N)}(x_0,\mathrm{},x_N^{})`$ in Eq. (44) into smaller pieces. To this end, Makri and Makarov use the fact that the real part of the integral kernel $`L_R(t)`$ typically exhibits a pronounced peak at $`t=0`$, and quickly approaches $`0`$ for $`t\pm \mathrm{}`$. The decay to zero depends naturally on the choice of the cut-off function $`f_c(\omega ,\omega _c)`$, see Eq. (4). This suggests the truncation of $`\eta (t)`$ after a certain number $`K`$ of time steps $`\mathrm{\Delta }t`$ and, correspondingly, to neglect $`\eta _{k^{}k}`$ if $`k^{}>k+K`$, i.e.
$$_{FV}^{(N)}(x_0,\mathrm{},x_N^{})\underset{k=0}{\overset{N}{}}\underset{k^{}=k}{\overset{\mathrm{min}\{N,k+K\}}{}}\mathrm{exp}\left\{\frac{1}{\mathrm{}}[x_k^{}x_k^{}^{}][\eta _{k^{}k}x_k\eta _{k^{}k}^{}x_k^{}]\right\}.$$
(45)
In doing so, we approximate $`L(t)`$ by zero for $`t>K\mathrm{\Delta }t`$, cf. Eqs. (13, 36). Of course, this truncation induces an error in the final result which has to be handled with care. The error becomes increasingly less important for increasing temperatures since then, the bath-induced correlations fall off increasingly faster. In other words, for higher temperatures the width of the response function $`L(t)`$ decreases. In the other limit of decreasing temperature however, the number $`K`$ of relevant time-intervals is increasing and in the limit of zero temperature $`T=0`$, it is well known that the response function $`L(t)`$ falls off only algebraically for $`t\pm \mathrm{}`$. Nevertheless, we will see that this approach allows to deal with quite low temperatures and produces qualitative agreement with analytic solutions.
The next goal is to approximate the spatially continuous integrals in Eq. (43) in terms of finite sums. To this end, Makri and Makarov perform a transformation into a basis given by the energy eigenstates $`|\varphi _m`$ of the bare system Hamiltonian $`๐_\mathrm{S}(t_r)`$ (1), but with the driving term clamped to an appropriate but fixed (!) reference time $`t_r`$, i.e.
$$๐_\mathrm{S}(t_r)|\varphi _m=E_m|\varphi _m,m=1,2,\mathrm{}.$$
(46)
$`E_m`$ denotes the energy eigenvalues of the static system Hamiltonian $`๐_\mathrm{S}(t_r)`$. In certain cases, symmetry properties suggest the choice of an appropriate $`t_r`$. Here, we choose the unperturbed harmonic oscillator as a reference configuration. This means for our choice of driving to use $`t_r=\pi /4`$, so that $`\mathrm{cos}(2t_r)=0`$ in Eq. (15). Reintroducing now the thermal bath but restricting ourselves to small-to-moderate temperatures $`T`$, the thermal occupation of high energy levels $`E_m`$ is expected to be negligible. This argument suggests that the $`|\varphi _m`$ provide a well adapted basis admitting a fast convergent truncation scheme. In other words, we may approximately project the dynamics onto the Hilbert subspace spanned by the first few energy eigenstates $`|\varphi _m`$, $`m=1,\mathrm{},M`$, corresponding to an approximate decomposition of the identity operator $`๐\mathrm{\Sigma }_{m=1}^M|\varphi _m\varphi _m|`$. Before doing so, we perform one more unitary transformation within that $`M`$-dimensional Hilbert space such that the position operator becomes diagonal \[discrete variable representation, DVR \]:
$$|u_m=\underset{m^{}=1}{\overset{M}{}}R_{mm^{}}|\varphi _m^{},u_m|๐ฑ|u_m^{}=x_m^{DVR}\delta _{mm^{}},m,m^{}=1,\mathrm{},M.$$
(47)
Exploiting the approximate decomposition of the identity $`๐\mathrm{\Sigma }_{m=1}^M|u_mu_m|`$ and the truncation of the bath-induced correlations in Eq. (45), it is a matter of straightforward but tedious manipulations โ starting from Eq. (43) โ to arrive at the final form of the QUAPI recursion scheme. In particular, the integrals in Eq. (43) turn into finite sums due to the transformation (47) into the DVR-basis. We do not present the detailed form here and refer the reader again to the original literature .
The above introduced restriction to a finite dimensional subspace induces an error in the evaluation of the reduced density matrix. However, as we will also discuss below, this error behaves in a controlled way if the relevant parameters such as the temperature and the damping are chosen in a moderate regime. This means that for increasing temperature increasingly more DVR-states are necessary to describe the dynamics appropriately. Note that in this regime however, the number $`K`$ of the relevant memory time steps is decreasing. In the opposite limit of decreasing temperature, the number $`M`$ of relevant basis states can be chosen rather small. In this low-temperature limit the number $`K`$ of memory time steps can therefore be increased. Moreover, we note that the restriction of the dynamics (at long times) to the $`M`$-dimensional Hilbert subspace is not allowed for systems with an inherent diverging dynamics. This is also seen in our example of the parametrically driven dissipative quantum harmonic oscillator for a parameter choice in an instability region of the Mathieu equation (15), see section IV below.
The efficiency of the QUAPI algorithm is based on the choice of the two free parameters $`M`$ (the number of basis states) and $`K`$ (the length of the memory). The numerical objects that one has to deal with are arrays of size $`M^{2K+2}`$ and $`M^{2K}`$. In practice, the calculations have been performed on conventional IBM RS/6000 workstations (43P-260 and 3CT). The computation time for an iteration over a typical time span $`[0,40]`$ depends strongly on the chosen parameters. It ranges from several milliseconds for $`M=3,K=2`$ (program size 7 MB) over several seconds for $`M=3,K=4`$ (program size 8 MB) up to several hours for $`M=5,K=4`$ (program size 176 MB). The strongly limiting factor is the program size since the size of the arrays grows exponentially with $`K`$. E.g., the parameter combination $`M=4,K=6`$ leads to too large arrays and cannot be treated by standard programming techniques. In practice, with the choice $`M=6,K=3`$ or $`M=5,K=4`$, we already are at the upper limit of the QUAPI algorithm.
## IV Results
We proceed in reporting our results for the specific example of the parametrically driven dissipative quantum harmonic oscillator. With the reduced density matrix (43) at hand, we can calculate the variances (19) within the QUAPI algorithm and compare them with the analytic predictions (II,II). Most of the figures contain results for rather extreme parameter values, e.g. low temperature and large driving amplitude, in order to show that the QUAPI algorithm performs satisfactorily also in these limits. For more moderate choices of the parameters, the agreement (not shown) between numerical and analytic results is much better.
Our main goal is to study the dependence of the variances (19) on the QUAPI parameters $`M,K`$ and $`\mathrm{\Delta }t`$. For finite $`M`$ and $`K`$, the deviation increases proportional to $`\mathrm{\Delta }t^3`$ due to the Trotter splitting in Eq. (38) with increasing $`\mathrm{\Delta }t`$. For decreasing $`\mathrm{\Delta }t`$, the Trotter error decreases but the error made by the memory truncation in Eq. (45) starts to dominate since more and more bath correlations are neglected. Thus, the overall error increases again. In between there exists an โoptimal time step of least dependenceโ, where the quantities are least sensitive to variations of $`\mathrm{\Delta }t`$. This represents the โprinciple of minimal sensitivityโ for the optimal choice of the time step $`\mathrm{\Delta }t`$ for the QUAPI algorithm (see also ). For $`M`$ finite and $`K\mathrm{}`$, the result would be independent of $`\mathrm{\Delta }t`$ for small $`\mathrm{\Delta }t`$ since the Trotter error would vanish and also the finite-memory error would not exist.
The choice of $`M`$ and $`K`$ should be adapted to the chosen bath parameters. In the case of no driving, if the temperature is low, only few energy eigenstates are required, i.e. $`M`$ may be chosen small. However, low temperature induces long-range bath correlations. Therefore, the memory length $`K`$ has to be assumed large. The opposite holds true in the other limit of high temperature. In the case of driving, the number $`M`$ of basis states is more important compared to the undriven case, since the variances oscillate strongly and higher lying energy states are excited. The memory length $`K`$ has to be reduced instead if one is interested in the oscillation amplitudes. However, for the mean value of the variances, the total memory length $`K`$ is again more important and should be maximized (see below).
We shall choose two representative parameter sets for our considerations. Since the memory in Eq. (44) is truncated in the QUAPI algorithm according to Eq. (45), the crucial parameters are the temperature $`T`$ and the damping strength $`\gamma `$. The relatively high temperature $`T=1.0`$ and the small damping $`\gamma =0.1`$ form the first parameter set (High temperature โ weak damping). For this choice, the numerical results are expected to agree well with the analytic results because large $`T`$ suppresses the long-time memory contributions in Eq. (44) and additionally, a small $`\gamma `$ diminishes the influence of the bath correlations (12). Our second parameter set is given by $`T=0.1,\gamma =1.0`$ (Low temperature โ strong damping). In this case, long-range bath correlations (12) play a major role and the truncation of them will induce an error which will be larger than in the case of a high temperature and weak damping. For intermediate parameter regimes, we find no qualitative differences.
In all our calculations, we set $`t_0=0`$ and choose as the initial state the ground-state of the maximally curved (i.e $`\omega _0^2\omega _0^2+2ฯต)`$ harmonic oscillator, i.e. $`\rho _S(t_0=0)=|00|`$. The corresponding initial variances in Eq. (II) readily follow as $`\sigma _{xx}^0=1/(2\sqrt{\omega _0^2+2ฯต})`$, $`\sigma _{xp}^0=0`$ and $`\sigma _{pp}^0=\sqrt{\omega _0^2+2ฯต}/2`$. Our standard choice for the cut-off function will be the exponential cut-off (5), if nothing else is stated. Furthermore, we always choose the dimensionless curvature $`\omega _0=1.0`$ in order to have a rather small separation of the energy levels in the undriven oscillator. This induces a high sensitivity on the number $`M`$ of basis states since the higher lying states are then easily populated thermally or by driving induced transitions. The choice of a larger $`\omega _0`$ would be more in favour of the numerical algorithm.
### A High temperature โ weak damping (no driving)
First, we consider the undriven case $`ฯต=0`$. Fig. 1 depicts the results for a high temperature $`T=1.0`$ and small friction $`\gamma =0.1`$. Here and in the following, we use the dimensionless quantities which have been introduced in Eq. (14). Moreover, $`\omega _0=1.0`$ and $`\omega _c=50.0`$. We find very good agreement with the analytic solution for the variances. The initial transient oscillations are reproduced and the asymptotic values for long times as well. The initial jump of $`\sigma _{xp}(t)`$ (of Eq. (26)) is still visible, while the jump of $`\sigma _{pp}(t)`$ is proportional to $`\gamma ^2`$ and is not visible on this plot.
To be able to study the dependence of the QUAPI algorithm on the parameters $`M,K`$ and $`\mathrm{\Delta }t`$ we consider the asymptotic values of the variances at long times. It is clear from Eq. (II) that $`\sigma _{xp}(\mathrm{})`$=0, so we focus in Fig. 2 on the two non-trivial variances $`\sigma _{xx}(\mathrm{})`$ and $`\sigma _{pp}(\mathrm{})`$. The qualitative dependence of both variances on the time step $`\mathrm{\Delta }t`$ is always similar: The deviation increases with increasing $`\mathrm{\Delta }t`$ due to the error proportional to $`\mathrm{\Delta }t^3`$ in the Trotter splitting in Eq. (38). For decreasing $`\mathrm{\Delta }t`$, this error decreases and the โfinite-Kโ-error takes over. The relevant $`\mathrm{\Delta }t`$-value on which we focus in the following is the one for which the numerical result varies the least, i.e. the minima in the curves in Fig. 2 (โprinciple of minimal sensitivityโ ).
The left column of Fig. 2 confirms that for a fixed memory length $`\mathrm{\Delta }tK`$, a smaller time step $`\mathrm{\Delta }t`$ induces a smaller Trotter-error whereas the finite-K-error remains roughly the same. While for a fixed $`M`$ (left column in Fig. 2) QUAPI tends to underestimate the analytic result as $`K`$ increases, a fixed $`K`$ and growing $`M`$ (right column) leads to an opposite trend, suggesting that indeed the analytic result will be approached best when both $`M`$ and $`K`$ become large (at a plateau $`\mathrm{\Delta }t`$-value tending towards zero).
### B Low temperature โ strong damping
#### 1 No driving
Fig. 3 depicts the time-dependence of the variances in satisfactory agreement with the analytic result. The initial jumps of $`\sigma _{xp}(t)`$ and of $`\sigma _{pp}(t)`$ are more pronounced in this strong damping case since the jumps are proportional to $`\gamma `$ and $`\gamma ^2`$ (see Eq. (26)). The deviations in the transient behavior are due to the assumption of a strictly Ohmic classical dynamics (infinite cut-off $`\omega _c`$) in the analytic solution, see the discussion at the end of section II. They become more pronounced for low temperatures and strong friction because this assumption induces deviations in the short-time evolution of the variances on a time-scale $`\omega _c^1`$. The bath-induced long-range memory at this low temperature carries the deviations over the whole range of the transient dynamics. The fact that the memory length $`K`$ is decisive for this low temperature is confirmed by the dashed-dotted line.
The dependence of the asymptotic values $`\sigma _{xx}(\mathrm{})`$ and $`\sigma _{pp}(\mathrm{})`$ on the QUAPI parameters is shown in Fig. 4. The number $`M`$ of basis states is not so important, while the memory length $`K`$ is decisive. Again, the analytic prediction is correctly approached when both $`M`$ and $`K`$ are increased.
#### 2 With driving
Fig. 5 demonstrates for a small driving amplitude reasonable agreement with the analytics. The long-memory parameter set with $`K=6`$ hits best the asymptotic mean value, but the oscillation amplitudes and frequencies are obtained best by the choice of a large $`M=5`$. In comparison to the undriven case ($`ฯต=0`$) the time averaged variances are almost unchanged (Figs. 4 and 6) while the time-resolved behavior (Figs. 3 and 5) displays notable differences.
Fig. 7 depicts the time evolution for the relatively large driving amplitude. As expected, for strong driving, a large number $`M`$ of basis states are required to describe the oscillations correctly. The averaged asymptotic values $`\overline{\sigma }_{xx}(\mathrm{})`$ and $`\overline{\sigma }_{pp}(\mathrm{})`$ are plotted in Fig. 8. Since the strong driving mixes high energy eigenstates, the results are considerably more sensitive to the choice of $`M`$ than for weak driving (Fig. 6 upper right panel). However, the same argumentation applies like in the undriven case (see Fig. 4). Considering the rather extreme parameters (small level-spacing, strong driving, low temperature, strong damping) the agreement with the analytic results is still satisfactory.
### C Diverging dynamics and dependence on the cut-off $`\omega _c`$
Fig. 9 shows $`\sigma _{xx}(t)`$ for parameters belonging to an instability region of the Mathieu oscillator (15) , i.e. the variances for the driven quantum harmonic oscillator diverge for long times. Since the QUAPI algorithm is restricted to a (finite) $`M`$-dimensional Hilbert subspace it cannot reproduce such an asymptotic divergence.
The last issue we address is the dependence of the dynamics on the cut-off parameter $`\omega _c`$ and on the explicit shape of the cut-off function (5,6). First, we keep an exponential cut-off but choose a smaller cut-off frequency $`\omega _c`$. It is well known that for the (undriven) quantum harmonic oscillator $`\sigma _{pp}(\mathrm{})`$ diverges with $`\omega _c`$, while $`\sigma _{xx}(\mathrm{})`$ is asymptotically independent of $`\omega _c`$. In Fig. 10, we choose the โworstโ case (i.e. low temperature and strong damping) without driving and decrease the cut-off to $`\omega _c=10.0`$. Compared to Fig. 3, the value of $`\sigma _{xx}(\mathrm{})`$ is indeed practically unchanged while $`\sigma _{pp}(\mathrm{})`$ has notably decreased.
Fig. 11 shows results for a step-like cut-off (6). First, we observe that mainly the short-time behavior of the relaxation process is affected. Clearly, QUAPI with its restriction to only a few energy eigenstates cannot reproduce the transient high-frequency oscillations of $`\sigma _{pp}(t)`$. Second, we note that a step-like cut-off affects the decay of the response function $`L(t)`$ from Eq. (13) for $`t\mathrm{}`$. The real/imaginary part of $`L(t)`$ decays qualitatively like an algebraically damped cos/sin-function. While this might suggest a strong dependence of the QUAPI results on the memory length, we actually find a rather weak dependence since the agreement between numeric and analytic results in Fig. 11 is not considerably worse than in Fig. 3. This means that the memory truncation in Eq. (45) is in fact not very sensitive to the choice of the cut-off function $`f_c(\omega ,\omega _c)`$ as long as one is not interested in the detailed short-time behavior.
## V Conclusions
We have studied the dependence of the QUAPI algorithm on its three numerical parameters, namely the time-step $`\mathrm{\Delta }t`$, the number $`M`$ of basis states, and the memory length $`K`$. As a test system, we have used the analytically solvable dissipative quantum harmonic oscillator and its parametrically driven generalization. The comparison shows a decent agreement of the approximative numerical result with the analytic solution, even in the case with driving. This means that a spatially continuous system can be described reasonably well by taking only a few basis states and a finite memory length into account. For low temperatures and weak-to-moderate driving, the number $`M`$ of basis states has to be chosen small and the memory length $`K`$ large, while in the opposite regime of high temperature, $`M`$ has to be large but $`K`$ may be chosen small. In both cases, satisfactorily large $`M`$ and $`K`$ values are still numerically feasible. For strong driving, the deviations increase but the QUAPI results are still in qualitative agreement with the analytic predictions.
Our findings demonstrate the reliability of the QUAPI algorithm even in driven, spatially continuous systems and not only in finite, discrete dissipative quantum systems such as the spin-Boson-system. Therefore, the QUAPI algorithm may become a standard procedure for simulating open quantum systems in the presence of a novel class of time-dependent, not necessarily periodic driving fields. This technique is especially interesting for the study of decoherence in interacting two-level-systems processing quantum bits. There, the quantum gate operation prescribes the time-dependence of the external control fields which may exhibit a complex non-periodic time-dependence.
## Acknowledgement
This work has been supported by the Deutsche Forschungsgemeinschaft Grant No. HA 1517/19-1 (P.H., M.T.), in part by the Sonderforschungsbereich 486 of the Deutsche Forschungsgemeinschaft (P.H.) and by the DFG-Graduiertenkolleg 283. |
warning/0001/quant-ph0001045.html | ar5iv | text | # Quantum key distribution relied on trusted information center
## Abstract
Quantum correlation between two particles and among three particles show nonclassic properties that can be used for providing secure transmission of information. In this paper, we propose two quantum key distribution schemes for quantum cryptographic network, which use the correlation properties of two and three particles. One is implemented by the Greenberger-Horne-Zeilinger state, and another is implemented by the Bell states. These schemes need a trusted information center like that in the classic cryptography. The optimal efficiency of the proposed protocols are higher than that in the previous schemes.
PACS: 03.67.Dd, 03.65.Bz, 03.67.-a
I. Introduction
Since the first finding that quantum effects may protect privacy information transmitted in an open quantum channel by S.Wiesner , and then by C.H.Bennett and G.Brassard , a remarkable surge of interest in the international scientific and industrial community has propelled quantum cryptography into mainstream computer science and physics. Furthermore, quantum cryptography is becoming increasingly practical at a fast pace. Quantum cryptography is a field that combines quantum theory with information theory. The goal of this field is to use the law of physics to provide secure information exchange, in contrast to classical methods based on (unproven) complexity assumption. Current investigations of quantum cryptography involve three aspects: quantum key distribution (QKD) \[3-19\], quantum secret sharing , and quantum bit commitment and its application \[22-23\]. In particular, quantum key distribution became especially important due to technological advances which allow their implementation in laboratory. The first quantum key distribution prototype, working over a distance of 32 centimeters in 1989, was implemented by means of laser transmitting in free space . Soon, experimental demonstrations by optical fibber were set up . Now the transmission distance is extended to more than 30Km in the fiber , and more than 205m in the free space .
Quantum key distribution is defined as a procedure allowing legitimate two (multi-) users of communication channel to establish exact two (multi-) copies, one copy for each user, of a random and secret sequence of bits. Quantum key distribution employs quantum phenomena such as the Heisenberg uncertainty principle and the quantum correlation to protect distributions of cryptographic keys. QKD is a technique that permits two (or multi-) parties, who share no secret information initially, to communicate over an open channel and to establish between themselves a shared secret sequence of bits. The presented QKD protocols are provably secure against eavesdropping attack, in that, as a matter of fundamental principle, the secret data can not be compromised unknowingly to the legitimate users of the channel. Several quantum key distribution protocols have been proposed, all these protocols can be classed into two kinds. i) The point-to-point (two parties) quantum key distribution (pQKD). Three main protocols of these are the BB84 protocol , B92 protocol and EPR protocol \[5-7\]. ii) The networking quantum key distribution (nQKD), e.g., the time-reserved EPR protocol . The physical implementation may be refer to Townsendโs works .
For the pQKD scheme, the first quantum key distribution scheme, i.e., the Bennett Brassard (BB84) scheme, was presented a decade ago. It is implemented by the four states $`\{|,|,|,|\}`$, where any of the two states $`\{|,|\}`$ and any of the two states $`\{|,|\}`$ are non-commuted, $`\{|,|\}`$ may be any orthogonal states of two-dimensional Hilbert space. Its security is warranted by the uncertainty principle of quantum mechanics. In 1992, Bennett devised another protocol, i.e., the B92 protocol, which is based on the transmission of nonorthogonal quantum states. This protocol uses any two nonorthogonal states to implement the QKD. Its security relies on the no-cloning of unknown two-nonorthogonal states. A further elegant scheme has been proposed by Ekert, which is implemented by the Einstein-Podolsky-Rosen (EPR) pairs . It is called the EPR protocol which relies on the violation of the Bell inequalities to provide the secret security. Consider the two-particle correlation, Bennett et al presented a modified version, in which the security is warranted by the quantum correlation.
Recently, Biham, Huttner, and Mor proposed a nQKD protocol, called time-reserved EPR protocol, in which users prestore quantum states in a trusted center, where their quantum states are preserved using quantum memories. The main procedures are as follows: users store quantum states in quantum memories, kept in a transmission center. Upon request from two users, the center uses two-bit gates to project the product state of two noncorrelated particles (one from each user) onto a fully entangled state. As a result, the two users can share a secret bit, which is unknown even to the center. The time-reserved EPR protocol was proposed to be used in a quantum cryptographic networking (QCN). The implementation of the time-reserved EPR scheme needs four particle for obtaining one qubit. One may ask that whether the nQKD scheme may be implemented by three or two particle or not? In this paper we propose several QKD protocols that use three or two particles to obtain one qubit. These protocols can also be used in networking QKD.
In this work we suggest two nQKD schemes. The suggested schemes need three parties: the trusted information center and two users, by conventional called Alice and Bob, here the center is trusted. One scheme is implemented by the Greenberger-Horne-Zeilinger (GHZ) triplet state, we call this protocol as GHZ-nQKD protocol, in which the centerโs role is to measure his/her particle (from the GHZ triplet) by the random measurement like that in BB84 protocol, and tell Alice and Bob the measurement results. Another is implemented by the Bell states, we call it as Bell-nQKD protocol. This scheme needs the center to measure the two-particle entanglement system by the Bell operators or the linear combination of Bell operators before Aliceโs and Bobโs measurement and send the users his/her results. By the centerโs assistance, the users can obtain the secret key, the cheating center as well as the eavesdropper (Eve) can not eavesdrop the key.
The paper is outlined as follows: First, in Sec. II, we review the two-particle maximally entangled states, the so-called Bell states and the three-particle maximally entangled states, the so-called GHZ triple state. In addition we investigate the correlation properties of the GHZ triplet and the Bell states. In Sec. III, we propose three protocols, which are implemented by the GHZ triplet. The efficiencies of these protocols are different, but they are more practical, especially the protocol 3. The securities of these protocols are analyzed. In Sec. IV, we propose two protocols implemented by the Bell states. The securities of these protocols are investigated. In Sec. V, we discuss the applications of our protocols in network QKD. Conclusions are presented in Sec VI.
II. The Bell states and the GHZ triplet states
First we review the two and three particles entanglement states. In general, N-particle entanglement states may be written as
$$|\psi =\underset{i=1}{\overset{N}{}}|u_i\pm \underset{i=1}{\overset{N}{}}|u_i^c,$$
(1)
where $`u_i`$ stands for a binary variable $`u_i\{|z+,|z\}`$ and $`u_i^c=1u_i`$, $`|z+`$ and $`|z`$ denote the spin eigenstates, or equivalently the horizontal and vertical polarization eigenstates, or equivalently any two-level system. For $`N=2`$ they reduce to the Bell states and $`N=3`$ and $`N=4`$ they represent the GHZ states. For a general $`N`$ we shall calling them cat states. In this paper, we are interested in the case of $`N=2`$ and $`N=3`$, i.e., the Bell states and the GHZ triplet state.
1. Bell states
Eq.(1) reduces to the Bell states when $`N=2`$
$$|\mathrm{\Psi }^+_c=\frac{1}{\sqrt{2}}(|z+_a|z+_b+|z_a|z_b),$$
(2)
$$|\mathrm{\Psi }^{}_c=\frac{1}{\sqrt{2}}(|z+_a|z+_b|z_a|z_b),$$
(3)
$$|\mathrm{\Phi }^+_c=\frac{1}{\sqrt{2}}(|z+_a|z_b+|z_a|z+_b),$$
(4)
$$|\mathrm{\Phi }^{}_c=\frac{1}{\sqrt{2}}(|z+_a|z_b|z_a|z+_b),$$
(5)
where the subscripts $`c,a,b`$ denote the states for the information center and the two communicators Alice and Bob. These Bell states can be generated from a type-II parametric down-conversion crystal . Define the $`x`$ eigenstates
$$|x+=\frac{1}{\sqrt{2}}(|z++|z),$$
(6)
$$|x=\frac{1}{\sqrt{2}}(|z+|z),$$
(7)
the four Bell states can be rewritten as
$$|\mathrm{\Psi }^+_c=\frac{1}{\sqrt{2}}(|x+_a|x+_b+|x_a|x_b),$$
(8)
$$|\mathrm{\Psi }^{}_c=\frac{1}{\sqrt{2}}(|x+_a|x_b+|x_a|x+_b),$$
(9)
$$|\mathrm{\Phi }^+_c=\frac{1}{\sqrt{2}}(|x+_a|x+_b|x_a|x_b),$$
(10)
$$|\mathrm{\Phi }^{}_c=\frac{1}{\sqrt{2}}(|x_a|x+_b|x+_a|x_b),$$
(11)
As should be noted, for example, the $`|\mathrm{\Psi }^+`$ states give correlated results in both the $`z`$ and $`x`$ bases, but the $`|\mathrm{\Psi }^{}`$ state give correlated results in the $`z`$ basis, but anticorrelated results in the $`x`$ basis. Summarizing these correlated or anticorrelated results of the Bell states $`\{\mathrm{\Psi }^+,\mathrm{\Psi }^{},\mathrm{\Phi }^+,\mathrm{\Phi }^{}\}`$ in the $`z`$ and $`x`$ bases, we get the following table:
Table I. The correlation of Bell states $`\{\mathrm{\Psi }^+,\mathrm{\Psi }^{},\mathrm{\Phi }^+,\mathrm{\Phi }^{}\}`$
| Trent | $`|\mathrm{\Psi }^+`$ | $`|\mathrm{\Psi }^{}`$ | $`|\mathrm{\Phi }^+`$ | $`|\mathrm{\Phi }^{}`$ |
| --- | --- | --- | --- | --- |
| Alice | $`|x+`$ | $`|x+`$ | $`|x+`$ | $`|x`$ |
| Bob | $`|x+`$ | $`|x`$ | $`|x+`$ | $`|x+`$ |
| Alice | $`|x`$ | $`|x`$ | $`|x`$ | $`|x+`$ |
| Bob | $`|x`$ | $`|x+`$ | $`|x`$ | $`|x`$ |
| Alice | $`|z+`$ | $`|z+`$ | $`|z+`$ | $`|z+`$ |
| Bob | $`|z+`$ | $`|z+`$ | $`|z`$ | $`|z`$ |
| Alice | $`|z`$ | $`|z`$ | $`|z`$ | $`|z`$ |
| Bob | $`|z`$ | $`|z`$ | $`|z+`$ | $`|z+`$ |
From Table I it is clear that after the center has projected the two-particle entanglement system onto any of the four Bell states $`\{\mathrm{\Psi }^+,\mathrm{\Psi }^{},\mathrm{\Phi }^+,\mathrm{\Phi }^{}\}`$, the state of any of two particles do not give determined results. For example, if the centerโs measurement basis is $`|\mathrm{\Psi }^+`$, the state of any of two particles may be $`|x+`$,or $`|x`$ with the probability $`\frac{1}{2}`$, or $`|z+`$ or $`|z`$ with the probability $`\frac{1}{2}`$. Even if Alice has measured her particle and announced her measurement basis, anyone, including Bob, can not knows Aliceโs results, because the probability of making error is $`\frac{1}{2}`$. However if the centerโs bases are public announced, Alice knows the Bobโs qubits and vice versa. These properties may be used to distribute the quantum key between Alice and Bob by the assistance of the trusted center.
By the four Bell states (Eq.2-Eq.5) one may obtain other correlated or anticorrelated results. Define a line combination of Bell states as
$$|\psi ^+_c=\frac{1}{\sqrt{2}}(|\mathrm{\Psi }^{}_c+|\mathrm{\Phi }^+_c),$$
(12)
$$|\psi ^{}_c=\frac{1}{\sqrt{2}}(|\mathrm{\Psi }^{}_c|\mathrm{\Phi }^+_c).$$
(13)
One may get
$$\begin{array}{cc}\hfill |\psi ^+_c=& \frac{1}{\sqrt{2}}(|x+_a|z+_b+|x_a|z_b)\hfill \\ & \\ & \frac{1}{\sqrt{2}}(|z+_a|x+_b+|z_a|x_b),\hfill \end{array}$$
(14)
$$\begin{array}{cc}\hfill |\varphi ^{}_c=& \frac{1}{\sqrt{2}}(|x+_a|z_b|x_a|z+_b)\hfill \\ & \\ & \frac{1}{\sqrt{2}}(|z+_a|x_b+|z_a|x+_b),\hfill \end{array}$$
(15)
We note that the set of states $`\{\mathrm{\Phi }^+,\mathrm{\Psi }^{},\varphi ^{},\psi ^+\}`$ have the following correlated or anticorrelated results
Table II. The correlation of states $`\{\mathrm{\Phi }^+,\mathrm{\Psi }^{},\varphi ^{},\varphi ^+\}`$
| Trent | $`|\mathrm{\Phi }^+`$ | $`|\mathrm{\Psi }^{}`$ | $`|\varphi ^{}`$ | $`|\psi ^+`$ |
| --- | --- | --- | --- | --- |
| Alice | $`|x+`$ | $`|x+`$ | $`|x+`$ | $`|x+`$ |
| Bob | $`|x+`$ | $`|x`$ | $`|z`$ | $`|z+`$ |
| Alice | $`|x`$ | $`|x`$ | $`|x`$ | $`|x`$ |
| Bob | $`|x`$ | $`|x+`$ | $`|z+`$ | $`|z`$ |
| Alice | $`|z+`$ | $`|z+`$ | $`|z+`$ | $`|z+`$ |
| Bob | $`|z`$ | $`|z+`$ | $`|x`$ | $`|x+`$ |
| Alice | $`|z`$ | $`|z`$ | $`|z`$ | $`|z`$ |
| Bob | $`|z+`$ | $`|z`$ | $`|x+`$ | $`|x`$ |
This table shows the states $`\{\mathrm{\Phi }^+,\mathrm{\Psi }^{},\varphi ^{},\psi ^+\}`$ also have the correlation properties in $`x`$ and $`z`$ direction. If the center projects the two-particle entanglement system onto any of the four bases $`\{\mathrm{\Phi }^+,\mathrm{\Psi }^{},\varphi ^{},\psi ^+\}`$, and sends respectively Alice and Bob one of two-particle entanglement, Aliceโs and Bobโs particles have yet not a determined results before their measurement. For example, if the center measure the two-particle system using the base $`\varphi ^{}`$, Aliceโs measurement may be $`|x+`$ or $`|x`$ if she measures her particle use $`x`$ basis. Before Alice reveals her measurement bases, anyone can not know Aliceโs results, even if Bob. However, if Alice and Bob know the state measured by the center and their measurement directions are determined, they can judge the qubits each other.
2. GHZ triplet states
Eq.(1) reduces to eight GHZ triplet states for $`N=3`$. In this paper we use the following state
$$|\psi =\frac{1}{\sqrt{2}}(|z+z+z++|zzz).$$
(16)
Suppose the center, Alice and Bob share one particle each from a three-particle entangled GHZ state, then the GHZ state may be represented by
$$|\psi _=\frac{1}{\sqrt{2}}(|z+_c|z+_a|z+_b+|z_c|z_a|z_b),$$
(17)
where the first particle is that of the center, the second that of Alice, and the third that of Bob. Define the $`y`$ eigenstates
$$|y+=\frac{1}{\sqrt{2}}(|z++i|z),$$
(18)
$$|y+=\frac{1}{\sqrt{2}}(|z+i|z),$$
(19)
and using the $`x`$ eigenstates defined in Eq.(6,7), the GHZ triplet state can be rewritten as
$$\begin{array}{cc}\hfill |\psi =& \frac{1}{2}[(|x+|x++|x|x)|x+\hfill \\ & \\ & +(|x+|x+|x|x+)|x],\hfill \end{array}$$
(20)
or
$$\begin{array}{cc}\hfill |\psi =& \frac{1}{2}[(|y+|y+|y|y+)|x+\hfill \\ & \\ & +(|x+|x+|x|x+)|x],\hfill \end{array}$$
(21)
or
$$\begin{array}{cc}\hfill |\psi =& \frac{1}{2}[(|y+|x+|y|x)|y+\hfill \\ & \\ & +(|y+|x++|y|x)|y],\hfill \end{array}$$
(22)
or
$$\begin{array}{cc}\hfill |\psi =& \frac{1}{2}[(|x+|y+|x|y+)|y+\hfill \\ & \\ & +(|x+|y++|x|y)|y].\hfill \end{array}$$
(23)
The above decomposition demonstrates the correlation among three particles. For example, in Eq.(20) if one particle is in the state $`|x+`$ and the second particle is in the state $`|x+`$, the third particle must be in the state $`|x+`$ because of the correlation of the GHZ triplet state. By Eqs.(20-23), one may construct a lock-up table to summarize these properties of GHZ states.
Table III. The correlation results of the GHZ triplet states
| Trent | $`|x+`$ | $`|x`$ | $`|y+`$ | $`|y`$ |
| --- | --- | --- | --- | --- |
| Alice | $`|x+`$ | $`|x+`$ | $`|x+`$ | $`|x+`$ |
| Bob | $`|x+`$ | $`|x`$ | $`|y`$ | $`|y+`$ |
| Alice | $`|x`$ | $`|x`$ | $`|x`$ | $`|x`$ |
| Bob | $`|x`$ | $`|x+`$ | $`|y+`$ | $`|y`$ |
| Alice | $`|y+`$ | $`|y+`$ | $`|y+`$ | $`|y+`$ |
| Bob | $`|y`$ | $`|y+`$ | $`|x`$ | $`|x`$ |
| Alice | $`|y`$ | $`|y`$ | $`|y`$ | $`|y`$ |
| Bob | $`|y+`$ | $`|y`$ | $`|x+`$ | $`|x`$ |
The table III shows several properties of the GHZ triplet state: i) anyone of the three parties, i.e., the center, Alice or Bob, can determine whether the other two participatorsโ results are the same or opposite and also that he (she) will gain no knowledge of what their results actually are, if he (she) knows what measurements have been made by the other two participators (that is $`x`$ or $`y`$). ii) From table III it is clear that allows two parties jointly, but only jointly, to determine which was the measurement outcome of the third party. So if the measurement directions of the three participators are public, the combined results of any two participators can determine what the result of the third partyโs measurement was.
III. GHZ-nQKD protocols
GHZ states has already found a number of uses. They form the basis of a very stringent test of local realistic theories. It was also proposed that they can be used for cryptographic conferencing or for multiparticle generations of superdense coding . In addition, related states can be used to reduce communication complexity. Recently, it was proposed that they can be used for quantum secret sharing and quantum information split . In this paper, we use the GHZ state to distribute quantum key between Alice and Bob by the centerโs assistance. In the following we present our quantum key distribution protocols.
A. The protocols
As discussed in Sec. II, the GHZ state has correlation properties that if only one communicatorโs measurement results is announced, the states of other two particles are still not determined, but two communicatorsโ results can determine the third result. These properties may be used in the QKD relying on a third party. Let us now show how to implement our quantum key distribution scheme by the GHZ state. There are several way to distribute the communicators the key by the centerโs assistance.
Protocol 1
1. The center measures his GHZ particle in the $`x`$ direction and obtains the result $`|x+`$ or $`|x`$
2. The center tell Alice and Bob his measurement results.
3. Alice and Bob make respectively the random measurement on their GHZ particles, either in the $`x`$ or $`y`$ direction.
4. Check the eavesdropping by using the correlation properties of the GHZ states.
5. Alice and Bob compare their bases. If their measurement bases are same, Alice and Bob keep their results, otherwise they discard their results.
6. Alice and Bob obtain the final key by using the data sifting, the error correction and the privacy amplification technologies.
In this protocol, we let the center firstly measures his particle from the GHZ triplet, and only measure it in the $`x`$ direction. The centerโs results will be $`|x+`$ or $`|x`$. After the center has finished the measurement, Alice and Bob measure their particles. This protocol only uses the correlation results in the first and second columns of table III. Of course, the center may randomly measure his GHZ particles, either in the $`x`$ or $`y`$ direction, but the efficiency is low by this way, because these results measured along $`y`$ direction have no use in this protocol and a half particles will be discarded, the efficiency is only 12.5%. The centerโs measurement collapses the GHZ triplet state to be a two-particle system. The state of the two-particle entanglement is not determined, because they may be any of the states
$$|\mathrm{\Psi }^1_{ab}=\frac{1}{\sqrt{2}}(|x+|x++|x|x),$$
$$|\mathrm{\Psi }^2_{ab}=\frac{1}{\sqrt{2}}(|x+|x+|x|x+),$$
$$|\mathrm{\Psi }^3_{ab}=\frac{1}{\sqrt{2}}(|y+|y+|y|y+),$$
$$|\mathrm{\Psi }^4_{ab}=\frac{1}{\sqrt{2}}(|y+|y++|y|y).$$
In step 4, we use the correlation properties of the GHZ states to check the eavesdropping. Having measured their particles, Bob randomly chooses a subset of qubits from his qubits and sends this subset to Alice. Alice compares the corresponding results from the center, Bob and Alice. If these results are correlation results, which are satisfy Eqs.(20-23) or the correlation of three states is in the table III, the results are perfect, otherwise it means eavesdropping or disturbed by noise.
In step 5, Alice and Bob compare their bases. Because Alice and Bob randomly measure their particles either in the $`x`$ or $`y`$ direction, some of their bases are different and some are same. If their bases are different, Aliceโs and Bobโs results are no correlation, thus Alice can not know Bobโs qubits and vice versa, in this case they need to discard these results. However if their bases are same, their results are correlated, Alice and Bob keep these results. So this protocol discards the results that corresponds the different bases.
The raw quantum key distribution is useless in practice because limited eavesdropping may be undetectable, yet it may leak some information, and errors are to be expected even in the absence of eavesdropping. For these reasons, our scheme needs to supplement some classical tools such as the privacy amplification, the error correction and the data sifting, so we use these technologies in our protocol. The implementation of these supplemented classic tools are the same as in the previous documents .
In quantum key distribution some qubits (henceforth $`l`$) will be wasted because of the loss and the inexactitude of equipment, so in order to be left with a key of $`L`$ qubits the center <sup>1</sup><sup>1</sup>1assume the particles are prepared by the center in this paper should prepare $`L^{}>2(L+l)`$. In this case the efficiency is
$$\eta _1=\frac{L}{2(L+l)}<50\%.$$
(24)
This efficiency is larger than that of the time-reserved EPR protocol, which is
$$\eta ^{}=\frac{L}{8(L+l)}<12.5\%.$$
(25)
Protocol 2
1. The center measures his GHZ particle either in the $`x`$ or $`y`$ direction and obtains any of the four states $`\{|x+,|x,|y+,|y\}`$
2. The center tells Alice and Bob his measurement results.
3. Alice and Bob make respectively a random measurement on their GHZ particles, either in the $`x`$ or $`y`$ direction.
4. Checking the eavesdropping like protocol 1.
5. Alice and Bob compare their bases. If the centerโs result is $`|x+`$ or $`|x`$ and their measurement bases are same, or if the centerโs result is $`|y+`$ or $`|y`$ and their measurement bases are different, Alice and Bob keep their results, otherwise they discard the results.
6. Alice makes her results to be consistent with Bobโs results according to table III.
7. Alice and Bob gain the final key by using the data sifting, the error correction and privacy amplification technologies.
The protocol 2 lets the center measure his/her particle either in the $`x`$ or $`y`$ direction. It is stresses that here the centerโs all measurement results are useful. When the centerโs result is the state $`|x+`$ or $`|x`$, Alice and Bob need to keep the results which have the same bases, but if the centerโs result is $`|y+`$ or $`|y`$, the communicators discard the results which have the same bases. The reason is that the results must be correlated or anticorrelated. This step is finshed in the step 5.
It needs to stress that Aliceโs results must be consistent with Bobโs results for getting the raw quantum key, so we have the step 6. By the properties of the GHZ triplet state, Alice (Bob) can judge Bobโs (Aliceโs) results by combining her (his) and the centerโs results. But the table III can not give completely a same results although Alice and Bob can know the qubits each other. For example, when the centerโs and Aliceโs results are respectively $`|y+`$ and $`|x+`$, Bobโs result should be $`|y`$, obviously, Aliceโs and Bobโs results are different. For obtaining a same key, Aliceโs (Bobโs) results need to be consistent with Bobโs (Aliceโs) results. The method is that Alice (Bob) transfers her (his) qubits to binary bits according to Bobโs (Aliceโs) results.
The efficiency of this protocol is the same as the protocol 1. After the center announced his results, Alice and Bob have a possibility of $`1/2`$ to obtain the correct results by the random measurement. Consider the wasted qubits ($`l`$), in order to be left with a key of $`L`$ qubits the center should send $`L^{}>2(L+l)`$, the efficiency
$$\eta _2=\frac{L}{2(L+l)}<50\%.$$
(26)
Protocol 3
1. Alice and Bob make respectively a random measurement on their GHZ particles, either in the $`x`$ or $`y`$ direction.
2. Alice and Bob send their measurement bases ($`x`$ or $`y`$) to the center, but not the qubit values.
3. The center randomly measures his particle according to Aliceโs and Bobโs measurement bases. If both Alice and Bob measure their particle using the same measurement basis, e.g. $`x`$ or $`y`$ direction, the center measures his particle using the $`x`$ measurement basis, otherwise, the center measures his particle using the $`y`$ measurement basis.
4. The center announces his measurement results, which is any of the four states $`\{|x+,|x,|y+,|y\}`$.
5. Check eavesdropping like the protocol 1.
6. Alice and Bob judge the quantum state each other according to table III. While the center announces his results, both Alice and Bob know the centerโs results. Then the Aliceโs (Bobโs) and the centerโs results can jointly determine what is the Bobโs (Aliceโ) measurement outcome.
7. Alice and Bob obtain a sharing key by using the data sifting, the error correction and privacy amplification technologies.
The important point is that Alice and Bob randomly measure their GHZ particles before the centerโs measurement in this protocol, it is different from the protocols 1 and 2, in which the centerโs measurement is completed before Aliceโs and Bobโs measurement. This change improves the efficiency of this protocol, because the center may measure his particle according to Aliceโs and Bobโs measurement bases. Although there are the situations that Aliceโs and Bobโs results are different, all their measurement states are useful. The correlation of the GHZ triplet state lets Alice know Bobโs qubits and vice versa, so Alice and Bob can know the qubits each other in the step 6. Consider the wasted qubits ($`l`$) in the measurement and the loss in the quantum channel, in order to be left with a key of $`L`$ qubits the center should send $`L^{}>(L+l)`$, the efficiency is
$$\eta _3=\frac{L}{(L+l)}<100\%.$$
(27)
B. security analysis
These presented schemes are secure against eavesdropping. Their securities are warranted by the correlation of the GHZ triplet. To see these in a sufficient way, we will consider several possible eavesdropping in the following.
1. The cheating centerโs attacks
The cheating center is impossible to know the quantum key. From the table III it is clear that if the center knows what measurements bases Alice and Bob made (that is, $`x`$ or $`y`$), he can determine whether their results are the same or opposite and also that the center will gain no knowledge of what Aliceโs and Bobโs actually are, because the cheating center will has the probability of 1/2 of making a mistake. If the center makes measurement on the three particles of the GHZ triplet, then send these particles to Alice and Bob, the centerโs measurement will introduce errors in Aliceโs and Bobโs results, thus Alice and Bob can check it like the four-state BB84 protocol. Of course, a cheating center may use the men-in-middle attack to obtain the key $`K_{ca}`$ and $`K_{cb}`$, where $`K_{ca}`$ represents the key between Alice and the cheating center, and $`K_{cb}`$ represents the key between Bob and the cheating center. For preventing this attacks, Alice and Bob may verify their identity using the identity verification technology . This method needs a sharing key between Alice and Bob, however, Alice and Bob, in general, have not sharing key, so this case needs the center to be trustworthy like the key distribution system used in the classic cryptography . So in our scheme we assume the center is trusted, e.g., the key distribution center (KDC) which is often used in the classic cryptography, but here the center process the qubits not the binary bits.
2. Intercept/resend attacks
Let us now consider the intercept/resend attack defined in . Suppose that the eavesdropper, by convention denoted by Eve, has managed to get a hold of Alice and Bobโs key, she then intercepts a communicatorโs (e.g. Alice) particle from the center and send another particle to Alice. In this case, three particles of the center, Bob and the Eve construct a GHZ triplet. However, because the Alice, Bob and the centerโs particles are not the GHZ triplet, there are no correlated or anticorrelated result, Eveโs interception will introduce error and can be detected by Alice and Bob when they check the eavesdropping.
3. The entanglement attacks
The entanglement attacks is no use in our protocol. To show that, Let us assume that the eavesdropper has been able to entangle an ancilla in state $`|A`$ with the GHZ triplet state that Alice and Bob are using. The state describing the state of the GHZ triplet and the ancilla is
$$|\mathrm{\Psi }=\frac{1}{\sqrt{2}}(|z+z+z++|zzz)|A.$$
(28)
By using the $`x`$ and $`y`$ eigenstates and Eq.(20), The eavesdropper get
$$\begin{array}{cc}\hfill U|\mathrm{\Psi }=& \frac{1}{2}(|x+x+|x+|A_1+|xx|x+|A_2\hfill \\ & \\ & +|x+x|x|A_3+|xx+|x|A_4).\hfill \end{array}$$
(29)
where $`U`$ denote the unitary transformation. By projecting the above state onto $`|\varphi =\alpha _1|x+x++\alpha _2|xx+\alpha _3|x+x+\alpha _4|xx+`$, the eavesdropper creates the states
$$|\mathrm{\Psi }_E=\frac{1}{2}(|x+(\alpha _1^{}|A_1+\alpha _2^{}|A_2)+|x(\alpha _3^{}|A_3+\alpha _4^{}|A_4)).$$
(30)
If Eve can gain Aliceโs (Bobโs) qubits, she can obtain the key. However, Eq.(30) shows that Eve can not obtain the results. By the similar method, Eve can not obtain Aliceโs (Bobโs) qubits when the GHZ triplet states satisfy Eqs.(21-23).
IV. Bell-nQKD protocols
The above schemes are efficient, however they need three particles. Can we implement the network QKD scheme only by using two particles? In this section we investigate the two-particle schemes. In the following we show that one can also use the Bell states to implement the above quantum key distribution procedure.
A. protocol
In Sec. II, we see that the two particles of the Bell states or the linear combination of Bell states have correlation properties, they are demonstrated in table I and II. These properties may be used in the QKD relied on a third party. Let us now show how to implement the quantum key distribution by Bell states.
Protocol 4
1. The center prepares a set of two-particle entanglement pairs and projects each pair onto any of the four Bell bases.
2. The center sends respectively Alice and Bob one of the two-particles entanglement and his measurement results.
3. Alice and Bob make respectively the random measurement on their particle, either in the $`x`$ or $`z`$ direction.
4. Alice and Bob check the eavesdropping by using the correlation of Bell states.
5. If their measurement bases are same, Alice and Bob keep their results, otherwise they discard the results.
6. Alice and Bob obtain a sharing key by using the data sifting, the error correction and the privacy amplification technologies.
This scheme is similar to the time-reserved EPR protocol, but there are several important dissimilarities. i) The time-reserved EPR protocol uses four particle and two particles were prestored in a transmission center, where their quantum states are preserved using quantum memories. Our scheme uses two particles and need not the quantum memories. ii) The efficiency of the time-reserved EPR protocol is $`\eta ^{}<12.5\%`$, but the efficiency of our protocol is
$$\eta _4=\frac{L}{2(L+l)}<50\%.$$
(31)
iii) The center only uses results of the singlet states and its correlation properties in the time-reserved EPR protocol, but the center uses all quantum states in our scheme.
We can also use the table II to design a nQKD protocol. The protocol goes as follows
Protocol 5
1. The center prepares a set of two-particle entanglement pairs and projects each pair onto any of the four bases $`\{\mathrm{\Phi }^+,\mathrm{\Psi }^{},\varphi ^{},\psi ^+,\}`$.
2. The center sends respectively Alice and Bob one of the two-particles entanglement and his measurement results.
3. Alice and Bob make respectively the random measurement on their particle, either in the $`x`$ or $`z`$ direction.
4. Check the eavesdropping by using the correlation demonstrated in Table II.
5. Alice and Bob compare their bases. If the centerโs result is one of the Bell states $`\{|\mathrm{\Phi }^+,|\mathrm{\Psi }^{}\}`$ and their measurement bases are same, or if the centerโs result is one of the states $`\{|\psi ^+,|\varphi ^{}\}`$ and their measurement bases are different, Alice and Bob keep their results, otherwise they discard the results.
6. Alice makes her results be consistent with Bobโs results.
7. Alice and Bob obtain a sharing key by using the data sifting, the error correction and privacy amplification technologies.
This protocol is similar to the protocol 2, but there are two dissimilarities: 1) The implementation of protocol 2 uses the GHZ triplet state, which needs three particles to obtain one qubit. The centerโs measurement result is one of the state $`\{|x+,|x,|y+,|y\}`$. But the protocol 5 uses the Bell states which only use two particles, the centerโs results is one of the states $`\{|\mathrm{\Phi }^+,|\mathrm{\Psi }^{},|\psi ^+,|\varphi ^{}\}`$. 2) The methods for checking eavesdropping are different. Protocol 2 uses the correlation of the GHZ states, and here protocol 5 uses the correlation demonstrated in the table II.
According to the protocol 5, we see the efficiency is same as protocol 2:
$$\eta _5=\frac{L}{2(L+l)}<50\%.$$
(32)
B. security analysis
In term of eavesdropping possibilities, protocols 4 and 5 have same security with the EPR protocol. After the center has measured the two-particle entanglement systems by using any Bell operators or the linear combination Bell operators, Alice and Bobโs particles are two-particle entanglement pairs, which is one of the four states $`\{|\mathrm{\Psi }^+,|\mathrm{\Psi }^{},|\mathrm{\Phi }^+,|\mathrm{\Phi }^{},\}`$ or $`\{|\mathrm{\Phi }^+,|\mathrm{\Psi }^{},|\varphi ^{},|\psi ^+,\}`$. It has the same correlation as the EPR pair, this is therefore equivalent to the EPR scheme. So the cheating center as well as the eavesdropper can not eavesdrop the key from the protocols 4 and 5 by the currently eavesdropping technologies, e.g., the intercept/resend attacks, the entanglement attacks etc..
V. Practical applications
Like the time-reserved EPR protocol, ours scheme can also be used in the quantum cryptographic network (QCN), not only in the quantum cryptographic network with a centers and multiusers but also in the QCN of worldwide network of many center. For the QCN of worldwide network of many centers, we can also implement the network QKD between two users by the teleporation scheme. The ways are similar to the time-reserved EPR protocol. As an example, we consider the QCN with a center and multiusers. Assume that there are $`N`$ users denoted by $`u_i,i=1,2,\mathrm{},N`$ and a trusted information center, all users have registered their $`ID`$ in this system at the initial phase. When any user $`u_i`$ wants to connect the other user $`u_j`$ who is in the same QCN, the user $`u_i`$ first tells the center this message. The center verifies the userโs identity by the quantum authentication technology or the classic authentication scheme . If the identity is correct, the center distributes GHZ particles or the two particles of Bell states to users $`u_i`$ and $`u_j`$, then they use the proposed protocol to distribute the key between two users. Here the users $`u_i`$ and $`u_j`$ are arbitrarily chosen.
VI. conclusion
We have show that the quantum correlation between two particles and among three particles can be used for the quantum key distribution relying on a trusted information center. Two schemes are proposed. One scheme is implemented by using the GHZ triplet states, in which three protocols are proposed, these protocols use three particles to obtain one qubit, and have optimal efficiency, especially the protocol 3. The other is implemented by the Bell state, in which two protocols are proposed, these protocols use one Bell particle pairs to obtain one qubit.
In the process of quantum key distribution, the center play an important role, however, the center can not gain any information by any method except for the classic attack, i.e., the men-in-middle attack. For preventing the men-in-middle attack, the presented schemes need the trusted information center, or the users can verify the communicatorsโ identity (in fact all previous QKD protocols, e.g., BB84, B92 and EPR protocol need this requisition in practical application).
Acknowledgments
This research was supported by the Natural Science Foundation (NSF) of China under the grants of No. 69803008
References
1. S. Wiesner, Sigact News, 15, 78 (1983); original manuscript written circa 1970.
2. C. H. Bennett, G. Brassard, S. Breidbart, and S. Wiesner, Advances in Cryptology: Proceedings of Crypto 82, August 1982, Plenum Press, New York, p. 267.
3. C. H. Bennett, and G.Brassard, Advances in Cryptology: Proceedings of Crypto 84, August 1984, Springer - Verlag, p. 475.
4. C. H.Bennett, Phys. Rev. Lett., 68, 3121, (1992).
5. A. K.Ekert, Phys. Rev. Lett., 67, 661, (1991).
6. A. K. Ekert, J. G. Rarity, P. R. Tapster, and G. M. Palma, Phys. Rev. Lett. 69, 1293 (1992).
7. C. H. Bennett, G. Brassard, and N. D. Mermin, Phys. Rev. Lett. 68, 557 (1992).
8. E. Biham, B. Huttner, and T. Mor, Phys. Rev. A, 54, 2651, (1996).
9. C. H. Bennett, F. Bessette, G. Brassard, L. Salvail and J. Smolin, J. Cryptology 5, 3 (1992).
10. A. Muller, J. Breguet, and N. Gisin, Europhysics Lett., vol. 23, no. 6, 383 (1993). J. Breguet, A. Muller, and N. Gisin, J. Modern Optics, 41, 2405, (1994).
11. C. Marand and P. D. Townsend, Optics Letters, 20, 1695 (1995).
12. W. T. Buttler, R. J. Hughes, P. G. Kwiat, et. al., Phys. Rev. A 57, 2379 (1998).
13. P. D. Townsend, J. G. Rarity, and P. R. Tapster, Electronics Letters, 29, 634, (1993).
14. P. D. Townsend, J. G. Rarity, and P. R. Tapster, Electronics Letters, 29, 1291, (1993)..
15. C. A. Fuchs, N. Gisin, R. B. Griffiths, C. S. Niu, and A. Peres, Phys. Rev. A, 56, 1163, (1997).
16. B. A. Slutsky, R. Rao, P. C. Sun, and Y. Fainman, Phys. Rev. A, 57, 2383, (1998).
17. Brandt, Howard E., John M. Meyers, And Samuel J.Lomonaco,Jr., Phys. Rev. A, 56, 4456, (1997).
18. C. Niu, and R. Griffiths, Phys. Rev. A, 58, 4377, (1998).
19. M. Hillery, V. Buzek, and A. Berthiaume, Phys. Rev. A, 59, 1829 (1999).
20. A. Karlsson, M. Koashi, and N. Imoto, Phys. Rev. A, 59, 162 (1999).
21. D. Mayers, Phys. Rev. Lett., 78, 3414 (1997).
22. H. K. Lo and H. F. Chau, Phys. Rev. Lett., 78, 3410 (1997).
23. L. Goldenberg, L. Vaidman, and S. Wiesner, Phys. Rev. Lett., 82, 3356 (1999).
24. J. S. Bell, Physics (Long Island City, N.Y.), 1, 195(1965).
25. A. Einstein, B. Podolsky, and N. Rosen, Phys. Rev. 47, 777 (1935).
26. D. Greenberger, M. A. Horne, and A. Zeilinger, in Bellโs Theorem, Quantum theory, and Conceptions of Universe, edited by M.Kaftos (Kluwer Academic, Dordrecht, 1989); D. Greenberger, M. A. Horne, A. Shimony, and A. Zeilinger, Am. J. Phys. 58, 1131 (1990).
27. S. Bose, V. Vedral, and P. L. Knight, Phys. Rev. A, 57, 822 (1998).
28. P. G. Kwiat, K. Mattle, H. Weinfurther, and A. Zeilinger, Phys. Rev. Lett., 75, 4337 (1995).
29. B. Schneier, Applied cryptographyโProtocols, Algorithms, and Source Code in C, John Wiley & Son, Inc., 1996.
30. G. Zeng and W. Zhang, Phys. Rev. A, 61 no 1, 2000 (In press).
31. G. Zeng, Phys. Lett. A, (Submitted).
32. M. N. Wegman, and J. L. Carter, J. Computer and System Sci., vol. 22, 265 (1981). |
warning/0001/cond-mat0001012.html | ar5iv | text | # Dissipative LandauโZener Tunneling at Marginal Coupling
## Abstract
The LandauโZener transition in a two level system can be suppressed or enhanced by coupling to an environment, depending on the temperature and the environment spectral function. We consider the marginal spectral function, when the dissipation effects are important for arbitrarily slow motion. LandauโZener transition rate demonstrates a non-trivial dependence of the on the โbiasโ, i. e., on the rate of the two energy levels relative motion. The LandauโZener transition is fully suppressed for the values of the bias below a threshold bias set by the coupling strength. Above the threshold, the transition rate for zero temperature is found using the instanton method. At finite temperature, the LandauโZener transition rate has a non-monotonic dependence on the coupling strength, being suppressed at the strong coupling.
Introduction. The LandauโZener problem deals with non-adiabatic transition in a two-state system under external bias. Physical examples include current states in small metallic loops, spin tunneling in magnetic molecules, Andreev states in SNS junctions , slow atomic and molecular collisions, electron transfer in biomolecules. Real-world two-state systems are typically coupled to the environment, and this coupling may lead to different physical effects. This coupling provides dissipation and slows the tunneling down. On the other hand, thermal noise in the environment may cause non-adiabatic transition. The competition between these two effects makes the problem of dissipative LandauโZener tunneling very interesting.
Also, the LandauโZener tunneling is of interest in the context of quantum computing, because it describes the adiabatic flip of a qubit. During the process of the flip, the qubit is in a superposition of the pure โonโ and โoffโ states and thus may be very sensitive to environmental noise. Thus, the problem of dissipative LandauโZener tunneling may be important for the physical implementation of the qubit.
Previous works on the dissipative LandauโZener tunneling treated the problem either perturbatively or phenomenologically . Perturbative treatment, however, is insufficient for adiabatic limit, because it poses too strong limitations on coupling strength. Phenomenological approach is satisfactory only for high temperatures. Also, previous works focused primarily on the case of Ohmic coupling which is critical for equilibrium tunneling. We will argue that the dissipation is relevant for LandauโZener problem when the coupling constant $`\alpha `$ has the same dimension as the velocity of levels relative motion $`\nu `$. (We call the latter bias in the following.) We call this coupling marginal. (A physical example of such coupling is an SNS junction with two Andreev states connected to resistor.) It turns out that the tunneling is blocked if the bias is small: $`\nu <\alpha `$. For larger bias, we derive an instanton solution describing dissipative LandauโZener tunneling at low temperatures and find the tunneling probability:
$$w\mathrm{exp}\left(\frac{\pi \mathrm{\Delta }^2}{\nu \alpha }\right),$$
(1)
where $`\mathrm{\Delta }`$ is the energy gap, $`\nu `$ is the bias, and $`\alpha `$ is the coupling strength. In the high temperature limit ($`T\mathrm{}\mathrm{\Delta }/2\nu `$) we derive master equation which takes into account both thermal noise and dissipation. This equation solves the problem for all temperatures larger than the crossover temperature, and can also incorporate non-equilibrium noise.
Model. It is conventional to represent the two states of LandauโZener model using the spin 1/2 basis. The Hamiltonian is
$$\widehat{H}=\nu t\widehat{\sigma }_z+\mathrm{\Delta }\widehat{\sigma }_x+\widehat{H}_c+\widehat{H}_{env},$$
(2)
where $`\nu `$ is the bias, $`\mathrm{\Delta }`$ is the energy gap, and $`\sigma _{x,z}`$ are Pauli matrices. The term $`\widehat{H}_c`$ describes the coupling to the environment with Hamiltonian $`\widehat{H}_{env}`$.
If the coupling $`\widehat{H}_c`$ is negligible, the model can be solved exactly , and the non-adiabatic transition probability is $`w=\mathrm{exp}(\pi \mathrm{\Delta }^2/\nu )`$. In the adiabatic limit, $`\nu \mathrm{\Delta }^2`$, the โadiabatically frozenโ eigenvalues are:
$$ฯต_\pm (t)=\pm \sqrt{(\nu t)^2+\mathrm{\Delta }^2}.$$
(3)
The characteristic time of tunneling is $`\tau _0=\mathrm{\Delta }/\nu `$.
Following , we model the environment by a set of oscillators with coordinates $`\widehat{x}_i`$ and momenta $`\widehat{p}_i`$:
$$\widehat{H}_{env}=\underset{i}{}\left(\frac{\widehat{p}_i^2}{2}+\frac{\omega _i^2\widehat{x}_i^2}{2}\right),$$
(4)
and couple the environment to the spin linearly:
$$\widehat{H}_c=\widehat{\sigma }_z\widehat{X}(t);\widehat{X}(t)=\underset{i}{}\gamma _i\widehat{x}_i(t).$$
(5)
We consider here only the diagonal coupling which is the main source of dephasing. The effect of environment depends only on the spectral function:
$$J(\omega )=\underset{i}{}\frac{\gamma _i^2}{2\omega _i}\delta (\omega \omega _i).$$
(6)
Without the loss of generality, in the limit of small $`\omega `$ one can consider power-like spectral functions: $`J(\omega )=\alpha _s\omega ^s`$. To estimate the effect of dissipation, we look at the dimensionless ratio $`\eta =\alpha _s/(\nu \tau _0^{s+1})`$, which gives the measure of the dissipation with respect to the bias $`\nu `$. In the adiabatic limit, when $`\tau _0`$ is large, the dissipation is weak ($`\eta 1`$) for $`s>1`$, and strong otherwise. The marginal situation arising when $`J=\alpha /\omega `$ is a subject of this work.
Marginal coupling describes, e. g., singleโchannel SNS junction connected in series to the resistor. There are two Andreev states in the junction. The dynamics of these states is governed by the Hamiltonian
$$\widehat{H}_{SNS}=\frac{\mathrm{\Delta }_0}{\mathrm{cos}}\frac{\varphi (t)}{2}\sigma _z+\mathrm{\Delta }_0\sqrt{1\tau }\mathrm{sin}\frac{\varphi (t)}{2}\sigma _x,$$
(7)
where $`\varphi (t)`$ is the superconducting phase difference across the junction, $`\mathrm{\Delta }_0`$ is the superconducting gap in the leads, and $`\tau `$ is the channel transmission coefficient. Let $`V`$ is voltage drop across the whole circuit, and $`V_R(t)`$ is the voltage drop across the resistor. Then, the phase difference is:
$$\varphi (t)=\frac{2eVt}{\mathrm{}}\frac{2e}{\mathrm{}}V_R(t)๐t.$$
(8)
In the vicinity of the point $`\varphi =\pi `$, where the energy difference of the two states is minimal, one may expand $`H_{SNS}`$ in $`\varphi `$. If there were no resistor, the system then would be described by LandauโZener model. To see what is the effect of the resistor, consider voltage fluctuations. According to quantum Nyquist formula, the voltage $`V_R`$ fluctuates at zero temperature as $`V_{R,\omega }V_{R,\omega }R\omega `$, where $`R`$ is the resistance, and $`\omega `$ is the frequency. Thus, phase fluctuations are:
$$\delta \varphi _\omega \delta \varphi _\omega \frac{R}{\omega }.$$
(9)
Comparing this correlation function to $`X_\omega X_\omega `$, one may see that the circuit is described by LandauโZener model with marginal dissipation. The parameters are identified as:
$$\nu =\frac{eV\mathrm{\Delta }_0}{\mathrm{}};\mathrm{\Delta }=\mathrm{\Delta }_0\sqrt{1\tau };\alpha =\frac{eR\mathrm{\Delta }_0^2}{2\pi \mathrm{}^2}.$$
(10)
There are two regimes of tunneling for different values of environment temperature $`T`$. For $`T\mathrm{}\mathrm{\Delta }/\nu `$ the tunneling is thermally assisted, and the tunneling probability obeys Arrhenius law: $`w\mathrm{exp}(2\mathrm{\Delta }/T)`$. For low temperatures, the tunneling is due to non-adiabaticity, and the probability saturates at $`T0`$. We consider these regimes separately.
Quantum regime. At low temperatures, we treat the nonequilibrium problem by the Keldysh technique. We introduce two time contours describing evolution of wave function and its complex conjugate. Standard contours going along the real time axis are inappropriate for the problem in question because the exponentially small adiabatic transition probability results from a delicate destructive interference of many oscillating contributions. We move Keldysh contours away from the real axis to avoid oscillating terms. Our choice of contours is similar to , where it was used in the context of exciton autolocalization. To illustrate the use of contours, consider the tunneling for $`\alpha =0`$. The transition probability is $`w=\mathrm{exp}(S_0)`$, with the action $`S`$ given by :
$$S_0=2Im\underset{t_1}{\overset{i\tau _0}{}}(ฯต_+(t)ฯต_{}(t))๐t=\frac{\pi \mathrm{\Delta }^2}{\nu }.$$
(11)
The integral is taken from any point $`t_1`$ on the real axis to $`i\tau _0`$, where the adiabatic energy $`ฯต(t)`$ has a square root singularity. The two adiabatic states correspond to the branches $`ฯต_\pm (t)`$, i. e., to the two different sheets of the Riemann surface.
The result (11) may be represented using Keldysh contours shown on Fig.1(a). In the Keldysh formalism, the forward contour always goes from $`t=\mathrm{}`$ on the real axis, where the initial state is prepared, to $`t=+\mathrm{}`$ where the final state is measured. We draw the contour through the branching point $`t=i\tau _0`$. At this point the two energies $`ฯต_+(t)`$ and $`ฯต_{}(t)`$ coincide, hence the adiabaticity is violated, and the transition occurs. It means that we change the branch $`ฯต_{}(t)ฯต_+(t)`$, and continue the contour on another sheet of Riemann surface. Then, the contour goes back to the real axis, and then to $`t=+\mathrm{}`$. The backward contour is a complex conjugate of the forward contour. In the adiabatic approximation, the segments parallel to the real axis give no contribution to the imaginary part of the action, and thus one may connect vertical segments into a single contour $`C`$ (see Fig.1(b)). In this representation, the action (11) is given by
$$S_0=\underset{C}{}\sqrt{(\nu t)^2+\mathrm{\Delta }^2}๐t$$
(12)
It turns out that the problem with $`\alpha 0`$ has a solution of a similar structure: there is a square root singularity at some point $`t=i\tau _\alpha `$ in the complex plane, and the branch is changed when the contour passes this point. In this case, the branching point $`i\tau _\alpha `$ has to be found self-consistently. We derive the effective action on the contour $`C`$, look for an instanton solution and find $`\tau _\alpha `$ from equations of motion. Let us introduce a field $`\mathrm{\Phi }(t)=\nu t+X(t)`$, so that the energy of the spin is $`\sqrt{\mathrm{\Delta }^2+\mathrm{\Phi }^2(t)}`$. Thus, the two-level system action is
$$S_{TLS}=i_C๐t\sqrt{\mathrm{\Delta }^2+\mathrm{\Phi }^2(t)}.$$
(13)
(The Berry phase is neglected here since the โmagnetic fieldโ acting on the spin is always in the $`xz`$โplane). The dynamics of the environment is described by $`S_{env}+S_\lambda `$, where
$$S_{env}=\frac{1}{2}\underset{i}{}\left(\dot{x}_i^2\omega _i^2x_i^2\right),$$
(14)
and the term $`S_\lambda `$ enforces the constraint $`\dot{\mathrm{\Phi }}\dot{X}=\nu `$:
$$S_\lambda =i_C๐t\lambda (t)\left(\dot{\mathrm{\Phi }}(t)\nu \underset{i}{}\gamma _i\dot{x}_i(t)\right)$$
(15)
We integrate out $`\mathrm{\Phi }(t)`$ and $`x_i(t)`$ in the saddle-point approximation and find the effective action in terms of the Lagrange multiplier $`\lambda (t)`$:
$`S_{eff}`$ $`=`$ $`i\mathrm{\Delta }{\displaystyle ๐t\sqrt{1\dot{\lambda }^2}}+i\nu {\displaystyle ๐t\lambda (t)}`$ (16)
$`+`$ $`{\displaystyle \frac{\alpha }{4\pi }}{\displaystyle ๐t๐t^{}\frac{(\lambda (t)\lambda (t^{}))^2}{(tt^{})^2}}.`$ (17)
The solution describing LandauโZener tunneling has different signs on different sheets of the Riemann surface: $`\lambda (t0)=\lambda (t+0)`$. Taking a saddle point with respect to $`\lambda (t)`$, one arrives to the equation of motion:
$$\mathrm{\Delta }\frac{d}{dt}\frac{\dot{\lambda }}{\sqrt{1\dot{\lambda }^2}}=\nu +\frac{i\alpha }{\pi }\underset{\tau _\alpha }{\overset{\tau _\alpha }{}}\frac{\lambda (t^{})dt^{}}{(tt^{})^2}.$$
(18)
For $`\alpha =0`$ the solution is $`\lambda _0(t)=\sqrt{\tau _0^2+t^2}`$. Interestingly, the integral in (18) with $`\lambda =\lambda _0(t)`$ is constant for $`t[i\tau _0;i\tau _0]`$, and the r.h.s. of (18) preserves its form. Thus, one may look for the solution of the form $`\lambda (t)=\sqrt{\tau _\alpha ^2+t^2}`$. From Eq. (18) one finds $`\tau _\alpha =\mathrm{\Delta }/(\nu \alpha )`$. Substituting it into (16), one obtains the action
$$S=\frac{\pi \mathrm{\Delta }^2}{\nu \alpha },$$
(19)
and the tunneling probability $`\mathrm{exp}(S)`$ given by Eq. (1).
This solution gives an instanton for $`\nu >\alpha `$. Otherwise, for $`\nu <\alpha `$, there is no saddle point solution and the tunneling is impossible. This happens because the friction force is larger than the bias, and the energy put into the system by the bias source, is fully dissipated into the environment. Because of that, there is no level crossing, even at complex times. The tunneling time $`\tau _\alpha `$ diverges at $`\nu =\alpha `$. Because of that, the dynamics far from level crossing point becomes essential. The linear approximation on which Eq. (2) is based may break down. Also, for $`\nu \alpha `$, the tunneling time may be comparable to the time $`t_{\mathrm{}}`$ when the system is prepared and measured. Thus, the LandauโZener problem makes no sense for $`\nu `$ too close to $`\alpha `$. The behaviour of the system is determined by the details of the dynamics far from the avoided crossing point.
In the perturbative regime, $`\alpha \nu `$, the action (19) can be expanded in $`\alpha `$:
$$S=S_0+\frac{\pi \mathrm{\Delta }^2\alpha }{\nu ^2}.$$
(20)
This result contradicts to Ao et al.,, who found an $`\alpha `$โindependent tunneling probability. We believe that the result is incorrect because of improper choice of shakeup force $`\zeta (t)`$ that does not describe tunneling.
The above solution shows that at $`T=0`$ the environment effect is purely dissipative, and the coupling to the environment reduces the transition probability. This happens because zero-point fluctuations of the environment do not lead to a real transition. At finite temperatures, the environment may transfer the energy to the two-level system, thus increasing the transition probability. If the temperature is low, $`TT_0=\mathrm{}/\tau _\alpha `$, the transition still has mostly quantum nature. In this limit, the finite temperature can be taken into account by requiring the instanton to be periodic in imaginary time, $`\lambda (t)=\lambda (t+i\beta )`$, with the period $`\beta =1/T`$. Thus, one has to consider a periodic system of cuts, $`t[i\tau +i\beta n,i\tau +i\beta n]`$. Note that the nonlocal term in (16) gives rise to interaction between instantons on different cuts. If $`\beta \tau _\alpha `$, the interaction is weak and can be treated perturbatively. The correction to the action per period is:
$$\delta S=\frac{\pi \alpha \tau _\alpha ^4}{2}\underset{n0}{}\frac{1}{\beta ^2n^2}$$
(21)
After computing the sum over $`n`$, one obtains:
$$S(T)=S(T=0)\left(1\frac{\pi ^2}{6}\frac{\alpha T^2\tau _\alpha ^2}{(\nu \alpha )}\right),$$
(22)
and the tunneling probability is $`w\mathrm{exp}(S(T))`$, as before. Eq. (22) is applicable when the correction is small and thus does not lead to nonโmonotonous $`T`$โdependence.
Classical regime. If the temperature of the environment is high ($`T\mathrm{}/\tau _\alpha `$), the tunneling is thermally assisted, and the effect of environment can be divided into slow regular motion with frequencies $`\omega 1/\tau _\alpha `$ and fast Langevin noise with $`\omega 1/\tau _\alpha `$. (Only noise with $`\omega >2\mathrm{\Delta }`$ contributes to the transition probability, and this separation is valid in the adiabatic limit.) Then the Hamiltonian is:
$$\widehat{H}=F(t)\widehat{\sigma }_z+\mathrm{\Delta }\widehat{\sigma }_x+\widehat{u}(t)\sigma _z,$$
(23)
where $`F(t)=\nu t+X(t)`$ is a regular part, and $`\widehat{u}(t)=\widehat{X}(t)X(t)`$ is a fluctuating part. Since the first two terms are slow functions of time, one may consider them in the adiabatic approximation, and treat the noise $`\widehat{u}(t)`$ as a perturbation causing non-adiabatic transition. This implies that the noise contribution to the transition amplitude is larger than the non-adiabatic correction. The eigenstates of the frozen Hamiltonian are:
$$\psi _+(t)=\left(\begin{array}{c}\mathrm{cos}\frac{\theta (t)}{2}\\ \mathrm{sin}\frac{\theta (t)}{2}\end{array}\right);\psi _{}(t)=\left(\begin{array}{c}\hfill \mathrm{sin}\frac{\theta (t)}{2}\\ \hfill \mathrm{cos}\frac{\theta (t)}{2}\end{array}\right),$$
(24)
where $`\mathrm{tan}\theta (t)=\mathrm{\Delta }/F(t)`$ . In this basis the Hamiltonian is:
$$\widehat{H}=ฯต(t)\widehat{\sigma }_z+\widehat{u}(t)(\mathrm{cos}\theta (t)\widehat{\sigma }_z+\mathrm{sin}\theta (t)\widehat{\sigma }_x),$$
(25)
where $`ฯต(t)=\mathrm{\Delta }/\mathrm{sin}\theta (t)`$ is the adiabatic energy. The perturbation theory with respect to $`\widehat{u}(t)`$ gives the transition rates
$$w_{\pm ,mn}=\left|\underset{\mathrm{}}{\overset{\mathrm{}}{}}e^{\pm i\lambda (t)}\mathrm{sin}\theta (t)n|\widehat{u}(t)|m๐t\right|^2$$
(26)
between the states $`|m`$ and $`|n`$ of the environment. In Eq. (26), $`\dot{\lambda }(t)=ฯต(t)`$, and the subscript $`(\pm )`$ denotes initial spin state. Tracing out the environment, one finds the LandauโZener transition rates
$$w_\pm =\underset{\mathrm{}}{\overset{\mathrm{}}{}}๐t\underset{\mathrm{}}{\overset{\mathrm{}}{}}๐t^{}e^{\pm i(\lambda (t)\lambda (t^{}))}\mathrm{sin}\theta (t)\mathrm{sin}\theta (t^{})\widehat{u}(t)\widehat{u}(t^{}).$$
(27)
This integral contains a fast oscillating function, and is determined by the region where $`|tt^{}|1/ฯต(t)\tau _0`$. Therefore, one may approximate in the prefactor $`tt^{}`$, integrate over $`tt^{}`$ and find the LandauโZener transition rates:
$$w_\pm =\underset{\mathrm{}}{\overset{\mathrm{}}{}}K(\omega =\pm 2ฯต(t))\mathrm{sin}^2\theta (t)๐t.$$
(28)
Here
$$K(\omega )=\underset{\mathrm{}}{\overset{\mathrm{}}{}}e^{i\omega t}\widehat{u}(t)\widehat{u}(0)๐t$$
(29)
is the noise spectrum. In equilibrium, it can be expressed in terms of the spectral function $`J(\omega )`$ and the Bose distribution function $`N(\omega )`$:
$$K(\omega )=J(|\omega |)(N(|\omega |)+\theta (\omega )).$$
(30)
The result (28) is also true for non-equilibrium noise with the proper choice of $`K(\omega )`$.
To find the still unknown function $`\theta (t)`$, consider the force exerted by the rotating spin on the environment, $`f(t)=\widehat{\sigma }_z(t)=\pm \mathrm{cos}\theta (t)`$. The response to that force is, in Fourier representation, $`X_\omega =iJ(\omega )f_\omega `$. Then, one may write an equation for $`\theta (t)`$:
$$\dot{F}(t)=\nu +\dot{X}(t)=\nu \alpha \mathrm{cos}\theta (t).$$
(31)
Since $`F(t)=\mathrm{\Delta }/\mathrm{tan}\theta (t)`$, one has:
$$\mathrm{\Delta }\dot{\theta }=\mathrm{sin}^2\theta (t)(\nu \alpha \mathrm{cos}\theta (t)).$$
(32)
For small bias, $`\nu <\alpha `$, there is no LandauโZener tunneling, and situation is similar to quantum limit. If the bias $`\nu >\alpha `$, the transition probability is:
$$w_\pm =\underset{0}{\overset{\pi }{}}r_\pm (\theta )๐\theta ;r_\pm (\theta )=\frac{\mathrm{\Delta }K(\pm \frac{2\mathrm{\Delta }}{\mathrm{sin}\theta })}{\nu \alpha \mathrm{cos}\theta }.$$
(33)
Note that the small bias $`\nu `$ appears in the denominator. Because of that, the probability computed from (33) can be quite large, and the perturbation theory on which this equation is based, appears to break down. This happens because the motion at small bias is slow, and a small noise acts on the system for a long time, giving rise to a large tunneling probability.
To solve the problem, note that according to (28), there is no interference between transition amplitudes at different times $`t`$. This happens because the transitions are due to random noise, and the amplitudes of separate transitions have random phase. In this situation, one may use master equation, with the transition rate per unit time from (28). Since $`\theta (t)`$ depends on the history of the system, it is convenient to use $`\theta `$ as an independent variable instead of $`t`$. Master equation, written in the variable $`\theta `$, has the form:
$$\frac{dP_+(\theta )}{d\theta }=\frac{dP_{}(\theta )}{d\theta }=r_+(\theta )P_+(\theta )+r_{}P_{}(\theta ),$$
(34)
where $`P_\pm (\theta )`$ are occupancies of two adiabatic states. The initial condition is $`P_+(0)=0`$, $`P_{}(0)=1`$.
Eq. (34) can be solved in a general form. However, for simplicity, we consider the two cases: (i) $`T\mathrm{\Delta }`$ and (ii) $`T\mathrm{\Delta }`$. For $`T\mathrm{\Delta }`$, the transition rates $`r_+`$ and $`r_{}`$ are equal. The noise correlator is $`K(\omega )=\alpha T/\omega ^2`$. The calculation gives for $`w=P_+(\theta =\pi )`$:
$$w=\frac{1}{2}(1e^{2R});R=\underset{0}{\overset{\pi }{}}r_{}(\theta )๐\theta .$$
(35)
Computing the integral, one obtains:
$$R=\frac{\pi T}{4\mathrm{\Delta }}\left(\frac{\nu }{\alpha }\sqrt{\frac{\nu ^2}{\alpha ^2}1}\right).$$
(36)
In the limit $`T\mathrm{}`$, Eq. (35) predicts $`w1/2`$, the result of a phenomenological approach. Note that (35) is true only when $`R\mathrm{\Delta }^2/\nu `$, otherwise non-adiabaticity is again important, and the exponential correction has different structure.
We consider now the case of intermediate temperatures: $`\mathrm{}/\tau _\alpha T\mathrm{\Delta }`$. The transition $`|\psi _{}|\psi _+`$ requires energy transfer $`2\mathrm{\Delta }`$, while the inverse transition does not, and the transition rates $`r_+`$ and $`r_{}`$ are not equal. Therefore, the transition from the lower state to the upper one occurs at $`\theta =\pi /2`$, when the energy difference is minimal. This transition can be considered as an instant excitation occurring at $`\theta \pi /2`$ during the time $`t_{ex}\sqrt{\mathrm{\Delta }T}/\nu `$. It is followed by a decay from the upper state. Thus, the transition probability is the product of the probability of excitation at $`\theta =\pi /2`$ and the probability that the excited state survives for $`\pi /2<\theta <\pi `$. From the master equation (34) one finds the excitation probability:
$$P_{ex}=\underset{0}{\overset{\pi }{}}r_{}(\theta )๐\theta =\frac{\alpha }{\nu }\sqrt{\frac{\pi T}{\mathrm{\Delta }}}e^{2\mathrm{\Delta }/T},$$
(37)
and the survival probability:
$$P_s=\mathrm{exp}\left(\underset{\pi /2}{\overset{\pi }{}}r_+(\theta )๐\theta \right)=\sqrt{1\frac{\alpha }{\nu }}.$$
(38)
Finally, the transition probability is:
$$w=P_{ex}P_s=\frac{\alpha }{\nu }\sqrt{1\frac{\alpha }{\nu }}\sqrt{\frac{\pi T}{\mathrm{\Delta }}}e^{2\mathrm{\Delta }/T}.$$
(39)
Note that the dependence of this expression on $`\alpha `$ is non-monotonous. This is due to the fact that the coupling to the environment provides both external noise and dissipation. The transition probability reaches the maximal value at $`\alpha =2\nu /3`$.
Conclusion. We studied dissipative LandauโZener tunneling marginally coupled to environment in both quantum and classical limits. At $`T=0`$ the effect of environment is purely dissipative, whereas at high temperatures thermal fluctuations in the environment increase the transition probability, leading to nonโmonotonous dependence of the transition rates on the coupling strength $`\alpha `$. At $`\nu <\alpha `$ the tunneling is blocked. Thus, the interplay between fluctuations and dissipation is manifest in the marginal coupling model.
Acknowledgements. I am grateful to M. V. Feigelโman, A. S. Ioselevich, L. S.Levitov, D. Esteve, and C. Urbina for stimulating and illuminating discussions. This research was supported by the Russian Ministry of Science via the program โPhysics of quantum computingโ, and by RFBR grant 98-02-19252. |
warning/0001/quant-ph0001082.html | ar5iv | text | # Quantum Diagonalization of Hermitean Matrices
(January 2000)
## Abstract
To measure an observable of a quantum mechanical system leaves it in one of its eigenstates and the result of the measurement is one of its eigenvalues. This process is shown to be a computational resource. It allows one, in principle, to diagonalize hermitean $`(N\times N)`$ matrices by quantum mechanical measurements only. To do so, one considers the given matrix as an observable of a single spin with appropriate length $`s`$ which can be measured using a generalized Stern-Gerlach apparatus. Then, each run provides one eigenvalue of the observable. As it is based on the โcollapse of the wave functionโ associated with a measurement, the procedure is neither a digital nor an analog calculationโit defines thus a new quantum mechanical method of computation.
Non-classical features of quantum mechanics such as Heisenbergโs uncertainty relation and entanglement have intrigued physicists for several decades. From a classical point of view, quantum mechanics imposes constraints on the ways to talk about nature. An electron does not โhaveโ position and momentum as does a billiard ball. Similarly, if a photon is entangled with a second oneโpossibly very far awayโone cannot ascribe properties to it as is done for an individual classical particle. The lesson to be learned is that classical intuition about the macroscopic world simply does not extrapolate into the microscopic world.
In recent years, an entirely different attitude towards quantum theory has been put forward. The focus is no longer on attempts to come to terms with its strange features but to capitalize on its both counter-intuitive and well-established properties. In this way, surprising methods have been uncovered to solve specific problems by means which have no classical equivalent: quantum cryptography, for example, allows one to establish secure keys for secret transmission of information ; entanglement is used as a tool to set up powerful quantum algorithms which do factor large integers much more efficiently than any classical algorithm . Throughout, these new techniques rely on the measurement of quantum mechanical observables as a reliable tool. This is also true for quantum error correction required to let any potential algorithm run.
Here the purpose is to point out that the bare โprojectionโ effected by a quantum mechanical measurement does possess computational power itself. As will be shown below, it can be used to solve explicitly at least one specific computational task, namely to determine eigenstates and eigenvalues of hermitean $`(N\times N)`$ matrices.
The diagonalization of hermitean matrices is a recurrent problem in mathematics, physics, and related fields. Using the notation of a quantum physicist the problem reads as follows. Given a self-adjoint operator $`\widehat{A}`$ acting on a Hilbert space $``$ of dimension $`N`$, one needs to determine its eigenstates $`|A_n,n=1,\mathrm{},N,`$ and its $`N`$ real eigenvalues $`A_n`$ satisfying $`\widehat{A}|A_n=A_n|A_n,n=1,\mathrm{},N.`$ If normalized to one, the eigenstates constitute a complete orthonormal basis of the space $``$: $`_{n=1}^N|A_nA_n|=1,A_n|A_n^{}=\delta _{nn^{}}.`$ The standard solution from linear algebra is to write down the eigenvalue equation with respect to a given orthonormal basis $`|k,k=1,\mathrm{},N`$, say. The $`N^2`$ matrix elements $`๐ _{kk^{}}=k|\widehat{A}|k^{}`$ determine the operator $`\widehat{A}`$ uniquely and its eigenstates are characterized by the coefficients $`(\stackrel{}{A}_n)_k=A_{nk}`$ in the expansion $`|A_n=_kA_{nk}|k`$. The number $`\lambda `$ is an eigenvalue of $`\widehat{A}`$ if the characteristic polynomial $`P_A(\lambda )`$ of the matrix $`๐ `$ vanishes, $`P_A(\lambda )=det\left(๐ \lambda ๐ค\right)=0,`$ where $`๐ค`$ is the $`(N\times N)`$ unit matrix. Once the $`N`$ roots $`A_n`$ of the polynomial $`P_A`$ are known, the non-zero solutions of the equation
$$\left(๐ A_n\right)\stackrel{}{A}_n=0,n=1,\mathrm{},N,$$
(1)
provide the eigenvectors $`|A_n`$ in the basis $`|k`$. Analytic expressions for the eigenvalues $`A_n`$ in terms of the elements of $`๐ `$ exist only if $`N4`$. In general, numerical methods are required to determine approximately the roots of $`P_A(\lambda )`$.
The quantum diagonalization of hermitean matrices is based on the assumption that the behaviour of a spin $`s`$ is described correctly by non-relativistic quantum mechanics. This method will make use of the โcollapse of the wave functionโ as computational resource. Note that the procedure does not depend on a particular interpretation of quantum mechanics. Five steps are necessary to achieve the diagonalization of a given matrix $`๐ `$ (supposed for simplicity not to have degenerate eigenvalues). The individual steps will be described first in a condensed form; subsequently, commentaries explain the technical details.
1. Standard form of $`๐ `$: Write the hermitean $`(N\times N`$) matrix $`๐ `$ as a combination of linearly independent hermitean *multipole* operators $`๐ณ_\nu ,\nu =0,\mathrm{},N^21,`$
$$๐ =\underset{\nu =0}{\overset{N^21}{}}๐_\nu ๐ณ_\nu ,๐_\nu =\frac{1}{N}\text{ Tr }\left[\mathrm{๐ ๐ณ}_\nu \right]๐.$$
(2)
2. Identification of an observable: Interpret the matrix $`๐ `$ as an observable $`๐ง_A`$ for a single quantum spin $`๐ฒ`$ with quantum number $`s=(N1)/2`$,
$$๐ง_A(๐ฒ)=\underset{\nu =0}{\overset{N^21}{}}๐_\nu ๐ณ_\nu (๐ฒ),$$
(3)
using the expression of the multipoles $`๐ณ_\nu (๐ฒ)`$ in terms of the components of a spin.
3. Setting up a measuring device: Construct an apparatus app($`๐ง_A`$) suitable to measure the observable $`๐ง_A`$.
4. Determination of the eigenvalues: Carry out measurements with the apparatus app($`๐ง_A`$) on a spin $`s`$ prepared in a homogeneous mixture $`\widehat{\rho }=๐จ/(2s+1)`$. The output of each measurement will be one of the eigenvalues $`A_n`$ of the matrix $`๐ .`$ After sufficiently many repetitions, all eigenvalues will be known.
5. Determination of the eigenstates: Calculate the eigenstates $`|A_n`$ of the matrix $`๐ `$ on the basis of Eq. (1) and the experimentally determined eigenvalues $`A_n`$. Alternatively, determine the eigenstates $`|A_n`$ *experimentally* by methods of state reconstruction.
Thus, the matrix $`๐ `$ has been diagonalized without calculating the zeroes of its characteristic polynomial by traditional means. The fourth step solves the hard part of the eigenvalue problem since it provides the eigenvalues $`A_n`$ of the matrix $`๐ `$. The comments to follow provide the background necessary to perform the individual steps. Emphasis will be both on the construction of a device measuring for a given hermitean operator (Step 3) and on the working of a quantum mechanical measurement (Step 4).
Ad 1: The $`N^2`$ self-adjoint multipole operators $`๐ณ_\nu =๐ณ_\nu ^{}`$ form a basis in the space of hermitean operators acting on an $`N`$-dimensional Hilbert space $``$ . Two multipoles are orthogonal with respect to a scalar product defined as the trace of their product: $`(1/N)\text{ Tr }\left[๐ณ_\nu ๐ณ_\nu ^{}\right]=\delta _{\nu \nu ^{}}.`$
Consider now a Hilbert space $`_s`$ of dimension ($`2s+1`$) which carries an irreducible representation of the group $`SU(2)`$ with the spin components $`(๐ฒ_\mathrm{๐ฃ},๐ฒ_\mathrm{๐ค},๐ฒ_\mathrm{๐ฅ})`$ as generators. Then, the multipoles $`๐ณ_\nu ,\nu =1,\mathrm{},N^21,`$ are given by the symmetrized products $`๐ฒ_{j_1}๐ฒ_{j_2}\mathrm{}๐ฒ_{j_a},j_i=1,2,3,`$ and $`a=0,1,\mathrm{},2s,`$ after subtracting off the trace (define $`๐ณ_0๐ณ^{(0)}=๐ค`$, the $`(N\times N)`$ unit matrix). The index $`a`$ labels $`(2s+1)`$ classes with $`(2a+1)`$ elements transforming among themselves under rotations; for the sake of brevity, a collective index $`\nu (a;j_1,\mathrm{},j_k)`$ is used. Explicitly, the lowest multipoles read
$$๐ณ_j^{(1)}=๐ฒ_j,๐ณ_{j_1j_2}^{(2)}=\frac{1}{2}\left(๐ฒ_{j_1}๐ฒ_{j_2}+๐ฒ_{j_2}๐ฒ_{j_1}\right)\frac{\delta _{i_1j_2}}{3}๐ฒ_{j_1}๐ฒ_{j_2}.$$
(4)
The set $`\{๐ณ_\nu \}`$ is a basis for the hermitean operators on $`_s`$.
Ad 2: Since the multipoles are expressed explicitly as a function of the spin components not exceeding the power $`2s`$, it is justified to consider them and, *a fortiori*, the quantity $`๐ง_A`$ as an *observable* for a spin $`s`$.
Ad 3: It is natural to expect that every self-adjoint operator $`\widehat{B}`$ comes along with an apparatus app($`\widehat{B}`$) capable of measuring it . For particle systems, setting up such a device remains a challenging task for an experimenter.
For spin systems, the situation is different, however. Swift and Wright have shown how to devise, in principle, a generalized Stern-Gerlach apparatus which measures any observable $`๐ง_A(๐ฒ)`$โjust as a traditional Stern-Gerlach apparatus measures the spin component $`๐ง๐ฒ`$ along the direction $`๐ง`$. The construction requires that arbitrary static electric and magnetic fields, consistent with Maxwellโs equations, can be created in the laboratory. To construct an apparatus app$`(๐ง_A)`$ means to identify a spin Hamiltonian $`๐ง(๐ซ,๐ฒ)`$ which splits an incoming beam of particles with spin $`s`$ into subbeams corresponding to the eigenvalues $`A_n`$. The most general Hamiltonian acting on the Hilbert space $``$ of a spin $`s`$ reads
$$๐ง(๐ซ,๐ฒ)=\underset{\nu =0}{\overset{N^21}{}}\mathrm{\Phi }_\nu (๐ซ)๐ณ_\nu ,$$
(5)
with traceless (except for $`\nu =0`$) symmetric expansion coefficients $`\mathrm{\Phi }_\nu (๐ซ)(\mathrm{\Phi }_{j_1j_2\mathrm{}j_k}^{(k)}(๐ซ))`$ which vary in space. Tune the electric and magnetic fields in such a way that the coefficients $`\mathrm{\Phi }_\nu (๐ซ)`$ and its first derivative with respect to some spatial direction, $`r_1`$, say, satisfy
$$\mathrm{\Phi }_\nu (๐ซ=0)=\frac{\mathrm{\Phi }_\nu (๐ซ=0)}{r_1}=๐บ_n.$$
(6)
This is always possible with realistic fields satisfying Maxwellโs equations. Then, the Hamiltonian in (5) has two important properties. (i) At the origin, $`๐ซ=0`$, it coincides with the matrix $`๐ง_A`$. (ii) Suppose that a beam of particles with spin $`s`$ enters the generalized Stern-Gerlach apparatus app($`๐ง_A`$) just described. At its center, particles in an eigenstate $`|A_n`$, say, will experience a force in the $`r_1`$ direction given (up to second order in distance from the center) by
$$F_1(๐ซ=\mathrm{๐})=\frac{A_n|๐ง(๐ซ=0,๐ฒ)|A_n}{r_1}=A_n,n=1,\mathrm{},2s+1.$$
(7)
Consequently, particles with a spin projected onto one of the eigenstates $`|A_n`$ of the operator $`๐ง_A`$ are separated spatially by this apparatus. The procedure is entirely analogous to that for a spin $`1/2`$ where a familiar Stern-Gerlach apparatus is used (see for details).
Ad 4: The โprojection postulateโ of quantum mechanics describes the effect of measuring an observable $`\widehat{B}`$ on a system $`๐ฎ`$ by means of an apparatus app($`\widehat{B}`$). If the system is prepared initially in a state with density matrix $`\widehat{\rho }`$ one has:
$$\text{app}(\widehat{B}):\widehat{\rho }\stackrel{p_n}{}(B_n;\widehat{\rho }_n),p_n=\text{ Tr }\left[\widehat{\rho }\widehat{\rho }_n\right].$$
(8)
The action of the apparatus is, with probability $`p_n`$, to throw the system into an eigenstate $`\widehat{\rho }_n|B_nB_n|`$ of the observable $`\widehat{B}`$; the *outcome* of the measurement is given by the associated eigenvalue $`B_n`$. By the way, the notion of โcollapseโ or โprojectionโ can be avoided by characterizing the process indirectly by refering to โrepeatable measurementsโ .
The outcome of an individual measurement cannot be predicted due to the probabilistic character of quantum mechanics. Therefore, the probabilities $`p_n`$, resulting from (infinitely often) repeated measurements on identically prepared systems, represent the essential link between theory and experiment. They provide information about the state of the system conditioned by the selected observable. Thus, a measurement reveals (or confirms) properties of the state $`\widehat{\rho }`$ of the system while the observable $`\widehat{B}`$ at hand is assumed to be known, including its eigenstates and eigenvalues. To put it differently, the observable defines the scope of the possible results of a measurement: the only possible outcomes are its eigenvalues $`B_n`$, and, directly after the measurement the system necessarily resides in the corresponding state $`|B_n`$.
As the occurrence of the eigenvalues is purely probabilistic, one needs to repeat the experiment until all values $`A_n`$ have been obtained. If the spin $`s`$ is prepared initially in a homogeneous mixture, $`\widehat{\rho }=๐ค/(2s+1)`$, the $`(2s+1)`$ possible outcomes occur with equal probability. The probability not to have obtained one specific value $`A_n`$ after $`N_0N`$ measurements equals $`1/(2s+1)^{N_0}`$, decreasing exponentially with $`N_0`$.
Ad 5: It would be very convenient now to โread outโ directly the quantum state $`\widehat{\rho }_n`$ obtained from a single measurement with result $`A_n`$. However, due to the no-cloning theorem , an unknown state cannot be determined if only one copy of it is available. Upon repeating the measurement a large number of times and keeping only those states with the same eigenvalue $`A_n`$, one produces an ensemble of systems prepared identically in the state $`\widehat{\rho }_n`$. This is sufficient to reconstruct an unknown state since a density matrix $`\widehat{\rho }`$ can be written as
$$\widehat{\rho }=\frac{1}{N}\underset{\mu =1}{\overset{N^2}{}}P_\mu \widehat{Q}^\mu ,N=2s+1,$$
(9)
where the coefficient $`P_\mu ๐ง_\mu |\widehat{\rho }|๐ง_\mu `$ is the probability to find the system in a coherent spin state $`|๐ง_\mu `$. The operators $`\widehat{Q}^\mu ,\mu =1,\mathrm{},N^2`$, form a basis for hermitian operators, similar to but different from the multipoles $`๐ณ_\nu `$ . Thus, Eq. (9) parametrizes $`\widehat{\rho }`$ by expectation values $`P_\mu `$ which can be measured by a standard Stern-Gerlach apparatus.-
In sum, the basic ingredient of quantum diagonalization is the โcollapseโ of the wave function projecting any state onto a randomly selected eigenstate of the measured observable. Generalizations of this approach are expected to include the diagonalization of unitary matrices and the determination of roots of polynomials.
Usually, a measurement is thought to confirm or reveal some information about the state of the system. Here, on the contrary, the idea is to learn something about the measured observable instead. Why is this possible? It is fundamental to realize that the input required to actually measure $`\widehat{A}`$ differs from the output of the experiment: for a measurement of $`\widehat{A}`$, the construction of an apparatus app($`\widehat{A}`$) is sufficient which is *possible without*knowing eigenvalues and eigenstates of $`\widehat{A}`$. Necessarily, after a measurement partial information about the spectral properties of the observable $`\widehat{A}`$ is available according to (8). This is due to the constraints (i) that the possible outcomes of measuring $`\widehat{A}`$ are its eigenvalues and (ii) that the system subsequently will occupy the corresponding eigenstate. Thus, if the eigenstates and eigenvalues of $`\widehat{A}`$ not known initially, one indeed acquires information about them by measuring $`\widehat{A}`$.
The quantum mechanical diagonalization appears to be neither an analog nor a digital calculation. It is not based on the representation of a mathematical equation in terms of a physical system which then would โsimulateโ it. Similarly, no โsoftware programโ is executed which would implement an diagonalization algorithm. One might best describe the measuring device app($`๐ง_A`$) as a โspecial purpose machineโ based on the projection postulate.
For the time being, the method introduced here is important from a conceptual but not a technological point of view. On the one hand, the diagonalization of matrices is not a hard problem such as factorization of large integer numbers; on the other, the actual implementation in the laboratory is challenging. It is important, however, that there is no physical principle which would forbid the construction of such a machine. Further, it is expected to be fruitful from a conceptual point of view since it provides a different perspective on the projection postulate . Quantum diagonalization as introduced here shows thatโin an unexpected wayโstandard quantum mechanics attributes computational power to the measurement of an observable. The fact that one can use a measurement to perform calculations might turn into an argument in favor of the โrealityโ of the quantum mechanical projection postulate. |
warning/0001/cond-mat0001154.html | ar5iv | text | # Excitation Energies from Time-Dependent Density Functional Theory Using Exact and Approximate Potentials
## 1 Introduction
The Hohenberg-Kohn theorem of ground-state density functional theory (DFT) guarantees that every observable of a stationary physical system can be expressed in terms of its ground-state density. In principle, this is also true for the set of excited-state energies, and several extensions of ground-state DFT have been proposed \- . Accurate calculations of excitation energies, however, remain a difficult subject. Recently, some of us proposed a different approach to the calculation of excitation energies , within the framework of time-dependent DFT (TDDFT) . The central idea is to use the fact that the linear density response has poles at the physical excitation energies and can be calculated from the response function of a noninteracting Kohn-Sham (KS) system and a frequency-dependent Kohn-Sham (KS) kernel. In this way, we obtain the shifts of the KS orbital differences (which are the poles of the KS response function) towards the true excitation energies. Recent applications \- suggest that this method may become a standard tool in quantum chemistry.
The success of any density functional method, however, depends on the quality of the functionals employed. In this article, we investigate the relative importance of the approximations inherent in the TDDFT formalism for the calculation of discrete excitation energies of finite systems. This mainly concerns the role of the ground-state XC potential, $`v_{\mathrm{xc}}(๐ซ)`$ compared to the dynamical XC kernel, $`f_{\mathrm{xc}}(๐ซ,๐ซ^{};\omega )`$. For the helium and beryllium atoms, we compare the results obtained from using the exact XC potentials and two orbital dependent potentials, one based on the exact exchange expression and the other on the self-interaction corrected local density approximation , evaluated with the method of Krieger, Li and Iafrate (KLI) \- in combination with three distinct approximations for the XC kernels, which are given in Sect. 2.2.
## 2 Formalism
### 2.1 Kohn-Sham equations for the frequency-dependent linear density response
The frequency dependent linear density response $`n_{1\sigma }(๐ซ,\omega )`$ of electrons with spin $`\sigma `$, reacting to a perturbation $`v_{1\sigma ^{}}`$ of frequency $`\omega `$ can be written in terms of the interacting density-density response function $`\chi _{\sigma \sigma ^{}}`$ by
$$n_{1\sigma }(๐ซ,\omega )=\underset{\sigma ^{}}{}d^3r^{}\chi _{\sigma \sigma ^{}}(๐ซ,๐ซ^{};\omega )v_{1\sigma ^{}}(๐ซ^{},\omega ).$$
(1)
In the spin-dependent version of time-dependent DFT , the density response $`n_{1\sigma }`$ can be expressed in terms of the response function $`\chi _{s\sigma \sigma ^{}}`$ of the non-interacting Kohn-Sham (KS) system :
$$n_{1\sigma }(๐ซ,\omega )=\underset{\sigma ^{}}{}d^3r^{}\chi _{\mathrm{s}\sigma \sigma ^{}}(๐ซ,๐ซ^{};\omega )v_{\mathrm{s},1\sigma ^{}}(๐ซ^{},\omega ).$$
(2)
The KS response function
$$\chi _{\mathrm{s}\sigma \sigma ^{}}(๐ซ,๐ซ^{};\omega )=\delta _{\sigma \sigma ^{}}\underset{j,k}{}\left(f_{k\sigma }f_{j\sigma }\right)\frac{\phi _{j\sigma }\left(๐ซ\right)\phi _{k\sigma }^{}\left(๐ซ\right)\phi _{j\sigma }^{}\left(๐ซ^{}\right)\phi _{k\sigma }\left(๐ซ^{}\right)}{\omega \left(ฯต_{j\sigma }ฯต_{k\sigma }\right)+i\eta }$$
(3)
is readily expressed in terms of the unperturbed static Kohn-Sham orbitals $`\phi _{k\sigma }`$ (with occupation numbers $`f_{j\sigma }`$). Relation (2) contains the linearized KS potential
$$v_{\mathrm{s},1\sigma }(๐ซ,\omega )=v_{1\sigma }(๐ซ,\omega )+\underset{\sigma ^{}}{}d^3r^{}\frac{n_{1\sigma ^{}}(๐ซ^{},\omega )}{\left|๐ซ๐ซ^{}\right|}+\underset{\sigma ^{}}{}d^3r^{}f_{\mathrm{xc}\sigma \sigma ^{}}(๐ซ,๐ซ^{};\omega )n_{1\sigma ^{}}(๐ซ^{},\omega ).$$
(4)
in which the spin-dependent exchange-correlation (XC) kernel $`f_{\mathrm{xc}}`$ is defined as the Fourier transform of
$$f_{\mathrm{xc}\sigma \sigma ^{}}[n_0,n_0](๐ซ,t,๐ซ^{},t^{}):=\frac{\delta v_{\mathrm{xc}\sigma }[n_{},n_{}](๐ซ,t)}{\delta n_\sigma ^{}(๐ซ^{},t^{})}|_{n_0,n_0}.$$
(5)
Given an approximation to $`f_{\mathrm{xc}}`$, Eqs. (2) and (4) can be solved self-consistently for every frequency $`\omega `$.
### 2.2 Approximations for the exchange-correlation kernel
For spin-unpolarized ground states, there are only two independent combinations of the spin components of the XC kernel, since $`f_{\mathrm{xc}}=f_{\mathrm{xc}}`$ and $`f_{\mathrm{xc}}=f_{\mathrm{xc}}`$:
$$f_{\mathrm{xc}}=\frac{1}{4}\underset{\sigma \sigma ^{}}{}f_{\mathrm{xc}\sigma \sigma ^{}}=\frac{1}{2}\left(f_{\mathrm{xc}}+f_{\mathrm{xc}}\right)G_{\mathrm{xc}}=\frac{1}{4}\underset{\sigma \sigma ^{}}{}\sigma \sigma ^{}f_{\mathrm{xc}\sigma \sigma ^{}}=\frac{1}{2}\left(f_{\mathrm{xc}}f_{\mathrm{xc}}\right),$$
(6)
(contrary to common usage, we have not separated the Bohr magneton in the definition of $`G_{\mathrm{xc}}`$). Note that $`f_\mathrm{x}=G_\mathrm{x}`$, as exchange contains only parallel spin contributions.
The simplest possible approximation is the adiabatic local (spin-)density approximation (ALDA). For spin-unpolarized ground states, this leads to
$$f_{\mathrm{xc}}^{\mathrm{ALDA}}\left[n\right](๐ซ,๐ซ^{})=\delta \left(๐ซ๐ซ^{}\right)\frac{d^2e_{\mathrm{xc}}^{\mathrm{hom}}}{d\rho ^2}|_{\rho =n\left(๐ซ\right),\zeta =0},G_{\mathrm{xc}}^{\mathrm{ALDA}}\left[n\right](๐ซ,๐ซ^{})=\delta \left(๐ซ๐ซ^{}\right)\frac{\alpha _{\mathrm{xc}}\left(n\left(๐ซ\right)\right)}{n\left(๐ซ\right)},$$
(7)
where $`e_{\mathrm{xc}}^{\mathrm{hom}}`$ is the exchange-correlation energy per unit volume of the homogeneous electron gas, $`\zeta `$ is the relative spin polarization, $`\left(n_{}n_{}\right)/n`$, and the spin-stiffness $`\alpha _{\mathrm{xc}}=\frac{\delta ^2}{\delta \zeta ^2}\left(e_{\mathrm{xc}}^{\mathrm{hom}}(\rho ,\zeta )/\rho \right)|_{\zeta =0}`$.
Approximate XC functionals derived from the homogeneous electron gas suffer from several shortcomings, such as spurious self-interaction contributions. These are very significant for calculations of orbital eigenvalues, as they affect the asymptotic decay of the ground-state potential. For example, the XC potential in the local density approximation decays exponentially, so rapidly that only one virtual state is bound. An alternative approach towards the construction of improved functionals is to use perturbation theory in the electron-electron coupling constant. This leads to orbital-dependent functionals, which can be solved self-consistently using the optimized effective potential method (OEP) \- . In the time-dependent case, this method takes as a starting point a given (approximate) expression for the quantum mechanical action integral as a functional of a set of orbitals. Variation with respect to a local effective potential then leads to an integral equation for the exchange-correlation potential. Given an exchange-correlation potential of that kind, the corresponding exchange-correlation kernel can be constructed in the same spirit . The essential steps are formally identical to the OEP construction of the exchange-correlation potential for the ground-state.
In the time-dependent X-only approximation, $`A_{\mathrm{xc}}`$ is replaced by<sup>1</sup><sup>1</sup>1In general, a Keldysh contour integral in complex time is needed to avoid causality difficulties, except when the action is local in the orbitals in time, as is the case with all approximations tested here.
$$A_{\mathrm{x}\mathrm{only}}=\left(1/2\right)\underset{\sigma }{}\underset{i,j}{\overset{N_\sigma }{}}_{\mathrm{}}^{t_1}๐td^3rd^3r^{}\varphi _{i\sigma }^{}\left(๐ซ^{}t\right)\varphi _{j\sigma }\left(๐ซ^{}t\right)\varphi _{i\sigma }\left(๐ซt\right)\varphi _{j\sigma }^{}\left(๐ซt\right)/\left|๐ซ๐ซ^{}\right|.$$
(8)
The orbital-dependent exchange kernel in the time-dependent KLI approximation is
$$f_{\mathrm{x}\mathrm{only}\sigma \sigma ^{}}^{\mathrm{TDOEP}}(๐ซ,๐ซ^{})=\delta _{\sigma \sigma ^{}}\frac{1}{\left|๐ซ๐ซ^{}\right|}\frac{\left|\underset{k}{}f_{k\sigma }\phi _{k\sigma }\left(๐ซ\right)\phi _{k\sigma }^{}\left(๐ซ^{}\right)\right|^2}{n_\sigma \left(๐ซ\right)n_\sigma \left(๐ซ^{}\right)}.$$
(9)
In general, the exact $`\mathrm{X}`$-only kernel carries a frequency-dependence. This is not accounted for in the present approximation (9). However, for one- and spin-unpolarized two-electron systems, Eqs. (9) is the exact solution of the respective integral equations in the limit of a time-dependent X-only theory. This yields
$$f_\mathrm{x}=G_\mathrm{x}=\frac{2\left|_kf_k\phi _k\left(๐ซ\right)\phi _k^{}\left(๐ซ^{}\right)\right|^2}{n\left(๐ซ\right)\left|๐ซ๐ซ^{}\right|n\left(๐ซ^{}\right)}=(\frac{1}{2\left|๐ซ๐ซ^{}\right|}\mathrm{for}2\mathrm{elec}).$$
(10)
Inherent to any X-only theory, the resulting kernels are lacking off-diagonal elements in spin space. To improve upon the X-only treatment, we use the self-interaction corrected (SIC) LDA for $`A_{\mathrm{xc}}`$:
$$A_{\mathrm{xc}}^{\mathrm{SIC}}=_{\mathrm{}}^{t_1}๐t\left(E_{\mathrm{xc}}^{\mathrm{LDA}}[n_{}\left(t\right),n_{}\left(t\right)]\underset{i\sigma }{}E_{\mathrm{xc}}^{\mathrm{LDA}}[n_{i\sigma }\left(t\right),0]\frac{1}{2}\underset{i\sigma }{}d^3rd^3r^{}\frac{n_{i\sigma }(๐ซ,t)n_{i\sigma }(๐ซ^{},t)}{\left|๐ซ๐ซ^{}\right|}\right)$$
(11)
which is an orbital-dependent functional as well due to the explicit dependence on the orbital densities
$$n_{i\sigma }(๐ซ,t)=\left|\varphi _{i\sigma }(๐ซ,t)\right|^2\left(i=1,2,\mathrm{},N/2\right).$$
(12)
\] An improvement over both ALDA and exact exchange might be provided by correcting ALDA for self-interaction error. Within the adiabatic SIC-LDA, the exchange-correlation kernel reads
$`f_{\mathrm{xc}\sigma \sigma ^{}}^{\mathrm{TDOEP}\mathrm{SIC}}(๐ซ,๐ซ^{},\omega )=f_{\mathrm{xc}\sigma \sigma ^{}}^{\mathrm{ALDA}}(๐ซ,๐ซ^{},\omega )`$ (13)
$`{\displaystyle \frac{\delta _{\sigma \sigma ^{}}}{n_{0\sigma }\left(๐ซ\right)n_{0\sigma }\left(๐ซ^{}\right)}}{\displaystyle \underset{k}{}}f_{k\sigma }\left|\phi _{k\sigma }\left(๐ซ\right)\right|^2\left|\phi _{k\sigma }\left(๐ซ^{}\right)\right|^2\left({\displaystyle \frac{\delta v_{\mathrm{xc}\sigma }^{\mathrm{LDA}}(n_{k\sigma }\left(๐ซ\right),0)}{\delta n_{k\sigma }\left(๐ซ^{}\right)}}+{\displaystyle \frac{1}{\left|๐ซ๐ซ^{}\right|}}\right).`$
This expression reduces to the exact result (Eq. (9)) for one electron. For more than one electron, spurious self-interaction parallel-spin contributions in ALDA are corrected, for both exchange and correlation. The correction has no affect on anti-parallel spin contributions, leaving simply the ALDA result. We find simply
$$f_{\mathrm{xc}}^{\mathrm{SIC}}=f_{\mathrm{xc}}^{\mathrm{ALDA}}+\mathrm{\Delta }f_{\mathrm{xc}}^{\mathrm{SIC}}G_{\mathrm{xc}}^{\mathrm{SIC}}=G_{\mathrm{xc}}^{\mathrm{ALDA}}+\mathrm{\Delta }f_{\mathrm{xc}}^{\mathrm{SIC}}$$
(14)
where
$$\mathrm{\Delta }f_{\mathrm{xc}}^{\mathrm{SIC}}=\frac{2_kf_kn_k\left(๐ซ\right)n_k\left(๐ซ^{}\right)}{n\left(๐ซ\right)n\left(๐ซ^{}\right)}\left\{\delta \left(๐ซ๐ซ^{}\right)\frac{v_{\mathrm{xc},}^{\mathrm{hom}}(n_k,0)}{n_k\left(๐ซ\right)}+\frac{1}{\left|๐ซ๐ซ^{}\right|}\right\}.$$
(15)
## 3 Calculation of excitation energies
The linear density response has poles at the exact excitation energies of the interacting system (see, e.g., ). The key idea is to start from a particular KS orbital energy difference $`ฯต_{j\sigma }ฯต_{k\sigma }`$ (at which the Kohn-Sham response function (3) has a pole) and to use the formally exact representation (2) of the linear density response to calculate the shifts of the Kohn-Sham excitation energies towards the true excitation energies $`\mathrm{\Omega }`$. To extract these shifts from the density response, we cast Eq. (2) together with (4) into the form of an integral equation for $`n_{1\sigma }`$:
$`{\displaystyle \underset{\nu ^{}}{}}{\displaystyle d^3y^{}}`$ $`\left[\delta _{\sigma \nu ^{}}\delta \left(๐ซ๐ฒ^{}\right){\displaystyle \underset{\nu }{}}{\displaystyle d^3y\chi _{\mathrm{s}\sigma \nu }(๐ซ,๐ฒ;\omega )\left(\frac{1}{\left|๐ฒ๐ฒ^{}\right|}+f_{\mathrm{xc}\nu \nu ^{}}(๐ฒ,๐ฒ^{};\omega )\right)}\right]n_{1\nu ^{}}(๐ฒ^{},\omega )`$ (16)
$`={\displaystyle \underset{\nu }{}}{\displaystyle d^3y\chi _{\mathrm{s}\sigma \nu }(๐ซ,๐ฒ;\omega )v_{1\nu }(๐ฒ,\omega )}.`$
In general, the true excitation energies $`\mathrm{\Omega }`$ are not identical with the Kohn-Sham excitation energies $`ฯต_{j\sigma }ฯต_{k\sigma }`$, and the right-hand side of Eq. (16) remains finite for $`\omega \mathrm{\Omega }`$. The exact spin-density response $`n_{1\sigma }`$, on the other hand, exhibits poles at the true excitation energies $`\mathrm{\Omega }`$. Hence, the integral operator acting on $`n_{1\sigma }`$ on the left-hand side of Eq. (16) cannot be invertible for $`\omega \mathrm{\Omega }`$. This means that the integral operator acting on the spin-density vector in Eq. (16) is non-invertible (i.e., has vanishing eigenvalues) at the physical excitation energies. Rigorously, the true excitation energies $`\mathrm{\Omega }`$ are those frequencies where the eigenvalues $`\lambda \left(\omega \right)`$ of
$`{\displaystyle \underset{\nu ^{}}{}}{\displaystyle d^3y^{}\underset{\nu }{}d^3y}`$ $`\chi _{\mathrm{s}\sigma \nu }(๐ซ,๐ฒ;\omega )\left({\displaystyle \frac{1}{\left|๐ฒ๐ฒ^{}\right|}}+f_{\mathrm{xc}\nu \nu ^{}}(๐ฒ,๐ฒ^{};\omega )\right)\gamma _\nu ^{}(๐ฒ^{},\omega )=`$ (17)
$`=\lambda \left(\omega \right)\gamma _\sigma (๐ซ,\omega )`$
satisfy
$$\lambda \left(\mathrm{\Omega }\right)=1.$$
(18)
For notational brevity, we use double indices $`q(j,k)`$ to characterize an excitation energy; $`\omega _{q\sigma }ฯต_{j\sigma }ฯต_{k\sigma }`$ denotes the excitation energy of the single-particle transition $`\left(k\sigma j\sigma \right)`$. Consequently, we set $`\alpha _{q\sigma }:=f_{k\sigma }f_{j\sigma }`$ and
$$\mathrm{\Phi }_{q\sigma }\left(๐ซ\right):=\phi _{k\sigma }^{}\left(๐ซ\right)\phi _{j\sigma }\left(๐ซ\right)$$
(19)
as well as
$$\xi _{q\sigma }\left(\omega \right):=\underset{\nu ^{}}{}d^3y^{}d^3y\mathrm{\Phi }_{q\sigma }\left(๐ฒ\right)^{}\left(\frac{1}{\left|๐ฒ๐ฒ^{}\right|}+f_{\mathrm{xc}\sigma \nu ^{}}(๐ฒ,๐ฒ^{};\omega )\right)\gamma _\nu ^{}(๐ฒ^{},\omega ).$$
(20)
Without any approximation, equation (17) can be cast into matrix form
$$\underset{\sigma ^{}}{}\underset{q^{}}{}\frac{M_{q\sigma q^{}\sigma ^{}}\left(\omega \right)}{\omega \omega _{q^{}\sigma ^{}}+i\eta }\xi _{q^{}\sigma ^{}}\left(\omega \right)=\lambda \left(\omega \right)\xi _{q\sigma }\left(\omega \right),$$
(21)
with the matrix elements
$$M_{q\sigma q^{}\sigma ^{}}\left(\omega \right)=\alpha _{q^{}\sigma ^{}}d^3rd^3r^{}\mathrm{\Phi }_{q\sigma }^{}\left(๐ซ\right)\left(\frac{1}{\left|๐ซ๐ซ^{}\right|}+f_{\mathrm{xc}\sigma \sigma ^{}}(๐ซ,๐ซ^{};\omega )\right)\mathrm{\Phi }_{q^{}\sigma ^{}}\left(๐ซ^{}\right).$$
(22)
At the frequencies $`\omega =\mathrm{\Omega }`$, Eq. (21) can be written as
$$\underset{q^{}\sigma ^{}}{}\left(M_{q\sigma q^{}\sigma ^{}}\left(\mathrm{\Omega }\right)+\delta _{q\sigma q^{}\sigma ^{}}\omega _{q\sigma }\right)\beta _{q^{}\sigma ^{}}\left(\mathrm{\Omega }\right)=\mathrm{\Omega }\beta _{q\sigma }\left(\mathrm{\Omega }\right),$$
(23)
where we have defined
$$\beta _{q\sigma }\left(\mathrm{\Omega }\right):=\xi _{q\sigma }\left(\mathrm{\Omega }\right)/\left(\mathrm{\Omega }\omega _{q\sigma }\right).$$
(24)
The solutions $`\mathrm{\Omega }`$ of the nonlinear matrix-equation (23) are the physical excitation energies. The inevitable truncation of the infinite-dimensional matrix in Eq. (23) amounts to the approximation of $`\chi ^{\left(0\right)}`$ by a finite sum
$$\chi ^{\left(0\right)}(๐ซ,๐ซ^{},\omega )\underset{\sigma =}{}\underset{q}{\overset{Q}{}}\alpha _q\frac{\mathrm{\Phi }_q\left(๐ซ\right)\mathrm{\Phi }_q\left(๐ซ^{}\right)}{\omega \omega _{q\sigma }}.$$
(25)
This truncation explicitly takes into account numerous poles of the noninteracting response function. In any adiabatic approximation to the XC kernel, the matrix elements $`M_{q\sigma q^{}\sigma ^{}}`$ are real and frequency independent. In this case the excitation energies $`\mathrm{\Omega }`$ are simply the eigenvalues of the ($`Q\times Q`$) matrix $`M_{q\sigma q^{}\sigma ^{}}\left(\mathrm{\Omega }=0\right)+\delta _{q\sigma ,q^{}\sigma ^{}}\omega _{q\sigma }`$. For bound states of finite systems we encounter well-separated poles in the linear density response. In our calculations, we include many such poles, but only those of bound states, ignoring continuum contributions. The nature and size of the error this introduces has been studied by van Gisbergen et al., and does not affect the qualitative conclusions found in this work.
A simple and extremely instructive case is when we expand about a single KS-orbital energy difference $`\omega _{p\tau }`$ . The physical excitation energies $`\mathrm{\Omega }`$ are then given by the solution of
$$\lambda \left(\mathrm{\Omega }\right)=\frac{A\left(\omega _{p\tau }\right)}{\mathrm{\Omega }\omega _{p\tau }}+B\left(\omega _{p\tau }\right)+\mathrm{}=1.$$
(26)
For non-degenerate single-particle poles $`\omega _{p\tau }`$, the coefficients in Eq. (26) are given by
$$A\left(\omega _{p\tau }\right)=M_{p\tau p\tau }\left(\omega _{p\tau }\right)$$
(27)
and
$$B\left(\omega _{p\tau }\right)=\frac{dM_{p\tau p\tau }}{d\omega }|_{\omega _{p\tau }}+\frac{1}{M_{p\tau p\tau }\left(\omega _{p\tau }\right)}\underset{q^{}\sigma ^{}p\tau }{}\frac{M_{p\tau q^{}\sigma ^{}}\left(\omega _{p\tau }\right)M_{q^{}\sigma ^{}p\tau }\left(\omega _{p\tau }\right)}{\omega _{p\tau }\omega _{q^{}\sigma ^{}}+i\eta }.$$
(28)
If the pole $`\omega _{p\tau }`$ is $`\mathrm{}`$-fold degenerate, $`\omega _{p_1\tau _1}=\omega _{p_2\tau _2}=\mathrm{}=\omega _{p_{\mathrm{}}\tau _{\mathrm{}}}\omega _0,`$ the lowest-order coefficient $`A`$ in Eq. (26) is determined by a $`\mathrm{}`$-dimensional matrix equation
$$\underset{k=1}{\overset{\mathrm{}}{}}M_{p_i\tau _ip_k\tau _k}\left(\omega _0\right)\xi _{p_k\tau _k}^{\left(n\right)}=A_n\left(\omega _0\right)\xi _{p_i\tau _i}^{\left(n\right)},i=1\mathrm{}\mathrm{},$$
(29)
with $`\mathrm{}`$ different solutions $`A_1\mathrm{}A_{\mathrm{}}`$. For excitation energies $`\mathrm{\Omega }`$ close to $`\omega _0`$, the lowest-order term of the above Laurent expansion will dominate the series. In this single-pole approximation (SPA), the excitation energies $`\mathrm{\Omega }`$ satisfy Eq. (26) reduces to
$$\lambda _n\left(\mathrm{\Omega }\right)\frac{A_n\left(\omega _0\right)}{\mathrm{\Omega }\omega _0}=1.$$
(30)
The condition (18) and its complex conjugate, $`\lambda ^{}\left(\mathrm{\Omega }\right)=1`$, finally lead to a compact expression for the excitation energies.
$$\mathrm{\Omega }_n\omega _0+\mathrm{}A_n\left(\omega _0\right).$$
(31)
For closed-shell systems, every Kohn-Sham orbital eigenvalue is degenerate with respect to spin, i.e. the spin multiplet structure is absent in the bare Kohn-Sham eigenvalue spectrum. Within the SPA, the dominant terms in the corrections to the Kohn-Sham eigenvalues towards the true multiplet energies naturally emerge from the solution of the ($`2\times 2`$) eigenvalue problem
$$\underset{\sigma ^{}=,}{}M_{p\sigma p\sigma ^{}}\left(\omega _0\right)\xi _{p\sigma ^{}}\left(\omega _0\right)=A\xi _{p\sigma }\left(\omega _0\right).$$
(32)
Then, the resulting excitation energies are:
$$\mathrm{\Omega }_{1,2}=\omega _0+\mathrm{}\left\{M_{pp}\pm M_{pp}\right\}.$$
(33)
Using the explicit form of the matrix elements (22) one finds<sup>2</sup><sup>2</sup>2Since we are dealing with spin saturated systems, we have dropped the spin-index of $`\mathrm{\Phi }_{p\sigma }`$.
$`\mathrm{\Omega }_1`$ $`=`$ $`\omega _0+2\mathrm{}{\displaystyle d^3rd^3r^{}\mathrm{\Phi }_p^{}\left(๐ซ\right)\left(\frac{1}{\left|๐ซ๐ซ^{}\right|}+f_{\mathrm{xc}}(๐ซ,๐ซ^{};\omega _0)\right)\mathrm{\Phi }_p\left(๐ซ^{}\right)}`$ (34)
$`\mathrm{\Omega }_2`$ $`=`$ $`\omega _0+2\mathrm{}{\displaystyle d^3rd^3r^{}\mathrm{\Phi }_p^{}\left(๐ซ\right)G_{\mathrm{xc}}(๐ซ,๐ซ^{};\omega _0)\mathrm{\Phi }_p\left(๐ซ^{}\right)}.`$ (35)
The kernel $`G_{\mathrm{xc}}`$ embraces the exchange and correlation effects in the Kohn-Sham equation for the linear response of the frequency-dependent magnetization density $`m(๐ซ,\omega )`$ . For unpolarized systems, the weight of the pole in the spin-summed susceptibility (both for the Kohn-Sham and the physical systems) at $`\mathrm{\Omega }_2`$ is exactly zero, indicating that these are the optically forbidden transitions to triplet states. The singlet excitation energies are at $`\mathrm{\Omega }_1`$. In this way, the SPA already gives rise to a spin-multiplet structure in the excitation spectrum. We use SPA to understand the results of different approximations, since it simply relates the calculated shifts from KS eigenvalues to matrix elements of the XC kernel.
At this point we stress that the TDDFT formalism for the calculation of excitation energies involves three different types of approximations:
1. In the calculation of the Kohn-Sham orbitals $`\phi _k\left(๐ซ\right)`$ and their eigenvalues $`ฯต_k`$, one employs some approximation of the static XC potential $`v_{\mathrm{xc}}`$.
2. Given the stationary Kohn-Sham orbitals and the ground state density, the functional form of the XC kernel $`f_{\mathrm{xc}\sigma \sigma ^{}}`$ needs to be approximated in order to calculate the matrix elements defined in Eq. (22).
3. Once the matrix elements are obtained, the infinite-dimensional eigenvalue problem (21) (or, equivalently, (23)) must be truncated in one way or another.
In the following, we are going to investigate the relative importance of the approximations (1.) and (2.). Furthermore, truncation effects will be estimated by comparing the results obtained in SPA (34,35) with the solution of the โfullโ problem (23) which is based on using up to 38 bound virtual orbitals.
## 4 Results for the Helium Atom
In this section we report numerical results for excitation energies of the He atom. The stationary Kohn-Sham equations were solved numerically on a radial grid (i.e. without basis set expansion) using a large number of semi-logarithmically distributed grid points up to a maximum radius of several hundred atomic units in order to achieve high accuracy the Rydberg states ($`n10`$) as well.
### 4.1 Exact Kohn-Sham potential
To eliminate the errors (1.) associated with the approximation for the ground-state KS potential, we employ the exact XC potential of the He atom to generate the stationary Kohn-Sham orbitals $`\phi _k\left(๐ซ\right)`$ and their eigenvalues $`ฯต_k`$. This isolates the effects which exclusively arise due to the approximations (2.) and (3.). The potential data provided by Umrigar and Gonze were interpolated nonlinearly for $`r10`$ atomic units. Around $`r=10`$ atomic units, the XC potential is almost identical to $`1/r`$. This behavior was used as an extrapolation of the exact exchange-correlation potential to larger distances.
Tables 1 and 2 show the excitation energies of neutral helium calculated with the exact exchange-correlation potential. The results are compared with a highly accurate nonrelativistic variational calculation of the eigenstates of Helium. It is a remarkable fact that the Kohn-Sham excitation energies $`\omega _{jk}=ฯต_jฯต_k`$ are already very close to the exact spectrum, and, at the same time, are always in between the singlet and the triplet energies.
Based on these eigenvalue differences, we have calculated the shifts towards the true excitation energies using several approximations for the exchange-correlation kernels $`f_{\mathrm{xc}}`$:
* The adiabatic local density approximation (ALDA), with the inclusion of correlation contributions in the parametrization of Vosko, Wilk and Nusair .
* The approximate X-only time-dependent OEP (TDOEP) kernel of Eq. (9), which is based on the time-dependent Fock expression, and
* The approximate TDOEP-SIC kernel from Eq. (13) with the parametrization of Ref. for the correlation contributions.
The columns denoted by โfullโ show the corresponding excitation energies $`\mathrm{\Omega }_i`$ which are obtained as eigenvalues obtained from the (truncated) matrix equation (23). To investigate the effects of the truncation of the matrix equation (23) we compare the difference between the single-pole approximation (SPA) and the fully coupled results. The matrix equation (23) was solved using $`N=34`$ unoccupied Kohn-Sham orbitals of $`s`$ or $`p`$ symmetry. For each symmetry class the resulting dimension of the (fully coupled but truncated) matrix in Eq. (23) is $`\left(4N\times 4N\right)`$ (due to the spin-degeneracy of the KS orbitals of Helium and the fact that the frequency-dependent Kohn-Sham response function is symmetric in the complex plane with respect to the imaginary axis). Thereby, convergence of the results to within $`10^6`$ atomic units was reached within the space of bound states. When comparing the results from the SPA with the results from the fully coupled matrix, we observe only a small change in the resulting excitation energies (from a few hundredth of a percent to at most one half percent), independent of the functional form of the exchange-correlation kernel. Thus we conclude that in helium the single-pole approximation gives the dominant correction to the Kohn-Sham excitation spectrum. Hence, starting from the Kohn-Sham eigenvalue differences as zeroth order approximation to the excitation energies, the SPA can be used for the assignment of the excitation energies which are obtained as eigenvalues from Eq. (23). Recent studies using basis set expansions indicate that further improvement of the fully coupled results can be expected from the inclusion of continuum states. The general trends of the results however, are not affected.
In figure 1 we have plotted some typical excitation energies taken from the column headed โfullโ of table 1 and 2,
We can understand the trends in this figure by analyzing the results in terms of the single-pole approximation. For the single-particle excitations in helium, the single-pole approximation leads to two-dimensional matrix equations for the excitation energies (c.f. Eqs. (32) - (34)). In the following, the notation
$$\widehat{๐ช}:=d^3rd^3r^{}\mathrm{\Phi }_p^{}\left(๐ซ\right)\widehat{๐ช}(๐ซ,๐ซ^{})\mathrm{\Phi }_p\left(๐ซ^{}\right)$$
(36)
will be used for the matrix elements of the two particle operators $`\widehat{๐ช}`$ involved in the calculation. Then, in the SPA,
$$\mathrm{\Omega }_p^{\mathrm{singlet}}=\omega _p+2W+2f_{\mathrm{xc}},\mathrm{\Omega }_p^{\mathrm{triplet}}=\omega _p+2G_{\mathrm{xc}},\mathrm{\Delta }\mathrm{\Omega }_p=2\left(W+f_\mathrm{c}G_\mathrm{c}\right),$$
(37)
where $`\mathrm{\Delta }\mathrm{\Omega }_p`$ is the singlet-triplet splitting. Within the various approximations to the kernel, these levels become
$`\mathrm{\Omega }_p^{\mathrm{sing}}`$ $`=`$ $`\omega _p+W,\mathrm{\Omega }_p^{\mathrm{triplet}}=\omega _pW,\left(\mathrm{X}\mathrm{only}\right)`$ (38)
$`=`$ $`\omega _p+2W+2f_{\mathrm{xc}}^{\mathrm{ALDA}},=\omega _p+2G_{\mathrm{xc}}^{\mathrm{ALDA}},\mathrm{ALDA}`$ (39)
$`=`$ $`\omega _p+W+2f_\mathrm{c}^{\mathrm{ALDA}}f_\mathrm{c}^{\mathrm{orb}},=\omega _pW+2G_\mathrm{c}^{\mathrm{ALDA}}f_\mathrm{c}^{\mathrm{orb}}.\mathrm{SIC}`$ (40)
We begin our analysis with the splitting. In the simplest case, the TDOEP X-only kernel, we see that the singlet transitions are always overestimated, while the triplets are always underestimated. Since our TDOEP treatment is exact for exchange in this case, this underscores the importance of correlation. In particular, since $`f_\mathrm{x}=G_\mathrm{x}=W/2`$, the splitting is just $`2W`$. This matrix element is always positive, correctly putting the singlet above the triplet, but the splitting is typically far too big. We demonstrate the effect of this in table 3, in which we compare splittings with and without correlation.
To see why inclusion of correlation always reduces the splitting, we note the sign and magnitude of matrix elements, within ALDA. Even though both $`f_{\mathrm{xc}}^{\mathrm{ALDA}}`$ and $`G_{\mathrm{xc}}^{\mathrm{ALDA}}`$ are negative, because they are dominated by their exchange contributions, we find
$$f_\mathrm{c}^{\mathrm{ALDA}}<G_\mathrm{c}^{\mathrm{ALDA}}<0$$
(41)
because in Eq. (6) antiparallel correlation dominates over parallel correlation. Thus the ALDA correlation contribution to the splitting is always negative in SPA. Note that the SIC treatment of the splitting is only marginally better than in ALDA because, within SPA, the SIC splitting is identical to that of ALDA.
To analyze the separate levels, we need the magnitude of the SIC corrections:
$$f_\mathrm{c}^{\mathrm{ALDA}}<f_\mathrm{c}^{\mathrm{orb}}:=\frac{\delta v_\mathrm{c}^{\mathrm{LDA}}[n_{k\sigma },0]}{\delta n_{k\sigma }}<0,$$
(42)
but the numerical values of both matrix elements differ by less than 8%. Moreover,
$$G_\mathrm{c}^{\mathrm{ALDA}}>\left|f_\mathrm{c}^{\mathrm{orb}}\right|>0.$$
(43)
Looking at the singlet excitation energies of table 1 we see that in ALDA, the s-levels are too high (up to 10 mH), whereas the p-levels are too low (by up to 0.4 mH). In X-only TDOEP, the s-levels drop (by up to 3 mH), approaching the exact values, but the rise of the p-states (by up to 8mH) is too high. Incorporating explicit correlation terms by using the TDOEP-SIC kernel, the singlet lines correctly drop further (in comparison to the X-only results by up to 1 mH) since $`2f_\mathrm{c}^{\mathrm{ALDA}}f_\mathrm{c}^{\mathrm{orb}}f_\mathrm{c}^{\mathrm{ALDA}}`$ in Eq. (40) is always a negative contribution. But still, the p-states are too high. Regarding the triplet excitation energies of table 2, the ALDA s-states are too high by at most 6mH, but the p-states are almost identical to the exact values. In X-only TDOEP, the triplet states experience a strong downshift from the Kohn-Sham excitation energies up to 25 mH, originating from the term $`W`$ (see Eq. (40)). In TDOEP-SIC, this downshift is partly screened by the positive correlation contributions $`2G_\mathrm{c}^{\mathrm{ALDA}}f_\mathrm{c}^{\mathrm{orb}}`$, as can be seen from Eqs. (40), (42) and (43). This leads to an excellent agreement with the exact values for the s-states. However, these correlation terms are too large for the p-states. Since $`G_\mathrm{c}^{\mathrm{ALDA}}>f_\mathrm{c}^{\mathrm{ALDA}}`$, the rise of the triplet is always bigger than the dropping of the singlet.
### 4.2 Approximate Kohn-Sham potentials
Next we explore the effect of approximate exchange-correlation potentials $`v_{\mathrm{xc}}`$ on the calculated excitation spectrum of the He atom. We do not even report results within LDA and generalized gradient approximations (GGA), since these potentials only support a few virtual states, so that many of the transitions reported here do not even exist in such calculations. (This problem is worst in small atoms, is less pronounced in molecules, and irrelevant in solids).
To produce a correct Rydberg series, the XC potential must decay as $`1/r`$, an exact exchange effect. Hence we examine the OEP X-only potential (which, for two-electron systems is identical to the Hartree-Fock potential) and the OEP-SIC potential. Both potentials show the correct behavior for large distances from the nucleus, and support of all the Rydberg states is guaranteed.
Tables 4 and 5 show the approximate Kohn-Sham excitation energies and the corresponding corrected excitation energies calculated from the approximate Kohn-Sham eigenvalues and orbitals of the X-only potential; tables 6 and 7 are their analogs from the OEP-SIC calculation. The Kohn-Sham orbital energy differences are almost uniformly shifted to larger values compared to the orbital energy differences of the exact Kohn-Sham potential. The shift ranges from 13.6 mH for the lowest excitation energy to 14.2 mH for excitation energies $`\mathrm{\Omega }_n`$ with $`n4`$ for the X-only potential. The latter shift is exactly the difference between the exact 1s eigenvalue ($`ฯต_{1s}^{\mathrm{exact}}=0.90372`$ a.u.) and the more strongly bound 1s eigenvalue of the X-only potential ($`ฯต_{1s}^{\mathrm{X}\mathrm{only}}=0.91796`$ a.u.). Similarly, the Kohn-Sham eigenvalue differences calculated in OEP-SIC are shifted by up to 44.5 mH, which again is equal to the difference between the 1s eigenvalues of the exact Kohn-Sham potential and the KS potential in OEP-SIC. In OEP-SIC, the correlation potential is attractive at all points in space. Hence, including SIC-correlation contributions into the OEP worsens the occupied orbital eigenvalue. To summarize, the inclusion of correlation contributions to the ground state potential mostly affects only the occupied state; the virtual states are almost exact, i.e., they are almost independent of the choice of the correlation potential. The He Kohn-Sham orbitals exhibit a Rydberg-like behavior already for relatively low quantum numbers $`n`$ : already the lower virtual states are mostly determined by the large-$`r`$ behavior of the Kohn-Sham potential, which is governed by the exchange contribution.
As a consequence, the corrections to the Kohn-Sham orbital energy differences, calculated on the approximate orbitals, are very close to the corrections calculated from the exact Kohn-Sham orbitals. This is most apparent from the singlet-triplet splittings given in tables 3 and 8: the splittings depend more strongly on the choice of the XC kernel than on the choice of the potential.
However, for the excitation energies, the differences among the various approximations of the exchange correlation kernel are smaller than the differences in the Kohn-Sham excitation energies coming from different potentials. This reflects the fact that the resulting orbitals are rather insensitive to different approximations of the potential. Hence, the corrections themselves, calculated with approximate XC kernels will not cancel the shortcomings of an approximate exchange potential. Tables 4 \- 7 show that the corrections go in the right direction only for the singlet states, which are always lower than the corresponding Kohn-Sham orbital energy differences. In other approximations, like the LDA and in the popular GGAs for instance, this will be even more severe: There the highest occupied orbital eigenvalue is in error by about a factor of two, due to spurious self-interaction. There may be error cancellations for the lower Kohn-Sham eigenvalue differences, but in general one should not expect to get a reliable (Kohn-Sham) spectrum in LDA and GGAs, because the respective potentials have the wrong behavior for large $`r`$. In addition, this causes the number of (unoccupied) bound KS states to be finite.
In total, the inaccuracies introduced by approximate ground state Kohn-Sham potentials are substantial, but mostly reside in the occupied eigenvalue for He. It is very unlikely that these defects will be cured by better approximations of $`f_{\mathrm{xc}}`$ alone, since the terms containing $`f_{\mathrm{xc}}`$ only give corrections to the underlying Kohn-Sham eigenvalue spectrum. Hence, the quantitative calculation of excitation energies heavily depends on the accuracy of the ground-state potential employed.
## 5 Results for the Beryllium Atom
### 5.1 Exact Kohn-Sham potential
The beryllium atom serves as a further standard example for first principles treatments: besides numerous quantum chemical studies (e.g. ), a highly accurate ground-state exchange-correlation potential, obtained from quantum Monte-Carlo methods , is available for this system. With this potential, we calculated accurate Kohn-Sham orbitals and orbital energies of the beryllium atom. In each symmetry class (s, p, and d), up to 38 virtual states were calculated on a radial grid similar to the one used in section 4.
In tables 9 and 10 we report the excitation energies for the 11 lowest excitations of singlet and triplet symmetry. As in helium, the orbital energies of the accurate potential lie always in between the experimental singlet and triplet energies. However, the experimentally measured singlet-triplet separations in beryllium are much larger than in the helium atom (cf. the last columns given in tables 3 and 11). Accordingly, to achieve agreement with the experimental data, appreciable shifts of the Kohn-Sham eigenvalue differences are needed.
For the singlet excitation spectrum, given in table 9, the TDDFT corrections yield significantly improved excitation energies compared to spectrum of the bare Kohn-Sham eigenvalue differences, with average errors reduced by a factor of about 3 regardless of which kernel is used. The most distinct improvement towards experiment is achieved for the singlet 2P excitation, where the Kohn-Sham eigenvalue difference is off by 32% (61 mHartree) from the experimental value.
For the remaining singlet excitations, the TDOEP-SIC kernel yields the best improvement upon the bare Kohn-Sham spectrum. From figure 2, where the errors for each singlet excitation energy are plotted, we see two competing effects: the errors increase with progressing angular momentum (with the error of the 3d-states being largest), but decrease with progressing principal quantum number $`n`$. Note that ALDA has the largest errors for the d-states, presumably due to its inability to account for orbital nodes.
For the triplet spectrum given in table 10, the transition to the 2p state is clearly problematic, presumably because of its small magnitude. In particular, the TDOEP X-only calculation greatly underestimates the downshift away from the KS eigenvalue difference. Because of this effect, we also report average errors with this transition excluded. All Kohn-Sham orbital excitations experience a downshift in the ALDA and TDOEP X-only calculation. In ALDA, this leads to an overall improvement of the spectrum by a more than a factor of 2. The downshift in TDOEP X-only results is too strong, and this behavior is partly corrected in the TDOEP-SIC. However, due to overcorrections for the higher excitation energies, the average reduction in error over the Kohn-Sham excitation spectrum is only a factor of 1.2.
The errors for the triplet excitation energies are plotted in figure 3. Clearly, the errors of both the Kohn-Sham eigenvalue spectrum and the corresponding corrections decrease again with progressing quantum number. Together with the errors plotted in figure 2 this signals that the Rydberg-like transitions to states with high principal quantum number $`n`$ are already close to the eigenvalue differences of the accurate Kohn-Sham potential.
The singlet-triplet separations from equation (23) are given in table 11 for the three different approximate XC kernels. Like in helium, the singlet-triplet splittings are overestimated by about a factor of two for the $`S`$ and $`P`$ transitions if the (diagonal) TDOEP x-only kernel is used. The splittings of the $`D`$ levels, however, appear too small by about a factor of two. By the inclusion of correlation contributions to the kernels, the splittings of the $`S`$ and $`P`$ levels are consistently (and usually correctly) reduced. However, for the D states, this correction is always too large, and leads to a reversal of the singlet and triplet energies.<sup>3</sup><sup>3</sup>3This effect can also be observed in the Helium atom. The exact values of the singlet-triplet splittings of the $`D`$ states in helium however, are by two orders of magnitude smaller than in beryllium. From the singlet-triplet splitting in Eq. (40) which, in the SPA, hold for any system since the diagonal terms of $`f_{\mathrm{xc}\sigma \sigma ^{}}`$ cancel, this behavior can be traced back to the overestimation of correlation contributions in LDA (in small systems). Self-interaction corrections are not expected to cure this shortcoming, for the reason that to leading order the self-interaction correction terms cancel in the expressions for the splittings, similar to the way shown in section 4. Accordingly, the separations in TDOEP-SIC and ALDA are of similar quality, which can be seen from columns one and three in table 11. The TDOEP X-only results on the other hand, although too small, show the correct ordering of singlet and triplet levels.
With increasing excitation energy, the difference between the results in SPA and the full solution is reduced, as was already observed in the case of helium. The drastic change of the triplet 2$`P`$ state in TDOEP X-only seems to be an artifact of the specific approximation to the exchange-correlation kernel, since the results in the SPA and the full calculation for this particular excitation energy only differ by 10% if the ALDA is used for $`f_{\mathrm{xc}\sigma \sigma ^{}}`$.
### 5.2 Approximate Kohn-Sham potentials
The results from using different approximate exchange-correlation potentials for the Be atom to calculate the Kohn-Sham eigenvalues and orbitals are given in tables 12 to 15.
The errors towards the experimental excitation energies are compiled in figures 2 and 3 for the singlet and triplet series. Looking first at the spectra of the bare Kohn-Sham eigenvalues (represented in figures 2 and 3 by the points connected with thick lines), we notice that the โHOMO-LUMOโ gap is almost independent of the approximation of $`v_{\mathrm{xc}}`$ employed. This is in sharp contrast with the He atom case. The correlation contributions cancel for the lowest excitation energy, and we must classify this as a non-Rydberg state. For the higher states, the situation is different: Starting from the excitation to the 3s level, the series of single-particle energy differences appear almost uniformly shifted with respect to the series of the exact potential, preserving the typical pattern of their deviation from the experimentally measured spectrum. The shifts amount to -14 mH for the OEP-SIC potential, and -34 mH for the X-only KLI potential. As in helium, these shifts are equal to the differences in the eigenvalues of the highest occupied Kohn-Sham orbital: For the accurate potential ($`ฯต_{2s}^{\mathrm{accurate}}=0.3426`$ a.u. ), the highest occupied orbitals are more strongly bound than in OEP-SIC ($`ฯต_{2s}^{\mathrm{OEP}\mathrm{SIC}}=0.3285`$ a.u.) and in X-only KLI ($`ฯต_{2s}^{\mathrm{X}\mathrm{onlyKLI}}=0.3089`$ a.u.). Thus, among the virtual states, only the 2p orbital is appreciably influenced by the details of the ground state potential. For the higher lying states, the long-range behavior of the Kohn-Sham potential dominates. Its $`1/r`$ behavior is correctly reproduced both in X-only KLI as well as in OEP-SIC. For larger systems, more low-lying excitations can be accurately approximated, but eventually, for any finite system, the Rydberg excitations will show errors due to errors in the ionization potential. Casida et al. have studied which excitations can be well-approximated with present functional approximations to the potential.
Regarding the corrections for the singlet excitation energies calculated from Eq. (23), the first excited state (2p) experiences the largest correction, irrespective of the exchange-correlation potential employed. Moreover, the results using different approximate exchange-correlation kernels agree within 10 mH. For the remaining singlet excitation energies, the calculated corrections using the approximate Kohn Sham orbitals are almost identical to the corrections which are obtained from using the accurate Kohn-Sham orbitals. Hence, in figure 2 the errors for the excitations to 3s through 6p show the same pattern of deviations, only shifted by the error in the respective eigenvalue of the 2s orbital. On average, the resulting singlet excitation energies are closest to experiment, if the approximate exchange-correlation potential $`v_{\mathrm{xc}}`$ is combined with the corresponding approximation of the exchange-correlation kernel $`f_{\mathrm{xc}\sigma \sigma ^{}}`$.
From tables 13 and 15 as well as from figure 3, the behavior of the triplet spectra is similar, but less unequivocal for the triplet 2p state. For this particular state, the corrections spread on the order of 100mH, prevalently due to the significant overcorrection of the X-only TDOEP kernels. However, the resulting triplet 2p excitation energy almost exclusively depends on the approximation to the exchange correlation kernel rather than on the exchange-correlation potential employed. On the average, apart from the higher excitations in OEP-SIC (c.f. table 14), the best triplet spectra are obtained if the ALDA is used for the exchange-correlation kernel, but this appears to be a fortuitous cancellation of errors. The approximate Kohn-Sham excitation energies, except for the 2p state, are already incorrectly lower than the experimental triplet levels. Any further lowering, although correct for the eigenvalue-differences of the exact Kohn-Sham potential, actually worsens the triplet spectra which are calculated on the basis of an approximate exchange-correlation potential. Since the shifts are reduced by correlation contributions in the kernels, the over-corrections become less severe for the ALDA and TDOEP-SIC kernels. Another apparent error cancellation is that when calculating the lowest excitation energy ($`2s2p`$) from approximate exchange-correlation potentials, the SPA-results are always closer to experiment than the full results. This might be related to the fact that for TDOEP X-only, SPA yields the exact first-order shift in energy levels in Gรถrling-Levy perturbation theory, while the โfullโ calculation does not. In cases where there are large differences between SPA and full results, the SPA might be more reliable for these reasons.
The fact that the corrections to the Kohn-Sham eigenvalue differences only weakly depend on the approximation of the exchange-correlation potential $`v_{\mathrm{xc}}`$, is also reflected in table 16, where the singlet-triplet separations in Be, calculated using the X-only KLI and OEP-SIC potentials are given. The numerical values are close to the results for the accurate Be exchange correlation potential in table 11. Again, the obtained splittings are more sensitive to the approximation of $`f_{\mathrm{xc}\sigma \sigma ^{}}`$ than to the approximation of the potential $`v_{\mathrm{xc}}`$.
## 6 Summary and Conclusion
In this work we aimed at an assessment of the influence of the three different types of approximations (i.e. (i) the XC potential $`v_{\mathrm{xc}}`$, (ii) the XC kernel $`f_{\mathrm{xc}}`$ and (iii) truncation of the space of virtual excitations) inherent in the calculation of excitation energies from TDDFT. We calculated the discrete optical spectra of helium and beryllium, two of the spectroscopically best known elements, using the exact exchange-correlation potential, the KLI-X-only potential and the the KLI-SIC potential for $`v_{xc}`$ (all three potentials are falling off like $`1/r`$ as $`r\mathrm{}`$). These were combined with three approximations for the XC kernel: The adiabatic LDA (ALDA), the TDOEP X-only kernel and the TDOEP-SIC kernel. The results are given both in the single-pole approximation (SPA) and for a โfullโ calculation, where as many virtual states as possible (typically about 30 to 40) entered the calculation. The analysis of these combinations reveals the following trends: First of all, the choice of $`v_{\mathrm{xc}}`$ on the calculated spectrum has the largest effect on the calculated spectra. The inaccuracies introduced by approximate ground state Kohn-Sham potentials (even those including exact exchange) can be quite substantial. This is especially true for the higher excited states, which appear almost uniformly shifted from the true excitation energies. We observe that this shift is closely related to the absolute value of the highest occupied eigenvalue, which, in exact DFT, is equal to the first ionization potential of the system at hand.
For the lower excitation energies, an error cancellation occurs, making these excitations less sensitive to the choice of the exchange-correlation potential. This error cancellation however, ceases to work the more the excited states behave like Rydberg states. For Helium, this is already the case for the first excited state. Hence, in improving the calculation of excitation energies from TDDFT requires an improved exchange-correlation potential in the first place. The most important requirement for such a potential would be that its highest occupied eigenvalue reproduces the experimental ionization potential as closely as possible. Empirically, one could introduce a โscissors-operatorโ similar to the one introduced by Levine and Allan , which shifts the Rydberg states by a constant being equal to the difference between the highest occupied eigenvalue and the negative of the experimental ionization potential. But such a procedure would not produce a first-principles calculation. In our opinion however, the construction of approximate exchange-correlation potentials based on orbital functionals would be the method of choice for the future.
The effect of the choice of the exchange-correlation kernel on the calculated spectra, in turn, is much less pronounced. However, its relative importance increases whenever the โfirst-orderโ effects, originating from $`v_{\mathrm{xc}}`$ cancel. This is the case for the values of the singlet-triplet splittings and the lower excitation energies of Be. For these โsecond-order effectsโ, the correlation contributions contained in $`f_{\mathrm{xc}}`$ are important. We observe that the ALDA for the XC kernels already leads to quite reasonable results which are only marginally improved by using the more complicated TDOEP-SIC kernel. Besides the missing frequency dependence, correlation contributions are hard to model on top of an exact exchange treatment and one might speculate that the ALDA takes advantage from a fortuitous error cancellation between exchange and correlation effects. Again, we expect only orbital functionals to manage a marked improvement over the ALDA, which, up to now, can has been the workhorse of TDDFT.
Finally, the inevitable truncation of the space of virtual excitations is appreciable only for the lowest lying states. In most cases, the results of the single-pole approximation (SPA), which, in the nondegenrate case, merely requires a pair of โinitialโ and โfinalโ KS states are close to the results obtained from using more configurations.
We thank C. Umrigar for providing us with the data of the exact exchange-correlation potentials for helium and beryllium. This work was supported in part by the Deutsche Forschungsgemeinschaft and by NATO. K.B. acknowledges support of the Petroleum Research Fund and of NSF grant CHE-9875091. |
warning/0001/nucl-th0001010.html | ar5iv | text | # Pion abundance and entropy in the hydrodynamical description of relativistic nuclear collisions
\[
## Abstract
We show that a hydrodynamical model with continuous particle emission instead of sudden freeze out can explain both the strange particle abundances and pion abundance from NA35 without extra assumption (e.g., sequential freeze out, modified equation of state, sudden plasma hadronization,โฆ). In this scenario, the observation of a larger pion abundance is natural and does not imply a higher initial entropy and early plasma phase.
\]
The main purpose of the ongoing and future heavy ion programs at the high energy laboratories CERN (Switzerland) and Brookhaven National Laboratory (U.S.A.) is to investigate the formation of hot dense matter and the possible transition from hadronic matter to quark gluon plasma. Various possible signatures of the appearence of a quark gluon plasma (thereafter QGP) have been suggested: entropy increase (due to the release of new degrees of freedom, namely color), strangeness increase (due to enhanced strange quark production and faster equilibration), J/$`\psi `$ supression (due to color screening), production of leptons and photons (emitted from a thermalized QGP and unaffected by strong interactions), etc. These signals have been studied extensively experimentally (see for exemple ). In this paper, we concentrate on two signatures: strangeness increase which has been observed between p-p and nucleus-nucleus at a fixed energy and entropy production which is studied via pion measurements.
A major problem to trace back any signature unambiguously to a quark gluon phase is that it is still unknown which theoretical description describes best high energy nuclear collisions. On one extreme, one might use a microscopic model. Hadronic microscopic models fail to reproduce simultaneously strange and non-strange particle data in nucleon-nucleon collisions and central nucleus-nucleus collisions at SPS energies (see ). Partonic microscopic models are expected to work at energies higher than SPS (however see and references therein).
On the other extreme, one might use a thermal or hydrodynamical model. In such models, it is assumed that a fireball (region filled with dense hadronic matter or QGP in local thermal and chemical equilibrium) is formed in a high energy heavy ion collision and evolves. Hydrodynamical models have been used successfully to describe various kinds of data at AGS and SPS. In particular they are able to account for strangeness data but in their simplest version, fail to predict big enough pion abundances.
Since as already mentioned, both strangeness and pion or entropy productions are expected to be modified by the appearence of a quark gluon plasma, looking for a joint explaination (with or without plasma) of the relevant data is crucial. In the case of the thermal and hydrodynamical models mentioned above, various ways out of the pion problem have been proposed (see next section). In this paper, we study another possibility: use a more accurate emission mecanism, namely continuous emission, rather than standard freeze out, in a hydrodynamical description. This explaination has the advantage that no extra assumption is needed: once the initial conditions of the hydrodynamical expansion are fixed, both strangeness and pion yields come out with the right magnitude.
Hydrodynamical or thermal description with (standard) freeze out emission - First we remind what is the status of the standard hydrodynamical or thermal description of relativistic nuclear collisions. In this kind of description, hadrons are kept in chemical or thermal equilibrium until some decoupling criterion has become satisfied. An exemple of freeze out criterion often used is that a certain temperature and baryonic potential have been reached.
Since abundances are fixed by the chemical freeze out, the chemical freeze out parameters can be extracted by analyzing experimental particle abundances. This has been done by many groups . The models have some variations between them, but as a general rule, while they can reproduce strange particle abundances, they underpredict the pion abundance. This was first noted by in a study of NA35 data and emphasized by in an analysis of the WA85 strange particle ratios and EMU05 specific net charge $`D_q(N^+N^{})/(N^++N^{})`$ (with $`N^+`$ and $`N^{}`$, the positive and negative charge multiplicity respectively).
Various possible mecanisms have been suggested so that a hadronic gas could yield both the correct strange particle ratios and pion multiplicity: sequential freeze out or separate chemical and thermal freeze outs, hadronic equation of state with excluded volume corrections, non-zero pion chemical potential, equilibrated plasma undergoing sudden hadronization and immediate decoupling.
All the physical points suggested to salvage the standard hadronic gas model might need to be contemplated in a precise hydrodynamical description of relativistic nuclear collisions, however it is somewhat suprising that one has to go to such kind of details to reconcile the strange particle and pion data. Is it not possible to build a simple hydrodynamical model that yield the correct abundances without extra assumption? We discuss this question in the next section.
Hydrodynamical description with continuous emission - In the standard hydrodynamical models, one assumes that the freeze out occurs on a sharp three-dimensional surface (defined for example by $`T(x,y,z,t)=\mathrm{const}`$). Before crossing it, particles have a hydrodynamical behavior, and after, they free-stream toward the detectors, keeping memory of the conditions (flow, temperature) of where and when they crossed the three dimensional surface. The Cooper-Frye formula gives the invariant momentum distribution in this case
$$Ed^3N/dp^3=_\sigma ๐\sigma _\mu p^\mu f(x,p).$$
(1)
$`d\sigma _\mu `$ is a normal vector to the freeze out surface $`\sigma `$ and $`f`$ the distribution function of the type of particles considered. This is the formula implicitly used in all standard thermal and hydrodynamical model calculations of the previous section.
The notion that particle emission does not necessarily occur on a three dimensional surface but may be continuous was incorporated in a hydrodynamical description in . In this model, the fluid is assumed to have two components, a free part plus an interacting part and its distribution function reads
$$f(x,p)=f_{\mathrm{f}ree}(x,p)+f_{\mathrm{i}nt}(x,p).$$
(2)
$`f_{\mathrm{f}ree}`$ counts all the particles that last scattered earlier at some point and are at time $`x^0`$ in $`\stackrel{}{x}`$. $`f_{\mathrm{i}nt}`$ describes all the particles that are still interacting (i.e., that will suffer collisions at time $`>x^0`$ and change momentum). The invariant momentum distribution is then
$$Ed^3N/dp^3=d^4xD_\mu [p^\mu f_{\mathrm{f}ree}(x,p)].$$
(3)
$`D_\mu [p^\mu f_{\mathrm{f}ree}(x,p)]`$ is a covariant divergence in general coordinates and $`d^4x`$ is the invariant volume element. A priori formula (3) is sensitive to the whole fluid history and not just to freeze out conditions as in formula (1).
To compare particle abundances in the continuous emission and freeze out scenarios, we use a simplified framework to describe the fluid expansion, namely we suppose longitudinal expansion only and longitudinal boost invariance. This approximation allows to carry out some calculations analytically and turns the physics involved more transparent. It is implicit however that this description applies at best to the midrapidity region. We will therefore concentrate on midrapidity abundances, precisely data from NA35.
In this simplified framework, in the case of a fluid with freeze out at a constant temperature and chemical potential, the Cooper-Frye formula (1) can be rewritten ignoring transverse expansion as
$`{\displaystyle \frac{dN}{dyp_{}dp_{}}}|_{y=0}`$ $`={\displaystyle \frac{gR^2}{2\pi }}\tau _{\mathrm{f}o}(T_{\mathrm{f}o},T_0,\tau _0)m_{}`$ (4)
$`\times `$ $`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}()^{n+1}\mathrm{exp}\left({\displaystyle \frac{n\mu _{\mathrm{f}o}}{T_{\mathrm{f}o}}}\right)K_1\left({\displaystyle \frac{nm_{}}{T_{\mathrm{f}o}}}\right).`$ (5)
(The plus sign corresponds to bosons and minus, to fermions.) It depends on the conditions at freeze out: $`T_{\mathrm{f}o}`$ and $`\mu _{\mathrm{f}o}=\mu _{b\mathrm{fo}}B+\mu _{S\mathrm{fo}}S`$, with $`B`$ and $`S`$ the baryon number and strangeness of the hadron species considered, and $`\mu _{S\mathrm{fo}}(\mu _{b\mathrm{fo}},T_{\mathrm{f}o})`$ obtained by imposing strangeness neutrality. So the experimental spectra of particles teach us in that case what the conditions were at freeze out.
For continuous emission, we can approximate equation (3) as
$`{\displaystyle \frac{dN}{dyp_{}dp_{}}}|_{y=0}{\displaystyle \frac{2g}{(2\pi )^2}}`$ (6)
$`\times {\displaystyle _{๐ซ=0.5}}d\varphi d\eta {\displaystyle \frac{m_{}\mathrm{cosh}\eta \tau _F\rho d\rho +p_{}\mathrm{cos}\varphi \rho _F\tau d\tau }{\mathrm{exp}\left[(m_{}\mathrm{cosh}\eta \mu )/T\right]\pm 1}},`$ (7)
where $`๐ซ`$ is the probabality to escape without collision calculated with a Glauber formula, $`\tau _F`$ (resp. $`\rho _F`$) is solution of $`๐ซ(\tau _F,\rho ,\varphi ,\eta ;v_{})=0.5`$ (resp. $`๐ซ(\tau ,\rho _F,\varphi ,\eta ;v_{})=0.5`$). In (7), various $`T`$ and $`\mu =\mu _bB+\mu _SS`$ appear (again $`\mu _S`$ is obtained from strangeness neutrality), reflecting the whole fluid history, not just $`T_{\mathrm{f}o}`$ and $`\mu _{b\mathrm{fo}}`$. This history is known by solving the hydrodynamical equations of a hadronic gas with continuous emisison. It depends only on the initial conditions $`T_0`$ and $`\mu _{b0}`$. Therefore (7) only depends on the initial conditions.
Once the spectra are known, they can be integrated to get abundances. Figure 1 shows the allowed region of initial conditions that lead to the experimental NA35 midrapidity values $`\mathrm{\Lambda }=1.26\pm 0.22_{p_{}>0.5\mathrm{GeV}}`$, $`\overline{\mathrm{\Lambda }}=0.44\pm 0.16_{p_{}>0.5\mathrm{GeV}}`$, $`K_S^0=1.30\pm 0.22_{p_{}>0.62\mathrm{GeV}}`$, $`h^{}=27\pm 1`$ and $`p\overline{p}=3.2\pm 0.4`$. We do not use the $`K^+`$ and $`K^{}`$ abundances because they were measured outside the mid-rapidity region. For heavy particles, initial conditions dominate so when looking at their abundances in finite $`p_{}`$ windows, transverse flow (assumed zero initially) can be neglected as we did. For negatives, the experimental abundance is for all $`p_{}`$ so tranverse flow does not affect their abundance either. The allowed window corresponds to $`T_0185`$ MeV, $`\mu _{b0}100`$ MeV , for an ideal gas equation of state and the strangeness saturation factor $`\gamma _s=1.3`$
Using a more sofisticated equation of state, the value of $`T_0`$ might be decreased by some 10-15%, i.e., to 155-165 MeV, compatible with (i.e., below) QCD lattice values for the phase transition temperature from QGP to hadronic matter. Our value of $`\gamma _S`$ is above 1 and this might looked surprising. However, its value is decreased by some 15% when looking at a more realistic equation of state. In addition, we have imposed strangeness neutrality, it is possible that this is a too strong constraint when analyzing data taken in a very restricted rapidity region (see where a similar problem was encountered).Using a larger value of $`\gamma _s`$, the size of the allowed window for initial conditions in figure 1 increases. There are other factors that influence the precise location and size of the window: values of the cross section needed to compute $`๐ซ`$ (taken constant and equal for simplicity here), value of the cutoff in $`๐ซ=0.5`$, etc. However the important point is that it is possible to find initial conditions of the hydrodynamical expansion such that strange and non-strange particle abundances can be reproduced simultaneously without extra assumption.
To illustrate why the continuous emission is able to reproduce both strangeness and pion data, in table 1, we compare results from the continuous emission scenario and the freeze out model. We took $`T_0=185`$ MeV and $`\mu _{b0}=100`$ MeV for the continuous emission case, $`T_{\mathrm{f}o}=185`$ MeV and $`\mu _{b\mathrm{fo}}=100`$ MeV for the freeze out case. We used $`\gamma _s=1.3`$ for both. We expect roughly similar results for both models for heavy particles: in the continuous emission case, due to thermal suppression, they are mostly emitted early, i.e., in similar conditions than in the freeze out model. Pions in the freeze out case are too few as discussed previously. Pions in the continuous emission case, on the other side, are emitted early and then on, so we expect to have more of them. This is precisely what we see in the table.
The initial conditions that we discussed so far correspond to a hadron gas, starting its hydrodynamical evolution. However the present continuous emission scenario is in fact probably compatible with the possibility that a QGP was created before for the following reason. Continuous emission is possible from a QGP but is inhibited by two factors: 1) only color singlet objects (not single quarks) can escape from it 2) the QGP core is surrounded by a dense hadronic region that the color singlets would have to cross without collisions to modify the previous results. In this case, particle emission would occur mostly in the hadronic phase. Numerical estimates are however necessary to back up these qualitative arguments.
Relation between pion number and entropy - In a hydrodynamical model without shocks and dissipation, entropy is conserved. In the usual freeze out scenario, this is an important point because the initial entropy can be determined from the final multiplicity. To illustrate this connection, let us consider a massless pion gas. Statistical mechanics yields the following relationship for the entropy density and the pion density $`s=3.6n_\pi `$. In the longitudinal boost invariant model, the entropy conservation equation gives: $`s\tau =\mathrm{const}`$. Using $`dV=\tau dy\pi R^2`$ at $`y=0`$, this can be written: $`dS(\tau )/(\pi R^2dy)=\mathrm{const}`$. So we can rewrite
$$\frac{dS}{dy}(\tau )=\frac{dS}{dy}=3.6\frac{dN_\pi }{dy}.$$
(8)
Therefore, experimental knowledge of $`dN_\pi /dy`$, permits to extract $`dS/dy`$, which are both independent of time.
In the continuous emission case, $`s3.6n_\pi `$ because the distribution function needed to compute the pion density now depends on the escape probability $`๐ซ`$, embodied in $`f_{\mathrm{f}ree}`$ (cf. eq. (2)). In fact, using similar methods than in , one can show that
$$s=\left[1+\frac{3\beta \alpha }{4(1+\alpha )}\right]3.6n_\pi ,$$
(9)
where $`\alpha (t,\rho )=\left(4\pi \right)^1๐\varphi ๐\theta \mathrm{sin}\theta \left[๐ซ/(1๐ซ)\right]_{z=0}`$ and $`\beta (t,\rho )=\left(4\pi \right)^1๐\varphi ๐\theta \mathrm{sin}\theta \mathrm{cos}^2\theta \left[๐ซ/(1๐ซ)\right]_{z=0}`$ . This relationship is plotted in figure 2. For all radii $`\rho `$, time $`\tau `$ and inicial condition $`T_0`$, $`s(\tau ,\rho )3.6n_\pi (\tau ,\rho )`$. One then gets in the longitudinal boost invariant model:
$$\frac{dS}{dy}3.6\frac{dN_\pi (\tau )}{dy},$$
(10)
in other words, for a fixed entropy, there are more pions in the continuous emission case than in the freeze out case (for all times $`\tau `$). This is in fact expected (even using a more realistic equation of state and including pions from decays): since pions are emitted continuously, there are more copious than in the the usual freeze out case for a given $`dS/dy`$ (itself fixed by the initial conditions).
As a consequence, a large experimental value of $`dN_\pi /dy`$ should not necessarily be associated to a large $`dS/dy`$ and be considered a hint of QGP formation as usually done in freeze out models. A larger $`dN_\pi /dy`$ and not large $`dS/dy`$ is a natural consequence of continuous emission, compared to freeze out. This result is in contrast for exemple with .
Conclusion - In this paper, we discussed data on strange and non-strange particles by NA35, from an hydrodynamical point of view. The standard model with sudden freeze out can reproduce the strange particle data but underpredicts the pion abundance, if no extra assumption is made. We showed that a hydrodynamical model with a more precise emission process, continuous emission, can reproduce both the strange and non-strange particle data without extra assumption.
This indicates the necessity of doing a more accurate description of particle emission in hydrodynamics, else some problems might artificially appear, such as a too low predicted pion abundance as discussed here or a too high freeze out density .
This point is reinforced by the fact that a large pion number is usually associated with a large entropy and QGP formation. Here we showed that a large pion number can be generated by continuous emission without modifying the entropy. (The larger pion emission will cause a faster cooling and shorter fluid lifetime.) In other words, a better understanding of particle emission in the hydrodynamical regime is also necessary to assess the possibility of QGP formation in relativistic heavy ion collisions.
There is a growing tendency to use hydrodynamical models to describe relativistic nuclear collisions, there is also a growing concern to modelize freeze out better . However it seems that the very notion of particle emission during the hydrodynamical expansion needs to be put under more scrutiny.
This work was partially supported by FAPESP (proc. 98/14990-0, 98/2249-4 and 99/0529-2) and CNPq (proc. 300054/92-0). |
warning/0001/hep-th0001202.html | ar5iv | text | # 1 Introduction
## 1 Introduction
In this paper we bigrade the Hopf algebra of undecorated rooted trees, and both of its closed Hopf subalgebras, taking account of Hochschild cohomology.
In we have exposed the connection between renormalization and Hopf algebra. The joblist of renormalization specifies a noncocommutative coproduct, $`\mathrm{\Delta }`$. On the left are products of divergent subdiagrams; on the right these shrink to points. An antipode, $`S`$, upgrades this bialgebra to a Hopf algebra, by specifying the procedure of subtracting subdivergences. If this antipode is twisted, by taking only the poles of the Laurent series in $`\epsilon `$, for dimensionally regularized diagrams in $`d:=42\epsilon `$ spacetime dimensions, the final subtraction delivers a finite renormalized Green function, in the limit $`d4`$, corresponding to the minimal subtraction scheme. Different twists correspond to different renormalization schemes .
The general problem of perturbative quantum field theory involves the Hopf algebra of decorated rooted trees. These decorations represent primitive divergences, coming from diagrams with no subdivergences. Restriction to the Hopf algebra $`_R`$ of undecorated rooted trees, generated by a single primitive divergence, reveals a remarkable feature. This apparently small problem in quantum field theory has a mathematical structure larger than a very general problem in noncommutative geometry, investigated by Alain Connes and Henri Moscovici , who showed that the composition of diffeomorphisms can be described algebraically, and hence extended to noncommutative manifolds, by making use of an appropriate Hopf algebra $`_{\mathrm{CM}}`$. In it was shown that the Hopf algebra $`_{\mathrm{CM}}`$ of is, in the one-dimensional case, the unique noncocommutative Hopf subalgebra of $`_R`$, corresponding to adding Feynman diagrams with weights determined by natural growth. The only other closed Hopf subalgebra is the cocommutative Hopf algebra $`_{\mathrm{ladder}}`$ of rooted trees whose nodes have fertility less than 2, corresponding to the ladder (or rainbow) diagrams of .
Suppose we are given an $`n`$-loop Feynman diagram that represents an $`n`$-node tree in $`_R`$. It involves $`n`$ ultraviolet-divergent integrations, and hence one may expect that it delivers, in dimensional regularization, a pole of $`n`$-th order. But then, combinations of diagrams corresponding to sums of products of rooted trees can provide cancellations of poles, and may hence eliminate leading pole terms. A prominent example of such a mechanism occurs in the calculation of an anomalous dimension, $`\gamma =d\mathrm{log}Z/d\mathrm{log}\mu `$, which detects only single-pole terms, after minimal subtraction of subdivergences. All higher order poles are determined by the requirement that they cancel when one takes the derivative of the logarithm of the renormalization factor $`Z`$ w.r.t. to the renormalization scale $`\mu `$. Every practitioner of multiloop quantum chromodynamics is vividly aware of the bigrading of her/his work, by loop number and degree of singularity. The slightest error in handling either the combinatorics or the integrations usually โ and mercifully โ reveals itself by a failure to get the uniquely finite answer that is ensured by the locality of counterterms. Thus there is deep โ and largely uncharted โ structure in the relations between Laurent expansions of products of Feynman diagrams, corresponding to forests of rooted trees.
In this work we make preparation for a cohomological approach to renormalization, by identifying and analyzing a combinatoric bigrading of linear combinations of undecorated rooted forests.
In sect. 2, we define this bigrading, in terms of the number of nodes $`n`$ and an index $`k`$ that classifies subspaces according to their projection into an augmentation ideal, analyzed by $`k`$-fold iterations of the coproduct $`\mathrm{\Delta }`$. We wish to learn the dimension, $`H_{n,k}`$, of the subspace with weight $`n`$ and index $`k`$. In sect. 3 we find that this problem has a very simple solution in the cocommutative subalgebra $`_{\mathrm{ladder}}`$, where the dimension is the number of ways of partitioning $`n`$ into $`k`$ positive integers, given in Table 1. In sect. 4 we find that the corresponding problem in the noncocommutative subalgebra $`_{\mathrm{CM}}`$ has the subtler solution of Table 2, which we find to be related to Table 1 by a remarkable transform, which preserves the sums of rows. In sect. 5 we team Neil Sloaneโs superseeker with Tony Hearnโs Reduce and find
$$H(x,y):=\underset{n,k}{}H_{n,k}x^ny^k=\frac{R(x)}{(1y)R(x)+xy}$$
(1)
for the generating function of the bigrading of $`_R`$, with results in Table 3 obtained from
$$R(x):=\underset{n>0}{}r_nx^n=x\underset{n>0}{}(1x^n)^{r_n}=x+x^2+2x^3+4x^4+9x^5+20x^6+\mathrm{}$$
(2)
which generates the number $`r_n`$ of rooted trees with $`n`$ nodes. Our discovery of the generating principle of Table 3 was triggered by superseeker analysis of merely the first 8 entries of its first column. After thorough study of the filtration in Table 4, we prove (1).
## 2 The second grading
The weight $`n`$ of a rooted tree $`t`$ is the number of its nodes. The weight of a forest $`F=_jt_j`$ is the sum of the weights of the trees $`t_j`$ in the product. This is the first grading.
To define the index $`k`$ for the second grading, $`k`$-primitivity, we use $`k`$-fold iterations of the coproduct $`\mathrm{\Delta }`$, defined by the highly nontrivial recursion
$$\mathrm{\Delta }(t)=te+(\mathrm{id}B_+)\mathrm{\Delta }B_{}(t)$$
(3)
for a nonempty tree $`t`$. Here $`e`$ is the empty tree, evaluating to unity, $`\mathrm{id}`$ is the identity map, $`B_{}`$ removes the root of $`t`$, and $`B_+`$ combines the trees of a product by appending them to a common root. The coproduct $`\mathrm{\Delta }`$ is coassociative. Hence it has a unique iteration, which may be written in a variety of equivalent ways. Since $`\mathrm{\Delta }`$ has only single trees on the right, the recursion
$$\mathrm{\Delta }^k=(\mathrm{id}\mathrm{\Delta }^{k1})\mathrm{\Delta }$$
(4)
is particularly convenient. For a forest $`F=_jt_j`$ we have $`\mathrm{\Delta }^k(F)=_j\mathrm{\Delta }^k(t_j)`$.
Let $`X`$ be a $`Q`$-linear combination of monomials of trees, i.e. of forests. We say that $`X`$ is $`k`$-primitive if every term of $`\mathrm{\Delta }^k(X)`$ has at least one empty tree $`e`$. Symbolically we may consider the composition of tensor products of the projection operator $`P:=\mathrm{id}E\overline{e}`$ with iterations of $`\mathrm{\Delta }`$. $`P`$ projects onto the augmentation ideal $`_c=\{X_RP(X)=X\}`$, where $`X=P(X)+E\overline{e}(X)`$. Here $`\overline{e}`$ is the counit, which annihilates everything except the empty tree, for which it gives $`\overline{e}(e)=1`$. The map from the rationals back to the algebra is simply $`E(q)=qe`$ for $`qQ`$. Hence $`P`$ annihilates $`e`$ and leaves everything else unchanged. Let $`U_0:=P`$ and
$$U_k:=(\underset{k+1\mathrm{times}}{\underset{}{P\mathrm{}P}})\mathrm{\Delta }^k=(PU_{k1})\mathrm{\Delta }$$
(5)
for $`k>0`$. In using the recursive form, note should be taken that, in general, the projection makes $`U_k(X_1X_2)U_k(X_1)U_k(X_2)`$: one should store results for forests; not just for trees.
We have said that $`X`$ is $`k`$-primitive if $`U_k(X)=0`$. Then clearly $`X`$ is $`(k+1)`$-primitive, since $`\mathrm{\Delta }^{k+1}(X)`$ has at least two empty trees $`e`$ in every term. We are interested in the number, $`H_{n,k}:=D_{n,k}D_{n,k1}`$, of weight-$`n`$ terms that are $`k`$-primitive but are not $`(k1)`$-primitive, where $`D_{n,k}`$ is the dimension of the subspace with weight $`n`$ and index $`k`$. To compute $`D_{n,k}`$ for specific (and rather modest) values of $`n,k`$ one considers the most general linear combination $`X`$ of weight-$`n`$ terms, with unknown coefficients, and solves $`U_k(X)=0`$. The rank deficiency of this large system of linear equations is $`D_{n,k}`$. From this one subtracts the number $`D_{n,k1}`$ of weight-$`n`$ terms that are $`(k1)`$-primitive. By this means we obtained the first 7 rows of Tables 1 and 2, for the Hopf subalgebras $`_{\mathrm{ladder}}`$ and $`_{\mathrm{CM}}`$, and inferred their generating principles. In the case of the full Hopf algebra $`_R`$, bigraded in Table 3, data were much harder to obtain. Fortunately the generating principle is very distinctive.
## 3 Bigrading the cocommutative subalgebra
We first consider the cocommutative Hopf algebra $`_{\mathrm{ladder}}`$ of rooted trees all of whose nodes have fertility less than 2, i.e. the Hopf algebra with linear basis $`l_n=B_+^n(e)`$, $`n0`$. In this very simple case, the recursive definition (3) linearizes on the left, giving
$$\mathrm{\Delta }(l_n)=\underset{k=0}{\overset{n}{}}l_{nk}l_k$$
(6)
for the unique $`n`$-node tree $`l_n_{\mathrm{ladder}}`$. Thanks to our recent work in we have an explicit construction of the weight-$`n`$ 1-primitive $`p_n_{\mathrm{ladder}}`$. First we compute the antipodes. In the cocommutative case, these are simply
$$S(l_n)=\underset{k=0}{\overset{n1}{}}S(l_{nk})l_k$$
(7)
with $`l_0=e`$ and $`S(e)=e`$. To construct the 1-primitives, we use the star product $`SY`$, where $`Y`$ is the grading operator, giving $`Y(l_k)=kl_k`$. In general, a star product of operators is defined by $`O_1O_2:=m(O_1O_2)\mathrm{\Delta }`$, where $`m`$ merely multiplies entries on the left and right of a tensor product. The ladder 1-primitives are given by
$$p_n:=\frac{1}{n}[SY](l_n)=\underset{k=0}{\overset{n}{}}\frac{k}{n}S(l_{nk})l_k.$$
(8)
Clearly $`p_1=l_1`$ and $`p_2=l_2\frac{1}{2}l_1^2`$ are 1-primitive. It takes some time to show that
$`p_8`$ $`=`$ $`l_8l_7l_1l_6l_2+l_6l_1^2l_5l_3+2l_5l_2l_1l_5l_1^3\frac{1}{2}l_4^2+2l_4l_3l_1+l_4l_2^23l_4l_2l_1^2+l_4l_1^4`$ (9)
$`+l_3^2l_2\frac{3}{2}l_3^2l_1^23l_3l_2^2l_1+4l_3l_2l_1^3l_3l_1^5\frac{1}{4}l_2^4+2l_2^3l_1^2\frac{5}{2}l_2^2l_1^4+l_2l_1^6\frac{1}{8}l_1^8`$
gives $`\mathrm{\Delta }(p_8)=p_8e+ep_8`$. We were able to compute this primitive with ease, using recursion (7) in the star product (8). From we know that $`[SY](t)`$ delivers a combination of diagrams whose singularity is a single pole, as $`d4`$, with a residue that determines the contribution of $`t`$ to the anomalous dimension. Moreover 1-primitives have only single poles. However the converse is not true in the full Hopf algebra: noncocommutativity implies that not every $`[SY](t)`$ is 1-primitive. Here, in the cocommutative subalgebra, there is a single 1-primitive for each weight $`n>0`$. Hence $`SY`$ delivers it.
From examples such as (9) we inferred the general result of (8). The 1-primitive $`p_n`$ contains all possible multiplicative partitions $`_jl_j^{n_j}`$ with weight $`n=_jn_jj`$. The coefficient of each partition is $`(1)^{k1}(k1)!/_jn_j!`$ where $`k=_jn_j`$ is the number of integers into which $`n`$ has been partitioned. For example the partition $`8=2+2+1+1+1+1`$, with $`k=6`$, gives the coefficient $`5!/2!4!=5/2`$ of $`l_2^2l_1^4`$ in (9). We have tested this Ansatz up to $`n=20`$, where $`p_{20}`$ contains 627 terms.
It is easy to understand the leading diagonal of Table 1: $`l_1^n`$ is $`n`$-primitive, but not $`(n1)`$-primitive. For $`8>n>k>1`$ we used Reduce to prove the results of Table 1. The entry in the $`n`$-th row and $`k`$-th column is $`\overline{H}_{n,k}=\overline{D}_{n,k}\overline{D}_{n,k1}`$, where $`\overline{D}_{n,k}`$ is the number of undetermined coefficients when one solves $`U_k(X)=0`$, with $`X`$ taken as an unknown linear combination of forests $`_jl_j^{n_j}`$ of weight $`n=_jn_jj`$. Clearly the generating principle is extremely simple: $`\overline{H}_{n,k}`$ is the number of partitions of $`n`$ into $`k`$ positive integers. This simply reflects the fact that solving $`U_k(X)=0`$ determines all and only the coefficients of partitions with $`_jn_jk`$. Hence the $`k`$-th column of Table 1 is generated by
$$\overline{H}_k(x):=\underset{n}{}\overline{H}_{n,k}x^n=\underset{jk}{}\frac{x}{1x^j}=\frac{x}{1x^k}\overline{H}_{k1}(x)$$
(10)
which yields the recursion of the tabular entry A048789 of :
$$\overline{H}_{n,k}=\overline{H}_{nk,k}+\overline{H}_{n1,k1}$$
(11)
seeded by the empty tree, which gives $`\overline{H}_{0,0}=1`$. We particularly note that for all $`j,k>0`$
$$\overline{H}_{j+k}(x)<\overline{H}_j(x)\overline{H}_k(x).$$
(12)
## 4 Bigrading the Connes-Moscovici subalgebra
To compute Table 2 we proceeded as above, now using the coproduct
$`\mathrm{\Delta }(\delta _n)`$ $`=`$ $`\delta _ne+e\delta _n+R_{n1}`$ (13)
$`R_n`$ $`=`$ $`[๐e+e๐+\delta _1๐,R_{n1}]+\delta _1Y(\delta _n)`$ (14)
with $`R_0=0`$, $`[๐,\delta _n]=\delta _{n+1}`$ increasing weight, and $`[๐,\delta _n]=Y(\delta _n)=n\delta _n`$ measuring weight. This is the noncocommutative coproduct of Connes and Moscovici , shown in to give the closed Hopf subalgebra of $`_R`$ that is realized by $`\delta _n=N^{n1}(l_1)`$, where $`N`$ is the natural growth operator, which appends a single node in all possible ways. Thus $`\delta _1=l_1`$ and $`\delta _2=l_2`$, while $`\delta _3=l_3+B_+(l_1^2)`$ differs from the ladder-algebra element $`l_3`$. Natural growth implies that $`\delta _n`$ is a sum over all weight-$`n`$ trees in $`_R`$, with nonzero Connes-Moscovici weights that we specified in , using an efficient recursive procedure.
Computation of the first 7 rows of Table 2 took longer than for Table 1, because of the proliferation of product terms on the left of the noncocommutative coproduct. These scanty data presented us with a pretty puzzle, which the reader might like to try to solve, after covering up the rows of Table 2 with $`n>7`$. What is the generating procedure? Recall that the sum of the $`n`$-th row in Table 2 must agree with that in Table 1, since each gives the total number of ways of partitioning the integer $`n`$. In Table 1 this is achieved with great simplicity: the $`k`$-th entry is the number of ways of partitioning $`n`$ into $`k`$ positive integers. In Table 2 it is achieved far more subtly, by the addition of only $`1+n/2`$ terms, since $`\stackrel{~}{H}_{n,k}`$ has support only for $`2knk`$.
Given merely data for $`n7`$, the most interesting feature is the second subleading diagonal $`1,2,4,6\mathrm{}`$ The leading diagonal is generated by $`G_0=1/(1z)`$, the first subleading diagonal by $`G_1=1/(1z)^2`$. The simplest Ansatz for the second is $`G_2=G_1/(1z^2)`$, which requires $`\stackrel{~}{H}_{8,6}=9`$. Then $`\stackrel{~}{H}_{8,5}=4`$ is required, so that $`1+\stackrel{~}{H}_{8,5}+9+7+1=22`$ is the number of ways of partitioning 8. A Reduce program, running for 24 hours, proved that indeed $`\stackrel{~}{H}_{8,5}=4`$. Next, the requirement $`\stackrel{~}{H}_{9,6}=7`$ comes from $`2+\stackrel{~}{H}_{9,6}+12+8+1=30`$, for the partitions of 9, taking $`\stackrel{~}{H}_{9,7}=12`$ from the hypothesis $`G_2=1/(1z)^2(1z^2)`$ for the second subleading diagonal. Then the third subleading diagonal is revealed as $`1,2,4,7\mathrm{}`$ which is nicely consistent with $`G_3=G_2/(1z^3)`$. Finally, it is easy to check that the recurrence relation $`G_k=G_{k1}/(1z^k)`$ for the diagonals makes the rows sum to the correct partitions. Later we shall prove this result by considering the Connes-Moscovici restriction of the filtration of the bigrading of the full Hopf algebra.
In words, the transformation is simple to state: the subleading diagonals of Table 2 are the partial sums of the columns of Table 1. This leads to the subtle recurrence relation
$$\stackrel{~}{H}_{n,k}=\stackrel{~}{H}_{k,2kn}+\stackrel{~}{H}_{n2,k1}$$
(15)
for the bigrading of the Connes-Moscovici Hopf subalgebra. We particularly note that for $`j,k>0`$ and $`j+k>2`$
$$\stackrel{~}{H}_{j+k}(x)<\stackrel{~}{H}_j(x)\stackrel{~}{H}_k(x)$$
(16)
while for $`j=k=1`$ we have the equality $`\stackrel{~}{H}_2(x)=\stackrel{~}{H}_1(x)\stackrel{~}{H}_1(x)=x^2(1+x)^2`$.
## 5 Bigrading the full Hopf algebra of rooted trees
Given how long it took to compute the data that eventually led to the generating principle for the Connes-Moscovici subalgebra, one might be daunted by the task of inferring the bigrading of the full Hopf algebra of undecorated rooted trees. In fact, we discovered this first, by mere consideration of the first 8 entries in the first column of Table 3. Thanks to we had an extremely efficient Reduce implementation of the coproduct (3). The severity of the challenge of understanding the range and kernel of $`U_k`$, i.e. the difficulty of the computation of $`\mathrm{\Delta }^k`$, increases drastically with $`k`$. At $`k=1`$ it was possible to solve $`U_1(X):=(PP)(\mathrm{\Delta }(X))=0`$, for weights $`n8`$, using a few hours of CPUtime, notwithstanding the fact that at $`n=8`$ the number of products of trees is $`r_9=286`$.
The book-keeping was very simple, since the defining property of rooted trees is that every weight-$`n`$ forest $`F=_jt_j`$ is uniquely labelled by the tree $`B_+(F)`$ with weight $`n+1`$. This clearly leads to the enumeration (2). More deeply, it shows that $`B_+`$ is Hochschild closed, and hence that the apparently simplistic quantum-field-theory task of handling a single primitive divergence solves a universal problem in Hochschild cohomology.
Submitting $`1,1,1,2,3,8,16,41`$ to Neil Sloaneโs superseeker , we learnt that it is generated by the first 8 terms of
$$H_1(x)=\frac{R(x)x}{R(x)}=1\underset{n>0}{}(1x^n)^{r_n}.$$
(17)
At the time, Sloane had no idea that we were studying the bigrading of rooted trees and told us โit is pretty unlikely this is your sequence, but I thought I should pass this along just in caseโ. In fact, his superseeker discovery unlocked our puzzle. We knew that
$$\frac{R(x)}{x}=\underset{k0}{}H_k(x)$$
(18)
where $`H_0(x):=1`$ and $`H_k(x):=_kH_{n,k}x^n`$ generates column $`k`$ of Table 3. We then construed (17) as
$$\frac{R(x)}{x}=\frac{1}{1H_1(x)}=\underset{k0}{}[H_1(x)]^k.$$
(19)
Comparison with (18) then led to the conjecture $`H_k(x)=[H_1(x)]^k`$, requiring that
$$H_{j+k}(x)=H_j(x)H_k(x).$$
(20)
To test this, we made intensive use of Reduce. At weight $`n=9`$ we computed the $`3214\times 719`$ matrix of integer contributions to the 3214 terms in $`\mathrm{\Delta }(X)`$ produced by $`r_{10}=719`$ weight-9 forests. The rank deficiency of the condition $`U_1(X)=0`$ was proven to be $`D_{9,1}=98`$, which is indeed the coefficient of $`x^9`$ in (17). We tested $`H_2(x)=[H_1(x)]^2`$ up to weight $`n=8`$, where $`\mathrm{\Delta }^2(X)`$ has 3651 terms in 286 unknowns. Here $`U_2(X)=0`$ gave $`D_{8,2}=41+58=99`$, where 41 and 58 are indeed the coefficients of $`x^8`$ in (17) and its square. Finally, we tested $`H_3(x)=H_1(x)H_2(x)`$ up to weight $`n=7`$, where $`\mathrm{\Delta }^3(X)`$ has 3168 terms in 115 unkowns, with $`U_3(X)=0`$ giving $`D_{7,3}=16+26+27=69`$, in agreement with the sum of the coefficients of $`x^7`$ in (17), its square and cube.
Hence we obtained compelling evidence for the bigrading (1) of the Hopf algebra of rooted trees, determined by the circumstance (20) that it saturates all inequalities. First we derive these general inequalities, for any commutative graded Hopf algebra. Then we prove that they are saturated in $`_R`$.
### 5.1 General inequalities
Let $``$ be a commutative graded Hopf algebra with unit $`e`$. Let $`\mathrm{deg}`$ be the grading, with $`\mathrm{deg}(X)N`$ for all $`X`$ and $`\mathrm{deg}(e)=0`$. We assume that $``$ is reduced to scalars by the counit $`\overline{e}`$.
Let $`_k`$ be the set of elements in the kernel of $`U_k`$ which are in the range of $`U_{k1}`$, so that $`U_k(X)=0`$ and $`U_{k1}(X)0`$, for $`X_k`$. Then we call $`k`$ the degree of primitivity of $`X`$, writing $`\mathrm{deg}_p(X)=k`$. We let $`_0`$ be the set of elements in the kernel of $`P=U_0`$, i.e. the scalars. The augmentation ideal fulfills $`_c=/_0=_{k=1}^{\mathrm{}}_k`$.
To show that $`\mathrm{deg}_p(X)\mathrm{deg}(X)`$, suppose that
$$\mathrm{\Delta }^{\mathrm{deg}(X)1}(X)=\underset{i}{}X_i^{(1)}\mathrm{}X_i^{(\mathrm{deg}(\mathrm{X}))}$$
(21)
has nonscalar entries in $`_c^{\mathrm{deg}(X)}`$. Then they are all formed from 1-primitives, in $`_1`$, since the coproduct is homogenous in $`\mathrm{deg}`$ and any element $`X`$ with $`\mathrm{deg}(X)=1`$ also has $`\mathrm{deg}_p(X)=1`$. Hence $`\mathrm{deg}_p`$ is majorized by $`\mathrm{deg}`$ and is thus finite for each $`X`$.
We denote by $`H_{n,k}`$ the number of linearly inequivalent terms $`X`$ with weight $`\mathrm{deg}(X)=n`$ and primitivity $`\mathrm{deg}_p(X)=k`$. Then the generators $`H_k(x):=_{nk}H_{n,k}x^n`$ satisfy
$$H_{j+k}(x)H_j(x)H_k(x),j,k>0.$$
(22)
Proof: It is sufficient to show that an element $`X_{j+k}`$ may be labelled by those terms in $`\mathrm{\Delta }(X)`$ that are in $`_j_k`$. To prove this, suppose that $`X_1`$ and $`X_2`$ give the same terms in $`_j_k`$. Now observe that $`U_{j,k}:=U_{j1}U_{k1}`$ projects onto $`_j_k`$, giving $`U_{j,k}\mathrm{\Delta }(X_1X_2)=0`$. Finally, observe that coassociativity gives
$$0=U_{j,k}\mathrm{\Delta }(X_1X_2)=P^{(j+k)}\mathrm{\Delta }^{j+k1}(X_1X_2):=U_{k+j1}(X_1X_2)$$
(23)
which shows that $`X_1X_2`$ is $`(j+k1)`$-primitive and hence that $`X_1`$ and $`X_2`$ are equivalent elements of $`_{j+k}`$$`\mathrm{}`$
In consequence of (22) we obtain
$$H_k(x)[H_1(x)]^k.$$
(24)
This reflects the fact that the terms in $`\mathrm{\Delta }^{k1}(X)`$ which belong to $`_1^k`$ are sufficient to label elements $`X_k`$. The remarkable feature of the Hopf algebra of rooted trees, to be proved below, is that all the elements of $`_1^k`$ are necessary to label elements of $`_k`$.
As a further comment, we note that (24) may be strengthened if the Hopf algebra is cocommutative, since then the order of labels is immaterial. In the case of $`_{\mathrm{ladder}}`$, with $`\overline{H}_1(x)=x/(1x)`$, one thus obtains
$$\overline{H}_k(x)\underset{jk}{}\frac{x}{1x^j}$$
(25)
which is in fact saturated by Table 1.
### 5.2 Saturation
We now seek to prove that $`H_k(x)=[H_1(x)]^k`$ in the case that $`=_R`$ is the Hopf algebra of undecorated rooted trees.
First we prove that $`\mathrm{deg}_p(X_jX_k)=\mathrm{deg}_p(X_j)+\mathrm{deg}_p(X_k)`$.
Proof: Suppose that $`X_j_j`$ and $`X_k_k`$. Then
$$U_{j+k1}(X_jX_k)=P^{(j+k)}(\mathrm{\Delta }^{j+k1}(X_j)\mathrm{\Delta }^{j+k1}(X_k))0$$
(26)
contains 1-primitives in all its slots, giving $`U_{j+k}(X_jX_k)=0`$, by coassociativity. $`\mathrm{}`$
It is instructive to see how this works out for the product $`ZX`$, when $`Z`$ is 1-primitive and $`X`$ is $`k`$-primitive. Then
$$\mathrm{\Delta }^k(Z)=\underset{j=1}{\overset{k+1}{}}e\mathrm{}Z_{_{j\text{th place}}}\mathrm{}e$$
(27)
consists of all $`k+1`$ terms with a single $`Z`$ and $`k`$ empty trees. As $`X`$ is $`k`$-primitive,
$$\mathrm{\Delta }^k(X)=\underset{j=1}{\overset{k+1}{}}\underset{i_j}{}X_{i_j}^{(1)}\mathrm{}e_{_{j\text{th place}}}\mathrm{}X_{i_j}^{(k+1)}+\mathrm{}$$
(28)
with the final ellipsis denoting omission of terms that contain more than one $`e`$. The latter make no contribution to
$$U_k(ZX)=\underset{j=1}{\overset{k+1}{}}\underset{i_j}{}X_{i_j}^{(1)}\mathrm{}X_{i_j}^{(j1)}Z_{_{j\text{th place}}}X_{i_j}^{(j+1)}\mathrm{}X_{i_j}^{(k+1)}$$
(29)
where $`Z`$ replaces a single $`e`$. By construction (29) has all its entries, namely $`Z`$ or $`X_{i_j}^{(r)}`$, in $`_1`$. Hence $`U_{k+1}(ZX)=0`$ and $`\mathrm{deg}_p(ZX)=k+1`$.
Iterating this result one immediately concludes that $`\mathrm{deg}_p(X_1\mathrm{}X_k)=k`$, for 1-primitive elements $`X_1,\mathrm{},X_k`$. This does not, of itself, allow us to conclude that $`H_k(x)=[H_1(x)]^k`$, since the products are commutative. Thus there are fewer $`k`$-fold products of 1-primitives than there are $`k`$-primitives.
To appreciate what is needed in the next step, we pause to consider $`_{\mathrm{CM}}`$. Its 1-primitives are $`\delta _1`$ and $`\stackrel{~}{\delta }_2=\delta _2\frac{1}{2}\delta _1^2`$. From these we can form the 2-primitive products $`\delta _1^2`$, $`\delta _1\stackrel{~}{\delta }_2`$, and $`\stackrel{~}{\delta }_2^2`$. Table 2 shows that there is a further inequivalent 2-primitive, at weight $`n=3`$. Direct computation shows that it may be taken as $`\stackrel{~}{\delta }_3=\delta _3\frac{1}{2}\delta _1^3`$. Then we may form 6 inequivalent 3-primitive products, namely $`\delta _1^3`$, $`\delta _1^2\stackrel{~}{\delta }_2`$, $`\delta _1\stackrel{~}{\delta }_3`$, $`\delta _1\stackrel{~}{\delta }_2^2`$, $`\stackrel{~}{\delta }_2\stackrel{~}{\delta }_3`$ and $`\stackrel{~}{\delta }_2^3`$. Table 2 shows that there is only one more 3-primitive, at weight $`n=4`$. It may be taken as $`\stackrel{~}{\delta }_4=\delta _4\frac{3}{4}\delta _1^4`$. The absence of a further inequivalent 3-primitive at weight $`n=5`$ means that $`\stackrel{~}{H}_3(x)<\stackrel{~}{H}_1(x)\stackrel{~}{H}_2(x)`$. This exercise reveals the filtration of the bigrading of Table 2: the generator is
$$\underset{n,k}{}\stackrel{~}{H}_{n,k}x^ny^k=\frac{1}{1xy}\underset{k>0}{}\frac{1}{1x^{k+1}y^k}$$
(30)
corresponding to products of $`l_1`$ and
$$\stackrel{~}{\delta }_{k+1}=N^k(l_1)\frac{k!}{2^k}l_1^{k+1}$$
(31)
with $`k`$-primitivity achieved by a subtraction at weight $`n=k+1>1`$. At $`y=1`$, the filtration (30) agrees with the ladder filtration
$$\underset{n,k}{}\overline{H}_{n,k}x^ny^k=\underset{k>0}{}\frac{1}{1x^ky}$$
(32)
generated by products of the 1-primitives (8).
Now consider the highly nontrivial filtration of the bigrading of rooted trees. Let $`P_{n,k}`$ be the number of weight-$`n`$ elements of $`_k`$ that cannot be expressed as products of elements of $`\{_jj<k\}`$. Then
$$H(x,y):=\underset{n,k}{}H_{n,k}x^ny^k=\underset{n,k}{}\frac{1}{(1x^ny^k)^{P_{n,k}}}$$
(33)
with $`P_{n,k}`$ telling us how many linearly independent combinations of weight-$`n`$ trees may be made $`k`$-primitive, but not $`(k1)`$-primitive, by suitable subtractions of products of trees of lesser weight. Setting $`y=1`$, taking logs, and using the unique property (2) of the enumeration of rooted trees, we obtain
$$\underset{n}{}r_n\mathrm{log}(1x^n)=\mathrm{log}x\mathrm{log}R(x)=\underset{n,k}{}P_{n,k}\mathrm{log}(1x^n)$$
(34)
and hence $`r_n=_kP_{n,k}`$.
Table 4 gives the filtration implied by $`H_k(x)=[H_1(x)]^k`$. The column generators are
$$P_k(x):=\underset{n}{}P_{n,k}x^n=\underset{j|k}{}\frac{\mu (j)}{k}\left(1\underset{n}{}(1x^{nj})^{r_n}\right)^{k/j}$$
(35)
where the Mรถbius function $`\mu (j)`$ vanishes if $`j`$ is divisible by a square and is equal to $`(1)^p`$ when $`j`$ is the product of $`p`$ distinct primes.
To proceed, we use the Hochschild property of $`B_+`$, namely
$$\mathrm{\Delta }B_+=B_+e+(\mathrm{id}B_+)\mathrm{\Delta }$$
(36)
which follows from the action of the coproduct (3) on the trees produced by $`B_+`$, using $`B_{}B_+=\mathrm{id}`$. Taking care to note that
$$C:=B_+B_{}B_{}B_+=\mathrm{id}$$
(37)
we obtain
$$\mathrm{\Delta }B_{}=(\mathrm{id}B_{})\mathrm{\Delta }C$$
(38)
by composition of (36) with $`\mathrm{id}B_{}`$ on the left and $`B_{}`$ on the right. It follows from (36) that if $`X`$ is $`k`$-primitive, then $`B_+(X)`$ has primitivity no greater than $`k+1`$.
Proof: Suppose that $`X_k`$. Then repeated application of (36) gives
$$U_{k+1}B_+(X)=(P^{(k+1)}B_+)\mathrm{\Delta }^{k+1}(X)=0$$
(39)
since every term in $`\mathrm{\Delta }^{k+1}(X)`$ contains at least two $`e`$โs, of which at most one is promoted to $`l_1`$ by $`B_+`$$`\mathrm{}`$
The presence of $`C`$ in (38) frustrates a parallel attempt to show that $`B_{}`$ decreases primitivity. Rather, we found that the kernel of $`B_{}`$ is an object of great interest. The action of $`B_{}`$ on a nonempty tree $`t`$ is simple: it removes the root to produce, in general, a forest of rooted trees, each of whose roots was originally connected to the root of $`t`$ by a single edge. Since $`B_{}`$ obeys the Leibniz rule
$$B_{}(X_1X_2)=X_1B_{}(X_2)+X_2B_{}(X_1),B_{}(e)=0,$$
(40)
its action on forests is less trivial. The action of $`B_{}`$ on a tree, $`t`$, is undone by $`B_+`$, giving $`C(t):=B_+(B_{}(t))=t`$. On a forest of more than one tree, $`C`$ does not degenerate to the identity map. It is this that makes the Hopf algebra of rooted trees such an amazingly rich structure. Another important feature is that the kernels of $`B_{}`$ and $`C`$ coincide, since $`C:=B_+B_{}`$ and $`B_{}=B_{}C`$. Moreover, $`C`$ is idempotent, since
$$(C\mathrm{id})C=B_+(B_{}B_+\mathrm{id})B_{}=0.$$
(41)
Hence there are two special types of object: trees, for which $`C`$ acts like the identity, and those linear combinations of forests that lie in the kernel of $`C`$. We shall show that the latter are the key to the filtration $`P_{n,k}`$ of Table 4. The first step is to prove that $`C(X)=0`$ for every $`X_1`$ with weight $`n>1`$.
Proof: The coproduct of tree $`t`$ has the form
$$\mathrm{\Delta }(t)=te+B_{}(t)l_1+\mathrm{}$$
(42)
where the ellipsis denotes terms with weight $`n>1`$ on the right. Now consider a forest $`F=_jt_j`$. The Leibniz rule (40) gives
$$\mathrm{\Delta }(F)=\underset{j}{}\mathrm{\Delta }(t_j)=Fe+B_{}(F)l_1+\mathrm{}$$
(43)
and hence $`\mathrm{\Delta }(X)`$ contains $`B_{}(X)l_1`$, for all $`X_R`$. Now suppose that $`X_1`$ has no weight-1 term. Then $`\mathrm{\Delta }(X)=Xe+eX`$ requires that $`B_{}(X)=0`$ and hence that $`X`$ is in the kernel of $`C`$$`\mathrm{}`$
To get acquainted with the problem in hand, consider a pair of 1-primitives, $`X_1`$ and $`X_2`$. Their product is 2-primitive, giving
$$U_1(X_1X_2)=X_1X_2+X_2X_1.$$
(44)
For every such pair, we require another 2-primitive construct, say $`W(X_1,X_2)`$, giving
$$U_1W(X_1,X_2)=X_1X_2X_2X_1.$$
(45)
This does not uniquely define $`W(X_1,X_2)`$, since we may add to any solution of (45) any combination of 1-primitives. The operative question is whether a solution exists, for each pair of distinct 1-primitives. This question does not arise in the ladder subalgebra, which is cocommutative. It is easily answered in the Connes-Moscovici subalgebra, where the asymmetry of
$$U_1(\stackrel{~}{\delta }_3)=3\delta _1\stackrel{~}{\delta }_2+\stackrel{~}{\delta }_2\delta _1$$
(46)
makes it simple to solve the single case of (45) by
$$W(\delta _1,\stackrel{~}{\delta }_2)=\stackrel{~}{\delta }_32\stackrel{~}{\delta }_2\delta _1=B_+(l_1^2)+l_32l_2l_1+\frac{1}{2}l_1^3.$$
(47)
More generally, the $`k`$-primitive nonproduct term $`\stackrel{~}{\delta }_{k+1}`$ accounts for the leading diagonal $`P_{k+1,k}=1`$ of Table 4.
In the full Hopf algebra, we must show the existence of $`P_{n,2}`$ asymmetric pairings enumerated by
$$P_2(x):=\underset{n}{}P_{n,2}x^n=\frac{1}{2}[H_1(x)]^2\frac{1}{2}H_1(x^2)=x^3+x^4+3x^5+5x^6+13x^7+28x^8+\mathrm{}$$
(48)
Part of what is required is clearly provided by
$$W(l_1,X)=l_1X2B_+(X)$$
(49)
since (36) shows that
$$U_1W(l_1,X)=l_1X+Xl_12(PB_+)(Xe+eX)=l_1XXl_1$$
(50)
has the desired antisymmetry. By this means we easily construct the elements of $`_2`$ with weight $`n<5`$ from products of 1-primitives and the action of $`B_+`$ on 1-primitives.
At weights $`n5`$ we need a further construction. There are $`P_{5,2}=3`$ weight-5 nonproduct 2-primitives, but only $`H_{4,1}=2`$ weight-4 1-primitives on which to act with $`B_+`$. We lack, thus far, a way of constructing $`W(p_2,p_3)`$, where $`p_2=l_2\frac{1}{2}l_1^2`$ and $`p_3=l_3l_2l_1+\frac{1}{3}l_1^3`$ are the 1-primitives at weights $`n=2,3`$, common to the cocommutative subalgebra $`_{\mathrm{ladder}}`$. At weight $`n=6`$, we lack $`W(p_2,p_4)`$ and $`W(p_2,p_4^{})`$, where
$`p_4`$ $`=`$ $`l_4l_3l_1\frac{1}{2}l_2^2+l_2l_1^2\frac{1}{4}l_1^4,`$ (51)
$`p_4^{}`$ $`=`$ $`B_+\left(2l_2l_1B_+(l_1^2)l_1^3\right)+l_1B_+(l_1^2)l_2^2,`$ (52)
are the weight-4 ladder and nonladder 1-primitives enumerated by $`H_{4,1}=2`$. It is simple to check that they are annihilated by the Leibniz action of $`B_{}`$, using $`B_{}B_+=\mathrm{id}`$ and $`B_{}(l_n)=l_{n1}`$ with $`l_0:=e`$ evaluating to unity.
At this juncture, it is instructive to compare Tables 3 and 4, which reveal that
$`P_{n,2}`$ $``$ $`2H_{n1,1}`$ (53)
$`P_{n,k}`$ $``$ $`H_{n1,k1},k>2.`$ (54)
In the Appendix, we show that these inequalities persist at large $`n`$, thanks to the fact that the Otter constant $`c:=lim_n\mathrm{}r_{n+1}/r_n=2.955765\mathrm{}`$ is slightly less than 3. Thus it is conceivable that for $`k>2`$ the action of $`B_+`$ might generate $`P_k(x)`$ from $`xH_{k1}(x)`$, but it is quite impossible for it to do this job at $`k=2`$. It appears from (53) that we need a second operator that increases $`n`$ and $`k`$ by unity.
### 5.3 Natural growth by a single node
There is a clear candidate for the second operator: the natural growth operator $`N`$, which appends a single node in all possible ways, and hence obeys a Leibniz rule.
The commutators of $`N`$ with $`B_\pm `$ are easily found, since we need only consider what is happening at the root. Defining the operator $`L`$ by $`L(X):=l_1X`$, we obtain
$`[N,B_+]`$ $`=`$ $`B_+L,`$ (55)
$`[B_{},N]`$ $`=`$ $`LB_{},`$ (56)
$`[N,C]`$ $`=`$ $`[N,B_+]B_{}B_+[B_{},N]=0.`$ (57)
The natural growth operator is a wonderful thing: it commutes with $`B_+B_{}`$, the operator that makes the Hopf algebra so structured; hence it preserves the kernel of $`B_{}`$; like $`B_{}`$, it acts as a derivative; like $`B_+`$, it adds a node and increases the degree of primitivity; finally, it identifies the unique noncocommutative Hopf subalgebra $`_{\mathrm{CM}}`$, with linear basis $`\delta _n:=N^{n1}(l_1)`$.
Constructing $`N(p_4)`$ and $`N(p_4^{})`$, we verified that they are in the kernel of $`U_2`$ and the range of $`U_1`$. It might thus appear that some linear combination of them with $`B_+(p_4)`$, $`B_+(p_4^{})`$ and the product terms $`\{p_1p_4,p_1p_4^{},p_2p_3\}`$ solves the problem of constructing $`W(p_2,p_3)`$. Remarkably, this turns out not to be the case. Rather, we find that application of
$$S_1:=N+(B_+L)Y$$
(58)
to a 1-primitive gives a 1-primitive of higher weight. Here $`Y`$ is the grading operator, which multiplies each tree by its weight and operates on products by a Leibniz rule. Thus $`N(p_4)`$ and $`N(p_4^{})`$ are linear combinations of $`\{B_+(p_4),B_+(p_4^{}),p_1p_4,p_1p_4^{}\}`$ and 1-primitives. Instead of constructing the missing weight-5 nonproduct 2-primitive, we discovered how to generate all the 1-primitives with weight $`n5`$. We have $`L(e)=p_1=l_1`$, at $`n=1`$; $`S_1(p_1)=2p_2`$, at $`n=2`$; $`S_1(p_2)=3p_3`$, at $`n=3`$. At $`n=4`$, we obtain $`S_1(p_3)=4p_4p_4^{}`$, to which we adjoin $`p_4`$, from the ladder construction (8) of sect. 3. Then we obtain the 1-primitives at $`n=5`$ as $`p_5`$, $`S_1(p_4)`$ and $`S_1^2(p_3)`$.
We then found a generalization of (58), which solves the problem of constructing $`W(p_2,p_3)`$. Operating on a weight-$`n`$ 1-primitive with
$$S_k:=\left(S_1\frac{k1}{2}(B_+L)\right)N^{k1}$$
(59)
we create a $`k`$-primitive of weight $`n+k`$. In particular,
$$W(p_2,p_3)=\frac{8S_2(p_3)7NS_1(p_3)}{12}p_2p_3$$
(60)
completes the construction of weight-5 2-primitives. More generally, we found that
$`W(p_2,X_n)`$ $`=`$ $`{\displaystyle \frac{2O_2(X_n)}{n(n+1)}}p_2X_n`$ (61)
$`O_2`$ $`:=`$ $`S_2(Y+\mathrm{id})NS_1(Y+\frac{1}{2}\mathrm{id})`$ (62)
gives $`U_1W(p_2,X_n)=p_2X_nX_np_2`$, where $`X_n`$ is a 1-primitive with weight $`n`$. We remark that (61) lies in the kernel of $`C`$, for all $`n>1`$.
However, it is not yet clear how to generalize this construction to obtain, for example, $`W(p_3,p_4)`$ and $`W(p_3,p_4^{})`$ at weight $`n=7`$. They key to this issue is an extension<sup>1</sup><sup>1</sup>1Our extension of natural growth allows a suitable extension of the Lie algebra dual to $`_R`$, as was observed by Alain Connes. This will be presented in a sequel to . of the concept of natural growth.
### 5.4 Natural growth by appending sums of forests
Let $`F`$ be a forest. We define $`N_F(X)`$ to be the sum of forests obtained by appending $`F`$ to every node of $`X`$, in turn. To append $`F=_jt_j`$ to a particular node, one connects the roots of all the $`t_j`$ to that node. We note that $`N_F`$ obeys a Leibniz rule, with $`N_F(e)=0`$ and $`N_F(l_1)=B_+(F)`$. We have already encountered two examples, namely the grading operator $`Y:=N_e`$, which merely counts nodes, and the simplest natural growth operator $`N:=N_{l_1}`$, which appends a single node. Finally, with $`Z=F_1+F_2`$, we make $`N_Z:=N_{F_1}+N_{F_2}`$ linear in its subscript, as well as its argument.
The commutation relations (55,56) then generalize to
$`[N_Z,B_+]`$ $`=`$ $`B_+L_Z`$ (63)
$`[B_{},N_Z]`$ $`=`$ $`L_ZB_{}`$ (64)
with $`L_Z(X):=ZX`$. Thus $`[N_Z,C]=0`$ and $`N_Z`$ preserves the kernel of $`C`$ for all $`Z_R`$.
The great virtue of this construct is that it gives
$$U_1N_Z(X)=ZY(X)$$
(65)
when both $`Z`$ and $`X`$ are 1-primitive.
Proof: We use the shorthand notation $`\mathrm{\Delta }(X)=Xe+eX+X^{}X^{\prime \prime }`$ for any Hopf algebra element $`X`$, with the final term denoting a sum over tensor products containing no scalars. Let $`Z`$ be any 1-primitive. Then
$$U_1N_Z(X)=N_Z(X^{})X^{\prime \prime }+X^{}N_Z(X^{\prime \prime })+(L_ZY)\mathrm{\Delta }(X)$$
(66)
consists of terms in which $`X^{}`$ or $`X^{\prime \prime }`$ grow naturally, with a final contribution where $`Z`$ is itself completely cut from any node to which it was connected by $`N_Z`$, with the grading operator $`Y`$ acting on the right, to count the number of cuts. The case with $`Z=l_1`$ was proven in , by an analysis of admissible cuts. Here, where $`Z`$ is 1-primitive, we obtain a result of the same form, since the internal cuts of $`Z`$ cancel when $`U_1(Z)=0`$. (A more general formula, for arbitrary $`Z`$, can be given but is not required here.) When $`X`$ is 1-primitive, with $`X^{}=X^{\prime \prime }=0`$, we obtain (65) from $`L_ZY`$ acting on the second term of $`\mathrm{\Delta }(X)=Xe+eX`$$`\mathrm{}`$
The result (65) immediately proves that $`H_2(x)=[H_1(x)]^2`$, since it shows that each pairing $`N_{X_1}(X_2)`$ of 1-primitives gives an element of $`_2`$ that is inequivalent to any other pairing. Hence (24) is saturated at $`k=2`$. Now we define the iteration
$$V_{k+1}(X_1,\mathrm{},X_k,X_{k+1}):=N_{V_k(X_1,\mathrm{},X_k)}(X_{k+1})$$
(67)
for $`k>0`$, with $`V_1:=\mathrm{id}`$. Then, for example, $`V_2(X_1,X_2):=N_{X_1}(X_2)`$ and
$$V_3(X_1,X_2,X_3):=N_{N_{X_1}(X_2)}(X_3)N_{X_1}\left(N_{X_2}(X_3)\right).$$
(68)
We remark that a Hochschild boundary can be defined for maps $`V_{k+1}:_R^{(k+1)}_R`$. For this, it is sufficient to define terms of the form $`V_k(X_1,\mathrm{},X_jX_{j+1},\mathrm{},X_{k+1})`$, where one argument is a product. Natural growth by forests supplies this. Consequences will be described in future work. For the present, we are content with the following result.
Theorem: The dimensions $`H_{n,k}`$ of the bigrading of the Hopf algebra of undecorated rooted trees, by weight $`n`$ and degree of primitivity $`k`$, are generated by (1).
Proof: Let $`X_1,X_2,\mathrm{},X_k`$ be 1-primitives, which need not be distinct. Then
$$U_{k1}V_k(X_1,X_2,\mathrm{},X_k)=X_1Y(X_2)\mathrm{}Y(X_k)$$
(69)
by coassociativity and iteration of the argument that led to (65). Thus $`H_k(x)=[H_1(x)]^k`$ saturates (24). Then $`H(x,y)=1/(1H_1(x)y)`$ gives $`R(x)=x/(1H_1(x))`$, at $`y=1`$. Solving for $`H_1(x)=1x/R(x)`$, we obtain (1). $`\mathrm{}`$
### 5.5 Comments on the main theorem
Four comments are in order. The first concerns the enumeration of the filtration. This follows from taking logs in (33), which gives
$$\mathrm{log}H(x,y)=\mathrm{log}(1H_1(x)y)=\underset{n,k}{}P_{n,k}\mathrm{log}(1x^ny^k).$$
(70)
Equating coefficients of $`y^j`$, and setting $`x=z^{1/j}`$, we obtain
$$[H_1(z^{1/j})]^j=\underset{k|j}{}kP_k(z^{1/k})$$
(71)
which is a classic problem in Mรถbius inversion, yielding (35), after use of (2).
Next, we remark on the number, $`C_{n,k}`$, of weight-$`n`$ elements of $`_k`$ that are in the kernel of $`C:=B_+B_{}`$. We have explicitly constructed a filtration of the bigrading, for weights $`n<7`$, in which the only element with $`C(X)0`$ is $`l_1`$. The iteration (67) proves that there is no obstacle to continuing this process, since the only restriction imposed by $`CV_{k+1}(X_1,X_2,\mathrm{},X_{k+1})=0`$ is $`X_{k+1}l_1`$. Thus $`_nC_{n,k+1}x^n=[H_1(x)]^k(H_1(x)x)`$ and the generating function
$$\underset{n,k}{}C_{n,k}x^ny^k=\frac{(1xy)R(x)}{(1y)R(x)+xy}$$
(72)
differs from (1) only by a factor of $`1xy`$, which removes $`l_1`$ from the filtration (33). In total, we have $`C_n:=_kC_{n,k}=r_{n+1}r_n`$ weight-$`n`$ solutions to $`C(X)=0`$. It is easy to see how that comes about: there are $`r_{n+1}`$ possible forests in $`X`$, subject to the $`r_n`$ conditions that the coefficient of every tree in $`C(X)`$ vanishes. The result $`C_n=r_{n+1}r_n`$ proves the independence of these conditions. Hence an element $`X`$ of the kernel of $`C`$ is uniquely identified by the contribution $`\overline{X}`$ that contains no pure trees, since $`X=\overline{X}C(\overline{X})`$. Finally, the filtration of the bigrading of the kernel of $`C`$ differs from that of the full Hopf algebra only by the absence of $`l_1`$. These distinctive features frustrate every attempt to decrease primitivity by the action of $`B_{}`$ on any nonproduct element except the single-node tree. One may climb up the ladder of primitivity with great ease, yet descent is impossible, save in one trivial case. In a sense, the second grading is characterized by the profound difficulty of constructing its 1-primitives. At first meeting, this makes it difficult to fathom. Then one realizes that the structure is beautifully tuned to prevent casual construction.
Our third comment concerns the remarkable operator $`O_2`$ in (62), which provides a way of solving $`U_1W(p_2,X)=p_2XXp_2`$. A second way is provided by $`N_{p_2}`$. These solutions need not be the same; they may differ by a 1-primitive. In general, they will differ, since $`N_{p_2}`$ acts by a Leibniz rule, while $`O_2`$ does not. Hence
$$T_2:=O_2N_{p_2}(Y+\mathrm{id})$$
(73)
provides a second shift operator that creates 1-primitives, when applied to 1-primitives. It gives information that is not provided by $`S_1`$ in (58). For example, at weight $`n=6`$ we already know how to construct 4 of the $`H_{6,1}=8`$ primitives, by applying powers of $`S_1`$ to the ladder primitives constructed in (8). Of the missing 4, the constructs $`T_2(p_4)`$ and $`T_2S_1(p_3)`$ provide 2. For the remaining 2, which are now proven to exist, we laboriously solved $`U_1(X)=0`$ at weight $`n=6`$, working with tensor products of the 38 forests with up to 6 nodes. At first sight, one might hope to add a few more shift operators, to arrive at a set that is sufficient to construct 1-primitives up to some large weight, without having to solve the fearsome explosion of linear equations required by the vanishing of all tensor products in $`U_1(X)=0`$. This seems not to be the case; the construction of 1-primitives appears to be a deeply nontrivial challenge. Asymptotically, no more than a fraction $`1/c`$ of what is necessary may be provided by $`S_1`$, and no more than $`1/c^2`$ by $`T_2`$, which increases weight by 2 units. The number of similarly constructed operators that change weight by $`n`$ cannot exceed the number $`H_{n,1}`$ of weight-$`n`$ 1-primitives. Constructing a finite number of these, we obtain merely an asymptotic fraction $`f<H_1(1/c)=11/c<1`$ of what is needed. Hence we envisage no easy route to the construction of 1-primitives, short of solving the tensorial defining property $`\mathrm{\Delta }(X)=Xe+eX`$. Thereafter, the problem of constructing $`k`$-primitives is completely solved by (67), which shows that the 1-primitives of weight $`n>1`$ are enumerated by those elements of the kernel of $`C`$ that cannot be generated by any process of natural growth acting on 1-primitives of lesser weight. This negative criterion appears even harder to implement than the tensorial definition $`U_1(X)=0`$, which we were able to solve at $`n=9`$, by explicit computation of the 98-dimensional kernel of a $`3214\times 719`$ matrix of integers.
Finally, we remark that we have explicit constructions of the bigradings (30,32) of the Connes-Moscovici and ladder subalgebras. In the case of $`_{\mathrm{CM}}`$ we have merely a pair of 1-primitives: $`\delta _1=l_1`$ and $`\stackrel{~}{\delta }_2=N_{\delta _1}(\delta _1)\frac{1}{2}\delta _1^2`$. The only form of natural growth that we are allowed is by a single node: this is the defining restriction. Then we easily construct $`\stackrel{~}{\delta }_{k+1}=N_{\delta _1}^k(\delta _1)2^kk!\delta _1^{k+1}`$ as a nonproduct $`k`$-primitive of weight $`k+1`$. This completes the filtration, since any further term would make the number of weight-$`n`$ products of filtered elements greater than the number of weight-$`n`$ products of the linear basis. Hence the construction of the Connes-Moscovici bigrading is particularly simple. In the case of $`_{\mathrm{ladder}}`$ the cocommutativity of the ladder restriction (6) of the coproduct means that all $`k`$-primitives are products at $`k>1`$. Here the problem of construction is more demanding, since it not clear how to generate an infinite set of 1-primitives. Hence one sees that detailed study of ladder diagrams, most notably by Bob Delbourgo and colleagues , addresses a problem more severe than that posed by the Connes-Moscovici prolegomenon to noncommutative geometry: ladder diagrams are a nontrivial infinite subset of perturbative quantum field theory; even after subtractions of products they provide an infinite subset of 1-primitives, when their bigrading is analyzed. Fortunately, our recent work in provides the explicit construction (8) of the ladder filtration. The reader may try to imagine what might be involved in giving an explicit construction of the 1-primitives of the full Hopf algebra of undecorated rooted trees. Then s/he should contemplate the true challenge of quantum field theory, by recalling that โ in physical reality โ every node of every rooted tree may be decorated in an infinite number of ways. After half a century, few physicists or mathematicians have even begun to grapple with the true legacy of Dyson, Feynman, Schwinger and Tomonaga.
## 6 Prospects
In this paper, we were content to study the bigrading of the Hopf algebra of undecorated rooted trees, by the number of nodes and a degree of primitivity analyzed by iterations of the coproduct. The extension of this bigrading to the decorated case is the obvious next step, in our plan to decode the rich structure of mature quantum field theory. The present work makes it clear that the key feature will be the nontriviality of $`C:=B_+B_{}B_{}B_+=\mathrm{id}`$. In the undecorated case, we have shown that the bifiltration of the Hopf algebra is obtained by adjoining the single-node tree to the bifiltration of the kernel of $`C`$. The proof of this lies in the powerful generalization (67) of the concept of natural growth, which diagonalizes (69). First results for the commutator $`[B_+,B_{}]=C\mathrm{id}`$ of the decorated Hopf algebra of full quantum field theory were recently given in . These increase our hopes that it will not take another 50 years to complete the characterization of the intricate interrelation of combinatorics and analysis that makes quantum field theory possible. We firmly believe that further elucidation of its structure has much to offer for wide areas of both physics and mathematics.
Acknowledgements: This study began during the workshop Number Theory and Physics at the ESI in November 1999, where we enjoyed discussions with Pierre Cartier, Werner Nahm, Ivan Todorov and Jean-Bernard Zuber. Work with Alain Connes at the IHES supports the present paper. System management by Chris Wigglesworth enabled accumulation of crucial data, which Neil Sloaneโs superseeker helped us to decode.
## Appendix: asymptotic enumerations
Here we consider inequalities inferred from Tables 3 and 4 and show that they persist at large weights, thanks to the upper bound $`c<3`$ on the Otter constant .
Asymptotically, the number of rooted trees is given by
$$r_n=c^nn^{3/2}(b+O(1/n))$$
(74)
with Otter constants that we evaluated in :
```
b = 0.43992401257102530404090339143454476479808540794011
98576534935450226354004204764605379862197779782334...
c = 2.95576528565199497471481752412319458837549230466359
65953504724789059647331395749510866682836765813525...
```
The asymptotic fraction of trees assigned to primitivity $`k`$ in the filtration of Table 4 is
$$f_k:=\underset{n\mathrm{}}{lim}\frac{P_{n,k}}{r_n}=\left(1\frac{1}{c}\right)^{k1}\frac{1}{c}$$
(75)
while the asymptotic fraction of forests in Table 3 is
$$g_k:=\underset{n\mathrm{}}{lim}\frac{H_{n,k}}{r_{n+1}}=\frac{kf_k}{c}=\left(1\frac{1}{c}\right)^{k1}\frac{k}{c^2}.$$
(76)
These follow by using $`|1R(x)|^2=O(1cx)`$, near $`x=1/c`$. Numerically,
$`g_1`$ $`=`$ $`0.1144616788557279695\mathrm{}`$
$`g_2`$ $`=`$ $`0.1514735822429146084\mathrm{}`$
$`g_3`$ $`=`$ $`0.1503401379409753267\mathrm{}`$
$`g_4`$ $`=`$ $`0.1326357110750687024\mathrm{}`$
$`g_5`$ $`=`$ $`0.1097026887662558145\mathrm{}`$
$`g_6`$ $`=`$ $`0.0871054456752243543\mathrm{}`$
$`g_7`$ $`=`$ $`0.0672417311397409555\mathrm{}`$
$`g_8`$ $`=`$ $`0.0508484386279160206\mathrm{}`$
$`g_9`$ $`=`$ $`0.0378509630072558308\mathrm{}`$
with $`k=2`$ giving the largest fraction of forests at large $`n`$. This was not apparent until $`n=28`$, where we found that $`H_{28,2}=\mathrm{20\hspace{0.17em}716\hspace{0.17em}895\hspace{0.17em}918}`$ exceeds $`H_{28,3}=\mathrm{20\hspace{0.17em}710\hspace{0.17em}700\hspace{0.17em}277}`$.
The asymptotic results establish inequalities (53,54) at large $`n`$, where it is sufficient that $`c<3`$. Amusingly, this upper bound and the condition $`R(1/c)=1`$ produce a rather tight lower bound
$$c=\mathrm{exp}\left(\underset{k>0}{}\frac{R(c^k)}{k}\right)>\mathrm{exp}\left(1+\underset{k>1}{}\frac{1}{(3^k1)k}\right)>2.943$$
(77)
from the rather loose lower bound $`R(x)R_{\mathrm{ladder}}(x)=x/(1x)`$.
Table 1: Dimensions $`\overline{H}_{n,k}`$ of the bigrading the cocommutative subalgebra, $`_{\mathrm{ladder}}`$
$$\begin{array}{cccccccccccccccccccc}& \hfill 1& \hfill 2& \hfill 3& \hfill 4& \hfill 5& \hfill 6& \hfill 7& \hfill 8& \hfill 9& \hfill 10& \hfill 11& \hfill 12& \hfill 13& \hfill 14& \hfill 15& \hfill 16& \hfill 17& \hfill 18& \hfill 19\\ & & & & & & & & & & & & & & & & & & & \\ \hfill 1& \hfill 1& & & & & & & & & & & & & & & & & & \\ \hfill 2& \hfill 1& \hfill 1& & & & & & & & & & & & & & & & & \\ \hfill 3& \hfill 1& \hfill 1& \hfill 1& & & & & & & & & & & & & & & & \\ \hfill 4& \hfill 1& \hfill 2& \hfill 1& \hfill 1& & & & & & & & & & & & & & & \\ \hfill 5& \hfill 1& \hfill 2& \hfill 2& \hfill 1& \hfill 1& & & & & & & & & & & & & & \\ \hfill 6& \hfill 1& \hfill 3& \hfill 3& \hfill 2& \hfill 1& \hfill 1& & & & & & & & & & & & & \\ \hfill 7& \hfill 1& \hfill 3& \hfill 4& \hfill 3& \hfill 2& \hfill 1& \hfill 1& & & & & & & & & & & & \\ \hfill 8& \hfill 1& \hfill 4& \hfill 5& \hfill 5& \hfill 3& \hfill 2& \hfill 1& \hfill 1& & & & & & & & & & & \\ \hfill 9& \hfill 1& \hfill 4& \hfill 7& \hfill 6& \hfill 5& \hfill 3& \hfill 2& \hfill 1& \hfill 1& & & & & & & & & & \\ \hfill 10& \hfill 1& \hfill 5& \hfill 8& \hfill 9& \hfill 7& \hfill 5& \hfill 3& \hfill 2& \hfill 1& \hfill 1& & & & & & & & & \\ \hfill 11& \hfill 1& \hfill 5& \hfill 10& \hfill 11& \hfill 10& \hfill 7& \hfill 5& \hfill 3& \hfill 2& \hfill 1& \hfill 1& & & & & & & & \\ \hfill 12& \hfill 1& \hfill 6& \hfill 12& \hfill 15& \hfill 13& \hfill 11& \hfill 7& \hfill 5& \hfill 3& \hfill 2& \hfill 1& \hfill 1& & & & & & & \\ \hfill 13& \hfill 1& \hfill 6& \hfill 14& \hfill 18& \hfill 18& \hfill 14& \hfill 11& \hfill 7& \hfill 5& \hfill 3& \hfill 2& \hfill 1& \hfill 1& & & & & & \\ \hfill 14& \hfill 1& \hfill 7& \hfill 16& \hfill 23& \hfill 23& \hfill 20& \hfill 15& \hfill 11& \hfill 7& \hfill 5& \hfill 3& \hfill 2& \hfill 1& \hfill 1& & & & & \\ \hfill 15& \hfill 1& \hfill 7& \hfill 19& \hfill 27& \hfill 30& \hfill 26& \hfill 21& \hfill 15& \hfill 11& \hfill 7& \hfill 5& \hfill 3& \hfill 2& \hfill 1& \hfill 1& & & & \\ \hfill 16& \hfill 1& \hfill 8& \hfill 21& \hfill 34& \hfill 37& \hfill 35& \hfill 28& \hfill 22& \hfill 15& \hfill 11& \hfill 7& \hfill 5& \hfill 3& \hfill 2& \hfill 1& \hfill 1& & & \\ \hfill 17& \hfill 1& \hfill 8& \hfill 24& \hfill 39& \hfill 47& \hfill 44& \hfill 38& \hfill 29& \hfill 22& \hfill 15& \hfill 11& \hfill 7& \hfill 5& \hfill 3& \hfill 2& \hfill 1& \hfill 1& & \\ \hfill 18& \hfill 1& \hfill 9& \hfill 27& \hfill 47& \hfill 57& \hfill 58& \hfill 49& \hfill 40& \hfill 30& \hfill 22& \hfill 15& \hfill 11& \hfill 7& \hfill 5& \hfill 3& \hfill 2& \hfill 1& \hfill 1& \\ \hfill 19& \hfill 1& \hfill 9& \hfill 30& \hfill 54& \hfill 70& \hfill 71& \hfill 65& \hfill 52& \hfill 41& \hfill 30& \hfill 22& \hfill 15& \hfill 11& \hfill 7& \hfill 5& \hfill 3& \hfill 2& \hfill 1& \hfill 1\end{array}$$
Table 2: Dimensions $`\stackrel{~}{H}_{n,k}`$ of the bigrading the noncocommutative subalgebra, $`_{\mathrm{CM}}`$
$$\begin{array}{cccccccccccccccccccc}& \hfill 1& \hfill 2& \hfill 3& \hfill 4& \hfill 5& \hfill 6& \hfill 7& \hfill 8& \hfill 9& \hfill 10& \hfill 11& \hfill 12& \hfill 13& \hfill 14& \hfill 15& \hfill 16& \hfill 17& \hfill 18& \hfill 19\\ & & & & & & & & & & & & & & & & & & & \\ \hfill 1& \hfill 1& & & & & & & & & & & & & & & & & & \\ \hfill 2& \hfill 1& \hfill 1& & & & & & & & & & & & & & & & & \\ \hfill 3& & \hfill 2& \hfill 1& & & & & & & & & & & & & & & & \\ \hfill 4& & \hfill 1& \hfill 3& \hfill 1& & & & & & & & & & & & & & & \\ \hfill 5& & & \hfill 2& \hfill 4& \hfill 1& & & & & & & & & & & & & & \\ \hfill 6& & & \hfill 1& \hfill 4& \hfill 5& \hfill 1& & & & & & & & & & & & & \\ \hfill 7& & & & \hfill 2& \hfill 6& \hfill 6& \hfill 1& & & & & & & & & & & & \\ \hfill 8& & & & \hfill 1& \hfill 4& \hfill 9& \hfill 7& \hfill 1& & & & & & & & & & & \\ \hfill 9& & & & & \hfill 2& \hfill 7& \hfill 12& \hfill 8& \hfill 1& & & & & & & & & & \\ \hfill 10& & & & & \hfill 1& \hfill 4& \hfill 11& \hfill 16& \hfill 9& \hfill 1& & & & & & & & & \\ \hfill 11& & & & & & \hfill 2& \hfill 7& \hfill 16& \hfill 20& \hfill 10& \hfill 1& & & & & & & & \\ \hfill 12& & & & & & \hfill 1& \hfill 4& \hfill 12& \hfill 23& \hfill 25& \hfill 11& \hfill 1& & & & & & & \\ \hfill 13& & & & & & & \hfill 2& \hfill 7& \hfill 18& \hfill 31& \hfill 30& \hfill 12& \hfill 1& & & & & & \\ \hfill 14& & & & & & & \hfill 1& \hfill 4& \hfill 12& \hfill 27& \hfill 41& \hfill 36& \hfill 13& \hfill 1& & & & & \\ \hfill 15& & & & & & & & \hfill 2& \hfill 7& \hfill 19& \hfill 38& \hfill 53& \hfill 42& \hfill 14& \hfill 1& & & & \\ \hfill 16& & & & & & & & \hfill 1& \hfill 4& \hfill 12& \hfill 29& \hfill 53& \hfill 67& \hfill 49& \hfill 15& \hfill 1& & & \\ \hfill 17& & & & & & & & & \hfill 2& \hfill 7& \hfill 19& \hfill 42& \hfill 71& \hfill 83& \hfill 56& \hfill 16& \hfill 1& & \\ \hfill 18& & & & & & & & & \hfill 1& \hfill 4& \hfill 12& \hfill 30& \hfill 60& \hfill 94& \hfill 102& \hfill 64& \hfill 17& \hfill 1& \\ \hfill 19& & & & & & & & & & \hfill 2& \hfill 7& \hfill 19& \hfill 44& \hfill 83& \hfill 121& \hfill 123& \hfill 72& \hfill 18& \hfill 1\end{array}$$
Table 3: Dimensions $`H_{n,k}`$ of the bigrading of the Hopf algebra of rooted trees, $`_R`$
$$\begin{array}{cccccccccccccc}& \hfill 1& \hfill 2& \hfill 3& \hfill 4& \hfill 5& \hfill 6& \hfill 7& \hfill 8& \hfill 9& \hfill 10& \hfill 11& \hfill 12& \hfill 13\\ & & & & & & & & & & & & & \\ 1\hfill & \hfill 1& & & & & & & & & & & & \\ 2\hfill & \hfill 1& \hfill 1& & & & & & & & & & & \\ 3\hfill & \hfill 1& \hfill 2& \hfill 1& & & & & & & & & & \\ 4\hfill & \hfill 2& \hfill 3& \hfill 3& \hfill 1& & & & & & & & & \\ 5\hfill & \hfill 3& \hfill 6& \hfill 6& \hfill 4& \hfill 1& & & & & & & & \\ 6\hfill & \hfill 8& \hfill 11& \hfill 13& \hfill 10& \hfill 5& \hfill 1& & & & & & & \\ 7\hfill & \hfill 16& \hfill 26& \hfill 27& \hfill 24& \hfill 15& \hfill 6& \hfill 1& & & & & & \\ 8\hfill & \hfill 41& \hfill 58& \hfill 63& \hfill 55& \hfill 40& \hfill 21& \hfill 7& \hfill 1& & & & & \\ 9\hfill & \hfill 98& \hfill 142& \hfill 148& \hfill 132& \hfill 100& \hfill 62& \hfill 28& \hfill 8& \hfill 1& & & & \\ 10\hfill & \hfill 250& \hfill 351& \hfill 363& \hfill 322& \hfill 251& \hfill 168& \hfill 91& \hfill 36& \hfill 9& \hfill 1& & & \\ 11\hfill & \hfill 631& \hfill 890& \hfill 912& \hfill 804& \hfill 635& \hfill 444& \hfill 266& \hfill 128& \hfill 45& \hfill 10& \hfill 1& & \\ 12\hfill & \hfill 1646& \hfill 2282& \hfill 2330& \hfill 2051& \hfill 1625& \hfill 1167& \hfill 742& \hfill 402& \hfill 174& \hfill 55& \hfill 11& \hfill 1& \\ 13\hfill & \hfill 4285& \hfill 5948& \hfill 6036& \hfill 5304& \hfill 4220& \hfill 3072& \hfill 2030& \hfill 1184& \hfill 585& \hfill 230& \hfill 66& \hfill 12& \hfill 1\end{array}$$
Table 4: Filtration $`P_{n,k}`$ of the bigrading of $`_R`$
$$\begin{array}{ccccccccccccc}& \hfill 1& \hfill 2& \hfill 3& \hfill 4& \hfill 5& \hfill 6& \hfill 7& \hfill 8& \hfill 9& \hfill 10& \hfill 11& \hfill 12\\ & & & & & & & & & & & & \\ 1\hfill & \hfill 1& & & & & & & & & & & \\ 2\hfill & \hfill 1& & & & & & & & & & & \\ 3\hfill & \hfill 1& \hfill 1& & & & & & & & & & \\ 4\hfill & \hfill 2& \hfill 1& \hfill 1& & & & & & & & & \\ 5\hfill & \hfill 3& \hfill 3& \hfill 2& \hfill 1& & & & & & & & \\ 6\hfill & \hfill 8& \hfill 5& \hfill 4& \hfill 2& \hfill 1& & & & & & & \\ 7\hfill & \hfill 16& \hfill 13& \hfill 9& \hfill 6& \hfill 3& \hfill 1& & & & & & \\ 8\hfill & \hfill 41& \hfill 28& \hfill 21& \hfill 13& \hfill 8& \hfill 3& \hfill 1& & & & & \\ 9\hfill & \hfill 98& \hfill 71& \hfill 49& \hfill 33& \hfill 20& \hfill 10& \hfill 4& \hfill 1& & & & \\ 10\hfill & \hfill 250& \hfill 174& \hfill 121& \hfill 79& \hfill 50& \hfill 27& \hfill 13& \hfill 4& \hfill 1& & & \\ 11\hfill & \hfill 631& \hfill 445& \hfill 304& \hfill 201& \hfill 127& \hfill 74& \hfill 38& \hfill 16& \hfill 5& \hfill 1& & \\ 12\hfill & \hfill 1646& \hfill 1137& \hfill 776& \hfill 510& \hfill 325& \hfill 192& \hfill 106& \hfill 49& \hfill 19& \hfill 5& \hfill 1& \\ 13\hfill & \hfill 4285& \hfill 2974& \hfill 2012& \hfill 1326& \hfill 844& \hfill 512& \hfill 290& \hfill 148& \hfill 65& \hfill 23& \hfill 6& \hfill 1\end{array}$$ |
warning/0001/gr-qc0001055.html | ar5iv | text | # References
A remark on Brans-Dicke cosmological
dust solutions with negative $`\omega `$
A.B. Batista<sup>1</sup><sup>1</sup>1e-mail brasil@cce.ufes.br, J. C. Fabris<sup>2</sup><sup>2</sup>2e-mail: fabris@cce.ufes.br and R. de Sรก Ribeiro<sup>3</sup><sup>3</sup>3e-mail: ribeiro@cce.ufes.br
Departamento de Fรญsica, Universidade Federal do Espรญrito Santo, 29060-900, Vitรณria, Espรญrito Santo, Brazil
## Abstract
Analysing the Brans-Dicke solutions for the dust phase, we show that, for negative values of $`\omega `$, they contain scenarios that display an initial subluminal expansion followed by an inflationary phase. We discuss these solutions with respect to the results of the observation of high redshif supernova as well as the age problem and structure formation. We stablish possible connections of these solutions with those emerging from string effective models.
PACS number(s): 04.20.Cv., 04.20.Me
The recent results coming from the high redshift supernova observations indicate that the Universe today may be in an accelerated phase . This suggests that the energy content of the Universe may be dominated today by a fluid with negative pressure. The cosmological constant is a natural candidate for this dark energy. The cosmological constant may be represented by a perfect fluid with an equation of state $`p=\rho `$. However, there is a controverse about the effective equation of state of the exotic fluid responsable for the acceleration of the expansion. In it is claimed that this exotic fluid may be represented by an equation $`p=\alpha \rho `$, with $`1\alpha 0.6`$, while in it is found the limits $`1\alpha 0.8`$. The confirmation of the preliminary results of the high redshif supernova program will bring new important questions about the nature of the dark matter in the Universe.
To take into account the possibility of an inflationary phase in the Universe today, it has been proposed recently a model where a scalar field is minimally coupled to gravity containing a suitable potential . The shape of the potential term is such that the effective equation of state evolves from $`\alpha =\frac{1}{3}`$, characteristic of a radiative phase, to a negative value near $`1`$. This model has been called quintessence. In this letter we would like to call attention to the fact that the prototype of a scalar-tensor theory, with a non-minimally coupled scalar field and no potential, the Brans-Dicke theory, has a class of solutions for the dust equation of state $`p=0`$ exhibiting a non-inflationary initial regime and an inflationary phase in the asymptotic limit. This class of asymptotic dust inflationary solutions can be consistent, at least from the kinetic point of view, with the recent results of an accelerating Universe. We stress that some studies of quintessence model have been carried out in the context of Brans-Dicke theory, but employing a sef-interacting scalar field , what is not the case analyzed here.
In fact, let us consider the Brans-Dicke Lagrangian,
$$L=\sqrt{g}(\varphi R\omega \frac{\varphi _{;\rho }\varphi ^{;\rho }}{\varphi }).$$
(1)
The parameter $`\omega `$ defines the coupling of the scalar field to gravity. Considering a Friedmann-Robertson-Walker flat Universe, this Lagrangian results in the following equations of motion:
$`3({\displaystyle \frac{\dot{a}}{a}})^2`$ $`=`$ $`{\displaystyle \frac{8\pi \rho }{\varphi }}+{\displaystyle \frac{\omega }{2}}\left({\displaystyle \frac{\dot{\varphi }}{\varphi }}\right)3{\displaystyle \frac{\dot{a}}{a}}{\displaystyle \frac{\dot{\varphi }}{\varphi }},`$ (2)
$`\ddot{\varphi }+3{\displaystyle \frac{\dot{a}}{a}}{\displaystyle \frac{\dot{\varphi }}{\varphi }}`$ $`=`$ $`{\displaystyle \frac{8\pi \rho }{3+2\omega }},`$ (3)
$`\dot{\rho }+3{\displaystyle \frac{\dot{a}}{a}}\rho `$ $`=`$ $`0.`$ (4)
The last equation leads to the $`\rho =\rho _0a^3`$. Inserting this relation in (3), we find the first integral, $`\dot{\varphi }=\frac{8\pi \rho t}{3+2\omega }C`$, where $`C`$ is a constant.
Following , we can define an auxiliary function $`u`$, satisfying the relation
$$\frac{\dot{u}}{u}=3\frac{\dot{a}}{a}+\frac{2}{t}\frac{u}{t}$$
(5)
which, in view of the equations of motion, results in the following integral relation
$$\frac{du}{u[u+4\pm a\sqrt{u^2+4u}]}=2\mathrm{ln}(tt_c),$$
(6)
where $`a=3\sqrt{1+\frac{2}{3}\omega }`$. This integral relation has three critical points $`u=0`$, $`u=4`$ and $`u=\frac{2}{4+3\omega }`$. The first one gives non physical results, while the second one leads to $`at^2`$ for any value of $`\omega `$. However, the energy density of ordinary matter associated to this solution is negative.
The third critical point of (6) leads to a particular solution in terms of power law function:
$$a=a_0t^{\frac{2+2\omega }{4+3\omega }},\varphi =\varphi _0t^{\frac{2}{4+3\omega }}.$$
(7)
A more general solution may be obtained through integration of (6). This has been donne in , resulting in the general flat solution
$`a`$ $`=`$ $`a_0(tt_{})^{\frac{1+\omega \pm \sqrt{1+\frac{2}{3}\omega }}{4+3\omega }}(tt_+)^{\frac{1+\omega \sqrt{1+\frac{2}{3}\omega }}{4+3\omega }},`$ (8)
$`\varphi `$ $`=`$ $`\varphi _0(tt_{})^{\frac{13\sqrt{1+\frac{2}{3}\omega }}{4+3\omega }}(tt_+)^{\frac{1\pm 3\sqrt{1+\frac{2}{3}\omega }}{4+3\omega }}.`$ (9)
Since $`t_+`$ and $`t_{}`$ are constants such that $`t_+>t_{}`$ , $`t=t_+`$ may be identified with the initial time.
Let us first inspect the solutions (8,9) in more details. They have two asymptotic regimes. Near the initial singularity, we find
$`a`$ $`=`$ $`a_0t^{\frac{1+\omega \pm \sqrt{1+\frac{2}{3}\omega }}{4+3\omega }},`$ (10)
$`\varphi `$ $`=`$ $`\varphi _0t^{\frac{13\sqrt{1+\frac{2}{3}\omega }}{4+3\omega }},`$ (11)
where we have redefined $`tt_+t`$. This is equivalent to the Brans-Dicke vacuum solution. This curious equivalence is easily understood if we look at equations (2,3): inserting in (2,3) the expressions (10,11), we see that the term $`\frac{\rho }{\varphi }`$ behaves as $`\frac{1}{t}`$, while the terms $`(\frac{\dot{a}}{a})^2,(\frac{\dot{\varphi }}{\varphi })^2,\frac{\dot{a}}{a}\frac{\dot{\varphi }}{\varphi }`$ behave as $`\frac{1}{t^2}`$; for small $`t`$, the latter terms dominate over the former one, the energy of the scalar field dominating over the energy of matter, and the vacuum solution is a good approximation. For large values of $`t`$, solutions (8,9) goes to (7).
In general, the solutions (8,9) represent a subluminal expansion, leading to a yonguer Universe than in the standard model based in the General Relativity theory. In the limit $`\omega \mathrm{}`$ they coincide with the dust standard model. The inverse of $`\varphi `$ is linked with the gravitational coupling through the relation $`G=\frac{4+2\omega }{3+2\omega }\frac{1}{\varphi }`$. Hence, in general, solutions (8,9) predict a decreasing gravitational coupling, what is reasonable since the Universe is going asymptotically to a flat geometry.
However, all the description made above is valid for a positive $`\omega `$. More precisely, it is valid for $`\omega >1`$. We intend to analyze here what happens if we allow $`\omega `$ to be negative. We will first make some general considerations. If we have $`\frac{4}{3}<\omega <1`$, the cosmic time must vary as $`\mathrm{}<t0`$, in order to have an expanding Universe. It represents what is generally called a pole-law inflation. If $`\frac{3}{2}<\omega <\frac{4}{3}`$, we have also inflation, but with $`0t<\mathrm{}`$. From these considerations, we have three special values of $`\omega `$: $`\omega =1,\frac{4}{3}`$ and $`\frac{3}{2}`$. The case $`\omega =\frac{3}{2}`$ represents a breakdown of the theory, since the Brans-Dicke Lagrangian may be recast in a form characteristic of a scalar field conformally coupled to gravity; $`\omega =1`$ leads to a constant scale factor with a varying cosmological constant (notice that it represents also the string Lagrangian in presence of ordinary matter); $`\omega =\frac{4}{3}`$ is a special case which we will discuss latter.
Let us consider the solutions (8,9) in the range $`\frac{3}{2}<\omega <\frac{4}{3}`$. Initially, for small $`t`$, we have two branches connected with the signs of the exponent of (10,11). The upper sign represents a subluminal expansion, while the lower sign a superluminal expansion. In the limit of large values of $`t`$, we obtain the solutions (7) which describe, in the same range of values of $`\omega `$, an inflationary Universe. Hence, it is possible to have the following scenario: the Universe enters in the dust phase with a subluminal expansion, evolving latter to a superluminal expansion. For example, if we choose $`\omega =1.4`$ and the upper sign in (10,11), the scale factor behaves initially as $`at^{0.7}`$ and as $`at^4`$ in a latter phase.
Concerning the observational limits for the effective equation of state, they are obtained essentially by inspecting the kinematical properties of the scale factor. In the standard model a barotropic equation of state leads to the solution
$$a=a_0t^{\frac{2}{3(1+\alpha )}}.$$
(12)
We will look for values of $`\omega `$ in (7) that give the same behaviour as (12) for a given value of $`\alpha `$. It has been argued that the most favoured value for the effective equation of state today seems to be $`\alpha 0.77`$. This can also be obtained from the matter dominated Brans-Dicke solutions, in the limit of large $`t`$, if $`\omega 1.4329`$. Notice that $`\alpha =1`$ corresponds to $`\omega =\frac{4}{3}`$. The other limit of validity of the Brans-Dicke cosmological solution, $`\omega =\frac{3}{2}`$, corresponds to $`\alpha =\frac{2}{3}`$ (a fluid of cosmic wall). Hence, $`\frac{3}{2}<\omega <\frac{4}{3}`$ corresponds, from the kinetic point of view, to $`1<\alpha <\frac{2}{3}`$ in the Standard Model.
We go back now to the case $`\omega =\frac{4}{3}`$. It can be seen from the solutions (7,10,11) that this value for $`\omega `$ is not allowed. In the case of the โvacuumโ solution, it can be shown that the power law solution is not possible for this value of $`\omega `$; however, we can find exponential solutions:
$$ae^t,\varphi e^{3t}.$$
(13)
For the deep material phase ($`t\mathrm{}`$), we can integrate exactly (6) for $`\omega =\frac{4}{3}`$, obtaining $`ate^{\frac{t}{3}}`$. However, for this solution the energy density is negative. Hence, there is a pathology for the dust solutions for this specific value of $`\omega `$.
The scenario described here must be tested against two important points. The first one concerns the evolution of density perturbations. The problem of density perturbations in Brans-Dicke theory has been studied in . It can be verified that the inflationary solutions in the dust Universe described here leads, in the large wavelength limit, only to decreasing modes. It must be noted that this is a generally feature of density perturbations in an inflationary phase . However, for the upper sign in (10,11), the initial behaviour is subluminal and very similar to the standard one based in the General Relativity theory. Consequently, there is initially growing modes for density perturbations, and we can expect the formation of galaxies in the same period as in the standard model. Moreover, even for large values of $`t`$, we must remark that, in opposition to the standard models with negative pressure, there is no instability in the small wavelength limit . In the quintessence model, there is only decreasing modes in the large wavelength limit, but the effective equation of state at small scale becomes positive, and the perturbations in this limit oscillate as an accoustic mode, and no instability is present also.
Since for the range of $`\omega `$ defined above the expansion of the Universe is always faster than in the standard model, the Universe in this case is older than in the standard model. This is a nice feature not only concerning the age problem, but also with respect to the structure formation problem.
Secondly, there is the question of the local tests. This leads to the problem of spherical symmetric solutions for the Brans-Dicke theory. It has been shown that there exist black holes in the Brans-Dicke theory only for negative values of $`\omega `$. In a matter of fact, black holes can exist only for $`\omega <\frac{4}{3}`$. These black holes are very special since their Hawking temperature is zero and the area of the horizon surface is infinite. The local test can be satisfied if $`|\omega |500`$, $`\omega `$ being positive or negative . This fact leads us to consider two main possibility to reconcile local test with the considerations made above: localy the scalar field could be essentially constant, what may suggest a scale dependent gravitational coupling, as it has already been evoked in the literature ; or we may consider a variable $`\omega `$, like a non-linear sigma model. Notice that it is possible to have, when $`\omega =`$ constant, cosmological solutions for $`\omega <\frac{3}{2}`$ , but it predicts, for the dust phase, a bouncing Universe; moreover, when transposed, through a conformal transformation, to the Einstein frame, such case exhibits negative kinetic energy, and we must be cautious about its stability.
We remember that the string theory predicts a low energy effective model with $`\omega =1`$. However, in the case of d-branes string model, the value of $`\omega `$ is given by
$$\omega =\frac{(D1)(d2)d^2}{(D2)(d2)d^2}$$
(14)
where $`D`$ and $`d1`$ are the dimension of space-time and of the brane, respectivelly. For example, for $`D=4`$, a 0-brane and 1-brane give $`\omega =\frac{4}{3}`$ and $`\omega =1`$ respectivelly. These values are not very satisfactory with respect to the range of $`\omega `$ considered above. But, it suggest that it might be possible that some specific configuration of the string effective model may lead to those values of $`\omega `$. |
warning/0001/math0001110.html | ar5iv | text | # On infinite tensor products of projective unitary representations
## 1 Introduction
The theory of infinite tensor products of Hilbert spaces started with the seminal paper by von Neumann . Later on, Guichardet approached this matter from a slightly different point of view and developed a unified framework for treating several related concepts involving operators, algebras and functionals. The notion of infinite tensor product has been mainly used in this form in operator algebras and quantum field theory over the last three decades (see
e. g. for a recent overview).
The existence of some infinite tensor product of representations of a group has been established and used in some recent works. For example, it was shown in that a locally compact group is $`\sigma `$-compact and amenable if and only if there exists an infinite tensor power of its regular representation. Such an infinite tensor power construction was then a useful tool for studying covariance of certain (induced) product-type representations of generalized Cuntz algebras with respect to natural product-type actions. This circle of ideas has been generalized and thoroughly investigated in . In another direction, the infinite tensor product of certain unitary representations of some group of diffeomorphisms was shown to exist under suitable assumptions in .
In this paper we initiate a study of infinite tensor products of *projective* unitary representations of a discrete group. It is in fact not obvious that such infinite tensor products exist at all. Indeed it is quite easy to realize that it is impossible to form the infinite tensor power of a single projective unitary representation unless the associated 2-cocycle vanishes. Besides its intrinsic interest, this new generality has the potential advantage to allow for extensions of the analysis given in to a broader class of product-type actions on the $`0^{th}`$-degree part of extended Cuntz algebras. It is also relevant when studying extensions of product-type actions from the algebraic to the von Neumann algebra level. Finally it provides a way to represent faithfully on infinite tensor product spaces some familiar C\*-algebras like non-commutative tori. To avoid technicalities, we stick to the case of a discrete group, although it could be of interest in the future to consider a locally compact (or even just a topological) group and strongly continuous projective unitary representations of such a group.
The paper is organized as follows. Section 2 is devoted to some preliminaries on projective unitary representations, product sequences of 2-cocycles and infinite tensor products. Section 3 contains our main existence results for infinite tensor products of projective unitary representations. We especially display some sufficient conditions for countable amenable groups in the case of projective regular representations and in the case of projective representations associated with CCR-representations of bilinear maps. To illustrate our work we specialize in section 4 to the case of finitely generated free abelian groups. The final section deals with infinite tensor products of actions of a discrete group $`G`$ on von Neumann algebras. We concentrate our attention to the existence problem of such product actions in the case of unitarily implemented actions. One of our result exhibits an obstruction for extending some algebraic tensor power action of $`G`$ to the weak closure that lies in the second cohomology group $`H^2(G,๐)`$. In another result, the obstruction lies in the non-amenability of $`G`$.
## 2 Preliminaries
Throughout this note $`G`$ denotes a non-trivial discrete group, with neutral element $`e`$.
A *2-cocycle* (or multiplier) on $`G`$ with values in the circle group $`๐`$ is a map $`u:G\times G๐`$ such that
$$u(x,y)u(xy,z)=u(y,z)u(x,yz)(x,y,zG),$$
see e.g. \[5, Chapter IV\]. We will consider only *normalized* 2-cocycles, satisfying
$$u(x,e)=u(e,x)=1(xG).$$
The set of all such 2-cocycles, which is denoted by $`Z^2(G,๐)`$, becomes an abelian group under pointwise product. We equip $`Z^2(G,๐)`$ with the topology of pointwise convergence.
A 2-cocycle $`v`$ on $`G`$ is called a *coboundary* whenever $`v(x,y)=\rho (x)\rho (y)\overline{\rho (xy)}`$ $`(x,yG)`$ for some $`\rho :G๐`$, $`\rho (e)=1`$, in which case we write $`v=\mathrm{d}\rho `$ (such a $`\rho `$ is uniquely determined up to multiplication by a character). The set of all coboundaries, which is denoted by $`B^2(G,๐)`$, is a subgroup of $`Z^2(G,๐)`$, which is easily seen to be closed. (Indeed, assume that $`(\mathrm{d}\rho _\alpha )`$ is a net in $`B^2(G,๐)`$ converging to $`vZ^2(G,๐).`$ Due to Tychonovโs theorem, we may, by passing to a subnet if necessary, assume that $`\rho _\alpha `$ converges pointwise to $`\rho `$, for some $`\rho :G๐`$, $`\rho (e)=1.`$ Then we have $`v=\mathrm{d}\rho .`$)
The quotient group $`H^2(G,๐):=Z^2(G,๐)/B^2(G,๐)`$ is called the *second cohomology group* of $`G`$ with values in $`๐`$. We denote elements in $`H^2(G,๐)`$ by $`[u]`$ and write $`vu`$ when $`[v]=[u](u,vZ^2(G,๐))`$. We also write $`v_\rho u`$ when we have $`v=(\mathrm{d}\rho )u`$ for some coboundary $`\mathrm{d}\rho `$.
We recall a few facts concerning infinite products of complex numbers (see ). Let $`(z_i)`$ denote a sequence of complex numbers. We say that the infinite product $`_iz_i`$ exists (or converges) if the limit of the net $`(_{iJ}z_i)_J`$ exists, where $``$ denotes the family of non-empty finite subsets of $``$ ordered by inclusion. We then also use $`_iz_i`$ to denote this limit. We will need the following result:
Assume that $`_i|1z_i|<\mathrm{}.`$ Then $`_iz_i`$ exists, and $`_iz_i0`$ if all $`z_i`$โs are non-zero. Conversely, assume that $`_iz_i`$ converges to a non-zero element. Then $`_i|1z_i|<\mathrm{}.`$
We shall be interested in *product* sequences in $`Z^2(G,๐)`$: we call a sequence $`(u_i)`$ in $`Z^2(G,๐)`$ a *product* sequence whenever the (pointwise) infinite product $`u=_iu_i`$ exists on $`G\times G`$ ($`u`$ being then obviously a 2-cocycle itself).
A cohomological problem concerning product sequences is that perturbing a product sequence (by a coboundary in each component) does not necessarily lead to a product sequence, as may be illustrated by taking all $`u_i`$โs to be 1 and perturbing by the same coboundary $`v1`$ in each component. The following lemma somewhat clarifies this problem.
###### Lemma 2.1.
Let $`(u_i)`$ and $`(v_i)`$ be two sequences in $`Z^2(G,๐)`$ satisfying $`v_i_{\rho _i}u_i`$ for every $`i`$.
i) Assume that $`\rho :=_i\rho _i`$ exists. Then $`(v_i)`$ is a product sequence if and only if $`(u_i)`$ is a product sequence, in which case we have $`_iv_i_\rho _iu_i`$.
ii) Assume that $`(u_i)`$ and $`(v_i)`$ are both product sequences. Then $`_iv_i_iu_i`$ (even if $`_i\rho _i`$ does not necessarily exist).
###### Proof.
As i) is straightforward, we only show ii). So we assume that $`u=_iu_i`$ and $`v=_iv_i`$ both exist. Then $`w:=_i\mathrm{d}\rho _i=_i\overline{u_i}v_i`$ also exists and is the limit of a net of 2-coboundaries. As $`B^2(G,๐)`$ is closed, this implies that $`wB^2(G,๐)`$. Since $`v=wu`$, this shows that $`vu`$, as asserted.
(To see that $`_i\rho _i`$ does not necessarily exist, assume that $`G`$ possess a non-trivial character $`\gamma .`$ Set $`u_i=v_i=1`$ and $`\rho _i=\gamma `$ for all $`i.`$ Then clearly $`v_i_{\rho _i}u_i`$ while $`_i\rho _i`$ does not exist.) โ
A *projective unitary representation* $`U`$ of $`G`$ on a Hilbert space $``$ associated with some $`uZ^2(G,๐)`$ is a map from $`G`$ into the group of unitaries on $``$ such that
$$U(x)U(y)=u(x,y)U(xy)(x,yG).$$
If we pick a $`\rho :G๐`$ satisfying $`\rho (e)=1`$ and set $`V=\rho U`$, then $`V`$ is also a projective unitary representation of $`G`$ on $``$ associated with a 2-cocycle $`v`$ satisfying $`v_\rho u`$. Such a $`V`$ is called a *perturbation* of $`U`$.
To each $`uZ^2(G,๐)`$ one may associate the left $`u`$-regular projective unitary representation $`\lambda _u`$ of $`G`$ on $`\mathrm{}^2(G)`$ defined by
$$(\lambda _u(x)f)(y)=u(y^1,x)f(x^1y)(f\mathrm{}^2(G),x,yG).$$
Choosing $`u=1`$ gives the left regular representation of $`G`$ which we will just denote by $`\lambda `$. It is well known (and easy to see) that if $`v_\rho u`$, then $`\lambda _v`$ is unitarily equivalent to $`\rho \lambda _u`$.
For $`i=1,2`$, let $`U_i`$ be a projective unitary representation of $`G`$ on a Hilbert space $`_i`$ associated with $`u_iZ^2(G,๐)`$. Then the naturally defined tensor product representation $`U_1U_2`$ is easily seen to be a projective unitary representation of $`G`$ on the Hilbert space $`_1_2`$ associated with the product cocycle $`u_1u_2`$. In the case of ordinary unitary representations of a group, it is a classical result of Fell (cf. , 13.11.3) that the left regular representation acts in an absorbing way with respect to tensoring (up to multiplicity and equivalence). In the projective case we have the following analogue.
###### Proposition 2.2.
Let $`u,v`$ be elements in $`Z^2(G,๐)`$ and let $`V`$ be any projective unitary representation of $`G`$ on a Hilbert space $``$ associated with $`v`$. Then the tensor product representation $`\lambda _uV`$ is unitarily equivalent to $`\lambda _{uv}id_{}`$, i.e. to (dim $`V)\lambda _{uv}`$.
###### Proof.
We leave to the reader to check that the same unitary operator $`W`$ as in the non-projective case ( which is determined on $`\mathrm{}^2(G)(\mathrm{}^2(G,))`$ by $`(W(f\psi ))(x)=f(x)V(x^1)\psi `$) implements the asserted equivalence. โ
We conclude this section with a short review on infinite tensor products of Hilbert spaces and operators. (See for more information.)
Let $`=\{_i\}`$ denote a sequence of Hilbert spaces and $`\varphi =\{\varphi _i\}`$ be a sequence of unit vectors where $`\varphi _i_i`$ for each $`i1`$. We denote by $`^\varphi `$ or by $`_i^\varphi _i`$ the associated infinite tensor product Hilbert space of the $`_i`$โs along the sequence $`\varphi `$.
For any sequence $`\psi _i_i`$ such that
$$\underset{i}{}|\mathrm{\hspace{0.17em}1}\psi _i|<\mathrm{}\text{and}\underset{i}{}|\mathrm{\hspace{0.17em}1}(\psi _i,\varphi _i)|<\mathrm{},$$
there corresponds a so-called decomposable vector in $`^\varphi `$ denoted by $`_i\psi _i`$. If $`_i\xi _i`$ is another decomposable vector in $`^\varphi `$, then
$$(_i\psi _i,_i\xi _i)=\underset{i}{}(\psi _i,\xi _i).$$
A decomposable vector of the form $`\psi _1\mathrm{}\psi _k\varphi _{k+1}\varphi _{k+2}\mathrm{}`$ is called elementary. The set of all elementary decomposable vectors is total in $`^\varphi `$.
Let $`T_1,T_2,\mathrm{}`$ be a sequence of bounded linear operators where each $`T_i`$ acts on $`_i`$. For each fixed $`n`$ there exists a unique bounded linear operator $`\stackrel{~}{T}_n`$ acting on $`^\varphi `$ which is determined by
$$\stackrel{~}{T}_n(_i\psi _i)=T_1\psi _1\mathrm{}T_n\psi _n\psi _{n+1}\psi _{n+2}\mathrm{}$$
for each decomposable vector $`_i\psi _i`$. Similarly, one may define $`\stackrel{~}{T}_J`$ for each (nonempty) finite $`J.`$ Under certain assumptions, the net $`\{\stackrel{~}{T}_J\}`$ converges in the strong operator topology to a bounded linear operator on $`^\varphi `$ which is then denoted by $`_iT_i`$.
By \[12, Part II, Proposition 6\]), a sufficient condition for $`_iT_i`$ to exist is that
$$\underset{i}{}T_i\text{exists},\underset{i}{}|1T_i\varphi _i|<\mathrm{}\text{and}\underset{i}{}|1(T_i\varphi _i,\varphi _i)|<\mathrm{},$$
in which case we have $`(_iT_i)(_i\psi _i)=_iT_i\psi _i`$ for all elementary decomposable vectors $`_i\psi _i.`$
When all $`T_i`$โs are unitaries (which is the case of interest in this paper) we have the following result, which will be used several times in the sequel.
###### Proposition 2.3.
Let $`(T_i)`$ be a sequence of unitaries where each $`T_i`$ acts on $`_i`$. Then $`_iT_i`$ exists on $`^\varphi `$ if and only if
$$()\underset{i}{}|1(T_i\varphi _i,\varphi _i)|<\mathrm{},$$
in which case $`_iT_i`$ is a unitary on $`^\varphi `$ satisfying $`(_iT_i)^{}=_iT_i^{}.`$
###### Proof.
Assume first that $`()`$ holds. It is then quite elementary to deduce from Guichardetโs result mentioned above that $`_iT_i`$ and $`_iT_i^{}`$ both exist. Moreover, these two operators are then isometries, being the strong limit of a net of unitaries, and they are easily seen to be the inverse of each other. So both are unitaries satisfying $`(_iT_i)^{}=_iT_i^{}.`$
Assume now that $`T:=_iT_i`$ exists on $`^\varphi .`$ Then $`T`$ is non-zero (being an isometry), so there are elementary decomposable vectors $`_i\psi _i`$ and $`_i\xi _i`$ such that
$$0c:=(T_i\psi _i,_i\xi _i).$$
Let $`J`$ be any finite subset of $``$ large enough so that $`\psi _i=\xi _i=\varphi _i`$ for all $`iJ.`$ Then we have
$$(\stackrel{~}{T}_J_i\psi _i,_i\xi _i)=\underset{iJ}{}(T_i\psi _i,\xi _i).$$
Since $`T=lim_J\stackrel{~}{T}_J`$, we get $`c=lim_J_{iJ}(T_i\psi _i,\xi _i)`$, i.e. $`_i(T_i\psi _i,\xi _i)`$ converges to a non-zero value.
Thus we get $`_i|1(T_i\psi _i,\xi _i)|<\mathrm{}`$ and therefore $`_i|1(T_i\varphi _i,\varphi _i)|<\mathrm{}`$ since $`\psi _i=\xi _i=\varphi _i`$ for all but finitely many $`i`$โs. โ
## 3 Infinite tensor products of projective unitary representations
Before attacking the main problem whether it is possible to form an infinite tensor product of a sequence of projective unitary representations, at least in some cases, we first show that this construction, when possible, produces a new projective unitary representation of $`G`$, and also make some general observations.
###### Theorem 3.1.
Let $`U_i`$ be a sequence of projective unitary representations of G acting respectively on a Hilbert space $`_i`$ and with associated $`u_iZ^2(G,๐)`$. Let $`\varphi =(\varphi _i)`$ be a sequence of unit vectors where each $`\varphi _i_i`$. Assume that $`_iU_i(x)`$ exists on $`^\varphi =_i^\varphi _i`$ for each $`xG`$. Then we have
i) $`(u_i)`$ is a product sequence in $`Z^2(G,๐).`$
ii) The map $`xU^\varphi (x):=_iU_i(x)`$ is a projective unitary representation of G on $`^\varphi `$ with $`u=_iu_i`$ as its associated 2-cocycle.
iii) If there exists one $`k`$ such that $`U_k`$ is unitarily equivalent to $`\lambda _{u_k}`$, then $`U^\varphi `$ is unitarily equivalent to $`\lambda _uid_{}`$, where $``$ denotes any infinite dimensional separable Hilbert space.
iv) $`\lambda U^\varphi `$ is unitarily equivalent to $`\lambda _uid_^\varphi `$.
###### Proof.
Notice first that Proposition 2.3 implies that each $`U^\varphi (x):=_iU_i(x)`$ is a unitary.
i) Let $`g,hG`$. We must show that $`_iu_i(g,h)`$ converges. Now
$$_iU_i(gh)$$
and
$$(_iU_i(g))(_iU_i(h))=_iU_i(g)U_i(h)=_iu_i(g,h)U_i(gh)$$
are both unitaries. Putting $`a_i=(U_i(gh))\varphi _i,\varphi _i)`$, we deduce from Proposition 2.3 that
$$\underset{i}{}|\mathrm{\hspace{0.17em}1}a_i|<\mathrm{}\text{and}\underset{i}{}|\mathrm{\hspace{0.17em}1}u_i(g,h)a_i|<\mathrm{}.$$
This implies that $`_i|\mathrm{\hspace{0.17em}1}u_i(g,h)|<\mathrm{}`$, and therefore that $`_iu_i(g,h)`$ converges, as desired. ( We use here implicitely that whenever $`z๐`$ and $`a`$, then $`|1z|=|1\overline{z}||1a|+|a\overline{z}|=|1a|+|za1|`$).
ii) Using i) we get
$$U^\varphi (x)U^\varphi (y)=_iu_i(x,y)U_i(xy)=(\underset{i}{}u_i(x,y))_iU_i(xy)=u(x,y)U^\varphi (xy)$$
for all $`x,yG`$, as asserted.
iii) and iv) follow easily from Proposition 2.2. โ
An obvious, but noteworthy consequence of part i) of this theorem is that it is impossible to form the infinite tensor power of a single projective unitary representation unless the associated 2-cocycle vanishes. In another direction, the case where infinitely many of the $`U_i`$โs are projective regular representations of $`G`$ can not occur in this theorem when $`G`$ is uncountable or non-amenable, as easily follows from our next theorem. (We refer to or for information on amenability).
###### Theorem 3.2.
Let $`(u_i)`$ be a sequence in $`Z^2(G,๐)`$ and set $`U_i=\lambda _{u_i}`$ for every $`i`$. Let $`\varphi =(\varphi _i)`$ be a sequence of unit vectors in $`\mathrm{}^2(G).`$ Assume that $`_iU_i(x)`$ exists on $`^\varphi =_i^\varphi \mathrm{}^2(G)`$ for each $`xG`$. Then $`G`$ is countable and amenable.
###### Proof.
Using Proposition 2.3, it follows that $`_i|1(U_i(x)\varphi _i,\varphi _i)|<\mathrm{}`$ for every $`xG`$. Notice that
$$|(U_i(x)\varphi _i,\varphi _i)|(\lambda (x)|\varphi _i|,|\varphi _i|)1.$$
Hence we get
$$(\lambda (x)|\varphi _i|,|\varphi _i|)1(xG).$$
This means that the trivial 1-dimensional representation of $`G`$ is weakly contained in $`\lambda `$ and the amenability of $`G`$ follows.
Moreover, setting $`f_i(x):=|(\lambda (x)\varphi _i,\varphi _i)|0`$ we have $`0f_i1`$, $`f_iC_0(G)`$ (cf. \[8, 13.4.11\]) and $`f_i1`$ pointwise. Then $`f_i^1([1/2,1])=:H_i`$ is finite and $`G=_iH_i`$, so $`G`$ is countable. โ
In view of this theorem, it is quite natural to wonder whether some converse holds. We shall provide a partial answer in Corollary 3.4. To ease our exposition, we introduce some terminology. A sequence $`(F_i)`$ of non-empty, finite subsets of $`G`$ will be called a $`F`$-sequence (resp. $`\sigma F`$-sequence) for $`G`$ whenever
$$\underset{i}{lim}\frac{\mathrm{\#}(F_ixF_i)}{\mathrm{\#}F_i}=1\text{for all}xG,$$
$$\text{(resp.}\underset{i}{}|1\frac{\mathrm{\#}(F_ixF_i)}{\mathrm{\#}F_i}|<\mathrm{}\text{for all}xG).$$
A $`F`$-sequence $`(F_i)`$ for $`G`$ is often called a *Fรถlner* sequence in the literature. We remark that the definition is usually phrased in a slightly different, but equivalent, way (involving the symmetric difference of sets) and that some authors also require that $`F_iF_{i+1}`$ for every $`i`$. Anyhow, thanks to Fรถlner (see ), we know that $`G`$ is countable and amenable if and onl y if $`G`$ has a $`F`$-sequence. Now, obviously, a $`\sigma F`$-sequence for $`G`$ is also a $`F`$-sequence. Moreover, any $`F`$-sequence has some subsequence which is a $`\sigma F`$-sequence, as is easily checked. Hence we can also conclude that $`G`$ is countable and amenable if and only if $`G`$ has a $`\sigma F`$-sequence.
When $`F`$ is a subset of $`G`$, we denote by $`\chi _F`$ its characteristic function.
###### Theorem 3.3.
Let $`(u_i)`$ be a sequence in $`Z^2(G,๐).`$ Assume that $`G`$ is countable and amenable, and has a $`\sigma F`$-sequence $`(F_i)`$ which satisfies
$$()\underset{i}{}\frac{1}{\mathrm{\#}F_i}\underset{yF_i}{}|1u_i(y^1,x)|<\mathrm{}\text{for all}xG.$$
Set $`U_i=\lambda _{u_i}`$ and $`\varphi _i:=\chi _{F_i}/(\mathrm{\#}F_i)^{1/2}`$ for every $`i`$.
Then $`\varphi =(\varphi _i)`$ is a sequence of unit vectors in $`\mathrm{}^2(G)`$ such that $`_iU_i`$ exists on $`^\varphi =_i^\varphi \mathrm{}^2(G)`$.
###### Proof.
We first record some easy calculations. Let $`F`$ be a finite (non-empty) subset of $`G`$ and set $`\varphi _F:=\chi _F/(\mathrm{\#}F)^{1/2}`$. Let $`uZ^2(G,๐)`$. Then we have
$$(\lambda (x)\varphi _F,\varphi _F)=\frac{1}{\mathrm{\#}F}\mathrm{\#}(FxF)$$
for every $`xG.`$ More generally we have
$$(\lambda _u(x)\varphi _F,\varphi _F)=\frac{1}{\mathrm{\#}F}\underset{yFxF}{}u(y^1,x)$$
and therefore
$$((\lambda (x)\lambda _u(x))\varphi _F,\varphi _F)=\frac{1}{\mathrm{\#}F}\underset{yFxF}{}(1u(y^1,x))$$
for all $`xG`$.
Using the triangle inequality and the above computations, we get
$$\underset{i}{}|1(U_i(x)\varphi _i,\varphi _i)|\underset{i}{}|1(\lambda (x)\varphi _i,\varphi _i)|+\underset{i}{}|((\lambda (x)U_i(x))\varphi _i,\varphi _i)|$$
$$=\underset{i}{}|1\frac{\mathrm{\#}(F_ixF_i)}{\mathrm{\#}F_i}|+\underset{i}{}\frac{1}{\mathrm{\#}F_i}|\underset{yF_ixF_i}{}(1u_i(y^1,x))|$$
$$\underset{i}{}|1\frac{\mathrm{\#}(F_ixF_i)}{\mathrm{\#}F_i}|+\underset{i}{}\frac{1}{\mathrm{\#}F_i}\underset{yF_i}{}|1u_i(y^1,x)|$$
for all $`xG.`$ Since $`(F_i)`$ is a $`\sigma F`$-sequence for $`G`$ satisfying $`()`$, both sums above converge for all $`xG`$. Hence, $`_i|1(U_i(x)\varphi _i,\varphi _i)|<\mathrm{}`$ for all $`xG`$ and the assertion follows from Proposition 2.3. โ
Clearly, if $`u_i=1`$ for all but finitely many $`i`$โs, any $`\sigma F`$-sequence $`(F_i)`$ for $`G`$ trivially satisfies $`()`$. In this case, the above theorem could also have been deduced from .
###### Corollary 3.4.
Let $`G`$ be countable and amenable, and let $`(v_j)`$ be a product sequence in $`Z^2(G,๐)`$. Then there exist a subsequence $`(u_i)`$ of $`(v_j)`$ and a sequence $`\varphi =(\varphi _i)`$ of unit vectors in $`\mathrm{}^2(G)`$ such that $`_i\lambda _{u_i}`$ exists on $`^\varphi =_i^\varphi \mathrm{}^2(G)`$.
###### Proof.
First we pick a $`\sigma F`$-sequence $`(F_i)`$ for $`G`$ and a growing sequence $`(H_i)`$ of non-empty finite subsets of $`G`$ satisfying $`_iH_i=G`$. Since the (pointwise) product $`_jv_j`$ exists, we can choose a subsequence $`(u_i)`$ of $`(v_j)`$ satisfying
$$|1u_i(y^1,x)|1/i^2\text{for all}xH_i,yF_i,i.$$
Let $`xG`$ and choose $`N`$ such that $`xH_N`$. Then we get
$$\underset{i}{}\frac{1}{\mathrm{\#}F_i}\underset{yF_i}{}|1u_i(y^1,x)|$$
$$\underset{i<N}{}2+\underset{iN}{}\frac{1}{\mathrm{\#}F_i}\underset{yF_i}{}1/i^2$$
$$=2(N1)+\underset{iN}{}1/i^2<\mathrm{}.$$
This shows that $`(F_i)`$ satisfies $`()`$ in Theorem 3.3, from which the result then clearly follows. โ
###### Corollary 3.5.
Let $`G`$ be countable and amenable. Then there always exist some product sequence $`(u_i)`$ in $`Z^2(G,๐)`$ satisfying $`u_i1`$ for all $`i`$ and some sequence $`\varphi =(\varphi _i)`$ of unit vectors in $`\mathrm{}^2(G)`$ such that $`_i\lambda _{u_i}`$ exists on $`^\varphi =_i^\varphi \mathrm{}^2(G)`$. If $`H^2(G,๐)`$ is non-trivial and $`1[u]H^2(G,๐)`$, then the sequence $`(u_i)`$ above may chosen so that $`u=_iu_i.`$
###### Proof.
We call a product sequence $`(u_i)`$ in $`Z^2(G,๐)`$ 1-free if $`u_i1`$ for all $`i`$. It is easy to see that 1-free product sequences do exist in $`B^2(G,๐)`$. As 1-freeness is clearly preserved when passing to subsequences, the first assertion follows from the previous corollary. The 1-free product sequence $`(u_i)`$ is then in $`B^2(G,๐)`$. Therefore (by closedness) $`_iu_iB^2(G,๐)`$, so we may write it as $`\mathrm{d}\rho `$ for some normalized $`\rho :G๐`$. Assume now $`H^2(G,๐)`$ is non-trivial and $`1[u]H^2(G,๐)`$. Set $`v_1=\overline{d\rho }u`$ and $`v_i=u_{i1},i>1`$. Then $`(v_i)`$ is a 1-free product sequence satisfying $`u=_iv_i.`$ Further we can define a new sequence $`\psi =(\psi _i)`$ of unit vectors in $`\mathrm{}^2(G)`$, by setting $`\psi _1=\delta _e`$ and $`\psi _i=\psi _{i1},i>1.`$ It is then obvious that $`_i\lambda _{v_i}`$ exists on $`^\psi `$, which proves the second assertion. โ
Remarks.
1) It follows from Theorem 3.1 iii) that representations obtained as the infinite tensor product of projective regular representations are never irreducible.
2) Let $`G`$ be countable and amenable, and let $`(u_i)`$ and $`(v_i)`$ be two sequences in $`Z^2(G,๐)`$ satisfying $`v_i_{\rho _i}u_i`$ for every $`i`$. Assume that $`_i\lambda _{u_i}`$ exists on $`^\varphi =_i^\varphi \mathrm{}^2(G)`$ for some sequence $`\varphi =(\varphi _i)`$ of unit vectors in $`\mathrm{}^2(G)`$. As $`_iv_i`$ does not necessarily exist, it may happen that $`_i\lambda _{v_i}`$ can not be formed at all (cf. Theorem 3.1). However, it is quite clear that $`\rho _1\lambda _{v_1}\rho _2\lambda _{v_2}\mathrm{}`$ exists on $`^{\psi _i}\mathrm{}^2(G)`$, where $`\psi _i`$ is defined by $`\psi _i(x)=\overline{\rho _i(x^1)}\varphi _i(x)`$, and this may be considered as a problem of gauge fixing. On the other hand, let us also assume that $`_i\lambda _{v_i}`$ exists on $`^\psi =_i^\psi \mathrm{}^2(G)`$ for some sequence $`\psi =(\psi _i)`$ of unit vectors in $`\mathrm{}^2(G)`$. Then we may conclude that $`_i\lambda _{v_i}`$ is, up to unitary equivalence, just a perturbation of $`_i\lambda _{u_i}`$.
(To prove this, we first appeal to Theorem 3.1 and obtain that both $`u=_iu_i`$ and $`v=_iv_i`$ exist. Using Lemma 2.1 we may then write $`v=\mathrm{d}\rho u`$ for some normalized $`\rho :G๐`$. Now, using that $`\lambda _v\rho \lambda _u`$ and Theorem 3.1, we get
$$_i\lambda _{v_i}\lambda _vid\rho (\lambda _uid)\rho _i\lambda _{u_i},$$
where $`id`$ denotes the identity representation of $`G`$ on any infinite separable Hilbert space.)
3) To produce examples of infinite tensor product of projective unitary representations of not necessarily amenable groups, one can proceed as follows. Let $`G`$ be any countable group possessing a non-trivial amenable factor group $`K`$ (one can here for instance let $`G`$ be any non-perfect, non-amenable group, e. g. any non-abelian countable free group, since the abelianized group $`G/[G,G]`$ is then non-trivial and abelian) and let $`(v_i)`$ be a sequence in $`Z^2(K,๐)`$ such that $`_i\lambda _{v_i}`$ exists on $`_i^\varphi \mathrm{}^2(K)`$. Using the canonical homomorphism $`\pi :GK`$, we may lift each $`v_i`$ to a $`u_iZ^2(G,๐)`$ in the obvious way. Set $`U_i(x):=\lambda _{v_i}(\pi (x)),xG,`$ for each $`i`$. It is then a simple matter to check that each $`U_i`$ is a projective unitary representation of $`G`$ on $`\mathrm{}^2(K)`$ associated to $`u_i`$, and that $`_iU_i`$ exists on $`_i^\varphi \mathrm{}^2(K)`$.
We now turn to another class of examples which is in spirit related to the setting of the Stone-Mackey-von Neumann theorem, i. e. with so-called CCR-representations of a locally compact abelian group and its dual (cf. ).
Let $`A`$ and $`B`$ be two discrete groups and $`\sigma :A\times B๐`$ be a bilinear map. We call a triple $`\{V,W,\}`$ for a *CCR-representation* of $`\sigma `$ whenever $`V`$ and $`W`$ are unitary representations of respectively $`A`$ and $`B`$ on $``$ which satisfy the CCR-relation
$$V(a)W(b)=\sigma (a,b)W(b)V(a)$$
for all $`aA,bB.`$
We now set $`G=A\times B`$ and define $`u_\sigma :G\times G๐`$ by
$$u_\sigma ((a_1,b_1),(a_2,b_2))=\overline{\sigma (a_2,b_1)}.$$
It is an easy exercise to check that $`u_\sigma `$ is a 2-cocycle on $`G`$ (in fact a bicharacter, i. e. a bilinear map on $`G\times G`$ into $`๐`$). When both $`A`$ and $`B`$ are abelian, then $`[u_\sigma ]1`$ in $`H^2(G,๐)`$ whenever $`\sigma `$ is non-trivial, as follows from since $`u_\sigma `$ is then clearly non-symmetric. Note that there is an 1-1 correspondence between CCR-representations of $`\sigma `$ and projective unitary representations of $`G`$ associated with $`u_\sigma `$ ( being given by setting $`U(a,b)=V(a)W(b)`$ whenever $`\{V,W,\}`$ is a CCR-representation of $`\sigma `$).
There is a canonical way to produce a CCR-representation of $`\sigma `$ on $`\mathrm{}^2(B)`$, to which we may associate a projective unitary representation $`U_\sigma `$ of $`G`$ on $`\mathrm{}^2(B)`$ associated with $`u_\sigma .`$ We recall this construction (and remark that a similar construction can be done on $`\mathrm{}^2(A)`$ in an analogous way):
For each $`aA,bB`$ we set $`\sigma _a(b)=\sigma (a,b)`$, so the map $`(a\sigma _a)`$ belongs to $`\mathrm{Hom}(A,\widehat{B})`$ where $`\widehat{B}:=\mathrm{Hom}(B,๐)`$. Let then $`V_\sigma (a)`$ denote the multiplication operator by the function $`\sigma _a`$ on $`\mathrm{}^2(B)`$ and $`\lambda _B`$ be the left regular representation of $`B`$ on $`\mathrm{}^2(B)`$. By computation we have
$$V_\sigma (a)\lambda _B(b)=\sigma (a,b)\lambda _B(b)V_\sigma (a)$$
for all $`aA,bB.`$ Hence, the triple $`\{V_\sigma ,\lambda _B,\mathrm{}^2(B)\}`$ is a CCR-representation of $`\sigma `$ and we can put $`U_\sigma (a,b):=V_\sigma (a)\lambda _B(b)`$ for all $`(a,b)G`$.
Assume now that $`(\sigma _i)`$ is a sequence of bilinear maps from $`A\times B`$ into $`๐.`$ The question whether is it possible to form $`_iU_{\sigma _i}`$ on $`_i^\varphi \mathrm{}^2(B)`$ for some sequence $`\varphi =(\varphi _i)`$ of unit vectors in $`\mathrm{}^2(B)`$ is then clearly equivalent to whether it is possible to form the infinite tensor product of the CCR-representations associated with the $`\sigma _i`$โs. In the case of a positive answer, the product $`_iu_{\sigma _i}`$ will exist (as a consequence of Theorem 3.1), so $`_i\sigma _i`$ will then exist too and the infinite tensor product of the CCR-representations associated with the $`\sigma _i`$โs will be a CCR-representation of this product map.
Quite similarly to Theorem 3.2 and Theorem 3.3 we have:
###### Theorem 3.6.
Let $`(\sigma _i)`$ be a sequence of bilinear maps from $`G=A\times B`$ into $`๐.`$ Set $`U_i:=U_{\sigma _i}.`$
i) Assume that $`_iU_i`$ exists on $`_i^\varphi \mathrm{}^2(B)`$ for some sequence $`\varphi =(\varphi _i)`$ of unit vectors in $`\mathrm{}^2(B).`$ Then $`B`$ is countable and amenable.
ii) Assume that $`B`$ is countable and amenable, and that $`(F_i)`$ be a $`\sigma `$F-sequence for $`B`$ satisfying
$$\underset{i}{}\frac{1}{\mathrm{\#}(F_i)}\underset{bF_i}{}|1\sigma _i(a,b)|<\mathrm{}$$
for every $`aA.`$ Set $`\varphi =(\varphi _i)`$ where $`\varphi _i:=\chi _{F_i}/\mathrm{\#}(F_i)^{1/2}.`$
Then $`_iU_i`$ exists on $`_i^\varphi \mathrm{}^2(B).`$
###### Proof.
i) Since $`U_i(e,b)=\lambda _B(b)`$, this follows from (or Theorem 3.2).
ii) Let $`B`$ be countable and amenable, and $`(F_i)`$ be as in ii). Since $`(F_i)`$ is a $`\sigma `$F-sequence for $`B`$ it follows from (or Theorem 3.3) that $`_iU_i(e,b)=_i\lambda _B(b)`$ exists on $`_i^\varphi \mathrm{}^2(B)`$ for every $`bB.`$ The existence of $`_iU_i`$ on $`_i^\varphi \mathrm{}^2(B)`$ reduces then to whether $`_iV_{\sigma _i}`$ exists on $`_i^\varphi \mathrm{}^2(B)`$, i. e. whether
$$\underset{i}{}|1(V_{\sigma _i}(a)\varphi _i,\varphi _i)|=\underset{i}{}|1((\sigma _i)_a\varphi _i,\varphi _i)|<\mathrm{}$$
holds for every $`aA`$. As we have
$$|1((\sigma _i)_a\varphi _i,\varphi _i)|=\frac{1}{\mathrm{\#}(F_i)}|\underset{bF_i}{}(1\sigma _i(a,b))|\frac{1}{\mathrm{\#}(F_i)}\underset{bF_i}{}|1\sigma _i(a,b)|$$
for every $`aA`$, this follows from the assumption on $`(F_i)`$. โ
We leave to the reader to deduce from this theorem the analogous versions of Corollary 3.4 and Corollary 3.5 in this setting.
## 4 The case of free abelian groups
The purpose of this section is to examplify the results of the previous section in the concrete case where G is a finitely generated free abelian group.
We let $`N`$ and set $`G=^N`$.
When $`x=(x_1,\mathrm{},x_N)G`$, we set $`|x|_1=_{j=1}^N|x_j|.`$
When $`m`$, we define $`K_mG`$ by
$$K_m=\{xG|0x_im,i=1\mathrm{}N\}(=\{0,1,\mathrm{},m\}^N).$$
To each $`N\times N`$ real matrix $`A`$, one may associate $`u_AZ^2(G,๐)`$ by
$$u_A(x,y)=e^{ix(Ay)}.$$
In fact, every element in $`H^2(G,๐)`$ may be written as $`[u_A]`$ for some skew-symmetric $`A`$ (see ). Without loss of generality, we can assume that $`AM_N((\pi ,\pi ])`$, i.e. all of $`A`$โs coefficients belong to $`(\pi ,\pi ]`$. We set
$$|A|_{\mathrm{}}=\mathrm{max}\{|a_{ij}|,1i,jN\}.$$
We first record a technical lemma.
###### Lemma 4.1.
Let $`AM_N((\pi ,\pi ])`$, $`x,yG`$ and $`m`$. Then
* $`|1u_A(x,y)||A|_{\mathrm{}}|x|_1|y|_1`$
* $`_{xK_m}|x|_1=\frac{Nm(m+1)^N}{2}`$
* $`1\frac{\mathrm{\#}((x+K_m)K_m)}{\mathrm{\#}K_m}\frac{|x|_1}{m+1}.`$
###### Proof.
1) follows from $`|1e^{ix(Ay)}||x(Ay)||A|_{\mathrm{}}|x|_1|y|_1`$.
2) $`_{xK_m}|x|_1=_{j=1}^N_{xK_m}|x_j|=N(m+1)^{N1}(_{k=0}^mk)=\frac{Nm(m+1)^N}{2}`$.
3) $`1\frac{\mathrm{\#}((x+K_m)K_m)}{\mathrm{\#}K_m}=\frac{\mathrm{\#}(K_m\backslash (x+K_m))}{\mathrm{\#}K_m}\frac{(m+1)^{N1}}{(m+1)^N}|x|_1=\frac{|x|_1}{m+1}.`$
###### Proposition 4.2.
Let $`(A_i)`$ be a sequence in $`M_N((\pi ,\pi ])`$ and $`(m_i)`$ be a sequence in $``$. For each $`i`$, we set
$$F_i=K_{m_i}G,$$
$$\varphi _i=\frac{1}{(\mathrm{\#}F_i)^{1/2}}\chi _{F_i}\mathrm{}^2(G),$$
$$u_i=u_{A_i}Z^2(G,๐).$$
Then we have:
* $`(F_i)`$ is a $`F`$-sequence for $`G`$ if and only if $`m_i+\mathrm{}`$.
* $`(F_i)`$ is a $`\sigma F`$-sequence for $`G`$ if and only if $`_{i=1}^{\mathrm{}}\frac{1}{m_i}<\mathrm{}`$.
* $`_iu_i`$ exists $``$ $`_i|A_i|_{\mathrm{}}<\mathrm{}`$.
* The projective unitary representation $`_i\lambda _{u_i}`$ of $`G`$ exists on $`_i^{\varphi _i}\mathrm{}^2(G)`$ whenever
$$\underset{i=1}{\overset{\mathrm{}}{}}\frac{1}{m_i}<\mathrm{}\text{and}\underset{i=1}{\overset{\mathrm{}}{}}m_i|A_i|_{\mathrm{}}<\mathrm{}$$
(and $`_iu_i`$ is then the associated 2-cocycle).
###### Proof.
The nontrivial parts of (1) and (2) are consequences of Lemma 4.1, part (3). Assertion (3) relies on the inequality $`2|\theta |/\pi |1e^{i\theta }||\theta |`$ which holds when $`|\theta |\pi `$. Concerning (4) let $`x,yG.`$ Then we have
$`{\displaystyle \frac{1}{\mathrm{\#}F_i}}{\displaystyle \underset{yF_i}{}}|1u_i(y,x)|`$ $`{\displaystyle \frac{1}{(m_i+1)^N}}({\displaystyle \underset{yF_i}{}}|A_i|_{\mathrm{}}|x|_1|y|_1)\text{(by Lemma 4.1, (1))}`$
$`={\displaystyle \frac{|x|_1|A_i|_{\mathrm{}}}{(m_i+1)^N}}{\displaystyle \underset{yF_i}{}}|y|_1`$
$`={\displaystyle \frac{|x|_1|A_i|_{\mathrm{}}}{(m_i+1)^N}}{\displaystyle \frac{Nm_i(m_i+1)^N}{2}}\text{(by Lemma 4.1, (2))}`$
$`={\displaystyle \frac{N|x|_1}{2}}m_i|A_i|_{\mathrm{}}`$
for every $`i`$. Hence we have
$$\underset{i}{}\frac{1}{\mathrm{\#}F_i}\underset{yF_i}{}|1u_i(y,x)|\frac{N|x|_1}{2}\underset{i}{}m_i|A_i|_{\mathrm{}}.$$
Now if we assume that $`_{i=1}^{\mathrm{}}\frac{1}{m_i}<\mathrm{}`$ and $`_{i=1}^{\mathrm{}}m_i|A_i|_{\mathrm{}}<\mathrm{}`$, then $`\{F_i\}`$ is a $`\sigma F`$-sequence for $`G`$ (by (2)) and $`_i\frac{1}{\mathrm{\#}F_i}_{yF_i}|1u_i(y,x)|<\mathrm{}`$ for all $`xG`$, and the conclusion follows from Theorem 3.3. โ
Example. Let $`AM_N((\pi ,\pi ]).`$ Set $`A_i=2^iA`$ and $`u_i=u_{A_i}`$ ($`i`$). Then clearly $`u_A=_iu_i`$. Further, if we let $`m_i=i^2`$, then $`_i1/m_i<\mathrm{}`$ and $`_im_i|A_i|_{\mathrm{}}=|A|_{\mathrm{}}_ii^2/2^i<\mathrm{}`$ so (4) in the above proposition applies. Theorem 3.1 then gives
$$\lambda _{u_A}id_i\lambda _{u_i},$$
thus producing an infinite tensor product decomposition of the amplification of $`\lambda _{u_A}`$. It is well known that the C\*-algebra $`C^{}(\lambda _{u_A})`$ generated by $`\lambda _{u_A}`$ on $`\mathrm{}^2(G)`$ is a so-called non-commutative $`N`$-torus. Using this decomposition result, we can clearly obtain a faithful representation of $`C^{}(\lambda _{u_A})`$ onto the $`C^{}`$-algebra generated by $`_i\lambda _{u_i}`$ on $`_i^\varphi \mathrm{}^2(G)`$ for some suitably chosen sequence $`\varphi `$ of unit vectors in $`\mathrm{}^2(G)`$.
We shall now exhibit projective unitary representations arising from CCR-representations of bilinear maps on some direct product decomposition of $`G`$.
We assume from now on that $`N2`$ and write $`G=^N^P\times ^Q`$ where $`1P,Q<N`$ and $`P+Q=N`$.
To each $`P\times Q`$ matrix $`D`$ with coefficients in $`(\pi ,\pi ]`$, we associate a bilinear map $`\sigma _D:^P\times ^Q๐`$ by
$$\sigma _D(a,b)=e^{ia(Db)}.$$
Using the construction described at the end of the previous section, we then obtain a CCR-representation of $`\sigma _D`$ on $`\mathrm{}^2(^Q)`$, or, equivalently, a projective unitary representation $`U_D`$ of $`G=^N`$ with associated 2-cocycle $`u^D`$. This cocycle is easy to describe: a simple computation gives
$$u^D(x,y)=e^{ix(\stackrel{~}{D}y)}(x,yG)$$
where $`\stackrel{~}{D}`$ is the $`N\times N`$ matrix given by
$$\stackrel{~}{D}=\left(\begin{array}{cc}0& 0\\ D^t& 0\end{array}\right).$$
Notice that $`u^D=u_{\stackrel{~}{D}}`$ and $`[u^D]`$ is non-trivial whenever $`D0`$.
###### Proposition 4.3.
Let $`(D_i)`$ be a sequence of $`P\times Q`$ matrices with coefficients in $`(\pi ,\pi ]`$, and let $`(U_i)=(U_{D_i})`$ be the associated sequence of projective unitary representations of $`G`$ on $`\mathrm{}^2(^Q)`$. Let $`(n_i)`$ be a sequence in $``$.
Set $`H_i=\{b^Q|0b_in_i,i=1\mathrm{}Q\}`$ and $`\psi _i=1/(\mathrm{\#}H_i)^{1/2}\chi _{H_i}(i)`$.
Then $`_iU_i`$ exists on $`_i^{\psi _i}\mathrm{}^2(^Q)`$ whenever $`_i1/n_i<\mathrm{}`$ and $`_in_i|D_i|_{\mathrm{}}<\mathrm{}.`$
###### Proof.
This follows from Theorem 3.6. As the details are quite similar to the proof of the previous proposition, we leave these to the reader. โ
Example. We take $`P=Q=1`$ so that $`G=\times =^2`$, and let $`(D_j)=(\theta _j)`$ be a sequence in $`(\pi ,\pi ]`$. This gives rise to the sequence $`(U_j)`$ of representations of $`^2`$ on $`\mathrm{}^2()`$ with associated 2-cocycles
$$u_j(x,y)=e^{i\theta _jx_1y_2}(x,y^2).$$
By Proposition 4.3 we can then form the infinite tensor representation $`_jU_j`$ whenever we can choose a sequence $`(n_j)`$ in $``$ such that $`_j1/n_j<\mathrm{}`$ and $`_jn_j|\theta _j|<\mathrm{}`$ (e.g. $`n_j=j^2`$ will do if $`(j^4|\theta _j|)`$ is bounded).
By a more careful analysis of this example involving the familiar Dirichlet sums, one can deduce that $`_jU_j`$ will exist whenever we can choose $`(n_j)`$ such that
$$\underset{j}{}\frac{1}{n_j}<\mathrm{}\text{and}\underset{j}{}|1\frac{1}{2n_j+1}\frac{\mathrm{sin}((2n_j+1)\theta _j/2)}{\mathrm{sin}(\theta _j/2)}|<\mathrm{}.$$
Assuming that $`_j|\theta _j|<\mathrm{}`$ (so $`_ju_j`$ exists), it would be interesting to know whether such a choice of $`(n_j)`$ can always be made.
## 5 Infinite products of actions
For each $`i`$ let $`_i`$ be a Hilbert space, $`\varphi _i_i`$ be a unit vector, $`_i(_i)`$ be a von Neumann algebra and $`\alpha _i:G\mathrm{Aut}(_i)`$ be an action of $`G`$ on $`_i`$. We denote by $`I_i`$ the identity operator on $`_i`$. We then form the $``$-algebra $`_i_i`$ (resp. von Neumann algebra $`_i(_i,\varphi _i)`$) acting on $`_i^{(\varphi _i)}_i`$ generated by operators of the form $`_iT_i`$ where $`T_i_i`$ and $`T_i=I_i`$ for all but finitely many $`i`$โs. At the $``$-algebraic level we define an action $`_i\alpha _i`$ of $`G`$ on $`_i_i`$ such that for every finite $`J`$ we have
$$_i\alpha _i((_{iJ}T_i)(_{iJ}I_i))=(_{iJ}\alpha _i(T_i))(_{iJ}I_i).$$
One natural question is whether $`_i\alpha _i`$ may be extended to an action of $`G`$ on the von Neumann algebra $`_i(_i,\varphi _i)`$. As we shall see, the answer may be negative in some situations, regardless of the choice of unit vectors $`\varphi _i`$.
We retrict ourselves to the case where each $`\alpha _i`$ is unitarily implemented, i. e. we assume that for every $`i`$ and $`g`$ there exists a unitary $`U_i(g)`$ on $`_i`$ such that $`\alpha _{i,g}=\mathrm{Ad}(U_i(g)).`$ This assumption is automatically satisfied for many classes of von Neumann algebras (see , ยง8). Note that if $`U_i(g)_i`$ for all $`gG`$ and $`M_i`$ is a factor, especially if $`_i=(_i),`$ then $`gU_i(g)`$ is a projective unitary representation of $`G`$ on $`_i.`$
We consider the following condition:
$$()\underset{i}{}(1|(U_i(g)\varphi _i,\varphi _i)|)<\mathrm{}\text{for all}gG.$$
###### Proposition 5.1.
Condition $`()`$ is equivalent to the following condition:
$$()\rho _i:G๐,\rho _i(e)=1,\text{such that}_i\rho _iU_i\text{exists on}_i^{\varphi _i}_i.$$
When $`()`$ holds, then $`_i\alpha _i`$ extends to a unitarily implemented action $`\alpha `$ on $`_i(_i,\varphi _i),`$ which is inner whenever $`U_i(g)_i`$ for every $`i`$ and $`gG.`$
###### Proof.
The first assertion follows from Proposition 2.3, using \[11, ยง1.2\]. When $`()`$ holds, then $`\alpha _g=\mathrm{Ad}(U(g))`$ where $`U(g)=_i\rho _i(g)U_i(g)`$ is well defined on $`_i^{\varphi _i}_i.`$ Clearly $`U(g)_i(_i,\varphi _i)`$ whenever $`U_i(g)_i`$ for every $`i`$ and $`gG,`$ and $`\alpha _g`$ is then inner for every $`gG`$. โ
We now treat the case where every $`_i`$ is a type $`I`$ factor. We use the well known fact that every automorphism of a type $`I`$ factor is inner and also that $`_i((_i),\varphi _i)=(_i^{\varphi _i}_i)`$ ( \[11, Proposition 1.6\]).
###### Theorem 5.2.
Assume that $`_i=(H_i)`$ for all $`i.`$ Then $`_i\alpha _i`$ extends (uniquely) to an action $`\alpha =\alpha _i`$ on $`_i((_i),\varphi _i)`$ if and only if condition $`()`$ holds.
###### Proof.
Assume that an extension $`\alpha `$ of $`_i\alpha _i`$ exists on $`^\varphi =_i(_i,\varphi _i).`$ Using the facts recalled above, we have $`\alpha _g=\mathrm{Ad}(U(g))`$ for some $`U(g)๐ฐ(_i^{\varphi _i}_i)`$ for every $`gG`$.
Let $`J`$ be a non-empty finite subset of $``$.
We identify $`^\varphi `$ with $`(_{iJ}_i){}_{J}{}^{}`$ where $`{}_{J}{}^{}:=_{iJ}(_i,\varphi _i)`$, and consider $`{}_{J}{}^{}`$ as a von Neumann subalgebra of $`^\varphi `$ in the obvious way. It is easy to see that $`\alpha `$ restricts to an action $`{}_{J}{}^{}\alpha `$ of $`G`$ on $`{}_{J}{}^{}`$ such that $`\alpha =(_{iJ}\alpha _i){}_{J}{}^{}\alpha `$. Since$`{}_{J}{}^{}`$ is a also type $`I`$ factor, we can write $`{}_{J}{}^{}\alpha _{g}^{}=\mathrm{Ad}({}_{J}{}^{}U(g))`$ for some $`{}_{J}{}^{}U(g)๐ฐ(_{iJ}^{\varphi _i}_i)`$ for each $`gG.`$
Set now $`U_J(g)=_{iJ}U_i(g)`$ for each $`gG`$. Then $`\alpha _g=\mathrm{Ad}(U_J(g){}_{J}{}^{}U(g))`$. Therefore, for each $`gG`$, there exists some $`z_J(g)๐`$ such that $`U(g)=z_J(g)U_J(g){}_{J}{}^{}U(g)`$.
Let $`gG.`$ Since $`U(g)0`$ we can pick two elementary decomposable vectors $`\psi _i`$ and $`\xi _i`$ in $`_i^{\varphi _i}_i`$ (which do not depend on $`J`$) satisfying
$$0c(g):=|(U(g)\psi _i,\xi _i)|=\underset{iJ}{}|(U_i(g)\psi _i,\xi _i)||({}_{J}{}^{}U(g)_{iJ}\psi _i,_{iJ}\xi _i)|$$
Since $`|({}_{J}{}^{}U(g)_{iJ}\psi _i,_{iJ}\xi _i)|1`$ we get
$$0<c(g)\underset{iJ}{}|(U_i(g)\psi _i,\xi _i)|.$$
As this holds for every $`J`$, one easily deduces that $`_i|(U_i(g)\psi _i,\xi _i)|`$ converges to a non-zero number. Since $`\psi _i=\xi _i=\varphi _i`$ for all but finitely many $`i`$โs, this implies that $`()`$ holds. Hence, we have shown the only if part of the assertion. The converse part follows from Proposition 5.1. โ
The proof of the above result is reminiscent of the proof of a lemma in (see also ). In the same line of ideas, we have the following result, which is related to \[6, Lemme 1.3.8\].
###### Theorem 5.3.
Assume that all $`_i`$โs are factors and that $`_i\alpha _i`$ extends to an action $`\alpha `$ on $`^\varphi =_i(_i,\varphi _i).`$ Then $`\alpha `$ is inner if and only if there exists for each $`gG`$ and each $`i`$ a unitary $`v_i(g)_i`$ implementing $`\alpha _{i,g}`$ such that the following condition holds:
$$(1)\underset{i}{}(1|(v_i(g)\varphi _i,\varphi _i)|)<\mathrm{}\text{for all}gG.$$
On the other hand, $`\alpha `$ is outer if and only if, for each $`gG,ge,`$ at least one of the $`\alpha _{i,g}`$ is outer or there exists for each $`i`$ a unitary $`v_i(g)_i`$ implementing $`\alpha _{i,g}`$ such that
$$(2)\underset{i}{}(1|(v_i(g)\varphi _i,\varphi _i)|)=\mathrm{}.$$
###### Proof.
Assume first that $`\alpha `$ is inner. So we have $`\alpha _g=\mathrm{Ad}(U(g))`$ for some unitary $`U(g)^\varphi `$ for every $`gG`$. Recall from that $`^\varphi `$ is a factor. Using \[14, Corollary 1.14\], it follows easily that each $`\alpha _i`$ is inner. Hence, there exists for each $`gG`$ and each $`i`$ a unitary $`v_i(g)_i`$ implementing $`\alpha _{i,g}.`$
Let $`J`$ be a non-empty finite subset of $``$. As in the previous proof, we identify $`^\varphi `$ with $`(_{iJ}_i){}_{J}{}^{}`$ where $`{}_{J}{}^{}:=_{iJ}(_i,\varphi _i)`$, We set $`V_J(g)=_{iJ}v_i(g)`$ for each $`gG`$ and $`W_J(g)=(V_J(g)(_{iJ}I_i))^{}U(g).`$ Then, using that we may write $`\alpha =(_{iJ}\alpha _i){}_{J}{}^{}\alpha `$, we get
$$W_J(g)(_i(_i,\varphi _i))((_{iJ}_i)(_{iJ}I_i))^{}.$$
Using that all $`_i`$ are factors, it is a simple exercise to deduce that $`W_J(g)(_{iJ}I_i)(_{iJ}(_i,\varphi _i))`$. We may therefore write $`W_J(g)=(_{iJ}I_i){}_{J}{}^{}V(g)`$ for some unitary $`{}_{J}{}^{}V(g)_{iJ}(_i,\varphi _i)`$. This gives $`U(g)=V_J(g){}_{J}{}^{}V(g)`$ and we can clearly proceed further in the same way as in the previous proof to show that $`(1)`$ holds, thereby proving the only if part of the first assertion. The converse part of this assertion follows from Proposition 5.1. The second assertion follows from a similar argument. โ
The following corollary may be seen as generalization of \[10, Theorem 6.7\].
###### Corollary 5.4.
Assume for each $`i`$ that $`\beta _i`$ is an action of $`G`$ on some von Neumann algebra $`๐ฉ_i`$ and that there exists a normal $`\beta _i`$-invariant state $`\tau _i`$ on $`๐ฉ_i.`$ Denote the GNS-triple of $`\tau _i`$ by $`(\pi _i,_i,\xi _i)`$ and set $`_i=\pi _i(๐ฉ_i).`$ Let $`\alpha _i`$ be the action of $`G`$ on $`_i`$ induced by $`\beta _i.`$ Then $`_i\alpha _i`$ extends to an action $`\alpha `$ of $`G`$ on $`_i(_i,\xi _i).`$
Assume further that all $`๐ฉ_i`$โs are factors and all $`\pi _i`$โs are faithful. Then $`\alpha `$ is inner if and only if there exists for each $`gG`$ and each $`i`$ a unitary $`v_i(g)๐ฉ_i`$ implementing $`\beta _{i,g}`$ such that the following condition holds:
$$(1)\underset{i}{}(1|\tau _i(v_i(g))|)<\mathrm{}\text{for all}gG.$$
On the other hand, $`\alpha `$ is outer if and only if, for each $`gG,ge,`$ at least one of the $`\beta _{i,g}`$ is outer or there exists each $`i`$ a unitary $`v_i(g)๐ฉ_i`$ implementing $`\beta _{i,g}`$ such that
$$(2)\underset{i}{}(1|\tau _i(v_i(g))|)=\mathrm{}.$$
###### Proof.
We first recall that there exists for each $`i`$ a unitary representation $`V_i:G(_i)`$ such that
$$\pi _i(\beta _{i,g}(x))=V_i(g)\pi _i(x)V_i(g)^{}\text{and}V_i(g)\pi _i(x)\xi _i=\pi _i(\beta _{i,g}(x))\xi _i$$
for all $`gG,x๐ฉ_i`$ (see ). The induced action $`\alpha _i`$ on $`_i`$ is then defined by $`\alpha _{i,g}(\pi _i(x))=\pi _i(\beta _{i,g}(x)).`$ As $`V_i(g)\xi _i=\xi _i`$ for all $`gG`$, the first assertion follows obviously from Proposition 3.1. The second assertion is then easily deduced from Theorem 5.3. โ
Example. Let $`u_i`$ be a sequence in $`Z^2(G,๐).`$ Set $`๐ฉ_i=\lambda _{u_i}(G)^{^{\prime \prime }}(\mathrm{}^2(G))`$ and let $`\beta _{i,g}`$ be the inner automorphism of $`๐ฉ_i`$ implemented by $`\lambda _{u_i}(g)`$ for all $`gG,i.`$ Let $`\tau _i`$ denote the canonical normal faithful tracial state of $`๐ฉ_i`$ (determined by $`\tau _i(\lambda _{u_i}(g))=1`$ if $`g=e`$ and $`0`$ otherwise), which is trivially $`\beta _i`$-invariant. If $`\xi `$ denote the normalized delta-function at $`e`$, then $`\tau _i=\omega _\xi |_{๐ฉ_i}.`$ So we may identify the GNS-triple of $`\tau _i`$ with $`(id_i,\mathrm{}^2(G),\xi _i)`$, where $`id_i`$ denotes the identity representation of $`๐ฉ_i`$ and $`\xi _i=\xi ,`$ i. e. we may take $`_i=๐ฉ_i`$ and $`\alpha _i=\beta _i`$ in the notation of Corollary 5.4. Hence, $`\alpha _i=\beta _i`$ extends to an action $`\alpha `$ on $`_i(\lambda _{u_i}(G)^{^{\prime \prime }},\xi _i).`$
Further, if all $`\lambda _{u_i}(G)^{^{\prime \prime }}`$ are factors, then $`\alpha `$ is outer, as
$$\underset{i}{}(1|\tau _i(\lambda _{u_i}(g))|)=\underset{i}{}1=\mathrm{}\text{for all}ge.$$
A necessary and sufficient condition for a twisted group von Neumann algebra $`\lambda _u(G)^{^{\prime \prime }}`$ to be a factor may be found in .
If we replace each $`๐ฉ_i`$ with $`(\mathrm{}^2(G))`$ in this example, the extended product action may be formed in many cases under the assumption that $`G`$ is countable and amenable, as follows from Teorem 3.3 and Proposition 5.1. This requires a suitable choice of unit vectors $`\varphi _i`$ in $`\mathrm{}^2(G).`$ This product action restricts then to an action on $`_i(\lambda _{u_i}(G)^{^{\prime \prime }},\varphi _i)`$ which is inner, in contrast to the factor case above. When $`G`$ is either uncountable or non-amenable, we have the following:
###### Theorem 5.5.
Let $`u_i`$ be a sequence in $`Z^2(G,๐)`$ and $`\alpha _i=\mathrm{Ad}\lambda _{u_i}`$ be the associated sequence of actions of $`G`$ on $`(\mathrm{}^2(G)).`$ If $`G`$ is either uncountable or non-amenable, then $`_i\alpha _i`$ does not extend to an action of $`G`$ on $`_i((\mathrm{}^2(G)),\varphi _i),`$ regardless of the choice of vectors $`\varphi _i`$.
###### Proof.
According to Proposition 5.1 and Theorem 5.2, the existence of such an extension $`_i((\mathrm{}^2(G)),\varphi _i)`$ would imply the existence of $`_i\rho _i\lambda _{u_i}`$ on $`_i^{\varphi _i}\mathrm{}^2(G)`$ for some choice of functions $`\rho _i:G๐`$ with $`\rho _i(e)=1`$. It is straightforward to see that this amounts to the existence of $`_i\lambda _{v_i}`$ on $`_i^{\psi _i}\mathrm{}^2(G)`$ for some $`v_iZ^2(G,๐)`$ with $`v_iu_i`$ and some sequence $`\psi _i`$ of unit vectors in $`\mathrm{}^2(G).`$ This is impossible if $`G`$ is either uncountable or non-amenable, as follows from Theorem 3.2. โ
Another type of possible obstruction for extending a product action from the $``$-algebraic level to the von Neumann algebra level is of cohomological nature, as we now illustrate:
###### Theorem 5.6.
Let $`\alpha _i`$ be a sequence of actions of $`G`$ on $`(_i)`$ and write each $`\alpha _i`$ as $`\mathrm{Ad}U_i(g)`$ where $`U_i`$ is a projective representation of $`G`$ with associated 2-cocycle $`u_i.`$ Assume that $`[u_i]=[u]`$ for every $`i`$ and $`[u][1]`$ in $`H^2(G,๐)`$.
Then $`_i\alpha _i`$ does not extend to an action of $`G`$ on $`_i((_i),\varphi _i),`$ regardless of the choice of vectors $`\varphi _i.`$
###### Proof.
Assume that such an extension exists $`_i((_i),\varphi _i).`$ Using Proposition 5.1 and Theorem 5.2, we deduce that $`\rho _iU_i`$ exists on $`_i^{\varphi _i}_i`$ for some choice of functions $`\rho _i:G๐`$ with $`\rho _i(e)=1`$. It follows then from Theorem 3.1 that $`_i(\mathrm{d}\rho _i)u_i`$ exists. Hence $`\mathrm{d}\rho _iu_i1`$ (in the pointwise topology). As each $`u_i=(\mathrm{d}\rho _i^{})u`$ for some $`\rho _i^{}`$, we get that $`u`$ is a limit of 2-coboundaries. Since $`B^2(G,๐)`$ is closed, this means that $`u`$ is itself a coboundary, i. e. $`[u]=1`$, which gives a contradiction. โ
Example. The simplest case where the above situation occurs is when $`G=_2\times _2`$. Indeed, let
$$V=\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right),W=\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right).$$
A projective unitary representation of $`G=_2\times _2`$ on $`^2`$ is then obtained by setting $`U((a,b))=V^aW^b(a,b_2)`$. Since $`V^aW^b=\sigma (a,b)W^bU^a`$ where $`\sigma (a,b)=1`$ if $`a=b=1`$ and $`1`$ otherwise, the associated cocycle $`u`$ is easily computed to be $`u((a_1,b_1),(a_2,b_2))=(1)^{a_2b_1}.`$ It is not difficult to check that $`[u]1.`$ Remark that $`U`$ is nothing but the projective representation associated to the CCR representation of $`\sigma `$ on $`^2=\mathrm{}^2(_2)`$ determined by $`V`$ and $`W.`$
For each $`i`$ consider the action $`\alpha _i`$ of $`G`$ on $`M_2()`$ given by $`\alpha _{i,(a,b)}=\mathrm{Ad}(U((a,b)))`$. Then, according to Theorem 5.6, the infinite tensor product of the $`\alpha _i`$ โs does never make sense as an action on $`_i(M_2(),\varphi _i)`$.
On the other hand, the canonical tracial state of $`M_2()`$ is trivially $`\alpha _i`$-invariant. Therefore we may use Corollary 5.4 to form the infinite tensor product action after passing to the GNS-representation with respect to this tracial state for each $`i.`$ As another application of Corollary 5.4, the resulting product action is easily seen to be outer.
Acknowledgements. Part of this work was made while R. C. was visiting the Department of Mathematics at the University of Oslo during the academic year 1998-1999 supported by the EU TMR network โNon Commutative Geometryโ. He is very grateful to the members of the Operator Algebra group in Oslo for their friendly hospitality and for providing perfect working conditions. Both authors would like to thank Ola Bratteli and Erling Stรธrmer for some helpful discussions.
Addresses of the authors:
Erik Bรฉdos, Institute of Mathematics, University of Oslo,
P.B. 1053 Blindern, 0316 Oslo, Norway. E-mail: bedos@math.uio.no.
Roberto Conti, Dipartimento di Matematica, Universitร di Roma
โTor Vergataโ, I-00133 Roma, Italy. E-mail: conti@mat.uniroma2.it. |
warning/0001/math0001096.html | ar5iv | text | # An Algebraic Duality Theory for Multiplicative Unitaries
## 1 Introduction
An interesting open problem in the duality theory of tensor $`C^{}`$โcategories is to decide which ones admit a (tensorโpreserving) embedding into the tensor $`C^{}`$-category of Hilbert spaces. A first positive result in this direction is the duality theorem of asserting that symmetric tensor $`C^{}`$โcategories with conjugates and irreducible tensor unit admit such an embedding. On the other hand, the theory of dimension introduced in allows one to see that certain tensor $`C^{}`$โcategories with conjugates and irreducible tensor unit cannot be embedded. For such tensor $`C^{}`$โcategories admit an intrinsic dimension function defined on objects. Under certain circumstances, the embedding functor must preserve dimensions. This is well known to be so in the rational case, i.e. when the set of equivalence classes of irreducibles is finite, since it is a simple consequence of the PerronโFrobenius Theorem. However, it is also true in the amenable case. We are in the amenable case whenever the category admits a unitary braiding, see Theorem 5.31 of . This means that the tensor $`C^{}`$โcategories with conjugates and a braiding appearing in low dimensional quantum field theory cannot be embedded whenever the dimensions are nonโintegral. On the other hand, the finite dimensional unitary representation theory of compact quantum groups provides us with examples of tensor $`C^{}`$โcategories with conjugates and irreducible tensor units with nonโintegral dimensions which are embeddable by construction. The duality theorem of Woronowicz generalizing TannakaโKrein to compact quantum groups provides a way of recognizing such categories if the embedding is given.
We will not treat the embedding problem here in full generality; we shall instead present a positive solution for abstract tensor $`C^{}`$โcategories with specific additional structure. Such structure is present in a class of model tensor $`C^{}`$โcategories namely the minimal tensor $`C^{}`$โcategories generated by the regular representation of locally compact groups or of multiplicative unitaries in the sense of Baaj and Skandalis . We prove that abstract tensor $`C^{}`$โcategories with this additional structure are isomorphic to a model tensor $`C^{}`$โcategory and are hence embeddable.
Our result, Thm. 6.13, is thus a duality theorem for multiplicative unitaries and is hence applicable to the case of locally compact quantum groups . Multiplicative unitaries have already played a role in Tatsuumaโs duality theorem for locally compact groups where the group elements are identified in the regular representation using the multiplicative unitary and in Takesakiโs Hopf von Neumann algebra version of duality . Multiplicative unitaries express the fundamental property of the regular representation $`V`$, namely that $`V\times V`$ is equivalent to a multiple of $`V`$. This can also be expressed through the existence of a remarkable Hilbert space $`H`$ of intertwiners from $`V`$ to $`V\times V`$ (cf also ). Our minimal model is the smallest tensor $`C^{}`$โcategory containing the regular representation as an object and $`H`$ as a subspace of arrows.
A closely related result, Thm. 6.11, characterizes a class of $`C^{}`$โalgebras $`๐`$ acted on by an endomorphism $`\rho `$ encoding the regular representation. Intertwiners between powers of this endomorphism encode intertwiners between tensor powers of the regular representation. The minimal model here is obtained as follows: take the regular representation $`V`$ considered as an object in the tensor $`C^{}`$โcategory of representations of the multiplicative unitary and associate to it as in the $`C^{}`$โalgebra $`๐ช_V`$ and its endomorphism $`\rho _V`$. Then the minimal model is the smallest $`\rho _V`$โstable $`C^{}`$โsubalgebra containing $`H`$ acted on by the restriction of $`\rho _V`$. The $`C^{}`$โalgebra obtained in this way is simple and is separable if and only if the given multiplicative unitary acts on a separable Hilbert space. Further variants of these duality results are Theorems 6.5, 6.6, 6.7 and 6.12.
In connection with the above results, attention should be drawn to Longoโs characterization of actions of finite dimensional Hopf algebras which has a similar algebraic and categorical flavour. However, his axiomatic structure involving $`Q`$โsystems is quite distinct from those used here. Despite this, the Hilbert space $`H`$ puts in an appearance here too and a multiplicative unitary appears in his proof.
The principal results in the remaining sections may be summarized as follows. Section 2 gathers together some elementary results on categories of Hilbert spaces. The discussion centres round the concept of shift, a category of Hilbert spaces with objects labelled by the integers, $`0`$ being irreducible, and equipped with a normal โfunctor adding one on objects. Such a structure is isomorphic to the category of tensor powers of a given Hilbert space $`K`$ with the functor being tensoring on the right by $`1_K`$.
In Section 3, it is shown how two commuting shifts of relative depth two are equivalent to giving a multiplicative unitary for the tensor product defined by one of the commuting shifts. Duality for multiplicative unitaries reflects the symmetry between the commuting shifts. The representation category for a multiplicative unitary finds a natural expression within this framework.
In Section 4, it is shown how two commuting shifts of relative depth two and hence multiplicative unitaries arise naturally in terms of ambidextrous Hilbert spaces in a tensor $`W^{}`$โcategory.
Section 5 is devoted to studying Hilbert spaces in and endomorphism of the Cuntz algebra which are algebraic with respect to the natural grading. We show, for example, how the problem of determining intertwiners between algebraic endomorphisms can be reduced to a purely algebraic problem. These results are used in the final section to arrive at the duality results already announced earlier in the introduction. The paper concludes with an appendix on braided symmetry.
In this paper we prefer to work with strictly associative tensor products and a simple way of achieving this is to use as the underlying Hilbert spaces the Hilbert spaces in some fixed von Neumann algebra since these are objects in a strict tensor $`W^{}`$โcategory. We will be concerned here with the representation categories of multiplicative unitaries and recall the basic definitions from . If $`K`$ is such a Hilbert space then a unitary $`V`$ on the tensor square $`K^2`$ is said to be multiplicative if
$$V_{12}V_{13}V_{23}=V_{23}V_{12},$$
where we use the usual convention regarding indices and tensor products. A representation of $`V`$ on a Hilbert space $`H`$ is a unitary $`W(HK,HK)`$ such that
$$W_{12}W_{13}V_{23}=V_{23}W_{12},\text{on}HK^2.$$
If $`W`$ and $`W^{}`$ are representations of $`V`$ on $`H`$ and $`H^{}`$ respectively, we say that $`T(H,H^{})`$ intertwines $`W`$ and $`W^{}`$ and write $`T(W,W^{})`$ if $`T\times 1_KW=W^{}T\times 1_K`$. We define the tensor product of $`W`$ and $`W^{}`$ to be the representation $`W\times W^{}`$ on $`HH^{}`$ given by $`W\times W^{}:=W_{13}W_{23}^{}`$. The usual tensor product of intertwiners is again an intertwiner and in this way we get a strict tensor $`W^{}`$โcategory $`(V)`$ of representations of $`V`$. In fact this assertion does not depend on $`V`$ being multiplicative. When it is then $`V`$ itself is a representation of $`V`$ called the regular representation.
A corepresentation of $`V`$ on $`H`$ is a unitary $`W(KH,KH)`$ such that
$$V_{12}W_{13}W_{23}=W_{23}V_{12}\text{on}K^2H.$$
If $`W`$ and $`W^{}`$ are corepresentations on $`H`$ and $`H^{}`$ respectively, we say that $`T(H,H^{})`$ intertwines $`W`$ and $`W^{}`$ and write $`T(W,W^{})`$ if $`1_K\times TW=W^{}1_K\times T`$. The tensor product $`W\times W^{}`$ of corepresentations is defined by $`W\times W^{}:=W_{12}W_{13}^{}`$. Just as in the case of representations we get a strict tensor $`W^{}`$โcategory now denoted by $`๐(V)`$. If $`\vartheta =\vartheta _{K,K}`$ denotes the flip on $`K^2`$ then $`\vartheta V^{}\vartheta `$ is again a multiplicative unitary and the mapping $`W\stackrel{~}{W}:=\vartheta _{H,K}W^{}\vartheta _{K,H}`$ defines a 1โ1 correspondence between representations of $`V`$ and corepresentations of $`\vartheta V^{}\vartheta `$. However, it does not define an isomorphism of tensor $`W^{}`$โcategories since $`W\times W^{}\stackrel{~}{W}_{13}^{}\stackrel{~}{W}_{12}`$ and so leads to an alternative definition of the tensor product of corepresentations. In fact the two expressions for the tensor product will be equal if and only if $`\vartheta _{W,W^{}}(W\times W^{},W^{}\times W)`$, and this corresponds to the case of a group cf. , Prop. 2.5. Thus exchanging the definitions of tensor product corresponds to exchanging representations of a multiplicative unitary and corepresentations of the dual multiplicative unitary.
## 2 Preliminaries on Categories of Hilbert Spaces
We begin our considerations with a simple but useful lemma on natural transformations in the context of $`W^{}`$โcategories. Elementary results and definitions on $`W^{}`$โcategories can be found in and Lemma 2.1 below is just a slight generalization of Corollary 7.4 in . The notion of direct sum will be used in the Hilbert space sense rather than in the purely algebraic sense. Thus $`A`$ is a direct sum of objects $`B_i`$, $`iI`$, if there are isometries $`W_i(B_i,A)`$ such that $`_iW_iW_i^{}=1_A`$, where the convergence is in, say, the $`s`$โtopology. In particular if the $`B_i=B`$ for all $`i`$, the condition amounts to saying that there is a Hilbert space of support $`1_A`$ in $`(B,A)`$. The isometries $`W_i`$ form an orthonormal basis of such a Hilbert space. An object $`B`$ has central support one or is a generator if given any object $`A`$ there are partial isometries $`W_{i,A}(B,A)`$ such that $`_iW_{i,A}W_{i,A}^{}=1_A`$, see Proposition 7.3 of .
2.1 Lemma Let $`๐ฏ`$ and $`๐ฆ`$ be $`W^{}`$โcategories and $`E`$ and $`F`$ be normal โfunctors from $`๐ฆ`$ to $`๐ฏ`$. Suppose $`๐ฆ`$ has a object $`B`$ of central support one. Then a natural transformation $`t`$ from $`E`$ to $`F`$ has form
$$t_A=\underset{i}{}F(W_{i,A})TE(W_{i,A})^{},$$
where the sum is taken over partial isometries $`W_{i,A}`$ of $`(B,A)`$ with $`_iW_{i,A}W_{i,A}^{}=1_A`$ and $`T`$ is an arbitrary element of $`(E(B),F(B))`$ satisfying the intertwining relation $`TE(S)=F(S)T`$, $`S(B,B)`$. $`t`$ is automatically bounded and $`t_B=T`$.
Proof. The sum defining $`t_A`$ converges in the $`s`$โtopology, say, and $`t_AT`$. (Consider a finite sum and use the $`C^{}`$โproperty of the norm.) Noting that $`t_B=T`$, we conclude that $`t=T`$. Pick $`Y(A,C)`$, then
$$t_CE(Y)=\underset{i,j}{}F(W_{i,C})TE(W_{i,C}^{}YW_{j,A}W_{j,A}^{}).$$
But $`W_{i,B}^{}YW_{j,A}(B,B)`$ so using the intertwining property of $`T`$, we deduce that $`t_CE(Y)=F(Y)t_A`$ and we have a bounded natural transformation. Conversely, suppose $`t(E,F)`$ and $`W(B,A)`$ then $`t_AE(W)=F(W)t_B`$ implying that $`t`$ is obtained by the above construction with $`T=t_B`$.
Note that if $`E(B)=F(B)`$ and $`B`$ is irreducible, then we even have a canonical natural transformation $`t(E,F)`$, satisfying $`t_B=1_{E(B)}`$. We use the notation $`(F,T,E)`$ for the natural transformation constructed as above from $`T(E(B),F(B))`$ satisfying the intertwining relation. The usual operations on natural transformations have simple expressions in this notation. Thus composition of natural transformations corresponds to composing these symbols in the obvious way:
$$(F,T,E)(E,S,D)=(F,TS,D).$$
In fact, the map $`tt_B`$ is a full and faithful โfunctor from the category of normal โfunctors from $`๐ฆ`$ to $`๐ฏ`$ to the category of normal representations of $`(B,B)`$. If $`G`$ is a normal โfunctor from $`๐ฏ`$, then acting on $`(F,T,E)`$ on the left with $`G`$ gives $`(GF,G(T),GE)`$. If $`D`$ is a normal โendofunctor of $`๐ฆ`$ then acting on $`(F,T,E)`$ on the right by $`D`$ gives $`(FD,S,ED)`$, where $`S=(F,T,E)_{D(B)}`$.
2.2 Lemma With the above notation, suppose that $`GE=ED`$ and $`GF=FD`$. Then if $`G(T)=S`$, $`G\times (F,T,E)=(F,T,E)\times D`$. If $`G\times (F,T,E)`$ is invertible, then
$$(F,T,E)\times D=G\times (F,T,E)(ED,G(T)^1S,ED).$$
Proof. As the natural transformations are uniquely determined by their values in $`B`$, the result follows by evaluating in $`B`$.
By a category of Hilbert spaces, we mean a $`W^{}`$โcategory whose objects are Hilbert spaces and whose mappings are all bounded linear mappings between these Hilbert spaces. Any object in such a category has central support one. Any $`W^{}`$โcategory with an irreducible object of central support one is a category of Hilbert spaces in a natural way.
We now consider a category $`๐ฆ`$ of Hilbert spaces whose objects are labelled by $`_0`$ with $`(0,0)=`$ and equipped with a normal โfunctor $`F`$ from $`๐ฆ`$ to $`๐ฆ`$ such that the action of $`F`$ on objects is given by $`F(n)=n+1`$, $`n_0`$.
Since $`(0,0)=`$, $`(0,n)`$ is a Hilbert space of support $`1`$ for each $`n`$ and we let $`\psi _{i,n}`$, $`iI_n`$, be an orthonormal basis of $`(0,n)`$ and set
$$\widehat{F}(X):=\underset{i}{}F^n(\psi _{i,1})XF^m(\psi _{i,1}^{}),X(m,n).$$
Then, with the notation of Lemma 2.1, $`\widehat{F}(X)=(F^n,X,F^m)_1`$ and $`\widehat{F}`$ is another normal -functor with $`\widehat{F}(n)=n+1`$, $`n_0`$.
It is obvious that $`F`$ and $`\widehat{F}`$ commute. If we iterate $`\widehat{F}`$, we find that
$$\widehat{F}^k(X)=\underset{i}{}F^n(\psi _{i,k})XF^m(\psi _{i,k}^{})=(F^n,X,F^m)_k.$$
Since $`\widehat{F}`$ has the properties assumed for $`F`$, we can form $`\widehat{\widehat{F}}`$. Calculating, we find that
$$\widehat{\widehat{F}}(X)=\underset{i,j,k}{}F(\psi _{j,m})\psi _{i,1}\psi _{j,n}^{}X\psi _{k,m}\psi _{i,1}^{}F(\psi _{k,m})^{}$$
$$=\underset{j,k}{}F(\psi _{j,m})F(\psi _{j,n}^{}X\psi _{k,m})F(\psi _{k,m})^{}=F(X).$$
Thus the operation $`\widehat{}`$ is involutive.
From Lemma 2.1, we know that the natural transformations between powers of the functor $`F`$ are automatically bounded, and furthermore, that a natural transformation $`t(F^r,F^s)`$ has the form $`t_n=\widehat{F}^n(T)`$ with $`T(r,s)`$. Finally, we know that $`Tt`$ is an isomorphism of $`W^{}`$โcategories between $`๐ฆ`$ and the category of natural transformations between the powers of $`F`$. The latter category is, however, a tensor $`W^{}`$โcategory, so we may use the isomorphism to equip $`๐ฆ`$ with a tensor product making it into a tensor $`W^{}`$โcategory. We compute this tensor product. If $`y(F^r,F^s)`$ and $`y^{}(F^r^{},F^s^{})`$, then $`y\times y^{}=y\times F^s^{}F^r\times y^{}`$. In other words, $`(y\times y^{})_n=y_{n+s^{}}F^r(y_n^{})=\widehat{F}^{n+s^{}}(Y)F^r\widehat{F}^n(Y^{})`$. Setting $`n=0`$, we see that the tensor product in $`๐ฆ`$ is given by
$$Y\times Y^{}:=\widehat{F}^s^{}(Y)F^r(Y^{}).$$
Thus $`\widehat{F}`$ is the functor of tensoring on the right by the object $`1`$ and $`F`$ the functor of tensoring on the left by the same object.
The above result can also be seen in a different way: there is a functor $``$ from $`๐ฆ`$ into the category of endofunctors of $`๐ฆ`$ defined on objects by $`(n)=\widehat{F}^n`$ and on arrows by $`(X)_r:=F^r(X)`$, $`X(m,n)`$. It combines the operations. In fact if $`T`$ is any arrow of $`๐ฆ`$,
$$\times 1_{(1)}=F(T),1_{(1)}\times (T)=\widehat{F}(T).$$
In view of this result, we refer to a normal โendofunctor on $`๐ฆ`$ with $`F(n)=n+1`$ as being a shift on $`๐ฆ`$.
As is well known, any two definitions of the tensor product on a category of Hilbert spaces are equivalent, i.e. the identity functor extends to a relaxed tensor functor. To be specific in the case at hand, if we have two shifts $`F`$ and $`G`$ on $`๐ฆ`$ and we define $`V_{m,n}:=(\widehat{G}^n,1_n,\widehat{F}^n)_m`$ then
$$V_{m^{},n^{}}\widehat{F}^n^{}(Y)F^m(Z)=\widehat{G}^n^{}(Y)G^m(Z)V_{m,n},$$
where $`Y(m,m^{})`$ and $`Z(n,n^{})`$. Furthermore,
$$\widehat{G}^p(V_{m,n})V_{m+n,p}=G^m(V_{n,p})V_{m,n+p}.$$
If we now define inductively $`K:=(0,1)`$ and $`K^n:=F(K^{n1})K`$, where the norm closed linear span is understood, then the above result shows that $`K^n=(0,n)`$.
We now make some remarks on the automorphisms of $`๐ฆ`$. Such an automorphism will be a normal โfunctor $`\mathrm{\Gamma }`$ and we suppose, as part of the definition that $`\mathrm{\Gamma }`$ leaves the objects fixed. Since an automorphism of a Hilbert space is given by a unitary operator, any automorphism of $`๐ฆ`$ will be inner, meaning that there is a unitary natural transformation $`u`$ from the identity functor to $`\mathrm{\Gamma }`$. Lemma 2.1 tells us that $`u`$ is determined by $`u_0`$. Since we are interested in $`\mathrm{\Gamma }`$ rather than $`u`$, we fix the free phase by requiring that $`u_0=1`$. An inner automorphism is then determined by a sequence of unitaries $`u_n(n,n)`$ with $`u_0=1`$. We now look for the inner automorphisms which commute with $`F`$. They must therefore commute with $`\widehat{F}`$ and preserve the tensor product structure determined by $`F`$. Applying Lemma 2.2, we derive the condition $`u_{n+1}=F(u_n)\widehat{F}^n(u_1)`$. Solving the recurrence relation gives $`u_{n+1}=F^n(u_1)F^{n1}\widehat{F}(u_1)\mathrm{}\widehat{F}^n(u_1)`$. In terms of the tensor product structure determined by $`F`$, this means that the $`u_n`$ are just tensor powers of $`u_1`$. If $`u_1`$ is a phase, we get automorphisms which commute with every shift and are the analogues of the grading automorphisms of the Cuntz algebra. Of course, if $`\mathrm{\Gamma }`$ does not commute with $`F`$ it maps the tensor product structure determined by $`F`$ onto that determined by $`\mathrm{\Gamma }F\mathrm{\Gamma }^1`$.
Our next goal is to characterize all normal โfunctors on $`๐ฆ`$ that commute with the given functor $`F`$. Note first that the action of such a functor $`G`$ on objects must be of the form $`G(n)=r+n=F^r(n)`$, $`n_0`$, for some $`r_0`$. Thus $`(G,1_r,F^r)`$ will be a natural unitary transformation from $`F^r`$ to $`G`$ and since $`F`$ commutes with these two functors, we may apply the second identity of Lemma 2.2 to deduce that
$$R_{n+1}=F(R_n)\widehat{F}^n(R_1),$$
where we have written $`R`$ for $`(F^r,1_r,G)`$.
Conversely, given a unitary $`R_1(r+1,r+1)`$, take $`R_0`$ to be the unit on $`r`$ and define $`R_n`$, $`n>1`$, inductively using the above formula. Finally, define
$$G(X)=R_nF^r(X)R_m^{}.$$
Then $`G`$ is obviously a normal โfunctor with $`G(n)=n+r`$, $`n_0`$. Furthermore, if $`X(m,n)`$,
$$GF(X)=R_{n+1}F^{r+1}(X)R_{m+1}^{}=F(R_n)\widehat{F}^n(R_1)F^{r+1}(X)R_{m+1}^{}$$
$$=F(R_n)F^{r+1}(X)F(R_m)^{}=FG(X).$$
Thus $`F`$ and $`G`$ commute and we have proved the following result.
2.3 Proposition Normal โfunctors $`G`$ commuting with $`F`$ on $`๐ฆ`$ are of the form
$$G(\psi )=RF^r(\psi ),\psi (0,1),R(r+1,r+1).$$
$`r`$ is called the rank of $`G`$.
Note that since we have already computed all natural transformations between the tensor powers of $`F`$, we have implicitly computed all natural transformations betwen two functors $`G`$ and $`G^{}`$ commuting with $`F`$. Note, too, that $`G`$ may be obtained from $`F^r`$ by acting on the left with an inner automorphism.
It is easy to compute what composition of functors means for the corresponding unitary operators. If we use the notation $`G_R`$ to denote the functor corresponding to the unitary $`R(r+1,r+1)`$, then $`G_RG_S=G_T`$, where $`T=G_R(S)F^s(R)`$. Note that Proposition 2.2 is closely related to Cuntzโs result characterizing the endomorphisms of the Cuntz algebra in terms of unitary operators.
We now consider another category $``$ of Hilbert spaces whose objects will be denoted $`H_n`$ with $`n_0`$. Suppose we have a normal โfunctor $`H`$ from $`๐ฆ`$ to $``$ whose action on objects takes $`n`$ to $`H_n`$. Then we may define a normal -endofunctor $`A`$ on $``$ by setting
$$A(X):=\underset{i}{}H\widehat{F}^n(\psi _{i,1})XH\widehat{F}^m(\psi _{i,1})^{}=(H\widehat{F}^n,X,H\widehat{F}^m)_1,X(H_m,H_n),$$
where, as before, the sum runs over an orthonormal basis. We see at once that, with this definition, $`AH=HF`$. $`H`$ is to be thought of as tensoring on the left by $`H_0`$ and $`A`$ as tensoring on the right by the object $`1`$ of $`๐ฆ`$. An easy calculation now shows that
$$A^s(X)H\widehat{F}^m(Y)=H\widehat{F}^n(Y)A^r(X),X(H_m,H_n),Y(r,s).$$
These equations show that we may define a functor $``$ from $`๐ฆ`$ to End$``$ by setting $`(r)=A^r`$ and $`(Y)_n=H\widehat{F}^n(Y)`$ for $`Y(r,s)`$.
2.4 Lemma There is a $`11`$ correspondence between normal โfunctors $`H`$ from $`๐ฆ`$ to $``$ with $`H(n)=H_n`$ and normal โfunctors $``$ from $`๐ฆ`$ to End$``$ such that $`(Y)\times \widehat{F}=\widehat{F}(Y)`$ given by $`H(Y):=(Y)_0`$.
Proof. We have seen above how to construct the functor $``$ from $`H`$.
$$((Y)\times F)_r=(Y)_{r+1}=HF^{n+1}(Y)=(F(Y))_r.$$
Conversely, given $``$, we obtain $`H`$ by evaluating in $`0`$: $`H(Y):=(Y)_0`$. Now $`HF^n(Y)=(F^n(Y))_0=(Y)_n`$. Thus $``$ is the functor associated with $`H`$.
We have seen how $`๐ฆ`$ with the functor $`F`$ is isomorphic to the category of tensor powers of $`(0,1)`$ such that $`F`$ becomes the functor of tensoring on the left by the object $`1`$. There is a similar result for $``$ with the functor $`H`$.
2.5 Proposition Let $``$ be a category of Hilbert spaces and $`H:๐ฆ`$ a normal โfunctor with $`H(r)=H_r`$. Then there is a unique isomorphism $`\mathrm{\Phi }`$ of $`(H_0,H_0)๐ฆ`$ into $``$ such that $`\mathrm{\Phi }(TY)=A^s(T)H(Y)`$, $`Y(r,s)`$.
Proof. Noting that $`A^s(T)H(Y)=H(Y)A^r(T)`$, we see that $`\mathrm{\Phi }`$ extends uniquely to the algebraic tensor product. Now given $`X(H_r,H_s)`$, write
$$X=\underset{i,j}{}H(\psi _{i,s})X_{ij}H(\psi _{j,r})^{},$$
where $`X_{ij}:=H(\psi _{i,s})^{}XH(\psi _{j,r})(H_0,H_0)`$. Hence $`X=_{i,j}A^s(X_{ij})H(\psi _{i,s}\psi _{j,r}^{})`$ showing that $`\mathrm{\Phi }`$ extends by continuity to a full functor. But any normal โfunctor from $`๐ฆ`$ is faithful hence $`\mathrm{\Phi }`$ is an isomorphism.
In view of the above results, given a shift $`F`$ on $`๐ฆ`$, we may say that a normal โfunctor $`H`$ from $`๐ฆ`$ to $``$ with $`H(r)=H_r`$ determines an action of $`(๐ฆ,F)`$ on $``$ via Lemma 2.4 making $``$ into a (right) $`(๐ฆ,F)`$โmodule.
We clearly should be able to define the tensor product of two $`๐ฆ`$โmodules $``$ and $`^{}`$ and it should just involve replacing $`(H_0,H_0)`$ and $`(H_0^{},H_0^{})`$ by $`(H_0,H_0)(H_0^{},H_0^{})`$. Let $`H`$ and $`H^{}`$ be actions of $`๐ฆ`$ on $``$ and $`^{}`$, respectively. Then the tensor product of the two actions is an action $`HH^{}`$ on a category of Hilbert spaces $`^{}`$ together with normal โfunctors $`D:^{}`$ and $`D^{}:^{}^{}`$ such that $`HH^{}=DH=D^{}H^{}`$ and $`DA=AA^{}D`$ and $`D^{}A^{}=AA^{}D^{}`$, where $`AA^{}`$ is the endofunctor on $`^{}`$ associated with $`HH^{}`$. In restriction to $`(H_0,H_0)`$ and $`(H_0^{},H_0^{})`$, we require that $`D`$ and $`D^{}`$ should define $`((HH^{})_0,(HH^{})_0)`$ as a tensor product of the von Neumann algebras $`(H_0,H_0)`$ and $`(H_0^{},H_0^{})`$. It should be noted that $`D`$ and $`D^{}`$ are uniquely determined by their values on $`(H_0,H_0)`$ and $`(H_0^{},H_0^{})`$, respectively. In fact, we have seen that with a suitable definition of tensor product the action $`H`$ becomes $`1_{(H_0,H_0)}:๐ฆ(H_0,H_0)๐ฆ`$ with a similar expression for $`H^{}`$. Taking $`^{}=(H_0,H_0)(H_0^{},H_0^{})๐ฆ`$ and $`HH^{}(Y):=1_{(H_0,H_0)(H_0^{},H_0^{})}Y`$ and defining $`D`$ to be the normal โfunctor such that $`D(TY):=T1_{(H_0^{},H_0^{})}Y`$, for $`T(H_0,H_0)`$ and $`Y(r,s)`$, with a similar expression for $`D^{}`$ we do get a tensor product of $`H`$ and $`H^{}`$.
In fact, every tensor product of actions is of this form, since, writing $`H^{\prime \prime }`$ for $`HH^{}`$, we may use $`D`$ and $`D^{}`$ to realize $`(H_0^{\prime \prime },H_0^{\prime \prime })`$ as a tensor product of $`(H_0,H_0)`$ and $`(H_0^{},H_0^{})`$. Then since $`H^{\prime \prime }`$ is an action, we have an isomorphism $`\mathrm{\Phi }^{\prime \prime }`$ from $`H^{\prime \prime }๐ฆ`$ to $`^{}`$ and $`\mathrm{\Phi }`$ and $`\mathrm{\Phi }^{}`$ from $`H๐ฆ`$ and $`H^{}๐ฆ`$ to $``$ and $`^{}`$, respectively. If we then define $`d(TY):=T1_{(H_0^{},H_0^{})}Y`$ and $`d^{}(T^{}Y):=1_{(H_0,H_0)}Y`$ using the tensor product structure on $`(H_0^{\prime \prime },H_0^{\prime \prime })`$ coming from $`D`$ and $`D^{}`$, we have $`\mathrm{\Phi }^{\prime \prime }d=D\mathrm{\Phi }`$ and $`\mathrm{\Phi }^{\prime \prime }d^{}=D^{}\mathrm{\Phi }`$. Thus the action $`H^{\prime \prime }`$ is ismorphic to an explicit tensor product of actions.
We now look for generalizations of our result characterizing categories $`๐ฆ`$ of Hilbert spaces with $`(0,0)=`$ which admit a shift by dropping the condition that $`0`$ should be irreducible.
2.6 Lemma Let $`H`$ and $`K`$ be Hilbert spaces and $`F`$ a unital normal homomorphism from $`(H,H)`$ to $`(K,K)`$. Then we can pick matrix units $`E_{ab}`$ in $`(H,H)`$ and $`E_{ar,bs}`$ in $`(K,K)`$ such that $`F(E_{bc})=_rE_{br,cr}`$.
Proof. Pick matrix units $`E_{ab}`$ in $`(H,H)`$ then for a fixed $`a`$, $`F(E_{aa})`$ is a non-zero projection and we may pick matrix units $`E_{ar,as}`$ for the Hilbert space $`F(E_{aa})K`$. Thus, in particular, $`F(E_{aa})=_rE_{ar,ar}`$. Now define
$$E_{br,cs}:=F(E_{ba})E_{ar,as}F(E_{ac}).$$
A routine computation shows that
$$E_{br,cs}E_{dt,ev}=\delta _{cd}\delta _{st}E_{br,ev}$$
$$\underset{r}{}E_{br,cr}=F(E_{bc})=I.$$
Thus we have defined a set of matrix units with the required properties.
2.7 Lemma Let $`H`$ and $`K`$ be Hilbert spaces and $`F`$ a unital normal homomorphism from $`(H,H)`$ to $`(K,K)`$ then there is a Hilbert space $`L`$ of support one in $`(H,K)`$ such that
$$\psi T=F(T)\psi ,\psi L,T(H,H).$$
Proof. Pick orthonormal bases $`e_a`$ and $`e_{ar}`$ in $`H`$ and $`K`$ in such a way that the corresponding matrix units are as in Lemma 2.6. Define $`\psi _r(H,K)`$ by $`\psi _re_a:=e_{ar}`$ and a computation shows that $`\psi _r`$ is a basis of a Hilbert space $`L`$ of support one in $`(H,K)`$. Now $`\psi _rE_{bc}e_d=\delta _{cd}e_{br}`$ and $`F(E_{bc})\psi _re_d=_sE_{bs,cs}e_{dr}=\delta _{cd}e_{br}`$, completing the proof.
Now let $``$ be a category of Hilbert spaces with objects labelled by $`_0`$ and $`F`$ a normal -functor from $``$ to $``$ acting on objects as $`F(n)=n+1`$. We have analyzed the situation when $`0`$ is irreducible and we begin a similar analysis in the general case.
By Lemma 2.7, there is a Hilbert space $`L`$ of support one in $`(0,1)`$ inducing $`F`$ on $`(0,0)`$. We now define as before
$$\widehat{F}(X)=\underset{i}{}F^n(\psi _i)XF^m(\psi _i^{}).$$
$`\widehat{F}`$ will be a normal โfunctor with $`\widehat{F}(n)=n+1`$ and commuting with $`F`$ and $`\widehat{\widehat{F}}=F`$. The
$$F^{n1}(\psi _{i_1})\mathrm{}F(\psi _{n1})\psi _{i_n}$$
form an orthonormal basis $`\psi _{i,n}`$ of a Hilbert space of support one in $`(0,n)`$. We find
$$\widehat{F}^k(X)=\underset{i}{}F^n(\psi _{i,k})XF^m(\psi _{i,k}^{}).$$
We can interchange the role of $`F`$ and $`\widehat{F}`$. But care is required as we need another Hilbert space of support one in $`(0,n)`$. It follows from Lemma 2.7 that $`\widehat{F}(T)=F(T)`$ for $`T(0,0)`$ so that $`L`$ remains the correct Hilbert space in $`(0,1)`$. In $`(0,n)`$ we need a Hilbert space with orthonormal basis
$$\widehat{F}^{n1}(\psi _{i_1})\mathrm{}\widehat{F}(\psi _{n1})\psi _{i_n}$$
denoted $`\widehat{\psi }_{i,n}`$. Then
$$F^k(X)=\underset{i}{}\widehat{F}^n(\widehat{\psi }_{i,k})X\widehat{F}^m(\widehat{\psi }_{i,k}^{}).$$
We now define a $`W^{}`$โsubcategory $`๐ฆ`$ of $``$ whose arrows consist of those $`S(m,n)`$ such that $`SF^m(T)=F^n(T)S`$. This subcategory is invariant under $`F`$ and the object $`0`$ is now irreducible. By Lemma 2.7, $`L`$ is a Hilbert space of support one in this category so that this is actually a category of Hilbert spaces. So our previous result is applicable and $`๐ฆ`$ is isomorphic to the category of tensor powers of a Hilbert space $`K`$ in such a way that $`F`$ is the functor of tensoring on the right by $`1_K`$.
But the inclusion functor of $`๐ฆ`$ in $``$ is now a normal โfunctor which is the identity on objects. The associated endofunctor $`A`$ is just $`F`$. Our previous analysis yields the following result.
2.8 Proposition Let $``$ be a category of Hilbert spaces with objects labelled by $`_0`$ and $`F`$ a normal -functor from $``$ to $``$ acting on objects as $`F(n)=n+1`$. Then $``$ is isomorphic to the category of Hilbert spaces whose objects are of the form $`HK^n`$, $`n_0`$, for Hilbert spaces of $`H`$ and $`K`$ in such a way that $`F`$ becomes the functor of tensoring on the right by $`1_K`$.
## 3 Shifts and Multiplicative Unitaries
We used above the characterization of endomorphisms of the Cuntz algebra in terms of unitaries given by Cuntz who also showed that if $`K`$ is a finite-dimensional Hilbert space then a unitary $`V(K^2,K^2)`$ is a multiplicative unitary if the endomorphism $`\lambda _R`$ of the Cuntz algebra $`๐ช_K`$ satisfies
$$\lambda _R\lambda _R=\rho \lambda _R,$$
where $`R=V\theta `$, $`\theta `$ being the flip on $`K^2`$, and $`\rho `$ the endomorphism generated by the defining Hilbert space $`K`$. This result remains valid in the extended Cuntz algebra if $`K`$ is infinite dimensional. The multiplicativity of $`V`$ is alternatively expressed by the identity $`\lambda _V^{}\lambda _R=\rho \lambda _V^{}`$. The corresponding identities for normal -functors on $`๐ฆ`$ commuting with $`F`$ imply that a unitary $`V(2,2)`$ is multiplicative for the tensor structure induced by $`\widehat{F}`$, i.e. where $`F`$ corresponds to tensoring on the right and $`\widehat{F}`$ to tensoring on the left by the object $`1`$. The role of the endomorphism $`\rho `$ of the Cuntz algebra is played by $`\widehat{F}=G_\theta `$. In other words, the following result holds.
3.1 Proposition Let $`F`$ and $`G`$ be two commuting shifts on $`๐ฆ`$. Let $`R(2,2)`$ be the unitary such that $`G(\psi )=RF(\psi )`$, $`\psi (0,1)`$. Set $`V:=R\theta `$, where $`\theta `$ is the flip on $`(2,2)`$ derived from the tensor structure induced by $`\widehat{F}`$, then the following conditions are equivalent.
a) $`V`$ is a multiplicative unitary.
b) $`GG=\widehat{F}G`$.
bโ) $`FF=\widehat{G}F`$.
c) $`nF^{n+1}(\psi )`$ is a natural transformation from $`G`$ to $`GG`$, $`\psi (0,1)`$.
cโ) $`nG^{n+1}(\psi )`$ is a natural transformation from $`F`$ to $`FF`$, $`\psi (0,1)`$.
d) $`GG(\psi )F(\psi ^{})=FF(\psi ^{})G(\psi )`$, $`\psi ,\psi ^{}(0,1)`$.
e) $`G_V^{}G=\widehat{F}G_V^{}`$.
f) $`nF^{n+1}(\psi )`$ is a natural transformation from $`G_V^{}`$ to $`G_V^{}G`$, $`\psi (0,1)`$.
g) $`G_V^{}G(\psi )F(\psi ^{})=FF(\psi ^{})G_V^{}(\psi )`$, $`\psi ,\psi ^{}(0,1)`$.
Proof. The equivalence of a), b) and e) is a simple computation whose origins were explained above. Suppose a) is valid and pick $`X(m,n)`$ then
$$\underset{i}{}F^{n+1}(\psi _{i,1})G(X)F^{m+1}(\psi _{i,1})^{}=\widehat{F}G(X)=GG(X).$$
Hence $`F^{n+1}(\psi )G(X)=G(X)F^{m+1}(\psi )`$ for $`\psi (0,1)`$, giving c). d) follows as a special case. But if d) holds, then
$$\widehat{F}G(\psi )=\underset{i}{}F^2(\psi _{i,1})G(\psi )F(\psi _{i,1})^{}=GG(\psi ).$$
But the set of arrows $`X`$ in $`๐ฆ`$ such that $`\widehat{F}G(X)=GG(X)`$, is a $`W^{}`$โsubcategory of $`๐ฆ`$ which is invariant under the action of $`F`$, seeing that $`F`$ commutes with $`\widehat{F}`$ and $`G`$. Thus d) implies b). In the same way, we can prove that e), f) and g) are equivalent. Finally, the symmetry between $`F`$ and $`G`$, visible in d), shows that bโ) and cโ) are also equivalent to the remaining conditions.
We have seen that the situation in Proposition 3.1 is symmetric in the two commuting shifts $`F`$ and $`G`$. Interchanging $`F`$ and $`G`$ obviously corresponds to replacing $`R`$ by $`R^1`$ and hence $`V:=R\theta `$ by $`\widehat{V}=R^1\theta `$, the dual multiplicative unitary. The multiplicative unitaries on $`(0,2)`$ for the tensor structure determined by $`F`$ are clearly in $`11`$ correspondence with the shifts commuting with $`F`$ and satisfying the equivalent conditions of Proposition 3.1. Equivalent multiplicative unitaries correspond to shifts conjugate under (inner) automorphisms of $`๐ฆ`$ commuting with $`F`$.
We can also characterize the intertwining operators between the tensor powers of $`V`$, regarded as an object in the category of representations of $`V`$, in terms of the commuting shifts.
3.2 Lemma Let $`F`$ and $`G`$ be commuting shifts on $`๐ฆ`$ with $`GG=\widehat{F}G`$ and let $`V`$ be the associated multiplicative unitary. Set $`G^{}:=G_{\vartheta V}`$ then
$$(V^r,V^s)=\{Y(r,s):G^{}(Y)=\widehat{F}(Y)\}.$$
Proof. Interpreting $`F`$ as tensoring on the right by $`1`$, the condition $`G^{}(Y)=\widehat{F}(Y)`$ reads
$$(\vartheta V)_sY\times 1(\vartheta V)_r^{}=\vartheta _sY\times 1\vartheta _r^{}.$$
Hence, it suffices to show that $`\vartheta _s^{}(\vartheta V)_s=V_{1s+1}V_{2s+1}\mathrm{}V_{ss+1}`$, where we have used the index notation on the right hand side. However, this may be proved by induction. In fact
$$\vartheta _s^{}(\vartheta V)_s=1_{s1}\times \vartheta \vartheta _{s1}^{}\times 1(\vartheta V)_{s1}\times 11_{s1}\times (\vartheta V)$$
$$=\vartheta _{ss+1}V_{1s}V_{2s}\mathrm{}V_{s1s}\vartheta _{ss+1}V_{ss+1}=V_{1s+1}V_{2s+1}\mathrm{}V_{ss+1}.$$
3.3 Corollary Let $`F`$ and $`G`$ be two commuting shifts on $`๐ฆ`$ such that $`GG=\widehat{F}G`$ and $`V`$ the corresponding multiplicative unitary. Let $`G^{}:=G_{\vartheta V}`$ and consider a sequence $`t_n(r+n,s+n)`$ that defines simultaneously a natural transformation $`t`$ in $`(\widehat{G}^{}_{}{}^{}r,\widehat{G}^{}_{}{}^{}s)`$ and $`(\widehat{F}^r,\widehat{F}^s)`$. Such natural transformations form a tensor $`W^{}`$โsubcategory of the tensor $`W^{}`$โcategory of all natural transformations between the powers of $`\widehat{G}^{}`$. Evaluating $`t`$ in $`0`$ establishes an isomorphism with the tensor $`W^{}`$โcategory of intertwiners between powers of $`V`$.
Proof. If $`t(\widehat{G}^{}_{}{}^{}r,\widehat{G}^{}_{}{}^{}s)(\widehat{F}^r,\widehat{F}^s)`$, then, by Lemma 2.1, $`t_n=G^{}_{}{}^{}n(t_0)=F^n(t_0)`$. Thus $`t`$ is uniquely determined by $`t_0`$ and, as $`G^{}`$ and $`F`$ commute, the only constraint on $`t_0`$ is that $`G^{}(t_0)=F(t_0)`$. Lemma 3.2 shows that we have an isomorphism of $`W^{}`$โcategories and a computation shows that it preserves tensor products.
We now comment on the role of multiplicative unitaries, seen in this light. As we have seen, $`F`$ determines a tensor product structure on $`๐ฆ`$ and $`G`$ determines another so it is natural to interpret $`R`$ as describing the transition from one tensor product to another. However, any unitary $`R(2,2)`$ would serve here. We do not need $`V:=R\theta `$ to be a multiplicative unitary. Although it is perfectly correct to say that $`F`$ determines a tensor product structure on $`๐ฆ`$, this structure really involves two commuting shifts $`F`$ and $`\widehat{F}`$ which we interpret as tensoring on the right by $`1`$ and on the left by $`1`$. Thus taking two commuting shifts $`F`$ and $`G`$ can be regarded as generalizing the idea of a tensor product. It is less symmetric in that $`\theta `$ has been replaced by $`R`$ and $`\widehat{F}`$ by $`G`$ and does not give rise to a bifunctor as a true tensor product unless $`G=\widehat{F}`$. Looked at this way, the condition of $`V`$ being multiplicative being equivalent to b) has a simple interpretation. It is a โdepth 2โ condition: it is not necessary to apply $`G`$ more than once since in successive applications it can always be replaced by $`\widehat{F}`$. The justification for adopting this terminology from the theory of subfactors is Lemma 6.3 of . We shall say that the two commuting shifts have relative depth two to stress that the notion involves the two shifts symmetrically.
Let us look at some examples of actions of $`๐ฆ`$ on $``$. A first example is suggested directly by Proposition 2.5. Given a Hilbert space $`H`$, take $`:=H๐ฆ`$ and then define $`H:๐ฆ`$ by $`H(Y):=1_HY`$. With this definition we find $`A(X)=X1_K`$. We get other normal โfunctors from $`๐ฆ`$ to $``$ by picking for each $`r`$ a unitary $`W_r(H_r,H_r)`$ and then defining $`E(Y):=W_s1_HYW_r^{}`$, $`Y(r,s)`$. In fact, we are in the situation of Lemma 2.1 and $`W_s=(E,1_{H_0},H)_s`$. A computation shows that in this case the corresponding endofunctor $`B`$, say, is given by
$$B(X)=W_{n+1}A(W_n^{}XW_m)W_{m+1}^{},X(H_m,H_n).$$
In particular, suppose we start from a multiplicative unitary $`V`$ on $`K^2`$ and a representation $`W`$ on $`H`$. We can define $`H`$ and $`E`$ as above but making the particular choice
$$W_s:=W_{12}W_{13}\mathrm{}W_{1s+1},$$
using index notation.
3.4 Lemma Let $`W`$ be a representation of a multiplicative unitary then defining functors $`E,H:๐ฆ`$, as above and letting $`G`$ be the endofunctor on $`๐ฆ`$ defined by $`R:=V\theta `$, as before, we have $`EG=HG`$.
Proof. It suffices to show that $`EG(\psi )=HG(\psi )`$ for each $`\psi (0,1)`$. Now $`HG(\psi )=H(R)HF(\psi )`$, whereas
$$EG(\psi )=E(R)EF(\psi )=E(R)W_{12}W_{13}HF(\psi )W_{12}^{}=E(R)W_{12}HF(\psi ).$$
However $`E(R)=W_{12}W_{13}R_{23}W_{13}^{}W_{12}^{}`$ and $`H(R)=R_{23}`$, so the identity in question follows from the definition of a representation $`W`$ of $`V`$, $`W_{12}W_{13}V_{23}=V_{23}W_{12}`$.
We can therefore adopt the following point of view. Regard a multiplicative unitary as a category $`๐ฆ`$ of Hilbert spaces, as above, equipped with two commuting shifts $`F`$ and $`G`$ satisfying $`GG=\widehat{F}G`$. A representation of such a multiplicative unitary is a category of Hilbert spaces $``$, as above, equipped with two normal โfunctors $`E`$ and $`H`$ from $`๐ฆ`$ to $``$ of rank zero such that $`EG=HG`$. An intertwining operator between two representations is a bounded linear operator $`T(H_0,H_0^{})`$ such that, setting
$$A(X):=\underset{i}{}H^{}\widehat{F}^n(\psi _{i,1})XH\widehat{F}^m(\psi _{i,1})^{},X(H_m,H_n^{}),$$
$`rA^r(T)`$ defines a natural transformation from $`E`$ to $`E^{}`$. In fact, if we pick $`\psi (0,1)`$, this gives
$$A(T)W1_H\psi =W^{}1_H^{}\psi T=W^{}A(T)1_H^{}\psi .$$
Since $`A(T)=T1_K`$, $`T(W,W^{})`$. On the other hand, if $`T(W,W^{})`$, then $`A^r(T)W_r=W_r^{}A^r(T)`$ and this implies that $`rA^r(T)`$ is a natural transformation from $`E`$ to $`E^{}`$. Note, however, that $`A^r(T)=(H^{},T,H)_r`$, so that $`T(H_0,H_0^{})`$ is in $`(W,W^{})`$ if and only if $`rA^r(T)`$ defines an element of $`(H,H^{})(E,E^{})`$.
Superficially, the relation $`GG=\widehat{F}G`$ looks like a special case of $`EG=HG`$. However, $`E`$ and $`H`$ have rank zero, whereas $`G`$ and $`\widehat{F}`$ have rank one. However, if we delete the object $`0`$ from $`๐ฆ`$ to give a full subcategory $`๐ฆ_+`$ and let $`G_+`$ and $`\widehat{F}_+`$ denote the functors obtained by restricting the range to $`๐ฆ_+`$, the relation $`G_+G=\widehat{F}_+G`$ is a special case of $`EG=HG`$ and we are considering $`V`$ as an object in its category of representations.
The relation $`EG=HG`$ is symmetrical in $`E`$ and $`H`$. Exchanging $`E`$ and $`H`$ corresponds to replacing $`W`$ by $`W^1`$ which will not, in general be a representation of the multiplicative unitary $`V`$. However, the tensor product notation involved in the definition of representation refers to $`H`$.
$`H`$ is to be interpreted here as the trivial representation on the same space $`H_0`$ as $`E`$. Since $`(E,E^{})(H,H^{})(H,H^{})`$, the category of representations of $`V`$ automatically comes equipped with a faithful functor into the subcategory of trivial representations.
When we have two commuting shifts and two actions $`E`$ and $`H`$, as above, it seems appropriate to refer to $``$ as a $`๐ฆ`$โbimodule. We have already considered the tensor product of $`๐ฆ`$โmodules in the last section and we now extend these considerations to bimodules. Given therefore two action $`E^{}`$ and $`H^{}`$ on $`^{}`$, we form $`^{}`$ and $`HH^{}`$. Of course, we could equally well have formed $`EE^{}`$ instead, but what we really need is the relation between the two actions. We therefore use the functors $`D`$ and $`D^{}`$, expressing $`^{}`$ as a tensor product and set
$$(EE^{})(Y)=D(W_s)D^{}E^{}(Y)D(W_r^{}),Y(r,s)$$
where $`W_r=(E,1_{H_0},H)_r`$. It is easy to check that $`EE^{}`$ is a normal โfunctor and that this definition applied to $`H`$ and $`H^{}`$ gives $`HH^{}`$. Expressing $`E^{}`$ in terms of $`H^{}`$ using $`W_r^{}=(E^{},1_{H_0^{}},H^{})_r`$, the definition is equivalent to
$$(EE^{},1_{(HH^{})_0},HH^{})_r=D(W_r)D^{}(W_r^{}).$$
There is obviously a convention involved here because we have chosen to write the primed terms to the right of the unprimed terms. However, this convention is consistent with that used for multiplicative unitaries in that it corresponds to taking the tensor product of representations $`W`$ and $`W^{}`$ as $`W_{13}W_{23}^{}`$. It is easily checked that $`EG=HG`$ and $`E^{}G=H^{}G`$ imply $`EE^{}G=HH^{}G`$.
## 4 Multiplicative Unitaries and Tensor Categories
We now exhibit a mechanism leading from objects in a tensor $`W^{}`$โcategory to multiplicative unitaries. It will involve a category of Hilbert spaces $`๐ฆ`$, equipped with two commuting shifts $`F`$ and $`G`$. Let $`\rho `$ be an object in a strict tensor $`W^{}`$โcategory and suppose that $`K`$ is a Hilbert space of support one contained in $`(\rho ,\rho ^2)`$. We then define inductively $`K^n:=K^{n1}\times 1_\rho K`$, where the norm closed linear span is understood. Then $`K^n`$ is a Hilbert space of support 1 in $`(\rho ,\rho ^{n+1})`$. $`K^0`$ will denote $`1_\rho `$. We now set
$$(K^m,K^n):=\{X(\rho ^{m+1},\rho ^{n+1}):XK^mK^n\}.$$
We see that $`K^n=(K^0,K^n)`$ and thus we have defined a $`W^{}`$โsubcategory $`๐ฆ`$ of Hilbert spaces of the tensor $`W^{}`$โcategory $`๐ฏ_\rho `$ whose objects are the tensor powers of $`\rho `$. We claim that $`๐ฆ`$ is invariant under tensoring on the right by $`1_\rho `$. In fact, it suffices to show that $`K\times 1_\rho (K,K^2)`$, i.e. that $`K\times 1_\rho KK^2`$ but this is true by construction. If $`F`$ denotes the restriction of $`\times 1_\rho `$ to $`๐ฆ`$, $`๐ฆ`$ has unique structure of tensor $`W^{}`$โcategory such that $`F`$ becomes the functor of tensoring on the right by $`1_K`$.
4.1 Lemma Let $`K(\rho ,\rho ^2)`$ be a Hilbert space of support one in a tensor $`W^{}`$โcategory and let $`๐ฆ`$ be the subcategory of Hilbert spaces defined as above, then the following conditions are equivalent.
a) $`1_\rho \times K(K,K^2)`$.
b) $`๐ฆ`$ is an invariant subcategory for $`1_\rho \times `$.
c) $`K^n=1_\rho \times K^{n1}K`$, $`n`$.
d) $`K^2=1_\rho \times KK`$.
Proof. $`๐ฆ`$ is the smallest $`W^{}`$โsubcategory containing $`K`$ and invariant under $`\times 1_\rho `$. Furthermore $`1_\rho \times `$ and $`\times 1_\rho `$ commute, thus a) implies b). Given b), we know from Proposition 2.3 that $`1_\rho \times \psi =R_n\psi \times 1_\rho `$, for $`\psi K^n`$ with $`R_n`$ unitary, giving c). c) implies d), trivially and d) implies a).
We call a Hilbert space $`K`$ satisfying the equivalent conditions of Lemma 4.1 ambidextrous.
4.2 Theorem Let $`K(\rho ,\rho ^2)`$ be an ambidextrous Hilbert space of support one in $`(\rho ,\rho ^2)`$. Let $`F`$ and $`G`$ denote the restrictions of $`\times 1_\rho `$ and $`1_\rho \times `$ to $`๐ฆ`$ then there is a unique $`V(K^2,K^2)`$ such that
$$G(\psi )=V\widehat{F}(\psi ),\psi K.$$
$`V`$ is a multiplicative unitary.
Proof. $`V`$ is unique and is unitary because it is given by
$$V=\underset{i}{}G(\psi _{i,1})\widehat{F}(\psi _{i,1})^{}$$
where the sum is taken over an orthonormal basis. Now since $`K(\rho ,\rho ^2)`$, for $`X(K^m,K^n),`$
$$\widehat{F}G(X)=\underset{i}{}F^{n+1}(\psi _{i,1})1_\rho \times XF^{m+1}(\psi _{i,1})^{}=1_{\rho ^2}\times X=G^2(X).$$
Thus $`G^2=\widehat{F}G`$ and $`V`$ is a multiplicative unitary by Proposition 2.3.
Every multiplicative unitary can be realized in this manner.
4.3 Proposition Let $`V`$ be a multiplicative unitary $`V(K^2,K^2)`$, then $`G_V^{}(K)`$ is an ambidextrous Hilbert space $`H(K,K^2)`$ of support one and if $`U(K,H)`$ is the unitary taking $`\psi `$ to $`G_V^{}(\psi )`$ the multiplicative unitary defined by $`H`$ is $`U\times UV(U\times U)^1`$.
Proof. $`H:=V^{}K\times 1_K=G_V^{}(K)`$ is obviously a Hilbert space of support one in $`(K,K^2)`$. To verify that $`H`$ is ambidextrous, it suffices by Proposition 2.4 to verify that
$$\psi _2^{}\times 1_KV1_K\times \psi _1^{}1_K\times VV^{}\times 1_K\psi _3\times 1_{K^2}V^{}\psi _4\times 1_K$$
is a multiple of $`1_K`$ for any choice of $`\psi _iK`$, $`i=1,2,3,4`$. Using the multiplicativity of $`V`$, this reduces to
$$(\psi _2^{}\times \psi _1^{}V^{}\psi _3\times \psi _4)\times 1_K$$
and hence is a multiple of $`1_K`$. It remains to compute the operator $`S`$ defined by
$$SV^{}\times 1_K\psi \times 1_{K^2}=1_K\times V^{}1_K\times \psi \times 1_K,\psi K.$$
Writing $`\varphi _iH`$ for $`V^{}\psi _i\times 1_K`$ and computing using the multiplicativity of $`V`$, we find
$$\varphi _1^{}\times \varphi _2^{}S\varphi _3\times \varphi _4=((\psi _1\times \psi _2)^{}R\psi _3\times \psi _4)\times 1_K.$$
Thus the multiplicative unitary associated with $`H`$ is as asserted.
An analogous computation shows that we may replace $`G_V^{}`$ by $`G_R`$ in Proposition 4.3.
We now ask how the multiplicative unitary depends on the choice of ambidextrous Hilbert space $`H(\rho ,\rho ^2)`$. Let $`H^{}`$ be another such Hilbert space then there is a unitary $`U(\rho ^2,\rho ^2)`$ such that $`UH=H^{}`$. Let $`V`$ and $`V^{}`$ denote the multiplicative unitary operators associated with $`H`$ and $`H^{}`$ and set $`R:=V\theta `$, $`R^{}:=V^{}\theta ^{}`$, where $`\theta `$ and $`\theta ^{}`$ are the flips on $`H^2`$ and $`H^{}_{}{}^{}2`$, respectively. We compute the relation between $`R`$ and $`R^{}`$. Let $`\psi H`$, then
$$R^{}(U\psi )\times 1_\rho =1_\rho \times (U\psi )=1_\rho \times UR\psi \times 1_\rho .$$
Thus $`R^{}=1_\rho \times URU^{}\times 1_\rho `$. This is in contrast to the transformation law of $`\theta `$, namely $`\theta ^{}=u_2\theta u_{2}^{}{}_{}{}^{}`$ where $`u_2:=U\times 1_H1_H\times U`$ is the tensor power of $`U`$. It should be remembered however that $`R^{}`$ is intrinsically determined by $`H^{}`$ whereas $`U`$ is not. The interesting question is whether the associated multiplicative unitaries $`V`$ and $`V^{}`$ are necessarily equivalent and for which unitaries $`U`$, $`UH`$ is ambidextrous.
We present an example. Let $`๐ฆ`$ denote the $`W^{}`$โtensor category of tensor powers of a Hilbert space $`K`$. Let $`V(K^2,K^2)`$ be a multiplicative unitary, then, as we have seen, $`H:=V^{}K`$ is an ambidextrous Hilbert space in $`(K,K^2)`$ whose associated multiplicative unitary is equivalent to $`V`$. Since we are free to choose any multiplicative unitary $`V`$, this makes it clear that the associated multiplicative unitaries can depend in an essential way on the ambidextrous Hilbert space.
We may sum up the results to date in this section as follows.
4.4 Theorem Let $`๐ฆ`$ be a category of Hilbert spaces with objects $`K_n`$, $`n_0`$ and $`(K_0,K_0)=`$ equipped with commuting shifts $`F`$ and $`G`$ such that $`GG=\widehat{F}G`$. Let $`V(K_2,K_2)`$ be the multiplicative unitary such that $`G(\psi )=V\widehat{F}(\psi )`$, $`\psi K_1`$, then $`H:=V^{}F(K_1)`$ is an ambidextrous Hilbert space. Let $``$ be the resulting category of Hilbert spaces with commuting shifts $`D`$ and $`E`$ obtained by restricting $`F`$ and $`\widehat{F}`$ to $``$, then the shift $`G^{}`$ on $`๐ฆ`$ defined by $`G^{}(\psi )=V^{}F(\psi )`$, $`\psi K_1`$ yields an isomorphism of $`๐ฆ,F,G`$ with $`,D,E`$.
Proof. We know from Proposition 3.1 that $`V`$ is a multiplicative unitary and from Proposition 4.3 that $`G^{}(K_1)`$ is an ambidextrous Hilbert space. The resulting category $``$ of Hilbert spaces is thus $`G^{}(๐ฆ)`$. Since $`G^{}`$ commutes with $`F`$ and by Proposition 3.1, $`G^{}G=\widehat{F}G^{}`$, $`G^{}`$ does yield the desired isomorphism.
Remark. The construction of this section deriving a multiplicative unitary from an ambidextrous Hilbert space is invariant under tensor โfunctors since the image of an ambidextrous Hilbert space of support one is again such a Hilbert space and a tensor โfunctor commutes with tensoring on the right and left.
We have seen in Theorem 4.2 how an ambidextrous Hilbert space leads to a multiplicative unitary. There is a variant of this result which instead yields a representation of a multiplicative unitary. To motivate this result, we let $`V`$ be a multiplicative unitary on the tensor power of a Hilbert space $`L`$ and $`W`$ a representation of $`V`$ on a Hilbert space $`M`$. Then $`V`$ and $`W`$ are objects of the tensor $`W^{}`$โcategory $`(V)`$ and, as we know, $`K:=V^{}L\times 1_L`$ is an ambidextrous Hilbert space in $`(V,V^2)`$. However, $`W`$ being a representation of $`V`$, $`H_0:=W^{}M\times 1_L`$ is a Hilbert space of support one in $`(V,WV)`$. Hence $`H_n:=H_{n1}\times 1_VK`$ is a Hilbert space of support one in $`(V,WV^{n+1})`$. Thus just as we have a category $`๐ฆ`$ of Hilbert spaces associated with $`K`$, there is a category $``$ associated with $`H_0`$. We claim that tensoring on the left with $`1_W`$ restricts to a โfunctor from $`๐ฆ`$ to $``$. It suffices to show that $`1_W\times KH_0H_1`$ and, expressing $`K`$ and $`H_0`$ in terms of $`L`$ and $`M`$, this is again a consequence of $`W`$ being a representation of $`V`$. We have here an obvious generalization of the notion of ambidextrous Hilbert space involving $`K`$ and $`H_0`$.
We use this as the basis of a definition. Let $`๐ฏ`$ be a tensor $`W^{}`$โcategory and $`\rho `$ and $`\sigma `$ objects of $`๐ฏ`$. Let $`K(\rho ,\rho ^2)`$ and $`H_0(\rho ,\sigma \rho )`$ be Hilbert spaces of support one and $`๐ฆ`$ and $``$ the corresponding categories of Hilbert spaces with objects $`K^n:=K^{n1}\times 1_\rho K`$ and $`H_n:=H_{n1}\times 1_\rho K`$, $`n_0`$, respectively. We say that $`H`$ is $`K`$โambidextrous if $`1_\sigma \times `$ restricts to a โfunctor from $`๐ฆ`$ to $``$.
Lemma 4.5 Let $`๐ฏ`$ be a tensor $`W^{}`$โcategory and $`K(\rho ,\rho ^2)`$ and $`H(\rho ,\sigma \rho )`$ be Hilbert spaces of support one then the following conditions are equivalent.
a) $`1_\sigma \times K(H_0,H_1)`$,
b) $`H`$ is $`K`$โambidextrous,
c) $`H_n=1_\sigma \times K^nH_0`$,
d) $`H_1=1_\sigma \times KH_0`$.
Proof. $`๐ฆ`$ is the smallest $`W^{}`$โsubcategory containing $`K`$ and invariant under $`\times 1_\rho `$. Furthermore $`1_\sigma \times `$ and $`\times 1_\rho `$ commute, thus a) implies b). Since both sides of c) are Hilbert spaces of support one, b) implies c). c) implies d), trivially and d) implies a).
4.6 Theorem Let $`K(\rho ,\rho ^2)`$ be an ambidextrous Hilbert space of support one in $`(\rho ,\rho ^2)`$ and $`H`$ a $`K`$โambidextrous Hilbert space in $`(\rho ,\sigma \rho )`$. Let $`F`$, $`G`$, $`E`$ and $`H`$ denote the restrictions of $`\times 1_\rho `$, $`1_\rho \times `$, $`1_\sigma \times `$ and $`1_{H_0}\times `$, respectively, to $`๐ฆ`$ then there is a unique $`V(K^2,K^2)`$ such that
$$G(\psi )=V\widehat{F}(\psi ),\psi K,$$
and a unique $`W(H_1,H_1)`$ such that
$$E(\psi )=WH(\psi ),\psi K.$$
$`V`$ is a multiplicative unitary and $`W`$ is a representation of $`V`$ on $`H_0`$.
Proof. In view of Theorem 4.2, we need only prove the assertions relating to $`W`$. $`W`$ is unique and is unitary because it is given by
$$W=\underset{i}{}E(\psi _{i,1})H(\psi _{i,1})^{}$$
where the sum is taken over an orthonormal basis of $`K`$. Now since $`H_0(\rho ,\sigma \rho )`$, for $`X(K^m,K^n)`$,
$$HG(X)=\underset{i}{}F^{n+1}(\varphi _i)1_\rho \times XF^{m+1}(\varphi _i)^{}=1_{\sigma \rho }\times X=EG(X),$$
where $`\varphi _i`$ is an orthonormal basis of $`H_0`$. Thus $`HG=EG`$ and $`W`$ is a representation of $`V`$ by the discussion following Lemma 3.4.
Theorem 4.6 does not really refer to the whole tensor $`W^{}`$โcategory $`๐ฏ`$ but only to the full subcategory with objects $`\rho ^n`$ and $`\sigma \rho ^n`$ and the structures induced on this subcategory by tensoring on the left and right by $`1_\rho `$ and on the left by $`1_\sigma `$. Introducing $`๐ฏ`$ enables us to avoid spelling out the structures.
We now show how a representation of multiplicative unitaries gives rise to an interchange law.
4.7 Proposition Let $`F`$ and $`G`$ be two commuting shifts on $`๐ฆ`$ with $`GG=\widehat{F}G`$ and $`E`$ and $`H`$ normal โfunctors from $`๐ฆ`$ to $``$ with $`EG=HG`$, then given $`X(H_m,H_n)`$ and $`Y(p,q)`$,
$$B^{q+1}(X)EG^{m+1}(Y)=EG^{n+1}(Y)B^{p+1}(X),$$
where $`B`$ is the normal โfunctor on $``$ associated with $`E`$:
$$B(X):\underset{i}{}EF^n(\psi _{i,1})XEF^m(\psi _{i,1})^{},X(H_m,H_n).$$
Thus defining $`X\times ^{}Y:=B^{q+1}(X)EG^{m+1}(Y)`$ gives an action of the tensor $`W^{}`$โcategory $`๐ฆ^+`$ on the category $``$ of Hilbert spaces. If $`๐ฆ`$ is given the tensor structure determined by $`F`$ then there is a unique normal tensor โfunctor $`G^{}`$ from $`๐ฆ^+`$ to $`๐ฆ`$ such that $`G^{}(\psi ):=V^{}F(\psi )`$, $`\psi K`$, where $`V`$ is the multiplicative unitary associated with $`F`$ and $`G`$.
Proof. To prove the interchange law, write $`EG^{m+1}=H\widehat{F}^mG`$ and $`EG^{n+1}=H\widehat{F}^nG`$ and use the interchange law between $`A`$ and $`H\widehat{F}`$ discussed before Lemma 2.4. The remarks above show that $`๐ฆ^+`$ is a tensor $`W^{}`$โcategory and allow us to check that $`G^{}`$ is a tensor โfunctor using Proposition 3.1e.
As we shall be considering Hilbert spaces $`L(K^r,K^{r+g})`$ in Section 5, it is natural to ask to what extent the results can be generalized to Hilbert spaces $`K`$ of support one in $`(\rho ^r,\rho ^{r+g})`$, where we suppose, of course, that $`g0`$. The initial construction can be easily modified. We define inductively $`K^n:=K^{n1}\times 1_{\rho ^g}K`$, the norm closed linear span being understood. $`K^n`$ is a Hilbert space of support 1 in $`(\rho ^r,\rho ^{r+ng})`$. We now set
$$(K^m,K^n):=\{X(\rho ^{r+mg},\rho ^{r+ng}):XK^mK^n\}.$$
In this way we have a $`W^{}`$โsubcategory $`๐ฆ`$ of Hilbert spaces whose objects are of the form $`\rho ^{r+ng}`$, $`n_0`$. $`๐ฆ`$ is now invariant under tensoring on the right by $`1_{\rho ^g}`$. Letting $`F`$ be the restriction of this functor to $`๐ฆ`$ we have a shift on $`๐ฆ`$ that can be regarded as tensoring on the right by $`1_K`$. The analogue of Lemma 4.1 now holds if we consider tensoring on the left by $`1_{\rho ^g}`$ and can be used to define the notion of ambidextrous Hilbert space. Thus we are again led to a category of Hilbert spaces $`๐ฆ`$ equipped with two commuting shifts. At this point there is the essential difference: we cannot use the exchange law in $`๐ฏ_\rho `$ to get an analogue of Theorem 4.2 unless $`rg`$.
4.8 Theorem Let $`K`$ be an ambidextrous Hilbert space of support one in $`(\rho ^r,\rho ^{r+g})`$ where $`rg`$. Let $`F`$ and $`G`$ denote the restrictions of $`\times 1_{\rho ^g}`$ and $`1_{\rho ^g}\times `$ to $`๐ฆ`$ then there is a unique $`V(K^2,K^2)`$ such that
$$G(\psi )=V\widehat{F}(\psi ),\psi K.$$
$`V`$ is a multiplicative unitary.
The proof follows that of Theorem 4.2, bearing in mind that $`rg`$. If $`r>g`$, there is the possibility of starting with the ambidextrous Hilbert space $`K^n`$, for $`n`$ sufficiently large.
There is another interesting general result involving Hilbert spaces in $`๐ฏ_\rho `$.
4.9 Theorem Let $`K`$ be a Hilbert space of support one in $`(\rho ^r,\rho ^{r+g})`$ and define the associated category of Hilbert spaces $`๐ฆ`$ as above. Suppose that
$$(\rho ^r,\rho ^r)\times 1_{\rho ^{rg}}(K^r,K^r).$$
Then
$$(\rho ^{r+mg},\rho ^{r+ng})\times 1_{\rho ^{rg}}(K^{r+m},K^{r+n}).$$
Hence defining, for $`X(\rho ^{r+mg},\rho ^{r+ng})`$,
$$F(X)\psi :=X\times 1_{\rho ^{rg}}\psi ,\psi K^{r+m},$$
$`F`$ is a faithful โfunctor from the full subcategory of $`๐ฏ_\rho `$ whose objects are $`\rho ^{r+ng}`$, $`n_0`$ to the category $`๐ฆ`$ of Hilbert spaces and
$$F(X\times 1_{\rho ^{rg}})=F(X)\times 1_K.$$
Proof. $`K(\rho ^r,\rho ^{r+g})`$ implies $`K^n(\rho ^r,\rho ^{r+ng})`$ and $`K^n(\rho ^{r+mg},\rho ^{r+ng})K^m(\rho ^r,\rho ^r)`$. Since $`(\rho ^r,\rho ^r)\times 1_{\rho ^{rg}}(K^r,K^r)`$, tensor the above on the right with $`1_{\rho ^{rg}}`$ and compose on the right with $`K^r`$ and on the left with $`K^r`$ to conclude that
$$K^{r+n}(\rho ^{r+mg},\rho ^{r+ng})\times 1_{\rho ^{rg}}K^{r+m}K^r(K^r,K^r)K^r1_{\rho ^r}.$$
Since $`K^{r+m}`$ has support one, we conclude that
$$(\rho ^{r+mg},\rho ^{r+ng})\times 1_{\rho ^{rg}}(K^{r+m},K^{r+n}).$$
The remaining assertions are now obvious.
We show that the hypotheses of Theorem 4.9 with $`r=g=1`$ are fulfilled, when $`V`$ is a multiplicative unitary, considered as an object of $`(V)`$. In fact, $`V(V^2,\iota (V)V)`$, so $`V^{}(,\iota (V))\times 1_V`$ is a Hilbert space $`H`$ of support one in $`(V,V^2)`$. Using the condition for $`T(K,K)`$ to be an arrow of $`(V,V)`$, we get
$$T\times 1_VV^{}=V^{}\iota (T)\times 1_V,$$
showing that $`(V,V)\times 1_VHH`$, as required. The hypothesis of Theorem 4.3 does imply for the tensor unit $`\iota `$ that $`(\iota ,\iota )=`$.
Theorem 4.9 raises some interesting questions. Let $`๐ฏ_{r,g}`$ denote the full subcategory of $`๐ฏ_\rho `$ whose objects are of the form $`\rho ^{r+ng}`$ with $`n_0`$. Then we have โfunctors $`X1_{\rho ^s}\times X\times 1_{\rho ^{gs}}`$, $`0sg`$, defined on $`๐ฏ_{r,g}`$ through the ambient tensor category $`๐ฏ_\rho `$. Restricting the domain of the functors to the subcategory $`๐ฆ`$ and composing with the functor $`F`$ of Theorem 4.9 gives us โendofunctors of $`๐ฆ`$ taking $`K^m`$ to $`K^{r+m}`$. This raises the question of whether the composition with $`F`$ is necessary, or, more precisely, whether $`1_{\rho ^s}\times K\times 1_{\rho ^{gs}}(K,K^2)`$? In fact, we know that this is true by construction for $`s=0`$ and the basis of the definition of ambidextrous for $`s=g`$. When it is valid for some $`s>r`$ there is some analogue of a multiplicative unitary.
We have seen how two commuting shifts $`F`$ and $`G`$ on $`๐ฆ`$ cannot be interpreted as tensoring on the right and tensoring on the left with the object $`1`$ unless $`G=\widehat{F}`$ because the interchange law would fail to hold. On the other hand, we have, in this section, been using the interchange law in a tensor category to produce commuting shifts tied to multiplicative unitaries. We want, now, to show how this process can be reversed. We first describe a tensor category in terms of tensoring on the right and tensoring on the left. Let $`๐ฏ`$ be a category and give for each object $`\rho `$ endofunctors $`F_\rho `$ and $`G_\rho `$ such that for $`X(\mu ,\nu )`$ and $`Y(\pi ,\rho )`$,
$$F_\rho (X)G_\mu (Y)=G_\nu (Y)F_\pi (X),$$
where this identity defines $`X\times Y`$ and expresses the interchange law. It is understood to imply that $`F_\rho (\mu )=G_\mu (\rho )`$ for each pair $`\rho `$, $`\mu `$ of objects. The set of these endofunctors is supposed to commute pairwise and $`F_{F_\rho (\sigma )}=F_\rho F_\sigma `$ and $`G_{G_\mu (\nu )}=G_\mu G_\nu `$ implying that $`\times `$ is associative. If we further require that for some object $`\iota `$ $`F_\iota `$ and $`G_\iota `$ are the identity functors then $`๐ฏ`$ becomes a (strict) monoidal category with monoidal unit $`\iota `$. A functor $`J`$ between two such monoidal categories is a (strict) monoidal functor if for each object $`\rho `$ of $`๐ฏ`$,
$$JF_\rho =F_{J(\rho )}J,JG_\rho =G_{J(\rho )}J.$$
To have corresponding statements for tensor categories or tensor $`C^{}`$โcategories or tensor $`W^{}`$โcategories we need only add the obvious conditions that the functors involved be linear, โpreserving or normal as the case may be.
4.10 Proposition Let $`F`$ and $`G`$ be two commuting shifts with $`GG=\widehat{F}G`$, then given $`X(m,n)`$ and $`Y(p,q)`$,
$$F^{q+1}(X)G^{m+1}(Y)=G^{n+1}(Y)F^{p+1}(X).$$
Thus defining $`X\times ^{}Y:=F^{q+1}(X)G^{m+1}(Y)`$ gives a tensor $`W^{}`$โcategory $`๐ฆ^+`$ after adjoining an irreducible tensor unit $``$ with no arrows to any other object. If $`๐ฆ`$ is given the tensor structure determined by $`F`$ then there is a unique normal tensor โfunctor $`G^{}`$ from $`๐ฆ^+`$ to $`๐ฆ`$ such that $`G^{}(\psi ):=V^{}F(\psi )`$, $`\psi K`$, where $`V`$ is the multiplicative unitary associated with $`F`$ and $`G`$.
Proof. Write $`G^{m+1}=\widehat{F}^mG`$ and $`G^{n+1}=\widehat{F}^nG`$ and use the interchange law between $`F`$ and $`\widehat{F}`$. The remarks above show that $`๐ฆ^+`$ is a tensor $`W^{}`$โcategory and allow us to check that $`G^{}`$ is a tensor โfunctor using Proposition 3.1e.
Note that $`\times ^{}`$ is not addition on the objects of $`๐ฆ`$. Instead we have $`m\times n=m+n+1`$. However as we have adjoined a tensor unit $``$ to give $`๐ฆ^+`$, it is natural to renumber the objects by adding one and $`\times ^{}`$ is then addition on the objects of $`๐ฆ^+`$.
We now generalize some of our results so that we can work with tensor $`C^{}`$โcategories rather than just tensor $`W^{}`$โcategories. We begin with an object $`\rho `$ in a strict tensor $`C^{}`$โcategory and a Hilbert space $`K`$ in $`(\rho ,\rho ^2)`$ such that $`K\times 1_{\rho ^m}`$ has left annihilator zero for $`m_0`$. We can define $`K^n`$ and $`(K^m,K^n)`$ as at the beginnning of this section to get a $`C^{}`$โcategory $`๐ฆ`$. $`๐ฆ`$ will now not be a category of Hilbert spaces but just some subcategory. $`๐ฆ`$ is obviously an invariant subcategory for $`\times 1_\rho `$. The category $`๐ฆ`$ can be completed in an obvious way to give a category of Hilbert spaces $`\stackrel{~}{๐ฆ}`$, say, by identifying $`X(K^m,K^n)`$ with the corresponding linear map $`\varphi X\varphi `$, $`\varphi K^m`$. The concept of ambidextrous Hilbert space is dealt with in the following lemma.
4.11 Lemma Let $`K(\rho ,\rho ^2)`$ be a Hilbert space such that $`K\times 1_{\rho ^m}`$ has left annihilator zero for $`m_0`$. Let $`๐ฆ`$ be the $`C^{}`$โcategory defined above, then the following conditions are equivalent.
a) $`๐ฆ`$ is an invariant subcategory for $`1_\rho \times `$.
b) $`1_\rho \times K(K,K^2)`$.
c) $`K^n=1_{\rho ^{n1}}\times KK^{n1}`$, $`nN_0`$.
d) $`K^2=1_\rho \times KK`$.
Proof. If b) holds, then $`1_\rho \times \psi \psi ^{{}_{}{}^{}}\times 1_\rho `$ maps $`K^2`$ into $`K^2`$ if $`\psi ,\psi ^{}K`$. Hence $`_i1_\rho \times \psi _i\psi _i^{}\times 1_\rho `$ converges in, say, the $`s`$โtopology to a unitary arrow $`U`$ in $`\stackrel{~}{๐ฆ}`$, where the sum is taken over an orthonormal basis in $`K`$. But $`U\psi \times 1_\rho \psi ^{}=1_\rho \times \psi \psi ^{}`$ and this proves d). c) follows from d) by induction since
$$K\times 1_{\rho ^{n1}}1_{\rho ^{n2}}\times K=1_{\rho ^{n1}}\times KK\times 1_{\rho ^{n2}},$$
Composing on the right with $`K^{n2}`$ and using the induction hypothesis we obtain c). But just as $`๐ฆ`$ is invariant under $`\times 1_\rho `$, c) implies that it is invariant under $`1_\rho \times `$. But b) implies a), trivially.
We can now prove an analogue of Theorem 4.2.
4.12 Theorem Let $`\rho `$ be an object in a tensor $`C^{}`$โcategory and $`K`$ an ambidextrous Hilbert space in $`(\rho ,\rho ^2)`$ such that $`K\times 1_{\rho ^m}`$ has left annihilator zero for $`m_0`$. Let $`๐ฆ`$ be as above and $`\stackrel{~}{๐ฆ}`$ its completion to a category of Hilbert spaces, then the endofunctors on $`๐ฆ`$ determined by $`\times 1_\rho `$ and $`1_\rho \times `$ extend uniquely to commuting shifts $`F`$ and $`G`$ on $`\stackrel{~}{๐ฆ}`$ with $`G^2=\widehat{F}G`$ and there is a multiplicative unitary $`V(\stackrel{~}{K}_2,\stackrel{~}{K}_2)`$, such that $`G(\psi )=V\widehat{F}(\psi )`$, $`\psi K`$.
Proof. The functor $`\times 1_\rho `$ defines each $`K^n`$ as a tensor power of $`K`$ and hence there is a unique shift $`F`$ on $`\stackrel{~}{๐ฆ}`$ such that $`F(\psi _n)=\psi _n\times 1_\rho `$, for $`\psi _nK_n`$ and $`n_0`$. If $`X(K^m,K^n)`$, then
$$X\times 1_\rho F(\psi _m)\psi =X\times 1_\rho \psi _m\times 1_\rho \psi =F(X\psi _m)\psi .$$
Thus $`F(X)=X\times 1_\rho `$. In the same way, in view of Lemma 4.11, there is a unique shift $`G`$ on $`\stackrel{~}{๐ฆ}`$ with $`G(X)=1_\rho \times X`$. $`F`$ and $`G`$ obviously commute and, as in Theorem 4.2, we see that $`GG=\widehat{F}G`$. Thus there is a multiplicative unitary $`V`$ with the properties claimed.
The asymmetry in the formulation of Theorem 4.12 is only apparent: its hypotheses imply that $`1_{\rho ^m}\times K`$ has left annihilator zero for $`m_0`$.
## 5 Algebraic Endomorphisms of the Cuntz Algebra
As a preliminary to our main duality result, we present in this section results on Hilbert spaces in the Cuntz algebra and endomorphisms of that algebra that are โalgebraicโ with respect to the natural grading of the Cuntz algebra.
When $`K`$ is a finite dimensional Hilbert space, $`๐ช_K`$ will denote the Cuntz algebra, a simple unital $`C^{}`$โalgebra introduced by Cuntz. When $`K`$ is infinite dimensional, it denotes the extended Cuntz algebra, a simple $`C^{}`$โalgebra introduced in . These algebras are special cases of a more general construction needed in ยง6, where the Hilbert space is replaced by an object in a tensor $`C^{}`$โcategory. $`๐ช_K`$ has a $``$โgrading derived from the automorphic action $`\alpha `$ of $`๐`$ with $`\alpha _\lambda (\psi )=\lambda \psi `$, $`\lambda ๐`$. Thus the part of $`๐ช_K`$ of grade $`k`$ is given by
$$๐ช_K^k:=\{X๐ช_K:\alpha _\lambda (X)=\lambda ^kX,\lambda ๐\}.$$
It is this grading in the case of the Cuntz algebra $`๐ช_K`$ which will allow us to refer to certain Hilbert spaces in this algebra and endomorphisms of this algebra as being โalgebraicโ. Here $`K`$ can be a finite dimensional or infinite dimensional Hilbert space and in the latter case $`๐ช_K`$ is the extended Cuntz algebra introduced in . The results can be roughly summarized by saying that computations involving these algebraic objects reduce to problems involving linear operators between Hilbert spaces that can be identified a priori. These results will prove useful in other contexts when dealing with concrete endomorphisms on the Cuntz algebra.
The problem has its origins in the relation between the Cuntz algebra $`๐ช_K`$ and the tensor $`W^{}`$โcategory of bounded linear mappings between tensor powers of $`K`$. The Cuntz algebra is obtained from the category by factoring out the operation of tensoring on the right by $`1_K`$ whilst the operation of tensoring on the left is retained in the shape of the canonical endomorphism $`\rho _K`$. Then a direct sum is taken over the grading and finally the algebra is completed in the unique $`C^{}`$โnorm. This raises certain questions when proving results using the Cuntz algebra. A result may be purely algebraic in nature involving only the algebraic part of the Cuntz algebra. The manipulations may have been simplified by factoring out the operation of tensoring on the right by $`1_K`$. At the same time their significance may have been obscured by the asymmetric treatment of tensoring on the two sides.
The problems treated in this section illustrate both the analytic and the algebraic aspects of the problem and we have had more success with the analytic aspects proving that the set of solutions of these problems involves only the algebraic part.
If $`H`$ is a Hilbert space in a $`C^{}`$โalgebra $`๐`$, $`Y๐`$, and $`L^1(H)`$ denotes the trace class operators on $`H`$, then there is a unique continuous linear mapping $`T\text{Tr}_H(YT)`$ from $`L^1(H)`$ to $`๐`$ such that
$$\text{Tr}_H(Y\psi ^{}\psi ^{})=\psi ^{}Y\psi ^{},\psi ,\psi ^{}M.$$
The norm of this mapping is $`Y`$ and Tr$`{}_{H}{}^{}(YT)=`$Tr$`{}_{H}{}^{}(TY)`$. As the notation suggests, we are taking a partial trace relative to $`H`$.
5.1 Lemma. Let $`๐`$ be a $``$โgraded โsubalgebra of $`๐ช_K`$, i.e. $`๐`$ is generated by the subspaces $`๐^k:=๐๐ช_{K}^{}{}_{}{}^{k}.`$ Then $`๐^kH`$ for some $`k`$ and some Hilbert space $`H`$ in $`๐ช_K`$ of dimension $`>1`$ implies $`๐^{nk}=0`$ for $`n0`$.
Proof. If $`n`$, then $`๐^{nk}H^n`$ so it suffices to show $`๐^k=0`$. Let $`\psi ๐^k`$, then $`\psi \psi \psi ^{}๐^kH`$. Hence $`\psi ^{}\psi \psi \psi ^{}I`$, but $`\psi \psi ^{}`$ is a multiple of a minimal projection in $`(H,H)`$ so $`\psi =0`$.
In our applications, the $``$โgrading will be that introduced above and $`H=K^k`$. We next give results on computing the relative commutant of certain Hilbert spaces of support $`I`$ in the Cuntz algebra. This amounts to the same thing as determining the fixed points under the inner endomorphism generated by the Hilbert spaces in question. If $`\lambda `$ is an endomorphism of the Cuntz algebra, then $`(\lambda ,\lambda )`$ is just the relative commutant of $`\lambda (K)`$ and we give our result in a generality to include computing certain spaces of intertwining operators. We consider Hilbert spaces of support $`I`$ that are algebraic in the sense that they are contained in some $`(K^r,K^{r+g})`$. $`g`$ will be referred to as the grade of the Hilbert space and we shall only consider the case that $`g1`$. Indeed if $`K`$ is finiteโdimensional $`g0`$ and $`g=0`$ implies that the Hilbert space has dimension one. The minimal value for $`r`$ will be referred to as the rank of the Hilbert space. If $`K`$ is finite dimensional, then every such Hilbert space is of the form $`WK`$, where $`W(K^{r+1},K^{r+g})`$ is an isometry. In infinite dimensions, we need to consider coisometries $`W`$, too. As these Hilbert spaces have a fixed grade, the endomorphisms they generate commute with $`\alpha _\lambda `$ and their fixedโpoint algebras are graded $`C^{}`$โsubalgebras of $`๐ช_K`$. The basic observation is that if $`L`$ and $`M`$ are such Hilbert spaces of grade $`g`$ and rank $`q`$ and $`r`$, respectively, then
$$L^{}(K^m,K^{m+k})M(K^{m1},K^{m1+k}),mqk+g,r+g,$$
$$L^{}(K^m,K^{m+k})M(K^r,K^{r+k}),r+gm,qrk,$$
$$L^{}(K^m,K^{m+k})M(K^{qk},K^q),qk+gm,qrk.$$
This will be used in the following way. Pick $`\phi L`$ and $`\psi M`$ of norm $`1`$ and consider the linear mapping $`X\phi ^{}X\psi `$ of norm $`1`$ on $`๐ช_K`$. Let $`\mathrm{\Phi }`$ denote a limit point of iterates of such mappings in the pointwise weak operator topology of some locally normal representation of $`๐ช_K`$ on some Hilbert space $``$. A priori $`\mathrm{\Phi }`$ maps $`๐ช_K`$ into $`()`$, however since the subspaces of the form $`(K^m,K^n)`$ are weak operator closed in such representations, $`\mathrm{\Phi }`$ will map $`๐ช_K^k`$ into $`(K^r,K^{r+k})`$ or $`(K^{qk},K^q)`$ according as $`k(qr)`$ or $`k(qr)`$. Hence $`\mathrm{\Phi }`$ maps $`๐ช_K`$ into itself.
We now consider the following intertwining problem. Given a bounded linear mapping $`Y(M,L)`$ of norm $`1`$, find the set $`๐`$ of elements $`X`$ of $`๐ช_K`$ satisfying one of the following equivalent conditions
a) $`X\psi =Y\psi X,\psi M`$,
b) $`\psi ^{{}_{}{}^{}}X\psi =\psi ^{{}_{}{}^{}}Y\psi X,\psi M,\psi ^{}L`$,
c) $`\psi ^{{}_{}{}^{}}X=X\psi ^{{}_{}{}^{}}Y,\psi ^{}L`$,
d) $`X^{}\psi ^{}=Y^{}\psi ^{}X^{},\psi ^{}L`$,
e) $`X=Y\rho _M(X)`$.
Notice that as we have chosen Hilbert spaces $`L`$ and $`M`$ of equal grade, $`๐`$ is stable under the automorphisms $`\alpha _\lambda `$ defining the $``$โgrading. To compute $`๐`$ it therefore suffices to compute $`๐^k`$ for each $`k`$. The first step is to use an appropriate mapping $`\mathrm{\Phi }`$. If we can pick $`\phi L`$ and $`\psi M`$ of norm $`1`$ such that $`\phi ^{}Y\psi =I`$, then by b), the map $`X\phi ^{}X\psi `$ leaves $`๐`$ pointwise invariant, and letting $`\mathrm{\Phi }`$ be a limit point of iterates of this particular mapping, we conclude that
$$๐^k(K^r,K^{r+k}),k(qr),$$
$$๐^k(K^{qk},K^q),k(qr).$$
In general, we could define $`\mathrm{\Phi }`$ to be a pointwise weak operator limit point of $`X\phi _n^nX\psi _n^n`$, where the $`\phi _n`$ and $`\psi _n`$ are chosen of norm $`1`$ such that $`(\phi _n^{}Y\psi _n)^nI`$ as $`n\mathrm{}`$, this being possible since $`Y1`$ by assumption. But we want to go further and reduce the problem of computing $`๐^k`$ to a purely local problem.
5.2 Proposition Let $`L`$ and $`M`$ be algebraic Hilbert spaces of equal grade and rank $`q`$ and $`r`$, respectively and $`Y(M,L)`$ of norm $`1`$. Let $`๐`$ denote the set of $`X๐ช_K`$ such that
$$X\psi =Y\psi X,\psi M,$$
then if $`kqr`$, $`X๐^k`$ if and only if $`X(K^r,K^{r+k})`$ and one of the following equivalent conditions hold
a) $`X\vartheta (K^r,M)=Y\vartheta (K^{r+k},M)X`$.
b<sub>n</sub>) $`X\text{Tr}_{M^n}(TY^{\times n}\vartheta (K^r,M))=\text{Tr}_{L^n}(Y^{\times n}TY\vartheta (K^{r+k},M))X`$, $`TL^1(L^n,M^n).`$
Here $`L^1`$ is used to denote the set of trace class operators. If $`kqr`$, then $`X(K^{qk},K^q)`$ and we need only replace $`r`$ by $`qk`$ in the above.
Proof. We have already seen that $`X(K^r,K^{r+k})`$ if $`kqr`$ and a) now follows noting that:
$$X\vartheta (K^r,M)=Y\rho _M(X)\vartheta (K^r,M)=Y\vartheta (K^{r+k},M)X.$$
Conversely, a) implies $`X=Y\rho _M(X)`$ since $`X(K^r,K^{r+k})`$. Now a) also implies that
$$X\psi ^{{}_{}{}^{}}Y\vartheta (K^r,M)\psi =\psi ^{{}_{}{}^{}}Y\vartheta (K^{r+k},M)Y\psi X,\psi ^{}L,\psi M.$$
This is b<sub>1</sub>) for rank one operators $`T`$ and hence equivalent to b<sub>1</sub>). On the other hand, a) follows from b<sub>1</sub>) since $`M`$ and $`L`$ have support one. The same argument shows that b<sub>n</sub>) is equivalent to b<sub>n-1</sub>), completing the proof.
Let us comment on these conditions: a) is a simple canonical condition that already serves to make the basic point that $`๐^k`$ is determined by intertwining conditions between fixed tensor powers of $`K`$ and is in this sense algebraic. However the permutation operators map between higher tensor powers of $`K`$ than is really necessary if $`X(K^r,K^{k+r})`$ is to intertwine. By using the partial trace, we can reduce the powers of the tensor spaces involved at the cost of increasing the number of intertwining relations. In fact, for $`n`$ sufficiently large, the operators involved on the left hand side are in $`(K^r,K^r)`$ and those on the right hand side in $`(K^q,K^q)`$.
In concrete cases, the following strategy for computing $`๐^k`$ proves useful. Let $`X๐^k`$ and $`V(K,M)`$; in practice, $`V`$ can usually be picked unitary. Then
$$\vartheta (K^n,M)\rho ^n(V)=V\vartheta (K^n,K),n_0,$$
where we have written $`\rho `$ for $`\rho _K`$. Since $`X(K^r,K^{r+k})`$, $`X\rho ^r(V)=\rho ^{r+k}(V)X`$, and using a) of Proposition 5.2, we get
$$XV\vartheta (K^r,K)=YV\vartheta (K^{r+k},K)X.$$
If $`V`$ has a right inverse, we can, conversely, deduce a) of Proposition 5.2 from this equation. If wished, the permutations operators can be eliminated in favour of the endomorphism $`\rho `$. In fact, since $`\vartheta (K^{r+k},K)X=\rho (X)\vartheta (K^r,K)`$, we get
$$XV=YV\rho (X),$$
but this is best derived directly from d), above. Similarly, if $`U(K,L)`$, we may conclude that
$$UX=\rho (X)UY.$$
This is equivalent to a) of Proposition 5.2, if $`V`$ has a left inverse.
After these results, let us try and clarify whether more might be expected by relating the set of solutions to questions posed entirely in terms of a category of Hilbert spaces and hence independent of the identifications used to define the Cuntz algebra.
In place of $`๐ช_K`$ we consider a category $`๐ฆ`$ of Hilbert spaces with objects $`_0`$ and equipped with a shift $`F`$ to be thought of as the tensor powers of a Hilbert space $`K`$, as described in detail in Section 2. Instead of considering a Hilbert spaces $`L(K^q,K^{q+g})`$ and $`M(K^r,K^{r+g})`$, we consider another such category $``$ with a shift $`G`$ and two normal $``$โfunctors $`J`$ and $`J^{}`$ from $``$ to $`๐ฆ`$ such that $`JG=F^gJ`$, $`J^{}G=F^gJ^{}`$ and $`J(0)=q`$ and $`J^{}(0)=r`$. We consider natural transformations $`t(J^{},J)(\widehat{F},\widehat{F})`$. As we know from computations in Section 2, such a natural transformation is uniquely determined by $`t_0:=X(K^q,K^r)`$ satisfying
$$F^g(X)=\underset{j}{}J(\psi _j)XJ^{}(\psi _j)^{},$$
where the sum is taken over an orthonormal basis of $`M`$. If we write this in the Cuntz algebra we obtain our condition e), $`X=Y\rho _M(X)`$, where $`Y=_jJ(\psi _j)J^{}(\psi _j)^{}`$ in the Cuntz algebra and is unitary. There is no difficulty in generalizing to include cases where $`Y`$ is not unitary.
We learn from this that it is quite natural to expect solutions of grade $`rq`$. Furthermore, by composing $`J`$ or $`J^{}`$ with tensoring on the right by $`1_K`$, we can replace $`q`$ by $`q+1`$ or $`r`$ by $`r+1`$. Thus we have potential solutions for any grade. We see, therefore that the identifications involved in defining the Cuntz algebra mean that one problem at the level of the Cuntz algebra involves a countable set of problems at the level of $`๐ฏ_K`$.
We conclude that the results obtained using the spaces $`(K^m,K^n)`$ in the Cuntz algebra are the best that can be expected in complete generality. However, we now show how the estimates on the localization of $`๐^k`$ can be improved under conditions involving the relative localization of $`YM`$ and $`K^g`$ or $`Y^{}L`$ and $`K^g`$. Note that $`M^nK^{gn}(K^r,K^r)`$ for all $`n`$.
5.3 Lemma Let $`m`$ denote the smallest integer $`\frac{r}{g}`$ and $`\mathrm{}`$ the smallest integer $`\frac{q}{g}`$. If the weak operator closed linear span of the $`(L^{}Y)^nK^{gn}`$, $`nm`$, in $`(K^r,K^r)`$ contains $`I`$ then $`๐^kK^k`$ for $`kq`$ and $`๐^k(K^{qk},K^q)`$ for $`qkqr`$. Similarly, if the weak operator closed linear span of the $`(M^{}Y^{})^nK^{gn}`$, $`n\mathrm{}`$, in $`(K^q,K^q)`$ contains $`I`$ then $`๐^kK^k`$ for $`kr`$ and $`๐^k(K^r,K^{r+k})`$ for $`rkqr`$.
Proof. We know that $`๐^k(K^r,K^{r+k})`$ if $`kqr`$. But $`L^n`$ has rank $`q`$ and grade $`ng`$ and $`K^{ng}`$ has rank $`0`$ and grade $`ng`$. Thus from previous computations
$$L^n๐^kK^{ng}K^k,kq,ngr,$$
$$L^n๐^kK^{ng}(K^{qk},K^q),qkqr,ngr.$$
But if $`X๐`$,
$$L^nXK^{gn}=X(L^{}Y)^nK^{gn}.$$
Thus if the weak operator closed linear span of the $`(L^{}Y)^nK^{gn}`$ contains $`I`$, $`X`$ will be in the weak operator closed linear span of the $`L^nXK^{gn}`$ and the first part follows. The second part can be proved similarly or deduced from the first by using $`X^{}`$ and $`Y^{}`$ in place of $`X`$ and $`Y`$.
Recalling Lemma 5.1 at this point, we get the following corollary.
5.4 Corollary Suppose $`L=M`$ and $`Y`$ is a projection or a unitary. Let $`m`$ denote the smallest integer $`\frac{q}{g}`$ and suppose the weak operator closed linear span of the $`(L^{}Y)^nK^{gn}`$, $`nm`$, in $`(K^r,K^r)`$ contains $`I`$, then $`๐^k=0`$ for $`kq`$ and $`kq`$ and $`๐^k(K^{qk},K^q)`$ for $`qk0`$.
Proof. When $`Y`$ is a projection, we need only remark that $`๐`$ is a $``$โgraded โsubalgebra of $`๐ช_K`$. If $`Y`$ is unitary, then $`X๐^k`$ implies $`XXX^{}๐^k`$ and this is all that is used in Lemma 5.1.
To give a simple example, $`\rho ^r(K)`$ is an algebraic Hilbert space of grade one and rank $`r`$. Its relative commutant is $`(\rho ^r,\rho ^r)`$ which was shown in , using techniques similar to those above, to be equal to $`(K^r,K^r)`$. In this case, $`\rho ^r(K^n)K^n`$ is the space of finite rank operators on $`K^r`$ for $`nr`$ and the space of compact operators from $`\rho ^{rn}(K^n)`$ to $`K^n`$ if $`1nr`$.
The theory of multiplicative unitaries provides us with further examples of algebraic Hilbert spaces $`L`$ and $`M`$ of equal grade, where the weak operator closure of $`L^{}M`$ and hence of $`M^{}L`$ contains $`I`$. For example, if $`V(K^2,K^2)`$ is a regular multiplicative unitary then the weak operator closures of $`K^{}VK`$, $`K^{}\vartheta V\vartheta K`$ and $`K^{}V\vartheta K`$ in $`(K,K)`$ are even โalgebras containing the unit . Of course, $`K`$ could be replaced here by any other algebraic Hilbert space of support $`I`$. Note that if $`L_i`$ and $`M_i`$ are algebraic Hilbert spaces such that $`I`$ is in the weak operator closure of $`L_i^{}M_i`$, $`i=1,2`$, then $`I`$ is also in the weak operator closure of $`L_1^{}L_2^{}M_2M_1`$.
Let us now return to the special case used to motivate our basic intertwining relation. If we take $`Y=1_M`$ then $`๐`$ is just the relative commutant of $`M`$ and it is of interest to ask when $`M`$ has trivial relative commutant. $`๐^0`$ will reduce to the complex numbers by b<sub>1</sub>) of Proposition 5.2 if $`M^{}\vartheta (K^r,M)M`$ has trivial commutant in $`(K^r,K^r)`$. We again have examples with $`r=1`$ drawn from the theory of a regular multiplicative unitary and this leads to the following result.
5.5 Proposition Let $`V`$ be a regular multiplicative unitary in $`K^2`$, then the following Hilbert spaces have trivial relative commutant in $`๐ช_K`$: $`V^{}K`$ and $`\vartheta V\vartheta K`$. In the case of the Hilbert spaces $`\vartheta VK`$ and $`V^{}\vartheta K`$, the relative commutants are the commutants of $`K^{}VK`$ and $`K^{}\vartheta V^{}\vartheta K`$ in $`(K,K)`$, respectively.
Proof. We need only remark that in each case we know that $`๐^k=0`$ for $`k1`$. Furthermore, $`๐^0`$ is, in each case, as claimed since, for a Hilbert space of the form $`UK`$ with $`U(K^2,K^2)`$ unitary, b<sub>1</sub> of Proposition 5.2 just reduces to saying that $`๐^0`$ is the commutant of the first component of $`U\vartheta `$.
Following , we denote $`K^{}VK`$ and $`K^{}\vartheta V^{}\vartheta K`$ by $`๐(V)`$ and $`\widehat{๐}(V)`$, respectively. If $`V`$ is a regular multiplicative unitary, these algebras are actually โalgebras .
We now come to the second application we had in mind, namely to study intertwiners between certain endomorphisms of the Cuntz algebra. We say that an endomorphism $`\tau `$ has grade $`g`$ if $`\tau (K)`$ has grade $`g+1`$ and is algebraic of rank $`r`$ if $`\tau (K)`$ is algebraic of rank $`r`$. There is a unique unitary $`V`$ such that $`\tau (\psi )=V\psi `$, for $`\psi K`$ and $`\tau `$ has rank $`r`$ if and only if $`r`$ is the smallest integer such that $`V(K^{r+1+g},K^{r+1+g})`$. If $`\sigma `$ is an algebraic endomorphism of grade $`f`$ and rank $`q`$ then the composition $`\sigma \tau `$ is of grade $`f+g`$ and rank $`q+r+g+(r+g)f`$.
Now suppose that we have endomorphisms $`\sigma `$ and $`\tau `$ as above of equal grade $`g`$, then the space of intertwiners $`(\tau ,\sigma )`$ is $``$โgraded and if $`X(\tau ,\sigma )`$, then
$$X\psi =Y\psi X,\psi \tau (K),$$
where $`Y(\tau (K),\sigma (K))`$ is the unitary taking $`\tau (\psi )`$ to $`\sigma (\psi )`$ for each $`\psi K`$. Thus the analysis of Proposition 5.2 holds. In particular, we have
$$(\tau ,\sigma )^k(K^r,K^{r+k}),kqr,$$
$$(\tau ,\sigma )^k(K^{q+k},K^q),kqr.$$
## 6 An Algebraic Version of TakesakiโTatsuuma Duality
After these results on algebraic Hilbert spaces and endomorphisms, our aim is to describe an algebraic model for a dual of a multiplicative unitary. We present a duality result for โlocally compactโ multiplicative unitaries in terms of the $`C^{}`$โalgebra generated by the regular representation considered as an object in the tensor $`C^{}`$โcategory of representations of the multiplicative unitary. Thus we get, in particular, an algebraic version of a duality result for the representation categories of locally compact groups.
We recall that $`๐ช_H`$ is a simple $`C^{}`$โalgebra and every unitary operator $`X(H,H^{})`$ extends to a unital morphism $`๐ช_H๐ช_H^{}`$, which we denote by $`\lambda _X`$.
Let $`V`$ be a regular multiplicative unitary acting on $`K^2`$ and $`W`$ a representation contained in $`M((H,H)๐(V)),`$ the multiplier algebra of the minimal tensor product $`(H,H)๐(V).`$ Let us identify $`๐(V)\rho _K(๐ช_H)=๐(V)๐ช_H.`$ Then
$$\lambda _{\vartheta _{H,K}W}(H^r,H^s)๐(V)+๐(V)\lambda _{\vartheta _{H,K}W}(H^r,H^s)๐(V)\rho _K(H^r,H^s).$$
Here in the definition of $`\lambda _{\vartheta _{H,K}W}`$ we consider $`H`$ and $`K`$ as Hilbert spaces of support $`I`$ in some $`()`$ and regard $`\vartheta _{H,K}W`$ as mapping $`H`$ onto $`\vartheta _{H,K}WH`$. The arguments of or ; Section 6 generalize to show that the monomorphism $`\lambda _{\vartheta _{H,K}W}:๐ช_HM(๐(V)๐ช_H)`$ defines a coaction of $`๐(V)`$ on $`๐ช_H,`$ i.e. $`\lambda _{\vartheta _{H,K}W}`$ is a unital โhomomorphism satisfying
$$\delta i\lambda _{\vartheta _{H,K}W}=i\lambda _{\vartheta _{H,K}W}\lambda _{\vartheta _{H,K}W},$$
where $`\delta `$ is the coproduct of $`๐(V)`$ induced by the adjoint action of$`\vartheta _{K,K}V`$ . The corresponding fixed point algebra is
$$๐ช^W=\{T๐ช_H:\lambda _{\vartheta _{H,K}W}(T)=\rho _K(T)\}.$$
Direct computations show that $`๐ช^W(H^r,H^s)=(W^{\times r},W^{\times s})`$. We recall that any object $`W`$ in a tensor $`C^{}`$โcategory $`๐ฏ`$ has a canonically associated $`C^{}`$โalgebra $`๐ช_W`$ with a unital endomorphism $`\rho _W`$. This construction can be used in two quite different ways. On the one hand it provides one with a large class of model endomorphisms with rather well defined properties. In favourable cases, the associated โfunctor $`F_W`$ from $`๐ฏ_W`$, the tensor $`C^{}`$โcategory whose objects are the tensor powers of $`W`$ with arrows taken from $`๐ฏ`$, to the tensor $`C^{}`$โcategory whose objects are the powers of $`\rho _W`$ and whose arrows are intertwining operators is even an isomorphism. This illustrates the second aspect of the construction that it also encodes properties of the object $`W`$ in question. $`W`$ is $`C^{}`$โamenable when $`F_W`$ is an isomorphism, in the terminology of , where related notions of amenability are discussed.
Now $`๐ฏ_W`$ carries an automorphic action of the circle group $`๐`$ defined by
$$\alpha _\lambda (T):=\lambda ^{sr}T,T(W^r,W^s),$$
and the induced automorphic action of $`๐`$ on $`(๐ช_W,\rho _W)`$ is also denoted by $`\alpha `$. The spectral subspaces of the action make $`๐ช_W`$ into a $``$โgraded $`C^{}`$โalgebra:
$$๐ช_{W}^{}{}_{}{}^{k}=\{T๐ช_W:\alpha _\lambda (T)=\lambda ^kT,\lambda ๐\}.$$
The construction is functorial so, given a โfunctor $`F`$ from $`๐ฏ`$, it yields a morphism $`F_{}:๐ช_W๐ช_{F(W)}`$ of $`C^{}`$โalgebras intertwining the canonical endomorphisms and the actions of $`๐`$. In particular, if we have a faithful functor into a tensor $`C^{}`$โcategory of Hilbert spaces as is the case for the categories $`(V)`$ or $`๐(V),`$ then it yields an inclusion $`๐ช_W๐ช_H`$ of $`C^{}`$โalgebras such that $`\rho _H๐ช_W=\rho _W.`$ Here $`H=F(W)`$ is the Hilbert space of $`W`$. If $`๐ช_W`$ has trivial relative commutant in $`๐ช_H`$ then the group of automorphisms of $`๐ช_H`$ leaving $`๐ช_W`$ pointwise fixed can be identified with
$$G_W:=\{U๐ฐ(H):TU^{\times r}=U^{\times s}T,T(W^{\times r},W^{\times s}),r,s\}.$$
Furthermore,
$$(\rho _W^r,\rho _W^s)=(H^r,H^s)๐ช_W.$$
Returning to the fixed point algebra under the above action, we have $`๐ช_W๐ช^W.`$ As in the case of a group action, equality follows from an amenability condition on $`V.`$
6.1 Theorem Let $`WM((H,H)๐(V))`$ be a unitary representation of $`V`$ and suppose that there is an invariant mean $`m`$ on $`(๐(V),\delta ).`$ Then there is a conditional expectation $`E:๐ช_H๐ช^W`$ satisfying $`E(H^r,H^s)=(W^{\times r},W^{\times s}).`$ In particular, $`๐ช_W=๐ช^W.`$
Proof. We extend $`m`$ to the multiplier algebra of $`๐(V)`$ via strict continuity. Let $`\omega `$ be a normal state of some faithful representation $`\pi `$ of $`๐ช_{}`$ where $`\pi (H)`$ has support $`I`$. Then $`i\omega `$ induces a strictly continuous positive map from $`M(๐(V)๐ช_H)`$ to $`M(๐(V)),`$ and setting
$$\omega (E(T)):=mi\omega \lambda _{\vartheta _{H,K}W}(T),T๐ช_H,$$
gives a positive map $`E`$ of norm one from $`๐ช_H`$ to $`(_\pi )`$ satisfying $`E(AT)=AE(T),A๐ช^W,T๐ช_H.`$ Now the arguments of ; Proposition 6.5, except that $`\rho _K(H^s)^{}\lambda _{\vartheta _{H,K}W}(H^r,H^s)\rho _K(H^r)M(๐(V))`$ here, show that $`E`$ is the desired conditional expectation.
In particular, if $`V`$ is compact and $`TK`$ is a fixed normalized vector then $`m(A)I=T^{}AT,A๐(V),`$ is the unique Haar measure on $`๐(V),`$ the conditional expectation corresponding to the representation $`W`$ is $`E(X)=T^{}\lambda _{\vartheta _{H,K}W}(X)T`$.
6.2 Theorem Let $`V`$ be a regular multiplicative unitary, then $`V`$ is $`C^{}`$โamenable as an object of $`(V)`$, i.e.
$$(\rho _V^r,\rho _V^s)=(V^r,V^s),r,s_0.$$
Proof. The pentagon equation expressing the multiplicativity of $`V`$ is equivalent to $`V^{}K(V,V^2)`$. $`V^{}K`$ has trivial relative commutant, by Proposition 5.5, if $`V`$ is regular. Thus $`๐ช_V^{}๐ช_K=`$ and, consequently,
$$(\rho _V^r,\rho _V^s)=(K^r,K^s)๐ช_V.$$
On the other hand, a computation, cf. ยง6 of , shows that
$$(V^r,V^s)=(K^r,K^s)๐ช^V,$$
where $`๐ช^V=\{X๐ช_K:\lambda _{\vartheta V}(X)=\rho _K(X)\}`$.
It is not clear whether $`V`$ is $`C^{}`$โamenable as an object of $`๐(V)`$, when $`V`$ is regular. The analogous proof does not work. The pentagon equation is now equivalent to
$$V\rho _K(K)=V\vartheta K(V,V^2).$$
But if $`V`$ is regular, the relative commutant of $`V\vartheta K`$ is, by Proposition 5.5, the commutant of $`K^{}\vartheta V\vartheta K`$, the first component of $`V`$, in $`(K,K)`$. This difference between $`๐(V)`$ and $`(V)`$ relates to the alternative definition of tensor product pointed out in the introduction. With the alternative tensor product for $`๐(V)`$, we would find $`\vartheta V\vartheta K(V,V^2)`$ and $`\vartheta V\vartheta K`$ does have trivial relative commutant in $`๐ช_K`$.
In virtue of Theorem 6.2, the model endomorphism $`(๐ช_V,\rho _V)`$ is a natural candidate for a dual of $`V`$. So we pose the following question. When does a pair $`(๐,\widehat{\rho })`$ consisting of a unital $`C^{}`$โalgebra and a unital endomorphism arise from a system of the form $`(๐ช_V,\rho _V)`$?
At the same time we have seen that the Cuntz algebra allows a simple description of a large variety of interesting model endomorphisms, giving rise to systems of the form $`(๐ช_K,\lambda _R)`$, where $`๐ช_K`$ is the (extended) Cuntz algebra over the Hilbert space $`K`$ and $`\lambda _R`$ the algebraic endomorphism determined by the unitary operator $`R`$, $`R(K^{r+1},K^{r+1+g})`$ for an algebraic endomorphism of grade $`g`$ and rank $`r`$. In fact, we would like our model systems to combine two features: they should be of the form $`(๐ช_\rho ,\widehat{\rho })`$, where $`\rho `$ is an object in a tensor $`C^{}`$โcategory. This means that $`๐ช_\rho `$ is generated by the intertwiners between the powers of $`\widehat{\rho }`$. The second feature is that there should be a Hilbert space $`K`$ of intertwiners between powers of $`\widehat{\rho }`$.
Now the Cuntz algebra and the extended Cuntz algebra $`๐ช_K`$ are derived from the tensor $`W^{}`$-category of Hilbert spaces whose objects are the tensor powers of $`K`$. If we begin, as in previous sections, simply with a $`W^{}`$โcategory of Hilbert spaces $`๐ฆ`$ we need a shift $`F`$ to give $`๐ช_K`$. Giving a second commuting shift $`G`$ amounts to giving an algebraic endomorphism $`\lambda _R`$ of grade zero and rank one. Here $`R(K^2,K^2)`$ is determined by $`G(\psi )=RF(\psi )`$, $`\psi K`$. Natural transformations yield intertwiners of endomorphisms: more precisely, if $`t(G^r,G^s)(\widehat{F}^r,\widehat{F}^s)`$ then $`t_0(\lambda _R^r,\lambda _R^s)`$ but it is not clear whether all intertwiners arise in this way.
Looking at Theorem 4.4 in this light gives us the following result.
6.3 Theorem Let $`V`$ be a multiplicative unitary on $`K^2`$, viewed as an element of $`๐ช_K`$. Let $`H:=V^{}K`$ and let $`\lambda ^{}`$ be the induced isomorphism of $`๐ช_K`$ onto $`๐ช_H๐ช_K`$, then $`\rho \lambda ^{}=\lambda ^{}\lambda _R`$, where $`R=V\theta `$ and $`\rho `$ is the canonical endomorphism of $`๐ช_K`$.
Remark In fact, $`๐ช_H=๐ช_V`$ since $`(V^r,V^s)\times 1_K(H^r,H^s)`$ for $`r,s>0`$ but this still leaves the finer details open on how the intertwiner spaces $`(V^r,V^s)`$ sit in $`๐ช_H`$.
We now take up the situation of Theorem 4.9. Thus we suppose that $`๐ฏ_\rho `$ is a tensor $`W^{}`$โcategory whose objects are the powers of $`\rho `$ with a Hilbert space $`K`$ of support one in $`(\rho ^r,\rho ^{r+g})`$ and suppose that $`(\rho ^r,\rho ^r)\times 1_{\rho ^{rg}}(K^r,K^r)`$. We now compare $`C^{}`$โalgebra $`๐ช_\rho `$ with the Cuntz algebra $`๐ช_K`$. Obviously, in constructing $`๐ช_K`$, we use only spaces of arrows between objects of the form $`\rho ^{r+ng}`$ with $`n_0`$. But these spaces define the $`C^{}`$โalgebra $`๐ช_{\rho ^g}`$. If we use $`\widehat{\rho }`$ to denote the endomorphism of $`๐ช_{\rho ^g}`$ induced by tensoring on the left by $`1_\rho `$, then the estimates of Theorem 4.9 show that $`\widehat{\rho }(K)(K^{r+1},K^{r+2})`$ in $`๐ช_{\rho ^g}`$ and that $`๐ช_{\rho ^g}`$ can be identified canonically with $`๐ช_K`$. Hence there is a unitary $`R(K^{r+2},K^{r+2})`$ such that $`R\psi =\widehat{\rho }(\psi )`$, $`\psi K`$. In other words $`\widehat{\rho }`$ can be identified with $`\lambda _R`$, an algebraic endomorphism of grade zero and rank $`r+1`$. This estimate on the rank is only an upper bound. In fact, if $`K`$ is ambidextrous, then $`\widehat{\rho }^g(K)(K,K^2)`$. We summarize the discussion as follows.
6.4 Proposition Let $`\rho `$ be an object in a tensor $`W^{}`$โcategory and $`K`$ a Hilbert space of support one in $`(\rho ^r,\rho ^{r+g})`$ with $`(\rho ^r,\rho ^r)\times 1_{\rho ^{rg}}(K^r,K^r)`$. Then $`(๐ช_{\rho ^g},\widehat{\rho })=(๐ช_K,\lambda _R)`$, where $`R(K^{r+2},K^{r+2})`$ is the unitary operator in $`๐ช_\rho `$ such that $`R\psi =\widehat{\rho }(\psi )`$, for all $`\psi K`$.
Returning to the question of when a pair $`(๐,\widehat{\rho })`$ consisting of a unital $`C^{}`$โalgebra and a unital endomorphism arises from a system of the form $`(๐ช_V,\rho _V)`$, where $`V`$ is a multiplicative unitary, we shall see that the following conditions are necessary and sufficient:
a) $`๐`$ is generated by a tensor $`W^{}`$โcategory which is a tensor subcategory of the category of intertwiners between the powers of $`\widehat{\rho }`$. The generating object will be denoted by $`\rho `$;
b) there is a Hilbert space of support $`I`$, $`K(\rho ,\rho ^2)`$;
c) $`\rho (K)(K,K^2)`$;
d) $`(\rho ,\rho )(K,K)`$.
We shall see in Theorem 6.11 that $`๐ช_\rho `$ is simple, so, in virtue of a), $`(๐,\widehat{\rho })`$ is the system $`(๐ช_\rho ,\widehat{\rho })`$ associated with the object $`\rho `$ of the tensor $`W^{}`$โcategory.
We recognize that c) just says that $`\lambda _R`$ of Proposition 6.4 has grade zero and rank $`1`$, or equivalently that $`R(K^2,K^2)`$. b), on the other hand, tells us that $`\lambda _R\lambda _R=\rho _K\lambda _R`$, or, equivalently that $`V:=R\vartheta `$ is a multiplicative unitary. If $`V`$ is even regular then c) can be strengthened to $`K^{}\rho (K)=KK^{}`$.
Notice that, as a consequence of Theorem 6.3 and the following remark, $`(๐ช_V,\rho _V)`$ satisfies a) to d) above relative to the Hilbert space $`H:=V^{}K`$.
As Theorem 6.3 establishes the isomorphism of $`(๐ช_V,\rho _V)`$ and $`(๐ช_K,\lambda _R)`$, $`(๐ช_K,\lambda _R)`$ satisfies a) to d), above, too. Indeed, except for d), this is easily seen directly. However, $`(\lambda _R,\lambda _R)`$ being just the relative commutant of the Hilbert space $`\lambda _R(K)=RK`$, we do get $`(\lambda _R,\lambda _R)(K,K)`$ by Corollary 5.4 if $`V`$ is regular. We now prove our first duality result.
6.5 Theorem Let $`(๐,\widehat{\rho })`$ satisfy a) to d) then there is a unique multiplicative unitary $`V`$ on the Hilbert space $`K^2,`$ such that $`(๐,\widehat{\rho },K)=(๐ช_V,\rho _V,V^{}K),`$ where $`V`$ is regarded as a representation of $`V`$. Two systems of the form $`(๐ช_V,\rho _V,V^{}K)`$ for multiplicative unitaries $`V`$ on $`K^2`$ are isomorphic if and only if the multiplicative unitaries are equivalent.
Proof. By Proposition 6.4, there is a unitary $`V๐`$ such that $`(๐,\widehat{\rho },K)=(๐ช_K,\lambda _{V\vartheta _{K,K}},K)`$. But c) implies that $`V(K^2,K^2)`$ and b) may be read as $`\widehat{\rho }^2=\rho _K\widehat{\rho }`$ so that $`V`$ is a multiplicative unitary on $`K^2`$ ,. By Theorem 6.3 again, we can consider isomorphisms of systems of the form $`(๐ช_K,\lambda _R,K)`$. So if $`\tau `$ is an isomorphism from $`๐ช_K`$ to $`๐ช_K^{}`$ with $`\tau (K)=K^{}`$ and $`\tau \lambda _R=\lambda _R^{}\tau `$. Then $`\tau (R)=R^{}`$ and, since $`\tau (\vartheta _{K,K})=\vartheta _{K^{},K^{}}`$, $`\tau (V)=V^{}`$. So if $`U`$ is the unitary from $`K`$ to $`K^{}`$ such that $`\tau (\psi )=U\psi `$, for $`\psi K`$, its second tensor power will intertwine $`V`$ and $`V^{}`$, realizing the desired equivalence. The converse is obvious.
In the case of a regular multiplicative unitary we can even obtain a categorical rather than an algebraic duality theorem. The following result characterizes tensor $`W^{}`$โcategories of the form $`๐ฏ_V`$ for a regular multiplicative unitary $`V`$.
6.6 Theorem Given a tensor $`W^{}`$โcategory $`๐ฏ_\rho `$ whose objects are the tensor powers of a $`C^{}`$-amenable object $`\rho `$, suppose there is an ambidextrous Hilbert space $`K`$ of support one in $`(\rho ,\rho ^2)`$ with $`(\rho ,\rho )\times 1_\rho (K,K)`$ and
$$K^{}\times 1_\rho 1_\rho \times K=KK^{}$$
then the unitary $`V`$ on $`K^2`$ defined by
$$V\psi _1\times 1_\rho \psi _2=1_\rho \times \psi _2\psi _1,\psi _1,\psi _2K,$$
is multiplicative and $`๐ฏ_\rho `$ and $`๐ฏ_V`$ are isomorphic.
Proof. The fact that $`V`$ is multiplicative follows from Theorem 4.2 and the condition $`K^{}\times 1_\rho 1_\rho \times K=KK^{}`$ just says that $`K`$ is regular. We now consider the image of $`๐ฏ_\rho `$ in $`๐ช_\rho `$ under the canonical map, then the conditions a) to d) above are satisfied by $`๐ช_\rho `$ and we conclude from Theorem 6.5 that $`(๐ช_\rho ,\widehat{\rho },K)=(๐ช_V,\rho _V,V^{}K)`$. Now $`\rho `$ is $`C^{}`$โamenable by assumption and $`V`$ is $`C^{}`$โamenable by Theorem 6.2 and, with an obvious notation, $`๐ฏ_{\widehat{\rho }}`$ and $`๐ฏ_{\rho _V}`$ are isomorphic hence $`๐ฏ_\rho `$ and $`๐ฏ_V`$ are isomorphic.
We now want to give an algebraic characterization of the situation where the multiplicative unitary $`V`$ on $`K^2`$ is endowed with a standard braided symmetry $`\epsilon `$, a concept explained in the appendix. In this case the system $`(๐ช_{\widehat{V}},\rho _{\widehat{V}},\widehat{V}^{}K),`$ with $`\epsilon =V\vartheta _{K,K}\widehat{V},`$ satisfies conditions a) to d). As pointed out in the appendix, $`(๐ช_{\widehat{V}},\rho _{\widehat{V}})=(๐ช_V,\rho _V),`$ where $`V`$ is the regular corepresentation and $`\widehat{V}`$ the regular representation. A direct computation shows that the condition for being standard, namely that $`\widehat{V}`$ and $`V_{23}`$ commute on $`(K^3,K^3),`$ is equivalent to $`\rho _V(\epsilon \psi )=\rho _H(\epsilon \psi )`$, $`\psi H:=\widehat{V}^{}K`$. We start with the following result.
6.7 Theorem Let $`(๐,\rho )`$ be a pair consisting of a $`C^{}`$โalgebra and a unital endomorphism satisfying a) to d). Let $`\epsilon `$ be a braided symmetry for $`\rho `$ satisfying
e) $`\rho (\epsilon \psi )=\rho _K(\epsilon \psi ),\psi K.`$
Then there is a multiplicative unitary $`V`$ on a Hilbert space $`H`$ whose regular corepresentation $`V`$ is endowed with a standard braided symmetry $`\epsilon _V`$ of $`V`$ and an isomorphism $`\mathrm{\Phi }:๐๐ช_V`$ such that
1) $`\mathrm{\Phi }\rho =\rho _V\mathrm{\Phi };`$
2) $`\mathrm{\Phi }(\epsilon )=\epsilon _V;`$
3) $`\mathrm{\Phi }(K)=\widehat{V}^{}H,`$ where $`\epsilon _V=V\vartheta \widehat{V}.`$
The multiplicative unitary $`V`$ is determined, up to equivalence, by the above conditions. If $`\epsilon `$ is a permutation symmetry then $`(๐,\rho )`$ correponds to a locally compact group $`G`$ and $`\epsilon `$ to the usual permutation symmetry.
Proof. By Theorem 6.5 we can find a regular representation $`\widehat{V}`$ acting on a Hilbert space $`H`$ such that the triples $`(๐,\rho ,K)`$ and $`(๐ช_{\widehat{V}},\rho _{\widehat{V}},\widehat{V}^{}H)`$ are isomorphic via an isomorphism $`\mathrm{\Phi }.`$ If we write $`\epsilon _V:=\mathrm{\Phi }(\epsilon )=V\vartheta \widehat{V}`$ then by e) $`V_{23}`$ and $`\widehat{V}`$ commute on $`H^3,`$ so as shown in the appendix, $`V`$ is multiplicative and $`\epsilon _V`$ is a standard braided symmetry of $`V.`$ It will follow from the next proposition that $`V`$ is unique up to equivalence. If $`\epsilon `$ is a permutation symmetry then, by Proposition A.4, $`V`$ is cocommutative, thus coming from a locally compact group $`G`$ .
6.8 Proposition. Let $`V`$ and $`V^{}`$ be multiplicative unitaries on Hilbert spaces $`H^2`$ and $`H_{}^{}{}_{}{}^{2}`$ endowed with standard braided symmetries $`\epsilon _V=V\vartheta _{H,H}\widehat{V}`$ and $`\epsilon _V^{}=V^{}\vartheta _{H^{},H^{}}\widehat{V^{}}`$ respectively. If there is an isomorphism $`\mathrm{\Phi }:๐ช_V๐ช_V^{}`$ satisfying
a) $`\mathrm{\Phi }\rho _V=\rho _V^{}\mathrm{\Phi };`$
b) $`\mathrm{\Phi }(\epsilon _V)=\epsilon _V^{};`$
c) $`\mathrm{\Phi }(K)=K^{},`$ where $`K=\widehat{V}^{}H`$ and $`K^{}=\widehat{V^{}}^{}H^{},`$
then $`V`$ and $`V^{}`$ are unitarily equivalent.
Proof. By c), $`\mathrm{\Phi }\rho _K=\rho _K^{}\mathrm{\Phi }`$ on $`๐ช_V`$. Thus by a) and b) $`\mathrm{\Phi }\lambda _{\widehat{V}^{}}(V)=\lambda _{\widehat{V^{}}^{}}(V^{})`$ since $`\lambda _{\widehat{V}^{}}(V)=\rho _V(\epsilon _V^{})\rho _K(\epsilon _V)`$ with a similar result for $`V^{}.`$ Now by c) there is a unitary operator $`U:HH^{}`$ extending to an isomorphism $`\alpha :๐ช_H๐ช_H^{}`$ such that $`\mathrm{\Phi }\lambda _{\widehat{V}^{}}=\lambda _{\widehat{V^{}}^{}}\alpha `$, on $`๐ช_H`$. It follows that $`\alpha (\widehat{V})=\widehat{V^{}}.`$ Let us define $`\stackrel{~}{\epsilon }_V=\widehat{V}\epsilon _V\widehat{V}^{}=\widehat{V}V\vartheta _{H,H}.`$ Then $`\epsilon _V=\lambda _{\widehat{V}^{}}(\stackrel{~}{\epsilon }_V)`$ since $`(V^{\times 2},V^{\times 2})`$ is generated by $`K(V,V)K^{}`$ as a weakly closed subspace and $`\widehat{V}`$ commutes with $`(V,V).`$ If we define the operator $`\stackrel{~}{\epsilon }_V^{}`$ corresponding to $`V^{},`$ as above, then we deduce $`\alpha (\widehat{V}V\vartheta _{H,H})=\widehat{V^{}}V^{}\vartheta _{H^{},H^{}},`$ from b) and the previous relation intertwining $`\mathrm{\Phi }`$ and $`\alpha `$. Clearly $`\alpha (\vartheta _{H,H})=\vartheta _{H^{},H^{}}`$, so $`\alpha (V)=V^{}`$. Now the adjoint action of $`U\times U`$ on $`(H^2,H^2)`$ implements $`\alpha `$, completing the proof.
Remark. We can now complement the discussion following Theorem 6.2. With a standard braided symmetry, $`๐(V)`$ and $`(V)`$ are isomorphic as tensor $`W^{}`$โcategories embedded in Hilbert spaces. Hence by Theorem 6.2, a $`V`$ with a standard braided symmetry and $`\widehat{V}`$ regular is $`C^{}`$โamenable in $`๐(V)`$. Indeed $`\widehat{V}^{}K(V,V^2)`$ has trivial relative commutant in $`๐ช_K`$.
We now examine model endomorphisms which are more $`C^{}`$โalgebraic in nature. In fact, when $`K`$ is infinite dimensional, the above systems are not really $`C^{}`$โalgebraic in nature since $`๐ช_K`$ is the norm closure of subspaces endowed with a $`W^{}`$โtopology and is not even separable when $`K`$ is an infinite dimensional separable Hilbert space. To cure this defect, we might replace $`(๐ช_K,\lambda _R)`$ by $`(๐ซ_K,\tau _R)`$, where $`๐ซ_K`$ is the smallest $`C^{}`$โsubalgebra of $`๐ช_K`$ containing $`K`$ and stable under $`\lambda _R`$ and $`\tau _R`$ denotes the restriction of $`\lambda _R`$ to $`๐ซ_K`$. Note that, since $`K๐ซ_K`$,
$$(\tau _R^m,\tau _R^n)=(\lambda _R^m,\lambda _R^n)๐ซ_K.$$
At the same time, we would now like our model endomorphisms to have the form $`(๐ช_\rho ,\widehat{\rho })`$ where $`\rho `$ is an object in a tensor $`C^{}`$-category. We still want a Hilbert space $`K`$ of intertwiners between powers of $`\rho `$. We say that a Hilbert space $`H(\rho ,\sigma )`$ in a $`C^{}`$โcategory has zero left annihilator if $`X\psi =0`$ for all $`\psi H`$ implies $`X=0`$. It obviously suffices to take $`X`$ to be a positive element of $`(\sigma ,\sigma )`$.
We first prove a result that will imply that the $`C^{}`$โalgebras of interest are simple $`C^{}`$โalgebras.
6.9 Theorem Let $`๐ฏ_\rho `$ be a tensor $`C^{}`$โcategory whose objects are the powers of the object $`\rho `$. Let $`K(\rho ^r,\rho ^{r+g})`$ be a Hilbert space such that $`K\times 1_{\rho ^m}`$ has left annihilator zero for $`m_0`$ and suppose that
$$(\rho ^r,\rho ^r)\times 1_{\rho ^r}(K^r,K^r).$$
Then $`๐ช_\rho `$ is a simple $`C^{}`$โalgebra.
Proof. The proof follows that of the simplicity of the extended Cuntz algebra Theorem 3.1 of . Obviously, $`K`$ must now play the role of the generating Hilbert space. We identify the arrows of $`๐ฏ_\rho `$ with their images in $`๐ช_\rho `$. It suffices to show that any non-degenerate representation $`\pi `$ of $`๐ช_\rho `$ is faithful. $`\pi `$ is trivially isometric on Hilbert spaces in $`๐ช_\rho `$ and in particular on $`K^n`$, $`n_0`$. But then $`\pi `$ is also isometric on $`(K^m,K^n)`$, defined as in the discussion preceding Lemma 4.11. Given $`k`$, we let
$${}_{}{}^{o}๐ช_{\rho }^{}{}_{}{}^{k}:=_k(K^r,K^{r+k}),r0,r+k0$$
and let $`{}_{}{}^{o}๐ช_{\rho }^{}`$ denote the โsubalgebra of $`๐ช_\rho `$ obtained by taking finite sums of elements from the $`{}_{}{}^{o}๐ช_{\rho }^{}{}_{}{}^{k}`$. We have seen that $`\pi `$ is isometric on each $`{}_{}{}^{o}๐ช_{\rho }^{k}`$. If $`K`$ is infinite dimensional, we can continue as in the proof of Theorem 3.1 of to show that $`\pi `$ is isometric on $`{}_{}{}^{o}๐ช_{\rho }^{}`$. However this algebra is dense in $`๐ช_\rho `$, since arguing as in Theorem 4.9, we have $`(\rho ^{r+m},\rho ^{r+n})\times 1_{\rho ^r}(K^{r+m},K^{r+n})`$. Hence $`\pi `$ is isometric and $`๐ช_\rho `$ is simple. If $`K`$ is finite dimensional, we need only remark that $`(K^m,K^n)`$ is the set of all linear mappings from $`K^m`$ to $`K^n`$, so that, as a $`C^{}`$โalgebra, $`๐ช_\rho =๐ช_K`$ which is a simple $`C^{}`$โalgebra.
Thus instead of considering $`(๐ช_V,\rho _V)`$, we consider the smallest $`C^{}`$โsubalgebra $`๐_V`$ of $`๐ช_V`$ containing $`H:=V^{}K`$ and stable under $`\rho _V`$, equipped with the endomorphism $`\sigma _V`$ obtained by restricting $`\rho _V`$. The system $`(๐_V,\sigma _V)`$ is then a natural candidate for such a minimal $`C^{}`$โmodel system and we therefore address the question of finding necessary and sufficient conditions on a system $`(๐,\widehat{\sigma })`$ for it to be of the form $`(๐_V,\sigma _V)`$. We shall only discuss the case where $`V`$ is its regular representation, as the case of a corepresentation can, as before, be reduced to this case, for systems endowed with a standard braided symmetry. (This is not completely trivial, in that the concept of a braided symmetry not taking values in the intertwiner spaces, but just in their weak closures in some Hilbert space representation with support $`I`$ has to be formalized.)
We start by pointing out that the smallest tensor $`C^{}`$โsubcategory $`๐ฎ_V`$ of $`๐ฏ_V`$ containing $`V`$ and the intertwining space $`H=V^{}K\times 1_K`$ is, up to tensoring on the right by $`1_V`$, $`W^{}`$โdense in $`๐ฏ_V`$. More precisely, $`(V^r,V^s)\times 1_V`$ is contained in the $`W^{}`$โclosure of $`H^sH_{}^{r}{}_{}{}^{}`$ in $`(V^{r+1},V^{s+1})`$. Note that $`(๐_V,\sigma _V)`$ is the canonical system derived from $`\sigma =V`$ regarded as an object of the tensor $`C^{}`$โcategory $`๐ฎ_V`$. It is therefore canonically associated with $`๐ฏ_V`$. The following simple result relates the systems associated with $`V`$ considered as an object of $`๐ฎ_V`$ and $`๐ฏ_V`$, respectively.
6.10 Proposition For $`r,s=0,1,2`$, $`(\sigma _{V}^{}{}_{}{}^{r},\sigma _{V}^{}{}_{}{}^{s})=(\rho _{V}^{}{}_{}{}^{r},\rho _{V}^{}{}_{}{}^{s})๐_V`$
Proof. As $`H`$ generates $`๐ช_V`$, $`T(\rho _{V}^{}{}_{}{}^{r},\rho _{V}^{}{}_{}{}^{s})`$ if it intertwines the restrictions of the corresponding endomorphisms to the space of intertwiners $`H=V^{}K`$. Now this space is contained in $`๐_V`$, therefore $`(\rho _{V}^{}{}_{}{}^{r},\rho _{V}^{}{}_{}{}^{s})๐_V(\sigma _{V}^{}{}_{}{}^{r},\sigma _{V}^{}{}_{}{}^{s})`$, and the reverse inclusion is obvious.
Summarizing, the above discussion and conditions a)โd) lead to the following necessary conditions for $`(๐,\widehat{\sigma })`$ to be of the form $`(๐_V,\sigma _V)`$ with $`V`$ a regular multiplicative unitary:
a) $`๐`$ is the smallest $`\widehat{\sigma }`$โstable $`C^{}`$โsubalgebra containing a Hilbert space $`K`$ with zero left annihilator,
b) $`K(\widehat{\sigma },\widehat{\sigma }^2)`$,
c) $`K^{}\widehat{\sigma }(K)=KK^{}`$,
d) $`(\widehat{\sigma },\widehat{\sigma })K=K`$.
To see that these conditions are necessary, we need only remark that c) just expresses the regularity of the multiplicative unitary and ensures that the hypotheses of Lemma 5.3 hold for $`L=M=\widehat{\sigma }(K)`$ as we have already remarked. d) is now a consequence of Corollary 5.4. The sufficiency of these conditions follows from the next result.
6.11 Theorem Let $`(๐,\widehat{\sigma },K)`$ satisfy conditions a) to d) then $`๐`$ is simple and there is a regular multiplicative unitary $`V`$, unique up to equivalence, such that $`(๐,\widehat{\sigma },K)`$ is isomorphic to the model dual object $`(๐_V,\sigma _V,V^{}K)`$, where $`V`$ is regarded as a representation of $`V`$.
Proof. $`๐`$ is simple by Theorem 6.9. From condition c) it follows that
$$K^{}K^{}\widehat{\sigma }(K)K$$
and hence, since $`K`$ has left annihilator zero that $`\widehat{\sigma }(K)(K,K^2)`$. We now may apply Lemma 4.11 and Theorem 4.12 to the tensor $`C^{}`$โcategory of intertwiners between the powers of $`\widehat{\sigma }`$ to conclude that there is a multiplicative unitary $`V`$ on $`K^2`$ with $`V\theta \psi =\widehat{\sigma }(\psi )`$, $`\psi K`$. $`V`$ is regular by c), therefore the remaining conclusions follow using arguments similar to those of Theorem 6.5.
We can also give necessary and sufficient conditions for $`(๐,\widehat{\sigma })`$ to be of the form $`(๐_V,\sigma _V)`$ for a general multiplicative unitary. We retain a) and b) above and replace c) and d) by
c<sup>โฒโฒ</sup>) $`\widehat{\sigma }(K)(K,K^2)`$,
d<sup>โฒโฒ</sup>) If we consider the tensor $`C^{}`$โsubcategory $`๐ฏ`$ of the tensor $`C^{}`$โcategory of intertwiners between the powers of $`\widehat{\sigma }`$ generated by $`K`$ and denote its objects by $`\sigma ^n`$, where $`n_0`$, then $`(\sigma ,\sigma )K=K`$.
6.12 Theorem Let $`(๐,\widehat{\sigma },K)`$ satisfy conditions a), b), c<sup>โฒโฒ</sup>) and d<sup>โฒโฒ</sup>) then $`๐`$ is simple and there is a multiplicative unitary $`V`$, unique up to equivalence, such that $`(๐,\widehat{\sigma },K)`$ is isomorphic to the model dual object $`(๐_V,\sigma _V,V^{}K)`$, where $`V`$ is regarded as a representation of $`V`$.
The proof is a simplified version of that of the previous theorem seeing that regularity now plays no role. Finally, as a pendant to Theorem 6.6, we give a characterization of tensor $`C^{}`$โcategories of the form $`๐ฎ_V`$ for a multiplicative unitary $`V`$
6.13 Theorem Let $`๐ฏ_\rho `$ be a tensor $`C^{}`$โcategory whose objects are the tensor powers of an object $`\rho `$ and suppose $`๐ฏ_\rho `$ is generated by an ambidextrous Hilbert space $`K`$ in $`(\rho ,\rho ^2)`$ such that $`K\times 1_{\rho ^m}`$ has left annihilator zero for $`m_0`$. Let $`V`$ be the associated multiplicative unitary of Theorem 4.12, then $`๐ฏ_\rho ,K`$ is isomorphic to $`๐ฎ_V,V^{}K\times 1_\rho `$.
Proof. We let $`\stackrel{~}{๐ฆ}`$ be the category of Hilbert spaces with commuting shifts $`F`$ and $`G`$ associated with the ambidextrous space $`K(\rho ,\rho ^2)`$ as in Theorem 4.12. Then the functor $`G_V^{}`$ not only commutes with $`F`$ but satisfies $`G_V^{}G=\widehat{F}G_V^{}`$ by Proposition 3.1e. If we interpret $`G_V^{}`$ as a tensor โfunctor as in Proposition 4.10, we recognize that the image of $`๐ฏ_\rho `$ is the tensor $`C^{}`$โsubcategory generated by $`H:=V^{}K\times 1_\rho ๐ฏ_V`$. But this is, by definition, the minimal $`C^{}`$โsubcategory $`๐ฎ_V`$.
## 7 Appendix. Braided Symmetry
This appendix is devoted to the notion of braided symmetry. This evolved from a notion of the same name introduced in in the context of endomorphisms of $`C^{}`$โalgebras to generalize a previous more restricted notion of permutation symmetry in . Expressed in the context of a general tensor category, this notion can be expressed as follows. Let $`\sigma `$ denote the endomorphism of the braid group $`๐น_{\mathrm{}}=๐น_n`$ that shifts the braids $`b๐น_n`$ on $`n`$ threads to the right. By a braided symmetry for an object $`V`$ in a tensor category $`๐ฏ`$ we mean a representation $`\epsilon `$ of $`๐น_{\mathrm{}}`$ in $`๐ฏ`$ such that
$`\epsilon (b)(V^{\times n},V^{\times n}),b๐น_n,`$
$`\epsilon (\sigma (b))=1_V\times \epsilon (b),b๐น_{\mathrm{}}=๐น_n,`$
$`\epsilon (s,1)X\times 1_V=1_V\times X\epsilon (r,1),X(V^{\times r},V^{\times s}),`$
where $`(1,1)=b_1`$ is the braid on the first two threads and $`(s,1)=b_1\sigma (b_1)\mathrm{}\sigma ^{s1}(b_1).`$
Obviously, if the full subcategory whose objects are the tensor powers of V can be made into a braided tensor category then this braiding does define a braided symmetry. However, not all braided symmetries arise in this way. An application of this notion of braided symmetry is Theorem 5.31 of relating notions of amenability to the existence of a unitary braided symmetry.
Whilst this definition, with a view to simplicity, focused attention on the full subcategory whose objects are the tensor powers of $`V`$, we here need to consider the full tensor category and we will hence call $`๐ฏ`$ braided relative to a distinguished object $`V๐ฏ`$ if for any object $`W`$ in $`๐ฏ`$ there is an invertible arrow $`\epsilon _W(W\times V,V\times W)`$ such that
$`\epsilon _{W\times W^{}}=\epsilon _W\times 1_W^{}1_W\times \epsilon _W^{},`$
$`\epsilon _W^{}T\times 1_V=1_V\times T\epsilon _W,T(W,W^{}).`$
The second equation just says that $`\epsilon `$ is a natural transformation from the functor of tensoring on the right by $`V`$ to that of tensoring on the left by $`V`$. The first equation implies in particular that this natural transformation takes the value $`1_V`$ on the tensor unit. If $`\epsilon `$ is a braided symmetry for $`V`$ then we may define braided symmetries for the tensor powers $`V^{\times n}`$ of $`V`$ inductively, setting
$$\epsilon _W^{\times n}:=1_V\times \epsilon _W^{\times n1}\epsilon _W\times 1_{V^{n1}}.$$
It is easy to see that if $`\epsilon _W`$ is defined on a subset of objects, closed under tensor products, and such that every object of $`๐ฏ`$ is a subobject of a (finite) direct sum of objects from the subset and the equations are satisfied for $`W`$ and $`W^{}`$ in the subset, then $`\epsilon _W`$ extends uniquely to a braided symmetry on the whole category. This remains true if $`๐ฏ`$ is a $`W^{}`$โcategory and we allow infinite direct sums.
Braided symmetries in this sense have appeared as the starting point of the centre construction in tensor categories, see eg. where references to the original articles are given. But in view of its simplicity, the notion may well have appeared in other contexts, still unknown to the authors. We recall, however, the notion of an arrow between braided symmetries. If $`\epsilon `$ and $`\epsilon ^{}`$ are braided symmetries for $`V`$ and $`V^{}`$, respectively, then an arrow $`T(\epsilon ,\epsilon ^{})`$ is an arrow $`T(V,V^{})`$ such that
$$\epsilon _W^{}1_W\times T=T\times 1_W\epsilon _W,$$
for each object $`W`$ of $`๐ฏ`$.
When does an arrow $`\epsilon _V(V\times V,V\times V)`$ define a braided symmetry? To give some kind of answer, we consider a strict tensor category $`๐ฏ`$ and an object $`V`$ with the property that the functor of tensoring on the right by $`V`$ is faithful and such that $`W\times V`$ is a direct sum of copies of $`V`$ for each object $`W`$ of $`๐ฏ`$. This property is related to the notion right regular representation. The corresponding property of $`V\times W`$ being a direct sum of copies of $`V`$ is similarly related to the notion of left regular representation.
Given an invertible $`\epsilon _V(V\times V,V\times V)`$ such that
$$\epsilon _VT\times 1_V=1_V\times T\epsilon _V,T(V,V),$$
there is, for each object $`W`$ of $`๐ฏ`$ a unique invertible
$$\epsilon _{W\times V}(W\times V\times V,V\times W\times V)$$
such that
$$\epsilon _{W\times V}S\times 1_V=1_V\times S\epsilon _V,S(V,W\times V).$$
In fact, we have only to pick $`X_i(V,W\times V)`$ and $`Y_i(W\times V,V)`$ such that $`_iX_iY_i=1_{W\times V}`$ and we see that we have no option but to set
$$\epsilon _{W\times V}:=\underset{i}{}1_V\times X_i\epsilon _VY_i\times 1_V.$$
A routine computation shows that
$$\epsilon _{W^{}\times V}S\times 1_V=1_V\times S\epsilon _{W\times V},S(W\times V,W^{}\times V).$$
Now consider the set $`\mathrm{\Sigma }`$ of objects $`W`$ such that
$$\epsilon _{W\times V}1_W\times \epsilon _V^1(W\times V,V\times W)\times 1_V$$
and define $`\epsilon _W`$ by
$$\epsilon _W\times 1_V=\epsilon _{W\times V}1_W\times \epsilon _V^1.$$
If $`W`$ and $`W^{}`$ are in $`\mathrm{\Sigma }`$ and $`T(W,W^{})`$,
$$\epsilon _W^{}\times 1_VT\times 1_{V\times V}=\epsilon _{W^{}\times V}1_W^{}\times \epsilon _V^1T\times 1_{V\times V}$$
$$=\epsilon _{W^{}\times V}T\times 1_{V\times V}1_W\times \epsilon _V^1=1_V\times T\times 1_V\epsilon _{W\times V}1_W\times \epsilon _V^1,$$
so that
$$\epsilon _W^{}T\times 1_V=1_V\times T\epsilon _W,T(W,W^{}).$$
The set $`\mathrm{\Sigma }`$ trivially contains the tensor unit. Suppose both $`W`$ and $`W^{}`$ are in $`\mathrm{\Sigma }`$ then
$$\epsilon _{W\times W^{}\times V}1_W\times \epsilon _{W^{}\times V}^11_{W\times V}\times S=\epsilon _{W\times W^{}\times V}1_W\times S\times 1_V1_W\times \epsilon _V^1$$
$$=1_V\times 1_W\times S\epsilon _{W\times V}1_W\times \epsilon _V^1$$
$$=1_V\times 1_W\times S\epsilon _W\times 1_V=\epsilon _W\times 1_{W^{}\times V}1_W\times 1_V\times S,$$
where $`S(V,W^{}\times V)`$. Now since $`W^{}\times V`$ is a direct sum of copies of $`V`$ we conclude that
$$\epsilon _{W\times W^{}\times V}1_W\times \epsilon _{W^{}\times V}^1=\epsilon _W\times 1_{W^{}\times V}.$$
But since $`W^{}\mathrm{\Sigma }`$, $`\epsilon _{W^{}\times V}=\epsilon _W^{}\times 1_V1_W^{}\times \epsilon _V`$. Thus
$$\epsilon _{W\times W^{}\times V}=\epsilon _W\times 1_{W^{}\times V}1_W\times \epsilon _W^{}\times 1_V,$$
so that $`W\times W^{}\mathrm{\Sigma }`$. Furthermore,
$$\epsilon _{W\times W^{}}=\epsilon _W\times 1_W^{}1_W\times \epsilon _W^{}.$$
It is also easy to see that $`\mathrm{\Sigma }`$ is closed under subobjects and direct sums.
We have now proved the following result.
A.1 Proposition Let $`๐ฏ`$ be a tensor category, $`V`$ an object such that $`W\times V`$ is a direct sum of copies of $`V`$ for each $`W`$. Let $`\epsilon _V(V\times V,V\times V)`$ be a unitary such that
$$\epsilon _VT\times 1_V=1_V\times T\epsilon _V,T(V,V),$$
$$\epsilon _V\times 1_V1_V\times \epsilon _VS\times 1_V=1_V\times S\epsilon _V,S(V,V\times V).$$
then there is a unique maximal tensor subcategory with a braided symmetry $`\epsilon `$ whose value at $`V`$ coincides with the given invertible $`\epsilon _V`$. This is a full subcategory closed under subobjects and direct sums.
Proof. We need only remark that the second equation above implies that
$$\epsilon _{V\times V}=\epsilon _V\times 1_V1_V\times \epsilon _V$$
and hence $`V`$ is in $`\mathrm{\Sigma }`$ and the two definitions of $`\epsilon _V`$ coincide.
We now consider the category $`๐(V)`$ of corepresentations of a multiplicative unitary $`V`$ with its forgetful functor $`\iota `$ into the underlying category of Hilbert spaces. We let $`\vartheta `$ denote the braided symmetry relative to $`\iota (V)`$ derived from the symmetry on the category of Hilbert spaces but write $`\vartheta _W`$ in place of $`\vartheta _{\iota (W)}`$. The relevance of braided symmetries to this paper lies in the fact that there are many cases where $`๐(V)`$ admits a braided symmetry $`\epsilon `$ relative to $`V`$ with the further property that $`\widehat{V}`$ defined by $`V\vartheta \widehat{V}=\epsilon _V`$ is another multiplicative unitary on the same space. Such a braided symmetry will be called standard. $`\widehat{V}`$ determines and is uniquely determined by $`\epsilon `$.
Theorem A.2 A braided symmetry $`\epsilon `$ on $`๐(V)`$ relative to $`V`$ is standard if and only if
$$\widehat{V}_{12}V_{23}=V_{23}\widehat{V}_{12}.$$
If $`\epsilon `$ is standard and $`W`$ is a corepresentation of $`V`$, then $`\widehat{W}`$ defined by
$$W\vartheta _{W,K}\widehat{W}=\epsilon _W$$
is a representation of $`\widehat{V}`$ and for any pair $`W`$, $`W^{}`$ of corepresentations, we have
$$\widehat{W}_{12}W_{23}^{}=W_{23}^{}\widehat{W}_{12},$$
$$\widehat{W\times W^{}}=\widehat{W}\times \widehat{W}^{}.$$
Proof. Whether $`\epsilon `$ is standard or not, a simple calculation shows that $`T(W,W^{})`$ if and only if $`T(\widehat{W},\widehat{W}^{})`$. Since $`\widehat{W}(W\times V,\iota (W)\times V)`$, this yields
$$\widehat{W}_{12}\widehat{W\times V}=\widehat{\iota (W)\times V}\widehat{W}_{12}.$$
Now
$$\widehat{W}_{13}\widehat{W}_{23}^{}=\widehat{W}_{13}\vartheta _{23}W_{23}^{}_{}{}^{}1\epsilon _{23}$$
$$=\vartheta _{23}\widehat{W}_{12}W_{23}^{}_{}{}^{}1\epsilon _{23}.$$
If $`\widehat{W}_{12}W_{23}^{}=W_{23}^{}\widehat{W}_{12}`$ then we get
$$\widehat{W}_{13}\widehat{W}_{23}^{}=\vartheta _{23}W_{23}^{}_{}{}^{}1\vartheta _{12}^1W_{12}^1\epsilon _{W\times W^{}}=\vartheta _{W\times W^{}}^1W_{13}^{}_{}{}^{}1W_{12}^1\epsilon _{W\times W^{}}.$$
So $`\widehat{W}_{13}\widehat{W}_{23}^{}=\widehat{W\times W^{}}`$. Thus if $`\widehat{W}_{12}`$ and $`V_{23}`$ commute, we could conclude from our first identity that
$$\widehat{W}_{12}\widehat{W}_{13}\widehat{V}_{23}=\widehat{V}_{23}\widehat{W}_{12},$$
and, in particular, if $`\widehat{V}_{12}`$ and $`V_{23}`$ commute that $`\widehat{V}`$ is a multiplicative unitary. We now show that $`\widehat{V}_{12}V_{23}=V_{23}\widehat{V}_{12}`$ implies $`\widehat{W}_{12}W_{23}^{}=W_{23}^{}\widehat{W}_{12}`$. Now
$$V_{23}\widehat{V}_{12}W_{24}W_{34}=\widehat{V}_{12}V_{23}W_{24}W_{34}=\widehat{V}_{12}W_{34}V_{23},$$
$$V_{23}W_{24}\widehat{V}_{12}W_{34}=V_{23}W_{24}W_{34}\widehat{V}_{12}=W_{34}V_{23}\widehat{V}_{12},$$
where we have used the multiplicativity of $`V`$. But these expressions are equal, so cancelling, we find $`\widehat{V}_{12}W_{24}=W_{24}\widehat{V}_{12}`$ and we have replaced $`V`$ by $`W`$. The proof is completed by a similar step using the multiplicativity of $`\widehat{V}`$.
$$\widehat{W}_{12}\widehat{W}_{13}W_{34}^{}\widehat{V}_{23}=\widehat{W}_{12}\widehat{W}_{13}\widehat{V}_{23}W_{34}^{}=\widehat{V}_{23}\widehat{W}_{12}W_{34}^{},$$
$$\widehat{W}_{12}W_{34}^{}\widehat{W}_{13}\widehat{V}_{23}=W_{34}^{}\widehat{W}_{12}\widehat{W}_{13}\widehat{V}_{23}=W_{34}^{}\widehat{V}_{23}\widehat{W}_{12}.$$
Again these two expressions are equal and cancelling gives $`\widehat{W}_{13}W_{34}^{}=W_{34}^{}\widehat{W}_{13}`$, as claimed. Hence, our previous computation shows that $`\widehat{W}`$ is a representation of $`\widehat{V}`$, completing the proof.
Remark. Since a braided symmetry relative to $`V`$ might not be defined on the whole of $`๐(V)`$, it is worth remarking that $`๐(V)`$ can be replaced by a full tensor subcategory in the above theorem.
Note that the above theorem shows that when $`\epsilon `$ is a standard braided symmetry, $`๐(V)`$ and $`(\widehat{V})`$ are canonically isomorphic as tensor $`W^{}`$โcategories. Of course, we could start with the multiplicative unitary $`\widehat{V}`$ and then use $`\epsilon `$ to define $`V`$ and $`V`$ would again be multiplicative if and only if $`\widehat{V}_{12}V_{23}=V_{23}\widehat{V}_{12}`$. We conclude by showing that the interesting standard braided symmetries cannot be permutation symmetries.
A.3 Lemma Let $`\epsilon =V\vartheta \widehat{V}`$ define a standard braided symmetry of $`V.`$ Then $`\widehat{V}^1\widehat{V}_{23}^1\widehat{V}=\epsilon _{23}^1\widehat{V}^1\epsilon _{23}\widehat{V}_{23}^1.`$
Proof.
$$\epsilon _{23}^1\widehat{V}^1\epsilon _{23}^1\widehat{V}_{23}=(V\vartheta \widehat{V})_{23}^1\widehat{V}^1(V\vartheta )_{23}=$$
$$\widehat{V}_{23}^1\widehat{V}_{13}^1=\widehat{V}_{12}^1\widehat{V}_{23}^1\widehat{V}_{12},$$
since $`V_{23}`$ and $`\widehat{V}`$ commute and $`\widehat{V}`$ is multiplicative.
A.4 Proposition Let $`\epsilon =V\vartheta \widehat{V}`$ be a standard braided symmetry for $`V.`$ Then $`\vartheta =\epsilon \widehat{V}_{23}^1\epsilon _{23}\widehat{V}(\epsilon _{23}^1)^2\widehat{V}^1\epsilon _{23}\widehat{V}_{23}.`$ In particular if $`\epsilon `$ is a permutation symmetry then $`\epsilon =\vartheta `$ and $`V`$ is cocommutative.
Proof. $`(V^{\times 2},V^{\times 2})`$ is generated, as a weakly closed subspace, by $`H(V,V)H^{},`$ with $`H=\widehat{V}^1K,`$ thus setting $`\stackrel{~}{\epsilon }=\widehat{V}\epsilon \widehat{V}^1=\widehat{V}V\vartheta `$ then $`\epsilon =\widehat{V}^1\widehat{V}_{23}^1\stackrel{~}{\epsilon }\widehat{V}_{23}\widehat{V}.`$ So $`\epsilon `$ commutes with $`\widehat{V}^1\widehat{V}_{23}^1\widehat{V}`$. Thus by the previous lemma
$$\epsilon =\widehat{V}_{23}\epsilon _{23}^1\widehat{V}\epsilon _{23}\epsilon \epsilon _{23}^1\widehat{V}^1\epsilon _{23}\widehat{V}_{23}^1.$$
Now $`\epsilon `$ defines a braided symmetry for $`V`$, thus for any $`\psi =\widehat{V}^1\phi H(V,V^{\times 2}),`$ $`\epsilon \epsilon _{23}\psi \epsilon ^1=(\widehat{V}^1\phi )_{23}=\widehat{V}^1\widehat{V}_{23}^1\widehat{V}\widehat{V}_{13}\vartheta \phi `$ since $`\widehat{V}`$ is multiplicative, and, again by Lemma A.3, $`\epsilon _{23}^1\widehat{V}^1\epsilon _{23}\widehat{V}_{23}^1\vartheta \widehat{V}_{23}=\epsilon \epsilon _{23}\widehat{V}^1\epsilon _{23}^1`$ hence
$$\vartheta =\widehat{V}_{23}\epsilon _{23}^1\widehat{V}\epsilon _{23}\epsilon \epsilon _{23}\widehat{V}^1\epsilon _{23}^1\widehat{V}_{23}^1$$
and the conclusion follows. |
warning/0001/hep-ph0001132.html | ar5iv | text | # References
The standard model (SM) of strong and electroweak interactions is presently in very good shape. The experimental data agree with SM and this continuing success is somewhat paradoxically, a principal factor in the intensive search for physics beyond the standard model. This search is conducted nowadays in various sectors of particles phenomena. Among these, rare b decays is considered to provide good opportunities for discovering new physics beyond SM .
Among the rare decays studied so far $`bs\gamma `$ plays prominent role. The measured rate in two different experiments is $`Br(bs\gamma )=[3.15\pm 0.35(stat)\pm `$ $`0.32(syst)\pm 0.26(mod)]\times 10^4`$ and $`Br(bs\gamma )=[3.11\pm 0.80(stat)\pm `$ $`0.72(syst)]\times 10^4`$ , to be compared with the latest theoretical calculations within the SM giving $`Br(BX_s\gamma )=(3.32\pm 0.30)\times 10^4`$ . The agreement with SM is impresive and it is doubtful that a deviation could be detected, even when the above figures are improved. This, in view of the fact that long-distance (LD) contributions are also present; although more difficult to calculate with good accuracy, the existing estimates concur that these are approximately $`(510)\%`$ of the short-distance (SD) amplitude .
An alternative approach to the identification of virtual effects from new particles in $`b`$ decays like $`bs\gamma `$ and $`bsl^+l^{}`$ is the consideration of rare decays which are of negligible strength in SM. In such cases the mere appearence of the decays at a rate much larger than it is possible in SM would be a clear sign of new physics.
Recently, the decays $`bss\overline{d}`$, $`bdd\overline{s}`$ were proposed as ideal prototypes of the latter method. As shown in Ref. , the $`bss\overline{d}`$ is mediated in the SM by box-diagram and its calculation results in a branching ratio nearly of $`10^{11}`$, the exact value depending on the relative unknown phase between t, c contributions in the box. The $`bdd\overline{s}`$ branching ratio is even smaller by a factor of about $`10^2`$, due to the relative $`|V_{td}/V_{ts}|`$ factor in the amplitudes. The authors of ref. have also calculated the $`bss\overline{d}`$ transition in various โbeyond the SMโ models. It appears that for certain plausible values of the parameters, this decay may proceed with a branching ratio of $`10^810^7`$ in the minimal supersymmetric standard model and in two Higgs doublet models.
Moreover, when one considers supersymmetric models with R-parity violating couplings, it turns out that the existing bounds on the involved couplings of the superpotential do not provide at present any constraint on the $`bss\overline{d}`$ mode . It has been pointed out in Ref. that the hadronic channels most suitable for the search of the $`bss\overline{d}`$ transition are the $`\mathrm{\Delta }S=2`$ decays $`B^{}K^{}K^{}\pi ^+`$ or $`\overline{B}^0K^{}K^{}\pi ^+\pi ^+`$. The appropriate exclusive channel for $`bdd\overline{s}`$ transition would be $`B^{}K^+\pi ^{}\pi ^{}`$. At present there is no published experimental limits on these modes.
In the analysis of Ref. only the short-distance contributions were considered in detail. However, it is well known that long-distance contributions which are associated with low-lying intermediate hadronic states are also present in particle transitions. As we mentioned above, such contributions to $`BX_s\gamma `$ are rather small. However, each specific decay mode requires the estimation of its LD contribution; this is imperative, since only when a trustworthy estimate of such contributions is available one may proceed to compare the specific transition to the theoretical SM treatment or use it for revealing new physics. This necessity is best exemplified by known occurences in K - physics : in some decays like $`K^+\pi ^+\pi ^0\gamma `$, $`K^+\pi ^+l^+l^{}`$ the SD contribution is obscured by LD contributions while for $`K_L^0\pi ^0\nu \overline{\nu }`$, $`K^+\pi ^+\nu \overline{\nu }`$, the LD contributions are considerably smaller than the standard model short-distance amplitude . The calculation of long-distance contributions to a specific process is not based on a well-defined theoretical procedure. Clearly, the intermediate states are the main contributions to this part of the amplitude. However, the technique of their inclusion, as well as the choice of relevant states will influence the final result. We shall rely on the accumulated experience from the treatment of long-distance contributions to various processes, like $`K\overline{K}`$ transition , $`\overline{D}D`$ transition , $`K^+\pi ^+\nu \overline{\nu }`$ , $`BX_s\gamma `$ and $`B_s\gamma \gamma `$ decays, in order to formulate our approach to the $`B^{}K^{}K^{}\pi ^+`$ process at hand.
We include two contributions in the calculation of the long distance amplitude $`B^{}K^{}K^{}\pi ^+`$: (I) the box diagram, shown in Fig. 1, which is essentially the LD analog of the SD calculation in the standard model of the $`bss\overline{d}`$ transition. (II) the contribution of virtual $`\mathrm{"}D^0\mathrm{"}`$ and $`\mathrm{"}\pi ^0\mathrm{"}`$ mesons, via the chain $`B^{}K^{}\mathrm{"}D^0\mathrm{"}(\mathrm{"}\pi ^0\mathrm{"})`$ $`K^{}K^{}\pi ^+`$. This contribution arises as a sequence of two $`\mathrm{\Delta }S=1`$ transitions and may lead to final $`K^{}K^{}\pi ^+`$ state as well. It is therefore necessary to have an estimate of its relevance vis - ร - vis the โdirectโ $`\mathrm{\Delta }S=2`$ transition. Let us consider firstly the amplitude arising from (I). The two diagrams (a) and (b) express the Glashow - Iliopoulos - Maiani symmetry, so that the decay amplitude vanishes in the limit $`m_c=m_u`$ $`(m_D=m_\pi )`$. Since each diagram contains two $`W^{}`$s, it is related to several semileptonic processes with one virtual meson. Thus diagram (a) relates to $`D^0K^{}e^+\nu _e`$, $`B^{}D^0e^{}\overline{\nu }_e`$ and $`D^0\pi ^+e^{}\overline{\nu }_e`$ involving the product $`V_{cb}V_{cs}^{}V_{cd}V_{cs}^{}`$ and diagram (b) relates $`D^0K^{}e^+\nu _e`$, $`B^{}D^0e^{}\overline{\nu }_e`$, $`K^{}\pi ^0e^{}\overline{\nu }_e`$ and $`\pi ^+\pi ^0e^+\nu _e`$ involving the product $`V_{cb}V_{cs}^{}V_{ud}V_{us}^{}`$.
The transition probabilty is given by
$`<K^{}K^{}\pi ^+|๐ฎ|B^{}>_{box}=({\displaystyle \frac{ig}{2\sqrt{2}}})^4V_{cb}V_{cs}^{}V_{cd}V_{cs}^{}{\displaystyle d^4q_1d^4q_2d^4Q_2d^4Q_1}`$
$`\{\delta ^4(Q_1+q_2k_2)\delta ^4(p_Bq_1Q_1)\delta ^4(Q_2p_\pi q_2)\delta ^4(q_1Q_2k_1)`$
$`<D^0|(\overline{c}b)^\nu |B^{}>{\displaystyle \frac{i}{q_1^2m_D^2}}<K^{}|(\overline{s}c)^\alpha |D^0>{\displaystyle \frac{i(g_{\nu \alpha }Q_{1\nu }Q_{1\alpha }/M_W^2)}{Q_1^2M_W^2}}`$
$`{\displaystyle \frac{i(g_{\mu \beta }Q_{2\mu }Q_{2\beta }/M_W^2)}{Q_2^2M_W^2}}[<K^{}|(\overline{s}c)^\mu |D^0>{\displaystyle \frac{i}{q_2^2m_D^2}}<\pi ^+D^0|(\overline{c}d)^\beta |0>`$
$`<K^{}|(\overline{s}u)^\mu |\pi ^0>{\displaystyle \frac{i}{q_2^2m_\pi ^2}}<\pi ^+\pi ^0|(\overline{u}d)^\beta |0>]`$
$`+\delta ^4(Q_1q_2k_2)\delta ^4(p_Bq_1Q_1)\delta ^4(Q_2p_\pi +q_2)\delta ^4(q_1Q_2k_1)`$
$`<D^0|(\overline{c}b)^\nu |B^{}>{\displaystyle \frac{i}{q_1^2m_D^2}}<K^{}|(\overline{s}c)^\mu |D^0>{\displaystyle \frac{i(g_{\nu \alpha }Q_{1\nu }Q_{1\alpha }/M_W^2)}{Q_1^2M_W^2}}`$
$`{\displaystyle \frac{i(g_{\mu \beta }Q_{2\mu }Q_{2\beta }/M_W^2)}{Q_2^2M_W^2}}[<K^{}\overline{D}^0|(\overline{s}c)^\alpha |0>{\displaystyle \frac{i}{q_2^2m_D^2}}<\pi ^+|(\overline{c}d)^\beta |\overline{D}^0>`$
$`<K^{}\pi ^0|(\overline{s}u)^\alpha |0>{\displaystyle \frac{i}{q_2^2m_\pi ^2}}<\pi ^+|(\overline{u}d)^\beta |\pi ^0>]+(k_1k_2)\},`$ (1)
where $`(\overline{q}_jq_i)^\alpha `$ stands for $`\overline{q}_j\gamma ^\alpha (1\gamma _5)q_i`$, while the rest of the notation is defined in Figure 1. The first part comes out from the diagrams on Figure 1, while the second results from the crossed diagrams. The calculation of (1) depends on the matrix elements $`<D^0|(\overline{c}b)^\nu |B^{}>`$, $`<K^{}|(\overline{s}c)^\mu |D^0>`$, $`<\pi ^+|(\overline{c}d)^\beta |\overline{D}^0>`$, $`<K^{}|(\overline{s}u)^\mu |\pi ^0>`$ and $`<\pi ^0|(\overline{u}d)^\beta |\pi ^{}>`$. Since only pseudoscalar states appear, we have to deal with transitions between such states induced by the vector current only,
$`<P^{}(p^{})|\overline{q}_j\gamma ^\mu q_i|P(p)>=f_+(q^2)(p^\mu +p^\mu )+f_{}(q^2)(p^\mu p^\mu ),`$ (2)
which may be rewritten as
$`<P^{}(p^{})|\overline{q}_j\gamma ^\mu q_i|P(p)>`$ $`=`$ $`F_1(q^2)(p^\mu +p^\mu {\displaystyle \frac{m_P^2m_P^{}^2}{q^2}}(p^\mu p^\mu ))`$ (3)
$`+`$ $`F_0(q^2){\displaystyle \frac{m_P^2m_P^{}^2}{q^2}}(p^\mu p^\mu ),`$
where $`F_1`$ and $`F_0`$ contain the contribution of vector and scalar states respectively and $`q^2=(pp^{})^2`$. Also, $`F_1(0)=F_0(0)`$ . For these form factors, one usually assumes pole dominance
$`F_1(q^2)`$ $`=`$ $`{\displaystyle \frac{F_1(0)}{1\frac{q^2}{m_V^2}}};F_0(q^2)={\displaystyle \frac{F_0(0)}{1\frac{q^2}{m_S^2}}}`$ (4)
and in order to simplify, we shall take $`m_V=m_S`$, from which results $`f_{}(q^2)=0`$. We shall assume that one can safely take $`f_+(q^2)1`$ and the limit $`Q_1^2`$ , $`Q_2^2M_W^2`$, which then leads to a more tractable expression for the real part of the amplitude
$`๐_r^{box}(B^{}(p_B)K^{}(k_1)K^{}(k_2)\pi ^+(p_\pi ))={\displaystyle \frac{G^2}{16\pi ^4}}V_{cb}V_{cs}^{}V_{cd}V_{cs}^{}{\displaystyle d^4q_1}`$
$`\{{\displaystyle \frac{1}{q_1^2m_D^2}}{\displaystyle \frac{1}{(q_1k_1p_\pi )^2m_D^2}}[(m_B^2+2k_2p_B)(m_K^2+2k_1p_\pi )`$
$`+q_1^2(m_K^2+2k_1p_\pi +m_B^22k_2p_B)+2k_2q_1(m_K^2+2k_1p_\pi )`$
$`+2p_\pi q_12k_2q_12k_2q_1q_1^2+2p_\pi q_1q_1^2+2p_\pi q_1(m_B^2+2p_\pi p_B)q_1^4]`$
$`{\displaystyle \frac{1}{q_1^2m_D^2}}{\displaystyle \frac{1}{(q_1k_1p_\pi )^2m_\pi ^2}}[(m_B^2+2k_2p_B)(m_K^2+2k_1p_\pi )`$
$`+q_1^2(m_K^2+2k_1p_\pi +m_B^22k_2p_B)+2k_2q_1(m_K^2+2k_1p_\pi )`$
$`+2p_\pi q_12k_2q_12k_2q_1q_1^2+2p_\pi q_1q_1^2+2p_\pi q_1(m_B^2+2p_\pi p_B)q_1^4]`$
$`+(k_1k_2)\},`$ (5)
where $`G=\sqrt{2}g^2/(8M_W^2)`$. The separate contributions of $`DD`$ and $`D\pi `$ intermadiate states diverge as fourth power. However, the GIM cancellation acts in such a way as to decrease the degree of divergence and finally the integral in (5) will give a quadratic divergence. Similar situations were encountered in previous LD calculations . We note that the explicit inclusion of the pole-type form factors for $`f_+(q^2)`$ would reduce the degree of divergence, in such a case, however, the evaluation of the integrals becomes very cumbersome and this effort is not justifed since as it will turn out the contribution of the real part is essentially negligible in comparison to that provided by the imaginary part. The integrals in (5) are calculated by using Feynman parametrization. The final result for the decay rate is
$`\mathrm{\Gamma }(B^{}K^{}K^{}\pi ^+)`$ $`=`$ $`{\displaystyle \frac{1}{2(2\pi )^332m_B^3}}{\displaystyle _{(m_\pi +m_K)^2}^{(m_Bm_K)^2}}๐s_2{\displaystyle _{(s_1)_1}^{(s_1)_2}}๐s_1|๐|^2,`$ (6)
where
$`(s_1)_{1,2}`$ $`=`$ $`m_K^2+m_\pi ^2{\displaystyle \frac{1}{2s_2}}[(s_2m_B^2+m_K^2)(s_2+m_\pi ^2m_K^2)`$
$`\pm `$ $`\lambda ^{1/2}(s_2,m_B^2,m_K^2)\lambda ^{1/2}(s_2,m_\pi ^2,m_K^2)]`$
and $`\lambda (a,b,c)=a^2+b^2+c^22ab2ac2ab`$. The $`๐_r^{box}`$ denotes the leading term of the amplitude, which results after using the primitive cut-off regularization:
$`๐_r^{box}(B^{}K^{}K^{}\pi ^+)`$ $``$ $`G^2V_{cb}V_{cs}^{}V_{cd}V_{cs}^{}{\displaystyle \frac{1}{16\pi ^2}}\mathrm{\Lambda }^2(m_D^2m_\pi ^2).`$ (8)
There is obviously the uncertainty in the value to be taken for $`\mathrm{\Lambda }`$. The momentum in the box cannot exceed $`m_B`$, and by taking $`\mathrm{\Lambda }10`$ $`\mathrm{GeV}`$ we obtain
$$BR(B^{}K^{}K^{}\pi ^+)_{(r)}^{(box)}8\times 10^{15}$$
(9)
for the real part of this contribution, using $`\mathrm{\Gamma }(B^{}all)=4\times 10^{13}`$ $`\mathrm{GeV}`$ .
Turning now to the imaginary part of the $`B^{}K^{}K^{}\pi ^+`$ amplitude provided by the $`DD`$ and $`D\pi `$ intermediate states, it is given by
$`๐_i^{box}(B^{}K^{}K^{}\pi ^+)={\displaystyle \frac{G^2}{32\pi ^2}}V_{cb}V_{cs}^{}V_{cd}V_{cs}^{}`$
$`{\displaystyle d^4q_1\delta (q_1^2m_D^2)[\delta ((q_1k_1k_2)^2m_D^2)\delta ((q_1k_1p_\pi )^2m_\pi ^2)]}`$
$`\{q_1^4+2k_2q_1q_1^2+2p_\pi q_1q_1^2+q_1^2(m_K^2+2k_1p_\pi +m_B^22k_2p_B)`$
$`+2p_\pi q_12k_2q_1+2p_\pi q_1(m_B^2+2k_2p_B)`$
$`+2k_2q_1(m_K^2+2p_\pi k_1)+(m_B^2+2k_2p_B)(m_K^2+2k_1p_\pi )`$
$`+(k_1k_2)\}.`$ (10)
Introducing now $`s_1=(p_Bk_1)^2=`$ $`(k_2+p_\pi )^2`$ and $`s_2=(p_Bk_2)^2=`$ $`(k_1+p_\pi )^2`$ one arrives at
$`๐_i^{box}(B^{}K^{}K^{}\pi ^+)={\displaystyle \frac{G^2}{32\pi ^2}}V_{cb}V_{cs}^{}V_{cd}V_{cs}^{}`$
$`\times \{F(m_D^2,s_1,s_2)F(m_\pi ^2,s_1,s_2)+(s_1s_2)\},`$ (11)
with
$`F(m_P,s_1,s_2)={\displaystyle \frac{\lambda ^{1/2}(m_D^2,m_P^2,s_1)}{m_D^2m_P^2+s_1}}`$
$`\{m_D^4+m_D^2(2s_1{\displaystyle \frac{3}{2}}m_K^2{\displaystyle \frac{3}{2}}m_\pi ^2{\displaystyle \frac{1}{2}}s_2)+(s_1m_\pi )(s_1+m_K^2)`$
$`+(s_1s_2)\}.`$ (12)
Using the expression (6) for the decay width we find
$$BR(B^{}K^{}K^{}\pi ^+)_{(i)}^{(box)}6\times 10^{12}.$$
(13)
We proceed now to estimate the second possibility for a LD part which may lead to a final $`K^{}K^{}\pi ^+`$ state. This possibility is expressed as two consecutive two-body nonleptonic transitions (see Fig. 2) in which the connecting single particles $`D^0`$ and $`\pi ^0`$ is virtual. An estimate of this contribution requires the knowledge of the $`<B^{}|_w|K^{}\mathrm{"}D^0\mathrm{"}>`$, $`<\mathrm{"}D^0\mathrm{"}|_w|K^{}\pi ^+>`$ amplitudes for virtual $`D^0`$, which is lacking. For a virtual $`\pi ^0`$ existing estimates for $`<\mathrm{"}\pi ^0\mathrm{"}|_w|K^{}\pi ^+>`$ indicate that it is smaller than the physical amplitude in a certain region. We rely in our estimation on the โphysicalโ amplitudes $`B^{}K^{}D`$, $`D^0K^{}\pi ^+`$ , keeping in mind that this induces an amount of uncertainty. However, the final numerical results will show that this is of no consequence in the present problem. In the diagram (2b) the $`D^0`$ may also be on the mass shell. Therefore, we must exclude in our calculations the region around physical $`D^0`$, which represents two $`\mathrm{\Delta }S=1`$ physical decays, $`B^{}D^0K^{}`$ followed by $`D^0K^{}\pi ^+`$, since we are pursuing the $`B^{}K^{}K^{}\pi ^+`$ outside the resonance region. We shall return to this point below.
The calculation of the virtual $`D^0`$ mediated part of the amplitude requires the use of the effective nonleptonic Lagrangian. The part relevant for the present calculation is
$`_{LD}`$ $`=`$ $`{\displaystyle \frac{G}{\sqrt{2}}}\{V_{cb}V_{us}^{}[a_1^{(b)}(\overline{c}b)^\mu (\overline{s}u)_\mu +a_2^{(b)}(\overline{c}u)^\mu (\overline{s}b)_\mu ]`$ (14)
$`+`$ $`V_{cs}V_{ud}^{}[a_1^{(c)}(\overline{c}s)^\mu (\overline{d}u)_\mu +a_2^{(c)}(\overline{c}u)^\mu (\overline{d}s)_\mu ]+h.c.\},`$
$`V_{q_1q_2}`$ are CKM matrix elements and $`a_1^{(c)}`$, $`a_2^{(c)}`$, $`a_1^{(b)}`$ and $`a_2^{(b)}`$ are effective Wilson coefficients (see Bauer et al., Ref. ) at the charm and beauty scales. We use factorization approximation for the two parts of the $`B^{}K^{}K^{}\pi ^+`$ amplitude and the expression obtained in for the $`<P_1P_2|D>`$ transition,
$`M_{<P_1P_2|D>}`$ $`=`$ $`{\displaystyle \frac{G}{\sqrt{2}}}C_{P_1P_2}if_{P_2}F_0^{DP_1}(m_{P_2}^2)(m_D^2m_{P_1}^2).`$ (15)
In (15) $`C_{P_1P_2}`$ contains CKM matrix elements and a Wilson coefficient. By using the explicit form of (15) we have neglected the small contribution from the annihilation part of the amplitude, which is proportional to $`a_2^{(b)}`$, $`a_2^{(c)}`$ . The part of the decay amplitude due to the $`\mathrm{"}D^0\mathrm{"}`$ pole is then given by
$`๐_{D^0}^{pole}(B^{}K^{}K^{}\pi ^+)={\displaystyle \frac{G^2}{2}}V_{cb}V_{us}^{}V_{cs}V_{ud}^{}a_1^{(b)}a_1^{(c)}f_Kf_\pi `$
$`\times F_0^{BD}(m_K^2)F_0^{DK}(m_\pi ^2){\displaystyle \frac{(m_B^2q^2)(q^2m_K^2)}{q^2m_D^2+im_D\mathrm{\Gamma }_D}}.`$ (16)
The decay width due to this contribution is given by
$`\mathrm{\Gamma }(B^{}K^{}K^{}\pi ^+)={\displaystyle \frac{1}{2(2\pi )^332m_B^3}}|C|^2|F_0^{BD}(m_K^2)F_0^{DK}(m_\pi ^2)|^2`$
$`\times {\displaystyle _{(m_\pi +m_K)^2}^{(m_Bm_K)^2}}ds{\displaystyle \frac{(m_B^2s)^2(sm_K^2)^2}{(sm_D^2)^2+(m_D\mathrm{\Gamma }_D)^2}}`$
$`{\displaystyle \frac{1}{s}}\lambda ^{1/2}(s,m_B^2,m_K^2)\lambda ^{1/2}(s,m_\pi ^2,m_K^2),`$ (17)
with $`C=(G^2/2)V_{cb}V_{us}^{}V_{cs}V_{ud}^{}a_1^{(b)}a_1^{(c)}f_Kf_\pi `$, for the resonance in a $`s`$ channel and the same for the resonance in a crossed channel.
Using for $`a_1^{(b)}`$, $`a_1^{(c)}`$, $`F_0^{DK}`$ and $`F_0^{BD}`$ the values of Bauer, Stech and Wirbel we calculate the virtual $`\mathrm{"}D^0\mathrm{"}`$ contribution by deleting a width of $`2\mathrm{\Delta }`$ around $`D`$ mass in the $`s`$ variable. The size of $`\mathrm{\Delta }`$ is related to the experimental accuracy of the $`D`$ \- determination in the final $`K^{}\pi ^+`$ state. In the various experiments it ranges between 1 and 10 $`\mathrm{MeV}`$. One should keep in mind that average accuracy of $`D^0`$ \- mass determination is $`0.5`$ $`\mathrm{MeV}`$ . Thus, in order to delete the physical $`D^0`$โs one must take at least $`\mathrm{\Delta }=1`$ $`\mathrm{MeV}`$. However, we shall check the $`\mathrm{\Delta }`$ dependence for a range of values to make sure that our conclusions are not affected.
For $`\mathrm{\Delta }=20`$, $`5`$, $`1`$, $`0.1`$ $`\mathrm{MeV}`$ we find the nonresonant $`D^0`$ contribution to be
$`BR(B^{}K^{}K^{}\pi ^+)_{(D^0)}^{pole}`$ $`=`$ $`(0.31;1.2;6.2;61)\times 10^{15}.`$ (18)
In all cases the result is much smaller than (13), though one should remember that $`\mathrm{\Delta }=(15)`$ $`\mathrm{MeV}`$ is the realistic option. A similar calculation for $`\pi ^0`$ intermediate contribution, i.e. $`B^{}K^{}\mathrm{"}\pi ^0\mathrm{"}`$ $`\mathrm{"}\pi ^0\mathrm{"}K^{}\pi ^+`$ yields a value smaller by four orders of magnitude, especially as a result of CKM angles. The pole contribution is therefore considerablly smaller than the LD box contribution calculated with $`DD`$ and $`D\pi `$ intermediate states; thus the total branching ratio from all diagrams we included is
$`BR(B^{}K^{}K^{}\pi ^+)_{LD}`$ $`=`$ $`6\times 10^{12}.`$ (19)
As a check, we used our $`๐_{(D^0)}^{pole}`$ amplitude to calculate $`B^{}K^{}K^{}\pi ^+`$ as given by decay via a physical $`D^0`$ and we find a branching ratio of $`7\times 10^6`$. This agrees very well with the experimental expectation of $`(9.9\pm 2.8)\times 10^6`$, obtained by using $`BR(B^{}D^0K^{})`$ $`=(2.57\pm 0.65\pm 0.32)\times 10^4`$ and $`BR(D^0K^{}\pi ^+)=3.85\times 10^2`$ .
A few remarks concerning our approximations. As we mentioned, we have neglected the form factor dependence in the calculation of $`๐_r^{box}`$. Their inclusion would have decreased the degree of divergence. However, in view of the smallness of the real part of the amplitude, this neglect is of no consequence. A possibly more serious uncertainty is caused by the fact that we used only $`DD`$ and $`D\pi `$ intermediate states in the box calculations. Additional intermediate states, all within the physical region could be $`DD^{}`$, $`D^{}D^{}`$, $`D\rho `$, $`D^{}\rho `$, $`D\eta (\eta ^{})`$. We did not consider these states for two main reasons: first, there is no knowledge of the matrix elements and the required form factors involved. Moreover, the inclusion of strongly decaying resonances $`D^{}`$, $`\rho `$ as intermediate states is questionable and some of their effects are taken into account by the form factors considered (4). We decided therefore to ignore these contributions, though we are aware of the possibilty that additional intermediate states in Fig. 1 might increase our result by a factor of, say, 2-3. Finally, it is interesting to note that our result, which indicates that the LD contribution in the $`B^{}K^{}K^{}\pi ^+`$ decay is smaller or at most comparable to the SM short-distance contribution, fits into the general picture of B decays. This is in contrast to the situation in the strange and charm sectors: in K transitions the two contributions are comparable in some cases, SD dominates in a few decays and LD in many others; in $`D`$ \- transitions LD contributions are generally larger than the SD ones, except for the unusual case of $`B_cB_u^{}\gamma `$ decay .
To summarize, we have shown that the long - distance contributions to $`B^{}K^{}K^{}\pi ^+`$ are smaller in the SM than the short - distance box diagram, and have the branching ratio in a $`10^{12}10^{11}`$ range. This is a most welcome feature since it strengthens the suitability of the $`B^{}K^{}K^{}\pi ^+`$ decay as an ideal testing ground for physics beyond the standard model, as originally suggested in ref. . We expect that this avenue will be explored experimentally in the near future and we note that the first analysis of this mode has just been completed by the OPAL Collaboration and an upper limit of $`1.29\times 10^4`$ at $`90\%`$ confidence level has been set for the branching ratio of this decay.
This work has been supported in part by the Ministry of Science of the Republic of Slovenia (SF) and by the Fund for Promotion of Research at the Technion (PS). We acknowledge with thanks discussion with Drs. Yoram Rozen and Shlomit Tarem on the experimental aspects of the problem. One of us (SF) thanks A. Ramลกak and D. Veberiฤ for their help in numerical calculations.
Figure Captions
Fig. 1. Long distance box-diagram contributions to $`B^{}K^{}K^{}\pi ^+`$.
Fig. 2. Pole contributions to the long distance amplitude of $`B^{}`$ $`K^{}K^{}\pi ^+`$, (a) quark picture, (b) hadronic picture. |
warning/0001/quant-ph0001072.html | ar5iv | text | # Quantum limit of optical magnetometry in the presence of ac-Stark shifts
## I Introduction
The detection of magnetic fields by optical means is a well developed technique with applications ranging from geology and medicine to fundamental tests of violations of parity and time-reversal symmetry .
In spite of their great variety, optical magnetometers can be divided in two basic classes. In the first class light absorption at a magnetic resonance is used to detect Zeeman level shifts, while the second class makes use of the associated changes of the index of refraction. So called optical pumping magnetometer (OPM) as well as dark-state magnetometers based on absorption measurements belong to the first class. The recently developed magnetometers based on phase-coherent atomic media and the mean-field laser magnetometer of ref. belong to the second class.
If systematic measurement errors can be avoided, which in practice can be a challenging task, the smallest detectable Zeeman shift (in units of frequency) is determined by the ratio of the noise level of the signal $`S`$ to its rate of change with respect to frequency
$$\mathrm{\Delta }\omega _{\mathrm{min}}=\frac{S_{\mathrm{noise}}}{\left|\mathrm{d}S/\mathrm{d}\omega \right|}.$$
(1)
A fundamental lower limit of $`S_{\mathrm{noise}}`$ results from photon counting errors due to shot-noise of the probe electromagnetic wave. $`\left(\mathrm{d}S/\mathrm{d}\omega \right)^1`$, which characterizes a โquality factorโ of the system, is determined by an effective width of the magnetic resonance. The ultimate goal of magnetometer design is to minimize the noise level and the effective width at the same time.
The width of magnetic resonances in optical magnetometers is subject to two types of broadening: resonant power-broadening due to the coupling of the optical fields to the probe-transition and a broadening due to ac-Stark shifts resulting from non-resonant couplings to other transitions. As shown in and power-broadening limits the simultaneous minimization of noise and $`\left(\mathrm{d}S/\mathrm{d}\omega \right)^1`$ in absorption based magnetometers. In such devices increasing the probe laser power reduces the shot-noise but does reduce the signal at the same time. As a consequence the sensitivity saturates at a rather low power level. On the other hand, as shown in and , this effect can be compensated in a magnetometer that detects phase shifts of the probe electromagnetic wave propagating in an optically thick atomic medium under conditions of electromagnetically induced transparency (EIT) . Theoretically a complete elimination is possible in a 3-level $`\mathrm{\Lambda }`$-type system.
In any real atomic system, however, there are non-resonant couplings to additional levels which lead to ac-Stark shifts and an additional broadening of the magnetic resonance proportional to the laser intensity. In the present paper we analyze the influence of ac-Stark shifts and show that they (i) can diminish the magnetometer signal and (ii) lead to additional noise contributions. We show that in absorption based devices ac-Stark broadening leads to a further reduction of the signal. In contrast it only gives rise to a bias phase shift in an phase-sensitive EIT magnetometer. This bias phase shift can be calibrated but is still a major source of systematic errors. It can be eliminated, if an EIT magnetometer with Faraday configuration is considered.
However, in both, absorptive and dispersive type devices, ac-Stark shifts give also rise to fundamental noise contributions which increase with the laser power more rapidly than shot noise. Hence the magnetometer sensitivity decreases above a certain power level. The maximum value of sensitivity constitutes the standard quantum limit. For an EIT magnetometer based on phase-shift measurements this limit is determined by the dispersion-absorption ratio of the medium and the intensity-phase noise coupling due to the self-phase modulation associated with ac-Stark shifts.
We also discuss the possibility of further increasing the sensitivity by means of non-classical light fields and show that the maximum sensitivity is essentially independent of the light statistics.
The paper is organized as follows: In Sec. II we discuss the fundamental broadening mechanisms of magnetic resonances, power-broadening and ac-Stark associated broadenings. It is shown in Sec. III that the classical signal reduction due to these broadenings can be compensated in phase-sensitive EIT magnetometers in contrast to absorption-based techniques. In Sec. IV fundamental quantum noise sources are discussed and the standard quantum limit of magnetometer sensitivity derived. A detailed analysis of an EIT-Faraday magnetometer is given in Sec. V and the prospects of using non-classical input states are discussed.
## II broadening of magnetic resonances
Optical magnetometers measure in essence the position of certain resonances which are sensitive to magnetic level shifts. An important quantity that determines the signal strength of such a measurement is the width of the magnetic resonance. As a rule the narrower the resonance, the easier it is to detect level shifts.
Magnetic resonances with small natural width can be obtained e.g. by coupling Zeeman or hyperfine components of ground states in atoms either with an RF field or via an optical Raman transition. In an optical magnetometer these ground-state sub-levels are then coupled by laser fields to excited atomic states. The optical coupling is also used to detect energy shifts of the ground-state sub-levels induced by a magnetic field. However, at the same time this coupling leads to a broadening of the magnetic resonances via two mechanisms: (i) power-broadening and (ii) broadening due to ac-Stark shifts.
### A Power-broadening
The first mechanism is power-broadening due to the resonant interaction with the probe transition. When the Rabi-frequency $`\mathrm{\Omega }`$ of the optical probe field exceeds the value
$$\mathrm{\Omega }_{\mathrm{crit}}^{(1)}\sqrt{\gamma \gamma _0},$$
(2)
where $`\gamma _0`$ is the unbroadened width of the magnetic resonance and $`\gamma `$ the homogeneous linewidth of the optical transition, the magnetic resonance becomes power-broadened. (Here and below we assume that $`\gamma \gamma _0`$.) The effective width scales linearly with the Rabi-frequency $`\mathrm{\Omega }`$ of the optical field or the square root of the corresponding power
$$\mathrm{\Gamma }_{\mathrm{eff}}=\gamma _0+\alpha \sqrt{\frac{\gamma _0}{\gamma }}|\mathrm{\Omega }|+\mathrm{}.$$
(3)
$`\alpha `$ is some numerical pre-factor of order unity that depends on the specific model . This broadening effect leads to a substantial limitation of the signal in an optical pumping magnetometer, as shown in and .
### B Broadening due to ac-Stark shifts
The second broadening mechanism is due to non-resonant couplings of the probe electromagnetic wave with other than the probe transition and the associated ac-Stark shifts. The ac-Stark effect leads to a shift of the magnetic resonance of
$$\mathrm{\Delta }\omega _{\mathrm{ac}\mathrm{Stark}}=\frac{|\mathrm{\Omega }|^2}{\mathrm{\Delta }_0}$$
(4)
where $`\mathrm{\Delta }_0`$ is some effective detuning of non-resonant transitions from the frequency of the probe field weighted with relative oscillator strengths. $`\mathrm{\Omega }`$ is again the Rabi-frequency of the probe field corresponding to the resonant probe transition. ($`\mathrm{\Delta }_0`$ is of course just a model-dependent coupling parameter. We have used this notation here for simplicity of the discussions.)
In the classical limit and for a homogeneous laser intensity throughout the atomic vapor, there is only a constant frequency shift due to the ac-Stark effect. This shift can be calibrated. However, maximum signal is usually achieved when the atomic density is chosen such that there is a substantial absorption of the probe field. Hence when the probe Rabi-frequency exceeds the value
$$\mathrm{\Omega }_{\mathrm{crit}}^{(2)}\sqrt{\mathrm{\Delta }_0\gamma _0}$$
(5)
the resonance frequency changes as a function of propagation through the medium. This leads to an effective inhomogeneous broadening of the magnetic resonance. For example, the transmission of a cell containing atoms with a Lorentzian magnetic resonance subject to ac-Stark shifts is determined by the integrated imaginary part of the susceptibility ($`\chi ^{\prime \prime }=\mathrm{Im}[\chi ]`$)
$$_0^Ldz\chi ^{\prime \prime }(z)_0^Ldz\frac{\gamma _0}{\gamma _0^2+(\mathrm{\Delta }+|\mathrm{\Omega }(z)|^2/\mathrm{\Delta }_0)^2}.$$
(6)
$`|\mathrm{\Omega }(z)|^2`$ characterizes the $`z`$-dependent power of the probe field and $`\mathrm{\Delta }`$ the detuning from the un-shifted transition frequency. It is easy to see, that there is a broadening of the magnetic resonance depending on the magnitude of the ac-Stark shifts and the details of the absorption process. An important feature is that this broadening is proportional to the square of the Rabi-frequency or the laser power. Thus above a certain power level, determined by Eq.(5) ac-Stark associated broadening can exceed power broadening, which leads e.g. to further reduction of the signal in an optical pumping magnetometer.
## III compensation of broadening effects in EIT magnetometer
We here demonstrate that the classical broadening mechanisms discussed in the previous section do not necessarily lead to a reduction of the magnetometer signal if phase measurement techniques are used. It has been shown in detail in and , that power-broadening can be completely compensated in a phase measurement by making use of EIT in optically dense $`\mathrm{\Lambda }`$-type systems.
The 3-level $`\mathrm{\Lambda }`$ configuration of an EIT magnetometer as well as the associated linear susceptibility spectrum of the probe field are shown in Fig.1. Here and in the following we consider closed systems i.e. we assume that there are no effective decay mechanism due to time-of-flight limitations. The upper level of the probe-field transition $`|a|b`$ is coupled to a meta-stable lower level $`|c`$ by a coherent and strong driving field of Rabi-frequency $`\mathrm{\Omega }_d`$. The probe field Rabi-frequency is denoted as $`\mathrm{\Omega }_p`$ ($`\mathrm{\Omega }_p\mathrm{\Omega }_d`$) and the coherence decay rate of the probe transition as $`\gamma `$. $`\mathrm{\Delta }`$ is the one-photon detuning of the drive field and $`\delta `$ the two-photon detuning. The transverse decay rate of the two-photon resonance (magnetic resonance) is denoted as $`\gamma _0`$. It is assumed that the corresponding population exchange between the ground-state sub-levels is small and will be neglected
As in the case of an OPM there is power-broadening in an EIT magnetometer as soon as $`|\mathrm{\Omega }_d|>\sqrt{\gamma \gamma _0}`$. A unique property of an EIT resonance is however that the dispersion-absorption ratio of the optical transition is given by the inverse of the width of the ground-state transition $`\gamma _0`$ and is independent on the drive power if $`|\mathrm{\Omega }_d|>\sqrt{\gamma \gamma _0}`$. Under conditions of one-photon resonance ($`\mathrm{\Delta }=0`$) one finds for small two-photon detuning
$`\chi ^{}\mathrm{Re}[\chi ]`$ $``$ $`{\displaystyle \frac{\delta }{|\mathrm{\Omega }_d|^2+\gamma \gamma _0}},`$ (7)
$`\chi ^{\prime \prime }\mathrm{Im}[\chi ]`$ $``$ $`{\displaystyle \frac{\gamma _0}{|\mathrm{\Omega }_d|^2+\gamma \gamma _0}}.`$ (8)
The residual absorption at the EIT resonance decreases with increasing laser power in the same way as the dispersion. Thus in a phase shift measurement power broadening can be compensated by increasing the density and keeping a constant optical depth of the medium.
Similarly one finds that as long as the drive-field Rabi-frequency is large compared to probe-induced ac-Stark shifts, which is very well satisfied, ac-Stark shifts of the magnetic resonance (eq.(4)) lead only to a bias phase shift.
$$\mathrm{\Delta }\varphi _{acStark}_0^Ldz\frac{|\mathrm{\Omega }(z)|^2}{\mathrm{\Delta }_0},$$
(9)
where $`L`$ is the length of the atomic vapor cell. This phase shift can in principle be calibrated but gives rise to systematic errors. As will be discussed in detail later on, there is no such bias phase shift in a resonant Faraday configuration.
We conclude this section by emphasizing that in phase-detection schemes based on EIT the detrimental (classical) effects of power-broadening and ac-Stark associated broadening are eliminated. In the following section we will discuss the fundamental quantum noise sources of such magnetometer schemes.
## IV quantum-noise limit of magnetic field detection via optical phase shifts in the presence of ac-Stark effects
The problem of sensitive detection of phase shifts is common in optics. On the quantum level, the sensitivity of such kind of measurements is restricted by (i) vacuum fluctuations in the system and (ii) self-action noise due to nonlinearities in the system, as for example caused by ac-Stark shifts. The simultaneous presence of both noises usually leads to an absolute limit of the sensitivity.
Let us discuss this problem for the particular case of optical magnetometry based on phase-shift measurements in an atomic medium. The ultimate limit for the smallest detectable phase shift is set by the generalized uncertainty relation between phase- $`\mathrm{\Delta }\varphi \varphi \varphi `$ and photon-number fluctuations $`\mathrm{\Delta }nnn`$ of the output field.
$$\mathrm{\Delta }\varphi ^2\mathrm{\Delta }n^21+\frac{1}{4}\{\mathrm{\Delta }\varphi ,\mathrm{\Delta }n\}^2,$$
(10)
where $`\{,\}`$ denotes the anti-commutator. If phase- and photon-number fluctuations are uncorrelated, the second term on the r.h.s. vanishes and one recovers the familiar Heisenberg relation. In any real magnetometer schemes phase and intensity fluctuations are however correlated due to e.g. ac-Stark shifts (self phase modulation), and thus the second term in Eq.(10) is in general nonzero. When the intensity-phase coupling is small, it can be characterized by a linear coupling coefficient $`\beta `$ in the form $`\mathrm{\Delta }\varphi =\mathrm{\Delta }\varphi _0+\beta \mathrm{\Delta }n`$, where $`\mathrm{\Delta }\varphi _0`$ denotes phase fluctuations not correlated to intensity fluctuations. Thus we find
$$\mathrm{\Delta }\varphi ^2\frac{1}{\mathrm{\Delta }n^2}+\beta ^2\mathrm{\Delta }n^2.$$
(11)
The signal phase accumulated during the propagation through an atomic vapor cell is proportional to the Zeeman splitting $`\mathrm{\Delta }\omega _B`$, the length of the cell $`L`$, and the dispersion of the real part of the susceptibility at the laser frequency $`\mathrm{d}\chi ^{}/\mathrm{d}\omega `$. The cell length is restricted by the absorption at the laser frequency, and a reasonable upper limit for $`L`$ is the (amplitude) absorption length $`L_{\mathrm{abs}}=(\pi \chi ^{\prime \prime }/\lambda )^1`$.
Thus the maximum phase shift is
$$\mathrm{\Delta }\varphi |_{\mathrm{max}}=\frac{1}{\chi ^{\prime \prime }}\frac{\mathrm{d}\chi ^{}}{\mathrm{d}\omega }\mathrm{\Delta }\omega _B.$$
(12)
One recognizes, that the sensitivity of phase measurements to Zeeman shifts is determined by the dispersion-absorption ratio $`\left(1/\chi ^{\prime \prime }\right)\mathrm{d}\chi ^{}/\mathrm{d}\omega `$.
The limit for the smallest detectable Zeeman shift is therefore given by
$$\mathrm{\Delta }\omega _B|_{\mathrm{min}}=\left[\frac{1}{\chi ^{\prime \prime }}\frac{\mathrm{d}\chi ^{}}{\mathrm{d}\omega }\right]^1\left[\mathrm{\Delta }n^2^1+\beta ^2\mathrm{\Delta }n^2\right]^{1/2}.$$
(13)
Under the condition, that the dispersion-absorption ratio is independent on the laser power, the r.h.s. of this expression is minimized when $`\mathrm{\Delta }n^2|_{\mathrm{opt}}=\beta ^1`$. Therefore there is an absolute lower limit or โquantum limitโ of magnetic field detection via phase-shift measurements independent on the photon-number fluctuations
$$\mathrm{\Delta }\omega _B|_{\mathrm{min}}=\left[\frac{1}{\chi ^{\prime \prime }}\frac{\mathrm{d}\chi ^{}}{\mathrm{d}\omega }\right]^1\sqrt{2\beta }.$$
(14)
The absorption-dispersion ratio of a magnetic resonance is usually given by its natural width, which can be rather small if a two-photon Raman process between Zeeman- or hyperfine components is used as in an EIT magnetometer.
We will show later on that different measurement strategies as well as the use of non-classical light fields do in general not improve this result.
## V EIT-based Faraday magnetometer
Let us now discuss in detail an EIT magnetometer in resonant nonlinear Faraday configuration. For this we consider the propagation of a strong, linear polarized light field through an optically dense medium, consisting of resonant $`\mathrm{\Lambda }`$-type systems (atoms, quantum wells etc.) as shown in Fig. 2. For simplicity we ignore optical pumping into lower states other than those shown in the figure and assume a closed system. For a resonant $`J=1J=0`$ transition (say), optical pumping into the lower $`m_J=0`$ state depletes both states $`m_J=\pm 1`$ in the same way and thus effectively diminishes the optical density but does not affect the signal. Symmetric re-pumping can be used to maintain the population in the relevant sub-system without affecting the detection scheme. We include a dephasing of the ground-state coherence with rate $`\gamma _0`$ and a population exchange rate between the ground states $`\gamma _{0r}`$.
The two circular components $`E_{}`$ and $`E_+`$ of the linear polarized light generate a coherent superposition (dark state) of the states $`|b_\pm |J=1,m_J=\pm 1`$. A magnetic field parallel to the propagation axis leads to an anti-symmetric level shift of $`|b_\pm `$ and thus by virtue of the large linear dispersion at an EIT-resonance to an opposite change in the index of refraction for both components. As a result the polarization direction is rotated, which is the so-called resonant nonlinear Faraday effect . The difference to the linear Faraday effect is the presence of the intensity-dependent dark resonance generated by the action of the strong laser field as opposed to a usual absorption resonance in the weak-field limit. The rotation of the plane of polarization at the output can be measured by detecting the intensity difference of two linear polarized components $`\pm `$45<sup>o</sup> rotated with respect to the input polarization.
An aspect of the system, which becomes particularly important when strong fields are considered, are non-resonant couplings of the two circular components to other levels, which to lowest order give rise to ac-Stark shifts of the states $`|b_\pm `$. In a Faraday configuration the ac-Stark shifts of $`|b_+`$ and $`|b_{}`$ are exactly equal and opposite in sign due to symmetry and thus there is no average effect on the signal and no bias phase shift or rotation. Thus the Faraday magnetometer is not subject to systematic errors associated with ac-Stark shifts. However, as mentioned before, ac-Stark shifts cause a coupling between intensity and phase fluctuations which need to be taken into account.
### A Detection scheme
We here consider the detection scheme shown in Fig. 3. A strong linear polarized field initially polarized in $`x`$ direction propagates through a cell of length $`L`$ with the magneto-optic medium. Due to the nonlinear Faraday effect the plane of polarization is rotated by an angle $`\varphi /2`$.
In order to detect this angle the intensity difference of the two orthogonal output directions $`1`$ and $`2`$ is measured. The operator for the number of counts is given by
$$\widehat{n}=C_{t_m}dt\left(\widehat{E}_2^{}(t)\widehat{E}_2^+(t)\widehat{E}_1^{}(t)\widehat{E}_1^+(t)\right).$$
(15)
where $`\widehat{E}^\pm `$ denote the positive and negative frequency part of the output electric field operators, $`t_m`$ is the measurement time, and $`C=2ฯต_0cA/\mathrm{}\nu _0`$, $`A`$ being the beam cross-section and $`\nu _0`$ the resonance frequency. Making use of the field commutation relations $`[\widehat{E}_{1,2}^+(L,t),\widehat{E}_{1,2}^{}(L,t^{})]=C^1\delta (tt^{})`$ and $`[\widehat{E}_1^\pm ,\widehat{E}_2^\pm ]=0`$, we can express the mean number of counts as well as the fluctuations in terms of normal-ordered correlation functions. The latter allows to apply a c-number approach where the operators $`\widehat{E}`$ are approximated by stochastic complex functions $`E`$.
$`\widehat{n}`$ $`=`$ $`n_2n_1,`$ (16)
$`\mathrm{\Delta }\widehat{n}^2`$ $`=`$ $`\mathrm{\Delta }n^2+n_1+n_2.`$ (17)
where $`n_{1,2}`$ follows form Eq.(15) by replacing the field operators by c-numbers
$$n_{1,2}=C_{t_m}dtE_{1,2}^{}(L,t)E_{1,2}^+(L,t).$$
(18)
In the usual configuration only the $`x`$-polarized component of the input field is excited and we will restrict the discussion to a vacuum input of the $`y`$-polarized component. The propagation of the field through the magneto-optical medium is most conveniently described in terms of right and left circular components $`E_\pm =(1/\sqrt{2})\left(E_x\pm iE_y\right)`$, and we therefore have
$$n=iC_{t_m}dt\left(E_{}^{}E_+^+E_+^{}E_{}^+\right).$$
(19)
The propagation of the circular components can be characterized by two parameters, the intensity transmission coefficient $`\eta `$ and the phase shift $`\varphi _\pm (L,t)`$ of the respective component at the output.
$`E_\pm ^+(L,t)`$ $`=`$ $`E_\pm ^+(0,t)\sqrt{\eta }\mathrm{e}^{i\varphi _\pm (L,t)}.`$ (20)
In the limit of small magnetic fields the absorption of both circular components is identical for symmetry reasons i.e. there is no dichroism. With this we obtain for cw-input fields
$`\widehat{n}=\eta n_x_{\mathrm{in}}\mathrm{sin}\varphi _{\mathrm{sig}}\eta n_x_{\mathrm{in}}\varphi _{\mathrm{sig}},`$ (21)
where $`\varphi _{\mathrm{sig}}=\varphi _+(L)\varphi _{}(L)`$ is the (stationary) signal phase shift. Similarly we can estimate the fluctuations in lowest order of the small rotation angle $`\varphi `$ in the case of an initially coherent field
$`\mathrm{\Delta }\widehat{n}^2=\eta n_x_{\mathrm{in}}+\eta ^2n_x_{\mathrm{in}}^2\delta \varphi ^2.`$ (22)
The first term corresponds to the vacuum noise level and the second term proportional to
$$\delta \varphi ^2=\frac{1}{t_m^2}dtdt^{}\delta \varphi (t),\delta \varphi (t^{})$$
(23)
describes fluctuations due to an intensity-phase noise coupling in the medium. ($`a,b[aa][bb]`$)
In the following we calculate the loss factor $`\eta `$, the signal phase shift $`\varphi _{\mathrm{sig}}`$ and the fluctuations $`\delta \varphi ^2`$ due to the intensity-phase noise coupling for the EIT-Faraday magnetometer.
### B Medium susceptibility and field propagation
The stationary propagation of the right and left circular polarized electric field components through the atomic vapor is described by Maxwell equations in slowly-varying amplitude and phase approximation
$$\frac{\mathrm{d}}{\mathrm{d}z}E_\pm ^+(z)=\frac{i\nu _0}{2cฯต_0}\mathrm{}_\pm N\sigma _{b\pm a}(z).$$
(24)
$`N`$ is the atomic number density, $`\mathrm{}_\pm `$ are the dipole moments of the respective transitions, and $`\sigma _{b\pm a}`$ are the c-number analogues of the atomic lowering operators $`\widehat{\sigma }_{b\pm a}=|b_\pm a|`$. Analytic expressions for $`\sigma _{b\pm a}`$ can be obtained from the stationary solution of the c-number Bloch equations for the atomic populations
$`\dot{\sigma }_{bb}`$ $`=`$ $`\gamma _{0\mathrm{r}}(\sigma _{bb}\sigma _{b+b+})+\gamma _\mathrm{r}\sigma _{aa}`$ (26)
$`i(\mathrm{\Omega }_{}\sigma _{ab}c.c.),`$
$`\dot{\sigma }_{b+b+}`$ $`=`$ $`\gamma _{0\mathrm{r}}(\sigma _{bb}\sigma _{b+b+})+\gamma _\mathrm{r}\sigma _{aa}`$ (28)
$`i(\mathrm{\Omega }_+\sigma _{ab+}c.c.),`$
and polarizations
$`\dot{\sigma }_{ab\pm }`$ $`=`$ $`\mathrm{\Gamma }_{ab\pm }\sigma _{ab\pm }i\mathrm{\Omega }_\pm ^{}(\sigma _{b\pm b\pm }\sigma _{aa})`$ (30)
$`i\mathrm{\Omega }_{}^{}\sigma _{bb\pm },`$
$`\dot{\sigma }_{bb+}`$ $`=`$ $`\mathrm{\Gamma }_{bb+}\sigma _{bb+}i\mathrm{\Omega }_{}\sigma _{ab_+}+i\mathrm{\Omega }_+^{}\sigma _{ba},`$ (31)
where
$`\mathrm{\Gamma }_{ab\pm }`$ $``$ $`\gamma +{\displaystyle \frac{\gamma _{0\mathrm{r}}}{2}}+i\left(\mathrm{\Delta }+\delta _\pm \pm {\displaystyle \frac{\delta _0}{2}}\right),`$ (32)
$`\mathrm{\Gamma }_{bb+}`$ $``$ $`\gamma _0+\gamma _{0\mathrm{r}}+i\left(\delta _0+\delta _+\delta _{}\right).`$ (33)
$`\gamma _\mathrm{r}`$ is the radiative linewidth of the transitions $`|a|b_\pm `$, and $`\gamma `$ is the homogeneous transverse linewidth of the optical transitions $`|a|b_\pm `$. $`\delta _0`$ is the Zeeman splitting and $`\delta _\pm `$ are the ac-Stark shifts of levels $`|b_\pm `$. $`\mathrm{\Omega }_\pm `$ are the complex Rabi-frequencies of the two optical fields, $`\mathrm{\Omega }_\pm =\mathrm{}_\pm E_\pm ^{}/\mathrm{}`$. We have disregarded Langevin noise forces in Eqs.(26-31) associated with spontaneous emission and collisional decay processes, since it was shown in that atomic noises have a negligible effect on the magnetometer sensitivity.
We calculate the stationary solutions of the Bloch-equations by considering only the lowest order in $`\gamma _0`$, $`\gamma _{0\mathrm{r}}`$, $`\delta _0`$ and $`\delta _\pm `$. In this limit we find
$`\sigma _{ab_\pm }`$ $`=`$ $`{\displaystyle \frac{i\mathrm{\Omega }_\pm (\gamma _0|\mathrm{\Omega }_{}|^2+\gamma _{0\mathrm{r}}|\mathrm{\Omega }_\pm |^2)}{|\mathrm{\Omega }|^2(2\gamma (2\gamma _{0\mathrm{r}}+\gamma _0)+|\mathrm{\Omega }|^2)}}`$ (36)
$`\left(\delta _\pm \pm {\displaystyle \frac{\delta _0}{2}}\right){\displaystyle \frac{2\mathrm{\Omega }_\pm |\mathrm{\Omega }_{}|^2}{|\mathrm{\Omega }|^2(2\gamma (2\gamma _{0\mathrm{r}}+\gamma _0)+|\mathrm{\Omega }|^2)}}`$
$`+{\displaystyle \frac{\mathrm{\Delta }}{\gamma }}{\displaystyle \frac{\mathrm{\Omega }_\pm \left[\gamma _{0\mathrm{r}}(|\mathrm{\Omega }_+|^4+|\mathrm{\Omega }_{}|^4)+2\gamma _0|\mathrm{\Omega }_{}|^2|\mathrm{\Omega }|^2\right]}{|\mathrm{\Omega }|^4(2\gamma (2\gamma _{0\mathrm{r}}+\gamma _0)+|\mathrm{\Omega }|^2)}}`$
where $`|\mathrm{\Omega }(z)|^2=|\mathrm{\Omega }_{}(z)|^2+|\mathrm{\Omega }_+(z)|^2`$. Usually the coherence decay between the ground levels dominates the population exchange and thus $`\gamma _0\gamma _{0\mathrm{r}}`$.
It is convenient to separately consider the spatial evolution of amplitudes and phases of the complex Rabi-frequencies $`\mathrm{\Omega }_\pm (z)=|\mathrm{\Omega }_\pm (z)|\mathrm{e}^{i\varphi _\pm (z)}`$. The intensities of the two fields are attenuated in the same way
$`{\displaystyle \frac{\mathrm{d}}{\mathrm{d}z}}|\mathrm{\Omega }_\pm |^2=\kappa {\displaystyle \frac{\gamma _0\gamma _\mathrm{r}}{|\mathrm{\Omega }|^2}}{\displaystyle \frac{|\mathrm{\Omega }_+|^2|\mathrm{\Omega }_{}|^2}{(2\gamma _0\gamma +|\mathrm{\Omega }|^2)}},`$ (37)
where $`\kappa =(3/4\pi )N\lambda ^2`$.
Eq. (37) can be easily solved when the length $`L`$ of the cell is small enough, such that $`|\mathrm{\Omega }(L)|^22\gamma \gamma _0`$. In the Faraday set-up discussed here $`\mathrm{\Omega }_\pm (0)=\mathrm{\Omega }(0)/\sqrt{2}`$, and therefore $`|\mathrm{\Omega }_\pm (z)|^2=|\mathrm{\Omega }(z)|^2/2`$. We thus arrive at
$`|\mathrm{\Omega }(z)|^2`$ $`=`$ $`|\mathrm{\Omega }(0)|^2\left(1{\displaystyle \frac{\gamma _0\gamma _\mathrm{r}\kappa z}{2|\mathrm{\Omega }(0)|^2}}\right)`$ (38)
$`=`$ $`|\mathrm{\Omega }(0)|^2\left(1\alpha _0z\right).`$ (39)
It is interesting to note that under conditions of EIT the residual absorption is not exponential but linear. The intensity transmission coefficient is then given by
$$\eta =\left(1\alpha _0L\right).$$
(40)
The approximation $`|\mathrm{\Omega }(L)|^22\gamma \gamma _0`$ sets an upper limit for the losses, such that $`1\eta 2\gamma \gamma _0/|\mathrm{\Omega }(0)|^2`$.
Similarly we find the phase equations
$`{\displaystyle \frac{\mathrm{d}}{\mathrm{d}z}}\varphi _{}`$ $`=`$ $`{\displaystyle \frac{\kappa \gamma _\mathrm{r}}{2\gamma }}{\displaystyle \frac{\mathrm{\Delta }\gamma _0+\gamma (\delta _0/2\delta _{})}{2\gamma _0\gamma +|\mathrm{\Omega }|^2}},`$ (41)
$`{\displaystyle \frac{\mathrm{d}}{\mathrm{d}z}}\varphi _+`$ $`=`$ $`{\displaystyle \frac{\kappa \gamma _\mathrm{r}}{2\gamma }}{\displaystyle \frac{\mathrm{\Delta }\gamma _0\gamma (\delta _0/2+\delta _+)}{2\gamma _0\gamma +|\mathrm{\Omega }|^2}}.`$ (42)
The contributions from the one-photon detuning $`\mathrm{\Delta }`$ cancel when the relative phase $`\varphi =\varphi _+\varphi _{}`$ is considered
$$\frac{\mathrm{d}}{\mathrm{d}z}\varphi =\frac{\kappa \gamma _\mathrm{r}}{2}\left(\frac{\delta _0}{|\mathrm{\Omega }|^2}+\frac{\delta _+\delta _{}}{|\mathrm{\Omega }|^2}\right).$$
(43)
The first term describes the signal-phase shift due to a magnetic field and the second term the ac-Stark contribution. Integration of Eq.(43) yields for the signal
$$\varphi _{\mathrm{sig}}=\frac{\delta _0}{\gamma _0}\mathrm{ln}\left|\frac{\mathrm{\Omega }(0)}{\mathrm{\Omega }(L)}\right|^2$$
(44)
and the ac-Stark contribution
$$\delta \varphi (t)=\frac{\kappa \gamma _\mathrm{r}}{2}_0^Ldz\frac{\delta _+(z,t)\delta _{}(z,t)}{|\mathrm{\Omega }(z)|^2}.$$
(45)
### C Ac-Stark shifts and associated noise
Let us now discuss the average ac-Stark shift and the corresponding noise contributions. For this we first consider the effect of an off-resonant quantized field on the energy of a single atom in lowest non vanishing order of perturbation. We then generalize the results for the average ac-Stark shift and its fluctuations to an ensemble of atoms by making the physically reasonable assumption that ac-Stark shifts of different atoms are uncorrelated.
We decompose the Hamiltonian of the single atom interacting with the quantized field in a rotating frame in the form $`H=H_0+H_\mathrm{S}`$, where $`H_0`$ is the unperturbed part
$`H_0`$ $`=`$ $`H_0^{\mathrm{field}}+\mathrm{}\omega _{b_{}b_+}|b_{}b_{}|`$ (47)
$`+\mathrm{}\mathrm{\Delta }_{ab_+}|aa|+\mathrm{}{\displaystyle \underset{j}{}}\mathrm{\Delta }_j|c_jc_j|.`$
$`\mathrm{\Delta }_{ab_+}=\omega _{ab_+}\nu _0`$ and $`\mathrm{\Delta }_j=\omega _{c_jb_+}\nu _0`$ are the detunings of the $`|a|b_+`$ and $`|c_j|b_+`$ transitions.
$`H_\mathrm{S}=\mathrm{}_{}|ab_{}|\widehat{E}_{}^+\mathrm{}_+|ab_+|\widehat{E}_+^+`$ (48)
$`{\displaystyle \underset{j}{}}\left(\mathrm{}_{j+}|c_jb_+|\widehat{E}_+^++\mathrm{}_j|c_jb_{}|\widehat{E}_{}^+\right)+\mathrm{adj}.`$ (49)
describes the resonant and non-resonant couplings of the quantized fields to the atom. The non-resonant couplings to the excited states $`|c_j`$ cause ac-Stark shifts. We here have assumed that both fields are nearly monochromatic and have set the energy of level $`|b_+`$ equal to zero. $`\mathrm{}_{j\pm }`$ are the dipole moments of the transitions $`|c_j|b_\pm `$.
We proceed by formally eliminating the excited states $`|c_j`$ by means of a canonical transformation in second order perturbation
$$\stackrel{~}{H}=\mathrm{exp}(S)H\mathrm{exp}(S)H+[S,H]+[S,[S,H]],$$
(50)
where $`S`$ obeys the equation
$$[S,H_0]=\underset{j}{}(\mathrm{}_{j+}|c_jb_+|\widehat{E}_+^++\mathrm{}_j|c_jb_{}|\widehat{E}_{}^++\mathrm{adj}.)$$
(51)
Under conditions of exact two-photon resonance for the fields we obtain the transformation operator
$$S=\underset{j}{}(\frac{\mathrm{}_{j+}}{\mathrm{\Delta }_j}|c_jb_+|\widehat{E}_+^++\frac{\mathrm{}_j}{\mathrm{\Delta }_j}|c_jb_{}|\widehat{E}_{}^+\mathrm{adj}.).$$
(52)
Assuming that the population of all excited levels is small, we eventually find for the transformed Hamiltonian
$`\stackrel{~}{H}H_0\mathrm{}_+|ab_+|\widehat{E}_+^+\mathrm{}_{}|ab_{}|\widehat{E}_{}^+`$ (53)
$`{\displaystyle \underset{j}{}}\left({\displaystyle \frac{\mathrm{}_{j+}^2}{\mathrm{}\mathrm{\Delta }_j}}|b_+b_+|\widehat{E}_+^{}\widehat{E}_+^++{\displaystyle \frac{\mathrm{}_j^2}{\mathrm{}\mathrm{\Delta }_j}}|b_{}b_{}|\widehat{E}_{}^{}\widehat{E}_{}^+\right)`$ (54)
$`{\displaystyle \underset{j}{}}{\displaystyle \frac{\mathrm{}_{j+}\mathrm{}_j}{\mathrm{}\mathrm{\Delta }_j}}\left(|b_+b_{}|\widehat{E}_+^{}\widehat{E}_{}^++|b_{}b_+|\widehat{E}_{}^{}\widehat{E}_+^+\right).`$ (55)
Let us assume now, that $`\mathrm{\Delta }_j`$ is much larger than the natural width of the excited states, and therefore the population transfer due to the non-resonant coupling is negligible. We identify $`_j\mathrm{}_{j\pm }^2/\mathrm{\Delta }_j\mathrm{}^2/\mathrm{\Delta }_0`$, where $`\mathrm{\Delta }_0`$ is some effective detuning. The dipole moments $`\mathrm{}_{j+}`$ and $`\mathrm{}_j`$ have usually alternating signs for different excited states $`|c_ศท`$. We therefore set $`_j\mathrm{}_{j+}\mathrm{}_j_{}/\mathrm{\Delta }_j=0`$. Then the ac-Stark shift of the single atom can be represented by the operator expression
$$\widehat{\delta }_\pm ^l(t)=\frac{\mathrm{}^2}{\mathrm{}^2\mathrm{\Delta }_0}\widehat{E}_\pm ^{}(z_l,t)\widehat{E}_\pm ^+(z_l,t),$$
(56)
where $`l`$ specifies the atom and $`z_l`$ its location. Thus we find for the average ac-Stark shift
$$\widehat{\delta }_\pm ^l(t)=\frac{\mathrm{}^2}{\mathrm{}^2\mathrm{\Delta }_0}\widehat{E}_\pm ^{}(z_l,t)\widehat{E}_\pm ^+(z_l,t)=\frac{|\mathrm{\Omega }(z_l,t)|^2}{2\mathrm{\Delta }_0},$$
(57)
where $`\mathrm{}^2|\widehat{E}_\pm (z_l,t)|^2/\mathrm{}^2=\mathrm{}^2|E(z_l,t)|^2/2\mathrm{}^2=|\mathrm{\Omega }(z_l,t)|^2/2`$. Similarly we obtain for the second-order moments of the ac-Stark shifts $`\widehat{x},\widehat{y}\left(\widehat{x}\widehat{x}\right)\left(\widehat{y}\widehat{y}\right)`$
$`\widehat{\delta }_\pm ^l(t)\widehat{\delta }_\pm ^l(t^{})`$ (58)
$`={\displaystyle \frac{\mathrm{}^4}{\mathrm{}^4\mathrm{\Delta }_0^2}}\widehat{E}_\pm ^{}(z_l,t)\widehat{E}_\pm ^+(z_l,t)\widehat{E}_\pm ^{}(z_l,t^{})\widehat{E}_\pm ^+(z_l,t^{}),`$ (59)
$`\widehat{\delta }_+^l(t)\widehat{\delta }_{}^l(t^{})`$ (60)
$`={\displaystyle \frac{\mathrm{}^4}{\mathrm{}^4\mathrm{\Delta }_0^2}}\widehat{E}_+^{}(z_l,t)\widehat{E}_+^+(z_l,t)\widehat{E}_{}^{}(z_l,t^{})\widehat{E}_{}^+(z_l,t^{}),`$ (61)
or after normal ordering
$`\widehat{\delta }_\pm ^l(t)\widehat{\delta }_\pm ^l(t^{})=`$ (62)
$`{\displaystyle \frac{\mathrm{}^4}{\mathrm{}^4\mathrm{\Delta }_0^2}}[E_\pm ^{}(z_l,t)E_\pm ^+(z_l,t)E_\pm ^{}(z_l,t^{})E_\pm ^+(z_l,t^{})`$ (63)
$`+{\displaystyle \frac{\delta (tt^{})}{C}}E_\pm ^{}(z_l,t)E_\pm ^+(z_l,t)],`$ (64)
$`\widehat{\delta }_+^l(t)\widehat{\delta }_{}^l(t^{})=`$ (65)
$`{\displaystyle \frac{\mathrm{}^4}{\mathrm{}^4\mathrm{\Delta }_0^2}}\left[E_+^{}(z_l,t)E_+^+(z_l,t)E_{}^{}(z_l,t^{})E_{}^+(z_l,t^{})\right].`$ (66)
The first terms in Eqs.(64) and (66) correspond to classical fluctuations, while the second term in (64) is vacuum or shot noise. If the applied fields are in a coherent state only the shot noise term survives. In any practical realizations there are however large excess noise contributions and the first terms are usually the dominant ones. We will show that all excess noise contributions are canceled in a Faraday magnetometer and only the vacuum contribution survives.
We generalize the above single-atom results to an ensemble of atoms assuming independent fluctuations of the ac-Stark shifts of different atoms, i.e.
$$\widehat{\delta }_\mu ^j\widehat{\delta }_\nu ^k\delta _{jk},$$
(67)
where $`\{\mu ,\nu \}\{+,\}`$. We introduce the continuous variable
$$\widehat{\delta }_\pm (z,t)=L\underset{j}{}\delta (zz_j)\widehat{\delta }_\pm ^j(t).$$
(68)
In a continuum approximation, $`_j(1/L)_Ldz`$, and we have
$$\widehat{\delta }_\pm (z,t)=\frac{\mathrm{}^2}{\mathrm{}^2\mathrm{\Delta }_0}\widehat{E}_\pm ^{}(z,t)\widehat{E}_\pm ^+(z,t)=\frac{|\mathrm{\Omega }(z,t)|^2}{2\mathrm{\Delta }_0}.$$
(69)
Similarly
$`\widehat{\delta }_\pm (z,t),\widehat{\delta }_\pm (z^{},t^{})={\displaystyle \frac{L\mathrm{}^4}{\mathrm{}^4\mathrm{\Delta }_0^2}}\delta (zz^{})`$ (70)
$`\times [E_\pm ^{}(z,t)E_\pm ^+(z,t),E_\pm ^{}(z,t^{})E_\pm ^+(z,t^{})`$ (71)
$`+{\displaystyle \frac{\delta (tt^{})}{C}}E_\pm ^{}(z,t)E_\pm ^+(z,t)],`$ (72)
and
$`\widehat{\delta }_+(z,t),\widehat{\delta }_{}(z^{},t^{})={\displaystyle \frac{L\mathrm{}^4}{\mathrm{}^4\mathrm{\Delta }_0^2}}\delta (zz^{})`$ (73)
$`\times \left[E_\pm ^{}(z,t)E_\pm ^+(z,t),E_\pm ^{}(z,t^{})E_\pm ^+(z,t^{})\right].`$ (74)
We here have used that in continuum approximation for any smooth function $`f(z)`$ holds
$$L\underset{l}{}\delta (zz_l)\delta (z^{}z_l)f(z_l)=\delta (zz^{})f(z).$$
(75)
It is now straight forward to evaluate the quadratic deviation of the relative ac-Stark shift
$`\left({\displaystyle \frac{\widehat{\delta }_+(z,t)\widehat{\delta }_{}(z,t)}{2|\mathrm{\Omega }(z)|^2}}\right),\left({\displaystyle \frac{\widehat{\delta }_+(z^{},t^{})\widehat{\delta }_{}(z^{},t^{})}{2|\mathrm{\Omega }(z^{})|^2}}\right)=`$ (76)
$`\delta (zz^{})\delta (tt^{}){\displaystyle \frac{\mathrm{}^2L}{2\mathrm{}^2C\mathrm{\Delta }_0^2|\mathrm{\Omega }(z)|^2}}.`$ (77)
We note that the classical excess noise contributions exactly cancel and only the vacuum contribution is left over. Due to the intrinsic balancing in the EIT-Faraday magnetometer excess noise contributions are automatically canceled. This is an important advantage of the Faraday configuration as compared to the asymmetric EIT-magnetometer discussed in and .
Using Eqs.(23), (45) and (77) we eventually find for the phase fluctuations due to ac-Stark shifts
$$\delta \varphi ^2=\frac{1}{t_m}\frac{\kappa ^2\gamma _\mathrm{r}^2}{4\mathrm{\Delta }_0^2}\frac{\mathrm{}^2L}{\mathrm{}^2C}_0^L\frac{1}{|\mathrm{\Omega }(z)|^2}๐z$$
(78)
### D Signal-to-noise ratio and minimum detectable Zeeman shift
The minimum detectable Zeeman shift is obtained by setting the mean number of counts
$$\widehat{n}=\eta n_x_{\mathrm{in}}\varphi _{\mathrm{sig}}=\eta n_x_{\mathrm{in}}\frac{\delta _0}{\gamma _0}\mathrm{ln}\left[\eta ^1\right]$$
(79)
equal to the quantum mechanical uncertainty
$`\mathrm{\Delta }\widehat{n}^2^{1/2}=\left[\eta n_x_{\mathrm{in}}+\eta ^2n_x_{\mathrm{in}}^2\delta \varphi ^2\right]^{1/2}`$ (80)
$`=\sqrt{\eta n_x_{\mathrm{in}}}\left[1+{\displaystyle \frac{|\mathrm{\Omega }(0)|^4}{\mathrm{\Delta }_0^2\gamma _0^2}}\eta (1\eta )\mathrm{ln}(\eta ^1)\right]^{1/2}`$ (81)
This yields the signal-to-noise ratio
$$\mathrm{SNR}=\left[\frac{{\displaystyle \frac{\delta _0^2}{\gamma _0^2}}n_x_{\mathrm{in}}\eta \mathrm{ln}^2(\eta ^1)}{1+{\displaystyle \frac{|\mathrm{\Omega }(0)|^4}{\mathrm{\Delta }_0^2\gamma _0^2}}\eta (1\eta )\mathrm{ln}(\eta ^1)}\right]^{1/2},$$
(82)
which is maximized for an optimal power of the field corresponding to
$$|\mathrm{\Omega }(0)|_{\mathrm{opt}}^2=\sqrt{\frac{\mathrm{\Delta }_0^2\gamma _0^2}{\eta (1\eta )\mathrm{ln}(\eta ^1)}}\mathrm{\Delta }_0\gamma _0.$$
(83)
Substituting the optimum Rabi-frequency (83) into (82) yields a maximum SNR for $`\eta 0.06`$. Thus we find the quantum limit for the detection of Zeeman level shifts
$$\delta _0^{\mathrm{SQL}}=\gamma _0f\left(\frac{\gamma _r}{\mathrm{\Delta }_0}\frac{3}{8\pi }\frac{\lambda ^2}{A}\frac{1}{\gamma _0t_m}\right)^{1/2},$$
(84)
where
$$f\left(\frac{1\eta }{\eta \mathrm{ln}^3(\eta ^1)}\right)^{1/4}$$
(85)
is a numerical factor which varies between 1 and 2 for $`\eta =0.01\mathrm{}0.8`$. (Note that $`\eta `$ is the transmission coefficient under conditions of EIT. Without EIT the medium would be totally opaque.) In Fig. 4 we have shown the minimum detectable Zeeman splitting (proportional to the magnetic field) as function of the laser input power for different transmission coefficients.
One clearly sees that for small laser powers shot-noise is dominant, while for larger laser powers ac-Stark associated fluctuations take over. Also shown is the saturation behavior of an OPM . Due to power broadening the sensitivity of an OPM saturates as soon as the Rabi-frequency reaches the value $`\sqrt{\gamma \gamma _0}`$. In the EIT-Faraday magnetometer, on the other hand, the optimum Rabi-frequency corresponding to the quantum limit is of the order of $`\sqrt{\mathrm{\Delta }_0\gamma _0}`$. Since $`\mathrm{\Delta }_0\gamma `$ much higher sensitivities can be achieved here.
### E Compensation of ac-Stark associated noise by use of non-classical input fields
It is well known, that the effect of self-phase modulation due to refractive nonlinearities can be compensated, at least in part, by means of an optimum detection procedure (for example, by measuring not the phase, but an appropriately chosen quadrature amplitude of the probe electromagnetic wave) and/or by using non-classical light . The properties of the input quantum state in the methods utilizing non-classical light are thereby chosen such that after the interaction the probe wave is in the coherent or phase-squeezed state.
In the case of an optical magnetometer, ac-Stark shifts appear due to non-resonant nonlinearities and it would seem that these shifts can in principle be compensated by an adapted measurement strategy and the use of non-classical light. An essential condition for such methods is however that the system is nearly lossless in order to preserve the non-classical state of light. On the other hand, as discussed above, the maximum signal in an optical magnetometer is achieved under conditions of substantial absorption. (We note that the SNR is proportional to ln$`(\eta )^2`$.) We will show in the following with simple estimates that this feature makes it impossible to increase the sensitivity by using non-classical light.
Let us consider the simplest example of compensation of ac-Stark associated noise by non-classical light. We assume, that the slowly varying field operators in the Heisenberg picture are represented in the form $`\widehat{E}_\pm =\widehat{E}+\widehat{e}_\pm `$, where $`\widehat{e}_\pm `$ is the fluctuation part. To discuss the compensation of ac-Stark effects let us disregard the resonant coupling with the medium and the associated absorption. Then we find that the field fluctuations at the end of the vapor cell can be written as
$`\widehat{e}_{}(t,L)=\widehat{e}_{}(t,0)+i{\displaystyle \frac{\kappa \gamma _\mathrm{r}L}{2\mathrm{\Delta }_0}}\left[\widehat{e}_{}(t,0)+\widehat{e}_{}^+(t,0)\right],`$ (86)
$`\widehat{e}_+(t,L)=\widehat{e}_+(t,0)i{\displaystyle \frac{\kappa \gamma _\mathrm{r}L}{2\mathrm{\Delta }_0}}\left[\widehat{e}_+(t,0)+\widehat{e}_+^+(t,0)\right].`$ (87)
The second terms in these equations are due to ac-Stark shifts. One can see that the uncertainty of the phase difference increases as a result of ac-Stark shifts, which leads to the sensitivity restriction, discussed above.
Let us assume now, that the incident field is squeezed in such a way, that the operators of the field fluctuations at the input obey the relations
$`\widehat{e}_{}(t,0)=\stackrel{~}{e}_{}i{\displaystyle \frac{\kappa \gamma _\mathrm{r}L}{2\mathrm{\Delta }_0}}(\stackrel{~}{e}_{}+\stackrel{~}{e}_{}^+),`$ (88)
$`\widehat{e}_+(t,0)=\stackrel{~}{e}_++i{\displaystyle \frac{\kappa \gamma _\mathrm{r}L}{2\mathrm{\Delta }_0}}(\stackrel{~}{e}_++\stackrel{~}{e}_+^+).`$ (89)
Here $`\stackrel{~}{e}_\pm `$ are free-field operators (the corresponding state is the field vacuum), which obey the commutation relations $`[\stackrel{~}{e}_\pm ^+(t),\stackrel{~}{e}_\pm ^{}(t^{})]=C^1\delta (tt^{})`$ and $`[\stackrel{~}{e}_\pm ^\pm (t),\stackrel{~}{e}_\pm ^\pm (t^{})]=0`$. Then, in the absence of losses, the effects of ac-Stark shifts are completely compensated in the output and the output fields are coherent.
$`\widehat{e}_{}(t,L)=\stackrel{~}{e}_{},`$ (90)
$`\widehat{e}_+(t,L)=\stackrel{~}{e}_+.`$ (91)
The sensitivity of the phase measurement would thus be determined by shot-noise only, $`\mathrm{\Delta }\varphi =1/\sqrt{n}`$.
In the absence of losses, the sensitivity of the detection can even be better than the shot-noise limit, if the initial state of the field is appropriately chosen . Making use of a SU(2) Lie-group description, Yurke showed that the sensitivity of a phase shift measurement in a Mach-Zehnder interferometer can approach the so-called Heisenberg limit $`\mathrm{\Delta }\varphi 1/n`$, where $`n`$ is the total number of registered quanta .
However, in the presence of losses resulting from the resonant coupling the noise compensation by means of non-classical light is only partial due to unwanted noises added by the medium. Taking into account linear losses and assuming, that the entrance field is squeezed in the way discussed above, we can rewrite the equation for the residual noises in the phase as follows:
$$\delta \varphi (t)=\frac{\kappa \gamma _\mathrm{r}}{2}_0^Ldz\frac{\delta _+(z,t)\delta _{}(z,t)}{|\mathrm{\Omega }(z)|^2}\sqrt{1\eta (z)}.$$
(92)
$`\eta (z)=1\alpha _0z`$ is the $`z`$-dependent transmission coefficient. The expression indicates, that for small losses in the medium, the noise can be almost completely suppressed. A maximum signal is achieved however when $`\eta 1`$ and thus the use of non-classical light only leads to a marginal reduction of the ac-Stark associated noise. This is in contrast to the measurement schemes discussed in which utilize squeezing to improve sensitivity. The change of the expression for the ac-Stark associated noise leads to a change of the sensitivity factor $`f`$ according to
$$f\stackrel{~}{f}=\left(\frac{(1\eta )(\mathrm{ln}(\eta ^1)+\eta 1)}{\eta \mathrm{ln}^4(\eta ^1)}\right)^{1/4}.$$
(93)
It is easy to see, that $`\stackrel{~}{f}f`$ for all relevant values of $`\eta `$, which means that squeezing does not improve the sensitivity of the detection.
The same conclusion can be drawn for any kind of optimal strategy of measurement to compensate ac Stark shifts. The main reason for this is that both, the magnitude of the signal and absorption losses increase with the density-length product of the atomic vapor cell.
## VI Summary
We have discussed the influence of ac-Stark shifts on the sensitivity of optical magnetometers. We have shown that these shifts cause a broadening of the relevant resonances and give rise to additional noise contributions. In absorption-type magnetometers, such as OPMs, the ac-Stark associated broadening as well as power-broadening lead to a reduction of the signal. We have shown that the classical part of these effects can be completely compensated in an EIT magnetometer in Faraday configuration where polarization rotation or, equivalently, the relative phase shift of two circular components is measured.
In a magnetometer based on phase measurements ac-Stark shifts lead also to a coupling between intensity and phase fluctuations. As a result there are additional, ac-Stark associated fluctuations which dominate over shot noise beyond a critical laser power. For a certain optimum intensity the fundamental signal-to-noise ratio attains a maximum value which represents the standard quantum limit of optical magnetometer based on phase-shift measurements. This quantum limit is determined by the dispersion-absorption ratio of the atomic medium and the strength of the intensity-phase noise coupling. The unique property of EIT is to provide a dispersion-absorption ratio which is independent of power-broadening and is given by the lifetime of a ground-state coherence. The minimum magnetic level shift corresponding to the quantum limit of EIT magnetometers can thus be orders of magnitude smaller than that of optical pumping devices.
We have shown that the best candidate to reach the standard quantum limit is a magnetometer in Faraday configuration, which has been analyzed in detail. In an EIT-Faraday magnetometer the signal reduction due to power- and ac-Stark broadenings is compensated by large densities of the atomic vapor. The influence of classical excess noise is completely eliminated due to symmetry and there are much less sources for systematic errors. We have also shown that the use of non-classical light and different detection techniques only marginally improves the attainable sensitivity since a maximum signal is associated with substantial losses in the atomic medium.
## Acknowledgements
The authors would like to thank M. Lukin for stimulating discussions on the role of ac-Stark shifts. A.M. and M.O.S. gratefully acknowledge further useful discussions with Y. Rostovtsev and the support from the Office of Naval Research, the National Science Foundation, the Welch Foundation, the Texas Advanced Research and Technology Program and the Air Force Research Laboratories. M.F. gratefully acknowledges the financial support of the Alexander-von-Humboldt foundation through the Feodor-Lynen Program. |
warning/0001/math-ph0001032.html | ar5iv | text | # 1 Introduction.
## 1 Introduction.
The relation between the affine Jacobian and integrable models is well known (cf. ). In the paper we have shown that the algebra of functions on the affine Jacobian is generated by action of hamiltonian vector fields from finite number of functions. The latter functions are coefficients of highest non-vanishing cohomologies of the affine Jacobian. Actually, the idea that such description of the algebra of functions is possible appeared from the paper which considers the structure of the algebra of observables for quantum and classical Toda chain.
In the present paper we give quantum version of . A quantum mechanical model is formulated which gives a quantization of the affine Jacobian. As usual in Quantum Mechanics we can describe not the variety itself but the algebra of functions on it (observables). We need to show that the quantum algebra of observables possesses essential property of corresponding classical algebra of functions. In our case this property is the possibility of creating every observable from a finite number of observables (cohomologies) by action of Hamiltonians.
In the process of realization of this program we find the Baxter equations which describe the spectrum of the model. It happens that these equations possess the property of duality: there is dual model with inverse Planck constant for which the eigen-vectors are the same. The algebras of observables of two dual models commute. The next ingredient of our study is the method of separation of variables developed by Sklyanin . Using this method we present the matrix elements of any observable in terms of certain integrals.
We show that the integrals in question are expresses in terms of deformed Abelian integrals (cf.). The observables for both dual models are defined in terms of cohomologies. The most beautiful feature of our construction is that in these cohomologies enter the integrals for matrix elements in such a way that the cohomologies of dual model play role of homologies for original one and vise a versa. We consider this relation between week-strong duality in quantum theory with duality between homologies and cohomologies as the most important conclusion of this paper.
## 2 Affine Jacobian.
In this section we briefly summarize necessary facts concerning relation between integrable models and algebraic geometry following the paper . The reason for repeating certain facts from is that we shall need them in slightly different situation.
Consider $`2\times 2`$ matrix which depends polynomially on the parameter $`z`$:
$`m(z)=\left(\begin{array}{cc}a(z)& b(z)\\ c(z)& d(z)\end{array}\right)`$
where the matrix elements are polynomials of the form:
$`a(z)=z^{g+1}+a_1z^g+\mathrm{}+a_{g+1},`$ (1)
$`b(z)=z^g+b_1z^{g1}+\mathrm{}+b_g,`$
$`c(z)=c_2z^g+c_3z^{g1}\mathrm{}+c_{g+2},`$
$`d(z)=d_2z^{g1}+d_3z^{g2}\mathrm{}+d_{g+1}`$
In the the affine space $`^{4g+2}`$ with coordinates $`a_1,\mathrm{},a_{g+1}`$, $`b_1,\mathrm{},b_g`$, $`c_2,\mathrm{},c_{g+2}`$, $`d_2,\mathrm{},d_{g+1}`$ consider the (2g+1)-dimensional affine variety $``$ defined as quadric
$`f(z)a(z)d(z)b(z)c(z)=1`$ (2)
We consider this simplest situation, but in principle it is possible to put arbitrary polynomial of degree $`2g`$ in RHS.
On the quadric $``$ let us consider the sections $`J_{\text{aff}}(t)`$ defined by the equations:
$`a(z)+d(z)=t(z)`$ (3)
where $`t(z)`$ is given polynomial of the form:
$`t(z)=z^{g+1}+z^gt_1+\mathrm{}+t_{g+1},`$ (4)
The notation $`J_{\text{aff}}(t)`$ stands for affine Jacobi variety. The definition of affine Jacobi variety and its equivalence to $`J_{\text{aff}}(t)`$ described above are given in the Appendix A. We include Appendix A because there is minor difference with the situation considered in and . The variety $``$ is foliated into the affine Jacobians $`J_{\text{aff}}(t)`$. Mechanical model described below provides a clever way of describing this foliation.
We would like to understand the geometrical meaning of quantum integrable models. The general philosophy teaches that in order to describe the quantization of a manifold one has to deform the algebra of functions on this manifold preserving certain essential properties of this algebra. The classical algebra must allow the Poisson structure in order that quantization is possible.
Certain Poisson brackets for the coefficients of matrix $`m(z)`$ can be introduced. We do not write them down explicitly, if needed they can be obtained taking classical limit of the commutation relations (14). The algebra
$$\widehat{๐}=[a_1,\mathrm{},a_{g+1},b_1,\mathrm{},b_g,c_2,\mathrm{},c_{g+2},d_2,\mathrm{},d_{g+1}]$$
becomes a Poisson algebra. The most important properties of this Poisson structure are the following. First, the coefficients of the determinant $`f(z)`$ belong to the center of the Poisson algebra, so, the equation (2) is consistent with Poisson structure. Second, the trace $`t(z)`$ generates commutative sub-algebra:
$$\{t(z),t(z^{})\}=0$$
It can be shown, actually, that the coefficient $`t_{g+1}`$ of the trace belongs to the center, it is convenient to put $`t_{g+1}=(1)^{g+1}2`$. The subject of our study is the algebra of functions on the $``$:
$$๐=\frac{\widehat{๐}}{\{f(z)=1,t_{g+1}=(1)^{g+1}2\}}$$
on which the Poisson structure is well defined.
The Poisson commutative algebra generated by the coefficients $`t_1,\mathrm{},t_g`$ is called the algebra of integrals of motion. Introduce the commuting vector-fields
$$_ig=\{t_i,g\},i=1,\mathrm{}g$$
The vector-fields $`_j`$ describe motion along the sub-varieties $`J_{\text{aff}}(t)`$.
One can think of these vector-fields as
$$_j=\frac{}{\tau _j}$$
where $`\tau _j`$ are โtimesโ corresponding to the integrals of motion $`t_j`$. Define the ring of integrals of motion
$`๐ฏ=[t_1,\mathrm{},t_g]`$ (5)
Introduce the space of differential forms $`C^k`$ with basis
$$xd\tau _{i_1}\mathrm{}d\tau _{i_k},x๐$$
and the differential
$$d=_jd\tau _j$$
Consider corresponding cohomologies $`H^k`$. In the paper the arguments are given in favour of the following Conjecture 1. The cohomologies $`H^k`$ are finite-dimensional over the ring $`๐ฏ`$, they are isomorphic to the cohomologies of the affine variety $`J_{\text{aff}}(t)`$ with $`t`$ in generic position. On the algebra $`๐`$ and on the spaces $`C^k`$ one can introduce degree . Take the basis of $`H^g`$ considered as a vector space over $`๐ฏ`$ which is composed of homogeneous representatives
$$\mathrm{\Omega }_\alpha =g_\alpha d\tau _1\mathrm{}d\tau _g$$
where $`\alpha `$ takes finite number of values. The fact of foliation of $``$ into varieties $`J_{\text{aff}}(t)`$ corresponds to the following statement concerning the algebra $`๐`$ Proposition 1. Every element $`x`$ of $`๐`$ can be presented in the form
$`x={\displaystyle \underset{\alpha }{}}p_\alpha (_1,\mathrm{},_g)g_\alpha `$ (6)
where $`p_\alpha (_1,\mathrm{},_g)`$ are polynomials of $`_1,\mathrm{},_g`$ with coefficients in $`๐ฏ`$. The representation (6) is not unique, the equations
$`{\displaystyle \underset{\alpha }{}}p_\alpha (_1,\mathrm{},_g)g_\alpha =0`$ (7)
are counted by $`H^{g1}`$ .
The formula (6) can be useful only if we are able to control the cohomologies. Concerning these cohomologies we adopt several conjectures following .
## 3 Conjectured form of cohomologies.
The affine variety $`J_{\text{aff}}(t)`$ allows the following description. Consider the hyper-elliptic curve $`X`$ of genus $`g`$:
$`w^2t(z)w+1=0`$ (8)
This curve has two points over the point $`z=\mathrm{}`$ which we denote by $`\mathrm{}^\pm `$.
Consider a matrix $`m(z)`$ satisfying (2). Take the zeros of $`b(z)`$:
$$b(z)=\underset{j=1}{\overset{g}{}}(zz_j)$$
and
$$w_j=d(z_j)$$
Obviously $`z_j,w_j`$ satisfy the equation of the curve $`X`$ (8). Thus $`m(z)`$ defines a point $`๐ซ`$ (divisor) on the symmetrized $`g`$-th power $`X[g]`$ of the curve $`X`$. The divisor $`๐ซ`$ consists of the points $`p_j=(z_j,w_j)X`$.
Oppositely one can reconstruct $`m(z)`$ starting form the divisor $`๐ซ`$. Corresponding map is singular, the singularities being located on
$`D=\{๐ซ|p_i=\sigma (p_j)\text{for some}i,j\text{or}p_i=\mathrm{}^\pm \text{for some}i\}`$ (9)
Where $`\sigma `$ is hyper-elliptic involution. Thus the alternative description of $`J_{\text{aff}}(t)`$ is
$$J_{\text{aff}}(t)=X[g]D$$
Consider the meromorphic differentials on $`X`$ with singularities at $`\mathrm{}^\pm `$. We chose the following basis of these differentials:
$`\mu _k(p)=z^{g+k}{\displaystyle \frac{dz}{y}},gk0`$
$`\mu _k(p)=\left[y{\displaystyle \frac{d}{dz}}(z^{kg1}y)\right]_{}{\displaystyle \frac{dz}{y}},k1`$ (10)
where $`p=(z,w)`$, $`y=2wt(z)`$, $`[]_{}`$ means that only non-negative degrees of Laurent series in the brackets are taken.
The form
$$\stackrel{~}{\mu }_k=\underset{i}{}\mu _k(p_i)$$
is viewed as a form on $`J_{\text{aff}}(t)`$. It is easy to see that the forms $`\mu _k`$ (hence $`\stackrel{~}{\mu }_k`$) with $`kg+1`$ are exact. Consider the space $`W^m`$ with the basis:
$$\mathrm{\Omega }_{k_1,\mathrm{},k_m}=\stackrel{~}{\mu }_{k_1}\mathrm{}\stackrel{~}{\mu }_{k_m}$$
where $`gk_jg`$. Following we adopt the Conjecture 2. We have
$`H^m={\displaystyle \frac{W^m}{\sigma W^{m2}}}`$ (11)
where
$$\sigma =\underset{j=1}{\overset{g}{}}\stackrel{~}{\mu }_j\stackrel{~}{\mu }_j$$
According to (9) the singularities of differential forms occur either at $`p_i=\sigma (p_j)`$ or at $`p_i=\mathrm{}^\pm `$. The non-trivial essence of Conjecture 2 is that the first kind of singularities can be eliminated by adding exact forms. There are $`(g1)`$-forms singular at $`p_i=\sigma (p_j)`$ such that these singularities disappear after applying $`d`$. This is the origin of the space $`\sigma W^{k2}`$ .
Consider briefly the dual picture. On the affine curve with punctures at $`\mathrm{}^\pm `$ there are $`2g+1`$ non-trivial cycles $`\delta _k`$ with $`k=g,\mathrm{},g`$. The cycles $`\delta _k`$ , $`k<0`$ are a-cycles, the cycles $`\delta _k`$ , $`k>0`$ are b-cycles and $`\delta _0`$ is the cycle around $`\mathrm{}^+`$. One defines the cycles $`\stackrel{~}{\delta }_k`$ on the symmetrical power of the affine curve. The $``$-operation is introduced for these cycles by duality with cohomologies. The non-trivial consequence of Conjecture 2 is that every cycle on $`J_{\text{aff}}(t)`$ can be constructed by wedging $`\stackrel{~}{\delta }_k`$. The formula dual to (11) is
$`H_m={\displaystyle \frac{W_m}{\sigma ^{}W_{m2}}}`$ (12)
where $`W_m`$ is spanned by
$$\mathrm{\Delta }_{k_1,\mathrm{},k_m}=\stackrel{~}{\delta }_{k_1}\mathrm{}\stackrel{~}{\delta }_{k_m}$$
and
$$\sigma ^{}=\underset{j=1}{\overset{g}{}}\stackrel{~}{\delta }_j\stackrel{~}{\delta }_j$$
We need to factorize over $`\sigma ^{}W_{m2}`$ because the 2-cycle $`\sigma ^{}`$ intersects with $`D`$.
Let us return to the relation of $`H^g`$ to the algebra $`๐`$. Notice that
$$d\tau _1\mathrm{}d\tau _g\stackrel{~}{\mu }_1\mathrm{}\stackrel{~}{\mu }_g\mathrm{\Omega }$$
The functions
$$x_{k_1,\mathrm{},k_g}=\mathrm{\Omega }^1\mathrm{\Omega }_{k_1,\mathrm{},k_g}$$
are symmetric polynomials of $`z_1,\mathrm{},z_g`$. Recall that $`b_1,\mathrm{},b_g`$ are nothing but elementary symmetric polynomials of $`z_1,\mathrm{},z_g`$. Hence the coefficients of cohomologies have the form:
$$g_\alpha =g_\alpha (b_1,\mathrm{},b_g)$$
The dimension of $`H^g`$ is determined by Conjecture 2:
$$\alpha =1,\mathrm{},\left(\genfrac{}{}{0pt}{}{2g+1}{g}\right)\left(\genfrac{}{}{0pt}{}{2g+1}{g2}\right),$$
The equations (7) are consequences of the following ones
$`{\displaystyle \underset{k=1}{\overset{g}{}}}_k\left(\mathrm{\Omega }^1(\mu _k\mathrm{\Omega }_{k_1,\mathrm{},k_{g1}})\right)=0,\mathrm{\Omega }_{k_1,\mathrm{},k_{g1}}W^{g1}`$ (13)
## 4 Quantization of affine Jacobian.
Let us consider a quanization of algebra $`๐`$. The parameter of deformation (Planck constant) is denoted by $`\gamma `$, we shall also use
$$q=e^{i\gamma }$$
Consider the 2$`\times `$ 2 matrix $`๐(z)`$ with noncommuting entries. Suppose that the dependence on the spectral parameter $`z`$ is exactly the same as in classical case (1). The variables $`๐_j`$, $`๐_j`$, $`๐_j`$, $`๐
_j`$ are subject to commutation relations which are summarized as follows:
$`r_{21}(z_1,z_2)๐_1(z_1)k_{12}(z_1)s_{12}๐_2(z_2)k_{21}(z_2)=`$
$`=๐_2(z_2)k_{21}(z_2)s_{21}๐_1(z_1)k_{12}(z_1)r_{12}(z_1,z_2)`$ (14)
where usual conventions are used: the equation (14) is written in the tensor product $`^2^2`$, $`a_1=aI`$, $`a_2=I2`$, $`a_{21}=Pa_{12}P`$ where $`P`$ is the operation of permuattions. The $``$-number matrices $`r`$, $`k`$, $`s`$ are:
$`r_{12}(z_1,z_2)={\displaystyle \frac{z_1qz_2}{1q}}(II)+{\displaystyle \frac{z_1+qz_2}{1+q}}(\sigma ^3\sigma ^3)+`$
$`+2(z_1\sigma ^{}\sigma ^++z_2\sigma ^+\sigma ^{}),`$
$`k_{12}(z)=I(I\sigma ^3)+\left(q^{\sigma ^3}+z(q^21)\sigma ^{}\right)(I+\sigma ^3),`$
$`s_{12}=II(qq^1)\sigma ^{}\sigma ^+`$ (15)
These commutation relations are important because they respect the form of matrix $`๐(z)`$ prescribed by (1), we shall explain how they are related to more usual r-matrix relations in the next section.
Define the polynomials:
$`๐(z)=q๐(z)+q^2๐
(z)z(q^21)๐(z)`$
$`๐(z)=q๐
(z)๐(zq^2)q^2๐
(z)๐
(zq^2)q๐(z)๐(zq^2)`$ (16)
The algebra $`\widehat{๐}(q)`$ is generated by $`๐_1,\mathrm{},๐_{g+1}`$, $`๐_1,\mathrm{},๐_g`$, $`๐_2,\mathrm{},๐_{g+2}`$, $`๐
_2,\mathrm{},๐
_{g+1}`$, The polynomial $`๐(z)`$ belongs to the center of $`\widehat{๐}(q)`$. The coefficients of $`๐(z)`$ are commuting, actually, $`๐_{g+1}`$ belongs to the center of $`\widehat{๐}(q)`$. We define:
$`๐(q)={\displaystyle \frac{\widehat{๐}(q)}{\{๐(z)=1,๐_{g+1}=(1)^{g+1}2\}}}`$
The non-commutative algebra $`๐(q)`$ defines a quantization of the algebra of function on the quadric $``$. However, we cannot define directly the quantization of the algebra of functions on the affine Jacobian because the coefficients of $`๐(z)`$ are not in the center of $`๐(q)`$. What we can do is to describe the quantum version of Proposition 1 and of description of cohomologies. The exposition will be more detailed than in the classical case.
Like in the paper we accept the following Conjecture 3. The algebra $`๐(q)`$ is spanned as linear space by elements of the form:
$`x=p_L(๐_1,\mathrm{},๐_g)g(๐_1,\mathrm{},๐_g)p_R(๐_1,\mathrm{},๐_g)`$ (17)
where $`p_L(๐ญ_1,\mathrm{},๐ญ_g),`$ $`g(๐_1,\mathrm{},๐_g),`$ $`p_R(๐ญ_1,\mathrm{},๐ญ_g)`$ are polynomials. We were not able to prove this statement, however, since the algebra $`๐(q)`$ is graded we can check it degree by degree. This has been done up to degree 8. Notice the similarity between the representation (17) and the representation for spin operators proved in . Conjecture 3 implies that certain generalization of the results of is possible. In fact the formula (17) is similar to the formula (6): we can either symmetrize or anti-symmetrize $`๐_j`$ in (17) which corresponds in classics to multipilation by $`t_j`$ or to applying $`_j`$. In order to have complete agreement with clasical case we have to show that only finitely many different polynomials $`g(๐_1,\mathrm{}๐_g)`$ (cohomologies) create entier algebra $`๐(q)`$.
Notice that the commutation relations (14) imply in particular that
$$[๐(z),๐(z^{})]=0$$
which means that we have the commutative family of operators $`๐_j`$ defined by
$$๐(z)=(z๐_j)$$
So, every polynomials $`g(๐_1,\mathrm{},๐_g)`$ can be considered as symmetric polynomial of $`๐_j`$ and vice versa.
It is very convenient to use the following formal definitions. Consider the ring $`๐ฏ`$ defined in (5). By $`๐ฑ^k`$ we denote the space of anti-symmetric polynomials of $`k`$ variables such that their degrees with respect to every variable is not less than $`1`$ with coefficients in $`๐ฏ๐ฏ`$. In other words $`๐ฑ^k`$ is the space spanned by the polynomials:
$$p_Lhp_Rp_L(t_1,\mathrm{},t_g)h(z_1,\mathrm{},z_k)p_R(t_1^{},\mathrm{},t_g^{})$$
where $`h`$ is anti-symmetric, vanishing when one of $`z_j`$ vanishes. The following operations can be defined.
1. Multiplication by $`t_j`$ and $`t_j^{}`$.
2. Operation $`:๐ฑ^k๐ฑ^l๐ฑ^{k+l}`$ which is defined as follows:
$$(p_Lhp_R)(p_L^{}h^{}p_R^{})=p_Lp_L^{}(hh^{})p_Rp_R^{}$$
where
$`(hh^{})`$ $`(z_1,\mathrm{},z_{k+l})=`$
$`={\displaystyle \frac{1}{k!l!}}{\displaystyle \underset{\pi S_{k+l}}{}}(1)^\pi h(z_{\pi (1)},\mathrm{},z_{\pi (k)})h^{}(z_{\pi (k+1)},\mathrm{},z_{\pi (k+l)})`$
We have a map
$$๐ฑ^g\stackrel{\chi }{}๐(q)$$
defined on the basis elements as
$$\chi (p_Lhp_R)=p_L(๐_1,\mathrm{},๐_g)\frac{h(๐_1,\mathrm{},๐_g)}{๐_i_{i<j}(๐_i๐_j)}p_R(๐_1,\mathrm{},๐_g)$$
and continued linearly. The Conjecture 3 states that this map is surjective. We want to describe the kernel of the map $`\chi `$.
First, consider the space $`๐ฑ^1`$. The elements of this space are polynomials of one variable $`z`$ with coefficients in $`๐ฏ๐ฏ`$. In Appendix B we describe certain basis in $`๐ฑ^1`$ considered as a linear space over $`๐ฏ๐ฏ`$. The basis in question consists of the polynomials: $`s_k`$ with $`kg`$ such that the degree of $`s_k`$ with respect to $`z`$ equals $`g+k+1`$. The kernel of $`\chi `$ is the joint of three sub-spaces, let us describe them. 1. For $`kg+1`$ we have:
$`\chi \left(s_k๐ฑ^{g1}\right)=0`$ (18)
2. Consider $`c๐ฑ^2`$ defined as
$$c=\underset{j=1}{\overset{g}{}}s_js_j$$
we have
$`\chi \left(c๐ฑ^{g2}\right)=0`$ (19)
3. Consider $`d๐ฑ^1`$ defined as
$$d=(t_jt_j^{})s_j$$
we have
$`\chi \left(d๐ฑ^{k1}\right)=0`$ (20)
The construction of the space
$$\frac{๐ฑ^g}{\text{Ker}(\chi )}๐(q)$$
is in complete correspondence with the classiccal case. In classics we start with all the 1-forms $`\stackrel{~}{\mu }_k`$. Imposing (18) corresponds to throwing away the exact forms and working with $`\stackrel{~}{\mu }_k`$ for $`k=g\mathrm{},g`$ only. Imposing (19) corresponds to factorizing over $`\sigma W^{k2}`$ in classics. Finally, (20) corresponds to the equation (13).
The origin of the equations (18), (19), (20) will be explained in the Section 9. There should be purely algebraic method of prooving these equations, but we do not know it. It is important to mention that accepting Conjecture 3 we are forced to conclude that the kernel of $`\chi `$ is completely described by the equations (18), (19), (20). This is proved by calculation of characters similarly to that of .
## 5 The realization of $`๐(๐ช)`$.
We want to describe a realization of the algebra $`๐(q)`$ in a space of functions. Consider the quantum mechanical system described by the operators $`x_j`$ with $`j=1,\mathrm{},2g+2`$ and $`y`$ (zero mode). The operators $`x_j`$ and $`y`$ are self-adjoint, they satisfy the commutation relations:
$`x_kx_l=q^2x_lx_kk<l,`$
$`yx_k=q^2x_kyk`$
The hamiltonian of the system is
$$๐=q^1\underset{k=1}{\overset{2g+2}{}}x_kx_{k1}^1$$
where
$$x_{2g+3}qyx_1$$
Physically this model defines the simplest lattice regularization of the chiral Bose field with modified energy-momentum tensor.
It is useful to double the number of degrees of freedom. Consider the algebra $`A`$ generated by two operators $`u`$ and $`v`$ satisfying the commutation relations:
$$uv=qvu$$
Take the algebra $`A^{(2g+2)}`$ the operators $`u_j`$, $`v_j`$ ($`j=1,\mathrm{},2g+2`$) are defined as $`u`$ and $`v`$ acting in $`j`$-th tensor component. The original operators $`x_i`$ are expressed in terms of $`u_i,v_i`$ as follows:
$$x_k=v_k\underset{j=1}{\overset{k1}{}}u_j^2,y=\underset{j=1}{\overset{2g+2}{}}u_j$$
Consider the โmonodromy matrixโ
$`\stackrel{\mathbf{~}}{๐}(z)=\left(\begin{array}{cc}\stackrel{\mathbf{~}}{๐}(z)& \stackrel{\mathbf{~}}{๐}(z)\\ \stackrel{\mathbf{~}}{๐}(z)& \stackrel{\mathbf{~}}{๐
}(z)\end{array}\right)=l_{2g+2}(z)\mathrm{}l_1(z)`$ (21)
where the l-operators are
$`l(z)={\displaystyle \frac{1}{\sqrt{z}}}\left(\begin{array}{cc}zu& qvu\\ zv^1u^1& 0\end{array}\right)`$ (22)
This is a particular case of more genaral l-operator $`l(z,\kappa )`$ in which the last matrix element is not $`0`$ but $`\kappa zu`$, the model corresponding to the latter l-operator is a subject of study in a series of papers .
The matrix elements of the matrix $`๐(z)`$ satisfy the commutation relations:
$`r_{12}(z_1,z_2)\stackrel{\mathbf{~}}{๐}_1(z_1)\stackrel{\mathbf{~}}{๐}_2(z_2)=\stackrel{\mathbf{~}}{๐}_2(z_2)\stackrel{\mathbf{~}}{๐}_1(z_1)r_{12}(z_1,z_2)`$ (23)
where the r-matrix $`r_{12}(z_1,z_2)`$ is defined ealier (15). These are canonical r-matrix commutation relations. The quantum determinant of the matrix $`\stackrel{\mathbf{~}}{๐}(z)`$ is defined by
$$๐(z)=\stackrel{\mathbf{~}}{๐
}(z)\stackrel{\mathbf{~}}{๐}(zq^2)\stackrel{\mathbf{~}}{๐}(z)\stackrel{\mathbf{~}}{๐}(zq^2),$$
it belongs to the center, in our realization of $`\stackrel{\mathbf{~}}{๐}(z)`$ one has $`๐(z)=1`$. The trace of $`\stackrel{\mathbf{~}}{๐}(z)`$ generates commuting quantities, we denote this trace as follows:
$$\stackrel{\mathbf{~}}{๐}(z)+\stackrel{\mathbf{~}}{๐
}(z)=y๐(z)$$
The matrix elements of the matrix $`\stackrel{~}{m}(z)`$ are of the form:
$`\stackrel{\mathbf{~}}{๐}(z)=\stackrel{\mathbf{~}}{๐}_0z^{g+1}+\stackrel{\mathbf{~}}{๐}_1z^g+\mathrm{}+\stackrel{\mathbf{~}}{๐}_{g+1},`$ (24)
$`\stackrel{\mathbf{~}}{๐}(z)=\stackrel{\mathbf{~}}{๐}_0z^g+\stackrel{\mathbf{~}}{๐}_1z^{g1}+\mathrm{}+\stackrel{\mathbf{~}}{๐}_g,`$
$`\stackrel{\mathbf{~}}{๐}(z)=\stackrel{\mathbf{~}}{๐}_1z^{g+1}+\stackrel{\mathbf{~}}{๐}_2z^g+\mathrm{}+\stackrel{\mathbf{~}}{๐}_{g+1}z,`$
$`\stackrel{\mathbf{~}}{๐
}(z)=\stackrel{\mathbf{~}}{๐
}_1z^g+\stackrel{\mathbf{~}}{๐
}_2z^{g1}+\mathrm{}+\stackrel{\mathbf{~}}{๐
}_gz`$
where, in particular, $`\stackrel{\mathbf{~}}{๐}_0=y`$. This form of polynomials $`\stackrel{\mathbf{~}}{๐}(z)`$, $`\stackrel{\mathbf{~}}{๐}(z)`$, $`\stackrel{\mathbf{~}}{๐}(z)`$, $`\stackrel{\mathbf{~}}{๐
}(z)`$ does not corresponds to what we have in the classical model of affine Jacobian. This is the reason for modifying the matrix $`\stackrel{~}{m}(z)`$ as follows:
$$๐(z)=\left(\begin{array}{cc}\stackrel{\mathbf{~}}{๐}_0\stackrel{\mathbf{~}}{๐}_0^1& 0\\ \stackrel{\mathbf{~}}{๐
}_1\stackrel{\mathbf{~}}{๐}_0^1& 1\end{array}\right)\stackrel{\mathbf{~}}{๐}(z)\left(\begin{array}{cc}\stackrel{\mathbf{~}}{๐}_0\stackrel{\mathbf{~}}{๐}_0^1& 0\\ q\stackrel{\mathbf{~}}{๐
}_1\stackrel{\mathbf{~}}{๐}_0^1& 1\end{array}\right)$$
The matrix elements of this matrix have structure (1), they satisfy closed commutation relations (14), the operators $`๐(z)`$ and $`๐(z)`$ defined for these two matrices coincide, in particular we have
$$๐_1=๐$$
Thus the modification of matrix $`\stackrel{\mathbf{~}}{๐}(z)`$ which is necessary for relation to the affine Jacobian is responcible for appearing of strangely looking commutation relations (14).
## 6 Q-operator.
Our first goal is to define Baxterโs Q-operator. Let us realize the operators $`v,u`$ in $`L_2()`$ as follows
$$v=e^\phi ,u=e^{i\gamma \frac{d}{d\phi }}$$
We shall work in the $`\phi `$-representation, i.e. in the space $`=(L_2())^{(2g+2)}`$. Following the standart procedure (cf. ) one introduces the vectors $`Q(\zeta |\psi _1,\mathrm{},\psi _{2g+2})`$ which depend on
$$\zeta =\frac{1}{2}\mathrm{log}z$$
and $`2g+2`$ additional parameters, $`\psi _j`$, and satisfy the equation:
$`(1)^{g+1}๐(z)Q(\zeta `$ $`|\psi _1,\mathrm{},\psi _{2g+2})=`$
$`=Q(\zeta +i\gamma |\psi _1,\mathrm{},\psi _{2g+2})+Q(\zeta i\gamma |\psi _1,\mathrm{},\psi _{2g+2})`$
In $`\phi `$-representation the โcomponentsโ of these vectors are given by
$`Q(\phi _1,`$ $`\mathrm{},\phi _{2g+2}|\zeta |\psi _1,\mathrm{},\psi _{2g+2})=`$
$`=e^{\frac{1}{2}(1+\frac{\pi }{\gamma })\zeta +\frac{1}{4i\gamma }\zeta ^2}{\displaystyle \underset{k=1}{\overset{2g+2}{}}}\lambda (\zeta |\phi _k\psi _k)\phi _k|\psi _{k1}`$ (25)
where $`\psi _0\psi _{2g+2}`$,
$`\phi |\psi =e^{\frac{1}{4i\gamma }(2\phi \psi \phi ^2)},`$ (26)
$`\lambda (\zeta |\psi )=e^{\frac{1}{2i\gamma }\zeta \psi }\mathrm{\Phi }(\psi \zeta )e^{\frac{\pi +\gamma }{\gamma }(\psi \zeta )},`$
and the function $`\mathrm{\Phi }(\phi )`$ satisfies the functional equation:
$`{\displaystyle \frac{\mathrm{\Phi }(\phi +i\gamma )}{\mathrm{\Phi }(\phi i\gamma )}}={\displaystyle \frac{1}{1+e^\phi }}`$ (27)
The solution to this equation is
$$\mathrm{\Phi }(\phi )=\mathrm{exp}\left(\underset{+i0}{}\frac{e^{ik\phi }}{4\mathrm{sinh}\gamma k\mathrm{sinh}\pi k}\frac{dk}{k}\right)$$
This wonderful function and its applications can be found in .
As usual we want to consider $`Q(\phi _1,\mathrm{},\phi _{2g+2}|\zeta |\psi _1,\mathrm{},\psi _{2g+2})`$ as the kernel of an operator:
$$Q(\phi _1,\mathrm{},\phi _{2g+2}|\zeta |\psi _1,\mathrm{},\psi _{2g+2})=\phi _1,\mathrm{},\phi _{2g+2}|๐ (\zeta )|\psi _1,\mathrm{},\psi _{2g+2}$$
The subtle point is that we have to use mixed representations: the vectors $`|\psi `$ are the eigenvectors of the operators
$$we^\psi =uvu$$
Notice that this justifies the notation $`\phi |\psi `$ in (26), and that
$$[\psi ,\phi ]=2i\gamma $$
The operators $`๐ (\zeta )`$ satisfy the equations
$`(1)^{g+1}๐(z)๐ (\zeta )=๐ (\zeta +i\gamma )+๐ (\zeta i\gamma )`$ (28)
This is famous Baxterโs equation.
Before going further let us discuss the properties of the operator $`๐ (\zeta )`$. We have
$`\overline{\mathrm{\Phi }(\phi )}=\mathrm{\Phi }(\overline{\phi });`$
$`\mathrm{\Phi }(\phi )\mathrm{exp}\left({\displaystyle \frac{1}{4i\gamma }}\phi ^2\right),\text{as}\phi \mathrm{}`$ (29)
so, the kernel of $`๐ (\zeta )`$ for $`\zeta `$ is an oscilating function, and it is rather clear that our operator is well defined on the functions of $`\psi _j`$ of Schwartz class ($`\text{Sch}_\psi `$) sending them to functions of $`\phi _j`$ which are also of Schwartz class ($`\text{Sch}_\phi `$). Using the equations (29) one easily finds the kernel $`\psi |๐ ^{}(\zeta )|\phi `$ of the adjoint operator $`๐ ^{}(\zeta )`$ (we consider the case of real $`\zeta `$). Further, notice that the l-operator can be rewritten as
$`l(z)={\displaystyle \frac{1}{\sqrt{z}}}\left(\begin{array}{cc}zu& qu^1w\\ zuw^1& 0\end{array}\right)`$ (30)
Applying to this l-operator the same procedure as before one finds that $`๐ ^{}(\zeta )`$ also solves the Baxter equation (28):
$$(1)^{g+1}๐(z)๐ ^{}(\zeta )=๐ ^{}(\zeta +i\gamma )+๐ ^{}(\zeta i\gamma )$$
It can be shown that actually
$$๐ (\zeta )=๐ ^{}(\zeta )\text{for}\zeta $$
Considering the kernel of the operator $`๐ ^{}(\zeta )`$ one finds that this operator acts from $`\text{Sch}_\phi `$ to $`\text{Sch}_\psi `$. So, the products $`๐ (\zeta )๐ (\zeta ^{})`$ are well defined at least for $`\zeta ,\zeta ^{}`$.
We want to show that the operators $`๐ (\zeta )`$ constitute a commutative family:
$`[๐ (\zeta ),๐ (\zeta ^{})]=0`$ (31)
To this end we want to show that the operator $`๐ (\zeta )`$ can be rewritten as
$`๐ (\zeta )=\text{tr}_a\left(_{a2q+2}(\zeta )\mathrm{}_{a1}(\zeta )\right)`$ (32)
where the operators $`_{aj}(\zeta )`$ act in the tensor product of the โauxilary spaceโ labeled by $`a`$ and of the โquantum spaceโ where $`\phi _j`$, $`\psi _j`$ act. Actually in our case the โauxilary spaceโ will be isomorphic to the โquantum spaceโ, i.e. we shall have a universal l-operator. If the operators $`_{aj}(\zeta )`$ satisfy Yang-Baxter equations with some R-matrix then the commutativity (31) follows from the standart argument.
To find the representation (32) rewrite (25) as
$`๐ (\zeta )=e^{\frac{1}{2}(1+\frac{\pi }{\gamma })\zeta +\frac{1}{4i\gamma }\zeta ^2}{\displaystyle \underset{j=1}{\overset{2g+2}{}}d\phi _j^{}d\psi _j^{}\psi _j^{}|_{aj}(\zeta )|\phi _j^{}\phi _j^{}|\psi _{j1}^{}}`$
where $`\phi _j^{}`$, $`\psi _j^{}`$ are operators acting in the โauxiliary spaceโ, $`\psi _0^{}=\psi _{2g+2}^{}`$. So, (32) indeed takes place if the kernel of the โuniversalโ l-operator is given by
$$\phi ^{}|\psi |(\zeta )|\psi ^{}|\phi =\delta (\phi \phi ^{})\delta (\psi \psi ^{})\lambda (\zeta |\phi \psi )$$
Hence the formula (32) holds for the operators $`_{aj}(\zeta )`$ of the form:
$$_{12}(\zeta )=๐ซ_{12}\widehat{}_{12}(\zeta )$$
where $`๐ซ_{12}`$ is the operator of permutation, and the operator $`\widehat{}_{12}(\zeta )`$ acts in the tensor product as follows:
$$\widehat{}_{12}(\zeta )=\lambda (\zeta |\phi II\psi )$$
Thus the operator $`๐ (\zeta )`$ can be considered as trace of โuniversalโ monodromy matrix and the commutativity (31) follows from the Yang-Baxter equation:
$`\widehat{}_{12}(\zeta _1\zeta _2)\widehat{}_{23}(\zeta _1)\widehat{}_{12}(\zeta _2)=\widehat{}_{23}(\zeta _2)\widehat{}_{12}(\zeta _1)\widehat{}_{23}(\zeta _1\zeta _2)`$ (33)
with the simple r-matrix:
$$\widehat{}_{12}(\zeta )=\mathrm{exp}\left(\frac{(I\psi \phi I)\zeta }{2i\gamma }\right)$$
The Yang-Baxter equation (33) in our case is almost trivial. In the case of the more general l-operator $`\widehat{l}(z,\kappa )`$ mentioned above we would need to use a more complicated r-matrix and the proof of Yang-Baxter equations needs some non-trivial identities .
The self-ajoint (for real $`\zeta _1,\zeta _2`$) operators $`๐ (\zeta _1)`$, $`๐ (\zeta _2)`$ commute, hence the eigen-vectors of $`๐ (\zeta )`$ do not depend on $`\zeta `$. Actually, the operator $`๐ (\zeta )`$ is an entire function of $`\zeta `$. The kernel of $`๐ (\zeta )`$ has poles, but in the process of analytical continuation the poles never pinch the contour of integration. The Baxterโs equation (28) implies that $`๐ (\zeta _1)`$ and $`๐(z_2)`$ also commute. Suppose that $`๐ฌ(\zeta )`$ ant $`t(z)`$ are eigen-values of these operators, due to the equation (28) they satisfy
$`(1)^{g+1}t(z)๐ฌ(\zeta )=๐ฌ(\zeta +i\gamma )+๐ฌ(\zeta i\gamma )`$ (34)
Let us discuss further analytical properties of $`๐ฌ(\zeta )`$. Since the operator $`๐ (\zeta )`$ is an entire functions of $`\zeta `$ the eigen-value $`๐ฌ(\zeta )`$ is an entire function as well. As it has been said $`๐(0)=๐_{g+1}`$ belongs to the center of the algebra defined by the commutation relations (23), so, we can fix it. It is convenient to put $`๐_{g+1}=(1)^{g+1}2`$ which allows to reqiure that
$`๐ฌ(\zeta )1,\zeta \mathrm{}`$ (35)
From quasi-classical consideration which are completely parallel to those from it is naturally to conjecture that the eigenvalues of $`๐ฌ(\zeta )`$ have zeros only on the real axis and that asymptotically for $`\zeta \mathrm{}`$ one has:
$`๐ฌ(\zeta )e^{(g+1)(1+\frac{\pi }{\gamma })\zeta }\mathrm{cos}\left({\displaystyle \frac{(g+1)\zeta ^2}{\gamma }}+{\displaystyle \frac{\pi }{4}}\right)`$ (36)
The important question is whether the equations (34) together with the analytical properties described above are sufficient to find the spectrum of commuting Hamiltonians. In our opinion it is impossible, the additional information is needed which is provided in the following section.
## 7 Duality.
Consider the function $`\mathrm{\Phi }(\zeta )`$. The most interesting property of this function is its duality: together with the equation (27) it satisfies the equation
$`{\displaystyle \frac{\mathrm{\Phi }(\phi +i\pi )}{\mathrm{\Phi }(\phi i\pi )}}={\displaystyle \frac{1}{1+e^{\frac{\pi }{\gamma }\phi }}}`$
Using this property and the definition of the operator $`๐ (\zeta )`$ one finds that there is dual equation for $`๐ (\zeta )`$:
$`(1)^{g+1}๐ป(Z)๐ (\zeta )=๐ (\zeta +\pi i)+๐ (\zeta \pi i)`$ (37)
where
$$Z=e^{\frac{2\pi }{\gamma }\zeta },$$
and $`๐ป(Z)`$ is the trace of the monodromy matrix
$$\stackrel{\mathbf{~}}{๐ด}(Z)=L_{2g+2}(Z)\mathrm{}L_1(Z)$$
with
$`L(Z)={\displaystyle \frac{1}{\sqrt{Z}}}\left(\begin{array}{cc}ZU^1& QVU\\ ZV^1U^1& 0\end{array}\right)`$
The dual operators
$$U=e^{\frac{\pi }{\gamma }\phi },V=e^{\pi i\frac{d}{d\phi }}$$
satisfy the commutation relations
$$UV=QVU$$
with dual
$$Q=e^{i\frac{\pi ^2}{\gamma }}$$
The only non-trivial commutation relations of $`u,v`$ with $`U,V`$ are
$$uV=Vu,vU=Uv$$
which means that
$$S(l(z)I)(IL(Z))=(IL(Z))(l(z)I)S$$
with $`S=\sigma ^3\sigma ^3`$. From here it is obvious that
$$[๐(z),๐ป(Z)]=0$$
All that is the result of manifest duality of the kernel of $`๐ (\zeta )`$ with respect to change:
$`\gamma {\displaystyle \frac{\pi ^2}{\gamma }},\zeta {\displaystyle \frac{\pi }{\gamma }}\zeta ,\phi _j{\displaystyle \frac{\pi }{\gamma }}\phi _j,\psi _j{\displaystyle \frac{\pi }{\gamma }}\psi _j`$
It is clear that $`๐ป(Z_1)`$ and $`๐ (\zeta _2)`$ commute, so, the equation (37) implies the equation for eigen-values:
$`(1)^{g+1}T(Z)๐ฌ(\zeta )=๐ฌ(\zeta +\pi i)+๐ฌ(\zeta \pi i)`$ (38)
The function $`๐ฌ(\zeta )`$ is not an entire function of $`z`$ as it is the case in other situations (for example ), that is why the equation (28) alone does not look strong enough to define it. However, the equation (37) controlling the behaviour of $`๐ฌ(\zeta )`$ under the $`2\pi i`$ -rotation in $`z`$-plane must provide the missing information.
So, our main conjecture is the following Conjecture 4. The spectrum on $`๐ญ(z)`$ (and, simultaneously, of $`๐(Z)`$) is described by all solutions of the equations (34) and (38) such that
1. $`t(z)`$ and $`T(Z)`$ are polynomials of degree $`g+1`$.
2. $`๐ฌ(\zeta )`$ is an entire function of $`\zeta `$.
3. $`๐ฌ(\zeta )`$ satisfies (35) and (36).
4. All the zeros of $`๐ฌ(\zeta )`$ in the strip $`(\pi +\gamma )<\text{Im}(\zeta )<(\pi +\gamma )`$ are real.
## 8 Separation of variables.
The relation of integrable models to the algebraic geometry can be completely understood in the framework of separation of variables.
We have already mentioned that
$$[๐(z),๐(z^{})]=0$$
which implies commutativity of the operators $`๐_j`$ defined as roots of $`๐(z)`$. Consider the operators
$$๐_j=(1)^{g+1}q๐
(\stackrel{}{๐}_j)$$
where $`d(\stackrel{}{๐_j})`$ means that $`๐_j`$ which does not commute with coefficients of $`๐
(z)`$ is substituted to this polynomial form the left. Following Sklyanin one shows that
$$๐_j๐_k=๐_k๐_j,jk;๐_j๐_j=q^2๐_j๐_j$$
and
$`๐_j^2๐_j๐(\stackrel{}{๐}_j)+1=0`$ (39)
Introduce the operators
$$๐ป_j=\frac{1}{2}\mathrm{log}(๐_j)$$
and consider the wave-function corresponding to given set of eigen-values of integral of motion $`๐_1,\mathrm{},๐_g`$ in $`๐ป`$-representation. The equation (39) implies that we can look for this wave function in the form:
$$\zeta _1,\mathrm{},\zeta _g|t_1,\mathrm{},t_g=๐ฌ(\zeta _1)\mathrm{}๐ฌ(\zeta _g)$$
where $`๐ฌ(\zeta )`$ satisfies the equation:
$$๐ฌ(\zeta +i\gamma )+๐ฌ(\zeta i\gamma )=(1)^{g+1}t(z)๐ฌ(\zeta )$$
where $`t(z)`$ is constructed from the eigen-values $`t`$. This equation coincides with the equation (28) written for particular eigen-values. So, following we claim that the wave function in separated variables is defined by eigen-value of the operator $`๐ `$ which connects two different approaches to integrable models.
Notice that the vector $`|t_1,\mathrm{},t_g`$ is eigen-vector for the operators $`๐ป_1,\mathrm{},๐ป_g`$ since the function $`๐ฌ(\zeta )`$ satisfies the equation (37). In order to identify explicitly the eigen-vales of $`๐ป_1,\mathrm{},๐ป_g`$ we shall write $`|t_1,\mathrm{},t_g;T_1,\mathrm{},T_g`$.
We have the algebra of operators $`๐(q)`$ and the dual algebra $`๐(Q)`$ which act in the same space $``$. All the operators from $`๐(q)`$ commute with the operators from $`๐(Q)`$. The fundamental property of $`๐(q)`$ is that it is spanned as linear space by the elements of the form (17) according to Conjecture 3. Similar fact must be true for $`๐(Q)`$. Taking these facts together one realizes the algebra $`๐(q)๐(Q)`$ is spanned by the elements of the form:
$`๐ณ=`$ $`xX=p_L(๐_1,\mathrm{},๐_g)P_L(๐ป_1,\mathrm{},๐ป_g)`$ (40)
$`\times g(๐_1,\mathrm{},๐_g)G(๐ฉ_1,\mathrm{},๐ฉ_g)P_R(๐ป_1,\mathrm{},๐ป_g)p_R(๐_1,\mathrm{},๐_g)`$
We denote by $`h(๐_1,\mathrm{},๐_g)`$ and $`H(๐_1,\mathrm{},๐_g)`$ anti-symmetric polynomials obtained from $`g(๐_1,\mathrm{},๐_g)`$ and $`G(๐ฉ_1,\mathrm{},๐ฉ_g)`$, for example:
$`h(๐_1,\mathrm{},๐_g)={\displaystyle ๐_i\underset{i<j}{}(๐_i๐_j)g(๐_1(๐_1,\mathrm{},๐_g),\mathrm{},๐_g(๐_1,\mathrm{},๐_g))}`$
Let us consider the matrix element of the operator $`๐ณ`$ between two eigen-vectors of Hamiltonians. The wave functions are real for real $`\zeta `$. By requirement of self-ajointness of $`๐(z)`$ and $`๐ป(Z)`$ one defines the scalar product (cf. ). The matrix element in question is
$`t_1,\mathrm{}t_g;T_1,\mathrm{},T_g|๐ณ|t_1^{},\mathrm{}t_g^{};T_1^{},\mathrm{},T_g^{}=`$
$`=p_L(t_1,\mathrm{},t_g)p_R(t_1^{},\mathrm{},t_g^{})P_L(T_1,\mathrm{},T_g)P_R(T_1^{},\mathrm{},T_g^{})`$ (41)
$`\times {\displaystyle \underset{\mathrm{}}{\overset{\mathrm{}}{}}}d\zeta _1\mathrm{}{\displaystyle \underset{\mathrm{}}{\overset{\mathrm{}}{}}}d\zeta _gh(z_1,\mathrm{},z_g)H(Z_1,\mathrm{},Z_g){\displaystyle \underset{j=1}{\overset{g}{}}}๐ฌ(\zeta _j)๐ฌ^{}(\zeta _j)`$
When does the integral for the matrix element (41) converge? Suppose that
$$hz_j^{g+k+1},HZ_j^{g+l+1}\text{when}\zeta _j\mathrm{}$$
then the integrand in the matrix element behaves when $`\zeta _j\mathrm{}`$ as
$$\mathrm{exp}2\zeta _j\left((k1)+\frac{\pi }{\gamma }(l1)\right)$$
hence for generic $`\gamma `$ the integral converges only if $`k=1,l=0`$ or $`k=0,l=1`$. When $`\gamma `$ is small we can allow the operators with $`l=0`$ and $`k<\frac{\pi }{\gamma }`$, oppositely, when $`\gamma `$ is big the operators with $`k=0`$ and $`l<\frac{\gamma }{\pi }`$ are allowed. The limits $`\gamma 0`$ and $`\gamma \mathrm{}`$ are two dual quasi-classical limits. For these limits the operators $`l=0,k`$ and $`k=0,l`$ respectively define the classical observables. At least these operators must be defined in the quantum case: if the quantization procedure makes sense the principle of correspondence must hold. Hence, the fact that in general only two operators with $`k=1,l=0`$ or $`k=0,l=1`$ lead to convergent integrals means that some regularization of these integrals is needed. The regularized integrals in question must allow to define the matrix element (41) for arbitrary $`k,l`$, they have to coincide with usual integrals whenever the latter are applicable, they must satisfy some additional requirements which will be discussed in the Section 9. The origin of these additional requirements is in the cohomological construction explained in Section 4.
Notice that any anti-symmetric with respect to $`z_1,\mathrm{},z_g`$ and $`Z_1,\mathrm{},Z_g`$ polynomial
$$h(z_1,\mathrm{},z_g)H(Z_1,\mathrm{},Z_g)$$
can be presented as linear combination of products of Schur-type determinants
$$\text{det}|z_i^{k_j}|\text{det}|Z_i^{l_j}|$$
where $`\{k_1,\mathrm{},k_g\}`$ and $`\{l_1,\mathrm{},l_g\}`$ are arbitrary sets of positive integers. So, the integrals (41) can be expressed in terms of 1-fold integrals
$`l|L{\displaystyle \underset{\mathrm{}}{\overset{\mathrm{}}{}}}๐ฌ(\zeta )๐ฌ^{}(\zeta )l(z)L(Z)๐\zeta `$ (42)
where $`l`$ and $`L`$ are polynomials such that $`l(0)=0`$ and $`L(0)=0`$. The symbol $``$ means that the integrals in RHS are not always defined, the regularization is defined in th Appendix B, in the next section we describe results of this regularization..
## 9 Deformed Abelian differentials.
In Appendix B we define the polynomials $`s_k(z)`$. These polynomials are of the form
$`s_k(z)=z^{g+1+k},gk0`$
$`s_k(z)={\displaystyle \frac{1}{i\gamma }}\left({\displaystyle \frac{q^k1}{q^k+1}}\right)z^{g+1+k}+\mathrm{},k1`$ (43)
where $`\mathrm{}`$ stands for terms of lower degree (containing $`t_j`$, $`t_j^{}`$ in coefficients) explicitly given in Appendix B.
In the classical case every polynomial defines an Abelian differentials on the affine curve $`X\mathrm{}^\pm `$. Similarly we consider the polynomials $`s_k`$ as corresponding to โdeformed Abelian differentials.โ Let us be more precise. The regularized integrals are defined in Appendix B in such a way that they satisfy several conditions. First of them is
$`s_k|S_l=0kg+1,l`$ (44)
$`s_k|S_l=0k,lg+1`$ (45)
Due to (44) we consider the polynomials $`s_k`$, $`kg+1`$ as corresponding to exact forms. The polynomials $`s_k`$ with $`k=1,\mathrm{},g`$ correspond to first kind differentials, $`s_o`$ corresponds to the third kind one, $`s_k`$ with $`k=1,\mathrm{},g`$ correspond to second kind differentials.
Explicitly the relation with classical case is as follows. Consider the case $`t(z)=t^{}(z)`$ and take the limit:
$$r_k=z^1\underset{\gamma 0}{lim}s_k(z)$$
Then the classical Abelian differential related to $`s_k`$ is
$$\mu _k=\frac{r_k(z)}{y}dz$$
Similar interpretation can be given to $`S_k`$ which correspond to Abelian differentials in dual classical limit $`\gamma \mathrm{}`$. However, the most interesting feature of our construction is that together with this cohomological interpretation an alternative โhomologicalโ one is possible. The polynomials $`S_k`$, $`kg+1`$ correspond to retractable cycles according to (45). The polynomials $`S_k`$ for $`k=\pm 1,\mathrm{},\pm g`$ are interpreted as analogues $`a`$ and $`b`$ cycles $`\delta _k`$ on the โdeformed affine curveโ, $`S_0`$ corresponds to cycle $`\delta _0`$ around $`\mathrm{}^+`$ which is non-trivial on the affine curve.The pairing $`l|L`$ defines the integral of differential defined by $`l`$ over cycle defined by $`L`$. The asymptotic of the integrals $`l|L`$ in the classical limit $`\gamma 0`$ are, indeed, described by Abelian integrals. Certainly, the opposite interpretation ($`l`$-cycle, $`L`$-differential ) is possible which corresponds to dual classical limit. It is not the first time that this kind of objects appears , but it is the first time that we observe real duality between two classical limits.
Let us define the pairing between two polynomials $`l_1`$ and $`l_2`$
$`l_1l_2=\underset{\mathrm{\Lambda }\mathrm{}}{lim}{\displaystyle \underset{\mathrm{\Lambda }}{\overset{\mathrm{\Lambda }+i\pi }{}}}[๐ฌ(\zeta )๐ฌ^{}(\zeta )l_1(z)\delta _\gamma ^1(๐ฌ๐ฌ^{}l_2)(\zeta i\pi )+`$
$`+๐ฌ(\zeta \pi i)๐ฌ^{}(\zeta i\pi )l_1(z)\delta _\gamma ^1(๐ฌ๐ฌ^{}l_2)(\zeta i\gamma )]d\zeta `$ (46)
One can show that these formulae give well-defined anti-symmetric pairings which correspond classically to natural pairing between meromorphic differentials
$$\omega _1\omega _2=\text{res}_{p=\mathrm{}^+}\left(\omega _1(p)\stackrel{p}{}\omega _2\right)$$
The polynomials $`s_{\pm j}`$ and for $`j=1,\mathrm{},g`$ constitute canonical basis:
$`s_ks_l=\text{sgn}(kl)\delta _{k,l}`$
Similarly, to introduce the definition of $`L_1L_2`$ it is sufficient to do necessary replacements in (46): $`l_iL_i`$, $`zZ`$, $`\gamma \pi `$. The polynomials $`S_{\pm j}`$ are canonically conjugated.
The following anti-symmetric polynomials play role of 2-forms $`\sigma `$ and $`\sigma ^{}`$ used in classics:
$`c(z_1,z_2)={\displaystyle \underset{j=1}{\overset{g}{}}}\left(s_j(z_1)s_j(z_2)s_j(z_1)s_j(z_2)\right),`$ (47)
$`C(Z_1,Z_2)={\displaystyle \underset{j=1}{\overset{g}{}}}\left(S_j(Z_1)S_j(Z_2)S_j(Z_1)S_j(Z_2)\right)`$
As usual the most important property of deformed Abelian integrals is that the Riemann bilinear relations remain valid after the deformation. Namely, consider the the following $`2g\times 2g`$ period matrix: $`P`$ with the matrix elements
$`P_{kl}=`$ $`s_k|S_l`$ $`k,l=g,\mathrm{},1,1,\mathrm{},g`$
The deformed Riemann bilinear identity is formulated as Proposition 2. The matrix $`P`$ belongs to the symplectic group:
$`PSp(2g)`$ (48)
This Proposition 2 is equivalent to a number of bilinear relations between the deformed Abelian integrals. Proving them it is convenient to consider the domain of small $`\gamma `$ ($`\gamma <\pi /n`$) when the regularization of integrals simplifies, and then continue analytically with respect to $`\gamma `$. Still the proof is rather complicated technically: it is based on non-trivial properties of the regularized integrals. We do not give this bulky proof here.
There is one more relation for deformed Abelian integrals. One can check that
$`d|S_k=0,k,d={\displaystyle \underset{j=1}{\overset{g}{}}}(t_jt_j^{})s_j`$ (49)
$`s_k|D=0,k,D={\displaystyle \underset{j=1}{\overset{g}{}}}(T_jT_j^{})S_j`$ (50)
The relations (49) do not have direct analogue in terms of Abelian integrals, recall that we put $`t_j=t_j^{}`$ taking classical limit which turns the relation into triviality. However, there is another way of taking the classical limit where this equation is important .
## 10 Return to quantization of affine Jacobian.
Let us return to the main subject of this paper: quantization of affine Jacobian. Consider any $`x๐(q)`$. Such $`x`$ is identified with $`xI๐(q)๐(Q)`$, so, due to Conjecture 3 the matrix element of $`x`$ can be presented as
$`t_1,\mathrm{}t_g|x|t_1^{},\mathrm{}t_g^{}=p_L(t_1,\mathrm{},t_g)p_R(t_1^{},\mathrm{},t_g^{})`$ (51)
$`\times {\displaystyle \underset{\mathrm{}}{\overset{\mathrm{}}{}}}d\zeta _1\mathrm{}{\displaystyle \underset{\mathrm{}}{\overset{\mathrm{}}{}}}d\zeta _gh(z_1,\mathrm{},z_g)H_I(Z_1,\mathrm{},Z_g){\displaystyle \underset{j=1}{\overset{g}{}}}๐ฌ(\zeta _j)๐ฌ^{}(\zeta _j)`$
where the eigen-values of $`๐ป_j`$ are the same at the left and at the right, so, we do not write them explicitly, the polynomial $`H_I`$ which corresponds to $`X=I`$ is given by
$$H_I(Z_1,\mathrm{},Z_g)=\underset{j=1}{\overset{g}{}}Z_j\underset{i<j}{}(Z_iZ_j)$$
notice that
$`H_I=S_1\mathrm{}S_g`$ (52)
The formula for the matrix elements (51) for small $`\gamma `$ (when no regularization of integrals is needed) can be deduced rigorously starting from realization of $`๐(q)`$ defined in Section 4. The following equations follow respectively from (44), (49), (48) (recall notations of Section 4):
$`s_k๐ฑ^{g1}0,kg+1`$ (53)
$`c๐ฑ^{g2}0,`$ (54)
$`d๐ฑ^{g1}0,`$ (55)
where $``$ means that these expression vanish being substitute into the integral (51). The equation (54) needs explanation. To prove this equation one has to take into account the Riemann bilinear identity (48) and the formula (52); notice that
$$S_iS_j=0,1i,jg$$
The formula for the matrix elements (51) can be rigorously deduced for small $`\gamma `$. Hence the equations (53), (54), (55) lead to equation for operators. The latter equations are obtained applying the operation $`\chi `$ (Section 4):
$`\chi \left(s_k๐ฑ^{g1}\right)=0,kg+1`$ (56)
$`\chi \left(c๐ฑ^{g2}\right)=0,`$ (57)
$`\chi \left(d๐ฑ^{k1}\right)=0`$ (58)
We conclude that the formulae for the polynomials $`s_k`$ needed in Section 4 are exactly the same as given in (43). Thus we put together the algebraic part of this work with the analytical one.
On the other hand the equations (56, 57, 58) are of purely algebraic character, so, if they are valid for small $`\gamma `$ they must be valid always. That is why we regularized the integrals for matrix element in order that the equations (53), (54), (55) hold for any $`\gamma `$.
Moreover, there is a dual model and we can consider the operators $`๐ณ=xX`$ from $`๐(q)๐(Q)`$. The equations (56), (57), (58) and dual equations still have to be valid. The regularized integrals are defined in such a way that it is the case. The equations (56), (58) and their duals clearly follow from (44), (45) and (49), (50). The most interesting is the equation (57). Due to the Riemann bilinear relation this equation follows from
$`c๐ฑ^{g2}0`$ (59)
which is true if the sub-space $`c๐ฑ^{g2}`$ of the space $`๐ฑ^g`$ is convoluted with the sub-space
$$\frac{๐ฑ_g}{C๐ฑ_{g2}}$$
where $`๐ฑ_k`$ is the same as $`๐ฑ^k`$ for dual model (this notation is not occasional: the space $`๐ฑ_k`$ plays role of $`k`$-cycles for $`k`$-forms from $`๐ฑ^k`$). In other words we impose the equation
$`C๐ฑ_{g2}=0`$
and the dual equation (59) is imposed automatically due to the Riemann bilinear relation.
Let us discuss the classical limit in some more details. Consider the hyper-elliptic curve $`X`$. If we realize this curve as characteristic equation of classical analogue of the monodromy matrix $`\stackrel{\mathbf{~}}{๐}(z)`$ (21) the branch points of the curve can be shown real non-negative. Actually requiring $`t_{g+1}=(1)^{g+1}2`$ we put one of branch points at $`z=0`$. Thus the branch points are $`0=q_1`$ $`<`$ $`\mathrm{}`$ $`<`$ $`q_{2g+2}`$. The Riemann surface is realized as two-sheet covering of the plane of $`z`$ with cuts $`I_k=[q_{2k1},q_{2k}]`$, $`k=1,\mathrm{},g+1`$. The canonical a-cycles $`\delta _j`$ and b-cycles $`\delta _j`$ are shown on the fig. 1 follows:
Under the classical dynamics every of the separated variables $`z_j`$ oscillates in the interval: $`q_{2j1}z_jq_{2j}`$, topologically it corresponds to motion along the a-cycle $`\delta _j`$. One can show that the integral $`s_k|S_j`$ is described in the classical limit $`\gamma 0`$ by $`\delta _j`$ of differential $`\mu _k`$. Thus the $`g`$-cycle (52) corresponds to classical trajectory $`\delta _1\mathrm{}\delta _g`$. Recall that the cycle (52) corresponds to insertion of unit operator of the dual model. Introducing other dual operators one gets integrals with respect to both a-cycles and b-cycles. Classically the coresponding trajectories are not real, but the factorization by $`\sigma ^{}W_{m2}`$ in (12) guarantees that the classical non-real trajectories are not singular. We would like to finish this paper with this topological interpretation of the dual model.
## 11 Appendix A.
In this Appendix we shall give the canonical definition of affine Jacobi variety $`J_{\text{aff}}(t)`$. Consider hyper-elliptic curve $`X`$ of genus $`g`$:
$`w^2t(z)w+1=0`$
We have the canonical basis with a-cycles $`\delta _k`$, $`gk<0`$ and b-cycles $`\delta _k`$, $`0<kg`$. Associate with this basis the basis of normalized holomorphic differentials $`\omega _j`$:
$$\underset{\delta _i}{}\omega _j=\delta _{ij},B_{ij}=\underset{\delta _i}{}\omega _j$$
The Jacobi variety of this curve is the $`g`$-dimensional complex torus:
$$J(t)=\frac{^g}{^g\times B^g}$$
With every point $`pX`$ we identify the point $`\alpha (p)J(t)`$ with coordinates:
$$\alpha _j(p)=\underset{b}{\overset{p}{}}\omega _j$$
for the reference point $`b`$ it is convenient to take one of the branch points. The curve $`X`$ has two points over the point $`z=\mathrm{}`$, denote them by $`\mathrm{}^\pm `$ and consider the $`(g1)`$-dimensional subvariety of $`J(t)`$ defined by
$$\mathrm{\Theta }^\pm =\{\zeta J(t)|\theta (\zeta +\alpha (\mathrm{}^{}))\theta (\zeta +\alpha (\mathrm{}^+))=0\}$$
where $`\theta `$ is Riemann theta-function. It can be shown that there exist an isomorphism:
$`J_{\text{aff}}(t)J(t)\mathrm{\Theta }^\pm `$ (60)
The equivalence of this description with the description in terms of divisors (Section 1) is due to the Abel map $`X[g]J(t)`$ explicitely given by
$$\zeta =\alpha (๐ซ)+\mathrm{\Delta },\alpha (๐ซ)=\alpha (p_j)$$
where $`\mathrm{\Delta }`$ is the Riemann characteristic.
## 12 Appendix B.
In this Appendix we describe the regularization of integrals which has been used in the paper.
Define:
$`\delta _\xi (f(\zeta ))=f(\zeta +i\xi )f(\zeta ),`$
$`\mathrm{\Delta }_\xi (f(\zeta ))=f(\zeta +i\xi )f(\zeta i\xi )`$
Introduce the polynomials
$`s_k(z)={\displaystyle \frac{1}{2i\gamma }}\{t(z)\mathrm{\Delta }_\gamma ^1[z^{kg1}t(z)]_>+t^{}(z)\mathrm{\Delta }_\gamma ^1[z^{kg1}t^{}(z)]_>`$
$`t(z)\mathrm{\Delta }_\gamma ^1[z^{kg1}q^{2(g+1k)}t^{}(zq^2)]_>`$
$`t^{}(z)\mathrm{\Delta }_\gamma ^1[z^{kg1}q^{2(g+1k)}t(zq^2)]_>`$
$`\frac{1}{2}\left(t^{}(z)[z^{kg1}t(z)]_>+t(z)[z^{kg1}t^{}(z)]_>\right)+`$
$`+(q^{2(g+1k)k}q^{2(kg1)})[z^{kg1}]_>\},k0;`$
$`s_k(z)=z^{g+1+k},gk0;`$
where the notation $`[]_>`$ means that only the positive degrees of Laurent series in brackets are taken. Obviously $`\text{deg}(s_k)=g+1+k`$. Further, with every function $`f(\zeta )`$ associate the functions:
$`u[f](\zeta )={\displaystyle \frac{1}{2i\gamma }}`$ $`\{t(z)\mathrm{\Delta }_\gamma ^1\left(f(\zeta )t(z)\right)+t^{}(z)\mathrm{\Delta }_\gamma ^1\left(f(\zeta )t^{}(z)\right)`$
$`t(z)\mathrm{\Delta }_\gamma ^1\left(f(\zeta i\gamma )t^{}(zq^2)\right)t^{}(z)\mathrm{\Delta }_\gamma ^1\left(f(\zeta i\gamma )t(zq^2)\right)`$
$`f(\zeta )t(z)t^{}(z)+f(\zeta +i\gamma )f(\zeta i\gamma )\},`$
$`v[f](\zeta )={\displaystyle \frac{1}{2i\gamma }}`$ $`\{(1)^{g+1}(\mathrm{\Delta }_\gamma ^1\left(f(\zeta i\gamma )t(zq^2)\right)๐ฌ(\zeta )๐ฌ^{}(\zeta i\gamma )+`$
$`+\mathrm{\Delta }_\gamma ^1\left(f(\zeta i\gamma )t^{}(zq^2)\right)๐ฌ(\zeta i\gamma )๐ฌ^{}(\zeta )`$
$`\mathrm{\Delta }_\gamma ^1\left(f(\zeta )t(z)\right)๐ฌ(\zeta i\gamma )๐ฌ^{}(\zeta )`$
$`\mathrm{\Delta }_\gamma ^1\left(f(\zeta )t^{}(z)\right)๐ฌ(\zeta )๐ฌ^{}(\zeta i\gamma ))+`$
$`+f(\zeta )๐ฌ(\zeta i\gamma )๐ฌ^{}(\zeta i\gamma )+f(\zeta i\gamma )๐ฌ(\zeta )๐ฌ^{}(\zeta )\}`$
Define
$$s_k^{}(\zeta )=\{\begin{array}{c}s_k(z)+u[f](\zeta ),f=z^{kg1},k1;\\ \\ s_0(z)+u[f](\zeta )+(1)^{g+1}2,f=\zeta z^{g1},k=0;\\ \\ z^{g+1+k},gk1\end{array}$$
and
$$p_k(\zeta )=\{\begin{array}{c}v[f](\zeta ),f=z^{kg1},k1;\\ \\ v[f](\zeta ),f=\zeta z^{g1},k=0;\\ \\ 0,gk1\end{array}$$
These definitions imply that
$`(s_k(z)+s_k^{}(z))๐ฌ(\zeta )๐ฌ^{}(\zeta )=\delta _\gamma (p_k(\zeta ))`$ (61)
Similarly one introduces the functions $`S_k(Z)`$, $`S_k^{}(\zeta )`$, $`P_k(\zeta )`$ changing everywhere $`z`$ by $`Z`$, $`q`$ by $`Q`$ and $`i\gamma `$-shift of $`\zeta `$ by $`i\pi `$-shift of $`\zeta `$. One has
$`(S_k(z)+S_k^{}(z))๐ฌ(\zeta )๐ฌ^{}(\zeta )=\delta _\pi (P_k(\zeta ))`$ (62)
Our goal is to define a pairing $`l|L`$ between two arbitrary polynomials $`l(z)`$ and $`L(Z)`$ such that $`l(0)=0`$, $`L(0)=0`$. Notice that every such polynomial $`l`$ ($`L`$) can be presented as linear combination of polynomials $`s_k`$ ($`S_k`$).
Consider the following two pictures:
We define
$`s_k|S_l{\displaystyle \underset{\mathrm{}}{\overset{\mathrm{\Lambda }_1}{}}}๐ฌ(\zeta )๐ฌ^{}(\zeta )s_k(z)S_l(Z)๐\zeta +`$
$`+{\displaystyle \underset{\mathrm{\Lambda }_1}{\overset{\mathrm{\Lambda }_2}{}}}๐ฌ(\zeta )๐ฌ^{}(\zeta )s_k^{}(\zeta )S_l(Z)๐\zeta +{\displaystyle \underset{\mathrm{\Lambda }_2}{\overset{\mathrm{}}{}}}๐ฌ(\zeta )๐ฌ^{}(\zeta )s_k^{}(\zeta )S_l^{}(\zeta )๐\zeta `$
$`{\displaystyle \underset{\mathrm{\Lambda }_1}{\overset{\mathrm{\Lambda }_1+i\gamma }{}}}S_l(Z)p_k(\zeta )๐\zeta {\displaystyle \underset{\mathrm{\Lambda }_2}{\overset{\mathrm{\Lambda }_2+i\pi }{}}}s_k^{}(\zeta )P_l(\zeta )๐\zeta ,`$ (63)
see (fig. 2a). The first integral in RHS converges at $`\mathrm{}`$ because $`l(0)=L(0)=0`$. The equations (61,62) guarantee that the regularization (63) does not depend on $`\mathrm{\Lambda }_1`$, $`\mathrm{\Lambda }_2`$ if they remain ordered: $`\mathrm{\Lambda }_1<\mathrm{\Lambda }_2`$. Moreover, let us transform (fig. 2a) into (fig. 2b). The alternative definition of regularized integral referring to (fig. 2b) is
$`s_k|S_l{\displaystyle \underset{\mathrm{}}{\overset{\mathrm{\Lambda }_2}{}}}๐ฌ(\zeta )๐ฌ^{}(\zeta )s_k(z)S_l(Z)๐\zeta +`$
$`+{\displaystyle \underset{\mathrm{\Lambda }_2}{\overset{\mathrm{\Lambda }_1}{}}}๐ฌ(\zeta )๐ฌ^{}(\zeta )s_k(z)S_l^{}(\zeta )๐\zeta +{\displaystyle \underset{\mathrm{\Lambda }_1}{\overset{\mathrm{}}{}}}๐ฌ(\zeta )๐ฌ^{}(\zeta )s_k^{}(\zeta )S_l^{}(\zeta )๐\zeta `$
$`{\displaystyle \underset{\mathrm{\Lambda }_1}{\overset{\mathrm{\Lambda }_1+i\gamma }{}}}s_k(z)P_l(\zeta )๐\zeta {\displaystyle \underset{\mathrm{\Lambda }_2}{\overset{\mathrm{\Lambda }_2+i\pi }{}}}S_l^{}(\zeta )p_k(\zeta )๐\zeta `$ (64)
The equivalence of the regularizations (63) and (64) is based on the following fact. It is easy to realize that for any $`l`$ and $`L`$ there exist a function $`X_{kl}(\zeta )`$ such that
$`(S_l(Z)S_l^{}(\zeta ))p_k(\zeta )=\delta _\pi (X_{kl}(\zeta )),`$
$`(s_k(z)s_k^{}(\zeta ))P_l(\zeta )=\delta _\gamma (X_{kl}(\zeta ))`$
The equivalence in question follows from the equality:
$`{\displaystyle \underset{\mathrm{\Lambda }}{\overset{\mathrm{\Lambda }+i\gamma }{}}}(S_l(Z)S_l^{}(\zeta ))g(\zeta )๐\zeta =\left({\displaystyle \underset{\mathrm{\Lambda }+i\pi }{\overset{\mathrm{\Lambda }+i\pi +i\gamma }{}}}{\displaystyle \underset{\mathrm{\Lambda }}{\overset{\mathrm{\Lambda }+i\gamma }{}}}\right)X_{kl}(\zeta )d\zeta =`$
$`=\left({\displaystyle \underset{\mathrm{\Lambda }+i\gamma }{\overset{\mathrm{\Lambda }+i\pi +i\gamma }{}}}{\displaystyle \underset{\mathrm{\Lambda }}{\overset{\mathrm{\Lambda }+i\pi }{}}}\right)X_{kl}(\zeta )d\zeta ={\displaystyle \underset{\mathrm{\Lambda }}{\overset{\mathrm{\Lambda }+i\pi }{}}}(s_k(z)s_k^{}(\zeta ))G(\zeta )๐\zeta `$ |
warning/0001/cond-mat0001448.html | ar5iv | text | # Ising Model and ๐ฟ โ Function
## 1 Introduction
Onsager and Kaufman invented the formula for the partition function of the two dimensional Ising model. Another method for the calculation of the partition function was proposed by Kac and Ward . They considered simultaneously two formulae: the determinant of the special matrix $`I+T`$ ($`I`$ is the identity matrix) is proportional to the partition function of the Ising model and it is proportional to the square of the partition function. For the proof of the first formula they used a topological statement. Sherman , constructed a counter โ example for this statement. Hurst and Green proposed to use for the calculation of the Ising model partition function not a determinant but a Pfaffian of some special matrix. This method was improved by Kasteleyn , Fisher , McCoy and Wu . McCoy and Wu obtained the wrong formula connecting the Pfaffians with the Ising model partition function. The proper formula of this type is obtained in the paper . Sherman , gave some arguments for the equality
$$Z^2=C(\beta )det(I+T)$$
(1.1)
where $`Z`$ is the partition function of the two dimensional Ising model with the free boundary conditions and $`C(\beta )`$ is the positive function of the inverse temperature $`\beta `$. In the paper the following formula
$$Z^2=C(\beta )det(IT)$$
(1.2)
is proved. For the rectangular lattice the expression (1.2) is independent of the sign of the matrix $`T`$. For an arbitrary lattice the formula (1.1) is wrong. If the matrices $`T^k`$ satisfy some estimate, then
$$det(IT)=\mathrm{exp}\{\underset{k=1}{\overset{\mathrm{}}{}}k^1\text{tr}T^k\}.$$
(1.3)
By the definition the partition function $`Z>0`$. If the numbers $`\text{tr}T^k`$ are real, then the equalities (1.2) and (1.3) imply
$$Z(C(\beta ))^{1/2}=\mathrm{exp}\{1/2\underset{k=1}{\overset{\mathrm{}}{}}k^1\text{tr}T^k\}.$$
(1.4)
The bulk of Sherman papers , and the subsequent Burgoyne paper were devoted to the proof of Feynman conjecture. Due to Feynman conjecture the Ising model partition function is proportional to some infinite formal product. The right hand side of the equality (1.4) is the product. By calculating the numbers $`\text{tr}T^k`$ it is possible to show that the equality (1.4) is a correct statement for Feynman conjecture.
Hashimoto studied some special infinite product. He called these products zeta functions of finite graphs in the paper and $`L`$ โ functions of finite graphs in the paper . It is possible to prove that the right hand side of the equality (1.3) is one over the Hashimoto $`L`$ โ function (zeta function). The definition (1.3) seems suitable since the series in (1.3) is convergent if the numbers $`\text{tr}T^k`$ satisfy some estimate.
The paper , , used the well โ known van der Waerden formula for the Ising model partition function. Similar formula for the correlation functions was obtained in the paper . By using these formulae and the formula (1.4) we caculate the thermodynamic limit of the free energy and the correlation functions of the two dimensional Ising model with free boundary conditions. Similar results are obtained for periodic boundary conditions.
In the second section we discuss Hashimoto results , . The third section is devoted to formula for the correlation functions. In the fourth section we study the two dimensional Ising model with free boundary conditions. The fifth section is devoted to the two dimensional Ising model with periodic boundary conditions.
## 2 $`L`$ โ Function
Let $`s`$ be a complex number. Then for $`\mathrm{}s>1`$ Riemann zeta function
$$\zeta (s)=\underset{n=1}{\overset{\mathrm{}}{}}n^s$$
(2.1)
is an analytic function of the variable $`s`$. Euler showed that for $`\mathrm{}s>1`$ (,formula 17.7.2)
$$\zeta (s)=\underset{p}{}(1p^s)^1$$
(2.2)
where the product is extended over the set of all prime numbers $`p=2,3,5,7,\mathrm{}`$.
Let $`n>1`$ be a fixed natural number and let $`m`$ be any natural number. Let us consider the functions $`\chi (m)`$ such that
$`\chi (m)=\chi (m^{})`$, if $`m=m^{}\text{mod}n`$,
$`\chi (1)=1`$,
$`\chi (m)=0`$, if the greatest common divisor $`(m,n)`$ of the natural numbers $`m`$ and $`n`$ is not one,
$$\chi (m)\chi (m^{})=\chi (mm^{}).$$
(2.3)
Such function $`\chi (m)`$ is called a modulo $`n`$ character.
Let $`n>1`$ be a fixed natural number and let $`\chi `$ be a modulo $`n`$ character. Then the series
$$L(s,\chi )=\underset{m=1}{\overset{\mathrm{}}{}}\chi (m)m^s$$
(2.4)
for $`\mathrm{}s>1`$ is called the $`L`$ โ series. The $`L`$ โ series was introduced by Dirichlet. The $`L`$ โ series and Riemann zeta function have many common properties. The Euler product (2.2) analogue is (, formula 17.8.5)
$$L(s,\chi )=\underset{m=1}{\overset{\mathrm{}}{}}(1\chi (p)p^s)^1$$
(2.5)
where $`\mathrm{}s>1`$ and the product is extended over the set of all prime numbers.
A function $`\chi (m)`$ of the natural number $`m`$ satisfying the condition (2.3) only is called a character. The function $`\chi (m,s)=m^s`$ gives an example of a character. Let us consider the series (2.4) with the character $`\chi (m)`$. When the character $`\chi (m)1`$ the series (2.4) coincides with Riemann zeta function (2.1). When the character $`\chi (m)`$ is a modulo $`n`$ character the series (2.4) is Dirichlet $`L`$ โ series.
Let us introduce $`L`$ โ function of the finite graph. Let $`X`$ be a finite graph and let $`๐`$ be an oriented edge of a graph $`X`$. The oriented edge $`๐`$ is defined by the pair of the vertices of a graph $`X`$: the beginning $`b(๐)`$ and the end $`f(๐)`$ of the oriented edge $`๐`$. A closed path is a sequence of the oriented edges $`C=(๐_1,\mathrm{},๐_k)`$ such that
$$b(๐_{i+1})=f(๐_i),i=1,\mathrm{},k1,b(๐_1)=f(๐_k).$$
(2.6)
Let us denote $`b(C)=b(๐_1)`$ and $`f(C)=f(๐_k)`$. For a closed path $`b(C)=f(C)`$. Let $`๐^1`$ be such oriented edge that $`b(๐^1)=f(๐)`$, $`f(๐^1)=b(๐)`$. The following pairs of the closed paths are regarded homotopic equivalent
$$(๐,๐^1)C(b(๐))$$
(2.7)
where the path $`C(b(๐))`$ consists of the only vertex $`b(๐)`$;
$$(๐_1,\mathrm{},๐_{i1},๐_i,๐_i^1,๐_{i+2},\mathrm{},๐_k)(๐_1,\mathrm{},๐_{i1},๐_{i+2},\mathrm{},๐_k)$$
(2.8)
where $`i=2,\mathrm{},k2`$, $`k>3`$;
$$(๐_1,๐_1^1,๐_3,\mathrm{},๐_k)(๐_3,\mathrm{},๐_k);$$
(2.9)
$$(๐_1,\mathrm{},๐_{k2},๐_{k1},๐_{k1}^1)(๐_1,\mathrm{},๐_{k2});$$
(2.10)
$$(๐_1,๐_2,\mathrm{},๐_{k1},๐_1^1)(๐_2,\mathrm{},๐_{k1}).$$
(2.11)
Two closed paths are regarded homotopic if they are related by the equivalence relation generated by the relations (2.7) โ (2.11).
Lemma 2.1. The closed paths $`(๐_k,๐_1,\mathrm{},๐_{k1})`$ and $`(๐_1,\mathrm{},๐_k)`$ are homotopic.
Proof. The equivalence relation (2.11) implies
$$(๐_k,๐_1,\mathrm{},๐_k,๐_k^1)(๐_1,\mathrm{},๐_k).$$
(2.12)
The equivalence relation (2.10) implies
$$(๐_k,๐_1,\mathrm{},๐_k,๐_k^1)(๐_k,๐_1,\mathrm{},๐_{k1}).$$
(2.13)
It follows from the equivalence relations (2.12) and (2.13) that the closed paths $`(๐_1,\mathrm{},๐_k)`$ and $`(๐_k,๐_1,\mathrm{},๐_{k1})`$ are homotopic. The lemma is proved.
The equivalence relations (2.7) โ (2.11) imply that every closed path is homotopic to a path consisting of the only vertex or is homotopic to a reduced closed path $`(๐_1,\mathrm{},๐_k)`$:
$$b(๐_{i+1})=f(๐_i),f(๐_{i+1})b(๐_i),i=1,\mathrm{},k1,b(๐_1)=f(๐_k),f(๐_1)b(๐_k).$$
(2.14)
In view of Lemma 2.1 $`k`$ reduced closed paths $`(๐_1,\mathrm{},๐_k)`$, $`(๐_k,๐_1,\mathrm{},๐_{k1})`$,โฆ, $`(๐_2,\mathrm{},๐_k,๐_1)`$ are homotopic to each other. They define the equivalence class called the oriented reduced cycle. We denote it by $`[(๐_1,\mathrm{},๐_k)]`$. An oriented reduced cycle, or a reduced closed path $`(๐_1,\mathrm{},๐_k)`$ representing it, is called non โ primitive, if there exists a positive integer $`l`$ ($`1l<k`$) such that $`(๐_1,\mathrm{},๐_k)=(๐_{l+1},\mathrm{},๐_k,๐_1,\mathrm{},๐_l)`$; and otherwise it is called primitive.
If two closed paths $`C_1=(๐_1,\mathrm{},๐_k)`$, $`C_2=(๐_{k+1},\mathrm{},๐_{k+l})`$ have the same beginning vertex: $`b(๐_1)=b(๐_{k+1})`$, it is possible to define the product $`C_1C_2=(๐_1,\mathrm{},๐_k,๐_{k+1},\mathrm{},๐_{k+l})`$. If $`b(๐_1)=b(๐_{k+1})`$ and $`๐_{k+1}๐_1`$, then the product $`C_1C_3=(๐_1,\mathrm{},๐_k,๐_{k+1},\mathrm{},๐_1^1)`$ of two reduced closed paths $`C_1=(๐_1,\mathrm{},๐_k)`$ and $`C_3=(๐_{k+1},..,๐_1^1)`$ is not a reduced closed path. The product $`C_1C_2=(๐_1,\mathrm{},๐_{k+l})`$ of two reduced closed paths $`C_1=(๐_1,\mathrm{},๐_k)`$ and $`C_2=(๐_{k+1},\mathrm{},๐_{k+l})`$ with the same beginning edge: $`๐_1=๐_{k+1}`$ is the reduced closed path.
Let a unitary matrix $`\rho (C)`$ in $`n`$ dimensional space correspond with any reduced closed path $`C`$ and satisfy the following conditions: if two reduced closed paths $`C_1=(๐_1,\mathrm{},๐_k)`$ and $`C_2=(๐_{k+1},\mathrm{},๐_{k+l})`$ have the same beginning edge: $`๐_1=๐_{k+1}`$ then
$$\rho (C_1C_2)=\rho (C_1)\rho (C_2);$$
(2.15)
there exists a unitary matrix $`\gamma `$ for any reduced closed path $`(๐_1,\mathrm{},๐_k)`$ such that
$$\rho ((๐_k,๐_1,\mathrm{},๐_{k1}))=\gamma \rho ((๐_1,\mathrm{},๐_k))\gamma ^1.$$
(2.16)
Lemma 2.2. Let a unitary matrix $`\rho (๐_1;๐_2)`$ correspond with any pair $`๐_1,๐_2`$ of the oriented edges of the graph $`X`$ such that $`f(๐_1)=b(๐_2)`$, $`b(๐_1)f(๐_2)`$. For any reduced closed path $`(๐_1,\mathrm{},๐_k)`$ we define the unitary matrix
$$\rho ((๐_1,\mathrm{},๐_k))=\rho (๐_1;๐_2)\rho (๐_2;๐_3)\mathrm{}\rho (๐_{k1};๐_k)\rho (๐_k;๐_1).$$
(2.17)
Then the matrix (2.17) satisfies the relations (2.15) and (2.16).
Proof. Let two reduced closed paths $`C_1=(๐_1,\mathrm{},๐_k)`$ and $`C_2=(๐_{k+1},\mathrm{},๐_{k+l})`$ have the same beginning edge: $`๐_1=๐_{k+1}`$. Then the definition (2.17) implies the relation (2.15).
Let $`(๐_1,\mathrm{},๐_k)`$ be a reduced closed path. Then the definition (2.17) implies the relation (2.16) with the unitary matrix $`\gamma =\rho (๐_k;๐_1)`$. The lemma is proved.
By a labelling on the set of non โ oriented edges of the graph $`X`$ we mean an assignment $`eu(e)=u(๐)=u(๐^1)`$, where $`u(e)`$ are idependent variables for the different non โ oriented edges. We denote them simply by $`๐ฎ=\{u(e)\}`$. We put
$$๐ฎ^C=\underset{i=1}{\overset{k}{}}u(๐_i)$$
(2.18)
where $`C=(๐_1,\mathrm{},๐_k)`$ is a reduced closed path. The $`L`$ โ function of $`X`$ attached to $`\rho `$ is defined by the following formal infinite product similar to the product (2.5)
$$L(๐ฎ,\rho ;X)=\underset{[C]:primitive}{}det(I_n\rho (C)๐ฎ^C)^1$$
(2.19)
where the product is extended over the set of primitive oriented reduced cycles $`[C]`$ and $`I_n`$ is the idetity matrix in the $`n`$ dimensional space.
If the graph $`X`$ is connected, then for any reduced closed path $`C`$ there exists a homotopic closed path $`C`$ such that $`b(C)=p`$ where $`p`$ is the fixed vertex. In view of the relation (2.9) we may consider that the beginning edge of the closed path $`C`$ is the fixed oriented edge $`๐`$ and $`b(๐)=p`$. Hashimoto , considered the unitary representation $`\rho `$ of the group of the classes of homotopically equivalent closed paths $`C`$ with the fixed initial vertex $`p`$. If we substitute the matrix $`\rho (C)`$ instead of the matrix $`\rho (C)`$ in the definition (2.19), we obtain Hashimoto definition . In the paper the function $`L(๐ฎ,\rho ;X)`$ is denoted by $`Z_X(๐ฎ;\rho )`$ and it is called a zeta function.
The definition (2.19) is formal. We transform it into another form. We denote by $`\alpha _i(C),i=1,\mathrm{},n`$, the eigenvalues of the unitary matrix $`\rho (C)`$. Since the matrix $`\rho (C)`$ is unitary,
$$|\alpha _i(C)|=1,i=1,\mathrm{},n.$$
(2.20)
Taking the logarithm of (2.19) we have
$$\mathrm{ln}L(๐ฎ,\rho ;X)=\underset{[C]:primitive}{}\mathrm{ln}[det(I_n\rho (C)๐ฎ^C)^1]$$
(2.21)
$`\mathrm{ln}[det(I_n\rho (C)๐ฎ^C)^1]={\displaystyle \underset{i=1}{\overset{n}{}}}\mathrm{ln}(1\alpha _i(C)๐ฎ^C)=`$
$`{\displaystyle \underset{i=1}{\overset{n}{}}}{\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}k^1(\alpha _i(C))^k(๐ฎ^C)^k={\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}k^1(\text{tr}\rho (C)^k)(๐ฎ^C)^k.`$ (2.22)
The substiturion of the equality (2) into the equality (2.21) gives
$$\mathrm{ln}L(๐ฎ,\rho ;X)=\underset{[C]:primitive}{}\underset{k=1}{\overset{\mathrm{}}{}}k^1(๐ฎ^C)^k\text{tr}\rho (C)^k.$$
(2.23)
Any primitive oriented reduced cycle $`[(๐_1,\mathrm{},๐_l)]`$ consists of $`l`$ different reduced closed paths: $`(๐_1,\mathrm{},๐_l)`$, $`(๐_l,๐_1,\mathrm{}๐_{l1})`$,โฆ, $`(๐_2,\mathrm{},๐_l,๐_1)`$. Due to the definition (2.18)
$$๐ฎ^{(e_l,e_1,\mathrm{},e_{l1})}=๐ฎ^{(e_1,\mathrm{},e_k)}.$$
(2.24)
for the reduced closed paths $`(๐_l,๐_1,\mathrm{},๐_{l1})`$ and $`(๐_1,\mathrm{},๐_l)`$. It follows from the equalities (2.16) and (2.24) that the equality (2.23) may be rewritten as
$$\mathrm{ln}L(๐ฎ,\rho ;X)=\underset{C:primitive}{}\underset{k=1}{\overset{\mathrm{}}{}}(k|C|)^1(๐ฎ^C)^k\text{tr}\rho (C)^k$$
(2.25)
where $`C`$ runs over the set of primitive reduced closed paths $`C=(๐_1,\mathrm{},๐_l)`$ and the length $`|C|=l`$.
The definition (2.18) implies
$$(๐ฎ^C)^k=๐ฎ^{C^{\times k}}.$$
(2.26)
The equalities (2.15), (2.25), (2.26) and the equality $`|C^{\times k}|=k|C|`$ imply
$$\mathrm{ln}L(๐ฎ,\rho ;X)=\underset{C:primitive}{}\underset{k=1}{\overset{\mathrm{}}{}}|C^{\times k}|^1๐ฎ^{C^{\times k}}\text{tr}\rho (C^{\times k}).$$
(2.27)
Any reduced closed path has the form $`C^{\times k}`$ where $`k`$ is a natural number and $`C`$ is a primitive reduced closed path. Then the equality (2.27) may be rewritten as
$$\mathrm{ln}L(๐ฎ,\rho ;X)=\underset{CRC(X)}{}|C|^1๐ฎ^C\text{tr}\rho (C)$$
(2.28)
where $`C`$ runs over the set $`RC(X)`$ of all reduced closed paths on the graph $`X`$.
The number of the non โ oriented edges which is incident to a vertex $`p`$ is called $`v(p)`$, the valency of $`p`$. Let
$$v=\underset{p}{\mathrm{max}}v(p).$$
(2.29)
Let us construct a reduced closed path with the initial vertex $`p`$. For the first edge we have $`v(p)`$ possibilities. For any other edge the number of possibilities is less than $`v1`$. Thus the total number of reduced closed paths of the length $`l`$ is less than $`\mathrm{\#}(VX)v(v1)^{l1}`$ where $`\mathrm{\#}(VX)`$ is the total number of vertices of the graph $`X`$. The relations (2.20) imply that the series (2.28) is absolutely convergent if the following estimate is valid
$$\underset{e}{\mathrm{max}}|u(๐)|<(v1)^1.$$
(2.30)
Definition. The $`L`$function of the finite graph $`X`$ attached to $`\rho `$ defined by
$$L(๐ฎ,\rho ;X)=\mathrm{exp}\{\underset{CRC(X)}{}|C|^1๐ฎ^C\text{tr}\rho (C)\}$$
(2.31)
where $`C`$ runs over the set $`RC(X)`$ of all reduced closed paths on the graph $`X`$.
Let $`\mathrm{\#}(๐X)`$ be the total number of oriented edges of the graph $`X`$.
Theorem 2.3. Let unitary matrix $`\rho (๐_1;๐_2)=\{(\rho (๐_1;๐_2))_{k_1k_2},k_1,k_2=1,\mathrm{},n\}`$ correspond with any pair $`๐_1,๐_2`$ of oriented edges of the graph $`X`$, such that $`f(๐_1)=b(๐_2)`$, $`b(๐_1)f(๐_2)`$. Let us define $`(n\mathrm{\#}(๐X))\times (n\mathrm{\#}(๐X))`$matrix
$$T(๐ฎ,\rho )_{(k_1,e_1),(k_2,e_2)}=\{\genfrac{}{}{0pt}{}{u(๐_1)(\rho (๐_1;๐_2))_{k_1k_2},f(๐_1)=b(๐_2),b(๐_1)f(๐_2),}{0,otherwise.}$$
(2.32)
If the estimate (2.30) is fulfilled, then
$$L(๐ฎ,\rho ;X)=det(IT(๐ฎ,\rho ))^1$$
(2.33)
where $`L`$fuction $`L(๐ฎ,\rho ;X)`$ is defined by the equality (2.31).
Proof. Analogously to the equality (2) we get
$$\mathrm{ln}[det(IT(๐ฎ,\rho ))^1]=\underset{k=1}{\overset{\mathrm{}}{}}k^1\text{tr}T(๐ฎ,\rho )^k.$$
(2.34)
Due to the definitions (2.17), (2.18) and (2.32) we have
$$\text{tr}T(๐ฎ,\rho )^k=\underset{\genfrac{}{}{0pt}{}{CRC(X),}{|C|=k}}{}๐ฎ^C\text{tr}\rho (C)$$
(2.35)
where $`C`$ runs over the set of all reduced closed paths of the length $`k`$. The substitution of the equality (2.35) into the right hand side of the equality (2.34) gives
$$\mathrm{ln}[det(IT(๐ฎ,\rho ))^1]=\underset{CRC(X)}{}|C|^1๐ฎ^C\text{tr}\rho (C)$$
(2.36)
where $`C`$ runs over the set $`RC(X)`$ of all reduced closed paths on the graph $`X`$. The equalities (2.31) and (2.36) imply the equality (2.33). The theorem is proved.
## 3 Ising Model
We consider a rectangular lattice on the plane formed by the points with integral Cartesian coordinates $`x=k_1`$, $`y=k_2`$, $`M_1^{}k_1M_1`$, $`M_2^{}k_2M_2`$, and the corresponding horizontal and vertical edges connecting these vertices. We denote this graph by $`G(M_1^{},M_2^{};M_1,M_2)`$. Let a graph $`G`$ be embedded in a rectangular lattice on the plane. Let all the vertices from the boundaries of all the edges of a graph $`G`$ be included into the set of the vertices of a graph $`G`$. The cell complex $`P(G)`$ is called the set consisting of the cells (vertices, edges, faces). A vertex of $`P(G)`$ is called a cell of dimension $`0`$. It is denoted by $`s_i^0`$. An edge of $`P(G)`$ is called a cell of dimension $`1`$. It is denoted by $`s_i^1`$. A face of $`P(G)`$ is called a cell of dimension $`2`$. It is denoted by $`s_i^2`$. We suppose that $`P(G)`$ contains all the faces whose all boundary edges are included into a graph $`G`$. We denote by $`๐_2^{add}`$ the group of modulo $`2`$ residuals. The modulo $`2`$ residuals are multiplied by each other and the group $`๐_2^{add}`$ is a field. To every pair of the cells $`s_i^p`$, $`s_j^{p1}`$ there corresponds the number $`(s_i^p:s_j^{p1})๐_2^{add}`$ (incidence number). If the cell $`s_j^{p1}`$ is included into the boundary of the cell $`s_i^p`$, then the incidence number $`(s_i^p:s_j^{p1})=1`$. Otherwise the incidence number $`(s_i^p:s_j^{p1})=0`$. For any pair of the cells $`s_i^2`$, $`s_j^0`$ the incidence numbers satisfy the condition
$$\underset{m}{}(s_i^2:s_m^1)(s_m^1:s_j^0)=0\text{mod}\mathrm{\hspace{0.17em}2}.$$
(3.1)
Indeed, if the vertex $`s_j^0`$ is not included into the boundary of the square $`s_i^2`$, then the condition (3.1) is fulfilled. If the vertex $`s_j^0`$ is included into the boundary of the square $`s_i^2`$, then it is included into the boundaries of four edges $`s_m^1`$ two of which are included into the boundary of the square $`s_i^2`$. The condition (3.1) is fulfilled again.
Let $`G(M_1^{},M_2^{};M_1,M_2)`$ be a rectangular lattice on the plane. We identify the opposite sides of the entire rectangular. The obtained graph is denoted by $`\stackrel{~}{G}(M_1^{},M_2^{};M_1,M_2)`$. It is called the rectangular lattice on the torus. Let a graph $`G`$ be embedded in the graph $`\stackrel{~}{G}(M_1^{},M_2^{};M_1,M_2)`$. Let all the vertices from the boundaries of all the edges of a graph $`G`$ be included into the set of the vertices of a graph $`G`$. The cell complex $`P(G)`$ is the set consisting of vertices, edges and faces. The sets of vertices and edges of $`P(G)`$ coincide with the sets of vertices and edges of a graph $`G`$. The set of faces of $`P(G)`$ consists of all the faces whose all edges of the boundaries are included into a graph $`G`$. The cells $`s_i^p`$ and the incidence numbers $`(s_i^p:s_j^{p1})`$ are defined similarly to the plane case. For any pair of the cells $`s_i^2`$, $`s_j^0`$ the incidence numbers satisfy the condition (3.1).
We suppose that a graph $`G`$ is embedded in a rectagular lattice on the plane or on the torus. A cochain $`c^p`$ of the complex $`P(G)`$ with the coefficients in the group $`๐_2^{add}`$ is a function on the $`p`$ dimensional cells taking values in the group $`๐_2^{add}`$. Usually the cell orientation is considered and the cochains are the antisymmetric functions: $`c^p(s^p)=c^p(s^p)`$. However, $`1=1\text{mod}\mathrm{\hspace{0.17em}2}`$ and we can neglect the cell orietation for the coefficients in the group $`๐_2^{add}`$. The cochains form an Abelian group
$$(c^p+c^p)(s_i^p)=c^p(s_i^p)+c^p(s_i^p)\text{mod}\mathrm{\hspace{0.17em}2}.$$
(3.2)
It is denoted by $`C^p(P(G),๐_2^{add})`$. The mapping
$$c^p(s_i^{p1})=\underset{j}{}(s_j^p:s_i^{p1})c^p(s_j^p)\text{mod}\mathrm{\hspace{0.17em}2}$$
(3.3)
defines the homomorphism of the group $`C^p(P(G),๐_2^{add})`$ into the group $`C^{p1}(P(G),๐_2^{add})`$. It is called the boundary operator. The mapping
$$^{}c^p(s_i^{p+1})=\underset{j}{}(s_i^{p+1}:s_j^p)c^p(s_j^p)\text{mod}\mathrm{\hspace{0.17em}2}$$
(3.4)
defines the homomorphism of the group $`C^p(P(G),๐_2^{add})`$ into the group $`C^{p+1}(P(G),๐_2^{add})`$. It is called the coboundary operator. The condition (3.1) implies $`=0`$, $`^{}^{}=0`$. The kernel $`Z_p(P(G),๐_2^{add})`$ of the homomorphism (3.3) on the group $`C^p(P(G),๐_2^{add})`$ is called the group of cycles of the complex $`P(G)`$ with the coefficients in the group $`๐_2^{add}`$. The image $`B_p(P(G),๐_2^{add})`$ of the homomorphism (3.3) in the group $`C^p(P(G),๐_2^{add})`$ is called the group of boundaries of the complex $`P(G)`$ with the coefficients in the group $`๐_2^{add}`$. Since $`=0`$, the group $`B_p(P(G),๐_2^{add})`$ is the subgroup of the group $`Z_p(P(G),๐_2^{add})`$. Analogously, for the coboundary operator $`^{}`$ the group of cocycles $`Z^p(P(G),๐_2^{add})`$ and the group of coboundaries $`B^p(P(G),๐_2^{add})`$ are defined.
It is possible to introduce the bilinear form on $`C^p(P(G),๐_2^{add})`$:
$$f^p,g^p=\underset{i}{}f^p(s_i^p)g^p(s_i^p)\text{mod}\mathrm{\hspace{0.17em}2}.$$
(3.5)
The definitions (3.3) and (3.4) imply
$`f^p,^{}g^{p1}=f^p,g^{p1}`$
$`f^p,g^{p+1}=^{}f^p,g^{p+1}.`$ (3.6)
Let a cochain $`\sigma C^0(P(G),๐_2^{add})`$. Let the energy be expressed in the form
$$H^{}(^{}\sigma )=\underset{s_i^1P(G)}{}h_i(^{}\sigma (s_i^1))$$
(3.7)
where $`h_i(ฯต)`$ is an arbitrary function on the group $`๐_2^{add}`$:
$$h_i(ฯต)=D_iE_i(1)^ฯต$$
(3.8)
and the constants
$`D_i=1/2(h_i(1)+h_i(0))`$
$`E_i=1/2(h_i(1)h_i(0)).`$ (3.9)
The substitution of the equality (3.8) into the equality (3.7) gives
$$H^{}(^{}\sigma )=\underset{s_i^1P(G)}{}D_i+H(^{}\sigma )$$
(3.10)
where the function
$$H(^{}\sigma )=\underset{s_i^1P(G)}{}E_i(1)^{^{}\sigma (s_i^1)}$$
(3.11)
is called the energy for the Ising model with zero magnetic field. The number $`E_i=E(s_i^1)`$ is the interaction energy attached to the edge $`s_i^1`$. The edge $`s_i^1`$ is given by its initial vertex and by its direction. For example, the edges of a rectangular lattice on the plane may be horizontal or vertical. If the interaction energy $`E_i=E(s_i^1)`$ is independent of the initial vertex of the edge $`s_i^1`$, the Ising model is called homogeneous. If the interaction energy $`E_i=E(s_i^1)`$ is independent of the direction of the edge $`s_i^1`$, the Ising model is called isotropic.
The equality (3.10) implies
$`Z_G^{}={\displaystyle \underset{\sigma C^0(P(G),Z_2^{add})}{}}\mathrm{exp}\{\beta H^{}(^{}\sigma )\}=`$
$`Z_G\mathrm{exp}\{\beta {\displaystyle \underset{s_i^1P(G)}{}}D_i\}`$ (3.12)
where the function
$$Z_G=\underset{\sigma C^0(P(G),Z_2^{add})}{}\mathrm{exp}\{\beta H(^{}\sigma )\}$$
(3.13)
is called the partition function of Ising model.
Let the cochain $`\chi C^0(P(G),๐_2^{add})`$ take the value $`1`$ at the vertices $`x_1`$,โฆ,$`x_m`$ and be equal to $`0`$ at all other vertices of the graph $`G`$. The correlation function at the vertices $`x_1`$,โฆ,$`x_m`$ of the lattice $`G`$ is the function
$`W_G(\chi )=(Z_G^{})^1{\displaystyle \underset{\sigma C^0(P(G),Z_2^{add})}{}}(1)^{\chi ,\sigma }\mathrm{exp}\{\beta H^{}(^{}\sigma )\}=`$
$`(Z_G)^1{\displaystyle \underset{\sigma C^0(P(G),Z_2^{add})}{}}(1)^{\chi ,\sigma }\mathrm{exp}\{\beta H(^{}\sigma )\}.`$ (3.14)
Proposition 3.1. The partition fuction of Ising model on the graph $`G`$
$`Z_G=2^{\mathrm{\#}(VG)}\left({\displaystyle \underset{s_i^1P(G)}{}}\mathrm{cosh}\beta E(s_i^1)\right)\times `$
$`{\displaystyle \underset{\xi ^1Z_1(P(G),Z_2^{add})}{}}{\displaystyle \underset{s_i^1P(G)}{}}(\mathrm{tanh}\beta E(s_i^1))^{1/2(1(1)^{\xi ^1(s_i^1)})}`$ (3.15)
where $`\mathrm{\#}(VG)`$ is the total number of the vertices of the graph $`G`$.
The correlation function of Ising model on the graph $`G`$
$`W_G(\chi )=(Z_G)^12^{\mathrm{\#}(VG)}\left({\displaystyle \underset{s_i^1P(G)}{}}\mathrm{cosh}\beta E(s_i^1)\right)\times `$
$`{\displaystyle \underset{\genfrac{}{}{0pt}{}{\xi ^1C^1(P(G),Z_2^{add}),}{\xi ^1=\chi }}{}}{\displaystyle \underset{s_i^1P(G)}{}}(\mathrm{tanh}\beta E(s_i^1))^{1/2(1(1)^{\xi ^1(s_i^1)})}.`$ (3.16)
Proof. The definition (3.11) implies
$$\mathrm{exp}\{\beta H(\sigma ^1)\}=\underset{s_i^1P(G)}{}\mathrm{exp}\{\beta E(s_i^1)(1)^{\sigma ^1(s_i^1)}\}$$
(3.17)
where $`\sigma ^1C^1(P(G),๐_2^{add})`$. It is easy to verify that for $`ฯต=0,1`$
$$\mathrm{exp}\{\beta E(s_i^1)(1)^ฯต\}=(\mathrm{cosh}\beta E(s_i^1))\underset{\xi =0,1}{}(1)^{\xi ฯต}(\mathrm{tanh}\beta E(s_i^1))^{1/2(1(1)^\xi )}.$$
(3.18)
The relations (3.17), (3.18) imply
$`\mathrm{exp}\{\beta H(\sigma ^1)\}=\left({\displaystyle \underset{s_i^1P(g)}{}}\mathrm{cosh}\beta E(s_i^1)\right)\times `$
$`{\displaystyle \underset{\xi ^1C^1(P(G),Z_2^{add})}{}}(1)^{\xi ^1,\sigma ^1}{\displaystyle \underset{s_i^1P(g)}{}}(\mathrm{tanh}\beta E(s_i^1))^{1/2(1(1)^{\xi ^1(s_i^1)})}.`$ (3.19)
The substitution of the equality (3) into the definition (3.13), the first relation (3) and the relation
$$\underset{\xi =0,1}{}(1)^{\xi ฯต}=\{\genfrac{}{}{0pt}{}{2,ฯต=0,}{0,ฯต=1,}$$
(3.20)
give the equality (3). The substitution of the equality (3) into the definition (3), the first relation (3) and the relation (3.20) give the equality (3). The proposition is proved.
Here we used the definitions and the methods of the paper . The equality (3) was proved for the first time in the paper . The equality (3) for homogeneous isotropic Ising model was proved in the paper .
The equality (3) implies that the correlation function $`W_G(\chi )`$ is not zero only for the cochains $`\chi B_0(P(G),๐_2^{add})`$. Therefore the cochain $`\chi `$ takes the value $`1`$ only at the ends of the broken lines. Any broken line has two ends. Hence the cochain $`\chi `$ takes the value $`1`$ at even number of vertices.
## 4 Free Boundary Conditions
We denote the one dimensional cell $`s_i^1`$ as the non โ oriented edge $`e_i`$ corresponding with two oppositely oriented edges $`๐_i`$ and $`๐_i^1`$. The interaction energy of Ising model is denoted by $`E(s_i^1)=E(e_i)=E(๐_i)=E(๐_i^1)`$. The equality (3) may be rewritten in the form
$$Z_G=2^{\mathrm{\#}(VG)}\left(\underset{eP(G)}{}\mathrm{cosh}\beta E(e)\right)Z_{r,G}$$
(4.1)
where
$$Z_{r,G}=\underset{\xi ^1Z_1(P(G),Z_2^{add})}{}๐ฎ^{\xi ^1},$$
(4.2)
$$๐ฎ^{\xi ^1}=\underset{eP(G)}{}u(e)^{1/2(1(1)^{\xi ^1(e)})},$$
(4.3)
$$u(e)=u(๐)=u(๐^1)=\mathrm{tanh}\beta E(e).$$
(4.4)
Let a graph $`G`$ be embedded in a rectangular lattice $`๐^{\times 2}`$ on the plane. Let with any pair $`๐_1`$, $`๐_2`$ of the oriented edges of a graph $`G`$ such that $`f(๐_1)=b(๐_2)`$, $`b(๐_1)f(๐_2)`$ there correspond the number
$$\rho (๐_1;๐_2)=\mathrm{exp}\{i/2\widehat{(๐_1,๐_2)}\}$$
(4.5)
where $`\widehat{(๐_1,๐_2)}`$ is the radian measure of the angle between the direction of the oriented edge $`๐_1`$ and the direction of the oriented edge $`๐_2`$. Due to the equalities (2.17) and (4.5) with any reduced closed path $`C`$ on the graph $`G`$ there corresponds the number $`\rho (C)=\mathrm{exp}\{i/2\varphi (C)\}`$ where $`\varphi (C)`$ is the total angle through which the tangent vector of the path $`C`$ turns along the path $`C`$.
For a graph $`G`$ embedded in the rectangular lattice $`๐^{\times 2}`$ on the plane the estimate (2.30) has the following form
$$|u(๐)|=|\mathrm{tanh}\beta E(๐)|<1/3.$$
(4.6)
Theorem 4.1. Let a finite graph $`G`$ be embedded in the rectangular lattice $`๐^{\times 2}`$ on the plane and let the estimate (4.6) be fulfilled. Then for the reduced partition function (4.2)
$$Z_{r,G}=\mathrm{exp}\{1/2\underset{CRC(G)}{}|C|^1๐ฎ^C\rho (C)\}$$
(4.7)
where $`C`$ runs over the set $`RC(G)`$ of all reduced closed paths on a graph $`G`$ and the number $`๐ฎ^C`$ is defined by the relations (2.18), (4.4).
Proof. Due to the paper
$$(Z_{r,G})^2=det(IT(๐ฎ,\rho ))$$
(4.8)
where $`(\mathrm{\#}(๐G))\times (\mathrm{\#}(๐G))`$ โ matrix $`T(๐ฎ,\rho )`$ is defined by the equalities (2.32), (4.4) and (4.5). In view of the definition the angle $`\varphi (C)=2\pi k`$ where $`k`$ is an integer. Hence the number $`\rho (C)=\mathrm{exp}\{i/2\varphi (C)\}`$ is real. Now the equality (4.7) follows from the equalities (2.36) and (4.8). The theorem is proved.
For the homogeneous two dimensional Ising model the interaction energy $`E(๐)`$ does not depend on an initial vertex of an edge $`๐`$. Let us denote $`E_1`$ ($`E_2`$) the interaction energy $`E(๐)`$ for horizotally (vertically) directed edges $`๐`$. The relations (4.1) โ (4.4) imply for the homogeneous Ising model
$$Z_{G(M_1^{},M_2^{};M_1,M_2)}=Z_{G(1,1;M_1M_1^{}+1,M_2M_2^{}+1)}.$$
(4.9)
Theorem 4.2. Let for the interaction energy of the homogeneous two dimensional Ising model the estimate (4.6) be valid. Then for the partition function (4.1) of the homogeneous Ising model on the rectangular lattice on the plane
$`\underset{M_i\mathrm{}i=1,2}{lim}(M_1M_2)^1\mathrm{ln}Z_{G(1,1;M_1,M_2)}=`$
$`\mathrm{ln}(2\mathrm{cosh}\beta E_1\mathrm{cosh}\beta E_2)+1/2(2\pi )^2{\displaystyle _0^{2\pi }}๐\theta _1{\displaystyle _0^{2\pi }}๐\theta _2`$
$`\mathrm{ln}[(1+z_1^2)(1+z_2^2)2z_1(1z_2^2)\mathrm{cos}\theta _12z_2(1z_1^2)\mathrm{cos}\theta _2]`$ (4.10)
where the variables $`z_i=\mathrm{tanh}\beta E_i`$, $`i=1,2`$.
Proof. It follows from the equalities (4.1) and (4.7) for the homogeneous Ising model on the rectangular lattice that
$`\underset{M_i\mathrm{}i=1,2}{lim}(M_1M_2)^1\mathrm{ln}Z_{G(1,1;M_1,M_2)}=`$
$`\mathrm{ln}(2\mathrm{cosh}\beta E_1\mathrm{cosh}\beta E_2)\underset{M_i\mathrm{}i=1,2}{lim}(2M_1M_2)^1{\displaystyle \underset{CRC(G)}{}}|C|^1๐ฎ^C\rho (C)`$ (4.11)
where $`C`$ runs over the set $`RC(G)`$ of all reduced closed paths on the graph $`G(1,1;M_1,M_2)`$.
The total number of all reduced closed paths of the length $`l`$ with the initial vertex $`(0,0)`$ on the lattice $`๐^{\times 2}`$ is less than $`43^{l1}`$. Due to definitions (2.17), (4.5) the number $`|\rho (C)|=1`$. Hence the estimate (4.6) implies the absolute convergence of the series in the right hand side of the equality (4) for finite $`M_1`$,$`M_2`$.
The interaction energy $`E(๐)`$ does not depend on the initial vertex of the oriented edge $`๐`$. Therefore due to definitions (2.17), (4.5), (2.18) and (4.4) the number $`|C|^1๐ฎ^C\rho (C)`$ does not depend on the initial vertex of the path $`C`$. Hence
$$\underset{CRC(G)}{}|C|^1๐ฎ^C\rho (C)=\underset{C_0RC_0(Z^{\times 2})}{}N(C_0)|C_0|^1๐ฎ^{C_0}\rho (C_0)$$
(4.12)
where $`C_0`$ runs over the set $`RC_0(๐^{\times 2})`$ of all reduced closed paths with the starting point at the vertex $`(0,0)`$ on the lattice $`๐^{\times 2}`$. The number $`N(C_0)`$ is the total number of shifted paths $`C_0`$ on the graph $`G(1,1;M_1,M_2)`$. If $`|C_0|M_1`$, $`|C_0|M_2`$, the following estimate is valid
$$(M_1|C_0|)(M_2|C_0|)N(C_0)M_1M_2.$$
(4.13)
The estimate (4.6) implies the absolute convergence of the series
$$\underset{C_0RC_0(Z^{\times 2})}{}|C_0|^k๐ฎ^{C_0}\rho (C_0)$$
(4.14)
for $`k=0,\pm 1`$. Thus it follows from the equalities (4), (4.12) and estimates (4.13) that
$`\underset{M_i\mathrm{}i=1,2}{lim}(M_1M_2)^1\mathrm{ln}Z_{G(1,1;M_1,M_2)}=`$
$`\mathrm{ln}(2\mathrm{cosh}\beta E_1\mathrm{cosh}\beta E_2)1/2{\displaystyle \underset{C_0RC_0(Z^{\times 2})}{}}|C_0|^1๐ฎ^{C_0}\rho (C_0).`$ (4.15)
Let us rewrite the expression (4) in more traditional form. On the rectangular lattice $`\stackrel{~}{G}(0,0;M_1,M_2)`$ on the torus the number $`u(๐)`$ is defined by the equality (4.4) and the number $`\rho (๐_1;๐_2)`$ is defined by the equality (4.5) taking into account the identification of the vertices and edges in the graph $`\stackrel{~}{G}(0,0;M_1,M_2)`$. The $`(4M_1M_2)\times (4M_1M_2)`$ โ matrix $`T(๐ฎ,\rho )`$ is defined by the relation (2.32). The equality (2.36) implies
$$(2M_1M_2)^1\mathrm{ln}[det(IT(๐ฎ,\rho ))]=(2M_1M_2)^1\underset{CRC(\stackrel{~}{G})}{}|C|^1๐ฎ^C\rho (C)$$
(4.16)
where $`C`$ runs over the set $`RC(\stackrel{~}{G})`$ of all reduced closed paths on the graph $`\stackrel{~}{G}(0,0;M_1,M_2)`$.
For the homogeneous Ising model the number $`|C|^1๐ฎ^C\rho (C)`$ does not depend on the starting point of the path $`C`$. The graph $`\stackrel{~}{G}(0,0;M_1,M_2)`$ is invariant under the shifts. Any reduced closed path $`C`$ is a shifted reduced closed path $`C_0`$ where a path $`C_0`$ has the starting point at the vertex $`(0,0)`$. Therefore the equality (4.16) implies
$$(2M_1M_2)^1\mathrm{ln}[det(IT(๐ฎ,\rho ))]=1/2\underset{C_0RC_0(\stackrel{~}{G})}{}|C_0|^1๐ฎ^{C_0}\rho (C_0)$$
(4.17)
where $`C_0`$ runs over the set $`RC_0(\stackrel{~}{G})`$ of all reduced closed paths with the starting point at the vertex $`(0,0)`$ on the graph $`\stackrel{~}{G}(0,0;M_1M_2)`$. When $`M_i\mathrm{}`$, $`i=1,2`$, the terms in the series (4.17) corresponding with the long paths connecting the vertices $`(0,k)`$ and $`(M_1,k)`$ or the $`(k,0)`$ and $`(k,M_2)`$, etc tend to zero due to the estimate (4.6). Hence
$$\underset{M_i\mathrm{}i=1,2}{lim}(2M_1M_2)^1\mathrm{ln}[det(IT(๐ฎ,\rho ))]=1/2\underset{C_0RC_0(Z^{\times 2})}{}|C_0|^1๐ฎ^{C_0}\rho (C_0)$$
(4.18)
where $`C_0`$ runs over the set $`RC_0(๐^{\times 2})`$ of all reduced closed paths with the starting point at the vertex $`(0,0)`$ on the lattice $`๐^{\times 2}`$.
Let us calculate the determinant of the matrix $`IT(๐ฎ,\rho )`$. The vertices of the graph $`\stackrel{~}{G}(0,0;M_1,M_2)`$ are defined by the vectors $`๐ฃ๐^{\times 2}`$, $`1j_1M_1`$, $`1j_2M_2`$. The oriented edge $`๐`$ of the graph $`\stackrel{~}{G}(0,0;M_1,M_2)`$ is defined by its initial vertex $`b(๐)=๐ฃ`$ and by its direction: the unit vector $`๐ฏ`$. The unit vector $`๐ฏ`$ is one of four vectors: $`(\pm 1,0)`$, $`(0,\pm 1)`$. Thus an oriented edge $`๐`$ is a pair $`(๐ฃ,๐ฏ)`$. Due to definitions (2.32), (4.4) and (4.5)
$`T(๐ฎ,\rho )_{(j,v),(j^{},v^{})}=u((๐ฃ,๐ฏ))\delta _{j_1+v_1j_1^{},M_1Z}\delta _{j_2+v_2j_2^{},M_2Z}\times `$
$`(1\delta _{v_1+v_1^{},\mathrm{\hspace{0.17em}0}}\delta _{v_2+v_2^{},\mathrm{\hspace{0.17em}0}})\mathrm{exp}\{i/2\widehat{(๐ฏ,๐ฏ^{})}\}`$ (4.19)
where Kronecker symbol
$$\delta _{j,M_kZ}=M_k^1\underset{l=1}{\overset{M_k}{}}\mathrm{exp}\{i2\pi M_k^1jl\}.$$
(4.20)
The symbol (4.20) equals $`1`$ if $`j=M_kl`$ where $`l`$ is an integer and it equals $`0`$ if $`jM_kl`$.
For the homogeneous Ising model the number $`u((๐ฃ,๐ฏ))`$ does not depend on the vector $`๐ฃ`$. We denote it by u(v). The equalities (4) and (4.20) imply
$$(IT(๐ฎ,\rho ))_{(j,v),(j^{},v^{})}=(CBC^1)_{(j,v),(j^{},v^{})}$$
(4.21)
where the matrices
$$C_{(j,v),(j^{},v^{})}=\delta _{v_1,v_1^{}}\delta _{v_2,v_2^{}}(M_1M_2)^{1/2}\mathrm{exp}\{i2\pi (M_1^1j_1j_1^{}+M_2^1j_2j_2^{})\},$$
(4.22)
$`B_{(j,v),(j^{},v^{})}=\delta _{j_1,j_1^{}}\delta _{j_2,j_2^{}}B(๐ฃ)_{v,v^{}},`$
$`B(๐ฃ)_{v,v^{}}=\delta _{v_1,v_1^{}}\delta _{v_2,v_2^{}}`$
$`\mathrm{exp}\{i2\pi (M_1^1v_1j_1+M_2^1v_2j_2)\}u(๐ฏ)\mathrm{exp}\{i/2\widehat{(๐ฏ,๐ฏ^{})}\}(1\delta _{v_1,v_1^{}}\delta _{v_2,v_2^{}}).`$ (4.23)
The matrix $`B_{(j,v),(j^{},v^{})}`$ is diagonal for the vectors $`๐ฃ`$,$`๐ฃ^{}`$. The second relation (4) defines $`4\times 4`$ โ matrix $`B(๐ฃ)_{v,v^{}}`$ for any vector $`๐ฃ๐^{\times 2}`$, $`1j_1M_1`$, $`1j_2M_2`$. It follows from the relations (4.21) and (4) that
$$det(IT(๐ฎ,\rho ))=\underset{j_1=1}{\overset{M_1}{}}\underset{j_2=1}{\overset{M_2}{}}detB(๐ฃ).$$
(4.24)
Due to relations (4.4) $`u((\pm 1,0))=\mathrm{tanh}\beta E_1=z_1`$, $`u(0,\pm 1)=\mathrm{tanh}\beta E_2=z_2`$. By using the definition (4) it is possible to calculate
$$detB(๐ฃ)=(1+z_1^2)(1+z_2^2)2z_1(1z_2^2)\mathrm{cos}2\pi M_1^1j_12z_2(1z_1^2)\mathrm{cos}2\pi M_2^1j_2.$$
(4.25)
The substitution of the equality (4.25) into the relation (4.24) yields
$`det(IT(๐ฎ,\rho ))=`$
$`{\displaystyle \underset{j_1=1}{\overset{M_1}{}}}{\displaystyle \underset{j_2=1}{\overset{M_2}{}}}[(1+z_1^2)(1+z_2^2)2z_1(1z_2^2)\mathrm{cos}2\pi M_1^1j_12z_2(1z_1^2)\mathrm{cos}2\pi M_2^1j_2].`$ (4.26)
The equalities (4), (4.18) and (4) imply the equality (4). The theorem is proved.
Let a cochain $`\xi ^1C^1(P(G),๐_2^{add})`$. The support $`\xi ^1`$ is the set of all non โ oriented edges of the graph $`G`$ on which a cochain $`\xi ^1`$ takes the value $`1`$. Let a cochain $`\chi C_0(P(G),๐_2^{add})`$. The support $`\xi ^1`$ is called $`\chi `$ โ connected if any connected component of the support $`\xi ^1`$ contains the non โ oriented edges incident to the vertices on which a cochain $`\chi `$ equals $`1`$. Let $`i(\xi ^1)`$ be the set of all non โ oriented edges incident to the vertices incident to the edges of the support $`\xi ^1`$. By using the relations (4.1) โ (4.4) we rewrite the correlation function (3) in the following form
$$W_G(\chi )=(Z_{r,G})^1\underset{\genfrac{}{}{0pt}{}{\xi ^1C^1(P(G),Z_2^{add}),\xi ^1=\chi ,}{\chi connected\xi ^1}}{}๐ฎ^{\xi ^1}Z_{r,Gi(\xi ^1)}$$
(4.27)
where the graph $`Gi(\xi ^1)`$ is obtained by deleting all edges of the set $`i(\xi ^1)`$ from the graph $`G`$.
If $`C`$ is a closed path on the graph $`G`$, the support $`C`$ is the set of all non โ oriented edges corresponding with the oriented edges from a closed path $`C`$.
Theorem 4.3. Let the estimate (4.6) be valid. Let the interaction energy $`E(๐)`$ be non โ negative. Let a cochain $`\chi C^0(P(G),๐_2^{add})`$ be equal to $`1`$ on the finite number of the vertices. Then for the correlation function (3) of the two dimensional Ising model with free boundary conditions
$`\underset{\genfrac{}{}{0pt}{}{M_k\mathrm{},M_k^{}\mathrm{},}{k=1,2}}{lim}W_{G(M_1^{},M_2^{};M_1,M_2)}(\chi )=`$
$`{\displaystyle \underset{\genfrac{}{}{0pt}{}{\xi ^1C^1(P(G),Z_2^{add}),\xi ^1=\chi ,}{\chi connected\xi ^1}}{}}๐ฎ^{\xi ^1}\mathrm{exp}\{1/2{\displaystyle \underset{\genfrac{}{}{0pt}{}{CRC(Z^{\times 2}),}{Ci(\xi ^1)\mathrm{}}}{}}|C|^1๐ฎ^C\rho (C)\}`$ (4.28)
where the number $`๐ฎ^{\xi ^1}`$ is defined by the relations (4.3), (4.4), the number $`๐ฎ^C`$ is defined by the relations (2.18), (4.4) and the number $`\rho (C)`$ is defined by the relations (2.17), (4.5).
Proof. Let us consider the equality (4.27) for the graph $`G=G(M_1^{},M_2^{};M_1,M_2)`$. Theorem 4.1 impiles
$$(Z_{r,G})^1Z_{r,Gi(\xi ^1)}=\mathrm{exp}\{1/2\underset{\genfrac{}{}{0pt}{}{CRC(G),}{Ci(\xi ^1)\mathrm{}}}{}|C|^1๐ฎ^C\rho (C)\}.$$
(4.29)
The total number of all reduced closed paths of the length $`l`$ with fixed initial vertex on the lattice $`๐^{\times 2}`$ is less than $`43^{l1}`$. Due to the estimate (4.6) the series (4.29) is absolutely convergent for $`G๐^{\times 2}`$. Thus every term of the sum (4.27) for the graph $`G=G(M_1^{},M_2^{};M_1,M_2)`$ converges to the term of the series (4) when $`G๐^{\times 2}`$. The equality (4) will be proved if the absolute convergence of the series (4) is proved when the estimate (4.6) is valid.
The correlation function (4.27) is not zero only for the cochains $`\chi B_0(P(G),๐_2^{add})`$ taking the value $`1`$ at the even number of vertices $`๐ฆ_1,\mathrm{},๐ฆ_{2k}`$ of the graph $`G๐^{\times 2}`$. Let a cochain $`\xi ^1C^1(P(G),๐_2^{add})`$ satisfy the condition $`\xi ^1=\chi `$. Then the vertex $`๐ฆ_1`$ is incident to one or three non โ oriented edges on which a cochain $`\xi ^1`$ takes the value $`1`$. We take one such edge. It corresponds with the oriented edge $`(๐ฆ_1,๐ฏ_1)`$. If $`๐ฆ_1+๐ฏ_1=๐ฆ_{j_1}`$, then we have constructed the path connecting the vertices $`๐ฆ_1`$ and $`๐ฆ_{j_1}`$. If $`๐ฆ_1+๐ฏ_1`$ does not coincide with any vertices $`๐ฆ_2,\mathrm{},๐ฆ_{2k}`$, then due to the condition $`\xi ^1=\chi `$ the vertex $`๐ฆ_1+๐ฏ_1`$ is incident to two or four non โ oriented edges on which a cochain $`\xi ^1`$ takes the value $`1`$. One of these non โ oriented edges corresponds to the oriented edge $`(๐ฆ_1,๐ฏ_1)`$. We take another edge. It corresponds with the oriented edge $`(๐ฆ_1+๐ฏ_1,๐ฏ_2)`$, $`๐ฏ_2๐ฏ_1`$. By repeating this process we obtain the path $`P_1=((๐ฆ_1^{},๐ฏ_1),\mathrm{},(๐ฆ_{q_1}^{},๐ฏ_{q_1}))`$ where $`๐ฆ_1^{}=๐ฆ_1`$, $`๐ฆ_{i+1}^{}=๐ฆ_i^{}+๐ฏ_i`$, $`i=1,\mathrm{},q_11`$, $`๐ฆ_{q_1}^{}+๐ฏ_{q_1}=๐ฆ_{j_1}`$ and $`๐ฆ_{j_1}`$ is one of the vertices $`๐ฆ_2,\mathrm{},๐ฆ_{2k}`$ on which the cochain $`\chi `$ takes the value $`1`$. Any non โ oriented edge may correspond with only one oriented edge from the path $`P_1`$. Let the cochain $`\xi ^1[P_1]C^1(P(G),๐_2^{add})`$ equal $`1`$ on all non โ oriented edges corresponding with the oriented edges from the path $`P_1`$. It equals $`0`$ on all other non โ oriented edges from the graph $`G`$. By construction $`\xi ^1=\xi ^1[P_1]+\eta ^1`$ where the supports of the cochains $`\xi ^1[P_1],\eta ^1C^1(P(G),๐_2^{add})`$ do not intersect each other. By repeating this process we construct the paths $`P_1,\mathrm{},P_k`$ connecting the vertices $`๐ฆ_{i_1},\mathrm{},๐ฆ_{i_k}`$, $`1=i_1<i_2<\mathrm{}<i_k`$ with the vertices $`๐ฆ_{j_1},\mathrm{},๐ฆ_{j_k}`$, $`i_l<j_l`$, $`l=1,\mathrm{},k`$. Any non โ oriented edge may correspond with only one oriented edge from the paths $`P_l`$, $`l=1,\mathrm{},k`$. These paths correspond with the cochains $`\xi ^1[P_l]`$, $`l=1,\mathrm{},k`$, such that $`\xi ^1=\xi ^1[P_1]+\mathrm{}+\xi ^1[P_k]+\eta ^1`$ where the supports of the cochains $`\xi ^1[P_1],\mathrm{},\xi ^1[P_k],\eta ^1C^1(P(G),๐_2^{add})`$ do not intersect each other and $`\eta ^1Z_1(P(G),๐_2^{add})`$. This decomposition is not unique in general. Therefore not an equality but the estimate is valid
$$W_G(\chi )(Z_{r,G})^1\underset{\genfrac{}{}{0pt}{}{\{i_l,j_l\},}{l=1,\mathrm{},k}}{}\underset{\genfrac{}{}{0pt}{}{P_l,}{l=1,\mathrm{},k}}{}\left(\underset{l=1}{\overset{k}{}}๐ฎ^{P_l}\right)Z_{r,G(_{l=1}^kP_l)}$$
(4.30)
where $`\{i_l,j_l\}`$ runs over the set of the subdivisions of the numbers $`1,\mathrm{},2k`$ into $`k`$ pairs: $`1=i_1<\mathrm{}<i_k`$, $`i_l<j_l`$ , $`l=1,\mathrm{},k`$, the paths $`P_l`$, $`l=1,\mathrm{},k`$, run over the set of the paths connecting the vertices $`๐ฆ_{i_l}`$ and $`๐ฆ_{j_l}`$, $`l=1,\mathrm{},k`$, any non โ oriented edge may correspond with only one oriented edge from the paths $`P_l`$, $`l=1,\mathrm{},k`$ and the graph $`G(_{l=1}^kP_l)`$ is obtained from the graph $`G`$ by deleting all edges from the supports $`P_l`$, $`l=1,\mathrm{},k`$. Due to the definition (4.4) the variable $`u(๐)`$ is non โ negative when the interaction energy $`E(๐)`$ is non โ negative. For the non โ negative variables $`u(๐)`$ the definition (4.2) implies the estimate
$$(Z_{r,G})^1Z_{r,G(_{l=1}^kP_l)}1.$$
(4.31)
It follows from the estimates (4.30), (4.31) that
$$W_G(\chi )\underset{\genfrac{}{}{0pt}{}{\{i_l,j_l\},}{l=1,\mathrm{},k}}{}\underset{\genfrac{}{}{0pt}{}{P_l,}{l=1,\mathrm{},k}}{}\underset{l=1}{\overset{k}{}}๐ฎ^{P_l}.$$
(4.32)
The total number of the reduced closed paths of the length $`l`$ starting at the fixed vertex on the lattice $`๐^{\times 2}`$ is less than $`43^{l1}`$. Hence the estimate (4.6) implies that for $`G๐^{\times 2}`$ the sum (4.32) converges to the absolutely convergent series. This series majorizes the series (4). Therefore the series (4) is absolutely convergent when the estimate (4.6) is valid. The theorem is proved.
Let a cycle $`\xi ^1Z_1(P(G),๐_2^{add})`$. By using the arguments of Theorem 4.3 we can construct the closed paths $`C_1`$,โฆ, $`C_m`$ where the supports $`C_1`$,โฆ,$`C_m`$ do not intersect each other and any non โ oriented edge may correspond with only one oriented edge from the closed paths $`C_1`$,โฆ,$`C_m`$. Any closed path on a rectangular lattice on the plane has an even number of the vertically directed edges and it has an even number of the horizontally directed edges. Let the interaction energy $`E(๐)`$ be non โ negative for the vertically directed edges $`๐`$ and let it be non โ positive for the horizontally directed edges $`๐`$. Due to the definitions (4.3), (4.4) $`๐ฎ^{\xi ^1}0`$. Hence the inequality (4.31) is fulfilled in this case. Therefore it is possible to prove the absolute convergence of the series (4). Thus Theorem 4.3 is valid when the interaction energy is non โ negative for the vertically directed edges and it is non โ positive for the horizontally directed edges. Theorem 4.3 is valid also for the case when the interaction energy is non โ positive for the vertically directed edges and it is non โ negative for the horizontally directed edges. If the interaction energy is non โ positive for all oriented edges, then Theorem 4.3 is also valid.
## 5 Periodic Boundary Conditions
Let us consider the rectagular lattice $`\stackrel{~}{G}(M_1^{},M_2^{};M_1,M_2)`$ on the torus introduced in Section 3. The number $`\rho (๐_1;๐_2)`$ is given by the relation (4.5) taking into account the identification of the vertices and the edges in the graph $`\stackrel{~}{G}(M_1^{},M_2^{};M_1,M_2)`$. With every reduced closed path $`C`$ on the graph $`\stackrel{~}{G}(M_1^{},M_2^{};M_1,M_2)`$ there corresponds the number $`\rho (C)=\mathrm{exp}\{i/2\varphi (C)\}`$ defined by the equality (2.17). Here $`\varphi (C)`$ is the total angle through which the tangent vector of the path $`C`$ turns along the path $`C`$.
If the reduced closed path $`C`$ lies on the rectangular lattice $`๐^{\times 2}`$ and the number $`\rho (C)`$ is defined by the relations (2.17), (4.5), then due to the number $`\rho (C)=(1)^{n(C)}`$ where $`n(C)`$ is the total number of the transversal self โ intersections of the path $`C`$. The papers , , , used this Whitney formula. Let us consider the line $`C`$ connecting the vertices $`(M_1^{},k)`$ and $`(M_1,k)`$ on the graph $`\stackrel{~}{G}(M_1^{},M_2^{};M_1,M_2)`$. The line has no self โ intersections. It is easy to see that $`\rho (C)=1`$. Thus Whitney formula is wrong for a torus in general. Therefore we can not use the results of the papers , , , for a torus.
Let us study the properties of the $`L`$ โ function (2.31) for a graph $`G`$ lying on the graph $`\stackrel{~}{G}(M_1^{},M_2^{};M_1,M_2)`$. Let for the reduced closed path $`C=(๐_1,\mathrm{},๐_p,๐_{p+1},\mathrm{},๐_{p+q})`$ the vertices $`b(๐_{p+1})=b(๐_1)`$. Then $`C=C_1C_2`$ where the closed paths $`C_1=(๐_1,\mathrm{},๐_p)`$ and $`C_2=(๐_{p+1},\mathrm{},๐_{p+q})`$ may be not reduced. Indeed, if $`๐_1=๐_p^1`$, then the closed path $`C_1`$ is not reduced. If $`๐_{p+1}=๐_{p+q}^1`$, then the closed path $`C_2`$ is not reduced. By the definition a reduced closed path does not contain the oppositely oriented edges $`๐`$, $`๐^1`$ if they are subsequent or if they are the first and the last edges of the path. A closed path is called completely reduced if it does not contained the oppositely oriented edges $`๐`$, $`๐^1`$ at any places. The multipliers $`C_1`$ and $`C_2`$ of any completely reduced closed path $`C=C_1C_2`$ are also completely reduced closed paths. The set of all completely reduced closed paths on the graph $`G`$ is denoted by $`CRC(G)`$.
Theorem 5.1. Let a graph $`G`$ be embedded in the rectangular lattice $`\stackrel{~}{G}(M_1^{},M_2^{};M_1,M_2)`$ on the torus. Let with any reduced closed path $`C`$ on the graph $`G`$ there correspond the number $`\rho (C)`$ given by the relations (2.17), (4.5) and the number $`๐ฎ^C`$ given by the relations (2.18), (4.4). If the estimate (4.6) is valid, then
$$\underset{CRC(G)}{}|C|^1๐ฎ^C\rho (C)=\underset{CCRC(G)}{}|C|^1๐ฎ^C\rho (C).$$
(5.1)
In the left hand side of the equality (5.1) the sum extends over the set $`RC(G)`$ of all the reduced closed paths on the graph $`G`$ and in the right hand side of the equality (5.1) the sum extends over the set $`CRC(G)`$ of all the completely reduced closed paths on the graph $`G`$.
Proof. Let the reduced closed path $`C=(๐_1,\mathrm{},๐_p,๐,๐_{p+1},\mathrm{},๐_{p+q},๐^1,๐_{p+q+1},\mathrm{},๐_{p+q+r})`$ contain the oppositely oriented edges $`๐`$ and $`๐^1`$. Then the closed path
$`C^{}=(๐_1,\mathrm{},๐_p,๐,๐_{p+q}^1,\mathrm{},๐_{p+1}^1,๐^1,๐_{p+q+1},\mathrm{},๐_{p+q+r})`$ is also reduced.
The path length definition and the definitions (2.18), (4.4) imply
$$|C|=|C^{}|,๐ฎ^C=๐ฎ^C^{}.$$
(5.2)
By using the definitions (2.17), (4.5) we get
$$\varphi (C)=\varphi _1+\varphi _2+\varphi _3,$$
(5.3)
$$\varphi _1=\underset{i=1}{\overset{p1}{}}\widehat{(๐_i,๐_{i+1})}+\widehat{(๐_p,๐)},$$
(5.4)
$$\varphi _2=\widehat{(๐,๐_{p+1})}+\underset{i=p+1}{\overset{p+q1}{}}\widehat{(๐_i,๐_{i+1})}+\widehat{(๐_{p+q},๐^1)},$$
(5.5)
$$\varphi _3=\widehat{(๐^1,๐_{p+q+1})}+\underset{i=p+q+1}{\overset{p+q+r1}{}}\widehat{(๐_i,๐_{i+1})}+\widehat{(๐_{p+q+r},๐_1)}.$$
(5.6)
It is easy to verify the relation
$$\widehat{(๐_1,๐_2)}=\widehat{(๐_2^1,๐_1^1)}$$
(5.7)
for the oriented edges $`๐_1`$, $`๐_2`$ such that $`f(๐_1)=b(๐_2)`$, $`b(๐_1)f(๐_2)`$. The definitions (2.17), (4.5) and the relations (5.7) imply
$$\varphi (C^{})=\varphi _1\varphi _2+\varphi _3.$$
(5.8)
Since the directions of the oriented edges $`๐`$ and $`๐^1`$ are opposite, due to the definition (5.5) $`\varphi _2=(2k+1)\pi `$ where $`k`$ is an integer. Hence $`\mathrm{exp}\{i/2\varphi _2\}=\mathrm{exp}\{i/2\varphi _2\}`$ and in view of the relations (5.3), (5.8)
$$\mathrm{exp}\{i/2\varphi (C)\}=\mathrm{exp}\{i/2\varphi (C^{})\}.$$
(5.9)
Due to the relations (5.2), (5.9) all terms $`|C|^1๐ฎ^C\rho (C)`$ in the left hand side sum (5.1) corresponding with the reduced closed paths $`C`$ containing the oppositely oriented edges $`๐`$ and $`๐^1`$ cancel each other. The theorem is proved.
Theorem 5.1 is valid also for any graph $`G`$ embedded in a rectangular lattice on the plane. Hence it is possible to change the summing over the reduced closed paths on the lattice $`๐^{\times 2}`$ in the equality (4) for the summing over the completely reduced closed paths on the lattice $`๐^{\times 2}`$.
The definitions (2.17) and (4.5) imply
$$\rho ((๐_k,๐_1,\mathrm{},๐_{k1}))=\rho ((๐_1,\mathrm{},๐_k)).$$
(5.10)
The number $`\rho ((๐_1,\mathrm{},๐_k))=\mathrm{exp}\{i/2\varphi ((๐_1,\mathrm{},๐_k))\}`$ where $`\varphi ((๐_1,\mathrm{},๐_k))`$ is the total angle through which the tangent vector of the path $`(๐_1,\mathrm{},๐_k)`$ turns along the path $`(๐_1,\mathrm{},๐_k)`$. Therefore $`\varphi ((๐_1,\mathrm{},๐_k))=2\pi m`$ where $`m`$ is an integer. Hence $`(\rho ((๐_1,\mathrm{},๐_k)))^1=`$
$`\rho ((๐_1,\mathrm{},๐_k))`$. The definitions (2.17), (4.5) and the relations (5.7) imply
$$\rho ((๐_k^1,๐_{k1}^1,\mathrm{},๐_1^1))=\rho ((๐_1,\mathrm{},๐_k)).$$
(5.11)
Due to the definitions (2.18), (4.4)
$$๐ฎ^{(e_k^1,e_{k1}^1,\mathrm{},e_1^1)}=๐ฎ^{(e_1,\mathrm{},e_k)}$$
(5.12)
for the reduced closed paths $`(๐_k^1,๐_{k1}^1,\mathrm{},๐_1^1)`$ and $`(๐_1,\mathrm{},๐_k)`$. It follows from the equalities (2.24), (5.10) that the numbers $`|C|^1๐ฎ^C\rho (C)`$ are equal to each other for $`k`$ reduced closed paths: $`(๐_1,\mathrm{},๐_k)`$, $`(๐_k,๐_1,\mathrm{},๐_{k1})`$,โฆ, $`(๐_2,\mathrm{},๐_k,๐_1)`$ which form the oriented reduced cycle $`[(๐_1,\mathrm{},๐_k)]`$. If the closed path $`(๐_1,\mathrm{},๐_k)`$ is completely reduced, the oriented cycle $`[(๐_1,\mathrm{},๐_k)]`$ is called completely reduced. The equalities (5.11), (5.12) imply that the numbers $`|C|^1๐ฎ^C\rho (C)`$ are equal to each other for two oriented reduced cycles $`[(๐_1,\mathrm{},๐_k)]`$ and $`[(๐_k^1,๐_{k1}^1,\mathrm{},๐_1^1)]`$. This pair of the oriented reduced cycles $`[(๐_1,\mathrm{},๐_k)]`$ and
$`[(๐_k^1,๐_{k1}^1,\mathrm{},๐_1^1)]`$ is called the non โ oriented reduced cycle $`[[(๐_1,\mathrm{},๐_k)]]`$. If the closed path $`(๐_1,\mathrm{},๐_k)`$ is completely reduced, the non โ oriented cycle $`[[(๐_1,\mathrm{},๐_k)]]`$ is called completely reduced. In other words by a non โ oriented completely reduced cycle is meant a definite sequence of oriented edges. There are no the oppositely oriented edges $`๐`$, $`๐^1`$ in this sequence. Each succeding edge starts at the vertex where the previous edge ended. The last edge must end at the vertex from which the first edge started. The direction in which the sequence of edges is traversed, and also the particular starting point are both immaterial. By a primitive non โ oriented completely reduced cycle is meant one which can not be constructed by exactly repeating some subpath of $`(๐_1,\mathrm{},๐_k)`$ two or more times.
A completely reduced closed path does not contain the oppositely oriented edges $`๐`$, $`๐^1`$. But it may contain the oriented edge $`๐`$ many times. Due to Lemma 2.1 such path is homotopic to the path $`C=(๐,๐_1,\mathrm{},๐_k,๐,๐_{k+1},\mathrm{},๐_{k+l})=C_1C_2`$ where the closed paths $`C_1=(๐,๐_1,\mathrm{},๐_k)`$ and $`C_2=(๐,๐_{k+1},\mathrm{},๐_{k+l})`$ are completely reduced. In view of the equality (2.15) $`\rho (C)=\rho (C_1)\rho (C_2)`$. A prime closed path is meant a completely reduced closed path which contains any of its oriented edge only once. A non โ oriented completely reduced cycle is called a prime non โ oriented cycle, if every its representative is a prime closed path. The prime non โ oriented cycles $`[[C_1]]`$,โฆ,$`[[C_k]]`$ are called disjoint if any its representatives $`C_1`$,โฆ,$`C_k`$ have no common oriented edges.
Theorem 5.2. Let a graph $`G`$ be embedded in the rectangular lattice $`\stackrel{~}{G}(M_1^{},M_2^{};M_1,M_2)`$ on the torus. Let with any reduced closed path $`C`$ on a graph $`G`$ there correspond the number $`\rho (C)`$ given by the relations (2.17), (4.5) and the number $`๐ฎ^C`$ given by the relations (2.18), (4.4). If the estimate (4.6) is valid, then
$$\mathrm{exp}\{1/2\underset{CRC(G)}{}|C|^1๐ฎ^C\rho (C)\}=1+\underset{k=1}{\overset{\mathrm{}}{}}\underset{\genfrac{}{}{0pt}{}{[[C_i]],i=1,\mathrm{},k:}{prime,disjoint}}{}(1)^k\left(\underset{i=1}{\overset{k}{}}๐ฎ^{C_i}\right)\left(\underset{i=1}{\overset{k}{}}\rho (C_i)\right)$$
(5.13)
where $`C`$ runs over the set $`RC(G)`$ of reduced closed paths on the graph $`G`$, $`[[C_i]]`$, $`i=1,\mathrm{},k`$, run over the set of prime non โ oriented cycles and the prime non โ oriented cycles $`[[C_1]]`$,โฆ,$`[[C_k]]`$ are disjoint.
Proof. Due to Theorem 5.1 it is possible to change the summing over the reduced closed paths on the graph $`G`$ in the left hand side of the equality (5.13) for the summing over the completely reduced closed paths on the graph $`G`$.
Every completely reduced closed path has the form $`C^{\times k}`$ where $`k`$ is an integer and $`C`$ is a primitive completely reduced closed path. By using the equality (2) we get
$$\mathrm{exp}\{1/2\underset{CRC(G)}{}|C|^1๐ฎ^C\rho (C)\}=\mathrm{exp}\{1/2\underset{\genfrac{}{}{0pt}{}{CCRC(G),}{primitive}}{}|C|^1\mathrm{ln}(1๐ฎ^C\rho (C))\}.$$
(5.14)
Let a closed path $`(๐_1,\mathrm{},๐_k)`$ be a primitive completely reduced one. Then $`k`$ primitive completely reduced closed paths $`(๐_1,\mathrm{},๐_k)`$, $`(๐_2,\mathrm{},๐_k,๐_1)`$,โฆ, $`(๐_k,๐_1,\mathrm{},๐_{k1})`$ are different. They form a primitive oriented completely reduced cycle $`[(๐_1,\mathrm{},๐_k)]`$. Due to the equalities (2.24), (5.10) the numbers are equal to each other for all representatives of this primitive oriented completely reduced cycle $`[(๐_1,\mathrm{},๐_k)]`$. Hence
$$\mathrm{exp}\{1/2\underset{CRC(G)}{}|C|^1๐ฎ^C\rho (C)\}=\mathrm{exp}\{1/2\underset{\genfrac{}{}{0pt}{}{[C]:CCRC(G),}{primitive}}{}\mathrm{ln}(1๐ฎ^C\rho (C))\}.$$
(5.15)
Due to the relations (5.11), (5.12) the numbers $`๐ฎ^C\rho (C)`$ are equal to each other for all representatives of the primitive non โ oriented completely reduced cycle $`[[(๐_1,\mathrm{},๐_k)]]`$. We change the summing over the set of primitive oriented completely reduced cycle in the right hand side of the equality (5.15) for the summing over the set of primitive non โ oriented completely reduced cycles. Thus the right hand side of the equality (5.15) is the product of the multipliers $`(1๐ฎ^C\rho (C))`$. The decomposition of this product into the series gives
$`\mathrm{exp}\{1/2{\displaystyle \underset{CRC(G)}{}}|C|^1๐ฎ^C\rho (C)\}=`$
$`1+{\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}{\displaystyle \underset{\genfrac{}{}{0pt}{}{[[C_i]],C_iCRC(G),i=1,\mathrm{},k:}{primitive,different}}{}}(1)^k\left({\displaystyle \underset{i=1}{\overset{k}{}}}๐ฎ^{C_i}\right)\left({\displaystyle \underset{i=1}{\overset{k}{}}}\rho (C_i)\right)`$ (5.16)
where $`[[C_i]]`$, $`i=1,\mathrm{},k`$, run over the set of primitive non โ oriented completely reduced cycles and the primitive non โ oriented completely reduced cycles $`[[C_1]]`$,โฆ,$`[[C_k]]`$ differ from each other.
Let us choose an oriented edge $`๐`$. Any completely reduced closed path does not contain the oppositely oriented edges $`๐`$ and $`๐^1`$ simultaneously. We choose the representatives $`C_i`$ of the non โ oriented completely reduced cycles $`[[C_i]]`$ in the right hand side of the equality (5) such that the paths $`C_i`$ do not contain the oriented edge $`๐^1`$. Thus the sum of all terms in the series (5) which contains $`u(๐)^n`$ is proportional to the sum
$$\underset{k=1}{\overset{n}{}}\underset{\genfrac{}{}{0pt}{}{[C_i],C_iCRC(G),i=1,\mathrm{},k:}{primitive,different}}{}(1)^k\left(\underset{i=1}{\overset{k}{}}๐ฎ^{C_i}\right)\left(\underset{i=1}{\overset{k}{}}\rho (C_i)\right)$$
(5.17)
where the primitive completely reduced closed path $`C_i`$ contains (it may be many times) the oriented edge $`๐`$. The different paths $`C_1`$,โฆ,$`C_k`$ all together contain the oriented edge $`๐`$ exactly $`n`$ times. Due to Lemma 2.1 we can choose such representatives $`C_1`$,โฆ,$`C_k`$ of the primitive oriented completely reduced cycles $`[C_1]`$,โฆ,$`[C_k]`$ that all paths start with the oriented edge $`๐`$. Every such path $`C_j`$ is the product $`C_{j1}^{}\mathrm{}C_{jq_j}^{}`$ where the completely reduced closed path $`C_{jl}^{}`$ starts with the oriented edge $`๐`$ and contains it exactly one time. All paths $`C_{jl}^{}`$ are primitive. Since the paths $`C_1`$,โฆ,$`C_k`$ all together contain the oriented edge $`๐`$ exactly $`n`$ times, they are decomposed into $`n`$ paths $`C_{jl}^{}`$. We change the numeration so that $`C_{jl}^{}=C_{i_{jl}}^{}`$ where the numbers $`(i_{j1},\mathrm{},i_{jq_j})`$, $`j=1,\mathrm{},k`$, give the subdivision of the numbers $`1,\mathrm{},n`$ into $`k`$ groups. Due to definitions (2.17), (4.5) the number $`\rho (C_j)=\rho (C_{i_{j1}}^{})\mathrm{}\rho (C_{i_{jq_j}}^{})`$. The definitions (2.18), (4.4) imply that $`๐ฎ^{C_j}=๐ฎ^{C_{i_{j1}}^{}}\mathrm{}๐ฎ^{C_{i_{jq_j}}^{}}`$. Thus the sum (5.17) has the following form
$$\underset{\genfrac{}{}{0pt}{}{[C_i^{}],C_i^{}CRC(G),i=1,\mathrm{},n:}{primitive}}{}\left(\underset{i=1}{\overset{n}{}}๐ฎ^{C_i^{}}\right)\left(\underset{i=1}{\overset{n}{}}\rho (C_i^{})\right)\underset{k=1}{\overset{n}{}}\underset{\genfrac{}{}{0pt}{}{[C_{i_{j1}}^{}\mathrm{}C_{i_{jq_j}}^{}],j=1,\mathrm{},k:}{primitive,different}}{}(1)^k$$
(5.18)
where the primitive completely reduced closed paths $`C_1^{}`$, โฆ,$`C_n^{}`$ start with the oriented edge $`๐`$ and contain it exactly one time. The numbers $`(i_{j1},\mathrm{},i_{jq_j})`$, $`j=1,\mathrm{},k`$, give a subdivision of the numbers $`1,\mathrm{},n`$ into $`k`$ groups.
If the sum (5.18) is equal to zero for $`n>1`$, then the left hand sides of the equalities (5.13) and (5) coincide and the theorem is proved.
Let us consider first the case when all completely reduced closed paths $`C_1^{}`$,โฆ,$`C_n^{}`$ are different. With every oriented completely reduced cycle $`[C_{i_{j1}}^{}\mathrm{}C_{i_{jq_j}}^{}]`$ there corresponds the permutation $`\pi [i_{j1},\mathrm{},i_{jq_j}]:i_{jl}i_{j,l+1}`$, $`l=1,\mathrm{},q_j1`$, $`i_{jq_j}i_{j1}`$. All the other numbers from $`1,\mathrm{},n`$ the permutation $`\pi [i_{j1},\mathrm{},i_{jq_j}]`$ leaves invariant. Since all completely reduced closed paths $`C_1^{}`$, โฆ,$`C_n^{}`$ are different, any permutation of the multipliers in the product $`C_{i_{j1}}^{}\mathrm{}C_{i_{jq_j}}^{}`$ gives another completely reduced closed path. The permutation $`\pi [i_{j1},\mathrm{},i_{jq_j}]`$ gives the completely reduced closed path $`C_{i_{j2}}^{}\mathrm{}C_{i_{jq_j}}^{}C_{i_{j1}}^{}`$. Due to Lemma 2.1 it is homotopic to the completely reduced closed path $`C_{i_{j1}}^{}\mathrm{}C_{i_{jq_j}}^{}`$. But the permutations $`\pi [i_{j2},\mathrm{},i_{jq_j},i_{j1}]`$ and $`\pi [i_{j1},\mathrm{},i_{jq_j}]`$ coincide. Hence the correspondence between the oriented completely reduced cycles $`[C_{i_{j1}}^{}\mathrm{}C_{i_{jq_j}}^{}]`$ and the permutations $`\pi [i_{j1},\mathrm{},i_{jq_j}]`$ is one โ to โ one. If the groups of numbers $`i_{j1}`$,โฆ,$`i_{jq_j}`$ and $`i_{j^{}1}`$,โฆ, $`i_{j^{}q_j^{}}`$ do not intersect, then the permutations $`\pi [i_{j1},\mathrm{},i_{jq_j}]`$ and $`\pi [i_{j^{}1},\mathrm{},i_{j^{}q_j^{}}]`$ commute with each other. Thus any subdivision of the numbers $`1,\mathrm{},n`$ into $`k`$ non โ intersecting groups $`i_{j1},\mathrm{},i_{jq_j}`$, $`j=1,\mathrm{},k`$ corresponds with the set of $`k`$ oriented completely reduced cycles $`[C_{i_{j1}}^{}\mathrm{}C_{i_{jq_j}}^{}]`$, $`j=1,\mathrm{},k`$, and with the permutation $`\pi [i_{11},\mathrm{},i_{1q_1}]\mathrm{}\pi [i_{k1},\mathrm{},i_{kq_k}]`$ of the numbers $`1,\mathrm{},n`$. Conversely, any permutation $`\pi `$ from the permutation group $`S_n`$ of the numbers $`1,\mathrm{},n`$ corresponds with a subdivision of the numbers $`1,\mathrm{},n`$ into the systems of transitivity of the permutation $`\pi `$: $`(i_{j1},\pi (i_{j1}),\mathrm{},\pi ^{q_j1}(i_{j1}))`$ where $`\pi ^{q_j}(i_{j1})=i_{j1}`$. The total number of these systems of transitivity is denoted by $`t(\pi )`$. Any system of transitivity of the permutation $`\pi `$ corresponds with an oriented completely reduced cycle $`[C_{i_{j1}}^{}C_{\pi (i_{j1})}^{}\mathrm{}C_{\pi ^{q_j1}(i_{j1})}^{}]`$. Therefore for the different completely reduced closed paths $`C_1^{}`$,โฆ,$`C_n^{}`$ the following relation is valid
$$\underset{k=1}{\overset{n}{}}\underset{[C_{i_{j1}}^{}\mathrm{}C_{i_{jq_j}}^{}],j=1,\mathrm{},k}{}(1)^k=\underset{\pi S_n}{}(1)^{t(\pi )}.$$
(5.19)
Since all completely reduced closed paths $`C_1^{}`$,โฆ, $`C_n^{}`$ are different, all oriented completely reduced cycles $`[C_{i_{j1}}^{}\mathrm{}C_{i_{jq_j}}^{}]`$, $`j=1,\mathrm{},k`$, are primitive and different.
Let us define $`n\times n`$ โ matrix $`A`$ whose matrix elements $`A_{ij}=1`$. We calculate the determinant of this matrix
$$detA=\underset{\pi S_n}{}(1)^{\sigma (\pi )}\underset{i=1}{\overset{t(\pi )}{}}A_{p_i\pi (p_i)}A_{\pi (p_i)\pi ^2(p_i)}\mathrm{}A_{\pi ^{q_i1}(p_i)p_i}$$
(5.20)
where $`\pi ^{q_i}(p_i)=p_i`$ and the numbers $`(p_i,\pi (p_i),\mathrm{},\pi ^{q_i1}(p_i))`$, $`i=1,\mathrm{},t(\pi )`$, give the subdivision of the numbers $`1,\mathrm{},n`$ into $`t(\pi )`$ groups. The parity of the permutation $`\pi `$ is equal to
$$\sigma (\pi )=\underset{i=1}{\overset{t(\pi )}{}}(q_i1)\text{mod}\mathrm{\hspace{0.17em}2}.$$
(5.21)
The substitution of the relation (5.21) and of the matrix elements $`A_{ij}=1`$ into the relation (5.20) gives
$$detA=\underset{\pi S_n}{}(1)^{t(\pi )}.$$
(5.22)
$`detA=0`$ for $`n>1`$. Hence the sum (5.19) equals zero for $`n>1`$.
If all completely reduced closed paths $`C_1^{}`$,โฆ,$`C_n^{}`$ coincide, then the completely reduced closed path $`C_{i_{j1}}^{}\mathrm{}C_{i_{jq_j}}^{}`$ is non โ primitive for $`q_j>1`$. If all $`q_j=1`$, the completely reduced closed paths $`C_1^{}`$,โฆ,$`C_n^{}`$ are not different. Thus for this case the sum (5.18) does not give the contribution into the sum (5).
Let us consider the case when there are $`m`$ groups in which $`n_i>1`$, $`i=1,\mathrm{},m`$, the completely reduced closed paths $`C_j^{}`$ coincide with each other. We prove the following equality
$`{\displaystyle \underset{k=1}{\overset{n}{}}}{\displaystyle \underset{\genfrac{}{}{0pt}{}{[C_{i_{j1}}^{}\mathrm{}C_{i_{jq_j}}^{}],j=1,\mathrm{},k:}{primitive,different}}{}}(1)^k=`$
$`\left({\displaystyle \underset{i=1}{\overset{m}{}}}(n_i)!\right)^1\left[{\displaystyle \underset{\pi S_n}{}}(1)^{t(\pi )}{\displaystyle \underset{l>1,q1}{\overset{}{}}}{\displaystyle \underset{\pi S_n}{\overset{}{}}}(1)^{t(\pi )lq}{\displaystyle \underset{\tau S_l}{}}(1)^{t(\tau )}\right].`$ (5.23)
In the right hand side of the equality (5) the second sum extends over certain numbers $`l,q`$ and over certain permutations $`\pi S_n`$.
Now the right hand side sum (5.19) contains the permutations corresponding with the non โ primitive and non โ different oriented completely reduced cycles. Let the completely reduced closed paths $`C_{i_{1j}}^{}`$,$`C_{i_{2j}}^{}`$,โฆ,$`C_{i_{lj}}^{}`$, $`j=1,\mathrm{},q`$, coincide with each other. Let $`\tau S_l`$ be a permutation of the numbers $`1,\mathrm{},l`$. The groups of the numbers $`(p_j,\tau (p_j),\mathrm{},\tau ^{d_j1}(p_j))`$, $`\tau ^{d_j}(p_j)=p_j`$, $`j=1,\mathrm{},t(\tau )`$, give the subdivision of the numbers $`1,\mathrm{},l`$ into the systems of transitivity of the permutation $`\tau `$. We consider the permutations
$`\pi \{\tau ,j\}\pi [i_{p_j1},\mathrm{},i_{p_jq},i_{\tau (p_j)1},\mathrm{},i_{\tau (p_j)q},i_{\tau ^{d_j1}(p_j)1},\mathrm{},i_{\tau ^{d_j1}(p_j)q}]`$ of the numbers $`1,\mathrm{},n`$ where $`j=1,\mathrm{},t(\tau )`$. The permutations $`\pi \{\tau ,j_1\}`$ and $`\pi \{\tau ,j_2\}`$ commute with each other. Let $`\pi \{\tau \}=\pi \{\tau ,1\}\mathrm{}\pi \{\tau ,t(\tau )\}`$. The permutation leaves invariant $`nql`$ numbers from $`1,\mathrm{},n`$. By the construction $`t(\pi \{\tau \})=t(\tau )+nql`$. Let a permutation $`\pi ^{}S_n`$ leave invariant the numbers $`i_{1j}`$,โฆ,$`i_{lj}`$, $`j=1,\mathrm{},q`$. Then
$$t(\pi ^{}\pi \{\tau \})=t(\pi ^{})lq+t(\tau ).$$
(5.24)
The permutation $`\pi ^{}\pi \{\tau \}S_n`$ corresponds with $`t(\pi ^{})lq+t(\tau )`$ oriented completely reduced cycles. From these cycles $`t(\tau )`$ oriented completely reduced cycles $`[C_{i_{p_j1}}^{}\mathrm{}C_{i_{p_jq}}^{}C_{i_{\tau (p_j)1}}^{}\mathrm{}C_{i_{\tau (p_j)q}}^{}\mathrm{}C_{i_{\tau ^{d_j1}(p_j)1}}^{}\mathrm{}C_{i_{\tau ^{d_j1}(p_j)q}}^{}]`$, $`j=1,\mathrm{},t(\tau )`$, are non โ primitive for $`d_j>1`$. If all numbers $`d_j=1`$ and $`t(\tau )=l`$, then these $`t(\tau )`$ oriented completely reduced cycles are non โ different. Hence the permutation $`\pi ^{}\pi \{\tau \}`$ must be subtracted from the right hand side sum (5.19). Now we explain the multiplier $`(_{i=1}^m(n_i)!)^1`$ in the right hand side of the equality (5).
Let a permutation $`\lambda S_n`$ rearrange only the numbers corresponding with the coinciding completely reduced closed paths $`C_1^{}`$,โฆ,$`C_n^{}`$. For this permutation $`C_{\lambda (i_{j1})}^{}\mathrm{}C_{\lambda (i_{jq_j})}^{}=C_{i_{j1}}^{}\mathrm{}C_{i_{jq_j}}^{}`$. But the permutations $`\pi [\lambda (i_{j1}),\mathrm{},\lambda (i_{jq_j})]`$ and $`\pi [i_{j1},\mathrm{},i_{jq_j}]`$ may coincide only in the case when the permutation $`\lambda `$ acts on the numbers $`i_{j1}`$,โฆ,$`i_{jq_j}`$ as a cyclic permutation. By the definition of the permutation $`\lambda `$ it is possible only for the coinciding completely reduced closed paths $`C_{i_{j1}}^{}`$,โฆ, $`C_{i_{jq_j}}^{}`$. But then the oriented completely reduced cycle $`[C_{i_{j1}}^{}\mathrm{}C_{i_{jq_j}}^{}]`$ is non โ primitive. Thus every set of the different primitive oriented completely reduced cycles $`[C_{i_{j1}}^{}\mathrm{}C_{i_{jq_j}}^{}]`$, $`j=1,\mathrm{},k`$, corresponds with $`_{i=1}^m(n_i)!`$ different permutations $`\pi [\lambda (i_{11}),\mathrm{},\lambda (i_{1q_1})]\mathrm{}\pi [\lambda (i_{k1}),\mathrm{},\lambda (i_{kq_k})]`$ of the numbers $`1,\mathrm{},n`$. The number $`_{i=1}^m(n_i)!`$ is the total number of the permutations $`\lambda `$ rearranging only the numbers corresponding with the coinciding completely reduced closed paths $`C_1^{}`$,โฆ,$`C_n^{}`$. The equality (5) is proved.
The equalities (5.22), (5) imply that the left hand side sum (5) is equal to zero for $`n>1`$. Hence the sum (5.18) equals zero for $`n>1`$ and the right hand sides of the equalities (5.13) and (5) coincide. The theorem is proved.
For the proof of Theorem 5.2 we used Theorem 5.1 and the definitions (2.17), (4.5), and (2.18), (4.4). Therefore Theorem 5.2 is valid also for a graph $`G`$ embedded in a rectangular lattice on the plane.
Let the left hand side of the inequality (2.30) be denoted by $`u`$. The inequality (4.6) is a particular case of the inequality (2.30). It may be rewritten as $`u<1/3`$.
Theorem 5.3. Let a graph $`G`$ be embedded in the rectangular lattice $`\stackrel{~}{G}(M_1^{},M_2^{};M_1,M_2)`$ on the torus. Let the estimate (4.6) be fulfilled and let interaction energy $`E(๐)`$ be non โ negative. Then for the reduced partition function (4.2) the following inequalities are valid
$`18/3\left({\displaystyle \underset{s=1}{\overset{2}{}}}(M_sM_s^{})\right)(13u)^1{\displaystyle \underset{s=1}{\overset{2}{}}}(3u)^{M_sM_s^{}}`$
$`(Z_{r,G})^1\mathrm{exp}\{1/2{\displaystyle \underset{CRC(G)}{}}|C|^1๐ฎ^C\rho (C)\}1`$ (5.25)
where $`C`$ run over the set $`RC(G)`$ of reduced closed paths on the graph $`G`$, the natural number $`|C|`$ is the length of the path $`C`$, the number $`๐ฎ^C`$ is given by the equalities (2.18), (4.4) and the number $`\rho (C)`$ is given by the equalities (2.17), (4.5).
Proof. Due to Theorem 5.2 the equality (5.13) is valid. Let four different oriented edges $`๐_i`$, $`i=1,\mathrm{},4`$, of the graph $`G`$ have the same initial vertex: $`b(๐_1)=\mathrm{}=b(๐_4)`$. Let the prime closed paths $`C_1=(๐_1,๐_5,\mathrm{},๐_m,๐_2^1)`$ and $`C_2=(๐_3,๐_{m+1},\mathrm{},๐_{m+n},๐_4^1)`$ on the graph $`G`$ correspond with the disjoint prime non โ oriented cycles $`[[C_1]]`$ and $`[[C_2]]`$. Then the products $`C_1C_2=(๐_1,๐_5,\mathrm{},๐_m,๐_2^1,๐_3,๐_{m+1},\mathrm{},๐_{m+n},๐_4^1)`$ and
$`C_1C_2^1=(๐_1,๐_5,\mathrm{},๐_m,๐_2^1,๐_4,๐_{m+n}^1,\mathrm{},๐_{m+1}^1,๐_3^1)`$ are the prime closed paths. Let $`C_3`$,โฆ,$`C_k`$ be the prime closed paths on the graph $`G`$ such that the non โ oriented cycles
$`[[C_1]]`$,$`[[C_2]]`$,$`[[C_3]]`$,โฆ,$`[[C_k]]`$ are disjoint. Then the prime non โ oriented cycles $`[[C_1C_2]]`$, $`[[C_3]]`$,โฆ,$`[[C_k]]`$ are disjoint and the prime non โ oriented cycles $`[[C_1C_2^1]]`$,$`[[C_3]]`$,โฆ,$`[[C_k]]`$ are disjoint too.
Let us consider the sum of three terms in the right hand side sum (5.13)
$`(1)^k\left({\displaystyle \underset{i=1}{\overset{k}{}}}๐ฎ^{C_i}\right)\left({\displaystyle \underset{i=1}{\overset{k}{}}}\rho (C_i)\right)+(1)^{k1}๐ฎ^{C_1C_2}\left({\displaystyle \underset{i=3}{\overset{k}{}}}๐ฎ^{C_i}\right)\rho (C_1C_2)\left({\displaystyle \underset{i=3}{\overset{k}{}}}\rho (C_i)\right)+`$
$`(1)^{k1}๐ฎ^{C_1C_2^1}\left({\displaystyle \underset{i=3}{\overset{k}{}}}๐ฎ^{C_i}\right)\rho (C_1C_2^1)\left({\displaystyle \underset{i=3}{\overset{k}{}}}\rho (C_i)\right).`$ (5.26)
The definitions (2.18), (4.4) imply that
$$๐ฎ^{C_1C_2}=๐ฎ^{C_1C_2^1}=๐ฎ^{C_1}๐ฎ^{C_2}.$$
(5.27)
The definitions (2.17), (4.5) and the relations (5.7) imply
$`\rho (C_1C_2)=\gamma (๐_1,๐_2,๐_3,๐_4)\rho (C_1)\rho (C_2),`$
$`\rho (C_1C_2^1)=\gamma (๐_1,๐_2,๐_4,๐_3)\rho (C_1)\rho (C_2)`$ (5.28)
where
$$\gamma (๐_1,๐_2,๐_3,๐_4)=\rho (๐_1^1;๐_2)\rho (๐_2^1;๐_3)\rho (๐_3^1;๐_4)\rho (๐_4^1;๐_1).$$
(5.29)
In the second equality (5) we took into account the value of the angle through which the tangent vector of the path $`C_2`$ turns along the path $`C_2`$: $`\varphi (C_2)=2\pi k`$ where $`k`$ is an integer. Hence $`\rho (C_2)=\mathrm{exp}\{i/2\varphi (C_2)\}=(\rho (C_2))^1`$.
The substitution of the equalities (5.27), (5) into the expression (5) gives
$$[1\gamma (๐_1,๐_2,๐_3,๐_4)\gamma (๐_1,๐_2,๐_4,๐_3)](1)^k\left(\underset{i=1}{\overset{k}{}}๐ฎ^{C_i}\right)\left(\underset{i=1}{\overset{k}{}}\rho (C_i)\right).$$
(5.30)
The definition (5.29) implies
$$\gamma (๐_2,๐_3,๐_4,๐_1)=\gamma (๐_1,๐_2,๐_3,๐_4).$$
(5.31)
Let for some index $`i=1,\mathrm{},3`$ the oriented edges $`๐_i`$ and $`๐_{i+1}`$ have the opposite directions or let the oriented edges $`๐_1`$ and $`๐_4`$ have the opposite directions. We shall prove that for all these cases $`\gamma (๐_1,๐_2,๐_3,๐_4)=1`$. Due to the relation (5.31) it is sufficient to prove this statement only for the case when the oriented edges $`๐_1`$ and $`๐_2`$ have the opposite directions. In this case the oriented edges $`๐_3`$ and $`๐_4`$ have also the opposite directions. Due to the definitions (4.5), (5.29) we have in this case
$$\gamma (๐_1,๐_2,๐_3,๐_4)=\rho (๐_2^1;๐_3)\rho (๐_4^1;๐_1).$$
(5.32)
The direction of the oriented edge $`๐_2^1`$ coincides with the direction of the oriented edge $`๐_1`$. The direction of the oriented edge $`๐_4^1`$ coincides with the direction of the oriented edge $`๐_3`$. Hence the definition (4.5) and the equality (5.32) imply $`\gamma (๐_1,๐_2,๐_3,๐_4)=1`$.
It is easy to verify that when the directions of the oriented edges $`๐_i`$ and $`๐_{i+1}`$, $`i=1,\mathrm{},3`$, $`๐_4`$ and $`๐_1`$ are orthogonal to each other the definitions (4.5), (5.29) imply $`\gamma (๐_1,๐_2,๐_3,๐_4)=1`$.
Let the directions of the oriented edges $`๐_i`$ and $`๐_{i+1}`$, $`i=1,\mathrm{},3`$, $`๐_4`$ and $`๐_1`$ be orthogonal to each other. Then the directions of the oriented edges $`๐_2`$ and $`๐_4`$ are opposite. Therefore $`\gamma (๐_1,๐_2,๐_3,๐_4)=1`$ and $`\gamma (๐_1,๐_2,๐_4,๐_3)=1`$. Hence due to the relations (5.27), (5) in this case the expression (5.30) is equal to
$$(1)^k\left(\underset{i=1}{\overset{k}{}}๐ฎ^{C_i}\right)\left(\underset{i=1}{\overset{k}{}}\rho (C_i)\right)=(1)^{k1}๐ฎ^{C_1C_2}\left(\underset{i=3}{\overset{k}{}}๐ฎ^{C_i}\right)\rho (C_1C_2)\left(\underset{i=3}{\overset{k}{}}\rho (C_i)\right).$$
(5.33)
The paths $`C_1`$,$`C_2`$ and $`C_1C_2`$ go subsequently through the oriented edges $`๐_i^{\pm 1}`$, $`i=1,\mathrm{},4`$, having the directions orthogonal to each other.
Let the oriented edges $`๐_1`$ and $`๐_2`$, $`๐_2`$ and $`๐_4`$, $`๐_4`$ and $`๐_3`$, $`๐_3`$ and $`๐_1`$ have the directions orthogonal to each other. Thus the oriented edges $`๐_2`$ and $`๐_3`$ have the opposite directions. Therefore $`\gamma (๐_1,๐_2,๐_3,๐_4)=1`$ and $`\gamma (๐_1,๐_2,๐_4,๐_3)=1`$. Hence in this case the expression (5.30) is equal to
$$(1)^k\left(\underset{i=1}{\overset{k}{}}๐ฎ^{C_i}\right)\left(\underset{i=1}{\overset{k}{}}\rho (C_i)\right)=(1)^{k1}๐ฎ^{C_1C_2^1}\left(\underset{i=3}{\overset{k}{}}๐ฎ^{C_i}\right)\rho (C_1C_2^1)\left(\underset{i=3}{\overset{k}{}}\rho (C_i)\right).$$
(5.34)
The paths $`C_1`$,$`C_2`$ and $`C_1C_2^1`$ go subsequently through the oriented edges $`๐_i^{\pm 1}`$, $`i=1,\mathrm{},4`$, having the directions orthogonal to each other.
Let the oriented edges $`๐_1`$ and $`๐_2`$ have the opposite directions. Hence $`\gamma (๐_1,๐_2,๐_3,๐_4)=\gamma (๐_1,๐_2,๐_4,๐_3)=1`$ and in this case the expression (5.30) is equal to
$`(1)^{k1}๐ฎ^{C_1C_2}\left({\displaystyle \underset{i=3}{\overset{k}{}}}๐ฎ^{C_i}\right)\rho (C_1C_2)\left({\displaystyle \underset{i=3}{\overset{k}{}}}\rho (C_i)\right)=`$
$`(1)^{k1}๐ฎ^{C_1C_2^1}\left({\displaystyle \underset{i=3}{\overset{k}{}}}๐ฎ^{C_i}\right)\rho (C_1C_2^1)\left({\displaystyle \underset{i=3}{\overset{k}{}}}\rho (C_i)\right).`$ (5.35)
Since the oriented edges $`๐_1`$ and $`๐_2`$ have the opposite directions, the oriented edges $`๐_3`$ and $`๐_4`$ have the opposite directions and the directions of the oriented edges $`๐_1`$ and $`๐_4`$, $`๐_4`$ and $`๐_2`$, $`๐_2`$ and $`๐_3`$, $`๐_3`$ and $`๐_1`$ are orthogonal to each other. Thus in this case the paths $`C_1C_2`$ and $`C_1C_2^1`$ go subsequently through the oriented edges $`๐_i^{\pm 1}`$, $`i=1,\mathrm{},4`$, having the directions orthogonal to each other.
We have considered all possible directions of the oriented edges $`๐_1`$,$`๐_2`$,$`๐_3`$,$`๐_4`$. In any case the sum (5) is equal to one of the expressions (5.33) โ (5) where the paths go subsequently through the oriented edges $`๐_i^{\pm 1}`$, $`i=1,\mathrm{},4`$, having the directions orthogonal to each other.
With the term
$$(1)^k\left(\underset{i=1}{\overset{k}{}}๐ฎ^{C_i}\right)\left(\underset{i=1}{\overset{k}{}}\rho (C_i)\right)$$
(5.36)
the right hand side sum (5.13) contains all the terms of type (5.36) where the disjoint prime non โ oriented cycles $`[[C_1^{}]]`$,โฆ,$`[[C_k^{}^{}]]`$ contain the same non โ oriented edges as the disjoint prime non โ oriented cycles $`[[C_1]]`$,โฆ,$`[[C_k]]`$. We have proved that the sum of these terms of type (5.36) is equal to the only term of type (5.36) where the prime closed paths $`C_1`$,โฆ,$`C_k`$ satisfy the condition: if four different oriented edges from the prime closed paths $`C_1`$,โฆ,$`C_k`$ are incident to one vertex, then the prime closed paths $`C_1`$,โฆ,$`C_k`$ go subsequently through the oriented edges from these four oriented edges having the directions orthogonal to each other.
Let us define the cochain $`\xi ^1[C]C^1(P(G),๐_2^{add})`$ equal $`1`$ on all non โ oriented edges from the prime non โ oriented cycle $`[[C]]`$ on the graph $`G`$. The cochain $`\xi ^1[C]`$ equals zero on all other non โ oriented edges of the graph $`G`$. Since $`[[C]]`$ is a prime non โ oriented cycle, $`\xi ^1[C]Z_1(P(G),๐_2^{add})`$. If the prime non โ oriented cycles $`[[C_1]]`$,โฆ,$`[[C_k]]`$ are disjoint, then the supports of the cycles $`\xi ^1[C_1]`$,โฆ,$`\xi ^1[C_k]`$ do not intersect. By the definition $`\rho (C_i)=\pm 1`$, $`i=1,\mathrm{},k`$. The interaction energy $`E(๐)`$ is non โ negative. Hence we obtain the estimate for the term (5.36)
$$(1)^k\left(\underset{i=1}{\overset{k}{}}๐ฎ^{C_i}\right)\left(\underset{i=1}{\overset{k}{}}\rho (C_i)\right)๐ฎ^{_1^k\xi ^1[C_i]}.$$
(5.37)
Since the sum of the terms of type (5.36) corresponding with the disjoint prime non โ oriented cycles containing the same set of the non โ oriented edges is again the term of type (5.36), it follows from the equality (4.2) and from the inequality (5.37) that
$$\mathrm{exp}\{1/2\underset{CRC(G)}{}|C|^1๐ฎ^C\rho (C)\}Z_{r,G}.$$
(5.38)
We shall prove that for any cycle $`\xi ^1Z_1(P(G),๐_2^{add})`$ there exist the disjoint prime non โ oriented cycles $`[[C_1]]`$,โฆ,$`[[C_k]]`$ such that $`\xi ^1=\xi ^1[C_1]+\mathrm{}+\xi ^1[C_k]`$. Let $`(๐ฆ,๐ฏ)`$ be an oriented edge corresponding with a non โ oriented edge on which the cycle $`\xi ^1`$ takes the value $`1`$. Then the vertex $`๐ฆ+๐ฏ`$ is incident to two or four non โ oriented edges on which the cycle $`\xi ^1`$ takes the value $`1`$. Let us choose an oriented edge $`(๐ฆ+๐ฏ,๐ฏ_1)`$, $`๐ฏ_1๐ฏ`$, corresponding with a non โ oriented edge on which the cycle $`\xi ^1`$ takes the value $`1`$. By repeating this process we obtain the prime closed path $`C_1`$ such that the support of the cycle $`\xi ^1[C_1]`$ is contained in the support of the cycle $`\xi ^1`$. Hence there exists a cycle $`\eta ^1Z_1(P(G),๐_2^{add})`$ such that $`\xi ^1=\xi ^1[C_1]+\eta ^1`$ and the supports of the cycles $`\xi ^1[C_1]`$ and $`\eta ^1`$ do not intersect. By applying the above procedure for a cycle $`\eta ^1`$ we construct a prime closed path $`C_2`$. By repeating this process we obtain the disjoint prime non โ oriented cycles $`[[C_1]]`$,โฆ,$`[[C_k]]`$ such that $`\xi ^1=\xi ^1[C_1]+\mathrm{}+\xi ^1[C_k]`$.
For a cycle $`\xi ^1B_1(P(G),๐_2^{add})`$ we can construct the disjoint prime non โ oriented cycles $`[[C_1]]`$,โฆ,$`[[C_k]]`$ such that $`\xi ^1=\xi ^1[C_1]+\mathrm{}+\xi ^1[C_k]`$ and the prime closed paths satisfy the above condition. If $`\xi ^1B_1(P(G),๐_2^{add})`$, then $`\xi ^1=\xi ^2`$ where a cochain $`\xi ^2C^2(P(G),๐_2^{add})`$. The support of the cochain $`\xi ^2`$ consists of the squares on which the cochain $`\xi ^2`$ takes the value $`1`$. Two squares $`s_i^2`$ and $`s_j^2`$ belong to one connected component of the support of the cochain $`\xi ^2`$ if there exist the squares $`s_{i_1}^2`$,โฆ,$`s_{i_k}^2`$ from the support of the cochain $`\xi ^2`$ such that the boundaries of the squares $`s_i^2`$ and $`s_{i_1}^2`$, $`s_{i_l}^2`$ and $`s_{i_{l+1}}^2`$, $`l=1,\mathrm{},k1`$, $`s_{i_k}^2`$ and $`s_j^2`$ contain the common non โ oriented edges. The boundaries of the squares from the different connected components of the support of the cochain $`\xi ^2`$ may contain the common vertices only. Thus $`\xi ^1=\xi _1^2+\mathrm{}+\xi _k^2`$ where the support of the cochain $`\xi _i^2`$ has the only connected component and for $`ij`$ the supports of the cochains $`\xi _i^2`$ and $`\xi _j^2`$ do not intersect. The support of the cochain $`\xi _i^2`$ corresponds with the prime non โ oriented cycle $`[[C_i]]`$ on the graph $`G`$. By the construction the prime non โ oriented cycles $`[[C_1]]`$,โฆ,$`[[C_k]]`$ are disjoint. Moreover, if four different oriented edges from the prime closed paths $`C_1`$,โฆ,$`C_k`$ are incident to one vertex, then the prime closed paths $`C_1`$,โฆ,$`C_k`$ go subsequently through such oriented edges from these four oriented edges that have the directions orthogonal to each other.
Let a prime non โ oriented cycle $`[[C]]`$ be the boundary of a connected set of the squares. We remove one square from this set, so that the new set is also connected. Let a prime non โ oriented cycle $`[[C^{}]]`$ be the boundary of this new connected set of the squares. Let $`C`$ and $`C^{}`$ be the representatives of the prime non โ oriented cycles $`[[C]]`$ and $`[[C^{}]]`$. By using the definitions (2.17), (4.5) it is possible to prove that $`\rho (C)=\rho (C^{})`$. By repeating this process we obtain $`\rho (C^{\prime \prime })=\rho (C)`$ where $`C^{\prime \prime }`$ is a representative of the boundary of one square. It is easy to calculate that $`\rho (C^{\prime \prime })=\rho (C)=1`$.
The previous arguments imply that the sum of the terms (5.36) corresponding with a cycle $`\xi ^1B_1(P(G),๐_2^{add})`$ is equal to a unique term (5.36) where $`\rho (C_i)=1`$, $`i=1,\mathrm{},k`$. For a cycle $`\xi ^1B_1(P(G),๐_2^{add})`$ we obtain a unique term (5.36) where $`\rho (C_i)=\pm 1`$, $`i=1,\mathrm{},k`$. The interaction energy $`E(๐)`$ is non โ negative. Therefore we get the estimate
$$Z_{r,G}\mathrm{exp}\{1/2\underset{CRC(G)}{}|C|^1๐ฎ^C\rho (C)\}2\underset{\genfrac{}{}{0pt}{}{\xi ^1Z_1(P(G),Z_2^{add}),}{\xi ^1B_1(P(G),Z_2^{add})}}{}๐ฎ^{\xi ^1}.$$
(5.39)
Let us evaluate the right hand side of the inequality (5.39). We have proved that any cycle $`\xi ^1Z_1(P(G),๐_2^{add})`$ has the form $`\xi ^1=\xi ^1[C_1]+\mathrm{}+\xi ^1[C_k]`$ where the prime non โ oriented cycles $`[[C_1]]`$,โฆ,$`[[C_k]]`$ are disjoint. If $`\xi ^1B_1(P(G),๐_2^{add})`$, then at least for one prime non โ oriented cycle $`[[C_i]]`$ the cochain $`\xi ^1[C_i]B_1(P(G),๐_2^{add})`$. Hence any cycle $`\xi ^1B_1(P(G),๐_2^{add})`$ has the form $`\xi ^1=\xi ^1[C]+\eta ^1`$ where the prime non โ oriented cycle $`[[C]]`$ corresponds with the cycle $`\xi ^1[C]B_1(P(G),๐_2^{add})`$ and the supports of the path $`C`$ and of the cycle $`\eta ^1Z_1(P(G),๐_2^{add})`$ do not intersect. By using this decomposition we obtain the estimate
$$\underset{\genfrac{}{}{0pt}{}{\xi ^1Z_1(P(G),Z_2^{add}),}{\xi ^1B_1(P(G),Z_2^{add})}}{}๐ฎ^{\xi ^1}Z_{r,G}\underset{\genfrac{}{}{0pt}{}{[[C]]:prime,}{\xi ^1[C]B_1(P(G),Z_2^{add})}}{}๐ฎ^C.$$
(5.40)
The total number of all reduced closed paths of the length $`l`$ with the fixed initial vertex on the lattice $`\stackrel{~}{G}(M_1^{},M_2^{};M_1,M_2)`$ is less than $`43^{l1}`$. The total number of vertices of the lattice $`\stackrel{~}{G}(M_1^{},M_2^{};M_1,M_2)`$ is equal to $`(M_1M_1^{})(M_2M_2^{})`$. Hence the total number of all reduced closed paths of the length $`l`$ on the lattice $`\stackrel{~}{G}(M_1^{},M_2^{};M_1,M_2)`$ is less than $`(M_1M_1^{})(M_2M_2^{})43^{l1}`$. Let for a prime closed path $`C`$ on the lattice $`\stackrel{~}{G}(M_1^{},M_2^{};M_1,M_2)`$ the cochain $`\xi ^1[C]B_1(P(G),๐_2^{add})`$. Hence the length $`|C|`$ of a path $`C`$ is more than $`M=\mathrm{min}(M_1M_1^{},M_2M_2^{})`$. It implies the following estimate
$`{\displaystyle \underset{\genfrac{}{}{0pt}{}{[[C]]:prime,}{\xi ^1[C]B_1(P(G),Z_2^{add})}}{}}๐ฎ^C{\displaystyle \underset{CRC(\stackrel{~}{G}),|C|M}{}}u^{|C|}`$
$`4/3(M_1M_1^{})(M_2M_2^{}){\displaystyle \underset{l=M}{\overset{\mathrm{}}{}}}(3u)^l`$
$`4/3\left({\displaystyle \underset{s=1}{\overset{2}{}}}(M_sM_s^{})\right)(13u)^1{\displaystyle \underset{s=1}{\overset{2}{}}}(3u)^{M_sM_s^{}}`$ (5.41)
where $`u`$ denotes the left hand side of the inequality (2.30). The inequalities (5.38) โ (5) imply the inequalities (5). The theorem is proved.
If a graph $`G`$ is embedded in a rectangular lattice on the plane, then $`Z_1(P(G),๐_2^{add})=B_1(P(G),๐_2^{add})`$. Therefore the inequality (5.39) becomes the equality (4.7). Thus we obtain a new proof of the equality (4.7) independent of the papers , , , .
It follows from the relations (4.1) โ (4.4) for the homogeneous Ising model on the torus that
$$Z_{\stackrel{~}{G}(M_1^{},M_2^{};M_1,M_2)}=Z_{\stackrel{~}{G}(0,0;M_1M_1^{},M_2M_2^{})}.$$
(5.42)
Let us denote $`E_1`$($`E_2`$) the interaction energy $`E(๐)`$ for horizontally (vertically) directed edges $`๐`$ of the lattice $`\stackrel{~}{G}(0,0;M_1,M_2)`$.
Theorem 5.4. Let for the non โ negative interaction energy of the homogeneous Ising model on a rectangular lattice on the torus the estimate (4.6) be valid. Then for the partition function (4.1) of the homogeneous Ising model on the rectangular lattice $`\stackrel{~}{G}(0,0;M_1,M_2)`$ on the torus
$`\underset{\genfrac{}{}{0pt}{}{M_1,M_2\mathrm{},}{M_1(M_2)^1+M_2(M_1)^1const}}{lim}(M_1M_2)^1\mathrm{ln}Z_{\stackrel{~}{G}(0,0;M_1,M_2)}=`$
$`\mathrm{ln}(2\mathrm{cosh}\beta E_1\mathrm{cosh}\beta E_2)+1/2(2\pi )^2{\displaystyle _0^{2\pi }}๐\theta _1{\displaystyle _0^{2\pi }}๐\theta _2`$
$`\mathrm{ln}[(1+z_1^2)(1+z_2^2)2z_1(1z_2^2)\mathrm{cos}\theta _12z_2(1z_1^2)\mathrm{cos}\theta _2]`$ (5.43)
where the variables $`z_i=\mathrm{tanh}\beta E_i`$, $`i=1,2`$.
Proof. The equality (2.36) and the inequalities (5) imply
$`\mathrm{ln}[18/3(M_1M_2)(13u)^1{\displaystyle \underset{s=1}{\overset{2}{}}}(3u)^{M_s}]`$
$`1/2\mathrm{ln}[det(IT(๐ฎ,\rho ))]\mathrm{ln}Z_{\stackrel{~}{G}(0,0;M_1,M_2)}0`$ (5.44)
where the matrix $`T(๐ฎ,\rho )`$ is given by the equalities (2.32) and (4.5) for the lattice
$`\stackrel{~}{G}(0,0;M_1,M_2)`$. Now the equality (5) follows from the inequalities (5) and from the equalities (4.1) and (4). The theorem is proved.
The equality (5) is proved in the paper for the arbitrary interaction energies $`E_i`$, $`i=1,2`$.
The relation (4.27) is valid also for the graph $`\stackrel{~}{G}=\stackrel{~}{G}(M_1^{},M_2^{};M_1,M_2)`$.
Theorem 5.5. Let for Ising model on the rectangular lattice $`\stackrel{~}{G}(M_1^{},M_2^{};M_1,M_2)`$ on the torus the estimate (4.6) be valid and let the interaction energy $`E(๐)`$ be non โ negative. Let a cochain $`\chi C^0(P(๐^{\times 2}),๐_2^{add})`$ be equal to $`1`$ on the finite number of the vertices. Then for the correlation function (3) of the two dimentional Ising model with periodic boundary conditions
$`\underset{\genfrac{}{}{0pt}{}{M_sM_s^{}\mathrm{},s=1,2,}{(M_1M_1^{})(M_2M_2^{})^1+(M_2M_2^{})(M_1M_1^{})^1const}}{lim}W_{\stackrel{~}{G}(M_1^{},M_2^{};M_1,M_2)}(\chi )=`$
$`{\displaystyle \underset{\genfrac{}{}{0pt}{}{\xi ^1C^1(P(G),Z_2^{add}),\xi ^1=\chi ,}{\chi connected\xi ^1}}{}}๐ฎ^{\xi ^1}\mathrm{exp}\{1/2{\displaystyle \underset{\genfrac{}{}{0pt}{}{CRC(Z^{\times 2}),}{Ci(\xi ^1)\mathrm{}}}{}}|C|^1๐ฎ^C\rho (C)\}`$ (5.45)
where the number $`๐ฎ^{\xi ^1}`$ is defined by the relations (4.3), (4.4), the number $`๐ฎ^C`$ is defined by the relations (2.18), (4.4) and the number $`\rho (C)`$ is defined by the relations (2.17), (4.5).
Proof. Let the estimate (4.6) be valid. Let the interaction energy be non โ negative. Let a cochain $`\xi ^1C^1(P(\stackrel{~}{G}),๐_2^{add})`$ satisfy the condition $`\xi ^1=\chi `$ and let its support $`\xi ^1`$ be $`\chi `$ โ connected. Let $`i(\xi ^1)`$ be the set of all non โ oriented edges incident to the vertices incident to the edges of the support $`\xi ^1`$. The inequalities (5) for the graphs $`\stackrel{~}{G}`$ and $`\stackrel{~}{G}i(\xi ^1)`$ imply
$`18/3\left({\displaystyle \underset{s=1}{\overset{2}{}}}(M_sM_s^{})\right)(13u)^1{\displaystyle \underset{s=1}{\overset{2}{}}}(3u)^{M_sM_s^{}}`$
$`Z_{r,\stackrel{~}{G}}(Z_{r,\stackrel{~}{G}i(\xi ^1)})^1\mathrm{exp}\{1/2{\displaystyle \underset{\genfrac{}{}{0pt}{}{CRC(\stackrel{~}{G}),}{Ci(\xi ^1)\mathrm{}}}{}}|C|^1๐ฎ^C\rho (C)\}`$
$`(18/3\left({\displaystyle \underset{s=1}{\overset{2}{}}}(M_sM_s^{})\right)(13u)^1{\displaystyle \underset{s=1}{\overset{2}{}}}(3u)^{M_sM_s^{}})^1.`$ (5.46)
The interaction energy $`E(๐)`$ is non โ negative. Then the definition (4.2) implies the following estimate
$$(Z_{r,\stackrel{~}{G}})^1Z_{r,\stackrel{~}{G}i(\xi ^1)}1.$$
(5.47)
It follows from the estimates (4.6) and (5.47) that for $`M_sM_s^{}\mathrm{}`$, $`s=1,2`$, the non โ zero contributions give only those terms of the sum (4.27) for the graph $`\stackrel{~}{G}`$ which correspond to the cochains $`\xi ^1C^1(P(๐^{\times 2}),๐_2^{add})`$ with the finite supports $`\xi ^1`$. Let a cochain $`\xi ^1C^1(P(๐^{\times 2}),๐_2^{add})`$ satisfy the condition $`\xi ^1=\chi `$ and let it have the finite $`\chi `$ โ connected support $`\xi ^1`$. For sufficiently large $`M_sM_s^{}`$, $`s=1,2`$, the set $`i(\xi ^1)\stackrel{~}{G}(M_1^{},M_2^{};M_1,M_2)`$. Let us prove that
$`\underset{\genfrac{}{}{0pt}{}{M_sM_s^{}\mathrm{},s=1,2,}{(M_1M_1^{})(M_2M_2^{})^1+(M_2M_2^{})(M_1M_1^{})^1const}}{lim}(Z_{r,\stackrel{~}{G}})^1Z_{r,\stackrel{~}{G}i(\xi ^1)}=`$
$`\mathrm{exp}\{1/2{\displaystyle \underset{\genfrac{}{}{0pt}{}{CRC(Z^{\times 2}),}{Ci(\xi ^1)\mathrm{}}}{}}|C|^1๐ฎ^C\rho (C)\}.`$ (5.48)
Indeed, for $`M_sM_s^{}\mathrm{}`$, $`s=1,2`$, and $`(M_1M_1^{})(M_2M_2^{})^1+(M_2M_2^{})(M_1M_1^{})^1const`$ the left and the right hand sides of the inequalities (5) tend to $`1`$. Since the set $`i(\xi ^1)`$ is finite, the last multiplier in the central part of the inequalities (5) tends to the right hand side of the equality (5). The series (5) coincides with the series (4). It was proved in Theorem 4.3 that the series (4) is absolutely convergent if the estimate (4.6) is fulfilled and the interaction energy is non โ negative. Thus the sum (4.27) for the graph $`\stackrel{~}{G}(M_1^{},M_2^{};M_1,M_2)`$ converges to the series (5). The theorem is proved.
The correlation functions (5) and (4) coincide. |
warning/0001/quant-ph0001112.html | ar5iv | text | # There is no spooky action-at-a-distance in quantum correlations: Resolution of the Einstein-Podolsky-Rosen nonlocality puzzle
## Abstract
The long-standing puzzle of the nonlocal Einstein-Podolsky-Rosen correlations is resolved. The correct quantum mechanical correlations arise for the case of entangled particles when strict locality is assumed for the probability amplitudes instead of locality for probabilities. Locality of amplitudes implies that measurement on one particle does not collapse the companion particle to a definite state.
Sixty five years ago, the most significant paper questioning a fundamental aspect of quantum phenomena was written by Einstein, Podolsky and Rosen (EPR) . They addressed the question whether the wave-function represented a complete description of reality in quantum mechanics, and argued that it didnโt. Bohrโs reply to this paper was not sufficient to resolve the fundamental issues raised by EPR. Decades later, Bohm rephrased the EPR problem in terms of particles correlated in their spin and this helped enormously in analyzing the problem with clarity. John Bell analyzed the EPR problem in the early sixties and established the Bellโs inequalities obeyed by any local hidden variable theory for the correlations of entangled particles . Quantum mechanical correlations calculated using the entangled wave-function and spin operators violate these inequalities. Experiments, the first of which was by Freedman and Clauser and the most remarkable by A. Aspect and collaborators , have established beyond doubt that there cannot be a viable local realistic hidden variable description of quantum mechanics . Further, these results also have been interpreted as evidence for nonlocal influences in quantum measurements involving entangled particles. Since no instruction set carried by the particles from their source of origin (possibly with the addition of several local hidden variables) can manage to create the correct correlations observed in experiments, the only way out seems to be that measurement of an observable on one of the particles in an entangled pair seems to convey the result of this measurement instantaneously to the other particle resulting in the correct behaviour of the other particle during a measurement on the second particle. Of course, the no signalling theorems in this context prohibit any faster than light signalling using this feature. Nevertheless, we seem to be stuck with the puzzling nonlocality which is probably the deepest mystery in the behaviour of entangled systems. In the quantum mechanical terminology, the measurement of an observable on one of the particles collapses the entire wave-function instantaneously and nonlocally and the second particle acquires a definite value for the same observable, consistent with the correlation determined by the relevant conservation law.
Apart from the disturbing aspect of accepting the concept of nonlocality without being able to understand its nature, there is serious conflict with the spirit of relativity. As soon as we bring in the concept of one measurement being influenced nonlocally by the other, the notion of simultaneity becomes important since both measurements can be labelled by local times. So, if one measurement precede the other in one frame, one can always find a moving frame in which the converse it true, the second measurement preceding the first .
In this paper we discuss the resolution of the quantum nonlocality puzzle. The crucial new idea is to assume locality at the level of probability amplitudes instead of at the level of probabilities. For quantum systems which show wave-like behaviour represented by complex numbers, this seems to be the physically correct assumption to make. The quantum correlation is encoded in the difference of an internal variable for the problem.
Consider the breaking up of a correlated state as in the standard Bohm version of the EPR problem . The two particle go off in opposite directions and are in space-like regions. Two observers make measurements on these particles individually at space like separated regions with time stamps such that these results can be correlated later through a classical channel. We assume that strict locality is valid at the level of probability amplitudes. A measurement changes probability amplitudes only locally. Measurements performed in one region do not change the magnitude or phase of the complex amplitude for the companion particle in a space-like separated region.
We assign local rules (probability amplitudes) for the outcome of a particular measurement on each of the two particles. We also assume the existence of an internal variable for each of these two particles. The correlation at source is encoded in the relative value, or the difference, of this internal variable for the two particles. For simplicity let us call this internal variable a โphaseโ, $`\varphi `$. Note that it is not a dynamical phase evolving as the particle propagates. It is an internal variable whose difference (possibly zero) remains constant for the particles of the correlated pair. The value of $`\varphi `$ can vary from particle to particle, but the relative phase between the two particles in all correlated pairs is constant. Consider $`\varphi `$ as a reference for the particles to determine the angle of a polarizer or analyzer encountered on their way, locally (we use the terms polarizer and analyzer in a generic way. They could be Stern-Gerlach like analyzers for spin $`1/2`$ particles). The first particle encounters analyzer #1 kept at an angle $`\theta _1`$ with respect to some global direction. We denote this angle of the analyzer with reference to $`\varphi `$ as $`\theta .`$ Similarly, the second particle which has the internal phase angle $`\varphi +\varphi _o,`$ where $`\varphi _o`$ is a constant, encounters the second analyzer oriented at angle $`\theta _2`$ at another space-like separated point. Let the orientation of this analyzer with respect to the internal phase angle of the second particle is $`\theta ^{}.`$ We have $`\theta \theta ^{}=\theta _1\theta _2+\varphi _o.`$ (The constant $`\varphi _o`$ characterizes the correlation.)
An experiment in which each particle is analyzed by orienting the analyzers at various angles $`\theta _1`$ and $`\theta _2`$ is considered next. At each location the result is two-valued denoted by ($`+1`$) for transmission and ($`1`$) for absorption of each particle, for any angle of orientation. The classical correlation function $`P(๐,๐)=\frac{1}{N}(A_iB_i)`$ satisfies $`1P(๐,๐)1.`$ Here $`(๐,๐)`$ denotes the two directions along which the analyzers are oriented and $`A_i`$ and $`B_i`$ are the two valued results. The Bell correlation $`P(๐,๐)`$ denotes the average of the quantity (number of detections in coincidence $``$ number of detections in anticoincidence), where โcoincidenceโ denotes both particles showing same value for the measurement and โanticoincidenceโ denotes those with opposite values. Our aim is to calculate the Bell correlation from our formalism employing local amplitudes and compare it with the quantum mechanical prediction obtained from the nonlocal entangled wave function and spin operators.
We specify the local rule for transmission as a complex number, whose square gives the probability of transmission. The complex amplitude associated with particle #1 is $`C_1=\frac{1}{\sqrt{2}}\mathrm{exp}(i\theta s)`$ for measurements at analyzer #1, and for particle #2 is $`C_2=\frac{1}{\sqrt{2}}\mathrm{exp}(i\theta ^{}s)`$ at analyzer #2. (For the maximally entangled particles, the amplitude for the alternate outcome at the analyzer differs only by a phase). In these expressions, the quantity $`s`$ is the spin (in units of $`\mathrm{}`$) of the particle, $`1`$ for photons and $`\frac{1}{2}`$ for spin-$`\frac{1}{2}`$ particles. The locality assumption is strictly enforced since the two complex functions depend only on local variables and on an internal variable determined at source and then individually carried by the particles without any subsequent communication of any sort. The probabilities for the outcomes of measurements at each end are now correctly reproduced, for any angle of orientation. These probabilities are $`Re(C_1C_1^{})=Re(C_2C_2^{})=\frac{1}{2}.`$ The correlation function for amplitudes is of the form $`Re(C_1C_2^{}).`$ The correlation amplitude for an outcome of either $`(++)`$ or $`()`$ of two maximally entangled particles is
$`U(\theta ,\theta ^{})`$ $`=`$ $`Re2(C_1C_2^{})=Re\{\mathrm{exp}is(\theta \theta ^{})\}`$ (1)
$`=`$ $`\mathrm{cos}\{s(\theta \theta ^{})\}=\mathrm{cos}\{s(\theta _1\theta _2)+s\varphi _o\}.`$
We rewrite this as $`U(\theta _1,\theta _2,\varphi _o)`$ since all references to the individual values of the hidden variable $`\varphi `$ has dropped out. The square of $`U(\theta _1,\theta _2,\varphi _o)`$ is the probability for coincidence detection of the two particles through the analyzers kept at angles $`\theta _1`$ and $`\theta _2`$. ( A distinction is made between what the quantum system uses as a rule for transmission, and what we can measure after the act of transmission. The correlation function is analogous to the two-point amplitude correlations of two independent electromagnetic fields).
Next we calculate the Bell correlation function $`P(๐,๐)`$ from the correlation function $`U(\theta _1,\theta _2,\varphi _o).`$ Since $`U^2(\theta _1,\theta _2,\varphi _o)`$ is the probability for a coincidence detection ($`++`$ or $``$), the quantity $`(1U^2(\theta _1,\theta _2,\varphi _o))`$ is the probability for an anticoincidence (events of the type $`+`$ and $`+`$). Since the average of the quantity (number of coincidences $``$ number of anticoincidences) =
$$U^2(\theta _1,\theta _2,\varphi _o)(1U^2(\theta _1,\theta _2,\varphi _o))=2U^2(\theta _1,\theta _2,\varphi _o)1,$$
(2)
the correspondence between $`P(๐,๐)`$ and $`U(\theta _1,\theta _2,\varphi _o)`$ is given by the general expression,
$$P(๐,๐)=2U^2(\theta _1,\theta _2,\varphi _o)1$$
(3)
Let us consider for discussion, the case of a correlated state of photons breaking up into orthogonal polarization states. This means that if one photon is transmitted through an analyzer on one side, the other one will not transmitted for the same orientation of the analyzer on the other side. So, perfect anti-correlation is implied for $`\theta _1\theta _2=0`$. The Bell correlation calculated from quantum mechanics for this case is given by $`\mathrm{cos}(2(`$ $`(\theta _1\theta _2)).`$ That is, if the analyzers are oriented at a relative angle of $`\pi /2,`$ perfect correlation is obtained. When the relative angle is $`\pi /4,`$ the quantum mechanical correlation defined in the Bell way is zero, since there are as many coincidences as anticoincidences.
The correlation function we derived give, for the case of the photons discussed above,
$$U(\theta _1,\theta _2,\varphi _o)=\mathrm{cos}\{(\theta _1\theta _2)+\varphi _o\}$$
(4)
We set $`\varphi _o=\pi /2`$ for denoting the correlation of the two orthogonal photons at source . Then we get
$`U(\theta _1,\theta _2,\varphi _o)`$ $`=`$ $`\mathrm{cos}\{(\theta _1\theta _2)+\pi /2\}`$ (5)
$`=`$ $`\mathrm{sin}(\theta _1\theta _2)`$
The probability for coincidence detection is
$$U^2(\theta _1,\theta _2,\varphi _o)=\mathrm{sin}^2(\theta _1\theta _2)$$
(6)
Correspondingly, the probability for anticoincidence is $`1\mathrm{sin}^2(\theta _1\theta _2).`$
We get for the Bell correlation,
$$P(๐,๐)=2\mathrm{sin}^2(\theta _1\theta _2)1=\mathrm{cos}(2((\theta _1\theta _2))$$
(7)
This agrees completely with the usual quantum mechanical prediction derived by applying the relevant spin operators on the correct entangled state of the two photons.
Another important example is the case of the singlet state breaking up into two spin $`1/2`$ particles propagating in opposite directions to spatially separated regions. We set $`\varphi _o=\pi .`$ Then our correlation function is
$`U(\theta _1,\theta _2,\varphi _o)`$ $`=`$ $`\mathrm{cos}\{s(\theta _1\theta _2)+s\varphi _o\}`$ (8)
$`=`$ $`\mathrm{cos}\{{\displaystyle \frac{1}{2}}(\theta _1\theta _2)+\pi /2\}`$
$`=`$ $`\mathrm{sin}{\displaystyle \frac{1}{2}}(\theta _1\theta _2)`$
The probability for joint detection through two Stern-Gerlach analyzers oriented at relative angle $`\theta _1\theta _2`$ is
$$U^2(\theta _1,\theta _2,\varphi _o)=\mathrm{sin}^2(\frac{1}{2}(\theta _1\theta _2))$$
(9)
For the case of the two particles of the singlet state,
$`2U^2(\theta _1,\theta _2,\varphi _o)1`$ $`=`$ $`2\mathrm{sin}^2({\displaystyle \frac{1}{2}}(\theta _1\theta _2))1`$ (10)
$`=`$ $`\mathrm{cos}(\theta _1\theta _2)=๐๐`$
This is again exactly same as the correct Bell correlation $`P(๐,๐)`$ for the quantum mechanical predictions obtained from the singlet entangled wave-function and the Pauli spin operators. Perfect correlation is obtained for oppositely oriented analyzers and perfect anticorrelation for similarly oriented analyzers. When the analyzers are orthogonal, the correlation is zero.
We have correctly reproduced the quantum mechanical correlation using local probability amplitudes. Bellโs theorem prohibiting local realistic theories is not violated since we used the concept of locality for probability amplitudes instead of locality at the level of probabilities. The correct correlation emerges from combining two local complex functions. Single events consisting of two independent measurements at the two analyzers obey the correlation we derived, and the probability for joint detection is given by the square of the correlation function. In particular if the two analyzers happen to be in the same orientation, perfect correlation is reproduced every time within the strict locality assumption. It is important to note that we have not used any information on the internal variable $`\varphi `$ even in terms of distributions. It may be considered as a hidden variable appearing in the measurement prescriptions only through a complex number and has the nature of the origin of a non-dynamical phase associated with the quantum system. In fact, such a variable is not an external input additional to what is already available in the quantum mechanical description, since the zero of the phase of a wave-function is unobservable.
All probabilities are guaranteed to be positive definite in our formalism since the correlation function is real. The nonlocality puzzle in the EPR correlations is resolved. Strict locality including Einstein locality is valid. An answer to the EPR query regarding the completeness of quantum description is found. It seems clear that even after performing a measurement on one of the particles of an entangled pair, the companion particle cannot be ascribed a reality in the sense of Einstein. The companion particleโs quantum properties remain as unmeasured and as โun-collapsedโ as ever, though the result of a measurement if performed, in the same direction, can be predicted with absolute certainty. Wave-function collapse in the sense of Copenhagen interpretation and realization of an outcome happens only during actually performed measurements and not as a consequence of a measurement on a subsystem of an entangled system. (I will argue in another paper that the results of the Popperโs experiment support this view).
The solution presented here resolves the problem, pointed out by EPR, of simultaneous reality of noncommuting observables. In fact the solution denies any reality to an actually unmeasured system. This suggests that there are physical systems in nature that are beyond the scope of the intuitive definition of EPR reality, just as the Copenhagen school maintained. The approach we have taken here gives predictions for correlations which are exactly the same as that would be obtained from the quantum wave-function and operators, without the apparent nonlocal influence of one measurement on the other. The nonlocality apparent in entanglement correlation in quantum mechanics is not an inherent feature, but a conclusion forced on us when using a restrictive definition of physical reality.
The same analysis works for particles entangled in other sets of variables like momentum and coordinate, and energy and time. The results follow from the fact that all these cases of two particle entanglement can be mapped on to the spin-$`\frac{1}{2}`$ singlet problem with two-valued outcomes. An experiment in which the particles entangled in momentum and position are used, with double slits for each of the particles, the amplitudes are
$`C_1`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}\mathrm{exp}(i\alpha k(x_1x_o)/2),`$
$`C_1`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}\mathrm{exp}(i\alpha k(x_2x_o)/2)`$ (11)
where $`x_1`$ and $`x_2`$ are the coordinates of the two detectors separated by a space-like interval. $`k`$ is the wave vector and $`\alpha `$ is a scaling factor for the angle subtended by the two slits at the detectors, source etc. The factor $`2`$ dividing the angular variable comes from the mapping with the spin-$`\frac{1}{2}`$ problem. The single particle data on either side separately do not show any interference. The correlation function is
$$U(x_1,x_2)=\mathrm{cos}(\alpha k(x_1x_2)/2)$$
(12)
Probability for coincidence detection is
$$P(x_1,x_2)=\mathrm{cos}^2(\alpha k(x_1,x_2)/2)=\frac{1}{2}(1+\mathrm{cos}k\alpha (x_1x_2))$$
(13)
This is the two photon correlation pattern with 100% visibility, derived assuming locality of probability amplitudes. This agrees with the quantum mechanical prediction from the relevant two-particle wave function.
We have also constructed local amplitudes for the Hardy experiment in which quantum mechanics predicts three particular zero joint probabilities are one nonzero joint probability (the other possible joint probabilities in the problem can be nonzero and are not relevant for the demonstration of nonlocality). Local complex amplitudes that reproduce the four relevant joint probabilities can be constructed easily. It is impossible to achieve this if locality at the level of probabilities are assumed, as in a local realistic theory.
Quantum entanglement swapping is understood within this frame work by noting that Bell state measurements choose subensembles of particle pairs that show a particular joint outcome. Particles entangled independently with the pair of particles that are subjected to the Bell state measurement will show a joint outcome consistent with swapped entanglement due to the correlation encoded in the internal variable. But the Bell state measurement does not collapse the distant particle into a definite state. Yet all correlations are correctly reproduced. This has important implication to the interpretation of quantum teleportation.
In summary, the long standing puzzle of nonlocality in the EPR correlations is resolved. There is no nonlocal influence between correlated particles separated into space-like regions. The solution has new physical and philosophical implications regarding the nature of reality, measurement and state reduction in quantum systems. Our approach shows that the EPR paradox of simultaneous reality for noncommuting physical variables arise from their restrictive definition of physical reality.
By restoring locality into the quantum measurements of entangled system and removing the undesirable โspooky action-at-a-distanceโ, one of Einsteinโs deepest wishes is realized. But his desire for a tangible concept of reality of unmeasured quantum systems does not look tenable.
E-mail address: unni@tifr.res.in, unni@iiap.ernet.in |
warning/0001/astro-ph0001298.html | ar5iv | text | # Direct Distance Measurements to Superluminal Radio Sources
## 1 Introduction
To answer fundamental questions about the geometry of the Universe and the distribution of matter on the largest scales we must be able to measure the distance to objects at large redshift. Type Ia supernovae have begun to fill this role and recent results suggest that the overall expansion of the Universe may be accelerating (Riess et al., 1998; Perlmutter et al., 1999). The great brightness of Type Ia supernovae, which can out-shine their parent galaxies, allow them to be observed up to redshifts of $`z=1`$. It is important to have cosmological probes at even higher redshifts where we are less sensitive to local perturbations (Tipler, 1999) and the predictions of different cosmological models become more distinct (Carroll, Press & Turner, 1992).
Quasars are regularly observed at very large redshifts and provide an alternative to gravitational lenses for high redshift cosmology studies. Optically, it is difficult to define a standard candle for quasars because of the wide range of intrinsic luminosities between objects and the high degree of variability of individual objects. Similar difficulties exist with using the radio luminosity for the $``$ 10% of objects that are radio loud. Global Very Long Baseline Interferometry (VLBI) arrays offer an alternative to defining a standard candle. The ability to resolve compact structures and observe proper motions in the radio jets of many sources suggests the possibility of defining a standard rod from which angular size distances may be derived.
Kellermann (1993) defined a standard rod for compact radio sources as the distance between the core and the most distant jet component whose peak brightness exceeds 2% of the core brightness. He used data from 82 compact sources (out to $`z3`$) and found an observed relation between angular size and redshift which is consistent with an Einstein-de Sitter Universe (deceleration parameter, $`q_0=0.5`$).
Lynden-Bell (1977) and Lynden-Bell and Liller (1978) combined early proper motion measurements with a light-echo model, e.g. (Couderc, 1939), in the first attempts to use proper motions observed in radio jets as cosmological distance indicators. Several authors have examined the statistics of proper motions of patterns in radio jets as a cosmological probe. Yahil (1979) proposed the idea of a proper motion-redshift diagram to measure Hubbleโs constant and the deceleration parameter. Cohen et al. (1988) used measured proper motion data and a simple beaming model to show that the upper envelope of the proper motion data decreased with redshift in a manner consistent with a Friedmann cosmology and inconsistent with several alternatives. More recently, Vermeulen and Cohen (1994) have refined these ideas and used a much larger sample to explore the possibility of measuring cosmological parameters with proper motion data. They found they could usefully constrain cosmological parameters and simultaneously learn about the distribution of jet parameters.
Our approach differs from these techniques in that we use VLBI observations to directly measure the distance to individual superluminal radio sources. We compare the proper motions of individual components in a parsec-scale radio jet with measurements of their Doppler factors. In addition, we must constrain the angle the component motion makes with the line of sight and separate the pattern speed (observed in proper motion measurements) from the flow speed (observed in Doppler factor measurements). In ยง2 we present a number of possibilities for constraining these parameters, and in ยง3 we evaluate the technique with an example, using a well defined component in the VLBI jet of the quasar 3C 279. Section 4 discusses the measurement errors and systematic uncertainties associated with this technique. Section 5 explores the application of such measurements to cosmological questions. Our conclusions are presented in ยง6.
## 2 Theory
The observed proper motion, $`\mu `$, of a pattern (component) in a parsec-scale radio jet depends on the intrinsic pattern speed, $`\beta _p`$, the angle between the jet axis and the line to the observer, $`\theta `$, and the angular size distance to the radio source, $`D_A`$.
$$\mu =\frac{c}{D_A(1+z)}\frac{\beta _p\mathrm{sin}\theta }{1\beta _p\mathrm{cos}\theta }=\frac{c}{D_A(1+z)}\beta _a$$
(1)
where $`\beta _a`$ is the apparent transverse velocity of the component in units of the speed of light.
We define a Doppler factor for the pattern, $`\delta _p`$, by
$$\delta _p=\frac{\sqrt{1\beta _p^2}}{1\beta _p\mathrm{cos}\theta }.$$
(2)
The aberrated angle, $`\theta ^{}`$, is the angle between the jet axis and the line to the observer in the frame that moves along the jet with the pattern speed. It is given by
$$\mathrm{cos}\theta ^{}=\frac{\mathrm{cos}\theta \beta _p}{1\beta _p\mathrm{cos}\theta }=\mathrm{cos}\theta \beta _a\mathrm{sin}\theta $$
(3)
and
$$\mathrm{sin}\theta ^{}=\delta _p\mathrm{sin}\theta .$$
(4)
We can use these relations to derive an expression for the angular size distance:
$$D_A=\frac{c}{\mu (1+z)}\left(\frac{\sqrt{\delta _p^2+\mathrm{cos}^2\theta ^{}1}\delta _p\mathrm{cos}\theta ^{}}{\sqrt{1\mathrm{cos}^2\theta ^{}}}\right).$$
(5)
Equation 5 is plotted for various values of $`\delta _p`$ in figure 1. In this equation $`\mu `$ is in natural units of rad/sec; however, in figure 1 $`\mu `$ is in units of mas/year.
The proper motion $`\mu `$ is a directly observed quantity. The Doppler factor of the pattern, $`\delta _p`$, is indirectly observed through the use of Synchrotron Self-Compton (SSC) or equipartition arguments (see ยง2.2 and the appendix) which measure a product of Doppler factors:
$$\delta _{SSC}=\delta _p(\delta _f^{})^{\frac{2\alpha +3}{2\alpha +4}}$$
(6)
and
$$\delta _{eq}=\delta _p(\delta _f^{})^{\frac{2\alpha +5}{2\alpha +6}}$$
(7)
where $`\alpha `$ is the spectral index ($`S\nu ^\alpha `$) for optically thin synchrotron radiation and $`\delta _f^{}`$ is the Doppler factor of the fluid frame relative to the pattern frame:
$$\delta _f^{}=\frac{\sqrt{1(\beta _f^{})^2}}{1\beta _f^{}\mathrm{cos}\theta ^{}}$$
(8)
where $`\theta ^{}`$ is the angle of observation in the pattern frame and $`\beta _f^{}`$ is the speed of the fluid in the pattern frame.
For comparison to proper motion measurements, we are interested in $`\delta _p`$. To determine $`\delta _p`$ from equipartition or SSC arguments we must measure or usefully constrain $`\delta _f^{}`$ which typically requires measuring the pattern versus flow speed; however, there is a useful constraint we can place on $`\delta _f^{}`$ if we know $`\theta ^{}`$. For a given $`\theta ^{}`$, the maximum in $`\delta _f^{}`$ is when $`\beta _f^{}=\mathrm{cos}\theta ^{}`$, so
$$\delta _f^{}1/\sqrt{1(\mathrm{cos}\theta ^{})^2}.$$
(9)
### 2.1 Additional Constraints
In the sections that follow, we explore a number of constraints that allow us to turn measurements of proper motions and Doppler factors into direct distance measurements. Some of these techniques allow measurement or constraint of $`\delta _f^{}`$ which is important for accurate determination of $`\delta _p`$ from Doppler factor measurements.
#### 2.1.1 Bent Jets
For a given pattern speed, $`\beta _p`$, the observed proper motion is maximized when $`\mathrm{cos}\theta =\beta _p`$. At this critical angle, $`\mathrm{cos}\theta ^{}=0`$ and equation 5 reduces to:
$$D_A=\frac{c\sqrt{\delta _p^21}}{\mu (1+z)}$$
(10)
Sources inside the critical angle will give an upper limit on $`D_A`$, and sources outside the critical angle will give a lower limit. However, these limits will only be useful for sources near the critical angle, and for these sources, we need an additional constraint to determine their orientation relative to the critical angle.
Jets which bend on VLBI scales give a unique opportunity for observing a source at or near its critical angle. As a component on a curved trajectory passes through the critical angle, a number of observable effects occur: the proper motion of the component maximizes, the orthogonal component of the magnetic field (projected in the plane of the sky) maximizes creating a maximum or minimum in the observed linear polarization, and thin features, such as shocks, minimize in observed aspect ratio.
We note that if the flow and pattern have different speeds they will also have different critical angles. The maximization of the proper motion occurs at the patternโs critical angle. Any maximum or minimum in the observed linear polarization depends on the flowโs critical angle. Observing both critical angles provides a method of resolving the difference between the flow and pattern speeds. If only the critical angle of the pattern is observed, a useful constraint is that the Doppler factor for the flow as observed from the pattern frame, $`\delta _f^{}`$, has a maximum value of 1 (see equation 9).
#### 2.1.2 Aspect Ratio
Several authors have made use of a sharp (narrow) feature in a radio jet to measure or constrain the jet angle to the line of sight (Eichler & Smith, 1983; Biretta, Owen & Hardee, 1983; Biretta, Owen & Cornwell, 1989; Unwin & Wehrle, 1992). A sharp feature, assumed to be oriented perpendicular to the jet direction, is a sign that the pattern is moving at close to its critical angle. The shape an observer sees for a component moving in a radio jet is governed by the aberration between the pattern frame and observer frame.
The observed aspect ratio of a component is the ratio of its extent along the jet to its extent transverse to the jet , $`\zeta =size_{}/size_{}`$ (see figure 2). The ratio $`\zeta `$ constrains $`\mathrm{cos}\theta ^{}`$ if the pattern is assumed to be axially symmetric and oriented perpendicular to the direction of motion. Under these circumstances, $`\zeta |\mathrm{cos}\theta ^{}|`$ and we obtain potentially useful limits on the angular size distance (from equation 5, corresponding to positive and negative signs for $`\mathrm{cos}\theta ^{}`$):
$$D_A\frac{c}{\mu (1+z)}\left(\frac{\sqrt{\delta _p^2+\zeta ^21}\delta _p\zeta }{\sqrt{1\zeta ^2}}\right)$$
(11)
and
$$D_A\frac{c}{\mu (1+z)}\left(\frac{\sqrt{\delta _p^2+\zeta ^21}+\delta _p\zeta }{\sqrt{1\zeta ^2}}\right)$$
(12)
Because the SSC technique produces only a lower limit on the Doppler factor (Marscher, 1987), only the lower limit on $`D_A`$ will be applicable when we determine $`\delta _p`$ using that technique. The equipartition assumption will also produce a lower limit on the Doppler factor if we only have an upper limit on the frequency of the self-absorption turnover in the synchrotron spectrum (Readhead, 1994).
It is important to note that the relations developed here assume that the pattern is perpendicular to the jet direction and not oblique. Obliqueness in the plane of the sky is observable from the orientation of the feature and perhaps its linear polarization. Assuming it is not very large, this kind of obliqueness can be corrected for; however, if a component can be oblique in the plane of the sky it may also be oblique in the plane of observation. Obliqueness in the plane of observation is indistinguishable from effects of aberration for determining the observed component dimensions and may cause the relation $`\zeta |\mathrm{cos}\theta ^{}|`$ to be violated. Uncertainty in the degree of obliqueness of a given component is equivalent to an added uncertainty in the measurement of $`\zeta `$ (Biretta, Owen & Cornwell, 1989; Unwin & Wehrle, 1992).
#### 2.1.3 Linear Polarization
The linear polarization of a pattern in a radio jet provides a measure of the magnetic field order. For a tangled magnetic field which has been compressed (due perhaps to a propagating shock), the degree of linear polarization observed depends both upon the degree of compression and the viewing angle (Laing, 1980; Hughes, Aller & Aller, 1985). The highest degrees of parallel linear polarization (for a given compression) will be observed when the flow is moving at or near its critical angle. The degree of compression can be related to the speeds of the flow (upstream and downstream) relative to the propagating shock, e.g. (Cawthorne & Wardle, 1988; Hughes, Aller & Aller, 1989).
Wardle et al. (1994) work out a complete model for deducing the jet angle to the line of sight and pattern versus flow speeds from detailed VLBI polarization data. They consider the general case of a compression in a jet with a tangled field plus a component of ordered field along the axis of compression. By measuring the degree of linear polarization and total intensity in both the shocked and un-shocked regions in the jet of 3C 345 (z = 0.595), they were able to constrain the flow speed relative to the shock and the inclination of the jet to the line of sight.
To illustrate the use of the constraints available from linear polarization observations, we will start with the results of Wardle et al. (1994) for jet component C3 of 3C 345. The reader is referred to that paper for details. For their 1984.2 epoch, they find nominal values of $`\theta =2.5^{}/D_A`$, $`\beta _d=0.5`$, and $`\beta _a=11.1D_A(\mu =0.44`$ mas/year). The flow speed of the shocked fluid towards the core in the frame of the shock-front is $`\beta _d`$. In the notation of this paper, $`\beta _f^{}=\beta _d`$. $`D_A`$ is measured in Gpc for the values given above.
We use equations 3 and 7 to calculate $`\mathrm{cos}\theta ^{}=0.5`$ and $`\delta _f^{}=0.7`$ which are essentially independent of distance. If we had observations of the synchrotron self-absorption turnover for component C3 in 3C 345 at this epoch, we could measure its total Doppler factor using equipartition or SSC arguments. We could then use the measurement of $`\delta _f^{}`$ to determine the Doppler factor of the pattern, $`\delta _p`$. With $`\delta _p`$, $`\mu `$, and $`\mathrm{cos}\theta ^{}`$ determined, equation 5 would allow calculation of the distance to 3C 345.
#### 2.1.4 Jet/Counter-Jet Ratio
Another potentially useful constraint is the observed jet/counter-jet brightness ratio, $`R`$, e.g. (Unwin & Wehrle, 1992).
$$R=\left(\frac{1+\beta \mathrm{cos}\theta }{1\beta \mathrm{cos}\theta }\right)^{n+\alpha }=(\beta _a^2+\delta ^2)^{n+\alpha }$$
(13)
where $`n=3`$ for discrete components and $`n=2`$ for continuous jet emission. The reduction of the equation to include $`\beta _a`$ is only valid if the pattern and flow speeds are the same. If they are the same or if we know the relationship between them, measurement of $`R`$ allows us to directly compare the Doppler factor to observed proper motion and deduce the distance to the source. In the event that the pattern and flow speeds are the same, the angular size distance is given by
$$D_A=\frac{c\sqrt{R^{\frac{1}{n+\alpha }}\delta _p^2}}{\mu (1+z)}$$
(14)
An attractive feature of this approach is that the final answer depends weakly on the measurement of $`R`$. For highly beamed sources, however, $`R`$ is huge and even high quality VLBI measurements cannot usefully constrain it. For less beamed sources, measuring or constraining $`R`$ is more promising. A major drawback to using $`R`$ to connect $`\mu `$ and $`\delta _p`$ is that $`R`$ is a global property of the jets rather than of an individual component. For $`R`$ to be useful, the flow speed and angle to the line of sight need to be the same for the jet and counter-jet and constant over the region for which $`R`$ is measured. In addition, there cannot be significant excess absorption of emission from the counter-jet, c.f. (Krichbaum et al., 1998).
### 2.2 Doppler Factors
The synchrotron spectral turnover provides a kind of natural (but broad) spectral line for homogeneous synchrotron sources. By carefully measuring the spectrum of a component and its angular size we can use limits on its SSC x-ray flux, e.g. (Marscher, 1987) or an assumption of equipartition between the field and particle energies (Readhead, 1994) to determine the Doppler factor. In the appendix these formulae are presented for arbitrary homogeneous geometry and for the specific case of a spherical geometry.
We have chosen to use a spherical component geometry for the calculations presented in this paper. Without detailed knowledge of the true geometry a spherical geometry is well-suited for calculation because it computes the angular area presented to the observer reasonably, gives a sensible line of sight through the component, and provides naturally for a range of optical depths across the component. Marscher (1987) suggests using $`\theta _d1.8\sqrt{\theta _{G_a}\theta _{G_b}}`$ to convert Gaussian FWHM dimensions (measured in model-fitting) to spherical diameters, and we adopt his approximation. (See appendix B for discussion of the effect of assumed model geometry on measured Doppler factor.)
## 3 Example: 3C 279
As an example of these ideas, we use a well defined component in the milli-arcsecond jet of the well known blazar 3C 279 at z = 0.536. We observed 3C 279 with the Very Long Baseline Array<sup>1</sup><sup>1</sup>1The VLBA is part of the National Radio Astronomy Observatory, which is a facility of the National Science Foundation operated under cooperative agreement by Associated Universities, Inc. (VLBA) for six epochs at 15 and 22 GHz during 1996 and at four frequencies (5.0, 8.4, 15.4, and 22.2 GHz) during December of 1997. These observations were all calibrated using standard techniques, e.g. (Cotton, 1993; Roberts, Wardle & Brown, 1994), using the National Radio Astronomy Observatoryโs Astronomical Image Processing System (AIPS). Model-fitting was performed in the (u,v)-plane with the Caltech VLBI program, DIFMAP.
### 3.1 Proper Motion
Figure 3 shows the structure of the inner jet of 3C 279 at 22 GHz. Table 1 gives detailed component data from model-fitting the inner jet of 3C 279 for the 1997.94 epoch. K1 is a well defined, strong component which has persisted for years, e.g. (Unwin et al., 1998). The component is located approximately 3 milli-arcseconds from the core at a position angle of $`115^{}`$. Over the course of our observations we observe this component to move radially from the core with a proper motion of $`\mu =0.24\pm 0.01`$ mas/year (see Figure 4). It maintains a structural position angle of $`114^{}\pm 1^{}`$ over the course of our observations.
### 3.2 Synchrotron Self-Absorption <br>Turnover
We fit the total intensity of K1 at all four frequencies in 1997.94 and have measured its spectral turnover. Figure 5 displays the fit of a synchrotron self-absorption spectrum to the data assuming a slab geometry. (Fitting the spectral shape of a homogeneous sphere gives a nearly identical result but a slightly smaller error range on the parameters.) The spectral turnover is located at $`\nu _{peak}=6.02`$ GHz $`(+0.33,0.49)`$ with a flux, $`S_{peak}=4.40`$ Jy $`(+0.19,0.07)`$, and a spectral index, $`\alpha =0.52\pm 0.05`$. The errors in the fit are approximately 1 $`\sigma `$ errors found by a Monte Carlo simulation. The simulation created and fit 1000 fictional data sets using the measured data and assuming the measurements are Gaussian distributed with 1 $`\sigma `$ deviations given by the measured error bars. The error bars on the fluxes were estimated by varying parameters in the model-fits using Jim Lovellโs Difwrap program, an interactive shell for the Caltech VLBI program, DIFMAP. A number of factors were used to gauge the size of the error bars including shape of the chi-square minimum, noise on the residual map, and direct comparison of model and data in the (u,v)-plane. We have no direct way of knowing if the errors estimated for the fluxes are genuinely 1 $`\sigma `$ errors; however, the spectral fit has 1 degree of freedom (4 data points and 3 parameters) so the $`\chi ^2`$ of the spectral fit should be near unity if the errors on the data are 1 $`\sigma `$. The measured $`\chi ^2`$ of the spectral fit is 1.0.
The chief uncertainty in the measured fluxes of K1 is due the presence of its poorly defined โtailโ, fit as component K2. While K1 is fit robustly by a sharp Gaussian component, K2 is more difficult to fit. This becomes more of a problem at the lower frequencies (especially 5 GHz) where K1 is not as well resolved.
To check the spectral fit for K1, we examined the spectrum of observed fractional linear polarization. Figure 6 displays the observed fractional polarization plotted together with a theoretical curve produced by numerical simulation. The simulation is of a homogeneous slab with the same total intensity spectrum as fit to K1. The simulation solved the full equations of polarized transfer, e.g. (Jones & OโDell, 1977), for a completely tangled magnetic field plus a small ordered component. The magnitude of the ordered component was scaled to give the observed fractional polarization at 22 GHz. To simplify the simulation, no internal Faraday rotation was allowed. It is clear that the spectrum of the fractional polarization is completely consistent with total intensity spectrum fit to K1.
### 3.3 Doppler Factor from Equipartition
We now use the measured angular size of K1 at 22 GHz and the fit to the spectral turnover to deduce an equipartition Doppler factor (derived in the appendix). The measured FWHM angular size at 22 GHz is $`\theta _{Ga}\times \theta _{Gb}=0.46(\pm 0.02)\times 0.20(\pm 0.01)`$ mas$`\times `$mas. The error bars were estimated by varying parameters in the model-fit and by comparison to the measurements at 15 GHz and 8 GHz. We use the Doppler factor formulation for a homogeneous sphere (equation A6) with $`\theta _d1.8\sqrt{\theta _{G_a}\theta _{G_b}}`$. So for K1, $`\theta _d=0.55\pm 0.02`$ mas and we obtain an equipartition Doppler factor for the pattern of
$$\delta _p=19.1_{2.9}^{+5.9}\left(\frac{\eta }{D_A}\right)^{1/7}(\delta _f^{})^{6/7}$$
(15)
where $`\eta =U_B/U_{rp}`$ is the equipartition factor ($`U_B=`$ magnetic field energy density; $`U_{rp}=`$ energy density in the radiating particles), $`D_A`$ is the angular size distance in Gpc, and $`\delta _f^{}`$ is the Doppler factor of the flow as viewed by an observer co-moving with the pattern. To do this computation, we have assumed energy spectrum limits of $`\gamma _1=10`$ and $`\gamma _2=1\times 10^6`$. For $`\alpha 0.5`$, the dependence on these limits is approximately $`[\mathrm{ln}(\gamma _2/\gamma _1)]^{1/7}`$.
It is interesting to compare this Doppler factor for K1 to a Doppler factor measured for the component K4. The spectral fit for K4 is given in Figure 7, we find $`\nu _{peak}=12.59`$ GHz $`(+0.70,0.41)`$, $`S_{peak}=10.86`$ Jy $`(+0.40,0.37)`$, and $`\alpha =0.70`$ $`(+0.17,0.16)`$ with $`\chi ^2=0.8`$. We have only an upper limit on the angular size of the component transverse to the jet direction, so we can only use the limit $`\theta _d0.42\pm 0.01`$. We calculate an equipartition pattern Doppler factor,
$$\delta _p16.7_{2.1}^{+3.6}\left(\frac{\eta }{D_A}\right)^{1/7.4}(\delta _f^{})^{6.4/7.4}$$
(16)
for the component K4. The dependence on the energy spectrum limits for $`\alpha 0.7`$ is approximately $`\gamma _1^{0.4/7.4}`$. Ghisellini et al. (1993) report an SSC Doppler factor for the core of 3C 279 of $`\delta _{SSC}18.0`$.
### 3.4 Aspect Ratio
K1 is a narrow component oriented perpendicular to its position angle. At 15 and 22 GHz the component is well resolved in both directions, at 8 GHz it is less well resolved, and at 5 GHz the component is unresolved along the direction of the jet. Assuming that the component is not oblique, the observed aspect ratio is $`\zeta =0.43\pm 0.03`$.
Component K1 shows no sign of significant obliqueness. At 22 GHz we measure its orientation to be $`7^{}\pm 11^{}`$ from perpendicular to its long term structural position angle. The high frequency linear polarization of K1 is aligned with its long term structural position angle to $`3^{}\pm 6^{}`$. (The overall calibration of the polarization position angle for our VLBA observations in Dec. 1997 was from simultaneous VLA observations of the compact source OJ287.) Using these estimates on the obliqueness in the plane of the sky as a guide, we estimate an uncertainty in the obliqueness in the plane of observation of $`\pm 5^{}`$. This uncertainty in the degree of obliqueness translates to an additional uncertainty in the measured aspect ratio, roughly $`\zeta =0.43\pm 0.08`$.
### 3.5 Measuring the Distance
Using our measurement of the observed proper motion, pattern Doppler factor, and aspect ratio of component K1, equation 11 gives the following limit on the angular size distance to 3C 279:
$$D_A1.8_{0.3}^{+0.5}\eta ^{1/8}Gpc$$
(17)
which depends only on the equipartition factor, $`\eta `$. Note that we have used equation 9 and our measurement of $`\zeta `$ ($`|\mathrm{cos}\theta ^{}|`$) to limit the Doppler factor of the flow relative to the pattern frame to $`\delta _f^{}1.1\pm 0.1`$. (Because we have only an upper limit on $`\delta _f^{}`$, the upper limit on the distance (equation 12) is undetermined.)
## 4 Discussion
Equation 17 provides only a lower limit on the angular size distance to 3C 279. The result is a limit because we could only make use of the aspect ratio constraint. In general this technique can provide direct measurements (not just limits) for sources where some of the other constraints in Section 2.1 can be applied successfully.
The 1 $`\sigma `$ errors on this limit are $`+28\%`$ and $`17\%`$. These errors are dominated by the uncertainty in the spectral turnover measurement of component K1. For K1, we have only one spectral point on the optically thick side of the turnover and this point is poorly constrained due reduced resolution at 5 GHz. The spectral turnover for component K4 is better determined giving a Doppler factor with 1 $`\sigma `$ errors of $`+22\%`$ and $`13\%`$. With better frequency coverage (perhaps by using widely separated IF channels near the spectral turnover) and better angular resolution at lower frequencies (through the use of space VLBI), we believe we can eventually reduce the 1 $`\sigma `$ measurement errors on Doppler factors from equipartition and SSC techniques to $`1015\%`$.
For our 3C 279 distance limit, the equipartition factor, $`\eta `$, is the most significant unknown quantity. Even though $`\eta `$ enters to only a small factor, it is poorly constrained. Readhead (1994), when proposing the technique, argued that sources should be near equipartition ($`\eta =1`$ for electron-positron jets) and suggested an error of $`13\%`$ in the Doppler factor for typical departures from equipartition. Singal (1986) calculated the diamagnetic effect of spiraling electrons in a magnetic field. He found that the energy density of the electrons could not exceed 6 times the energy of the magnetic field and still maintain synchrotron radiation. Bobo, Ghisellini & Trussoni (1992) repeated Singalโs calculation and included a surface current term. They found that the energy density of the electrons could not exceed the energy density of the applied field by more than a factor of 3 although they note that the energy density of the effective field in the region could be much smaller than the applied field.
Another issue important to calculating precise values for equipartition Doppler factors is the cutoffs in the power law particle energy spectrum. The computed Doppler factors depend only weakly on the assumed value of the cutoffs, but different, reasonable assumptions for the cutoffs may lead to a $`510\%`$ uncertainty in the computed Doppler factor. Energy spectrum cutoffs can be measured or constrained, however. In Wardle et al. (1998) we used VLBI circular polarization observations to show the lower energy cutoff in 3C 279 was $`\gamma _120`$. The uncertainty in the low energy cutoff dominates when $`\alpha >0.5`$.
One way around the uncertainties in assuming equipartition and energy spectrum cutoffs is to calculate Doppler factors using measured x-ray fluxes. A drawback to this approach is that the SSC Doppler factor calculated for a given component is only a lower limit because, with the capabilities of current instruments, observed x-ray fluxes include contributions from all parts of the parsec-scale source. Such a limit is unlikely to be useful for jet components like K1 in 3C 279 which contribute a very small fraction of the total x-ray flux of the source. Assuming $`\eta =1`$ and inverting equation A9 to solve for the x-ray flux of K1 yields $`0.01\mu `$Jy of 2 KeV x-rays. This flux is less than $`1\%`$ of the total 2 KeV x-ray flux reported by Wehrle et al. (1998) for 3C 279 in its quiescent state in January of 1996. It is interesting to note that the capabilities of the proposed MAXIM program (http://maxim.gsfc.nasa.gov/) would make x-ray observation of individual jet components possible.
The final area of systematic uncertainty is the assumed geometry for the pattern. We found that we obtained essentially identical spectral fits when using the functional form for a uniform slab as for a uniform sphere, so we can safely say that the assumed geometry has little affect on the spectral fit. In appendix B we explore the remaining dependence of derived Doppler factors on assumed pattern geometry. The main result is that a spherical geometry should be a good approximation for calculational purposes and tends to produce a lower limit on the Doppler factor if the true geometry is non-spherical.
## 5 Application to Cosmology
The angular size distance in terms of redshift, $`z`$, Hubble constant, $`H_0`$, matter density, $`\mathrm{\Omega }_M`$, and cosmological constant, $`\mathrm{\Omega }_\mathrm{\Lambda }=\mathrm{\Lambda }/(3H_0^2)`$ is given by, e.g. (Carroll, Press & Turner, 1992)
$$\begin{array}{c}D_A=\frac{c}{H_0\sqrt{|\kappa |}(1+z)}\hfill \\ \hfill \times ๐(\sqrt{|\kappa |}_0^z[(1+z^{})^2(1+\mathrm{\Omega }_Mz^{})\\ \hfill z^{}(2+z^{})\mathrm{\Omega }_\mathrm{\Lambda }]^{1/2}dz^{})\end{array}$$
(18)
where if $`\mathrm{\Omega }_M+\mathrm{\Omega }_\mathrm{\Lambda }>1`$ then $`๐(x)=\mathrm{sin}(x)`$ and $`\kappa =1\mathrm{\Omega }_M\mathrm{\Omega }_\mathrm{\Lambda }`$, if $`\mathrm{\Omega }_M+\mathrm{\Omega }_\mathrm{\Lambda }<1`$ then $`๐(x)=\mathrm{sinh}(x)`$ and $`\kappa =1\mathrm{\Omega }_M\mathrm{\Omega }_\mathrm{\Lambda }`$, and if $`\mathrm{\Omega }_M+\mathrm{\Omega }_\mathrm{\Lambda }=1`$ then $`๐(x)=x`$ and $`\kappa =1`$.
In figure 8 we plot our lower limit on the distance to 3C 279 (assuming $`\eta =1`$) against two cosmological models for a flat universe ($`\mathrm{\Omega }_M+\mathrm{\Omega }_\mathrm{\Lambda }=1`$). The first case is consistent with recent type Ia supernovae results (Perlmutter et al., 1999; Riess et al., 1998). The second case is the standard Einstein-de Sitter universe. The Hubble constant is maintained as a free parameter which scales the ordinate of figure 8.
Our single distance limit cannot distinguish between these cosmological models; however, we can investigate whether this technique holds promise for distinguishing these cases in the future. At redshifts larger than $`1.0`$ these cosmological models differ by as much as $`3040`$%. Superluminal radio sources are regularly observed at these large redshifts. For 3C 279 our lower limit has 1 $`\sigma `$ measurement errors on the order of $`2025\%`$ and we believe it is reasonable obtain distance measures (or limits) good to $`1015\%`$ for carefully planned observations on well selected objects. Even a handful objects (over a range of redshift values) could provide strong constraints on cosmological models.
The systematic uncertainties (discussed in ยง4) in measuring the Doppler factor present the largest difficulty here. We can constrain the range of allowed values for the equipartition factor, $`\eta `$, by comparing distance measurements to sources at similar redshifts. To detect any systematic offset in $`\eta `$ from unity, we must have a way of calibrating our Doppler factor measurements. Fortunately, all of the poorly constrained quantities ($`\eta `$, energy spectrum cutoffs, geometric dependence) can be grouped as a single multiplicative parameter in the equipartition Doppler factor. We can calibrate any systematic offset in this parameter by making distance measurements to sources at moderate redshifts (where the effects of differing cosmological models are not strong) and comparing the result to other distance measurement techniques. Such a scheme would introduce a dependence on the cosmological distance ladder, so we should actively seek other techniques for calibrating systematic effects in Doppler factor measurements.
## 6 Conclusions
We have demonstrated a new technique for directly measuring the distances to high redshift, superluminal radio sources. This technique involves the comparison of Doppler factor and proper motion measurements for individual jet components; we must also determine the jet angle to the line of sight and pattern versus flow velocity. We have presented several techniques for measuring or usefully constraining these parameters. In general these techniques will be applicable only to selected sources which have jet components with the right characteristics; however, with hundreds of currently known superluminal sources and new jet components emerging frequently from many of them, it seems reasonable to assume that we will find a significant number of candidates. One interesting possibility is that some sources could have more than one component to which we can apply these techniques, giving us multiple, independent distance determinations to the same object.
To begin to usefully constrain cosmological parameters we need to obtain high quality distance measurements or limits to several sources over a range of redshift. We have performed a detailed analysis of the measurement error associated with our limit on the distance to 3C 279. We found the measurement error was $`2025`$% and concluded that carefully planned, high-quality observations could reduce measurement error to $`1015`$%. The systematic error was more difficult to quantify, although we outlined the sources of systematic error and presented rough estimates. Applying this technique to a larger sample of sources will be important not only for probing cosmological parameters but also for investigating the sources of systematic error and how close these sources are to equipartition.
## 7 Acknowledgments
This work has been supported by NASA Grants NGT-51658 and NGT5-50136 and NSF Grants AST 92-24848, AST 95-29228, and AST 98-02708.
## Appendix A Doppler Factor Formulae
While the relativistic plasma in radio jets does not contain atoms and molecules whose characteristic spectra we could use to directly measure their bulk Doppler factors, they do contain a unique (but broad) spectral feature: the synchrotron self-absorption turnover. The location of the turnover due to synchrotron self-absorption depends on the magnetic field strength, particle density, and size of the emission region. For a volume of homogeneous plasma, we can use the extrapolated optically thin flux at the turnover frequency and the optical depth at the turnover frequency to solve for the magnetic field and the particle density, e.g. (Marscher, 1987).
$$B=\frac{\delta _p\delta _f^{}}{(1+z)}C_3\tau _m^2\nu _m^5S_m^2\mathrm{\Omega }^2$$
(A1)
and
$$K=\frac{(1+z)^{4+2\alpha }}{\delta _p^{4+2\alpha }(\delta _f^{})^{3+2\alpha }}C_4\tau _m^{(2\alpha +2)}\nu _m^{(4\alpha +5)}S_m^{2\alpha +3}\mathrm{\Omega }^{(2\alpha +3)}(D_A\xi _c)^1$$
(A2)
where $`K`$ is the constant in the power law particle density, $`N_\gamma d\gamma =K\gamma ^{(2\alpha +1)}d\gamma `$. The sign of the spectral index, $`\alpha `$, for optically thin emission is given by $`S\nu ^\alpha `$. The parameter $`\mathrm{\Omega }=\left[\frac{V^{}}{D_A^2s_c^{}}\right]`$ and $`\xi _c=s_c^{}/D_A`$, where $`V^{}`$ is the volume in the pattern frame and $`s_c^{}`$ is the line of sight through the center of the volume along which the optical depth at the turnover, $`\tau _m`$, is defined. The extrapolated optically thin flux<sup>2</sup><sup>2</sup>2The factor by which $`S_m`$ over-predicts the observed peak flux, $`S_o`$, is tabulated in table 2 for a spherical geometry. at the turnover frequency, $`\nu _m`$, is given by $`S_m`$. Constants $`C_3`$ and $`C_4`$ are tabulated in table 2 for $`\mathrm{\Omega }`$ in mas<sup>2</sup>, $`\nu _m`$ in GHz, $`S_m`$ in Jy, $`D_A`$ in Gpc, and $`\xi _c`$ in mas.
Assuming we can translate the source dimensions, $`V^{}`$ and $`s_c^{}`$, into observable quantities (e.g. observed angular size), these expressions provide us with two equations and three unknowns, $`B`$, $`K`$, and the Doppler factor. To solve for these quantities, it is necessary to have a third constraint. Two possibilities for a third constraint are equipartition (Readhead, 1994) and Synchrotron Self-Compton (SSC) x-ray flux (Marscher, 1987).
### A.1 Equipartition
Equipartition assumes that the energy density of the magnetic field and the energy density of the particles are equal. We will parameterize the relationship between the energy density of the magnetic field and the energy density in the radiating particles:
$$U_B=\eta U_{rp}$$
(A3)
where $`U_B=B^2/8\pi `$, $`U_{rp}=mc^2\gamma N_\gamma ๐\gamma `$. For an electron-positron jet, $`\eta =1`$ for equipartition. In an electron-proton jet, the value of $`\eta `$ for equipartition will depend on the details of the particle acceleration within the jet. (See ยง4 for discussion of the value of $`\eta `$.)
The equipartition condition is
$$B^2/8\pi =\eta mc^2g(\alpha ,\gamma _1,\gamma _2)K,$$
(A4)
where for $`\alpha 0.5`$
$$g(\alpha ,\gamma _1,\gamma _2)=\frac{1}{2\alpha 1}\left(\gamma _1^{(2\alpha 1)}\gamma _2^{(2\alpha 1)}\right)$$
and for $`\alpha =0.5`$
$$g(\alpha ,\gamma _1,\gamma _2)=\mathrm{ln}\left(\frac{\gamma _2}{\gamma _1}\right).$$
$`\gamma _1`$ and $`\gamma _2`$ represent the lower and upper cutoffs for the particle energy distribution.
Substituting the expressions for $`B`$ and $`K`$ into the equipartition condition and solving for the Doppler factor yields
$$\delta _{eq}=\delta _p(\delta _f^{})^{\frac{2\alpha +5}{2\alpha +6}}=F_{eq}\tau _m^1(1+z)\left[\frac{g(\alpha ,\gamma _1,\gamma _2)\eta S_m^{2\alpha +7}}{D_A\xi _c\mathrm{\Omega }^{2\alpha +7}\nu _m^{4\alpha +15}}\right]^{\frac{1}{2\alpha +6}}$$
(A5)
where $`F_{eq}`$ is tabulated in table 2 for $`\mathrm{\Omega }`$ in mas<sup>2</sup>, $`\nu _m`$ in GHz, $`S_m`$ in Jy, $`D_A`$ in Gpc, and $`\xi _c`$ in mas.
##### Homogeneous Sphere Geometry:
The spherical geometry assumes that the components emitted from the AGN are homogeneous spheres of radiating plasma. The sphere has a radius, $`R`$, and we can define an angular diameter, $`\theta _d=2R/D_A`$. Therefore $`\mathrm{\Omega }=(\pi /6)\theta _d^2`$ and $`\xi _c=\theta _d`$. With these identifications, the equipartition Doppler factor is given by
$$\delta _{eq}=F_{eq}\tau _m^1(1+z)\left(\frac{6}{\pi }\right)^{\frac{2\alpha +7}{2\alpha +6}}\left[\frac{g(\alpha ,\gamma _1,\gamma _2)\eta S_m^{2\alpha +7}}{D_A(\theta _d\nu _m)^{4\alpha +15}}\right]^{\frac{1}{2\alpha +6}}.$$
(A6)
The equipartition Doppler factor derived by Readhead (1994) has a slightly different functional form than our expression. Readheadโs expression is derived by comparing observed brightness temperature, $`T_b^{}`$, to an (rest-frame) equipartition brightness temperature, $`T_{eq}`$: $`\delta _{eq}=T_b^{}/T_{eq}`$. For 3C 279 we can turn this expression around and calculate $`T_{eq}`$ for component K1 from our measured $`\delta _{eq}`$ and observed brightness temperature. We find $`T_{eq}=4\times 10^{10}K`$ (for $`\eta =1`$) which is close to $`T_{eq}5\times 10^{10}K`$ which Readhead argues is a typical upper cutoff for powerful extra-galactic radio sources in their rest-frame.
### A.2 Synchrotron Self-Compton Emission
A second possible constraint is synchrotron self-Compton x-ray flux, e.g. (Marscher, 1987). The observed x-ray flux from the SSC process is given by (adapted from Rybicki & Lightman (1979))
$$S_X(\nu _c)=\mathrm{\Xi }\frac{3\sigma _T}{8}KA(p)l^{\prime \prime }S_m\left(\frac{\nu _m}{\nu _c}\right)^\alpha \mathrm{ln}\left[\frac{\nu _b}{\nu _m}\right]$$
(A7)
where
$$A(p)=2^{p+3}\frac{p^2+4p+11}{(p+3)^2(p+5)(p+1)}p=2\alpha +1$$
and $`\mathrm{\Xi }`$ is a factor ($`1`$) that accounts for the differences in photon number density (resulting from edge effects) within the emitting volume, $`l^{\prime \prime }`$ is the average line of sight as seen from the center of the geometry in the fluid frame, $`\nu _b`$ is the upper cutoff frequency of the synchrotron emission spectrum, and $`\nu _c`$ is the x-ray observation frequency. As defined earlier, $`S_m`$ is the extrapolated, optically thin synchrotron flux at the turnover frequency, $`\nu _m`$.
Substituting for $`K`$ and solving for the Doppler factor gives
$$\delta _{SSC}=\delta _p(\delta _f^{})^{\frac{2\alpha +3}{2\alpha +4}}=F_{SSC}S_m(1+z)\left[\frac{\mathrm{\Xi }l^{\prime \prime }}{s_c^{}}\frac{\mathrm{ln}\left[\frac{\nu _b}{\nu _m}\right]\nu _m^{(3\alpha +5)}\tau _m^{(2\alpha +2)}}{S_X(h\nu _c)_{KeV}^\alpha \mathrm{\Omega }^{2\alpha +3}}\right]^{\frac{1}{2\alpha +4}}$$
(A8)
where $`F_{SSC}`$ is a constant which is tabulated in table 2 for $`\mathrm{\Omega }`$ in mas<sup>2</sup>, $`\nu _m`$ in GHz, $`S_m`$ in Jy, and $`S_X`$ in $`\mu `$Jy.
It is important to note that, in practical application, the SSC Doppler factor is only a lower limit. With the capabilities of current instruments, observed x-ray fluxes include contributions from all parts of the parsec scale jet, thermal x-ray emission from the accretion disk, and inverse-Compton x-rays which are not the result of the SSC process. We therefore obtain only an upper limit on the SSC x-ray flux of a given component and thus a lower limit on its Doppler factor.
##### Homogeneous Sphere Geometry:
$`\mathrm{\Omega }=(\pi /6)\theta _d^2`$ and $`s_c^{}=2R`$. Also, for a spherical geometry<sup>3</sup><sup>3</sup>3This is strictly true only if the pattern and flow are moving with the same speed. $`l^{\prime \prime }=R`$ and $`\mathrm{\Xi }=3/4`$ (Gould, 1979). Making these substitutions gives the following expression for the SSC Doppler factor:
$$\delta _{SSC}=F_{SSC}S_m(1+z)\left(\frac{6}{\pi }\right)^{\frac{2\alpha +3}{2\alpha +4}}\left[\frac{3}{8}\frac{\mathrm{ln}\left[\frac{\nu _b}{\nu _m}\right]\nu _m^{(3\alpha +5)}\tau _m^{(2\alpha +2)}}{S_X(h\nu _c)_{KeV}^\alpha \theta _d^{4\alpha +6}}\right]^{\frac{1}{2\alpha +4}}.$$
(A9)
## Appendix B Choice of Model Geometry for Doppler Factor Measurements
Differences between model geometries show up in the range of optical depths across the component (which affects both $`\tau _m`$ and the ratio $`S_m/S_o`$) and conversion of measured Gaussian FWHM diameters, $`\theta _G`$, to the angular dimensions of the assumed geometry. These conversion factors can be estimated for simple geometries by matching the second moment of their Fourier transforms which are fit in the (u,v)-plane by Gaussian models. For a uniformly bright disk, $`\theta _d1.7\theta _G`$. For a homogeneous sphere, $`\theta _d1.9\theta _G`$.
For the purposes of measuring the Doppler factor, the effect of having a wide range of optical depths for a homogeneous sphere nearly offsets the larger conversion factor for Gaussian measured component dimensions. If we assume that a given component is a sphere, when in reality it is a uniformly bright disk, we will calculate a Doppler factor that is about $`10\%`$ too small using the equipartition formula or about $`5\%`$ too small using the SSC formula. In this scenario there is an additional factor by which assuming a sphere will under-predict the Doppler factor. This factor is due to the unknown physical depth of the uniformly bright disk. For the equipartition formula, this factor comes in as $`(\theta _d/\xi )^{1/7}`$, where $`\xi `$ is the angular thickness of the disk. For the SSC formula, this factor is $`\left[(8/3)(l^{\prime \prime }/s^{})\mathrm{\Xi }\right]^{1/5}`$, where $`l^{\prime \prime }`$ is the mean line of sight as seen by a photon at the center of the disk (in the fluid frame), $`s^{}`$ is the physical depth of the disk (pattern frame), and $`\mathrm{\Xi }1`$ accounts for differences in photon density throughout the disk.
In general, the patterns we observe are likely to be some compromise between these geometries, perhaps a cylindrical disk viewed nearly edge on. Without detailed knowledge of the true geometry a spherical geometry is well-suited for calculation because it computes the angular area presented to the observer reasonably, gives a sensible line of sight through the component, and provides naturally for a range of optical depths across the component. Marscher (1987) suggests using $`\theta _d1.8\sqrt{\theta _{G_a}\theta _{G_b}}`$ to convert Gaussian FWHM dimensions to spherical diameters, and we have adopted his approximation. |
warning/0001/physics0001045.html | ar5iv | text | # Large-scale Breit-Pauli R-matrix calculations for transition probabilities of Fe V
## 1. Introduction
Transition probabilities of heavy elements, particularly the iron group, are of great importance in astrophysical and laboratory sources. Fuhr et al. have compiled data from a number of available sources. However, the accuracy and the extent of these data is largely inadequate for many general applications such as the calculation of local thermodynamic equilibrium (LTE) stellar opacities , and radiative levitation and accelerations of heavy elements . Among the particular applications including Fe V as a prominent spectral constituent are the the non-LTE models of Fe V spectra in hot stars , and the observed extreme ultraviolet Fe V emission from young white dwarfs . For example, currently available data for Fe V fails to account for the observed opacity of iron in the XUV region where observations of newly formed hot and young white dwarfs clearly show Fe V lines . In all of these applications it is highly desirable to have as complete a dataset of radiative transition probabilities as possible. While the twin problems of completeness and accuracy pose a challenge to the theoretical methods, they are of interest not only in various applications but may also be of use in the analysis of experimental measurements of observed energy levels of complex atomic systems from the iron group.
The Opacity Project (OP) and the Iron Project (IP) laid the foundation for large-scale theoretical calculations using ab intio methods. The R-matrix method , based on atomic collision theory techniques and adapted for the OP and the IP , has proven to be very efficient for these calculations. Whereas the OP calculations were all in the LS coupling approximation, with no relativistic effects included, the subsequent IP work is in intermediate coupling using the Breit-Pauli extension of the R-matrix method . While most the IP work has concentrated on collisional calculations, recent works have extended the BPRM method to radiative bound-bound and bound-free calculations for transition probabilities , photoionization , and (electro-ion) recombination . The first comprehensive BPRM calculation of fine structure transition probabilities was carried out for the highly charged ions Fe XXIV and Fe XXV that are of special interest in X-ray astronomy. Very good agreement was found with existing results available for a limited number of transitions but using very accurate theoretical methods including relativistic and QED effects , thus establishing the achievable accuracy for the BPRM calculations. However those He-like and Li-like atomic systems are relatively simple, and the electron correlation effects relatively weak, compared to the low ionization stages of iron group elements. The present work attempts to enlarge the scope of the possible BPRM calculations to include the iron group elements, as well as to solve some outstanding problems related to level identifications in ab initio theoretical calculations using collision theory methods.
Unlike atomic structure calculations, where the electronic configurations are pre-specified and the levels identified, the bound levels calculated by collision theory methods adopted in the OP and the IP need to be identified since only the channel quantum numbers are known for the bound states corresponding to the (e + ion) Hamiltonian of a given total angular and spin symmetry $`SL\pi `$ or $`J\pi `$. The precise correspondence between the channels of the collision complex, and the bound levels, must therefore be determined. The problem is non-trivial for complex atoms and ions with many highly mixed levels due to configuration interaction. In the OP work, carried out in LS coupling, this problem was solved by an analysis based on quantum defects and the numerical components of wavefunctions in the region outside the R-matrix boundary (that envelops the target ion orbitals). The present work extends that treatment to the analysis of fine structure levels computed in intermediate coupling. In addition, considerable effort is devoted to the determination of the completeness of the set of computed bound levels; comparison with the expected levels derived from all possible combination of angular and spin quantum numbers reveals the missing levels. The general procedure could be applied to spectroscopic measurements and the analysis of observed levels of a given atomic system by comparison with the theoretical predictions.
## 2. Theory
The general theory for the calculation of bound states in the close coupling (CC) approximation of atomic collision theory, using the R-matrix method, is described by Burke and Seaton and Seaton . The application to the Opacity Project work is described by Seaton , Berrington et al. , and Seaton et al. . The relativistic extensions of the R-matrix method in the Breit-Pauli approximation are discussed by Scott and Taylor , and the computational details by Berrington, Eissner, and Norrington . The application to the Iron Project work is outlined in Hummer et al. .
In the present work we describe the salient features of the theory and computations as they pertain to large-scale BPRM calculations for complex atomic systems. Identification of fine structure energy levels is discussed in detail.
Following standard collision theory nomenclature, we refer to the (e + ion) complex in terms of the โtargetโ ion, with N bound electrons, and a โfreeโ electron that may be either bound or continuum. The total energy of the system is either negative or positive; negative eigenvalues of the (N + 1)-electron Hamiltonian correspond to bound states of the (e + ion) system. In the coupled channel or close coupling (CC) approximation the wavefunction expansion, $`\mathrm{\Psi }(E)`$, for a total spin and angular symmetry $`SL\pi `$ or $`J\pi `$, of the (N+1) electron system is represented in terms of the target ion states as:
$$\mathrm{\Psi }_E(e+ion)=A\underset{i}{}\chi _i(ion)\theta _i+\underset{j}{}c_j\mathrm{\Phi }_j,$$
(1)
where $`\chi _i`$ is the target ion wave function in a specific state $`S_iL_i\pi _i`$ or level $`J_i\pi _i`$, and $`\theta _i`$ is the wave function for the (N+1)th electron in a channel labeled as $`S_iL_i(J_i)\pi _ik_i^2\mathrm{}_i(SL\pi )[J\pi ]`$; $`k_i^2(=ฯต_i)`$ is the incident kinetic energy. In the second sum the $`\mathrm{\Phi }_j`$โs are correlation wavefunctions of the (N+1) electron system that (a) compensate for the orthogonality conditions between the continuum and the bound orbitals, and (b) represent additional short-range correlation that is often of crucial importance in scattering and radiative CC calculations for each $`SL\pi `$.
The functions $`\mathrm{\Psi }(E)`$ are given by the R-matrix method in an inner region $`ra`$. These are bounded at the origin and contain radial functions that satisfy a logarithmic boundary condition at $`r=a`$ . In the outer region $`r>a`$ the inner region functions are matched to a set of linearly independent functions that correspond to all possible (e + ion) channels of a given symmetry $`SL\pi `$ or $`J\pi `$. The outer region wavefunctions are computed for all channels, $`(C_tS_tL_t\pi _t)ฯตl`$, where $`C_t`$ is the target configuration, and used to determine the individual channel contributions (called โchannel weightsโ).
In the relativistic BPRM calculations the set of $`SL\pi `$ are recoupled to obtain (e + ion) levels with total $`J\pi `$, followed by diagonalisation of the (N+1)-electron Hamiltonian,
$$H_{N+1}^{BP}\mathrm{\Psi }=\mathrm{E}\mathrm{\Psi }.$$
(2)
The BP Hamiltonian is
$$H_{N+1}^{\mathrm{BP}}=H_{N+1}+H_{N+1}^{\mathrm{mass}}+H_{N+1}^{\mathrm{Dar}}+H_{N+1}^{\mathrm{so}},$$
(3)
where $`H_{N+1}`$ is the nonrelativistic Hamiltonian,
$$H_{N+1}=\underset{i=1}{\overset{N+1}{}}\left\{_i^2\frac{2Z}{r_i}+\underset{j>i}{\overset{N+1}{}}\frac{2}{r_{ij}}\right\},$$
(4)
and the additional terms are the one-body terms, the mass correction term, the Darwin term and the spin-orbit term respectively. Spin-orbit interaction, $`H_{N+1}^{so}`$, splits the LS terms into fine-structure levels labeled by $`J\pi `$, where $`J`$ is the total angular momentum. Other terms of the Breit-interaction ,
$$H^B=\underset{i>j}{}[g_{ij}(\mathrm{so}+\mathrm{so}^{})+g_{ij}(\mathrm{ss}^{})],$$
(5)
representing the two-body spin-spin and the spin-other-orbit interactions are not included.
The positive and negative energy states (Eq. 1) define continuum or bound (e + ion) states,
$$\begin{array}{c}E=k^2>0continuum(scattering)channel\hfill \\ E=\frac{z^2}{\nu ^2}<0boundstate,\hfill \end{array}$$
(6)
where $`\nu `$ is the effective quantum number relative to the core level. If $`E<`$ 0 then all continuum channels are โclosedโ and the solutions represent bound states. Determination of the quantum defect ($`\mu (\mathrm{}))`$, defined as $`\nu _i=n\mu (\mathrm{})`$ where $`\nu _i`$ is relative to the core level $`S_iL_i\pi _i`$, is helpful in establishing the $`\mathrm{}`$-value associated with a given channel (level).
At E $`<`$ 0 a scattering channel may represent a bound state at the proper eigenvalue of the Hamiltonian (Eq. 2). A large number of channels are considered for the radiative processes of Fe V. Each SL$`\pi `$ or J$`\pi `$ symmetry is treated independently and corresponds to a large number of channels. Therefore, the overall configuration interaction included in the total (e + ion) wavefunction expansion is quite extensive. This is the main advantage of the CC method in representing electron correlation accurately.
### a) Level identification and coupling schemes
The BPRM calculations in intermediate coupling employ the pair-coupling representation
$$\begin{array}{c}S_i+L_iJ_i\hfill \\ J_i+\mathrm{}K\hfill \\ K+\mathrm{s}J\hfill \end{array},$$
(7)
where the โiโ refers to the target ion level and $`\mathrm{},s`$ are the orbital angular momemtum (partial wave) and spin of the additional electron. According to designations of a collision complex, a channel is fully specified by the quantum numbers
$$(S_iL_iJ_i)\pi _iฯต_i\mathrm{}_iKs[J\pi ]$$
(8)
The main problem in identification of the fine structure levels stems from the fact that the bound levels are initially given only as eigenvalues of the (e + ion) Hamiltonian of a given symmetry $`J\pi `$. Each level therefore needs to be associated with the quantum numbers characterizing a given collision channel. Subsequently, three main parameters are to be determined: (i) the parent or the target ion level, (ii) the orbital, effective and principal quantum numbers $`(l,\nu ,n)`$ of the (N+1)th electron, and (iii) the symmetry, $`SL\pi `$. The task is relatively straightforward for simple few-electron atomic systems. For example, in a recent work Nahar and Pradhan have calculated a large number of transition probabilities for Li-like Fe XXIV and He-like Fe XXV, where the problem of level identification is trivial, compared to the present work, since the bound levels are well separated in energy and in $`\nu `$. However when a number of mixed bound levels fall within a given interval $`(\nu ,\nu +1)`$, for the same $`J\pi `$, the quantum numbers and the magnitude of the components in all associated channels must be analysed. A scheme for identification of levels is developed (discussed later) that rests mainly on an analysis of quantum defects of the bound levels and their orbital angular momenta, and the percentage of the total wavefunction in all channels of a given $`J\pi `$.
Following level identification, further work is needed to enable a direct correspondence with standard spectroscopic designations that follow different coupling schemes, such as between $`LS`$ and $`JJ`$, appropriate for atomic structure calculations as, for example, in the NIST tables of observed energy levels . The correspondence provides the check for completeness of calculated set of levels or the levels missing. The level identification procedure involves considerable manipulation of the bound level data and, although it has been encoded for general applications, still requires analysis and interpretation of problem cases of highly mixed levels that are difficult to identify.
### b) Oscillator strengths and transition probabilities
The oscillator strength (or photoionization cross section) is proportional to the generalized line strength defined, in either length form or velocity form, by the equations
$$S_\mathrm{L}=\left|\mathrm{\Psi }_f|\underset{j=1}{\overset{N+1}{}}r_j|\mathrm{\Psi }_i\right|^2$$
(9)
and
$$S_\mathrm{V}=\omega ^2\left|\mathrm{\Psi }_f|\underset{j=1}{\overset{N+1}{}}\frac{}{r_j}|\mathrm{\Psi }_i\right|^2.$$
(10)
In these equations $`\omega `$ is the incident photon energy in Rydberg units, and $`\mathrm{\Psi }_\mathrm{i}`$ and $`\mathrm{\Psi }_\mathrm{f}`$ are the wave functions representing the initial and final states respectively. The boundary conditions satisfied by a bound state with negative energy correspond to exponentially decaying partial waves in all โclosedโ channels, whilst those satisfied by a free or continuum state correspond to a plane wave in the direction of the ejected electron momentum $`\underset{ยฏ}{\overset{^}{k}}`$ and ingoing waves in all open channels.
Using the energy difference, $`E_{ji}`$, between the initial and final states, the oscillator strength, $`f_{ij}`$, for the transition can be obtained from $`S`$ as
$$f_{ij}=\frac{E_{ji}}{3g_i}S,$$
(11)
and the Einsteinโs A-coefficient, $`A_{ji}`$, as
$$A_{ji}(a.u.)=\frac{1}{2}\alpha ^3\frac{g_i}{g_j}E_{ji}^2f_{ij},$$
(12)
where $`\alpha `$ is the fine structure constant, and $`g_i`$, $`g_j`$ are the statistical weight factors of the initial and final states, respectively. In terms of c.g.s. unit of time,
$$A_{ji}(s^1)=\frac{A_{ji}(a.u.)}{\tau _0},$$
(13)
where $`\tau _0=2.4191^{17}`$s is the atomic unit of time.
## 3. Computations
The target wavefunctions of Fe VI were obtained by Chen and Pradhan from an atomic structure calculation using the Breit-Pauli version of the SUPERSTRUCTURE program , intended for electron collision calculations with Fe VI using the Breit-Pauli R-matrix method. Present work employs their optimized target of 19 fine structure levels corresponding to the 8-term $`LS`$ basis set of $`3d^3(^4F`$, $`{}_{}{}^{4}P`$, $`{}_{}{}^{2}G`$, $`{}_{}{}^{2}P`$, $`{}_{}{}^{2}D2`$, $`{}_{}{}^{2}H`$, $`{}_{}{}^{2}F`$, $`{}_{}{}^{2}D1)`$. The set of correlation configurations used were $`3s^23p^63d^24s`$, $`3s^23p^63d^24d`$, $`3s3p^63d^4`$, $`3p^63d^5`$, $`3s^23p^43d^5`$, and $`3p^63d^44s`$. The values of the scaling parameter in the Thomas-Fermi potential for each orbital of the target ion are given in Ref. . Table I lists the 19 fine structure energy levels of Fe VI used in the eigenfunction expansion where the energies are the observed ones. Most bound levels in low ionization stages correspond to the level of excitation of the parent ion involving the first few excited states. The criterion remains the accuracy of the target represetation that constitute the core ion states. The (N+1) electron configurations, $`\mathrm{\Phi }_j`$, which meet the orthogonality condition for the CC expansion (the second term of the wavefunction, Eq. (1)) are given below Table I. The same set of configurations is used for all the states considered in this work. STG1 of the BPRM codes computes the one- and two-electron radial integrals using the one-electron target orbitals generated by SUPERSTRUCTURE. The number of continuum basis functions is 12.
The present calculations are concerned with all possible bound levels with $`n10,\mathrm{}n1`$. These correspond to total (e + Fe VI) symmetries ($`SL\pi `$) with (2$`S`$ \+ 1) = 1,3,5 and $`L`$ = 0 - 10 (even and odd parities). The intermediate coupling calculations are carried out on recoupling these $`LS`$ symmetries in a pair-coupling representation, Eq. 6, in stage RECUPD. The computer memory requirement for this stage is the maximum, since it carries out angular algebra of dipole matrix elements of a large number of fine structure levels. The (e + Fe VI) Hamiltonian is diagonalized for each resulting $`J\pi `$ in STGH. The negative eigenvalues of the Hamiltonian correspond to the bound levels of Fe V, that are found according to the procedure described below.
### a) Calculation of bound levels
The eigenenergies of the Hamiltonian for each $`J\pi `$ are determined with a numerical search on an effective quantum number mesh, with an interval $`\mathrm{\Delta }\nu `$, using the code STGB. In the relativistic case, the number of Rydberg series of levels increases considerably from those in $`LS`$ coupling due to splitting of the target states into their fine structure components. This results in a large number of fine structure levels in comparatively narrow energy bands. A mesh with $`\mathrm{\Delta }(\nu )=0.01`$ is usually adequate to scan for $`LS`$ term energies; however, it is found to be of insufficient resolution for fine structure energy levels. The mesh needs to be finer by an order of magnitude, i.e., $`\mathrm{\Delta }(\nu )=0.001`$, so as not to miss out any significant number of bound levels. This considerably increases the computational requirements for the intermediate coupling calculations of bound levels over the LS coupling case by orders of magnitude. The calculations take up to several CPU hours per $`J\pi `$ in order to determine the corresponding eigenvalues. All bound levels of total $`J`$ 8, of both parities, are considered. However, a further search with an even finer $`\mathrm{\Delta }\nu `$ reveals that a few levels are still missing for some $`J\pi `$ symmetries.
### b) Procedure for level identification
The energy levels in the BPRM approximation (from STGB) are identified by $`J\pi `$ alone. This is obviously insufficient information to identify all associated quantum numbers of a level from among a large set levels for each $`J\pi `$, typically a few hundred for Fe V. A sample set of energy levels for $`J=2`$, even parity, obtained from the BPRM calculations is presented in Table II. The table shows energies and effective quantum number $`\nu _g`$, as calculated relative to the ground level $`(3d^{34}F_{3/2})`$ of the core ion Fe VI. The complexity of the calculations, and that of level identification, may be gauged from the fact that 30 of these levels have nearly the same $`\nu _g`$. Further, the $`\nu _g`$ do not in general correspond to the actual effective quantum number of the Fe V level since it may belong to an excited parent level, and not the ground level, of Fe VI.
A scheme has been developed to identify the levels with complete spectroscopic information consisting of
$$(C_tS_tL_tJ_t\pi _t\mathrm{}[K]\mathrm{s})J\pi ,$$
(14)
and also to designate the levels with a possible $`SL\pi `$ symmetry. The designation of the $`SL\pi `$, from the identifications denoted above, is generally ambiguous since the collision channels are all in intermediate coupling. However, in most cases we are able to carry through the identification procedure to the $`LS`$ term designation. An advantage of identification is that it greatly facilitates the completeness check for all possible LS terms and locate any missing levels. A computer code PRCBPID has been developed to identify all the quantum numbers relevant to the $`J\pi `$ and the $`LS`$ term assignments. Identification is carried out for all the levels belonging to a $`J\pi `$ symmetry at a time.
The components of the total wavefunction of a given fine structure energy level span all closed โcollisionโ channels $`(C_tS_tL_t(J_t)\pi _t)ฯตl`$. Each channel contains the information of the relevant core and the outer electron angular momentum. The โchannel weightsโ, mentioned earlier, determine the magnitude of the wavefunction in the outer R-matrix region of each channel evaluated in STGB. A bound level may be readily assigned to the quantum numbers of a given channel provided the corresponding channel weight (in percentage terms) dominates the other channels.
The number of channels can be large especially for complex ions. For Fe V, for example, each level with $`J>2`$ corresponds to several hundred channels. As the first step in the level identification scheme we isolate the two most dominant channels by comparing all channel percentage weights. The reason is that the largest channel percentage weight may not uniquely determine the identifications since the channel weights are evaluated from the outer region contributions ($`r>a`$); the inner region contributions are unknown. Also, many levels are often heavily mixed and no assignment for the dominant channel may be made.
The program, PRCBPID, sorts out the duplicate identifications in all the levels of the $`J\pi `$ symmetry. Two levels with the same configuration and set of quantum numbers can actually be two independent levels due to outer electron spin addition/ subtraction to/from the parent spin angular momentum, i.e. $`S_t\pm s=S`$. The identical pair of levels are tagged with positive and negative signs indicating higher and lower multiplicity respectively. The lower energies are normally assigned with the higher spin multiplicity. However, the energies and effective quantum numbers ($`\nu `$) of levels of higher and lower spin multiplicity can be very close to each other, in which case the spin multiplicity assignment may be uncertain.
One important identification criterion is the analysis of the quantum defect, $`\mu `$, or the effective quantum number, $`\nu `$, of the outer or the valence electron. The principle quantum number, $`n`$, of the outer electron of a level is determined from its $`\nu `$, and a Rydberg series of levels can be identified from the effective quantum number. Hence, in the identification procedure, $`\nu `$ of the lowest member (level with the lowest principal quantum number of the valence electron) of a Rydberg series is determined from quantum defect analysis of all the computed levels for each partial wave $`l`$. The lowest partial wave has the highest quantum defect. A check is maintained to differentiate the quantum defect of a $`{}_{}{}^{}s_{}^{}`$ electron with that of an equivalent electron state which has typically a large value in the close coupling calculations. The principle quantum number, $`n`$, of the lowest member of the series is determined from the orbital angular momentum of the outer electron and the target or core configuration. Once $`\nu `$ and $`\mu =n\nu `$ of the lowest member are known, the $`n`$-values of all levels can be assigned for each paritial wave, $`l`$. The relevant Rydberg series of levels is also identified from the levels that have the same symmetry, $`J\pi `$, core configuration, $`C_tS_tL_t\pi _t`$ and outer electron orbital angular momentum $`l`$, but different $`\nu `$ that differs between successive levels by $``$ 1. While the $`\nu (n\mathrm{})`$ are more accurate for the higher members of the series, they are more approximate for the lowest ones. The quantum defect of a given partial wave $`\mathrm{}`$ also varies slightly with different parent core levels and final $`SLJ`$ symmetries.
Of the two most dominant channels the proper one for each bound level is determined based on several criteria. There are cases when more than two levels are found to have identical identifications. These levels are checked individually for proper identification. Often a swap of identifications is needed between the two sets of dominant channels since the second dominating channel is more likely to be associated with the given level, consistent with all other criteria. In some cases the most dominant channel (largest percentage weight in the outer region) may correspond to comparatively larger $`\nu `$ for the partial wave $`\mathrm{}`$, than to a reasonable $`\nu `$ for the second channel, indicating that the identification should correspond to the second channel.
In a few cases a level is found not to correspond to any of the two dominant channels, predetermined from the channel weights. At the same time often a level is found to be missing in the same energy range. In such case the level is assigned to a channel of lower percentage weight that has a reasonable core configuration and term, $`nl`$ quantum numbers for the outer electron and effective quantum number that match the missing level.
There are a number of levels belonging to equivalent-electron configurations and require different identification criteria from those of the Rydberg states. These levels usually have: (i) a number of approximately equal channel weights, and (ii) quantum defects that are larger than that of the lowest partial wave, or an inconsistent $`\nu `$ that does not match with any reasonable $`n`$. Once these levels are singled out, they are identified with the possible configurations of the core level, augmented by one electron in the existing orbital sub-shell. These low-lying levels are often assigned to those identified from the small experimentally available set of observed levels. The levels that can not be identified in the above procedure, such as by swapping of channels, or maching to a missing level, are assumed to belong to mixed states. These are not analysed futher by quantum defects.
Two additional (and related) problems, as mentioned above, are addressed in the identification work: (A) standard $`LS`$ coupling designation, $`SL\pi `$, and (B) the completeness check for the set of all fine structure components within an $`LS`$ multiplet. Identification according to collision channel quantum numbers is not quite sufficient to establish a direct correspondence with the standard spectroscopic notation employed in atomic structure calculations, or in the compiled databases such as those by the U.S. National Institute for Standards and Technology (NIST).
The possible set of $`SL\pi `$s of a level is obtained from the target term, $`S_tL_t\pi _t`$, and the valence electron angular momentum, $`l`$, at the first occurance of the level in the set. The total spin multiplicity of the level is defined according to the energy level position as discussed above. For example, the core $`3d^3(^4F^e)`$ combining with a $`4d`$ electron forms the terms $`{}_{}{}^{5}(P,D,F,G,H)_{}^{e}`$ and $`{}_{}{}^{3}(P,D,F,G,H)_{}^{e}`$ (Table IV) where the quintet for each $`L`$ should be lower than the triplet.
To each $`LS`$ symmetry, $`SL\pi `$, of the set belongs a set of predetermined $`J`$-levels. The set of total J-values of same spin multiplicity is then calculated from all possible $`LS`$ terms, equal to $`|๐+๐|`$. The program sorts out all calculated fine stucture levels with the same configuration, but with different sets of $`J_t`$ and $`J`$, e.g. $`(C_tS_tL_tJ_t\pi _tn\mathrm{})J\pi `$ (including the sign for the upper or lower spin multiplicity), compares them with the predetermined set, and groups them together. Thus a correspondence is made between the set of $`SL\pi `$ and the calculated fine structure levels of same configuration.
In addition to the correspondence between the two sets, the program PRCBPID also calculates the possible set of $`SL\pi `$โs for each single J-level in above group. In the set of $`SL\pi `$s, the total spin is fixed while the angular momentum, $`L`$, varies. In the above example for the quintets, $`{}_{}{}^{5}(P,D,F,G,H)_{}^{e}`$, each $`J`$=1 level is assigned to a possible set of terms, $`{}_{}{}^{5}(P,D,F)`$ (Table IV). However, these levels can be futher identified uniquely following Hundโs rule that the term with the larger angular momentum, $`L`$, is the lower one, i.e., the first or the lowest $`J`$=1 level should correspond to $`{}_{}{}^{5}F`$, the second one to $`{}_{}{}^{5}D`$ and the last one to $`{}_{}{}^{5}P`$.
The completeness of sets of fine structure levels with respect to the $`LS`$ terms are checked. As mentioned above, PRCBPID determines the possible sets of $`SL\pi `$ from the target term and valence electron angular momentum of a level at its first occurance and calculates the total J-values of the set of $`LS`$ terms. The number of these $`J`$-values, $`Nlv`$, is compared with that of calculated levels, $`Ncal`$ to check the completeness. For example, for the above case of $`{}_{}{}^{5}(P,D,F,G,H)_{}^{e}`$ in Table IV discussed above, $`{}_{}{}^{5}P`$ can have $`J`$ = 1,2,3, $`{}_{}{}^{5}D`$ can have $`J`$ = 0,1,2,3,4, and so on, giving a total of 23 fine structure levels for this set of $`LS`$ terms. The one $`J`$ = 0 level belongs to $`{}_{}{}^{5}D`$, the three $`J`$ = 1 levels belong to $`{}_{}{}^{5}(P,D,F)`$, and so on. All 23 levels of this set are found in the computed levels (Table IV), thus making the computed set complete. This procedure, in addition to finding the link between the two diiferent coupling schemes, enables an independent counting of the number of levels obtained, and ascertains missing or mis-identified levels.
### c) Transition probabilities
The oscillator strengths (f-values) and transition probabilites (A-values) for bound-bound fine structure level transitions in Fe V are calculated for levels up to $`J8`$. Computations are carried out using STGBB of the BPRM codes.
The f-values are initially calculated by the program STGBB with level designations given by $`J\pi `$ only. However, the transitions may be fully described following the level identifications as described in the previous section. Work is in progress to identify all the transitions with proper quantum numbers, configurations and possible $`SL\pi `$โs.
A subset of the large number of transitions has been processed with complete identifications. Among these transitions are those that correspond to the experimentally observed levels . As these levels have been identified, their oscillator strenghts could be sorted out from the file of f-values. Another subsidiary code, PRCBPRAD, is developed to reprocess the transition probabilities where the calculated transition energies are replaced by the observed ones for improved accuracy.
The computation time required for the BPRM calculations was orders of magnitude longer compared to oscillator strengths calculations in LS coupling, as carried out under the OP for example. The time excludes that needed for creating the necessary bound state wavefunctions and calculating dipole matrix elements using the R-matrix package of codes. Computations are carried out for one or a few pairs of symmetries at a time requiring several hours of CPU time on the Cray T94. The memory requirement was over 30 MWords.
## 4. Results and discussion
Theoretical spectroscopic data are calculated on a large-scale with relativistic fine structure included in an ab initio manner, and ensuring completeness in terms of obtaining nearly all possible energy levels and transition probabilities for Fe V for the total angular symmetries considered. The results are described below.
### a) Energy Levels
We have calculated 3,865 fine structure bound levels, with 0 $`J`$ 8, for Fe V. Following level identification, as explained in the previous section, the energy levels are arranged according to ascending order in energy.
The present energies are compared with the relatively small set of experimentally observed levels compiled by NIST in Table III. All 179 observed levels are obtained and identified. Asterisks attached to levels in Table III indicate an incomplete set of observed levels corresponding to the $`LS`$ term. Often in experimental measurements the weak lines are not observed. The theoretical datasets on the other hand are usually complete.
We find some discrepancies regarding the identification of a couple of levels in the NIST tabulation. The $`J=2`$ level at 2.9395 Ry identified in the NIST table as $`3d^3(^4P)4p(^5S^o)_2`$, from the maximum leading percentage, may have been misidentified. Present analysis for the completeness of a set of fine structure levels belonging to a term indicates it as an extra level for the given configuration and that the possible $`LS`$ terms for this level are $`3d^3(^2D2)4p(^3PDF^o)`$, possibly $`{}_{}{}^{3}F_{}^{o}`$. Similarly the NIST identification for the $`J`$=3 level at 2.8968 Ry is $`3d^3(^2P)4p(^3D^o)_3`$, from the maximum leading percentage. Present calculations however assign the level to possible LS terms, $`3d^3(^2D2)4p(^3DF^o)`$, and most likely to $`{}_{}{}^{3}D_{}^{o}`$.
In the computed set of fine structure levels the observed levels are usually the ones with the lowest energy in each subset of $`J\pi `$. The lowest calculated levels are the 34 levels of the ground configuration $`3d^4`$ of Fe V, in agreement with the observed ones. The agreement between the observed and calculated energies for these levels is within 1%. The calculated energies agree to about 1% with the measured ones for most of the observed levels. Although the energies are exoected to be highly accurate, but the uncertainty in the calculations is not comparable to that in spectroscopic observations (of the order of few wavenumbers).
Employing the completeness procedure the computed fine structure levels are tabulated, according to the two sets of cross-correlating quantum numbers: one according to the collision channels identified as $`(C_tS_tL_tJ_t\pi _tn\mathrm{}[K]\mathrm{s})J\pi `$, and the other according to the complete set of J-values for each multiplicity $`(2S+1)`$, $`L`$ and $`\pi `$. A subset of the complete table of fine structure levels is presented in Table IV. (The complete table will be available electronically). Each set of levels is grouped by the possible set of $`LS`$ terms followed by the levels of same configuration, core term, total spin multiplicity and parity, and with different $`J`$-values. The header for each group contains the total number of possible $`J`$-levels, $`Nlv`$, total spin multiplicity, parity, and all possible $`L`$ values formed from the core and the outer electron. The possible $`J`$-values for each $`SL\pi `$ are given within parentheses next to each $`L`$ value.
The two sets of quantum numbers are compared. The levels that may be missing or mis-identified are thereby checked out. The number of computed levels, $`Ncal`$, is compared with that expected from angular and spin couplings, $`Nlv`$. For most of the configurations the set of levels is complete except for the high lying ones. The comparison detects missing levels. An example is shown in the the set of $`3d^32(^2D)5d^3(S,P,D,F,G)^e`$ in Table IV where one level with $`J^e`$ = 4 is missing.
In Table IV, the effective quantum number $`\nu `$ is specified alongwith other quantum numbers for each level. The consistency in $`\nu =\frac{z}{\sqrt{(}EE_t)}`$, where $`E_t`$ is the corresponding target energy, for each set of levels may be noted.
The possible $`SL\pi `$s for each level are given in the last column. The levels with a single possible term only are uniquely defined. However, those with two or multiple term assignments can be defined uniquely applying Hundโs rule that the higher $`L`$ corresponds to the lower energy of same $`J\pi `$ as explained in the previous section (we note that Hundโs rule may not always apply in cases of strong CI).
There are 112 levels of odd parity that we could not properly identify. Some of these levels are given in Table IV. These levels could be equivalent electron levels of configuration, $`3p^5(^2P^o)3d^5`$. The 16 LS terms of $`3d^5`$, which are $`{}_{}{}^{2}D1`$, $`{}_{}{}^{2}P3`$, $`{}_{}{}^{2}D3`$, $`{}_{}{}^{2}F3`$, $`{}_{}{}^{2}G3`$, $`{}_{}{}^{2}H3`$, $`{}_{}{}^{4}P5`$, $`{}_{}{}^{4}F5`$, $`{}_{}{}^{2}S5`$, $`{}_{}{}^{2}D5`$, $`{}_{}{}^{2}F5`$, $`{}_{}{}^{2}G5`$, $`{}_{}{}^{2}I5`$, $`{}_{}{}^{4}D5`$, $`{}_{}{}^{4}G5`$, and $`{}_{}{}^{6}S5`$, in combination with the parent core $`{}_{}{}^{2}P_{}^{o}`$, form 88 LS terms with 31 singlets, 43 triplets, 13 quintets and 1 septet. The number of fine structure levels from these terms exceed the 112 computed levels that have not been identified.
This new procedure of cross-correlation between two coupling schemes thus provides a powerful check on the completeness and level identification, and is expected to be of use in further BPRM work on complex atomic systems.
### b) Transition Probabilities
The oscillator strengths (f-values) and transition probabilites (A-values) for fine structure level transitions in Fe V are obtained for $`J8`$. The allowed $`\mathrm{\Delta }J=0,\pm 1`$ transitions include both the dipole allowed ($`\mathrm{\Delta }S=0,\pm 1`$) and the intercombination ($`\mathrm{\Delta }S`$ 0) transitions. The total number of computed transition probabilities is well over a million, approximately $`1.46\times 10^6`$. For most allowed pairs of $`J\pi `$ symmetries, there are about $`10^310^5`$ transitions.
As explained in the previous section, a subset of the encoded transitions have been processed to present them with proper identifications. These correspond to the levels that have been observed. A sample of these is presented in Table V. In all of the f-values presented the calculated transition energy has been replaced by the observed one, using the BPRM line strengths ($`S`$) which are energy independent. Since measured energies in general have smaller uncertainties than the calculated ones, this replacement improves the accuracy of the oscillator strengths. The transitions among the 179 observed levels correspond to 3727 oscillator strengths. (The complete set of transition probabilities will be available electronically.)
The f-values in Table V have been reordered to group the transitions of the same multiplet together. This enables a check on the completeness of the set of transitions. As this table corresponds to transitions among observed levels only, the completeness depends on the set of observed levels belonging to the LS terms. For the dipole allowed transitions, the LS multiplets are also given at the end of $`jj^{}`$ transitions.
To our knowledge, no measured f-values for Fe V are available for comparison. Current NIST compilation contains no f-values for any allowed transition. On the other hand, Fe V oscillator strengths for a large number of transitions were obtined in the close coupling approximation under the OP and the IP . Both of these datasets are non-relativistic calculations in LS coupling and do not compute fine structure transitions. Fawcett has carried out semi-empirical relativistic atomic structure calculations for fine structure transitions in Fe V. Comparison of the present $`f`$-values is made with the previous ones in Table VI, showing various degrees of agreement. Present values agree within 10% with those by Fawcett for a number of fine structure transitions of multiplets, $`3d^4(^5D)3d^3(^4F)4p(^5D^o)`$, and $`3d^4(^5D)3d^3(^4P)4p(^5P^o)`$, and the disagreement is large with other as well as with those of $`3d^4(^5D)3d^3(^4F)4p(^5F^o)`$. The agreement of the present $`LS`$ multiplets with the others is good for transitions $`3d^4(^5D)3d^3(^4F)4p(^5F^o,^5D^o,^5P^o)`$. More detailed comparisons will be made at the completion of this work.
The procedure of substitution of experimental for calculated energies provides an indication of uncertainties in the calculated $`f`$-values. The difference between the $`f`$-values obtained using the calculated transition energies and the observed ones is only a few percents ($`<`$ 5%) for most of the allowed transitions. The difference is usually larger for the intercombination transitions which have lower transition probabilities. In atomic structure calculations, it is possible to re-adjust eigenenergies of the Hamiltonian to match the observed ones and then use the wavefunctions to obtain the transition probabilities. Such a re-adjustment is not carried out in the BPRM calculations of bound states, which are entirely ab initio, with the associated advantage of consistent uncertainties for most transitions considered.
To obtain an estimate of the accuracy of the wavefunctions employed in the length and the velocity formulations, we plot, for example, the $`gf`$-values for transitions $`(J=1)^e(J=2)^o`$ and $`(J=3)^e(J=4)^o`$ in Fig. 1. The top panel contains over 13,300 transitions between the pair of symmetries $`(J=1)^e(J=2)^o`$, and the bottom panel contains over 20,200 transitions between the pair $`(J=3)^e(J=4)^o`$. The plots show practically no dispersion for the strongest transitions with $`gf510`$, and some dispersion around 10-20% for others with $`gf<3`$. Up to $`gf<0.1`$ the dispersion in length and velocity remains around the 10-20% level for most of the transitions, although the number of outlying transitions increases with decreasing $`gf`$. Given the large number of points in the figures, the relatively low dispersion of $`gf_L`$ and $`gf_V`$ indicates that the $`f`$-values ($`gf`$ divided by the statistical weight factor, $`2J`$+1) for most of the transitions with $`gf`$1 should be within 20% uncertainty. The $`f_L`$โs are usually more accurate than $`f_V`$โs since the asymptotic region wavefunctions are more accurately represented in the close coupling calculations using the R-matrix method.
In general the intercombination transitions are weaker than the dipole allowed ones; the f-values can be orders of magnitude lower. The BP Hamiltonian in the present work (Eq. 2) does not include the two-body spin-spin and spin-other-orbit terms of Breit interaction . A discussion of these terms is given by Mendoza et al. in a recent IP paper . Their study on the intercombination transitions in C-like ions shows that the effect of the two-body Breit terms, relative to the one-body operators, decreases with Z such that for Z = 26 the computed A-values with and without the two-body Breit terms differ by less than 0.5 %. However, the differences towards the neutral end of the C-sequence is up to about 20%. It may therefore be expected that for Fe V the weaker intercombination f-values may also be systematically affected to a similar extent (the uncertainties in the dipole allowed f-values should be much less). Further studies of the Breit interaction in complex atoms are needed to ascertain this effect more precisely.
Several aspects of the present work are targets for future studies, such as atomic structure calculations to study the effect of configuration interaction and relativistic effects on different types of transitions, and a detailed quantum defect analysis along interacting Rydberg series of levels in intermediate coupling. These studies should provide information on the accuracy of particular type of transitions and groups of levels, as well as address general problems in the analysis of complex spectra.
## 5. Conclusion
The present work is the first study of large-scale transition probabilities computed using the accurate BPRM method for a highly complex ion. Some of the results obtained herein are expected to form the basis for future computational spectroscopy of heretofore intractable complex atomic systems using efficient collision theory methods. The computational procedures developed for such undertakings are described, and illustrative results are presented from the ab initio Breit-Pauli R-matrix calculations for Fe V. Detailed analysis for the identification of over 3,800 fine structure levels of Fe V is carried out using a combination of methods that include quantum defect theory. Further work on the analysis of relativistic quantum defects in intermediate coupling is planned.
Following the completion of all computations and identifications, the dataset of approximately 1.5 million oscillator strengths will be described in another publication with a view towards astrophysical and laboratory applications. In order to complete the dataset for practical applications calculations are also in progress for the forbidden electric quadrupole and magnetic dipole transition probabilties using the atomic structure program SUPERSTRUCTURE.
The newly acquired theoretical capability to obtain an essentially complete description of radiative transitions for an atomic system should enable several new advances such as: (a) the synthesis of highly detailed monochromatic opacity spectra , (b) the simulation of โquasi-continuumโ line spectra from iron ions , (c) high resolution spectral diagnostics of iron in laboratory fusion and astrophysical sources, and (d) the analysis of experimentally measured spectra of complex iron ions.
Acknowledgements
This work was supported partially by the U.S. National Science Foundation (AST-9870089) and the NASA Astrophysical Theory Program. The computational work was carried out at the Ohio Supercomputer Center in Columbus Ohio.
References
1. Fuhr, J.R., Martin, G.A., Wiese, W.L., J. Phys. Chem. Ref. Data 17, Suppl No. 4 (1988)
2. Seaton, M.J., Yu, Y., Mihalas, D., Pradhan, A.K., MNRAS 266. 805 (1994).
3. Rogers F.J. and Iglesias C.A., Science 263, 50 (1994)
4. Chayer, P., Fontaine, G., and Wesemael, F., Astrophys.J. 99, 189 (1995).
5. Becker, S.R. and Butler, K., Astron. Astrophys. , 265, 647 (1992)
6. Vennes, S., Astrophysics in the Extreme Ultraviolet (Ed: Stuart Bowyer and Roger F. Malina), Kluwer, p. 185; (1996); Pradhan, A.K., Ibid., p. 569
7. Seaton M.J. 1987, J.Phys.B 20, 6363.
8. Hummer, D.G., Berrington, K.A., Eissner, W., Pradhan, A.K., Saraph, H.E., Tully, J.A., Astron. Astrophys. 279, 298 (1993)
9. Burke, P.G., Hibbert, A., and Robb, Journal Of Physics B 4, 153 (1971)
10. Berrington K.A., Burke P.G., Butler K., Seaton M.J., Storey P.J., Taylor K.T. and Yu Yan, J.Phys.B 20 6379 (1987).
11. Nahar S.N. and Pradhan, A.K., Astron. Astrophys. Suppl. Ser. 135, 347 (1999).
12. Zhang, H.L., Physical Review A 57, 2640 (1998)
13. Zhang H.L., Nahar S.N., and Pradhan A.K., J. Phys. B 32, 1459 (1999).
14. Johnson W.R., Liu Z.W., and Sapirstein J., At. Data Nucl. Data 64, 279 (1996).
15. Yan Z-C, Tambasco M., and Drake G.W.F., Phys. Rev. A 57, 1652 (1998)
16. Burke P.G. and Seaton M.J., Journal Of Physics B 17, L683 (1984)
17. Seaton M.J. J.Phys.B 18, 2111 (1985)
18. Scott N.S., Taylor K.T., Comput. Phys. Commun. 25, 347 (1982)
19. Berrington K.A., Eissner, W., Norrington P.H. Comput. Phys. Commun. 92, 290 (1995)
20. Burke, P.G. and Berrington, K.A., Atomic and Molecular Processes, an R-matrix Approach, Institute of Physics Publishing, Bristol (1993)
21. Chen, G.X. and Pradhan, A.K., Journal Of Physics B 32, 1809 (1999a); Astron. Astrophys. Suppl. 136, 395 (1999b)
22. Eissner, W., Jones, M., Nussbaumer, H., Comput. Phys. Commun 8, 270 (1974); W. Eissner, J. Phys. IV (Paris) C1, 3 (1991), Eissner W. (in preparation, 1999)
23. Sugar, J. and Corliss, C., J. Phys. Chem. Ref. Data 14, Suppl. 2 (1985)
24. Butler, K. (unpublished); data are available through the OP database, TOPbase (W. Cunto, C. Mendoza, F. Ochsenbein, C.J. Zeippen, Astron. Astrophys 275, L5 (1993))
25. Bautista M.A., A&A Suppl. Ser. 119, 105 (1996)
26. Fawcett, B.C. At. Data Nucl. Data Tables 41, 181 (1989)
27. Mendoza C., Zeippen C.J. and Storey P.J., A&A Suppl.Ser. 135, 159 (1999)
28. Beirsdorfer, P. Lepson, J.K., Brown, G.V., Utter, S.B., Kahn, S.M., Liedahl, D.A., and Mauche, C.W., Astrophys. J. 519, L185 (1999)
Figure captions:
1. Comprarison of $`gf_L`$ versus $`gf_V`$ for bound-bound fine structure level transitions in Fe V obtained in BPRM approximation. |
warning/0001/hep-ph0001122.html | ar5iv | text | # 1 Introduction
## 1 Introduction
Initially, one of the most attractive features of Grand Unified Theories (GUTs) was the prospect that they might provide an explanation for the matter-antimatter asymmetry of the Universe, via their new interactions that violate baryon and/or lepton number. Subsequently, it has been realized that, even in the Standard Model, at the non-perturbative level there are sphaleron interactions that violate both baryon and lepton number. This discovery has given rise to new scenarios for baryogenesis, at the electroweak phase transition or via leptogenesis followed by sphaleron reprocessing . Supersymmetric extensions of the Standard Model offer yet more scenarios for baryogenesis. For example, they may facilitate electroweak baryogenesis by permitting a first-order electroweak phase transition despite the constraints imposed by LEP . There is also the possbility that they may contain perturbative interactions that violate baryon and/or lepton number via a breakdown of $`R`$ parity, which under certain circumstances can induce baryogenesis.
However, perhaps the most attractive mechanism offered by supersymmetry is that proposed by Affleck and Dine, according to which a condensate of a combination of squark and/or slepton fields may have formed during an inflationary epoch in the early universe, causing the vacuum to carry a large net baryon and/or lepton number, which is then transferred to matter particles when the condensate eventually decays. We recall that the condensate forms along some flat direction of the effective potential of the theory, which we take to be the Minimal Supersymmetric extension of the Standard Model (MSSM) at low energies. In the conventional approach to Affleck-Dine baryogenesis, the condensate is essentially static until a relatively late cosmological epoch, when it starts to oscillate. In turn, the termination of the period of oscillation has been calculated in terms of the magnitudes of the soft supersymmetry-breaking terms present in the effective potential, which become significant only at low temperatures, and of the thermalization effects of inflaton decay .
The purpose of this paper is to re-examine this Affleck-Dine mechanism by incorporating a more complete treatment of the reheating of the universe after the inflationary epoch. We argue that the flat directions are in general coupled to other fields that are kinematically accessible to inflaton decay. These fields therefore have non-trivial statistical densities, and become thermalized. The couplings of these densities to the flat directions induce effective supersymmetry-breaking masses and $`A`$ terms for the erstwhile flat fields. As a result, the โflatโ directions start oscillating earlier than previously estimated. Subsequently, the oscillations also terminate earlier, as the flat-direction condensate interacts with the plasma of inflaton decay products and evaporates. The bottom line is that previous estimates of the resulting baryon/lepton asymmetry of the universe may be substantially altered, and we estimate some orders of magnitude for different representative parameter choices.
## 2 Flat Directions
The $`D`$-flat directions of the MSSM are classified by gauge-invariant monomials in the fields of the theory. These monomials have been classified in , and, for directions which are also $`F`$-flat for renormalizable standard model superpotential interactions, the dimension of the non-renormalizable term in the superpotential which first lifts the respective $`D`$-flat direction has also been derived. Hereafter, we consider only those $`D`$-flat directions which are not lifted by renormalizable superpotential interactions. These correspond to 14 independent monomials, and each monomial represents a complex $`D`$-flat direction: one vev magnitude and one phase (all fields in the monomial have the same vev). Since the monomials are gauge-invariant, appropriate gauge transformations generated by non-diagonal generators can be used to remove that part of the $`D`$-term contribution to the potential which comes from the non-diagonal generators. Also, any relative phase among the fields in the monomial can be rotated away by those gauge transformations which are generated by diagonal generators. There remains only one overall phase, i.e., the phase of the flat direction, which can be absorbed by redefinition of the scalar fields. Note that the $`D`$-term and $`F`$-term parts of the scalar potential are invariant under such a redefinition (which is equivalent to a U(1) symmetry transformation) while the soft-breaking terms and fermionic Yukawa terms generally are not. So we can always arrange the vev of the fields in the monomial to be initially along the real axis. We also note that such a non-zero vev breaks spontaneously the MSSM gauge group.
As an explicit example, consider the simplest case, which is the $`H^uL`$ flat direction. If the $`T_3=\frac{1}{2}`$ component of $`H^u`$ and the $`T_3=\frac{1}{2}`$ component of $`L`$ have the same vev, then all the $`D`$ terms from both diagonal and non-diagonal generators of the MSSM are zero. The non-diagonal ones are identically zero and the equality of the vev makes the diagonal ones zero as well. These vevโs can then be chosen along the real axis as noted above. There are eight real degrees of freedom in the $`H^u`$ and $`L`$ doublets. Two of them comprise the flat direction and another three are Goldstone bosons eaten by the gauge fields of the spontaneously-broken symmetries. The remaining three are physical scalars which are coupled to the flat direction, and are massive due to its vev.
Now that all fields in the monomial have the same vev and are real, by an orthogonal transformation we can go to a new basis where there is only one direction with a non-zero vev. Let us label this direction $`\alpha `$ and the orthogonal directions generically as $`\varphi `$, therefore $`\alpha _R0`$ while $`\alpha _I=\varphi _R=\varphi _I=0`$. For the specific $`H^uL`$ example, these are the following combinations after the Goldstone bosons are absorbed by the Higgs mechanism:
$`\sqrt{2}\alpha _R`$ $`=`$ $`(H_1)_R+(L_2)_R`$
$`\sqrt{2}\alpha _I`$ $`=`$ $`(H_1)_I+(L_2)_I`$
$`\sqrt{2}\varphi _1`$ $`=`$ $`(H_1)_R(L_2)_R`$
$`\sqrt{2}\varphi _2`$ $`=`$ $`(H_2)_R(L_1)_I`$
$`\sqrt{2}\varphi _3`$ $`=`$ $`(H_2)_I+(L_1)_R`$
The $`D`$ terms from the $`T_3`$ and $`U(1)_Y`$ generators give terms $`g^2\alpha ^2\varphi _{1}^{}{}_{}{}^{2}`$ (up to numerical factors) in the potential, whilst those from $`T_1`$ and $`T_2`$ give $`g^2\alpha ^2\varphi _{2}^{}{}_{}{}^{2}`$ and $`g^2\alpha ^2\varphi _{3}^{}{}_{}{}^{2}`$ terms (up to numerical factors). It is a generic feature that all fields entering in the flat direction monomial which are left after the Higgs mechanism (except the linear combination which receives the vev after diagonalization) have masses of order $`g\alpha `$ due to their $`D`$-term couplings to the flat-direction vev.
We now consider supergravity effects, both in minimal models with soft-breaking terms at the tree level, and in no-scale models , where such terms are absent at tree-level but arise from quantum corrections . The superpotential consists of the tree-level MSSM terms and a series of non-renormalizable terms of successively higher dimension, which are induced in the effective theory by the dynamics of whatever is the underlying more fundamental theory. Without imposing $`R`$ parity (or any other symmetry) all gauge-invariant terms of higher dimension would exist in the superpotential. We may, however, also wish to impose $`R`$ parity on the higher-dimensional terms, as we have done on the renormalizable interactions, to prevent substantial $`R`$-parity violation being fed down from high scales by the renormalization-group running of the soft mass terms . If we assume that $`R`$ parity is a discrete gauge symmetry of the theory, then it would be respected by all gauge-invariant superpotential terms of arbitrary dimension. Relevant higher-dimensional superpotential terms which lift the flat direction $`\alpha `$ are of the form:
$$W\lambda _n\frac{\alpha ^n}{nM^{n3}}$$
(1)
where $`\lambda _n`$ is a number of order one and $`M`$ is a large mass scale, e.g., the GUT or Planck scale.
During inflation, supersymmetry is strongly broken by the non-zero energy of the vacuum. In minimal models this is transferred to the observable sector through the Kรคhler potential at tree level , while in no-scale models this happens at the one-loop level . Inflation then induces the soft-breaking terms
$$C_IH_{I}^{}{}_{}{}^{2}|\alpha |^2+a\lambda _nH_I\frac{\alpha ^n}{nM^{n3}}+h.c.$$
(2)
where $`C_I`$ and $`a`$ are numbers depending on the sector in which the inflaton lies, and $`H_I`$ is the Hubble constant during inflation. We shall assume here that $`C_I`$ is positive and not unnaturally small . In the presence of the $`A`$ term, the potential along the angular direction has the form $`\mathrm{cos}(n\theta +\theta _a)`$, where $`\theta _a`$ is the phase of $`a`$. Due to its negative mass-squared, the flat direction rolls down towards one of the discrete minima at $`n\theta +\theta _a=\pi `$ and $`|\alpha |=(\frac{C_I}{(n1)\lambda _n}H_IM^{n3})^{\frac{1}{n2}}`$, and quickly settles at one of the minima ($`\frac{C_I}{(n1)\lambda _n}`$ is $`O(1)`$). Therefore, at the end of inflation $`\alpha `$ can be at any of the above-mentioned minima.
In the absence of thermal effects, $`\alpha `$ would track the instantaneous minimum $`|\alpha |(HM^{n3})^{\frac{1}{n2}}`$ from the end of inflation until the time when $`Hm_{\frac{3}{2}}`$, where $`m_{\frac{3}{2}}1`$ TeV is the low-energy supersymmetry-breaking scale . At $`Hm_{\frac{3}{2}}`$ the low-energy soft terms
$$m_{\frac{3}{2}}^{}{}_{}{}^{2}|\alpha |^2+A\lambda _nm_{\frac{3}{2}}\frac{\alpha ^n}{nM^{n3}}+h.c.$$
(3)
would take over, with the mass-squared of $`\alpha `$ becoming positive, and $`\alpha `$ would then start its oscillations. Also, the minima along the angular direction would then move in a non-adiabatic way, due generally to different phases for $`A`$ and $`a`$. As a result, $`\alpha `$ starts its free oscillations around the origin with an initial vev $`\alpha _{osc.}(m_{\frac{3}{2}}M^{n3})^{\frac{1}{n2}}`$ and frequency $`m_{\frac{3}{2}}`$ and, at the same time, the torque exerted on it causes motion along the angular direction. In the case that the flat direction carries a baryon (lepton) number this will lead to a baryon (lepton) asymmetry $`n_B`$ given by $`n_B=\alpha _R\frac{\alpha _I}{t}\alpha _I\frac{\alpha _R}{t}`$. At $`m_{\frac{3}{2}}t1`$ the upper bound on $`n_B`$ may be written
$$n_B\frac{1}{m_{\frac{3}{2}}^{}{}_{}{}^{2}t^2}\alpha _{osc.}^{}{}_{}{}^{3}(\frac{\alpha _{osc.}}{M})^{n3}$$
(4)
which, after transition to a radiation-dominated universe, results in an $`\frac{n_B}{s}`$ that remains constant as long as there is no further entropy release.
As we will see in subsequent sections, thermal effects of inflaton decay products with superpotential couplings to the flat direction can fundamentally alter the dynamics of the flat direction oscillation, and necessitate revision of the estimates for the resulting baryon/lepton asymmetries produced.
## 3 Flat-Direction Superpotential Couplings and Finite Temperature Effects
As we have seen, the flat direction $`\alpha `$ has couplings of the form $`g^2\alpha ^2\varphi ^2`$ to the fields $`\varphi `$ which are in the monomial that represents it. Besides these $`D`$-term couplings, it also has $`F`$-term couplings to other fields $`\chi `$ which are not present in the monomial. These come from renormalizable superpotential Yukawas, and have the form <sup>1</sup><sup>1</sup>1We note that for $`F`$-flat directions of the renormalizable piece of the superpotential, which are only lifted by higher-dimensional nonrenormalizable terms, $`\alpha `$ cannot have such superpotential couplings to $`\varphi `$ fields which appear in the monomial.
$$Wh\alpha \chi \chi $$
(5)
which results in a term $`h^2|\alpha |^2|\chi |^2`$ in the scalar potential. Again for illustration, consider the $`H^uL`$ flat direction: $`H^u`$ has Yukawa couplings to left-handed and right-handed (s)quarks while $`L`$ has Yukawa couplings to $`H^d`$ and right-handed (s)leptons.
In the class of models that we consider, the inflaton is assumed to be in a sector which is coupled to ordinary matter by interactions of gravitational strength only. In this case, the inflaton decay always occurs in the perturbative regime and we need not worry about parametric-resonance decay effects . The inflaton decay rate is $`\mathrm{\Gamma }_d\frac{m^3}{M_{Pl}^{}{}_{}{}^{2}}`$, where $`m`$ is the inflaton mass and $`m10^{13}`$ GeV from the COBE data on the CMBR anisotropy . Efficient inflaton decay occurs at the time when $`H\mathrm{\Gamma }_d`$ and the effective reheat temperature at that time will be $`T_R(\mathrm{\Gamma }_dM_{Pl})^{\frac{1}{2}}`$. For $`m10^{13}`$ GeV we get $`T_R10^{10}`$ GeV, which is in the allowed range to avoid the gravitino problem .
The crucial point to note is that, although inflaton decay effectively completes much later than the start of its oscillation, nonetheless decay occurs throughout this period. In fact, a dilute plasma with temperature T<
(HฮdMPl2)14๐<
superscript๐ปsubscriptฮ๐superscriptsubscript๐๐๐214T{\ \lower-1.2pt\vbox{\hbox{\hbox to0.0pt{$<$\hss}\lower 5.0pt\vbox{\hbox{$\sim$}}}}\ }{(H{\Gamma}_{d}{{M}_{Pl}}^{2})}^{{1\over 4}} (assuming instant thermalization: we address thermalization below) is present from the first several oscillations, until the effective completion of the inflaton decay . It is easily seen that it has the highest instantaneous temperature at the earliest time, which can reach $`T10^{13}`$ GeV. This plasma, however, carries a relatively small fraction of the cosmic energy density, with the bulk still in inflaton oscillations. The dilution of relics produced from this plasma by the entropy release from the subsequent decay of the bulk of the inflaton energy is the reason that it does not lead to gravitino overproduction. It is important to note that the energy density in the plasma may be comparable to the energy density stored in the condensate along a flat direction. As a result, the thermal effects from the plasma may affect the dynamics of flat direction evolution which, as we see below, occurs in many cases.
All fields with mass less than $`T`$, and gauge interactions with the plasma particles, can reach thermal equilibrium with the plasma. Those fields which are coupled to the flat direction have generically large masses in the presence of its vev, and might not be excited thermally. These include the $`\varphi `$ fields which are gauge-coupled to $`\alpha `$ and have a mass $`g\alpha `$ (up to numerical factors of $`O(1)`$) and many of the $`\chi `$ fields which have superpotential couplings to $`\alpha `$, and hence have a mass $`h\alpha `$ (also up to numerical factors of $`O(1)`$). For $`g\alpha >T`$ or $`h\alpha >T`$, the former or the latter are not in thermal equilibrium, respectively. We recall that, in the presence of Hubble-induced soft-breaking terms, the minimum of the potential for the flat direction determines that $`\alpha (HM^{n3})^{\frac{1}{n2}}`$, and the plasma temperature is $`T(H\mathrm{\Gamma }_dM_{Pl}^{}{}_{}{}^{2})^{\frac{1}{4}}`$. So a field with a coupling $`h`$ to the flat direction can be in thermal equilibrium provided that $`h\alpha T`$, which implies that
$$H^{6n}\frac{\mathrm{\Gamma }_{d}^{}{}_{}{}^{n2}}{h^{4(n2)}}\frac{M_{Pl}^{}{}_{}{}^{2(n3)}}{M^{4(n3)}}$$
(6)
and similarly for the gauge coupling $`g`$.
The back-reaction effect of the plasma of quanta of this field will then induce a mass-squared $`+h^2T^2`$ for the flat direction to which it is coupled. If this exceeds the negative Hubble-induced mass-squared $`H^2`$, the flat direction starts its oscillation. This happens for $`hTH`$, i.e., for
$$H^3h^4\mathrm{\Gamma }_dM_{Pl}^{}{}_{}{}^{2}$$
(7)
and similarly for back-reaction from plasma fields with gauge coupling $`g`$ to the flat direction. Therefore, a flat direction will start its oscillations if both of the above conditions are satisfied simultaneously. We note that the finite-temperature effects of the plasma can lead to a much earlier oscillatory regime for the flat direction, i.e., when $`Hm_{\frac{3}{2}}`$.
It is clear that, in order for a plasma of the quanta of a field to be produced, the coupling of that field to a flat direction should not be so large that its induced mass prevents its thermal excitation. On the other hand, in order for its thermal plasma to have a significant reaction back on the flat direction, its coupling to the flat direction should not be so small that the thermal mass-squared induced for the flat direction will be smaller than the Hubble-induced contribution. Therefore, to have significant thermal effects, we need couplings of intermediate strength in order to have both conditions simultaneously satisfied. For the fields $`\varphi `$ which have $`D`$-term couplings of gauge strength $`g`$ to flat directions, this is usually not the case: As will be seen shortly, in most cases their couplings are too large to satisfy the equilibrium condition. For the fields $`\chi `$ which have $`F`$-term couplings of Yukawa strength $`h`$ to the flat direction, the existence of significant thermal effects depends on the value of $`h`$, as well as on the initial value of the flat-direction vev $`\alpha `$, which in turn depends on the mass scale and the dimension of the higher-dimensional operator which lifts the flatness.
To organize our discusion then, we first assess the typical values of $`h`$ to be expected for couplings to the flat directions. For these typical values, we then estimate the importance of thermal effects on vevโs determined by higher-dimensional operators ranging over the various different dimensions that can lift the flat direction, for both the case of lifting by the GUT scale: $`O(10^{16})\mathrm{GeV}`$, and by the Planck scale: $`O(10^{19})\mathrm{GeV}`$.
We now list the Yukawa couplings of the MSSM. For low $`tan\beta `$, the ratio of $`H^u`$ and $`H^d`$ vevโs, we have
$$\begin{array}{ccc}h_{}^{u}{}_{1}{}^{}10^4& h_{}^{d}{}_{1}{}^{}10^5& h_{}^{l}{}_{1}{}^{}10^6\\ h_{}^{u}{}_{2}{}^{}10^2& h_{}^{d}{}_{2}{}^{}10^3& h_{}^{l}{}_{2}{}^{}10^3\\ h_{}^{u}{}_{3}{}^{}1& h_{}^{d}{}_{3}{}^{}10^2& h_{}^{l}{}_{3}{}^{}10^2\end{array}$$
(8)
whilst the $`h^u`$โs and $`h^d`$โs tend to be more similar for high $`tan\beta `$. The only Yukawa couplings which are significantly different from $`O(10^2)`$ are $`h_{}^{u}{}_{1}{}^{}`$, $`h_{}^{d}{}_{1}{}^{}`$, $`h_{}^{l}{}_{1}{}^{}`$, and $`h_{}^{u}{}_{3}{}^{}`$. Only flat directions which include only the left- and right-handed up squark, the left- and right-handed down squark, the left- and right-handed selectron and the left-handed sneutrino will have an $`h`$ significantly less than $`O(10^2)`$.
For low $`tan\beta `$, any flat direction which includes right-handed top squarks has a Yukawa coupling of $`O(1)`$ to some $`\chi `$โs, too large for those $`\chi `$โs to be in thermal equilibrium, given the expected range of flat direction vevโs $`\alpha `$. The left-handed squarks are coupled to both $`H^u`$ and $`H^d`$, so any flat direction which includes a left-handed top squark has a Yukawa coupling of order $`10^2`$ to $`H^d`$ as well. For high $`tan\beta `$, any flat direction which includes the left- or right-handed top or bottom squarks has a Yukawa coupling of order 1 since the top and bottom Yukawas are of the same order. In general, any flat direction which consists only of the above-mentioned scalars has a Yukawa coupling of order 1 to some $`\chi `$โs and/or a Yukawa coupling significantly less than $`O(10^2)`$ to other $`\chi `$โs.
Among all flat directions which are not lifted by the renormalizable superpotential terms, there is only one which allows such a flavor choice: $`uude`$ with one $`u`$ in the third generation and all other scalars in the first generation (i.e., $`tude`$). This exceptional flat direction still has a coupling of $`O(10^4)`$ to some $`\chi `$ fields, since it includes the right-handed up squark. Taking into account all flavor choices for all flat directions which are not lifted at the renormalizable superpotential level, we can use $`h10^2`$ for the coupling of a generic flat directions to $`\chi `$ fields. For the above-mentioned exceptional case we shall use $`h10^4`$.
So, for our discussion of the dynamics of flat direction oscillations we will consider three representative cases. We will analyze the dynamics when inflaton decay plasmons are coupled to the flat direction by: gauge couplings with coupling $`g10^1`$, generic Yukawa couplings of order $`h10^2`$, or suppressed Yukawa couplings of order $`h10^4`$. Consideration of these cases should allow us to explore the generic range of physical effects that arise in flat direction oscillations, from a plasma of inflaton decay products.
We now undertake a detailed analysis to determine in which cases a plasma of inflaton decay products can be produced, and can initiate the flat-direction oscillations by the reaction they induce on the flat direction. Whether this occurs or not depends on the vev of the flat direction, and the strength of the coupling of the plasma quanta to the flat direction. The initial vev of the flat direction is set by both the underlying scale of the physics of the higher-dimensional operators that lift the flat direction, and, for a given flat direction, by the dimension of the gauge-invariant operator of lowest dimension which can be induced by the underlying dynamics to lift the flat direction.
In order to categorize systematically the various cases which arise, we organize them as follows. First, we divide them into two cases, depending on whether the underlying scale of the new physics responsible for the higher-dimensional operators which lift the flat direction and stabilize the vev at the end of inflation are GUT-scale: $`O(10^{16})`$ GeV, or Planck-scale: $`O(10^{19})`$ GeV. Each of these cases is subdivided according to whether the coupling between the flat direction and the inflaton decay products is of gauge strength ($`g10^1`$), standard superpotential Yukawa strength ($`h10^2`$), or exceptional suppressed Yukawa strength ($`h10^4`$). As noted above, this covers the generic range of couplings exhibited by fields in flat directions in the supersymmetric standard model. Finally, each of these cases is subdivided and tabulated according to the dimension of the operator that stabilizes the flat-direction vev, setting (given the possibilities listed above for the underlying scale of the new physics responsible for the operators) the initial vev of the flat direction. These higher-dimensional operators are listed by the order of the monomial in the superfields which appears in the superpotential and is responsible for the operator. We tabulate against the order of the higher-dimensional superpotential term the following quantities (in Planck units <sup>2</sup><sup>2</sup>2From now on, we express some dimensionful quantities in Planck units.): the Hubble constant $`H`$, the temperature $`T`$, and the value of the flat-direction vev $`\alpha `$ at the onset of oscillations, as well as the combination $`\frac{hT^2}{H\alpha }`$ ($`\frac{gT^2}{H\alpha }`$ for the case of the gauge coupling) which will be useful when we discuss the produced baryon asymmetry in the next section. We also explain the reasons for the values of the entries appearing, in the light of the two necessary conditions introduced above for inducing the flat-direction oscillations by plasma effects, i.e., that on the one hand the mass of the plasmon induced by the coupling to the flat direction is small enough that it can be populated in the thermal bath from inflaton decay, and, on the other hand, that the coupling is large enough for back-reaction effects from the plasma to lift the flat direction sufficiently to start oscillation despite the effects of the Hubble-induced mass.
First, let us consider the case that the scale of the new physics that induces operators that stabilize the flat direction is of order the GUT scale: $`O(10^{16})`$ GeV. We then subdivide this case according to the strength of the coupling of the inflaton decay products to the flat direction. To start, we consider the gauge-coupled case with $`g=10^1`$. In this case, it is only for initial flat-direction vevโs fixed by either quartic or quintic higher-dimensional terms in the superpotential that the plasma effects can accelerate the onset of flat direction oscillation, with the results shown in the following Table. Physically, for superpotential monomials of sixth order or higher, the initial flat-direction vev is sufficiently large that the mass generated by its gauge coupling to the prospective inflaton decay products is large enough to prevent them from being kinematically accessible for thermal excitation. In the case of a quintic superpotential monomial this is also initially the case, and it is only after Hubble expansion has reduced $`\alpha `$, and hence the induced plasmon mass, that thermalized products of inflaton decay can back-react to induce flat-direction oscillation. However, this only occurs for $`H<10^{16}`$, by which time the low-energy soft supersymmetry breaking has already initiated flat-direction oscillation.
GUT scale $`M=10^{16}`$ GeV, gauge coupling $`g=10^1`$
$`H`$ $`T`$ $`\alpha `$ $`\frac{gT^2}{H\alpha }`$ $`n=4`$ $`10^8`$ $`10^{\frac{13}{2}}`$ $`10^{\frac{11}{2}}`$ $`10^{\frac{1}{2}}`$ $`n=5`$ $`10^{18}`$ $`10^9`$ $`10^8`$ $`10^7`$
For the GUT case $`M=10^{16}`$ GeV with generic Yukawa coupling $`h=10^2`$, we have the results shown in the following Table for lifting of the flat direction by monomials of the orders listed. In the cases that the order of the monomial is four or five we have no difficulty satisfying the condition that $`h\alpha T`$, so that they are (thermally) populated in the inflaton decay plasma. For monomials of order six, seven or eight, the induced mass of the prospective plasmon is, in fact, of the same order or slightly larger than the instantaneous effective temperature. So thermally they are present, albeit now with some Boltzman suppression. Moreover, we also note that these induced masses are less than the mass of the decaying inflaton, and so they will be produced in the cascade of inflaton decay products, though, as noted above, after complete thermalization they will be subject to some Boltzmann suppression. In all cases the value of the Hubble constant at the onset of oscillation will be determined by the second condition ($`hTH`$), which requires that the back-reaction-induced mass overcome the Hubble-induced mass to initiate oscillation. By comparing the results of the Tables for $`g=10^1`$ and $`h=10^2`$, we note that for a general flat direction with $`h=10^2`$ which is lifted at the $`n=4`$ superpotential level, the values at the onset of oscillations should be taken from the gauge analysis. The reason is that, in this case, the back-reaction of the inflaton decay products which have gauge coupling to the flat direction act at an earlier time than the back-reaction of those decay products which have Yukawa couplings to it.
GUT scale $`M=10^{16}`$ GeV, standard Yukawa coupling $`h=10^2`$
$`H`$ $`T`$ $`\alpha `$ $`\frac{hT^2}{H\alpha }`$ $`n=4`$ $`10^{\frac{26}{3}}`$ $`10^{\frac{20}{3}}`$ $`10^{\frac{35}{6}}`$ $`10^{\frac{2}{3}}`$ $`n=5`$ $`10^{\frac{26}{3}}`$ $`10^{\frac{20}{3}}`$ $`10^{\frac{44}{9}}`$ $`10^{\frac{16}{9}}`$ $`n=6`$ $`10^9`$ $`10^{\frac{27}{4}}`$ $`10^{\frac{64}{15}}`$ $`10^2`$ $`n=7`$ $`10^{\frac{28}{3}}`$ $`10^{\frac{41}{6}}`$ $`10^{\frac{64}{15}}`$ $`10^{\frac{31}{15}}`$ $`n=8`$ $`10^{\frac{34}{3}}`$ $`10^{\frac{22}{3}}`$ $`10^{\frac{79}{18}}`$ $`10^{\frac{17}{18}}`$
For $`M=10^{16}`$ GeV, $`h=10^4`$, as a function of the order of the superpotential monomial lifting the flat direction we have the results shown in the next Table. For these cases, the flat-direction-induced mass is always less than the instantaneous temperature, due to the weak coupling of the flat direction to the plasmons. The only non-trivial condition now is the second one ($`hTH`$), which determines how long one must wait before the Hubble-induced mass is sufficiently reduced that the back-reaction-induced flat-direction mass can overcome it to initiate oscillation. This fixes the value of $`H`$ at the onset of oscillation. Comparing the results of the Tables for $`g=10^1`$ and $`h=10^4`$, we note that for an exceptional flat direction with $`h=10^4`$ which is lifted at the $`n=4`$ superpotential level, the values at the onset of oscillations should also be taken from the gauge analysis.
GUT scale $`M=10^{16}`$ GeV, exceptional Yukawa coupling $`h=10^4`$
$`H`$ $`T`$ $`\alpha `$ $`\frac{hT^2}{H\alpha }`$ $`n=4`$ $`10^{\frac{34}{3}}`$ $`10^{\frac{22}{3}}`$ $`10^{\frac{43}{6}}`$ $`10^{\frac{1}{2}}`$ $`n=5`$ $`10^{\frac{34}{3}}`$ $`10^{\frac{22}{3}}`$ $`10^{\frac{52}{9}}`$ $`10^{\frac{14}{9}}`$ $`n=6`$ $`10^{\frac{34}{3}}`$ $`10^{\frac{22}{3}}`$ $`10^{\frac{61}{12}}`$ $`10^{\frac{9}{4}}`$ $`n=7`$ $`10^{\frac{34}{3}}`$ $`10^{\frac{22}{3}}`$ $`10^{\frac{14}{3}}`$ $`10^{\frac{8}{3}}`$ $`n=8`$ $`10^{\frac{34}{3}}`$ $`10^{\frac{22}{3}}`$ $`10^{\frac{79}{18}}`$ $`10^{\frac{53}{18}}`$ $`n=9`$ $`10^{\frac{34}{3}}`$ $`10^{\frac{22}{3}}`$ $`10^{\frac{88}{21}}`$ $`10^{\frac{22}{7}}`$
We now turn to the case that the underlying scale of new physics responsible for generating the higher-dimensional operators is the Planck scale. This means that the values of the flat direction vevโs after inflation will be larger, raising the mass of prospective plasmons to which they couple, and making it harder to satisfy the constraint that these putative plasmons be generated thermally, or even be kinematically accessible to inflaton decay.
For $`M=10^{19}`$ GeV, $`g=10^1`$, we have significant effects only for flat directions lifted by superpotential terms arising from quartic or quintic monomials. In all other cases ($`n6`$) the flat direction vev is so large that quanta gauge-coupled to it receive sufficiently large masses that they can not be thermally populated at the instantaneous temperature of the inflaton decay products. For the two non-trivial cases we have the results shown in the following Table. Only in the $`n=4`$ case can we produce thermally a number of plasma quanta sufficient to induce enough mass for the flat direction to initiate its oscillation at an earlier time. In the $`n=5`$ case, back-reaction from the plasma of inflaton decay products only manages to induce flat-direction oscillation after $`H10^{18}`$, by which time the low-energy soft supersymmetry breaking has already acted to start the oscillation and also the inflaton decay has been completed.
Planck scale $`M=10^{19}`$ GeV, gauge coupling $`g=10^1`$
$`H`$ $`T`$ $`\alpha `$ $`\frac{gT^2}{H\alpha }`$ $`n=4`$ $`10^{14}`$ $`10^8`$ $`10^7`$ $`10^4`$ $`n=5`$ $`10^{42}`$ $`10^{15}`$ $`10^{14}`$ $`10^{25}`$
For $`M=10^{19}`$ GeV and $`h=10^2`$, we again have a case where flat directions lifted by superpotential monomials of order six or higher result in such a large flat-direction vev that quanta coupled to it receive too large a mass for them to be thermally excited in the plasma of inflaton decay products. For the cases of quartic or quintic superpotential monomials, we have the results shown in the following Table. We again find that only in the $`n=4`$ case can thermal effects actually induce sufficient mass for the flat direction to initiate oscillation earlier. In the $`n=5`$ case, back-reaction from the plasma of inflaton decay products only manages to induce flat-direction oscillation after the low-energy soft supersymmetry breaking has already done so, and the inflaton decay has been completed. By comparing the results of the Tables for $`g=10^1`$ and $`h=10^2`$, we note that, for a generic flat direction, i.e., one with $`h=10^2`$, the initial values at the onset of oscillations should be taken from the latter. The reason is that, in this case, the back-reaction of the inflaton decay products which have Yukawa couplings to the flat direction act at an earlier time than the back-reaction of those decay products with a gauge coupling to it.
Planck scale $`M=10^{19}`$ GeV, standard Yukawa coupling $`h=10^2`$
$`H`$ $`T`$ $`\alpha `$ $`\frac{hT^2}{H\alpha }`$ $`n=4`$ $`10^{10}`$ $`10^7`$ $`10^5`$ $`10^1`$ $`n=5`$ $`10^{30}`$ $`10^{12}`$ $`10^6`$ $`10^{14}`$
Finally, in the case $`M=10^{19}`$ GeV, $`h=10^4`$, for flat directions lifted by monomials higher than sixth order the resulting flat-direction vevs are sufficiently large that quanta coupled to it with this coupling are too massive to be excited at the instantaneous temperature of the inflaton decay products. So the nontrivial cases are those in the following Table. For $`n=6`$, we marginally satisfy the requirement that $`hT\alpha `$, necessary for thermal production of quanta coupled to the flat direction, while for $`n=4`$ and $`n=5`$ we do so comfortably. The second condition, that $`hTH`$ for effective back-reaction, then serves to determine the value of $`H`$ at the onset of the thermally-induced oscillation. By comparing the results of the Tables for $`g=10^1`$ and $`h=10^4`$, we note that when the exceptional flat direction with $`h=10^4`$ is lifted at the $`n=4`$ superpotential level, the values at the onset of oscillations should be taken from the latter. This is because the back-reaction of the inflaton decay products with Yukawa coupling to the flat direction act at an earlier time than the back-reaction of those decay products with gauge couplings to it.
Planck scale $`M=10^{19}`$ GeV, exceptional Yukawa coupling $`h=10^4`$
$`H`$ $`T`$ $`\alpha `$ $`\frac{hT^2}{H\alpha }`$ $`n=4`$ $`10^{\frac{34}{3}}`$ $`10^{\frac{22}{3}}`$ $`10^{\frac{17}{3}}`$ $`10^{\frac{5}{3}}`$ $`n=5`$ $`10^{\frac{34}{3}}`$ $`10^{\frac{22}{3}}`$ $`10^{\frac{34}{9}}`$ $`10^{\frac{32}{9}}`$ $`n=6`$ $`10^{\frac{38}{3}}`$ $`10^{\frac{23}{3}}`$ $`10^{\frac{19}{6}}`$ $`10^{\frac{7}{2}}`$
We noted above that thermal effects from the plasma can be important up to $`h\alpha T`$ or even somewhat higher. For $`\alpha `$ less than this, they change the convexity of the effective potential in the $`\alpha `$ direction at much earlier times, inducing the onset of flat-direction oscillations. We should note that since $`\alpha H^{\frac{1}{n2}}`$ and $`TH^{\frac{1}{4}}`$, then $`\alpha `$ decreases at the same rate as, or more slowly than, $`T`$ for $`7n9`$. This means that if $`h\alpha T`$ right after the end of inflation, it will remain so for later times as well. Therefore, in the $`7n9`$ cases for $`M=10^{19}`$ GeV, the Hubble-induced negative mass-squared is dominant and $`\alpha `$ will not be lifted until $`H10^{16}`$, if $`h\alpha T`$ at $`H10^6`$.
In sum, we conclude that for $`M=M_{GUT}`$, a general flat direction with $`h10^2`$ starts oscillating at $`H10^{16}`$ in the $`4n8`$ cases. For the exceptional one with $`h10^4`$ it is true in the $`n=9`$ case as well. For $`M=M_{Planck}`$ in the denominator, only in the $`n=4`$ case do oscillations of a general flat direction start at $`H10^{16}`$. In the $`5n9`$ cases, the flat direction is protected from thermal effects because its large vev induces such a large mass for fields coupled to it that they cannot be thermally excited in the plasma of inflaton decay products. For the exceptional flat direction with $`h10^4`$, this protection is weaker because of the smaller Yukawa coupling to $`\chi `$ (which therefore are lighter and can be excited in thermal equilibrium) and, as a result, oscillations start at $`H10^{16}`$ in the $`n=5,6`$ cases also.
We need to elaborate on the implicit assumption that the $`\chi `$โs ($`\varphi `$โs) are effectively thermal upon production. In the model that we study, the inflaton decays in the perturbative regime, and the decay products have a momentum less than, or comparable to, the inflaton mass $`m10^6`$. The $`\chi `$โs ($`\varphi `$โs) which are produced in two-body decays have a momentum of order $`m`$ <sup>3</sup><sup>3</sup>3The $`\varphi `$โs generically have larger $`\alpha `$-induced masses than do the $`\chi `$โs, so their production may be delayed until $`\alpha `$ Hubble-dilutes to a smaller value.. It can easily be seen that the temperature at which oscillations start (assuming thermal equilibrium) is $`10^7`$ in all the above cases. Since the momentum of produced particles is greater than the the average thermal momentum, the dominant process to reach equilibrium is through the decay of $`\chi `$โs ($`\varphi `$โs) to other particles with smaller momenta. However, the momentum of $`\chi `$โs ($`\varphi `$โs) is very close to the average thermal momentum. Since thermalization does not change the energy density in the plasma, the number density of $`\chi `$โs ($`\varphi `$โs) is also close to its thermal distribution. Therefore, the plasma-induced mass-squared $`h^2\frac{n_\chi }{E_\chi }`$ ($`g^2\frac{n_\varphi }{E_\varphi }`$) from $`\chi `$โs ($`\varphi `$โs) is of the same order as $`h^2T^2`$ ($`g^2T^2`$).
## 4 Thermal $`A`$ Terms and Baryo/Leptogenesis
Motion along the angular direction is required for the build-up of a baryon or lepton asymmetry. This is possible if a torque is exerted on $`\alpha `$ or, equivalently, if $`\alpha `$ is not in one of the discrete minima along the angular direction, when it starts oscillating. These discrete minima are due to the $`A`$ term part of the potential. Before the start of oscillations, the Hubble-induced $`A`$ terms are dominant, and the locations of the minima are determined by them. During inflation, $`\alpha `$ rolls down towards one of these minima and rapidly settles there. After inflation it tracks that minimum and there is no motion along the angular direction . What is necessary then is a non-adiabatic change in the location of the minima, such that at the onset of oscillations $`\alpha `$ is no longer in a minimum along the angular direction. In the absence of thermal effects, $`\alpha `$ would start its angular motion (as well as its linear oscillations) at $`Hm_{\frac{3}{2}}`$. This occurs as a result of uncorrelated phases of the $`A`$ terms induced by the Hubble expansion and low-energy supersymmetry breaking. At this time, the latter takes over from the former, and $`\alpha `$ will in general no longer be in a minimum along the angular direction. This will lead to the generation of a baryon or lepton asymmetry if $`\alpha `$ carries a non-zero number of either .
As we have seen above, due to thermal effects, in many cases the flat directions start oscillating at much larger $`H`$. At this time the Hubble-induced $`A`$ terms are still much larger than the low-energy ones from hidden-sector supersymmetry breaking. In order to have angular motion for $`\alpha `$, another $`A`$ term of size comparable to the Hubble-induced one, but with uncorrelated phase, is required. Since it is finite-temperature effects from the plasma that produce a mass-squared which dominates the Hubble-induced one, one might expect that the same effects also produce an $`A`$ term which dominates the Hubble-induced $`A`$ term. This is the only new effect that could produce such an $`A`$ term with uncorrelated phase, as the thermal plasma is the only difference from the standard scenario.
The simplest such thermal $`A`$ terms arise at tree-level from cross terms from the following two terms in superpotential
$$h\alpha \chi \chi +\lambda _n\frac{\alpha ^n}{nM^{n3}}$$
(9)
which results in the contribution
$$h\lambda _n\frac{\chi _{}^{}{}_{}{}^{2}\alpha ^{n1}}{M^{n3}}+h.c.$$
(10)
in the scalar potential. In thermal equilibrium, $`<\chi _{}^{}{}_{}{}^{2}>`$ can be approximated by $`T^2`$ and therefore the thermal $`A`$ term is of order
$$h\lambda _n\frac{T^2\alpha ^{n1}}{M^{n3}}$$
(11)
There is another thermal $`A`$ term that arises from one-loop diagrams with gauginos and fermionic partners of $`\alpha `$. It results in a contribution, in the thermal bath, of order:
$$\lambda _n\left(\frac{gT}{4\pi \alpha }\right)^2\frac{T\alpha ^n}{M^{n3}}$$
(12)
We have checked that, for the parameter range of interest for this process, this has the same order of magnitude as the tree-level $`A`$ term. In the following, we use the tree-level term for our estimates.
The ratio of the thermal $`A`$ term to the Hubble-induced one is $`\frac{hT^2}{H\alpha }`$. It is clear from the results summarized in the Tables that, at the onset of $`\alpha `$ oscillations, the thermal $`A`$ term is weaker than the Hubble-induced one in all cases. Therefore, at this time, the minimum along the angular direction is slightly shifted, the curvature at the minimum is still determined by the Hubble-induced $`A`$ term, and the force in the angular direction is of order $`\lambda _n\frac{hT^2\alpha ^{n2}}{M^{n3}}`$. The ratio of the thermal $`A`$ term to the Hubble-induced one grows, however, as $`\frac{hT^2}{H\alpha }`$ increases in time. Therefore, in what follows we keep both $`A`$ terms in the equation of motion of $`\alpha `$.
We consider the case where oscillations start because of the back-reaction of the $`\chi `$ fields, as is the most common case. The masses of $`\alpha _R`$ and $`\alpha _I`$ are then of order $`hT`$ <sup>4</sup><sup>4</sup>4We use the estimate $`gT`$ for the plasma-induced masses if oscillations start because of the back-reaction of the $`\varphi `$ fields.. The equation of motion for the flat direction is then:
$$\ddot{\alpha }+3H\dot{\alpha }+h^2T^2\alpha +(n1)h\lambda _n\frac{T^2\alpha _{}^{}{}_{}{}^{n2}}{M^{n3}}+A\lambda _n\frac{H\alpha _{}^{}{}_{}{}^{n1}}{M^{n3}}+(n1)\lambda _{n}^{}{}_{}{}^{2}\frac{|\alpha |^{2(n2)}}{M^{2(n3)}}\alpha =\mathrm{\hspace{0.33em}0}$$
(13)
At this time, the universe is matter-dominated, by the oscillating inflaton field, and thus $`H=\frac{2}{3t}`$. Also, $`T^2=(H\mathrm{\Gamma }_dM_{Pl}^{}{}_{}{}^{2})^{\frac{1}{2}}t^{\frac{1}{2}}`$ for $`H10^{18}`$. After re-scaling $`\alpha (\frac{H_{osc}M^{n3}}{\lambda _n})^{\frac{1}{n2}}\alpha `$ and $`tH_{osc}t`$, where $`H_{osc}`$ is the Hubble constant at the onset of oscillations, we get the following equations of motion for the real and imaginary components of $`\alpha `$:
$$\begin{array}{c}\ddot{\alpha _R}+\frac{2}{t}\dot{\alpha _R}+a\frac{\alpha _R}{t^{\frac{1}{2}}}+b\frac{|\alpha |^{n2}}{t^{\frac{1}{2}}}\mathrm{cos}((n2)\theta +\phi )+A\frac{|\alpha |^{n1}}{t}\mathrm{cos}((n1)\theta )+(n1)|\alpha |^{2(n2)}\alpha _R=\mathrm{\hspace{0.33em}0}\\ \\ \\ \ddot{\alpha _I}+\frac{2}{t}\dot{\alpha _I}+a\frac{\alpha _I}{t^{\frac{1}{2}}}b\frac{|\alpha |^{n2}}{t^{\frac{1}{2}}}\mathrm{sin}((n2)\theta +\phi )A\frac{|\alpha |^{n1}}{t}\mathrm{sin}((n1)\theta )+(n1)|\alpha |^{2(n2)}\alpha _I=\mathrm{\hspace{0.33em}0}\end{array}$$
(14)
Here $`a\frac{h^2T_{osc}^{}{}_{}{}^{2}}{H_{osc}^{}{}_{}{}^{2}}`$: $`T_{osc}=(H_{osc}\mathrm{\Gamma }_dM_{Pl}^{}{}_{}{}^{2})^{\frac{1}{2}}`$ is the plasma temperature at the onset of oscillations, $`b(n1)a\lambda _n\frac{(\frac{H_{osc}M^{n3}}{\lambda _n})^{\frac{n3}{n2}}}{M^{n3}}`$, and $`\phi =O(1)`$ is the relative phase between the thermal and Hubble-induced $`A`$ terms.
The first two terms and the superpotential term in each of these equations are the same as in the equations derived in , but there are some important differences. First of all, the flat-direction mass-squared is not the (constant) low-energy value $`m_{\frac{3}{2}}^{}{}_{}{}^{2}`$, but the thermal mass which is redshifted as $`t^{\frac{1}{2}}`$. Also, the Hubble-induced $`A`$ term with coefficient $`H`$ appears instead of the low-energy one with coefficient $`m_{\frac{3}{2}}`$, which is negligible for $`H10^{16}`$. This explains the $`\frac{1}{t}`$ factor in front of the Hubble-induced $`A`$ term. Finally, there is another $`A`$ term, the thermal one, which is also redshifted as $`t^{\frac{1}{2}}`$, because of its $`T^2`$ dependence.
At the onset of oscillations, $`t=t_i=\frac{2}{3}`$ and $`\alpha `$ is in one of the minima which are determined by the Hubble soft terms. Therefore, $`|\alpha |_i`$, which was $`(\frac{H_{osc}M^{n3}}{\lambda _n})^{\frac{1}{n2}}`$ before re-scaling, is scaled to $`|\alpha |_i=1`$, and also $`\theta _i=n\pi `$. We have solved these equations numerically for $`\theta _i=0`$, $`A=(n1)`$ and $`\lambda _n=1`$, and calculated $`n_B=\alpha _R\frac{\alpha _I}{t}\alpha _I\frac{\alpha _R}{I}`$. We find that, among the cases listed in the above-mentioned Tables, only for the following ones do we get an $`\frac{n_B}{s}`$ of order $`10^{11}`$ or larger, before any subsequent dilution after reheating.
The value of $`\frac{n_B}{s}`$ for flat directions which undergo plasma-induced oscillations
| | $`M=10^{16}`$ GeV | | $`M=10^{19}`$ GeV | |
| --- | --- | --- | --- | --- |
| | $`h=10^2`$ | $`h=10^4`$ | $`h=10^2`$ | $`h=10^4`$ |
| $`n=4`$ | $`<10^{11}`$ | $`<10^{11}`$ | $`<10^{11}`$ | $`3\times 10^{11}`$ |
| $`n=5`$ | $`<10^{11}`$ | $`3\times 10^{11}`$ | $`\mathrm{no}\mathrm{plasma}\mathrm{effect}`$ | $`3\times 10^9`$ |
| $`n=6`$ | $`10^{11}`$ | $`4\times 10^{11}`$ | $`\mathrm{no}\mathrm{plasma}\mathrm{effect}`$ | $`10^7`$ |
| $`n=7`$ | $`4\times 10^{11}`$ | $`10^{10}`$ | $`\mathrm{no}\mathrm{plasma}\mathrm{effect}`$ | $`\mathrm{no}\mathrm{plasma}\mathrm{effect}`$ |
| $`n=8`$ | $`10^{10}`$ | $`3\times 10^{10}`$ | $`\mathrm{no}\mathrm{plasma}\mathrm{effect}`$ | $`\mathrm{no}\mathrm{plasma}\mathrm{effect}`$ |
| $`n=9`$ | $`\mathrm{no}\mathrm{plasma}\mathrm{effect}`$ | $`5\times 10^{10}`$ | $`\mathrm{no}\mathrm{plasma}\mathrm{effect}`$ | $`\mathrm{no}\mathrm{plasma}\mathrm{effect}`$ |
We see that, in some cases, $`\frac{n_B}{s}`$ is near the observed value of $`5\times 10^{10}`$. However, in the most general case, when the standard model gauge group is the only symmetry group, these viable flat directions constitute only a small subset of all flat directions. We also see that $`\frac{n_B}{s}`$ is larger for the exceptional flat directions, when $`M=10^{19}`$ GeV, and for flat directions which are lifted by terms of higher order $`n`$. This is easily understandable, as for larger $`M`$ and $`n`$, and for smaller $`h`$, plasma-induced oscillations start later and closer to the efficient reheat epoch $`H=10^{18}`$. Larger $`M`$ and $`n`$ lead to a larger vev for the flat direction and, therefore, the condition $`h\alpha T`$ will be satisfied at a later time. A smaller value of $`h`$, on the other hand, implies that the condition $`hTH`$ will be satisfied at a later time. Later oscillations mean less dilution of the generated lepton/baryon asymmetry by the plasma of inflaton decay products (we recall that $`sT^3`$ is redshifted only as $`t^{\frac{3}{4}}`$ for $`H10^{18}`$).
Now we comment how our results may be affected by changes in the model-dependent constants involved in the calculations: the reheat temperature $`T_R`$ (or equivalently the inflaton decay rate), and the constant $`\frac{C_I}{(n1)\lambda _n}`$ which appears in the expression for the flat-direction vev. There are two concerns in this regard. First, whether the two conditions for plasma-induced $`\alpha `$ oscillations still result in a consistent value for $`H_{osc}`$ which is greater than $`10^{16}`$, and, secondly, what is the corresponding change in the estimated value for $`\frac{n_B}{s}`$. In our calculations, we have used $`T_R10^9`$ and $`\frac{C_I}{(n1)\lambda _n}1`$. If we assume instead that $`T_R10^{10}`$ and $`\frac{C_I}{(n1)\lambda _n}10^1`$, it turns out that for all cases except the marginal ones (the $`n=6,7,8`$ cases for $`h=10^2`$ and $`M=10^{16}`$ GeV, and the $`n=6`$ case for $`h=10^4`$ and $`M=10^{19}`$ GeV), plasma effects still trigger the oscillations for $`H10^{16}`$, though at a somewhat smaller value of $`H`$. Moreover, the value of $`\frac{n_B}{s}`$ remains the same within an order of magnitude. Therefore, the plasma-induced oscillations of the (non-marginal) flat directions, and the resulting value of $`\frac{n_B}{s}`$, are rather insensitive to the exact order of magnitude of $`T_R`$ and $`\frac{C_I}{(n1)\lambda _n}`$, at least for our purposes.
## 5 Evaporation of the Flat Direction
Now let us find the time when the $`\alpha `$ condensates are knocked out of the zero mode by the thermal bath. For the evaporation to happen, it is necessary that the thermal bath includes those particles which are coupled to $`\alpha `$. Then, two conditions should be satisfied: first, the scattering rate of $`\alpha `$ off the thermal bath must be sufficient for equilibriation, and secondly, the energy density in the bath must be greater than that in the condensate. The flat direction has couplings both of Yukawa strength $`h`$ to $`\chi `$โs and of gauge strength $`g`$ to $`\varphi `$โs. The conditions for thermal production of $`\chi `$โs and $`\varphi `$โs are $`h\alpha T`$ and $`g\alpha T`$, respectively. Since $`h<g`$, the $`\chi `$โs will come to thermal equilibrium at an earlier time. On the other hand, the scattering rate of $`\alpha `$ off the thermal $`\chi `$โs is $`\mathrm{\Gamma }_{scatt.}h^4T`$ while the rate for scattering of $`\alpha `$ off the thermal $`\varphi `$โs is $`\mathrm{\Gamma }_{scatt.}g^4T`$. Therefore, $`\chi `$โs are produced earlier but in general have a smaller scattering rate. The competition between the $`\chi `$โs and $`\varphi `$โs, and between the ratio of the energy density of the flat direction to the energy density in the plasma will determine whether and how the flat direction evaporates.
First we consider those flat directions which have plasma-induced oscillations. If oscillations start due to the back-reaction of $`\chi `$โs (which is the situation for most cases) $`\mathrm{\Gamma }_{scatt.}h^4T`$. For a general flat direction with $`h10^2`$, this is comparable to $`H`$ at $`H10^{17}`$, while for the exceptional flat direction with $`h10^4`$ this occurs at a much smaller $`H`$. However, after $`\alpha `$ starts its oscillation, it is redshifted as $`t^{\frac{7}{8}}`$ while $`T`$ is redshifted as $`t^{\frac{1}{4}}`$. This implies that $`\frac{g\alpha }{T}`$ decreases rapidly and soon the $`\varphi `$โs will be in thermal equilibrium. The rate for scattering of $`\alpha `$ off thermal $`\varphi `$โs is $`\mathrm{\Gamma }_{scatt.}g^4T`$ and $`\mathrm{\Gamma }_{scatt.}H`$ at $`H10^{12}`$. The energy density in the condensate at the onset of oscillations is $`h^2\alpha ^2T^2T^4`$ (recall that $`h\alpha T`$ at this time). The ratio of the two energy densities is further redshifted as $`t^{\frac{5}{4}}`$ (for $`H10^{18}`$) which ensures the second necessary condition for the evaporation of condensate, i.e., that the plasma energy density is dominant over the energy density in the condensate. It can easily be checked that the condensate evaporates at $`H10^{18}`$, before the inflaton decay is completed <sup>5</sup><sup>5</sup>5If $`\alpha `$ oscillations start due to the back-reaction of $`\varphi `$โs, from the beginning $`g\alpha T`$ and $`\mathrm{\Gamma }_{scatt.}g^4T`$. Therefore, there is no need to wait for further redshift of $`\alpha `$ and again the condensate evaporates at $`H10^{18}`$..
In those cases in which the plasma effects do not lead to an early oscillation of the flat direction, oscillations start at $`H10^{16}`$, when the low-energy supersymmdetry breaking takes over the Hubble-induced one. It is important to find the time when the condensate will evaporate in these cases too. For such flat directions, the ratio of the baryon number density to the condensate density is of order one . Therefore, if the condensate dominates the energy density of the universe before evaporation, the resulting $`\frac{n_B}{s}`$ will also be of order one. Some regulating mechanism is then needed in order to obtain the value for successful big bang nucleosynthesis: $`\frac{n_B}{s}10^{10}`$ .
Now consider a general flat direction with $`h10^2`$. As we showed, in the $`5n9`$ cases for $`M=10^{19}`$ GeV, and the $`n=9`$ case for $`M=10^{16}`$ GeV, plasma effects are not important and the flat direction starts oscillating at $`H10^{16}`$. By $`H10^{18}`$ the inflaton has efficiently decayed and $`\alpha `$ has been redshifted by a factor of $`10^2`$. From then on, the universe is radiation-dominated, so $`\alpha t^{\frac{3}{4}}`$ and $`Tt^{\frac{1}{2}}`$. Therefore, the energy density in the condensate is redshifted as $`t^{\frac{3}{2}}`$ whilst the energy density in radiation is redshifted as $`t^2`$. If the condensate does not evaporate (or decay) until very late times, its energy density dominates that of the radiation and universe will again be matter-dominated. At the beginning of oscillations, i.e., at $`H10^{16}`$, $`\alpha `$ has the largest vev in the $`n=9`$ case for $`M=10^{19}`$ GeV, which is $`\alpha 10^{\frac{16}{7}}`$. At $`H10^{18}`$ this is redshifted to $`\alpha 10^{\frac{30}{7}}`$ which still leaves $`h\alpha >T`$, so plasmons with this Yukawa coupling to the flat direction cannot be produced. However, since $`\alpha `$ redshifts more rapidly than $`T`$, eventually $`h\alpha `$ becomes of order $`T`$, after a time such that
$$T10^{\frac{101}{7}},\alpha 10^{\frac{87}{7}}$$
(15)
It is easily seen that at this time the energy density in the condensate and in the radiation are of the same order. Moreover, $`\mathrm{\Gamma }_{scatt.}10^8TH`$ and the condensate evaporates promptly. This case is marginal as the condensate almost dominates the energy density of the universe at evaporation.
In the $`5n8`$ cases for $`M=10^{19}`$ GeV and the $`n=9`$ case for $`M=10^{16}`$ GeV the vev is considerably smaller and the energy density in radiation is even more dominant. Therefore, a general flat direction with $`h10^2`$ will evaporate before dominating the energy density of the universe. We summarize the situation for a general flat direction with $`h10^2`$, regarding both the early, i.e., plasma-induced, oscillation, and evaporation, in the following Table.
Viability of scenarios with generic Yukawa coupling $`h=10^2`$
| | $`M=10^{16}`$ GeV | | $`M=10^{19}`$ GeV | |
| --- | --- | --- | --- | --- |
| | Early Oscillation | Evaporation | Early Oscillation | Evaporation |
| $`n=4`$ | $``$ | $``$ | $``$ | $``$ |
| $`n=5`$ | $``$ | $``$ | | $``$ |
| $`n=6`$ | $`\mathrm{marginal}`$ | $``$ | | $``$ |
| $`n=7`$ | $`\mathrm{marginal}`$ | $``$ | | $``$ |
| $`n=8`$ | $`\mathrm{marginal}`$ | $``$ | | $``$ |
| $`n=9`$ | | $``$ | | $`\mathrm{marginal}`$ |
For the exceptional flat direction with $`h10^4`$ the situation is different. Here plasma effects are not important in the $`7n9`$ cases for $`M=10^{19}`$ GeV. In the $`n=9`$ case the condition for thermal production of $`\chi `$โs, $`h\alpha =T`$ gives
$$T10^{\frac{73}{7}},\alpha 10^{\frac{45}{7}}$$
(16)
which means we do not need as much redshift to reduce $`\alpha `$, so $`\chi `$โs are produced earlier and at a higher temperature. However, $`\mathrm{\Gamma }_{scatt.}10^{16}T`$, which is much smaller than $`H`$ at this time. Therefore, the condensate cannot evaporate by scattering off the $`\chi `$โs. It is easily seen that $`hT>m_{\frac{3}{2}}10^{16}`$ when $`h\alpha =T`$. This implies that the mass and energy density of the flat direction are $`hT`$ and $`h^2\alpha ^2T^2`$, respectively, upon thermal production of $`\chi `$โs, and the energy density in the flat direction and the thermal bath are comparable. As long as $`hTm_{\frac{3}{2}}`$, $`\alpha `$ and $`T`$ are both redshifted as $`t^{\frac{1}{2}}`$. During this interval $`\frac{\alpha }{T}`$ remains constant and the flat direction and plasma energy densities remain comparable. Later, when $`T<10^{12}`$ we have $`hT<m_{\frac{3}{2}}`$, and the energy density in the condensate is $`m_{\frac{3}{2}}^{}{}_{}{}^{2}\alpha ^2`$ and begins to dominate the thermal energy density. At some point, $`g\alpha <T`$ and $`\varphi `$โs can be produced thermally. The scattering rate of the condensate off the $`\varphi `$โs is $`\mathrm{\Gamma }_{scatt.}10^4T`$ which is clearly at equilibrium. However, the energy density in the condensate is now overwhelmingly dominant and evaporation does not occur. For the $`n=7,8`$ cases the situation is similar and the condensate does not evaporate. The summary for the exceptional flat direction, regarding both the early, i.e., plasma-induced, oscillation, and evaporation, is illustrated in the Table below.
Viability of scenarios with exceptional Yukawa coupling $`h=10^4`$
| | $`M=10^{16}`$ GeV | | $`M=10^{19}`$ GeV | |
| --- | --- | --- | --- | --- |
| | Early Oscillation | Evaporation | Early Oscillation | Evaporation |
| $`n=4`$ | $``$ | $``$ | $``$ | $``$ |
| $`n=5`$ | $``$ | $``$ | $``$ | $``$ |
| $`n=6`$ | $``$ | $``$ | $`\mathrm{marginal}`$ | $``$ |
| $`n=7`$ | $``$ | $``$ | | |
| $`n=8`$ | $``$ | $``$ | | |
| $`n=9`$ | $``$ | $``$ | | |
In summary: a general flat direction, i.e., with $`h10^2`$, which does not have plasma-induced early oscillation, does not come to dominate the energy density of the universe (the $`n=9`$ case for $`M=10^{19}`$ GeV is marginal). For the exceptional flat direction, i.e., with $`h10^4`$, the situation is different and it dominates the energy density of the universe before decay.
## 6 Discussion
We have found that all flat directions, except those which are lifted by nonrenormalizable superpotential terms of high dimension and with a large mass scale in the denominator, start oscillating at early times due to plasma effects. For a general flat direction with $`h10^2`$ these are the $`n=4`$ case for $`M=10^{19}`$ GeV and the $`4n8`$ cases for $`M=10^{16}`$ GeV (with the $`6n8`$ cases being marginal and sensitive to model-dependent parameters). For the exceptional flat direction with $`h10^4`$ these are the $`4n6`$ cases for $`M=10^{19}`$ GeV (with the $`n=6`$ case being marginal and sensitive to model-dependent parameters) and all $`n`$ for $`M=10^{16}`$ GeV. In these cases it is difficult to achieve efficient baryon asymmetry generation by the oscillation of the condensate along the flat direction. We showed that a general flat direction, i.e., one with $`h10^2`$, which is not lifted by thermal effects, still evaporates before dominating the energy density of the universe. This is not important for baryogenesis, however, and the resulting dilution by the thermal bath can be used to regulate the $`\frac{n_B}{s}`$ which is initially of order one. On the other hand, the exceptional flat direction, i.e., one with $`h10^4`$, which is not lifted by plasma effects, dominates the energy density of the universe before its decay.
For models with supersymmetry breaking via low-energy gauge mediation, on the other hand, the evaporation of the condensate has yet another implication. In such models there is a candidate for cold dark matter, the so called Q-ball . In order to have stable Q-balls as dark matter candidates, some flat directions must dominate the energy density of the universe. This means that any flat direction which is evaporated by the thermal bath cannot be used to form a Q-ball.
Now the question is which flat directions are lifted by $`n>4`$ terms. A look at reveals that only 18 out of 295 directions which are $`D`$\- and $`F`$-flat at the renormalizable level in the MSSM are not lifted at the $`n=4`$ level. Even a smaller subset of only 2 flat directions are not lifted at the $`n=6`$ level. If nonrenormalizable terms with $`n=4`$ and $`n=5`$ are not forbidden by imposing other symmetries, only a very few flat directions in the MSSM can be used for baryogenesis and even fewer for Q-ball formation (regardless of the mass scale in the denominator or Yukawa couplings of these flat directions). This is if all higher-order terms which respect gauge symmetry exist in the superpotential. With other symmetries (discrete or continuous) imposed on the model, a specific flat direction will, in general, be lifted at a higher level. The initial vev of $`\alpha `$ can then be larger and $`\chi `$ and $`\varphi `$ quanta may not be produced thermally, and the standard treatment of the Affleck-Dine baryogenesis may be valid. Model-dependent analysis is needed to identify at which level a given flat direction is actually lifted, in a given model.
Finally, an interesting possibility is the parametric-resonance decay of a supersymmetric flat direction to the fields $`\varphi `$ to which it is gauge-coupled. The occurence and implications of a potential parametric resonance are more pronounced for those flat directions which start their oscillations at $`H10^{16}`$, as in the standard scenario. They have an incredibly large $`q=(\frac{g\alpha }{2m_{\frac{3}{2}}})^2`$ <sup>6</sup><sup>6</sup>6For those flat directions which have plasma-induced oscillations, $`m_{\frac{3}{2}}`$ is replaced by $`hT`$ or $`gT`$, leading to a considerably smaller $`q`$. which could be as large as $`O(10^{20})`$ (the parameter $`q`$ determines the strength of resonance ). Explosive resonance decay could also prevent these flat directions from dominating the energy density of the universe. However, the situation is too complicated to allow simple estimates based on the results of parametric-resonance decay of a real scalar field. First of all, the renormalizable part of the scalar potential (including the $`D`$-term part which is responsible for parametric-resonance decay to $`\varphi `$โs) is fully known and very complicated. Moreover, the flat direction itself is a complex scalar field. This may result in out-of-phase oscillations in the imaginary part of the flat direction, as well as in other scalar fields which are coupled to the same $`\varphi `$, which can then substantially alter the outcome of simple parametric resonance .
## 7 Conclusion
In conclusion, we have seen that many of the MSSM flat directions may start their oscillations differently than in the standard scenario, where the low-energy supersymmetry breaking determines the onset of oscillations. The two key ingredients for such a different behaviour are: superpotential Yukawa couplings of the flat directions to other fields, and the thermal plasma from partial inflaton decay, whose instantaneous temperature is higher than the reheat temperature. Together, these lead to an earlier start of the oscillations. On the one hand, the masses of those fields which are coupled to the flat direction that are induced by the flat-direction vev are then small enough to be kinematically accessible to inflaton decay and, on the other hand, induce large enough thermal masses for flat directions from the back-reaction of those fields to overcome the negative Hubble-induced mass-squared of the flat directions. Subsequently, thermal masses and $`A`$ terms may be responsible for baryo/leptogenesis, but typically result in an insufficient baryon/lepton asymmetry of the universe. The oscillations are also terminated earlier, due to evaporation of the flat direction through its interactions with the thermal plasma. It was also shown that even for many flat directions whose oscillations are not initiated by plasma effects, these effects cause them to evaporate before dominating the energy density of the universe.
Acknowledgements
The work of RA and BAC was supported in part by the Natural Sciences and Engineering Research Council of Canada, and they would also like to thank the CERN Theory Division for kind hospitality during part of this research. |
warning/0001/hep-ph0001217.html | ar5iv | text | # Relativistic Charge Form Factor of the Deuteron from (np)โScattering Phase Shifts
## A Kinematic variables.
By definition $`s`$ is the invariant mass of $`np`$ system squared:
$`s=(p_n+p_p)_\mu ^2.`$
In laboratory (LS) and center-of-mass (CMS) systems we have
$`s=4M^2+2E=4M^2+4p^2,`$
where $`E`$ is the nucleon energy in LS and $`p`$ is the magnitude of the nucleon 3โmomentum in CMS.
$`Q^2`$ is the magnitude of the 4โmomentum transfer squared:
$`Q^2q_\mu ^2t>0.`$
## B Jost matrices $`B,\stackrel{~}{B}`$.
The formulae for pairs ($`S,B`$) and ($`\stackrel{~}{S},\stackrel{~}{B}`$) have the most convenient form in the $`p`$โplane:
$`S(p)B_+(p)=B_{}(p),\text{ }\mathrm{}<p<\mathrm{},`$
where $`SS[\delta (p),\eta (p),\epsilon (p)]`$, see eq.(5). Let us introduce two new matrices $`\stackrel{~}{S}`$ and $`\stackrel{~}{B}`$:
$`\begin{array}{ccc}\hfill \stackrel{~}{B}_\pm (p)& =R(p)B_\pm (p),\hfill & \\ \hfill R(p)& =I\frac{2i\alpha }{(p+i\alpha )(1+\rho ^2)}\left(\begin{array}{cc}1\hfill & \hfill \rho \\ \rho \hfill & \hfill \rho ^2\end{array}\right),\hfill & \hfill (\alpha ^2=M\epsilon ).\end{array}`$
Now the equation for $`\stackrel{~}{B}`$ has the form
$$\{\begin{array}{ccc}\hfill \stackrel{~}{S(p)}\stackrel{~}{B}_+(p)=& \stackrel{~}{B}_{}(p),\hfill & \hfill \mathrm{}<p<\mathrm{},\\ \hfill \stackrel{~}{S}(p)=& R(p)S(p)R^1(p)\stackrel{~}{S}[\stackrel{~}{\delta },\stackrel{~}{\eta },\stackrel{~}{\epsilon }].\hfill & \end{array}$$
(B1)
The last equation defines the reduced phase shifts $`\stackrel{~}{\delta },\stackrel{~}{\epsilon },\stackrel{~}{\eta }`$ as functions of input experimental phase shifts $`\delta ,\epsilon ,\eta `$.
The solution of eq.(B1) was found in ref. in the form of series
$`\stackrel{~}{B}_\pm (p)=\stackrel{~}{B}_{\pm ,0}(p)[I+{\displaystyle \underset{m=1}{\overset{\mathrm{}}{}}}\stackrel{~}{B}_{\pm ,m}(p)],`$
where
$`\stackrel{~}{B}_{\pm ,0}(p)=\left(\begin{array}{cc}\phi _1(p)e^{\stackrel{~}{\delta }(p)}\hfill & \hfill 0\\ 0\hfill & \hfill \phi _2(p)e^{\stackrel{~}{\eta }(p)}\end{array}\right),`$
$`\phi _1(p)=\mathrm{exp}[{\displaystyle \frac{1}{\pi }}V.P.{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{\stackrel{~}{\delta }(p^{})dp^{}}{p^{}p}}],`$
$`\phi _2(p)=\mathrm{exp}[{\displaystyle \frac{1}{\pi }}V.P.{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{\stackrel{~}{\eta }(p^{})dp^{}}{p^{}p}}],`$
$$\stackrel{~}{B}_{\pm ,m}(p)=\frac{1}{2\pi i}_{\mathrm{}}^{\mathrm{}}\frac{dp^{}}{pp^{}\pm i0}\underset{n=1}{\overset{m}{}}G_n(p^{})\stackrel{~}{B}_{+,0}(p^{})[\stackrel{~}{B}_{+,mn}(p^{})]^{1\delta _{mn}}.$$
(B2)
In eq.(B2) for odd $`n`$
$`G_n(p)=i(1)^{\frac{n1}{2}}{\displaystyle \frac{1}{n!}}[2\stackrel{~}{\epsilon }(p)]^n\left(\begin{array}{cc}0\hfill & e^{i(\stackrel{~}{\delta }+\stackrel{~}{\eta })}\hfill \\ e^{i(\stackrel{~}{\delta }+\stackrel{~}{\eta })}\hfill & 0\hfill \end{array}\right)`$
and for even $`n`$
$`G_n(p)=i(1)^{\frac{n}{2}}{\displaystyle \frac{1}{n!}}[2\stackrel{~}{\epsilon }(p)]^n\left(\begin{array}{cc}e^{2i\stackrel{~}{\delta }}\hfill & 0\hfill \\ 0\hfill & e^{2i\stackrel{~}{\eta }}\hfill \end{array}\right),`$
$`\delta _{mn}`$ is the Kroneker delta.
## C $`g_c^{ll^{}}`$โmatrix.
In terms of invariant variables $`s,s^{},t`$ and the nucleon EM form factors the matrix elements have the form:
$`g_c^{00}(s,s^{},t)=`$
$`g(s,s^{},t)[g_1(s,s^{},t)(\mathrm{cos}\alpha _1\mathrm{cos}\alpha _2{\displaystyle \frac{1}{3}}\mathrm{sin}\alpha _1\mathrm{sin}\alpha _2)G_{EN}^s(Q^2)+`$
$`+{\displaystyle \frac{1}{2M}}g_2(s,s^{},t)({\displaystyle \frac{1}{3}}\mathrm{sin}\alpha _1\mathrm{cos}\alpha _2\mathrm{cos}\alpha _1\mathrm{sin}\alpha _2)G_{MN}^s(Q^2)],`$
$`g_c^{02}(s,s^{},t)=`$
$`g(s,s^{},t)\{g_1(s,s^{},t)(\sqrt{2}P_{20}\mathrm{cos}\alpha _1\mathrm{sin}\alpha _2+{\displaystyle \frac{1}{\sqrt{2}}}P_{21}\mathrm{sin}\alpha _1\mathrm{cos}\alpha _2)G_{EN}^s`$
$`{\displaystyle \frac{1}{2M}}g_2(s,s^{},t)(\sqrt{2}P_{20}\mathrm{sin}\alpha _1\mathrm{cos}\alpha _2+{\displaystyle \frac{1}{\sqrt{2}}}P_{21}\mathrm{cos}\alpha _1\mathrm{sin}\alpha _2)G_{MN}^s\},`$
$`g_c^{20}(s,s^{},t)=g_c^{02}(s^{},s,t),`$
$`g_c^{22}(s,s^{},t)=`$
$`g(s,s^{},t)\{g_1(s,s^{},t)[({\displaystyle \frac{1}{3}}P_{21}P_{21}^{}+{\displaystyle \frac{2}{3}}P_{20}P_{20}^{})\mathrm{cos}(\alpha _1\alpha _2)+`$
$`+({\displaystyle \frac{1}{12}}P_{22}P_{22}^{}+{\displaystyle \frac{1}{3}}P_{20}P_{20}^{})\mathrm{cos}\alpha _1\mathrm{cos}\alpha _2+`$
$`+\left({\displaystyle \frac{1}{12}}(P_{22}P_{21}^{}P_{21}P_{22}^{})+{\displaystyle \frac{1}{2}}(P_{21}P_{20}^{}P_{20}P_{21}^{})\right)\mathrm{sin}(\alpha _1\alpha _2)`$
$`{\displaystyle \frac{1}{6}}(P_{22}P_{20}^{}+P_{20}P_{22}^{})\mathrm{sin}\alpha _1\mathrm{sin}\alpha _2]G^s_{EN}{\displaystyle \frac{1}{2M}}g_2(s,s^{},t)`$
$`[{\displaystyle \frac{1}{12}}((P_{21}P_{22}^{}P_{22}P_{21}^{})+{\displaystyle \frac{1}{2}}(P_{20}P_{21}^{}P_{21}P_{20}^{}))`$
$`\mathrm{cos}(\alpha _1\alpha _2)({\displaystyle \frac{1}{12}}P_{22}P_{22}^{}+{\displaystyle \frac{1}{3}}P_{20}P_{20}^{})\mathrm{cos}\alpha _1\mathrm{sin}\alpha _2`$
$`{\displaystyle \frac{1}{6}}(P_{22}P_{20}^{}P_{20}P_{22}^{})\mathrm{sin}\alpha _1\mathrm{cos}\alpha _2+`$
$`+({\displaystyle \frac{1}{3}}P_{21}P_{21}^{}+{\displaystyle \frac{2}{3}}P_{20}P_{20}^{})\mathrm{sin}(\alpha _1\alpha _2)]G_{MN}^s\},`$
where
$`\begin{array}{cc}\hfill g(s,s^{},t)=& \frac{g_1(s,s^{},t)(t)}{\sqrt{(s4M^2)(s^{}4M^2)}}\frac{1}{[\lambda (s,s^{},t)]^{3/2}}\frac{1}{\sqrt{1+\tau }},\hfill \\ \hfill g_1(s,s^{},t)=& s+s^{}t,\hfill \\ \hfill g_2(s,s^{},t)=& [(1)(M^2\lambda (s,s^{},t)+ss^{}t\left)\right]^{1/2},\hfill \\ \hfill \lambda (s,s^{},t)=& s^2+s^2+t^22(ss^{}+st+s^{}t).\hfill \end{array}`$
$`P_{lm}`$ are the Legendre polynomials, $`P_{lm}P_{lm}(x)`$ and $`P_{lm}^{}P_{lm}(x^{})`$, where
$`\begin{array}{cc}\hfill x(s,s^{},t)=& \frac{\sqrt{s^{}(s^{}st)}}{\sqrt{(s^{}4M^2)\lambda (s,s^{},t)}},\hfill \\ \hfill x^{}(s,s^{},t)=& x(s^{},s,t).\hfill \end{array}`$
The angles $`\alpha _1,\alpha _2`$ of the relativistic rotation of nucleon spins in deuteron are
$`\begin{array}{cc}\alpha _1=\hfill & \mathrm{arctan}\frac{g_2(s,s^{},t)}{M\left[(\sqrt{s}+\sqrt{s^{}})^2t\right]+\sqrt{ss^{}}(\sqrt{s}+\sqrt{s^{}}+2M)},\hfill \\ \alpha _2=\hfill & \mathrm{arctan}\frac{g_2(s,s^{},t)(\sqrt{s}+\sqrt{s^{}}+2M)}{M(s+s^{}t)(\sqrt{s}+\sqrt{s^{}}+2M)+\sqrt{ss^{}}(4M^2t)}.\hfill \end{array}`$
$`\tau =Q^2/4M^2`$ ; $`G_{E,MN}^s=\frac{1}{2}(G_{E,Mp}+G_{E,Mn})`$ are the nucleon isoscalar charge and magnetic form factors. |
warning/0001/hep-ph0001005.html | ar5iv | text | # 1 Introduction
## 1 Introduction
One of the most promising candidates for the cold dark matter believed to pervade the Universe is the lightest supersymmetric particle (LSP) , commonly expected to be the lightest neutralino $`\chi `$, which is stable in the minimal supersymmetric extension of the Standard Model (MSSM) with conserved $`R`$ parity . The quantum stability of the gauge hierarchy suggests that sparticles weigh less than about 1 TeV , which is also the range favoured for a cold dark matter particle , and there are indeed generic domains of the MSSM parameter space in which the relic LSP density falls within the range $`0.1\mathrm{\Omega }_\chi h^20.3`$ favoured by astrophysics and cosmology . The unsuccessful laboratory searches for sparticles impose non-trivial constraints on the MSSM parameter space, suggesting that the LSP $`\chi `$ is mainly a $`U(1)`$ gaugino (Bino) .
Many non-accelerator strategies to search for cosmological relic neutralinos have been proposed , including indirect searches for products of their annihilations in free space or inside astrophysical bodies, and direct searches for their scattering on target nuclei in low-background underground laboratories . The rates for such experiments typically have larger uncertainties than those for producing sparticles at accelerators, since they involve some astrophysical and/or cosmological uncertainties as well as those due to simulations of the signatures, over and above the common uncertainties in the MSSM parameters. Nevertheless, such dark matter searches offer interesting prospects for beating accelerators to the discovery of supersymmetry, particularly during the coming years before the LHC enters operation.
In this paper, we embark on a programme to clarify the extents of the uncertainties in searches for supersymmetric dark matter, by re-evaluating the rates to be expected for the elastic scattering of relic LSPs on protons and neutrons . Large ranges for these rates are often quoted , reflecting general explorations of the MSSM parameter space. Our approach is to establish as accurately as possible a baseline set of predictions based on the most plausible assumptions commonly used in constrained MSSM phenomenology, such as universality in the soft supersymmetry-breaking parameters as suggested by minimal supergravity models, and requiring the cosmological relic density to lie within the range favoured by astrophysics and cosmology, namely $`0.1\mathrm{\Omega }_\chi h^20.3`$. These assumptions can and should be questioned, but they are well motivated and good candidates for default options in analyses of the MSSM and cold dark matter.
In the course of this re-evaluation of elastic $`\chi p,n`$ scattering cross sections, we re-analyze the relevant spin-independent and spin-dependent matrix elements of scalar densities and axial currents in protons and neutrons. We update previous analyses using further information from chiral symmetry , low-energy $`\pi p,n`$ scattering and deep-inelastic lepton-nucleon scattering . We include a discussion of uncertainties in the values of the scalar and axial-current matrix elements.
We perform a systematic scan of the region of the $`m_0,m_{1/2}`$ parameter space of the MSSM with supergravity-inspired universality that is consistent with accelerator constraints and yields a cosmological relic density within the favoured range $`0.1\mathrm{\Omega }_\chi 0.3`$ . We treat $`\mu `$ as a dependent parameter (modulo a sign ambiguity), and our results are not very sensitive to $`A`$. We order our results in terms of $`m_\chi `$ which closely tracks $`m_{1/2}`$. For any given choice of $`m_\chi ,\mathrm{tan}\beta `$ and the sign of $`\mu `$, we find a relatively narrow band of possible cross sections, reflecting the fact that the accelerator and cosmological constraints favour a predominant $`U(1)`$ gaugino (Bino) composition for the LSP. Our results fall considerably below many of the possible predictions in the literature , and may discourage some faint-hearted experimentalists. However, we think they provide a realistic estimate of the target sensitivity required for an experiment to have a good chance of success.
## 2 Theoretical Framework
We review in this Section the theoretical framework we use in the context of the MSSM . The neutralino LSP is the lowest-mass eigenstate combination of the Bino $`\stackrel{~}{B}`$, Wino $`\stackrel{~}{W}`$ and Higgsinos $`\stackrel{~}{H}_{1,2}`$, whose mass matrix $`N`$ is diagonalized by a matrix $`Z`$: $`diag(m_{\chi _1,..,4})=Z^{}NZ^1`$. The composition of the lightest neutralino may be written as
$$\chi =Z_{\chi 1}\stackrel{~}{B}+Z_{\chi 2}\stackrel{~}{W}+Z_{\chi 3}\stackrel{~}{H_1}+Z_{\chi 4}\stackrel{~}{H_2}$$
(1)
As already mentioned, we assume universality at the supersymmetric GUT scale for the $`U(1)`$ and $`SU(2)`$ gaugino masses: $`M_{1,2}=m_{1/2}`$, so that $`M_1=\frac{5}{3}\mathrm{tan}^2\theta _WM_2`$ at the electroweak scale. We denote by $`\mathrm{tan}\beta `$ the ratio of Higgs vacuum expectation values, and $`\mu `$ is the Higgsino mass-mixing parameter. We also assume GUT-scale universality for the soft supersymmetry-breaking scalar masses $`m_0`$, for the Higgs bosons as well as the squarks and sleptons. We further assume GUT-scale universality for the soft supersymmetry-breaking trilinear terms $`A`$. Our treatment of the sfermion mass matrices $`M`$ follows . As discussed there, the sfermion mass-squared matrix is diagonalized by a matrix $`\eta `$: $`diag(m_1^2,m_2^2)\eta M^2\eta ^1`$, which can be parameterized for each flavour $`f`$ by an angle $`\theta _f`$ and phase $`\gamma _f`$:
$$\left(\begin{array}{cc}\mathrm{cos}\theta _f& \mathrm{sin}\theta _fe^{i\gamma _f}\\ \mathrm{sin}\theta _fe^{i\gamma _f}& \mathrm{cos}\theta _f\end{array}\right)\left(\begin{array}{cc}\eta _{11}& \eta _{12}\\ \eta _{21}& \eta _{22}\end{array}\right)$$
As a simplification, we neglect CP violation in this paper, so that $`\gamma _f=0`$ and there are no CP-violating phases in the neutralino mass matrix, either. We treat $`m_{1/2},m_0,A`$ and $`\mathrm{tan}\beta `$ as free parameters, and $`\mu `$ and the pseudoscalar Higgs mass $`m_A`$ as dependent parameters specified by the electroweak vacuum conditions, which we calculate using $`m_t=175`$ GeV <sup>1</sup><sup>1</sup>1We have checked that varying $`m_t`$ by $`\pm 5`$ GeV has a negligible effect on our results..
The MSSM Lagrangian leads to the following low-energy effective four-fermi Lagrangian suitable for describing elastic $`\chi `$-nucleon scattering :
$$=\overline{\chi }\gamma ^\mu \gamma ^5\chi \overline{q_i}\gamma _\mu (\alpha _{1i}+\alpha _{2i}\gamma ^5)q_i+\alpha _{3i}\overline{\chi }\chi \overline{q_i}q_i+\alpha _{4i}\overline{\chi }\gamma ^5\chi \overline{q_i}\gamma ^5q_i+\alpha _{5i}\overline{\chi }\chi \overline{q_i}\gamma ^5q_i+\alpha _{6i}\overline{\chi }\gamma ^5\chi \overline{q_i}q_i$$
(2)
This Lagrangian is to be summed over the quark generations, and the subscript $`i`$ labels up-type quarks ($`i=1`$) and down-type quarks ($`i=2`$). The terms with coefficients $`\alpha _{1i},\alpha _{4i},\alpha _{5i}`$ and $`\alpha _{6i}`$ make contributions to the elastic scattering cross section that are velocity-dependent, and may be neglected for our purposes. In fact, if the CP violating phases are absent as assumed here, $`\alpha _5=\alpha _6=0`$ . The coefficients relevant for our discussion are:
$`\alpha _{2i}`$ $`=`$ $`{\displaystyle \frac{1}{4(m_{1i}^2m_\chi ^2)}}\left[\left|Y_i\right|^2+\left|X_i\right|^2\right]+{\displaystyle \frac{1}{4(m_{2i}^2m_\chi ^2)}}\left[\left|V_i\right|^2+\left|W_i\right|^2\right]`$ (3)
$`{\displaystyle \frac{g^2}{4m_Z^2\mathrm{cos}^2\theta _W}}\left[\left|Z_{\chi _3}\right|^2\left|Z_{\chi _4}\right|^2\right]{\displaystyle \frac{T_{3i}}{2}}`$
and
$`\alpha _{3i}`$ $`=`$ $`{\displaystyle \frac{1}{2(m_{1i}^2m_\chi ^2)}}Re\left[\left(X_i\right)\left(Y_i\right)^{}\right]{\displaystyle \frac{1}{2(m_{2i}^2m_\chi ^2)}}Re\left[\left(W_i\right)\left(V_i\right)^{}\right]`$ (4)
$`{\displaystyle \frac{gm_{qi}}{4m_WB_i}}[Re\left(\delta _{1i}[gZ_{\chi 2}g^{}Z_{\chi 1}]\right)D_iC_i({\displaystyle \frac{1}{m_{H_1}^2}}+{\displaystyle \frac{1}{m_{H_2}^2}})`$
$`+Re\left(\delta _{2i}[gZ_{\chi 2}g^{}Z_{\chi 1}]\right)({\displaystyle \frac{D_i^2}{m_{H_2}^2}}+{\displaystyle \frac{C_i^2}{m_{H_1}^2}})]`$
where
$`X_i`$ $``$ $`\eta _{11}^{}{\displaystyle \frac{gm_{q_i}Z_{\chi 5i}^{}}{2m_WB_i}}\eta _{12}^{}e_ig^{}Z_{\chi 1}^{}`$
$`Y_i`$ $``$ $`\eta _{11}^{}\left({\displaystyle \frac{y_i}{2}}g^{}Z_{\chi 1}+gT_{3i}Z_{\chi 2}\right)+\eta _{12}^{}{\displaystyle \frac{gm_{q_i}Z_{\chi 5i}}{2m_WB_i}}`$
$`W_i`$ $``$ $`\eta _{21}^{}{\displaystyle \frac{gm_{q_i}Z_{\chi 5i}^{}}{2m_WB_i}}\eta _{22}^{}e_ig^{}Z_{\chi 1}^{}`$
$`V_i`$ $``$ $`\eta _{22}^{}{\displaystyle \frac{gm_{q_i}Z_{\chi 5i}}{2m_WB_i}}+\eta _{21}^{}\left({\displaystyle \frac{y_i}{2}}g^{}Z_{\chi 1}+gT_{3i}Z_{\chi 2}\right)`$ (5)
where $`y_i,T_{3i}`$ denote hypercharge and isospin, and
$`\delta _{1i}=Z_{\chi 3}(Z_{\chi 4})`$ , $`\delta _{2i}=Z_{\chi 4}(Z_{\chi 3}),`$
$`B_i=\mathrm{sin}\beta (\mathrm{cos}\beta )`$ , $`A_i=\mathrm{cos}\beta (\mathrm{sin}\beta ),`$
$`C_i=\mathrm{sin}\alpha (\mathrm{cos}\alpha )`$ , $`D_i=\mathrm{cos}\alpha (\mathrm{sin}\alpha )`$ (6)
for up (down) type quarks. We denote by $`m_{H_2}<m_{H_1}`$ the two scalar Higgs masses, and $`\alpha `$ denotes the Higgs mixing angle <sup>2</sup><sup>2</sup>2We note that (4) is taken from and corrects an error in , and that (3, 4) agree with and the published version of ..
## 3 Hadronic Matrix Elements
The elastic cross section for scattering off a nucleus can be decomposed into a scalar (spin-independent) part obtained from the $`\alpha _{2i}`$ term in (2), and a spin-dependent part obtained from the $`\alpha _{3i}`$ term. Each of these can be written in terms of the cross sections for elastic scattering for scattering off individual nucleons, as we now review and re-evaluate.
The scalar part of the cross section can be written as
$$\sigma _3=\frac{4m_r^2}{\pi }\left[Zf_p+(AZ)f_n\right]^2$$
(7)
where $`m_r`$ is the reduced LSP mass,
$$\frac{f_p}{m_p}=\underset{q=u,d,s}{}f_{Tq}^{(p)}\frac{\alpha _{3q}}{m_q}+\frac{2}{27}f_{TG}^{(p)}\underset{c,b,t}{}\frac{\alpha _{3q}}{m_q}$$
(8)
and $`f_n`$ has a similar expression. The parameters $`f_{Tq}^{(p)}`$ are defined by
$$m_pf_{Tq}^{(p)}p|m_q\overline{q}q|pm_qB_q$$
(9)
whilst $`f_{TG}^{(p)}=1_{q=u,d,s}f_{Tq}^{(p)}`$ . We observe that only the products $`m_qB_q`$, the ratios of the quark masses $`m_q`$ and the ratios of the scalar matrix elements $`B_q`$ are invariant under renormalization and hence physical quantities.
We take the ratios of the quark masses from :
$$\frac{m_u}{m_d}=0.553\pm 0.043,\frac{m_s}{m_d}=18.9\pm 0.8$$
(10)
In order to determine the ratios of the $`B_q`$ and the products $`m_qB_q`$ we use information from chiral symmetry applied to baryons. Following , we have:
$$z\frac{B_uB_s}{B_dB_s}=\frac{m_{\mathrm{\Xi }^0}+m_\mathrm{\Xi }^{}m_pm_n}{m_{\mathrm{\Sigma }^+}+m_\mathrm{\Sigma }^{}m_pm_n}$$
(11)
Substituting the experimental values of these baryon masses, we find
$$z=1.49$$
(12)
with an experimental error that is negligible compared with others discussed below. Defining
$$y\frac{2B_s}{B_d+B_u},$$
(13)
we then have
$$\frac{B_d}{B_u}=\frac{2+((z1)\times y)}{2\times z((z1)\times y)}$$
(14)
The experimental value of the $`\pi `$-nucleon $`\sigma `$ term is :
$$\sigma \frac{1}{2}(m_u+m_d)\times (B_d+B_u)=45\pm 8\mathrm{MeV}$$
(15)
and octet baryon mass differences may be used to estimate that
$$\sigma =\frac{\sigma _0}{(1y)}:\sigma _0=36\pm 7\mathrm{MeV}$$
(16)
Comparing (15) and (16), we find a central value of $`y=0.2`$, to which we assign an error $`\pm 0.1`$, which yields
$$\frac{B_d}{B_u}=0.73\pm 0.02$$
(17)
The formal error in $`y`$ derived from (15) and (16) is actually $`\pm `$0.2, which would double the error in $`B_d/B_u`$. We have chosen the smaller uncertainty because we consider a value of y in excess of 30% rather unlikely. However, we do illustrate later by one example the potential consequences of a larger error in $`y`$.
The numerical magnitudes of the individual renormalization-invariant products $`m_qB_q`$ and hence the $`f_{Tq}^{(p)}`$ may now be determined:
$$f_{Tu}^{(p)}=0.020\pm 0.004,f_{Td}^{(p)}=0.026\pm 0.005,f_{Ts}^{(p)}=0.118\pm 0.062$$
(18)
where essentially all the error in $`f_{Ts}^{(p)}`$ arises from the uncertainty in $`y`$. The corresponding values for the neutron are
$$f_{Tu}^{(n)}=0.014\pm 0.003,f_{Td}^{(n)}=0.036\pm 0.008,f_{Ts}^{(n)}=0.118\pm 0.062.$$
(19)
It is clear already that the difference between the scalar parts of the cross sections for scattering off protons and neutrons must be rather small.
The spin-dependent part of the elastic $`\chi `$-nucleus cross section can be written as
$$\sigma _2=\frac{32}{\pi }G_F^2m_r^2\mathrm{\Lambda }^2J(J+1)$$
(20)
where $`m_r`$ is again the reduced neutralino mass, $`J`$ is the spin of the nucleus, and
$$\mathrm{\Lambda }\frac{1}{J}(a_pS_p+a_nS_n)$$
(21)
where
$$a_p=\underset{i}{}\frac{\alpha _{2i}}{\sqrt{2}G_f}\mathrm{\Delta }_i^{(p)},a_n=\underset{i}{}\frac{\alpha _{2i}}{\sqrt{2}G_f}\mathrm{\Delta }_i^{(n)}$$
(22)
The factors $`\mathrm{\Delta }_i^{(p,n)}`$ parametrize the quark spin content of the nucleon. A recent global analysis of QCD sum rules for the $`g_1`$ structure functions , including $`๐ช(\alpha _s^3)`$ corrections, corresponds formally to the values
$$\mathrm{\Delta }_u^{(p)}=0.78\pm 0.02,\mathrm{\Delta }_d^{(p)}=0.48\pm 0.02,\mathrm{\Delta }_s^{(p)}=0.15\pm 0.02$$
(23)
whilst perturbative QCD fits to the data for $`g_1`$ tend to give broader ranges . In our numerical analysis, we double the formal errors in (23) to $`\pm 0.04`$, essentially 100% correlated for the three quark flavours. In the case of the neutron, we have $`\mathrm{\Delta }_u^{(n)}=\mathrm{\Delta }_d^{(p)},\mathrm{\Delta }_d^{(n)}=\mathrm{\Delta }_u^{(p)}`$, and $`\mathrm{\Delta }_s^{(n)}=\mathrm{\Delta }_s^{(p)}`$.
## 4 Cosmological and Experimental Constraints
The domain of MSSM parameter space that we explore in this paper is that defined in . Several convergent measures of cosmological parameters suggest that the cold dark matter density $`\mathrm{\Omega }_{CDM}=0.3\pm 0.1`$ and that the Hubble expansion rate $`Hh\times 100`$ km/s/Mpc: $`h=0.7\pm 0.1`$, leading to the preferred range $`0.1\mathrm{\Omega }_{CDM}h^20.3`$. The upper limit on $`\mathrm{\Omega }_{CDM}`$ can be translated directly into the corresponding upper limit on $`\mathrm{\Omega }_\chi `$. However, it is possible that there is more than one component in the cold dark matter, so that $`\mathrm{\Omega }_\chi <\mathrm{\Omega }_{CDM}`$, opening up the possibility that $`\mathrm{\Omega }_\chi <0.1`$. Although the MSSM parameters which lead to $`\mathrm{\Omega }_\chi <0.1`$ tend to give larger elastic scattering cross sections, the detection rate also must be reduced because of the corresponding reduction in the density of LSPs in the Galactic halo. Here we shall neglect this possibility, assuming instead that essentially all the cold dark matter is composed of LSPs, so that $`\mathrm{\Omega }_\chi 0.1`$.
For the calculation of the relic LSP density, we follow , where coannihilations between $`\chi `$ and the sleptons $`\stackrel{~}{\mathrm{}}`$, particularly the lighter stau $`\stackrel{~}{\tau }_1`$, were shown to play an important role. As we discuss in more detail later, $`m_\chi `$ depends essentially on $`m_{1/2}`$, and coannihilation increases by a factor $`2`$ the cosmological upper limit on $`m_{1/2}`$ to $`1400`$ GeV, allowing $`m_\chi <600`$ GeV. At this upper limit on $`m_{1/2}`$, there is a unique allowed value of $`m_0350`$ GeV, but for lower values of $`m_{1/2}`$ the width of the allowed range of $`m_0`$ expands, reaching $`50<m_0<150`$ GeV when $`m_{1/2}200`$ GeV. At this value of $`m_{1/2}`$ and scanning across the cosmological range in $`m_0`$, we find that $`m_\chi 80`$ GeV, with a small variation by $`0.4`$ GeV. These numbers are not very sensitive to $`\mathrm{tan}\beta `$ in the range from 3 to 10 studied in and here, nor are they very sensitive to the chosen value of $`A`$.
The lower limit on $`m_{1/2}`$ and hence $`m_\chi `$ depends on the sparticle search limits provided by LEP . The most essential of these for our current purposes are those provided by the experimental lower limits on the lighter chargino mass $`m_{\chi ^\pm }`$ and the lighter scalar Higgs mass $`m_{H_2}`$. A lower limit $`m_{\chi ^\pm }95`$ GeV was assumed in : unsuccessful chargino searches during higher-energy runs of LEP have now increased this lower limit to $`m_{\chi ^\pm }100`$ GeV , which does not reduce very much the range allowed in .
The impact of the recently-improved lower limits on the Higgs mass is potentially more significant, particularly for $`\mathrm{tan}\beta =3`$, as displayed in Figs. 6 and 7 of . The present experimental lower limit for $`\mathrm{tan}\beta =3`$ is probably $`m_{H_2}>105`$ to 109 GeV . The $`m_{H_2}`$ contours shown in Figs. 6a,b and 7b of were not calculated with the most recent two-loop MSSM code , so we take the $`m_{H_2}=100`$ GeV lines in as indicative constraints. These correspond to $`m_{1/2}340(720)`$ GeV for $`\mu >(<)0`$, corresponding in turn to $`m_\chi >140(310)`$ GeV. On the other hand, for $`\mathrm{tan}\beta =10`$, the LEP lower limit on $`m_{H_2}`$ is considerably weaker than $`100`$ GeV , and hence does not constrain significantly the allowed parameter space, as seen in Figs. 6c,d and 7c of .
We note in passing that requiring our present electroweak vacuum to be stable against transitions to a lower-energy state in which electromagnetic charge and colour are broken (CCB) would divide the parameter regions allowed in into two parts: one at large $`m_{1/2}`$ and the other at small $`m_{1/2}`$ and relatively large $`m_0`$. We do not implement the CCB constraint in our analysis, since it may be considered optional. Nor do we implement any constraint due to the observed rate of $`bs\gamma `$ decay , but it is well known that this reduces very substantially the parameter space allowed for $`\mu <0`$.
## 5 Results
As discussed above, we scan the cosmologically preferred set of parameters which yield $`0.1\mathrm{\Omega }_\chi h^20.3`$ and are consistent with the recent LEP accelerator bounds. For each value of $`\mathrm{tan}\beta `$ and sign of $`\mu `$, we vary $`m_{1/2}`$ and $`m_0`$ over all the allowed range. As default, we choose $`A_0=m_{1/2}`$ in most of our computations. Then, using the hadronic inputs described in section 3, we compute separately the spin-dependent and scalar contributions from the $`\alpha _2`$ and $`\alpha _3`$ coefficients, respectively, to the elastic scattering of LSPs on both protons and neutrons.
In Figure 1, we show the resulting spin-dependent elastic cross section as a function of the LSP mass, $`m_\chi `$. Although it is barely discernible, the thicknesses of the central curves in the panels show the ranges in the cross section for fixed $`m_\chi `$ that are induced by varying $`m_0`$. At large $`m_\chi `$ where coannihilations are important, the range in the allowed values of $`m_0`$ is small and particularly little variation in the cross section is expected. The shaded regions in this and the following figures show the effects of the uncertainties in the input values of the $`\mathrm{\Delta }_i^{(p)}`$ (23). In Figure 1a, for $`\mathrm{tan}\beta =3,\mu <0`$, we see at small $`m_\chi `$ the effect of a cancellation induced by the difference in signs between $`\mathrm{\Delta }_u`$ and $`\mathrm{\Delta }_{d,s}`$. Cancellations are possible for the other values of $`\mathrm{tan}\beta `$ and sign of $`\mu `$, but not in the preferred range of $`m_{1/2}`$ and $`m_0`$ used here. Aside from the cancellation, the spin-dependent cross section peaks at about $`10^4`$ pb and drops rapidly as $`m_\chi `$ increases.
In Figure 2, we show the corresponding result for the scalar cross section, based on $`\alpha _3`$. As in Figure 1, the thickness of the central curve reflects the range in $`m_0`$ sampled. The shaded region now corresponds to the uncertainties in the inputs given in (18). The scalar cross section is, in general, more sensitive to the sign of $`\mu `$ than is the spin-dependent cross section. Notice that, in Figure 2c for $`\mathrm{tan}\beta =10`$ and $`\mu <0`$, there is another cancellation. In this case, Higgs exchange is dominant in $`\alpha _3`$. We first note that, for $`\mu <0`$, both $`Z_{\chi 3}`$ and $`Z_{\chi 4}`$ are negative, as is the Higgs mixing angle $`\alpha `$. Inserting the definitions of $`\delta _{1i(2i)}`$, we see that there is a potential cancellation of the Higgs contribution to $`\alpha _3`$ for both up-type and down-type quarks. Whilst there is such a cancellation for the down-type terms, which change from positive to negative as one increases $`m_\chi `$, such a cancellation does not occur for the up-type terms, which remain negative in the region of parameters we consider. The cancellation that is apparent in the figure is due to the cancellation in $`\alpha _3`$ between the up-type contribution (which is negative) and the down-type contribution, which is initially positive but decreasing, eventually becoming negative as we increase $`m_\chi `$.
In Figure 3, we show the effects of varying some of the input assumptions made earlier. For example, when the assumed uncertainty in $`y`$ is taken to be $`\pm `$0.2, we get a thicker shaded region, as shown for $`\mathrm{tan}\beta =3,\mu <0`$ in Figure 3a. In Figure 3b, we give one example of the cross section for the elastic scattering of neutralinos on neutrons. This particular case was chosen because it displays the largest difference between the neutron and proton cross sections among those tested. As one can see, our results for the LSP scattering on neutrons and protons are almost identical. Similarly, the effects of changing $`A_0`$ are also relatively small, as illustrated by two cases with $`A_0=0`$ in Figures 3c and 3d. In the latter example, there is almost a factor of 2 difference at higher values of $`m_\chi `$, which is due to yet another cancellation, this time between the squark-exchange and $`Z`$-exchange terms in $`\alpha _{2u}`$.
Finally, we show in Figure 4 compilations of our results for the spin-dependent and -independent cross sections, compared with current and projected experimental limits obtained from . The shaded region in panel (a) is the union of the shaded regions in Figure 1, and the shaded region in panel (b) is the union of the shaded regions in Figure 2.
## 6 Discussion
As seen in Figure 4, the present experimental upper limit on the spin-independent part of the elastic scattering of the LSP on a nucleon is around $`10^5`$ pb for $`50\mathrm{GeV}<m_\chi <100`$ GeV. On the other hand, the maximum scalar cross section we find is around $`10^8`$ pb, which is attained for $`m_\chi 50`$ GeV. This means that present experiments searching directly for supersymmetric dark matter are far from constraining the parameter space of our baseline theoretical framework, in which LEP constraints are applied to MSSM models with universal supergravity-inspired soft supersymmetry-breaking parameters $`m_{1/2},m_0`$.
The literature contains predictions for the elastic LSP-nucleus scattering rates that vary considerably, with some estimates lying considerably higher than ours . There are various ways in which such differences might arise, of which we mention a few here. We have imposed the requirement that the LSP relic density lie in the favoured range $`0.1\mathrm{\Omega }_\chi h^20.3`$, whereas other calculations often include models with lower relic densities. Such models would normally have larger $`\chi \chi `$ annihilation cross sections, and correspondingly larger elastic scattering cross sections. Hence the predicted scattering rates would be larger, if the conventional halo density $`\rho 0.3`$ GeV/cm<sup>3</sup> is assumed for the LSP . However, we believe this assumption is unreasonable: if not all the total cold dark matter density $`\mathrm{\Omega }_{CDM}`$ is composed of LSPs, the density of LSPs in the halo should be reduced by the corresponding factor $`\mathrm{\Omega }_\chi /\mathrm{\Omega }_{CDM}`$.
Other possible differences may arise in the treatment of the LEP constraints: we find it to be almost excluded that the LSP be Higgsino-like , even if the assumptions of universal soft supersymmetry breaking are relaxed, and Higgsino dark matter is certainly excluded if universality is assumed, as is the case here. In addition to the LEP constraints, this is because the value of $`\mu `$ is predicted as a function of $`m_{1/2}`$ and $`m_0`$, placing the LSP firmly in the Bino-like region. The same considerations exclude an LSP with mixed Higgsino/gaugino content.
The prediction of $`\mu `$ may be circumvented by postulating non-universality for the soft scalar supersymmetry-breaking parameters in the Higgs sector, which might have appeared to resurrect the possibility of a Higgsino-like LSP . However, such a possibility goes beyond the universality framework adopted here, and, moreover, the LEP constraints now appear to exclude this possibility , as mentioned above.
There are no differences between the effective Lagrangians we and others use to describe the four-fermion $`\chi q`$ interaction that determines the elastic $`\chi p,n`$ scattering cross sections. We have found differences of detail between our and other treatments of the hadronic matrix elements of the scalar and axial-current $`\overline{q}q`$ operators appearing in this Lagrangian , but this is not responsible for any big difference between the results.
We should not want our experimental colleagues to be too downcast by the long road they appear to have to cover in order to probe the minimal universal MSSM framework utilized here. For example, there are surely some supersymmetric models that predict larger scattering rates. However, we think it best to have in mind a plausible and realistic target sensitivity, which is what our universal framework and implementation of the LEP and cosmological constraints provide. Our results also have the merit of being relatively specific: as seen in Figure 4, the elastic scattering cross sections we predict for any given value of the LSP mass $`m_\chi `$ lie in a comparatively narrow band. As discussed earlier, this is essentially because the LSP is always mainly Bino-like in our framework, so its couplings do not depend greatly on other MSSM parameters such as $`m_0`$. The principal causes of broadening are the uncertainties in the hadronic inputs and the possibilities of cancellations that may reduce the cross sections for some specific values of the constrained MSSM parameters.
This tight correlation we find between the LSP mass and its elastic scattering rate means that future experiments should be able to phrase their sensitivities directly in terms of the LSP mass in the universal supergravity-inspired version of the MSSM. For example, our results suggest that the proposed Genius experiment would be sensitive to $`m_\chi <100`$ GeV for almost all MSSM parameter choices in Figure 5b. More optimistically, if/when a signal is observed, its plausibility would be enhanced if its recoil spectrum was correlated with the rate in the manner suggested by Figure 5. Thus our analysis provides experiments with an additional tool that may assist in the extraction of a signal that might be significantly smaller than they could have hoped. In any case, the importance of the search for supersymmetric matter remains unchanged, and there are still several years before the LHC comes into operation, so these experiments still have both motivation and opportunity.
Acknowledgments
We thank Toby Falk and Gerardo Ganis for many related discussions. The work of K.A.O. was supported in part by DOE grant DEโFG02โ94ERโ40823. |
warning/0001/quant-ph0001087.html | ar5iv | text | # Quantum secret sharing for general access structures
## 1 Introduction
A classical secret sharing scheme is a (usually) randomized encoding of a secret $`s`$ into a $`n`$-tuple, the coordinates of which are each given to different players in the player set $`P`$. The encoding is a secret sharing scheme if there exists a collection $`๐`$ of subsets of $`P`$ (called the *adversary structure*) such that no set of players in $`๐`$ gets any information about $`s`$ from their shares, but any set of players not in $`๐`$ will be able to compute $`s`$. The classic example of this is due to Shamir . He gives a construction based on polynomials over a finite field of a *threshold* secret-sharing scheme for any threshold $`t`$ and any number of players (in such a scheme, $`๐=\{BP:|B|t\}`$).
The idea of sharing *quantum* secrets was first described and solved for the case $`t=1,n=2`$ by Hillery *et al.* in <sup>1</sup><sup>1</sup>1In fact, shows how efficency can be gained in the insecure channels model by combining the key distribution and secret-sharing layers of the protocol. An even more efficient protocol was suggested in .. A more general solution, for all $`t>\frac{n}{2}1`$, was recently given by Cleve et al. (CGL, ). Their scheme is a direct generalization of the well-known Shamir scheme , with all calculations done unitarily and โat the quantum levelโ, i. e. replacing random choices with equal superpositions over those choices.
In next section we give definitions and background. In section 3, we then prove that classical *linear* secret-sharing schemes, with an appropriate adversary structure, can be converted into quantum schemes with the same complexity, both in terms of share size and encoding/reconstruction. This gives another proof of theorem 8 from . In the last section, we give a necessary and sufficient condition for (not necessarily linear) classical ss schemes to become quantum ones when run at the quantum level, and observe that all group homomorphic schemes obey this condition.
## 2 Preliminaries
### 2.1 Adversary structures
Given a set of players $`P`$, an adversary structure $`๐`$ over $`P`$ is a set of subsets of players which is downward-closed under inclusion:
$$(B๐\text{ and }B^{}B)B^{}๐.$$
Normally such a structure is used to represent the collection of all coalitions of players which a given protocol can tolerate without losing security: as long as the set of cheating players is in $`๐`$, the cheaters cannot breach the security of the protocol.
Secret-sharing schemes usually tolerate *threshold structures*, which are of the form $`๐=\{BP:|B|t\}`$ for some $`t`$. However, when working with more general structures, the following definitions prove useful.
###### Definition 1
An adversary structure $`๐2^P`$ is $`๐ฌ^2`$ if no two sets in $`๐`$ cover $`P`$, that is
$$B_1,B_2๐:B_1B_2=P.$$
###### Definition 2
The *dual* of an adversary structure $`๐`$ over $`P`$ is the collection
$$๐^{}=\{BP:B^c๐\}$$
where $`B^c`$ denotes the complement $`PB`$.
###### Definition 3
A structure $`๐`$ over $`P`$ is $`๐ฌ^2`$ if its dual $`๐^{}`$ is $`๐ฌ^2`$. This means that any two sets not in $`๐`$ will have a non-empty intersection.
It is interesting to note that $`๐`$ is $`๐ฌ^2`$ iff $`๐๐^{}`$. Dually, $`๐`$ is $`๐ฌ^2`$ iff $`๐๐^{}`$. Consequently, a collection is *self-dual* iff it is both $`๐ฌ^2`$ and $`๐ฌ^2`$.
#### 2.1.1 Monotone functions
We can define a partial order on $`\{0,1\}^n`$ by the rule โ$`๐ฑ๐ฒ`$ iff each coordinate of $`๐ฑ`$ is smaller than the corresponding coordinate of $`๐ฒ`$.โ
By identifying $`\{0,1\}^n`$ with $`2^{\{1,\mathrm{},n\}}`$, the relation $``$ on $`\{0,1\}^n`$ corresponds to inclusion ($``$) in $`2^{\{1,\mathrm{},n\}}`$. Then a monotone function $`f`$ corresponds to a function from $`2^{\{1,\mathrm{},n\}}`$ to $`\{0,1\}`$ such that $`ABf(A)f(B)`$.
Such a monotone function $`f`$ naturally defines an adversary structure $`๐_f=f^1(\{0\})=\{BP:f(B)=0\}`$. Moreover, $`f`$ is called $`๐ฌ^2`$ (resp. $`๐ฌ^2`$) iff $`๐_f`$ is $`๐ฌ^2`$ (or $`๐ฌ^2`$).
### 2.2 Monotone span programs
Span programs were introduced as a model of computation in . They were first used for multiparty protocols in under this name, although a similar construction, attributed to Brickell, already existed (). In this section we define some concepts related to monotone span programs.
###### Definition 4
A monotone span program (MSP) over a set $`P`$ is a triple $`(K,M,\psi )`$ where $`K`$ is a finite field, $`M`$ is a $`d\times e`$ matrix over $`K`$ and $`\psi :\{1,\mathrm{},d\}P`$ is a function which effectively labels each row of $`M`$ by a member of $`P`$.
The MSP associates to each subset $`BP`$ a subset of the rows of $`M`$: the set of rows $`l`$ such that $`\psi (l)B`$. This corresponds to a linear subspace of $`K^e`$ (the span of those rows). The monotone function $`f:2^P\{0,1\}`$ defined by a MSP is given by the rule โ$`f(B)=1`$ if and only if the target vector $`ฯต=(1,0,0,\mathrm{},0)`$ is in the subspace associated with $`B`$โ. If we denote by $`M_B`$ the submatrix of $`M`$ formed of the rows $`l`$ such that $`\psi (l)B`$ then we get that
$$f(B)=1ฯตIm(M_B^T).$$
In fact, given any monotone function $`f`$, we can construct a MSP which computes it. The size of the MSP will be at most proportional to the size of the smallest monotone threshold formula for $`f`$, but may in some cases be exponentially smaller .
The proof uses the following fact from linear algebra. Here the dual of a vector subspace $`W`$ is denoted $`W^{}=\{๐ฎ:๐ฎ^{}๐ฐ=0๐ฐW\}`$.
Denote the dual of a vector subspace $`W`$ by $`W^{}=\{๐ฎ:๐ฎ^{}๐ฐ=0๐ฐW\}`$. For any matrix $`M`$ we have $`Im(M^{})=ker(M)^{}`$. Thus, $`f(B)=0`$ iff $`๐ฏ:M_B๐ฏ=\mathrm{๐}`$ and $`ฯต^{}๐ฏ0`$.
#### 2.2.1 Secret-sharing from MSPโs
Given a MSP $`(K,M,\psi )`$, we can define a classical secret sharing scheme which tolerates the adversary structure $`๐_f`$ induced by the MSP. Say the dealer has a secret $`sK`$. He extends it to an $`e`$-rowed vector by adding random field elements $`a_2,\mathrm{},a_e`$ to make a vector $`๐ฌ_{}=(s,a_2,\mathrm{},a_e)`$. The dealer gives the $`l`$th component of $`\widehat{๐ฌ}=M๐ฌ_{}`$ to player $`P_{\psi (l)}`$. If $`\widehat{๐ฌ}_A`$ denotes the elements of $`\widehat{๐ฌ}`$ with indices in $`A`$ where $`A\{1,\mathrm{},d\}`$, then each $`P_i`$ receives $`\widehat{๐ฌ}_{\psi ^1(i)}`$.
The ss scheme thus defined tolerates exactly the adversary structure $`๐_f`$.
Note that the concept of MSPโs is very general: any linear secret-sharing scheme (i.e. one in which the encoding of the secret is given by a linear map over a field) can be formulated as a MSP-based scheme . The Shamir scheme is a special case, where $`M`$ is a $`n\times (k+1)`$ Vandermonde matrix, $`e=k+1`$, $`d=n`$, and $`\psi `$ is the identity on $`\{1,\mathrm{},n\}`$.
### 2.3 Secret sharing with general access structures
With classical data, secret sharing is possible for any access structure. Given a monotone threshold formula for a function $`f`$, Benaloh and Leichter gave a construction for $`๐_f`$ with efficency proportional to the size of the formula. This is improved on by constructions based on monotone span programs (section 2.2.1), which are always at least as efficient as the Benaloh-Leichter scheme but can be super-polynomially more so.
When sharing quantum data, the situation is slightly different. Because of the no-cloning theorem, it is impossible to share secrets with an adversary structure which is not $`๐ฌ^2`$ (since then one can find two disjoint sets which can reconstruct the secret based on their shares). Because a pure-state qss scheme is also a quantum code correcting erasures on the sets described by its adversary structure, we also get that any pure-state qss scheme has an adversary structure which is in fact self-dual . The natural converse to this is
###### Theorem 1
Given any $`๐ฌ^2`$ structure $`๐`$, we can find a qss scheme for $`๐`$. If $`๐`$ is self-dual, then the scheme can be a pure-state one.
This was proved for the case of threshold structures in : their construction works when the number of cheaters $`t`$ is more than $`\frac{n}{2}1`$ (i. e. it takes more than $`\frac{n}{2}`$ players to reconstruct the secret). Moreover, theirs is a pure-state scheme when $`n=2t+1`$ (these correspond to the $`๐ฌ^2`$ and self-dual conditions, respectively).
The full theorem was stated but not proved in . We give a proof here, based on monotone span programs. Another proof, due to Daniel Gottesman and based on purification of quantum superoperators, appeared in .
## 3 Quantum secret-sharing from classical linear schemes
We assume that the reader is familiar with the notation and basic concepts of quantum computing. For clarity, we will ignore normalization factors.
### 3.1 Pure-state linear QSS
Cramer et al. pointed out that any linear secret-sharing scheme can be realized as a MSP-based scheme. In this section, I show that any MSP with adversary structure $`๐`$ gives rise to a quantum erasure-correcting code for erasures occuring on any set of positions in $`๐๐^{}`$. In the case where $`๐`$ is self-dual, this yields a pure-state quantum secret-sharing scheme for $`๐`$.
The idea is the same as that for the CGL scheme . First choose a MSP, say $`(K,M,\psi )`$. Note that wlog all $`e`$ rows of $`M`$ are linearly independent and so we can extend $`M`$ to an invertible $`d\times d`$ matrix $`M^{}`$. We can construct a quantum circuit $`\stackrel{~}{M}`$ implementing multiplication by $`M^{}`$ and thus encode a basis state $`|s`$, for $`sK`$, as
$`\stackrel{~}{M}\left(|s{\displaystyle \underset{๐K^{e1}}{}}|a_1\mathrm{}a_{e1}|0\mathrm{}0\right)`$
$`={\displaystyle \underset{๐K^{e1}}{}}|M\left({\displaystyle \genfrac{}{}{0pt}{}{s}{๐}}\right)`$
(The expression $`\left(\genfrac{}{}{0pt}{}{s}{๐}\right)`$ denotes the column vector obtained by adjoining $`s`$ to the beginning of the vector a).
This scheme can be extended by linearity to arbitrary states $`|\varphi =_{sK}\alpha _s|s`$. The pieces of the encoded state are then distributed according to the function $`\psi `$. We have:
###### Theorem 2
Let $`(K,M,\psi )`$ be a MSP with a.s. $`๐`$. Then the encoding above is corrects erasures on any set of positions in $`๐๐^{}`$.
To prove this, we need to show for any set $`B`$ which is in $`๐`$ but whose complement is not, the players in $`A`$ can reconstruct the encoded data. We give a reconstruction procedure. The proof consists of the two following lemmas.
First we show the existence of certain vectors used in the reconstruction process.
###### Lemma 3
Let $`(K,M,\psi )`$ be a MSP with a.s. $`๐`$. Suppose $`B๐๐^{}`$ (i.e. $`A=PB`$ is in $`๐`$). Then there exists an invertible linear transformation $`U`$ on the shares of $`A`$ such that after the transformation,
1. the first share contains the secret $`s`$;
2. all remaining shares, including those of players in $`B`$, are distributed independently of $`s`$ when the $`e1`$ other components of $`๐ฌ_{}`$ are chosen at random.
Say $`A`$ contains $`m`$ shares. Then we must construct $`m`$ linearly independent vectors $`๐ฎ_1,๐ฎ_2,\mathrm{},๐ฎ_m`$ such that
1. $`๐ฎ_1^{}M_A\left(\genfrac{}{}{0pt}{}{s}{๐}\right)=s`$;
2. If $`U^{}`$ is the matrix with rows given by $`๐ฎ_2,\mathrm{},๐ฎ_m`$, then the value
$$\left(\genfrac{}{}{0pt}{}{U^{}M_A}{M_B}\right)\left(\genfrac{}{}{0pt}{}{s}{๐}\right)$$
is distributed independently of $`s`$.
To satisfy the first condition, pick any $`๐ฎ_1`$ such that $`๐ฎ_1^{}M_A=ฯต^{}`$. Such a vector must exist since by hypothesis the players in $`A`$ can reconstruct the secret.
To satisfy the second condition, itโs enough to ensure there exists $`๐ฏ`$ such that $`\left(\genfrac{}{}{0pt}{}{U^{}M_A}{M_B}\right)๐ฏ=\mathrm{๐}`$ and $`ฯต^{}๐ฏ0`$ (see section 2.2).
Since $`B๐`$, we know that there is a $`๐ฏ`$ such that $`ฯต^{}๐ฏ0`$ and $`M_B๐ฏ=\mathrm{๐}`$. Furthermore, the subspace $`W=\{๐ฎK^m:๐ฎ^{}M_A๐ฏ=0\}`$ has dimension $`m1`$, and $`๐ฎ_1`$ is not in that space since $`๐ฎ_1^{}M_A๐ฏ=ฯต^{}๐ฏ0`$. Hence any basis $`\{๐ฎ_2,\mathrm{},๐ฎ_m\}`$ of $`W`$ will do.
The matrix $`U`$ whose rows are given by the $`\text{u}_i`$โs gives the desired transformation. Note that the $`U`$ doesnโt depend on a. $`\mathrm{}`$
Finally we show that the reconstruction process works:
###### Lemma 4
Let $`(K,M,\psi )`$ be a MSP and let $`B๐๐^{}`$, $`A=PB`$. Suppose a quantum state $`|\varphi =_{sK}\alpha _s|s`$ is encoded as described at the beginning of this section. Then the shares in $`A`$ can be used to reconstruct $`|\varphi `$. Consequently, no information on $`|\varphi `$ can be obtained from the shares in $`B`$.
Consider the case when $`|\varphi =|s`$ for some $`sK`$. Then the encoded state can be written
$$\underset{๐K^{e1}}{}|M_A\left(\genfrac{}{}{0pt}{}{s}{๐}\right)|M_B\left(\genfrac{}{}{0pt}{}{s}{๐}\right)$$
Construct a quantum circuit for the map $`๐U๐`$, where $`U`$ is constructed as in lemma 3. Denote by $`U^{}`$ the matrix obtained by removing the first row of $`U`$. Applying the circuit for $`U`$ only to the components of the encoded state corresponding to $`A`$, we get
$$\begin{array}{c}\underset{๐K^{e1}}{}|UM_A\left(\genfrac{}{}{0pt}{}{s}{๐}\right)|M_B\left(\genfrac{}{}{0pt}{}{s}{๐}\right)\hfill \\ \hfill =|s\underset{๐K^{e1}}{}|U^{}M_A\left(\genfrac{}{}{0pt}{}{s}{๐}\right)|M_B\left(\genfrac{}{}{0pt}{}{s}{๐}\right)\end{array}$$
However, by construction the joint distribution of $`U^{}M_A\left(\genfrac{}{}{0pt}{}{s}{๐}\right)`$ and $`M_B\left(\genfrac{}{}{0pt}{}{s}{๐}\right)`$ is independent of $`s`$ when a is chosen uniformly at random (lemma 3). Hence, for an arbitrary state $`|\varphi `$ this procedure yields
$$|\varphi \underset{๐K^{e1}}{}|U^{}M_A\left(\genfrac{}{}{0pt}{}{0}{๐}\right)|M_B\left(\genfrac{}{}{0pt}{}{0}{๐}\right)$$
By a strong form of the no cloning theorem, the correctness of the reconstruction implies that the shares of $`B`$ give no information at all on $`|\varphi `$. $`\mathrm{}`$
(This completes the proof of theorem 2).
When the adversary structure $`๐`$ defined by a MSP is $`๐ฌ^2`$, we have $`๐๐^{}`$. Hence, the previous theorem shows that erasures on any set of coordinates in $`๐`$ can be corrected. In addition, if $`๐`$ is self-dual (i. e. both $`๐ฌ^2`$ and $`๐ฌ^2`$) then the qualified sets are precisely the complements of sets in $`๐`$ and hence every qualified set can reconstruct the secret but no unqualified set gets any information on it. Thus we have shown theorem 1 for the case of self-dual structures.
### 3.2 Mixed-state linear QSS
To handle structures which are simply $`๐ฌ^2`$, we follow the strategy of : first extend to a self-dual structure and then โtrace-outโ the new share(s).
To extend a structure $`๐`$ over a player set $`P`$, add a new player to $`P`$ (say $`\tau `$):
###### Lemma 5
For any $`๐ฌ^2`$ adversary structure $`๐`$ over a player set $`P`$, the structure $`๐^{}`$ over the set $`P^{}=P\{\tau \}`$ given by
$$๐^{}=๐\{B\{\tau \}:B๐^{}\}$$
is self-dual and its restriction to $`P`$ yields $`๐`$.
Elementary, using the fact that
$`๐`$ is $`๐ฌ^2๐^{}๐`$. $`\mathrm{}`$
Thus, a pure-state QSS scheme for $`๐^{}`$ will yield a mixed-state scheme for $`๐`$ by throwing out the share corresponding to $`\tau `$. For the construction to be efficient, we need the following:
###### Lemma 6
Given a MSP for $`๐`$, an MSP for $`๐^{}`$ can be efficiently constructed.
Note that the new access structure is $`\mathrm{\Gamma }^{}=\mathrm{\Gamma }\{B\{\tau \}:B\mathrm{\Gamma }^{}\}`$ (here $`\mathrm{\Gamma },\mathrm{\Gamma }^{},\mathrm{\Gamma }^{}`$ are the complements of $`๐,๐^{},๐^{}`$ resp.). Thus if $`f,f^{},f^{}`$ are functions detecting membership in $`๐,๐^{},๐^{}`$ respectectively, and if $`f_\tau `$ detects the presence of $`\tau `$ in a set, then $`f^{}=f(f^{}f_\tau )`$.
Now to construct the desired MSP, first obtain an MSP for $`๐^{}`$ according to . The MSP for $`A^{}`$ can then be constructed by composition from MSPโs calculating and and or. $`\mathrm{}`$
The resulting MSP is at most a constant times the size of the original.
## 4 QSS from classical SS
A natural conjecture given the results of the previous section is that *any* classical secret-sharing scheme for an adversary structure will give a quantum erasure-coorecting for erasures in $`๐๐^{}`$. I show here a condition on the scheme for this to be the case. Not all schemes satisfy the condition, though a large class of them does, in particular group-homomorphic ones.
The corollary to this, as before, is that when $`๐`$ is self-dual, the resulting quantum scheme is a qss scheme for $`๐`$. Note that the main difference between the proof we give here and that of the preivous section is that here we donโt guarantee that the reconstructon procedure is efficient, only that it exists and is unitary.
### 4.1 A general condition
A classical secret sharing scheme can be thought of as a probabilistic map $`E`$ from a secret space $`๐ฎ`$ into $`n`$ โshare spacesโ $`๐ด_1,\mathrm{},๐ด_n`$. The random input can be modeled as a choice from some set $``$ with a given probability distribution. Now consider some set $`U๐๐^{}`$ and let $`Q=U^c`$ be its complement ($`Q`$ is qualified). Let $`S`$ be the random variable corresponding to the secret and let $`Y_u`$ and $`Y_q`$ be those corresponding to the shares in $`U`$ and $`Q`$ respectively. Denote their concatenation $`E(S)=Y=Y_uY_q`$. Finally, let $`๐ด_u,๐ด_q`$ be the share spaces for $`U`$ and $`Q`$ and let $`๐ด=๐ด_u\times ๐ด_q`$ be the global share space.
Note that for the SS scheme to be perfect we must have
$`H(S|Y_q)=0`$. Equivalently, $`S=f(Y_q)`$ for some deterministic function $`f`$.
$`I(S;Y_u)=0`$. Equivalently, $`P(Y_u=y_u|S=s)=P(Y_u=y_u|S=s^{})=P(Y_u=y_u)s,s^{}๐ฎ`$.
Suppose now we have a quantum secret which is a linear superposition of shares in $`๐ฎ`$ and a unitary map $`\stackrel{~}{E}`$ such that for $`s๐ฎ`$:
$$\stackrel{~}{E}|s=\underset{y๐ด}{}\sqrt{P(Y=y|S=s)}|y$$
This can in fact be rewritten as
$$\begin{array}{c}\underset{y_q:f(y_q)=s}{}\sqrt{P(Y_q=y_q|S=s)}|y_q\hfill \\ \hfill \underset{y_u๐ด_u}{}\sqrt{P(Y_u=y_u|Y_q=y_q)}|y_u\end{array}$$
We want to decide if this is can correct erasures on $`U`$. To do so requires showing that the density matrix of the $`U`$ component is independent of the secretโs state. Note that it is not sufficient to show that the density matrix is the same for all $`|s`$. We have to show this for all choices of the $`\alpha _s`$โs in $`_{s๐ฎ}\alpha _s|s`$. We can compute the density matrix explicitly by imagining that a measure is made on the $`Q`$ component of the code and the secret. We can then consider $`P(S=s)`$ to be $`|\alpha _s|^2`$. In what follows $`๐_{๐ผ\mathbf{|}๐_๐}`$ is the density matrix of $`U`$ given $`Y_q=y_q`$.
$$\begin{array}{c}\rho _u\hfill \\ \hfill =\underset{s๐ฎ}{}|\alpha _s|^2\underset{y_q๐ด_q}{}P(Y_q=y_q|S=s)๐_{๐ผ\mathbf{|}๐_๐}\\ \hfill =\underset{y_q๐ด_q}{}P(Y_q=y_q)๐_{๐ผ\mathbf{|}๐_๐}\\ \hfill =\underset{y_q๐ด_q}{}P(Y_q=y_q)\\ \hfill \left(\underset{y_u^{(1)}๐ด_u}{}\sqrt{P(Y_u=y_u^{(1)}|Y_q=y_q)}\right|y_u^{(1)})\\ \hfill \left(\underset{y_u^{(2)}๐ด_u}{}\sqrt{P(Y_u=y_u^{(2)}|Y_q=y_q)}y_u^{(2)}|\right)\\ \hfill =\underset{y_u^{(1)},y_u^{(2)}๐ด_u}{}\underset{y_q๐ด_q}{}\sqrt{P(Y_u=y_u^{(1)},Y_q=y_q)}\\ \hfill \sqrt{P(Y_u=y_u^{(2)},Y_q=y_q)}|y_u^{(1)}y_u^{(2)}|\end{array}$$
The matrices in the set
$$\left\{|y_u^{(1)}y_u^{(2)}|:y_u^{(1)},y_u^{(2)}๐ด_u\right\}$$
are linearly independent. Their coefficients are
$$\begin{array}{c}\underset{y_q๐ด_q}{}\sqrt{P(Y_u=y_u^{(1)},Y_q=y_q)P(Y_u=y_u^{(2)},Y_q=y_q)}\hfill \\ \hfill =\underset{s๐ฎ}{}|\alpha _s|^2\underset{y_q:f(y_q)=s}{}\sqrt{P(Y_u=y_u^{(1)},Y_q=y_q|S=s)}\\ \hfill \sqrt{P(Y_u=y_u^{(2)},Y_q=y_q|S=s)}\end{array}$$
For $`\rho _u`$ to be independent of the choice of $`\alpha _s`$ we must therefore have
$$\begin{array}{c}\underset{y_q:f(y_q)=s}{}\sqrt{P(Y_u=y_u^{(1)},Y_q=y_q|S=s)}\hfill \\ \hfill \sqrt{P(Y_u=y_u^{(2)},Y_q=y_q|S=s)}\end{array}$$
(1)
independent of $`s`$ for all $`y_u^{(1)},y_u^{(2)}๐ด_u`$. Thus
###### Theorem 7
Given a classical SS scheme for an adversary structure $`๐`$, the correspnding quantum scheme corects erasures on $`U๐๐^{}`$, iff Equation (1) is independent of $`s`$ for all $`y_u^{(1)},y_u^{(2)}๐ด_u`$.
As unnatural as this condition seems, it is nonetheless satisfied by many SS schemes:
* If $`Y_u`$ is a function of $`Y_q`$ (as is the case in the Shamir scheme) then we have the expression (1) equal to 0 whenever $`y_u^{(1)}y_u^{(2)}`$. Furthermore, when $`y_u^{(1)}=y_u^{(2)}=y_u`$ the expression reduces to $`_{y_q:f(y_q)=s}P(Y_u=y_u,Y_q=y_q|S=s)`$, which sums to $`P(Y_u=y_u|S=s)`$. This is independent of $`s`$ by the secrecy assumption above. Thus this type of scheme yields a secure QSS.
* A group homomorphic secret sharing scheme is based on an injective homomorphism $`h:G\times G^mG^n`$ for some group $`G`$. The secret $`s`$ is an element of $`G`$ and the $`n`$ shares are obtained by picking $`๐ฏ_RG^m`$ and calculating $`h(s,๐ฏ)`$.
In this case, the independence of expression (1) from $`s`$ is guaranteed by the following fact: in any homomorphic ss scheme, either two words $`y_u^{(1)},y_u^{(2)}`$ never appear with the same word $`y_q`$ (that is
$$P(Y_u=y_u^{(1)}|Y_q=y_q)P(Y_u=y_u^{(2)}|Y_q=y_q)=0$$
for all $`y_q`$) or they always appear with the same probability:
$$\begin{array}{c}\sqrt{P(Y_u=y_u^{(1)}|Y_q=y_q)P(Y_u=y_u^{(2)}|Y_q=y_q)}\hfill \\ \hfill =P(Y_u=y_u^{(1)}|Y_q=y_q).\end{array}$$
The same analysis as before applies: qss schemes constructed from homomorphic schemes are secure. Interestingly, there seem to be no cases where non-homomorphic schemes provide any advantage over homomorphic ones .
Thus, it seems that although not all classical ss schemes yield a qss scheme directly, the most important ones do. However, the proof given does not give the reconstruction procedure; it only proves its existence. It is not *a priori* clear that all classical SS schemes which yield a secure QSS scheme will have efficient (quantum) reconstruction procedures.
### Acknowledgements
I would like to thank Richard Cleve, Claude Crรฉpeau, Daniel Gottesman and Paul Dumais for helpful discussions. |
warning/0001/hep-ph0001028.html | ar5iv | text | # On the Resummed Hadronic Spectra of Inclusive ๐ต Decays
## I Introduction
Inclusive $`B`$ decays are considered fertile ground for precision tests of the standard model. The process $`BX_ue\nu `$ can be used to extract the all important Cabibbo-Kobayashi-Maskawa (CKM) matrix element $`V_{ub}`$, while $`BX_s\gamma `$ decays are important for discovering new physics. However, the utility of experimental measurements of these processes is bounded by our ability to control the theoretical errors. Tremendous effort has gone into determining ways to calculate these rates in a systematic fashion. Indeed, the algorithm for calculating these rates is now part of the theoretical canon . Unfortunately, experimental cuts complicate life for theorists. In particular, these cuts often force us to work near the boundary of the phase space, where the aforementioned canonical techniques break down. Only now are we learning how to retool our calculations to accommodate these highly non-trivial issues.
The complications arise if the cut forces us into a corner of phase space, since the calculation can now depend on a new parameter, $`\rho `$, which is a measure of the relative size of phase space of interest. When this parameter becomes parametrically small, the systematics of the calculation usually break down; perturbative QCD corrections become enhanced by large logs of the form $`\mathrm{log}\rho `$, while the non-perturbative expansion in $`\mathrm{\Lambda }/m_b`$ becomes an expansion in $`\mathrm{\Lambda }/(\rho m_b)`$.
The most relevant cut rate arises in semi-leptonic $`B`$ decays, where one wishes to measure $`V_{ub}`$ by eliminating the large background from charmed transitions. To eliminate this background, we have a choice of variables with which to cut. Perhaps the simplest choice is the electron energy, which is the oldest method used for extracting $`|V_{ub}|`$. Unfortunately, as has been widely discussed in the literature, such extractions are typically model dependent since the rate in this window is sensitive to the Fermi motion of the heavy quark. There is no way to write down a meaningful theoretical error for such extractions. It is only very recently that a model independent method has been proposed, within a well defined systematic scheme , that could lead to an extraction with a well defined error.
It is also possible to remove the background from charmed transitions by cutting on the hadronic invariant mass . While this choice presents a greater experimental challenge, it benefits from the fact that, unlike the electron spectrum, most of the $`BX_ue\nu `$ decays are expected to lie within the region $`s_H<M_D^2`$. Furthermore, it is believed that even though both the invariant mass region $`s_H<M_D^2`$ and electron energy regions $`M_B/2>E_e>(M_B^2M_D^2)/(2M_B)`$ receive contributions from hadronic final states with invariant mass up to $`M_D`$, the cut mass spectrum will be less sensitive to local duality violations. This belief rests on the fact that the contribution of large mass states is kinematically suppressed for the electron energy spectrum in the region of interest.
The goal of this paper is to study the viability of extracting $`V_{ub}`$ from the invariant mass spectrum. In particular, we are interested in studying the breakdown of the perturbative expansion for the cut rate, and whether or not a reorganized expansion can be used reliably.
Building upon the work of Korchemsky and Sterman , we begin by discussing how the doubly differential decay rate factorizes in moment space. We then use the recent results of to calculate a closed form expression for the inverse Mellin transform at next-to-leading logarithmic (NLL) order. The result for the resummed rate is presented in terms of the partonic as well as hadronic invariant mass. This result is used to extract a piece of the two loop rate, the size of which can be compared to the Brodsky-Lepage-Mackenzie (BLM) two loop correction. We then determine the region of invariant mass where resummation is necessary as well as the region where the reorganized expansion breaks down. We conclude with a brief discussion of the phenomenology, saving a complete discussion, including the effects of Fermi motion, for a later publication.
## II Factorization in Hadronic Variables
The problem of summing large threshold logarithms in perturbative expansions, which arise due to incomplete KLN cancellation of the IR sensitivity at the edge of phase space, has been addressed for various processes . The technique relies on factorization, which allows for resummations via a renormalization group equation. The factorization in $`B`$ decays has been previously discussed in leptonic variables in . Here we review the arguments that are germane to our discussion of factorization in terms of hadronic variables.
Consider the inclusive semi-leptonic decay of the $`b`$ quark into a lepton pair with momenta $`q=(p_e+p_\nu )`$ and a hadronic jet of momenta $`p_h`$. It is convenient to define the following partonic kinematic variables in the rest frame of the $`B`$ meson $`v=(1,\stackrel{}{0})`$,
$$\widehat{s}_0=\frac{p_h^2}{m_b^2},h=\frac{2vp_h}{m_b},x=\frac{2vp_e}{m_b},$$
(1)
with phase space boundaries
$$0\widehat{s}_01,2\sqrt{\widehat{s}_0}h1+\widehat{s}_0,1\frac{h}{2}\frac{1}{2}\sqrt{h^24\widehat{s}_0}x1\frac{h}{2}+\frac{1}{2}\sqrt{h^24\widehat{s}_0}.$$
(2)
In addition, it is customary to define the leptonic variables
$$y_0=\frac{2vq}{m_b},y=\frac{q^2}{m_b^2}.$$
(3)
In terms of the leptonic variables, $`\widehat{s}_0=(1y_0+y)`$ and $`h=2y_0`$, one can see that in the endpoint region of the electron energy spectrum when $`x1`$ with $`y<1`$, the invariant mass of the jet approaches zero with its energy held fixed. In addition, the jet hadronizes at a much later time in the rest frame of the $`B`$ meson, due to the time dilation. Factorization exploits this and separates the particular differential rate under consideration into subprocesses with disparate scales. This factorization fails when the jet energy vanishes in the dangerous region $`yx1`$. However, this problematic region of phase space is suppressed because the rate to produce soft massless fermions vanishes at tree level.
The infrared sensitive regions, which give rise to the large logarithms, can be determined by constructing a reduced diagram, as shown in Fig. 1. According the Coleman-Norton theorem , a diagram at the infrared singular point must describe a physically realizable process after contracting all off-shell lines to a point. In the figure, $`S`$ denotes a soft blob which interacts with the jet and the $`b`$ quark via soft lines. $`J`$ denotes the hadronic jet and $`H`$ the hard scattering amplitude. Thus, the reduced diagram in Fig. 1 is simply a visualization of factorization. An important consequence of this factorization is that the soft function $`S`$ is universal. Moreover, it has been shown that the soft function $`S`$ contains the non-perturbative structure function introduced in Ref. . It is this universality that will eventually allow us to eliminate the dependence on unknown non-perturbative hadronic dynamics.
From the above discussions one can see that in terms of the variables introduced earlier, factorization holds when $`\widehat{s}_00`$ and $`h1`$, or equivalently $`1\widehat{s}_0/h`$. The typical momenta flowing through the hard subprocesses are $`๐ช(m_b)`$. Thus, $`H`$ does not contain any large threshold logarithms and has a well-defined perturbative expansion in $`\alpha _s(m_b)`$. The soft function $`S`$ contains typical momentum $`k`$, with $`k^+k^{}k_{}=๐ช(m_b\widehat{s}_0/h)`$. By soft we mean soft compared to $`m_b`$, but still larger than $`\mathrm{\Lambda }_{\mathrm{QCD}}`$. For an energetic $`u`$ quark moving in the โ$``$โ direction, the jet subprocess has typical momenta $`p`$ such that $`p^{}p^+,p_{}`$ with $`p^+=๐ช(m_b\widehat{s}_0/h)`$, $`p_{}^2=๐ช(m_b^2\widehat{s}_0/h)`$, and $`p^{}=๐ช(m_b)`$. In order to delineate between momentum regimes, a factorization scale $`\mu `$ is introduced. The fact that the process is independent of the factorization scale $`\mu `$ is utilized to sum the large threshold logarithms in the soft and jet functions. The reduced diagram for the inclusive radiative decays $`BX_s\gamma `$ is exactly the same as above if we replace the lepton pair with a photon and ignore the strange quark mass.
In terms of $`\widehat{s}_0`$ and $`h`$, the triply differential rate, which factorizes into hard, jet and soft subprocesses, may be written as
$`{\displaystyle \frac{1}{\mathrm{\Gamma }_0}}{\displaystyle \frac{d^3\mathrm{\Gamma }}{d\widehat{s}_0dhdx}}=`$ (4)
$`12(2hx)(x+h1\widehat{s}_0){\displaystyle _\xi ^{M_B/m_b}}๐zS(z)m_b^2J[m_b^2h(z\xi ),\mu ]H(m_bh/\mu ),`$ (5)
$`\mathrm{\Gamma }_0={\displaystyle \frac{G_F^2}{192\pi ^3}}|V_{ub}|^2m_b^5,`$ (6)
where $`\xi =1\widehat{s}_0/h`$ is analogous to the Bjorken scaling variable in deep inelastic scattering. $`z=1+k_+/m_b`$, where $`k_+`$ is the heavy quark light cone residual momentum. $`S(z)`$ essentially describes the probability for the $`b`$ quark to carry light cone momentum fraction $`z`$ and allows for a leakage past the partonic endpoint, as can be seen explicitly in the upper limit of $`z`$. A similar factorized expression holds for the inclusive radiative decays near the endpoint,
$`{\displaystyle \frac{1}{\mathrm{\Gamma }_0^\gamma }}{\displaystyle \frac{d\mathrm{\Gamma }^\gamma }{d\widehat{s}_0}}`$ $`=`$ $`{\displaystyle _{1\widehat{s}_0}^{M_B/m_b}}๐zS(z)m_b^2J[m_b^2(z1+\widehat{s}_0),\mu ^2]H(m_b/\mu ),`$ (7)
$`\mathrm{\Gamma }_0^\gamma `$ $`=`$ $`{\displaystyle \frac{G_F^2}{32\pi ^4}}|V_{ts}^{}V_{tb}|^2\alpha C_7^2m_b^5.`$ (8)
For the inclusive semi-leptonic decays, the integration over $`x`$ can be done in the endpoint region and the resulting doubly differential rate is<sup>*</sup><sup>*</sup>* Here we confirm explicitly that the dangerous region $`h0`$, where the energy of the hadronic jet vanishes and factorization fails, is suppressed by the pre-factor and therefore not important.
$$\frac{1}{\mathrm{\Gamma }_0}\frac{d^2\mathrm{\Gamma }}{d\widehat{s}_0dh}=2h^2(32h)_\xi ^{M_B/m_b}๐zS(z)m_b^2J[m_b^2h(z\xi ),\mu ]H(m_bh/\mu )+๐ช(\widehat{s}_0).$$
(9)
This is where the factorization in the invariant mass spectrum is simpler than in the electron energy spectrum. In the latter case, none of the integrals in the triply differential rate can be done trivially and one has to take an extra derivative with respect to $`x`$ to arrive at an expression similar to Eq. (9). An interesting consequence of factorization is that it connects the electron energy spectrum in the region $`x1`$ with the invariant mass spectrum, in the region $`\widehat{s}_00`$,
$$\frac{d\mathrm{\Gamma }}{d\widehat{s}_0}|_{\widehat{s}_01x}=\frac{1}{2}\frac{d}{dx}\frac{d\mathrm{\Gamma }}{dx},$$
(10)
which can be verified explicitly at the one loop level using the corresponding expressions in Ref. .
## III The Perturbative Resummation
At one loop level the differential rates are
$`{\displaystyle \frac{1}{\mathrm{\Gamma }_0}}{\displaystyle \frac{d^2\mathrm{\Gamma }}{d\widehat{s}_0dh}}`$ $`=`$ $`2h^2(32h)\left[\delta (\widehat{s}_0)+{\displaystyle \frac{C_F\alpha _s}{4\pi }}E_1(h,\widehat{s}_0)\right]+{\displaystyle \frac{C_F\alpha _s}{4\pi }}E_2(h,\widehat{s}_0),`$ (11)
$`{\displaystyle \frac{1}{\mathrm{\Gamma }_0^\gamma }}{\displaystyle \frac{d\mathrm{\Gamma }^\gamma }{d\widehat{s}_0}}`$ $`=`$ $`\delta (\widehat{s}_0)\left[1{\displaystyle \frac{C_F\alpha _s}{4\pi }}\left(13+{\displaystyle \frac{4}{3}}\pi ^2\right)\right]`$ (13)
$`+{\displaystyle \frac{C_F\alpha _s}{4\pi }}\left[6+3\widehat{s}_02\widehat{s}_0^22(2\widehat{s}_0)\mathrm{log}\widehat{s}_0\left({\displaystyle \frac{7}{\widehat{s}_0}}+4{\displaystyle \frac{\mathrm{log}\widehat{s}_0}{\widehat{s}_0}}\right)_+\right],`$
where
$`E_1(h,\widehat{s}_0)`$ $`=`$ $`\delta (\widehat{s}_0)\left[8\mathrm{log}^2(h)10\mathrm{log}h+{\displaystyle \frac{2\mathrm{log}h}{1h}}+4\mathrm{Li}_2(1h)+5+{\displaystyle \frac{4}{3}}\pi ^2\right]`$ (15)
$`4\left({\displaystyle \frac{\mathrm{log}\widehat{s}_0}{\widehat{s}_0}}\right)_++(8\mathrm{log}h7)\left({\displaystyle \frac{1}{\widehat{s}_0}}\right)_++{\displaystyle \frac{1}{\widehat{s}_0}}\left[8\mathrm{log}\left({\displaystyle \frac{1+t}{2}}\right)+7(1t)\right],`$
$`E_2(h,\widehat{s}_0)`$ $`=`$ $`\delta (\widehat{s}_0){\displaystyle \frac{4h^3\mathrm{log}h}{1h}}4\left[2h(34h)3(12h)\widehat{s}_02\widehat{s}_0^2\right]\mathrm{log}\left({\displaystyle \frac{1+t}{1t}}\right)`$ (17)
$`+\mathrm{\hspace{0.17em}4}ht(1015h+8\widehat{s}_0),`$
and $`t=\sqrt{14\widehat{s}_0/h^2}`$. We also adopt the following definition for the โ+โ distributions
$$\left(\frac{\mathrm{log}^n(\widehat{s}_0)}{\widehat{s}_0}\right)_+=\underset{ฯต0}{lim}\left[\theta (\widehat{s}_0ฯต)\frac{\mathrm{log}^n(\widehat{s}_0)}{\widehat{s}_0}+\delta (\widehat{s}_0)\frac{\mathrm{log}^{n+1}(ฯต)}{n+1}\right].$$
(18)
This definition is such that
$$_0^\rho ๐\widehat{s}_0F(\widehat{s}_0)\left(\frac{\mathrm{log}^n(\widehat{s}_0)}{\widehat{s}_0}\right)_+=F(0)\frac{\mathrm{log}^{n+1}(\rho )}{n+1}+_0^\rho ๐\widehat{s}_0\left[F(\widehat{s}_0)F(0)\right]\frac{\mathrm{log}^n(\widehat{s}_0)}{\widehat{s}_0}.$$
(19)
Note that if the parameter $`\rho `$ becomes parametrically small, the first term on the right hand side of Eq. (19) will give large logarithms thereby spoiling the systematics of the perturbative expansions, while the second term must be regular as $`\rho 0`$. To perform the resummation we go into the moment space where the amplitudes factorize completely. In the case of inclusive semi-leptonic decays, it is convenient to define a new variable $`\lambda =\widehat{s}_0/h`$ with kinematic range
$`0h1;0\lambda {\displaystyle \frac{h}{4}},`$ (20)
$`1h2;1{\displaystyle \frac{1}{h}}\lambda {\displaystyle \frac{h}{4}}.`$ (21)
The region with $`h1`$ is populated with real gluon emissions only, whereas the region with $`h1`$ has both real and virtual gluon corrections. In the relevant region $`h1`$ and $`\lambda 0`$ the contributions from real and virtual gluon emissions combine to give terms which are โ+โ distributions, which upon integrating up to a cut, lead to the large logs we wish to resum.
To proceed, we take the $`N`$th moment with respect to $`\xi =1\lambda `$ in the large $`N`$ limit. In the region $`\widehat{s}_00`$ and $`z\xi 1`$, one can replace $`J[m_b^2h(z\xi )]`$ in Eq. (9) with $`J[m_b^2h(1\xi /z)]`$. This replacement is permissible to the order we are working. We then obtain
$`M_N`$ $`=`$ $`{\displaystyle _0^1}๐\lambda (1\lambda )^{N1}{\displaystyle \frac{1}{\mathrm{\Gamma }_0}}{\displaystyle \frac{d^2\mathrm{\Gamma }}{d\lambda dh}}`$ (22)
$`=`$ $`2h^2(32h)S_NJ_N(m_b^2h/\mu ^2)H(m_b/\mu )+๐ช(1/N),`$ (23)
$`J_N(m_b^2/\mu ^2)`$ $`=`$ $`m_b^2{\displaystyle _0^1}๐yy^{N1}J[m_b^2(1y),\mu ],`$ (24)
$`S_N`$ $`=`$ $`{\displaystyle _0^{M_B/m_b}}๐zz^NS(z).`$ (25)
The soft moment $`S_N`$ further decomposes into a perturbative soft piece, which accounts for soft gluon radiation and a non-perturbative piece which incorporates bound state dynamics and serves as the boundary condition for the renormalization group equation .
$`S(z)`$ $`=`$ $`{\displaystyle _z^{M_b/m_b}}{\displaystyle \frac{dy}{y}}f[m_b(1y)]\sigma (z/y),`$ (26)
$`S_N`$ $`=`$ $`f_N\sigma _N,`$ (27)
where $`f(y)=B(v)|\overline{b}_v\delta (yiD_+)b_v|B(v)`$ is the non-perturbative structure function defined in Ref. . A similar expression holds for the inclusive radiative decays $`BX_s\gamma `$
$`M_N^\gamma `$ $`=`$ $`{\displaystyle _0^1}๐\widehat{s}_0(1\widehat{s}_0)^{N1}{\displaystyle \frac{1}{\mathrm{\Gamma }_0^\gamma }}{\displaystyle \frac{d\mathrm{\Gamma }^\gamma }{d\widehat{s}_0}}`$ (28)
$`=`$ $`f_N\sigma _NJ_NH^\gamma +๐ช(1/N),`$ (29)
Subsequently, we will ignore the non-perturbative structure function $`f(y)`$ and concentrate on the perturbative resummations.
It merits emphasizing that the large $`N`$ asymptotics of the moments corresponds to the behavior of the spectra in the region $`\widehat{s}_0\lambda 0`$. Taking the large $`N`$ limit also enables us to extend the integration limit of $`\lambda `$ in Eq. (22) up to $`1`$, despite the fact that the kinematic range of $`\lambda `$ never goes up to $`1`$. In this limit the contribution from the region $`\lambda 1`$ is power suppressed.
Comparing Eq. (22) with the corresponding expression for the electron energy spectrum in Ref. , one sees that the moments $`\sigma _N`$ and $`J_N`$ are identical with those in the electron energy spectrum, with change of variables $`x1\lambda `$ and $`2y_0h`$. A similar identification for the resummed radiative decays, Eq. (28), can be made with the change of variables $`x1\widehat{s}_0`$. In moment space the soft and jet functions have been calculated to NLL order and are given by
$`\sigma _NJ_N`$ $`=`$ $`\mathrm{exp}[\mathrm{log}(N)g_1(\chi )+g_2(\chi )]`$ (30)
$`\sigma _NJ_N^\gamma `$ $`=`$ $`\mathrm{exp}[\mathrm{log}(N)g_1(\chi )+g_2^\gamma (\chi )],`$ (31)
where $`\chi =\alpha _s(m_b^2)\beta _0\mathrm{log}N`$, and $`g_1`$ and $`g_2`$ are given explicitly as
$`g_1(\chi )`$ $`=`$ $`{\displaystyle \frac{2}{3\pi \beta _0\chi }}[(12\chi )\mathrm{log}(12\chi )2(1\chi )\mathrm{log}(1\chi )],`$ (32)
$`g_2(\chi )`$ $`=`$ $`g_2^\gamma (\chi )+g_{sl}(\chi ,h),`$ (33)
$`g_2^\gamma (\chi )`$ $`=`$ $`{\displaystyle \frac{k}{3\pi ^2\beta _0^2}}[\mathrm{\hspace{0.17em}2}\mathrm{log}(1\chi )\mathrm{log}(12\chi )]{\displaystyle \frac{2\beta _1}{3\pi \beta _0^3}}[\mathrm{log}(12\chi )2\mathrm{log}(1\chi ){\displaystyle \frac{}{}}`$ (36)
$`+{\displaystyle \frac{1}{2}}\mathrm{log}^2(12\chi )\mathrm{log}^2(1\chi )]{\displaystyle \frac{1}{\pi \beta _0}}\mathrm{log}(1\chi ){\displaystyle \frac{2}{3\pi \beta _0}}\mathrm{log}(12\chi )`$
$`+{\displaystyle \frac{4\gamma _E}{3\pi \beta _0}}\left[\mathrm{log}(12\chi )\mathrm{log}(1\chi )\right],`$
$`g_{sl}(\chi ,h)`$ $`=`$ $`{\displaystyle \frac{4}{3\pi \beta _0}}\mathrm{log}(h)\mathrm{log}(1\chi ).`$ (37)
In the above, $`\beta _0=(11C_A2N_f)/(12\pi )`$, $`\beta _1=(17C_A^25C_AN_f3C_FN_f)/(24\pi ^2)`$, $`k=C_A(67/18\pi ^2/6)10T_RN_f/9`$, and $`\gamma _E=0.577216\mathrm{}`$ is the Euler-Mascheroni constant. In our case, $`C_A=3`$, $`C_F=4/3`$, and $`T_R=1/2`$. To get back the physical spectra from the moment space, the inverse Mellin transform has to be evaluated at NLL accuracy as well. To this end, we apply the identity derived in the Appendix of Ref.
$`{\displaystyle \frac{1}{2\pi i}}{\displaystyle _{Ci\mathrm{}}^{C+i\mathrm{}}}๐Nx^Ne^{\mathrm{log}(N)F_1[\alpha _s\mathrm{log}N]+F_2[\alpha _s\mathrm{log}N]}`$ (38)
$`=x{\displaystyle \frac{d}{dx}}\left\{\theta (1x){\displaystyle \frac{e^{lF_1(\alpha _sl)+F_2(\alpha _sl)}}{\mathrm{\Gamma }\left[1F_1(\alpha _sl)\alpha _slF_1^{}(\alpha _sl)\right]}}\times \left[1+(\alpha _s,l)\right]\right\},`$ (39)
where $`l=\mathrm{log}(\mathrm{log}x)\mathrm{log}(1x)`$, and
$$(\alpha _s,l)=\underset{k=1}{\overset{\mathrm{}}{}}\alpha _s^k\underset{j=0}{\overset{k1}{}}f_{kj}l^j$$
(40)
represents next-to-next-to-leading log contributions. Changing variables from $`\lambda `$ back to $`\widehat{s}_0/h`$, we obtain
$`{\displaystyle \frac{1}{\mathrm{\Gamma }_0}}{\displaystyle \frac{d^2\mathrm{\Gamma }}{d\widehat{s}_0dh}}`$ $`=`$ $`2h^2(32h)H(h)`$ (42)
$`\times {\displaystyle \frac{d}{d\widehat{s}_0}}\left\{\theta \left({\displaystyle \frac{\widehat{s}_0}{h}}\eta \right){\displaystyle \frac{e^{lg_1(\alpha _s\beta _0l)+g_2(\alpha _s\beta _0l)}}{\mathrm{\Gamma }\left[1g_1(\alpha _s\beta _0l)\alpha _s\beta _0lg_1^{}(\alpha _s\beta _0l)\right]}}\right\},`$
$`{\displaystyle \frac{1}{\mathrm{\Gamma }_0^\gamma }}{\displaystyle \frac{d\mathrm{\Gamma }^\gamma }{d\widehat{s}_0}}`$ $`=`$ $`H^\gamma {\displaystyle \frac{d}{d\widehat{s}_0}}\left\{\theta (\widehat{s}_0\eta ){\displaystyle \frac{e^{lg_1(\alpha _s\beta _0l^\gamma )+g_2^\gamma (\alpha _s\beta _0l^\gamma )}}{\mathrm{\Gamma }\left[1g_1(\alpha _s\beta _0l^\gamma )\alpha _s\beta _0l^\gamma g_1^{}(\alpha _s\beta _0l^\gamma )\right]}}\right\},`$ (43)
where $`l=\mathrm{log}\left(\mathrm{log}(1\lambda )\right)\mathrm{log}(\widehat{s}_0/h)`$, and $`l^\gamma =\mathrm{log}\left(\mathrm{log}(1\widehat{s}_0)\right)\mathrm{log}(\widehat{s}_0)`$. The $`\theta `$-functions define the differential rates in a distribution sense, as $`\eta 0`$, and turn the singular terms into the โ$`+`$โ distributions, as can be seen explicitly by expanding in power series of $`\mathrm{log}\widehat{s}_0`$ and using the definition Eq. (18).
The hard parts can be obtained through the one loop results Eq. (11) and Eq. (13)
$`H(h)`$ $`=`$ $`1{\displaystyle \frac{2\alpha _s}{3\pi }}\left[4\mathrm{log}^2(h)5\mathrm{log}h+{\displaystyle \frac{3\mathrm{log}h}{32h}}+2\mathrm{L}\mathrm{i}_2(1h)+{\displaystyle \frac{5}{2}}+{\displaystyle \frac{2\pi ^2}{3}}\right]`$ (44)
$`H^\gamma `$ $`=`$ $`1{\displaystyle \frac{2\alpha _s}{3\pi }}\left({\displaystyle \frac{13}{2}}+{\displaystyle \frac{2\pi ^2}{3}}\right).`$ (45)
Eq. (42) and Eq. (43) reproduce the dominate contribution at one loop level in the limit $`\lambda 0`$ and $`\widehat{s}_00`$, respectively, and include the infinite set of terms of the form $`\alpha _s^n\mathrm{log}^{n+1}(\widehat{s}_0)`$ and $`\alpha _s^n\mathrm{log}^n(\widehat{s}_0)`$ in the Sudakov exponent for both semi-leptonic and radiative decays.
## IV The Integrated Cut Invariant Mass Spectrum
As previously mentioned, it has been proposed that we measure the modulus of the CKM matrix element $`V_{ub}`$ from inclusive decays by making a cut on the hadronic invariant mass below $`M_D^2`$. This cut eliminates the overwhelming background from bottom to charm transitions. While it is the hadronic invariant mass which is of interest, we shall first consider the cut partonic invariant mass, as it will be relevant to our conclusions.
The use of an upper cut $`c_0`$ on the partonic invariant mass introduces large logs of the form $`\alpha _s\mathrm{log}^2(c_0)`$. As $`c_0`$ approaches zero, the logs become parametrically large and need to be resummed. Using the resummation formulas in the previous section, it is simple to generate an expression for the resummed cut rate, since our expression can be written as a total derivative with respect to $`\widehat{s}_0`$. The cut rate may be written as
$$\frac{1}{\mathrm{\Gamma }_0}\mathrm{\Gamma }(c_0)=_0^1๐h\mathrm{\hspace{0.33em}2}h^2(32h)\left[\xi \left(\frac{c_0}{h}\right)\xi _{\alpha _s}\left(\frac{c_0}{h}\right)\right]+_0^{c_0}๐\widehat{s}_0_{2\sqrt{\widehat{s}_0}}^{1+\widehat{s}_0}๐h\gamma _{\alpha _s}(h,\widehat{s}_0),$$
(46)
where
$$\xi (c_0/h)=\frac{e^{lg_1(\alpha _s\beta _0l)+g_2(\alpha _s\beta _0l)}}{\mathrm{\Gamma }\left[1g_1(\alpha _s\beta _0l)\alpha _s\beta _0lg_1^{}(\alpha _s\beta _0l)\right]}|_{l=\mathrm{log}(c_0/h)},$$
(47)
and $`\gamma _{\alpha _s}(h,\widehat{s}_0)`$ is simply the one loop rate defined in Eq. (11). $`\xi _{\alpha _s}`$ stands for terms up to $`๐ช(\alpha _s)`$ when expanding $`\xi `$ in power series of $`\alpha _s`$. We subtracted it from $`\xi `$ in order to ensure that we correctly reproduce the one loop result at order $`\alpha _s`$. From Eq. (46) one can see that the large logs arise when $`\widehat{s}_0`$ approaches zero, which is not only the lower kinematic limit for $`\widehat{s}_0`$, but also the phase space boundary for virtual and real gluon emissions.
The perturbative expansion has been reorganized into an expansion in the exponent. The systematics of this expansion have been discussed at length in . Here we just recall that we may test the convergence of the reorganized expansion by comparing the NLL resummed result with the leading-logarithm (LL) resummed result. Fig. 2 shows the cut rate as a function of $`c_0`$. In this figure we show the one loop cut rate as well as the resummed cut rate with and without NLL corrections. We see that for $`0.1<c_0<0.2`$ the resummation becomes necessary, while for $`c_0<0.1`$ the NLL dominates the LL, so that we can no longer trust our results. Indeed, this breakdown occurs well before we reach the Landau pole at $`\widehat{s}_p=e^{1/(2\alpha _s\beta _0)}0.028`$.
We now expand this result to pick out the leading and next to leading infrared log contribution to the two loop differential rate. This contribution is given by
$$\frac{1}{\mathrm{\Gamma }_0}\frac{d\mathrm{\Gamma }}{d\widehat{s}_0dh}|_{๐ช(\alpha _s^2)}=2h^2(32h)\frac{\alpha _s^2}{3\pi ^2}\left[\frac{8}{3}\left(\frac{\mathrm{log}^3(\widehat{s}_0)}{\widehat{s}_0}\right)_++(146\pi \beta _0)\left(\frac{\mathrm{log}^2(\widehat{s}_0)}{\widehat{s}_0}\right)_+\right]$$
(48)
As expected, we see that the most singular contribution at $`๐ช(\alpha _s^2)`$ doesnโt have any terms proportional to $`\beta _0`$. It may be the case that, in this particular region of phase space the infrared logs terms may dominate over the BLM terms. Such a conclusion was reached in for the two loop contribution to lepton and photon spectra in semi-leptonic and radiative decays, respectively. Naturally , this does not preclude the possibility that there exist a cancellation with other uncalculated terms, such that the $`\alpha _s^2\beta _0`$ still dominate.
Let us now consider the physical case, where we are interested in placing an upper cut on the hadronic invariant mass. The hadronic invariant mass may be written as
$`\widehat{s}_H`$ $`=`$ $`{\displaystyle \frac{s_H}{m_b^2}}=\widehat{s}_0+ฯตh+ฯต^2,`$ (49)
$`ฯต`$ $`=`$ $`{\displaystyle \frac{\overline{\mathrm{\Lambda }}}{m_b}},`$ (50)
where $`\overline{\mathrm{\Lambda }}`$ is the mass difference $`M_Bm_b`$ in the infinite $`b`$ quark mass limit, which is a measure of binding energy for the $`b`$ quark inside the $`B`$ meson. Thus, given a cut on $`\widehat{s}_H`$, $`c`$, we may translate this into an $`h`$ dependent cut on $`\widehat{s}_0`$. After changing the order of integration we find that the cut rate may be written as
$$\frac{1}{\mathrm{\Gamma }_0}\mathrm{\Gamma }(c)=_0^{\frac{cฯต(1+ฯต)}{1+ฯต}}๐\widehat{s}_0_{2\sqrt{\widehat{s}_0}}^{1+\widehat{s}_0}๐h\left(\frac{d\mathrm{\Gamma }}{d\widehat{s}_0dh}\right)+_{\frac{cฯต(1+ฯต)}{1+ฯต}}^{(\sqrt{c}ฯต)^2}๐\widehat{s}_0_{2\sqrt{\widehat{s}_0}}^{\frac{c\widehat{s}_0ฯต^2}{ฯต}}๐h\left(\frac{d\mathrm{\Gamma }}{d\widehat{s}_0dh}\right),$$
(51)
if $`c>ฯต(1+ฯต)`$, whereas if $`c<ฯต(1+ฯต)`$ then
$$\frac{1}{\mathrm{\Gamma }_0}\mathrm{\Gamma }(c)=_0^{(\sqrt{c}ฯต)^2}๐\widehat{s}_0_{2\sqrt{\widehat{s}_0}}^{\frac{c\widehat{s}_0ฯต^2}{ฯต}}๐h\left(\frac{d\mathrm{\Gamma }}{d\widehat{s}_0dh}\right).$$
(52)
This situation is different from the partonic invariant mass spectrum. In the present case, the lower kinematic limit of $`\widehat{s}_H`$ is $`ฯต^2`$, while the phase space boundary for virtual and real gluon emissions has been moved to $`\rho _ฯต=ฯต(1+ฯต)`$, above which only real gluon emissions contribute.A more general discussion for this kind of phenomenon in other observables can be found in Ref. . Hence there could potentially be two kinds of parametrically large logarithms $`\mathrm{log}(cฯต^2)`$ and $`\mathrm{log}(c\rho _ฯต)`$ appearing. However, as can be seen from Eq. (49), $`\widehat{s}_Hฯต^2`$ corresponds to vanishing $`\widehat{s}_0`$ and $`h`$, which is the dangerous region where the infra-red factorization fails and is kinematically suppressed by the tree level rate. More explicitly, $`\mathrm{log}(cฯต^2)`$ will be killed by the pre-factor $`2(cฯต^2)^2`$ so that the rate vanishes as $`cฯต^2`$. The only important parametrically large logarithms are those resulting from the incomplete cancellation between virtual and real corrections at the phase space boundary $`\rho _ฯต`$. From this argument it would seem that we only need to resum logs of the form $`\mathrm{log}(c\rho _ฯต)`$. However, in this particular case, the distance from $`\rho _ฯต`$ to $`ฯต^2`$ is only $`ฯต`$, which is expected to be a numerically small quantity at around $`0.08`$. If the experimental cut $`c`$ lies below $`\rho _ฯต`$, the partially integrated rate could still be sensitive to $`\mathrm{log}(cฯต^2)`$ due to the smallness of $`ฯต`$. We thus chose to resum all logs of the form $`\mathrm{log}(cฯต^2)`$ as well. Fortunately, since we have the resummed rate at the doubly differential level, this is not a problem. The resummed rate with hadronic mass cut $`c>\rho _ฯต`$ is given by
$`{\displaystyle \frac{1}{\mathrm{\Gamma }_0}}\mathrm{\Gamma }(c)_>=`$ $`{\displaystyle _0^1}๐h\mathrm{\hspace{0.17em}2}h^2(32h)\left\{\xi \left[{\displaystyle \frac{c\rho _ฯต}{h(1+ฯต)}}\right]\xi _{\alpha _s}\left[{\displaystyle \frac{c\rho _ฯต}{h(1+ฯต)}}\right]\right\}`$ (53)
$`+`$ $`{\displaystyle _{\frac{c\rho _ฯต}{1+ฯต}}^{(\sqrt{c}ฯต)^2}}๐\widehat{s}_0{\displaystyle _{2\sqrt{\widehat{s}_0}}^{\frac{c\widehat{s}_0ฯต^2}{ฯต}}}๐h\mathrm{\hspace{0.17em}2}h^2(32h)\left[\xi ^{}(\widehat{s}_0/h)\xi _{\alpha _s}^{}(\widehat{s}_0/h)\right]`$ (54)
$`+`$ $`{\displaystyle _0^{\frac{c\rho _ฯต}{1+ฯต}}}๐\widehat{s}_0{\displaystyle _{2\sqrt{\widehat{s}_0}}^{1+\widehat{s}_0}}๐h\gamma _{\alpha _s}(h,\widehat{s}_0)+{\displaystyle _{\frac{c\rho _ฯต}{1+ฯต}}^{(\sqrt{c}ฯต)^2}}๐\widehat{s}_0{\displaystyle _{2\sqrt{\widehat{s}_0}}^{\frac{c\widehat{s}_0ฯต^2}{ฯต}}}๐h\gamma _{\alpha _s}(h,\widehat{s}_0),`$ (55)
where
$$\xi ^{}(\widehat{s}_0/h)=\frac{d}{d\widehat{s}_0}\xi (\widehat{s}_0/h).$$
(56)
While the resummed rate with hadronic mass cut $`c<\rho _ฯต`$ is given by
$`{\displaystyle \frac{1}{\mathrm{\Gamma }_0}}\mathrm{\Gamma }(c)_<=`$ $`{\displaystyle _0^{\frac{cฯต^2}{ฯต}}}๐h\mathrm{\hspace{0.33em}2}h^2(32h)\left\{\xi \left[{\displaystyle \frac{(\sqrt{c}ฯต)^2}{h}}\right]\xi _{\alpha _s}\left[{\displaystyle \frac{(\sqrt{c}ฯต)^2}{h}}\right]\right\}`$ (57)
$`+`$ $`{\displaystyle _0^{\left(\sqrt{c}ฯต\right)^2}}๐\widehat{s}_0{\displaystyle _{2\sqrt{\widehat{s}_0}}^{1+\widehat{s}_0}}\gamma _{\alpha _s}(h,\widehat{s}_0).`$ (58)
Analytic expressions for the partially integrated rate at one loop level, last lines in Eq. (53) and Eq. (57), can be found in Ref. .
## V Results and Conclusions
In Fig. 3 we show the one loop, LL resummed (including $`g_1`$ only) and NLL resummed (including $`g_1`$ and $`g_2`$) results for $`\overline{\mathrm{\Lambda }}=0.39\text{GeV}`$. We see the for $`c0.18`$, the next to leading order approximation breaks down, as the next to leading order piece becomes just as large as the leading order piece. Notice that when $`c0.18`$, the effective cut on the partonic invariant mass is $`c_00.09`$, which from Fig. 2 we see is consistent with the breakdown of the resummed expansion. This is contrast with the result that an energy cut of $`2.1\mathrm{GeV}`$, on the rate for $`BX_s\gamma `$ does not necessitate resummation, as the argument of the logs in the case of the radiative decay is about $`0.12`$.
It is clear from Fig. 3 that resummation shifts the whole spectrum toward the high invariant mass region such that the number of events which lie below the cut $`c`$ is decreased. This occurs because the high invariant mass region with $`\widehat{s}_H>\rho _ฯต`$ is populated with real gluon emissions only.
In Fig. 4 we show the cut rate for several different values of $`\overline{\mathrm{\Lambda }}`$. It is clear that the cut rate can be very sensitive to the value of the unphysical parameter $`\overline{\mathrm{\Lambda }}`$. This occurs because the argument of the large logs is now $`c\rho _ฯต`$. The parameter $`\overline{\mathrm{\Lambda }}`$ is well defined at a given fixed order in perturbation theory and the values chosen in Fig. 3 are the mean value and one sigma values extracted in Ref. at one loop. Soon the errors in the extraction of $`\overline{\mathrm{\Lambda }}`$ will become less significant. In this paper we will not delve into the phenomenology of the extraction of $`V_{ub}`$, as the calculation we have discussed here has not included the important non-perturbative corrections coming from the Fermi motion. These corrections are parameterized in terms of a well defined structure function, $`f(y)`$ defined in Eq. (26), and as discussed in , can be very large. We will reexamine the issue of the structure function in a future publication, where following , we will eliminate the structure function from the cut rate prediction by utilizing the data from the end-point of radiative $`B`$ decays, which in turn encodes all the information contained in the structure function.
From Fig. 4 we see that, for large values of $`\overline{\mathrm{\Lambda }}`$, the resummation is not under control for phenomenologically interesting cuts and thus it may not be possible to extract $`V_{ub}`$ in this way. However, when modding out by the soft function using the $`bs\gamma `$ rate, there are cancellations which lead to a perturbative series which is better behaved. This is indeed what happened in the case for the electron energy spectrum. Thus, we refrain from drawing any conclusions regarding the viability of extracting $`V_{ub}`$ from the invariant mass spectrum at this time.
The purpose of this paper was to determine the effect of threshold resummation on the rate for semi-leptonic $`B`$ decays with a cut on the hadronic invariant mass. We first showed how the rate factorizes when written in terms of hadronic variables, generalizing the results of . Using this factorization, we resummed the cut rate at next to leading order in the infrared logs and found that, for cuts of interest, the resummation is crucial, and that for $`c<0.18`$, even the next to leading order resummed rate is no longer reliable. However, this breakdown point depends on the value of $`\overline{\mathrm{\Lambda }}`$, and becomes smaller as $`\overline{\mathrm{\Lambda }}`$ is decreased. A more phenomenological analysis, including the effects of the structure function responsible for the Fermi motion, is forthcoming.
###### Acknowledgements.
This work was supported in part by the Department of Energy under grant number DOE-ER-40682-143. I. L. would like to thank the hospitality of the National Center for Theoretical Sciences at National Tsing Hua University in Taiwan where part of this work was completed. |
warning/0001/cond-mat0001239.html | ar5iv | text | # Effect of Electron Correlation on the Bragg Reflection
## Abstract
We study the effect of correlation on the Bragg reflection in the 3D electron gas, the 1D Luttinger liquid, and the 1D Hubbard model in an alternating periodic potential at half-filling. In the last system, we suggest a Luttinger-liquid-type quasi-metallic state in the crossover region from the band insulator to the Mott insulator. We explain the appearance of this state in terms of the incompatibility of the Bragg reflection with the concept of Luttinger liquids.
One of the fundamental issues in solid state physics is to elucidate to what extent the concept of โelectronic band structureโ is relevant in a strongly-correlated system. In the free-electron gas, the band gap is formed by the quantum interference between an incident electron plane wave, $`e^{i๐ค๐ซ}`$, with a reflected one, $`e^{i(๐ค+๐)๐ซ}`$, where $`๐`$ is a reciprocal-lattice vector related to periodicity of $`V`$ an electron-lattice potential. In this sense, the quantum coherence leading to the Bragg reflection is a key to the band structure and thus we consider it quite important to study the correlation effect on the Bragg reflection.
The study is conceptually simple in the Fermi liquid as represented by the three-dimensional electron gas (3DEG) at metallic densities. The point is that we should grasp the band-gap formation in terms of the Bragg reflection of a quasi-particle rather than a free electron, because quasi-particles or wave packets composed of a complicated combination of plane waves due to $`U`$ the electron-electron interaction manifest themselves in low-energy physics. In the one-dimensional Luttinger liquid (1DLL), however, the situation is not so simple ; we should ask even the very existence of the Bragg reflection and this constitutes one of the aims of this paper.
Basically there exist two complementary approaches to the many-electron system in a crystal described by the Hamiltonian $`H`$ composed of $`T`$ the kinetic energy, $`V`$, and $`U`$. One is โthe band approachโ or $`(T+V)+U`$ in which the problem is reduced to the self-consistent determination of an effective one-body potential $`\stackrel{~}{V}`$ by combining $`V`$ with the effect of $`U`$ in the Hartree-Fock-like mean-field approximation. If desired, the correlation effect due to $`U`$ (which is missed in $`\stackrel{~}{V}`$) can be included by perturbation in $`U`$ with respect to the unpertubed part $`T+\stackrel{~}{V}`$. Another is โthe correlated-electron approachโ or $`(T+U)+V`$ in which we consider a correlated-electron state defined by $`T+U`$ first and then include $`V`$ perturbatively. Note that these two approaches do not always provide the same conclusion, as the discussion on the Mott transition indicates. We can even imagine situations in which the competition between $`V`$ and $`U`$ brings about a state which neither approach describes well.
In this paper, an example of such intriguing situations is shown on the basis of our finding that the Bragg reflection is usually incompatible with the concept of Luttinger liquids. More specifically we shall treat the 1D Hubbard model with an alternating periodic potential at half-filling, a system attracting much attention in relation to the neutral-ionic transition , the ferroelectric perovskites , and the crossover from the band insulator (BI) at $`VU`$ to the Mott insulator (MI) at $`VU`$ . We have made a study in the density-matrix renormalization group (DMRG) to obtain the charge and spin gaps, $`\mathrm{\Delta }_c`$ and $`\mathrm{\Delta }_s`$, as well as the electron localization length $`\lambda `$ with the system size up to 400 sites (which is larger than any previous calculations by an order of magnitude). By comparing these exact results with approximate ones given in both the band and correlated-electron approaches and also by referring to the very recent result of Fabrizio et al. , we suggest the appearance of a Luttinger-liquid-type quasi-metallic state at the BI-MI crossover. Here by โquasi-metallicโ we mean that it is not the same as โmetallicโ because of nonzero $`\mathrm{\Delta }_c`$, but the state is distinct from either BI or MI by such features as very small $`\mathrm{\Delta }_c`$ and long $`\lambda `$.
For better illustration of the Bragg reflection of a quasi-particle, let us start with 3DEG in a weak periodic potential $`V`$ for which $`H`$ is given in second quantization with the plane-wave basis as
$`H`$ $`=`$ $`T+V+U={\displaystyle \underset{๐ค\sigma }{}}\epsilon _๐คc_{๐ค\sigma }^+c_{๐ค\sigma }+{\displaystyle \underset{๐\mathrm{๐}}{}}{\displaystyle \underset{๐ค\sigma }{}}V(๐)c_{๐ค\sigma }^+c_{๐ค+๐\sigma }`$ (2)
$`+{\displaystyle \frac{1}{2}}{\displaystyle \underset{๐ช\mathrm{๐}}{}}{\displaystyle \underset{๐ค\sigma }{}}{\displaystyle \underset{๐ค^{}\sigma ^{}}{}}U(๐ช)c_{๐ค+๐ช\sigma }^+c_{๐ค^{}๐ช\sigma ^{}}^+c_{๐ค^{}\sigma ^{}}c_{๐ค\sigma },`$
where $`c_{๐ค\sigma }`$ annihilates an electron specified by momentum $`๐ค`$ and spin $`\sigma `$, $`\epsilon _๐ค=๐ค^2/2m\mu `$ with $`m`$ the mass of a free electron and $`\mu `$ the chemical potential, $`V(๐)`$ the local electron-ion pseudopotential, and $`U(๐ช)=4\pi e^2/๐ช^2`$.
Single-electron properties can be analyzed by the study of the thermal Greenโs function $`G_{๐ค,๐ค+๐}(i\omega _n)`$ with $`\omega _n`$ a fermion Matsubara frequency, defined conventionally as
$`G_{๐ค,๐ค+๐}(i\omega _n){\displaystyle _0^{1/T}}๐\tau T_\tau c_{๐ค\sigma }(\tau )c_{๐ค+๐\sigma }^+e^{i\omega _n\tau }.`$ (3)
In the correlated-electron approach, we consider 3DEG without $`V`$ in the first step. Here $`G_{๐ค,๐ค+๐}(i\omega _n)`$ is not zero only for $`๐=\mathrm{๐}`$. Thus we simply write $`G_๐ค^{\mathrm{EG}}(i\omega _n)`$ and this function can be determined by the formally exact Dyson equation as shown diagrammatically in Fig. 1(a) with use of the vertex function $`\mathrm{\Lambda }(๐ค^{}i\omega _n^{},๐คi\omega _n)`$ defined in Fig. 1(c). In the second step, we include $`V`$ in its lowest order by solving the equation in Fig. 1(b) to obtain $`G_{๐ค,๐ค+๐}(i\omega _n)`$ in terms of $`G_๐ค^{\mathrm{EG}}(i\omega _n)`$ and $`\mathrm{\Lambda }(๐ค^{}i\omega _n^{},๐คi\omega _n)`$.
In order to make the physics as clear as possible, we shall be concerned only with the most interesting case in which $`|๐|`$ is equal to $`2k_\mathrm{F}`$ with $`k_\mathrm{F}`$ the Fermi wave number of 3DEG. Then, a band gap $`\mathrm{\Delta }`$ opens at the Fermi level with the value of $`2V(๐)`$ in the noninteracting electron gas, while in the interacting system, by expanding the self-energy $`\mathrm{\Sigma }_๐ค^{\mathrm{EG}}(i\omega _n)`$ $`[=i\omega _n\epsilon _๐คG_๐ค^{\mathrm{EG}}(i\omega _n)^1]`$ with respect to $`\omega _n`$ up to first order in line with weak $`V`$, we find easily that $`\mathrm{\Delta }`$ is given exactly as $`2z_{k_\mathrm{F}}\mathrm{\Lambda }(k_\mathrm{F}0,k_\mathrm{F}0)V(๐)`$ with $`G_{k_\mathrm{F},k_\mathrm{F}}(i\omega _n)=z_{k_\mathrm{F}}\mathrm{\Delta }/2[(i\omega _n)^2(\mathrm{\Delta }/2)^2]`$ where $`z_{k_\mathrm{F}}`$ is the quasi-particle renormalization factor at the Fermi surface. Thus the ratio, $`z_{k_\mathrm{F}}\mathrm{\Lambda }(k_\mathrm{F}0,k_\mathrm{F}0)`$, quantifies the effect of correlation on the Bragg reflection.
We estimate this ratio quantitatively by employing the local-field correction $`G_+(q)`$ to represent the effect of the vertex function. Using the values of $`G_+(q)`$ as supplied by quantum Monte Carlo , we can calculate the ratio, together with its components, $`z_{k_\mathrm{F}}`$ and $`\mathrm{\Lambda }(k_\mathrm{F}0,k_\mathrm{F}0)=\{1+\alpha r_s[1G_+(2k_\mathrm{F})]/2\pi \}^1`$, as a function of $`r_s`$ the interelectron distance in units of the Bohr radius with $`\alpha =(4/9\pi )^{1/3}0.521`$. The results are shown in Fig. 1(d) in which we see that $`\mathrm{\Lambda }(k_\mathrm{F}0,k_\mathrm{F}0)1`$.
Deviation of the vertex function from unity accounts for the many-body effects on the electron-ion interaction $`V_{\mathrm{el}\mathrm{ion}}`$ and physically this should be small at metallic densities ($`1<r_s<6`$) for $`q=2k_\mathrm{F}`$ corresponding to the interelectron distance in real space; $`V_{\mathrm{el}\mathrm{ion}}`$ is not modified much from the bare one in the neighborhood of an electron at this distance due to the absence of other electrons by correlation. Thus the ratio is essentially determined only by $`z_{k_\mathrm{F}}`$ or the weight of the coherent part, confirming a naively anticipated result that the Bragg reflection occurs only in the coherent part of a quasi-particle.
The importance of $`z_{k_\mathrm{F}}`$ in the Bragg reflection prompts us to investigate 1DLL in which $`z_{k_\mathrm{F}}`$ vanishes. The Hamiltonian is basically the 1D version of Eq. (2); $`\epsilon _k`$ is linearized as $`\epsilon _k=v_\mathrm{F}(kk_\mathrm{F})[v_\mathrm{F}(kk_\mathrm{F})]`$ for the right- \[left-\]moving branch in $`T`$ and coupling constants, $`g_1`$, $`g_2`$, and $`g_4`$, corresponding to the backward scattering, the forward scattering between the opposite branches, and the forward one within the same branch, respectively, are introduced in $`U`$ . Using $`a_{k\sigma }[b_{k\sigma }]`$ the annihilation operator for an electron in the right- \[left-\]moving branch, we can write $`V`$ as
$`V=v{\displaystyle \underset{k\sigma }{}}(a_{k\sigma }^+b_{k2k_\mathrm{F}\sigma }+b_{k2k_\mathrm{F}\sigma }^+a_{k\sigma }),`$ (4)
in which only the $`K=2k_\mathrm{F}`$ part for $`V(K)`$ is retained.
The assertion we shall make is that $`\mathrm{\Delta }_c`$ due to $`V`$ corresponding to $`\mathrm{\Delta }`$ in the Fermi liquid vanishes in 1DLL. Our strategy to prove it is to find a criterion as to when $`V`$ turns out to be an irrelevant perturbation. Since the key quantity is $`a_{k_\mathrm{F}\sigma }^+b_{k_\mathrm{F}\sigma }`$ or $`T_{\omega _n}G_{k_\mathrm{F},k_\mathrm{F}}(i\omega _n)`$, we evaluate the expectation value by treating $`V`$ as a linear perturbation to the system decribed by the Hamiltonian $`T+U`$. Then, the Kuboโs formula provides us
$`a_{k_\mathrm{F}\sigma }^+b_{k_\mathrm{F}\sigma }={\displaystyle \frac{v}{2}}\underset{\omega 0}{lim}N(2k_\mathrm{F},\omega ),`$ (5)
where $`N(q,\omega )`$ is the charge-density response function. Thus the problem is reduced to evaluating $`N(2k_\mathrm{F},\omega )`$ in the $`\omega 0`$ limit for the system of $`T+U`$ . The behavior of this function is known well and the result is $`N(2k_\mathrm{F},\omega 0)\omega ^{\gamma _\rho 1}`$ with $`\gamma _\rho `$, given by
$`\gamma _\rho =\sqrt{{\displaystyle \frac{1+(g_4+g_12g_2)/2\pi v_\mathrm{F}}{1+(g_4g_1+2g_2)/2\pi v_\mathrm{F}}}},`$ (6)
for $`g_10`$. This leads us to conclude that $`a_{k_\mathrm{F}\sigma }^+b_{k_\mathrm{F}\sigma }`$ vanishes for $`g_1>2g_2`$ even in the presence of $`V`$. More generally, the expectation value is zero in the phases characterized by $`N(2k_\mathrm{F},\omega 0)=0`$, indicating the irrelevance of $`V`$. This irrelevance implies that $`\mathrm{\Delta }_c`$ due to $`V`$ vanishes, because the system is the same as that with $`v=0`$.
Physically low-lying excitations in 1DLL are rigorously represented by sound waves with wavelengths much longer than $`1/2k_\mathrm{F}`$. Thus the effect of $`V`$ manifests itself after the average of $`V`$ over a distance longer than its periodicity, which is null and leads to the complete absence of the Bragg reflection. This statement ceases to be valid if $`V`$ is large enough to destroy the Luttinger-liquid state itself. In this sense, the concept of the Bragg reflection is incompatible with that of Luttinger liquids.
So far no lattice periodicity is considered in $`T+U`$ and thus the concept of electron filling is irrelevant. Now we include it by treating a 1D system at half-filling on the lattice prescribed by $`T+U`$ with $`V`$ possessing periodicity of two lattice units. In site representation, $`H`$ is given by
$`H`$ $`=`$ $`T+V+U=t{\displaystyle \underset{j\sigma }{}}(c_{j\sigma }^+c_{j+1\sigma }+c_{j+1\sigma }^+c_{j\sigma })`$ (8)
$`+v{\displaystyle \underset{j\sigma }{}}(1)^jc_{j\sigma }^+c_{j\sigma }+u{\displaystyle \underset{j}{}}c_j^+c_jc_j^+c_j.`$
For the study of competition between $`V`$ and $`U`$ in the whole region from $`vu`$ to $`vu`$, we implement DMRG to calculate $`E(N_{},N_{})`$ the ground-state energy with $`N_\sigma `$ the fixed number of $`\sigma `$-spin electrons in the $`L`$-site system under the open-boundary condition. At size $`L`$, the charge and spin gaps, $`\mathrm{\Delta }_c(L)`$ and $`\mathrm{\Delta }_s(L)`$, are given as
$`\mathrm{\Delta }_c(L)`$ $`=`$ $`E({\displaystyle \frac{L}{2}}+1,{\displaystyle \frac{L}{2}})+E({\displaystyle \frac{L}{2}}1,{\displaystyle \frac{L}{2}})2E({\displaystyle \frac{L}{2}},{\displaystyle \frac{L}{2}}),`$ (9)
$`\mathrm{\Delta }_s(L)`$ $`=`$ $`E({\displaystyle \frac{L}{2}}+1,{\displaystyle \frac{L}{2}}1)E({\displaystyle \frac{L}{2}},{\displaystyle \frac{L}{2}}).`$ (10)
By using a finite-size scaling as $`\mathrm{\Delta }_i(L)=(\mathrm{\Delta }_i^2+A_i/L^2+B_i/L^3+\mathrm{})^{1/2}`$ for $`i=c`$ or $`s`$ , we extrapolate the data at $`L=50`$, 100, 200, and 400 to obtain the values at $`L=\mathrm{}`$, $`\mathrm{\Delta }_c`$ and $`\mathrm{\Delta }_s`$. In Fig. 2, the results thus obtained are plotted as a function of $`u`$ at $`v=0.5t`$. As $`u`$ increases, both $`\mathrm{\Delta }_c`$ and $`\mathrm{\Delta }_s`$ decrease from $`2v`$ (the band gap at $`u=0`$) and become very small at around $`u=2.6t`$. With the further increase of $`u`$, $`\mathrm{\Delta }_c`$ increases very rapidly, while $`\mathrm{\Delta }_s`$ remains to be zero. In the inset of Fig. 2, the gaps near $`u=2.6t`$ are shown in detail. The same overall behavior of the gaps is seen for other values of $`v`$ of the order of $`t`$.
Let us analyze these results by the comparison with those in both band and correlated-electron approaches. The former approach begins with the Hartree-Fock (HF) approximation which amounts to the one-body problem described by $`T+\stackrel{~}{V}`$ where we define $`\stackrel{~}{V}`$ in the same form of $`V`$ in Eq. (8) by replacing $`v`$ into $`\stackrel{~}{v}`$, determined through
$`\stackrel{~}{v}=vu{\displaystyle _{\frac{\pi }{2}}^{\frac{\pi }{2}}}{\displaystyle \frac{dk}{2\pi }}{\displaystyle \frac{\stackrel{~}{v}}{\sqrt{\stackrel{~}{v}^2+4t^2\mathrm{cos}^2k}}}.`$ (11)
In HF we obtain the band insulator (BI) brought about by the full Bragg reflection with $`\mathrm{\Delta }_c=\mathrm{\Delta }_s=2\stackrel{~}{v}`$. Since $`\stackrel{~}{v}`$ is always positive, these gaps never vanish, which clearly contradicts the exact results for large $`u`$. Even for $`u`$ as small as $`t`$, they do not agree well with the exact ones, as seen by the dotted-dashed curve in Fig. 2.
However, we achieve a surprisingly good improvement by including the correlation effect in second-order perturbation in $`U`$ with the unperturbed basis in $`T+\stackrel{~}{V}`$, as shown by the dashed curve in the inset of Fig. 2. In fact, the exact gaps are reproduced quite accurately in this $`(T+V)+U`$ approach for $`u`$ from $`0`$ up to about $`u_{c0}2.45t`$. Note that the correlation effect included in this way does not separate $`\mathrm{\Delta }_c`$ from $`\mathrm{\Delta }_s`$. Physical mechanism to reduce the gaps from those in HF is the same as explained in 3DEG, namely, the reduction of the the Bragg reflection on the conversion from a free electron to a quasi-particle. Thus we conclude that the state realized in the system for $`u<u_{c0}`$ is the correlated BI.
For $`u`$ larger than about $`u_{c2}2.90t`$, on the other hand, $`\mathrm{\Delta }_s`$ vanishes, while $`\mathrm{\Delta }_c`$ does not, indicating that the state is the MI. The effect of $`v`$ can be included in $`\mathrm{\Delta }_c`$ in the $`(T+U)+V`$ approach; by examining the exact solution in $`T+U`$ , we can deduce an expansion for $`\mathrm{\Delta }_c`$ in $`u^1`$ up to third order as
$`\mathrm{\Delta }_cu2\sqrt{v^2+4t^2}+8\mathrm{ln}2{\displaystyle \frac{t^2}{u}}{\displaystyle \frac{u^2+4v^2}{u^2}}6\zeta (3){\displaystyle \frac{t^4}{u^3}},`$ (12)
with $`\zeta (3)1.202`$. The above result is plotted by the dotted curve in Fig. 2 in which we see that the exact result is reproduced quite well for $`u`$ larger than $`5t`$.
Now we need to clarify the nature of the state for $`u`$ at the BI-MI crossover, ranging from $`u_{c0}`$ to $`u_{c2}`$. For that purpose, we calculate $`D_L`$ a dimensionless localization parameter introduced by Resta and Sorella as
$`D_L=L\mathrm{ln}\left|\mathrm{\Psi }|\mathrm{exp}\left(i{\displaystyle \frac{2\pi }{L}}{\displaystyle \underset{j}{}}x_j\right)|\mathrm{\Psi }\right|^2,`$ (13)
with $`\mathrm{\Psi }`$ the ground-state wavefunction at $`N_{}=N_{}=L/2`$ under the open-boundary condition for the $`L`$-site system and $`x_j`$ the position operator at site $`j`$. Extrapolation of $`D_L`$ to the $`L\mathrm{}`$ limit gives the value $`D`$ which is related to $`\lambda `$ through $`\lambda =\sqrt{D}/2\pi `$ in units of the lattice spacing.
The calculated results for both $`D`$ and $`D_L`$ at various $`L`$โs are shown as a function of $`u`$ in Fig. 3(a). For either $`u<u_{c0}`$ or $`u>u_{c2}`$, we see that $`D_L`$ converges to $`D`$ at $`L`$ as small as 100, while for $`u`$ in-between, even $`L=400`$ is not large enough for the convergence. In particular, $`D`$ seems to diverge for $`u`$ around $`u_{c1}2.65t`$. Divergence in $`D`$, or equivalently that in $`\lambda `$, implies the appearance of a metallic state, indicating the vanishment of $`\mathrm{\Delta }_c`$.
As for $`\mathrm{\Delta }_c=0`$ at $`u=u_{c1}`$ and the state for $`u_{c1}<u<u_{c2}`$, Fabrizio et al. have suggested a spontaneously dimerized insulating phase (SDI) by analytical arguments. In fact we find that $`\lambda `$ decreases quite rapidly to be less than twice the lattice spacing as $`u`$ increases from $`u_{c1}`$, implying an insulating behavior. Thus we conclude that an insulating phase, most likely SDI, is realized for $`u_{c1}<u<u_{c2}`$.
A remaining problem is that $`\mathrm{\Delta }_c`$ never becomes zero in Fig. 2, although we now know that it should be zero at least at $`u=u_{c1}`$. Thus we reexamine the size dependence of both $`\mathrm{\Delta }_c(L)`$ and $`\mathrm{\Delta }_s(L)`$ carefully and the typical results are plotted in Fig. 3(b). Let us first analyze them in terms of $`\mathrm{\Delta }_c\mathrm{\Delta }_s`$ or the spin-charge separation. Because of one dimensionality, one might assume that $`\mathrm{\Delta }_c\mathrm{\Delta }_s`$ should be the case as long as $`u0`$, but this is not true; at $`u=2t`$ in the BI region, we find that both $`\mathrm{\Delta }_c`$ and $`\mathrm{\Delta }_s`$ coincide up to at least five digits (which exceeds numerical accuracy) at $`L=200`$ or larger. On the other hand, we know that $`\mathrm{\Delta }_c\mathrm{\Delta }_s=\mathrm{\Delta }_s0`$ at $`u=u_{c1}`$. Therefore there definitely exists a value of $`u`$ at which $`\mathrm{\Delta }_c\mathrm{\Delta }_s`$ begins to deviate from zero. We identify $`u_{c0}`$ as such a value and thus only for $`u`$ larger than $`u_{c0}`$ the 1DLL-like spin-charge separation occurs. In this sense, we consider that $`u_{c0}`$ gives a sharp phase boundary.
In order to find more detailed features of the state at $`u`$ from $`u_{c0}`$ to $`u_{c1}`$, let us look at Fig. 3 again. As represented at $`u=3t`$ in Fig. 3(b), $`\mathrm{\Delta }_s(L)`$ converges to zero very nicely with the increase of $`L`$ in MI. Similar convergence of both charge and spin gaps is obtained in SDI as well. However, a distinct behavior is seen in the gaps for $`u_{c0}<u<u_{c1}`$ as illustrated at $`u=2.6t`$; both $`\mathrm{\Delta }_c(L)`$ and $`\mathrm{\Delta }_s(L)`$ at $`L=400`$ seem to be larger than those extrapolated from the data at smaller $`L`$, implying that our data for the gaps are not accurate enough for these $`u`$โs. This inaccuracy should be due to the large $`\lambda `$ which is always longer than the two lattice units for these $`u`$โs as indicated in Fig. 3(a). This difficulty in obtaining exact values for the gaps in this region can be overcome only by a more accurate calculation of energies at much larger $`L`$. Such a calculation is not feasible at present, but we can safely conclude even at the present time that $`\mathrm{\Delta }_c`$ is very small, i.e., much smaller than $`0.1t`$ at most of the values of $`u`$ in this phase. The nature of small $`\mathrm{\Delta }_c`$ and long $`\lambda `$ suggests us that a Luttinger-liquid-type quasi-metallic phase appears for $`u_{c0}<u<u_{c1}`$. Incidentally, each electron in this phase feels $`V`$ with the spatial average over $`\lambda `$ which is longer than the periodicity of $`V`$. This implies that the effect of $`V`$ on the electrons is very small and thus this should be the reason why the feature of BI is lost in the state for $`u_{c0}<u<u_{c1}`$. Here again we find that the Bragg reflection, a crucial concept to define BI, is incompatible with the Luttinger-liquid feature.
Finally we note that both this phase and SDI deserve special attention, because they are not anticipated in both band and correlated-electron approaches; their existence is entirely due to the competition of $`V`$ and $`U`$.
In conclusion, we have investigated the effect of electron correlation on the Bragg reflection in a variety of situations in various approaches. We have found the incompatibility of the Bragg reflection with the Luttinger liquid, based on which we have suggested a Luttinger-liquid-type quasi-metallic state at the crossover from BI to MI via SDI.
Y.T. is supported by the Grant-in-Aid for Scientific Research (C) from the Ministry of Education, Science, Sports, and Culture of Japan. |
warning/0001/nucl-th0001022.html | ar5iv | text | # Dressing the Nucleon in a Dispersion Approach
## I Introduction
The properties of unitarity and analyticity have been exploited in various theoretical approaches to pion-nucleon scattering. Unitarity is usually implemented either by solving a relativistic wave equation for the scattering amplitude or using a K-matrix formalism . Analytic properties, which are related to the condition of causality, have allowed one to derive useful dispersion relations for various amplitudes .
In the present work we have developed an approach in which constraints due to both unitarity and analyticity are incorporated by using both a K-matrix method and dispersion relations . The approch used consists of two separate stages. In the first effective 2- and 3-point Greenโs functions (i.e. propagators and vertices) are built which incorporate non-perturbative dressing due to regular (non-pole) parts of loop diagrams. In the second stage a K-matrix formalism is used to calculate the T-matrix, where the kernel, the K-matrix, is constructed from tree-level diagrams using the dressed vertices and propagators calculated in the first stage. Through the use of the K-matrix formalism the pole contributions are taken into account which were left out in the first stage. The T-matrix obtained from thus constructed K-matrix will contain the principal value parts of loop integrals, which is important for implementing analyticity in the K-matrix framework. The model as presented is geared to the calculation of pion-nucleon scattering and includes a consistent dressing of interacting nucleons, expanding the method of Ref. . Since the dressing is formulated in terms of effective vertices and propagators through the use of form factors and self-energies, a broader application might be possible.
In the calculation of the propagators and vertices a technique based on the use of dispersion relations is used. This allows us to arrange the dressing of the nucleon such that the only $`\pi NN`$ vertices needed throughout the procedure are those with one virtual external nucleon line (half-off-shell vertices). For the construction of the K-matrix we also need only these $`\pi NN`$ vertices. Being interacting 3-point Greenโs functions, such vertices are not measurable quantities. In particular, they depend on the representation of interpolating fields in the Lagrangian. A wide class of field transformations leave the S-matrix (and therefore the observables) invariant while changing the interaction Lagrangian and hence the vertices . In this work all calculations are performed in a particular representation in which self-energies and vertex modifications are calculated consistently with the K-matrix. In particular, 4- and higher-point Greenโs functions are absent in this representation. In Section V we show that the field ambiguity can be taken advantage of to change to a different representation in which the nucleon self-energy vanishes and no 4- (or higher-) point vertices are introduced. While leading to the same observables in virtue of the equivalence theorem , this representation is convenient for interpreting the effects of nucleon dressing in terms of effective $`\pi NN`$ vertices. Rather soft form factors in 3-point vertices are dynamically generated through the dressing.
In the dressing it is essential to include the low-lying meson and baryon degrees of freedom, in particular, beside the pion, the $`\rho `$โ and $`\sigma `$โ mesons and the $`\mathrm{\Delta }`$โresonance. The associated coupling parameters were fixed by considering phase shifts for pion-nucleon scattering. A good agreement could be obtained for pion energies exceeding 400 MeV, adjusting only 5 parameters. The requirement that the dressing procedure converges for a given bare $`\pi NN`$ form factor puts additional constraints on the allowed range of these parameters.
In discussing the calculated phase shifts, we focus primarily on effects of the dressing. These can be regarded as effects of the the explicit inclusion of the principal value parts of loop integrals (which are omitted in the usual approximation for the K-matrix ). Effects of principal value parts were also considered in Ref. in a calculation based on three-dimentional reductions of the Bethe-Salpeter equation. In our approach we have kept the most general Lorentz structure for the $`\pi NN`$ vertex. Another difference with is that we obtain the principal value parts through the use of dispersion integrals.
A general description of the model is given in Section II. Details of the calculation of the main building blocks of the K-matrix โ dressed vertices and propagators โ are given in Sections III and IV, respectively. In Section V we construct a representation which is convenient for interpreting results of the dressing. Effects of the dressing on calculated phase shifts for pion-nucleon scattering are discussed in Section VI. Concluding remarks are made in Section VII.
## II Outline of the model
We start by giving the main formulae of the K-matrix approach in the context of pion-nucleon scattering. The S-matrix S is expressed in terms of the scattering amplitude $`๐ฏ`$ (the T-matrix) by
$$S=1+2i๐ฏ.$$
(1)
In principle, the T-matrix can be obtained by solving the Bethe-Salpeter equation,
$$๐ฏ=V+V๐ข๐ฏ,$$
(2)
where the kernel (potential) $`V`$ is the sum of irreducible diagrams describing the scattering, and $`๐ข`$ is a dressed $`\pi N`$ propagator.
In general, any integral over the 4-momenta in $`๐ข`$ can be split into its pole and principal value (regular) parts,
$$๐ข=i\delta +๐ข^R,$$
(3)
where the pole part $`i\delta `$ is the contribution of the real (on-mass-shell) $`\pi N`$ state, and $`๐ข^R`$ corresponds to the propagation of virtual (off-shell) nucleon and pion. According to Cutkosky rules , $`i\delta `$ contains the imaginary parts of the invariant functions which parametrize $`๐ข`$, and $`๐ข^R`$ contains the real parts.
It is this separation of the pole and principal value parts of propagators that is exploited in the K-matrix formalism. Namely, on defining the K-matrix by the equation
$$K=V+V๐ข^RK,$$
(4)
Eq. (2) can be written in the form
$$๐ฏ=K+Ki\delta ๐ฏ.$$
(5)
This can be formally solved, yielding the central equation of the K-matrix method :
$$๐ฏ=\frac{1}{1Ki\delta }K.$$
(6)
The S-matrix can be obtained in a two step approach: 1) given a hermitian potential $`V`$, calculate $`K`$ according to Eq. (4), and 2) solve Eq. (5) to calculate the amplitude $`๐ฏ`$ and use Eq. (1) to calculate the S-matrix. This scheme is equivalent to solving the Bethe-Salpeter equation Eq. (2) and as such should provide the full unitary and analytic S-matrix. Given a hermitian K-matrix, Eq. (5) can be solved relatively easily (for instance, by expanding $`K`$ in partial waves and using Eq. (6)), and a unitary S-matrix is obtained. The simplicity of Eq. (5) is due to the fact that it involves integrals only over the mass-shells of internal particles. The problem of solving Eq. (4) is harder since one has to integrate over off-shell 4-momenta of $`๐ข^R`$. For this reason one usually avoids solving Eq. (4) in K-matrix models, setting $`K=V`$ . The main drawback of this approximation is that the principal value parts of the integrals are completely ignored, i. e. $`๐ข^R=0`$, and therefore analyticity of the amplitude cannot be fulfilled.
We take the potential $`V`$ as a sum of tree diagrams. These include the s- and u-channel diagrams with an intermediate nucleon and $`\mathrm{\Delta }`$-resonance plus the t-channel diagrams with an intermediate $`\rho `$\- and $`\sigma `$-meson. The tree diagrams are calculated using the free propagators and bare vertices. Note that we require that 4- and higher-point vertices do not participate in the construction of this potential. In other words, the bare interaction Lagrangian is assumed to contain 3-point vertices only. According to Eq. (4), the K-matrix should be obtained by dressing $`V`$ with the principal value parts of loop integrals. We construct the K-matrix as the sum of skeleton diagrams shown in Fig. (1). It has the same form as $`V`$, except that it contains dressed $`\pi NN`$ vertices and nucleon propagators and dressed propagators of the $`\mathrm{\Delta }`$, $`\rho `$ and $`\sigma `$.
Once the K-matrix is constructed, the T-matrix is calculated from Eq. (6) using a partial wave decomposition as in Refs. . The form of the K-matrix in Fig. (1) implies that physical one-nucleon โ one-pion intermediate states are explicitly included in the unitarization procedure. Thus, the S-matrix obtained from Eq. (1) obeys $`1N1\pi `$ unitarity exactly.
### A Dressing procedure
The calculation of the dressed $`\pi NN`$ vertex and the nucleon propagator is based on a system of integral equations, shown diagrammatically in Fig. (2). In the equation for the vertex, the external on-shell lines for the outgoing nucleon (on the right) and the pion, as well as the off-shell line for the incoming nucleon, are stripped away, as indicated by dashes on these lines. The solution is obtained in an iteration procedure and is described in detail in Ref. (for the case including the nucleon and pion only). Here we shall repeat the main points.
Every iteration step (say, step number $`n`$) proceeds as follows. The imaginary or pole contributions of the loop integrals are obtained by applying cutting rules to both the propagators and the vertices. Since the outgoing nucleon and the pion are on-shell, the only kinematically allowed cuts for the vertex loops are those shown by the curved lines in Fig. (2). In calculating these pole contributions, we retain only real parts of the loop integrals for vertices and self-energies from the previous step $`n1`$, as dictated by Eq. (4).
The real parts of the form factors and self-energy functions are calculated at every iteration step by applying dispersion relations to the imaginary parts. For example, for the form factors at the iteration step $`n`$ we have
$$ReG_i^n(p^2)=G_i^0(p^2)+\frac{๐ซ}{\pi }_{(M+\mu )^2}^{\mathrm{}}๐p^2\frac{ImG_i^n(p^2)}{p^2p^2},$$
(7)
where $`i`$ labels the structure of the vertex, pseudoscalar ($`S`$) or pseudovector ($`V`$), see Eq. (8). $`M`$ and $`\mu `$ are the nucleon and pion masses. $`G_i^0(p^2)`$ are the form factors in a bare $`\pi NN`$ vertex, the first term on the right-hand side of Fig. (2).
This procedure is repeated until a converged solution is reached. We use a normalized root-mean-square difference $`d_n`$ between two subsequent iteration steps $`n`$ and $`n+1`$ for the form factors and self-energy functions. The convergence criterion is that $`d_n<10^4`$ for at least a hundred iteration steps. As zeroth iteration step the bare $`\pi NN`$ vertex and the free nucleon propagator are taken. The solution of the equations in Fig. (2) is equivalent to a dressing of the potential $`V`$ with the principal value parts of loop integrals.
Despite the explicit use of dispersion integrals, analyticity in the model is obeyed only โapproximatelyโ. The violation of analyticity comes in at the level of bare form factors needed as part of the regularization of dispersion integrals. Strictly speaking, singularities of a bare form factor should give rise to additional residue contributions to the dispersion relation. To evaluate such a residue, one would have to know the behaviour of the function to which the dispersion relation is applied at the singularity of the bare form factor. In general however, this behaviour is not known. One way to circumvent this difficulty is to choose the bare form factor with singularities that are as remote from the region of physical interest as possible, implying a large width in general. This is supported by the fact that the width of the form factor should be larger than the masses of mesons included explicitly.
We remark that โ as a consequence of Liouvilleโs theorem โ the difficulty with additional singularities in the complex plane is inherent in any approach where phenomenological form factors are used.
## III Vertices
### A The $`\pi NN`$ vertex
The general $`\pi NN`$ vertex can be parametrized in terms of four Lorentz-invariant functions (form factors) . The form factors may depend on the 4-momenta squared of each of the three external lines. For the K-matrix in this model we need $`\pi NN`$ vertices with one virtual nucleon (4-momentum $`p`$), while the other nucleon ($`p^{}`$) and the pion are on the mass-shell. Such vertices are conventionally called โhalf-off-shellโ vertices. The general Lorentz and isospin covariant form of this vertex can be written <sup>*</sup><sup>*</sup>*Here and throughout the paper, we use the notation of Ref. .
$$\tau _\alpha \mathrm{\Gamma }(p)=\tau _\alpha P_+(p^{})\gamma ^5\left[G_S(p^2)+P_+(p)G_V(p^2)\right],$$
(8)
for an incoming virtual nucleon. Here $`G_S(p^2)`$ and $`G_V(p^2)`$ are pseudoscalar and pseudovector form factors, $`\tau _\alpha ,\alpha =1,2,3`$, are Pauli isospin matrices, and $`P_+(p)(p/+M)/(2M)`$. In the course of the dressing procedure the most general structure Eq. (8) of the $`\pi NN`$ vertex is maintained.
The bare vertex in the dressing procedure is chosen as
$$G_V(p^2)=f_N(1\chi )G^0(p^2),G_S(p^2)=f_N\chi G^0(p^2)$$
(9)
with
$$G^0(p^2)=\mathrm{exp}\left[\mathrm{ln}2\frac{(p^2M^2)^2}{\mathrm{\Lambda }_N^4}\right],$$
(10)
Here $`\mathrm{\Lambda }_N^2`$ is the half-width of the bare form factor, the parameter $`\chi `$ is the amount of pseudoscalar admixture in the bare vertex, and $`f_N`$ is a bare coupling constant. The latter is fixed from the renormalization condition imposed on the dressed vertex at the on-shell point,
$$\overline{u}(p^{})\mathrm{\Gamma }(M)u(p)=\overline{u}(p^{})\gamma ^5g_{\pi NN}u(p),$$
(11)
where $`g_{\pi NN}`$ is the physical pion-nucleon coupling constant (for which we take the value $`13.02`$ ) and $`u(p)`$ is the positive-energy nucleon spinor. The renormalization conditions for the nucleon propagator are described in Section IV. Because of the coupled structure of the equations in Fig. (2), the renormalization of the vertex and that of the propagator are not independent of each other.
The role of the bare $`\pi NN`$ vertex is two-fold. On the one hand, it serves as the driving vertex at the zeroth iteration step. On the other hand, it is used for regularization of the dispersion integrals. The bare vertex is supposed to encapsulate the physics due to degrees of freedom not included explicitly in the dressing.
### B Vertices with $`\mathrm{\Delta }`$, $`\rho `$ and $`\sigma `$
In principle, the dressing procedure should also apply to vertices describing the coupling to the $`\mathrm{\Delta }`$-resonance and to the $`\rho `$\- and $`\sigma `$-mesons. In such an approach one would have to solve a system of 10 coupled equations, instead of the system of two equations in Fig. (2). Clearly, pursuing this is hardly feasible.
For all other vertices except $`\pi NN`$ we ignore the dressing and restrict ourselves to one particular Lorentz covariant form. With each vertex a form factor is associaled which is required for regularization of the loop integrals. The propagators of the $`\mathrm{\Delta }`$-resonance and the $`\rho `$\- and $`\sigma `$-mesons are dressed by the standard summation of loop insertions as discussed in Section IV.
The $`\sigma \pi \pi `$ and $`\rho \pi \pi `$ vertices for a $`\sigma `$ or $`\rho `$ meson with 4-momentum $`p=q+q^{}`$ are taken as
$`(\mathrm{\Gamma }_{\rho \pi \pi })_{\alpha \beta \gamma }^\nu `$ $`=`$ $`(\widehat{e}_{\alpha \beta \gamma })ig_{\rho \pi \pi }F_\rho (p^2)\left[k^\nu {\displaystyle \frac{(pk)}{p^2}}p^\nu \right],`$ (12)
$`(\mathrm{\Gamma }_{\sigma \pi \pi })_{\alpha \beta }`$ $`=`$ $`i{\displaystyle \frac{g_{\sigma \pi \pi }}{\mu }}F_\sigma (p^2)\delta _{\alpha \beta }(qq^{}),`$ (13)
where $`q`$ and $`q^{}`$ are the 4-momenta of the pions with isospin indices $`\alpha `$ and $`\beta `$ respectively, $`k=qq^{}`$ and $`(\widehat{e}_{\alpha \beta \gamma })=iฯต_{\alpha \beta \gamma }`$. The $`\rho `$-meson carries the isospin index $`\gamma `$ and the Lorentz vector index $`\nu `$. $`g_{\rho \pi \pi }`$ and $`g_{\sigma \pi \pi }`$ are coupling constants (the values of all coupling constants will be given later).
For the vertices discussed in this section a generic form factor $`F_r`$ is introduced whose functional form is similar to that of the bare $`\pi NN`$ form factor given in Eq. (10):
$$F_r(p_r^2)=\mathrm{exp}\left[\mathrm{ln}2\frac{(p_r^2\stackrel{~}{m}_r^2)^2(m_r^2\stackrel{~}{m}_r^2)^2}{\mathrm{\Lambda }^4}\right].$$
(14)
normalized to unity at the on-shell point $`p_r^2=m_r^2`$ with a half-width of $`\mathrm{\Lambda }^2`$, the latter taken the same for all vertices considered in this subsection. For the $`\rho \pi \pi `$ and the $`\sigma \pi \pi `$ vertices, $`\stackrel{~}{m}_r`$, the position of the maximum of the form factor, is set equal to the mass of the meson, $`\stackrel{~}{m}_r=m_r`$.
The $`\rho \pi \pi `$ vertex, Eq. (12) chosen such that it vanishes when contracted with the 4-momentum $`p`$ of the $`\rho `$-meson. As a consequence, the spin-0 part of the $`\rho `$ propagator does not contribute to any matrix element (because the projection operator on the spin-0 component is $`๐ซ_{\mu \nu }^0(p)=p_\mu p_\nu /p^2`$). The apparent singularity at $`p^2=0`$ of the vertex Eq. (12) lies outside the kinematical range covered in the calculations. In any case, the $`1/p^2`$-pole behaviour could be compensated by choosing in Eq. (12) a form factor with a zero at $`p^2=0`$.
The $`\rho NN`$ and $`\sigma NN`$ vertices are taken as
$`(\mathrm{\Gamma }_{\rho NN})_\gamma ^\nu `$ $`=`$ $`ig_{\rho NN}F_N(p_N^2){\displaystyle \frac{\tau _\gamma }{2}}\left[\gamma ^\nu +i\kappa _\rho {\displaystyle \frac{\sigma ^{\nu \lambda }q_\lambda }{2M}}\right],`$ (15)
$`\mathrm{\Gamma }_{\sigma NN}`$ $`=`$ $`ig_{\sigma NN}F_N(p_N^2),`$ (16)
where $`q`$ is the (incoming) momentum of the $`\rho `$-meson, and $`g_{\rho NN}`$, $`\kappa _\rho `$ and $`g_{\sigma NN}`$ are coupling constants. The form factor $`F_N(p_N^2)`$, where $`p_N`$ is the 4-momentum of the off-shell nuceon, is given in Eq. (14) with $`\stackrel{~}{m}_N=M`$, the nucleon mass.
The $`\pi N\mathrm{\Delta }`$ vertex used in this calculation can be written as
$$(\mathrm{\Gamma }_{\pi N\mathrm{\Delta }})_\alpha ^\nu =i\frac{g_{\pi N\mathrm{\Delta }}}{\mu ^2}T_\alpha F_\mathrm{\Delta }(p^2)F_N(p_N^2)\left[p/q^\nu (pq)\gamma ^\nu \right],$$
(17)
where $`p`$ is the (incoming) 4-momentum of the $`\mathrm{\Delta }`$-resonance and $`p_N=pq`$ is the (outgoing) nucleon 4-momentum, $`g_{\pi N\mathrm{\Delta }}`$ is a coupling constant and $`T_\alpha ,\alpha =1,2,3`$, are isospin 3/2 to 1/2 transition operators. The form factors $`F_\mathrm{\Delta }`$ and $`F_N`$ are taken as in Eq. (14) with $`\stackrel{~}{m}_N=M`$ and $`\stackrel{~}{m}_\mathrm{\Delta }^2<M_\mathrm{\Delta }^2`$, the mass squared of the $`\mathrm{\Delta }`$, to obtain a reasonable description of the $`P33`$\- phase shift in pion-nucleon scattering (see the discussion of results below). Indications in favour of a $`\pi N\mathrm{\Delta }`$ form factor slightly assymmetric with respect to the $`\mathrm{\Delta }`$ mass have been also found in other works , even though $`\pi N\mathrm{\Delta }`$ vertices different to Eq. (17) have been used there. The dependence of the form factor in Eq. (17) on $`p_N^2`$ turns out to be neccesary to regularize the contribution of the third loop diagram on the righ-hand side of the equation in Fig. (2).
The reason for the particular structure Eq. (17) of the $`\pi N\mathrm{\Delta }`$ vertex is that it possesses the property $`p(\mathrm{\Gamma }_{\pi N\mathrm{\Delta }})_\alpha =0`$. As a consequence, the โsandwichโ of the spin-1/2 part of the Rarita-Schwinger $`\mathrm{\Delta }`$ propagator between two $`\pi N\mathrm{\Delta }`$ vertices vanishes since every term in the spin-1/2 part of the $`\mathrm{\Delta }`$ propagator is proportional to either $`p_\mu `$ or $`p_\nu `$. Thus only the spin-3/2 part of the $`\mathrm{\Delta }`$ propagator gives rise to non-vanishing matrix elements , and it suffices to calculate only the spin-3/2 part of the $`\mathrm{\Delta }`$ self-energy.
## IV Dressed propagators
The inverse of the dressed nucleon propagator can be written as
$$S^1(p)=p/M\mathrm{\Sigma }(p),$$
(18)
with the self-energy given by
$$\mathrm{\Sigma }(p)=\mathrm{\Sigma }_L(p)(Z_21)(p/M)Z_2\delta M.$$
(19)
Here $`\mathrm{\Sigma }_L(p)`$ is the contribution of pion loops,
$$\mathrm{\Sigma }_L(p)=A(p^2)p/+B(p^2)M,$$
(20)
parametrized by the โself-energy functionsโ $`A(p^2)`$ and $`B(p^2)`$. The field and mass renormalization constants $`Z_2`$ and $`\delta M`$ are fixed by requiring that the propagator have a simple pole with a unit residue at $`p/=M`$. This yields
$`Z_2`$ $`=`$ $`1+ReA(M^2)+2M^2{\displaystyle \frac{d}{d(p^2)}}\left[ReA(p^2)+ReB(p^2)\right]|_{p^2=M^2},`$ (21)
$`\delta M`$ $`=`$ $`{\displaystyle \frac{M}{Z_2}}\left[ReA(M^2)+ReB(M^2)\right].`$ (22)
For dressing the $`\mathrm{\Delta }`$-propagator only a one $`\pi N`$ loop approximation is used. Similar to the nucleon self-energy, the imaginary parts of the resonance self-energy are calculated using cutting rules and the real parts are obtained from dispersion integrals.
As explained in the previous Section, the choice of the $`\pi N\mathrm{\Delta }`$ vertex Eq. (17) allows us to retain the spin-3/2 part of the $`\mathrm{\Delta }`$ propagator only,
$$P_{\mu \nu }(p)=\frac{1}{p/M_\mathrm{\Delta }\mathrm{\Sigma }_\mathrm{\Delta }(p)}๐ซ_{\mu \nu }^{3/2}(p),$$
(23)
with the spin-3/2 projection operator
$$๐ซ_{\mu \nu }^{3/2}(p)=g_{\mu \nu }\frac{1}{3}\gamma _\mu \gamma _\nu \frac{1}{3p^2}(p/\gamma _\mu p_\nu +p_\mu \gamma _\nu p/).$$
(24)
Due to the elimination of the spin-1/2 components the $`\mathrm{\Delta }`$ self-energy can be parametrized by only two Lorentz invariant functions ($`A_\mathrm{\Delta }(p^2)`$ and $`B_\mathrm{\Delta }(p^2)`$) instead of 10 functions which would be needed in the general case with the spin-1/2 part present. The structure of the self-energy is thus the same as for the nucleon, Eq. (20). The counter-term contribution to the self-energy contains the real constants $`Z_2^\mathrm{\Delta }`$ and $`\delta M_\mathrm{\Delta }`$ fixed by the renormalization as described for the nucleon. The term $`1/p^2`$ in Eq. (24) does not lead to a singularity if the $`\pi N\mathrm{\Delta }`$ vertices from Eq. (17) are used.
The meson propagators are dressed through the insertion of a $`\pi \pi `$ loop as described in Section II A. The pion propagator thus remains undressed.
The dressed propagator of the $`\sigma `$-meson has the form
$$D(p^2)=\frac{1}{p^2m_\sigma ^2\mathrm{\Pi }_\sigma (p^2)},$$
(25)
where $`m_\sigma `$ is the physical mass of the meson and $`\mathrm{\Pi }_\sigma (p^2)`$ is its self-energy. The latter can be written as a sum of the loop and counter-term contributions,
$$\mathrm{\Pi }_\sigma (p^2)=\mathrm{\Pi }_{\sigma ,L}(p^2)(Z_\sigma 1)(p^2m_r^2)Z_\sigma \delta m_\sigma ^2,$$
(26)
where $`Z_\sigma `$ and $`\delta m_\sigma ^2`$ play the role of the field and mass renormalization constants. These constants are fixed by requiring that the expansion of $`Re\mathrm{\Pi }_\sigma (p^2)`$ contain only second and higher powers of $`(p^2m_\sigma ^2)`$ . In other words,
$$\frac{1}{p^2m_\sigma ^2Re\mathrm{\Pi }_\sigma (p^2)}$$
(27)
is required to have a simple pole with a unit residue at $`p^2=m_\sigma ^2`$. This yields $`Z_\sigma `$ and $`\delta m_\sigma ^2`$ in terms of $`Re\mathrm{\Pi }_{\sigma ,L}(p^2)`$:
$$Z_\sigma =1+\frac{d}{d(p^2)}Re\mathrm{\Pi }_{\sigma ,L}(p^2)|_{p^2=m_\sigma ^2},$$
(28)
$$\delta m_\sigma ^2=\frac{Re\mathrm{\Pi }_{\sigma ,L}(m_\sigma ^2)}{Z_\sigma }.$$
(29)
Following similar arguments as for the $`\mathrm{\Delta }`$, the structure of the vertex Eq. (12) has been chosen such that only the spin-1 part of the dressed $`\rho `$ propagator can be retained,
$$D_{\mu \nu }(p)=\frac{1}{p^2m_\rho ^2\mathrm{\Pi }_\rho (p^2)}๐ซ_{\mu \nu }^1(p),$$
(30)
where
$$๐ซ_{\mu \nu }^1(p)=g_{\mu \nu }\frac{p_\mu p_\nu }{p^2},$$
(31)
is the spin-1 projection operator and $`\mathrm{\Pi }_\rho (p^2)`$ is the self-energy which has the same structure as for a scalar particle.
## V Changing representation
It is known that interacting Greenโs functions depend on the representation of fields in the Lagrangian. There exist a wide class of field transfomations which do not affect the asymptotic behaviour of the fields and hence leave the S-matrix (and thus all observables) invariant . The invariance of the S-matrix under field transformations is known as the equivalence theorem .
In this Section we will take advantage of this irrelevance of representation and transform the effect of the dressing of the nucleon propagator into new $`\pi NN`$ vertices. The representation constructed in this Section is an example of the โphysical representationโ discussed in Ref. .
We introduce the notation where the subscript $`\mathrm{\Sigma }`$ labels the representation where the propagator $`S`$ contains a non-trivial self-energy. The new representation is defined by the two requirements: 1) the nucleon propagator must be equal to the free propagator $`S^0(p)`$, 2) it must be possible to construct the K-matrix as in Fig. (1), i. e. solely in terms of 2- and 3-point Greenโs functions. The new $`\pi NN`$ vertex $`\mathrm{\Gamma }`$ must thus be a solution of the equation
$$\mathrm{\Gamma }(p)S^0(p)\overline{\mathrm{\Gamma }}(p)=\mathrm{\Gamma }_\mathrm{\Sigma }(p)S(p)\overline{\mathrm{\Gamma }}_\mathrm{\Sigma }(p),$$
(32)
where only the dependence on the internal, off-shell, 4-momentum, $`p`$, is indicated. All effects of the dressing are now contained in the difference between the new dressed and the bare vertex, $`\mathrm{\Gamma }\mathrm{\Gamma }^0`$.
The solution of Eq. (32) can be written as
$$\left[ReG(p^2)\pm \frac{W}{2M}ReG_V(p^2)\right]=\left[ReG_\mathrm{\Sigma }(p^2)\pm \frac{W}{2M}ReG_{\mathrm{\Sigma },V}(p^2)\right]\sqrt{\frac{S(\pm W)}{S^0(\pm W)}}$$
(33)
where $`G=G_S+G_V/2`$ and $`W=\sqrt{p^2}0`$ is the invariant mass of the virtual nucleon. Also, $`S^0(\pm W)=1/(MW)`$ and
$$S(\pm W)=\frac{1}{Z_2(MW)Z_2\delta M\pm ReA(p^2)W+ReB(p^2)M}$$
(34)
are positive- and negative-energy parts of the free and dressed nucleon propagators, respectively.
In the present work we consider only those solutions for the dressed propagator $`S`$ that do not have real poles in addition to the nucleon pole $`W=M`$. With this qualification, the solution Eq. (33) for the $`\pi NN`$ vertex in the new representation is well defined, since, due to the renormalization procedure, the ratio $`S(\pm W)/S^0(\pm W)`$ is positive.
An extra pole at positive $`W`$ would correspond to an additional asymptotic state, different to the free nucleon. To take it into account properly, certain modifications would be necessary of the standard renormalization of the nucleon field Eq. (21): an additional field renormalization constant would probably be required to account for the fact that a new particle species occured as a result of the dressing . Such a study lies outside the scope of the present work.
## VI Discussion
The masses of the particles included in the model are given in Table I and kept fixed in the calculation of pion-nucleon phase shifts.
An important characteristic of the bare vertex is its half-width $`\mathrm{\Lambda }_N^2`$, see Eq. (10). To investigate the dependence of the dressing on $`\mathrm{\Lambda }_N^2`$, calculations have been done for two values, $`\mathrm{\Lambda }_N^2=2`$ GeV<sup>2</sup> and $`\mathrm{\Lambda }_N^2=3`$ GeV<sup>2</sup>, referred to as calculations (I) and (II), respectively. The requirement that a converged solution of the dressing procedure can be obtained, without developing additional poles of the propagator (see Section V), puts an upper bound on $`\mathrm{\Lambda }_N^2`$. While the exact value of this limit depends also on other parameters of the model, it is certain that the bare form factor cannot be arbitrarily hard. Note however that the scale introduced by the bare form factor, which is of the order of $`M^2+\mathrm{\Lambda }_N^2`$, is larger than the scale due to the degrees of freedom explicitly included in the dressing. The values of the bare coupling constant $`f_N`$, introduced in Eq. (9), are given in Table II, where also the values of the field and mass renormalization constants are listed.
We find that a sizable pseudoscalar admixture in the bare vertex (with $`|\chi |>0.1`$ in Eq. (9)) leads to a poor description of low energy phase shifts. This is intimately related to the smallness of explicit chiral-symmetry breaking. Besides, even without resorting to phenomenology, the range of variation of $`\chi `$ is severely constrained by the requirement of convergence. Both calculations presented in this work were done with $`\chi =0.055`$.
The values of the parameters in the vertices for the $`\mathrm{\Delta }`$-resonance and $`\rho `$\- and $`\sigma `$-mesons, Eqs. (12 \- 17), are summarized in Table III. The constants $`g_{\pi N\mathrm{\Delta }}`$, $`g_{\rho \pi \pi }`$ and $`g_{\sigma \pi \pi }`$ were fixed from the decay widths of the $`\mathrm{\Delta }`$, $`\rho `$ and $`\sigma `$ . The value of half-width $`\mathrm{\Lambda }^2`$, see Eq. (14), was kept fixed and had to be sufficiently soft to provide convergence of the dressing procedure.
The coupling constants $`g_{\rho NN}`$, $`\kappa _\rho `$, $`g_{\sigma NN}`$, as well as the parameter $`\stackrel{~}{m}_\mathrm{\Delta }^2`$, were chosen from a comparison of the calculated $`\pi N`$ phase shifts with the data, taken from . Together with $`\chi `$, discussed above, the five adjustable parameters are given in the last five columns in Table III. It should be stressed that only for a rather restricted range of these constants a convergent solution of the dressing procedure could be found.
The phase shifts in pion-nucleon scattering are shown in Figs. (3) and (5) as function of the pion kinetic energy in the laboratory system, corresponding to calculations (I) and (II), respectively. The solid lines are the phase shifts calculated with the dressed K-matrix as shown in Fig. (1). The dashed lines are obtained in the approximation where $`K`$ is set equal to the potential $`V`$, hence without taking the dressing into account.
The effect of the dressing on the $`\pi NN`$ vertex, can be seen more clearly from Figs. (4) (calculation(I)) and (6) (calculation (II)). The form factors are shown as functions of $`p^2`$, the invariant mass squared of the virtual nucleon. The representation constructed in Section V is particularly useful because in it the nucleon propagator is free and the effects of nucleon dressing are encapsulated solely in the difference $`G_{V,S}(p^2)G_{V,S}^0(p^2)`$ between the dressed and bare $`\pi NN`$ form factors. For this reason we do not present results for the self-energy functions. It should be stressed that in virtue of the equivalence theorem , either of the two representations described in Section V lead to identical results for the phase shifts. The upper and lower panels contain pseudovector and pseudoscalar form factors, respectively (please note that Fig. (4) and Fig. (6) have different vertical scales). The dotted lines are the bare form factors, see Eq. (9), with the constants $`f_N`$ and $`\chi `$ given in Tables II and III. The dashed lines are the form factors obtained after the first iteration step (essentially, a one-loop correction to the bare vertex) and the solid lines are the fully dressed form factors. A comparison of the solid and dashed lines exhibits a non-perturbative aspect of the dressing in the sense that it goes beyond an inclusion of few loop corrections. It can be seen that the ratio of pseudoscalar and pseudovector form factors remains small if the nucleon is not far off the mass-shell. The dash-dotted curves correspond to the form factors in the representation where the nucleon self-energy has not been eliminated. We see that the dressed $`\pi NN`$ vertex may depend significantly on the representation chosen.
Comparing the form factors in Fig. (4) with those in Fig. (6), we conclude that the converged solution depends strongly on the width of the bare form factor. However, independent of this width, the dressing causes considerable softening of the form factor at higher invariant masses. The results shown in Figs. (3) and (5) suggest that it is possible to obtain a reasonable description of phase shifts up to pion laboratory energies of about 400 MeV starting from bare form factors with rather different widths.
In this model only one resonance of the $`\pi N`$ scattering, the $`\mathrm{\Delta }`$, was included. The lack of other resonances becomes especially conspicuous at higher energies. In fact, in the calculations of phase shifts we also included the Roper resonance, though it is not taken into account in the dressing. This improved the calculated P11 phase shift at energies of about 300 MeV and higher, with a negligible effect on the other phase shifts. In principle, the Roper can be easily included in the dressing, as well as other important degrees of freedom (for example, the S11-resonance). In the present version of the model we have forgone doing so, limiting ourselves to the lowest lying $`\mathrm{\Delta }`$-resonance only.
## VII Conslusions
We have presented a model in which considerations of unitarity and anlyticity (causality) are implemented in the K-matrix approach to pion-nucleon scattering. The principal ingredient of the model is the dressing procedure, formulated in terms of half-off-shell vertices and propagators, the building blocks of the K-matrix. Analyticity properties are exploited through the use of dispersion relations to obtain the principal value parts of loop integrals required for the unitarization of the scattering amplitude.
The five parameters of the model are constrained rather much by the requirement of convergence of the dressing procedure. By the same token, there is an implicit interdependence of the parameters. This means that a comparison with experiment is an important test for this approach. We showed that a good overall description of phase shifts in pion-nucleon scattering can be achieved at the energies exceeding the scale due to the degrees of freedom explicit in the model. This suggests that the developed dressing procedure provides a physically reasonable method for studying higher-order correction to the (unitarized) Born approximation traditionally adopted in the K-matrix approach.
###### Acknowledgements.
This work is part of the research program of the โStichting voor Fundamenteel Onderzoek der Materieโ (FOM) with financial support from the โNederlandse Organisatie voor Wetenschappelijk Onderzoekโ (NWO). We would like to thank Alex Korchin, Rob Timmermans and John Tjon for discussions. |
warning/0001/astro-ph0001443.html | ar5iv | text | # Scale Dependent Dimension of Luminous Matter in the Universe
## Abstract
We present a geometrical model of the distribution of luminous matter in the universe, derived from a very simple reaction-diffusion model of turbulent phenomena. The apparent dimension of luminous matter, $`D(l)`$, depends linearly on the logarithm of the scale $`l`$ under which the universe is viewed: $`D(l)3\mathrm{log}(l/l_0)/\mathrm{log}(\xi /l_0)`$, where $`\xi `$ is a correlation length. Comparison with data from the SARS red-shift catalogue, and the LEDA database provides a good fit with a correlation length $`\xi 300`$ Mpc. The geometrical interpretation is clear: At small distances, the universe is zero-dimensional and point-like. At distances of the order of 1 Mpc the dimension is unity, indicating a filamentary, string-like structure; when viewed at larger scales it gradually becomes 2-dimensional wall-like, and finally, at and beyond the correlation length, it becomes uniform.
Uniformity of the background radiation requires that the universe must be homogeneous at the largest scale; this is known as the cosmological principle. However, a decade ago, Coleman and Pietronero suggested that the universe, at length scales $`L`$ up to a couple of Mpc is fractal with fractal dimension, $`D1.2`$, based on a study of the CfA galaxy catalogue. Subsequent studies seemed to confirm this picture: Guzzo et al. found $`D=1.2`$ for $`L=13`$ Mpc, increasing to $`D=2.2`$ for $`L=310`$ Mpc, from the Perseus-Pisces catalogue. Martinez and Coles found that the dimension gradually increases from 2.25 to 2.77 at length scales increasing from $`150`$ Mpc. These empirical studies have recently been reviewed by Wu et al . Even though there is a general agreement about the existence of fractal galactic structures at moderate scales, there is still intense debate whether or not the universe is homogeneous at very large scales and, if so, how the transition to homogeneity takes place . The value of the homogeneity scale and the matter distribution within such scale have great cosmological consequences.
We propose that the distribution of luminous matter in the Universe can be described by a new geometric scaling form that we discovered recently in a different context. This description leads to a reconciliation of observational data at various scales and a consistent phenomenology of the crossover to homogeneity. A sharp transition to homogeneity at $`300`$Mps is predicted.
The model is a simple non-equilibrium reaction-diffusion โforest-fireโ model , proposed to capture the essential features of turbulent systems, where energy is injected at the largest scale, and dissipated at a small length scale. We found that in a range of length scales between these two limits the dimension of the luminous field (fire distribution) varies gradually from zero to three, and that the distribution becomes homogeneous beyond a correlation length which depends on the energy injection rate. The model operates near a dynamical โcritical pointโ, with diverging correlation length.
As we will show below, analysis of galaxy maps indicates that the geometrical structure of luminous matter in the universe is very similar to that of the forest-fire model. Our alternative form provides a better fit to the data than conventional models. The underlying picture is one where luminous matter is being created and destroyed in an ongoing non-equilibrium dynamical process. This similarity is appealing in that it suggests that the universe shares the basic characteristic features of other dynamical systems, so perhaps the dynamics of the universe is not unique, but belongs to a more general universality class of non-equilibrium turbulent systems.
Usually, systems near equilibrium criticality are self-similar, or fractal, for length scales below the correlation length; hence fractal behavior can often be viewed as a consequence of criticality . However, the forest-fire model does not show simple power-law (fractal) scaling below the correlation length. Numerical studies show that the average amount of dissipation $`n(l)`$, within a cube box of size $`l`$ that contains dissipation, obeys
$$\mathrm{log}(n)\left(\frac{3}{2}\frac{\mathrm{log}(l/l_0)}{\mathrm{log}(\xi /l_0)}\right)\mathrm{log}(l/l_0),$$
(1)
where $`l_01`$ lattice spacing for the forest fire model. At the correlation length $`\xi `$, there is a sharp cross-over to a homogeneous 3d structure. Thus, the correlation length is identical to the homogeneity length, as is usually the case in critical phenomena. However, the lack of self-similarity implies that one can derive the correlation length from observations in a range of much smaller length scales.
This equation can be interpreted in terms of an apparent fractal dimension for luminous matter that varies linearly with the logarithm of the length scale:
$$D(l)=\frac{d\mathrm{log}(n)}{d\mathrm{log}(l)}3\left(\frac{\mathrm{log}(l/l_0)}{\mathrm{log}(\xi /l_0)}\right).$$
(2)
We suggest that the length scale dependent behaviour observed in this model may be sufficiently general that it is worthwhile to make a detailed comparison with real astronomical data. The underlying viewpoint is that the galactic dynamics is turbulent, with stellar objects interacting with one another in reaction-diffusion type processes through shock waves, super-novae explosions, galaxy mergers, etc. Apart from an overall amplitude, there are only two fitting parameters in our proposed galaxy distribution, namely the upper length scale $`\xi `$, where the distribution becomes uniform, and the lower cutoff, $`l_0`$, where the distribution becomes point like.
In their seminal work, Sylos-Labini, Pietronero and coworkers analysed several database catalogues of galaxy maps. From the databases, they created volume-limited (VL) samples containing all galaxies exceeding a certain absolute luminosity within a given volume. Then they calculated the conditional density $`\mathrm{\Gamma }^{}(l)`$, which is the average density of galaxies within a sphere of size $`l`$. This quantity corresponds to the density $`n(l)`$ defined above, divided by the volume $`l^3`$. Thus, the resulting prediction for $`\mathrm{\Gamma }^{}(l)`$ becomes
$$\mathrm{log}\left(\mathrm{\Gamma }^{}(l)\right)\left(\frac{3}{2}\left(\frac{\mathrm{log}(l/l_0)}{\mathrm{log}(\xi /l_0)}\right)3\right)\mathrm{log}(l/l_0).$$
(3)
Namely, on a log-log plot there is a pure quadratic dependence, rather than the linear dependence found for self-similar fractal structures.
We have fitted the above expression to the conditional densities extracted by Pietronero et al. from two widely different data bases with consistent results. The LEDA database is a heterogeneous compilation of data from the literature containing more than 200,000 galaxies. The Stromlo-APM red shift survey (SARS ) consists of 1797 galaxies. Figure 1 shows results from the fits, with two different cut-offs for the LEDA database. The labeling follows Sylos Labini et al., with the numbers representing the lower luminosity cut-offs. Obviously, there are larger fluctuation for the sparser, but perhaps higher quality, Stromlo-APM data set than for the LEDA database. The fits are very good in view of the fact that the only fitting parameters are the upper and lower length scales, $`\xi `$ and $`l_0`$, respectively. In contrast to conventional critical phenomena, the correlation length enters the expression for length scales below the correlation length. We are therefore able to fit the correlation length to the data, despite the fact that no data is available at and beyond the projected correlation length.
The upper length scale is the one where the curves become flat, $`d=3`$. The three fits yield very consistent values of this length scale, $`\xi =260`$ Mpc from the LEDA16 data, $`\xi =275`$ Mpc from the LEDA14 data, and $`\xi =380`$ Mpc from the APM data.
The empirical logarithmic scale dependence of the dimension can be seen directly by re-plotting the data in figure 1: Figure 2 shows $`D(l)=2\times (\mathrm{log}(\mathrm{\Gamma }^{}(l)/\mathrm{\Gamma }(l_0))/\mathrm{log}(l/l_0)+3)`$. All data sets yield linear behavior. The correlation length is found by linear extrapolation to the point where $`D(l)`$ assumes the value of 3. The dimensions derived from the intense galaxies, LEDA 16 and APM 18, are essentially identical, but the LEDA 14 data yield a somewhat steeper scale dependence. However, they all converge at essentially the same homogeneity length.
We predict a sharp crossover to uniformity, i. e. a sharp kink in the curve, at this length, which will be readily observable once data becomes available. Actually, there is a recent analysis based on ESO Slice Project galaxy redshift survey which indicates that the fractal dimension is close to 3 for the length scale greater than 300 Mpc ). Also, the intermediate data points are predicted to follow the straight lines in figure 2.
The lower cut-off, $`l_0`$, is the scale at which the slope of the curves in the figure assumes the value of -3. We find $`l_0=370`$ light-years, $`l_0=3700`$ light-years, and $`l_0=330`$ light years for the three samples, respectively. This scale is determined with less precision than the correlation length $`\xi `$. It is not clear how well our scaling form applies to the analysis of the galaxy distribution at small length scales.
The geometry of the luminous set is not fractal when viewed over the entire range of scales, since there is no self-similarity for different scales. Nevertheless, the scale dependent dimension has a clear geometrical interpretation: At small distances, the universe is zero-dimensional and point-like. Indeed, energy dissipation takes place on individual point objects, like stars and galaxies. At distances of the order of 1 Mpc the dimension is unity, indicating a filamentary, string-like structure; when viewed at larger scales it gradually becomes 2-dimensional wall-like, and finally at the correlation length, $`\xi `$, it becomes uniform.
It might be instructive to compare with more conventional interpretations of the large scale structure . The conditional density can be related to a correlation function $`g(r)`$ through
$$\mathrm{\Gamma }^{}(l)<n>(1+g(l)),$$
(4)
where $`<n>`$ is the mean density of galaxies. For instance, the field theory of de Vega et al. yields an expression of this form. The correlation function is often assumed to be of the form $`g(l)=(r_0/l)^\gamma `$. Figure 1 also shows a fit to this expression, with $`r_0=10`$ Mpc and $`\gamma =1.3`$. The fit is clearly inferior, flattening out at too small length scales. This is in accordance with the observations by Sylos Labini et al. that the value of fitted parameter $`r_0`$ depends heavily on the range of length scales used. At larger scales, the difference between the two fits is even more pronounced; when further data becomes available in the near future, one should be able to discriminate even better between the two pictures. In this traditional view, there is a smooth crossover to homogeneity when the amplitude, expressed in terms of $`r_0`$ reaches unity. In contrast, following our โcritical phenomenaโ viewpoint, there is a sharp, possibly exponential, cutoff of the non-uniform part of the correlation function at the correlation length.
This has some important cosmological consequences. In the traditional formulation, one usually visualizes that the amplitude $`r_0`$ of the power-law fluctuations increases with time, starting from the time of the decoupling of radiation from hadronic matter, leading to an increase of the cross-over length to homogeneity. In our phenomenology, the correlation length $`\xi `$ is the only parameter, so it is this quantity which is increasing with time. The average density of galaxies in the universe is equal to the density within the correlation length, i.e. $`<n>=\mathrm{\Gamma }^{}(\xi )1/\xi ^{3/2}`$. Thus, once the correlation length has been determined, one knows the density of galaxies. Assuming that the entire density of hadronic matter scales as that of the luminous galaxies studied here, one might get an estimate of the mass of the universe. From the fit to LEDA 14 one gets that the density of galaxies in the entire universe with apparent magnitude greater than 14 is $`<n>=2\times 10^3Mpc^3`$. From the fit to the APS 18 data we find that the density of galaxies with apparent luminosity greater than 18 is $`<n>=3\times 10^4Mpc^3`$. The traditional fits give much larger values for the density of galaxies in the universe, depending on the range of length scales used in the fit .
In the forest fire model the energy flux (which determines the average density of fires) is an independent parameter, namely the growth rate of trees, whereas for the universe it is self-consistently determined by the dynamics. The forest fire model exhibits self-organised criticality , in the sense that the correlation length diverges as the tree growth rate $`p0`$. All fire goes extinct as the correlation length reaches the system size. As the universe expands, the correlation length $`\xi (t)`$ increases faster than the size of the universe $`R(t)`$ and the universe become more and more inhomogeneous. One might speculate that as the correlation length reaches the size of the universe, all the luminous matter is extinguished, and we are left with a universe without luminous matter!
Acknowledgment. We thank Maya Paczuski, Kim Sneppen, and Jakob Bak for helpful discussions and comments on the manuscript.
Figure Captions
Figure 1. Conditional average densities for various galaxy catalogues (arbitrary scale), as derived by Sylos Labini et al , compared with fits to equation 1, yielding $`\xi =260`$ Mpc from the LEDA16 data, $`\xi =275`$ Mpc from the LEDA14 data, and $`\xi =380`$ Mpc from the APM data. The broken line is a conventional fit to equation 4 with $`\gamma =1.3`$, $`r_0=10`$ Mpc.
Figure 2. Scale dependent dimension $`D(l)`$ derived from the data points in figure 1 as explained in text. We conjecture that future data points follow the straight lines, and saturate sharply to $`D=3`$ at the correlation length. |
warning/0001/astro-ph0001235.html | ar5iv | text | # Abundances in Very Metal-Poor Stars
## 1 Introduction
Halo stars can be viewed in several contexts. They constitute the oldest and most extended stellar population known in the Galaxy. As probes of Galactic chemical evolution (GCE), they are the oldest objects and have the lowest metallicities, and hence provide the first data in the evolutionary sequence. In a third context, the early evolution of the universe, Figure 1 shows the metallicity distributions of 36 halo globular clusters (Laird et al. 1988a), 373 halo field stars (Laird, Carney & Latham 1988b; Ryan & Norris 1991) and 34 damped Lyman-$`\alpha `$ systems (DLAs; Pettini et al. 1997). Not only are the field and cluster metallicity distributions comparable, they are lower in metallicity than the DLAs having redshifts $`z`$ 1โ3. That is, very metal poor stars are amongst the lowest metallicity objects in the known universe.
The surprise some people express in discovering that DLAs are generally more metal rich than the Galactic halo emphasises that our knowledge of the DLAs has yet to mature. There is ongoing debate about what they really are, possibilities including:
$``$ spiral disks/protodisks/thick disks (e.g. Wolfe et al. 1986; Lu et al. 1996)
$``$ dwarf galaxies (Pettini, Boksenberg & Hunstead 1990; Pettini et al. 1999a)
$``$ ejecta from dwarf galaxies (Nulsen, Barcons & Fabian 1998).
In examining Galactic stars, we have the advantage of studying objects with reasonably well understood histories and physical states, whose spectra are not blended, and whose abundances, which are measurable for many elements, are unaffected by depletion onto dust grains.
The value of halo stars for studying early epochs of the universe may be further illustrated by considering additional objects in metallicity-redshift space. Figure 2 shows a number of Galactic and high-redshift objects, along with three GCE models assuming outflow, no outflow (closed box) and inflow for Galaxy formation assumed to begin at redshift $`z=5`$ (Edmunds & Phillipps 1997). (A Hubble constant $`H_0=50`$ km s<sup>-1</sup> Mpc<sup>-1</sup> and flat cosmology \[deceleration parameter $`q_0`$ = 0.5\] were used to establish the age-redshift relation.) The disk star sequence and bulge of the Galaxy are based on the observations of Edvardsson et al. (1993) and Sadler, Rich & Terndrup (1996). The high-redshift objects are the DLAs from Pettini et al. (1997) binned in redshift, Molaro et al. (1996) and Lu et al. (1996), a region corresponding to the Lyman-$`\alpha `$ forest (Hellsten et al. 1997), and one Lyman break galaxy (LBG) from Pettini et al. (1999b). The redshift distribution in the Hubble Deep Field (Bouwens, Broadhurst & Silk 1998) is shown as an inset against the vertical axis. In adjoining panels are the extinction-corrected star formation rate (Steidel et al. 1999) and the quasar space density (Warren, Hewitt & Osmer 1994) whose steep fall at redshift $`z`$ $`>`$ 3.5 indicates that these objects were still forming prior to this epoch, presumably along with galaxies. Several points can be noted.
$``$ The occurrence of systems covering a wide metallicity range at the same high redshift โ the Lyman-$`\alpha `$ forest at $`3`$ $`<`$ \[M/H\] $`<`$ $`2`$, DLAs at $`2`$ $`<`$ \[M/H\] $`<`$ $`1`$, and a Lyman break galaxy at \[M/H\] $`>`$ $`1`$ โ suggests that we are probing diverse objects, not necessarily an evolutionary sequence, in the high redshift universe.
$``$ The overlapping of high redshift and Galactic objects in this epoch-metallicity plane (redshift translates to epoch), e.g. the LBG and the Galactic bulge, and the Lyman-$`\alpha `$ forest and metal poor stars, emphasises that these objects provide complementary views on the formation and evolution of galaxies. No one class should be considered in isolation from the others.
$``$ Galactic stars with \[Fe/H\] $`<`$ $`3`$, corresponding to $`z`$ $`{}_{}{}^{}{}_{}{}^{>}`$ 4โ5, uniquely probe the earliest star formation events. A high level of detail is achievable because many elements can be measured in well understood objects. Furthermore, the elements in these objects owe their existence to very few previous generations of stars, possibly only one (Ryan, Norris & Bessell 1991). They mark the beginning of GCE, and as such will be the main topic of this review.
## 2 How Many Supernovae Make a Population II Star?
Disk stars formed from gas enriched by many previous generations of stars. Population II stars formed earlier and have lower metallicities, consistent with fewer supernovae being involved. A simple closed box model of GCE (e.g. Searle & Sargent 1972; Pagel & Patchett 1975) establishes a framework based on a single free parameter โ the fraction of enriched matter returned to the interstellar medium by a stellar generation, known as the yield. No check is kept on timescales or stellar generations; with instantaneous processing and complete mixing, the model follows essentially the average evolution of a galaxy. Despite these gross simplifications, the model compares very favourably with the metallicity distribution of the Galactic halo; Figure 1 compares halo field star data with a simple model with yield $`y=10^{1.6}`$ (Ryan & Norris 1991). However, models of this type are incapable of indicating how many supernovae are required to enrich a stellar Population to a given metallicity, and alternatives treating the formation and evolution of stars more realistically had to be developed.
One model which considered individual generations of stars was Searleโs (1977) stochastic model, which postulated that separate star forming fragments underwent star formation according to Poisson statistics, the mean number of enrichment events increasing with time and the enrichment from each being constant. This model again has only one free parameter, the mean number of enrichments prior to the termination of star formation, $`\mu `$, but it is capable of quantifying the number of supernovae involved. Its metallicity distribution is broadly similar to that of the simple model. Applied to halo field stars, Searleโs stochastic model was not obviously better than the simple model, but did give a marginally better fit to the globular cluster distribution, implying a mean of ten enrichments per fragment (Ryan & Norris 1991). With the mean metallicity of the halo globular clusters being \[Fe/H\] $``$ $`1.5`$, this suggested that a single event could enrich a fragment to \[Fe/H\] $``$ $`2.5`$.
Truran (1981) argued that the atmospheres of second generation stars would contain r- but no s-process isotopes, due to the latterโs secondary nature, whereas subsequent generations would contain both types. Although it was not clear at which metallicity this would be achieved, it forced people to think about the first few generations of stars.<sup>1</sup><sup>1</sup>1 It is uncertain that we will actually see a clear division between r- and s-process elements in second versus third generation stars, because most neutron-capture elements have contributions from both the s- and r-process. Large star-to-star variations in the neutron-capture abundances of stars with \[Fe/H\] $`<`$ $`2.5`$ (e.g. Gilroy et al. 1988) suggested that prior nucleosynthesis involved small numbers of supernovae (SN). Related observations of Sr led Ryan et al. (1991) to consider stochastic enrichment by only a few stellar generations, and to calculate the metallicity produced by single supernova as \[Fe/H\] = $`3.8`$, based on a typical SN II progenitor mass of 25 M and an assumed primordial cloud mass of $`10^6`$M.<sup>2</sup><sup>2</sup>2The SN mass was a compromise between higher mass stars being rarer and lower mass stars having lower yields. The cloud mass was based on large globular clusters, giant molecular clouds, and the collapse of metal poor gas. This coincided with the observed onset of huge abundance variations of strontium (by a factor of 100 or more), and was consistent with the lowest stellar metallicities then observed.<sup>3</sup><sup>3</sup>3The vanishing of Sr in some stars at this metallicity suggested to me, at the time, that Truranโs mechanism was possibly being observed, and that genuine second generation stars were being identified. However, the more recent availability of data on Ba has altered my views on this; see Footnote 1 and later sections of this paper.
Star-to-star differences in neutron-capture element abundances at the lowest metallicities also required that chemical inhomogeneities existed around the time these stars were forming. Audouze & Silk (1995) showed further that mixing timescales in the halo were sufficiently long that inhomogeneities of this type would not be erased on the timescale over which stars would form, supporting the proposition that the progeny of the first supernovae would be found around this metallicity. Mounting examples of neutron-capture element variations from star to star (e.g. Norris, Peterson & Beers 1993; McWilliam et al. 1995; Ryan 1996; Ryan et al. 1996) strengthened the view that the most metal-poor stars exhibit the ejecta of very small numbers of supernova.
The need to integrate small number statistics of the first supernova with GCE models led Ryan, Norris & Beers (1996) to examine the enrichment sphere of a single SN in a primordial cloud. Adopting the supernova remnant (SNR) model of Cioffi, McKee & Bertschinger (1988), they calculated the cloud mass with which the SN ejecta would mix as
$$m_{\mathrm{ISM}}=3.4\times 10^4E_{51}^{0.95}n_0^{0.10}\zeta _m^{0.15}(\beta C_{06})^{1.29}\mathrm{M}_{},$$
where $`E`$, $`n`$, $`\zeta `$, and $`\beta C_{06}`$ refer to the explosion energy, cloud density, cloud metallicity, and shock speed in appropriate units. The main features of this result were:
$``$ the mass of gas enriched is almost independent of the cloud characteristics ($`n,\zeta `$) and depends strongly (almost linearly) on the energy of the SN;
$``$ the typical enriched mass of the cloud would be $`7\times 10^4`$M;
$``$ the typical metallicity following this first enrichment would be \[Fe/H\] = $`2.7`$, matching (perhaps coincidentally) the changes in the behaviour of iron-peak and neutron-capture elements and the lowest metallicity globular clusters.
Many other GCE models have been developed that incorporate the initial mass function, the mass-dependence of stellar lifetimes, the mass- and metallicity-dependence of supernova and stellar-wind yields, and more. In the light of observational and theoretical reasons for expecting the first star forming regions to be poorly mixed, new GCE models have been forthcoming that include SNR physics and inhomogeneous mixing (e.g. Shigeyama & Tsujimoto 1998; Tsujimoto & Shigeyama 1998; Ishimaru & Wanajo 1999a; Argast & Samland 1999), against which the abundances of very metal poor stars can be compared. It is then possible to invert the problem and use the observed abundances to constrain the model inputs. As stars at these low abundances are believed to be second generation stars, of particular interest will be the shape of the IMF of their progenitors (Population III stars!), the mass limits for the production of SN of Population III stars, and the yields of individual Population III objects.
## 3 Abundances: Can You Believe What You Read?
Weak lines have the greatest sensitivity to abundance and the least sensitivity to uncertain parameters of the stellar atmosphere. However, the lines that are weak in very metal poor stars are strong in the sun, so completely different lines are measured in the two cases, the former often also being of lower excitation potential and possibly of a different ionisation state. Photometric temperature calibrations and stellar atmosphere models also depend on metallicity. These factors give rise to potential systematic differences between analyses conducted for metal rich compared to metal poor stars. The overall rarity of spectral lines in metal poor stars also limits consistency checks between several lines of one element. There can also be substantial differences in the approaches of investigators, who may adopt different reference solar abundances, model atmosphere grids, and atmospheric parameters. Differences of 10% in the equivalent widths measured in two studies of one star are not uncommon. Differences of 0.2 dex in the absolute abundances can accumulate through these effects.
Fortunately, for many species relative abundances (ratios) \[X/Fe\] are less susceptible to errors than \[X/H\]; although errors in \[Fe/H\] better than 0.10 dex are rare, \[X/Fe\] can often be believed at the level of 0.05 dex (though in some cases only 0.15 dex may be achieved). Homogeneous studies, where abundances are derived almost identically for all stars, can achieve better internal accuracy.
Additional errors can arise from the assumptions of the analysis:
$``$ Effective temperature scales are a large source of error. Infra-red flux method (IRFM) temperatures, often argued to be preferable to other photometric calibrations, are now available for many halo stars (Alonso, Arribas & Martรญnez-Roger 1996), although the uncertainties on any individual star are currently large.
$``$ Collisional damping constants have improved to the extent that strong lines may now provide more reliable abundances than weak lines (Anstee, OโMara & Ross 1997). Damping constants for many neutral transitions have been published (Anstee & OโMara 1995; Barklem & OโMara 1997,1998; Barklem, OโMara & Ross 1998); comparisons with older computations are presented elsewhere (Ryan 1998).
$``$ Corrections for non-LTE have been published for several elements: e.g. Li (Carlsson et al. 1994; Pavlenko & Magazzu 1996), Be (Garcรญa Lรณpez, Severino & Gomez 1995), B (Kiselman 1994; Kiselman & Carlsson 1996), O (Kiselman 1991), Na (Baumรผller, Butler & Gehren 1998), Mg (Zhao, Butler & Gehren 1998), Al (Baumรผller & Gehren 1997), and Ba (Mashonkina, Gehren & Bikmaev 1999).
$``$ 3-D hydrodynamical models are being computed to investigate the errors introduced by 1-D models. Preliminary work signals some interesting results (Asplund et al. 1999), including the primordial Li abundance having been overestimated.
## 4 The Lightest Elements
The primordial and spallative elements Li, Be, and B will be thoroughly considered by IAU Symposium 198. In the context of globular clusters versus field stars (see Figure 3), it is useful to compare M92 subgiants (Boesgaard et al. 1998) and field halo turnoff stars (Ryan, Norris & Beers 1999). The latter show no intrinsic spread ($`\sigma _{\mathrm{int}}<0.02`$ dex) once a small metallicity dependence (believed to be due to GCE) is taken into account, whereas M92 shows a range of Li. This suggests a difference between the field and globular cluster populations, presumably related to their very different environments, specifically the stellar density. Although single stars seldom interact, protostellar disks may have done so in nascent globular clusters but not in lower density clusters destined ultimately to produce field stars (Kraft 1998, private communication). Such speculation falls short, however, of explaining why the M92 subgiants might have a higher Li spread than in the field. Boesgaard et al. favoured rotationally-induced turbulence resulting in a spread of Li depletion factors from a higher initial value. One might imagine that different angular momentum histories of cluster and field stars could lead to differences of this sort, though the extreme thinness of the field star A(Li) distribution (Ryan et al. 1999) is a lingering difficulty with that scenario.
## 5 Intermediate-Mass Elements
As two of the most abundant metals, C and O are very important in stellar evolution and as diagnostics of GCE, but our views of O have undergone numerous revisions in the last decade. Measurements are presented in Figure 4, which isolates giants from dwarfs, and shows separately the abundances derived from the forbidden lines \[O I\], near-infrared triplet O I, and ultraviolet OH lines.
Barbuy (1988), studying the forbidden line in giants, found a result similar to that for $`\alpha `$-elements in the halo, i.e. an overabundance by 0.3โ0.4 dex irrespective of metallicity, shown as a dotted line in Figure 4. Results for the infra-red triplet, however, have been inconsistent. Abia & Rebolo (1989) found a progressive increase in the overabundance, approaching 1.5 dex at \[Fe/H\] = $`3`$. Subsequent studies found different results, arguing inconsistencies between different lines (Magain 1988), $`T_{\mathrm{eff}}`$ dependences, and errors in the $`T_{\mathrm{eff}}`$ scale and equivalent widths. The most recent studies used the UV OH lines, Israelian et al. (1998) and Boesgaard et al. (1999) finding almost identical trends towards high O overabundances in the most metal-poor stars. Note that although Bessell, Sutherland & Ruan (1991) concluded that their OH line measurements were consistent with Barbuyโs (1988) giants, Bessell et alโs dwarfs and Nissen et alโs (1994) almost fit the recent OH trend, albeit displaced by $`0.3`$ dex. Gratton (1999) warns of the difficulty of fitting the continuum near the UV OH lines for the more metal-rich stars, the need for accurate molecular line parameters, and the need to apply NLTE corrections for the high-excitation O I triplet. The lack of agreement between measurements of the triplet (Figure 4) attests to the difficulties in using these lines.
Barbuyโs finding that O behaves like the $`\alpha `$-elements was attractive under the paradigm that copious production of Fe in SN Ia was responsible for the decrease in \[$`\alpha `$/Fe\] at \[Fe/H\] $`>`$ $`1`$. However, O and the $`\alpha `$ elements are formed separately, so their yields relative to Fe do not have to exhibit the same dependence on metallicity. The difficulty posed by the observations is that current stellar models predict O and $`\alpha `$ to evolve together, and do not show high overabundances of \[O/Fe\]. The models of Timmes, Woosley & Weaver (1995), for example, achieve \[O/Fe\] = +0.4 at \[Fe/H\] $``$ $`3`$, or even +0.6 if the iron yield is halved, but they donโt reach 1.0 as the OH data do. Furthermore, it is not sufficient to claim that the Fe yields could be wrong, for while they could be wrong, that would not help the problem with O; the OH observations require that \[O/$`\alpha `$\] is also strongly dependent upon metallicity, and that is not seen in the models either. So, if the theoretical yields are to fit the OH line data, significantly higher O production will be required, irrespective of what changes are made to Fe.
Carbon and nitrogen measurements in very metal-poor stars (Israelian & Rebolo 1999) generally give approximately solar ratios. However, there have also been significant numbers of C-rich stars found at the lowest iron abundances (Beers, Preston & Shectman 1992). Possibly as many as 10% of stars with \[Fe/H\] $`<`$ $`2`$ have high CH-band strengths. While one of these, CS 22892-052, has huge r-process element overabundances (Sneden et al. 1994, 1996), the majority exhibit s-process patterns (Norris, Ryan & Beers 1997a; Barbuy et al. 1997; Norris et al. 2000). The high C abundances are not predicted by theoretical yields of supernovae (e.g. Timmes et al. 1995), but might be explained by enrichment of the early halo by mass loss from high mass Population III stars which have synthesized C via the triple-alpha process and mixed it to their surfaces (e.g. Marigo 1999). The frequency of C-rich stars and their tendency to be accompanied by s-process rather than r-process heavy element signatures (suggestive of AGB star evolution rather than supernovae) will be important to understanding their origin. (Note, however, that not all C-rich stars exhibit heavy element anomalies \[Norris, Ryan & Beers 1997b\]).
The $`\alpha `$-elements (Mg, Si, Ca) have fairly uniform overabundances relative to iron extending down to the lowest metallicities (Ryan et al. 1991; Norris et al. 1993; McWilliam et al. 1995; Ryan et al. 1996). Recent studies have begun to concentrate on star-to-star variations, with King (1997), Carney et al. (1997), and Nissen & Schuster (1997) identifying stars with low \[$`\alpha `$/Fe\] ratios, predominantly retrograde kinematics, and young ages. The low \[$`\alpha `$/Fe\] ratios are reminiscent of those proposed by Matteucci & Brocatto (1990) and Gilmore & Wyse (1991) for star formation in dwarf galaxies where star formation ceased before metallicities typical of Galactic disk stars were reached, so that the appearance of SN Ia would lead to low \[$`\alpha `$/Fe\] ratios even at metallicities typical of halo stars.<sup>4</sup><sup>4</sup>4A possible conflict is raised by the work of Kobayashi et al. (1998) who propose that the SN Ia mechanism would have lower efficiency at \[Fe/H\] $`<`$ $`1`$ due to the weaker winds in metal-deficient stars. Similarly, the dual halo models of Zinn (1993), based on globular clusters, and Norris (1994), based on field stars, fit with the characteristics of the stars now observed. Clearly there is still much to learn about the Galactic halo by exploiting relative abundances.
For \[Na/Fe\] and \[Mg/Fe\], Hanson et al. (1998) have found different behaviours for globular cluster and field stars, even though they have examined the same evolutionary states in each sample. They find the two elements to be positively correlated in the field, signifying common production, and find correlations with kinematics in the sense that the youngest stars (inferred from lower \[Na/Fe\] and \[Mg/Fe\] values) have predominantly retrograde kinematics. In clusters, on the other hand, they find overall an anti-correlation of \[Na/Fe\] and \[Mg/Fe\] suggestive of the proton-capture chain having been active (converting Ne to Na, and Mg to Al, during H-burning via the CNO cycle). The appearance at the stellar surface of these changes requires deep mixing, so one may conclude that deep mixing has occurred in the globular cluster giants but not in the field giants even at the same evolutionary state. Why this should be so has yet to be resolved; the Li results (ยง4) may be related.
## 6 The Iron-Peak Elements
The iron-peak elements most easily observed in very metal-poor stars are Sc, Ti, Cr, Mn, Fe, Co, and Ni. For stars with \[Fe/H\] $`>`$ $`2.5`$, most of these species exhibit solar abundance ratios. One exception is Mn, which is underabundant by $``$0.3 dex in halo stars. A second exception is Ti, but whether or not Ti should be included in the iron-peak group at all is unclear (e.g. Lambert 1987). While it is believed to be produced in the same region of stars as the other iron-peak elements, its abundance appears more like the $`\alpha `$-elements in that it is overabundant by $``$0.3โ0.4 dex in stars with \[Fe/H\] $`<`$ $`1`$.<sup>5</sup><sup>5</sup>5Truran (1999, private communication) reports that the experimental <sup>44</sup>Ti($`\alpha `$,p)<sup>47</sup>V rate is five times higher than the theoretical one used in previous nucleosynthesis calculations, so the theoretical yields of some iron-peak elements will be subject to revision. McWilliam et al. (1995) and Ryan et al. (1996) showed that although the Sc and Ti abundance trends persist to the lowest metallicities known, \[Fe/H\] = $`4.0`$, Mn and Cr become very underabundant in stars with \[Fe/H $`<`$ $`2.5`$, while in the same objects Co becomes overabundant. (See Figure 5.) There is also limited evidence for Ni overabundances in some of these objects. These changes were not predicted by supernova computations, and provide a recent example of observations leading theory in new directions.
The iron-peak species are synthesised deep in the star, close to mass-cut (the division between matter ejected from the supernova and that which collapses onto the stellar remnant). Supernova calculations are unable to eject material naturally from the physical conditions of the star, foreshadowing the difficulty of accurately predicting the yields of the iron-peak species which depend sensitively on the explosive conditions. The position of the mass cut cannot be predicted, but must be constrained by the observed abundances. Cr and Mn are produced only in a shell towards the outside of the Fe (and Ti) core, while Co and Ni are produced internal to that, so the resulting relative yields of iron-peak species can be adjusted by moving the mass-cut even slightly within the star (e.g. Nakamura et al. 1999).
Efforts to understand the abundance trends have focussed on the possible $`\alpha `$-rich freeze-out (Woosley & Hoffman 1992; McWilliam et al. 1995), the location of the mass-cut (Nakamura et al. 1999), and the dependence of yields on stellar mass, metallicity, and neutron excess (Nakamura et al. 1999; Hix & Thielemann 1996). The explanation must account for a handful of elements with different behaviours, whilst not distorting the \[X/Fe\] ratios of species such as Mg, Si, and Ca which are formed well outside the mass-cut. This is particularly challenging for explanations which alter the Fe yield! Progress may increase when 3D hydrodynamical models provide more realistic treatments of the supernova explosion (e.g. Mรผller, Fryxell & Arnett 1991). Observational evidence for mixing of material from the mass-cut to the photosphere appears to be growing (Stathakis 1996; Fassia et al. 1998), and provides an additional set of clues and constraints as we seek more realistic supernova models.
## 7 Neutron-Capture Elements
Spite & Spite (1978) showed from observations of Ba (of mixed s- and r-process origin) and Eu (essentially pure r-process) that the Ba in halo stars owed its origin primarily to the r-process. Truran (1981) provided a theoretical insight into the relative contributions of the processes by examining the roles of seed nuclei, and argued that the s-process should be viewed as a secondary process while the r-process was primary, and thus that first generation โ Population III โ stars would not execute the s-process. Gilroy et al. (1988) showed that the neutron-capture elements exhibited abundance patterns more consistent with the r-process than the s-process. McWilliam et al. (1995, 1998) extended the Ba and Eu comparison to the lowest metallicities (\[Fe/H\] = $`4`$), confirming the existence of r-process (rather than s-process) ratios.
An alternative means of examining the s- and r-process contributions would be via isotope ratios. Heavy element isotope lines are invariably blended in stellar spectra, but in a small number of cases they give rise to significant differences in the line profile allowing constraints to be placed on the isotope ratios. Magain (1995) obtained observations of this effect, and in HD 140283, at \[Fe/H\] = $`2.6`$, inferred an s-process rather than r-process pattern for Ba. This result was unexpected given the r-process framework that had been established over the preceeding nearly 20 years. In contrast, Gacquer & Francois (1998, private communication) find an r-process signature for this star.
The reliance almost solely on Ba and Eu as diagnostics of the s- vs r-process fractions has been diminished by HST data for other important neutron-capture elements having UV spectra. Included in this list are Ag, Pt, Os, and Pb (Crawford et al. 1998; Cowan et al. 1996; Sneden et al. 1998).
Gilroy et al. (1988) also found that neutron-capture elements showed significant star-to-star variations in the most metal poor objects (\[Fe/H\] $`{}_{}{}^{}{}_{}{}^{<}`$ $`2.5`$), building upon similar cases reported earlier during the decade. The abundance variations are greatest for Sr, where ranges of a factor of more than 100 were found amongst dwarfs (Ryan et al. 1991) and giants (Norris et al. 1993). Moreover, as Figure 6 shows, such extreme variations are not shared by Ba, which exhibits a fairly steady trend towards lower \[Ba/Fe\] at lower \[Fe/H\].
Reconciling Sr and Ba is a challenge. If the r-process alone is responsible for their synthesis in very metal-poor halo stars, then Figure 6 tells us that the r-process cannot be universal. Whilst there is no theoretical reason why it must be universal, halo star neutron-capture element abundance patterns generally resemble the r-process contribution in the sun (e.g. Gilroy et al. 1988; McWilliam et al. 1995; Cowan et al. 1995, Sneden et al. 1996). This cannot be ignored, even if non-uniqueness (Goriely & Arnould 1997) is possible. Hypothesising that the lower envelope to the \[Sr/Fe\] observations is the โnormalโ r-process behaviour, corresponding to a universal \[Ba/Sr\] r-process value, we seek a source of additional Sr in the lowest metallicity objects. Figure 6 emphasises that the process responsible must involve low neutron exposures that synthesise only species around the atomic number of Sr. Other species near Sr are also enhanced in these high Sr stars (Ryan et al. 1996, 2000; see Figure 7.) Whilst the weak s-process would produce primarily low atomic-numbered neutron-capture species, and would be active in normal high mass ($`>`$15 M) stars, the difficulties related to the lack of seed nuclei and a suitable neutron source would severely hamper this in low metallicity stars. It may be necessary to look for a new, low-neutron-exposure r-process (Ishimaru & Wanajo 1999b), reflecting two different types of core collapse supernovae.
GCE models need to reproduce the climb in \[Ba/Fe\] and \[Eu/Fe\] for stars with \[Fe/H\] $`<`$ $`2`$, which suggests that these species are not produced abundantly in the most massive metal-poor supernovae, for if they were then they would exhibit higher ratios (to iron) even in the lowest metallicity stars (e.g. Mathews & Cowan 1990). Mathews & Cowan associated their production with 10โ11 M stars. Although the evolution in their models is now seen to be too steep for todayโs data, the concept of low-mass stars being responsible has survived, and a more gradual emergence is seen in the models of Travaglio et al. (1999) which use a mass range 8โ10 M for the process. Travaglio et al. also treat the enrichment of the cloud inhomogeneously, along the lines described in ยง2 of this review (also Ishimaru & Wanajo 1999a).
The final comments on the neutron-capture elements are reserved for the atypical r-process-rich star, CS 22892-052. Assuming that the neutron-capture elements in this star are present in the same proportions as the r-process contribution in the sun, Cowan et al. (1997) derive an age of 17$`\pm `$4 Gyr, based on the radioactive Th abundance. This is consistent with current estimates of globular cluster ages from isochrone fitting, of around 12.8$`{}_{2.8}{}^{}{}_{}{}^{+4.2}`$ Gyr (Chaboyer 1999). Improved data and GCE models will hopefully reduce the errors to provide a chronometer of better accuracy than the globular cluster technique.
## 8 Final Comments and Summary
Stars with \[Fe/H\] $`<`$ $`3`$ formed at the earliest epochs of star formation, corresponding to redshifts $`z{}_{}{}^{>}4`$. They formed from gas clouds enriched by single supernovae, and hence allow us to investigate the first (Population III) stars which enriched primordial (big bang) material with heavy elements. Even though that stellar population was small in number and no surviving example has been detected today, it is still possible to investigate its mass function and evolution via the nucleosynthetic yields frozen into the next generation of stars, the extremely metal-poor Population II stars.
Extremely metal-poor stars exist in two main environments โ the field and globular clusters โ and the two appear not to be identical. Building on previous differences (e.g. CN variations which appear more common in globular cluster than field stars), we now see a spread of Li in M92 and evidence of deep mixing in globular cluster giants (via the proton-chain signature) which do not appear in the field. These may indicate that the clusters in which todayโs field stars formed were not the same as the dense globular clusters that survive to the present. In using Population II objects to study GCE, it would then seem safer to rely on field stars which sample a much larger volume of the Galaxy and are less likely to have interacted with other objects in a dense environment, rather than trusting objects which sample the chemical evolution of a small region of space with high stellar density.
Recent measurements have highlighted the state of confusion which exists over oxygen abundances in halo stars. Meanwhile, some elements have shown remarkably robust abundance patterns as more metal-deficient stars are sampled, such as Ti (though we know not why!). Greater accuracy has enabled star-to-star abundance variations to be examined in intermediate atomic-mass elements, shedding new light on the chemical and kinematic composition of the Galaxy, and its evolution. Likewise, the huge abundance spread in Sr, but not Ba, is allowing us to probe the nucleosynthesis processes and ultimately the mass function of the first stars. These advances accompany more realistic GCE models that incorporate the stochastic nature of star formation which is potentially of great importance in the earliest epochs where small numbers of supernovae were involved.
## References
> Abia, C. & Rebolo, R. 1989, ApJ, 347, 186
>
> Alonso, A., Arribas, S. & MartรญnezโRoger, C. 1996, A&AS, 117, 227
>
> Anstee, S. D. & OโMara, B. J. 1995, MNRAS, 276, 859
>
> Anstee, S. D., OโMara, B. J. & Ross, J. E. 1997, MNRAS, 284, 202
>
> Argast, D. & Samland, M. 1999, The Galactic Halo: from Globular Clusters to Field Stars, 35$`^{\text{th}}`$ Liรจge Int. Astroph. Coll., in press
>
> Asplund, M., Nordlund, ร
., Trampedach, R. & Stein, R. F. 1999, astro-ph/9905059
>
> Audouze, J. & Silk, J. 1995, ApJ, 451, L49
>
> Barbuy, B. 1988, A&A, 191, 121
>
> Barbuy, B. Cayrel, R., Spite, M., Beers, T. C., Spite, F., Nordstrom, B. & Nissen, P. E. 1997, A&A, 317, L63
>
> Barbuy, B., Spite, F., & Spite, M. 1985, A&A, 144, 343
>
> Barklem, P. S. & OโMara, B. J. 1997, MNRAS, 290, 102
>
> Barklem, P. S. & OโMara, B. J. 1998, MNRAS, 300, 863
>
> Barklem, P. S., OโMara, B. J. & Ross, J. E. 1998, MNRAS, 296, 1057
>
> Baumรผller, D., Butler, K. & Gehren, T. 1998, A&A, 338, 637
>
> Baumรผller, D. & Gehren, T. 1997, A&A, 325, 1088
>
> Beers, T. C., Preston, G. W. & Shectman, S. A. 1992, AJ, 103, 1987
>
> Bessell, M. S. & Norris, J. E. 1984, ApJ, 285, 622
>
> Bessell, M. S., Sutherland, R. S. & Ruan, K. 1991, ApJ, 383, L71
>
> Boesgaard, A. M., Deliyannis, C. P., Stephens, A. & King, J. R. 1998, ApJ, 493, 206
>
> Boesgaard, A. M., King, J. R., Deliyannis, C. P. & Vogt, S. S. 1999, AJ, 117, 492
>
> Bouwens, R., Broadhurst, T., & Silk, J. 1998, ApJ, 506, 557
>
> Carlsson, M., Rutten, R. J., Bruls, J. H. M. J. & Shchukina, N. G. 1994, A&A, 288, 860
>
> Carney, B. W. & Peterson, R. C. 1981, ApJ, 245, 238
>
> Carney, B. W., Wright, J. S., Sneden, C., Laird, J. B., Aguilar, L. A. & Latham, D. W. 1997, AJ, 114, 363
>
> Cavallo, R. M., Pilachowski, C. A. & Rebolo, R. 1997, PASP, 109, 226
>
> Chaboyer, B. 1999, The Galactic Halo: from Globular Clusters to Field Stars, 35$`^{\text{th}}`$ Liรจge Int. Astroph. Coll., in press
>
> Cioffi, D. F., McKee, C. F. & Bertschinger, E. 1988, ApJ, 334, 252
>
> Cowan, J. J., Burris, D. L., Sneden, C., McWilliam, A. & Preston, G. W. 1995, ApJ, 439, L51
>
> Cowan, J. J., McWilliam, A., Sneden, C., & Burris, D. L. 1997, ApJ, 480, 246
>
> Cowan, J. J., Sneden, C., Truran, J. W. & Burris, D. L. 1996, ApJ, 460, L115
>
> Crawford, J. L., Sneden, C., King, J. R., Boesgaard, A. M. & Deliyannis, C. P. 1998, AJ, 116, 2489
>
> Edmunds, M. G. & Phillipps, S. 1997, MNRAS, 292, 733
>
> Edvardsson, B., Andersen, J., Gustafsson, B., Lambert, D. L., Nissen, P. E. & Tomkin, J. 1993, A&A, 275, 101
>
> Fassia, A., Meikle, W. P. S., Geballe, T. R., Walton, N. A., Pollaco, D. L., Rutten, R. G. M. & Tinney, C. 1998, MNRAS, 299, 150
>
> Garcรญa Lรณpez, R. J., Severino, G. & Gomez, M. T. 1995, A&A, 297, 787
>
> Gilmore, G. & Wyse, R. F. G. 1991, ApJ, 367, L55
>
> Gilroy, K. K., Sneden, C., Pilachowski, C. A. & Cowan, J. J. 1988, ApJ, 327, 298
>
> Goriely, S. & Arnould, M. 1997, A&A, 322, L29
>
> Gratton, R. G. 1989, A&A, 208, 171
>
> Gratton, R. 1999, The Galactic Halo: from Globular Clusters to Field Stars, 35$`^{\text{th}}`$ Liรจge Int. Astroph. Coll., in press
>
> Gratton, R. G. & Sneden, C. 1987, A&A, 178, 179
>
> Gratton, R. G. & Sneden, C. 1988, A&A, 204, 193
>
> Gratton, R. & Sneden, C. 1994, A&A, 287, 927
>
> Hanson, R. B., Sneden, C., Kraft, R. P. & Fulbright, J. 1998, AJ, 116, 1286
>
> Hellsten, U., Dave, R., Hernquist, L., Weinberg, D. H. & Katz, N. 1997, ApJ, 487, 482
>
> Hix, W. R. & Thielemann, K.-F. , 1996, ApJ, 460, 869
>
> Ishimaru, Y. & Wanajo, S. 1999a, ApJ, 511, L33
>
> Ishimaru, Y. & Wanajo, S. 1999b, MPA/ESO Workshop on the First Stars, Garching 1999, in press.
>
> Israelian, G., Garcรญa Lรณpez, R. & Rebolo, R. 1998, ApJ, 507, 805
>
> Israelian, G. & Rebolo, R. 1999, The First Stars, MPA/ESO Workshop, Garching August 1999, in press
>
> King, J. R. 1993, AJ, 106, 1206
>
> King, J. R. 1997, AJ, 113, 2302
>
> Kiselman, D. 1991, A&A, 245, L9
>
> Kiselman, D. 1994, A&A, 286, 169
>
> Kiselman, D. & Carlsson, M. 1996, A&A, 311, 680
>
> Kobayashi, C., Tsuijmoto, T., Nomoto, K. Hachisu, I. & Kato, M. 1998, ApJ, 503, L155
>
> Laird, J. B., Carney, B. W. & Latham, D. W. 1988b, AJ, 95, 1843
>
> Laird, J. B., Rupen, M. P., Carney, B. W. & Latham, D. W. 1988a, AJ, 96, 1908
>
> Lambert, D. L. 1987, JApA, 8, 103
>
> Lu, L., Sargent, W. L. W., Barlow, T. A., Churchill, C. W. & Vogt, S. S. 1996, ApJS, 107, 475
>
> Magain, P. 1988, The Impact of Very High S/N Spectroscopy on Stellar Physics, ed. G. Cayrel de Strobel & M. Spite (Dordrecht, Kluwer), 485
>
> Magain, P. 1989, A&A, 209, 211
>
> Magain, P. 1995, A&A, 297, 686
>
> Marigo, P. 1999, MPA/ESO Workshop on The First Stars, Garching August 1999, in press
>
> Mashonkina, L., Gehren, T. & Bikmaev, I. 1999, A&A, preprint
>
> Mathews, G. J. & Cowan, J. J. 1990, Nature, 345, 491
>
> Matteucci, F. & Brocato, E. 1990, ApJ, 365, 539
>
> McWilliam, A., 1998, AJ, 115, 1640
>
> McWilliam, A., Preston, G. W., Sneden, C. & Searle, L. 1995, AJ, 109, 2757
>
> Molaro, P. & Bonifacio, P. 1990, A&A, 236, L5
>
> Molaro, P. & Castelli, F. 1990, A&A, 228, 426
>
> Molaro, P., DโOdorico, S., Fontana, A., Savaglio, S. & Vladilo, G. 1996, A&A, 308, 1
>
> Mรผller, E., Fryxell, B. & Arnett, D. 1991, A&A, 251, 505
>
> Nakamura, T., Umeda, Nomoto, K., Thielemann, F.-K. & Burrows, I. 1999, ApJ, in press (astro-ph/9809307)
>
> Nissen, P. E., Gustafsson, B., Edvardsson, B., & Gilmore, G. 1994, A&A, 285, 440
>
> Nissen, P. E. & Schuster, W. M. 1997, A&A, 326, 751
>
> Norris, J. E. 1994, ApJ, 431, 645
>
> Norris, J. E., Peterson, R. C. & Beers, T. C. 1993, ApJ, 415, 797
>
> Norris, J. E., Ryan, S. G. & Beers, T. C. 1997a, ApJ, 488, 350
>
> Norris, J. E., Ryan, S. G. & Beers, T. C. 1997b, ApJ, 489, L169
>
> Norris, J. E., Ryan, S. G., Beers, T. C. & Aoki, W. 2000, in prep.
>
> Nulsen, P. E. J., Barcons, X. & Fabian, A. C. 1998, MNRAS, 301, 168
>
> Pagel, B. E. J. & Patchett, B. E. 1975, MNRAS, 172, 13
>
> Pavlenko, Ya. V. & Magazzu, A. 1996, A&A, 311, 961
>
> Peterson, R. 1981, ApJ, 244, 989
>
> Peterson, R., Kurucz, R. L., & Carney, B. W. 1990, ApJ, 350, 173
>
> Pettini, M., Boksenberg, A. & Hunstead, R. W. 1990, ApJ, 348, 48
>
> Pettini, M., Ellison, S. L., Steidel, C. C. & Bowen, D. V. 1999a, ApJ, 510, 576
>
> Pettini, M., Smith, L. J., King, D. L. & Hunstead, R. W. 1997, ApJ, 486, 665
>
> Pettini, M., Steidel, C. C., Adelberger, K. L., Dickinson, M. & Giavalisco, M. 1999b, preprint
>
> Primas, F., Molaro, P., & Castelli, F. 1994, A&A, 290, 885
>
> Ryan, S. G. 1996, ASP Conf.Ser. 92, 113
>
> Ryan, S. G. 1998, A&A, 331, 1051
>
> Ryan, S. G., Blake, L., Norris, J. E. & Beers, T. C. 2000, in preparation
>
> Ryan, S. G. & Norris, J. E. 1991, AJ, 101, 1865
>
> Ryan, S. G., Norris, J. E. & Bessell, M. S. 1991, AJ, 102, 303
>
> Ryan, S. G., Norris, J. E., & Beers, T. C. 1996, ApJ, 471, 254
>
> Ryan, S. G., Norris, J. E. & Beers, T. C. 1999, ApJ, 523, 654
>
> Sadler, E. M., Rich, R. M. & Terndrup, D. M. 1996, AJ 112, 171
>
> Searle, L. 1977 In: Tinsley, B. M. & Larson, R. B. (eds) The Evolution of Galaxies and Stellar Populations, Yale Uni. Obs., New Haven, p.219
>
> Searle, L. & Sargent, W. L. W. 1972, ApJ, 173, 25
>
> Shigeyama, T. & Tsujimoto, T. 1998, ApJ, 507, L135
>
> Sneden, C., Cowan, J. J., Burris, D. L. & Truran, J. W. 1998, ApJ, 496, 235
>
> Sneden, C., Preston, G. W., McWilliam, A. & Searle, L. 1994, ApJ, 431, L27
>
> Sneden, C., McWilliam, A., Preston, G. P., Cowan, J. J., Burris, D. L. & Armosky, B. J., 1996, ApJ, 467, 819
>
> Spite, M. & Spite, F. 1978, A&A, 67, 23
>
> Spite, M. & Spite, F. 1991, A&A, 252, 689
>
> Stathakis, R. A. 1996, PhD Thesis, Sydney Uni.
>
> Steidel, C. C., Adelberger, K. L., Giavalisco, M., Dickinson, M. & Pettini, M., 1999, astro-ph/9811399
>
> Timmes, F. X., Woosley, S. E., & Weaver, T. A. 1995, ApJS, 98, 617
>
> Tomkin, J., Lemke, M., Lambert, D. L. & Sneden, C. 1992, AJ, 104, 1568
>
> Travaglio, C., Galli, D., Gallino, R., Busso, M., Ferrini, F. & Straniero, O. 1999, ApJ, 521, 691
>
> Truran, J. W. 1981, A&A, 97, 391
>
> Tsujimoto, T. & Shigeyama, T. 1998, ApJ, 508, L151
>
> Warren, S. J., Hewett, P. C. & Osmer, P. S. 1994, ApJ, 421, 412
>
> Wolfe, A. M., Turnshek, D. A., Smith, H. E. & Cohen, R. D. 1986, ApJS, 61, 249
>
> Woosley, S. E. & Hoffman, R. D. 1992, ApJ, 395, 202
>
> Zhao, G., Butler, K. & Gehren, T. 1998, A&A, 333, 219
>
> Zinn, R. 1993, ASP.Conf.Ser, 48, 38 |
warning/0001/astro-ph0001283.html | ar5iv | text | # Induction of supernova-like explosions by ๐พ-ray bursts in close binary systems
## 1 Introduction
The prompt localization of gamma-ray bursts (GRBs) by BeppoSAX led to the discovery of long lived GRB afterglows spanning the energy range from X-ray to radio (Costa et al. (1997); van Paradijs et al. (1997); Frail et al. (1997)), and of associated host galaxies (Kulkarni et al. 1998a ). Detections of absorption and emission features at high redshifts ($`0.69z3.42`$) in optical afterglows of GRBs and in their host galaxies (e.g., Kulkarni et al. (1999)) have clearly tipped the scale in favour of the cosmological origin of GRBs sources (Usov & Chibisov (1975); van den Berg (1983); Paczyลski (1986); Goodman (1986); Eichler et al. (1989)).
Despite such great advances the exact nature of the GRB progenitors is still unknown. Several currently popular models posit as the energy-releasing event: coalescence of two neutron stars (Blinnikov et al. (1984); Paczyลski (1986); Eichler et al. (1989)), the collapse of a massive star (Woosley (1993); Paczyลski (1998)); or the formation of a millisecond pulsar with extremely strong magnetic field $`(10^{15}10^{16}`$ G) (Usov (1992); Thompson & Duncan (1993); Blackman, Yi, & Field (1996); Katz (1997); Kluลบniak & Ruderman (1998); Vietri & Stella (1998)).
The energetic ejecta of GRBs may affect their close surrounding in various observable ways. For example, the reprocessing of some of the GRB energy in the atmosphere of a companion might produce an optical afterglow, as discussed, e.g., by London & Cominsky (1983) and by Melia, Rappaport, & Joss (1986) for galactical GRBs, and recently by Blinnikov & Postnov (1998) for GRBs at cosmological distances. In this Letter, we discuss another possible interaction of the GRB ejecta with a companion: Some GRB explosions might occur in a close binary companionship with a white dwarf (WD). This, in fact, is a prerequisite in the GRB model of Usov (1992). The interaction of the ejecta with the companion then may induce its explosion and appearance of a supernova-like phenomenon. The recently claimed supernovae in association with GRBs (e.g., Wheeler (1999)) are unlikely to have been produced in this way. In ยง2 we consider the process of induced explosion, and estimate the parameter values needed to actuate it. In ยง3 we discuss possible observational consequences and other pertinent issues.
## 2 Induction of white-dwarf explosions by GRBs in close binaries
The observed fluxes of GRBs at cosmological distances imply total radiation energy output per unit solid angle within the beam of radiation of $`Q_\gamma 10^{51}10^{52}`$ ergs/st on a time scale of seconds. The total angular energy output, $`Q`$, is even larger, as only part of it is converted into the observed radiation. If the GRB source has a close binary companion that lies within the main beam, a powerful flux of energy impacts the latter. On near enough a companion the effects may be staggering. We consider such possible effects, and, in particular, the possibility that the impact can induce a supernova-like explosion of a white-dwarf companion. The energy that falls on a unit area of the surface of the secondary that is within the GRB beam is $`qQ/D^2`$, where $`D=10^{10}D_{10}`$ cm is the binary distance. The chances of an induced explosion are maximized when the companion is fully within the GRB beam, as we assume in our discussion below. In this case the total energy that hits the companion is $`\mathrm{\Delta }Q\pi R_s^2q`$, where $`R_s=10^9R_9`$ cm is the radius of the companion.
Our proposal applies more generally, but, for the sake of concreteness, we consider system parameters that are natural in the GRB model that involves a strongly magnetized millisecond pulsar produced by accretion-induced collapse of a white dwarf in a close binary (Usov (1992)). An inherent feature of this model is that GRBs occur in binaries. The GRB progenitor is a strongly magnetized white dwarf with a mass near the Chandrasekhar limit, $`1.4M_{}`$. The secondary is a white dwarf with a mass $`M_s0.30.5M_{}`$, and fills its Roche lobe. For such a binary with $`M_s0.5M_{}`$, we have $`R_91`$, $`D_{10}0.7`$ and $`\mathrm{\Delta }Q5\times 10^{50}Q_{52}`$, where $`Q_{52}=Q/10^{52}`$ ergs/st. This can easily be more than the binding energy of the secondary ($`GM_s^2/2R_s3\times 10^{49}`$ ergs) (e.g., Shapiro & Teukolsky (1983); Nomoto (1982)). Therefore, for strong GRBs the energy $`\mathrm{\Delta }Q`$ suffices to completely evaporate the secondary. This is still true for a $`1.4M_{}`$ companion.
The interaction between the relativistic GRB wind and the secondary is very complicated; it depends, among other factors, on the properties of the winds. For the GRB model involving a strongly magnetized, millisecond pulsar, the outflowing wind is dominated by a Poynting flux. The luminosity in electron-positron pairs and radiation is only $`10^2`$ of the Poynting luminosity (e.g., Usov (1994)). The Lorentz factor of the wind is $`10^210^3`$. Typically, the diameter of the secondary white dwarf is rather smaller than the thickness of the wind shell which is $`c\tau `$, where $`\tau 110`$ s is the characteristic time of deceleration of the pulsar rotation due to the action of the electromagnetic torque; this is roughly the GRB duration or less. The action of the relativistic, strongly magnetized wind on the secondary may be roughly modeled by assuming that the external pressure on its surface facing the GRB increases instantly to
$$P_{\mathrm{ext}}\frac{Q}{D^2c\tau }3\times 10^{21}Q_{52}D_{10}^2\tau _1^1\mathrm{ergs}\mathrm{cm}^3$$
(1)
for $`\tau _1=\tau /1`$ s $`110`$.
Inside white dwarfs, electrons are free and strongly degenerated (except for a very thin surface layer with the density $`\rho 10^2`$ g cm<sup>-3</sup>). At high density and low temperature, these electrons give the main contribution to the gas pressure irrespective of the element abundances. The equation of state is (e.g., Shapiro & Teukolsky (1983))
$$P_e(\rho )10^{23}\times \{\begin{array}{ccc}(\rho _6/\mu _e)^{5/3}\mathrm{ergs}\mathrm{cm}^3,\hfill & & \mathrm{for}\mathrm{\hspace{0.17em}\hspace{0.17em}10}^4\rho _6<\mu _e,\hfill \\ (\rho _6/\mu _e)^{4/3}\mathrm{ergs}\mathrm{cm}^3,\hfill & & \mathrm{for}\rho _6>\mu _e,\hfill \end{array}$$
(2)
where $`\rho _6=\rho /10^6\mathrm{g}\mathrm{cm}^3`$, and $`\mu _e`$ is the mean molecular weight per electron. In white dwarfs helium, carbon, and oxygen dominate, so $`\mu _e`$ is nearly 2. The upper limit on the density for the validity of equation (2) is determined by neutronization, and, for example, for helium it is $`\rho _610^5`$.
The instantaneous increase of external pressure from zero to $`P_{\mathrm{ext}}`$ results in formation of a strong shock that propagates into the high density region. Equations (1) and (2) imply that when the density in front of the shock is
$$\rho \stackrel{~}{\rho }10^5Q_{52}^{3/5}D_{10}^{6/5}\tau _1^{3/5}\mathrm{g}\mathrm{cm}^3,$$
(3)
the pressure may be neglected. In this case, the temperature behind the shock is about
$$T_9=T/10^9\mathrm{K}Q_{52}^{1/4}D_{10}^{1/2}\tau _1^{1/4}.$$
(4)
For helium white dwarfs, this temperature may be higher than the ignition temperature which varies from about $`6\times 10^8`$ K at $`\rho 10^5`$ g cm<sup>-3</sup> to $`10^8`$ K at $`\rho 10^9`$ g cm<sup>-3</sup> (e.g., Nomoto (1982)). In this case, the nuclear energy of the shocked matter is released within a dynamical time scale, i.e., almost instantaneously. At $`\rho >\stackrel{~}{\rho }10^5`$ g cm<sup>-3</sup>, the process of thermonuclear burning propagates in the white dwarf either as a supersonic detonation wave or as a subsonic deflagration wave (Khokhlov, Mรผller, & Hรถflich (1993) and references therein). It is worth noting that a transition from a deflagration to a detonation is possible in the process of burning propagation (e.g., Khokhlov, Oran, & Wheeler (1997)).
For a helium secondary of reasonable mass (not too close to the Chandrasekhar limit), the nuclear energy is enough for its complete disruption. The detonation of such explosions is then similar in some respects to that of Type I supernovae (see below).
The temperature given by equation (4) is at least a few times smaller than the ignition temperature for carbon-oxygen mixtures at $`\rho 10^510^6`$ g cm<sup>-3</sup> (e.g., Nomoto (1982)). However, an explosion might still occur for a carbon-oxygen WD if there is enough amplification of the inward shock due to the converging geometry of the phenomenon. For an exactly spherical geometry the amplification is very large (e.g., Zeldovich & Raizer (1969)), but ours is only a semi-spherical implosion.
The equation of state for matter of hot white dwarfs is (e.g., Cox & Giuli (1968); Shapiro & Teukolsky (1983))
$$P_e(\rho ,T)=P_e(\rho )+\mathrm{\Delta }P_e(\rho ,T),$$
(5)
where $`P_e(\rho )`$ is the pressure of completely degenerated electrons, given by equation (2), and $`\mathrm{\Delta }P_e(\rho ,T)`$ is the thermal part of the electron pressure $`P_e(\rho ,T)`$. For $`\rho >10^6\mu _e`$ g cm<sup>-3</sup>, we have
$$\mathrm{\Delta }P_e(\rho ,T)=2\left(\frac{\pi }{3}\right)^{2/3}\left(\frac{kT}{c\mathrm{}}\right)^2\left(\frac{m_p\mu _e}{\rho }\right)^{2/3}P_e(\rho ),$$
(6)
where $`k`$ is the Boltzmann constant, $`c`$ is the speed of light, $`\mathrm{}`$ is the Planck constant, and $`m_p`$ is the proton mass.
For $`\mu _e=2`$, equations (2) and (6) yield
$$\mathrm{\Delta }P_e(\rho ,T)4\times 10^{22}\rho _6^{2/3}T_9^2\mathrm{ergs}\mathrm{cm}^3.$$
(7)
Without detailed calculations similar, e.g., to the two-dimensional hydrodynamic simulations of supernova models by Livne & Arnett (1995) and Livne (1999) we cannot tell whether, when the disturbance reaches the center of the white dwarf, the temperature there is raised high enough for burning to occur there. We do not know how effective the shock convergence may be in amplifying the shock. We just parameterize the convergence effect by $`\alpha `$, the ratio of the thermal, electron pressure reached at the center, $`\mathrm{\Delta }P_e(\rho ,T)`$, to that produced by the impact on the surface, $`P_{\mathrm{ext}}`$. Using equations (1) and (7) we then have
$$T_{c,9}\left(\frac{\alpha }{10}\right)^{1/2}Q_{52}^{1/2}D_{10}^1\tau _1^{1/2}\rho _{c,6}^{1/3},$$
(8)
where $`\rho _c`$ is the density at the stellar center. For $`Q_{52}3`$, $`D_{10}0.7`$, $`\tau _11`$ and $`\rho _{c,6}10`$, from equation (8) we can see that for $`\alpha >20`$ the temperature $`T_c`$ is higher than the ignition temperature which is $`1.7\times 10^9`$ K for carbon-oxygen mixtures at the density of $`10^7`$ g cm<sup>-3</sup> (e.g., Nomoto (1982)). In this case explosion is expected to occur at the center.
Since detonation waves in WD matter have a finite width it is important to compare this with the size of the star. In carbon-oxygen WD matter the detonation wave has roughly three spatially separated zones. The foremost is a sharp shock. The shock compresses and heats the material behind it, and carbon burning can start. This reaches a peak energy output within some distance $`\mathrm{\Delta }l__\mathrm{C}`$, which typifies the carbon-burning zone. This scale roughly equals to the speed of propagation of the shock multiplied by the carbon burning time. The third zone is the region where matter is incinerated into nuclear-statistical-equilibrium (NSE) composition. The energy release in this layer is rather small, so it is not so important in the dynamics of the detonation wave, but it is important in determining the composition of the ashes and the subsequent appearance of the explosion remnant.
For carbon-oxygen mixtures, the width of NSE-relaxation layer is many orders larger than the width of carbon burning, $`\mathrm{\Delta }l_{_{\mathrm{NSE}}}\mathrm{\Delta }l__\mathrm{C}`$. For $`\rho _610`$, we have $`\mathrm{\Delta }l__\mathrm{C}1`$ cm and $`\mathrm{\Delta }l_{_{\mathrm{NSE}}}10^8`$ cm (e.g., Khokhlov (1989)). Since $`\mathrm{\Delta }l__\mathrm{C}R_s`$, the detonation wave may form in the vicinity of the stellar center and propagates to the surface. In the process of propagation of the outward detonation wave, the width of carbon burning is small, $`\mathrm{\Delta }l__\mathrm{C}R_s`$, except of a rather thin surface layer of the white dwarf, and therefore most of the nuclear energy is released by the time the detonation wave reaches the surface. As noted above, the nuclear energy released is enough for the white dwarf to be completely disrupted.
The kinetics of helium burning differs substantially from that of carbon-oxygen burning. The leading reaction is $`3^4`$He $`^{12}`$ C. At the same densities, the rate of this reaction is much smaller than that of carbon burning. As a result the width of the nuclear-burning zone is many orders larger (e.g., Khokhlov (1989)). Still, for $`\rho _60.11`$ the width of the helium-burning zone is small enough, $`\mathrm{\Delta }l_{_{\mathrm{He}}}10^8`$ cm $`R_s`$, so that we can take this zone as small compared with the other relevant scales. Thus, an inward detonation wave can lead to the white dwarf explosion as we discussed above.
For rather massive white dwarfs, $`M_s0.8M_{}`$, which can undergo accretion-induced explosions, the mean density of the bulk matter is $`10^6`$ g cm<sup>-3</sup>, and the NSE relaxation width is small, $`\mathrm{\Delta }l_{_{\mathrm{NSE}}}R_s`$. In this case, incineration is effective, and the remaining ashes consist mostly of <sup>56</sup>Ni, which decays and provides energy for the long-time radiation of Type I supernovae (e.g. Nomoto (1982); Woosley, Taam, & Weaver (1986)). In contrast, for low-mass white dwarfs, $`M_s0.3M_{}`$, which cannot be exploded by gas accretion without significant increase of their masses, the great fraction of the mass is at lower densities ($`\rho 10^5`$ g cm<sup>-3</sup>) for which $`\mathrm{\Delta }l_{_{\mathrm{NSE}}}R_s`$ and the production of <sup>56</sup>Ni is strongly suppressed irrespective the abundance (Khokhlov (1989); Nomoto (1982); Woosley, Taam, & Weaver (1986)). Therefore, for such a low-mass white dwarf the mass of <sup>56</sup>Ni that is produced in the GRB-induced explosion is very low, and this explosion leads to a weak supernova-like phenomenon, which differs qualitatively from known supernovae. Observation of such a phenomenon correlated with GRBs could confirm our idea on the GRB-induced explosions of secondary white dwarfs.
## 3 Discussion
It is generally believed that Type I supernovae are produced by thermonuclear explosions of white dwarfs (e.g., Nomoto (1982); Woosley, Taam, & Weaver (1986); Niemeyer & Woosley (1997)). Such explosions may be brought about by accretion of matter onto the white dwarfs. In the process, thermonuclear burning that is triggered near the surface propagates either in the form of a supersonic detonation wave, or as a subsonic deflagration. This results in the incineration of most of the white dwarf matter into <sup>56</sup>Ni, which is ejected from the star. The radioactive decay <sup>56</sup>Ni $``$ <sup>56</sup>Co $``$ <sup>56</sup>Fe can provide a sufficient amount of late-time energy input to power the light curves of Type I supernovae.
In this paper we have argued that a similar fate may befall a white dwarf that is exposed to the ejecta of a GRB explosion in a very close binary companion. The GRB angular energy output, $`Q`$, near the binary plane which is necessary for the WD explosion is $`10^{52}`$ ergs/st. This is a typical value of $`Q`$ in the model of GRBs we considered. Indeed, the rotational energy of millisecond pulsars that is a plausible source of energy for GRBs may be as high as a few $`\times 10^{53}`$ ergs, and almost all this energy may be transformed into the energy of a relativistic, strongly magnetized wind (e.g., Usov (1994)). The angular distribution of the wind flux depends on the angle $`\vartheta `$ between the rotational and magnetic axes and varies within a factor of 2-3 or so. Such a moderate collimation of the outflowing wind may be either along the rotational axes of the pulsar at $`\vartheta 0`$ (Benford (1984); Michel (1985)) or near the equator at $`\vartheta \pi /2`$ (Belinsky et al. (1994)). For a very close WD + WD binary that is the predecessor of the GRB source, one expects the secondary white dwarf to be near the equator of the millisecond pulsar which forms by accretion-induced collapse of the primary white dwarf. In this case, the $`Q`$ value in the WD direction is typically $`10^{52}10^{53}`$ ergs/st, the maximum value of it being reached when both the pulsar rotation is extremely fast and the magnetic axis of the pulsar is perpendicular to its rotational axis.
We suggest that the resulting explosion is similar in some respects to Type I supernova, but may differ substantially in others, especially if the mass of the WD is small, $`M_s0.3M_{}`$. First, because the trigger mechanism is different, a different elemental abundance may result. Second, the post-explosion WD remnant has ample time ($`10`$ s or more) to interact with the relativistic wind outflowing from the GRB source. This can lead to additional acceleration of the explosion debris even, for some parts, up to relativistic velocities. Therefore, for GRB-induced supernovae the maximum of their light curves is expected to be observed substantially earlier than for typical Type I supernovae. Taking also into account that typically the amount of radioactive <sup>56</sup>Ni produced in GRB-induced supernovae is low, the luminosities of these supernovae may decrease fast after the maximum without long lived tails.
At present, several SN/GRB associations have been suggested (for a review, see Wheeler (1999)). Among them, SN 1998bw, possibly associated with GRB 980425 (Galama et al. (1998)), is the most famous and best established candidate for such an association. SN 1998bw that was very powerful could not have been produced by our mechanism, which produces rather weak optical supernovae. The amount of radioactive <sup>56</sup>Ni produced in SN 1998bw has been estimated to be $`0.50.75M_{}`$ (e.g., Iwamoto et al. (1998); Woosley, Eastman, & Schmidt (1999)), much more than the explosion we discuss can make. Also, if GRB 980425 is connected with SN 1998bw its total energy released, even if it is isotropic, is only $`10^{48}`$ ergs. This is about four orders less than what is necessary for induction of a WD explosion. And third, the optical properties of SN 1998bw indicate that the progenitor star (like the progenitor stars of all other supernovae possibly associated with GRBs) was a massive star with a mass at least a several times larger that the maximum possible mass of WDs (Iwamoto et al. (1998); Woosley, Eastman, & Schmidt (1999)). While the observations of GRB 980425-SN 1998bw may be explained fairly well in the collapsar model (e.g., MacFadyen & Woosley (1999)), we suggest that at least some cosmological GRBs may be associated with rather weak supernova-like explosions of low-mass $`(0.30.5M_{})`$ white dwarfs. In our scenario, GRBs and SNs associated with each other are different phenomena while in the collapsar model the two events are one.
Recently, possible evidence for the existence of iron K-shell emission lines has been found in two GRBs: GRB 970508 and GRB 970828 (Piro et al. (1999); Yoshida et al. (1999)). The presence of dense matter very close ($`10^{16}`$ cm) to the GRBs, which is not expanding relativistically, is required by these observations. The remnants of the GRB-induced explosions of white dwarfs may be responsible for emission of the iron lines. For this, it is necessary that a small fraction of the remnant matter with iron mass of $`10^510^4M_{}`$ is accelerated by the GRB wind to subrelativistic velocities and generates the Fe line emission at the distance of $`(13)\times 10^{15}`$ cm from the GRB source. It is worth noting that rather strong, high-redshift ($`z1`$) GRBs, like GRB 970828, with X-ray afterglows (for their prompt localization) and without standard optical afterglow are the best candidates for searching the possible, weak, supernova-like explosions posited here.
Induction of supernova-like explosions by GRBs is similar in many respects to ablation in laser and heavy-ion fusion (for review, see Meyer-ter-Vehn, Atzeni, & Ramis (1998)). It is well-known that the symmetry of the irradiation of the fusion fuel is crucial for its successful explosion. Otherwise, only the outer layers may be affected. The same may be true for carbon-oxygen WDs if the driving external pressure $`P_{\mathrm{ext}}`$ is not spherical enough. Other obstacles to successful explosion might result from numerous instabilities that may develop at the surface. In our case, though, plasma instabilities at the surface where the GRB wind interacts with the WD matter may be suppressed by a very strong magnetic field of the GRB wind.
We thank an anonymous referee for useful suggestions. This research was supported by the MINERVA Foundation, Munich, Germany. |
warning/0001/math0001014.html | ar5iv | text | # Theorem 0.1
## 1 A $`K`$-theory version of the equivariant family index theorem
In this section, we recall a $`K`$-theory version of the equivariant family index theorem \[LiuMaZ, Theorem 1.2\] for $`S^1`$-actions, which will play a crucial role in the following sections.
This section is organized as follows: In Section 1.1, we recall the $`K`$-theory version of the equivariant family index theorem for $`S^1`$-actions on a family of Spin<sup>c</sup> manifolds. In Section 1.2, as a simple application of Theorem 1.1, we obtain a $`K`$-theory version of the vanishing theorem of Hattori \[Ha\] for the case of almost complex manifolds.
### 1.1 A $`K`$-theory version of the equivariant family index theorem
Let $`M,B`$ be two compact manifolds, let $`\pi :MB`$ be a fibration with compact fibre $`X`$ such that $`dimX=2l`$ and that $`S^1`$ acts fiberwise on $`M`$. Let $`h^{TX}`$ be a metric on $`TX`$. We assume that $`TX`$ is oriented. Let $`(W,h^W)`$ be a Hermitian complex vector bundle over $`M`$.
Let $`V`$ be a $`2p`$ dimensional oriented real vector bundle over $`M`$. Let $`L`$ be a complex line bundle over $`M`$ with the property that the vector bundle $`U=TXV`$ obeys $`\omega _2(U)=c_1(L)\mathrm{mod}(2)`$. Then the vector bundle $`U`$ has a Spin<sup>c</sup>-structure. Let $`h^V,h^L`$ be the corresponding metrics on $`V,L`$. Let $`S(U,L)`$ be the fundamental complex spinor bundle for $`(U,L)`$ \[LaM, Appendix D.9\] which locally may be written as
(1.1) $`S(U,L)=S_0(U)L^{1/2},`$
where $`S_0(U)`$ is the fundamental spinor bundle for the (possibly non-existent) spin structure on $`U`$, and where $`L^{1/2}`$ is the (possibly non-existent) square root of $`L`$.
Assume that the $`S^1`$-action on $`M`$ lifts to $`V`$, $`L`$ and $`W`$, and assume the metrics $`h^{TX},h^V,h^L,h^W`$ are $`S^1`$-invariant. Also assume that the $`S^1`$-actions on $`TX,V,L`$ lift to $`S(U,L)`$.
Let $`^{TX}`$ be the Levi-Civita connection on $`(TX,h^{TX})`$ along the fibre $`X`$. Let $`^V`$, $`^L`$ and $`^W`$ be the $`S^1`$-invariant and metric-compatible connections on $`(V,h^V)`$, $`(L,h^L)`$ and $`(W,h^W)`$ respectively. Let $`^{S(U,L)}`$ be the Hermitian connection on $`S(U,L)`$ induced by $`^{TX}^V`$ and $`^L`$ (cf. \[LaM, Appendix D\], \[LiuMaZ, ยง1.1\]). Let $`^{S(U,L)W}`$ be the tensor product connection on $`S(U,L)W`$ induced by $`^{S(U,L)}`$ and $`^W`$,
(1.2) $`^{S(U,L)W}=^{S(U,L)}1+1^W.`$
Let $`\{e_i\}_{i=1}^{2l}`$ (resp. $`\{f_j\}_{j=1}^{2p}`$) be an oriented orthonormal basis of $`(TX,h^{TX})`$ (resp. $`(V,h^V)`$). We denote by $`c()`$ the Clifford action of $`TXV`$ on $`S(U,L)`$. Let $`D^XW`$ be the family Spin<sup>c</sup>-Dirac operator on the fiber $`X`$ defined by
(1.3) $`D^XW={\displaystyle \underset{i=1}{\overset{2l}{}}}c(e_i)_{e_i}^{S(U,L)W}.`$
There are two canonical ways to consider $`S(U,L)`$ as a $`๐_2`$-graded vector bundle. Let
(1.6) $`\begin{array}{c}\tau _s=i^lc(e_1)\mathrm{}c(e_{2l}),\hfill \\ \tau _e=i^{l+p}c(e_1)\mathrm{}c(e_{2l})c(f_1)\mathrm{}c(f_{2p})\hfill \end{array}`$
be two involutions of $`S(U,L)`$. Then $`\tau _s^2=\tau _e^2=1`$. We decompose $`S(U,L)=S^+(U,L)S^{}(U,L)`$ corresponding to $`\tau _s`$ (resp. $`\tau _e`$) such that $`\tau _s|_{S^\pm (U,L)}=\pm 1`$ (resp. $`\tau _e|_{S^\pm (U,L)}=\pm 1`$).
For $`\tau =\tau _s`$ or $`\tau _e`$, by \[LiuMa1, Proposition 1.1\], the index bundle $`\mathrm{Ind}_\tau (D^X)`$ over $`B`$ is well-defined in the equivariant $`K`$-group $`K_{S^1}(B)`$.
Let $`F=\{F_\alpha \}`$ be the fixed point set of the circle action on $`M`$. Then $`\pi :F_\alpha B`$ (resp. $`\pi :FB`$) is a smooth fibration with fibre $`Y_\alpha `$ (resp. $`Y`$). Let $`\stackrel{~}{\pi }:NF`$ denote the normal bundle to $`F`$ in $`M`$. Then $`N=TX/TY`$. We identify $`N`$ as the orthogonal complement of $`TY`$ in $`TX_{|F}`$. Let $`h^{TY},h^N`$ be the corresponding metrics on $`TY`$ and $`N`$ induced by $`h^{TX}`$. Then, we have the following $`S^1`$-equivariant decomposition of $`TX`$ over $`F`$,
$`TX_{|F}=N_{m_1}\mathrm{}N_{m_l}TY,`$
where each $`N_\gamma `$ is a complex vector bundle such that $`gS^1`$ acts on it by $`g^\gamma `$. To simplify the notation, we will write simply that
(1.7) $`TX_{|F}=_{v0}N_vTY,`$
where $`N_v`$ is a complex vector bundle such that $`gS^1`$ acts on it by $`g^v`$ with $`v๐^{}`$. Clearly, $`N=_{v0}N_v`$. We will denote by $`N`$ a complex vector bundle, and $`N_๐`$ the underlying real vector bundle of $`N`$.
Similarly let
(1.8) $`W_{|F}=_vW_v`$
be the $`S^1`$-equivariant decomposition of the restriction of $`W`$ over $`F`$. Here $`W_v(v๐)`$ is a complex vector bundle over $`F`$ on which $`gS^1`$ acts by $`g^v`$.
We also have the following $`S^1`$-equivariant decomposition of $`V`$ restricted to $`F`$,
(1.9) $`V_{|F}=_{v0}V_vV_0^๐,`$
where $`V_v`$ is a complex vector bundle such that $`g`$ acts on it by $`g^v`$, and $`V_0^๐`$ is the real subbundle of $`V`$ such that $`S^1`$ acts as identity. For $`v0`$, let $`V_{v,๐}`$ denote the underlying real vector bundle of $`V_v`$. Denote by $`2p^{}=dimV_0^๐`$ and $`2l^{}=dimY`$.
Let us write
(1.10) $`L_F=L\left({\displaystyle \underset{v0}{}}detN_v{\displaystyle \underset{v0}{}}detV_v\right)^1.`$
Then $`TYV_0^๐`$ has a Spin<sup>c</sup> structure as $`\omega _2(TYV_0^๐)=c_1(L_F)\mathrm{mod}(2)`$. Let $`S(TYV_0^๐,L_F)`$ be the fundamental spinor bundle for $`(TYV_0^๐,L_F)`$ \[LaM, Appendix D, pp. 397\].
Let $`D^Y,D^{Y_\alpha }`$ be the families of Spin<sup>c</sup> Dirac operators acting on $`S(TYV_0^๐,L_F)`$ over $`F,F_\alpha `$ as (1.3). If $`R`$ is an Hermitian complex vector bundle equipped with an Hermitian connection over $`F`$, let $`D^YR,D^{Y_\alpha }R`$ denote the twisted Spin<sup>c</sup> Dirac operators on $`S(TYV_0^๐,L_F)R`$ and on $`S(TY_\alpha V_0^๐,L_F)R`$ respectively.
Recall that $`N_{v,๐}`$ and $`V_{v,๐}`$ are canonically oriented by their complex structures. The decompositions (1.7), (1.9) induce the orientations on $`TY`$ and $`V_0^๐`$ respectively. Let $`\{e_i\}_{i=1}^{2l^{}}`$, $`\{f_j\}_{j=1}^{2p^{}}`$ be the corresponding oriented orthonormal basis of $`(TY,h^{TY})`$ and $`(V_0^๐,h^{V_0^๐})`$. There are two canonical ways to consider $`S(TYV_0^๐,L_F)`$ as a $`๐_2`$-graded vector bundle. Let
(1.13) $`\begin{array}{c}\tau _s=i^l^{}c(e_1)\mathrm{}c(e_{2l^{}}),\hfill \\ \tau _e=i^{l^{}+p^{}}c(e_1)\mathrm{}c(e_{2l^{}})c(f_1)\mathrm{}c(f_{2p^{}})\hfill \end{array}`$
be two involutions of $`S(TYV_0^๐,L_F)`$. Then $`\tau _s^2=\tau _e^2=1`$. We decompose $`S(TYV_0^๐,L_F)=S^+(TYV_0^๐,L_F)`$ $`S^{}(TYV_0^๐,L_F)`$ corresponding to $`\tau _s`$ (resp. $`\tau _e`$) such that $`\tau _s|_{S^\pm (TYV_0^๐,L_F)}=\pm 1`$ (resp. $`\tau _e|_{S^\pm (TYV_0^๐,L_F)}=\pm 1`$).
Upon restriction to $`F`$, one has the following isomorphism of $`๐_2`$-graded Clifford modules over $`F`$,
(1.14) $`S(U,L)S(TYV_0^๐,L_F)\widehat{{\displaystyle \underset{v0}{}}}\mathrm{\Lambda }N_v\widehat{{\displaystyle \underset{v0}{}}}\mathrm{\Lambda }V_v.`$
We denote by $`\mathrm{Ind}_{\tau _s}`$, $`\mathrm{Ind}_{\tau _e}`$ the index bundles corresponding to the involutions $`\tau _s,\tau _e`$ respectively.
Let $`S^1`$ act on $`L`$ by sending $`gS^1`$ to $`g^{l_c}`$ $`(l_c๐)`$ on $`F`$. Then $`l_c`$ is locally constant on $`F`$. We define the following elements in $`K(F)[[q^{1/2}]]`$,
(1.19) $`\begin{array}{c}R_\pm (q)=q^{\frac{1}{2}\mathrm{\Sigma }_v|v|dimN_v\frac{1}{2}\mathrm{\Sigma }_vvdimV_v+\frac{1}{2}l_c}_{0<v}\left(\mathrm{Sym}_{q^v}(N_v)detN_v\right)\hfill \\ _{v<0}\mathrm{Sym}_{q^v}(\overline{N}_v)_{v0}\mathrm{\Lambda }_{\pm q^v}(V_v)_vq^vW_v=_nR_{\pm ,n}q^n,\hfill \\ R_\pm ^{}(q)=q^{\frac{1}{2}\mathrm{\Sigma }_v|v|dimN_v\frac{1}{2}\mathrm{\Sigma }_vvdimV_v+\frac{1}{2}l_c}_{0<v}\mathrm{Sym}_{q^v}(\overline{N}_v)\hfill \\ _{v<0}(\mathrm{Sym}_{q^v}(N_v)detN_v)_{v0}\mathrm{\Lambda }_{\pm q^v}(V_v)_vq^vW_v=_nR_{\pm ,n}^{}q^n.\hfill \end{array}`$
The following result was proved in \[LiuMaZ, Theorem 1.2\]:
###### Theorem 1.1
For $`n๐`$, we have the following identity in $`K(B)`$,
(1.24) $`\begin{array}{c}\mathrm{Ind}_{\tau _s}(D^XW,n)=_\alpha (1)^{\mathrm{\Sigma }_{0<v}dimN_v}\mathrm{Ind}_{\tau _s}(D^{Y_\alpha }R_{+,n})\hfill \\ =_\alpha (1)^{\mathrm{\Sigma }_{v<0}dimN_v}\mathrm{Ind}_{\tau _s}(D^{Y_\alpha }R_{+,n}^{}),\hfill \\ \mathrm{Ind}_{\tau _e}(D^XW,n)=_\alpha (1)^{\mathrm{\Sigma }_{0<v}dimN_v}\mathrm{Ind}_{\tau _e}(D^{Y_\alpha }R_{,n})\hfill \\ =_\alpha (1)^{\mathrm{\Sigma }_{v<0}dimN_v}\mathrm{Ind}_{\tau _e}(D^{Y_\alpha }R_{,n}^{}).\hfill \end{array}`$
Remark 1.1. If $`TX`$ has an $`S^1`$-equivariant Spin structure, by setting $`V=0,L=๐`$, we get \[LiuMaZ, Theorem 1.1\].
### 1.2 $`K`$-theory version of the vanishing theorem of Hattori
In this subsection, we assume that $`TX`$ has an $`S^1`$-equivariant almost complex structure $`J`$. Then one has the canonical splitting
(1.25) $`TX_๐๐=T^{(1,0)}XT^{(0,1)}X,`$
where
(1.28) $`\begin{array}{c}T^{(1,0)}X=\{zTX_๐๐,Jz=\sqrt{1}z\},\hfill \\ T^{(0,1)}X=\{zTX_๐๐,Jz=\sqrt{1}z\}.\hfill \end{array}`$
Let $`K_X=det(T^{(1,0)}X)`$ be the determinant line bundle of $`T^{(1,0)}X`$ over $`M`$. Then the complex spinor bundle $`S(TX,K_X)`$ for $`(TX,K_X)`$ is $`\mathrm{\Lambda }(T^{(0,1)}X)`$. In this case, the almost complex structure $`J`$ on $`TX`$ induces an almost complex structure on $`TY`$. Then we can rewrite (1.7) as,
(1.29) $`T^{(1,0)}X=_{v0}N_vT^{(1,0)}Y,`$
where $`N_v`$ are complex vector subbundles of $`T^{(1,0)}X`$ on which $`gS^1`$ acts by multiplication by $`g^v`$.
We suppose that $`c_1(T^{(1,0)}X)=0\mathrm{mod}(N)`$ $`(N๐,N2)`$. Then the complex line bundle $`K_X^{1/N}`$ is well defined over $`M`$. After replacing the $`S^1`$ action by its $`N`$-fold action, we can always assume that $`S^1`$ acts on $`K_X^{1/N}`$. For $`s๐`$, let $`D^XK_X^{s/N}`$ be the twisted Dirac operator on $`\mathrm{\Lambda }(T^{(0,1)}X)K_X^{s/N}`$ defined as in (1.3).
The following result generalizes the main result of \[Ha\] to the family case.
###### Theorem 1.2
We assume that $`M`$ is connected and that the $`S^1`$ action is nontrivial. If $`c_1(T^{(1,0)}X)=0\mathrm{mod}(N)`$ $`(N๐,N2)`$, then for $`s๐,N<s<0`$,
(1.30) $`\mathrm{Ind}(D^XK_X^{s/N})=0\mathrm{in}K_{S^1}(B).`$
Proof : Consider $`R_+(q),R_+^{}(q)`$ of (1.19) with $`V=0,W=K_X^{s/N}`$. We know
(1.33) $`\begin{array}{c}R_{+,n}=0\mathrm{if}n<a_1=\mathrm{inf}_\alpha (\frac{1}{2}_v|v|dimN_v+(\frac{1}{2}+\frac{s}{N})_vvdimN_v),\hfill \\ R_{+,n}^{}=0\mathrm{if}n>a_2=\mathrm{sup}_\alpha (\frac{1}{2}_v|v|dimN_v+(\frac{1}{2}+\frac{s}{N})_vvdimN_v).\hfill \end{array}`$
As $`N<s<0`$, by (1.33), we know that $`a_10,a_20`$, with $`a_1`$ or $`a_2`$ equal to zero iff $`_v|v|dimN_v=0`$ for all $`\alpha `$, which means that the $`S^1`$ action does not have fixed points.
From Theorem 1.1 (cf. \[Z, Theorem A.1\]) and the above discussion, we get Theorem 1.2. $`\mathrm{}`$
Remark 1.2. From the proof of Theorem 1.2, one also deduces that $`D^XK_X^1,D^X`$ are rigid on the equivariant $`K`$-theory level (cf. \[Z, (2.17)\]).
## 2 Rigidity and vanishing theorems in K-Theory
The purpose of this section is to establish the main results of this paper: the rigidity and vanishing theorems on the equivariant $`K`$-theory level for a family of Spin<sup>c</sup> manifolds. The results in this section refine some of the results in \[LiuMa2\] to the $`K`$-theory level.
This section is organized as follows: In Section 2.1, we state our main results, the rigidity and vanishing theorems on the equivariant $`K`$-theory level for a family of Spin<sup>c</sup> manifolds. In Section 2.2, we state two intermediate results which will be used to prove our main results stated in Section 2.1. In Section 2.3, we prove the family rigidity and vanishing theorems.
Throughout this section, we keep the notations of Section 1.1.
### 2.1 Family rigidity and vanishing Theorem
Let $`\pi :MB`$ be a fibration of compact manifolds with fiber $`X`$ and $`dimX=2l`$. We assume that $`S^1`$ acts fiberwise on $`M`$, and $`TX`$ has an $`S^1`$-invariant Spin<sup>c</sup> structure. Let $`V`$ be an even dimensional real vector bundle over $`M`$. We assume that $`V`$ has an $`S^1`$-invariant spin structure. Let $`W`$ be an $`S^1`$-equivariant complex vector bundle of rank $`r`$ over $`M`$. Let $`K_W=det(W)`$ be the determinant line bundle of $`W`$.
Let $`K_X`$ be the $`S^1`$-equivariant complex line bundle over $`M`$ which is induced by the $`S^1`$-invariant Spin<sup>c</sup> structure of $`TX`$. Its equivariant first Chern class $`c_1(K_X)_{S^1}`$ may also be written as $`c_1(TX)_{S^1}`$.
Let $`S(TX,K_X)`$ be the complex spinor bundle of $`(TX,K_X)`$ as in Section 1.1. Let $`S(V)=S^+(V)S^{}(V)`$ be the spinor bundle of $`V`$.
We define the following elements in $`K(M)[[q^{1/2}]]`$:
(2.6) $`\begin{array}{c}Q_1(W)=_{n=0}^{\mathrm{}}\mathrm{\Lambda }_{q^n}(\overline{W})_{n=1}^{\mathrm{}}\mathrm{\Lambda }_{q^n}(W),\hfill \\ R_1(V)=(S^+(V)+S^{}(V))_{n=1}^{\mathrm{}}\mathrm{\Lambda }_{q^n}(V),\hfill \\ R_2(V)=(S^+(V)S^{}(V))_{n=1}^{\mathrm{}}\mathrm{\Lambda }_{q^n}(V),\hfill \\ R_3(V)=_{n=1}^{\mathrm{}}\mathrm{\Lambda }_{q^{n1/2}}(V),\hfill \\ R_4(V)=_{n=1}^{\mathrm{}}\mathrm{\Lambda }_{q^{n1/2}}(V).\hfill \end{array}`$
For $`N๐^{}`$, let $`y=e^{2\pi i/N}๐`$. Let $`G_y`$ be the multiplicative group generated by $`y`$. Following Witten \[W\], we consider the fiberwise action $`G_y`$ on $`W`$ and $`\overline{W}`$ by sending $`yG_y`$ to $`y`$ on $`W`$ and $`y^1`$ on $`\overline{W}`$. Then $`G_y`$ acts naturally on $`Q_1(W)`$.
Recall that the equivariant cohomology group $`H_{S^1}^{}(M,๐)`$ of $`M`$ is defined by
(2.7) $`H_{S^1}^{}(M,๐)=H^{}(M\times _{S^1}ES^1,๐),`$
where $`ES^1`$ is the usual universal $`S^1`$-principal bundle over the classifying space $`BS^1`$ of $`S^1`$. So $`H_{S^1}^{}(M,๐)`$ is a module over $`H^{}(BS^1,๐)`$ induced by the projection $`\overline{\pi }:M\times _{S^1}ES^1BS^1`$. Let $`p_1(V)_{S^1},p_1(TX)_{S^1}H_{S^1}^{}(M,๐)`$ be the $`S^1`$-equivariant first Pontrjagin classes of $`V`$ and $`TX`$ respectively. As $`V\times _{S^1}ES^1`$ is spin over $`M\times _{S^1}ES^1`$, one knows that $`\frac{1}{2}p_1(V)_{S^1}`$ is well-defined in $`H_{S^1}^{}(M,๐)`$ (cf. \[T, pp. 456-457\]). Also recall that
(2.8) $`H^{}(BS^1,๐)=๐[[u]]`$
with $`u`$ a generator of degree $`2`$.
In the following, we denote by $`D^XR`$ the family of Dirac operators acting fiberwise on $`S(TX,K_X)R`$ as was defined in Section 1.1.
We can now state the main results of this paper as follows.
###### Theorem 2.1
If $`\omega _2(W)_{S^1}=\omega _2(TX)_{S^1}`$, $`\frac{1}{2}p_1(V+WTX)_{S^1}=e\overline{\pi }^{}u^2`$ $`(n๐)`$ in $`H_{S^1}^{}(M,๐)`$, and $`c_1(W)=0\mathrm{mod}(N)`$. For $`i=1,2,3,4`$, consider the family of $`G_y\times S^1`$-equivariant elliptic operators
$$D^X(K_WK_X^1)^{1/2}_{n=1}^{\mathrm{}}\mathrm{Sym}_{q^n}(TX)Q_1(W)R_i(V).$$
i) If $`e=0`$, then these operators are rigid on the equivariant K-theory level for the $`S^1`$ action.
ii) If $`e<0`$, then the index bundles of these operators are zero in $`K_{G_y\times S^1}(B)`$. In particular, these index bundles are zero in $`K_{G_y}(B)`$.
###### Remark 2.1
As $`\omega _2(W)_{S^1}=\omega _2(TX)_{S^1}`$, $`\frac{1}{2}p_1(WTX)_{S^1}H_{S^1}^{}(M,๐)`$ is well defined. The condition $`\omega _2(W)_{S^1}=\omega _2(TX)_{S^1}`$ also means $`c_1(K_WK_X^1)_{S^1}=0\mathrm{mod}(2)`$, by \[HaY, Corollary 1.2\], the $`S^1`$-action on $`M`$ can be lifted to $`(K_WK_X^1)^{1/2}`$ and is compatible with the $`S^1`$ action on $`K_WK_X^1`$.
###### Remark 2.2
If we assume $`c_1(W)_{S^1}=c_1(TX)_{S^1}`$ in $`H_{S^1}^{}(M,๐)`$ instead of $`\omega _2(W)_{S^1}=\omega _2(TX)_{S^1}`$ in Theorem 2.1, then $`K_WK_X^1`$ is a trivial line bundle over $`M`$, and $`S^1`$ acts trivially on it. In this case, Theorem 2.1 gives the family version of the results of \[De\].
###### Remark 2.3
The interested reader can apply our method to get various rigidity and vanishing theorems, for example, to get a generalization of Theorem1.2 for the elements \[W, (65)\].
Actually, as in \[LiuMaZ\], our proof of these theorems works under the following slightly weaker hypothesis. Let us first explain some notations.
For each $`n>1`$, consider $`๐_nS^1`$, the cyclic subgroup of order $`n`$. We have the $`๐_n`$ equivariant cohomology of $`M`$ defined by $`H_{๐_n}^{}(M,๐)=H^{}(M\times _{๐_n}ES^1,๐)`$, and there is a natural โforgetfulโ map $`\alpha (S^1,๐_n):M\times _{๐_n}ES^1M\times _{S^1}ES^1`$ which induces a pullback $`\alpha (S^1,๐_n)^{}:H_{S^1}^{}(M,๐)H_{๐_n}^{}(M,๐)`$. The arrow which forgets the $`S^1`$ action altogether we denote by $`\alpha (S^1,1)`$. Thus $`\alpha (S^1,1)^{}:H_{S^1}^{}(M,๐)H^{}(M,๐)`$ is induced by the inclusion of $`M`$ into $`M\times _{S^1}ES^1`$ as a fiber over $`BS^1`$.
Finally, note that if $`๐_n`$ acts trivially on a space $`Y`$, then there is a new arrow $`t^{}:H^{}(Y,๐)H_{๐_n}^{}(Y,๐)`$ induced by the projection $`Y\times _{๐_n}ES^1=Y\times B๐_n\stackrel{t}{}Y`$.
We let $`๐_{\mathrm{}}=S^1`$. For each $`1<n+\mathrm{}`$, let $`i:M(n)M`$ be the inclusion of the fixed point set of $`๐_nS^1`$ in $`M`$ and so $`i`$ induces $`i_{S^1}:M(n)\times _{S^1}ES^1M\times _{S^1}ES^1`$.
In the rest of this paper, we suppose that there exists some integer $`e๐`$ such that for $`1<n+\mathrm{}`$,
(2.9) $`\alpha (S^1,๐_n)^{}i_{S^1}^{}\left({\displaystyle \frac{1}{2}}p_1(V+WTX)_{S^1}e\overline{\pi }^{}u^2\right)`$
$`=t^{}\alpha (S^1,1)^{}i_{S^1}^{}\left({\displaystyle \frac{1}{2}}p_1(V+WTX)_{S^1}\right).`$
###### Remark 2.4
The relation (2.9) clearly follows from the hypothesises of Theorem 2.1 by pulling back and forgetting. Thus it is weaker.
We can now state a slightly more general version of Theorem 2.1.
###### Theorem 2.2
Under the hypothesis (2.9), we have
i) If $`e=0`$, then the index bundles of the elliptic operators in Theorem 2.1 are rigid on the equivariant K-theory level for the $`S^1`$-action.
ii) If $`e<0`$, then the index bundles of the elliptic operators in Theorem 2.1 are zero as elements in $`K_{G_y\times S^1}(B)`$. In particular, these index bundles are zero in $`K_{G_y}(B)`$.
The rest of this section is devoted to a proof of Theorem 2.2.
### 2.2 Two intermediate results
Let $`F=\{F_\alpha \}`$ be the fixed point set of the circle action. Then $`\pi :FB`$ is a fibration with compact fibre denoted by $`Y=\{Y_\alpha \}`$.
As in \[LiuMaZ, ยง2\], we may and we will assume that
(2.12) $`\begin{array}{c}TX_{|F}=TY_{0<v}N_v,\hfill \\ TX_๐๐=TY_๐๐_{0<v}(N_v\overline{N}_v),\hfill \end{array}`$
where $`N_v`$ is the complex vector bundle on which $`S^1`$ acts by sending $`g`$ to $`g^v`$ (Here $`N_v`$ can be zero). We also assume that
(2.15) $`\begin{array}{c}V_{|F}=V_0^๐_{0<v}V_v,\hfill \\ W_{|F}=_vW_v,\hfill \end{array}`$
where $`V_v`$, $`W_v`$ are complex vector bundles on which $`S^1`$ acts by sending $`g`$ to $`g^v`$, and $`V_0^๐`$ is a real vector bundle on which $`S^1`$ acts as identity.
By (2.12), as in (1.14), there is a natural isomorphism between the $`๐_2`$-graded $`C(TX)`$-Clifford modules over $`F`$,
(2.16) $`S(TY,K_X_{0<v}(detN_v)^1)\widehat{}_{0<v}\mathrm{\Lambda }N_vS(TX,K_X)_{|F}.`$
For $`R`$ a complex vector bundle over $`F`$, let $`D^YR`$, $`D^{Y_\alpha }R`$ be the twisted Spin<sup>c</sup> Dirac operator on $`S(TY,K_X_{0<v}(detN_v)^1)R`$ on $`F,F_\alpha `$ respectively.
On $`F`$, we write
(2.20) $`\begin{array}{c}e(N)=_{0<v}v^2dimN_v,d^{}(N)=_{0<v}vdimN_v,\hfill \\ e(V)=_{0<v}v^2dimV_v,d^{}(V)=_{0<v}vdimV_v,\hfill \\ e(W)=_vv^2dimW_v,d^{}(W)=_vvdimW_v.\hfill \end{array}`$
Then $`e(N),e(V),e(W),d^{}(N),d^{}(V)`$ and $`d^{}(W)`$ are locally constant functions on $`F`$.
By \[H, ยง8\], we have the following property,
###### Lemma 2.1
If $`c_1(W)=0\mathrm{mod}(N)`$, then $`d^{}(W)\mathrm{mod}(N)`$ is constant on each connected component of $`M`$.
Proof : As $`c_1(W)=0\mathrm{mod}(N)`$, $`(K_W)^{1/N}`$ is well defined. Consider the $`N`$-fold covering $`S^1S^1`$, with $`\mu \lambda =\mu ^N`$, then $`\mu `$ acts on $`M`$ and $`K_W`$ through $`\lambda `$. This action can be lift to $`(K_W)^{1/N}`$. On $`F`$, $`\mu `$ acts on $`(K_W)^{1/N}`$ by multipication by $`\mu ^{d^{}(W)}`$. However, if $`\mu =\zeta =e^{2\pi i/N}`$, then it operates trivially on $`M`$. So the action of $`\zeta `$ in each fibre of $`L`$ is by multiplication by $`\zeta ^a`$, and $`a\mathrm{mod}(N)`$ is constant on each connected component of $`M`$.
The proof of Lemma 2.1 is complete. $`\mathrm{}`$
Let us write
(2.24) $`\begin{array}{c}L(N)=_{0<v}(detN_v)^v,L(V)=_{0<v}(detV_v)^v,\hfill \\ L(W)=_{v0}(detW_v)^v,\hfill \\ L=L(N)^1L(V)L(W).\hfill \end{array}`$
We denote the Chern roots of $`N_v`$ by $`\{x_v^j\}`$ (resp. $`V_v`$ by $`u_v^j`$ and $`W_v`$ by $`w_v^j`$), and the Chern roots of $`TY_๐๐`$ by $`\{\pm y_j\}`$ (resp. $`V_0=V_0^๐_๐๐`$ by $`\{\pm u_0^j\}`$). Then if we take $`๐_{\mathrm{}}=S^1`$ in (2.9), we get
(2.27) $`\begin{array}{c}\frac{1}{2}(\mathrm{\Sigma }_{v,j}(u_v^j+vu)^2+\mathrm{\Sigma }_{v,j}(w_v^j+vu)^2\mathrm{\Sigma }_j(y_j)^2\mathrm{\Sigma }_{v,j}(x_v^j+vu)^2)eu^2\hfill \\ =\frac{1}{2}(\mathrm{\Sigma }_{v,j}(u_v^j)^2+\mathrm{\Sigma }_{v,j}(w_v^j)^2\mathrm{\Sigma }_j(y_j)^2\mathrm{\Sigma }_{v,j}(x_v^j)^2).\hfill \end{array}`$
By (2.8), (2.27), we get
(2.31) $`\begin{array}{c}c_1(L)=\mathrm{\Sigma }_{v,j}vu_v^j+\mathrm{\Sigma }_{v,j}vw_v^j\mathrm{\Sigma }_{v,j}vx_v^j=0,\hfill \\ e(V)+e(W)e(N)\hfill \\ =_{0<v}v^2dimV_v+_vv^2dimW_v_{0<v}v^2dimN_v=2e,\hfill \end{array}`$
which does not depends on the connected components of $`F`$. This means $`L`$ is a trivial complex line bundle over each component $`F_\alpha `$ of $`F`$, and $`S^1`$ acts on $`L`$ by sending $`g`$ to $`g^{2e}`$, and $`G_y`$ acts on $`L`$ by sending $`y`$ to $`y^{d^{}(W)}`$. By Lemma 2.1, we can extend $`L`$ to a trivial complex line bundle over $`M`$, and we extend the $`S^1`$-action on it by sending $`g`$ on the canonical section $`1`$ of $`L`$ to $`g^{2e}1`$, and $`G_y`$ acts on $`L`$ by sending $`y`$ to $`y^{d^{}(W)}`$.
The line bundles in (2.24) will play important roles in the next two sections which consist of the proof of Theorems 2.3, 2.4 to be stated below.
In what follows, if $`R(q)=_{m\frac{1}{2}๐}R_mq^mK_{S^1}(M)[[q^{1/2}]]`$, we will also denote $`\mathrm{Ind}(D^XR_m,h)`$ by $`\mathrm{Ind}(D^XR(q),m,h)`$. For $`k=1,2,3,4`$, set
(2.32) $`R_{1k}=(K_WK_X^1)^{1/2}Q_1(W)R_k(V).`$
We first state a result which expresses the global equivariant family index via the family indices on the fixed point set.
###### Proposition 2.1
For $`m\frac{1}{2}๐`$, $`h๐`$, $`1k4`$, we have the following identity in $`K_{G_y}(B)`$,
(2.36) $`\begin{array}{c}\mathrm{Ind}(D^X_{n=1}^{\mathrm{}}\mathrm{Sym}_{q^n}(TX)R_{1k},m,h)\hfill \\ =_\alpha (1)^{\mathrm{\Sigma }_{0<v}dimN_v}\mathrm{Ind}(D^{Y_\alpha }_{n=1}^{\mathrm{}}\mathrm{Sym}_{q^n}(TX)R_{1k}\hfill \\ \mathrm{Sym}(_{0<v}N_v)_{0<v}detN_v,m,h)\hfill \end{array}`$
Proof : This follows directly from Theorem 1.1 and (2.16). $`\mathrm{}`$
For $`p๐`$, we define the following elements in $`K_{S^1}(F)[[q]]`$:
(2.41) $`\begin{array}{c}_p(X)=_{0<v}\left(_{n=1}^{\mathrm{}}\mathrm{Sym}_{q^n}(N_v)_{n>pv}\mathrm{Sym}_{q^n}(\overline{N}_v)\right)_{n=1}^{\mathrm{}}\mathrm{Sym}_{q^n}(TY),\hfill \\ _p^{}(X)=_{\stackrel{0<v}{0npv}}\left(\mathrm{Sym}_{q^n}(N_v)detN_v\right),\hfill \\ \\ ^p(X)=_p(X)_p^{}(X).\hfill \end{array}`$
Then, from (2.12), over $`F`$, we have
(2.42) $`^0(X)=_{n=1}^{\mathrm{}}\mathrm{Sym}_{q^n}(TX)\mathrm{Sym}(_{0<v}N_v)_{0<v}detN_v.`$
We now state two intermediate results on the relations between the family indices on the fixed point set. They will be used in the next subsection to prove Theorem 2.2.
###### Theorem 2.3
For $`1k4`$, $`h,p๐`$, $`p>0`$, $`m\frac{1}{2}๐`$, we have the following identity in $`K_{G_y}(B)`$,
(2.46) $`\begin{array}{c}_\alpha (1)^{\mathrm{\Sigma }_{0<v}dimN_v}\mathrm{Ind}(D^{Y_\alpha }^0(X)R_{1k},m,h)\hfill \\ =_\alpha (1)^{pd^{}(N)+\mathrm{\Sigma }_{0<v}dimN_v}\mathrm{Ind}(D^{Y_\alpha }^p(X)R_{1k},\hfill \\ m+\frac{1}{2}p^2e(N)+\frac{1}{2}pd^{}(N),h).\hfill \end{array}`$
###### Theorem 2.4
For each $`\alpha `$, $`1k4`$, $`h,p๐`$, $`p>0`$, $`m\frac{1}{2}๐`$, we have the following identity in $`K_{G_y}(B)`$,
(2.49) $`\begin{array}{c}\mathrm{Ind}(D^{Y_\alpha }^p(X)R_{1k},m+\frac{1}{2}p^2e(N)+\frac{1}{2}pd^{}(N),h)\hfill \\ =(1)^{pd^{}(W)}\mathrm{Ind}(D^{Y_\alpha }^0(X)R_{1k}L^p,m+ph+p^2e,h).\hfill \end{array}`$
Theorem 2.3 is a direct consequence of Theorem 2.5 to be stated below, which will be proved in Section 4, while Theorem 2.4 will be proved in Section 3.
To state Theorem 2.5, let $`J=\{v๐|`$ There exists $`\alpha `$ such that $`N_v0`$ on $`F_\alpha \}`$ and
(2.50) $`\mathrm{\Phi }=\{\beta ]0,1]|\mathrm{There}\mathrm{exists}vJ\mathrm{such}\mathrm{that}\beta v๐\}.`$
We order the elements in $`\mathrm{\Phi }`$ so that $`\mathrm{\Phi }=\{\beta _i|1iJ_0,J_0๐\mathrm{and}\beta _i<\beta _{i+1}\}`$. Then for any integer $`1iJ_0`$, there exist $`p_i,n_i๐,0<p_in_i`$, with $`(p_i,n_i)=1`$ such that
(2.51) $`\beta _i=p_i/n_i.`$
Clearly, $`\beta _{J_0}=1`$. We also set $`p_0=0`$ and $`\beta _0=0`$.
For $`1jJ_0`$, $`p๐^{}`$, we write
(2.55) $`\begin{array}{c}I_0^p=\varphi ,\text{the empty set},\hfill \\ I_j^p=\{(v,n)|vJ,(p1)v<npv,{\displaystyle \frac{n}{v}}=p1+{\displaystyle \frac{p_j}{n_j}}\},\hfill \\ \overline{I}_j^p=\{(v,n)|vJ,(p1)v<npv,{\displaystyle \frac{n}{v}}>p1+{\displaystyle \frac{p_j}{n_j}}\}.\hfill \end{array}`$
For $`0jJ_0`$, set
(2.56)
$`_{p,j}(X)=_p(X)_{p1}^{}(X){\displaystyle \underset{(v,n)_{i=1}^jI_i^p}{}}\left(\mathrm{Sym}_{q^n}(N_v)detN_v\right){\displaystyle \underset{(v,n)\overline{I}_j^p}{}}\mathrm{Sym}_{q^n}(\overline{N}_v).`$
Then
(2.59) $`\begin{array}{c}_{p,0}(X)=^{p+1}(X),\hfill \\ _{p,J_0}(X)=^p(X).\hfill \end{array}`$
For $`s๐`$, let $`[s]`$ denote the greatest integer which is less than or equal to the given number $`s`$. For $`0jJ_0`$, denote by
(2.62) $`\begin{array}{c}e(p,\beta _j,N)=\frac{1}{2}_{0<v}(dimN_v)\left((p1)v+[\frac{p_jv}{n_j}]\right)\left((p1)v+[\frac{p_jv}{n_j}]+1\right),\hfill \\ d^{}(p,\beta _j,N)=_{0<v}(dimN_v)([\frac{p_jv}{n_j}]+(p1)v).\hfill \end{array}`$
Then $`e(p,\beta _j,N)`$ and $`d^{}(p,\beta _j,N)`$ are locally constant functions on $`F`$. And
(2.66) $`\begin{array}{c}e(p,\beta _0,N)=\frac{1}{2}(p1)^2e(N)+\frac{1}{2}(p1)d^{}(N),\hfill \\ e(p,\beta _{J_0},N)=\frac{1}{2}p^2e(N)+\frac{1}{2}pd^{}(N),\hfill \\ d^{}(p,\beta _{J_0},N)=d^{}(p+1,\beta _0,N)=pd^{}(N).\hfill \end{array}`$
###### Theorem 2.5
For $`1k4`$, $`1jJ_0`$, $`p๐^{}`$, $`h๐`$, $`m\frac{1}{2}๐`$, we have the following identity in $`K_{G_y}(B)`$,
(2.71) $`\begin{array}{c}_\alpha (1)^{d^{}(p,\beta _{j1},N)+\mathrm{\Sigma }_{0<v}dimN_v}\mathrm{Ind}(D^{Y_\alpha }_{p,j1}(X)R_{1k},\hfill \\ m+e(p,\beta _{j1},N),h)\hfill \\ =_\alpha (1)^{d^{}(p,\beta _j,N)+\mathrm{\Sigma }_{0<v}dimN_v}\mathrm{Ind}(D^{Y_\alpha }_{p,j}(X)R_{1k},\hfill \\ m+e(p,\beta _j,N),h).\hfill \end{array}`$
Proof: The proof is delayed to Section 4. $`\mathrm{}`$
Proof of Theorem 2.3 : From (2.59), (2.66), and Theorem 2.5, for $`1k4`$, $`h๐`$, $`p๐^{}`$ and $`m\frac{1}{2}๐`$, we have the following identity in $`K_{G_y}(B)`$:
(2.76) $`\begin{array}{c}_\alpha (1)^{d^{}(p,\beta _{J_0},N)+\mathrm{\Sigma }_{0<v}dimN_v}\mathrm{Ind}(D^{Y_\alpha }^p(X)R_{1k},\hfill \\ m+\frac{1}{2}p^2e(N)+\frac{1}{2}pd^{}(N),h)\hfill \\ =_\alpha (1)^{d^{}(p,\beta _0,N)+\mathrm{\Sigma }_{0<v}dimN_v}\mathrm{Ind}(D^{Y_\alpha }^{p+1}(X)R_{1k},\hfill \\ m+\frac{1}{2}(p1)^2e(N)+\frac{1}{2}(p1)d^{}(N),h).\hfill \end{array}`$
From (2.66), (2.76), we get Theorem 2.3. $`\mathrm{}`$
### 2.3 Proof of Theorem 2.2
As $`\frac{1}{2}p_1(TXW)_{S^1}H_{S^1}^{}(M,๐)`$ is well defined, by (2.20), and (2.27),
(2.77) $`d^{}(N)+d^{}(W)=0\mathrm{mod}(2).`$
From Proposition 2.1, Theorems 2.3, 2.4, (2.62), (2.77), for $`1k4`$, $`h,p๐`$, $`p>0`$, $`m\frac{1}{2}๐`$, we get the following identity in $`K_{G_y}(B)`$,
(2.80) $`\begin{array}{c}\mathrm{Ind}(D^X_{n=1}^{\mathrm{}}\mathrm{Sym}_{q^n}(TX)R_{1k},m,h)\hfill \\ =\mathrm{Ind}(D^X_{n=1}^{\mathrm{}}\mathrm{Sym}_{q^n}(TX)R_{1k}L^p,m^{},h),\hfill \end{array}`$
with
(2.81) $`m^{}=m+ph+p^2e.`$
Note that from (2.1), (2.32), if $`m<0`$, or $`m^{}<0`$, then two side of (2.80) are zero in $`K_{G_y}(B)`$. Also recall that $`yG_y`$ acts on the trivial line bundle $`L`$ by sending $`y`$ to $`y^{d^{}(W)}`$.
i) Assume that $`e=0`$. Let $`h๐,m_0\frac{1}{2}๐`$, $`h0`$ be fixed. If $`h>0`$, we take $`m^{}=m_0`$, then for $`p`$ big enough, we get $`m<0`$ in (2.81). If $`h<0`$, we take $`m=m_0`$, then for $`p`$ big enough, we get $`m^{}<0`$ in (2.81).
So for $`h0`$, $`m_0\frac{1}{2}๐`$, $`1k4`$, we get
(2.82) $`\mathrm{Ind}(D^X_{n=1}^{\mathrm{}}\mathrm{Sym}_{q^n}(TX)R_{1k},m_0,h)=0\mathrm{in}K_{G_y}(B).`$
ii) Assume that $`e<0`$. For $`h๐`$, $`m_0\frac{1}{2}๐`$, we take $`m=m_0`$, then for $`p`$ big enough, we get $`m^{}<0`$ in (2.81), which again gives us (2.82).
The proof of Theorem 2.2 is complete. $`\mathrm{}`$
Remark 2.5: Under the condition of Theorem 2.2 i), if $`d^{}(W)0\mathrm{mod}(N)`$, we canโt deduce these index bundles are zero in $`K_{G_y}(B)`$. If in addition, $`M`$ is connected, by (2.80), for $`1k4`$, in $`K_{G_y}(B)`$, we get
(2.85) $`\begin{array}{c}\mathrm{Ind}(D^X_{n=1}^{\mathrm{}}\mathrm{Sym}_{q^n}(TX)R_{1k})\hfill \\ =\mathrm{Ind}(D^X_{n=1}^{\mathrm{}}\mathrm{Sym}_{q^n}(TX)R_{1k})[d^{}(W)].\hfill \end{array}`$
Here we denote by $`[d^{}(W)]`$ the one dimensional complex vector space on which $`yG_y`$ acts by multiplication by $`y^{d^{}(W)}`$. In particular, if $`B`$ is a point, by (2.85), we get the vanishing theorem analogue to the result of \[H, ยง10\].
Remark 2.6: If we replace $`c_1(W)=0\mathrm{mod}(N),y=e^{2\pi i/N}`$ by $`c_1(W)=0,y=e^{2\pi ci}`$, with $`c๐๐`$ in Theorem 2.2, then by Lemma 2.1, $`d^{}(W)`$ is constant on each connected component of $`M`$. In this case, we still have Theorem 2.2. In fact, we only use $`c_1(W)=0\mathrm{mod}(N)`$ to insure the action $`G_y`$ on $`L`$ is well defined. So we also generalize the main result of \[K\] to family case.
## 3 Proof of Theorem 2.4
This section is organized as follows: In Section 3.1, we introduce some notations. In Section 3.2, we prove Theorem 2.4 by introducing some shift operators as in \[LiuMaZ, ยง3\].
Throughout this section, we keep the notations of Section 2.
### 3.1 Reformulation of Theorem 2.4
To simplify the notations, we introduce some new notations in this subsection. For $`n_0๐^{}`$, we define a number operator $`P`$ on $`K_{S^1}(M)[[q^{\frac{1}{n_0}}]]`$ in the following way: if $`R(q)=_{n\frac{1}{n_0}๐}q^nR_nK_{S^1}(M)[[q^{\frac{1}{n_0}}]]`$, then $`P`$ acts on $`R(q)`$ by multiplication by $`n`$ on $`R_n`$. From now on, we simply denote $`\mathrm{Sym}_{q^n}(TX),\mathrm{\Lambda }_{q^n}(V)`$ by $`\mathrm{Sym}(TX_n),\mathrm{\Lambda }(V_n)`$ respectively. In this way, $`P`$ acts on $`TX_n`$, $`V_n`$ by multiplication by $`n`$, and the action $`P`$ on $`\mathrm{Sym}(TX_n),\mathrm{\Lambda }(V_n)`$ is naturally induced by the corresponding action of $`P`$ on $`TX_n`$, $`V_n`$. So the eigenspace of $`P=n`$ is just given by the coefficient of $`q^n`$ of the corresponding element $`R(q)`$. For $`R(q)=_{n\frac{1}{n_0}๐}q^nR_nK_{S^1}(M)[[q^{\frac{1}{n_0}}]]`$, we will also denote
(3.1) $`\mathrm{Ind}(D^XR(q),m,h)=\mathrm{Ind}(D^XR_m,h).`$
Let $`H`$ be the canonical basis of $`\mathrm{Lie}(S^1)=๐`$, i.e., $`\mathrm{exp}(tH)=\mathrm{exp}(2\pi it)`$ for $`t๐`$. If $`E`$ is an $`S^1`$-equivariant vector bundle over $`M`$, on the fixed point set $`F`$, let $`J_H`$ be the representation of $`\mathrm{Lie}(S^1)`$ on $`E|_F`$. Then the weight of $`S^1`$ action on $`\mathrm{\Gamma }(F,E|_F)`$ is given by the action
(3.2) $`๐_H={\displaystyle \frac{1}{2\pi }}\sqrt{1}J_H.`$
Recall that the $`๐_2`$ grading on $`S(TX,K_X)_{n=1}^{\mathrm{}}\mathrm{Sym}(TX_n)`$ (resp. $`S(TY,K_X_{0<v}(detN_v)^1)^p(X)`$) is induced by the $`๐_2`$-grading on $`S(TX,K_X)`$ (resp. $`S(TY,K_X_{0<v}(detN_v)^1)`$). Let
(3.6) $`\begin{array}{c}F_V^1=S(V)_{n=1}^{\mathrm{}}\mathrm{\Lambda }(V_n),\hfill \\ F_V^2=_{n๐+\frac{1}{2}}\mathrm{\Lambda }(V_n),\hfill \\ Q(W)=_{n=0}^{\mathrm{}}\mathrm{\Lambda }(\overline{W}_n)_{n=1}^{\mathrm{}}\mathrm{\Lambda }(W_n)\hfill \end{array}`$
There are two natural $`๐_2`$ gradings on $`F_V^1,F_V^2`$ (resp. $`Q(W)`$). The first grading is induced by the $`๐_2`$-grading of $`S(V)`$ and the forms of homogeneous degree in $`_{n=1}^{\mathrm{}}\mathrm{\Lambda }(V_n)`$, $`_{n๐+\frac{1}{2}}\mathrm{\Lambda }(V_n)`$ (resp. $`Q(W)`$). We define $`\tau _{e|F_V^{i\pm }}=\pm 1`$ (resp. $`\tau _{1|Q(W)^\pm }=\pm 1`$) to be the involution defined by this $`๐_2`$-grading. The second grading is the one for which $`F_V^i`$ $`(i=1,2)`$ are purely even, i.e., $`F_V^{i+}=F_V^i`$. We denote by $`\tau _s=\mathrm{Id}`$ the involution defined by this $`๐_2`$ grading. Then the coefficient of $`q^n`$ $`(n\frac{1}{2}๐)`$ in (2.1) of $`R_1(V)`$ or $`R_2(V)`$ (resp. $`R_3(V),R_4(V)`$, or $`Q_1(W)`$) is exactly the $`๐_2`$-graded vector subbundle of $`(F_V^1,\tau _s)`$ or $`(F_V^1,\tau _e)`$ (resp. $`(F_V^2,\tau _e)`$, $`(F_V^2,\tau _s)`$ or $`(Q(W),\tau _1)`$), on which $`P`$ acts by multiplication by $`n`$.
We denote by $`\tau _e`$ (resp. by $`\tau _s`$) the $`๐_2`$-grading on $`S(TX,K_X)_{n=1}^{\mathrm{}}\mathrm{Sym}(TX_n)F_V^k`$ ($`k=1,2`$) induced by the above $`๐_2`$-gradings. We will denote by $`\tau _{e1}`$ (resp. by $`\tau _{s1}`$) the $`๐_2`$-gradings on $`S(TX,K_X)_{n=1}^{\mathrm{}}\mathrm{Sym}(TX_n)F_V^kQ(W)`$ defined by
(3.7) $`\tau _{e1}=\tau _e1+1\tau _1,\tau _{s1}=\tau _s1+1\tau _1.`$
Let $`h^{V_v}`$ be the metric on $`V_v`$ induced by the metric $`h^V`$ on $`V`$. In the following, we identify $`\mathrm{\Lambda }V_v`$ with $`\mathrm{\Lambda }\overline{V}_v^{}`$ by using the Hermitian metric $`h^{V_v}`$ on $`V_v`$. By (2.15), as in (1.14), there is a natural isomorphism between $`๐_2`$-graded $`C(V)`$-Clifford modules over $`F`$,
(3.8) $`S(V_0^๐,_{0<v}(detV_v)^1)\widehat{}_{0<v}\mathrm{\Lambda }V_vS(V)_{|F}.`$
By using the above notations, we rewrite (2.41), on the fixed point set $`F`$, for $`p๐`$,
(3.12) $`\begin{array}{c}_p(X)=_{0<v}\left(_{n=1}^{\mathrm{}}\mathrm{Sym}(N_{v,n})_{\stackrel{n๐,}{n>pv}}\mathrm{Sym}(\overline{N}_{v,n})\right)_{n=1}^{\mathrm{}}\mathrm{Sym}(TY_n),\hfill \\ _p^{}(X)=_{\stackrel{0<v,n๐,}{0npv}}\left(\mathrm{Sym}(N_{v,n})detN_v\right),\hfill \\ ^p(X)=_p(X)_p^{}(X).\hfill \end{array}`$
Let $`V_0=V_0^๐_๐๐`$. From (2.12), (3.8), we get
(3.19) $`\begin{array}{c}^0(X)=_{n=1}^{\mathrm{}}\mathrm{Sym}\left(_{0<v}(N_{v,n}\overline{N}_{v,n})\right)_{n=1}^{\mathrm{}}\mathrm{Sym}(TY_n)\hfill \\ \mathrm{Sym}(_{0<v}N_{v,0})det(_{0<v}N_v),\hfill \\ F_V^1=_{n=1}^{\mathrm{}}\mathrm{\Lambda }(_{0<v}(V_{v,n}\overline{V}_{v,n})V_{0,n})\hfill \\ S(V_0^๐,_{0<v}(detV_v)^1)_{0<v}\mathrm{\Lambda }(V_{v,0}),\hfill \\ F_V^2=_{0<n๐+1/2}\mathrm{\Lambda }(_{0<v}(V_{v,n}\overline{V}_{v,n})V_{0,n}),\hfill \\ Q(W)=_{n=0}^{\mathrm{}}\mathrm{\Lambda }(_v\overline{W}_{v,n})_{n=1}^{\mathrm{}}\mathrm{\Lambda }(_vW_{v,n}).\hfill \end{array}`$
Now we can reformulate Theorem 2.4 as follows.
###### Theorem 3.1
For each $`\alpha `$, $`h,p๐`$, $`p>0`$, $`m\frac{1}{2}๐`$, for $`i=1,2`$, $`\tau =\tau _{e1}`$ or $`\tau _{s1}`$, we have the following identity in $`K_{G_y}(B)`$,
(3.24) $`\begin{array}{c}\mathrm{Ind}_\tau (D^{Y_\alpha }(K_WK_X^1)^{1/2}^p(X)F_V^iQ(W),\hfill \\ m+\frac{1}{2}p^2e(N)+\frac{1}{2}pd^{}(N),h)\hfill \\ =(1)^{pd^{}(W)}\mathrm{Ind}_\tau (D^{Y_\alpha }(K_WK_X^1)^{1/2}^0(X)F_V^i\hfill \\ Q(W)L^p,m+ph+p^2e,h).\hfill \end{array}`$
Proof : The rest of this section is devoted to a proof of Theorem 3.1. $`\mathrm{}`$
### 3.2 Proof of Theorem 3.1
Inspired by \[T, ยง7\], as in \[LiuMaZ, ยง3\], for $`p๐^{}`$, we define the shift operators,
(3.28) $`\begin{array}{c}r_{}:N_{v,n}N_{v,n+pv},r_{}:\overline{N}_{v,n}\overline{N}_{v,npv},\hfill \\ r_{}:W_{v,n}W_{v,n+pv},r_{}:\overline{W}_{v,n}\overline{W}_{v,npv},\hfill \\ r_{}:V_{v,n}V_{v,n+pv},r_{}:\overline{V}_{v,n}\overline{V}_{v,npv}.\hfill \end{array}`$
Recall that $`L(N),L(W),L(V)`$ are the complex line bundles over $`F`$ defined by (2.24). Recall also that $`L=L(N)^1L(W)L(V)`$ is a trivial complex line bundle over $`F`$, and $`gS^1`$ acts on it by multiplication by $`g^{2e}`$.
###### Proposition 3.1
For $`p๐`$, $`p>0`$, $`i=1,2`$, there are natural isomorphisms of vector bundles over $`F`$,
(3.31) $`\begin{array}{c}r_{}(^p(X))^0(X)L(N)^p,\hfill \\ r_{}(F_V^i)F_V^iL(V)^p.\hfill \end{array}`$
For any $`p๐`$, $`p>0`$, there is a natural $`G_y\times S^1`$-equivariant isomorpism of vector bundles over $`F`$,
(3.32) $`r_{}(Q(W))Q(W)L(W)^p.`$
Proof : The equation (3.31) was proved in \[LiuMaZ, Prop. 3.1\]. To prove (3.32), we only need to consider the shift operator on the following elements,
(3.34) $`\begin{array}{c}Q_W=_{n=0}^{\mathrm{}}\mathrm{\Lambda }(_{v0}\overline{W}_{v,n})_{n=1}^{\mathrm{}}\mathrm{\Lambda }(_{v0}W_{v,n}).\hfill \end{array}`$
We compute easily that
(3.36) $`\begin{array}{c}r_{}Q_W=_{n=0}^{\mathrm{}}\mathrm{\Lambda }(_{v0}\overline{W}_{v,npv})_{n=1}^{\mathrm{}}\mathrm{\Lambda }(_{v0}W_{v,n+pv}).\hfill \end{array}`$
Let $`h^W`$ be a Hermitian metric on $`W`$. Let $`h^{W_v}`$ be the metric on $`W_v`$ induced by $`h^W`$. As in \[LiuMaZ, ยง3\], the hermitian metric $`h^{W_v}`$ on $`W_v`$ induces a natural isomorphism of complex vector bundles over $`F`$,
(3.37) $`\mathrm{\Lambda }^i\overline{W}_v\mathrm{\Lambda }^{dimW_vi}W_vdet\overline{W}_v.`$
$``$ If $`v>0`$, for $`n๐,0n<pv`$, $`0idimW_v`$, (3.37) induces a natural $`G_y\times S^1`$-equivariant isomorphism of complex vector bundles
(3.39) $`\begin{array}{c}\mathrm{\Lambda }^i\overline{W}_{v,npv}\mathrm{\Lambda }^{dimW_vi}W_{v,n+pv}det\overline{W}_v.\hfill \end{array}`$
$``$ If $`v<0`$, for $`n๐,0<npv`$, $`0idimW_v`$, (3.37) induces a natural $`G_y\times S^1`$-equivariant isomorphism of complex vector bundles
(3.41) $`\begin{array}{c}\mathrm{\Lambda }^iW_{v,n+pv}\mathrm{\Lambda }^{dimW_vi}\overline{W}_{v,npv}(det\overline{W}_v)^1.\hfill \end{array}`$
From (2.24), (3.39) and (3.41), we have
(3.44) $`\begin{array}{c}{\displaystyle \underset{\stackrel{n๐,v>0,}{0n<pv}}{}}\mathrm{\Lambda }^{i_n}\overline{W}_{v,npv}{\displaystyle \underset{\stackrel{n๐,v<0,}{0<npv}}{}}\mathrm{\Lambda }^{i_n^{}}W_{v,n+pv}\hfill \\ {\displaystyle \underset{\stackrel{n๐,v>0,}{0n<pv}}{}}\mathrm{\Lambda }^{dimW_vi_n}W_{v,n+pv}{\displaystyle \underset{\stackrel{n๐,v<0,}{0<npv}}{}}\mathrm{\Lambda }^{dimW_vi_n^{}}\overline{W}_{v,npv}L(W)^p.\hfill \end{array}`$
From (3.36), (3.44), we get (3.32).
The proof of Proposition 3.1 is complete. $`\mathrm{}`$
###### Proposition 3.2
For $`p๐`$, $`p>0`$, $`i=1,2`$, the $`G_y`$-equivariant bundle isomorphism induced by (3.31) and (3.32),
(3.49) $`\begin{array}{c}r_{}:S(TY,K_X_{0<v}(detN_v)^1)(K_WK_X^1)^{1/2}\hfill \\ ^p(X)F_V^iQ(W)\hfill \\ S(TY,K_X_{0<v}(detN_v)^1)(K_WK_X^1)^{1/2}\hfill \\ ^0(X)F_V^iQ(W)L^p,\hfill \end{array}`$
verifies the following identities
(3.52) $`\begin{array}{c}r_{}^1๐_Hr_{}=๐_H,\hfill \\ r_{}^1Pr_{}=P+p๐_H+p^2e\frac{1}{2}p^2e(N)\frac{p}{2}d^{}(N).\hfill \end{array}`$
For the $`๐_2`$-gradings, we have
(3.55) $`\begin{array}{c}r_{}^1\tau _er_{}=\tau _e,r_{}^1\tau _sr_{}=\tau _s,\hfill \\ r_{}^1\tau _1r_{}=(1)^{pd^{}(W)}\tau _1.\hfill \end{array}`$
Proof : We divide the argument into several steps.
1) The first equation of (3.52) is obvious.
2) a) From \[LiuMaZ, (3.23)\] and (2.20), for $`i=1,2`$, on $`F_V^i`$, we have
(3.56) $`r_{}^1Pr_{}=P+p๐_H+{\displaystyle \frac{1}{2}}p^2e(V).`$
b) Note that on $`_{0<v,0npv}detN_v`$, $`๐_H`$ acts as $`pe(N)+d^{}(N)`$. On $`S(TY,K_Xdet(_{0<v}N_v)^1)(K_WK_X^1)^{1/2}`$, $`๐_H`$ acts as $`\frac{1}{2}d^{}(N)+\frac{1}{2}d^{}(W)`$. From (2.20), (3.12), on $`S(TY,K_Xdet(_{0<v}N_v)^1)(K_WK_X^1)^{1/2}^p(X)`$,
(3.57) $`r_{}^1Pr_{}=P+p๐_Hp^2e(N){\displaystyle \frac{1}{2}}p(d^{}(N)+d^{}(W)).`$
c) From (2.20), (3.44), on $`_{\stackrel{n๐,v>0,}{0n<pv}}\mathrm{\Lambda }^{i_n}\overline{W}_{v,n}_{\stackrel{n๐,v<0,}{0<npv}}\mathrm{\Lambda }^{i_n^{}}W_{v,n}`$, one has
(3.58)
(3.62) $`\begin{array}{c}r_{}^1Pr_{}={\displaystyle \underset{\stackrel{n๐,v>0,}{0n<pv}}{}}(dimW_vi_n)(n+pv)+{\displaystyle \underset{\stackrel{n๐,v<0,}{0<npv}}{}}(dimW_vi_n^{})(npv)\hfill \\ =P+p๐_H+{\displaystyle \underset{\stackrel{n๐,v>0,}{0n<pv}}{}}(dimW_v)(n+pv)+{\displaystyle \underset{\stackrel{n๐,v<0,}{0<npv}}{}}(dimW_v)(npv)\hfill \\ =P+p๐_H+\frac{1}{2}p^2e(W)+\frac{1}{2}pd^{}(W).\hfill \end{array}`$
From (2.31), (3.56), (3.57) and (3.58), we get the second equality of (3.52).
3) The first two identities of (3.55) were proved in \[LiuMaZ, Proposition 3.2\].
For the $`๐_2`$-grading $`\tau _1`$, it changes only on $`_{\stackrel{n๐,v>0,}{0n<pv}}\mathrm{\Lambda }^{i_n}\overline{W}_{v,n}_{\stackrel{n๐,v<0,}{0<npv}}\mathrm{\Lambda }^{i_n^{}}W_{v,n}`$. From (2.20), (3.44), we get the last equality of (3.55).
The proof of Proposition 3.2 is complete. $`\mathrm{}`$
Proof of Theorem 3.1 : From (2.31), (3.7) and Propositions 3.2, we easily obtain Theorem 3.1. $`\mathrm{}`$
## 4 Proof of Theorem 2.5
In this section, we prove Theorem 2.5. As in \[LiuMaZ, ยง4\], we will construct a family twisted Dirac operator on $`M(n_j)`$, the fixed point set of the induced $`๐_{n_j}`$ action on $`M`$. By applying our $`K`$-theory version of the equivariant family index theorem to this operator, we prove Theorem 2.5.
This section is organized as follows: In Section 4.1, we construct a family Dirac operator on $`M(n_j)`$. In Section 4.2, by introducing a shift operator, we will relate both sides of equation (2.71) to the index bundle of the family Dirac operator on $`M(n_j)`$. In Section 4.3, we prove Theorem 2.5.
In this section, we make the same assumptions and use the same notations as in Sections 2, 3.
### 4.1 The Spin<sup>c</sup> Dirac operator on $`M(n_j)`$
Let $`\pi :MB`$ be a fibration of compact manifolds with fiber $`X`$ and $`dim_๐X=2l`$. We assume that $`S^1`$ acts fiberwise on $`M`$, and $`TX`$ has an $`S^1`$-invariant Spin<sup>c</sup> structure. Let $`F=\{F_\alpha \}`$ be the fixed point set of the $`S^1`$-action on $`M`$. Then $`\pi :FB`$ is a fibration with compact fiber $`Y`$. For $`n๐,n>0`$, let $`๐_nS^1`$ denote the cyclic subgroup of order $`n`$.
Let $`V`$ be a real even dimensional vector bundle over $`M`$ with an $`S^1`$-invariant spin structure. Let $`W`$ be an $`S^1`$-equivariant complex vector bundle over $`M`$.
For $`n_j๐`$, $`n_j>0`$, let $`M(n_j)`$ be the fixed point set of the induced $`๐_{n_j}`$-action on $`M`$. Then $`\pi :M(n_j)B`$ is a fibration with compact fiber $`X(n_j)`$. Let $`N(n_j)M(n_j)`$ be the normal bundle to $`M(n_j)`$ in $`M`$. As in \[LiuMaZ, ยง4.1\], we see that $`N(n_j)`$ and $`V`$ can be decomposed, as real vector bundles over $`M(n_j)`$, to
(4.3) $`\begin{array}{c}N(n_j){\displaystyle \underset{0<v<n_j/2}{}}N(n_j)_vN(n_j)_{\frac{n_j}{2}}^๐,\hfill \\ V|_{M(n_j)}V(n_j)_0^๐{\displaystyle \underset{0<v<n_j/2}{}}V(n_j)_vV(n_j)_{\frac{n_j}{2}}^๐\hfill \end{array}`$
respectively. In (4.3), the last term is understood to be zero when $`n_j`$ is odd. We also denote by $`V(n_j)_0`$, $`V(n_j)_{\frac{n_j}{2}}`$, $`N(n_j)_{\frac{n_j}{2}}`$ the corresponding complexification of the real vector bundles $`V(n_j)_0^๐`$, $`V(n_j)_{\frac{n_j}{2}}^๐`$ and $`N(n_j)_{\frac{n_j}{2}}^๐`$ on $`M(n_j)`$. Then $`N(n_j)_v`$, $`V(n_j)_v`$โs are complex vector bundles over $`M(n_j)`$ with $`g๐_{n_j}`$ acting by $`g^v`$ on it.
Similarly, we also have the following $`๐_{n_j}`$-equivariant decomposition of $`W`$ on $`M(n_j)`$,
(4.4) $`W=_{0v<n_j}W(n_j)_v,`$
Here $`W(n_j)_v`$ is a complex vector bundle over $`M(n_j)`$ with $`g๐_{n_j}`$ acting by $`g^v`$ on it.
It is essential for us to know that the vector bundles $`TX(n_j)`$ and $`V(n_j)_0^๐`$ are orientable. For this we have the following lemma which generalizes \[BT, Lemmas 9.4, 10.1\],
###### Lemma 4.1
Let $`R`$ be a real, even dimensional orientable vector bundle over a manifold $`M`$. Let $`G`$ be a compact Lie group. We assume that $`G`$ acts on $`M`$, and lifts to $`R`$. We assume that $`R`$ has a $`G`$-invariant Spin<sup>c</sup> structure. For $`gG`$, let $`M^g`$ be the fixed point set of $`g`$ on $`M`$. Let $`R_0`$ be the subbundle of $`R`$ over $`M^g`$ on which $`g`$ acts trivially. Then $`R_0`$ is even dimensional and orientable.
Proof : Let $`h^R`$ be the metric on $`R`$ which is induced from the Spin<sup>c</sup> structure on $`R`$. As $`g`$ preserves the Spin<sup>c</sup> structure of $`R`$, $`g`$ is an isometry on $`R`$ and preserves the orientation of $`R`$. On $`M^g`$, we have the following decomposition of real vector bundles,
$`R=R_0R_1.`$
Since the only possible real eigenvalue of $`g`$ on $`R_1`$ is $`1`$, and $`det(g_{|R_1})=1`$ on $`M^g`$, we know that $`dim_๐R_1=dim_๐Rdim_๐R_0`$ must be even. So $`dim_๐R_0`$ is even.
Let $`K_R`$ be the $`G`$-equivariant complex line bundle over $`M`$ which is induced by the Spin<sup>c</sup> structure of $`R`$. Now the action of $`g`$ on the fiber of the complex spinor bundle $`S(R,K_R)`$ at $`xM^g`$ gives an element $`\stackrel{~}{g}\mathrm{Spin}^c(R_x)=\mathrm{Spin}(R_x)\times _{๐_2}S^1C(R_x)_๐๐`$, here $`C(R_x)`$ is the Clifford algebra of $`R_x`$. Let $`\stackrel{~}{g}=(\stackrel{~}{g}_1,s)`$, with $`\stackrel{~}{g}_1\mathrm{Spin}(R_x)`$, $`sS^1`$. Let $`\rho :\mathrm{Spin}(R_x)SO(R_x)`$ be the standard representation of $`\mathrm{Spin}(R_x)`$, then $`\rho (\stackrel{~}{g}_1)=g`$. So $`\stackrel{~}{g}c(a)=c(ga)\stackrel{~}{g}`$ for $`aR_x`$. Here we denote by $`c()`$ the Clifford action. This means that $`\stackrel{~}{g}`$ commutes $`c(a)`$ for $`aR_{0x}`$, so $`\stackrel{~}{g}\mathrm{Spin}^c(R_{1x})=\mathrm{Spin}(R_{1x})\times _{๐_2}S^1C(R_{1x})_๐๐`$ and $`\stackrel{~}{g}_1\mathrm{Spin}(R_{1x})`$.
Let $`e_1,\mathrm{},e_{2k}`$ be an orthonormal basis of $`R_{1x}`$, then $`e_{i_1}\mathrm{}e_{i_j}`$ $`(1i_1<\mathrm{}<i_j2k)`$ is an orthonormal basis of the complex vector space $`C(R_{1x})_๐๐`$. We define $`\sigma :C(R_{1x})_๐๐det(R_{1x})_๐๐`$ by
(4.7) $`\begin{array}{c}\sigma (e_{i_1}\mathrm{}e_{i_j})=e_1\mathrm{}e_{2k}\mathrm{if}j=2k=dim_๐R_1,\hfill \\ =0\mathrm{otherwise}.\hfill \end{array}`$
By \[BGV, Lemma 3.22\],
(4.8) $`|\sigma (\stackrel{~}{g})|=|\sigma (\stackrel{~}{g}_1)|=\stackrel{1/2}{det}((1g_{|R_1})/2).`$
So $`\sigma (\stackrel{~}{g})`$ is a nonvanishing section of $`det(R_1)_๐๐`$, $`det(R_1)_๐๐`$ is a trivial complex line bundle on $`F`$. So $`det(R_1)`$ is trivial, and $`R_1`$ is orientable. So $`R_0`$ is orientable.
This completes the proof of Lemma 4.1. $`\mathrm{}`$
By Lemma 4.1, $`TX(n_j)`$ and $`V(n_j)_0^๐`$ are even dimensional and orientable over $`M(n_j)`$. Thus $`N(n_j)`$ is orientable over $`M(n_j)`$. By (4.3), $`N(n_j)_{\frac{n_j}{2}}^๐`$ and $`V(n_j)_{\frac{n_j}{2}}^๐`$ are also even dimensional and orientable over $`M(n_j)`$. In the following, we fix the orientations of $`N(n_j)_{\frac{n_j}{2}}^๐`$ and $`V(n_j)_{\frac{n_j}{2}}^๐`$ over $`M(n_j)`$. We also fix the orienations of $`TX(n_j)`$ and $`V(n_j)_0^๐`$ which are induced by (4.3) and the orientations on $`TX,V`$, $`N(n_j)_{\frac{n_j}{2}}^๐`$ and $`V(n_j)_{\frac{n_j}{2}}^๐`$.
Let
(4.9) $`r(n_j)={\displaystyle \frac{1}{2}}(1+(1)^{n_j}).`$
###### Lemma 4.2
Assume that (2.9) holds. Let
(4.12) $`\begin{array}{c}L(n_j)=_{0<v<n_j/2}(det(N(n_j)_v)det(\overline{V(n_j)_v})\hfill \\ det(\overline{W(n_j)}_v)det(W(n_j)_{n_jv}))^{(r(n_j)+1)v}\hfill \end{array}`$
be the complex line bundle over $`M(n_j)`$. Then we have
i) $`L(n_j)`$ has an $`n_j^{\mathrm{th}}`$ root over $`M(n_j)`$.
ii) Let
(4.16) $`\begin{array}{c}L_1=K_X_{0<v<n_j/2}\left(det(N(n_j)_v)det(\overline{V(n_j)_v})\right)\hfill \\ det(W(n_j)_{n_j/2})L(n_j)^{r(n_j)/n_j},\hfill \\ L_2=K_X_{0<v<n_j/2}\left(det(N(n_j)_v)\right)det(W(n_j)_{n_j/2})L(n_j)^{r(n_j)/n_j}.\hfill \end{array}`$
Let $`U_1=TX(n_j)V(n_j)_0^๐`$ and $`U_2=TX(n_j)V(n_j)_{\frac{n_j}{2}}^๐`$. Then $`U_1`$ (resp. $`U_2`$) has a Spin<sup>c</sup> structure defined by $`L_1`$ (resp. $`L_2`$).
Proof : Both statements follow from the proof of \[BT, Lemmas 11.3 and 11.4\]. $`\mathrm{}`$
Lemma 4.2 allows us, as we are going to see, to apply the constructions and results in Section 1.1 to the fibration $`M(n_j)B`$, which is the main concern of this section.
For $`p_j๐`$, $`p_j<n_j`$, $`(p_j,n_j)=1`$, $`\beta _j=\frac{p_j}{n_j}`$, let us write
(4.19) $`\begin{array}{c}(\beta _j)=_{0<n๐}\mathrm{Sym}(TX(n_j)_n)_{0<v<n_j/2}\mathrm{Sym}(_{0<n๐+p_jv/n_j}N(n_j)_{v,n}\hfill \\ _{0<n๐p_jv/n_j}\overline{N(n_j)}_{v,n})_{0<n๐+\frac{1}{2}}\mathrm{Sym}(N(n_j)_{\frac{n_j}{2},n}),\hfill \end{array}`$
(4.25) $`\begin{array}{c}F_V^1(\beta _j)=\mathrm{\Lambda }(_{0<n๐}V(n_j)_{0,n}_{0<v<n_j/2}(_{0<n๐+p_jv/n_j}V(n_j)_{v,n}\hfill \\ _{0<n๐p_jv/n_j}\overline{V(n_j)}_{v,n})_{0<n๐+\frac{1}{2}}V(n_j)_{\frac{n_j}{2},n}),\hfill \\ F_V^2(\beta _j)=\mathrm{\Lambda }(_{0<n๐}V(n_j)_{\frac{n_j}{2},n}_{0<v<n_j/2}(_{0<n๐+p_jv/n_j+\frac{1}{2}}V(n_j)_{v,n}\hfill \\ _{0<n๐p_jv/n_j+\frac{1}{2}}\overline{V(n_j)}_{v,n})_{0<n๐+\frac{1}{2}}V(n_j)_{0,n}),\hfill \\ Q_W(\beta _j)=\mathrm{\Lambda }\left(_{0v<n_j}\left(_{0<n๐+p_jv/n_j}W(n_j)_{v,n}_{0n๐p_jv/n_j}\overline{W(n_j)}_{v,n}\right)\right).\hfill \end{array}`$
We denote by $`D^{X(n_j)}`$ the $`S^1`$-equivariant Spin<sup>c</sup>-Dirac operator on $`S(U_1,L_1)`$ or $`S(U_2,L_2)`$ along the fiber $`X(n_j)`$ defined as in Section 1.1. We denote by $`D^{X(n_j)}(\beta _j)F_V^i(\beta _j)Q_W(\beta _j)`$ $`(i=1,2)`$ the corresponding twisted Spin<sup>c</sup> Dirac operator on $`S(U_i,L_i)(\beta _j)F_V^i(\beta _j)Q_W(\beta _j)`$ along the fiber $`X(n_j)`$.
Remark 4.1. In fact, to define an $`S^1`$ (resp. $`G_y`$)-action on $`L(n_j)^{r(n_j)/n_j}`$, one must replace the $`S^1`$-action by its $`n_j`$-fold action (resp. the $`G_y`$-action by $`G_{y^{1/n_j}}`$-action). Here by abusing notation, we still say an $`S^1`$ (resp. $`G_y`$)-action without causing any confusion.
In the rest of this subsection, we will reinterpret all of the above objects when we restrict ourselves to $`F`$, the fixed point set of the $`S^1`$ action. We will use the notation of Sections 1.1 and 2.
Let $`N_{F/M(n_j)}`$ be the normal bundle to $`F`$ in $`M(n_j)`$. Then by (2.12),
(4.28) $`\begin{array}{c}N_{F/M(n_j)}=_{0<v:vn_j๐}N_v,\hfill \\ TX(n_j)_๐๐=TY_๐๐_{0<v,vn_j๐}(N_v\overline{N}_v).\hfill \end{array}`$
By (2.12), (2.15) and (4.3), the restriction to $`F`$ of $`N(n_j)_v`$, $`V(n_j)_v`$ $`(1vn_j/2)`$ is given by
(4.31) $`\begin{array}{c}N(n_j)_v={\displaystyle \underset{0<v^{}:v^{}=v\mathrm{mod}(n_j)}{}}N_v^{}{\displaystyle \underset{0<v^{}:v^{}=v\mathrm{mod}(n_j)}{}}\overline{N}_v^{},\hfill \\ V(n_j)_v={\displaystyle \underset{0<v^{}:v^{}=v\mathrm{mod}(n_j)}{}}V_v^{}{\displaystyle \underset{0<v^{}:v^{}=v\mathrm{mod}(n_j)}{}}\overline{V}_v^{}.\hfill \end{array}`$
And
(4.32) $`V(n_j)_0=V_0^๐_๐๐{\displaystyle \underset{0<v,v=0\mathrm{mod}(n_j)}{}}(V_v\overline{V}_v).`$
By (4.28)-(4.32), we have the following identifications of real vector bundles over $`F`$,
(4.37) $`\begin{array}{c}N(n_j)_{\frac{n_j}{2}}^๐=_{0<v,v=\frac{n_j}{2}\mathrm{mod}(n_j)}N_v,\hfill \\ TX(n_j)=TY_{0<v,v=0\mathrm{mod}(n_j)}N_v,\hfill \\ V(n_j)_0^๐=V_0^๐_{0<v,v=0\mathrm{mod}(n_j)}V_v,\hfill \\ V(n_j)_{\frac{n_j}{2}}^๐=_{0<v,v=\frac{n_j}{2}\mathrm{mod}(n_j)}V_v.\hfill \end{array}`$
By (2.15) and (4.4), the restriction to $`F`$ of $`W(n_j)_v`$ $`(0v<n_j)`$ is given by
(4.38) $`W(n_j)_v=_{v^{}=v\mathrm{mod}(n_j)}W_v^{}.`$
We denote by $`V_0=V_0^๐_๐๐`$ the complexification of $`V_0^๐`$ over $`F`$. As $`(p_j,n_j)=1`$, we know that for $`v๐`$, $`p_jv/n_j๐`$ iff $`v/n_j๐`$. Also, $`p_jv/n_j๐+\frac{1}{2}`$ iff $`v/n_j๐+\frac{1}{2}`$. From (4.28)-(4.38), we then get
(4.39)
(4.43) $`\begin{array}{c}(\beta _j)=_{0<n๐}\mathrm{Sym}(TY_n)_{0<v,v=0,\frac{n_j}{2}\mathrm{mod}(n_j)}_{0<n๐+\frac{p_jv}{n_j}}\mathrm{Sym}(N_{v,n}\overline{N}_{v,n})\hfill \\ _{0<v^{}<n_j/2}\mathrm{Sym}(_{v=v^{}\mathrm{mod}(n_j)}\left(_{0<n๐+\frac{p_jv}{n_j}}N_{v,n}_{0<n๐\frac{p_jv}{n_j}}\overline{N}_{v,n}\right)\hfill \\ _{v=v^{}\mathrm{mod}(n_j)}\left(_{0<n๐+\frac{p_jv}{n_j}}N_{v,n}_{0<n๐\frac{p_jv}{n_j}}\overline{N}_{v,n}\right)),\hfill \end{array}`$
(4.49) $`\begin{array}{c}F_V^1(\beta _j)=\mathrm{\Lambda }[_{0<n๐}V_{0,n}_{0<v,v=0,\frac{n_j}{2}\mathrm{mod}(n_j)}\left(_{0<n๐+\frac{p_jv}{n_j}}V_{v,n}_{0<n๐\frac{p_jv}{n_j}}\overline{V}_{v,n}\right)\hfill \\ _{0<v^{}<n_j/2}\left(_{v=v^{},v^{}\mathrm{mod}(n_j)}\left(_{0<n๐+\frac{p_jv}{n_j}}V_{v,n}_{0<n๐\frac{p_jv}{n_j}}\overline{V}_{v,n}\right)\right)],\hfill \\ F_V^2(\beta _j)=\mathrm{\Lambda }[_{0<n๐+\frac{1}{2}}V_{0,n}_{0<v,v=0,\frac{n_j}{2}\mathrm{mod}(n_j)}\left(_{0<n๐+\frac{p_jv}{n_j}+\frac{1}{2}}V_{v,n}_{0<n๐\frac{p_jv}{n_j}+\frac{1}{2}}\overline{V}_{v,n}\right)\hfill \\ _{0<v^{}<n_j/2}\left(_{v=v^{},v^{}\mathrm{mod}(n_j)}\left(_{0<n๐+\frac{p_jv}{n_j}+\frac{1}{2}}V_{v,n}_{0<n๐\frac{p_jv}{n_j}+\frac{1}{2}}\overline{V}_{v,n}\right)\right)],\hfill \\ Q_W(\beta _j)=\mathrm{\Lambda }\left(_v\left(_{0<n๐+p_jv/n_j}W_{v,n}_{0n๐p_jv/n_j}\overline{W}_{v,n}\right)\right).\hfill \end{array}`$
Now, we want to compare the spinor bundles over $`F`$. From (4.12), (4.16), (4.31) and (4.38), we get that over $`F`$ we have the identities
(4.56) $`\begin{array}{c}L(n_j)^{\frac{r(n_j)}{n_j}}=_{0<v^{}<n_j/2}(_{v=v^{}\mathrm{mod}(n_j)}(detN_vdet\overline{V}_vdet\overline{W}_v)^{2v^{}}\hfill \\ _{v=v^{}\mathrm{mod}(n_j)}(detN_vdet\overline{V}_vdet\overline{W}_v)^{2v^{}})^{r(n_j)/n_j},\hfill \\ L_1=K_XL(n_j)^{r(n_j)/n_j}_{0<v^{}<n_j/2}(_{v=v^{}\mathrm{mod}(n_j)}(detN_vdet\overline{V}_v)\hfill \\ _{v=v^{}\mathrm{mod}(n_j)}(detN_vdet\overline{V}_v)^1)_{v=\frac{n_j}{2}\mathrm{mod}(n_j)}detW_v,\hfill \\ L_2=K_XL(n_j)^{r(n_j)/n_j}_{0<v^{}<n_j/2}(_{v=v^{}\mathrm{mod}(n_j)}detN_v\hfill \\ _{v=v^{}\mathrm{mod}(n_j)}(detN_v)^1)_{v=\frac{n_j}{2}\mathrm{mod}(n_j)}detW_v.\hfill \end{array}`$
From (4.37), we have, over $`F`$,
(4.59) $`\begin{array}{c}TX(n_j)V(n_j)_0^๐=TYV_0^๐_{0<v,v=0\mathrm{mod}(n_j)}(N_vV_v),\hfill \\ TX(n_j)V(n_j)_{\frac{n_j}{2}}^๐=TY_{0<v,v=0\mathrm{mod}(n_j)}N_v_{0<v,v=\frac{n_j}{2}\mathrm{mod}(n_j)}V_v.\hfill \end{array}`$
Recall that the Spin<sup>c</sup> vector bundles $`U_1`$, $`U_2`$ have been defined in Lemma 4.2. Denote by
(4.60)
(4.63) $`\begin{array}{c}S(U_1,L_1)^{}=S(TYV_0^๐,L_1{\displaystyle \underset{\stackrel{0<v,}{v=0\mathrm{mod}(n_j)}}{}}(detN_vdetV_v)^1){\displaystyle \underset{\stackrel{0<v,}{v=0\mathrm{mod}(n_j)}}{}}\mathrm{\Lambda }V_v,\hfill \\ S(U_2,L_2)^{}=S(TY,L_2{\displaystyle \underset{\stackrel{0<v,}{v=0\mathrm{mod}(n_j)}}{}}(detN_v)^1{\displaystyle \underset{\stackrel{0<v,}{v=\frac{n_j}{2}\mathrm{mod}(n_j)}}{}}(detV_v)^1){\displaystyle \underset{\stackrel{0<v,}{v=\frac{n_j}{2}\mathrm{mod}(n_j)}}{}}\mathrm{\Lambda }V_v.\hfill \end{array}`$
Then from (1.14) and (4.60), for $`i=1,2`$, we have the following isomorphism of Clifford modules over $`F`$,
(4.65) $`\begin{array}{c}S(U_i,L_i)S(U_i,L_i)^{}\mathrm{\Lambda }(_{0<v,v=0\mathrm{mod}(n_j)}N_v).\hfill \end{array}`$
We define the $`๐_2`$ gradings on $`S(U_i,L_i)^{}(i=1,2)`$ induced by the $`๐_2`$-gradings on $`S(U_i,L_i)`$ $`(i=1,2)`$ and on $`\mathrm{\Lambda }(_{0<v,v=0\mathrm{mod}(n_j)}N_v)`$ such that the isomorphism (4.65) preserves the $`๐_2`$-grading.
We introduce formally the following complex line bundles over $`F`$,
(4.68) $`\begin{array}{c}L_1^{}=\left[L_1^1_{\stackrel{0<v,}{v=0\mathrm{mod}(n_j)}}(detN_vdetV_v)_{0<v}(detN_vdetV_v)^1K_X\right]^{1/2},\hfill \\ L_2^{}=\left[L_2^1_{\stackrel{0<v,}{v=0\mathrm{mod}(n_j)}}detN_v_{\stackrel{0<v,}{v=n_j/2\mathrm{mod}(n_j)}}detV_v_{0<v}(detN_v)^1K_X\right]^{1/2}.\hfill \end{array}`$
From (1.14), Lemma 4.2 and the assumption that $`V`$ is spin, one verifies easily that $`c_1(L_{i}^{}{}_{}{}^{2})=0\mathrm{mod}(2)`$ for $`i=1,2`$. Thus $`L_1^{},L_2^{}`$ are well defined complex line bundles over $`F`$. For the later use, we also write down the following expressions of $`L_i^{}`$ ($`i=1,2`$) which can be deduced from (4.56):
(4.73) $`\begin{array}{c}L_1^{}=\left[L(n_j)^{r(n_j)/n_j}_{v=\frac{n_j}{2}\mathrm{mod}(n_j)}(detN_vdet\overline{V}_vdet\overline{W}_v)\right]^{\frac{1}{2}}\hfill \\ _{0<v\frac{n_j}{2}\mathrm{mod}(n_j)}(detN_v)^1_{\frac{n_j}{2}<v<n_j\mathrm{mod}(n_j)}(detV_v)^1,\hfill \\ L_2^{}=\left[L(n_j)^{r(n_j)/n_j}_{v=\frac{n_j}{2}\mathrm{mod}(n_j)}(detN_vdetV_vdet\overline{W}_v)\right]^{\frac{1}{2}}\hfill \\ _{0<v\frac{n_j}{2}\mathrm{mod}(n_j)}(detN_v)^1.\hfill \end{array}`$
From (4.56), (4.60), and the definition of $`L_i^{}`$ $`(i=1,2)`$, we get the following identifications of Clifford modules over $`F`$,
(4.77) $`\begin{array}{c}S(U_1,L_1)^{}L_1^{}=S(TY,K_X_{0<v}(detN_v)^1)S(V_0^๐,_{0<v}(detV_v)^1)\hfill \\ \mathrm{\Lambda }(_{0<v,v=0\mathrm{mod}(n_j)}V_v),\hfill \\ S(U_2,L_2)^{}L_2^{}=S(TY,K_X_{0<v}(detN_v)^1)\mathrm{\Lambda }(_{0<v,v=\frac{n_j}{2}\mathrm{mod}(n_j)}V_v).\hfill \end{array}`$
Let
(4.80) $`\begin{array}{c}\mathrm{\Delta }(n_j,N)={\displaystyle \underset{\frac{n_j}{2}<v^{}<n_j}{}}{\displaystyle \underset{0<v=v^{}\mathrm{mod}(n_j)}{}}dimN_v+o(N(n_j)_{\frac{n_j}{2}}^๐),\hfill \\ \mathrm{\Delta }(n_j,V)={\displaystyle \underset{\frac{n_j}{2}<v^{}<n_j}{}}{\displaystyle \underset{0<v=v^{}\mathrm{mod}(n_j)}{}}dimV_v+o(V(n_j)_{\frac{n_j}{2}}^๐),\hfill \end{array}`$
with $`o(N(n_j)_{\frac{n_j}{2}}^๐)=0\mathrm{or}1`$ (resp. $`o(V(n_j)_{\frac{n_j}{2}}^๐)=0\mathrm{or}1`$), depending on whether the given orientation on $`N(n_j)_{\frac{n_j}{2}}^๐`$ ( resp. $`V(n_j)_{\frac{n_j}{2}}^๐`$) agrees or disagrees with the complex orientation of $`_{v=\frac{n_j}{2}\mathrm{mod}(n_j)}N_v`$ (resp. $`_{v=\frac{n_j}{2}\mathrm{mod}(n_j)}V_v`$).
By \[LiuMaZ, ยง4.1\], (4.38) and (4.65), for the $`๐_2`$-gradings induced by $`\tau _s`$, the difference of the $`๐_2`$-gradings of (4.77) is $`(1)^{\mathrm{\Delta }(n_j,N)}`$; for the $`๐_2`$-gradings induced by $`\tau _e`$, the difference of the $`๐_2`$-gradings of the first (resp. second) equation of (4.77) is $`(1)^{\mathrm{\Delta }(n_j,N)+\mathrm{\Delta }(n_j,V)}`$ (resp. $`(1)^{\mathrm{\Delta }(n_j,N)+o(V(n_j)_{\frac{n_j}{2}}^๐)}`$).
### 4.2 The Shift operators
Let $`p๐^{}`$ be fixed. For any $`1jJ_0`$, inspired by \[T, ยง9\], as in \[LiuMaZ, ยง4\], we define the following shift operators $`r_j`$:
(4.84) $`\begin{array}{c}r_j:N_{v,n}N_{v,n+(p1)v+p_jv/n_j},r_j:\overline{N}_{v,n}\overline{N}_{v,n(p1)vp_jv/n_j},\hfill \\ r_j:W_{v,n}W_{v,n+(p1)v+p_jv/n_j},r_j:\overline{W}_{v,n}\overline{W}_{v,n(p1)vp_jv/n_j},\hfill \\ r_j:V_{v,n}V_{v,n+(p1)v+p_jv/n_j},r_j:\overline{V}_{v,n}\overline{V}_{v,n(p1)vp_jv/n_j}.\hfill \end{array}`$
If $`E`$ is a combination of the above bundles, we denote by $`r_jE`$ the bundle on which the action of $`P`$ is changed in the above way.
Recall that the vector bundles $`F_V^i`$ $`(i=1,2)`$ have been defined in (3.19). From (2.56), we see that
(4.87) $`\begin{array}{c}_{p,j}(X)=_p(X)_{p1}^{}(X)_{(v,n)_{i=1}^jI_i^p}\left(\mathrm{Sym}(N_{v,n})detN_v\right)\hfill \\ _{(v,n)\overline{I}_j^p}\mathrm{Sym}(\overline{N}_{v,n}).\hfill \end{array}`$
###### Proposition 4.1
There are natural isomorphisms of vector bundles over $`F`$,
(4.96) $`\begin{array}{c}r_j_{p,j1}(X)(\beta _j)_{0<v,v=0\mathrm{mod}(n_j)}\mathrm{Sym}(\overline{N}_{v,0})\hfill \\ _{0<v}(detN_v)^{[\frac{p_jv}{n_j}]+(p1)v+1}_{0<v,v=0\mathrm{mod}(n_j)}(detN_v)^1,\hfill \\ r_j_{p,j}(X)(\beta _j)_{0<v,v=0\mathrm{mod}(n_j)}\mathrm{Sym}(N_{v,0})_{0<v}(detN_v)^{[\frac{p_jv}{n_j}]+(p1)v+1},\hfill \\ r_jF_V^1S(V_0^๐,_{0<v}(detV_v)^1)F_V^1(\beta _j)_{0<v,v=0\mathrm{mod}(n_j)}\mathrm{\Lambda }(V_{v,0})\hfill \\ _{0<v}(det\overline{V}_v)^{[\frac{p_jv}{n_j}]+(p1)v},\hfill \\ r_jF_V^2F_V^2(\beta _j)_{0<v,v=\frac{n_j}{2}\mathrm{mod}(n_j)}\mathrm{\Lambda }(V_{v,0})_{0<v}(det\overline{V}_v)^{[\frac{p_jv}{n_j}+\frac{1}{2}]+(p1)v},\hfill \\ r_jQ(W)Q_W(\beta _j)_{0<v}(det\overline{W}_v)^{[\frac{p_jv}{n_j}]+(p1)v+1}_{0<v,v=0\mathrm{mod}(n_j)}(det\overline{W}_v)^1\hfill \\ _{v<0}(detW_v)^{[\frac{p_jv}{n_j}](p1)v}.\hfill \end{array}`$
Proof : The proof is similar to the proof of Proposition 3.1.
Note that, by (2.51), for $`vJ=\{v๐|`$ There exists $`\alpha `$ such that $`N_v0`$ on $`F_\alpha \}`$, there are no integer in $`]\frac{p_{j1}v}{n_{j1}},\frac{p_jv}{n_j}[`$. So for $`vJ`$, the elements $`(v,n)_{i=1}^{i_0}I_i^p`$ are $`(v,(p1)v+1)`$, $`\mathrm{},(v,(p1)v+[\frac{p_{i_0}v}{n_{i_0}}])`$ for $`i_0=j1,j`$. Furthermore,
(4.99) $`\begin{array}{c}[{\displaystyle \frac{p_{j1}v}{n_{j1}}}]=[{\displaystyle \frac{p_jv}{n_j}}]1\mathrm{if}v=0\mathrm{mod}(n_j),\hfill \\ [{\displaystyle \frac{p_{j1}v}{n_{j1}}}]=[{\displaystyle \frac{p_jv}{n_j}}]\mathrm{if}v0\mathrm{mod}(n_j).\hfill \end{array}`$
By using (3.19), (4.84), (4.87), (4.99), we can prove the first four equalities of (4.96) as in the proof of \[LiuMaZ, Proposition 4.1\].
From (3.37), we have the natural $`G_y\times S^1`$-equivariant isomorphisms of complex vector bundles over $`F`$,
(4.104) $`\begin{array}{c}{\displaystyle \underset{\stackrel{n๐,v>0,}{0n<(p1)v+\frac{p_jv}{n_j}}}{}}\mathrm{\Lambda }^{i_n}\overline{W}_{v,n(p1)v\frac{p_jv}{n_j}}{\displaystyle \underset{\stackrel{n๐,v>0,}{0n<(p1)v+\frac{p_jv}{n_j}}}{}}\mathrm{\Lambda }^{dimW_vi_n}W_{v,n+(p1)v+\frac{p_jv}{n_j}}\hfill \\ {\displaystyle \underset{0<v}{}}(det\overline{W}_v)^{[\frac{p_jv}{n_j}]+(p1)v+1}{\displaystyle \underset{0<v,v=0\mathrm{mod}(n_j)}{}}(det\overline{W}_v)^1,\hfill \\ {\displaystyle \underset{\stackrel{n๐,v<0,}{0<n(p1)v\frac{p_jv}{n_j}}}{}}\mathrm{\Lambda }^{i_n}W_{v,n+(p1)v+\frac{p_jv}{n_j}}{\displaystyle \underset{\stackrel{n๐,v<0,}{0<n(p1)v\frac{p_jv}{n_j}}}{}}\mathrm{\Lambda }^{dimW_vi_n}\overline{W}_{v,n(p1)v\frac{p_jv}{n_j}}\hfill \\ {\displaystyle \underset{v<0}{}}(detW_v)^{[\frac{p_jv}{n_j}](p1)v}.\hfill \end{array}`$
From (3.19), (4.39), (4.104), we get the last equation of (4.96).
The proof of Proposition 4.1 is complete. $`\mathrm{}`$
###### Lemma 4.3
Let us write
(4.111) $`\begin{array}{c}L(\beta _j)_1=L_1^{}_{0<v}(detN_v)^{[\frac{p_jv}{n_j}]+(p1)v+1}_{0<v}(det\overline{V}_v)^{[\frac{p_jv}{n_j}]+(p1)v}\hfill \\ _{0<v,v=0\mathrm{mod}(n_j)}(detN_v)^1_{v<0}(detW_v)^{[\frac{p_jv}{n_j}](p1)v}\hfill \\ _{0<v}(det\overline{W}_v)^{[\frac{p_jv}{n_j}]+(p1)v+1}_{0<v,v=0\mathrm{mod}(n_j)}(det\overline{W}_v)^1,\hfill \\ L(\beta _j)_2=L_2^{}_{0<v}(detN_v)^{[\frac{p_jv}{n_j}]+(p1)v+1}_{0<v}(det\overline{V}_v)^{[\frac{p_jv}{n_j}+\frac{1}{2}]+(p1)v}\hfill \\ _{0<v,v=0\mathrm{mod}(n_j)}(detN_v)^1_{v<0}(detW_v)^{[\frac{p_jv}{n_j}](p1)v}\hfill \\ _{0<v}(det\overline{W}_v)^{[\frac{p_jv}{n_j}]+(p1)v+1}_{0<v,v=0\mathrm{mod}(n_j)}(det\overline{W}_v)^1.\hfill \end{array}`$
Then $`L(\beta _j)_1,L(\beta _j)_2`$ can be extended naturally to $`G_y\times S^1`$-equivariant complex line bundles which we will still denote by $`L(\beta _j)_1,L(\beta _j)_2`$ respectively over $`M(n_j)`$.
Proof : Write
(4.112) $`[{\displaystyle \frac{p_jv}{n_j}}]={\displaystyle \frac{p_jv}{n_j}}{\displaystyle \frac{\omega (v)}{n_j}}.`$
Note that for $`v=\frac{n_j}{2}\mathrm{mod}(n_j)`$, $`\frac{\omega (v)}{n_j}=\frac{1}{2}`$.
We introduce the following line bundle over $`M(n_j)`$,
(4.115) $`\begin{array}{c}L^\omega (\beta _j)=_{0<v<\frac{n_j}{2}}(det(N(n_j)_v)det(\overline{V(n_j)}_v)\hfill \\ det(\overline{W(n_j)}_v)det(W(n_j)_{n_jv}))^{\omega (v)r(n_j)v}.\hfill \end{array}`$
As in \[LiuMaZ, (4.38)\], Lemma 4.2 implies $`L^\omega (\beta _j)^{1/n_j}`$ is well defined over $`M(n_j)`$.
The contributions of $`N`$ and $`V`$ in $`L(\beta _j)_1,L(\beta _j)_2`$ are the same as given in \[LiuMaZ, Lemma 4.2\], we only need to calculate the contribution of $`W`$ in $`L(\beta _j)_1,L(\beta _j)_2`$. Actually from \[LiuMaZ, (4.37), (4.44)\], (2.24), (4.38), (4.73), (4.111), (4.112) and (4.115), we get
(4.119) $`\begin{array}{c}L(\beta _j)_1=L^{(p1)p_j/n_j}L^\omega (\beta _j)^{1/n_j}{\displaystyle \underset{0<v\frac{n_j}{2}}{}}det(\overline{W(n_j)}_v),\hfill \\ L(\beta _j)_2=L^{(p1)\frac{p_j}{n_j}}L^\omega (\beta _j)^{\frac{1}{n_j}}{\displaystyle \underset{0<v\frac{n_j}{2}}{}}det(\overline{W(n_j)}_v)\hfill \\ {\displaystyle \underset{1mp_j/2}{}}{\displaystyle \underset{m\frac{1}{2}<p_jv^{}/n_j<m}{}}det(\overline{V(n_j)}_v^{}).\hfill \end{array}`$
The proof of Lemma 4.3 is complete. $`\mathrm{}`$
Let us write
(4.124) $`\begin{array}{c}\epsilon (W)=\frac{1}{2}_{0<v}(dimW_v)[([\frac{p_jv}{n_j}]+(p1)v)([\frac{p_jv}{n_j}]+(p1)v+1)\hfill \\ (\frac{p_jv}{n_j}+(p1)v)(2([\frac{p_jv}{n_j}]+(p1)v)+1)]\hfill \\ \frac{1}{2}_{v<0}(dimW_v)[([\frac{p_jv}{n_j}](p1)v)([\frac{p_jv}{n_j}](p1)v+1)\hfill \\ +(\frac{p_jv}{n_j}+(p1)v)(2([\frac{p_jv}{n_j}](p1)v)+1)],\hfill \end{array}`$
(4.131) $`\begin{array}{c}\epsilon _1=\frac{1}{2}_{0<v}(dimN_vdimV_v)[([\frac{p_jv}{n_j}]+(p1)v)([\frac{p_jv}{n_j}]+(p1)v+1)\hfill \\ (\frac{p_jv}{n_j}+(p1)v)(2([\frac{p_jv}{n_j}]+(p1)v)+1)],\hfill \\ \epsilon _2=\frac{1}{2}_{0<v}(dimN_v)[([\frac{p_jv}{n_j}]+(p1)v)([\frac{p_jv}{n_j}]+(p1)v+1)\hfill \\ (\frac{p_jv}{n_j}+(p1)v)(2([\frac{p_jv}{n_j}]+(p1)v)+1)]\hfill \\ \frac{1}{2}_{0<v}(dimV_v)[([\frac{p_jv}{n_j}+\frac{1}{2}]+(p1)v)^2\hfill \\ 2(\frac{p_jv}{n_j}+(p1)v)([\frac{p_jv}{n_j}+\frac{1}{2}]+(p1)v)].\hfill \end{array}`$
Then $`\epsilon (W),\epsilon _1,\epsilon _2`$ are locally constant functions on $`F`$.
Recall that the involutions $`\tau _e,\tau _s`$ and $`\tau _1`$ were defined in Section 3.1. Also recall that if $`E`$ is a $`S^1`$-equivariant vector bundle over $`M`$, then the weight of the $`S^1`$-action on $`\mathrm{\Gamma }(F,E)`$ is given by the action $`๐_H`$ (cf. ยง3.1).
###### Proposition 4.2
For $`i=1,2`$, the $`G_y`$-equivariant isomorphisms induced by (4.77) and (4.96),
(4.140) $`\begin{array}{c}r_{i1}:S(TY,K_X_{0<v}(detN_v)^1)(K_WK_X^1)^{1/2}\hfill \\ _{p,j1}(X)F_V^iQ(W)\hfill \\ S(U_i,L_i)^{}(K_WK_X^1)^{1/2}(\beta _j)F_V^i(\beta _j)\hfill \\ Q_W(\beta _j)L(\beta _j)_i_{\stackrel{0<v,}{v=0\mathrm{m}\mathrm{o}\mathrm{d}(n_j)}}\mathrm{Sym}(\overline{N}_{v,0}),\hfill \\ r_{i2}:S(TY,K_X_{0<v}(detN_v)^1)(K_WK_X^1)^{1/2}\hfill \\ _{p,j}(X)F_V^iQ(W)\hfill \\ S(U_i,L_i)^{}(K_WK_X^1)^{1/2}(\beta _j)F_V^i(\beta _j)\hfill \\ Q_W(\beta _j)L(\beta _j)_i_{\stackrel{0<v,}{v=0\mathrm{mod}(n_j)}}(\mathrm{Sym}(N_{v,0})detN_v),\hfill \end{array}`$
have the following properties : 1) for $`i=1,2`$, $`\gamma =1,2`$,
(4.143) $`\begin{array}{c}r_{i\gamma }^1๐_Hr_{i\gamma }=๐_H,\hfill \\ r_{i\gamma }^1Pr_{i\gamma }=P+(\frac{p_j}{n_j}+(p1))๐_H+\epsilon _{i\gamma },\hfill \end{array}`$
where
(4.146) $`\begin{array}{c}\epsilon _{i1}=\epsilon _i+\epsilon (W)e(p,\beta _{j1},N),\hfill \\ \epsilon _{i2}=\epsilon _i+\epsilon (W)e(p,\beta _j,N).\hfill \end{array}`$
2) Recall that $`o(V(n_j)_{\frac{n_j}{2}}^๐)`$ was defined in (4.80). Let
(4.151) $`\begin{array}{c}\mu _1=_{0<v}[\frac{p_jv}{n_j}]dimV_v+\mathrm{\Delta }(n_j,N)+\mathrm{\Delta }(n_j,V)\mathrm{mod}(2),\hfill \\ \mu _2=_{0<v}[\frac{p_jv}{n_j}+\frac{1}{2}]dimV_v+\mathrm{\Delta }(n_j,N)+o(V(n_j)_{\frac{n_j}{2}}^๐)\mathrm{mod}(2),\hfill \\ \mu _3=\mathrm{\Delta }(n_j,N)\mathrm{mod}(2),\hfill \\ \mu _4=_v(dimW_v)([\frac{p_jv}{n_j}]+(p1)v)+dimW+dimW(n_j)_0\mathrm{mod}(2).\hfill \end{array}`$
Then for $`i=1,2`$; $`\gamma =1,2`$,
(4.154) $`\begin{array}{c}r_{i\gamma }^1\tau _er_{i\gamma }=(1)^{\mu _i}\tau _e,r_{i\gamma }^1\tau _sr_{i\gamma }=(1)^{\mu _3}\tau _s,\hfill \\ r_{i\gamma }^1\tau _1r_{i\gamma }=(1)^{\mu _4}\tau _1.\hfill \end{array}`$
Proof : The first equality of (4.143) is trivial. From (2.62) and (4.99), one has
(4.155) $`e(p,\beta _j,N)=e(p,\beta _{j1},N)+{\displaystyle \underset{0<v,v=0\mathrm{mod}(n_j)}{}}\left((p1)v+{\displaystyle \frac{p_jv}{n_j}}\right)dimN_v.`$
Denote by $`\epsilon _i(V)(i=1,2)`$ the contribution of $`dimV`$ in $`\epsilon _i`$ $`(i=1,2)`$ respectively. Then from \[LiuMaZ, (4.52), (4.53)\], on $`F_V^i`$, we have
(4.157) $`\begin{array}{c}r_j^1Pr_j=P+((p1)+\frac{p_j}{n_j})๐_H+\epsilon _i(V).\hfill \end{array}`$
From (4.104), as in (3.58), on $`Q(W)`$, we get
(4.158) $`r_j^1Pr_j=P+((p1)+{\displaystyle \frac{p_j}{n_j}})๐_H+\epsilon (W)+{\displaystyle \frac{1}{2}}\left((p1)+{\displaystyle \frac{p_j}{n_j}}\right)d^{}(W).`$
From (4.155), (4.157), (4.158), and by proceeding as in the proof of Proposition 3.2, as in \[LiuMaZ, Proposition 4.2\], one deduces easily the second equation of (4.143).
Finally from the discussion following (4.80), and \[LiuMaZ, (4.50)\], we get the first two equations of (4.154). By (4.38) and (4.104), we get the last equation of (4.154).
The proof of Proposition 4.2 is complete. $`\mathrm{}`$
###### Lemma 4.4
For each connected component $`M^{}`$ of $`M(n_j)`$, $`\epsilon _1+\epsilon (W)`$, $`\epsilon _2+\epsilon (W)`$ are independent on the connected component of $`F`$ in $`M^{}`$.
Proof : From (2.31), (4.32), (4.38), (4.112) and (4.124), we have
(4.164) $`\begin{array}{c}\epsilon _1={\displaystyle \frac{1}{2}}{\displaystyle \underset{0v^{}<n_j}{}}{\displaystyle \underset{v=v^{}\mathrm{mod}(n_j)}{}}(dimN_vdimV_vdimW_v)\hfill \\ \left[({\displaystyle \frac{p_jv}{n_j}}+(p1)v)^2{\displaystyle \frac{\omega (v^{})(n_j\omega (v^{}))}{n_j^2}}\right]\hfill \\ =(p1+{\displaystyle \frac{p_j}{n_j}})^2e{\displaystyle \frac{1}{16}}\left(dim_๐N(n_j)_{\frac{n_j}{2}}^๐dim_๐V(n_j)_{\frac{n_j}{2}}^๐2dimW(n_j)_{\frac{n_j}{2}}\right)\hfill \\ {\displaystyle \frac{1}{2}}{\displaystyle \underset{0<v^{}<n_j/2}{}}(dimN(n_j)_v^{}dimV(n_j)_v^{}dimW(n_j)_v^{}\hfill \\ dimW(n_j)_{n_jv^{}}){\displaystyle \frac{\omega (v^{})(n_j\omega (v^{}))}{n_j^2}}.\hfill \end{array}`$
By (4.124), $`\epsilon _2\epsilon _1`$ was given in \[LiuMaZ, (4.49)\], it is independent on the connected component of $`F`$ in $`M^{}`$.
The proof of Lemma 4.4 is complete. $`\mathrm{}`$
The following Lemma was proved in \[BT, Lemma 9.3\] and \[T, Lemma 9.6\] (cf. \[LiuMaZ, Lemma 4.6\]).
###### Lemma 4.5
Let $`M`$ be a smooth manifold on which $`S^1`$ acts. Let $`M^{}`$ be a connected component of $`M(n_j)`$, the fixed point set of the subgroup $`๐_{n_j}`$ of $`S^1`$ on $`M`$. Let $`F`$ be the fixed point set of the $`S^1`$-action on $`M`$. Let $`VM`$ be a real, oriented, even dimensional vector bundle to which the $`S^1`$-action on $`M`$ lifts. Assume that $`V`$ is Spin over $`M`$. Let $`p_j]0,n_j[,p_j๐`$ and $`(p_j,n_j)=1`$, then
(4.167) $`\begin{array}{c}_{0<v}(dimV_v)[\frac{p_jv}{n_j}]+\mathrm{\Delta }(n_j,V)\mathrm{mod}(2),\hfill \\ _{0<v}(dimV_v)[\frac{p_jv}{n_j}+\frac{1}{2}]+o(V(n_j)_{n_j/2}^๐)\mathrm{mod}(2)\hfill \end{array}`$
are independent on the connected components of $`F`$ in $`M^{}`$.
Recall that the number $`d^{}(p,\beta _j,N)`$ has been defined in (2.62).
###### Lemma 4.6
For each connected component $`M^{}`$ of $`M(n_j)`$, $`d^{}(p,\beta _j,N)+\mu _i+\mu _4\mathrm{mod}(2)`$ $`(i=1,2,3)`$ are independent on the connected component of $`F`$ in $`M^{}`$.
Proof : By (4.151), and Lemma 4.5, to prove Lemma 4.6, we only need to prove
$$\underset{0<v}{}(dimN_v)([\frac{p_jv}{n_j}]+(p1)v)+\mathrm{\Delta }(n_j,N)+\mu _4\mathrm{mod}(2)$$
is independent on the connected components of $`F`$ in $`M^{}`$. But by \[BT, Lemma 9.3\], as $`\omega _2(TXW)_{S^1}=0`$, we know that, $`\mathrm{mod}(2)`$,
(4.168) $`{\displaystyle \underset{0<v}{}}(dimN_v)[{\displaystyle \frac{p_jv}{n_j}}]+\mathrm{\Delta }(n_j,N)+{\displaystyle \underset{v}{}}(dimW_v)[{\displaystyle \frac{p_jv}{n_j}}]`$
is independent on the connected components of $`F`$ in $`M^{}`$. From (2.62), (2.77), (4.168), we get Lemma 4.6.
The proof of Lemma 4.6 is complete. $`\mathrm{}`$
### 4.3 Proof of Theorem 2.5
From (2.62), (4.31), (4.38) and (4.99), we see that
(4.171) $`\begin{array}{c}{\displaystyle \underset{0<v}{}}dimN_v={\displaystyle \underset{0<v<\frac{n_j}{2}}{}}dimN(n_j)_v+{\displaystyle \frac{1}{2}}dim_๐N(n_j)_{n_j/2}^๐+{\displaystyle \underset{0<v,v=0\mathrm{mod}(n_j)}{}}dimN_v,\hfill \\ d^{}(p,\beta _j,N)=d^{}(p,\beta _{j1},N)+{\displaystyle \underset{0<v,v=0\mathrm{mod}(n_j)}{}}dimN_v.\hfill \end{array}`$
By Lemma 4.6, (4.171), $`d^{}(p,\beta _{j1},N)+_{0<v}dimN_v+\mu _i+\mu _4\mathrm{mod}(2)`$ $`(i=1,2,3)`$ are constant functions on each connected component of $`M(n_j)`$.
From Lemma 4.3, one knows that the Dirac operator $`D^{X(n_j)}F(\beta _j)F_V^i(\beta _j)Q_W(\beta _j)L(\beta _j)_i`$ $`(i=1,2)`$ is well-defined on $`M(n_j)`$. Thus, by using Proposition 4.2, Lemma 4.4, (4.65) and (4.171), for $`i=1,2`$, $`h๐`$, $`m\frac{1}{2}๐`$, $`\tau =\tau _{e1}`$ or $`\tau _{s1}`$, and by applying both the first and the second equations of Theorem 1.1 to each connected component of $`M(n_j)`$ separately, we get the following identity in $`K_{G_y}(B)`$,
(4.172)
(4.179) $`\begin{array}{c}_\alpha (1)^{d^{}(p,\beta _{j1},N)+_{0<v}dimN_v}\mathrm{Ind}_\tau (D^{Y_\alpha }(K_WK_X^1)^{1/2}_{p,j1}(X)\hfill \\ F_V^iQ(W),m+e(p,\beta _{j1},N),h)\hfill \\ =_\beta (1)^{d^{}(p,\beta _{j1},N)+_{0<v}dimN_v+\mu }\mathrm{Ind}_\tau (D^{X(n_j)}(K_WK_X^1)^{1/2}F(\beta _j)\hfill \\ F_V^i(\beta _j)Q_W(\beta _j)L(\beta _j)_i,m+\epsilon _i+\epsilon (W)+(\frac{p_j}{n_j}+(p1))h,h)\hfill \\ =_\alpha (1)^{d^{}(p,\beta _j,N)+_{0<v}dimN_v}\mathrm{Ind}_\tau (D^{Y_\alpha }(K_WK_X^1)^{1/2}_{p,j}(X)\hfill \\ F_V^iQ(W),m+e(p,\beta _j,N),h).\hfill \end{array}`$
Here $`_\beta `$ means the sum over all connected components of $`M(n_j)`$. In (4.172), if $`\tau =\tau _{s1}`$, then $`\mu =\mu _3+\mu _4`$; if $`\tau =\tau _{e1}`$, then $`\mu =\mu _i+\mu _4`$.
The proof of Theorem 2.5 is complete. $`\mathrm{}`$
โโโโโโโโ
Kefeng LIU, Department of Mathematics, Stanford University, Stanford, CA 94305, USA.
E-mail address: kefeng@math.stanford.edu
Xiaonan MA, Humboldt-Universitat zu Berlin, Institut fรผr Mathematik, unter den Linden 6, D-10099 Berlin, Germany.
E-mail address: xiaonan@mathematik.hu-berlin.de
Weiping ZHANG, Nankai Institute of Mathematics, Nankai university, Tianjin 300071, P. R. China.
E-mail address: weiping@nankai.edu.cn |
warning/0001/math0001023.html | ar5iv | text | # NUMERICALLY TRIVIAL FIBRATIONS
## 1 Introduction
Let $`X`$ be a smooth projective variety and let $`L`$ be a line bundle on $`X`$. It is fundamental to study the ring
$$R(X,L):=_{m0}\mathrm{\Gamma }(X,๐ช_X(mL))$$
(in more geometric language to study the Iitaka fibration associated with $`L`$) in algebraic geometry. In most case, to show the nonvanishing ,i.e.,
$`\mathrm{\Gamma }(X,๐ช_X(mL))0`$ for some $`m>0`$ is a central problem.
Because $`R(X,L)`$ C, if $`L`$ is not pseudoeffective (cf. Definition 2.4), the problem is meaningful only when $`L`$ is pseudoeffective.
If $`L`$ is big, then for a sufficiently large $`m`$, the linear system $`mL`$ gives a birational rational embedding of $`X`$ into a projective space. But if $`L`$ is not big, there are very few tools to study $`R(X,L)`$ except Shokurovโs nonvanishing theorem . Moreover even if $`L`$ is big, to study $`R(X,L)`$ we often need to study the restriction of $`R(X,L)`$ on the subvarieties on which the restriction of $`L`$ is not big (e.g. ).
When $`L`$ is not big, a natural approach is to distinguish the null direction of $`L`$. Then we may consider that $`L`$ has positivity in the transverse direction.
If $`L`$ has a $`C^{\mathrm{}}`$-hermitian metric $`h`$ such that the cuvature form $`\mathrm{\Theta }_h`$ is semipositive, the null foliation
$$_{xX}\{vTX_x\mathrm{\Theta }_h(v,\overline{v})=0\}$$
defines a $`C^{\mathrm{}}`$-foliation on the open subset where the rank of the semipositive form $`\mathrm{\Theta }_h`$ is maximal and every leaf is a complex submanifold on the set. In this case the null direction is given by this foliation.
But in general, a pseudoeffective line bundle on a smooth projective variety does not admit a $`C^{\mathrm{}}`$-hermitian metric with semipositive curvature, even if it is nef, although it admits a singular hermitian metric with positive curvature current<sup>1</sup><sup>1</sup>1Here we note that โpositiveโ does not mean strict positivity (cf. Definition 2.2). This terminology may be misleading for algebraic geometers. For this reason I include a subsection which summarize the notion of closed positive currents. . Hence we need to consider a singular hermitian metric on $`L`$ in order to study $`R(X,L)`$.
In this paper we develop an intersection theory for singular hemitian line bundles with positive curvature current and curves on a smooth projective variety. The new intersection number measures the intersection of the positive part of the singular hermitian line bundle and the curve. This intersection theory is not cohomological.
We obtain a natural rational fibration structure in terms of this intersection theory as follows.
###### Theorem 1.1
(Fibration theorem) Let $`(L,h)`$ be a pseudoeffective singular hermitian line bundle (cf. Definition 2.4) on a smooth projective variety $`X`$. Then there exists a unique (up to birational equivalence) rational fibration
$$f:X\mathrm{}Y$$
such that
1. $`f`$ is regular over the generic point of $`Y`$,
2. for every very general fiber $`F`$, $`(L,h)_F`$ is well defined and is numerically trivial (cf. Definition 2.9,2.10),
3. $`dimY`$ is minimal among such fibrations.
We call the above fibration $`f:X\mathrm{}Y`$ the numerically trivial fibration associated with $`(L,h)`$.
###### Remark 1.1
Let $`X`$,$`(L,h)`$ be as above. Then for any smooth divisor $`D`$ on $`X`$, there exists a numerically trivial fibration
$$f_D:D\mathrm{}W.$$
This is simply because the restriction of the intersection theory on $`D`$ exists (cf. Section 2.5) and the proof of the above theorem essentially does not require the existence of the restriction of $`\mathrm{\Theta }_h`$ to $`D`$.
###### Remark 1.2
By the proof of Theorem 1.1 below, we see that the 3-rd condition in Theorem 1.1 is equivalent to :
$`3^{}.`$ for a very general point $`xX`$ and any irreducible horizontal curve (with respect to $`f`$) $`C`$ containing $`x`$, $`(L,h)C>0`$ holds (cf. Definition 2.9 for the definition of $`(L,h)C`$).
Theorem 1.1 singles out the null direction of $`(L,h)`$ as fibers. But this direction is only a part of the null direction as is shown by the following example. This example also shows that in general $`dimY`$ may be strictly larger than the numerical dimension of $`L`$.
###### Example 1.1
Let $`X`$ be an irreducible quotient of the open unit bidisk $`\mathrm{\Delta }^2`$ in $`\text{C}^2`$, i.e.,
$$X=\mathrm{\Delta }^2/\mathrm{\Gamma },$$
where $`\mathrm{\Gamma }`$ is an irreducible cocompact torsion free lattice. Let $`(L,h)`$ denotes the hermitian line bundle whose curvature form comes from the Poincarรฉ metric on the first factor. Then one see that $`L`$ is nef and $`L^2=0`$ holds. In particular $`L`$ is not big. In this case the null foliation of $`\mathrm{\Theta }_h`$ is nothing but the projection of the fibers of the first projection and every leaf of the foliation is Zariski dense (actually even topologically dense in usual topology) in $`X`$. This implies that $`L`$ (and hence also $`(L,h)`$) is numerically positive and the numerically trivial fibration is the identity.
By using an AZD (cf. Definition 2.8, Theorem 2.4 and Proposition 2.1 below), we have the following corollary.
###### Corollary 1.1
Let $`L`$ be a pseudoeffective line bundle on a smooth projective variety $`X`$ and let $`h`$ be a canonical AZD of $`L`$ (cf. Section 2.3). Then there exists a unique rational fibration (up to birational equivalence):
$$f:X\mathrm{}Y$$
such that
1. $`f`$ is regular over the generic point of $`Y`$,
2. for every very general fiber $`F`$, $`(L,h)`$ is numerically trivial on $`F`$.
3. $`dimY`$ is minimal among such fibrations.
Also $`f`$ does not depend on the choice of the canonical AZD $`h`$ (see Proposition 2.1).
We call the above fibration $`f:X\mathrm{}Y`$ the numerically trivial fibration associated with $`L`$.
The poof of Theorem 1.1 is done by finding a dominating family of maximal dimensional subvarieties on which the restriction of $`(L,h)`$ is numerically trivial. The heart of the proof is to prove that this family actually gives a rational fibration by showing that the generic point of a general member of the family does not intersect other members.
The structure of numerically trivial singular hermitian line bundles with positive curvature current is given as follows.
###### Theorem 1.2
Let $`(L,h)`$ be a singular hermitian line bundle on a smooth projective variety $`X`$. Suppose that $`\mathrm{\Theta }_h`$ is closed positive and $`(L,h)`$ is numerically trivial on $`X`$. Then there exist at most countably many prime divisors $`\{D_i\}`$ and nonnegative numbers $`\{a_i\}`$ such that
$$\mathrm{\Theta }_h=2\pi a_iD_i$$
holds, where we have identified each $`D_i`$ with a closed positive current. More generally let $`Y`$ be a subvariety of $`X`$ such that the restriction $`h_Y`$ is well defined. Suppose that $`(L,h)`$ is numerically trivial on $`Y`$. Then the restriction $`\mathrm{\Theta }_h_Y`$ is a sum of at most countably many prime divisors with nonnegative coefficients on $`Y`$.
###### Remark 1.3
For a divisor $`D`$, the current associated with $`D`$ is often denoted by $`[D]`$. But this notation is confusing with the round down of $`D`$ in algebraic geometry. Hence we do not use this notation in this paper.
This paper is a byproduct of the proof of the nonvanishing theorem (\[18, Theorem 5.1\]).
In this paper, I cannot refer to applications of the above theorems because of the length. These will be published separately.
In this paper โvery generalโ means outside of at most countably many union of proper Zariski closed subsets and โgeneralโ means in the sense of usual Zariski topology.
I intended the paper to be readable for algebraic geometers who are not familiar with complex analytic background.
I would like to express hearty thanks to the referee for his careful reading and a lot of useful comments.
## 2 Intersection theory for singular hermitian line bundles
In this section we define an intersection number for a singular hermitian line bundle with positive curvature current on a smooth projective variety and an irreducible curve on it. This intersection number is different from the usual intersection number of the underlying line bundle and the curve.
### 2.1 Closed positive currents
In this subsection we shall review the definition and basic notions of closed positive $`(p,p)`$-currents on a complex manifold. For the general facts about the theory of currents, see for example \[6, Chapter 3\]. Let $`M`$ be a complex manifold of dimension $`n`$ and let $`A_c^{p,q}(M)`$ denote the space of $`C^{\mathrm{}}`$ $`(p,q)`$-forms with compact support. We define a topology on $`A_c^{p,q}(M)`$ such that a sequence $`\{\phi _i\}_{i=1}^{\mathrm{}}`$ in $`A_c^{p,q}(M)`$ converges, if and only if there exists a compact subset $`K`$ of $`M`$ such that $`\text{Supp}\phi _iK`$ holds for every $`i`$ and $`\{\phi _i\}_{i=1}^{\mathrm{}}`$ converges in $`C^k`$-topology on $`K`$ for every $`k`$ to a $`C^{\mathrm{}}`$ $`(p,q)`$-form $`\phi _{\mathrm{}}`$.
###### Definition 2.1
Let $`M`$ be a complex manifold of dimension $`n`$. The space of $`(p,q)`$-currents $`๐^{p,q}(M)`$ on $`M`$ is the dual space of $`A_c^{np,nq}(M)`$. We define
$$:๐^{p,q}(M)๐^{p+1,q}(M)$$
and
$$\overline{}:๐^{p,q}(M)๐^{p,q+1}(M)$$
by
$$T(\phi ):=(1)^{p+q+1}T(\phi )(T๐^{p,q}(M),\phi A^{np,nq}(M))$$
and
$$\overline{}T(\phi ):=(1)^{p+q+1}T(\overline{}\phi )(T๐^{p,q}(M),\phi A^{np,nq}(M))$$
We define the exterior derivative $`d`$ by
$$d:=+\overline{}.$$
###### Definition 2.2
$`T๐^{p,q}(M)`$ is said to be closed, if $`dT=0`$ holds. A $`(p,p)`$-current $`T`$ is real in case $`T=\overline{T}`$ in the sense that $`\overline{T(\phi )}=T(\overline{\phi })`$ holds for all $`\phi A_c^{np,np}(M)`$. A real $`(p,p)`$-current $`T`$ on $`M`$ is said to be positive, if
$$(\sqrt{1})^{\frac{p(p1)}{2}}T(\eta \overline{\eta })0$$
holds for every $`\eta A_c^{np}(M)`$.
The above definition of positivity of currents is somewhat misleading for algebraic geometers. It might be appropriate to say pseudoeffective currents instead of positive currents.
###### Example 2.1
Let $`V`$ be a subvariety of codimension $`p`$ in $`M`$. Then $`V`$ is a closed positive $`(p,p)`$-current on $`M`$ by
$$V(\phi ):=_{V_{reg}}\phi (\text{for}\phi A_c^{np,np}(M)).$$
###### Example 2.2
Let $`\varphi `$ be a $`C^{\mathrm{}}`$-closed $`(p,q)`$-form on $`M`$. Then $`\varphi `$ is a closed $`(p,q)`$-current on $`M`$ by
$$\varphi (\phi ):=_M\varphi \phi (\text{for}\phi A_c^{np,nq}(M))$$
### 2.2 Multiplier ideal sheaves
In this subsection $`L`$ will denote a holomorphic line bundle on a complex manifold $`M`$.
###### Definition 2.3
A singular hermitian metric $`h`$ on $`L`$ is given by
$$h=e^\phi h_0,$$
where $`h_0`$ is a $`C^{\mathrm{}}`$-hermitian metric on $`L`$ and $`\phi L_{loc}^1(M)`$ is an arbitrary function on $`M`$. We call $`\phi `$ a weight function of $`h`$.
The curvature current $`\mathrm{\Theta }_h`$ of the singular hermitian line bundle $`(L,h)`$ is defined by
$$\mathrm{\Theta }_h:=\mathrm{\Theta }_{h_0}+\sqrt{1}\overline{}\phi ,$$
where $`\overline{}`$ is taken in the sense of a current. The $`L^2`$-sheaf $`^2(L,h)`$ of the singular hermitian line bundle $`(L,h)`$ is defined by
$$^2(L,h):=\{\sigma \mathrm{\Gamma }(U,๐ช_M(L))h(\sigma ,\sigma )L_{loc}^1(U)\},$$
where $`U`$ runs over the open subsets of $`M`$. In this case there exists an ideal sheaf $`(h)`$ such that
$$^2(L,h)=๐ช_M(L)(h)$$
holds. We call $`(h)`$ the multiplier ideal sheaf of $`(L,h)`$. If we write $`h`$ as
$$h=e^\phi h_0,$$
where $`h_0`$ is a $`C^{\mathrm{}}`$ hermitian metric on $`L`$ and $`\phi L_{loc}^1(M)`$ is the weight function, we see that
$$(h)=^2(๐ช_M,e^\phi )$$
holds. Also we define
$$_{\mathrm{}}(h)=^{\mathrm{}}(๐ช_M,e^\phi )$$
and call it the $`L^{\mathrm{}}`$-multiplier ideal sheaf of $`(L,h)`$.
Let $`D`$ be an effective R-divisor on $`M`$ and let
$$\underset{i}{}a_iD_i$$
be the irreducible decomposition of $`D`$. Let $`\sigma _i`$ be a global section of $`๐ช_M(D_i)`$ with divisor $`D_i`$. Let $`h_i`$ be a $`C^{\mathrm{}}`$-hermitian metric on $`๐ช_M(D_i)`$. Then
$$h=\frac{_ih_i^{a_i}}{_ih_i(\sigma _i,\sigma _i)^{a_i}}$$
is a singular hermitian metric on the R-line bundle $`๐ช_M(D)`$. It is clear that $`h`$ is independent of the choice of $`h_i`$โs. We define the multiplier sheaf $`(D)`$ associated with $`D`$ by
$$(D):=(h)=^2(๐ช_X,\frac{1}{_ih_i(\sigma _i,\sigma _i)^{a_i}}).$$
If $`\text{Supp}D`$ is a divisor with normal crossings,
$$(D)=๐ช_M([D])$$
holds, where $`[D]:=_i[a_i]D_i`$ (for a real number $`a`$, $`[a]`$ denotes the largest integer smaller than or equal to $`a`$).
The following terminology is fundamental in this paper.
###### Definition 2.4
$`L`$ is said to be pseudoeffective, if there exists a singular hermitian metric $`h`$ on $`L`$ such that the curvature current $`\mathrm{\Theta }_h`$ is a closed positive current.
Also a singular hermitian line bundle $`(L,h)`$ is said to be pseudoeffective, if the curvature current $`\mathrm{\Theta }_h`$ is a closed positive current.
It is easy to see that a line bundle $`L`$ on a smooth projective manifold $`M`$ is pseudoeffective, if and only if for an ample line bundle $`H`$ on $`M`$, $`L+ฯตH`$ is Q-effective (or big) for every positive rational number $`ฯต`$ (cf. ).
If $`\{\sigma _i\}`$ are a finite number of global holomorphic sections of $`L`$, for every positive rational number $`\alpha `$ and a $`C^{\mathrm{}}`$-function $`\varphi `$,
$$h:=e^\varphi \frac{h_0^\alpha }{(_ih_0(\sigma _i,\sigma _i))^\alpha }$$
defines a singular hermitian metric on $`\alpha L`$, where $`h_0`$ is a $`C^{\mathrm{}}`$-hermitian metric on $`L`$ (note that the righthandside is independent of $`h_0`$). We call such a metric $`h`$ a singular hermitian metric on $`\alpha L`$ with algebraic singularities. Singular hermitian metrics with algebraic singularities are particulary easy to handle, because its multiplier ideal sheaf or that of the multiple of the metric can be controlled by taking suitable successive blowing ups such that the total transform of the divisor $`_i(\sigma _i)`$ is a divisor with normal crossings.
By definition a multiplier ideal sheaf has the following property which will be used later.
###### Lemma 2.1
Let $`(L,h)`$ be a singular hermitian line bundle on a complex manifold $`M`$ such that $`\mathrm{\Theta }_h`$ is bounded from below by a $`C^{\mathrm{}}`$-$`(1,1)`$-form. Let $`f:NM`$ be a modification. Then $`(f^{}L,f^{}h)`$ is a singular hermitian line bundle on $`N`$ and
$$f_{}(f^{}h)(h)$$
holds.
Proof. First we note that $`f_{}(f^{}L)=L`$ holds. Let $`xM`$ be an arbitrary point of $`M`$. Let $`U`$ be a neighbourhood of $`x`$ and let $`\sigma `$ be a holomorphic section of $`L`$ on $`U`$ such that
$$_{f^1(U)}f^{}h(\sigma ,\sigma )๐V_N<\mathrm{}$$
holds, where $`dV_N`$ denote a $`C^{\mathrm{}}`$ volume form on $`N`$. Let $`dV_M`$ be a $`C^{\mathrm{}}`$-volume form on $`M`$. Then if we shrink $`U`$ a little bit, we may assume that there exists a positive constant $`C`$ such that
$$f^{}dV_MCdV_N$$
holds on $`f^1(U)`$. Hence we see that
$$_Uh(\sigma ,\sigma )๐V_M<\mathrm{}$$
holds. Q.E.D.
The following theorem is fundamental in the applications of multiplier ideal sheaves.
###### Theorem 2.1
(Nadelโs vanishing theorem \[11, p.561\]) Let $`(L,h)`$ be a singular hermitian line bundle on a compact Kรคhler manifold $`M`$ and let $`\omega `$ be a Kรคhler form on $`M`$. Suppose that $`\mathrm{\Theta }_h`$ is strictly positive, i.e., there exists a positive constant $`\epsilon `$ such that
$$\mathrm{\Theta }_h\epsilon \omega $$
holds. Then $`(h)`$ is a coherent sheaf of $`๐ช_M`$ ideal and for every $`q1`$
$$H^q(M,๐ช_M(K_M+L)(h))=0$$
holds.
We note that the multiplier ideal sheaf of a singular hermitian R-line bundle is well defined because the multiplier ideal sheaf is defined in terms of the weight function. Sometimes it is useful to consider the following variant of multiplier ideal sheaves.
###### Definition 2.5
Let $`h_L`$ be a singular hermitian metric on a line bundle $`L`$. Suppose that the curvature of $`h_L`$ is a positive current on $`X`$. We set
$$\overline{}(h_L):=\underset{\epsilon 0}{lim}(h_L^{1+\epsilon })$$
and call it the closure of $`(h_L)`$.
As you see later, the closure of a multiplier ideal sheaf is easier to handle than the original multiplier ideal sheaf in some respect.
Next we shall consider the restriction of singular hermitian line bundles to subvarieties.
###### Definition 2.6
Let $`h`$ be a singular hermitian metric on $`L`$ given by
$$h=e^\phi h_0,$$
where $`h_0`$ is a $`C^{\mathrm{}}`$-hermitian metric on $`L`$ and $`\phi L_{loc}^1(M)`$ is an uppersemicontinuous function. Here $`L_{loc}^1(M)`$ denotes the set of locally integrable functions (not the set of classes of almost everywhere equal locally integrable functions on $`M`$).
For a subvariety $`V`$ of $`M`$, we say that the restriction $`h_V`$ is well defined, if $`\phi `$ is not identically $`\mathrm{}`$ on $`V`$.
Let $`(L,h)`$,$`h_0`$,$`V`$, $`\phi `$ be as in Definition 2.6. Suppose that the curvature current $`\mathrm{\Theta }_h`$ is bounded from below by some $`C^{\mathrm{}}`$-(1,1)-form. Then $`\phi `$ is an almost plurisubharmonic function, i.e. locally a sum of a plurisubharmonic function and a $`C^{\mathrm{}}`$-function. Let $`\pi :\stackrel{~}{V}V`$ be an arbitrary resolution of $`V`$. Then $`\pi ^{}(\phi _V)`$ is locally integrable on $`\stackrel{~}{V}`$, since $`\phi `$ is almost plurisubharmonic. Hence
$$\pi ^{}(\mathrm{\Theta }_h_V):=\mathrm{\Theta }_{\pi ^{}h_0_V}+\sqrt{1}\overline{}\pi ^{}(\phi _V)$$
is well defined.
###### Definition 2.7
Let $`\phi `$ be a plurisubharmonic function on a unit open polydisk $`\mathrm{\Delta }^n`$ with center $`O`$. We define the Lelong number of $`\phi `$ at $`O`$ by
$$\nu (\phi ,O):=\underset{xO}{lim\; inf}\frac{\phi (x)}{\mathrm{log}x},$$
where $`x=(x_i^2)^{1/2}`$. Let $`T`$ be a closed positive $`(1,1)`$-current on a unit open polydisk $`\mathrm{\Delta }^n`$. Then by $`\overline{}`$-Poincarรฉ lemma there exists a plurisubharmonic function $`\varphi `$ on $`\mathrm{\Delta }^n`$ such that
$$T=\frac{\sqrt{1}}{\pi }\overline{}\varphi .$$
We define the Lelong number $`\nu (T,O)`$ at $`O`$ by
$$\nu (T,O):=\nu (\varphi ,O).$$
It is easy to see that $`\nu (T,O)`$ is independent of the choice of $`\varphi `$ and local coordinates around $`O`$. For an analytic subset $`V`$ of a complex manifold $`X`$, we set
$$\nu (T,V)=\underset{xV}{inf}\nu (T,x).$$
###### Remark 2.1
More generally the Lelong number is defined for a closed positive $`(k,k)`$-current on a complex manifold.
###### Theorem 2.2
(\[13, p.53, Main Theorem\]) Let $`T`$ be a closed positive $`(k,k)`$-current on a complex manifold $`M`$. Then for every $`c>0`$
$$\{xM\nu (T,x)c\}$$
is a subvariety of codimension $`k`$ in $`M`$.
The following lemma shows a rough relationship between the Lelong number of $`\nu (\mathrm{\Theta }_h,x)`$ at $`xX`$ and the stalk of the multiplier ideal sheaf $`(h)_x`$ at $`x`$.
###### Lemma 2.2
(\[1, p.284, Lemma 7\],\[13, p.85, Lemma 5.3\]) Let $`\phi `$ be a plurisubharmonic function on the open unit polydisk $`\mathrm{\Delta }^n`$ with center $`O`$. Suppose that $`e^\phi `$ is not locally integrable around $`O`$. Then we have that
$$\nu (\phi ,O)2$$
holds. And if
$$\nu (\phi ,O)>2n$$
holds, then $`e^\phi `$ is not locally integrable around $`O`$.
Let $`(L,h)`$ be a pseudoeffective singular hermitian line bundle on a complex manifold $`M`$. The closure $`\overline{}(h)`$ of the multiplier ideal sheaf $`(h)`$ can be analysed in terms of Lelong numbers in the following way. We note that $`\overline{}(h)`$ is coherent ideal sheaf on $`M`$ by Theorem 2.1.
In the case of $`dimM=1`$, we can compute $`\overline{}(h)`$ in terms of the Lelong number $`\nu (\mathrm{\Theta }_h,x)(xM)`$. In fact in this case $`\overline{}(h)`$ is locally free and
$$\overline{}(h)=๐ช_M(\underset{xM}{}[\nu (\mathrm{\Theta }_h,x)]x)$$
holds by Lemma 2.2, because $`2=2dimM`$.
In the case of $`dimM2`$, let $`f:NM`$ be a modification such that $`f^{}\overline{}(h)`$ is locally free. If we take $`f`$ properly, we may assume that there exists a divisor $`F=_iF_i`$ with normal crossings on $`Y`$ such that
$$K_N=f^{}K_M+\underset{i}{}a_iF_i$$
and
$$\overline{}(h)=f_{}๐ช_N(\underset{i}{}b_iF_i)$$
hold on $`Y`$ for some nonnegative integers $`\{a_i\}`$ and $`\{b_i\}`$. Let $`yF_i_{ji}F_j`$ and let $`(U,z_1,\mathrm{},z_n)`$ be a local corrdinate neighbourhood of $`y`$ which is biholomorphic to the open unit disk $`\mathrm{\Delta }^n`$ with center $`O`$ in $`\text{C}^n(n=dimM)`$ and
$$UF_i=\{pUz_1(p)=0\}$$
holds. For $`q\mathrm{\Delta }^{n1}`$, we set $`\mathrm{\Delta }(q):=\{pU(z_2(p),\mathrm{},z_n(p))=q\}`$. Then considering the family of the restriction $`\{\mathrm{\Theta }_h_{\mathrm{\Delta }(q)}\}`$ for very general $`q\mathrm{\Delta }^{n1}`$, by Lemma 2.2, we see that
$$b_i=[\nu (f^{}\mathrm{\Theta }_h,F_i)a_i]$$
holds for every $`i`$. In this way $`\overline{}(h)`$ is determined by the Lelong numbers of the curvature current on some modification. This is not the case, unless we take the closure as in the following example.
###### Example 2.3
Let $`h_P`$ be a singular hermitian metric on the trivial line bundle on the open unit polydisk $`\mathrm{\Delta }`$ with center $`O`$ in C defined by
$$h_P=\frac{^2}{z^2(\mathrm{log}z)^2}.$$
Then $`\nu (\mathrm{\Theta }_{h_P},0)=1`$ holds. But $`(h_P)=๐ช_\mathrm{\Delta }`$ holds. On the other hand $`\overline{}(h_P)=_0`$ holds, where $`_0`$ is the ideal sheaf of $`0\mathrm{\Delta }`$.
### 2.3 Analytic Zariski decompositions
In this subsection we shall introduce the notion of analytic Zariski decompositions. By using analytic Zariski decompositions, we can handle big line bundles like nef and big line bundles.
###### Definition 2.8
Let $`M`$ be a compact complex manifold and let $`L`$ be a holomorphic line bundle on $`M`$. A singular hermitian metric $`h`$ on $`L`$ is said to be an analytic Zariski decomposition, if the followings hold.
1. $`\mathrm{\Theta }_h`$ is a closed positive current,
2. for every $`m0`$, the natural inclusion
$$H^0(M,๐ช_M(mL)(h^m))H^0(M,๐ช_M(mL))$$
is an isomorphim.
###### Remark 2.2
If an AZD exists on a line bundle $`L`$ on a smooth projective variety $`M`$, $`L`$ is pseudoeffective by the condition 1 above.
###### Theorem 2.3
() Let $`L`$ be a big line bundle on a smooth projective variety $`M`$. Then $`L`$ has an AZD.
As for the existence for general pseudoeffective line bundles, now we have the following theorem.
###### Theorem 2.4
() Let $`X`$ be a smooth projective variety and let $`L`$ be a pseudoeffective line bundle on $`X`$. Then $`L`$ has an AZD.
Proof of Theorem 2.4. Let $`h_0`$ be a fixed $`C^{\mathrm{}}`$-hermitian metric on $`L`$. Let $`E`$ be the set of singular hermitian metric on $`L`$ defined by
$$E=\{h;h:\text{lowersemicontinuous singular hermitian metric on }L,$$
$$\mathrm{\Theta }_h\text{is positive},\frac{h}{h_0}1\}.$$
Since $`L`$ is pseudoeffective, $`E`$ is nonempty. We set
$$h_L=h_0\underset{hE}{inf}\frac{h}{h_0},$$
where the infimum is taken pointwise. The supremum of a family of plurisubharmonic functions uniformly bounded from above is known to be again plurisubharmonic, if we modify the supremum on a set of measure $`0`$(i.e., if we take the uppersemicontinuous envelope) by the following theorem of P. Lelong.
###### Theorem 2.5
(\[10, p.26, Theorem 5\]) Let $`\{\phi _t\}_{tT}`$ be a family of plurisubharmonic functions on a domain $`\mathrm{\Omega }`$ which is uniformly bounded from above on every compact subset of $`\mathrm{\Omega }`$. Then $`\psi =sup_{tT}\phi _t`$ has a minimum uppersemicontinuous majorant $`\psi ^{}`$ which is plurisubharmonic.
###### Remark 2.3
In the above theorem the equality $`\psi =\psi ^{}`$ holds outside of a set of measure $`0`$(cf.\[10, p.29\]).
By Theorem 2.5 we see that $`h_L`$ is also a singular hermitian metric on $`L`$ with $`\mathrm{\Theta }_h0`$. Suppose that there exists a nontrivial section $`\sigma \mathrm{\Gamma }(X,๐ช_X(mL))`$ for some $`m`$ (otherwise the second condition in Definition 3.1 is empty). We note that
$$\frac{1}{\sigma ^{\frac{2}{m}}}$$
gives the weihgt of a singular hermitian metric on $`L`$ with curvature $`2\pi m^1(\sigma )`$, where $`(\sigma )`$ is the current of integration along the zero set of $`\sigma `$. By the construction we see that there exists a positive constant $`c`$ such that
$$\frac{h_0}{\sigma ^{\frac{2}{m}}}ch_L$$
holds. Hence
$$\sigma H^0(X,๐ช_X(mL)(h_L^m))$$
holds. This means that $`h_L`$ is an AZD of $`L`$. Q.E.D.
The following proposition implies that the multiplier ideal sheaves of $`h_L^m(m1)`$ constructed in the proof of Theorem 2.4 are independent of the choice of the $`C^{\mathrm{}}`$-hermitian metric $`h_0`$. The proof is trivial. Hence we omit it.
###### Proposition 2.1
$`h_0,h_0^{}`$ be two $`C^{\mathrm{}}`$-hermitian metrics on a pseudoeffective line bundle $`L`$ on a smooth projective variety $`X`$. Let $`h_L,h_L^{}`$ be the AZDโs constructed as in the proof of Theorem 2.4 associated with $`h_0,h_0^{}`$ respectively. Then
$$(\underset{xX}{\mathrm{min}}\frac{h_0}{h_0^{}}(x))h_L^{}h_L(\underset{xX}{\mathrm{max}}\frac{h_0}{h_0^{}}(x))h_L^{}$$
hold. In particular
$$(h_L^m)=((h_L^{})^m)$$
holds for every $`m1`$.
We call the AZD constructed as in the proof of Theorem 2.4 a canonical AZD of $`L`$. Proposition 2.1 implies that the multiplier ideal sheaves associated with the multiples of the canonical AZD are independent of the choice of the canonical AZD.
### 2.4 Intersection numbers
In this subsection we shall define the intersection number for a singular hermitian line bundle with positive curvature current and an irreducible curve such that the restriction of the singular hermitian metric is well defined.
###### Definition 2.9
Let $`(L,h)`$ be a pseudoeffective singular hermitian line bundle on a smooth projective variety $`X`$. Let $`C`$ be an irreducible curve on $`X`$ such that the natural morphism $`(h^m)๐ช_C๐ช_C`$ is an isomorphism at the generic point of $`C`$ for every $`m0`$.
The intersection number $`(L,h)C`$ is defined by
$$(L,h)C:=\overline{lim}_m\mathrm{}m^1dimH^0(C,๐ช_C(mL)(h^m)/tor),$$
where $`tor`$ denotes the torsion part of $`๐ช_C(mL)(h^m)`$.
If the natural morphism $`(h^m)๐ช_C๐ช_C`$ is $`0`$ at the generic point of $`C`$ for some $`m1`$, to define $`(L,h)C`$, $`H^0(C,๐ช_C(mL)(h^m)/tor)`$ cannot be considered as a subspace of $`H^0(C,๐ช_C(mL))`$. A special important case will be treated in Section 2.5.
###### Remark 2.4
Let $`(L,h)`$, $`C`$ be as above. Let $`\pi :\stackrel{~}{C}C`$ be the normalization of $`C`$. Then we see that
$$(L,h)C=\overline{lim}_m\mathrm{}m^1dimH^0(\stackrel{~}{C},๐ช_{\stackrel{~}{C}}(m\pi ^{}L)\pi ^{}(h^m)/tor)$$
holds. This is verified as follows. First it is clear that
$$(L,h)C\overline{lim}_m\mathrm{}m^1dimH^0(\stackrel{~}{C},๐ช_{\stackrel{~}{C}}(m\pi ^{}L)\pi ^{}(h^m)/tor)$$
holds. On the other hand, there exists a nonzero ideal sheaf $`๐ฅ`$ independent of $`m0`$ on $`\stackrel{~}{C}`$ such that
$$H^0(\stackrel{~}{C},(๐ช_{\stackrel{~}{C}}(m\pi ^{}L)\pi ^{}(h^m)/tor)๐ฅ)\pi ^{}H^0(C,๐ช_C(mL)(h^m)/tor)$$
holds. For example, we can take $`๐ฅ`$ to be $`(\pi ^{}_{\text{Sing}(C)})^r`$ where $`_{\text{Sing}(C)}`$ denotes the ideal sheaf of the singular locus of $`C`$ and $`r`$ is a sufficiently large positive integer. Because $`V()`$ consists of a finite number of points, this implies that
$$(L,h)C\overline{lim}_m\mathrm{}m^1dimH^0(\stackrel{~}{C},๐ช_{\stackrel{~}{C}}(m\pi ^{}L)\pi ^{}(h^m)/tor)$$
holds. The above two inequalities imply the assertion.
###### Remark 2.5
Let $`(L,h)`$, $`C`$ be as in Definition 2.9. We see that
$$(L,h)C=\overline{lim}_m\mathrm{}m^1dimH^0(C,๐ช_C(mL)\overline{}(h^m)/tor)$$
always holds.
This can be verified as follows. First we shall assume that $`C`$ is smooth. By the assumption $`(h^m)/tor`$ is an ideal sheaf on $`C`$. If
$$\mathrm{deg}_C๐ช_C(mL)(h^m)/tor>2g(C)2$$
holds, where $`g(C)`$ denotes the genus of $`C`$, then
$$H^1(C,๐ช_C(mL)(h^m)/tor)=0$$
holds. On the other hand if
$$\mathrm{deg}_C๐ช_C(mL)(h^m)/tor2g(C)2$$
holds, then there exists a constant $`K`$ independent of such $`m`$ such that
$$dimH^0(C,๐ช_C(mL)(h^m)/tor)K$$
holds.
Hence we see that
$$(\mathrm{})(L,h)C=\overline{lim}_m\mathrm{}m^1\mathrm{deg}_C๐ช_C(mL)(h^m)/tor$$
holds by the Riemann-Roch theorem. By the same reason, we see that
$$\overline{lim}_m\mathrm{}m^1dimH^0(C,๐ช_C(mL)\overline{}(h^m)/tor)=\overline{lim}_m\mathrm{}m^1\mathrm{deg}_C๐ช_C(mL)\overline{}(h^m)/tor$$
holds. On the other hand, for every $`ฯต>0`$
$$\overline{lim}_m\mathrm{}m^1\mathrm{deg}_C๐ช_C(mL)(h^m)/tor$$
$$\overline{lim}_m\mathrm{}m^1\mathrm{deg}_C๐ช_C((1+2ฯต)mL)\overline{}(h^{(1+ฯต)m})/tor$$
holds, since $`(h^{(1+2ฯต)m})\overline{}(h^{(1+ฯต)m})`$ holds for every $`m0`$. And also
$$\underset{ฯต0}{lim}(\overline{lim}_m\mathrm{}m^1\mathrm{deg}_C๐ช_C((1+2ฯต)mL)\overline{}(h^{(1+ฯต)m})/tor)$$
$$=\underset{ฯต0}{lim}((\overline{lim}_m\mathrm{}m^1\mathrm{deg}_C๐ช_C((1+ฯต)mL)\overline{}(h^{(1+ฯต)m})/tor)+ฯตLC)$$
$$=\overline{lim}_m\mathrm{}m^1\mathrm{deg}_C๐ช_C(mL)\overline{}(h^m)/tor$$
hold. We note that $`\overline{}(h^m)(h^m)`$ holds for every $`m0`$ by their definitions. Hence we have that
$$\overline{lim}_m\mathrm{}m^1\mathrm{deg}_C๐ช_C(mL)(h^m)/tor=\overline{lim}_m\mathrm{}m^1\mathrm{deg}_C๐ช_C(mL)\overline{}(h^m)/tor$$
holds. By the above argument we see that
$$(L,h)C=\overline{lim}_m\mathrm{}m^1dimH^0(C,๐ช_C(mL)\overline{}(h^m)/tor)$$
holds.
If $`C`$ is singular, by the argument as in Remark 2.4, we can easily deduce the same conclusion by considering the normalization $`\pi :\stackrel{~}{C}C`$.
Since the closure of multiplier a multiplier ideal sheaf is easier to handle as you see in this paper, it might be better to use the above formula as the definition of the intersection number.
Let $`(L,h)`$ and $`C`$ be as above. Assume that $`h_C`$ is well defined. Let
$$\pi :\stackrel{~}{C}C$$
be the normalization of $`C`$. We define the multiplier ideal sheaf
$`(h^m_C)(m0)`$ on $`C`$ by
$$(h^m_C):=\pi _{}(\pi ^{}h^m_C).$$
We note that $`(h^m_C)`$ is not necessary a subsheaf of $`๐ช_C`$, if $`C`$ is nonnormal. And the Lelong number $`\nu (\mathrm{\Theta }_h_C,x)(xC)`$ by
$$\nu (\mathrm{\Theta }_h,x)=\underset{\stackrel{~}{x}\pi ^1(x)}{}\nu (\pi ^{}\mathrm{\Theta }_h_C,\stackrel{~}{x}).$$
###### Proposition 2.2
Let $`(L,h)`$ be a pseudoeffective singular hermitian line bundle on a smooth projective variety $`X`$. Let $`C`$ be an irreducible curve on $`X`$ such that $`h_C`$ is well defined. Suppose that $`(L,h)C=0`$ holds. Then
$$\mathrm{\Theta }_h_C=2\pi \underset{xC}{}\nu (\mathrm{\Theta }_h_C,x)x$$
holds in the sense that
$$\pi ^{}(\mathrm{\Theta }_h_C)=2\pi \underset{\stackrel{~}{x}\stackrel{~}{C}}{}\nu (\pi ^{}\mathrm{\Theta }_h_C,\stackrel{~}{x})\stackrel{~}{x}$$
holds.
Proof of Proposition 2.2. First we quote the following $`L^2`$-extension theorem.
###### Theorem 2.6
(\[12, p.197, Theorem\]) Let $`\mathrm{\Omega }`$ be a bounded pseudoconvex domain in $`๐^n`$, $`\psi :\mathrm{\Omega }๐\{\mathrm{}\}`$ a plurisubharmonic function and $`H๐^n`$ a complex hyperplane.
Then there exists a constant $`C`$ depending only on the diameter of $`\mathrm{\Omega }`$ such that for any holomorphic function $`f`$ on $`\mathrm{\Omega }H`$ satisfying
$$_{\mathrm{\Omega }H}e^\psi f^2๐V_{n1}<\mathrm{},$$
where $`dV_{n1}`$ denotes the $`(2n2)`$-dimensional Lebesgue measure, there exists a holomorphic function $`F`$ on $`\mathrm{\Omega }`$ satisfying $`F_{\mathrm{\Omega }H}=f`$ and
$$_\mathrm{\Omega }e^\psi F^2๐V_nC_{\mathrm{\Omega }H}e^\psi f^2๐V_{n1}.$$
###### Lemma 2.3
Let $`S`$ be the singular points of $`C`$ with reduced structure and let $`_S`$ denote the ideal of $`S`$. Then there exists a positive integer $`a`$ such that
$$(h^m_C)_S^a(h^m)๐ช_C$$
hold for every $`m`$.
Proof of Lemma 2.3. In fact let
$$f:\stackrel{~}{X}X$$
be an embedded resolution of $`C`$ and let $`\stackrel{~}{C}`$ denote the strict transform of $`C`$ in $`\stackrel{~}{X}`$. Since $`\stackrel{~}{C}`$ is locally a smooth complete intersection of smooth divisors, for $`\stackrel{~}{x}\stackrel{~}{C}`$, by the successive use of Theorem 2.6 every element of $`(f^{}h^m_{\stackrel{~}{C}})_{\stackrel{~}{x}}`$ can be extended to an element of $`(f^{}h^m)_{\stackrel{~}{x}}`$. This means that
$$(f^{}h^m_{\stackrel{~}{C}})(f^{}h^m)_{\stackrel{~}{C}}$$
holds. By the definition of $`(h^m_C)`$ we see that
$$(h^m_C)=f_{}((f^{}h^m_{\stackrel{~}{C}}))$$
holds. Hence we have that
$$\text{(}+\text{)}(h^m_C)f_{}((f^{}h^m)_{\stackrel{~}{C}})$$
holds.
Let $`f^{}(C)`$ be the total transform of $`C`$.
First we note that if a germ of $`(f^{}h^m)_{\stackrel{~}{C}}`$ is identically $`0`$ along the scheme theoretic intersection $`(f^{}(C)\stackrel{~}{C})\stackrel{~}{C}`$, it extends to a germ of $`(f^{}h^m)_{f^{}(C)}`$ by setting identically $`0`$ on the branches of $`f^{}(C)`$ except $`\stackrel{~}{C}`$.
Next we note that
$$f_{}((f^{}h^m)๐ช_{f^{}(C)})(h^m)๐ช_C$$
holds by Lemma 2.1.
By these facts and ($`+`$), we see that there exists a positive integer $`a`$ independent of $`m`$ such that
$$(h^m_C)_S^a(h^m)๐ช_C$$
holds. This completes the proof of Lemma 2.3. Q.E.D.
By \[13, p.111, Lemma 9.5\] we see that
$$\mathrm{\Theta }_h_C2\pi \underset{xC}{}\nu (\mathrm{\Theta }_h_C,x)x$$
is a positive current on $`C`$. Hence
$$LC\underset{xC}{}\nu (\mathrm{\Theta }_h_C,x)$$
is a nonnegative number. Let $`\pi :\stackrel{~}{C}C`$ be the normalization of $`C`$. Then by Lemma 2.2 and the definition of $`\nu (\mathrm{\Theta }_h_C)`$, we see that
$$\mathrm{deg}_{\stackrel{~}{C}}(๐ช_{\stackrel{~}{C}}(m\pi ^{}L)(\pi ^{}(h^m_C))(LC\underset{xC}{}\nu (\mathrm{\Theta }_h_C,x))m$$
holds. Hence we see that
$$\underset{m\mathrm{}}{lim}m^1\mathrm{deg}_{\stackrel{~}{C}}(๐ช_{\stackrel{~}{C}}(m\pi ^{}L)(\pi ^{}(h^m_C))LC\underset{xC}{}\nu (\mathrm{\Theta }_h_C,x)0$$
hold. By the Riemann-Roch theorem for curves and the Kodaira vanishing theorem, we see that if
$$LC\underset{xC}{}\nu (\mathrm{\Theta }_h_C,x)>0$$
holds, then
$$\underset{m\mathrm{}}{lim}m^1dimH^0(\stackrel{~}{C},๐ช_{\stackrel{~}{C}}(m\pi ^{}L)(\pi ^{}(h^m_C))LC\underset{xC}{}\nu (\mathrm{\Theta }_h_C,x)$$
holds.
By Lemma 2.3, this means that $`(L,h)C`$ is always nonnegative and
$$(L,h)C>0$$
holds, when
$$LC\underset{xC}{}\nu (\mathrm{\Theta }_h_C,x)>0$$
holds. Hence if $`(L,h)C=0`$ holds, then
$$LC=\underset{xC}{}\nu (\mathrm{\Theta }_h_C,x)$$
holds. This implies that
$$\mathrm{\Theta }_h_C=2\pi \underset{xC}{}\nu (\mathrm{\Theta }_h_C,x)x$$
holds. This completes the proof of Proposition 2.2. Q.E.D.
###### Definition 2.10
Let $`(L,h)`$ be a pseudoeffective singular hermitian line bundle on a smooth projective variety $`X`$. $`(L,h)`$ is said to be numerically trivial, if for every irreducible curve $`C`$ on $`X`$ such that $`h_C`$ is well defined,
$$(L,h)C=0$$
holds.
### 2.5 Restriction of the intersection theory to divisors
In the previous subsection we define an intersection number of a singular hermitian line bundle with positive curvature and an irreducible curve on which the restriction of the singular hermitian metric is well defined. In this subsection we shall consider the case that the restriction of the singular hermitian metric is not well defined.
Let $`(L,h)`$ be a pseudoeffective singular hermitian line bundle on a smooth projective variety $`X`$.
Let $`D`$ be a smooth divisor on $`X`$. We set
$$v_m(D)=\text{mult}_D\text{Spec}(๐ช_X/(h^m))$$
and
$$\stackrel{~}{}_D(h^m)=๐ช_D(v_m(D)D)(h^m).$$
Then $`\stackrel{~}{}_D(h^m)`$ is an ideal sheaf on $`D`$ (it is torsion free, since $`D`$ is smooth).
Let $`xD`$ be an arbitrary point of $`D`$ and let $`(U,z_1,\mathrm{},z_n)(n:=dimX)`$ be a local coordinate neighbourhood of $`x`$ which is biholomorphic to the unit open polydisk $`\mathrm{\Delta }^n`$ with center $`O`$ in $`\text{C}^n`$ and
$$UD=\{pUz_1(p)=0\}$$
holds. For $`q\mathrm{\Delta }^{n1}`$, we set $`\mathrm{\Delta }(q):=\{pU(z_2(p),\mathrm{},z_n(p))=q\}`$. Then considering the family of the restriction $`\{\mathrm{\Theta }_h_{\mathrm{\Delta }(q)}\}`$ for very general $`q\mathrm{\Delta }^{n1}`$, by Lemma 2.2, we see that
$$m\nu (\mathrm{\Theta }_h,D)1v_m(D)m\nu (\mathrm{\Theta }_h,D)$$
holds.
We define the ideal sheaves $`\sqrt[m]{\stackrel{~}{}_D(h^m)}`$ on $`D`$ by
$$\sqrt[m]{\stackrel{~}{}_D(h^m)}_x:=(\frac{1}{m}(\sigma ))_x(xD),$$
where $`\sigma `$ runs all the germs of $`\stackrel{~}{}_D(h^m)_x`$. And we set
$$_D(h):=_{m1}\sqrt[m]{\stackrel{~}{}_D(h^m)}$$
and call it the multipler ideal of $`h`$ on $`D`$. Also we set
$$\overline{}_D(h):=\underset{\epsilon 0}{lim}_D(h^{1+\epsilon }).$$
See Theorem 2.8 below for the reason why we define $`_D(h)`$ in this way.
Let $`C`$ be an irreducible curve in $`D`$ such that the natural morphism
$$\stackrel{~}{}_D(h^m)๐ช_C๐ช_C$$
is an isomorphism at the generic point of $`C`$ for every $`m0`$. In this case we can define the intersection number $`(L,h)C`$ by
$$(L,h)C:=\overline{lim}_m\mathrm{}m^1dimH^0(C,๐ช_C(mLv_m(D)D)\stackrel{~}{}_D(h^m)/tor).$$
Then as the formla $`(\mathrm{})`$ in Remark 2.5, we see that
$$(\mathrm{})(L,h)C=(L\nu (\mathrm{\Theta }_h,D)D)C+\overline{lim}_m\mathrm{}m^1\mathrm{deg}_C\stackrel{~}{}_D(h^m)๐ช_C$$
holds.
We may define the Lelong number $`\nu _D(\mathrm{\Theta }_h,x)(xD)`$ by
$$\nu _D(\mathrm{\Theta }_h,x):=\overline{lim}_m\mathrm{}m^1\text{mult}_x\text{Spec}(๐ช_D/\stackrel{~}{}_D(h^m)).$$
Then we see that the set
$$S_D:=\{xD\nu (\mathrm{\Theta }_h_D,x)>0\}$$
consists of a countable union of subvarieties on $`D`$. This follows from the approximation theorem \[4, p.380, Proposition 3.7\].
### 2.6 Another definition of the intersection numbers
Let $`(L,h)`$ be a pseudoeffective singular hermitian line bundle on a smooth projective variety $`X`$. And let $`C`$ be an irreducible curve on $`X`$ such that the restriction $`h_C`$ is well defined. Another candidate for the intersection number of $`(L,h)`$ and $`C`$ is :
$$(L,h)C:=LC\underset{xC}{}\nu (\mathrm{\Theta }_h_C,x).$$
But we have the following theorem.
###### Theorem 2.7
With the above notations
$$(L,h)C=(L,h)C$$
holds.
Proof of Theorem 2.7. To prove Theorem 2.7, by taking an embedded resolution of $`C`$, we may assume that $`C`$ is smooth, since the intersection number $`(L,h)C`$ is defined by using the asymptotics of the dimension of sections and $`\nu (\mathrm{\Theta }_h_C,x)(xC)`$ is defined in terms of the normalization of $`C`$.
In fact let $`\varpi :YX`$ be an embedded resolution of $`C`$ and let $`\stackrel{~}{C}`$ be the strict transform of $`C`$. Since by the definition of multiplier ideal sheaves
$$๐ช_X(K_X)(h^m)=\varpi _{}(๐ช_Y(K_Y)(\varpi ^{}h^m))$$
holds for every $`m0`$, we have that
$$\varpi ^{}(h^m)๐ช_Y(\varpi ^{}K_XK_Y)(\varpi ^{}h^m)$$
holds for every $`m0`$. Then by Remark 2.4, we have that
$$(L,h)C\overline{lim}_m\mathrm{}m^1dimH^0(\stackrel{~}{C},๐ช_{\stackrel{~}{C}}(m\varpi ^{}L)(\varpi ^{}h^m)/tor)$$
holds. Suppose that
$$(\varpi ^{}L,\varpi ^{}h)\stackrel{~}{C}=(\varpi ^{}L,\varpi ^{}h)\stackrel{~}{C}$$
holds. Then by the above inequality, we have that
$$(L,h)C(\varpi ^{}L,\varpi ^{}h)\stackrel{~}{C}$$
holds. Since by definition
$$(L,h)C=(\varpi ^{}L,\varpi ^{}h)\stackrel{~}{C}$$
holds, we have that
$$(L,h)C(L,h)C$$
holds. On the other hand by Lemma 2.3 and Lemma 2.2 (see also the explanataion right after Lemma 2.2), we see that the opposite inequality :
$$(L,h)C(L,h)C$$
holds. Hence we conclude that
$$(L,h)C=(L,h)C$$
holds.
Hereafter we shall assume that $`C`$ is smooth. First we note that for every ample line bundle $`H`$ on $`X`$ and a $`C^{\mathrm{}}`$-hermitian metric $`h_H`$ on $`H`$ with strictly positive curvature
$$(LH,hh_H)C=HC+(L,h)C$$
holds by the formula $`(\mathrm{})`$ and
$$(LH,hh_H)C=HC+(L,h)C$$
hold. Hence we may assume that $`h`$ is strictly positive.
Since we already have the inequality :
$$(L,h)C(L,h)C$$
as above, we only have to show the opposite inequality
$$(L,h)C(L,h)C$$
holds.
First we shall consider the case that $`h`$ has algebraic singularities. In this case by taking a suitable modification
$$f:\stackrel{~}{X}X$$
we see that there exists an effective Q-divisor $`D`$ with normal crossings on $`\stackrel{~}{X}`$ such that
$$(f^{}h^m)=๐ช_{\stackrel{~}{X}}(mD)$$
holds for every $`m0`$, where $``$ denotes the round up. Let $`\stackrel{~}{C}`$ denote the strict transform of $`C`$. We may assume that $`\stackrel{~}{C}`$ is smooth. By this
$$\mathrm{deg}_C((h^m)_C)=[mD]\stackrel{~}{C}+(K_{\stackrel{~}{X}}f^{}K_X)\stackrel{~}{C}$$
holds. On the other hand
$$\mathrm{deg}_C(h^m_C)=\mathrm{deg}_{\stackrel{~}{C}}[mD_{\stackrel{~}{C}}]$$
holds. Then since
$$\underset{m\mathrm{}}{lim}\frac{1}{m}[mD]\stackrel{~}{C}=\underset{m\mathrm{}}{lim}\frac{1}{m}\mathrm{deg}_C[mD_{\stackrel{~}{C}}]$$
holds, we have that
$$\underset{m\mathrm{}}{lim}\frac{1}{m}\mathrm{deg}_C((h^m)_C)=\underset{m\mathrm{}}{lim}\frac{1}{m}\mathrm{deg}_C(h^m_C)$$
holds. The lefthandside is equal to
$$LC(L,h)C$$
by the argument in Remark 2.5 (especially by the formula $`(\mathrm{})`$) and the righthandside is equal to
$$LC(L,h)C$$
by Lemma 2.2. Hence if $`h`$ has algebraic singularities,
$$(L,h)C=(L,h)C$$
holds.
On the other hand by (a slight generalization of) the approximation theorem of \[4, p.380, Proposition 3.7\], there exists a sequence of singular hermitian metrics $`\{h_j\}_{j=1}^{\mathrm{}}`$ satisfying the following 6-conditions :
1. $`\mathrm{\Theta }_{h_j}`$ is positive for every $`j`$,
2. $`lim_j\mathrm{}h_j=h`$ holds in the sense of the convergence of the weight functions as currents on $`M`$ and $`C`$,
3. $`h_j`$ has algebraic singularities,
4. $`(h^{jm})(h_j^{jm})`$, holds for every $`m0`$ and $`j1`$,
5. $`lim_j\mathrm{}\overline{}(h_j^m)=\overline{}(h^m)`$ holds for every $`m`$,
6. $`lim_j\mathrm{}\overline{}(h_j^m_C)=\overline{}(h^m_C)`$ holds for every $`m`$.
The third condition looks a little bit different from \[4, Proposition 3.7\]. But it is essentially the same by Lemma 2.2 and the construction of $`\{h_j\}`$ below. The 4-th condition cannot be deduced directly by the approximation theorem of \[4, p.380, Proposition 3.7\].
Let us briefly show how to construct $`\{h_j\}`$. The following argument is a slight modification of that in . First we shall consider the local approximation of a plurisubharmonic function by a sequence of plurisubharmonic functions with algebraic singularities.
Let $`\phi `$ be a plurisubharmonic function on $`\mathrm{\Delta }^n`$. Let $`C=\{p\mathrm{\Delta }^nz_2(p)=\mathrm{}z_n(p)=0\}`$. Suppose that $`\phi `$ is not identically $`\mathrm{}`$ on $`C`$. That is to say we are considering the case that $`h=e^\phi `$ and $`C`$ is a smooth curve in $`\mathrm{\Delta }^n`$. Let $`m`$ be a positive integer. Let $`(j\phi )_C`$ be the Hilbert space defined by
$$(j\phi )_C:=\{f๐ช(\mathrm{\Delta }^n)_{\mathrm{\Delta }^n}f^2e^{j\phi }๐\lambda <\mathrm{}\text{and}_Cf^2e^{j\phi }๐\lambda _C<\mathrm{}\}$$
with the inner product
$$(f,g):=\frac{1}{2}_{\mathrm{\Delta }^n}f\overline{g}e^{j\phi }๐\lambda +\frac{1}{2}_Cf\overline{g}e^{j\phi }๐\lambda _C$$
where $`d\lambda `$ and $`d\lambda _C`$ is the usual Lebesgue measure on $`\mathrm{\Delta }^n`$ and $`C`$ respectively. Let $`\{\sigma _{\mathrm{}}\}`$ be an orthonormal basis of $`(j\phi )_C`$ and let
$$\phi _j:=\frac{1}{2j}\mathrm{log}\sigma _{\mathrm{}}^2.$$
Let $`\psi `$ is the plurisubharmonic function on $`\mathrm{\Delta }^n`$ defined by
$$\psi =(n1)\mathrm{log}(\underset{i=2}{\overset{n}{}}z_i^2).$$
###### Proposition 2.3
There exist positive constants $`K_1,K_2>0`$ independent of $`m`$ such that
1. $$\phi (z)\frac{K_1}{j}\phi _j(z)\underset{\zeta z<r}{sup}\phi (\zeta )+\frac{1}{j}\mathrm{log}(\frac{K_2}{r^n})$$
holds for every $`zC`$ and $`r<d(z,\mathrm{\Delta }^n)`$ and
$$\phi (z)+\frac{1}{2j}\psi (z)\frac{K_1}{j}\phi _j(z)\underset{\zeta z<r}{sup}\phi (\zeta )+\frac{1}{j}\mathrm{log}(\frac{K_2}{r^n})$$
holds for every $`z\mathrm{\Delta }^nC`$ and $`r<d(z,\mathrm{\Delta }^n)`$,
2. $`\nu (\phi ,z)n/j\nu (\phi _j,z)\nu (\phi ,z)`$ holds for every $`z\mathrm{\Delta }^n`$.
Proof of Proposition 2.3. We note that
$$\phi _j(z)=\underset{fB(1)}{sup}\frac{1}{j}\mathrm{log}f(z)$$
holds, where $`B(1)`$ is the unit ball of $`(j\phi )_C`$. For $`r<\text{dist}(z,\mathrm{\Delta }^n)`$ and $`fB(1)`$, the mean value inequality applied to the plurisubharmonic function $`f^2`$ implies
$`f(z)^2`$ $``$ $`{\displaystyle \frac{1}{\pi ^nr^{2n}/n!}}{\displaystyle _{\zeta z<r}}f(\zeta )^2๐\lambda (\zeta )`$
$``$ $`{\displaystyle \frac{1}{\pi ^nr^{2n}/n!}}\mathrm{exp}(2j\underset{\zeta z<r}{sup}\phi (\zeta )){\displaystyle _{\mathrm{\Delta }^n}}f^2e^{2j\phi }๐\lambda `$
holds. If we take the supremum over all $`fB(1)`$ we have
$$\phi _j(z)\underset{\zeta z<r}{sup}\phi (\zeta )+\frac{1}{2j}\mathrm{log}\frac{1}{\pi ^nr^{2n}/n!}$$
holds.
Conversely, the $`L^2`$-extension theorem (Theorem 2.6) applied twice to the zero dimensional subvariety $`\{z\}C\mathrm{\Delta }^n`$ shows that for any $`a\text{C}`$ there is a holomorphic function $`f`$ on $`\mathrm{\Delta }^n`$ such that $`f(z)=a`$ and
$$_{\mathrm{\Delta }^n}f^2e^{j\phi }๐\lambda +_Cf^2e^{j\phi }๐\lambda _C2K_1a^2e^{2j\phi (z)},$$
where $`K_1`$ only depends on $`n`$. We fix $`a`$ such that the righthandside is $`1`$. This gives the other inequality
$$\phi _j(z)\frac{1}{j}\mathrm{log}a=\phi (z)\frac{\mathrm{log}K_1}{2j}.$$
If $`z\mathrm{\Delta }^nC`$, there is a holomorphic function $`f`$ on $`\mathrm{\Delta }^n`$ such that $`f(z)=a`$ and
$$_{\mathrm{\Delta }^n}f^2e^{j\phi \psi }๐\lambda K_1a^2e^{2j\phi (z)\psi (z)}$$
holds. In particular $`f_C0`$ holds in this case. This implies the inequality
$$\phi _j(z)\phi (z)+\frac{1}{2j}\psi (z)\frac{\mathrm{log}K_1}{2j}.$$
Hence we see that
$$\nu (\phi _j,z)\nu (\phi ,z)$$
holds for every $`z\mathrm{\Delta }^n`$. In the opposite direction we find
$$\underset{xz<r}{sup}\phi _j(x)\underset{\zeta z<2r}{sup}\phi (\zeta )+\frac{1}{j}\mathrm{log}\frac{K_2}{r^n}$$
holds, where $`K_2`$ is a positive constant independent of $`j`$. Thus we obtain
$$\nu (\phi _j,x)\nu (\phi ,x)\frac{n}{j}.$$
Q.E.D.
To construct $`\{h_j\}`$ we need to globalize the above argument,i.e. we need to glue local approximations. But this is completely parallel to the argument in \[4, pp. 377-380\]. Hence we omit it. We note that the glueing process in \[4, pp. 377-380, see especially p.377, Lemma 3.5\] does not change singularities of the sequence of approximations (up to quasi-isometry) on $`X`$ (hence in particular on $`C`$).
By the construction we have the following lemma.
###### Lemma 2.4
$$(h^{jm})(h_j^{jm})$$
holds for every $`j`$ and $`m0`$.
Proof of Lemma 2.4. By the construction of $`h_j`$ we see that
$$(h^j)_{\mathrm{}}(h_j^j)$$
holds (for the definition of $`_{\mathrm{}}`$ see Section 2.2). By the subadditivity theorem (), we see that
$$(h^{jm})(h^j)^m_{\mathrm{}}(h_j^j)^m_{\mathrm{}}(h_j^{jm})(h_j^{jm})$$
hold for every $`m0`$. Q.E.D.
By Lemma 2.4 the sequence $`\{h_j\}`$ satisfies the 4-th condition above. The 3-rd and 5-th conditions are satisfied by the convergences of the Lelong numbers
$$\underset{j\mathrm{}}{lim}\nu (\mathrm{\Theta }_{h_j})=\nu (\mathrm{\Theta }_h)$$
and
$$\underset{j\mathrm{}}{lim}\nu (\mathrm{\Theta }_{h_j}_C)=\nu (\mathrm{\Theta }_h_C)$$
which follow from Proposition 2.3. Since the first and the second conditions are cleary satisfied, $`\{h_j\}`$ is a desired sequence of singular hermitian metrics on $`L`$.
We note that for every $`m0`$, $`๐ช_C(mL)(h^m)`$ is torsion free, since it is a subsheaf of a locally free sheaf on a smooth variety $`C`$. Since $`dimC=1`$, this means that for every $`m0`$ $`๐ช_C(mL)(h^m)`$ is invertible on $`C`$. Since for every $`0k<j`$
$$\mathrm{deg}_C๐ช_C((jm+k)L)(h^{jm+k})\mathrm{deg}_C๐ช_C(jmL)(h^{jm})+k(LC)$$
holds, by Lemma 2.4 we see that for every $`0k<j`$
$$\mathrm{deg}_C๐ช_C((jm+k)L)(h^{jm+k})\mathrm{deg}_C๐ช_C(jmL)(h_j^{jm})+k(LC)$$
hold. Then by the Riemann-Roch theorem and the Kodaira vanishing theorem imply that
$$(L,h_j)C(L,h)C$$
holds. In particular we see that
$$\overline{lim}_j\mathrm{}(L,h_j)C(L,h)C$$
holds.
On the other hand since $`h_j`$ has algebraic singularities,
$$(L,h_j)C=(L,h_j)C$$
holds. This implies that
$$\overline{lim}_j\mathrm{}(L,h_j)C=\overline{lim}_j\mathrm{}(L,h_j)C=(L,h)C$$
hold. The last equality comes from the 2-nd condition. Combining the above inequalities, we have that
$$(L,h)C(L,h)C$$
holds. Since we already have the opposite inequality, we see that
$$(L,h)C=(L,h)C$$
holds. This completes the proof of Theorem 2.7. Q.E.D.
###### Corollary 2.1
Let $`(L,h)`$ be a pseudoeffective singular hermitian line bundle on a smooth projective variety $`X`$. Let $`Y`$ be a subvariety such that the restriction $`h_Y`$ is well defined. Then for every irreducible curve $`C`$ on $`Y`$ such that $`h_C`$ is well defined,
$$(L,h)C=(L,h)_YC$$
holds. In other words, the intersection theory is compactible with restrictions. In particular $`(L,h)_Y`$ is numerically trivial, if and only if $`(L,h)`$ is numerically trivial on $`Y`$.
By the additivity of Lelong numbers we have the following corollary.
###### Corollary 2.2
Let $`(L,h),(L^{},h^{})`$ be singular hermitian line bundles on a smooth projective variety $`X`$ such that the curvature currents $`\mathrm{\Theta }_h`$,$`\mathrm{\Theta }_h^{}`$ are positive. Then for an irreducible curve $`C`$ such that $`h_C`$ and $`h^{}_C`$ are both well defined,
$$(LL^{},hh^{})C=(L,h)C+(L^{},h^{})C$$
holds.
###### Theorem 2.8
Let $`(L,h)`$ be a singular hermitian line bundle on a smooth projective variety $`X`$. Suppose that $`\mathrm{\Theta }_h`$ is bounded from below by some negative multiple of a $`C^{\mathrm{}}`$-Kรคhler form on $`X`$. Let $`D`$ be a smooth divisor on $`X`$. If $`h_D`$ is well defined, then
$$\overline{}_D(h)=\overline{}(h_D)$$
holds.
Proof of Theorem 2.8. Since the statement is local, we may assume that $`X`$ is the unit open polydisk $`\mathrm{\Delta }^n=\{(z_1,\mathrm{},z_n)\text{C}^n;z_i<1,1in\}`$, $`D`$ is the divisor $`(z_n)`$ and $`L`$ is a trivial bundle with singular hermitian metric $`e^\phi `$, where $`\phi `$ is a plurisubharmonic function on $`\mathrm{\Delta }^n`$.
The proof of Theorem 2.8 is parallel to that of Theorem 2.7 , if we replace the curve $`C`$ by the divisor $`D`$.
First we shall consider the case that $`h`$ has algebraic singularities. Let
$$f:YX$$
be a modification such that there exits an effective Q-divisor $`F`$ with normal crossings on $`Y`$ such that
$$(f^{}h^m)=๐ช_Y([mF])$$
holds for every $`m0`$. Let $`E`$ be the strict transform of $`D`$ in $`Y`$. By the assumption the support of $`F`$ does not contain $`E`$. We may and do assume that $`E+F`$ is a divisor with normal crossings. Then we have that
$$_E(f^{}h^m)=๐ช_E([mF])$$
holds for every $`m0`$. And
$$()\stackrel{~}{}_D(h^m)=๐ช_X(K_X)f_{}(๐ช_Y(K_Y)๐ช_Y([mF]))๐ช_D$$
holds. Let
$$F=\underset{i=1}{\overset{\mathrm{}}{}}a_iF_i$$
be the irreducible decomposition of $`F`$. Let us fix an arbitrary point $`x`$ on $`X`$. For every $`1i\mathrm{}`$ and $`m1`$, we define the number $`b_i(m)`$ by
$$b_i(m):=\underset{\sigma }{inf}\text{mult}_{F_i}f^{}(\sigma ),$$
where $`\sigma `$ runs all the nonzero element of $`\stackrel{~}{}_D(h^m)_x`$.
Let $`(m\phi )`$ be the Hilbert space defined by
$$(m\phi ):=\{\varphi ๐ช(\mathrm{\Delta }^n)_{\mathrm{\Delta }^n}\varphi ^2e^{m\phi }๐\lambda <\mathrm{}\},$$
with the inner product
$$(\varphi ,\varphi ^{}):=_{\mathrm{\Delta }^n}\varphi \overline{\varphi }^{}e^{m\phi }๐\lambda ,$$
where $`d\lambda `$ is the usual Lebesgue measure on $`\mathrm{\Delta }^n`$. Let $`\{\sigma _{\mathrm{}}\}`$ be an orthonormal basis of $`(m\phi )`$ and let
$$\phi _m:=\frac{1}{2m}\mathrm{log}\underset{\mathrm{}}{}\sigma _{\mathrm{}}^2.$$
Clearly
$$b_i(m)=m\nu (f^{}\phi _m,F_i)$$
holds. We define the nonnegative numbers $`\{r_i\}`$ by
$$K_Y=f^{}K_X+\underset{i}{}r_iF_i+\text{other components}.$$
To estimate $`b_i(m)`$ we shall prove the following lemma.
###### Lemma 2.5
1. $`\nu (f^{}\phi _m,F_i)a_i`$ holds,
2. $$\nu (f^{}\phi _m,F_i)a_i\frac{n+r_i}{m}$$
holds for every $`m1`$.
Proof of Lemma 2.5. The first assertion follows from the parallel argument as in the proof of Proposition 2.3. In fact the $`L^2`$-extension theorem (Theorem 2.6) applied to the zero dimensional subvariety $`\{z\}\mathrm{\Delta }^n`$ shows that for any $`a\text{C}`$ there is a holomorphic function $`f`$ on $`\mathrm{\Delta }^n`$ such that $`f(z)=a`$ and
$$_{\mathrm{\Delta }^n}f^2e^{j\phi }๐\lambda 2K_1a^2e^{2j\phi (z)},$$
where $`K_1`$ only depends on $`n`$. This gives the inequality :
$$\phi _m\frac{1}{m}\mathrm{log}a=\phi \frac{\mathrm{log}K_1}{2m}.$$
This implies the first assertion.
Let us prove the second assertion. Let $`yY`$ be a general point on $`F_i`$ and let $`(U,w_1,\mathrm{},w_n)`$ be a local coordinate around $`y`$ such that $`U`$ is biholomorphic to the unit open polydisk in $`\text{C}^n`$ by the coordinate $`(w_1,\mathrm{},w_n)`$. We set
$$J=\frac{f^{}dz_1\mathrm{}dz_n}{dw_1\mathrm{}dw_n}.$$
Then $`J`$ is a holomorphic function on $`U`$. Let $`\varphi (m\phi )`$ be an arbitrary element.
Then my mean value inequality
$`f^{}\varphi (w)J(w)^2`$ $``$ $`{\displaystyle \frac{1}{\pi ^nr^{2n}/n!}}{\displaystyle _{\zeta w<r}}f^{}\varphi ^2J^2๐\lambda (\zeta )`$
$``$ $`{\displaystyle \frac{1}{\pi ^nr^{2n}/n!}}\mathrm{exp}(2m\underset{\zeta w<r}{sup}f^{}\phi (\zeta )){\displaystyle _{\mathrm{\Delta }^n}}f^{}\varphi (\xi )J^2f^{}e^{2m\phi }๐\lambda (\xi )`$
$``$ $`{\displaystyle \frac{1}{\pi ^nr^{2n}/n!}}\mathrm{exp}(2m\underset{\zeta w<r}{sup}f^{}\phi (\zeta )){\displaystyle _{\mathrm{\Delta }^n}}\varphi (z)^2e^{2m\phi }๐\lambda (z)`$
hold, where $`d\lambda `$ denotes the usual Lebesgue measure on the unit open polydisk.
We note that
$$\phi _m(z)=\underset{fB(1)}{sup}\frac{1}{m}\mathrm{log}f(z)$$
holds, where $`B(1)`$ is the unit ball of $`(m\phi )`$. If we take the supremum over all $`\varphi `$ in $`B(1)`$ in $`(m\phi )`$, we have that
$$\phi _m(w)\underset{\zeta w<r}{sup}\phi (\zeta )+\frac{1}{2m}\mathrm{log}\frac{1}{\pi ^nJ(w)^2r^{2n}/n!}$$
holds. Hence we have that
$$\nu (f^{}\phi _m,y)\nu (f^{}\phi ,y)\frac{n+r_i}{m}=a_i\frac{n+r_i}{m}$$
hold. This completes the proof of Lemma 2.5. Q.E.D.
We note that for every positive number $`ฯต`$ and positive integer $`m`$,
$$(h^{1+2ฯต}_D)\sqrt[m]{\stackrel{~}{}_D(h^{(1+ฯต)m})}$$
holds by the formula $`()`$ and the definition of $`\sqrt[m]{\stackrel{~}{}_D(h^m)}`$. Hence we have that
$$(h^{1+2ฯต}_D)_D(h^{1+ฯต})$$
holds. By the definition of the closure of multiplier ideal sheaves, letting $`ฯต`$ tend to $`0`$, we have that
$$\overline{}(h_D)\overline{}_D(h)$$
holds.
On the other hand by Lemma 2.5 we have that
$$\underset{m\mathrm{}}{lim}\frac{1}{m}b_i(m)=\underset{m\mathrm{}}{lim}\nu (f^{}\phi _m,F_i)=a_i$$
hold. By the definitions of $`b_i(m)`$ and $`\sqrt[m]{\stackrel{~}{}(h^m)}`$, we see that the opposite inclusion :
$$\overline{}(h_D)\overline{}_D(h)$$
holds. Hence
$$\overline{}_D(h)=\overline{}(h_D)$$
holds.
If $`h`$ is not of algebraic sigularities, by approximating $`h`$ by a sequence of singular hermitian metrics with algebraic singularities as in Section 2.6, we completes the proof of Theorem 2.8. Q.E.D.
## 3 Characterization of numerically trivial singular hermitian line bundles
In this section we prove Theorem 1.2. Let $`(L,h)`$ be a singular hermitian line bundle on a smooth projective variety $`X`$ with positive curvature current. Suppose that $`(L,h)`$ is numerically trivial on $`X`$. Let us define the closed positive current $`T`$ on $`X`$ by
$$T:=\frac{1}{2\pi }\mathrm{\Theta }_h\underset{D}{}\nu (\mathrm{\Theta }_h,D)D$$
where $`D`$ runs all the prime divisors on $`X`$.
Let us define the subset $`S`$ of $`X`$ by
$$S:=\{xX\nu (T,x)>0\}.$$
Then $`S`$ consists of at most countable union of subvarieties of codimension greater than or equal to $`2`$ by a theorem of Siu (). Let $`n`$ be the dimension of $`X`$. Let $`H`$ be a very ample divisor and let $`C`$ be a very general complete intersection curve of $`(n1)`$-members of $`H`$. If we take $`H`$ sufficiently ample and take $`C`$ very general we may assume that
$$CS=\mathrm{}$$
holds and $`C`$ intersects every prime divisor $`D`$ with $`\nu (\mathrm{\Theta }_h,D)>0`$ (such prime divisors are at most coutably many) at $`D_{reg}`$ transversally. Let $`\omega `$ be a Kรคhler form which represents $`c_1(H)`$. Let $`\sigma \mathrm{\Gamma }(X,๐ช_X(H))`$ be a very general nonzero element such that $`D=(\sigma )`$ is smooth and $`T_D`$ is well defined. Then by Stokesโ theorem,
$$T(\omega ^{n1})=_DT\omega ^{n2}$$
holds. Hence inductively we have that
$$T(\omega ^{n1})=_CT=LC\underset{D}{}\nu (\mathrm{\Theta }_h,D)DC$$
hold. On the other hand, by the choice of $`C`$ and Lemma 2.2 we see that
$$(h^m_C)๐ช_C([m\underset{D}{}\nu (\mathrm{\Theta }_h,D)D])$$
holds for every $`m0`$ (since $`C`$ is smooth, the both sides are torsion free). Hence if
$$LC\underset{D}{}\nu (\mathrm{\Theta }_h,D)DC>0$$
holds, then
$`(L,h)C`$ $``$ $`LC{\displaystyle \underset{D}{}}\nu (\mathrm{\Theta }_h,D)DC`$
$`=`$ $`\overline{lim}_m\mathrm{}{\displaystyle \frac{1}{m}}\mathrm{deg}_C๐ช_C(mL)๐ช_C([m{\displaystyle \nu (\mathrm{\Theta }_h,D)D}])`$
$`=`$ $`LC{\displaystyle \underset{D}{}}\nu (\mathrm{\Theta }_h,D)DC>0`$
hold by the Riemann-Roch theorem and the Kodaira vanishing theorem. This is the contradiction. Hence we see that
$$T(\omega ^{n1})=_CT=LC\underset{D}{}\nu (\mathrm{\Theta }_h,D)DC=0$$
hold. Since $`T`$ is closed positive, this implies that $`T0`$. Hence we conclude that
$$\mathrm{\Theta }_h=2\pi \underset{D}{}\nu (\mathrm{\Theta }_h,D)D$$
holds. This completes the proof of Theorem 1.2. Q.E.D.
By the proof of Theorem 1.2, we obtain the following.
###### Theorem 3.1
Let $`(L,h)`$ be a pseudoeffective singular hermitian line bundle on a smooth projective variety $`X`$. Then $`(L,h)`$ is numerically trivial if and only if for every irreducible curve $`C`$ such that the restriction $`h_C`$ is well defined
$$\mathrm{\Theta }_h_C=2\pi \underset{xC}{}\nu (\mathrm{\Theta }_h_C,x)x$$
holds.
By using the intersection theory on smooth divisors (cf. Section 2.5), we have the following corollary.
###### Corollary 3.1
Let $`X`$ be a smooth projective variety and let $`(L,h)`$ be a pseudoeffective singular hermitian line bundle on $`X`$. Let $`D`$ be a smooth divisor on $`X`$. Suppose that $`(L,h)`$ is numerically trivial on $`D`$. Then
$$S_D:=\{xD\nu _D(\mathrm{\Theta }_h,x)>0\}$$
is a sum of countably many prime divisors on $`D`$. And for every $`m0`$,
$$\overline{}_D(h^m)=(m\underset{E}{}\nu _D(\mathrm{\Theta }_h,E)E)$$
holds, where $`E`$ runs all prime divisors on $`D`$.
Here we do not need assume that the restriction $`h_D`$ is well defined.
Proof of Corollary 3.1. The proof is essentially same as that of Theorem 1.2.
Let $`X`$,$`D`$,$`(L,h)`$ be as above. Let $`\{F_i\}_{iI}`$ be the set of divisorial components of $`S_D`$. Let $`C`$ be a very general complete intersection curve of a sufficiently ample linear system $`H`$ on $`D`$ which does not intersects $`S_D_{iI}F_i`$ and meets every $`F_i(iI)`$ transversally. We set
$$a_i:=\nu _D(\mathrm{\Theta }_h.F_i).$$
By the definition of the intersection number
$$(L,h)C=(L\nu (\mathrm{\Theta }_h,D)D)C+\overline{lim}_m\mathrm{}m^1\mathrm{deg}_C\stackrel{~}{}_D(h^m)๐ช_C=0$$
hold. Hence if we take $`C`$ very general, we see that
$$(L\nu (\mathrm{\Theta }_h.D)D)C=\underset{iI}{}a_i(F_iC)$$
holds by the definition of $`\nu _D`$.
Suppose that $`S_D_{iI}F_i`$ is nonempty. Let $`\{C_t\}_{t\mathrm{\Delta }}`$ be a family of complete intersection curve of $`(dimD1)`$-members of $`H`$ such that
1. $`C_0`$ is not contained in $`S_D`$,
2. $`C_0(S_D_{iI}F_i)\mathrm{}`$, a very general member of $`\{C_t\}`$ does not intersects $`S_D_{iI}F_i`$ and meets every $`F_i`$ transversally.
Then by the uppersemicontinuity of Lelong numbers $`\nu _D`$ in countable Zariski topology (the uppersemicontinuity is ovbious by the definition of $`\nu _D`$ (cf. Section 2.5)), we see that
$$(L,h)C_0<0$$
holds. This is the contradiction, since $`(L,h)C_00`$ holds by the definition of the intersection number. Hence we have that $`S_D=_{iI}F_i`$ holds. This completes the proof of the first assertion.
Let us prove the second assertion. Let us fix an irreducible component $`F_i`$ of $`S_D`$. Let $`U`$ be a Stein open subset on $`D`$ and let $`\sigma \stackrel{~}{}(h^{\mathrm{}})(U)`$ be a very general element. Let $`ฯต`$ be any small positive number. Then by the definition of $`\nu _D(\mathrm{\Theta }_h)`$, we see that for every sufficiently large $`\mathrm{}`$
$$\text{(}\text{)}(1ฯต)\nu _D(\mathrm{\Theta }_h,F_i)\frac{1}{\mathrm{}}\text{mult}_{F_i}(\sigma )\nu _D(\mathrm{\Theta }_h,F_i)$$
hold. Then by the definition of $`_D(h^m)`$ and Lemma 2.2, we see that
$$\overline{}_D(h^m)(m\underset{iI}{}\nu _D(\mathrm{\Theta }_h,F_i)F_i)$$
holds.
Let $`C`$ be a very general smooth complete intersection of $`(dimD1)`$-members of $`H`$. Then as above
$$((L\nu (\mathrm{\Theta }_h,D)D)_D\underset{iI}{}a_iF_i)C=0$$
holds.
We claim that $`L\nu (\mathrm{\Theta }_h,D)D_{iI}a_iF_i`$ is pseudoeffective in the sense that $`c_1((L\nu (\mathrm{\Theta }_h,D)D)_D_{iI}a_iF_i)`$ is on the closure of effective cone of $`D`$. Let $`G`$ be an ample line bundle on $`X`$ such that $`๐ช_X(G+mL)(h^m)`$ is globally generated on $`X`$ for every $`m0`$. This is possible by \[14, p.664, Proposition 1\]. By the formula ($``$), this implies that for any sufficiently small $`ฯต>0`$ and every finite subset $`I_0`$ of $`I`$
$$(L\nu (\mathrm{\Theta }_h,D)D)_D\underset{iI_0}{}a_iF_i$$
is pseudoeffective. Hence this we see that $`(L\nu (\mathrm{\Theta }_h,D)D)_D_{iI}a_iF_i`$ is pseudoeffective.
Since
$$((L\nu (\mathrm{\Theta }_h,D)D)_D\underset{iI}{}a_iF_i)H^{dimD1}=0$$
holds, this implies that $`L\nu (\mathrm{\Theta }_h,D)D)_D_{iI}a_iF_i`$ is numerically trivial.
Let $`f:\stackrel{~}{D}D`$ be any composition of successive blowing ups with smooth center, then by the same argument as above, we see that
$$f^{}(L\nu (\mathrm{\Theta }_h,D)D)_D\underset{\stackrel{~}{E}}{}\nu _{\stackrel{~}{D}}(f^{}\mathrm{\Theta }_h,\stackrel{~}{E})\stackrel{~}{E}$$
is numerically trivial, where $`\stackrel{~}{E}`$ runs all the prime divisors on $`\stackrel{~}{D}`$. We note that by the definitions of $`\nu _D`$ and $`\nu _{\stackrel{~}{D}}`$, we see that
$$\underset{\stackrel{~}{E}}{}\nu _{\stackrel{~}{D}}(f^{}\mathrm{\Theta }_h,\stackrel{~}{E})\stackrel{~}{E}f^{}(\underset{iI}{}a_iF_i)$$
is effective, i.e. a sum of prime divisors with nonnegative coefficients. Since $`f^{}((L\nu (\mathrm{\Theta }_h,D)D)_D_{iI}a_iF_i)`$ is numerically trivial on $`\stackrel{~}{D}`$, we see that
$$\underset{\stackrel{~}{E}}{}\nu _{\stackrel{~}{D}}(f^{}\mathrm{\Theta }_h,\stackrel{~}{E})\stackrel{~}{E}=f^{}(\underset{iI}{}a_iF_i).$$
Let $`m`$ be any positive integer and $`f_m:D_mD`$ be a modification such that $`f_m^{}\overline{}(h^m)`$ is locally free. Then by the definition of $`\overline{}_D(h^m)`$, it is determined by the Lelong numbers $`\nu _{D_m}`$ on prime divisors on $`D_m`$. Applying the above argument by taking $`\stackrel{~}{D}`$ to be $`D_m`$, we see that
$$\overline{}_D(h^m)=(m\underset{iI}{}a_iF_i)$$
holds. This completes the proof of Corollary 3.1. Q.E.D.
###### Remark 3.1
By the above proof Corollary 3.1 still holds for a subvariety $`V`$ on $`D`$, if there exists a curve on $`V`$ such that $`(L,h)C`$ is well defined (cf. \[18, Remark 3.1\]).
## 4 Numerical triviality and the growth of $`H^0`$
In this section we shall relate the numerical triviality of singular hermitian line bundles with positive curvature current and the growth of dimension of global sections.
###### Definition 4.1
Let $`(L,h)`$ be a singular hermitian line bundle on a smooth projective variety $`X`$. Let $`H`$ be an ample line bundle on $`X`$. We define the number $`\mu _h(X,H+mL)`$ by
$$\mu _h(X,H+mL):=(dimX)!\overline{lim}_{\mathrm{}\mathrm{}}\mathrm{}^{dimX}dimH^0(X,๐ช_X(\mathrm{}(H+mL))(h^m\mathrm{}))$$
For a subvariety $`Y`$ in $`X`$ such that $`(L,h)_Y`$ is well defined, we define
$$\mu _h(Y,H+mL):=(dimY)!\overline{lim}_{\mathrm{}\mathrm{}}\mathrm{}^{dimY}dimH^0(Y,๐ช_Y(\mathrm{}(H+mL))(h^m\mathrm{})/tor),$$
where $`tor`$ denotes the torsion part of $`๐ช_Y(H+mL)(h^m)`$.
We note that $`\mu _h(Y,H+mL)`$ is different from $`\mu _h(Y,H+mL_Y)`$ in general. By Corollary 2.1, we note that if $`(L,h)`$ is numerically trivial on $`Y`$ if and only if $`(L,h)_Y`$ is numerically trivial.
###### Lemma 4.1
Suppose that $`(L,h)`$ is pseudoeffective and is not numerically trivial on $`X`$. Then
$$\overline{lim}_m\mathrm{}m^1\mu _h(X,H+mL)>0$$
holds for every ample line bundle $`H`$ on $`X`$.
Proof of Lemma 4.1. Let $`n`$ be the dimension of $`X`$. We prove this lemma by induction on $`n`$. If $`n=1`$, then for every $`\mathrm{}0`$
$$๐ช_X(\mathrm{}L)(h^{\mathrm{}})๐ช_X(\mathrm{}L\underset{xX}{}[\mathrm{}\nu (\mathrm{\Theta }_h,x)])$$
holds by Lemma 2.2. Hence by Theorem 1.2, we see that
$$\underset{\mathrm{}\mathrm{}}{lim}\mathrm{deg}_X๐ช_X(\mathrm{}L)(h^{\mathrm{}})=+\mathrm{}.$$
This implies Lemma 4.1.
Let $`\pi :\stackrel{~}{X}๐^1`$ be a Lefschetz pencil associated with a very ample linear system say $`H`$ on $`X`$. If we take the pencil very general, we may assume that $`(h^{\mathrm{}})`$ is an ideal sheaf on all fibers of $`\pi `$ for every $`\mathrm{}`$. Let
$$b:\stackrel{~}{X}X$$
be the modification associated with the pencil and let $`E`$ be the exceptional locus of $`b`$. We note that on the Hilbert scheme of curves in $`X`$, the intersection number $`(L,h)C`$ is lower semicontinuous in countable Zariski topology by the upper semicontinuity of the Lelong number (or by the $`L^2`$-extension theorem (Theorem 2.6)), where $`C`$ moves in the Hilbert scheme. Then by the inductive assumption for a general fiber $`F`$ of $`\pi `$ we see that
$$\overline{lim}_m\mathrm{}m^1\mu _h(F,b^{}(H+mL))>0$$
holds. Let us consider the direct image
$$_{m,\mathrm{}}:=\pi _{}๐ช_{\stackrel{~}{X}}(\mathrm{}b^{}(H+mL)(b^{}(h^m\mathrm{}))).$$
By Grothendiekโs theorem, we see that
$$_{m,\mathrm{}}_{i=1}^r๐ช_{๐^1}(a_i)$$
for some $`a_i=a_i(m,\mathrm{})`$ and $`r=r(m,\mathrm{})`$. By the inductive assumption we see that
$$\overline{lim}_m\mathrm{}m^1(\overline{lim}_{\mathrm{}\mathrm{}}\mathrm{}^{(n1)}r(m,\mathrm{}))>0$$
holds. We note that $`\mathrm{}_0b^{}HE`$ is ample for some large positive integer $`\mathrm{}_0`$. Hence we see that
$$๐ช_{\stackrel{~}{X}}(\mathrm{}_0b^{}HE)$$
admits a $`C^{\mathrm{}}`$-hermitian metric $`h_0`$ with strictly positive curvature.Let $`h_1`$ be a $`C^{\mathrm{}}`$-hermitian metric on $`๐ช_{\text{P}^1}(1)`$. Then there exists a positive rational number $`c`$ such that
$$\frac{1}{\mathrm{}_0}\mathrm{\Theta }_{h_0}c\pi ^{}\mathrm{\Theta }_{h_1}$$
is a Kรคhler form on $`\stackrel{~}{X}`$. By Nadelโs vanishing theorem (Theorem 2.1),
$$H^1(\stackrel{~}{X},๐ช_{\stackrel{~}{X}}(\mathrm{}(b^{}(H+mL)\frac{1}{\mathrm{}_0}E))(b^{}(h^m\mathrm{}))\pi ^{}๐ช_{๐^1}(c\mathrm{}))=0$$
holds for every sufficiently large $`\mathrm{}`$ such that $`\mathrm{}/\mathrm{}_0`$ and $`c\mathrm{}`$ are integers. Also by Nadelโs vanishing theorem, we see that
$$R^1\pi _{}๐ช_{\stackrel{~}{X}}(\mathrm{}(b^{}(H+mL)\frac{1}{\mathrm{}_0}E))(b^{}(h^m\mathrm{}))$$
is the $`0`$-sheaf on $`\text{P}^1`$ for every sufficiently large $`\mathrm{}`$ divisible by $`\mathrm{}_0`$. Hence we see that $`_{m,\mathrm{}}๐ช_{\text{P}^1}(c\mathrm{}+1)`$ is globally generated on $`\text{P}^1`$ for every sufficiently large $`\mathrm{}`$ such that $`\mathrm{}/\mathrm{}_0`$ and $`c\mathrm{}`$ are integers. This implies that
$$\overline{lim}_{\mathrm{}\mathrm{}}\mathrm{}^1\underset{i}{\mathrm{min}}a_ic$$
holds for every $`i`$. Hence
$$\overline{lim}_{\mathrm{}\mathrm{}}\mathrm{}^ndimH^0(\stackrel{~}{X},๐ช_{\stackrel{~}{X}}(\mathrm{}b^{}(H+mL))(b^{}(h^m\mathrm{})))$$
$$c\overline{lim}_{\mathrm{}\mathrm{}}\mathrm{}^{(n1)}r(m,\mathrm{})$$
holds. By this we see that
$$\overline{lim}_m\mathrm{}m^1(\overline{lim}_{\mathrm{}\mathrm{}}\mathrm{}^ndimH^0(\stackrel{~}{X},๐ช_{\stackrel{~}{X}}(\mathrm{}b^{}(H+mL))(b^{}(h^m\mathrm{})))>0$$
holds. Since
$$b_{}(b^{}h^m\mathrm{})((h^m\mathrm{}))$$
holds by Lemma 2.1, we see that
$$\overline{lim}_m\mathrm{}m^1\mu _h(X,H+mL)>0$$
holds. Here we have assumed that $`H`$ to be sufficiently very ample. To prove the general case of Lemma 4.1, we argue as follows. Let $`H`$ be any ample line bundle on $`X`$. Then thanks to Nadelโs vanishing theorem
$$\mu _h(X,a(H+mL))=a^n\mu _h(X,H+mL)$$
holds for every positive integer $`a`$. Now it is clear that Lemma 4.1 holds for any ample line bundle $`H`$. This completes the proof of Lemma 4.1. Q.E.D.
###### Theorem 4.1
Let $`(L,h)`$ be a pseudoeffective singular hermitian line bundle on a smooth projective variety $`X`$. Then $`(L,h)`$ is numerically trivial if and only if
$$\overline{lim}_m\mathrm{}\mu _h(X,H+mL)<\mathrm{}$$
holds for every ample line bundle $`H`$ on $`X`$.
Proof of Theorem 4.1. By Lemma 4.1, $`(L,h)`$ is numerically trivial, if
$$\overline{lim}_m\mathrm{}\mu _h(X,H+mL)<\mathrm{}$$
holds for every ample line bundle $`H`$ on $`X`$.
Let us prove the converse. Suppose that
$$\overline{lim}_m\mathrm{}\mu _h(X,H+mL)=\mathrm{}$$
holds for some ample line bundle $`H`$ on $`X`$. Let $`x`$ be a very general point of $`X`$ such that
$$(h^m)_x=๐ช_{X,x}$$
holds for every $`m0`$.
###### Lemma 4.2
For every postive integer $`N`$ there exists a positive integer $`m_0`$ such that for every sufficienly large $`\mathrm{}`$ there exists a section
$$\sigma _{\mathrm{}}H^0(X,๐ช_X(\mathrm{}(H+m_0L))(h^{m_0\mathrm{}})_x^N\mathrm{})\{0\}.$$
Proof of Lemma 4.2.
$$\overline{lim}_m\mathrm{}\mu _h(X,H+mL)=\mathrm{}$$
holds by the assumption. Hence there exists a positive integer $`m_0`$ such that
$$\mu _h(X,H+m_0L)>N^{dimX}+1$$
holds. Then
$$dimH^0(X,๐ช_X(\mathrm{}(H+m_0L))(h^{m_0\mathrm{}}))\frac{N^{dimX}+1}{(dimX)!}\mathrm{}^{dimX}+o(\mathrm{}^{dimX})$$
holds. We consider the exact sequence
$$0H^0(X,๐ช_X(\mathrm{}(H+m_0L))(h^{m_0\mathrm{}})_x^N\mathrm{})H^0(X,๐ช_X(\mathrm{}(H+m_0L))(h^{m_0\mathrm{}}))$$
$$H^0(X,๐ช_X(\mathrm{}(H+m_0L))(h^{m_0\mathrm{}})๐ช_X/_x^N\mathrm{}).$$
Since
$$(h^m)_x=๐ช_{X,x}$$
holds for every $`m0`$, we see that
$$dimH^0(X,๐ช_X(\mathrm{}(H+m_0L))(h^{m_0\mathrm{}})๐ช_X/_x^N\mathrm{})=\frac{N^{dimX}}{(dimX)!}\mathrm{}^{dimX}+o(\mathrm{}^{dimX})$$
holds. Combining the above facts, we see that
$$H^0(X,๐ช_X(\mathrm{}(H+m_0L))(h^{m_0\mathrm{}})_x^N\mathrm{})0$$
holds for every sufficiently large $`\mathrm{}`$. This completes the proof of Lemma 4.2. Q.E.D.
Let us continue the proof of Theorem 4.1. Let $`H_0`$ be a sufficiently ample line bundle. Let $`C`$ be a very general complete intersection of $`(dimX1)`$-members of $`H_0`$ such that $`xC`$. We may assume that $`h_C`$ is well defined and
$$\sigma _{\mathrm{}}_C0$$
holds for every sufficiently large $`\mathrm{}`$. This implies by a degree argument that
$$\overline{lim}_m\mathrm{}\mu _h(C,H+mL)N$$
holds. Since $`N`$ is arbitrary, we may take $`m_0`$ so that
$$\mu _h(C,H+m_0L)3HC$$
holds. Then for every sufficiently large $`\mathrm{}`$
$$dimH^0(C,๐ช_C(\mathrm{}(H+m_0L)\mathrm{}H))(h^m))(HC)\mathrm{}$$
holds. Hence
$$(L,h)C>0$$
holds. This completes the proof of Theorem 4.1. Q.E.D.
###### Theorem 4.2
Let $`f:YX`$ be a surjective morphism between smooth projective varieties. Let $`(L,h)`$ be a pseudoeffective singular hermitian line bundle on $`X`$. Then $`(L,h)`$ is numerically trivial on $`X`$ if and only if $`f^{}(L,h)`$ is numerically trivial on $`Y`$.
Proof of Theorem 4.2. If $`(L,h)`$ is numerically trivial on $`X`$, then by Theorem 1.2, $`\mathrm{\Theta }_h`$ is a sum of at most countably many prime divisors with nonnegative coefficients. Hence $`f^{}\mathrm{\Theta }_h`$ is at most countably many prime divisors with nonnegative coefficients. Hence by Theorem 3.1, $`f^{}(L,h)`$ is numerically trivial on $`Y`$.
Suppose that $`(L,h)`$ is not numerically trivial on $`X`$. Let $`H`$ be a sufficiently very ample line bundle on $`Y`$ and let $`C`$ be a very general complete intersection curve of $`dimY1`$ members of $`H`$. We may assume that
1. $`C`$ is smooth,
2. $`f(C)`$ is a smooth curve,
3. $`f_C:Cf(C)`$ is unramified on $`\{yC\nu (f^{}\mathrm{\Theta }_h_C,y)>0\}`$,
4. $`(L,h)f(C)>0`$ holds.
Then we have that
$$\frac{1}{2\pi }_Cf^{}\mathrm{\Theta }_h\underset{yC}{}\nu (f^{}\mathrm{\Theta }_h,y)=\mathrm{deg}(f_C)(\frac{1}{2\pi }_{f(C)}\mathrm{\Theta }_h\underset{xf(C)}{}\nu (\mathrm{\Theta }_h_C,x))>0$$
holds. Hence $`f^{}(L,h)`$ is not numerically trivial on $`Y`$. This completes the proof of Theorem 4.2. Q.E.D.
By Theorem 4.2, we may define the numerical triviality of pseudoeffective singular hermitian line bundles on singular varieties.
###### Definition 4.2
Let $`X`$ be a singular variety and let
$$\pi :\stackrel{~}{X}X$$
be a resolution of singularities. Let $`L`$ be a line bundle on $`X`$. A hermitian metric $`h`$ on $`L_{X_{reg}}`$ is said to be a singular hermitian metric on $`X`$, if $`\pi ^{}h`$ is a singular hermitian metric with curvature current bounded from below by a $`C^{\mathrm{}}`$-from on $`\stackrel{~}{X}`$.
$`(L,h)`$ is said to be pseudoeffective, if $`\pi ^{}(L,h)`$ is pseudoeffective.
Suppose that $`X`$ is proper and $`\stackrel{~}{X}`$ is smooth projective . A singular hermitian line bundle $`(L,h)`$ is said to be numerically trivial, if $`\pi ^{}(L,h)`$ is numerically trivial.
The above definition is independent of the choice of the resolution $`\pi `$, by the $`L^1`$-property of almost plurisubharmonic functions and Theorem 4.2.
## 5 The fibration theorem
In this section we shall prove Theorem 1.1.
### 5.1 Key lemma
The following lemma is the key for the proof of Theorem 1.1.
###### Lemma 5.1
Let $`f:MB`$ be an algebraic fiber space and let $`(L,h)`$ be a pseudoeffective singular hermitian line bundle on $`M`$. Suppose that for every very general fiber $`F`$, $`(L,h)`$ is numerically trivial on $`F`$ and there exists a subvariety $`W`$ of $`M`$ such that
1. $`h_W`$ is well defined,
2. $`(L,h)`$ is numerically trivial on $`W`$.
3. $`f(W)=B`$.
Then $`(L,h)`$ is numerically trivial on $`M`$.
Proof of Lemma 5.1. Taking a suitable modification of $`M`$, by Theorem 1.2 and Theorem 4.2 we may assume that $`W`$ is a smooth divisor.
Suppose that $`(L,h)`$ is not numerically trivial on $`M`$. Then there exists an ample line bundle $`H`$ on $`M`$ such that
$$\overline{lim}_m\mathrm{}\mu _h(M,mL+H)=\mathrm{}$$
holds. We may assume that $`H`$ is very ample on $`M`$.
By the assumption we see that
$$(h^m)_x=๐ช_{M,x}$$
for a very general point $`xW`$ and every $`m0`$. Let $`x_0`$ be a very general point of $`W`$ such that
$$(h^m)_{x_0}=๐ช_{M,x_0}$$
holds for every $`m0`$. The proof of the following lemma is identical to that of Lemma 4.2. Hence we omit it.
###### Lemma 5.2
For any positive integer $`N`$ there exists a positive integer $`m_0`$ such that
$$H^0(M,๐ช_M(\mathrm{}(m_0L+H))(h^{\mathrm{}m_0})_{x_0}^N\mathrm{})0$$
holds for every sufficiently large $`\mathrm{}`$.
Let us continue the proof of Lemma 5.1. Let $`N`$ be a sufficiently large positive integer and let $`m_0`$ be the integer as in Lemma 5.2. For every sufficiently large $`\mathrm{}`$, we take an element
$$\sigma _{\mathrm{}}H^0(M,๐ช_M(\mathrm{}(m_0L+H))(h^{\mathrm{}m_0})_{x_0}^N\mathrm{})\{0\}.$$
Let $``$ be the family of smooth curves which are complete intersection of $`dimW1`$ members of $`H_W`$ on $`W`$. Let $`d_0`$ be a large positive integer such that
$$H^{dimF1}F(Hd_0W)<0$$
holds for every general fiber $`F`$ of $`f`$. Since $`(L,h)`$ is numerically trivial on $`W`$, we have the following lemma.
###### Lemma 5.3
There exists a positive constant $`A_0`$ independent of $`m_0`$ such that for every member $`R`$ of $``$ such that the restriction $`h_R`$ is well defined,
$$()dimH^0(R,๐ช_R(\mathrm{}(H+m_0L)sW)(h^{m_0\mathrm{}}))A_0\mathrm{}+o(\mathrm{})$$
holds for every $`0sd_0\mathrm{}`$.
Proof of Lemma 5.3. We note that $`(h^{m_0\mathrm{}})_R`$ is torsion free, hence locally free (note that $`dimR=1`$), since $`R`$ is smooth and $`(h^{m_0\mathrm{}})_R`$ is a subsheaf of $`๐ช_R`$. Then since $`(L,h)_W`$ is numerically trivial, by Corollary 2.1, by the formula $`(\mathrm{})`$ in Remark 2.5 there exists a positive constant $`A_0`$ independent of $`m_0`$ such that
$$\mathrm{deg}_R(๐ช_R(\mathrm{}(H+m_0L)sW)(h^{m_0\mathrm{}}))A_0\mathrm{}+o(\mathrm{})$$
holds for every $`\mathrm{}0`$ and $`0sd_0\mathrm{}`$.
First let us consider the case that $`WR0`$ holds. Since $`H`$ is ample, we have that
$$H^1(R,๐ช_R(\mathrm{}(H+m_0L)sW)(h^{m_0\mathrm{}}))=0$$
holds for every sufficiently large $`\mathrm{}`$ and $`0sd_0\mathrm{}`$. By the Riemann-Roch theorem we have that
$$dimH^0(R,๐ช_R(\mathrm{}(H+m_0L)sW)(h^{m_0\mathrm{}}))A_0\mathrm{}+o(\mathrm{})$$
holds for every $`0sd_0\mathrm{}`$.
Next let us consider the case that $`WR>0`$ holds. Then there exists a positive integer $`a_0`$ such that for every $`sa_0`$, $`H^0(R,๐ช_R(sW))0`$ holds. This implies that
$$\text{(1)}dimH^0(R,๐ช_R(\mathrm{}(H+m_0L)sW)(h^{m_0\mathrm{}}))dimH^0(R,๐ช_R(\mathrm{}(H+m_0L))(h^{m_0\mathrm{}}))$$
holds for every $`sa_0`$. Hence as before we see that there exists a positive constant $`A_0`$ such that
$$dimH^0(R,๐ช_R(\mathrm{}(H+m_0L)sW)(h^{m_0\mathrm{}}))A_0\mathrm{}+o(\mathrm{})$$
holds for every $`sa_0`$. On the other hand since $`H`$ is ample, for every sufficiently large $`\mathrm{}`$ and every $`0sa_0`$, we see that
$$H^1(R,๐ช_R(\mathrm{}(H+m_0L)sW)(h^{m_0\mathrm{}}))=0$$
holds. Hence we see that by the Riemann-Roch theorem
$$\text{(2)}dimH^0(R,๐ช_R(\mathrm{}(H+m_0L)sW)(h^{m_0\mathrm{}}))=$$
$$=1g(R)+\mathrm{deg}_R๐ช_R(\mathrm{}(H+m_0L)sW)(h^{m_0\mathrm{}})$$
$$1g(R)+\mathrm{deg}_R๐ช_R(\mathrm{}(H+m_0L))(h^{m_0\mathrm{}})$$
hold for every sufficiently large $`\mathrm{}`$ and every $`0sa_0`$, where $`g(R)`$ denotes the genus of $`R`$. Combining the inequalities (1) and (2) above, we see that there exists a positive constant $`A_0`$ such that
$$dimH^0(R,๐ช_R(\mathrm{}(H+m_0L)sW)(h^{m_0\mathrm{}}))A_0\mathrm{}+o(\mathrm{})$$
holds for every sufficiently large $`\mathrm{}`$ and $`0sd_0\mathrm{}`$, also in this case. Q.E.D.
Take $`N>A_0`$ and the corresponding $`m_0`$ in Lemma 5.2. We see that using the case $`s=0`$ of $`()`$, for every member $`R`$ of $``$ containing $`x_0`$, by a degree argument
$$\sigma _{\mathrm{}}_R0$$
holds for every sufficiently large $`\mathrm{}`$. Since the members of $``$ containing $`x_0`$ dominates $`W`$, we see that
$$\sigma _{\mathrm{}}_W0$$
holds for every sufficiently large $`\mathrm{}`$. Next we consider the vanishing order of $`\sigma _{\mathrm{}}`$ along the divisor $`W`$. Repeating the same arugument we see that
$$\sigma _{\mathrm{}}H^0(M,๐ช_M(\mathrm{}(H+m_0L)d_0\mathrm{}W)(h^{m_0\mathrm{}}))$$
holds. Let $`F`$ be a very general fiber of $`f`$ such that $`(L,h)_F`$ well defined and is numerically trivial.
Let $`๐ฎ_F`$ denote the family of smooth curves complete intersection of $`dimF1`$ members of the very ample linear system $`H_F`$ on $`F`$. We note that $`F`$ is dominated by a family $`๐ฎ_F`$ of smooth curves passing through $`WF`$ and
$$H^{dimF1}F(Hd_0W)<0$$
holds. Since $`\sigma _{\mathrm{}}_F`$ has the vanishing order at least $`d_0\mathrm{}`$ along $`WF`$ and $`(L,h)`$ is numerically trivial on $`F`$, for every sufficiently large $`\mathrm{}`$, $`\sigma _{\mathrm{}}`$ is identically $`0`$ along any members of $`๐ฎ_F`$. In fact for every $`[S]๐ฎ_F`$ and every sufficiently large $`\mathrm{}`$
$$\mathrm{deg}_S๐ช_S(\mathrm{}(H+m_0L)d_0\mathrm{}W)(h^{m_0\mathrm{}})$$
is negative, since
$$\overline{lim}_m\mathrm{}\frac{1}{m}\mathrm{deg}_S๐ช_S(mL)(h^m)=(L,h)S=0$$
hold (cf. the formula $`(\mathrm{})`$ in Remark 2.5). Hence
$$\sigma _{\mathrm{}}_F0$$
holds for every sufficiently large $`\mathrm{}`$. Moving smooth fibers $`F`$, $`๐ฎ_F`$ forms a dominating family of curves $`๐ฎ`$ on $`M`$. We may take such $`\mathrm{}`$ independent of a very general $`F`$, since there exists a nonempty Zariski open subset $`๐ฎ_{\mathrm{}}`$ of $`๐ฎ`$ such that for every $`[S]๐ฎ_{\mathrm{}}`$, $`(h^{m_0\mathrm{}})๐ช_S`$ is an ideal sheaf on $`S`$ and
$$\mathrm{deg}_S๐ช_S(\mathrm{}(H+m_0L)d_0\mathrm{}W)(h^{m_0\mathrm{}})$$
is independent of $`[S]๐ฎ_{\mathrm{}}`$. This implies that
$$\sigma _{\mathrm{}}0$$
holds on $`M`$ for every sufficiently large $`\mathrm{}`$. This is the contradiction. This completes the proof of Lemma 5.1. Q.E.D.
### 5.2 Proof of Theorem 1.1
Let $`x`$ be an arbitrary point on $`X`$. We set
$$๐ฉ(x):=\{V\text{a subvariety of }X\text{ such that }xV\text{ and }(L,h)\text{ is }$$
$$\text{numerically trivial on }V\}.$$
Let $`\nu (x)`$ denote the maximal dimension of the member of $`๐ฉ(x)`$ and we set
$$\nu :=\underset{xX}{inf}\nu (x).$$
Then for very general $`xX`$, we see that $`\nu =\nu (x)`$ holds. We note that on the Hilbert scheme of curves in $`X`$, the intersection number $`(L,h)C`$ is lower semicontinuous in countable Zariski topology by the upper-semicontinuity of the Lelong number (or by the $`L^2`$-extension theorem (Theorem 2.6)), where $`C`$ moves in the Hilbert scheme. Hence for every irreducible component of the Hilbert scheme of $`X`$, the set of members on which the restriction of $`(L,h)`$ is well defined and numerically trivial is locally closed in countable Zariski topology. Since the Hilbert scheme of $`X`$ has only countably many components, this implies that there exists an irreducible subvariety $`๐ฉ^0`$ in the Hilbert scheme of $`X`$ whose members dominate $`X`$ and for a very general point $`x`$, there exists a member $`V`$ of $`๐ฉ^0`$ such that
1. $`xV`$,
2. $`dimV=\nu `$,
3. $`(L,h)`$ is numerically trivial on $`V`$.
Let
$$\phi :๐ฑ๐ฉ^0$$
be the universal family and let
$$p:๐ฑX$$
be the natural morphism.
###### Lemma 5.4
Let $`V`$ be a very general member of $`๐ฉ_0`$. Then there exists a Zariski open subset $`V_0V`$ such that $`V`$ is the unique member of $`๐ฉ_0`$ which intersects $`V_0`$.
Proof of Lemma 5.4. Supppose the contrary. Let $`V`$ be a very general member of $`๐ฉ_0`$. Let $`\eta `$ denote the generic point of $`V`$. We define the closed subset $`V_1`$ of $`X`$ by
$$V_1=\text{the closure of}p(\phi ^1(\phi (p^1(\eta )))).$$
By the assumption we see that $`dimV_1>dimV`$ holds (if there are only finitely many members of $`๐ฉ_0`$ which intersect $`V`$, for a suitable choice of $`V_0`$ the assertion is cleary satisfied). We note that $`V_1`$ may be reducible. Let $`S_1`$ be the closed subset of the closure of $`\phi ^1(\phi (p^1(\eta )))`$ defined by
$$S_1:=\text{the closure of}p^1(\eta ).$$
We note that $`p^{}(L,h)`$ is numerically trivial on $`S_1`$ by Theorem 4.2, since $`(L,h)`$ is numerically trivial on $`V`$ and $`p(S_1)=V`$ holds. By Lemma 5.1, we see that $`p^{}(L,h)`$ is numerically trivial on
$$V^{(1)}:=\text{the closure of }\phi ^1(\phi (p^1(\eta )),$$
since $`p^{}(L,h)`$ is numerically trivial on $`S_1`$ and every very general fiber of $`\phi :๐ฑ๐ฉ^0`$ by the definition of $`๐ฉ_0`$. Here we have used the fact that the numerical triviality is invariant under modifications, hence we may use the notion of numerical triviality on singular varieties by Theorem 4.2 (cf. Definition 4.2). Again by Theorem 4.2, we see that $`(L,h)`$ is numerically trivial on $`V_1`$. Since $`dimV_1>dimV`$ holds, this contradicts the definition of $`\nu `$. This completes the proof of Lemma 5.4. Q.E.D.
Let us continue the proof of Theorem 1.1. Let $`๐ฉ^{}`$ be another subvariety of the Hilbert scheme of $`X`$ whose members dominate $`X`$ and for a very general point $`x`$ of $`X`$ there exists a member $`V^{}`$ of $`๐ฉ^{}`$ such that
1. $`xV^{}`$,
2. $`dimV^{}=\nu `$,
3. $`(L,h)`$ is numerically trivial on $`V^{}`$.
Let
$$\phi ^{}:๐ฑ^{}๐ฉ^{}$$
be the universal family and let
$$p^{}:๐ฑ^{}X$$
be the natural morphism. Then we set
$$V_1^{}:=p^{}((\phi ^{})^1(\phi ^{}((p^{})^1(V))))$$
for a very general member $`V`$ of $`๐ฉ^0`$. Repeating the same argument as above we see that $`(L,h)`$ is numerically trivial on $`V_1^{}`$. Hence by the definition of $`\nu `$, we see that $`๐ฉ^{}=๐ฉ^0`$ holds.
Hence by Lemma 5.4, we see that for a very general point $`xX`$, there exists a unique member $`V`$ of $`๐ฉ(x)`$ such that
1. $`[V]๐ฉ^0`$,
2. $`(L,h)`$ is numerically trivial on $`V`$,
3. $`dimV=\nu `$.
Hence there exists a complement $`U_0`$ of at most countably many union of proper Zariski closed subsets in $`X`$ such that for every $`xU_0`$,
$$f(x)=[V]๐ฉ^0,xV$$
is a well defined morphism. Hence $`f`$ defines a rational fibration
$$f:X\mathrm{}Y$$
by setting
$$Y:=๐ฉ^0.$$
If we replace the second condition on $`V^{}`$,i.e. $`dimV^{}=\nu `$ by $`dimV^{}>0`$, by repeating the same argument, we see that $`V^{}`$ is contained in a member of $`๐ฉ^0`$. This implies the 3-rd assertion of Theorem 1.1. Lemma 5.1 implies the first assertion of Theorem 1.1.
By the construction this is the desired fibration. This completes the proof of Theorem 1.1. Q.E.D.
## 6 An algebraic counterpart of the fibration theorem
An algebraic counterpart of Theorem 1.1 would be the following theorem.
###### Theorem 6.1
Let $`X`$ be a normal projective variety and let $`L`$ be a nef line bundle on $`X`$. Then there exists a unique (up to birational equivalence) rational fibration
$$f:X\mathrm{}Y$$
such that
1. $`f`$ is regular over the generic point of $`Y`$,
2. $`L`$ is numerically trivial on every fibers of $`f`$,
3. $`dimY`$ is minimal among such fibrations.
The proof of the above theorem is essentially same as the proof of Theorem 1.1 and is much easier. Hence we omit it.
We should note that the fibrations given by Corollary 1.1 and Theorem 6.1 may not be same in general. I do not know how to generalize Theorem 6.1 to the case that $`L`$ is pseudoeffective.
Authorโs address
Hajime Tsuji
Department of Mathematics
Tokyo Institute of Technology
2-12-1 Ohokayama, Megro 152-8551
Japan
e-mail address: tsuji@math.titech.ac.jp |
warning/0001/hep-ph0001266.html | ar5iv | text | # ๐ตฬโ๐ท^(โ)โข๐โป beyond naive factorization in the heavy quark limit
## Acknowledgements
The author was supported in part by the Ministry of Education grants KRF-99-042-D00034 D2002, BSRI 98-2408, and the German-Korean scientific exchange program DFG-446-KOR-113/72/0. The author would like to thank Ahmed Ali, Christoph Greub, and Pyungwon Ko for stimulating discussions.
Figure caption
Figure 1: Feynman diagrams for nonfactorizable contribution. The dots denote the operator $`O_2`$. |
warning/0001/hep-ph0001086.html | ar5iv | text | # Contents
## 1 Introduction
ISAJET is a Monte Carlo program which simulates $`pp`$, $`\overline{p}p`$ and $`e^+e^{}`$ interactions at high energies. ISAJET is based on perturbative QCD plus phenomenological models for parton and beam jet fragmentation. Events are generated in four distinct steps:
* A primary hard scattering is generated according to the appropriate QCD cross section.
* QCD radiative corrections are added for both the initial and the final state.
* Partons are fragmented into hadrons independently, and particles with lifetimes less than about $`10^{12}`$ seconds are decayed.
* Beam jets are added assuming that these are identical to a minimum bias event at the remaining energy.
ISAJET incorporates ISASUSY, which evaluates branching ratios for the minimal supersymmetric extension of the standard model. H. Baer and X. Tata are coauthors of this package, and they have done the original calculations with various collaborators. See the ISASUSY documentation in the patch Section 12.
ISAJET is supported for ANSI Fortran and for Cray, DEC Ultrix, DEC VMS, HP/9000 7xx, IBM VM/CMS 370 and 30xx, IBM AIX RS/6000, Linux, Silicon Graphics 4D, and Sun computers. The CDC 7600 and ETA 10 versions are obsolete and are no longer supported. It is written mainly in ANSI standard FORTRAN 77, but it does contain some extensions except in the ANSI version. The code is maintained with a combination of RCS, the Revision Control System, and the Patchy code management system, which is part of the CERN Library. The original sources are kept on physgi01.phy.bnl.gov in `~isajet/isalibrary/RCS`; decks revised in release `n.nn` are kept in `~isajet/isalibrary/nnn`. ISAJET is supplied to BNL, CERN, Fermilab, and SLAC; it is also available by anonymous ftp from
```
ftp://penguin.phy.bnl.gov/pub/isajet
```
or by request from the authors.
Patch ISAPLT contains the skeleton of an HBOOK histogramming job, a trivial calorimeter simulation, and a jet-finding algorithm. (The default is HBOOK4; HBOOK3 can be selected with a Patchy switch.) These are provided for convenience only and are not supported.
## 2 Physics
ISAJET is a Monte Carlo program which simulates $`pp`$, $`\overline{p}p`$ and $`e^+e^{}`$ interactions at high energy. The program incorporates perturbative QCD cross sections, initial state and final state QCD radiative corrections in the leading log approximation, independent fragmentation of quarks and gluons into hadrons, and a phenomenological model tuned to minimum bias and hard scattering data for the beam jets.
### 2.1 Hard Scattering
The first step in simulating an event is to generate a primary hard scattering according to some QCD cross section. This has the general form
$$\sigma =\sigma _0F(x_1,Q^2)F(x_2,Q^2)$$
where $`\sigma _0`$ is a cross section calculated in QCD perturbation theory, $`F(x,Q^2)`$ is a structure function incorporating QCD scaling violations, $`x_1`$ and $`x_2`$ are the usual parton model momentum fractions, and $`Q^2`$ is an appropriate momentum transfer scale.
For each of the processes included in ISAJET, the basic cross section $`\sigma _0`$ is a two-body one, and the user can set limits on the kinematic variables and type for each of the two primary jets. For DRELLYAN and WPAIR events the full matrix element for the decay of the Wโs into leptons or quarks is also included.
The following processes are available:
#### 2.1.1 Minbias
No hard scattering at all, so that the event consists only of beam jets. Note that at high energy the jet cross sections become large. To represent the total cross section it is better to use a sample of TWOJET events with the lower limit on pt chosen to give a cross section equal to the inelastic cross section or to use a mixture of MINBIAS and TWOJET events.
#### 2.1.2 Twojet
All order $`\alpha _s^2`$ QCD processes, which give rise in lowest order to two high-$`p_t`$ jets. Included are, e.g.
$`g+g`$ $``$ $`g+g`$
$`g+q`$ $``$ $`g+q`$
$`g+g`$ $``$ $`q+\overline{q}`$
Masses are neglected for $`c`$ and lighter quarks but are taken into account for $`b`$ and $`t`$ quarks. The $`Q^2`$ scale is taken to be
$$Q^2=2stu/(s^2+t^2+u^2)$$
The default parton distributions are those of the CTEQ Collaboration, fit CTEQ3L, using lowest order QCD evolution. Two older fits, Eichten, Hinchliffe, Lane and Quigg (EHLQ), Set 1, and Duke and Owens, Set 1, are also included. There is also an interface to the CERN PDFLIB compilation of parton distributions. Note that structure functions for heavy quarks are included, so that processes like
$$g+tg+t$$
can be generated. The Duke-Owens parton distributions do not contain b or t quarks.
Since the $`t`$ is so heavy, it decays before it can hadronize, so instead of $`t`$ hadrons a $`t`$ quark appears in the particle list. It is decayed using the $`VA`$ matrix element including the $`W`$ propagator with a nonzero width, so the same decays should be used for $`m_t<m_W`$ and $`m_t>m_W`$; the $`W`$ should not be listed as part of the decay mode. The partons are then evolved and fragmented as usual; see below. The real or virtual $`W`$ and the final partons from the decay, including any radiated gluons, are listed in the particle table, followed by their fragmentation products. Note that for semileptonic decays the leptons appear twice: the lepton parton decays into a single particle of the same type but in general somewhat different momentum. In all cases only particles with $`\mathrm{๐ธ๐ณ๐ฒ๐ฐ๐}=0`$ should be included in the final state.
A fourth generation $`x,y`$ is also allowed. Fourth generation quarks are produced only by gluon fusion. Decay modes are not included in the decay table; for a sequential fourth generation they would be very similar to the t decays. In decays involving quarks, it is essential that the quarks appear last.
#### 2.1.3 Drellyan
Production of a $`W`$ in the standard model, including a virtual $`\gamma `$, a $`W^+`$, a $`W^{}`$, or a $`Z^0`$, and its decay into quarks or leptons. If the transverse momentum QTW of the $`W`$ is fixed equal to zero then the process simulated is
$`q+\overline{q}W`$ $``$ $`q+\overline{q}`$
$``$ $`\mathrm{}+\overline{\mathrm{}}`$
Thus the $`W`$ has zero transverse momentum until initial state QCD corrections are taken into account. If non-zero limits on the transverse momentum $`q_t`$ for the $`W`$ are set, then instead the processes
$`q+\overline{q}`$ $``$ $`W+g`$
$`g+q`$ $``$ $`W+q`$
are simulated, including the full matrix element for the $`W`$ decay. These are the dominant processes at high $`q_t`$, but they are of course singular at $`q_t=0`$. A cutoff of the $`1/q_t^2`$ singularity is made by the replacement
$$1/q_t^21/\sqrt{q_t^4+q_{t0}^4}q_{t0}^2=(.2\mathrm{GeV})M$$
This cutoff is chosen to reproduce approximately the $`q_t`$ dependence calculated by the summation of soft gluons and to give about the right integrated cross section. Thus this option can be used for low as well as high transverse momenta.
The scale for QCD evolution is taken to be proportional to the mass for lowest order Drell-Yan and to the transverse momentum for high-$`p_t`$ Drell-Yan. The constant is adjusted to get reasonable agreement with the $`W+n\mathrm{jet}`$ cross sections calculated from the full QCD matrix elements by F.A. Berends, et al., Phys. Lett. B224, 237 (1989).
For the processes $`g+bW+t`$ and $`g+tZ+t`$, cross sections with a non-zero top mass are used for the production and the $`W/Z`$ decay. These were calculated using FORM 1.1 by J. Vermaseren. The process $`g+tW+b`$ is not included. Both $`g+bW^{}+t`$ and $`g+\overline{t}W^{}+\overline{b}`$ of course give the same $`W^{}+t+\overline{b}`$ final state after QCD evolution. While the latter process is needed to describe the $`m_t=0`$(!) mass singularity for $`q_tm_t`$, it has a pole in the physical region at low $`q_t`$ from on-shell $`tW+b`$ decays. There is no obvious way to avoid this without introducing an arbitrary cutoff. Hence, selecting only $`W+b`$ will produce a zero cross section. The $`Q^2`$ scale for the parton distributions in these processes is replaced by $`Q^2+m_t^2`$; this seems physically sensible and prevents the cross sections from vanishing at small $`q_t`$.
#### 2.1.4 Photon
Single and double photon production through the lowest order QCD processes
$`g+q`$ $``$ $`\gamma +q`$
$`q+\overline{q}`$ $``$ $`\gamma +g`$
$`q+\overline{q}`$ $``$ $`\gamma +\gamma `$
Higher order corrections are not included. But $`\gamma `$โs, $`W`$โs, and $`Z`$โs are radiated from final state quarks in all processes, allowing study of the bremsstrahlung contributions.
#### 2.1.5 Wpair
Production of pairs of W bosons in the standard model through quark-antiquark annihilation,
$`q+\overline{q}`$ $``$ $`W^++W^{}`$
$``$ $`Z^0+Z^0`$
$``$ $`W^++Z^0,W^{}+Z^0`$
$``$ $`W^++\gamma ,W^{}+\gamma `$
The full matrix element for the W decays, calculated in the narrow resonance approximation, is included. However, the higher order processes, e.g.
$$q+qq+q+W^++W^{}$$
are ignored, although they in fact dominate at high enough mass. Specific decay modes can be selected using the WMODEi keywords.
#### 2.1.6 Higgs
Production and decay of the standard model Higgs boson. The production processes are
$`g+g`$ $``$ $`H\text{(through a quark loop)}`$
$`q+\overline{q}`$ $``$ $`H\text{(with }t+\overline{t}\text{ dominant)}`$
$`W^++W^{}`$ $``$ $`H\text{ (with longitudinally polarized }W\text{)}`$
$`Z^0+Z^0`$ $``$ $`H\text{ (with longitudinally polarized }Z\text{)}`$
If the (Standard Model) Higgs is lighter than $`2M_W`$, then it will decay into pairs of fermions with branching ratios proportional to $`m_f^2`$. If it is heavier than $`2M_W`$, then it will decay primarily into $`W^+W^{}`$ and $`Z^0Z^0`$ pairs with widths given approximately by
$`\mathrm{\Gamma }(HW^+W^{})`$ $`=`$ $`{\displaystyle \frac{G_FM_H^3}{8\pi \sqrt{2}}}`$
$`\mathrm{\Gamma }(HZ^0Z^0)`$ $`=`$ $`{\displaystyle \frac{G_FM_H^3}{16\pi \sqrt{2}}}`$
Numerically these give approximately
$$\mathrm{\Gamma }_H=0.5\mathrm{TeV}\left(\frac{M_H}{1\mathrm{TeV}}\right)^3$$
The width proportional to $`M_H^3`$ arises from decays into longitudinal gauge bosons, which like Higgs bosons have couplings proportional to mass.
Since a heavy Higgs is wide, the narrow resonance approximation is not valid. To obtain a cross section with good high energy behavior, it is necessary to include a complete gauge-invariant set of graphs for the processes
$`W^+W^{}`$ $``$ $`W^+W^{}`$
$`W^+W^{}`$ $``$ $`Z^0Z^0`$
$`Z^0Z^0`$ $``$ $`W^+W^{}`$
$`Z^0Z^0`$ $``$ $`Z^0Z^0`$
with longitudinally polarized $`W^+`$, $`W^{}`$, and $`Z^0`$ bosons in the initial state. This set of graphs and the corresponding angular distributions for the $`W^+`$, $`W^{}`$, and $`Z^0`$ decays have been calculated in the effective $`W`$ approximation and included in HIGGS. The $`W`$ structure functions are obtained by integrating the EHLQ parameterization of the quark ones term by term. The Cabibbo-allowed branchings
$`q`$ $``$ $`W^++q^{}`$
$`q`$ $``$ $`W^{}+q^{}`$
$`q`$ $``$ $`Z^0+q`$
are generated by backwards evolution, and the standard QCD evolution is performed. This correctly describes the $`W`$ collinear singularity and so contains the same physics as the effective $`W`$ approximation.
If the Higgs is lighter than $`2M_W`$, then its decay to $`\gamma \gamma `$ through $`W`$ and $`t`$ loops may be important. This is also included in the HIGGS process and may be selected by choosing `GM` as the jet type for the decay.
If the Higgs has $`M_Z<M_H<2M_Z`$, then decays into one real and one virtual $`Z^0`$ are generated if the `Z0 Z0` decay mode is selected, using the calculation of Keung and Marciano, Phys. Rev. D30, 248 (1984). Since the calculation assumes that one $`Z^0`$ is exactly on shell, it is not reliable within of order the $`Z^0`$ width of $`M_H=2M_Z`$; Higgs and and $`Z^0Z^0`$ masses in this region should be avoided. The analogous Higgs decays into one real and one virtual charged W are not included.
Note that while HIGGS contains the dominant graphs for Higgs production and graphs for $`W`$ pair production related by gauge invariance, it does not contain the processes
$`q+\overline{q}`$ $``$ $`W^+W^{}`$
$`q+\overline{q}`$ $``$ $`Z^0Z^0`$
which give primarily transverse gauge bosons. These must be generated with WPAIR.
If the `MSSMi` or `SUGRA` keywords are used with HIGGS, then one of the three MSSM neutral Higgs is generated instead using gluon-gluon and quark-antiquark fusion with the appropriate SUSY couplings. Since heavy CP even SUSY Higgs are weakly coupled to W pairs and CP odd ones are completely decoupled, $`WW`$ fusion and $`WWWW`$ scattering are not included in the SUSY case. ($`WWWW`$ can be generated using the Standard Model process with a light Higgs mass, say 100 GeV.) The MSSM Higgs decays into both Standard Model and SUSY modes as calculated by ISASUSY are included. For more discussion see the SUSY subsection below and the writeup for ISASUSY. The user must select which Higgs to generate using HTYPE; see Section 6 below. If a mass range is not specified, then the range mass $`M_H\pm 5\mathrm{\Gamma }_H`$ is used by default. (This cannot be done for the Standard Model Higgs because it is so wide for large masses.) Decay modes may be selected in the usual way.
#### 2.1.7 WHiggs
Generates associated production of gauge and Higgs bosons, i.e.,
$$q+\overline{q}H+W,H+Z,$$
in the narrow resonance approximation. The desired subprocesses can be selected with JETTYPEi, and specific decay modes of the $`W`$ and/or $`Z`$ can be selected using the WMODEi keywords. Standard Model couplings are assumed unless SUSY parameters are specified, in which case the SUSY couplings are used.
#### 2.1.8 SUSY
Generates pairs of supersymmetric particles from gluon-quark or quark-antiquark fusion. If the MSSMi or SUGRA parameters defined in Section 6 below are not specified, then only gluinos and squarks are generated:
$`g+g`$ $``$ $`\stackrel{~}{g}+\stackrel{~}{g}`$
$`q+\overline{q}`$ $``$ $`\stackrel{~}{g}+\stackrel{~}{g}`$
$`g+q`$ $``$ $`\stackrel{~}{g}+\stackrel{~}{q}`$
$`g+g`$ $``$ $`\stackrel{~}{q}+\stackrel{~}{\overline{q}}`$
$`q+\overline{q}`$ $``$ $`\stackrel{~}{q}+\stackrel{~}{\overline{q}}`$
$`q+q`$ $``$ $`\stackrel{~}{q}+\stackrel{~}{q}`$
Left and right squarks are distinguished but assumed to be degenerate. Masses can be specified using the `GAUGINO`, `SQUARK`, and `SLEPTON` parameters described in Section 6. No decay modes are specified, since these depend strongly on the masses. The user can either add new modes to the decay table (see Section 9) or use the `FORCE` or `FORCE1` commands (see Section 6).
If `MSSMA`, `MSSMB`, and `MSSMC` are specified, then the ISASUSY package is used to calculate the masses and decay modes in the minimal supersymmetric extension of the standard model (MSSM), assuming SUSY grand unification constraints in the neutralino and chargino mass matrix but allowing some additional flexibility in the masses. The scalar particle soft masses are input via `MSSMi`, so that the physical masses will be somewhat different due to $`D`$-term contributions and mixings for 3rd generation sparticles. $`\stackrel{~}{t}_1`$ and $`\stackrel{~}{t}_2`$ production and decays are now included. The lightest SUSY particle is assumed to be the lightest neutralino $`\stackrel{~}{Z}_1`$. If the `MSSMi` parameters are specified, then the following additional processes are included using the MSSM couplings for the production cross sections:
$`g+q`$ $``$ $`\stackrel{~}{Z}_i+\stackrel{~}{q},\stackrel{~}{W}_i+\stackrel{~}{q}`$
$`q+\overline{q}`$ $``$ $`\stackrel{~}{Z}_i+\stackrel{~}{g},\stackrel{~}{W}_i+\stackrel{~}{g}`$
$`q+\overline{q}`$ $``$ $`\stackrel{~}{W}_i+\stackrel{~}{Z}_j`$
$`q+\overline{q}`$ $``$ $`\stackrel{~}{W}_i^++\stackrel{~}{W}_j^{}`$
$`q+\overline{q}`$ $``$ $`\stackrel{~}{Z}_i+\stackrel{~}{Z}_j`$
$`q+\overline{q}`$ $``$ $`\stackrel{~}{\mathrm{}}^++\stackrel{~}{\mathrm{}}^{},\stackrel{~}{\nu }+\stackrel{~}{\nu }`$
Processes can be selected using the optional parameters described in Section 6 below.
Beginning with Version 7.42, matrix elements are taken into account in the event generator as well as in the calculation of decay widths for MSSM three-body decays of the form $`\stackrel{~}{A}\stackrel{~}{B}f\overline{f}`$, where $`\stackrel{~}{A}`$ and $`\stackrel{~}{B}`$ are gluinos, charginos, or neutralinos. This is implemented by having ISASUSY save the poles and their couplings when calculating the decay width and then using these to reconstruct the matrix element. Other three-body decays may be included in the future. Decays selected with `FORCE` use the appropriate matrix elements.
An optional keyword `MSSMD` can be used to specify the second generation masses, which otherwise are assumed degenerate with the first generation. An optional keyword `MSSME` can be used to specify values of the $`U(1)`$ and $`SU(2)`$ gaugino masses at the weak scale rather than using the default grand unification values. The chargino and neutralino masses and mixings are then computed using these values.
Instead of using the `MSSMi` parameters, one can use the `SUGRA` parameter to specify in the minimal supergravity framework. This assumes that the gauge couplings unify at a GUT scale and that SUSY breaking occurs at that scale with universal soft breaking terms, which are related to the weak scale using the renormalization group. The parameters of the model are
* $`m_0`$: the common scalar mass at the GUT scale;
* $`m_{1/2}`$: the common gaugino mass at the GUT scale;
* $`A_0`$: the common soft trilinear SUSY breaking parameter at the GUT scale;
* $`\mathrm{tan}\beta `$: the ratio of Higgs vacuum expectation values at the electroweak scale;
* $`sgn\mu =\pm 1`$: the sign of the Higgsino mass term.
The renormalization group equations are solved iteratively to determine all the electroweak SUSY parameters from these data assuming radiative electroweak symmetry breaking but not other possible constraints such as b-tau unification or limits on proton decay.
The assumption of universality at the GUT scale is rather restrictive and may not be valid. A variety of non-universal SUGRA (NUSUGRA) models can be generated using the `NUSUG1`, โฆ, `NUSUG5` keywords. These might be used to study how well one could test the minimal SUGRA model.
An alternative to the SUGRA model is the Gauge Mediated SUSY Breaking (GMSB) model of Dine, Nelson, and collaborators. In this model SUSY breaking is communicated through gauge interactions with messenger fields at a scale $`M_m`$ small compared to the Planck scale and are proportional to gauge couplings times $`\mathrm{\Lambda }_m`$. The messenger fields should form complete $`SU(5)`$ representations to preserve the unification of the coupling constants. The parameters of the GMSB model, which are specified by the `GMSB` keyword, are
* $`\mathrm{\Lambda }_m=F_m/M_m`$: the scale of SUSY breaking, typically 10โ$`100\mathrm{TeV}`$;
* $`M_m>\mathrm{\Lambda }_m`$: the messenger mass scale;
* $`N_5`$: the equivalent number of $`5+\overline{5}`$ messenger fields.
* $`\mathrm{tan}\beta `$: the ratio of Higgs vacuum expectation values at the electroweak scale;
* $`sgn\mu =\pm 1`$: the sign of the Higgsino mass term;
* $`C_{\mathrm{grav}}1`$: the ratio of the gravitino mass to the value it would have had if the only SUSY breaking scale were $`F_m`$.
In GMSB models the lightest SUSY particle is always the nearly massless gravitino $`\stackrel{~}{G}`$. The parameter $`C_{\mathrm{grav}}`$ scales the gravitino mass and hence the lifetime of the next lightest SUSY particle to decay into it. The `NOGRAV` keyword can be used to turn off gravitino decays.
A variety of non-minimal GMSB models can be generated using additional parameters set with the GMSB2 keyword. These additional parameters are
* $`\text{ / }R`$, an extra factor multiplying the gaugino masses at the messenger scale. (Models with multiple spurions generally have $`\text{ / }R<1`$.)
* $`\delta M_{H_d}^2`$, $`\delta M_{H_u}^2`$, Higgs mass-squared shifts relative to the minimal model at the messenger scale. (These might be expected in models which generate $`\mu `$ realistically.)
* $`D_Y(M)`$, a $`U(1)_Y`$ messenger scale mass-squared term ($`D`$-term) proportional to the hypercharge $`Y`$.
* $`N_{5_1}`$, $`N_{5_2}`$, and $`N_{5_3}`$, independent numbers of gauge group messengers. They can be non-integer in general.
For discussions of these additional parameters, see S. Dimopoulos, S. Thomas, and J.D. Wells, hep-ph/9609434, Nucl. Phys. B488, 39 (1997), and S.P. Martin, hep-ph/9608224, Phys. Rev. D55, 3177 (1997).
Gravitino decays can be included in the general MSSM framework by specifying a gravitino mass with `MGVTNO`. The default is that such decays do not occur.
Another alternative SUSY model choice allowed is anomaly-mediated SUSY breaking, developed by Randall and Sundrum. In this model, it is assumed that SUSY breaking takes place in other dimensions, and SUSY breaking is communicated to the visible sector via the superconformal anomaly. In this model, the lightest SUSY particle is usually the neutralino which is nearly pure wino-like. The chargino is nearly mass degenerate with the lightest neutralino. It can be very long lived, or decay into a very soft pion plus missing energy. The model incorporated in ISAJET, based on work by Ghergetta, Giudice and Wells (hep-ph/9904378), and by Feng and Moroi (hep-ph/9907319) adds a universal contribution $`m_0^2`$ to all scalar masses to avoid problems with tachyonic scalars. The parameter set is $`m_0,m_{3/2},\mathrm{tan}\beta ,sign(\mu )`$, and can be input via the $`AMSB`$ keyword. Care should be taken with the chargino decay, since it may have macroscopic decay lengths, or even decay outside the detector.
Since neutrinos seem to have mass, the effect of a massive right-handed neutrino has been included in ISAJET, when calculating the sparticle mass spectrum. If the keyword $`SUGRHN`$ is used, then the user must input the 3rd generation neutrino mass (at scale $`M_Z`$) in units of GeV, and the intermediate scale right handed neutrino Majorana mass $`M_N`$, also in GeV. In addition, one must specify the soft SUSY-breaking masses $`A_n`$ and $`m_{\stackrel{~}{\nu }_R}`$ valid at the GUT scale. Then the neutrino Yukawa coupling is computed in the simple see-saw model, and renormalization group evolution includes these effects between $`M_{GUT}`$ and $`M_N`$. Finally, to facilitate modeling of $`SO(10)`$ SUSY-GUT models, loop corrections to 3rd generation fermion masses have been included in the ISAJET SUSY models.
The ISASUSY program can also be used independently of the rest of ISAJET, either to produce a listing of decays or in conjunction with another event generator. Its physics assumptions are described in more detail in Section 12. The ISASUGRA program can also be used independently to solve the renormalization group equations with SUGRA, GMSB, or NUSUGRA boundary conditions and then to call ISASUSY to calculate the decay modes.
Generally the MSSM, SUGRA, or GMSB option should be used to study supersymmetry signatures; the SUGRA or GMSB parameter space is clearly more manageable. The more general option may be useful to study alternative SUSY models. It can also be used, e.g., to generate pointlike color-3 leptoquarks in technicolor models by selecting squark production and setting the gluino mass to be very large. The MSSM or SUGRA option may also be used with top pair production to simulate top decays to SUSY particles.
#### 2.1.9 $`e^+e^{}`$
An $`e^+e^{}`$ event generator is also included in ISAJET. The Standard Model processes included are $`e^+e^{}`$ annihilation through $`\gamma `$ and $`Z`$ to quarks and leptons, and production of $`W^+W^{}`$ and $`Z^0Z^0`$ pairs. In contrast to WPAIR and HIGGS for the hadronic processes, the produced $`W`$โs and $`Z`$โs are treated as particles, so their spins are not properly taken into account in their decays. (Because the $`W`$โs and $`Z`$โs are treated as particles, their decay modes can be selected using `FORCE` or `FORCE1`, not `WMODEi`. See Section below.) Other Standard Model processes, including $`e^+e^{}e^+e^{}`$ ($`t`$-channel graph) and $`e^+e^{}\gamma \gamma `$, are not included. Once the primary reaction has been generated, QCD radiation and hadronization are done as for hadronic processes.
The $`e^+e^{}`$ generator can be run assuming no initial state radiation (the default), or an initial state electron structure function can be used for bremsstrahlung or the combination bremsstrahlung/beamstrahlung effect. Bremsstrahlung is implemented using the Fadin-Kuraev $`e^{}`$ distribution function, and can be turned on using the `EEBREM` command while stipulating the minimal and maximal subprocess energy. Beamstrahlung is implemented by invoking the `EEBEAM` keyword. In this case, in addition the beamstrahlung parameter $`\mathrm{{\rm Y}}`$ and longitudinal beam size $`\sigma _z`$ (in mm) must be given. The definition for $`\mathrm{{\rm Y}}`$ in terms of other beam parameters can be found in the article Phys. Rev. D49, 3209 (1994) by Chen, Barklow and Peskin. The bremsstrahlung structure function is then convoluted with the beamstrahlung distribution (as calculated by P. Chen) and a spline fit is created. Since the cross section can contain large spikes, event generation can be slow if a huge range of subprocess energy is selected for light particles; in these scenarios, `NTRIES` must be increased well beyond the default value.
$`e^+e^{}`$ annihilation to SUSY particles is included as well with complete lowest order diagrams, and cascade decays. The processes include
$`e^+e^{}`$ $``$ $`\stackrel{~}{q}\stackrel{~}{q}`$
$`e^+e^{}`$ $``$ $`\stackrel{~}{\mathrm{}}\stackrel{~}{\mathrm{}}`$
$`e^+e^{}`$ $``$ $`\stackrel{~}{W}_i\stackrel{~}{W}_j`$
$`e^+e^{}`$ $``$ $`\stackrel{~}{Z}_i\stackrel{~}{Z}_j`$
$`e^+e^{}`$ $``$ $`H_L^0+Z^0,H_H^0+Z^0,H_A^0+H_L^0,H_A^0+H_H^0,H^++H^{}`$
Note that SUSY Higgs production via $`WW`$ and $`ZZ`$ fusion, which can dominate Higgs production processes at $`\sqrt{s}>500\mathrm{GeV}`$, is not included. Spin correlations are neglected, although 3-body sparticle decay matrix elements are included.
$`e^+e^{}`$ cross sections with polarized beams are included for both Standard Model and SUSY processes. The keyword `EPOL` is used to set $`P_L(e^{})`$ and $`P_L(e^+)`$, where
$$P_L(e)=(n_Ln_R)/(n_L+n_R)$$
so that $`1P_L+1`$. Thus, setting `EPOL` to $`.9,0`$ will yield a 95% right polarized electron beam scattering on an unpolarized positron beam.
#### 2.1.10 Technicolor
Production of a technirho of arbitrary mass and width decaying into $`W^\pm Z^0`$ or $`W^+W^{}`$ pairs. The cross section is based on an elastic resonance in the $`WW`$ cross section with the effective $`W`$ approximation plus a $`W`$ mixing term taken from EHLQ. Additional technicolor processes may be added in the future.
#### 2.1.11 Extra Dimensions
The possibility that there might be more than four space-time dimensions at a distance scale $`R`$ much larger than $`G_N^{1/2}`$ has recently attracted interest. In these theories,
$$G_N=\frac{1}{8\pi R^\delta M_D^{2+\delta }},$$
where $`\delta `$ is the number of extra dimensions and $`M_D`$ is the $`4+\delta `$ Planck scale. Gravity deviates from the standard theory at a distance $`R10^{22/\delta 19}\mathrm{m}`$, so $`\delta 2`$ is required. If $`M_D`$ is of order $`1\mathrm{TeV}`$, then the usual heirarchy problem is solved, although there is then a new heirarchy problem of why $`R`$ is so large.
In such models the graviton will have many Kaluza-Klein excitations with a mass splitting of order $`1/R`$. While any individual mode is suppressed by the four-dimensional Planck mass, the large number of modes produces a cross section suppressed only by $`1/M_D^2`$. The signature is an invisible massive graviton plus a jet, photon, or other Standard Model particle. The `EXTRADIM` process implements this reaction using the cross sections of Giudice, Rattazzi, and Wells, hep-ph/9811291. The number $`\delta `$ of extra dimensions, the mass scale $`M_D`$, and the logical flag `UVCUT` are specified using the keyword `EXTRAD`. If `UVCUT` is `TRUE`, the cross section is cut off above the scale $`M_D`$; the model is not valid if the results depend on this flag.
### 2.2 Multiparton Hard Scattering
All the processes listed in Section 2.1 are either $`22`$ processes like `TWOJET` or $`21`$ $`s`$-channel resonance processes followed by a 2-body decay like `DRELLYAN`. The QCD parton shower described in Section 2.3 below generates multi-parton final states starting from these, but it relies on an approximation which is valid only if the additional partons are collinear either with the initial or with the final primary ones. Since the QCD shower uses exact non-colliear kinematics, it in fact works pretty well in a larger region of phase space, but it is not exact.
Non-collinear multiparton final states are interesting both in their own right and as backgrounds for other signatures. Both the matrix elements and the phase space for multiparton processes are complicated; they have been incorporated into ISAJET for the first time in Version 7.45. To calculate the matrix elements we have used the MadGraph package by Stelzer and Long, Comput. Phys. Commun. 81, 357 (1994), hep-ph/9401258. This automatically generates the amplitude using `HELAS`, a formalism by Murayama, Watanabe, and Hagiwarak KEK-91-11, that calculates the amplitude for any Feynman diagram in terms of spinnors, vertices, and propagators. The MadGraph code has been edited to incorporate summations over quark flavors. To do the phase space integration, we have used a simple recursive algorithm to generate $`n`$-body phase space. We have included limits on the total mass of the final state using the `MTOT` keyword. Limits on the $`p_T`$ and rapidity of each final parton can be set via the `PT` and `Y` keyworks, while limits on the mass of any pair of final partons can be set via the `MIJTOT` keyword. These limits are sufficient to shield the infrared and collinear singularities and to render the result finite. However, the parton shower populates all regions of phase space, so careful thought is needed to combine the parton-shower based and multiparton based results.
While the multiparton formalism is rather general, it still takes a substantial amount of effort to implement any particular process. So far only one process has been implemented.
#### 2.2.1 $`Z+2\mathrm{jets}`$
The `ZJJ` process generates a $`Z`$ boson plus two jets, including the $`q\overline{q}Zq\overline{q}`$, $`ggZq\overline{q}`$, $`q\overline{q}Zgg`$, $`qqZqq`$, and $`gqZgq`$ processes. The $`Z`$ is defined to be jet 1; it is treated in the narrow resonance approximation and is decayed isotropically. The quarks, antiquarks, and gluons are defined to be jets 2 and 3 and are symmetrized in the usual way.
### 2.3 QCD Radiative Corrections
After the primary hard scattering is generated, QCD radiative corrections are added to allow the possibility of many jets. This is essential to get the correct event structure, especially at high energy.
Consider the emission of one extra gluon from an initial or a final quark line,
$$q(p)q(p_1)+g(p_2)$$
From QCD perturbation theory, for small $`p^2`$ the cross section is given by the lowest order cross section multiplied by a factor
$$\sigma =\sigma _0\alpha _s(p^2)/(2\pi p^2)P(z)$$
where $`z=p_1/p`$ and $`P(z)`$ is an Altarelli-Parisi function. The same form holds for the other allowed branchings,
$`g(p)`$ $``$ $`g(p_1)+g(p_2)`$
$`g(p)`$ $``$ $`q(p_1)+\overline{q}(p_2)`$
These factors represent the collinear singularities of perturbation theory, and they produce the leading log QCD scaling violations for the structure functions and the jet fragmentation functions. They also determine the shape of a QCD jet, since the jet $`M^2`$ is of order $`\alpha _sp_t^2`$ and hence small.
The branching approximation consists of keeping just these factors which dominate in the collinear limit but using exact, non-collinear kinematics. Thus higher order QCD is reduced to a classical cascade process, which is easy to implement in a Monte Carlo program. To avoid infrared and collinear singularities, each parton in the cascade is required to have a mass (spacelike or timelike) greater than some cutoff $`t_c`$. The assumption is that all physics at lower scales is incorporated in the nonperturbative model for hadronization. In ISAJET the cutoff is taken to be a rather large value, $`(6\mathrm{GeV})^2`$, because independent fragmentation is used for the jet fragmentation; a low cutoff would give too many hadrons from overlapping partons. It turns out that the branching approximation not only incorporates the correct scaling violations and jet structure but also reproduces the exact three-jet cross section within factors of order 2 over all of phase space.
This approximation was introduced for final state radiation by Fox and Wolfram. The QCD cascade is determined by the probability for going from mass $`t_0`$ to mass $`t_1`$ emitting no resolvable radiation. For a resolution cutoff $`z_c<z<1z_c`$, this is given by a simple expression,
$$P(t_0,t_1)=\left(\alpha _s(t_0)/\alpha _s(t_1)\right)^{2\gamma (z_c)/b_0}$$
where
$$\gamma (z_c)=_{z_c}^{1z_c}๐zP(z),b_0=(332n_f)/(12\pi )$$
Clearly if $`P(t_0,t_1)`$ is the integral probability, then $`dP/dt_1`$ is the probability for the first radiation to occur at $`t_1`$. It is straightforward to generate this distribution and then iteratively to correct it to get a cutoff at fixed $`t_c`$ rather than at fixed $`z_c`$.
For the initial state it is necessary to take account of the spacelike kinematics and of the structure functions. Sjostrand has shown how to do this by starting at the hard scattering and evolving backwards, forcing the ordering of the spacelike masses $`t`$. The probability that a given step does not radiate can be derived from the Altarelli-Parisi equations for the structure functions. It has a form somewhat similar to $`P(t_0,t_1)`$ but involving a ratio of the structure functions for the new and old partons. It is possible to find a bound for this ratio in each case and so to generate a new $`t`$ and $`z`$ as for the final state. Then branchings for which the ratio is small are rejected in the usual Monte Carlo fashion. This ratio suppresses the radiation of very energetic partons. It also forces the branching $`gt+\overline{t}`$ for a $`t`$ quark if the $`t`$ structure function vanishes at small momentum transfer.
At low energies, the branching of an initial heavy quark into a gluon sometimes fails; these events are discarded and a warning is printed.
Since $`t_c`$ is quite large, the radiation of soft gluons is cut off. To compensate for this, equal and opposite transverse boosts are made to the jet system and to the beam jets after fragmentation with a mean value
$$p_t^2=(.1\mathrm{GeV})\sqrt{Q^2}$$
The dependence on $`Q^2`$ is the same as the cutoff used for DRELLYAN and the coefficient is adjusted to fit the $`p_t`$ distribution for the $`W`$.
Radiation of gluons from gluinos and scalar quarks is also included in the same approximation, but the production of gluino or scalar quark pairs from gluons is ignored. Very little radiation is expected for heavy particles produced near threshold.
Radiation of photons, $`W`$โs, and $`Z`$โs from final state quarks is treated in the same approximation as QCD radiation except that the coupling constant is fixed. Initial state electroweak radiation is not included; it seems rather unimportant. The $`W^+`$โs, $`W^{}`$โs and $`Z`$โs are decayed into the modes allowed by the `WPMODE`, `WMMODE`, and `Z0MODE` commands respectively. Warning: The branching ratios implied by these commands are not included in the cross section because an arbitrary number of $`W`$โs and $`Z`$โs can in principle be radiated.
### 2.4 Jet Fragmentation:
Quarks and gluons are fragmented into hadrons using the independent fragmentation ansatz of Field and Feynman. For a quark $`q`$, a new quark-antiquark pair $`q_1\overline{q}_1`$ is generated with
$$u:d:s=.43:.43:.14$$
A meson $`q\overline{q}_1`$ is formed carrying a fraction $`z`$ of the momentum,
$$E^{}+p_z^{}=z(E+p_z)$$
and having a transverse momentum $`p_t`$ with $`p_t=0.35\mathrm{GeV}`$. Baryons are included by generating a diquark with probability 0.10 instead of a quark; adjacent diquarks are not allowed, so no exotic mesons are formed. For light quarks $`z`$ is generated with the splitting function
$$f(z)=1a+a(b+1)(1z)^b,a=0.96,b=3$$
while for heavy quarks the Peterson form
$$f(z)=x(1x)^2/((1x)^2+ฯตx)^2$$
is used with $`ฯต=.80/m_c^2`$ for $`c`$ and $`ฯต=.50/m_q^2`$ for $`q=b,t,y,x`$. These values of $`ฯต`$ have been determined by fitting PEP, PETRA, and LEP data with ISAJET and should not be compared with values from other fits. Hadrons with longitudinal momentum less than zero are discarded. The procedure is then iterated for the new quark $`q_1`$ until all the momentum is used. A gluon is fragmented like a randomly selected $`u`$, $`d`$, or $`s`$ quark or antiquark.
In the fragmentation of gluinos and scalar quarks, supersymmetric hadrons are not distinguished from partons. This should not matter except possibly for very light masses. The Peterson form for $`f(x)`$ is used with the same value of epsilon as for heavy quarks, $`ฯต=0.5/m^2`$.
Independent fragmentation correctly describes the fast hadrons in a jet, but it fails to conserve energy or flavor exactly. Energy conservation is imposed after the event is generated by boosting the hadrons to the appropriate rest frame, rescaling all of the three-momenta, and recalculating the energies.
### 2.5 Beam Jets
There is now experimental evidence that beam jets are different in minimum bias events and in hard scattering events. ISAJET therefore uses similar a algorithm but different parameters in the two cases.
The standard models of particle production are based on pulling pairs of particles out of the vacuum by the QCD confining field, leading naturally to only short-range rapidity correlations and to essentially Poisson multiplicity fluctuations. The minimum bias data exhibit KNO scaling and long-range correlations. A natural explanation of this was given by the model of Abramovskii, Kanchelli and Gribov. In their model the basic amplitude is a single cut Pomeron with Poisson fluctuations around an average multiplicity $`n`$, but unitarity then produces graphs giving $`K`$ cut Pomerons with multiplicity $`Kn`$.
A simplified version of the AKG model is used in ISAJET. The number of cut Pomerons is chosen with a distribution adjusted to fit the data. For a minimum bias event this distribution is
$$P(K)=(1+4K^2)\mathrm{exp}1.8K$$
while for hard scattering
$$P(1)0.1P(1),P(2)0.2P(2),P(3)0.5P(3)$$
For each side of each event an $`x_0`$ for the leading baryon is selected with a distribution varying from flat for $`K=1`$ to like that for mesons for large K:
$$f(x)=N(K)(1x_0)^c(K),c(K)=1/K+(11/K)b(s)$$
The $`x_i`$ for the cut Pomerons are generated uniformly and then rescaled to $`1x_0`$. Each cut Pomeron is then hadronized in its own center of mass using a modified independent fragmentation model with an energy dependent splitting function to reproduce the rise in $`dN/dy`$:
$$f(x)=1a+a(b(s)+1)^b(s),b(s)=b_0+b_1\mathrm{log}(s)$$
The energy dependence is put into $`f(x)`$ rather than $`P(K)`$ because in the AKG scheme the single particle distribution comes only from the single chain. The probabilities for different flavors are taken to be
$$u:d:s=.46:.46:.08$$
to reproduce the experimental $`K/\pi `$ ratio.
## 3 Sample Jobs
The simplest ISAJET job reads a user-supplied parameter file and writes a data file and a listing file. The following is an example of a parameter file which generates each type of event:
```
SAMPLE TWOJET JOB
800.,100,2,50/
TWOJET
PT
50,100,50,100/
JETTYPE1
โGLโ/
JETTYPE2
โUPโ,โUBโ,โDNโ,โDBโ,โSTโ,โSBโ/
END
SAMPLE DRELLYAN JOB
800.,100,2,50/
DRELLYAN
QMW
80,100/
WTYPE
โW+โ,โW-โ/
END
SAMPLE MINBIAS JOB
800.,100,2,50/
MINBIAS
END
SAMPLE WPAIR JOB
800.,100,2,50/
WPAIR
PT
50,100,50,100/
JETTYPE1
โW+โ,โW-โ,โZ0โ/
JETTYPE2
โW+โ,โW-โ,โZ0โ/
WMODE1
โE+โ,โE-โ,โNUSโ/
WMODE2
โQUARKSโ/
END
SAMPLE HIGGS JOB FOR SSC
40000,100,1,1/
HIGGS
QMH
400,1600/
HMASS
800/
JETTYPE1
โZ0โ/
JETTYPE2
โZ0โ/
WMODE1
โMU+โ,โMU-โ/
WMODE2
โE+โ,โE-โ/
PT
50,20000,50,20000/
END
SAMPLE SUSY JOB
1800,100,1,10/
SUPERSYM
PT
50,100,50,100/
JETTYPE1
โGLSSโ,โSQUARKSโ/
JETTYPE2
โGLSSโ,โSQUARKSโ/
GAUGINO
60,1,40,40/
SQUARK
80.3,80.3,80.5,81.6,85,110/
FORCE
29,30,1,-1/
FORCE
21,29,1/
FORCE
22,29,2/
FORCE
23,29,3/
FORCE
24,29,4/
FORCE
25,29,5/
FORCE
26,29,6/
END
SAMPLE MSSM JOB FOR TEVATRON
1800.,100,1,1/
SUPERSYM
BEAMS
โPโ,โAPโ/
MSSMA
200,-200,500,2/
MSSMB
200,200,200,200,200/
MSSMC
200,200,200,200,200,0,0,0/
JETTYPE1
โGLSSโ/
JETTYPE2
โSQUARKSโ/
PT
100,300,100,300/
END
SAMPLE MSSM SUGRA JOB FOR LHC
14000,100,1,10/
SUPERSYM
PT
50,500,50,500/
SUGRA
247,302,-617.5,10,-1/
TMASS
175/
END
SAMPLE SUGRA HIGGS JOB USING DEFAULT QMH RANGE
14000,100,20,50/
HIGGS
SUGRA
200,200,0,2,+1/
HTYPE
โHA0โ/
JETTYPE1
โGAUGINOSโ,โSLEPTONSโ/
JETTYPE2
โGAUGINOSโ,โSLEPTONSโ/
END
SAMPLE E+E- TO SUGRA JOB WITH POLARIZED BEAMS AND BREM/BEAMSTRAHLUNG
500.,100,1,1/
E+E-
SUGRA
125,125,0,3,1/
TMASS
175,-1,1/
EPOL
-.9,0./
EEBEAM
200.,500.,.1072,.12/
JETTYPE1
โALLโ/
JETTYPE2
โALLโ/
NTRIES
10000/
END
SAMPLE WH JOB
2000,100,0,0/
WHIGGS
BEAMS
โPโ,โAPโ/
HMASS
100./
JETTYPE1
โW+โ,โW-โ,โHIGGSโ/
JETTYPE2
โW+โ,โW-โ,โHIGGSโ/
WMODE1
โALLโ/
WMODE2
โALLโ/
PT
10,300,10,300/
END
SAMPLE EXTRA DIMENSIONS JOB
14000,100,1,100/
EXTRADIM
QMW
5,1000/
QTW
500,1000/
EXTRAD
2,1000,.FALSE./
END
SAMPLE ZJJ JOB AT LHC
14000,100,1,100/
ZJJ
PT
20,7000,20,7000,20,7000/
MIJLIM
0,0,20,7000/
MTOT
100,500/
NSIGMA
200/
NTRIES
10000/
END
STOP
```
See Section 6 of this manual for a complete list of the possible commands in a parameter file. Note that all input to ISAJET must be in UPPER case only.
Subroutine RDTAPE is supplied to read events from an ISAJET data file, which is a machine-dependent binary file. It restores the event data to the FORTRAN common blocks described in Section 7. The skeleton of an analysis job using HBOOK and PAW from the CERN Program Library is provided in patch ISAPLT but is not otherwise supported. A Zebra output format based on code from the D0 Collaboration is also provided in patch ISAZEB; see the separate documentation in patch ISZTEXT.
### 3.1 DEC VMS
On a VAX or ALPHA running VMS, ISAJET can be compiled by executing the .COM file contained in P=ISAUTIL,D=MAKEVAX. Extract this deck as ISAMAKE.COM and type
```
@ISAMAKE
```
This will run YPATCHY with the pilot patches described in Section 4 and the VAX flag to extract the source code, decay table, and documentation. The source code is compiled and made into a library, which is linked with the following main program,
```
PROGRAM ISARUN
C MAIN PROGRAM FOR ISAJET
OPEN(UNIT=1,STATUS=โOLDโ,FORM=โFORMATTEDโ,READONLY)
OPEN(UNIT=2,STATUS=โNEWโ,FORM=โUNFORMATTEDโ)
OPEN(UNIT=3,STATUS=โOLDโ,FORM=โFORMATTEDโ)
OPEN(UNIT=4,STATUS=โNEWโ,FORM=โFORMATTEDโ)
CALL ISAJET(-1,2,3,4)
STOP
END
```
to produce ISAJET.EXE. Two other executables, ISASUSY.EXE and ISASUGRA.EXE, will also be produced to calculate SUSY masses and decay modes without generating events. Temporary files can be removed by typing
```
@ISAMAKE CLEAN
```
Create an input file `JOBNAME.PAR` following the examples above or the instructions in Section 6 and run ISAJET with the command
```
@ISAJET JOBNAME
```
using the ISAJET.COM file contained P=ISAUTIL,D=RUNVAX. This will create a binary output file `JOBNAME.DAT` and a listing file `JOBNAME.LIS`. Analyze the output data using the commands described in Section 8.
There is also an simple interactive interface to ISAJET which will prompt the user for commands, write a parameter file, and optionally execute it.
### 3.2 IBM VM/CMS
On an IBM mainframe running VM/CMS, run YPATCHY with the pilot patches described in Section 4 and the IBM flag to extract the source code, decay table, and documentation. Compile the source code and link it with the main program
```
PROGRAM ISARUN
C MAIN PROGRAM FOR ISAJET
OPEN(UNIT=1,STATUS=โOLDโ,FORM=โFORMATTEDโ)
OPEN(UNIT=2,STATUS=โNEWโ,FORM=โUNFORMATTEDโ)
OPEN(UNIT=3,STATUS=โOLDโ,FORM=โFORMATTEDโ)
OPEN(UNIT=4,STATUS=โNEWโ,FORM=โFORMATTEDโ)
CALL ISAJET(-1,2,3,4)
STOP
END
```
to make ISAJET MODULE.
Create a file called `JOBNAME INPUT` containing ISAJET input commands following the examples above or the instructions in Section 6. Then run ISAJET using ISAJET EXEC, which is contained in P=ISAUTIL,D=RUNIBM. The events will be produced on `JOBNAME DATA A` and the listing on `JOBNAME OUTPUT A`.
### 3.3 Unix
The Makefile contained in P=ISAUTIL,D=MAKEUNIX has been tested on DEC Ultrix, Hewlett Packard HP-UX, IBM RS/6000 AIX, Linux, Silicon Graphics IRIX, Sun SunOS, and Sun Solaris. It should work with minor modifications on almost any Unix system with /bin/csh, `ypatchy` or `nypatchy`, and a reasonable Fortran 77 compiler. Extract the Makefile and edit it, changing the installation parameters to reflect your system. Note in particular that CERNlib is usually compiled with underscores postpended to all external names; you must choose the appropriate compiler option if you intend to link with it. Then type
```
make
```
This should produce an executable `isajet.x` for the event generator, which links the code with the following main program:
```
PROGRAM RUNJET
CHARACTER*60 FNAME
READ 1000, FNAME
1000 FORMAT(A)
PRINT 1020, FNAME
1020 FORMAT(1X,โData file = โ,A)
OPEN(2,FILE=FNAME,STATUS=โNEWโ,FORM=โUNFORMATTEDโ)
READ 1000, FNAME
PRINT 1030, FNAME
1030 FORMAT(1X,โParameter file = โ,A)
OPEN(3,FILE=FNAME,STATUS=โOLDโ,FORM=โFORMATTEDโ)
READ 1000, FNAME
PRINT 1040, FNAME
1040 FORMAT(1X,โListing file = โ,A)
OPEN(4,FILE=FNAME,STATUS=โNEWโ,FORM=โFORMATTEDโ)
READ 1000, FNAME
OPEN(1,FILE=FNAME,STATUS=โOLDโ,FORM=โFORMATTEDโ)
CALL ISAJET(-1,2,3,4)
STOP
END
```
Two other executables, `isasusy.x` and `isasugra.x`, will also be produced to calculate SUSY masses and decay modes without generating events. Type
```
make clean
```
to delete the temporary files.
Most Unix systems do not allow two jobs to read the same decay table file at the same time. The shell script in P=ISAUTIL,D=RUNUNIX copies the decay table to a temporary file to avoid this problem. Extract this file as `isajet`. Create an input file `jobname.par` following the examples above or the instructions in Section 6 and run ISAJET with the command
```
isajet jobname
```
This will create a binary output file `jobname.dat` and a listing file `jobname.lis`. Analyze the output data using the commands described in Section 8.
This section only describes running ISAJET as a standalone program and generating output in machine-dependent binary form. The user may elect to analyze events as they are generated; this is discussed in Section 5 of this manual.
## 4 Patchy and PAM Organization
Patchy is a code management system developed at CERN and used to maintain the CERN Library. It is used to provide versions of ISAJET for a wide variety of computers. Instructions for using PATCHY are available from `http://wwwinfo.cern.ch/asdoc/Welcome.html`.
A master source file in Patchy is called a โPAM.โ The ISAJET PAM contains all the source code and documentation plus Patchy commands to include common blocks and to select the desired version. It is divided into the following patches:
`ISACDE`: contains all common blocks, etc. These are divided into decks based on their usage.
`ISADATA`: contains block data ALDATA. This must always be loaded when using ISAJET.
`ISAJET`: contains the code for generating events. Each subroutine is in a separate deck with the same name.
`ISASSRUN`: contains the main program for ISASUSY, which prompts for input parameters and prints out all the decay modes. It is selected by `*ISASUSY`.
`ISASUSY`: contains code to calculate all the decay widths and branching fractions in the minimal supersymmetric model.
`ISATAPE`: contains the code for reading and writing tapes, again with each subroutine on a separate deck.
`ISARUN`: contains a main program and a simple interactive interface. It is selected by `IF=INTERACT`.
`ISAZEB`: contains Zebra format output routines, an alternative to the ISATAPE routines.
`ISZRUN`: contains the analog of ISAPLT for the Zebra format.
`ISAPLT`: contains a simple calorimeter simulation and the skeleton of a histogramming job using HBOOK.
`ISATEXT`: contains the instructions for using ISAJET, i.e. the text of this document. It also includes the documentation for ISASUSY.
`ISZTEXT`: contains the instructions for the Zebra output routines and a description of the Zebra banks.
`ISADECAY`: contains the input decay table.
The code is actually maintained using RCS on a Silicon Graphics computer at BNL. Patchy is used primarily to handle common blocks and machine dependent code.
The input to YPATCHY must contain both `+USE` cards, which define the wanted program version, and `+EXE` cards, which determine which patches or decks are written to the ASM file. To facilitate this selection, the ISAJET PAM contains the following pilot patches:
`*ISADECAY`: USE selects ISADECAY and all corrections to it.
`*ISAJET`: USE selects ISACDE, ISADATA, ISAJET, ISATAPE, ISARUN and all corrections to them. Note that ISARUN is not actually selected without `+USE,INTERACT`.
`*ISAPLT`: USE selects ISACDE, ISAPLT, and all corrections to them.
`*ISASUSY`: USE selects CDESUSY, ISASUSY, and ISASSRUN to create a program to calculate all the MSSM decay modes.
`*ISATEXT`: USE selects ISACDE, ISATEXT, and all corrections to them.
`*ISAZEB`: USE selects ISAJET with a Zebra output format.
`*ISZRUN`: USE selects the Zebra analysis package.
Patches are provided to select the machine dependent features for specific computers or operating systems:
`ANSI`: ANSI standard Fortran (no time or date functions)
`APOLLO`: APOLLO โ only tested by CERN
`CDC`: CDC 7600 and 60-bit CYBER (obsolete)
`CRAY`: CRAY with UNICOS
`DECS`: DEC Station with Ultrix
`ETA`: ETA 10 running Unix System V (obsolete)
`HPUX`: HP/9000 7xx running Unix System V
`IBM`: IBM 370 and 30xx running VM/CMS
`IBMRT`: IBM RS/6000 running AIX 3.x or 4.x
`LINUX`: PC running Linux with f2c/gcc or g77 compiler
`SGI`: Silicon Graphics running IRIX
`SUN`: Sun Sparcstation running SUNOS or Solaris
`VAX`: DEC VAX or Alpha running VMS
These patches in turn select a variety of patches and IF flags, allowing one to select more specific features, as indicated below. (Replace `&` by `+` everywhere.)
```
&PATCH,ANSI. GENERIC ANSI FORTRAN.
&USE,DOUBLE. DOUBLE PRECISION.
&USE,STDIO. STANDARD FORTRAN 77 TAPE INPUT/OUTPUT.
&USE,MOVEFTN. FORTRAN REPLACEMENT FOR MOVLEV.
&USE,RANFFTN,IF=-CERN. FORTRAN RANF.
&USE,RANFCALL. STANDARD RANSET AND RANGET CALLS.
&USE,NOCERN,IF=-CERN. NO CERN LIBRARY.
&EOD
&PATCH,APOLLO.
&DECK,BLANKDEK.
&USE,DOUBLE. DOUBLE PRECISION.
&USE,STDIO. STANDARD FORTRAN 77 TAPE INPUT/OUTPUT.
&USE,MOVEFTN. FORTRAN REPLACEMENT FOR MOVLEV.
&USE,RANFFTN,IF=-CERN. FORTRAN RANF.
&USE,RANFCALL. STANDARD RANSET AND RANGET CALLS.
&USE,NOCERN,IF=-CERN. NO CERN LIBRARY.
&USE,IMPNONE. IMPLICIT NONE
&EOD.
&PATCH,CDC. CDC 7600 OR CYBER 175.
&USE,SINGLE. SINGLE PRECISION.
&USE,LEVEL2. LEVEL 2 STORAGE.
&USE,CDCPACK. PACK 2 WORDS PER WORD FOR INPUT/OUTPUT.
&USE,RANFCALL. STANDARD RANSET AND RANGET CALLS.
&USE,NOCERN,IF=-CERN. NO CERN LIBRARY.
&EOD
&PATCH,CRAY. CRAY XMP OR 2.
&USE,SINGLE. SINGLE PRECISION.
&USE,STDIO. STANDARD FORTRAN 77 TAPE INPUT/OUTPUT.
&USE,MOVEFTN. FORTRAN REPLACEMENT FOR MOVLEV.
&USE,NOCERN,IF=-CERN. NO CERN LIBRARY.
&EOD
&PATCH,DECS. DEC STATION (ULTRIX)
&USE,SUN.
&EOD
&PATCH,ETA. ETA-10.
&USE,SINGLE. SINGLE PRECISION.
&USE,STDIO. STANDARD FORTRAN 77 TAPE INPUT/OUTPUT.
&USE,MOVEFTN. FORTRAN REPLACEMENT FOR MOVLEV.
&USE,RANFCALL. STANDARD RANSET AND RANGET CALLS.
&USE,NOCERN,IF=-CERN. NO CERN LIBRARY.
&EOD
&PATCH,HPUX. HP/9000 7XX RUNNING UNIX.
&USE,DOUBLE. DOUBLE PRECISION.
&USE,STDIO. STANDARD FORTRAN 77 TAPE INPUT/OUTPUT.
&USE,MOVEFTN. FORTRAN REPLACEMENT FOR MOVLEV.
&USE,RANFFTN,IF=-CERN. FORTRAN RANF.
&USE,RANFCALL. STANDARD RANSET AND RANGET CALLS.
&USE,NOCERN,IF=-CERN. NO CERN LIBRARY.
&USE,IMPNONE. IMPLICIT NONE
&EOD
&PATCH,IBM. IBM 370 OR 30XX.
&USE,DOUBLE. DOUBLE PRECISION.
&USE,STDIO. STANDARD FORTRAN 77 TAPE INPUT/OUTPUT.
&USE,MOVEFTN. FORTRAN REPLACEMENT FOR MOVLEV.
&USE,RANFFTN,IF=-CERN. FORTRAN RANF.
&USE,RANFCALL. STANDARD RANSET AND RANGET CALLS.
&USE,NOCERN,IF=-CERN. NO CERN LIBRARY.
&EOD
&PATCH,IBMRT. IBM RS/6000 WITH AIX 3.1
&USE,DOUBLE. DOUBLE PRECISION.
&USE,STDIO. STANDARD FORTRAN 77 TAPE INPUT/OUTPUT.
&USE,MOVEFTN. FORTRAN REPLACEMENT FOR MOVLEV.
&USE,RANFFTN,IF=-CERN. FORTRAN RANF.
&USE,RANFCALL. STANDARD RANSET AND RANGET CALLS.
&USE,NOCERN,IF=-CERN. NO CERN LIBRARY.
&USE,IMPNONE. IMPLICIT NONE
&EOD
&PATCH,LINUX. IBM PC WITH LINUX 1.X
&USE,DOUBLE. DOUBLE PRECISION.
&USE,STDIO. STANDARD FORTRAN 77 TAPE INPUT/OUTPUT.
&USE,MOVEFTN. FORTRAN REPLACEMENT FOR MOVLEV.
&USE,RANFFTN,IF=-CERN. FORTRAN RANF.
&USE,RANFCALL. STANDARD RANSET AND RANGET CALLS.
&USE,NOCERN,IF=-CERN. NO CERN LIBRARY.
&USE,IMPNONE. IMPLICIT NONE
&EOD
&PATCH,SGI.
SILICON GRAPHICS 4D/XX.
&USE,DOUBLE. DOUBLE PRECISION.
&USE,STDIO. STANDARD FORTRAN 77 TAPE INPUT/OUTPUT.
&USE,MOVEFTN. FORTRAN REPLACEMENT FOR MOVLEV.
&USE,RANFFTN,IF=-CERN. FORTRAN RANF.
&USE,RANFCALL. STANDARD RANSET AND RANGET CALLS.
&USE,NOCERN,IF=-CERN. NO CERN LIBRARY.
&EOD
&PATCH,SUN. SUN (SPARC)
&USE,DOUBLE. DOUBLE PRECISION.
&USE,STDIO. STANDARD FORTRAN 77 TAPE INPUT/OUTPUT.
&USE,MOVEFTN. FORTRAN REPLACEMENT FOR MOVLEV.
&USE,RANFFTN,IF=-CERN. FORTRAN RANF.
&USE,RANFCALL. STANDARD RANSET AND RANGET CALLS.
&USE,NOCERN,IF=-CERN. NO CERN LIBRARY.
&EOD
&PATCH,VAX. DEC VAX 11/780 OR 8600.
&USE,DOUBLE. DOUBLE PRECISION.
&USE,STDIO. STANDARD FORTRAN 77 TAPE INPUT/OUTPUT.
&USE,MOVEFTN. FORTRAN REPLACEMENT FOR MOVLEV.
&USE,RANFFTN,IF=-CERN. FORTRAN RANF.
&USE,RANFCALL. STANDARD RANSET AND RANGET CALLS.
&USE,NOCERN,IF=-CERN. NO CERN LIBRARY.
&USE,IMPNONE. IMPLICIT NONE
&EOD
```
An empty patch INTERACT selects a main program and an interactive interface which will prompt the user for parameters and do some error checking. A patch CERN allows ISAJET to take the random number generator RANF and several other routines from the CERN Library; to use this include the Patchy command
```
&USE,CERN.
```
Similarly, a patch PDFLIB enables the interface to the PDFLIB parton distribution compilation by H. Plothow-Besch:
```
&USE,PDFLIB
```
The only internal links with PDFLIB are calls to the routines PDFSET, PFTOPDG, and DXPDF, and the common blocks W50510 and W50517,
```
C Copy of PDFLIB common block
COMMON/W50510/IFLPRT
INTEGER IFLPRT
SAVE /W50510/
C Copy of PDFLIB common block
COMMON/W50517/N6
INTEGER N6
SAVE /W50517/
```
which are used to specify the level of output messages and the logical unit number for them.
In general it should be sufficient to run YPATCHY with the following cradle (replace `&` with `+` everywhere):
```
&USE,(*ISAJET,*ISATEXT,*ISADECAY,*ISAPLT). CHOOSE ONE.
&USE,ANSI,DECS,HPUX,IBM,IBMRT,SGI,SUN,.... CHOOSE ONE.
&[USE,INTERACT]. FOR INTERACTIVE MODE.
&[USE,CERN.] FOR CERN LIBRARY.
&[USE,HBOOK3.] HBOOK 3 FOR ISAPLT.
&EXE.
&PAM.
&QUIT.
```
The input to YPATCHY can also contain changes by the user. It is suggested that these not be made permanent parts of the PAM to avoid possible conflicts with later corrections.
## 5 Main Program
A main program is not supplied with ISAJET. To generate events and write them to disk, the user should provide a main program which opens the files and then calls subroutine ISAJET. In the following sample, i,j,m,n are arbitrary unit numbers.
Main program for VMS:
```
PROGRAM RUNJET
C
C MAIN PROGRAM FOR ISAJET ON BNL VAX CLUSTER.
C
OPEN(UNIT=i,FILE=โ$2$DUA14:[ISAJET.ISALIBRARY]DECAY.DATโ,
$STATUS=โOLDโ,FORM=โFORMATTEDโ,READONLY)
OPEN(UNIT=j,FILE=โmyjob.datโ,STATUS=โNEWโ,FORM=โUNFORMATTEDโ)
OPEN(UNIT=m,FILE=โmyjob.parโ,STATUS=โOLDโ,FORM=โFORMATTEDโ)
OPEN(UNIT=n,FILE=โmyjob.lisโ,STATUS=โNEWโ,FORM=โFORMATTEDโ)
C
CALL ISAJET(+-i,+-j,m,n)
C
STOP
END
```
Main program for IBM (VM/CMS)
```
PROGRAM RUNJET
C
C MAIN PROGRAM FOR ISAJET ON IBM ASSUMING FILES HAVE BEEN
C OPENED WITH FILEDEF.
C
CALL ISAJET(+-i,+-j,m,n)
C
STOP
END
```
Main program for Unix:
```
PROGRAM RUNJET
C
C Main program for ISAJET on Unix
C
CHARACTER*60 FNAME
C
C Open user files
READ 1000, FNAME
1000 FORMAT(A)
PRINT 1020, FNAME
1020 FORMAT(1X,โData file = โ,A)
OPEN(2,FILE=FNAME,STATUS=โNEWโ,FORM=โUNFORMATTEDโ)
READ 1000, FNAME
PRINT 1030, FNAME
1030 FORMAT(1X,โParameter file = โ,A)
OPEN(3,FILE=FNAME,STATUS=โOLDโ,FORM=โFORMATTEDโ)
READ 1000, FNAME
PRINT 1040, FNAME
1040 FORMAT(1X,โListing file = โ,A)
OPEN(4,FILE=FNAME,STATUS=โNEWโ,FORM=โFORMATTEDโ)
C Open decay table
READ 1000, FNAME
OPEN(1,FILE=FNAME,STATUS=โOLDโ,FORM=โFORMATTEDโ)
C
C Run ISAJET
CALL ISAJET(-1,2,3,4)
C
STOP
END
```
The arguments of ISAJET are tape numbers for files, all of which should be opened by the main program.
`TAPEi`: Decay table (formatted). A positive sign prints the decay table on the output listing. A negative sign suppress printing of the decay table.
`TAPEj`: Output file for events (unformatted). A positive sign writes out both resonances and stable particles. A negative sign writes out only stable particles.
`TAPEm`: Commands as defined in Section 6 (formatted).
`TAPEn`: Output listing (formatted).
In the sample jobs in Section 3, TAPEm is the default Fortran input, and TAPEn is the default Fortran output.
### 5.1 Interactive Interface
To use the interactive interface, replace the call to ISAJET in the above main program by
```
CALL ISASET(+-i,+-j,m,n)
CALL ISAJET(+-i,+-j,m,n)
```
ISASET calls DIALOG, which prompts the user for possible commands, does a limited amount of error checking, and writes a command file on TAPEm. This command file is rewound for execution by ISAJET. A main program is included in patch ISARUN to open the necessary files and to call ISASET and ISAJET.
### 5.2 User Control of Event Loop
If the user wishes to integrate ISAJET with another program and have control over the event generation, he can call the driving subroutines himself. The driving subroutines are:
`ISAINI(+-i,+-j,m,n)`: initialize ISAJET. The arguments are the same as for subroutine ISAJET.
`ISABEG(IFL)`: begin a run. IFL is a return flag: IFL=0 for a good set of commands; IFL=1001 for a STOP; any other value means an error.
`ISAEVT(I,OK,DONE)` generate event I. Logical flag OK signifies a good event (almost always .TRUE.); logical flag DONE signifies the end of a run.
`ISAEND`: end a run.
There are also subroutines provided to write standard ISAJET records, or Zebra records if the Zebra option is selected:
`ISAWBG` to write a begin-of-run record, should be called immediately after ISABEG
`ISAWEV` to write an event record, should be called immediately after ISAEVT
`ISAWND` to write an end-of-run record, should be called immediately after ISAEND
The control of the event loop is somewhat complicated to accomodate multiple evolution and fragmentation as described in Section 11. Note in particular that after calling ISAEVT one should process or write out the event only if OK=.TRUE. The check on the DONE flag is essential if one is doing multiple evolution and fragmentation. The following example indicates how events might be generated, analyzed, and discarded (replace `&` by `+` everywhere):
```
PROGRAM SAMPLE
C
&SELF,IF=IMPNONE
IMPLICIT NONE
&SELF
&CDE,ITAPES
&CDE,IDRUN
&CDE,PRIMAR
&CDE,ISLOOP
C
INTEGER JTDKY,JTEVT,JTCOM,JTLIS,IFL,ILOOP
LOGICAL OK,DONE
SAVE ILOOP
C---------------------------------------------------------------------
C> Open files as above
C> Call user initialization
C---------------------------------------------------------------------
C
C Initialize ISAJET
C
CALL ISAINI(-i,0,m,n)
1 IFL=0
CALL ISABEG(IFL)
IF(IFL.NE.0) STOP
C
C Event loop
C
ILOOP=0
101 CONTINUE
ILOOP=ILOOP+1
C Generate one event - discard if .NOT.OK
CALL ISAEVT(ILOOP,OK,DONE)
IF(OK) THEN
C---------------------------------------------------------------------
C> Call user analysis for event
C---------------------------------------------------------------------
ENDIF
IF(.NOT.DONE) GO TO 101
C
C Calculate cross section and luminosity
C
CALL ISAEND
C---------------------------------------------------------------------
C> Call user summary
C---------------------------------------------------------------------
GO TO 1
END
```
### 5.3 Multiple Event Streams
It may be desirable to generate several different kinds of events simultaneously to study pileup effects. While normally one would want to superimpose minimum bias or low-pt jet events on a signal of interest, other combinations might also be interesting. It would be very inefficient to reinitialize ISAJET for each event. Therefore, a pair of subroutines is provided to save and restore the context, i.e. all of the initialization information, in an array. The syntax is
```
CALL CTXOUT(NC,VC,MC)
CALL CTXIN(NC,VC,MC)
```
where VC is a real array of dimension MC and NC is the number of words used, about 20000 in the standard case. If NC exceeds MC, a warning is printed and the job is terminated. The use of these routines is illustrated in the following example, which opens the files with names read from the standard input and then superimposes on each event of the signal sample three events of a pileup sample. It is assumed that a large number of events is specified in the parameter file for the pileup sample so that it does not terminate.
```
PROGRAM SAMPLE
C
C Example of generating two kinds of events.
C
CHARACTER*60 FNAME
REAL VC1(20000),VC2(20000)
LOGICAL OK1,DONE1,OK2,DONE2
INTEGER NC1,NC2,IFL,ILOOP,I2,ILOOP2
C
C Open decay table
READ 1000, FNAME
1000 FORMAT(A)
OPEN(1,FILE=FNAME,STATUS=โOLDโ,FORM=โFORMATTEDโ)
C Open user files
READ 1000, FNAME
OPEN(3,FILE=FNAME,STATUS=โOLDโ,FORM=โFORMATTEDโ)
READ 1000, FNAME
OPEN(4,FILE=FNAME,STATUS=โNEWโ,FORM=โFORMATTEDโ)
READ 1000,FNAME
OPEN(13,FILE=FNAME,STATUS=โOLDโ,FORM=โFORMATTEDโ)
READ 1000,FNAME
OPEN(14,FILE=FNAME,STATUS=โNEWโ,FORM=โFORMATTEDโ)
C
C Initialize ISAJET
CALL ISAINI(-1,0,3,4)
CALL CTXOUT(NC1,VC1,20000)
CALL ISAINI(-1,0,13,14)
IFL=0
CALL ISABEG(IFL)
IF(IFL.NE.0) STOP1
CALL CTXOUT(NC2,VC2,20000)
ILOOP2=0
CALL user_initialization_routine
C
1 IFL=0
CALL CTXIN(NC1,VC1,20000)
CALL ISABEG(IFL)
CALL CTXOUT(NC1,VC1,20000)
IF(IFL.NE.0) GO TO 999
ILOOP=0
C
C Main event
C
101 CONTINUE
ILOOP=ILOOP+1
CALL CTXIN(NC1,VC1,20000)
CALL ISAEVT(ILOOP,OK1,DONE1)
CALL CTXOUT(NC1,VC1,20000)
IF(.NOT.OK1) GO TO 101
CALL user_analysis_routine
C
C Pileup
C
CALL CTXIN(NC2,VC2,20000)
I2=0
201 CONTINUE
ILOOP2=ILOOP2+1
CALL ISAEVT(ILOOP2,OK2,DONE2)
IF(OK2) I2=I2+1
IF(DONE2) STOP2
CALL user_analysis_routine
IF(I2.LT.3) GO TO 201
CALL CTXOUT(NC2,VC2,20000)
C
IF(.NOT.DONE1) GO TO 101
C
C Calculate cross section and luminosity
C
CALL CTXIN(NC1,VC1,20000)
CALL ISAEND
GO TO 1
C
999 CALL CTXIN(NC2,VC2,20000)
CALL ISAEND
CALL user_termination_routine
STOP
END
```
It is possible to superimpose arbitrary combinations of events, including events of the same reaction type with different parameters. In general the number of events would be selected randomly based on the cross sections and the luminosity.
At this time CTXOUT and CTXIN cannot be used with the Zebra output routines.
## 6 Input
### 6.1 Input Format
ISAJET is controlled by commands read from the specified input file by subroutine READIN. (In the interactive version, this file is first created by subroutine DIALOG.) Syntax errors will generate a message and stop execution. Based on these commands, subroutine LOGIC will setup limits for all variables and check for inconsistencies. Several runs with different parameters can be combined into one job. The required input format is:
```
Title
Ecm,Nevent,Nprint,Njump/
Reaction
(Optional parameters)
END
(Optional additional runs)
STOP
```
with all lines starting in column 1 and typed in upper case. These lines are explained below.
Title line: Up to 80 characters long. If the first four letters are STOP, control is returned to main program. If the first four letters are SAME, the parameters from previous run are used excepting those which are explicitly changed.
Ecm line: This line must always be given even if the title is SAME. It must give the center of mass energy (Ecm) and the number of events (Nevent) to be generated. One may also specify the number of events to be printed (Nprint) and the increment (Njump) for printing. The first event is always printed if Nprint $`>`$ 0. For example:
```
800.,1000,10,100/
```
generates 1000 events at $`E_{\mathrm{cm}}=800\mathrm{GeV}`$ and prints 10 events. The events printed are: 1,100,200,โฆ. Note that an event typically takes several pages of output. This line is read with a list directed format (READ\*).
After Nprint events have been printed, a single line containing the run number, the event number, and the random number seed is printed every Njump events (if Njump is nonzero). This seed can be used to start a new job with the given event if in the new run NSIGMA is set equal to zero:
```
SEED
value/
NSIGMA
0/
```
In general the same events will only be generated on the same type of computer.
Reaction line: This line must be given unless title is SAME, when it must be omitted. It selects the type of events to be generated. The present version can generate TWOJET, E+E-, DRELLYAN, MINBIAS, WPAIR, SUPERSYM, HIGGS, PHOTON, TCOLOR, or WHIGGS events. This line is read with an A8 format.
### 6.2 Optional Parameters
Each optional parameter requires two lines. The first line is a keyword specifying the parameter and the second line gives the values for the parameter. The parameters can be given in any order. Numerical values are read with a list directed format (READ\*), jet and particle types are read with a character format and must be enclosed in quotes, and logical flags with an L1 format. All momenta are in GeV and all angles are in radians.
The parameters can be classified in several groups:
| Jet Limits: | W/H Limits: | Decays: | Constants: | Other: |
| --- | --- | --- | --- | --- |
| JETTYPE1 | HTYPE | FORCE | AMSB | BEAMS |
| JETTYPE2 | PHIW | FORCE1 | CUTJET | EPOL |
| JETTYPE3 | QMH | NODECAY | CUTOFF | EEBEAM |
| MIJLIM | QMW | NOETA | EXTRAD | EEBREM |
| MTOT | QTW | NOEVOLVE | FRAGMENT | NPOMERON |
| P | THW | NOFRGMNT | GAUGINO | NSIGMA |
| PHI | WTYPE | NOGRAV | GMSB | NTRIES |
| PT | XW | NOPI0 | GMSB2 | PDFLIB |
| TH | YW | | HMASS | SEED |
| X | | | HMASSES | STRUC |
| Y | | | LAMBDA | WFUDGE |
| WMODE1 | | | MGVTNO | WMMODE |
| WMODE2 | | | MSSMA | WPMODE |
| | | | MSSMB | Z0MODE |
| | | | MSSMC | |
| | | | MSSMD | |
| | | | MSSME | |
| | | | NUSUG1 | |
| | | | NUSUG2 | |
| | | | NUSUG3 | |
| | | | NUSUG4 | |
| | | | NUSUG5 | |
| | | | SIGQT | |
| | | | SIN2W | |
| | | | SLEPTON | |
| | | | SQUARK | |
| | | | SUGRA | |
| | | | SUGRHN | |
| | | | TCMASS | |
| | | | TMASS | |
| | | | WMASS | |
It may be helpful to know that the TWOJET, WPAIR, PHOTON, SUPERSYM, and WHIGGS processes use the same controlling routines and so share many of the same variables. In particular, PT limits should normally be set for these processes, and JETTYPE1 and JETTYPE2 are used to select the reactions. Similarly, the DRELLYAN, HIGGS, and TCOLOR processes use the same controlling routines since they all generate s-channel resonances. The mass limits for these processes are set by QMW. Normally the QMW limits will surround the $`W^\pm `$, $`Z^0`$, or Higgs mass, but this is not required. (QMH acts like QMW for the Higgs process.) For historical reasons, JETTYPE1 and JETTYPE2 are used to select the W decay modes in DRELLYAN, while WMODE1 and WMODE2 select the W decay modes for WPAIR, HIGGS, and WHIGGS. Also, QTW can be used to generate DRELLYAN events with non-zero transverse momentum, whereas HIGGS automatically fixes QTW to be zero. (Of course, non-zero transverse momentum will be generated by gluon radiation.)
For example the lines
```
P
40.,50.,10.,100./
```
would set limits for the momentum of jet 1 between 40 and 50 GeV, and for jet 2 between 10 and 100 GeV. As another example the lines
```
WTYPE
โW+โ/
```
would specify that for DRELLYAN events only W+ events will be generated. If for a kinematic variable only the lower limit is specified then that parameter is fixed to the given value. Thus the lines
```
P
40.,,10./
```
will fix the momentum for jet 1 to be 40 GeV and for jet 2 to be 10 GeV. If only the upper limit is specified then the default value is used for the lower limit. Jet 1 or jet 2 parameters for DRELLYAN events refer to the W decay products and cannot be fixed. If QTW is fixed to 0, then standard Drell-Yan events are generated.
A complete list of keywords and their default values follows.
| Keyword | | Explanation |
| --- | --- | --- |
| Values | Default values | |
| AMSB | | Anomaly-mediated SUSY breaking |
| $`m_0`$,$`m_{3/2}`$,$`\mathrm{tan}\beta `$,$`sgn\mu `$ | none | scalar mass, gravitino mass, |
| | | VEV ratio, sign |
| BEAMS | | Initial beams. Allowed are |
| type<sub>1</sub>,type<sub>2</sub> | โPโ,โPโ | โPโ,โAPโ,โNโ,โANโ. |
| CUTJET | | Cutoff mass for QCD jet |
| $`\mu _c`$ | 6. | evolution. |
| CUTOFF | | Cutoff $`qt^2=\mu ^2Q^\nu `$ for |
| $`\mu ^2`$, $`\nu `$ | .200,1.0 | DRELLYAN events. |
| EEBEAM | | impose brem/beamstrahlung |
| $`\sqrt{\widehat{s}}_{min}`$, $`\sqrt{\widehat{s}}_{max}`$, $`\mathrm{{\rm Y}}`$, $`\sigma _z`$ | none | min and max subprocess energy, |
| | | beamstrahlung parameter $`\mathrm{{\rm Y}}`$ |
| | | longitudinal beam size $`\sigma _z`$ in mm |
| EEBREM | | impose bremsstrahlung for $`e^+e^{}`$ |
| $`\sqrt{\widehat{s}}_{min}`$, $`\sqrt{\widehat{s}}_{max}`$ | none | min and max subprocess energy |
| EPOL | | Polarization of $`e^{}`$ ($`e^+`$) beam, |
| $`P_L(e^{}),P_L(e^+)`$ | 0,0 | $`P_L(e)=(n_Ln_R)/(n_Ln_R)`$, |
| | | so that $`1P_L1`$ |
| EXTRAD | | Parameters for EXTRADIM process |
| $`\delta `$,$`M_D`$,UVCUT | None | UVCUT is logical flag |
| FORCE | | Force decay of particles, |
| $`i,i_1,\mathrm{},i_5`$/ | None | $`\pm i\pm (i1+\mathrm{}+i5)`$. |
| | | Can call 20 times. |
| | | See note for $`i`$ = quark. |
| FORCE1 | | Force decay $`ii1+\mathrm{}+i5`$. |
| $`i,i_1,\mathrm{},i_5`$/ | None | Can call 40 times. |
| | | See note for $`i`$ = quark. |
| FRAGMENT | | Fragmentation parameters. |
| $`P_{ud}`$,โฆ | .4,โฆ | See also SIGQT, etc. |
| GAUGINO | | Masses for $`\stackrel{~}{g}`$, $`\stackrel{~}{\gamma }`$, |
| $`m_1`$,$`m_2`$,$`m_3,m_4`$ | 50,0,100,100 | $`\stackrel{~}{W}^+`$, and $`\stackrel{~}{Z}^0`$ |
| GMSB | | GMSB messenger SUSY breaking, |
| --- | --- | --- |
| $`\mathrm{\Lambda }_m`$,$`M_m`$,$`N_5`$ | none | mass, number of $`5+\overline{5}`$, VEV |
| $`\mathrm{tan}\beta `$,$`sgn\mu `$,$`C_{\mathrm{gr}}`$ | | ratio, sign, gravitino scale |
| GMSB2 | | non-minimal GMSB parameters |
| $`\text{ / }R`$,$`\delta M_{H_d}^2`$,$`\delta M_{H_u}^2`$,$`D_Y(M)`$ | 1,0,0,0 | gaugino mass multiplier |
| $`N_{5_1}`$,$`N_{5_2}`$,$`N_{5_3}`$ | $`N_5`$ | Higgs mass shifts, D-term mass<sup>2</sup> |
| | | indep. gauge group messengers |
| HMASS | 0 | Mass for standard Higgs. |
| $`m`$ | | |
| HMASSES | | Higgs meson masses for |
| $`m_1`$,โฆ,$`m_9`$ | 0,โฆ,0 | charges 0,0,0,0,0,1,1,2,2. |
| HTYPE | | One MSSM Higgs type (โHL0โ, |
| โHL0โ/ orโฆ | none | โHH0โ, or โHA0โ) |
| JETTYPE1 | | )Select types for jets: |
| โGLโ,โUPโ,โฆ | โALLโ | )โALLโ; โGLโ; โQUARKSโ=โUPโ, |
| | | )โUBโ,โDNโ,โDBโ,โSTโ,โSBโ, |
| JETTYPE2 | | )โCHโ,โCBโ,โBTโ,โBBโ,โTPโ, |
| โGLโ,โUPโ,โฆ | โALLโ | )โTBโ,โXโ,โXBโ,โYโ,โYBโ; |
| | | )โLEPTONSโ=โE-โ,โE+โ,โMU-โ, |
| JETTYPE3 | | )โMU+โ,โTAU-โ,โTAU+โ; โNUSโ; |
| โGLโ,โUPโ,โฆ | โALLโ | )โGMโ,โW+โ,โW-โ,โZ0โ |
| | | ) See note for SUSY types. |
| LAMBDA | | QCD scale |
| $`\mathrm{\Lambda }`$ | .2 | |
| MGVTNO | | Gravitino mass โ ignored for |
| $`M_{\mathrm{gravitino}}`$ | $`10^{20}`$ GeV | GMSB model |
| MIJLIM | | Multimet mass limits |
| $`i`$,$`j`$,$`M_{\mathrm{min}}`$,$`M_{\mathrm{max}}`$ | 0,0,$`1\mathrm{GeV}`$,$`1\mathrm{GeV}`$ | |
| MSSMA | | MSSM parameters โ |
| $`m(\stackrel{~}{g})`$,$`\mu `$, | Required | Gluino mass, $`\mu `$, $`A`$ mass, |
| $`m(A)`$,$`\mathrm{tan}\beta `$ | | $`\mathrm{tan}\beta `$ |
| MSSMB | | MSSM 1st generation โ |
| $`m(q_1)`$,$`m(d_r)`$,$`m(u_r)`$, | Required | Left and right soft squark and |
| $`m(l_1)`$,$`m(e_r)`$ | | slepton masses |
| MSSMC | | MSSM 3rd generation โ |
| --- | --- | --- |
| $`m(q_3)`$,$`m(b_r)`$,$`m(t_r)`$, | Required | Soft squark masses, slepton |
| $`m(l_3)`$,$`m(\tau _r)`$, | | masses, and squark and slepton |
| $`A_t`$,$`A_b`$,$`A_\tau `$ | | mixings |
| MSSMD | | MSSM 2nd generation โ |
| $`m(q_2)`$,$`m(s_r)`$,$`m(c_r)`$, | from MSSMB | Left and right soft squark and |
| $`m(l_2)`$,$`m(mu_r)`$ | | slepton masses |
| MSSME | | MSSM gaugino masses โ |
| $`M_1`$,$`M_2`$ | MSSMA + GUT | Default is to scale from gluino |
| MTOT | | Mass range for multiparton |
| $`M_{\mathrm{min}}`$,$`M_{\mathrm{max}}`$ | None | processes |
| NODECAY | | Suppress all decays. |
| TRUE or FALSE | FALSE | |
| NOETA | | Suppress eta decays. |
| TRUE or FALSE | FALSE | |
| NOEVOLVE | | Suppress QCD evolution and |
| TRUE or FALSE | FALSE | hadronization. |
| NOGRAV | | Suppress gravitino decays in |
| TRUE or FALSE | FALSE | GMSB model |
| NOHADRON | | Suppress hadronization of |
| TRUE or FALSE | FALSE | jets and beam jets. |
| NONUNU | | Suppress $`Z^0`$ neutrino decays. |
| TRUE or FALSE | FALSE | |
| NOPI0 | | Suppress $`\pi ^0`$ decays. |
| TRUE or FALSE | FALSE | |
| NPOMERON | | Allow $`n_1<n<n_2`$ cut pomerons. |
| $`n_1`$,$`n_2`$ | 1,20 | Controls beam jet mult. |
| NSIGMA | | Generate n unevolved events |
| $`n`$ | 20 | for SIGF calculation. |
| NTRIES | | Stop if after n tries |
| $`n`$ | 1000 | cannot find a good event. |
| NUSUG1 | | Optional non-universal SUGRA |
| --- | --- | --- |
| $`M_1`$,$`M_2`$,$`M_3`$ | none | gaugino masses |
| NUSUG2 | | Optional non-universal SUGRA |
| $`A_t`$,$`A_b`$,$`A_\tau `$ | none | $`A`$ terms |
| NUSUG3 | | Optional non-universal SUGRA |
| $`M_{H_d}`$,$`M_{H_u}`$ | none | Higgs masses |
| NUSUG4 | | Optional non-universal SUGRA |
| $`M_{u_L}`$,$`M_{d_R}`$,$`M_{u_R}`$, | none | 1st/2nd generation masses |
| $`M_{e_L}`$,$`M_{e_R}`$ | | |
| NUSUG5 | | Optional non-universal SUGRA |
| $`M_{t_L}`$,$`M_{b_R}`$,$`M_{t_R}`$, | none | 3rd generation masses |
| $`M_{\tau _L}`$,$`M_{\tau _R}`$ | | |
| P | | Momentum limits for jets. |
| $`p_{\mathrm{min}}(1)`$,โฆ,$`p_{\mathrm{max}}(3)`$ | 1.,$`0.5E_{\mathrm{cm}}`$ | |
| PDFLIB | | CERN PDFLIB parton distribution |
| โname<sub>1</sub>โ,val<sub>1</sub>,โฆ | None | parameters. See PDFLIB manual. |
| PHI | | Phi limits for jets. |
| $`\varphi _{\mathrm{min}}(1)`$,โฆ,$`\varphi _{\mathrm{max}}(3)`$ | 0,$`2\pi `$ | |
| PHIW | | Phi limits for W. |
| $`\varphi _{\mathrm{min}}`$,$`\varphi _{\mathrm{max}}`$ | 0,$`2\pi `$ | |
| PT or PPERP | | $`p_t`$ limits for jets. |
| $`p_{t,\mathrm{min}}(1)`$,โฆ,$`p_{t,\mathrm{max}}(3)`$ | $`.05E_{\mathrm{cm}}`$,$`.2E_{\mathrm{cm}}`$ | Default for TWOJET only. |
| QMH | | Mass limits for Higgs. |
| $`q_{\mathrm{min}}`$,$`q_{\mathrm{max}}`$ | $`.05E_{\mathrm{cm}}`$,$`.2E_{\mathrm{cm}}`$ | Equivalent to QMW. |
| QMW | | Mass limits for $`W`$. |
| $`q_{\mathrm{min}}`$,$`q_{\mathrm{max}}`$ | $`.05E_{\mathrm{cm}}`$,$`.2E_{\mathrm{cm}}`$ | |
| QTW | | $`q_t`$ limits for $`W`$. Fix $`q_t=0`$ |
| $`q_{t,\mathrm{min}}`$,$`q_{t,\mathrm{max}}`$ | .1,$`.025E_{\mathrm{cm}}`$ | for standard Drell-Yan. |
| SEED | | Random number seed (double |
| real | 0 | precision if 32 bit). |
| SIGQT | | Internal $`k_t`$ parameter for |
| --- | --- | --- |
| $`\sigma `$ | .35 | jet fragmentation. |
| SIN2W | | Weinberg angle. See WMASS. |
| $`\mathrm{sin}^2(\theta _W)`$ | .232 | |
| SLEPTON | | Masses for $`\stackrel{~}{\nu }_e`$, $`\stackrel{~}{e}`$, $`\stackrel{~}{\nu }_\mu `$ |
| $`m_1`$,โฆ,$`m_6`$ | 100,โฆ,101.8 | $`\stackrel{~}{\mu }`$, $`\stackrel{~}{\nu }_\tau `$, $`\stackrel{~}{\tau }`$ |
| SQUARK | | Masses for $`\stackrel{~}{u}`$, $`\stackrel{~}{d}`$, $`\stackrel{~}{s}`$, |
| $`m_1`$,โฆ,$`m_6`$ | 100.3,โฆ,240. | $`\stackrel{~}{c}`$, $`\stackrel{~}{b}`$, $`\stackrel{~}{t}`$ |
| STRUC | | Structure functions. CTEQ3L, |
| name | โCTEQ3Lโ | CTEQ2L, EHLQ, OR DO |
| SUGRA | | Minimal supergravity parameters |
| $`m_0`$,$`m_{1/2}`$,$`A_0`$, | none | scalar M, gaugino M, trilinear |
| $`\mathrm{tan}\beta `$,$`sgn\mu `$ | | breaking term, vev ratio, +-1 |
| TH or THETA | | Theta limits for jets. Do not |
| $`\theta _{\mathrm{min}}(1)`$,โฆ,$`\theta _{\mathrm{max}}(3)`$ | 0,$`\pi `$ | also set Y. |
| SUGRHN | | SUGRA see-saw $`\nu `$-effect |
| $`m_{\nu _\tau }`$,$`M_N`$,$`A_n`$,$`m_{\stackrel{~}{\nu }_R}`$ | $`0,1E20,0,0`$ | nu-mass, int. scale, |
| | | GUT scale nu SSB terms |
| THW | | Theta limits for W. Do not |
| $`\theta _{\mathrm{min}}`$,$`\theta _{\mathrm{max}}`$ | 0,$`\pi `$ | also set YW. |
| TCMASS | | Technicolor mass and width. |
| $`m`$,$`\mathrm{\Gamma }`$ | 1000,100 | |
| TMASS | | t, y, and x quark masses. |
| $`m_t`$,$`m_y`$,$`m_x`$ | 180.,-1.,-1. | |
| WFUDGE | | Fudge factor for DRELLYAN |
| factor | 1.85 | evolution scale. |
| WMASS | | W and Z masses. See SIN2W. |
| $`M_W`$,$`M_Z`$ | 80.2, 91.19 | |
| WMMODE | | Decay modes for $`W^{}`$ in parton |
| โUPโ,โฆ,โTAU+โ | โALLโ | cascade. See JETTYPE. |
| WMODE1 | | ) |
| --- | --- | --- |
| โUPโ,โUBโ,โฆ | โALLโ | )Decay modes for WPAIR. |
| | | )Same code for quarks and |
| WMODE2 | | )leptons as JETTYPE. |
| โUPโ,โUBโ,โฆ | โALLโ | ) |
| WPMODE | | Decay modes for $`W^+`$ in parton |
| โUPโ,โฆ,โTAU+โ | โALLโ | cascade. See JETTYPE. |
| WTYPE | | Select W type: W+,W-,GM,Z0. |
| type<sub>1</sub>,type<sub>2</sub> | โGMโ,โZ0โ | Do not mix W+,W- and GM,Z0. |
| X | | Feynman x limits for jets. |
| $`x_{\mathrm{min}}(1)`$,โฆ,$`x_{\mathrm{max}}(3)`$ | $`1`$,1 | |
| XGEN | | Jet fragmentation, Peterson |
| a(1),โฆ,a(8) | .96,3,0,.8,.5,โฆ | with $`ฯต=a(n)/m^2`$, $`n=4`$-8. |
| XGENSS | | Fragmentation of GLSS, UPSS, |
| a(1),โฆ,a(7) | .5,.5,โฆ | etc. with $`ฯต=a(n)/m2`$ |
| XW | | Feynman x limits for W. |
| $`x_{\mathrm{min}}`$,$`x_{\mathrm{max}}`$ | $`1`$,1 | |
| Y | | Y limits for each jet. |
| $`y_{\mathrm{min}}(1)`$,โฆ,$`y_{\mathrm{max}}(3)`$ | from PT | Do not also set TH. |
| YW | | Y limits for W. |
| $`y_{\mathrm{min}}`$,$`y_{\mathrm{max}}`$ | from QTW,QMW | Do not set both YW and THW. |
| Z0MODE | | Decay modes for $`Z^0`$ in parton |
| โUPโ,โฆ,โTAU+โ | โALLโ | cascade. See JETTYPE. |
### 6.3 Kinematic and Parton-type Parameters
While the TWOJET PT limits and the DRELLYAN QMW limits are formally optional parameters, they are set by default to be fractions of $`\sqrt{s}`$. Thus, for example, the parameter file
```
DEFAULT TWOJET JOB
14000,100,1,100/
TWOJET
END
STOP
```
will execute, but it will generate jets between 5% and 20% of $`\sqrt{s}`$, which is probably not what is wanted. Similarly, the parameter file
```
DEFAULT DRELLYAN JOB
14000,100,1,100/
DRELLYAN
END
STOP
```
will generate $`\gamma +Z`$ events with masses between 5% and 20% of $`\sqrt{s}`$, not masses around the $`Z`$ mass, and transverse momenta between $`1\mathrm{GeV}`$ and 2.5% of $`\sqrt{s}`$.
Normally the user should set PT limits for TWOJET, PHOTON, WPAIR, SUPERSYM, and WHIGGS events and QMW and QTW limits for DRELLYAN, HIGGS, and TCOLOR events. If these limits are not set, they will be selected as fractions of $`E_{\mathrm{cm}}`$. This can give nonsense. For TWOJET the $`p_t`$ range should usually be less than about a factor of two except for $`b`$ and $`t`$ jets at low $`p_t`$ to produce uniform statistics. For $`W^+`$, $`W^{}`$, or $`Z^0`$ events or for Higgs events the QMW (QMH) range should usually include the mass. But one can select different limits to study, e.g., virtual $`W`$ production or the effect of a lighter or heavier Higgs on WW scattering. If only $`t`$ decays are selected, then the lower QMW limit must be above the $`t`$ threshold. For standard Drell-Yan events QTW should be fixed to zero,
```
QTW
0/
```
Transverse momenta will then be generated by initial state gluon radiation. A range of QTW can also be given. For SUPERSYM either the masses and decay modes should be specified, or the MSSM, SUGRA, GMSB, or AMSB parameters should be given. For fourth generation quarks it is necessary to specify the quark masses.
Note that if the limits given cover too large a kinematic range, the program can become very inefficient, since it makes a fit to the cross section over the specified range. NTRIES has to be increased if narrow limits are set for X, XW or for jet 1 and jet 2 parameters in DRELLYAN events. For larger ranges several runs can be combined together using the integrated cross section per event SIGF/NEVENT as the weight. This cross section is calculated for each run by Monte Carlo integration over the specified kinematic limits and is printed at the end of the run. It is corrected for JETTYPEi, WTYPE, and WMODEi selections; it cannot be corrected for branching ratios of forced decays or for WPMODE, WMMODE, or Z0MODE selections, since these can affect an arbitrary number of particles.
To generate events over a large range, it is much more efficient to combine several runs. This is facilitated by using the special job title SAME as described above. Note that SAME cannot be used to combine standard DRELLYAN events (QTW fixed equal to 0) and DRELLYAN events with nonzero QTW.
The cross sections for multiparton final states in general have infrared and collinear singularities. To obtain sensible results, it is in general essential to set limits both on the $`p_T`$ of each final parton using PT and on the mass of each pair of partons using MIJLIM. The default lower limits are all $`1\mathrm{GeV}`$. Using these default limits without thought will likely give absurd results.
For TWOJET, DRELLYAN, and most other processes, the JETTYPEi and WTYPEi keywords should be used to select the subprocesses to be included. For $`e^+e^{}W^+W^{}`$, $`Z^0Z^0`$, use FORCE and FORCE1 instead of WMODEi to select the $`W`$ decay modes. Note that these do not change the calculated cross section. (In the E+E- process, the $`W`$ and $`Z`$ decays are currently treated as particle decays, whereas in the WPAIR and HIGGS processes they are treated as $`24`$ parton processes.)
For HIGGS with $`W^+W^{}`$ or $`Z^0Z^0`$ decays allowed it is generally necessary to set PT limits for the Wโs, e.g.
```
PT
50,20000,50,20000/
```
If this is not done, then the default lower limit of 1 GeV is used, and the $`t`$-channel exchanges will dominate, as they should in the effective $`W`$ approximation. Depending on the other parameters, the program may fail to generate an event in NTRIES tries.
### 6.4 SUSY Parameters
SUPERSYM (SUSY) by default generates just gluinos and squarks in pairs. There are no default masses or decay modes. Masses can be set using GAUGINO, SQUARK, SLEPTON, and HMASSES. Decay modes can be specified with FORCE or by modifying the decay table. Left and right squarks are distinguished but assumed to be degenerate, except for stops. Since version 7.11, types must be selected with JETTYPEi using the supersymmetric names, e.g.
```
JETTYPE1
โGLSSโ,โUPSSLโ,โUPSSRโ/
```
Use of the corresponding standard model names, e.g.
```
JETTYPE1
โGLโ,โUPโ/
```
and generation of pure photinos, winos, and zinos are no longer supported.
If MSSMA, MSSMB and MSSMC are given, then the specified parameters are used to calculate all the masses and decay modes with the ISASUSY package assuming the minimal supersymmetric extension of the standard model (MSSM). There are no default values, so you must specify values for each MSSMi, i=A-C. MSSMD can optionally be used to set the second generation squark and slepton parameters; if it is omitted, then the first generation ones are used. MSSME can optionally be used to set the U(1) and SU(2) gaugino masses; if it is omitted, then the grand unification values are used. The parameters and the use of the MSSM is preserved if the title is SAME. FORCE can be used to override the calculated branching ratios.
The MSSM option also generates charginos and neutralinos with cross sections based on the MSSM mixing angles in addition to squarks and sleptons. These can be selected with JETTYPEi; the complete list of supersymmetric options is:
```
โGLSSโ,
โUPSSLโ,โUBSSLโ,โDNSSLโ,โDBSSLโ,โSTSSLโ,โSBSSLโ,โCHSSLโ,โCBSSLโ,
โBTSS1โ,โBBSS1โ,โTPSS1โ,โTBSS1โ,
โUPSSRโ,โUBSSRโ,โDNSSRโ,โDBSSRโ,โSTSSRโ,โSBSSRโ,โCHSSRโ,โCBSSRโ,
โBTSS2โ,โBBSS2โ,โTPSS2โ,โTBSS2โ,
โW1SS+โ,โW1SS-โ,โW2SS+โ,โW2SS-โ,โZ1SSโ,โZ2SSโ,โZ3SSโ,โZ4SSโ,
โNUELโ,โANUELโ,โEL-โ,โEL+โ,โNUMLโ,โANUMLโ,MUL-โ,โMUL+โ,โNUTLโ,
โANUTLโ,โTAU1-โ,โTAU1+โ,โER-โ,โER+โ,โMUR-โ,โMUR+โ,โTAU2-โ,โTAU2+โ,
โZ0โ,โHL0โ,โHH0โ,โHA0โ,โH+โ,โH-โ,
โSQUARKSโ,โGAUGINOSโ,โSLEPTONSโ,โALLโ.
```
Note that mixing between $`L`$ and $`R`$ stop states results in 1 (light) and 2 (heavy) stop, sbottom and stau eigenstates, which depend on the input parameters of left- and right- scalar masses, plus $`A`$ terms, $`\mu `$ and $`\mathrm{tan}\beta `$. The last four JETTYPEโs generate respectively all allowed combinations of squarks and antisquarks, all combinations of charginos and neutralinos, all combinations of sleptons and sneutrinos, and all SUSY particles.
For SUSY Higgs pair production or associated production in E+E-, select the appropriate JETTYPEโs, e.g.
```
JETTYPE1
โZ0โ/
JETTYPE2
โHL0โ/
```
As usual, this gives only half the cross section. For single production of neutral SUSY Higgs in $`pp`$ and $`\overline{p}p`$ reactions, use the HIGGS process together with the MSSMi, SUGRA, GMSB, or AMSB keywords. You must specify one and only one Higgs type using
```
HTYPE
โHL0โ or โHH0โ or โHA0โ/ <<<<< One only!
```
If no QMH range is given, one is calculated using $`M\pm 5\mathrm{\Gamma }`$ for the selected Higgs. Decays into quarks, leptons, gauge bosons, lighter Higgs bosons, and SUSY particles are generated using the on-shell branching ratios from ISASUSY. You can use JETTYPEi to select the allowed Higgs modes and WMODEi to select the allowed decays of W and Z bosons. Since heavy SUSY Higgs bosons couple weakly to W pairs, WW fusion and WW scattering are not included.
SUGRA can be used instead of MSSMi to generate MSSM decays with parameters determined from $`m_0`$, $`m_{1/2}`$, $`A_0`$, $`\mathrm{tan}\beta `$, and $`sgn\mu =\pm 1`$ in the minimal supergravity framework. The NUSUGi keywords can optionally be used to specify additional parameters for non-universal SUGRA models, while SUGRHN is used to specify the parameterf of an optional right-handed neutrino. Similarly, the GMSB keyword is used to specify the $`\mathrm{\Lambda }`$, $`M_m`$, $`N_5`$, $`\mathrm{tan}\beta `$, $`sgn\mu =\pm 1`$, and $`C_{\mathrm{grav}}`$ parameters of the minimal Gauge Mediated SUSY Breaking model. GMSB2 can optionally be used to specify additional parameters of non-minimal GMSB models. The AMSB keyword is used to specify $`m_0`$, $`m_{3/2}`$, $`\mathrm{tan}\beta `$, and $`sgn\mu `$ for the minimal Anomaly Mediated SUSY Breaking model. Note that $`m_{3/2}`$ is much larger than the weak scale, typically 50 TeV.
WHIGGS is used to generate $`W`$ plus neutral Higgs events. For the Standard Model the JETTYPE is `HIGGS`. If any of the SUSY models is specified, then the appropriate SUSY Higgs type should be used, most likely `HL0`. In either case WMODEi is used to specify the $`W`$ decay modes. The Higgs is treated as a particle; its decay modes can be set using FORCE.
### 6.5 Forced Decay Modes
The FORCE keyword requires special care. Its list must contain the numerical particle IDENT codes, e.g.
```
FORCE
140,130,-120/
```
The charge-conjugate mode is also forced for its antiparticle. Thus the above example forces both $`\overline{D}^0K^+\pi ^{}`$ and $`D^0K^{}\pi ^+`$. If only a specific decay is wanted one should use the FORCE1 command; e.g.
```
FORCE1
140,130,-120/
```
only forces $`\overline{D}^0K^+\pi ^{}`$.
To force a heavy quark decay one must generally separately force each hadron containing it. If the decay is into three leptons or quarks, then the real or virtual W propagator is inserted automatically. Since Version 7.30, top and fourth generation quarks are treated as particles and decayed directly rather than first being made into hadrons. Thus for example
```
FORCE1
6,-12,11,5/
```
forces all top quarks to decay into an positron, neutrino and a b-quark (which will be hadronized). For the physical top mass, the positron and neutrino will come from a real W. Note that forcing $`tW^+b`$ and $`W^+e^+\nu _e`$ does not give the same result; the first uses the correct $`VA`$ matrix element, while the second decays the $`W`$ according to phase space.
Forced modes included in the decay table or generated by ISASUSY will automatically be put into the correct order and will use the correct matrix element. Modes not listed in the decay table are allowed, but caution is advised because a wrong decay mode can cause an infinite loop or other unexpected effects.
FORCE (FORCE1) can be called at most 20 (40) times in any run plus all subsequent โSAMEโ runs. If it is called more than once for a given parent, all calls are listed, and the last call is used. Note that FORCE applies to particles only, but that for gamma, W+, W-, Z0 and supersymmetric particles the same IDENT codes are used both as jet types and as particles.
### 6.6 Parton Distributions
The default parton distributions are fit CTEQ3L from the CTEQ Collaboration using lowest order QCD. The CTEQ and the older EHLQ and Duke-Owens distributions can be selected using the STRUC keyword.
If PDFLIB support is enabled (see Section 4), then any of the distributions in the PDFLIB compilation by H. Plothow-Besch can be selected using the PDFLIB keyword and giving the proper parameters, which are identical to those described in the PDFLIB manual and are simply passed to the routine PDFSET. For example, to select fit 29 (CTEQ3L) by the CTEQ group, leaving all other parameters with their default values, use
```
PDFLIB
โCTEQโ,29D0/
```
Note that the fit-number and the other parameters are of type DOUBLE PRECISION (REAL on 64-bit machines). There is no internal passing of parameters except for those which control the printing of messages.
### 6.7 Multiparton Processes
For multiparton final states one should in general set limits on the total mass `MTOT` of the final state, on the minimum `PT` of each light parton, and on the minimum mass `MIMLIM` of each pair of light partons. Limits for `PT` are set in the ususal way. Limits for the mass $`M_{ij}`$ of partons $`i,j`$ are set using
```
MIJLIM
i,j,Mmin,Mmax
```
If $`i=j=0`$, the limit is applied to all jet pairs. For example the following parameter file generates `ZJJ` events at the LHC with a mimimum $`p_T`$ of $`20\mathrm{GeV}`$ and a minimum mass of $`20\mathrm{GeV}`$ for all jet pairs:
```
GENERATE ZJJ with PTMIN = 20 GEV AND MMIN = 20 GEV
14000,100,1,100/
ZJJ
PT
20,7000,20,7000,20,7000/
MIJLIM
0,0,20,7000/
MTOT
100,500/
NSIGMA
200/
NTRIES
10000/
END
STOP
```
The default lower limits for `PT` and `MIJLIM` are $`1\mathrm{GeV}`$. While these limits are sufficient to make the cross sections finite, they will in general not give physically sensible results. Thus, the user must think carefully about what limits should be set.
## 7 Output
The output tape or file contains three types of records. A beginning record is written by a call to ISAWBG before generating a set of events; an event record is written by a call to ISAWEV for each event; and an end record is written for each run by a call to ISAWND. These subroutines load the common blocks described below into a single
```
COMMON/ZEVEL/ZEVEL(1024)
```
and write it out when it is full. A subroutine RDTAPE, described in the next section, inverts this process so that the user can analyze the event.
ZEVEL is written out to TAPEj by a call to BUFOUT. For the CDC version IF = PAIRPAK is selected; BUFOUT first packs two words from ZEVEL into one word in
```
COMMON/ZVOUT/ZVOUT(512)
```
using subroutine PAIRPAK and then does a buffer out of ZVOUT to TAPEj. Typically at least two records are written per event. For all other computers IF=STDIO is selected, and ZEVEL is written out with a standard FORTRAN unformatted write.
### 7.1 Beginning Record
At the start of each run ISAWBG is called. It writes out the following common blocks:
```
COMMON/DYLIM/QMIN,QMAX,QTMIN,QTMAX,YWMIN,YWMAX,XWMIN,XWMAX,THWMIN,
2 THWMAX,PHWMIN,PHWMAX
3 ,SETLMQ(12)
SAVE /DYLIM/
LOGICAL SETLMQ
EQUIVALENCE(BLIM1(1),QMIN)
REAL QMIN,QMAX,QTMIN,QTMAX,YWMIN,YWMAX,XWMIN,XWMAX,THWMIN,
+ THWMAX,PHWMIN,PHWMAX,BLIM1(12)
```
| QMIN,QMAX | = | $`W`$ mass limits |
| --- | --- | --- |
| QTMIN,QTMAX | = | $`W`$ $`q_t`$ limits |
| YWMIN,YWMAX | = | $`W`$ $`\eta `$ rapidity limits |
| XWMIN,XWMAX | = | $`W`$ $`x_F`$ limits |
| THWMIN,THWMAX | = | $`W`$ $`\theta `$ limits |
| PHWMIN,PHWMAX | = | $`W`$ $`\varphi `$ limits |
```
COMMON/IDRUN/IDVER,IDG(2),IEVT,IEVGEN
SAVE /IDRUN/
INTEGER IDVER,IDG,IEVT,IEVGEN
```
| IDVER | = | program version |
| --- | --- | --- |
| IDG(1) | = | run date (10000$`\times `$month+100$`\times `$day+year) |
| IDG(2) | = | run time (10000$`\times `$hour+100$`\times `$minute+second) |
| IEVT | = | event number |
```
C Jet limits
INTEGER MXLIM
PARAMETER (MXLIM=8)
INTEGER MXLX12
PARAMETER (MXLX12=12*MXLIM)
COMMON/JETLIM/PMIN(MXLIM),PMAX(MXLIM),PTMIN(MXLIM),PTMAX(MXLIM),
$YJMIN(MXLIM),YJMAX(MXLIM),PHIMIN(MXLIM),PHIMAX(MXLIM),
$XJMIN(MXLIM),XJMAX(MXLIM),THMIN(MXLIM),THMAX(MXLIM),
$SETLMJ(12*MXLIM)
SAVE /JETLIM/
COMMON/FIXPAR/FIXP(MXLIM),FIXPT(MXLIM),FIXYJ(MXLIM),
$FIXPHI(MXLIM),FIXXJ(MXLIM),FIXQM,FIXQT,FIXYW,FIXXW,FIXPHW
SAVE /FIXPAR/
COMMON/SGNPAR/CTHS(2,MXLIM),THS(2,MXLIM),YJS(2,MXLIM),XJS(2,MXLIM)
SAVE /SGNPAR/
REAL PMIN,PMAX,PTMIN,PTMAX,YJMIN,YJMAX,PHIMIN,PHIMAX,XJMIN,
+ XJMAX,THMIN,THMAX,BLIMS(12*MXLIM),CTHS,THS,YJS,XJS
LOGICAL SETLMJ
LOGICAL FIXQM,FIXQT,FIXYW,FIXXW,FIXPHW
LOGICAL FIXP,FIXPT,FIXYJ,FIXPHI,FIXXJ
EQUIVALENCE(BLIMS(1),PMIN(1))
```
| PMIN,PMAX | = | jet momentum limits |
| --- | --- | --- |
| PTMIN,PTMAX | = | jet $`p_t`$ limits |
| YJMIN,YJMAX | = | jet $`\eta `$ rapidity limits |
| PHIMIN,PHIMAX | = | jet $`\varphi `$ limits |
| THMIN,THMAX | = | jet $`\theta `$ limits |
```
INTEGER MXKEYS
PARAMETER (MXKEYS=20)
COMMON/KEYS/IKEYS,KEYON,KEYS(MXKEYS)
COMMON/XKEYS/REAC
SAVE /KEYS/,/XKEYS/
LOGICAL KEYS
LOGICAL KEYON
CHARACTER*8 REAC
INTEGER IKEYS
```
| KEYON | = | normally TRUE, FALSE if no good reaction |
| --- | --- | --- |
| KEYS | = | TRUE if reaction I is chosen |
| | | 1 for TWOJET |
| | | 2 for E+E- |
| | | 3 for DRELLYAN |
| | | 4 for MINBIAS |
| | | 5 for SUPERSYM |
| | | 6 for WPAIR |
| REAC | = | character reaction code |
```
COMMON/PRIMAR/NJET,SCM,HALFE,ECM,IDIN(2),NEVENT,NTRIES,NSIGMA
SAVE /PRIMAR/
INTEGER NJET,IDIN,NEVENT,NTRIES,NSIGMA
REAL SCM,HALFE,ECM
```
| NJET | = | number of jets per event |
| --- | --- | --- |
| SCM | = | square of com energy |
| HALFE | = | beam energy |
| ECM | = | com energy |
| IDIN | = | ident code for initial beams |
| NEVENT | = | number of events to be generated |
| NTRIES | = | maximum number of tries for good jet parameters |
| NSIGMA | = | number of extra events to determine SIGF |
```
INTEGER MXGOQ,MXGOJ
PARAMETER (MXGOQ=85,MXGOJ=8)
COMMON/Q1Q2/GOQ(MXGOQ,MXGOJ),GOALL(MXGOJ),GODY(4),STDDY,
$GOWW(25,2),ALLWW(2),GOWMOD(25,MXGOJ)
SAVE /Q1Q2/
LOGICAL GOQ,GOALL,GODY,STDDY,GOWW,ALLWW,GOWMOD
```
| GOQ(I,K) | = | TRUE if quark type I allowed for jet k |
| --- | --- | --- |
| | | I = 1 2 3 4 5 6 7 8 9 10 11 12 13 |
| | | $``$ $`g`$ $`u`$ $`\overline{u}`$ $`d`$ $`\overline{d}`$ $`s`$ $`\overline{s}`$ $`c`$ $`\overline{c}`$ $`b`$ $`\overline{b}`$ $`t`$ $`\overline{t}`$ |
| | | I = 14 15 16 17 18 19 20 21 22 23 24 25 |
| | | $``$ $`\nu _e`$ $`\overline{\nu }_e`$ $`e^{}`$ $`e^+`$ $`\nu _\mu `$ $`\overline{\nu }_\mu `$ $`\mu ^{}`$ $`\mu ^+`$ $`\nu _\tau `$ $`\overline{\nu }_\tau `$ $`\tau ^{}`$ $`\tau ^+`$ |
| GOALL(K) | = | TRUE if all jet types allowed |
| GODY(I) | = | TRUE if $`W`$ type I is allowed. |
| I= 1 2 3 4 | | |
| GM W+ W- Z0 | | |
| STDDY | = | TRUE if standard DRELLYAN |
| GOWW(I,K) | = | TRUE if I is allowed in the decay of K for WPAIR. |
| ALLWW(K) | = | TRUE if all allowed in the decay of K for WPAIR. |
```
COMMON/QCDPAR/ALAM,ALAM2,CUTJET,ISTRUC
SAVE /QCDPAR/
INTEGER ISTRUC
REAL ALAM,ALAM2,CUTJET
```
| ALAM | = | QCD scale $`\mathrm{\Lambda }`$ |
| --- | --- | --- |
| ALAM2 | = | QCD scale $`\mathrm{\Lambda }^2`$ |
| CUTJET | = | cutoff for generating secondary partons |
| ISTRUC | = | 3 for Eichten (EHLQ), |
| | = | 4 for Duke (DO) |
| | = | 5 for CTEQ 2L |
| | = | 6 for CTEQ 3L |
| | = | $`999`$ for PDFLIB |
```
COMMON/QLMASS/AMLEP(100),NQLEP,NMES,NBARY
SAVE /QLMASS/
INTEGER NQLEP,NMES,NBARY
REAL AMLEP
```
| AMLEP(6:8) | = | $`t`$,$`y`$,$`x`$ masses, only elements written |
| --- | --- | --- |
### 7.2 Event Record
For each event ISAWEV is called. It writes out the following common blocks:
```
COMMON/FINAL/NKINF,SIGF,ALUM,ACCEPT,NRECS
SAVE /FINAL/
INTEGER NKINF,NRECS
REAL SIGF,ALUM,ACCEPT
```
| SIGF | = | integrated cross section, only element written |
| --- | --- | --- |
```
COMMON/IDRUN/IDVER,IDG(2),IEVT,IEVGEN
SAVE /IDRUN/
INTEGER IDVER,IDG,IEVT,IEVGEN
```
| IDVER | = | program version |
| --- | --- | --- |
| IDG | = | run identification |
| IEVT | = | event number |
```
COMMON/JETPAR/P(3),PT(3),YJ(3),PHI(3),XJ(3),TH(3),CTH(3),STH(3)
1 ,JETTYP(3),SHAT,THAT,UHAT,QSQ,X1,X2,PBEAM(2)
2 ,QMW,QW,QTW,YW,XW,THW,QTMW,PHIW,SHAT1,THAT1,UHAT1,JWTYP
3 ,ALFQSQ,CTHW,STHW,Q0W
4 ,INITYP(2),ISIGS,PBEAMS(5)
SAVE /JETPAR/
INTEGER JETTYP,JWTYP,INITYP,ISIGS
REAL P,PT,YJ,PHI,XJ,TH,CTH,STH,SHAT,THAT,UHAT,QSQ,X1,X2,
+ PBEAM,QMW,QW,QTW,YW,XW,THW,QTMW,PHIW,SHAT1,THAT1,UHAT1,
+ ALFQSQ,CTHW,STHW,Q0W,PBEAMS
```
| P | = | jet momentum $`|\stackrel{}{p}|`$ |
| --- | --- | --- |
| PT | = | jet $`p_t`$ |
| YJ | = | jet $`\eta `$ rapidity |
| PHI | = | jet $`\varphi `$ |
| XJ | = | jet $`x_F`$ |
| TH | = | jet $`\theta `$ |
| CTH | = | jet $`\mathrm{cos}(\theta )`$ |
| STH | = | jet $`\mathrm{sin}(\theta )`$ |
| JETTYP | = | jet type. The code is listed under /Q1Q2/ above |
| | | continuedโฆ |
| SHAT,THAT,UHAT | = | hard scattering $`\widehat{s}`$, $`\widehat{t}`$, $`\widehat{u}`$ |
| --- | --- | --- |
| QSQ | = | effective $`Q^2`$ |
| X1,X2 | = | initial parton $`x_F`$ |
| PBEAM | = | remaining beam momentum |
| QMW | = | $`W`$ mass |
| QW | = | $`W`$ momentum |
| QTW | = | $`W`$ transverse momentum |
| YW | = | $`W`$ rapidity |
| XW | = | $`W`$ $`x_F`$ |
| THW | = | $`W`$ $`\theta `$ |
| QTMW | = | $`\sqrt{q_{t,W}^2+Q^2}`$ |
| PHIW | = | $`W`$ $`\varphi `$ |
| SHAT1,THAT1,UHAT1 | = | invariants for $`W`$ decay |
| JWTYP | = | $`W`$ type. The code is listed under /Q1Q2/ above. |
| ALFQSQ | = | QCD coupling $`\alpha _s(Q^2)`$ |
| CTHW | = | $`W`$ $`\mathrm{cos}(\theta )`$ |
| STHW | = | $`W`$ $`\mathrm{sin}(\theta )`$ |
| Q0W | = | $`W`$ energy |
```
INTEGER MXJSET,JPACK
PARAMETER (MXJSET=400,JPACK=1000)
COMMON/JETSET/NJSET,PJSET(5,MXJSET),JORIG(MXJSET),JTYPE(MXJSET),
$JDCAY(MXJSET)
SAVE /JETSET/
INTEGER NJSET,JORIG,JTYPE,JDCAY
REAL PJSET
```
| NJSET | = | number of partons |
| --- | --- | --- |
| PJSET(1,I) | = | $`p_x`$ of parton I |
| PJSET(2,I) | = | $`p_y`$ of parton I |
| PJSET(3,I) | = | $`p_z`$ of parton I |
| PJSET(4,I) | = | $`p_0`$ of parton I |
| PJSET(5,I) | = | mass of parton I |
| JORIG(I) | = | JPACK\*JET+K if I is a decay product of K. |
| | | IF K=0 then I is a primary parton. |
| | | (JET = 1,2,3 for final jets.) |
| | | (JET = 11,12 for initial jets.) |
| JTYPE(I) | = | IDENT code for parton I |
| JDCAY(I) | = | JPACK\*K1+K2 if K1 and K2 are decay products of I. |
| | | If JDCAY(I)=0 then I is a final parton |
| MXJSET | = | dimension for /JETSET/ arrays. |
| JPACK | = | packing integer for /JETSET/ arrays. |
```
INTEGER MXSIGS,IOPAK
PARAMETER (MXSIGS=3000,IOPAK=100)
COMMON/JETSIG/SIGMA,SIGS(MXSIGS),NSIGS,INOUT(MXSIGS),SIGEVT
SAVE /JETSIG/
INTEGER NSIGS,INOUT
REAL SIGMA,SIGS,SIGEVT
```
| SIGMA | = | cross section summed over types |
| --- | --- | --- |
| SIGS(I) | = | cross section for reaction I (not written) |
| NSIGS | = | number of nonzero cross sections (not written) |
| INOUT(I) | = | packed partons for process I (not written) |
| MXSIGS | = | dimension for JETSIG arrays (not written) |
| SIGEVT | = | partial cross section for selected channel |
```
INTEGER MXPTCL,IPACK
PARAMETER (MXPTCL=4000,IPACK=10000)
COMMON/PARTCL/NPTCL,PPTCL(5,MXPTCL),IORIG(MXPTCL),IDENT(MXPTCL)
1,IDCAY(MXPTCL)
SAVE /PARTCL/
INTEGER NPTCL,IORIG,IDENT,IDCAY
REAL PPTCL
```
| NPTCL | = | number of particles |
| --- | --- | --- |
| PPTCL(1,I) | = | $`p_x`$ for particle I |
| PPTCL(2,I) | = | $`p_y`$ for particle I |
| PPTCL(3,I) | = | $`p_z`$ for particle I |
| PPTCL(4,I) | = | $`p_0`$ for particle I |
| PPTCL(5,I) | = | mass for particle I |
| IORIG(I) | = | IPACK\*JET+K if I is a decay product of K. |
| | = | -(IPACK\*JET+K) if I is a primary particle from |
| | | parton K in /JETSET/. |
| | = | 0 if I is a primary beam particle. |
| | | (JET = 1,2,3 for final jets.) |
| | | (JET = 11,12 for initial jets.) |
| IDENT(I) | = | IDENT code for particle I |
| IDCAY(I) | = | IPACK\*K1+K2 if decay products are K1-K2 inclusive. |
| | | If IDCAY(I)=0 then particle I is stable. |
| MXPTCL | = | dimension for /PARTCL/ arrays. |
| IPACK | = | packing integer for /PARTCL/ arrays. |
```
COMMON/PINITS/PINITS(5,2),IDINIT(2)
SAVE /PINITS/
INTEGER IDINIT
REAL PINITS
```
| PINITS(1,I) | = | $`p_x`$ for initial parton I |
| --- | --- | --- |
| PINITS(2,I) | = | $`p_y`$ for initial parton I |
| PINITS(3,I) | = | $`p_z`$ for initial parton I |
| PINITS(4,I) | = | $`p_0`$ for initial parton I |
| PINITS(5,I) | = | mass for initial parton I |
| IDINIT(I) | = | IDENT for initial parton I |
```
INTEGER MXJETS
PARAMETER (MXJETS=10)
COMMON/PJETS/PJETS(5,MXJETS),IDJETS(MXJETS),QWJET(5),IDENTW
$,PPAIR(5,4),IDPAIR(4),JPAIR(4),NPAIR,IFRAME(MXJETS)
SAVE /PJETS/
INTEGER IDJETS,IDENTW,IDPAIR,JPAIR,NPAIR,IFRAME
REAL PJETS,QWJET,PPAIR
```
| PJETS(1,I) | = | $`p_x`$ for jet I |
| --- | --- | --- |
| PJETS(2,I) | = | $`p_y`$ for jet I |
| PJETS(3,I) | = | $`p_z`$ for jet I |
| PJETS(4,I) | = | $`p_0`$ for jet I |
| PJETS(5,I) | = | mass for jet I |
| IDJETS(I) | = | IDENT code for jet I |
| QWJET(1) | = | $`p_x`$ for $`W`$ |
| QWJET(2) | = | $`p_y`$ for $`W`$ |
| QWJET(3) | = | $`p_z`$ for $`W`$ |
| QWJET(4) | = | $`p_0`$ for $`W`$ |
| QWJET(5) | = | mass for $`W`$ |
| IDENTW | = | IDENT CODE for $`W`$ |
| PPAIR(1,I) | = | $`p_x`$ for WPAIR decay product I |
| PPAIR(2,I) | = | $`p_y`$ for WPAIR decay product I |
| PPAIR(3,I) | = | $`p_z`$ for WPAIR decay product I |
| PPAIR(4,I) | = | $`p_0`$ for WPAIR decay product I |
| PPAIR(5,I) | = | mass for WPAIR decay product I |
| IDPAIR(I) | = | IDENT code for WPAIR product I |
| JPAIR(I) | = | JETTYPE code for WPAIR product I |
| NPAIR | = | 2 for $`W^\pm \gamma `$ events, 4 for $`WW`$ events |
```
COMMON/TOTALS/NKINPT,NWGEN,NKEEP,SUMWT,WT
SAVE /TOTALS/
INTEGER NKINPT,NWGEN,NKEEP
REAL SUMWT,WT
```
| NKINPT | = | number of kinematic points generated. |
| --- | --- | --- |
| NWGEN | = | number of W+jet events accepted. |
| NKEEP | = | number of events kept. |
| SUMWT | = | sum of weighted cross sections. |
| WT | = | current weight. (SIGMA$`\times `$WT = event weight.) |
```
COMMON/WSIG/SIGLLQ
SAVE /WSIG/
REAL SIGLLQ
```
| SIGLLQ | = | cross section for $`W`$ decay. |
| --- | --- | --- |
Of course irrelevant common blocks such as /WSIG/ for TWOJET events are not written out.
### 7.3 End Record
At the end of a set ISAWND is called. It writes out the following common block:
```
COMMON/FINAL/NKINF,SIGF,ALUM,ACCEPT,NRECS
SAVE /FINAL/
INTEGER NKINF,NRECS
REAL SIGF,ALUM,ACCEPT
```
| NKINF | = | number of points generated to calculate SIGF |
| --- | --- | --- |
| SIGF | = | integrated cross section for this run |
| ALUM | = | equivalent luminosity for this run |
| ACCEPT | = | ratio of events kept over events generated |
| NRECS | = | number of physical records for this run |
Events within a given run have uniform weight. Separate runs can be combined together using SIGF/NEVENT as the weight per event. This gives a true cross section in mb units.
The user can replace subroutines ISAWBG, ISAWEV, and ISAWND to write out the events in a different format or to update histograms using HBOOK or any similar package.
## 8 File Reading
The FORTRAN instruction
```
CALL RDTAPE(IDEV,IFL)
```
will read a beginning record, an end record or an event (which can be more than one record). IDEV is the tape number and
```
IFL=0 for a good read,
IFL=-1 for an end of file.
```
The information is restored to the common blocks described above. The type of record is contained in
```
COMMON/RECTP/IRECTP,IREC
SAVE /RECTP/
INTEGER IRECTP,IREC
```
| IRECTP | = | 100 for an event record |
| --- | --- | --- |
| IRECTP | = | 200 for a beginning record |
| IRECTP | = | 300 for an end record |
| IREC | = | no. of physical records in event record, 0 otherwise |
The parton momenta from the primary hard scattering are contained in /PJETS/. The parton momenta generated by the QCD cascade are contained in /JETSET/. The hadron momenta both from the QCD jets and from the beam jets are contained in /PARTCL/. The final hadron momenta and the associated pointers should be used to calculate the jet momenta, since they are changed both by the QCD cascade and by hadronization. Particles with IDCAY=0 are stable, while the others are resonances.
The weight per event needed to produce a weighted histogram in millibarn units is SIGF/NEVENT. The integrated cross section SIGF is calculated by Monte Carlo integration during the run for the given kinematic limits and JETTYPE, WTYPE, and WMODE selections. Any of three methods can be used to find the value of SIGF:
(1) The current value, which is written out with each event, can be used. To prevent enormous fluctuations at the beginning of a run, NSIGMA extra primary parton events are generated first. The default value, NSIGMA = 20, gives negligible overhead but may not be large enough for good accuracy.
(2) The value SIGF calculated with the full statistics of the run can be obtained by reading through the tape until an end record (IRECTP=300) is found. After SIGF is saved with a different name, the first event record for the run can be found by backspacing the tape NRECS times.
(3) Unweighted histograms can be made for the run and the weight added after the end record is found. An implementation of this using special features of HBOOK is contained in ISAPLT.
The functions AMASS(IDENT), CHARGE(IDENT), and LABEL(IDENT) are available to determine the mass, charge, and character label in A8 format. Subroutine FLAVOR returns the quark content of any hadron and may be useful to convert IDENT codes to other schemes. CALL PRTEVT(0) prints an event.
## 9 Decay Table
ISAJET uses an external table of decay modes. Particles can be put into the table in arbitrary order, but all modes for each particle must be grouped together. The table is rewound and read in before each run with a READ\* format. Beginning with Version 7.41, the decay table must begin with a comment of the form
```
โ ISAJET V7.41 11-JAN-1999 20:41:57โ
```
If this does not match the internal version number, a warning is printed. After this initial line, each entry must have the form
```
IDENT,MELEM,CBR,ID1,ID2,ID3,ID4,ID5/
```
where IDENT is the code for the parent particle, MELEM specifies the decay matrix element, CBR is the cumulative branching ratio, and ID1,โฆ,ID5 are the IDENT codes for the decay products. The currently defined values of MELEM are:
| MELEM | Matrix Element |
| --- | --- |
| 0 | Phase Space |
| 1 | Dalitz decay |
| 2 | $`\omega /\varphi `$ decay |
| 3 | $`VA`$ decay |
| 4 | top decay: $`VA`$ plus $`W`$ propagator |
| 5 | $`\tau \mathrm{}\nu \overline{\nu }`$ |
| 6 | $`\tau \nu \pi `$, $`\nu K`$ |
| 7 | $`\tau \nu \rho `$, $`\nu a_1`$ |
The parent IDENT must be positive; the charge conjugate mode is used for the antiparticle. The values of CBR must of course be positive and monotonically increasing for each mode, with the last value being 1.00 for each parent IDENT. The last parent IDENT code must be zero. Care should be taken in adding new modes, since there is no checking for validity. In some cases order is important; note in particular that quarks and gluons must always appear last so that they can be removed and fragmented into hadrons.
The format of the decay table for Versions 7.41 and later is incompatible with that for Versions 7.40 and earlier. Using an obsolete decay table will produce incorrect results.
The decay table is contained in patch ISADECAY.
## 10 IDENT Codes
ISAJET uses a numerical ident code for particle types. Quarks and leptons are numbered in order of mass:
```
UP = 1 NUE = 11
DN = 2 E- = 12
ST = 3 NUM = 13
CH = 4 MU- = 14
BT = 5 NUT = 15
TP = 6 TAU- = 16
```
with a negative sign for antiparticles. Arbitrary conventions are:
```
GL = 9
GM = 10
KS = 20
KL =-20
W+ = 80
Z0 = 90
```
The supersymmetric particle IDENT codes distinguish between the partners of left and right handed fermions and include the Higgs sector of the minimal supersymmetric model:
```
UPSSL ... TPSS1 = 21 ... 26
NUEL ... TAU1- = 31 ... 36
UPSSR ... TPSS2 = 41 ... 46
NUER ... TAU2- = 51 ... 56
GLSS = 29
Z1SS = 30 Z2SS = 40
Z3SS = 50 Z4SS = 60
W1SS+ = 39 W2SS+ = 49
HL0 = 82 HH0 = 83
HA0 = 84 H+ = 86
```
Finally, the gravitino and graviton are
```
GVSS = 91 GRAV = 92
```
The same symbol is used for the graviton and its (possible) Kaluza-Klein excitations.
The code for a meson is a compound integer +-JKL, where J.LE.K are the quarks and L is the spin. The sign is for the J quark. Glueball IDENT codes have not been selected, but the choice GL=9 clearly allows 990, 9990, etc. Flavor singlet mesons are ordered by mass,
```
PI0 = 110
ETA = 220
ETAP = 330
ETAC = 440
```
which is natural for the heavy quarks. Similarly, the code for a baryon is a compound integer +-IJKL formed from the three quarks I,J,K and a spin label L=0,1. The code for a diquark is +-IJ00. Additional states are distinguished by a fifth integer, e.g.,
```
A1+ = 10121
```
These and a few J=2 mesons are used in some of the B decays.
A routine PRTLST is provided to print out a complete list of valid IDENT codes and associated information. The usage is CALL PRTLST(LUN, AMY, AMX) where LUN is the unit number and AMY and AMX are the masses of the Y and X quarks respectively. This routine should be linked with the ISAJET library and with ALDATA.
The complete list of ident codes follows. (Hadrons containing $`t`$ quarks are defined but are no longer listed since the $`t`$ quark is treated as a particle.)
```
IDENT LABEL MASS CHARGE
1 UP .30000E+00 .67
-1 UB .30000E+00 -.67
2 DN .30000E+00 -.33
-2 DB .30000E+00 .33
3 ST .50000E+00 -.33
-3 SB .50000E+00 .33
4 CH .16000E+01 .67
-4 CB .16000E+01 -.67
5 BT .49000E+01 -.33
-5 BB .49000E+01 .33
6 TP .17500E+03 .67
-6 TB .17500E+03 -.67
9 GL 0. 0.00
10 GM 0. 0.00
11 NUE 0. 0.00
-11 ANUE 0. 0.00
12 E- .51100E-03 -1.00
-12 E+ .51100E-03 1.00
13 NUM 0. 0.00
-13 ANUM 0. 0.00
14 MU- .10566E+00 -1.00
-14 MU+ .10566E+00 1.00
15 NUT 0. 0.00
-15 ANUT 0. 0.00
16 TAU- .18070E+01 -1.00
-16 TAU+ .18070E+01 1.00
20 KS .49767E+00 0.00
-20 KL .49767E+00 0.00
21 UPSSL none 0.67
-21 UBSSL none -0.67
22 DNSSL none -0.33
-22 DBSSL none 0.33
23 STSSL none -0.33
23 SBSSL none 0.33
24 CHSSL none 0.67
-24 CBSSL none -0.67
25 BTSS1 none -0.33
-25 BBSS1 none 0.33
26 TPSS1 none 0.67
-26 TBSS1 none -0.67
29 GLSS none 0.00
30 Z1SS none 0.00
31 NUEL none 0.00
-31 ANUEL none 0.00
32 EL- none -1.00
-32 EL+ none +1.00
33 NUML none 0.00
-33 ANUML none 0.00
34 MUL- none -1.00
-34 MUL+ none +1.00
35 NUTL none 0.00
-35 ANUTL none 0.00
36 TAU1- none -1.00
-36 TAU1+ none -1.00
39 W1SS+ none 1.00
-39 W1SS- none -1.00
40 Z2SS none 0.00
41 UPSSR none 0.67
-41 UBSSR none -0.67
42 DNSSR none -0.33
-42 DBSSR none 0.33
43 STSSR none -0.33
43 SBSSR none 0.33
44 CHSSR none 0.67
-44 CBSSR none -0.67
45 BTSS2 none -0.33
-45 BBSS2 none 0.33
46 TPSS2 none 0.67
-46 TBSS2 none -0.67
49 W2SS+ none 1.00
-49 W2SS- none -1.00
50 Z3SS none 0.00
51 NUER none 0.00
-51 ANUER none 0.00
52 ER- none -1.00
-52 ER+ none +1.00
53 NUMR none 0.00
-53 ANUMR none 0.00
54 MUR- none -1.00
-54 MUR+ none +1.00
55 NUTR none 0.00
-55 ANUTR none 0.00
56 TAU2- none -1.00
-56 TAU2+ none -1.00
60 Z4SS none 0.00
80 W+ .80200E+02 1.00
81 HIGGS .80200E+02 0.00
82 HL0 none 0.00
83 HH0 none 0.00
84 HA0 none 0.00
86 H+ none 1.00
90 Z0 .91190E+02 0.00
91 GVSS 0 0.00
92 GRAV 0 0.00
110 PI0 .13496E+00 0.00
120 PI+ .13957E+00 1.00
-120 PI- .13957E+00 -1.00
220 ETA .54745E+00 0.00
130 K+ .49367E+00 1.00
-130 K- .49367E+00 -1.00
230 K0 .49767E+00 0.00
-230 AK0 .49767E+00 0.00
330 ETAP .95760E+00 0.00
140 AD0 .18645E+01 0.00
-140 D0 .18645E+01 0.00
240 D- .18693E+01 -1.00
-240 D+ .18693E+01 1.00
340 F- .19688E+01 -1.00
-340 F+ .19688E+01 1.00
440 ETAC .29788E+01 0.00
150 UB. .51700E+01 1.00
-150 BU. .51700E+01 -1.00
250 DB. .51700E+01 0.00
-250 BD. .51700E+01 0.00
350 SB. .53700E+01 0.00
-350 BS. .53700E+01 0.00
450 CB. .64700E+01 1.00
-450 BC. .64700E+01 -1.00
550 BB. .97700E+01 0.00
111 RHO0 .76810E+00 0.00
121 RHO+ .76810E+00 1.00
-121 RHO- .76810E+00 -1.00
221 OMEG .78195E+00 0.00
131 K*+ .89159E+00 1.00
-131 K*- .89159E+00 -1.00
231 K*0 .89610E+00 0.00
-231 AK*0 .89610E+00 0.00
331 PHI .10194E+01 0.00
141 AD*0 .20071E+01 0.00
-141 D*0 .20071E+01 0.00
241 D*- .20101E+01 -1.00
-241 D*+ .20101E+01 1.00
341 F*- .21103E+01 -1.00
-341 F*+ .21103E+01 1.00
441 JPSI .30969E+01 0.00
151 UB* .52100E+01 1.00
-151 BU* .52100E+01 -1.00
251 DB* .52100E+01 0.00
-251 BD* .52100E+01 0.00
351 SB* .54100E+01 0.00
-351 BS* .54100E+01 0.00
451 CB* .65100E+01 1.00
-451 BC* .65100E+01 -1.00
551 UPSL .98100E+01 0.00
112 F2 .12750E+01 0.00
132 K2*+ .14254E+01 1.00
-132 K2*- .14254E+01 -1.00
232 K2*0 .14324E+01 0.00
-232 AK2*0 .14324E+01 0.00
10110 F0 .98000E+00 0.00
10111 A10 .12300E+01 0.00
10121 A1+ .12300E+01 1.00
-10121 A1- .12300E+01 -1.00
10131 K1+ .12730E+01 1.00
-10131 K1- .12730E+01 -1.00
10231 K10 .12730E+01 0.00
-10231 AK10 .12730E+01 0.00
30131 K1*+ .14120E+01 1.00
-30131 K1*- .14120E+01 -1.00
30231 K1*0 .14120E+01 0.00
-30231 AK1*0 .14120E+01 0.00
10441 PSI(2S) .36860E+01 0.00
20440 CHI0 .34151E+01 0.00
20441 CHI1 .35105E+01 0.00
20442 CHI2 .35662E+01 0.00
1120 P .93828E+00 1.00
-1120 AP .93828E+00 -1.00
1220 N .93957E+00 0.00
-1220 AN .93957E+00 0.00
1130 S+ .11894E+01 1.00
-1130 AS- .11894E+01 -1.00
1230 S0 .11925E+01 0.00
-1230 AS0 .11925E+01 0.00
2130 L .11156E+01 0.00
-2130 AL .11156E+01 0.00
2230 S- .11974E+01 -1.00
-2230 AS+ .11974E+01 1.00
1330 XI0 .13149E+01 0.00
-1330 AXI0 .13149E+01 0.00
2330 XI- .13213E+01 -1.00
-2330 AXI+ .13213E+01 1.00
1140 SC++ .24527E+01 2.00
-1140 ASC-- .24527E+01 -2.00
1240 SC+ .24529E+01 1.00
-1240 ASC- .24529E+01 -1.00
2140 LC+ .22849E+01 1.00
-2140 ALC- .22849E+01 -1.00
2240 SC0 .24525E+01 0.00
-2240 ASC0 .24525E+01 0.00
1340 USC. .25000E+01 1.00
-1340 AUSC. .25000E+01 -1.00
3140 SUC. .24000E+01 1.00
-3140 ASUC. .24000E+01 -1.00
2340 DSC. .25000E+01 0.00
-2340 ADSC. .25000E+01 0.00
3240 SDC. .24000E+01 0.00
-3240 ASDC. .24000E+01 0.00
3340 SSC. .26000E+01 0.00
-3340 ASSC. .26000E+01 0.00
1440 UCC. .35500E+01 2.00
-1440 AUCC. .35500E+01 -2.00
2440 DCC. .35500E+01 1.00
-2440 ADCC. .35500E+01 -1.00
3440 SCC. .37000E+01 1.00
-3440 ASCC. .37000E+01 -1.00
1150 UUB. .54700E+01 1.00
-1150 AUUB. .54700E+01 -1.00
1250 UDB. .54700E+01 0.00
-1250 AUDB. .54700E+01 0.00
2150 DUB. .54700E+01 0.00
-2150 ADUB. .54700E+01 0.00
2250 DDB. .54700E+01 -1.00
-2250 ADDB. .54700E+01 1.00
1350 USB. .56700E+01 0.00
-1350 AUSB. .56700E+01 0.00
3150 SUB. .56700E+01 0.00
-3150 ASUB. .56700E+01 0.00
2350 DSB. .56700E+01 -1.00
-2350 ADSB. .56700E+01 1.00
3250 SDB. .56700E+01 -1.00
-3250 ASDB. .56700E+01 1.00
3350 SSB. .58700E+01 -1.00
-3350 ASSB. .58700E+01 1.00
1450 UCB. .67700E+01 1.00
-1450 AUCB. .67700E+01 -1.00
4150 CUB. .67700E+01 1.00
-4150 ACUB. .67700E+01 -1.00
2450 DCB. .67700E+01 0.00
-2450 ADCB. .67700E+01 0.00
4250 CDB. .67700E+01 0.00
-4250 ACDB. .67700E+01 0.00
3450 SCB. .69700E+01 0.00
-3450 ASCB. .69700E+01 0.00
4350 CSB. .69700E+01 0.00
-4350 ACSB. .69700E+01 0.00
4450 CCB. .80700E+01 1.00
-4450 ACCB. .80700E+01 -1.00
1550 UBB. .10070E+02 0.00
-1550 AUBB. .10070E+02 0.00
2550 DBB. .10070E+02 -1.00
-2550 ADBB. .10070E+02 1.00
3550 SBB. .10270E+02 -1.00
-3550 ASBB. .10270E+02 1.00
4550 CBB. .11370E+02 0.00
-4550 ACBB. .11370E+02 0.00
1111 DL++ .12320E+01 2.00
-1111 ADL-- .12320E+01 -2.00
1121 DL+ .12320E+01 1.00
-1121 ADL- .12320E+01 -1.00
1221 DL0 .12320E+01 0.00
-1221 ADL0 .12320E+01 0.00
2221 DL- .12320E+01 -1.00
-2221 ADL+ .12320E+01 1.00
1131 S*+ .13823E+01 1.00
-1131 AS*- .13823E+01 -1.00
1231 S*0 .13820E+01 0.00
-1231 AS*0 .13820E+01 0.00
2231 S*- .13875E+01 -1.00
-2231 AS*+ .13875E+01 1.00
1331 XI*0 .15318E+01 0.00
-1331 AXI*0 .15318E+01 0.00
2331 XI*- .15350E+01 -1.00
-2331 AXI*+ .15350E+01 1.00
3331 OM- .16722E+01 -1.00
-3331 AOM+ .16722E+01 1.00
1141 UUC* .26300E+01 2.00
-1141 AUUC* .26300E+01 -2.00
1241 UDC* .26300E+01 1.00
-1241 AUDC* .26300E+01 -1.00
2241 DDC* .26300E+01 0.00
-2241 ADDC* .26300E+01 0.00
1341 USC* .27000E+01 1.00
-1341 AUSC* .27000E+01 -1.00
2341 DSC* .27000E+01 0.00
-2341 ADSC* .27000E+01 0.00
3341 SSC* .28000E+01 0.00
-3341 ASSC* .28000E+01 0.00
1441 UCC* .37500E+01 2.00
-1441 AUCC* .37500E+01 -2.00
2441 DCC* .37500E+01 1.00
-2441 ADCC* .37500E+01 -1.00
3441 SCC* .39000E+01 1.00
-3441 ASCC* .39000E+01 -1.00
4441 CCC* .48000E+01 2.00
-4441 ACCC* .48000E+01 -2.00
1151 UUB* .55100E+01 1.00
-1151 AUUB* .55100E+01 -1.00
1251 UDB* .55100E+01 0.00
-1251 AUDB* .55100E+01 0.00
2251 DDB* .55100E+01 -1.00
-2251 ADDB* .55100E+01 1.00
1351 USB* .57100E+01 0.00
-1351 AUSB* .57100E+01 0.00
2351 DSB* .57100E+01 -1.00
-2351 ADSB* .57100E+01 1.00
3351 SSB* .59100E+01 -1.00
-3351 ASSB* .59100E+01 1.00
1451 UCB* .68100E+01 1.00
-1451 AUCB* .68100E+01 -1.00
2451 DCB* .68100E+01 0.00
-2451 ADCB* .68100E+01 0.00
3451 SCB* .70100E+01 0.00
-3451 ASCB* .70100E+01 0.00
4451 CCB* .81100E+01 1.00
-4451 ACCB* .81100E+01 -1.00
1551 UBB* .10110E+02 0.00
-1551 AUBB* .10110E+02 0.00
2551 DBB* .10110E+02 -1.00
-2551 ADBB* .10110E+02 1.00
3551 SBB* .10310E+02 -1.00
-3551 ASBB* .10310E+02 1.00
4551 CBB* .11410E+02 0.00
-4551 ACBB* .11410E+02 0.00
5551 BBB* .14710E+02 -1.00
-5551 ABBB* .14710E+02 1.00
1100 UU0. .60000E+00 0.67
-1100 AUU0. .60000E+00 -0.67
1200 UD0. .60000E+00 0.33
-1200 AUD0. .60000E+00 -0.33
2200 DD0. .60000E+00 -0.67
-2200 ADD0. .60000E+00 0.67
1300 US0. .80000E+00 0.33
-1300 AUS0. .80000E+00 -0.33
2300 DS0. .80000E+00 -0.67
-2300 ADS0. .80000E+00 0.67
3300 SS0. .10000E+01 -0.67
-3300 ASS0. .10000E+01 0.67
1400 UC0. .19000E+01 1.33
-1400 AUC0. .19000E+01 -1.33
2400 DC0. .19000E+01 0.33
-2400 ADC0. .19000E+01 -0.33
3400 SC0. .21000E+01 0.33
-3400 ASC0. .21000E+01 -0.33
4400 CC0. .32000E+01 1.33
-4400 ACC0. .32000E+01 -1.33
1500 UB0. .49000E+01 0.33
-1500 AUB0. .49000E+01 -0.33
2500 DB0. .49000E+01 -0.67
-2500 ADB0. .49000E+01 0.67
3500 SB0. .51000E+01 -0.67
-3500 ASB0. .51000E+01 0.67
4500 CB0. .65000E+01 0.33
-4500 ACB0. .65000E+01 -0.33
5500 BB0. .98000E+01 -0.67
-5500 ABB0. .98000E+01 0.67
```
## 11 Higher Order Processes
Higher order processes can be generated either by the QCD evolution or by supplying partons from an external generator.
Frequently it is interesting to generate higher-order processes with a particular branching in the QCD evolution or with a particular particle or group of particles being produced from the fragmentation. Examples include
1. Branching of jets into heavy quarks (e.g., $`gb+\overline{b}`$);
2. Decay of such a heavy quark into a lepton or neutrino;
3. Radiation of a photon, $`W`$, or $`Z`$ from a jet.
It is important to realize that all of the cross sections and the QCD evolution in ISAJET are based on leading-log QCD, so generating such processes does not give the correct higher order QCD cross sections or โK factorsโ, even though it may produce better agreement with them in some cases.
ISAJET does produce events with particular topologies which in many cases are the most important effect of higher order processes. In the heavy quark example, the lowest order process
$$g+gQ+\overline{Q}$$
produces back-to-back heavy quark pairs, whereas the splitting process
$$g+gg+g,gQ+\overline{Q}$$
produces collinear pairs. Such collinear pairs are essential to obtain agreement with experimental data on $`b\overline{b}`$ production, and they often are the dominant background for processes of interest.
Branchings such as the emission of a heavy quark pair, a photon, or a $`W^\pm `$ or $`Z^0`$ are rare, and since they may occur at any step in the evolution, one cannot force them to occur. Therefore, generation of such events is very slow. M. Della Negra (UA1) suggested first doing $`n_1`$ QCD evolutions for each hard scattering and rejecting events without the desired partons, then doing $`n_2`$ fragmentations for each successful evolution. This generates the equivalent of $`n_1n_2`$ events for each hard scattering, so the cross section must be divided by $`n_1n_2`$. This algorithm can speed up the generation of $`gb+\overline{b}`$ splitting by a factor of ten for $`n_1=n_2=10`$.
Since the evolution and fragmentation steps are executed $`n_1n_2`$ times even if good events are found, a single hard scattering can lead to multiple events. This does not change the inclusive cross sections, but it does mean that the fluctuations may be larger than expected. Hence it is important to choose the numbers $`n_1`$ and $`n_2`$ carefully.
The following entities are used in ISAJET for generating events with multiple evolution and fragmentation:
`NEVENT`: The number of primary hard scatterings to be generated. Set as usual on the input line with the energy.
`SIGF`: The cross section for the selected hard scatterings divided by $`n_1\times n_2`$. Hence the correct weight is SIGF/NEVENT, just as for normal running. (The cross section printed at the end of a run does not contain this factor.)
`NEVOLVE`: The number $`n_1`$ of evolutions per hard scattering. This should never be set unless you supply a REJJET function. Do not confuse this with NOEVOLVE.
`NHADRON`: The number $`n_2`$ of fragmentations for a given evolution. This should never be set unless you supply a REJFRG function. Do not confuse this with NOHADRON.
`REJJET`: A logical function which if true causes the evolution to be rejected. The user must supply one to make the selections which he wants. The default always .FALSE. but includes an example as a comment.
`REJFRG`: A logical function which if true causes the fragmentation to be rejected. The user must supply one to make the selections which he wants. The default always .FALSE. but includes an example as a comment.
Note that one can also use function EDIT to make a final selection of the events. Of course ISAJET must be relinked if EDIT, REJJET or REJFRG is modified.
At the end of a run, the jet cross section, the cross section for the selected events, and the number and fraction of events selected are printed. The cross section SIGF stored internally is divided by $`n_1\times n_2`$ so that if the events are used to make histograms, then the correct weight per event is
```
SIGF/NEVENT
```
just as for normal events. Of course NEVENT now has a different meaning; it is in general larger than the number of events in the file but might be smaller if NEVOLVE and NHADRON are badly chosen.
NEVOLVE and NHADRON are set as parameters in the input. One wants to choose them to give better acceptance of the primary hard scatterings but not to give multiple events for one hard scattering. For lepton production from heavy quarks the values
```
NEVOLVE
10/
NHADRON
10/
```
seem appropriate, giving reasonable efficiency. For radiation of photons from jets, NEVOLVE can be somewhat larger but NHADRON should be one, and REJFRG should always return .FALSE., since the selection is just on the parton process, not on the hadronization.
The loops over evolutions and fragmentations are done inside of subroutine ISAEVT and are always executed the same number of times even though ISAEVT returns after each generated event. Logical flag OK signals a good event, and logical flag DONE signals that the run is finished. If you control the event generation loop yourself, you should make use of these flags as in the following extract from subroutine ISAJET:
```
ILOOP=0
101 CONTINUE
ILOOP=ILOOP+1
CALL ISAEVT(ILOOP,OK,DONE)
IF(OK) CALL ISAWEV
IF(.NOT.DONE) GO TO 101
```
Otherwise you may get the wrong weights.
It is possible to supply to ISAJET events with partons generated by some other program that may have more accurate matrix elements for higher order processes. Because any such calculation must involve cutoffs ISAJET assumes that the partons were generated imposing some $`R`$ cutoff, where $`R=\sqrt{\varphi ^2+\eta ^2}`$, and some $`E_t`$ cutoff. Given that information ISAJET will generate initial state radiation partons only below the Et cutoff and final state radiation inside the $`R`$ cutoff. The external partons can be supplied to ISAJET by calls to 2 subroutines. To initialize ISAJET for externally supplied partons, use
```
CALL INISAP(CMSE,REACTION,BEAMS,WZ,NDCAYS,DCAYS,ETMIN,RCONE,OK)
```
where the inputs are
| CMSE | = | center of mass energy |
| --- | --- | --- |
| REACTION | = | reaction (only TWOJET and DRELLYAN are |
| | | implemented so far) |
| BEAMS(2) | = | chose โP โ or โAPโ |
| ETMIN | = | minimum ET of supplied partons |
| RCONE | = | minimum cone (R) between supplied partons |
| WZ | = | option โWโ, โZโ, or โ โ no $`W`$โs or $`Z`$โs |
| NDCAYS | = | number of decay options (if 0, assume decay has |
| | | already been done) |
| DCAYS | = | list of particles W or Z can decay into |
and the output is
| OK | = | TRUE if initialization is possible |
| --- | --- | --- |
Then for each event use
```
CALL IPARTNS(NPRTNS,IDS,PRTNS,IDQ,WEIGHT,WZDK)
```
where the inputs are
| NPRTNS | = | number of partons, $`10`$ |
| --- | --- | --- |
| IDS(NPRTNS) | = | ids of final partons |
| PRTNS(4,NPRTNS) | = | parton 4 vectors |
| IDQ(2) | = | ids of initial partons |
| WEIGHT | = | weight |
| WZDK | = | if true last 2 partons are from W,Z decay |
Further QCD radiation is then generated consistent with ETMIN and RCONE, and the partons are fragmented into hadrons as usual. If RCONE is set to a value greater than 1.5 no cone restriction is applied during parton evolution.
## 12 ISASUSY: Decay Modes in the Minimal Supersymmetric Model
The code in patch ISASUSY of ISAJET calculates decay modes of supersymmetric particles based on the work of H. Baer, M. Bisset, M. Drees, D. Dzialo (Karatas), X. Tata, J. Woodside, and their collaborators. The calculations assume the minimal supersymmetric extension of the standard model. The user specifies the gluino mass, the pseudoscalar Higgs mass, the Higgsino mass parameter $`\mu `$, $`\mathrm{tan}\beta `$, the soft breaking masses for the first and third generation left-handed squark and slepton doublets and right-handed singlets, and the third generation mixing parameters $`A_t`$, $`A_b`$, and $`A_\tau `$. Supersymmetric grand unification is assumed by default in the chargino and neutralino mass matrices, although the user can optionally specify arbitrary $`U(1)`$ and $`SU(2)`$ gaugino masses at the weak scale. The first and second generations are assumed by default to be degenerate, but the user can optionally specify different values. These inputs are then used to calculate the mass eigenstates, mixings, and decay modes.
Most calculations are done at the tree level, but one-loop results for gluino loop decays, $`H\gamma \gamma `$ and $`Hgg`$, loop corrections to the Higgs mass spectrum and couplings, and leading-log QCD corrections to $`Hq\overline{q}`$ are included. The Higgs masses have been calculated using the effective potential approximation including both top and bottom Yukawa and mixing effects. Mike Bisset and Xerxes Tata have contributed the Higgs mass, couplings, and decay routines. Manuel Drees has calculated several of the three-body decays including the full Yukawa contribution, which is important for large tan(beta). Note that e+e- annihilation to SUSY particles and SUSY Higgs bosons have been included in ISAJET versions $`>7.11`$. ISAJET versions $`>7.22`$ include the large $`\mathrm{tan}\beta `$ solution as well as non-degenerate sfermion masses.
Other processes may be added in future versions as the physics interest warrants. Note that the details of the masses and the decay modes can be quite sensitive to choices of standard model parameters such as the QCD coupling ALFA3 and the quark masses. To change these, you must modify subroutine SSMSSM. By default, ALFA3=.12.
All the mass spectrum and branching ratio calculations in ISASUSY are performed by a call to subroutine SSMSSM. Effective with version 7.23, the calling sequence is
```
SUBROUTINE SSMSSM(XMG,XMU,XMHA,XTANB,XMQ1,XMDR,XMUR,
$XML1,XMER,XMQ2,XMSR,XMCR,XML2,XMMR,XMQ3,XMBR,XMTR,
$XML3,XMLR,XAT,XAB,XAL,XM1,XM2,XMT,IALLOW)
```
where the following are taken to be independent parameters:
| XMG | = | gluino mass |
| --- | --- | --- |
| XMU | = | $`\mu `$ = SUSY Higgs mass |
| | = | $`2m_1`$ of Baer et al. |
| XMHA | = | pseudo-scalar Higgs mass |
| XTANB | = | $`\mathrm{tan}\beta `$, ratio of vevโs |
| | = | $`1/R`$ (of old Baer-Tata notation). |
| XMQ1 | = | $`\stackrel{~}{q}_l`$ soft mass, 1st generation |
| --- | --- | --- |
| XMDR | = | $`\stackrel{~}{d}_r`$ mass, 1st generation |
| XMUR | = | $`\stackrel{~}{u}_r`$ mass, 1st generation |
| XML1 | = | $`\stackrel{~}{\mathrm{}}_l`$ mass, 1st generation |
| XMER | = | $`\stackrel{~}{e}_r`$ mass, 1st generation |
| XMQ2 | = | $`\stackrel{~}{q}_l`$ soft mass, 2nd generation |
| XMSR | = | $`\stackrel{~}{s}_r`$ mass, 2nd generation |
| XMCR | = | $`\stackrel{~}{c}_r`$ mass, 2nd generation |
| XML2 | = | $`\stackrel{~}{\mathrm{}}_l`$ mass, 2nd generation |
| XMMR | = | $`\stackrel{~}{\mu }_r`$ mass, 2nd generation |
| XMQ3 | = | $`\stackrel{~}{q}_l`$ soft mass, 3rd generation |
| XMBR | = | $`\stackrel{~}{b}_r`$ mass, 3rd generation |
| XMTR | = | $`\stackrel{~}{t}_r`$ mass, 3rd generation |
| XML3 | = | $`\stackrel{~}{\mathrm{}}_l`$ mass, 3rd generation |
| XMTR | = | $`\stackrel{~}{\tau }_r`$ mass, 3rd generation |
| XAT | = | stop trilinear term $`A_t`$ |
| XAB | = | sbottom trilinear term $`A_b`$ |
| XAL | = | stau trilinear term $`A_\tau `$ |
| XM1 | = | U(1) gaugino mass |
| | = | computed from XMG if ยฟ 1E19 |
| XM2 | = | SU(2) gaugino mass |
| | = | computed from XMG if ยฟ 1E19 |
| XMT | = | top quark mass |
The variable IALLOW is returned:
| IALLOW | = | 1 if Z1SS is not LSP, 0 otherwise |
| --- | --- | --- |
All variables are of type REAL except IALLOW, which is INTEGER, and all masses are in GeV. The notation is taken to correspond to that of Haber and Kane, although the Tata Lagrangian is used internally. All other standard model parameters are hard wired in this subroutine; they are not obtained from the rest of ISAJET. The theoretically favored range of these parameters is
$`50<M(\stackrel{~}{g})<2000\mathrm{GeV}`$
$`50<M(\stackrel{~}{q})<2000\mathrm{GeV}`$
$`50<M(\stackrel{~}{\mathrm{}})<2000\mathrm{GeV}`$
$`1000<\mu <1000\mathrm{GeV}`$
$`1<\mathrm{tan}\beta <m_t/m_b`$
$`M(t)175\mathrm{GeV}`$
$`50<M(A)<2000\mathrm{GeV}`$
$`M(\stackrel{~}{t}_l),M(t_r)<M(\stackrel{~}{q})`$
$`M(\stackrel{~}{b}_r)M(\stackrel{~}{q})`$
$`1000<A_t<1000\mathrm{GeV}`$
$`1000<A_b<1000\mathrm{GeV}`$
It is assumed that the lightest supersymmetric particle is the lightest neutralino $`\stackrel{~}{Z}_1`$, the lighter stau $`\stackrel{~}{\tau }_1`$, or the gravitino $`\stackrel{~}{G}`$ in GMSB models. Some choices of the above parameters may violate this assumption, yielding a light chargino or light stop squark lighter than $`\stackrel{~}{Z}_1`$. In such cases SSMSSM does not compute any branching ratios and returns IALLOW = 1.
SSMSSM does not check the parameters or resulting masses against existing experimental data. SSTEST provides a minimal test. This routine is called after SSMSSM by ISAJET and ISASUSY and prints suitable warning messages.
SSMSSM first calculates the other SUSY masses and mixings and puts them in the common block /SSPAR/:
```
C SUSY parameters
C AMGLSS = gluino mass
C AMULSS = up-left squark mass
C AMELSS = left-selectron mass
C AMERSS = right-slepton mass
C AMNiSS = sneutrino mass for generation i
C TWOM1 = Higgsino mass = - mu
C RV2V1 = ratio v2/v1 of vevโs
C AMTLSS,AMTRSS = left,right stop masses
C AMT1SS,AMT2SS = light,heavy stop masses
C AMBLSS,AMBRSS = left,right sbottom masses
C AMB1SS,AMB2SS = light,heavy sbottom masses
C AMLLSS,AMLRSS = left,right stau masses
C AML1SS,AML2SS = light,heavy stau masses
C AMZiSS = signed mass of Zi
C ZMIXSS = Zi mixing matrix
C AMWiSS = signed Wi mass
C GAMMAL,GAMMAR = Wi left, right mixing angles
C AMHL,AMHH,AMHA = neutral Higgs h0, H0, A0 masses
C AMHC = charged Higgs H+ mass
C ALFAH = Higgs mixing angle
C AAT = stop trilinear term
C THETAT = stop mixing angle
C AAB = sbottom trilinear term
C THETAB = sbottom mixing angle
C AAL = stau trilinear term
C THETAL = stau mixing angle
C AMGVSS = gravitino mass
COMMON/SSPAR/AMGLSS,AMULSS,AMURSS,AMDLSS,AMDRSS,AMSLSS
$,AMSRSS,AMCLSS,AMCRSS,AMBLSS,AMBRSS,AMB1SS,AMB2SS
$,AMTLSS,AMTRSS,AMT1SS,AMT2SS,AMELSS,AMERSS,AMMLSS,AMMRSS
$,AMLLSS,AMLRSS,AML1SS,AML2SS,AMN1SS,AMN2SS,AMN3SS
$,TWOM1,RV2V1,AMZ1SS,AMZ2SS,AMZ3SS,AMZ4SS,ZMIXSS(4,4)
$,AMW1SS,AMW2SS
$,GAMMAL,GAMMAR,AMHL,AMHH,AMHA,AMHC,ALFAH,AAT,THETAT
$,AAB,THETAB,AAL,THETAL,AMGVSS
REAL AMGLSS,AMULSS,AMURSS,AMDLSS,AMDRSS,AMSLSS
$,AMSRSS,AMCLSS,AMCRSS,AMBLSS,AMBRSS,AMB1SS,AMB2SS
$,AMTLSS,AMTRSS,AMT1SS,AMT2SS,AMELSS,AMERSS,AMMLSS,AMMRSS
$,AMLLSS,AMLRSS,AML1SS,AML2SS,AMN1SS,AMN2SS,AMN3SS
$,TWOM1,RV2V1,AMZ1SS,AMZ2SS,AMZ3SS,AMZ4SS,ZMIXSS
$,AMW1SS,AMW2SS
$,GAMMAL,GAMMAR,AMHL,AMHH,AMHA,AMHC,ALFAH,AAT,THETAT
$,AAB,THETAB,AAL,THETAL,AMGVSS
REAL AMZISS(4)
EQUIVALENCE (AMZISS(1),AMZ1SS)
SAVE /SSPAR/
```
It then calculates the widths and branching ratios and puts them in the common block /SSMODE/:
```
C MXSS = maximum number of modes
C NSSMOD = number of modes
C ISSMOD = initial particle
C JSSMOD = final particles
C GSSMOD = width
C BSSMOD = branching ratio
C MSSMOD = decay matrix element pointer
C LSSMOD = logical flag used internally by SSME3
INTEGER MXSS
PARAMETER (MXSS=1000)
COMMON/SSMODE/NSSMOD,ISSMOD(MXSS),JSSMOD(5,MXSS),GSSMOD(MXSS)
$,BSSMOD(MXSS),MSSMOD(MXSS),LSSMOD
INTEGER NSSMOD,ISSMOD,JSSMOD,MSSMOD
REAL GSSMOD,BSSMOD
LOGICAL LSSMOD
SAVE /SSMODE/
```
Decay modes for a given particle are not necessarily adjacent in this common block. Note that the branching ratio calculations use the full matrix elements, which in general will give nonuniform distributions in phase space, but this information is not saved in /SSMODE/. In particular, the decays $`HZ+Z^{}Z+f+\overline{f}`$ give no indication that the $`f\overline{f}`$ mass is strongly peaked near the upper limit.
All IDENT codes are defined by parameter statements in the PATCHY keep sequence SSTYPE:
```
C SM ident code definitions. These are standard ISAJET but
C can be changed.
INTEGER IDUP,IDDN,IDST,IDCH,IDBT,IDTP
INTEGER IDNE,IDE,IDNM,IDMU,IDNT,IDTAU
INTEGER IDGL,IDGM,IDW,IDZ,IDH
PARAMETER (IDUP=1,IDDN=2,IDST=3,IDCH=4,IDBT=5,IDTP=6)
PARAMETER (IDNE=11,IDE=12,IDNM=13,IDMU=14,IDNT=15,IDTAU=16)
PARAMETER (IDGL=9,IDGM=10,IDW=80,IDZ=90,IDH=81)
C SUSY ident code definitions. They are chosen to be similar
C to those in versions < 6.50 but may be changed.
INTEGER ISUPL,ISDNL,ISSTL,ISCHL,ISBT1,ISTP1
INTEGER ISNEL,ISEL,ISNML,ISMUL,ISNTL,ISTAU1
INTEGER ISUPR,ISDNR,ISSTR,ISCHR,ISBT2,ISTP2
INTEGER ISNER,ISER,ISNMR,ISMUR,ISNTR,ISTAU2
INTEGER ISZ1,ISZ2,ISZ3,ISZ4,ISW1,ISW2,ISGL
INTEGER ISHL,ISHH,ISHA,ISHC
INTEGER ISGRAV
PARAMETER (ISUPL=21,ISDNL=22,ISSTL=23,ISCHL=24,ISBT1=25,ISTP1=26)
PARAMETER (ISNEL=31,ISEL=32,ISNML=33,ISMUL=34,ISNTL=35,ISTAU1=36)
PARAMETER (ISUPR=41,ISDNR=42,ISSTR=43,ISCHR=44,ISBT2=45,ISTP2=46)
PARAMETER (ISNER=51,ISER=52,ISNMR=53,ISMUR=54,ISNTR=55,ISTAU2=56)
PARAMETER (ISGL=29)
PARAMETER (ISZ1=30,ISZ2=40,ISZ3=50,ISZ4=60,ISW1=39,ISW2=49)
PARAMETER (ISHL=82,ISHH=83,ISHA=84,ISHC=86)
PARAMETER (ISGRAV=91)
```
These are based on standard ISAJET but can be changed to interface with other generators. Since masses except the t mass are hard wired, one should check the kinematics for any decay before using it with possibly different masses.
Instead of specifying all the SUSY parameters at the electroweak scale using the MSSMi commands, one can instead use the SUGRA parameter to specify in the minimal supergravity framework the common scalar mass $`m_0`$, the common gaugino mass $`m_{1/2}`$, and the soft trilinear SUSY breaking parameter $`A_0`$ at the GUT scale, the ratio $`\mathrm{tan}\beta `$ of Higgs vacuum expectation values at the electroweak scale, and $`sgn\mu `$, the sign of the Higgsino mass term. The `NUSUGi` keywords allow one to break the assumption of universality in various ways. `NUSUG1` sets the gaugino masses; `NUSUG2` sets the $`A`$ terms; `NUSUG3` sets the Higgs masses; `NUSUG4` sets the first generation squark and slepton masses; and `NUSUG5` sets the third generation masses.
The renormalization group equations are solved iteratively using Runge-Kutta numerical integration to determine the weak scale parameters from the GUT scale ones:
1. The RGEโs are run from the weak scale $`M_Z`$ up to the GUT scale, where $`\alpha _1=\alpha _2`$, taking all thresholds into account. We use two loop RGE equations for the gauge couplings only.
2. The GUT scale boundary conditions are imposed, and the RGEโs are run back to $`M_Z`$, again taking thresholds into account.
3. The masses of the SUSY particles and the values of the soft breaking parameters B and mu needed for radiative symmetry are computed, e.g.
$$\mu ^2(M_Z)=\frac{M_{H_1}^2M_{H_2}^2\mathrm{tan}^2\beta }{\mathrm{tan}^2\beta 1}M_Z^2/2$$
These couplings are frozen out at the scale $`\sqrt{M(t_L)M(t_R)}`$.
4. The 1-loop radiative corrections are computed.
5. The process is then iterated until stable results are obtained.
This is essentially identical to the procedure used by several other groups. Other possible constraints such as b-tau unification and limits on proton decay have not been included.
An alternative to the SUGRA model is the Gauge Mediated SUSY Breaking (GMSB) model of Dine and Nelson, Phys. Rev. D48, 1277 (1973); Dine, Nelson, Nir, and Shirman, Phys. Rev. D53, 2658 (1996). In this model SUSY is broken dynamically and communicated to the MSSM through messenger fields at a messenger mass scale $`M_m`$ much less than the Planck scale. If the messenger fields are in complete representations of $`SU(5`$), then the unification of couplings suggested by the LEP data is preserved. The simplest model has a single $`5+\overline{5}`$ messenger sector with a mass $`M_m`$ and and a SUSY-breaking VEV $`F_m`$ of its auxiliary field $`F`$. Gauginos get masses from one-loop graphs proportional to $`\mathrm{\Lambda }_m=F_m/M_m`$ times the appropriate gauge coupling $`\alpha _i`$; sfermions get squared-masses from two-loop graphs proportional to $`\mathrm{\Lambda }_m`$ times the square of the appropriate $`\alpha _i`$. If there are $`N_5`$ messenger fields, the gaugino masses and sfermion masses-squared each contain a factor of $`N_5`$.
The parameters of the GMSB model implemented in ISAJET are
* $`\mathrm{\Lambda }_m=F_m/M_m`$: the scale of SUSY breaking, typically 10โ$`100\mathrm{TeV}`$;
* $`M_m>\mathrm{\Lambda }_m`$: the messenger mass scale, at which the boundary conditions for the renormalization group equations are imposed;
* $`N_5`$: the equivalent number of $`5+\overline{5}`$ messenger fields.
* $`\mathrm{tan}\beta `$: the ratio of Higgs vacuum expectation values at the electroweak scale;
* $`sgn\mu =\pm 1`$: the sign of the Higgsino mass term;
* $`C_{\mathrm{grav}}1`$: the ratio of the gravitino mass to the value it would have had if the only SUSY breaking scale were $`F_m`$.
The solution of the renormalization group equations is essentially the same as for SUGRA; only the boundary conditions are changed. In particular it is assumed that electroweak symmetry is broken radiatively by the top Yukawa coupling.
In GMSB models the lightest SUSY particle is always the nearly massless gravitino $`\stackrel{~}{G}`$. The phenomenology depends on the nature of the next lightest SUSY particle (NLSP) and on its lifetime to decay to a gravitino. The NLSP can be either a neutralino $`\stackrel{~}{\chi }_1^0`$ or a slepton $`\stackrel{~}{\tau }_1`$. Its lifetime depends on the gravitino mass, which is determined by the scale of SUSY breaking not just in the messenger sector but also in any other hidden sector. If this is set by the messenger scale $`F_m`$, i.e., if $`C_{\mathrm{grav}}1`$, then this lifetime is generally short. However, if the messenger SUSY breaking scale $`F_m`$ is related by a small coupling constant to a much larger SUSY breaking scale $`F_b`$, then $`C_{\mathrm{grav}}1`$ and the NLSP can be long-lived. The correct scale is not known, so $`C_{\mathrm{grav}}`$ should be treated as an arbitrary parameter. More complicated GMSB models may be run by using the GMSB2 keyword.
Patch ISASSRUN of ISAJET provides a main program SSRUN and some utility programs to produce human readable output. These utilities must be rewritten if the IDENT codes in /SSTYPE/ are modified. To create the stand-alone version of ISASUSY with SSRUN, run YPATCHY on isajet.car with the following cradle (with `&` replaced by `+`):
```
&USE,*ISASUSY. Select all code
&USE,NOCERN. No CERN Library
&USE,IMPNONE. Use IMPLICIT NONE
&EXE. Write everything to ASM
&PAM,T=C. Read PAM file
&QUIT. Quit
```
Compile, link, and run the resulting program, and follow the prompts for input. Patch ISASSRUN also contains a main program SUGRUN that reads the minimal SUGRA, non-universal SUGRA, or GMSB parameters, solves the renormalization group equations, and calculates the masses and branching ratios. To create the stand-alone version of ISASUGRA, run YPATCHY with the following cradle:
```
&USE,*ISASUGRA. Select all code
&USE,NOCERN. No CERN Library
&USE,IMPNONE. Use IMPLICIT NONE
&EXE. Write everything to ASM
&PAM. Read PAM file
&QUIT. Quit
```
The documentation for ISASUSY and ISASUGRA is included with that for ISAJET.
ISASUSY is written in ANSI standard Fortran 77 except that IMPLICIT NONE is used if +USE,IMPNONE is selected in the Patchy cradle. All variables are explicitly typed, and variables starting with I,J,K,L,M,N are not necessarily integers. All external names such as the names of subroutines and common blocks start with the letters SS. Most calculations are done in double precision. If +USE,NOCERN is selected in the Patchy cradle, then the Cernlib routines EISRS1 and its auxiliaries to calculate the eigenvalues of a real symmetric matrix and DDILOG to calculate the dilogarithm function are included. Hence it is not necessary to link with Cernlib.
The physics assumptions and details of incorporating the Minimal Supersymmetric Model into ISAJET have appeared in a conference proceedings entitled โSimulating Supersymmetry with ISAJET 7.0/ISASUSY 1.0โ by H. Baer, F. Paige, S. Protopopescu and X. Tata; this has appeared in the proceedings of the workshop on Physics at Current Accelerators and Supercolliders, ed. J. Hewett, A. White and D. Zeppenfeld, (Argonne National Laboratory, 1993). Detailed references may be found therein. Users wishing to cite an appropriate source may cite the above report.
## 13 Changes in Recent Versions
This section contains a record of changes in recently released versions of ISAJET, taken from the memoranda distributed to users. Note that the released version numbers are not necessarily consecutive.
### 13.1 Version 7.47, December 1999
There are several improvements in the treatment of supersymmetry. The Anomaly Mediated SUSY Breaking model of of Randall and Sundrum and of Gherghetta, Giudice, and Wells (hep-ph/9904378) has been added. The parameters of the model are a universal scalar mass $`m_0`$ at the GUT scale, a gravitino mass $`m_{3/2}`$, and the usual $`\mathrm{tan}\beta `$ and $`sgn\mu `$. These are set by the `AMSB` keyword. The renormalization group equations have been extended to include two-loop Yukawa terms and right-handed sneutrinos (with default masses above the Planck scale). The $`\stackrel{~}{\nu }_R`$ play a role in the evolution for the inverted hierarchy models of Bagger, Feng, and Polonsky, hep-ph/9905292. SUSY loop corrections to Yukawa couplings have been incorporated in the SUSY mass calculations.
The Helas library of Murayama, Watanabe, and Hagiwara has been incorporated together with a simple multi-body phase space generator. This makes it possible to use code generated by MadGraph to produce multi-body hard scattering processes. As a first example, a `ZJJ` process that generates $`Z+\text{2 jets}`$ has been added, with the $`Z`$ treated as a narrow resonance. Additional processes may be added in future releases.
A new `EXTRADIM` process has been added to generate Kaluza-Klein graviton production in association with a jet or photon in models with extra dimensions at the TeV scale. The cross sections are from G.F.Giudice et al., hep-ph/9811291. We thank I. Hinchliffe and L. Vacavant for providing this.
A number of bugs have been fixed, including in particular one in the decay $`\stackrel{~}{W}_i\stackrel{~}{Z}_j\tau \nu `$.
### 13.2 Version 7.44, April 1999
A serious bug introduced in Version 7.42 that could lead to matrix elements being stored for the wrong mode has been corrected. Some sign errors in the matrix elements for gaugino decays have also been corrected.
### 13.3 Version 7.42, January 1999
Beginning with this version, matrix elements are taken into account in the event generator as well as in the calculation of decay widths for MSSM three-body decays of the form $`\stackrel{~}{A}\stackrel{~}{B}f\overline{f}`$, where $`\stackrel{~}{A}`$ and $`\stackrel{~}{B}`$ are gluinos, charginos, or neutralinos. This is implemented by having ISASUSY save the poles and their couplings when calculating the decay width and then using these to reconstruct the matrix element. Other three-body decays may be included in the future. Decays selected with `FORCE` use the appropriate matrix elements.
As part of the changes to implement these matrix elements, the format of the decay table has changed. It now starts with a header line; if this does not match the internal version, then a warning is printed. The decay table now includes an index MELEM that specifies the matrix element to be used for all processes. This is also used for `FORCE` decays and is printed on the run listing for them. SUSY 3-body decays have internally generated negative values of MELEM.
This version also includes both initial state radiation and beamstrahlung for $`e^+e^{}`$ interactions. For initial state radiation (bremsstrahlung), if the `EEBREM` keyword is selected, an electron structure function will be used. For a convolution of both bremsstrahlung and beamstrahlung, the keyword `EEBEAM` must be used, with appropriate inputs (see documentation).
### 13.4 Version 7.40, October 1998
A new process WHIGGS generates $`W^\pm +H`$ and $`Z+H`$ events for both the Standard Model and SUSY models and also Higgs pair production for SUSY models. The types and $`W`$ decay modes are selected with JETTYPE and WMODE as for WPAIR events. This process is of particular interest for producing fairly light Higgs bosons at the Tevatron. See the documentation for more details.
Some non-minimal GMSB models can be generated using a new keyword GMSB2. The optional parameters are an extra factor between the gaugino and scalar masses, shifts in the Higgs masses, a $`D`$-term proportional to hypercharge, and independent numbers of messenger fields for the three gauge groups. The documentation gives more details and references.
The default for SUGRA models has been changed to use $`\alpha _s(M_Z)=0.118`$, the experimental value. This means that the couplings do not exactly unify at the GUT scale, presumably because of the effects of heavy particles. The keyword AL3UNI can be used to select exact unification, which produces too large a value for $`\alpha _s(M_Z)`$.
A number of three-body slepton decays that occur through left-right mixing are now included. These are obviously small but might compete with gravitino decays. In particular, a decay like $`\stackrel{~}{\mu }_R\stackrel{~}{\tau }_1\nu \overline{\nu }`$ might lead to a wrong momentum measurement in the muon system. So far we have found no case in which this is probable.
The new release also includes a separate Unix tar file `mcpp.tar` containing C++ code to read a standard ISAJET output file and copy all the information into C++ classes. The tar file contains makefiles for Software Release Tools, documentation, and examples as well as the code.
### 13.5 Version 7.37, April 1998
Version 7.37 incorporates Gauge Mediated SUSY Breaking models for the first time. In these models, SUSY is broken in a hidden sector at a relatively low scale, and the masses of the MSSM fields are then produced through ordinary gauge interactions with messenger fields. The parameters of the GMSB model in ISAJET are $`M_m`$, the messenger mass scale; $`\mathrm{\Lambda }_m=F_m/M_m`$, where $`F_m`$ is the SUSY breaking scale in the messenger sector; $`N_5`$, the number of messenger fields; the usual $`\mathrm{tan}\beta `$ and $`sgn\mu `$; and $`C_{\mathrm{grav}}1`$, a factor which scales the gravitino mass and hence the lifetime for the lightest MSSM particle to decay into it.
GMSB models have a light gravitino $`\stackrel{~}{G}`$ as the lightest SUSY particle. The phenomenology of the model depends mainly on the nature of the next lightest SUSY particle, a $`\stackrel{~}{\chi }_1^0`$ or a $`\stackrel{~}{\tau }_1`$, which changes with the number $`N_5`$ of messengers. The phenomenology also depends on the lifetime for the $`\stackrel{~}{\chi }_1^0\stackrel{~}{G}\gamma `$ or $`\stackrel{~}{\tau }_1\stackrel{~}{G}\tau `$ decay; this lifetime can be short or very long. All the relevant decays are included except for $`\stackrel{~}{\mu }\nu \nu \stackrel{~}{\tau }_1`$, which is very suppressed.
The keyword MGVTNO allows the user to independently input a gravitino gravitino mass for the MSSM option. This allows studies of SUGRA (or other types) of models where the gravitino is the LSP.
Version 7.37 also contains an extension of the SUGRA model with a variety of non-universal gaugino and sfermion masses and $`A`$ terms at the GUT scale. This makes it possible to study, for example, how well the SUGRA assumptions can be tested.
Two significant bugs have also been corrected. The decay modes for $`B^{}`$ mesons were missing from the decay table since Version 7.29 and have been restored. A sign error in the interference term for chargino production has been corrected, leading to a larger chargino pair cross section at the Tevatron.
### 13.6 Version 7.32, November 1997
This version makes several corrections in various chargino and neutralino widths, thus changing the branching ratios for large $`\mathrm{tan}\beta `$. For $`\stackrel{~}{\chi }_2^0`$, for example, the $`\stackrel{~}{\chi }_1^0b\overline{b}`$ branching ratio is decreased significantly, while the $`\stackrel{~}{\chi }_1^0\tau ^+\tau ^{}`$ one is increased. Thus the SUGRA phenomenology for $`\mathrm{tan}\beta 30`$ is modified substantially.
The new version also fixes a few bugs, including a possible numerical precision problem in the Drell-Yan process at high mass and $`q_T`$. It also includes a missing routine for the Zebra interface.
### 13.7 Version 7.31, August 1997
Version fixes a couple of bugs in Version 7.29. In particular, the JETTYPE selection did not work correctly for supersymmetric Higgs bosons, and there was an error in the interactive interface for MSSM input. Since these could lead to incorrect results, users should replace the old version. We thank Art Kreymer for finding these problems.
Since top quarks decay before they have time to hadronize, they are now put directly onto the particle list. Top hadrons ($`t\overline{u}`$, $`t\overline{d}`$, etc.) no longer appear, and FORCE should be used directly for the top quark, i.e.
```
FORCE
6,11,-12,5/
```
The documentation has been converted to LaTeX. Run either LaTeX 2.09 or LaTeX 2e three times to resolve all the forward references. Either US (8.5x11 inch) or A4 size paper can be used.
### 13.8 Version 7.30, July 1997
This version fixes a couple of bugs in the previous version. In particular, the JETTYPE selection did not work correctly for supersymmetric Higgs bosons, and there was an error in the interactive interface for MSSM input. Since these could lead to incorrect results, users should replace the old version. We thank Art Kreymer for finding these problems.
Since top quarks decay before they have time to hadronize, they are now put directly onto the particle list. Top hadrons ($`t\overline{u}`$, $`tud`$, etc.) no longer appear, and FORCE should be used directly for the top quark, i.e.
```
FORCE
6,11,-12,5/
```
The documentation has been converted to . Run either 2.09 or 2e three times to resolve all the forward references. Either US ($`8.5\times 11`$ inch) or A4 size paper can be used.
### 13.9 Version 7.29, May 1997
While the previous version was applicable for large as well as small $`\mathrm{tan}\beta `$, it did contain approximations for the 3-body decays $`\stackrel{~}{g}t\overline{b}\stackrel{~}{W}_i`$, $`\stackrel{~}{Z}_ib\overline{b}\stackrel{~}{Z}_j,\tau \tau \stackrel{~}{Z}_j`$, and $`\stackrel{~}{W}_i\tau \nu \stackrel{~}{Z}_j`$. The complete tree-level calculations for three body decays of the gluino, chargino and neutralino, with all Yukawa couplings and mixings, have now been included (thanks mainly to M. Drees). We have compared our branching ratios with those calculated by A. Bartl and collaborators; the agreement is generally good.
The decay patterns of gluinos, charginos and neutralinos may differ from previous expectations if $`\mathrm{tan}\beta `$ is large. In particular, decays into $`\tau `$โs and $`b`$โs are often enhanced, while decays into $`e`$โs and $`\mu `$โs are reduced. It could be important for experiments to study new types of signatures, since the cross sections for conventional signatures may be considerably reduced.
We have also corrected several bugs, including a fairly serious one in the selection of jet types for SUSY Higgs. We thank A. Kreymer for pointing this out to us.
### 13.10 Version 7.27, January 1997
The new version contains substantial improvements in the treatment of the Minimal Supersymmetric Standard Model (MSSM) and the SUGRA model. The squarks of the first two generations are no longer assumed to be degenerate. The mass splittings and all the two-body decay modes are now correctly calculated for large $`\mathrm{tan}\beta `$. While there are still some approximations for three-body modes, ISAJET is now usable for the whole range $`1<\mathrm{tan}\beta <M_t/M_b`$. The most interesting new feature for large $`\mathrm{tan}\beta `$ is that third generation modes can be strongly enhanced or even completely dominant.
To accomodate these changes it was necessary to change the MSSM input parameters. To avoid confusion, the MSSM keywords have been renamed MSSM\[A-C\] instead of MSSM\[1-3\], and the order of the parameters has been changed. See the input section of the manual for details.
Treatment of the MSSM Higgs sector has also been improved. In the renormalization group equations the Higgs couplings are frozen at a higher scale, $`Q=\sqrt{M(\stackrel{~}{t}_L)M(\stackrel{~}{t}_R)}`$. Running $`t`$, $`b`$ and $`\tau `$ masses evaluated at that scale are used to reproduce the dominant 2-loop effects. There is some sensitivity to the choice of $`Q`$; our choice seems to give fairly stable results over a wide range of parameters and reasonable agreement with other calculations. In particular, the resulting light Higgs masses are significantly lower than those from Version 7.22.
The default parton distributions have been updated to CTEQ3L. A bug in the PDFLIB interface and other minor bugs have been fixed.
### 13.11 Version 7.22, July 1996
The new version fixes errors in $`\stackrel{~}{b}\stackrel{~}{W}t`$ and in some $`\stackrel{~}{t}`$ decays and Higgs decays. It also contains a new decay table with updated $`\tau `$, $`c`$, and $`b`$ decays, based loosely on the QQ decay package from CLEO. The updated decays are less detailed than the full CLEO QQ program but an improvement over what existed before. The new decays involve a number of additional resonances, including $`f_0(980)`$, $`a_1(1260)`$, $`f_2(1270)`$, $`K_1(1270)`$, $`K_1^{}(1400)`$, $`K_2^{}(1430)`$, $`\chi _{c1,2,3}`$, and $`\psi (2S)`$, so users may have to change their interface routines.
A number of other small bugs have been corrected.
### 13.12 Version 7.20, June 1996
The new version corrects both errors introduced in Version 7.19 and longstanding errors in the final state QCD shower algorithm. It also includes the top mass in the cross sections for $`gbWt`$ and $`gtZt`$. When the $`t`$ mass is taken into account, the process $`gtWb`$ can have a pole in the physical region, so it has been removed; see the documentation for more discussion.
Steve Tether recently pointed out to us that the anomalous dimension for the $`qqg`$ branching used in the final state QCD branching algorithm was incorrect. In investigating this we found an additional error, a missing factor of $`1/3`$ in the $`gq\overline{q}`$ branching. The first error produces a small but non-negligible underestimate of gluon radiation from quarks. The second overestimates quark pair production from gluons by about a factor of 3. In particular, this means that backgrounds from heavy quarks $`Q`$ coming from $`gQ\overline{Q}`$ have been overestimated.
The new version also allows the user to set arbitrary masses for the $`U(1)`$ and $`SU(2)`$ gaugino mases in the MSSM rather than deriving these from the gluino mass using grand unification. This could be useful in studying one of the SUSY interpretations of a CDF $`ee\gamma \gamma \text{ / }E_T`$ event recently suggested by Ambrosanio, Kane, Kribs, Martin and Mrenna. Note, however, that radiative decay are not included, although the user can force them and multiply by the appropriate branching ratios calculated by Haber and Wyler, Nucl. Phys. B323, 267 (1989). No explicit provision for the decay $`\stackrel{~}{Z}_1\stackrel{~}{G}\gamma `$ of the lightest zino into a gravitino or goldstino and a photon has been made, but forcing the decay $`\stackrel{~}{Z}_1\nu \gamma `$ has the same effect for any collider detector.
A number of other minor bugs have also been corrected.
### 13.13 Version 7.16, October 1995
The new version includes $`e^+e^{}`$ cross sections for both SUSY and Standard Model particles with polarized beams. The $`e^{}`$ and $`e^+`$ polarizations are specified with a new keyword EPOL. Polarization appears to be quite useful in studying SUSY particles at an $`e^+e^{}`$ collider.
The new release also includes some bug fixes for $`pp`$ reactions, so you should upgrade even if you do not plan to use the polarized $`e^+e^{}`$ cross sections.
### 13.14 Version 7.13, September 1994
Version 7.13 of ISAJET fixes a bug that we introduced in the recently released 7.11 and another bug in $`\stackrel{~}{g}\stackrel{~}{q}\overline{q}`$. We felt it was essential to fix these bugs despite the proliferation of versions.
The new version includes the cross sections for the $`e^+e^{}`$ production of squarks, sleptons, gauginos, and Higgs bosons in Minimal Supersymmetric Standard Model (MSSM) or the minimal supergravity (SUGRA) model, including the effects of cascade decays. To generate such events, select the `E+E-` reaction type and either SUGRA or MSSM, e.g.,
```
SAMPLE E+E- JOB
300.,50000,10,100/
E+E-
SUGRA
100,100,0,2,-1/
TMASS
170,-1,1/
END
STOP
```
The effects of spin correlations in the production and decay, e.g., in $`e^+e^{}\stackrel{~}{W}_1^+\stackrel{~}{W}_1^{}`$, are not included.
It should be noted that the Standard Model $`e^+e^{}`$ generator in ISAJET does not include Bhabba scattering or $`W^+W^{}`$ and $`Z^0Z^0`$ production. Also, its hadronization model is cruder than that available in some other generators.
### 13.15 Version 7.11, September 1994
The new version includes the cross sections for the $`e^+e^{}`$ production of squarks, sleptons, gauginos, and Higgs bosons in Minimal Supersymmetric Standard Model (MSSM) or the minimal supergravity (SUGRA) model including the effects of cascade decays. To generate such events, select the `E+E-` reaction type and either SUGRA or MSSM, e.g.,
```
SAMPLE E+E- JOB
300.,50000,10,100/
E+E-
SUGRA
100,100,0,2,-1/
TMASS
170,-1,1/
END
STOP
```
The effects of spin correlations in the production and decay, e.g., in $`e^+e^{}\stackrel{~}{W}_1^+\stackrel{~}{W}_1^{}`$, are not included.
It should be noted that the Standard Model $`e^+e^{}`$ generator in ISAJET does not include Bhabba scattering or $`W^+W^{}`$ and $`Z^0Z^0`$ production. Also, its hadronization model is cruder than that available in some other generators.
### 13.16 Version 7.10, July 1994
This version adds a new option that solves the renormalization group equations to calculate the Minimal Supersymmetric Standard Model (MSSM) parameters in the minimal supergravity (SUGRA) model, assuming only that the low energy theory has the minimal particle content, that electroweak symmetry is radiatively broken, and that R-parity is conserved. The minimal SUGRA model has just four parameters, which are taken to be the common scalar mass $`m_0`$, the common gaugino mass $`m_{1/2}`$, the common trilinear SUSY breaking term $`A_0`$, all defined at the GUT scale, and $`\mathrm{tan}\beta `$; the sign of $`\mu `$ must also be given. The renormalization group equations are solved iteratively using Runge-Kutta integration including the correct thresholds. This program can be used either alone or as part of the event generator. In the latter case, the parameters are specified using
> SUGRA
> $`m_0`$, $`m_{1/2}`$, $`A_0`$, $`\mathrm{tan}\beta `$, $`sgn\mu `$
While the SUGRA option is less general than the MSSM, it is theoretically attractive and provides a much more managable parameter space.
In addition there have been a number of improvements and bug fixes. An occasional infinite loop in the minimum bias generator has been fixed. A few SUSY cross sections and decay modes and the JETTYPE flags for SUSY particles have been corrected. The treatment of $`B`$ baryons has been improved somewhat. |
warning/0001/astro-ph0001351.html | ar5iv | text | # ๐ฟ Scuti Stars in Stellar Systems: a Review
## 1. Introduction
It is well known that the vast majority of stars belong to a binary or a multiple system, irrespective of their spectral type. Recent surveys with improved astrometric accuracy, either from space or from the ground, present clear evidence that the higher the accuracy, the larger the number of binary detections. Up to 3000 new binaries have thus been discovered during the Hipparcos satellite mission (Lindegren 1997). The frequency of binaries is estimated to be at least 60% in the solar neighbourhood (Duquennoy & Mayor 1991) but this is probably an underestimation as modelisation tends to show. For example, Odenkirchen and Brosche (1999) found that a frequency of at least 70% was needed in some models to account for the existence of another 2400 (up to now undetected) astrometric binaries in the Hipparcos catalogue. The same is also true with respect to the improvement of the photometric accuracy: the higher the accuracy, the larger the number of variable star detections. The results of the Hipparcos photometric survey are again quite illustrative: some 8000 new variable stars have been identified (van Leeuwen 1997). It is thus not at all a surprise to encounter pulsating $`\delta `$ Scuti stars as components of stellar systems. The catalogue of Seeds and Yanchak (1972) contains a third of double stars over a total of 155 $`\delta `$ Sct variables or suspected ones. Szatmรกry (1990) also quotes a binary fraction of about 30 %. To study such cases presents some obvious advantages: we have additional constraints on some important physical parameters such as distance, mass, radiusโฆ but this of course will depend on the type of binary or multiple system of which the pulsating star is a member. We may find them among the wide visual binaries (VB), the closer visual binaries with orbital motion (VB/O), the (unresolved) astrometric binaries (AB), the spectroscopic binaries (SB), or even eclipsing and ellipsoidal binary systems (E). In almost all these cases it is generally accepted that both components pertaining to the system originate from the same parent cloud, therefore have the same age and chemical composition. Additional information on the non-variable component can then be extrapolated unto the variable one and used to better constrain the variability characteristics such as the pulsation type and modes. But we could go further and address such questions as: does binarity modify the pulsation properties or even trigger the pulsation? Is there any correlation between rotation, pulsation and orbital motion ?
For all these reasons the identification of such objects offers an interesting challenge toward a better comprehension of pulsation in the $`\delta `$ Scuti instability strip. A comparative study between the two companions allows to restrict for example the domain in mass in which the $`\delta `$ Sct phenomenon takes place. Various recent studies exist that investigate the pulsating $`\delta `$ Scuti stars in open clusters for similar reasons (e.g. the Praesepe cluster (Peรฑa et al. 1998; Alvarez et al. 1998)). One should however also keep in mind that star formation in some open clusters is not always a single-epoch process (e.g. M16 (Hillenbrand et al. 1993)).
## 2. Summary tables
We here summarize the available information for as complete a sample as possible of $`\delta `$ Scuti stars in stellar systems in two tables, one for the binarity and one for the variability characteristics. The basis was drawn on the catalogue work by Garcรญa et al. (1995), supplemented by our own list and literature search. The Catalogue of the Components of Double and Multiple Stars (CCDM, Dommanget & Nys 1995) was used to retrieve information on the visual components of the stellar systems. Table 1. lists the name and the HD number (or BD number if HD does not exist), the Hipparcos number, the V magnitude, remarks concerning the multiplicity, the Hipparcos parallax and error. Next comes the information regarding the duplicity in the Hipparcos Catalogue (ESA 1997): a flag referring to that part of the Annex of Double and Multiple Systems (DMSA) where the star has been classified, the component designations, the angular separation (in arcsec), followed by the information about the variability: the periodicity, a flag indicating in which Variability Annex the star has been classified (P/1=periodic variable; U/2=unsolved; M=possible microvariable; D=โduplicity-induced variabilityโ; C=โconstantโ). The final columns list the median Hp magnitude with standard error and the differential Hp magnitude with standard error.
Table 2 lists the name and the HD number (or BD number if HD does not exist), remarks concerning the variability, the number of frequencies, the dominant periodicity and semi-amplitude, the projected rotational velocity of the $`\delta `$ Sct star, the spectral types of the $`\delta `$ Sct and additional components. Among the remarks the following notations are used: N=narrow range of frequencies; W=wide range of frequencies; R=radial pulsator; NR=non-radial pulsator.
## 3. Visual binaries
### 3.1. Orbits and physical parameters
In Table 3 we present additional information regarding the physical parameters of the components of some of the visual binaries listed in Tables 1 and 2 Given are the name and HD designation, the orbital period in years, the semi-axis major in AU, the sum of the masses, and if known the radii $`R_1`$, $`R_2`$ and the individual masses $`M_1`$, $`M_2`$ as well as a reference.
We recall that the sum of the masses, $`\mathrm{\Sigma }M`$ (in solar mass), and the orbital elements, P<sub>orb</sub>, the orbital period (expressed in years), and A, the true semi-axis major (in AU), are linked through Keplerโs third law:
$$\mathrm{\Sigma }M=\frac{\mathrm{A}^3}{\mathrm{P}^2}.$$
There are three VB/O cases ($`\delta `$ Ser, $`\gamma `$ CrB, $`\tau `$ Cyg) but only two have a reliable orbit determination. Of course, only if the mass ratio is known from either the absolute astrometry or the spectroscopy can precise individual masses be obtained.
$`\sigma ^2`$ CrB is not listed since it is very probably not a $`\delta `$ Scuti star. This orbital pair is in fact a triple system. The visual binary has a period of $``$ 1000 yr and a semi-axis major of 140 AU. It also contains a double-lined and chromospherically active spectroscopic binary. From differential spectrophotometry, Frasca et al. (1997) deduce that there is no evidence for a 0.1 day periodicity but that all the variation is linked with the period of rotation. They conclude that the photometric variability is due to dark spots on the secondary component of the SB and very probably not of $`\delta `$ Scuti type. This conclusion is also supported by the spectral classification (F6+G0). We also discarded the following two visual double stars: V377 Cas (sep=2.1<sup>โฒโฒ</sup>) because the light variations are not typical of the $`\delta `$ Scuti type (Lowder 1989) and DL Dra (HR 5492, sep=3.9<sup>โฒโฒ</sup>), shown to be probably constant already in 1990 (Paparรณ et al. 1990).
We next focus on some particular objects of this category.
### 3.2. Selected stars
#### VW Ari A
This very wide binary consists of an A-type primary showing $`\lambda `$ Boo peculiarities in the spectrum and a F-type secondary of solar-like composition in common spatial motion (similar proper motions (CCDM, Dommanget & Nys 1995) and radial velocities (Fehrenbach et al. 1987)). Because of its very wide separation and the different chemical composition of its components, stellar capture was proposed as the probable mechanism of formation. However, Chernyshova et al. (1998) recently argumented that capture is improbable and also no longer needed to explain the differences in composition since these could be due to the specific evolution of the primary star solely. The primary is the $`\delta `$ Scuti star that was intensively observed by the STEPHI network in 1993 (Liu et al. 1996). They detected more than seven frequencies, more or less grouped. Non-radial rotationally split modes might therefore be present in this star, a medium to fast rotator with vsini=90 kms<sup>-1</sup>. Andrievsky et al. (1995) concluded from their spectral analysis that VW Ari A has no sharp lines while the lines of VW Ari B are โsharp and strongโ, so fast rotation is not applicable to the companion. Such a difference in rotational velocity between both components was not considered by Liu et al. (1996), who also found a much smaller surface gravity value for component A via photometric calibration, thereby resulting in masses surprisingly small to match the given spectral types.
#### $`\theta ^2`$ Tau
$`\theta ^2`$ Tau is the most massive main-sequence star of the Hyades cluster, the primary of a common proper motion pair with $`\theta ^1`$ Tau (at a separation of 5$`.^{}`$6), and a member of a single-lined spectroscopic binary (SB1) of period 140.7 days with an (highly eccentric) interferometric orbit from the Mark III optical interferometer (Pan, Shao, & Colavita 1992; Hummel & Armstrong 1992). Torres, Stefanik & Latham (1997; hereafter TSL97) determined individual masses and the distance of the system by treating it as a double-lined spectroscopic binary (SB2), thereby exploring a range of values for the mass ratio and the rotational velocity. The derived orbital and the Hipparcos trigonometric parallaxes agree very well. This spectroscopic binary is formed by two stars of nearly identical colour and mass but with different projected rotational velocities: TSL97 obtained a best fit with v<sub>B</sub>sini= 110 kms<sup>-1</sup> while v<sub>A</sub>sini= 80 kms<sup>-1</sup>. Both are therefore considered to be rapid rotators. From the location in a colour-magnitude diagram and a best fit with an isochrone of age $``$ 630 Myr, they conclude that the primary is in a phase near H core exhaustion, immediately preceding the phase of overall contraction. However, because both the binarity and the fast rotation may affect the colour indices, its evolutionary status may still be ambiguous (see also Krรณlikowska 1992).
Various multi-site campaigns have been conducted. Breger et al. (1989; hereafter BG89) obtained five closely spaced and stable frequencies, all of which had amplitudes below 0.01 mag. They discarded rotational splitting since it could not explain the observed frequency separations and proposed a mixture of modes of different l and m values. Kennelly et al. (1996) discussed a large set of radial velocity and line profile data from which up to seven frequencies emerged with only three frequencies in common with the previous analysis. They suggested long-term ($`>`$ 6 yr) amplitude variability and a combination of low and high degree modes. Amplitude variability on a 10 yr time scale is also claimed by Li, Zhou & Yang (1997). Both components lie within the $`\delta `$ Scuti instability strip but it seems well established that the more massive primary is the pulsating star (BG89, KW96). Even though a wealth of information about physical parameters is known for $`\theta ^2`$ Tau, the situation regarding variability is very confused and there is up to now no clear mode identification possible for the apparently very complex (not solved) frequency spectrum.
#### KW Aur A
KW Aur A is an ellipsoidal, single-lined spectroscopic binary system (Harper 1938) as well as the primary component of the visual multiple system ADS 3824 that is in common proper motion with the tertiary component situated at a distance of some 1000 AU with a period of about 24000 yr (Tokovinin 1997). Though well-detached, both components of the inner system must be tidally deformed due to the short period of the binary (P=3.79d). Photometric and radial velocity variations analyzed together gave evidence for three non-radial modes (Fitch & Wisniewski 1979) (FW79). In addition, they also identified a frequency that corresponds to twice the orbital frequency due to the ellipticity effects on the mean light curve. They suggested a single pulsation frequency split by tidal modulation (instead of rotational modulation) as the cause for the observed non-radial triplet (same l, different m). In an analysis of line profile variations for this star, Smith (1982) corroborated their conclusion.
Rotational effects are expected to be small due to the fact that there is synchronization of the rotation with the orbital motion (see also Table 2). More theoretical work needs to be done in this context: computations should be done for inhomogeneous models, with a density gradient from core to envelope, better applicable to $`\delta `$ Scuti stars in general and to KW Aur A in particular.
#### $`\kappa ^2`$ Boo A
$`\kappa ^2`$ Boo forms a common proper motion system with $`\kappa ^1`$ Boo, itself consisting of a spectroscopic binary with a period of 1791 days and a high eccentricity (Batten, Fletcher, & MacCarthy 1989). Both stars have colours that locate them in or near the $`\delta `$ Scuti instability strip. The intrinsic separation is $``$ 600 AU, consequently the orbital period could be about 8000 yr long if one adopts 3.5 M as the sum of the masses. The primary is a fast rotator but this is improbable for the secondary ($`v_B`$sini=40 kms<sup>-1</sup>). Frandsen et al. (1995) (FJ95) observed both adopting a scheme of programme star (comp. A) versus comparison star (comp. B) in order to study the pulsation behaviour of $`\kappa ^2`$ Boo. They detected a multiperiodic pattern with up to four frequencies of which three are very close and one is a probable radial mode. All associated amplitudes are below 0.01 mag. FJ95 found a matching model by fitting a common isochrone through both stars and derive a distance that is in perfect agreement with the (later published) Hipparcos distance. They are hesitant to invoke rotationally split modes since their model produces a good match of the observed set of frequencies, even though rotation was not considered! One problem of this analysis is that they were not able to identify which component is responsible for the variability, let stand for what frequencies. Since component B is an early F-type binary, in principle it could also be a pulsating variable star (as admitted by the authors themselves). It would thus be very interesting to precisely identify the source of the multiple frequencies detected in this triple system.
#### HR 8895
An interesting common origin pair is formed by HD 220392 and HD 220391. Both components are located in the $`\delta `$ Scuti instability strip but up to now only the primary seems to present short-period pulsations of the $`\delta `$ Scuti type. More observational effort should be spent on both stars to investigate their behaviour with respect to pulsation. For a more detailed analysis we refer to Lampens & Van Camp (1999). A very recent and exciting information concerns new radial velocity measurements: Grenier et al. (1999) confirm that the radial velocities are in agreement and furthermore also show that component B has a variable radial velocity!
#### $`\gamma `$ CrB
This is an orbital binary with a period of 93 yr, a high eccentricity and a semi-axis major of some 33 AU (Hartkopf & McAlister 1989). The system consists of a B9/A0 IV primary, a suspected Maia star, and a 1.5 mag fainter A3/A4 main-sequence secondary. Lehmann et al. (1997) measured the radial velocity of the system and concluded that stars of the Maia type can pulsate in the same way as $`\delta `$ Scuti stars. In a first possible scenario they invoke an additional mechanism for the amplitude variation of the NR modes, while a second possible scenario implies spontaneous excitation and damping of these modes. One more alternative is that component B could well be the $`\delta `$ Scuti variable star.
#### DG Leo
This system is also triple, consisting of a double-lined spectroscopic binary and a visual component B that is the $`\delta `$ Scuti pulsating star. The spectroscopic double has an orbital period of 4.147 days and a mass ratio close to unity. All three stars are possibly nearly identical with a global spectral type A8IV and with marginal metallicity. Rosvick & Scarfe (1991) (RS91) detected a velocity change of the visual component with respect to the centre of mass of the spectroscopic pair that is due to the orbital motion. Speckle measurements show that this system is highly inclined, possibly also highly eccentric (Hartkopf, McAlister, & Mason, 1999). This was also claimed by RS91, who presented evidence for the occurrence of very shallow eclipses with a possible inclination of $``$ 70 . They obtained a mass of some 2 M for each spectroscopic component. Resolving the visual orbit will lead to more accurate masses for all three stars.
Only one periodicity seems to be known for the $`\delta `$ Scuti variable companion but more may be present since amplitude and phase changes have been reported (RS91). It is a very interesting multiple system where both metallicity and binarity effects can potentially influence the pulsational characteristics (on stabilization and tidal mixing see Budaj (1996; 1997)).
#### $`ฯต`$ Cephei
This is the primary of a visual double system that is a chance alignment of a bright foreground and a much fainter background star ($`\rho `$=128โ; $`\mathrm{\Delta }\mathrm{m}`$=5). However, $`ฯต`$ Cephei was recently also found to be a new astrometric binary discovered by the Hipparcos satellite (ESA 1997, part DMSA/G). We have no further binarity information. Observations by Lopรฉz de Coca et al. (1979) in the B band indicate two periods, perhaps three. Ratios are compatible with the second overtone and the fundamental radial modes. But the amplitudes are very small. From an analysis of the line profiles, Baade et al. (1993) found evidence for intermediate to high order NR p-modes with 6$`m`$8. Why such high order p-modes are also detected in integrated light data is unclear. This star shows a rich spectrum of modes according to Horner et al. (1996) and it is not improbable that it still belongs to the main sequence. Given its spectral type of F0, it is a medium to fast rotator.
#### $`\beta `$ Cas
ADS 107 A was reported to show radial velocity variations with an apparent period of about 27 days (Mellor 1917) and was therefore subsequently classified as SB (for example in $`\delta `$ Scuti stars catalogues). However Abt (1965) already concluded that there was no evidence for such a binarity, as was later on confirmed by the results of Yang, Walker, & Fahlman (1982). They showed that the radial velocity curve varied with a very short period, close to the photometric period of pulsation. In conclusion, $`\beta `$ Cas is a monoperiodic $`\delta `$ Scuti pulsator with a very stable small amplitude but for which the mode identification is uncertain (Rodrรญguez et al. 1992; Riboni, Poretti, & Galli 1994). We have classified it among the probable optical visual double stars and do not consider it as a pulsating star in a binary system (cfr. bottom part of Tables 1 and 2).
## 4. Spectroscopic and eclipsing binaries
An interesting feature of pulsating stars in binary systems is the so-called light-time effect: the orbital motion of the variable star produces a Doppler effect, and the observed period of variation will decrease and increase. This effect, together with mass transfer in a semi-detached system and apsidal motion, can be the cause of regular period changes in a binary system. For a circular orbit of radius $`a`$, orbital period $`P`$ and inclination $`i`$, the time of light maximum is given by (e.g. Barnes & Moffett, 1975):
$$T_{\mathrm{max}}=T_o+EP_o\frac{a\mathrm{sin}i}{c}\mathrm{cos}2\pi \left(\frac{EP_o}{P}\varphi \right),$$
(1)
where $`E`$ is the number of elapsed periods, $`T_o`$ is the initial epoch of maximum, $`P_o`$ is the pulsation period and $`c`$ is the speed of light. The light-time phenomenon has been observed in many binary $`\delta `$ Scuti stars and offers a useful way to obtain the orbital period.
Another feature of very short-period binaries is tidal deformation: the components can no longer be considered as spherical objects but have a triaxial symmetry with the longest axis directed toward the companion. The effects on the pulsation may be quite diverse and very difficult to detect: a) frequency shifts; b) amplitude modulation in the case of radial pulsation (e.g. $`\theta `$ Tuc, De Mey et al. 1998); c) frequency splitting in the case of non-radial pulsation (e.g. KW Aur, Fitch & Wiลniewski 1979)โฆ.
### 4.1. Single-lined spectroscopic binaries
Table 4 contains the orbital elements for the $`\delta `$ Scuti stars which are reported in the literature as SB1. The table lists the name, HD designation (or BD number if HD does not exist), spectral type, orbital period in days, eccentricity, semi-amplitude of the radial velocity in kms<sup>-1</sup>, mass function in solar mass and finally the reference of the orbit. For simplicity, when the orbit was published in the Catalogue of Batten et al. (1989), we only refer them and not the original papers. In this case, the spectral type is also the one they quote. Let us recall here that the mass function is given by
$$f(m)=\frac{M_2}{(M_1+M_2)^2}\mathrm{sin}^3i,$$
(2)
where $`M_1`$ and $`M_2`$ are the mass of the primary and secondary, respectively, and $`i`$ is the unknown inclination of the orbit with respect to the line-of-sight.
We do not claim to have the complete list of $`\delta `$ Scuti members of spectroscopic binaries. The classification of a $`\delta `$ Scuti star as a spectroscopic binary is indeed not easy if no detailed study is done. A simple variation of the radial velocity over the years may not be an indication of duplicity but solely the result of pulsation. The misclassification of AI CVn (King & Liu, 1990) is very illustrative in this respect. Szatmรกry (1990) gave a list of $`\delta `$ Scuti stars in binary systems, either eclipsing or spectroscopic. Some stars of his table do not appear in our Table 4 : ET And, DV Aqr, MM Cas, RX Cas, AZ CMi, ZZ Cyg and $`\theta `$ Vir.
* The binary system ET And contains a B9p primary which was believed to be pulsating. However, Weiss et al. (1998) have found that it is the main comparison star, HD 219891, which pulsates, with a period of 0.1 day and a semi-amplitude of 2.5 millimag. Thus, HD 219891 is the $`\delta `$ Scuti star while ET And appears very stable!
* The $`\delta `$ Scuti nature of DV Aqr needs to be confirmed.
* MM Cas is an eclipsing binary reported by Chaubey (1983) to have brightness fluctuations of 0.08 mag amplitude and a period of $`3^\mathrm{h}40`$. Chaubey therefore conjectured that the star is a $`\delta `$ Scuti variable. This needs confirmation.
* RX Cas is a semi-detached system whose orbital period is increasing. The A-type primary spectrum is that of a shell or disc that completely conceals from view the primary star.
* AZ CMi is a pulsating star of spectral type F0 III. The long orbital period (2625 days) quoted by Szatmรกry (1990) is derived from a sinusoidal fit to the O-C data, assuming that they can be attributed to the light-time effect. A spectroscopic study of this object would thus be useful.
* Frolov et al. (1982) could not confirm the $`\delta `$ Scuti status of ZZ Cyg. We did not retain it in our list.
* Finally, $`\theta `$ Vir (HR 4963 = HD 114330) also deserves some attention. This star is a spectroscopic binary with a 17.84 years orbit, which is itself a visual binary. The star has been classified as a hot Am star and Beardsley & Zizka (1977) detected a variability at the level of 5 kms<sup>-1</sup> with a period of $`0.^\mathrm{d}152360`$. Adelman (1997), however, found no evidence for variability, although Scholtz et al. (1998) note that there is โremarkable scatter which could possibly indicate real variationsโ. These authors also remark that Hipparcos revealed a variability at the 8 millimag level with a period of $`0.^\mathrm{d}697382`$. From their spectroscopic investigation, they detected a well defined period of $`0.^\mathrm{d}0614`$, with an amplitude of 0.4 kms<sup>-1</sup>. It could be the case that $`\theta `$ Vir is another example of a pulsating Am star in a binary system. This needs however to be more definitively ascertained.
There are some other stars which could enter our list of $`\delta `$ Scuti in multiple systems. Among these pending cases, let us consider the high amplitude $`\delta `$ Scuti variable, V474 Mon (HR 2107) which is quoted in the Yale Bright Star Catalogue (see King & Liu 1990) as a $`15.^\mathrm{d}492`$ spectroscopic binary, although no trace of any orbit could been found in the literature. It is also noteworthy that the quoted binary period is exactly twice the value of the period of the Blazhko-effect (Romanov & Fedotov 1979). Another example is V650 Tau (HD 23643), which is quoted by Garcia et al. (1995) as spectroscopic binary. No mention of this could be found in the literature. Abt et al. (1965) give several radial velocity measurements in the range -35 to 2 kms<sup>-1</sup>, with errors up to about 10 kms<sup>-1</sup>. Unless a detailed study is performed, this is however not a proof for binarity.
CC And is probably a binary with an orbital period of 10.469 days (Fitch, 1976) and an eccentricity of 0.12 (Fitch, 1969). The shape of the light curve does indeed vary with this period, a phenomenon attributed to tidal modulation of the fundamental by a faint companion. Fitch (1967) detected 6 pulsation frequencies in the light curve, while 7 were found more recently by Fu & Jiang (1995).
We will now further discuss some of the stars listed in Table 4.
#### UZ Lyn
The orbital elements for UZ Lyn (2 Lyn = HD 43378 = HR 2238) quoted in Table 4 are only preliminary and need confirmation. It was obtained by Scholtz et al. (1998), although in their spectroscopic run covering 240 min, the radial velocity was found to be constant. However, Caliskan & Adelman (1997) had published some radial velocity measurements, suggesting the star to be a spectroscopic binary. Combining their velocities with the one obtained by Caliskan & Adelman (1997), Scholtz et al. (1998) found three orbital solutions with periods of 21, 33 and 87 days. They kept the 21 days solution as it had the smaller residuals. Additional data are clearly called for.
#### SZ Lyn
The pulsation behaviour of SZ Lyn was discovered by Hoffmeister (1949). It was later classified as a dwarf Cepheid by Broglia (1963) and is now considered as a monoperiodic (0.<sup>d</sup>12) high-amplitude $`\delta `$ Scuti star. Moffett et al. (1988) improved the value of the pulsation period to 0.12052115 days. This period is apparently undergoing a secular change of 3 $`10^{12}`$ d/cycle (Soliman et al. 1986). McNamara (1997) derived a semi-empirical $`PL`$ relation of SX Phe and large-amplitude $`\delta `$ Scuti stars :
$$<M_v>=3.725\mathrm{log}P1.933$$
(3)
For SZ Lyn, he quotes an absolute magnitude of $`M_v`$=1.35 and a period of 0.1205 d, as well as a mass of 1.92 M and a radius of 3.18 R. Rodrรญguez et al. (1996) found that, like all other high-amplitude $`\delta `$ Scuti and SX Phe variables, it is a radial pulsator. Rodrรญguez (1999), analyzing all available photometric datasets, did not find any long-term change of amplitude of the light curve.
The binary nature of SZ Lyn was first suggested by Barnes & Moffett (1975) as an explanation for the periodic ephemeris needed to account for the observed times of maximum. The expected period was around 1146 days. The binary nature was confirmed by CORAVEL radial velocity measurements by Bardin & Imbert (1981) who obtained a preliminary eccentricity of 0.26. They also obtained a total amplitude in radial velocity of 39.9 kms<sup>-1</sup>, which, combined with their preliminary orbit, suggests a variation of 0.115 R over one pulsation cycle for the radius of the star. With additional observations, Bardin & Imbert (1984) however obtained a slightly longer (1181.5 d) and less eccentric (0.191) orbit. Using photometric data, Soliman et al. (1986) found the orbital period to be 1173.5 $`\pm `$ 2 days. Moffett et al. (1988), using both photometric and spectroscopic data, determined an orbital period of 1181.1 days and an eccentricity of 0.188. The value of the mass function, 0.101 M, implies that the unseen companion is most likely on the main sequence with a spectral type between F2 and K3, that is, in the mass range 0.7 - 1.6 M.
#### V644 Her
Like FM Vir (see below), this star was first known as a spectroscopic binary before being noticed as a variable. The first published orbits gave orbital periods of 11.<sup>d</sup>848, 11.<sup>d</sup>857, 11.<sup>d</sup>878 and 11.<sup>d</sup>851. Bardin & Imbert (1982) used CORAVEL to derive more precisely the orbital elements and obtained a period of 11.858592 days and an eccentricity of 0.365. From Hipparcos data, the absolute magnitude of the system is M<sub>v</sub> = 1.87, a possible value for a F2 IV star of about 1.5 - 2 M. The absence of the secondary from the spectra as well as from the CORAVEL trace, implies that the secondary is fainter by at least two magnitudes, corresponding to a star cooler than F6-8 V, and therefore less massive than about 1.2 M. Breger (1973) showed the variable nature of the star, with a period of 0.098 day and an amplitude of 0.02 mag. Elliott (1974) found the period to be 0.1150 day with an amplitude of 0.044 mag. It might be of interest to note that the ratio of the orbital to the pulsation period is exactly an integer value, 121 when using Bregerโs period, and close to 103 when using Elliottโs one (see also Sect. 5.).
#### GX Peg
From the results of their three weeks multi-site campaign, Michel et al. (1992) unambiguously detected five frequencies. Goupil et al. (1993) identified two radial modes of order n=2,3 and one non radial mode l=1, n=3 split by rotation. The orbital elements of GX Peg were first determined by Albitzky (1933) and by Harper (1933). Although the two sets of data were contemporary (data obtained between 1928 and 1932, and between 1926 and 1933, respectively), there is a difference of 6 kms<sup>-1</sup> in the radial velocity amplitude of the derived orbits. Bolton & Geffken (1976) obtained 25 spectrograms betwen 1971 and 1974 to derive a new orbit. They also recomputed the orbit from Albitzky and Harper data. Although they found a good agreement betwen Albitzkyโs orbit and theirs, the significant discrepancy between the periastron longitudes lead them to conclude to the possibility of apsidal motion with a period of about 260 years. The very small eccentricity of the orbit makes this possibility unlikely. The Lucy & Sweeney (1971) test indeed indicates that the eccentricity is compatible with zero. As the typical time to have synchronization between orbital and rotation motion is smaller than the time needed to circularize an orbit, this small orbital period (2.34 days) system has certainly achieved synchronization. The rotational velocity therefore implies a value of R $`\mathrm{sin}i2.75`$ R. Goupil et al. (1993) concluded that the rotational splitting of the modes are indicative of the fact that the star cannot rotate as a a solid body and is thus not synchronized down to the core.
### 4.2. Double-lined spectroscopic binaries
Table 5 gives the orbital elements of those $`\delta `$ Scuti stars which are classified as SB2 and for which we could find details in the literature. The outline of the table is similar to Table 4 except that we mention the two spectral types when available, as well as two semi-amplitudes of radial velocity. We also list the mass ratio instead of the mass function.
To this list, we should also add BQ Cnc (HD 73729) which has been classified as an SB2 (Abt & Biggs, 1972) but for which no orbital elements are known. Again, some cases of either misclassification, either lack of precise elements can be mentioned. 56 Ser (HD 160613) for example, is listed in the Bright Star Catalogue as having two spectra, while only one spectrum is visible on the Parkins plate (Slettebak, 1954). On the other hand, DL UMa (HD 82620) is quoted by Henriksson (1979) as an eclipsing binary with a period of $`0.^\mathrm{d}42`$, while he previously (Henriksson 1977) considered it as a $`\delta `$ Sct variable with an amplitude of 0.056 mag and a period of $`0.^\mathrm{d}0831`$. If confirmed, the almost exact integer ratio between the pulsation period and the orbital period might be of great interest.
When studying CE Oct (HD 188520), Kurtz (1980) discovered a second period of $`0.^\mathrm{d}21`$ which may result from a g-mode pulsation or ellipsoidal variability. Morris (1985) rejected the latter as being the least likely.
#### $`\theta `$ Tuc
Cousins & Lagerwey (1971) were the first to notice the variability of $`\theta `$ Tuc and to derive a period of variation around 70-80 minutes. Later, Stobie & Shobbrook (1976) classified the star as a $`\delta `$ Sct star and Kurtz (1980) determined a set of 8 stable frequencies. This was later extended to 10 highly-stable frequencies by Paparรฒ et al (1996). From their high-resolution spectra, De Mey et al. (1998) could derive 4 frequencies, the most significant one corresponding to the main photometric frequency. They showed that this pulsation mode is radial. A new frequency, $`f_2=18.82`$ c/d was found, not appearing in the photometric data. It must therefore correspond to a high degree pulsation mode.
Paparรฒ et al (1996), noting periodicity of the long-term variations, suggested for the first time that $`\theta `$ Tuc might be member of a binary system. This was confirmed by Sterken (1997) which showed the star to be in a non-eclipsing binary system with ellipsoidal variations, with a period of 7.<sup>d</sup>04 and a mass ratio of about 0.1-0.15. Both the primary and the secondary minima were observable in the light curve. De Mey, Daems & Sterken (1998) made an extensive study of this object. Using high-resolution spectroscopy, they could classify the system as a double-lined spectroscopic binary with a circular orbit and they derived the orbital elements listed in Table 5. The orbital period (7.1036 d) is not very far from the previously determined photometric period. The resulting mass ratio is $`q=0.0896`$, a rather low value for a SB2. De Mey et al. (1998) concluded that the mass of the components are constrained by $`M_1<4.3M_{}`$ and $`M_2<0.4M_{}`$, while the radii of the primary and secondary lie between 1.7 $`R_{}`$ and 2.6 $`R_{}`$, and 1.6 $`R_{}`$ and 2.2 $`R_{}`$, respectively. One can therefore believe that $`\theta `$ Tuc is a post-mass transfer binary, in which the secondary is probably the remnant of an Algol-like mass-losing star. Sterken (1997) emphasizes the strong similarities between the orbital light curves of $`\theta `$ Tuc and HD 96008, a system with a very small mass ratio. It has to be noted however that while in the case of HD 96008, one of the component nearly fills its Roche lobe, this is not the case for $`\theta `$ Tuc. Using De Mey et al. (1998) results, one can see that the primary lies well inside its Roche lobe, while the secondary fills maybe only about 50 % of it. It is also noteworthy that $`\theta `$ Tuc seems to show a rotational velocity (v$`\mathrm{sin}i`$=80 kms<sup>-1</sup>) too large for synchronization between the orbital and the rotational motion.
#### FM Vir
FM Vir (=32 Vir) is one of the very few Am stars reported as a pulsating star. FM Vir was recognized as a metallic-line star by Roman, Morgan & Eggen (1948). The Am phenomenon is generally attributed to slow rotation and the effect of diffusion (e.g. Abt & Morrell, 1995). Bertiau (1957) quotes v$`\mathrm{sin}i`$ of 20 km<sup>-1</sup> for FM Vir. The velocity variation was detected by Adams (1914) and the first orbit was determined by Cannon (1915) who found an orbital period of 38.3 days and also detected double-lines. Petrie (1950) confirmed the double-lined nature of FM Vir and derived a magnitude difference between the two components of 0.43 $`\pm `$ 0.11 mag. Bertiau (1956) derived a new orbit, with an orbital period of 38.3 days, a rather typical value for Am stars. In the spectral region in which he was observing ($`\lambda \lambda `$ 4350-4650), he could not observe the lines of the fainter star. However, the lines of the primary are shallow, as if filled in by the continuous spectrum of a companion. He also observed a fairly large scatter of the individual velocities around the mean velocity curve, something which can now be explained by the variability of the component. Bartolini, Grilli & Parmeggiani (1972) found the star FM Vir to be variable with an amplitude ranging from $`0.^\mathrm{m}02`$ to $`0^\mathrm{m}.05`$ and a period of about $`0.^\mathrm{d}07`$. Bartolini et al. (1983) confirmed that the star is pulsating with a strongly variable amplitude. The period P=$`0.^\mathrm{d}07188`$ has the highest amplitude and is constant. Kurtz et al. (1976) performed a photometric and spectroscopic study of FM Vir. Their light curves had an observed range in the visual amplitudes between 0.01 and 0.035 mag, while the derived periods for the different nights range from 0.07 to 0.084 days, with an average of $`0.^\mathrm{d}0756`$, clearly suggesting the $`\delta `$ Scuti character of the star. The Am character of the star was also confirmed. They also concluded that FM Vir is a double-lined binary system and obtained v$`\mathrm{sin}i`$ values of 24 $`\pm `$ 6 kms<sup>-1</sup> and 140 $`\pm `$ 25 kms<sup>-1</sup> for the primary and the secondary, respectively. Therefore, while the primary has a rotation slow enough for the Am phenomenon to appear, the secondary is above the observed cutoff for Am stars. The secondary must then have normal abundances. The properties of the components place them both inside the instability strip. Because Breger (1970), from an extensive survey, found that classical Am stars do not pulsate, the suggestion was made that it is the secondary which pulsates. Assuming a magnitude difference between the two components of 0.43 mag and a light variability of the system of 0.035 mag indicates that the pulsational amplitude of the secondary should be 0.09 mag. Mitton and Stickland (1979) suggested that the primary is an evolved Am star (or $`\delta `$ Delphini star) of subgiant luminosity, which pulsates, while the secondary is a main-sequence star of spectral type near A7. They were able to detect the secondary in their spectra and to derive a magnitude difference between the two components between 0.6 and 0.9 mag. A semi-amplitude for the secondary could also be determined, leading to a mass ratio of 0.92 $`\pm `$ 0.03. This, combined with the mass function of the system, implies that the system must be nearly edge-on for the masses not to be unrealistically large. Masses around 2.05 M and 1.89 M are derived for the primary and the secondary, respectively.
FM Vir is clearly a system of great interest. The fact that we have two components of almost the same mass, but with different chemical composition should lead to a better understanding of stellar evolution mechanisms. Although it is clear that the primary owns its Am status to its slow rotation, one could wonder why two very similar stars in a close binary would have a different rotation. This should certainly not be the case if the slow rotation was due to tidal effects. However, the orbital period might be too large for synchronization to have taken place (e.g. Levato 1976). The clue may lie in the subgiant status of the primary : being more evolved, the starโs rotation decreased. This should not be much more than a factor 2 however, so that we conclude that there was already a large difference in rotational velocity between the two components when the system was formed.
#### $`\delta `$ Del
The spectroscopic binary nature of $`\delta `$ Del was discovered by Frost (1924). Eggen (1956) discovered its photometric variability with a period of 0.<sup>d</sup>13505 and a variable range in brightness, with a mean around 0.05 mag. Spectroscopic observations by Struve, Sahade & Zebergs (1957) indicated that the radial-velocity is variable with a period of 0.<sup>d</sup>13447.
$`\delta `$ Del is the prototype of a sub-class among the $`\delta `$ Scuti stars with metal-line subgiant or giant spectra which might be characteristic of evolved Am stars (Breger 1979). Preston (1973, quoted in Reimers 1976) found the star to be an eccentric double-lined spectroscopic binary with a 40.5 days orbital period. Both components are $`\delta `$ Scuti variables and Reimers (1976) showed that they have identical chemical compositions, i.e. all metals up to the iron group are deficient relative to the Sun by a factor 2, while the abundance of heavy elements (Sr, Y, Zr, Ba, Ce, La and Eu) are enhanced by factors between 4 and 8 relative to iron. Smith (1982) considered the fact that both components could be $`\delta `$ Scuti stars and concluded that comp A is probably a monoperiodic radial pulsator while comp B, with the strongest $`\delta `$ Del peculiarity, oscillates in a mixture of radial and non-radial modes. Baade et al. (1993) confirmed that $`\delta `$ Del is a SB2 with well separated and almost identical spectra.
### 4.3. Eclipsing binaries
Among the stars quoted in Tables 4 and 5, some are eclipsing systems. Such systems are very powerful tools in astrophysics as they allow the determination of the individual masses and radii of the components.
Here again, some misleading examples appeared in the literature. For example, RY Lep is among the list of southern eclipsing binaries observed by Popper (1966) but he could only observe sharp single lines. Diethelm (1985) showed that this star is actually a relatively bright high-amplitude $`\delta `$ Scuti star with a period of $`0.^\mathrm{d}2254`$ and an amplitude of 0.35 mag.
#### AB Cas
As reviewed by Rodrรญguez et al. (1998), AB Cas is one of the clear examples where the light curves simultaneously and clearly show both types of variability: binarity and pulsation. It is an Algol-type binary system with a period of $`1.^\mathrm{d}3668`$, while the primary component is a monoperiodic $`\delta `$ Sct star with a pulsation period of $`0.^\mathrm{d}0583`$. The star is pulsating in the fundamental radial mode. Rodrรญguez et al. (1998) using a previously determined value for the mass ratio of 0.22 and assuming the secondary to fill the Roche lobe, found the parameters quoted in Table 6.
#### Y Cam
Together with AB Cas this is one of the oldest $`\delta `$ Sct variables known to be in an eclipsing system: its variability was discovered in 1903 by Mrs. Ceraski. Its components (A9 IV and K1 IV) have rather different masses (1.9 M and 0.4 M) but rather similar radii (3.15 R and 3.05 R, respectively). The light curve is very reminiscent of the Algol-type and the system is thought to be semi-detached, with the K star filling its Roche lobe. Shapley (1917), using the data of Miss Harwood, computed the first orbital elements of Y Cam, although a light curve was already obtained by Nijland as quoted in the catalogue of Shapley (1913). Dugan (1924) obtained a complete visual light curve and computed a solution for the system. It was shown that the dimensions of the two components were nearly the same. Szcepanowska (1955), on the basis of 176 times of minimum, found it necessary to introduce a sinusoidal term into the elements of Y Cam which Plavec et al. (1961) thought to be only the main part of the whole period variation. They moreover concluded that the large periodic term could not be due to a third body nor to apsidal motion. Y Cam nevertheless appeared in the catalogue of systems with apsidal motions of Petrova & Orlov (1999), with an apsidal period of 60 years. Broglia & Marin (1974) revealed the primary to be a $`\delta `$ Sct star with a period of $`0.^\mathrm{d}0634697`$ and a variable light amplitude. The amplitude variation is not correlated with the phase in the orbital motion which led them to conclude that the โamplitude variations in $`\delta `$ Sct stars are not necessarily caused by a companionโ. Broglia & Marin (1974) confirmed the variation of the orbital period, which they considered as a proof for mass transfer in the system.
## 5. Discussion
Does the binarity influence the pulsation properties of $`\delta `$ Scuti stars? In the light of the previous sections, it is obvious that there is no clear and easy answer. Every discussed case seems to be particular in its own way. Apart from the binarity, many other phenomena are involved that may even have a stronger impact on the pulsation properties of the stars: evolution, chemical composition, rotational effectsโฆ In an effort to generalize, some authors consider the ratio between the pulsation period and the orbital period in order to search whether a resonance mechanism between both can occur (e.g. Frolov et al. 1980, Tsvetkov & Petrova 1993). This is however only possible if both periods are determined with very high accuracy. For example, in the case of SZ Lyn, the ratio between both periods is almost exactly 9800 (taking the orbital period from Moffett et al. (1988)). But with a slightly different value of the orbital period (1181.5 instead of 1181.1), this ratio, 9803.26, is not indicative. Such an exercise is therefore only applicable to very close systems for which tidal interactions are expected to be important! We computed the ratio between the orbital and pulsation periods for the closest binary systems of our list (with orbital periods up to 20 days) and, in general, we do not find values close to an integer. Five systems however form an exception: KW Aur, CC And, DL Uma, V644 Her and WX Eri. DL Uma and WX Eri have very short orbital periods (0.42 and 0.82 days, respectively) and in both cases, this is almost exactly 5 times the pulsation period. For KW Aur, CC And and V644 Her, this ratio is 43, 84 and 103, respectively. It is not clear whether these last values have any physical meaning at all.
In very close systems tidal forces may force the system towards synchronization. For spectral types typical of $`\delta `$ Scuti stars, this happens for binary systems with orbital periods of a few days up to 10 days if the stars are evolved. Synchronization will slow down the star. Thus, an indirect effect of binarity may be the change of the rotational velocity. Possible correlations between the amplitude of the pulsation and the rotation have been discussed before (e.g. Breger 1980; Solano & Fernley 1997). Such a correlation is also seen in the data presented here.
Some cases present better evidence than others: e.g. amplitude modulation in the case of the radial pulsator $`\theta `$ Tuc (De Mey et al. 1998) or possible frequency splitting in the case of the non-radial pulsator KW Aur (Fitch & Wiลniewski 1979) but other effects may also occur (e.g. frequency shifts) that are almost impossible to detect. We need better models to check whether any of the above presented assumptions on the link between binarity and pulsation are valid and whether these can explain the observational facts in the most interesting cases.
Nevertheless, we should carefully investigate $`\delta `$ Scuti stars in binary systems from the observational point-of-view since they provide additional constraints on the physical parameters of the pulsating star, and therefore also on the characteristics of the pulsation. For single field $`\delta `$ Scuti stars, the position in the HR diagram and thus the evolutionary phase may be ambiguous, especially at the end of the core H burning phase, while $`\delta `$ Scuti stars in binary systems can be located with much better accuracy. It may also be especially worthwhile to study the differences in variability between two nearly identical components of a binary system, of which one or both may be $`\delta `$ Scuti stars. Various such cases have appeared in the present discussions and many deserve further observations and study.
##### Acknowledgments.
This research made use of the Simbad database, operated at CDS, Strasbourg, France, as well as of the NASA Astrophysics Data System. We also acknowledge funding by project G.00265.97 of the Fonds voor Wetenschappelijk Onderzoek (FWO, Belgium).
## References
Abt, H.A. 1965, ApJS, 11, 429
Abt, H.A., Biggs, E.S. 1972, in Bibliography of Stellar Radial Velocities, Kitt Peak National Observatory
Abt, H.A., Levy, S.G. 1985, ApJS, 59, 229
Abt, H.A., Morrell, N.I. 1995, ApJS, 99, 135
Adams, W.S. 1914, PASP, 26, 261
Adelman, S.J. 1997, A&AS, 125, 497
Albitzky, V.A. 1933, Pulkova Circ. No. 8
Andersen, J. 1991, Astron.Astrophys.Review, 3, 91
Andrievsky, S.M., Chernyshova, I.V., Usenko, I.A., Kovtyukh, V.V., Panchuk, V.E., & Galazutdinov, G.A. 1995, PASP, 107, 219 (AC95)
Baade, D., Bardelli, S., Beaulieu, J.Ph., Vogel, S. 1993, A&A, 269, 195
Baize, P. & Petit, M. 1989, A&AS, 77, 497 (BP89)
Bardin, C., Imbert, M. 1981, A&A, 98, 198
Bardin, C., Imbert, M. 1982, A&AS, 47, 319
Bardin, C., Imbert, M. 1984, A&AS, 57, 249
Barnes, T.G., Moffett, T.J. 1975, AJ, 80, 48
Bartolini, C., Grilli, F., Parmeggiani, G. 1972, IBVS 704
Bartolini, C., Grilli, F., Parmeggiani, G., Piccioni, A., Silveri, P. 1983, A&AS, 53, 139
Batten, A.H., Fletcher, J.M., MacCarthy, D.G. 1989, Publ. Dom. Astrophys. Obs. Victoria, 17, 1
Beardsley, W.R., Zizka, E.R. 1977, Rev.Mexicana de Astro. y Astrofisica, 3, 109
Bertiau, S.J. 1957, ApJ, 125, 696
Bolton, C.T., Geffken N. 1976, PASP, 88, 195
Brancewicz, H.K.., Dworak, T.Z. 1980, Acta Astron., 30, 501
Breger, M. 1970, ApJ, 162, 597
Breger, M. 1973, A&A, 22, 247
Breger, M. 1979, PASP, 91, 5
Breger, M. 1980, ApJ, 235, 153
Breger, M., Garrido, R., Huang, L., Jiang, S-.Y., Guo, Z.-H., Frueh, M., Paparo, M. 1989, A&A, 214, 209 (BG89)
Broglia, P. 1963, Mem.Soc. Astron.Ital., 34, 431
Broglia, P., Marin, F. 1974, A&A, 34, 89
Budaj, J. 1996, A&A, 313, 513
Budaj, J. 1997, A&A, 326, 655
Caliskan, H., Adelman, S.J. 1997, MNRAS, 288, 501
Cannon, J.B. 1915, Pub.Dom.Obs.Ottawa. 2, 369
Chaubey, U.S. 1983, IBVS, 2348
Chernyshova, I.V., Andrievsky, S.M., Kovtyukh, V.V. & Mkrtichian, D.E. 1998, Contrib. Astron. Obs. Skalnatรฉ Pleso 27, 332
De Mey, K., Daems, K., Sterken, C. 1998, A&A, 336, 527
Diethelm, R. 1985, A&A, 149, 465
Dommanget, J., & Nys, O. 1995, Bull. DโInf. du CDS de Strasbourg 46, 3
Dugan, R.S. 1924, Princeton Contr. No.6
Duquennoy, A. & Mayor, M. 1991, A&A, 248. 485
Eggen, O.J. 1956, PASP, 68, 541
Elliott, J.E. 1974, AJ, 79, 1082
ESA 1997, The Hipparcos and Tycho Catalogues, ESA SPโ1200
Fehrenbach, C., Duflot, M., Burnage, R., Mannone, C., Peton, A., & Genty, V. 1987, A&AS, 71, 275
Fitch, W.S. 1967, ApJ, 148, 481
Fitch, W.S. 1969, ApJ, 158, 269
Fitch, W.S. 1976, IAU Coll. 29, p. 168
Fitch, W.S., & Wiลniewski, W.Z. 1979, ApJ, 231, 808 (FW79)
Frandsen S., Jones A., Kjeldsen H., Viskum, M., Hjorth, J., Andersen, N.H., & Thomsen, B. 1995, A&A, 301, 123 (FJ95)
Frasca, A., Catalano, S., & Mantovani, D. 1997, A&A, 320, 101
Frolov, M.S., Pastukhova, E.N., Mironov, A.V., Moshkalev, V.G. 1980, IBVS 1894
Frolov, M.S., Pastukhova, E.N., Mironov, A.V. 1982, IBVS 2117
Frost, E.B. 1924, ApJ, 60, 319
Fu, J.-N., Jiang, S.Y. 1995, A&AS, 110, 303
Garcรญa, J.R., Cebral, J.R., Scoccimarro, E.R., Wahnon, P., Arena, R., Bazterra, V., Pellizza, L., Risi, A., Rodr\]ฬiguez, M.L., Zimmermann, M. 1995 A&AS, 109, 201
Goupil, M.J., Michel, E., Lebreton, Y., Baglin, A. 1993, A&A, 268, 546
Grenier, S., Burnage, R., Faraggiana, R., Gerbaldi, M., Delmas, F., Gomez, A.E., Sabas, V.,& Sharif, L. 1999, A&AS, 135, 503
Harper, W.E. 1933, Pub. Dom. Astrophys. Obs., 6, 203
Harper, W.E. 1938, Pub. Dom. Astrophys. Obs., 6, 207
Hartkopf, W.I., McAlister, ,H.A., & Mason, B.D. 1999, CHARA Contrib. No. 4, Georgia State University, Atlanta
Hartkopf, W.I., & McAlister, ,H.A. 1989, AJ, 98, 1014
Hillenbrand, L.A., Massey, Ph., Strom, S.E., & Merrill, K.M. 1993, AJ, 106, 1906
Hoffmeister, C. 1949, Astron. Abhand. 12, 1
Hummel, C.A. & Armstrong, J.T. 1992 Complementary Approaches to Double and Multiple Star Research, ASP Conference Series, Vol. 32, ed. H.A. McAlister & W.I. Hartkopf, p.552
Kaye, A.B. 1999, IBVS, 469
Kennelly, E.J. & Walker, G.A.H. 1996 PASP, 108, 327 (KW96)
Kennelly, E.J., Walker, G.A.H., Catala, C., Foing, B.H., Huang, L., Jiang, S., Hao, J., Zhai, D., Zhao, F., Neff, J.E., Houdebine, E.R., Ghosh, K.K., Charbonneau, P. 1996, A&A, 313, 571
King, J., Liu, T. 1990, PASP, 102, 328
Krรณlikowska, M. 1992 A&A, 260, 183
Kurtz, D.W. 1980, MNRAS, 193, 51
Kurtz, D.W., Breger, M., Evans, S.W., Sandmann, W.H. 1976, ApJ, 207, 181
Lampens, P.,& Van Camp, M. 1999, this volume (LV99)
Lehmann H., Scholz, G., Hildebrandt, G., & Yang, S. 1997, A&A, 327, 167
Levato, H. 1976, ApJ, 203, 680
Li, Z.-P., Zhou, A.-Y., & Yang, D. 1997 PASP, 109, 217
Lindegren, L. 1997, ESA SP-402, 13
Liu Y.Y., Jiang, S.Y., Michel, E., Hernรกndez, M.M., Chevreton, M., Auvergne, M., Mangeney, A., Belmonte, J.A., Goupil, M.J., Baglin, A., Roca Cortรฉs, T., & Vidal, I. 1996, A&AS, 120, 179
Lopez de Coca, P., Costa, V., Rolland, A., & Walker, E.N. 1979, A&AS, 36, 61
Lowder, W.M. 1989, JAVSO 18, 116
Lucy, L.B., Sweeney, M.A. 1971, AJ, 76, 544
MacNamara, D.H. 1997, PASP, 109, 1221
Mellor L.L. 1917, Pub. Obs. Univ. Mich. 3, 61
Mitton, J., Stickland, D.J. 1979, MNRAS, 186, 189
Michel, E., et al. 1992, A&A, 255, 139
Michel, E., Hernรกndez, M.M., Houdek, G., Goupil, M.J., Lebreton, Y., Pรฉrez Hernรกndez, F., Baglin, A., Belmonte, J.A., & Soufi, F. 1999, A&A, 342, 153
Moffett, J.J., Barnes III, T.G., Fekel, F.C.Jr, Jefferys, W.H., Achtermann, J.M. 1988, AJ, 95, 1534
Morris, S.L. 1985, ApJ, 295, 143
Pan, X., Shao, M., Colavita, M.M. 1992 Complementary Approaches to Double and Multiple Star Research, ASP Conference Series, Vol. 32, ed. H.A. McAlister & W.I. Hartkopf, p.502
Paparรณ M., Jiang S.-Y., Li Z.-P., Kollath Z. 1990, IBVS 3541
Petrie, R.M. 1950, Pub. Dom. Astrophys. Obs. Victoria, 8, 319
Petrova, A.V.,, Orlov, V.V. 1999, AJ, 117, 587
Plavec, M., Smetanova, M., Pekny, Z. 1961, Bull. Astron. Inst. Czechoslovakia, 12, 117
Popper, D.M. 1966, AJ, 71, 153
Reimers, D. 1976, A&A, 53, 377
Riboni E., Poretti E., & Galli G. 1994, A&AS, 108, 55
Rodrรญguez, E. 1999, PASP, 111, 709
Rodrรญguez, E., Rolland, A., Lopez de Coca, P., Garrido, R., & Gonzalez-Bedolla, S.F. 1992, A&AS, 96, 429
Rodrรญguez, E., Rolland, A., Lopez de Coca, P., Martin, S. 1996, A&A, 307, 539
Rodrรญguez, E. Claret, A., Sedano, J.L., Garcรญa, J.M., Garrido, R. 1998, รฅ, 340, 196
Roman, N.G., Morgan, W.W., Eggen, O.J. 1948, ApJ, 107, 107
Romanov, Yu. S., Fedotov, Yu. T. 1979, Astron.Tsirk., No. 1071
Rosvick, J.M., & Scarfe, C.D. 1991, PASP, 103, 628
Scholtz, G., Lehmann, H., Hildebrandt, G., Panov, K., Iliev, L. 1998, รฅ, 337, 447
Seeds, M.A., Yanchak, G.A. 1972, IBVS, 625
Shapley, H. 1913, ApJ, 38, 158
Shapley, H. 1917, ApJ, 46, 56
Slettebak, A. 1954, ApJ, 119, 146
Smith, M.A. 1982, ApJ, 254, 242
Sรถderhjelm, S. 1999, A&AS, 341, 121
Solano, E., Fernley, S. 1997, A&AS, 122, 131
Soliman, M.A., Hamdy, M.A., Szeidl, B., Szabados, L. 1986, Commun. Konkoly Obs. Hung. Acad. Sci., Budapest, No.88
Struve, O., Sahade, J., Zebergs, V. 1957, ApJ, 125, 692
Szatmรกry, K. 1990, J. AAVSO, 19, 52
Szcepanowska, A. 1955, Acta Astron., 6, 134
Tokovinin, A. 1997, A&AS, 121, 71
Torres, G., Stefanik, R.P., & and Latham, D.W. 1997, ApJ, 485, 167 (TSL97)
Tsvetkov, TS.G., Petrova, TS.C. 1993, Ap&SS, 203, 257
van Leeuwen, F. 1997, ESA SP-402, 19
Vivekananda-Rao, P., Sarma, M.B.K., Agrawal, P.C., & Apparao, M.V.K. 1985, IBVS, 2721
Weiss, W.W., et al. 1998, A&A, 338, 919
Yang, S., Walker, G.A.H., Fahlman, G.G., & Campbell, B. 1982, PASP, 94, 317 |
warning/0001/hep-lat0001034.html | ar5iv | text | # 1 Introduction
## 1 Introduction
The investigation of the confinement problem is now one of the important subjects of QCD. One of the most popular schemes of confinement is the dual superconductor mechanism. According to this mechanism the quark- antiquark string appears in the dual representation of the theory in the way similar to the appearance of the Abrikosov string in Ginsburg - Landau theory. But now it is not clear how the analogue of the Ginsburg - Landau theory can be derived from the QCD.
The relativistic analogue of Ginsburg - Landau theory is the abelian Higgs model. Thus the dual superconductor theory should be the nonlocal variant of abelian Higgs model. There exists the point of view that the field correspondent to the Maximal Abelian monopoles should play the role of Higgs field. The fact that the condensate of that monopoles is the order parameter and is not equal to zero in the confinement phase partially justifies this hypothesis. This picture seems to be quite natural in the $`SU(2)`$ theory, but in the $`SU(3)`$ theory two maximal abelian monopoles appear and the dual superconductor mechanism becomes too complex.
Recently the approach alternative to the Maximal Abelian projection was investigated. That approach is the Maximal Center projection. It occurs that the Center dominance also takes place as the Abelian dominance . And furthermore in the Maximal Center projected theory also one can construct the monopoles which are condensed in the confinement phase . The pleasant feature of that monopole is that there is only one center monopole in the $`SU(3)`$ theory.
In this work we derive the representation of the Maximal Center projected gluodynamics which has the form of the dual superconductor theory. The field of the center monopole plays the role of the Higgs field. The quark \- antiquark string appears as the topological defect in this theory. Also we construct the monopole creation and annihilation operators.
## 2 The Maximal Center Projection.
The Maximal Center projection is the procedure of partial gauge fixing. The gauge ambiquity is used to make all the link variables $`USU(3)`$ as close as possible to the elements of the center $`Z_3`$ of $`SU(3)`$:
$`Z_3=\{\mathrm{diag}(\mathrm{e}^{(2\pi i/3)N},\mathrm{e}^{(2\pi i/3)N},\mathrm{e}^{(2\pi i/3)N}\}`$, where $`N\{1,0,1\}`$.
There is a lot of versions of the Maximal Center projection. That versions differ from each other by the choice of the gauge fixing potential $`O(U)`$. The projection is achieved by the minimizing of the potential $`O(U)`$ with respect to the gauge transformations $`U_{xy}g_xU_{xy}g_y^1`$.
This gauge condition is invariant under the central subgroup $`Z_3`$ of $`SU(3)`$. Thus the Maximal center projected theory is nonlocal $`Z_3`$ gauge theory. The well - known center vortices are defined as follows. After fixing the Maximal Center gauge we define the integer-valued link variable $`N`$:
$`N_{xy}=0`$ $`\mathrm{if}`$ $`(\mathrm{Arg}(U_{11})+\mathrm{Arg}(U_{22})+\mathrm{Arg}(U_{33}))/3]\pi /3,\pi /3],`$
$`N_{xy}=1`$ $`\mathrm{if}`$ $`(\mathrm{Arg}(U_{11})+\mathrm{Arg}(U_{22})+\mathrm{Arg}(U_{33}))/3]\pi /3,\pi ],`$
$`N_{xy}=1`$ $`\mathrm{if}`$ $`(\mathrm{Arg}(U_{11})+\mathrm{Arg}(U_{22})+\mathrm{Arg}(U_{33}))/3]\pi ,\pi /3].`$ (1)
In other words $`N=0`$ if $`U`$ is close to $`1`$, $`N=1`$ if $`U`$ is close to $`\mathrm{e}^{2\pi i/3}`$ and $`N=1`$ if $`U`$ is close to $`\mathrm{e}^{2\pi i/3}`$. Next we define the plaquette variable:
$$\sigma _{xywz}=N_{xy}+N_{yw}N_{zw}N_{xz}$$
(2)
In terms of the calculus of the differential forms on the lattice this equation looks like
$$\sigma =dN$$
(3)
Then we introduce the dual lattice and define the variable $`\sigma ^{}`$ dual to $`\sigma `$: if plaquette $`{}_{}{}^{}\mathrm{\Omega }`$ is dual to plaquette $`\mathrm{\Omega }`$, then $`\sigma _{}_{}{}^{}\mathrm{\Omega }^{}=\sigma _\mathrm{\Omega }`$. One can easily check that the variable $`\sigma `$ represents a closed surface. This surface is known as the worldsheet of the center vortex.
We express the $`SU(3)`$ gauge field $`U`$ as the product of $`\mathrm{exp}((2\pi i/3)N)`$ and $`V`$, where $`V`$ is the $`SU(3)/Z_3`$ variable $`(\mathrm{Arg}(V_{11})+\mathrm{Arg}(V_{22})+\mathrm{Arg}(V_{33}))/3]\pi /3,\pi /3]`$. Then $`U=\mathrm{exp}((2\pi i/3)N)V`$
After that we represent the action of the Wilson loop $`C`$ as follows:
$$W_C=\mathrm{\Pi }_CU=\mathrm{exp}((2\pi i/3)L(C,\sigma ))\mathrm{\Pi }_CV$$
(4)
The term $`(2\pi i/3)L(C,\sigma )`$ is known as the Aharonov - Bohm interaction term. The quantity $`L(C,\sigma )`$ is the linking number of the loop $`C`$ and the closed surface $`\sigma ^{}`$.
The center dominance means that after the Maximal Center projection the Aharonov - Bohm interaction term causes confinement and produces the full string tension.
The center monopole is just the $`Z_3`$ analogue of the monopole in $`U(1)`$ theory. Let us recall that monopoles in $`U(1)`$ theory are constructed as loops on which the force lines of the gauge field end. It is well known that in electrodynamics the Maxwell equations $`dF=0`$ restrict the existence of magnetic charges. But in the compact theory values of F which differ from each other by $`2\pi `$, are equivalent. Thus the correct field strength is $`F\mathrm{mod}\mathrm{\hspace{0.17em}2}\pi `$ and $`{}_{}{}^{}d(F\mathrm{mod}\mathrm{\hspace{0.17em}2}\pi )=2\pi j_m`$, where $`j_m`$ is the monopole current.
In the $`Z_3`$ theory $`\sigma =dN`$ is the analogue of the field strength $`F=dA`$. The Aharonov - Bohm interaction between the center vortex and the quark depends only on $`[\sigma ]\mathrm{mod}\mathrm{\hspace{0.17em}3}`$. The variable $`[\sigma ]\mathrm{mod}\mathrm{\hspace{0.17em}3}`$ represents the surface with boundary. This boundary is a closed line. We assume that this line represents the world trajectory of the particle, which we call a center monopole:
$$3j_m=^{}d([\sigma ]\mathrm{mod}\mathrm{\hspace{0.17em}3})=\delta ([\sigma ^{}]\mathrm{mod}\mathrm{\hspace{0.17em}3}).$$
(5)
## 3 The derivation of the dual superconductor theory
We start with the Wilson loop average in the $`SU(3)`$ theory.
$$<W(C)>=DUexp(\underset{plaq}{}\beta (11/3ReTrU_{plaq})ReTr\mathrm{\Pi }_CU$$
(6)
To consider the Maximal Center projection we have to use the Faddeev-Popov unity
$$1=\underset{\alpha \mathrm{}}{lim}Dgexp(\alpha O(g_xU_{xy}g_y^1))\mathrm{\Delta }_{FP}(U,\alpha )$$
(7)
Here the gauge condition is that we assume $`U`$ to provide a minimum of the functional $`O(U)`$.
This functional is the given measure of the distance in the functional space between the matrix function $`U`$ and the set of functions, which attach one of the center elements of $`SU(3)`$ to each link. It is obvious, that the functional $`O`$ is invariant under the remaining $`Z_3`$ symmetry.
As in the previous section we consider the following representation of the projected $`U`$;
$$U=e^{2\pi iN/3}V$$
(8)
where $`V`$ is closer to $`1`$ than to other center elements.
It follows from the invariance of $`O`$ under the $`Z_3`$ symmetry, that the Faddeev-Popov determinant does not depend upon $`N`$.
Thus we have
$`<W(C)>=lim_\alpha \mathrm{}{\displaystyle D_{VSU(3)/Z_3}V\underset{N=1,0,1}{}}`$
$`exp({\displaystyle \underset{plaq}{}}\beta (11/3ReTre^{2\pi dN/3}V_{plaq})`$
$`\alpha O(V)+(2\pi i/3)(N,C))\mathrm{\Delta }_{FP}(V,\alpha )ReTr\mathrm{\Pi }_CV`$ (9)
The center dominance means, that we can omit the expression $`ReTr\mathrm{\Pi }_CV`$ to calculate the correct string tension. Thus we are considering the following expression for the $`Z_3`$ Wilson loop:
$`<Z(C)>=lim_\alpha \mathrm{}{\displaystyle D_{VSU(3)/Z_3}V\underset{N=1,0,1}{}}`$
$`exp({\displaystyle \underset{plaq}{}}\beta (11/3ReTre^{2\pi dN/3}V_{plaq})`$
$`\alpha O(V)+(2\pi i/3)(N,C))\mathrm{\Delta }_{FP}(V,\alpha )`$ (10)
After the integration over short - ranged field $`V`$ we obtain the resulting $`Z_3`$ theory with nonlocal action
$$<Z(C)>=\underset{N=1,0,1}{}exp(S(dNmod3)+(2\pi i/3)(N,C))$$
(11)
It seems that this theory behaves like an usial $`Z_3`$ theory in the confinement phase.
Now we are going to transfer ourselves into the dual representation of the above theory to understand how the superconductor appears. First let us remember, that the center monopoles are defined as $`j=1/3^{}d(dNmod3)`$.
Then we apply the dual transformation.
$`<Z(C)>={\displaystyle \underset{N=1,0,1}{}}{\displaystyle \underset{m=1,0,1}{}}{\displaystyle \underset{j,n,\delta j=0}{}}exp(S(dNmod3)`$
$`+(2\pi i/3)(N,C))\delta (mdN3n)\delta (3^{}jdm)`$ (12)
We use the formulas
$`{\displaystyle \underset{k=0,1,1}{}}e^{(2\pi i/3)(Z,k)}={\displaystyle \underset{n}{}}\delta (Z3n);`$
$`{\displaystyle _\pi ^\pi }๐he^{ihZ}=\delta (Z),`$ (13)
to obtain:
$`<Z(C)>={\displaystyle \underset{N,m,k=1,0,1}{}}{\displaystyle \underset{j,\delta j=0}{}}{\displaystyle _\pi ^\pi }Dhexp(S(dNmod3)`$
$`+(2\pi i/3)(N,C)+(mdN,k)2\pi i/3+i(h,3^{}jdm))`$ (14)
Then we use the expression $`C=\delta A[C]`$, where $`A`$ is some surface, spanned on the quark loop.
$`<Z(C)>={\displaystyle \underset{N,m,k=1,0,1}{}}{\displaystyle }\delta j=0{\displaystyle _\pi ^\pi }Dhexp(S(m)`$
$`+(m,\delta h+(2\pi /3)(A[C]+k))(2\pi i/3)(N,\delta k)+i(h,3^{}j))`$ (15)
We can perform the summation over $`N`$ to obtain the constraint $`\delta k=3l`$ for some integer $`l`$. Also we can perform the summation over $`m`$, obtaining
$$exp(Q(f))=\underset{m=1,0,1}{}exp(s(m)+i(m,f))$$
(16)
Itโs obvious, that $`Q(f)`$ is periodic with the period $`2\pi `$. Thus we get
$`<Z(C)>={\displaystyle \underset{k=1,0,1;\delta k=3l;l;j}{}}{\displaystyle _\pi ^\pi }Dhexp(Q(\delta h+(2\pi /3)(A[C]+k)))`$
$`exp(i(h,3j))`$ (17)
We can solve the constraint $`\delta k=3l`$ : $`k=3A[l]+\delta z`$. Due to the periodicity of $`Q`$, $`l`$ is eliminated. Then we redefine $`h(h+2\pi /3\delta z)mod2\pi `$, and finally get:
$$<Z(C)>=\underset{\delta j=0}{}_\pi ^\pi Dhexp(Q(\delta h+(2\pi /3)A[C])+i(h,3^{}j))$$
(18)
Thus we have obtained that the theory dual to the original $`Z_3`$ projected gluodynamics is just the nonlocal $`U(1)`$ gauge theory with additional summation over the worldlines of the center monopoles, which carry the charge $`3`$ with respect to the mentioned $`U(1)`$ gauge field.
We can rewrite the summation over the worldlines of the monopoles as the integral over the Higgs field of charge $`3`$:
$$\underset{j}{}exp(i(H,3j))=D_{\mathrm{\Phi }C}\mathrm{\Phi }exp(\underset{xy}{}\mathrm{\Phi }_xe^{3iH_{xy}}\mathrm{\Phi }_y^+V(|\mathrm{\Phi }|)),$$
(19)
where the potential $`V`$ is infinitely deep, and the vacuum average of $`|\mathrm{\Phi }|`$ is infinitely large. So $`V(r)=a(r^2b)^2`$, where $`a,b\mathrm{}`$. Here we denoted $`{}_{}{}^{}h=H`$.
Finally we have
$`<Z(C)>={\displaystyle _\pi ^\pi }DH{\displaystyle }D_{\mathrm{\Phi }C}\mathrm{\Phi }exp(Q(dH+(2\pi /3)^{}A[C])`$
$`{\displaystyle \underset{xy}{}}\mathrm{\Phi }_xe^{3iH_{xy}}\mathrm{\Phi }_y^+V(|\mathrm{\Phi }|))`$ (20)
The last representation is the dual superconductor representation of the $`SU(3)`$ theory. Here the field $`\mathrm{\Phi }`$ is the field of our center monopole. When the monopole is condensed, the Abrikosov-Nielsen Olesen strings appear. That strings carry the magnetic flow $`2\pi /3`$, thus connecting the quarks, which play the role of monopoles here. Also the usial monopoles, existing due to the periodicity of the action, create the Abrikosov-Nielsen-Olesen strings. But that strings should carry the magnetic flow $`2\pi `$, which is clear from the above expression. Thus that monopoles indeed create $`3`$ strings and do not influence the confinement mechanism.
## 4 The center monopole creation operator
We define the monopole creation operator in the way similar to that of in the $`U(1)`$ theory representing the Maximal Abelian projected $`SU(2)`$ \- gluodynamics. Following , we obtain the vacuum average of the monopole - antimonopole correlator in the dual theory:
$`<\mathrm{\Phi }(z1)\mathrm{\Phi }^{}(z2)>={\displaystyle _\pi ^\pi }DH{\displaystyle }D_{\mathrm{\Phi }C}\mathrm{\Phi }exp(Q(dH)`$
$`{\displaystyle \underset{xy}{}}\mathrm{\Phi }_xe^{3iH_{xy}}\mathrm{\Phi }_y^+V(|\mathrm{\Phi }|))\mathrm{\Phi }(z1)\mathrm{\Phi }^{}(z2)exp(i(D(z1)D(z2),H)),`$ (21)
where $`\delta D(z)=3\delta _z`$.
After coming back to the representation throw the worldlines of the monopoles we get
$`<\mathrm{\Phi }(z1)\mathrm{\Phi }^{}(z2)>={\displaystyle \underset{\delta j=\delta _{z1}\delta _{z2}}{}}{\displaystyle _\pi ^\pi }DHexp(Q(dH)+`$
$`i(H,3j(D(z1)D(z2)))`$ (22)
Repeating the steps back to the original representation we obtain
$`<\mathrm{\Phi }(z1)\mathrm{\Phi }^{}(z2)>=lim_\alpha \mathrm{}{\displaystyle D_{VSU(3)/Z_3}V\underset{N=1,0,1}{}}`$
$`exp({\displaystyle \underset{plaq}{}}\beta (11/3ReTre^{2\pi (dN)/3}V_{plaq})`$
$`\alpha O(V))\mathrm{\Delta }_{FP}(V,\alpha )`$
$`Q(^{}d([dN]mod3)(3j_{z1,z2}D(z1)+D(z2)))`$ (23)
Here $`j_{z1,z2}`$ is the line connecting points $`z1`$ and $`z2`$ of the dual lattice. And
$$Q(x)=\underset{\delta j=0}{}\mathrm{\Pi }_{{}_{}{}^{}links}sin(\pi (xj))/(xj),$$
(24)
where the summation is over the closed integer 1-forms on the dual lattice.
In other words
$`<\mathrm{\Phi }(z1)\mathrm{\Phi }^{}(z2)>={\displaystyle DUexp(\underset{plaq}{}\beta (11/3ReTrU_{plaq}))}`$
$`Q(3^{}j_{Z_3}(3j_{z1,z2}D(z1)+D(z2))),`$ (25)
where $`j_{Z_3}`$ is the center monopole trajectory extracted from the field configuration $`U`$.
## 5 Conclusions
In this work we made an attempt to extract the kind of superconductor theory from the $`SU(3)`$ gluodynamics in the Maximal Center Gauge. We obtained the theory, which contains $`U(1)`$ gauge field and the scalar field charged with respect to that $`U(1)`$ field. The action is essentially nonlocal. The usial quarks play the role of monopoles in this theory. The scalar field is condensed in the confinement phase and is not condensed in the deconfinement phase. The worldlines of the particle correspondent to that scalar field are just the worldlines of the center monopoles.
Of course, the analogous representation one can obtain for the gluodynamics in the Maximal Abelian Gauge. But in that case two scalar fields and two $`U(1)`$ fields appear. Thus the dual superconductor mechanism becomes too complex and unnatural.
## Acknowledgments
The author is grateful to B.L.G. Bakker, J. Greensite, and A. Veselov for useful discussions.
This work was supported by the JSPS Program on JapanโFSU scientists collaboration, by the grants INTAS-RFBR-95-0681 and RFBR-97-02-17491. |
warning/0001/cond-mat0001096.html | ar5iv | text | # Irreversibility temperature from magnetic relaxation
## 1 Introduction
Magnetic relaxation in high-$`T_c`$ superconductors is experimentally known to be non-logarithmic over a large time interval . Though the non-logarithmic behaviour is related to a diverging pinning potential within vortex glass theory and collective creep theory , the details of the experimental behaviour is not consistent with the theoretical predictions. Specifically, the normalized relaxation rate is found to be temperature independent with a value that cannot be accounted by the theory . Moreover, no clear picture exists for the change in relaxation behaviour across the irreversibility temperature $`T_{irr}`$ and into a true superconducting state with finite persistent current.
## 2 The model
In order to understand generic processes involved during the relaxation in hard type-II superconductors, we have simulated relaxation of the thermoremanent magnetisation in a 2D Josephson junction array at finite temperatures. As have been shown previously , a 2D Josephson junction array (JJA) with finite screening current is a realistic model for magnetic behaviour of hard superconductors. We consider a 2D JJA of size $`N\times N`$ in a transverse magnetic field. The equation of motion for the gauge invariant phase difference across a junction $`\varphi `$ is given by
$`{\displaystyle \frac{d\varphi }{d\tau }}`$ $`=`$ $`๐ฌ^TI_m\mathrm{sin}\varphi +X(\tau ),`$
$`๐ฌ\varphi `$ $`=`$ $`2\pi f{\displaystyle \frac{1}{\lambda _J^2}}I_m.`$ (1)
Here, the current in a cell $`I_m`$ is scaled by the critical current $`I_c`$ of the junction. The $`\tau =\frac{2\pi RI_c}{\mathrm{\Phi }_0}t`$ represents dimensionless time, and $`\lambda _J^2=\frac{\mathrm{\Phi }_0}{2\pi L_0I_c}`$ is the dimensionless penetration depth analogous to Londonโs penetration depth of a bulk superconductor ($`\mathrm{\Phi }_0`$ is a quantum of flux, $`L_0`$ is the self-inductance of a unit cell, and $`R`$ is the normal state resistance of the junction). The applied magnetic flux is represented by $`f=\mathrm{\Phi }_{ext}/\mathrm{\Phi }_0`$. The matrix $`๐ฌ`$ is directed loop-sum operator and is equivalent to lattice curl operation. The temperature bath is attached through the noise term $`X(\tau )`$ with $`X_๐ซ(t)=0`$ and $`X_๐ซ(\tau )X_๐ซ^{}(\tau ^{})=2T\delta (\tau \tau ^{})\delta _{๐ซ,๐ซ^{}}`$. The temperature is in units of $`I_c\mathrm{\Phi }_0/2\pi k_B`$. We employed free-end boundary conditions for all the variables. Further details regarding the simulation and setting up of the equation can be found in ref. .
## 3 Results and discussions
We present here the results for an array of size $`N=16`$. The Fig.LABEL:fig1 shows the relaxation over 6 decades in $`\tau `$ for some selected temperatures. The relaxation behaviour changes markedly across two temperatures : $`T_{cr}0.24`$ and $`T_{sc}0.04`$. At $`T_{cr}`$, the relaxation curve develops a kink in an intermediate time interval which at lower temperature develops into a plateau on which the dynamics is almost frozen. The width of this plateau thus sets a new time scale $`\tau _\beta `$ for the relaxation of the magnetisation. Experimentally, the $`T_{irr}`$ is obtained as the temperature at which the field cooled (FC) and zero field cooled (ZFC) susceptibility $`\chi (T)`$ differs. Inset of Fig.LABEL:fig1 shows $`\chi (T)`$ for the model. The temperature $`T_{cr}`$ at which the kink appears in the relaxation curve is also the temperature at which the $`\chi _{FC}\chi _{ZFC}>0`$. This allow us to conclude that $`T_{cr}`$ is the irreversibility temperature appearing in the relaxation. This is also consistent with the previous simulation study in which magnetic irreversibility is found to set below this temperature . For $`T_{sc}<T<T_{cr}`$, the magnetisation $`M(\tau \mathrm{})=0`$. At $`T_{sc}`$, the relaxation is frozen on the plateau as $`\tau \mathrm{}`$, thus establishing a finite persistent current density (hence, remanent magnetisation). We consider this as the transition into the true superconducting state. By explicitly observing the flux distribution, we find that the plateau is related to the crossover from a supercritical state with $`J>J_c`$ to a subcritical state with $`J<J_c`$. We attribute this crossover as arising due to self-organization during the relaxation process. The $`\tau _\beta `$ increases rapidly with decreasing $`T`$.
For $`T_{sc}<T`$, the long time relaxation fits to $`\mathrm{exp}[(\tau /\tau _\alpha )^\alpha ]`$. The characteristic time scale $`\tau _\alpha `$ increases rapidly below $`T_{cr}`$. One also observes $`\mathrm{log}(\tau /\tau _0)`$ behaviour for 1-2 decades far from the plateau. This is the regime in which thermal activation is a dominant process and can be used to extract the characteristic time scale $`\tau _0`$ for relaxation at long time. Fig.LABEL:fig1 inset (b) shows the temperature dependence of $`\tau _0^1`$ (magnetic diffusivity) and is plotted as a function of $`TT_{sc}`$. Though $`\tau _0`$ follows the Arrhenius law, a power law at low temperatures as indicated by the straight line in the plot is surprising. The $`T_{sc}`$ obtained from fitting the time scale for long time relaxation matches well with the $`T`$ at which $`M(\tau \mathrm{})=M_0>0`$. The $`T_{cr}`$ appears as the temperature at which $`\tau _0^1`$ deviates from the power law as marked in Fig. 1 inset(b).
In conclusion, we have shown that a new time scale govern the flux dynamics below $`T_{irr}`$ which arises due to self-organization of the magnetic flux during the relaxation. This conclusively proves that $`T_{irr}`$ is associated with the crossover in dynamical behaviour for magnetic flux in type-II superconductors.
Figure Caption The normalized magnetisation $`M(\tau )/M(0)`$ for some selected temperatures (marked along the curves) for $`f=5`$ and $`N=16`$. Inset: (a)shows the $`\chi (T)`$ for FC and ZFC conditions ($`T_{irr}`$ is marked on the curve). (b) The $`\tau _0^1`$ and $`\tau _\beta `$ as a function of $`TT_{sc}`$ on log-log plot. The vertical axis label indicates the exponent only. |
warning/0001/astro-ph0001090.html | ar5iv | text | # Emission Lines in the Spectrum of the 3He Star 3 Cen A *footnote **footnote *Based on observations obtained with the Canada-France-Hawaii telescope, operated by the National Research Council of Canada, the Centre National de Scientifique of France, and the University of Hawaii.
## 1 Introduction
3 Centauri A (HD 120709) is a B5 III-IVp He-weak star of the Ga-P subclass. Its spectrum is characterized by a highly non-solar chemical composition, with large overabundances (1-2 dex) of P, Sc, Mn, and Cu, extreme overabundances (3-4 dex) of Ga, Kr, and Hg, and underabundances (0.5-1 dex) of He, B, C, N, Mg, and Al (Castelli, Parthasarathy & Hack 1997). In addition, most of the helium in its photosphere is in the form of <sup>3</sup>He, with a $`{}_{}{}^{3}\mathrm{He}/^4\mathrm{He}`$ ratio of $`2.5`$ as compared to the solar ratio of $`1.410^4`$ (Sargent & Jugaku 1961, Hartoog & Cowley 1979). These abundance peculiarities are thought to have their ultimate explanation in terms of elemental diffusion operating in the photosphere. The mechanism of light-induced-drift seems capable of explaining the very large <sup>3</sup>He to <sup>4</sup>He ratio (Michaud & Proffitt 1992, LeBlanc & Michaud 1993). No longitudinal magnetic field has been detected in 3 Cen A to an upper limit of $`100`$G (Borra, Landstreet & Thompson 1983). Chemical abundances derived from optical spectra have been recently reported by Pintado, Adelman, & Gulliver (1998) and UV spectral synthesis has been performed by Castelli et al. (1997).
In this Letter, we add to the peculiarities of 3 Cen A by reporting the presence of several emission lines of Mn II (multiplet 13) and P II, and one of Hg II ($`\lambda \mathrm{\hspace{0.17em}6149.2}`$), in its red spectrum. We also report emission in the same Mn II multiplet in the spectrum of the prototype hot, mildMild in this context refers to the relatively moderate manganese photospheric abundance enhancement. HgMn star 46 Aquila (HD 186122).
## 2 Observations
Spectra are available for 3 Cen A in two wavelength regions, $`\lambda \lambda \mathrm{\hspace{0.17em}6105}6190`$ and $`\lambda \lambda \mathrm{\hspace{0.17em}6350}6425`$, which were taken with the coudรฉ f/4 (Gecko) spectrograph at the Canada-France-Hawaii telescope in May 1999. These observations were obtained as part of a program to study neon abundances in late and mid B stars in an attempt to detect very weak stellar winds, as suggested by Landstreet, Dolez & Vauclair (1998). The CCD detector EEV2 was used, giving 200 x 4500 pixels, each with a size of 13 $`\mu m`$. The spectra were bias subtracted, flat-fielded, and continuum normalized using the IRAF reduction package (Tody 1993). The spectra were wavelength calibrated from a ThAr comparison arc spectrum taken after each grating setting; the resulting dispersion relation limits the relative wavelengths to an accuracy of $`\pm 0.003`$ร
. The Julian dates of each observation, as well as other details, are summarized in Table 1. As the Gecko spectrograph offers a resolution of $`1.210^5`$, it is a superb instrument with which to study ultra-sharp-lined stars such as 3 Cen A.
In the following discussion, the spectra of 46 Aql (B9 III HgMn, HD 186122) and $`\kappa `$ Cnc (B8 IIIp HgMn, HD 78316), obtained during the same observing run, will be used for comparison with the spectra of 3 Cen A; their observational details are also summarized in Table 1.
## 3 Results
Figure 5 shows the spectrum of 3 Cen A in both spectral regions. Prominent absorption lines dues to O I, Si II, Fe II, Ne I, and P II are indicated. The final wavelength correction for the radial velocity of 3 Cen A was made using the laboratory wavelengths of these strong absorption lines. After applying a mean shift, the wavelengths of these reference lines were reproduced to within $`\pm 0.01`$ร
. As no variability was detected between observations on successive nights, all eight 300 second exposures for each wavelength region of 3 Cen A listed in Table 1 were combined.
### 3.1 Mn II Emission
Figure 5 shows the spectral region of $`\lambda \lambda \mathrm{\hspace{0.17em}6120}6135`$ which contains Mn II multiplet 13, $`3\mathrm{d}^5(^6\mathrm{S})4\mathrm{d}^5\mathrm{D}3\mathrm{d}^5(^6\mathrm{S})4\mathrm{f}^5\mathrm{F}^\mathrm{o}`$. These are transitions between highly excited states, with the lower levels lying $`10.2`$eV above ground. In the Figure, the multiplet structure is shown on top, with the individual line lengths proportional to the $`gf`$ values of the transitions. The required atomic data for this multiplet were obtained from the Kurucz & Bell (1995) database. As can be seen, the entire multiplet structure is accounted for and the identification is not in doubt. For reference, the predicted LTE spectrum for 3 Cen A using the abundances of Pintado et al. (1998), with a manganese enhancement of 1.5 dex, and a Kurucz model atmosphere of parameters $`T_{\mathrm{eff}}=\mathrm{17\hspace{0.17em}500}`$K and $`\mathrm{log}(g)=3.8`$, is shown in the figure. The spectrum of the HgMn star $`\kappa `$ Cnc in the same wavelength range is also shown, which presents this Mn II multiplet in strong absorption. Note that $`\kappa `$ Cnc is significantly cooler than 3 Cen A with atmospheric parameters of $`T_{\mathrm{eff}}=\mathrm{13\hspace{0.17em}470}`$K and $`\mathrm{log}(g)=3.76`$ as derived by Smith (1992).
3 Cen A is not the only star in our sample to show emission in multiplet 13 of Mn II. Weak emission in this multiplet is also definitely detected in the spectrum of 46 Aql (B9 III) which is the prototype of the hot, mild HgMn stars (Cowley 1980, Smith 1993). Its spectrum is also shown in Figure 5. Smith (1992) derives atmospheric parameters for 46 Aql of $`T_{\mathrm{eff}}=\mathrm{13\hspace{0.17em}000}`$K and $`\mathrm{log}(g)=3.65`$, making it significantly cooler than 3 Cen A and comparable to $`\kappa `$ Cnc.
For 3 Cen A, the observed wavelengths of the Mn II emission lines have a mean shift relative to the photospheric absorption lines of $`0.015`$ร
, at the limit of the wavelength calibration. Hence, there is no evidence that the emission is Doppler shifted relative to the photospheric absorption lines. The lines in the 3 Cen A spectra are resolved and symmetric, and the FWHM of the emission features are not noticeably different from the absorption lines.
The relative intensities of the emission lines in the Mn II multiplet in both 3 Cen A and 46 Aql are not exactly proportional to their $`gf`$ values, suggesting optically thick, not optically thin, formation. Note that the laboratory spectrum of this multiplet presented by Johansson et al. (1995) gives relative intensities in excellent agreement with those predicted by the Kurucz & Bell (1995) $`gf`$ values, suggesting that the $`gf`$ values have high relative accuracy, as is frequently the case for Kuruczโs calculations within a single multiplet.
There are no traces of absorption components to the emission lines, i.e. none of the emission lines sit on an absorption trough. Hence either the emission has completely filled in the photospheric absorption components expected to be present in LTE (see later discussion) due to the high Mn abundance of 3 Cen A derived from other wavelengths (see Figure 5), or the emission is the photospheric spectrum.
### 3.2 Hg II and P II Emission
Figure 5 shows the spectral region region $`\lambda \mathrm{\hspace{0.17em}6144}6160`$ of 3 Cen A which contains two prominent Fe II transitions of multiplet 74, $`\lambda \mathrm{\hspace{0.17em}6147.7}`$ and $`\lambda \mathrm{\hspace{0.17em}6149.2}`$. The emission feature close to the Fe II line at $`\lambda \mathrm{\hspace{0.17em}6149.2}`$ is definitely identified with Hg II $`\lambda \mathrm{\hspace{0.17em}6149.5}`$. Identification of this feature, and its impact on using the relative strengths of these two Fe II lines to detect magnetic fields, is discussed by Takada-Hida & Jugaku (1992) and Hubrig, Castelli & Wahlgren (1999). In Figure 5, the spectrum of $`\kappa `$ Cnc is shown for reference, and the blend of Hg II in the wing of the Fe II line is clearly seen.
We note that the apparent increase in strength of Fe II $`\lambda \mathrm{\hspace{0.17em}6147.7}`$ over that of $`\lambda \mathrm{\hspace{0.17em}6149.2}`$ is probably due to a blend in the red wing of $`\lambda \mathrm{\hspace{0.17em}6147.7}`$ with another (high-excitation) transition of Fe II and thus there is no evidence for a magnetic field in 3 Cen A using the technique of Mathys & Lanz (1990). However, the ultra-sharp lines of 3 Cen A can be used to constrain any non-thermal broadening due to rotation, microturbulence, or a magnetic field. The largest possible value for each (assuming the other two are negligible) consistent with the line widths have been determined by modeling the Fe II and Ne I line profiles. The resulting limits are $`v\mathrm{sin}i2\mathrm{km}\mathrm{s}^1`$, $`\zeta _t1.5\mathrm{km}\mathrm{s}^1`$, and $`B_s1500`$G (where $`B_s`$ is the mean field modulus averaged over the visible hemisphere).
Many of the remaining weak emission features in the observed spectrum of 3 Cen A can be identified with high-excitation P II transitions, having as their lower levels symmetries of the $`3\mathrm{s}^2\mathrm{\hspace{0.17em}3}\mathrm{p}(^2\mathrm{P}^\mathrm{o})\mathrm{\hspace{0.17em}4}\mathrm{f}`$ configuration near 16.2 eV, and their upper levels, symmetries of the $`3\mathrm{s}^2\mathrm{\hspace{0.17em}3}\mathrm{p}(^2\mathrm{P}^\mathrm{o})\mathrm{\hspace{0.17em}6}\mathrm{g}`$ configuration. Several P II transitions are also identified in the second spectral window of $`\lambda \lambda \mathrm{\hspace{0.17em}6350}6425`$. These identifications have been made with the atomic line list of van Hoofhttp://www.pa.uky.edu/$``$peter/atomic/ which is based on the energy level data available in the NIST Atomic Spectroscopic Database.
The measured wavelengths of all of the emission features in both spectral windows are given in Table 2, along with their identifications and $`\mathrm{log}(gf)`$ values, where available. A few weak emission features still remain unidentified. In this latter group, we have noted a few wavelength co-incidences with permitted transitions of Si I from the $`3\mathrm{s}^23\mathrm{p}\mathrm{\hspace{0.17em}4}\mathrm{p}`$ configuration. However, as these transitions are all to highly excited states, $`n>10`$, and as 3 Cen A shows absorption lines of only Si II and Si III in its optical and UV spectrum, consistent with its $`T_{\mathrm{eff}}`$, we regard these identifications as extremely unlikely. Given 3 Cen Aโs P-Ga classification, emission lines due to Ga II may be a interesting possibility, but the energy level identifications of Ga II are incomplete and no line lists of Ga II transitions in the red ($`\lambda >6000`$ร
) are currently available.
## 4 Formation Mechanisms
The emission lines described in the previous section appear to be photospheric, not circumstellar, in origin. The lines are resolved and symmetric, and their FWHM are similar to the absorption lines present in each spectrum. The radial velocity derived from the absorption and emission lines agrees to within the uncertainty in the wavelength calibration.
Weak emission lines of photospheric origin are not unknown. Vega exhibits weak emission lines of Fe II in the wings of Lyman-$`\alpha `$ (van Noort et al. 1998) and weak emission lines from iron and rare earth elements appear in the wings of the Ca II H & K lines in the solar spectrum (Cram, Rutten & Lites 1980). In both instances, it is strong wing opacity in Lyman-$`\alpha `$ or Ca II which shifts the depth of formation of the weaker lines to small optical depths where interlocked non-LTE effects produce source functions which rise with height (decreasing optical depth). Interlocked non-LTE effects also produce the well known solar Mg I 12 $`\mu m`$ lines (Carlsson, Rutten & Shchukina 1992). None of these mechanisms require a temperature rise in a chromosphere to produce the emission.
Mn II multiplet 13 is clearly seen in emission in the spectrum of $`\eta `$ Carinรฆ (Johansson et al. 1995); however, this spectrum is of circumstellar origin. Johansson et al. suggest that this Mn II emission is formed by fluorescent excitation via a wavelength coincidence of Mn II multiplet 15 and Si II multiplet UV 5. However, we are doubtful that a fluorescent origin for the Mn II lines is possible in the present case. For fluorescent excitation to occur, the source function of the pumping transition must be different from the Planck function in the line forming layers, which, in the current case, would be deep in the photosphere where the collisional rates are large (although perhaps chemical stratification could play a role - see later discussion).
A non-LTE origin for these emission lines is suggested as only certain transitions for each ion are in emission. Both absorption and emission lines of P II are present in our spectra, and Mn II and P II absorption lines appear in optical and UV spectra forming the basis for the the published abundance studies. Interlocked non-LTE effects are capable of producing line source functions that rise with height (or decreasing optical depth) without the requirement of a chromospheric temperature rise. As $`h\nu <kT`$ for transitions at the wavelengths considered, photon losses in the lines themselves are insufficient to significantly influence the departure coefficients of their upper and lower levels. The departure coefficients can be controlled by losses in other lines, or bound-free rates, which can produce rising source functions in the same manner as for the solar Mg I 12 $`\mu `$m lines (Carlsson et al. 1992). As this mechanism is transition specific, it does not follow that the entire manganese, phosphorus, or mercury spectrum need be in emission.
There is an additional interesting point to consider: the large overabundances of manganese, phosphorus, and mercury are thought to result from diffusive processes which will act to concentrate these elements in thin layers at small optical depths (Michaud, Reeves & Charland 1974, Alecian & Michaud 1981). If this is the case, then for example, the entire manganese or mercury spectrum would be produced at small optical depths where the collisional rates are very small and non-LTE effects can dominate. The chemical stratification would then play a role similar to the overlapping wing opacity in the previous cases cited. While this mechanism is plausible, detailed non-LTE calculations are required to confirm this hypothesis, and in particular, to determine if chemical stratification of the abundance distribution plays a critical role. Such calculations are planned for the near future. Additional observations are required to capture more of the Mn II spectrum. If our hypothesis is indeed correct, these emission features may provide rather direct evidence of chemical stratification in the photosphere of 3 Cen A providing addition constraints on the diffusive model for chemically peculiar stars.
## 5 Conclusions
We have discovered weak emission due to Mn II, Hg II and P II in the spectrum of 3 Cen A in the wavelength regions $`\lambda \lambda \mathrm{\hspace{0.17em}6105}6190`$ and $`\lambda \lambda \mathrm{\hspace{0.17em}6350}6425`$. We have also discovered Mn II emission in the spectrum of the hot, mild HgMn star 46 Aql. We suggest that these emission lines are of photospheric origin and are produced by interlocked non-LTE effects. As the elements observed to produce the emission are very probably made overabundant by chemical diffusion, it is speculated that the concentration of absorbers in thin layers at small optical depth may play an important role in the formation of the emission lines. Additional observations of more He-weak and HgMn stars, as well as wider wavelength coverage for 3 Cen A, are required. A more complete survey of these spectral regions, for both normal and chemically peculiar stars, would be particularly valuable.
TAAS wishes to thank J. M. Marlborough and J. D. Landstreet for support through their NSERC grants. We particularly thank Elizabeth Griffin for suggesting the P II and possible Si I identifications. We acknowledge the assistance of Simon Strasser with the IRAF reductions. |
warning/0001/gr-qc0001023.html | ar5iv | text | # Frequency-domain P-approximant filters for time-truncated inspiral gravitational wave signals from compact binaries
## I Introduction and Summary
The discovery of the first binary pulsar in 1974 has had a very important impact on gravitational wave research. First, it proved the reality of gravitational radiation by measuring the orbital period decay entailed by the propagation at the velocity of light of the gravitational interaction between the two neutron stars making up the system . Second, it provided the first experimental evidence that General Relativity correctly describes gravity in the strong-field regime . Third, it led to a shift in perception regarding the most promising sources for future gravitational wave (GW) detectors, away from the then assumed, violent โ but less predictable โ gravitational collapse associated with supernovae, to the more predictable, final inspiralling phase of compact binaries of neutron stars and black holes driven by gravitational radiation-reaction. This also led to the thrust in the laser interferometric gravitational wave detectors which are inherently broad-band rather than in the narrow-band bar detectors.
### A Data analysis algorithms for inspiral wave searches
Consider a compact binary system like the binary pulsar after it has been inspiralling inwards for three hundred million years due to gravitational radiation-reaction. The inspiral waveform enters the detector bandwidth during the last few minutes of evolution of the binary. Our ability, in principle, to compute the waveform very accurately, allows us to track the gravitational wave phase and enhance the signal-to-noise ratio by integrating the signal for the interval during which it lasts in the detector band. This, in turn, requires a template with which the detector output may be filtered. Though template waveforms should, optimally, be exact copies of the expected signal, in practice, they are constructed by some approximation scheme and will differ from the actual signal in the detector output. Consequently, the overlap of template and signal waveforms will be less than if they had exactly matched, leading to a loss of potential events. Data analysis issues like these for inspiralling compact binaries of neutron stars and black holes have been formulated and addressed for the last twelve years , even though interferometric gravitational wave detectors like the GEO600 or LIGO and VIRGO are a year or three in the future. Much of the work in this area has addressed practical issues of direct relevance to data analysis strategies. These include: construction of templates for detection, the number of templates, their placement, spacing, the required computing power and the storage or memory requirement , the order of post-Newtonian (PN) approximation adequate for detection , parameter estimation by covariance matrix and Monte Carlo simulations, determination of cosmological parameters , tests of general relativity , one step versus hierarchical searches , effects of precession and of eccentricity . For the time-domain waveform, all of these works use the restricted post-Newtonian approximation to quasi-circular inspiral. This keeps the crucial phase information to the best order of approximation then available , but restricts the amplitude to be Newtonian and the harmonic to the second harmonic of the orbital frequency. Such an approximation should be adequate for the on-line search of gravitational wave signals . Evidently, it is assumed that the offline analysis of the data will use the best available (unrestricted post-Newtonian) representation of the inspiral signals.
### B Modelling inspiral waveforms
The post-Newtonian approximation is basically a Taylor expansion (in powers of $`v/c`$) and all the above treatments use as building blocks the straightforward Taylor expansions in $`v/c`$ of some intermediate quantities (orbital energy and gravitational-wave flux). We shall refer to the templates based on such straightforward PN expansions as โTaylor approximantsโ (or simply T-approximants). The very slow convergence and oscillatory behaviour of the PN expansion, and therefore of the sequence of Taylor approximants, made imperative a search for better approximants for phasing. This prompted us (later referred to as DIS) to propose new approximants, with much improved convergence properties, for application to gravitational-wave data analysis problems.
In DIS , we showed how to construct a new type of time-domain approximant, called โP-approximantsโ, which not only converged faster and more monotonically, but were also more effectual (larger overlaps for detection) and faithful (smaller biases for parameter estimation) than the standard T-approximants. Our construction was two-pronged: on the one hand, it introduced new basic energy and flux functions, and on the other hand, it made a systematic use of Padรฉ techniques (a well-known convergence-acceleration technique) to construct successive approximants of our new basic energy and flux functions. These new functions form a pivotal aspect of our construction and successfully handle issues related to appearance of non-rational functions in the energy function and logarithmic terms in the flux function that for long proved to be hurdles to the application of well-known Padรฉ techniques to this problem. For initial LIGO, the 2.5PN P- approximants are likely to provide overlaps in excess of 96.5% with exact waveforms<sup>*</sup><sup>*</sup>* This statement was proven in DIS by quantifying the convergence of the sequence of P-approximants towards some โfiducial exactโ waveform. In the test-particle case this waveform used the known Schwarzschildโs energy function $`E(v)`$ and Poissonโs numerically computed GW flux . In the comparable mass case, it was constructed by modelling the $`\eta `$\- dependent higher post-Newtonian corrections to the best known analytical results. so that more than 90% of the potential events can be detected. In contrast, the corresponding 2.5PN Taylor approximants can only detect about 50% of the potential events for massive systems (at the price of large biases $`15\%`$). Later studies have confirmed the performance of these P-approximants and assessed their need in related contexts of space based interferometers like LISA.
### C Fourier representation of inspiral signals and validity of the stationary phase approximation (SPA)
Independent of the choice between T- and P-approximants, another desirable approximation in data analysis for inspiralling compact binaries is the stationary phase approximation (SPA), which is a simple, explicit analytic approximation to the Fourier transform of the time-domain chirp (see, e.g., ). In fact, most work on inspiral waveforms (except DIS) has used only SPA approximants to the frequency-domain chirps. In the course of our P-approximant work we noticed a progressive worsening of the overlap between the SPA and the โexactโ Fourier transform โ numerically computed by a fast Fourier transform (FFT) of the time-domain signal โ (see Table II of DIS) and commented on these โinaccuracies of the SPAโ. In the above by SPA one means not only the problem of the formal accuracy of the stationary phase estimate to the Fourier transform of an analytically extended, mathematical signal but also some issues linked to the physics, and observability, of the real signal. In particular, in DIS, we were considering templates which are shut off, in the time-domain, at the last stable orbit (LSO). The present paper will also consider such time-truncated inspiral signals. We shall discuss this point in more detail below, but the idea is that the post-inspiral signal (plunge + merger) will have a frequency content very different from the inspiral one (probably pushed to much higher frequencies). It should, therefore make sense to try to construct filters that represent as best as possible an inspiral signal which lasts only up to some maximum time (time-windowing). For such signals, DIS noted a worsening of the usual (frequency-windowed) SPA approximation, both as the total mass of the system increases and as the post-Newtonian approximation order is increased, and mentioned that this worsened performance was due to the fact that โsuch systems emit many less wave cycles in the effective detector bandwidthโ centered (for initial LIGO) near $`f_{\mathrm{det}}=167`$ Hz. In this paper $`f_{\mathrm{det}}`$ denotes the frequency at which the noise power spectrum per logarithmic bin of the detector is the least (or equivalently the frequency at which the detector is most sensitive to a broad-band burst). To avoid irrelevant, uncontrolled sources of inaccuracy, DIS used the FFT of the time-windowed chirp rather than its SPA to generate the frequency-domain waveform and make comparisons between the T- and P-approximants.
The use of FFT rather than SPA in DIS makes the P-approximant computationally expensive. As will be discussed in detail in Sec. VI, the use of SPA or similar frequency-domain representations is far less expensive. The obvious need to incorporate this desirable feature makes urgent and mandatory a critical investigation of the possibility of marrying together the excellent performance of the P-approximants to the relative inexpensiveness of the SPA without a serious loss in event rate.
Recently, some issues related to the accuracy of the SPA have been investigated. For general chirps, Chassande-Mottin and Flandrin have studied whether the usual conditions assumed for the validity of the SPA are necessary and sufficient and attempted a quantitative control of the approximation. Droz et al have examined other issues related to the accuracy of the SPA of particular relevance to gravitational wave data analysis. Unlike DIS, by SPA, Droz et al imply only the stationary phase estimate of the Fourier transform and discuss separately the issue of windowing โ the fact that the signal in the time-domain lasts only from $`t_{\mathrm{min}}`$ to $`t_{\mathrm{max}}`$ or a time-window. To improve the SPA estimate of a Newtonian chirp, they compute the next order contribution We shall give below the general result for any chirp. (to the Fourier integral) by the method of steepest descent, show that it is of order $`v_{}^{10}`$ relative to the leading order SPA estimate and conclude that it is small enough to be justifiably neglected. \[Here $`v_{}`$ is an invariantly defined โvelocityโ Note that, following DIS, we shall use $`v(\pi mF)^{1/3}`$, instead of $`v_{}=\eta ^{1/5}v`$, in all our analysis. We also use units such that $`G=c=1`$. related to the instantaneous gravitational wave frequency $`F`$ and chirp mass $``$ by $`v_{}=(\pi F)^{1/3}`$. The chirp mass is related to the total mass $`m=m_1+m_2`$ and dimensionless mass ratio $`\eta =\frac{m_1m_2}{(m_1+m_2)^2}`$ by $`=\eta ^{3/5}m`$.\] They point out the importance of windowing, estimate the amplitude and phase modulations induced in the frequency-domain by the time-window and conclude that in all cases these modulations have negligible effect on overlaps. However, their analytic expression for the effects of windowing is only valid for values of frequencies well away from the boundaries of the natural frequency-window induced by the time-window, denoted by $`F_{\mathrm{min}}=F(t_{\mathrm{min}})`$ and $`F_{\mathrm{max}}=F(t_{\mathrm{max}})`$ โ the gravitational wave frequencies at times $`t_{\mathrm{min}}`$ and $`t_{\mathrm{max}}`$, respectively. In this paper we provide a formalism allowing one to compute analytic approximations to the Fourier-transform of a time-windowed signal in the most crucial edge-frequency-domains $`fF_{\mathrm{min}}`$ and $`fF_{\mathrm{max}}`$ (including $`f<F_{\mathrm{min}}`$ and $`f>F_{\mathrm{max}}`$). As first noticed in DIS and discussed in detail in this present work, the effect of window oscillations on overlaps (claimed to be negligible in ) starts to be noticeable when the total mass $`m13M_{}`$ and becomes very significant for $`m20M_{}`$. \[Here we consider equal mass systems $`\eta =1/4`$.\] Since the difference between the statements in DIS and Droz et al can be disconcerting and a serious source of confusion to the potential user community, we discuss this in further detail next.
In DIS, what was meant in Table II (the only place where it was used) by โstationary phase approximationโ was the product of the usual SPA by a simple Heaviside step function $`\theta (F_{\mathrm{max}}f)`$ i.e. $`\stackrel{~}{h}(f)`$ was truncated above a Fourier frequency $`f=F_{\mathrm{max}}`$ where $`F_{\mathrm{max}}`$ is the instantaneous gravitational wave frequency at which the time-domain signal is itself terminated, assumed to be (in DIS and here) the frequency at the last stable orbit $`F_{\mathrm{LSO}}`$. \[In the following, we shall, for brevity, refer to this frequency Windowed Usual Stationary Phase Approximation as the โuSPAwโ.\] We were motivated to do this from the stationary phase result itself. The SPA (to the Fourier transform of the chirp) says that the dominant contribution to a certain Fourier amplitude $`\stackrel{~}{h}(f)`$ comes from a neighbourhood of time (in the Fourier integral) when the instantaneous frequency $`F(t)`$ numerically reaches the corresponding Fourier frequency $`f`$. It is therefore to be expected that the signal essentially terminates at $`f=F_{\mathrm{LSO}}`$ i.e. that there is no significant power in the Fourier transform of the signal beyond $`F_{\mathrm{LSO}}`$. This is indeed true in the first approximation, as is evident from Fig. 6 below, which shows that the power in the exact Fourier transform of the time-windowed signal \[computed via a discrete Fourier transform (DFT)\] falls off much faster than the SPA for $`f>F_{\mathrm{LSO}}`$. Moreover, as is discussed in detail below, in the relativistic case the usual SPA breaks down at $`F_{\mathrm{LSO}}`$ and cannot be meaningfully extended for $`f>F_{\mathrm{LSO}}`$. Hence the values quoted in Table II of DIS were obtained by computing the overlap of the DFT of the truncated time-domain waves with the truncated SPA representation of the wave.
On the other hand, a critical examination of Droz et al reveals that their claim regarding the adequacy of the SPA in fact has only a restricted domain of validity. It is relevant to SPA considered as a mathematical algorithm to be applied to a generic smooth signal and low mass binaries ($`m13M_{}`$). As acknowledged by the authors, they do not address physical issues related to an eventual time-domain cut-off of the signal at $`F_{\mathrm{LSO}}`$. What they call โNewtonian signalsโ are unphysical, formally defined chirps whose instantaneous frequencies are extended to $`F_{\mathrm{max}}=F_{\mathrm{Nyquist}}F_{\mathrm{LSO}}`$, in fact, better described as โanalytically extended Newtonian signalsโ. It is the SPA of this formal, analytically extended signal which is shown to produce overlaps with exact FFTs better than $`0.99`$ even for massive binary systems of chirp mass $`=10M_{}`$, corresponding to a total mass of $`m23M_{}`$ for an equal mass system ($`\eta =1/4`$). These large overlaps, in our view, are not a proof of the validity of the SPA to compute, physically relevant, accurate frequency-domain inspiral templates, as they do not address the important issue of inspiral-signal termination at or near the $`F_{\mathrm{LSO}}`$ when $`F_{\mathrm{LSO}}f_{\mathrm{det}}`$, the frequency at which the broad-band noise of the detector is the least. It turns out that for binary systems of total mass $`m28M_{}`$ the power in the Fourier domain beyond $`f=F_{\mathrm{LSO}}`$ for a relativistic signal, is a significant fraction $`(>10\%)`$ of the total power. If the usual (frequency-windowed) SPA is used in constructing frequency-domain inspiral waves we are risking the loss of more than 30 % of the events from binaries with masses $`28M_{}`$. \[This will be illustrated in Fig. 1 below.\] This is in addition to the losses induced by the inaccuracy of the post-Newtonian waveforms and the discreteness of the bank of templates used in data analysis.
### D Massive black hole binaries and first detections in LIGO/VIRGO
Let us first establish our notation. We define the Fourier transform (FT) $`\stackrel{~}{h}(f)`$ of a time-domain signal $`h(t)`$ by
$$h(t)=_{\mathrm{}}^{\mathrm{}}๐fe^{2\pi ift}\stackrel{~}{h}(f);\stackrel{~}{h}(f)=_{\mathrm{}}^{\mathrm{}}๐te^{2\pi ift}h(t).$$
(1)
We write the (suitably transformed) output of the detector as
$$h_{\mathrm{out}}=h(t)+n(t),$$
(2)
where $`h(t)`$ is the signal and $`n(t)`$ the noise. The correlation function of the noise reads
$$\overline{n(t_1)n(t_2)}=C_n(t_1t_2)=_{\mathrm{}}^{\mathrm{}}๐fS_n(f)e^{2\pi if(t_1t_2)},$$
(3)
where $`S_n(f)=S_n(f)`$ is the two-sided noise power spectral density. In all the present work, we shall consider a noise curve of the type expected for initial interferometers. For initial LIGO we take,
$`S_n(f)`$ $`=`$ $`{\displaystyle \frac{S_0}{2}}\left[2+2\left({\displaystyle \frac{f}{f_0}}\right)^2+\left({\displaystyle \frac{f}{f_0}}\right)^4\right],ff_s,`$ (5)
$`=`$ $`\mathrm{},f<f_s.`$ (6)
with $`f_s=40`$ Hz, $`f_0=200`$ Hz and $`S_0=1.47\times 10^{46}`$ Hz<sup>-1</sup>. In the above we have included a factor of $`1/2`$ \[$`S_n^{\mathrm{one}\mathrm{sided}}2S_n^{\mathrm{two}\mathrm{sided}}`$\] because Eq. (5) gives the two-sided noise; the one-sided noise would be given by the same formula without the factor of $`1/2`$. The minimum of $`S_n(f)`$ is at $`f=f_0`$ and is equal to $`S_{\mathrm{min}}=2.5S_0`$. However, a physically more relevant quantity is the minimum of the dimensionless quantity $`h_n^2(f)fS_n(f)`$ (effective GW noise, see below). This is reached at the characteristic detection frequency $`f=f_{\mathrm{det}}=0.8347f_0`$, and is equal to $`(h_n^{\mathrm{min}})^2=2.2761f_0S_0`$. The above numerical value for $`f_0`$ and $`S_0`$ leads to $`f_{\mathrm{det}}=167`$ Hz and corresponds to $`h_n^{\mathrm{min}}=2.5868\times 10^{22}`$.
For VIRGO on the other hand, the corresponding noise curve is given by
$`S_n(f)`$ $`=`$ $`{\displaystyle \frac{S_0}{2}}\left[10^3\left({\displaystyle \frac{f_s}{f}}\right)^5+2\left({\displaystyle \frac{f_0}{f}}\right)+1+\left({\displaystyle \frac{f}{f_0}}\right)^2\right],ff_s,`$ (8)
$`=`$ $`\mathrm{},f<f_s,`$ (9)
In this case, $`f_s=20`$ Hz, $`f_0=500`$ Hz while $`S_0=3.24\times 10^{46}`$ Hz<sup>-1</sup>. The minimum of $`h_n(f)=\sqrt{fS_n(f)}`$ is reached at $`f=103`$ Hz and is equal to $`h_n^{\mathrm{min}}=4.2902\times 10^{22}`$. It should be noted that the VIRGO noise curve is used only in this Section, while discussing Fig. 1. In the rest of the paper and all the Figures and Tables, the scalar product is defined using the LIGO noise curve.
Anticipating on formulas to be discussed in II B, the square of the signal to noise ratio (SNR) is given by
$$\rho ^2=\left(\frac{S}{N}\right)^2=\frac{k,h^2}{k,k},$$
(10)
where the scalar product is defined by
$$k,h_{\mathrm{}}^+\mathrm{}๐f\frac{\stackrel{~}{k}^{}(f)\stackrel{~}{h}(f)}{S_n(f)}.$$
(11)
Here, $`h`$ denotes the exact signal, and $`k`$ the filter used in the data analysis. We assume in this paper that the signal $`h`$ is given by a time-truncated adiabatic inspiral signal. \[For simplicity, we consider in this subsection Newtonian waveforms, and we approximate the Fourier transform of a time-truncated Newtonian signal by the very accurate improved Newtonian stationary phase approximation (inSPA) to be constructed below.\] In computing $`\rho ^2`$ we average over all the angles (determining both the detector and the source orientations), and we place the source at a fiducial distance of 100 Mpc. \[Note that a coalescence rate of $`10^5`$ per galaxy and per year implies that in two years one event should happen within 100 Mpc.\]
In most of the literature one uses as Fourier-domain filter $`\stackrel{~}{k}(f)`$ the frequency-windowed usual stationary phase approximation (uSPAw) to estimate the SNR for an inspiral signal. We illustrate in Fig. 1 the loss in signal strength extracted by using as filter the uSPAw in LIGO and VIRGO \[cf. Eqs. (5) and (8)\], instead of using the optimal filter $`k=h`$ (leading to the optimal SNR $`\rho ^2=h,h`$). The plot also shows on the top horizontal axis the last stable orbit frequency corresponding to the total mass in question. The left vertical axis shows the SNR extracted and the right vertical axis shows the sensitivity, $`h_n^1[fS_n(f)]^1`$, (both of which are dimensionless) of LIGO-I and VIRGO instruments. While reading the sensitivity curve one should use the top and right axes and while reading the SNR curve one should use the bottom and left axes. The SNR values plotted in Fig. 1 have been computed numerically by inserting the relevant values of $`\stackrel{~}{h}(f)`$ and $`\stackrel{~}{k}(f)`$ in Eq. (10). Though we did not use it, it might help the reader to see the analytical expression of $`\rho ^2`$ obtained in the simple approximation where $`\stackrel{~}{k}(f)\stackrel{~}{h}(f)\stackrel{~}{h}^{\mathrm{uspaw}}(f)`$. Using the equations of Sec. II below (and averaging over angles as explained in Sec. IV A below) leads to
$$\rho ^2\frac{\eta }{15\pi }\left(\frac{m}{d}\right)^2_0^{F_{\mathrm{LSO}}}\frac{df}{f}\frac{1}{v(f)}\frac{1}{fS_n(f)},$$
(12)
where $`..`$ denotes the angular average, $`d`$ the distance to the source, $`v(f)=(\pi mf)^{1/3}`$ and $`S_n(f)`$ the two-sided noise given by Eq. (5). We indicated no precise detection threshold in Fig. 1 because this depends on many parameters (like the number of detectors involved). The reader should, however, have in mind that a reasonable detection threshold is, at least, $`\rho _{\mathrm{threshold}}5`$.
We note that the effective sensitivity of LIGO-I peaks near a frequency of 167 Hz which is the last stable orbit frequency for a binary of total mass of about $`27M_{}`$. The effective sensitivity of VIRGO peaks at a much lower frequency of 103 Hz. This low-frequency sensitivity of VIRGO means two important things: Firstly, lighter binaries (i.e., $`m30M_{}`$) are integrated for a longer time in the low frequency regime and, therefore, the corrections to the Fourier transform introduced in this paper are less important for such systems. This means that the uSPAw is quite good in extracting the full signal power of such binaries as evidenced in Fig. 1. Notice that, for VIRGO, the uSPAw curve follows the inSPA curve for $`m30M_{}.`$ On the contrary, LIGOโs lower sensitivity to lower frequencies makes it important to include the corrections to the Fourier transform of LSO-truncated signals from binaries of mass $`m15M_{}`$. In LIGOโs case uSPAw extracts only 75% of the full SNR implying a loss of more than 40 % of all massive binary coalescences. On the other hand, the low-frequency peak of VIRGO sensitivity means that we will have to employ the accurate Fourier domain models discussed in this paper for more massive binaries, i.e. $`m30M_{}`$. \[Note, however, that the low-frequency sensitivity of VIRGO means that, for low-mass and medium-mass binaries, it is even more crucial to use P-approximants (instead of the usually considered T-approximants) than for LIGO, in order to accurately keep track of the phasing of the many cycles accumulated at low frequencies.\]
It is fair to say that at present the most well-understood gravitational waveform is the inspiral one and thus the only reliable templates correspond to inspiral signals. It is also generally believed that binary black holes are better candidates for gravitational wave sources than binary neutron stars due to their larger masses (the average mass of observed black hole candidates is around $`8M_{}`$ ). Theoretical computations based on stellar evolution indicate that binary black holes with individual masses $`15M_{}`$ may be the only known sources that exist (hopefully) in sufficient numbers . When looking at Fig. 1, one clearly sees the importance of dealing with binary black holes with total masses in the range of $`2830M_{}`$. They lead to signals with the best SNR. However, it is precisely for such systems that the $`F_{\mathrm{LSO}}`$ is around the middle of the detection bandwidth for initial LIGO i.e. $`F_{\mathrm{LSO}}f_{\mathrm{det}}`$. The most likely sources to detect are in the problematic region discussed in this paper. This makes it imperative to not lose SNR when dealing with such signals and provides the other major motivation for our work. If our assumption that the best models of inspiral waveforms must be abruptly shut off in the time domain holds, it is essential to use the improved SPA formulas discussed in this work in order to maximise our chances of detecting inpiralling binaries. The analysis presented in this paper provides insights and techniques to deal with binary black hole signals in probably the most crucial mass range.
### E Summary of the present paper and proposals for data analysis groups
In this paper we propose analytical approximations to the Fourier-transform of the LSO-truncated time-domain inspiral waves that are very accurate even for the massive black hole binaries, the most likely sources for LIGO and VIRGO \[overlaps with FFT $`>0.99`$ for $`m40M_{}`$\] and are at the same time computationally inexpensive. We call our final new, frequency-domain filters the SPP approximants because they combine the computational convenience of stationary phase approximants with the accuracy of the (time-domain) P-approximants. Our strategy is two-fold: On the one hand we introduce a correction factor $`๐(f)`$ to the usual SPA for $`f<f_{\mathrm{up}}F_{\mathrm{LSO}}`$ which improves the SPA by taking into account the โedgeโ oscillations present when $`ff_{\mathrm{up}}`$. ($`f_{\mathrm{up}}`$ will be defined below. For Newtonian-like signals $`f_{\mathrm{up}}=F_{\mathrm{LSO}}`$, while for relativistic signals $`f_{\mathrm{up}}<F_{\mathrm{LSO}}`$.) On the other hand, we introduce a new approximation to the Fourier transform for $`f>f_{\mathrm{up}}`$ which efficiently recovers the signal power around and beyond the frequency corresponding to the last stable orbit. These features are important new steps forward as there was no formalism until now that could compute (especially for relativistic signals) Fourier transforms analytically for $`fF_{\mathrm{LSO}}`$, and in particular $`f>F_{\mathrm{LSO}}`$. These new features now make it possible to generate templates directly in the Fourier-domain, leading to a saving on the computational cost of template generation by a factor of 10 or more.
Our concrete proposal to the interferometer data analysis groups that are building the gravitational wave search software and wish to have Fourier-domain filters which are both accurate and fast-computed, is thus the following: First, we confirm that for accurate post-Newtonian template generation of binary systems of total mass $`m40M_{}`$ one needs to use a frequency-domain version of the P-approximant (previously defined only in the time-domain). For $`m<5M_{}`$ a straightforward (uncorrected for edge-effects) SPA of the P-approximants is sufficient. (They match with the exact DFT of the same time-signal with overlaps $`>0.999`$.) On the other hand, in the total mass range $`5M_{}m40M_{}`$, and assuming that one wishes an accurate frequency-domain (f-domain) representation of a time-windowed signal it is crucial to use our new SPP approximants. For $`m40M_{}`$ a straightforward DFT is recommended (but, anyway, the signal is not known with enough precision in this high mass range, where the plunge and merger signals become observationally important).
It is important to stress the position we assume in this paper: Given the absence of any detailed and precise information about the plunge signal today, we suggest that a time-truncated chirp (time-windowed signal) is currently our best bet and the modified SPA presented in this paper is the appropriate Fourier-domain representation one must use. However, this should not be taken to imply that we are claiming to have logically excluded the other possibility that the f-window may turn out to be the better choice, when we have further details about the transition from inspiral to plunge and, about the plunge waveform. Even so, we emphasize that a definitive contribution of the present work is to provide explicitly for the first time the frequency-domain version of the time-domain P-approximants which were shown in DIS to bring indispensible improvements over the usually considered T-approximants. Consequently, even in the unlikely case where a straightforward frequency-window turns out to be a better model than the time-window assumed in most of this work, one will still require the formulas given in this paper \[with the trivial change of replacing the correction factors $`๐(\zeta )`$ by a $`\theta `$ function $`\theta (F_{\mathrm{LSO}}f)`$\] to generate sufficiently accurate f-domain filters. Thus this work may not be the complete final answer but only a step ahead and a partial contribution towards defining good f-domain filters. Assumptions that seem the best we can accept, require special tools for their analysis and this paper provides them.
This paper is organised as follows: In Sec. II A, II B, II C as a prelude to later technical material, we introduce several useful physical notions and employ them to give a preliminary discussion of the questions raised by the detectability of massive-binary signals. In Sec. II D we summarise the mathematical tools used in the paper to estimate the time-truncated chirps. In Sec. III we consider time-windowed Newtonian-like signals. Sec. III A provides a short summary of the well known SPA. Sec. III B sets up the basic equations to discuss the Fourier transform of time-windowed signals. Sec. III C estimates the edge contribution to the Fourier transform coming from the non-resonant integral. This is followed by Sec. III D and III E where we elaborate in detail the construction of optimal analytic approximations to the Fourier transform of the time-windowed gravitational wave chirp (improved Newtonian SPA). In Sec. III F we compare and contrast in detail the usual SPA with our improved SPA for Newtonian-like signals. Sec. IV A addresses the new issues related to the Fourier-transform of time-windowed relativistic signals. In Sec IV B we present a new method to estimate the small non-resonant contribution in the relativistic case. In Sec. IV C we construct a new form of improved SPA for such signals (improved relativistic SPA). Combining this improved relativistic SPA with the P-approximants of DIS leads to the construction of the frequency-domain SPP approximants. In Sec. V we use the SPP approximants constructed earlier and investigate their faithfulness and effectualness in detail for the test mass case. Based on this we comment on the corresponding situation in the comparable mass case. In Sec. VI we compare the computing costs for template generation using the time-domain FFT with corresponding costs for the frequency-domain SPA and improved SPA both for Newtonian and relativistic cases. Sec. VII contains our concluding remarks.
## II Preliminary Discussion
As a preface to the technical treatments of the following Sections in which we shall construct optimal analytic approximations to the Fourier transform of the gravitational wave inspiral signal $`h(t)`$ let us start by discussing some general issues which are central to this paper.
### A Wiener filters and time-truncated inspiral signals
We briefly recall the principle underlying the optimal linear filter technique (Wiener filter). A (real) linear filter is a linear functional of the detectorโs output, $`h_{\mathrm{out}}`$, Eq.(2), say
$$K[h_{\mathrm{out}}]=_{\mathrm{}}^{\mathrm{}}๐tK(t)h_{\mathrm{out}}(t)=_{\mathrm{}}^{\mathrm{}}๐f\stackrel{~}{K}(f)\stackrel{~}{h}_{\mathrm{out}}(f)=_{\mathrm{}}^{\mathrm{}}๐f\stackrel{~}{K}^{}(f)\stackrel{~}{h}_{\mathrm{out}}(f).$$
(13)
Let us associate to any $`K(t)`$ the time-domain function $`k(t)`$ such that its FT equals $`\stackrel{~}{k}(f)S_n(f)\stackrel{~}{K}(f)`$ and let us introduce the Wiener scalar product (defined on real time-domain functions)
$$g,h_{\mathrm{}}^{\mathrm{}}\frac{df}{S_n(f)}\stackrel{~}{g}^{}(f)\stackrel{~}{h}(f)=_{\mathrm{}}^{\mathrm{}}_{\mathrm{}}^{\mathrm{}}๐t_1๐t_2g(t_1)w_1(t_1t_2)h(t_2),$$
(14)
$$\mathrm{where},w_1(\tau )=_{\mathrm{}}^{\mathrm{}}\frac{df}{S_n(f)}e^{2\pi if\tau }=w_1(\tau ),$$
(15)
is the convolution inverse of the noise correlation function $`C_n(\tau )=C_n(\tau )`$ i.e.
$$(w_1C_n)(t)=\delta (t).$$
(16)
\[Here $``$ denotes the convolution product: $`(gh)(t)_{\mathrm{}}^{\mathrm{}}๐\tau g(\tau )h(t\tau )`$\]. With this notation the action of the filter $`K`$ on $`h_{\mathrm{out}}`$ reads
$$K[h_{\mathrm{out}}]=k,h_{\mathrm{out}}S+N,$$
(17)
where $`S`$ is the filtered โsignalโ and $`N`$ the filtered โnoiseโ, defined by
$$SK[h]=k,h;NK[n]=k,n.$$
(18)
The definition of the symmetric Wiener scalar product Eq. (14) is such that the statistical average of a product of filtered noises simplify: $`\overline{k_1,nk_2,n}=\overline{k_1,nn,k_2}=k_1,k_2`$. In particular, the variance of the filtered noise $`N`$ reads $`\overline{N^2}=\overline{k,n^2}=k,k`$, so that the square of the signal-to-noise ratio (SNR) for the filter defined by any function $`k(t)`$ reads
$$\rho ^2\frac{S^2}{\overline{N^2}}=\frac{k,h^2}{k,k}=|๐ช(k,h)|^2h,h,$$
(19)
where we have defined the โoverlapโ, or normalized ambiguity function, between $`k`$ and $`h`$
$$๐ช(k,h)\frac{k,h}{\sqrt{k,kh,h}}.$$
(20)
Schwarzโs inequality guarantees that $`|๐ช(k,h)|1`$, the equality being reached only when $`k(t)=\lambda h(t)`$. (We work here in the space of real signals.) For a given signal $`h(t)`$, the choice of filter $`Kk`$ which maximizes the SNR $`\rho `$ is, in view of Eq. (19), $`k(t)=\lambda h(t)`$, where the proportionality constant is unimportant and can be taken to be one (โWiener Theoremโ). This optimal linear filter theorem applies when the full time-development of the signal $`h(t)`$ is known, the noise is stationary and has a known spectral distribution $`S_n(f)`$.
It is important to note that, saying that the best โassociatedโ filter $`k(t)`$ is simply $`k(t)=h(t)`$, means that the best time-domain filter $`K(t)`$, which must be directly correlated with the detectorโs output, $`K[h_{\mathrm{out}}]=_{\mathrm{}}^{\mathrm{}}๐tK(t)h_{\mathrm{out}}(t)`$, is a non-local (in time) functional of $`h(t)`$. Explicitly,
$$K(t)=(w_1h)(t)=_{\mathrm{}}^{\mathrm{}}๐\tau w_1(\tau )h(t\tau ).$$
(21)
This poses the question whether, for an off-line analysis of the data, one would like to store a bank of these non-local time-domain filters $`K(t)`$, which densely cover the expected parameter space. In the present paper, we shall assume that one anyway computes (nearly) on-line the Fast Fourier Transform (FFT) of the detectorโs output (which is anyway needed to factor out the frequency dependent effect of the interferometer on the GW signal), and we shall set ourselves the task of providing the best possible analytical representations of the Fourier transform of the expected signals $`\stackrel{~}{h}(f)`$. Moreover, the availability of a fast Fourier algorithm makes the filtering problem computationally less intensive in the Fourier-domain. It is well-known that the computation of a discrete correlation for all (discrete) time lags between the output and the filter \[ie the discrete version of Eq. (13) requires $`๐ช[N^2]`$ operations in the time-domain while it takes only $`๐ช[N\mathrm{log}_2(N)]`$ operations in the Fourier domain (because a time lag $`\tau `$ adds a factor $`\mathrm{exp}(2\pi if\tau )`$ in the f-domain version of Eq. (13), which is equivalent to computing a certain inverse Fourier transform). Discrete correlation in the f-domain suffers from spurious correlations for non-zero time lags but this is easily taken care of by padding the tail part of a template with large number of zeroes (see, eg. Ref for details.)
As the problem of the locality/non-locality in time will be crucial to our discussion of the inspiral signals, we wish to give an alternative discussion of Wienerโs optimal linear filter theorem. Indeed, another proof of the theorem can be obtained by introducing a โwhiteningโ transformation say $`w_{\frac{1}{2}}`$, which simplifies the properties of the noise. We define the โwhitening-kernelโ $`w_{\frac{1}{2}}(\tau )`$ by
$$w_{\frac{1}{2}}(\tau )=_{\mathrm{}}^+\mathrm{}\frac{df}{\sqrt{S_n(f)}}e^{2\pi i\tau f};\stackrel{~}{w}_{\frac{1}{2}}(f)\frac{1}{\sqrt{S_n(f)}}.$$
(22)
For any function $`g(t)`$, the action of the kernel $`w_{\frac{1}{2}}`$ on $`g`$ (i.e. the โwhitenedโ version of the function $`g`$) can be denoted by
$$g_{\frac{1}{2}}(t)(w_{\frac{1}{2}}g)(t)=_{\mathrm{}}^{\mathrm{}}๐\tau w_{\frac{1}{2}}(\tau )g(t\tau ).$$
(23)
The name โwhitening kernelโ comes from the fact that the transformed noise $`n_{\frac{1}{2}}(t)`$ is โwhiteโ i.e. uncorrelated:
$$\overline{n_{\frac{1}{2}}(t_1)n_{\frac{1}{2}}(t_2)}=_{\mathrm{}}^{\mathrm{}}๐fe^{2\pi if(t_1t_2)}=\delta (t_1t_2).$$
(24)
Note that $`w_{\frac{1}{2}}`$ is simply the convolution square-root of the Wiener kernel $`w_1`$ introduced above: $`w_{\frac{1}{2}}w_{\frac{1}{2}}=w_1`$. The Wiener theorem states then that, after having whitened all the functions, the optimal filter is simply the usual straightforward correlation between the (whitened) output and the (whitened) signal i.e.
$$K_{\mathrm{optimal}}[h_{\mathrm{out}}]=_{\mathrm{}}^{\mathrm{}}๐th_{\frac{1}{2}}(t)h_{\frac{1}{2}}^{\mathrm{out}}(t),$$
(25)
where $`h_{\frac{1}{2}}=w_{\frac{1}{2}}h`$, $`h_{\frac{1}{2}}^{\mathrm{out}}=w_{\frac{1}{2}}h^{\mathrm{out}}`$. In other words, we can think of the optimal filter as being local-in-time after the application, to all signals, of the convolution kernel $`w_{\frac{1}{2}}`$. When working with the transformed time-domain functions $`h_{\frac{1}{2}}(t)`$, $`h_{\frac{1}{2}}^{\mathrm{out}}(t)\mathrm{}`$, we shall say that we work in the โwhitened time-domainโ. In this language, the best filter in the whitened time-domain is to simply correlate (as when trying to visually superpose two time-functions) the output with a copy of the signal. This โwhitened time domainโ is conceptually useful in the present context because it introduces only a small non-locality (by small we mean here much smaller than the non-locality introduced by the Fourier transformation). Indeed, as the function $`1/\sqrt{S_n(f)}`$ has a rather flat maximum, its Fourier transform $`w_{\frac{1}{2}}(\tau )`$ is nearly a delta function as seen in Fig. 2 wherein we have plotted the whitening kernel for the initial LIGO interferometer. More precisely, $`w_{\frac{1}{2}}(\tau )`$ is an even function made of a positive spike around $`\tau =0`$, followed (on each side) by a slightly negative wing which decays fast towards zero as $`|\tau |+\mathrm{}`$. The half-width at half-maximum of the central spike is 0.18 ms. The location of the wings is around $`\tau =\pm 0.002`$ s. Therefore the non-locality contained in the whitening transformation is only between 0.2 ms and 2 ms (depending on the function on which it acts). This non-locality together with the one and a half cycle in $`w_{\frac{1}{2}}`$ is sufficient to efficiently damp both high and low frequencies so that $`h_{\frac{1}{2}}=w_{\frac{1}{2}}h`$ is a chirp whose amplitude is important only when the instantaneous frequency is around $`f_\mathrm{p}=165`$ Hz (see below). We shall use below this whitened time domain picture to discuss the important features of the expected chirp that we should try to model as well as possible.
In the present paper, we shall be primarily interested in massive compact binaries with total mass $`m=m_1+m_2`$ in the approximate range $`3M_{}m40M_{}`$. We recall that the GW signal from a compact binary is made of an inspiral signal, followed, after the last stable orbit is reached, by a plunge signal which leads to a final merger signal. Thanks to the analytical work on the motion of and GW emission from general relativistic binary systems we have quite a good analytical control of the inspiral signal. In the present paper, we shall further argue that we have also a rather good analytical control on the location of the last stable orbit (LSO), i.e. on the transition between the inspiral and plunge. First, DIS introduced a new, more robust approach to the determination of LSO based on the invariant function $`e(v)`$. More recently , a new approach to the dynamics of binary systems has confirmed the result of DIS (saying that the LSO was slightly more โinwardsโ for comparable mass systems than as predicted by the test-mass limit) and predicts values for the important physical quantities at the LSO (notably the orbital frequency) which are even nearer to the (Schwarzschild-like) ones obtained in the test-mass limit. We anticipate that further analytical progress in the problem of motion of binary systems , combined with the LSO-determination techniques given in and will soon allow one to know with more certainty and more precision the gravitational wave frequency at the LSO. Pending such a determination, we shall use as fiducial value for the GW frequency at the LSO โ when we need it for simple analytical estimates โ the usual โSchwarzschild-likeโ approximation
$$F_{\mathrm{LSO}}=4400\left(\frac{M_{}}{m}\right)\mathrm{Hz}.$$
(26)
However, in our actual numerical calculations and plots we shall use the $`\eta `$-dependent $`F_{\mathrm{LSO}}`$ corresponding to the approximation used for the energy function $`E_\mathrm{A}(v)`$. For instance, in the case of the 2PN P-approximant $`P_4`$ we have
$$F_{\mathrm{LSO}}^{P_4}=4397.2\left(\frac{1+\frac{1}{3}\eta }{1\frac{35}{36}\eta }\right)^{\frac{3}{2}}\left[2\frac{1+\frac{1}{3}\eta }{\sqrt{1\frac{9}{16}\eta +\frac{1}{36}\eta ^2}}\right]^{\frac{3}{2}}\frac{M_{}}{m}\mathrm{Hz}.$$
(27)
In the equal-mass case ($`\eta =1/4`$) this yields $`F_{\mathrm{LSO}}=5719.4M_{}/m`$ Hz. Note that the most recent determination of the LSO suggests that when $`\eta 0`$ the GW frequency at the LSO lies between Eq. (26) and Eq. (27):
$$F_{\mathrm{LSO}}=4397.2(1+0.3155\eta )\left(\frac{M_{}}{m}\right)\mathrm{Hz}.$$
(28)
Preliminary studies indicate that the plunge signal, emitted during the fast fall of the two masses towards each other following the crossing of the LSO, will last (when $`4\eta 1`$) only for a fraction of an orbital period (see and )<sup>ยง</sup><sup>ยง</sup>ยง We assume here that we are in the generic case where the spins of the coalescing objects are smallish compared to their maximal Kerr value.. As usual, one can also assume that the subsequent merger signal linked to the formation of a black hole of total mass $`m`$ contains significantly higher frequencies than the inspiral ones. Indeed, the characteristic frequency of the merger signal may be taken to be given by the real part of the most slowly damped quadrupolar normal mode of a black hole (which when neglecting the black hole spin, has a complex circular frequency $`\omega _1m_{\mathrm{bh}}=0.373670.08896i`$ ), i.e. (with $`m_{\mathrm{bh}}m`$)
$$f_{\mathrm{merger}}f_{\mathrm{bh}}\frac{0.374}{2\pi m}12000\frac{M_{}}{m}\mathrm{Hz}.$$
(29)
Eqs. (26) and (29) lead us to accept that there is a significant frequency separation between the inspiral and plunge signals and the merger one: $`f_{\mathrm{merger}}/F_{\mathrm{LSO}}2.75`$. Therefore, if we restrict our attention to systems such that the characteristic detection frequency $`f_{\mathrm{det}}`$ defined by the noise curve, stays logarithmically nearer to $`F_{\mathrm{LSO}}`$ than to $`f_{\mathrm{merger}}`$, it seems plausible that a good filter to use for GW detection can neglect the (ill-known) merger signal, but should try to model as accurately as possible the inspiral and plunge signal. For the initial LIGO noise curve, Eq. (5), the characteristic detection frequency $`f_{\mathrm{det}}`$ is 167 Hz. It is then for a total mass $`m43.5M_{}`$ that $`f_{\mathrm{det}}/F_{\mathrm{LSO}}f_{\mathrm{bh}}/f_{\mathrm{det}}`$.
We shall see below that, just before reaching the LSO, the inspiral signal is still significantly โquasi-periodicโ (with $`6`$ cycles before a significant change in instantaneous frequency). By contrast, though the plunge signal may not decay monotonically and may be oscillating, it seems reasonable to assume that the plunge lasts only for a fraction of the orbital period $`T_{\mathrm{LSO}}=2F_{\mathrm{LSO}}^1`$. Thus, in the absence of a precise knowledge of the plunge signal, a good model of the time-domain signal consists in abruptly shutting off, by a step function $`\theta (t_{\mathrm{LSO}}t)`$ the (adiabatic) inspiral signal beyond the time $`t_{\mathrm{LSO}}`$ when the last stable orbit is reached. We also tested formally the robustness of our approach by showing that our model above has a good overlap with a signal which decays smoothly on a time scale of a few (up to 3) $`F_{\mathrm{LSO}}^1`$ beyond the LSO. Because of the likely oscillatory behaviour of the plunge signal, details of the oscillations are necessary for any further improvements and we are currently working towards improving our understanding of the transition between the inspiral and the plunge .
These considerations motivate us to propose that, in the absence of knowledge of the optimal filter which should be $`k_{\mathrm{optimal}}(t)=k_{\mathrm{exact}}(t)`$, our best bet is to use as (sub-optimal) filter the time-truncated inspiral signal $`h_{\mathrm{inspiral}}(t)\theta (t_{\mathrm{LSO}}t)`$. In other words, we think that the best strategy is to use all the information available, in the time-domain, about the signal and to replace the transient plunge and higher frequency merger signals by zero, as a measure of our current ignorance. But having settled on this tactic in the time-domain, the aim of this paper is to provide the best possible frequency-domain description of such a time-windowed signal. We shall see in detail below that the Fourier transform $`\stackrel{~}{h}_{\mathrm{tw}}(f)=\mathrm{FT}[h_{\mathrm{tw}}(t)]`$ of the time-windowed signal $`h_{\mathrm{tw}}(t)h_{\mathrm{inspiral}}(t)\theta (t_{\mathrm{LSO}}t)`$ has a non-trivial structure which is not captured by the usually considered frequency-windowed stationary phase approximation $`\stackrel{~}{h}_{\mathrm{spaw}}(f)`$. In particular, for massive systems, a significant fraction of the total power is contained in the โtailโ of $`\stackrel{~}{h}_{\mathrm{tw}}(f)`$ beyond $`f=F_{\mathrm{LSO}}`$. The general result Eq. (19) for SNR obtained with any filter then says that the Fourier-domain filter $`\stackrel{~}{h}_{\mathrm{tw}}(f)`$, though sub-optimal, should be still significantly better than $`\stackrel{~}{h}_{\mathrm{spaw}}(f)`$ because (under our assumptions about the plunge + merger signal) it has better overlaps with the exact signal. We shall return below to this important issue and give further arguments (in the time-domain) to confirm the superiority of $`\stackrel{~}{h}_{\mathrm{tw}}(f)`$ over $`\stackrel{~}{h}_{\mathrm{spaw}}(f)`$ (see Fig. 9 and Fig. 10 and text around it).
### B The number of useful cycles
Often one mentions that, in the total time development of an inspiral signal, the total number of gravitational wave cycles
$$N_{\mathrm{tot}}=\frac{1}{2\pi }(\varphi _{\mathrm{end}}\varphi _{\mathrm{begin}})=_{F_{\mathrm{beg}}}^{F_{\mathrm{end}}}๐F\left(\frac{1}{2\pi }\frac{d\varphi }{dF}\right),$$
(30)
is very large. Here $`\varphi `$ is the gravitational wave phase, $`\varphi _{\mathrm{end}}`$ is the phase at the end of the inspiral regime (defined by the last stable orbit for sufficiently massive systems, i.e. for the black-hole-neutron-star and the black-hole-black-hole systems), while $`\varphi _{\mathrm{begin}}`$ is the phase when the signal enters the lower frequency (seismic) cutoff of the detector bandwidth. We have also rewritten $`N_{\mathrm{tot}}`$ as an integral over the running instantaneous gravitational wave frequency $`F`$. However, the large number $`N_{\mathrm{tot}}`$, Eq. (30), is not significant because the only really useful cycles are those which contribute most to the signal-to-noise ratio (SNR). To have a clearer idea of what one might want to mean by the notion of a useful number of cycles, let us first introduce the instantaneous number of cycles spent near some instantaneous frequency $`F`$. It is naturally defined by multiplying the integrand in Eq. (30) by $`F`$, considered as the length of an interval $`\pm \mathrm{\Delta }F=\pm \frac{F}{2}`$ around $`F`$ i.e.
$$N(F)\frac{F}{2\pi }\frac{d\varphi }{dF}\frac{F^2}{dF/dt},$$
(31)
where we have used $`d\varphi /dt2\pi F(t)`$. Note that the instantaneous $`N`$ can be considered either as a function of the running frequency $`F(t)`$, or, directly, of time.
The instantaneous number of cycles plays an important role both in defining the observability of a signal, and in controlling partially the validity of the stationary phase approximation. The square of the optimal SNR reads
$$\rho ^2\left(\frac{S}{N}\right)^2=_{\mathrm{}}^+\mathrm{}๐f\frac{|\stackrel{~}{h}(f)|^2}{S_n(f)}.$$
(32)
In the stationary phase approximation (discussed at length and improved below; but here we use standard results for orientation) the modulus of the Fourier transform of the real signal $`h(t)=2a(t)\mathrm{cos}\varphi (t)`$, reads $`|\stackrel{~}{h}(f)|a(t_f)/\sqrt{\dot{F}(t_f)}`$ where $`t_f`$ is the time when the instantaneous frequency $`F(t)`$ reaches the value $`f`$. In the following, it will be necessary to distinguish carefully between the instantaneous frequency $`F`$ and the Fourier variable $`f`$. In the present Section this distinction is not very important and we shall freely change notation $`fF`$. Similarly, the gravitational wave flux and factored flux functions which following standard notation was denoted by $`F(v)`$ and $`f(v)`$ in are denoted by $`(v)`$ and $`l(v)`$ in this paper to avoid confusion with the instantaneous gravitational wave frequency and Fourier variable respectively. Therefore, the squared modulus can be written as
$$|\stackrel{~}{h}(f)|^2\frac{a^2(f)}{df/dt}\frac{1}{f^2}N(f)a^2(f).$$
(33)
Finally, the SNR can be rewritten as
$$\rho ^2\left(\frac{S}{N}\right)^2=_{\mathrm{}}^+\mathrm{}\frac{df}{f}\frac{N(f)a^2(f)}{fS_n(f)}=_{\mathrm{}}^+\mathrm{}\frac{df}{f}\frac{h_s^2(f)}{h_n^2(f)},$$
(34)
where we have introduced the notation $`h_s^2(f)N(f)a^2(f)`$ and $`h_n^2(f)fS_n(f)`$. Here, $`h_n^2(f)`$ is the usual squared amplitude of the effective gravitational wave noise at the frequency $`f`$, i.e. the minimum gravitational wave amplitude detectable in a bandwidth $`\pm f/2`$ around frequency $`f`$. Eq. (34) exhibits that the squared amplitude of the corresponding effective gravitational wave signal is $`h_s^2(f)N(f)a^2(f)=f^2|\stackrel{~}{h}(f)|^2`$, i.e. that the โbareโ amplitude $`a(f)a(t(f))`$ is effectively multiplied by $`\sqrt{N(f)}`$ . Eq. (34) also exhibits the relative weight with which each cycle counts for detectability purposes. Per logarithmic frequency interval this weight is simply
$$w(f)a^2(f)/h_n^2(f).$$
(35)
Therefore it is natural to define the number of useful cycles as
$$N_{\mathrm{useful}}\left(_{F_{\mathrm{min}}}^{F_{\mathrm{max}}}\frac{df}{f}w(f)N(f)\right)\left(_{F_{\mathrm{min}}}^{F_{\mathrm{max}}}\frac{df}{f}w(f)\right)^1,$$
(36)
where $`F_{\mathrm{min}}`$ is the low-frequency seismic-cutoff below which $`h_n^2(f)`$ is essentially infinite and where the upper-cutoff $`F_{\mathrm{max}}`$ is the frequency at which the signal itself shuts off. For illustration, we list in Table I the number of useful cycles and the total number of cycles for representative systems and orders of approximation. For Newtonian chirps the total number of cycles is
$$N_{\mathrm{tot}}^{\mathrm{Newt}}=\frac{(\pi mf_\mathrm{s})^{5/3}(\pi mf_{\mathrm{max}})^{5/3}}{32\pi \eta },$$
(37)
where $`f_\mathrm{s}`$ is the seismic cutoff. The total number of cycles for relativistic chirps is always smaller than $`N_{\mathrm{tot}}^{\mathrm{Newt}}`$. From Table I it is clear that the number $`N_{\mathrm{useful}}`$ is usually much smaller than $`N_{\mathrm{tot}}`$, Eq. (30). Note that for massive systems $`N_{\mathrm{useful}}`$ becomes quite small. The number of useful cycles given in Table I have been computed for the initial LIGO noise curve Eq. (5). The corresponding numbers for the VIRGO noise curve Eq. (8) would be larger both because the VIRGO sensitivity curve peaks at a lower frequency, and because it is flatter.
To explore in more detail the case of systems with a small number of useful cycles we display in Fig. 3 (on a linear-log plot) the various factors of the logarithmic integrand on the RHS of Eq. (34) for two different binary systems. The instantaneous number of cycles $`N(f)`$ in the Newtonian approximation is plotted, together with square of the amplitude $`a^2(f)`$, their product the effective gravitational wave amplitude $`h_s^2(f)=N(f)a^2(f)`$, the reciprocal of effective noise $`h_\mathrm{n}^2(f)=fS_n(f)`$, (cut off after $`F_{\mathrm{max}}=F_{\mathrm{LSO}}`$) and the power per log bin of the square of the SNR $`d\rho ^2/d(\mathrm{log}f)`$. On the left, the scale on the y-axis corresponds to $`N(f)`$. On the right it corresponds to the amplitude on an arbitrary scale. Other quantities are on an arbitrary scale. The top panel is for a lighter mass binary ($`m_1=1.4M_{}`$, $`m_2=10M_{}`$) and the bottom panel for a heavier one ($`m_1=m_2=10M_{}`$).
The last Figure exhibits two useful lessons which are well-known but are particularly important to keep in mind when reading the present paper. First, because of the mass-scaling of the gravitational wave frequency at the last stable orbit, given in the lowest (Schwarzschild-like) approximation by Eq. (26), it is only for systems with total mass $`m=m_1+m_213M_{}`$ that the peak of the SNR logarithmic frequency-distribution $`f_p`$ becomes comparable, within a factor of two, to $`F_{\mathrm{max}}=F_{\mathrm{LSO}}`$. This statement critically depends on the characteristic frequency entering the considered noise-curve. For instance, in Fig. 3 we have used the initial LIGO curve Eq. (5) for which $`f_p=0.825f_0=165`$ Hz. Note that the peak of the logarithmic SNR integrand, $`f_p`$, is very close to the minimum of the effective noise amplitude $`h_n^2(f)=fS_n(f)`$, which is located (as mentioned above) at $`f_{\mathrm{det}}=0.8347f_0167`$ Hz. This is because, the frequency dependence of the factor $`N(f)(\eta v^5)^1f^{5/3}`$ (which in the effective signal $`h_s^2(f)=N(f)a^2(f)`$, favours lower frequencies) is nearly compensated by the frequency dependence of the bare amplitude $`a^2(f)v^4f^{4/3}`$ (which favours higher frequencies).
A second lesson, to be drawn from Fig. 3 is that the number of useful cycles also becomes small in the same problematic case of massive systems. To see this more clearly, let us write down the explicit expression for the instantaneous number of cycles. In the Newtonian case (for which the basic formulas are recalled in Sec. III A below), one has
$$N_{\mathrm{Newtonian}}(f)=\frac{5}{24\pi }\frac{1}{4\eta }\frac{1}{v^5},$$
(38)
where $`v=(\pi mf)^{1/3}`$. The lowest value of $`N`$ is physically that formally reached at the upper frequency-cutoff $`f=F_{\mathrm{max}}=F_{\mathrm{LSO}}`$. For $`v_{\mathrm{LSO}}=1/\sqrt{6}`$ (the โSchwarzschildโ value), the above equation reads
$$N_{\mathrm{Newtonian}}(F_{\mathrm{LSO}})=\frac{5}{24\pi }6^{5/2}\frac{1}{4\eta }\frac{5.8477}{4\eta },$$
(39)
where we recall that $`\eta 1/4`$ and that the upper value $`\eta _{\mathrm{max}}=1/4`$ is reached for equal mass systems $`m_1=m_2`$. Therefore for comparable-mass, massive systems, the Newtonian approximation suggests that the useful number of cycles will be rather small $`(6)`$ and concentrated near the LSO. As we shall see later, if we were interested in estimating the Fourier transform (FT) of an analytic Newtonian-like signal, even such a smallish number of cycles (and even a smaller one, down to $`N1)`$ would be enough to ensure that the leading correction to the stationary phase approximation is small. However, the complication comes from the combination of two facts: (i) the signal essentially terminates at the LSO crossing time $`t_{\mathrm{LSO}}`$, and (ii) one crucially needs a relativistic description of the evolution near the LSO. Using the formulas and the notation of Sec. IV below, we find that the relativistic prediction for the instantaneous number of cycles (in the adiabatic inspiral approximation) is
$$N_{\mathrm{relat}}(f)=\frac{1}{3\pi }v^4\frac{E^{}(v)}{(v)}.$$
(40)
By definition of the LSO (see e.g. the discussion in DIS) the derivative $`E^{}(v)`$ vanishes at $`v=v_{\mathrm{LSO}}`$. Therefore, the instantaneous number of cycles is smaller in the relativistic case than in the Newtonian one and actually tends to zero near the LSO. We shall tackle in Sec. IV below the problem that this vanishing of $`N(F_{\mathrm{LSO}})`$ causes for the stationary phase approximation. In this introductory Section let us only illustrate the problem by plotting the Newtonian and relativistic values of $`N(F(t))`$. In Fig. 4 we plot the instantaneous number of cycles for the Newtonian and second post-Newtonian $`P`$\- approximant waveforms in the last few cycles of the binary inspiral for a $`(20M_{}`$, $`20M_{})`$ system. We also show the development of the waveform in this interval on an arbitrary scale. These plots demonstrate how the number of useful cycles diminishes as one gets close to the LSO and lead us to anticipate the subtleties in the detectability of signals whose LSO is near the most sensitive part of the frequency response of the detector.
### C Loss of SNR due to edge effects
As we already mentioned, in addition to the problem of a vanishing instantaneous number of cycles near the LSO in the relativistic case, the main problem with the acccuracy of the stationary phase approximation comes from the fact that the Fourier transform of a time-windowed signal $`\stackrel{~}{h}_{\mathrm{tw}}(f)=\mathrm{FT}[h_{\mathrm{tw}}(t)]`$ differs from the frequency-windowed SPA $`\stackrel{~}{h}_{\mathrm{spaw}}(f)=\theta (F_{\mathrm{max}}f)\stackrel{~}{h}_{\mathrm{spa}}(f)`$ because of some โedge effectsโ in the frequency-domain, linked to the abrupt termination of the signal in the time-domain. These edge-effects comprise some additional oscillatory behaviour in $`\stackrel{~}{h}(f)`$ for $`f<F_{\mathrm{LSO}}`$, as well as a decaying oscillatory tail in the usually disregarded frequency-domain for $`f>F_{\mathrm{LSO}}`$.
Let us here anticipate on the results below and use a first-order approximation to discuss the main features of the corrections brought by the time-windowing. Roughly (see below) the exact Fourier transform can be written as
$$\stackrel{~}{h}_X(f)\stackrel{~}{h}_{\mathrm{win}}^{\mathrm{spa}}(f)+\epsilon (f),$$
(41)
where $`\stackrel{~}{h}_{\mathrm{win}}^{\mathrm{spa}}(f)=\stackrel{~}{h}^{\mathrm{spa}}(f)\theta (F_{\mathrm{max}}f)`$ is the usually considered frequency-windowed SPA. The difference $`\epsilon (f)`$ is approximately of the form
$$\epsilon (f)๐(f)\stackrel{~}{h}^{\mathrm{spa}}(f),$$
(42)
$$\mathrm{where},๐(f)๐(f)\theta (F_{\mathrm{max}}f),$$
(43)
with the correction factor $`๐(f)`$ given by Eq. (101) below (with any choice of $`\zeta (f)`$ in the present approximation) and where $`\stackrel{~}{h}^{\mathrm{spa}}(f)`$ is some smooth continuation of $`\stackrel{~}{h}_{\mathrm{win}}^{\mathrm{spa}}(f)`$ from the domain $`f<F_{\mathrm{LSO}}`$ to the domain $`f>F_{\mathrm{LSO}}`$. (For the present purpose one can think that $`\stackrel{~}{h}^{\mathrm{spa}}(f)`$ is simply given by the Newtonian approximation).
Starting from Eq. (41) one can compute the overlap between the exact $`\stackrel{~}{h}(f)`$ and the usually considered frequency-windowed SPA $`\stackrel{~}{h}_{\mathrm{win}}^{\mathrm{spa}}(f)`$
$$๐ช=\frac{h_X,h_{\mathrm{win}}^{\mathrm{spa}}}{\sqrt{h_X,h_Xh_{\mathrm{win}}^{\mathrm{spa}},h_{\mathrm{win}}^{\mathrm{spa}}}}.$$
(44)
As recalled above \[see Eq. (19)\] this overlap, if it is significantly smaller than one, represents a loss in SNR. To lowest order in $`\epsilon `$ the overlap Eq. (44) differs from $`1`$ by
$$1๐ช\frac{1}{2h_X^2}(\epsilon ^2|\epsilon ,\widehat{h}_X|^2),$$
(45)
where $`\epsilon ^2\epsilon ,\epsilon `$, and $`\widehat{h}_Xh_X/h_X`$. In inserting the explicit result Eqs.(42) and (43) for $`\epsilon `$ one sees that the second term on the RHS of Eq. (45) is much smaller than the first (because the oscillations in $`๐(f)`$ are integrated against the smooth variation of $`\stackrel{~}{h}^{\mathrm{spa}}(f)`$). Finally, if we define the weight function
$$\sigma (f)\frac{f|\stackrel{~}{h}^{\mathrm{spa}}(f)|^2}{S_n(f)}\frac{N(f)a^2(f)}{h_n^2(f)}=\frac{h_s^2(f)}{h_n^2(f)},$$
(46)
which is the full logarithmic weight function appearing in the squared SNR, Eq. (34), we can write
$$1๐ช\frac{1}{2}_0^{\mathrm{}}\frac{df}{f}\sigma (f)|๐(f)|^2\left(_0^{F_{\mathrm{max}}}\frac{df}{f}\sigma (f)\right)^1.$$
(47)
As will be discussed later (see Fig. 5) the function $`๐(f)=๐(f)\theta (F_{\mathrm{max}}f)`$ is concentrated in an interval of order $`\sqrt{\dot{F}(t_{\mathrm{max}})}`$ around $`fF_{\mathrm{max}}`$ and decays on both sides \[like $`1/\zeta (f)1/(fF_{\mathrm{max}})`$\] when $`f`$ gets away from $`F_{\mathrm{max}}`$. The total integral of $`|๐(f)|^2`$ is finite and of order unity. Thus, we see from Eq. (47) that when the characteristic frequency $`f_\mathrm{p}`$ around which $`\sigma (f)`$ is concentrated satisfies $`f_\mathrm{p}F_{\mathrm{max}}`$ we shall have a rough estimate
$$1๐ช\frac{\sigma (F_{\mathrm{max}})}{\sigma (f_\mathrm{p})}\frac{\sqrt{\dot{F}(t_{\mathrm{max}})}}{F_{\mathrm{max}}}=\frac{\sigma (F_{\mathrm{max}})}{\sigma (f_\mathrm{p})}\frac{1}{\sqrt{N(F_{\mathrm{max}})}},$$
(48)
while in the opposite limit where $`f_\mathrm{p}F_{\mathrm{max}}`$, we get the rough estimate
$$1๐ช\frac{\sqrt{\dot{F}(t_{\mathrm{max}})}}{F_{\mathrm{max}}}=\frac{1}{\sqrt{N(F_{\mathrm{max}})}}.$$
(49)
In the case where $`f_\mathrm{p}F_{\mathrm{max}}`$ the factor $`\sigma (F_{\mathrm{max}})/\sigma (f_\mathrm{p})`$ in the RHS of Eq. (48) is very small. Therefore, even if the number of cycles $`N(F_{\mathrm{max}})`$ is not very large, Eq. (48) predicts that a simple frequency-windowed SPA will have excellent overlap with the exact $`\stackrel{~}{h}(f)`$. On the other hand, Eq. (49) shows that in the reverse limit $`f_\mathrm{p}F_{\mathrm{max}}`$ which means in fact when $`f_\mathrm{p}F_{\mathrm{max}}`$, the overlap will become bad if $`\sqrt{N(F_{\mathrm{max}})}`$ is not very large. As we have seen that $`\sqrt{N(F_{\mathrm{max}})}`$ gets as low as $`\sqrt{5.85}2.4`$ in the Newtonian case, and reaches smaller values in the relativistic case, we expect that the cases where the frequency-windowed SPA has a bad overlap with $`\stackrel{~}{h}(f)`$ are those where $`f_\mathrm{p}`$ becomes comparable, say within a factor of two, to $`F_{\mathrm{max}}=F_{\mathrm{LSO}}`$. We recover the same conclusion as above, which was the conclusion already pointed out in DIS: namely the signal from massive systems \[$`m13M_{}`$ if $`f_\mathrm{p}=165`$ Hz, corresponding to $`f_0=200`$ Hz, more generally, $`m13\left(165\mathrm{Hz}/f_\mathrm{p}\right)M_{}`$\], when treated (as they should be) relativistically will be badly represented by the usual frequency-windowed SPA. This conclusion obtained from an analytical approximation is borne out by the numerical computations shown in Fig. 1 above (see also Table II below). It is mainly for such systems that the work presented in the following Sections will be mandatory. But even for lower mass systems, we shall construct here for the first time the Fourier-domain version of the (time-domain) P-approximants introduced in DIS. Since P-approximants provide better templates than the usually considered T-approximants , the work of this paper will be useful for all types of systems, even the less massive ones.
### D Fourier transform of time-truncated chirps
To introduce the detailed analysis that we shall give in the following Sections, let us start by delineating some general mathematical facts about the integrals we have to deal with. We will be interested in evaluating the Fourier-transform $`\stackrel{~}{h}(f)`$ of a time-truncated chirp $`h(t)=2a(t)\mathrm{cos}\varphi (t)\theta (t_{\mathrm{max}}t)`$. After decomposing the cosine into complex exponentials, the Fourier integral leads to a sum of two integrals of the form $`_{\mathrm{}}^{t_{\mathrm{max}}}๐ta(t)e^{i\psi _f^\pm (t)}`$ with phases $`\psi _f^\pm =2\pi ft\pm \varphi (t)`$. Let us then, for generality, discuss the properties of integrals of the type
$$I(\epsilon )=_{t_a}^{t_b}๐ta(t)e^{i\frac{\psi (t)}{\epsilon }}.$$
(50)
Here, we have introduced a formal โsmall parameterโ $`\epsilon `$ (set to unity at the end of the calculation) to formalize the fast variation of the phase compared to that of the amplitude.
Let us first note that: (i) if the phase has no stationary point $`\dot{\psi }(t)0`$ for $`t[t_a,t_b]`$, (ii) if the amplitude vanishes smoothly at the edge points $`t_a`$ and $`t_b`$, which might be pushed to $`\pm \mathrm{}`$ and (iii) if the functions $`a(t)`$ and $`\psi (t)`$ are smooth $`(๐^{\mathrm{}})`$ within the interval $`[t_a,t_b]`$, the integral $`I(\epsilon )`$ tends to zero with $`\epsilon `$, faster than any power. This can be seen by integrating by parts. To simplify the calculation we can \[thanks to the assumption (i)\] use $`\psi `$ as integration variable. This yields
$$I=_{\psi _a}^{\psi _b}๐\psi A(\psi )e^{i\psi },$$
(51)
where $`A(\psi )=\left[a(t)/\dot{\psi }(t)\right]_{t(\psi )}`$ where $`t(\psi )`$ denotes the (unique) solution in $`t`$ of $`\psi =\psi (t)`$ and where $`\psi _a=\psi (t_a),\psi _b=\psi (t_b)`$. Using $`e^{i\psi /\epsilon }=\frac{\epsilon }{i}\frac{d}{d\psi }\left(e^{i\psi /\epsilon }\right)`$, integrating by parts, and using \[thanks to assumption (ii)\] the vanishing of $`A(\psi )`$ at the edges, leads to
$$I(\epsilon )=i\epsilon _{\psi _a}^{\psi _b}๐\psi A^{}(\psi )e^{\frac{i\psi }{\epsilon }}.$$
(52)
Using \[assumption (iii)\] the vanishing of all the derivatives of $`A(\psi )`$ at the edges, we can iterate the result Eq. (52) to any order:
$$I(\epsilon )=(i\epsilon )^n_{\psi _a}^{\psi _b}๐\psi A^{(n)}(\psi )e^{\frac{i\psi }{\epsilon }}.$$
(53)
The result Eq. (53) means that, when $`\epsilon 0`$, $`I(\epsilon )=๐ช(\epsilon ^n)`$ for any integer $`n`$, i.e. $`I(\epsilon )`$ vanishes faster than any power of $`\epsilon `$. It does not mean that $`I(\epsilon )`$ is zero for any finite (but small) value of $`\epsilon `$. For instance, under stronger assumptions about the existence and properties of an analytic continuation of the function $`A(\psi )`$ in the complex $`\psi `$ plane it follows that $`I(\epsilon )Ae^{\frac{B}{\epsilon }}`$ for some constants $`A`$ and $`B`$. For reasonably small values of $`\epsilon `$ such exponentially small contributions are numerically negligible. \[We shall see later that the โsmall parameterโ $`\epsilon `$ (or better $`\epsilon /B`$) is typically of order $`1/(2\pi N)`$ where $`NF^2/\dot{F}`$, is the instantaneous number of cycles.\]
We conclude, therefore that the integral $`I`$ will be (in most relevant cases) numerically non-negligible only if the assumptions above are violated. In other words, the value of $`I(\epsilon )`$ will be dominated by the contributions coming from either (i) from stationary-phase points, $`\dot{\psi }(t_s)=0`$, or (ii) from the edge points $`t_a`$ and/or $`t_b`$. Let us (for simplicity) assume that there is (at most) one stationary-phase point $`t_s`$, and that it is of the normal parabolic type, i.e. $`\psi (t)=\psi _s+\frac{1}{2!}\ddot{\psi }_s(tt_s)^2+๐ช((tt_s)^3)`$ with $`\ddot{\psi }_s0`$. Let us also assume that $`a(t_a)0`$, $`a(t_b)0`$. \[We maintain here for the moment, the assumption (iii) above about the regularity of the functions $`a(t)`$, $`\psi (t)`$ in the closed interval $`[t_a,t_b]`$.\] Then, assuming analyticity of the involved functions, the mathematically most rigorous way of decomposing $`I`$ as the sum (modulo nonperturbative small contributions of the type discussed above) of a stationary-point contribution $`I_{\mathrm{stationary}}`$ and of edge contributions $`I_{\mathrm{edge}}=I_{\mathrm{edge}}^a+I_{\mathrm{edge}}^b`$ is to deform the original (real) contour of integration into the complex plane . The deformed contour must be such that near $`t_s`$ it leads to a basic integral of the type $`_\alpha ^\beta ๐xe^{bx^2}\left[c_0+c_1x+c_2x^2+\mathrm{}\right]`$, while near each end point it leads to integrals of the type $`_0^\gamma ๐ye^{cy}\left[d_0+d_1y+d_2y^2+\mathrm{}\right]`$. It is then easy to find the structure of the expansion of both $`I_{\mathrm{stationary}}`$ and $`I_{\mathrm{edge}}`$ in powers of $`\epsilon `$, as $`\epsilon 0`$. For instance, it is convenient near $`t_s`$ to introduce a scaled variable: $`tt_s=\epsilon ^{\frac{1}{2}}\tau `$ (before rotating $`\tau `$ to complex values $`\tau =e^{\pm \frac{i\pi }{4}}x`$), so that the phase scales as
$$\frac{\psi }{\epsilon }=\frac{\psi _s}{\epsilon }+\frac{1}{2!}\ddot{\psi }_s\tau ^2+\frac{\sqrt{\epsilon }}{3!}\psi _s^{(3)}\tau ^3+\mathrm{},$$
(54)
where $`\psi _s^{(3)}d^3\psi _s/dt^3`$. Expanding then the integrand of this $`\tau `$-integral in powers of $`\epsilon ^{\frac{1}{2}}`$ leads to an integral of the symbolic type
$$I_{\mathrm{stationary}}\epsilon ^{\frac{1}{2}}๐xe^{x^2}\left[1+\epsilon ^{\frac{1}{2}}x^{\mathrm{odd}}+\epsilon x^{\mathrm{even}}+\epsilon ^{\frac{3}{2}}x^{\mathrm{odd}}+\mathrm{}\right],$$
(55)
where $`x^{\mathrm{odd}}(x^{\mathrm{even}})`$ denotes a sum of terms $`x^{2k+1}(x^{2k})`$. This yields (using the fact that the terms $`_{\alpha /\sqrt{\epsilon }}^{\beta /\sqrt{\epsilon }}๐xe^{x^2}x^{\mathrm{odd}}`$ are exponentially small) an expansion of the type
$$I_{\mathrm{stationary}}\epsilon ^{\frac{1}{2}}\left[C_0+C_1\epsilon +C_2\epsilon ^2+\mathrm{}\right].$$
(56)
The structure of the โedgeโ contribution $`I_{\mathrm{edge}}`$ can be obtained by a similar technique. Now the appropriate scaling is different, e.g. $`tt_a=\epsilon \tau `$, and one ends up with integrals of the type $`_0^{\gamma /\epsilon }๐ye^yy^n`$. This yields,
$$I_{\mathrm{edge}}\epsilon \left[D_0+D_1\epsilon +D_2\epsilon ^2\mathrm{}\right].$$
(57)
The aim of this preliminary discussion was to point out the structures Eqs. (56) and (57) of the two main contributions to a generic oscillatory integral of the form Eq. (50). Note that while the leading contribution is given by the lowest order term in the stationary-point or saddle-point expansion Eq. (56), the next-to-leading contribution comes from the lowest order edge-correction Eq. (57). One generally expects that $`I_{\mathrm{edge}}`$ will be only $`\sqrt{\epsilon }`$, i.e. $`1/\sqrt{2\pi N}`$, smaller than $`I_{\mathrm{stationary}}`$. We shall give below the explicit expressions for the first two terms in both expansions Eqs. (56) and (57). We shall see that each coefficient $`C_0,C_1,C_2\mathrm{}\mathrm{}D_0,D_1,\mathrm{}`$ in Eqs. (56) and (57) is a combination of derivatives (of increasing total order) of $`a(t)`$ and $`\psi (t)`$ evaluated at $`t_s`$ for Eq. (56) and at $`t_a`$ or $`t_b`$ for Eq. (57). We note in advance that, for actual calculations, the simplest way to evaluate the explicit forms of the expansions Eqs. (56) and (57) is not necessarily to follow the complex-contour route sketched above. In the case of Eq. (56) one can deal directly with the original stationary-point-expanded integral written as $`\epsilon ^{\frac{1}{2}}๐\tau e^{i\ddot{\psi }_s\tau ^2}\left[\tau _0+c_1\tau +\mathrm{}\right]`$, and in the case of Eq. (57) the simplest is to keep the boundary terms in the integration-by-parts approach Eqs. (52) and (53) given above for the simple case where they were neither stationary phase points, nor boundary contributions.
To put in context the analysis that we shall perform below, let us finally mention two serious limitations of the assumptions leading to Eqs. (56) and (57). First, in the analysis above, based on the introduction of the formal parameter $`\epsilon 0`$, we have assumed that the stationary-phase point $`t_s`$ was parametrically separated from the edges $`t_a`$ or $`t_b`$, i.e. that $`|t_st_a|`$ and $`|t_st_b|`$ were much larger than the characteristic Gaussian width $`\mathrm{\Delta }t=\sqrt{\epsilon /\ddot{\psi }_s}=๐ช(\sqrt{\epsilon })`$ associated to the stationary point. As we shall see, this limitation is unacceptable for the application we have in mind and we shall have to introduce new tools to overcome it. A second limitation (which compounds with the first and will lead to an unavoidable complexity of our treatment) is the seemingly innocent assumption (iii) above, namely the hypothesis that the functions $`a(t)`$ and $`\psi (t)`$ are infinitely differentiable within the closed interval $`[t_a,t_b]`$ (i.e. including also the end points). As we shall see, in the physically relevant case of a relativistic (adiabatic) chirp the functions $`a(t)`$ and $`\psi (t)`$ will not be $`๐^{\mathrm{}}`$ at the physically imposed upper cutoff $`t_b=t_{\mathrm{LSO}}`$. This will require us to introduce new types of expansions and new tools to deal with the relativistic edge contribution in addition to the modification of the stationary-phase approximation needed in the case where $`t_s`$ is near the edge, in the sense that $`|t_st_b|=๐ช(\sqrt{\epsilon })`$.
## III Improved stationary phase approximation for time-windowed Newtonian-like signals
### A The usual stationary phase approximation
Let us begin by a quick recall of the usual treatment of the stationary phase approximation to a chirp. Consider a signal,
$`h(t)`$ $`=`$ $`2a(t)\mathrm{cos}\varphi (t)=a(t)e^{i\varphi (t)}+a(t)e^{i\varphi (t)},`$ (59)
$`\mathrm{where},{\displaystyle \frac{d\varphi (t)}{dt}}`$ $``$ $`2\pi F(t)>0.`$ (60)
We shall say that a signal is โNewtonian-likeโ if the instantaneous frequency $`F(t)`$, defined by Eq. (60), increases without limit when $`t`$ runs over its full (mathematically allowed) range. (Note that we conventionally consider only positive instantaneous frequencies.) For instance, at the Newtonian order, the explicit forms for the chirp amplitude, phase and frequency of the gravitational waves are respectively given by:
$`a(t)`$ $`=`$ $`๐_{}(\pi F(t))^{2/3},`$ (62)
$`\varphi (t)`$ $`=`$ $`\varphi _c2\left[{\displaystyle \frac{(t_ct)}{5}}\right]^{5/8},`$ (63)
$`\pi F(t)`$ $`=`$ $`\left[{\displaystyle \frac{5}{256(t_ct)}}\right]^{3/8},`$ (64)
where $``$ is the chirp mass given by $`=\eta ^{3/5}m`$ in terms of the total mass $`m`$ and the symmetric mass ratio $`\eta `$; $`\varphi _c`$ the gravitational wave phase at instant of coalescence $`t_c`$ and $`๐_{}`$ is the product of a function of different angles, characterising the relative orientations of the binary and the detector, with the ratio $`/d`$ where $`d`$ is the distance to the source (see below). Note that the function $`F(t)`$ increases without limit as $`t`$ tends to the formal coalescence time $`t_c`$.
Coming back to a general signal, the Fourier transform is defined by Eq. (1). Because the signal $`h(t)`$ is real, we have the identity $`\stackrel{~}{h}(f)(\stackrel{~}{h}(f))^{}`$. It therefore suffices to compute the Fourier transform for positive values of the frequency $`f`$. \[Note that we use a lower case letter to distinguish the Fourier variable $`f`$ from the instantaneous frequency $`F(t)`$.\] The Fourier transform of a generic signal of the form Eq. (59) reads
$`\stackrel{~}{h}(f)`$ $`=`$ $`\stackrel{~}{h}_{}(f)+\stackrel{~}{h}_+(f),`$ (66)
$`\mathrm{where},\stackrel{~}{h}_{}(f)`$ $``$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}}๐ta(t)e^{i(2\pi ft\varphi (t))},`$ (67)
$`\stackrel{~}{h}_+(f)`$ $``$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}}๐ta(t)e^{i(2\pi ft+\varphi (t))}.`$ (68)
The integrands of $`\stackrel{~}{h}_\pm (f)`$ are violently oscillating and thus their dominant contributions come from the vicinity of the stationary points of their phase (when such points exist). When $`f>0`$ (which we shall henceforth assume), only the $`\stackrel{~}{h}_{}(f)`$ term has such a saddle-point. Therefore, we can write the approximation
$`\stackrel{~}{h}(f)`$ $``$ $`\stackrel{~}{h}_{}(f){\displaystyle _{\mathrm{}}^{\mathrm{}}}๐ta(t)e^{i\psi _f(t)},`$ (70)
$`\mathrm{where},\psi _f(t)`$ $``$ $`2\pi ft\varphi (t).`$ (71)
The saddle-point of the phase $`\psi _f(t)`$ is the value, say $`t_f`$, of the time variable $`t`$ where $`d\psi _f(t)/dt=0`$, i.e. it is the solution of the equation $`F(t_f)=f`$. The dominant contribution to the integral Eq. (70) now comes from a time interval near $`t=t_f`$. When the second time derivative of the phase at the saddle-point does not vanish, i.e. when $`\dot{F}(t_f)0`$, one can estimate Eq. (70) by replacing $`\psi _f(t)`$ \[and $`a(t)`$\] by truncated Taylor expansions near $`t=t_f`$, namely
$`\psi _f(t)`$ $``$ $`\psi _f(t_f)\pi \dot{F}(t_f)(tt_f)^2,`$ (73)
$`a(t)`$ $``$ $`a(t_f).`$ (74)
(The zeroth-order term in Eq. (74) is enough because the first order term $`\dot{a}(t_f)(tt_f)`$ vanishes after integration over $`t`$). This leads to a Gaussian integral
$$\stackrel{~}{h}(f)_{\mathrm{}}^{\mathrm{}}๐ta(t_f)e^{i\psi _f(t_f)i\pi \dot{F}(t_f)(tt_f)^2}.$$
(75)
Evaluating this Gaussian integral, one finally obtains the well-known expression for the usual SPA (hereafter abbreviated as uSPA):
$$\stackrel{~}{h}^{\mathrm{uspa}}(f)=\frac{a(t_f)}{\sqrt{\dot{F}(t_f)}}e^{i\left[\psi _f(t_f)\pi /4\right]},$$
(76)
where $`\psi _f(t_f)`$ is the value of $`\psi _f(t)`$ at $`t=t_f`$.
The conditions for the validity of the SPA are usually assumed to be $`\epsilon _11`$, $`\epsilon _21`$, where
$$\epsilon _1\left|\frac{\dot{a}(t)}{a(t)\dot{\varphi }(t)}\right|;\epsilon _2\left|\frac{\ddot{\varphi }(t)}{\dot{\varphi }^2(t)}\right|=\left|\frac{1}{2\pi }\frac{\dot{F}(t)}{F^2(t)}\right|=\frac{1}{2\pi N}.$$
(77)
One can assess in a more precise quantitative manner the accuracy of the SPA by computing the leading correction to the integral Eq. (75). This leading correction will be given by keeping more terms in the Taylor expansions Eqs. (73) and (74). To keep track of what one means by the โnext order termโ in the SPA expansion it is convenient (as in Section II D above) to formalize the fast variation of the phases $`\varphi (t)`$ and $`\psi (t)`$ by considering an integral of the form $`I=๐ta(t)\mathrm{exp}(i\psi (t)/\epsilon )`$ with a โsmallโ parameter $`\epsilon `$ (set to unity at the end of the calculation). It is then easy to see \[e.g. after the introduction of a rescaled time-variable: $`tt_f=\epsilon ^{1/2}\tau `$ where $`t_f`$ denotes, as above, the saddle point of the phase $`\psi _f(t)`$\] that the leading correction to the result Eq. (76) will be of fractional order $`\epsilon `$ \[as exhibited in Eq. (56)\] and will come from keeping two more terms in both the Taylor expansions Eqs. (73) and (74). Expanding in powers of $`\epsilon ^{1/2}`$ leads to integrals of the form $`_{\mathrm{}}^{\mathrm{}}๐\tau \tau ^n\mathrm{exp}(i\pi \dot{F}(t_f)\tau ^2)`$ with $`n6`$. Finally, one finds that the sum of the usual SPA and of its leading correction is equivalent to multiplying Eq. (76) by the correcting phase factor $`e^{i\delta }`$ where ($`F^{(3)}`$ denoting $`d^3F/dt^3`$)
$$\delta =\frac{1}{2\pi \dot{F}(t_f)}\left[\frac{1}{2}\frac{\ddot{a}}{a}+\frac{1}{2}\frac{\dot{a}}{a}\frac{\ddot{F}}{\dot{F}}+\frac{1}{8}\frac{F^{(3)}}{\dot{F}}\frac{5}{24}\left(\frac{\ddot{F}}{\dot{F}}\right)^2\right]_{t=t_f}.$$
(78)
Therefore, a quantitatively precise criterion for the local validity of the SPA is $`\epsilon _{\mathrm{loc}}|\delta |1`$. In the case of power-law chirps, $`\epsilon _{\mathrm{loc}}`$ is equal to one-fifth of the criterion explicitly given in the recent study of the validity of the SPA. In the case of Newtonian chirps, Eq. (78) yields
$$\delta =\frac{23}{24}\left(\frac{1}{9\pi }\frac{\dot{F}}{F^2}\right)=\frac{23}{24}\left(\frac{1}{9\pi N}\right).$$
(79)
Written in terms of $`v=(\pi mF)^{1/3}`$ this reads $`\delta =(92/45)\eta v^5`$ which agrees with the corresponding result in . It is interesting to note that Eq. (79) formally predicts, at the LSO, $`\delta _{LSO}=(4\eta )\times 0.58\%`$ which is quite small. Alternatively, one can say that Eq. (79) predicts that even if the instantaneous number of cycles were as small as $`N1`$, the local correction to the SPA would be small ($`\delta =0.0339/N`$). This result does not mean, however, that we can use the usual SPA Eq. (76) to estimate with sufficient accuracy the FT of a real inspiral signal. Indeed, even if we were considering a Newtonian-like signal (where $`N`$ stays away from zero at the LSO) the correcting phase factor Eq. (78) represents just the local correction to the SPA, i.e. the correction due to higher-terms in the local expansion near the saddle-point. There are also global corrections to the SPA coming from the entire integration domain, and, most importantly (as emphasized in ) from the end-points of the time-integration. In addition, there is also a correction coming from the neglected contribution $`\stackrel{~}{h}_+(f)`$ in Eq. (67). Before considering them in detail, let us also note that Eq. (78) indicates that $`\delta `$ blows up to infinity, at the LSO, in the case of a relativistic GW chirp (because $`F(t)F_{\mathrm{LSO}}a(t_{\mathrm{LSO}}t)^{1/2}`$ there; see below). This shows again that, independently of the problems linked to the time-windowing, relativistic signals will pose special difficulties. But let us start by studying the simpler case of time-truncated Newtonian-like signals, by which we mean that $`NF^2/\dot{F}`$ stays away from zero at the upper time cut-off.
### B Beyond the usual stationary phase approximation
We therefore consider time-domain signals of the form
$$h(t)=2a(t)\mathrm{cos}\varphi (t)\theta (t_{\mathrm{max}}t),$$
(80)
where $`\theta `$ denotes the Heaviside step function, ($`\theta (x)=1`$, for $`x>0`$ and $`\theta (x)=0`$, for $`x<0`$). This time-windowing has three effects: (i) it induces oscillations in the usually considered frequency-domain $`f<F_{\mathrm{max}}`$, (ii) it generates a tail in the usually disregarded frequency-domain $`f>F_{\mathrm{max}}`$, and (iii) it renders non-negligible the โnon-resonantโ contribution $`\stackrel{~}{h}_+(f)`$. Here, $`F_{\mathrm{max}}`$ denotes the instantaneous gravitational wave frequency reached at $`t=t_{\mathrm{max}}`$, i.e. $`F_{\mathrm{max}}F(t_{\mathrm{max}})`$. The main purpose of the present paper is to model and estimate, analytically, as accurately as possible all these effects. The case where the saddle-point $`t_f`$ is (below and) far away from the upper-cutoff $`t_{\mathrm{max}}`$ has been recently considered in . However, this case is not the physically relevant one. As pointed out in DIS, the case where the usual SPA becomes unacceptably inaccurate is the case of massive systems for which the signal emits not many cycles in the detectorโs bandwidth before crossing the last stable orbit. In this case the most important frequencies are located around the effective-cutoff frequency $`F_{\mathrm{max}}F_{\mathrm{LSO}}`$ (and as we shall see below, it is important to estimate the Fourier transform accurately, both for $`f<F_{\mathrm{max}}`$ and for $`f>F_{\mathrm{max}}`$). Here, we shall provide an approximate analytical treatment valid in this crucial range of frequencies.
Let us first state clearly our notation. We decompose the FT of the time-windowed signal Eq. (80) as
$`\stackrel{~}{h}(f)`$ $`=`$ $`\stackrel{~}{h}_{}(f)+\stackrel{~}{h}_+(f),`$ (82)
$`\mathrm{where},\stackrel{~}{h}_{}(f)`$ $`=`$ $`{\displaystyle _{\mathrm{}}^{t_{\mathrm{max}}}}๐ta(t)e^{i\psi _f^{}(t)},`$ (83)
$`\stackrel{~}{h}_+(f)`$ $`=`$ $`{\displaystyle _{\mathrm{}}^{t_{\mathrm{max}}}}๐ta(t)e^{i\psi _f^+(t)},`$ (84)
$`\psi _f^{}(t)\psi _f(t)`$ $``$ $`2\pi ft\varphi (t),`$ (85)
$`\psi _f^+(t)`$ $``$ $`2\pi ft+\varphi (t).`$ (86)
We shall refer to the contribution $`\stackrel{~}{h}_{}(f)`$ as the โresonantโ contribution because the equation defining the saddle point it contains, $`F(t_f)=f`$, corresponds to a resonance between the Fourier-frequency $`f`$ and the instantaneous gravitational wave frequency $`F(t)`$.
### C Edge contribution to the non-resonant integral $`\stackrel{~}{h}_+(f)`$
Before dealing with the oscillatory and tail corrections to the resonant contribution $`\stackrel{~}{h}_{}(f)`$ of $`\stackrel{~}{h}(f)`$, we shall first deal with the non-resonant contribution $`\stackrel{~}{h}_+(f)`$. When $`f>0`$, the phase $`\psi _f^+(t)`$ in contrast to $`\psi _f^{}(t)`$ has no stationary point. Let us therefore consider the general problem of approximating an integral of the form
$$I=_{t_a}^{t_b}๐ta(t)e^{i\psi (t)},$$
(87)
where $`\psi (t)`$ is always monotonically varying $`(\dot{\psi }0)`$. We can then use $`\psi `$ as integration variable. This yields, as above,
$$I=_{\psi _a}^{\psi _b}๐\psi A(\psi )e^{i\psi },$$
(88)
where $`A(\psi )=\left[a(t)/\dot{\psi }(t)\right]_{t(\psi )}`$ where $`t(\psi )`$ denotes the (unique) solution in $`t`$ of $`\psi =\psi (t)`$ and where $`\psi _a=\psi (t_a),\psi _b=\psi (t_b)`$. One can treat $`\psi `$ as a fast varying phase, so that by comparison, the amplitude $`A(\psi )`$ varies slowly when $`\psi `$ varies by $`2\pi `$ (as above this could be formalized by the formal replacement $`\psi \psi /\epsilon `$). In other words, instead of a SPA, we are in a WKB like approximation where one can expand in the slowness of variation of $`A(\psi )`$, i.e. we can expand in successive derivatives $`d^nA(\psi )/d\psi ^n`$. This expansion is obtained by successively integrating Eq. (51) by parts \[using $`e^{i\psi }\frac{d}{d\psi }(e^{i\psi }/i)`$\]. Contrary to Section II D above we now keep the edge contribution coming from the boundaries. For instance at second order this leads to
$$I=\left[\frac{e^{i\psi }}{i}\left[A(\psi )+iA^{}(\psi )\right]\right]_a^b+i^2_{\psi _a}^{\psi _b}๐\psi A^{\prime \prime }(\psi )e^{i\psi }.$$
(89)
It is easy to re-express this result in terms of the original time variable $`t`$ by replacing $`A(\psi )a(t)/\dot{\psi }(t)`$ and $`d/d\psi =(\dot{\psi })^1d/dt`$. We deduce from Eq. (89) the full โedgeโ contribution to the integral $`I`$ (as explained in Section II D, the โbulkโ contribution is exponentially small):
$$I_{\mathrm{edge}}=\left[\frac{e^{i\psi }}{i}\left(A(\psi )+iA^{}(\psi )+\mathrm{}+i^nA^{(n)}(\psi )+\mathrm{}\right)\right]_a^b.$$
(90)
This is the explicit form of the parametric expansion sketched in Eq. (57). It can be expressed in terms of the time derivatives of $`a(t)`$ and $`\psi (t)`$ by using the replacement rules just mentioned. In particular, at the leading and next-to-leading order it reads explicitly
$$I_{\mathrm{edge}}=\left[\frac{a(t)}{i\dot{\psi }(t)}e^{i\psi (t)}\left\{1+\frac{1}{i\dot{\psi }(t)}\left(\frac{\ddot{\psi }(t)}{\dot{\psi }(t)}\frac{\dot{a}(t)}{a(t)}\right)\right\}\right]_{t_a}^{t_b}.$$
(91)
If we apply this general result to $`\stackrel{~}{h}_+(f)`$ Eq. (84), we get the estimate
$`\stackrel{~}{h}_+(f)`$ $``$ $`\stackrel{~}{h}_+^{\mathrm{edge}}(f){\displaystyle \frac{a(t_{\mathrm{max}})}{i\dot{\psi }_f^+(t_{\mathrm{max}})}}e^{i\psi _f^+(t_{\mathrm{max}})}\left[1+{\displaystyle \frac{1}{i\dot{\psi }_f^+(t_{\mathrm{max}})}}\left({\displaystyle \frac{\ddot{\psi }_f^+(t_{\mathrm{max}})}{\dot{\psi }_f^+(t_{\mathrm{max}})}}{\displaystyle \frac{\dot{a}(t_{\mathrm{max}})}{a(t_{\mathrm{max}})}}\right)\right]`$ (93)
$``$ $`{\displaystyle \frac{a(t_{\mathrm{max}})}{2\pi i(f+F_{\mathrm{max}})}}e^{i\psi _f^+(t_{\mathrm{max}})}\left[1+{\displaystyle \frac{1}{2\pi i(f+F_{\mathrm{max}})}}\left({\displaystyle \frac{\dot{F}_{\mathrm{max}}}{(f+F_{\mathrm{max}})}}{\displaystyle \frac{\dot{a}(t_{\mathrm{max}})}{a(t_{\mathrm{max}})}}\right)\right].`$ (94)
Note that, in the approximation where the amplitude is taken as Newtonian-like, i.e., $`a(t)=Cv_F^2(F(t))^{2/3}`$, the last term in Eq. (94) reads $`\dot{a}(t_{\mathrm{max}})/a(t_{\mathrm{max}})=\frac{2}{3}\dot{F}_{\mathrm{max}}/F_{\mathrm{max}}.`$
### D Improved stationary phase approximation when $`f<F_{\mathrm{max}}`$
Let us now consider the dominant contribution $`\stackrel{~}{h}_{}(f)`$, Eq. (83). We start by considering the case where the Fourier variable $`f`$ \[of the Fourier transform of the time-windowed signal Eq. (80)\] is smaller than $`F_{\mathrm{max}}`$, but near $`F_{\mathrm{max}}`$. As will become clear from the formulas below, the interval around $`F_{\mathrm{max}}`$ where it is needed to improve the usual SPA, is the range,
$$|fF_{\mathrm{max}}|(\mathrm{few})\sqrt{\dot{F}(t_f)}.$$
(95)
In the case where $`f`$ is in the range Eq. (95) with $`f<F_{\mathrm{max}}`$, there is a saddle-point $`t_f`$ in the first term of the exact integral Eq. (67), and one can still use the parabolic approximation Eq. (73) to the phase $`\psi _f(t)`$, and the lowest approximation Eq. (74) to the amplitude $`a(t)`$. \[Indeed, the work of the previous Section has shown that the local corrections to the integral Eq. (70), coming from the inclusion of more terms in Eqs. (73) and (74), were quite small as long as $`N1`$\]. Therefore, in this case, the resonant contribution to the Fourier transform becomes:
$`\stackrel{~}{h}_{}(f)`$ $``$ $`{\displaystyle _{\mathrm{}}^{t_{\mathrm{max}}}}๐ta(t)e^{i\psi _f(t)},`$ (97)
$``$ $`a(t_f)e^{i\psi _f(t_f)}{\displaystyle _{\mathrm{}}^{t_{\mathrm{max}}}}๐te^{i\pi \dot{F}(t_f)(tt_f)^2}.`$ (98)
The crucial difference between the Eqs. (75) and (98) is that the full Gaussian integral has become a complex Fresnel integral which may be evaluated in terms of the complementary error function. Let us recall that the complementary error function $`\mathrm{erfc}(z)1\mathrm{erf}(z)`$ is defined by
$$\mathrm{erfc}(z)=\frac{2}{\sqrt{\pi }}_z^+\mathrm{}e^{x^2}๐x.$$
(99)
It takes on the real axis the particular values $`\mathrm{erfc}(+\mathrm{})=0`$, $`\mathrm{erfc}(0)=1`$, and $`\mathrm{erfc}(\mathrm{})=2`$. By rotating the integration contour in the complex plane $`(x=e^{\frac{i\pi }{4}}\xi )`$ and shifting the new integration variable $`\xi `$, we get the following useful integration formula:
$$_{\mathrm{}}^{\tau _m}๐\tau e^{i(a\tau ^2+2b\tau +c)}=\frac{1}{2}\sqrt{\frac{\pi }{a}}e^{\frac{i\pi }{4}}e^{i\frac{b^2ac}{a}}\mathrm{erfc}\left[e^{\frac{i\pi }{4}}\sqrt{a}(\tau _m+\frac{b}{a})\right].$$
(100)
The formula Eq. (100) motivates us to define the following auxiliary function
$$๐(\zeta )\frac{1}{2}\mathrm{erfc}\left(\mathrm{e}^{\frac{\mathrm{i}\pi }{4}}\zeta \right).$$
(101)
It is useful to note that $`๐(+\mathrm{})=0`$, $`๐(0)=1/2`$, and $`๐(\mathrm{})=1`$. Moreover, the leading terms in the asymptotic expansions of $`๐(\zeta )`$ as $`\zeta \pm \mathrm{}`$ are
$`\zeta \mathrm{},๐(\zeta )`$ $``$ $`1+{\displaystyle \frac{e^{\frac{i\pi }{4}}}{2\sqrt{\pi }}}{\displaystyle \frac{e^{i\zeta ^2}}{\zeta }},`$ (103)
$`\zeta +\mathrm{},๐(\zeta )`$ $``$ $`{\displaystyle \frac{e^{\frac{i\pi }{4}}}{2\sqrt{\pi }}}{\displaystyle \frac{e^{i\zeta ^2}}{\zeta }}.`$ (104)
As the auxiliary function $`๐(\zeta )`$ plays an important role in our work we plot it in Fig. 5. From the Figure we note that the real part of the above complex function is a softened version of a step function $`\theta (\zeta )`$ and the imaginary part an oscillating function vanishing in the limits $`\pm \mathrm{}`$, as well as at $`\zeta =0`$.
Armed with this definition one finds that the right hand side of Eq. (98) yields the approximation
$`\stackrel{~}{h}_{}(f)`$ $``$ $`๐(\zeta _0(f))\stackrel{~}{h}^{\mathrm{uspa}}(f),`$ (106)
$`\mathrm{where},\zeta _0(f)`$ $``$ $`\sqrt{\pi \dot{F}(t_f)}(t_ft_{\mathrm{max}}).`$ (107)
In words, when $`f<F_{\mathrm{max}}`$ one can correct for the โedge effectsโ caused by the cutoff at $`t_{\mathrm{max}}`$ by multiplying the usual SPA $`\stackrel{~}{h}^{\mathrm{uspa}}(f)`$ given in Eq. (76) by a complex โcorrection factorโ $`๐(\zeta _0(f))`$. The expression Eq. (106) gives very good overlaps with the exact discrete Fourier transform (DFT) of the time-windowed signal Eq. (80). However, it is possible to do even better by a slight modification of the argument $`\zeta _0(f)`$, Eq. (107).
To understand how a slight modification of the argument $`\zeta (f)`$ of the auxiliary function $`๐(\zeta )`$ can improve both the visual agreement (even quite far away from $`F_{\mathrm{max}}`$) and the overlap with the exact DFT of the time-windowed signal Eq. (80) we have to take into account the asymptotic expansion Eq. (103). Indeed, on the one hand, when inserting the expansion Eq. (103) into Eq. (106), using Eq. (76) for $`\stackrel{~}{h}^{\mathrm{uspa}}(f)`$ and allowing for a more general frequency dependent argument $`\zeta (f)`$, we find that $`\stackrel{~}{h}(f)`$ differs from $`\stackrel{~}{h}^{\mathrm{uspa}}(f)`$ (in the domain $`\zeta (f)\mathrm{}`$ i.e. $`fF_{\mathrm{max}}`$) by a correction term proportional to $`e^{i\pi /2}e^{i[\psi _f(t_f)\zeta ^2]}/\zeta `$. On the other hand, a different way of estimating this edge correction consists in writing Eq. (98) as an integral between $`\mathrm{}`$ and $`+\mathrm{}`$ minus a โcorrectingโ integral between $`t_{\mathrm{max}}`$ and $`+\mathrm{}`$. As was discussed in Section III D the latter integral can be estimated by successive integration by parts. This gives \[see Eq. (91)\] a first order correction term proportional to $`e^{+i\pi /2}a(t_{\mathrm{max}})e^{i\psi _f(t_{\mathrm{max}})}/\dot{\psi }_f(t_{\mathrm{max}})`$. The phasing of this edge correction can be made to agree perfectly with the phasing predicted by the form Eq. (106) written with a generalized argument $`\zeta (f)`$ if $`(\psi _f(t_f)\zeta ^2)=\psi _f(t_{\mathrm{max}})`$. This leads us to define, in the domain $`\zeta <0`$, i.e. $`f<F_{\mathrm{max}}`$, the new argument
$$\zeta _<\sqrt{\psi _f(t_f)\psi _f(t_{\mathrm{max}})}.$$
(108)
In the left part of the crucial region Eq. (95) the argument $`\zeta _<(f)`$ Eq. (108) is nearly identical to the previous result $`\zeta _0(f)`$ Eq. (107), as is seen from Eq. (73). However, we have checked that the replacement of $`\zeta _0`$ by $`\zeta _<`$ improves both the visual agreement and the overlap with the exact $`\stackrel{~}{h}(f)`$. Let us also note in passing that an amplitude proportional to $`\zeta ^1`$ of the correction term to $`\stackrel{~}{h}^{\mathrm{uspa}}(f)`$ derived from the asymptotic expansion of Eq. (103) is consistent with the different analytical treatment used in which was valid only for $`fF_{\mathrm{max}}`$, i.e. large, negative $`\zeta `$. By contrast our approach based on the function $`๐(\zeta )`$ is adequate in the full range $`\mathrm{}<\zeta 0`$ without exhibiting any fictitious blowup at $`\zeta =0`$ (remember $`๐(0)=1/2`$). \[Our approach is also valid in the region $`\zeta >0`$, i.e. $`f>F_{\mathrm{max}}`$, but for an improved treatment of this domain, we shall find it convenient to modify further the argument $`\zeta (f)`$ in the following Section.\]
Summarizing: we propose as final result for the resonant part of our improved SPA for Newtonian-like signals (or inSPA)
$$fF_{\mathrm{max}}:\stackrel{~}{h}_<^{\mathrm{inspa}}=๐(\zeta _<(f))\frac{a(t_f)}{\sqrt{\dot{F}(t_f)}}e^{i\left(\psi _f(t_f)\pi /4\right)}.$$
(109)
The corresponding total improved approximation $`\stackrel{~}{h}^{\mathrm{intot}}`$ to the Fourier transform $`\stackrel{~}{h}=\stackrel{~}{h}_{}(f)+\stackrel{~}{h}_+(f)`$ is the sum of Eqs. (94) and (109). Note that the ratio $`\stackrel{~}{h}_{}/\stackrel{~}{h}_+`$ is (when $`t_f`$ is near $`t_{\mathrm{max}}`$) of order $`4\pi F_{\mathrm{max}}/\sqrt{\dot{F}_{\mathrm{max}}}=4\pi \sqrt{N_{\mathrm{max}}}`$ (this is consistent with the $`\epsilon `$ scaling of Eqs. (56) and (57), remembering that $`\epsilon 1/2\pi N`$). The contribution of $`\stackrel{~}{h}_+`$ is expected to be non-negligible only for signals which are really discontinuous in time. As the real signal (whatever be the subsequent plunge signal) will be continuous (and even smooth) it is clear that one should not add any contribution from $`\stackrel{~}{h}_+`$ when applying our above treatment to real signals. In fact, we shall see below that, even for discontinuous signals, the addition of $`h_+`$ has only a minute effect on overlaps.
### E Approximate Fourier transform when $`f>F_{\mathrm{max}}`$
Let us now consider the evaluation of $`\stackrel{~}{h}_{}(f)`$ in the case when the Fourier variable $`f`$ is larger than $`F_{\mathrm{max}}`$ \[but near $`F_{\mathrm{max}}`$, in the sense of Eq. (95)\]. In that case the integral Eq. (83) giving $`\stackrel{~}{h}_{}(f)`$ no longer has a saddle-point. However, it โnearlyโ has a saddle-point and therefore we expect that
$$\stackrel{~}{h}_{}(f)=_{\mathrm{}}^{t_{\mathrm{max}}}๐ta(t)e^{i\psi _f(t)},$$
(110)
will still dominate over $`\stackrel{~}{h}_+(f)`$. One could think of two ways of analytically approximating the integral Eq. (110). A first way is to still use the fact that (for Newtonian-like signals where the mathematical function $`F(t)`$ continues to exist and increase beyond $`t=t_{\mathrm{max}}`$) though there is no saddle-point in the domain of integration $`[\mathrm{},t_{\mathrm{max}}]`$, there exists a nearby saddle-point of the analytically continued phase function $`\psi _f(t)`$. More precisely, for Newtonian-like signals the mathematical equation $`F(t_f)=f`$ still defines a unique value $`t_f`$ (with $`t_f>t_{\mathrm{max}}`$ when $`f>F_{\mathrm{max}}`$). Capitalizing on the existence of this nearby saddle-point one can still try to insert the expansions Eqs. (73) and (74). This leads to a result of the form Eq. (109) with the correction factor Eq. (107), i.e. now considered for positive values of the argument $`\zeta _0(f)\sqrt{\pi \dot{F}(t_f)}(t_ft_{\mathrm{max}})`$. In other words, a simple uniform expansion to $`\stackrel{~}{h}_{}(f)`$ on both sides of $`fF_{\mathrm{max}}`$ would seem to be simply
$`\stackrel{~}{h}_{}^{\mathrm{cspa}}(f)`$ $`=`$ $`๐(\zeta _0(f)){\displaystyle \frac{a(t_f)}{\sqrt{\dot{F}(t_f)}}}e^{i\left[\psi _f(t_f)\pi /4\right]},`$ (112)
$`\mathrm{with}\zeta _0(f)`$ $`=`$ $`\sqrt{\pi \dot{F}(t_f)}(t_ft_{\mathrm{max}}).`$ (113)
Here โcspaโ means (zeroth order) corrected SPA. Note that when $`f>F_{\mathrm{max}}`$, $`t_f`$, and therefore all the quantities evaluated at $`t_f`$, are defined by using the (supposedly existing) analytic continuation of the mathematical function $`F(t)`$ beyond $`t=t_{\mathrm{max}}`$.
In our first attempts at improving the SPA in presense of a time-windowing we came up with the simple proposal Eqs. (112)-(113) and it gave excellent overlaps with the exact Fourier transform. However, we realised later that we could further improve on this simple proposal. We already stated that for $`f<F_{\mathrm{max}}`$ our best proposal is to modify the argument Eq. (113) into Eq. (108). In the case where $`f>F_{\mathrm{max}}`$ our best proposal is neither to use the straightforward argument Eq. (113), nor the โimproved-phasingโ argument Eq. (108) with a positive sign in front of the square-root \[which, however, still improves over the choice Eq. (113)\] but to follow a different tack which will turn out to be useful when considering the case of relativistic-like signals in the next Section.
To motivate our proposal in the case $`f>F_{\mathrm{max}}`$, let us remark that the integral to be approximated, i.e. Eq. (110) having no saddle-point in the domain of integration, is formally of the general type Eq. (88) with the phase $`\psi (t)=\psi _f(t)`$ being a monotonically increasing function of $`t`$. The important information we wish to deduce from Eqs. (89) and (91) is that there exists an expansion (valid when $`fF_{\mathrm{max}}`$, i.e. $`\zeta _0(f)`$ is large and positive) in which $`\stackrel{~}{h}_{}(f)`$ is entirely expressed in terms of the values of the functions $`\psi _f(t)`$ and $`a(t)`$, and their derivatives, evaluated at the edge point $`t=t_{\mathrm{max}}`$. This contrasts with the โcorrectedโ result Eqs. (112)โ(113) which relied on the existence of the functions $`\psi _f(t)`$ and $`a(t)`$ in the โunphysicalโ region $`t>t_{\mathrm{max}}`$. This motivates us to look for an approximation to Eq. (51) valid all over the domain $`\zeta _0(f)>0`$ \[and not only when $`\zeta _01`$ which will be seen to be the domain of validity of Eqs. (89) and (91)\] but expressed entirely in terms of the edge values of $`\psi _f(t)`$ and $`a(t)`$. We propose to define such an approximation by replacing the phase and amplitude in Eq. (110) by
$`\psi _f(t)`$ $``$ $`\psi _f(t_{\mathrm{max}})+2\pi (fF_{\mathrm{max}})(tt_{\mathrm{max}})\pi \dot{F}(t_{\mathrm{max}})(tt_{\mathrm{max}})^2,`$ (115)
$`a(t)`$ $``$ $`a(t_{\mathrm{max}}).`$ (116)
Thanks to the parabolic nature of the approximation Eq. (115) this again leads to an incomplete complex Gaussian integral (i.e. a Fresnel integral) which can be evaluated as before in terms of the complementary error function. Using Eq. (100), this leads to our final proposal for the (nearly) resonant part of $`\stackrel{~}{h}_{}(f)`$ for Newtonian-like signals
$`fF_{\mathrm{max}}:\stackrel{~}{h}_>^{\mathrm{inspa}}(f)`$ $`=`$ $`๐\left(\zeta _>(f)\right){\displaystyle \frac{a(t_{\mathrm{max}})}{\sqrt{\dot{F}(t_{\mathrm{max}})}}}\mathrm{exp}i\left[\psi _f(t_{\mathrm{max}})+{\displaystyle \frac{\pi (fF_{\mathrm{max}})^2}{\dot{F}(t_{\mathrm{max}})}}\pi /4\right],`$ (118)
$`\zeta _>(f)`$ $`=`$ $`{\displaystyle \frac{\sqrt{\pi }(fF_{\mathrm{max}})}{\sqrt{\dot{F}(t_{\mathrm{max}})}}}.`$ (119)
Note that, in the parabolic approximation where Eq. (73) or Eq. (115) hold, the function $`\zeta _>(f)`$ is approximately equal both to $`\zeta _0(f)`$, Eq. (113), and to the analytic continuation of Eq. (108), i.e. $`\zeta _<(f)\mathrm{sign}(fF_{\mathrm{max}})\sqrt{\psi _f(t_f)\psi _f(t_{\mathrm{max}})}`$. Note also that the phase factor in Eq. (118) (which is explicitly expressed in terms of edge quantities) is nearly equal to the analytic continuation of the usual SPA phase factor appearing in Eq. (109), i.e. $`\mathrm{exp}[i\psi _f(t_f)i\pi /4]`$. Finally, as required, the expressions, Eqs. (109) and (118) match continuously at $`f=F_{\mathrm{max}}`$ with common value
$$\stackrel{~}{h}_<^{\mathrm{inspa}}(F_{\mathrm{max}})=\stackrel{~}{h}_>^{\mathrm{inspa}}(F_{\mathrm{max}})=\frac{1}{2}\frac{a(t_{\mathrm{max}})}{\sqrt{\dot{F}(t_{\mathrm{max}})}}e^{i\left(\psi _f(t_{\mathrm{max}})\pi /4\right)}.$$
(120)
Summarizing: our best analytical estimate for the Fourier transform of discontinuous Newtonian-like signals is the sum
$$\stackrel{~}{h}^{\mathrm{intot}}(f)=\stackrel{~}{h}_{}^{\mathrm{inspa}}(f)+\stackrel{~}{h}_+^{\mathrm{edge}}(f),$$
(121)
where $`\stackrel{~}{h}_+^{\mathrm{edge}}(f)`$ is approximated by Eq. (94) and where $`\stackrel{~}{h}_{}^{\mathrm{inspa}}(f)`$ is given, when $`fF_{\mathrm{max}}`$ by Eq. (109) and for $`fF_{\mathrm{max}}`$ by Eq. (118). As stated earlier, we shall in fact recommend that the edge correction $`h_+^{\mathrm{edge}}`$ be not included when applying our result to real signals (we shall also see that it brings only a negligible improvement to the overlaps of time-windowed signals).
### F Comparison between the improved SPA, the usual SPA and the โexactโ SPA (numerical DFT)
Before proceeding to a quantitative comparison of the various approximants in Table II, it is important to remark that one needs to be more specific when using the terminology uSPA. One could compute the uSPA truncated at the $`F_{\mathrm{LSO}}`$ that we refer to as uSPAw ( where โwโ stands for โwindowedโ) or the uSPA truncated at the Nyquist frequency that we designate as uSPAn (where โnโ stands for โNyquistโ). In Table II we have listed the overlaps, as defined by Eq. (20), of a signal model generated in the time-domain and then Fourier transformed using a numerical DFT algorithm This defines for us the โexactโ Fourier representation of the signal, after due care has been taken to use a smooth time-window below $`f_s`$, and a high enough sampling rate. with the same signal model but directly generated in the frequency-domain using the usual SPA (uSPA), the corrected SPA (cSPA) and the improved Newtonian SPA (inSPA) discussed earlier. For simplicity, we consider only equal-mass systems $`(\eta =1/4)`$ and parametrize them by the total mass $`m=m_1+m_2`$. The total mass is the crucial parameter which measures the location of the $`F_{\mathrm{LSO}}`$ with respect to the bandwidth of the detector. The parameter $`\eta `$ is also important because it determines the number of cycles near the LSO \[Eq. (39) shows that $`N(F_{\mathrm{LSO}})`$ scales as $`1/\eta `$\]. The worst case (for the sensibility to the shutting off of the signal after the LSO) is $`\eta =\eta _{\mathrm{max}}=1/4`$, and this is why we focus on this case. \[We are also motivated by the fact that 1/4 being the maximum value of the function $`\eta (m_1,m_2)`$, the observed values of $`\eta `$, corresponding to a random sample of of $`m_1`$ and $`m_2`$, are expected to have an accumulation point at $`\eta _{\mathrm{max}}=1/4`$.\] As we have checked, if our filters exhibit good overlaps for $`\eta =1/4`$ they will have even better overlaps for $`\eta <1/4`$ and the same value for $`m`$.
The error function needed in computing $`๐(f)`$ is numerically computed using the NAG library S15DDF. The overlaps are shown for the usual SPA with a frequency-windowing (uSPAw) together with the overlaps for the uSPAn, cSPA and inSPA, computed up to $`F_{\mathrm{Nyquist}}`$. Table II shows that the improvements on the SPA that we propose in this paper (both the simple cspa and our final inspa) succeed very well in modelling the edge effects due to time-windowing. The overlaps in the case of cSPA/inSPA are better than 0.99 for equal-mass systems with total mass $`m<40M_{}`$. For a system of $`m=40M_{}`$ uSPAw gives an overlap of 0.8589 resulting in a loss in the number of events by 37%. Although the overlaps of cSPA and inSPA seem to be always about the same, we think that inSPA is a better representation of $`\stackrel{~}{h}(f)`$; it has better overlaps in the case of the most massive systems (see the first lines of Table II), and, as shown by Fig. 6, it captures better the decay of $`\stackrel{~}{h}(f)`$ beyond $`F_{\mathrm{max}}`$. In the Table for the usual SPA we have listed the overlaps for uSPAw i.e., uSPA terminated at $`f=F_{\mathrm{LSO}}=4400(m_{})^1`$ Hz and the overlap (uSPAn) up to the Nyquist frequency $`f_{\mathrm{Nyquist}}=2\mathrm{kHz}`$. As is very clear from these entries, windowing of the SPA improves the overlaps for massive systems very much. As remarked earlier, this is why in DIS, while comparing the DFT to the SPA, the uSPAw was used. On the other hand, computing overlaps up to the Nyquist frequency, i.e. uSPAn, produces much smaller overlaps.
To understand this further we plot in Fig. 6 the power per logarithmic bin of the squared SNR, $`d\rho ^2/d\mathrm{log}f=f|\stackrel{~}{h}(f)|^2/S_n(f)`$, which is the Fourier-domain quantity of most significance when discussing overlaps. We compare this quantity for various approximations to the Fourier-transform of an (arbitrarily-normalized) time-windowed signal: DFT, uSPA, cSPA and inSPA. In the important range of frequencies our best analytical approximant inSPA agrees with the exact result (FFT) quite well, the uSPA grossly overestimates and cSPA somewhat underestimates the actual signal power. This is why, though analytically continued up to $`f_{\mathrm{Nyquist}}`$, the uSPAn returns a smaller overlap as compared to the windowed SPA (uSPAw) because it overestimates the power in the signal beyond $`F_{\mathrm{LSO}}`$ .
In all the comparisons above, it is worth stressing that the FFT calculation is delicate: The โexactโ time-domain chirp contains an infinite number of cycles in the far past, with instantaneous frequencies tending to zero. As what happens to frequencies below the seismic cut-off $`f_\mathrm{s}=40`$ Hz is not physically important, we wish to simplify the numerical calculation of the FFT by essentially discarding the (infinite) part of the signal, having instantaneous frequencies $`F(t)<f_\mathrm{s}F(t_\mathrm{s})`$. We started doing that by simply time-windowing the signal for $`t<t_{\mathrm{min}}<t_\mathrm{s}`$ by a sharp, lower time-window $`\theta (tt_{\mathrm{min}})`$. However, this method introduces physically spurious oscillations (which are the lower-cutoff analogue of the physically important upper-cutoff oscillations) present in both $`\stackrel{~}{h}_+(f)`$ and $`\stackrel{~}{h}_{}(f)`$. One way to deal with this problem is to subtract out from the FFT these spurious edge oscillations by using the general formula Eq. (91), which in the present context, can be applied both to $`\stackrel{~}{h}_+^{\mathrm{FFT}}(f)`$ and $`\stackrel{~}{h}_{}^{\mathrm{FFT}}(f)`$. For instance, to lowest order the FFT corrected for these oscillations would read
$`\stackrel{~}{h}_{\mathrm{corrected}}^{\mathrm{FFT}}`$ $`=`$ $`\stackrel{~}{h}^{\mathrm{FFT}}+\mathrm{\Delta }_{\mathrm{min}}^++\mathrm{\Delta }_{\mathrm{min}}^{},`$ (123)
$`\mathrm{where},\mathrm{\Delta }_{\mathrm{min}}^\pm `$ $`=`$ $`{\displaystyle \frac{a(t_{\mathrm{min}})}{2\pi i(f\pm F_{\mathrm{min}})}}e^{i\psi _f^\pm (t_{\mathrm{min}})},`$ (124)
where $`F_{\mathrm{min}}=F(t_{\mathrm{min}})<f_\mathrm{s}`$ \[a better approximation can be derived from Eq. (91)\]. However, it seems better to use an alternative approach which does not require one to correct by hand the FFT. The alternative approach we have actually used in our calculations consists in imposing a smooth (rather than a sharp) lower time-window on the exact chirp, acting below $`t_\mathrm{s}`$ and in a smooth-enough manner that it does not introduce spurious edge-oscillations in the frequency-domain. The smooth time-window that we used consists in multiplying the chirp by the function
$$\sigma (t,t_1,t_2)=\frac{1}{e^z+1},z=\frac{t_2t_1}{tt_1}+\frac{t_2t_1}{tt_2},$$
(125)
which smoothly interpolates between 0 when $`t=t_1+0`$ and 1 when $`t=t_20`$. We used $`t_1`$ such that $`F_1=F(t_1)=30`$ Hz and $`t_2=t_\mathrm{s}`$ i.e. $`F(t_2)=f_\mathrm{s}`$. Moreover, we need to be careful with sampling and phase factors to correctly reproduce the edge correction to $`\stackrel{~}{h}_+^{\mathrm{edge}}`$.
In addition to the comparative evaluation of the various approximants, Table II also provides a numerical proof regarding the effect of $`h_+^{\mathrm{edge}}`$ on the overlaps. It is quite important to note that the inclusion of the non-resonant edge term $`h_+^{\mathrm{edge}}`$ has only a very minute (but positive) effect on overlaps. This is good news for our formal time-windowing ansatz, because we expect that this contribution will be (exponentially) negligible in the case of real (continuous) signals. We interpret the fact that even for our formal discontinuous model $`h_+^{\mathrm{edge}}`$ is negligible<sup>\**</sup><sup>\**</sup>\** Note that our statement here is only that $`h_+^{\mathrm{edge}}(f)`$ can be effectively omitted without significantly worsening the overlaps. We are not claiming that $`h_+^{\mathrm{edge}}(f)`$ is pointwise numerically negligible compared to $`h_{}(f)`$. Indeed, because the instantaneous number of cycles is rather small near the LSO, our analytical estimates above show that $`h_+^{\mathrm{edge}}(f)`$ is not very much smaller than $`h_{}(f)`$ near $`f=F_{\mathrm{LSO}}`$. as a confirmation that our improved SPA can adequately model not only signals that vanish after the LSO, but also signals that shut off rather quickly (on the $`F_{\mathrm{LSO}}^1`$ time-scale) after the LSO. It leads us also to propose, finally, to use as analytical representative of the FT of real signals the $`h_{}^{\mathrm{inspa}}`$ part of our formula above (without the edge term). \[To simplify the notation, we shall henceforth drop the extra subscript minus on $`h^{\mathrm{inspa}}`$.\]
In computing the above overlaps we have matched all the parameters of the two waveforms, including the time of arrival and the starting phase<sup>โ โ </sup><sup>โ โ </sup>โ โ The lag is set equal to zero in testing the accuracy of the Fourier representation but chosen optimally when testing faithfulness of a family of templates e.g. in Sec. V.. The overlaps in this Table as well as all other Tables in this paper are found to be insensitive to the sampling rate at the level of a fraction of a percent, provided that it is large enough to obey Shannonโs sampling theorem.
## IV Improved stationary phase approximation for relativistic signals in the adiabatic approximation: The SPP-approximants
Though one might a priori think that it is a simple matter to generalize the improved SPA discussed above for Newtonian-like signals to the relativistic case, it does not turn out to be so. What complicates matters is that there are serious qualitative, โnon-perturbativeโ differences between the two cases: first, the value of $`\dot{F}(t)`$ formally tends to $`+\mathrm{}`$ at the last stable orbit (LSO) which physically defines the upper-cutoff $`t_{\mathrm{max}}`$ of the inspiral signal, and, second, the mathematical function $`F(t)`$ does not admit a unique real analytic continuation beyond $`t_{\mathrm{max}}=t_{\mathrm{LSO}}`$. (These two facts are evidently related; indeed we shall see that $`F(t)`$ behaves in the non-analytic manner $`F(t)c_1+c_2(t_{\mathrm{LSO}}t)^{\frac{1}{2}}`$ when $`tt_{\mathrm{LSO}}^{}`$). Remembering the crucial role of a finite $`\dot{F}(t)`$ in the results Eqs. (109) and (118), it is clear that we need to tackle afresh the problem of finding a good, analytic approximation to $`\stackrel{~}{h}_{}(f)`$. Similarly, in view of the appearance of $`\dot{F}^{+1}(t_{\mathrm{max}})`$ in the next-to-leading contribution to $`\stackrel{~}{h}_+(f)`$ Eq. (94), we shall also need to revisit the calculation of $`h_+(f)`$ (though we shall, again, find that it makes only a negligible contribution to the overlaps.)
### A The phasing formula for relativistic signals in the adiabatic approximation
To extend the treatment of the previous Section and go beyond the Newtonian approximation, let us begin with the phasing formulas for gravitational waves from compact binaries written in a parametric form in terms of the variable $`v_F(\pi mF)^{1/3}`$ defined by the total mass $`m=m_1+m_2`$ and instantaneous gravitational wave frequency $`F`$
$$t(v_F)=t_{\mathrm{LSO}}+m_{v_F}^{v_{\mathrm{LSO}}}๐v\frac{E^{}(v)}{(v)},$$
(126)
$$\varphi (v_F)=\varphi _{\mathrm{LSO}}+2_{v_F}^{v_{\mathrm{LSO}}}๐vv^3\frac{E^{}(v)}{(v)},$$
(127)
where $`E(v)`$ is the dimensionless energy function related to the total relativistic energy or Bondi mass by $`E_{\mathrm{total}}=m(1+E)`$, $`(v)`$ the flux function denoting the gravitational wave luminosity of the system and $`t_{\mathrm{LSO}}`$ is the time and $`\varphi _{\mathrm{LSO}}`$ is the phase of the signal when $`v=v_{\mathrm{LSO}}`$. The parametric representaion Eqs. (126 (127 of the phasing formula $`\varphi =\varphi (t)`$ holds under the assumption of โadiabatic inspiralโ, i.e., that gravitational radiation damping can be treated as an adiabatic perturbation of a circular motion. See for a treatment of radiation damping going beyond this approximation.
In the restricted post-Newtonian approximation, one uses a Newtonian approximation for the amplitude . However, in order to extract an inspiral signal that may be buried in noisy data by the method of matched filtering, we need to employ post-Newtonian accurate representations for the two functions $`E^{}(v)`$ and $`(v)`$ that appear in the above phasing formulas. To any approximant $`E_A(v)`$, $`_A(v)`$, correspond \[by replacing $`E(v)E_A(v)`$, $`(v)_A(v)`$ in Eqs. (126) and (127)\] some approximate parametric representation $`t=t_A(v_F)`$, $`\varphi =\varphi _A(v_F)`$, and therefore a corresponding approximate time-domain template
$$h^A=h^A(t;๐,t_{\mathrm{LSO}},\varphi _{\mathrm{LSO}},m,\eta ),$$
(128)
obtained by replacing $`v_F`$, in the following $`v_F`$-parametric representation of the waveform
$$h^A(v_F)=๐v_F^2\mathrm{cos}\varphi _A(v_F),$$
(129)
by the function of time $`v_F=v_A(t)`$ obtained by inverting $`t=t_A(v_F)`$.
The standard approximants for $`E(v)`$ and $`(v)`$ are simply their successive Taylor approximants $`E_{T_n}`$ and $`_{T_n}`$ respectively. The DIS strategy for constructing new approximants to $`E(v)`$ and $`(v)`$ is two-pronged: Starting from the more basic energy-type and flux-type functions, $`e(v)`$ and $`l(v)`$ we construct Padรฉ-type approximants, say $`e_{P_n}`$, $`l_{P_n}`$, of the โbasicโ functions $`e(v)`$, $`l(v)`$<sup>โกโก</sup><sup>โกโก</sup>โกโกFor explicit formulas representing $`E(v)`$ and $`(v)`$ see Eqs. (3.8),(4.2) and (4.3) of DIS. The associated $`e(v)`$ and $`l(v)`$ functions are given by Eqs. (3.7), (3.9) and Eqs. (4.4)-(4.9) in DIS . See also Eqs. (3.5), (3.11) and Eqs. (3.18)-(3.23) there.. We then compute the required energy and flux functions entering the phasing formula. The successive approximants $`E[e_{P_n}]`$ and $`[e_{P_n},l_{P_n}]`$ have better convergence properties than their Taylor counterparts $`E_{T_n}[e_{T_n}]`$ and $`_{T_n}[e_{T_n},l_{T_n}]`$. In DIS we were working directly with the time-domain signal $`h(t)`$. As explained above this necessarily requires a numerical inversion of the parametric representation $`t=t(v_F)`$. By contrast, if one wants to compute the usual stationary phase approximation of $`h(t)=๐v_F^2(t)\mathrm{cos}\varphi (v_F(t))`$ there is no need to invert this parametric representation. Indeed, from Eq. (76), it is sufficient to know the instantaneous amplitude and the phase at the time $`t_f`$ where $`f=F(t_f)`$. This time is simply given by the same expression Eq. (126) as above with the replacement of $`v_F(\pi mF)^{1/3}`$ by $`v_f(\pi mf)^{1/3}`$, i.e. the stationary point $`t_f`$ is given by
$$t_f=t_{\mathrm{LSO}}+m_{v_f}^{v_{\mathrm{LSO}}}\frac{E^{}(v)}{(v)}๐v.$$
(130)
One then substitutes this value of $`t_f`$ in Eq. (127) to compute the phase $`\psi _f(t_f)2\pi ft_f\varphi (t_f)`$ of the Fourier component:
$$\psi _f(t_f)=2\pi ft_{\mathrm{LSO}}\varphi _{\mathrm{LSO}}+2_{v_f}^{v_{\mathrm{LSO}}}(v_f^3v^3)\frac{E^{}(v)}{(v)}๐v.$$
(131)
In terms of these quantities one has
$$\stackrel{~}{h}^{\mathrm{uspa}}(f)=\frac{1}{2}๐\frac{v_f^2}{\sqrt{\dot{F}(t_f)}}e^{i\left[\psi _f(t_f)\frac{\pi }{4}\right]}$$
(132)
The inclusion of relativistic effects in $`\stackrel{~}{h}^{\mathrm{uspa}}(f)`$ is then simply accomplished by using relativistic accurate expressions for $`E^{}(v)`$ and $`(v)`$ in the formulas giving $`\psi _f(t_f)`$ and $`\dot{F}(t_f)`$. The coefficient $`๐`$, in Eq. (129), determining the actual amplitude of the waveform reads:
$$๐(r,i,\theta ,\overline{\varphi },\overline{\psi })=(4\eta )\left(\frac{m}{d}\right)C(i,\theta ,\overline{\varphi },\overline{\psi }),$$
(133)
where $`d`$ is the distance to the source, and where
$`C(i,\theta ,\overline{\varphi },\overline{\psi })`$ $`=`$ $`\sqrt{A^2+B^2},`$ (135)
$`\mathrm{with},A`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left(1+\mathrm{cos}^2i\right)F_+;B=\mathrm{cos}iF_\times ,`$ (136)
with the beam-pattern factors
$`F_+(\theta ,\overline{\varphi },\overline{\psi })`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left(1+\mathrm{cos}^2\theta \right)\mathrm{cos}2\overline{\varphi }\mathrm{cos}2\overline{\psi }\mathrm{cos}\theta \mathrm{sin}2\overline{\varphi }\mathrm{sin}2\overline{\psi },`$ (138)
$`F_\times (\theta ,\overline{\varphi },\overline{\psi })`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left(1+\mathrm{cos}^2\theta \right)\mathrm{cos}2\overline{\varphi }\mathrm{sin}2\overline{\psi }+\mathrm{cos}\theta \mathrm{sin}2\overline{\varphi }\mathrm{cos}2\overline{\psi }.`$ (139)
In these formulas the angle $`i`$ denotes the inclination of the orbit with respect to the plane of the sky, and the angles $`\theta `$, $`\overline{\varphi }`$, and $`\overline{\psi }`$ parametrize both the propagation direction and the polarization of the gravitational wave with respect to the detector (see for exact definitions; we added a bar over $`\varphi `$ and $`\psi `$ to distinguish them from the GW phase $`\varphi `$ and Fourier phase $`\psi `$ respectively). Performing averages over the angles in the squared SNR leads to :
$$F_+^2_{\theta ,\overline{\varphi },\overline{\psi }}=F_\times ^2_{\theta ,\overline{\varphi },\overline{\psi }}=\frac{1}{5},$$
(141)
and finally
$$C^2_{i,\theta ,\overline{\varphi },\overline{\psi }}=\frac{4}{25}.$$
(142)
We are finally in a position to write down the rms and ideal SNRs. For a binary at a distance $`d`$ from the earth consisting of stars of individual masses $`m_1`$ and $`m_2`$ (total mass $`mm_1+m_2`$ and symmetric mass ratio $`\eta =m_1m_2/m^2)`$ the rms and ideal SNRs, obtained by using the rms and ideal values of $`C,`$ namely $`C=2/5`$ and $`C=1,`$ respectively, when replacing Eq. (132) in Eq. (32), or equivalently, when replacing $`a(f)=(1/2)๐v^2(f)=2\eta md^1Cv^2(f)`$ in Eq. (34) (with Eq. (38) and a truncation at $`F_{\mathrm{LSO}}`$), are given by
$$\rho _{\mathrm{rms}}=\frac{m^{5/6}}{d\pi ^{2/3}}\left(\frac{\eta }{15}\right)^{1/2}\left[_0^{F_{\mathrm{LSO}}}๐f\frac{f^{7/3}}{S_n(f)}\right]^{1/2},\rho _{\mathrm{ideal}}=\frac{5}{2}\rho _{\mathrm{rms}}.$$
(143)
Note that the SNR depends only on the combination $`=m\eta ^{3/5}`$ โ the chirp mass (see e.g. ), and that the first Eq. (143) is equivalent to Eq. (12).
Let us next delineate the qualitative differences between the relativistic and non-relativistic cases by considering the function appearing as denominator in the uSPA, Eq. (132)
$$\dot{F}(t)=\frac{1}{2\pi }\frac{d^2\varphi }{dt^2}=\frac{3v^2}{\pi m^2}\frac{(v)}{E^{}(v)}.$$
(144)
At the LSO, the gravitational wave flux $`(v)`$ is finite (it blows up only later, when reaching the light ring ) while, by definition, $`E^{}(v)`$ vanishes linearly, $`E^{}(v)vv_{\mathrm{LSO}}`$. As we shall see below this means that $`\dot{F}(t)`$ blows up as $`(t_{\mathrm{LSO}}t)^{1/2}`$. A consequence of this blow up is that the last two terms in Eq. (78) blow up like $`(t_{\mathrm{LSO}}t)^{3/2}`$ confirming the need for a special treatment of the Fourier transform near the LSO. We are here speaking of the exact behaviour of the functions $`E(v)`$ and $`(v)`$, as supposedly known from combining the test-mass limit results with the best available results on the physics underlying the existence of the LSO , and the emission of gravitational waves in comparable mass systems . In DIS, we have incorporated this information so that all the P-approximants $`E_{P_n}E(e_{P_n}),_{P_n}[e_{P_n},l_{P_n}]`$ that we define share, with the โexactโ functions $`E`$ and $``$ the crucial properties mentioned above (i.e. finite $`(v_{\mathrm{LSO}})`$ and $`E^{}(v)vv_{\mathrm{LSO}}`$). The (less-convergent) successive T-approximants $`E_{T_n}`$ and $`_{T_n}`$ do not incorporate this information exactly, and only as $`n`$ increases they tend to incorporate it. In our opinion the $`T_n`$ approximants disqualify as โrelativisticโ approximants since they do not consistently incorporate the expectation (based on several different methods; see references in ) that the frequency at the LSO is (for any $`\eta 1/4`$) numerically near the Schwarzschild-like prediction, Eq. (26). Indeed, if we define the 2PN Taylor estimate of $`F_{\mathrm{LSO}}`$ by the value of $`v=(\pi mF)^{1/3}`$ where the straightforward Taylor approximant $`E_{T_4}(v)=_{k=0}^4E_k(\eta )v^k`$ reaches a minimum, we find, e.g. that (i) when $`m=40M_{}`$ and $`\eta =0`$, $`F_{T_4}=200`$ Hz, very different from the exact value of 110 Hz, and (ii) when $`m=40M_{}`$ and $`\eta =1/4`$, that $`F_{\mathrm{LSO}}^{T_4}=221.4`$ Hz, very different from the other predictions $`F_{\mathrm{LSO}}^{P_4}=143`$ Hz and $`F_{\mathrm{LSO}}^{\mathrm{Ref}.\text{[40]}}=118.6`$ Hz. We compare and contrast in Fig. 7 the Newtonian and relativistic behaviours of the wave amplitude and instantaneous frequency $`F(t)`$ during the last couple of orbits before the LSO. The blow up of $`\dot{F}(t)`$, i.e. the fact that the slope of $`F(t)`$ becomes vertical is an effect which is localized in the last part of the last cycle before the LSO. Note also in Fig. 7 that a less localized consequence of this blow up is that the average frequency a few cycles before the LSO is smaller (for a given $`F_{\mathrm{LSO}}`$) in the relativistic case, than in the (unphysical) Newtonian one. Note that the physical origin of the blow up of $`\dot{F}`$ is that, just before the LSO the โeffective potentialโ for the radial motion becomes very flat (before having an inflection point at the LSO). In picturesque terms, the radial motion becomes โgroundlessโ at the LSO. Evidently, the blow up of $`\dot{F}`$ is due to our use of the โadiabaticโ approximation down to the LSO. In reality, radiation reaction will cause a progressive transition between the inspiral and plunge which will modify the evolution of $`F(t)`$ in the last cycle before the LSO. We shall discuss this issue in detail in a forthcoming paper and subsequently its data analysis consequences.
### B Edge contribution to the non-resonant relativistic $`\stackrel{~}{h}_+(f)`$
As in the Newtonian-like case we decompose $`\stackrel{~}{h}(f)`$ in two contributions, Eqs. (83) and (84). The non-resonant contribution $`\stackrel{~}{h}_+(f)`$ will be dominated by the โedgeโ contribution to an integral of our usual type Eq. (50). Though the problem is similar to the one we have generically solved in Section III D we cannot apply the results Eqs. (90), (91), (93), (94), because of the limiting hypothesis (iii) mentioned in our introductory discussion Section II D. Indeed, the problem is that, in the (physically relevant) case of relativistic signals the functions $`a(t)`$ and $`\psi (t)`$ are not smooth at the upper edge $`t=t_{\mathrm{LSO}}`$. Let us see explicitly in what way they violate smoothness there. Let us first define,
$$e_1(\eta )\left[\frac{d}{dv}\left(\frac{E^{}(v)}{(v)}\right)\right]_{v_{\mathrm{LSO}}},$$
(145)
so that near the LSO we may write:
$$\frac{E^{}(v)}{(v)}=e_1(vv_{\mathrm{LSO}})+๐ช[(vv_{\mathrm{LSO}})^2].$$
(146)
If we were to use the test-mass approximation for the energy function $`E^{}(v)`$ and the Newtonian (quadrupole) one for the flux function $`(v)`$ this would give
$$e_1^{P_0(\mathrm{tm})}(\eta )\frac{15}{2}\frac{1}{4\eta }\frac{1}{v_{\mathrm{LSO}}^8(13v_{\mathrm{LSO}}^2)^{\frac{3}{2}}}\frac{27492}{4\eta }.$$
(147)
We have numerically estimated the function $`\overline{e}_1^{P4}(\eta )4\eta e_1(\eta )`$, when using the $`P_4`$-approximant of in the definition of Eq. (145). We find that to a good approximation
$$4\eta e_1^{P_4}(\eta )\overline{e}_1^{P4}(\eta )26091.61194\mathrm{exp}(4.474405683\eta ).$$
(148)
In terms of
$$\tau \frac{t_{\mathrm{LSO}}t}{m},\tau 0\mathrm{for}vv_{\mathrm{LSO}},$$
(149)
and using
$$\tau =_v^{v_{\mathrm{LSO}}}๐v\frac{E^{}(v)}{(v)}=\frac{1}{2}e_1(vv_{\mathrm{LSO}})^2+๐ช[(vv_{\mathrm{LSO}})^3],$$
(150)
and Eq. (127) for $`\varphi (v)`$, we find the following approximate representation (valid near the LSO) for the phase $`\varphi (t)`$:
$`tt_{\mathrm{LSO}}`$ $`=`$ $`m\tau ,`$ (152)
$`\varphi (t)\varphi (t_{\mathrm{LSO}})`$ $``$ $`2v_{\mathrm{LSO}}^3\tau +{\displaystyle \frac{4\sqrt{2}}{\sqrt{e_1}}}v_{\mathrm{LSO}}^2\tau ^{3/2}.`$ (153)
Note also that Eq. (150) gives the following representation for $`v(t)`$, and therefore for the amplitude $`a(t)=๐v^2(t)`$
$`v`$ $``$ $`v_{\mathrm{LSO}}{\displaystyle \frac{\sqrt{2}}{\sqrt{e_1}}}\tau ^{\frac{1}{2}},`$ (155)
$`a(t)`$ $``$ $`a_{\mathrm{LSO}}\left[1{\displaystyle \frac{2\sqrt{2}}{\sqrt{e_1}}}{\displaystyle \frac{1}{v_{\mathrm{LSO}}}}\tau ^{\frac{1}{2}}\right].`$ (156)
We are interested in evaluating the edge contribution to the integral
$$I=\stackrel{~}{h}_+(f)=_{\mathrm{}}^{t_{\mathrm{LSO}}}๐ta(t)e^{i\psi _f^+}(t)=m_0^+\mathrm{}๐\tau a(\tau )e^{i\psi _f^+(\tau )}.$$
(157)
Near the LSO boundary i.e. near the edge $`\tau =0`$ in the $`\tau `$-form of the integral, the amplitude behaves as Eq. (156) while the appropriate phase $`\psi _f^+(\tau )`$ behaves, from Eqs. (152) and (153) as
$$\psi _f^+(\tau )\psi _{f\mathrm{LSO}}^+2\pi m(F_{\mathrm{LSO}}+f)\tau +\frac{4\sqrt{2}}{\sqrt{e_1}}v_{\mathrm{LSO}}^2\tau ^{3/2},$$
(158)
where
$$\psi _{f\mathrm{LSO}}^+\psi _f^+(t_{\mathrm{LSO}})=2\pi ft_{\mathrm{LSO}}+\varphi _{\mathrm{LSO}}.$$
(159)
The appearance of fractional powers of $`\tau `$ in the expansions Eqs. (155) and (156) show explicitly the violation of the $`๐^{\mathrm{}}`$ property of $`a(t)`$ and $`\psi (t)`$ at the edge. We cannot use the integration-by-parts method to evaluate the expansion of $`I_{\mathrm{edge}}`$. However, we can still use the general method sketched in Section II D. Without rotating explicitly the the $`\tau `$-contour in the complex plane the edge contribution to $`I`$ is obtained by inserting the expansions Eqs. (156) and (158) in Eq. (157) and expanding everything out, except for the main phase, $`\psi _{f\mathrm{LSO}}^+2\pi m(F_{\mathrm{LSO}}+f)\tau `$ which must be kept in the exponent. This yields
$`I_{\mathrm{edge}}`$ $`=`$ $`ma_{\mathrm{LSO}}e^{i\psi _{f\mathrm{LSO}}^+}{\displaystyle _0^{\mathrm{}}}๐\tau e^{iy\tau }\left(1{\displaystyle \frac{2\sqrt{2}}{\sqrt{e_1}}}{\displaystyle \frac{1}{v_{\mathrm{LSO}}}}\tau ^{\frac{1}{2}}+{\displaystyle \frac{i4\sqrt{2}}{\sqrt{e_1}}}v_{\mathrm{LSO}}^2\tau ^{\frac{3}{2}}\right),`$ (161)
$`\mathrm{where},y`$ $``$ $`2\pi m\left(F_{\mathrm{LSO}}+f\right).`$ (162)
Note that, instead of rotating $`\tau `$ in the complex plane, we can (equivalently) consider that $`y`$ possesses a small negative imaginary contribution: $`yyi0`$. The integrals appearing in Eq. (162) are evaluated by the general formula
$$i_\alpha =_0^{\mathrm{}}๐\tau e^{iy\tau }\tau ^\alpha =\frac{e^{i\frac{\pi }{2}(\alpha +1)}}{y^{\alpha +1}}\mathrm{\Gamma }(\alpha +1).$$
(163)
This yields finally
$$\stackrel{~}{h}_+^{\mathrm{edge}}(f)\frac{ma_{\mathrm{LSO}}e^{i[\psi _{f\mathrm{LSO}}^+]}}{iy}\left[1+\left(\frac{3}{2}\frac{F_{\mathrm{LSO}}}{F_{\mathrm{LSO}}+f}1\right)e^{i\pi /4}\frac{\sqrt{2\pi }}{\sqrt{e_1}}\frac{1}{v_{LSO}\sqrt{y}}\right].$$
(164)
The leading contribution $`((iy)^1)`$ to the relativistic result Eq. (164) agrees with the leading contribution in Eq. (94). Note that the next-to-leading contribution does not have the same dependence on $`f+F_{\mathrm{LSO}}`$ as the corresponding term in the non-relativistic result in Eq. (94). In spite of the breakdown of the formal expansion Eq. (94) the fractional correction given by the last term in the bracket of Eq. (164) is checked to be numerically small. This check was the main motivation for us to compute $`\stackrel{~}{h}_+^{\mathrm{edge}}`$ to next-to-leading order in the relativistic case. The lesson is that the formal blow up of $`\dot{F}`$ near the LSO has only a small numerical effect on $`\stackrel{~}{h}_+^{\mathrm{edge}}`$. This is again a confirmation that our results are robust under a refinement of our knowledge of the signal. We shall further check below that, as in the Newtonian case, $`\stackrel{~}{h}_+^{\mathrm{edge}}`$ has only a negligible effect on overlaps.
### C Improved stationary phase approximation for relativistic signals
Let us now consider the resonant contribution $`\stackrel{~}{h}_{}(f)`$, considered in the crucial domain where the stationary point is near the edge $`t_{\mathrm{LSO}}`$. As before also, the optimal approximants to $`\stackrel{~}{h}_{}(f)`$ that we can construct are given by different analytical expressions according to the value of $`f`$. However, we need now to introduce a new definition of the two ranges of frequencies in which one must (minimally) divide the $`f`$-axis. More precisely, we introduce a frequency $`f_{\mathrm{up}}`$, near but below $`F_{\mathrm{max}}`$, and we shall construct a โlowerโ approximation $`\stackrel{~}{h}_<(f)`$ in the range $`f<f_{\mathrm{up}}`$, and an โupperโ one $`\stackrel{~}{h}_>(f)`$ in the range $`f>f_{\mathrm{up}}`$ (which includes $`f=F_{\mathrm{max}}`$). The optimal value of $`f_{\mathrm{up}}`$ will be determined below.
In the lower range, $`f<f_{\mathrm{up}}`$, we can draw on the work of Sec. III D. Indeed, in that range there exists a saddle-point in the domain of integration. However, as that saddle-point can become rather near $`t_{\mathrm{max}}`$ (because $`f_{\mathrm{up}}`$ is near $`F_{\mathrm{max}}`$), we can significantly improve the usual SPA estimate by using our previous result, i.e. by defining
$`ff_{\mathrm{up}}:\stackrel{~}{h}_<^{\mathrm{irspa}}(f)`$ $`=`$ $`๐\left(\zeta _<(f)\right){\displaystyle \frac{a(t_f)}{\sqrt{\dot{F}(t_f)}}}e^{i\left[\psi _f(t_f)\pi /4\right]},`$ (166)
$`\zeta _<(f)`$ $`=`$ $`\sqrt{\psi _f(t_f)\psi _f(t_{\mathrm{max}})}.`$ (167)
The label โirspaโ in Eq. (166) stands for improved relativistic SPA.
Let us finally explore the optimal analytic approximation to $`\stackrel{~}{h}_{}(f)`$ in the upper range $`ff_{\mathrm{up}}`$. Proceeding as in Section IV B in this case one has
$$\psi _f^{}(t)\psi _{f\mathrm{LSO}}^{}+2\pi m(F_{\mathrm{LSO}}f)\tau \frac{4\sqrt{2}}{\sqrt{e_1}}v_{\mathrm{LSO}}^2\tau ^{3/2},$$
(168)
where
$$\psi _{f\mathrm{LSO}}^{}\psi _f^{}(t_{\mathrm{LSO}})=2\pi ft_{\mathrm{LSO}}\varphi _{\mathrm{LSO}}.$$
(169)
We shall use this expansion (which replaces the parabolic approximation Eq. (73) used in the Newtonian case) to evaluate the Fourier integral Eq. (110). To this end we must introduce a new special function (characteristic of the relativistic phasing near the LSO) to replace the error function. Let us define the function
$$g_{\frac{3}{2}}(x)_0^{\mathrm{}}๐\widehat{\tau }e^{i\left(3x\widehat{\tau }2\widehat{\tau }^{3/2}\right)},$$
(170)
where the new variable $`\widehat{\tau }`$ is related to $`\tau `$ by $`\tau =\alpha \widehat{\tau }`$ where
$$\alpha =\frac{1}{2}v_{\mathrm{LSO}}^{4/3}e_1^{1/3}.$$
(171)
In the test mass case corresponding to Eq. (147) the value of $`\alpha `$ is $`49.83/(4\eta )^{1/3}`$. The value defined by the $`P_4`$ approximant on the other hand is given by combining Eqs. (27) and (148). In particular, $`\alpha `$ equals 30.055 (54.578) for $`\eta =`$ 0.25 (0.1) respectively. The index $`\frac{3}{2}`$ in $`g_{\frac{3}{2}}(x)`$ alludes to the power $`\widehat{\tau }^{3/2}`$ replacing the power $`\widehat{\tau }^2`$ in the usual error function, and where the conventional coefficients $`3`$ and $`2`$ have been chosen to simplify some formulas (although they complicate others!). The final result is conveniently written in terms of a variable $`x`$ given by
$$x=\frac{2\pi }{3}\alpha m(F_{\mathrm{LSO}}f).$$
(172)
This improved relativistic SPA is thus written as
$$ff_{\mathrm{up}}:\stackrel{~}{h}_>^{\mathrm{irspa}}(f)=m\alpha e^{i\psi _f^{\mathrm{LSO}}}a(t_{\mathrm{LSO}})g_{\frac{3}{2}}(x).$$
(173)
Note that $`f<F_{\mathrm{LSO}}`$ corresponds to $`x>0`$ (saddle-point domain), while $`f>F_{\mathrm{LSO}}`$ corresponds to $`x<0`$ (absence of a saddle-point). Roughly speaking the variable $`x(f)`$ corresponds to $`\zeta (f)`$ of the non-relativistic case, and $`g_{\frac{3}{2}}(x)`$ is the relativistic analogue of the combination $`๐(\zeta )e^{i\zeta ^2}`$ appearing in the previous treatment (see, e.g., Eq. (118)).
It is useful to summarise some properties of the function $`g_{\frac{3}{2}}(x)`$:
$`g_{\frac{3}{2}}(0)`$ $`=`$ $`{\displaystyle \frac{1}{3}}{\displaystyle \frac{(1i\sqrt{3})}{4^{1/3}}}\mathrm{\Gamma }\left({\displaystyle \frac{2}{3}}\right)=0.2843470.492503i,`$ (175)
$`g_{\frac{3}{2}}(x)`$ $``$ $`\sqrt{{\displaystyle \frac{4\pi x}{3}}}e^{i(x^3\pi /4)},x>0,x1,`$ (176)
$`g_{\frac{3}{2}}(x)`$ $``$ $`{\displaystyle \frac{i}{3x}};x<0,x1.`$ (177)
By expanding the integrand of $`g_{\frac{3}{2}}(x)`$ in powers of $`x`$, and integrating term by term \[using the properties of the Euler $`\mathrm{\Gamma }`$-integral after having changed the variable of integration: $`\widehat{\tau }=e^{\frac{i\pi }{3}}(u/2)^{\frac{2}{3}}`$\], one proves that $`g_{\frac{3}{2}}(x)`$ is given by the following, everywhere convergent, Taylor-Maclaurin expansion:
$$g_{\frac{3}{2}}(x)=\frac{2^{1/3}}{3}e^{i\pi /3}\underset{n=0}{\overset{n=\mathrm{}}{}}\frac{\mathrm{\Gamma }[\frac{2}{3}(n+1)]}{n!}\left(\frac{3x}{2^{2/3}}e^{\frac{i\pi }{6}}\right)^n.$$
(178)
With about 300 terms the above series represents $`g_{\frac{3}{2}}(x)`$ accurately enough for values of $`x`$ in the range $`x[2.3,\mathrm{\hspace{0.17em}2.3}]`$. We used this series to generate the plot of $`g_{\frac{3}{2}}(x)`$ represented in Fig. 8. Though we do not use it in this paper, note that for $`x\mathrm{}`$, the following (divergent) asymptotic expansion is also valid:
$$g_{\frac{3}{2}}(x)\frac{1}{3x}\underset{n=0}{\overset{n=\mathrm{}}{}}\frac{\mathrm{\Gamma }(\frac{3}{2}n+1)}{n!}\left[\frac{2}{(3x)^{3/2}}\right]^ne^{\frac{i\pi }{4}(n+2)}.$$
(179)
In all our calculations of overlaps we shall define the frequency $`f_{\mathrm{up}}`$ separating the lower range from the upper range by choosing $`x_{\mathrm{up}}=0.36`$ as the right hand side of Eq. (172). This value is chosen so that at $`x_{\mathrm{up}}`$ one has a smooth transition from the lower to the upper approximation. We have also checked that the overlaps do not change very significantly for $`x_{\mathrm{up}}`$ between 0.2 and 0.4.
In summary our best analytic representation of time-windowed relativistic signals in the Fourier-domain would be defined by combining the P-approximant construction of the functions $`E^{}(v)`$, $`(v)`$ with the total improved relativistic approximants (irtot) defined as
$$\stackrel{~}{h}^{\mathrm{irtot}}(f)=\stackrel{~}{h}_{}^{\mathrm{irspa}}+\stackrel{~}{h}_+^{\mathrm{edge}},$$
(180)
where $`\stackrel{~}{h}_+^{\mathrm{edge}}`$ is defined in Eq. (164) and $`\stackrel{~}{h}_{}^{\mathrm{irspa}}`$ is defined for $`ff_{\mathrm{up}}`$ by Eqs. (166), (167), and, for $`ff_{\mathrm{up}}`$ by Eq. (173). Actually, as in the case of Newtonian signals, we have found that the inclusion of $`\stackrel{~}{h}_+^{\mathrm{edge}}`$ has only a minimal (though favorable) effect on overlaps. Moreover, such a contribution is absent in the case of real signals. Therefore, our final practical and best proposal consists in using only $`\stackrel{~}{h}_{}^{\mathrm{irspa}}`$ (For simplicity we henceforth drop the subscript minus). We shall henceforth refer to the improved frequency domain stationary phase P-approximants based on the irSPA as the SPP approximants.
### D Comparison between the usual SPA, the improved relativistic SPA and the โexactโ SPA (numerical DFT)
In this Section, we test the accuracy of our analytical approximations in various ways. Fig. 9 compares an inspiral wave from a (20,20) $`M_{}`$ binary generated by three different methods: (i) directly in the time-domain and terminated when the instantaneous gravitational wave frequency reaches the value at the LSO (solid line), (ii) in the Fourier domain using the usual SPA but with a square window between $`f_{\mathrm{min}}`$=40 Hz, $`f_{\mathrm{max}}=F_{\mathrm{LSO}}`$ (uSPAw) and then inverse Fourier transformed (dashed line) and (iii) again in the Fourier domain but using irSPA with Fourier components computed up to Nyquist frequency and then inverse Fourier transformed to obtain its time-domain representation (dotted line). We only exhibit the comparison near the crucial LSO region \[Much before the LSO the uSPA is nearly equivalent to the irSPA and they both do a good job in representing the actual signal\]. We observe that the uSPAw begins to get out of phase with the wave directly generated in the time-domain during the last cycle and rings a few times beyond the shut-off point. Our new proposal, irSPA, keeps in phase with the time-domain signal until the last moment although it too has a couple of low amplitude cycles beyond the LSO.
Matched filtering involves not just the correlation of two signals but rather their weighted correlation โ the weight coming from the detector spectral noise density. To further compare and contrast our new $`f`$-domain approximants to the usual frequency-windowed SPA it is conceptually useful to compare various approximations in the โwhitened-time-domainโ introduced in Section II A above. As discussed above, in this picture (and only in this picture) the optimal filter consists of correlating the output of the detector with an exact copy of the expected signal. The whitened \[i.e., convolved with the whitening kernel $`w_{\frac{1}{2}}`$ Eq. (22)\] signals are plotted and compared in Fig. 10 which is the same as Fig. 9 except that all the waves here are whitened (i.e. divided by $`\sqrt{S_n(f)}`$ and then inverse Fourier transformed). The inset in Fig. 10 shows the full whitened signal that was originally generated in the time domain. Several observations are in order. First, we see how low frequency components are suppressed relative to high frequency components which occur in a more sensitive band of the detector. Second, we can very clearly see the non-local behaviour of the whitening kernel. It has the effect of softening the window imposed on the wave that was directly generated in the time-domain and curbing the oscillations in the irSPA beyond $`F_{\mathrm{LSO}}.`$ Finally, this same whitening is seen to have worsened the mismatch of uSPAw with the whitened version of the original time-truncated signal. The conclusions drawn from these visual comparisons are borne out by detailed numerical experiments we performed.
To compare the approximants more quantitatively, in Table III we list the overlaps of the exact Fourier representation of a model waveform (i.e., a signal generated in the time-domain and then Fourier transformed using a DFT algorithm) with their approximate Fourier representations analytically computed using one of the following: the frequency-windowed usual SPA \[i.e., uSPAw, cf.Eq. (76)\], the improved Newtonian SPA \[inSPA is the same as irSPAw, cf. Eq. (167)\] and the improved relativistic SPA \[cf. Eq. (180)\]. The uSPA and the inSPA used in computing these overlaps are terminated at $`f=F_{\mathrm{LSO}}^\mathrm{A},`$ where $`F_{\mathrm{LSO}}^\mathrm{A}`$ is the last stable orbit frequency determined by the condition $`E_\mathrm{A}^{}(v)=0`$ (hence the labels SPAw and inSPAw where โwโ stands for โwindowedโ โ in the frequency-domain). This is because both uSPA and inSPA vanish at the LSO (due to the factor $`1/\sqrt{\dot{F}_{\mathrm{LSO}}}`$) and are either not defined (in the case of the usual SPA) or formally vanishing (according to the definition Eq. (118) in the case of inSPA) beyond the LSO. Contrast this with the Newtonian case where it is possible to analytically extend the usual SPA beyond $`F_{\mathrm{LSO}}`$.
It is generally true, as stated in DIS, that the stationary phase approximation to the Fourier transform worsens very significantly as we consider more massive binaries. In this sense the uSPA poorly represents the exact chirp. We conclude that, for massive systems with total mass $`m=m_1+m_240M_{}`$ the only uniformly acceptable analytic representation of the Fourier transform is the irSPA.
## V Faithfulness and Effectualness of SPP approximants
So far we have concentrated on developing an accurate Fourier representation of the inspiral waveform at various levels of approximation from Newtonian to P-approximants. In order to quantify the accuracy, we used the overlap of the DFT of the waveform computed using a FFT of the time-domain signal with an analytical approximation of the Fourier transform of the same time-domain signal using the improved SPA suggested in Sec. III and IV. However, an important question still remains: What is the total loss of accuracy due to combining the loss of precision entailed by the use of an analytical approximant to the FT (loss that we have shown how to minimize by defining the irspa) with the loss of accuracy<sup>\**</sup><sup>\**</sup>\** We distinguish precision and accuracy in the same way that they are distinguished in Metrology. entailed by the use of some finite-order in the post-Newtonian approximation of the exact signal. In other words, how accurate is the approximate frequency-domain representation of a post-Newtonian approximant in modelling the exact FT of the exact general relativistic signal? More precisely, what fraction of the SNR of a true signal is the Fourier-domain approximant likely to extract? Additionally, one is also interested in knowing the biases induced in the estimation of parameters when using the frequency-domain approximants introduced in this work.
We shall follow DIS in saying that a representation of a signal is faithful if it has a good overlap<sup>\*โ </sup><sup>\*โ </sup>\*โ When discussing faithfulness and effectualness we always assume, as in DIS, Eq 2.17 there, that the overlap Eq. (20) is first maximized with respect to the relative time lag (and relative phase). with the exact signal for the same values of the (dynamical) parameters (or more precisely, if the overlap is maximized for template parameters which have acceptably small biases with respect to the exact signal parameters). As in DIS, we employ as necessary criterion for faithfulness the requirement that the โdiagonalโ ambiguity function be larger than 0.965. On the other hand, we shall say that a representation of a signal is effectual if the overlap, maximized over the template parameters is very near one. To use these definitions we follow DIS in introducing a fiducial exact general relativistic signal. In Table IV and Table VI we use as fiducial exact signal the formal โtest-mass caseโ for which the function $`E(v)`$ is known analytically and $`(v)`$ numerically . In Table V we use as fiducial exact signal the one defined in DIS for comparable masses \[see Eq. (4.11) there for the definition of the exact new energy function, and Eqs. (7.1) (7.2) for the exact factored flux function; we took the value $`\kappa _0=47/39`$ for the parameter defining formal higher PN-effects in Eq. (4.11) \]. As above we consider that the exact time-domain signal is shut off after the LSO. \[For each considered waveform, defined by some approximate energy and flux functions $`E_A(v)`$ and $`_A(v)`$, we shut it off at the LSO defined by the corresponding energy function $`E_A(v)`$.\]
In Tables IV and V we list the overlaps for different approximants for the three โmassiveโ archetypal binaries \[$`(1.4M_{},10M_{}),(10M_{},10M_{}),`$ and $`(20M_{},20M_{})`$\] that could be searched for in GEO/LIGO/VIRGO data. These overlaps are computed using the expected LIGO noise Eq. (5) by maximising over the lag parameter $`\tau `$ and phase $`\varphi _c`$ but without re-adjusting the intrinsic parameters i.e., the masses of the two stars in the approximants, to maximise the overlap. \[This implies that the value of $`F_{\mathrm{LSO}}`$ used in the approximant is different from that in the โexactโ signal.\] The overlaps are therefore a reflection of how accurate the various representations are in an absolute sense. In other words, they compare the faithfulness of the different approximants. Two independent aspects of approximation are investigated in these Tables. Firstly, the comparison between the two alternatives in the frequency domain: the usual SPA (uspaw) and our improved relativistic SPA (irspa). And secondly, the post-Newtonian order to which the phasing is computed. To investigate further the performance of these approximants we summarise in Table VI the overlaps obtained by maximising over all the parameters in the approximants including the intrinsic ones. Thus in addition to maximising over the lag parameter $`\tau `$ and the phase $`\varphi _c`$ one also extremises over the masses of the two stars $`m_1`$ and $`m_2`$. In other words, we compare the effectualness of the various approximants. We also compute the bias introduced in the total mass $`m`$.
From Tables IVVI one can conclude the following: (i) The improved relativistic SPA is significantly more faithful and more effectual for massive systems with total mass $`m20M_{}`$, and mandatory for $`m26M_{}`$, (ii) Comparing with DIS, we see that the frequency-domain irspa does as well as the time-domain waveform even for massive binaries up to $`40M_{}`$; (iii) The 2.5PN SPP approximant is both a faithful and an effectual approximant for a wide range of binary systems $`(m40M_{})`$. It only introduces a small bias. Note also that, in regard to effectualness, the gain in going from 2PN to 2.5 PN accuracy is quite significant (mainly in decreasing the biases) and especially for low-mass systems (which have many useful cycles), while the gain in going from 2.5PN to 3PN seems very slight.
To summarise: If one would like to lose no more than a tenth of the events that would be observable had one known the exact general-relativistic signal, then the 2.5PN SPP-approximants are a must. Furthermore, unbiased parameter estimation requires 2.5PN SPP-approximants in all cases.
## VI Why are Time Domain Relativistic Signals more expensive to compute?
The main purpose of this work is to provide a set of tools to the experimenters so that they can generate templates with a minimal computational cost. We next, therefore, address the issue of computational costs of various algorithms for template generation.
First, though the signal is initially given in the time-domain, the time-domain version of the Wiener filter contains a double time integration \[see second form of Eq. (14)\] which is (given the existence of FFT algorithms) much more computationally expensive than the single frequency-domain version of the Wiener filter \[see first form of Eq. (14)\]. Therefore, in the computation of the correlation of a template with the detector output what is required is the Fourier transform of the matched filter. However, the DIS proposal was to compute the templates in the time-domain and compute their exact DFT using FFT algorithms. Admissibly, this procedure is still highly computation-intensive. Let us reason out why this is so.
To compute the time-domain signal we need a phasing formula $`\varphi =\varphi (t)`$. Since there is no explicit expression for the phasing of inspiral waves as a function of time the standard approach is to use the implicit formula, Eqs. (126)-(127). The binding energy $`E(v)`$ and the gravitational wave flux $`(v)`$ have been computed, e.g. using Padรฉ techniques, as explicit functions of $`v`$ and these when used in Eq. (126) and (127) yield an implicit relation between $`\varphi `$ and $`t`$. However, the problem is that we need $`\varphi `$ at equal intervals of time (to enable us to use the standard FFT algorithms) and this makes the computation of $`\varphi (t)`$ expensive: every time-sample $`\varphi _i\varphi (t_i)`$ is computed by first solving Eq. (126) iteratively for $`v_i`$, the lower limit in the integral for a given $`t_i`$, and then using this $`v_i`$ as the lower limit in the integral of Eq. (127). Though the second step is the computation of a single integral, the first step is a rather slowly converging ($`10`$ iterations for every $`t_i`$) computation.
This problem could have been circumvented if it had been adequate to use the explicit analytical expression $`\varphi (t)=b_0(t_{\mathrm{LSO}}t)^{5/8}+_{k1}b_k(t_{\mathrm{LSO}}t)^{(4k)/8}`$ (modulo logarithms) obtained by : (i) expanding the quantity $`E^{}(v)/(v)`$ in the integrands, in a straightforward expansion in powers of $`v`$, (ii) integrating term by term, and (iii) inverting analytically by successive iterations (see e.g. ). However, this straightforward PN expansion of the phasing formula defeats the very purpose of P-approximants and loses all the benefits brought by the constructions given in . Consequently, DIS had to use the iterative procedure to compute the signal phasing. By contrast, using (any form) of SPA, i.e. an explicit analytical f-domain expression, brings a tremendous reduction in computational costs. On the one hand, as we shall discuss below there is no iterative procedure involved in computing SPA. Secondly they are computed directly in the frequency-domain and hence lead to a further cost reduction, since time-domain waveforms need to be Fourier transformed using FFTs โ costing $`N\mathrm{log}_2N`$ floating point operations โ in addition to floating point operations required to compute time-domain templates.
Let us recall that the usual SPA is given by Eq. (76). In this expression $`t_f`$ is the stationary point of the phase in the integral of Eq. (71). At a Fourier frequency $`f=v_f^3/\pi m`$ the stationary point $`t_f`$ is given by Eq. (130), which is a non-iterative computation. One then substitutes this value of $`t_f`$ in Eq. (131) to compute $`\psi _f`$ โ the phase of the Fourier component. Moreover, the derivative of the frequency which occurs in the amplitude of the Fourier transform can be computed using Eq. (144) while the factor $`a(t_f)f^{2/3}`$ from Eq. (62). Every quantity that appears in the SPA is computed using a straighforward integral or a mere algebraic expression. Hence, from the computational-cost point-of-view, it is desirable to use some SPA to generate templates. Since the usual SPA has been shown to be inadequate for representing time-windowed signals from massive binaries, we have proposed the use of corrections to (for $`ff_{\mathrm{up}}F_{\mathrm{LSO}}`$) and analytic extensions of (for $`fF_{\mathrm{up}})`$ the usual SPA. In Table VII we compare for archetypal binaries, the computational costs of templates that are generated in the time-domain and Fourier transformed using an FFT algorithm with the computational costs for the uSPA, inSPA and irSPA. This Table clearly shows that it is sensible to generate templates in the Fourier domain. The SPA is up to a factor 100 times faster and the irSPA is up to a factor 10 times faster than the corresponding time-domain construction and Fourier transformation. Table VII together with Table V (of overlaps) demonstrates that SPP approximants while more expensive to generate than the usual SPA are nevertheless โaffordableโ, and are anyway necessary for efficient searches of inspiral signals in gravitational wave interferometer data.
## VII Concluding Remarks
After nearly two decades of detector-technology development long-baseline interferometric gravitational wave antennas LIGO/VIRGO are scheduled to become operational in about 2-4 years with target sensitivities that are good enough to detect inspiral events from massive ($`m>20M_{}`$) binaries at an optimistic rate of a few per year. Searches are planned to be carried out over a range of $`0.2`$-$`50M_{}`$ by the method of matched filtering.
An important issue in matched filtering is the number of cycles accumulated in the correlation integral since the SNR grows as the square-root of the number of cycles. While this is strictly true, if the noise power spectrum of the instrument is independent of frequency, in practice one can only improve the SNR in proportion to the square-root of a โusefulโ number of cycles $`N_{\mathrm{useful}}`$ which is determined by a combination of the detector noise power spectrum and the signalโs power-spectrum. We have pointed out how the number of useful cycles can be a lot smaller than the actual number of cycles for massive and relativistic systems: e.g. a $`(10M_{},10M_{})`$ \[ $`(20M_{},20M_{})`$\] binary system has only 7.6 \[3.4\] useful cycles in the detectorโs bandwidth (see Table I). A priori , it may seem that the fewer number of cycles should make it easier to model the massive black hole binaries compared to the lighter neutron star-neutron star ones with its corresponding large number of cycles to phase. Tables IVVI show that there is some truth in this, but that for very massive black hole binaries, these fewer cycles are in fact more difficult to model than the neutron star-neutron star, or neutron star-black hole cases for two reasons: (i) they are near the end of the inspiral, i.e. when the radiation reaction effects drives a faster drift of the frequency which has to be modelled accurately (this is why we need P-approximants introduced in DIS); (ii) they might terminate due to the transition from inspiral to plunge while in the detectorโs bandwidth, and this poses the problem of accurately describing the Fourier transform of a time-windowed signal (this requires the correction factors introduced in this paper). All this places stringent demands in modelling the waveform in the Fourier-domain and due attention needs to be paid to delicate issues of detail. This task is all the more important that the first detections expected from LIGO/VIRGO are likely to concern massive systems with $`m25\pm 5M_{}`$, for which the LSO frequency lies near the middle of the sensitivity curve \[see Fig. 1\].
To this end, the present work makes two new robust (i.e. assumption-independent) contributions:
* the proposal of stationary phase P-approximants (SPP) which combine the excellent performance of our time-domain P-approximants with the analytic convenience of the stationary phase approximation without serious loss of event-rate. These Fourier-domain P-approximants perform as well as their time-domain counterparts in extracting the true general relativistic signal.
* the definition of a universal Newtonian-like โedge-correctionโ factor $`๐(\zeta (f))`$, as well as its relativistic complement $`g_{\frac{3}{2}}(x(f))`$ which take into account the frequency-domain effects, concentrated around (and on both sides) of $`F_{\mathrm{max}}=F(t_{\mathrm{max}})`$, for signals which are abruptly shut off, in the time-domain, after $`t_{\mathrm{max}}`$.
In addition to these new achievements, let us mention two other useful contributions, of a more technical nature: (i) our recommendation to systematically use a smooth time window at the lower frequency side to conveniently and efficiently suppress spurious oscillations due to a numerical low frequency cutoff and (ii) the emphasis on the comparison of the form of the signals in the โwhitenedโ time-domain.
Based on the detailed analysis presented in this paper we find that for post-Newtonian template generation of binary systems of total mass $`m5M_{}`$ it suffices to use the usual SPA (without correction factor) of the P-approximants defined in DIS. On the other hand, in the total mass range $`5M_{}m40M_{}`$, it is crucial to use our new SPP approximants to construct the frequency-domain templates.
In addition to the construction of the SPP approximants, the paper has examined in detail the Fourier-domain effects entailed by a sharp time-domain windowing. As emphasized in the introduction, at our present stage of knowledge, one cannot be sure that a template waveform terminated (in the time-domain) at the LSO is an accurate-enough representation of a real GW signal coming from massive binaries (say with $`m<40M_{}`$). We have given several plausibility arguments towards justifying this assumption: brevity of the plunge, and an expected frequency separation from the merger signal. In the absence of knowledge of the transient plunge signal and of the final merger signal, we have argued that it is best to use a template waveform which is terminated at the LSO. \[Actually, we anticipate that the effectualness of the template waveform will be increased if we allow it to be terminated at a frequency somewhat larger than $`F_{\mathrm{LSO}}`$ (thereby allowing it to approximately represent the plunge waveform).\] Consequently, this work has concentrated on signal models that are truncated in the time-domain by a step-function and has aimed at constructing the best associated Fourier-domain analytical representation for this possibility.
We have also pointed out that the opposite assumption of an abrupt termination at $`F_{\mathrm{LSO}}`$ of the usual SPA in the frequency domain implies, when viewed from either the time-domain or the whitened time-domain, the existence of some coherent oscillations โringingโ after the LSO crossing. We have done another numerical experiment on this issue, by appending to the inspiral signal a smooth decay taking place over less than $`3F_{\mathrm{LSO}}^1`$ time-scales. We have found that our improved SPA was a reasonably good representation of (the FT of) such a signal, and definitely a better representation than the usual SPA one. Let us finally reiterate that, we do not claim to have conclusively ruled out the possibility that a frequency-windowed SPA may perform better compared to the time-windowed SPA we propose here. This important issue is not settled though we conjecture that this is unlikely. Anyway, this paper is the first one to explicitly construct the frequency domain version of the time-domain P-approximants which were shown in DIS to bring indispensible improvements over the usually considered T-approximants. Therefore, even in the unlikely case where a straightforward frequency-window turns out to be a better model than the time-window assumed in most of this work, one will still require the formulas given in this paper (with the trivial change of replacing the correction factors $`๐(\zeta )`$ by a $`\theta `$ function $`\theta (F_{\mathrm{LSO}}f)`$) to generate sufficiently accurate f-domain filters. In view of these comments, we feel there is a urgent need to model more precisely the transition from the inspiral signal to the plunge signal close to the last stable orbit. We hope that the techniques (if not all the details of our construction) used in this work to handle the blow up of $`\dot{F}(t)`$ at the LSO will be useful (maybe with some modifications) even if, on a later examination, this blow up turns out to be an artefact of an approximation which may drastically alter with a better treatment of the transition to the plunge. Only with this improved understanding and its implications for the construction of templates can one build even more optimal templates for massive binaries and maximise our chance of detecting them. Independently of issues such as windowing in time versus windowing in frequency or the nature of the plunge we feel that in general P-approximants are much better tools than the Taylor approximants. We hope to come back to this question in a future work .
Another aspect that needs to be looked into is the issue of whether whether the interferometers will work in the time-domain or the frequency-domain. If indeed, they would decide to work in the time domain: i.e., to store the raw output, and to transform it nearly online in the defiltered time-domain equivalent GW amplitude $`h(t)`$ the analysis of this paper would be irrelevant. In that case, one should store the Wiener transformed time-domain filter $`K(t)=w_1h(t)`$. However, with the presently available computational resources it seems hopeless to filter in the time-domain. We therefore anticipate that, though the raw detector output will be stored in the time-domain, all filtering will be done in the Fourier domain. In this event, the robust aspects of the present analysis will be relevant even if not the details.
The formalism developed in this paper can be applied not only to initial interferometers but also to future generations of interferometers. We have refrained from applying our formalism to the case of LIGO II since the LIGO II design is at the moment in a state of flux and any quantitative results we may quote will soon be irrelevant. However, we should expect the results of this work to be important for any detector that works with a lower seismic cutoff and a broader bandwidth than LIGO I, since in such cases we will have to match the signalโs phase for a larger number of effective cycles.
There are several notable and obvious improvements that need to be pursued. The sensitivity to the value of $`F_{\mathrm{LSO}}`$ needs to be investigated \[in particular, our improved SPA will probably maximize their overlaps with the real signals if we allow some flexibility in the choice of $`F_{\mathrm{LSO}}`$ (within some limits)\]. Once the results of 3PN generation of gravitational waves are available and are combined with the 3PN results on the dynamics they must be included in the construction of templates. In our discussion we have not considered waveforms from binaries with spinning compact objects. Nor have we included the effect of eccentricity on the detectability . These are unarguably important physical effects that need to be incorporated in later data analysis algorithms. Future research in this area should shed light on these issues.
###### Acknowledgements.
We thank J.Y. Vinet for informative communications concerning the VIRGO noise curve. BRI and BSS thank A. Gopakumar and B.J. Owen for discussions on the validity of the stationary phase approximation. BRI would like to thank AEI, Germany and PPARC, U.K. for visiting fellowships.
## A List of Symbols
| $`a(t)`$ | GW amplitude; $`h(t)=2a(t)\mathrm{cos}\varphi (t)`$ |
| --- | --- |
| $`\alpha `$ | $`=\frac{1}{2}v_{\mathrm{LSO}}^{4/3}e_1^{1/3}`$; Eq. (171) |
| cspa | corrected SPA; Eqs. (112)-(113) |
| $`C_n(t_1t_2)`$ | correlation function of noise |
| $`๐(\zeta )`$ | $`\frac{1}{2}\mathrm{erfc}(e^{i\pi /4}\zeta )`$; correction factor; softened step function |
| $`\delta `$ | leading phase correction to SPA; Eq. (78) |
| $`\eta `$ | symmetric mass ratio $`m_1m_2/(m_1+m_2)^2`$ |
| $`\mathrm{erfc}(x)`$ | complementary error function; Eq. (99) |
| $`e_1(\eta )`$ | $`\left[\frac{d}{dv}\left(\frac{E^{}(v)}{(v)}\right)\right]_{v_{\mathrm{LSO}}}`$; Eq. (145) |
| $`E(v)`$ | dimensionless energy function |
| $`\epsilon _1`$ | $`|\frac{\dot{a}(t)}{a(t)\dot{\varphi }(t)}|`$; Eq. (77) |
| $`\epsilon _2`$ | $`|\frac{\ddot{\varphi }(t)}{\dot{\varphi }^2(t)}|=|\frac{1}{2\pi }\frac{\dot{F}(t)}{F^2(t)}|=\frac{1}{2\pi N}`$; Eq. (77) |
| FFT | Fast Fourier transform |
| f-domain | frequency-domain |
| f-window | frequency window |
| $`F(t)`$ | instantaneous GW frequency |
| $`(v)`$ | flux function |
| $`F_{\mathrm{min}}(F_{\mathrm{max}})`$ | GW frequency at $`t_{\mathrm{min}}(t_{\mathrm{max}})`$ |
| $`F_{\mathrm{LSO}}`$ | GW frequency at LSO |
| $`F_{\mathrm{Nyquist}}`$ | Nyquist Frequency |
| $`f`$ | Fourier frequency |
| $`f_{\mathrm{det}}`$ | characteristic detection frequency; minima of effective GW noise $`\sqrt{fS_n(f)}`$ |
| $`f_\mathrm{p}`$ | frequency at which $`d\mathrm{SNR}^2/d(\mathrm{ln}f)`$ peaks |
| $`f_\mathrm{s}`$ | seismic frequency |
| $`f_{\mathrm{up}}`$ | transition frequency between the low and high frequency approximations for the irSPA |
| $`g_{\frac{3}{2}}(x)`$ | $`_0^{\mathrm{}}๐\widehat{\tau }e^{i\left(3x\widehat{\tau }2\widehat{\tau }^{3/2}\right)}`$; Eq. (170) |
| GW | Gravitational wave |
| $`\mathrm{\Gamma }`$ | Gamma function |
| $`h_s^2(f)`$ | squared amplitude of effective GW signal; $`N(f)a^2(f)`$ |
| $`h_n^2(f)`$ | squared amplitude of effective GW noise; $`fS_n(f)`$ |
| $`h(t)`$ | time domain signal |
| $`\stackrel{~}{h}(f)`$ | Fourier transform of $`h(t)`$; $`\stackrel{~}{h}(f)_{\mathrm{}}^{\mathrm{}}๐te^{2\pi ift}h(t)`$ |
| $`\stackrel{~}{h}_+(f)`$ | Fourier transform of non-resonant part of $`h(t)`$ |
| $`\stackrel{~}{h}_{}(f)`$ | Fourier transform of resonant part of $`h(t)`$ |
| $`h_{}^{\mathrm{inspa}}(f)`$ | improved Newtonian SPA corresponding to $`\stackrel{~}{h}_{}(f)`$; Eqs. (109) (118) |
| $`h_+^{\mathrm{edge}}(f)`$ | edge approximation to $`\stackrel{~}{h}_+(f)`$; Eq. (94) |
| $`h^{\mathrm{intot}}(f)`$ | $`=h_+^{\mathrm{edge}}+h_{}^{\mathrm{inspa}}`$: total improved Newtonian SPA of $`h(t)`$ |
| $`h_{}^{\mathrm{irspa}}(f)`$ | improved relativistic SPA corresponding to $`\stackrel{~}{h}_{}(f)`$; Eqs. (166), (167), (173) |
| $`h^{\mathrm{irtot}}(f)`$ | $`=h_+^{\mathrm{edge}}+h_{}^{\mathrm{irspa}}`$: total relativistic SPA of $`h(t)`$ |
| inspa | improved Newtonian SPA |
| irspa | improved relativistic SPA |
| $`l(v)`$ | factored flux function |
| $`m`$ | total mass of the binary |
| $``$ | chirp mass $`\eta ^{3/5}m`$ |
| $`n(t)`$ | noise |
| $`n_{\frac{1}{2}}(t)`$ | whitened noise; $`w_{\frac{1}{2}}(t)n(t)`$ |
| $`N_{\mathrm{tot}}`$ | total number of cycles; Eq. (30) |
| $`N(F)`$ | instantaneous number of cycles; Eq. (31) |
| $`N_{\mathrm{new}}(F)`$ | instantaneous number of cycles in Newtonian case; Eq. (38) |
| $`N_{\mathrm{rel}}(F)`$ | instantaneous number of cycles in relativistic case; Eq. (40) |
| $`N_{\mathrm{useful}}`$ | useful number of cycles; Eq. (36) |
| $`๐ช`$ | overlap (Normalised Ambiguity Function); Eq. (20) |
| $`\varphi (t)`$ | GW phase |
| $`P_n`$ | P-approximant of order $`v^n`$ |
| $`\rho `$ | signal to noise ratio |
| SPA | stationary phase approximation |
| $`S_n(f)`$ | two-sided noise power spectral density |
| $`\sigma (f)`$ | weight function in $`\rho ^2`$; Eq. (46) |
| $`\sigma (t,z_1,z_2)`$ | smoothing time window; Eq. (125) |
| $`t_{\mathrm{min}}`$ | starting time of the signal |
| $`t_{\mathrm{max}}`$ | time at which the signal terminates or is terminated |
| $`T_n`$ | Taylor approximant of order $`v^n`$ |
| $`\tau `$ | $`\frac{t_{\mathrm{LSO}}t}{m}`$ |
| $`\theta `$ | Heaviside step function |
| uspa | usual SPA |
| uspaw | usual SPA frequency windowed |
| uspan | usual SPA up to Nyquist frequency |
| $`v`$ | invariant velocity $`(\pi mF)^{1/3}`$ |
| $`v_{}`$ | invariant velocity $`(\pi F)^{1/3}`$ |
| $`w(f)`$ | weight factor $`\frac{a^2(f)}{h_n^2(f)}`$ |
| $`w_1(t)`$ | correlation inverse of noise correlation function; Eq. (15) |
| $`w_{\frac{1}{2}}(\tau )`$ | whitening kernel; Eq. (22) |
| $`x`$ | $`=\frac{2\pi }{3}\alpha m(F_{\mathrm{LSO}}f)`$; Eq. (172) |
| $`\zeta _0(f)`$ | $`\sqrt{\pi \dot{F}(t_f)}(t_ft_{\mathrm{max}})`$; Eq. (113) |
| $`\zeta _<(f)`$ | $`=\sqrt{\psi _f(t_f)\psi _f(t_{\mathrm{max}})}`$; Eq. (108) |
| $`\zeta _>(f)`$ | $`=\frac{\sqrt{\pi }(fF_{\mathrm{max}})}{\sqrt{\dot{F}(t_{\mathrm{max}})}}`$; Eq. (119) | |
warning/0001/hep-th0001035.html | ar5iv | text | # Untitled Document
IFT-P.005/2000
Super-Poincarรฉ Covariant Quantization of the Superstring
Nathan Berkovits<sup>1</sup> e-mail: nberkovi@ift.unesp.br
Instituto de Fรญsica Teรณrica, Universidade Estadual Paulista
Rua Pamplona 145, 01405-900, Sรฃo Paulo, SP, Brasil
Using pure spinors, the superstring is covariantly quantized. For the first time, massless vertex operators are constructed and scattering amplitudes are computed in a manifestly ten-dimensional super-Poincarรฉ covariant manner. Quantizable non-linear sigma model actions are constructed for the superstring in curved backgrounds, including the $`AdS_5\times S^5`$ background with Ramond-Ramond flux.
January 2000
1. Introduction
There are many motivations for covariantly quantizing the superstring. As in any theory, it is desirable to make all physical symmetries manifest in order to reduce the amount of calculations and simplify any cancellations coming from the symmetry. Recently, an additional motivation has come from the desire to construct a quantizable sigma model action for the superstring in curved backgrounds with Ramond-Ramond flux.
Most attempts to covariantly quantize the superstring have started from the classical super-Poincarรฉ invariant version of the Green-Schwarz (GS) action . One quantization approach is based on gauge-fixing the fermionic symmetries to get to โsemi-light-coneโ gauge where $`(\gamma ^+\theta )_\alpha =0`$ and $`\gamma ^+=\gamma ^0+\gamma ^9`$ . In this gauge, the covariant Green-Schwarz action simplifies to $`S=d^2z[x^m\overline{}x_m+x^+(\theta \gamma ^{}\overline{}\theta )].`$ However, even this simplified action cannot be easily quantized since the propagator for $`\theta `$ involves $`(x^+)^1`$ which is not well-defined.<sup>2</sup> On a genus $`g`$ worldsheet with $`N`$ punctures, $`x^+`$ vanishes at $`2g+N2`$ points on the worldsheet. This fact is related to the need for interaction-point operators in the light-cone GS superstring. For this reason, it has not yet been possible to use this approach to construct physical vertex operators or compute scattering amplitudes, except in the $`p^+0`$ limit that reproduces the light-cone gauge computations . Another approach to quantizing the covariant Green-Schwarz action is based on replacing the fermionic second-class constraints with an appropriate set of first-class constraints , sometimes using SO(9,1)/SO(8) harmonic variables which covariantize the semi-light-cone gauge choice. However, despite numerous attempts , noone was able to find an appropriate set of first-class constraints which allows the covariant computation of scattering amplitudes.
In the absence of Ramond states, it is possible to quantize the superstring in a manifestly Lorentz-covariant manner using the standard Ramond-Neveu-Schwarz (RNS) formalism. However, none of the spacetime supersymmetries are manifest in the RNS formalism and, in order to explicitly construct the spin field for Ramond states, manifest SO(9,1) Lorentz invariance must be broken (after Wick-rotation) to a U(5) subgroup . Recently, an alternative formalism for the superstring was constructed which manifestly preserves this same U(5) subgroup in addition to manifestly preserving six of the sixteen spacetime supersymmetries . The worldsheet variables of this supersymmetric U(5) formalism are related to those of the RNS formalism by a field redefinition, allowing one to prove that physical vertex operators and scattering amplitudes in the two formalisms are equivalent. However, the lack of manifest Lorentz invariance makes it difficult to use this formalism to describe the superstring in curved (Wick-rotated) backgrounds which do not preserve U(5) holonomy.
In this paper, a new formalism for the superstring will be presented which can be quantized in a manifestly super-Poincarรฉ covariant manner. The worldsheet variables of this formalism will consist of the usual ten-dimensional superspace variables in addition to a bosonic spacetime spinor $`\lambda ^\alpha `$ satisfying the โpureโ spinor condition
$$\lambda ^\alpha \gamma _{\alpha \beta }^m\lambda ^\beta =0$$
for $`m=0`$ to 9. $`\lambda ^\alpha `$ must be complex to satisfy (1.1) and, after Wick-rotating SO(9,1) to SO(10), can be parameterized by eleven complex variables. One of these eleven variables is an overall scale factor, and the other ten parameterize the coset space SO(10)/U(5). So this new formalism is probably related to a covariantization of the U(5) formalism of . Although the precise relation between the two formalisms is still unclear, it will be argued in section 2 that pure spinor variables are necessary for equating RNS vertex operators with the GS vertex operators proposed in .
In section 3, physical states will be defined as elements in the cohomology of the BRST-like operator
$$Q=๐z\lambda ^\alpha d_\alpha $$
where $`d_\alpha `$ is the generator of supersymmetric derivatives as defined in . Since $`d_\alpha (y)d_\beta (z)2(yz)^1\mathrm{\Pi }_m(z)\gamma _{\alpha \beta }^m`$ where $`\mathrm{\Pi }_m`$ is the supersymmetric translation generator, (1.1) implies that $`Q^2=0`$. Note that the operator of (1.1) was used in by Howe to show that the constraints of ten-dimensional super-Yang-Mills and supergravity can be understood as integrability conditions on pure spinor lines.
Using this definition of physical states, one can easily construct the physical massless vertex operators. For the open superstring, the massless vertex operator in unintegrated form is $`V=\lambda ^\alpha A_\alpha (x,\theta )`$ and in integrated form is
$$V=๐z(\mathrm{\Pi }^mA_m+\theta ^\alpha A_\alpha +d_\alpha W^\alpha +N^{mn}F_{mn})$$
where $`A_M`$ are the super-Yang-Mills prepotentials, $`W^\alpha `$ and $`F_{mn}`$ are the gauge-invariant superfields whose lowest components are the gluino and the gluon field strength, and $`N^{mn}`$ is the pure spinor contribution to the Lorentz generator. Except for the $`N^{mn}`$ term, the vertex operator of (1.1) is that proposed by Siegel in . As will be shown in section 4, these vertex operators can be used to compute scattering amplitudes in a manifestly super-Poincarรฉ covariant manner.
The physical vertex operators for the closed superstring can be obtained by taking the โleft-rightโ product of two open superstring vertex operators. In section 5, the integrated form of the closed superstring massless vertex operator will be used to construct a quantizable sigma model action for the superstring in a curved superspace background. As a special case, a quantizable sigma model action will be constructed for the superstring in an $`AdS_5\times S^5`$ background with Ramond-Ramond flux. This action differs from that of Metsaev and Tseytlin in containing a kinetic term for the fermions which allows quantization.
In section 6, further evidence will be given for equivalence with the RNS formalism and some possible applications of the new formalism will be discussed.
2. Pure Spinors and Lorentz Currents
In conformal gauge, the left-moving contribution to the covariant Green-Schwarz superstring action can be written as
$$S=d^2z(\frac{1}{2}x^m\overline{}x_m+p_\alpha \overline{}\theta ^\alpha )$$
where $`p_\alpha `$ is related to $`x^m`$ and $`\theta ^\alpha `$ by the constraint $`d_\alpha =0`$ with
$$d_\alpha =p_\alpha +\gamma _{\alpha \beta }^mx_m\theta ^\beta +\frac{1}{2}\gamma _{\alpha \beta }^m\gamma _{m\gamma \delta }\theta ^\beta \theta ^\gamma \theta ^\delta .$$
Since $`d_\alpha (y)d_\beta (z)2(yz)^1\gamma _{\alpha \beta }^m\mathrm{\Pi }_m(z)`$ where $`\mathrm{\Pi }^m=x^m+\theta ^\alpha \gamma _{\alpha \beta }^m\theta ^\beta `$, $`d_\alpha =0`$ involves first and second-class constraints. The idea of is to find an appropriate set of first-class constraints constructed from $`d_\alpha `$ which can replace the second-class constraints. In such a framework, $`p_\alpha `$ is treated as an independent field and physical vertex operators are annihilated by the first-class constraints. Although an appropriate set of first-class constraints were not found in , Siegel used supersymmetry arguments to conjecture that the massless open superstring vertex operator should have the form
$$V=๐z(\mathrm{\Pi }^mA_m+\theta ^\alpha A_\alpha +d_\alpha W^\alpha )$$
where $`A_M`$ are the super-Yang-Mills prepotentials and $`W^\alpha `$ is the super-Yang-Mills spinor field strength.
For a gluon, the vertex operator of (2.1) reduces to $`V=๐z(x^m๐_m(x)+\frac{1}{2}(p\gamma ^{mn}\theta )_{mn}(x))`$ where $`๐_m`$ and $`_{mn}`$ are the ordinary $`\theta `$-independent gluon gauge field and field strength, which closely resembles the gluon vertex operator in the RNS formalism $`V=๐z(x^m๐_m+\psi ^m\psi ^n_{mn})`$. However, there is a crucial difference between the OPEโs of the SO(9,1) Lorentz currents $`M^{mn}=\frac{1}{2}p\gamma ^{mn}\theta `$ and $`\widehat{M}^{mn}=\psi ^m\psi ^n`$ which will force the introduction of pure spinors. Namely, the OPE of $`M^{kl}`$ with $`M^{mn}`$ has a double pole proportional to $`\frac{16}{4}(\eta ^{kn}\eta ^{lm}\eta ^{km}\eta ^{ln})`$ where the factor of 16 comes from the spinor dimension. However, the double pole in the OPE of $`\widehat{M}^{kl}`$ with $`\widehat{M}^{mn}`$ is proportional to $`(\eta ^{kn}\eta ^{lm}\eta ^{km}\eta ^{ln})`$ without the factor of $`\frac{16}{4}`$. So the vertex operator of can only be equivalent at the quantum level to the RNS vertex operator if one adds a new term to the Lorentz current $`M^{mn}=\frac{1}{2}p\gamma ^{mn}\theta +N^{mn}`$ where $`N^{mn}`$ satisfies the OPE <sup>3</sup> In four dimensions, $`\frac{1}{2}p\gamma ^{mn}\theta `$ has a double pole proportional to $`\frac{4}{4}(\eta ^{kn}\eta ^{lm}\eta ^{km}\eta ^{ln})`$, so there is no need to add new Lorentz degrees of freedom when quantizing the four-dimensional superstring . In six dimensions, $`\frac{1}{2}p_j\gamma ^{mn}\theta ^j`$ (where $`j=1`$ to 2 is an internal SU(2) index) has a double pole proportional to $`\frac{8}{4}(\eta ^{kn}\eta ^{lm}\eta ^{km}\eta ^{ln})`$, so one needs to add degrees of freedom whose Lorentz current has a double pole with itself proportional to $`(\eta ^{kn}\eta ^{lm}\eta ^{km}\eta ^{ln})`$. These degrees of freedom are a bosonic spinor $`u^\alpha `$ and its conjugate momentum $`v_\alpha `$ for $`\alpha =1`$ to 4. They are the ghosts for the โharmonicโ constraints $`\stackrel{~}{d}_\alpha =d_{\alpha 2}e^{\rho i\sigma }d_{\alpha 1}`$ of whose contribution was incorrectly ignored in . The correct massless six-dimensional open superstring vertex operator is $`V=๐z(\mathrm{\Pi }^mA_m+\theta ^{\alpha j}A_{\alpha j}+d_{\alpha j}W^{\alpha j}+\frac{1}{2}(u\gamma ^{mn}v)F_{mn})`$ where $`F_{mn}`$ is a superfield whose lowest component is the gluon field strength. Note that this vertex operator is annihilated on-shell by the โharmonicโ BRST-like operator $`Q=๐zu^\alpha \stackrel{~}{d}_\alpha `$ and the central charge contribution from $`v_\alpha u^\alpha `$ cancels the contribution from $`p_{\alpha 2}\theta ^{\alpha 2}`$ in the stress tensor to give a vanishing conformal anomaly.
$$N^{kl}(y)N^{mn}(z)\frac{\eta ^{m[l}N^{k]n}(z)\eta ^{n[l}N^{k]m}(z)}{yz}3\frac{\eta ^{kn}\eta ^{lm}\eta ^{km}\eta ^{ln}}{(yz)^2}.$$
As will now be shown, such a Lorentz current $`N^{mn}`$ can be explicitly constructed from a pure spinor $`\lambda ^\alpha `$, i.e. a complex bosonic spinor satisfying (1.1). To parameterize the eleven independent complex degrees of freedom of $`\lambda ^\alpha `$, it is convenient to Wick-rotate and temporarily break SO(10) to SU(5)$`\times `$ U(1) as in . The sixteen complex components of $`\lambda ^\alpha `$ split into $`(\lambda ^+,\lambda _{ab},\lambda ^a)`$ for $`a,b=1`$ to 5, which transform respectively as $`(1_{\frac{5}{2}},\overline{10}_{\frac{1}{2}},5_{\frac{3}{2}})`$ representations of SU(5)$`\times `$ U(1) where the subscript denotes the U(1) charge. In terms of the eleven independent complex variables $`(\gamma ,u_{ab})`$ transforming as $`(1_5,\overline{10}_2)`$ representations, one can check that
$$\lambda ^+=\gamma ,\lambda _{ab}=\gamma u_{ab},\lambda ^a=\frac{1}{8}\gamma ฯต^{abcde}u_{bc}u_{de}$$
satisfies the pure spinor condition of (1.1). Note that $`\gamma `$-matrices in U(5) notation satisfy $`\lambda \gamma ^a\lambda =\lambda ^+\lambda ^a+\frac{1}{8}ฯต^{abcde}\lambda _{bc}\lambda _{de}`$ and $`\lambda \gamma _a\lambda =\lambda _{ab}\lambda ^b`$ where the SO(10) vector has been split into a $`5_1`$ and $`\overline{5}_1`$ representation.
In conformal gauge, the worldsheet action for the left-moving variables will be defined as
$$S=d^2z(\frac{1}{2}x^m\overline{}x_m+p_\alpha \overline{}\theta ^\alpha +\frac{1}{2}v^{ab}\overline{}u_{ab}+\beta \overline{}\gamma )$$
with the left-moving stress tensor
$$T=\frac{1}{2}x^mx_m+p_\alpha \theta ^\alpha +\frac{1}{2}v^{ab}u_{ab}+\beta \gamma $$
where $`(\beta ,v^{ab})`$ are the conjugate momenta for $`(\gamma ,u_{ab})`$. As desired, $`T`$ has no conformal anomaly since the central contribution for the new degrees of freedom is $`22`$, which cancels the central charge contribution from the $`x^m`$ and $`(\theta ^\alpha ,p_\alpha )`$ variables.
In U(5) notation, the SO(10) Lorentz currents $`N^{mn}`$ split into $`(N,N_a^b,N^{ab},N_{ab})`$ which transform respectively as $`(1_0,24_0,10_2,\overline{10}_2)`$ representations. After fermionizing $`\gamma =\eta e^\varphi `$ and $`\beta =\xi e^\varphi `$ as in , $`N^{mn}`$ will be defined as
$$N=\frac{1}{\sqrt{5}}(u_{ab}v^{ab}+\frac{25}{4}\eta \xi +\frac{15}{4}\varphi ),N_a^b=u_{ac}v^{bc}\frac{1}{5}\delta _a^bu_{cd}v^{cd},$$
$$N^{ab}=v^{ab},N_{ab}=3u_{ab}+u_{ac}u_{bd}v^{cd}+u_{ab}(\frac{5}{2}\eta \xi +\frac{3}{2}\varphi ).$$
Using the free-field OPEโs,
$$\eta (y)\xi (z)(yz)^1,\varphi (y)\varphi (z)\mathrm{log}(yz),v^{ab}(y)u_{cd}(z)\delta _c^{[a}\delta _d^{b]}(yz)^1,$$
one can check that
$$N_a^b(y)N_c^d(z)\frac{\delta _c^bN_a^d(z)\delta _a^dN_c^b}{yz}3\frac{\delta _a^d\delta _c^b\frac{1}{5}\delta _a^b\delta _c^d}{(yz)^2},N(y)N(z)\frac{3}{(yz)^2},$$
$$N_{ab}(y)N^{cd}(z)\frac{\delta _{[a}^{[c}N_{b]}^{d]}(z)\frac{2}{\sqrt{5}}\delta _a^{[c}\delta _b^{d]}N(z)}{yz}+3\frac{\delta _a^{[c}\delta _b^{d]}}{(yz)^2},$$
which correctly reproduces the OPE of (2.1).
Furthermore, $`[๐zN^{mn},\lambda ^a]=\frac{1}{2}(\gamma ^{mn}\lambda )^\alpha `$ as can be easily shown by noting that $`N^{mn}=\frac{1}{2}\omega \gamma ^{mn}\lambda `$ where $`\omega _\alpha `$ is a spinor of the opposite chirality to $`\lambda ^\alpha `$ with components<sup>4</sup> Terms coming from normal-ordering ambiguities in $`\omega \gamma ^{mn}\lambda `$ can be ignored since they only involve $`\varphi +\eta \xi `$ and $`u_{ab}`$, which have no singularities with $`\lambda ^\alpha `$.
$$\omega _+=\xi e^\varphi (\eta \xi \frac{1}{2}u_{ab}v^{ab}),\omega ^{ab}=\xi e^\varphi v^{ab},\omega _a=0.$$
Note that
$$\omega _\alpha (y)\lambda ^\beta (z)(yz)^1\delta _\alpha ^\beta \frac{1}{2}(yz)^1\gamma _m^{\beta +}\xi e^\varphi (\gamma ^m\lambda )_\alpha $$
where the $`+`$ in $`\gamma _m^{\beta +}`$ signifies the $`1_{\frac{5}{2}}`$ spinor component in the SU(5) notation of (2.1). The second term in the OPE of (2.1) is necessary for $`\omega `$ to have no singularity with $`\lambda \gamma ^m\lambda `$, however, it does not contribute to the commutator $`[๐z\omega \gamma ^{mn}\lambda ,\lambda ]`$ since $`\lambda \gamma ^m\gamma ^{np}\lambda =0.`$
So after introducing pure spinors, it is possible to obtain vanishing conformal anomaly and to relate the RNS gluon vertex operator with the proposal of Siegel in . It will now be shown how these pure spinors can be used to define physical vertex operators and compute scattering amplitudes in a super-Poincarรฉ covariant manner.
3. Physical Vertex Operators
Since the stress-tensor of (2.1) has vanishing central charge, one can require that physical vertex operators in unintegrated form are primary fields of dimension zero. However, this requirement is clearly insufficient since, for a massless vertex operator depending only on the zero modes of the worldsheet fields, it implies $`_m^m\mathrm{\Phi }(x,\theta ,\lambda )=0`$ which has far more propagating fields than super-Yang-Mills. One therefore needs a further constraint on physical vertex operators, and using the intuition of , this constraint should be constructed from $`d_\alpha `$ of (2.1).
Using the pure spinor $`\lambda ^\alpha `$ defined in terms of $`\gamma `$ and $`u_{ab}`$ as in (2.1), one can define a nilpotent BRST-like operator
$$Q=๐z\lambda ^\alpha (z)d_\alpha (z).$$
Defining ghost charge to be $`q_{ghost}=๐z\gamma \beta `$, $`Q`$ carries ghost-number one. So it is natural to define physical vertex operators as states of ghost-number 1 in the cohomology of $`Q`$. Note that after Wick rotation, $`\theta ^\alpha `$ and $`\lambda ^\alpha `$ are complex spinors, so the Hilbert space of states should be restricted to analytic functions of these variables.<sup>5</sup> Although it is difficult to impose reality conditions on the states in Euclidean space, this is not a problem for computing scattering amplitudes since it will be trivial to Wick-rotate the final result back to Minkowski space where the reality conditions are easily defined.
It will now be shown for the massless sector of the open superstring that this definition of physical states reproduces the desired super-Yang-Mills spectrum. Massless vertex operators of dimension zero can only depend on the worldsheet zero modes, so the most general such vertex operator of ghost number 1 is $`U=\gamma Y(x,\theta ,u_{ab})`$ where $`Y`$ is an analytic function of $`\theta ^\alpha `$ and $`u_{ab}`$. Since $`Q`$ is Lorentz invariant (after including the contribution of $`N^{mn}`$ (2.1) in the Lorentz generators), elements in its cohomology must transform Lorentz covariantly. But because of the non-linear nature of the Lorentz transformations generated by $`N^{mn}`$, the only finite-dimensional covariantly transforming object which is linear in $`\gamma `$ is $`\lambda ^\alpha `$. So if the cohomology is restricted to finite-dimensional elements, the most general massless vertex operator of ghost number 1 is
$$U=\lambda ^\alpha A_\alpha (x,\theta )$$
where $`A_\alpha (x,\theta )`$ is a generic spinor function of $`x^m`$ and $`\theta ^\alpha `$.
The constraint $`QU=0`$ implies that $`\lambda ^\alpha \lambda ^\beta D_\beta A_\alpha =0`$ where $`D_\alpha =\frac{}{\theta ^\alpha }+\gamma _{\alpha \beta }^m\theta ^\beta \frac{}{x^m}`$. Since $`\lambda \gamma ^m\lambda =0`$, this implies that $`D_\alpha (\gamma ^{mnpqr})^{\alpha \beta }A_\beta =0`$, which is the on-shell constraint for the spinor prepotential of super-Yang-Mills. Furthermore, the gauge transformation
$$\delta U=Q\mathrm{\Omega }(x,\theta )=\lambda ^\alpha D_\alpha \mathrm{\Omega }(x,\theta )$$
reproduces the usual super-Yang-Mills gauge transformation $`\delta A_\alpha =D_\alpha \mathrm{\Omega }`$ where $`\mathrm{\Omega }(x,\theta )`$ is a generic scalar superfield. So the ghost number 1 cohomology of $`Q`$ for the massless sector reproduces the desired super-Yang-Mills spectrum.
To compute scattering amplitudes, one also needs vertex operators in integrated form, i.e. integrals of dimension 1 primary fields. Normally, these are obtained from the unintegrated vertex operator by anti-commuting with $`b(z)`$. But in this formalism, there are no $`(b,c)`$ ghosts, so it is presently unclear how to relate the two types of vertex operators. Nevertheless, one can define physical integrated vertex operators as elements in the BRST cohomology of ghost-number zero.
In the massless sector, there is an obvious candidate which is the dimension 1 vertex operator of (2.1) suitably modified to include the pure spinor contribution to the Lorentz current, i.e.
$$V=\mathrm{\Pi }^mA_m+\theta ^\alpha A_\alpha +d_\alpha W^\alpha +N^{mn}F_{mn}$$
where $`F_{mn}`$ is the superfield whose lowest component is the gluon field strength. To show that $`[Q,๐zV]=0`$, first note that
$$d_\alpha (y)๐zv(z)\frac{1}{2}๐z(yz)^1F_{mn}(z)(d(z)\gamma ^{mn})_\alpha $$
where $`v(z)=\mathrm{\Pi }^mA_m+\theta ^\alpha A_\alpha +d_\alpha W^\alpha .`$ Since $`\lambda ^\alpha (y)N^{mn}(z)\frac{1}{2}(yz)^1(\lambda (z)\gamma ^{mn})^\alpha `$,
$$[Q,๐zV]=๐zN^{mn}\lambda ^\alpha D_\alpha F_{mn}=๐zN^{mn}(\lambda \gamma _m_nW)$$
where $`W^\alpha `$ is the spinor field strength. But using $`N^{mn}=\frac{1}{2}\omega \gamma ^{mn}\lambda `$ from (2.1),
$$N^{mn}(\lambda \gamma _m)_\alpha =\frac{1}{2}(w\gamma ^{mn}\lambda )(\lambda \gamma _m)_\alpha =\frac{1}{2}(w_\beta \lambda ^\beta )(\lambda \gamma ^n)_\alpha $$
since $`(\gamma ^m\lambda )_\alpha (\gamma _m\lambda )_\beta =\frac{1}{2}\gamma _{\alpha \beta }^m(\lambda \gamma _m\lambda )=0`$ from ten-dimensional $`\gamma `$-matrix identities. Finally, using the gluino equation of motion,
$$[Q,๐zV]=\frac{1}{2}๐z(w_\beta \lambda ^\beta )(\lambda \gamma ^n_nW)=0$$
so $`๐zV`$ describes a physical integrated vertex operator. Note that the super-Yang-Mills gauge transformation $`\delta A_M=_M\mathrm{\Omega }`$ transforms $`V`$ by the total derivative $`_z\mathrm{\Omega }`$, so $`๐zV`$ is manifestly gauge-invariant.
4. Computation of Scattering Amplitudes
In this section, it will be shown how to compute tree-level open superstring scattering amplitudes in a manifestly super-Poincarรฉ covariant manner. To compute $`N`$-point tree-level scattering amplitudes, one needs three vertex operators in unintegrated form and $`N3`$ vertex operators in integrated form. Since only the massless vertex operators are known explicitly, only scattering of massless states will be considered here.
The two-dimensional correlation function which needs to be evaluated for computing tree-level scattering of $`N`$ super-Yang-Mills multiplets is
$$๐=U_1(z_1)U_2(z_2)U_3(z_3)๐z_4V_4(z_4)\mathrm{}๐z_NV_N(z_N)$$
where $`U_r`$ is the dimension 0 vertex operator of (3.1), $`V_r`$ is the dimension 1 vertex operator of (3.1), and the locations of $`(z_1,z_2,z_3)`$ can be chosen arbitrarily because of SL(2,R) invariance.
The functional integral over the non-zero modes of the various worldsheet fields is completely straightforward using the free-field OPEโs. For example, the dimension 1 worldsheet fields $`(x^m,\theta ^\alpha ,d_\alpha ,N^{mn})`$ can be integrated out by contracting with other dimension 1 fields or with $`(x^m,\theta ^\alpha ,\lambda ^\alpha )`$. Note that manifest Lorentz invariance is preserved by the contractions of $`N^{mn}`$ because its only singular OPEโs are $`N^{mn}(y)\lambda ^\alpha (z)\frac{1}{2}(yz)^1(\gamma ^{mn}\lambda )^\alpha `$ and (2.1).
However, the functional integral prescription for the zero modes of the worldsheet fields needs to be explained. Besides the zero modes of $`x^m`$ (which are treated in the usual manner using conservation of momentum), there are the eleven bosonic zero modes of $`(\gamma ,u_{ab})`$ and the sixteen fermionic zero modes of $`\theta ^\alpha `$. After integrating out the non-zero modes, one gets an expression
$$๐=๐z_4\mathrm{}๐z_N\lambda ^\alpha \lambda ^\beta \lambda ^\gamma f_{\alpha \beta \gamma }(z_r,k_r,\theta )$$
where only the zero modes of $`\lambda ^\alpha `$ contribute and $`f_{\alpha \beta \gamma }`$ is a function which depends on $`z_4\mathrm{}z_N`$, on the momenta $`k_r^m`$ for $`r=1`$ to $`N`$, and on the zero modes of $`\theta ^\alpha `$.
The prescription for integration over the remaining worldsheet zero modes will be
$$\lambda ^\alpha \lambda ^\beta \lambda ^\gamma f_{\alpha \beta \gamma }(z_r,k_r,\theta )๐\mathrm{\Omega }(\overline{\lambda }_\delta \lambda ^\delta )^3(\frac{}{\theta }\gamma ^{lmn}\frac{}{\theta })(\overline{\lambda }\gamma _l\frac{}{\theta })(\overline{\lambda }\gamma _m\frac{}{\theta })(\overline{\lambda }\gamma _n\frac{}{\theta })$$
$$\lambda ^\alpha \lambda ^\beta \lambda ^\gamma f_{\alpha \beta \gamma }(z_r,k_r,\theta )$$
where $`\overline{\lambda }_\alpha `$ is the complex conjugate of $`\lambda ^\alpha `$ in Euclidean space and $`d\mathrm{\Omega }`$ is an integration over the different possible orientions of $`\lambda ^\alpha `$. Although this prescription is defined in Euclidean space, it is trivial to Wick-rotate the result back to Minkowski space using the fact that
$$๐\mathrm{\Omega }(\overline{\lambda }_\delta \lambda ^\delta )^3\lambda ^\alpha \lambda ^\beta \lambda ^\gamma \overline{\lambda }_\rho \overline{\lambda }_\sigma \overline{\lambda }_\tau =T_{\rho \sigma \tau }^{\alpha \beta \gamma }\frac{1}{4032}[\delta _\rho ^{(\alpha }\delta _\sigma ^\beta \delta _\tau ^{\gamma )}\frac{1}{40}\gamma _m^{(\alpha \beta }\delta _{(\rho }^{\gamma )}\gamma _{\sigma \tau )}^m].$$
Equation (4.1) can be derived using the fact that there is a unique covariantly transforming tensor $`T_{\rho \sigma \tau }^{\alpha \beta \gamma }`$ which is symmetrized with respect to its upper and lower indices and which satisfies $`T_{\alpha \beta \gamma }^{\alpha \beta \gamma }=1`$ and $`T_{\rho \sigma \tau }^{\alpha \beta \gamma }\gamma _{\alpha \beta }^m=T_{\rho \sigma \tau }^{\alpha \beta \gamma }\gamma _m^{\rho \sigma }=0.`$
So the amplitude of (4.1) can be written in SO(9,1) Lorentz-covariant notation as
$$๐=T_{\rho \sigma \tau }^{\alpha \beta \gamma }(\frac{}{\theta }\gamma ^{lmn}\frac{}{\theta })(\gamma _l\frac{}{\theta })^\rho (\gamma _m\frac{}{\theta })^\sigma (\gamma _n\frac{}{\theta })^\tau ๐z_4\mathrm{}๐z_Nf_{\alpha \beta \gamma }(z_r,k_r,\theta )$$
where $`T_{\rho \sigma \tau }^{\alpha \beta \gamma }`$ is defined in (4.1). By expanding $`f_{\alpha \beta \gamma }(z_r,k_r,\theta )`$ as a power series in $`\theta ^\alpha `$, one can check that the prescription of (4.1) selects out the term
$$f_{\alpha \beta \gamma }(z_r,k_r,\theta )=\mathrm{}+a(z_r,k_r)(\theta \gamma ^{lmn}\theta )(\gamma _l\theta )_\alpha (\gamma _m\theta )_\beta (\gamma _n\theta )_\gamma +\mathrm{}$$
in the power series, i.e. $`\lambda ^\alpha \lambda ^\beta \lambda ^\gamma f_{\alpha \beta \gamma }(z_r,k_r,\theta )=a(z_r,k_r)`$.
This prescription for integrating out the zero modes is reasonable since it is Lorentz invariant and since the eleven bosonic zero mode integrations are expected to cancel eleven of the sixteen fermionic zero mode integrations, leaving five zero modes of $`\theta `$ which are removed with five $`\frac{}{\theta }`$โs. Further evidence for this zero-mode prescription comes from the fact that it is gauge invariant and spacetime supersymmetric, as will now be shown.
To show that $`๐`$ is invariant under a gauge transformation $`\delta U_1(z_1)=[๐z\lambda ^\alpha d_\alpha ,\mathrm{\Omega }(z_1)]`$, note that $`๐z\lambda ^\alpha d_\alpha `$ commutes with $`U_r`$ and $`๐z_rV_r`$, so
$$\delta ๐=[๐z\lambda ^\alpha d_\alpha ,\mathrm{\Omega }(z_1)U_2(z_2)U_3(z_3)๐z_4V_4(z_4)\mathrm{}๐z_NV_N(z_N)]$$
$$=๐z_4\mathrm{}๐z_N\lambda ^\alpha \lambda ^\beta \lambda ^\gamma D_\alpha h_{\beta \gamma }(z_r,k_r,\theta )$$
for some $`h_{\beta \gamma }`$ after integrating out the non-zero modes. Using the zero-mode prescription of (4.1),
$$\delta ๐=T_{\rho \sigma \tau }^{\alpha \beta \gamma }(\frac{}{\theta }\gamma ^{lmn}\frac{}{\theta })(\gamma _l\frac{}{\theta })^\rho (\gamma _m\frac{}{\theta })^\sigma (\gamma _n\frac{}{\theta })^\tau \frac{}{\theta ^\alpha }๐z_4\mathrm{}๐z_Nh_{\beta \gamma }(z_r,k_r,\theta ),$$
where conservation of momentum has been used to replace $`D_\alpha `$ with $`\frac{}{\theta ^\alpha }`$ in (4.1). But using anti-symmetry properties of $`\frac{}{\theta }`$, one can show that
$$T_{\rho \sigma \tau }^{\alpha \beta \gamma }(\frac{}{\theta }\gamma ^{lmn}\frac{}{\theta })(\gamma _l\frac{}{\theta })^\rho (\gamma _m\frac{}{\theta })^\sigma (\gamma _n\frac{}{\theta })^\tau \frac{}{\theta ^\alpha }=0,$$
so $`\delta ๐=0`$.
It will now be shown that the prescription of (4.1) is invariant under spacetime supersymmetry transformations, implying that the amplitudes are SO(9,1) super-Poincarรฉ invariant. Under a spacetime supersymmetry transformation with global parameter $`ฯต^\kappa `$, the term $`a(z_r,k_r)`$ of (4.1) transforms as $`\delta a(z_r,k_r)=ฯต^\kappa \xi _\kappa (z_r,k_r)`$ where $`\xi _\kappa `$ appears in the power series for $`f_{\alpha \beta \gamma }`$ as
$$f_{\alpha \beta \gamma }(z_r,k_r,\theta )=\mathrm{}+(\theta \gamma ^{lmn}\theta )(\gamma _l\theta )_\alpha (\gamma _m\theta )_\beta (\gamma _n\theta )_\gamma [a(z_r,k_r)+\theta ^\kappa \xi _\kappa (z_r,k_r)]+\mathrm{}.$$
But $`๐z_4\mathrm{}๐z_N\lambda ^\alpha \lambda ^\beta \lambda ^\gamma f_{\alpha \beta \gamma }`$ comes from vertex operators which commute with $`๐z\lambda ^\delta d_\delta `$, so it must satisfy the constraint
$$๐z_4\mathrm{}๐z_N\lambda ^\alpha \lambda ^\beta \lambda ^\gamma \lambda ^\delta D_\delta f_{\alpha \beta \gamma }=0$$
for any pure spinor $`\lambda ^\alpha `$. Plugging (4.1) into (4.1) and using $`(\lambda \gamma ^{lmn}\theta )(\lambda \gamma _l\theta )(\lambda \gamma _m\theta )(\lambda \gamma _n\theta )=0`$, one finds that
$$๐z_4\mathrm{}๐z_N(\theta \gamma ^{lmn}\theta )(\lambda \gamma _l\theta )(\lambda \gamma _m\theta )(\lambda \gamma _n\theta )\lambda ^\kappa \xi _\kappa (z_r,k_r)$$
must vanish for any pure spinor $`\lambda ^\alpha `$. But this is only possible if $`๐z_4\mathrm{}๐z_N\xi _\kappa (z_r,k_r)=0`$, implying that
$$\delta ๐=๐z_4\mathrm{}๐z_N\delta a(z_r,k_r)=0$$
so the amplitude prescription of (4.1) is spacetime supersymmetric.
5. Superstring Action in a Curved Background
In this section, the massless integrated vertex operator for the closed superstring will be used to construct a quantizable action for the superstring in a curved background. As a special case, a quantizable action will be constructed for the Type IIB superstring in an $`AdS_5\times S^5`$ background with Ramond-Ramond flux.
In bosonic string theory and in the Neveu-Schwarz sector of superstring theory, the action in a curved background (ignoring the Fradkin-Tseytlin term for dilaton coupling) can be constructed by โcovariantizingโ the massless closed string vertex operator with respect to target-space reparameterization invariance. As in and , this procedure can also be used here after constructing the massless closed string vertex operator from the โleft-rightโ product of two massless open string vertex operators of (3.1).
To do this, one first needs to introduce right-moving analogs of the worldsheet fields described in (2.1). The complete worldsheet action for the Type II superstring in a flat background in conformal gauge is
$$S=d^2z(\frac{1}{2}x^m\overline{}x_m+p_\alpha \overline{}\theta ^\alpha +\widehat{p}_{\widehat{\alpha }}\widehat{\theta }^{\widehat{\alpha }}+\frac{1}{2}(v^{ab}\overline{}u_{ab}+\widehat{v}^{ab}\widehat{u}_{ab})+\beta \overline{}\gamma +\widehat{\beta }\widehat{\gamma })$$
where $`\widehat{\lambda }^{\widehat{\alpha }}`$ is constructed from $`\widehat{\gamma }`$ and $`\widehat{u}_{ab}`$ in a manner similar to (2.1). Note that $`\widehat{\theta }^{\widehat{\alpha }}`$ and $`\widehat{\lambda }^{\widehat{\alpha }}`$ are independent of $`\theta ^\alpha `$ and $`\lambda ^\alpha `$, and are not related by complex conjugation. For the Type IIA superstring, the hatted spinor index has the opposite chirality to the unhatted spinor index while, for the Type IIB superstring, the hatted spinor index has the same chirality as the unhatted spinor index.
The action for the Type II superstring in a curved background obtained by โcovariantizingโ the massless closed superstring vertex operator with respect to target-space super-reparameterization invariance is<sup>6</sup> In the action for the superstring in a curved six-dimensional background, there are terms coming from the bosonic ghosts $`(u^\alpha ,v_\alpha )`$ described in footnote 3 which were incorrectly omitted from the action of . The correct action should have terms of the type $`d^2z(d_{\alpha j}+u_\alpha v^\beta D_{\beta j})E_M^{\alpha j}\overline{}Y^M+\mathrm{})`$, as well as a kinetic action for the $`(u^\alpha ,v_\alpha )`$ and $`(\widehat{u}^{\widehat{\alpha }},\widehat{v}_{\widehat{\alpha }})`$ ghosts.
$$S=d^2z[\frac{1}{2}(G_{MN}+B_{MN})y^M\overline{}y^N$$
$$+(d_\alpha +N_\alpha ^\beta D_\beta )E_M^\alpha \overline{}y^M+(\widehat{d}_{\widehat{\alpha }}+\widehat{N}_{\widehat{\alpha }}^{\widehat{\beta }}\widehat{D}_{\widehat{\beta }})E_M^{\widehat{\alpha }}y^M+(d_\alpha +N_\alpha ^\beta D_\beta )(\widehat{d}_{\widehat{\gamma }}+\widehat{N}_{\widehat{\gamma }}^{\widehat{\delta }}\widehat{D}_{\widehat{\delta }})P^{\alpha \widehat{\gamma }}$$
$$+\frac{1}{2}(v^{ab}\overline{}u_{ab}+\widehat{v}^{ab}\widehat{u}_{ab})+\beta \overline{}\gamma +\widehat{\beta }\widehat{\gamma }]$$
where $`y^M`$ parameterizes the curved superspace background, $`E_M^\alpha `$ and $`E_M^{\widehat{\alpha }}`$ are the spinor parts of the super-vierbein $`E_M^A`$, $`N_\alpha ^\beta N^{mn}(\gamma _{mn})_\alpha ^\beta `$ and $`\widehat{N}_{\widehat{\alpha }}^{\widehat{\beta }}\widehat{N}^{mn}(\gamma _{mn})_{\widehat{\alpha }}^{\widehat{\beta }}`$ with $`N_{mn}`$ being defined by (2.1) and $`\widehat{N}_{mn}`$ being defined similarly in terms of the hatted variables, and $`P^{\alpha \widehat{\gamma }}`$ is the superfield whose lowest components are the bispinor Ramond-Ramond field strengths. The operators $`D_\alpha `$ and $`\widehat{D}_{\widehat{\alpha }}`$ in (5.1) are understood to act on the superfield to their left, e.g. on $`E_M^\alpha `$, $`E_M^{\widehat{\alpha }}`$ or $`P^{\alpha \widehat{\gamma }}`$.
Note that the first line of (5.1) is identical to the Green-Schwarz action in a curved background, however, the second and third lines are crucial for quantization since they provide an invertible propagator for $`\theta ^\alpha `$ and $`\widehat{\theta }^{\widehat{\alpha }}`$. Furthermore, since there is no fermionic $`\kappa `$-symmetry which needs to be preserved, there is no problem with adding a Fradkin-Tseytlin term to (5.1) of the type $`\alpha ^{}d^2z\mathrm{\Phi }(x,\theta ,\widehat{\theta })R`$ where $`R`$ is the worldsheet curvature and $`\mathrm{\Phi }`$ is a scalar superfield whose lowest component is the dilaton.
When the background superfields satisfy their effective low-energy equations of motion, the action of (5.1) together with the Fradkin-Tseytlin term is expected to be conformally invariant where the left-moving stress tensor is
$$T=\frac{1}{2}G_{MN}y^My^N+(d_\alpha +N_\alpha ^\beta D_\beta )E_M^\alpha y^M+\frac{1}{2}v^{ab}u_{ab}+\beta \gamma +\alpha ^{}^2\mathrm{\Phi }.$$
Furthermore, the current $`\lambda ^\alpha d_\alpha `$ is expected to be holomorphic and nilpotent when the background superfields are on-shell.
5.1. Superstring Action in $`AdS_5\times S^5`$ background
In this subsection, the action of (5.1) will be explicitly constructed for the special case of the Type IIB superstring in an $`AdS_5\times S^5`$ background with Ramond-Ramond flux. As discussed in , this background can be conveniently described by a coset supergroup $`g`$ taking values in $`PSU(2,2|4)/SO(4,1)\times SO(5)`$ where the super-vierbein $`E_M^A`$ satisfies
$$E_M^Ady^M=(g^1dg)^A$$
and $`A=(c,\alpha ,\widehat{\alpha })`$ ranges over the 10 bosonic and 32 fermionic entries in the Lie-algebra valued matrix $`g^1dg`$. Furthermore, as discussed in , the only non-zero components of $`B_{AB}=E_A^ME_B^NB_{MN}`$ and $`P^{\alpha \widehat{\beta }}`$ are
$$B_{\alpha \widehat{\beta }}=B_{\widehat{\beta }\alpha }=\frac{1}{2ng_s}\delta _{\alpha \widehat{\beta }},P^{\alpha \widehat{\beta }}=ng_s\delta ^{\alpha \widehat{\beta }},$$
where $`n`$ is the value of the Ramond-Ramond flux, $`g_s`$ is the string coupling constant, and $`\delta _{\alpha \widehat{\beta }}=(\gamma ^{01234})_{\alpha \widehat{\beta }}`$ with $`01234`$ being the directions of $`AdS_5`$.
Plugging these background superfields into the action of (5.1), one finds
$$S_{AdS}=d^2z[\frac{1}{2}\eta _{cd}(g^1g)^c(g^1\overline{}g)^d\frac{1}{4ng_s}\delta _{\alpha \widehat{\beta }}[(g^1g)^\alpha (g^1\overline{}g)^{\widehat{\beta }}(g^1\overline{}g)^\alpha (g^1g)^{\widehat{\beta }}]$$
$$+d_\alpha (g^1\overline{}g)^\alpha +\widehat{d}_{\widehat{\alpha }}(g^1g)^{\widehat{\alpha }}+ng_s\delta ^{\alpha \widehat{\beta }}d_\alpha \widehat{d}_{\widehat{\beta }}$$
$$+N_{cd}(g^1\overline{}g)^{cd}+\widehat{N}_{cd}(g^1g)^{cd}+\frac{1}{2}(v^{ab}\overline{}u_{ab}+\widehat{v}^{ab}\widehat{u}_{ab})+\beta \overline{}\gamma +\widehat{\beta }\widehat{\gamma }]$$
where $`(g^1dg)^{cd}`$ are the $`SO(4,1)\times SO(5)`$ coset elements of $`g^1dg`$. This action is $`PSU(2,2|4)`$-invariant since under $`gMg\mathrm{\Omega }`$ with $`M`$ a global $`SU(2,2|4)`$ matrix and $`\mathrm{\Omega }`$ a local $`SO(4,1)\times SO(5)`$ matrix assumed to be close to the identity,
$$\delta (g^1dg)^A=([g^1dg,\mathrm{\Omega }])^A,\delta (g^1dg)^{cd}=([g^1dg,\mathrm{\Omega }])^{cd}+(d\mathrm{\Omega })^{cd},$$
$$\delta (\frac{1}{2}(v^{ab}\overline{}u_{ab}+\widehat{v}^{ab}\widehat{u}_{ab})+\beta \overline{}\gamma +\widehat{\beta }\widehat{\gamma })=N_{cd}(\overline{}\mathrm{\Omega })^{cd}\widehat{N}_{cd}(\mathrm{\Omega })^{cd},$$
where $`(d_\alpha ,\widehat{d}_{\widehat{\alpha }})`$ and $`(\lambda ^\alpha ,\widehat{\lambda }^{\widehat{\alpha }})`$ are defined to transform as Lorentz-covariant spinors under the local $`SO(4,1)\times SO(5)`$ transformation.
Because of the $`ng_s\delta ^{\alpha \widehat{\beta }}d_\alpha \widehat{d}_{\widehat{\beta }}`$ term, $`d_\alpha `$ and $`\widehat{d}_{\widehat{\alpha }}`$ are auxiliary fields which can be integrated out as was done in . Their auxiliary equations of motion are
$$d_\alpha =\frac{1}{ng_s}\delta _{\alpha \widehat{\beta }}(g^1g)^{\widehat{\beta }},\widehat{d}_{\widehat{\beta }}=\frac{1}{ng_s}\delta _{\alpha \widehat{\beta }}(g^1g)^\alpha ,$$
which can be substituted into (5.1) to give
$$S_{AdS}=d^2z[\frac{1}{2}\eta _{cd}(g^1g)^c(g^1\overline{}g)^d$$
$$\frac{3}{4ng_s}\delta _{\alpha \widehat{\beta }}(g^1\overline{}g)^\alpha (g^1g)^{\widehat{\beta }}\frac{1}{4ng_s}\delta _{\alpha \widehat{\beta }}(g^1g)^\alpha (g^1\overline{}g)^{\widehat{\beta }}$$
$$+N_{cd}(g^1\overline{}g)^{cd}+\widehat{N}_{cd}(g^1g)^{cd}+\frac{1}{2}(v^{ab}\overline{}u_{ab}+\widehat{v}^{ab}\widehat{u}_{ab})+\beta \overline{}\gamma +\widehat{\beta }\widehat{\gamma }].$$
Finally, one can perform the rescaling
$$E_M^c(ng_s)^1,E_M^\alpha (ng_s)^{\frac{1}{2}},E_M^{\widehat{\alpha }}(ng_s)^{\frac{1}{2}},$$
to obtain the action<sup>7</sup> Two quantizable actions have been proposed for the superstring in an $`AdS_3\times S^3`$ background with Ramond-Ramond flux . One of the actions includes eight left and right-moving $`\theta `$ coordinates and is based on the coset supergroup $`PSU(1,1|2)\times PSU(2|2)/SU(2)\times SU(2)`$ . For the reasons discussed in footnotes 3 and 6, this action requires the addition of terms involving the bosonic ghosts $`(u^\alpha ,v_\alpha )`$ and $`(\widehat{u}^{\widehat{\alpha }},\widehat{v}_{\widehat{\alpha }})`$. The other action includes four left and right-moving $`\theta `$ coordinates and is based on the supergroup $`PSU(2|2)`$ . This action does not involve the โharmonicโ constraints discussed in and therefore does not require the addition of any terms involving bosonic ghosts.
$$S_{AdS}=\frac{1}{n^2g_s^2}d^2z[\frac{1}{2}\eta _{cd}(g^1g)^c(g^1\overline{}g)^d$$
$$\frac{3}{4}\delta _{\alpha \widehat{\beta }}(g^1\overline{}g)^\alpha (g^1g)^{\widehat{\beta }}\frac{1}{4}\delta _{\alpha \widehat{\beta }}(g^1g)^\alpha (g^1\overline{}g)^{\widehat{\beta }}]$$
$$+d^2z[N_{cd}(g^1\overline{}g)^{cd}+\widehat{N}_{cd}(g^1g)^{cd}+\frac{1}{2}(v^{ab}\overline{}u_{ab}+\widehat{v}^{ab}\widehat{u}_{ab})+\beta \overline{}\gamma +\widehat{\beta }\widehat{\gamma }].$$
Except for the third line of (5.1) involving the pure spinor fields, this is precisely the action which was proposed in . Note that unlike the action proposed in , the action of (5.1) is straightforward to quantize and sigma model loop computations can be performed in the manner of .
6. Concluding Remarks
In this paper, a new formalism has been presented for covariantly quantizing the superstring. For the first time, vertex operators have been constructed and scattering amplitudes have been computed in a manifestly ten-dimensional super-Poincarรฉ invariant manner. A quantizable action has been proposed for the superstring in any curved background, including the $`AdS_5\times S^5`$ background with Ramond-Ramond flux.
There are various possible generalizations of the new formalism which should be possible. For example, one should be able to generalize the massless vertex operators to massive vertex operators and generalize the tree-level amplitude prescription to a multiloop amplitude prescription. One should also be able to construct physical vertex operators in the $`AdS_5\times S^5`$ background in a manner similar to the construction of $`AdS_3\times S^3`$ vertex operators in .
A more ambitious application would be to use the new formalism to construct a second-quantized superstring field theory. Note that the physical state condition $`QV=0`$ is easily generalized to the non-linear equation of motion $`QV+VV=0`$ where the $``$-product is defined as in . However, since the scattering amplitudes of section 4 are only spacetime-supersymmetric when the states satisfy the on-shell condition $`QV=0`$, it is unlikely that the action which produces this equation of motion will be manifestly spacetime-supersymmetric.
Perhaps the most important unresolved issue is to prove the equivalence between this formalism and the RNS formalism. In section 2, preliminary evidence for this equivalence came from comparing the gluon vertex operator in the two formalisms. Further evidence for this equivalence will now be shown by considering the pure spinor formalism in the โU(5) gaugeโ defined by setting $`u_{ab}=\theta _{ab}=0`$.
In this U(5) gauge, the operator $`๐z\lambda ^\alpha d_\alpha `$ reduces to $`๐z\gamma d_+`$. If one compares with the U(5)-invariant formalism of , it is natural to identify<sup>8</sup> In the following discussion, $`\gamma `$ always refers to $`\lambda ^+`$ and not to the RNS bosonic ghost. $`\gamma =e^{\frac{1}{2}(\rho i\sigma )}`$, or using the field redefinition of to RNS worldsheet variables, $`\gamma =ce^{\frac{3}{2}\varphi }\mathrm{\Sigma }^+`$ where $`\mathrm{\Sigma }^\alpha `$ is the RNS spin field. Since $`d_+=b\eta e^{\frac{3}{2}\varphi }\mathrm{\Sigma }_+`$ using this same field redefinition, one finds $`๐z\lambda ^\alpha d_\alpha =๐z\eta `$ in the U(5) gauge. So the restriction that $`๐z\lambda ^\alpha d_\alpha `$ commutes with physical states maps in this gauge to the restriction that physical states should be independent of the $`\xi `$ zero mode.
Note that the unintegrated vertex operator for super-Yang-Mills, $`U=\lambda ^\alpha A_\alpha `$, reduces in the U(5) gauge to
$$U=\gamma (A_++\theta ^aA_a+\frac{1}{2}\theta ^a\theta ^bW_{ab}+\frac{1}{12}ฯต_{abcde}\theta ^a\theta ^b\theta ^cF^{de}+\frac{1}{24}ฯต_{abcde}\theta ^a\theta ^b\theta ^c\theta ^d^eW^+)$$
where $`A_a`$ is the $`\overline{5}_1`$ component of the gluon gauge field and $`F^{ab}`$ is the $`10_2`$ component of the gluon field strength. Using the field redefinition of to write $`U`$ in terms of RNS variables where $`\theta ^a=e^{\frac{1}{2}\varphi }\mathrm{\Sigma }^a`$, one finds
$$U=c(e^{\frac{3}{2}\varphi }\mathrm{\Sigma }^+A_++e^\varphi \psi ^aA_a+\frac{1}{2}e^{\frac{1}{2}\varphi }\mathrm{\Sigma }^{ab}W_{ab}+\frac{1}{2}\psi _d\psi _eF^{de}+e^{\frac{1}{2}\varphi }:\mathrm{\Sigma }^{}\psi _e:^eW^+)$$
where $`:\mathrm{\Sigma }^{}\psi _e:`$ signifies $`lim_{yz}(yz)^{\frac{1}{2}}\mathrm{\Sigma }^{}(y)\psi _e(z)`$. Except for the term proportional to $`^eW^+`$, all of the terms in $`U`$ can be recognized as pieces of the RNS gluon and gluino vertex operators in various pictures.
Furthermore, if only the contribution from $`u_{ab}=0`$ is included, the integration prescription of (4.1) implies that
$$\lambda ^\alpha \lambda ^\beta \lambda ^\gamma f_{\alpha \beta \gamma }(z_r,k_r,\theta )(\frac{}{\theta })^5f_{+++}(z_r,k_r,\theta ),$$
i.e. $`\gamma ^3(\theta )^5=1`$ where $`(\theta )^5=(5!)^1ฯต_{abcde}\theta ^a\theta ^b\theta ^c\theta ^d\theta ^e`$. But under the field redefinition of , $`\gamma ^3(\theta )^5`$ maps to $`cc^2ce^{2\varphi }`$, as expected from $`\gamma ^3(\theta )^5=1`$.
So it appears reasonable that the U(5) gauge-fixed version of the new formalism is equivalent to the RNS formalism with a fixed choice of picture for the vertex operators. Removing the gauge-fixing condition, i.e. integrating over the $`u_{ab}`$ and $`\theta _{ab}`$ variables using the prescription of (4.1), might then be equivalent to summing over the different possible pictures in the RNS formalism. It would be very useful to be able to prove such an equivalence.
Acknowledgements: I would like to thank CNPq grant 300256/94-9 for partial financial support.
References
relax M.B. Green and J.H. Schwarz, Covariant Description of Superstrings, Phys. Lett. B136 (1984) 367. relax S. Carlip, Heterotic String Path Integrals with the Green-Schwarz Covariant Action, Nucl. Phys. B284 (1987) 365 ; R. Kallosh, Quantization of Green-Schwarz Superstring, Phys. Lett. B195 (1987) 369. relax G. Gilbert and D. Johnston, Equivalence of the Kallosh and Carlip Quantizations of the Green-Schwarz Action for the Heterotic String, Phys. Lett. B205 (1988) 273. relax W. Siegel, Classical Superstring Mechanics, Nucl. Phys. B263 (1986) 93. relax E. Sokatchev, Harmonic Superparticle, Class. Quant. Grav. 4 (1987) 237 ; E.R. Nissimov and S.J. Pacheva, Manifestly Super-Poincarรฉ Covariant Quantization of the Green-Schwarz Superstring, Phys. Lett. B202 (1988) 325; R. Kallosh and M. Rakhmanov, Covariant Quantization of the Green-Schwarz Superstring, Phys. Lett. B209 (1988) 233. relax S.J. Gates Jr, M.T. Grisaru, U. Lindstrom, M. Rocek, W. Siegel, P. van Nieuwenhuizen and A.E. van de Ven, Lorentz-Covariant Quantization of the Heterotic Superstring, Phys. Lett. B225 (1989) 44; R.E. Kallosh, Covariant Quantization of Type IIA,B Green-Schwarz Superstring, Phys. Lett. B225 (1989) 49; M.B. Green and C.M. Hull, Covariant Quantum Mechanics of the Superstring, Phys. Lett. B225 (1989) 57. relax D. Friedan, E. Martinec and S. Shenker, Conformal Invariance, Supersymmetry and String Theory, Nucl. Phys. B271 (1986) 93. relax N. Berkovits, Quantization of the Superstring with Manifest U(5) Super-Poincarรฉ Invariance, Phys. Lett. B457 (1999) 94, hep-th/9902099. relax P.S. Howe, Pure Spinor Lines in Superspace and Ten-Dimensional Supersymmetric Theories, Phys. Lett. B258 (1991) 141, Addendum-ibid.B259 (1991) 511; P.S. Howe, Pure Spinors, Function Superspaces and Supergravity Theories in Ten Dimensions and Eleven Dimensions, Phys. Lett. B273 (1991) 90. relax R. Metsaev and A. Tseytlin, Type IIB superstring action in $`AdS_5\times S^5`$ background, Nucl.Phys. B533 (1998) 109, hep-th/9805028. relax N. Berkovits, Covariant Quantization Of The Green-Schwarz Superstring in a Calabi-Yau Background, Nucl. Phys. B431 (1994) 258; N. Berkovits, A New Description Of The Superstring, Proceedings to VIII Jorge Swieca Summer School on Particles and Fields, p. 490, World Scientific Publishing, 1996, hep-th/9604123. relax N. Berkovits, Quantization of the Type II Superstring in a Curved Six-Dimensional Background, to appear in Nucl. Phys. B, hep-th/9908041. relax N. Berkovits and W. Siegel, Superspace Effective Actions for 4D Compactifications of Heterotic and Type II Superstrings, Nucl. Phys. B462 (1996) 213, hep-th 9510106. relax N. Berkovits, M. Bershadsky, T. Hauer, S. Zhukov and B. Zwiebach, Superstring Theory on $`AdS_2\times S^2`$ as a Coset Supermanifold, to appear in Nucl. Phys. B, hep-th/9907200. relax N. Berkovits, C. Vafa and E. Witten, Conformal Field Theory of AdS Background with Ramond-Ramond Flux, JHEP 9903 (1999) 018, hep-th/9902098. relax L. Dolan and E. Witten, Vertex Operators for $`AdS_3`$ Background with Ramond-Ramond Flux, JHEP 9911 (1999) 003, hep-th/9910205. relax E. Witten, Noncommutative Geometry and String Field Theory, Nucl. Phys. B268 (1986) 253. |
warning/0001/hep-ph0001328.html | ar5iv | text | # The most energetic particles in the Universe
## Acknowledgments
Work partially supported by ANPCyT, CONICET and Fundaciรณn Antorchas, Argentina. |
warning/0001/hep-th0001208.html | ar5iv | text | # Untitled Document
* Supported in part by the Robert A. Welch Foundation, N.S.F. grants DMS9706707 and PHY-9511632.
UTTG-06-99
hep-th/0001208
29 January 2000
Toric Calabi-Yau Fourfolds
Duality Between N=1 Theories and
*Divisors that Contribute to the Superpotential *
<sup>1,3</sup>Volker Braun<sup>1,3</sup>, Philip Candelas<sup>2,3</sup>, Xenia de la Ossa<sup>2,3</sup>
and
<sup>4</sup>Antonella Grassi<sup>4</sup>
<sup>1</sup>Institut fรผr Physik Humboldt Universitรคt Invalidenstrasse 110, 10115 Berlin, Germany <sup>2</sup>Mathematical Institute Oxford University 24-29 St. Gilesโ Oxford OX1 3LB, England
<sup>3</sup>Theory Group Department of Physics University of Texas Austin, TX 78712, USA <sup>4</sup>Department of Mathematics 209 South 33rd St. University of Pennsylvania Philadelphia, PA 19104, USA
ABSTRACT
We study issues related to $`F`$-theory on Calabi-Yau fourfolds and its duality to heterotic theory for CalabiโYau threefolds. We discuss principally fourfolds that are described by reflexive polyhedra and show how to read off some of the data for the heterotic theory from the polyhedron. We give a procedure for constructing examples with given gauge groups and describe some of these examples in detail. Interesting features arise when the local pieces are fitted into a global manifold. An important issue is how to compute the superpotential explicitly. Witten has shown that the condition for a divisor to contribute to the superpotential is that it have arithmetic genus 1. Divisors associated with the short roots of non-simply laced gauge groups do not always satisfy this condition while the divisors associated to all other roots do. For such a โdissidentโ divisor we distinguish cases for which $`\chi (๐ช_D)>1`$ corresponding to an $`X`$ that is not general in moduli (in the toric case this corresponds to the existence of non-toric parameters). In these cases the โdissidentโ divisor $`D`$ does not remain an effective divisor for general complex structure. If however $`\chi (๐ช_D)0`$, then the divisor is general in moduli and there is a genuine instability.
Contents
1. Introduction
2. Toric Preliminaries
3. Heterotic Structure
3.1 The Fibrations
4. Construction of $`X`$
4.1 The KLRY Spaces
4.2 Constructing the Fan for $`\stackrel{~}{X}`$
4.3 Extending the Groups
4.4 New groups
5. The Yukawa Couplings and Mori Cone
5.1 the Yukawa Couplings
5.2 The Mori Cone of $`X`$
5.3 Volumes of the Divisors
6. The Superpotential
6.1 Characterization of the Divisors Contributing to the Worldsheet Instantons
6.2 Comparison with the Heterotic Superpotential
6.3 Andreasโ Examples
6.4 Comparison with $`d=3`$ Dimensional Yang-Mills Theory
6.5 New Features for a Non-Simply Laced Group
A. Appendix: Geometry of $`Z`$
A.1 The Divisors
A.2 Projection to $`B^Z`$
B. Appendix: The Divisors for the Spaces $`Y^\pm `$
1. Introduction
This paper is devoted to a number of issues pertaining to the compactification of F-theory on CalabiโYau fourfolds. For fourfolds, $`X`$, of particular structure there are believed to be interesting dualities
$$\text{F}[X]=\text{IIB}[B^X]=\text{Het}[Z^X,V^X],$$
$`(1.1)`$
that relate F-theory on $`X`$ to IIB string theory compactified on a (non CalabiโYau) threefold $`B^X`$ and also to a heterotic compactification on a CalabiโYau threefold $`Z^X`$ with vector bundle $`V^X`$. The relation between heterotic compactification on threefolds and CalabiโYau fourfolds that these dualities entail is particularly interesting since it offers the hope of insight into the important but so far poorly understood $`(0,2)`$ compactifications of the heterotic string. Relation (1.1) suggests that heterotic theories on CalabiโYau threefolds are, in some sense, classified by CalabiโYau fourfolds. We will here concern ourselves largely with CalabiโYau fourfolds that are themselves described by reflexive polyhedra. Thus a class of heterotic theories on CalabiโYau threefolds are described by reflexive five dimensional polyhedra. The question that we seek to answer is to what extent we can read off the data for the heterotic theory from this polyhedron.
A basic question is how to read off the the threefold of the heterotic theory from the data. This is accomplished by observing that the dual polyhedron for the heterotic threefold is obtained via a certain projection, which we explain, of the dual polyhedron of the fourfold. This toric description is dual to the picture of the heterotic threefold in terms of a maximal degeneration of the fourfold. We will argue that, at least for the models we consider here, the relation between the fourfold and the heterotic threefold involves mirror symmetry in an interesting way:
$$\begin{array}{cc}\hfill \text{F}[X]& =\text{Het}[Z,V^X],\stackrel{~}{X}=(\stackrel{~}{Z},\mathrm{IP}{}_{1}{}^{})\hfill \\ \multicolumn{2}{c}{}\\ \hfill \text{F}[\stackrel{~}{X}]& =\text{Het}[\stackrel{~}{Y},V^{\stackrel{~}{X}}],X=(Y,\mathrm{IP}{}_{1}{}^{})\hfill \end{array}$$
where tildes are used to denote mirror manifolds and $`X=(Y,\mathrm{IP}{}_{1}{}^{})`$, for example, denotes that $`X`$ is a fibration by a (CalabiโYau) threefold $`Y`$ fibered over a $`\mathrm{IP}_1`$.
Another important question is how to compute explicitly the $`F`$-theory superpotential on a given fourfold and try to understand this as a superpotential for the corresponding heterotic theory. When we do this several problems arise. At this point we are unable to give a complete answer; we are presenting here various examples and related observations.
An immediate question that arises on computing a superpotential from the fourfold $`X`$ is that the superpotential is a cubic function of the Kรคhler parameters of $`X`$, while the heterotic superpotential is a function of the volumes of the curves, and thus linear. The comparison makes sense (so far) only for smooth divisors $`D`$ contributing to worldsheet instantons; curiously this question has not arisen in examples that have been previously considered owing to the fact that in these examples the cubic expression in the Kรคhler parameters is in fact trilinear. We discuss these previous calculations in Section 7 (see also Section 5.2). It seems that the resolution is to regard the volume of the elliptic fibre as an infinitessimal and in this limit the volumes of the divisors are indeed trilinear.
Other questions arise when we compare the $`F`$-theory superpotential corresponding to gauge groups, with a $`d=3`$ dimensional Yang-Mills theory, following . Their work deals with divisors corresponding to gauge groups , arising as resolution of the Weierstrass model of the elliptic fourfold: each (irreducible) divisor mapping via the elliptic fibration to the same surface can be identified with a node in the extended Dynkin diagram of the group. A key point in their computation is that all (or none) of the irreducible divisors of a Dynkin diagram contribute to the superpotential. In Section 4 we consider explicit examples where โmixed configurationsโ occur, that is when some divisors contribute to the superpotential, and some do not (following the criterion of ). This happens when the gauge group is not simply laced ($`SO(odd),Sp(n),G_2,F_4`$). A non simply laced group arises via the action of monodromy on a group that is simply laced. In such a case the divisor (two divisors for the case of $`F_4`$) that arise through the identification of divisors of the simply laced group may have $`\chi (๐ช_D)1`$, violating the condition to contribute to the superpotential. We shall refer to such divisors as being dissident. In Section 6 we show that there are many such examples. If $`\chi (๐ช_D)>1`$, $`X`$ turns out not to be general in moduli (in the toric case this correspond to the existence of non-toric parameters) and that the โdissidentโ divisor $`D`$ will not remain an effective divisor for general values of the complex moduli space. If $`\chi (๐ช_D)0`$, then the divisor will be general in moduli and we cannot reproduce Vafaโs computations. On the other hand we show that in the toric case $`h^{2,1}(X)>0`$: this always leads to an interesting structure on the heterotic dual (see ).
Many of our observations will be recognisable to those who have developed a local description of the duality in terms of branes wrapping the singular fibres of the fourfold seen as an elliptic fibration over the threefold base $`B^X`$. On the other hand our observation is that interesting features arise precisely as a result of trying to fit the local pieces together into a global manifold. In particular there is a tendency for the gauge group of the effective theory to โgrowโ since maintaining the fibration structure of the fourfold $`X`$, when we put together the local singularities, requires further resolution of these singularities.
The layout of this paper is the following: in $`\mathrm{\S }`$2 we gather together some expressions that compute the Hodge numbers of a CalabiโYau fourfold $`X`$, corresponding to a reflexive polyhedron $``$, as well as the arithmetic genus of its divisors in terms of the combinatoric properties of $``$. We find also interesting expressions, whose significance we do not properly understand, that relate the arithmetic genera of the divisors of a manifold to the arithmetic genera of the divisors of the mirror. In $`\mathrm{\S }`$3 we introduce our basic model fourfold $`X`$ and describe its structure. This model is perhaps as simple as one can have without taking a model that is completely trivial. The gauge group that we have is $`SU(2)\times G_2`$. In $`\mathrm{\S }`$4 we explain the structure of this model. We set out originally to construct a model with group $`SU(2)\times SU(3)`$ however maintaining the fibration structure of our particular choice $`X`$ required extending the $`SU(3)`$ to $`G_2`$ in a way that we explain in detail. We also show how to extend the gauge group by taking the degenerate fibers to have a more complicated structure. The new element here is the process by which we show how to divide the fans in such a way as to maintain the fibration. In $`\mathrm{\S }`$5 we analyze the Yukawa coupling and the structure of the Mori cone; for the Mori Cone we implement a procedure advocated in . It is clear however that our implementation of this procedure leads to a cone that is too large. This is similar to a recent result in . In $`\mathrm{\S }`$7 we discuss the computation of the superpotential and compare this calculation to that for previous examples.
2. Toric Preliminaries
We draw together here some essential results that we will need. CalabiโYau fourfolds, $`X`$, for which the holonomy is SU(4) rather than a subgroup have a Hodge diamond whose top half is of the form
$$\begin{array}{cccccccccc}& & & & & 1& & & & \\ & & & & 0& & 0& & & \\ & & & 0& & h^{11}& & 0& & \\ & & 0& & h^{21}& & h^{21}& & 0& \\ & 1& & h^{31}& & h^{22}& & h^{31}& & 1.\end{array}$$
There is also a linear relation between the Hodge numbers
$$h^{22}=2(22+2h^{11}+2h^{31}h^{21}).$$
Thus there are three independent Hodge numbers and the Euler number is given in terms of the Hodge numbers by
$$\chi _E(X)=6(8+h^{11}+h^{31}h^{21}).$$
$`(2.1)`$
The three independent Hodge numbers $`h^{31}`$, $`h^{21}`$ and $`h^{11}`$ may be determined directly from the Newton polyhedron, $`\mathrm{\Delta }`$, and its dual, $``$, (our convention is that the fan of $`X`$ is that fan over the faces of $``$) via the expressions \[10-12\]
$$\begin{array}{cc}\hfill h^{31}& =\text{pts}(\mathrm{\Delta })\underset{\text{codim}\theta =1}{}\text{int}(\theta )+\underset{\text{codim}\theta =2}{}\text{int}(\theta )\text{int}(\stackrel{~}{\theta })6\hfill \\ \multicolumn{2}{c}{}\\ \hfill h^{11}& =\text{pts}()\underset{\text{codim}\stackrel{~}{\theta }=1}{}\text{int}(\stackrel{~}{\theta })+\underset{\text{codim}\stackrel{~}{\theta }=2}{}\text{int}(\stackrel{~}{\theta })\text{int}(\theta )6\hfill \\ \multicolumn{2}{c}{}\\ \hfill h^{21}& =\underset{\text{codim}\theta =3}{}\text{int}(\stackrel{~}{\theta })\text{int}(\theta )\hfill \end{array}$$
$`(2.2)`$
In these expressions $`\text{pts}(\mathrm{\Delta })`$ denotes the number of lattice points in $`\mathrm{\Delta }`$, while $`\theta `$ runs over the faces of $`\mathrm{\Delta }`$, $`\text{int}(\theta )`$ denotes the number of lattice points strictly interior to $`\theta `$ and $`\stackrel{~}{\theta }`$ denotes the face of $``$ dual to $`\theta `$. It is interesting to note that Batyrev gives also an expression for the โphysicistโs Euler numberโ in terms of the volumes of the faces of the polyhedra
$$\chi _E^{}=\underset{\text{codim}\theta =2}{}\text{vol}(\theta )\text{vol}(\stackrel{~}{\theta })2\underset{\text{codim}\theta =3}{}\text{vol}(\theta )\text{vol}(\stackrel{~}{\theta })+\underset{\text{codim}\theta =4}{}\text{vol}(\theta )\text{vol}(\stackrel{~}{\theta })$$
$`(2.3)`$
in which the volumes are normalized such that the volume of a fundamental lattice simplex is unity (rather than $`1/d!`$ in $`d`$ dimensions). If the fourfold is nonsingular then these two expressions for the Euler number yield the same result. There are cases however of fourfolds $`X_{mnp}`$ defined by (4.1) for which $`\chi _E^{}\chi _E`$ indicating that these fourfolds are singular (for all values of their parameters).
Each point $`q`$ that is not interior to a facet (codimension one face) corresponds to a divisor $`D_q`$ of $`X`$. Such a point $`q`$ defines a face $`\stackrel{~}{\theta }_q`$ of $``$, the unique face to which it is interior<sup>1</sup><sup>1</sup> A point $`q`$ that is not the interior point is contained in some face of $``$. We ask whether $`q`$ is interior to this face or if it lies in the boundary. If it lies in the boundary then it lies in a face of lower dimension and we ask again if it lies in the interior to this face or in the boundary. Proceeding in this way we either arrive at the unique face to which $`q`$ is interior or we find that $`q`$ is a vertex. It is therefore convenient to adopt the convention that the vertices are interior to themselves., and in virtue of duality also a face $`\theta _q`$ of $`\mathrm{\Delta }`$. Klemm et al. establish an elegant expression for the arithmetic genus, $`\chi _q`$, of $`D_q`$:
$$\chi _q=\chi \left(๐ช(D_q)\right)=1(1)^{\text{dim}(\stackrel{~}{\theta }_q)}\text{int}(\theta _q).$$
$`(2.4)`$
An observation, made already in , is that manifolds containing divisors of arithmetic genus one are abundant. These exist whenever $``$ has faces that (a) have interior points and that (b) are dual to faces of $`\mathrm{\Delta }`$ that have no interior points. Klemm et al. study and interesting class of manifolds that we will term the KLRY spaces, and whose construction we review in Sect. 4. All these manifold and all their mirrors have such faces<sup>2</sup><sup>2</sup> The number of these divisors is by (2.4) always finite. The infinite number found in is associated with the fact the manifold under consideration is not in the class that we consider here, since it cannot be realised as a hypersurface in a toric variety given by a single equation, and hence by a single polyhedron. It would be of interest to study this example from the toric perspective. The formalism however is much less developed for the cases that require more than one equation..
Witten shows that $`\chi (๐ช(D_q))=1`$ is a necessary condition to contribute to the superpotential, while
$$h^0(๐ช(D_q))=1,h^i(๐ช(D_q))=0,i>0$$
is a sufficient condition.
It is of interest to note that in the expressions (2.2) for the Hodge numbers there occur crossterms that involve both $``$ and $`\mathrm{\Delta }`$
$$\begin{array}{cc}\hfill \overline{\delta }& =\underset{\text{codim}\stackrel{~}{\theta }=2}{}\text{int}(\theta )\text{int}(\stackrel{~}{\theta })=\underset{q(\text{dim}\stackrel{~}{\theta }=3)}{}(\chi _q1)\hfill \\ \multicolumn{2}{c}{}\\ \hfill h^{21}& =\underset{\text{codim}\stackrel{~}{\theta }=3}{}\text{int}(\theta )\text{int}(\stackrel{~}{\theta })=\underset{q(\text{dim}\stackrel{~}{\theta }=2)}{}(\chi _q1)\hfill \\ \multicolumn{2}{c}{}\\ \hfill \delta & =\underset{\text{codim}\stackrel{~}{\theta }=4}{}\text{int}(\theta )\text{int}(\stackrel{~}{\theta })=\underset{q(\text{dim}\stackrel{~}{\theta }=1)}{}(\chi _q1)\hfill \\ \multicolumn{2}{c}{}\end{array}$$
$`(2.5)`$
The quantities $`\delta `$ and $`\overline{\delta }`$ are respectively the number of non-toric deformations of $`X`$ and its mirror, while $`2h^{21}`$ is the number of cohomology classes of three cycles. Since the KLRY formula(2.4) shows that all $`q`$โs interior to a given $`\stackrel{~}{\theta }_q`$ have the same arithmetic genus it follows that these crossterms are also equal to $`\pm (\chi _q1)`$ as given. If $`\text{dim}(\stackrel{~}{\theta })=4`$ then $`\theta _q`$ is a vertex and $`\text{int}(\theta _q)=1`$ by convention so
$$\underset{\text{codim}(\stackrel{~}{\theta })=1}{}(\chi _q1)=\underset{\text{codim}(\stackrel{~}{\theta })=1}{}\text{int}(\stackrel{~}{\theta }_q)$$
which counts the divisors of $`\mathrm{IP}_{}`$ that do not intersect the hypersurface. A dual relation obtains also for $`\text{dim}(\stackrel{~}{\theta })=0`$. We summarize these relations in the following table:
| $`\text{dim}(\stackrel{~}{\theta }_q)`$ | $`\text{dim}(\theta _q)`$ | $`(\chi _q1)`$ contributes to |
| --- | --- | --- |
| | | |
| $`4`$ | $`0`$ | $`q`$โs that are not divisors |
| $`3`$ | $`1`$ | $`\stackrel{~}{\delta }`$ |
| $`2`$ | $`2`$ | $`h^{21}`$ |
| $`1`$ | $`3`$ | $`\delta `$ |
| $`0`$ | $`4`$ | irrelevant monomials |
| | | |
Note that if a divisor has $`\chi 1`$ then this divisor contributes to precisely one of the quantities in the third column of the table.
There are also curious relations whose significance we do not understand such as
$$\underset{q\stackrel{~}{\theta }}{}(\chi _q1)=\underset{\stackrel{~}{q}\theta }{}(\chi _{\stackrel{~}{q}}1)$$
that relate $`\chi 1`$ for divisors in the manifold and its mirror. Another such relation is
$$\underset{q}{}(\chi _q1)=\frac{\chi _E(X)}{6}\left(\text{pts}()+\text{pts}(\mathrm{\Delta })\right)+4$$
or equivalently
$$\underset{q}{}\chi _q=\frac{\chi _E(X)}{6}\text{pts}(\mathrm{\Delta })+3$$
where $``$ denotes $``$ less the interior point and owing to our convention $`\chi _q1=1`$ for $`q`$ in a codimension one face of $``$.
3. Heterotic Structure
3.1. Fibrations A first statement of the relation between the manifolds that appear in (1.1) may be made in terms of fibrations. A poor but useful notation that we employ is to write $`(,)`$ for a manifold that is a fibration over a base $``$ with generic fiber $``$. The notation is poor since a manifold is not uniquely specified. There may well be different manifolds that can be realized as fibrations over a given base with the same generic fiber, the difference being due to the manner in which the fibers degenerate over subvarieties of the base. This said, the relation between the manifolds of (1.1) is believed to be the following:
$$\begin{array}{cc}\hfill X=(,B^X)=(K3^Y,& B^Z),Z=(,B^Z)\hfill \\ \hfill B^X=(\mathrm{IP}{}_{1}{}^{},B^Z),& K3^Y=(,\mathrm{IP}{}_{1}{}^{})\hfill \end{array}$$
$`(3.1)`$
with $``$ denoting an elliptic curve and $`K3^Y`$ denoting a $`K3`$ manifold. The superfix $`Y`$ refers to another CalabiโYau threefold to which we shall refer as we proceed. In other words, $`X`$ is an elliptic fibration over a base $`B^X`$, with $`B^X`$ a $`\mathrm{IP}_1`$-fibration over a two dimensional base $`B^Z`$. The manifold $`Z`$ of the heterotic compactification is then an elliptic fibration over this same two dimensional base. The second representation of $`X`$ states that it is also a fibration over $`B^Z`$ with fiber an elliptic K3.
The purpose of the present paper is, in part, to study the relations between these manifolds in the toric context for which some degree of control is afforded by the relation between CalabiโYau manifolds and reflexive polyhedra and the observation of that the fibration structure of a manifold specified by such a polyhedron is visible in the polyhedron. Although the nature of the bundle $`V^X`$ is not properly understood in terms of the toric data, nevertheless some information is available. For example, the structure group $`G^X`$ of $`V^X`$ can be read off from the polyhedron . Another issue that we study in the toric context is that of the existence of divisors that give rise to a superpotential through non-perturbative effects. Proceeding loosely, the integral points, $`q`$, of the dual polyhedron, $`^X`$, of $`X`$ are in direct correspondence with the divisors of $`X`$. Moreover, Klemm et al. have formulated a simple and elegant criterion that distinguishes the points corresponding to the divisors of arithmetic genus one. The toric context, though not general, permits a study of examples and a certain systematization which we find useful.
As noted above a reflexive polyhedron corresponding to a CalabiโYau manifold that is a fibration $`(,)`$ has a slice corresponding to the dual polyhedron of the fiber $``$ and this enables us to establish a standard coordinate system for the polyhedra. It is perhaps easiest to see this at work in an example. The first column of Table 3.1 lists the integral points of the dual polyhedron of an interesting example which we shall denote by $`X`$ throughout this article.
$`^X`$ ( $`x_1`$, $`x_2`$, $`x_3`$, $`x_4`$, $`x_5`$ ) ( -1, 0, 0, 2, 3 ) ( 0, -1, 0, 2, 3 ) ( 0, 0, -1, 2, 3 ) ( 0, 0, -1, 1, 2 ) ( 0, 0, 0, -1, 0 ) ( 0, 0, 0, 0, -1 ) ( 0, 0, 0, 0, 0 ) ( 0, 0, 0, 0, 1 ) ( 0, 0, 0, 1, 1 ) ( 0, 0, 0, 1, 2 ) ( 0, 0, 0, 2, 3 ) ( 0, 0, 1, 1, 2 ) ( 0, 0, 1, 2, 3 ) ( 0, 0, 2, 2, 3 ) ( 0, 0, 1, 1, 1 ) ( 0, 1, 2, 2, 3 ) ( 0, 1, 3, 2, 3 ) ( 1, 0, 4, 2, 3 )
$`^Y`$ $`(\mathrm{\hspace{0.33em}0},x_2,x_3,x_4,x_5)`$ ( -1, 0, 2, 3 ) ( 0, -1, 2, 3 ) ( 0, -1, 1, 2 ) ( 0, 0, -1, 0 ) ( 0, 0, 0, -1 ) ( 0, 0, 0, 0 ) ( 0, 0, 0, 1 ) ( 0, 0, 1, 1 ) ( 0, 0, 1, 2 ) ( 0, 0, 2, 3 ) ( 0, 1, 1, 2 ) ( 0, 1, 2, 3 ) ( 0, 2, 2, 3 ) ( 0, 1, 1, 1 ) ( 1, 2, 2, 3 ) ( 1, 3, 2, 3 )
$`^Z`$ $`(x_1,x_2,\widehat{x_3},x_4,x_5)`$ ( -1, 0, 2, 3 ) ( 0, -1, 2, 3 ) ( 0, 0, -1, 0 ) ( 0, 0, 0, -1 ) ( 0, 0, 0, 0 ) ( 0, 0, 0, 1 ) ( 0, 0, 1, 1 ) ( 0, 0, 1, 2 ) ( 0, 0, 2, 3 ) ( 0, 1, 2, 3 ) ( 1, 0, 2, 3 )
$`^{K3^Y}`$ $`(\mathrm{\hspace{0.33em}0},\mathrm{\hspace{0.33em}0},x_3,x_4,x_5)`$ ( -1, 2, 3 ) ( -1, 1, 2 ) ( 0, -1, 0 ) ( 0, 0, -1 ) ( 0, 0, 0 ) ( 0, 0, 1 ) ( 0, 1, 1 ) ( 0, 1, 2 ) ( 0, 2, 3 ) ( 1, 1, 2 ) ( 1, 2, 3 ) ( 2, 2, 3 ) ( 1, 1, 1 )
$`^{K3^Z}`$ $`(x_1,\widehat{x_2},\widehat{x_3},x_4,x_5)`$ ( -1, 2, 3 ) ( 0, -1, 0 ) ( 0, 0, -1 ) ( 0, 0, 0 ) ( 0, 0, 1 ) ( 0, 1, 1 ) ( 0, 1, 2 ) ( 0, 2, 3 ) ( 1, 2, 3 )
$`^{}`$ $`(\mathrm{\hspace{0.33em}0},\mathrm{\hspace{0.33em}0},\mathrm{\hspace{0.33em}0},x_4,x_5)`$ $`(\widehat{x_1},\widehat{x_2},\widehat{x_3},x_4,x_5)`$ ( -1, 0 ) ( 0, -1 ) ( 0, 0 ) ( 0, 1 ) ( 1, 1 ) ( 1, 2 ) ( 2, 3 )
Table 3.1: The dual polyhedron for $`X`$ with the data of the associated fibrations. The $``$โs on the upper level are linked by a series of injections while those on the lower level are related by projections. We take the Cartesian coordinates $`(x_1,x_2,x_3,x_4,x_5)`$ for the $`\mathrm{IR}^5`$ in which the polyhedron is embedded. The points that lie in the hyperplane $`\{x_1=0\}`$ form $`^Y`$, the dual polyhedron of a CalabiโYau threefold, $`Y`$. These points are listed in the upper table of column two of Table 3.1. The points with $`\{x_1=x_2=0\}`$ form $`^{K3^Y}`$, the dual polyhedron of a $`K3`$ surface associated to $`Y`$. The points with $`\{x_1=x_2=x_3=0\}`$ form the dual polyhedron, $`^{}`$, of the Weierstrass torus $`=\mathrm{IP}^{(1,2,3)}[6]`$. Finally the three points with $`\{x_1=x_2=x_3=x_4=0\}`$ are the dual polyhedron of a zero-dimensional CalabiโYau manifold<sup>3</sup><sup>3</sup> We can think of a zero-dimensional CalabiโYau manifold as $`\mathrm{IP}_1[2]`$ with equation $`\xi _1^2+\xi _1\xi _2+\xi _2^2=0`$. The Newton polyhedron associated with this equation consists of three points in a straight line. After a change of coordinates these become the points $`x=1,0,1`$. This trivial reflexive polyhedron is self-dual. The lower route through the table is realised by making the indicated projections. The coordinates with hats are projected out in this process. Thus $`^X^Z`$ corresponds to the projection $`(x_1,x_2,x_3,x_4,x_5)(x_1,x_2,x_4,x_5)`$. Note that the fibration manifests additional structure, not only does the dual polyhedron of $``$ appear as a slice but also as a projection onto the last two coordinates. This is related to the fact that $`^{}`$ is self dual.
We would like now to discuss the bases of the elliptic CalabiโYau and $`K3`$ fibrations which we denote by $`B^X`$, $`B^Y`$ and $`B^Z`$ respectively. We have noted that we can see the dual polyhedron of the fiber, $``$, of the fibration $`X=(,B^X)`$ as the slice $`\{x_1=x_2=x_3=0\}`$ of $`^X`$. The base $`B^X`$ may also be seen as the projection onto the first three coordinates. This gives us the points of the first column of Table 3.2 the rays from the origin through these points yield the fan of $`B^X`$. Note also that $`B^X`$ can be obtained not only as the projection to the first three coordinates but also as the slice $`\{x_4=2,x_5=3\}`$, which is a three-face of $`^X`$. It is one of the observations of that the roles of injections and projections are interchanged by mirror symmetry so the fact that $`^{}`$ and $`B^X`$ are visible both as injections and projections has the consequence that the mirror, $`\stackrel{~}{X}`$, of $`X`$ is also an elliptic fibration $`\stackrel{~}{X}=(,B^{\stackrel{~}{X}})`$, where in this relation $`B^{\stackrel{~}{X}}`$ denotes the base of the mirror fibration and we write $``$ for the fiber in place of $`\stackrel{~}{}`$ since $`^{}`$ is self mirror. $`B^X`$ is not CalabiโYau so there is no notion of a mirror of $`B^X`$. For the CalabiโYau fibration $`X=(Y,\mathrm{IP}{}_{1}{}^{})`$ we have already noted that $`^Y`$ is the slice $`\{x_1=0\}`$ of $`^X`$. The fan of the $`\mathrm{IP}_1`$ consists of the three points $`x_1=1,0,1`$ obtained by projecting $`^X`$ onto the first coordinate. The threefold $`Y`$ is itself an elliptic fibration, $`Y=(,B^Y)`$, over a base $`B^Y`$ whose toric data are obtained by projecting $`^Y`$ to its first two coordinates. This gives the upper table of the second column of Table 3.2. As for the $`K3`$-fibration $`X=(K3^Y,B^Z)`$, we have seen the $`K3^Y`$ as the slice $`\{x_1=x_2=0\}`$ and we see $`B^Z`$ as the projection onto the first two coordinates, the result being the lower table of the second column of Table 3.2. The base of the $`K3`$-fibration is in fact the base of the fibration $`Z=(,B^Z)`$ as we shall see. Note however that $`Y`$ and the $`K3`$ are not projections onto any slices so we might not expect the mirror of $`X`$ to be a CalabiโYau and $`K3`$-fibration, although we shall see presently that it is.
$`B^X`$ $`(x_1,x_2,x_3,\widehat{x_4},\widehat{x_5})`$ $`(x_1,x_2,x_3,\mathrm{\hspace{0.33em}2},\mathrm{\hspace{0.33em}3})`$ ( -1, 0, 0 ) ( 0, -1, 0 ) ( 0, 0, -1 ) ( 0, 0, 0 ) ( 0, 0, 1 ) ( 0, 0, 2 ) ( 0, 1, 2 ) ( 0, 1, 3 ) ( 1, 0, 4 )
$`B^Y`$ $`(\mathrm{\hspace{0.33em}0},x_2,x_3,\widehat{x_4},\widehat{x_5})`$ $`(\mathrm{\hspace{0.33em}0},x_2,x_3,\mathrm{\hspace{0.33em}2},\mathrm{\hspace{0.33em}3})`$ ( -1, 0 ) ( 0, -1 ) ( 0, 0 ) ( 0, 1 ) ( 0, 2 ) ( 1, 2 ) ( 1, 3 )
$`B^Z`$ $`(x_1,x_2,\widehat{x_3},\widehat{x_4},\widehat{x_5})`$ $`(x_1,x_2,\widehat{x_3},\mathrm{\hspace{0.33em}2},\mathrm{\hspace{0.33em}3})`$ ( -1, 0 ) ( 0, -1 ) ( 0, 0 ) ( 0, 1 ) ( 1, 0 )
$`\mathrm{IP}_{}^{Z}{}_{1}{}^{}`$ $`(0,x_2,\widehat{x_3},\widehat{x_4},\widehat{x_5})`$ $`(0,x_2,\widehat{x_3},\mathrm{\hspace{0.33em}2},\mathrm{\hspace{0.33em}3})`$ ( -1 ) ( 0 ) ( 1 )
$`\mathrm{IP}_{}^{Y}{}_{1}{}^{}`$ $`(x_1,\widehat{x_2},\widehat{x_3},\widehat{x_4},\widehat{x_5})`$ $`(x_1,\widehat{x_2},\widehat{x_3},\mathrm{\hspace{0.33em}2},\mathrm{\hspace{0.33em}3})`$ ( -1 ) ( 0 ) ( 1 )
Table 3.2: The points corresponding to the bases of the fibrations. To summarize thus far, the coordinates of $`^X`$ relate $`B^X`$, $`B^Y`$, $`B^Z`$ and $``$ as follows
$$\begin{array}{cc}& B^X\hfill \\ \multicolumn{2}{c}{}\\ & \stackrel{}{}\stackrel{}{}\hfill \\ \multicolumn{2}{c}{}\\ \hfill \stackrel{}{x}=& (\underset{}{x_1,x_2},x_3,x_4,x_5)\hfill \\ \multicolumn{2}{c}{}\\ & B^Z\hfill \\ \multicolumn{2}{c}{}\\ & \underset{}{}\hfill \\ \multicolumn{2}{c}{}\\ & B^Y\hfill \end{array}$$
The reader wishing to acquire dexterity with seeing the various fibrations should check from Table 3.2 that
$$B^X=(\mathrm{IP}{}_{1}{}^{},B^Z)\text{and}B^X=(B^Y,\mathrm{IP}{}_{1}{}^{}).$$
We want to find $`Z=(,B^Z)`$ in $`^X`$ so we need to project out the coordinate $`x_3`$. Thus $`^Z`$ can be realized by performing the projection $`(x_1,x_2,x_3,x_4,x_5)(x_1,x_2,x_4,x_5)`$. This yields the sixth column of Table 3.1. Another way of stating this is that $`Z`$ is being identified as the base of a fibration. This statement seems to involve mirror symmetry in a nontrivial way. The fibrations being
$$\begin{array}{cc}\hfill X& =(Y,\mathrm{IP}{}_{1}{}^{})=(\text{ZZ}_2,\mathrm{IP}(^Z))\hfill \\ \hfill \stackrel{~}{X}& =(\stackrel{~}{Z},\mathrm{IP}{}_{1}{}^{})=(\text{ZZ}_2,\mathrm{IP}(\mathrm{\Delta }^Y))\hfill \end{array}$$
where $`\mathrm{IP}(^Z)`$ and $`\mathrm{IP}(\mathrm{\Delta }^Y)`$ denote the toric manifolds corresponding to the fans over the faces of $`^Z`$ and $`\mathrm{\Delta }^Y`$ respectively.
A class of fourfolds $`X`$ with the structure $`X=(,B^X)`$ with $`B^X=(\mathrm{IP}{}_{1}{}^{},B^Z)`$ has been discussed in . The idea is to take $`B^X`$ of the form $`(\mathrm{IP}{}_{1}{}^{},(\mathrm{IP}{}_{1}{}^{},\mathrm{IP}{}_{1}{}^{}))`$. The inner fibration $`(\mathrm{IP}{}_{1}{}^{},\mathrm{IP}{}_{1}{}^{})`$ is taken to be a fiber bundle, the Hirzebruch surface<sup>4</sup><sup>4</sup> Notice that this already shows how loose the notation $`(\mathrm{IP}{}_{1}{}^{},\mathrm{IP}{}_{1}{}^{})`$ is since even for the case of a fiber bundle there is one of these for each integer $`m`$. $`\mathrm{IF}_m`$. The wrapping of the outer $`\mathrm{IP}_1`$ is specified by two further integers corresponding to the wrapping of this $`\mathrm{IP}_1`$ about each of the other two. The resulting manifold is denoted by $`\mathrm{IF}_{mnp}`$ in . For many values of these integers, it is possible to define in terms of toric data manifolds $`X=(,\mathrm{IF}_{mnp})`$ in such a way that $`X`$ is CalabiโYau. By this we mean here that we can associate a Newton polyhedron with $`X`$ and this polyhedron has the property of being reflexive.
These reflexive polyhedra have interesting structure corresponding to the fact that in addition to the structure (1.1) we have also
$$X=(Y,\mathrm{IP}{}_{1}{}^{}),Y=(K3^Y,\mathrm{IP}{}_{1}{}^{}),K3^Y=(,\mathrm{IP}{}_{1}{}^{}),=(\text{ZZ}_2,\mathrm{IP}{}_{1}{}^{}),$$
$`(3.2)`$
with $`Y`$ a CalabiโYau threefold. Now, this structure manifests itself in the dual polyhedron $`^X`$ of $`X`$ in a very simple way: $`^X`$ contains a codimension one slice which is $`^Y`$, the dual polyhedron of $`Y`$ and this structure is repeated; $`^Y`$ contains $`^{K3^Y}`$, the dual polyhedron of the $`K3`$ surface as a slice and $`^{K3^Y}`$ contains $`^{}`$ the triangle of the Weierstrass polynomial as a slice. Finally, $`^{}`$ contains a line with three points corresponding to a zero dimensional CalabiโYau manifold. In other words we have a series of injections
$$^X^Y^{K3^Y}^{}^{\text{ZZ}_2}.$$
$`(3.3)`$
A point that is important is that this structure imposes a natural coordinate system on the polyhedron. $`^X`$ is five-dimensional; within it is a four-plane containing $`^Y`$; within this a three-plane containing $`^{K3}`$; within this a two-plane containing $`^{}`$; and finally within this a line corresponding to the zero dimensional CalabiโYau $`^{\text{ZZ}_2}`$.
There is a further important property of this class of manifolds which is that there is also a hierarchy of projections that relate $`X`$ to $`Z`$:
$$^X^Z^{K3^Z}^{}^{\text{ZZ}_2}$$
$`(3.4)`$
It is one of the observations of that the roles of injections and projections are interchanged by mirror symmetry so the mirror of each such $`X`$ has also a structure analogous to (3.3) and (3.4) with the replacements $`Y\stackrel{~}{Z}`$ and $`Z\stackrel{~}{Y}`$. Here and in the following tildes are used to denote the mirror of a given manifold.
In these notes we shall be primarily concerned with the KLRY spaces and their mirrors (these spaces will be discussed in details in $`\mathrm{\S }`$4). The polyhedra for these classes are very different, the KLRY spaces have dual polyhedra that are small, typically with 20โ70 points, while their Newton polyhedra (which are the dual polyhedra of the mirrors) are large with typically 20,000โ70,000 points. The structure of these fourfolds suggests a precise specification of the threefold $`Z^X`$, given $`X`$, by defining $`Z`$ in terms of its mirror
$$\text{F}[X]=\text{Het}[Z,V^X],\stackrel{~}{X}=(\stackrel{~}{Z},\mathrm{IP}{}_{1}{}^{}),$$
$`(3.5)`$
where in the second relation $`\stackrel{~}{X}`$ and $`\stackrel{~}{Z}`$ are the mirrors of $`X`$ and $`Z`$ respectively. As we show in $`\mathrm{\S }`$2, the relation $`\stackrel{~}{X}=(\stackrel{~}{Z},\mathrm{IP}{}_{1}{}^{})`$, and hence $`Z`$, can be given a precise meaning in virtue of the natural projections mentioned above. For $`X`$ the KLRY space $`X_{mnp}`$, this relation gives what we would expect $`Z=Z_m=(,\mathrm{IF}_m)`$, the elliptic fibration of the Hirzebruch surface that is familiar from . However, for $`X`$ the mirror of a KLRY space, the fact that we obtain a sensible definition of $`Z`$ in this way is far from trivial.
The occurrence of divisors that lead to superpotentials turns out to be a rather involved subject. A first observation that is perhaps counterintuitive given is that divisors of the fourfold with arithmetic genus one are ubiquitous at least in the class of fourfolds that we study. (Of more than 3000 manifolds all have these divisors and all their mirrors have them also.) What is less clear is how to deal with compactifications to four dimensions for which we are interested in divisors of arithmetic genus one that are of the form $`\pi ^1(R)`$ where $`\pi `$ denotes the projections $`\pi :XB^X`$ onto the base of the fibration and $`R`$ denotes a divisor of $`B^X`$. Here the situation is clearest when $`\pi ^1(R)`$ consists of a single component which is a divisor of $`X`$ of arithmetic genus one. Frequently however, the preimage $`\pi ^1(R)`$ consists of several irreducible components with non-trivial intersection. These preimages are precisely the ones that give rise to the gauge group $`G^X`$ of the heterotic model.
4. Construction of X
4.1. The KLRY Spaces The KLRY spaces, $`X_{mnp}`$, provide many examples of elliptically fibered fourfolds with the structure (3.1). These spaces are defined by toric data
$$Q_{mnp}=\left(\begin{array}{ccccccccc}1& 1& m& 0& p& 0& 2(m+p+2)& 3(m+p+2)& 0\\ 0& 0& 1& 1& n& 0& 2(m+2)& 3(m+2)& 0\\ 0& 0& 0& 0& 1& 1& 4& 6& 0\\ 0& 0& 0& 0& 0& 0& 2& 3& 1\end{array}\right)$$
$`(4.1)`$
with the three integers having the ranges $`1m,n12`$ and $`0p\frac{1}{2}n(m+2)`$. What is meant by this is, of course, that we have 9 coordinates $`(\xi _1,\xi _2,\mathrm{},\xi _9)`$ that are identified under 4 scaling symmetries with weights given by the rows of $`Q`$. Now the allowed monomials (those that have the same multidegree as the fundamental monomial $`\xi _1\xi _2\mathrm{}\xi _9`$) are points in a 9-dimensional space. However in virtue of the scaling relations these points lie in a 5-dimensional plane. Thus the monomials associated with (4.1) give rise to a 5-dimensional Newton polyhedron, $`\mathrm{\Delta }`$. This polyhedron is reflexive and so corresponds to a CalabiโYau fourfold.
Perhaps the simplest way to motivate the particular fourfold that we study is to describe first a simple but inconsistent model associated with the manifold $`X_{234}`$ which we denote by $`\stackrel{~}{X}`$ to save writing. The model is inconsistent for our purposes since, despite the fact that it appears to be an elliptic fibration by construction nevertheless, as we shall see, it fails to be an elliptic fibration. We shall discuss this carefully in the following however the point at issue is whether the embedding space $`\mathrm{IP}^{}`$ is an elliptic fibration. Such a fibration is expressed torically if every cone in the fan of $`\stackrel{~}{X}`$ projects to some cone of the fan of the base. This is what fails for $`\stackrel{~}{X}`$. Repairing the fibration leads to a more complicated but viable model.
To begin however consider then the dual polyhedron $`\stackrel{~}{}`$ corresponding to $`\stackrel{~}{X}`$ shown below. The divisors $`E_1`$ and $`\stackrel{~}{E}`$ are associated with an $`SU(2)`$ gauge group. The fact that the groups are as seen from the polyhedron can be verified by performing a calculation in the Cox coordinate ring. In our case, this has been checked by A. Klemm. As we have already stressed, the fibration $`\stackrel{~}{X}B`$ is important for us and it is related to the projection onto the first three coordinates of the points of $`\stackrel{~}{}`$.
| Vertex | $`\stackrel{~}{}`$ | | | | | | | Divisor |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| $`V_1`$ | ( | -1, | 0, | 0, | 2, | 3 | ) | |
| $`V_2`$ | ( | 0, | -1, | 0, | 2, | 3 | ) | |
| $`V_3`$ | ( | 0, | 0, | -1, | 2, | 3 | ) | |
| $`V_4`$ | ( | 0, | 0, | 0, | -1, | 0 | ) | |
| $`V_5`$ | ( | 0, | 0, | 0, | 0, | -1 | ) | |
| $`\frac{1}{2}(V_3+E_1)`$ | ( | 0, | 0, | 0, | 2, | 3 | ) | $`B`$ |
| $`\frac{1}{3}(V_2+V_4+V_7)`$ | ( | 0, | 0, | 1, | 1, | 2 | ) | $`\stackrel{~}{E}`$ |
| $`\frac{1}{2}(V_2+F)`$ | ( | 0, | 0, | 1, | 2, | 3 | ) | $`E_1`$ |
| $`\frac{1}{2}(V_1+V_6)`$ | ( | 0, | 1, | 2, | 2, | 3 | ) | $`F`$ |
| $`V_7`$ | ( | 0, | 1, | 3, | 2, | 3 | ) | $`G`$ |
| $`V_6`$ | ( | 1, | 2, | 4, | 2, | 3 | ) | $`Y`$ |
4.2. Constructing the Fan for $`\stackrel{~}{X}`$ The polyhedron $`\stackrel{~}{}`$ is not a simplex. However, we can start by taking the cones over its faces
$$\begin{array}{cc}\hfill \{& V_2V_4V_5V_6V_7,V_2V_3V_4V_5V_6,V_1V_2V_3V_4V_5,V_1V_2V_3V_4V_6V_7,\hfill \\ & V_1V_2V_3V_5V_6V_7,V_1V_3V_4V_5V_6,V_1V_2V_4V_5V_7,V_1V_4V_5V_6V_7\}\hfill \end{array}$$
Two of these cones $`V_1V_2V_3V_4V_6V_7`$ and $`V_1V_2V_3V_5V_6V_7`$ are not simplicial. They correspond to facets of $`\stackrel{~}{}`$ that share a common 3-face $`V_1V_2V_3V_6V_7`$. We can see how to perform a triangulation by drawing the $`V_2V_3V_7F`$-plane as well as the $`V_2V_4V_7`$ 2-face. The following rules effect the triangulation:
$$\begin{array}{cc}\hfill V_1V_6& \{V_1F,FV_6\}\hfill \\ \hfill V_2V_3V_7F& \{V_2V_3B,V_2BE_1,V_2E_1V_7,BE_1V_7,BFV_7,V_3BF\}\hfill \\ \hfill V_2V_4V_7& \{V_2\stackrel{~}{E}V_4,V_4\stackrel{~}{E}V_7,V_7\stackrel{~}{E}V_2\}\hfill \end{array}$$
It is necessary to check that the $`\stackrel{~}{X}`$ is actually an elliptic fibration, that is, that the map $`\pi :\stackrel{~}{X}B`$ is smooth. We may ensure this by requiring that each cone of the fan for $`\stackrel{~}{X}`$ should project onto some cone of the fan for the base $`B`$. As we will show now, this is not the case for $`\stackrel{~}{}`$. The problem arises in connection with the triangulation of the three face $`(x_1,x_2,x_3,2,3)`$ corresponding to $`\mathrm{\Sigma }^B`$. In Figure4.1, we draw the two plane $`(0,x_2,x_3,2,3,)`$ that lies within this face. The cones of $`\mathrm{\Sigma }^{\stackrel{~}{X}}`$ intersect this plane in the regions indicated. The problem comes from the cone from the origin of $`\stackrel{~}{}`$ (which is out of the plane) which interects this plane in the triangle $`V_2E_1G`$. If we project this cone onto the plane clearly it projects onto the union of the cone generated by $`V_1`$ and $`E_1`$ and the cone generated by $`E_1`$ and $`G`$. The simplest way to fix the problem is to add the point $`E_2(0,0,2,2,3)`$ to $`\stackrel{~}{}`$ as in Figure 4.1b. This divides the troublesome cone in two so that the fan for $`\stackrel{~}{X}`$ now projects nicely.
Figure 4.1: The two-plane $`(0,x_2,x_3,2,3)`$ of $``$. In (a) the plane is given for $`\stackrel{~}{X}`$. The bad cone is subdivided in (b) corresponding to the improved manifold $`X`$. The cones now project properly onto the fan (c).
$`Y^{}`$ $`C_1`$ $`F`$ $`G`$ $`Y^{}`$ $`C_1`$ $`F`$ $`G`$ $`E_1`$ $`E_2`$ When we take the convex hull of $`\stackrel{~}{}\{E_2\}`$ we find that the new polyhedron contains also the point $`E_3(0,0,1,1,1)`$. Moreover, the point $`\stackrel{~}{E}`$ now lies in the interior of a codimension one face. In this way, the $`SU(2)`$ associated with $`\{E_1,\stackrel{~}{E}\}`$ is replaced by a $`G_2`$ associated with $`\{E_1,E_2,E_3\}`$.
At the level of practical calculation, note that the preimage of the ray $`(0,0,1)`$ of the base of $`\stackrel{~}{X}`$ is the divisor $`E_1+\stackrel{~}{E}`$. It follows that $`(E_1+\stackrel{~}{E})^4`$ should vanish since the intersection calculation pulls back from the intersection calculation on the base. This consistency check fails for the fan for $`\stackrel{~}{X}`$.
We are not yet quite done with the changes. In order to enforce this condition, we have to add the point $`C_2(1,0,1,1,2)`$. The data for our consistent fourfold is displyed in Table 4.1. Note that $`\{C_1,C_2\}`$ correspond to an additional $`SU(2)`$ gauge group. We have added the divisor $`C_2`$, which could have been omitted, in order to show how we may build up the group. We make some further comments about how to build up the groups in $`\mathrm{\S }`$4.3 below. Table 4.1 summarizes the polyhedron and divisors for $`X`$.
| Relation to vertices | $`\chi `$ | $`^X`$ | | | | | | | Divisor |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| $`V_1`$ | $`0`$ | ( | -1, | 0, | 0, | 2, | 3 | ) | $`Y^+`$ |
| $`V_2`$ | $`0`$ | ( | 0, | -1, | 0, | 2, | 3 | ) | $`Y^{}=F+G`$ |
| $`V_3`$ | $`1^{}`$ | ( | 0, | 0, | -1, | 2, | 3 | ) | $`C_1`$ |
| $`V_9`$ | $`1^{}`$ | ( | 0, | 0, | -1, | 1, | 2 | ) | $`C_2`$ |
| $`V_4`$ | $`89`$ | ( | 0, | 0, | 0, | -1, | 0 | ) | $`2H+E_3C_2`$ |
| $`V_5`$ | $`368`$ | ( | 0, | 0, | 0, | 0, | -1 | ) | $`3H+E_3C_2`$ |
| $`\frac{1}{2}(V_3+E_1)`$ | $`1^{}`$ | ( | 0, | 0, | 0, | 2, | 3 | ) | $`B`$ |
| $`\frac{1}{2}(V_3+2E_2)`$ | $`1^{}`$ | ( | 0, | 0, | 1, | 2, | 3 | ) | $`E_1`$ |
| $`\frac{1}{2}(V_1+V_6)`$ | $`1^{}`$ | ( | 0, | 0, | 2, | 2, | 3 | ) | $`E_2`$ |
| $`\frac{1}{2}(V_5+E_2)`$ | $`0`$ | ( | 0, | 0, | 1, | 1, | 1 | ) | $`E_3=C_1+C_2(E_1+2E_2+2F+3G+4Y^+)`$ |
| $`V_7`$ | $`1^{}`$ | ( | 0, | 1, | 2, | 2, | 3 | ) | $`F`$ |
| $`V_8`$ | $`1^{}`$ | ( | 0, | 1, | 3, | 2, | 3 | ) | $`G`$ |
| $`V_6`$ | $`0`$ | ( | 1, | 0, | 4, | 2, | 3 | ) | $`Y^+`$ |
| $`h^{11}=8,h^{31}=2897,h^{21}=1,h^{22}=11662,\chi _E=17472`$ | | | | | | | | | |
| $`H=B+C_1+C_2+E_1+E_2+2F+2G+2Y^+`$ | | | | | | | | | |
Table 4.1: The divisors for the manifold $`X`$.
Figure 4.2: A sketch of how the various divisors intersect the base $`B`$ of the elliptic fibration showing also the degenerate fibers corresponding to the groups $`SU_2`$ and $`G_2`$. The surfaces shown as $`E_3Y^+`$ are really a single connected surface. The โcomponentsโ that are shown meet in pairs. The surface $`C_2Y^+`$ is ruled by quadrics which degenerate, exceptionally, into a pair of lines. This explains the $`1/2`$ that appears in the relation $`\mathrm{}^3=\frac{1}{2}C_2FY^+`$.
$`B^Z`$
$`Y^+`$
$`\begin{array}{c}\\ & G\\ & \end{array}`$
$`\begin{array}{cc}& B\\ \\ \end{array}`$
$`\begin{array}{ccc}& & \\ & F\end{array}`$
$`E_1E_2G`$
$`E_2Y^+`$
$`E_3Y^+\{`$
$`C_1Y^+`$
$`C_2Y^+`$
ยฟFrom we also see that
$$\begin{array}{c}\text{ }h^{k,0}(E_1)=h^{k,0}(E_2)=h^{k,0}(C_j)=h^{k,0}(F)=h^{0,0}(G)=0\text{for}k=1,2,3\text{ }\hfill \\ \text{ }\text{and}h^{0,0}(E_i)=h^{0,0}(C_j)=h^{0,0}(F)=h^{0,0}(G)=1,\text{while}h^{2,0}(E_3)=0\text{and}h^{1,0}(E_3)=1.\text{ }\hfill \end{array}$$
For our model $`X`$ our construction of the fan proceeds similarly to the case of $`\stackrel{~}{X}`$. We begin with the cones over the facets of the polyhedron that is the convex hull of the vertices $`V_1,V_2,\mathrm{},V_8`$. This yields the cones
$$\begin{array}{cc}\hfill \{& V_4V_5V_6V_7V_8,V_2V_3V_4V_5V_6,V_1V_3V_4V_5V_6,V_1V_2V_3V_4V_5,V_1V_2V_3V_4V_6V_7V_8,\hfill \\ & V_1V_2V_3V_5V_6V_7V_8,V_1V_4V_5V_6V_7,V_1V_2V_4V_5V_8,V_1V_4V_5V_7V_8,V_2V_4V_5V_6V_8\}\hfill \end{array}$$
There are two facets which are not simplices, $`V_1V_2V_3V_4V_6V_7V_8`$ and $`V_1V_2V_3V_5V_6V_7V_8`$, and these facets have a common 3-face $`V_1V_2V_3V_6V_7V_8=V_1V_6Y^{}C_1FG`$ which corresponds to the fan for the base $`B`$ of the elliptic fibration (see Table 4.1). To see how to perform a triangulation we make reference to Figure 4.1 which depicts the two-plane $`V_1V_2V_3V_6V_7V_8`$ that lies within this three-face. By associating divisors to the points of the figure and noting that $`V_1`$ and $`V_6`$ are the only points of the three plane that do not lie in the two plane and that these points are joined by a line that passes through $`F`$. We see that the following rules effect the triangulation:
$$\begin{array}{cc}\hfill V_1V_6& \{V_1F,FV_6\}\hfill \\ \hfill V_2V_3V_7V_8F& \{V_2V_3B,V_2BE_1,V_2E_1V_8,E_1V_7V_8,BE_1V_7,BFV_7,V_3BF\}\hfill \\ \hfill V_5V_8& \{V_5E_3,E_3V_8\}\hfill \end{array}.$$
This yields a simplicial fan. Finally, we insert $`V_9`$ which lies in the cone $`V_3V_4V_5`$ by means of the rule:
$$V_3V_4V_5\{V_3V_4V_9,V_4V_5V_9,V_3V_5V_9\}.$$
The result is a fan of 54 cones $`\mathrm{\Sigma }^X=V_1\mathrm{\Sigma }^YV_6\mathrm{\Sigma }^Y`$, where $`\mathrm{\Sigma }^Y`$ denotes the fan of 27 cones
$$\begin{array}{cc}\hfill \{& BE_1V_2V_4,BFC_1V_4,BV_2C_1V_4,E_3E_1V_2V_5,BE_1V_2V_5,BFC_1V_5,BV_2C_1V_5,E_3V_2V_4V_5,\hfill \\ & BFV_4G,BE_1V_4G,BFV_5G,E_3E_1V_5G,BE_1V_5G,E_3V_4V_5G,FV_4V_5G,E_3E_1V_2E_2,\hfill \\ & E_3V_2V_4E_2,E_1V_2V_4E_2,E_3E_1GE_2,E_3V_4GE_2,E_1V_4GE_2,FC_1V_4C_2,V_2C_1V_4C_2,\hfill \\ & FC_1V_5C_2,V_2C_1V_5C_2,FV_4V_5C_2,V_2V_4V_5C_2\}\hfill \end{array}$$
and $`V_1\mathrm{\Sigma }^Y`$ denotes the set obtained by appending $`V_1`$ to each cone of $`\mathrm{\Sigma }^Y`$ (and similarly for $`V_6\mathrm{\Sigma }^Y`$).
Figure 4.3: The polyhedron for $`K^Y`$. The horizontal triangle $`^{}`$ divides the polyhedron into a top and a bottom. The extended Dynkin diagrams corresponding to the groups $`G_2`$ and $`SU(2)`$ are seen as the blue lines.
Figure 4.4: The two faces of $`^{K^Y}`$ that admit more than one triangulation.
We can in fact express the combinatorics of the fan in a better way. To do this we write each cone as a product and a fan as a sum of cones. Thus with this understanding $`\mathrm{\Sigma }^Y=BE_1V_2V_4+BFC_1V_4+\mathrm{}+V_2V_4V_5C_2`$ and $`\mathrm{\Sigma }^X=(V_1+V_6)\mathrm{\Sigma }^Y`$. Now with this notation we may express $`\mathrm{\Sigma }^Y`$ in the form
$$\mathrm{\Sigma }^Y=(F+V_2)\mathrm{\Sigma }^C+(G+V_2)\mathrm{\Sigma }^E+FG\mathrm{\Sigma }^{}$$
where $`\mathrm{\Sigma }^C`$ and $`\mathrm{\Sigma }^E`$ denote the fans over the faces of the half-polyhedra corresponding to the $`SU(2)`$ and $`G_2`$. and $`\mathrm{\Sigma }^{}`$ denotes the fan over the faces of $`^{}`$.
$$\begin{array}{cc}\hfill \mathrm{\Sigma }^C& =BC_1V_4+C_1C_2V_4+BC_1V_5+C_1C_2V_5+C_2V_4V_5\hfill \\ \hfill \mathrm{\Sigma }^E& =E_1E_2E_3+BE_1V_4+E_1E_2V_4+E_2E_3V_4+BE_1V_5+E_1E_3V_5+E_3V_4V_5\hfill \\ \hfill \mathrm{\Sigma }^{}& =BV_4+BV_5+V_4V_5.\hfill \end{array}$$
Could we have chosen different fans? The answer is yes however if we restrict to fans that project to the fan for $`B^X`$ then there are just four choices and these are related by flops corresponding to the fact that two of the faces of $`^{K^Y}`$ each admit two different triangulations as in Figure 4.4. The yukawa couplings for these four fans are however the same which indicates that the flops affect the embedding space but not the hypersurface $`X`$.
4.3. Extending the Groups It is of course also possible to extend the gauge group by adding points to $``$. We think of this as building up the half-polyhedra that project down onto the divisors of $`B`$ and wish to show how this may be implemented on the fan. Suppose we begin by seeking to build up a half-polyhedron over the divisor $`F`$. We change notation by denoting $`F`$ by $`F_1`$, so that the new points are $`F_2`$, $`F_3`$ etc. and by denoting $`G`$ by $`G_1`$ since we will want also to add a group over $`G`$. Our starting point is the fan
$$\mathrm{\Sigma }^Y=(F_1+V_2)\mathrm{\Sigma }^{\text{bot}}+(G_1+V_2)\mathrm{\Sigma }^{\text{top}}+F_1G_1\mathrm{\Sigma }^{}$$
One checks that the new points $`F_2`$, $`F_3`$,โฆ all lie in the cone $`F_1V_4V_5`$ of $`\mathrm{\Sigma }^Y`$. This cone occurs as a face in two of the terms in $`\mathrm{\Sigma }^Y`$. It occurs both in $`F_1\mathrm{\Sigma }^{}`$ and $`F_1\mathrm{\Sigma }^C`$ since both $`\mathrm{\Sigma }^C`$ and $`\mathrm{\Sigma }^{}`$ contain the cone $`V_4V_5`$. Let $`\mathrm{\Sigma }^{F_1}`$ denote the trivial half-polyhedron over $`F_1`$, that is the half-polyhedron with no extra points, and denote by $`\mathrm{\Sigma }^F`$ the half-polyhedron corresponding to the new group. We will write $`\mathrm{\Sigma }^{F_1}\mathrm{\Sigma }^F`$ for the process of extending the fan to the fan over the faces of the new half-polyhedron.
For the term $`F_1\mathrm{\Sigma }^{}`$ we observe that
$$F_1\mathrm{\Sigma }^{}=\mathrm{\Sigma }^{F_1}\mathrm{\Sigma }^F.$$
While for the term $`F_1V_4V_5`$ in $`F_1\mathrm{\Sigma }^C`$ we note that
$$F_1V_4V_5=\mathrm{\Sigma }^{F_1}F_1B(V_4+V_5)$$
so we have
$$\begin{array}{cc}\hfill F_1\mathrm{\Sigma }^C& =F_1\left(\mathrm{\Sigma }^CC_2V_4V_5\right)+C_2(F_1V_4V_5)\hfill \\ & =F_1\left(\mathrm{\Sigma }^CC_2V_4V_5\right)+C_2\left(\mathrm{\Sigma }^{F_1}F_1B(V_4+V_5)\right)\hfill \\ & F_1\left(\mathrm{\Sigma }^CC_2V_4V_5\right)+C_2\left(\mathrm{\Sigma }^FF_1B(V_4+V_5)\right).\hfill \end{array}$$
In this way we see that $`\mathrm{\Sigma }^Y\stackrel{~}{\mathrm{\Sigma }}^Y`$ with
$$\stackrel{~}{\mathrm{\Sigma }}^Y=(F_1+V_2)\mathrm{\Sigma }^C+(G_1+V_2)\mathrm{\Sigma }^E+(G_1+C_2)\mathrm{\Sigma }^FF_1C_2\mathrm{\Sigma }^{}.$$
Note that the function of the last term is to remove terms that are present in $`F_1\mathrm{\Sigma }^C`$ so there are really no minus signs in this expression.
Now let us add a group over $`G_1`$. The new points all lie in the cone $`G_1V_4V_5`$ which is contained in the terms $`G_1\mathrm{\Sigma }^F`$ and $`G_1\mathrm{\Sigma }^E`$. In $`\mathrm{\Sigma }^F`$ there is a single cone of the form $`F_jV_4V_5`$ and we denote this divisor $`F_j`$ by $`F_{\text{max}}`$. For $`G_1\mathrm{\Sigma }^F`$ we write
$$\begin{array}{cc}\hfill G_1\mathrm{\Sigma }^F& =G_1(\mathrm{\Sigma }^FF_{\text{max}}V_4V_5)+F_{\text{max}}\left(\mathrm{\Sigma }^{G_1}G_1B(V_4+V_5)\right)\hfill \\ & G_1(\mathrm{\Sigma }^FF_{\text{max}}V_4V_5)+F_{\text{max}}\left(\mathrm{\Sigma }^GG_1B(V_4+V_5)\right)\hfill \\ & =G_1\mathrm{\Sigma }^F+F_{\text{max}}\mathrm{\Sigma }^GF_{\text{max}}G_1\mathrm{\Sigma }^{}.\hfill \end{array}$$
Similarly
$$G_1\mathrm{\Sigma }^EG_1\mathrm{\Sigma }^E+E_3\mathrm{\Sigma }^GE_3G_1\mathrm{\Sigma }^{}.$$
In this way we arrive at a fan
$$\begin{array}{cc}\hfill \stackrel{~}{\stackrel{~}{\mathrm{\Sigma }}}& =(F_1+V_2)\mathrm{\Sigma }^C+(G_1+V_2)\mathrm{\Sigma }^E+(G_1+C_2)\mathrm{\Sigma }^F+(E_3+F_{\text{max}})\mathrm{\Sigma }^G\hfill \\ & (E_3G_1+F_{\text{max}}G_1+C_2F_1)\mathrm{\Sigma }^{}.\hfill \end{array}$$
4.4. New groups The groups may be extended by adding points to the polyhedron and we have carried out this procedure to extend the group $`G_2`$. A non-simply laced group results from the effect of monodromy on a simply laced group. Some of the divisors for the simply laced group are identified under the monodromy. The resulting divisor(s) of the non-simply laced group are the ones that may have $`\chi 1`$. It seems that while it is frequently the case that these divisors that result from identification under monodromy have $`\chi 1`$ that this is not always the case. Our first extension $`G_2SO(7)`$ leads to a group all of whose divisors have $`\chi =1`$. however as we extend the group further to $`SO(9)`$, $`F_4`$, $`SO(11)`$ and $`SO(13)`$ we find that these cases all exhibit dissident divisors. The data is summarized by the following table.
| Apparent Group | $`h^{11}`$ | $`h^{31}`$ | $`h^{22}`$ | $`h^{21}`$ | $`\delta `$ | $`\overline{\delta }`$ | $`\chi `$ ofp dissident(s) |
| --- | --- | --- | --- | --- | --- | --- | --- |
| | | | | | | | |
| $`G_2`$ | $`8`$ | $`2897`$ | $`11662`$ | $`1`$ | $`0`$ | $`0`$ | $`\{0\}`$ |
| $`SO(7)`$ | $`9`$ | $`2895`$ | $`11660`$ | $`0`$ | $`0`$ | $`0`$ | no dissidents |
| $`SO(9)`$ | $`10`$ | $`2894`$ | $`11660`$ | $`0`$ | $`2`$ | $`0`$ | $`\{3\}`$ |
| $`F_4`$ | $`10`$ | $`2894`$ | $`11660`$ | $`0`$ | $`4`$ | $`0`$ | $`\{3,3\}`$ |
| $`SO(10)`$ | $`11`$ | $`2869`$ | $`11564`$ | $`0`$ | $`0`$ | $`0`$ | no dissidents |
| $`SO(11)`$ | $`11`$ | $`2869`$ | $`11564`$ | $`0`$ | $`12`$ | $`0`$ | $`\{13\}`$ |
| $`SO(13)`$ | $`13`$ | $`2787`$ | $`11244`$ | $`0`$ | $`25`$ | $`0`$ | $`\{26\}`$ |
| $`E_{7b}`$ | $`14`$ | $`2790`$ | $`11236`$ | $`12`$ | $`0`$ | $`0`$ | $`\{11\}`$ |
| $`E_8`$ | $`15`$ | $`2825`$ | $`11292`$ | $`56`$ | $`0`$ | $`0`$ | $`\{55\}`$ |
We will discuss in $`\mathrm{\S }`$6 the contribution to the superpotential from the corresponding divisors; here we make some further comments on the table:
$`\underset{ยฏ}{G_2\text{}}`$ The dissident divisor, $`E_3`$ corresponds to the end of the Dynkin diagram away from the extending root. This divisor is the one that corresponds to the three nodes of the Dynkin diagram for $`SO(8)`$ that are identified under a $`\text{ZZ}_3`$ monodromy. The dissident divisor contributes one to $`h^{21}`$, and so also to $`h^{12}`$ and hence contains two three-cycles.
$`\underset{ยฏ}{SO(7)\text{}}`$ All the divisors have arithmetic genus unity even though the group is not simply laced.
$`\underset{ยฏ}{SO(9)\text{}}`$ There is a dissident divisor again corresponding to the end of the Dynkin diagram away from the extending root. This is the node that corresponds to the two nodes of the $`SO(10)`$ that are identified under a $`\text{ZZ}_2`$ monodromy. In this case the dissident divisor contributes to $`\delta `$, the number of non-toric parameters.
$`\underset{ยฏ}{F_4\text{}}`$ There are now two dissidents at the end of the Dynkin diagram away from the extending root which correspond to the nodes of the $`E_6`$ that are identified under a $`\text{ZZ}_2`$ monodromy. The dissidents contribute to $`\delta `$. The Hodge numbers of the manifolds corresponding to $`SO(9)`$ and $`F_4`$ are the same suggesting that they are in fact the same manifold. The number of non-toric parameters is less for $`SO(9)`$ than for $`F_4`$ suggesting that $`SO(9)`$ gives the better description and that the true group is perhaps $`SO(8)`$.
$`\underset{ยฏ}{SO(10)\text{}}`$ We include this group for purposes of comparison with $`SO(11)`$. The group $`SO(10)`$ is simply laced so all the nodes of the Dynkin diagram correspond to divisors with arithmetic genus unity. For this case there are no non-toric parameters.
$`\underset{ยฏ}{SO(11)\text{}}`$ This case is similar to $`SO(9)`$. There is a dissident divisor which contributes to $`\delta `$. The Hodge numbers for this manifold are the same as those for $`SO(10)`$ which suggests that under a generic deformation the group becomes $`SO(10)`$.
$`\underset{ยฏ}{SO(13)\text{}}`$ This case is interesting and exhibits a new phenomenon which it has in common with the following two examples. The group is not simply laced nevertheless all the nodes of the Dynkin diagram correspond to divisors with arithmetic genus unity. The dissident divisor, in the toric description, arises not from the group per se but because the half polyhedron that projects down to a ray in the fan of $`B^X`$ contains the divisor $`\{0,0,1,0,0\}`$ as a point interior to a facet. This divisor is associated with a further blow-up of the base.
More precisely: after resolving the general singularities of the Weirstrass model, which gives rise to the $`S(13)`$ configuration, the fourfold is still singular; the curve of singularities is a rational curve. It turns out that after one blow up the fourfold $`X`$ is smooth and still satisfies the Calabi-Yau condition; $`X`$ is elliptically fibered over a threefold $`B`$, which is the blow up of $`B^X`$ (the common base of the other examples) along a rational curve $`\mathrm{\Gamma }`$. The fiber of this new fibration are all curves, and there are no new gauge groups.
The elliptic threefold over the surface $`B^Y`$, obtained after resolving the general singularities over $`\sigma _{\mathrm{}}`$ is singular at one point, after blowing up this point to another surface $`B`$ and normalizing we obtain a new Calabi-Yau, mapping to $`B`$, with one dimensional fibers.
$`\underset{ยฏ}{E_8\text{}}`$ Again the nodes of the Dynkin diagram correspond to divisors with arithmetic genus unity and the dissident divisor is as in the previous case.
A more careful statement is again that after resolving the general singularities of the Weirstrass model, which gives rise to the $`E_8`$ configuration, the fourfold is still singular and further blow ups are needed. It turns out that after all the necessary blow ups the fourfold $`X`$ still satisfies the Calabi-Yau condition; $`X`$ is elliptically fibered over a threefold $`B`$, which is the blow up of $`B^X`$ (the common base of the other examples) along a curve $`\mathrm{\Gamma }`$. The fiber of this new fibration are all curves, and there are no new gauge groups. We can also consider the elliptic threefold over the surface $`B^Y`$ ($`B^X`$ is fibered by $`B^Y`$, the Calabi-Yau fourfold is fibered by such Calabi-Yau threefolds). $`B^Y`$ is the Hirzebruch surface $`\mathrm{IF}_3`$ blown up at a point. We denote by $`f`$ the exceptional divisor of this blow up, $`g`$ the strict transform of a fiber of the fibration $`\mathrm{IF}_3\mathrm{IP}^1`$ and $`\sigma _{\mathrm{}}`$ the section with negative self-intersection. Our models have a $`SU(2)`$ gauge group over a section $`\sigma _0`$, such that $`\sigma _0g=0,\sigma _0f=1`$. A quick check shows that the general elliptic fibration with an $`E_8`$ gauge group over $`\sigma _{\mathrm{}}`$, acquires extra singularities over $`9`$ different points $`P_i,i=1,\mathrm{}9`$, at the intersection of $`\sigma _{\mathrm{}}`$ with the divisor of $`I_1`$ singular fibers. These $`9`$ points are the intersection of $`\mathrm{\Gamma }`$ and $`B^Y`$. It can be easily verified that the resolution of these singularities introduces $`9`$ new divisors on the threefold: if we want to mantain the equidimensionality conditions (all the fibers being curves), then we have to blow up the base at each $`P_i`$. The new threefold is Calabi-Yau and there are no new gauge groups.
$`\underset{ยฏ}{E_{7b}\text{}}`$ This is one of the ways of realizing $`E_7`$ and is similar to the two cases above. All the nodes of the Dynkin diagram correspond to divisors with arithmetic genus unity. The dissident divisor is $`\{0,0,1,0,0\}`$ which is a point interior to a facet.
5. The Yukawa Couplings and Mori Cone
5.1. The Yukawa Couplings The topological yukawa couplings $`D_iD_jD_kD_l`$ are calculated from the fan by means of the program SCHUBERT. The most important vanishing relations in the intersection ring can be read off directly from the fan or more simply from Figure 4.1. We see, for example, that $`V_1`$ and $`V_6`$ always lie in different cones and that $`G`$ never occurs in the same cone with $`C_1`$ or $`C_2`$ and that $`F`$ never occurs in the same cone with any of the $`E_i`$. It follows that these divisors do not intersect in $`\mathrm{IP}_{}`$ and hence do not intersect in $`X`$. In this way we learn that
$$\begin{array}{c}\text{ }GC_i=0,FE_j=0,C_iE_j=0,BC_2=BE_2=BE_3=0\text{ }\hfill \\ \text{ }F(F+G)=G(F+G)=0,Y^2=0,(3H+E_3)E_2=0\text{ }\hfill \end{array}$$
where the last of these identities follows from the fact that $`V_5`$ and $`E_2`$ never lie in the same cone. Four further quadratic identities follow by examining the intersection numbers
$$E_1E_3=0,BH=0,E_1H=0,\text{and}(2HC_2)C_1=0.$$
Taken together with the previous identities these furnish a basis for the quadratic relations between the divisors. The nonzero intersection numbers are given below for a slightly redundant basis that includes also the divisor $`E_3`$:
| $`B^4=82`$ | $`B^3C_1=48`$ | $`B^3E_1=4`$ | $`B^3F=8`$ | $`B^3Y=7`$ |
| --- | --- | --- | --- | --- |
| $`B^2C_1^2=28`$ | $`B^2C_1F=6`$ | $`B^2C_1Y=4`$ | $`B^2E_1^2=10`$ | $`B^2E_1G=2`$ |
| $`B^2E_1Y=1`$ | $`B^2F^2=2`$ | $`B^2FG=2`$ | $`B^2FY=1`$ | $`B^2G^2=2`$ |
| $`B^2GY=1`$ | $`BC_1^3=16`$ | $`BC_1^2F=4`$ | $`BC_1^2Y=2`$ | $`BC_1FY=1`$ |
| $`BE_1^3=24`$ | $`BE_1^2G=4`$ | $`BE_1^2Y=3`$ | $`BE_1GY=1`$ | $`BF^3=4`$ |
| $`BF^2G=4`$ | $`BF^2Y=1`$ | $`BFG^2=4`$ | $`BFGY=1`$ | $`BG^3=4`$ |
| $`BG^2Y=1`$ | $`C_1^4=144`$ | $`C_1^3C_2=192`$ | $`C_1^3F=8`$ | $`C_1^3Y=8`$ |
| $`C_1^2C_2^2=272`$ | $`C_1^2C_2F=16`$ | $`C_1^2C_2Y=12`$ | $`C_1^2FY=2`$ | $`C_1C_2^3=384`$ |
| $`C_1C_2^2F=24`$ | $`C_1C_2^2Y=16`$ | $`C_1C_2FY=2`$ | $`C_2^4=528`$ | $`C_2^3F=32`$ |
| $`C_2^3Y=20`$ | $`C_2^2FY=2`$ | $`E_1^4=56`$ | $`E_1^3G=8`$ | $`E_1^3Y=8`$ |
| $`E_1^2E_2^2=2`$ | $`E_1^2E_2Y=1`$ | $`E_1^2GY=2`$ | $`E_1E_2^3=4`$ | $`E_1E_2^2G=2`$ |
| $`E_1E_2^2Y=1`$ | $`E_1E_2GY=1`$ | $`E_2^4=48`$ | $`E_2^3E_3=36`$ | $`E_2^3G=8`$ |
| $`E_2^3Y=8`$ | $`E_2^2E_3^2=18`$ | $`E_2^2E_3G=6`$ | $`E_2^2E_3Y=9`$ | $`E_2^2GY=2`$ |
| $`E_2E_3^2Y=9`$ | $`E_2E_3GY=3`$ | $`E_3^4=72`$ | $`E_3^3G=24`$ | $`E_3^2GY=6.`$ |
5.2. The Mori Cone of $`X`$ The Mori cone (what is this and why do we even care) of the embedding space $`\mathrm{IP}_{}`$ may be found by the method of positive piecewise linear functions as explained in KLRY (see also ). Recall that a dimension is added to the vector space defined by the points of $``$ and a 1 is prepended to each of the points corresponding to the divisors (so that $`D_0(1,0,0,0,0)`$, etc.). The cones of the fan extend to simplicial cones whose vertex is at the origin of this extended space.
A piecewise linear function $`m`$ is defined on the extended space so as to be linear on each cone. If $`u`$ is a point in a cone $`\sigma `$ with generators $`u_i`$, $`iI`$, then $`u`$ can be expressed uniquely in the form
$$\stackrel{}{u}=\underset{iI}{}\lambda _i\stackrel{}{u}_i;\lambda _i0.$$
The function $`m(\stackrel{}{u})`$ is given in terms of $`m_i=m(\stackrel{}{u}_i)`$ by
$$m(\stackrel{}{u})=\underset{iI}{}\lambda _im_i$$
which is equivalent to giving a vector $`\stackrel{}{m}_\sigma `$ on each cone such that
$$m(\stackrel{}{u})=\stackrel{}{m}_\sigma \stackrel{}{u}\mathrm{for}\stackrel{}{u}\sigma .$$
The function $`m`$ is a positive piecewise linear function if
$$\begin{array}{cc}\hfill m(\stackrel{}{u})& =\stackrel{}{m}_\sigma \stackrel{}{u};u\sigma \hfill \\ \hfill m(\stackrel{}{u})& \stackrel{}{m}_\sigma \stackrel{}{u};u\sigma \hfill \end{array}$$
$`(5.1)`$
The system of inequalities (5.1), being linear, is specified by the coefficients that appear. In this way, they specify a cone which is identified with the Mori cone of $`\mathrm{IP}_{}`$. The integral basis for the system (5.1) may be identified with the generators of the Mori cone. (This process is performed in detail for the simple case of the threefold $`Z`$ in appendix A.) For fourfolds this process requires computer calculation.
For the case at hand, the relations between the divisor classes may be used to express the system (5.1) entirely is terms of a basis consisting of the divisor classes
$$\{B,C_1,C_2,E_1,E_2,F,G,Y\}.$$
$`(5.2)`$
In other words, the $`m_i`$ corresponding to the remaining divisors may be set to zero. The system (5.1) then consists of 83 inequalities which are generated by the following coefficient vectors
$$\begin{array}{cc}& a^1=(1,2,2,0,0,1,0,0),\hfill \\ \multicolumn{2}{c}{}\\ & a^0=(1,1,1,1,0,0,0,0),\hfill \\ \multicolumn{2}{c}{}\\ & a^1=(1,1,0,0,0,1,1,0),\hfill \\ \multicolumn{2}{c}{}\\ & a^2=(0,0,0,1,1,0,1,0),\hfill \\ \multicolumn{2}{c}{}\\ & a^3=(1,0,0,1,0,1,1,0),\hfill \\ \multicolumn{2}{c}{}\\ & a^4=(0,0,0,1,2,0,0,0),\hfill \\ \multicolumn{2}{c}{}\\ & a^5=(0,0,0,0,1,0,0,0),\hfill \\ \multicolumn{2}{c}{}\\ & a^6=(0,1,1,0,0,0,0,0),\hfill \\ \multicolumn{2}{c}{}\\ & a^7=(1,3,3,0,0,0,0,0),\hfill \\ \multicolumn{2}{c}{}\\ & a^8=(1,0,0,2,0,0,0,0),\hfill \\ \multicolumn{2}{c}{}\\ & a^9=(0,0,0,0,2,0,0,1),\hfill \end{array}$$
The vectors on the right are the coefficients of the inequalities that generate (5.1). We may associate them also with the curves that generate the Mori cone. The components of the vectors are also the intersection numbers
$$a^i{}_{j}{}^{}=a^iD_j$$
of the curves with the divisors $`D_j`$ of the basis (5.2). The Mori cone that we have obtained has nine edges $`a^i,i=1,\mathrm{},9`$. There are in addition two further generators $`a^1`$ and $`a^0`$ which are internal to the cone and which are required because in this case the edges do not generate the cone
$$\begin{array}{cc}& a^0=\frac{1}{2}(a^6+a^7+a^8),\hfill \\ \multicolumn{2}{c}{}\\ & a^1=\frac{1}{2}(a^0+a^4+a^7).\hfill \end{array}$$
Note that since there are nine edges rather than eight the cone is not simplicial. A ninth edge is however necessary since the relation of linear dependence is
$$a^1a^23a^3+4a^4+2a^5=0$$
and so no edge can be written as a positive combination of the others.
The complication is that what we have calculated is the Mori cone of the embedding space $`\mathrm{IP}_{}`$ rather than the Mori cone of $`X`$. The procedure advocated by Cox and Katz is to
(i) compute all the possible fans for $`\mathrm{IP}_{}`$ that is triangulate $``$ in all possible ways. The program PUNTOS can accomplish this in cases that are not too complicated.
(ii) Compute using the program SCHUBERT the yukawa couplings $`Y^{ijkl}=D_iD_jD_kD_l`$ corresponding to each fan. Fans that lead to different couplings $`Y^{ijkl}`$ correspond to different phases of the theory in the sense of the linear sigma-model. Fans that lead to the same $`Y^{ijkl}`$ correspond to different resolutions of the embedding space that do not affect the CalabiโYau hypersurface $`X`$. That is the corresponding $`\mathrm{IP}_{}^{}s`$ are related by flopping curves that do not intersect $`X`$. Thus the fans should be grouped into classes classified by the $`Y^{ijkl}`$.
(iii) Within a given class the Mori cones for the embedding spaces will in general be different however the true Mori cone, i.e., the Mori cone for $`X`$, should be contained in each of them. Thus we proceed, for a given class, by computing the intersection of all the corresponding Mori cones of the embedding spaces.
We have carried through this program for our $``$; though we shall see that the resulting cone is still too large. A total of 990 fans were found which when classified by the yukawa couplings fall into 7 classes. These 7 classes turn out to correspond to the 7 ways of triangulating the $`(0,x_2,x_3,2,3)`$ plane shown in Figure 4.1. The class corresponding to our couplings, i.e., the Table at the beginning of this section, comprises 20 fans all of which correspond to the triangulation of the $`(0,x_2,x_3,2,3)`$ plane shown in the central figure in Figure 4.1. Although they coincide in this plane these 20 fans are different leading to 20 distinct Mori cones. Of the 20 only 4 satisfy the condition that was discussed previously that each cone of the fan of $`\mathrm{IP}_{}`$ project onto some cone of the fan for $`B`$. The other fans in this class correspond to the same yukawa couplings as the fans that do satisfy the projection criterion so it must be the case that although these fans do not realize $`\mathrm{IP}_{}`$ as an elliptic fibration they do realize $`X`$ as an elliptic fibration.
The task of finding the intersection of the twenty cones can be done by means of a computer program however in this particular case it is easy to do by hand. It happens that among the twenty cones there are many edges that appear in a certain cone and then with opposite sign in another cone. In such a case it is easy to see that the intersection of the two cones is contained in the cone formed by discarding the edges that appear with opposite sign and taking the union of the remaining edges for the two cones. Thus we may discard all the edges that appear with opposite sign and take the union of all the remaining edges. In this way we see that the intersection must be contained in the following cone that has the ten edges $`\mathrm{}^i,i=1,\mathrm{},10`$
$$\begin{array}{cc}& \mathrm{}^0=(1,1,1,1,1,0,0,0)=a^0+a^5\hfill \\ \multicolumn{2}{c}{}\\ & \mathrm{}^1=(0,0,0,1,1,0,1,0)=E_1E_2Y^+=a^2\hfill \\ \multicolumn{2}{c}{}\\ & \mathrm{}^2=(1,2,2,0,0,0,0,0)=C_1FY^+=a^6+a^7\hfill \\ \multicolumn{2}{c}{}\\ & \mathrm{}^3=(0,1,1,0,0,0,0,0)=\frac{1}{2}C_2FY^+=a^6\hfill \\ \multicolumn{2}{c}{}\\ & \mathrm{}^4=(1,0,0,2,1,0,0,0)=E_1GY^+=a^5+a^8\hfill \\ \multicolumn{2}{c}{}\\ & \mathrm{}^5=(0,0,0,1,2,0,0,0)=E_2GY^+=a^4\hfill \\ \multicolumn{2}{c}{}\\ & \mathrm{}^6=(0,0,0,0,1,0,0,0)=\frac{1}{3}E_3GY^+=a^5\hfill \\ \multicolumn{2}{c}{}\\ & \mathrm{}^7=(1,1,0,0,0,1,1,0)=FBY^+=a^1\hfill \\ \multicolumn{2}{c}{}\\ & \mathrm{}^8=(1,0,0,1,0,1,1,0)=GBY^+=a^3\hfill \\ \multicolumn{2}{c}{}\\ & \mathrm{}^9=(0,0,0,0,2,0,0,1)=E_1E_2G=\frac{1}{3}E_2E_3G=a^9\hfill \\ \multicolumn{2}{c}{}\\ & \mathrm{}^{10}=(1,0,1,0,0,0,0,0)=a^0+a^1+a^3\hfill \end{array}$$
$`(5.3)`$
It is easy to see also that each of the edges of this cone is contained in each of the twenty cones with which we started so that it is in fact the intersection we were seeking. Again in this case the edges do not generate the cone and we require also the internal generator
$$\mathrm{}^0=\frac{1}{2}(\mathrm{}^2+\mathrm{}^4+\mathrm{}^6)$$
This cone contains the true Mori cone and by inspection we identify curves of $`X`$ with each of the generators apart from $`\mathrm{}^0`$ and $`\mathrm{}^{10}`$. Thus the edges $`\mathrm{}^1,\mathrm{},\mathrm{}^9`$ are true edges. Notice also that the divisors of our basis appear in the curves as
$$\begin{array}{cc}\hfill \{C_1,C_2\}FY^+=\{C_1,C_2\}Y^+Y^{}& ,\{E_1,E_2,E_3\}GY^+=\{E_1,E_2,E_3\}Y^+Y^{},\hfill \\ \multicolumn{2}{c}{}\\ \hfill \text{and}& \{F,G\}BY.\hfill \end{array}$$
It is interesting to note that if we compute the intersection matrix between the divisors $`C_i`$ and the curves $`C_jFY^+`$ with $`i,j=1,2`$ we find
$$C_iC_jFY^+=\left(\begin{array}{cc}2& 2\\ 2& 2\end{array}\right)$$
which we recognise as the extended Cartan matrix for $`SU(2)`$ by which we mean the matrix corresponding to including the extending root of the algebra. This is in accord with the observations of Intriligator et al. . If we consider the intersections between the divisors $`E_i`$ and the curves $`\mathrm{}^j`$ with $`i=1,2,3`$ and $`j=4,5,6`$ then we find the extended Cartan matrix corresponding to $`G_2`$
$$E_i\mathrm{}^j=\left(\begin{array}{ccc}2& 1& 0\\ 1& 2& 3\\ 0& 1& 2\end{array}\right).$$
The curve $`\mathrm{}^9=E_1E_2G`$ maps to the $`\mathrm{IP}_1`$ of the base of CalabiโYau-fibration $`X=(Y,\mathrm{IP}{}_{1}{}^{})`$ while $`\mathrm{}^1`$ maps to a fiber of $`B^Z`$. Since there are ten edges there are two linear relations between them one of these involves $`\mathrm{}^{10}`$ the other relation is more interesting and permits the elliptic fiber to be expressed in two different ways
$$=FGY=(1,0,0,0,0,0,0,0)=\mathrm{}^2+2\mathrm{}^3=\mathrm{}^4+2\mathrm{}^5+3\mathrm{}^6.$$
$`(5.4)`$
We can show that $`\mathrm{}^0`$ and $`\mathrm{}^{10}`$ cannot be generators of the true cone. Consider first $`\mathrm{}^{10}`$. If this is a generator of the Mori Cone then it is an irreducible curve that is contained in $`B`$, since $`\mathrm{}^{10}B=1`$, and which also intersects $`C_2`$, since $`\mathrm{}^{10}C_2=1`$. But this is impossible since $`B`$ and $`C_2`$ do not intersect. In a similar way we see that $`\mathrm{}^0`$ is not a true generator since such a curve would have to be contained in both $`C_1`$ and $`E_1`$ which however do not intersect.
Note now that the generators $`\mathrm{}^1,\mathrm{}^2,\mathrm{}^4,\mathrm{}^5,\mathrm{}^7,\mathrm{}^8,\mathrm{}^9`$, that is the generators $`\mathrm{}^1,\mathrm{},\mathrm{}^9`$ with $`\mathrm{}^3`$ and $`\mathrm{}^6`$ omitted, define a seven-plane $`L`$, say, within the eight dimensional cone. For any vector $`k`$ in the lattice we can define a height relative to $`L`$
$$h(k)=det(k,\mathrm{}^1,\mathrm{}^2,\mathrm{}^4,\mathrm{}^5,\mathrm{}^7,\mathrm{}^8,\mathrm{}^9)=h.k$$
where on the right $`h`$ denotes the vector $`(6,8,5,4,2,8,6,4)`$. Now $`h(\mathrm{}^0)=1`$, $`h(\mathrm{}^3)=3`$, $`h(\mathrm{}^6)=2`$ and $`h(\mathrm{}^{10})=1`$ so $`\mathrm{}^{10}`$ lies on one side of $`L`$ and $`\mathrm{}^0`$, $`\mathrm{}^3`$ and $`\mathrm{}^6`$ lie on the other side. This seems to provide a counterexample to the conjecture of Cox and Katz (see also ) that the procedure we have followed should yield the true Mori cone.
We have seen that $`\mathrm{}^{10}`$ is not a true generator and the question arises as to whether we can discard all the points that have negative height with respect to $`L`$. Now we have to express $`\mathrm{}^0`$ as a positive integral combination of generators, since $`\mathrm{}^0`$ cannot be a generator, and since $`h(\mathrm{}^0)=1`$ which is less than $`h(\mathrm{}^3)`$ and $`h(\mathrm{}^6)`$ we see that we must have at least one generator with negative height.
5.3. Volumes of the Divisors There are two natural ways to parametrize the Kรคhler-form. The first is to write it directly in terms of the basis of divisors
$$J=tB+s_1C_1+s_2C_2+s_3E_1+s_4E_2+s_5F+s_6G+vY.$$
in this expression $`t=J`$ is the volume of the elliptic fiber. Another useful parametrization of the Kรคhler-form is obtained by taking the volumes of the curves $`\mathrm{}^1,\mathrm{},\mathrm{}^9`$ (defined in (5.3)) as coordinates. We set
$$J\mathrm{}^i=(\mu _+,\gamma _1,\gamma _2,ฯต_1,ฯต_2,ฯต_3,\delta _F,\delta _G,\mu _{}),i=1,\mathrm{},9.$$
$`(5.5)`$
There are 9 parameters on the right so there is a linear relation which is a consequence of (5.4)
$$t=\gamma _1+2\gamma _2=ฯต_1+2ฯต_2+3ฯต_3.$$
$`(5.6)`$
We wish now to examine the relation between the volumes of the b-divisors and the volumes of the Mori generators. This is of interest since the superpotential arises, in the M-theory description, through contributions of the form $`\mathrm{exp}(2\pi i\text{vol}(D))`$ while in the heterotic description it arises through instanton corrections and the instantons are linear combinations of the Mori generators. Now
$$\text{vol}(D)=\frac{1}{3!}J^3D$$
which is cubic in the parameters of $`J`$ while the volumes of the Mori curves and so of course the instantons are linear in the parameters of $`J`$. The volumes of the b-divisors are for the most part complicated cubic expressions in the parameters. The simplest of these expressions are those for the volumes of $`E_1`$ and $`E_2`$
$$\begin{array}{cc}\hfill \text{vol}(E_1)& =ฯต_1\left(\mu _{}\mu _++ฯต_1(\frac{1}{2}\mu _{}+\mu _+)+\frac{4}{3}ฯต_1^2\right)\hfill \\ \multicolumn{2}{c}{}\\ \hfill \text{vol}(E_2)& =ฯต_2\mu _{}\left(\mu _++\frac{1}{2}ฯต_2\right).\hfill \end{array}$$
It seems that one should consider the limit with $`t`$ small. Since the parameters that we have introduced through (5.5) are all positive it follows in virtue of (5.6) that the $`\gamma _i`$ and $`ฯต_j`$ tend to zero with $`t`$. We expect $`\text{vol}(D)`$ to tend to zero linearly with $`t`$ for every b-divisor $`D`$ so the neglect of terms of $`๐ช(t^2)`$ leads to expressions with a term of $`๐ช(t)`$ as a factor. For $`F`$ we have in this limit $`\text{vol}(F)t\mu _{}\delta _F`$. The corresponding linearized expressions for the other b-divisors are as follows
$$\begin{array}{cc}\hfill \text{vol}(C_1)& \gamma _1(2\delta _F+3\delta _G+\mu _+)(4\delta _F+4\delta _G+\mu _{})\hfill \\ \multicolumn{2}{c}{}\\ \hfill \text{vol}(C_2)& 2\gamma _2(2\delta _F+3\delta _G+\mu _+)(4\delta _F+4\delta _G+\mu _{})\hfill \\ \multicolumn{2}{c}{}\\ \hfill \text{vol}(E_1)& ฯต_1\mu _{}\mu _+\hfill \\ \multicolumn{2}{c}{}\\ \hfill \text{vol}(E_2)& ฯต_2\mu _{}\mu _+\hfill \\ \multicolumn{2}{c}{}\\ \hfill \text{vol}(E_3)& 3ฯต_3\mu _{}\mu _+\hfill \\ \multicolumn{2}{c}{}\\ \hfill \text{vol}(F)& t\delta _F(\mu _{}+2\delta _F+4\delta _G)\hfill \\ \multicolumn{2}{c}{}\\ \hfill \text{vol}(G)& t\delta _G(\mu _{}+2\delta _G).\hfill \end{array}$$
6. The Superpotential
6.1. Characterization of the Divisors Contributing to the Worldsheet Instantons Let us denote with $`\pi :XB^X`$ the elliptic fibration, $`p:B^XB^Z`$ the fibration by rational fibers (generally $`\mathrm{IP}^1`$), and by $`ฯต:XB^Z`$ the composed K3-fibration. We assume that $`p`$ is equidimensional (replacing, if necessary, $`B^Z`$ by a suitable blow up). The divisors that contribute to the superpotential are the divisors $`D`$ such that $`\pi (D)`$ is a divisor, $`C`$, of $`B^X`$ and $`p(\pi (D))`$ is a curve, $`\gamma `$, in $`B^Z`$:
$$\begin{array}{c}X\\ \\ \pi & ฯต\\ \\ B^X& ^p& B^Z\end{array}\begin{array}{c}D\\ \\ \pi & ฯต\\ \\ C& ^p& \gamma \end{array}$$
Most the divisors that we construct via the toric construction contribute to space-time instantons. It turns out that divisors contributing to the worldsheet instantons are nicely divided in 3 different types, each of which has a distinct meaning in physics:
(a) If $`D`$ does not correspond to a non-abelian gauge group and $`D=ฯต^{}(\gamma )`$, where $`\gamma =ฯต(D)`$. In this case $`1=\chi (D)=\gamma ^2`$. In fact,
$$\chi (D)=D^2c_2/24=\gamma ^2\chi _{top}(S)/24=\gamma ^2,$$
where $`S`$ is the general fiber of the $`K_3`$-fibration $`XB^Z`$. Furthermore, the adjunction formula shows that $`\gamma `$ is a smooth rational curve. Such curve is necessarily an edge of the Mori cone of $`B^Z`$.
(b) If $`D`$ does not correspond to a non-abelian gauge group and $`Dฯต^{}(\gamma )`$, then $`p:B^XB^Z`$ is not a $`\mathrm{IP}^1`$-bundle; not all such divisors contribute to the superpotential, even if $`\chi (๐ช_D)=1`$ (see also ). HOwever the divisors $`F`$ and $`G`$ do contribute(see Section 4).
(c) If $`D`$ corresponds to a gauge group, $`D`$ arises from a degeneration of the K3 fiber $`S`$: therefore $`D`$ generates a non-perturbative gauge group and the corresponding heterotic model will have singularities. In particular $`ฯต(D)=\gamma `$ is a component of the discriminant locus (with gauge groups) of the elliptic fibration $`ZB^Z`$. We have the following 2 cases:
$``$ c.1) $`C=\pi (D)=p^{}(\gamma )`$
$``$ c.2) $`C`$ is a component of $`\pi (D)p^{}(\gamma )`$
6.2. Comparison with the Heterotic Superpotential At this point we do not know much about cases b) and c), so we are concentrating on the divisors of type a). Most of the explicit examples in the literature are of type a). The corresponding heterotic superpotential (via the F-theory/heterotic duality) is expected to be a function of the volume of the curves $`p(\pi (D))`$ and thus to be linear. While we do not see any mathematical a priori reason of why this should be true, it turns out to be so in all the examples examined hitherto. Typically the divisors contributing to the superpotential on $`F`$-theory compactifications are finite in number: this is because $`B^X`$, the base of the elliptic fibration needs to have an effective first Chern class ($`c_1(B^X)0`$); in the toric case also the number is always finite .
The examples in and have only a finite number of divisors and these are of type (a) and a simple computation shows that the superpotential is linear in the volumes of these divisors, up to an overall factor. The computation is more complicated in the case where there are infinitely many divisors, as in the following examples:
6.3. Andreasโ Examples In his paper on heterotic/F-theory duality Andreas considers certain examples closely related to the one in : there are infinitely many divisors contributing to the superpotential. We summarize his argument here:
$`\underset{ยฏ}{\text{The threefold }B_n=B^X\text{.}}`$ $`S\mathrm{IP}^1`$ is $`\mathrm{IP}^2`$ blown up at 9 points and $`\mathrm{IF}_r\mathrm{IP}^1`$ a Hirzebruch surface. $`B_n=B^X`$ is the fiber product of $`S`$ and $`\mathrm{IF}_r`$ with base $`\mathrm{IP}^1`$, with $`2r=n`$. In particular $`B^X`$ is a $`\mathrm{IP}^1`$-bundle over $`S`$, but is not a product unless, $`r=0`$ (where $`B^X=S\times \mathrm{IP}^1`$, as in ).
$`\underset{ยฏ}{\text{The threefold }_r\text{.}}`$ Let $`X_n^3\mathrm{IF}_n`$, $`n=2r`$ be a smooth Calabi-Yau 3-fold, with an involution $`\tau `$ compatible with the involution on $`\mathrm{IP}^1`$: $`zz`$. (Such threefolds can be obtained by choosing appropriate coefficients for the Weierstrass model.) Set $`X_n^3/\tau =_r`$; $`_r`$ is a smooth threefold. There is a natural elliptic fibration $`_r\mathrm{IF}_r`$. (Note: if $`r=n=0,_0=B`$, as in .)
$`\underset{ยฏ}{\text{The fourfold }X_n^4=X_r\text{.}}`$ $`X_r`$ is the fiber product of $`S\mathrm{IP}^1`$ and $`_r\mathrm{IP}^1`$. By construction $`X_r`$ is elliptically fibered over $`B^X`$, while is fibered by K3 surfaces over $`S=B^Z`$ (the basis of the heterotic dual.
It can be verified that $`X_r`$ is a Calabi-Yau 4-fold, with heterotic dual $`X_n^3`$. (If $`r=0`$, then $`X_0`$ is the Weierstrass model of the $`X`$ in ; the calculation in computes also the superpotential for $`X_0`$, up to a factor.) We are interested in the contribution to the superpotential from worldsheet instantons.
$`\underset{ยฏ}{\text{The divisors contributing to the superpotential}}`$ The divisors contributing to the superpotential via worldsheet instanton are, as in the inverse images of the section of the fibration $`S\mathrm{IP}^1`$. As in , they are all isomorphic to $`_r`$. If $`r>0`$, there are other divisors contributing to the superpotential via spacetime instantons (some correspond to gauge groups).
Denote by $`\{\mathrm{\Gamma }_0,\mathrm{\Gamma }_1,\mathrm{},\mathrm{\Gamma }_s\}`$ the generators of $`H^2(B^X,\mathrm{IR})`$ and by
$$\{B,\pi ^{}(\mathrm{\Gamma }_0),\mathrm{},\pi ^{}(\mathrm{\Gamma }_s),\mathrm{\Lambda }_1,\mathrm{},\mathrm{\Lambda }_w\}$$
the generators $`H^2(X,\mathrm{IR})`$, where $`BB^X`$ and the $`\mathrm{\Lambda }_j`$ correspond to gauge groups (there are the exceptional divisors of the morphism to the Weierstrass model). Without loss of generality we take $`\mathrm{\Lambda }_jB=0`$ and take $`\mathrm{\Gamma }_0SB^Z`$ and $`\mathrm{\Gamma }_i=p^{}(\gamma _j)`$, where $`\{\gamma _j\}`$ generate the Kรคhler cone of $`S`$. As before we can express the Kรคhler form as a linear combination of the divisors
$$J=tB+\stackrel{~}{\mathrm{\Gamma }}(u)+\mathrm{\Lambda }(v),\text{with}\stackrel{~}{\mathrm{\Gamma }}(u)=\underset{j}{}u^j\pi ^{}(\mathrm{\Gamma }_j)\text{and}\mathrm{\Lambda }(v)=\underset{k}{}v^k\mathrm{\Lambda }_k.$$
The volume form for a threefold is then
$$\frac{1}{3!}J^3=\frac{1}{3!}\left(tB+\stackrel{~}{\mathrm{\Gamma }}(u)+\mathrm{\Lambda }(v)\right)^3$$
We will now compute the volume of a worldsheet instanton, that is, $`\text{vol}(D_\gamma )`$ for $`D_\gamma =\pi ^{}(p^{}\gamma )`$, with $`\gamma `$ a section of the elliptic fibration $`B^Z\mathrm{IP}^1`$. We use the following facts: $`B^2=B_{|B}=K_B`$, and $`K_{B^Z}`$ is a fiber of the elliptic fibration $`B^Z\mathrm{IP}^1`$. By the adjunction formula $`B^2=K_B=2S(r+1)p^{}(F)`$, where $`F`$ is a fiber of the fibration $`S\mathrm{IP}^1`$ and $`B^3=K_BK_{B}^{}{}_{|B}{}^{}`$. Note that again we have $`t=\text{vol}`$, and that, if $`f`$ is the homology class of the $`\mathrm{IP}^1`$ bundle $`p:B^XS`$, $`t_0=(t_j\pi ^{}\mathrm{\Gamma }_j)f=Sf=\text{vol}f.`$ Using the geometry of the fiber products involved we see that:
$$\begin{array}{cc}\hfill B^2\stackrel{~}{\mathrm{\Gamma }}(u)D_\gamma =& (r1)\text{vol}_B(f)2\text{vol}_S(\gamma )\hfill \\ \multicolumn{2}{c}{}\\ \hfill B^3D=& 4\hfill \\ \multicolumn{2}{c}{}\\ \hfill \stackrel{~}{\mathrm{\Gamma }}(u)^3D_\gamma =& 0\hfill \\ \multicolumn{2}{c}{}\\ \hfill \mathrm{\Lambda }(v)^3D_\gamma =& d(r,v)\hfill \\ \multicolumn{2}{c}{}\\ \hfill B^2\mathrm{\Lambda }(v)D_\gamma =& 0\hfill \\ \multicolumn{2}{c}{}\\ \hfill \mathrm{\Lambda }(v)\stackrel{~}{\mathrm{\Gamma }}(u)^2D_\gamma =& 0\hfill \\ \multicolumn{2}{c}{}\\ \hfill B\mathrm{\Lambda }(v)^2D_\gamma =& 0\hfill \\ \multicolumn{2}{c}{}\\ \hfill \stackrel{~}{\mathrm{\Gamma }}(u)\mathrm{\Lambda }(v)^2D_\gamma =& c(r,u,v)\text{vol}_{B^X}(f)+c^{}(r,u,v)\text{vol}_S(\gamma )\hfill \\ \multicolumn{2}{c}{}\\ \hfill \stackrel{~}{\mathrm{\Gamma }}(u)^2BD_\gamma =& 2\text{vol}_S(\gamma )\text{vol}_{B^X}(f)r\text{vol}_{B^X}^2(f)\hfill \\ \multicolumn{2}{c}{}\\ \hfill B\mathrm{\Lambda }(v)\stackrel{~}{\mathrm{\Gamma }}(u)=& 0,\hfill \end{array}$$
where $`d(r,v),c(r,u,v),c^{}(r,u,v)`$ are linear function on $`r`$, which do not depend on the choice of $`\gamma `$, and are zero for $`r=0`$. Then:
$$\mathrm{exp}(\text{vol}(D))=\mathrm{exp}(A)\times \mathrm{exp}\left[C\text{vol}(\gamma )\right],$$
where $`A`$ and $`C`$ are the same for every divisor $`D_\gamma `$. This function, up to a constant depending only on $`r`$, is the expression in (and for $`r=0`$ is equal to this expression); then, up to a constant the superpotential is as in .
6.4. Comparison with $`d=3`$ Dimensional Yang-Mills Theory Katz and Vafa in consider divisors contributing to the superpotential arising from resolution of singularities of the Weierstrass model; the divisors are associated to a simple gauge group $`G`$. They assume that there is no adjoint matter. By the chain of duality in , $`F`$-theory compactified on a circle is dual to $`M`$-theory compactified on the Calabi-Yau manifold $`X`$; the radius of the circle is the inverse to the Kรคhler class of the elliptic fiber. In this way, one obtains a $`N=2`$ theory with $`d=3`$; Katz and Vafa show that, under this duality the $`F`$-theory superpotentials become the expected superpotential. A key point in their computation is that each divisor $`D_i`$ corresponding to the nodes of the affine Dynkin diagram of the Group $`G`$ contribute to the superpotential, that is they satisfy the conditions (2.4) and (2.5). For examples, in the simply laced cases, one can write $`[]=_{i=1}^{r+1}a_i[e_i]`$, where $``$ is the class of the elliptic fiber, $`e_i`$ is the fiber of each $`D_i`$ (a ruled surface) and $`a_i`$ is the Dynkin index of the corresponding node of the Dynkin diagram).
6.5. New Features for a Non-Simply Laced Group While the argument implied by the chain of dualities should imply the same conclusion, we are unable to make the argument work for the cases where not all the divisors satisfy the condition that $`\chi (๐ช_D)=1`$, as in the example in Sections 5 and 6. These mixed configurations, in which some but not all of the nodes of the Dynkin diagram contribute to the superpotential, correspond to a genuine instability since in these cases it is not possible to satisfy the conditions $`dW=0`$ corresponding to the supersymmetric vacuum states.
If $`\chi (๐ช_D)>1`$, then $`h^2(D)>0`$. It follows that $`D`$ is not general in the moduli of $`X`$, that is the locus for which $`D`$ deforms in the family $`๐ณ`$ (of complex deformation of $`X`$) is a complex submanifold of codimension $`h^2(D)`$. The argument needed is a modification of the corresponding statement for Calabi-Yau threefolds in . In this case $`h^2(D)`$ contributes to the number of non-toric parameters. We will argue in this case that the Calabi-Yau fourfold is non-general and the usual techniques for counting the divisors contributing to the superpotential do not suffice. For example, there is evidence that one should also consider the contribution of reducible divisors.
If $`\chi (๐ช_D)0`$, then $`h^1>0`$; in the our examples $`h^2(D)=0`$ and there are no non-toric parameter, so the divisor $`D`$ will be effective (with the same Hodge number), for all points of the complex moduli space of $`X`$. Following we see that in the toric case
$$h^{2,1}(X)=\underset{qdim\stackrel{~}{\theta }=2}{}(1\chi _q)=h^1(D).$$
This gives an interesting, yet-little studied structure on the heterotic dual .
By construction the Calabi-Yau fourfold $`X`$ is fibered by the family of Calabi-Yau threefolds $`Y`$. It follows that the locus for which $`D_Y`$ deforms in the family $`๐ด`$ (of complex deformation of $`Y`$) is a complex submanifold of codimension $`h^1(D_Y)`$. At this point we are not sure of the implication of this fact.
6.6. The prefactor. Ganor argued that the contribution of a divisor $`D`$ to the superpotential is multiplied by a pre-factor $`f`$. In most cases, the pre-factor is non-zero; Ganor gives a necessary and sufficient condition for this prefactor to vanish. Interestingly this can happen only when $`h^{2,1}(X)>0`$. So one would hope that in the Dynkin diagram configurations with โdissident divisorsโ the prefactor would actually be zero for the non dissident ones. An easy computation in the case of $`G_2`$, $`\mathrm{\S }`$4.4, shows that the prefactor is nevertheless non-zero for the divisors contributing to the superpotential.
Appendix A: Geometry of Z
A.1. The Divisors The polyhedron that we obtain by deleting the third column of $`^X`$ has the structure
| Relation to vertices | $`\chi `$ | $`^Z`$ | | | | | | Divisor |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| $`v_1`$ | $`0`$ | ( | -1, | 0, | 2, | 3 | ) | $`K_+`$ |
| $`v_2`$ | $`0`$ | ( | 0, | -1, | 2, | 3 | ) | $`K_{}`$ |
| $`v_3`$ | $`24`$ | ( | 0, | 0, | -1, | 0 | ) | $`2H^Z`$ |
| $`v_4`$ | $`57`$ | ( | 0, | 0, | 0, | -1 | ) | $`3H^Z`$ |
| $`\frac{1}{2}(v_1+v_6)=\frac{1}{2}(v_2+v_5)`$ | $`1`$ | ( | 0, | 0, | 2, | 3 | ) | $`B^Z`$ |
| $`v_5`$ | $`0`$ | ( | 0, | 1, | 2, | 3 | ) | $`K_{}`$ |
| $`v_6`$ | $`0`$ | ( | 1, | 2, | 2, | 3 | ) | $`K_+`$ |
| $`(h^{11},h^{21})=(3,243),\chi _E=480,(\delta ,\stackrel{~}{\delta })=(0,0)`$ | | | | | | | | |
| $`H^Z=B^Z+2K_++2K_{}`$ | | | | | | | | |
The manifold $`Z`$ is elliptically fibered over a base $`B^Z=\mathrm{IP}{}_{1}{}^{}\times \mathrm{IP}_1`$. We denote these two $`\mathrm{IP}{}_{1}{}^{}{}_{}{}^{}s`$ by $`_+`$ and $`_{}`$ and we see that the elliptic fibers over $`_\pm `$ form K3-surfaces
$$K_\pm =(,_\pm ).$$
The fan for $`Z`$ is
$$\mathrm{\Sigma }^Z=(v_1+v_6)(v_2+v_5)\mathrm{\Sigma }^{},\text{with}\mathrm{\Sigma }^{}=B^Z(v_3+v_4)+v_3v_4.$$
Given the fan SCHUBERT immediately provides the intersection numbers:
$$(B^Z)^3=8,(B^Z)^2.K_\pm =2,B^ZK_+K_{}=1,K_\pm ^2=0.$$
Finding the Kรคhler and Mori cones of $`Z`$ is an easy exercise in the method of piecewise linear functions (though more elementary procedures work in this case also). In a notation analogous to that introduced in Sect. 5.2 one is led to the following inequalities
$$\begin{array}{cc}\hfill 0& 2m_b+m_1+m_6\hfill \\ \hfill 0& 2m_b+m_2+m_5\hfill \\ \hfill 0& m_b6m_0+2m_3+3m_4\hfill \\ \hfill 0& 12m_0+m_1+4m_3+6m_4+m_6\hfill \\ \hfill 0& 12m_0+m_2+4m_3+6m_4+m_5.\hfill \end{array}$$
A basis is provided by the first three inequalities so taking a basis of divisors to be $`(B^Z,v_6,v_5)=(B^Z,K_+,K_{})`$ we see that the curves that generate the Mori cone are
$$\begin{array}{cc}& (2,1,0)\hfill \\ & (2,0,1)\hfill \\ & (1,0,0)\hfill \end{array}$$
which are the curves
$$_+=B^ZK_+,_{}=B^ZK_{},=K_+K_{}.$$
The generators of the Kรคhler cone are the divisors $`(K_{},K_+,H^Z)`$ that are dual to these curves.
We may write the Kรคhler-form as a linear combination of the generators
$$J^Z=tH^Z+u_{}K_++u_+K_{}.$$
Written this way the parameters are the volumes of the dual curves:
$$J^Z=t,J^Z_\pm =u_\pm .$$
We record here also the volume of $`Z`$ itself as well as that of the base $`B^Z`$ and that of the $`K_\pm `$.
$$\begin{array}{c}\text{ }\frac{1}{3!}(J^Z)^3=t\left[u_+u_{}+t(u_++u_{})+\frac{4}{3}t^2\right]\text{ }\hfill \\ \\ \text{ }\frac{1}{2!}(J^Z)^2B^Z=u_+u_{},\frac{1}{2!}(J^Z)^2K_\pm =t(u_\pm +t).\text{ }\hfill \end{array}$$
Note that, owing to the fact that the elliptic fiber varies over the base, the volume of $`Z`$ is not simply the volume of the base multiplied by the volume of the fiber unless $`t`$ is small.
Figure A.1: A sketch of $`Z`$ as an elliptic fibration over $`_+\times _{}`$ and as a $`K3`$-fibration in two ways as $`(K_+,_{})`$ and $`(K_{},_+)`$.
$``$ $`_+`$ $`_{}`$ $`K_+`$ $`K_{}`$
A.2. Projection to $`B^Z`$ As discussed in Section 3 the projection to $`Z`$ corresponds to projecting out the third component of the points of $`^X`$. In reality the projection to $`Z`$ exists in only a limited sense. There is a well defined projection to $`B^Z`$ which is a section of the fibration $`B=(\mathrm{IP}{}_{1}{}^{},B^Z)`$. The section is not unique nevertheless there are projections onto each of these sections.
We may also project the divisors of $`X`$ onto the divisors of $`Z`$. This proceeds in the following way. The projection of $`C_2(0,0,1,1,2)`$ is $`(0,0,1,2)`$ which is interior to a codimension one face of $`^Z`$. We therefore take $`C_2`$ to project to zero. In an analogous way we see that we should also take $`E_3`$ to project to zero. Now we see from the polyhedra that we should take
$$HH^Z,\text{and}Y^\pm K_\pm .$$
Now $`H=B+C_1+C_2+E_1+E_2+2Y^++2Y^{}`$ and $`H^Z=B^Z+2K_++2K_{}`$. So we take also
$$BB^Z\text{and}C_i0,E_j0\text{for all}i,j.$$
Now there is an element of choice in what we wish to call the preimage of $`B^Z`$ under this projection since we are free to add multiples of the divisors that project to zero. For the intersection calculation that follows it is sufficient to take this preimage to be $`\widehat{B}=B+C_1+E_1`$. We check that we obtain the correct values for the intersection numbers on $`Z`$:
| $`(B^Z)^3`$ | $`=\widehat{B}^2B^Z`$ | $`=\widehat{B}^2BE_1`$ | $`=8`$ |
| --- | --- | --- | --- |
| $`(B^Z)^2K_\pm `$ | $`=\widehat{B}Y^\pm B^Z`$ | $`=\widehat{B}Y^\pm BE_1`$ | $`=2`$ |
| $`B^ZK_+K_{}`$ | $`=Y^+Y^{}B^Z`$ | $`=Y^+Y^{}BE_1`$ | $`=1.`$ |
Note that we could take instead $`B^Z=BC_1`$ and these intersection numbers would still be correct. Now observe that since the images of $`F`$ and $`G`$ under the projection are both multiples of $`K_{}`$ we should set
$$F\alpha K_{}\text{and}G(1\alpha )K_{}$$
for some $`\alpha `$. It turns out that $`\alpha `$ is 0 or 1 depending on whether we take $`B^Z`$ to be $`BE_1`$ or $`BC_1`$ since in the first case $`B^Z`$ intersects $`G`$ but not $`F`$ while in the second $`B^Z`$ intersects $`F`$ but not $`G`$.
Appendix B: The Divisors for the Spaces $`\text{}^\pm `$
We record here Tables for the divisors of $`Y^+`$ and $`Y^{}`$. It is evident from the topological numbers that the two manifolds $`Y^\pm `$ are different.
| Relation to vertices | $`\chi `$ | $`^{Y^+}`$ | | | | | | Divisor |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| $`V_2^+`$ | $`0`$ | ( | -1, | 0, | 2, | 3 | ) | $`Y^+=F^++G^+`$ |
| $`V_3^+`$ | $`1`$ | ( | 0, | -1, | 2, | 3 | ) | $`C_1^+`$ |
| $`V_9^+`$ | $`1`$ | ( | 0, | -1, | 1, | 2 | ) | $`C_2^+`$ |
| $`V_4^+`$ | $`13`$ | ( | 0, | 0, | -1, | 0 | ) | $`2H^++E_3^+C_2^+`$ |
| $`V_5^+`$ | $`38`$ | ( | 0, | 0, | 0, | -1 | ) | $`3H^++E_3^+C_2^+`$ |
| $`\frac{1}{2}(V_3^++E_1^+)`$ | $`1`$ | ( | 0, | 0, | 2, | 3 | ) | $`B^+`$ |
| $`\frac{1}{2}(V_2^++V_F)`$ | $`1`$ | ( | 0, | 1, | 2, | 3 | ) | $`E_1^+`$ |
| $`V_8^+`$ | $`1`$ | ( | 0, | 2, | 2, | 3 | ) | $`E_2^+`$ |
| $`\frac{1}{2}(V_5^++V_8^+)`$ | $`2`$ | ( | 0, | 1, | 1, | 1 | ) | $`E_3^+=C_1^++C_2^+(E_1^++2E_2^++2F^++3G^+)`$ |
| $`V_F`$ | $`1`$ | ( | 1, | 2, | 2, | 3 | ) | $`F^+`$ |
| $`V_7^+`$ | $`1`$ | ( | 1, | 3, | 2, | 3 | ) | $`G^+`$ |
| $`H^+=B^++C_1^++C_2^++E_1^++E_2^++2Y^+`$ | | | | | | | | |
| $`(h^{11},h^{21})=(7,169),`$ $`(\delta ,\stackrel{~}{\delta })=(0,1),`$ $`\chi _E=324`$ | | | | | | | | |
Table B1: The divisors for $`Y^+`$.
| Relation to vertices | $`\chi `$ | $`^Y^{}`$ | | | | | | Divisor |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| $`V_1^{}`$ | $`0`$ | ( | -1, | 0, | 2, | 3 | ) | $`Y^{}`$ |
| $`V_3^{}`$ | $`1`$ | ( | 0, | -1, | 2, | 3 | ) | $`C_1^{}`$ |
| $`V_9^{}`$ | $`1`$ | ( | 0, | -1, | 1, | 2 | ) | $`C_2^{}`$ |
| $`V_4^{}`$ | $`14`$ | ( | 0, | 0, | -1, | 0 | ) | $`2H^{}+E_3^{}C_2^{}`$ |
| $`V_5^{}`$ | $`45`$ | ( | 0, | 0, | 0, | -1 | ) | $`3H^{}+E_3^{}C_2^{}`$ |
| $`\frac{1}{3}(2V_3^{}+E_2^{})`$ | $`1`$ | ( | 0, | 0, | 2, | 3 | ) | $`B^{}`$ |
| $`\frac{1}{2}(B^{}+E_2^{})`$ | $`1`$ | ( | 0, | 1, | 2, | 3 | ) | $`E_1^{}`$ |
| $`\frac{1}{2}(V_1^{}+V_6^{})`$ | $`1`$ | ( | 0, | 2, | 2, | 3 | ) | $`E_2^{}`$ |
| $`\frac{1}{2}(V_5^{}+E_2^{})`$ | $`1`$ | ( | 0, | 1, | 1, | 1 | ) | $`E_3^{}=C_1^{}+C_2^{}(E_1^{}+2E_2^{}+4Y^{})`$ |
| $`V_6^{}`$ | $`0`$ | ( | 1, | 4, | 2, | 3 | ) | $`Y^{}`$ |
| $`H^{}=B^{}+C_1^{}+C_2^{}+E_1^{}+E_2^{}+2Y^{}`$ | | | | | | | | |
| $`(h^{11},h^{21})=(8,194),`$ $`(\delta ,\stackrel{~}{\delta })=(0,2),`$ $`\chi _E=372`$ | | | | | | | | |
Table B2: The divisors for $`Y^{}`$.
References
relax1. relax C. Vafa, โEvidence for $`F`$-Theoryโ, Nucl. Phys. B469 (1996) 403, hep-th/9602022. relax relax2. relax S. Katz and C. Vafa, โGeometric engineering of $`N=1`$ quantum field theoriesโ, Nucl. Phys. B497 (1997) 196, hep-th/9611090. relax relax3. relax C. Vafa, โOn $`N=1`$ Yang-Mills in four dimensionsโ, hep-th/9801139. relax relax4. relax M. Bershadsky, K. Intriligator, S. Kachru, D. R. Morrison, V. Sadov, and C. Vafa, โGeometric singularities and enhanced gauge symmetriesโ, Nucl. Phys. B481 (1996) 215, hep-th/9605200. relax relax5. relax E. Witten, โNonperturbative Superpotentials in String Theoryโ, Nucl. Phys. B474 (1996) 343, hep-th/9604030. relax relax6. relax D. Diaconescu, private communication relax relax7. relax O. J. Ganor, โA Note On Zeroes Of Superpotentials in FโTheoryโ, Nucl. Phys. B499 55 (1997), hep-th/9612077 relax relax8. relax D. A. Cox and S. Katz, โMirror Symmetry and Algebraic Geometryโ, AMS 1999. relax relax9. relax B. Szendrรถi, โOn an Example of Aspinwall and Morrisonโ, math.AG/9911064. relax relax10. relax V. Batyrev, J. Alg. Geom. 3 (1994) 493โ535, alg-geom/9310003. relax relax11. relax P. S. Aspinwall, B. R. Greene and D. R. Morrison, โCalabiโYau Moduli Space, Mirror Manifolds and Space-Time Topology Change in String Theoryโ, Nucl. Phys. B416 (1994) 414, hep-th/9309097. relax relax12. relax A. Klemm, B. Lian, S. S. Roan and S. T. Yau, โCalabiโYau Fourfolds for $`M`$ Theory and $`F`$ Theory Compactificationsโ, Nucl. Phys. B518 (1998) 515, hep-th/9701023. relax relax13. relax D. Dais and V. Batyrev, โStrong McKay Correspondence, String-Theoretic Hodge Numbers and Mirror Symmetryโ, Topology 35, No.4, 901 (1996), alg-geom/9410001. relax relax14. relax R. Donagi, A. Grassi and E. Witten, โA non-perturbative superpotential with $`E_8`$ symmetryโ, Mod. Phys. Lett. A17 (1996), hep-th/9607091. relax relax15. relax A. Avram, M. Mandelberg, M. Kreuzer and H. Skarke, โSearching for $`K3`$ Fibrationsโ, Nucl. Phys. B505 (1997) 625, hep-th/9610154. relax relax16. relax P. Candelas and A. Font, โDuality Between the Webs of Heterotic and Type II Vacuaโ, Nucl. Phys. B511 (1998) 295, hep-th/9603170. relax relax17. relax P. Candelas, E. Perevalov and G. Rajesh, โToric Geometry and Enhanced Gauge Symmetry of F Theory/Heterotic Vacuaโ, Nucl. Phys. B507 (1997) 445, hep-th/9704097 relax relax18. relax D. Morrison and C. Vafa, โCompactifications of $`F`$-theory on Calabi-Yau threefolds, I, IIโ, Nucl. Phys. B473 (1996) 476, hep-th/9602114/9603161. relax relax19. relax T. Oda and H. S. Park, โLinear Gale Transforms and Gelโfand-Kapranov-Zelevinskij Decompositionsโ, Tohoku Math. J. (2) 43 (1991), no. 3, 375โ399. relax relax20. relax K. Intriligator, D. R. Morrison and N. Seiberg, โFive-Dimensional Supersymmetric Gauge Theories and Degenerations of CalabiโYau Spacesโ, Nucl. Phys. B497 (1997) 56, hep-th/9702198. relax relax21. relax D. Diaconescu and G. Rajesh, โGeometrical Aspects of Five-Branes in Heterotic / F Theory Duality in Four-Dimensionsโ, JHEP 9906 (1999) 002, hep-th/9903104 relax relax22. relax A. Grassi, โDivisors 0n Elliptic CalabiโYau Four Folds and the Superpotential in F Theory. 1โ, J. Geom. Phys. 28 (1998) 289 relax relax23. relax P. Mayr,โMirror Symmetry, $`N=1`$ superpotentials and Tensionless Strings on Calabi-Yau Four-Foldsโ, hepth/9610162. relax relax24. relax B. Andreas, โN=1 Heterotic / F Theory Dualityโ, Fortsch. Phys. 47 (1999) 587, hep-th/9808159. relax relax25. relax P. M. H. Wilson, โThe Kaehler Cone on Calabi-Yau Threefoldsโ, Inventiones mathematicae 107 (1992). relax relax26. relax D. Diaconescu, Private comunication. relax relax27. relax O. J. Ganor, โA Note on Zeros of Superpotentials in F Theoryโ, Nucl. Phys. B499 (1997) 55, hep-th/9612077. relax |
warning/0001/hep-ph0001147.html | ar5iv | text | # On conditions for the nonperturbative equivalence of ultraviolet cut-off and dimensional regularization schemes
## I Introduction
In order to study quantum field theories in the nonperturbative regime it is essential to have appropriate regularization schemes which respect the symmetries of the underlying theory. In the lattice approach to nonperturbative studies of gauge theories the gauge fields are represented by links and the action is formulated in terms of these links in order to maintain gauge invariance by construction. This is the case even though the finite lattice spacing is acting as a form of ultraviolet regulator. It is useful to augment lattice studies by other nonperturbative methods such as studies of Dyson-Schwinger equations (DSE). Studies using the DSE must always involve some form of truncation of the infinite tower of coupled integral equations and hence can never be a first principles approach. Nevertheless, DSE are a useful complement to lattice studies and through their self-consistent nature can provide important insights into the nonperturbative behavior of quantum field theories. In addition, information obtained from the lattice can place significant constraints on the DSE approach, which can further enhance its usefulness.
One difficulty facing DSE integral equation studies is that the use of an ultraviolet (UV) cut-off to regulate the integrations will in general lead to an explicit violation of gauge invariance. In other words, the renormalized Greenโs functions calculated within a DSE study using a UV cut-off will in general contain unacceptable explicit gauge-invariance violating contributions, unless specific steps are taken to remove them. On the other hand, DSE studies implemented using a gaugeโinvariant regularization scheme such as dimensional regularization will have no such undesirable explicit gauge-invariance violating contributions.
Only recently have explicit numerical DSE studies been succesfully performed using dimensional regularization. The subject of these initial studies was quenched QED<sub>4</sub>, which, while not a physically realistic theory, has the advantage of being simple enough that it is an excellent testing ground for nonperturbative techniques such as DSE and lattice studies. Renormalized quantities calculated within a dimensional regularization scheme can be compared directly with those obtained using a UV cut-off scheme, (provided of course that the same renormalization conditions are imposed). This is just what was done for studies of the fermion propagator and dynamical chiral symmetry breaking using dimensional regularization in Refs. , where direct comparisons of results for the fermion propagator and critical coupling $`\alpha _c`$ were made with results obtained using ultraviolet cut-off regularization . It was found that, with an appropriate modification to the naive UV cut-off treatment and within the currently achieved numerical precision, the results were the same as those obtained from the more computationally demanding dimensional regularization approach. In nonperturbative studies the UV cut-off regularization scheme can have significant computational advantages over the dimensional regularization scheme, where a careful extrapolation to the $`ฯต0`$ limit must be taken numerically.
The purpose of the present work is to exploit this recent development of nonperturbative dimensional regularization in order to help motivate and establish general principles for removing the unwanted explicit gauge-violating contributions in the UV cut-off regularization approach. This is to be achieved by imposing translational invariance on to cut-off regularization. In order to achieve this we are free to add terms which would vanish in any translationally invariant scheme. Hence one must choose an arbitrary centre for the 4-dimensional momentum cut-off hypersphere and then add terms which will be designed to produce a translationally invariant result. In order for this program to be successful, one needs sufficient constraints to fix the free parameters in order to arrive at a uniquely defined, translationally invariant answer. The parameters are fixed by eliminating worse than logarithmically divergent terms, and by requiring consistency with perturbation theory dimensional regularization in the weak-coupling limit. In Sec. II we present the formalism for the fermion DSE and briefly summarize and compare the numerical studies for the renormalized nonperturbative fermion propagator using the two schemes. In Sec. III we analyse translational invariance theoretically for perturbative massless and massive QED<sub>4</sub>. We also provide a derivation of the modified cut-off regularization scheme which agrees with the translationally invariant dimensional regularization scheme. Finally, in Sec. IV we summarize and conclude.
## II Fermion Dyson-Schwinger Equation in Quenched QED<sub>4</sub>
The DSE for fermion propagator in quenched QED<sub>4</sub> can be represented diagramatically as :
Making use of the Feynman rules for this diagram leads to :
$`\mathrm{๐๐}_F^1=\mathrm{๐๐}_{F}^{0}{}_{}{}^{1}{\displaystyle _M}{\displaystyle \frac{d^4k}{(2\pi )^4}}(ie\mathrm{\Gamma }_F^\mu )iS_F(k^2)(ie\gamma ^\nu )i\mathrm{\Delta }_{\mu \nu }(q^2),`$ (1)
where introducing several frequently used notations at once :
The Full Fermion Propagator is :
$`iS_Fi{\displaystyle \frac{F(p^2)}{\overline{)}pM(p^2)}}i{\displaystyle \frac{Z(p^2)}{\overline{)}pM(p^2)}}i{\displaystyle \frac{1}{A(p^2)\overline{)}pB(p^2)}},`$ (2)
and $`F(p^2)Z(p^2)1/A(p^2)`$ is the fermion wave-function renormalization, $`M(p^2)B(p^2)/A(p^2)Z(p^2)B(p^2)`$ is the dynamical fermion mass.
The Full Photon Propagator is :
$`i\mathrm{\Delta }_{\mu \nu }(p^2)={\displaystyle \frac{i}{p^2}}\left[G(p^2)\left(g_{\mu \nu }{\displaystyle \frac{p_\mu p_\nu }{p^2}}\right)+\xi {\displaystyle \frac{p_\mu p_\nu }{p^2}}\right],`$ (3)
where $`G(p^2)1/(1+\mathrm{\Pi }(p^2))`$ is the boson wave-function renormalization which is $`1`$ in the quenched approximation, and $`\xi `$ the covariant gauge parameter. We shall call $`\left(g_{\mu \nu }p_\mu p_\nu /p^2\right)`$ the transverse part and $`p_\mu p_\nu /p^2`$ the longitudinal part. Finally, $`\mathrm{\Gamma }^\mu `$ is the full fermion-boson vertex for which we use the CP ansatz, namely ($`qkp`$)
$`\mathrm{\Gamma }^\mu (k,p,q)=\mathrm{\Gamma }_{\mathrm{BC}}^\mu (k,p)+\tau _{\mathrm{CP}}(k^2,p^2)\left[\gamma ^\mu (p^2k^2)+(p+k)^\mu \overline{)}q\right],`$ (4)
where $`\mathrm{\Gamma }_{\mathrm{BC}}`$ is the usual Ball-Chiu part of the vertex which satisfies the Ward-Takahashi identity
$`\mathrm{\Gamma }_{\mathrm{BC}}^\mu (k,p)`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left({\displaystyle \frac{1}{F(k^2)}}+{\displaystyle \frac{1}{F(p^2)}}\right)\gamma ^\mu `$ (5)
$`+`$ $`{\displaystyle \frac{(k+p)^\mu }{k^2p^2}}\left\{\left({\displaystyle \frac{1}{F(k^2)}}{\displaystyle \frac{1}{F(p^2)}}\right){\displaystyle \frac{(\overline{)}k+\overline{)}p)}{2}}\left({\displaystyle \frac{M(k^2)}{F(k^2)}}{\displaystyle \frac{M(p^2)}{F(p^2)}}\right)\right\},`$ (6)
and the coefficient function $`\tau _6`$ is that chosen by Curtis and Pennington, i.e.,
$`\tau _{\mathrm{CP}}(k^2,p^2)={\displaystyle \frac{1}{2d(k,p)}}\left({\displaystyle \frac{1}{F(k^2)}}{\displaystyle \frac{1}{F(p^2)}}\right),`$ (7)
where
$`d(k,p)={\displaystyle \frac{(k^2p^2)^2+[M^2(k^2)+M^2(p^2)]^2}{k^2+p^2}}.`$ (8)
For studies of the nonperturbative renormalized fermion propagator, dynamical chiral symmetry breaking, and critical coupling $`\alpha _c`$ in quenched QED<sub>4</sub>, it is necessary to choose a specific form for the fermion-photon proper vertex. While the exact form of this vertex is not known, there are constraints and symmetries which strongly restrict the allowable form. These include the Ward-Takahashi identity (WTI)), the absence of artificial kinematic singularities, the requirements of multiplicative renormalizability (MR), and the need to agree with perturbation theory in the weak coupling limit. Furthermore, one should eventually ensure that the gauge dependence of the resulting fermion propagator is consistent with the Landau-Khalatnikov transformation and that the value of the dynamical chiral symmetry breaking critical coupling ($`\alpha _c`$) should be gauge independent.
A number of discussions of the choice of the transverse part of the proper vertex can be found in the literature, e.g., Refs. . We consider for our illustration here only the Curtis-Pennington (CP) vertex, which satisfies both the WTI and the constraints of multiplicative renormalizability. It is known that this vertex is not entirely adequate to ensure the gauge invariance of $`\alpha _c`$, although it is superior in this regard to the bare vertex for example. The choice of a vertex satisfying the WTI is a necessary but not sufficient condition to ensure the full gauge covariance of the Greenโs functions of the theory and the gauge invariance of physical observables. Bashir and Pennington have proposed alternatives to the CP vertex, which ensure by construction that the critical coupling indeed becomes strictly gauge independent. All of the above studies were carried out with a UV cut-off regularization, where it was necessary to remove by hand an obvious explicit gauge-invariance violating term which arose from the cut-off itself, (see for example Ref. for a detailed discussion of this). We emphasize that the issue of interest here is not the construction of the ideal fermion-photon proper vertex, but rather to understand and remove explicit gauge-invariance violations arising from the UV cut-off regulator itself. For that purpose the CP vertex provides a useful illustrative example.
### A Numerical Studies of the Fermion DSE
In quenched QED there is no renormalization of the electron charge and the appropriate photon propagator is just the bare one. The resulting nonlinear integral equation for the fermion propagator is solved numerically. The general features of dimensional regularization can be found in any recent textbook (e.g., Ref. ), and we use the notation $`D=42ฯต<4`$ for the dimension of Euclidean space. Successive calculations with decreasing $`ฯต`$ must be numerically extrapolated to $`ฯต=0`$. For the UV cut-off regularization we simply cut-off the Euclidean four-dimensional loop integral at $`\mathrm{\Lambda }`$ and verify that $`\mathrm{\Lambda }`$ is sufficiently large. Clearly there is an ambiguity about exactly which loop momentum variable has the cut-off applied to it. In all calculations to date which use a UV cut-off a formal numerical extrapolation to $`\mathrm{\Lambda }\mathrm{}`$ has not been necessary.
The formalism is presented in Minkowski space and the Wick rotation into Euclidean space can then be performed once the equations to be solved have been written down. Although we use dimensional regularization, we cannot make use of the popular perturbative renormalization schemes such as $`MS`$ or $`\overline{MS}`$, since they cannot be applied in a nonperturbative context.
In the following equations and definitions, we will use $`ฯต`$ to denote generic regularization dependence, where we must take the generic $`ฯต0`$ in the limit where the regularization is removed. For the UV cut-off case we understand $`ฯต1/\mathrm{\Lambda }`$. The renormalized inverse fermion propagator is defined through
$`S_R^1(\mu ;p)={\displaystyle \frac{1}{F_R(\mu ;p^2)}}\overline{)}p{\displaystyle \frac{M_R(\mu ;p^2)}{F_R(\mu ;p^2)}}`$ $`=`$ $`Z_2(\mu ,ฯต)[\overline{)}pm_0(ฯต)]\mathrm{\Sigma }_0(\mu ,ฯต;p),`$ (9)
$`=`$ $`\overline{)}pm(\mu )\mathrm{\Sigma }_R(\mu ;p),`$ (10)
where $`\mu `$ is the chosen renormalization scale, $`m(\mu )`$ is the value of the renormalized mass at $`p^2=\mu ^2`$, $`m_0(ฯต)`$ is the bare mass and $`Z_2(\mu ,ฯต)`$ is the wave-function renormalization constant. Due to the WTI for the fermion-photon proper vertex, we have for the vertex renormalization constant $`Z_1(\mu ,ฯต)=Z_2(\mu ,ฯต)`$. The renormalized and unrenormalized fermion self-energies are denoted as $`\mathrm{\Sigma }_R(\mu ;p)`$ and $`\mathrm{\Sigma }_0(\mu ,ฯต;p)`$ respectively. These can be expressed in terms of Dirac and scalar pieces
$$\mathrm{\Sigma }_0(\mu ,ฯต;p)=\mathrm{\Sigma }_0^d(\mu ,ฯต;p^2)\overline{)}p+\mathrm{\Sigma }_0^s(\mu ,ฯต;p^2),$$
(11)
and similarly for $`\mathrm{\Sigma }_R(\mu ;p)`$. We do not explicitly indicate the dependence on $`ฯต`$ of the renormalized quantities $`1/F_R(\mu ;p^2)`$, $`M_R(\mu ;p^2)/F_R(\mu ;p^2)`$ and $`\mathrm{\Sigma }_R(\mu ;p)`$, since for renormalized quantities we will always be interested in their $`ฯต0`$ limit. The renormalized mass function $`M(p^2)`$ is renormalization point independent due to the nature of multiplicative renormalizability . The renormalization point boundary condition
$$S_R^1(\mu ;p)|_{p^2=\mu ^2}=\overline{)}pm(\mu ),$$
(12)
implies that $`F_R(\mu ;\mu ^2)1`$ and $`m(\mu )M(\mu ^2)`$ and gives
$$\mathrm{\Sigma }_R^{d,s}(\mu ;p^2)=\mathrm{\Sigma }_0^{d,s}(\mu ,ฯต;p^2)\mathrm{\Sigma }_0^{d,s}(\mu ,ฯต;\mu ^2).$$
(13)
Also, the wave-function renormalization is given by
$$Z_2(\mu ,ฯต)=1+\mathrm{\Sigma }_0^d(\mu ,ฯต;\mu ^2),$$
(14)
and for the bare mass $`m_0(ฯต)`$
$$m_0(ฯต)=\left[m(\mu )\mathrm{\Sigma }_0^s(\mu ,ฯต;\mu ^2)\right]/Z_2(\mu ,ฯต).$$
(15)
Under a renormalization point transformation $`\mu \mu ^{}`$, $`m(\mu ^{})=M(\mu _{}^{}{}_{}{}^{2})`$ and $`Z_2(\mu ^{},ฯต)=Z_2(\mu ,ฯต)/F(\mu ^{};\mu ^2)`$ as discussed in Ref. . Since we are working here in the quenched approximation we have $`Z_3(\mu ,ฯต)=1`$, $`e_0e(\mu )`$, and the photon propagator $`\mathrm{\Delta }^{\mu \nu }(\mu ;q)`$ has its perturbative form where $`G(p^2)=1`$.
The unrenormalized self-energy is given by the integral
$$\mathrm{\Sigma }_0(\mu ,ฯต;p)=i(e(\mu )\nu ^ฯต)^2\frac{d^Dk}{(2\pi )^D}\gamma ^\lambda S(\mu ;k)\mathrm{\Gamma }^\nu (\mu ;k,p)\mathrm{\Delta }_{\lambda \nu }(\mu ;pk),$$
(16)
where $`\nu `$ is an arbitrary mass scale introduced in $`D42ฯต`$ dimensions so that the renormalized coupling $`e(\mu )`$ remains dimensionless in the dimensional regularization scheme. For the UV cut-off case we have no need of $`\nu `$ since $`ฯต=0`$ but instead we integrate over a four-dimensional sphere whose radius is the UV cut-off $`\mathrm{\Lambda }`$. The center of this sphere is often taken to be $`k_\mu =0`$, although one could equally well choose it to be at any location, e.g., at any $`(k_\mu +cp_\mu +b_\mu )=0`$, where $`c`$ is an arbitrary real constant and $`b_\mu `$ is an arbitrary Euclidean four-momentum.
### B Numerical Comparison
In Ref. , the renormalized dimensionally regularized fermion DSE for the Curtis-Pennington vertex in quenched approximation was studied numerically. Therein, it was noted that the fermion propagator extrapolated to $`ฯต=0`$ using dimensional regularization differed from that obtained from using a cut-off regulator โas isโ, but agreed with that obtained by using a cut-off regulator with the modification proposed by Ref. , within the numerical accuracy of the study. This was observed for a massive solution with the coupling $`\alpha =1.5`$ and the gauge parameter $`\xi =0.25`$. Here we explore whether this agreement holds at much increased numerical precision for the more numerically tractable case of $`\alpha =0.6`$ in a variety of gauges for both massive and massless solutions of the quenched fermion DSE.
#### 1 Massive Case
Fig 4 shows a family of solutions calculated in dimensional regularization scheme with the regulator parameter $`ฯต`$ decreased from 0.08 to 0.03 for the coupling $`\alpha =0.6`$. The gauge parameter is $`\xi =0.25`$, the renormalization point $`\mu ^2=10^8`$ and the renormalized mass is $`m(\mu )=400`$. \[Note that we have chosen our units such that $`\mu ^2=10^8`$ and $`m(\mu )=400`$ in those units. We could equally well choose these units to be MeV, eV, GeV, etc. For a given solution we can simply multiply all mass scales in the problem by the same arbitrary constant and we still have a valid solution\]. It is important to note the strong dependence of the solutions on $`ฯต`$, even though this parameter is already rather small. The ultraviolet is most sensitive to this regulator, however even in the infrared there is considerable dependence due to the intrinsic coupling between these regions by the renormalization procedure. This strong dependence on $`ฯต`$ should be contrasted with the situation in cut-off based studies where at rather modest cut-offs ($`\mathrm{\Lambda }^210^{10}`$) the renormalized functions $`A`$ and $`M`$ had already reached their asymptotic limits.
Also shown is the result of extrapolating these solutions to $`ฯต=0`$ by fitting a polynominal quartic in $`ฯต`$ at each momentum point. As was observed in , the linearity and stability of the extrapolation may be improved by a suitable choice of scale,$`\nu `$ : of the scales 1, 10, 100, 1000, and 10000, the latter two fit these criteria best. We show $`\nu =10000`$ on the graphs.
Fig. 5 shows this extrapolated solution along with the corresponding cut-off results, both with and without the aforementioned modification which eliminates a spurious term induced by the cut-off which breaks translational invariance. As may be clearly seen from the insert of Fig 5 the modified ultaviolet cut-off curve is indisguishable from the scaled dimensionally regularized one, while the naive cut-off curve clearly deviates from the others in the infrared. This observation is quantified by tables I and II which show absolute percentage comparisons of the finite renormalization and the mass function for the extrapolated solution with the modified and naive UV cut-off massive solutions respectively, with parameters as in Fig. 4 for two different scales and two different polynomial degrees. The agreement with the modified cut-off solution is seen to be excellent, and is three orders of magnitude better than the agreement with the naive cut-off.
Finally, Fig 6 shows a comparison in three different gauges, namely $`\xi =0`$, $`\xi =0.25`$ and $`\xi =1`$, of solutions of the fermion DSE extrapolated to $`ฯต=0`$, and solutions using naive and modified UV cut-off regulators, with other parameters the same as in Fig. 4. The $`A(p^2)`$ solutions are identical in Landau gauge, and the agreement between the extrapolated solution and the modified cut-off solution is readily distinguished in Feynman gauge. Owing to the approximate gauge invariance of the mass function, an insert is neccessary to reveal the same holds for $`M(p^2)`$.
#### 2 Massless Case
Fig 7 shows a family of solutions calculated in the dimensional regularization scheme with the regulator parameter $`ฯต`$ from from 0.04 to 0.005 for the coupling $`\alpha =0.6`$, the gauge parameter $`\xi =0.25`$ and the renormalization point $`\mu ^2=10^8`$. As in the massive case, there is a strong dependence of the renormalized function $`A(p^2)`$ on the regulator $`ฯต`$, although here the infrared is even more sensitive than the ultraviolet to $`ฯต`$. This contrasts with cut-off solutions which reach their asymptotic limit at rather modest cut-offs.
Also shown is the result of extrapolating these solutions to $`ฯต=0`$ by fitting a polynominal quartic in $`ฯต`$ at each momentum point to $`\mathrm{log}_{10}(A)`$. This was appropriate because the the logarithmically scaled axes reveals the power-law character of the extrapolated solution.
In Fig 8 we show this extrapolated solution along with curves of the naive and modified cut-off solutions based on the power-behaved analytical formulae Eqs. (72) and (77) discussed in the next section of the form
$`F(p^2)=\left(p^2/\mathrm{\Lambda }^2\right)^\gamma .`$ (17)
The curves are indistinguishable on the main figure: an insert reveals the extrapolated solution agrees with the modified cut-off solution. This is quantified by table III which shows absolute percentage comparisons of $`A(p^2)`$ for these solutions. As in the massive case, the agreement with the modified cut-off solution is many orders of magnitude better than that with the naive cut-off.
Finally Fig 9 shows a comparison in three different gauges, namely $`\xi =0`$, $`\xi =0.25`$ and $`\xi =1`$, of solutions of the fermion DSE extrapolated to $`ฯต=0`$, and solutions using naive and modified UV cut-off regulators, with other parameters the same as in Fig. 7. As in the massive case, the $`A(p^2)`$ solutions are identical in Landau gauge, and the agreement between the extrapolated and modified cut-off solution is clear in Feynman gauge.
## III Theoretical Studies on Translational Invariance of the DSE
The DSE approach to calculating any nonperturbative renormalized Greenโs function in a renormalizable quantum field theory involves an integral over a loop momentum, where the integrand involves two or more renormalized nonperturbative Green functions. If the Greenโs function to be calculated from the loop integral corresponds to one of the primitively divergent diagrams, then regularization of the loop integration and renormalization will in general be necessary. The primitively divergent diagrams in QED are the 2 and 3-point Green functions, i.e., the fermion and photon propagators and the fermion-photon proper (i.e., one-particle irreducible) vertex. For other higher $`n`$-point Greenโs functions the loop integrations in the DSE formalism are necessarily finite and renormalization of these quantities is not needed once the nonperturbative primitively divergent diagrams have been renormalized.
For instance the fermion self energy part of Eq. (1), as it stands, has a linear ultraviolet divergence because the integrand behaves like $`\frac{d^4k}{k^3}`$ for large $`k`$. Therefore, an ultraviolet regulator must be introduced in order to perform the loop integral. When the regulator is removed by renormalizing the theory, one should be left with a finite quantity which is independent of the scheme used. If two regularization schemes give different results then some symmetries have been violated by one (or both) of the regulators . The regulatization scheme should be chosen carefully so that gauge invariance and Poincare symmmetry in QED are preserved. For instance, while the Pauli Villars and dimensional regularization schemes respect gauge and translational invariance, UV cut-off regularization does not. However, one can attempt to use a UV cut-off regulator and still preserve these symmetries by imposing them on the regulator itself. Whether or not this procedure will be unique and conserve all symmetries is the key question. Here, a translationally invariant regularization scheme is defined to mean that the same results are obtained after arbitrary shifts in the definition of the loop mometum variable in the limit that the regularization is removed. Since a UV cut-off regularization is a restriction of the Euclidean loop-momentum integral to a four-dimensional hypersphere of radius $`\mathrm{\Lambda }`$, we wish to ensure that in the limit $`\mathrm{\Lambda }\mathrm{}`$ we find that the results are insensitive to the location of the centre of this hypersphere.
We are interested in establishing necessary and sufficient conditions for a UV cut-off regularization scheme to reproduce the results of a dimensional regularization scheme for the renormalized $`n`$-point Greenโs functions. The scope of this present work is to establish a procedure for this for the electron self-energy. The procedures for removing unwanted contributions in a UV cut-off scheme are straightforward: (1) The best way to begin identifying such terms is to replace nonperturbative quantities in the integrand by their perturbative form; (2) Test that the resulting expression for the integral of the nonperturbative renormalized quantity is independent of the location of the hyperspheres center in the limit $`\mathrm{\Lambda }\mathrm{}`$. Let us begin with an analysis of perturbation theory in order to understand the problem :
### A Perturbation Theory as a Guide
Taking the weak coupling limit of Eq. (1) for massless QED<sub>4</sub> up to $`๐ช(\alpha )`$, substituting in the fermion and photon propagators and the 3-point vertex function, multiplying it by $`\overline{)}p`$ and taking its trace leads to the following expression :
$`{\displaystyle \frac{1}{F(p^2)}}=1+{\displaystyle \frac{i\alpha }{4\pi ^3p^2}}{\displaystyle _M}{\displaystyle \frac{d^4k}{k^2q^4}}\left\{\left(2kpq^2\right)+(\xi 1)\left(k^2pqp^2kq\right)\right\},`$ (18)
$`=`$ $`1+{\displaystyle \frac{i\alpha }{4\pi ^3p^2}}{\displaystyle _M}{\displaystyle \frac{d^4k}{k^2q^4}}\left\{\underset{}{3kp(k^2+p^2)+4(kp)^2+2k^2p^2}+\xi \underset{}{\left(k^2pqp^2kq\right)}\right\}.`$
$`I_T+I_L`$
where โMโ denotes Minkowski space.
ANALYSIS:
(1) :
First we shall calculate the fermion wave-function renormalization within Dimensional Regularization, which respects the symmetries of QED. It is important to note that the first term and the coefficient of $`(\xi 1)`$ in the curly bracket of Eq. (18) give the same answer implying that the non-$`\xi `$ part, $`I_T`$, of Eq. (LABEL:eq:perfermion2) vanishes and the $`\xi `$ part, $`I_L`$, yields the result as below :
$`{\displaystyle \frac{1}{F(p^2)}}=1{\displaystyle \frac{\alpha \xi }{4\pi }}\left[{\displaystyle \frac{1}{ฯต}}+\mathrm{ln}\left({\displaystyle \frac{p^2}{\nu ^2}}\right)1+\gamma \mathrm{ln}(4\pi )\right].`$ (20)
(2) :
Repeating the same calculation as above except this time using Cut-off Regularization we get with the hypersphere centre at $`k_\mu =0`$ :
$`{\displaystyle \frac{1}{F(p^2)}}=1{\displaystyle \frac{\alpha \xi }{4\pi }}\left(\mathrm{ln}{\displaystyle \frac{p^2}{\mathrm{\Lambda }^2}}\text{ }{\displaystyle \frac{\mathrm{๐}}{\mathrm{๐}}}\text{ }\right).`$ (21)
Once again the integral of $`I_T`$ from Eq. (LABEL:eq:perfermion2) is zero.
The $`1/2`$ term in Eq. (21) is a consequence of cut-off regularization not preserving translational invariance and gauge covariance; in other words it is due to the non-conservation of current. The Landau-Khalatnikov (LK) transformations determines the gauge covariance of the theory in that if any Greenโs functions of the theory are known in one gauge they are also known for any other gauge via this transformation. If in the Landau gauge, the fermion wave-function renormalization is $`F(p^2,\mathrm{\Lambda }^2)=A_0(p^2/\mathrm{\Lambda }^2)^{\gamma _0}`$ then this transforms to the covariant gauge as $`F(p^2,\mathrm{\Lambda }^2)=A(p^2/\mathrm{\Lambda }^2)^\gamma `$ where $`\gamma =\gamma _0+\alpha \xi /4\pi `$ and $`A,A_0`$ are constants,. $`\gamma _0`$ is gauge independent term and in perturbation theory of $`๐ช(\alpha ^2)`$, . Therefore, if one performs the perturbative expansion of $`F(p^2)=A(p^2/\mathrm{\Lambda }^2)^\gamma `$ for small $`\alpha `$, then the $`1/2`$ term in Eq. (21) should be absent in order to ensure that the solution of the DSE for fermion wave-function renormalization is LK covariant.
As we shall see below this term can be removed by making use of the WTI or symmetry properties.
*
* The WTI follows from gauge invariance. Applying it to the $`\xi `$-part of photon propagator in Eq. (1), separates out the term, $`pq/q^4`$, which is zero in any translational invariant scheme (odd in $`q`$). On the other hand, in cut-off regularization it is the source of the $`1/2`$ term in Eq. (21).
* To analyse the translational invariance in cut-off regularization we shall shift the centre of the sphere from $`k_\mu =0`$ to $`k_\mu =cp_\mu `$, where $`c`$ is an arbitrary real constant, in Eq. (18) and Eq. (LABEL:eq:perfermion2). In so doing we obtain the following equations :
$`{\displaystyle ^\mathrm{\Lambda }}d^4k{\displaystyle \frac{2kp}{k^2(kp)^2}}`$ $`=`$ $`\pi ^2p^2\left\{\mathrm{ln}{\displaystyle \frac{p^2}{\mathrm{\Lambda }^2}}{\displaystyle \frac{1}{2}}\right\}`$ (23)
$`\stackrel{kkcp}{}\pi ^2p^2\left\{\mathrm{ln}{\displaystyle \frac{p^2}{\mathrm{\Lambda }^2}}+\left({\displaystyle \frac{1}{2}}+c\right)+๐ช(\mathrm{\Lambda }^1)\right\},`$
$`{\displaystyle ^\mathrm{\Lambda }}d^4k{\displaystyle \frac{kqp^2}{k^2(kp)^4}}`$ $`=`$ $`\pi ^2p^2\left\{\mathrm{ln}{\displaystyle \frac{p^2}{\mathrm{\Lambda }^2}}\right\}`$ (25)
$`\stackrel{kkcp}{}\pi ^2p^2\left\{\mathrm{ln}{\displaystyle \frac{p^2}{\mathrm{\Lambda }^2}}+๐ช(\mathrm{\Lambda }^1)\right\},`$
$`{\displaystyle ^\mathrm{\Lambda }}d^4k{\displaystyle \frac{pq}{(kp)^4}}`$ $`=`$ $`\pi ^2p^2\left\{{\displaystyle \frac{1}{2}}\right\}`$ (27)
$`\stackrel{kkcp}{}\pi ^2p^2\left\{{\displaystyle \frac{(c+1)}{2}}\right\},`$
$`{\displaystyle ^\mathrm{\Lambda }}d^4k{\displaystyle \frac{3kp(k^2+p^2)+4(kp)^2+2k^2p^2}{k^2(kp)^4}}`$ $`=`$ $`0`$ (28)
$`\stackrel{kkcp}{}`$ $`\pi ^2p^2\left\{{\displaystyle \frac{3c}{2}}+๐ช(\mathrm{\Lambda }^1)\right\}.`$ (29)
Examination of Eqs. (23)- (29) reveals that with the exception of Eq. (25) (which is logarithmically divergent) all others are linearly divergent and VIOLATE translational invariance :
Looking at the first term in $`I_L`$, which corresponds to Eq. (27), the WTI helps us to immediately recognize it as odd and linearly divergent. In cut-off regularization the position of the sphere is very important for the regularized quantities. If it is placed at $`k_\mu =0`$ then the $`1/2`$ term is generated in Eq. (21) and Eq. (27) which is consequence of the violation of translational invariance. On the other hand if the centre is located at $`k_\mu =p_\mu `$ i.e. $`c=1`$ then Eq. (27) is zero, translational invariance is preserved and cut-off is consistent with dimensional regularization.
The second term in $`I_L`$ has only a logarithmic divergence and when it is shifted arbitrarily, the difference between the shifted and unshifted value vanishes as $`\mathrm{\Lambda }\mathrm{}`$ so this term is translational invariant even under cut-off regularization, Eq. (25).
Finally we shall consider the transverse part, $`I_T`$ of Eq. (LABEL:eq:perfermion2) or Eq. (29). For large $`k`$ it becomes :
$`I_T`$ $`=`$ $`{\displaystyle d^4k\left\{\frac{3kp}{k^4}+\frac{1}{k^6}\underset{}{\left(4(kp)^2+k^2p^2\right)}\right\}},`$ (31)
$`0`$
as can be seen this expression, is zero (convergent) in any translationally invariant scheme. The reason for this is that the integral of the linearly divergent (also odd in $`k`$) term ($`3kp/k^4`$) is zero and the logarithmically divergent second and third terms cancel each other out in any translationally invariant scheme, since they cancel after the angular integration about $`k_\mu =0`$. If Eq. (31) is centred at $`k_\mu =0`$ then even in cut-off regularization, the integral of the first term is identically zero. Conversely, if the hypersphere is centred at $`k_\mu 0`$ the integral of the first term is non-zero. For instance, in Eq. (29), $`k_\mu =cp_\mu `$ will give $`3c/2`$. Hence, in order to be consistent with the translational invariant regularization $`c`$ must be $`\mathrm{๐ณ๐๐ซ๐จ}`$, i.e. when the transverse part is calculated one should keep the centre at $`k_\mu =0`$ or one must add suitable terms to compansate. Up until now, the general covariant gauge case has been discussed. However, often it is easier to calculate the fermion wave-function renormalization in a specific gauge. Take for example Feynman gauge, $`\xi =1`$, in this case Eq. (18) greatly simplifies and we are left with only the first term of the integral, Eq. (23). From this equation we can see that in order to be consistent with the arbitrary covariant gauge calculation and dimensional regularization calculation $`c`$ can conveniently be chosen to be $`1/2`$. In the arbitrary covariant gauge calculation, the non-$`\xi `$ part of the integral in Eq. (18) is given by Eqs. (23-25-27). In this case, the terms violating translational invariance cancel out in Eq. (23) and Eq. (27) if and only if $`c=0`$. In the Feynman gauge no cancellation occurs so that the term violating the translational invariance in Eq. (23) must vanish identically. This can only be done if the hypersphere is centred at $`k_\mu =p_\mu /2`$, namely $`c=1/2`$. We emphasize that we can equally well choose either centre for any gauge. The two choices described here give the same translationally invariant result since the two choices can be seen to differ by just the right term (that vanishes in a translationally invariant scheme).
To summarize: In an arbitrary covariant gauge $`I_T`$ vanishes as it should, if the centre of the hypersphere is located at $`k_\mu =0`$. In that case, centering the cut-off integration for $`I_L`$ at $`k_\mu =q_\mu `$ gives the result consistent with the translationally invariant dimensional regularization. This is one convenient procedure. In the specific case of the Feynman gauge this same translationally invariant result can also, for example, be conveniently obtained by centering the hypersphere of the entire integrand at $`k_\mu =p_\mu /2`$. These two ways of proceeding, of course, lead to the same expression for the total integral in Feynman gauge as they should.
(3) Renormalization :
(a) : Applying multiplicative renormalization (MR) requires the following relations between renormalized, $`F_R`$, and unrenormalized, $`F_0`$, fermion wave-function renormalization :
$`F_R=Z_2^1F_0,`$ (32)
where $`Z_2`$ denotes the fermion renormalization constant. If we apply MR to the perturbation theory order by order, then the unrenormalized fermion wave-function renormalization which is calculated in an uncorrected cut-off scheme up to order $`\alpha `$ is :
$`F_0(p^2,\mathrm{\Lambda }^2)`$ $`=`$ $`1+{\displaystyle \frac{\alpha \xi }{4\pi }}\left(\mathrm{ln}{\displaystyle \frac{p^2}{\mathrm{\Lambda }^2}}{\displaystyle \frac{1}{2}}\right)+๐ช(\alpha ^\mathcal{2}),`$ (33)
and the fermion renormalization constant is :
$`Z_2(\mu ^2,\mathrm{\Lambda }^2)=1+{\displaystyle \frac{\alpha \xi }{4\pi }}\left(\mathrm{ln}{\displaystyle \frac{\mu ^2}{\mathrm{\Lambda }^2}}{\displaystyle \frac{1}{2}}\right)+๐ช(\alpha ^2)`$ (34)
One can renormalize $`F_0`$ by choosing $`F_R(p^2,\mu ^2)=1`$ at $`p^2=\mu ^2`$ and find the renormalized wave-function renormalization from Eq. (32) as :
$`F_R(p^2,\mu ^2)=1+{\displaystyle \frac{\alpha \xi }{4\pi }}\mathrm{ln}{\displaystyle \frac{p^2}{\mu ^2}}+๐ช(\alpha ^2),`$ (35)
which can be summed to all orders as :
$`F_R(p^2)=\left({\displaystyle \frac{p^2}{\mu ^2}}\right)^{\frac{\alpha \xi }{4\pi }}.`$ (36)
(b) : Applying subtractive renormalization to the dimensional regularization calculation, Eq. (20), we get :
$`F_R(p^2)`$ $`=`$ $`1+\mathrm{\Sigma }_R(p^2;\mu ^2)`$ (37)
$`=`$ $`1+\left(\mathrm{\Sigma }_0(p^2;\mu ^2,\mathrm{\Lambda }^2)\mathrm{\Sigma }_0(\mu ^2;\mu ^2,\mathrm{\Lambda }^2)\right)`$ (38)
$`=`$ $`1+{\displaystyle \frac{\alpha \xi }{4\pi }}\mathrm{ln}{\displaystyle \frac{p^2}{\mu ^2}}+๐ช(\alpha ^2)`$ (39)
$`=`$ $`\left({\displaystyle \frac{p^2}{\mu ^2}}\right)^{\frac{\alpha \xi }{4\pi }},`$ (40)
where $`\mathrm{\Sigma }_R(p^2;\mu ^2)`$ is renormalized fermion self energy. As we see, even if we start with the incorrect unrenormalized fermion wave-function renormalization, we get the same renormalized results for both cut-off and dimensional regularization schemes in perturbation theory. This is because the WTI is taken care of automatically in perturbation theory; however this is not the case in nonperturbative theory. So, starting with the wrong quantity, gives the wrong answer in nonperturbative theory. If one does not impose translational invariance on the regulator as a necessary condition then the result will be different in cut-off and dimensional regularization schemes.
Since the violation of translational invariance in a naive UV cut-off scheme appears to be the only source of gauge-covariance violation, the restoration of translational invariance should also remove any explicit source of gauge-covariance violation. Translational invariance is certainly a necessary condition, but one should ask whether it is a sufficient condition. At present we know of no rigorous mathematical argument that proves such a sufficiency. All that can be said is that it is difficult to conceive of integrands where this would not be the case. Indeed, field theories which yield different behaviors from dimensional regularization and translationally invariant UV cut-off approaches would need to be specified by both a Lagrangian density and by a particular choice of regularization scheme.
We now move on to nonperturbative QED and investigate how much information from perturbation theory we can make use of. Let us start with Ball-Chiu (BC) plus Curtis-Pennington (CP) vertex in massless QED<sub>4</sub>.
### B Prescription for Consistency with Dimensional Regularization
Thus far we have seen how translational invariance is violated in the UV cut-off regularization and have described what must be done to obtain the same result as dimensional regularization for the fermion self-energy in quenched QED<sub>4</sub>. In this section we shall formulate a more general prescription.
In any translationally invariant regularization scheme the following expression is true:
translationally (41)
invariant scheme (42)
$`I(p^2)={\displaystyle d^4kI(k,p,\theta )}`$ $`=`$ $`{\displaystyle d^4kI(k^{},p,\theta )},`$ (43)
where the integrand $`I(k,p,\theta )`$ is related to the shifted integrand $`I(k^{},p,\theta )`$ through $`k_\mu ^{}=k_\mu +cp_\mu +b_\mu `$. Unfortunately, in cut-off regularization the above equality is not true for renormalized integrands in the limit $`\mathrm{\Lambda }\mathrm{}`$ unless the integrand, $`I(k,p,\theta )`$, is at worst logarithmically divergent, i.e., in general
$`\underset{\mathrm{\Lambda }\mathrm{}}{lim}\left({\displaystyle ^\mathrm{\Lambda }}d^4kI(k,p,\theta ){\displaystyle ^\mathrm{\Lambda }}d^4kI(k^{},p,\theta )\right)0.`$ (45)
Since in the limit $`\mathrm{\Lambda }\mathrm{}`$ the renormalized Green functions (and their various component parts) are necessarily finite, the contribution from the integrands must be vanishingly small at infinity. This ensures that the result of the integral is independent of whether the shape of the integral region was hyperspherical, hypercubic or any other shape one might construct. In other words, once translational invariance in the UV cut-off approach has been ensured and the limit $`\mathrm{\Lambda }\mathrm{}`$ has been taken, the resulting renormalized nonperturbative quantities are entirely independent of the details of how the limit was taken. To ensure the above equality we need to develop an appropriate, unique, translationally invariant cut-off regularization scheme. This can be summarized by the following prescription :
1. Start with the integral $`^\mathrm{\Lambda }d^4kI(k,p,\theta )`$.
2. Choose any centre for the four-dimensional hypersphere with radius $`\mathrm{\Lambda }`$.
3. Add $`d^4k\mathrm{\Delta }I_i`$ terms which would vanish in any translational invariant regularization scheme to eliminate the defect of all linearly divergent terms in perturbation theory, where
$`{\displaystyle d^4k\mathrm{\Delta }I_i}={\displaystyle d^4k\left[I_i(k^{},p,\theta )I_i(k,p,\theta )\right]}=\mathrm{\hspace{0.17em}0}.`$ (46)
and where $`I_i(k,p,\theta )`$ is some integrand. The above difference is zero in dimensional regularization but will introduce an artificial term which contributes to the next to leading order terms in a cut-off scheme. The number of $`\mathrm{\Delta }I_i`$ terms needed is related to the number of linearly divergent terms in the integrand ,
$`{\displaystyle ^\mathrm{\Lambda }}d^4kI^{}(k,p,\theta )={\displaystyle ^\mathrm{\Lambda }}d^4k\left[I(k,p,\theta )+{\displaystyle \underset{i}{}}d_i\mathrm{\Delta }I_i(k,k^{},p,\theta )\right].`$ (47)
where $`d_i`$ are constants.
4. The $`d_i`$โs are to be fixed by equating Eq. (47) with the translational invariant results (dimensional regularization results) from perturbation theory.
### C Application of the Prescription
Let us consider the fermion wave-function renormalization, $`F(p^2)`$, in perturbation theory as an example. As we shall disscuss later one can see the residue of the violated symmetries in the cut-off scheme in the nonperturbative case even after renormalization. Of course this is not the case in perturbation theory. Therefore, since in this section we deal with the perturbative expansion of the fermion wave-function renormalization, we shall use regularized quantities in order to pin down the terms which cause problems in the nonperturbative case.
The integrand in Eq. (LABEL:eq:perfermion2) can be divided into three parts :
$`I(k,p,\theta )=I_0(k,p,\theta )+\xi [I{}_{}{}^{\mathrm{odd}}(k,p,\theta )+I{}_{}{}^{\mathrm{trans}.}(k,p,\theta )],`$ (48)
where
$`\begin{array}{ccc}I_0(k,p,\theta )& \hfill & {\displaystyle \frac{\left(3kp(k^2+p^2)+4(kp)^2+2k^2p^2\right)}{k^2q^4}},\hfill \\ I{}_{}{}^{\mathrm{odd}}(k,p,\theta )& \hfill & {\displaystyle \frac{k^2pq}{k^2q^4}},\hfill \\ I{}_{}{}^{\mathrm{trans}.}(k,p,\theta )& \hfill & {\displaystyle \frac{p^2kq}{k^2q^4}},\hfill \end{array}\}`$ (52)
where $`I^{\mathrm{odd}}`$ is odd in $`q`$ and $`I^{\mathrm{trans}}`$ is a translationally invariant integrand (since it is only logarithmically divergent).
Let us apply the prescription to the $`\xi `$ part in Eq. (48) first :
1. Start with,
$`I_\xi (p){\displaystyle ^\mathrm{\Lambda }}d^4k[I{}_{}{}^{\mathrm{odd}}(k,p,\theta )+I{}_{}{}^{\mathrm{trans}.}(k,p,\theta )].`$ (53)
2. Choose $`k_\mu `$ as the centre of the hypersphere $`\mathrm{\Lambda }`$.
3. Add $`d^4k\mathrm{\Delta }I_i`$ terms :
$`I_\xi ^{}(p)`$ $`=`$ $`{\displaystyle ^\mathrm{\Lambda }}d^4k[I^{}{}_{}{}^{\mathrm{odd}}(k,p,\theta )+I^{}{}_{}{}^{\mathrm{trans}.}(k,p,\theta )],`$ (54)
$`=`$ $`{\displaystyle ^\mathrm{\Lambda }}d^4k\{I{}_{}{}^{\mathrm{odd}}(k,p,\theta )+d_1[I{}_{}{}^{\mathrm{odd}}(k^{},p,\theta )I{}_{}{}^{\mathrm{odd}}(k,p,\theta )]`$ (56)
$`+I{}_{}{}^{\mathrm{trans}}(k,p,\theta )+d_2[I{}_{}{}^{\mathrm{trans}}(k^{},p,\theta )I{}_{}{}^{\mathrm{trans}}(k,p,\theta )]\},`$
where $`k^{}`$ is the shifted $`k`$, in general $`k_\mu ^{}=k_\mu +cp_\mu +b_\mu `$.
4. Fix $`d_1`$ by using the following :
* In any translational invariant scheme, any odd part of the integral should be zero :
$`{\displaystyle }d^4kI^{}{}_{}{}^{\mathrm{odd}}(k,p)={\displaystyle }d^4kI^{}{}_{}{}^{\mathrm{odd}}(k^{},p)=0`$ (57)
Therefore the odd part of Eq. (56) can be written as :
$`I^{{}_{}{}^{}\mathrm{odd}}(p)=\mathrm{\hspace{0.17em}0}={\displaystyle ^\mathrm{\Lambda }}d^4k\{(1d_1)I{}_{}{}^{\mathrm{odd}}(k,p,\theta )+d_1I{}_{}{}^{\mathrm{odd}}(k^{},p,\theta )\}.`$ (58)
For instance, if we locate the centre of the sphere at $`k_\mu =0`$ and then we shift the loop momentum to $`k_\mu k_\mu +p_\mu `$ then :
$`{\displaystyle ^\mathrm{\Lambda }}d^4kI{}_{}{}^{\mathrm{odd}}(k,p,\theta )`$ $`=`$ $`{\displaystyle \frac{1}{2}},`$ (59)
$`{\displaystyle ^\mathrm{\Lambda }}d^4kI{}_{}{}^{\mathrm{odd}}(k+p,p,\theta )`$ $`=`$ $`{\displaystyle ^\mathrm{\Lambda }}d^4k{\displaystyle \frac{(pk)}{k^4}}=0.`$ (60)
Hence, in order to satisfy Eq. (58),
$`d_1=1`$
* The $`I^{\mathrm{trans}}`$ part is independent of shifts in $`k`$, i.e. it is translationally invariant since it is only logarithmically divergent :
$`I^{{}_{}{}^{}\mathrm{trans}}(p)={\displaystyle ^\mathrm{\Lambda }}I{}_{}{}^{trans}(k,p,\theta )={\displaystyle ^\mathrm{\Lambda }}I{}_{}{}^{trans}(k^{},p,\theta )\text{independent of }d_2`$ (61)
Now let us consider the $`I_0`$ part of Eq. (48) :
* We know from dimensional regularization that in perturbation theory $`d^4kI_0(k,p,\theta )=0`$ .
$`I_0^{}(p)=\mathrm{\hspace{0.17em}0}={\displaystyle ^\mathrm{\Lambda }}d^4k\left\{(1d_3)I_0(k,p,\theta )+d_3I_0(k^{},p,\theta )\right\}.`$ (62)
Within cut-off regularization :
$`{\displaystyle ^\mathrm{\Lambda }}d^4kI_0(k,p,\theta )`$ $`=`$ $`0,`$ (63)
$`{\displaystyle ^\mathrm{\Lambda }}d^4kI_0(k^{},p,\theta )`$ $`=`$ $`{\displaystyle \frac{3}{2}},`$ (64)
where again here the first integral is centred at $`k_\mu =0`$ and the second integral at $`k_\mu =p_\mu `$. So, for Eq. (62) to be true
$`d_3`$ $`=`$ $`0.`$
The above prescription for the cut-off regularization scheme should ensure the same result as the dimensional regularization scheme.
### D Massless, Quenched QED<sub>4</sub> with CP-Vertex
Inserting the full fermion and bare photon propagators and full fermion-photon vertex (BC+CP) into Eq. (1), then multiplying the result by $`\overline{)}p`$, and taking its trace we get :
$`{\displaystyle \frac{1}{F(p^2)}}`$ $`=`$ $`1+{\displaystyle \frac{i\alpha }{8\pi ^3p^2}}{\displaystyle _M}{\displaystyle \frac{d^4k}{k^2q^4}}\times `$
$`\{(1+{\displaystyle \frac{F(k^2)}{F(p^2)}})[(3kp(k^2+p^2)+4(kp)^2+2k^2p^2)+\xi (k^2pqp^2kq)]`$
$`+(1{\displaystyle \frac{F(k^2)}{F(p^2)}})[{\displaystyle \frac{(k^2+p^2)}{(k^2p^2)}}(3kp(k^2+p^2)+4(kp)^2+2k^2p^2)+\xi (k^2pq+p^2kq)]\}.`$
After moving from Minkowski space to Euclidean space by performing a Wick rotation, we can carry out the above integrals. By looking at Eq. (LABEL:eq:bccp), we notice that the fermion wave-function, $`F(p^2)`$, is the only nonperturbative quantity and does not depend on the angle between $`k`$ and $`p`$. At the level of calculating angular integrals, everything is the same as for the perturbation theory case, hence we know how to deal with these integrals. As we have seen before, the first term of the integral in Eq. (LABEL:eq:bccp), earlier called $`I_T`$, is zero,( Eq. (29)). So then we have:
$`{\displaystyle \frac{1}{F(p^2,\mathrm{\Lambda }^2)}}=1{\displaystyle \frac{\alpha \xi }{4\pi ^3}}{\displaystyle _E}{\displaystyle \frac{d^4k}{k^2q^4}}\left(k^2pqp^2kq{\displaystyle \frac{F(k^2,\mathrm{\Lambda }^2)}{F(p^2,\mathrm{\Lambda }^2)}}\right).`$ (66)
The above equation can be expressed in terms of renormalized quantities as :
$`{\displaystyle \frac{1}{Z_2(\mu ^2,\mathrm{\Lambda }^2)F_R(p^2,\mu ^2)}}=1{\displaystyle \frac{\alpha \xi }{4\pi ^3p^2}}{\displaystyle _E}{\displaystyle \frac{d^4k}{k^2q^4}}\left(k^2pqp^2kq{\displaystyle \frac{F_R(k^2,\mu ^2)}{F_R(p^2,\mu ^2)}}\right).`$ (67)
Multiplying this equation by $`F_R(p^2,\mu ^2)`$ to leave the fermion renormalization constant, $`Z_2`$, alone on the left hand side and performing some of the integrals, we find :
$`{\displaystyle \frac{1}{Z_2(\mu ^2,\mathrm{\Lambda }^2)}}=F_R(p^2,\mu ^2)+{\displaystyle \frac{\alpha \xi }{4\pi }}{\displaystyle _0^{\mathrm{\Lambda }^2}}๐k^2\left({\displaystyle \frac{k^2}{p^4}}F_R(p^2,\mu ^2)\theta (p^2k^2)+{\displaystyle \frac{1}{k^2}}F_R(k^2,\mu ^2)\theta (k^2p^2)\right).`$
If we write the same expression for $`p^2=\mu ^2`$ and subtract it from Eq. (LABEL:eq:peqmu), we get :
$`0=F_R(p^2,\mu ^2)F_R(\mu ^2,\mu ^2)+{\displaystyle \frac{\alpha \xi }{4\pi }}\left\{{\displaystyle \frac{1}{2}}F_R(p^2,\mu ^2){\displaystyle \frac{1}{2}}F_R(\mu ^2,\mu ^2)+{\displaystyle _{p^2}^{\mu ^2}}{\displaystyle \frac{dk^2}{k^2}}F_R(k^2,\mu ^2)\right\}.`$
Considering that the fermion wave-function renormalization $`F(p^2)`$ must obey the power behaviour $`F_R(p^2,\mu ^2)=(p^2/\mu ^2)^\nu `$ in the nonperturbative massless case, then after carrying out the radial integral Eq. (LABEL:eq:differ) can be written as:
$`F_R(p^2,\mu ^2)F_R(\mu ^2,\mu ^2)={\displaystyle \frac{\alpha \xi }{4\pi }}\left\{\left({\displaystyle \frac{1}{2}}{\displaystyle \frac{1}{\nu }}\right)\left[F_R(p^2,\mu ^2)F_R(\mu ^2,\mu ^2)\right]\right\}.`$ (70)
Comparing both sides of Eq. (70) we find that,
$`\nu ={\displaystyle \frac{2\alpha \xi }{\alpha \xi +8\pi }}`$ (71)
giving the solution
$`F_R(p^2/\mu ^2)`$ $`=`$ $`\left({\displaystyle \frac{p^2}{\mu ^2}}\right)^{2\alpha \xi /(\alpha \xi +8\pi )}.`$ (72)
Unlike perturbation theory, the third and fourth terms of Eq. (LABEL:eq:differ) do not cancel each other, so that the first term, $`1/2`$, on the right hand side of Eq. (70) survives. We have seen in perturbation theory that keeping $`d^4kpq/q^4`$ which is the source of this term did not make any difference in the renormalized quantities, because in perturbation theory it is :
$`{\displaystyle d^4k\frac{pq}{q^4}}{\displaystyle d^4k\frac{\mu q^{}}{q^4}}=\mathrm{\hspace{0.17em}0},`$ (73)
where $`q^{}=k\mu `$. BUT it does make a difference in nonperturbative studies. Assuming the transverse vertex vanishes in the Landau gauge means that $`\nu =\alpha \xi /4\pi `$, i.e. $`F_R(p^2,\mu ^2)=(p^2/\mu ^2)^{\alpha \xi /4\pi }`$, from the LKF transformation. Eq. (71) is different from $`\alpha \xi /4\pi `$ due to the fact that translational invariance is broken by cut-off regularization. Therefore, one must cancel the $`d^4kpq/q^4`$ term in Eq. (66) in order to recover the correct behaviour of the fermion wave-function renormalization. Removing this term from Eq. (66), we find :
$`{\displaystyle \frac{1}{F(p^2,\mathrm{\Lambda }^2)}}=1+{\displaystyle \frac{\alpha \xi }{4\pi ^3}}{\displaystyle _E}{\displaystyle \frac{d^4k}{k^2q^4}}\left(p^2kq{\displaystyle \frac{F(k^2,\mathrm{\Lambda }^2)}{F(p^2,\mathrm{\Lambda }^2)}}\right),`$ (75)
so that
$`\nu `$ $`=`$ $`{\displaystyle \frac{\alpha \xi }{4\pi }}`$ (76)
giving the solution of this equation as
$`F_R(p^2/\mu ^2)`$ $`=`$ $`\left({\displaystyle \frac{p^2}{\mu ^2}}\right)^{\alpha \xi /4\pi }.`$ (77)
This is exactly the same as the result obtained from a nonperturbative dimensional regularization scheme . As a result of the above discussions we see that the modified cut-off prescription can be succesfully applied to massless quenched QED. Therefore, after applying the prescription for this case one can see the agreement between dimensional regularization and the modified cut-off result numerically in Figs.8 and 9.
### E Massive, Quenched QED<sub>4</sub> with CP-Vertex
The fermion wave-function renormalization for the massive QED<sub>4</sub> case using BC and CP vertices can be written as :
$`{\displaystyle \frac{1}{F(p^2)}}`$ $`=`$ $`1+{\displaystyle \frac{i\alpha }{8\pi ^3p^2}}{\displaystyle _M}{\displaystyle \frac{d^4k}{\left(k^2M^2(k^2)\right)q^4}}\times `$ (82)
$`\{(1+{\displaystyle \frac{F(k^2)}{F(p^2)}})[(3kp(k^2+p^2)+4(kp)^2+2k^2p^2)+\xi (k^2pqp^2kq)]`$
$`+\left(1{\displaystyle \frac{F(k^2)}{F(p^2)}}\right)\left[{\displaystyle \frac{1}{\left(k^2p^2\right)}}{\displaystyle \frac{\left(3kp\left(k^2+p^2\right)+4(kp)^2+2k^2p^2\right)}{d}}+\xi \left(k^2pq+p^2kq\right)\right]`$
$`\left(1{\displaystyle \frac{F(k^2)}{F(p^2)}}\right)\left[{\displaystyle \frac{1}{\left(k^2p^2\right)}}{\displaystyle \frac{2\mathrm{\Delta }^2\left(M^2(k^2)+M^2(p^2)\right)^2}{d}}\right]`$
$`+(M^2(k^2)M(k^2)M(p^2){\displaystyle \frac{F(k^2)}{F(p^2)}}){\displaystyle \frac{1}{(k^2p^2)}}(4\mathrm{\Delta }^2+\xi (k^2p^2)^2pq)\},`$
where
$`d`$ $`=`$ $`{\displaystyle \frac{(k^2p^2)^2+[M^2(k^2)+M^2(p^2)]^2}{k^2+p^2}},`$
$`\mathrm{\Delta }^2`$ $`=`$ $`(kp)^2k^2p^2.`$
In this case, the second and third lines of Eq. (82) are exactly the same as in the massless case except for the mass term in the denominator. Hence, for calculating these lines, the only difference between the massless and massive cases will come from radial integrals and the presence of a mass term in the denominator only makes the calculation convergent more quickly. So we do not encounter any worse than a logarithmic divergence. The third and fourth lines of Eq. (82) will introduce new terms which depend on the mass function but the integrals do not have any worse divergence than the logarithmic divergence because for large momenta the mass function behaves like $`M(k^2)\alpha k^\gamma ,0<\gamma <2`$, , ,. Therefore, there is no danger of violating translational invariance.
Of course, in the massive case with the Curtis-Pennington or the real transverse vertex, the fermion wave-function renormalization will not give $`1`$ in Landau gauge. In other words, the transversality condition , is not applicable for these vertices. As a result of that, for such vertices, we can not use the condition $`F(p^2)=1`$ in Landau gauge in the prescription. Consequently, the third item, Eq. (62), in the prescription must be changed to :
$`I_0^{}(p)=\overline{f(p)}`$ $`=`$ $`{\displaystyle ^\mathrm{\Lambda }}d^4kI_0^{}(k,p,\theta )={\displaystyle ^\mathrm{\Lambda }}d^4kI_0^{}(k^{},p,\theta )`$ (83)
$`=`$ $`{\displaystyle ^\mathrm{\Lambda }}d^4k\left\{(1d_3)I_0(k,p,\theta )+d_3I_0(k^{},p,\theta )\right\}`$ (84)
$`=`$ $`(1d_3)f(p)+d_3f^{}(p).`$ (85)
Knowing that $`k_\mu =0`$ is the right centre for $`I_0`$ term in massless QED and required to satisfy the massless limit when $`M(p^2)0`$, we should also choose the centre at $`k_\mu =0`$ for the massive case. This means $`\overline{f(p)}=f(p)`$, then Eq. (85) becomes :
$`0=d_3\left[f^{}(p)f(p)\right],`$ (86)
Due to the fact that \[$`f^{}(p)f(p)`$\], $`d_3`$ should be :
$`d_3=0.`$ (87)
This is just the prescription used in the modified UV cut-off scheme which we have already seen gives such excellent numerical agreement with the dimensional regularization studies.
## IV Conclusions and Outlook
Studying Quantum Electrodynamics necessarily introduces divergences. We have seen explicitly that the violation of translational invariance in nonperturbative studies using an ultraviolet cut-off leaves an error in the renormalized result. Hence, if one wants to use cut-off regularization and find the translationally invariant answer for the calculated renormalized quantity then a modification is needed. Since the nonperturbative quantity must give the perturbative result in the weak coupling limit, it is simplest to attempt to identify these modifications within perturbation theory. Once this is done we can attempt to generalise it to the nonperturbative case. Fortunately, in the case of the fermion self-energy calculation one can establish a nonperturbative framework on top of the perturbative one.
In this work the violation of translational invariance for the electron self-energy is analyzed in detail and a prescription is presented in order to calculate the quantity without breaking translational invariance in cut-off regularization. More precisely we mean that violations of translational invariance are no worse than logarithmic and so for the subtracted (i.e., renormalized) integral in the limit $`\mathrm{\Lambda }\mathrm{}`$, translational invariance is restored. In this regard, the electron self-energy is used a test case and as a result we have seen that a suitably modified cut-off scheme and the dimensional regularization scheme should be in agreement. Careful numerical studies (see Tables IIII and Figs. 56 and 89) have demonstrated this agreement to high precision. In closing we note that while we can always add terms with arbitrary coefficients which would vanish in a translationally invariant regularization scheme, our approach will only be useful when there are sufficient known constraints to determine these coefficients uniquely. We are currently attemping to extend this approach to include the photon self energy, so that we can study unquenched QED<sub>4</sub> using a translationally invariant ultraviolet cut-off regularization scheme.
###### Acknowledgements.
We thank A. Schreiber for numerious helpful discussions. AGW also acknowledges support from the Department of Energy Contract No. DE-FG05-86ER40273 and by the Florida State University Supercomputations Research Institute which is partially funded by the Department of Energy through contract No. DE-FC05-85ER25000 |
warning/0001/hep-ph0001323.html | ar5iv | text | # 1 The ๐ตโข๐ term is generated in the two-loop diagram.
## Acknowledgements
A part of this work was done at the Department of Physics Engineering, Mie University, Tsu, Japan. The authors are grateful to Professor Y. Abe and Assistant Professor M. Matsunaga for their hospitality. One of the authors (T. M.) is supported in part by a Grant-in-Aid for Scientific Research from the Ministry of Education, Science, Sports and Culture, Japan (No . 10640256). |
warning/0001/astro-ph0001326.html | ar5iv | text | # Identifying the Reionization Redshift from the Cosmic Star Formation Rate
## 1 Introduction
Current observations reveal the existence of galaxies out to redshifts as high as $`z6.7`$ (Chen et al. 1999; Weymann et al. 1998; Dey et al. 1998; Spinrad et al. 1998; Hu, Cowie, & McMahon 1998) and bright quasars out to $`z5`$ (Fan et al. 1999). The detection of transmitted flux at rest-frame wavelengths shorter than Ly$`\alpha `$ in the spectrum of some of these sources implies that most of the intergalactic medium (IGM) is ionized at $`z5`$ (Songaila et al. 1999), and that a substantial population of ionizing sources should exist at higher redshifts (Madau, Haardt, & Rees 1999). This inference is consistent with theoretical expectations in Cold Dark Matter (CDM) models of galaxy formation (Shapiro, Giroux, & Babul 1994; Gnedin & Ostriker 1997; Haiman & Loeb 1998a,b; Valageas & Silk 1999; Miralda-Escudรฉ, Haehnelt, & Rees 1999; Chiu & Ostriker 1999; Gnedin 1999).
One of the major goals of the study of galaxy formation is to achieve an observational determination and a theoretical understanding of the cosmic star formation history. By now, this history has been sketched out only to a redshift $`z4`$ (e.g., see the compilation of Blain et al. 1999). The Next Generation Space Telescope (NGST )<sup>1</sup><sup>1</sup>1See http://www.ngst.stsci.edu/, planned for launch in 2008, is expected to reach an imaging sensitivity better than 1 nJy in the infrared, which will allow it to detect galaxies and hence determine the star formation history at $`z10`$.
The redshift of reionization, $`z_{\mathrm{reion}}`$, defines the epoch when the H 2 regions surrounding individual sources in the IGM overlapped. At this time, the cosmic ionizing background increased sharply, and the IGM was heated by the ionizing radiation to a temperature $`10^4`$ K (see, e.g., Gnedin 1999). In this Letter, we explore the distinct signature that this heating process is expected to have left on the global star formation rate in the universe. Due to the substantial increase in the IGM temperature, the intergalactic Jeans mass increased dramatically, changing the minimum mass of forming galaxies (Gnedin & Ostriker 1997; Miralda-Escudรฉ & Rees 1998). Reionization is therefore predicted to cause a drop in the cosmic star formation rate (SFR) at $`z_{\mathrm{reion}}`$. This drop is accompanied by a dramatic fall in the number counts of faint galaxies. Detection in future galaxy surveys of this fall in the faint luminosity function could be used to identify $`z_{\mathrm{reion}}`$ observationally. This method for inferring $`z_{\mathrm{reion}}`$ is more straightforward than alternative proposals which rely on high resolution spectroscopy of individual sources (Miralda-Escudรฉ 1998; Haiman & Loeb 1999; Loeb & Rybicki 1999) or on the detection of faint spectral features in the background radiation (Gnedin & Ostriker 1997; Shaver et al. 1999).
In ยง2 we describe the galaxy formation model used to calculate the cosmic SFR and the luminosity function at high redshift. The results from this model, including the reionization signature, are described in ยง3. We dedicate ยง4 to a detailed comparison with previous semi-analytic models and numerical simulations. Finally, we summarize our main conclusions in ยง5.
## 2 Galaxy Formation Model
We model galaxy formation within a hierarchical CDM model of structure formation. We obtain the abundance of dark matter halos using the Press-Schechter (1974) model. Relevant expressions for various CDM cosmologies are given, e.g., in Navarro, Frenk, & White (1997). The abundance of halos evolves with redshift as each halo gains mass through mergers with other halos. If $`dp[M_1,t_1M,t]`$ is the probability that a halo of mass $`M_1`$ at time $`t_1`$ will have merged to form a halo of mass between $`M`$ and $`M+dM`$ at time $`t>t_1`$, then in the limit where $`t_1`$ tends to $`t`$ we obtain an instantaneous merger rate $`d^2p[M_1M,t]/(dMdt)`$. This quantity was evaluated by Lacey & Cole \[1993, their Eq. (2.18)\].
Once a dark matter halo has collapsed and virialized, the two requirements for forming new stars are gas infall and cooling. Gas of primordial composition can cool through atomic transitions in halos with virial temperatures $`T_{\mathrm{vir}}10^{3.8}`$K (Haiman, Abel, & Rees 1999), and through molecular hydrogen ($`\mathrm{H}_2`$) transitions in smaller halos. However, $`\mathrm{H}_2`$ molecules are fragile, and could have been easily photo-dissociated throughout the Universe by trace amounts of starlight (Stecher & Williams 1967; Haiman, Rees, & Loeb 1996) that were well below the level required for complete reionization of the IGM. We therefore assume that, in the redshift interval covered by our models, efficient cooling can only occur with atomic transitions, in halos above a circular velocity of $`V_c13.3\mathrm{km}\mathrm{s}^1`$. Haiman et al. (1999) showed that if quasars contributed significantly to the UV background before reionization, their soft X-rays produced free electrons in small halos and may have catalyzed the formation of molecular hydrogen. In this case, smaller halos than we assume could have formed stars prior to reionization, down to $`T_{\mathrm{vir}}10^{2.4}`$K. However, the same X-rays may also have begun to heat the IGM before reionization, hindering gas infall into the smallest halos. Even if this heating mechanism was ineffective, most of the additional galaxies would be too faint to detect with NGST. Regardless of the effectiveness of $`H_2`$ cooling, the formation of new galaxies after reionization would still be greatly suppressed since it is limited by gas infall.
Gas infall depends sensitively on the Jeans mass. When a halo more massive than the Jeans mass begins to form, the gravity of its dark matter overcomes the gas pressure. Even in halos below the Jeans mass, although the gas is initially held up by pressure, once the dark matter collapses its increased gravity pulls in some gas (Haiman, Thoul, & Loeb 1997). Before reionization, the IGM is cold and neutral, and the Jeans mass plays a secondary role compared to cooling. After reionization, the Jeans mass is increased by several orders of magnitude due to the photoionization heating of the IGM, and hence begins to play a dominant role in limiting the formation of stars. Gas infall in a reionized and heated Universe has been investigated in a number of numerical simulations. Thoul & Weinberg (1996) found a reduction of $`50\%`$ in the collapsed gas mass due to heating, for a halo of circular velocity $`V_c50\mathrm{km}\mathrm{s}^1`$ at $`z=2`$, and a complete suppression of infall below $`V_c30\mathrm{km}\mathrm{s}^1`$. Kitayama & Ikeuchi (2000) also performed spherically-symmetric simulations but they included self-shielding of the gas, and found that it lowers the circular velocity thresholds by $`5\mathrm{km}\mathrm{s}^1`$. Three dimensional numerical simulations (Quinn, Katz, & Efstathiou 1996; Weinberg, Hernquist, & Katz 1997; Navarro & Steinmetz 1997) found a significant suppression of gas infall in even larger halos ($`V_c75\mathrm{km}\mathrm{s}^1`$), but this was mostly due to a suppression of late infall at $`z2`$. Thus, we adopt a prescription for the suppression of gas infall based on the spherically symmetric simulations. Each of these simulations follows an isolated halo, rather than a set of merging progenitors, and the halo is followed for longer than it would typically survive before merging into a larger halo. Therefore, we adopt slightly higher $`V_c`$ thresholds than indicated by the spherically symmetric simulations.
When a volume of the IGM is ionized by stars, the gas is heated to a temperature $`T_{\mathrm{IGM}}10^4`$ K. If quasars dominate the UV background at reionization, their harder photon spectrum leads to $`T_{\mathrm{IGM}}2\times 10^4`$ K. Including the effects of dark matter, a given temperature results in a linear Jeans mass (e.g., ยง6 in Peebles 1993) corresponding to a halo circular velocity of
$$V_J=93\left(\frac{T_{\mathrm{IGM}}}{2\times 10^4\mathrm{K}}\right)^{1/2}\left[\frac{1}{\mathrm{\Omega }(z)}\frac{\mathrm{\Delta }_c}{18\pi ^2}\right]^{1/6}\mathrm{km}\mathrm{s}^1,$$
(1)
where $`\mathrm{\Delta }_c`$ is the mean collapse overdensity (Bryan & Norman 1998) and $`\mathrm{\Omega }(z)`$ is the total density in units of the critical density, evaluated at redshift $`z`$. In halos with $`V_c>V_J`$, the gas fraction equals the universal mean of $`\mathrm{\Omega }_b/\mathrm{\Omega }_0`$. We assume that the gas fraction is suppressed by $`50\%`$ at a circular velocity $`V_c=V_J\times 55/93`$ and a complete suppression occurs below $`V_c=V_J\times 35/93`$. We set $`T_{\mathrm{IGM}}=2\times 10^4`$ K, although a lower temperature would result in a slightly weaker suppression of gas infall.
Recent semi-analytic models (Miralda-Escudรฉ et al. 1999) and numerical simulations (Gnedin 1999) show that the process of reionization is gradual. When stars or quasars form in galaxies, each source must first ionize gas in its host object (Wood & Loeb 2000). Individual ionization bubbles then grow for about a Hubble time at the relevant redshifts, until they begin to overlap. Since overlapping H 2 regions are filled with radiation from several sources, the ionizing intensity increases rapidly at each overlap and this leads to a quick ionization of almost the entire volume of the universe. Only the highest density gas remains neutral and is gradually ionized subsequently.
The reionization feedback on galaxy formation depends on the fraction of the IGM which is ionized at each redshift. Thus, we focus on the overlap stage of reionization. We refer to the reionization redshift $`z_{\mathrm{reion}}`$ as the redshift at which $`50\%`$ of the volume of the universe is ionized, and assume that a significant volume is first ionized at the starting redshift $`z_{\mathrm{start}}`$ and that most of the volume has been ionized by the ending redshift $`z_{\mathrm{end}}`$. We take the Gnedin (1999) simulation as a guide for the redshift interval of reionization, adopting $`z_{\mathrm{start}}=11`$ and $`z_{\mathrm{end}}=6.5`$ for the case of $`z_{\mathrm{reion}}=8.2`$. When we vary $`z_{\mathrm{reion}}`$ we scale the values of $`z_{\mathrm{start}}`$ and $`z_{\mathrm{end}}`$ in order to keep the reionization era equal to a constant fraction of the age of the universe at $`z_{\mathrm{reion}}`$.
At each redshift, we thus obtain the fraction of the Universe which is ionized, and, in both the neutral and ionized IGM, we obtain the gas mass which falls and cools into halos as a function of the halo mass $`M`$. If the cold gas mass is denoted by $`f_{\mathrm{cold}}(M,z)M\mathrm{\Omega }_b/\mathrm{\Omega }_0`$, then $`f_{\mathrm{cold}}(M,z)=0`$ in halos below a minimum mass $`M_{\mathrm{min}}(z)`$, set by the minimum circular velocity. Given $`f_{\mathrm{cold}}(M,z)`$, we then derive the SFR directly from halo mergers, and the luminosity of each galaxy directly from its SFR.
New star formation in a given galaxy can occur either from primordial gas or from recycled gas which has already undergone a previous burst of star formation. Consider first a region with neutral IGM, in which $`M_{\mathrm{min}}(z)`$ is set by the need for gas to cool. In this case, halos below $`M_{\mathrm{min}}(z)`$ contain roughly the universal fraction $`\mathrm{\Omega }_b/\mathrm{\Omega }_0`$ of primordial gas, which collapsed along with the dark matter but could not cool. Halos with $`M>M_{\mathrm{min}}(z)`$ contain the same gas fraction, but this gas is able to cool \[i.e., $`f_{\mathrm{cold}}(M,z)=1`$\]. Consider halos of mass $`M_1`$ and $`M_2`$ merging to yield a halo of total mass $`M`$. We assume that the merger triggers star formation only if $`M>M_{\mathrm{min}}(z)`$. If $`M_1<M_{\mathrm{min}}(z)`$, then the merger will trigger star formation in the primordial gas contained in this halo. Assuming a star formation efficiency $`\eta `$, we obtain a starburst with a total stellar mass of $`\eta M_1\mathrm{\Omega }_b/\mathrm{\Omega }_0`$. Similarly, if $`M_2<M_{\mathrm{min}}(z)`$ then the starburst has an additional contribution equal to a mass of $`\eta M_2\mathrm{\Omega }_b/\mathrm{\Omega }_0`$. If $`M_1>M_{\mathrm{min}}(z)`$, we assume that the gas in this halo has already undergone a previous burst of star formation, and thus, before the merger, the gas fraction is only $`(1\eta )\times \mathrm{\Omega }_b/\mathrm{\Omega }_0`$. We assume that the merger triggers another episode of star formation in this gas if $`M_2`$ is sufficiently massive. Numerical simulations of starbursts in interacting $`z=0`$ galaxies (e.g., Mihos & Hernquist 1994; 1996) found that a merger triggers significant star formation even if $`M_2`$ is a small fraction of $`M_1`$. Preliminary results (Somerville, private communication) from simulations of mergers at $`z3`$ find that they remain effective at triggering star formation even when the initial disks are dominated by gas. We adopt a conservative threshold mass ratio of $`r_{\mathrm{merge}}=1/2`$ and assume that if $`M_2>r_{\mathrm{merge}}M_1`$ then star formation is triggered in $`M_1`$, producing a stellar mass of $`\eta (1\eta )M_1\mathrm{\Omega }_b/\mathrm{\Omega }_0`$ (where we take the same efficiency, $`\eta `$, for star formation from primordial and from recycled gas). Similarly, if $`M_2>M_{\mathrm{min}}(z)`$, then a stellar mass of $`\eta (1\eta )M_2\mathrm{\Omega }_b/\mathrm{\Omega }_0`$ is produced as long as $`M_1>r_{\mathrm{merge}}M_2`$.
Star formation proceeds along similar lines in a region with ionized IGM, except that the value of $`M_{\mathrm{min}}(z)`$, the lowest mass of a halo which can accumulate gas from the IGM, is set by gas infall. An additional complication in this case is that some galaxies in halos below $`M_{\mathrm{min}}(z)`$ could have survived photo-ionization heating (Barkana & Loeb 1999a) and kept their cold gas reservoir after reionization. Although gas infall into these galaxies was subsequently suppressed, they may have continued to merge and thus form stars. In order to be conservative about the suppression of star formation, we assume that all gas which cooled before reionization settled into dense gas disks and did not photo-evaporate subsequently. We estimate the contribution of this gas to merger-induced star formation as follows. At a given redshift $`z`$ after reionization has started, we find the redshift $`z_{\mathrm{half}}`$ at which the filling factor of ionized regions was equal to half its value at $`z`$. Thus, roughly half of the volume which is ionized at $`z`$ was ionized after $`z_{\mathrm{half}}`$. We compute the gas fraction $`f_{\mathrm{coll}}`$ which had collapsed into halos in neutral regions by redshift $`z_{\mathrm{half}}`$. This gas lies in halos that will have merged into larger halos by redshift $`z`$. To approximate the resulting gas content of halos, we assume that at redshift $`z`$ all halos down to a mass $`M_{\mathrm{merge}}(z)`$ contain a gas fraction equal to the cosmic mean of $`\mathrm{\Omega }_b/\mathrm{\Omega }_0`$, where $`M_{\mathrm{merge}}(z)`$ is set by requiring the total gas fraction in these halos to equal $`f_{\mathrm{coll}}`$. Since $`M_{\mathrm{merge}}(z)`$ may be lower than $`M_{\mathrm{min}}(z)`$, we include by this prescription halos which had formed prior to reionization but could not have begun to form in a fully ionized universe. As discussed above, if we only considered infall in an ionized region then halos with $`M<M_{\mathrm{min}}(z)`$ would contain no gas, and the gas fraction would increase gradually up to $`f_{\mathrm{cold}}(M,z)=1`$ for all $`M`$ greater than the Jeans mass at $`z`$. We modify this prescription by increasing $`f_{\mathrm{cold}}(M,z)`$ to unity whenever $`M>M_{\mathrm{merge}}(z)`$. Thus, when two halos with masses $`M_1`$ and $`M_2=MM_1`$ merge, the resulting halo accretes from the IGM an additional primordial gas mass of $`[f_{\mathrm{cold}}(M,z)Mf_{\mathrm{cold}}(M_1,z)M_1f_{\mathrm{cold}}(M_2,z)M_2][\mathrm{\Omega }_b/\mathrm{\Omega }_0]`$ and turns a fraction $`\eta `$ of it into stars. In addition, the initial stellar content of the two halos may undergo merger-induced recycled star formation, as discussed above in the pre-reionization case.
In order to determine the typical SFR in a halo of mass $`M`$, we average the SFR over all halos $`M_1<M`$ that can merge to produce a halo of fixed total mass $`M`$. We use the Press-Schechter (1974) abundance of halos of mass $`M_1`$, and the halo merger rate $`d^2p[M_1M,t]/(dMdt)`$ introduced above. By averaging the SFR over all possible mergers we neglect the possibility of episodic star formation in individual galaxies, with merger-triggered peaks reaching star formation rates well above the average. However, for the halo mass scales relevant to galaxy formation at high redshift, the slope of the primordial power spectrum approaches $`n=3`$, which implies that the merger rate is high and galaxies double their mass in a short time compared to the age of the universe. Galaxies can be brighter than we calculate only if they undergo star formation on a timescale much shorter than the merger timescale. We expect that at each redshift, galaxies observed during an exceptional burst of star formation would, on the whole, account for only a small fraction of the total cosmic SFR.
Having obtained the average SFR for a halo of mass $`M`$, we calculate the luminosity of this halo self-consistently from this SFR. At a given redshift, we assume that the halo contains a total stellar mass of $`\eta f_{\mathrm{cold}}(M,z)M\mathrm{\Omega }_b/\mathrm{\Omega }_0`$ which has formed at a steady rate equal to the calculated average SFR. The total stellar mass in each halo may be slightly different from a fraction $`\eta `$ of the cold gas mass, but the luminosity depends mostly on the number of young, bright stars, and this number depends only on the SFR at the time that the galaxy is observed.
## 3 Results
We assume a $`\mathrm{\Lambda }`$CDM cosmology, with matter and vacuum density parameters $`\mathrm{\Omega }_0=0.3`$ and $`\mathrm{\Omega }_\mathrm{\Lambda }=0.7`$, respectively. We also assume a Hubble constant $`H_0=70\text{ km s}^1\text{Mpc}^1`$, and a primordial scale invariant ($`n=1`$) power spectrum with $`\sigma _8=0.9`$, where $`\sigma _8`$ is the root-mean-square amplitude of mass fluctuations in spheres of radius $`8h^1`$ Mpc. We adopt a cosmological baryon density of $`\mathrm{\Omega }_b=0.04`$.
In calculating the stellar emission spectrum, we assume a universal Salpeter initial mass function with a metallicity $`Z=0.001`$, and use the stellar population model results of Leitherer et al. (1999) <sup>2</sup><sup>2</sup>2Model spectra of star-forming galaxies were obtained from http://www.stsci.edu/science/starburst99/. We also include a Ly$`\alpha `$ cutoff in the spectrum due to absorption by the dense Ly$`\alpha `$ forest. We do not, however, include dust extinction, which is expected to be less significant at high redshift, when the mean metallicity was low. We describe the sensitivity of NGST by $`F_\nu ^{\mathrm{ps}}`$, the minimum spectral flux at wavelengths 0.6โ3.5$`\mu `$m required to detect a point source. The detection threshold of extended sources is higher, and we therefore incorporate a probability distribution of disk sizes at each value of halo mass and redshift (see Barkana & Loeb 1999b for details). We adopt a value of $`F_\nu ^{\mathrm{ps}}=0.25`$ nJy <sup>3</sup><sup>3</sup>3We obtained the flux limit using the NGST calculator at http://www.ngst.stsci.edu/nms/main/, assuming a very deep 300-hour integration on an 8-meter NGST and a spectral resolution of 10:1. This resolution should suffice for a $`10\%`$ redshift measurement, based on the Ly$`\alpha `$ cutoff.
Figure 1 shows our predictions for the star formation history of the Universe, with $`\eta =10\%`$. We show the SFR for $`z_{\mathrm{reion}}=7`$ (solid curves), $`z_{\mathrm{reion}}=10`$ (dashed curves), and $`z_{\mathrm{reion}}=13`$ (dotted curves). In each pair of curves, the upper one is the total SFR, and the lower one is the fraction detectable with NGST. As noted in ยง2, if star formation is episodic then halos are much brighter than average a small fraction of the time, and the detectable SFR could be somewhat higher than shown in the figure, although it would always lie below the total SFR. As discussed above, photoionization directly suppresses new gas infall after reionization, but it does not immediately affect mergers which continue to trigger star formation in gas which had cooled prior to reionization. Indeed, we find a stronger suppression of primordial star formation (which declines by a factor of 3.4 for $`z_{\mathrm{reion}}=10`$) than recycled star formation (which only declines by a factor of 1.7 for $`z_{\mathrm{reion}}=10`$). The contribution from merger-induced star formation is comparable to that from primordial gas at $`z<z_{\mathrm{reion}}`$, and it is smaller at $`z>z_{\mathrm{reion}}`$. However, the recycled gas contribution to the detectable SFR is dominant at the highest redshifts, since infalling halos more massive than $`M_{\mathrm{min}}(z)`$ dominate the star formation in the most massive halos and only these most massive halos are bright enough to be detected from the pre-reionization era. Although most stars at $`zz_{\mathrm{reion}}`$ form out of primordial, zero-metallicity gas, a majority of stars in detectable galaxies may form out of the small gas fraction that has already been enriched by the first generation of stars.
Points with error bars in Figure 1 are observational estimates of the cosmic SFR per comoving volume at various redshifts (see Blain et al. 1999 for the original references). The highest SFR estimates at $`z3`$โ4 are based on sub-millimeter observations or on extinction-corrected observations at shorter wavelengths, and all are fairly uncertain. We choose $`\eta =10\%`$ to obtain a rough agreement between the models and these observations. The SFR curves are roughly proportional to the value of $`\eta `$. Note that in reality $`\eta `$ may depend on the halo mass, since the effect of supernova feedback may be more pronounced in small galaxies.
Figure 1 shows a sharp rise in the total SFR at redshifts higher than $`z_{\mathrm{reion}}`$. Although only a fraction of the total SFR can be detected with NGST, the detectable SFR displays a definite signature of the reionization redshift. As noted in ยง2, if quasars were abundant before reionization then stars may have formed in smaller halos through $`\mathrm{H}_2`$ cooling<sup>4</sup><sup>4</sup>4Note, however, that the harder quasar spectra may lead to broader ionization fronts and to pre-heating of the neutral IGM that would suppress gas infall into the lowest-mass halos. This negative effect was not considered by Haiman et al. (1999). (Haiman et al. 1999). In this case, the SFR rise would be even larger than in the cases shown in Figure 1, but the detected SFR would be essentially unchanged since the additional galaxies would be extremely faint.
Most of the increase in SFR beyond the reionization redshift is due to star formation occurring in very small, and thus faint, galaxies. This evolution in the faint luminosity function constitutes the clearest observational signature of the suppression of star formation after reionization. Figure 2 shows the predicted luminosity function of galaxies at various redshifts. The curves show $`d^2N/(dzd\mathrm{log}F_\nu ^{\mathrm{ps}})`$, where $`N`$ is the total number of galaxies in a single field of view of NGST. Results are shown at redshift $`z=7`$ (solid curves) or $`z=13`$ (dashed curves). In each pair of curves, the upper one at 0.1 nJy assumes a permanently neutral IGM (i.e., $`z_{\mathrm{reion}}z`$), and the lower one assumes a permanently fully-ionized IGM (i.e., $`z_{\mathrm{reion}}z`$). Although our models assign a fixed, average, luminosity to all halos of a given mass and redshift, in reality such halos would have some dispersion in their merger histories and thus in their luminosities. We thus include smoothing in the plotted luminosity functions, assuming that the flux of each halo can vary by up to a factor of 2 around the mean for the set of halos with its mass and redshift. As noted in ยง2, if star formation is episodic than the distribution of fluxes about the mean could have a significant tail toward high luminosities. Note the enormous increase in the number density of faint galaxies in a pre-reionization universe. Observing this dramatic increase toward high redshift would constitute a clear detection of reionization and of its major effect on galaxy formation. The effect is greatest below a flux limit of 1 nJy, and detecting it therefore requires the capabilities of an 8-meter NGST. In the case of effective $`\mathrm{H}_2`$ cooling before reionization, the luminosity functions in a neutral IGM would follow the corresponding curves in Figure 2 down to their peaks, but they would continue to rise toward even fainter fluxes. This additional rise, though, would occur well below the detection limit of NGST .
## 4 Comparison with previous work
The reionization of the IGM has been explored in the past, based on semi-analytic models as well as numerical simulations (see ยง1). In this section we discuss the relation between our work and previous models that considered the evolution of star formation through the epoch of reionization.
The key physical ingredient which leads to the SFR drop after reionization is the temperature gap between the warm ionized regions (where the Jeans mass is high) and the cold neutral regions (where the Jeans mass is low). During reionization, the IGM consists of these two physically-independent phases which maintain their temperature gap even as the filling factor of the ionized regions increases. This temperature gap was not included in the semi-analytic models of Shapiro, Giroux, & Babul (1994) and Valageas & Silk (1999), who assumed a single-phase IGM with a uniform temperature. Gnedin & Ostriker (1997) also assumed a uniform photo-heating of the IGM in their simulation.
A recent numerical simulation by Gnedin (1999) accounted for the inhomogeneous distribution of the ionizing sources, with individual ionization fronts surrounding each source. For a stellar spectrum, most ionizing photons are just above the ionizing threshold of 13.6 eV, where the absorption cross-section is high and the mean-free-path is correspondingly short. Therefore, the ionization fronts have a sharp edge, and the reheating of the IGM occurs nearly simultaneously with reionization (see Figure 3 in Gnedin 1999). Once reheating occurs, the subsequent formation of low-mass galaxies is suppressed as predicted by our semi-analytic model (see Figure 8 in Gnedin 1999). The suppression occurs more gradually in Gnedin (1999) than in our model, because of a difference in the assumptions about star formation. In the simulation of Gnedin (1999), gas which cools can continue to form stars for much longer than a dynamical time, until all of it has turned into stars. In our model, we assume that $`10\%`$ of the gas turns into stars, and that feedback heats or expels the rest of the gas and thus halts star formation. We assume that additional star formation in the gas which is left over requires another trigger such as a merger. Our assumption of an upper limit on the efficiency of star formation appears to be required by observations. Local galaxies contain a mass fraction in stars and stellar remnants up to $`1\%`$ of the total halo mass (e.g., Fukugita, Hogan, & Peebles 1998), which is an order of magnitude smaller than the cosmic baryon fraction of $`10\%`$ indicated by observations of X-ray clusters (White & Fabian 1995; Arnaud & Evrard 1999). Feedback is expected to have been most effective at limiting star formation in the shallow potential wells which characterize high redshift galaxies.
Chiu & Ostriker (1999) included a two-phase IGM in their semi-analytic model of reionization, and our results are generally consistent with theirs. While we have focused on the star formation rate and the faint luminosity function, Chiu & Ostriker focused on the reionization process itself and did not explicitly highlight the suppression of star formation or the evolution of the luminosity function.
## 5 Conclusions
We have shown that the reionization of the IGM is expected to have left a clear signature on the history of the cosmic star formation rate. In particular, the photo-ionization heating of the IGM resulted in a suppression of the formation of low-mass galaxies, and in a fairly rapid drop in the average SFR by a factor of $`3`$ relative to the case of no reionization.
Most of the additional star formation in the pre-reionization era occurred in low-mass, faint galaxies. This means that NGST can detect only a fraction of this additional star formation, but this should still suffice to identify the reionization redshift. Even more striking is the substantial increase (by 1โ2 orders of magnitude) in the number density of faint galaxies before reionization. These high-redshift galaxies are expected to appear highly irregular because of their high merger rates. Although the precise shape of the luminosity function depends on the detailed model assumptions, the increase in the number density is dramatic and robust. Much of this increase occurs at a flux limit $`1`$ nJy and should be detectable with an 8-meter NGST.
###### Acknowledgements.
We thank Andrew Blain for providing SFR data from an earlier compilation. We are also grateful to Jerry Ostriker, Nick Gnedin, and Weihsueh Chiu for valuable discussions. RB acknowledges support from Institute Funds. This work was supported in part by NASA grants NAG 5-7039 and NAG 5-7768 for AL. |
warning/0001/hep-ph0001131.html | ar5iv | text | # 1 Introduction
## 1 Introduction
The precise determination of the machine luminosity is necessary for the successful accomplishment of the physics program of $`e^+e^{}`$ colliders operating in the region of the low-lying hadronic resonances, such as DA$`\mathrm{\Phi }`$NE (Frascati) , VEPP-2M (Novosibirsk) , as well as for the BABAR and BELLE experiments at PEP-II and KEKB. In particular, the precise measurement of the hadronic cross section requires a luminosity determination with a total relative error better than 1% . It is well known that the luminosity of $`e^+e^{}`$ colliders can be precisely derived by the relation $`L=N/\sigma _{th}`$, where $`N`$ and $`\sigma _{th}`$ are the number of events and the theoretical cross section of a given reference reaction. In order to make the total luminosity error as small as possible, the reference process should be characterized by a large cross section and calculable with high theoretical accuracy. At low-energy $`e^+e^{}`$ machines, the best candidate fulfilling the above criteria is the Bhabha process ($`e^+e^{}e^+e^{}`$) detected at large scattering angles.
On the theoretical side, precision calculations of the large-angle Bhabha (LABH) cross-section are therefore demanded, with a theoretical accuracy at the $`O(10^3)`$ level. This requires the inclusion in the calculation of all the relevant radiative corrections, in particular the large effects due to photonic radiation. The complete and exclusive simulation of events in generators is also strongly required by the experimental analysis.
## 2 Theoretical approach
The calculation of the Bhabha scattering cross-section, corrected by the effects of photon radiation, and the corresponding event generation is performed according to the master formula
$`\sigma _{corrected}={\displaystyle }dx_{}dx_+dy_{}dy_+{\displaystyle }d\mathrm{\Omega }_{lab}D(x_{},Q^2)D(x_+,Q^2)\times `$
$`D(y_{},Q^2)D(y_+,Q^2){\displaystyle \frac{d\sigma _0}{d\mathrm{\Omega }_{cm}}}(x_{}x_+s,\theta _{cm})J(x_{},x_+,\theta _{lab})\mathrm{\Theta }(cuts).`$ (1)
In the previous equation, the electron Structure Function (SF) $`D(x,Q^2)`$ is the solution of DGLAP equation in QED. It takes into account soft-photon exponentiation and multiple hard bremsstrahlung emission in the leading log (LL) approximation , both for the QED initial-state (ISR) and final-state radiation (FSR). The QED-DGLAP equation can be exactly solved by means of the QED Parton Shower (PS) algorithm , which allows also exclusive photon generation in the LL approximation. A more detailed discussion about the implementation of the PS algorithm as adopted in the present analysis will be given elsewhere . In eq. 1, $`d\sigma _0/d\mathrm{\Omega }`$ is the โhard-scatteringโ differential cross section relevant for centre of mass (c.m.) energy around 1 GeV, including the photonic $`s`$\- and $`t`$-channel diagrams, their interference and the (small) contribution due to $`\mathrm{\Phi }`$ exchange. In the hard-scattering cross section, the correction due to vacuum polarization is taken into account as well, according to the parameterization and the recipe given in ref. . The effect of the running coupling constant at $`\sqrt{s}M_\mathrm{\Phi }`$ is to enhance the cross section by $`2(2.5)\%`$ for $`20^{}(50^{})\vartheta _\pm 160^{}(130^{})`$. The jacobian factor $`J(x_{},x_+,\theta _{lab})`$ in eq. 1 accounts for the boost from the c.m. to the laboratory frame due to emission by initial state $`e^+`$ and $`e^{}`$ of unbalanced radiation, while $`\mathrm{\Theta }(cuts)`$ stands for (arbitrary) experimental cuts implementation.
Upon the above-sketched theoretical background, a new Monte Carlo (MC) generator (BABAYAGA) for simulation of the LABH process at $`\mathrm{\Phi }`$-factories has been developed. In the program both ISR and FSR are simulated and the complete kinematics of the generated events is reconstructed in the LL approximation. The possibility of performing an up to $`O(\alpha )`$ calculation of eq. 1 is included as well, in view of a comparison with the exact $`O(\alpha )`$ perturbative results.
In order to test the precision and the reliability of the Bhabha generator, an exact $`O(\alpha )`$ calculation has also been addressed, by computing the up to $`O(\alpha )`$ corrected cross-section, consisting of soft+virtual and hard photon corrections . Moreover, higher-order LL terms can be summed on top of the exact $`O(\alpha )`$ cross section whitin the collinear SF approach, following the algorithm of ref. . This formulation is available in the form of a MC integrator (LABSPV), which is a suitable modification of the SABSPV code described in ref. .
## 3 Numerical results
In the following, the selection criteria adopted for the analysis and the simulations correspond to realistic data taking at DA$`\mathrm{\Phi }`$NE and VEPP-2M, at c.m. energy $`\sqrt{s}=1.019`$ GeV. The energy cut imposed on the final-state electron and positron is $`E_{min}^\pm =0.4`$ GeV, with the angular acceptance of $`20^{}\vartheta _\pm 160^{}`$ or $`50^{}\vartheta _\pm 130^{}`$ and the (maximum) acollinearity cut allowed to vary in the range $`\xi _{max}=5^{}`$-$`25^{}`$.
A sample of simulations obtained by means of BABAYAGA is shown in fig. 1, where the energy, the cosine of the angle, the $`p_{}`$ of the most energetic photon of each event and the missing mass of the event are plotted. As expected, the behaviour of photonic radiation (soft and collinear to charged particles) is well reproduced by the PS. The effect of FSR has also been investigated. The corrected cross section including only ISR has been compared with the corrected cross section including both ISR and FSR. We noticed that, as a consequence of the rather severe cuts, the total effects of photon radiation is to reduce the integrated cross section by an $`O(10\%)`$ amount. As expected, half of the whole effect must be ascribed to FSR when non-calorimetric (โbareโ) event selection is adopted.
The comparison between the exact $`O(\alpha )`$ calculation and the $`O(\alpha )`$ predictions of the PS generator allows to evaluate the size of the $`O(\alpha )`$ next-to-leading-order (NLO) corrections missing in the LL approximation PS predictions. Moreover, this comparison can be a useful guideline to improve the agreement between perturbative and PS results, for example, by properly choosing the virtuality $`Q^2`$ in the electron SF in such a way that the bulk of $`O(\alpha )`$ NLO terms is effectively reabsorbed into the LL contributions. The scale choice $`Q^2=st/u`$ ($`s`$, $`t`$ and $`u`$ are the usual Mandelstam variables) allows to keep under control the dominant structure due to initial-, final- and initial-final-state interference radiation . As a function of the acollinearity cut, the relative difference between the exact $`O(\alpha )`$ cross section and the corresponding PS one is shown in fig. 2, for the angular acceptances $`20^{}\vartheta _\pm 160^{}`$ and $`50^{}\vartheta _\pm 130^{}`$ and for two different choices of the $`Q^2`$ scale in the PS, i.e. $`Q^2=st/u`$ and $`Q^2=0.75st/u`$. It can be seen that, with the scale $`Q^2=0.75st/u`$, the difference between the exact $`O(\alpha )`$ calculation and the PS predictions is within 0.5%. This naive example illustrates how, for a given selection criterion, the level of agreement can be substantially improved by a simple redefinition of the maximum virtuality of the electromagnetic shower. Going beyond this simple recipe would require a true merging between perturbative calculation and PS scheme, which is beyond the scope of the present analysis.
In addition to the evaluation of the $`O(\alpha )`$ NLO corrections, it is important, for an assessment of the theoretical precision, to quantify the amount of the higher-order LL contributions with typical experimental cuts. The size of LL $`O(\alpha ^nL^n`$) ($`n2`$) corrections can be derived in the PS scheme by comparing the full all-order predictions with the corresponding up to $`O(\alpha )`$ truncation, as shown in fig. 3. The comparison shows that the $`O(\alpha ^nL^n`$) corrections are unavoidable for a theoretical precision better than 1%, being their contribution 0.7% at $`\xi _{max}=5^{}`$ and 0.3-0.4% for larger acollinearity cuts in the angular acceptance $`20^{}\vartheta _\pm 160^{}`$. The $`O(\alpha ^nL^n`$) corrections are even more important, of the order of 1.5%, in the narrower angular range $`50^{}\vartheta _\pm 130^{}`$. It is worth noticing that the generators so far used by the experimental groups at Frascati and Novosibirsk include only $`O(\alpha )`$ corrections, missing the important effect of higher-order contributions.
## 4 Conclusions and perspectives
In order to provide predictions of interest for the luminosity determination at $`e^+e^{}`$ flavour factories, a precision calculation of the LABH process has been addressed. It is based on a QED PS algorithm (the details of the formulation as adopted in the present paper will be given elsewhere ), which accounts for corrections due to ISR and FSR (and interference) in the LL approximation and allows the complete event generation. A new MC event generator (BABAYAGA) has been developed and is available for a full experimental simulation; actually, it is under test at Frascati and Novosibirsk. The overall precision of the PS approach has been checked by means of a benchmark calculation, which includes exact $`O(\alpha )`$ and higher-order LL corrections and is available as a MC integrator (LABSPV), allowing for precise cross section calculations. Critical comparisons between the exact $`๐ช(\alpha )`$ and the $`๐ช(\alpha )`$ PS calculations pointed out that the contribution of the $`O(\alpha )`$ NLO corrections is important for the required theoretical precision. Moreover, the effect of higher-order $`O(\alpha ^nL^n)`$ LL corrections has been evaluated to be at the 1-2% level.
By virtue of its generality, the PS approach could be employed to simulate and to evaluate radiative corrections to other large-angle QED processes, as for example $`e^+e^{}\gamma \gamma `$ or $`e^+e^{}\mu ^+\mu ^{}`$. An interesting application of PS would be the simulation of processes with tagged photons, e.g. $`e^+e^{}hadrons+\gamma `$.
In conclusion, our analysis points out that theoretical predictions aiming at a $`O(10^3)`$ precision must include the effects of both $`O(\alpha )`$ NLO terms and $`O(\alpha ^nL^n)`$ LL contributions. As a consequence of that, we can estimate the present accuracy of our generator BABAYAGA to be at 0.5% level and the accuracy of the integrator LABSPV at 0.1% level. In the future, an improvement of the presented approach is needed by means of an appropriate merging of the exact $`O(\alpha )`$ matrix element with the exclusive photon exponentiation realized by the PS algorithm.
## 5 Acknowledgements
C.M. Carloni Calame is grateful to the organisers for the kind invitation and the pleasant atmosphere during the workshop. The authors are indebted with A. Bukin, G. Capon, G. Cabibbo, A. Denig, S. Eidelman, V.N. Ivanchenko, F. Jegerlehner, V.A. Khoze, G.A. Kukartsev, J. Lee-Franzini, G. Pancheri, I. Peruzzi, Z.A. Silagadze and G. Venanzoni for useful discussions, remarks and interest in their work. The authors acknowledge partial support from the EEC-TMR Program, Contract N. CT98-0169. C.M. Carloni Calame and C. Lunardini wish to thank the INFN, Sezione di Pavia, for the use of computer facilities. |
warning/0001/cond-mat0001218.html | ar5iv | text | # Holon Pairing Instability based on the Bethe-Salpeter Equation obtained from the t-J Hamiltonians of both U(1) and SU(2) Slave-boson Symmetries
## Abstract
We investigate a possibility of holon pairing for bose condensation based on the Bethe-Salpeter equation obtained from the use of the t-J Hamiltonians of both the U(1) and SU(2) slave-boson symmetries. It is shown that the vertex function contributed from ladder diagram series involving holon-holon scattering channel in the Bethe-Salpeter equation leads to a singular behavior at a critical temperature at each hole doping concentration, showing the instability of the normal state against holon pairing. We find that the holon pairing instability occurs only in a limited range of hole doping, by showing an โarchโ shaped bose condensation line in agreement with observation for high $`T_c`$ cuprates. It is revealed that this is in agreement with a functional integral approach of the slave-boson theories.
Since the advent of high $`T_c`$ superconductivity, both the U(1) and SU(2) slave-boson approaches to the t-J Hamiltonian have been proposed to study superconductivity for the two-dimensional systems of copper oxides-. Unlike the U(1) slave-boson mean field theory, the recently proposed SU(2) theory of Wen and Lee has a merit of treating the low energy phase fluctuations of order parameters. Recently we proposed a theory of bose condensation by paying attention to the holon pairing channel in the U(1) slave-boson theory, in contrast to other earlier studies concerned with the single holon condensation. In this paper the phase fluctuations of the spinon pairing and hopping order parameters are taken care of by the SU(2) theory. In addition, following our earlier study the spinon and holon degrees of freedom are introduced into the Heisenberg term on the basis of on-site charge fluctuations which arise as a result of site-to-site electron(and thus holon) hopping in two-dimensional quantum systems of hole-doped high $`T_c`$ cuprates. Most of the slave-boson theories have been concerned with single holon bose condensation. These theories predicted a linear increase of bose condensation temperature with doping concentration, contrary to the observed bose condensation line of an โarchโ shape which manifests the presence of optimal doping rate. Most recently Wen and Lee questioned whether the single holon or the holon pair bose condensation is favored with the SU(2) slave-boson theory. In this work we pay attention to holon pair bose condensation by demonstrating a possibility of holon pairing instability from the study of the Bethe-Salpeter equation obtained from the use of the t-J Hamiltonians of both the U(1) and SU(2) slave-boson symmetries. The present approach is shown to predict the observed characteristics of the bose condensation line of an arch shape in excellent agreement with a functional integral approach to the slave-boson theories.
We write the t-J Hamiltonian,
$`H`$ $`=`$ $`t{\displaystyle \underset{<i,j>}{}}(c_{i\sigma }^{}c_{j\sigma }+c.c.)+J{\displaystyle \underset{<i,j>}{}}(๐_i๐_j{\displaystyle \frac{1}{4}}n_in_j),`$ (1)
where $`๐_i๐_j\frac{1}{4}n_in_j=\frac{1}{2}(c_i^{}c_j^{}c_i^{}c_j^{})(c_jc_ic_jc_i)`$. Here $`๐_i`$ is the electron spin operator at site $`i`$, $`๐_i=\frac{1}{2}c_{i\alpha }^{}๐_{\alpha \beta }c_{i\beta }`$ with $`๐_{\alpha \beta }`$, the Pauli spin matrix element and $`n_i`$, the electron number operator at site $`i`$, $`n_i=c_{i\sigma }^{}c_{i\sigma }`$. In the slave-boson representation- $`c_\sigma `$, the electron annihilation operator of spin $`\sigma `$ can be written as a composite of spinon and holon operators. That is, $`c_\sigma =b^{}f_\sigma `$ in the U(1) theory and $`c_\alpha =\frac{1}{\sqrt{2}}h^{}\psi _\alpha `$ in the SU(2) theory with $`\alpha =1,2`$, where $`f_\sigma `$($`b`$) is the spinon(holon) annihilation operator in the U(1) theory, and $`\psi _1=\left(\begin{array}{c}f_1\\ f_2^{}\end{array}\right)`$ and $`\psi _2=\left(\begin{array}{c}f_2\\ f_1^{}\end{array}\right)`$ and $`h=\left(\begin{array}{c}b_1\\ b_2\end{array}\right)`$ are respectively the doublets of spinon and holon annihilation operators in the SU(2) theory.
After Hubbard Stratonovich transformations in association with the direct, exchange and pairing channels and the saddle point approximation, the effective Hamiltonian is decomposed into the spinon sector and the holon sector separately. Here to explore the instability of the normal state against holon pairing, we pay attention to the holon sector of the Hamiltonian. The holon Hamiltonian is derived to be, for the U(1) slave-boson theory,
$`H_{tJ,U(1)}^b`$ $`=`$ $`t{\displaystyle \underset{<i,j>}{}}\chi _{ij}b_i^{}b_j+c.c.`$ (2)
$``$ $`{\displaystyle \frac{J}{2}}{\displaystyle \underset{<i,j>}{}}|\mathrm{\Delta }_{ij}^f|^2b_i^{}b_j^{}b_jb_i\mu _0{\displaystyle \underset{i}{}}b_i^{}b_i,`$ (3)
where $`\chi _{ij}=<f_{j\sigma }^{}f_{i\sigma }+\frac{4t}{J(1\delta )^2}b_j^{}b_i>`$ is the hopping order parameter and $`\mathrm{\Delta }_{ij}^f=<f_jf_if_jf_i>`$, the spinon pairing order parameter, and for the SU(2) slave-boson theory,
$`H_{tJ,SU(2)}^b={\displaystyle \frac{t}{2}}{\displaystyle \underset{<i,j>}{}}h_i^{}U_{ij}h_j+c.c.`$ (5)
$`{\displaystyle \frac{J}{2}}{\displaystyle \underset{<i,j>,\alpha ,\beta }{}}|\mathrm{\Delta }_{ij}^f|^2b_{i\alpha }^{}b_{j\beta }^{}b_{j\beta }b_{i\alpha }\mu _0{\displaystyle \underset{i}{}}h_i^{}h_i,`$
where $`U_{i,j}=\left(\begin{array}{cc}\chi _{ij}& \mathrm{\Delta }_{ij}^f\\ \mathrm{\Delta }_{ij}^f& \chi _{ij}^{}\end{array}\right)`$ is the order parameter matrix of hopping order, $`\chi _{ij}`$ and spinon pairing order, $`\mathrm{\Delta }_{ij}^f`$ with $`\chi _{ij}=<f_{j\sigma }^{}f_{i\sigma }+\frac{2t}{J(1\delta )^2}(b_{j1}^{}b_{i1}b_{i2}^{}b_{j2})>`$ and $`\mathrm{\Delta }_{ij}^f=<f_{j1}f_{i2}f_{j2}f_{i1}>`$.
Introducing a uniform hopping order parameter, $`\chi _{ji}=\chi `$, and a d-wave spinon pairing order parameters, $`\mathrm{\Delta }_{ji}^f=\pm \mathrm{\Delta }_f`$(the sign $`+()`$ is for the $`\mathrm{๐ข๐ฃ}`$ link parallel to $`\widehat{x}`$ ($`\widehat{y}`$)), we obtain from Eq.(3) the momentum space representation of the Hamiltonian, for the U(1) theory,
$`H_{tJ,U(1)}^b={\displaystyle \underset{๐ค}{}}ฯต(๐ค)b_๐ค^{}b_๐ค`$ (7)
$`+{\displaystyle \frac{1}{2N}}{\displaystyle \underset{๐ค,๐ค^{^{}},๐ช}{}}v(๐ค๐ค^{^{}})b_{๐ค^{^{}}+๐ช}^{}b_๐ค^{^{}}^{}b_๐คb_{๐ค+๐ช},`$
where $`ฯต(๐ค)=2t\chi \gamma _๐ค\mu _0`$, the energy dispersion of holon with $`\gamma _๐ค=(\mathrm{cos}k_x+\mathrm{cos}k_y)`$ and $`v(๐ค^{^{}}๐ค)=J|\mathrm{\Delta }_f|^2\gamma _๐ค`$, the momentum space representation of holon-holon interaction. Likewise we obtain from Eq.(5),
$`H_{tJ,SU(2)}^b={\displaystyle \underset{๐ค}{}}h_๐ค^{}\left(\begin{array}{cc}t\chi \gamma _๐ค\mu _0& t\mathrm{\Delta }_f\phi _๐ค\\ t\mathrm{\Delta }_f\phi _๐ค& t\chi \gamma _๐ค\mu _0\end{array}\right)h_๐ค`$ (11)
$`+{\displaystyle \frac{1}{2N}}{\displaystyle \underset{๐ค,๐ค^{^{}},๐ช,\alpha ,\beta }{}}v(๐ค๐ค^{^{}})b_{๐ค^{^{}}+๐ช,\beta }^{}b_{๐ค^{^{}},\alpha }^{}b_{๐ค,\alpha }b_{๐ค+๐ช,\beta },`$
where $`\alpha `$, $`\beta `$ ($`=1,2`$) represent two isospin components of SU(2) holon and $`\phi _๐ค=(\mathrm{cos}k_x\mathrm{cos}k_y)`$. $`N`$ is the total number of sites for the two dimensional lattice of interest.
Considering the t-matrix of involving the particle-particle(holon-holon) scattering channels, we obtain the Bethe-Salpeter equations in the U(1) slave-boson representation,
$`<k^{^{}},k^{^{}}+q|t|k,k+q>_{U(1)}=v(๐ค^{^{}}๐ค)`$ (14)
$`{\displaystyle \frac{1}{N\beta }}{\displaystyle \underset{k^{^{\prime \prime }}}{}}v(๐ค^{^{}}๐ค^{^{\prime \prime }})g(k^{^{\prime \prime }})g(k^{^{\prime \prime }}+q)\times `$
$`<k^{^{\prime \prime }},k^{^{\prime \prime }}+q|t|k,k+q>_{U(1)},`$
where $`g(k)=_0^\beta ๐\tau e^{ik_0\tau }<T_\tau [b_๐ค(\tau )b_๐ค^{}(0)]>`$, the holon Matsubara Greenโs function in the U(1) slave-boson representation, and in the SU(2) slave-boson representation,
$`<k^{^{}},\alpha ^{^{}};k^{^{}}+q,\beta ^{^{}}|t|k,\alpha ;k+q,\beta >_{SU(2)}`$ (18)
$`=v(๐ค^{^{}}๐ค)\delta _{\alpha ^{^{}},\alpha }\delta _{\beta ^{^{}},\beta }`$
$`{\displaystyle \frac{1}{N\beta }}{\displaystyle \underset{k^{^{\prime \prime }},\alpha ^{^{\prime \prime }},\beta ^{^{\prime \prime }}}{}}v(๐ค^{^{}}๐ค^{^{\prime \prime }})g_{\alpha ^{^{}}\alpha ^{^{\prime \prime }}}(k^{^{\prime \prime }})g_{\beta ^{^{}}\beta ^{^{\prime \prime }}}(k^{^{\prime \prime }}+q)`$
$`\times <k^{^{\prime \prime }},\alpha ^{^{\prime \prime }};k^{^{\prime \prime }}+q,\beta ^{^{\prime \prime }}|t|k,\alpha ;k+q,\beta >_{SU(2)},`$
where $`g_{\alpha \beta }(k)=_0^\beta ๐\tau e^{ik_0\tau }<T_\tau [b_{\alpha ๐ค}(\tau )b_{\beta ๐ค}^{}(0)]>`$, the holon Matsubara Greenโs function in the SU(2) slave-boson representation. Here $`k(k_0,๐ค)`$ is the three component vector of the energy-momentum of holon. $`q=(q_0,๐ช)`$ is the three component vector of the total energy-momentum of holon pair.
After summing over the Matsubara frequencies in Eqs.(14) and (18), we obtain the following matrix equation for the t-matrix, in the U(1) slave-boson representation,
$`{\displaystyle \underset{๐ค^{^{\prime \prime }}}{}}\left(\delta _{๐ค^{^{}},๐ค^{^{\prime \prime }}}m_{๐ค^{^{}},๐ค^{^{\prime \prime }}}(q_0,๐ช)\right)t_{๐ค^{^{\prime \prime }},๐ค}(q_0,๐ช)=v(๐ค^{^{}}๐ค),`$ (19)
where $`t_{๐ค^{^{}},๐ค}(q_0,๐ช)<k^{^{}},k^{^{}}+q|t|k,k+q>_{U(1)}`$,
$`m_{๐ค^{^{}},๐ค^{^{\prime \prime }}}(q_0,๐ช){\displaystyle \frac{1}{N\beta }}{\displaystyle \underset{k_0^{^{\prime \prime }}}{}}v(๐ค^{^{}}๐ค^{^{\prime \prime }})g(k^{^{\prime \prime }})g(k^{^{\prime \prime }}+q)`$ (22)
$`={\displaystyle \frac{v(๐ค^{^{}}๐ค^{^{\prime \prime }})}{N\beta }}\left[{\displaystyle \underset{k_0^{^{\prime \prime }}}{}}{\displaystyle \frac{1}{ik_0^{^{\prime \prime }}ฯต(๐ค^{^{\prime \prime }})}}{\displaystyle \frac{1}{i(k_0^{^{\prime \prime }}+q_0)ฯต(๐ค^{^{\prime \prime }}+๐ช)}}\right]`$
$`={\displaystyle \frac{1}{N}}v(๐ค^{^{}}๐ค^{^{\prime \prime }}){\displaystyle \frac{n(ฯต(๐ค^{^{\prime \prime }}))+e^{\beta ฯต(๐ค^{^{\prime \prime }}+๐ช)}n(ฯต(๐ค^{^{\prime \prime }}+๐ช))}{iq_0(ฯต(๐ค^{^{\prime \prime }}+๐ช)+ฯต(๐ค^{^{\prime \prime }}))}},`$
and $`n(ฯต)=\frac{1}{e^{\beta ฯต}1}`$, the boson distribution function. Similarly, we obtain, in the SU(2) slave-boson representation,
$`{\displaystyle \underset{๐ค^{^{\prime \prime }},\alpha ^{^{\prime \prime }},\beta ^{^{\prime \prime }}}{}}\left(\delta _{๐ค^{^{}},๐ค^{^{\prime \prime }}}\delta _{\alpha ^{^{}}\alpha ^{^{\prime \prime }}}\delta _{\beta ^{^{}}\beta ^{^{\prime \prime }}}m_{๐ค^{^{}},๐ค^{^{\prime \prime }}}^{\alpha ^{^{}}\beta ^{^{}}\alpha ^{^{\prime \prime }}\beta ^{^{\prime \prime }}}(q_0,๐ช)\right)t_{๐ค^{^{\prime \prime }},๐ค}^{\alpha ^{^{\prime \prime }}\beta ^{^{\prime \prime }}\alpha \beta }(q_0,๐ช)`$ (24)
$`=v(๐ค^{^{}}๐ค)\delta _{\alpha ^{^{}}\alpha }\delta _{\beta ^{^{}}\beta },`$
where $`t_{๐ค^{^{}},๐ค}^{\alpha ^{^{}}\beta ^{^{}}\alpha \beta }(q_0,๐ช)`$ $`<k^{^{}},\alpha ^{^{}};k^{^{}}+q,\beta ^{^{}}|t|k,\alpha ;k+q,\beta >_{SU(2)}`$ and $`m_{๐ค^{^{}},๐ค^{^{\prime \prime }}}^{\alpha ^{^{}}\beta ^{^{}}\alpha ^{^{\prime \prime }}\beta ^{^{\prime \prime }}}(q_0,๐ช)`$ $`\frac{1}{N}_{\alpha _1^{^{}}\beta _1^{^{}}}v(๐ค^{^{}}๐ค^{^{\prime \prime }})\frac{n(E_{\alpha _1^{^{}}}(๐ค^{^{\prime \prime }}))+e^{\beta E_{\beta _1^{^{}}}(๐ค^{^{\prime \prime }}+๐ช)}n(E_{\beta _1^{^{}}}(๐ค^{^{\prime \prime }}+๐ช))}{iq_0(E_{\alpha _1^{^{}}}(๐ค^{^{\prime \prime }})+E_{\beta _1^{^{}}}(๐ค^{^{\prime \prime }}+๐ช))}\times `$ $`U_{\alpha ^{^{}}\alpha _1^{^{}}}(๐ค^{^{\prime \prime }})U_{\beta ^{^{}}\beta _1^{^{}}}(๐ค^{^{\prime \prime }}+๐ช)U_{\alpha _1^{^{}}\alpha ^{^{\prime \prime }}}^{}(๐ค^{^{\prime \prime }})U_{\beta _1^{^{}}\beta ^{^{\prime \prime }}}^{}(๐ค^{^{\prime \prime }}+๐ช)`$. Here $`E_1(๐ค)=E_๐ค\mu _0`$ and $`E_2(๐ค)=E_๐ค\mu _0`$ are the energy dispersions of the upper and lower bands of holons with $`E_๐ค=t\sqrt{(\chi \gamma _๐ค)^2+(\mathrm{\Delta }_f\phi _๐ค)^2}`$. $`U_{\alpha \beta }(๐ค)=\left(\begin{array}{cc}u_๐ค& v_๐ค\\ v_๐ค& u_๐ค\end{array}\right)`$ is the unitary transformation matrix used for the diagonalization of the one-body holon Hamiltonian in Eq.(11) with $`u_๐ค=\frac{1}{\sqrt{2}}\sqrt{1\frac{t\chi \gamma _๐ค}{E_๐ค}}`$ and $`v_๐ค=\frac{1}{\sqrt{2}}\sqrt{1+\frac{t\chi \gamma _๐ค}{E_๐ค}}`$. It is noted that the t-matrices, $`t_{๐ค^{^{}},๐ค}(q_0,๐ช)`$ and $`t_{๐ค^{^{}},๐ค}^{\alpha ^{^{}}\beta ^{^{}}\alpha \beta }(q_0,๐ช)`$ for the U(1) and SU(2) theories respectively are independent of the Matsubara frequencies $`k_0^{^{}}`$ and $`k_0`$, owing to the consideration of instantaneous holon-holon interaction, $`v(๐ค^{^{}}๐ค)`$.
As mentioned earlier, the t-matrix elements are obtained from Eqs.(14) and (18), both of which involve summation over the Matsubara frequencies. Using the matrix equations (19) and (24), the poles of the t-matrices are searched for as a function of energy $`q_0`$ with $`๐ช=0`$, i.e., the zero total momentum of the holon pair. The t-matrices are numerically evaluated from the use of Eqs.(19) and (24) for each doping rate and temperature. The hopping and spinon pairing order parameters in Eqs.(3) and (5) are the saddle point values evaluated from the usual partition functions involving the functional integrals of slave-boson representation. For both the U(1) and SU(2) t-matrix calculations we choose $`J/t=0.3`$ in Eqs.(3) and (5).
At high temperatures, computed poles are found to be positive and real, indicating that there exist no bound states. As temperature is lowered to a critical(onset) value $`T_c`$, we find that with the U(1) theory one pole changes its sign from positive to negative, indicating that their exists an instability of the normal state against holon pairing at the onset(critical) temperature $`T_c`$. Similarly, with the SU(2) slave-boson theory two poles change their signs from positive to negative at $`T_c`$; these two poles correspond to the particle-particle(holon-holon) scattering channels of $`b_1`$-$`b_1`$ and $`b_2`$-$`b_2`$ respectively. In Fig.1. the onset temperature $`T_c`$(denoted by solid square in Fig.1) for the appearance of such negative pole(s) is predicted to occur only in a limited range of hole doping concentration, by revealing an โarchโ shaped feature of bose condensation line in agreement with observation in the phase diagram of high $`T_c`$ cuprates. The onset temperature of negative pole(s) coincide(s) surprisingly well with the critical temperature of holon pair condensation obtained from the functional integral approach(denoted by open square in Fig.1) of the slave-boson theories. Holon-holon scattering occurs above the lowest possible single particle energy $`ฯต(๐ค=0)`$. The negative pole corresponds to the binding energy of the holon pair. This is analogous to the binding energy of electron pairs which results from the Cooperโs two particle problem of electron-electron scattering only above the Fermi energy $`ฯต(๐ค_f)`$, i.e., the Fermi-surface.
In order to find the symmetry of the holon pairing order parameter we now compute the eigenvectors of the t-matrix whose poles change their signs at a critical(onset) temperature $`T_c`$; the eigenvalue equations are
$`{\displaystyle \underset{๐ค^{^{}}}{}}t_{๐ค,๐ค^{^{}}}(q_0=0,๐ช=0)W(๐ค^{^{}})=\lambda W(๐ค),`$ (25)
for the U(1) theory, where $`W(๐ค)`$ is the eigenvector and $`\lambda `$, the eigenvalue and
$`{\displaystyle \underset{๐ค^{^{}},\alpha ^{^{}},\beta ^{^{}}}{}}t_{๐ค,๐ค^{^{}}}^{\alpha \beta \alpha ^{^{}}\beta ^{^{}}}(q_0=0,๐ช=0)W_{\alpha ^{^{}}\beta ^{^{}}}(๐ค^{^{}})=\lambda W(๐ค)_{\alpha \beta },`$ (26)
for the SU(2) theory, where $`W_{\alpha \beta }(๐ค)`$ is the eigenvector concerned with the SU(2) isospin channels $`\alpha `$ and $`\beta `$($`\alpha =1,2`$ and $`\beta =1,2`$ ) and $`\lambda `$, the corresponding eigenvalue. For the case of U(1) there are $`N`$ eigenvectors corresponding to $`N`$ discrete values of momenta for an $`N\times N`$ reciprocal lattice. For SU(2), $`4N`$ eigenvectors are available in accordance with the isospin channels of holon($`b_\alpha `$)-holon($`b_\beta `$) scattering. We search for the eigenvectors whose eigenvalues diverge in the limit of $`q_0=0`$ and $`๐ช=0`$ at an onset(critical) temperature $`T_c`$, that is, the eigenvectors corresponding to the pole of the t-matrix whose sign changes at $`T_c`$. We choose $`N=10\times 10`$ for a reciprocal lattice to determine the symmetry of holon pairing at an underdoping rate, $`\delta =0.05`$.
The computed eigenvectors for both the U(1) and SU(2) slave-boson approaches yielded good fits to the s-wave symmetry of the form $`\mathrm{cos}k_x+\mathrm{cos}k_y`$ in momentum space ; for the U(1) theory we have
$`W(๐ค)=A(\mathrm{cos}k_x+\mathrm{cos}k_y),`$ (27)
where $`A`$ is the normalization constant to satisfy $`_๐ค|W(๐ค)|^2=1`$, and for the SU(2) theory,
$`W_{\alpha \beta }^e(๐ค)=A^e(\mathrm{cos}k_x+\mathrm{cos}k_y)(\delta _{\alpha ,1}\delta _{\beta ,1}+\delta _{\alpha ,2}\delta _{\beta ,2}),`$ (28)
and
$`W_{\alpha \beta }^o(๐ค)=A^o(\mathrm{cos}k_x+\mathrm{cos}k_y)(\delta _{\alpha ,1}\delta _{\beta ,1}\delta _{\alpha ,2}\delta _{\beta ,2}),`$ (29)
for $`b_1`$ holon and $`b_2`$ holon pairing, where $`A^{e(o)}`$ is the normalization constant to satisfy $`_{๐ค,\alpha ,\beta }|W_{\alpha \beta }^{e(o)}(๐ค)|^2=1`$. One of the two computed eigenvectors shows a phase difference of even multiples of $`\pi `$ between the $`b_1`$ and $`b_2`$ holon pairing order parameters and is well fitted by Eq.(28); the other computed eigenvector is well fitted by Eq.(29), by showing a phase difference of odd multiples of $`\pi `$. The divergence of eigenvalues was seen to occur nearly at the same temperature(within a numerical accuracy of $`10^4t`$). Fig.2 displays the U(1) result of the s-wave symmetry of the holon pairing order. In Figs.2(a) and (b), we show the SU(2) results of the s-wave symmetry of opposite phase between the $`b_1`$ and $`b_2`$ holon pairing order parameters. Both results demonstrated excellent fits to the s-wave symmetry form of Eqs.(27) and (29). Since the computed results which fits Eq.(28) are indistinguishable with the ones shown in both Fig.2 and Figs.3 (a) and thus are not displayed. In short, both the U(1) and SU(2) theories predicted the s-wave symmetry of holon pairing order. This is equivalent to the d-wave symmetry of hole pairing order, by noting that the hole is a composite of a holon(boson) of spin $`0`$ with charge $`+e`$ and a spinon(fermion) of spin $`1/2`$ with charge $`0`$ and realizing the s-wave holon pairing in the presence of the d-wave spinon pairing-.
In the present study, we found that the bose condensation of an arch shape occurs only in a limited range of hole doping concentration, by searching for the onset temperature of the holon pairing instability. It is found that the predicted holon pair bose condensation temperature is in precise agreements with the functional integral approaches of both the U(1) and the SU(2) slave-boson theories. Most importantly, both of these approaches were found to satisfactorily predict the experimentally observed bose condensation (superconducting) temperature as a function of the hole doping rate, by revealing the presence of the optimal doping rate in high $`T_c`$ cuprates.
One(SHSS) of us acknowledges the generous supports of Korea Ministry of Education(BSRI-98 and 99) and the Center for Molecular Science at Korea Advanced Institute of Science and Technology. We thank Tae-Hyoung Gimm, Jae-Hyun Uhm and Ki-Suk Kim for helpful discussions. |
warning/0001/quant-ph0001111.html | ar5iv | text | # Noether Theorem and the quantum mechanical operators
## Abstract
We show that the quantum mechanical momentum and angular momentum operators are fixed by the Noether theorem for the classical Hamiltonian field theory we proposed.
Recently, we have proposed a classical Hamiltonian field theory ,which in the limit of very large Planck frequency,mimics many aspects of a quantum mechanical system. In particular, the Schrodinger Equation will follow from the Hamiltonโs Field Equation, and the Hamiltonian of the classical field theory will become the energy expectation value of the corresponding quantum mechanical system.
The Hamiltonian density for the classical field theory we proposed contains (1+2n) pairs of canonical conjugate variables ($`p,q`$) ( $`P_j,Q_j`$), ($`\pi _j,\eta _j`$ ), j=1,โฆ,n.All of these canonical variables are functions of $`x=(x_1,..,x_n)`$ and t.And it reads as
$$H=(1/2h)(V(x))(p^2+q^2)(1/2h)(mc^2)(P_j^2+Q_j^2+\pi _j^2+\eta _j^2)(c/2)p_j(Q_j+\eta _j)(c/2)(P_j+\pi _j)_jq$$
(1)
Independent variations of the field variables generate the Hamiltonian Field Equation. If we are interested in the case in which the field variables oscillate with frequencies far smaller than the Planck frequency h/mc, then the variables $`P_j,Q_j`$ and $`\pi _j,\eta _j`$ will be related to $`p,`$q through
$`Q_i`$ $`=`$ $`h/2mc_ip`$
$`P_i`$ $`=`$ $`h/2mc_iq`$
$`\eta _i`$ $`=`$ $`h/2mc_ip`$
$`\pi _i`$ $`=`$ $`h/2mc_iq,j=1,..n`$ (2)
In this paper, we will explore the translational and rotational symmetries of this classical Hamiltonian field theory. It is well known in the literature that these continuous symmetries will lead to conserved physical quantities ; a result called the Noether theorem .And the aim of this paper is to understand how these conserved physical quantities coming from a classical field theory are related to the measurable quantities of the corresponding quantum mechanical system.
Let us first consider the $`V(x)=0`$ case.When there is no external potential present, there will be both translational and rotational symmetries for the classical field system. And by Noether theorem, there exist some corresponding conserved quantities. For the translational invariance, the resulting conserved quantities are the components of the second rank tress-energy tensor $`T_\nu ^\mu `$,given by Noether as
$$T_\nu ^\mu =\mathrm{\Sigma }(L/(_\mu u))(_\nu u)L\delta _\nu ^\mu $$
(3)
where L is the underlying Lagrangian density for the classical field theory. u stands collectively for all the field variables.Noether theorem requires the conservation law
$$_\mu T_\nu ^\mu =0$$
(4)
We are particularly interested in the vector $`m_j`$ defined as
$$m_j=d^nxT_{0j}$$
(5)
These $`m_j`$,other than a multiplicative constant that we shall fix later,are always taken as the mementum components carried by the classical fields because they are generated by the translational symmetries. A close look at $`T_j^0`$ will show that they are independent of the detailed structures of the Lagrangian density L and has the simple form of
$$T_j^0=\mathrm{\Sigma }(L/(_tu))(_ju)=p_jq+P_i_jQ_i+\pi _i_j\eta _i$$
(6)
Using the result given in Eq(2),$`T_j^0`$ can be written in terms of $`p`$ and $`q`$ as
$$T_j^0=p_jq2(h/2mc)^2_ip_j_iq$$
(7)
For a very large Planck frequency , the second term of Eq(7) drops out, and hence
$$m_j=d^nx(p_jq)$$
(8)
If we define the corresponding quantum mechanical wave function by $`\psi (x,t)`$=($`q(x,t)+ip(x,t)`$)/$`\sqrt{2}`$. It can be seen immediately that
$$m_j=d^nx\psi (i_j)\psi ,$$
(9)
after integration by parts.
The physical meaning of the above result is the following: If we use $`p_j`$ to denote the quantum mechanical operator for the j the component of the momentum, and if we use the above $`\psi `$ to compute the expectation value of the momentum components, then $`p_j`$ must be of the form
$$p_j=i\beta _j$$
(10)
where $`\beta `$ is a proportional constant that will be shown to be h later.This result can be regarded as a derivation of the most fundamental quantum mechanical prescription
$$p_j=ih_j$$
(11)
For the rotational invariance, we assume that $`p`$,$`q`$,$`P_j`$,$`Q_j`$,$`\pi _j`$ and $`\eta _j`$ all transform as scalars under the rotation group.The resulting conserved quantities will then be the components of the third rank angular momentum tensor$`M_{\lambda \mu }^\beta `$, given by Noether as
$$M_{\lambda \mu }^\beta =x_\mu T_\lambda ^\beta x_\lambda T_\mu ^\beta $$
(12)
The components of this third rank tensor that are related to the angular momentum components of the classical fields are $`M_{lk}^0`$.Using the result given in Eq(7), it can be shown easily that the integrated components
$$L_{lk}=d^nxM_{lk}^0$$
(13)
can be written as
$$L_{lk}=d^nx\psi (ix_l_k+ix_k_l)\psi $$
(14)
And hence the orbital angular momentum $`L`$ in quantum mechanics will have the familiar form
$$L=rXp$$
(15)
In the presence of the potential V(x), we will no longer have translational invariance, and so no more conservation law. Instaed we shall have
$$d/dx_\mu (T_\nu ^\mu )=_\nu L=_\nu H$$
(16)
The integrated spatial parts for the above equation read as
$$_td^nxT_{0j}+d^nxd/๐x_lT_{lj}=d^nx_jH$$
(17)
Throwing away the surface term,and using the results given in Eq(1),Eq(8) and Eq(9), Eq(17) will become
$$_td^nx\psi (i_j)\psi =1/hd^nx\psi (_jV)\psi $$
(18)
or
$$_td^nx\psi (ih_j)\psi =d^nx\psi (_jV)\psi $$
(19)
This is the Ehrenfest theorem that we always encounter in quantum mechanics. And as we have promised before, we have fixed the proportional constant $`\beta `$ that appeared in Eq(10) to be the Planck constant h.
An equation similar to Eq(18) can also be derived for the angular momentum which relates the rate of change of angular momentum with the external applied torque.
So we may conclude our paper by saying that the quantum mechanical operators for the momentum and angular momentum variables will be fixed by the Noether theorem for our classical Hamiltonian field theory. |
warning/0001/hep-th0001214.html | ar5iv | text | # 1 Introduction
## 1 Introduction
For supersymmetric theories superalgebras are powerful tool to explore non-perturbative aspects about BPS states . About a decade ago Green showed the possibility of the central extension of the Super-Poincar$`\stackrel{ยด}{\mathrm{e}}`$ algebra , where the momentum charge does not commute with the supercharge and gives rise to the fermionic central charge<sup>1</sup><sup>1</sup>1$`Q,P,Z`$ are a supercharge, a momentum and a fermionic central charge. $`\mathrm{\Gamma }`$ and $`C=i\mathrm{\Gamma }^0`$ are gamma matrix and a charge conjugation matrix, a slash represents contraction with $`\mathrm{\Gamma }^m`$, e.g. $`/P=P_m\mathrm{\Gamma }^m`$. :
$`\{Q_\alpha ,Q_\beta \}=2(C/P)_{\alpha \beta },[P_m,Q_\alpha ]=(Z\mathrm{\Gamma }_m)_\alpha .`$ (1.1)
Siegel proposed a good use of this algebra that the Wess-Zumino action of the Green-Schwarz superstring can be obtained as a local bilinear form , namely an element of the Chevalley-Eilenberg cohomology . Recently this idea has been applied to superbrane theories and super-D-brane theories . As well as a superstring these objects require the Wess-Zumino actions to include the kappa symmetry. The Wess-Zumino actions for $`p`$-branes and D-$`p`$branes are higher rank tensors and/or containing U(1) fields, so integrating and obtaining their local expressions are complicated . These complications can be simplified, if Siegelโs method is used. As long as the algebra (1.1) tells the truth in this sense, its physical interpretation will be the next question.
It is interesting to compare the global superalgebra with the local superalgebra. In order to realize the global superalgebra one needs the local supersymmetry constraints $`F_\alpha =0`$ to remove extra fermionic degrees of freedom. They anticommute with the global supercharges, $`\{F,Q\}=0`$ . ($`Q_\alpha ,P_m`$) are global charges, while ($`F_\alpha ,D_m`$) are covariant derivatives satisfying following algebras;
$`\{F_\alpha ,F_\beta \}=2(C/D)_{\alpha \beta },[D_m,F_\alpha ]=(W\mathrm{\Gamma }_m)_\alpha .`$ (1.2)
$`W`$ is named the super-Yang-Mills field strength . It is also expected that $`W_\alpha `$ and $`Z_\alpha `$ are a global charge and a covariant derivative respectively in a larger superspace. In order to see the physical role of $`Z_\alpha `$, it is useful to take a concrete model. For a superstring case charges and covariant derivatives are given in terms of space-time coordinates $`X`$ and $`\theta `$, for example
$`W_\alpha \overline{\theta }_\alpha ^{}`$ (1.3)
where stands for the derivative with respect to worldsheet spacial coordinate $`\sigma `$. (1.3) suggests that the new fermionic central charges are realized by $`Z_\alpha \overline{\theta }_\alpha ^{}`$. This is a surface term, so nontrivial $`Z_\alpha `$ will require nontrivial boundary values of the fermionic coordinates.
Bergshoeff and Sezgin applied the Siegelโs method to a M2 brane where commutators of supercharges and not only the momentum charge but also brane charges allow fermionic central extension . Further studies have been done for NS1 (IIA,IIB) and D1 , for D3, D5 and NS5 and for D2 and M5 . In the work on SL(2,R)$``$SO(2,1) covariant central extension , the momentum and the NS/NS and R/R brane charges are treated equally as SO(2,1) triplet elements. Then the SO(2,1) triplet fermionic central charges appear. It was shown that there exists the Rarita-Schwinger type constraint on the new fermionic central charges and one of the triplet elements is redundant. In other words, the momentum or the brane charge can be chosen to be super-invariant. In this paper we will consider the super-invariant momentum and the super-noninvariant brane charges. In the super-noninvariant momenta can be realized by introducing auxiliary fields. Instead, in this paper we will examine the super-noninvariant brane charges by introducing a nontrivial fermionic boundary condition without any auxiliary fields.
Organization of the present paper is as follows: In section 2 we discuss the fermionic central extension of the superalgebra from the BPS state point of view. The Jacobi identity of three supercharges restricts that the commutator of the brane charge and the broken supersymmetry charge allows the fermionic central extension. In section 3 we will justify this by examining equations of motion in a suitable gauge condition. We show that there exists nontrivial fermionic solution which can give non-zero values of new fermionic brane charges. In section 4 we will also consider the supergravity coupling with this state. The quantum states and the vertex operators for D-branes are not known because of the difficulty of the soliton quantization. On the other hand classical soliton property may suggest how to couple with the supergravity fields. Since it is supposed that D-branes are static objects, the nonrelativistic approximation will be a good approximation. So we begin by the classical Gauss law equation, then examine its consistency under the broken supersymmetry. The unbroken supersymmetry transformation of a purely bosonic soliton solution has been studied in . Usually, including this reference, the fermionic supergravity fields are set to be zero. However we rather consider nontrivial fermionic supergravity fields. It turns out that there can be nonzero dilatino fields whose some mode couple to the new fermionic brane currents.
## 2 BPS state with new fermionic brane charges
We begin by a D-string case as a simplest example. The D-string superalgebra is given by<sup>2</sup><sup>2</sup>2Notation follows where $`(X,\theta )`$ are space-time coordinates and their conjugates are $`(p,\zeta )`$.
$`\{Q_{A\alpha },Q_{B\beta }\}`$ $`=`$ $`2(C\mathrm{\Xi }_G)_{A\alpha B\beta },\mathrm{\Xi }_G=\mathrm{\Gamma }_{\alpha \beta }^m\left(\mathrm{๐}_{AB}P_m(\tau _I)_{AB}\mathrm{\Sigma }_m^I\right).`$ (2.1)
Concrete expression of the global supercharges, the total momenta and the brane charges are
$`Q_{A\alpha }`$ $`=`$ $`{\displaystyle ๐\sigma [\zeta \overline{\theta }/p+\overline{\theta }/X^{}\tau _q\frac{1}{6}\overline{\theta }\mathrm{\Gamma }\tau _q\theta ^{}\overline{\theta }\mathrm{\Gamma }\frac{1}{6}\overline{\theta }\mathrm{\Gamma }\theta ^{}\overline{\theta }\mathrm{\Gamma }\tau _q]_{A\alpha }},`$
$`P_m`$ $`=`$ $`{\displaystyle ๐\sigma p_m},`$ (2.2)
$`\mathrm{\Sigma }_m^I`$ $`=`$ $`q^I{\displaystyle _{\sigma _I}^{\sigma _F}}๐\sigma X_m^{}=q^I(X_FX_I)_m,`$ (2.3)
where $`\sigma _{F,I}`$ are D-string end points. The massiveness of a D-string is reflected to its infinity length of $`\mathrm{\Sigma }L|_L\mathrm{}`$.
One may notice that there is no way to centrally extend of $`[P,Q]`$,
$`[P_m,Q]`$ $`=`$ $`[{\displaystyle ๐\sigma _1p_m(\sigma _1)},{\displaystyle ๐\sigma _2\overline{\theta }/X^{}(\sigma _2)\tau _q}]`$ (2.4)
$`=`$ $`{\displaystyle ๐\sigma _2_2\left(๐\sigma _1\delta (\sigma _1\sigma _2)\right)\overline{\theta }(\sigma _2)\mathrm{\Gamma }_m\tau _q}=0`$
with $`\tau _q=q^I\tau _I=q^{NS}\tau _3+q^R\tau _1`$ (I=NS,R or 3,1). On the other hand, the brane charges $`\mathrm{\Sigma }^I`$ allow to have the central extension in a commutator with the supercharges:
$`[\mathrm{\Sigma }_m^I,Q]`$ $`=`$ $`q^I{\displaystyle ๐\sigma _1_1\left(๐\sigma _2\overline{\theta }(\sigma _2)\mathrm{\Gamma }_m\delta (\sigma _1\sigma _2)\right)}`$ (2.5)
$`=`$ $`Z^I\mathrm{\Gamma }_m.`$
Now $`Z`$ may be interpreted as a fermionic brane charge
$`Z^I=q^I{\displaystyle _{\sigma _I}^{\sigma _F}}๐\sigma \overline{\theta }^{}=q^I(\overline{\theta }_F\overline{\theta }_I).`$ (2.6)
This charge will be infinity volume $`ZL\overline{\eta }|_L\mathrm{}`$ with a constant spinor $`\eta `$, as same as a usual bosonic brane charge because the supertranslation does not change the volume in general.
The superalgebra (2.1) shows that a D-string ground state is BPS saturated where 1/2 supersymmetry are broken. The right hand side of the (2.1) is a projection operator
$`C\mathrm{\Xi }_G=2iM๐ซ_{},M=T_qL`$ (2.7)
with a D-string mass $`M`$, a D-string tension $`T_q=\sqrt{(q^I)^2}`$ and a string length $`L`$, and supersymmetry(SUSY) charges are projected into unbroken SUSY charges ($`๐ฌ`$) and broken SUSY charges ($`๐ฎ`$)
$`Q`$ $`=`$ $`\{\begin{array}{ccc}๐ฌ& & Q๐ซ_+\\ ๐ฎ& & Q๐ซ_{}\end{array}`$ (2.10)
in such a way that
$`\{๐ฌ,๐ฌ\}=0=\{๐ฌ,๐ฎ\},\{๐ฎ,๐ฎ\}0`$ (2.11)
then
$`๐ฌ|0=0,๐ฎ|00,`$ (2.12)
for a ground state $`|0`$. The projection operators $`๐ซ_\pm `$ defined as (2.7) become simple form<sup>3</sup><sup>3</sup>3Gamma matrices is denoted as $`\mathrm{\Gamma }^{m_1\mathrm{}m_N}=`$ $`\frac{1}{N!}{\displaystyle \underset{\mathrm{antisymmetrized}\mathrm{m}}{}}`$ $`\mathrm{\Gamma }^{m_1}\mathrm{}\mathrm{\Gamma }^{m_N}`$. for a ground state in the static gauge
$`๐ซ_\pm ={\displaystyle \frac{1}{2}}\left(\mathrm{๐}_{\alpha \beta }\mathrm{๐}_{AB}\pm (\mathrm{\Gamma }_{01})_{\alpha \beta }(\widehat{\tau }_q)_{AB}\right),\widehat{\tau }_q=\tau _q/|\tau _q|.`$ (2.13)
It is shown that the central extension (2.5) is possible only for broken SUSY
$`[\mathrm{\Sigma }_m,๐ฎ_\alpha ]=(Z\mathrm{\Gamma }_m)_\alpha ,`$ (2.14)
since
$`[\mathrm{\Sigma }_m,๐ฌ_\alpha ]=0`$ (2.15)
from the Jacobi identity of three $`Q`$โs by using with (2.11)
$`[\{Q_\alpha ,Q_\beta \},๐ฌ_\gamma ]`$ $`=`$ $`(C\mathrm{\Gamma }^m\tau _I)_{\alpha \beta }[\mathrm{\Sigma }_m^I,๐ฌ_\gamma ]`$ (2.16)
$`=`$ $`[\{๐ฌ_\gamma ,Q_{(\alpha }\},Q_{\beta )}]=0`$
with $`A_{(\alpha }B_{\beta )}=A_\alpha B_\beta +A_\beta B_\alpha `$.
The algebra discussed above is summarized as follows:
$`\begin{array}{cccc}\{๐ฌ,๐ฌ\}=\{๐ฌ,๐ฎ\}=0\hfill & ,& \{๐ฎ,๐ฎ\}=2C\mathrm{\Xi }_G\hfill & ,\\ [\mathrm{\Sigma }_m^I,๐ฌ]=0\hfill & ,& [\mathrm{\Sigma }_m^I,๐ฎ]=Z^I\mathrm{\Gamma }_m\hfill & ,\\ [P_m,๐ฌ]=0\hfill & ,& [P_m,๐ฎ]=0\hfill & .\end{array}`$ (2.20)
## 3 Gauge fixing and the equation of motion
We examine the equation of motion in a suitable gauge condition and will find the ground state solution to give the algebra (2.20). In addition to the static gauge, we impose the following gauge condition
$`\theta _+๐ซ_+\theta =0`$ (3.1)
by using the kappa invariance. The local superalgebra for a D-string is given by
$`\{F_{A\alpha }(\sigma _1),F_{B\beta }(\sigma _2)\}`$ $`=`$ $`2(C\mathrm{\Xi })_{A\alpha B\beta }(\sigma _1)\delta (\sigma _1\sigma _2),`$ (3.2)
$`(\mathrm{\Xi })_{A\alpha B\beta }`$ $`=`$ $`\mathrm{\Gamma }_{\alpha \beta }^m\left(\mathrm{๐}_{AB}\stackrel{~}{p}_m(\tau _q)_{AB}(\mathrm{\Pi }_1)_m\right),`$
where
$`F`$ $`=`$ $`\zeta +\overline{\theta }(\stackrel{~}{p}/\mathrm{\Pi }_1\tau _q){\displaystyle \frac{1}{2}}(\overline{\theta }\mathrm{\Gamma }\theta ^{}\overline{\theta }\mathrm{\Gamma }\tau _q+\overline{\theta }\mathrm{\Gamma }\tau _q\theta ^{}\overline{\theta }\mathrm{\Gamma }),`$
$`\stackrel{~}{p}_m`$ $`=`$ $`p_m+\overline{\theta }\mathrm{\Gamma }_m\tau _q\theta ^{},`$
$`(\mathrm{\Pi }_1)_m`$ $`=`$ $`X_m^{}\overline{\theta }\mathrm{\Gamma }_m\theta ^{}.`$ (3.3)
$`\mathrm{\Xi }`$ is a rank half and nilpotent operator so that a half of $`F`$ are first class constraints $`\stackrel{~}{F}_{A\alpha }(F\mathrm{\Xi })_{A\alpha }=0`$ generating the kappa symmetry and another half are second class. The counting of the physical degrees of freedom is following: The degrees of freedom of $`\theta ^{A\alpha }`$ and $`\zeta _{A\alpha }`$ are $`16\times 2\times 2`$. The number of the second class constraints is $`16`$. The number of the first class constraints is $`16`$, and the one for the gauge fixing conditions is $`16`$. So totally surviving degrees of freedom are $`16\times 2\times 2(16+16\times 2)=16`$, so $`8\theta `$โs and $`8\zeta `$โs are physically dynamical variables.
For a ground state in the static gauge, the right hand side of the (3.2) becomes the same form of the unbroken SUSY projection operator $`C\mathrm{\Xi }=2iM๐ซ_{}`$ of (2.13). This coincidence occurs to guarantee the universal property, $`\{Q,F\}=0`$. In order to solve BPS fermionic solutions explicitly, we restrict ourself to discuss only on the ground state in the rest of this section. It should be noticed that the kappa projection is not $`C\mathrm{\Xi }`$ but $`\mathrm{\Xi }`$. The kappa symmetry is generated by
$`{\displaystyle \stackrel{~}{F}\kappa ^{}}={\displaystyle F\mathrm{\Xi }\kappa ^{}}={\displaystyle F๐ซ_+(2M\mathrm{\Gamma }^0\kappa ^{})}={\displaystyle F๐ซ_+\kappa _+},`$ (3.4)
therefore the kappa parameter is projected along the unbroken SUSY direction. The gauge condition (3.1) can be possible by using the kappa transformation with $`\delta _\kappa \theta _+=\kappa _+=\theta _+`$.
Under the global translation, $`X_m`$ is transformed as $`\delta _aX_m=a_m`$ . Analogously under the global SUSY, $`\theta _{}`$ is transformed $`\delta _ฯต\theta _{}=ฯต_{}`$, while $`\theta _+=0`$ is preserved by the kappa transformation $`(\delta _ฯต+\delta _\kappa )\theta _+=ฯต_++\kappa _+`$ with $`\kappa _+=ฯต_+`$. In this gauge $`X^m`$ is transformed under the global SUSY and the kappa transformation
$`(\delta _ฯต+\delta _\kappa )X^m`$ $`=`$ $`\overline{ฯต}_{}\mathrm{\Gamma }^m\theta _{}+\overline{ฯต}_+\mathrm{\Gamma }^m\theta _{}\overline{\kappa }_+\mathrm{\Gamma }^m\theta _{}=\overline{ฯต}_{}\mathrm{\Gamma }^m\theta _{}+2\overline{ฯต}_+\mathrm{\Gamma }^m\theta _{},`$ (3.5)
where $`ฯต_{}`$ is nothing but the broken SUSY parameter. Of course the static gauge is always recovered by the reparametrization invariance. The transverse coordinates $`X^i`$ and $`\theta _{}`$ are Nambu-Goldstone modes associated with the broken translation invariance and the broken supertranslation by a brane .
The action for a D-$`p`$-brane is
$`I`$ $`=`$ $`{\displaystyle d^{p+1}\sigma L}={\displaystyle d^{p+1}\sigma \left[T\sqrt{det(G_{\mu \nu }+_{\mu \nu })}+_{WZ}\right]}`$ (3.6)
$`G_{\mu \nu }`$ $`=`$ $`\mathrm{\Pi }_\mu ^m\mathrm{\Pi }_{\nu m},\mathrm{\Pi }_\mu ^m=_\mu X^m\overline{\theta }\mathrm{\Gamma }^m_\mu \theta ,_{\mu \nu }=F_{\mu \nu }\mathrm{\Omega }_{[\mu \nu ]}^3`$
$`\mathrm{\Omega }_{\mu \nu }^I`$ $`=`$ $`(\overline{\theta }\mathrm{\Gamma }^m\tau ^I_\mu \theta )(\mathrm{\Pi }_{\nu m}+{\displaystyle \frac{1}{2}}\overline{\theta }\mathrm{\Gamma }_m_\nu \theta )(I=1,3).`$
In the fermionic gauge $`๐ซ_+\theta =\theta _+=0`$ and the static gauge $`X^\mu =\sigma ^\mu `$, $`\mu =0,1,\mathrm{},p`$, it becomes
$`G_{\mu \nu }`$ $`=`$ $`\eta _{\mu \nu }\overline{\theta }_{}\mathrm{\Gamma }_{(\mu }_{\nu )}\theta _{}+(\overline{\theta }_{}\mathrm{\Gamma }^\rho _\mu \theta _{})(\overline{\theta }_{}\mathrm{\Gamma }_\rho _\nu \theta _{})+_\mu \varphi ^i_\nu \varphi ^i`$ (3.7)
$`q^I\mathrm{\Omega }_{\mu \nu }^I`$ $`=`$ $`\overline{\theta }_{}\mathrm{\Gamma }_\nu \tau _q_\mu \theta _{}+{\displaystyle \frac{1}{2}}(\overline{\theta }_{}\mathrm{\Gamma }^\rho \tau _q_\mu \theta _{})(\overline{\theta }_{}\mathrm{\Gamma }_\rho _\nu \theta _{})`$
where $`\varphi ^i`$, $`i=p+1,\mathrm{},9`$ are perpendicular modes to a $`p`$-brane. It is used that $`[๐ซ_\pm ,C\mathrm{\Gamma }^\mu ]=0`$, $`๐ซ_\pm (C\mathrm{\Gamma }^i)=(C\mathrm{\Gamma }^i)๐ซ_{}`$ and $`[๐ซ_\pm ,\tau _q]=0`$. The expression of the Dirac-Born-Infeld part is available for all $`p`$. In this section we take D1 case as an example where the Wess-Zumino part is given by
$`_{WZ}|_{p=1}`$ $`=`$ $`Tฯต^{\mu \nu }\mathrm{\Omega }_{\mu \nu }^1,`$ (3.8)
and we will discuss general D-$`p`$-brane cases in the last section. Under an arbitrary variation of $`\delta \theta _{}`$ it becomes
$`\delta L`$ $`=`$ $`{\displaystyle \frac{TG}{2\sqrt{det(G+)}}}G^{\mu \nu }\delta G_{\mu \nu }+q^Iฯต^{\mu \nu }\delta \mathrm{\Omega }_{\mu \nu }^I`$ (3.9)
$`\delta G_{\mu \nu }`$ $`=`$ $`(\delta \overline{\theta }_{}\mathrm{\Gamma }_{(\mu }_{\nu )}\theta _{}_{(\mu }\delta \overline{\theta }_{}\mathrm{\Gamma }_{\nu )}\theta _{})`$
$`+2(\delta \overline{\theta }_{}\mathrm{\Gamma }^\rho _\mu \theta _{})(\overline{\theta }_{}\mathrm{\Gamma }_\rho _\nu \theta _{})2(_\mu \delta \overline{\theta }_{}\mathrm{\Gamma }^\rho \theta _{})(\overline{\theta }_{}\mathrm{\Gamma }_\rho _\nu \theta _{})`$
$`q^I\delta \mathrm{\Omega }_{\mu \nu }^I`$ $`=`$ $`(\delta \overline{\theta }_{}\mathrm{\Gamma }_\nu \tau _q_\mu \theta _{}_\mu \delta \overline{\theta }_{}\mathrm{\Gamma }_\nu \tau _q\theta _{})`$
$`+{\displaystyle \frac{1}{2}}\{(\delta \overline{\theta }_{}\mathrm{\Gamma }^\rho \tau _q_\mu \theta _{})(\overline{\theta }_{}\mathrm{\Gamma }_\rho _\nu \theta _{})(_\mu \delta \overline{\theta }_{}\mathrm{\Gamma }^\rho \tau _q\theta _{})(\overline{\theta }_{}\mathrm{\Gamma }_\rho _\nu \theta _{})`$
$`(\delta \overline{\theta }_{}\mathrm{\Gamma }^\rho _\mu \theta _{})(\overline{\theta }_{}\mathrm{\Gamma }_\rho \tau _q_\nu \theta _{})+(_\mu \delta \overline{\theta }_{}\mathrm{\Gamma }^\rho \theta _{})(\overline{\theta }_{}\mathrm{\Gamma }_\rho \tau _q_\nu \theta _{})\}.`$
For a ground state where $`G_{\mu \nu }`$ and $`_{\mu \nu }`$ are constant and terms in $`O(\theta ^3)`$ vanishes, the equation of motion $`\delta L/\delta \overline{\theta }_{}=0`$ becomes quite simple
$`\mathrm{\Gamma }_\mu (\eta ^{\mu \nu }\widehat{\tau }_qฯต^{\mu \nu })_\nu \theta _{}=0`$ (3.10)
where $`q^I`$ are given by
$`\{\begin{array}{ccc}q^{NS}\hfill & =& q^3=E^1=T\frac{_{01}}{\sqrt{det(G+)}}\hfill \\ q^R\hfill & =& q^1=T\hfill \end{array}`$ (3.13)
and we use $`(q^I)^2=T^2G/det(G+)`$. If we take the following ground state ansatz
$`\theta _{}=\sigma \mathrm{\Theta }`$ (3.14)
with a constant spinor $`\mathrm{\Theta }`$, then the equation of motion (3.10) reduces into
$`(\mathrm{\Gamma }_1\widehat{\tau }_q\mathrm{\Gamma }_0)_1\theta _{}`$ $`=`$ $`\mathrm{\Gamma }_1(1+\widehat{\tau }_q\mathrm{\Gamma }_{01})\mathrm{\Theta }=0.`$ (3.15)
Then there exists nontrivial ground state solution of $`\theta _{}`$ as the form of (3.14) with
$`๐ซ_+\mathrm{\Theta }=0.`$ (3.16)
The boundary term of the equation motion also allows the solution (3.14)
$`\mathrm{\Gamma }_\mu (\eta ^{\mu \nu }\widehat{\tau }_qฯต^{\mu \nu })\theta _{}|_{\mathrm{boundary}}=0.`$ (3.17)
As a result there can be a fermionic brane charge for this solution
$`Z_\alpha ^I=q^I{\displaystyle _{\sigma _I}^{\sigma _F}}๐\sigma \overline{\theta }_\alpha ^{}=q^IL\overline{\mathrm{\Theta }}_\alpha |_L\mathrm{}.`$ (3.18)
It is natural to set the volume of $`Z`$ to be infinity, since it is obtained from infinity volume brane charge $`\mathrm{\Sigma }`$ by the supertransformation.
The choice $`\mathrm{\Theta }=0`$ or $`\theta _{}=`$const. brings back to the usual Super-Poincar$`\stackrel{ยด}{\mathrm{e}}`$ algebra which is usually considered as a ground state solution.
## 4 Supergravity coupling
A D-string is described by the NS/NS and R/R worldvolume currents:
$`(J^I)^{nl}(x)`$ $`=`$ $`q^I{\displaystyle ๐\tau ๐\sigma _\tau X^{[n}_\sigma X^{l]}\delta ^{(10)}(xX(\tau ,\sigma ))}.`$ (4.1)
In this section we take a โnonrelativistic gaugeโ $`X^0=\tau `$ ($`_\tau X^0=1`$) rather than the static gauge, in order to keep a spacial rotational symmetry. The static property, equally nonrelativistic property, of D-branes is assumed. For these objects the spacial covariance will be a suitable guiding principle to determine the background coupling. To preserve this gauge, additional restriction is imposed on $`\theta _{}`$:
$`\delta _ฯต_{}(_\tau X^01)=\overline{ฯต}_{}\mathrm{\Gamma }^0_\tau \theta _{}=0_\tau \theta _{}=0.`$ (4.2)
The worldvolume currents (4.1) are related to the brane charge as
$`{\displaystyle d^9x(J^I)^{0m}(x)}`$ $`=`$ $`{\displaystyle d^9xq^I๐\tau ๐\sigma _\tau X^0_\sigma X^m\delta ^{(9)}(xX)\delta (x^0\tau )}`$ (4.3)
$`=`$ $`q^I{\displaystyle ๐\sigma _\sigma X^m}`$
$`=`$ $`(\mathrm{\Sigma }^I)^m(\tau )`$
in the nonrelativistic gauge.
The currents (4.1) are sources of two form gauge fields $`B_{mn}^I`$. In a flat background if we put a test D-brane, field equations for $`B_{mn}^I`$ become
$`^l(H^I)_{lmn}(x)`$ $`=`$ $`(J^I)_{mn}(x)`$ (4.4)
where $`H_{lmn}^I`$ are fluctuations.
Now let us consider the broken supersymmetry transformation of (4.4). Because the brane charge $`(\mathrm{\Sigma }^I)^m`$ transforms into the fermionic brane charge $`(Z^I)^\alpha `$ under the broken supersymmetry and because of (4.3), a bosonic brane current $`(J^I)^{mn}`$ also transforms into a fermionic brane current $`(๐ฅ^I)^{m\alpha }`$ :
$`\delta _ฯต_{}(J^I)^{mn}(x)`$ $`=`$ $`q^I{\displaystyle ๐\tau ๐\sigma \left(_\tau \delta _ฯต_{}X^{[m}_\sigma X^{n]}+_\tau X^{[m}_\sigma \delta _ฯต_{}X^{n]}\right)\delta ^{(10)}(xX)}`$ (4.5)
$`=`$ $`q^I{\displaystyle ๐\tau ๐\sigma _\tau X^{[m}_\sigma \overline{\theta }_{}\mathrm{\Gamma }^{n]}ฯต_{}\delta ^{(10)}(xX)}`$
$``$ $`(๐ฅ^I)^{[m|\alpha |}(x)(\mathrm{\Gamma }^{n]}ฯต_{})_\alpha .`$
It is noted that under the broken supersymmetry brane coordinates transform into $`\delta _ฯต_{}X^m=\overline{ฯต}_{}\mathrm{\Gamma }^m\theta _{}=\overline{\theta }_{}\mathrm{\Gamma }^mฯต_{}`$. The fermionic currents $`๐ฅ^{m\alpha }`$
$`(๐ฅ^I)^{n\alpha }(x)=q^I{\displaystyle ๐\tau ๐\sigma _\tau X^n_\sigma \overline{\theta }_{}^\alpha \delta ^{(10)}(xX(\tau ,\sigma ))}`$ (4.6)
are also conserved currents
$`_n(๐ฅ^I)^{n\alpha }(x)`$ $`=`$ $`q^I{\displaystyle ๐\tau ๐\sigma _\tau X^n_\sigma \overline{\theta }_{}^\alpha _n\delta ^{(10)}(xX(\tau ,\sigma ))}`$ (4.7)
$`=`$ $`q^I{\displaystyle ๐\tau ๐\sigma _\sigma \overline{\theta }_{}^\alpha (\frac{X^n}{\tau }\frac{}{X^n})\delta ^{(10)}(xX(\tau ,\sigma ))}`$
$`=`$ $`q^I{\displaystyle ๐\tau ๐\sigma _\tau _\sigma \overline{\theta }_{}^\alpha \delta ^{(10)}(xX(\tau ,\sigma ))}=0,`$
and are related with the fermionic charges $`(Z^I)^\alpha `$ as
$`{\displaystyle d^9x(๐ฅ^I)^{0\alpha }(x)}`$ $`=`$ $`q^I{\displaystyle ๐\sigma _\sigma \overline{\theta }_{}^\alpha }`$ (4.8)
$`=`$ $`(Z^I)^\alpha (\tau ).`$
If the right hand side of the equation (4.4) transforms under the broken supersymmetry of the brane, the left hand side should also transform. In another word the total supersymmetry charge is sum of the supergravity part and the brane part. Transformation rules of supergravity fields are given in , and
$`\delta _ฯต(B^I)_{mn}=(2\overline{\psi }_{[m}\mathrm{\Gamma }_{n]}\overline{\lambda }\mathrm{\Gamma }_{mn})\tau ^Iฯต,`$ (4.9)
where $`\psi `$ is a gravitino and $`\lambda `$ is a dilatino. In the first order of this test brane perturbation the background metric is not affected, since the energy momentum tensor which is a source of the metric contains square of $`H`$ as the fluctuation. So gravitinos must be set to zero $`\psi _m=0`$ because $`\delta _ฯตe_m^a\overline{\psi }_m\mathrm{\Gamma }^aฯต=0`$. In order to have nontrivial transformation rule of $`B_{mn}^I`$ with the broken SUSY parameter $`ฯต_{}`$, we will consider nonzero $`\lambda _+`$ so that $`\delta B_{01}=\overline{\lambda }_+\mathrm{\Gamma }_{01}\tau ฯต_{}0`$ as their fluctuations.
Under the broken global SUSY a dilatino does not transform $`\delta _ฯต\lambda (\varphi ,H)\mathrm{\Gamma }ฯต=0`$ in a flat background $`(\varphi =0,H=0)`$. Then the left hand side of (4.4) transforms into
$`\delta _ฯต_{}^l(H^I)_{lmn}`$ $`=`$ $`\mathrm{}\delta _ฯต_{}(B^I)_{mn}`$ (4.10)
$`=`$ $`{\displaystyle \frac{1}{2}}(\mathrm{}\overline{\lambda }_+\mathrm{\Gamma }_{[m}\tau ^I)\mathrm{\Gamma }_{n]}ฯต_{}`$
in the Lorentz gauge $`^mB_{mn}=0`$. As a result of (4.5) and (4.10), under the broken SUSY (4.4) transforms into
$`{\displaystyle \frac{1}{2}}\left(\mathrm{}\overline{\lambda }_+\tau ^I\mathrm{\Gamma }^n\right)^\alpha (x)=(๐ฅ^I)^{n\alpha }(x).`$ (4.11)
Therefore this new fermionic brane charge becomes a source of dilatino fields.
It is also confirmed that another current does not follow. Even if one more broken supersymmetry is performed on the fermionic brane current, it vanishes by (4.2)
$`\delta _ฯต_{}(๐ฅ^I)^{n\alpha }=q^I{\displaystyle _\tau \overline{\theta }_{}^\beta _\sigma \overline{\theta }_{}^\alpha \delta ^{(10)}(xX)(\mathrm{\Gamma }^nฯต_{})_\beta }=0.`$ (4.12)
This concludes that a D-string action allows the fermionic brane currents coupled with some modes of dilatino fields as well as the usual brane current coupling $`B_{mn}J^{mn}`$, when a D-string has a nontrivial fermionic boundary condition.
## 5 Discussions
Generalization to arbitrary D-$`p`$-branes is straightforward. The global supersymmetry and the local supersymmetry algebras for D-$`p`$-branes are given in , as the same form as (2.1) and (3.2). The projection (2.13) has the form of
$`๐ซ_\pm ={\displaystyle \frac{1}{2}}(\mathrm{๐}\pm \widehat{\mathrm{\Gamma }}),`$ (5.1)
and $`\widehat{\mathrm{\Gamma }}`$ is replaced by suitable $`p`$-brane projection $`\mathrm{\Gamma }_{01\mathrm{}p}\widehat{\tau }_p`$. For example $`\widehat{\mathrm{\Gamma }}=\mathrm{\Gamma }_0\mathrm{\Gamma }_{11}`$ for D0, $`\widehat{\mathrm{\Gamma }}=\mathrm{\Gamma }_0\mathrm{\Gamma }_1\mathrm{\Gamma }_2`$ for D2 and so on. By using with these projection operators, global supersymmetry charges are separated into broken supersymmetries and unbroken supersymmetries. Breaking 16 supersymmetry, $`๐ฎ|00`$, gives rise to 16 Nambu-Goldstone fermions $`\theta _{}=๐ซ_{}\theta `$. Half of this, i.e. 8 $`\theta _{}`$โs are dynamical and the rest are non-dynamical by the equations of motion in the excited states. Another set of 16 fermionic coordinates are gauged away $`\theta _+=๐ซ_+\theta =0`$ by using the kappa symmetry generated by the first class constraints $`\stackrel{~}{F}=0`$.
Equations of motions can be analyzed analogously. In the same static gauge fixing condition $`๐ซ_+\theta =0`$ of (3.1) with (5.1) and $`X^\mu =\sigma ^\mu `$, equations of motions are given as
$`\mathrm{\Gamma }_\mu \left(\eta ^{\mu \nu }+ฯต^{\mu \rho _1\mathrm{}\rho _{p1}\nu }\mathrm{\Gamma }_{\rho _1\mathrm{}\rho _{p1}}(\widehat{\tau }_๐ฉ)/(p1)!\right)_\nu \theta _{}=0,`$ (5.2)
and the nontrivial ground state solution of (3.14) can be extended as
$`\theta _{}={\displaystyle \underset{l}{}}\sigma _l\mathrm{\Theta }_l(l=1,\mathrm{},p)\mathrm{with}๐ซ_+\mathrm{\Theta }_l=0.`$ (5.3)
For D-$`p`$-brane cases, brane charges are $`p`$-rank tensors. Each commutator with the broken supersymmetry charges replaces a vector index of the brane charges by a spinor index. When $`\theta _{}`$ has nontrivial solution of (5.3), totally $`p+1`$ kinds of brane charges with vector indices or spinor indices have nonzero values. For example a D2 brane carries totally $`3`$ kinds of brane charges, among which $`2`$ kinds of bosonic brane charges $`\mathrm{\Sigma }_{mn}`$ and $`Z_{\alpha \beta }`$ are obtained in the static gauge as
$`\mathrm{\Sigma }_{12}=TL^2,Z_{\alpha \beta }=TL^2\overline{\mathrm{\Theta }}_{1(\alpha }\overline{\mathrm{\Theta }}_{2\beta )},`$ (5.4)
and $`1`$ kind of fermionic brane charges $`Z_{m\alpha }`$ are obtained as
$`Z_{1\alpha }=TL^2\overline{\mathrm{\Theta }}_{2\alpha },Z_{2\alpha }=TL^2\overline{\mathrm{\Theta }}_{1\alpha }.`$ (5.5)
The supergravity coupling will follow the argument of the section 4 replacing $`B_{mn}`$ by a $`p+1`$-rank gauge field $`C_{m_1\mathrm{}m_{p+1}}`$ for a D-$`p`$-brane respectively.
It is curious that considering that the equation (4.11) is second order although usual equation of motion for a spinor field is first order. Even in a flat background this fermionic brane current will couple to some mode of dilatino. So the coupling will be with the dilaton potential mode $`(/D/\mathrm{})\lambda `$ or nonlocal. Further studies are necessary to clarify these points and to apply many other systems.
Acknowledgments
M.H. wishes to thank Joaquim Gomis, Ken-ji Hamada, Nobuyuki Ishibashi and Shunya Mizoguchi for fruitful discussions, and M.S. would like to express his gratitude to Hiroshi Kunitomo and wishes to thank the theory group of KEK for the kind hospitality. We also thank Kiyoshi Kamimura for helpful discussions, and Mitsuko Abe and Nathan Berkovits for a question โwhat is $`Z_\alpha `$?โ which is our motivation of this work. |
warning/0001/quant-ph0001065.html | ar5iv | text | # Optical Fock-state synthesizer
## I Introduction
In the last decade, much attention has been devoted to the quantum engineering of nonclassical light. Different types of nonclassical states can be now prepared, and their quantum properties can be entirely characterized by accessible measurement schemes. Among the various quantum states of the optical field, states with a definite number of photons deserve a special attention. Indeed, besides the interest for fundamental tests of Quantum Mechanics , the Fock (number) states are also relevant for many applications, as for example, to achieve the optimal capacity coding in quantum communication channels , or the Heisenberg sensitivity limit in high-precision quantum interferometry .
Different methods for the generation of Fock states have been proposed, both for traveling-wave and cavity fields. For traveling-wave fields, these methods are principally based on tailored nonlinear interactions , conditional measurements , or state engineering . The experimental realizability or effectiveness of these proposals is often challenging. On the other hand, Fock states have been generated into cavities , i.e. using the micromaser trapped states.
In this paper we address the problem of Fock state generation in a traveling mode. We suggest an optical device based on an avalanche triggering photodetector and a ring cavity coupled to an external traveling wave through a cross-Kerr medium. Remarkably, the input states of our proposal are just customary coherent states. Our scheme differs from conventional setups involving conditional photon counters , since we use simple onโoff detection of an intense coherent field, allowing very low quantum efficiency at the photodetector. Moreover, as we will see, the scheme can also be used to engineer superpositions of few Fock states, which are a crucial resource for optical quantum computers , and quantum tomography of optical Hamiltonians .
The main advantages of the present scheme are its tunability in preparing any chosen number state and selected superpositions, along with its robustness against the imperfections of the triggering photodetector. Indeed, low quantum efficiency at the photodetector has no detrimental effect on the filtering process, but only reduces the state-synthesizing rate.
The paper is structured as follows. In the next Section the scheme is introduced, and its use in preparing Fock states is described. In Section III we show how the same scheme can be used to prepare superpositions of few Fock states, whereas in Section IV the effects of non-unit quantum efficiency at the photodetector are taken into account. Section V closes the paper by summarizing results, and discussing the feasibility of the scheme.
## II Synthesis of Fock states
The schematic setup of the synthesizer is depicted in Fig. 1. The ring cavity is build by two mirrors and two high reflectivity beam splitters. Here, for simplicity, we suppose that the beam splitters have the same transmissivity $`\tau `$. The cavity is fed by a coherent state in the mode $`a_1`$, whereas the mode $`a_2`$ is left unexcited. Through the cross-Kerr interaction, the cavity mode $`d`$ is coupled to an external traveling mode $`c_1`$, according to the unitary evolution
$`\widehat{U}_K=\mathrm{exp}(i\chi td^{}dc_1^{}c_1),`$ (1)
$`\chi `$ being the nonlinear susceptibility of the medium and $`t`$ the interaction time. The signal mode $`c_1`$ is prepared in a coherent state, and, additionally, a tunable phase shift $`\psi `$ is introduced in the cavity mode.
The scheme is suitable for applications with both continuous waves and pulses. In a situation involving pulses, however, in order have states with a definite number of photons we need a long enough quantization time. Indeed, the coherence time of the signal mode should be quite long compared to the probe one. We assume that the coherence time of the input signal is of the order of the photon flight time in the cavity, thus assuring that the cavity mode effectively couples with the signal through the cross-Kerr medium. Moreover, the coherence time of the cavity mode should be shorter than the signal one, thus allowing the Kerr phase to cumulate in the loop. In summary, the coherence time of the probe mode should be shorter than the signal one, which, in turn, is determined by the photon flight time in the cavity.
At the output of the cavity the field is monitored by an avalanche photodetector. For the purpose of our scheme, we only need to verify the presence or absence of the field, at the output port of the cavity through the triggering photodiode D. Let us initially assume unit quantum efficiency at photodetection. The measurement is described by the two-value probability operator measure (POM) $`\widehat{\mathrm{\Pi }}_n`$
$`\widehat{\mathrm{\Pi }}_0|00|,\widehat{\mathrm{\Pi }}_1\widehat{I}_{b_2}|00|,`$ (2)
where $`|0`$ is the vacuum and $`\widehat{I}_{b_2}`$ is the identity for mode $`b_2`$. As we will show in the following, due to the very steep dependence of the cavity transmissivity on the total phase shiftโincluding both cross Kerr interaction and phase $`\psi `$โthe detection of the field at photodetector D guarantees that the free mode $`c_2`$ at the output of the Kerr medium is reduced into a Fock state or a superposition of few Fock states.
The mode transformations of the ring cavity are
$`\{\begin{array}{c}b_1=\kappa (\phi )a_1+e^{i\phi }\sigma (\phi )a_2\hfill \\ b_2=\sigma (\phi )a_1+\kappa (\phi )a_2\hfill \end{array},`$ (5)
where the phase-dependent cavity transmissivity $`\sigma `$ and reflectivity $`\kappa `$ are given by
$`\kappa (\phi )`$ $``$ $`{\displaystyle \frac{\sqrt[]{1\tau }(e^{i\phi }1)}{1e^{i\phi }(1\tau )}}`$ (6)
$`\sigma (\phi )`$ $``$ $`{\displaystyle \frac{\tau }{1e^{i\phi }(1\tau )}}`$ (7)
with $`\left|\kappa (\phi )\right|^2+\left|\sigma (\phi )\right|^2=1`$. The transformations (5) and (7) are rigorously obtained by quantizing the e.m. field modes which solve the Helmholtz equation of the etalon, as in Ref. , and taking the input/output modes of the asymptotic free plane waves. However, a naive solution of the etalon as a loop of beam splitters gives the same result, with the internal modes having a reduced commutator (this point is well explained in Ref. ). For $`c_1`$ in the Fock state $`|n`$, the total phase shift is given by
$`\phi =\psi \chi nt\phi _n.`$ (8)
To simplify the notation, we write $`\sigma _n\sigma (\phi _n)`$ and analogously for $`\kappa `$. Let us now consider the input state
$`\widehat{\varrho }_{in}=|\alpha \alpha ||00|\widehat{\nu }_{in},`$ (9)
namely a generic state $`\widehat{\nu }_{in}`$ for mode $`c_1`$, a coherent state $`|\alpha `$ for mode $`a_1`$, and vacuum for $`a_2`$. In the Schrรถdinger picture the output state can be written in the form
$`\widehat{\varrho }_{out}={\displaystyle \underset{n,m=0}{\overset{\mathrm{}}{}}}\nu _{nm}|\kappa _n\alpha \kappa _m\alpha ||\sigma _n\alpha \sigma _m\alpha ||nm|.`$ (10)
The process of filtering the desired Fock state from the input state $`\widehat{\nu }_{in}`$ is triggered by the photodetector D as follows. The probabilities corresponding to the outcomes $`1`$ (detector D on) and $`0`$ (detector D off) are given by
$`P_1=\text{Tr}\left[\widehat{\varrho }_{out}\widehat{\mathrm{\Pi }}_1\right]={\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}\nu _{nn}\left(1e^{|\alpha |^2|\sigma _n|^2}\right),`$ (11)
and $`P_0=1P_1`$. By means of Eq. (7) we have
$`|\sigma _n|^2=\left(1+4{\displaystyle \frac{1\tau }{\tau ^2}}\mathrm{sin}^2{\displaystyle \frac{\psi \chi nt}{2}}\right)^1,`$ (12)
which is a periodic function sharply peaked at
$`n={\displaystyle \frac{\psi +2\pi j}{\chi t}}n^{}+{\displaystyle \frac{2\pi }{\chi t}}j,j,`$ (13)
with unit maximum height and width of the same order of the beam splitter transmissivity $`\tau `$ (typically $`\tau 10^4รท10^6`$). The value $`n^{}`$ in Eq. (13) can be adjusted to an arbitrary integer by tuning the phase-shift $`\psi `$ as a multiple of $`\chi t`$, whereas multiple resonances are avoided by using small nonlinearity $`\chi t1`$, so that the values of $`n`$ satisfying Eq. (13) for $`j0`$ correspond to vanishing matrix elements $`\nu _{ni}0i`$. In this case for high-quality cavities $`\tau \chi t`$ in Eq. (11) we have $`|\sigma _n|^2\delta _{nn^{}}`$, and the detection probability $`P_1`$ rewrites $`P_1\nu _{n^{}n^{}}\left(1e^{|\alpha |^2}\right)`$. Notice that increasing the amplitude of $`\alpha `$ will enhance the detection probability $`P_1`$. Moreover, one can optimize also the input state $`\widehat{\nu }_{in}`$ to achieve the highest $`\nu _{n^{}n^{}}`$. For example, in the case of a coherent input $`\widehat{\nu }_{in}=|\beta \beta |`$, one could select $`|\beta |\sqrt{n^{}}`$.
We now evaluate the conditional state $`\widehat{\nu }_{out}`$ at the output of the Kerr medium for detector D on. One has
$`\widehat{\nu }_{out}={\displaystyle \frac{1}{P_1}}\text{Tr}_{a_1a_2}\left[\widehat{\varrho }_{out}\widehat{\mathrm{\Pi }}_1\right]={\displaystyle \frac{e^{|\alpha |^2}}{P_1}}{\displaystyle \underset{n,m=0}{\overset{\mathrm{}}{}}}\nu _{nm}e^{|\alpha |^2\kappa _n\kappa _m^{}}\left(e^{|\alpha |^2\sigma _n\sigma _m^{}}1\right)|nm|,`$ (14)
where the partial trace is performed over the ring cavity modes. The argument $`\theta `$ of $`\sigma (\phi )=|\sigma (\phi )|\mathrm{exp}[i\theta (\phi )]`$ is given by $`\theta (\phi )=\mathrm{arctan}\left[\frac{(1\tau )\mathrm{sin}\phi }{1(1\tau )\mathrm{cos}\phi }\right]`$. For $`\tau 1`$, as already noticed, $`|\sigma _n|`$ is nonzero only for $`n=n^{}`$, and correspondingly we have $`\theta (\phi _n)=0`$. Therefore, for all practical purposes we can write $`\sigma _n\sigma _m^{}|\sigma _n||\sigma _m^{}|\delta _{nn^{}}\delta _{mn^{}}`$, and the output state (14) becomes
$`\widehat{\nu }_{out}|n^{}n^{}|,\tau \chi t,`$ (15)
i.e . the Fock component $`|n^{}`$ has been filtered from the initial state $`\widehat{\nu }_{in}`$.
In Fig. 2 we report the number distribution of the conditional output state $`\widehat{\nu }_{out}`$, with $`\psi `$ tuned to obtain $`|n^{}=4`$ for different values of the beam splitter transmissivity.
## III Synthesis of superpositions
We now show how the same setup may be used to produce superpositions of Fock states. By choosing higher nonlinearities, the quantity $`2\pi /(\chi t)`$ decreases and $`|\sigma _n|^2`$ can be significantly different from zero for more than one value of $`n`$ corresponding to sizeable components of the input state $`\widehat{\nu }_{in}`$. If there are only two of these โresonantโ values $`n_1=n^{}`$ and $`n_2=n^{}+2\pi /\chi t`$, we have $`|\sigma _n|\delta _{nn_1}+\delta _{nn_2}`$ and $`P_1`$ now reads
$`P_1(\nu _{n_1n_1}+\nu _{n_2n_2})\left(1e^{|\alpha |^2}\right),\tau \chi t.`$ (16)
Accordingly, the conditional state after a successful photodetection becomes
$`\widehat{\nu }_{out}{\displaystyle \frac{1}{\nu _{n_1n_1}+\nu _{n_2n_2}}}[\nu _{n_1n_1}|n_1n_1|+\nu _{n_2n_2}|n_2n_2|`$ (17)
$`+`$ $`\nu _{n_1n_2}|n_1n_2|+\nu _{n_2n_1}|n_2n_1|]\tau \chi t,`$ (18)
which is a pure state if and only if $`\nu _{n_1n_1}\nu _{n_2n_2}=\nu _{n_1n_2}\nu _{n_2n_1}`$, namely for $`\widehat{\nu }_{in}`$ in a pure state. In Fig. 3 we report the density matrix of the conditional output state $`\widehat{\nu }_{out}`$, with $`\psi `$ tuned to obtain a superposition of the Fock states $`|n_110`$ and $`|n_220`$ for different values of the beam splitter transmissivity. It is worth noting that the coefficients of the superposition in Eq. (18) are selected by the input state $`\widehat{\nu }_{in}`$. Therefore, in order to have a superposition with equal weights starting from a coherent state $`|\beta `$, it is sufficient to choose its amplitude in such a way that $`|\beta |^2=\left(n_1!/n_2!\right)^{1/(n_1n_2)}`$. Notice, however, that with this choice of $`\beta `$ we find a small contribution due to the term with $`j=2`$ in Eq. (13).
## IV Effects of imperfect photodetection
Let us now take into account the quantum efficiency $`\eta `$ at the photodetector. In this case the POM $`\widehat{\mathrm{\Pi }}_n`$ is replaced with $`\widehat{\mathrm{\Pi }}_n^{(\eta )}`$, where
$`\widehat{\mathrm{\Pi }}_0^{(\eta )}{\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}(1\eta )^k|kk|,\widehat{\mathrm{\Pi }}_1^{(\eta )}I_{b_2}\widehat{\mathrm{\Pi }}_0^{(\eta )}.`$ (19)
The probability $`P_1^{(\eta )}`$ of having outcome $`1`$ at the photodetector and the conditional output state now are the following
$`P_1^{(\eta )}=\text{Tr}[\widehat{\mathrm{\Pi }}_1^{(\eta )}\widehat{\varrho }_{out}]={\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}\nu _{nn}\left(1e^{\eta |\alpha |^2|\sigma _n|^2}\right),`$ (20)
$`\widehat{\nu }_{out}^{(\eta )}={\displaystyle \frac{e^{|\alpha |^2}}{P_1^{(\eta )}}}{\displaystyle \underset{n,m=0}{\overset{\mathrm{}}{}}}\nu _{nm}e^{|\alpha |^2[\kappa _n\kappa _m^{}+\sigma _n\sigma _m^{}]}\left(1e^{\eta |\alpha |^2\sigma _n\sigma _m^{}}\right)|nm|.`$ (21)
Remarkably, low values for the quantum efficiency $`\eta `$ can reinforce the process of filtering, though at the cost of lowering the probability $`P_1^{(\eta )}`$ of photodetection. In fact, $`\eta `$ scales the term $`\sigma _p\sigma _q^{}`$ in the exponential in Eq. (21), thus lowering the off-resonance contributions. In Fig. 4a we actually purify the superposition shown in Fig. 3a by lowering the quantum efficiency to the value $`\eta =20\%`$. The probability of obtaining the state is correspondingly lowered from the value $`P_1=0.205`$ to $`P_1^{(\eta )}=0.116`$. An analogous argument holds for the dependence of the output state $`\widehat{\rho }_{out}`$ on the input coherent amplitude $`\alpha `$. In Fig. 4b we report the conditional state obtained by choosing $`\alpha =3.58`$, which corresponds to the same detection probability $`P_1^{(\alpha )}=0.116`$. Obviously the above discussion on the effect of non-unit quantum efficiency and on the intensity of the input coherent state $`|\alpha `$ holds also for the generation of single Fock states.
## V Conclusions and outlook
The present proposal is characterized by two relevant features: its tunability in the preparation of any chosen number state and of selected superpositions, and its robustness against the imperfections of the photodetection process. Therefore, it is a matter of interest to discuss its feasibility in practical applications. This mostly depends on two parameters: the quality factor of the ring cavity (governed by the beam splitters transmissivity $`\tau `$) and the value of the nonlinear coupling $`\chi t`$. For a good cavity (i.e. a cavity with large quality factor) $`\tau `$ should be quite small, usual values achievable in quantum optical labs being of the order of $`\tau 10^4`$$`10^6`$. Remarkably, losses due to absorption processes are of the order of $`10^7`$, at least one order of magnitude smaller than $`\tau `$. The scheme also requires an appreciable value for the nonlinear cross-Kerr coupling between the signal and the cavity modes. The coupling can be realized by optical nonlinear couplers, either codirectional or contradirectional, in the case of an all-optical fiber implementation, or by direct matching of modes on a third-order crystal in the case of a free field. Usually, third-order nonlinearities are small, and may be masked by the concurrent self-modulation and absorption processes. Although the effects of self phase modulation can be avoided using resonant $`\chi ^{(3)}`$ media , currently available nonlinearities can be used in the present scheme only for the preparation of number states . However, recent breakthroughs in nonlinear Kerr physics open new perspectives for applications in quantum optics. In fact, methods based on dark atomic resonance and electromagnetically induced transparency have been suggested to strongly enhance nonlinearity while suppressing absorption. These results (especially for atomic gases) indicate that giant Kerr nonlinear shifts of the order of 1 radiant per photon may be obtained by methods not too far from present technology .
As regards the power of the output signal (and the probe mode), we mention that this should not be too large, since the $`2\pi `$-periodic resonance structure of the cavity transmissivity relies on the linear phase-shift imposed to the cavity mode. If the power is too large the nonlinear contribution drastically alters this situation, and gives rise to bistability or period-doubling instabilities. The actual value of the allowed power strongly depends on the specific nonlinear crystal employed in the setup. We do not discuss here the technical details.
In conclusion, we have suggested a novel scheme to generate arbitrary optical photon-number eigenstates in a traveling wave mode. The scheme uses onโoff photodetection of the field mode exiting a high-Q cavity, which, in turn, is coupled to the traveling-wave by nonlinear Kerr interaction. The input fields for the setup are just customary coherent states. After a successful photodetection, the traveling mode is found in a photon-number eigenstate, or, for sufficiently high Kerr nonlinearity, in a superposition of Fock states. The photodetector is needed just to test the presence of an intense coherent output field and, in fact, we have shown that non-unit quantum efficiency at photodetection improves the quality of the state synthesis, however at the expenses of the synthesizing rate. |
warning/0001/gr-qc0001038.html | ar5iv | text | # 1 Introduction
## 1 Introduction
Ever since the discovery of Black Hole (BH) evaporation it has been evident that quantum processes involving a BH can exhibit quite unusual properties. In particular, it is not clear whether such a basic property of the $`S`$-matrix as unitarity can be preserved if BHs are present in intermediate states. At the present level of understanding of quantum general relativity it seems quite impossible to give a satisfactory description of such processes in the framework of full four-dimensional theory. Therefore, the study of simplified models is so important, among which spherically reduced Einstein gravity (SRG) is the physically most relevant one. Some important information can be collected already at the classical limit. Due to the progress in computer simulations the understanding of spherically symmetric collapse of classical matter towards a BH has reached a remarkable level . That collapse is governed by a critical threshold which even allows a simplified discussion in terms of self-similar solutions . On the other hand, it has already been conjectured that for (in $`d=2`$) minimally coupled matter (i.e. not properly restricted to its s-wave part) this critical behaviour is absent, i.e. a BH is produced by an arbitrarily small amount of matter.
One approach to obtain the quantum version of such processes is the use of Dirac quantization for suitably defined operators which describe the collapse of matter. A very popular model in this context is the study of thin spherical shells. Using the Kuchaล decomposition , especially for null (lightlike) shells important progress has been made recently . We should also mention some earlier papers where the problem of physical states in dilaton gravity without matter was addressed. Although certain aspects of the correspondence between physical states and BHs could be clarified, the application of these results to the quantum $`S`$ matrix for matter fields was not possible in the framework of the reduced phase space formalism.
From the point of view of usual quantum field theory the path integral approach seems the most natural one. There (properly defined) S-matrix elements directly determine the โphysical observablesโ, which in the Dirac approach are โ in the opinion of the present authors โ not so easily to be extracted from the โDirac observablesโ. During the last years the path integral quantization of general 2d gravity theories, including SRG coupled to matter, has shown considerable progress . Based upon the (global and local) dynamical equivalence of (torsionless) dilaton theories with first order 2d gravity in Cartan variables (with nonvanishing torsion) it turned out that the 2d geometry can be integrated much more easily, if a specific gauge, the โlight-coneโ gauge for the Cartan variables is chosen. The 2d metric corresponding to this gauge coincides with the Eddington-Finkelstein (EF) metric which has the advantage of avoiding coordinate singularities at an eventual horizon. As shown in the explicit one loop contributions in the generating functional originate from the Gaussian integral of the scalars and from the โback reactionโ due to the scalars through the covariant measure.
Our present work concentrates on the classical part of this functional, i.e. on the zero loop order (tree approximation) in terms of the scalar matter fields. Clearly here features similar to the ones in classical collapse can be expected, although those โmacroscopicโ effects could well be modified at the โmicroscopicโ level, when we refer in the latter case, say, to the scattering of individual scalar quanta in 2d gravity theory. As will be shown, these effects are indeed present, emerging naturally without any further ad hoc assumptions from the 2d quantum gravity formalism.
In order to make this evident we shortly review the main arguments of the latter in section 2, generalizing it to the case of massive scalars. The classical vertex of scalar fields is extracted in section 3. The problems arising for the scattering amplitude from our restriction to minimally coupled massless and massive scalars are discussed in section 4. In section 5 we summarize our results.
## 2 2d quantum gravity with massive scalars
### 2.1 Path integral quantization
The total action $`L=L^{(g)}+L^{(m)}`$ consists of a geometric part which is most conveniently written in first order form involving Cartan variables
$`L^{(g)}`$ $`=`$ $`{\displaystyle _{M_2}}\left[X^+De^{}+X^{}De^++Xd\omega +ฯต๐ฑ(X^+X^{},X)\right],`$ (1)
$`๐ฑ`$ $`=`$ $`X^+X^{}U(X)+V(X)`$ (2)
where $`De^a=de^a+(\omega e)^a`$ is the torsion two form, the scalar curvature $`R`$ is related to the spin connection $`\omega `$ by $`\frac{R}{2}=d\omega `$ and $`ฯต`$ denotes the volume two form $`ฯต=\frac{1}{2}\epsilon _{ab}e^ae^b=d^2xdete_\mu ^a=d^2xe`$. Our conventions are determined by $`\eta =diag(1,1)`$ and $`\epsilon ^{ab}`$ by $`\epsilon ^{01}=\epsilon ^{10}=1`$. We also have to stress that even with holonomic indices, $`\epsilon ^{\mu \nu }`$ is always understood to be the antisymmetric Levi-Civitรก symbol and never the corresponding tensor. $`L^{(g)}`$ is globally and locally equivalent to the general dilaton theory
$$L_{(dil)}=d^2x\sqrt{g}\left(X\frac{R}{2}V(X)+\frac{U(X)}{2}(X)^2\right),$$
(3)
determined by the same functions $`V`$ and $`U`$ of the dilaton field $`X`$ as in (1) and (2). In (3) $`g_{\mu \nu }`$ is the 2d metric, $`R`$ the Ricci scalar. Spherically reduced gravity (SRG) is the special case $`U_{SRG}=(2X)^1,V_{SRG}=2`$.
The matter action we write directly in terms of components of the zweibeine $`e_\mu ^a`$ in $`e_\mu ^ae_\nu ^b\eta _{ab}`$, converting the usual expression for the Lagrangian with nonminimally coupled scalar fields $`S`$ ($`(X)=X`$ for SRG)<sup>1</sup><sup>1</sup>1A further generalization, including selfcouplings is possible without difficulties. Those terms could also provide the necessary counterterms for the renormalization when quantum corrections to scalar vertices are included. Note, however, that such a self-interaction only gives rise to local contributions, while the matter-vertices derived by means of our effective theory are non-local in general (see below).
$$^{(m)}=\frac{(X)}{2}\sqrt{g}\left(g^{\mu \nu }_\mu S_\nu Sm^2S^2\right)$$
(4)
into
$$^{(m)}=\frac{(X)}{2}\left[\frac{\epsilon ^{\alpha \mu }\epsilon ^{\beta \nu }}{e}\eta _{ab}e_\mu ^ae_\nu ^b_\alpha S_\beta Sm^2S^2e\right].$$
(5)
In our paper, as in , we treat the simple case $`=1`$ of minimal coupling. This will be enough to see some of the basic features.
For the quantum theory โ as well as for the much simplified treatment of the exact classical solutions to (1) or (3) โ the use of the Eddington-Finkelstein (EF) gauge
$$e_0^+=\omega _0=0,e_0^{}=1$$
(6)
has been found to be useful. It is convenient to introduce the shorthand notation for โcoordinatesโ, โmomentaโ and related sources
$`q_i`$ $`=`$ $`(\omega _1,e_1^{},e_1^+),`$
$`p_i`$ $`=`$ $`(X,X^+,X^{}),`$ (7)
$`j_i`$ $`=`$ $`(j,j^+,j^{}),`$
$`J_i`$ $`=`$ $`(J,J^{},J^+).`$
Following the canonical steps of constructing the path integral , after integrating out the auxiliary variables and the conjugate momentum to $`S`$ for the gauge (6) the path integral reads
$$W=\sqrt{detq_3}(๐S)(๐^3q)(๐^3p)detF\mathrm{exp}i\left(\frac{_{(1)}^{\text{eff}}}{\mathrm{}}+_{(s)}\right)d^2x$$
(8)
where the effective Lagrangian, derived from (3) and (5) becomes
$$_{(1)}^{\text{eff}}=q_i\dot{p_i}+q_1p_2q_3๐ฑq_2(_0S)^2+(_0S)(_1S)q_3\frac{m^2}{2}S^2.$$
(9)
It is well known that the correct diffeomorphism invariant measure for a scalar field $`S`$ on a curved background $`e_\mu ^a`$ is $`d((g)^{1/4}S)=(d\sqrt{e}S)`$, where $`e=dete_\mu ^a`$ . Note, that $`\sqrt{g}=e=e_1^+=q_3`$ in the EF gauge is a consequence of the gauge choice (6). $`detF`$ is the determinant resulting from the integration of the auxiliary variables of the extended Hamiltonian. It depends on the differential operator
$$F=_0+p_2U(p_1)$$
(10)
The source term includes also the source $`Q`$ for the scalar field as well as sources $`J_i`$ for the momenta:
$$_{(s)}=j_iq_i+J_ip_i+SQ.$$
(11)
Eq. (9) possesses the crucial property to be linear in the โcoordinatesโ $`q_i`$. Also the factor $`\sqrt{detq_3}`$ in the measure may be either lifted into a Lagrangian type term by the integration over auxiliary ghost fields that can be integrated out later again, or by introduction of an auxiliary metric . In both cases the linearity of the effective action is preserved. Thus integrating $`d^3q`$ simply produces three $`\delta `$-functions
$`\delta \left(_0\left(p_1\widehat{B_1}\right)\right)`$ (12)
$`\delta \left(_0\left(p_2\widehat{B_2}\right)\right)`$ (13)
$`\delta \left(F\left(p_3\widehat{B_3}\right)\right).`$ (14)
where $`F`$ is defined in (10). $`\widehat{B}_i`$ are functions of the sources $`j_i`$ and matter fields and will be given below. Using these three $`\delta `$-functions the integrations over $`(d^3p)`$ yield directly
$$p_i=\widehat{B}_i.$$
(15)
This simply means that in the phase-space (path-) integral only classical paths contribute to the $`p_i`$ and the remaining continuous physical degrees of freedom are represented by the scalar field alone, since all integrations over geometric variables have been performed exactly. Note that integration over $`p_3`$ from (14) produces another factor $`(detF)^1`$ so that the total Faddeev-Popov determinant is one โ a result consistent with experience from Yang-Mills fields in temporal gauges like (6).
By $`_0=_0+i(\mu +i\epsilon )`$ we define a regularized time derivative with $`ฯต,\mu +0`$ for describing the IR and UV regularized one-dimensional associated Green function in loop integrals. Homogeneous modes always appear when we invert the operator $`_0`$. Such modes $`(_0\overline{p}_i=0)`$ must be included in $`\widehat{B}_i`$ where they completely describe the (eventual) classical background <sup>2</sup><sup>2</sup>2E.g. for SRG that background may be a BH or flat Minkowski spacetime, depending on the choice of modes $`\overline{p}_i`$. The freedom still encoded in the homogeneous modes can be reduced by fixing the residual gauge freedom of the EF gauge (6) and solving Ward identities. This procedure is described in detail in . .
$`\widehat{B}_1`$ $`=`$ $`\underset{}{\overline{p}_1+_0^1\overline{p}_2+\mathrm{}(_0^1j_1+_0^2j_2)}_0^2(_0S)^2,`$
$`:=B_1`$
$`\widehat{B}_2`$ $`=`$ $`\underset{}{\overline{p}_2+\mathrm{}_0^1j_2}_0^1(_0S)^2,`$
$`:=B_2`$
$`\widehat{B}_3`$ $`=`$ $`e^{\widehat{T}}\left[_0^1e^{\widehat{T}}(\mathrm{}j_3V(\widehat{B}_1){\displaystyle \frac{m^2}{2}}S^2)+\overline{p}_3\right]`$ (18)
$`=`$ $`\underset{}{e^T\left[_0^1e^T(\mathrm{}j_3V(B_1))+\overline{p}_3\right]}+\text{terms }๐ช(S^2).`$
$`:=B_3`$
$$\widehat{F}=e^{\widehat{T}}_0e^{\widehat{T}},\widehat{T}=_0^1(\widehat{U}\widehat{B}_2),\widehat{U}=U(\widehat{B}_1).$$
(20)
In the abbreviations an exponential representation for the operator $`F`$ is used. There is still an ambiguity in the path integral. Indeed, the term $`J_3\widehat{B}_3`$ can be formally rewritten as
$$\left(e^{\overline{T}}(\mathrm{}j_3V(\widehat{B}_1)\frac{m^2}{2}S^2)(_0)e^{\widehat{T}}J_3+J_3\overline{p}_3\right)$$
(21)
We have a freedom to add a homogeneous solution $`_0\stackrel{~}{g}=0`$ to the term $`e^{\widehat{T}}J_3`$. This amounts to adding to the effective Lagrangian the term
$$\stackrel{~}{}=\stackrel{~}{g}e^{\widehat{T}}\left(\mathrm{}j_3\widehat{V}\frac{m^2}{2}S^2\right).$$
(22)
The same procedure applied to the terms $`J_1\widehat{B}_1`$ and $`J_2\widehat{B}_2`$ just leads to trivial contributions. Clearly $`\stackrel{~}{}`$ alone survives when the sources $`J_i`$ for the momenta are switched off. Nevertheless, those sources are technically important for a simple definition of an overall conservation law $`d๐=0`$, peculiar to all $`2d`$ theories, even with interacting matter . Its geometric part $`(Q=^{p_1}U(y)๐y)`$
$$๐^{(g)}=e^{Q(p_1)}p_2p_3+^{p_1}V(u)e^{Q(u)}๐u$$
(23)
for SRG by fixing integration constants in a specific way may be defined as
$$๐_{SRG}^{(g)}=\frac{p_2p_3}{\sqrt{p_1}}4\sqrt{p_1}$$
(24)
### 2.2 Effective scalar theory
Having performed the integral $`(๐^3q)`$ and then $`(๐^3p)`$ which is possible only in the chosen gauge (6) in a straightforward way, the generating functional (8) becomes
$$W(j,J,Q)=(๐S)\mathrm{exp}id^2x\left(\frac{_{(2)}^{eff}}{\mathrm{}}+SQ\right).$$
(25)
Scalar fields can be integrated only perturbatively. Let us separate different orders of $`S`$ in the effective Lagrangian:
$$_{(2)}^{eff}=_0+SQ+_2+^{int}$$
(26)
where $`_0`$ does not contain $`S`$ and $`_2`$ is quadratic in $`S`$. All higher powers are collected in $`^{int}`$. According to (25) the quadratic part $`_2`$ describes a free minimally coupled scalar field on the effective background geometry with the zweibein expressed in terms of the external sources ($`T=\widehat{T}(S=0)`$):
$$_2=\left(_0S\right)\left(_1S\right)E_1^{}\left(_0S\right)^2\frac{m^2}{2}S^2E_1^+$$
(27)
$$E_1^+=e^T,E_1^{}=_0^2\stackrel{~}{g}e^T\left(V^{}+UV\right)$$
(28)
We use capital letters $`E_1^\pm (J,j)`$ to distinguish the effective values from the fundamental zweibein fields $`e_1^\pm `$ which are already integrated out.
The interaction Lagrangian can be represented as
$$^{int}(S)^{int}\left(\frac{1}{i}\frac{\delta }{\delta Q}\right)$$
(29)
and pulled out from the integral over $`S`$. As shown in the path integral measure for $`S`$ by a straightforward redefinition can be reduced to just the standard Gaussian one. In the generating functional
$`W`$ $`=`$ $`\mathrm{exp}\left\{{\displaystyle \frac{i}{\mathrm{}}}{\displaystyle d^2x^{int}}\right\}`$ (30)
$`\times \mathrm{exp}\left\{{\displaystyle \frac{i\mathrm{}}{2}}{\displaystyle _x}{\displaystyle _y}Q(x)G_{xy}Q(y)+{\displaystyle _x}(J_iB_i)+i\mathrm{\Gamma }^{1loop}(j,J)\right\}`$
where $`G_{xy}`$ is the scalar field propagator on the effective background (28) and $`\mathrm{\Gamma }^{1loop}`$ is the logarithm of the determinant which for $`m=0`$ may be expressed as a Polyakov action. We do not have to go into details on the one-loop contributions since the present paper deals with the tree-level diagrams which are determined by the first two terms in (30), where $`^{int}`$ is to be interpreted as in (29).
## 3 Vertices of scalar field
In our present paper of primary interest are the effective scalar vertices contained in $`^{int}`$. At vanishing sources $`J_i=0`$ for the momenta, $`^{int}`$ in (26) reduces to the corresponding contribution from $`\stackrel{~}{}`$ as defined in (22).
### 3.1 Massless scalars
Let us first consider the case $`m=0`$. Then the scalar field $`S`$ enters the interaction Lagrangian only as $`(_0S)^2`$. Moreover, according to (15) and (2.1), which, in turn, are the input in (18) and (22), it always appears in the combination $`[\mathrm{}j_2(_0S)^2]`$ . Therefore, the effective vertex of order $`2n`$ in $`^{int}`$ of (26) has the generic form<sup>3</sup><sup>3</sup>3From now on we put $`\mathrm{}=1`$ for simplicity.:
$$S^{(2n)}=d^2x_1\mathrm{}d^2x_nS^{(2n)}(x_1,\mathrm{},x_n)(_0S)_{x_1}^2\mathrm{}(_0S)_{x_n}^2$$
(31)
where
$$S^{(2n)}=\frac{(1)^n}{n!}\frac{\delta ^n}{\delta j_2^n}\stackrel{~}{}|_{j=0}=\frac{(1)^{n1}}{(n)!}\left(\frac{\delta ^{n1}}{\delta j_2^{n1}}E_1^{}\right)|_{j=0}$$
(32)
with $`E_1^{}`$ defined in (28).
To obtain the $`(n1)^{th}`$ functional derivative of $`E_1^{}`$ it is enough to take $`j_2`$ localized at $`n1`$ different points:
$$j_2(x)=\underset{k=1}{\overset{n1}{}}c_k\delta (y_kx)$$
(33)
then we can expand $`E_1^{}(j_2,x)`$ in a power series of $`c_k`$. In the resulting sum the coefficient of the term with $`_{k=1}^{n1}c_k`$ will give the desired functional derivative.
As seen from (20), (28) $`E_1^{}`$ is a nonlocal functional of the $`p_i`$ which are again nonlocal functionals of the sources $`j_i`$. Our aim is the determination of the classical vertices. Therefore, their regularizations introduced in $`_0`$ may be removed and $`_0^1`$ simply becomes an integration $`๐x_0`$. However, instead of applying this integration several times it is more convenient to solve the corresponding differential equations with suitable boundary conditions. This may seem surprising, because (32) in principle already represents the solution in closed form. But the treatment of multiple integrations is very involved with many, at first, undetermined integration ranges and integration constants, which โ as for the BH in SRG โ have singularities. Also the trick to go back to the classical equations which determine $`\stackrel{~}{}`$ (cf. ) will give us important additional information to be used for the physical interpretation of the results.
Our starting point are the three differential equations for $`p_i`$ which follow from solving the $`\delta `$-functions (12)-(14) (cf. eqs. (39-41) of ) in the presence of a source term $`j_2q_2`$ in $``$ whose solution is (15), in the special case $`S=0,j_1=j_3=0`$, i.e. $`(_0a=\dot{a})`$:
$`\dot{p}_1p_2`$ $`=`$ $`0`$ (34)
$`\dot{p}_2`$ $`=`$ $`j_2`$ (35)
$`\dot{p}_3+p_2Up_3+V`$ $`=`$ $`0`$ (36)
with $`j_2`$ of (33). The quantity $`E_1^{}=q_2(j_2)`$ in the notation (7) may be calculated from the classical e.o.m.โs for $`q_i`$ or, equivalently, by suitable differentiations of (28) ($`U^{}=dU/dp_1`$ etc. ):
$`\dot{q}_1p_2p_3q_3U^{}q_3V^{}`$ $`=`$ $`0`$ (37)
$`\dot{q}_2+q_1q_3p_3U`$ $`=`$ $`0`$ (38)
$`\dot{q}_3p_2q_3U`$ $`=`$ $`0`$ (39)
From (38) and (39), eliminating $`\dot{q}_1`$ by (37), or directly differentiating twice (28) the simple differential equations for $`q_2=E_1^{}`$ may be obtained:
$$\ddot{q}_2+q_3(V^{}+UV)=0$$
(40)
where $`q_3=E_1^+`$ is already determined by the first Eq. (28).
In the following we restrict ourself to the vertex $`S^{(4)}(n=2)`$ in (31). It is depicted in Fig. 1 (the momenta and corresponding arrows we included for later reference in the basic $`S`$-matrix elements with Minkowski modes). Then in (33) only one term with $`c_k=c`$ is needed and
$$S^{(4)}(x,y)=\frac{1}{2}\frac{\delta q_2(x)}{\delta j_2(y)}.$$
(41)
In $`S^{(4)}`$ only the part of (41) contributes that is symmetric in $`xy`$. For $`E_1^{}`$ in (40) the only input from the momentum equation are the contributions to $`p_1`$ and $`p_2`$. Fixing the residual gauge transformations these solutions with (33) at $`k=1`$ can be written as $`(x^0=t,y^0=s)`$
$$p_2=1+c\left[\alpha +\mathrm{\Theta }(ts)\right]$$
(42)
$$p_1=t+c\left[\alpha +\mathrm{\Theta }(ts)\right](ts)$$
(43)
where one integration constant has been absorbed in the definition of $`t`$. An overall factor $`\delta (x^1y^1)`$ of the square brackets in (42) and (43), expressing the locality in $`x^1`$, is not written explicitly but will be taken into account in the end. The constant<sup>4</sup><sup>4</sup>4Here and in the following a โconstantโ means also functions of $`x^1`$ and $`s=y^0`$. $`\alpha `$ parametrizes different possible solutions. There are three main cases:
a) $`p_21`$ for $`t>s`$ only: $`\alpha =0`$ b) $`p_21`$ for $`t<s`$ only: $`\alpha =1`$ c) โsymmetricโ solution: $`\alpha =\frac{1}{2}`$ .
In the last case the square bracket in (42) and (43) may be replaced by half the sign function $`\epsilon (ts)`$. Although $`p_3`$ does not enter eq. (40), it is necessary to compute the effective BH mass. From the general solution of (36)
$$p_3=e^Q\left(\overline{p}_3^{p_1}๐uV(u)e^{Q(u)}\right)$$
(44)
with $`\dot{\overline{p}}_3=0`$ for each of the cases a) b) c) it is simple to find a solution with $`p_3^{ts}(t=s)=p_3^{ts}(t=s)`$. The most interesting application is SRG with $`U_{SRG}=(2p_1)^1,V_{SRG}=2`$. E.g. for case a) one finds easily with (34) for $`๐_{SRG}^{(g)}`$
$$๐^{(g)}|_{t<s}=\overline{p}_3,๐^{(g)}|_{t>s}=\overline{p}_3+c\left(\overline{p}_3+4\sqrt{s}\right)$$
(45)
Here the integration constant $`\overline{p}_3`$ must be independent of $`c`$. Thus in the term $`O(c)`$, relevant for our vertex $`S^{(4)}`$ a nonvanishing effective โBH massโ has to be present. Since $`๐^{(g)}m_{BH}`$ (cf. e.g. the last ref. ) with a โnaturalโ choice $`\overline{p}_3=0`$ for the solution without source $`j_2`$, a BH mass proportional $`(\sqrt{4}s)`$ will be switched on for $`t>s`$. As will be clarified below โ despite our suggestive notation โ $`t`$ and $`s`$ refer to a space coordinate. Our gauge choice (42) and (43) for SRG has placed the singularity at $`t=0`$ which, however, in this case would not lie in the region $`t>s`$ where $`๐^{(g)}`$ differs from zero. Thus case a) suggests the interpretation of a shell with negative mass, situated at $`t>s`$.
For case b) an analogous computation gives $`m_{ADM}+4\sqrt{s}`$ for $`t<s`$ only, i.e. a proper BH at $`t=0`$, whose effect is switched off for $`ts`$. For c) $`๐^{(g)}`$ jumps from $`2\sqrt{s}`$ at $`t<s`$ to $`+2\sqrt{s}`$ at $`t>s`$. The common feature of this apparently highly ambiguous situation (also other values of $`\alpha `$ may be taken in (42) and (43)!) is the discontinuity in the effective BH mass at $`t=s`$ which will make the appearance of the a singularity in $`S^{(4)}`$ unavoidable. We call this phenomenon โvirtual Black Holeโ (VBH). Actually the ambiguity in $`\alpha `$ disappears alltogether in the vertex $`S^{(4)}(x,y)=S^{(4)}(y,x)`$ so that the different interpretations in a), b), c) and for other values for $`\alpha `$ should not be taken at face value. It should be noted that the range of variables $`t=x^0,s=y^0`$ are not to be identified as the variables to be used in a scattering amplitude $`S+SS+S`$ connecting asymptotic Minkowski space scalar fields (see below).
We now turn to the solution of (39) and (40), using case a) for (42) and (43) in anticipation of the fact that it will be symmetrized anyhow to the only relevant contribution to $`S^{(4)}`$. On the other hand, for another vertex to appear below for massive scalars, only that case will produce a finite result.
The continuous solution to (39) for SRG is (the indices $`\left(0\right)`$ and $`\left(1\right)`$ refer to $`t<s`$ and $`t>s`$, respectively; the integration constant in $`Q`$ is fixed according to $`Q_{SRG}=T_{SRG}=\frac{1}{2}\mathrm{ln}t)`$
$`q_3^{(0)}`$ $`=`$ $`{\displaystyle \frac{\overline{q}_3}{\sqrt{t}}},`$ (46)
$`q_3^{(1)}`$ $`=`$ $`{\displaystyle \frac{\overline{q}_3}{\sqrt{t}}}\left[1c\left(1{\displaystyle \frac{s}{t}}\right)\right].`$ (47)
Introducing (46), (47) into (40) for continuous $`q_2`$ and $`\dot{q}_2`$ at $`s=t`$
$`q_2^{(0)}`$ $`=`$ $`\overline{q}_3\left[4\sqrt{t}{\displaystyle \frac{2t}{\sqrt{s}}}2\sqrt{s}+\overline{a}t+\overline{b}\right],`$ (48)
$`q_2^{(1)}`$ $`=`$ $`\overline{q}_3[{\displaystyle \frac{4}{(1+c)^2}}\sqrt{t(1+c)sc}+{\displaystyle \frac{2}{1+c}}(\sqrt{s}{\displaystyle \frac{t}{\sqrt{s}}})`$ (49)
$`{\displaystyle \frac{4\sqrt{s}}{(1+c)^2}}+\overline{a}t+\overline{b}],`$
there is still a dependence on integration constants $`\overline{q}_3,\overline{a},\overline{b}`$. It should be noted that $`\overline{q}_3`$ in a direct calculation from $`\stackrel{~}{}`$ in (21) would be replaced by the factor $`\stackrel{~}{g}(x^1)`$.
Next we will see that these constants can be fixed uniquely by natural assumptions for the effective line element, computed in the gauge (6) from $`q_2=E_1^{},q_3=E_1^+`$ (we set $`x^1=x`$):
$$(ds)^2=2q_3(dt+q_2dx)dx$$
(50)
For $`t<s`$ with (46) and (48) in case a) we require the line element to describe flat (Minkowski) space, i.e. with a new coordinate $`\overline{t}`$
$$(ds)_{(0)}^2=2d\overline{t}dx+(dx)^2=(d\tau )^2(dz)^2$$
(51)
This completely (only up to a sign in $`\overline{q}_3`$ which we chose to be positive) fixes
$`b=sa=2\sqrt{s},\overline{q}_3={\displaystyle \frac{1}{2\sqrt{2}}},`$ (52)
$`\sqrt{2}\overline{t}=\sqrt{t}`$ (53)
Otherwise a BH and an acceleration term (Rindler metric) would be present. In the last equality (51) the transition from (outgoing) EF coordinates to usual Minkowski coordinates
$$x=\tau z,\overline{t}=z$$
(54)
has been made. The relations (52) also lead to a unique result for the term of first order in $`c`$ which determines
$$\frac{\delta q_2(x)}{\delta j_2(y)}=\frac{1}{2\sqrt{2}}\frac{|\sqrt{x^0}\sqrt{y^0}|^3}{\sqrt{x^0y^0}}\delta (x^1y^1).$$
(55)
Here the symmetrization has been performed and the overall factor $`\delta (x^1y^1)`$ included. This result is proportional to the (symmetrized) vertex calculated in , where the overall constant had not been determined.
It is also essential to study the effective line element $`(ds)_{(1)}^2`$ valid in the range $`ts`$. Again we may first bring it into EF form in terms of a new coordinate $`\overline{t}`$. Joining $`\overline{t}`$ smoothly to the corresponding variable for $`ts`$ we get
$$\overline{t}_{(1)}=\frac{1}{\sqrt{2}}\left[\sqrt{t}\frac{1}{\sqrt{t}}+\sqrt{s}+c(\sqrt{s}\sqrt{t})\right]$$
(56)
which for large $`t`$ and $`\overline{t}_{(1)}`$ reduces to $`\overline{t}_{(1)}\left(\sqrt{t}(1c)+\sqrt{s}(1+c)\right)/\sqrt{2}`$. In the line element
$$(ds)_{(1)}^2=2d\overline{t}_{(1)}dx+K_{(1)}(dx)^2$$
(57)
the Killing norm $`K_{(1)}`$ in terms of $`\overline{t}_{(1)}`$ (for case a)) can be calculated easily to the required $`๐ช(c)`$ from $`K_{(1)}=2q_2^{(1)}q_3^{(1)}`$ for asymptotic values of the radial variable $`\overline{t}_{(1)}`$:
$$\underset{\overline{t}_{(1)}\mathrm{}}{lim}K=1+c\left[\frac{3\sqrt{s}}{2\sqrt{2}\overline{t}_{(1)}}+\frac{\overline{t}_{(1)}}{\sqrt{2s}}3+๐ช\left(\frac{1}{\overline{t}_{(1)}^2}\right)\right]$$
(58)
The first term in the square bracket, in agreement with (45), describes the VBH as a massive object with (negative) effective mass proportional to $`\sqrt{s}`$. Eq. (58) is valid asymptotically, but the linear dependence on $`\overline{t}_{(1)}`$ indicates that it corresponds to a uniformly accelerated coordinate system with respect to Minkowski space. The acceleration is proportional to $`s^{1/2}=(y_0)^{1/2}`$. Thus in the simultaneous limit with $`y_0\mathrm{}`$ the asymptotic scalar fields, entering an $`S`$-matrix element to be computed from (55) in a certain sense may be determined by Minkowski modes after all.
It is obvious that similar arguments for case b) will produce flat Minkowski space for $`t>s`$. At $`t<s`$ something like a genuine BH again together with linear terms in the radial variable $`\overline{t}_{(1)}`$ appears. However, the restriction of the BH-like structure to the interval $`0ts`$, at least for any finite $`s`$ does not permit the definition of an asymptotic radial variable associated with some corresponding Rindler space. The โsymmetricโ case c) and all other situations with general $`\alpha `$ in (42)-(43) have Rindler terms in the whole range of $`t`$. Thus the presence of an asymptotically flat Minkowski space on the sense of a double limit $`x^0\mathrm{},y^0\mathrm{}`$ (subjected to a very special sequence of those limits) is restricted to case a).
### 3.2 Massive scalars
For massive scalars the last term in (22) again occurs in combination with a source for the zweibeine, namely $`j_3`$. For the vertex $`S^{(4)}`$ now another term is created from an $`(_0S)^2`$ in $`e^{\widehat{T}}`$ and that mass term. Therefore in the new vertex contribution
$$R^{(4)}=d^2xd^2yS_x^2R^{(4)}(x,y)\left(_0S\right)_y^2$$
(59)
we get $`(\mathrm{}=1)`$
$$R^{(4)}=\frac{m^2}{2}\frac{\delta ^2\stackrel{~}{}}{\delta j_3(x)\delta j_2(y)}=\frac{m^2}{2}\frac{\delta q_3(x)}{\delta j_2(y)}$$
(60)
This expression is simply the factor of $`c`$ in (47). Together with the normalization of $`\overline{q}_3`$ and the overall $`\delta (x^1y^1)`$ we obtain in case a) of our different solutions for (42) and (43)
$$\frac{\delta q_3(x)}{\delta j_2(y)}=\frac{\mathrm{\Theta }(x^0y^0)}{2\sqrt{2}(x_0)^{3/2}}(x_0y_0).$$
(61)
The crucial difference to the other cases b) and c) consists in the property of (61) that only here the step function โprotectsโ the $`x`$ and $`y`$ integrations from the singularity at $`x^0=0`$ in $`R^{(4)}`$ (see below).
## 4 Scattering Amplitude
For the scattering process of two scalars $`S+SS+S`$ through the two vertex contributions $`S^{(4)}+R^{(4)}`$ of the previous section we first transform both $`x`$ and $`y`$ to the asymptotically flat coordinates (54). Clearly, in view of the remarks after (58), it may seem questionable that this is consistent with the properties of the vertex at asymptotic distances. Only in case a) an effective line element, say in $`x`$, exists which is asymptotically flat. But it refers to an accelerated system whose acceleration is proportional to $`(y_0)^{1/2}`$. On the other hand, the free fields $`S(x)`$ in the interaction picture approach to standard scattering theory cannot show a dependence on the variable $`y`$ of a different vertex. This, in our opinion, justifies the assumption that (for case a)) there must be an asymptotic limit towards โindependentโ flat Minkowski space for $`S(x)`$ and $`S(y)`$ for a (hopefully) gauge fixing independent $`S`$-matrix element to exist. Then (31) with (41) and (55) yields $`(x^1=\overline{x}^1)`$ the manifestly nonlocal vertex
$$S^{(4)}=\frac{1}{64}d^2\overline{x}d^2\overline{y}\mathrm{\Theta }(\overline{x}_0)\mathrm{\Theta }(\overline{y}_0)\frac{\left(_{\overline{x}_0}S\right)^2}{\overline{x}_0^2}\frac{\left(_{\overline{y}_0}S\right)^2}{\overline{y}_0^2}\left|\overline{x}_0\overline{y}_0\right|^3\delta (x_1y_1)$$
(62)
which is the same for $`m=0`$ and $`m0`$ as well as for all possible $`\alpha `$-prescriptions.
In a similar manner the additional vertex for massive scalars from (6), (60) and (61) becomes
$`R^{(4)}`$ $`=`$ $`{\displaystyle \frac{m^2}{8}}{\displaystyle d^2\overline{x}d^2\overline{y}\mathrm{\Theta }(\overline{x}_0)\mathrm{\Theta }(\overline{y}_0)\delta (x_1y_1)}`$ (63)
$`\times {\displaystyle \frac{\mathrm{\Theta }(\overline{x}_0\overline{y}_0)(\overline{x}_0^2\overline{y}_0^2)}{\overline{x}_0^2\overline{y}_0}}S_{\overline{x}_0}^2\left(_{\overline{y}_0}S\right)^2.`$
The vertex has been given for the case a) ($`\alpha =0`$) in (43).
### 4.1 Massless Scalars
Evidently both vertices (62) and (63) exhibit singularities at $`\overline{x}^0=0,\overline{y}^0=0`$. Let us consider asymptotic massless scalars first. Without imposing any boundary condition at $`\overline{x}^0=0`$ the usual decomposition of the scalar field into Minkowski space modes will be
$$S=\frac{1}{\sqrt{2\pi }}\frac{dk}{\sqrt{2k}}\left(a_R^+e^{ik(\tau z)}+a_L^+e^{ik(\tau +z)}+a_R^{}e^{ik(\tau z)}+a_L^{}e^{ik(\tau +z)}\right),$$
(64)
where the indices $`R`$ and $`L`$ denote the right and left moving parts in our one-dimensional situation. In the โoutgoing EF coordinatesโ (54) used in (6) the arguments are simply $`\tau z=\overline{x}^1,\tau +z=\overline{x}^1+2\overline{x}^0`$ . The singularity of (65) in a scattering matrix element for two ingoing and outgoing Minkowski quanta of the $`S`$-field with momenta $`q,q^{}`$ and $`k,k^{}`$, respectively
$$T(q,q^{};k,k^{})=\frac{1}{2}0\left|a^{}(k)a^{}(k^{})S^{(4)}a^+(q)a^+(q^{})\right|0$$
(65)
for any $`R`$ and $`L`$ can be interpreted as an indication that in this case the formation of a BH is โinevitableโ. Indeed regularizing (62) with $`(\overline{x}^0)^2lim_{\delta 0}(\overline{x}^0{}_{}{}^{2}+\delta ^2)^1`$ and defining a left moving wave packet formally by $`S\delta ^{3/4}S`$ would yield a finite result (up to an undetermined factor). The observed divergence of the scattering amplitude (65) for our vertex which contributes to the tree approximation thus seems to be in qualitative agreement with the conjecture , when scalar fields with minimal coupling are introduced at the reduced ($`1+1`$) level: Then the BH is formed for arbitrary small amounts of collapsing matter. A threshold for BH formation only occurs if nonminimally coupled scalars are considered, corresponding to proper taking into account of the $`s`$-wave nature in the spherically reduced situation .
A radical solution to the divergence problem consists in imposing a suitable boundary condition<sup>5</sup><sup>5</sup>5This boundary condition implies that $`S`$ becomes a self-dual scalar field at the origin - an essential difference to Dirichlet or Neumann boundary conditions, which - in a certain sense - are dual to each other . for the scalar field to make (6) finite:
$$\frac{S}{\overline{x}^0}|_{\overline{x}^0=0}=\left(\frac{S}{\tau }+\frac{S}{z}\right)=0.$$
(66)
However, this eliminates the left-movers $`(a_L^\pm =0)`$ in (62) also at $`\overline{x}^00`$, and (63), as well as in all higher order vertices. The physical system now consists of a background, eventually describing a fixed BH by a singularity at $`\overline{x}^0=z=0`$, and free right-moving scalars which run away from it. No genuine BH formation occurs in this setting.
### 4.2 Massive scalars
Although also in this case half of the modes are eliminated, a nontrivial result is obtained for free massive scalars with energy $`E_k=\sqrt{k^2+m^2}`$, obeying the boundary condition (66). The modes can be extracted from
$`S`$ $`=`$ $`{\displaystyle \underset{0}{\overset{\mathrm{}}{}}}dkN(k)\{a^+(k)e^{iE(\overline{x}^0+\overline{x}^1)}[(E_k+k)e^{ik\overline{x}^0}`$
$`(E_kk)e^{ik\overline{x}^0}]+h.c.\}`$
with the normalization factor
$$N(k)=\left[4\pi E_k(E_k^2+k^2)\right]^{1/2}$$
(68)
determined such that the Hamiltonian is $`H=_0^{\mathrm{}}๐kE_ka^+a^{}`$.
Now
$$\frac{S}{\overline{x}_0}=2m^2_0^{\mathrm{}}dkN(k)\mathrm{sin}k\overline{x}_0(a^+e^{iE_k(\overline{x}_0+\overline{x}_1)}+h.c.)$$
(69)
clearly obeys (66), but for $`m0`$ it does not vanish identically any more. In the presence of boundary conditions, $`a^\pm (k)`$ for positive, respectively negative energy in the S-matrix element (65) are related to values $`k0`$. Each $`k`$ labels a mixture of left and right moving โparticlesโ. With (69) the matrix element $`T^{(S)}`$ of (65) is regular at $`\overline{x}^0=\overline{y}^0=0`$. The same is true for the analogous one from $`R^{(4)}`$. In the latter the singularity $`\overline{x}_0^2`$ is absent only for the solution a) ( $`\alpha =0`$ in (42) and (43) ) thanks to the step functions. Both contributions to the total matrix element can be integrated completely yielding distributions. Some details of the calculation are described in Appendix A.
For โincomingโ momenta $`q,q^{}`$ with energies $`E_q,E_q^{}`$ and outgoing ones $`k,k^{}`$ with $`E_k,E_k^{}`$, taking into account symmetries and ensuing factors we obtain from (65)
$`T^{(S)}(q,q^{};k,k^{})`$ $`=`$ $`{\displaystyle \frac{m^8}{64}}N(q)N(q^{})N(k)N(k^{})\delta (E_k+E_k^{}E_qE_q^{})`$ (70)
$`[I(q,q^{};k,k^{})+I(q,k;q^{},k^{})+`$
$`+I(k^{},q^{};k,q)+I(k,k^{};q,q^{})]`$
with
$$I:=\underset{abrs}{}abrsK_{ab,rs}$$
(71)
where the sum extends over $`a,b,r,s,`$ each taking the values $`\pm 1`$. With the abbreviations $`(E=E_k+E_k^{}=E_q+E_q^{}`$)
$`u_{ab}`$ $`:=`$ $`ak+bk^{}+E=\stackrel{~}{u}_{ab}+E`$
$`v_{rs}`$ $`:=`$ $`rq+sq^{}+E=\stackrel{~}{v}_{rs}+E`$ (72)
the quantity $`K_{ab,rs}:=lim_{\epsilon 0}K(u_{ab},v_{rs})`$ is given by
$`K(u,v)`$ $`:=`$ $`2\pi i{\displaystyle \frac{(uv)^3}{u^2v^2}}\mathrm{ln}(uv+i\epsilon )`$ (73)
$`+{\displaystyle \frac{\pi i}{v^2}}(3vu)\mathrm{ln}(u+i\epsilon )`$
$`{\displaystyle \frac{\pi i}{u^2}}(3uv)\mathrm{ln}(vi\epsilon ).`$
The logarithms of complex arguments are defined as usual, i.e.:
$$\underset{\epsilon 0}{lim}\mathrm{ln}(r\pm i\epsilon )=\mathrm{ln}\left|r\right|\pm i\pi \mathrm{\Theta }(r)$$
(74)
The final result for the second contribution (63) inserted in (65) has a similar structure:
$`T^{(R)}`$ $`=`$ $`{\displaystyle \frac{m^6}{16}}N(q)N(q^{})N(k)N(k^{})\delta (E_q+E_q^{}E_kE_k^{})`$ (75)
$`[V(q,q^{};k,k^{})+V(q,k^{};k,q^{})+`$
$`+V(k^{},q;k,q)+V(k,k^{};q,q^{})]`$
$`V`$ $`:=`$ $`V^{(1)}+V^{(2)}`$ (76)
$`V^{(1)}`$ $`:=`$ $`{\displaystyle \underset{abrs}{}}rs(E_q+rq)(E_q^{}+sq^{})J_{abrs}`$ (77)
$`V^{(2)}`$ $`:=`$ $`{\displaystyle \underset{abrs}{}}rsb(E_q+rq)(E_q^{}+sq^{})L_{abrs}`$ (78)
In terms of $`\stackrel{~}{u}_{ab}`$ and $`\stackrel{~}{v}_{rs}`$ defined in (72) we have for $`J_{abrs}:=lim_{\epsilon 0}J(\stackrel{~}{u}_{ab},\stackrel{~}{v}_{rs})`$, $`L_{abrs}:=lim_{\epsilon 0}L(\stackrel{~}{u}_{ab},\stackrel{~}{v}_{rs})`$ the expressions
$`J(\stackrel{~}{u},\stackrel{~}{v})`$ $`:=`$ $`{\displaystyle \frac{2\pi i}{\stackrel{~}{v}}}\mathrm{ln}(\stackrel{~}{u}+\stackrel{~}{v}+i\epsilon )`$ (79)
$`L(\stackrel{~}{u},\stackrel{~}{v})`$ $`:=`$ $`2\pi i{\displaystyle \frac{(E\stackrel{~}{v})}{(E\stackrel{~}{u})^2}}\mathrm{ln}(E\stackrel{~}{v}i\epsilon )`$ (80)
For both contributions the common overall sign has been fixed by the choice $`2\sqrt{2}\overline{q}_3=+1`$ in (52). Both terms also share the energy conservation factor. Therefore, only a probability per unit of time with factor $`\frac{1}{2\pi }\delta (E_k+E_k^{}E_qE_q^{})`$ is a well defined quantity. The situation with respect to momenta is different because of the nonlocality of the vertex. Thus we encounter a situation similar to scattering at a fixed external โpotentialโ in ordinary quantum mechanics (or, equivalently, in $`D=0+1`$ dimensional QFT).
It should be noted that in the infrared limit both amplitudes are proportional to $`m`$ and hence vanish for the massless case in agreement with the previous discussion. For large energies<sup>6</sup><sup>6</sup>6Note that for very large energies our perturbation theory breaks down since the effects from the scalar field are not โsmallโ anymore; therefore, the energy should lie in the range $`mEE_{Planck}`$. the amplitudes decrease rapidly:
$$T^{(S)}|_{UV}m^8\frac{\mathrm{ln}E}{E^7},T^{(R)}|_{UV}m^6\frac{\mathrm{ln}E}{E^5}$$
(81)
## 5 Summary and Outlook
Two dimensional path integral quantum gravity can now be based upon a well-defined formalism which โ in a very specific gauge โ allows to separate the exact, almost trivial, quantum integral of the geometric variables from the loop-wise effects of the scalars. In our present work we considered the classical, tree-approximation, limit for minimally interacting scalar fields $`S`$, starting from the path integral formalism. This implies the appearance of effective (classical) $`2n`$-vertices of scalar fields ($`n2`$). Those vertices are highly nontrivial, because they yield โ through the natural appearance of classical background phenomena โ mathematical structures which allow the interpretation that an intermediate โvirtualโ BH is involved. We have studied this for the geometric action as derived by spherically reducing Einstein gravity. The scalar matter field was assumed to be coupled minimally at the $`d=2`$ level. We also concentrated upon the simplest nontrivial vertex $`S^{(4)}`$ with four scalar fields.
For the massless case we found that the resulting nonlocal matrix element for unrestricted left- and right-moving scalar fields diverges as $`_0^{z_f}\frac{dz}{z^2}`$ at that point in space-time which can be identified with the โlocationโ of a singularity. However, imposing a suitable boundary condition upon $`S`$ completely eliminates the scalar excitation moving towards the singularity, whereas the ones moving away decouple from the theory: The manifold has been โpluggedโ at the place where an eventual BH may have been formed. We believe that this result at the particle level shows a qualitative relation to a conjecture for macroscopic BH formation: There for minimally coupled scalars a BH forms without any threshold . The divergence of a probability amplitude or, alternatively, an amplitude which is finite only for a wave packet, properly rescaled to tend to zero width at $`z=0`$, seems to imply the same phenomenon.
As an example for a system where finite amplitudes for minimally coupled scalars can be obtained we also studied massive scalars, where the necessary boundary condition no longer prevents BH effects. Both, the vertex from the massless case, and another new one, induced by the mass-term, yield finite results which can be even represented by (complicated) sums of directions (plus or minus) of momenta in terms of (simple) functions and distributions. Overall energy conservation holds in the process $`S+SS+S`$. Momenta are not conserved, in general. Here we note parallels to recent work of P. Hรกjรญฤek on massless, but non-minimally coupled thin spherical shells. He found no residual BH for a collapsing shell with Dirac quantization. Also in that work the phenomenon has been observed which we have called โvirtual BHโ, consisting in a certain sense of a black and a white hole.
The next task is to take into account also the proper nonminimal coupling of scalars at the $`2d`$ level. Superficially no essential basic changes for the vertices may be expected: On the one hand, the measure of the integral will change as $`dzd\tau z^2dzd\tau `$, because $`z`$ will become a radial variable. On the other hand the scalar field will be reduced to the one describing $`s`$-waves in $`d=4`$, i.e. $`SS/z`$. But e.g. the threshold effect known for macroscopic studies should show up. In that case a detailed comparison with Dirac quantization as treated in will be possible.
Our formalism is general enough so that any other 2d gravity theory, produced e.g. by spherical reduction of generalized Einstein gravities in $`d=4`$, can be covered as well.
Of course, also the study of higher loop orders in the scalar fields, based upon the one-loop determinant (Polyakov type action) in the path integral, as well as of higher loops involving the vertices discussed here, together with propagators of the scalars, remains a wide field of possible further applications.
## Acknowledgment
The authors thank their colleagues at the Institute for Theoretical Physics of the Vienna University of Technology for many discussions. This paper has been supported by project P-12815-TPH of the Austrian Science Foundation (FWF). One of the authors (D.V.) has been supported in part by the Alexander von Humboldt foundation and by RFBR, grant 97-01-01186.
## Appendix A: Derivation of Scattering Amplitudes
The explicit computation of the scattering amplitude $`T^{(S)}`$ in (70) and $`T^{(R)}`$ in (75) for Minkowski modes in the initial and final state is most conveniently based upon suitable Fourier transforms of the rational factors in $`\overline{x}^0`$ and $`\overline{y}^0`$ in (62) and (63). From the identity
$$_0^{\mathrm{}}x^\lambda e^{i(\sigma +i\epsilon )}๐x=ie^{\frac{i\lambda \pi }{2}}\mathrm{\Gamma }(\lambda +1)(\sigma +i\epsilon )^{\lambda 1}$$
(82)
the required singular limits $`\delta +0,\epsilon +0`$ for the Fourier transforms at $`\lambda =2+\delta `$, resp. $`\lambda =1+\delta `$ are
$$_{\mathrm{}}^{\mathrm{}}x^{2+\delta }\mathrm{\Theta }(x)e^{i(\sigma +i\epsilon )}=f_2(\sigma ,\epsilon )\left[\left(\frac{1}{\delta }+(1+\frac{i\pi }{2}\gamma )\right)\mathrm{ln}(\sigma +i\epsilon )+๐ช(\delta )\right]$$
(83)
resp.
$$_{\mathrm{}}^{\mathrm{}}x^{1+\delta }\mathrm{\Theta }(x)e^{i(\sigma +i\epsilon )}=f_1(\sigma ,\epsilon )\left[\left(\frac{1}{\delta }+(\frac{i\pi }{2}\gamma )\right)\mathrm{ln}(\sigma +i\epsilon )+๐ช(\delta )\right],$$
(84)
with
$$f_n:=i(1)^{(n1)}e^{i\frac{n\pi }{2}}\left(\sigma +i\epsilon \right)^{(n1)}$$
(85)
Introducing (83), resp. (84) into (70) resp. (75) and using ($`P`$ is Cauchyโs principal value)
$`\epsilon (x)={\displaystyle \frac{1}{\pi i}}P{\displaystyle ๐\tau \frac{e^{i\tau x}}{\tau }}`$
$`\mathrm{\Theta }(x)={\displaystyle \frac{1}{2\pi i}}{\displaystyle ๐\tau \frac{e^{i\tau x}}{\tau i\epsilon }}`$ (86)
all the terms proportional to $`\delta ^1`$ from (83) and (84) cancel together with the constant contributions<sup>7</sup><sup>7</sup>7These cancellations are a direct consequence of the boundary condition (66). โ as they should in these finite integrals. Furthermore in (70) it is useful to replace the factor $`(\overline{x}^0\overline{y}^0)^3`$ by a third derivative with respect to $`E=E_k+E_k^{}`$. Then the generic integral
$$A(a,b)=๐\tau \mathrm{ln}(a+\tau +i\epsilon _1)\mathrm{ln}(b\tau +i\epsilon _2)(a+\tau )(b+\tau )(\frac{1}{\tau +i\epsilon _3}+\frac{1}{\tau i\epsilon _3})$$
(87)
remains which after three differentiations with respect to $`E`$ in $`a=E+q+q^{},b=E+k+k^{}`$ becomes a contribution of integrals with one logarithm multiplied by a factor with two or three poles. These integrals are straightforward and can be most conveniently done using the contour depicted in Fig. 3.
In the vertex $`T^{(R)}`$ in (75) for the first contribution $`V^{(1)}`$ the procedure is the same, not even requiring some differentiation at an intermediate step. $`V^{(2)}`$ originates from the term with factor $`(\overline{y}^0\overline{x}^2`$). Here $`\overline{y}^0`$ may be expressed first by a derivative with respect to one of the momenta in the sine factor from $`(_{\overline{y}}S)^2`$ (cf. (69)). |
warning/0001/math0001092.html | ar5iv | text | # The Orbit Method for Finite Groups of Nilpotency Class Two of Odd Order
## 1 Introduction
Fortunately, Kirillov published recently a survey of merits and demerits of the orbit method , so I can refer to it and skip a historical introduction here.
## 2 Groups of nilpotency class 2
By definition, a group $`B`$ of nilpotency class 2 is a central extension of an abelian group, i.e. there is an exact sequence
$$0ABC0$$
(2.1)
where $`A`$ and $`C`$ are abelian groups and $`A`$ is the center of $`B`$. In other words, elements of the group $`B`$ of nilpotency class 2 can be written as pairs $`b=(a,c)`$ with $`aA`$ and $`cC`$ so that
$$(a_1,c_1)(a_2,c_2)=(a_1+a_2+\psi (c_1,c_2),c_1+c_2).$$
(2.2)
The associativity of group operation (2.2) is equivalent to the following identity:
$$\psi (c_1,c_2)+\psi (c_1+c_2,c_3)=\psi (c_1,c_2+c_3)+\psi (c_2,c_3)$$
(2.3)
which means that $`\psi `$ is a 2-cocycle, $`\psi \text{C}^2(C,A)`$ supposing that the action of $`C`$ on $`A`$ is trivial. Note that substituting either $`c_1=c_2=0`$, or $`c_2=c_3=0`$ in (2.3), we obtain the identities
$$\psi (0,c)=\psi (c,0)=\psi (0,0).$$
(2.4)
For every such cocycle $`\psi `$, the element $`(\psi (0,0),0)`$ is an identity of $`B`$ and
$$(a,c)^1=(a\psi (c,c)\psi (0,0),c)$$
(2.5)
is a left inverse to $`(a,c)`$, which means that formula (2.2) defines a group structure in set $`B=A\times C`$ for every cocycle $`\psi \text{C}^2(C,A)`$. In particular, the left inverse (2.5) is a right inverse of $`(a,c)`$ as well, thus formula (2.3) implies the identity
$$\psi (c,c)=\psi (c,c).$$
(2.6)
In this construction, both mappings
$`AA\times \{0\},`$ $`a(a\psi (0,0),0);`$ (2.7)
$`CB/A,`$ $`c(0,c)A`$ (2.8)
are isomorphisms. Choosing other representatives of cosets than $`(0,c)`$ in (2.8), for any function $`q:CA`$,
$$CB/A,c(0,c)(q(c)\psi (0,0),0)A=(q(c),c)A$$
(2.9)
we obtain the same group structure in $`B`$. Renaming
$$(a,c)_{\text{new}}=(a+q(c),c),$$
(2.10)
we obtain from (2.2) that the same group structure in $`B`$ can be defined by 2-cocycle
$$\psi _{\text{new}}(c_1,c_2)=\psi (c_1,c_2)+q(c_1)+q(c_2)q(c_1+c_2).$$
(2.11)
The difference between these old and new cocycles is equal to the coboundary of 1-chain $`q`$, thus the group structure in $`B`$ is uniquely determined by 2-cocycles modulo coboundaries of 1-chains, i.e. by elements of $`\text{H}^2(C,A)`$ with trivial action of $`C`$ on $`A`$.
###### Definition 2.1.
Weโll call a cocycle $`\psi \text{C}^2(C,A)`$ centered iff $`\psi (0,0)=0`$.
From (2.4), for a centered cocycle $`\psi `$, for every $`cC`$,
$$\psi (0,c)=\psi (c,0)=\psi (0,0)=0.$$
(2.12)
Choosing $`q(0)=\psi (0,0)`$ in (2.11), we obtain a centered cocycle. Comparing (2.12) with the formula for the identity in $`B`$, we see that we can always choose a centered cocycle $`\psi `$ in the same cohomology class so the identity element of $`B`$ was $`(0,0)`$.
###### Definition 2.2.
An abelian group $`A`$ is 2-divisible if homomorphism
$$AA,aa+a$$
(2.13)
is an automorphism. For an element $`a`$ of a 2-divisible abelian group, we denote $`a/2`$ the image of $`a`$ under the automorphism inverse to (2.13).
###### Lemma 2.3.
A finite abelian group is 2-divisible iff it has an odd order. An abelian $`p`$-group is 2-divisible iff $`p`$ is odd.
###### Proof.
The kernel of (2.13) is the subgroup of elements of order 2. So abelian 2-groups and abelian finite groups of even order canโt be 2-divisible. Finite abelian groups of odd order donโt have elements of order 2, so homomorphism (2.13) is injective, thus its image contains as many elements as $`A`$ does, so it is $`A`$. For abelian $`p`$-groups with odd $`p`$, homomorphism (2.13) is injective, and its restriction on every cyclic subgroup is an automorphism of that subgroup, so (2.13) is surjective as well. โ
###### Definition 2.4.
Weโll call a cocycle $`\psi \text{C}^2(C,A)`$ equalized iff $`\psi (c,c)=0`$ for all $`cC`$.
For an equalized cocycle $`\psi `$, from (2.5), we have
$$(a,c)^1=(a,c)$$
(2.14)
for every element $`(a,c)B`$.
###### Lemma 2.5.
For a 2-divisible abelian group $`A`$ and an abelian group $`C`$, every element of $`\text{H}^2(C,A)`$ can be represented by an equalized cocycle.
###### Proof.
If $`A`$ is a 2-divisible abelian group, for a centered cocycle $`\psi `$, choosing $`q(c)=\psi (c,c)/2`$ in (2.11), and combining it with (2.6), we obtain an equalized cocycle. โ
For any $`A`$, if $`C`$ doesnโt have elements of order 2, an equalized cocycle can be constructed from a centered cocycle $`\psi `$ by choosing $`\{q(c),q(c)\}=\{\psi (c,c),0\}`$ with an arbitrary choice of which of them must be equal 0. Here is the example showing that it canโt be obtained for all cases.
###### Example 2.6.
Let either $`B\text{D}_8`$, a dihedral group of order 8, or $`B\text{Q}_8`$, a quaternionic group of order 8 . In both cases $`A\text{C}_2`$ and $`C\text{C}_2\text{C}_2`$ where $`\text{C}_n`$ denotes a cyclic group of order $`n`$. We have $`a=a`$ and $`c=c`$ for all elements $`aA`$, $`cC`$. If formula (2.14) was true, then all nontrivial elements of $`B`$ would have order 2, which is false.
If we have a centered or an equalized cocycle and we want to change it adding the coboundary of a 1-chain $`q`$ so that the new cocycle was still centered or equalized, we need the following conditions on $`q`$: $`q(0)=0`$ for centered cocycles, or $`q(c)=q(c)`$ for all $`cC`$ for equalized cocycles. Note also that for an equalized cocycle $`\psi `$ for any $`c_1,c_2C`$,
$$\psi (c_1,c_2)=\psi (c_2,c_1),$$
(2.15)
that follows from the identity $`(b_1b_2)^1=b_2^1b_1^1`$.
All what was told from (2.1) until now, was applicable to all central group extensions. We didnโt explore the fact that $`A`$ is exactly the center of $`B`$. What do we need for that? We need that cocycle $`\psi `$ was non-degenerate.
###### Definition 2.7.
A cocycle $`\psi `$ is non-degenerate iff for every $`c_1C`$, $`c_10`$ there exist such $`c_2C`$ that
$$\psi (c_1,c_2)\psi (c_2,c_1).$$
(2.16)
For a non-degenerate cocycle $`\psi `$, element $`b=(a_1,c_1)B`$ with $`c_10`$ canโt be central since it is not commute with $`(0,c_2)`$ with $`c_2`$ satisfying (2.16). Every coboundary of a 1-chain $`q`$ is symmetric, so adding it to $`\psi `$ doesnโt change inequality (2.16). That means that all cocycles representing a cohomology class from $`H^2(C,A)`$, are either non-degenerate, in which case weโll call that class non-degenerate; or degenerate, then weโll call that class degenerate as well.
## 3 Lie rings of nilpotency class 2
By a Lie ring I mean an abelian group with a bilinear commutator $`[.,.]`$ such that $`[a,a]=0`$ for all $`a`$, it implies $`[a,b]=[b,a]`$, and satisfying Jacobi identity. A commutative Lie ring means zero commutator.
By definition, a Lie ring $`๐`$ of nilpotency class 2 is a central extension of a commutative Lie ring, i.e. there is an exact sequence
$$0๐๐๐ 0$$
(3.1)
where $`๐`$ and $`๐ `$ are commutative Lie rings and $`๐`$ is the center of $`๐`$. In other words, elements of the Lie ring $`๐`$ of nilpotency class 2 can be written as pairs $`b=(a,c)`$ with $`a๐`$ and $`c๐ `$ so that
$$(a_1,c_1)+(a_2,c_2)=(a_1+a_2+\varphi (c_1,c_2),c_1+c_2).$$
(3.2)
and
$$[(a_1,c_1),(a_2,c_2)]=[\eta (c_1,c_2)\varphi (0,0),0].$$
(3.3)
The associativity and commutativity of the operation in $`๐`$ defined in (3.2) are equivalent to $`\varphi `$ being a symmetric 2-cocycle $`\varphi \text{C}^2(C,A)`$ where $`C`$ and $`A`$ are underlying abelian groups of $`๐ `$ and $`๐`$, correspondingly, supposing that the action of $`C`$ on $`A`$ is trivial. The commutator properties are equivalent to the fact that $`\eta :C\times CA`$ is a skew-symmetric bihomomorphism where skew-symmetric means $`\eta (c,c)=0`$ for all $`cC`$ which implies $`\eta (c_1,c_2)=\eta (c_2,c_1)`$. In other words, $`\eta `$ is a 2-cocycle $`\eta \text{C}^2(๐ ,A)`$ with trivial action of $`๐ `$ on A.
In this construction, both mappings
$`๐๐\times \{0\},`$ $`a(a\varphi (0,0),0);`$ (3.4)
$`๐ ๐/๐,`$ $`c(0,c)+๐`$ (3.5)
are Lie ring isomorphisms. Choosing other representatives of cosets than $`(0,c)`$ in (3.5), for any function $`q:CA`$,
$$๐ ๐/๐,c(0,c)+(q(c)\varphi (0,0),0)+๐=(q(c),c)+๐$$
(3.6)
we obtain the same Lie ring structure in $`๐`$. Renaming
$$(a,c)_{\text{new}}=(a+q(c),c),$$
(3.7)
we obtain from (3.2) that the same Lie ring structure in $`๐`$ can be defined by 2-cocycles
$$\varphi _{\text{new}}(c_1,c_2)=\varphi (c_1,c_2)+q(c_1)+q(c_2)q(c_1+c_2),$$
(3.8)
obtaining from $`\varphi `$ by adding a 1-chain, and $`\eta `$. Note that the coboundary of a 1-chain $`r\text{C}^1(๐ ,A)`$ with the trivial action of $`๐ `$ on $`A`$, is $`r(c_1,c_2)=r([c_1,c_2])=0`$. Thus the Lie ring structure in $`๐`$ is uniquely determined by elements of $`\text{H}_{\text{sym}}^2(C,A)\text{H}^2(๐ ,A)`$ with trivial actions of $`C`$ and $`๐ `$ on $`A`$, where $`\text{H}_{\text{sym}}^2(C,A)`$ denotes the subgroup of 2-cohomology classes defined by symmetric cocycles. The same as in the previous section, choosing $`q(c)=\varphi (0,0)`$ in (3.8), we get $`\varphi _{\text{new}}(0,0)=0`$, and $`(0,0)=0๐`$.
All what was told from (3.1) until now, was applicable to all central Lie ring extensions. We didnโt explore the fact that $`๐`$ is exactly the center of $`๐`$. What do we need for that? We need that cocycle $`\eta `$ was non-degenerate.
###### Definition 3.1.
A skew-symmetric bihomomorphism $`\eta :C\times CA`$ is non-degenerate iff for every $`c_1C`$, $`c_10`$ there exist such $`c_2C`$ that
$$\eta (c_1,c_2)0.$$
(3.9)
For a non-degenerate cocycle $`\eta `$, element $`b=(a_1,c_1)๐`$ with $`c_10`$ canโt be central since its commutator with $`(0,c_2)`$ with $`c_2`$ satisfying (3.9), is not 0.
## 4 Lie correspondence
###### Theorem 4.1.
For a 2-divisible (topological) abelian group $`A`$ and an abelian (topological) group $`C`$ acting trivially on $`A`$, the following mapping:
$$\mathrm{L}_\text{c}:\text{C}^2(\mathrm{C},\mathrm{A})\text{C}_{\text{sym}}^2(\mathrm{C},\mathrm{A})\text{C}^2(๐ ,\mathrm{A}),\psi (\varphi ,\eta )$$
(4.1)
where $`๐ `$ is the commutative (topological) Lie ring, underlying abelian group of which is $`C`$, acting trivially on $`A`$, $`\text{C}_{\text{sym}}`$ denotes symmetric cocycles, and
$$\varphi (c_1,c_2)=\frac{\psi (c_1,c_2)+\psi (c_2,c_1)}{2},$$
(4.2)
$$\eta (c_1,c_2)=\psi (c_1,c_2)\psi (c_2,c_1),$$
(4.3)
is an isomorphism, factor of which by coboundaries of 1-chains is an isomorphism of the cohomology groups
$$\mathrm{L}_\text{h}:\text{H}^2(\mathrm{C},\mathrm{A})\text{H}_{\text{sym}}^2(\mathrm{C},\mathrm{A})\text{H}^2(๐ ,\mathrm{A}).$$
(4.4)
For a 2-divisible (topological) commutative Lie ring $`๐`$ underlying abelian group of which is $`A`$, and a commutative (topological) Lie ring $`๐ `$ acting trivially on $`A`$, the following mapping:
$$\mathrm{E}_\text{c}:\text{C}_{\text{sym}}^2(\mathrm{C},\mathrm{A})\text{C}^2(๐ ,\mathrm{A})\text{C}^2(\mathrm{C},\mathrm{A}),(\varphi ,\eta )\psi $$
(4.5)
where
$$\psi (c_1,c_2)=\varphi (c_1,c_2)+\frac{\eta (c_1,c_2)}{2},$$
(4.6)
is an isomorphism, factor of which by coboundaries of 1-chains is an isomorphism of the cohomology groups
$$\mathrm{E}_\text{h}:\text{H}_{\text{sym}}^2(\mathrm{C},\mathrm{A})\text{H}^2(๐ ,\mathrm{A})\text{H}^2(\mathrm{C},\mathrm{A}).$$
(4.7)
The isomorphisms $`L_\text{c}`$ and $`E_\text{c}`$ are mutually inverse and the isomorphisms $`L_\text{h}`$ and $`E_\text{h}`$ are mutually inverse. Cocycle $`\varphi `$ is centered or equalized iff $`\psi `$ is centered or equalized, correspondingly. Cocycle $`\eta `$ is non-degenerate iff $`\psi `$ is non-degenerate.
###### Proof.
From (4.2), $`\varphi `$ is a cocycle, by linearity, and it is symmetric,
$$\varphi (c_1,c_2)=\varphi (c_2,c_1).$$
(4.8)
$`\eta `$ is skew-symmetric by (4.3). To check that it is a bihomomorphism, add the following cocycle identities:
$`\psi (c_1+c_2,c_3)+\psi (c_1,c_2)`$ $`=\psi (c_1,c_2+c_3)+\psi (c_2,c_3),`$ (4.9)
$`\psi (c_3+c_1,c_2)+\psi (c_3,c_1)`$ $`=\psi (c_3,c_1+c_2)+\psi (c_1,c_2),`$ (4.10)
$`\psi (c_1+c_3,c_2)\psi (c_1,c_3)`$ $`=\psi (c_1,c_3+c_2)\psi (c_3,c_2).`$ (4.11)
After cancelling equal items, we get
$$\psi (c_1+c_2,c_3)+\psi (c_3,c_1)\psi (c_1,c_3)=\psi (c_3,c_1+c_2)+\psi (c_2,c_3)\psi (c_3,c_2),$$
(4.12)
or
$$\eta (c_1+c_2,c_3)=\eta (c_1,c_3)+\eta (c_2,c_3).$$
(4.13)
$`L_\text{c}`$ is a homomorphism by linearity of (4.2) and (4.3). Adding coboundary to $`\psi `$ adds the same coboundary to $`\varphi `$, and inverse from (4.6), and we donโt have to worry about adding coboundaries to $`\eta `$ since $`\text{H}^2(๐ ,A)=\text{C}^2(๐ ,A)`$, so $`L_\text{c}`$ defines $`L_\text{h}`$. From the other side, $`\psi `$ defined in (4.6) is a cocycle by linearity (note, that $`\eta `$ is a cocycle since every bihomomorphism from $`C\times C`$ to $`A`$ is an element of $`\text{C}^2(C,A)`$). $`E_\text{c}`$ is a homomorphism by linearity of (4.6). It is easy to check that $`L_\text{c}E_\text{c}=\mathrm{id}`$ and $`E_\text{c}L_\text{c}=\mathrm{id}`$. Since $`L_\text{h}`$ and $`E_\text{h}`$ are defined from $`L_\text{c}`$ and $`E_\text{c}`$ by a factorization by the same coboundaries, they are mutually inverse isomorphisms as well. The last statement of the theorem immediately follows from the definitions. โ
For a group $`B`$ of nilpotency class 2 with 2-divisible center denote $`L(B)`$ the Lie ring of nilpotency class 2 defined on the underlying set of $`B`$ by cocycles (4.2) and (4.3). Theorem 4.1 tells that this construction doesnโt depend on the choice of the cocycle $`\psi `$ defining a group structure in $`B`$. For a Lie ring $`๐`$ of nilpotency class 2 with 2-divisible center denote $`E(๐)`$ the group of nilpotency class 2 defined on the underlying set of $`๐`$ by a cocycle (4.6). Theorem 4.1 tells that this construction doesnโt depend on the choice of the cocycles $`\varphi `$ and $`\eta `$ defining a Lie ring structure in $`๐`$.
For a homomorphism $`f:B_1B_2`$ of groups $`B_1`$ and $`B_2`$ of nilpotency class 2 with 2-divisible centers denote $`L(f)=f`$ considered as a function from $`L(B_1)`$ to $`L(B_2)`$. Analogously, for a homomorphism $`f:๐_1๐_2`$ denote $`E(f)=f`$ considered as a function from $`E(๐_1)`$ to $`E(๐_1)`$.
###### Theorem 4.2.
$`L`$ and $`E`$ are mutually inverse functors defining an isomorphism between categories of groups of nilpotency class 2 with 2-divisible center and Lie rings of nilpotency class 2 with 2-divisible center.
###### Proof.
By Theorem 4.1, L and E are mutually inverse if they are functors. All categorical properties would follow immediately from definitions if we showed that $`L(f)`$ is a Lie ring homomorphism for every group homomorphism $`f`$ and $`E(f)`$ is a group homomorphism for every Lie ring homomorphism $`f`$.
By Lemma 2.5, we can choose equalized cocycles $`\psi _1`$ and $`\psi _2`$ defining group structures in $`B_1`$ and $`B_2`$. Then symmetric cocycles $`\varphi _1`$ and $`\varphi _2`$ defined by Theorem 4.1, are equalized as well. Multiplying the left hand sides and the right hand sides of the following identities:
$$\begin{array}{c}f(a_1+a_2+\psi _1(c_1,c_2),c_1+c_2)=f(a_1,c_1)f(a_2,c_2)\hfill \\ \hfill =(\alpha _1,\gamma _1)(\alpha _2,\gamma _2)=(\alpha _1+\alpha _2+\psi _2(\gamma _1,\gamma _2),\gamma _1+\gamma _2),\end{array}$$
(4.14)
$$\begin{array}{c}f(a_1a_2\psi _1(c_2,c_1),c_1c_2)=(f(a_2,c_2)f(a_1,c_1))^1\hfill \\ \hfill =((\alpha _2,\gamma _2)(\alpha _1,\gamma _1))^1=(\alpha _1\alpha _2\psi _2(\gamma _2,\gamma _1),\gamma _1\gamma _2)\end{array}$$
(4.15)
we get
$$f(\eta _1(c_1,c_2),0)=(\eta _2(\gamma _1,\gamma _2),0).$$
(4.16)
Dividing the central elements by 2 and inverting, we get
$$f(\frac{\eta _1(c_1,c_2)}{2},0)=(\frac{\eta _2(\gamma _1,\gamma _2)}{2},0).$$
(4.17)
Multiplying (4.17) and (4.14), we get
$$f(a_1+a_2+\varphi _1(c_1,c_2),c_1+c_2)=(\alpha _1+\alpha _2+\varphi _2(\gamma _1,\gamma _2),\gamma _1+\gamma _2).$$
(4.18)
So
$$L(f)(b_1+b_2)=L(f)(b_1)+L(f)(b_2)$$
(4.19)
for $`b_1=(a_1,c_1)`$, $`b_2=(a_2,c_2)`$. This formula together with (4.16) rewritten as
$$L(f)([b_1,b_2])=[L(f)(b_1),L(f)(b_2)]$$
(4.20)
means that $`L(f)`$ is a Lie ring homomorphism.
Similarly, for a Lie ring homomorphism $`f`$, multiplying (4.18) and the identity obtained from (4.17) by changing $``$ to $`+`$, we get (4.14) which means that $`E(f)`$ is a group homomorphism. โ
###### Lemma 4.3.
For any elements $`b,b_1,b_2`$ of a group $`B`$ of nilpotency class 2 with 2-divisible center,
$$b^1=b,$$
(4.21)
$$b_1b_2=b_1+b_2+\frac{[b_1,b_2]}{2},$$
(4.22)
$$b_1b_2b_1^1=b_2+[b_1,b_2],$$
(4.23)
$$b_1b_2b_1^1b_2^1=[b_1,b_2].$$
(4.24)
If $`b_1`$ and $`b_2`$ commute, then
$$b_1b_2=b_1+b_2.$$
(4.25)
###### Proof.
Formula (4.21) is true because by Lemma 2.5 we can choose an equalized cocycle $`\psi `$ defining group structure in $`B`$ and it corresponds by Theorem 4.1 to the equalized cocycle $`\varphi `$ defining additive group structure in $`L(B)`$. Formula (4.25) is true because $`\psi (c_1,c_2)=\varphi (c_1,c_2)`$ for commuting $`b_1=(a_1,c_1)`$ and $`b_2=(a_2,c_2)`$. We already used (4.22) at the end of the proof of Theorem 4.2. Formula (4.24) telling that the group commutator in $`B`$ coincides with the Lie ring commutator in $`L(B)`$ is true since the product of left hand sides of formulas (4.14) and (4.15) equals left hand side of (4.16). Formula (4.23) can be obtained by right multiplication of both sides of (4.24) by $`b_2`$ and using (4.25). โ
###### Definition 4.4.
A group $`B`$ is 2-rootable if mapping
$$BB,bb^2$$
(4.26)
is a bijection.
###### Corollary 4.5.
A finite group of nilpotency class 2 is 2-rootable iff it has an odd order. A $`p`$-group of nilpotency class 2 is 2-rootable iff $`p`$ is odd.
###### Proof.
Groups of even order and 2-groups have elements of order 2, so the identity covers more than once by (4.26), and these groups canโt be 2-rootable. Finite groups of odd order and $`p`$-groups with odd $`p`$ have 2-divisible center, by Lemma 2.3, so $`b^2=b+b`$ by (4.25), and Lemma 2.3 applied to $`L(B)`$, completes the proof. โ
## 5 The orbit method
###### Lemma 5.1.
For a group $`B`$ of nilpotency class 2 with 2-divisible center, formula
$$\mathrm{Ad}(b)(l)=L(blb^1)$$
(5.1)
defines a structure of left $`B`$-module in the underlying abelian group of $`L(B)`$.
###### Proof.
Conjugation by $`bB`$ is an automorphism of $`B`$, so $`\mathrm{Ad}(b)`$ is a Lie ring automorphism by Theorem 4.2. The formula
$$\mathrm{Ad}(b_1)\mathrm{Ad}(b_2)=\mathrm{Ad}(b_1b_2)$$
(5.2)
follows directly from the definition (5.1). โ
As usual, weโll call $`\mathrm{Ad}`$ adjoint representation of $`B`$. Denote $`L(B)^{}`$ the group of (unitary) characters of the underlying abelian group of $`L(B)`$. Define coadjoint representation $`\mathrm{Ad}^{}`$ as the dual to $`\mathrm{Ad}`$ left action of $`B`$ in $`L(B)^{}`$:
$$\mathrm{Ad}^{}(b)(\chi )(l)=\chi (\mathrm{Ad}(b^1)(l)).$$
(5.3)
###### Lemma 5.2.
For a group $`B`$ of nilpotency class 2 with 2-divisible center, for every $`bB`$, $`lL(B)`$, $`\chi L(B)^{}`$,
$$\mathrm{Ad}(b)(l)=l+[b,l],$$
(5.4)
$$\mathrm{Ad}^{}(b)(\chi )(l)=\chi (l[b,l]).$$
(5.5)
###### Proof.
Both formulas follow from (4.23) and definitions (5.1) and (5.3). โ
Denote $`๐ช(B)`$ the set of orbits of the coadjoint representation of $`B`$.
###### Theorem 5.3 (Orbit method).
For a finite group $`B`$ of nilpotency class 2 of odd order, elements
$$\left(X_\chi =\underset{lL(B)}{}\chi (l)l\right)_{\chi L(B)^{}}$$
(5.6)
form an orthonormal basis in $`[B]`$. For every orbit $`\mathrm{\Omega }`$ of the coadjoint representation of $`B`$, the subspace
$$V_\mathrm{\Omega }=\mathrm{span}(X_\chi )_{\chi \mathrm{\Omega }}$$
(5.7)
is a two-side ideal of a group algebra $`[B]`$, the restriction of the regular representation of $`B`$ in $`[B]`$ on $`V_\mathrm{\Omega }`$ is isotypic, and
$$[B]=\underset{\mathrm{\Omega }๐ช(B)}{}V_\mathrm{\Omega }$$
(5.8)
is the decomposition of the regular representation of $`B`$ in $`[B]`$ in the direct sum of isotypic components.
###### Proof.
Elements (5.6) form an orthonormal basis in $`[L(B)]`$ and dot products in $`[L(B)]`$ and $`[B]`$ are the same.
$$\begin{array}{c}bX_\chi =\underset{lL(B)}{}\chi (l)bl=\underset{gB}{}\chi (b^1g)g=\underset{gB}{}\chi \left(b+g+\frac{[b,g]}{2}\right)g\hfill \\ \hfill =\chi (b)\underset{gB}{}\mathrm{Ad}^{}\left(\frac{b}{2}\right)(\chi )(g)g=\chi (b)X_{\mathrm{Ad}^{}(b/2)(\chi )}V_\mathrm{\Omega }.\end{array}$$
(5.9)
Similarly,
$$\begin{array}{c}X_\chi b=\underset{lL(B)}{}\chi (l)lb=\underset{gB}{}\chi (gb^1)g=\underset{gB}{}\chi \left(gb+\frac{[g,b]}{2}\right)g\hfill \\ \hfill =\chi (b)\underset{gB}{}\mathrm{Ad}^{}\left(\frac{b}{2}\right)(\chi )(g)g=\chi (b)X_{\mathrm{Ad}^{}(b/2)(\chi )}V_\mathrm{\Omega }.\end{array}$$
(5.10)
So $`V_\mathrm{\Omega }`$ is a two-side ideal of $`[B]`$. Because $`X_\chi `$ are orthogonal for different $`\chi `$, these ideals $`V_\mathrm{\Omega }`$ are orthogonal for different $`\mathrm{\Omega }`$. Since $`[B]`$ is a semi-simple associative algebra and every component in the direct sum of the isotypic components
$$[B]=\underset{\tau \widehat{B}}{}\mathrm{Iso}(\tau )$$
(5.11)
is simple, every $`V_\mathrm{\Omega }`$ is either one of these components, or a direct sum of a few of them. But the number of coadjoint orbits coincides with the number of conjugate classes of $`B`$, by Duality Lemma 5.4, i.e. with the number of irreducible unitary representations of $`B`$, i.e. with the number of components in the sum (5.11). Thus every $`V_\mathrm{\Omega }`$ coincides with isotypic component $`\mathrm{Iso}(\tau )n\tau `$ for an irreducible unitary representation $`\tau `$ where $`n=dim\tau `$, and different orbits correspond to different irreducible representations. โ
###### Lemma 5.4 (Duality Lemma).
Let a finite abelian group $`M`$ be a left $`G`$-module for a finite group $`G`$. Denote $`M^{}`$ the group of unitary characters of $`M`$ and define the structure of a dual left $`G`$-module in $`M^{}`$ by formula
$$g\chi (m)=\chi (g^1m).$$
(5.12)
Then the number of $`G`$-orbits in $`M^{}`$ is the same as the number of $`G`$-orbit in $`M`$.
###### Proof.
Denote $`T_M`$ the representation of $`G`$ in $`[M]`$ defined by formula $`T_M(g)m=gm`$. If element
$$x=\underset{mM}{}x_mm$$
(5.13)
is an invariant of $`T_M`$, then for every orbit $`\mathrm{\Xi }`$ of $`G`$ in $`M`$ all coefficients $`x_m`$ with $`m\mathrm{\Xi }`$ are the same. That means that elements
$$e_\mathrm{\Xi }=\underset{m\mathrm{\Xi }}{}m$$
(5.14)
form a basis in the space of invariants of $`T_M`$, and the dimension of the space of invariants equals number of $`G`$-orbits in $`M`$. The same is true for $`M^{}`$: the number of $`G`$-orbits in $`M^{}`$ equals the dimension of the space of invariants of representation $`T_M^{}`$ defined by formula $`T_M^{}(g)\chi =g\chi `$. By construction, the space $`[M^{}]`$ is the dual space to $`[M]`$ and representations $`T_M`$ and $`T_M^{}`$ are dual. Thus if
$$T_M\underset{\tau \widehat{G}}{}n_\tau \tau $$
(5.15)
is the decomposition of $`T_M`$ in the sum of irreducible representations, then
$$T_M^{}\underset{\tau \widehat{G}}{}n_\tau \tau ^{}$$
(5.16)
is the decomposition of $`T_M^{}`$ in the sum of irreducible representations. The dimensions of the spaces of the invariants equal to multiplicities of the trivial representation in (5.15) and (5.16), which are the same since the trivial representation is self-dual. โ
Formulas (5.8) and (5.11) set two different isomorphisms between the set of coadjoint orbits $`๐ช`$ and the set of classes of equivalency of irreducible unitary representations $`\widehat{B}`$, depending on what regular representation of $`B`$ in $`C(B)`$ we use.
###### Definition 5.5.
Denote $`\tau (\mathrm{\Omega })`$ and $`\mathrm{\Omega }(\tau )`$ the irreducible unitary representation class and coadjoint orbit so that
$$V_{\mathrm{\Omega }(\tau )}=\mathrm{Iso}(\tau (\mathrm{\Omega }))n\tau (\mathrm{\Omega })$$
(5.17)
where $`n=dim\tau (\mathrm{\Omega })`$ and the regular representation of $`B`$ in $`[B]`$ used in (5.11) and (5.17), is defined as
$$R(g)x=xg^1.$$
(5.18)
We need to use the regular representation given by (5.18) to ensure for abelian $`B`$ for an orbit containing one character, correspondence to that character.
###### Corollary 5.6 (Dimension formula).
For a finite group $`B`$ of nilpotency class 2 of odd order, every coadjoint orbit $`\mathrm{\Omega }`$ of $`B`$ has $`n^2`$ elements where $`n=dim\tau (\mathrm{\Omega })`$. In other words,
$$dim\tau (\mathrm{\Omega })=\sqrt{\mathrm{\#}\mathrm{\Omega }(\tau )}.$$
(5.19)
###### Proof.
From (5.7),
$$dimV_\mathrm{\Omega }=\mathrm{\#}\mathrm{\Omega }.$$
(5.20)
Formula (5.19) follows directly from here and (5.17) โ
###### Lemma 5.7 (Stabilizer lemma).
For a group $`B`$ of nilpotency class 2 with 2-divisible center, for any coadjoint orbit $`\mathrm{\Omega }`$ and $`\chi _1,\chi _2\mathrm{\Omega }`$
$$\mathrm{Stab}\chi _1=\mathrm{Stab}\chi _2$$
(5.21)
and
$$\chi _1(b)=\chi _2(b)$$
(5.22)
for any $`b\mathrm{Stab}\chi _1=\mathrm{Stab}\chi _2`$ where $`\mathrm{Stab}\chi `$ denotes the stabilizer of $`\chi `$.
###### Proof.
If $`b\mathrm{Stab}\chi _1`$, then for all $`lL(B)`$
$$\chi _1(l)=\chi _1(l[b,l])=\chi _1(l)/\chi _1([b,l])$$
(5.23)
so $`\chi _1([b,l])=1`$. Since $`\chi _1`$ and $`\chi _2`$ are on the same orbit, there is such $`gB`$ that for all $`lL(B)`$
$$\chi _2(l)=\chi _1(l[g,l]).$$
(5.24)
Then
$$\chi _2([b,l])=\chi _1([b,l][g,[b,l]])=\chi _1([b,l])=1$$
(5.25)
so
$$\chi _2(l[b,l])=\chi _2(l)/\chi _2([b,l])=\chi _2(l)$$
(5.26)
so $`b\mathrm{Stab}\chi _2`$ and we proved (5.21). Now
$$\chi _2(b)=\chi _1(b[g,b])=\chi _1(b)\chi _1([b,g])=\chi _1(b)$$
(5.27)
###### Theorem 5.8 (Character formula).
For a finite group $`B`$ of nilpotency class 2 of odd order, for any $`bB`$ and $`\chi \mathrm{\Omega }`$
$$\mathrm{char}\tau (\mathrm{\Omega })(b)=\{\begin{array}{cc}n\chi (b)\hfill & \text{if }b\mathrm{Stab}\chi ,\hfill \\ 0\hfill & \text{otherwise},\hfill \end{array}$$
(5.28)
where $`n=dim\tau (\mathrm{\Omega })`$.
###### Proof.
Find the character of the restriction of regular representation (5.18) to $`V_\mathrm{\Omega }`$. From (5.10), if $`b/2\mathrm{Stab}\chi `$, then from Stabilizer Lemma 5.7, $`b`$ acts in $`V_\mathrm{\Omega }`$ by scalar multiplication on $`\chi (b)`$, so
$$\mathrm{char}n\tau (\mathrm{\Omega })(b)=dimV_\mathrm{\Omega }\chi (b)=n^2\chi (b)$$
(5.29)
Note that the cyclic subgroup generated by $`b`$ is of odd order, so it contains $`b/2`$, that means that $`b`$ and $`b/2`$ either belong to $`\mathrm{Stab}\chi `$, or not, simultaneously. Now, if $`b\mathrm{Stab}\chi `$, then again from (5.10) and from Stabilizer Lemma 5.13, all diagonal elements of the matrix of the action of $`b`$ in $`V_\mathrm{\Omega }`$ are zeroes, so
$$\mathrm{char}n\tau (\mathrm{\Omega })(b)=0$$
(5.30)
Dividing (5.29) and (5.30) by $`n`$, we get (5.28). โ
###### Corollary 5.9.
For a finite abelian group $`B`$ of odd order, for $`\mathrm{\Omega }=\{\chi \}`$,
$$\tau (\mathrm{\Omega })=\chi $$
(5.31)
###### Proof.
It follows from (5.30) for $`n=1`$ and $`\mathrm{Stab}\chi =B`$. โ
###### Corollary 5.10.
For a finite group $`B`$ of nilpotency class 2 of odd order,
$$\tau (\mathrm{\Omega })=\tau (\mathrm{\Omega })^{}.$$
(5.32)
###### Proof.
Again it follows directly from (5.28). โ
###### Acknowledgments.
I would like to thank Martin Isaacs, Wolfgang Kappe and Peter Morris for useful discussions. |
warning/0001/cond-mat0001043.html | ar5iv | text | # BARRIER CROSSING BY A LONG CHAIN MOLECULE - THE KINK MECHANISM
## I Introduction
The escape of a particle over a one dimensional barrier has been the subject of a large number of investigations. The problem, often referred to as the Kramers problem , has been the subject of detailed reviews . Kramers found solutions in the limit of weak friction and also in the limit of moderate to strong damping . The intermediate regime has been an active area of investigation . The reason for this extensive activity is that this forms a model for a chemical reaction occuring in a condensed medium. Kramers problem for few degrees of freedom has also been the topic of study . The quantum problem of escape/tunneling through a barrier too is of considerable interest. In the case where the system has an infinite number of degrees of freedom, this has been referred to as the decay of metastable vaccum, a problem that has attracted quite a bit of atttention in field theory, cosmology and mesoscopic quantum phenomena . In this paper, we consider a similar situation involving only classical physics. The trapped object has $`N`$ $`\left(\mathrm{}\right)`$ degrees of freedom, and is a polymer (a string). Though there are no quantum effects, the problem is similar, and some experiments are already available, that the results of the theory are easily verified. Further, the mathematics is considerably simpler than in the other cases, being equivalent to that of quantum mechanical tunneling of a single particle in a bistable potential.
The way that the $`N`$ degrees of freedom are connected (a chain or a string) leads to interesting new aspects to the problem that are not present in the case where there are only finite number of degrees of freedom. Also, the problem is of great interest in biology as many biological processes involve the translocation of a chain molecule from one side of a membrane to the other, through a pore in the membrane. The translocation of proteins from the cytosol into the endoplasmic reticulum, or into mitochondria or chloroplasts are processes of great interest and importance. Often, the proteins are hydrophilic and the pore in the membrane forms a hydrophobic region, through which it has to pass through , resulting in an increase in the free energy for the portion of the chain inside the pore. In infection by bacteriophages, conjugative DNA transfer etc, long chain DNA molecules snake through pores in membranes . In all these cases, the chain molecule seems to get across the membrane rather easily, contrary to the expectation that one gets from the theoretical analysis available in the literature on the subject (see below). Bezerukov et. al. have studied the partitioning of polymer molecule into a nanoscale pore. Chipot and Pohoille have carried out a molecular dynamics simulation of a polypeptide, translocating through the interface between hexane and water. They found that the polypeptide (undecamer of poly-L-leucine), initially placed in a random coil conformation on the aqueous side of the interface rapidly translocates to the interfacial region and then folds. In another interesting experiment, Han et. al. observed the forced movement of long, double stranded DNA molecules through microfabricated channels which have regions that present an entropic barrier for the entry of the molecules.
All these problems involve the passage of a long chain molecule, through a region in space, where the free energy per segment is higher, thus effectively presenting a barrier for the motion of the molecule. This problem forms the generalization that we refer to as the Kramers problem for a chain molecule. On the theoretical side, a variety of studies exist on this kind of problem. Muthukumar and Baumgartner studied the movement of self avoiding polymer molecules between periodic cubic cavities seperated by bottlenecks. The passage through the bottleneck presents an entropic barrier to the motion, and they show that it leads to an exponetial slowing down of diffusion with the number of segments $`N`$ in the chain. Baumgartner and Skolnick studied the movement of polymers through a membrane driven by an external bias and membrane asymmetry. Park and Sung , have studied the translocation through a pore. They analyze the passage through a pore on a flat membrane, with only the effects of entropy included. The resultant entropic barrier is rather broad, its width being proportional to $`N`$. Consequently, they consider the translocation process as being equivalent to the motion of the center of mass of the molecule. Using the result of the Rouse model that the diffusion coefficient of the center of mass is proportional to $`1/N`$, they effectively reduce the problem to the barrier crossing of single particle having a diffusion coefficient proportional to $`1/N`$. As the translocation involve motion of $`N`$ segments across the pore, the time taken to cross, $`t_{cross}`$ scales as $`N^3`$. They also show that in cases where there is adsorption on the trans side, translocation is favored and then $`t_{cross}`$ scales as $`N^2`$. In a very recent paper, Park and Sung have given a detailed investigation of the dynamics of a polymer surmounting a potential barrier. They use multidimensional barrier crossing theory to study the motion of a chain molecule over a barrier, in the limit where the width of the barrier is much larger than the lateral dimension of the molecule. In an interesting recent paper, Lubensky and Nelson study a case where they assume the interaction of the segments of the polymer with the pore to be strong. They argue that effectively, the dynamics of the portion of the chain inside the pore is the one that is important and this, they show, can give rise to $`t_{cross}`$ proportional to $`N`$. Again, they assume diffusive dynamics. In a recent paper, we have suggested a kink mechanism for the motion of the chain across a barrier and it is our aim to give details of this mechanism in this paper.
We consider a polymer undergoing activated crossing over a barrier. This can form a model for a polymer going through a pore too, as the pore can cause an increase in the free energy of the segments inside it, as they would interact with the walls of the pore. The width $`w`$ of the barrier is assumed to be much larger than the Kuhn length $`l`$ of the polymer, but small in comparison with the total length $`Nl`$ of the polymer. That is, $`l<<w<<Nl`$ . For example, in eukaryotic cells, the length of the nuclear pore is about 100 ร
, while the Kuhn length for a single stranded DNA is perhaps around 15 ร
. Therefore, one is justified in using a continuum approach to the dynamics of the long chain. (It is possible to retain the discrete approach, and develop the ideas based on them, but this is more involved mathematically). Our approach is the following: We describe the motion of the polymer using the Rouse model. The force that the barrier exerts on the chain appears as an additional, non-linear term in the model. We refer to this as the non-linear Rouse model. The non-linear term causes a distortion of the portion of the chain inside the barrier, which we refer to as the kink. Movement of the chain across the barrier is equivalent to the motion of the kink in the reverse direction. The kink is actually a special solution of the non-linear Rouse model, arising because of the non-linearity. In the presence of a free energy difference between the two sides, the kink moves with a definite velocity and hence the polymer would cross the barrier with $`t_{cross}`$ proportional to $`N`$. Traditionally, the non-linear models that one studies (for example, the $`\varphi ^4`$or the sine-Gordon model ) have potentials that are translationally invariant, and hence the kink can migrate freely in space. In comparison, in our problem, the non-linear term is fixed in position space and hence the kink too is fixed in space. However, the chain molecule (modelled as a string ) can move in space and hence the kink migrates, not in space, but on the chain. As far as we know, such a suggestion has never been made in the past and we believe that this is a very useful idea in understanding polymer translocation.
In general, the polymer can escape by essentially two mechanisms. The first, which we refer to as end crossing, involves the passage of one end of the polymer over the barrier, by thermal activation. This leads to the formation of the kink, which is then driven the free energy difference between the two sides of the barrier. The second is by the escape of any portion of the polymer over the barrier, in the form of a hairpin. The hairpin is a kink-antikink pair. For a flexible polymer, the hairpin crossing has twice the activation energy for end escape and hence one expects it to be less probable. However, as it can take place anywhere on the chain, the frequency factor for it is proportional to $`N`$ and hence for a sufficiently long chain, this can become the dominant mechanism for the escape. Hairpin crossing leads to the formation of a kink-antikink pair. The pair moves apart on the chain, driven by the free energy gain and hence the time of crossing is still proportional to $`N`$, though one expects that it is roughly half the time of crossing in the end crossing case. In principle, in addition to these, it is possible for more than one hairpin to be formed. However, it is obvious that in passage through a pore, unless the pore is rather wide, only the end-crossing mechanism would operate.
In all our mathematical development, we use the one dimensional version of the Rouse model. This is no limitation, if one is concerned with translocation across the interface between two immiscible liquids or the experiments of Han et. al., which involve motion in a channel, whose width is large in comparison with the size of the molecule. On the other hand, if one is interested in translocation through a pore, strictly speaking, one has to consider the full three dimensional nature of the problem, which at present seems rather involved. However, we believe that the one dimensional model captures the essential physics of the problem. Our analysis should also be useful in situtations where the whole of the polymer is in a pore, so that the dynamics may be taken to be one dimensional, with the chain trying to cross a region of high free energy.
## II The Model
### A The Free Energy Landscape
The considerations in this section are quite general and do not depend on the model that one uses to describe the polymer dynamics but involves the assumption that the polymer is flexible over a length scale comparable to the width of the barrier. We start by considering the free energy landscape for the crossing of the barrier. The barrier and the polymer stretched across it are shown in the figure 1. The polymer has initially all its units on the cis side, where its free energy per segment is taken to be zero. So the initial state, corresponds to a free energy of zero in the free energy hypersurface shown in figure 2. In crossing over to the trans side, it has to go over a barrier, as in the figure 1. The transition state for the crossing can be easily found, from physical considerations. It is the state shown in figure 3. In it, the configuration of the polymer is such that the end of the polymer on the trans side is located exactly at the point on the trans side at which its free energy per segment is zero, with the other end on the cis side, and the chain is such that the free energy of the whole chain is a minimum. This is so because if one moves the end either in the forward or in the backward direction (and the rest of the chain adjusted so that the free energy of the chain as a whole minimum), then the total free energy of the system would decrease. Hence in the free energy hypersurface figure 2, the configuration shown in the figure 3 corresponds to the maximum (i.e. transition state). Once the system has crossed the transition state, the chain is stretched across the barrier. The path of steepest descent then corresponds to moving segments from the cis side to the trans side, with out changing the configuration of the polymer in the barrier region. As there is a free energy difference $`\mathrm{\Delta }V`$ between the two sides, this would lead to a lowering of the free energy by $`\mathrm{\Delta }V`$ per segment, and this leads to a path on the free energy surface with a constant slope, and of width $`W`$ proportional to $`N`$ (see figure 3). Such a landscape implies that the translocation process would involve two steps. First step is going through the transition state by the overcoming of the activation barrier. Once the system has done this, it encounters a rather wide region of width proportional to the length of the chain. Crossing this region is the second step. As this region has a constant slope, the motion in this region is driven and it is similar to that of a Brownian particle subject to a constant force. Such a particle would take a time $`t_{cross}`$, proportional to $`N`$ to cross this region.
Till now, we considered the case of end-crossing. The scenario for hairpin crossing is similar. However, the activation energy is higher for hairpin crossing. In hairpin crossing, the transition state is equivalent to the one end crossing, repeated two times. Hence the activation energy for the process is two times larger. Once a hairpin crossing occur, a kink-antikink pair is formed and the kink and the anti-kink separate rapidly, due to the driving force of the free energy gain. Then futher crossing occurs by the movement of these two on the chain, which again leads to a time of crossing proportional to $`N`$.
In the following we make all these considerations quantitative, using the Rouse model to describe the dynamics of the chain.
### B The Dynamics
We consider the continuum limit of the Rouse model, discussed in detail by Doi and Edwards . The chain is approximated as a string, with segments (beads) labelled by their position $`n`$ along the chain. $`n`$ is taken to be a continuous variable, having values ranging from $`0`$ to $`N`$. The position of the $`n^{th}`$ segment in space is denoted by $`R(n,t)`$, where $`t`$ is time. In the Rouse model, this position undergoes overdamped Brownian motion and its time development is described by the equation
$$\zeta \frac{R(n,t)}{t}=m\frac{^2R(n,t)}{n^2}V^{}(R(n,t))+f(n,t).$$
(1)
In the above, $`\zeta `$ is a friction coefficient for the $`n^{th}`$ segment. The term $`m\frac{^2R(n,t)}{n^2}`$ comes from the fact that stretching the chain can lower its entropy and hence increase its free energy. Consequently, the parameter $`m=3k_BT/l^2`$ (see Doi and Edwards , equation (4.5). They use the symbol $`k`$ for the quantity that we call $`m`$) . As the ends of the string are free, the boundary conditions to be satisfied are $`\left\{\frac{R(n,t)}{n}\right\}_{n=0}=\left\{\frac{R(n,t)}{n}\right\}_{n=N}=0`$. $`V(R)`$ is the free energy of a segment of chain, located at the position $`R`$. We assume that $`V(R)`$ leads to a barrier located near $`R=0`$. $`f(n,t)`$ are random forces acting on the $`n^{th}`$ segment and have the correlation function $`f(n,t)f(n_1,t_1)=2\zeta k_BT\delta (nn_1)\delta (tt_1)`$(see , equation (4.12)). The deterministic part of the equation (1), which will play a key role in our analysis, is obtained by neglecting the random noise term in (1). It is:
$$\zeta \frac{R(n,t)}{t}=m\frac{^2R(n,t)}{n^2}V^{}(R(n,t))$$
(2)
This may also be written as:
$$\zeta \frac{R(n,t)}{t}=\frac{\delta E[R(n,t)]}{\delta R(n,t)}$$
(3)
where $`E[R(n,t)]`$ is the free energy functional for the chain given by:
$$E[R(n,t)]=_0^N๐n\left[\frac{m}{2}\left(\frac{R(n,t)}{n}\right)^2+V(R(n,t))\right]$$
(4)
### C The form of the barrier
The chain is assumed to be subject to a biased double well potential (BDW), of the form shown in the figure 1. The two minima are at $`a_0`$ and $`a_1`$, with $`a_0<a_1`$. There is assumed to be a maximum at $`R=0`$. Further, we take $`V(a_0)=0`$. All these conditions can be satisfied if one takes $`V^{}(R)=2kR\left(R+a_0\right)\left(Ra_1\right)`$. Here, $`k`$ is a constant and will determine the height of the barrier. Integrating this and using $`V(a_0)=0`$, we get
$$V(R)=\frac{k}{6}(R+a_0)^2(3R^22Ra_04Ra_1+a_0^2+2a_0a_1)$$
(5)
The barrier height for the forward crossing is $`V_f=V(0)V(a_0)=\frac{\mathrm{\hspace{0.17em}1}}{6}ka_0{}_{}{}^{3}(a_0+2a_1)`$ and for the reverse process, it is $`V_b=V(0)V(a_1)=\frac{1}{6}ka_1{}_{}{}^{3}(2a_0+a_1)`$. On crossing the barrier, a unit of the polymer lowers its free energy by $`\mathrm{\Delta }V=V(a_1)V(a_0)=\frac{1}{6}k\left(a_0a_1\right)\left(a_0+a_1\right)^3`$. The form of the potential is shown in the figure 3.
### D The Activation Free Energy for End and Hairpin Crossings
In this section, we consider the first step and calculate the activation free energy for both end and hairpin crossing. Activation free energy can be obtained from the free energy functional of equation (2). This free energy functional implies that at equilibrium, the probability distribution functional is $`\mathrm{exp}\left[\frac{1}{k_BT}๐n\left\{\frac{1}{2}m\left(\frac{dR}{dn}\right)^2+V(R(n))\right\}\right]`$. The configurations of the polymer which makes free energy a minimum are found from $`\frac{\delta E[R(n)]}{\delta R(n)}=0`$, which leads to the equation
$$m\frac{d^2R}{dn^2}=V^{}(R)$$
(6)
Notice that this is just a Newtonโs equation for a particle (ficticious, ofcourse) of mass $`m`$ moving in a potential $`V(R)`$. This equation has four solutions that are of interest to us. The first two are: (1) $`R(n)=a_0`$, (2) $`R(n)=a_1`$ which are the minima of the free energy. The first solution is the initial state, where the polymer is trapped in the vicinity of $`a_0`$. The second is the most stable minimum, at $`R(n)=a_1`$. In addition to these, there are two more solutions which are of interest to us. These are $`n`$ dependent and correspond to end and hairpin crossings.
#### 1 End Crossing
As we are interested in the case where the polymer is very long, we can imagine $`n`$ to vary from $`\mathrm{}`$ to $`0`$ and find a saddle point in the free energy surface by searching for a solution satisfying $`R(\mathrm{})=a_0`$ and the other end of the polymer to be at a point with $`R>R_{\mathrm{max}}`$, where $`R_{\mathrm{max}}`$ is the point where $`V(R)`$ has its maximum value. For the Newtonโs equation (6) the conserved energy is $`E_c=\frac{1}{2}m\left(\frac{dR}{dn}\right)^2V(R(n))`$. For the extremum path, $`E_c=0`$. Thus, the particle starts at $`R(\mathrm{})=a_0`$ with the velocity zero (this follows from the boundary conditons of the Rouse model) and ends up at $`R_f`$ at the โtimeโ $`n=0`$. Here $`R_f(>R_{\mathrm{max}}),`$ is the point such that $`V(R_f)=0`$, again with the velocity zero. Further, free energy of this configuration is activation free energy for end crossing. As for this configuration, $`\frac{1}{2}m\left(\frac{dR}{dn}\right)^2=V(R(n))`$, we find the activation free energy to be given by
$$E_{a,end}=_{a_0}^{R_f}\sqrt{2mV(R)}๐R.$$
(7)
The end crossing is illustrated in figure 4.
#### 2 Hairpin Crossing
If one imagines $`n`$ to vary in the range $`(\mathrm{},\mathrm{})`$ a second saddle point may be found by taking $`R(\mathrm{})=a_0`$ and $`R(\mathrm{})=a_0`$, so that the Newtonian particle starts at $`a_0`$, makes a round trip in the inverted potential $`V(R)`$ and gets back to its starting point. This obviously has an activation energy
$$E_{a,hp}=2_{a_0}^{R_f}\sqrt{2mV(R)}๐R=2E_{a,end}$$
(8)
Thus the activation energy is exactly two times for end crossing. The hairpin crossing is shown in figure 5
#### 3 The Temperature dependence
As the parameter $`m`$ is proportional to the temperature ( $`=3k_BT/l^2`$ ), we arrive at the general conclusion that both the activation energies $`E_{a,end}`$ and $`E_{a,hp}`$ are proportional to $`\sqrt{T}`$. For our model potential of equation (5) we find $`R_f=a_0(\gamma \sqrt{\gamma ^2\gamma })`$ where $`\gamma =(1+2\frac{a_1}{a_0})\frac{1}{3}`$ and
$$E_{a,end}=\frac{\sqrt{mk}a_0^3}{6}\left[(3\gamma ^2+1)\sqrt{1+3\gamma }3\gamma (\gamma ^21)\mathrm{ln}\left(\sqrt{\gamma (\gamma 1)}/\left(1+\gamma \sqrt{1+3\gamma }\right)\right)\right].$$
(9)
The Boltzmann factor $`e^{\frac{E_{act}}{k_BT}}`$ for the crossing of one end of the polymer over the barrier thus has the form $`e^{\text{constant}/\sqrt{T}}`$. Further, we find that it is independent of $`N`$ for large $`N`$.
## III The Rate of Crossing
### A Hairpin Crossing
We now calculate the rate of crossing in the two cases. We first consider the hairpin crossing, as this has connections with material available in the literature . The methods that we use are quite well known in the soliton literature and hence we give just enough details to make the approach clear. The Rouse model in the equation (1) leads to the functional Fokker Planck equation
$$\frac{P}{t}=\frac{1}{\zeta }_0^N๐n\frac{\delta }{\delta R(n)}\left[k_BT\frac{\delta P}{\delta R(n)}+\frac{\delta E[R(n)]}{\delta R(n)}P\right]$$
(10)
for the probability distribution functional $`P`$. This equation implies that the flux associated with the co-ordinate $`R(n)`$ is
$$j(R(n))=\frac{1}{\zeta }\left[k_BT\frac{\delta P}{\delta R(n)}+\frac{\delta E[R(n)]}{\delta R(n)}P\right]$$
(11)
We now consider the initial, metastable state. As the rate of escape is small, we can assume the probability distribution to be the equilibrium one, which is
$$P=\frac{1}{Z_0}\mathrm{exp}\left\{E[R(n)]/k_BT\right\}$$
(12)
To determine $`Z_0`$ we use the condition $`D[R(n)]`$ $`P=1`$, where $`D[R(n)]`$ stands for functional integration. It is convenient to introduce the normal co-ordinates for small amplitude motion around the metastable minimum and do the functional integration using them. For this, we expand $`E[R(n)]`$ around the metastable minimum, by putting $`R(n)=a_0+\delta R(n)`$, and expanding as a functional Taylor series in $`\delta R(n)`$ and keeping terms up to second order in $`\delta R(n)`$. Then
$$E[R(n)]=\frac{1}{2}m_0^N\delta R(n)\left(\frac{^2}{n^2}+\omega _0^2\right)\delta R(n)$$
(13)
We have defined $`\omega _0`$ by putting $`m\omega _0^2=\left[\frac{^2V(R)}{R^2}\right]_{R=a_0}`$. The normal (Rouse) modes are just the eigenfunctions $`\psi _k(n)`$ of the operator $`\widehat{H}^{ms}=\left(\frac{^2}{n^2}+\omega _0^2\right)`$, having the eigenvalue $`\epsilon _k`$ and satisfying the Rouse boundary conditions $`\frac{\psi _k(n)}{n}=0`$ at the two ends of the string. (The superscript โmsโ in $`\widehat{H}^{ms}`$s$`\mathrm{tan}`$ds for metastable). Now we can expand $`\delta R(n)`$ as $`\delta R(n)=_kc_k\psi _k(n)`$ so that the expression for energy (13) becomes
$$E[R(n)]=\frac{1}{2}m\underset{k}{}\epsilon _kc_k^2$$
(14)
We now do the functional integration using the variables $`c_k`$. Then the normalization condition $`D[R(n)]`$ $`P=1`$ becomes $`\frac{1}{Z_0}\underset{k}{}`$ $`๐c_k\mathrm{exp}\left[\frac{1}{2}m\beta \epsilon _kc_k^2\right]=1`$. This leads to $`Z_0=\underset{k}{}\left(\frac{2\pi }{m\beta \epsilon _k}\right)^{1/2}`$. Now we consider the vicinity of a saddle point, where the probability distribution deviates from the equilibrium one. We first consider the saddle point which corresponds to hairpin crossing. The potential of the equation (5) is rather difficult to handle as we have not been able to obtain analytic solutions to the Newtonโs equation (6). In determining the crossing of the barrier, the key role is played by the quantities $`\omega _0`$ and the height of the barrier for crossing in the forward direction $`V_f`$. The quantities that we calculate in this section have no dependence of the behavior of the potential near the stable minimum. So, instead of using the quartic potential of the equation (5), we use the simpler cubic potential of equation (15). This has no stable minimum (corresponding to the final state), but that does not matter, because the quantities that we calculate do not depend on its existence. Thus we use the potential:
$$V_c(R)=V_0\left(\frac{R+a_0}{R_0}\right)^2\left(1\frac{R+a_0}{R_0}\right)$$
(15)
where we adjust $`V_0`$ and $`R_0`$ to reproduce the values for $`\omega _0`$ and the barrier height $`V_f`$. Solving the equation (6) for this potential, in the limit of an infinitely long chain extending from $`n=\mathrm{}`$ to $`+\mathrm{}`$, the saddle point that corresponds to hairpin crossing is easily found to be given by the equation
$$R_{hp}(n)=a_0+R_0\left\{\mathrm{sec}h\left(\sqrt{\frac{V_0}{2m}}n\right)\right\}^2$$
(16)
In fact one has a continuous family of solutions of the form $`R_{hp}(nn_0)`$, where $`n_0(\mathrm{},\mathrm{})`$ is arbitrary and determines center of the kink-antikink pair. Now expanding the energy $`E[R(n)]`$ about this saddle, by writing $`R(n)=R_{hp}(nn_0)+`$ $`\delta R(n)`$ we get
$$E[R(n)]=E_{a,hp}+\frac{1}{2}m๐n\delta R(n)\left[\frac{^2}{n^2}+\omega _0^2\left\{13\mathrm{sec}h^2\left(\omega _0(nn_0)/2\right)\right\}\right]\delta R(n)$$
(17)
For the potential of equation (15) $`E_{a,hp}=\left(8R_0/15\right)\sqrt{2mV_0}`$. The normal modes for fluctuations around the saddle are determined by the eigenfunctions of the operator $`\widehat{H}^{}=\frac{^2}{n^2}+\omega _0^2\left\{13sech^2\left(\omega _0(nn_0)/2\right)\right\}.`$ ($``$ is used to denote the saddle point). The eigenfunctions are: (a) the discrete states $`\psi _0^{}`$, $`\psi _1^{}`$ and $`\psi _2^{}`$ having the eigenvalues $`\epsilon _0^{}=5\omega _0^2/4`$, $`\epsilon _1^{}=0`$ and $`\epsilon _2^{}=3\omega _0^2/4`$ and (b) the continuum of eigenstates with eigenvalues of the form $`\epsilon _k^{}=\omega _0^2+k^2`$(more details are given in appendix A). We denote the eigenfunctions by $`\psi _k^{}`$. The existence of the eigenvalue $`\epsilon _1^{}=0`$ comes from the freedom of the kink-antikink pair to have its center anywhere on the chain (the hairpin can be formed anywhere). In the following, $`\underset{k}{}`$ would stand for summation over all the eigenstates, including both the discrete and continuum states while a symbol like $`\underset{k1}{}`$ means that the bound state $`\psi _1^{}`$ is to be excluded from the sum. Now writing $`\delta R(n)=\underset{k1}{}c_k^{}\psi _k^{}`$, we get
$$E[R(n)]=E_{a,hp}+\frac{1}{2}m\underset{k1}{}\epsilon _k^{}\left(c_k^{}\right)^2$$
We write the probability density near the saddle as
$$P=\frac{\theta (c_0^{},c_1^{}\mathrm{})}{Z_0}\mathrm{exp}\left\{\frac{E[R(n)]}{k_BT}\right\}$$
(18)
where $`\theta (c_0^{},c_1^{}\mathrm{})`$, is a function that must approach unity in the vicinity of the metastable minimum. Near the saddle, one can calculate the flux $`j_k^{}`$ in the direction of $`c_k^{}`$.
$$j_k^{}=\frac{1}{\zeta }\left[k_BT\frac{P}{c_k^{}}+\frac{E[R(n)]}{c_k^{}}P\right]$$
Using the equations (17 ) and (18 ) we get
$$j_k^{}=\frac{k_BT}{Z_0\zeta }\frac{\theta (c_0^{},c_1^{}\mathrm{})}{c_k^{}}\mathrm{exp}\left\{\frac{1}{k_BT}\left(E_{a,hp}+\frac{1}{2}m\underset{k1}{}\epsilon _k^{}\left(c_k^{}\right)^2\right)\right\}$$
(19)
In a steady state, there is flux only in the unstable direction. That is, only $`j_0^{}`$ is non-zero. This means that $`\theta `$ can depend only on $`c_0^{}`$, which impies that $`j_0^{}`$ must have the form
$$j_0^{}=A\mathrm{exp}\left\{\frac{1}{k_BT}\left(\frac{1}{2}m\underset{k>1}{}\epsilon _k^{}\left(c_k^{}\right)^2\right)\right\}$$
(20)
where $`A`$ is a constant, to be determined. Using the equation (20) in (19) we get $`\frac{\theta (c_0^{})}{c_0^{}}=A\mathrm{exp}\left\{\frac{m}{2k_BT}\left|\epsilon _0^{}\right|\left(c_0^{}\right)^2\right\}`$. The fact that $`\theta (c_0^{})`$ must approach unity as $`c_0^{}\mathrm{}`$, enables one to get $`A=\left(\frac{m\left|\epsilon _0^{}\right|}{2\pi k_BT}\right)^{1/2}`$ . Hence $`\theta (c_0^{})=\left(\frac{m\left|\epsilon _0^{}\right|}{2\pi k_BT}\right)^{1/2}_{c_0^{}}^{\mathrm{}}๐z\mathrm{exp}\left\{\frac{1}{2k_BT}m\left|\epsilon _0^{}\right|z^2\right\}.`$ Now the net flux crossing the barrier is found by integrating $`j_0^{}`$ over all directions other than $`c_0^{}`$ . The integrals over all $`c_k^{}`$, except $`c_1^{}`$ is straightforward. As $`\epsilon _1^{}=0`$, $`๐c_1^{}`$ needs special handling. The integral, as is well-known, is performed by converting it to an integral over the kink-antikink position, $`n_0`$. That is, $`๐c_1^{}=\alpha ๐n_0,`$ where $`\alpha ^2=_{\mathrm{}}^{\mathrm{}}๐n\left(\frac{R_{hp}(n)}{n}\right)^2=\frac{E_{a,hp}}{m}`$. Hence the rate becomes
$$k_{hp}=\frac{k_BT}{Z_0\zeta }\left(\frac{m\left|\epsilon _0^{}\right|}{2\pi k_BT}\right)^{1/2}\underset{k>1}{}\left(\frac{2\pi k_BT}{m\left|\epsilon _k^{}\right|}\right)^{1/2}\left(\frac{E_{a,hp}}{m}\right)^{1/2}N\mathrm{exp}\left(E_{a,hp}/k_BT\right)$$
(21)
The notation $`\underset{k>1}{}`$ is used to indicate product over all eigenvalues of $`\widehat{H}^{}`$, except the first two. On using the expression for $`Z_0`$,
$$k_{hp}=\frac{k_BT}{\zeta }\left(\frac{m}{2\pi k_BT}\right)^{3/2}I_{hp}\left(\frac{\left|\epsilon _0^{}\right|E_{a,hp}}{\left|\epsilon _2^{}\right|m}\right)^{1/2}N\mathrm{exp}\left(E_{a,hp}/k_BT\right)$$
(22)
where $`I_{hp}=\left(\frac{\underset{k}{}\epsilon _k}{\underset{k>2}{}\epsilon _k^{}}\right)^{1/2}`$. This infinite product is evaluated in the appendix B and is found to be $`I_{hp}=\frac{15}{2}\omega _0^3`$. This leads to
$$k_{hp}=\frac{5Nm\omega _0^3}{4\pi \zeta }\left(\frac{15E_{a,hp}}{2\pi k_BT}\right)^{1/2}\mathrm{exp}\left(E_{a,hp}/k_BT\right)$$
(23)
### B End Crossing
In this case, the analysis is similar to the above. The operator $`\widehat{H}^{}`$ is the same as earlier. However, there is an interesting difference. In the hairpin case, the boundary conditions on $`\psi _k^{}`$ ( $`\frac{d\psi _k^{}}{dn}=0`$, at the two ends) were at $`n=\pm \mathrm{}`$, while in this case, they are at $`n=0`$ and at $`n=\mathrm{}`$ (i.e. the boundary value problem is now on the half-line). Due to this, one has to rule out the odd $`\psi _k^{}`$ that exists in the hairpin case as they do not satisfy the Rouse boundary condition$`\frac{d\psi _k^{}}{dn}=0`$ at $`n=0`$. So we consider only the even solutions. Thus the eigenvalue at zero is ruled out (which is quite alright as end crossing can occur only at the end and not anywhere else, but we will put in additional factor of 2 as it can occur at the two ends). The discrete spectrum now has only the eigenvalues $`\epsilon _0^{}=5\omega _0^2/4`$, and $`\epsilon _2^{}=3\omega _0^2/4`$. The expression for the rate is
$$k_{end}=\frac{k_BT}{\zeta }\left(\frac{m\left|\epsilon _0^{}\right|}{2\pi k_BT}\right)^{1/2}\stackrel{~}{I}_{end}\mathrm{exp}\left(E_{a,end}/k_BT\right)$$
(24)
where $`\stackrel{~}{I}_{end}=\frac{\underset{k0}{}\left(\frac{2\pi k_BT}{m\epsilon _k^{}}\right)^{1/2}}{\underset{k}{}\left(\frac{2\pi k_BT}{m\epsilon _k}\right)^{1/2}}`$. In this product, there are $`N1`$ terms in the numerator and $`N`$ terms in the denominator. One of the $`N1`$ terms is the bound state with an eigenvalue $`\epsilon _2^{}=3\omega _0^2/4`$. Separating this out from the product, one can write $`\stackrel{~}{I}_{end}=\left(\frac{2m}{3\pi k_BT\omega _0^2}\right)^{1/2}I_{end}`$, where $`I_{end}=\left(\frac{\underset{k}{}\epsilon _k}{\underset{k>2}{}\epsilon _k^{}}\right)^{1/2}`$. the evaluation of this product involves some subtelity and is done in the Appendix B. The result is
$$k_{end}=\frac{5m\omega _0^2}{2\sqrt{2}\pi \zeta }\mathrm{exp}\left(E_{a,end}/k_BT\right)$$
Accounting for the existence of two ends leads to
$$k_{twoends}=\frac{5m\omega _0^2}{\sqrt{2}\pi \zeta }\mathrm{exp}\left(E_{a,end}/k_BT\right)$$
(25)
## IV The kink and its motion
### A The kink solution and its velocity
Having overcome the activation barrier, how much time would the polymer take to cross it? We denote this time by $`t_{cross}`$. To calculate this, we first look at the mathematical solutions of the deterministic equation (2). The simplest solutions of this equation are: $`R(n,t)=a_0`$ and $`R(n,t)=a_1`$. These correspond to the polymer being on either side of the barrier and these are just mean values of the position on the two sides. Thermal noise makes $`R(n,t)`$ fluctuate about the mean position which may be analyzed using the normal co-ordinates for fluctuations about this mean position. Each normal mode obeys a Langevin equation similar to that for a harmonic oscillator, executing Brownian motion. In addition to these two time independent solutions, the above equation has a time dependent solution (a kink) too, which corresponds to the polymer crossing the barrier. We analyze the dynamics of the chain, with the kink in it, using the normal modes for fluctuations about this kink configuration. Our analysis makes use of the techniques that have been used to study the diffusion of solitons
As is usual in the theory of non-linear wave equations, a kink solution moving with a velocity $`v`$ may be found using the ansatz $`R(n,t)=R_s(\tau )`$ where $`\tau =nvt`$ . Then the equation (2) reduces to
$$m\frac{d^2R_s}{d\tau ^2}+v\zeta \frac{dR_s}{d\tau }=V^{}(R_s).$$
(26)
If one imagines $`\tau `$ as time, then this too is a simple Newtonian equation for the motion of particle of mass $`m`$, moving in the upside down potential $`V(R)`$. However, in this case, there is a frictional term too, and $`v\zeta /m`$ is the coefficient of friction. This term makes it possible for us to find a solution for quite general forms of potential, with $`V^{}(R)0`$ as $`R\pm \mathrm{}`$. For the potential of the equation (6), we can easily find a solution of this equation, obeying the conditions $`R_s(\tau )=a_0`$ for $`\tau \mathrm{}`$ and $`R_s(\tau )=a_1`$ for $`\tau \mathrm{}.`$ The solution is
$$R_s(\tau )=\left(a_0+e^{\tau \omega \left(a_0+a_1\right)}a_1\right)\left(1+e^{\tau \omega \left(a_0+a_1\right)}\right)^1,$$
(27)
with $`\omega =\sqrt{k/m}`$ . The solution exists only if the velocity $`v=\frac{\sqrt{mk}}{\zeta }(a_0a_1).`$ This solution is a kink, occurring in the portion of the chain inside the barrier. We shall refer to the point with $`\tau =0`$ as the center of the kink. (Actually one has a one-parameter family of solutions of the form $`R_s(\tau +\tau _0)`$, where $`\tau _0`$ is any arbitrary contant). As $`\tau =nvt`$, the center of the kink moves with a constant velocity $`v`$. Note that this velocity depends on the shape of the barrier. Thus for our model potential, if $`a_0<a_1`$, then $`V_f<V_b`$, and this velocity is negative. This implies that the kink is moving in the negative direction, which corresponds to the chain moving in the positive direction. That is, the chain moves to the lower free energy region, with this velocity. If the barrier is symmetric, then $`a_0=a_1`$( $`V_f=V_b`$) the velocity of the kink is zero.
### B Fluctuations about the kink
We now analyze the effect of the noise term present in the equation (1). The center of the kink can be anywhere on the chain - which means that the kink is free to move on the chain. Actually, as the position of the kink is fixed in space, this means that the polymer is moving across the barrier. The kink would also execute Brownian motion, due to the noise term. The motion of the kink caused by the noise terms is a well studied problem in the literature and one can make use of these methods. Following โInstanton methodsโ of field theory , we write
$$R(n,t)=R_s(na(t))+\underset{p=1}{\overset{\mathrm{}}{}}X_p(t)\varphi _p(na(t),t)$$
(28)
We have allowed for the motion of the kink by taking the kink center to be at $`a(t)`$, where $`a(t)`$ is a random function of time which is to be determined. $`\varphi _p`$ are a set of functions (the Rouse modes) below and $`X_p(t)`$ are expansion coefficients. This may be put into the equation (1) to derive an equation of motion for $`a(t).`$ Neglecting kink-phonon scattering leads to
$$\stackrel{}{a}(t)=v+\xi _0(t)/C$$
(29)
where we define $`\psi _0(n)`$ by $`_nR_s(n)=C\psi _0^{}(n)`$ with
$$C^2=_nR_s(n)\left|e^{v\zeta \overline{n}/m}\right|_nR_s(n)=\frac{2}{3}\pi \omega \mathrm{csc}(2\pi \frac{a_1a_0}{a_0+a_1})\left(a_1a_0\right)a_0a_1.$$
(30)
and
$$\xi _0(t)=\frac{1}{\zeta }_{N/2}^{N/2}๐n\psi _0^{}(n)e^{v\zeta \overline{n}/(2m)}f(n+a(t),t).$$
(31)
$`\xi _0(t)`$ is a random function of time, having the correlation function
$$\xi _0(t)\xi _0(t_1)=\delta (tt_1)(2k_BT/\zeta )_{over\text{ }the\text{ }kink}๐ne^{vn\zeta /m}\left[\psi _0(n)\right]^2$$
(32)
.For the potential given by the equation (5) one gets
$$\xi _0(t)\xi _0(t_1)=\delta (tt_1)k_BT/(2\zeta a_0a_1)\mathrm{sec}(2\pi \frac{a_1a_0}{a_0+a_1})\left(3a_1a_0\right)\left(3a_0a_1\right).$$
(33)
The equations (29) and (30) imply that the kink position $`a(t)`$ executes Brownian motion with drift. As $`v`$ is negative, the drift is in the negative direction.
### C The crossing time $`t_{cross}`$
For the polymer to cross the barrier, the kink has to go in the reverse direction, by a distance equal to $`N`$. As the equation (29) is just that for a particle executing Brownian motion with drift, we can estimate the time of crossing as a first passage time. As the kink starts at one end, we take the initial position of the particle, $`a`$ to be $`N`$ and calculate the average time required for it to attain the value $`0`$, which would correspond to the polymer crossing the barrier fully. Writing the diffusion equation for the survival probability $`P(a,t)`$ for a particle starting at $`a=N`$ at the time $`t=0`$ and being absorbed at $`a=0`$, we get
$$\frac{P(a,t)}{t}=D\frac{^2P(a,t)}{a^2}v\frac{P(a,t)}{a}$$
(34)
. Here, the diffusion coefficient
$$D=\frac{1}{2tC^2}_0^t๐t_1_0^t๐t_2\xi _0(t_1)\xi _0(t_2)$$
$$=\frac{3k_BT}{8\pi \zeta }\sqrt{\frac{m}{k}}\frac{\left(3a_1a_0\right)\left(3a_0a_1\right)}{a_0{}_{}{}^{2}a_{1}^{}{}_{}{}^{2}(a_1a_0)}\mathrm{tan}(2\pi \frac{a_1a_0}{a_0+a_1}).$$
(35)
The equation (34) is to be solved, subject to the initial condition $`P(a,0)=\delta (aN)`$ and with absorbing boundary condition at $`a=0`$ (i.e. $`P(0,t)=0`$) and $`P(\mathrm{},t)=0`$. It is easy to solve the above equation in the Laplace domain. The result for the Laplace transform $`\overline{P}(a,s)=_0^{\mathrm{}}๐tP(a,t)\mathrm{exp}(st)`$ is:
$$\overline{P}(a,s)=\frac{1}{\sqrt{4Ds+v^2}}\left[e^{\frac{\left(aN\right)v\sqrt{4Ds+v^2}\left|aN\right|}{2D}}e^{\frac{\left(aN\right)v\sqrt{4Ds+v^2}\left|a\right|\sqrt{4Ds+v^2}N}{2D}}\right]$$
(36)
The Laplace transform of the survival probability is given by $`\overline{P}(s)=_{\mathrm{}}^{\mathrm{}}๐a\overline{P}(a,s)`$ and is found to be
$$\overline{P}(s)=\frac{1}{s}\left[1e^{\frac{Nv\sqrt{4Ds+v^2}N}{2D}}\right]$$
(37)
The average crossing time is given by $`t_{cross}=Limit_{s>0}\overline{P}(s)=N/(v)`$, if $`v<0`$. As $`v`$ is proportional $`\sqrt{mk}`$, assuming $`V(R)`$ to be temperature independent we find $`t_{cross}N/\sqrt{T}`$. This is a general conclusion, independent of the model that we assume for the potential. If the barrier is symmetric, the kink moves with an average velocity $`v=0`$. Taking the $`v0`$ limit of $`\overline{P}(s)`$, we get
$$\overline{P}(s)=\frac{1}{s}\left(1e^{\frac{\sqrt{s}N}{\sqrt{D}}}\right)$$
(38)
so that the survival probability becomes
$$P(t)=Erf(\frac{N}{2\sqrt{Dt}}).$$
(39)
This expression for the survival probability implies that the average time that the particle survives is $`t_{cross}N^2/D.`$ For the symmetric barrier, the value of $`D`$ may be obtained by taking the limit $`a_1a_0`$, and one finds $`D=\frac{3k_BT}{4\zeta a_0^3}\sqrt{\frac{m}{k}}`$ and thus $`t_{cross}N^2/T^{3/2}`$.
In their analysis, Park and Sung considered the passage of a polymer through a pore for which the barrier is entropic in origin. Consequently it is very broad, the width being of the order of $`N`$. Hence they consider the movement as effectively that of the center of mass of the polymer which diffuses with a coefficient proportional to $`1/N`$. As the center of mass has to cover a distance $`N`$, the time that it takes is proportional to $`N^3`$. If there is a free energy difference driving the chain from one side to the other, then the time is proportional to $`N^2`$. In comparison, we take the barrier to be extrinsic in origin and assume its width to be small in comparison with the length of the chain. The crossing occurs by the motion of the kink, which is a localized non-linear object in the chain whose width is of the same order as that of the barrier. As the polymer is intially subject to a potential well, the entropic contribution to the barrier that Park and Sung consider does not exist in our case. Such a potential is realistic, in cases where the polymer is subjected to a driving force (for example an electric field). As the kink is a localized object, its diffusion coefficient has no $`N`$ dependence and our results are different from those of Park and Sung . In the case where there is no free energy difference, our crossing time is proportional to $`N^2`$(in contrast to $`N^3`$ of Park and Sung) , while if there is a free energy difference, our crossing time is proportional to $`N`$ (in contrast to $`N^2`$ of Park and Sung). In a very recent paper , Park and Sung have considered the Rouse dynamics of a short polymer surmounting a barrier. The size of the polymer is assumed to be small in comparison with the width of the potential barrier. Consequently, the transition state has almost all the units at the top of the barrier, leading to the prediction that the activation energy is proportional to $`N`$. This leads to a crossing probability that decreases exponentially with $`N`$. In comparison, as found in section II D, the free energy of activation does not depend on the length of the chain. Hence, the mechanism is the favoured one for long chains.
### D The net rate
As the actual crossing is a two step process, with activation as the first step and kink motion as the second step, the net rate of the two has to be a harmonic mean of the two rates. For a very long chain, the motion of the kink has to become rate determining. In the case of translocation of biological macromolecules, considered in section V there does not seem to be any free energy of activation and then the rate is determined by $`t_{cross}`$ alone. Recently, the motion of long chains in microfabricated channels have been investigated by Han et al . In contrast to the situtation for a pore, there is an additional direction is available for the molecule to form a hairpin, viz. perpendicular to the direction of movement of the molecule. Consequently, in overcoming the barrier, both end crossing and hairpin crossing can occur (see figures 4 and 5). Experimental results show that the longer molecule crosses the barrier faster. This means that the $`N`$-dependence of $`k_{hp}`$ causes the hairpin crossing to be the dominant mechanism of crossing in these experiments.
## V How do biological systems lower the activation energy?
If there was a high activation energy ($`>>k_BT`$) for the translocation, the process would be unlikely and hence, biological systems would not be able to function, if they depended crucially on such transfers. As translocation seem to be very efficient in biological systems, one needs to look at the mechanism that evolution has designed to reduce the barrier. The destination (referred to as sorting) of a biological long chain molecule is determined by a sequence of units at the begining of the chain, referred to as the signal sequence. For example, proteins destined to the endoplasmic reticulum possess an amino-terminal signal sequence, while those destined to remain in the cytosol do not have this. If one attaches this sequence to a cytosolic protein, then the protein is found to end up in the endoplasmic reticulum (see reference , figure 14.6). The way the sequence works is simple. If the pore is hydrophobic and the chain hydrophilic, then the signal sequence is hydrophobic, so that the signal sequence has a low free energy inside the pore.
We qualitatively analyze this type of problem in the following, using the Rouse model. The way to model the situation would be to have a potential that is dependent upon the segment number $`n`$ in the chain. Hence, in the equations of the Rouse model the potential term would have an explicit dependence on $`n`$. Let us denote the length of the signal sequence by $`s`$. The simplest model would be to have a potential which is attractive, for $`0<n<s`$ and which has the shape of a barrier for $`s<n<N`$. The transition state is determined by the the Newton-like equation
$$m\frac{d^2R}{dn^2}=V_{new}^{}(n,R),$$
(40)
with $`n`$ playing the role of time (in the following we shall refer to $`n`$ as the time for the motion of this ficticious particle). We take the potential to be such that
$$V_{new}(n,R)=V(R)\text{ if }0<n<s\text{ and }$$
$$=V(R)\text{ if }s<n<N\text{ }$$
This corresponds to a particle moving in a time dependent potential, which switches from being repulsive to attractive at the time $`s`$. The shape of this time dependent potential is shown in the figure 8. The boundary conditions $`\left\{\frac{dR(n)}{dn}\right\}_{n=0}=\left\{\frac{dR(n)}{dn}\right\}_{n=N}=0`$ imply that the particle has to start and end with zero velocity. Let us imagine that the particle starts at the point $`R_0`$ (see figure 8). As the potential that it feels up to the time $`s`$ is repulsive, it follows the path indicated by the dashed line in the figure, and the conservation of energy may be written as: $`\frac{1}{2}m\left(\frac{dR(n)}{dn}\right)^2+V(R)=V(R_0)`$. Let it reach the point $`R_s`$ after a time $`s`$. At this time, the potential is switched from $`V(R)`$ to $`V(R)`$. From this time on, the equation of conservation of energy would be:
$$\frac{1}{2}m\left(\frac{dR(n)}{dn}\right)^2V(R)=V(R_0)2V(R_s).$$
(41)
This is the equation of motion of the particle for $`s<n<N`$. We are interested in $`N\mathrm{}`$ limit and we have to satisfy the boundary condition $`\left\{\frac{dR(n)}{dn}\right\}_{n=N}=0`$ at the end of the chain. In the particle picture, this is equivalent to the condition that the total energy of the particle obeying the equation 40 must be zero. This implies that $`V(R_0)=2V(R_s)`$. For a given $`s`$, this uniquely fixes the values of the two variables $`R_0`$ and $`R_s`$.
The net transition state is shaped like a hook and the hydrophobic part of the chain is completely in the short arm of the hook (see figure 9) . A configuration like the one in the figure 10 where the whole of the hook is formed by the hydrophobic part is not a transition state. The transition state in figure 9, though it seems likely to occur in crossing between liquid liquid interfaces, it seems rather difficult to form in the case of passage through a pore as there are two difficulties: (1) the chain has to bend to form the hook (2) the pore has to be wide enough to accommodate the two strands of the hook simultaneously. Inspite of these, nature does seem to use this as an inspection of the figure 14-14 of reference shows.
## VI Conclusions
We have considered the generalization of the Kramers escape over a barrier problem to the case of a long chain molecule. It involves the motion of chain molecule of $`N`$ segments across a region where the free energy per segment is higher, so that it has to cross a barrier. We consider the limit where the width of the barrier $`w`$ is large in comparison with the Kuhn length $`l`$, but small in comparison with the total length $`Nl`$ of the molecule. The limit where $`Nl<<w`$ has been considered in a recent paper by Park and Sung . We use the Rouse model and find there are two possible mechanism that can be important - end crossing and hairpin crossing. We calculate the free energy of activation for both and show that both have a square root dependence on the temperature $`T`$, leading to a non-Arrhenius form for the rate. We also find that the activation energy for hairpin crossing is two times the activation energy for end-crossing. Inspite of this, for long enough chains, where the geometry of the systems permits, hairpin formation can be the dominant mode of escape as seen in the experiments of Han et. al.. The width of the barrier in these experiments is rather large in comparison with the length of the polyme so that the kink mechanism of crossing seems to be unlikely in this case.
While in the short chain limit Park and Sung find the activation energy to be linearly dependent on $`N`$, we find that for long chains, the activation energy is independent of $`N`$. We also show that there is a special time dependent solution of the model, which corresponds to a kink in the chain, confined to the region of the barrier. In usual non-linear problems with a kink solution, the problem has translational invariance and the soliton/kink can therefore migrate. In our problem, the translational invariance is not there, due to the presence of the barrier and the kink solution is not free to move in space. However, the polymer on which the kink exists, can move, though the kink is fixed in space. Thus, the polymer goes from one side to the other by the motion of the kink in the reverse direction on the chain. If there is no free energy difference between the two sides of the barrier, then the kink moves by diffusion and the time of crossing $`t_{cross}N^2/T^{3/2}`$. If there is a free energy difference, then the kink moves with a non-zero velocity from the lower free energy side to the other, leading to $`t_{cross}N/\sqrt{T}`$. We also consider the translocation of hydrophilic polypeptides across hydrophobic pores. Biological systems accompolish this by having a hydrophobic signal sequence at the end that goes in first. Our analysis leads to the conclusion that for such a molecule, the configuration of the molecule in the transition state is similar to a hook, and this is in agreement with presently accepted view in cell biology . It is also possible that a kink movement mechanism might operate in other biological phenomena, like protein folding
## VII Acknowledgements
K.L. Sebastian is deeply indebted to Professor K. Kishore for the encouragement that he has given over the years and this paper is dedicated to his memory. He thanks Professors S. Vasudevan and Diptiman Sen for discussions and Professors B. Cherayil and Indrani Bose for their comments.
## A appendix
### I The eigenfunctions of the Hamiltonian $`\widehat{H}^{}`$
The Hamiltonian $`\widehat{H}^{}=\frac{^2}{n^2}+\omega _0^2\left\{13sech^2\left(\omega _0n/2\right)\right\}`$ has the following eigenfunctions (functions are not normalised) and eigenvalues, if $`n`$ allowed to be in the range $`(\mathrm{},\mathrm{})`$.
#### 1 Discrete States
1) $`\psi _0(n)=sech(\frac{\omega _0n}{2})^3;`$ $`\epsilon _0=5\omega _0^2/4`$
2) $`\psi _1(n)=sech(\frac{\omega _0n}{2})^2\mathrm{tanh}(\frac{\omega _0n}{2});`$ $`\epsilon _1=0`$
3) $`\psi _2(n)=\left\{3+2\mathrm{cosh}(\omega _0n)\right\}sech(\frac{\omega _0n}{2})^3;`$ $`\epsilon _2=3\omega _0^2/4`$
#### 2 Continuum States
The continous part of the spectrum starts at $`\omega _0^2`$. The potential is reflectionless. Corresponding to an eigenvalue $`\omega _0^2+k^2`$, there are two eigenfunctions, which we write as an odd function and an even function. They are:
1) $`\psi _{even}(n)=8k\left(k^2+\omega _0^2\right)\mathrm{cos}(kn)3\omega _0\left(8k^2+3\omega _0^2\right)\mathrm{sin}(kn)\mathrm{tanh}(\frac{\omega _0n}{2})`$
$`30k\omega _0^2\mathrm{cos}(kn)\mathrm{tanh}(\frac{\omega _0n}{2})^2+15\omega _0^3\mathrm{sin}(kn)\mathrm{tanh}(\frac{\omega _0n}{2})^3`$
2) $`\psi _{odd}(n)=8k\left(k^2+\omega _0^2\right)\mathrm{sin}(kn)3\omega _0\left(8k^2+3\omega _0^2\right)\mathrm{cos}(kn)\mathrm{tanh}(\frac{\omega _0n}{2})`$
$`+30k\omega _0^2\mathrm{sin}(kn)\mathrm{tanh}(\frac{\omega _0n}{2})^2+15\omega _0^3\mathrm{cos}(kn)\mathrm{tanh}(\frac{\omega _0n}{2})^3`$
In the limit $`n\pm \mathrm{}`$, the even function becomes like
$`\psi _{even}(n)=2k\left(4k^211\omega _0^2\right)\mathrm{cos}(kx)\pm 6\omega _0\left(4k^2+\omega _0^2\right)\mathrm{sin}(kx)`$, which may be written as $`(Constant)\mathrm{cos}(kx\delta (k))`$, so that the phase shift $`\delta (k)=\mathrm{arctan}(\frac{3\omega _0\left(\omega _0^24k^2\right)}{k\left(11\omega _0^2+4k^2\right)})`$. The phase shift for the odd solution is just the same. Hence the total change in the density of states is given by $`\mathrm{\Delta }n(k)=\frac{2}{\pi }\frac{d\delta (k)}{dk}=\frac{2}{\pi }\left(\frac{\omega _0}{k^2+\omega _0^2}+\frac{2\omega _0}{4k^2+\omega _0^2}+\frac{6\omega _0}{4k^2+9\omega _0^2}\right)`$. On integration, $`_0^{\mathrm{}}๐k\mathrm{\Delta }n(k)=3`$ as it should be, as there are three bound states for $`\widehat{H}^{}`$.
### II Evaluation of the Infinite Products
#### 1 Hairpin Crossing
The infinite product is:
$$I_{hp}=\left(\frac{\underset{k}{}\epsilon _k}{\underset{k>2}{}\epsilon _k^{}}\right)^{1/2}$$
(A-1)
where $`\epsilon _k`$ represent the eigenvalues of the continuum states of the hamiltonian $`\widehat{H}^{ms}=\left(\frac{^2}{n^2}+\omega _0^2\right)`$and $`\epsilon _k^{}`$are the eigenvalues of $`\widehat{H}^{}`$, satisfying the boundary conditions at $`n=\pm \mathrm{}`$. The above product involves only the continuum eigenvalues of the two Hamiltonians. Now,
$$\mathrm{ln}I_{hp}=\frac{1}{2}\left(\underset{k}{}\mathrm{ln}\epsilon _k\underset{k>2}{}\mathrm{ln}\epsilon _k^{}\right)$$
$$=\frac{1}{2}_0^{\mathrm{}}๐k\mathrm{ln}\left(\omega _0^2+k^2\right)\left(n(k)n_{hp}^{}(k)\right)$$
where the $`n(k)`$ stands for the density of states in the continuum, for the Hamiltonian $`\widehat{H}^s`$and $`n_{hp}^{}(k)`$ for the Hamiltonian $`\widehat{H}^{}`$. The change in the density of states is $`\mathrm{\Delta }n_{hp}(k)=n(k)+n_{hp}^{}(k)`$ and is easily evaluated from the information given in subsection A I. It is $`\mathrm{\Delta }n_{hp}(k)=\left(\frac{\omega _0}{\omega _0^2+k^2}+\frac{2\omega _0}{\omega _0^2+4k^2}+\frac{6\omega _0}{9\omega _0^2+4k^2}\right)\frac{2}{\pi }`$. Using this, we get
$$I_{hp}=\frac{15}{2}\omega _0^3$$
(A-2)
#### 2 End Crossing
The product that we wish to evaluate is
$$I_{end}=\left(\frac{\underset{k}{}\epsilon _k}{\underset{k>2}{}\epsilon _k^{}}\right)^{1/2}$$
(A-3)
This infinite product in the above equation is over the continuous spectra of the two Hamiltonians and may be evaluated. The change in the density of states is now just half of the density of states density of states for the hairpin case. That is, $`\mathrm{\Delta }n_{end}(k)=\frac{1}{2}\mathrm{\Delta }n_{hp}(k)`$. At the first sight, this leads to a problem, as $`_0^{\mathrm{}}๐k`$ $`\mathrm{\Delta }n_{end}(k)=3/2`$, instead of the expected $`2`$ (as $`\widehat{H}^{}`$ has two bound states while $`\widehat{H}^{ms}`$ has none). The solution to ths is quite well known - $`\widehat{H}^{ms}`$ has a state with eigenvalue $`\omega _0^2`$ where its continuous spectrum starts, and half of this state is to be considered as a bound state. Then, we can write the above as:
$$I_{end}=\mathrm{exp}\left(\frac{1}{2}_0^{\mathrm{}}๐k\mathrm{\Delta }n_{end}(k)\mathrm{ln}\left(\omega _0^2+k^2\right)\right)\sqrt{\omega _0}$$
Hence we find
$$I_{end}=\left(\frac{15}{2}\right)^{1/2}\omega _0^2$$
(A-4)
. |
warning/0001/gr-qc0001062.html | ar5iv | text | # 1 Introduction
## 1 Introduction
Supernovae have been historically the first envisaged sources of gravitational waves (GW). Although binary inspirals or even periodic GW emitters like pulsars seem to be nowadays more promising sources, impulsive sources of GW such as supernovae should also be considered in the data analysis design of interferometric detectors currently under construction (LIGO, VIRGO). Impulsive GW sources are typically collapses of massive stars, leading to the birth of a neutron star (type II supernova) <sup>?,?,?</sup> or of a black hole <sup>?</sup>; mergers of compact binaries can also be considered as impulsive sources <sup>?</sup>.
The problem with such sources is that the emitted waveforms are very poorly predicted, unlike the binary inspirals. As a consequence, this forbids the use of matched filtering for the detection of GW bursts. The filtering of such bursts should therefore be as general and robust as possible and with minimal a priori assumptions on the waveforms. A drawback is of course that such filters will be sensitive to non-stationary noise as well as to GW bursts; spurious events, e.g. generated by transient noise, should be eliminated afterwards when working in coincidence with other detectors. But, on the other hand, burst filters could help to identify and understand these noise sources, which would be useful especially during the commissioning phase of the detector.
All the filters presented here are dedicated to GW bursts detection and are compared by studying their performance to detect a reference sample of GW burst signals, numerically computed by Zwerger and Mรผller (ZM).<sup>?</sup>
Throughout the following, we assume that the detector noise is white, stationary and Gaussian with zero mean. For numerical estimates, we chose the flat (amplitude) spectral density to be $`h_n4\times 10^{23}/\sqrt{\mathrm{H}z}`$ and the sampling frequency $`f_s20`$ kHz, so the standard deviation of the noise is $`\sigma _n=h_n\sqrt{f_s/2}4\times 10^{21}`$. The value chosen for $`h_n`$ corresponds approximately to the minimum of the sensitivity curve of the VIRGO detector <sup>?</sup>; around this minimum, the sensitivity is rather flat, in the range \[200 Hz,1kHz\], which is precisely the range of interest for the gravitational wave bursts we are interested in. This validates then our assumption of a white noise ; otherwise, we can always assume that the detector output has been first whitened by a suitable filter <sup>?</sup>.
## 2 General filters
### 2.1 Filters based on the autocorrelation
The noise being whitened, its autocorrelation is ideally a Dirac function and in practice vanishes outside of zero. The autocorrelation of the data $`x(t)`$
$$A_x(\tau )=x(t)x(t+\tau )๐t$$
(1)
should then reveal the presence of some signal (which is surely correlated). The information contained in the autocorrelation function can be extracted in different ways. We have studied two of them and built so two non-linear filters. The first one computes the maximum of $`A_x(\tau )`$ and has already been described in <sup>?</sup>. In the following, we will refer to this filter as the Norm Filter (NF). A similar approach has been developed independently by Flanagan and Hughes in the context of the detection of binary black hole mergers <sup>?</sup>.
Another possibility is to look at the norm of the autocorrelation function (NA Filter) :
$$A=\sqrt{\frac{1}{N}\underset{k=2}{\overset{N}{}}A(k)^2},$$
(2)
where $`A(k)`$ denotes the discrete autocorrelation of $`N`$ data $`x_i`$. The sum is here initiated at the second bin according to the fact that the noise (uncorrelated) contributes essentially to the first bin. Note that the only parameter for these two filters is the window size $`N`$. The behavior of the NA filter with noise only is not known analytically and its characteristics (mean and standard deviation) have to be found numerically (adding some complexity to this filtering method).
### 2.2 The Bin Counting method
This filter (BC) computes the number of bins in a window of size N whose value exceeds some threshold $`s\times \sigma _n`$. The threshold $`s`$ is chosen by maximizing the signal to noise ratio (SNR) when detecting the signals of the ZM catalogue (for more details, see <sup>?</sup>).
### 2.3 Linear Fit Filters
This filter fits the data to a straight line in a window of size $`N`$. If the data are pure white noise with zero mean the slope and the offset of the fitted line are zero on average, so this can well discriminate between the two cases : only noise or noise+signal. The slope and the offset of the fit can easily be computed as a function of time and of the $`x_i`$ (see <sup>?</sup>).
In fact, the slope and the offset are two correlated random variables. By computing their Covariance Matrix, one reduces them to two uncorrelated normal random variables, $`X_+`$ and $`X_{}`$, with
$$X_\pm =\left(\frac{X\pm Y}{\sqrt{2(1\pm \alpha )}}\right),\text{with }\alpha =\sqrt{\frac{3}{2}\left(\frac{N+1}{2N+1}\right)}=COV(X,Y)$$
(3)
where X = SNR(slope) = $`|slope|/\sigma _{slope}`$, Y = SNR(offset). Finally, the sum $`X_+^2+X_{}^2`$ is a $`\chi ^2`$ like random variable and gives us a new filter, ALF (Advanced Linearfit Filter).
### 2.4 The Peak Correlator
Filtering by correlating the data with peak (or pulse) templates is justified by the fact that simulated supernovae GW signals exhibit one (or more) peaks. The pulse templates have been built from truncated Gaussian functions. The method and results are explained in <sup>?</sup>.
## 3 Performance and efficiency of the filters
### 3.1 Definition of a false alarm rate
We arbitrarily set the false alarm rate for each of the filters to be $`10^6`$ (72 false alarms per hour for a sampling frequency $`f_s=20`$ kHz). This high rate is required because the signals we look for are very weak. False alarms will be discarded later when working in coincidence.
### 3.2 The Zwerger and Mรผller Catalogue
The catalogue of Zwerger and Mรผller <sup>?</sup> contains 78 gravitational-wave signals. Each of them corresponds to a particular set of parameters (e.g initial distribution of angular momentum). All the signals are computed for a source located at 10 Mpc. We can then re-scale the waveforms in order to locate the source at any distance $`d`$.
Since the signal waveforms are here known, we can explicitly derive the optimal SNR provided by the Wiener filter matched to each of them, and then compute the maximal distance of detection. We will then be able to build a benchmark for the different filters by comparing their results (detection distances) to the results of the Wiener filter (we consider here optimally polarized GWโs, along the interferometer arms).
The mean distance obtained for the Wiener Filter, averaged over all the signals, is about $`\overline{d}_{\mathrm{o}pt}25.4`$ kpc, which is of the order of the diameter of the Milky Way.
### 3.3 Estimating a filter power
The optimal (Wiener) filtering allows to detect the $`i^{\mathrm{t}h}`$ signal in the Catalogue emitted by a source located up to a distance $`d_i^{(๐ฒ)}`$. Similarly, a filter $``$ is able to detect the same signal up to a distance $`d_i^{()}`$; of course $`d_i^{()}`$ is averaged over many noise realizations (about 1000) in a Monte Carlo simulation. The detection performance of the filter $``$ for this signal is simply defined as the distance of detection relative to the optimal distance of detection : $`d_i^{()}/d_i^{(๐ฒ)}`$. The global performance of $``$ is then estimated as the detection performance averaged over all the waveforms of the catalogue:
$$\rho =\frac{1}{78}\underset{i=1}{\overset{78}{}}\frac{d_i^{()}}{d_i^{(๐ฒ)}}.$$
(4)
For a given filter, and a given source located at a distance $`d`$, one can also evaluate a detection efficiency $`ฯต`$, which is the number of detections $`n`$ over the total number of noise realisations $`๐ฉ`$. This efficiency (averaged over all the signals of the catalogue) will characterize the practical behaviour of the filter.
### 3.4 Comparison of the filtering methods : Performance
The results for the different filters are reported in Table 1. We also give the average distance of detection $`\overline{d}=\frac{1}{78}_{i=1}^{78}d_i^{()}`$ for all the filters.
The three first filters NF, NA and BC (all non-linear) have a performance slightly below one half, while the PC have a performance greater than 0.7. Both $`X_+`$ (or $`X_{}`$) and ALF can reach a performance around 0.8. Note that ALF has been in fact implemented with a sampling of 20 different window sizes, sufficient to cover the variety of signals (with a non-significant loss of generality). If implemented with a single window size, as the other filters NF, NA and BC, its performance decreases down to 0.72.
### 3.5 Comparison of the filtering methods : Detection efficiency
For the Wiener filter, one can show that the mean efficiency is roughly $`50\%`$ for signals located at a distance $`d_i^{(๐ฒ)}`$. If $`\overline{\rho }`$ is the mean performance of a given filter, one would expect a $`50\%`$ detection efficiency at $`d_i^{()}\overline{\rho }\times d_i^{(๐ฒ)}`$. In fact, such efficiency is reached for a smaller distance.
If one defines the effective performance $`\rho _{eff}`$ as the ratio $`d_i^{()}/d_i^{(๐ฒ)}`$ for which the detection efficiency is about $`50\%`$, then $`\rho _{eff}0.74`$ for ALF and $`X_{}`$ and $`\rho _{eff}0.71`$ for $`X_+`$. This definition gives an idea of the efficiency one can reach in practice, and has to be taken into account when choosing between different online triggers.
## 4 Conclusion
We have discussed several filters to be used as triggers for detecting GW bursts in interferometric detectors. They are all sub-optimal but their performance is close to the one obtained with the Wiener Filter.
Concerning the detection of Zwerger-Mรผller-like signals, we note that none of the BC, NF and NA filters is efficient enough to cover the whole Galaxy in average (but their window sizes have not yet been optimized), contrary to ALF and PC (and optimal) filters. A few signals can be detected at distances beyond 50 kpc, the distance of the Large Magellanic Cloud (LMC). It is clear that this class of signals will be detected by the first generation interferometric detectors such as VIRGO only if the supernovae occur inside our Galaxy or in the very close neighbourhood.
Finally, all the filters studied here can be implemented on line without problem, due to use of FFTโs (for the NA and the PC) or to simple recursive relations between filter outputs in successive windows (NF,BC or ALF).
Correlations and coincidences between those filters are under study in order to either reduce background (hence a quantifiable loss of signal) or lower detection thresholds (hence a gain of a few $`\%`$ in performance and efficiency).
References |
warning/0001/cond-mat0001217.html | ar5iv | text | # Stripes: Why hole rich lines are antiphase domain walls?
## I Introduction
An anti-phase domain wall in stripes is the state where the local antiferromagnetic (AFM) spin order parameter undergo a $`\pi `$ phase shift across a hole rich line. Such periodic stripe structures were experimentally found in doped Copper oxides and Nickel oxides. Historically, this well accepted feature was first considered to be a natural outcome of mean-field theory Fermi surface instability. Yet, no similar rigorous microscopic explanation was given within the frustrated phase separation picture which is currently regarded as underlying the microscopic origin of stripes in the relevant materials. One of our goals here is to close this gap in the theory.
In contrast with common folklore, we show that hole rich lines are not necessarily anti-phase domain walls of AFM spin domains. First, on simple general grounds, we argue that there must be a phase transition from anti-phase to in-phase domain wall configuration as a function of increased electron filling fraction $`\delta `$ of the domain wall. We then proceed to construct microscopic t-J models of the local electronic dynamics which determines the resulting spin order across a hole line.
It is frequently argued theoretically and exemplified experimentally that the charge segregation into hole rich stripe lines is prior to the establishment antiphase spin domains. Therefore, the microscopic mechanism by which the hole lines enslave the spin order should be distinguished and considered separately.
We develope a qualitative and quantitative microscopic understanding of the domain wall magnetic order. Our analysis accounts for the electronic dynamics both transverse and along a pre-established hole rich line between two AFM domains. Thus we focus solely on the mechanism which give rise to the spin antiphase domain wall feature, by examining the competition between anti-phase and in-phase configurations at given electron filling fraction $`\delta `$ of the hole rich line.
In the stripes literature it is more common to describe the domain wall filling in terms of the number of holes (below half-filling) per site along the domain wall;
$$n^h=12\delta .$$
(1)
Experimentally, anti-phase domain wall stripes in Nickel-Oxides have $`n^h1`$ (corresponding to one hole per site along the domain wall, or equivalently an electron empty domain wall, $`\delta =0`$), while anti-phase domain wall stripes in Copper-Oxides have $`n^h\frac{1}{2}`$ (corresponding to one hole per two sites along the domain wall, or electron 1/4 filled domain wall, $`\delta 0.25`$). These domain wall filling fractions remain roughly constant over a wide range of dopings in the respective materials.
We find that the transition from anti-phase to in-phase stripe domain wall occurs at critical filling fraction
$$0.28<\delta _c<0.30$$
(2)
depending on the value of $`\left(\frac{J}{t}\right)`$. In other words, for example, we predict that stripe domain walls with a hole density of one hole per three sites are always in-phase and not anti-phase. Appropriate numerical simulations can be constructed to get exact numbers beyond the limits of our analytical approximations. Yet, since our analysis indicates that the sensitivity to $`\left(\frac{J}{t}\right)`$ is rather weak, we do not expect the exact numerical results to deviate much from our predictions.
We further apply our model to evaluate the of spin-wave velocity in a stripe systems, and compare with experiments. In addition, we relate the competing interactions in our microscopic model to previous Landau theory approach to stripes order. Thus, we advance towards a coherent microscopic and phenomenological understanding of the stripes structure within the frustrated phase-separation picture.
### A Overview of the model construction, analysis and main results
Since the paper is quite long, we here supply the reader with an overview of the gist of our model development and main results.
The general intuitive argument for the domain wall transition is the following: In one extreme case, $`n^h=1`$, where there is a hole on each site along the domain wall (i.e., it is empty of electrons, $`\delta =0`$, as in Nickel-Oxides), it is clear that an anti-phase domain wall configuration is favored by transverse hole fluctuations. Now, consider the state of domain walls with increasing electron filling fraction. In the opposite extreme case where the domain wall is half-filled with electrons (i.e., $`n^h=0`$, $`\delta =\frac{1}{2}`$), we should recover the undoped ordered AFM state. Therefore, there must be some intermediate critical electron filling fraction $`\delta _c`$ of the domain wall at which there is a transition from an anti-phase to in-phase local AFM spin configuration across a hole rich line.
The main quantitative objective of this paper is to determine the critical domain wall filling fraction $`\delta _c`$ as a function of the t-J model parameters. In the process, we develope an advanced qualitative and quantitative understanding of the various competing local interactions in the single stripe physics. Additional applications are discussed in sections-V and VI.
Intuitively, increase of electron filling of the domain wall amounts to increase of magnetic interactions across the domain wall until they are strong enough to dominate over the charge fluctuation dynamics (due to holes). The possibility of a transition from topological (anti-phase) to non-topological (in-phase) stripes, due to increase of AFM interactions, was first speculated by Neto & Hone (though, on not quite rigorous grounds, it was somehow related to the spin correlation length).
We find that the topological/non-topological nature of the stripe charge wall can be completely determined by the local dynamics, which in turn is determined by the electron filling fraction of the wall (assuming fixed given t-J model parameters). We do not see a way by which interaction between domain walls, the width of the spin domains or any similar long length scale are significantly relevant.
In section-II, we list some preliminary assumptions of our model of the stripe domain wall: (a) The effective electron dynamics is captured by a one band t-J model for electrons hopping between Copper sites (i.e., in which the Oxygen sites are integrated out). (b) It is assumed that the hole line, its mean position, and hole density are already pre-established by some higher energy processes (presumably phase separation and coulomb frustration). We hence focus solely on determining the preferred spin configuration across a pre-established hole line. (c) The spin order is determined by comparing the energetics of anti-phase and in-phase domain wall configurations (and not with the motion of dilute holes in a uniform AFM as was done in most previous work).
In section-III, we introduce our microscopic model and start with the consideration of only transverse interactions and dynamics, (i.e., ignoring the effects of dynamics along the domain wall). The kinetic energy due to transverse hopping of holes favors an anti-phase domain wall configuration, while magnetic interaction between electrons favors an in-phase domain configuration (see figure-1 below). These competing interactions determine the preferred local spin configuration. All calculations are done to the first significant order in $`\left(\frac{J}{t}\right)`$. Our analysis surprisingly shows that if dynamics along the domain wall is ignored then transition to an in-phase domain wall would have occurred already at $`n^h2/3`$ (i.e., at a density of two holes per three sites), in conflict with experimental observations of anti-phase stripes in $`\left(LaNd\right)SrCuO`$ materials with $`n^h1/2`$. Hence, it is essential to investigate the effect of kinetic dynamics along the domain wall, which we undertake in section-IV.
In section-IV, we model the kinetic dynamics of the electrons moving along a stripe domain wall in an effective external magnetic mean field due to its AFM environment. Along an in-phase domain wall there is an effective net staggered external magnetic field, while along an anti-phase domain wall the net external field is zero on each site. We analyze two extreme limits: (a) Non-interacting electrons moving along the domain wall, and (b) Large $`Ut`$ limit for the interaction of electrons along the domain wall. In both cases, the essential result is that kinetic fluctuations along the domain wall weaken the average magnetic interaction energy which favors an in-phase domain wall, and thus extend the stability of an anti-phase domain wall configuration to higher electron filling fractions (i.e., lower hole densities).
In section-V, we relate the competing interactions in our microscopic model to previous Landau theory approach to stripes order .
In section-VI, we apply our model to evaluate the spin-wave velocity $`v_{}`$ in a stripe state compared with $`v_0`$ in the parent AFM material (where $`v_{}`$ is velocity perpendicular to the stripes).
We argue that our results are quantitatively accurate well beyond the seemingly rough approximations of our simple models. The heart of the matter is the fact that our quantitative results are sensitive only to the energetic difference between an anti-phase and in-phase domain wall configuration. Processes which are neglected in our treatment (e.g., deeper penetration of hole hopping into the AFM environment) have the same contribution in both domain wall configurations and thus only very weakly affect the energetic difference between them.
## II Preliminary assumptions
The stripes characteristics can vary to include both diagonal and vertical stripes, which may be insulating or conducting. Therefore, the anti-phase domain wall mechanism should be rather simple, robust, and not too sensitive to the above mentioned variations.
1. The most microscopic Hubbard type model of the CuO<sub>2</sub> or NiO<sub>2</sub> planes includes distinct Oxygen and Copper (or Nickel) orbitals bands. Yet, as commonly argued, we assume that the effective dynamic is captured by a one-band t-J model (in which the oxygen sites degrees of freedom have been integrated out) where electrons hop directly between Copper lattice sites. In the context of stripes theory, the above assumption is supported by the fact that correct domain wall configurations turned out in numerical simulations of one band t-J models.
2. Experimental evidence and theoretical considerations suggest that stripe formation is commonly charge driven, i.e., that periodic hole line stripes form first and enslave the formation of the AFM spin domain. Therefore, for our purpose in this paper we take the hole lines to be pre-formed.
3. Stripes were found in both Spin-$`\frac{1}{2}`$ and Spin-$`1`$ doped antiferromagnets. A doped hole can thus correspond to a spin-$`0`$ or a spin-$`\frac{1}{2}`$ site respectively. Yet, in both cases the hole and its dynamics are carried on only within the $`d_{x^2y^2}`$ orbital band. Indeed, the model which we construct and analyze works equally well for both doped Spin-$`\frac{1}{2}`$ and Spin-$`1`$ antiferromagnets. We chose to present our analysis in terms of a doped Spin-$`\frac{1}{2}`$ AFM (i.e., corresponding to stripes in a CuO<sub>2</sub> plane).
4. As with superexchange mechanism of antiferromagnetism in the undoped parent system, we argue that the domain wall spin order is determined by local interactions across the hole rich line. Hence, it is sufficient to consider a single stripe segment in isolation (see figure-1).
5. Thereโs a non-trivial distinction between site-centered and bond-centered domain walls, in the sense that the spin alignment across an anti-phase domain wall is antiferromagnetic for site-centered stripe and ferromagnetic for bond-centered stripe. The presentation in this paper is conducted in terms of site-centered stripes, and elsewhere we will show that the same principles apply for bond-centered stripes as well.
6. Probably the main quantitative approximation in our model is that we treat the AFM regions between the hole lines as if they were antiferromagnetically ordered. We neglect spin exchange fluctuations and the quantum nature of the AFM correlations in the rather narrow ladder geometry of the stripes. In other words, our quantitative results are rigorously valid for an Ising model approximation of the AFM regions. Yet, in all of our calculations we use a parameter $`\epsilon `$ which is defined to be the energy difference between parallel and anti-parallel near-neighbor spin states (see equation (6)). Only in the end we substitute $`\epsilon =J`$ for intuitive concreteness. In principle, all the effects of fluctuations etc. can be incorporated into a renormalized value of $`\epsilon `$ without changing our results.
7. Our quantitative results are sensitive only to the energetic difference between an anti-phase and in-phase domain wall configuration. Therefore, many processes which are neglected in our treatment (e.g., deeper penetration of hole hopping into the AFM environment, magnetic interaction between electrons on the wall) have the same contribution in both domain wall configurations and thus only very weakly affect the energetic difference between them.
## III The effect of transverse interaction
Our analysis proceed in two stages; First, in this section, we consider solely the transverse fluctuations of a hole and magnetic interaction of electrons in the stripe, while the holes/electrons configuration along the domain wall is taken as static. In a second stage (section-IV), we consider the implications of electron mobility along the stripe.
### A Model of site-centered stripe
The average hole line position is fixed by introducing a chemical potential shift ($`\mu `$) on particular sites. (In Figure-1, hole line $`\mu `$-shifted sites are represented by dark circles). In the ground state, holes are preferentially situated on the $`\mu `$ shifted sites. Hence, the effective local chemical potential shift, $`\mu `$, incorporates the net effect of the high energy dynamics (e.g., coulomb frustrated phase separation ) which lead to the stripe formation. The magnitude of $`\mu `$ determines the stripe stability, and can be associated with the stripe creation temperature for the purpose of extracting an experimental estimate of itโs magnitude. Thus, it is assumed that $`\mu <J<t`$.
An antiferromagnetic spin exchange interaction $`J`$ exists between electron spins on nearest neighbors sites. In order to fix the spin order, it is enough to determine the relative orientation of the spins immediately to the left and to the right of the hole line. Therefore, we include a transverse hoping interaction, $`t`$, only between a domain wall site (indicated by a dark circle in figure-1a) and its right and left nearest neighbors in the AFM regions. As noted before, we argue that processes of further penetration of a hole into the AFM regions are not significantly affecting our results. Hopping interaction $`t_{}`$ along the domain wall will be considered in section-IV.
In the absence of kinetic motion along the domain wall, the Hamiltonian describing a single site centered domain wall is
$`H`$ $`=`$ $`{\displaystyle \underset{i}{}}H_i`$ (3)
$`H_i`$ $`=`$ $`\mu n_{0,i}+J{\displaystyle \frac{1}{2}}{\displaystyle \underset{jj^{^{}}ii^{^{}}}{}}\text{S}_{j,i}\text{S}_{j^{},i^{}}`$ (5)
$`t{\displaystyle \underset{i,s}{}}[(c_{1,i,s}^{}c_{0,i,s}+c_{1,i,s}^{}c_{0,i,s})+\mathrm{h}.\mathrm{c}.]`$
where S<sub>j,i</sub> is the spin of an electron at column $`j`$ and line $`i`$ (where $`j=0`$ is the domain wall position). $`n_{0,i}`$ is the occupation number of a domain wall site at position $`i`$ along the wall. $`c_{1,i,s}^{}`$ and $`c_{1,i,s}^{}`$ are electron creation operators respectively to the right and left of the domain wall site $`n_{0,i}`$. (t-J model projection operator which excludes double occupancy is implicitly assumed through out the paper).
### B Kinetic transverse hole fluctuations
In the absence of hoping interaction, $`t`$, the anti-phase stripe groundstate (a) and the in-phase stripe groundstate (aโ) are degenerate. But, the energy of the corresponding excited states (b) and (bโ) differ. As a result, transverse hole hoping interaction removes the ground state degeneracy in favor of anti-phase domain wall.
To be general, we define $`\epsilon `$ to be the energy difference between an antiferromagnetic โgood bondโ and a ferromagnetic โbad bondโ of two neighboring spins. Obviously, $`\epsilon `$ is proportional to $`J`$ (and $`\epsilon =J`$ in the case of Ising model of the AFM spin domains). We label by $`E_1^{anti}`$ and $`E_1^{in}`$ the bare excited states energy in the case of an anti-phase and in-phase domain walls as depicted in figure ($`1b`$) and ($`1b^{^{}}`$) respectively. Clearly,
$$E_1^{in}E_1^{anti}=\epsilon $$
(6)
The Hamiltonian (5) is thus given by the matrix
$$H^\alpha =\left(\begin{array}{ccc}0& t& t\\ t& E_1^\alpha & 0\\ t& 0& E_1^\alpha \end{array}\right)$$
(7)
($`\alpha =`$anti/in) which can be diagonalized exactly, resulting with the ground state energy $`E_g^\alpha `$
$`E_g^\alpha `$ $`=`$ $`{\displaystyle \frac{1}{2}}\left(E_1^\alpha \sqrt{\left(E_1^\alpha \right)^2+8t^2}\right)`$ (8)
$``$ $`\{\begin{array}{cc}\frac{1}{2}\left[E_1^\alpha \sqrt{8}t\left(1+\frac{\left(E_1^\alpha \right)^2}{16t^2}\right)\right]\hfill & \text{ for }Jt\hfill \\ \frac{2t^2}{E_1}\hfill & \text{ for }Jt\hfill \end{array}`$ (11)
The difference in kinetic energy gain due to transverse hoping, between an anti-phase and an in-phase domain wall configurations, is given by
$`\mathrm{\Delta }E_1^{kin}`$ $`=`$ $`E_g^{anti}E_g^{in}`$ (12)
$``$ $`\{\begin{array}{cc}\frac{\epsilon }{2}+๐ช\left(\frac{\epsilon ^2}{t}\right)\hfill & \text{ for }Jt\hfill \\ \frac{t^2}{\epsilon }\hfill & \text{ for }Jt\hfill \end{array}`$ (15)
Our result agrees with a previous large-d calculation (valid only in the limit $`Jt`$). But, in the experimental systems of interest $`\frac{J}{t}\frac{1}{3}`$, and thus they are better approximated by calculations in the limit $`Jt`$ which will be assumed for the rest of our discussion. From equation (12) we conclude that, in a site-centered stripe geometry, the zero point transverse kinetic fluctuation of the holes favor an antiferromagnetic alignment of the neighboring spins.
Note that the above result is in stark contrast with the conventional wisdom that holeโs zero-point kinetic fluctuations always favors ferromagnetic alignment of their environment, which is based mostly on isolated hole models. It is a rather simple demonstration of the difference between collective and single hole properties in doped antiferromagnets.
### C Competing magnetic interactions at electron filling $`\delta 0`$
From the analysis of the above model, we conclude that the transverse kinetic fluctuations of holes are sufficient to induce an anti-phase domain wall ground state in an empty domain wall $`\delta =0`$ (i.e., one hole per site along the wall, $`n^h=1`$) of stripes.
For a domain wall with electron filling fraction of $`\delta `$, $`2\delta `$ is the number of electrons per site along the domain wall (in the large U limit), and $`n_h=\left(12\delta \right)`$ is the number of holes in the lower Hubbard band. Each hole contributes a kinetic energy difference $`\mathrm{\Delta }E_1^{kin}`$. Hence, at an arbitrary electron filling fraction $`\delta `$ of the domain wall, the average transverse kinetic energy gain per site $`\mathrm{\Delta }E_{}^{kin}\left(\delta \right)`$ of an anti-phase domain wall in comparison with an in-phase domain wall is
$$\mathrm{\Delta }E_{}^{kin}\left(\delta \right)=\frac{\mathrm{\Delta }E_1^{kin}}{N_{site}}=\frac{\epsilon }{2}\left(12\delta \right)$$
(16)
For each electron on the domain wall there is an energy difference $`\mathrm{\Delta }E_1^{mag}=+\epsilon `$ in favor of an in-phase domain wall, due to spin exchange interaction with the AFM environment. At electron filling $`\delta `$, the average magnetic energy difference (of an anti-phase domain wall in comparison with an in-phase domain) per site along the domain wall is
$$\mathrm{\Delta }E_{}^{mag}\left(\delta \right)=\frac{\mathrm{\Delta }E_1^{mag}}{N_{site}}=+\epsilon \left(2\delta \right).$$
(17)
The competition of these transverse interaction alone would predict a transition as a function of domain wall filling from anti-phase to in-phase domain wall when
$$\mathrm{\Delta }E_{}\left(\delta \right)=\mathrm{\Delta }E_{}^{kin}+\mathrm{\Delta }E_{}^{mag}>0$$
(18)
at
$$\delta \frac{1}{6}$$
(19)
(corresponding to $`n^h2/3`$, i.e., at a density of two holes per three sites), in conflict with experimental observations of anti-phase stripes in $`\left(LaNd\right)SrCuO`$ materials with $`\delta 1/4`$. This conflict is resolved in the next section, where we account for effect of kinetic dynamics along the domain wall.
## IV Implications of electron dynamics along the domain wall
In this section we investigate the consequences of electron dynamics along the domain wall for the competition between anti-phase and in-phase domain wall configurations. As a first order approximation, we assume static spin configuration of the antiferromagnetic domains on the left and right of the domain wall. Thus, electrons moving along the domain wall are effectively modeled as a one-dimensional electron gas (1DEG) in a static external magnetic field. At this mean-field level, electrons moving in an anti-phase domain wall experience a net zero external field on each site, while electrons moving on an in-phase domain wall experience a staggered external magnetic field of magnitude proportional to the spin interaction strength $`J`$. Hence, the effective domain wall 1DEG Hamiltonian has the form
$`H`$ $`=`$ $`H_0\left(B\right)+U{\displaystyle \underset{j}{}}n_jn_j`$ (20)
$`H_0\left(B\right)`$ $`=`$ $`t_{}{\displaystyle \underset{j\sigma }{}}(c_{j\sigma }^{}c_{j+1,\sigma }+h.c.)\mu _F{\displaystyle \underset{j\sigma }{}}c_{j\sigma }^{}c_{j\sigma }`$ (22)
$`+2B{\displaystyle \underset{j\sigma }{}}\sigma \left(1\right)^jc_{j\sigma }^{}c_{j\sigma }`$
where $`\sigma =\pm 1`$ for spin $`,`$ respectively, and
$$B=\{\begin{array}{cc}0& \text{ for anti-phase}\\ \frac{J}{4}& \text{ for in-phase}\end{array}$$
(23)
When considering the competition of stripes versus droplets forms of local phase separation, Nayak&Wilzek proposed that a significant energy is gained by the increased mobility along an anti-phase domain wall. But we note that the kinetic energy gain in an in-phase domain wall is practically the same as in anti-phase domain wall state. Therefore, it is essential to make a more careful analysis of the electronic dynamics along the domain wall in both cases.
Below, we extract quantitative results in two limits: (a) For non-interacting electrons moving along the domain wall, and (b) In the large $`Ut`$ limit for the interaction of electrons along the domain wall. In both cases, the essential result is that kinetic fluctuations along the domain wall weakens the magnetic interaction energy gain which favors an in-phase domain wall environment, and thus extends the stability of anti-phase domain wall configuration to higher electron filling fractions (i.e., lower hole densities). In particular, we conclude that dynamics fluctuations along the domain wall allow for the establishment of anti-phase stripes with domain wall filling $`\delta \frac{1}{4}`$ (as in Copper-Oxides systems).
### A Effect of motion along the domain wall for non-interacting electrons at arbitrary filling
We here model the dynamics along the hole rich domain wall by an effective one dimensional lattice model (20) of non-interacting electrons ($`U=0`$). For an in-phase domain wall, the exchange coupling to the spins in the AFM environment result with an effective external staggered magnetic field on the electrons moving along the domain wall (see figure-1b). Our model of kinetic motion along the domain wall included only band structure effects due to a static spin configuration of the AFM environment, but no dynamic scattering interactions (leading to finite resistivity).
The anti-phase domain wall spectrum ($`B=0`$) is
$`E_n^{(anti)}`$ $`=`$ $`2t_{}\mathrm{cos}\left(k_na\right)\mu _F^{(anti)}`$ (24)
$`k_na`$ $`=`$ $`{\displaystyle \frac{2\pi }{N}}n\text{ (}{\displaystyle \frac{N}{2}}n{\displaystyle \frac{N}{2}}\text{)}`$ (25)
$`\mu _F^{(anti)}`$ $`=`$ $`2t_{}\mathrm{cos}\left({\displaystyle \frac{2\pi }{N}}n_F\right)`$ (26)
where $`N`$ is the number of sites.
For an in-phase domain wall ($`B0`$), the staggered field doubles the unit cell. The non-interacting Hamiltonian (22) $`H_0\left(B0\right)`$ is diagonalized by the appropriate Bloch states;
$`\psi _{k,\sigma }^{}`$ $`=`$ $`\sqrt{{\displaystyle \frac{2}{N}}}{\displaystyle \underset{j=1}{\overset{N/2}{}}}e^{+ik\left(2ja\right)}W_{j,\sigma }\left(k\right)`$ (27)
$`W_{j,\sigma }\left(k\right)`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{1+\left|f_{k,\sigma }\right|^2}}}\left[c_{2j,\sigma }^{}+f_{k,\sigma }e^{ika}c_{2j1,\sigma }^{}\right]`$ (28)
$`f_{k,\sigma }`$ $`=`$ $`{\displaystyle \frac{\sigma \left(\frac{B}{t}\right)\sqrt{\left(\frac{B}{t}\right)^2+\mathrm{cos}^2\left(k\right)}}{\mathrm{cos}\left(k\right)}}`$ (29)
The resulting energy spectrum for the in-phase domain wall is
$`E_n^{(in)}=\pm 2t_{}\sqrt{\left({\displaystyle \frac{B}{t_{}}}\right)^2+\mathrm{cos}^2\left({\displaystyle \frac{2\pi }{N}}n\right)}\mu _F^{(in)}`$ (30)
$`\text{(}{\displaystyle \frac{N}{4}}n{\displaystyle \frac{N}{4}}\text{)}`$ (31)
$$\delta =\frac{2n_F}{N}$$
(32)
In this context we comment that for a Hubbard model in an external staggered magnetic field, unlike the case of staggered charge potential/interaction, there is no charge gap opening at 1/4 filling. For simplicity, the rest of the formulas are written for the case where the odd-sites are with the magnetic field anti-parallel to the spin. Moreover, we remind the reader that we assume $`BJt_{}`$. Well below half filling, (i.e., $`t_{}\mathrm{cos}\left(k\right)>0`$),
$`f_{k,}`$ $`=`$ $`{\displaystyle \frac{B+\sqrt{B^2+t_{}{}_{}{}^{2}\mathrm{cos}_{}^{2}\left(k\right)}}{t_{}\mathrm{cos}\left(k\right)}}`$ (33)
$``$ $`1{\displaystyle \frac{B}{t_{}\mathrm{cos}\left(k\right)}}+๐ช\left({\displaystyle \frac{B}{t_{}\mathrm{cos}\left(k\right)}}\right)^2`$ (34)
The gain in magnetic energy per $`k`$-state is due to the difference between the occupation probability of odd and even sites for a given electron spin, (this is the main difference between 1DEG in an in-phase vs. anti-phase domain wall),
$`\left|c_{2j,\sigma }^{}c_{2j,\sigma }_k\right|^2`$ $`\left|c_{2j1,\sigma }^{}c_{2j1,\sigma }_k\right|^2`$ (35)
$``$ $`{\displaystyle \underset{j=1}{\overset{N/2}{}}}{\displaystyle \frac{2}{N}}{\displaystyle \frac{1}{1+\left|f_{k,\sigma }\right|^2}}\left(2{\displaystyle \frac{B}{t_{}\mathrm{cos}\left(k\right)}}\right)`$ (36)
$``$ $`\left({\displaystyle \frac{B}{t_{}\mathrm{cos}\left(k\right)}}\right)+๐ช\left({\displaystyle \frac{B}{t_{}\mathrm{cos}\left(k\right)}}\right)^2`$ (37)
Therefore, for filling up to $`k_F`$, the magnetic interaction energy gain for an in-phase domain wall in comparison with an anti-phase domain wall is given approximately by,
$`\mathrm{\Delta }E^{mag}\left(\delta \right)`$ $``$ $`2B\left[2{\displaystyle _{\pi \delta }^{\pi \delta }}{\displaystyle \frac{B}{t_{}\mathrm{cos}\left(k\right)}}๐k\right]`$ (38)
$`=`$ $`2B\left[2\left({\displaystyle \frac{B}{t_{}}}\right)\mathrm{ln}\left({\displaystyle \frac{1+\mathrm{sin}\left(\pi \delta \right)}{1\mathrm{sin}\left(\pi \delta \right)}}\right)\right]`$ (39)
$``$ $`{\displaystyle \frac{J}{4}}\left[{\displaystyle \frac{J}{t_{}}}\mathrm{ln}\left({\displaystyle \frac{1+\mathrm{sin}\left(\pi \delta \right)}{1\mathrm{sin}\left(\pi \delta \right)}}\right)\right]`$ (40)
(Note that the limit $`t_{}0`$ of the previous section cannot be recovered since we used approximations based on $`t_{}BJ/4`$).
We are now in a position to give a first analytical estimate of the critical transition point filling $`\delta _c`$ determined given by (using (16) and (40))
$`0`$ $`=`$ $`\mathrm{\Delta }E_{}\left(\delta _c\right)E_{}^{kin}+\mathrm{\Delta }E^{mag}`$ (41)
$``$ $`{\displaystyle \frac{J}{2}}\left(12\delta _c\right)+{\displaystyle \frac{J}{4}}\left[{\displaystyle \frac{J}{t}}\mathrm{ln}\left({\displaystyle \frac{1+\mathrm{sin}\left(\pi \delta _c\right)}{1\mathrm{sin}\left(\pi \delta _c\right)}}\right)\right]`$ (42)
The geometrical solution for $`\frac{J}{t}=\frac{1}{3},\frac{1}{4},\frac{1}{6},\frac{1}{8},\frac{1}{10}`$ is plotted in figure-2
The main qualitative result is that electron hoping fluctuations along a partially filled domain wall inhibit the effective magnetic interaction energy across the domain wall. As a result, an anti-phase domain wall groundstate configuration can be obtained beyond $`\delta =\frac{1}{6}`$, depending on the value of $`J/t`$. For $`\frac{1}{3}\frac{J}{t}\frac{1}{10}`$ we find $`0.3<\delta _c<0.4`$. Note that the critical filling fraction $`\delta _c`$ is not very sensitive to the value of $`J/t`$. Most importantly, it is above the value $`\delta 0.25`$ of anti-phase stripes in $`\left(LaNd\right)SrCuO`$ systems.
### B Large $`Ut`$ model, and a second estimate of the critical domain wall filling $`\delta _c`$
In this subsection we examine the competition between anti-phase and in-phase domain wall configurations for strongly interacting electrons, $`Ut`$, along the domain wall at arbitrary filling fraction.
The exact ground-state energy of the Hubbard model (20) is well known for the case $`B=0`$ from Bethe ansatz solution. Unfortunately, we do not know of an established solution for the groundstate energy of a one dimensional Hubbard model in a staggered magnetic field $`B0`$. In particular, when $`B=0`$ the Hubbard interaction in momentum space takes the form $`Un_kn_k`$, and the $`U=\mathrm{}`$ limit is effectively captured by imposing a restriction of one electron per $`k`$state. But such a simple representation does not exist in terms of the $`W_{j,\sigma }\left(k\right)`$ basis (27) in a staggered external magnetic field. Therefore, we here develope a way to approximate the two competing wall configuration energies at the same level of approximation.
We wish to tress that the qualitative effect of longitudinal hoping fluctuations - to inhibit the effective magnetic interaction energy across the domain wall - was already established in the preceding subsection. The purpose of the current subsection is to investigate the quantitative effect of including large Hubbard interactions between the electrons. Though we indeed resort to approximations at several stages of our model and calculations, we do capture the substantial effect. Once the core physics established, the degree to which our final numerical values are exact can be checked (and improved) in numerical simulations on finite systems.
Our main result is that the large Hubbard interaction, $`U`$, delimitate the critical filling fraction to be in the narrow range
$$0.28<\delta _c<0.30$$
(43)
for any value of $`\frac{J}{t}<\frac{1}{3}`$.
Below, we explain our modeling approach and calculation. We first apply the model to examine the simplest case of 1/4 filled domain wall, and show that always an anti-phase is obtained. Then, we apply the model to re-estimate the critical filling fraction for a transition to an in-phase domain wall, and thus establish the result noted above in equation (43).
#### 1 Model of characteristic unit cells
The unit cell of a stripe domain wall comprise of two lines as depicted in figure-3. Our modeling idea in this subsection is to solve exactly the Hamiltonian of isolated prototypical unit cells, and then approximate the full stripe as assembled of a collection of such unit cells. Such an approximation is building on our findings in the previous subsection, where we argued that it is not the global longitudinal mobility, but only the local kinetic fluctuations (in the occupation of odd/even numbered sites along the wall) which distinguish anti-phase and in-phase domain walls.
As shown in figure-3, in the $`U=\mathrm{}`$ limit, there are only three possible electron occupation numbers of a unit cell: one, two or zero electrons on the domain wall sites. In our model approximation, we ignore the hopping from one unit cell to another. As we shall later explain, for filling fractions $`\delta \frac{1}{4}`$ it will suffice to consider only unit cells with one and two electrons on the domain wall sites (i.e., as in figure-3(a1,a2 and b1,b2)).
We are interested only in the effect of hopping dynamics $`t_{}`$ of electron along the domain wall. Therefore we examine first the unit cells with one hole and one electron on the domain wall sites. Thereโs spin exchange interactions of strength $`J`$ between nearest-neighbor electrons. Yet, to zeroth level of approximation, the spin configuration is assumed to be static (apart from $`t_{}`$ electron hoping). Thus, we consider only the electron configurations of an anti-phase (a1,a1โ) and in-phase (b1,b1โ) domain wall unit cells.
Consider the case of unit cells with a single hole and spin-$``$ electron hopping between the domain wall sites. We observe that while the anti-phase domain wall states (a1,a1โ) are degenerate, the in-phase domain wall states (b1 and b1โ) have an energy difference of magnitude $`2\epsilon `$ (where $`\epsilon J`$ is the energy difference between parallel and anti-parallel neighboring spin states). Thus, the Hamiltonian of a single electron in an anti-phase domain wall unit cell is
$$H_{\text{hole-cell}}^{anti}=\left(\begin{array}{cc}0& t_{}\\ t_{}& 0\end{array}\right)$$
(44)
with groundstate energy $`t`$, while for an in-phase domain wall
$$H_{\text{hole-cell}}^{in}=\left(\begin{array}{cc}\epsilon & t_{}\\ t_{}& +\epsilon \end{array}\right)$$
(45)
with groundstate energy $`\sqrt{\left(\epsilon ^2+t_{}^2\right)}\left[t_{}+\frac{1}{2}\left(\frac{\epsilon }{t_{}}\right)\epsilon \right]`$. (Note that the magnetic interaction energy is already fully accounted for in the eigenvalues, so it shouldnโt be counted again). Thus we find that, for a stripe unit cell with one hole and one electron on the domain wall sites, the ground state magnetic interaction energy difference between an anti-phase and in-phase domain wall is
$`\mathrm{\Delta }E_{\text{hole-cell}}^{mag}`$ $`=`$ $`E_{\text{hole-cell}}^{anti}E_{\text{hole-cell}}^{in}`$ (46)
$`=`$ $`t_{}+\sqrt{\left(\epsilon ^2+t_{}^2\right)}`$ (47)
$``$ $`\{\begin{array}{cc}+\frac{1}{2}\left(\frac{\epsilon }{t_{}}\right)\epsilon \hfill & \text{ for }t_{}J\hfill \\ +\epsilon \hfill & \text{ for }t_{}0\hfill \end{array}`$ (50)
The above needs to be divided by $`2`$ in order to get the energy difference per site, since the unit cell has two sites along the domain wall. In the static limit $`t_{}0`$, we recover equation (17) of section-III, $`\frac{\mathrm{\Delta }E^{mag}\left(\delta =1/4\right)}{N}+\frac{\epsilon }{2}`$. But for the rest of the paper, we focus on the experimentally more relevant limit $`t_{}J`$.
We now proceed to demonstrate first the essence of our approach for the simplest case of $`\delta =\frac{1}{4}`$ filling.
#### 2 The special case of 1/4 filling
In section-III, we have shown that dynamics of only transverse interactions (hole hopping and Heisenberg interaction) will not suffice to stabilize an anti-phase domain wall for $`\delta >1/6`$. Therefore, by investigating the case of $`\delta =1/4`$ we can already answer the question of principle about the effect of electron dynamics along a stripe domain wall. Moreover, it is a case of particular interest for stripes in HTc, which are found with a filling fraction very near $`\delta =1/4`$.
In the limit of large on-site interaction $`Ut`$, at $`\delta =1/4`$ there is effectively one electron for every two sites along the wall. Moreover, to mimic the expected effect of coulomb interactions, it makes sense to supplement the model with a repulsive interaction also between near-neighbor holes (i.e., a $`tUV`$ model along the domain wall). Thus, in the absence of kinetic hoping dynamics along the domain wall, the groundstate of the one dimensional electron gas is a periodic charge density wave (CDW) with one electron per every second site. Even with the inclusion of kinetic fluctuations, the same CDW still dominate the electron correlations at low temperatures. As a result, we argue that the occurrence of unit cells with an occupation of zero or two electrons on the domain wall sites (as in figure-3(a2,a3)) are rare events and thus their contribution may be safely neglected.
Therefore, a two-line stripes segment with one hole and one electron on the domain wall sites can be regarded as the characteristic unit cell along a 1/4 filled domain wall (as in figure-3(a1,b1)). By solving such unit cell model exactly, we are able to account for the effect of local kinetic fluctuations along the domain wall. Moreover, since we are interested only in the difference between anti-phase and in-phase wall, and since the net magnetic field averages to zero along both anti-phase and in-phase wall, it is intuitively expected that the significant difference is not in the net electron mobility but mostly in the local kinetic fluctuations (Though magnetic scattering from AFM environment, which we neglect, may also differ).
Let us recall first the result of section-III. In the absence of hopping along the wall, the average transverse kinetic energy per site of an anti-phase domain wall in comparison with an in-phase domain wall was (using equation (16))
$`\mathrm{\Delta }E_{}^{kin}\left(\delta ={\displaystyle \frac{1}{4}}\right)={\displaystyle \frac{E_{}^{kin}}{N}}={\displaystyle \frac{\epsilon }{2}}\left(12\delta \right)={\displaystyle \frac{\epsilon }{4}}`$
and the competing transverse magnetic energy due to Heisenberg interaction with the AFM environment was (using equation (17))
$`\mathrm{\Delta }E_{}^{mag}\left(\delta ={\displaystyle \frac{1}{4}}\right)={\displaystyle \frac{E_{}^{mag}}{N}}=+\epsilon \left(2\delta \right)=+{\displaystyle \frac{\epsilon }{2}}.`$
Hence, it was the in-phase domain wall configuration which had the lowest energy ground state. Now, with the inclusion of the effect of the kinetic fluctuations along the domain wall, the average magnetic interaction energy is modified and (using (46) instead of (17)) the average energy difference per site is
$`{\displaystyle \frac{\mathrm{\Delta }E}{N}}`$ $`=`$ $`\mathrm{\Delta }E_{}^{kin}+{\displaystyle \frac{1}{2}}\mathrm{\Delta }E_{\text{hole-cell}}^{mag}`$ (51)
$`{\displaystyle \frac{\mathrm{\Delta }E\left(\delta =1/4\right)}{N}}`$ $``$ $`{\displaystyle \frac{\epsilon }{4}}+\left({\displaystyle \frac{\epsilon }{t_{}}}\right){\displaystyle \frac{\epsilon }{4}}<0`$ (52)
where $`N`$ is the number of sites along the domain wall, (The factor 1/2 in (51) is due to having two domain wall sites per unit cell). Therefore, we find that due to the kinetic fluctuations along the domain wall, the magnetic interaction energy gain of an in-phase domain wall configuration is reduced enough to make the anti-phase domain wall more favorable at 1/4 filling for any value of $`J\epsilon <t_{}`$.
#### 3 Transition at domain wall filling $`\delta _c`$
We now attempt to determine the contribution of the effect described above at arbitrary electron filling $`2\delta >\frac{1}{2}`$, and thus estimate the transition point.
As argued before, the 1DEG correlations on the domain wall are dominated by a CDW correlations where, at electron filling $`2\delta >1/2`$, the occurrence of two near-neighbor holes is statistically negligible compared with other configurations in the ground state. Therefore, the two-sites unit cells along the domain wall are basically only of two kinds: (a) There are $`\left(12\delta \right)`$ unit cells containing one hole, which are the type analyzed in the previous subsection. (b) There are $`\frac{1}{2}\left(12\delta \right)=\left(2\delta \frac{1}{2}\right)`$ unit cells containing no holes.
To estimate the average energy contribution per site, we can use the result of the previous subsection for the hole cells, only with the added factor $`\left(12\delta \right)`$;
$`\mathrm{\Delta }E^a\left(\delta \right)`$ $`=`$ $`\mathrm{\Delta }E_{}^{kin}+{\displaystyle \frac{1}{2}}\mathrm{\Delta }E_{\text{hole-cell}}^{mag}`$ (53)
$`=`$ $`{\displaystyle \frac{\epsilon }{4}}\left[1\left({\displaystyle \frac{\epsilon }{t_{}}}\right)\right]\left(12\delta \right)`$ (54)
and add the contribution of magnetic energy difference per site due to unit cells with no holes,
$$\mathrm{\Delta }E^b\left(\delta \right)=\frac{1}{2}E_{\text{2e-cell}}^{mag}=+\frac{1}{2}\left(2\delta \frac{1}{2}\right)2\epsilon .$$
(55)
The preferred groundstate configuration is determined by
$`{\displaystyle \frac{\mathrm{\Delta }E\left(\delta \right)}{N}}`$ $`=`$ $`\mathrm{\Delta }E^a\left(\delta \right)+\mathrm{\Delta }E^b\left(\delta \right)`$ (56)
$`=`$ $`{\displaystyle \frac{\epsilon }{4}}\left[1\left({\displaystyle \frac{\epsilon }{t_{}}}\right)\right]\left(12\delta \right)+\left(2\delta {\displaystyle \frac{1}{2}}\right)\epsilon .`$ (57)
A transition from anti-phase to in-phase domain wall occurs when
$`\mathrm{\Delta }E\left(\delta \right)>0`$
Therefore, the critical filling fraction $`\delta _c`$ is given by
$`0`$ $`=`$ $`{\displaystyle \frac{\epsilon }{2}}\left[1\left({\displaystyle \frac{\epsilon }{t_{}}}\right)\right]\left(12\delta _c\right)+2\left(2\delta _c{\displaystyle \frac{1}{2}}\right)\epsilon `$ (58)
$`\delta _c`$ $`=`$ $`{\displaystyle \frac{\frac{3}{2}\frac{1}{2}\left(\frac{\epsilon }{t_{}}\right)}{5\left(\frac{\epsilon }{t_{}}\right)}}`$ (59)
The predicted $`\delta _c`$ for several values of $`\frac{\epsilon }{t}`$ is given below,
$`\delta _c\left({\displaystyle \frac{\epsilon }{t}}={\displaystyle \frac{1}{3}}\right)`$ $`=`$ $`\mathrm{0.\hspace{0.17em}285\hspace{0.17em}71}`$
$`\delta _c\left({\displaystyle \frac{\epsilon }{t}}={\displaystyle \frac{1}{6}}\right)`$ $`=`$ $`\mathrm{0.\hspace{0.17em}293\hspace{0.17em}1}`$
$`\delta _c\left({\displaystyle \frac{\epsilon }{t}}0\right)`$ $``$ $`\mathrm{0.\hspace{0.17em}3}`$
We find that $`0.28<\delta _c<0.3`$, (for $`\frac{J}{t}<\frac{1}{3}`$), and is only weakly sensitive to the value of $`\frac{J}{t}`$.
## V Connection with a Landau theory of stripes order
The properties of a general Ginzburg-Landau free energy (61) of stripes were previously investigated in. Specifically, an ordered stripe phase is a unidirectional density wave which consists of coupled spin-density wave (SDW) and charge-density wave (CDW) order parameters.
$`_{q,k}`$ $`=`$ $`r_{1\mathrm{s}}\left|๐_\stackrel{}{q}\right|^2+r_{2\mathrm{s}}\left|๐_{\stackrel{}{q}+\stackrel{}{k}}\right|^2+r_\mathrm{c}\left|\rho _\stackrel{}{k}\right|^2`$ (61)
$`+\gamma [๐_\stackrel{}{q}^{}๐_{\stackrel{}{q}+\stackrel{}{k}}\rho _\stackrel{}{k}^{}+\mathrm{c}.\mathrm{c}.]+\mathrm{},`$
where $`\rho _\stackrel{}{k}`$ is the complex-valued CDW amplitude with wave vector $`\stackrel{}{k}`$, $`\rho _\stackrel{}{k}^{}\rho _\stackrel{}{k}`$, and similarly $`๐_\stackrel{}{q}`$ is a complex vector amplitude of the SDW. The quartic (and higher order) terms required for stability are omitted.
Zachar et al. have considered the phase transition between a stripe phase and a high-temperature disordered state, as involving only a single SDW wave vector (i.e., $`๐_\stackrel{}{q}^{}=๐_{\stackrel{}{q}+\stackrel{}{k}}`$). The existence of the cubic $`\gamma `$ term, coupling these two order parameters in the Landau free energy, drives the period of the SDW to be twice that of the CDW, and the absence of any net AFM ordering is equivalent to the statement that the stripes are topological.
In contrast, it was shown by Pryadko et al., that the same sort of cubic term in a Landau theory of the transition from a homogeneous ordered antiferromagnetic phase to a stripe ordered phase produces a state in which the antiferromagnetic magnetization does not change its sign between the domains, i.e. the stripes are non-topological.
When investigating the transition from a well-developed antiferromagnet with a modulation vector $`\stackrel{}{\pi }=(\pi ,\pi )`$ to an incommensurate modulated phase, we must account for both the original AFM order parameter $`๐_\stackrel{}{\pi }`$ (which cannot be assumed small), and the spin density wave $`๐_{\stackrel{}{\pi }+\stackrel{}{k}}`$. The most relevant terms in the Landau expansion of the effective free energy are then
$``$ $`=`$ $`r_{1\mathrm{s}}\left|๐_\stackrel{}{\pi }\right|^2+r_{2\mathrm{s}}\left|๐_{\stackrel{}{\pi }+\stackrel{}{k}}\right|^2+r_\mathrm{c}\left|\rho _\stackrel{}{k}\right|^2`$ (63)
$`+\gamma [๐_\stackrel{}{\pi }^{}๐_{\stackrel{}{\pi }+\stackrel{}{k}}\rho _\stackrel{}{k}^{}+\mathrm{c}.\mathrm{c}.]+\mathrm{},`$
where a finite $`r_{1\mathrm{s}}<0`$ is assumed as given. This expression implies that an instability in either spin or charge sector generates both spin- and charge-density waves at the wave vectors $`\stackrel{}{q}=\stackrel{}{\pi }+\stackrel{}{k}`$ and $`\stackrel{}{k}`$, respectively. Near the transition the magnitude of the incommensurate peak is necessarily much smaller then the commensurate AFM modulation $`|๐_{\stackrel{}{\pi }+\stackrel{}{k}}||๐_\stackrel{}{\pi }|`$. It is easy to see that this corresponds to in-phase domain walls; The derived relationship between $`\stackrel{}{q}`$ and $`\stackrel{}{k}`$ implies that the periods of spin and charge modulation are equal.
Our analysis in this paper supply microscopic insight to the Landau theory results. First, we find that the possibility of both topological and non-topological stripe phases can indeed be realized within a single microscopic model (as anti-phase and in-phase domain wall groundstates respectively).
Second, we find that non-topological stripes (corresponding to in-phase domain wall state) are indeed established due to enhanced antiferromagnetic interactions (as first speculated by Castro-Neto). Yet, thus far the connection is more heuristic than rigorous.
Furthermore, after analyzing the effect higher order derivative terms in a gradient expansion of the Ginzburg-Landau free energy, Pryadko et al. argue that: When there is no frustration, topological stripes are not established in the ground state. However they speculate that frustration, such as competing first and second neighbor interactions, or opposite sign terms in the gradient expansion of the Ginzburg-Landau model (i.e. below a Lifshitz point), can stabilize topological collinear domain walls. In other words, topological stripes are a consequence of physics on short length scale, and they do not appear in a theory that considers only long-distance or low-energy physics.
Pryadko et al. speculate that the missing frustrating ingredient may be due to inter-stripe interactions such as long range coulomb interactions which are expected to exist between the charged domain walls. Our analysis in this paper demonstrate that such interactions are not required. We have shown that local single stripe dynamic interactions, (albeit involving finite strength competing interactions, i.e., expectantly beyond the reach of perturbative gradient expansion) are sufficient to realize both alternative possible topological magnetic order across the charge line.
## VI Estimate of spin-wave velocity
In this section, we use our microscopic domain wall model to estimate spin-wave velocity reduction
$$\alpha _v=\frac{v_{}}{v_0},$$
(64)
where $`v_{}`$ is the spin-wave velocity perpendicular to the stripes (i.e., along the modulation direction), and $`v_0`$ is the spin-wave velocity in the undoped parent antiferromagnetic material. Equivalently, it is the predicted spin-wave velocity anisotropy in stripes (where $`v_0`$ is the same as the velocity along the stripes). Castro-Neto & Hone were the first to point out that within the stripes model, one of the main effects of the hole rich lines would be to weaken the effective exchange interaction between spins on either side across the domain wall. Thus, a cardinal parameter in the model is the effective weakening factor of the magnetic interaction $`J_W`$ across the stripes domain wall as compared with the Heisenberg interaction $`J`$ within the hole free AFM regions (and hence along the stripes).
$$\alpha _J=\frac{J_W}{J}$$
(65)
The key new ingredient is our analytical estimate of $`\alpha _J`$. Consequently, while previous model estimates always involved fitting of free parameters, we are able to estimate analytically the spin-wave velocity anisotropy (without any free parameters).
Neto&Hone proposed an effective anisotropic Heisenberg model for describing the low temperature spin dynamics in a striped Cu-O plane. Within the anisotropic Heisenberg model , they derived the spin-wave velocity anisotropy
$$v_{}\sqrt{\frac{\alpha _J}{2}}v_{0.}$$
(66)
i.e., $`\alpha _v=\sqrt{\frac{\alpha _J}{2}}`$. By fitting other consequences of their model to experimental results, Neto&Hone extracted the value
$`\alpha _J0.01.`$
The fitted value of $`\alpha _J`$, together with (66) $`\mathrm{}v_{}50meV`$ร
. As discussed by Tranquada et al. , this value is much too small to be compatible with inelastic neutron scattering experiments. It is further incompatible with the range of energies over which stripes incommensurability is observed.
Aiming specifically to model the interaction between narrow stripe AFM regions, Tworzydlo et al. made predictions based on models of coupled narrow spin ladders (e.g., 2-leg or 3-leg ladders). The resulting expressions for the spin wave velocity anisotropy were
$$\alpha _v=\{\begin{array}{cc}\sqrt{\frac{2\alpha _J}{1+\alpha _J}}& \text{ for 2-leg ladder}\\ \sqrt{\frac{3\alpha _J}{1+2\alpha _J}}& \text{ for 3-leg ladder}\end{array}$$
(67)
In terms of 2-leg ladder model, Tranquada et al. proclaimed to fit the experiments well with a value of
$`\alpha _J0.35.`$
giving
$`\alpha _v\mathrm{0.\hspace{0.17em}72}`$
In principle, the ladder model results are quite sensitive to the width of the stripes (which affects the ladder spin gap magnitude). Yet accidentally, the above value of $`\alpha _J0.35`$ would give $`\alpha _v\mathrm{0.\hspace{0.17em}78}`$ for 3-leg ladder, i.e., practically the same as for 2-leg.
Our microscopic model is based solely on the local physics in the neighborhood of the hole line. As such, it is insensitive to details of the width of the AFM spin domains. We argue that, by definition, the energy difference between alternative spin configurations across the domain wall is a measure of the effective spin interaction across the domain wall in any appropriate low energy theory. Therefore, using our microscopic model calculation, we suggest that
$$J_W=\mathrm{\Delta }E\left(\delta \right)=E^{in}E^{anti}$$
(68)
were $`\mathrm{\Delta }E\left(\delta \right)`$ is given in equation (56)
$`\mathrm{\Delta }E\left(\delta \right){\displaystyle \frac{\epsilon }{2}}\left[1\left({\displaystyle \frac{\epsilon }{t}}\right)\right]\left(12\delta \right)+2\left(2\delta {\displaystyle \frac{1}{2}}\right)\epsilon `$
(where as discussed previously, $`\epsilon J`$). Thus we can calculate $`\alpha _J`$ and then, following the anisotropic Heisenberg model reasoning of Castro-Neto & Hone, use (66) to calculate the spin-wave velocity anisotropy.
For the particular case of stripes observed in LaNdSrCuO $`\delta 1/4`$, we obtain
$$J_W\mathrm{\Delta }E\left(\delta =\frac{1}{4}\right)\frac{J}{4}\left[1\left(\frac{J}{t}\right)\right]$$
(69)
Substituting $`\frac{1}{3}>\frac{J}{t}>\frac{1}{10}`$, we obtain
$$\mathrm{0.\hspace{0.17em}17}<\alpha _J\left(\delta 1/4\right)<\mathrm{0.\hspace{0.17em}23}$$
(70)
Importantly, note that $`\alpha _J\left(\delta 1/4\right)`$ is rather insensitive to the variation of $`\frac{J}{t}<\frac{1}{3}`$. (where $`\delta 1/4`$ is the supposed electron filling fraction of the stripe domain walls in Copper-Oxides).
Using equation (66) thus give an estimate of
$$0.29<\alpha _v<\mathrm{0.\hspace{0.17em}34}$$
(71)
or using equation (67) give an estimate of
$$\begin{array}{cc}0.67<\alpha _v<0.71& \text{ for 2-leg ladder domains}\\ 0.74<\alpha _v<0.78& \text{ for 3-leg ladder domains}\end{array}$$
(72)
which fit remarkably well with the experimentally deduced spin wave velocity inhibition $`0.60<\alpha _v<0.72`$ in doped Copper-Oxides compared with the undoped parent AFM material. Note that our estimate is obtained by directly substituting the experimental value of $`v_0`$ into the analytical results for $`\alpha _J`$ (using our model equation (56)) and for $`\alpha _v`$ (from the ladder models or the anisotropic Heisenberg model), without any fitting of free parameters.
Acknowledgments: I thank Natan Andrei, Thierry Giamarchi, Baruch Horovitz, Efstratios Manousakis, John Tranquada, and particularly Steve Kivelson for fruitful discussions. This work was partly supported by the TMR#ERB4001GT97294 fellowship. |
warning/0001/hep-ph0001178.html | ar5iv | text | # An improved timeโdependent HartreeโFock approach for scalar ฯโด QFT
## I Introduction
A great effort has been devoted in the last few years in order to develop a deeper qualitative and quantitative understanding of systems described by interacting quantum fields out of equilibrium. There is a class of physical problems that requires the consistent treatment of time dependent meanโfields in interaction with their own quantum or thermal fluctuations. We may mention, among others, the problem of reheating of the universe after the inflationary era of exponential growth and cooling, and the time evolution of the scalar order parameter through the chiral phase transition, soon to be probed in the forthcoming heavyโion experiments at CERNโSPS, BNLโRHIC and CERNโLHC. In these situations, a detailed description of the timeโdependent dynamics is necessary to calculate the nonโequilibrium properties of the system. Indeed, the development of practical general techniques and the advent of faster and cheaper computers have made possible the discovery of novel and unexpected phenomena, ranging from dissipative processes via particle production to novel aspects of symmetry breaking .
From the technical point of view, it should be pointed out, first of all, that a perturbative treatment of this dynamical problem is meaningful only when the early time evolution is considered. The presence of parametric resonant bands or spinodal instabilities (in the case, respectively, of unbroken or spontaneously broken symmetries) rapidly turns the dynamics completely nonโlinear and nonโperturbative. Thus, the asymptotic evolution at late time can be consistently studied only if approximate nonโperturbative methods are applied to the problem .
Quite recently one of these schemes, namely the large $`N`$ expansion at leading order (LN) , has been used in order to clarify some dynamical aspects of the $`\varphi ^4`$ theory in $`3`$ spatial dimensions, reaching the conclusion that the nonโperturbative and nonโlinear evolution of the system might eventually produce the onset of a form of nonโequilibrium BoseโEinstein condensation (BEC) of the longโwavelength Goldstone bosons usually present in the broken symmetry phase . Another very interesting result in concerns the dynamical Maxwell construction, which reproduces the flat region of the effective potential in case of broken symmetry as asymptotic fixed points of the background evolution.
In a companion work we have addressed the question of whether a standard BEC could actually take place as time goes on, by putting the system in a finite volume (a periodic box of size $`L`$) and carefully studying the volume dependence of outโofโequilibrium features in the broken symmetry phase. We summarize here the main result contained in . The numerical solution shows the presence of a time scale $`\tau _L`$, proportional to the linear size $`L`$ of the system, at which finite volume effects start to manifest, with the remarkable consequence that the zero-mode quantum fluctuations cannot grow macroscopically large if they start with microscopic initial conditions. In fact, the size of lowโlying widths at time $`\tau _L`$ is of order $`L`$, to be compared to order $`L^{3/2}`$ for the case of standard BEC. In other words we confirmed that the linear growth of the zero mode width, as found also by the authors of , really signals the onset of a novel form of dynamical BEC, quite different from the standard one described by equilibrium finiteโtemperature field theory. This interpretation is reinforced by the characteristics of the longโwavelength fluctuationsโ spectrum.
Since after all the large $`N`$ approximation is equivalent to a Gaussian ansatz for the timeโdependent density matrix of the system , one might still envisage a scenario in which, while gaussian fluctuations would stay microscopic, nonโgaussian fluctuations would grow in time to a macroscopic size, leading to an occupation number for the zero mode proportional to the volume $`L^3`$ of the system. Therefore, in order to go beyond the gaussian approximation, we will consider in this work a timeโdependent HF approach capable in principle of describing the dynamics of some nonโgaussian fluctuations of a single scalar field with $`\varphi ^4`$ interaction.
Before going into the details of the analysis, let us briefly summarize the main limitations and the most remarkable results of the study of a scalar field out of equilibrium within the gaussian HF scheme . First of all, this scheme has the advantage of going beyond perturbation theory, in the sense that the (numerical) solution of the evolution equations will contain arbitrary powers of the coupling constant, corresponding to a nonโtrivial resummation of the perturbative series. For this reason, the method is able to take into account the quantum backโreaction on the fluctuations themselves, which shuts off their early exponential growth. This is achieved by the standard HF factorization of the quartic interaction, yielding a time dependent selfโconsistently determined mass term, which stabilizes the modes perturbatively unstable. The detailed numerical solution of the resulting dynamical equations clearly shows the dissipation associated with particle production, as a result of either parametric amplification in case of unbroken symmetry or spinodal instabilities in case of broken symmetry, as well as the shut off mechanism outlined above.
However, the standard HF method is really not controllable in the case of a single scalar field, while it becomes exact only in the $`N\mathrm{}`$ limit. Moreover, previous approaches to the dynamics in this approximation scheme had the unlikely feature of maintaining a weak (logarithmic) cutโoff dependence on the renormalized equations of motion of the order parameter and the mode functions .
In this article we consider the case of a single scalar field (i.e. $`N=1`$). With the aim of studying the dynamics of the model with the inclusion of some nonโgaussian contributions, we introduce an improved timeโdependent HartreeโFock approach. Even if it is still based on a factorized trial wavefunction(al), it has the merit to keep the quartic interaction diagonal in momentum space, explicitly in the hamiltonians governing the evolution of each mode of the field. In this framework, issues like the static spontaneous symmetry breaking can be better understood, and the further gaussian approximation needed to study the dynamics can be better controlled. In particular, questions like outโofโequilibrium โquantum phase orderingโ and โdynamical BoseโEinstein condensationโ can be properly posed and answered within a verifiable approximation.
We also perform a detailed study of the asymptotic dynamics in infinite volume, with the aim of clarifying the issue of Maxwell construction in this approximation scheme. In fact, in the $`O(N)\mathrm{\Phi }^4`$ model at leading order, the asymptotic dynamical evolution of the mean field completely covers the spinodal region of the classical potential, which coincides with the flatness region of the effective potential. This is what is called dynamical Maxwell construction . When we use the HF approximation for the case of $`N=1`$, we find that the spinodal region and the flatness region are different and the question arise of whether a full or partial dynamical Maxwell construction still takes place.
In section II we set up the model in finite volume, defining all the relevant notations and the quantum representation we will be using to study the evolution of the system.
We introduce in section III our improved timeโdependent HartreeโFock (tdHF) approximation, which generalizes the standard gaussian self-consistent approach to nonโgaussian waveโfunctionals; we then derive the meanโfield coupled timeโdependent Schroedinger equations for the modes of the scalar field, under the assumption of a uniform condensate, see eqs (5), (6) and (7). A significant difference with respect to previous tdHF approaches concerns the renormalization of ultraviolet divergences. In fact, by means of a single substitution of the bare coupling constant $`\lambda _\text{b}`$ with the renormalized one $`\lambda `$ in the HartreeโFock hamiltonian, we obtain cut-off independent equations (apart from corrections in inverse powers, which are there due to the Landau pole). The substitution is introduced by hand, but is justified by simple diagrammatic considerations.
One advantage of not restricting a priori the self-consistent HF approximation to gaussian waveโfunctionals, is in the possibility of a better description of the vacuum structure in case of broken symmetry. In fact we can show quite explicitly that, in any finite volume, in the ground state the zeroโmode of $`\varphi `$ field is concentrated around the two vacua of the broken symmetry, driving the probability distribution for any sufficiently wide smearing of the field into a two peaks shape. This is indeed what one would intuitively expect in case of symmetry breaking. On the other hand none of this appears in a dynamical evolution that starts from a distribution localized around a single value of the field in the spinodal region, confirming what already seen in the large $`N`$ approach . More precisely, within a further controlled gaussian approximation of our tdHF approach, one observes that initially microscopic quantum fluctuations never becomes macroscopic, suggesting that also nonโgaussian fluctuations cannot reach macroscopic sizes. As a simple confirmation of this fact, consider the completely symmetric initial conditions $`\varphi =\dot{\varphi }=0`$ for the background: in this case we find that the dynamical equations for initially gaussian field fluctuations are identical to those of large $`N`$ (apart for a rescaling of the coupling constant by a factor of three; cfr. ref. ), so that we observe the same asymptotic vanishing of the effective mass. However, this time no interpretation in terms of Goldstone theorem is possible, since the broken symmetry is discrete; rather, if the width of the zeroโmode were allowed to evolve into a macroscopic size, then the effective mass would tend to a positive value, since the mass in case of discrete symmetry breaking is indeed larger than zero.
Anyway, also in the gaussian HF approach, we do find a whole class of cases which exhibit the time scale $`\tau _L`$. At that time, finite volume effects start to manifest and the size of the lowโlying widths is of order $`L`$. We then discuss why this undermines the selfโconsistency of the gaussian approximation, imposing the need of further study, both analytical and numerical.
In section IV we study the asymptotic evolution in the broken symmetry phase, in infinite volume, when the expectation value starts within the region between the two minima of the potential. We are able to show by precise numerical simulations, that the fixed points of the background evolution do not cover the static flat region completely. On the contrary, the spinodal region seems to be absolutely forbidden for the late time values of the mean field. Thus, as far as the asymptotic evolution is concerned, our numerical results lead to the following conclusions. We can distinguish the points lying between the two minima in a fashion reminiscent of the static classification: first, the values satisfying the property $`v/\sqrt{3}<\left|\overline{\varphi }_{\mathrm{}}\right|v`$ are metastable points, in the sense that they are fixed points of the background evolution, no matter which initial condition comprised in the interval $`(v,v)`$ we choose for the expectation value $`\overline{\varphi }`$; secondly, the points included in the interval $`0<\left|\overline{\varphi }_{\mathrm{}}\right|<v/\sqrt{3}`$ are unstable points, because if the mean field starts from one of them, after an early slow rolling down, it starts to oscillate with decreasing amplitude around a point inside the classical metastable interval. Obviously, $`\overline{\varphi }=v`$ is the point of stable equilibrium, and $`\overline{\varphi }=0`$ is a point of unstable equilibrium. Actually, it should be noted that our data do not allow a precise determination of the border between the dynamical unstable and metastable regions; thus, the number we give here should be looked at as an educated guess inspired by the analogous static classification and based on considerations about the solutions of the gap equation \[see eq. (36)\]
Finally, in section VI we give a brief summary of the results presented in this article and we outline some interesting open questions that need more work before being answered properly.
## II Cutoff field theory
Let us consider the scalar field operator $`\varphi `$ and its canonically conjugated momentum $`\pi `$ in a $`D`$dimensional periodic box of size $`L`$ and write their Fourier expansion as customary
$$\varphi (x)=L^{D/2}\underset{k}{}\varphi _ke^{ikx},\varphi _k^{}=\varphi _k$$
$$\pi (x)=L^{D/2}\underset{k}{}\pi _ke^{ikx},\pi _k^{}=\pi _k$$
with the wavevectors $`k`$ naturally quantized: $`k=(2\pi /L)n`$, $`n^D`$.
The canonical commutation rules are $`[\varphi _k,\pi _k^{}]=i\delta _{kk^{}}^{(D)}`$, as usual. The introduction of a finite volume should be regarded as a regularization of the infrared properties of the model, which allows to โcountโ the different field modes and is needed especially in the case of broken symmetry.
To keep control also on the ultraviolet behavior and manage to handle the renormalization procedure properly, we restrict the sums over wavevectors to the points lying within the $`D`$dimensional sphere of radius $`\mathrm{\Lambda }`$, that is $`k^2\mathrm{\Lambda }^2`$, with $`๐ฉ=\mathrm{\Lambda }L/2\pi `$ some large integer. Till both the cutโoffs remain finite, we have reduced the original fieldโtheoretical problem to a quantumโmechanical framework with finitely many (of order $`๐ฉ^{D1}`$) degrees of freedom.
The $`\varphi ^4`$ Hamiltonian is
$$\begin{array}{cc}\hfill H& =\frac{1}{2}d^Dx\left[\pi ^2+(\varphi )^2+m_\text{b}^2\varphi ^2+\lambda _\text{b}\varphi ^4\right]=\hfill \\ & =\frac{1}{2}\underset{k}{}\left[\pi _k\pi _k+(k^2+m_\text{b}^2)\varphi _k\varphi _k\right]+\frac{\lambda }{4}\underset{k_1,k_2,k_3,k_4}{}\varphi _{k_1}\varphi _{k_2}\varphi _{k_3}\varphi _{k_4}\delta _{k_1+k_2+k_3+k_4,0}^{(D)}\hfill \end{array}$$
where $`m_\text{b}^2`$ and $`\lambda _\text{b}`$ are the bare parameters and depend on the UV cutoff $`\mathrm{\Lambda }`$ in such a way to guarantee a finite limit $`\mathrm{\Lambda }\mathrm{}`$ for all observable quantities. It should be noted here that, being the theory trivial (as is manifest in the resummed oneโloop approximation due to the Landau pole) the ultraviolet cutโoff should be kept finite and much smaller than the renormalon singularity. In this case, we must regard the $`\varphi ^4`$ model as an effective lowโenergy theory (here lowโenergy means practically all energies below Planckโs scale, due to the large value of the Landau pole for renormalized coupling constants of order one or less).
We shall work in the wavefunction representation where $`\phi |\mathrm{\Psi }=\mathrm{\Psi }(\phi )`$ and
$$(\varphi _0\mathrm{\Psi })(\phi )=\phi _0\mathrm{\Psi }(\phi ),(\pi _0\mathrm{\Psi })(\phi )=i\frac{}{\phi _0}\mathrm{\Psi }(\phi )$$
while for $`k>0`$ (in lexicographic sense)
$$(\varphi _{\pm k}\mathrm{\Psi })(\phi )=\frac{1}{\sqrt{2}}\left(\phi _k\pm i\phi _k\right)\mathrm{\Psi }(\phi ),(\pi _{\pm k}\mathrm{\Psi })(\phi )=\frac{1}{\sqrt{2}}\left(i\frac{}{\phi _k}\pm \frac{}{\phi _k}\right)\mathrm{\Psi }(\phi )$$
Notice that by construction the variables $`\phi _k`$ are all real.
In practice, the problem of studying the dynamics of the $`\varphi ^4`$ field out of equilibrium consists now in trying to solve the time-dependent Schroedinger equation given an initial wavefunction $`\mathrm{\Psi }(\phi ,t=0)`$ that describes a state of the field far away from the vacuum. This approach could be very well generalized in a straightforward way to mixtures described by density matrices, as done, for instance, in . Here we shall restrict to pure states, for sake of simplicity and because all relevant aspects of the problem are already present in this case.
We shall consider here the time-dependent HartreeโFock (tdHF) approach (an improved version with respect to what is presented, for instance, in ), being the large $`N`$ expansion to leading order treated in another work . In fact these two methods are very closely related (see, for instance in ). However, before passing to any approximation, we would like to stress that the following rigorous result can be immediately established in this model with both UV and IR cutoffs.
### A A rigorous result: the effective potential is convex
This is a well known fact in statistical mechanics, being directly related to stability requirements. It would therefore hold also for the field theory in the Euclidean functional formulation. In our quantumโmechanical context we may proceed as follow. Suppose the field $`\varphi `$ is coupled to a uniform external source $`J`$. Then the ground state energy $`E_0(J)`$ is a concave function of $`J`$, as can be inferred from the negativity of the second order term in $`\mathrm{\Delta }J`$ of perturbation around any chosen value of $`J`$. Moreover, $`E_0(J)`$ is analytic in a finite neighborhood of $`J=0`$, since $`J\varphi `$ is a perturbation โsmallโ compared to the quadratic and quartic terms of the Hamiltonian. As a consequence, this effective potential $`V_{\text{eff}}(\overline{\varphi })=E_0(J)J\overline{\varphi }`$, $`\overline{\varphi }=E_0^{}(J)=\varphi _0`$, that is the Legendre transform of $`E_0(J)`$, is a convex analytic function in a finite neighborhood of $`\overline{\varphi }=0`$. In the infrared limit $`L\mathrm{}`$, $`E_0(J)`$ might develop a singularity in $`J=0`$ and $`V_{\text{eff}}(\overline{\varphi })`$ might flatten around $`\overline{\varphi }=0`$. Of course this possibility would apply in case of spontaneous symmetry breaking, that is for a doubleโwell classical potential. This is a subtle and important point that will play a crucial role later on, even if the effective potential is relevant for the static properties of the model rather than the dynamical evolution out of equilibrium that interests us here. In fact such evolution is governed by the CTP effective action and one might expect that, although nonโlocal in time, it asymptotically reduces to a multiple of the effective potential for trajectories of $`\overline{\varphi }(t)`$ with a fixed point at infinite time. In such case there should exist a oneโtoโone correspondence between fixed points and minima of the effective potential.
## III Time-dependent HartreeโFock
In order to follow the time evolution of the nonโgaussian quantum fluctuations we consider in this section a timeโdependent HF approximation capable in principle of describing the dynamics of nonโgaussian fluctuations of a single scalar field with $`\varphi ^4`$ interaction.
We examine in this work only states in which the scalar field has a uniform, albeit possibly timeโdependent expectation value. In a tdHF approach we may then start from a wavefuction of the factorized form (which would be exact for free fields)
$$\mathrm{\Psi }(\phi )=\psi _0(\phi _0)\underset{k>0}{}\psi _k(\phi _k,\phi _k)$$
(1)
The dependence of $`\psi _k`$ on its two arguments cannot be assumed to factorize in general since space translations act as $`SO(2)`$ rotations on $`\phi _k`$ and $`\phi _k`$ (hence in case of translation invariance $`\psi _k`$ depends only on $`\phi _k^2+\phi _k^2`$). The approximation consists in assuming this form as valid at all times and imposing the stationarity condition on the action
$$\delta ๐ti_tH=0,\mathrm{\Psi }(t)||\mathrm{\Psi }(t)$$
(2)
with respect to variations of the functions $`\psi _k`$. To enforce a uniform expectation value of $`\varphi `$ we should add a Lagrange multiplier term linear in the single modes expectations $`\phi _k`$ for $`k0`$. The multiplier is then fixed at the end to obtain $`\phi _k=0`$ for all $`k0`$. Actually one may verify that this is equivalent to the simpler approach in which $`\phi _k`$ is set to vanish for all $`k0`$ before any variation. Then the only non trivial expectation value in the Hamiltonian, namely that of the quartic term, assumes the form
$$\begin{array}{cc}\hfill d^Dx\varphi (x)^4=& \frac{1}{L^D}\left[\phi _0^43\phi _0^2^2\right]+\frac{3}{2L^D}\underset{k>0}{}\left[(\phi _k^2+\phi _k^2)^22\left(\phi _k^2+\phi _k^2\right)^2\right]\hfill \\ & +\frac{3}{L^D}\left(\underset{k}{}\phi _k^2\right)^2\hfill \end{array}$$
(3)
Notice that the terms in the first row would cancel completely out for gaussian wavefunctions $`\psi _k`$ with zero mean value. The last term, where the sum extends to all wavevectors $`k`$, corresponds instead to the standard mean field replacement $`\varphi ^43\varphi ^2^2`$. The total energy of our trial state now reads
$$E=H=\frac{1}{2}\underset{k}{}\frac{^2}{\phi _k^2}+(k^2+m_\text{b}^2)\phi _k^2+\frac{\lambda _\text{b}}{4}d^Dx\varphi (x)^4$$
(4)
and from the variational principle (2) we obtain a set of simple Schroedinger equations
$$i_t\psi _k=H_k\psi _k$$
(5)
$$\begin{array}{cc}\hfill H_0& =\frac{1}{2}\frac{^2}{\phi _0^2}+\frac{1}{2}\omega _0^2\phi _0^2+\frac{\lambda _\text{b}}{4L^D}\phi _0^4\hfill \\ \hfill H_k& =\frac{1}{2}\left(\frac{^2}{\phi _k^2}+\frac{^2}{\phi _k^2}\right)+\frac{1}{2}\omega _k^2(\phi _k^2+\phi _k^2)+\frac{3\lambda _\text{b}}{8L^D}\left(\phi _k^2+\phi _k^2\right)^2\hfill \end{array}$$
(6)
which are coupled in a meanโfield way only through
$$\omega _k^2=k^2+m_\text{b}^2+3\lambda _\text{b}\mathrm{\Sigma }_k,\mathrm{\Sigma }_k=\frac{1}{L^D}\underset{\genfrac{}{}{0pt}{}{q^2\mathrm{\Lambda }^2}{qk,k}}{}\phi _q^2$$
(7)
and define the HF time evolution for the theory. By construction this evolution conserves the total energy $`E`$ of eq. (4).
It should be stressed that in this particular tdHF approximation, beside the meanโfield backโreaction term $`\mathrm{\Sigma }_k`$ of all other modes on $`\omega _k^2`$, we keep also the contribution of the diagonal scattering through the diagonal quartic terms. In fact this is why $`\mathrm{\Sigma }_k`$ has no contribution from the $`k`$mode itself: in a gaussian approximation for the trial wavefunctions $`\psi _k`$ the Hamiltonians $`H_k`$ would turn out to be harmonic, the quartic terms being absent in favor of a complete backโreaction
$$\mathrm{\Sigma }=\mathrm{\Sigma }_k+\frac{\phi _k^2+\phi _k^2}{L^D}=\frac{1}{L^D}\underset{k}{}\phi _k^2$$
(8)
Of course the quartic selfโinteraction of the modes as well as the difference between $`\mathrm{\Sigma }`$ and $`\mathrm{\Sigma }_k`$ are suppressed by a volume effect and could be neglected in the infrared limit, provided all wavefunctions $`\psi _k`$ stays concentrated on mode amplitudes $`\phi _k`$ of order smaller than $`L^{D/2}`$. This is the typical situation when all modes remain microscopic and the volume in the denominators is compensated only through the summation over a number of modes proportional to the volume itself, so that in the limit $`L\mathrm{}`$ sums are replaced by integrals
$$\mathrm{\Sigma }_k\mathrm{\Sigma }_{k^2\mathrm{\Lambda }^2}\frac{d^Dk}{(2\pi )^D}\phi _k^2$$
Indeed we shall apply this picture to all modes with $`k0`$, while we do expect exceptions for the zeroโmode wavefunction $`\psi _0`$.
The treatment of ultraviolet divergences requires particular care, since the HF approximation typically messes things up (see, for instance, ). Following the same login of the large $`N`$ approximation , we could take as renormalization condition the requirement that the frequencies $`\omega _k^2`$ are independent of $`\mathrm{\Lambda }`$, assuming that $`m_\text{b}^2`$ and $`\lambda _\text{b}`$ are functions of $`\mathrm{\Lambda }`$ itself and of renormalized $`\mathrm{\Lambda }`$independent parameters $`m^2`$ and $`\lambda `$ such that
$$\omega _k^2=k^2+m^2+3\lambda \left[\mathrm{\Sigma }_k\right]_{\text{finite}}$$
(9)
where by $`[.]_{\text{finite}}`$ we mean the (schemeโdependent) finite part of some possibly ultraviolet divergent quantity. Unfortunately this would not be enough to make the spectrum of energy differences cutoffโindependent, because of the bare coupling constant $`\lambda _\text{b}`$ in front of the quartic terms in $`H_k`$ and the difference between $`\mathrm{\Sigma }`$ and $`\mathrm{\Sigma }_k`$ \[such problem does not exist in large $`N`$ because that is a purely gaussian approximation\]. Again this would not be a problem whenever these terms become negligible as $`L\mathrm{}`$. At any rate, to be ready to handle the cases when this is not actually true and to define an ultravioletโfinite model also at finite volume, we shall by hand modify eq. (3) as follows:
$$\begin{array}{cc}\hfill \lambda _\text{b}d^Dx\varphi (x)^4=& \lambda L^D\left\{\phi _0^43\phi _0^2^2+\frac{3}{2}\underset{k>0}{}\left[(\phi _k^2+\phi _k^2)^22\left(\phi _k^2+\phi _k^2\right)^2\right]\right\}\hfill \\ & +3\lambda _\text{b}L^D\mathrm{\Sigma }^2\hfill \end{array}$$
(10)
We keep the bare coupling constant in front of the term containing $`\mathrm{\Sigma }^2`$ because that part of the hamiltonian is properly renormalized by means of the usual cactus resummation which corresponds to the standard HF approximation. On the other hand, within the same approximation, it is not possible to renormalize the part in curly brackets of the equation above, because of the factorized form (1) that we have assumed for the wavefunction of the system. In fact, the $`4`$legs vertices in the curly brackets are diagonal in momentum space; at higher order in the loop expansion, when we contract two or more vertices of this type, no sum over internal loop momenta is produced, so that all higher order perturbation terms are suppressed by volume effects. However, we know that in the complete theory, the wavefunction is not factorized and loops contain all values of momentum. This suggests that, in order to get a finite hamiltonian, we need to introduce in the definition of our model some extra resummation of Feynmann diagrams, that is not automatically contained in this selfโconsistent HF approach. The only choice consistent with the cactus resummation performed in the twoโpoint function by the HF scheme is the resummation of the $`1`$-loop fish diagram in the fourโpoint function. This amounts to the change from $`\lambda _\text{b}`$ to $`\lambda `$ and it is enough to guarantee the ultraviolet finiteness of the hamiltonian through the redefinition
$$H_0H_0+\frac{\lambda \lambda _\text{b}}{4L^D}\phi _0^4,H_kH_k+\frac{3(\lambda \lambda _\text{b})}{8L^D}\left(\phi _k^2+\phi _k^2\right)^2$$
(11)
At the same time the frequencies are now related to the widths $`\phi _k^2`$ by
$$\begin{array}{cc}\hfill \omega _k^2& =k^2+M^23\lambda L^D(\phi _k^2+\phi _k^2),k>0\hfill \\ \hfill M^2& \omega _0^2+3\lambda L^D\phi _0^2=m_\text{b}^2+3\lambda _\text{b}\mathrm{\Sigma }\hfill \end{array}$$
(12)
Apart for $`O(L^D)`$ corrections, $`M`$ plays the role of timeโdependent mass for modes with $`k0`$, in the harmonic approximation.
In this new setup the conserved energy reads
$$E=\underset{k0}{}H_k\frac{3}{4}\lambda _\text{b}L^D\mathrm{\Sigma }^2+\frac{3}{4}\lambda L^D\left[\phi _0^2^2+\underset{k>0}{}\left(\phi _k^2+\phi _k^2\right)^2\right]$$
(13)
Since the gapโlike equations (12) are stateโdependent, we have to perform the renormalization first for some reference quantum state, that is for some specific collection of wavefunctions $`\psi _k`$; as soon as $`m_\text{b}^2`$ and $`\lambda _\text{b}`$ are determined as functions $`\mathrm{\Lambda }`$, ultraviolet finiteness will hold for the entire class of states with the same ultraviolet properties of the reference state. Then an obvious consistency check for our HF approximation is that this class is closed under time evolution.
Rather than a single state, we choose as reference the family of gaussian states parametrized by the uniform expectation value $`\varphi (x)=L^{D/2}\phi _0=\overline{\varphi }`$ (recall that we have $`\phi _k=0`$ when $`k0`$ by assumption) and such that the HF energy $`E`$ is as small as possible for fixed $`\overline{\varphi }`$. Then, apart from a translation by $`L^{D/2}\overline{\varphi }`$ on $`\phi _0`$, these gaussian $`\psi _k`$ are ground state eigenfunctions of the harmonic Hamiltonians obtained from $`H_k`$ by dropping the quartic terms. Because of the $`k^2`$ in the frequencies we expect these gaussian states to dominate in the ultraviolet limit also at finite volume (as discussed above they should dominate in the infiniteโvolume limit for any $`k0`$). Moreover, since now
$$\phi _0^2=L^D\overline{\varphi }^2+\frac{1}{2\omega _0},\phi _{\pm k}^2=\frac{1}{2\omega _k},k0$$
(14)
the relation (12) between frequencies and widths turn into the single gap equation
$$M^2=m_\text{b}^2+3\lambda _\text{b}\left(\overline{\varphi }^2+\frac{1}{2L^D}\underset{q^2\mathrm{\Lambda }^2}{}\frac{1}{\sqrt{k^2+M^2}}\right)$$
(15)
fixing the self-consistent value of $`M`$ as a function of $`\overline{\varphi }`$. It should be stressed that (12) turns through eq. (14) into the gap equation only because of the requirement of energy minimization. Generic $`\psi _k`$, regarded as initial conditions for the Schroedinger equations (5), are in principle not subject to any gap equation.
The treatment now follows closely that in the large $`N`$ approximation , the only difference being in the value of the coupling, now three times larger. In fact, in case of $`O(N)`$ symmetry, the quantum fluctuations over a given background $`\mathit{\varphi }(x)=\overline{\mathit{\varphi }}`$ decompose for each $`k`$ into one longitudinal mode, parallel to $`\overline{\mathit{\varphi }}`$, and $`N1`$ transverse modes orthogonal to it; by boson combinatorics the longitudinal mode couples to $`\overline{\mathit{\varphi }}`$ with strength $`3\lambda _\text{b}/N`$ and decouple in the $`N\mathrm{}`$ limit, while the transverse modes couple to $`\overline{\mathit{\varphi }}`$ with strength $`(N1)\lambda _\text{b}/N\lambda _\text{b}`$; when $`N=1`$ only the longitudinal mode is there.
As $`L\mathrm{}`$, $`\omega _k^2k^2+M^2`$ and $`M`$ is exactly the physical mass gap. Hence it must be $`\mathrm{\Lambda }`$independent. At finite $`L`$ we cannot use this request to determine $`m_\text{b}^2`$ and $`\lambda _\text{b}`$, since, unlike $`M`$, they cannot depend on the size $`L`$. At infinite volume we obtain
$$M^2=m_\text{b}^2+3\lambda _\text{b}[\overline{\varphi }^2+I_D(M^2,\mathrm{\Lambda })],I_D(z,\mathrm{\Lambda })_{k^2\mathrm{\Lambda }^2}\frac{d^Dk}{(2\pi )^D}\frac{1}{2\sqrt{k^2+z}}$$
(16)
When $`\overline{\varphi }=0`$ this equation fixes the bare mass to be
$$m_\text{b}^2=m^23\lambda _\text{b}I_D(m^2,\mathrm{\Lambda })$$
(17)
where $`m=M(\overline{\varphi }=0)`$ may be identified with the equilibrium physical mass of the scalar particles of the infiniteโvolume Fock space without symmetry breaking (see below). Now, the coupling constant renormalization follows from the equalities
$$\begin{array}{cc}\hfill M^2& =m^2+3\lambda _\text{b}[\overline{\varphi }^2+I_D(M^2,\mathrm{\Lambda })I_D(m^2,\mathrm{\Lambda })]\hfill \\ & =m^2+3\lambda \overline{\varphi }^2+3\lambda \left[I_D(M^2,\mathrm{\Lambda })I_D(m^2,\mathrm{\Lambda })\right]_{\text{finite}}\hfill \end{array}$$
(18)
and reads when $`D=3`$
$$\frac{\lambda }{\lambda _\text{b}}=1\frac{3\lambda }{8\pi ^2}\mathrm{log}\frac{2\mathrm{\Lambda }}{m\sqrt{e}}$$
(19)
that is the standard result of the oneโloop renormalization group . When $`D=1`$, that is a $`1+1`$dimensional quantum field theory, $`I_D(M^2,\mathrm{\Lambda })I_D(m^2,\mathrm{\Lambda })`$ is already finite and the dimensionfull coupling constant is not renormalized, $`\lambda _\text{b}=\lambda `$.
The Landau pole in $`\lambda _\text{b}`$ prevents the actual UV limit $`\mathrm{\Lambda }\mathrm{}`$. Nonetheless, neglecting all inverse powers of the UV cutoff when $`D=3`$, it is possible to rewrite the gap equation (18) as
$$\frac{M^2}{\widehat{\lambda }(M)}=\frac{m^2}{\widehat{\lambda }(m)}+3\overline{\varphi }^2$$
(20)
in terms of the oneโloop running coupling constant
$$\widehat{\lambda }(\mu )=\lambda \left[1\frac{3\lambda }{8\pi ^2}\mathrm{log}\frac{\mu }{m}\right]^1$$
It is quite clear that the HF states for which the renormalization just defined is sufficient are all those that are gaussianโdominated in the ultraviolet, so that we have \[compare to eq. (14)\]
$$\phi _{\pm k}^2\frac{1}{2\omega _k},k^2\mathrm{\Lambda }^2,\mathrm{\Lambda }\mathrm{}$$
(21)
If this property holds at a certain time, then it should hold at all times, since the Schroedinger equations (5) are indeed dominated by the quadratic term for large $`\omega _k`$ and $`\omega _k^2k^2+\text{const}+O(k^1)`$ as evident from eq. (9). Thus this class of states is indeed closed under time evolution and the parameterizations (17) and (19) make our tdHF approximation ultraviolet finite. Notice that the requirement (21) effectively always imposes a gap equation similar to eq. (15) in the deep ultraviolet.
Another simple check of the selfโconsistency of our approach, including the change in selected places from $`\lambda _\text{b}`$ to $`\lambda `$, as discussed above, follows from the energy calculation for the gaussian states with $`\varphi (x)=\overline{\varphi }`$ introduced above. Using eq. (4) and the standard replacement of sums by integrals in the infinite volume limit, we find
$$(\overline{\varphi })=\underset{L\mathrm{}}{lim}\frac{E}{L^D}=\frac{1}{2}\overline{\varphi }^2(M^2\lambda \overline{\varphi }^2)+\frac{1}{2}_{k^2\mathrm{\Lambda }^2}\frac{d^Dk}{(2\pi )^D}\sqrt{k^2+M^2}\frac{3}{4}\lambda _\text{b}\left[\overline{\varphi }^2+I_D(M^2,\mathrm{\Lambda })\right]^2$$
where $`M=M(\overline{\varphi })`$ depends on $`\overline{\varphi }`$ through the gap equation (18). The explicit calculation of the integrals involved shows that the energy density difference $`(\overline{\varphi })(0)`$ \[which for unbroken symmetry is nothing but the effective potential $`V_{\text{eff}}(\overline{\varphi })`$\], is indeed finite in the limit $`\mathrm{\Lambda }\mathrm{}`$, as required by a correct renormalization scheme. Notice that the finiteness of the energy density difference can be shown also by a simpler and more elegant argument, as presented below in section III B. This check would fail instead when $`D=3`$ if only the bare coupling constant $`\lambda _\text{b}`$ would appear in the last formula.
The tdHF approximation derived above represents a huge simplification with respect to the original problem, but its exact solution still poses itself as a considerable challenge. As a matter of fact, a numerical approach is perfectly possible within the capabilities of modern computers, provided the number of equations (5) is kept in the range of few thousands. As will become clear later on, even this numerical workout will turn out not to be really necessary in the form just alluded to, at least for the purposes of this paper.
### A On symmetry breaking
Quite obviously, in a finite volume and with a UV cutoff there cannot be any symmetry breaking, since the ground state is necessarily unique and symmetric when the number of degrees of freedom is finite . However, we may handily envisage the situation which would imply symmetry breaking when the volume diverges.
Let us first consider the case that we would call of unbroken symmetry. In this case the HF ground state is very close to the member with $`\overline{\varphi }=0`$ of the family of gaussian states introduced before. The difference is entirely due to the quartic terms in $`H_k`$. This correction vanish when $`L\mathrm{}`$, since all wavefunctions $`\psi _k`$ have $`L`$independent widths, so that one directly obtains the symmetric vacuum state with all the right properties of the vacuum (translation invariance, uniqueness, etc.) upon which a standard scalar massive particle Fock space can be based. The HF approximation then turns out to be equivalent to the resummation of all โcactus diagramsโ for the particle selfโenergy . In a finite volume, the crucial property of this symmetric vacuum is that all frequencies $`\omega _k^2`$ are strictly positive. The generalization to nonโequilibrium initial states with $`\overline{\varphi }0`$ is rather trivial: it amounts to a shift by $`L^{D/2}\overline{\varphi }`$ on $`\psi _0(\phi _0)`$. In the limit $`L\mathrm{}`$ we should express $`\psi _0`$ as a function of $`\xi =L^{D/2}\phi _0`$ so that, $`|\psi _0(\xi )|^2\delta (\xi \overline{\varphi })`$, while all other wavefunctions $`\psi _k`$ will reconstruct the gaussian wavefunctional corresponding to the vacuum $`|0,M`$ of a free massive scalar theory whose mass $`M=M(\varphi )`$ solves the gap equation (18). The absence of $`\psi _0`$ in $`|0,M`$ is irrelevant in the infinite volume limit, since $`\phi _0^2=L^D\overline{\varphi }^2+`$ terms of order $`L^0`$. The effective potential $`V_{\text{eff}}(\overline{\varphi })=(\overline{\varphi })(0)`$, where $`(\overline{\varphi })`$ is the lowest energy density at fixed $`\overline{\varphi }`$ and infinite volume, is manifestly a convex function with a unique minimum in $`\overline{\varphi }=0`$.
Now let us consider a different situation in which one or more of the $`\omega _k^2`$ are negative. Quite evidently, this might happen only for $`k`$ small enough, due to the $`k^2`$ in the gap equation \[thus eq. (21) remains valid and the ultraviolet renormalization is the same as for unbroken symmetry\]. Actually we assume here that only $`\omega _0^2<0`$, postponing the general analysis. Now the quartic term in $`H_0`$ cannot be neglected as $`L\mathrm{}`$, since in the ground state $`\psi _0`$ is symmetrically concentrated around the two minima of the potential $`\frac{1}{2}\omega _0^2\phi _0^2+\frac{\lambda }{4L^D}\phi _0^4`$, that is $`\phi _0=\pm (\omega _0^2L^D/\lambda )^{1/2}`$. If we scale $`\phi _0`$ as $`\phi _0=L^{D/2}\xi `$ then $`H_0`$ becomes
$$H_0=\frac{1}{2L^D}\frac{^2}{\xi ^2}+\frac{L^D}{2}\left(\omega _0^2\xi ^2+\frac{\lambda }{2}\xi ^4\right)$$
(22)
so that the larger $`L`$ grows the narrower $`\psi _0(\xi )`$ becomes around the two minima $`\xi =\pm (\omega _0^2/\lambda )^{1/2}`$. In particular $`\xi ^2\omega _0^2/\lambda `$ when $`L\mathrm{}`$ and $`\phi _0^2L^D\xi ^2`$. Moreover, the energy gap between the ground state of $`H_0`$ and its first, odd excited state as well as difference between the relative probability distributions for $`\xi `$ vanish exponentially fast in the volume $`L^D`$.
Since by hypothesis all $`\omega _k^2`$ with $`k0`$ are strictly positive, the ground state $`\psi _k`$ with $`k0`$ are asymptotically gaussian when $`L\mathrm{}`$ and the relations (12) tend to the form
$$\begin{array}{cc}\hfill \omega _k^2& =k^2+M^2k^2+m^2\hfill \\ \hfill M^2& =2\omega _0^2=m_\text{b}^2+3\lambda _\text{b}(L^D\phi _0^2+\mathrm{\Sigma }_0)=m_\text{b}^2+3\lambda _\text{b}\omega _0^2+3\lambda _\text{b}I_D(m^2,\mathrm{\Lambda })]\hfill \end{array}$$
This implies the identification $`\omega _0^2=m^2/2`$ and the bare mass parameterization
$$m_\text{b}^2=\left(1\frac{3}{2}\lambda _\text{b}/\lambda \right)m^23\lambda _\text{b}I_D(m^2,\mathrm{\Lambda })$$
(23)
characteristic of a negative $`\omega _0^2`$ \[compare to eq. (17)\], with $`m`$ the physical equilibrium mass of the scalar particle, as in the unbroken symmetry case. The coupling constant renormalization is the same as in eq. (19) as may be verified by generalizing to the minimum energy states with given field expectation value $`\overline{\varphi }`$; this minimum energy is nothing but the HF effective potential $`V_{\text{eff}}^{\text{HF}}(\overline{\varphi })`$, that the effective potential in this nonโgaussian HF approximation; of course, since $`\psi _0`$ is no longer asymptotically gaussian, we cannot simply shift it by $`L^{D/2}\overline{\varphi }`$ but, due to the concentration of $`\psi _0`$ on classical minima as $`L\mathrm{}`$, one readily finds that $`V_{\text{eff}}(\overline{\varphi })`$ is the convex envelope of the classical potential, that is its Maxwell construction. Hence we find
$$\phi _0^2\underset{L\mathrm{}}{}\{\begin{array}{cc}L^D\omega _0^2/\lambda ,\hfill & \lambda \overline{\varphi }^2\omega _0^2\hfill \\ L^D\overline{\varphi }^2,\hfill & \lambda \overline{\varphi }^2>\omega _0^2\hfill \end{array}$$
and the gap equation for the $`\overline{\varphi }`$dependent mass $`M`$ can be written, in terms of the step function $`\mathrm{\Theta }`$ and the extremal ground state field expectation value $`v=m/\sqrt{2\lambda }`$,
$$M^2=m^2+3\lambda _\text{b}(\overline{\varphi }^2v^2)\mathrm{\Theta }(\overline{\varphi }^2v^2)+3\lambda _\text{b}\left[I_D(M^2,\mathrm{\Lambda })I_D(m^2,\mathrm{\Lambda })\right]$$
(24)
We see that the specific bare mass parameterization (23) guarantees the nonโrenormalization of the treeโlevel relation $`v^2=m^2/2\lambda `$ ensuing from the typical symmetry breaking classical potential $`V(\varphi )=\frac{1}{4}\lambda (\varphi ^2v^2)^2`$. With the same finite part prescription as in eq. (18), the gap equation (24) leads to the standard coupling constant renormalization (19) when $`D=3`$.
In terms of the probability distributions $`|\psi _0(\xi )|^2`$ for the scaled amplitude $`\xi =L^{D/2}\phi _0`$, the Maxwell construction corresponds to the limiting form
$$|\psi _0(\xi )|^2\underset{L\mathrm{}}{}\{\begin{array}{cc}\frac{1}{2}(1+\overline{\varphi }/v)\delta (\xi v)+\frac{1}{2}(1\overline{\varphi }/v)\delta (\xi +v),\hfill & \overline{\varphi }^2v^2\hfill \\ \delta (\xi \overline{\varphi }),\hfill & \overline{\varphi }^2>v^2\hfill \end{array}$$
(25)
On the other hand, if $`\omega _0^2`$ is indeed the only negative squared frequency, the $`k0`$ part of this minimum energy state with arbitrary $`\overline{\varphi }=\varphi (x)`$ is better and better approximated as $`\mathrm{\Lambda }\mathrm{}`$ by the same gaussian state $`|0,M`$ of the unbroken symmetry state. Only the effective mass $`M`$ has a different dependence $`M(\overline{\varphi })`$, as given by the gap equation (24) proper of broken symmetry.
At infinite volume we may write
$$\phi _k^2=C(\overline{\varphi })\delta ^{(D)}(k)+\frac{1}{2\sqrt{k^2+M^2}}$$
where $`C(\overline{\varphi })=\overline{\varphi }^2`$ in case of unbroken symmetry (that is $`\omega _0^2>0`$), while $`C(\overline{\varphi })=\text{max}(v^2,\overline{\varphi }^2)`$ when $`\omega _0^2<0`$. This corresponds to the field correlation in space
$$\varphi (x)\varphi (y)=\frac{d^Dk}{(2\pi )^D}\phi _k^2e^{ik(xy)}=C(\overline{\varphi })+\mathrm{\Delta }_D(xy,M)$$
where $`\mathrm{\Delta }_D(xy,M)`$ is the massive free field equalโtime two points function in $`D`$ space dimensions, with selfโconsistent mass $`M`$. The requirement of clustering
$$\varphi (x)\varphi (y)\varphi (x)^2=v^2$$
contradicts the infinite volume limit of
$$\varphi (x)=L^{D/2}\underset{k}{}\varphi _ke^{ikx}=\phi _0=\overline{\varphi }$$
except at the two extremal points $`\overline{\varphi }=\pm v`$. In fact we know that the $`L\mathrm{}`$ limit of the finite volume states with $`\overline{\varphi }^2<v^2`$ violate clustering, because the two peaks of $`\psi _0(\xi )`$ have vanishing overlap in the limit and the first excited state becomes degenerate with the vacuum: this implies that the relative Hilbert space splits into two orthogonal Fock sectors each exhibiting symmetry breaking, $`\varphi (x)=\pm v`$, and corresponding to the two independent equal weight linear combinations of the two degenerate vacuum states. The true vacuum is either one of these symmetry broken states. Since the two Fock sectors are not only orthogonal, but also superselected (no local observable interpolates between them), linear combinations of any pair of vectors from the two sectors are not distinguishable from mixtures of states and clustering cannot hold in nonโpure phases. It is perhaps worth noticing also that the Maxwell construction for the effective potential, in the infinite volume limit, is just a straightforward manifestation of this fact and holds true, as such, beyond the HF approximation.
To further clarify this point and in view of subsequent applications, let us consider the probability distribution for the smeared field $`\varphi _f=d^Dx\varphi (x)f(x)`$, where
$$f(x)=f(x)=\frac{1}{L^D}\underset{k}{}f_ke^{ikx}\underset{L\mathrm{}}{}\frac{d^Dk}{(2\pi )^D}\stackrel{~}{f}(k)e^{ikx}$$
is a smooth real function with $`d^Dxf(x)=1`$ (i.e. $`f_0=1`$) localized around the origin (which is good as any other point owing to translation invariance). Neglecting in the infinite volume limit the quartic corrections for all modes with $`k0`$, so that the corresponding ground state wavefunctions are asymptotically gaussian, this probability distribution evaluates to
$$\text{Pr}(u<\varphi _f<u+du)=\frac{du}{(2\pi \mathrm{\Sigma }_f)^{1/2}}_{\mathrm{}}^+\mathrm{}๐\xi |\psi _0(\xi )|^2\mathrm{exp}\left\{\frac{(u\xi )^2}{2\mathrm{\Sigma }_f}\right\}$$
where
$$\mathrm{\Sigma }_f=\underset{k0}{}\phi _k^2f_k^2\underset{L\mathrm{}}{}\frac{d^Dk}{(2\pi )^D}\frac{\stackrel{~}{f}(k)^2}{2\sqrt{k^2+m^2}}$$
In the unbroken symmetry case we have $`|\psi _0(\xi )|^2\delta (\xi \overline{\varphi })`$ as $`L\mathrm{}`$, while the limiting form (25) holds for broken symmetry. Thus we obtain
$$\text{Pr}(u<\varphi _f<u+du)=p_f(u\overline{\varphi })du,p_f(u)\left(2\pi \mathrm{\Sigma }_f\right)^{1/2}\mathrm{exp}\left(\frac{u^2}{2\mathrm{\Sigma }_f}\right)$$
for unbroken symmetry and
$$\text{Pr}(u<\varphi _f<u+du)=\{\begin{array}{cc}\frac{1}{2}(1+\overline{\varphi }/v)p_f(uv)du+\frac{1}{2}(1\overline{\varphi }/v)p_f(u+v)du,\hfill & \overline{\varphi }^2v^2\hfill \\ p_f(u\overline{\varphi })du,\hfill & \overline{\varphi }^2>v^2\hfill \end{array}$$
for broken symmetry. Notice that the momentum integration in the expression for $`\mathrm{\Sigma }_f`$ needs no longer an ultraviolet cutoff; of course in the limit of deltaโlike test function $`f(x)`$, $`\mathrm{\Sigma }_f`$ diverges and $`p_f(u)`$ flattens down to zero. The important observation is that $`\text{Pr}(u<\varphi _f<u+du)`$ has always a single peak centered in $`u=\overline{\varphi }`$ for unbroken symmetry, while for broken symmetry it shows two peaks for $`\overline{\varphi }^2v^2`$ and $`\mathrm{\Sigma }_f`$ small enough. For instance, if $`\overline{\varphi }=0`$, then there are two peaks for $`\mathrm{\Sigma }_f<v^2`$ \[implying that $`\stackrel{~}{f}(k)`$ has a significant support only up to wavevector $`k`$ of order $`v`$, when $`D=3`$, or $`m\mathrm{exp}(\text{const }v^2)`$ when $`D=1`$\].
To end the discussion on symmetry breaking, we may now verify the validity of the assumption that only $`\omega _0^2`$ is negative. In fact, to any squared frequency $`\omega _k^2`$ (with $`k0`$) that stays strictly negative as $`L\mathrm{}`$ there corresponds a wavefunction $`\psi _k`$ that concentrates on $`\phi _k^2+\phi _k^2=\omega _k^2L^D/\lambda `$ ; then eqs. (12) implies $`2\omega _k^2=k^2+m^2`$ for such frequencies, while $`\omega _k^2=k^2+m^2`$ for all frequencies with positive squares; if there is a macroscopic number of negative $`\omega _k^2`$ (that is a number of order $`L^D`$), then the expression for $`\omega _0^2`$ in eq. (12) will contain a positive term of order $`L^D`$ in the r.h.s., clearly incompatible with the requirements that $`\omega _0^2<0`$ and $`m_\text{b}^2`$ be independent of $`L`$; if the number of negative $`\omega _k^2`$ is not macroscopic, then the largest wavevector with a negative squared frequency tends to zero as $`L\mathrm{}`$ (the negative $`\omega _k^2`$ clearly pile in the infrared) and the situation is equivalent, if not identical, to that discussed above with only $`\omega _0^2<0`$.
### B Outโofโequilibrium dynamics
We considered above the lowest energy states with a predefinite uniform field expectation value, $`\varphi (x)=\overline{\varphi }`$, and established how they drastically simplify in the infinite volume limit. For generic $`\overline{\varphi }`$ these states are not stationary and will evolve in time. By hypothesis $`\psi _k`$ is the ground state eigenfunction of $`H_k`$ when $`k>0`$, and therefore $`|\psi _k|^2`$ would be stationary for constant $`\omega _k`$, but $`\psi _0`$ is not an eigenfunction of $`H_0`$ unless $`\overline{\varphi }=0`$. As soon as $`|\psi _k|^2`$ starts changing, $`\phi _0^2`$ changes and so do all frequencies $`\omega _k`$ which are coupled to it by the eqs. (12). Thus the change propagates to all wavefunctions. The difficult task of studying this dynamics can be simplified with the following scheme, that we might call gaussian approximation. We first describe it and discuss its validity later on.
Let us assume the usual gaussian form for the initial state \[see eq. (14) and the discussion following it\]. We know that it is a good approximation to the lowest energy state with given $`\phi _0`$ for unbroken symmetry, while it fails to be so for broken symmetry, only as far as $`\psi _0`$ is concerned, unless $`\overline{\varphi }^2v^2`$. At any rate this is an acceptable initial state: the question is about its time evolution. Suppose we adopt the harmonic approximation for all $`H_k`$ with $`k>0`$ by dropping the quartic term. This approximation will turn out to be valid only if the width of $`\psi _k`$ do not grow up to the order $`L^D`$ (by symmetry the center will stay in the origin). In practice we are now dealing with a collection of harmonic oscillators with timeโdependent frequencies and the treatment is quite elementary: consider the simplest example of one quantum degree of freedom described by the gaussian wavefunction
$$\psi (q,t)=\frac{1}{(2\pi \sigma ^2)^{1/4}}\mathrm{exp}\left[\frac{1}{2}\left(\frac{1}{2\sigma ^2}i\frac{s}{\sigma }\right)q^2\right]$$
where $`s`$ and $`\sigma `$ are timeโdependent. If the dynamics is determined by the timeโdependent harmonic hamiltonian $`\frac{1}{2}[_q^2+\omega (t)^2q^2]`$, then the Schroedinger equation is solved exactly provided that $`s`$ and $`\sigma `$ satisfy the classical Hamilton equations
$$\dot{\sigma }=s,\dot{s}=\omega ^2\sigma +\frac{1}{4\sigma ^3}$$
It is not difficult to trace the โcentrifugalโ force $`(4\sigma )^3`$ which prevents the vanishing of $`\sigma `$ to Heisenberg uncertainty principle .
The extension to our case with many degrees of freedom is straightforward and we find the following system of equations
$$i\frac{}{t}\psi _0=H_0\psi _0,\frac{d^2\sigma _k}{dt^2}=\omega _k^2\sigma _k+\frac{1}{4\sigma _k^3},k>0$$
(26)
coupled in a meanโfield way by the relations (12), which now read
$$\begin{array}{cc}\hfill \omega _k^2& =k^2+M^26\lambda L^D\sigma _k^2,k>0\hfill \\ \hfill M^2& =m_\text{b}^2+3\lambda _\text{b}\left(L^D\phi _0^2+\mathrm{\Sigma }_0\right),\mathrm{\Sigma }_0=\frac{1}{L^D}\underset{k0}{}\sigma _k^2\hfill \end{array}$$
(27)
This stage of a truly quantum zeroโmode and classical modes with $`k>0`$ does not appear fully consistent, since for large volumes some type of classical or gaussian approximation should be considered for $`\phi _0`$ too. We may proceed in two (soon to be proven equivalent) ways:
1. We shift $`\phi _0=L^{D/2}\overline{\varphi }+\eta _0`$ and then deal with the quantum mode $`\eta _0`$ in the gaussian approximation, taking into account that we must have $`\eta _0=0`$ at all times. This is most easily accomplished in the Heisenberg picture rather than in the Schroedinger one adopted above. In any case we find that the quantum dynamics of $`\phi _0`$ is equivalent to the classical dynamics of $`\overline{\varphi }`$ and $`\sigma _0\eta _0^2^{1/2}`$ described by the ordinary differential equations
$$\frac{d^2\overline{\varphi }}{dt^2}=\omega _0^2\overline{\varphi }\lambda \overline{\varphi }^3,\frac{d^2\sigma _0}{dt^2}=\omega _0^2\sigma _0+\frac{1}{4\sigma _0^3}$$
(28)
where $`\omega _0^2=M^23\lambda L^D\phi _0^2`$ and $`\phi _0^2=L^D\overline{\varphi }^2+\sigma _0^2`$.
2. We rescale $`\phi _0=L^{D/2}\xi `$ right away, so that $`H_0`$ takes the form of eq. (22). Then $`L\mathrm{}`$ is the classical limit such that $`\psi _0(\xi )`$ concentrates on $`\xi =\overline{\varphi }`$ which evolves according to the first of the classical equations in (28). Since now there is no width associated to the zeroโmode, $`\overline{\varphi }`$ is coupled only to the widths $`\sigma _k`$ with $`k0`$ by $`\omega _0^2=M^23\lambda \overline{\varphi }^2`$, while $`M^2=m_\text{b}^2+3\lambda _\text{b}(\overline{\varphi }^2+\mathrm{\Sigma }_0)`$.
It is quite evident that these two approaches are completely equivalent in the infinite volume limit, and both are good approximation to the original tdHF Schroedinger equations, at least provided that $`\sigma _0^2`$ stays such that $`L^D\sigma _0^2`$ vanishes in the limit for any time. In this case we have the evolution equations
$$\frac{d^2\overline{\varphi }}{dt^2}=(2\lambda \overline{\varphi }^2M^2)\overline{\varphi },\frac{d^2\sigma _k}{dt^2}=(k^2+M^2)\sigma _k+\frac{1}{4\sigma _k^3}$$
(29)
meanโfield coupled by the $`L\mathrm{}`$ limit of eqs. (27), namely
$$M^2=m^2+3\lambda _\text{b}\left[\overline{\varphi }^2+\mathrm{\Sigma }I_D(m^2,\mathrm{\Lambda })\right]$$
(30)
for unbroken symmetry \[that is $`m_\text{b}^2`$ as in eq. (17)\] or
$$M^2=m^2+3\lambda _\text{b}\left[\overline{\varphi }^2v^2+\mathrm{\Sigma }I_D(m^2,\mathrm{\Lambda })\right],m^2=2\lambda v^2$$
(31)
for broken symmetry \[that is $`m_\text{b}^2`$ as in eq. (23)\]. In any case we define
$$\mathrm{\Sigma }=\frac{1}{L^D}\underset{k}{}\sigma _k^2\underset{L\mathrm{}}{}_{k^2\mathrm{\Lambda }^2}\frac{d^Dk}{(2\pi )^D}\sigma _k^2$$
as the sum, or integral, over all microscopic gaussian widths \[N.B.:this definition differs from that given before in eq. (8) by the classical term $`\overline{\varphi }^2`$\]. Remarkably, the equations of motion (29) are completely independent of the ultraviolet cutโoff and this is a direct consequence of the substitution (11). Had we kept the bare coupling constant everywhere in the expression (10), we would now have $`\lambda _\text{b}`$ also in front of the $`\overline{\varphi }^3`$ in the r.h.s. of the first of the two equations (29) \[cfr., for instance, ref. \].
The conserved HF energy (density) corresponding to these equations of motion reads
$$\begin{array}{cc}\hfill & =๐ฏ+๐ฑ,๐ฏ=\frac{1}{2}(\dot{\overline{\varphi }})^2+\frac{1}{2L^D}\underset{k}{}\dot{\sigma }_k^2\hfill \\ \hfill ๐ฑ& =\frac{1}{2L^D}\underset{k}{}\left(k^2\sigma _k^2+\frac{1}{4\sigma _k^2}\right)+\frac{1}{2}m_\text{b}^2(\overline{\varphi }^2+\mathrm{\Sigma })+\frac{3}{4}\lambda _\text{b}(\overline{\varphi }^2+\mathrm{\Sigma })^2\frac{1}{2}\lambda \overline{\varphi }^4\hfill \end{array}$$
(32)
Up to additive constants and terms vanishing in the infinite volume limit, this expression agrees with the general HF energy of eq. (13) for gaussian wavefunctions. It holds both for unbroken and broken symmetry, the only difference being in the parameterization of the bare mass in terms of UV cutoff and physical mass, eqs. (17) and (23). The similarity to the energy functional of the large $`N`$ approach is evident; the only difference, apart from the obvious fact that $`\overline{\varphi }`$ is a single scalar rather than a $`O(n)`$ vector, is in the meanโfield coupling $`\sigma _k`$$`\overline{\varphi }`$ and $`\sigma _k`$$`\mathrm{\Sigma }`$, due to different coupling strength of transverse and longitudinal modes (cfr. ref. ).
This difference between the HF approach for discrete symmetry (i.e $`N=1`$) and the large $`N`$ method for the continuous $`O(N)`$-symmetry is not very relevant if the symmetry is unbroken \[it does imply however a significantly slower dissipation to the modes of the background energy density\]. On the other hand it has a drastic consequence on the equilibrium properties and on the outโofโequilibrium dynamics in case of broken symmetry (see below), since massless Goldstone bosons appear in the large $`N`$ approach, while the HF treatment of the discrete symmetry case must exhibits a mass also in the broken symmetry phase.
The analysis of physically viable initial conditions proceeds exactly as in the large $`N`$ approach and will not be repeated here, except for an important observation in case of broken symmetry. The formal energy minimization w.r.t. $`\sigma _k`$ at fixed $`\overline{\varphi }`$ leads again to eqs.
$$\dot{\sigma }_k=0,\sigma _k^2=\frac{1}{2\sqrt{k^2+M^2}}$$
(33)
and again these are acceptable initial conditions only if the gap equation that follows from eq. (31) in the $`L\mathrm{}`$ limit, namely
$$M^2=m^2+3\lambda _\text{b}\left[\overline{\varphi }^2v^2+I_D(M^2,\mathrm{\Lambda })I_D(m^2,\mathrm{\Lambda })\right]$$
(34)
admits a nonnegative, physical solution for $`M^2`$. Notice that there is no step function in eq. (34), unlike the static case of eq. (24), because $`\sigma _0^2`$ was assumed to be microscopic, so that the infinite volume $`\sigma _k^2`$ has no deltaโlike singularity in $`k=0`$. Hence $`M=m`$ solves eq. (34) only at the extremal points $`\overline{\varphi }=\pm v`$, while it was the solution of the static gap equation (24) throughout the Maxwell region $`v\overline{\varphi }v`$. The important observation is that eq. (34) admits a positive solution for $`M^2`$ also within the Maxwell region. In fact it can be written, neglecting as usual the inverseโpower corrections in the UV cutoff
$$\frac{M^2}{\widehat{\lambda }(M)}=\frac{m^2}{\lambda }+3(\overline{\varphi }^2v^2)=3\overline{\varphi }^2v^2$$
(35)
and there exists indeed a positive solution $`M^2`$ smoothly connected to the ground state, $`\overline{\varphi }^2=v^2`$ and $`M^2=m^2`$, whenever $`\overline{\varphi }^2v^2/3`$. The two intervals $`v^2\overline{\varphi }^2v^2/3`$ correspond indeed to the metastability regions, while $`\overline{\varphi }^2<v^2/3`$ is the spinodal region, associated to a classical potential proportional to $`(\overline{\varphi }^2v^2)^2`$. This is another effect of the different coupling of transverse and longitudinal modes: in the large $`N`$ approach there are no metastability regions and the spinodal region coincides with the Maxwell one. As in the large $`N`$ approach in the spinodal interval there is no energy minimization possible, at fixed background and for microscopic widths, so that a modified form of the gap equation
$$M^2=m^2+3\lambda _\text{b}\left[\overline{\varphi }^2v^2+\frac{1}{L^D}\underset{k^2<|M^2|}{}\sigma _k^2+\frac{1}{L^D}\underset{k^2>|M^2|}{}\frac{1}{2\sqrt{k^2|M^2|}}I_D(0,\mathrm{\Lambda })\right]$$
(36)
should be applied to determine ultravioletโfinite initial conditions.
The main question now is: how will the gaussian widths $`\sigma _k`$ grow with time, and in particular how will $`\sigma _0`$ grow in case of method 1 above, when we start from initial conditions where all widths are microscopic? For the gaussian approximation to remain valid through time, all $`\sigma _k`$, and in particular $`\sigma _0`$, must at least not become macroscopic. In fact we have already positively answered this question in the large $`N`$ approach and the HF equations (29) do not differ so much to expect the contrary now. In particular, if we consider the special initial condition $`\overline{\varphi }=\dot{\overline{\varphi }}=0`$, the dynamics of the widths is identical to that in the large $`N`$ approach, apart from the rescaling by a factor of three of the coupling constant.
In fact, if we look at the time evolution of the zeroโmode amplitude $`\sigma _0`$ \[see Fig. 1\], we can see the presence of the timeโscale $`\tau _L`$ at which finite volume effects start to manifest. The time scale $`\tau _L`$ turns out to be proportional to the linear size of the box $`L`$ and its presence prevents $`\sigma _0`$ from growing to macroscopic values. Thus our HF approximation confirms the large $`N`$ approach in the following sense: even if one considers in the variational ansatz the possibility of nonโgaussian wavefunctionals, the time evolution from gaussian and microscopic initial conditions is effectively restricted for large volumes to nonโmacroscopic gaussians.
Strictly speaking, however, this might well not be enough, since the infrared fluctuations do grow beyond the microscopic size to become of order $`L`$ \[see Fig. 2, where the evolution of the mode with momentum $`k=2\pi /L`$ is plotted\]. Then the quartic term in the low$`k`$ Hamiltonians $`H_k`$ is of order $`L`$ and therefore it is not negligible by itself in the $`L\mathrm{}`$ limit, but only when compared to the quadratic term, which for a fixed $`\omega _k^2`$ of order $`1`$ would be of order $`L^2`$. But we know that, when $`\overline{\varphi }=0`$, after the spinodal time and before the $`\tau _L`$, the effective squared mass $`M^2`$ oscillates around zero with amplitude decreasing as $`t^1`$ and a frequency fixed by the largest spinodal wavevector. In practice it is โzero on averageโ and this reflect itself in the average linear growth of the zeroโmode fluctuations and, more generally, in the average harmonic motion of the other widths with nonโzero wavevectors. In particular the modes with small wavevectors of order $`L^1`$ feel an average harmonic potential with $`\omega _k^2`$ of order $`L^2`$. This completely compensate the amplitude of the mode itself, so that the quadratic term in the low$`k`$ Hamiltonians $`H_k`$ is of order $`L^0`$, much smaller than the quartic term that was neglected beforehand in the gaussians approximation. Clearly the approximation itself no longer appears fully justified and a more delicate analysis is required. We intend to return on this issue in a future work, restricting ourselves in the next section to the gaussians approximation.
## IV Lateโtime evolution and dynamical Maxwell construction
By definition, the gaussian approximation of the effective potential $`V_{\text{eff}}(\overline{\varphi })`$ coincides with the infiniteโvolume limit of the potential energy $`๐ฑ(\overline{\varphi },\{\sigma _k\})`$ of eq. (32) when the widths are of the $`\overline{\varphi }`$dependent, energyโminimizing form (33) with the gap equation for $`M^2`$ admitting a nonnegative solution. As we have seen, this holds true in the unbroken symmetry case for any value of the background $`\overline{\varphi }`$, so that the gaussian $`V_{\text{eff}}`$ is identical to the HF one, since all wavefunctions $`\psi _k`$ are asymptotically gaussians as $`L\mathrm{}`$. In the presence of symmetry breaking instead, this agreement holds true only for $`\overline{\varphi }^2v^2`$; for $`v^2/3\overline{\varphi }^2<v^2`$ the gaussian $`V_{\text{eff}}`$ exists but is larger than the HF potential $`V_{\text{eff}}^{\text{HF}}`$, which is already flat. In fact, for any $`\overline{\varphi }^2v^2/3`$, we may write the gaussian $`V_{\text{eff}}`$ as
$$V_{\text{eff}}(\overline{\varphi })=V_{\text{eff}}(\overline{\varphi })=V_{\text{eff}}(v)+_v^{|\overline{\varphi }|}๐uu[M(u)^22\lambda u^2]$$
where $`M(u)^2`$ solves the gap equation (35), namely $`M(u)^2=\widehat{\lambda }(M(u))(3u^2v^2)`$. In each of the two disjoint regions of definition this potential is smooth and convex, with unique minima in $`+v`$ and $`v`$, respectively. These appear therefore as regions of metastability (states which are only locally stable in the presence of a suitable uniform external source). The HF effective potential is identical for $`\overline{\varphi }^2v^2`$, while it takes the constant value $`V_{\text{eff}}(v)`$ throughout the internal region $`\overline{\varphi }^2<v^2`$. It is based on truly stable (not only metastable) states. The gaussian $`V_{\text{eff}}`$ cannot be defined in the spinodal region $`\overline{\varphi }^2<v^2/3`$, where the gap equation does not admit a nonnegative solution in the physical region far away from the Landau pole.
Let us first compare this HF situation with that of large $`N`$ . There the different coupling of the transverse modes, three time smaller than the HF longitudinal coupling, has two main consequences at the static level: the gap equation similar to (35) does not admit nonnegative solutions for $`\overline{\mathit{\varphi }}^2<v^2`$, so that the spinodal region coincides with the region in which the effective potential is flat, and the physical mass vanishes. The outโofโequilibrium counterpart of this is the dynamical Maxwell construction: when the initial conditions are such that $`\overline{\mathit{\varphi }}^2`$ has a limit for $`t\mathrm{}`$, the set of all possible asymptotic values exactly covers the flatness region (and the effective mass vanishes in the limit). In practice this means that $`|\overline{\mathit{\varphi }}|`$ is not the true dynamical order parameter, whose large time limit coincides with $`v`$, the equilibrium field expectation value in a pure phase. Rather, one should consider as order parameter the renormalized local (squared) width
$$\underset{N\mathrm{}}{lim}\frac{\mathit{\varphi }(x)\mathit{\varphi }(x)_\text{R}}{N}=\overline{\mathit{\varphi }}^2+\mathrm{\Sigma }_\text{R}=v^2+\frac{M^2}{\lambda }$$
where the last equality follows from the definition itself of the effective mass $`M`$ (see ref. ). Since $`M`$ vanishes as $`t\mathrm{}`$ when $`\overline{\mathit{\varphi }}^2`$ tends to a limit within the flatness region, we find the renormalized local width tends to the correct value $`v`$ which characterizes the broken symmetry phase, that is the bottom of the classical potential. We may say that the spinodal region, perturbatively unstable, at the nonโperturbative level corresponds to metastable states, all reachable through the asymptotic time evolution with a vanishing effective mass.
In the HF approximation, where at the static level the spinodal region $`\overline{\varphi }^2<v^2/3`$ is smaller than the flatness region $`\overline{\varphi }^2<v^2`$, the situation is rather different. Our numerical solution shows that, $`\overline{\varphi }`$ oscillates around a certain value $`\overline{\varphi }_{\mathrm{}}`$ with an amplitude that decreases very slowly. As in large $`N`$, the asymptotic value $`\overline{\varphi }_{\mathrm{}}`$ depends on the initial value $`\overline{\varphi }(0)`$. But, if the background $`\overline{\varphi }`$ starts with zero velocity from a nonโzero value inside the spinodal interval, then it always leaves this region and eventually oscillates around a point between the spinodal point $`v/\sqrt{3}`$ and the minimum of the tree level potential $`v`$ (see Fig.s 3 and 4). In other words, if we start with a $`\overline{\varphi }`$ in the interval $`[v,v]`$, except the origin, we end up with a $`\overline{\varphi }_{\mathrm{}}`$ in the restricted interval $`[v,v/\sqrt{3}][v/\sqrt{3},v]`$. The spinodal region is completely forbidden for the late time evolution of the mean field, as is expected for an unstable region. We stress that we are dealing with true fixed points of the asymptotic evolution since the force term on the mean field \[cfr. eq. (29), $`f=(2\lambda \overline{\varphi }^2M^2)\overline{\varphi }`$\] does vanish in the limit. In facts its time average $`\overline{f}=^Tf(t)๐t/T`$ tends to zero as $`T`$ grows and its mean squared fluctuations around $`\overline{f}`$ decreases towards zero, although very slowly (see Fig.s 5 and 6). Moreover, for $`N=1`$ the order parameter reads as $`t\mathrm{}`$
$$\varphi (x)^2_\text{R}=\overline{\varphi }^2+\mathrm{\Sigma }_\text{R}=\frac{v^2}{3}+\frac{M^2}{3\lambda },\mathrm{\Sigma }_\text{R}=\frac{v^2\overline{\varphi }^2}{3}$$
(37)
where the last equality is valid for the asymptotic values and follows from the vanishing of the force term $`f`$. From the last formula we see that when $`\overline{\varphi }=0`$ at the beginning, and then at all times, the renormalized backโreaction tends to $`v^2/3`$, not $`v^2`$. It โstops at the spinodal lineโ. The same picture applies for a long time, all during the โslow rolling downโ (see section V), to evolutions that start close enough to $`\overline{\varphi }=0`$. This fact is at the basis of the soโcalled spinodal inflation.
In any case, the dynamical Maxwell construction, either complete or partial, poses an interesting question by itself. In fact it is not at all trivial that the effective potential, in any of the approximation previously discussed, does bear relevance on the asymptotic behavior of the infiniteโvolume system whenever a fixed point is approached. Strictly speaking in fact, even in such a special case it is not directly related to the dynamics, since it is obtained from a static minimization of the total energy at fixed mean field, while the energy is not at its minimum at the initial time and is exactly conserved in the evolution. On the other hand, if a solution of the equations of motion (29) exists in which the background $`\overline{\varphi }`$ tends to a constant $`\overline{\varphi }_{\mathrm{}}`$ as $`t\mathrm{}`$, one might expect that the effective action (which however is nonlocal in time) somehow reduces to a (infinite) multiple of the effective potential, so that $`\overline{\varphi }_{\mathrm{}}`$ should be an extremal of the effective potential. This is still an open question that deserves further analytic studies and numerical confirmation.
It is worth noticing also that when the field starts very close to the top of the potential hill, it remains there for a very long time and evolves through a very slow rolling down, before beginning a damped oscillatory motion around a point in the metastability region. During the slow roll period, $`M^2`$ oscillates around zero with decreasing amplitude and the โphenomenologyโ is very similar to the evolution from symmetric initial conditions, as can be seen comparing Fig.s 7 and 8. Fig. 9 shows the evolution of the zero mode amplitude in case of a very slow rolling down. In such a case, after a very short (compared to the time scale of the figure) period of exponential growth (the spinodal time), the quantum fluctuations start an almost linear growth, very similar to the evolution starting from a completely symmetric initial state. This, obviously, corresponds to the vanishing of the effective mass. In the meanwhile, $`\overline{\varphi }`$ keeps growing and rolling down the potential hill with increasing speed towards the minimum of the classical potential, eventually entering the metastable region. At that time, the effective mass starts to increase again and the zero mode stops its linear growth, turns down and enters a phase of โwildโ evolution. This time scale, let us call it $`\tau _{\text{s}rd}`$, depends on the initial value of the condensate: the smaller $`\overline{\varphi }(t=0)`$ is, the longer $`\tau _{\text{s}rd}`$ will be. We find numerically that $`\tau _{\text{s}rd}\left(\overline{\varphi }(t=0)\right)^{1/2}`$.
If we now study the dynamics in finite volume, starting from condensates different from zero, we will find a competition between $`\tau _{\text{s}rd}`$ and $`\tau _L`$, the time scale characteristic of the finite volume effects, that is proportional to the linear size of the box we put the system in. Fig. 10 shows clearly that when $`L/2\pi =100`$ and $`\overline{\varphi }=10^5`$, we have $`\tau _{\text{s}rd}\tau _L`$. In any case, either one or the other effect will prevent the zero mode amplitude from growing to macroscopic values for any initial condition we may start with.
It should be noted, also, that the presence of the time scale $`\tau _{\text{s}rd}`$ does not solve the internal inconsistency of the gaussian approximation described above in section III B. In fact, for any fixed value $`L`$ for the linear size of the system, we can find a whole interval of initial conditions for the mean field, which leave enough time to the fluctuations for growing to order $`L`$, much before the field itself had rolled down towards one of the minima of the classical potential. For those particular evolutions, we would need to consider the quartic terms in the hamiltonians that the gaussian approximation neglects, as already explained.
In addition, there will be also initial conditions for which $`\tau _L>\tau _{\text{s}rd}`$. In that case, the effective mass soon starts oscillating around positive values and it is reasonable to think that it will take a much longer time than $`\tau _L`$ for the finite volume effects to manifest. In we have interpreted the proportionality between $`\tau _L`$ and $`L`$ as an auto interference effect (due to periodic boundary conditions) suffered by a Goldstone boson wave, traveling at speed of light, at the moment it reaches the borders of the cubic box. Here, the massless wave we have in the early phase of the evolution, rapidly acquires a positive mass, as soon as the condensate rolls down; this decelerates the waveโs propagation and delays the onset of finite volume effects. The gaussian approximation appears to be fully consistent when we limit ourselves to the evolution of these particular configurations.
## V Numerical analysis
We discuss in this section the asymptotic behavior of the dynamical evolution as it turns out from our numerical results in the gaussian approximation.
Let us begin with the precise form of the evolution equations for the field background and the quantum mode widths, as described in sections III B (cfr eq. (29).
$$\left[\frac{d^2}{dt^2}+\left(M^22\lambda \varphi ^2\right)\right]\varphi =0,\left[\frac{d^2}{dt^2}+k_n^2+M^2\right]\sigma _n\frac{1}{4\sigma _n^3}=0$$
(38)
where the index $`n`$ labels the discrete set of values used to perform the sum (finite volume) or the integral (infinite volume) over momenta in the quantum backโreaction $`\mathrm{\Sigma }`$, while $`M^2(t)`$ is defined by the eq. (30) in case of unbroken symmetry and by eq. (31) in case of broken symmetry. The backโreaction $`\mathrm{\Sigma }`$ reads, in the notations of this appendix
$$\mathrm{\Sigma }=\underset{n=0}{\overset{๐ฉ}{}}g_n\sigma _n^2$$
where $`g_n`$ is the appropriate โdegeneracyโ factor and $`๐ฉ`$ is the number of modes with distinct dynamics. Technically it is simpler to treat an equivalent set of equations, which are formally linear and do not contain the singular Heisenberg term $`\sigma _n^3`$. This is done by introducing the complex mode amplitudes $`z_n=\sigma _n\mathrm{exp}(i\theta _n)`$, where the phases $`\theta _n`$ satisfy $`\sigma _n^2\dot{\theta }_n=1`$. Then we find a discrete version of the equations studied for instance in ref , namely
$$\left[\frac{d^2}{dt^2}+k_n^2+M^2\right]z_n=0,\mathrm{\Sigma }=\frac{1}{L^D}\underset{n=0}{\overset{๐ฉ}{}}g_n|z_n|^2$$
(39)
subject to the Wronskian condition
$$z_n\dot{\overline{z_n}}\overline{z_n}\dot{z}_n=i$$
One realizes that the Heisenberg term in $`\sigma _n`$ corresponds to the centrifugal potential for the motion in the complex plane of $`z_n`$. Looking at the figs. 2 or 9, we can see that the motions of the quantum modes correspond qualitatively to orbits with very large eccentricities. In fact, there are istants in which $`\sigma _n`$ is very little and the angular velocity $`\dot{\theta }_n`$ is very large. This is the technical reason for preferring the equations in the form (39).
To solve these evolution equations, we have to choose suitable initial conditions respecting the Wronskian condition. In case of unbroken symmetry, the requirement of minimum energy for the quantum fluctuations leads to the massive particle spectrum:
$$z_n(0)=\frac{1}{\sqrt{2\mathrm{\Omega }_n}}\frac{dz_n}{dt}(0)=ฤฑ\sqrt{\frac{\mathrm{\Omega }_n}{2}}$$
where $`\mathrm{\Omega }_n=\sqrt{k_n^2+M^2(0)}`$ and the initial squared effective mass $`M^2(t=0)`$, has to be determined self-consistently, by means of its definition (30).
In case of broken symmetry, the gap equation is a viable mean for fixing the initial conditions only when $`\varphi `$ lies outside the spinodal region \[cfr. eq (35)\]; otherwise, the gap equation does not admit a positive solution for the squared effective mass and we cannot minimize the energy of the fluctuations. Following the discussion presented in III B, one possible choice is to set $`\sigma _k^2=\frac{1}{2\sqrt{k^2+|M^2|}}`$ for $`k^2<|M^2|`$ and then solve the corresponding gap equation (36). We will call this choice the โflippedโ initial condition. An other acceptable choice would be to solve the gap equation, setting a massless spectrum for all the spinodal modes but the zero mode, which is started from an arbitrary, albeit microscopic, value. This choice will be called the โmasslessโ initial condition.
Before passing to discuss the influence of different initial conditions on the results, let us present the asymptotic behavior we find when we choose the flipped initial condition. In Fig. 13 we have plotted the asymptotic values of the mean field versus the initial values, for $`\lambda =0.1`$. All dimensionful quantities are expressed in terms of the suitable power of the equilibrium mass $`m`$. For example, the vev of the field is equal to $`\sqrt{5}`$ in these units. First of all, consider the initial values for the condensate far enough from the top of the potential hill, say between $`\overline{\varphi }(t=0)=0.88`$ and $`\overline{\varphi }(t=0)=2.64`$. In that region the crosses seem to follow a smooth curve, that has its maximum exactly at $`\overline{\varphi }_{\mathrm{}}=\sqrt{5}`$ (the point of stable equilibrium). When we start from an initial condition smaller than $`\overline{\varphi }(t=0)=0.88`$, the asymptotic value $`\overline{\varphi }_{\mathrm{}}`$ is not guaranteed to be positive anymore. On the contrary, it is possible to choose the initial condition in such a way that the condensate will oscillate between positive and negative values for a while, before settling around an asymptotic value near either one or the other minimum, as fig 11 clearly shows. Fig.s 14, 15 and 16 helps to understand this behavior by consideration on the energy balance. Both the evolutions are such that the classical energy, defined as $`\lambda (\overline{\varphi }^2v^2)/4`$, is not a monotonically decreasing function of time. Indeed, energy is continuously exchanged between the classical degree of freedom and the quantum fluctuations bath, in both directions. However, the two rates of energy exchange are not exactly the same and an effective dissipation of classical energy on average can be seen, at long time at least. Of course, this is not the case for the initial transient part of the evolution starting from the initial condition $`\overline{\varphi }(t=0)=0.08`$; there, the condensate absorbs energy (on average) from the quantum fluctuations, being able to go beyond the top of the potential hill, towards the negative minimum. This happens because in case of broken symmetry, the minimization of the fluctuation energy, within microscopic gaussian states, is not possible for initial conditions in the spinodal region \[cfr. the discussion about the gap equation (35) in section III B\]. After a number of oscillations, the energy starts to flow from the condensate to the quantum bath again (on the average), constraining the condensate to oscillate around a value close to one of the two minima. If we look at fig. 13 again, we can find positive asymptotic values as well as negative ones, without a definite pattern, in the whole interval $`[0.01,0.8]`$. If we start with $`0<\overline{\varphi }(t=0)<0.01`$ we have the slow rolling down, already described in section IV and the mean field oscillates around a positive value from the beginning, never reaching negative values. A further note is worth being added here. During the phase of slow rolling down, the evolution is very similar to a symmetric evolution starting from $`\overline{\varphi }(t=0)=0`$; in that case, the dissipation mechanism works through the emission of (quasi-)massless particles and it is very efficient because it has not any perturbative threshold. If the field stays in this slow rolling down phase for a time long enough, it will not be able to absorb the sufficient energy to pass to the other side ever again and it will be confined in the positive valley for ever. Evidently, when $`\overline{\varphi }(t=0)>0.01`$ this dissipative process might not be so efficient to prevent the mean field from sampling also the other valley.
Which one of the two valleys will be chosen by the condensate is a matter of initial conditions and it is very dependent from the energy stored in the initial state, as is shown in fig. 12, where two evolutions are compared, starting from the same value for the condensate, but with the two initial conditions, โflippedโ and massless, for the quantum fluctuations.
## VI Conclusions and Perspectives
In this work we have extended the standard time dependent Hartree-Fock approximation for the $`\varphi ^4`$ QFT, to include some non-gaussian features of the complete theory. We have presented a rather detailed study of the dynamical evolution out of equilibrium, in finite volume (a cubic box of size $`L`$ in $`3`$D), as well as in infinite volume. For comparison, we have also analyzed some static characteristics of the theory both in unbroken and broken symmetry phases.
By means of a proper substitution of the bare coupling constant with the renormalized coupling constant (fully justified by diagrammatic consideration), we have been able to obtain equations of motion completely independent of the ultraviolet cut-off (apart from a slight dependence on inverse powers, that is, however, ineluctable because of the Landau pole). We have described in detail the shape of the ground state, showing how a broken symmetry scenario can be recovered from the quantum mechanical model, when the volume diverges.
Moreover, we have shown that, within this slightly enlarged tdHF approach that allows for nonโgaussian wavefunctions, one might recover the usual gaussian HF approximation in a more controlled way. In fact, studying the late time dynamics, we have confirmed the presence of a time scale $`\tau _L`$, proportional to the linear size $`L`$ of the box, at which the evolution ceases to be similar to the infinite volume one. At the same time, the lowโlying modes amplitudes have grown to order $`L`$. The same phenomenon has been observed in the $`O(N)`$ model . Looking at this result in the framework of our extended tdHF approximation, one realizes that the growth of longโwavelength fluctuations to order $`L`$ in fact undermines the selfโconsistency of the gaussian HF itself. In fact, in our tdHF approach the initial gaussian wavefunctions are allowed to evolve into nonโgaussian forms, but they simply do not do it in a macroscopic way, within a further harmonic approximation for the evolution, so that in the infiniteโvolume limit they are indistinguishable from gaussians at all times. But when $`M^2`$ is on average not or order $`L^0`$, but much less, as it happens for suitable initial conditions, infrared modes of order $`L`$ will be dominated by the quartic term in our Schroedinger equations (5), showing a possible internal inconsistency of the gaussians approximation.
An other manifestation of the weakness of the HF scheme is the curious โstopping at the spinodal lineโ of the width of the gaussian quantum fluctuations, when the initial configuration does not break the symmetry. This does not happen in the large $`N`$ approach because of different coupling of transverse mode (the only ones that survive in the $`N\mathrm{}`$ limit) with respect to the longitudinal modes of the $`N=1`$ case in the HF approach.
We have also described the nonโtrivial phenomenology of the infiniteโvolume lateโtime evolution in the gaussian approximation, showing how the dynamical Maxwell construction differs from the $`N=\mathrm{}`$ case. In fact, we have observed the presence of an unstable interval, contained in the static flat region which is forbidden as attractor of the asymptotic evolution. This region corresponds, more or less, to the spinodal region of the classical potential, with the obvious exception of the origin. In particular, we have found that the energy flux between the classical degree of freedom and the bath of quantum fluctuations is quite complex and not monotonous. In other words, since we start from initial conditions where the fluctuation energy is not minimal, there are special situations where enough energy is transferred from the bath to the condensate, pushing it beyond the top of the potential hill.
Clearly further study, both analytical and numerical, is needed in our tdHF approach to better understand the dynamical evolution of quantum fluctuations in the broken symmetry phase coupled to the condensate. An interesting direction is the investigation of the case of finite $`N`$, in order to interpolate smoothly the results for $`N=1`$ to those of the $`1/N`$ approach. It should be noted, in fact, that the theory with a single scalar field contains only the longitudinal mode (by definition), while only the transverse modes are relevant in the large $`N`$ limit. Hence a better understanding of the coupling between longitudinal and transverse modes is necessary.
In this direction, another relevant point is whether the Goldstone theorem is respected in the HF approximation . It would be interesting also to study the dynamical realization of the Goldstone paradigm, namely the asymptotic vanishing of the effective mass in the broken symmetry phases, in different models; this issue needs further study in the $`2D`$ case , where it is known that the Goldstone theorem is not valid.
## VII Acknowledgements
C. D. thanks D. Boyanovsky, H. de Vega, R. Holman and M. Simionato for very interesting discussions. C. D. and E. M. thank MURST and INFN for financial support. |
warning/0001/physics0001065.html | ar5iv | text | # Beam Coupling Impedances of Small DiscontinuitiesLectures presented at the US Particle Accelerator School in Tucson, AZ on January 17-21, 2000, as a part of R.L. Glucksternโs course โAnalytic Methods for Calculating Coupling Impedancesโ
## 1 Introduction
A common tendency in design of modern accelerators is to minimize beam-chamber coupling impedances to avoid beam instabilities and reduce heating. Even contributions from tiny discontinuities like pumping holes have to be accounted for because of their large number. Numerical time-domain methods are rather involved and time-consuming for small obstacles, especially for those protruding inside the beam pipe. This makes analytical methods for calculating the impedances of small discontinuities very important.
A discontinuity on the wall of a beam pipe โ a hole, cavity, post, mask, etc. โ is considered to be small when its typical size is small compared to the size of the chamber cross section and the wavelength. Analytical calculations of the coupling impedances of small obstacles are based on the Bethe theory of diffraction of electromagnetic waves by a small hole in a metal plane . The methodโs basic idea is that the hole in the frequency range where the wavelength is large compared to the typical hole size, can be replaced by two induced dipoles, an electric and a magnetic one. The dipole magnitudes are proportional to the beam fields at the hole location, with the coefficients called polarizabilities . The fields diffracted by a hole into the vacuum chamber can be found as those radiated by these effective electric and magnetic dipoles, and integrating the fields gives us the coupling impedances. Following this path, the impedances produced by a small hole on the round pipe were calculated first in .
Since essentially the same idea works for any small obstacle, the method can be extended for arbitrary small discontinuities on the pipe with an arbitrary-shaped cross section. Analytical expressions for the coupling impedances of various small discontinuities on the wall of a cylindrical beam pipe with an arbitrary single-connected cross section have been obtained in Refs. -.
All the dependence on the discontinuity shape enters only through its polarizabilities. Therefore, the problem of calculating the impedance contribution from a given small discontinuity was reduced to finding its electric and magnetic polarizabilities, which can be done by solving proper electro- or magnetostatic problem. Useful analytical results have been obtained for various axisymmetric obstacles (cavities and irises) , as well as for holes and slots: circular and elliptic hole in a zero-thickness wall, circular and elliptic hole in a thick wall, various slots (some results are compiled in ), and for a ring-shaped cut . The impedances of various protrusions (post, mask, etc) have been calculated in .
This text is organized as follows. In Sections 2 and 3 a general derivation of the coupling impedances for a small discontinuity is given, and then in Sect. 4 the trapped modes are discussed. This part mostly follows Ref. . In Sect. 5 we collected for the reader convenience some practical formulas for the coupling impedances of various small discontinuities. In a compact form, many of these formulas are included in the handbook .
## 2 Fields
### 2.1 Beam Fields
Let us consider an infinite cylindrical pipe with an arbitrary cross section $`S`$ and perfectly conducting walls. The $`z`$ axis is directed along the pipe axis, a hole (or another small discontinuity like a post, cavity or mask) is located at the point ($`\stackrel{}{b},z=0`$), and its typical size $`h`$ satisfies $`hb`$. To evaluate the coupling impedance one has to calculate the fields induced in the chamber by a given current. If an ultrarelativistic point charge $`q`$ moves parallel to the chamber axis with the transverse offset $`\stackrel{}{s}`$ from the axis, the fields harmonics $`\stackrel{}{E}^b,\stackrel{}{H}^b`$ produced by this charge on the chamber wall without discontinuity would be
$`E_\nu ^b(\stackrel{}{s},z;\omega )`$ $`=`$ $`Z_0H_\tau ^b(\stackrel{}{s},z;\omega )`$
$`=`$ $`Z_0qe^{ikz}{\displaystyle \underset{n,m}{}}k_{nm}^2e_{nm}(\stackrel{}{s})_\nu e_{nm}(\stackrel{}{b}),`$
where $`Z_0=120\pi `$ Ohms is the impedance of free space, and $`k_{nm}^2`$, $`e_{nm}(\stackrel{}{r})`$ are eigenvalues and orthonormalized eigenfunctions (EFs) of the 2D boundary problem in $`S`$:
$$\left(^2+k_{nm}^2\right)e_{nm}=0;e_{nm}|_S=0.$$
(2)
Here $`\stackrel{}{}`$ is the 2D gradient in plane $`S`$; $`k=\omega /c`$; $`\widehat{\nu }`$ means an outward normal unit vector, $`\widehat{\tau }`$ is a unit vector tangent to the boundary $`S`$ of the chamber cross section $`S`$, and $`\{\widehat{\nu },\widehat{\tau },\widehat{z}\}`$ form a right-handed basis. The differential operator $`_\nu `$ is the scalar product $`_\nu =\widehat{\nu }\stackrel{}{}`$. The eigenvalues and EFs for circular and rectangular cross sections are given in the Appendix.
Let us introduce the following notation for the sum in Eq. (2.1)
$$e_\nu (\stackrel{}{s})=\underset{g}{}k_g^2e_g(\stackrel{}{s})_\nu e_g(\stackrel{}{b})$$
(3)
where $`g=\{n,m\}`$ is a generalized index. From a physical viewpoint, this is just a normalized electrostatic field produced at the hole location by a filament charge displaced from the chamber axis by distance $`\stackrel{}{s}`$. It satisfies the normalization condition
$$_S๐le_\nu (\stackrel{}{s})=1,$$
(4)
where integration goes along the boundary $`S`$, which is a consequence of the Gauss law. It follows from the fact that Eq. (3) gives the boundary value of $`\stackrel{}{e}_\nu (\stackrel{}{s})\stackrel{}{}\mathrm{\Phi }(\stackrel{}{r}\stackrel{}{s})`$, where $`\mathrm{\Phi }(\stackrel{}{r}\stackrel{}{s})`$ is the Green function of boundary problem (2): $`^2\mathrm{\Phi }(\stackrel{}{r}\stackrel{}{s})=\delta (\stackrel{}{r}\stackrel{}{s})`$. For the symmetric case of an on-axis beam in a circular pipe of radius $`b`$ from Eq. (4) immediately follows $`e_\nu (0)=1/(2\pi b)`$. It can also be derived by directly summing up the series in Eq. (3) for this particular case.
### 2.2 Effective Dipoles and Polarizabilities
At distances $`l`$ such that $`hlb`$, the fields radiated by the hole into the pipe are equal to those produced by effective dipoles
$`P_\nu `$ $`=`$ $`\chi \epsilon _0E_\nu ^h/2;M_\tau =(\psi _{\tau \tau }H_\tau ^h+\psi _{\tau z}H_z^h)/2;`$
$`M_z`$ $`=`$ $`(\psi _{z\tau }H_\tau ^h+\psi _{zz}H_z^h)/2,`$ (5)
where superscript โ$`h`$โ means that the fields are taken at the hole. Polarizabilities $`\psi ,\chi `$ are related to the effective ones $`\alpha _e,\alpha _m`$ used in as $`\alpha _e=\chi /2`$ and $`\alpha _m=\psi /2`$, so that for a circular hole of radius $`a`$ in a thin wall $`\psi =8a^3/3`$ and $`\chi =4a^3/3`$ . In general, $`\psi `$ is a symmetric 2D-tensor, which can be diagonalized. If the hole is symmetric, and its symmetry axis is parallel to $`\widehat{z}`$, the skew terms vanish, i.e. $`\psi _{\tau z}=\psi _{z\tau }=0`$. In a more general case of a non-zero tilt angle $`\alpha `$ between the major symmetry axis and $`\widehat{z}`$,
$`\psi _{\tau \tau }`$ $`=`$ $`\psi _{}\mathrm{cos}^2\alpha +\psi _{}\mathrm{sin}^2\alpha ,`$
$`\psi _{\tau z}`$ $`=`$ $`\psi _{z\tau }=(\psi _{}\psi _{})\mathrm{sin}\alpha \mathrm{cos}\alpha ,`$ (6)
$`\psi _{zz}`$ $`=`$ $`\psi _{}\mathrm{sin}^2\alpha +\psi _{}\mathrm{cos}^2\alpha ,`$
where $`\psi _{}`$ is the longitudinal magnetic susceptibility (for the external magnetic field along the major axis), and $`\psi _{}`$ is the transverse one (the field is transverse to the major axis of the hole). When the effective dipoles are obtained, e.g., by substituting beam fields (2.1) into Eqs. (5), one can calculate the fields in the chamber as a sum of waveguide eigenmodes excited in the chamber by the dipoles, and find the impedance. This approach has been carried out for a circular pipe in , and for an arbitrary chamber in . The polarizabilities for various types of small discontinuities are discussed in detail in Section 5.
### 2.3 Radiated Fields
The radiated fields in the chamber can be expanded in a series in TM- and TE-eigenmodes as
$`\stackrel{}{F}={\displaystyle \underset{nm}{}}\left[A_{nm}^+\stackrel{}{F}_{nm}^{(E)+}\theta (z)+A_{nm}^{}\stackrel{}{F}_{nm}^{(E)}\theta (z)\right]+`$ (7)
$`{\displaystyle \underset{nm}{}}\left[B_{nm}^+\stackrel{}{F}_{nm}^{(H)+}\theta (z)+B_{nm}^{}\stackrel{}{F}_{nm}^{(H)}\theta (z)\right],`$
where $`\stackrel{}{F}`$ means either $`\stackrel{}{E}`$ or $`\stackrel{}{H}`$, superscripts โ$`\pm `$โ denote waves radiated respectively in the positive (+, $`z>0`$) or negative ($``$, $`z<0`$) direction, and $`\theta (z)`$ is the Heaviside step function. The fields $`F_{nm}^{(E)}`$ of $`\{n,m\}`$th TM-eigenmode in Eq. (7) are expressed in terms of EFs (2)
$`E_z^{}`$ $`=`$ $`k_{nm}^2e_{nm}\mathrm{exp}(\pm \mathrm{\Gamma }_{nm}z);H_z^{}=0;`$
$`\stackrel{}{E}_t^{}`$ $`=`$ $`\pm \mathrm{\Gamma }_{nm}\stackrel{}{}e_{nm}\mathrm{exp}(\pm \mathrm{\Gamma }_{nm}z);`$ (8)
$`\stackrel{}{H}_t^{}`$ $`=`$ $`{\displaystyle \frac{ik}{Z_0}}\widehat{z}\times \stackrel{}{}e_{nm}\mathrm{exp}(\pm \mathrm{\Gamma }_{nm}z),`$
where propagation factors $`\mathrm{\Gamma }_{nm}=(k_{nm}^2k^2)^{1/2}`$ should be replaced by $`i\beta _{nm}`$ with $`\beta _{nm}=(k^2k_{nm}^2)^{1/2}`$ for $`k>k_{nm}`$. For given values of dipoles (5) the unknown coefficients $`A_{nm}^\pm `$ can be found using the Lorentz reciprocity theorem
$$A_{nm}^\pm =a_{nm}M_\tau \pm b_{nm}P_\nu ,$$
(9)
with
$$a_{nm}=\frac{ikZ_0}{2\mathrm{\Gamma }_{nm}k_{nm}^2}_\nu e_{nm}^h;b_{nm}=\frac{1}{2\epsilon _0k_{nm}^2}_\nu e_{nm}^h.$$
(10)
The fields $`F_{nm}^{(H)}`$ of the TE<sub>nm</sub>-eigenmode in Eq. (7) are
$`H_z^{}`$ $`=`$ $`k_{nm}^2h_{nm}\mathrm{exp}(\pm \mathrm{\Gamma }_{nm}^{}z);E_z^{}=0;`$
$`\stackrel{}{H}_t^{}`$ $`=`$ $`\pm \mathrm{\Gamma }_{nm}^{}\stackrel{}{}h_{nm}\mathrm{exp}(\pm \mathrm{\Gamma }_{nm}^{}z);`$ (11)
$`\stackrel{}{E}_t^{}`$ $`=`$ $`ikZ_0\widehat{z}\times \stackrel{}{}h_{nm}\mathrm{exp}(\pm \mathrm{\Gamma }_{nm}^{}z),`$
with propagation factors $`\mathrm{\Gamma }_{nm}^{}=(k_{nm}^2k^2)^{1/2}`$ replaced by $`i\beta _{nm}^{}=i(k^2k_{nm}^2)^{1/2}`$ when $`k>k_{nm}^{}`$. Here EFs $`h_{nm}`$ satisfy the boundary problem (2) with the Neumann boundary condition $`_\nu h_{nm}|_S=0`$, and $`k_{nm}^2`$ are corresponding eigenvalues, see in Appendix. The TE-mode excitation coefficients in the expansion (7) for the radiated fields are
$$B_{nm}^\pm =\pm c_{nm}M_\tau +d_{nm}P_\nu +q_{nm}M_z,$$
(12)
where
$`c_{nm}`$ $`=`$ $`{\displaystyle \frac{1}{2k_{nm}^2}}_\tau h_{nm}^h;q_{nm}={\displaystyle \frac{1}{2\mathrm{\Gamma }_{nm}^{}}}h_{nm}^h;`$
$`d_{nm}`$ $`=`$ $`{\displaystyle \frac{ik}{2Z_0\epsilon _0\mathrm{\Gamma }_{nm}^{}k_{nm}^2}}_\tau h_{nm}^h.`$ (13)
### 2.4 Fields near Hole with Radiation Corrections
A more refined theory should take into account the reaction of radiated waves back on the hole. Adding corrections to the beam fields (2.1) due to the radiated waves in the vicinity of the hole gives
$`E_\nu `$ $`=`$ $`{\displaystyle \frac{E_\nu ^b+\psi _{z\tau }\mathrm{\Sigma }_x^{}Z_0H_\tau +\psi _{zz}\mathrm{\Sigma }_x^{}Z_0H_z}{1\chi (\mathrm{\Sigma }_1\mathrm{\Sigma }_1^{})}},`$ (14)
$`H_\tau `$ $`=`$ $`{\displaystyle \frac{H_\tau ^b+\psi _{\tau z}(\mathrm{\Sigma }_2\mathrm{\Sigma }_2^{})H_z}{1\psi _{\tau \tau }(\mathrm{\Sigma }_2\mathrm{\Sigma }_2^{})}},`$ (15)
$`H_z`$ $`=`$ $`{\displaystyle \frac{\chi \mathrm{\Sigma }_x^{}E_\nu /Z_0+\psi _{z\tau }\mathrm{\Sigma }_3^{}H_\tau }{1\psi _{zz}\mathrm{\Sigma }_3^{}}},`$ (16)
where ($`s=\{n,m\}`$ is a generalized index)
$`\mathrm{\Sigma }_1`$ $`=`$ $`{\displaystyle \frac{1}{4}}{\displaystyle \underset{s}{}}{\displaystyle \frac{\mathrm{\Gamma }_s\left(_\nu e_s^h\right)^2}{k_s^2}};\mathrm{\Sigma }_2={\displaystyle \frac{k^2}{4}}{\displaystyle \underset{s}{}}{\displaystyle \frac{\left(_\nu e_s^h\right)^2}{\mathrm{\Gamma }_sk_s^2}};`$
$`\mathrm{\Sigma }_1^{}`$ $`=`$ $`{\displaystyle \frac{k^2}{4}}{\displaystyle \underset{s}{}}{\displaystyle \frac{\left(_\tau h_s^h\right)^2}{\mathrm{\Gamma }_s^{}k_s^2}};\mathrm{\Sigma }_2^{}={\displaystyle \frac{1}{4}}{\displaystyle \underset{s}{}}{\displaystyle \frac{\mathrm{\Gamma }_s^{}\left(_\tau h_s^h\right)^2}{k_s^2}};`$
$`\mathrm{\Sigma }_x^{}`$ $`=`$ $`i{\displaystyle \frac{k}{4}}{\displaystyle \underset{s}{}}{\displaystyle \frac{h_s^h_\tau h_s^h}{\mathrm{\Gamma }_s^{}}};\mathrm{\Sigma }_3^{}={\displaystyle \frac{1}{4}}{\displaystyle \underset{s}{}}{\displaystyle \frac{k_s^2\left(h_s^h\right)^2}{\mathrm{\Gamma }_s^{}}}.`$ (17)
Since this consideration works at distances larger than $`h`$, one should restrict the summation in Eq. (17) to the values of $`s=\{n,m\}`$ such that $`k_sh1`$ and $`k_s^{}h1`$.
## 3 Impedance
### 3.1 Longitudinal Impedance
The generalized longitudinal impedance of the hole depends on the transverse offsets from the chamber axis $`\stackrel{}{s}`$ of the leading particle and $`\stackrel{}{t}`$ of the test particle, and is defined as
$$Z(k;\stackrel{}{s},\stackrel{}{t})=\frac{1}{q}_{\mathrm{}}^{\mathrm{}}๐ze^{ikz}E_z(\stackrel{}{t},z;\omega ),$$
(18)
where the longitudinal field $`E_z(\stackrel{}{t},z;\omega )`$ is taken along the test particle path. The displacements from the axis are assumed to be small, $`sb`$ and $`tb`$. The impedance $`Z(k;\stackrel{}{s},\stackrel{}{t})`$ includes higher multipole longitudinal impedances, and in the limit $`s,t0`$ gives the usual monopole one $`Z(k)=Z(k;0,0)`$. To calculate $`E_z(\stackrel{}{t},z;\omega )`$, we use Eq. (7) with coefficients (9) and (12) in which the corrected near-hole fields (14)-(16) are substituted \[a dependence on $`\stackrel{}{s}`$ enters via beam fields (2.1)\]. It yields
$`Z(k;\stackrel{}{s},\stackrel{}{t})={\displaystyle \frac{ikZ_0e_\nu (\stackrel{}{s})e_\nu (\stackrel{}{t})}{2}}\times `$ (19)
$`\times \left[{\displaystyle \frac{\psi _{\tau \tau }}{1\psi _{\tau \tau }(\mathrm{\Sigma }_2\mathrm{\Sigma }_2^{})}}+\psi _{\tau z}^2\mathrm{\Sigma }_3^{}{\displaystyle \frac{\chi }{1\chi (\mathrm{\Sigma }_1\mathrm{\Sigma }_1^{})}}\right],`$
where $`e_\nu (\stackrel{}{r})`$ is defined above by Eq. (3). In practice, we are usually interested only in the monopole term $`Z(k)=Z(k;0,0)`$, and will mostly use Eq. (19) with replacement $`e_\nu (\stackrel{}{s})e_\nu (\stackrel{}{t})\stackrel{~}{e}_\nu ^2`$, where $`\stackrel{~}{e}_\nu e_\nu (0)`$. In deriving Eq. (19) we have neglected the coupling terms between $`E_\nu `$, $`H_\tau `$ and $`H_z`$, cf. Eqs. (14)-(16), which contribute to the third order of an expansion discussed below, and also have taken into account that $`\psi _{\tau z}=\psi _{z\tau }`$.
For a small discontinuity, polarizabilities $`\psi ,\chi =O(h^3)`$, and they are small compared to $`b^3`$. If we expand the impedance (19) in a perturbation series in polarizabilities, the first order gives
$$Z_1(k)=\frac{ikZ_0\stackrel{~}{e}_{\nu }^{}{}_{}{}^{2}}{2}\left(\psi _{\tau \tau }\chi \right),$$
(20)
that is exactly the inductive impedance obtained in for an arbitrary cross section of the chamber. For a particular case of a circular pipe, from either direct summation in (2.1) or applying the Gauss law, one gets $`\stackrel{~}{e}_\nu =1/(2\pi b)`$. Substituting that into Eq. (20) leads to a well-known result :
$$Z(k)=ikZ_0\frac{\psi _{\tau \tau }\chi }{8\pi ^2b^2}=ikZ_0\frac{\alpha _e+\alpha _m}{4\pi ^2b^2},$$
(21)
where we recall that two definitions of the polarizabilities are related as $`\alpha _e=\chi /2`$ and $`\alpha _m=\psi _{\tau \tau }/2`$.
From a physical point of view, keeping only the first order term (20) corresponds to dropping out all radiation corrections in Eqs. (14)-(16). These corrections first reveal themselves in the second order term
$`Z_2(k)={\displaystyle \frac{ikZ_0\stackrel{~}{e}_{\nu }^{}{}_{}{}^{2}}{2}}[\psi _{\tau \tau }^2(\mathrm{\Sigma }_2\mathrm{\Sigma }_2^{})+\psi _{\tau z}^2\mathrm{\Sigma }_3^{}`$ (22)
$`+\chi ^2(\mathrm{\Sigma }_1^{}\mathrm{\Sigma }_1)],`$
which at frequencies above the chamber cutoff has both a real and imaginary part. The real part of the impedance is
$`ReZ_2(k)`$ $`=`$ $`{\displaystyle \frac{k^3Z_0\stackrel{~}{e}_{\nu }^{}{}_{}{}^{2}}{8}}\{\psi _{\tau z}^2{\displaystyle \underset{s}{\overset{<}{}}}{\displaystyle \frac{k_s^2\left(h_s^h\right)^2}{k^2\beta _s^{}}}`$
$`+`$ $`\psi _{\tau \tau }^2\left[{\displaystyle \underset{s}{\overset{<}{}}}{\displaystyle \frac{\left(_\nu e_s^h\right)^2}{\beta _sk_s^2}}+{\displaystyle \underset{s}{\overset{<}{}}}{\displaystyle \frac{\beta _s^{}\left(_\tau h_s^h\right)^2}{k^2k_s^2}}\right]`$
$`+`$ $`\chi ^2[{\displaystyle \underset{s}{\overset{<}{}}}{\displaystyle \frac{\beta _s\left(_\nu e_s^h\right)^2}{k^2k_s^2}}+{\displaystyle \underset{s}{\overset{<}{}}}{\displaystyle \frac{\left(_\tau h_s^h\right)^2}{\beta _s^{}k_s^2}}]\},`$
where the sums include only a finite number of the eigenmodes propagating in the chamber at a given frequency, i.e. those with $`k_s<k`$ or $`k_s^{}<k`$.
The dependence of $`ReZ`$ on frequency is rather complicated; it has sharp peaks near the cutoffs of all propagating eigenmodes of the chamber, and increases in average with the frequency increase. Well above the chamber cutoff, i.e. when $`kb1`$ (but still $`kh1`$ to justify the Bethe approach), this dependence can be derived as follows. If the waveguide cross section $`S`$ is a simply connected region, the average number $`n(k)`$ of the eigenvalues $`k_s`$ (or $`k_s^{}`$) which are less than $`k`$, for $`kb1`$, is proportional to $`k^2`$ :
$$n(k)\frac{S}{4\pi }k^2+O(k),$$
where $`S`$ is the area of the cross section. Using this property, and taking into account that $`_\nu e_s^hk_se_s^h`$, and $`_\tau h_s^hk_s^{}h_s^h`$, we replace sums in the RHS of Eq. (3.1) by integrals as $`_s^<^k๐k\frac{d}{dk}n(k)`$. It turns out that all sums in Eq. (3.1) have the same asymptotic behavior, being linear in $`k`$, and as a result, $`ReZk^4`$. Obtaining the exact coefficient in this dependence seems rather involved for a general $`S`$, but it can be easily done for a rectangular chamber, see in Appendix B. The result is
$$ReZ=\frac{Z_0k^4\stackrel{~}{e}_{\nu }^{}{}_{}{}^{2}}{12\pi }(\psi _{\tau \tau }^2+\psi _{\tau z}^2+\chi ^2).$$
(24)
Remarkably, the same answer (for $`\psi _{\tau z}=0`$) has been obtained in Ref. simply by calculating the energy radiated by the dipoles into a half-space. The physical reason for this coincidence is clear: at frequencies well above the cutoff the effective dipoles radiate into the waveguide the same energy as into an open half-space.
Strictly speaking, the real part of impedance is non-zero even below the chamber cutoff, due to radiation outside. In the case of a thin wall, $`ReZ`$ below the cutoff can be estimated by Eq. (24), and twice that for high frequencies, $`kb1`$. For a thick wall, the contribution of the radiation outside to $`ReZ`$ is still given by Eq. (24), but with the outside polarizabilities substituted, and it decreases exponentially with the thickness increase .
The real part of the impedance is related to the power $`P`$ scattered by the hole into the beam pipe as $`ReZ=2P/q^2`$. These energy considerations can be used as an alternative way for the impedance calculation. The radiated power is
$$P=\underset{s}{}\left[A_s^2P_s^{(E)}+B_s^2P_s^{(H)}\right],$$
where we sum over all propagating modes in both directions, and $`P_s`$ means the time-averaged power radiated in $`s`$th eigenmode:
$$P_s^{(E)}=k\beta _sk_s^2/(2Z_0)\text{ and }P_s^{(H)}=Z_0k\beta _s^{}k_s^2/2.$$
Substituting beam fields (2.1) into Eqs. (9)-(13) for the coefficients $`A_s`$ and $`B_s`$ and performing calculations gives us exactly the result (3.1). Such an alternative derivation of the real part has been carried out in Ref. for a circular pipe with a symmetric untilted hole ($`\psi _{\tau z}=0`$). Our result (3.1) coincides, in this particular case, with that of Reference . It is appropriate to mention also that in this case at high frequencies the series has been summed approximately using asymptotic expressions for roots of the Bessel functions, and the result, of course, agrees with Eq. (24).
One should note that the additional $`\psi _{\tau z}^2`$-term in Eq. (3.1) is important in some particular cases. For example, this skew term gives a leading contribution to $`ReZ`$ for a long and slightly tilted slot, because $`\psi _{\tau z}`$ can be much larger than $`\psi _{\tau \tau }`$ in this case, since $`\psi _{}\psi _{}`$, cf. Eqs. (6).
### 3.2 Transverse Impedance
We will make use of the expression for the generalized longitudinal impedance $`Z(k;\stackrel{}{s},\stackrel{}{t})`$, Eq. (19). According to the Panofsky-Wenzel theorem, the transverse impedance can be derived as $`\stackrel{}{Z}_{}(k;\stackrel{}{s},\stackrel{}{t})=\stackrel{}{}Z(k;\stackrel{}{s},\stackrel{}{t})/(ks)`$, see, e.g., for details. This way leads to the expression
$`\stackrel{}{Z}_{}(k;\stackrel{}{s},\stackrel{}{t})={\displaystyle \frac{iZ_0e_\nu ^{dip}(\stackrel{}{s})\stackrel{}{}e_\nu (\stackrel{}{t})}{2s}}\times `$ (25)
$`\times \left[{\displaystyle \frac{\psi _{\tau \tau }}{1\psi _{\tau \tau }(\mathrm{\Sigma }_2\mathrm{\Sigma }_2^{})}}+\psi _{\tau z}^2\mathrm{\Sigma }_3^{}{\displaystyle \frac{\chi }{1\chi (\mathrm{\Sigma }_1\mathrm{\Sigma }_1^{})}}\right],`$
where $`e_\nu ^{dip}(\stackrel{}{s})=\stackrel{}{s}\stackrel{}{}e_\nu (\stackrel{}{s})`$.
Going to the limit $`st0`$, we get the usual dipole transverse impedance
$`\stackrel{}{Z}_{}(k)=iZ_0{\displaystyle \frac{1}{2}}(d_x^2+d_y^2)\stackrel{}{a}_d\mathrm{cos}(\phi _b\phi _d)\times `$ (26)
$`\times \left[{\displaystyle \frac{\psi _{\tau \tau }}{1\psi _{\tau \tau }(\mathrm{\Sigma }_2\mathrm{\Sigma }_2^{})}}+\psi _{\tau z}^2\mathrm{\Sigma }_3^{}{\displaystyle \frac{\chi }{1\chi (\mathrm{\Sigma }_1\mathrm{\Sigma }_1^{})}}\right].`$
Here $`x,y`$ are the horizontal and vertical coordinates in the chamber cross section; $`d_x_xe_\nu (0)`$, $`d_y_ye_\nu (0)`$; $`\phi _b=\phi _s=\phi _t`$ is the azimuthal angle of the beam position in the cross-section plane; $`\stackrel{}{a}_d=\stackrel{}{a}_x\mathrm{cos}\phi _d+\stackrel{}{a}_y\mathrm{sin}\phi _d`$ is a unit vector in this plane in direction $`\phi _d`$, which is defined by conditions $`\mathrm{cos}\phi _d=d_x/\sqrt{d_x^2+d_y^2}`$, $`\mathrm{sin}\phi _d=d_y/\sqrt{d_x^2+d_y^2}`$. It is seen from Eq. (26) that the angle $`\phi _d`$ shows the direction of the transverse-impedance vector $`\stackrel{}{Z}_{}`$ and, therefore, of the beam-deflecting force. Moreover, the value of $`Z_{}`$ is maximal when the beam is deflected along this direction and vanishes when the beam offset is perpendicular to it.
The equation (26) includes the corrections due to waves radiated by the hole into the chamber in exactly the same way as Eq. (19) for the longitudinal impedance. If we expand it in a series in the polarizabilities, the first order of the square brackets in (26) gives $`(\psi _{\tau \tau }\chi )`$, and the resulting inductive impedance is :
$$\stackrel{}{Z}_{}(k)=iZ_0\frac{1}{2}(d_x^2+d_y^2)\stackrel{}{a}_d\mathrm{cos}(\phi _b\phi _d)\left(\psi _{\tau \tau }\chi \right).$$
(27)
For a circular pipe, $`d_x=\mathrm{cos}\phi _h/(\pi b^2)`$ and $`d_y=\mathrm{sin}\phi _h/(\pi b^2)`$, where $`\phi _h`$ is the azimuthal position of the discontinuity (hole). As a result, $`\phi _d=\phi _h`$, and the deflecting force is directed toward (or opposite to) the hole. Note that in axisymmetric structures the beam-deflecting force is directed along the beam offset; the presence of an obstacle obviously breaks this symmetry. For a general cross section, the direction of the deflecting force depends on the hole position in a complicated way, see for rectangular and elliptic chambers.
The transverse impedance of a discontinuity on the wall of a circular pipe has a simple form :
$$\stackrel{}{Z}_{circ}(k)=iZ_0\frac{\psi _{\tau \tau }\chi }{2\pi ^2b^4}\stackrel{}{a}_h\mathrm{cos}(\phi _b\phi _d)=iZ_0\frac{\alpha _m+\alpha _e}{\pi ^2b^4}\stackrel{}{a}_h\mathrm{cos}(\phi _b\phi _d),$$
(28)
where $`\stackrel{}{a}_h`$ is a unit vector in the direction from the chamber axis to the discontinuity, orthogonal to $`\widehat{z}`$.
For $`M`$ ($`M3`$) holes uniformly spaced in one cross section of a circular beam pipe, a vector sum of M expressions (28) gives the transverse impedance as
$$\stackrel{}{Z}_{circM}(k)=iZ_0\frac{\alpha _m+\alpha _e}{\pi ^2b^4}\frac{M}{2}\stackrel{}{a}_b,$$
(29)
where $`\stackrel{}{a}_b=\stackrel{}{s}/|\stackrel{}{s}|`$ is a unit vector in the direction of the beam transverse offset. One can see that the deflecting force is now directed along the beam offset, i.e. some kind of the axial symmetry restoration occurs. The maximal value of $`Z_{}`$ for $`M`$ holes which are uniformly spaced in one cross-section is only $`M/2`$ times larger than that for $`M=1`$. Moreover, the well-known empirical relation $`Z_{}=(2/b^2k)Z`$, which is justified only for axisymmetric structures, holds in this case also.
The second order term in Eq. (26) includes $`ReZ_{}`$, cf. Sect. 3.1 for the longitudinal impedance.
## 4 Trapped Modes
So far we considered the perturbation expansion of Eq. (19) implicitly assuming that correction terms $`O(\psi )`$ and $`O(\chi )`$ in the denominators of its right-hand side (RHS) are small compared to 1. Under certain conditions this assumption is incorrect, and it leads to some non-perturbative results. Indeed, at frequencies slightly below the chamber cut-offs, $`0<k_skk_s`$ (or the same with replacement $`k_sk_s^{}`$), a single term in sums $`\mathrm{\Sigma }_1^{}`$, $`\mathrm{\Sigma }_2`$, or $`\mathrm{\Sigma }_3^{}`$ becomes very large, due to very small $`\mathrm{\Gamma }_s=(k_s^2k^2)^{1/2}`$ (or $`\mathrm{\Gamma }_s^{}`$) in its denominator, and then the โcorrectionsโ $`\psi \mathrm{\Sigma }`$ or $`\chi \mathrm{\Sigma }`$ can be of the order of 1. As a result, one of the denominators of the RHS of Eqs. (19) can vanish, which corresponds to a resonance of the coupling impedance. On the other hand, vanishing denominators in Eqs. (14)-(16) mean the existence of non-perturbative eigenmodes of the chamber with a hole, since non-trivial solutions $`E,H0`$ exist even for vanishing external (beam) fields $`E^b,H^b=0`$. These eigenmodes are nothing but the trapped modes studied in for a circular waveguide with a small discontinuity. In our approach, one can easily derive parameters of trapped modes for waveguides with an arbitrary cross section.
### 4.1 Frequency Shifts
Let us for brevity restrict ourselves to the case $`\psi _{\tau z}=0`$ and consider Eq. (15) in more detail. For $`H^b=0`$ we have
$$H_\tau \left[1\psi _{\tau \tau }\frac{k^2\left(_\nu e_s^h\right)^2}{4\mathrm{\Gamma }_sk_s^2}+\mathrm{}\right]=0,$$
(30)
where $`s\{nm\}`$ is the generalized index, and $`\mathrm{}`$ mean all other terms of the series $`\mathrm{\Sigma }_2,\mathrm{\Sigma }_2^{}`$. At frequency $`\mathrm{\Omega }_s`$ slightly below the cutoff frequency $`\omega _s=k_sc`$ of the TM<sub>s</sub>-mode, the fraction in Eq. (30) is large due to small $`\mathrm{\Gamma }_s`$ in its denominator, and one can neglect the other terms. Then the condition for a non-trivial solution $`H_\tau 0`$ to exist is
$$\mathrm{\Gamma }_s\frac{1}{4}\psi _{\tau \tau }\left(_\nu e_s^h\right)^2.$$
(31)
In other words, there is a solution of the homogeneous, i.e., without external currents, Maxwell equations for the chamber with the hole, having the frequency $`\mathrm{\Omega }_s<\omega _s`$ โ the $`s`$th trapped TM-mode. When Eq. (31) is satisfied, the series (7) is obviously dominated by the single term $`A_sF_s^E`$; hence, the fields of the trapped mode have the form \[cf. Eq. (8)\]
$`_z`$ $`=`$ $`k_s^2e_s\mathrm{exp}(\mathrm{\Gamma }_s|z|);_z=0;`$
$`\stackrel{}{}_t`$ $`=`$ $`\text{sgn}(z)\mathrm{\Gamma }_s\stackrel{}{}e_s\mathrm{exp}(\mathrm{\Gamma }_s|z|);`$ (32)
$`Z_0\stackrel{}{}_t`$ $`=`$ $`ik\widehat{z}\times \stackrel{}{}e_s\mathrm{exp}(\mathrm{\Gamma }_s|z|),`$
up to some arbitrary amplitude. Strictly speaking, these expressions are valid at distances $`|z|>b`$ from the discontinuity. Typically, $`\psi _{\tau \tau }=O(h^3)`$ and $`_\nu e_s^h=O(1/b)`$, and, as a result, $`\mathrm{\Gamma }_sb1`$. It follows that the field of the trapped mode extends along the vacuum chamber over the distance $`1/\mathrm{\Gamma }_s`$, large compared to the chamber transverse dimension $`b`$.
The existence of the trapped modes in a circular waveguide with a small hole was first proved in , and conditions similar to Eq. (31) for this particular case were obtained in , using the Lorentz reciprocity theorem. From the general approach presented here for the waveguide with an arbitrary cross section, their existence follows in a natural way. Moreover, in such a derivation, the physical mechanism of this phenomenon becomes quite clear: a tangential magnetic field induces a magnetic moment on the hole, and the induced magnetic moment supports this field if the resonance condition (31) is satisfied, so that the mode can exist even without an external source. One should also note that the induced electric moment $`P_\nu `$ is negligible for the trapped TM-mode, since $`P_\nu =O(\mathrm{\Gamma }_sb)M_\tau `$, as follows from Eq. (32).
The equation (31) gives the frequency shift $`\mathrm{\Delta }\omega _s\omega _s\mathrm{\Omega }_s`$ of the trapped $`s`$th TM-mode down from the cutoff $`\omega _s`$
$$\frac{\mathrm{\Delta }\omega _s}{\omega _s}\frac{1}{32k_s^2}\psi _{\tau \tau }^2\left(_\nu e_s^h\right)^4.$$
(33)
In the case of a small hole this frequency shift is very small, and for the trapped mode (32) to exist, the width of the resonance should be smaller than $`\mathrm{\Delta }\omega _s`$. Contributions to the resonance width come from energy dissipation in the waveguide wall due to its finite conductivity, and from energy radiation inside the waveguide and outside, through the hole. Radiation escaping through the hole is easy to estimate , and for a thick wall it is exponentially small, e.g., . The damping rate due to a finite conductivity is $`\gamma =P/(2W)`$, where $`P`$ is the time-averaged power dissipation and $`W`$ is the total field energy in the trapped mode, which yields
$$\frac{\gamma _s}{\omega _s}=\frac{\delta }{4k_s^2}๐l\left(_\nu e_s\right)^2,$$
(34)
where $`\delta `$ is the skin-depth at frequency $`\mathrm{\Omega }_s`$, and the integration is along the boundary $`S`$. The evaluation of the radiation into the lower waveguide modes propagating in the chamber at given frequency $`\mathrm{\Omega }_s`$ is also straightforward , if one makes use of the coefficients of mode excitation by effective dipoles on the hole, Eqs. (9)-(13). The corresponding damping rate $`\gamma _R=O(\psi ^3)`$ is small compared to $`\mathrm{\Delta }\omega _s`$. For instance, if there is only one TE<sub>p</sub>-mode with the frequency below that for the lowest TM<sub>s</sub>-mode, like in a circular waveguide (H<sub>11</sub> has a lower cutoff than E<sub>01</sub>),
$$\frac{\gamma _R}{\mathrm{\Delta }\omega _s}=\frac{\psi _{\tau \tau }\beta _p^{}}{k_p^2}\left(_\nu h_s^h\right)^2,$$
(35)
where $`\beta _p^{}(k_s^2k_p^2)^{1/2}`$ because $`kk_s`$.
One can easily see that denominator $`[1\chi (\mathrm{\Sigma }_1\mathrm{\Sigma }_1^{})]`$ in Eq. (14) does not vanish because singular terms in $`\mathrm{\Sigma }_1^{}`$ have a โwrongโ sign. However, due to the coupling between $`E_\nu `$ and $`H_z`$, a non-trivial solution $`E_\nu ,H_z0`$ of simultaneous equations (14) and (16) can exist, even when $`E^b=0`$. The corresponding condition has the form
$$\mathrm{\Gamma }_{nm}^{}\frac{1}{4}\left[\psi _{zz}k_{nm}^2\left(h_{nm}^h\right)^2\chi \left(_\tau h_{nm}^h\right)^2\right],$$
(36)
which gives the frequency of the trapped TE<sub>nm</sub>-mode, provided the RHS of Eq. (36) is positive.
### 4.2 Impedance
The trapped mode (32) gives a resonance contribution to the longitudinal coupling impedance at $`\omega \mathrm{\Omega }_s`$
$$Z_s(\omega )=\frac{2i\mathrm{\Omega }_s\gamma _sR_s}{\omega ^2(\mathrm{\Omega }_si\gamma _s)^2},$$
(37)
where the shunt impedance $`R_s`$ can be calculated as that for a cavity with given eigenmodes, e.g. ,
$$R_s=\frac{\sigma \delta \left|๐z\mathrm{exp}(i\mathrm{\Omega }_sz/c)_z(z)\right|^2}{_{S_w}๐s|_\tau |^2}.$$
(38)
The integral in the denominator is taken over the inner wall surface, and we assume here that the power losses due to its finite conductivity dominate. Integrating in the numerator one should include all TM-modes generated by the effective magnetic moment on the hole using Eqs. (9)-(13), in spite of a large amplitude of only the trapped TM<sub>s</sub> mode. While all other amplitudes are suppressed by factor $`\mathrm{\Gamma }_sb1`$, their contributions are comparable to that from TM<sub>s</sub>, because this integration produces the factor $`\mathrm{\Gamma }_qb`$ for any TM<sub>q</sub> mode. The integral in the denominator is obviously dominated by TM<sub>s</sub>. Performing calculations yields
$$R_s=\frac{Z_0\stackrel{~}{e}_\nu ^2\psi _{\tau \tau }^3k_s\left(_\nu e_s^h\right)^4}{8\delta ๐l\left(_\nu e_s\right)^2},$$
(39)
where $`\stackrel{~}{e}_\nu =e_\nu (0)`$ is defined by Eq. (3).
Results for a particular shape of the chamber cross section can be obtained from the equations above by substituting the corresponding eigenfunctions (see Appendix).
One should note that typically the peak value $`R_s`$ of the impedance resonance due to one small hole is rather small except for the limit of a perfectly conducting wall, $`\delta 0`$ โ indeed, $`R_s(h/b)^9b/\delta `$, and $`hb`$. However, for many not-so-far separated holes, the resulting impedance can be much larger. The trapped modes for many discontinuities on a circular waveguide has been studied in Ref. , and the results can be readily transferred to the considered case of an arbitrary shape of the chamber cross section. In particular, it was demonstrated that the resonance impedance in the extreme case can be as large as $`N^3`$ times that for a single discontinuity, where $`N`$ is the number of discontinuities. It strongly depends on the distribution of discontinuities, or on the distance between them if a regular array is considered.
After the trapped modes in beam pipes with small holes were predicted theoretically , their existence was proved by experiments with perforated waveguides at CERN .
## 5 Analytical Formulae for Some Small Discontinuities
For reader convenience, in this section we collected analytical expressions for the coupling impedances of various small discontinuities. The expressions give the inductive part of the impedance and work well at frequencies below the chamber cutoff, and, in many cases, even at much higher frequencies. However, there can also exist resonances of the real part at frequencies near the cutoff for holes and cavities due to the trapped modes, as was shown in , and the real part of the impedance due to the energy radiated into the beam pipe should be taken into account at frequencies above the cutoff, see .
It is worth noting that both the longitudinal and transverse impedance are proportional to the same combination of polarizabilities, $`\alpha _e+\alpha _m`$, for any cross section of the beam pipe. Here we use the effective polarizabilities $`\alpha _e,\alpha _m`$ as defined in ; they are related to the magnetic susceptibility $`\psi `$ and the electric polarizability $`\chi `$ of an obstacle as $`\alpha _e=\chi /2`$ and $`\alpha _m=\psi /2`$. The real part of the impedance is proportional to $`\alpha _e^2+\alpha _m^2`$, and is usually small compared to the reactance at frequencies below the chamber cutoff. While the impedances below are written for a round pipe, more results for the other chamber cross sections, the impedance dependence on the obstacle position on the wall and on the beam position can be found in .
The longitudinal impedance of a small obstacle on the wall of a cylindrical beam pipe with a circular transverse cross section of radius $`R`$ is simply (up to notations, it is the same Eq. (21) above)
$$Z(k)=ikZ_0\frac{\alpha _e+\alpha _m}{4\pi ^2R^2},$$
(40)
where $`Z_0=120\pi `$ Ohms is the impedance of free space, $`k=\omega /c`$ is the wave number, and $`\alpha _e,\alpha _m`$ are the electric and magnetic polarizabilities of the discontinuity. The polarizabilities depend on the obstacle shape and size.
The transverse dipole impedance of the discontinuity for this case is
$$\stackrel{}{Z}_{}(\omega )=iZ_0\frac{\alpha _m+\alpha _e}{\pi ^2R^4}\stackrel{}{a}_h\mathrm{cos}(\phi _h\phi _b),$$
(41)
where $`\stackrel{}{a}_h`$ is the unit vector directed to the obstacle in the chamber transverse cross section containing it, $`\phi _h`$ and $`\phi _b`$ are azimuthal angles of the obstacle and beam in this cross section.
### 5.1 Holes and Slots
For a circular hole with radius $`a`$ in a thin wall, when thickness $`ta`$, the polarizabilities are $`\alpha _m=4a^3/3,\alpha _e=2a^3/3`$, so that the impedance Eq. (40) takes a simple form
$$Z(k)=ikZ_0\frac{a^3}{6\pi ^2R^2},$$
(42)
and similarly for Eq. (41). For the hole in a thick wall, $`ta`$, the sum $`(\alpha _m+\alpha _e)=2a^3/3`$ should be multiplied by a factor 0.56, see . There are also analytical expressions for polarizabilities of elliptic holes in a thin wall , and paper gives thickness corrections for this case. Surprisingly, the thickness factor for $`(\alpha _m+\alpha _e)`$ exhibits only a weak dependence on ellipse eccentricity $`\epsilon `$, changing its limiting value for the thick wall from 0.56 for $`\epsilon =0`$ to 0.59 for $`\epsilon =0.99`$.
For a longitudinal slot of length $`l`$ and width $`w`$, $`w/l1`$, in a thin wall, useful approximations have been obtained : for a rectangular slot
$$\alpha _m+\alpha _e=w^3(0.18140.0344w/l);$$
and for a rounded end slot
$$\alpha _m+\alpha _e=w^3(0.13340.0500w/l);$$
substituting of which into Eqs. (40)-(41) gives the impedances of slots. Figure 1 compares impedances for different shapes of pumping holes.
### 5.2 Annular Cut
The polarizabilities of a ring-shaped cut in the wall of an arbitrary thickness have been calculated in . Such an aperture can serve as an approximation for a electrode of a button-type beam position monitor, for a thin wall, or a model of a coax attached to the vacuum chamber, when the wall thickness is large. If $`a`$ and $`b`$ denote the inner and outer radii of the annular cut, $`abR`$, the magnetic susceptibility of a narrow ($`w=bab`$) annular slot in a thin plate is
$$\psi \frac{\pi ^2b^2a}{\mathrm{ln}(32b/w)2}.$$
(43)
For a narrow annular gap in the thick wall the asymptotic behavior is $`\psi 2\pi b^2w`$. The approximation (43) works well for narrow gaps, $`w/b0.15`$, while the thick wall result is good only for $`w/b0.05`$.
Analytical results for the electric polarizability of a narrow annular cut are: for a thin wall
$$\chi \pi ^2w^2(b+a)/8,$$
(44)
and for a thick wall
$$\chi w^2(b+a).$$
(45)
These estimates work amazingly well even for very wide gaps, up to $`w/b0.85`$. The electric polarizability depends on the wall thickness rather weakly.
The difference $`(\psi \chi )/b^3=2(\alpha _m+\alpha _e)/b^3`$ for an annular cut, calculated by variational methods in , is plotted in Fig. 2 for a few values of the wall thickness.
### 5.3 Protrusions
For a protrusion inside the beam pipe having the shape of a half ellipsoid with semiaxis $`a`$ in the longitudinal direction (along the chamber axis), $`b`$ in the radial direction, and $`c`$ in the azimuthal one, with $`a,b,cR`$, the polarizabilities are
$$\alpha _e=\frac{2\pi abc}{3I_b},$$
(46)
and
$$\alpha _m=\frac{2\pi abc}{3(I_c1)},$$
(47)
where
$$I_b=\frac{abc}{2}_0^{\mathrm{}}\frac{ds}{(s+b^2)^{3/2}(s+a^2)^{1/2}(s+c^2)^{1/2}},$$
(48)
and $`I_c`$ is given by Eq. (48) with $`b`$ and $`c`$ interchanged.
In the particular case $`a=c`$, $`b=h`$ we have an ellipsoid of revolution, and the polarizabilities are expressed in terms of the hypergeometric function $`{}_{2}{}^{}F_{1}^{}`$:
$$\alpha _e=\frac{2\pi a^2h}{{}_{2}{}^{}F_{1}^{}(1,1;5/2;1h^2/a^2)},$$
(49)
and
$$\alpha _m=\frac{2\pi a^2h}{{}_{2}{}^{}F_{1}^{}(1,1;5/2;1a^2/h^2)3}.$$
(50)
#### 5.3.1 Post
In the limit $`a=ch`$, corresponding to a pinlike obstacle, we get a simple expression for the inductive impedance of a narrow pin (post) of height $`h`$ and radius $`a`$, protruding radially into the beam pipe:<sup>1</sup><sup>1</sup>1One could use the known result for the induced electric dipole of a narrow cylinder parallel to the electric field . It will only change $`\mathrm{ln}(2h/a)1`$ in Eq. (51) to $`\mathrm{ln}(4h/a)7/3`$.
$$Z(k)ikZ_0\frac{h^3}{6\pi R^2\left(\mathrm{ln}(2h/a)1\right)}.$$
(51)
The factor $`F(\alpha _e+\alpha _m)/V`$, where $`V=2\pi a^2h/3`$ is the volume occupied by the obstacle, is plotted in Fig. 3 versus the ratio $`h/a`$. The figure also shows comparison with the asymptotic approximation given by Eq. (51).
#### 5.3.2 Mask
One more particular case of interest is $`h=a`$, i.e. a semispherical obstacle of radius $`a`$. From Eqs. (49)-(50) the impedance of such a discontinuity is
$$Z(k)=ikZ_0\frac{a^3}{4\pi R^2},$$
(52)
which is $`3\pi /2`$ times that for a circular hole of the same radius in a thin wall, cf. Eq. (42).
Another useful result that can be derived from the general solution, Eqs. (46)-(48), is the impedance of a mask intended to intercept synchrotron radiation. We put $`b=c=h`$, so that our model mask has the semicircular shape with radius $`h`$ in its largest transverse cross section. Then the integral in Eq. (48) is reduced to
$$I_b=I_c=\frac{1}{3}{}_{2}{}^{}F_{1}^{}(1,\frac{1}{2};\frac{5}{2};1\frac{h^2}{a^2}),$$
and we can further simplify the result for two particular cases.
The first one is the thin mask, $`ah`$, in which case $`\alpha _e8h^3/3`$ , and again it dominates the magnetic term, $`\alpha _mV=2\pi ah^2/3`$. The coupling impedance for such an obstacle โ a half disk of radius $`h`$ and thickness $`2a`$, $`ah`$, transverse to the chamber axis โ is therefore
$$Z(k)=ikZ_0\frac{2h^3}{3\pi ^2R^2}\left[1+\left(\frac{4}{\pi }\frac{\pi }{4}\right)\frac{a}{h}+\mathrm{}\right],$$
(53)
where the next-to-leading term is shown explicitly.
In the opposite limit, $`ha`$, which corresponds to a long (along the beam) mask, the leading terms $`\alpha _e\alpha _m4\pi ah^2/3`$ cancel each other. As a result, the impedance of a long mask with length $`l=2a`$ and height $`h`$, $`hl`$, is
$$Z(k)ikZ_0\frac{4h^4}{3\pi R^2l}\left(\mathrm{ln}\frac{l}{h}1\right),$$
(54)
which is relatively small due to the โaerodynamicโ shape of this obstacle, in complete analogy with results for long elliptic slots .
Figure 4 shows the impedance of a mask with a given semicircular transverse cross section of radius $`h`$ versus its normalized half length, $`a/h`$. The comparison with the asymptotic approximations Eqs. (53) and (54) is also shown. One can see, that the asymptotic behavior (54) starts to work well only for very long masks, namely, when $`l=2a10h`$. Figure 4 demonstrates that the mask impedance depends rather weakly on the length. Even a very thin mask ($`ah`$) contributes as much as $`8/(3\pi )0.85`$ times the semisphere ($`a=h`$) impedance, Eq. (52), while for long masks the impedance decreases slowly: at $`l/h=20`$, it is still 0.54 of that for the semisphere.
In practice, however, the mask has usually an abrupt cut toward the incident synchrotron radiation, so that it is rather one-half of a long mask. From considerations above one can suggest as a reasonable impedance estimate for such a discontinuity the half sum of the impedances given by Eqs. (53) and (54). This estimate is corroborated by 3D numerical simulations using the MAFIA code, at least, for the masks which are not too long. In fact, the low-frequency impedances of a semisphere and a half semisphere of the same depth โ which can be considered as a relatively short realistic mask โ were found numerically to be almost equal (within the errors), and close to that for a longer half mask. From these results one can conclude that a good estimate for the mask impedance is given simply by Eq. (52).
### 5.4 Axisymmetric Discontinuities
Following a similar procedure one can also easily obtain the results for axisymmetric irises having a semi-elliptic profile in the longitudinal chamber cross section, with depth $`b=h`$ and length $`2a`$ along the beam. For that purpose, one should consider limit $`c\mathrm{}`$ in Eq. (48) to calculate the effective polarizabilities $`\stackrel{~}{\alpha }_e`$ and $`\stackrel{~}{\alpha }_m`$ per unit length of the circumference of the chamber transverse cross section. The broad-band impedances of axisymmetric discontinuities have been studied in , and the longitudinal coupling impedance is given by
$$Z(k)=ikZ_0\frac{\stackrel{~}{\alpha }_e+\stackrel{~}{\alpha }_m}{2\pi R},$$
(55)
quite similar to Eq. (40). As $`c\mathrm{}`$, the integral $`I_c0`$, and $`I_b`$ is expressed in elementary functions as
$$I_b=\frac{1}{2}{}_{2}{}^{}F_{1}^{}(1,\frac{1}{2};2;1\frac{h^2}{a^2})=\frac{a}{a+h}.$$
It gives us immediately
$$\stackrel{~}{\alpha }_e=\frac{\pi }{2}h(h+a);\stackrel{~}{\alpha }_m=\frac{\pi }{2}ah,$$
(56)
and the resulting impedance of the iris of depth $`h`$ with the semielliptic profile is simply
$$Z(k)=ikZ_0\frac{h^2}{4R},$$
(57)
which proves to be independent of the iris thickness $`a`$. The same result has been obtained using another method , and also directly by conformal mapping in , following the general method of .
Using a conformal mapping, one can readily obtain an answer also for irises having the profile shaped as a circle segment with the chord of length $`s`$ along the chamber wall in the longitudinal direction, and opening angle $`2\phi `$, where $`0\phi \pi `$. The impedance of such an exotic iris, expressed in terms of its height $`h=s(1\mathrm{cos}\phi )/(2\mathrm{sin}\phi )`$:
$`Z(k)`$ $`=`$ $`ikZ_0{\displaystyle \frac{h^2}{2R(1\mathrm{cos}\phi )^2}}\times `$
$`\times `$ $`\left[{\displaystyle \frac{\phi (2\pi \phi )}{3(\pi \phi )^2}}\mathrm{sin}^2\phi {\displaystyle \frac{2\phi \mathrm{sin}2\phi }{2\pi }}\right].`$
Again, the impedance is proportional to $`h^2`$, but the coefficient now depends (in fact, rather weakly) on $`\phi `$.
A few useful results for low-frequency impedances of axisymmetric cavities and irises with a rectangular, trapezoidal and triangular transverse profile have been obtained in using conformal mapping to calculate the electric polarizability.
The low-frequency impedance of the small short pill-box whose length $`g`$ is not large than depth $`h`$ is
$$Z(\omega )=ikZ_0\frac{1}{2\pi R}\left(gh\frac{g^2}{2\pi }\right),$$
(59)
The low-frequency impedance of the shallow enlargement takes the form
$$Z(\omega )=ikZ_0\frac{h^2}{2\pi ^2R}\left(2\mathrm{ln}(2\pi g/h)+1\right),$$
(60)
where $`gh`$, but still less than $`R`$.
The low-frequency impedance of a small step of depth $`hR`$ is
$$Z(\omega )=ikZ_0\frac{h^2}{4\pi ^2R}\left(2\mathrm{ln}(2\pi R/h)+1\right).$$
(61)
The inductance produced by the transition with the slope angle $`\theta =\pi \nu `$ has the form:
$$Z(\omega )=i\frac{Z_0kh^2}{2\pi ^2R}\left\{\mathrm{ln}\left[\pi \nu \left(\frac{b}{h}2\mathrm{cot}\pi \nu \right)\right]+\frac{3}{2}\gamma \psi (\nu )\frac{\pi }{2}\mathrm{cot}\pi \nu \frac{1}{2\nu }\right\},$$
(62)
where $`\gamma =0.5772\mathrm{}`$ is Eulerโs constant, $`\psi (\nu )`$ is the psi-function and the transition is assumed to be short compared to the chamber radius, i.e. transition length $`l=h\mathrm{cot}\pi \nu R`$. The ratio of this inductance to that of the abrupt step ($`\nu =1/2`$, Eq. (61)) with the same height is plotted in Fig. 5 as a function of the slope angle.
The impedance of a thin (or deep) iris, $`gh`$, has the form
$$Z(\omega )=ikZ_0\frac{1}{4R}\left[h^2+\frac{gh}{\pi }\left(\mathrm{ln}(8\pi g/h)3\right)\right].$$
(63)
This formula works well even for rather large $`h`$, when $`h`$ is close to $`R`$.
More generally, the low-frequency impedance of the iris having a rectangular profile with an arbitrary aspect ratio is
$$Z(\omega )=ikZ_0\frac{gh}{2\pi R}F\left(\frac{h}{g}\right),$$
(64)
where function F(x) is plotted in Fig. 6.
The impedances of discontinuities having a triangle-shaped cross section with height (depth) $`h`$ and base $`g`$ along the beam are given below. When $`gh`$, the low-frequency impedance of a triangular enlargement is
$$Z(\omega )=ikZ_0\frac{1}{4\pi R}\left(gh\frac{g^2}{\pi }\right),$$
(65)
and that of a triangular iris is
$$Z(\omega )=ikZ_0\frac{1}{4R}\left[h^2+\frac{2gh}{\pi }(1\mathrm{ln}2)\right].$$
(66)
For the case of shallow triangular perturbations, $`hg<R`$, both the enlargement and contraction of the chamber have the same inductance,
$$Z(\omega )=ikZ_0\frac{h^22\mathrm{ln}2}{\pi ^2R},$$
(67)
which is independent of $`g`$.
## Appendix
### Circular Chamber
For a circular cross section of radius $`b`$ the eigenvalues $`k_{nm}=\mu _{nm}/b`$, where $`\mu _{nm}`$ is $`m`$th zero of the Bessel function $`J_n(x)`$, and the normalized EFs are
$$e_{nm}(r,\phi )=\frac{J_n(k_{nm}r)}{\sqrt{N_{nm}^E}}\left\{\begin{array}{c}\mathrm{cos}n\phi \\ \mathrm{sin}n\phi \end{array}\right\},$$
(68)
with $`N_{nm}^E=\pi b^2ฯต_nJ_{n+1}^2(\mu _{nm})/2`$, where $`ฯต_0=2`$ and $`ฯต_n=1`$ for $`n0`$. For TE-modes, $`k_{nm}^{}=\mu _{nm}^{}/b`$ with $`J_n^{}(\mu _{nm}^{})=0`$, and
$$h_{nm}(r,\phi )=\frac{J_n(k_{nm}^{}r)}{\sqrt{N_{nm}^H}}\left\{\begin{array}{c}\mathrm{cos}n\phi \\ \mathrm{sin}n\phi \end{array}\right\},$$
(69)
where $`N_{nm}^H=\pi b^2ฯต_n(1n^2/\mu _{nm}^2)J_n^2(\mu _{nm}^{})/2`$. In this case $`\stackrel{~}{e}_\nu =1/(2\pi b)`$, which also follows from the Gauss law, and the formula for the inductive impedance takes an especially simple form, cf. .
### Rectangular Chamber
For a rectangular chamber of width $`a`$ and height $`b`$ the eigenvalues are $`k_{nm}=\pi \sqrt{n^2/a^2+m^2/b^2}`$ with $`n,m=1,2,\mathrm{}`$, and the normalized EFs are
$$e_{nm}(x,y)=\frac{2}{\sqrt{ab}}\mathrm{sin}\frac{\pi nx}{a}\mathrm{sin}\frac{\pi my}{b},$$
(70)
with $`0xa`$ and $`0yb`$. Let a hole be located in the side wall at $`x=a,y=y_h`$. From Eq. (3) after some algebra follows
$$\stackrel{~}{e}_\nu =\frac{1}{b}\mathrm{\Sigma }(\frac{a}{b},\frac{y_h}{b}),$$
(71)
where
$$\mathrm{\Sigma }(u,v)=\underset{l=0}{\overset{\mathrm{}}{}}\frac{(1)^l\mathrm{sin}[\pi (2l+1)v]}{\mathrm{cosh}[\pi (2l+1)u/2]}$$
(72)
is a fast converging series; the behavior of $`\mathrm{\Sigma }(u,v)`$ versus $`v`$ for different values of the aspect ratio $`u`$ is plotted in Ref. .
## 6 Summary
A review of calculating the beam coupling impedances of small discontinuities was presented. We also collected some analytical formulas for the inductive contributions due to various small obstacles to the beam coupling impedances of the vacuum chamber.
An importance of understanding these effects can be illustrated by the following example. An original design of the beam liner for the LHC vacuum chamber anticipated a circular liner with many circular holes of 2-mm radius, providing the pumping area about 5% of the liner surface. Their total contribution to the low-frequency coupling impedance was calculated to be
$$|Z/n|=0.53\mathrm{\Omega },|Z_{}|=20M\mathrm{\Omega }/m,$$
which was close or above (for $`Z_{}`$) the estimated instability threshold. A modified liner design had about the same pumping area provided by rounded-end slots $`1.5\times 6`$ mm<sup>2</sup>, which were placed near the corners of the rounded-square cross section of the liner . As a result of these changes, the coupling impedances were reduced by more than an order of magnitude, 30-50 times:
$$|Z/n|=0.017\mathrm{\Omega },|Z_{}|=0.4M\mathrm{\Omega }/m.$$
Now the pumping slots are not among the major contributors to the impedance budget of the machine.
One should mention that these notes do not include more recent developments, in particular, results for coaxial structures, frequency corrections for polarizabilities, etc. Some of these new results and proper references can be found in . |
warning/0001/astro-ph0001100.html | ar5iv | text | # Spectral Energy Distributions for Disk and Halo MโDwarfs
## 1 Introduction
Until very recently the observational data for lowโmass stars could not be well reproduced by synthetic spectra or photometry. The cool, highโpressure atmospheres are difficult to model, due in particular to complex opacity sources: strong molecular bands and, for the halo stars and very lowโmass objects, pressureโinduced molecular hydrogen opacity (see e.g. Borysow, Jรธrgensen & Zheng (1997)). The situation is now much improved as demonstrated by for example the โNextGenโ models of Hauschildt, Allard & Baron (1999) (see also Allard et al. (1997)). The observational side of the study of lowโmass stars has also changed remarkably with a large increase in the known number of lowโmass stars, brown dwarfs, and even giant planets.
For the last several years we have been obtaining infrared spectra for a sample of halo and disk stars approaching and even below the stellar/subโstellar boundary. This work extends the similar spectroscopic study presented by Leggett et al. (1996) to more metalโpoor and lower mass regimes, and builds on the photometric study published recently by us (Leggett et al. (1998)). The spectra are compared to synthetic spectra generated from an improved version of Allard & Hauschildtโs NextGen model atmospheres (Hauschildt, Allard & Baron (1999)). In this paper we will present the results for the hotter stars in the sample, those with effective temperature higher than 2500 K, where neither grain condensation or extinction is significant. A subsequent paper will present our results for the cooler objects.
In ยง2 we describe the various instruments that have been used for this project. The target sample is described in ยง3 where we illustrate the likely range of metallicity and effective temperature through colorโcolor and magnitudeโcolor diagrams. ยง4 gives the observational results in the form of selected sets of spectra as well as integrated fluxes and bolometric magnitudes. ยง5 gives a brief description of the models and the comparison process. The results of the comparison of the data to the synthetic spectra are presented in ยง6 and our conclusions given in ยง7.
## 2 Instrumentation
We have been obtaining 1โ2.5$`\mu `$m spectra of low mass stars for the last few years, where the targets have been selected to sample a broad range of temperature and metallicity. Here we present the results for forty-two of the stars, those with spectral type earlier than M7, and effective temperature hotter than 2500 K. The observing runs occurred in 1999 April, 1998 January, 1997 November and 1997 April using CGS4 at UKIRT on Mauna Kea, and 1994 July using KSPEC at the University of Hawaiiโs 88โinch telescope on Mauna Kea. KSPEC is a crossโdispersed spectrometer with a 1024$`\times `$1024 HgCdTe detector. CGS4 is a grating spectrometer with a 256$`\times `$256 InSb detector. Fainter stars could be observed with CGS4 than with KSPEC, however KSPEC provided spectra from 0.9$`\mu `$m to 2.4$`\mu `$m in a single exposure, compared to the four grating settings required by CGS4.
For 9 of the stars we have obtained optical spectra using spectrometers in Hawaii in 1996 April, and in Flagstaff Arizona in 1989 December and 1989 September. In Arizona we used the B & C spectrograph on the 72โinch Lowell telescope and the spectrograph on the Naval Observatory Flagstaff Stationโs 40โinch telescope. In Hawaii we used the HARIS spectrograph at the UH 88โinch telescope.
Table 1 lists the observing dates, grating and slit information, and resolutions, provided by these instruments. In the cases where we did not have our own optical spectra we have obtained spectra from the literature. These sources are described below.
Data reduction was carried out in the usual way using the IRAF and Figaro software packages. For the infrared data, the effect of the terrestrial atmosphere was removed by dividing by the spectrum of a nearby earlyโtype star, after removing hydrogen lines seen in this reference spectrum. The shape of the infrared spectrum is corrected for the known flux distribution of the earlyโtype star. The spectral segments were individually flux calibrated using the targets known IJHK photometry. Each section was integrated over the appropriate filter profile (Cousins I or UKIRT JHK); the observed flux from Vega was integrated over the same profile. Vega was assumed to be zero magnitude at all wavelengths, and the target flux was scaled to match the broadband photometry. Where we obtained optical spectra from the literature these were flux calibrated by us in the same way (using either an R or I filter as appropriate).
## 3 The Sample
### 3.1 Sample Selection
Infrared spectra were obtained for a selection of the very lowโmass stars of the halo and disk. The halo stars were selected from studies of known high properโmotion stars by Gizis (1997); Gizis & Reid (1997) and Monet et al. (1992). The disk stars are also known proper motion objects, and were selected from Leggett (1992) to sample metallicity and a range of effective temperature. The targets are listed in Table 2. We give LHS or LP number (Luyten (1979)), and/or Gliese or Gliese/Jahreiss number (Gliese & Jahreiss (1991)), and/or Giclas number (Giclas, Burnham & Thomas (1971)), for each star. An abbreviated RA/Dec is also given to aid identification. Note that LHS 421 is the wellโknown eclipsing binary CM Draconis (also known as Gliese 630.1A, Metcalfe et al. (1966)).
The spectral types in Table 2 are taken from various sources and there may be discrepancies or errors at the level of one subโclass. For the halo stars the classifications primarily are from Gizis (1997) and Gizis & Reid (1997), but for LHS 2045 and LHS 3390 they are based on our own optical spectra. For the disk stars the classifications are taken from Gizis (1997); Kirkpatrick, Henry & Simons (1995) and Kirkpatrick, Henry & Irwin (1997).
The kinematic populations have been taken from Leggett et al. (1998); Leggett (1992). For a few objects the classifications of Leggett (1992) have been updated using radial velocities from Reid, Hawley & Gizis (1995) and Dawson & de Robertis (1998). The classification schemes for young disk (YD), young/old disk (Y/O), old disk (OD), old disk/halo (O/H) and halo (H) are described in Leggett (1992).
Table 2 also lists the instrumentation used for each object using the configuration names given in Table 1. Where we did not have our own optical data available we used published spectra taken from the sources listed in the Table. The instrumental resolutions for these data are given in the notes to the Table.
### 3.2 Photometrically Implied Properties of the Sample
Table 3 lists distance moduli and VIJHKL colors for the sample, taken primarily from the compilations by Leggett (1992); Leggett et al. (1998) โ the reader is asked to refer to these papers for the data sources. V,I for LHS 425 are unpublished Naval Observatory Flagstaff Station data. New L data have been obtained by us for LHS 1174, LHS 523 and LHS 5328 in 1999 September using the reconfigured IRCAM/TUFTI camera on UKIRT at Mauna Kea. These data are presented in Table 3. As shown by Leggett et al. (1998) the NextGen models (Hauschildt, Allard & Baron (1999)) do a very good job of reproducing the observed photometry. The most recent models include a new linelist for TiO (Schwenke (1998)) which has improved the match to the energy distribution in the optical to red regime (compare our Figure 1 to Figure 1 of Leggett et al. (1998)). The new models also include a new linelist for H<sub>2</sub>O (Partridge & Schwenke (1997)) however the calculated and observed water band strengths still show some discrepancies. This is discussed further below.
Figures 1 to 4 show VIJHKL colorโcolor diagrams with model synthetic color sequences overlaid. The most metal-poor stars and coolest stars are identified. Figure 5 shows M<sub>J</sub>:J$``$K with isochrones from the structural models by Baraffe et al. (1997); Chabrier & Baraffe (1997). These use the NextGen atmospheres but not the most recent versions with the improved linelists. Nevertheless the agreement is good and we can estimate mass and metallicities for our objects from this diagram. Note that the empirical masses based on the Henry & McCarthy (1993) scale (shown on the right axis) agree well with these structural models. The more recent massโluminosity paper by Henry et al. (1999) supports their previous work.
Figures 1 through 5 imply that our sample covers a range of metallicity of about solar to about 3% of solar (m/H$`1.5`$) and the range in effective temperature is about 3800 K to 2400 K. The implied mass range is 0.3โ0.1M/M for the halo stars, 0.6โ0.09M/M for the disk stars. A grid of synthetic spectra were calculated covering this range of likely values, and more exact determinations of effective temperature and metallicity are presented later as a result of comparison to the synthetic spectra.
## 4 Observational Results
### 4.1 Spectroscopic Sequences
Figure 6 shows a representative set of spectra for approximately solar metallicity objects with a range of effective temperature and spectral type. The obvious features to note, which strengthen with decreasing temperature, are: the CO bands at 2.3$`\mu `$m; the water bands at around 1.4$`\mu `$m, 1.8$`\mu `$m and 2.4$`\mu `$m; the FeH band at 0.99$`\mu `$m; and the KI doublets at around 1.18$`\mu `$m and 1.24$`\mu `$m. A more complete list of spectral features seen in the Mโdwarfs is given in Leggett et al. (1996).
Figure 7 demonstrates the effect of metallicity at T$`{}_{eff}{}^{}3000`$ K. The shape of the infrared energy distribution for the metalโpoor stars becomes dominated by pressureโinduced H<sub>2</sub> opacity. There are no strong absorption features seen in the subdwarfs in the infrared, but in the optical the hydride features are a good indicator of a subdwarf nature โ the hydride features, especially CaH at 0.69$`\mu `$m, become very strong relative to the TiO bands.
### 4.2 Integrated Fluxes and Bolometric Corrections
Table 3 gives integrated fluxes for the sample, expressed as flux at the Earth, bolometric magnitude and intrinsic stellar luminosity. The integrated fluxes were obtained by integrating our spectroscopic data over wavelength, and adding the flux contributions at shorter and longer wavelengths. Some stars had gaps in our spectroscopic data around 1$`\mu `$m. The contributions from these regions were estimated using stars of similar spectral type with complete spectral coverage.
The flux contributions at wavelengths beyond 2.4$`\mu `$m were calculated by deriving the flux at L using an effective wavelength approach, summing the contribution from the end of the Kโband spectrum to this point with a linear interpolation, and assuming a RayleighโJeans tail beyond L. Theoretical energy distributions imply that the error in a RayleighโJeans assumption is $``$1% for this sample. For stars without L photometry, L was estimated from stars of similar J$``$K color and metallicity.
Most of our stars have spectra available starting around 0.6$`\mu `$m. For these stars the shorter wavelength flux contribution was adopted to be a simple linear extrapolation to zero flux at zero wavelength from the flux at 0.6$`\mu `$m, except for the hottest stars where the flux at B was estimated using the effective wavelength approach, and linear interpolations used from zero wavelength to B, and from B to the start of the red spectrum. For the two stars without optical spectra the contribution in this region was estimated from stars of similar temperature and metallicity. For the hotter of these two stars, LHS 5327, there is a relatively large uncertainty in the total flux due to the lack of optical data; for this star the uncertainty in total flux is 10%, leading to an uncertainty of 0.10mag in the bolometric correction and 0.05dex in log$`{}_{10}{}^{}L/L_{}`$. For the rest of our sample the uncertainties are 5%, 0.05 mag and 0.02dex, respectively.
Figure 8 plots Kโband bolometric correction against I$``$K color. We have included the results from Leggett et al. (1996). The approximate metallicities of the stars are indicated, based on kinematic population. Model sequences are overlaid as dashed lines, where again these model calculations have not been upgraded to include the new TiO or H<sub>2</sub>O linelists but still match the observations well. The metalโpoor stars are confined to small values of BC<sub>K</sub> due to the onset of pressureโinduced H<sub>2</sub> opacity, reducing the flux at K. For the disk stars the relationship between Kโband bolometric correction and I$``$K color can be well represented by the cubic polynomial:
$$BC_K=2.741+5.452(IK)1.824(IK)^2+0.211(IK)^3$$
for 1.8 $``$ I$``$K $``$ 3.3. This fit is indicated by the solid line in Figure 8.
## 5 Models and Synthetic Spectra, and Comparison Process
We have calculated the models presented in this paper using our multipurpose model atmosphere code Phoenix, version 10.7. Details of the code and the general input physics setup are discussed in the description of the NextGen grid of model atmospheres presented by Hauschildt, Allard & Baron (1999) and references therein. The model atmospheres presented here were calculated with the same general input physics as the NextGen models. However, a change of the linelists has significant impact on the model structure and synthetic spectra. The most important difference from our NextGen grid is the replacement of TiO and H<sub>2</sub>O linelists with the newer linelist calculated by the NASAโAMES group, Schwenke (1998) for TiO (about 175 million lines of 5 isotopes) and Partridge & Schwenke (1997) for H<sub>2</sub>O (about 350 million lines in 2 isotopes). Our combined molecular line list includes about 550 million molecular lines. These lines are treated with a direct opacity sampling technique where each line has its individual Voigt (for strong lines) or Gauss (weak lines) line profile, see Hauschildt, Allard & Baron (1999) and references therein for details. The number of lines selected by this procedure depends on the the model parameters. In addition to the new line data, we have also included dust formation and opacities in the models used in this paper. However, the lowest effective temperatures of the stars we consider here are slightly above the regime were dust formation and opacities are important. A complete description of the models and the differences to the NextGen models will be given in Allard & Hauschildt (1999, in preparation).
The fitting of the synthetic to the observed spectra was done using an automatic IDL program. First, the resolution of the synthetic spectra is reduced to that of each individual observed spectrum and the spectra are normalized to unit area for scaling. The comparison was done using a model atmosphere grid that covers the range 1500 K$``$ T$`{}_{eff}{}^{}`$ 4000 K, 3.5 $``$ log g $``$ 5.5 and $`1.5`$ \[m/H\] $`0.0`$, with a total of 221 model atmospheres. For each observed spectrum we then calculate the $`\chi ^2`$ value for the comparison with all synthetic spectra in the grid. In order to avoid known problematic spots in either the observations or the synthetic spectra, the wavelength ranges 1.35โ1.5$`\mu `$m and 1.8โ1.95$`\mu `$m were excluded from the comparison, however, tests showed this did not significantly change the results. We selected the models that resulted in the lowest 3โ5 $`\chi ^2`$ values as the most probable parameter range for each individual star. The โbestโ value was then chosen by visual inspection. This procedure allows a rough estimate of the probable range in the stellar parameters. Note that systematic errors due to missing or incomplete opacity sources are not eliminated, however investigations of the different linelists available for TiO and H<sub>2</sub>O indicate that differences in the treatment of these opacity sources effect the implied effective temperatures in opposite senses (Allard, Hauschildt & Schwenke (2000)) โ i.e. the systematic errors should be small.
## 6 Results of Comparison of Data and Models
The automatic comparison described in the previous section resulted in $`\chi ^2`$ values between 0.02 and 0.10, with an average value of 0.05. The best fits were inspected by eye and in some cases a match with a slightly higher $`\chi ^2`$ value than minimum was adopted. We did not try to match the bottom of the water bands, but did look at the depth of the CO and TiO bands, and tried to match the overall โcontinuumโ in the optical and infrared regimes. For the metalโpoor stars the flatness of the infrared continuum could be used to constrain the derived metallicities.
Based on visual inspection and the $`\chi ^2`$ values, the uncertainty in the derived values of effective temperature is $`\pm 100`$ K, and in metallicity (\[m/H\]) $`\pm 0.25`$ dex. Gravity could not be well constrained by data of this relatively low resolution โ spectra generated with log g values of 5.0 $`\pm 0.5`$ dex all matched the data well. Table 4 lists the derived parameters for the sample. For the two stars in common with Leggett et al. (1996) who used earlier NextGen atmospheres, the agreement is within the quoted 150 K errors for that work (for LHS 57 the temperature derived here is identical and for LHS 377 it is 150 K cooler).
Figure 9 shows comparisons of synthetic and observed spectral energy distributions for approximately solar-metallicity stars with effective temperatures of 3600 K (LHS 65), 3100 K (LHS 421) and 2600 K (LHS 523). Also shown is a more metalโpoor star with T$`{}_{eff}{}^{}=`$3200 K (LHS 3061). The fits are good except for the coolest temperatures where problems with the water opacity became apparent, and where details of grain formation and settling may become important. The strength of the FeH line at 1$`\mu `$m is also overestimated in the models. These new models have resolved the discrepancy between the infrared and optical regions seen by Viti et al. (1997) for the eclipsing binary CM Draconis (LHS 421); Figure 9 shows that the entire energy distribution is well matched by a model with T$`{}_{eff}{}^{}=`$3100 K and \[m/H\]$`=0.5`$.
Figures 10โ12 show derived T<sub>eff</sub> as a function of various colors. Symbols indicate the metallicity implied by the energy distribution comparison (given in Table 4) โ metalโpoor stars are bluer at a constant temperature, except in V$``$I at T$`{}_{eff}{}^{}2900`$K. The solid lines are modelโpredicted temperature:color relationships; the apparent offset for the infrared colors of the disk stars is probably due to the remaining problems with water opacity, which is more significant at cooler temperatures. The error in color due to absolute calibration or observational error is smaller than the apparent offset. We note that the observed trend of T<sub>eff</sub> with V$``$I and I$``$K colors agrees with that implied by our earlier work using the NextGen atmospheres (Leggett et al. (1996), Figure 17) but the agreement between observation and theory for V$``$I is much improved with the new TiO linelist.
Diameters have been derived for the stars in the sample in two ways. One was to use the scaling factors necessary to match the synthetic stellar surface spectra to that observed at the Earth, which requires use of the trigonometric parallax and which results in errors in diameter of around 10% โ or that in the parallax if that is larger. The other method was to use Stefanโs Law to derive diameter from the observed stellar luminosity at the Earth and the effective temperature. Here the largest uncertainty comes from that in T<sub>eff</sub> which enters as the fourth power; the typical uncertainty in diameter is then 13%. These two determinations give values for diameter that agree very well โ to typically 1.5%. Diameters are also given in Table 4.
Figure 13 shows diameter (determined from scaling factor) versus effective temperature with symbols indicating the metallicity implied by the energy distribution comparison. Metalโpoor stars have a smaller diameter for a constant temperature. Open symbols are known multiple systems which would have larger diameters; typical error bars as well as larger individual errors are indicated. The dashed lines are structural model predictions from Baraffe et al. (1997); Chabrier & Baraffe (1997) (which use the NextGen atmospheres without the improved linelists) for an age of 10 Gyr. The models for 1 Gyr are not significantly different. The agreement is good except for known multiple systems and for LHS 1183 (a young flare star), LHS 135, LHS 2945, LHS 5327 and LHS 549 (an old variable star), which also seem to have too large a diameter. As far as we are aware none of these are known to be multiple, and LHS 549 has been searched for close companions using speckle by Leinert et al. (1997) with a negative result. These five stars merit further study for multiplicity.
## 7 Conclusions
We have obtained 1โ2.5$`\mu `$m spectra for 42 disk and halo M1โM6.5 dwarfs. These data have been combined with new or published optical spectra, and energy distributions derived by flux calibrating and combining the individual spectral regions for each object. Bolometric luminosities have been determined and a relationship between bolometric magnitude and I$``$K color given.
The colors and energy distributions have been compared to synthetic photometry and spectroscopy generated by an upgraded version of the NextGen models, the AMES-Dusty models (Hauschildt, Allard & Baron (1999)). These models use more recent TiO and H<sub>2</sub>O linelists, and include grain condensation and extinction. The agreement is much improved in the red region compared to earlier comparisons (e.g. Leggett et al. (1996)). Problems remain with the match to the observed FeH features and also to the water bands. These problems are being addressed with ongoing work to calculate more complete linelists. Nevertheless we can determine effective temperatures for the sample to $`\pm `$100 K, metallicities to $`\pm `$0.25 dex, and radii to typically 10%. Recent structural models by Baraffe et al. (1997); Chabrier & Baraffe (1997) agree with the luminosities and radii derived except for five stars which may be previously unknown multiple systems: LHS 1183, LHS 135, LHS 2945, LHS 5327 and LHS 549.
We are very grateful to the staff at UKIRT, the University of Hawaiiโs 88โinch telescope, the Lowell Observatory and the Naval Observatory Flagstaff Station for their assistance in obtaining the data presented in this paper. UKIRT, the United Kingdom Infrared Telescope, is operated by the Joint Astronomy Centre Hilo Hawaii on behalf of the U.K. Particle Physics and Astronomy Research Council. FA acknowledges support from NASA LTSA NAG5-3435 and NASA EPSCoR grants to Wichita State University, and support from CNRS. PHH acknowledges partial support from the Pรดle Scientifique de Modรฉlisation Numรฉrique at ENS-Lyon. This work was also supported in part by NSF grants AST-9417242, AST-9731450, and NASA grant NAG5-3505. Some of the calculations presented in this paper were performed on the IBM SP and the SGI Origin 2000 of the UGA UCNS and on the IBM SP of the San Diego Supercomputer Center (SDSC), with support from the National Science Foundation, on the Cray T3E of the NERSC with support from the DoE, and on the IBM SP2 of the French Centre National Universitaire Sud de Calcul (CNUSC) and the Cray T3E of the Commissariat a lโEnergie Atomique (CEA). We thank all these institutions for a generous allocation of computer time. |
warning/0001/nlin0001046.html | ar5iv | text | # Multi-component optical solitary waves
## I Introduction
Rapid progress in the design and manufacture of optical fiber systems is a result of worldwide demand for ultra-high bit-rate optical communications. This explains the growing interest of the soliton community in soliton-based optical fiber communication systems. This area of research was considerably advanced in recent years . The most remarkable results include the application of the concept of the dispersion management to temporal optical solitons and soliton-based optical transmission systems, and the discovery of the so-called dispersion managed soliton. High-speed optical communications require effective components such as high-performance broadband computer networks that can be developed by employing the concept of the bit-parallel-wavelength (BPW) pulse transmission that offers many of the advantages of both parallel fiber ribbon cable and conventional wavelength-division-multiplexing (WDM) systems .
Expanding development in the study of the soliton fiber systems has been observed in parallel with impressive research on their spatial counterparts, optical self-trapped beams or spatial optical solitons. One of the key concepts in this field came from the theory of multi-frequency wave mixing and cascaded nonlinearities where a nonlinear phase shift is produced as a result of the parametric wave interaction . In all such systems, the nonlinear interaction between the waves of two (or more) frequencies is the major physical effect that can support coupled-mode multi-frequency solitary waves.
The examples of temporal and spatial solitons mentioned above have one common feature: they involve the study of solitary waves in multi-component nonlinear models. The main purpose of this paper is to overview several different physical examples of multi-mode and/or multi-frequency solitary waves that occur for the pulse or beam propagation in nonlinear optical fibers and waveguides. For these purposes, we select three different cases: multi-wavelength solitary waves in bit-parallel-wavelength optical fiber links, multi-colour spatial solitons due to multistep cascading in optical waveguides with quadratic nonlinearities, and quasiperiodic solitons in the Fibonacci superlattices. We believe these examples display both the diversity and richness of the multi-mode soliton systems, and they will allow further progress to be made in the study of nonlinear waves in multi-component nonintegrable physical models.
## II Temporal and spatial solitons
Because the phenomenon of the long-distance propagation of temporal optical solitons in optical fibers is known to a much broader community of researchers in optics and nonlinear physics, first we emphasize the difference between temporal and spatial solitary waves. Indeed, for a long time stationary beam propagation in planar waveguides has been considered somewhat similar to the pulse propagation in fibers. This approach is based on the so-called spatio-temporal analogy in wave propagation, meaning that the propagation coordinate $`z`$ is treated as the evolution variable and the spatial beam profile along the transverse direction in waveguides, is similar to the temporal pulse profile in fibers. This analogy is based on a simple notion that both beam evolution and pulse propagation can be described by the cubic nonlinear Schrรถdinger (NLS) equation.
However, contrary to the widely accepted opinion, there is a crucial difference between temporal and spatial solitons. Indeed, in the case of the nonstationary pulse propagation in fibers, the operation wavelength is usually selected near the zero point of the group-velocity dispersion. This means that the absolute value of the fiber dispersion is small enough to be compensated by a weak nonlinearity such as that produced by the (very weak) Kerr effect in optical fibers which leads to a relative nonlinearity-induced change in the refractive index. Therefore, nonlinearity in such systems is always weak and it should be well modeled by a cubic NLS equation which is known to be integrable by means of the inverse-scattering technique. However, for very short (e.g., fs) pulses the cubic NLS equation describing the long-distance propagation of pulses should be corrected to include additional terms that would account for such effects as higher-order dispersion, Raman scattering, etc. All such corrections can be taken into account with the help of the perturbation theory . Thus, in fibers nonlinear effects are weak and they become important only when dispersion is small (near the zero-dispersion point) affecting the pulse propagation over large distances (of order of hundreds of meters or even kilometers). The situation changes dramatically when we consider the propagation of multi-wavelength pulses with almost equal group velocities. The corresponding model is described by a nonintegrable and rather complicated system of coupled NLS equations, which we briefly discuss below.
In contrary to the pulse propagation in optical fibers, the physics underlying the stationary beam propagation in planar waveguides and bulk media is different. In this case an optical beam is generated by a continuous wave (CW) source and it is time independent. However, when the beam evolves with the propagation distance $`z`$, it diffracts in the transverse spatial directions. Then, a nonlinear change in the refractive index should compensate for the beam spreading caused by diffraction which is not a small effect. That is why to observe spatial solitons as self-trapped optical beams, much larger nonlinearities are usually required, and very often such nonlinearities are not of the Kerr type (e.g. they saturate at higher intensities). This leads to the models of generalized nonlinearities with the properties of solitary waves different from those described by the integrable cubic NLS equation. Propagation distances involved in the phenomenon of the beam self-focusing and spatial soliton propagation are of the order of millimeters or centimeters. To achieve such large nonlinearities, one needs to use the optical materials with large nonlinearity-induced refractive index. One of the possible way to overcome this difficulty is to use the so-called cascaded nonlinearities of noncentrosymmetric optical materials where nonlinear effects are accumulated due to parametric wave interaction under the condition of the wave phase matching. Such parametric wave-mixing effects generate novel classes of spatial optical solitons where resonant parametric coupling between the envelopes of two (or more) beams of different frequencies supports stable spatially localised waves even in a bulk medium (see details in Ref. ). It is this kind of multi-component solitary waves that we discuss below.
## III BIT-PARALLEL-WAVELENGTH SOLITONS
A growing demand for high-speed computer communications requires an effective and inexpensive computer interconnection. One attractive alternative to the conventional WDM systems is BPW single-mode fiber optics links for very high bandwidth computer communications . They differ from the WDM schemes in that no parallel to serial conversion is necessary, and parallel pulses are launched simultaneously on different wavelengths.
When the pulses of different wavelengths are transmitted simultaneously, the cross-phase modulation can produce an interesting pulse shepherding effect , when a strong (โshepherdโ) pulse enables the manipulation and control of pulses co-propagating on different wavelengths in a multi-channel optical fiber link.
To describe the simultaneous transmission of $`N`$ different wavelengths in a nonlinear optical fiber, we follow the standard derivation and obtain a system of $`N`$ coupled nonlinear Schrรถdinger (NLS) equations $`(0jN1`$):
$$\begin{array}{c}i\frac{A_j}{z}+iv_{gj}^1\frac{A_j}{t}\frac{\beta _{2j}}{2}\frac{^2A_j}{t^2}\hfill \\ +\chi _j\left(|A_j|^2+2\underset{mj}{}|A_m|^2\right)A_j=0,\hfill \end{array}$$
(1)
where, for the $`j`$th wave, $`A_j(z,t)`$ is the slowly varying envelope, $`v_{gj}`$ and $`\beta _{2j}`$ are the group velocity and group-velocity dispersion, respectively, and the nonlinear coefficients $`\chi _j`$ characterize the Kerr effect. Equations (1) do not include the fiber loss, since the fiber lengths involved in bit-parallel links are only a small fraction of the attenuation length.
We measure the variables in the units of the central wavelength channel (say, $`j=0`$), and obtain the following normalized system of the $`N`$ coupled NLS equations,
$$\begin{array}{c}i\frac{u_j}{z}+\frac{1}{2}\alpha _j\frac{^2u_j}{t^2}\hfill \\ +\gamma _j\left(|u_j|^2+2\underset{mj}{}|u_m|^2\right)u_j=0,\hfill \end{array}$$
(2)
where $`u_j=A_j/\sqrt{P_0}`$, $`P_0`$ is the incident optical power in the central channel, $`\alpha _j(\beta _{2j}/|\beta _{20}|)`$, $`\gamma _j\chi _j/\chi _0`$, so that $`\alpha _0=\gamma _0=1`$. For the operating wavelengths spaced $`1`$nm apart within the band $`1530รท1560`$ nm, the coefficients $`\alpha _j`$ and $`\gamma _j`$ are different but close to $`1`$. Initially, in Eq. (2), we omit the mode walk-off effect described by the parameters $`\delta _j=v_{g0}^1v_{gj}^1`$ (so that $`\delta _0=0`$). This effect will be analysed later in this section.
To analyze the nonlinear modes, i.e. localized states of the BPW model (2), we look for stationary solutions in the form,
$$u_j(z,t)=u_j(t)e^{i\beta _jz},$$
(3)
and therefore obtain the system of equations for the normalized mode amplitudes,
$$\begin{array}{c}\hfill \frac{1}{2}\frac{d^2u_0}{dt^2}+\left(|u_0|^2+2\underset{n=1}{\overset{N1}{}}|u_n|^2\right)u_0=\frac{1}{2}u_0,\\ \hfill \frac{1}{2}\alpha _n\frac{d^2u_n}{dt^2}+\gamma _n\left(|u_n|^2+2\underset{mn}{}|u_m|^2\right)u_n=\lambda _nu_n,\end{array}$$
(4)
where $`n=1,2,\mathrm{},N1`$, the amplitudes and time are measured in the units of $`\sqrt{2\beta _0}`$ and $`(2\beta _0)^{1/2}`$, respectively, and $`\lambda _n=\beta _n/2\beta _0`$.
System (4) has exact analytical solutions for $`N`$ coupled components, the so-called BPW solitons. Indeed, looking for solutions in the form $`u_0(t)=U_0\mathrm{sech}t`$, $`u_n(t)=U_n\mathrm{sech}t`$, we obtain the constraint $`\lambda _n=\alpha _n/2`$, and a system of $`N`$ coupled algebraic equations for the wave amplitudes,
$`U_0^2+2{\displaystyle \underset{n=1}{\overset{N1}{}}}U_n^2=1,U_n^2+2{\displaystyle \underset{mn}{}}U_m^2=\alpha _n/\gamma _n.`$
In a special symmetric case, we take $`\alpha _n=\gamma _n=1`$, and the solution of those equations is simple :
$`U_0=U_nU_{}=[1+2(N1)]^{1/2}`$.
Analytical solutions can also be obtained in the linear limit, when the central frequency pulse (at $`n=0`$) is large. Then, linearizing Eqs. (4) for small $`|u_n||u_0|`$, we obtain a decoupled nonlinear equation for $`u_0`$ and $`N1`$ decoupled linear equations for $`u_n`$. Each of the latter possess a localized solution provided $`\lambda _n=\lambda _n^{(0)}`$, where $`\lambda _n^{(0)}=(\alpha _n/8)[1\sqrt{1+16(\gamma _n/\alpha _n)}]^2`$. In this limit the central soliton pulse $`u_0`$ (โshepherd pulseโ) can be considered as inducing an effective waveguide that supports a fundamental mode $`u_n`$ with the corresponding cutoff $`\lambda _n^{(0)}`$. Since, by definition, the parameters $`\alpha _n`$ and $`\gamma _n`$ are close to $`1`$, we can verify that the soliton-induced waveguide supports maximum of two modes (fundamental and the first excited one). This is an important physical result that explains the effective robustness of the pulse guidance by the shepherding pulse.
To demonstrate a number of unique properties of the multi-channel BPW solitons, we consider the case $`N=4`$ in more details. A comprehensive discussion of the case $`N=2`$ can be found in the preprint . We select the following set of the normalized parameters: $`\alpha _{0,3}=\gamma _{0,3}=1`$, $`\alpha _1=\gamma _1=1.1`$, and $`\alpha _2=\gamma _2=1.05`$. Solitary waves of this four-wavelength BPW system can be found numerically as localized solutions of Eqs. (4). Figure 1 presents the lowest-order families of such localized solutions. In general, they are characterized by $`N1`$ parameters, but we can capture the characteristic features by presenting power dependencies along the line $`\lambda \lambda _1=\lambda _2=\mathrm{}`$ in the parameter space $`\{\lambda _n\}`$. The power of the central-wavelength component ($`n=0`$) does not depend on $`\lambda `$ (straight line $`P_0=2`$). Thin dashed, dotted, and dash-dotted curves correspond to the three separate single-mode solitons of the multi-channel BPW system, (1), (2), and (3), respectively, shown with the corresponding branches of (0+1), (0+2), and (0+3) two-mode solitons. The latter curves start off from the bifurcation points on the $`u_0`$ branch at $`\lambda _1^{(0)}`$, $`\lambda _2^{(0)}`$, and $`\lambda _3^{(0)}`$, respectively. Close separation of those curves is the result of closeness of the parameters $`\alpha _n`$ and $`\gamma _n`$ for $`n=1,2,3`$.
Thick solid curves in Fig. 1 correspond to the two- (1+2) and three-mode (0+1+2) localized solutions. The latter solutions bifurcate and give birth to four-wavelength solitons (0+1+2+3) (branch A-B). Two examples of such four-wave composite solitons are shown in Fig. 1 (bottom row). The solution B is close to an exact sech-type solution at $`\lambda =0.5`$ (described above) for $`N=4`$, whereas the solution A is close to that approximately described in the linear limit in the vicinity of a bifurcation point. Importantly, for different values of the parameters $`(\alpha _n,\gamma _n)`$, the uppermost bifurcation point for this branch (open circle in Fig. 1) is not predicted by a simple linear theory and, due to the nonlinear mode coupling, it gets shifted from the branch of the central-wavelength soliton (straight line) to a two-mode branch (0+1+2) (thick solid curve).
As a result, if we start on the right end of the horizontal branch and follow the lowest branches of the total power $`P(\lambda )`$ in Fig. 1, we pass the following sequence of the soliton families and bifurcation points: $`(0)(0+1)(0+1+2)(0+1+2+3)(1+2+3)(2+3)(3)`$. If the modal parameters are selected closer to each other, the first two links of the bifurcation cascade disappear (i.e. the corresponding bifurcation points merge), and the four-mode soliton bifurcates directly from the central-wavelength pulse, as predicted by the linear theory. Note however that the sequence and location of the bifurcation points is a function of the cross-section of the parameter space $`\{\lambda _n\}`$, and the results presented above correspond to the choice $`\lambda =\lambda _1=\lambda _2=\mathrm{}`$.
The qualitative picture of the cascading bifurcations preserves for other values of $`N`$. In particular, near the bifurcation point a mixed-mode soliton corresponds to the localized modes guided by the central-wavelength soliton (shepherd) pulse. The existence of such soliton solutions is a key concept of BPW transmission when the data are launched in parallel carrying a desirable set of bits of information, all guided by the shepherd pulse at a selected wavelength.
Effects of the walk-off on the multi-channel BPW solitons seems to be most dangerous for the pulse alignment in the parallel links. For nearly equal soliton components, it was shown long time ago that nonlinearity can provide an effective trapping mechanism to keep the pulses together. For the shepherding effect, the corresponding numerical simulations are presented in Figs. 2(a-d) for the four-channel BPW system. Initially, we launch a composite four-mode soliton as an unperturbed solution A \[see Fig. 1\] of Eqs. (2), without walk-off and centered at $`t=0`$. When this solution evolves along the propagation direction $`z`$ in the presence of small to moderate relative walk-off ($`\delta _n0`$ for $`n0`$), its components remain strongly localized and mutually trapped \[Fig. 2(a,b)\], whereas it loses some energy into radiation for much larger values of the relative mode walk-off \[Fig. 2(c,d)\].
## IV PARAMETRIC OPTICAL SOLITONS DUE TO MULTISTEP CASCADING
### A Concept of multistep cascading
Recent progress in the study of cascading effects in optical materials with quadratic (second-order or $`\chi ^{(2)}`$) nonlinear response has offered new opportunities for all-optical processing, optical communications, and optical solitons . Most of the studies of cascading effects employ parametric wave mixing processes with a single phase-matching and, as a result, two-step cascading . For example, the two-step cascading associated with type I second-harmonic generation (SHG) includes the generation of the second harmonic ($`\omega +\omega =2\omega `$) followed by reconstruction of the fundamental wave through the down-conversion frequency mixing (DFM) process ($`2\omega \omega =\omega `$). These two processes are governed by one phase-matched interaction and they differ only in the direction of power conversion.
The idea to explore more than one simultaneous nearly phase-matched process, or double-phase-matched (DPM) wave interaction, became attractive only recently , for the purposes of all-optical transistors, enhanced nonlinearity-induced phase shifts, and polarization switching. In particular, it was shown that multistep cascading can be achieved by two second-order nonlinear cascading processes, SHG and sum-frequency mixing (SFM), and these two processes can also support a novel class of multi-colour parametric solitons , briefly discussed below.
### B Multistep cascading solitons
To introduce the simplest model of multistep cascading, we consider the fundamental beam with frequency $`\omega `$ entering a noncentrosymmetric nonlinear medium with a quadratic response. As a first step, the second-harmonic wave with frequency $`2\omega `$ is generated via the SHG process. As a second step, we expect the generation of higher order harmonics due to SFM, for example, a third harmonic ($`\omega +2\omega =3\omega `$) or even fourth harmonic ($`2\omega +2\omega =4\omega `$. When both such processes are nearly phase matched, they can lead, via down-conversion, to a large nonlinear phase shift of the fundamental wave . Additionally, the multistep cascading can support a novel type of three-wave spatial solitary waves in a diffractive $`\chi ^{(2)}`$ nonlinear medium, multistep cascading solitons.
We start our analysis with the reduced amplitude equations derived in the slowly varying envelope approximation with the assumption of zero absorption of all interacting waves (see, e.g., Ref. ). Introducing the effect of diffraction in a slab waveguide geometry, we obtain
$$\begin{array}{c}2ik_1\frac{A_1}{z}+\frac{^2A_1}{x^2}+\chi _1A_3A_2^{}e^{i\mathrm{\Delta }k_3z}\hfill \\ +\chi _2A_2A_1^{}e^{i\mathrm{\Delta }k_2z}=0,\hfill \\ 4ik_1\frac{A_2}{z}+\frac{^2A_2}{x^2}+\chi _4A_3A_1^{}e^{i\mathrm{\Delta }k_3z}\hfill \\ +\chi _5A_1^2e^{i\mathrm{\Delta }k_2z}=0,\hfill \\ 6ik_1\frac{A_3}{z}+\frac{^2A_3}{x^2}+\chi _3A_2A_1e^{i\mathrm{\Delta }k_3z}=0,\hfill \end{array}$$
(5)
where $`\chi _{1,2}=2k_1\sigma _{1,2}`$, $`\chi _3=6k_1\sigma _3`$, and $`\chi _{4,5}=4k_1\sigma _{4,5}`$, and the nonlinear coupling coefficients $`\sigma _k`$ are proportional to the elements of the second-order susceptibility tensor which we assume to satisfy the following relations (no dispersion), $`\sigma _3=3\sigma _1`$, $`\sigma _2=\sigma _5`$, and $`\sigma _4=2\sigma _1`$.
In Eqs. (5), $`A_1`$,$`A_2`$ and $`A_3`$ are the complex electric field envelopes of the fundamental harmonic (FH), second harmonic (SH), and third harmonic (TH), respectively, $`\mathrm{\Delta }k_2=2k_1k_2`$ is the wavevector mismatch for the SHG process, and $`\mathrm{\Delta }k_3=k_1+k_2k_3`$ is the wavevector mismatch for the SFM process. The subscripts โ1โ denote the FH wave, the subscripts โ2โ denote the SH wave, and the subscripts โ3โ, the TH wave. Following the technique earlier employed in Refs. , we look for stationary solutions of Eq. (5) and introduce the normalised envelope $`w(z,x)`$, $`v(z,x)`$, and $`u(z,x)`$ according to the relations,
$$\begin{array}{c}A_1=\frac{\sqrt{2}\beta k_1}{\sqrt{\chi _2\chi _5}}e^{i\beta z}w,\hfill \\ A_2=\frac{2\beta k_1}{\chi _2}e^{2i\beta z+i\mathrm{\Delta }k_2z}v,\hfill \\ A_3=\frac{\sqrt{2\chi _2}\beta k_1}{\chi _1\sqrt{\chi _5}}e^{3i\beta z+i\mathrm{\Delta }kz}u,\hfill \end{array}$$
(6)
where $`\mathrm{\Delta }k\mathrm{\Delta }k_2+\mathrm{\Delta }k_3`$. Renormalising the variables as $`zz/\beta `$ and $`xx/\sqrt{2\beta k_1}`$, we finally obtain a system of coupled equations,
$$\begin{array}{c}i\frac{w}{z}+\frac{^2w}{x^2}w+w^{}v+v^{}u=0,\hfill \\ 2i\frac{v}{z}+\frac{^2v}{x^2}\alpha v+\frac{1}{2}w^2+w^{}u=0,\hfill \\ 3i\frac{u}{z}+\frac{^2u}{x^2}\alpha _1u+\chi vw=0,\hfill \end{array}$$
(7)
where $`\alpha =2(2\beta +\mathrm{\Delta }k_2)/\beta `$ and $`\alpha _1=3(3\beta +\mathrm{\Delta }k)/\beta `$ are two dimensionless parameters that characterise the nonlinear phase matching between the parametrically interacting waves. Dimensionless material parameter $`\chi \chi _1\chi _3/\chi _2^2=9(\sigma _1/\sigma _2)^2`$ depends on the type of phase matching, and it can take different values of order of one. For example, when both SHG and SFM are due to quasi-phase matching (QPM), we have $`\sigma _j=(2/\pi m)(\pi /\lambda _1n_1)\chi ^{(2)}[\omega ;(4j)\omega ;(3j)\omega ]`$, where $`j=1,2`$. Then, for the first-order $`(m=1)`$ QPM processes (see, e.g., Ref. ), we have $`\sigma _1=\sigma _2`$, and therefore $`\chi =9`$. When SFM is due to the third-order QPM process (see, e.g., Ref. ), we should take $`\sigma _1=\sigma _2/3`$, and therefore $`\chi =1`$. At last, when SFM is the fifth-order QPM process, we have $`\sigma _1=\sigma _2/5`$ and $`\chi =9/25`$.
Dimensionless equations (7) present a fundamental model for three-wave multistep cascading solitons in the absence of walk-off. Additionally to the type I SHG solitons (see, e.g., Refs ), the multistep cascading solitons involve the phase-matched SFM interaction ($`\omega +2\omega =3\omega `$) that generates a third harmonic wave. Two-parameter family of localised solutions consists of three mutually coupled waves. It is interesting to note that, similar to the case of nondegenerate three-wave mixing , Eqs. (7) possess an exact solution. To find it, we make a substitution $`w=w_0\mathrm{sech}^2(\eta x)`$, $`v=v_0\mathrm{sech}^2(\eta x)`$ and $`u=u_0\mathrm{sech}^2(\eta x)`$, and obtain unknown parameters from the following algebraic equations
$$w_0^2=\frac{9v_0}{3+4\chi v_0},\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}4}\chi v_0^2+6v_0=9,u_0=\frac{2}{3}\chi w_0v_0,$$
(8)
valid for $`\eta =\frac{1}{2}`$ and $`\alpha =\alpha _1=1`$. Equations (8) have two solutions corresponding to positive and negative values of the amplitude ($`w_0`$). This indicates a possibility of multi-valued solutions, even within the class of exact solutions.
In general, three-wave solitons of Eqs. (7) can be found only numerically. Figures 3(a) and 3(b) present two examples of solitary waves for different sets of the mismatch parameters $`\alpha `$ and $`\alpha _1`$. When $`\alpha _11`$ \[see Fig. 3(a)\], which corresponds to an unmatched SFM process, the amplitude of the third harmonic is small, and it vanishes for $`\alpha _1\mathrm{}`$.
To summarise different types of three-wave solitary waves, in Fig. 4 we plot the dependence of the total soliton power defined as
$$P=_{\mathrm{}}^+\mathrm{}๐x\left(|w|^2+4|v|^2+\frac{9}{\chi }|u|^2\right),$$
(9)
on the mismatch parameter $`\alpha _1`$, for fixed $`\alpha =1`$. It is clearly seen that for some values of $`\alpha _1`$ (including the exact solution at $`\alpha _1=1`$ shown by two filled circles), there exist two different branches of three-wave solitary waves, and only one of those branches approaches, for large values of $`\alpha _1`$, a family of two-wave solitons of the cascading limit (Fig. 4, dashed). The slope of the branches changes from negative (for small $`\alpha _1`$) to positive (for large $`\alpha _1`$), indicating a possible change of the soliton stability. However, the detailed analysis of the soliton stability is beyond the scope of this paper (see, e.g., Refs. ).
### C Parametric soliton-induced waveguides
Another type of multistep cascading parametric processes which involve only two frequencies, i.e. two-colour multistep cascading, can occur due to the vectorial interaction of waves with different polarization. We denote two orthogonal polarization components of the fundamental harmonic (FH) wave ($`\omega _1=\omega `$) as A and B, and two orthogonal polarizations of the second harmonic (SH) wave ($`\omega _2=2\omega `$), as S and T. Then, a simple multistep cascading process consists of the following steps. First, the FH wave A generates the SH wave S via type I SHG process. Then, by down-conversion SA-B, the orthogonal FH wave B is generated. At last, the initial FH wave A is reconstructed by the processes SB-A or AB-S, SA-A. Two principal second-order processes AA-S and AB-S correspond to two different components of the $`\chi ^{(2)}`$ susceptibility tensor, thus introducing additional degrees of freedom into the parametric interaction. Different cases of such type of multistep cascading processes are summarized in Table I.
To demonstrate some of the unique properties of the multistep cascading, we discuss here how it can be employed for soliton-induced waveguiding effects in quadratic media. For this purpose, we consider a model of two-frequency multistep cascading described by the principal DPM process (c) (see Table I above) in the planar slab-waveguide geometry. Using the slowly varying envelope approximation with the assumption of zero absorption of all interacting waves, we obtain
$`\begin{array}{c}2ik_1{\displaystyle \frac{A}{z}}+{\displaystyle \frac{^2A}{x^2}}+\chi _1SA^{}e^{i\mathrm{\Delta }k_1z}=0,\hfill \\ 2ik_1{\displaystyle \frac{B}{z}}+{\displaystyle \frac{^2B}{x^2}}+\chi _2SB^{}e^{i\mathrm{\Delta }k_2z}=0,\hfill \\ 4ik_1{\displaystyle \frac{S}{z}}+{\displaystyle \frac{^2S}{x^2}}+2\chi _1A^2e^{i\mathrm{\Delta }k_1z}+2\chi _2B^2e^{i\mathrm{\Delta }k_2z}=0,\hfill \end{array}`$ (13)
where $`\chi _{1,2}=2k_1\sigma _{1,2}`$, the nonlinear coupling coefficients $`\sigma _k`$ are proportional to the elements of the second-order susceptibility tensor, and $`\mathrm{\Delta }k_1`$ and $`\mathrm{\Delta }k_2`$ are the corresponding wave-vector mismatch parameters.
To simplify the system (13), we look for its stationary solutions and introduce the normalized envelopes $`u`$, $`v`$, and $`w`$ according to the following relations, $`A=\gamma _1u\mathrm{exp}(i\beta z\frac{i}{2}\mathrm{\Delta }k_1z)`$, $`B=\gamma _2v\mathrm{exp}(i\beta z\frac{i}{2}\mathrm{\Delta }k_2z)`$, and $`S=\gamma _3w\mathrm{exp}(2i\beta z)`$, where $`\gamma _1^1=2\chi _1x_0^2`$, $`\gamma _2^1=2x_0^2(\chi _1\chi _2)^{1/2}`$, and $`\gamma _3^1=\chi _1x_0^2`$, and the longitudinal and transverse coordinates are measured in the units of $`z_0=(\beta \mathrm{\Delta }k_1/2)^1`$ and $`x_0=(z_0/2k_1)^{1/2}`$, respectively. Then, we obtain a system of normalized equations,
$$\begin{array}{c}i\frac{u}{z}+\frac{^2u}{x^2}u+u^{}w=0,\hfill \\ i\frac{v}{z}+\frac{^2v}{x^2}\alpha _1v+\chi v^{}w=0,\hfill \\ 2i\frac{w}{z}+\frac{^2w}{x^2}\alpha w+\frac{1}{2}(u^2+v^2)=0,\hfill \end{array}$$
(14)
where $`\chi =(\chi _2/\chi _1)`$, $`\alpha _1=(\beta \mathrm{\Delta }k_2/2)(\beta \mathrm{\Delta }k_1/2)^1`$, and $`\alpha =4\beta (\beta \mathrm{\Delta }k_1/2)^1`$.
First of all, we notice that for $`v=0`$ (or, similarly, $`u=0`$), the dimensionless Eqs. (14) reduce to the corresponding model for the two-step cascading due to type I SHG discussed earlier , and its stationary solutions are defined by the equations for real $`u`$ and $`w`$,
$$\begin{array}{c}\frac{d^2u}{dx^2}u+uw=0,\hfill \\ \frac{d^2w}{dx^2}\alpha w+\frac{1}{2}u^2=0,\hfill \end{array}$$
(15)
that possess a one-parameter family of two-wave localized solutions $`(u_0,w_0)`$ found earlier numerically for any $`\alpha 1`$, and also known analytically for $`\alpha =1`$, $`u_0(x)=\left(3/\sqrt{2}\right)\mathrm{sech}^2(x/2)=\sqrt{2}w_0(x)`$ (see Ref. ).
Then, in the small-amplitude approximation, the equation for real orthogonally polarized FH wave $`v`$ can be treated as an eigenvalue problem for an effective waveguide created by the SH field $`w_0(x)`$,
$$\frac{d^2v}{dx^2}+[\chi w_0(x)\alpha _1]v=0.$$
(16)
Therefore, an additional parametric process allows to propagate a probe beam of one polarization in an effective waveguide created by a two-wave spatial soliton in a quadratic medium with FH component of another polarization. However, this type of waveguide is different from what has been studied for Kerr-like solitons because it is coupled parametrically to the guided modes and, as a result, the physical picture of the guided modes is valid, rigorously speaking, only in the case of stationary phase-matched beams. As a result, the stability of the corresponding waveguide and localized modes of the orthogonal polarization it guides is a key issue. In particular, the waveguide itself (i.e. two-wave parametric soliton) becomes unstable for $`\alpha <\alpha _{\mathrm{cr}}0.2`$ .
In order to find the guided modes of the parametric waveguide created by a two-wave quadratic soliton, we have to solve Eq. (16) where the exact solution $`w_0(x)`$ is to be found numerically. Then, to address this problem analytically, approximate solutions can be used, such as those found with the help of the variational method . However, the different types of the variational ansatz used do not provide a very good approximation for the soliton profile at all $`\alpha `$. For our eigenvalue problem (16), the function $`w_0(x)`$ defines parameters of the guided modes and, in order to obtain accurate results, it should be calculated as close as possible to the exact solutions found numerically. To resolve this difficulty, below we suggest a novel โalmost exactโ solution that would allow to solve analytically many of the problems involving quadratic solitons, including the eigenvalue problem (16).
First, we notice that from the exact solution at $`\alpha =1`$ and the asymptotic result for large $`\alpha `$, $`wu^2/\left(2\alpha \right)2\mathrm{sech}^2(x)`$, it follows that the SH component $`w_0(x)`$ of Eqs. (15) remains almost self-similar for $`\alpha 1`$. Thus, we look for the SH field in the form $`w_0(x)=w_m\mathrm{sech}^2(x/p)`$, where $`w_m`$ and $`p`$ are to be defined. The solution for $`u_0(x)`$ should be consistent with this choice of the shape for SH, and it is defined by the first (linear for $`u`$) equation of the system (15). Therefore, we can take $`u`$ in the form of the lowest guided mode, $`u_0(x)=u_m\mathrm{sech}^p(x/p)`$, that corresponds to an effective waveguide $`w_0(x)`$. By matching the asymptotics of these trial functions with those defined directly from Eqs. (15) at small and large $`x`$, we obtain the following solution,
$$u_0(x)=u_m\mathrm{sech}^p(x/p),w_0(x)=w_m\mathrm{sech}^2(x/p),$$
(17)
$$u_m^2=\frac{\alpha w_m^2}{\left(w_m1\right)},p=\frac{1}{\left(w_m1\right)},\alpha =\frac{4\left(w_m1\right)^3}{\left(2w_m\right)}.$$
(18)
Here, the third relation allows us to find $`w_m`$ for arbitrary $`\alpha `$ as a solution of a cubic equation, and then to find all other parameters as functions of $`\alpha `$. For mismatches in the interval $`0<\alpha <+\mathrm{}`$, the parameter values change monotonically in the regions: $`0<u_m<+\mathrm{}`$, $`1<w_m<2`$, and $`+\mathrm{}>p>1`$. It is really amazing that the analytical solution (17),(18) provides an excellent approximation for the profiles of the two-wave parametric solitons found numerically, with the relative errors not exceeding 1%โ3% for stable solitons (e.g. when $`\alpha >\alpha _{\mathrm{cr}}`$). As a matter of fact, we can treat Eqs. (17) and (18) as an approximate scaling transformation of the family of two-wave bright solitons. Moreover, this solution allows us to capture some remarkable internal similarities and distinctions between the solitons existing in different types of nonlinear media. In particular, as follows from Eqs. (17) and (18), the FH component and the self-consistent effective waveguide (created by the SH field) have approximately the same stationary transverse profiles as for one-component solitons in a Kerr-like medium with power-law nonlinear response . For $`\alpha =1`$ ($`p=2`$) and $`\alpha 1`$ ($`p=1`$) our general expressions reduce to the known analytical solutions, and the FH profile is exactly the same as that for solitons in quadratic and cubic Kerr media, respectively. On the other hand, the strength of self-action for quadratic solitons depends on the normalized phase mismatch $`\alpha `$ and, in general, the beam dynamics for parametric wave mixing can be very different from that observed in Kerr-type media.
Now, the eigenvalue problem (16) can be readily solved analytically. The eigenmode cutoff values are defined by the parameter $`\alpha _1`$ that takes one of the discrete values, $`\alpha _1^{(n)}=(sn)^2/p^2`$, where $`s=(1/2)+[(1/4)+w_m\chi p^2]^{1/2}`$. Number $`n`$ stands for the mode order $`(n=0,1,\mathrm{})`$, and the localized solutions are possible provided $`n<s`$. The profiles of the corresponding guided modes are
$`v_n(x)=V\mathrm{sech}^{sn}(x/p)H(n,2sn+1,sn+1;\zeta /2),`$
where $`\zeta =1\mathrm{tanh}(x/p)`$, $`H`$ is the hypergeometric function, and $`V`$ is the modeโs amplitude which cannot be determined within the framework of the linear analysis.
According to these results, a two-wave parametric soliton creates, a multi-mode waveguide and larger number of the guided modes can be observed for smaller $`\alpha `$. Figures 5(a,b) show the dependence of the mode cutoff values $`\alpha _1^{(n)}(\alpha )`$ for a fixed $`\chi `$, and $`\alpha _1^{(n)}(\chi )`$ for a fixed $`\alpha `$, respectively. For the case $`\chi =1`$, the dependence has a simple form: $`\alpha _1^{(n)}(\alpha )=[1n(w_m1)]^2`$.
Because a two-wave soliton creates an induced waveguide parametrically coupled to its guided modes of the orthogonal polarization, the dynamics of the guided modes may differ drastically from that of conventional waveguides based on the Kerr-type nonlinearities. Figures 6(a-d) show two examples of the evolution of guided modes. In the first example \[see Fig. 6(a-c)\], a weak fundamental mode is amplified via parametric interaction with a soliton waveguide, and the mode experiences a strong power exchange with the orthogonally polarized FH component through the SH field. This process is accompanied by only a weak deformation of the induced waveguide \[see Fig. 6(a) โ dotted curve\]. The resulting effect can be interpreted as a power exchange between two guided modes of orthogonal polarizations in a waveguide created by the SH field. In the second example, the propagation is stable \[see Fig. 6(d)\].
When all the fields in Eq. (14) are not small, i.e. the small-amplitude approximation is no longer valid, the profiles of the three-component solitons should be found numerically. However, some of the lowest-order states can be calculated approximately using the approach of the โalmost exactโ solution (17),(18) described above, which is presented in detail elsewhere . Moreover, a number of the solutions and their families can be obtained in an explicit analytical form. For example, for $`\alpha _1=1/4`$, there exist two families of three-component solitary waves for any $`\alpha 1`$, that describe soliton branches starting at the bifurcation points $`\alpha _1=\alpha _1^{(1)}`$ at $`\alpha =1`$: (i) the soliton with a zero-order guided mode for $`\chi =1/3`$: $`u(x)=\left(3/\sqrt{2}\right)\mathrm{sech}^2\left(x/2\right)`$, $`v(x)=c_2\mathrm{sech}\left(x/2\right)`$, $`w(x)=\left(3/2\right)\mathrm{sech}^2\left(x/2\right)`$, and (ii) the soliton with a first-order guided mode for $`\chi =1`$: $`u(x)=c_1\mathrm{sech}^2\left(x/2\right)`$, $`v(x)=c_2\mathrm{sech}^2\left(x/2\right)\mathrm{sinh}\left(x/2\right)`$, $`w(x)=\left(3/2\right)\mathrm{sech}^2\left(x/2\right)`$, where $`c_2=\sqrt{3\left(\alpha 1\right)}`$ and $`c_1=\sqrt{\left(9/2\right)+c_2^2}`$.
For a practical realization of the DPM processes and the soliton waveguiding effects described above, we can suggest two general methods. The first method is based on the use of two commensurable periods of the quasi-phase-matched (QPM) periodic grating. Indeed, to achieve DPM, we can employ the first-order QPM for one parametric process, and the third-order QPM, for the other parametric process. Taking, as an example, the parameters for LiNbO<sub>3</sub> and AA-S $`(xxz)`$ and BB-S $`(zzz)`$ processes , we find two points for DPM at about 0.89 $`\mu `$m and 1.25 $`\mu `$m. This means that a single QPM grating can provide simultaneous phase-matching for two parametric processes. For such a configuration, we obtain $`\chi 1.92`$ or, interchanging the polarization components, $`\chi 0.52`$. The second method to achieve the conditions of DPM processes is based on the idea of quasi-periodic QPM grating . Specifically, Fibonacci optical superlattices provide an effective way to achieve phase-matching at several incommensurable periods allowing multi-frequency harmonic generation in a single structure. We describe the properties of such structures in the next section.
## V ENVELOPE SOLITONS IN FIBONACCI SUPERLATTICES
### A Incoherence and solitary waves
For many years, solitary waves have been considered as coherent localized modes of nonlinear systems, with particle-like dynamics quite dissimilar to the irregular and stochastic behavior observed for chaotic systems . However, about 20 years ago Akira Hasegawa, while developing a statistical description of the dynamics of an ensemble of plane waves in nonlinear strongly dispersive plasmas, suggested the concept of a localized envelope of random phase waves . Because of the relatively high powers required for generating self-localized random waves, this notion remained a theoretical curiosity until recently, when the possibility to generate spatial optical solitons by a partially incoherent source was discovered in a photorefractive medium .
The concept of incoherent solitons can be compared with a different problem: the propagation of a soliton through a spatially disordered medium. Indeed, due to random scattering on defects, the phases of the individual components forming a soliton experience random fluctuations, and the soliton itself becomes partially incoherent in space and time. For a low-amplitude wave (linear regime) spatial incoherence is known to lead to a fast decay. As a result, the transmission coefficient vanishes exponentially with the length of the system, the phenomenon known as Anderson localization . However, for large amplitudes (nonlinear regime), when the nonlinearity length is much smaller than the Anderson localization length, a soliton can propagate almost unchanged through a disordered medium as predicted theoretically in 1990 and recently verified experimentally .
These two important physical concepts, spatial self-trapping of light generated by an incoherent source in a homogeneous medium, and suppression of Anderson localization for large-amplitude waves in spatially disordered media, both result from the effect of strong nonlinearity. When the nonlinearity is sufficiently strong it acts as an effective phase-locking mechanism by producing a large frequency shift of the different random-phase components, and thereby introducing an effective order into an incoherent wave packet, thus enabling the formation of localized structures. In other words, both phenomena correspond to the limit when the ratio of the nonlinearity length to the characteristic length of (spatial or temporal) fluctuations is small. In the opposite limit, when this ratio is large, the wave propagation is practically linear.
Below we show that, at least for aperiodic inhomogeneous structures, solitary waves can exist in the intermediate regime in the form of quasiperiodic nonlinear localized modes. As an example, we consider SHG and nonlinear beam propagation in Fibonacci optical superlattices, and demonstrate numerically the possibility of spatial self-trapping of quasiperiodic waves whose envelope amplitude varies quasiperiodically, while still maintaining a stable, well-defined spatially localized structure, a quasiperiodic envelope soliton.
### B Quasi-phase-matching in optical superlattices
We consider the interaction of a fundamental wave with the frequency $`\omega `$ (FH) and its SH in a slab waveguide with quadratic (or $`\chi ^{(2)}`$) nonlinearity. Assuming the $`\chi ^{(2)}`$ susceptibility to be modulated and the nonlinearity to be of the same order as diffraction, we write the dynamical equations in the form
$$\begin{array}{c}i\frac{u}{z}+\frac{1}{2}\frac{^2u}{x^2}+d(z)u^{}ve^{i\beta z}=0,\hfill \\ i\frac{w}{z}+\frac{1}{4}\frac{^2w}{x^2}+d(z)u^2e^{i\beta z}=0,\hfill \end{array}$$
(19)
where $`u(x,z)`$ and $`w(x,z)`$ are the slowly varying envelopes of the FH and SH, respectively. The parameter $`\beta =\mathrm{\Delta }k|k_\omega |x_0^2`$ is proportional to the phase mismatch $`\mathrm{\Delta }k=2k_\omega k_{2\omega }`$, $`k_\omega `$ and $`k_{2\omega }`$ being the wave numbers at the two frequencies. The transverse coordinate $`x`$ is measured in units of the input beam width $`x_0`$, and the propagation distance $`z`$ in units of the diffraction length $`l_d=x_0^2|k_\omega |`$. The spatial modulation of the $`\chi ^{(2)}`$ susceptibility is described by the quasi-phase-matching (QPM) grating function $`d(z)`$. In the context of SHG, the QPM technique is an effective way to achieve phase matching, and it has been studied intensively .
Here we consider a QPM grating produced by a quasiperiodic nonlinear optical superlattice. Quasiperiodic optical superlattices, one-dimensional analogs of quasicrystals , are usually designed to study the effect of Anderson localization in the linear regime of light propagation. For example, Gellermann et al. measured the optical transmission properties of quasiperiodic dielectric multilayer stacks of SiO<sub>2</sub> and TiO<sub>2</sub> thin films and observed a strong suppression of the transmission . For QPM gratings, a nonlinear quasiperiodic superlattice of LiTaO<sub>3</sub>, in which two antiparallel ferro-electric domains are arranged in a Fibonacci sequence, was recently fabricated by Zhu et al. , who measured multi-colour SHG with energy conversion efficiencies of $`5\%20\%`$. This quasiperiodic optical superlattice in LiTaO<sub>3</sub> can also be used for efficient direct third harmonic generation .
The quasiperiodic QPM gratings have two building blocks A and B of the length $`l_A`$ and $`l_B`$, respectively, which are ordered in a Fibonacci sequence \[Fig. 7(a)\]. Each block has a domain of length $`l_{A_1}`$=l ($`l_{B_1}`$=l) with $`d`$=$`+1`$ (shaded) and a domain of length $`l_{A_2}`$=$`l(1+\eta )`$ \[$`l_{B_2}`$=$`l(1\tau \eta )`$\] with $`d`$=$`1`$ (white). In the case of $`\chi ^{(2)}`$ nonlinear QPM superlattices this corresponds to positive and negative ferro-electric domains, respectively. The specific details of this type of Fibonacci optical superlattices can be found elsewhere . For our simulations presented below we have chosen $`\eta `$= $`2(\tau 1)/(1+\tau ^2)`$= 0.34, where $`\tau `$= $`(1+\sqrt{5})/2`$ is the so-called golden ratio. This means that the ratio of length scales is also the golden ratio, $`l_A/l_B`$= $`\tau `$. Furthermore, we have chosen $`l`$=0.1.
The grating function $`d(z)`$, which varies between $`+1`$ and $`1`$ according to the Fibonacci sequence, can be expanded in a Fourier series
$$d(z)=\underset{m,n}{}d_{m,n}e^{iG_{m,n}z},G_{m,n}=\frac{2\pi (m+n\tau )}{D},$$
(20)
where $`D`$=$`\tau l_A+l_B`$=0.52 for the chosen parameter values. Hence the spectrum is composed of sums and differences of the basic wavenumbers $`\kappa _1`$=$`2\pi /D`$ and $`\kappa _2`$=$`2\pi \tau /D`$. These components fill the whole Fourier space densely, since $`\kappa _1`$ and $`\kappa _2`$ are incommensurate. Figure 7(b) shows the numerically calculated Fourier spectrum $`G_{m,n}`$. The lowest-order โFibonacci modesโ are clearly the most intense.
### C Quasiperiodic optical solitons
To analyze the beam propagation and SHG in a quasiperiodic QPM grating one could simply average Eqs. (19). To lowest order this approach always yields a system of equations with constant mean-value coefficients, which does not allow to describe oscillations of the beam amplitude and phase. However, here we wish to go beyond the averaged equations and consider the rapid large-amplitude variations of the envelope functions. This can be done analytically for periodic QPM gratings . However, for the quasiperiodic gratings we have to resolve to numerical simulations.
Thus we have solved Eqs. (19) numerically with a second-order split-step routine. At the input of the crystal we excite the fundamental beam (corresponding to unseeded SHG) with a Gaussian profile,
$$u(x,0)=A_ue^{x^2/10},w(x,0)=0.$$
(21)
We consider the quasiperiodic QPM grating with matching to the peak at $`G_{2,3}`$, i.e., $`\beta `$=$`G_{2,3}`$=82.25. First, we study the small-amplitude limit when a weak FH is injected with a low amplitude. Figures 8(a,b) show an example of the evolution of FH and SH in this effectively linear regime. As is clearly seem from Fig. 8(b) the SH wave is excited, but both beams eventually diffract.
When the amplitude of the input beam exceeds a certain threshold, self-focusing and localization should be observed for both harmonics. Figures 8(c,d) show an example of the evolution of a strong input FH beam, and its corresponding SH. Again the SH is generated, but now the nonlinearity is so strong that it leads to self-focusing and mutual self-trapping of the two fields, resulting in a spatially localized two-component soliton, despite the continuous scattering of the quasiperiodic QPM grating.
It is important to notice that the two-component localized beam created due to the self-trapping effect is quasiperiodic by itself. As a matter of fact, after an initial transient its amplitude oscillates in phase with the quasiperiodic QPM modulation $`d(z)`$. This is illustrated in Fig. 9, where we show in more detail the peak intensities in the asymptotic regime of the evolution. The oscillations shown in Fig. 9 are in phase with the oscillations of the QPM grating $`d(z)`$, and we indeed found that their spectra are similar.
Our numerical results show that the quasiperiodic envelope solitons can be generated for a broad range of the phase-mismatch $`\beta `$. The amplitude and width of the solitons depend on the effective mismatch, which is the separation between $`\beta `$ and the nearest strong peak $`G_{m,n}`$ in the Fibonacci QPM grating spectrum \[see Fig. 7(b)\]. Thus, low-amplitude broad solitons are excited for $`\beta `$-values in between peaks, whereas high-amplitude narrow solitons are excited when $`\beta `$ is close to a strong peak, as shown in Fig. 8(c,d).
To analyse in more detail the transition between the linear (diffraction) and nonlinear (self-trapping) regimes, we have made a series of careful numerical simulations . In Fig. 10 we show the transmission coefficients and the beam widths at the output of the crystal versus the intensity of the FH input beam, for a variety of $`\beta `$-values. These dependencies clearly illustrate the universality of the generation of localised modes for varying strength of nonlinearity, i.e. a quasiperiodic soliton is generated only for sufficiently high amplitudes. This is of course a general phenomenon also observed in many nonlinear isotropic media. However, here the self-trapping occurs for quasiperiodic waves, with the quasiperiodicity being preserved in the variation of the amplitude of both components of the soliton.
## VI CONCLUSION
We have overviewed several important physical examples of the multi-component solitary waves which appear due to multi-mode and/or multi-frequency coupling in nonlinear optical fibers and waveguides. We have described several types of such multi-component solitary waves, including: (i) multi-wavelength solitary waves in multi-channel bit-parallel-wavelength fiber transmission systems, (ii) multi-colour parametric spatial solitary waves due to multistep cascading in quadratic materials, and (iii) quasiperiodic envelope solitons in Fibonacci optical superlattices. These examples reveal some general features and properties of multi-component solitary waves in nonintegrable nonlinear models, also serving as a stepping stone for approaching other problems of the multi-mode soliton coupling and interaction.
## ACKNOWLEDGMENTS
The work was supported by the Australian Photonics Cooperative Research Centre and by a collaborative Australia-Denmark grant of the Department of Industry, Science, and Tourism (Australia). |
warning/0001/quant-ph0001047.html | ar5iv | text | # Quantum Diffusions and Appell Systems 11footnote 1Submitted to Journal of Computational and Applied Mathematics. Special Issue of Proccedings of Fifth Inter. Symp. on Orthogonal Polynomaials, Special Functions and their Applications. Keywords: Random walks, Hopf Algebras, Appell Systems.
## Abstract
Within the algebraic framework of Hopf algebras, random walks and associated diffusion equations (master equations) are constructed and studied for two basic operator algebras of Quantum Mechanics i.e the Heisenberg-Weyl algebra ($`hw`$) and its $`q`$-deformed version $`hw_q`$. This is done by means of functionals determined by the associated coherent state density operators. The ensuing master equations admit solutions given by $`hw`$ and $`hw_q`$-valued Appell systems.
1. Introduction. We work in the general framework of the so called quantum probability theory and more specifically along the research line relating random walks, diffusions and Markov transition operators to Lie-Hopf algebras. Our aim is to construct algebraic random walks and their diffusion limit in terms of master equations. We work with two basic operator algebras of Quantum Mechanics i.e the Heisenberg-Weyl algebra ($`hw`$) and its $`q`$-deformed version $`hw_q`$, and use their Hopf algebra like structures for our construction (Chapt. 2). The density of the two functionals needed are constructed by the associated to those algebras coherent states vectors. As the random walks take place on the manifold of these coherent states vectors it is important to investigate the geometrical features of them (Chapt. 3). Then a limiting procedure leads to the master (diffusion) equations for the case of $`hw`$ random walk (Chapt. 5) and the case of $`hw_q`$ random walk (Chapt. 6), correspondingly. The solutions of the resulting master equations of motion for certain general elements of the respective operator algebras are obtained in terms of the associated operator valued Appell systems. Certain generalities of classical Appell systems are discussed in Chapt. 4. Finally, some technicalities such as ordering formulae for generators of the two $`hw`$ algebras, as well as some Baker-Campbell-Hausdorff decompositions formulae for the $`SU(1,1)`$ group elements are summarized in Appendices A and B.
2. Hopf Algebras. A Hopf algebra $`๐=๐(\mu ,\eta ,\mathrm{\Delta },ฯต,S)`$ over a field $`k`$ is a vector space equipped with an algebra structure with homomorphic associative product map $`\mu :๐\times ๐๐`$, and a homomorphic unit map $`\eta :k๐`$, that are related by $`\mu (\eta id)=id=\mu (id\eta )`$, together with a coalgebra structure with a homomorphic coassociative coproduct map $`\mathrm{\Delta }:๐๐๐`$ and a homomorphic counit map $`ฯต:๐k`$, that are related between them by $`(ฯตid)\mathrm{\Delta }=id=(idฯต)\mathrm{\Delta }`$. Both products satisfy the compatibility condition of bialgebra i.e $`(\mu \mu )(id\tau id)(\mathrm{\Delta }\mathrm{\Delta })=\mathrm{\Delta }\mu `$, where $`\tau (xy)=yx`$ stands for the twist map. If $`\eta `$ or $`ฯต`$ is not defined in $`๐`$ we speak about non unital or non counital Hopf algebra.
Suppose we have a functional $`\varphi :๐๐`$, defined on $`๐`$, let us define the operator $`T_\varphi :๐๐`$ as $`T_\varphi =(\varphi id)\mathrm{\Delta }`$, then $`ฯตT_\varphi =\varphi `$, namely the counit aids to pass from the operator to its associated functional. From this relation we can define the convolution product $`\psi \varphi `$, between functionals as follows :
$`ฯตT_\psi T_\varphi `$ $`=`$ $`ฯต(\psi id)\mathrm{\Delta }(\varphi id)\mathrm{\Delta }=(\varphi \psi )(ididฯต)(id\mathrm{\Delta })\mathrm{\Delta }`$ (1)
$`=`$ $`(\varphi \psi )\mathrm{\Delta }=\varphi \psi ,`$
and in general $`ฯตT_\varphi ^n=ฯตT_{\varphi ^n}=\varphi ^n`$. These last relations imply that the transition operators form a discrete semigroup wrt their composition with identity element $`T_ฯตid`$ (due to the axioms of Hopf algebra) and generator $`T_\varphi `$, while the functionals form a dual semigroup wrt the convolution with identity element $`e`$ and generator $`\varphi `$, and that these two semigroups are homomorphic to each other. and the
We recall now two algebras and their structural maps that concerns us here:
i) Heisenberg-Weyl algebra $`hw`$: this is the algebra of the quantum mechanical oscillator and is generated by the creation, annihilation and the unit operator $`\{a^{},a,\mathrm{๐}\}`$ respectively which satisfy the commutation relation (Lie bracket) $`[a,a^{}]=\mathrm{๐}`$, while $`\mathrm{๐}`$ commutes with the other elements. This algebra possesses a natural non counital Hopf algebra structure (or bialagebra-like cf. , Chap. 3), with comultiplication defined as
$`\mathrm{\Delta }^{(n1)}a`$ $`=`$ $`n^{\frac{1}{2}}(a\mathrm{}\mathrm{๐}+\mathrm{๐}a\mathrm{}\mathrm{๐}+\mathrm{๐}\mathrm{}a),`$
$`\mathrm{\Delta }^{(n1)}a^{}`$ $`=`$ $`n^{\frac{1}{2}}(a^{}\mathrm{}\mathrm{๐}+\mathrm{๐}a\mathrm{}\mathrm{๐}+\mathrm{๐}\mathrm{}a^{}),`$
$`\mathrm{\Delta }\mathrm{๐}`$ $`=`$ $`\mathrm{๐}\mathrm{๐}+\mathrm{๐}\mathrm{๐}.`$ (2)
Let us also define the so called number operator $`N=a^{}a`$ with the following commutation relations with the generators of $`hw`$:
$$[a,a^{}]=\mathrm{๐},[N,a^{}]=a^{},[N,a]=a.$$
(3)
The module which carries the unique irreducible and infinite dimensional representation of the oscillator algebra is the Hilbert-Fock space $`_F`$ which is generated by a starting (or โvacuumโ ) state vector $`|0`$ and is given as $`=\{|n=\frac{(a^{})^n}{n!}|0,n๐_+`$.
ii) The $`q`$-deformed Heisenberg-Weyl algebra $`hw_q`$: The $`q`$-deform Heisenberg-Weyl algebra is generated by the elements $`hw_q=<b,b^{},q^N,q^N,\mathrm{๐}>`$ that satisfy the relations
$`bb^{}q^1b^{}b=q^N,q^Nq^N=\mathrm{๐},`$
$`q^Nbq^N=q^1b,q^Nb^{}q^N=qb^{}.`$ (4)
For real $`q`$ the Fock representation space is spanned by the vectors $`\{|n=\frac{(b^{})^n}{\sqrt{[n]_q!}}|0,n๐_+\}`$, where $`[n]_q=\frac{q^nq^n}{qq^1}`$ and $`[n]_q=[1]_q[2]_q\mathrm{}[n]_q`$. In the Fock space representation of this algebra we have the additional relations $`b^{}b=[N]_q`$, $`bb^{}=[N+1]_q`$. This algebras has no satisfactory Hopf structure but still as will be seen below we can define algebraic random walks on it and study their diffusion limit. To this end let us make the transformations $`a_q=q^{N/2}b`$ and $`a_q^{}=b^{}q^{N/2}`$, and obtain the resulting algebra
$$a_qa_q^{}q^2a_q^{}a_q=\mathrm{๐}$$
(5)
which is the new form of the $`hw_q`$ algebra. Although not an algebra homomorphisms we will use below the coassociative maps
$`\mathrm{\Delta }a_q`$ $`=`$ $`a_q\mathrm{๐}+\mathrm{๐}a_q,\mathrm{\Delta }a_q^{}=a_q^{}\mathrm{๐}+\mathrm{๐}a_q^{}.`$ (6)
3. Coherent States . For our needs here a brief introduction to the concept of coherent states (CS) on Lie groups goes as follows: consider a Lie group $`๐ข`$, with a unitary irreducible representation $`T(g)`$, $`g๐ข`$, in a Hilbert space $``$. We select a reference vector $`|\mathrm{\Psi }_0`$, to be called the โvacuumโ state vector, and let $`๐ข_0๐ข`$ be its isotropy subgroup, i.e for $`h๐ข_0`$, $`T(h)|\mathrm{\Psi }_0=e^{i\phi (h)}|\mathrm{\Psi }_0`$. The map from the factor group $`=๐ข/๐ข_0`$ to the Hilbert space $``$, introduced in the form of an orbit of the vacuum state under a factor group element, defines a CSV $`|x=T(๐ข/๐ข_0)|\mathrm{\Psi }_0`$ labelled by points $`x`$ of the coherent state manifold. Coherent states form an (over)complete set of states, since by means of the Haar invariant measure of the group $`๐ข`$ viz. $`d\mu (x),x`$, they provide a resolution of unity, $`\mathrm{๐}=_{}d\mu (x)\left|xx\right|`$. As a consequence, any vector $`|\mathrm{\Psi }`$ is analyzed in the CS basis, $`|\mathrm{\Psi }=_{}๐\mu (x)\mathrm{\Psi }(x)|x`$, with coefficients $`\mathrm{\Psi }(x)=x|\mathrm{\Psi }`$. We should note here that the square integrability of the vectors $`\mathrm{\Psi }`$ will impose some limits on the growth parameters of the functions $`\mathrm{\Psi }(x)`$ (cf. and references therein).
What concerns us here is mostly the geometry of the CS manifold $``$. This is due to the fact that the random walks and their diffusion limits that will be study below will be given in terms of functionals associated with coherent states so that the random walks will be induced on the functions defined on $``$ (passive description) or on the operators acting on the functions defined on $``$ (active description). Although only the latter description will be studied here in terms of the quantum master equations, it should be obvious that the geometry of the background manifold $``$ namely both the Riemannian and the symplectic geometry (the symplectic geometry especially in the case of non stationary random walks), will manifest itself in the associated diffusion equations. Specifically below it will be shown that the $`hw`$ random walk takes place on the flat complex plane $`๐`$ with canonical symplectic structure, while the deformed $`hw_q`$ random walk takes place on a $`q`$-deformed surface of revolution with modified, due to $`q`$-deformation, Riemannian and symplectic geometry. This fact provides a further motivation for studying random walks and diffusions within the present algebraic framework since in this way we are able to study these phenomena taking place on non trivial spaces. Details constructions and studies can be found elsewhere, here we summarize some relevant information:
Let us first specialize to the $`HW`$ group: The $`hw`$-CS is defined by the realation
$$|\alpha =e^{\alpha a^{}\overline{\alpha }a}|0=๐ฉe^{\alpha a^{}}|0=e^{\frac{1}{2}\left|\alpha \right|^2}\underset{n=0}{\overset{\mathrm{}}{}}\frac{\alpha ^n}{\sqrt{n!}}|n.$$
(7)
It is an (over)complete set of states with respect to the measure $`d\mu (\alpha )=\frac{1}{\pi }e^{\left|\alpha \right|^2}d^2\alpha `$ for the non-normalized CS, and $`\alpha =HW/U(1)๐`$ is the CS manifold. Since $`a|\alpha =\alpha |\alpha `$, $``$ is the flat canonical phase plane with the standard line element $`ds^2=d\alpha d\overline{\alpha }`$. Also the symplectic 2-form $`\omega =id\alpha d\overline{\alpha }`$ is associated to the canonical Poisson bracket $`\{f,g\}=i(_\alpha ^f_{\overline{\alpha }}^g_{\overline{\alpha }}^f_\alpha ^g)`$.
Next we turn to the $`hw_q`$ case: The definition of the $`hw_q`$-CS reads
$$||\alpha _q=e_q^{\alpha a_q^{}}|0=e^{\alpha A_q^{}}|0=\underset{n=0}{\overset{\mathrm{}}{}}\frac{\alpha ^n}{\sqrt{[n]!}}|n,$$
(8)
where $`[n]=\frac{q^{2n}1}{q^21}`$. The states are first defined in terms of the $`q`$-deformed exponential function $`e_q^x=_{no}\frac{x^n}{[n]!}`$ and the $`q`$-creation operator and then equivalently by exponentiation of the operator $`A_q^{}=\frac{N}{[N]}a_q^{}`$, that satisfies with the $`hw_q`$ elements the $`hw`$ algebra relations
$$[a_q,A_q^{}]=\mathrm{๐},[A_q,a_q^{}]=\mathrm{๐}.$$
(9)
The $`q`$-CS is an (over)complete set of states with respect to the measure $`d\mu (\alpha )_q=\frac{1}{\pi }(e_q^{\left|\alpha \right|^2})^1d_q^2\alpha `$, and wrt the Jackson $`q`$-integral. If $`q=e^\lambda `$, then since $`a_q|\alpha _q=\alpha |\alpha _q`$, the $`q`$-CS manifold $``$ is a non flat surface of revolution with $`q`$-deformed induced curvature with curvature scalar $`R=\lambda ^212(1+2\left|\alpha \right|^2+๐ช(\lambda ^3))`$. Also the symplectic 2-form $`\omega `$ is modified by the $`q`$-deformation as $`\omega =\{i\frac{\lambda ^2}{2}\left|\alpha \right|^2(\left|\alpha \right|^2+2)+๐ช(\lambda ^3)\}d\alpha d\overline{\alpha }`$ .
The density operator (state) $`\rho `$ will be used below to determine functionals of some Hopf operator algebras $`๐`$, so here we introduce the general concept and give its construction in terms of convex combinations of projectors of coherent states. Let a Hilbert vector space $``$ that carries a unitary irreducible representation of $`๐`$ of finite or infinite dimension. The set
$$๐ฎ=\{\rho \mathrm{End}():\rho 0,\rho ^{}=\rho ,tr\rho =1\},$$
(10)
namely the set of non-negative, Hermitian, trace-one operators acting on $``$ form a convex subspace of $`\mathrm{End}()`$, which is the convex hull of the set
$$๐ฎ_P=\{\rho ๐ฎ,\rho ^2=\rho \}/U(1),$$
(11)
namely of the set of pure density operators (states), that are in one-to-one correspondance with the state vectors of $``$. Two kinds of $`\rho `$ density operators that will be used in the sequel are constructed by $`hw`$-CS and $`hw_q`$-CS. Explicitly from the pure density operators $`|\pm \alpha \pm \alpha |๐ฎ_P`$ and the $`q`$-deformed ones $`|\alpha _{qq}\alpha ||\alpha \alpha |_q๐ฎ_P`$, we form convex combination belonging to the convex hull of $`๐ฎ_P`$ i.e
$`\rho `$ $`=`$ $`p\left|\alpha \alpha \right|+(1p)|\alpha \alpha |,`$
$`\rho _q`$ $`=`$ $`p|\alpha \alpha |_q+(1p)|\alpha \alpha |_q.`$ (12)
4. Appell Systems. Classical Appell polynomials on the real line are polynomials $`\{h_n(x);n๐\}`$ of degree $`n`$ that satisfy the condition $`(d/dx)h_n(x)=nh_n(x)`$. A class of such systems is the shifted moment sequences $`h_n(x)=_{\mathrm{}}^{\mathrm{}}(x+y)^np(dy)`$, for some positive real measure $`p`$ with finite moments. The class of Appell polynomials includes cases such as the divided sequences, the Bernoulli polynomials and the Hermite polynomials, which correspond to the Gaussian measure $`p=p(dy)=\frac{1}{\sqrt{2\pi }}e^{y^2/2}dy`$. Some important properties of the Appell polynomial sets that have been investigated are the following: Hermite polynomials are the only Appell polynomials associated to the ordinary derivative operator that are also orthogonal (e, b, c), similarly Charlier polynomials are the only Appell systems associated to the difference operator that are also orthogonal(d), while the Rogergs $`q`$-Hermite polynomials are the only Appell systems associated to Askey-Wilson $`q`$-derivative operator that are orthogonal too(e).
The following Hopf algebraic reformulation of the real line Appell systems (i.e non polynomials necesserily) motivates their generalization to more general spaces. Let $`๐=๐[[X]]`$ the algebra of the real formal power series generated by pointwise multiplication $`fg(x)=f(x)g(x),f,g๐`$. Then $`๐`$ becomes a Hopf algebra with comultiplication $`(\mathrm{\Delta }f)(x,y)=f(x+y)`$ and counit $`ฯต(id)=1`$, $`ฯต(X)=0`$, where $`id`$ is the identity function and $`X(x)=x`$ stands for the coordinate function. For a given functional $`\varphi :๐๐`$ and a chosen basis $`(x^n),n๐_+`$ in $`๐`$, it is easy to verify that the relation $`h_n(x)=(\varphi id)\mathrm{\Delta }x^n=T_\varphi x^n`$ defines an Appell systems and is equivalent to the preceding definition. Specifically for $`\varphi =_{\mathrm{}}^{\mathrm{}}p(dy)`$ with $`p`$ the Gaussian measure we obtain the Hermite polynomials if we make the identifications $`x1x`$ and $`1xiy`$. This algebraic definition has been used extensively to introduce Appell systems in non commuting algebras. Here we will utilize it to define below Appell systems on two important operator algebras of Quantum Mechanics i.e the Heisenberg algebra and the $`q`$-deformed Heisenberg algebra and to show that the resulting operator valued Appell systems are solutions of quantum master equations that are constructed respectively as limits of random walks defined on these algebras.
5. Diffusion on C. Let $`\varphi ()=\mathrm{Tr}\rho ()<\rho ,>`$, a functional defined on the enveloping Heisenberg-Weyl algebra $`๐ฐ(hw)`$, where $`\rho =p\left|\alpha \alpha \right|+(1p)|\alpha \alpha |`$, i.e the $`\rho `$ density operator is given as a convex sum of pure state density operators. The action of the transition operator $`T_\varphi =(\varphi id)\mathrm{\Delta }`$ on the generating monomials of $`๐ฐ(hw)`$ (where we ignore the numerical factors in the comultiplication of eq.(2)) reads,
$`T_\varphi ((a^{})^ma^n)`$ $`=`$ $`(\varphi id)\mathrm{\Delta }((a^{})^ma^n)`$ (17)
$`=`$ $`{\displaystyle \underset{i=0}{\overset{m}{}}}{\displaystyle \underset{j=0}{\overset{n}{}}}\left(\begin{array}{c}m\\ i\end{array}\right)\left(\begin{array}{c}n\\ j\end{array}\right)[p\alpha ^i\alpha ^j+(1p)(\alpha )^i(\alpha )^j](a^{})^{mi}a^{nj}`$
$`=`$ $`p(a^{}+\alpha ^{})^m(a+\alpha )^n+(1p)(a^{}\alpha ^{})^m(a\alpha )^n.`$ (18)
For a general element $`f(a,a^{})๐ฐ(hw)`$ that is normally ordered, namely the annihilation operator $`a`$ is placed to the right of the creation operator $`a^{}`$, denoted by $`\widehat{f}(a,a^{})=_{m,n0}c_{mn}(a^{})^ma^n`$, the action of the linear operator $`T_\varphi `$ becomes
$$T_\varphi (\widehat{f}(a,a^{}))=p\widehat{f}(a+\alpha ,a^{}+\alpha ^{})+(1p)\widehat{f}(a\alpha ,a^{}\alpha ^{})$$
(19)
By means of the CS eigenvector property and the normal ordering of the $`f`$ element we also compute the value of functional viz.
$$\varphi (\widehat{f}(a,a^{}))=p\widehat{f}(\alpha ,\alpha ^{})+(1p)\widehat{f}(\alpha ,\alpha ^{}).$$
(20)
Let us consider the displacement operator $`D_\alpha =e^{\alpha a^{}\alpha ^{}a}`$ which acts with the group adjoint action on any element $`f`$ of the $`๐ฐ(hw)`$ algebra viz.
$$AdD_a(f)=Ade^{\alpha a^{}\alpha ^{}a}(f)=Ade^{ad(\alpha a^{}\alpha ^{}a)}(f)=D_\alpha fD_\alpha ^{},$$
(21)
where $`ad(X)f=[X,f]`$ and $`ad(X)ad(X)f=[X,[X,f]]`$ and similarly for higher powers, stands for the Lie algebra adjoint action that is defined in terms of the Lie commutator. Explicitly the action of the displacement operator on the generators of $`๐ฐ(hw)`$ reads $`AdD_{\pm \alpha }(a)=a\alpha `$ and $`AdD_{\pm \alpha }(a^{})=a^{}\alpha ^{}`$. By means of these expressions we rewrite the action of the preceding transition operator as
$$T_\varphi (\widehat{f}(a,a^{}))=[pAdD_\alpha +(1p)AdD_\alpha ]\widehat{f}$$
(22)
Next we want to compute the limiting transition operator
$`T_t`$ $``$ $`T_{\varphi _t}\underset{n\mathrm{}}{lim}T_\varphi ^n`$ (23)
$`=`$ $`\underset{n\mathrm{}}{lim}[p(1+ad(\alpha a^{}+\alpha ^{}a)+{\displaystyle \frac{1}{2}}adad(\alpha a^{}+\alpha ^{}a)+\mathrm{})`$
$`+`$ $`(1p)(1+ad(\alpha a^{}\alpha ^{}a)+{\displaystyle \frac{1}{2}}adad(\alpha a^{}\alpha ^{}a)+\mathrm{}]^n.`$
If we introduce the parameters $`t๐`$ and $`c,\gamma ๐`$ by means of the relations,
$$2\alpha (p\frac{1}{2})=\frac{tc}{n},\frac{\alpha ^2}{2}=\frac{t\gamma }{n},$$
(24)
and then take $`\alpha 0,n\mathrm{}`$, with $`t,c,\gamma `$ fixed, we use the limit $`(1+\frac{Z}{n})^ne^Z`$, to arrive at the limiting Markov operator $`T_t=e^{tad}`$, where
$$=ca^{}+c^{}a+\gamma (a^{})^2\gamma ^{}a^2\left|\gamma \right|(a^{}a+aa^{}).$$
(25)
By construction $`T_t`$ is the time evolution operator for any element $`f`$ of $`๐ฐ(hw)`$ i.e $`f_t=T_t(f)`$ and forms a continous semigroup $`T_tT_t^{}=T_{t+t^{}}`$ under composition. This yields the diffusion equation obeyed by $`f_t`$, which will be taken to be normally ordered hereafter. By time derivation of the equation
$$\varphi _t(\widehat{f})=<\rho ,\widehat{f}_t>=<\rho ,e^{tad}\widehat{f}>=<e^{tad^{}}\rho ,\widehat{f}>=<\rho _t,\widehat{f}>,$$
(26)
we obtain the diffusion equation $`\frac{d}{dt}\widehat{f}_t=\widehat{f}_t`$, as well as the dual one satisfied by the $`\rho `$ density operator viz. $`\frac{d}{dt}\rho _t=^{}\rho _t`$. To simplify and eventually solve the ensuing equations we will assume here that the parameter $`\gamma `$ introduced above is a complex variable with random argument of zero average and constant non zero magnitude. Then if we average over random $`\gamma `$ the equations of motion only the term proportional to the amplitude of $`\gamma `$ will be retained. If in addition we consider the case of an symmetric random walk i.e $`p=1/2,c=0`$ the equation of motion becomes
$$\frac{d}{dt}\widehat{f}_t=2\left|\gamma \right|\left[a^{}\widehat{f}_ta+a\widehat{f}_ta^{}N\widehat{f}\widehat{f}(N+1)\right].$$
(27)
This is a quantum master equation of the Lindblad type which will be shown to admit a solution in terms of a operator valued Appell system associated with the generator of that equation. We may introduce the following operators
$$K_+f=a^{}faK_{}f=afa^{}K_0f=\frac{1}{2}(a^{}af+faa^{}),$$
(28)
and $`K_cf=[a^{}a,f]`$. These operators acting on the elements $`f`$ of the enveloping algebra $`๐ฐ(hw)`$, generate the $`su(1,1)`$ Lie algebra defined by the commutation relations
$$[K_{},K_+]=2K_0,[K_0,K_\pm ]=\pm K_\pm ,$$
(29)
where $`K_c`$ is the central element (Casimir operator ) of the algebra. In terms of these operators the quantum master equation (27) is cast in the form
$$\frac{d}{dt}\widehat{f}_t=2\left|\gamma \right|(2K_0+K_++K_{})\widehat{f}_t.$$
(30)
Use of the disentangling theorem (Baker-Campbell-Hausdorff formula) of a general $`SU(1,1)`$ group element (c.f Appendix A), allows to express the solution of the quantum master equation in the form
$$\widehat{f}_t=\mathrm{exp}(A_+K_+)\mathrm{exp}(\mathrm{ln}A_0K_0)\mathrm{exp}(A_{}K_{})(\widehat{f})=\mathrm{exp}(B_{}K_{})\mathrm{exp}(\mathrm{ln}B_0K_0)\mathrm{exp}(B_+K_+)(\widehat{f}),$$
(31)
if the normally or respectively antinormally ordered BCH decomposition is used. Above $`\widehat{f}=_{n0}c_{mn}(a^{})^sa^t`$, stands for the initial time operator which can be a general element of the enveloping algebra $`๐ฐ(hw)`$. Specifically in the case of normally ordered decomposition with initial operator taken as $`\widehat{f}=(a^{})^ma^n`$ the solution of the quantum master equation is obtained by means of the actions issued in eq.(28) and by the antinormal-to-normal reordering relations among the generators of the $`๐ฐ(hw)`$ algebra (c.f Appemdix B). An arduous but straightforward calculation yields the normal ordered solution:
$`\widehat{f}_t=\mathrm{exp}(A_+K_+)\mathrm{exp}(\mathrm{ln}A_0K_0)\mathrm{exp}(A_{}K_{})((a^{})^sa^t)=`$
$`{\displaystyle \underset{k0}{}}{\displaystyle \underset{l0}{}}{\displaystyle \underset{m0}{}}{\displaystyle \underset{i=0}{\overset{min(k,s)}{}}}{\displaystyle \underset{j=0}{\overset{min(k+ti,k)}{}}}{\displaystyle \underset{u=0}{\overset{l}{}}}{\displaystyle \underset{v=0}{\overset{u}{}}}{\displaystyle \underset{q=0}{\overset{lu}{}}}{\displaystyle \underset{w=0}{\overset{v}{}}}{\displaystyle \underset{f=0}{\overset{min(q,x)}{}}}{\displaystyle \underset{h=0}{\overset{min(y+qf,w)}{}}}\times `$
$`{\displaystyle \frac{A_{}^k}{k!}}{\displaystyle \frac{\overline{A}_0^l}{l!}}{\displaystyle \frac{A_+^m}{m!}}d_{k,s}^id_{k+ti,k}^jd_{q,x}^fd_{y+qf,w}^h\overline{d}_{lu,q}\overline{d}_{v,w}{\displaystyle \frac{1}{2^l}}\left(\begin{array}{c}l\\ u\end{array}\right)\left(\begin{array}{c}u\\ v\end{array}\right)(a^{})^{x+w+mfh}a^{y+q+mfh}`$ (36)
where $`x=s+k+qij`$, $`y=t+k+wij`$ and $`\overline{A}_0=\mathrm{ln}A_0`$, with $`A_0=\frac{1}{14\left|\gamma \right|t}`$ and $`A_\pm =\frac{2\left|\gamma \right|t}{12\left|\gamma \right|t}`$. A similar solution can be obtained for the antinormal BCH decomposition. We can therefore state the results in the following
Proposition 1. The solution of the quantum master equation $`\frac{d}{dt}\widehat{f}_t=\widehat{f}_t`$ where the generator $`(\widehat{f}_t)=2\left|\gamma \right|\left[a^{}\widehat{f}_ta+a\widehat{f}_ta^{}N\widehat{f}\widehat{f}(N+1)\right]`$ of Lindblad type generates the semigroup of Markov transition operators $`T_t=e^t`$ acting on the enveloping algebra $`๐ฐ(hw)`$, is given by the associated $`๐ฐ(hw)`$-valued Appell system which in its normally ordered form is given by equation (36).
We note also that the dual master equation satisfied by the density operator can easily be solved along the above lines in terms of the associated Appell system.
6. $`q`$-Diffusion. Let $`\varphi _\varphi ()=\mathrm{Tr}\rho _q()<\rho _q,>`$, a functional defined on the enveloping $`q`$-Heisenberg-Weyl algebra $`๐ฐ_q(hw)`$, where $`\rho _q=p|\alpha \alpha |_q+(1p)|\alpha \alpha |_q`$ is the $`\rho `$ density operator given as a convex sum of pure state $`q`$-density operators. The action of transition operator $`T_\varphi ^q=(\varphi _qid)\mathrm{\Delta }`$ on the monomials of $`๐ฐ_q(hw)`$, with $`\mathrm{\Delta }`$ map given is eq.(6) reads,
$`T_{\varphi _q}(a^{})_q^ma_q^n)`$ $`=`$ $`(\varphi _qid)\mathrm{\Delta }((a^{})_q^ma_q^n)`$ (41)
$`=`$ $`{\displaystyle \underset{i=0}{\overset{m}{}}}{\displaystyle \underset{j=0}{\overset{n}{}}}\left(\begin{array}{c}m\\ i\end{array}\right)\left(\begin{array}{c}n\\ j\end{array}\right)[p\alpha ^i\alpha ^j+(1p)(\alpha )^i(\alpha )^j](a^{})_q^{mi}a_q^{nj}`$
$`=`$ $`p(a_q^{}+\alpha ^{})^m(a_q+\alpha )^n+(1p)(a_q^{}\alpha ^{})^m(a_q\alpha )^n.`$ (42)
On an element $`f(a_q,a_q^{})`$ of the enveloping algebra $`๐ฐ_q(hw)`$ that is normally ordered, namely the annihilation operator $`a_q`$ is placed to the right of the creation operator $`a_q^{}`$, that is expressed as $`\widehat{f}(a_q,a_q^{})=_{m,n0}c_{mn}(a^{})_q^ma_q^n`$, the action of the linear operator $`T_{\varphi _q}`$ becomes
$$T_{\varphi _q}(\widehat{f}(a_q,a_q^{}))=p\widehat{f}(a_q+\alpha ,a_q^{}+\alpha ^{})+(1p)\widehat{f}(a_q\alpha ,a_q^{}\alpha ^{}).$$
(43)
By means of the $`q`$-CS eigenvector property and the normal ordering of the element $`f`$ we also compute the value of functional viz.
$$\varphi _q\widehat{f}(a_q,a_q^{})=p\widehat{f}(\alpha _q,\alpha _q^{})+(1p)\widehat{f}(\alpha ,\alpha ^{}).$$
(44)
Let us now consider the displacement operator $`D_\alpha ^q=e^{\alpha A_q^{}\alpha ^{}a_q}`$, which acts with the following adjoint action on any element $`f`$ of the $`๐ฐ_q(hw)`$ algebra, $`AdD_a^q(f)=Ade^{\alpha A_q^{}\alpha ^{}a_q}(f)=e^{ad(\alpha A_q^{}\alpha ^{}a_q)}(f)=D_\alpha ^qfD_\alpha ^q`$. We should emphasize at this point that $`D_\alpha ^qD_\alpha ^q`$. This is an important difference from the preceding undeformed case with $`q=1`$, which stems from the fact the though eq.(5) is valid the two involved operators are not Hermitian conjugate to each other. This fact would not permit us to proceed for the construction of quantum diffusion equation in a manner analogous to the $`q=1`$ case. Instead here we will restrict the space of solutions of the resulting $`q`$-master equation from the whole algebra $`๐ฐ_q(hw)`$ to the commuting subalgebra generated either by monomials of the creation operator $`\{(a^{})_q^m,m๐_+\}`$ or of the annihilation operator $`\{a_q^m,m๐_+\}`$ alone. Notice however that such a choice would be undesirable from the physical point of view since it would not allow us to study Hermitian solutions of the ensuing master equation.
Then the explicit action of the $`q`$-displacement operator on the generators of $`๐ฐ_q(hw)`$ reads $`AdD_{\pm \alpha }^q(a_q)=a_q\alpha `$ and $`AdD_\alpha ^q(a_q^{})=a_q^{}\alpha ^{}`$. By means of these expressions we rewrite the action of the preceding $`q`$-transition operator on an analytic formal power series $`f(a_q)`$ as
$$T_{\varphi _q}(f(a_q))=[pAdD_\alpha ^q+(1p)AdD_\alpha ^q](f(a_q)).$$
(45)
We wish to compute the limiting transition operator
$`T_t^q`$ $``$ $`T_{\varphi _t^q}\underset{n\mathrm{}}{lim}(T_{\varphi _q})^n`$ (46)
$`=`$ $`\underset{n\mathrm{}}{lim}[p(1+ad(\alpha A_q^{}+\alpha ^{}a_q)+{\displaystyle \frac{1}{2}}adad(\alpha A_q^{}+\alpha ^{}a_q)+\mathrm{})`$
$`+`$ $`(1p)(1+ad(\alpha A_q^{}\alpha ^{}a_q)+{\displaystyle \frac{1}{2}}adad(\alpha A_q^{}\alpha ^{}a_q)+\mathrm{}]^n.`$
If we introduce the parameters $`t๐`$ and $`c,\gamma ๐`$ by means of the same relations (24) as in the $`q=1`$ case, then we will obtain the limiting $`q`$-transition operator $`T_t^q=e^{tad_q}`$, where $`_q=cA_q^{}+c^{}a_q+\gamma (A_q^{})^2\gamma ^{}a_q^2\left|\gamma \right|(A_q^{}a_q+a_qA_q^{})`$.
To simplify this $`q`$-master equation we will assume as in the undeformed case that the parameter $`\gamma `$ is a complex variable with random argument of zero average and constant non zero magnitude. Then if we average over random $`\gamma `$ the equation of motion then only terms proportional to the amplitude of $`\gamma `$ will be retained. If in addition we consider the case of an symmetric random walk i.e $`p=1/2,c=0`$ the equation of motion becomes
$$\frac{d}{dt}f_t=2\left|\gamma \right|\left[A_q^{}f_ta+af_tA_q^{}Nf_tf_t(N+1)\right].$$
(47)
This is a $`q`$-quantum master equation of the Lindblad type which will be shown to admit a solution in terms of a operator valued Appell system associated with the generator of that equation. We may introduce as in the preceding undeformed case the following operators
$$K_+f=A_q^{}fa_qK_{}f=a_qfA_q^{}K_0f=\frac{1}{2}(A_q^{}a_qf+fa_qA_q^{}),$$
(48)
and $`K_cf=[A_q^{}a_q,f]`$. These operators acting on the elements $`f`$ of the enveloping algebra $`๐ฐ_q(hw)`$, generate the $`su(1,1)`$ Lie algebra defined as in eq. (29). In terms of these operators the $`q`$-quantum master equation (47) is cast in the form
$$\frac{d}{dt}f_t=2\left|\gamma \right|(2K_0+K_++K_{})f_t.$$
(49)
Use of the disentangling theorem (Baker-Campbell-Hausdorff formula) of a general $`SU(1,1)`$ group element (c.f Appendix A), allows to express the solution of the quantum $`q`$-master equation in the form
$$\widehat{f}_t=\mathrm{exp}(A_+K_+)\mathrm{exp}(\mathrm{ln}A_0K_0)\mathrm{exp}(A_{}K_{})(\widehat{f})=\mathrm{exp}(B_{}K_{})\mathrm{exp}(\mathrm{ln}B_0K_0)\mathrm{exp}(B_+K_+)(\widehat{f}),$$
(50)
if the normally or respectively the antinormally ordered BCH decomposition is used. Above we choose $`f=_{n0}c_na_q^n`$, to stand for the initial time operator which can be a general element of the subalgebra of $`๐ฐ_q(hw)`$ that is generated by the $`q`$-annihilation operator. Specifically in the case of normally ordered decomposition with initial operator taken as $`f=a_q^t`$ the solution of the quantum $`q`$-master equation is obtained by means of the actions issued in eq.(48). A straightforward calculation yields the solution:
$`f_t=\mathrm{exp}(A_+K_+)\mathrm{exp}(\mathrm{ln}A_0K_0)\mathrm{exp}(A_{}K_{})(a^t)={\displaystyle \underset{k0}{}}{\displaystyle \underset{l0}{}}{\displaystyle \underset{m0}{}}{\displaystyle \underset{r=0}{\overset{l}{}}}\times `$
$`{\displaystyle \frac{A_{}^k}{k!}}{\displaystyle \frac{\overline{A}_0^l}{l!}}{\displaystyle \frac{A_+^m}{m!}}{\displaystyle \frac{1}{2^m}}\left(\begin{array}{c}l\\ r\end{array}\right)(a^{})_q^{k+m}(N+k)^{lr}(N+k+t+1)^ra_q^{k+t+m},`$ (53)
where the $`A`$โs have the same values as before. A similar solution can be obtained for the antinormal BCH decomposition. We can therefore state the results in the following
Proposition 2. The solution of the quantum $`q`$-master equation $`\frac{d}{dt}f_t=_qf_t`$ where the operator $`_q(f_t)=2\left|\gamma \right|\left[A_q^{}f_ta_q+a_qf_tA_q^{}Nf_tf_t(N+1)\right]`$ of Lindblad type generates the semigroup of $`q`$-Markov transition operators $`T_t^q=e^{t_q}`$ acting on the enveloping algebra $`๐ฐ_q(hw)`$, is given by the associated $`a_q^{}a_q`$-valued Appell system which is given by equation (53).
We note also that the dual $`q`$-master equation satisfied by the density operator can easily be solved along the above lines in terms of the associated Appell system.
7. Discussion. A novel way for constructing quantum master equations has been provided with solutions given by certain sets of operator valued functions that constitute a generalization of the concept of classical Appell polynomial. This entire approach is algebraic and utilizes concepts and tools from the powerfully structure of Hopf algebra. The choice of the dual partner of that algebra structure, namely the coherent states and their adjoint density operators, offers a chance to investigate random walks on non trivial geometries.
The prospect of such a framework is rich enough to allow for random walks constructed on e.g non commuting spaces with braided/smash structure or on Lie groups, quantum groups and quantum modules and comodules. The kinds of Appell systems resulting in those cases might provide new challences to the theory of Special Functions. Some of these issues will be taken up in a forthcoming communication.
Appendix A. The disentangling theorem (Baker-Campbell-Hausdorff formula) of a general $`SU(1,1)`$ group element $`g(a_+,a_0,a_{})`$ in the normal $`\{K_+^aK_0^bK_{}^c:a,b,c๐_+\}`$, and antinormal $`\{K_{}^aK_0^bK_+^c:a,b,c๐_+\}`$ ordering of the generators of the enveloping algebra $`๐ฐ(su(1,1))`$ reads respectively:
$`g(a_+,a_0,a_{})`$ $`=`$ $`\mathrm{exp}(\alpha _+K_++\alpha _0K_0+\alpha _{}K_{})`$ (54)
$`=`$ $`\mathrm{exp}(A_+K_+)\mathrm{exp}(\mathrm{ln}A_0K_0)\mathrm{exp}(A_{}K_{}),`$
$`=`$ $`\mathrm{exp}(B_{}K_{})\mathrm{exp}(\mathrm{ln}B_0K_0)\mathrm{exp}(B_+K_+),`$
where $`A_\pm (a_0)=\frac{(a_\pm /\varphi )\mathrm{sinh}\varphi }{\mathrm{cosh}\varphi (a_0/2\varphi )\mathrm{sinh}\varphi }`$, $`A_0=(\mathrm{cosh}\varphi (a_0/2\varphi )\mathrm{sinh}\varphi )^2`$ and $`B_\pm (a_0)=A_\pm (a_0)`$, $`B_0=(\mathrm{cosh}\varphi +(a_0/2\varphi )\mathrm{sinh}\varphi )^2`$, with $`\varphi ^2=((\alpha _0/2)^2a_+a_{})`$. The relations between the two types of ordered decompositions is based on the formulae $`A_\pm =\frac{B_0B_\pm }{1B_0B_+B_{}}`$, $`A_0=\frac{B_0}{(1B_0B_+B_{})^2}`$, and $`B_\pm =\frac{A_\pm }{A_0A_+A_{}}`$, $`B_0=1/A_0(A_0A_+A_{})^2`$.
Appendix B. Relations among ordered basic monomials of the enveloping algebra $`๐ฐ(hw)`$. From antinormal to normal ordering:
$$a^i(a^{})^j=\underset{l=0}{\overset{min(i,j)}{}}d_{i,j}^l(a^{})^{jl}a^{jl}=\underset{l=0}{\overset{min(i,j)}{}}l!\left(\begin{array}{c}i\\ l\end{array}\right)\left(\begin{array}{c}j\\ l\end{array}\right)(a^{})^{jl}a^{jl},$$
(55)
From number operator to normal ordering:
$$N^k=\underset{l=1}{\overset{k}{}}c_{k,l}(a^{})^la^l,$$
(56)
where $`\overline{c}_{k+1,l}=\overline{c}_{k,l1}+l\overline{c}_{k,l}`$, and these coefficients are recognized as the Stirling numbers of second kind.
From number operator to antinormal ordering:
$$N^k=\underset{l=1}{\overset{k}{}}\overline{d}_{k,l}a^l(a^{})^l,$$
(57)
where $`\overline{d}_{k+1,l}=\overline{d}_{k,l1}(l+1)\overline{d}_{k,l}`$, with $`\overline{d}_{0,0}=1`$.
Relations among ordered basic monomials of the enveloping algebra $`๐ฐ_q(hw)`$. From antinormal to normal ordering:
$$a_q^i(a^{})_q^j=\underset{l=0}{\overset{min(i,j)}{}}\overline{b}_{i,j}^l(a^{})_q^{jl}a_q^{jl}=\underset{l=0}{\overset{min(i,j)}{}}q^{l(lij)+ij}[l]!\left[\begin{array}{c}i\\ l\end{array}\right]_q\left[\begin{array}{c}j\\ l\end{array}\right]_q(a^{})_q^{jl}a_q^{jl}.$$
(58)
We note that for $`q1`$ the $`\overline{b}_{i,j}^ld_{i,j}^l`$. From normal to antinormal ordering:
$$(a^{})^ia^j=\underset{l=0}{\overset{min(i,j)}{}}b_{i,j}^la^{jl}(a^{})^{il}=\underset{l=0}{\overset{min(i,j)}{}}()^lq^{l(lij)ij)}[l]!\left[\begin{array}{c}i\\ l\end{array}\right]_q\left[\begin{array}{c}j\\ l\end{array}\right]_qa^{jl}(a^{})^{il}.$$
(59)
Acknowledgement. Discussions with I. Tsohantjis are gratefully acknowledged. |
warning/0001/quant-ph0001119.html | ar5iv | text | # Quantum tunneling dynamics using hydrodynamic trajectories.
## I Introduction
Trajectory based constructions of quantum behavior are ubiquitous throughout quantum physics. In the semi-classical limit, quantum dynamics is approximated by classical equations of motion whereby the transition amplitudes and wavefunctions are computed using the classical action connecting the initial and final states. Hydrodynamic constructions date back to early work by de Broglieand Madelung \[ref5a\] later by Bohm. The so called quantum trajectories arising from this formalism have been the subject of a large number of ontological and philosophical papers seeking a causal interpretation of quantum mechanics. While a comprehensive overview of these works is far beyond the scope of this paper, Hollandโs book provides perhaps the most comprehensive technical overview of the approach and contains many examples of how the formalism can be applied in a wide variety of cases. In short, the quantum trajectories themselves are relatively easy to compute once one has obtained a wavefunction solution of Schrรถdinger equation, $`\psi `$. The velocity of a trajectory at a given point in space-time is computed as
$`v(x,t)={\displaystyle \frac{j(x,t)}{\rho (x,t)}}`$ (1)
where $`j(x,t)`$ is the quantum current, and $`\rho =|\psi |^2`$. In this โpilot waveโ approach, $`\psi `$ acts as a guiding field for the trajectories. This approach is useful in that once the wave function has been obtained, is generally easy to compute the trajectories. While there are a number of papers which have computed quantum trajectories having obtained the wave function, relatively little work has been done in developing computational methods based upon the description which does not rely upon first constructing the wave function. However, Wyatt and co-workers have recently described a mesh-less finite-element method for integrating the de Broglie-Bohm equations using Lagrangian hydrodynamic trajectories. Similar methods are widely used in computational fluid dynamics (CFD) to simulate fluid flow dynamics in porous media (such as an oil reservoir) and other systems with complex topologies. Wyattโs method represents the quantum density using a cloud of Lagrangian fluid points which themselves evolve according to the de Broglie-Bohm hydrodynamic equations of motion. The method features a moving weighted least-squares (MWLS) approach to compute the various derivatives and gradients required by the de Broglie-Bohm equations.
In this work, we report on our implementation of the MWLS methodology and present an assessment of its difficulties and where improvements can be made. Previous applications of the approach have focused entirely upon reactive scattering type calculations. While this class of problems has its own set of associated difficulties, we have elected to focus upon the dynamics of systems trapped in various one dimensional potential wells. These problems, while perhaps too idealistic, allow one to compare trajectories and results obtained via a โhydrodynamicโ calculation to those obtained via analytical or numerical solution of the time-dependent Schrodinger equation. Specifically, we first examine the dynamics of harmonic oscillators and then move onto a more challenging problem of tunneling in double well potential. Our results indicate that the methodology reproduces results obtained via more traditional grid based approaches; however, we note that the long time stability of the methodology needs to be improved before it can be used as an alternative to wave-packet based calculation.
## II Quantum Equations of Motion
The de Broglie-Bohm equations are derived directly from the time dependent Schrรถdinger equation. The derivation is initialized by writing the wave function in polar form:
$`\psi (x,t)=R(x,t)e^{iS(x,t)/\mathrm{}}.`$ (2)
When this substituted into the Schrรถdinger equation, the real and imaginary components can be collected into a continuity equation:
$`_t\rho (x,t)+{\displaystyle \frac{1}{m}}(\rho S)=`$ (3)
and the quantum Hamilton-Jacobi equation:
$`_tS+{\displaystyle \frac{|S|^2}{2m}}+V+Q=0`$ (4)
where $`Q`$ is the so called quantum potential which is non-local and arises from the quantum kinetic energy operator in the Schrรถdinger equation.
$`Q(x,t)={\displaystyle \frac{\mathrm{}^2}{2m}}{\displaystyle \frac{1}{\sqrt{\rho }}}^2\sqrt{\rho }`$ (5)
Taking the gradient of Eq.4, one obtains the equations of motion for the trajectories:
$`D_tv(x,t)`$ $`=`$ $`{\displaystyle \frac{1}{m}}(V(x(t))+Q(x(t)))`$ (6)
where we define $`x(t)`$ and $`v(x(t))=S/m`$ as the trajectory and velocity of Lagrangian fluid elements. The notation $`D_tf`$ denotes the โmaterialโ or โconvectiveโ derivative of the function $`f`$
$`D_tf=\left(_t+v\right)f`$ (7)
which gives the rate of change of $`f((x(t))`$ as observed while moving with the particle along the trajectory, $`x(t)`$. Hence in the Lagrangian picture of fluid mechanics, we โgo with the flowโ. The equation of continuity can also be written in terms of the convective derivative:
$`D_t\rho +(v)\rho =0`$ (8)
From this we can deduce a short time propagator for $`\rho `$:
$`\rho (x,t+\delta t)=e^{v\delta t}\rho (x,t)`$ (9)
which is local in space and is dependent upon the divergence of the velocity field at the point $`x`$. To apply Eq. 6 and Eq. 9 in a computational scheme, we first discretize the system into an ensemble of Lagrangian fluid elements labeled by their position vectors, $`x`$. These elements carry information regarding the local density about that point and the phase is determined by integrating Eq. 5 along the path. From these two bits of information, the quantum wavefunction can be reconstructed by writing
$`\psi (x_i,t)=\sqrt{\rho (x_i,0)}\mathrm{exp}\left({\displaystyle \frac{i}{\mathrm{}}}{\displaystyle _0^t}L(x_i,v_i,\tau )๐\tau +{\displaystyle \frac{i}{\mathrm{}}}S(x,0)\right)\mathrm{exp}\left({\displaystyle \frac{1}{2}}{\displaystyle _0^t}v(\tau )๐\tau \right)`$ (10)
where
$`L(x,v,t)={\displaystyle \frac{1}{2}}mv^2VQ`$ (11)
is the quantum Lagrangian and $`S(x,0)`$ represents any initial spatial dependent phase present in the initial wave function, such as $`S(x,0)=\mathrm{}kx`$.
In order to compute expectation values we need to be able to integrate over the spatial domain spanned by either $`\rho `$ or $`\psi `$. If we require that each trajectory element carry a volume element, $`dx(t)`$, then by normalization
$`1={\displaystyle ๐x(0)\rho (x,0)}={\displaystyle \underset{i=1}{\overset{N}{}}}\rho _i(0)dx_i(0)={\displaystyle \underset{i=1}{\overset{N}{}}}\rho _i(t)dx_i(t),`$ (12)
where $`\rho _i(t)=\rho (x_i(t))`$ is the density carried by the $`i`$th trajectory.
To compute $`dx_i(t)`$ we need the Jacobian which transforms the volume element $`dx_i(0)`$ to the volume element, $`dx_i(t)`$ some time later along trajectory $`x_i(t)`$.
$`J_i={\displaystyle \frac{x_i(0)}{x_i(t)}}.`$ (13)
Taking the material time derivative of $`J`$ yields along each trajectory:
$`D_t\mathrm{log}J_i=v_i`$ (14)
Thus, an amplitude element of the wavefunction is propagated along a trajectory $`x_i(t)`$ as
$`dx_i(t)\psi _i(t)`$ $`=`$ $`dx_i(0)\sqrt{\rho _i(0)}\mathrm{exp}\left({\displaystyle \frac{i}{\mathrm{}}}{\displaystyle _0^t}L(x_i,\dot{x}_i,\tau )๐\tau +{\displaystyle \frac{i}{\mathrm{}}}S(x_i,0)\right)`$ (15)
$`\times `$ $`\mathrm{exp}\left({\displaystyle _0^t}v_i(\tau )๐\tau \right).`$ (16)
A typical propagation cycle consists of computing the quantum potential from the density at each fluid element, taking the gradient of $`Q`$ and $`V`$ to compute the acceleration of the fluid element and computing divergence of the velocity field to determine how the density in given fluid element changes over one discrete time interval. Thus, the trajectories, density, and velocity fields contain all of the essential information required to construct the full quantum mechanical wave function.
## III Moving Least Squares Approximation
Since our goal is to be able to compute the quantum trajectories without computing the wavefunction a priori, we need to be able to be able to discretize the system into a set of Lagrangian fluid elements, compute the various derivatives required to compute the equations of motion and finally, advance the fluid elements to a new set of coordinates. This we accomplish through the use of an adaptive meshless-cloud method described recently by Wyatt . The method itself is described in detail by Liszka and co-workers and is similar to methods used extensively in computational fluid dynamics. First, we review the finite element approximation (FEA) and its implementation in a movable weighted least-squares scheme (MWLS).
In the moving least squares approximation, we assume that we have a function, $`g=g(x)`$ defined over a finite set of points and we seek an approximate value of $`g`$ at one of these points, $`x_o`$, based upon the values at neighboring points. We assume that the function we seek is smooth enough to be expanded about the point $`x_o`$ in a finite polynomial basis
$$p=\{x,\frac{x^2}{2},\mathrm{}\frac{x^{n_p}}{n_p!},\}$$
(17)
Thus, we can write
$`f(x)={\displaystyle \underset{i=1}{\overset{n_p}{}}}a_ip_i(xx_o)+f(x_o)`$ (18)
where $`a_i`$ is a vector of coefficients. For a simple polynomial as above, the derivatives of $`f`$ at $`x_o`$ are then just the coefficients themselves. If we have a data point at $`x_o`$ and we want the approximation to pass though the remaining data points, we can write a set of least-squares equations
$`f_j={\displaystyle \underset{i=1}{\overset{n_p}{}}}a_ip_i(x_jx_o),`$ (19)
where $`f_j=f(x_j)f(x_o)`$, $`p_i(x_jx_o)`$ is the $`i`$th polynomial basis member evaluated at $`x_jx_o`$, and $`a_i`$ are the expansion coefficients. In other words, we seek a vector $`a`$ which satisfies the linear equation:
$`f=Pa`$ (20)
where $`f`$, $`P`$, and $`a`$ are the terms in Eq. 19 written in matrix/vector form.
In order to solve for $`a`$, we need at least as many data points in the neighborhood about $`x_o`$ as we have terms in the polynomial expansion. Rather than trying to pick out exactly the right number of points, the most efficient procedure is to simply select more data points than basis functions. Since the system of equations is overdetermined, $`P`$, is a rectangular matrix and we use must either singular value decomposition or other pseudo-inversion method to invert $`P`$. Furthmore, we can improve the stability of the least squares procedure by weighting each data point so that points farther away from the central point receive the least weight. Thus, for the โweightedโ least squares procedure we solve
$`f^\mathrm{\Omega }=P^\mathrm{\Omega }a`$ (21)
where $`f_j^\mathrm{\Omega }=(f_jf(x_o))\omega _j`$, $`P_{ij}^\mathrm{\Omega }=\omega _jP_{ij}`$, $`a`$ is a vector of undetermined coefficients, and $`\omega _j`$ is the weight assigned to point $`x_j`$.
We have found that the convergence of the least-squares procedure is greatly improved by using a logarithmic form of the density,
$`\rho =e^{g(x)},`$ (22)
where $`g`$ is a polynomial of order $`n_p`$. This representation is a useful way to transform a non-linear model to a linear one. For example, in computing the quantum potential we can write
$`{\displaystyle \frac{1}{\sqrt{\rho }}}^2\sqrt{\rho }=e^{g/2}^2e^{g/2}`$ (23)
Working through the derivative terms yields:
$`e^{g/2}^2e^{g/2}={\displaystyle \frac{1}{2}}\left(g_{,\mu \mu }+{\displaystyle \frac{1}{2}}g_{,\mu }^2\right)`$ (24)
where the โcommaโ delimited subscripts in $`g_{,u}=_\mu g`$ denote taking derivative with respect to coordinate $`\mu `$.
Finally we write the equations of motion for the velocity in terms of the derivatives of the various fields in component form:
$`ma_\mu (x)`$ $`=`$ $`_\mu \left(V(x){\displaystyle \frac{\mathrm{}^2}{4m}}\left(g_{,\nu \nu }+{\displaystyle \frac{1}{2}}g_{,\nu }^2\right)\right)`$ (25)
$`\rho (x,t+\delta t)`$ $`=`$ $`e^{_\mu v_\mu (x)\delta t}\rho (x,t).`$ (26)
Some final notes regarding our implementation of the MWLS scheme are also in order. We propagate trajectories using the sympletic Verlet algorithm . Convergence with respect to time step is checked by comparing the results of two 1/2 time steps to one full time step. If the differences between position, energy, velocity, or $`\rho `$ computed via the two separate procedures varied by more than $`1:10^6`$, the shorter time step results were taken, $`\delta t`$ was reduced by a factor of 0.75 and propagation continued. However, if the longer time step results were sufficiently accurate, $`\delta t`$ was increased by a factor of 2 for the next iteration cycle. For the least squares fitting procedures, we found that a Hermite polynomial basis provided the most robust basis as measured by the $`\chi ^2`$ for each fit. We also weighted the fits using a Gaussian weighting function centered about $`x_o`$ adjusted so that the point farthest from $`x_o`$ received a weight of 0.01. For the one dimensional results presented here, our code reproduces results obtained via Wyattโs MWLS particle code described in Ref.
## IV Example Calculations
### A Harmonic Oscillator
As a primary example, we consider the quantum trajectories for a harmonic oscillator and compare these to the classical trajectories starting from identical starting points. In Fig. 1 we show the time evolution of an ensemble of quantum trajectories for a harmonic system with period of $`\tau =2\pi /\omega =888.57a.u.`$ Unless otherwise noted explicitly, we shall use atomic units from here on, setting $`\mathrm{}=1`$. The first two cases (A & B) are the quantum trajectories for a particle with mass $`m=2000a.u`$ and with mass $`m=200a.u.`$ respectively, each starting with identical initial quantum densities
$`\rho (x)=e^{\beta (xx_o)^2}`$ (27)
with $`x_o=3.0`$ and $`\beta =0.3`$.
In each of the cases shown here, we used 100 points spaced evenly about the center of the initial density. In fourth plot (D) we show the trajectories obtain from a purely classical calculation involving the same 100 points. The effect of the quantum potential is clearly seen as the trajectories traverse the minimum of the well at $`x=0`$. For the heavier mass (in A), the trajectories tend to focus at the well minimum, just as in the classical calculation (in D). However, the trajectories do not cross each other as do the classical trajectories. This is a crucial characteristic of Bohmian trajectories. In essence, the density must be single valued and the Jacobian of the volume element carried by each point must remain positive definite for all times. Hence, if any pair of trajectories were to cross, these conditions would be violated. For case B, in which $`m=200a.u`$ , the quantum potential is $`10\times `$ stronger than in case A and hence the trajectories become more diffuse as the particle traverses the bottom of the well. Interestingly enough, if we instead choose $`\beta =1/m\omega `$, the trajectory lines remain evenly spaced throughout the calculation as shown in C.
This last result can be understood from the fact that the wavepacket used in Fig. 1.C is a coherent state and hence the time evolution of the density is given by
$`\rho (x,t)=e^{m\omega (xx_o\mathrm{cos}(\omega t))^2}`$ (28)
where $`x_o`$ is the original centroid of the wavepacket. The quantum potential can be easily derived from Eq. 26 and the action along any given Bohmian trajectory is given by
$`S(x,t)={\displaystyle \frac{1}{2}}\mathrm{}\omega t{\displaystyle \frac{1}{2}}m\omega \left(2xx_o\mathrm{sin}(\omega t){\displaystyle \frac{1}{2}}x_o^2\mathrm{sin}(2\omega t)\right).`$ (29)
Thus, each trajectory is a solution of
$`m\dot{x}=_xS=m\omega x_o\mathrm{sin}\omega t.`$ (30)
and belongs to the family of cosine curves of period $`\omega `$:
$`x_i(t)=x_i(0)+x_o(\mathrm{cos}(\omega t)1).`$ (31)
I.e. the Bohmian particle undergoes simple harmonic motion of amplitude $`x_o`$ about the point $`x_i(0)x_o`$. Furthermore, the energy along a given Bohmian trajectory is computed by taking the time derivative of the action (Eq. 29):
$`E(x,t)=_tS={\displaystyle \frac{\mathrm{}\omega }{2}}{\displaystyle \frac{1}{2}}m\omega ^2(2xx_o\mathrm{cos}(t\omega )x_o^2\mathrm{cos}(2t\omega )).`$ (32)
Note that this is not constant in time as would be expected for a purely classical system. The quantum force thus acts as a time dependent driving force on the ensemble of trajectories. If we were to have chosen the initial center of the density to be at exactly $`x=0`$, then $`E=\mathrm{}\omega /2`$ and the quantum force would exactly counter-balance the classical potential force. In essence, the particles would remain fixed in their spatial positions.
In the case of the quantum harmonic oscillator, the quantum trajectories computed using the MWLS procedure agree with the exact analytical results one can obtain for this system. Furthermore, the procedure appears to be stable for long periods of time. We have carred out calculations up to 10 oscillation periods without considerable accumulation of error. We now move on to the more formidable problem of tunneling in a double well potential.
### B Quantum Tunneling in a Double Well Potential
Tunneling and barrier penetration represent phenomena which are unique to quantum mechanics. It is well understood from classical mechanics that if a particle lacks sufficient energy to surmount a potential barrier, the particle will be reflected and the barrier is impenetrable. However in quantum mechanics, a particle can โpass throughโ a barrier. This can be rationalized in a variety of ways. Perhaps the most consistent is that if the particle is in brought to a halt at the classical turning point, both its position and momentum are defined to exact precision, which would violate the uncertainty principle. In the quantum trajectory approach, the quantum potential acts to enforce this behavior by effectively โloweringโ the barrier to permit trajectories to penetrate through the barrier and raises the barrier once sufficient amplitude has passed. To see how this comes about in the context of a particle based approach, let us consider tunneling through the double well potential:
$`V(x)=ax^4bx^2`$ (33)
In this example, we take $`a=0.007`$ and $`b=0.01`$ in atomic units, which gives two tunneling states below the barrier at $`E=369.827\mathrm{cm}^1`$ and -313.918 $`\mathrm{cm}^1`$ and a barrier height of $`V_b=786.24\mathrm{cm}^1`$. The initial Gaussian density was centered about the center of the right side well at $`x_o=\sqrt{b/2a}`$ with $`\beta =\sqrt{4bm}`$ corresponding to a harmonic approximation to a state localized in the right hand well. In the lower frame of Fig. 2 we show quantum trajectories for the initial crossing of the wavefunction from the right to the left side of the well and in Fig. 3 we show snapshots of the quantum density, the quantum potential, $`V+Q`$, and $`V(x)`$ for various times. The trajectories cross from $`+x`$ to $`x`$ corresponding to top to bottom in Fig 2. The initial expansion of the paths is primarily due to the expansion of the wavefunction in the well. Since the initial state is not in a stationary state of the potential, there is a net force on the particles due to the difference between the $`Q^{}`$ and $`V^{}`$. Once the particles have passed through the barrier region (at $`x=0`$), they encounter the repulsive potential wall and are reflected. Note that the particles penetrate fairly deeply into the repulsive region. However, they carry very little net density. Unfortunately as far as our calculations are concerned, we run into numerical difficulties in computing the quantum potential in this region. As a result, the congruency condition fails to hold and trajectories begin to cross (after $`t=650`$) and our results are meaningless beyond this point.
However, prior to this point in time, we can compute the effective barrier height due to the quantum potential at $`x=0`$. In Fig. 4 we plot the effective barrier height, $`V_{eff}=Q(0)+V_b`$, as a function of time. As indicated above, the quantum potential from the initial Gaussian distribution effectively โlowersโ the barrier allowing trajectories to pass from the right to the left. However, once the initial expansion of the density is accomplished, the barrier begins to rise and the effective barrier height appears to approach an asymptotic value.
If we assume that the lowest two eigenstates of the double well can be approximated as symmetric and antisymmetric linear combinations of ground state harmonic oscillators centered about the left and right minima in the well, we can compute the approximate quantum potential at $`x=0`$ quite easily. Let us write the approximate wavefunction as (neglecting a common phase factor),
$`\psi (x,t)={\displaystyle \frac{1}{\sqrt{2}}}(e^{i\omega t}\varphi _+(x)+e^{+i\omega t}\varphi _{}(x)),`$ (34)
where $`\varphi _\pm (x)`$ are the symmetric and anti-symmetric tunneling states split by $`\mathrm{}\omega =(E_{}E_+)/2`$. Taking,
$`\varphi _\pm (x){\displaystyle \frac{1}{\sqrt{2}}}(\varphi _R(x)\pm \varphi _L(x)),`$ (35)
and using the approximation
$`\varphi _{R,L}(x)=\left({\displaystyle \frac{m\omega }{\pi \mathrm{}}}\right)^{1/4}e^{m\omega _o(x\pm x_o)^2/2\mathrm{}},`$ (36)
the quantum potential at the barrier is given by
$`Q(0)=\left({\displaystyle \frac{\mathrm{}\omega _o}{2}}+{\displaystyle \frac{1}{4}}m\omega _o^2x_o^2(\mathrm{cos}(4t\omega )3)\right)`$ (37)
Thus, the effective barrier separating the right and left sides of the system is at its minimum when $`\rho `$ is localized in the right or left hand well and greatest when $`\rho `$delocalized in a 50:50 mixture of right and left hand states. Moreover, as the separation between the respective wells increases, the harmonic oscillator approximation to the lowest eigenstates becomes better and better. Correspondingly, the quantum potential becomes the parabola,
$`Q(x){\displaystyle \frac{\mathrm{}\omega _o}{2}}{\displaystyle \frac{1}{2}}m\omega _o^2(x\pm x_o)^2.`$ (38)
Since the quantum force and the classical potential force will be nearly equal and opposite, the particles will remain almost motionless in the initial well.
## V Pilot wave trajectories
As noted above, the trajectories themselves can be obtained by propagating a solution to the time-dependent Schrรถdinger equation. In this โpilot waveโ scheme, the quantum potential and equations of motion for the Bohmian particles may be computed directly from the wave function itself. Alternatively, if we have an accurate representation of $`\psi `$ and can compute $`v[\psi ]`$ via Eq. 1, we can obtain what should be a set of โexactโ Bohmian trajectories associated with the quantum wavefunction. To compute such trajectories for the double well example shown above, we used a discrete variable representation (DVR) of Gauss-Tchebychev quadrature points to represent both the wave packet and the time-evolution operator on a finite spatial grid of 200 quadrature points spanning 2.5 Bohr on either side of the barrier. For dynamics at low total energy, this representation is certainly more than adequate as we were able to converge the lowest 20 eigenstates of this well to $`1:10^6`$. Because the instantaneous position of a Bohmian particle at $`x(t)`$ will not generally correspond to the position of a quadrature point, the velocity can not be computed directly from the DVR wavefunction, which is only defined on the quadrature points them selves. Rather, we transform $`\psi `$ from the DVR to a finite basis representation (FBR) using the unitary transformation,
$`\psi ^{FBR}=U\psi ^{DVR},`$ (39)
where $`U`$ is the unitary transformation associated with a $`N`$-point Gauss-Tchebychev DVR.
$`U_{ij}=\sqrt{{\displaystyle \frac{2}{N+1}}}\mathrm{sin}\left(ij{\displaystyle \frac{\pi }{N+1}}\right).`$ (40)
The elements of the vector $`\psi ^{FBR}`$ are the expansion coefficients of the wavefunction in a finite polynomial basis.
$`\psi (x,t)={\displaystyle \underset{i=1}{\overset{n}{}}}\psi _i^{FBR}(t)T_i(x).`$ (41)
For the Gauss-Tchebychev quadrature scheme used here, basis functions are:
$`T_i(x)=\sqrt{{\displaystyle \frac{2}{\mathrm{\Delta }}}}\mathrm{sin}(i\pi (x\mathrm{\Delta }/2)/\mathrm{\Delta }).`$ (42)
Eq. 37 gives the wavefunction at any point $`x[\mathrm{\Delta }/2,\mathrm{\Delta }/2]`$ and we can use this information to compute the velocity of a particle guided by $`\psi `$. In the top frame of Fig. 2 we show the Bohm trajectories for the double well system choosing identical conditions as above, this time computed via the DVR pilot wave based procedure. For clarity we show only those trajectories for particles starting to the left of the initial centroid of the density. At long times, the trajectories transmitted to the left hand well tend to bunch together. This is due to the fact that the finite basis functions in Eq. 38 go to zero at $`x=\pm \mathrm{\Delta }/2`$ and force the system to have an artificial node at the boundary points. Consequently, the quantum potential is infinitely repulsive at the end of the grid and all particle trajectories are repelled from these regions. For practical purposes, the density carried by these trajectories is negligible and the overall effect on the wavepacket dynamics is minimal.
Notice, however, that between $`t=600`$ and $`t=800`$, a number of trajectories near the barrier at $`x=0`$ are deflected from each other. This is due to the fact that $`\rho `$ develops oscillatory structure due to constructive and destructive interferences between on coming and reflected components of the wavefunction near the barrier. The density eventually develops a node at about $`t=900`$ a.u. Surprisingly, and disturbingly, this behavior is absent in the MWLS trajectories. In Figure 5 and Figure 6 we compare the evolution of $`\rho `$ using the DVR $`(\rho ^{DVR})`$ and MWLS $`(\rho ^{MWLS})`$ methods. Initially, the agreement is quite good. At longer times the agreement is very poor with $`\rho ^{MWLS}`$ even โlooping backโ through itself. At intermediate times, as $`\rho `$ is reflected by the repulsive barrier in the $`x`$ half of the double well, oscillations develop in $`\rho ^{DVR}`$ whereas no such oscillations are seen in $`\rho ^{MWLS}`$. Since the quantum potential is a measure of the local curvature of $`\rho `$, these oscillations translate into attractive and repulsive regions in the quantum potential. Eventually as $`\rho ^{DVR}`$ goes on to form nodes (at approximately $`t=700`$ au and 900 au trajectories are guided away from such regions via the quantum force. This accounts for the significant deflections in the DVR trajectories but does not tell us why this fails to occur in the MWLS case.
The answer is basically that of sampling resolution. As the MWLS trajectories evolve, they by in large tend to spread apart. This restricts our ability to resolve any structure in $`\rho ^{MWLS}`$ which occurs on lengthscales finer than the distance of separation of the MWLS trajectories at a given point in time. Hence, the oscillatory structure seen in the DVR calculations (e.g. at $`t`$=550 au) are too fine to be resolved by the MWLS trajectories. Thus, the quantum potential will be too smooth in this region. The DVR calculations are also band width limited due to the finite spacing of the DVR points. However, in the DVR grid used in these examples, this spacing is fixed at $`\delta x=0.049`$ Bohr, where as in the MWLS case the spacing between trajectories in the region of the barrier at $`t=550`$ is $`0.10`$ Bohr, which slightly courser than the structure seen in $`\rho ^{DVR}`$ in this region. Consequently, the MWLS trajectories lose band width resolution precisely where it is needed the most in this case.
Let us compare the DVR and MWLS based trajectories on a head to head basis. Four sets of trajectories extracted from Fig. 2 are shown in Figure 7. First, for trajectories originating near the leading edge of the MWLS grid, we expect that the two results should deviate rather early due to the fact that the fitting procedure used to compute the quantum potential using the MWLS is least accurate near the end of the grid (Trajectory # 9). Surprisingly, however, the agreement is quite good. In fact, it is appearent that the deviatiation is more likely due to the artificial boundary conditions imposed by the basis function used to represent the quantum wave packet in the DVR calculations. For trajectories originating near the maximum of the initial wavepacket (Trajectory #50), we see quantative agreement between DVR and MWLS based trajectories since $`\rho `$ is smooth function of $`x`$ in this region and remains so over the course of both calculations. However, let us compare the two trajectories labeled #38 and #39 from each calculation. These two trajectories originate side by side separated by $`\delta x=0.02Bohr`$. In the DVR calculation, these trajectories are deflected by the oscillations in $`\rho `$ with #38 deflected slightly towards the $`x`$ direction and #39 in the $`+x`$ direction. These deflections are not observed in the corresponding MWLS trajectories.
## VI Discussion
In this paper, we present our implementation and computational analysis of the de Broglie-Bohm hydrodynamic equations using a particle based description. The formalism itself allows an elegant interpretation of quantum dynamics in geometric terms. Instead of wave functions, we have geometric ray lines. Surprisingly, this intuitive connection between quantum wave mechanics and geometry has not made it into the โmain streamโ even though the original seeds for this interpretation can be found at the very advent of quantum theory. However, we do note that this view point is seeing a resurgence in popularity as measured by the number of papers which have used the Bohmian construction in vastly different areas of quantum physics.
What is important, however, to point out that we are only showing the best case scenarios in this paper. We have found that trajectories tend to cross cases in which $`\rho `$ is very small, trajectory lines are far apart, or when the potential is significantly anharmonic. Some of this is exhibited in Figures 2 and 3. This is most certainly a result of numerical instabilities in the MWLS procedure which we have elaborated upon in this paper.
We also run into particular dificulties in dealing with nodal points in the density, in other words, when $`\rho (x)=\psi (x)|^2=0`$. At such points, it is not possible to fit the coefficients of a polynomial expansion using points on either side of the node since the function does not have continuous derivatives at the node. We have tried various computational alternatives to dealing with such points. One alternative is to use only trajectories and densities on one side of the node or the other when computing $`Q(x)`$. Under this approach we compute derivatives by approaching the node from the right and from the left. Another alternative we have tried is to introduce a $`\mathrm{log}(xx_n)`$ basis function for stars which include nodal points. Both of these approaches give very good results for cases in which the initial wavefunction was taken as an odd parity eigenstate in a symmetric well. However, both cases require prior knowledge of the location of the node and that parity be a constant of the motion so that the node does not move. For cases in which parity is not a constant of the motion, nodal points can occur over the natural course of the evolution of the quantum wavefunction, even if the initial wavefunction has no nodes initially. We speculate that a solution may be to construct
$`R_i=\sqrt{\rho _i}e^{i\mu \pi }`$ (43)
along each $`x_i(t)`$ trajectory where $`\mu =0,1,2,\mathrm{}`$ accounts for the branch cut taken when evaluating $`\sqrt{\rho }`$ on either side of a node. In essence, on one side of a node $`R>0`$ and $`\mu =0,2,\mathrm{}`$, while on the other side, $`R<0`$ and $`\mu =1,3,\mathrm{}`$ This would preserve analyticity across the node allowing us to calculate $`Q`$ using a simple polynomial basis. At present, the treatment of nodes and their evolution is an open problem which we will take up in a future work.
###### Acknowledgements.
This work was supported by grants from the Robert A. Welch Foundation and the National Science Foundation. The author also wishes to thank Prof. Bob Wyatt (Univ. of Texas), Prof. Don Kouri (U. Houston) and Dr. Pablo Yepes (Rice) for stimulating discussions. |
warning/0001/math0001148.html | ar5iv | text | # Topological Representations of Posets
## 1. Introduction
Since Stone introduced the topological representation of Boolean algebras there was a lot of attempts to generalize this result: the Stone-like representations of orthopsets by Mayet and Tkadlec , different topological representations of distributive and arbitrary (by Hartonas, Dunn and Urquhart) lattices. We follow the construction introduced in where algebraic dual space $`P^{}`$ is endowed with two closures $`\text{C}_1`$ and $`\text{C}_2`$ in such a way that the collection of all subsets of $`P^{}`$ which are closed in $`\text{C}_1`$ and open in $`\text{C}_2`$ ordered by set inclusion (we denote this collection by $`\text{C}\text{1}\text{O}\text{2}(P^{},\text{C}_1,\text{C}_2)`$) is isomorphic to the initial poset $`P`$:
(1)
$$\text{C}\text{1}\text{O}\text{2}(A,\text{C}_1,\text{C}_2)P$$
The representation (1) of $`P`$ works for arbitrary poset $`P`$. However, for particular classes of posets the โuniversal setโ $`P^{}`$ can be contracted to a smaller one $`AP^{}`$ with the closures $`\text{C}_1`$, $`\text{C}_2`$ induced from $`P^{}`$. In this paper we show that the representations of specific classes of posets mentioned above all have the form
$$\text{C}\text{1}\text{O}\text{2}(A,\text{C}_1,\text{C}_2)P$$
and differ only by the choice of $`AP^{}`$.
### 1.1. Spaces with two closures
Mapping $`\text{C}:\mathrm{exp}()\mathrm{exp}()`$ we call closure if
1. $`A\text{C}(A)`$;
2. $`\text{C}(\text{C}(A))=\text{C}(A)`$;
3. if $`AB`$ then $`\text{C}(A)\text{C}(B)`$.
A set $`A`$ is closed (or C-closed) if $`A=\text{C}(A)`$, $`A`$ is open if $`\overline{A}=A`$ is closed and clopen if it is both closed and open. Note, that any intersection of closed sets is closed, and $`\text{C}(A)`$ is the intersection of all closed sets which contain $`A`$. $`๐ฆ\mathrm{exp}()`$ is called the base of closure C ($`\text{C}=\text{clos }(๐ฆ)`$) if any closed set is an intersection of elements of $`๐ฆ`$.
The closure C is exact if $`\text{C}(\mathrm{})=\mathrm{}`$, and topological if $`\text{C}(AB)=\text{C}(A)\text{C}(B)`$. Note, that exact topological closure defines topology on $``$. For a closure C on $``$ define $`\text{CO}(,\text{C})`$ to be the collection of all clopen subsets of $``$. Obviously $`\text{CO}(,\text{C})`$ ordered by set inclusion is a bounded orthoposet. It was shown by Mayet and Tkadlec , that for an arbitrary bounded orthoposet $`P`$ there is a space $``$ with closure C such that $`P\text{CO}(,\text{C})`$.
If we define two closures on $`\text{C}_1`$ and $`\text{C}_2`$ $``$, then by $`\text{C}\text{1}\text{O}\text{2}(,\text{C}_1,\text{C}_2)`$ we denote the collection of all subsets of $``$ which are both $`\text{C}_1`$-closed and $`\text{C}_2`$-open, ordered by set inclusion. We can say nothing about the structure of $`\text{C}\text{1}\text{O}\text{2}(,\text{C}_1,\text{C}_2)`$ except it is a poset, moreover, as it was shown in for an arbitrary poset $`P`$ one can build a space with two closures such that $`P\text{C}\text{1}\text{O}\text{2}(,\text{C}_1,\text{C}_2)`$.
### 1.2. Algebraic duality for posets
For a poset $`P`$ its algebraic dual space $`P^{}`$ is the set of all isotone mappings from $`P`$ to poset $`\mathrm{๐}=\{0,1\}`$ with $`0<1`$. Here we develop the techniques needed to build the representation.
Consider $`AP^{}`$. A set $`I`$ we call an ideal (with respect to $`A`$ or $`A`$-ideal) if $`I`$ is an intersection of kernels of some mappings $`xA`$ (i.e. $`I=x^1(0)`$). Dually, the intersection of co-kernels $`F`$ we call a filter ($`F=x^1(1)`$).
For $`BP^{}`$ we define an ideal $`I(B)`$ (filter $`F(B)`$) to be the intersection of kernels (co-kernels, respectively) of $`xB`$.
For $`QP`$ define an ideal $`\mathbf{\left[}๐ธ\mathbf{\right)}_A`$ (resp., filter $`\mathbf{\left(}๐ธ\mathbf{\right]}_A`$) โ the intersection of ideals (resp., filters) containing $`Q`$.
Note that ideals with respect to $`P^{}`$ coincide with order ideals ($`I`$ is an order ideal if $`qI`$ and $`pq`$ implies $`pI`$). In general $`A`$-ideals are always order ideals, but the converse is not always true.
We say that $`AP^{}`$ is full if for all $`pq`$ there exists $`xA`$ such that $`x(p)=1`$, $`x(q)=0`$.
$`AP^{}`$ is called separating if for any disjoint ideal $`I`$ and filter $`F`$ there exists $`xA`$ such that $`x|_I=0`$ and $`x|_F=1`$.
In some cases discussed in section 3 $`\mathbf{\left(}๐\mathbf{\right]}_A`$ and $`\mathbf{\left(}๐\mathbf{\right]}_A`$ coincide with lower and upper cones of $`p`$ respectively. Due to the following obvious lemma the separating set is full in this case.
###### Lemma 1.
Let $`A`$ be a separating subset of $`P^{}`$ and $`\mathbf{\left[}๐ฉ\mathbf{\right)}_A\mathbf{\left(}๐ช\mathbf{\right]}_A=\mathrm{}`$ for all $`qpP`$. Then $`A`$ is full.
## 2. Topological representation: the general case
Define two closures on $`P^{}`$. For $`pP`$ consider two subsets of $`P^{}`$:
$$๐ฐ๐ซ(p)=\{xx(p)=1\}๐ช(p)=\{xx(p)=0\}.$$
Then define closures $`\text{C}_1`$, $`\text{C}_2`$ in the following way:
$$\text{C}_1=\text{clos }\{๐ฐ๐ซ(p)\}_{pP}\text{and}\text{C}_2=\text{clos }\{๐ช(p)\}_{pP}.$$
Note, that since $`\overline{๐ฐ๐ซ(p)}=๐ช(p)`$ all $`๐ฐ๐ซ(p)`$ are C<sub>1</sub>O<sub>2</sub>-sets.
On $`AP^{}`$ consider closures $`\text{C}_{1}^{}{}_{A}{}^{},\text{C}_{2}^{}{}_{A}{}^{}`$ induced by $`\text{C}_1`$ and $`\text{C}_2`$ (i.e. $`\text{C}_{i}^{}{}_{A}{}^{}(X)=\text{C}(X)A`$). Let $`๐ฐ๐ซ_A(p)=๐ฐ๐ซ(p)A`$ and $`๐ช_A(p)=๐ช(p)A`$, then
$$\text{C}_{1}^{}{}_{A}{}^{}=\text{clos }\{๐ฐ๐ซ_A(p)\}_{pP}\text{and}\text{C}_{2}^{}{}_{A}{}^{}=\text{clos }\{๐ช_A(p)\}_{pP}.$$
We omit the index $`A`$ in $`\text{C}_{i}^{}{}_{A}{}^{}`$, $`๐ฐ๐ซ_A`$ etc. when it is clear which subspace is meant.
The following equations show the relation between the closures introduced on $`A`$ and $`A`$-ideals:
$$\text{C}_1(X)=\underset{pF(X)}{}๐ฐ๐ซ(p)\text{and}\text{C}_2(X)=\underset{pI(X)}{}๐ช(p).$$
###### Theorem 2.
Let $`AP^{}`$. Consider $`\sigma :P\text{C}\text{1}\text{O}\text{2}(A,\text{C}_{1}^{}{}_{A}{}^{},\text{C}_{2}^{}{}_{A}{}^{})`$ which maps $`p`$ to $`๐ฐ๐ซ(p)`$, then
* $`\sigma `$ is isotone;
* if $`A`$ is full then $`\sigma `$ is injective;
* if $`A`$ is separating then $`\sigma `$ is surjective.
###### Proof.
(1) Since $`pq`$ implies $`x(p)x(q)`$ for all $`xP^{}`$ then $`pq`$ implies $`๐ฐ๐ซ(p)๐ฐ๐ซ(q)`$, so $`\sigma `$ is isotone.
(2) For $`pq`$ either $`pq`$ or $`qp`$, so there exists $`xA:x(p)x(q)`$, then exactly one of $`๐ฐ๐ซ(p)`$, $`๐ฐ๐ซ(q)`$ contains $`x`$ and $`๐ฐ๐ซ(p)๐ฐ๐ซ(q)`$.
(3) Let $`B\text{C}\text{1}\text{O}\text{2}(A,\text{C}_{1}^{}{}_{A}{}^{},\text{C}_{2}^{}{}_{A}{}^{})`$, then $`B=\text{C}_{1}^{}{}_{A}{}^{}(B)`$ and $`\overline{B}=\text{C}_{2}^{}{}_{A}{}^{}(\overline{B})`$. Consider $`Q=I(\overline{B})F(B)=IF`$. If $`Q=\mathrm{}`$ there exists $`xA:x|_I=0`$ and $`x|_F=1`$, so $`x๐ฐ๐ซ(p)`$ for all $`pF`$ and $`x๐ช(q)`$ for all $`qI`$. Thus $`xB`$ and $`x\overline{B}`$ simultaniously, so $`Q\mathrm{}`$. For $`pQ`$ we have $`B๐ฐ๐ซ(p)`$, $`\overline{B}๐ช(p)=\overline{๐ฐ๐ซ(p)}`$ and $`B=๐ฐ๐ซ(p)`$. โ
###### Corollary 2.1.
Let $`A`$ be a full and separating subspace of $`P^{}`$, then $`P\text{C}\text{1}\text{O}\text{2}(A,\text{C}_{1}^{}{}_{A}{}^{},\text{C}_{2}^{}{}_{A}{}^{})`$.
To get the topological representation of an arbitrary poset we prove
###### Lemma 3.
$`P^{}`$ is full and separating.
###### Proof.
For disjoint ideal $`I`$ and filter $`F`$, which are in this case order ideal and filter, consider $`x:x(p)=0`$ for $`pI`$ and $`x(p)=1`$ otherwise. Obviously $`xP^{}`$ and separates $`I`$ and $`F`$. Applying lemma 1 we see that $`P^{}`$ is full. โ
This leads us to the following theorem:
###### Theorem 4.
Let $`P`$ be an arbitrary poset, then $`P\text{C}\text{1}\text{O}\text{2}(P^{},\text{C}_1,\text{C}_2)`$.
Due to the following lemma in the case of bounded poset $`P`$ subspaces $`A`$ of $`P^{}`$ can be reduced:
###### Lemma 5.
Let $`P`$ be a bounded poset, $`AP^{}`$ be full and separating, then $`A\{\mathrm{๐},\mathrm{๐}\}`$, where $`\mathrm{๐},\mathrm{๐}P^{}`$ are constant mappings, is also full and separating.
###### Proof.
Note that the ideals (filters) with respect to $`A\{\mathrm{๐},\mathrm{๐}\}`$ coincide with the proper $`A`$-ideals ($`A`$-filters) and for disjoint nonempty $`I`$ and $`F`$ the separating mapping $`xA`$ is not constant. โ
## 3. Topological representations: special cases
We apply the results of previous section to some special classes of posets.
### 3.1. Orthoposets
The bounded poset $`P`$ is called an orthoposet if there exists an anti-isotone mapping $`()^{}:PP`$ (orthocomplementation) such that $`p=(p^{})^{}`$, $`pp^{}=1`$ and $`pp^{}=0`$. For an orthoposet define its orthodual space $`P^{}`$ to be the set of all $`xP^{}`$ such that $`x(p^{})=(x(p))^{}`$.
###### Lemma 6.
$`P^{}`$ is full and separating.
###### Proof.
For disjoint ideal $`I`$ and filter $`F`$ consider $`x:x(p)=0`$ for $`pIF^{}`$, $`x(p)=1`$ for $`pI^{}F`$, otherwise $`x(p)=y(p)`$ for some $`yP^{}`$. Obviously $`xP^{}`$ and separates $`I`$ and $`F`$, so $`P^{}`$ is separating. As $`\mathbf{\left[}๐\mathbf{\right)}_P^{}`$ is the lower cone of $`p`$ for all $`pP`$ $`P^{}`$ is full according to lemma 1. โ
Since $`๐ฐ๐ซ_P^{}(p^{})=๐ช_P^{}(p)`$, the bases of closures $`\text{C}_1`$ and $`\text{C}_2`$ coincide and $`\text{C}_1=\text{C}_2`$. Denote
$$\text{C}=\text{C}_1=\text{C}_2$$
Then C<sub>1</sub>O<sub>2</sub>-sets are C-clopen. Applying theorem 2 we have
###### Theorem 7.
Let $`P`$ be an orthoposet, then there exists a closure space $`(,\text{C})`$ such that $`P\text{CO}(,\text{C})`$.
The representation obtained in previous theorem coicides with that described by Mayet and Tkadlec .
Now we use the notion of full separating subspace to characterize all orthocomplementations which can be introduced on a bounded poset $`P`$. Any orthocomplementation $`()^{}`$ defines a full separating subspace of $`P^{}`$ on which the closures $`\text{C}_1`$ and $`\text{C}_2`$ coincide. Let $`๐ฎ`$ be the collection of full separating subspaces of $`P^{}`$ where $`\text{C}_1=\text{C}_2`$. Consider $`A๐ฎ`$ then the set complementation on $`\text{C}\text{1}\text{O}\text{2}(A,\text{C}_1,\text{C}_2)P`$ is an orthocomplementation, so with every $`A๐ฎ`$ we can associate an orthocomplementation $`()^{}^A`$ on $`P`$.
###### Theorem 8.
All orthocomplementations on $`P`$ are in one-to-one correspondence with maximal (with respect to set inclusion) elements of $`๐ฎ`$.
###### Proof.
For $`A๐ฎ`$ all $`xA`$ preserves $`()^{}^A`$ because $`x(p)=x(p^{}{}_{}{}^{A})=1`$ implies $`x๐ฐ๐ซ(p)`$ and $`x๐ฐ๐ซ(p^{}{}_{}{}^{A})=\overline{๐ฐ๐ซ(p)}`$ (the similar contradiction holds for $`x(p)=0`$). It means that $`AP^A`$, so all maximal elements of $`๐ฎ`$ are of the form $`P^A`$. Thus any orthodual space $`P^{}`$ is a subspace of $`P^A`$ for some $`A`$. Obviously, orthocomplementation associated with $`P^A`$ is $`()^{}^A`$ and the one associated with $`P^{}`$ is $`()^{}`$. Since $`P^{}P^A`$ and orthocomplementations are induced by set complementation we get that $`()^{}=()^{}^A`$ and $`P^{}=P^A`$, so all orthodual spaces, defined by different orthocomplementations on $`P`$, are maximal in $`๐ฎ`$. โ
### 3.2. Distributive lattices
According to the Stone representation theorem any Boolean algebra is isomorphic to the collection of all clopen sets in some topological space. Since Boolean algebra is an orthocomplemented distributive lattice one can expect distributive lattice to be represented as the collection of C<sub>1</sub>O<sub>2</sub>-sets of some space with two topological closures. We are going to construct such a representation which follows from theorem 2 and is different from Priestley and Rieger .
For a lattice $`L`$ let $`L^{}L^{}`$ be the set of all lattice morphisms (isotone mappings preserving lattice operations) from $`L`$ to $`\mathrm{๐}`$. Note that $`L^{}`$-ideal is always lattice ideal (an order ideal $`I`$ is called lattice ideal if $`a,bI`$ implies $`abI`$).
###### Lemma 9.
For any distributive lattice $`L`$ the ideals (filters) with respect to $`L^{}`$ coincide with the lattice ideals (filters). Besides that, $`L^{}`$ is full and separating.
###### Proof.
First we prove that for disjoint lattice ideal $`I`$ and filter $`F`$ there exists $`xL^{}`$ such that $`x|_I=0;x|_F=1`$ (it means that $`L^{}`$ separates lattice ideals). Suppose $`I_0`$ to be the maximal lattice ideal containing $`I`$ which is disjoint with $`F`$. The set-complement of $`I_0`$ is a filter , thus the mapping $`x`$: $`x|_{I_0}=0`$, $`x|_{LI_0}=1`$ preserves $``$ and $``$. For an arbitrary $`pL`$ the upper cone of $`p`$ is a lattice filter. Then we get every lattice ideal $`I`$ to be the intersection of kernels of all $`x_p`$, which separates $`I`$ and the upper cone of $`p`$, over all $`pI`$, so I is an ideal with respect to $`L^{}`$ (recall the definition of ideal in section 1.2). Hence, the separating property for $`L^{}`$ is equivalent to the fact that $`L^{}`$ separates lattice ideals, which was proved above. $`L^{}`$ is full by lemma 1. โ
###### Theorem 10.
For any distributive lattice $`L`$ there exists a space with two topological closures $`(,\text{C}_1,\text{C}_2)`$ such that $`L\text{C}\text{1}\text{O}\text{2}(,\text{C}_1,\text{C}_2)`$.
###### Proof.
The only thing we need to prove is that the closures $`\text{C}_1`$, $`\text{C}_2`$ induced on $`L^{}`$ are topological. Since elements of $`L^{}`$ preserve both $``$ and $``$ we have $`๐ฐ๐ซ(pq)=๐ฐ๐ซ(p)๐ฐ๐ซ(q)`$ and $`๐ช(pq)=๐ช(p)๐ช(q)`$, so the bases of $`\text{C}_1`$ and $`\text{C}_2`$ are closed under finite set union, therefore the closures themselves are topological. โ
###### Corollary 10.1.
A lattice $`L`$ is distributive iff $`L^{}`$ is a full separating subspace of $`L^{}`$.
###### Proof.
This follows from lemma 9, the fact that for any lattice $`L`$ the closures induced on $`L^{}`$ are topological, and that for any space $``$ with two topological closures $`\text{C}\text{1}\text{O}\text{2}(,\text{C}_1,\text{C}_2)`$ is a distributive lattice. โ
### 3.3. Boolean algebras
Here we present a proof of the Stone representation theorem:
###### Theorem 11 (Stone).
Any Boolean algebra $`B`$ is isomorphic to the collection of all clopen subsets of a topological space.
###### Proof.
Since $`B`$ is a bounded distributive lattice, $`B^{}\{\mathrm{๐},\mathrm{๐}\}`$ is full and separating. Every lattice morphism of Boolean algebras preserves orthocomplementation and, as in the case of orthoposets, the topological closures $`\text{C}_1`$ and $`\text{C}_2`$ do coincide.
Associating with every element of $`B^{}\{\mathrm{๐},\mathrm{๐}\}`$ its kernel (that is a maximal lattice ideal) one get the Stone space of Boolean algebra originally described in . โ
###### Corollary 11.1.
Let $`L`$ be a distributive lattice, then $`L`$ is a Boolean algebra iff closures $`\text{C}_1`$ and $`\text{C}_2`$ coincide on $`L^{}\{\mathrm{๐},\mathrm{๐}\}`$. |
warning/0001/astro-ph0001487.html | ar5iv | text | # The Impact of Nuclear Reaction Rate Uncertainties on Evolutionary Studies of the Nova Outburst
## I Nuclear Uncertainties in Novae
Observations of nova outbursts have revealed an elemental composition that differs markedly from solar. Theoretical studies indicate that these differences are caused by the combination of convection with explosive hydrogen burning which results in a unique nucleosynthesis that is rich in odd-numbered nuclei such as $`{}_{}{}^{15}\mathrm{N}`$, $`{}_{}{}^{17}\mathrm{O}`$ and $`{}_{}{}^{13}\mathrm{C}`$. These nuclei are difficult to form in other astrophysical events. Many of the proton-rich nuclei produced in nova outbursts are radioactive, offering the possibility of direct observation with $`\gamma `$-ray instruments. Potentially important $`\gamma `$-ray producers include $`{}_{}{}^{26}\mathrm{Al}`$, $`{}_{}{}^{22}\mathrm{Na}`$, $`{}_{}{}^{7}\mathrm{Be}`$ and $`{}_{}{}^{18}\mathrm{F}`$.
The observable consequences of a nova outburst depend sensitively on the details of the thermonuclear runaway which initiates the outburst. One of the more important sources of uncertainty is the nuclear reaction data used as input for the evolutionary calculations STWS98 . A number of features conspire to magnify the effects of nuclear uncertainties on nova nucleosynthesis. Many reactions of relevance to novae involve unstable proton-rich nuclei, making experimental rate determinations difficult. For hydrodynamic conditions typical of novae, many rates depend critically on the properties of a few individual resonances, resulting in wide variation between different rate determinations. Statistical model (Hauser-Feshbach) calculations, which are employed with great success for a large number of reactions RaTK97a , are unreliable for rates dominated by individual resonances. The similarity of the nuclear burning and convective timescales results in nuclear burning in novae which is far from the steady state which typifies quiescent burning.
## II Monte Carlo Estimates of Uncertainties
Though analysis of the impact of variations in the rates of a few individual reactions has recently been performed using one-dimension hydrodynamic models JoCH99 , analysis of the impact of the complete set of possible reaction rate variations in such hydrodynamic models remains computationally prohibitive. We therefore begin by examining in detail the nucleosynthesis of individual zones, using hydrodynamic trajectories (temperature and density as a function of time) drawn from nova outburst models. Such one zone, post-processing nucleosynthesis simulations are a common means of estimating nova nucleosynthesis (see e.g., HiTh82 ; WGTR86 ). For this presentation we are using a hydrodynamic trajectory for an inner zone of a $`1.35M_{}`$ ONeMg WD which is similar to that described in PSTW95 . These calculations were performed using a nuclear network with 87 species, composed of elements from n and H to S, including all isotopes between the proton drip line and the most massive stable isotope.
Figure 1 shows the abundances of each species at the end of the simulation, $`5.2\times 10^5\mathrm{sec}`$ after peak temperature. Because of the long time which has elapsed, the unstable proton-rich nuclei have decayed, reducing their abundances to less than $`10^{20}`$. To investigate the extent to which nuclear reaction uncertainties translate into abundance differences, we use a Monte Carlo technique which assigns to each reaction rate in the nuclear network a random enhancement factor. The error bars displayed in Fig. 1 are the 90% confidence intervals which result from 992 Monte Carlo iterations. Monte Carlo methods have been employed with great success in the analysis of Big Bang nucleosynthesis SmKM93 , but have not previously been applied to other thermonuclear burning environments.
The reaction rate enhancement factors are distributed according to the log-normal distribution, which is the correct uncertainty distribution for quantities like reaction rates which are manifestly positive Smit91 ,
$$p_{lognormal}(x)=\frac{1}{\sqrt{2}\pi \beta x}\mathrm{exp}\left(\frac{(\mathrm{ln}x\alpha )^2}{2\beta ^2}\right),$$
(1)
where $`\alpha `$ and $`\beta `$ are the (logarithmic) mean and standard deviation. For small uncertainties $`(<20\%)`$, the difference between the log-normal distribution and the normal (Gaussian) distribution is small. However, for uncertainties of larger sizes such as those encountered in this problem, the difference is important. For this preliminary analysis, we have chosen to assign uncertainties of $`50\%(\beta =\mathrm{ln}(1.5))`$ both to rates calculated by Hauser-Feshbach methods and also to rates whose measurement require radioactive ion beams. For all other rates we assign $`\beta =\mathrm{ln}(1.2)`$. Figure 2 plots the resulting abundance distributions for two representative nuclei. Fig. 2 also demonstrates the differences between normal and log-normal distributions for widths of these sizes. These are very conservative uncertainties; relatively few reactions, especially among unstable nuclei, have measurement uncertainties this small.
## III Results
As evidenced by the error bars in Fig. 1, the impact of even our conservatively chosen variations in reaction rates on the nucleosynthesis is large. While broader conclusions will require analysis of additional hydrodynamic trajectories, a number of interesting points can be made from the analysis of this single trajectory. The impact on the rate of energy production is small. At the 90% confidence level, variations in the amount of hydrogen consumed represent $`10\%`$ variation in the thermonuclear energy released. For the most abundant metals (those which represent more than 1% of the mass), 2$`\sigma `$ variations by factors of 1.1 to 1.4 are common, with some of these nuclei showing 2$`\sigma `$ variations as large as a factor of 2, for example, $`{}_{}{}^{15}\mathrm{N}`$ ($`2.1\times `$) and $`{}_{}{}^{30}\mathrm{Si}`$ ($`2.3\times `$). For the $`\gamma `$-ray source nuclei $`{}_{}{}^{22}\mathrm{Na}`$ and $`{}_{}{}^{26}\mathrm{Al}`$, the 90% confidence interval includes variations of nearly a factor of 2, representing almost a factor of 4 difference in the distance from which novae may be observed by $`\gamma `$-ray telescopes. For $`{}_{}{}^{7}\mathrm{Be}`$, the 90% confidence interval spans more than 2 orders of magnitude.
Such large uncertainties in the nucleosynthesis, resulting from poorly known nuclear reaction rates, constrain our ability to make detailed comparisons between theoretical models for the nova outburst and astrophysical observations to a degree which is often ignored. Improved knowledge of these uncertain rates, both experimental and theoretical, is necessary to provide tight constraints on the nova outburst from its nucleosynthetic products. |
warning/0001/cond-mat0001145.html | ar5iv | text | # Quantum Dot Self-assembly in Growth of Strained-Layer Thin Films: a Kinetic Monte-Carlo Study
## I. Introduction
There has been considerable attention in recent years on the nature of the formation of three dimensional (3D) islands called quantum dots (QD) during the growth of strained-layer superlattices. For Ge grown on Si(001), for example, the nature of islands seen have been characterized by Mo et al . โHutโ clusters are the first type of islands to appear with well defined (105) facets, tilted at 11.3 to the surface, then there is a transition to larger islands with (11n)-like faces and finally to even larger but dislocated islands. It is possible to bypass the hut cluster stage by growing at slightly higher temperatures . Another much studied system is the growth of InAs on GaAs(001) (mismatch $`7\%`$); here the particular interest is in the uniformity in the size of islands formed . This uniformity, with dispersions of 10% in height and 7% in diameter of the islands at the initial stages of formation, decreases with coverage $`\theta `$. There seems to be a distinct coverage $`\theta _c`$ ( = 1.5 monolayers (ML), 1.75ML, 1.7ML ) at which the transition from two dimensional (2D) to three dimensional (3D) growth occurs for the InAs/GaAs system. This critical thickness transition is slightly dependent on growth conditions; the work of Gerard et al shows that by substantially increasing the deposition rate, for example, it is possible to shift it from 1.7ML to 1.95ML. There is evidence that the material to build an island comes mainly by depleting its immediate environment: the thickness of the InAs layer before islanding occurs, which is between one and two MLs, is reduced to one ML in the immediate region surrounding the island . This suggests that the critical layer thickness for InAs/GaAs(001), beyond which it is energetically favorable to form islands, is actually one ML and that the extra thickness before islanding may be due to the presence of a barrier at the 2D to 3D transition. There are signs that some depletion is also present in the growth of Ge on Si . It is known that island shapes and sizes can depend on growth conditions , so that kinetic effects are important. Under much higher deposition rates and lower growth temperatures than those used by Leonard et al and Moisson et al , Ruvimov et al found that islands also exhibit size uniformity ($`<20\%`$);however, island size increases with coverage . While Moisson et al observed (104) and (110) facets on the islands, Grundmann et al and Ruvimov et al saw only (110) facets; Moll et al showed that the equilibrium shape of an InAs island involves (111), $`(\overline{1}\overline{1}\overline{1})`$ and (110) facets in proportions which change with the size of the island.
Kinetic effects clearly change the shape and size of islands, and may even affect the critical thickness for the 2D-3D transition; however, the uniformity of islands seems to be robust for the highly strained InAs/GaAs system. We first look at the experimental results of islanding in InAs/GaAs systems because of the availability of data at small increments of coverage $`\theta `$ . There are a number of observations that need to be discussed. The first is the narrow distribution in width and height of the 3D islands. The second is the existence of a sharp (possibly first order) 2D-3D transition at a critical coverage $`\theta _c`$ . There is also the phenomenon of fast depletion (of the order of seconds ), where a 3D island is created quickly (compared to a deposition rate of .01ML/sec ) largely out of the atoms from the its 2D environment. Finally, it is also seen that under conditions of slow deposition 3D islands remain essentially constant in size over a coverage interval of $`\mathrm{\Delta }\theta 0.4`$. Note that these results are affected by growth conditions. Under high deposition rates (compared to diffusion rates) which is possible at low temperatures, the sharp island size distribution may disappear, see, for example, growth of Ge on Si(001) , where the lattice mismatch is smaller ($``$ 4%). For the Ge/Si system, where the strain is much less than that in the InAs/GaAs system, depletion seems to take a time of the order of minutes at 550C and recent growth experiments were carried out at typical deposition rates of a few MLโs/sec . Under these growth conditions, even the sharp 2D-3D transition may disappear.
In this study then we focus on the early stages of growth for thin films which grow in the Stranski-Krastanov (SK) mode. We study growth under conditions where diffusion is fast compared to deposition, so that effects due to the process of depletion can be distinguished from those due to deposition. In a previous work , we investigated the energetics of the 2D to 3D transition in detail by means of molecular dynamics simulation, using an empirical potential that has been appropriately tuned . We argued that the 2D to 3D transition occurred when 2D islands had grown much larger than the size $`s_o`$, when 3D islands first become energetically favorable; this effectively is a barrier, which once scaled by a 2D island, allows it to reorganize itself into a 3D shape, with an immediate gain of energy. This gain, which is more pronounced for the highly mismatched InAs/GaAs system than for the Ge/Si, can be quite substantial, about 5 - 10 meV/atom for the former. We feel that this is the underlying factor for the uniformity of sizes of islands seen in this system. Priester et al have attempted to provide an explanation for the uniformity of the 3D islands, but have not taken into account the factor of the barrier, which should affect their considerations.
It is known for the growth of Si/Si(001) that islands nucleate with a critical size of one to three atoms and then grow two-dimensionally. This picture of nucleation is also supported by the results of Chen and Washburn, who used a critical nucleus of $`i=1`$ in the scaling function $`\mathrm{\Phi }(N/\overline{N})`$ in fitting the island density results of Leonard al. Island nucleation of Ge on Si(001) should be similar (in both cases the dimer is the stable nucleus). We suggest that the 2Dโ3D transition picture is the following: 2D islands nucleate with critical nuclei of about one atom and grow two-dimensionally until a critical size $`s_c`$ when strain makes it favorable for there to be a transition to 3D growth. This size $`s_c`$ is quite large, roughly a few hundred angstroms. There is direct experimental evidence for this picture of growth. Mo and Lagally observe, after growth of about 3ML of Ge on Si at 500C, a growth front roughness of three layers over an area of 60nm x 60nm . Gerard et al observe one layer roughness over extensive 2D areas ($`2000\AA `$ ) for the growth of InAs on GaAs(001) at 520C. We stress that $`s_c>>s_o`$, the size at which 3D clusters have just become energetically favorable. Indeed the 2D island must reach a size comparable to that of the two-layer island when the latter becomes energetically favorable. Once this size barrier is reached, the transition to islands of two or more layers in height is possible since taller islands are already favorable at smaller sizes. There is a rapid rearrangement of its atoms in order to achieve the shape of the optimally energetic (105) facetted clusters. There is an immediate gain in energy of 1-2 meV/atom for the Ge/Si system; for InAs/GaAs, we estimate this gain, assuming that the elastic energy scales with the square of misfit, to be 5-10 meV/atom. This latter amount is substantial and is probably the reason for the phenomenon of depletion seen in the highly mismatched systems.
The above picture obtained from an energetics study is complemented by our work here on kinetics. In this study, we approach island growth on strained-layer superlattices, by using finite temperature non-equilibrium Monte Carlo (MC) simulations, where diffusion rates of adatoms depend on strain as well as the usual local bonding. Computational time and size constraints force us to carry out our kinetic MC simulations in 1+1 dimensions, i.e., in our MC simulations the substrate is one dimensional and the growth is two dimensional. We do not believe that our 1+1 dimensional simulations introduce any qualitative complications, although it will be necessary in the future to verify our proposed picture using the full three dimensional MC simulations. Our results show that under growth conditions of fast diffusion relative to deposition, i.e., not very low growth temperature, the picture obtained from energetics is largely correct. There is a sharp 2Dโ3D transition which occurs at an island size s<sub>c</sub> which is well beyond the critical size s<sub>o</sub> at which the 3D islands first become energetically favorable. Depletion is observed and narrow 3D island distributions are obtained. The average size of 3D islands does not change with coverage. In this work, we attempt to understand the microscopic dynamics and mechanisms underlying these results. In the following section we describe the simulation method and the parameters chosen. Then we present detailed results of the simulation in Sec. III and discuss the results in Sec. IV. We conclude in Sec. V.
## II. The Simulation Model
In our MC growth simulations (which is done in 1+1 dimensions), an adatom moves (under solid-on-solid restrictions (SOS)) by hopping randomly to neighboring sites at a rate that depends on its bonding. (We obey detailed balance in our kinetic MC simulation.) The hopping activation energy depends on the bonding environment and the elastic energy associated with strain The hopping activation energy depends on the bonding environment and the elastic energy associated with strain. The hopping rate is given by the expression,
$$R_n=R_oexp^{\frac{E}{k_bT}},$$
(1)
where $`R_o=d^{}kT/h`$ is a characteristic vibrational frequency and $`d^{}=1`$ is the substrate dimension. The activation energy $`E=E_{bond}E_{strain}`$, with $`E_{bond}`$ being determined by the number of nearest neighbors(nn) and next nearest neighbors(nnn). The elastic energy is given by harmonic interactions between an atom and its nn and nnn neighbors, using spring constants k. Following Orr et al, we obtain $`E_{strain}`$ for a particular site by taking the difference in elastic energies of the system when the site is unoccupied and when the site is occupied. This energy is calculated by allowing atoms in a 5(height)x7(width) cell centered at the site first to equilibrate under molecular dynamics simulation and then to relax to its minimum energy configuration by means of the method of steepest descent. Every 100 time steps or so the entire system is allowed to relax globally to avoid any local strain accumulation. $`E_{bond}`$ is chosen in the following way,
$$E_{bond}=\{\begin{array}{cc}E_o=(0.7NN+0.2NNN)eV,\hfill & \text{if NN }2\hfill \\ E_1=4.0eV,\hfill & \text{ if NN=3}\hfill \\ E_2=1.45eV,\hfill & \text{ steps of height }2\hfill \end{array}$$
(2)
where NN is the number of nnโs and NNN is the number of nnnโs. $`E_o`$ applies to single adatoms or atoms at step edges, except when step heights are two layers or greater. Then $`E_2`$, a reduced barrier height, is applied to the surface atoms on top of these steps, so that inclined (11) island facets are favored over vertical ones. $`E_1`$ is the barrier for the rest of the surface atoms which have three nnโs. It is chosen a little higher than that given by bond counting to eliminate intrasubstrate breakaway (especially at the foot of islands) and therefore to avoid substrate roughening, which is not seen experimentally. For simplicity we have also used the same barrier for midisland surface atoms; results are not different from those using bond counting for these atoms. The parameters have been chosen so that diffusion will dominate over deposition, for example, a single adatom will diffuse a distance of approximately 600 unit cells for each deposition event at 750K. This is about 50-100 times the width of the islands that form. Using diffusion rates from Mo et al and others and experimental deposition rates and island sizes, we get comparable results of the ratio diffusion distance/island size$`100`$. We choose the spring constant $`k=200eV200`$ times the diffusion barrier for a single adatom, and a deposition rate of 0.01-0.2 MLs/sec. We carried out simulations for strained-layer lattices with misfits of $`07\%`$, at temperatures of 700 to 800K. We start with systems at thicknesses of 11MLs, with the three top layers at the larger lattice constant. System sizes vary from 500 to 8000 cells. At zero strain, growth was layer by layer as would be expected under the above conditions of fast diffusion โ there is no kinetic roughening at this โhigh temperatureโ growth in the absence of strain.
## III. Results
We report on two preliminary studies that will help in understanding the final results. First we carry out simulations for the unstrained system, varying the diffusion barrier for atoms at the ends of islands, $`E_{end}`$, from 1.3 to 1.8eV, for temperatures T, from 700-800K. and deposition rates 0.1-0.2 MLs/sec and over coverages of $`\theta `$ from .5 to 0.8. In Table I, we display the results for two growth temperatures, 750 and 800K, with a deposition rate of 0.2MLs/sec, a system size of $`10^4`$ cells and a coverage $`\theta 0.6`$. We calculate a roughness index (R.I.) as the percentage of sites in islands, which have heights $`>1`$, i.e., R.I. is a rough measure of the deviation from โtwo-dimensionalityโ (one-dimensionality in our simulations) in the islands. For $`E_{end}1.5eV`$ growth is smooth, islands are flat (very small R.I.), but growth is distinctly rough for $`E_{end}1.4eV`$, there being much larger proportions of islands with 2 or more layers in height. Clearly the transition from smooth to rough growth is sharp.
Next we look at island-end energies $`E_{end}`$ for some island configurations when elastic interactions are included. Specifically we carry out calculations for misfits of 5%. In Fig. 1, we plot $`E_{end}`$ against island volume (number of atoms), for seven island configurations, comprising (a) 1-level islands (h=1), (b) 2-level islands (with 1 atom (h=2a) and 2-atoms (h=2b) on the second level, (c) 3 level islands with 1 (h=3a) and 2 atoms (h=3b) on the third level and 3 and 4 atoms respectively on the second and (d) 4 and 5 level islands each with 1 atom on the top level (h=4,5 respectively) and the same shape as islands in (c). Island volumes are varied by changing the length of level 1 of the islands, while keeping upper configurations fixed. If we take $`E_{end}<1.5`$eV as the condition for rough growth, then islands with volumes $`>`$ 15 will have end atoms with diffusion barriers $`<`$ 1.5eV for all the consequtive configurations 1,2,3 and higher levels. The following picture of 3D islanding is suggested: 2D islands grow two dimensionally until a certain size when end atoms are promoted to the second level; this process becomes more rapid as it proceeds because $`E_{end}`$ increases with the number of atoms on the second level (while island volume is kept constant). This process then continues in the same fashion with the subsequent promotion of atoms to the third and higher levels. This, we believe, is the mechanism for the phenomenon of depletion seen experimentally .
We now present results of our full kinetic MC simulation done on systems of substrate sizes L=2000, 4000 and 8000 cells. The observations we report below are true of all these sizes and so are not affected by finite size effects. For these simulations, we also consider the effect of a strain enhancing factor $`F_{end}=1.0,1.2`$ and 1.5 on the first level end atoms of islands. It is known that there is tremendous strain at the foot of islands. Our results are not particularly sensitive to variations in this strain, aside from making islands a little smaller as $`F_{end}`$ is increased. In Fig. 2 , we follow the development of a single island over a growth period of about 0.2ML ($`<3`$secs.). Figs. 2a,b,c show a 1-level(2D) island of volume 19 atoms being folded up into a 2-level island in 0.5 secs. The material for this 2-level island (volume=20) comes almost completely from the original 1-level island. In the rest of Figs. 2d,e,f and g, we see similarly rapid buildups of the third and fourth levels after a brief waiting period. The whole process starting from Fig. 2a to Fig. 2g takes less than 3 secs. The bulk of the material ( $`80\%`$ for the 3-level island and $`65\%`$ for the 4-level island) for the formation of the 3D island comes from the original 2D island (compare with the experimental results of the three groups above). Fig. 2 shows a typical 2D-3D transition sequence for islands in our simulation. It clearly illustrates the process of depletion seen experimentally. (Note that in our simulations what we refer to above as 2D and 3D are really 1D and 2D respectively since we are using 1+1 dimensional simulation).
In Fig. 3 we display width and height distributions of islands for a range of coverages $`\theta =0.3930.87`$. There is uniformity in the island size distributions which are sharply clustered around the mean width or height, each with a half-width of $`1`$ cell. Furthermore, while island density increases with coverage, the average island size remains essentially constant. In Table II, we show the average volumes at which islands undergo transitions from the first to the second levels, from the second to the third, and from the third to the fourth level. The root mean square deviation is 2 atoms in each case, showing that transitions occur at sharp distinct sizes.
We plot in Fig. 4 the total number of islands with 3 or more levels as a function of coverage $`\theta `$ for systems of size L=4000. The results are the same for systems of other sizes (L=2000 and 8000) when appropriately normalized. We see that island density is zero until a certain coverage $`\theta _c`$ is reached, when the density increases rapidly. Leonard et al observed this experimentally and fitted the island density $`\rho _{isl}`$ with the function $`\rho _{isl}=\rho _o(\theta \theta _c)^\alpha `$. They obtained a value of $`\alpha =1.76`$ while we get $`\alpha =1.34`$. The difference in the value of $`\alpha `$ could be due to our using a 1+1 dimensional simulation. We arrive at similar conclusions if we look at islands with 2 or more levels instead of the $``$3 levels we have chosen above.
In Table III, we show the energies of islands of various configurations comprising 2,3 and 4 levels, relative to the energies of their corresponding 1-level configuration at the same volume. We see that the first energetically favorable 2-level island is the one whose volume is 8 atoms with a configuration of 3 atoms on the second level and 5 on the first. 3-level islands become favorable at a volume of about 12 atoms but for this volume the 2-level configuration has the best energetics. 3 and 4-level islands are energetically optimal at volumes of 15 atoms and 24-28 atoms respectively. These figures can be compared to the transition volumes of Table II. There clearly is a correspondence between energetics and kinetics. However, one interesting point emerges, although a 2-level island becomes energetically favorable at a volume of 8 atoms, kinetically the transition occurs at a volume well beyond that (around 19 atoms). Energetics sets the lower size limit for the beginning of depletion, but it is kinetics that determines the actual point. This is the s$`{}_{c}{}^{}>>`$ s<sub>o</sub> kinetically driven scenario we discussed before. Note also that because of the small size of the islands we encounter here, it is the 2-level island that first becomes energetically favorable before the taller islands; we expect the situation to be reversed when mean island sizes are larger as surface energies become less significant - this would be the case with the sizes actually seen experimentally. This aspect of physics is not appropriately captured in our small system 1+1 dimensional simulations.
## IV. Discussion
The experimental results of islanding in InAs/GaAs systems of a number of groups are shown in Table III. The first four groups observed uniformity in the size distributions of the islands, in particular, Leonard et al reported dispersions of 10% in height and 7% in diameter at the first appearance of the islands, at $`\theta \theta _c`$; with further deposition, this uniformity is reduced, island density increases but sizes remain essentially the same . The first three groups concluded that there is depletion-like behavior. Leonard et al show that more than 80% of the atoms to form an island comes from its environment, rather than from additional deposition. Gerard et al display an atomic force micrograph ( Fig. 3 in ref.) of the depletion zone around an island, whose size is $`1000\AA `$. They also show that the timescale of this mass movement to form an island is from 2 to 10 seconds. This phenomenon of depletion is clearly consistent with the results of our simulation. It takes a few seconds in a highly mismatched system, but is much longer, $``$minutes, in the Ge/Si system; so in this system it is probably masked by the deposition rates used and only two groups have reported seeing it in this system . From an energetics perspective, beyond the critical coverage $`\theta _c`$, as deposition continues, there is much more energy to be gained for the new material to create new 3D islands than to grow existing ones. So there is an increase of island density but little size gain. Our simulation shows that the process of depletion is driven by two factors. First 2D islands are grown well past the size $`s_o`$ at which 3D islands become energetically favorable. At a distinct critical size $`s_c`$ determined by kinetics, atoms at the 2D island ends are, as a result of strain, easily promoted up to the next level. Secondly, strain continues to be adequate to keep island-end atoms mobile even with little further island growth, so that the next higher levels are formed quickly, also at distinct sizes. The process of depletion lasting seconds only, then, is largely the pulling in of existing material to form 3D islands. It is completed when the tallest island is formed. Subsequent growth of these islands is mainly by the formation of new facets. Facet formation is generally much harder than adding atoms to the ends of a 2D island. As can be seen from Fig. 1, the diffusion barrier for an atom at the end of a 2D island of size 10-15 atoms is $`1.5`$eV while it is $`1.381.4`$eV at the bottom edge of a 3-4 level island. This difference translates into a substantial difference in mobility, as we have seen above, so that as long as 2D islands are present, their growth is strongly favored over that of 3D islands. The uniformity of islands at coverage of $`\theta _c`$ is due to depletion occurring at distinct sizes. The continuing size uniformity coupled with constant mean 3D island size while 3D island density increases, as deposition proceeds, especially for $`\theta \theta _c0.5`$ , is due to the preferred growth of 2D over that of 3D islands. Chen and Washburn obtained results of continuous increase in the size of 3D islands (see their Fig. 5) with the rate being the largest at the smallest size. Clearly this can only apply after most of the 2D islands have disappeared.
As we have noted before, there should be a correlation between kinetics and energetics. Our kinetic MC simulation shows that the depletion process begins once strain enables 2D island edge atoms to be mobile enough to be promoted to the upper levels. But this process must be also favored by the energetics; we expect the corresponding 3D island to be energetically more favorable than the 2D island. For the systems studied here we have observed this correspondence. In a previous paper we studied the energetics of (10n) facetted Ge โhutโ clusters on Si substrate, using the atomic configuration of (105) side facets suggested by Mo et al. The general conclusion was that taller (105) facetted islands become energetically favorable at smaller sizes than islands with (10n) facets for n$``$7. (103) facetted islands are excluded because these faces require costly double steps so that they are not observed experimentally. (105) facetted hut clusters become favorable only when they have at minimum, heights of 12 layers (at size s<sub>o</sub>, say). Islands with lower aspect ratios have to reach greater sizes to become energetically favorable. For systems which are growing with growth front roughness of 1-2MLs, it is then necessary for 2D islands to grow well beyond the size s<sub>o</sub> before the transition to a 3D shape can begin. This size s$`{}_{c}{}^{}{}_{}{}^{}`$ may be comparable to the size at which a 2-level island first becomes favorable. As we have noted in the simulations above, the actual transition size, s<sub>c</sub> is determined by kinetics but this size must be such that s$`{}_{c}{}^{}`$s$`{}_{c}{}^{}{}_{}{}^{}`$. So energetics sets the lower size limit at which a 2Dโ3D transition can occur. In Fig. 5, (see Khor et al) (111) facetted islands are shown to become energetically favorable at sizes and heights greater than those for (105) islands at size s<sub>o</sub>. With increasing size, (111) facetted islands quickly become more favorable than the (105) hut clusters. These results are consistent with experimental observations of Hansson etal who obtained (111)-facetted islands under near equilibrium conditions and also the results of Moet al where macroscopic structures were seen to be the stable ones.
Three groups, Leonard et al, Moisson et al and Gerardet al, observe the presence of a critical coverage $`\theta _c`$ below which no 3D islands are seen. Leonard et al characterise this transition to be like that of a first-order phase transition. We observe a similar transition in our simulations. However, in contrast to the results above , Polimeni et al report a smooth 2Dโ3D transition for the growth of InAs on GaAs(001). In Table IV, we compare the growth conditions for the different groups. The growth temperature used by Polimeni et al at $`420^{}`$C, is substantially lower than those ($`500530^{}`$C) used by the other groups. At these temperature differences, the diffusion rate D could differ by more than an order of magnitude. Assuming a behavior for D to be similar to that observed for Si adatoms on Si(001), we calculate the ratio R/D, where R is the deposition rate. This ratio is $`N^3`$ (N= island density), for the same coverage; this tendency to nucleate islands should correlate with growth front roughness. We see from the last column of Table IV, this ratio for Polimeni et al is about 50 times that for Leonard et al. Since higher effective deposition rates contribute to rougher growth, it may make level to level transitions less distinct than those we have seen in the simulations above. At some point kinetic roughness at the growth front arising from fast (slow) deposition (diffusion) may mask the phenomenon of depletion and give rise to an apparently smooth 2Dโ3D transition.
Solomonet al have shown that 3D island density, at fixed coverage and temperature, is increased when either the growth rate R is reduced or the diffusion D is increased. (The latter is done by increasing the flux V/III ratio). A related observation is made by Mo and Lagally for Ge/Si(100) growth. When they deposited Ge at 850K , they found the concentration of macroscopic clusters to be higher than that at T$`<`$800K. This result is unexpected for as we have seen above from nucleation theory for regular island growth that island density goes as $`R^p/D^p`$ where p is positive ; but this applies, in our case to the 2D islands. Increasing deposition rate or decreasing diffusion then increases the 2D island density and correspondingly decreases average 2D island size at a given coverage. Assuming that there is an average 2D-3D island transition size s<sub>c</sub> for the growth regimes of Solomon et al (this must be the case since they observe the constant 3D island diameter throughout their experiments), this means that fewer 2D islands reach this size at that coverage. The density of 3D islands then increases by reducing growth rate or increasing D. We see from table IV, that its relative R/D ratio of 24-52 may put it in a โroughโ growth regime closer to that of Polimeni et al than to that of Leonard et al, so that the existence of a sharp $`\theta _c`$ is uncertain.
In Table IV it is interesting to note that even for a relative R/D=120, much larger than that of Polimeni et al, Ruvimov et al still observed 3D island size uniformity $`<20\%`$. This is true even when their observations were carried out for coverages of $`\theta `$ = 2-4MLs, which is much greater than the $`\theta _c`$ of Leonard et al. They did not specifically study if the 2D-3D transition is sharp or smooth.
We have noted above that depletion is seen to occur on a timescale of minutes in Ge/Si systems compared to seconds for InAs/GaAs systems. We should expect to see results in the former similar to those observed for the latter if the deposition rates and growth temperatures are appropriately scaled. Shklyaev et al have carried out growth experiments of Ge on Si(111) at small increments of coverage; they used a growth rate of 0.004 bilayer(BL)/sec and a temperature of $`480^{}`$C ( R/D$``$0.8, see Table IV). They observed the growth of two types of islands which were called large flat islands and 3D islands. The latter appear โabruptlyโ, there being a distinct jump in 3D island density over a growth interval of 0.1 BL. Much of the material for the formation of these islands come from the substrate. Annealing experiments suggest that this depletion occurs over a time period of about 10 minutes. They did not measure island size distribution but their Fig. 1 shows 3D island images which appear quite uniform in size. Many of the experiments for the growth of Ge on Si were carried out with quite high relative R/D values, for example, R/D=7,7 and 4 for Voigtlander et al , Medeiros-Ribeiro et al and Kastner et al respectively. It is in this growth regime that the last two groups observed rectangularly shaped โhutโ islands. Voigtlander et al saw the aspect ratio of a single island change over a coverage interval of $``$1BL(20min), indicating that depletion probably takes that long.
As noted above, the time ($`\tau )`$ for depletion increases with decreasing lattice mismatch $`x`$. In general it probably goes as $`\tau x^\eta D^\gamma `$, where $`\eta `$ and $`\gamma `$ are some positive constants. We have seen above that depletion occurs over a large range of growth conditions as determined by the ratio R/D. We would expect depletion to fail to occur only when it is completely overwhelmed by deposition, that is, when 1/R $`\tau `$, or when $`xR^{1/\eta }D^{\gamma /\eta }`$. This must be the condition for smooth (non-islanding) growth at low temperatures or high deposition rates. The relationship must be applicable to the temperature-concentration phase curve, delineating smooth from rough growth for the deposition of Si<sub>1-x</sub>Ge<sub>x</sub> on Si(001) obtained by Bean et al. In Fig. 6, (Khor et al), a replot of the experimental data of Fig.1 of Bean et al shows a linear relationship of ln(x) versus 1/T, (except for the point at $`x=1`$ where a minor temperature change from 550 to 527C would put the point on the line), which supports this conclusion.
## V. Conclusion
In conclusion, we find that in general, for strained heteroepitaxial growth of semiconductors, there exists an effective kinetic barrier for the 2D to 3D transition. Under conditions of slow deposition and fast diffusion, islands initially grow two-dimensionally to a size $`s_c>>s_o`$ well beyond the size $`s_o`$ at which a 3D island first becomes energetically favorable. At this size $`s_c`$ atoms at the edge of the 2D island become mobile as a result of strain, and are promoted to the next level. Promotion of atoms to the next levels occurs in quick succession because edge atoms continue to be highly strained and so remain mobile. The process of depletion is completed when the island attains it highest aspect ratio. This size $`s_c`$ is sharply defined, and there is a correlation with energetics. This is a robust result that should apply to a wide range of semiconductor systems. For highly mismatched systems, it is the underlying microscopic reason for the uniformity in the sizes of islands seen experimentally. It is consistent with other experimental results such as the increase in island density with coverage with no corresponding increase in size, the phenomenon of depletion, the (initially unexpected), result that island density increases with reduced growth rate or enhanced diffusion.
This work is supported by the U.S.-ONR and NSF-MRSEC. |
warning/0001/hep-th0001156.html | ar5iv | text | # Chaotic Symmetry Breaking and Dissipative Two-Field Dynamics
## I Introduction
The study and understanding of the dynamics of fields are a timely subject and of broad interest, with applications in diverse areas like in particle physics, cosmology and in condensed matter (for a recent review, see and references therein). Additional interest on the subject comes from the fact that many of the theoretical ideas and models can be tested in ongoing experiments, as those been performed in condensed matter systems, and in the future ones, in the entering of operation of the RHIC and LHC heavy ion colliders, which will be able to probe possible new phenomena at the QCD scale and space based experiments, which will be putting on test different cosmological models. It is then becoming urgent the detailed investigation of the underline field dynamics that may be common to all these very different areas of physical research.
In this paper we are mainly concerned with the connection between the development of strong nonlinearities in the time evolving system of equations of motion of a given field theory model and the possible chaotic behavior associated to them. Lets recall that in symmetry breaking phase transitions we are usually interested in the study of the evolution of a given order parameter, for example the magnetization in spin systems in statistical physics or a vacuum expectation value of some scalar field (the Higgs) in particle physics, which gives a measure of the degree of organization of the system at the macroscopic level. However, at the microscopic level disorder is related to the nonlinearities and fluctuations responsible to chaotic behavior in the system and these chaotic motion phenomena can reflect in a nontrivial time dependence of the macroscopic quantities and, therefore, influencing all the dynamics of the system. This is clear once several properties of the system at longer times are closely related to the microscopic physics, like relaxation to equilibrium, phase ordering, thermalization and so on. Thus we expect that chaos will be not only an important ingredient in determining the final states of a given system but also in how it gets there.
Previous studies on chaos in field theory have mostly emphasized chaotic behavior in gauge theory models (for a review and additional references and applications, see ). In special, in homogeneous Yang-Mills-Higgs models we can reduce the system of classical equations of motion to ones analogous to those of nonlinear coupled oscillators, which is well known to exhibit chaotic motion.
While in all the previous works on chaotic dynamics of fields dealt with (conservative) Hamiltonian systems, here we will be mainly concerned with the effective field evolution equations, which are known to be intrinsically dissipative and, therefore, the dynamical system we will be studying is non-Hamiltonian.
Thus, we will be concerned with the influence of field dissipation, due to field decaying modes, on the degree of chaoticity of the field dynamics. Typically we expect that dissipation damps the fluctuations on the system and consequently tends to suppress possible chaotic motions and makes field trajectories in phase space to tend faster to the system asymptotic states. On the other hand there are well known examples of dynamical systems, as the Lorenz system , which are dissipative ones and at the same time they display a very rich evolution on phase space in which, under appropriate system parameters, field trajectories may be lead towards strange attractors. The verification of the same properties in a model motivated by particle physics would be a novel result with possible consequences to, for example, particle physics phenomenology and cosmology.
We study chaos in our dynamical system of equations by means of the measure of the fractal dimension (or dimension information) \[for a review and definitions, see e.g., \], which gives a topological measure of chaos for different space-time settings and it is a quantity invariant under coordinate transformations, providing then an unambiguous signal for chaos . The method we apply in this work for quantifying chaos will then be particularly useful in our planned future applications of our model and it extensions to a cosmological context, in which case other methods may be ambiguous, like, for example, the determination of Lyapunov exponents, which does not give a coordinate invariant measure for chaos, as discussed in . Also, other methods for studying chaotic systems, like for example by Poincarรฉ sections, are not suitable in the case we are interested here, in which chaos is a transitory phenomenon as we will see.
This work is organized as follows. In Sec. II we introduce the model and discuss its general properties at the classical level. In Sec. III we obtain the one-loop effective equations of motion for the fields and we determine the general form of the nonlocal (non-Markovian) dissipative kernels. In Sec. IV we then discuss the validity of the one-loop approximation and the Markovian approximation for the dissipative kernels appearing in the one-loop effective equations. We couple our system of fields to a set of $`N`$ other fields making up the bath (or environment), in which the system evolves, and the large N behavior of the various important quantities is determined. In Sec. V we then present the Markovian form for the effective equations of motion and our main numerical results, where we determine the fractal dimension. In Sec. VI our concluding remarks are given.
## II The Two Interacting Scalar Fields Model
The model we will study consists of two scalar fields in interaction with Lagrangian density given by
$`[\mathrm{\Phi },\mathrm{\Psi }]`$ $`=`$ $`{\displaystyle \frac{1}{2}}(_\mu \mathrm{\Phi })^2{\displaystyle \frac{m_\varphi ^2}{2}}\mathrm{\Phi }^2{\displaystyle \frac{\lambda _\varphi }{4!}}\mathrm{\Phi }^4`$ (1)
$`+`$ $`{\displaystyle \frac{1}{2}}(_\mu \mathrm{\Psi })^2{\displaystyle \frac{m_\psi ^2}{2}}\mathrm{\Psi }^2{\displaystyle \frac{\lambda _\psi }{4!}}\mathrm{\Psi }^4{\displaystyle \frac{g^2}{2}}\mathrm{\Phi }^2\mathrm{\Psi }^2.`$ (2)
All coupling constants are positive and $`m_\varphi ^2>0`$, but we choose $`m_\psi ^2<0`$, such that we allow for spontaneous symmetry breaking in the $`\mathrm{\Psi }`$-field direction. Additionally, note that from the above Lagrangian, that for values of $`\mathrm{\Phi }`$ larger than a $`\mathrm{\Phi }_{\mathrm{cr}}`$, where $`\mathrm{\Phi }_{\mathrm{cr}}^2=|m_\psi ^2|/g^2`$, there is no symmetry breaking in the $`\mathrm{\Psi }`$-field direction. Thus, for example, if we have an initial state prepared at $`\mathrm{\Phi }>\mathrm{\Phi }_{\mathrm{cr}}`$, the $`\mathrm{\Psi }`$ field will move towards zero and remain around that state till eventually $`\mathrm{\Phi }`$ crosses below the critical value inducing a (dynamical) symmetry breaking in the $`\mathrm{\Psi }`$-field direction, after which the fields evolve to their vacuum values at $`\mathrm{\Psi }_v=\pm \sqrt{6}|m_\psi |/\sqrt{\lambda _\psi }`$ and $`\mathrm{\Phi }_v=0`$.
The classical equations of motion for field configurations $`\mathrm{\Phi }=\varphi _c`$ and $`\mathrm{\Psi }=\psi _c`$ can be readily be obtained from Eq. (2):
$$\mathrm{}\varphi _c+m_\varphi ^2\varphi _c+\frac{\lambda _\varphi }{6}\varphi _c^3+g^2\varphi _c\psi _c^2=0,$$
(3)
$$\mathrm{}\psi _c+m_\psi ^2\psi _c+\frac{\lambda _\psi }{6}\psi _c^3+g^2\psi _c\varphi _c^2=0.$$
(4)
For homogeneous fields $`\varphi _c`$ and $`\psi _c`$ (in which case Eqs. (3) and (4) are equivalent to the equations of motion of two particles with quartic potentials and quadratic interaction between them) the above equations are well known to lead to chaotic trajectories in phase space. The chaotic behavior of very similar classical equations have been studied recently in Ref. for a model with $`Z_2\times Z_2`$ symmetry and the corresponding dynamical system shown to be chaotic for symmetry breaking in one of the field directions.
## III The One-Loop Equations of Motion
In the following we study how quantum effects, which will be responsible for the appearance of dissipative dynamics in the system evolution, will influence this dynamical symmetry breaking process and we identify and quantify possible chaotic behavior in the system evolution, as a function of the fields dissipations.
Roughly speaking, chaos means extreme sensitivity to small changes in the initial conditions. Due to nonlinearity, fluctuations in the initial conditions of chaotic systems evolve such that they can completely alter the asymptotic outcome of the unperturbed trajectories in phase space. We here quantify the chaoticity of the system by measuring the fractal dimension. The fractal dimension is associated with the possible different exit modes under small changes of initial conditions and it gives a measure of the degree of chaos of a dynamical system . The exit modes we refer to above are one of the symmetry breaking minima in the $`\mathrm{\Psi }`$-field direction, which are attractors of field trajectories in phase space. The method we employ to determine the fractal dimension is the box-counting method, whose definition and the specific numerical implementation we use here have been described in details in Ref. .
Quantum corrections are taken into account in the effective equations of motions (EOMs) by use of the tadpole method . Let $`\varphi _c`$ and $`\psi _c`$ be the expectation values for $`\mathrm{\Phi }`$ and $`\mathrm{\Psi }`$, respectively. Splitting the fields in (2) in the expectation values and fluctuations,
$$\mathrm{\Phi }\varphi _c+\varphi \mathrm{and}\mathrm{\Psi }\psi _c+\psi ,$$
(5)
where $`\mathrm{\Phi }=\varphi _c`$ and $`\mathrm{\Psi }=\psi _c`$, the EOMs for $`\varphi _c`$ and $`\psi _c`$ are obtained by imposing that $`\varphi =0`$ and $`\psi =0`$, which lead to the condition that the sum of all tadpole terms for each field vanishes. Restricting our analysis of the EOMs to homogeneous fields ($`\varphi _c\varphi _c(t)`$, $`\psi _c\psi _c(t)`$), thus, at the one-loop order we can write the following EOMs for $`\varphi _c`$ and $`\psi _c`$, respectively, as<sup>*</sup><sup>*</sup>*Note that the mixed field averages in (7) and (9) can only be treated within perturbation theory. (where overdots mean time derivatives)
$`\ddot{\varphi }_c`$ $`+`$ $`m_\varphi ^2\varphi _c+{\displaystyle \frac{\lambda _\varphi }{6}}\varphi _c^3+g^2\varphi _c\psi _c^2+{\displaystyle \frac{\lambda _\varphi }{2}}\varphi _c\varphi ^2`$ (6)
$`+`$ $`g^2\varphi _c\psi ^2+2g^2\psi _c\varphi \psi =0`$ (7)
and
$`\ddot{\psi }_c`$ $`+`$ $`m_\psi ^2\psi _c+{\displaystyle \frac{\lambda _\psi }{6}}\psi _c^3+g^2\psi _c\varphi _c^2+{\displaystyle \frac{\lambda _\psi }{2}}\psi _c\psi ^2`$ (8)
$`+`$ $`g^2\psi _c\varphi ^2+2g^2\varphi _c\varphi \psi =0,`$ (9)
where $`\varphi ^2`$ and $`\psi ^2`$ are given in terms of the coincidence limit of the (causal) two-point Greenโs functions $`G_\varphi ^{++}(x,x^{})`$ and $`G_\psi ^{++}(x,x^{})`$, which are obtained from the $`(1,1)`$-component of the real time matrix of full propagators which satisfy the appropriate Schwinger-Dyson equations (see, e.g., Refs. and for further details):
$`\left[\mathrm{}+m_\varphi ^2+{\displaystyle \frac{\lambda _\varphi }{2}}\varphi _c^2+g^2\psi _c^2\right]G_\varphi (x,x^{})`$ (11)
$`+{\displaystyle d^4z\mathrm{\Sigma }_\varphi (x,z)G_\varphi (z,x^{})}=i\delta (x,x^{}),`$
and
$`\left[\mathrm{}+m_\psi ^2+{\displaystyle \frac{\lambda _\psi }{2}}\psi _c^2+g^2\varphi _c^2\right]G_\psi (x,x^{})`$ (13)
$`+{\displaystyle d^4z\mathrm{\Sigma }_\psi (x,z)G_\psi (z,x^{})}=i\delta (x,x^{}),`$
where $`\mathrm{\Sigma }_\varphi (x,x^{})`$ and $`\mathrm{\Sigma }_\psi (x,x^{})`$ are the self-energies for $`\mathrm{\Phi }`$ e $`\mathrm{\Psi }`$, respectively. The momentum-space Fourier transform of $`G(x,x^{})`$ (for both fields) can be expressed in the form
$$G(x,x^{})=i\frac{d^3q}{(2\pi )^3}e^{i๐ช.(๐ฑ๐ฑ^{})}\left(\begin{array}{cc}G^{++}(๐ช,tt^{})\hfill & G^+(๐ช,tt^{})\hfill \\ G^+(๐ช,tt^{})\hfill & G^{}(๐ช,tt^{})\hfill \end{array}\right),$$
(14)
where
$`G^{++}(๐ช,tt^{})=G^>(๐ช,tt^{})\theta (tt^{})+G^<(๐ช,tt^{})\theta (t^{}t)`$ (18)
$`G^{}(๐ช,tt^{})=G^>(๐ช,tt^{})\theta (t^{}t)+G^<(๐ช,tt^{})\theta (tt^{})`$
$`G^+(๐ช,tt^{})=G^<(๐ช,tt^{})`$
$`G^+(๐ช,tt^{})=G^>(๐ช,tt^{}),`$
and the fully dressed two-point functions, at a finite temperature $`T=1/\beta `$, are given byR.O.R. thanks I. Lawrie for pointing him the correct form for these expressions.
$`G^>(๐ช,tt^{})`$ $`=`$ $`{\displaystyle \frac{1}{2\omega }}\left\{[1+n(\omega i\mathrm{\Gamma })]e^{i(\omega i\mathrm{\Gamma })(tt^{})}+n(\omega +i\mathrm{\Gamma })e^{i(\omega +i\mathrm{\Gamma })(tt^{})}\right\}\theta (tt^{})`$ (19)
$`+`$ $`{\displaystyle \frac{1}{2\omega }}\left\{[1+n(\omega +i\mathrm{\Gamma })]e^{i(\omega +i\mathrm{\Gamma })(tt^{})}+n(\omega i\mathrm{\Gamma })e^{i(\omega i\mathrm{\Gamma })(tt^{})}\right\}\theta (t^{}t),`$ (20)
$`G^<(๐ช,tt^{})`$ $`=`$ $`G^>(๐ช,t^{}t),`$ (21)
where $`n(\omega )`$ is the Bose distribution, $`\omega \omega (๐ช)`$ is the particleโs dispersion relation and $`\mathrm{\Gamma }`$ is the decay width, defined as usual in terms of the field self-energy by
$$\mathrm{\Gamma }(q)=\frac{\mathrm{Im}\mathrm{\Sigma }(๐ช,\omega )}{2\omega }.$$
(22)
Assuming the couplings $`g,\lambda _\varphi ,\lambda _\psi 1`$, so that perturbation theory can be consistently formulated and subleading terms can be neglected, then by perturbatively expanding the field averages in Eqs. (7) and (9), we can write for the EOMs
$`\ddot{\varphi }_c(t)+\stackrel{~}{m}_\varphi ^2\varphi _c(t)+{\displaystyle \frac{\lambda _\varphi }{6}}\varphi _c^3(t)+g^2\varphi _c(t)\psi _c^2(t)`$ (26)
$`+\lambda _\varphi \varphi _c(t){\displaystyle _{\mathrm{}}^t}๐t^{}\left[{\displaystyle \frac{\lambda _\varphi }{2}}\varphi _c^2(t^{})+g^2\psi _c^2(t^{})\right]{\displaystyle \frac{d^3๐ช}{(2\pi )^3}\mathrm{Im}\left[G_\varphi ^{++}(๐ช,tt^{})\right]^2}`$
$`+2g^2\varphi _c(t){\displaystyle _{\mathrm{}}^t}๐t^{}\left[{\displaystyle \frac{\lambda _\psi }{2}}\psi _c^2(t^{})+g^2\varphi _c^2(t^{})\right]{\displaystyle \frac{d^3๐ช}{(2\pi )^3}\mathrm{Im}\left[G_\psi ^{++}(๐ช,tt^{})\right]^2}`$
$`+8g^4\psi _c(t){\displaystyle _{\mathrm{}}^t}๐t^{}\varphi _c(t^{})\psi _c(t^{}){\displaystyle \frac{d^3๐ช}{(2\pi )^3}\mathrm{Im}\left[G_\varphi ^{++}(๐ช,tt^{})G_\psi ^{++}(๐ช,tt^{})\right]}=0`$
and
$`\ddot{\psi }_c(t)+\stackrel{~}{m}_\psi ^2\psi _c(t)+{\displaystyle \frac{\lambda _\psi }{6}}\psi _c^3(t)+g^2\psi _c(t)\varphi _c^2(t)`$ (30)
$`+\lambda _\psi \psi _c(t){\displaystyle _{\mathrm{}}^t}๐t^{}\left[{\displaystyle \frac{\lambda _\psi }{2}}\psi _c^2(t^{})+g^2\varphi _c^2(t^{})\right]{\displaystyle \frac{d^3๐ช}{(2\pi )^3}\mathrm{Im}\left[G_\psi ^{++}(๐ช,tt^{})\right]^2}`$
$`+2g^2\psi _c(t){\displaystyle _{\mathrm{}}^t}๐t^{}\left[{\displaystyle \frac{\lambda _\varphi }{2}}\varphi _c^2(t^{})+g^2\psi _c^2(t^{})\right]{\displaystyle \frac{d^3๐ช}{(2\pi )^3}\mathrm{Im}\left[G_\varphi ^{++}(๐ช,tt^{})\right]^2}`$
$`+8g^4\varphi _c(t){\displaystyle _{\mathrm{}}^t}๐t^{}\varphi _c(t^{})\psi _c(t^{}){\displaystyle \frac{d^3๐ช}{(2\pi )^3}\mathrm{Im}\left[G_\varphi ^{++}(๐ช,tt^{})G_\psi ^{++}(๐ช,tt^{})\right]}=0,`$
where
$`\stackrel{~}{m}_\varphi ^2`$ $`=`$ $`m_\varphi ^2+{\displaystyle \frac{\lambda _\varphi }{2}}\varphi ^2_0+g^2\psi ^2_0,`$ (31)
$`\stackrel{~}{m}_\psi ^2`$ $`=`$ $`m_\psi ^2+{\displaystyle \frac{\lambda _\psi }{2}}\psi ^2_0+g^2\varphi ^2_0,`$ (32)
with $`\mathrm{}_0`$ meaning fields independent averages. The nonlocal terms of the type appearing in the above equations have been shown in Refs. to lead to dissipative dynamics in the EOMs. This can be made more explicit by an appropriate integration by parts in the time integrals in Eqs. (26) and (30) (see Ref. ) to obtain the result
$`\ddot{\varphi }_c(t)+\stackrel{~}{m}_\varphi ^2\varphi _c(t)+{\displaystyle \frac{\stackrel{~}{\lambda }_\varphi }{6}}\varphi _c^3(t)+\stackrel{~}{g}^2\varphi _c(t)\psi _c^2(t)`$ (35)
$`+\varphi _c(t){\displaystyle _{\mathrm{}}^t}๐t^{}\varphi _c(t^{})\dot{\varphi }_c(t^{})F_1(t,t^{})+\varphi _c(t){\displaystyle _{\mathrm{}}^t}๐t^{}\psi _c(t^{})\dot{\psi }_c(t^{})F_3(t,t^{})`$
$`+\psi _c(t){\displaystyle _{\mathrm{}}^t}๐t^{}\left[\varphi _c(t^{})\dot{\psi }_c(t^{})+\dot{\varphi }_c(t^{})\psi _c(t^{})\right]F_4(t,t^{})=0,`$
and
$`\ddot{\psi }_c(t)+\stackrel{~}{m}_\psi ^2\psi _c(t)+{\displaystyle \frac{\stackrel{~}{\lambda }_\psi }{6}}\psi _c^3(t)+\stackrel{~}{g}^2\psi _c(t)\varphi _c^2(t)`$ (38)
$`+\psi _c(t){\displaystyle _{\mathrm{}}^t}๐t^{}\psi _c(t^{})\dot{\psi }_c(t^{})F_2(t,t^{})+\psi _c(t){\displaystyle _{\mathrm{}}^t}๐t^{}\varphi _c(t^{})\dot{\varphi }_c(t^{})F_3(t,t^{})`$
$`+\varphi _c(t){\displaystyle _{\mathrm{}}^t}๐t^{}\left[\varphi _c(t^{})\dot{\psi }_c(t^{})+\dot{\varphi }_c(t^{})\psi _c(t^{})\right]F_4(t,t^{})=0,`$
where the dissipative kernels $`F_1,F_2,F_3`$ and $`F_4`$ are given by
$`F_1(t,t^{})`$ $`=`$ $`{\displaystyle ๐t^{}\frac{d^3๐ช}{(2\pi )^3}\left\{\lambda _\varphi ^2\mathrm{Im}\left[G_\varphi ^{++}(๐ช,tt^{})\right]^2+4g^4\mathrm{Im}\left[G_\psi ^{++}(๐ช,tt^{})\right]^2\right\}},`$ (39)
$`F_2(t,t^{})`$ $`=`$ $`{\displaystyle ๐t^{}\frac{d^3๐ช}{(2\pi )^3}\left\{\lambda _\psi ^2\mathrm{Im}\left[G_\psi ^{++}(๐ช,tt^{})\right]^2+4g^4\mathrm{Im}\left[G_\varphi ^{++}(๐ช,tt^{})\right]^2\right\}},`$ (40)
$`F_3(t,t^{})`$ $`=`$ $`2g^2{\displaystyle ๐t^{}\frac{d^3๐ช}{(2\pi )^3}\left\{\lambda _\varphi \mathrm{Im}\left[G_\varphi ^{++}(๐ช,tt^{})\right]^2+\lambda _\psi \mathrm{Im}\left[G_\psi ^{++}(๐ช,tt^{})\right]^2\right\}},`$ (41)
$`F_4(t,t^{})`$ $`=`$ $`8g^4{\displaystyle ๐t^{}\frac{d^3๐ช}{(2\pi )^3}\mathrm{Im}\left[G_\varphi ^{++}(๐ช,tt^{})G_\psi ^{++}(๐ช,tt^{})\right]},`$ (42)
and $`\overline{\lambda _\varphi }`$, $`\overline{\lambda _\psi }`$ and $`\overline{g}^2`$ (the one-loop effective coupling constants) in Eqs. (35) and (38) are given by
$`\overline{\lambda _\varphi }`$ $`=`$ $`\lambda _\varphi +{\displaystyle ๐t^{}\frac{d^3๐ช}{(2\pi )^3}\left\{\frac{\lambda _\varphi ^2}{2}\mathrm{Im}\left[G_\varphi ^{++}(๐ช,tt^{})\right]^2+2g^4\mathrm{Im}\left[G_\psi ^{++}(๐ช,tt^{})\right]^2\right\}}|_{t^{}=t},`$ (43)
$`\overline{\lambda _\psi }`$ $`=`$ $`\lambda _\psi +{\displaystyle ๐t^{}\frac{d^3๐ช}{(2\pi )^3}\left\{\frac{\lambda _\psi ^2}{2}\mathrm{Im}\left[G_\psi ^{++}(๐ช,tt^{})\right]^2+2g^4\mathrm{Im}\left[G_\varphi ^{++}(๐ช,tt^{})\right]^2\right\}}|_{t^{}=t},`$ (44)
$`\overline{g}^2`$ $`=`$ $`g^2+g^2{\displaystyle }dt^{}{\displaystyle }{\displaystyle \frac{d^3๐ช}{(2\pi )^3}}\{\lambda _\varphi \mathrm{Im}\left[G_\varphi ^{++}(๐ช,tt^{})\right]^2+\lambda _\psi \mathrm{Im}\left[G_\psi ^{++}(๐ช,tt^{})\right]^2`$ (45)
$`+`$ $`8g^2\mathrm{Im}\left[G_\varphi ^{++}(๐ช,tt^{})G_\psi ^{++}(๐ช,tt^{})\right]\left\}\right|_{t^{}=t},`$ (46)
## IV Coupling to Bath Degrees of Freedom and an Approximate Form for the EOMs
The two coupled nonlocal EOMs, Eqs. (35) and (38) (or equivalently, Eqs. (26) and (30)) are too complicated to directly numerically work with them. The main difficulty in handling these equations comes from the dissipative like terms in Eqs. (35) and (38), which have the non-Markovian kernels shown in Eq. (42). If we suppose that there is a Markovian limit for those kernels, then by using Eqs. (18) and (21) in Eq. (42), we find that at $`T=0`$ the kernels diverge logarithmic (a result also found in Ref. for the single scalar field case). However, in the high temperature limit ($`\mathrm{\Gamma }/\omega 1`$ and $`\mathrm{\Gamma }/T1`$) it has been argued in Ref. that a Markovian approximation exists and a finite result for the dissipation coefficients can be found. Such an approximation for the kernels given in Eq. (42), we can also find here for our specific problem, in which case we may then, in principle, write for the dissipation terms in Eqs. (35) and (38) the approximate expressions
$`\varphi _c(t){\displaystyle _{\mathrm{}}^t}๐t^{}\varphi _c(t^{})\dot{\varphi }_c(t^{})F_1(t,t^{})`$ $``$ $`\varphi _c^2(t)\dot{\varphi }_c(t)\eta _1,`$ (47)
$`\psi _c(t){\displaystyle _{\mathrm{}}^t}๐t^{}\psi _c(t^{})\dot{\psi }_c(t^{})F_2(t,t^{})`$ $``$ $`\psi _c^2(t)\dot{\psi }_c(t)\eta _2,`$ (48)
$`\varphi _c(t){\displaystyle _{\mathrm{}}^t}๐t^{}\psi _c(t^{})\dot{\psi }_c(t^{})F_3(t,t^{})`$ $``$ $`\varphi _c(t)\psi _c(t)\dot{\psi }_c(t)\eta _3,`$ (49)
$`\psi _c(t){\displaystyle _{\mathrm{}}^t}๐t^{}\varphi _c(t^{})\dot{\varphi }_c(t^{})F_3(t,t^{})`$ $``$ $`\varphi _c(t)\psi _c(t)\dot{\varphi }_c(t)\eta _3,`$ (50)
$`\psi _c(t){\displaystyle _{\mathrm{}}^t}๐t^{}\left[\varphi _c(t^{})\dot{\psi }_c(t^{})+\dot{\varphi }_c(t^{})\psi _c(t^{})\right]F_4(t,t^{})`$ $``$ $`\psi _c(t)\left[\varphi _c(t)\dot{\psi }_c(t)+\dot{\varphi }_c(t)\psi _c(t)\right]\eta _4,`$ (51)
$`\varphi _c(t){\displaystyle _{\mathrm{}}^t}๐t^{}\left[\varphi _c(t^{})\dot{\psi }_c(t^{})+\dot{\varphi }_c(t^{})\psi _c(t^{})\right]F_4(t,t^{})`$ $``$ $`\varphi _c(t)\left[\varphi _c(t)\dot{\psi }_c(t)+\dot{\varphi }_c(t)\psi _c(t)\right]\eta _4,`$ (52)
where $`\eta _1,\eta _2,\eta _3`$ and $`\eta _4`$ in the above expressions are given by (using Eqs. (18) and (21) in Eqs. (35) and (38), and in the high temperature approximation $`\mathrm{\Gamma }_\varphi /T,\mathrm{\Gamma }_\psi /T1`$, with $`\mathrm{\Gamma }_\varphi /\omega _\varphi ,\mathrm{\Gamma }_\psi /\omega _\psi ,1`$ and $`\overline{m}_\varphi ^2\overline{m}_\psi ^2`$)
$`\eta _1`$ $``$ $`{\displaystyle \frac{\lambda _\varphi ^2}{8}}\beta {\displaystyle \frac{d^3q}{(2\pi )^3}\frac{n_\varphi (1+n_\varphi )}{\omega _\varphi ^2\mathrm{\Gamma }_\varphi }}+{\displaystyle \frac{g^4}{2}}\beta {\displaystyle \frac{d^3q}{(2\pi )^3}\frac{n_\psi (1+n_\psi )}{\omega _\psi ^2\mathrm{\Gamma }_\psi }}+๐ช(\lambda _\varphi ^2{\displaystyle \frac{\mathrm{\Gamma }_\varphi }{\omega _\varphi }},g^4{\displaystyle \frac{\mathrm{\Gamma }_\psi }{\omega _\psi }}),`$ (53)
$`\eta _2`$ $``$ $`{\displaystyle \frac{\lambda _\psi ^2}{8}}\beta {\displaystyle \frac{d^3q}{(2\pi )^3}\frac{n_\psi (1+n_\psi )}{\omega _\psi ^2\mathrm{\Gamma }_\psi }}+{\displaystyle \frac{g^4}{2}}\beta {\displaystyle \frac{d^3q}{(2\pi )^3}\frac{n_\varphi (1+n_\varphi )}{\omega _\varphi ^2\mathrm{\Gamma }_\varphi }}+๐ช(\lambda _\psi ^2{\displaystyle \frac{\mathrm{\Gamma }_\psi }{\omega _\psi }},g^4{\displaystyle \frac{\mathrm{\Gamma }_\varphi }{\omega _\varphi }}),`$ (54)
$`\eta _3`$ $``$ $`\beta {\displaystyle \frac{\lambda _\varphi g^2}{4}}{\displaystyle \frac{d^3p}{(2\pi )^3}\frac{n_\varphi (1+n_\varphi )}{\omega _\varphi ^2\mathrm{\Gamma }_\varphi }}+\beta {\displaystyle \frac{\lambda _\psi g^2}{4}}{\displaystyle \frac{d^3p}{(2\pi )^3}\frac{n_\psi (1+n_\psi )}{\omega _\psi ^2\mathrm{\Gamma }_\psi }}+๐ช(\lambda _\varphi g^2{\displaystyle \frac{\mathrm{\Gamma }_\varphi }{\omega _\varphi }},\lambda _\psi g^2{\displaystyle \frac{\mathrm{\Gamma }_\psi }{\omega _\psi }}),`$ (55)
$`\eta _4`$ $``$ $`8g^4\beta {\displaystyle \frac{d^3q}{(2\pi )^3}\frac{1}{(\omega _\varphi ^2\omega _\psi ^2)^2}\left[n_\varphi (1+n_\varphi )\mathrm{\Gamma }_\varphi +n_\psi (1+n_\psi )\mathrm{\Gamma }_\psi \right]}+๐ช(g^4{\displaystyle \frac{\mathrm{\Gamma }_\varphi ^2}{\omega _\varphi ^2}},g^4{\displaystyle \frac{\mathrm{\Gamma }_\psi ^2}{\omega _\psi ^2}}).`$ (56)
Similar time non-localities as the ones appearing in Eqs. (26) and (30) have also been dealt with in Refs. by using an adiabatic (or sudden) approximation for the fields. The dissipation coefficients obtained by that approximation are just the same as the ones given in (56). As shown in Ref. the adiabatic approximation for the nonlocal kernels is a consistent approximation in the case the fields are in an overdamped regime. The strong field dissipation responsible for the overdamped regime can be attained by coupling both $`\mathrm{\Phi }`$ and $`\mathrm{\Psi }`$ fields to a large set of other fields making the bath in which $`\mathrm{\Phi }`$ and $`\mathrm{\Psi }`$ evolve, which then enlarges the number of field decay channels available for both fields. This idea was used in Ref. for the construction of an alternative inflationary model.
### A The Validity of the One-Loop and Markovian Approximations
The consideration of the coupling of $`\mathrm{\Phi }`$ and $`\mathrm{\Psi }`$ to a large set of bath field degrees of freedom is particularly relevant to the validity of the Markovian approximation taken for the dissipative kernels and in order to make a clear assessment of the regime of validity not only of this approximation but also for the validity of the one-loop approximation we used to derive the EOMs. The study of the validity of these approximations are clearly important to our study which is the study of chaotic behavior in the dynamics of our system of equations, and by making clear that chaos in our dynamical system is not just an artifact of the approximations taken. In order to address these important issues, let us consider the coupling of both $`\mathrm{\Phi }`$ and $`\mathrm{\Psi }`$ to a set of $`N`$ (scalar) fields $`\chi `$ making the bath, which are coupled in the following way:
$$f_\varphi ^2\underset{i=1}{\overset{N}{}}\mathrm{\Phi }^2\chi _i^2+f_\psi ^2\underset{i=1}{\overset{N}{}}\mathrm{\Psi }^2\chi _i^2.$$
(57)
We next study how our main quantities scale with the various coupling constants and $`N`$ in the large N limitR.O.R. deeply thanks S. Jeon for correspondence regarding the large N contributions to dissipation and the consistency of the adiabatic approximation in the large N limit.. This analysis will allow us to make a qualitative and clear assessment of our main approximations.
Let us first consider that the various coupling constants we have scale with $`N`$ in the following way:
$$\lambda _\varphi \lambda _\psi g^2\frac{\alpha }{\sqrt{N}},$$
(58)
and
$$f_\varphi ^2f_\psi ^2\frac{\alpha ^2}{N},$$
(59)
with $`\alpha 1`$. The choice for the coupling constants taken above is consistent in the sense that higher order quantum corrections to quantities like the field masses and coupling constants will not blow up in the large N limit. For instance, as $`N\mathrm{}`$ the effective masses $`\overline{m}_\varphi ^2`$ and $`\overline{m}_\psi ^2`$ scale as (in the high temperature limit)
$`\overline{m}_\varphi ^2`$ $`=`$ $`m_\varphi ^2+๐ช(\lambda _\varphi T^2)+๐ช(g^2T^2)+๐ช(Nf_\varphi ^2T^2)m_\varphi ^2+๐ช(\alpha ^2T^2),`$ (60)
$`\overline{m}_\psi ^2`$ $`=`$ $`m_\psi ^2+๐ช(\lambda _\psi T^2)+๐ช(g^2T^2)+๐ช(Nf_\psi ^2T^2)m_\psi ^2+๐ช(\alpha ^2T^2),`$ (61)
and the effective coupling constants go like
$`\overline{\lambda }_\varphi `$ $`=`$ $`\lambda _\varphi +๐ช(\lambda _\varphi ^2)+๐ช(g^4)+๐ช(Nf_\varphi ^4)\lambda _\varphi +๐ช\left({\displaystyle \frac{\alpha ^2}{N}}\right),`$ (62)
$`\overline{\lambda }_\psi `$ $`=`$ $`\lambda _\psi +๐ช(\lambda _\psi ^2)+๐ช(g^4)+๐ช(Nf_\psi ^4)\lambda _\psi +๐ช\left({\displaystyle \frac{\alpha ^2}{N}}\right),`$ (63)
$`\overline{g}^2`$ $`=`$ $`g^2+๐ช(\lambda _\varphi g^2)+๐ช(\lambda _\psi g^2)+๐ช(g^4)+๐ช(Nf_\varphi ^2f_\psi ^2)g^2+๐ช\left({\displaystyle \frac{\alpha ^2}{N}}\right).`$ (64)
We next determine the scaling of the fields decay widths $`\mathrm{\Gamma }_\varphi `$, $`\mathrm{\Gamma }_\psi `$ and for those of the bath, $`\mathrm{\Gamma }_{\chi _i}`$, which determines the time-scale for collisions for the system in interaction with the (thermal) bath. We begin by noticing that the imaginary part of the field self-energies contributing to the decay widths are typically dominated by momenta $`|๐ช|T`$ . We can then find for the decay widths the relations:
$`\mathrm{\Gamma }_\varphi `$ $`=`$ $`๐ช(\lambda _\varphi ^2T)+๐ช(g^4T)+๐ช(Nf_\varphi ^4T)๐ช\left({\displaystyle \frac{\alpha ^2}{N}}T\right),`$ (65)
$`\mathrm{\Gamma }_\psi `$ $`=`$ $`๐ช(\lambda _\psi ^2T)+๐ช(g^4T)+๐ช(Nf_\psi ^4T)๐ช\left({\displaystyle \frac{\alpha ^2}{N}}T\right),`$ (66)
$`\mathrm{\Gamma }_{\chi _i}`$ $`=`$ $`๐ช(f_\varphi ^4T)+๐ช(f_\psi ^4T)๐ช\left({\displaystyle \frac{\alpha ^4}{N^2}}T\right).`$ (67)
Using these in the expressions for the dissipative kernels, Eq. (42), we can then show that the dissipation coefficients $`\eta _1,\mathrm{},\eta _4`$, appearing in Eq. (52), have magnitude given by
$`\eta _1`$ $``$ $`๐ช\left({\displaystyle \frac{\lambda _\varphi ^2}{\mathrm{\Gamma }_\varphi }}\right)+๐ช\left({\displaystyle \frac{g^4}{\mathrm{\Gamma }_\psi }}\right)+๐ช\left(N{\displaystyle \frac{f_\varphi ^4}{\mathrm{\Gamma }_{\chi _i}}}\right)๐ช\left({\displaystyle \frac{N}{T}}\right),`$ (68)
$`\eta _2`$ $``$ $`๐ช\left({\displaystyle \frac{\lambda _\psi ^2}{\mathrm{\Gamma }_\psi }}\right)+๐ช\left({\displaystyle \frac{g^4}{\mathrm{\Gamma }_\varphi }}\right)+๐ช\left(N{\displaystyle \frac{f_\psi ^4}{\mathrm{\Gamma }_{\chi _i}}}\right)๐ช\left({\displaystyle \frac{N}{T}}\right),`$ (69)
$`\eta _3`$ $``$ $`๐ช\left({\displaystyle \frac{g^2\lambda _\varphi }{\mathrm{\Gamma }_\varphi }}\right)+๐ช\left({\displaystyle \frac{g^2\lambda _\psi }{\mathrm{\Gamma }_\psi }}\right)+๐ช\left(N{\displaystyle \frac{f_\varphi ^2f_\psi ^2}{\mathrm{\Gamma }_{\chi _i}}}\right)๐ช\left({\displaystyle \frac{N}{T}}\right),`$ (70)
$`\eta _4`$ $``$ $`๐ช\left(g^4\mathrm{\Gamma }_\varphi +g^4\mathrm{\Gamma }_\psi \right)๐ช\left({\displaystyle \frac{\alpha ^4}{N^2T}}\right).`$ (71)
Therefore, the above estimates show that, for $`\alpha ^21`$ and in the large N limit, the dissipation kernel $`F_4`$ and its related dissipation coefficient in the Markov approximation are negligible compared with the others one. The behavior of the remaining coefficients reproduce the typical behavior found previously for the dissipation coefficients in Refs. and . Additionally, we can also find that higher loop contributions to dissipation (for instance the ones coming from two-loop diagrams and higher) are all at most of order $`๐ช(\lambda _\varphi ^2\mathrm{\Gamma }_\varphi )`$, $`๐ช(\lambda _\psi ^2\mathrm{\Gamma }_\psi )`$, $`๐ช(g^4\mathrm{\Gamma }_\varphi +g^4\mathrm{\Gamma }_\psi )`$ or $`๐ช(Nf^4\mathrm{\Gamma }_\chi )`$, which are all subleading in the perturbative regime. All these simple qualitative estimative allow us then to assess the validity of both the Markov and one-loop approximations used in this study.
## V The Dynamical System and Chaotic Behavior
From the results obtained in the last Section, we can then write Eqs. (35) and (38) in the local form shown below and more suitable for the numerical analysis as:
$`\ddot{\varphi }_c+\overline{m}_\varphi ^2\varphi _c+{\displaystyle \frac{\overline{\lambda }_\varphi }{6}}\varphi _c^3+\overline{g}^2\varphi _c\psi _c^2+\eta _1\varphi _c^2\dot{\varphi _c}+\eta _3\varphi _c\psi _c\dot{\psi }_c=0`$ (72)
and
$`\ddot{\psi }_c+\overline{m}_\psi ^2\psi _c+{\displaystyle \frac{\overline{\lambda }_\psi }{6}}\psi _c^3+\overline{g}^2\psi _c\varphi _c^2+\eta _2\psi _c^2\dot{\psi _c}+\eta _{\varphi \psi }\varphi _c\psi _c\dot{\varphi }_c=0,`$ (73)
where $`\overline{\lambda }_\varphi \lambda _\varphi ,\overline{\lambda }_\psi \lambda _\psi `$ and $`\overline{g}^2g^2`$ denote the renormalized couplings <sup>ยง</sup><sup>ยง</sup>ยงThe renormalization of the coupling constants and masses can be made by just the standard way, by introducing the appropriate counterterms of renormalization in our model Lagrangian .. $`\eta _1`$, $`\eta _2`$ and $`\eta _3`$ denote the dissipation coefficients and we have neglect the term involving $`\eta _4`$ due to the estimates shown in (71). Although these coefficients can be explicitly evaluated, however, we here refrain ourselves from an explicit evaluation of these terms, which, as shown in the previous Section, depend on the various parameters of the model, temperature and on the number $`N`$ of other field degrees making the environment that $`\mathrm{\Phi }`$ and $`\mathrm{\Psi }`$ may be coupled to. In fact, the magnitude of the dissipation terms may be controlled by these additional field couplings, as shown in the simple estimates obtained in the previous Section. Therefore, for the sake of simplicity we just take $`\eta _1`$, $`\eta _2`$ and $`\eta _3`$ as additional free constant parameters.
Next, let us define the following constants
$`a^2`$ $`=`$ $`{\displaystyle \frac{\overline{m}_\varphi ^2}{6|\overline{m}_\psi ^2|}},`$ (74)
$`G^2`$ $`=`$ $`{\displaystyle \frac{\overline{g}^2}{a^2\overline{\lambda }_\psi }},`$ (75)
$`\lambda _x`$ $`=`$ $`{\displaystyle \frac{\overline{\lambda }_\varphi }{a^4\overline{\lambda }_\psi }},`$ (76)
in term of which we can define the dimensionless variables:
$`x`$ $`=`$ $`\sqrt{\overline{\lambda }_\psi }{\displaystyle \frac{a^2}{\overline{m}_\varphi }}\varphi _c,`$ (77)
$`z`$ $`=`$ $`{\displaystyle \frac{a^2}{\overline{m}_\varphi \mathrm{\Psi }_v}}\dot{\varphi }_c,`$ (78)
$`y`$ $`=`$ $`{\displaystyle \frac{1}{\mathrm{\Psi }_v}}\psi _c,`$ (79)
$`w`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{6|\overline{m}_\psi ^2|}\mathrm{\Psi }_v}}\dot{\psi }_c,`$ (80)
and by also rescaling time and dissipation coefficients as $`t^{}=\sqrt{6|\overline{m}_\psi ^2|}t`$, $`\eta _1=\eta _xa^4\sqrt{\overline{\lambda }_\psi }/\mathrm{\Psi }_v`$, $`\eta _2=\eta _y\sqrt{\overline{\lambda }_\psi }/\mathrm{\Psi }_v`$ and $`\eta _3=\eta _{xy}a^2\sqrt{\overline{\lambda }_\psi }/\mathrm{\Psi }_v`$, respectively, we can then write Eqs. (72) and (73) in terms of the following dimensionless first-order differential system of equations:
$`\dot{x}`$ $`=`$ $`z`$ (81)
$`\dot{z}`$ $`=`$ $`a^2\left(x+{\displaystyle \frac{\lambda _x}{6}}x^3+G^2xy^2+\eta _xx^2z+\eta _{xy}xyw\right)`$ (82)
$`\dot{y}`$ $`=`$ $`w`$ (83)
$`\dot{w}`$ $`=`$ $`{\displaystyle \frac{y}{6}}{\displaystyle \frac{y^3}{6}}G^2x^2y\eta _yy^2w\eta _{xy}xyz.`$ (84)
For convenience we also choose parameters such that $`a^2=1`$, $`G^2=1`$ and $`\lambda _x=1`$, which means we consider $`\overline{m}_\varphi ^2=6|\overline{m}_\psi ^2|`$ and $`\overline{g}^2=\overline{\lambda }_\varphi =\overline{\lambda }_\psi `$, which is a choice consistent with the considerations used in the previous Section, Eq. (58). We also take as base values for the (dimensionless) dissipation coefficients the values $`\{\eta \}=(\eta _x,\eta _y,\eta _{xy})=(1/120,1/240,1/200)`$ This is consistent with the general expressions shown for the dissipation kernels $`F_1,F_2`$ and $`F_3`$, which for $`\lambda _\varphi =\lambda _\psi =g^2`$, satisfy the simple relation between then: $`2(F_1+F_2)=5F_3`$., for which the dynamics displayed by (84) happen in the weakly damped regime. We also make the special consideration that the temperature of the system, which is kept fixed at a value $`T`$, is large enough such that the high temperature approximation involved in the Markov limit for the dissipation kernels is valid, but smaller than the critical temperature, $`T_{c_\psi }`$, for phase transition in the $`\mathrm{\Psi }`$ field direction, $`T_{c_\psi }>T\mathrm{\Gamma }_\varphi ,\mathrm{\Gamma }_\psi ,\overline{m}_\varphi ,\overline{m}_\psi `$, where $`T_{c_\psi }^212|m_\psi ^2|/(g^2+\lambda _\psi /2)`$.
We then numerically solve the dynamical system with initial conditions taken such that at $`t=0`$ the potential in the Lagrangian density (2) is symmetrical in both field directions. The system is then evolved in time till the symmetry breaking in the $`\mathrm{\Psi }`$-field direction occurs. We then look for chaotic regimes as $`\eta _\varphi `$, $`\eta _\psi `$ and $`\eta _{\varphi \psi }`$ are changed. Our choice for initial conditions to numerically solving the (dissipative) dynamical system (84) is as follows. At the initial time we consider
$$(\varphi _c,\dot{\varphi }_c,\psi _c,\dot{\psi }_c)|_{t=0}=(4\mathrm{\Phi }_{\mathrm{cr}},0,0,0).$$
(85)
A typical result for the evolution of the fields in time is shown in Fig. 1, which already shows a highly chaotic dynamics prior to symmetry breaking.
Following the method of Box-Counting , around the initial condition (85) it is then considered a box in phase space (for the dimensionless variables) of size $`10^5`$, inside which a large number of random points are taken (a total of $`200.000`$ random points were used in each run). All initial conditions are then numerically evolved by using an eighth-order Runge-Kutta integration method and the fractal dimension is obtained by statistically studying the outcome of each initial condition at each run of the large set of points. Special care is taken to keep the statistical error in the results always below $`1\%`$. The results obtained are shown in Table I for different values of dissipation coefficients. In this table we also show the uncertainty exponent $`ฯต`$ (see Ref. ), which gives a measure of how chaotic is the system. Qualitatively speaking we can say that the closest is $`ฯต`$ of zero, the more chaotic is the system. On the contrary, the closest is $`ฯต`$ of unit, the less chaotic is the system. We see clearly the effect of dissipation on the nonlinearities of the system. It can change fast from a chaotic to an integrable regime with a relatively small increase of the dissipation. Larger dissipations tend also to destroy fast the chaotic attractors. In Fig. 2 we show an example of the structure of the chaotic attractors in the $`\psi _c,\dot{\psi }_c`$ plane ($`y,w`$ plane).
## VI Conclusions
In this paper we have studied the dynamical system made by the leading order effective equations of motion for a model of two coupled scalar fields. The chaotic behavior for (ensemble averaged) field trajectories has been demonstrated and we have also shown that in the overdamped regime chaos gets completely suppressed. Chaotic motion can only develop in the underdamped or very weakly damped regime, in which case enough energy can be exchanged fast enough from one field to the other, making both $`\varphi _c`$ and $`\psi _c`$ to fluctuate with large enough amplitudes, leading to a highly nonlinear behavior that then precludes the chaotic motion of the system.
It seems also that we can indirectly associate the chaoticity in the system with the equilibration rates of the fields, in close analogy with the one found between the Lyapunov exponent and the thermalization rate in perturbative thermal gauge theory . We note from the results obtained here and from the numerical simulations we performed, that the smaller is the fractal dimension (or the larger is $`ฯต`$) the fastest the fields equilibrate to their asymptotic states by loosing their energies to radiation, which will then eventually thermalize. A clear assessment to this interesting point, however, must be carefully studied from the complete nonlocal EOMs, which then restores the time reversal invariance of the equations of motion.
Finally we would like to point out that the kind of model we have studied here and its generalization to larger number of fields is natural to be found in extensions of the standard model with large scalar sectors. Physical implementations of the model can be found, for example, in particle physics or condensed matter models displaying multiple stages of phase transitions, in which case the dynamics we have studied here would be likely to manifest between any of that stages and, therefore, with consequences to the phenomenology of that models. The model studied here has also a strong motivation from inflationary models (hybrid inflation) . In special, in this context, a model Lagrangian of a similar form of the one we studied here has been studied in Ref. , showing the possibility of chaotic behavior during the final stages of inflation. However, the authors in make use of the classical equations of motion. It would be extremely interesting to assess the effect of quantum effects (and particle productions) and consequently dissipation also in that context, which is of fundamental relevance for the description of the process of reheating. In the particular study performed here, our results could apply instead to the description of the pre-inflationary stage, with possible contributions to the discussion of the fine-tuning problem of the initial field configuration in hybrid inflation . These problems are currently being examined by us and more results and details will be reported elsewhere.
###### Acknowledgements.
R.O.R. is partially supported by CNPq and F. A. R. N. was supported by a M.Sc. scholarship from CAPES. |
warning/0001/math0001189.html | ar5iv | text | # 1 Introduction
## 1 Introduction
We investigate the system
$$\begin{array}{cc}& \psi _1=(|\psi _1|^2+|\psi _2|^2)\psi _2\hfill \\ & \overline{}\psi _2=(|\psi _1|^2+|\psi _2|^2)\psi _1\hfill \end{array}$$
(1.1)
which has been derived in and governs constant mean curvature (CMC) surfaces in the conformal parametrization $`z,\overline{z}`$ $`(=_z,\overline{}=_{\overline{z}}`$). This system was subsequently discussed in . In this paper, we demonstrate that system (1.1) can be decoupled into a direct sum of the elliptic Sh-Gordon and Laplace equations. Firstly, we change from $`\psi _1`$, $`\psi _2`$ to the new dependent variables $`Q,R`$
$$Q=2(\psi _2\overline{\psi }_1\overline{\psi }_1\psi _2),R=2(|\psi _1|^2+|\psi _2|^2)^2/|Q|.$$
Secondly, we introduce new independent variables $`\eta `$, $`\overline{\eta }`$ according to
$$d\eta =\sqrt{Q}dz,d\overline{\eta }=\sqrt{\overline{Q}}d\overline{z}$$
(these formulae are correct since $`Q`$ is holomorphic). In the new variables $`Q,R,\eta ,\overline{\eta }`$, system (1.1) assumes the decoupled form
$$(\mathrm{ln}R)_{\eta \overline{\eta }}=\frac{1}{R}R,$$
$$Q_{\overline{\eta }}=\overline{Q}_\eta =0,$$
which is a direct sum of the elliptic sh-Gordon and Laplace equations. This transformation is an immediate corollary of the known properties of CMC surfaces. Connection of system (1.1) with the sigma-model equations
$$\overline{}\rho \frac{2\overline{\rho }}{1+|\rho |^2}\rho \overline{}\rho =0,$$
($`\rho =i\overline{\psi }_1/\psi _2`$) is also discussed. In terms of $`\rho `$, our transformation adopts the form:
$$Q=\frac{2\rho \overline{\rho }}{(1+|\rho |^2)^2},R=\left|\frac{\overline{\rho }}{\rho }\right|.$$
## 2 Generalized Weierstrass representation of surfaces in $`^3`$
Following the results of , with any solution $`\psi _1`$, $`\psi _2`$ of the Dirac equations
$$\psi _1=p\psi _2,\overline{}\psi _2=p\psi _1,$$
(2.1)
($`p(z,\overline{z})`$ is a real potential), we associate a surface $`M^2E^3`$ with the radius-vector $`๐ซ(z,\overline{z})`$ defined by the formulae
$$๐ซ=(i(\psi _2^2+\overline{\psi }_1^2),\overline{\psi }_1^2\psi _2^2,2\psi _2\overline{\psi }_1),$$
$$\overline{}๐ซ=(i(\overline{\psi }_2^2+\psi _1^2),\psi _1^2\overline{\psi }_2^2,2\psi _1\overline{\psi }_2).$$
The latter are compatible by virtue of (2.1). The unit normal $`๐ง`$ of the surface $`M^2`$ can be calculated according to the formula $`๐ง=\frac{1}{i}๐ซ\times \overline{}๐ซ/|๐ซ\times \overline{}๐ซ|`$ and is represented as follows:
$$๐ง=\frac{1}{q}(i(\overline{\psi }_1\overline{\psi }_2\psi _1\psi _2),\overline{\psi }_1\overline{\psi }_2+\psi _1\psi _2,\psi _1\overline{\psi }_1\psi _2\overline{\psi }_2),$$
(2.2)
$$q=|\psi _1|^2+|\psi _2|^2.$$
One can verify directly that the scalar products in $`E^3`$ are $`(๐ง,๐ง)=1`$, $`(๐ง,๐ซ)=(๐ง,\overline{}๐ซ)=0`$. Equations of motion of the complex frame $`๐ซ`$, $`\overline{}๐ซ`$, $`๐ง`$ are of the form
$$\begin{array}{cc}& \left(\begin{array}{c}๐ซ\\ \overline{}๐ซ\\ ๐ง\end{array}\right)=\left(\begin{array}{ccc}2\frac{q}{q}& 0& Q\\ 0& 0& 2Hq^2\\ H& \frac{Q}{2q^2}& 0\end{array}\right)\left(\begin{array}{c}๐ซ\\ \overline{}๐ซ\\ ๐ง\end{array}\right)\hfill \\ & \overline{}\left(\begin{array}{c}๐ซ\\ \overline{}๐ซ\\ ๐ง\end{array}\right)=\left(\begin{array}{ccc}0& 0& 2Hq^2\\ 0& 2\frac{\overline{}q}{q}& \overline{Q}\\ \frac{\overline{Q}}{2q^2}& H& 0\end{array}\right)\left(\begin{array}{c}๐ซ\\ \overline{}๐ซ\\ ๐ง\end{array}\right)\hfill \end{array}$$
(2.3)
where $`Q=2(\psi _2\overline{\psi }_1\overline{\psi }_1\psi _2)`$ and $`H=p/q`$ is the mean curvature. Formulae (2.3) are compatible with the scalar products
$$(๐ง,๐ง)=1,(๐ซ,\overline{}๐ซ)=2q^2$$
(all other scalar products being equal to zero). Using (2.3), one can derive the following useful equation for the unit normal $`๐ง`$:
$$\overline{}๐ง+(๐ง,\overline{}๐ง)๐ง+\overline{}H๐ซ+H\overline{}๐ซ=0.$$
(2.4)
The first fundamental form $`I=(d๐ซ,d๐ซ)`$ and the second fundamental form $`II=(d^2๐ซ,๐ง)`$ of the surface $`M^2`$ are given by
$$\begin{array}{cc}& I=4q^2dzd\overline{z},\hfill \\ & II=Qdz^2+4Hq^2dzd\overline{z}+\overline{Q}d\overline{z}^2.\hfill \end{array}$$
(2.5)
The quantity $`Qdz^2`$ is called the Hopf differential. The real potential $`p`$ and the spinors $`\psi _1,\psi _2`$ satisfying (1.1) can be viewed as the โgeneralized Weierstrass dataโ of the surface $`M^2`$. The corresponding Gauss-Codazzi equations which are the compatibility conditions for (2.3), are of the form
$$\begin{array}{cc}& \overline{}(\mathrm{ln}q^2)=\frac{1}{2}\frac{Q\overline{Q}}{q^2}2H^2q^2,\hfill \\ & \overline{}Q=2q^2H,\hfill \\ & \overline{Q}=2q^2\overline{}H.\hfill \end{array}$$
(2.6)
In fact, equations (2.6) are a direct differential consequence of (2.1), as can be checked by a straightforward calculation. Gauss-Codazzi equations of surfaces in conformal parametrization $`z,\overline{z}`$ have been discussed in . We recall also that the Gaussian curvature $`K`$ of the surface $`M^2`$ can be calculated as follows:
$$K=\frac{1}{q^2}\overline{}(\mathrm{ln}q).$$
(2.7)
The unit normal $`๐ง=(n_1,n_2,n_3)`$, given by (2.2), maps a surface $`M^2`$ onto the unit sphere $`S^2`$. Combining this map with the stereographic projection, we obtain a map $`\rho `$ of the surface $`M^2`$ onto the complex plane, called the complex Gauss map. In our notation, $`\rho `$ assumes the form
$$\rho =\frac{n_1+in_2}{1n_3}=i\frac{\overline{\psi }_1}{\psi _2}.$$
(2.8)
According to the results of , Gauss map $`\rho `$ satisfies the nonlinear equation
$$(\overline{}\rho \frac{2\overline{\rho }}{1+|\rho |^2}\rho \overline{}\rho )H=H\overline{}\rho ,$$
(2.9)
which formally can be viewed as a differential consequence of (2.1). In terms of $`\rho `$ and $`H`$, the initial data $`\psi _1`$, $`\psi _2`$ and $`p`$ assume the form
$$\psi _1=\frac{\overline{\rho }}{\sqrt{H}}\frac{\sqrt{i\overline{}\rho }}{1+|\rho |^2},\psi _2=\frac{1}{\sqrt{H}}\frac{\sqrt{i\overline{\rho }}}{1+|\rho |^2},p=\frac{|\overline{\rho }|}{1+|\rho |^2}$$
while the expressions for $`q`$ and $`Q`$ take the form
$$q=\frac{1}{H}\frac{\overline{}\rho \overline{\rho }}{1+|\rho |^2},Q=\frac{2}{H}\frac{\rho \overline{\rho }}{(1+|\rho |^2)^2}.$$
###### Remark 1.
Linear problem (2.3) can be rewritten in terms of $`\psi _1`$, $`\psi _2`$ as follows. First of all, we point out that
$$q=\psi _1\overline{\psi }_1+\overline{\psi }_2\psi _2.$$
Combining this equation with the definition of $`Q`$:
$$\frac{1}{2}Q=\psi _2\overline{\psi }_1\overline{\psi }_1\psi _2,$$
and solving these two equations for $`\overline{\psi }_1`$, $`\psi _2`$, we can โcloseโ system (2.1) as follows:
$$\begin{array}{cc}& \left(\begin{array}{c}\psi _1\\ \psi _2\end{array}\right)=\left(\begin{array}{cc}0& qH\\ \frac{Q}{2q}& \frac{q}{q}\end{array}\right)\left(\begin{array}{c}\psi _1\\ \psi _2\end{array}\right),\hfill \\ & \overline{}\left(\begin{array}{c}\psi _1\\ \psi _2\end{array}\right)=\left(\begin{array}{cc}\frac{\overline{}q}{q}& \frac{\overline{Q}}{2q}\\ qH& 0\end{array}\right)\left(\begin{array}{c}\psi _1\\ \psi _2\end{array}\right).\hfill \end{array}$$
(2.10)
The compatibility conditions for system (2.10) coincide with (2.6). We point out that the $`2\times 2`$ matrix approach to surfaces in $`E^3`$ has been extensively developed in . From the point of view of the theory of integrable systems, linear system (2.3) can be regarded as the squared eigenfunction equations corresponding to (2.10) (indeed, $`๐ซ`$, $`\overline{}๐ซ`$ and $`๐ง`$ are quadratic expressions in $`\psi _1`$ and $`\psi _2`$).
## 3 CMC-1 surfaces
The class of CMC-1 surfaces is characterized by the constraint $`H=1`$ or, equivalently, $`p=q`$. Introducing this ansatz in (2.1), we arrive at the nonlinear system (1.1)
$$\psi _1=(|\psi _1|^2+|\psi _2|^2)\psi _2,$$
$$\overline{}\psi _2=(|\psi _1|^2+|\psi _2|^2)\psi _1,$$
which is the main subject of our study. According to the previous section, system (1.1) is equivalent to
$$\begin{array}{cc}& \overline{}\mathrm{ln}q^2=\frac{1}{2}\frac{Q\overline{Q}}{q^2}2q^2,\hfill \\ & \overline{}Q=\overline{Q}=0,\hfill \end{array}$$
(3.1)
where $`q=|\psi _1|^2+|\psi _2|^2`$, $`Q=2(\psi _2\overline{\psi }_1\overline{\psi }_1\psi _2)`$. Thus, for CMC surfaces the Hopf differential $`Qdz^2`$ is holomorphic. Applying to system (3.1) the reciprocal transformation
$$d\eta =\sqrt{Q}dz,d\overline{\eta }=\sqrt{\overline{Q}}d\overline{z}$$
(that is, changing from $`z,\overline{z}`$ to the new independent variables $`\eta ,\overline{\eta }`$ which are correctly defined in view of the holomorphicity of $`Q`$) and introducing
$$R=\frac{2q^2}{|Q|},$$
we transform system (3.1) into the decoupled form
$$\begin{array}{cc}& (\mathrm{ln}R)_{\eta \overline{\eta }}=\frac{1}{R}R,\hfill \\ & Q_{\overline{\eta }}=\overline{Q}_\eta =0.\hfill \end{array}$$
(3.2)
This result provides the rationale for the change of variables which we introduce in Section 1.
###### Remark 2.
System (1.1) is invariant under the $`SU(2)`$-symmetry
$$\zeta _1=\alpha \psi _1+\beta \overline{\psi }_2,\zeta _2=\beta \overline{\psi }_1+\alpha \psi _2,$$
(3.3)
where $`\alpha ,\beta `$ are complex constants subject to the constraint $`\alpha \overline{\alpha }+\beta \overline{\beta }=1`$. One can check directly, that the quantities $`Q`$ and $`R`$ are invariant under transformations (3.3), so that the surfaces, corresponding to $`(\psi _1,\psi _2)`$ and $`(\zeta _1,\zeta _2)`$, have coincident fundamental forms. Thus, they are identical up to a rigid motion in $`E^3`$. The passage from $`\psi _1,\psi _2`$ to $`Q,R`$ can thus be viewed as a passage to the differential invariants of the point symmetry group (3.3).
###### Remark 3.
For CMC-1 surfaces, system (2.10) allows the introduction of a spectral parameter
$$\begin{array}{cc}& \left(\begin{array}{c}\psi _1\\ \psi _2\end{array}\right)=\left(\begin{array}{cc}0& q\\ \lambda \frac{Q}{2q}& \frac{q}{q}\end{array}\right)\left(\begin{array}{c}\psi _1\\ \psi _2\end{array}\right)\hfill \\ & \overline{}\left(\begin{array}{c}\psi _1\\ \psi _2\end{array}\right)=\left(\begin{array}{cc}\frac{\overline{}q}{q}& \frac{1}{\lambda }\frac{\overline{Q}}{2q}\\ q& 0\end{array}\right)\left(\begin{array}{c}\psi _1\\ \psi _2\end{array}\right)\hfill \end{array}$$
(3.4)
where $`\lambda `$ is a unitary constant, $`|\lambda |=1`$. The gauge transformation
$$\stackrel{~}{\psi }_1=\psi _1,\stackrel{~}{\psi }_2=q^1\psi _2$$
reduces linear spectral problem (3.4) to the $`SL(2)`$ form
$$\begin{array}{cc}& \left(\begin{array}{c}\stackrel{~}{\psi }_1\\ \stackrel{~}{\psi }_2\end{array}\right)=\left(\begin{array}{cc}0& q^2\\ & \\ \lambda \frac{Q}{2q^2}& 0\end{array}\right)\left(\begin{array}{c}\stackrel{~}{\psi }_1\\ \stackrel{~}{\psi }_2\end{array}\right)\hfill \\ & \overline{}\left(\begin{array}{c}\stackrel{~}{\psi }_1\\ \stackrel{~}{\psi }_2\end{array}\right)=\left(\begin{array}{cc}\frac{\overline{}q}{q}& \frac{\overline{Q}}{2\lambda }\\ & \\ 1& \frac{\overline{}q}{q}\end{array}\right)\left(\begin{array}{c}\stackrel{~}{\psi }_1\\ \stackrel{~}{\psi }_2\end{array}\right).\hfill \end{array}$$
(3.5)
The compatibility conditions for both systems (3.4) and (3.5) coincide with (3.1). From the linear system (3.5) the radius-vector $`๐ซ`$ can be recovered via the so-called Sym formula: we refer to and for the further discussion of the Sym approach.
Linear system (3.5) can be readily rewritten in terms of $`\psi _1`$, $`\psi _2`$. Indeed, observing that (3.5) implies
$$\mathrm{ln}q=\mathrm{ln}(|\psi _1|^2+|\psi _2|^2),\overline{}\mathrm{ln}q=\overline{}\mathrm{ln}(|\psi _1|^2+|\psi _2|^2),$$
we can take $`q=c(|\psi _1|^2+|\psi _2|^2)`$, $`c`$, which, upon substitution in (3.5), produces system (1.1). Thus transformation from (3.1) to (1.1) consists of rewriting (3.5) in terms of the $`\psi `$. Representations in terms of $`\psi `$ are called eigenfunction equations, and are fundamental in soliton theory โ see eg . The Lax pair for system (1.1) is of the form
$$\left(\begin{array}{c}\varphi _1\\ \varphi _2\end{array}\right)=\frac{2}{\mu +1}\left(\begin{array}{cc}\overline{\psi }_1\psi _2+\frac{Q}{2q^2}\psi _1\overline{\psi }_2& \overline{\psi }_1^2\frac{Q}{2q^2}\overline{\psi }_2^2\\ & \\ \psi _2^2+\frac{Q}{2q^2}\psi _1^2& \overline{\psi }_1\psi _2\frac{Q}{2q^2}\psi _1\overline{\psi }_2\end{array}\right)\left(\begin{array}{c}\varphi _1\\ \varphi _2\end{array}\right)$$
$$\overline{}\left(\begin{array}{c}\varphi _1\\ \varphi _2\end{array}\right)=\frac{2}{\mu 1}\left(\begin{array}{cc}\psi _1\overline{\psi }_2+\frac{\overline{Q}}{2q^2}\overline{\psi }_1\psi _2& \overline{\psi }_2^2+\frac{\overline{Q}}{2q^2}\overline{\psi }_1^2\\ & \\ \psi _1^2\frac{\overline{Q}}{2q^2}\psi _2^2& \psi _1\overline{\psi }_2\frac{\overline{Q}}{2q^2}\overline{\psi }_1\psi _2\end{array}\right)\left(\begin{array}{c}\varphi _1\\ \varphi _2\end{array}\right).$$
It is interesting to note that the compatibility conditions for this linear system, which is of the first order in the derivatives of $`\psi `$, give us exactly system (1.1). The latter is also of the first order in $`\psi `$.
## 4 CMC surfaces and sigma model equations
For CMC surfaces, equations (2.9) imply the nonlinear sigma model
$$\overline{}\rho \frac{2\overline{\rho }}{1+|\rho |^2}\rho \overline{}\rho =0,$$
(4.1)
descriptions of the stationary two-dimensional $`SU(2)`$ magnet. The transformation of the system (4.1) into the decoupled form (3.2) now assumes the form
$$\begin{array}{cc}& Q=2\frac{\rho \overline{\rho }}{(1+|\rho |^2)^2},R=|\frac{\overline{\rho }}{\rho }|,\hfill \\ & d\eta =\sqrt{Q}dz,d\overline{\eta }=\sqrt{\overline{Q}}d\overline{z}.\hfill \end{array}$$
(4.2)
In terms of the unit normal vector $`๐ง`$, equation (2.4) adopts the form of the $`SO(3)`$ sigma model
$$\overline{}๐ง+(๐ง,\overline{}๐ง)๐ง=0,(๐ง,๐ง)=1.$$
(4.3)
Formula (2.8) establishes a link between sigma models (4.1) and (4.3). The topological charge
$$\frac{1}{4\pi }(๐ง,[๐ง\times \overline{}๐ง])๐zd\overline{z}$$
can be written as
$$\frac{1}{2\pi i}\overline{}\mathrm{ln}qdzd\overline{z}$$
which, in view of (2.7), is the topologically invariant integral curvature of the surface $`M^2`$.
Instanton solutions of system (4.3) are specified by the ansatz
$$๐ง=\pm i๐ง\times ๐ง,\overline{}๐ง=i๐ง\times \overline{}๐ง,$$
which, after a simple calculation, implies $`Q=0`$. Solutions of system (1.1) specified by a constraint $`Q=0`$, can be represented in the form
$$\psi _1=\frac{\rho \sqrt{\overline{}\overline{\rho }}}{1+|\rho |^2},\psi _2=\frac{\sqrt{\rho }}{1+|\rho |^2},p=\frac{|\rho |}{1+|\rho |^2},$$
(4.4)
where $`\rho (z)`$ is an arbitrary holomorphic function. In the case when the energy
$$E=\frac{\rho \overline{}\rho }{1+|\rho |^2}๐zd\overline{z}$$
is finite, the function $`\rho (z)`$ is rational in $`z`$ .
Geometrically, instanton solutions (4.4) parametrize the standardly embedded sphere $`S^2E^3`$. This can be readily seen from formulae (2.5), which, in case $`Q=0`$, imply the proportionality of fundamental forms I and II. This example shows that different Weierstrass data $`(\psi _1,\psi _2,p)`$ can correspond to different parametrizations of one and the same surface $`M^2E^3`$.
Introducing the two-component complex vector
$$๐=(\frac{\psi _1}{\sqrt{q}},\frac{\overline{\psi }_2}{\sqrt{q}}),q=|\psi _1|^2+|\psi _2|^2,$$
one can check that $`๐`$ satisfies the equations of the $`P^1`$ sigma model
$$\begin{array}{cc}& (๐,\overline{๐})=1,\hfill \\ & \overline{}๐=(\overline{๐},\overline{}๐)๐+(\overline{๐},๐)\overline{}๐k๐,\hfill \end{array}$$
(4.5)
where
$$k=2(\overline{๐},๐)(๐,\overline{}\overline{๐})+\frac{1}{2}(๐,\overline{}\overline{๐})+\frac{1}{2}(\overline{}๐,\overline{๐}).$$
Equations (4.5) are associated with the Lagrangian
$$L=\{(\overline{}๐,\overline{๐})+(๐,\overline{}\overline{๐})+2(\overline{๐},๐)(\overline{๐},\overline{}๐)2k[(๐,\overline{๐})1]\}๐z๐\overline{z},$$
where $`k`$ is the Lagrange multiplier.
### Acknowledgments
One of the authors (E.V.F.) would like to thank Centre de Recherches Mathรฉmatiques, Universitรฉ de Montrรฉal for generous hospitality and financial support during the period when this investigation was performed. This work was supported by research grant from NSERC of Canada and the Fonds FCAR du Gouvernment du Quรฉbec. We would like to thank Professor B. G. Konopelchenko for drawing our attention to this subject and for helpful discussions. We would like to thank the referee for useful comments. |
warning/0001/hep-th0001198.html | ar5iv | text | # 1 Introduction
## 1 Introduction
Open strings have enjoyed a rather varied amount of interest during the past three decades of string theory. In the early days open strings were used to describe mesons, with quarks attached to the endpoints. Indeed, the Chan-Paton labels still used today date back to as early as 1969 . This provided the first method for obtaining gauge groups in string theory. The possible gauge groups were classified much later , in a period when string theory in general had fallen into decline. In 1983 Alvarez-Gaumรฉ and Witten showed that open superstrings (type-I) were plagued by chiral anomalies for any gauge group, whereas closed superstrings (type-II) were automatically anomaly free. Although this looked like a fatal blow, there was a brief revival after Green and Schwarz found a novel mechanism to cancel the anomaly for the gauge group $`SO(32)`$. But within months the heterotic string was discovered . This theory could in addition to $`SO(32)`$ have the gauge group $`E_8\times E_8`$ , which at first sight seemed phenomenologically more attractive. During the subsequent ten years open strings were almost completely neglected in favor of heterotic strings, which went through a phase of rapid development. Apart from being phenomenologically disfavored, open strings looked ugly and complicated: their description requires world-sheets with boundaries and crosscaps, and to obtain finite one-loop diagrams one has to cancel tadpoles by hand.
All this changed drastically in 1995, for several reasons. First of all open strings were found to be part of the duality picture, and in particular in ten dimensions the strong coupling limit of the type-I string was conjectured to be the heterotic string . Secondly, the discovery of D-branes swept out by the endpoints of open strings made boundaries more respectable . Furthermore in some cases tadpole cancellation was found to be equivalent to charge cancellation between D-branes and orientifold planes , which sounds somewhat less ad hoc. Furthermore it was pointed out that the relation between the gauge and gravitational coupling is different in open string theories and in closed ones, which makes it possible to separate the unification scale and the string scale . This has opened new avenues in string phenomenology, involving open strings (see e.g. ). These developments make it clear that open strings must be considered seriously again.
## 2 Closed strings, CFT and modular invariance
During the period 1984-1994 there has been a lot of progress in the description of lower-dimensional closed strings in terms of conformal field theory (here and in the following โclosedโ is an abbreviation for โclosed and orientedโ). This had led to a very economical formalism based on a few simple, algebraic constraints, from which very general theorems can be derived. It would be nice to have a similar description of open strings. The algebraic constraints in the closed string case are (essentially) Lorentz invariance in the number of dimensions one considers, (super)conformal invariance and modular invariance. If one builds the internal (โcompactifiedโ) part of the theory out of some (super)CFT, the constraint of modular invariance is that the integer matrix $`Z`$ appearing in the one-loop partition function
$$P(\tau ,\overline{\tau })=\underset{ij}{}\chi _i(\overline{\tau })Z_{ij}\chi _j(\tau )$$
(1)
must commute with the generators $`S`$ and $`T`$ of the modular group of the torus. Here $`\chi _i(\tau )`$ are the Virasoro characters of the internal CFT (which may be combined with the space-time CFT in a more intricate way than suggested here, but this is easy to take into account). Modular invariance is a simple and powerful constraint, from which one can for example derive the Green-Schwarz factorization of chiral anomalies or prove the existence of fractional charges in a large class of heterotic string theories .
Unfortunately all this is limited to perturbative, closed string theories. There may be non-perturbative states in the spectrum of closed strings that are not controlled by the modular invariant partition function. Nevertheless modular invariance is, in my opinion, too nice a principle to simple give up. I would hope to find some sort of generalization, a principle that governs the presence or absence of states in string theory, M-theory or whatever string theory generalizes to. Such a principle would justify the term โtheory of everythingโ. Although this expression has fallen out of favor because it is usually maliciously misinterpreted by adversaries of string theory in particular and science in geneneral, it is justified in the following precise sense. In field-theoretic descriptions of our world it is always possible to add some new particles to a successful theory (and in particular to the standard model). There are few theoretical constraints, but if one makes the new particles sufficiently massive, unstable and weakly coupled, it is not hard to evade all experimental and cosmological constraints. This implies that one can never claim to have arrived at a complete description of all physics in our universe, since experimental and cosmological constraints are always limited to some subset of any relevant parameter space. This seems like an inevitable fact: one can never know more than one has measured. However, string theory provides a potential way out. Adding extra particles to a given string theory makes it inconsistent. The only thing that one may do is remove some states, and add others in their place. This is perhaps best known in the example of orbifold constructions, where one removes states from the spectrum that are not invariant under a certain symmetry, and replaces them by โtwisted statesโ. This is not limited to orbifold constructions, but is in fact a general property of modular invariant, perturbative closed strings. If it generalizes beyond perturbative closed strings it would in principle be possible to make a unique choice among the huge amount of string vacua, based on a finite number of experimental results. If further experiments find additional particles not predicted by this particular string theory, then they can only be accomodated at the expense of some particles that weregeneral found before. In other words, any further experiments would rule out string theory as a whole, or in still other words, we would have a very strong prediction for the existence or non-existence of any other particle, no matter how massive or weakly coupled, in our universe. This would certainly deserve the name โtheory of everythingโ.
The argument given above is summarized in the above two pictures, with on the left the string theory way of going from theory A to a different theory B, and on the right the field theory way.
At present these pictures are only a caricature of reality. There is little hope of finding the right string โvacuumโ without additional information, even if the first picture is the right one. But in addition non-perturbative effects, and in particular open and unoriented strings (which are non-perturbative from the point of view of closed strings) pose a serious challenge to this picture.
Open and unoriented strings are usually constructed by starting from a consistent closed, oriented theory. To the one-loop closed string amplitude, the torus, one adds an unoriented closed string diagram, the Klein bottle. This acts as a projection, removing certain states from the spectrum. Furthermore one adds open string diagrams, which at the one-loop level are the annulus and the Moebius strip. These diagrams come with a free parameter, the Chan-Paton multiplicity, for each boundary. Until this point this looks reminiscent of the orbifold procedure, with the Klein bottle projection playing the rรดle of the removal of non-invariant states, and with open strings playing the rรดle of twisted sectors. However, the presence of โtwistedโ sectors is in this case not governed by modular invariance, but by a different principle, the cancellation of massless tadpoles that lead to infinities. Unlike modular invariance this is a target space criterion, and hence one loses the nice feature of closed oriented theories that a consistent world sheet theory is sufficient to get a consistent target space theory. Worse yet, there exist unoriented string theories for which the tadpole cancellation conditions require all Chan-Paton labels to vanish . This means that there are no open string states at all, even though the Klein bottle still acts as a projection. Hence in this case the states of the closed, unoriented theory are a subset of those of the closed oriented theory, as in the second picture above. Note that this can never happen in modular invariant partition functions. It is easy to show that if $`Z_{ij}`$ yields a modular invariant, then any matrix $`0\widehat{Z}_{ij}Z_{ij}`$ can only be modular invariant if $`\widehat{Z}=Z`$ (this follows from $`Z_{00}=\widehat{Z}_{00}=1`$ and $`S_{0i}>0`$).
This implies that the first picture does not hold for perturbative states. It might still be saved by non-perturbative states, if the unoriented theory has non-perturbative states not present in the oriented theory. On the other hand it is possible that the first picture has to given up, and that one has to allow for a discrete set of consistent truncations of certain string theories. This does not necessarily invalidate the discussion given above. Most people would probably agree with the statement that in string theory and its generalizations the possibilities for adding or removing states are severely limited by consistency requirements. It would be nice to make that more precise.
## 3 Simple currents and fixed points
In search of a principle that governs the presence of states in a string theory, I now turn to something much more down-to-earth, namely a tool that plays an essential rรดle in dealing with modular invariance in the closed, oriented case: simple currents . Simple currents are primary fields that upon fusion with any other field yield just one field . They can be used to build a non-diagonal partition function. If one has a closed set of integral spin simple currents, these currents can extend the chiral algebra, and one obtains a partition function
$$\underset{Q(i)=0}{}N_i\left|\underset{j\mathrm{Orbit}(\mathrm{i})}{}\chi _j\right|^2$$
(2)
Here $`Q`$ is a charge (or set of charges) defined for each current, and the orbit is the set of distinct fields generated by the set of currents acting on $`i`$. In general some currents may fix $`i`$, and then the action of the currents covers the orbit $`N_i`$ times. Fractional spin currents also generate modular invariants, but they correspond to automorphisms of the fusion algebra, which pair the left and right representations in an off-diagonal way.
Simple currents are used for a variety of purposes in the construction of closed string theories, such as
* Field identification in coset CFTโs
* World-sheet supersymmetry projections
* Space-time supersymmetry projections
* D-type invariants
* Inverse orbifolds (under conditions discussed in )
As we will see, they play a useful rรดle in open string constructions as well.
Another concept that comes back in open string theories is the resolution of fixed points . The presence of the multiplicities $`N_i`$ in partition functions implies usually (but not always ), that the corresponding terms are reducible representations of the extended chiral algebra. To describe the modular properties of the characters of the extended algebra (the orbit sums) one needs a set of matrices that act on the resolved fixed points. There is such a matrix $`S_{ij}^J`$ for any current that has fixed points, and it is defined only on the fixed points $`i`$ and $`j`$ of $`J`$. In terms of these matrices, and the original matrix $`SS^0`$, the matrix $`S`$ of the extended theory takes the form
$$S_{(i,\mu )(j,\nu )}=\frac{|๐ข|}{\sqrt{|๐ฎ_i||๐ฐ_i||๐ฎ_j||๐ฐ_j|}}\underset{J}{}\mathrm{\Psi }_\mu ^JS_{ij}^J(\mathrm{\Psi }_\nu ^J)^{}$$
(3)
Here $`\mu ,\nu `$ label the components into which the orbits of $`i`$ and $`j`$ are resolved, $`๐ข`$ is the simple current group that extends the chiral algebra, and $`๐ฎ_i`$ (the stabilizer) is the subgroup that fixes $`i`$. The group $`๐ฐ_i`$ is a subgroup of $`๐ฎ_a`$ called the untwisted stabilizer; for the precise definition see . The factors $`\mathrm{\Psi }`$ are discrete group characters of $`๐ฐ_i`$, and $`i`$ is resolved into $`|๐ฐ_i|`$ components.
In general one can derive a list of properties that the matrices $`S^J`$ should satisfy in order for $`S_{(i,\mu )(j,\nu )}`$ to be a proper modular transformation matrix. In the case of WZW-models one can find a natural set of matrices that satisfy all those properties, and that are therefore obvious candidates for $`S^J`$ . This works as follows . To each WZW-model belongs a Dynkin diagram, which is an extended Dynkin diagram of an ordinary Lie algebra. Simple currents are related to symmetries of this extended Dynkin diagram that move the extended root (with one exception for $`E_8`$ level 2 ). Given a Dynkin diagram symmetry one can define a folded diagram in a fairly obvious way (for details see ), and to that diagram one can associate a Cartan matrix. This defines a new algebra, which we call the orbit Lie algebra. There is such an algebra for any simple current, and the modular transformation matrices of the characters of this orbit Lie algebra are equal to the matrices $`S^J`$, up to a calculable phase. The orbit Lie algebras of simple currents are affine Lie algebras, so that their modular transformation matrices are calculable using the Kac-Peterson formula . In one case the orbit Lie algebra is a twisted affine Lie algebra, but luckily this is precisely the only twisted affine Lie algebra whose characters have good modular properties.
The generalization of the foregoing to arbitrary CFTโs is not completely understood yet. The concept of an orbit Lie algebra seems restricted to WZW-models. But in any case most known rational CFTโs are related to WZW-models by the coset construction. Since field identification can be described in terms of simple currents (with a few exceptions), formula (3) applies to those cases. One can apply formula (3) to a combination of field identification and any chiral algebra extension of the coset theory, and read off the matrices $`S^J`$ of the coset theory. In a formula for these matrices was derived.
## 4 Open string CFT
Here a very brief introduction to open string conformal field theory is given. Only those aspects needed in the rest of the paper are mentioned.
Open string conformal field theory is defined on surfaces with handles, boundaries and crosscaps. Any such surface has a double cover which only has handles, and on which one defines a closed, oriented conformal field theory. This CFT is the starting point for constructions of open (and unoriented) strings, which are referred to as โopen descendantsโ of the closed string theories . The presence of boundaries and crosscaps is described by boundary and crosscap โstatesโ, which are not really states themselves, but in fact non-normalizable linear combinations of states in the closed string Hilbert space.
The closed string CFT has a chiral algebra which includes the Virasoro algebra. The boundary may preserve all or only part of the closed string chiral algebra, but must at least preserve the Virasoro algebra. Here we will assume that the entire chiral algebra remains unbroken (โtrivial gluingโ). The condition that a symmetry is not broken by a boundary or a crosscap is
$$(J_n+(1)^{h_J}\stackrel{~}{J}_n)|B=0;(J_n+(1)^{h_J+n}\stackrel{~}{J}_n)|C=0$$
(4)
where $`J_n`$ is a mode of a chiral current, $`\stackrel{~}{J}_n`$ a mode of an anti-chiral current, $`|B`$ a boundary state and $`|C`$ a crosscap state. A basis for the solutions to these conditions is formed by the Ishibashi states
$$|B_i=\underset{I}{}|I_iU_B|I_{i^c};|C_i=\underset{I}{}|I_iU_C|I_{i^c}.$$
(5)
Here the $`i`$ labels a representation of the chiral algebra and $`i^c`$ its charge conjugate. The sum is over all states in the representation, and $`U_B`$ and $`U_C`$ are operators satisfying
$$\stackrel{~}{J}_nU_B=(1)^{h_J}U_B\stackrel{~}{J}_n;\stackrel{~}{J}_nU_C=(1)^{h_J+n}U_C\stackrel{~}{J}_n$$
(6)
Any boundary state must be a linear combination of these Ishibashi states, i.e.
$$|B_a=\underset{i}{}B_{ai}|B_i;|C=\underset{i}{}\mathrm{\Gamma }_i|C_i$$
(7)
It turns out that in general one can allow for several boundary states, labelled by a boundary label $`a`$, but for only one crosscap state.
A choice of a set of boundary labels $`a`$, and a set of coefficients $`B_{ai}`$ and $`\mathrm{\Gamma }_i`$ form part of the data that define an open string CFT. Although more is required to specify all correlation functions on arbitrary surfaces, this information is sufficient to compute the one-loop diagrams without external lines that contribute to the open and closed string partition functions. Hence we can at least compute the spectrum of the theory. The diagrams are computed in the transverse channel, in which closed strings propagate between two boundaries, a boundary and a crosscap, or two crosscaps.
Transverse Annulus: $`N_aN_bB_a^c|e^{i\tau H}|B_b`$ (8)
Transverse Moebius strip: $`N_a\left[B_a^c|e^{i\tau H}|C+C^c|e^{i\tau H}|B_a\right]`$ (9)
Transverse Klein bottle: $`N_aC^c|e^{i\tau H}|C`$ (10)
Here $`H`$ is the closed string Hamiltonian: $`H=2\pi (L_0+\stackrel{~}{L}_0c/12)`$, and $`\tau `$ is a real number representing the length of the cylinder. The subscript โ$`c`$โ indicates that a CPT conjugate state is to be used. The integers $`N_a`$ are the Chan-Paton multiplicities. One can express these amplitudes in terms of characters of the representation $`i`$. By means of a transformation of the parameter $`\tau `$ one can then compute the corresponding amplitudes in the direct channel (the open and closed string loop channels). In the case of the Klein bottle and the annulus this transformation acts on the characters as the modular transformation matrix $`S`$, whereas in the case of the Moebius strip one uses the matrix $`P=\sqrt{T}ST^2S\sqrt{T}`$, with $`\sqrt{T}`$ defined as $`\mathrm{exp}i\pi (L_0c/24)`$. Then one arrives at the following expressions
Direct Annulus: $`{\displaystyle \frac{1}{2}}N_aN_bA_{ab}^i\chi _i({\displaystyle \frac{1}{2}}\tau )`$ (11)
Direct Moebius strip: $`{\displaystyle \frac{1}{2}}N_aM_a^i\widehat{\chi }_i({\displaystyle \frac{1}{2}}+{\displaystyle \frac{1}{2}}\tau )`$ (12)
Direct Klein bottle: $`{\displaystyle \frac{1}{2}}K^i\chi _i(2\tau )`$ (13)
Here $`\widehat{\chi }_iT^{\frac{1}{2}}\chi _i`$, and the parameter $`\tau `$ is purely imaginary. The coefficients are
$$A_{ab}^i=\underset{m}{}S_m^iB_{am}B_{am};M_a^i=\underset{m}{}P_m^iB_{am}\mathrm{\Gamma }_m;K^i=\underset{m}{}S_m^i\mathrm{\Gamma }_m\mathrm{\Gamma }_m;$$
(14)
### 4.1 Constraints
These coefficients are subject to several constraints. Since the direct channel contributions yield the open and closed string partition functions, the coefficients are subject to the requirement that all multiplicities should be positive integers (if one applies the formalism to fermionic strings one would like to see negative integers for space-time fermions, but those signs come out automatically if one takes into account ghosts properly). This yields two important conditions:
Closed sector: $`|K^i|=Z_{ii}`$ (15)
Open sector: $`M_a^i=A_{aa}^i\mathrm{mod}2,|M_a^i|A_{aa}^i`$ (16)
In writing down the first condition we assume that $`Z_{ii}1`$. One can write down modular invariant partition functions for which that is not the case, but this always means that some fields must be resolved into irreducible components first. Then the combination $`\frac{1}{2}(Z_{ii}+K_i)`$ is either a symmetric or anti-symmetric projection, or vanishes completely. In the second condition it is assumed that $`A_{aa}^i`$ is non-negative. Then $`\frac{1}{2}(N_aN_aA_{aa}^i+N_aM_a^i)`$ can be interpreted as $`\frac{1}{2}(A_{aa}^i+M_a^i)`$ symmetric tensors of the Chan-Paton group, plus $`\frac{1}{2}(A_{aa}^iM_a^i)`$ anti-symmetric tensors.
There are other constraints that are easy to check. First of all the โcompleteness conditionsโ
$`A_{ia}^bA_{jb}^c`$ $`=N_{ij}^kA_{ka}^c`$ (17)
$`A_{iab}A_{cd}^i`$ $`=A_{iac}A_{bd}^i`$ (18)
Here the boundary indices are raised and lowered with the โboundary metricโ $`A_{ab}^0=_nS_{0n}B_{na}B_{nb}`$, which must be order-2 permutation. In particular the matrix $`A_{ab}^0`$ must have entries $`0`$ or $`1`$, and must be numerically equal to its own inverse $`A^{0ab}`$. The three allowed Chan-Paton gauge groups correspond precisely to the three allowed combinations of $`A`$ and $`M`$: $`A_{aa}^0=M_a^0=1`$ gives $`Sp(N_a)`$ (if $`N_a`$ is even), $`A_{aa}^0=M_a^0=1`$ gives $`SO(N_a)`$, $`A_{ab}^0=A_{ba}^0=1,M_a^0=M_b^0=0,ab`$ gives $`U(N_a)`$ (if $`N_a=N_b`$).
It is convenient to define the matrices
$$R_{ia}=B_{ia}\sqrt{S_{i0}}$$
(19)
A sufficient condition for the completeness conditions as well as the properties of $`A_{ab}^0`$ is then
$$R_{ia}R_{ib}^{}=\delta _{ab}$$
$$R_{ib}R_{jb}^{}=\delta _{ij}$$
This may not be the most general solution, but the only one considered here. Given the matrices $`R`$ it is natural to define also
$$U_i=\mathrm{\Gamma }_i\sqrt{S_{i0}}.$$
(20)
Another easy constraint is the โKlein bottle constraintโ
$$K_iK_jK_k0\text{ if }N_{ijk}0$$
(21)
This ensures that the Klein bottle defines a consistent truncation on the spectrum. If this were not satisfied two states that are in the projected spectrum can have couplings to a third state that is not in the projected spectrum, so that the latter can appear in the intermediate channels of tree-diagrams. A slightly weaker form of this constraint was conjectured in .
There are other constraints (see ), but they involve additional quantities (such as OPE coefficients and duality matrices), that are not readily available, except in a few special cases. We will see, however, that the constraints described above are already very restrictive.
## 5 Open strings and simple currents
In this section various simple current modifications of the basic open descendant construction are discussed. This basic construction is often referred to as the โCardy caseโ, and consists of a natural ansatz for the boundaries, supplemented with an ansatz for the crosscap. We will discuss this case first, and show that the crosscap ansatz follows directly from the boundary ansatz and the positivity requirements.
### 5.1 Uniqueness of crosscaps in the Cardy case
Cardy conjectured a general ansatz for the annulus
$$B_{ai}=\frac{S_{ai}}{\sqrt{S_{0i}}}$$
This ansatz satisfies the completeness condition, and yields an annulus coefficient equal to the Verlinde fusion coefficients for $`S`$. It follows that all fields propagate in the transverse channel, and hence that all fields $`\varphi _{i,i^c}`$ must appear in the bulk theory. Therefore this ansatz requires the torus partition function to be defined in terms of $`Z=C`$, the charge conjugation matrix.
To check any of the other conditions we need to know the crosscap coefficients $`\mathrm{\Gamma }_i`$. They were first determined in an $`SU(2)`$ model by Sagnotti et. al. and this result was used as a conjecture for all other cases. Additional support for this conjecture was given in , where it was shown that this conjecture satisfies all positivity and integrality constraints; see also for related results and for a discussion of the Klein bottle constraint.
The results of the latter paper can in fact be turned around: we may impose positivity and integrality and derive the crosscap coefficients. To do this we make use of the fact that the set of boundaries in the Cardy case is in one-to-one correspondence with the bulk labels, and in particular there is a boundary โ0โ. Then $`A_{00}^i=N_{00}^i=\delta _{i0}`$, and hence $`M_0^i=\pm 1`$. Hence
$$\underset{m}{}P_m^iB_{0m}\mathrm{\Gamma }_m=\pm \delta _{i0}$$
(22)
from which we read off immediately
$$\mathrm{\Gamma }_m=\pm \frac{P_{0m}}{\sqrt{S_{0m}}},$$
(23)
where the sign is undetermined, but does not depend on $`m`$ (it would ultimately determined by the tadpole cancellation condition). We conclude that the crosscap coefficients are in fact uniquely determined by the boundary coefficients (up to an overall sign, which is not fixed by any CFT constraint). Recently the crosscap coefficient has also been determined using 3-dimensional TFT , but the argument given here has the advantage of being considerably simpler. It does not necessarily generalize to other annuli, because the boundary โ0โ need not exist. But it will generalize to the cases considered below.
### 5.2 Simple current modifications
Various simple current related modifications of the Cardy ansatz have been studied. A possibly incomplete list is
* Extensions of the closed chiral algebra
* Non-trivial Klein bottle projections
* Simple current automorphisms
* Broken bulk symmetries
The first item is dealt with entirely within the closed string theory and requires no further discussion. Examples of the second kind have been around for a while , and were studied in general in where also the consistency conditions were shown to hold in general. In the third class the bulk theory is defined by means of a simple current automorphism of the fusion rules. Such examples were first studied in . In automorphisms of spin-$`\frac{1}{2}`$ simple currents were studied in general. These authors gave an interpretation of the boundary label โ$`a`$โ in terms of the label of representations of a โclassifying algebraโ, which in the C-diagonal case is just the Verlinde algebra. A remarkable feature of this case is the appearance of the fixed point resolution matrix of the spin-$`\frac{1}{2}`$ current in the formulas for the boundary coefficient. This matrix does not play a direct rรดle in the closed string theory. The fourth case is studied for example in and . The latter papers reveal an even more interesting appearance of fixed point resolution matrices. I will not discuss any of these results in detail here, but in the rest of this section I will show how in the second and third case one may also derive the crosscap coefficients directly from the positivity constraints.
### 5.3 A formula for $`P`$
To derive these results I will need a formula relating matrix elements of the matrix $`P`$ on simple current orbits. It is analogous to the well-known formula for $`S`$
$$S_{a,J^{\mathrm{}}b}=e^{2\pi i\mathrm{}Q_J(a)}S_{ab}$$
(24)
The corresponding relation for $`P`$ is more complicated due to the factors $`\sqrt{T}`$ in the definition of $`P`$, and only works if the indices are shifted by even powers of the current
$$P_{a,J^2\mathrm{}b}=\rho (\mathrm{})e^{2\pi i\mathrm{\Delta }(2l,b)}e^{2\pi iQ_J(a)}P_{ab}$$
(25)
where
$$\mathrm{\Delta }(\mathrm{},c)=h_{J^{\mathrm{}}c}h_J^{\mathrm{}}h_c+\mathrm{}Q(c)$$
(26)
with $`0Q<1`$, and
$$\rho (\mathrm{})=e^{\pi i(r\mathrm{}+M_2\mathrm{})}$$
(27)
where
$$M_{\mathrm{}}=h_J^{\mathrm{}}\frac{r\mathrm{}(N\mathrm{})}{2N},$$
(28)
where $`r`$ is the monodromy parameter of the current. The derivation is straightforward, provided one replaces the ill-defined quantity $`\sqrt{T}`$ systematically by the well-defined quantity $`\mathrm{exp}(i\pi (hc/24))`$. The first two factors in (25) are signs.
### 5.4 Non-trivial Klein bottle projection
The first simple current modification I will consider was called a โnon-trivial Klein bottle projectionโ in , because in some cases it produces sign changes in the coefficients $`K_i`$ with respect to the Cardy case. Here I will study it from a different starting point, namely the annulus.
Consider the following set of reflection coefficients, $`R_{ma}=S_{ma}\sqrt{\frac{S_{m0}}{S_{mJ}}}`$, which satisfy the completeness conditions. Then
$$A_{00}^i=\underset{m}{}\frac{S_m^iR_{ma}R_{mb}}{S_{mJ}}=N_{00}^{Ji}$$
Hence $`M_0^i=\pm \delta _0^{Ji}=\pm \delta _{J^c}^i`$. On the other hand
$$M_0^i=\underset{m}{}P_m^iU_m\sqrt{\frac{S_{m0}}{S_{mJ}}}$$
so that
$`U_m`$ $`={\displaystyle \underset{i}{}}P_{mi}M_0^i\sqrt{{\displaystyle \frac{S_{mJ}}{S_{m0}}}}`$ (29)
$`=\pm P_{mJ^c}\sqrt{{\displaystyle \frac{S_{mJ}}{S_{m0}}}}`$ (30)
Now we use the formula
$$P_{a,K^2c}=ฯต(K,c)e^{2\pi iQ_K(a)}P_{ac}$$
where $`ฯต(K,c)`$ is a sign. We choose $`K=J^c`$ and $`c=J`$. Then
$`U_m`$ $`=\pm P_{mJ^c}\sqrt{{\displaystyle \frac{S_{mJ}}{S_{m0}}}}`$ (32)
$`=\pm e^{2\pi iQ_{J^c}(m)}P_{mJ}\sqrt{{\displaystyle \frac{S_{mJ}}{S_{m0}}}}`$ (33)
$`=\pm P_{mJ}\sqrt{{\displaystyle \frac{S_{m0}}{S_{mJ}}}}`$ (34)
This is the formula used in , which turns out to be the only possibility, given the annulus coefficients. The proof that the other constraints are satisfied can be found in that paper.
### 5.5 Non-trivial simple current automorphism
Up to now the bulk modular invariant was the charge conjugation invariant, $`Z=C`$. A general modular invariant is characterized by a left and right extension of the chiral algebra, the modules of which are paired by an automorphism. For the construction of โopen descendantsโ the left and right extensions must be identical, and the automorphism symmetric. The extension can be dealt with at the closed string level, which leaves the possibility of non-trivial automorphisms. One may distinguish three basic types: the charge conjugation invariant, simple current invariants and anything else, which by definition is โexceptionalโ.
A solution is known for simple current automorphisms generated by $`๐_{\mathrm{odd}}`$ and $`๐_2`$ simple currents. Both cases are described by the following formula for $`R`$
$$R_{m,a_i}=\frac{1}{\sqrt{|G|}}\underset{JS_a}{}\underset{KG/S_a}{}\stackrel{ห}{S}_{m,Ka}^J\psi _i^J$$
Here $`G`$ is the full simple current group that produces the automorphism, $`S_aG`$ is the stabilizer of $`a`$, $`\stackrel{ห}{S}^J`$ is the (appropriate generalization of) the orbit Lie algebra $`S`$-matrix corresponding to $`J`$ and $`\psi _i^J`$ is a discrete group character of $`S_a`$. The boundary labels are taken to correspond to be $`G`$-orbits, with each orbit split into $`|S_a|`$ components. The label $`m`$ ranges over all fields with $`Z_{mm^c}=1`$. This formula is in any case correct for $`G=๐_2`$ (in which case it summarizes the four expressions given in ) and $`G=๐_{\mathrm{odd}}`$, and is a good candidate for a general formula, although this has not been investigated yet. From here on we will assume that $`G=๐_N`$, with $`N`$ odd or equal to 2.
In a consistent ansatz was presented for the crosscap coefficient. Here I will show how to derive it, demonstrating that this ansatz is in fact unique (given the reflection coefficients). Consider the โzero-boundaryโ $`a=0`$. It is not hard to show that
$$A_{00}^i=\underset{J}{}\delta _J^i,$$
so that
$$M_0^i=\eta (i)\underset{J}{}\delta _J^i,$$
where $`\eta (i)`$ is a sign.
Computing $`M_0^i`$ directly gives
$$M_0^i=\underset{m,Q(m)=0}{}P_m^i\sqrt{|G|}U_m.$$
(35)
The restriction to zero charge is due to the fact that $`R_{m0}`$ vanishes if $`m`$ has non-zero charge. Using the inverse of $`P`$ we get, for $`Q(m)=0`$
$$U_m=\frac{1}{\sqrt{|G|}}\underset{JG}{}\eta (J)P_{Jm},$$
(36)
Note that we get no information about the coefficients for non-zero charge. On the other hand, the fact that the terms with $`Q(m)0`$ do not contribute to the sum (35) implies that
$$\underset{JG}{}\eta (J)P_{Jm}=0\text{ for }Q(m)0$$
This puts strong constraints on the possible choices for $`\eta (J)`$.
We can write (36) as a sum over even and odd elements:
$$V_m=\frac{1}{\sqrt{N}}\left[\underset{\mathrm{}=0}{}\eta (2\mathrm{})P_{J^2\mathrm{},m}+\underset{\mathrm{}=0}{}\eta (2\mathrm{}+1)P_{J^{2\mathrm{}+1},m}\right]$$
Here $`V_m=U_m`$ if $`Q(m)=0`$ and $`V_m=0`$ (with $`U_m`$ undetermined) if $`Q(m)0`$. Using the relation (25) we can express the first terms in terms of $`P_{0m}`$ and the second ones in terms of $`P_{Jm}`$. If $`N`$ is odd we can furthermore express $`P_{Jm}=P_{J^{N+1}m}`$ in terms of $`P_{0m}`$. Then the final result can be expressed as
$$V_m=\frac{1}{\sqrt{N}}P_{0m}\underset{\mathrm{}=0}{\overset{N1}{}}\sigma (\mathrm{})e^{2\pi ilQ(m)},$$
where $`\sigma (\mathrm{})`$ is a sign. This can only vanish for all $`Q(m)0`$ if $`\sigma (\mathrm{})=\pm 1`$, independent of $`\mathrm{}`$. Then also $`U_m=V_m`$ for $`Q(m)=0`$ is determined,
$$U_m=\pm \sqrt{N}P_{0m}$$
(37)
For $`N=2`$ we find
$$V_m=\frac{1}{\sqrt{2}}\left[\eta (0)P_{0m}+\eta (J)P_{Jm}\right].$$
Since $`P_{0m}`$ and $`P_{Jm}`$ are unrelated we cannot simplify the result further. In addition $`P_{0m}=P_{Jm}=0`$ for $`Q(m)=\frac{1}{2}`$, so that we get no further constraints. It may seem that there are two solutions now (not counting the overall sign), but closer inspection of the positivity constraints of fixed point boundary labels reveals that for each CFT only one definite relative sign $`\eta (0)/\eta (J)`$ is allowed. The other sign is the correct one for a different choice of reflection coefficients (see for more details).
To determine the coefficients $`U_m`$ for charged $`m`$ we may use closed sector positivity for bulk label $`i=0`$. This leads to the requirement $`_mU_m^2=1`$. For $`N`$ odd there are no transverse channel fields with non-zero charge, and it is easy to show that (37) satisfies this condition. In the case $`N=2`$ there is an additional problem, namely that we do not know $`U_m`$ for fixed points. Allowing for an unknown contribution on the fixed points we get
$$K^i=\underset{m,Q(m)=0}{}\frac{S_m^iU_mU_m}{S_{0m}}+\underset{f,Jf=f}{}\frac{S_f^iU_fU_f}{S_{0f}}=K_1^i+K_2^i,$$
where $`U_m=V_m`$ given above, and $`U_f`$ is unknown. It can be shown that $`K_1`$ already satisfies the positivity constraints. Then the only allowed values for $`K_2`$ are $`K_2^i=k_iK_1^i`$, where $`k_i=2`$ or $`0`$. But it is easy to show that $`K_2^{Ji}=K_2^i`$ whereas $`K_1^{Ji}=K_1^i`$. Then we must have $`K_2^i=0`$ and hence $`U_f=0`$. So also in this case the crosscap is unique.
### 5.6 Other automorphisms
For automorphisms that are not simple current modifications of the $`C`$-invariant there is an amusing observation to be made. The coefficients $`U_m`$ must vanish whenever $`m`$ is not paired with its charge conjugate. The coefficients $`K^i`$ must vanish if $`i`$ is not paired with itself. This leads to the sum rule
$$\underset{i,Z_{ii}=1}{}K^iS_{im}=0,\text{if }Z_{mm^c}=0$$
For example in the interesting case $`Z=\mathrm{๐}`$ this leads to the sum rule $`_iK^iS_{im}=0`$ for complex fields $`m`$. Empirically this rule is indeed satisfied, with all $`K^i`$ equal to 1. Although in some cases (e.g $`A_2`$ level 1) this sum rule is satisfied in a trivial way, there are many other examples (e.g $`A_2`$ level 3) where it is non-trivial, and implies relations between matrix elements of $`S`$ that are hard to derive in any other way. This illustrates that by studying open, unoriented CFT one may learn something interesting about the closed, oriented case.
Acknowledgements
I would like to thank Lennaert Huiszoon and Nuno Sousa for discussions and Christoph Schweigert for comments on the manuscript. Special thanks to the organizers for inviting me to give this talk, and for making this a successful conference despite what turned out to be very difficult circumstances. |
warning/0001/astro-ph0001541.html | ar5iv | text | # ALTAI: Computational code for the simulations of TeV air showers as observed with the ground-based imaging atmospheric ฤerenkov telescopes.
## 1 Introduction
Recent exciting detections and observations of TeV $`\gamma `$-ray emission from a number of galactic and extragalactic objects (Ong 1998, Catanese & Weekes 1998) have shown the high performance of currently operating imaging air ฤerenkov telescopes (IACTs). Several projects for future detectors have been proposed lately. The major forthcoming instruments, such as CANGAROO IV, H.E.S.S., MAGIC and VERITAS, will have significantly better sensitivity to $`\gamma `$-ray fluxes in the energy range from 50 GeV up to 50 TeV.
For a lack of a collimated test beam of TeV $`\gamma `$-ray photons, the ground-based ฤerenkov detectors heavily rely on the Monte Carlo simulations of the ฤerenkov light emission from air showers which are used to understand the performance of detector. Basic parameters of the instrument (detection area, angular and energy resolution, efficiency of cosmic ray rejection, etc) can be derived from the simulations. The crucial point is a precise measurement of the primary shower energy. For that one should include properly into the simulations all processes of ฤerenkov light emission in the air shower as well as photon propagation on the way from the emitting shower particle to the telescope camera. Measurements of $`\gamma `$-ray fluxes, energy spectra, upper limits strongly depend on the absolute energy calibration of a telescope.
Here we give a description of ALTAI <sup>1</sup><sup>1</sup>1ALTAI is the abbreviation for Atmospheric Light Telescope Array Image. Mountain Altai is the pristine wilderness in the south-west of Russia. computational code developed for detailed Monte Carlo simulations of the ฤerenkov light emission in TeV air showers. Among the other existing codes intended for such simulations, MOCCA (Hillas 1979), KASCADE (Kertzman & Sembroski 1994), CORSIKA (Heck et al. 1997), this code has a distinctive advantage โ itโs rather high speed of shower simulations due to a specific algorithm used for processing the multiple scattering of charged shower particles. This approach does not consume a lot of CPU time and allows to perform fast and accurate simulations. Together with the additional routine recently developed for the simulations of telescope response (Hemberger 1998) the ALTAI code was extensively used for production of a standard Monte Carlo database used in the HEGRA (High Energy Gamma Ray Astronomy) VHE $`\gamma `$-ray experiment (Konopelko et al. 1999a).
The ALTAI code consists of two major programs which simulate the development of the electromagnetic (EMCCS) (hereafter we put in brackets the name of the corresponding subroutine of the code) and proton-nuclei (STEPAD, MULTIP, XPI) cascade in Earth atmosphere. We discuss in detail the procedure of simulating the electromagnetic and hadron-nuclei cascade in Sections 2 and 3, respectively. Section 2 also describes the algorithms of ฤerenkov light emission by the shower charged particle. Section 4 deals with nucleus-nucleus interactions. In Section 5 we review the results of the ฤerenkov light simulations using ALTAI code compared with other relevant simulations and observational data.
## 2 Electromagnetic cascade
Here we overview the part of the code intended for the generation of an electromagnetic cascade in the atmosphere (EMCCS). Note that at present all cross-sections for electron and photon interactions are established and are well described in detail in the relevant energy range. An electromagnetic part of the code accounts for the following interaction processes: electron-positron pair production, Compton scattering for the primary photons and bremsstrahlung, ionization losses, Coulomb scattering for the primary electrons and positrons.
The cross-section of $`e^{},e^+`$ pair production by primary photon was calculated according to the Bethe-Heitler formula taken from Motz et al. (1964) (MPRH). The emission angle of $`e^{},e^+`$ components, defined with respect to the direction of initial photon, was sampled using the Bethe distribution (Motz et al. 1964). We used the Klein-Nishina formula (see e.g., Gaisser 1990) for the cross-section of Compton photon scattering (COMPH). The dependence of photon cross-sections on energy and the atomic number of a medium was derived from data taken from Storm & Israel (1973) and Hubbel (1969) (FMPM).
We simulated the bremsstrahlung interaction process (MRDH) according to the Bethe-Heitler formula (see Koch & Motz, 1959). The angle of emitted photon was sampled from the Schieff distribution (see Koch & Motz, 1959). The Bethe-Bloch formula, with the corrections for the density effect (Sternheimer 1952), was used for simulation of the mean ionization losses (MIONH) whereas for the differential cross-section of $`e^{},e^+`$ ionization collisions we used the Mรถller formula (Mรถller 1932). Note that in the electromagnetic shower positrons were treated similarly to electrons.
### 2.1 Multiple scattering
The relativistic electron suffers an enormous number of interactions along its path length in the matter. Apparently, a direct simulation of all interactions is not possible. That is why all numerical codes for the shower simulation use a specific technique for grouping electron interactions (see for review Berger 1963, Akkerman 1991, Gaisser 1990). In this approach one can simulate only the so-called catastrophic $`e^{},e^+`$ interactions. These interactions lead to emission of photon/electron of a relatively high energy, $`\mathrm{E}>\mathrm{Q}`$, which exceeds the intermediate energy threshold of $`\mathrm{Q}`$. The same approach was used for the simulation of the $`e^{},e^+`$ ionization process ($`\mathrm{Q}=\mathrm{Q}_\mathrm{I}`$) and bremsstrahlung ($`\mathrm{Q}=\mathrm{Q}_\mathrm{R}`$). The emitted electron must have a kinetic energy above a certain intermediate energy threshold, $`\mathrm{T}>\mathrm{Q}_\mathrm{I}`$.
A random path length between neighboring catastrophic collisions was divided into small segments of $`\mathrm{\Delta }\mathrm{l}5\mathrm{gr}/\mathrm{cm}^2`$. We sampled the phase coordinates of a charged particle at the end of the segment $`\mathrm{\Delta }\mathrm{l}`$ according to the cumulative effect of all low energy collisions along the segment (SGMEH, SGMQH). Thus we simulated at the end of each segment the loss of the $`e^{},e^+`$ kinetic energy, $`\mathrm{\Delta }\mathrm{T}=\mathrm{T}_1\mathrm{T}_2`$, the scattering angle, $`\mathrm{\Theta }=\mathrm{arccos}(\stackrel{}{\mathrm{\Omega }}_1\stackrel{}{\mathrm{\Omega }}_2)`$, the azimuth angle of scattered particle, $`\phi `$, the longitudinal $`\mathrm{Z}`$ and lateral $`\stackrel{}{\rho }`$ displacement of initial particle while passing over the segment (see Figure 1). Appropriate transport equations were used in order to derive the probability distributions for the phase coordinates at the end of a segment. We have obtained the analytical solutions for these distributions.
#### 2.1.1 Energy losses
The energy losses of a charged particle, $`\mathrm{\Delta }\mathrm{E}`$, in the segment, $`\mathrm{\Delta }\mathrm{l}`$, were sampled according to the Landau-Vavilov formula (Landau 1944, Vavilov 1957). Plyasheshnikov & Kolchuzkin (1975) have tabulated this formula for a specific conditions of shower simulations. The distribution of the energy losses at the end of a multiple-scattering segment was described as follows:
$`P(\mathrm{\Delta }T)d\mathrm{\Delta }T=f(\mu ,\lambda )d\lambda ,`$
$`f(\mu ,\lambda )={\displaystyle \frac{1}{2\pi i}}{\displaystyle _{ci\mathrm{}}^{c+i\mathrm{}}}\mathrm{exp}\left\{\lambda t\mu \varphi (t)\right\}๐t,`$
$`\varphi (t)=t{\displaystyle _0^t}{\displaystyle \frac{(1\tau exp(\tau ))}{\tau ^2}}๐\tau `$
$`\mu =\mathrm{\Sigma }_h(T_1)\mathrm{\Delta }l,`$
$`\lambda =(\mathrm{\Delta }T\beta _l(T_1)\mathrm{\Delta }l)/Q_I,`$
$`\mathrm{\Sigma }_h(T)={\displaystyle _{Q_I}^{T/2}}W_I(T,Q)๐Q,`$ (1)
where
$`\beta _l(T)=\beta _I+\beta _R`$
$`\beta _I={\displaystyle _0^{Q_I}}W_I(T,Q)Q๐Q`$
$`\beta _R={\displaystyle _0^{Q_R}}W_R(T,Q)Q๐Q`$ (2)
is the mean energy loss per unit path length due to the inelastic collisions of a charged particle by production of low energy secondaries. $`\mathrm{W}_\mathrm{I}(\mathrm{Q}),\mathrm{W}_\mathrm{R}(\mathrm{Q})`$ are the differential cross-sections for the ionization ($`\mathrm{I}`$) and bremsstrahlung ($`\mathrm{R}`$) interactions.
The standard inverse function technique was applied in order to simulate the energy loss $`\mathrm{\Delta }T`$ according to the two-dimensional distribution $`f(\mu ,\lambda )`$ (Eqn. (1)). For that one needs to compute the tables of the function $`\stackrel{~}{\lambda }=\stackrel{~}{\lambda }(\mu ,\alpha )`$, which is inverse to the integral distribution
$$F(\mu ,\lambda )=_{\mathrm{}}^\lambda f(\mu ,\stackrel{~}{\lambda })๐\stackrel{~}{\lambda }$$
(3)
These tables can be found, e.g., in Akimov et.al. (1981) where $`\stackrel{~}{\lambda }=\stackrel{~}{\lambda }(\mu ,\alpha )`$ was tabulated over a two-dimensional lattice with the steps of 0.025 and 0.01 over $`\alpha `$ and $`\mu `$, respectively. First, it is necessary to generate the random number $`\alpha `$ uniformly distributed in the interval (0,1) and calculate parameter $`\mu `$ according to Eqn. (1). After that one should interpolate $`\lambda =\lambda (\mu ,\alpha )`$ using the above mentioned tables and finally determine $`\mathrm{\Delta }T`$ as
$$\mathrm{\Delta }T=\beta _I\mathrm{\Delta }l+\lambda Q_I$$
(4)
where
$`\lambda =\{\begin{array}{c}\mu ^{1/2}\stackrel{~}{\lambda },\mu <0.1;\hfill \\ \mu (\stackrel{~}{\lambda }+\mathrm{ln}\mu ),\mu 0.1\hfill \end{array}`$ (7)
Formula (5) was derived using two asymptotics of initial distribution (1) for $`\mu =0`$ and $`\mu =\mathrm{}`$, which were sewn up at $`\mu =0.1`$. The subroutine ELSH makes calculations according to these algorithms.
#### 2.1.2 Angular deflection
We simulated the angle of multiple scattering using the Moliere theory (Moliere 1948, Bethe 1953). In addition we have improved the Moliere distribution by taking into account the energy losses at the multiple scattering segment. This distribution may be described as follows:
$$P(\mathrm{\Theta })\mathrm{\Theta }d\mathrm{\Theta }=\left[f_0(\stackrel{~}{\mathrm{\Theta }})+\frac{f_1(\stackrel{~}{\mathrm{\Theta }})}{B}+\frac{f_2(\stackrel{~}{\mathrm{\Theta }})}{B^2}\right]\stackrel{~}{\mathrm{\Theta }}d\stackrel{~}{\mathrm{\Theta }},$$
(8)
where
$`\stackrel{~}{\mathrm{\Theta }}^2={\displaystyle \frac{\mathrm{\Theta }^2}{\stackrel{~}{\chi }_c^2}}B,`$
$`\stackrel{~}{\chi }_c^2={\displaystyle _{T_o}^{T_1}}\chi _c^2(T)/\beta _l(T)๐T,`$
$`T_o=T_1\beta _l(T_1)\mathrm{\Delta }l.`$ (9)
Parameter $`\mathrm{B}`$ was determined by resolving the transcendental equation $`B\mathrm{ln}B=\mathrm{ln}(0.856(\stackrel{~}{\chi }_c/\chi _a)^2)`$. Functions $`f_n(n=0,1,2)`$ were calculated as
$$f_n(\stackrel{~}{\mathrm{\Theta }})=_0^{\mathrm{}}(\frac{u^2lnu}{4})^ne^{\frac{u^2}{4}}J_o(\stackrel{~}{\mathrm{\Theta }}u)u๐u.$$
(10)
Quantities $`\chi _c`$ and $`\chi _a`$ are closely related to the differential cross-section for Coulomb scattering (see formula (10) for definition of the Coulomb cross-section).
For a fixed segment length $`\mathrm{\Delta }\mathrm{l}`$ and kinetic energy of a particle $`\mathrm{T}`$ one can determine in a few iterations the parameter $`\mathrm{B}`$, which defines the shape of the distribution $`\mathrm{P}(\mathrm{\Theta })`$. Such calculations were done by use of subroutine ANGMH.
#### 2.1.3 Space displacement
At the multiple scattering segments we simulated the longitudinal displacement $`\mathrm{Z}`$ of a charged particle. For that we used the Yang-Spenser distribution (Yang 1951, Spencer & Coune 1962). This distribution was adapted for the air shower simulations by Plyasheshnikov & Kolchuzhkin (1975). This distribution may be represented by the following expression
$`P(Z)dZ=e^\nu f(\nu ,\xi )d\xi `$
$`f(\nu ,\xi )={\displaystyle \frac{1}{2\pi i}}{\displaystyle _{ci\mathrm{}}^{c+i\mathrm{}}}u^{1/2}cosch(u^{1/2})e^q๐u`$
$`q=\xi u\nu u^{1/2}ctg(u^{1/2}),`$ (11)
where
$`\nu ={\displaystyle \frac{1cos\mathrm{\Theta }}{\eta \mathrm{\Delta }l}},\xi ={\displaystyle \frac{\mathrm{\Delta }lZ}{\eta \mathrm{\Delta }l^2}},`$
$`\eta =2\pi {\displaystyle _1^{+1}}W_c(T_1,\theta )(1cos\theta )d\mathrm{cos}\theta `$
$`W_c(T,\theta )=2\chi _c^2/(\theta ^2+\chi _a^2)^2.`$ (12)
Here $`\mathrm{W}_\mathrm{c}(\mathrm{T},\theta )`$ is a cross-section of the Coulomb scattering which determines parameters $`\chi _\mathrm{a}`$ and $`\chi _c`$.
To simulate the longitudinal displacement $`Z`$ we use approach similar to that used in Section 2.1.1 for simulation of the energy loss $`\mathrm{\Delta }T`$. It is based on the interpolation of the two-dimensional function $`\xi =\xi (\nu ,\alpha )`$, which is the inverse function to the integral distribution
$$F(\nu ,\xi )=_0^\xi f(\nu ,\stackrel{~}{\xi })๐\stackrel{~}{\xi }$$
(13)
One can find these tables, e.g., in Akimov et.al (1981). This approach includes (i) sampling of the random number $`\alpha `$, (ii) calculation of $`\nu `$ on the basis of quantities $`T_1`$ and $`\mathrm{\Delta }l`$, (iii) determination of $`\xi `$ on the basis of the two-dimensional interpolation using the above mentioned tables and, finally, calculation of $`Z`$ according to the formula
$$Z=\mathrm{\Delta }l\xi \eta \mathrm{\Delta }l^2.$$
(14)
For the lateral displacement of a charged particles at the multiple scattering segment we defined the corresponding probability distribution using Fermi formula (see e.g., Kolchuzhkin & Plyasheshnikov 1975). This distribution was derived by solving the transport equations in the Focker-Planck approximation where the collision integral corresponding to the Coulomb scattering was replaced by a second order differential operator (see e.g., Kolchuzhkin & Uchaikin 1978). For the fixed angle of multiple-scattering $`\stackrel{}{\mathrm{\Theta }}=`$ $`(\mathrm{\Theta }_x,\mathrm{\Theta }_y)=`$ $`(\mathrm{\Theta }\mathrm{cos}\phi ,\mathrm{\Theta }\mathrm{sin}\phi )`$ this distribution was as follows
$`P(\stackrel{}{\psi })={\displaystyle \frac{6}{\pi \gamma }}\mathrm{exp}\left[{\displaystyle \frac{6}{\gamma }}(\stackrel{}{\psi }\stackrel{}{\mathrm{\Theta }}/2)^2\right],`$
$`\stackrel{}{\psi }=\stackrel{}{\rho }/\mathrm{\Delta }l,`$ (15)
where the parameter $`\gamma =\eta \mathrm{\Delta }l`$ defines the width of the distribution. As was shown by Kolchuzhkin & Plyasheshnikov (1975) more accurate numerical solution of the transport equations reduces by a factor of 1.5 the width of the final distribution. This difference may be corrected by introducing another definition for the parameter $`\gamma `$:
$$\gamma =\chi _c^2\stackrel{~}{B}.$$
(16)
where $`\stackrel{~}{\mathrm{B}}`$ is determined from $`\stackrel{~}{B}ln\stackrel{~}{B}=ln((\stackrel{~}{\chi }_c/\chi _a)^2/1.80)`$. The complete procedure for charged particle transport in multi-dimensional phase space of energy, angular and space coordinates was defined by a few parameters: two threshold energies $`\mathrm{Q}_\mathrm{I}`$, $`\mathrm{Q}_\mathrm{R}`$, and the length of a multiple-scattering segment $`\mathrm{\Delta }\mathrm{l}`$. The extensive test calculations using the ALTAI code revealed the optimum values of parameters which allow to perform rather fast shower simulations without introducing systematic errors. Thus we used the following parameters for the calculational procedure: $`\mathrm{Q}_\mathrm{I}0.5\mathrm{T}_\mathrm{o}`$, $`\mathrm{Q}_\mathrm{R}0.1\mathrm{T}_\mathrm{o}`$, where $`\mathrm{T}_\mathrm{o}20`$ MeV is the threshold energy for ฤerenkov light emission by cascade electrons in air.
Apparently the length of a multiple scattering segment, $`\mathrm{\Delta }\mathrm{l}`$, is one of the basic parameters of this method. By use of rather small segments one can reduce systematic error introduced by approximations of analytical solutions for the phase transformations at the multiple scattering segment. On the other hand this may slow down the procedure of shower simulation. We found that the optimum length of the multiple-scattering segment is within the range of $`15\mathrm{gr}/\mathrm{cm}^2`$. Very accurate analytical solutions of a multiple-scattering process for a $`e^{},e^+`$ transport allow us to use in calculations such segment length without introducing systematic errors in three-dimensional shower development.
In comparison to other computational codes we used more accurate distributions derived analytically from the multiple scattering theory. On the contrary the standard approach used for example in EGS-IV code does not include fluctuations of the energy losses at the multiple-scattering segment. In addition, in the EGS-IV code the lateral displacement of charged particle at the multiple-scattering segment is taken as $`\stackrel{}{\rho }=0`$ and correspondingly the longitudinal displacement is of $`z=\mathrm{\Delta }l`$. All this necessitates a small segment length and makes significantly more time consuming the shower simulations. Thus for the simulations of ฤerenkov light from the TeV air shower ALTAI code is faster by as much as a few times compared with the EGS-IV.
### 2.2 Emission of ฤerenkov Light
A charged particle ($`e,\mu `$) in an air shower can emit ฤerenkov light when its energy exceeds a certain threshold energy $`\mathrm{E}>\mathrm{E}_{\mathrm{th}}`$. The threshold energy is determined as (Frank & Tamm 1937)
$$E_{th}=\frac{m_oc^2}{\sqrt{2(n1)}}$$
(17)
where $`\mathrm{n}`$ is the refraction index in air, $`\mathrm{m}_\mathrm{o}`$ is the particle mass. Thus at sea level the energy threshold is $`E_{th}`$20 MeV for electron and $``$4 GeV for muons. To describe the altitude dependence of the refraction index we used the following expression (Beliaev et al. 1980)
$$n=1+\eta ,\eta (H)=\eta _o/\rho _o\rho (H)$$
(18)
where $`\eta _o=2.910^4`$, $`\rho _o=1.2210^3\mathrm{g}/\mathrm{cm}^3`$, and $`\rho (H)`$ is the air density at a height H above the sea. The model of a standard atmosphere (Elterman 1968) was used in the simulations.
According to Frank & Tamm (1937) the mean number of ฤerenkon photons emitted in a 1 cm pathlength by electron is described as
$`{\displaystyle \frac{dQ}{dl}}=2\pi \alpha ({\displaystyle \frac{1}{\lambda _1}}{\displaystyle \frac{1}{\lambda _2}})\mathrm{sin}^2\theta _c`$
$`\mathrm{sin}^2\theta _c=1{\displaystyle \frac{1}{(1(mc^2/E)^2)n^2}}`$ (19)
where $`\alpha `$ is a fine structure constant $`\alpha =1/137`$ and $`\theta _c1^o`$ is the opening angle of the ฤerenkov light cone. The spectral region of emission is defined by wavelengths $`\lambda _1`$ and $`\lambda _2`$. In simulations we included sampling of the ฤerenkov light attenuation in the atmosphere due to the Raleigh scattering, ozone and aerosol absorption. The cross-sections for these processes were calculated using the data of Driscoll & Vaughan (1978).
In the shower simulation procedure (PARCHR, GSTCHR) (Konopelko 1990), we first define the total number of ฤerenkov photons $`<\mathrm{n}_{\mathrm{ph}}>`$ emitted at the multiple scattering segment $`\mathrm{\Delta }\mathrm{l}`$. As was discussed above we used rather small multiple scattering segments. Therefore, we can assume that all ฤerenkov photons are emitted from the center of the multiple scattering segment (see Figure 1). To a rather good approximation an intersection of the ฤerenkov light cone with the observation plane forms a circle of a radius $`\mathrm{r}_\mathrm{c}`$ which can be calculated as
$$r_c=(ZZ_{obs})\theta _c$$
(20)
where $`Z`$ and $`Z_{obs}`$ are the height of ฤerenkov light emission and the height of the observation level above the sea, respectively. The number of photons hitting the telescope mirror can be calculated as
$$<\stackrel{~}{n}_{ph}>=\frac{\mathrm{\Delta }L}{2\pi r_c}(1p(Z,Z_{obz}))<n_{ph}>$$
(21)
where $`\mathrm{\Delta }L`$ is length of the circle arc between points B and C (see Figure 2). $`\mathrm{p}(\mathrm{Z},\mathrm{Z}_{\mathrm{obs}})`$ is a probability of the ฤerenkov light attenuation in the atmosphere along the way of a photon propagation. For specific response functions of the telescope camera one may calculate corresponding average number of photoelectrons as $`<\mathrm{n}_{\mathrm{ph}.\mathrm{e}.}>=\xi <\stackrel{~}{\mathrm{n}}_{\mathrm{ph}}>`$, where $`\xi `$ is a photon-to-photoelectrons conversion efficiency. Finally one can use the Poisson distribution in order to simulate the random number of photons (photoelectrons) hitting the telescope. The photons reaching the telescope mirror were uniformly distributed over the circle arc BC (see Figure 2).
In the latest version of the ALTAI code we save all, or a certain fraction, of ฤerenkov photons hitting the telescope. Each photon has a set of 6 variables, $`\stackrel{}{\upsilon }(x,y,\theta _x,\theta _y,z_o,t)`$. $`\mathrm{z}_\mathrm{o}`$ is a height of the photon emission (in $`\mathrm{g}/\mathrm{cm}^2`$) and $`t`$ is a time of photon arrival to the detector. Such database of the photon is used for detailed simulations of the telescope camera response (Hemberger 1998).
## 3 Hadron-nuclei cascade
For simulations of the hadron-nuclei cascade we used the phenomenological model of hadron interactions (Konopelko 1990) which, in the most part, is based on the available accelerator data. The energy spectra of secondary hadrons generated in a $`pA`$ interactions were approximated by use of the radial scaling model (Hillas 1979). In this approach the energy spectra may be presented as
$$xdN/dx=F_{(pq)}(x)H_q(x,E),x=E/E_o$$
(22)
where $`E_o,E`$ are the energies of the primary and secondary particles, respectively. Indeces $`p,q`$ denote the type of primary and secondary particle ($`p,n,\pi ^+,\pi ^{},\pi ^o`$). The basic functions are given below
$`F_{(p,n)\pi ^\pm }=`$
$`F_{(p,n)\pi ^0}=`$
$`1.22(1x)^{3.5}+0.98e^{18x},`$
$`F_{\pi ^\pm \pi ^0}=1.3(1+{\displaystyle \frac{x}{0.45}})^3,`$
$`F_{\pi ^\pm \pi ^\pm }=F_{\pi ^\pm \pi ^0}+0.57e^{4(x1)},`$
$`H_{(p,n)(\pi ^\pm ,\pi ^o)}=[1+{\displaystyle \frac{0.4}{E+0.14}}]^1,`$
$`H_{\pi ^\pm (\pi ^\pm \pi ^o)}=10.88e^{1.8x}`$ (23)
For the total cross-sections of inelastic hadron interactions we used the data given by Shabelski (1986,1987) in the following form
$`\sigma _{\pi ^\pm }=1.2\sigma _{(p,n)},`$
$`\sigma _{(p,n)}=\{\begin{array}{c}258E<E_1,\hfill \\ 273(1+3.2910^2y+\hfill \\ 3.8810^3y^2)E_1<E<E_2,\hfill \\ 258(1+0.07y)E>E_2\hfill \end{array}`$ (28)
$`E_1=100GeV,E_2=10^4GeV,`$
$`y=ln(E/10^3GeV)`$ (29)
The cross-sections from Eqn. (22) are measured in mBarn and correspond to the particle interaction with air. We assume that the mean value of atomic number for air nuclei is of 14.4.
We used a simplified geometrical representation (Murzin 1988) for the total cross-section of the nucleus-nucleus interactions as follows
$`\sigma _A=R^2\sigma _N,`$
$`R={\displaystyle \frac{A^{1/3}+A_0^{1/3}b}{1+A_0^{1/3}b}},`$ (30)
where $`A_0`$ is the effective atomic mass of air ($`A_o=14.4`$) and $`b`$ is the effective radius of nuclei overlapping zone ($`b=1.17`$). $`\sigma _N`$ defines the cross-section of hadron interaction with air.
The transverse momenta of secondary hadrons were calculated according to the distribution given by Ranft (1972)
$$f_{(p,n,\pi ^\pm )}(p_{})=p_{}\frac{e^{Bp_{}^2}+Ce^{Dp_{}}}{\frac{1}{2D}+\frac{C}{D^2}}$$
(31)
where the corresponding parameters (B,C,D) were defined for the interactions with air as described by Ranft et al. (1972). $`p_{}`$ is measured in GeV/c.
The ALTAI code was developed for the simulations in the energy range relevant for the very high energy $`\gamma `$-ray astronomy $`\mathrm{E}50\mathrm{TeV}`$. In this energy range the mean number of kaons ($`\mathrm{K}^\pm ,\mathrm{K}^\mathrm{o}`$) produced in hadron-nuclei interactions is very small compared with the number of emitted pions ($`\pi ^\pm ,\pi ^\mathrm{o}`$). Besides this, in this energy region kaons and pions exhibit similar properties of inelastic interactions (see, e.g., Grishin 1982). Thus in the shower simulations for this restricted energy range we can exclude the kaon production process by introducing appropriate corrections for the probabilities of pion production in the $`(p,n,\pi ^\pm )X`$ inelastic collisions.
For each inelastic hadron interaction (MULTIP) we sampled first the energy of the so-called leading particle ($`p,n,\pi ^\pm `$). This particle carries away the bulk of the primary energy. We assume that the energy of a leading particle E is uniformly distributed within the range $`(0,E_0)`$ where $`E_0`$ is the energy of a primary interacting particle. Note that the type of the leading particle was selected randomly amongst proton and neutron for interaction $`(p,n)+A(p,n)`$, and amongst charged pion or neutral pion for interaction $`(\pi ^\pm ,\pi ^o)+A(\pi ^\pm ,\pi ^o)`$. Thus the total inelasticity coefficient was determined as $`\kappa =1E/E_0`$. At the next step we sampled multiple production of other secondary particles ($`\pi ^\pm ,\pi ^o`$). The type of secondary particle was randomized assuming that $`\pi ^o`$ pions forms, on average, one third of all pions generated in hadron interactions. The energy of secondary particle was simulated using the spectra given by Eqn. (22). We have completed the production of secondary particles when their total energy exceeds the energy of a primary particle. The energy of the last simulated particle was renormalized in order to conserve the total energy in each inelastic interaction.
The transverse momentum of the secondaries was simulated using Eqn. (31). By the renormalization of the transverse momentum of leading particles we allowed to fulfill the momentum conservation for each hadron interaction without distortion of stochastic properties of a multi-particle production process. The test calculations have shown that such an approach describes very well the initial inclusive spectra of the secondary particles if the number of secondaries is relatively high ($`1`$). Note that as almost identical algorithm was developed and tested by Barashenkov & Toneev (1972).
Note that we have included in the code the propagation of a single muon generated in the hadron-nuclei cascade due to the $`\pi ^\pm `$ decay (SGG,TRMUON). The emission of the ฤerenkov light from the muons was included in simulations according to the scheme described above.
## 4 Nucleus-nucleus interactions
We have implemented in the ALTAI code a model of independent nucleon interactions of colliding nuclei with nuclei fragmentation included (TAFRAC, TRAN, FRAG, OVERLAP). In this approach we assume that all nucleons of the projectile nucleus have the same energy determined as $`E_0/A_p`$. $`E_0`$ and $`A_p`$ are the energy and atomic number of projectile nucleus, respectively. All nucleons of the projectile nucleus which overlap with the target nucleus interacted independently with each other. The non-overlapping part of the projectile nucleus decayed into individual nucleons and heavier fragments. The energy of fragment with the atomic number $`A`$ is defined as $`E_f=(E_0A)/A_p`$. We simulated a random number of a fragments according to the probabilities of different channels summarized in Table 1.
## 5 Comparison with other codes and data
An overview of Monte Carlo results on lateral, temporal, and angular characteristics of the ฤerenkov light in air showers of 10 GeV - 1 TeV was recently given by Konopelko (1997). Most of the ฤerenkov light characteristics calculated with the ALTAI code reproduce rather well the results obtained with the MOCCA code (Hillas 1996). The stereoscopic observations of BL Lac object Mkn 501 in 1997 with the HEGRA system of imaging air ฤerenkov telescopes allowed the first measurements of the parameters of the ฤerenkov light images from the $`\gamma `$-ray-induced air showers. The HEGRA data are in excellent agreement with the results of Monte Carlo simulations obtained with ALTAI code (Aharonian et al. 1999a). Using the Mkn 501 data sample comprising 38,000 $`\gamma `$-ray events we have tested in great detail the parameters of ฤerenkov light image orientation and shape (Konopelko et al. 1999a) (see Figure 3) as predicted by the simulations.
The HEGRA stereoscopic system of 5 IACTs was used for the measurements of the lateral distribution of the ฤerenkov light in the $`\gamma `$-ray-induced air showers. These measurements have been compared again with the Monte Carlo simulations using the ALTAI code (Aharonian et al. (1998)). The simulations using the ALTAI code reproduce very well the measured lateral distributions of the ฤerenkov light. Note that the measured shape of the ฤerenkov light lateral distributions is almost independent of the detector simulation procedure but strongly depends on the development in space of a multi-TeV $`\gamma `$-ray shower in the atmosphere.
We made several comparisons of the shape and size (total number of photoelectrons) of the ฤerenkov light images for the proton- and nuclei-induced air shower calculated with the ALTAI and CORSIKA codes (Heck et al. 1997). The CORSIKA code was used with the HDPM model of the proton-nuclei interactions. This model is based on supercollider data and describes rather well the proton-nuclei air showers of energy below 10 TeV. In addition, the HDPM model of CORSIKA allows to perform relatively fast calculations with respect to other more modern shower generators, e.g. VENUS/QGSJET. Despite the different algorithms and schemes in the two codes the results appeared to be almost identical (Plyasheshnikov et al. 1997). The simulations with the ALTAI and CORSIKA codes well reproduce each other. Note that simulations with the ALTAI code are essentially less time consuming. The HEGRA data for the cosmic ray air showers were compared with the ALTAI simulations. Good agreement between simulations and data provided a precise measurement of the cosmic ray proton spectrum in the energy range $``$1-5 TeV (Aharonian et al. 1999b).
We show in Figure 3 the distribution of so-called mean scaled Width parameter for primary $`\gamma `$-rays and cosmic rays extracted from the HEGRA data as well as from the Monte Carlo simulations. The simulations fit very well the data. Recently, we compared the relevant results of the ALTAI simulations with the cosmic ray data taken with a 10 m Whipple imaging air ฤerenkov telescope during the 1995/1996 Crab Nebula observations (Konopelko 1999b). For these simulations we have used the standard detector response functions offered by the Whipple collaboration. The simulations reproduced very well the shape of recorded ฤerenkov light images (see Figure 4).
As mentioned above the upper energy of the shower simulations is about 50 TeV. It is limited by the simplified phenomenological model of the hadronic cascade. Although in the case of pure electromagnetic showers one can extend simulations using the ALTAI code to much higher energies without breaking any model restrictions such extension may need the introduction of a number of changes into the code for the correct treatment of a larger number of particles.
## 6 Summary
Here we have presented a detailed description of the numeric code ALTAI, which was developed for simulations of ฤerenkov light from extensive air showers. This code allows to make very fast and accurate calculations of the response of the ground-based ฤerenkov detectors used in VHE $`\gamma `$-ray astronomy. Although the code was designed for calculating the parameters and characteristics of the imaging air ฤerenkov telescopes it can be rearranged with minor changes in order to simulate the responses of showerfront sampling experiments like MILAGRO, or Tibet AS-$`\gamma `$.
We tested our computational code against the data taken with two currently operating detectors, the HEGRA system of imaging air ฤerenkov telescopes and the state-of-the-art single 10 m Whipple telescope. These comparisons have proven the high precision of the simulations. The computational code ALTAI is an effective tool for producing the simulated data for VHE $`\gamma `$-ray astronomy. Note that the forthcoming imaging ฤerenkov detectors (CANGAROO IV, H.E.S.S., MAGIC, VERITAS) will need a large amount of simulated data. Therefore the performance of these instruments may benefit from the use of the ALTAI code.
## 7 Acknowledgments
The work for the ALTAI code was primarily initiated and substantially advanced at the Tomsk Technological University, Tomsk, Russia, where the major algorithms of the shower simulation scheme were developed. Authors thank Prof. A.M. Kolchuzhkin for a significant contribution for all these studies. In the most part the ALTAI code was developed and programmed at the Altai State University, Barnaul, Russia. The authors thank Dr K.V. Vorobjev, Prof A.M. Lagutin, Dr V.A. Litvinov and Prof V.V. Uchaikin for their valuable input and support of this activity. The computational code ALTAI was used and further developed at the Max-Planck-Institut fรผr Kernphysik, Heidelberg. The authors would like to acknowledge the contribution and support of all members of the Heidelberg group in particular Prof F. Aharonian, Dr M. Hemberger, Prof W. Hofmann, J. Kettler and Prof H.J. Vรถlk. AKK thanks Prof T.C. Weekes for support and supervision of a short-term project at the University of Arizona, Tucson, and at the Whipple Observatory, Harvard-Smithsonian Center for Astrophysics, Amado.
We are grateful to an anonymous referee for detailed and helpful comments.
## 8 Availability
The ALTAI code is written in FORTRAN and may be easily installed at any computer platforms maintaining FORTRAN compiler. Regarding the availability of the code contact: alexander.konopelko@mpg.mpi-hd.de |
warning/0001/astro-ph0001141.html | ar5iv | text | # CTQ 839: Candidate for the Smallest Projected Separation Binary Quasar Based on observations carried out at the Cerro Tololo Interamerican Observatory (CTIO), the Las Campanas Observatory (LCO), and and the National Radio Astronomy Observatory (NRAO) Very Large Array (VLA). CTIO is part of the National Optical Astronomy Observatories, which are operated by the Association of Universities for Research in Astronomy, Inc., under cooperative agreement with the National Science Foundation. The NRAO is a facility of the National Science Foundation operated under cooperative agreement by Associated Universities, Inc.
## 1 INTRODUCTION
The discovery of similar redshift, small separation optical-optical ($`O^2`$) double quasars (pairs where both components are optically bright and radio faint; see Kochanek, Falco, and Muรฑoz (1999)) can yield a range of information on cosmological scales. Such systems are intensively investigated as gravitational lens candidates, and if confirmed, can yield measurements of the Hubble constant (Refsdal 1964) as well as statistical constraints on the cosmological constant (Kochanek 1996). If an $`O^2`$ pair is confirmed as a binary quasar, it can provide clues regarding the triggering of nuclear activity in galaxies (Osterbrock 1993) as well as the evolution of early ($`z>2`$) galaxy mergers (Barnes 1999). The observed frequency of binary quasars are also important in understanding gravitational lensing statistics; Kochanek, Falco, and Muรฑoz (1999) have recently used the observed paucity of $`O^2R^2`$ quasar pairs (pairs with both components bright in optical and radio) as compared to the number of $`O^2`$ pairs to conclude that the majority of known wide separation quasar pairs must be binary quasars. In this paper, we report the discovery of the new small separation $`O^2`$ quasar pair CTQ 839 and investigate the nature of the system as either a gravitational lens or binary quasar.
CTQ 839 (2$`^\text{h}`$ 52$`^\text{m}`$ 57$`\stackrel{\mathrm{s}}{\mathrm{.}}`$86, -32$`\mathrm{ยฐ}`$ 49$`\mathrm{}`$ 8$`\stackrel{}{\mathrm{.}}`$6, J2000.0) was originally identified as a $`z=2.24`$ quasar from the Calรกn-Tololo Survey (CTS) (Maza et al. 1996). The CTS is an objective prism survey conducted at Cerro Calรกn using photographic plates obtained at the Cerro Tololo Interamerican Observatory (CTIO) and is aimed at discovering quasars and emission-line galaxies in the southern hemisphere. To date, the CTS has identified $``$ 1000 southern hemisphere quasars as well as two confirmed gravitational lenses: CTQ 286 (Claeskens et al. 1996) and CTQ 414 (Morgan et al. 1999). During November 1998, approximately 100 CTS quasars were observed with the 1.5 m telescope at CTIO as part of a five night observing program to discover new gravitationally lensed quasars. This particular run has yielded one definite lensed system, the complex gravitational lens HE 0230-2130 (Wisotzki et al. 1999), in addition to the double quasar CTQ 839 presented here. Optical images of CTQ 839 immediately revealed two components in the system, with a separation of 2$`\stackrel{}{\mathrm{.}}`$1 evident in all observed filters. We describe these observations, as well as followup $`R`$ and $`H`$ band observations conducted at the Las Campanas Observatory (LCO), in ยง2. In ยง3, we present our analysis of the quasar components from spectra taken at CTIO, while ยง4 presents radio observations taken with the Very Large Array (VLA) in July of 1999. In ยง5 we discuss lens modeling and interpretation of the system, while ยง6 summarizes our findings and conclusions for CTQ 839.
## 2 OBSERVATIONS AND REDUCTION
### 2.1 Initial Optical Imaging
Initial optical observations of CTQ 839 were taken with the CTIO 1.5 m telescope by two of us (N.D.M. and P.L.S.) on the nights of 1998 November 12 and 15. The SITe 2048 #6 CCD camera was used, although only the central 1536 $`\times `$ 1536 of the array was read out. The telescope was operated at an $`f`$ ratio of $`f/13.5`$, providing a field of view of 6.1 arcmin square and a scale of 0$`\stackrel{}{\mathrm{.}}`$2407 per pixel. The gain and read noise of the detector were 2.9 e<sup>-</sup>/ADU and 4.0 e<sup>-</sup>, respectively. Multiple 300 s exposures of CTQ 839 were taken in Johnson $`B`$ and $`R`$ and Cousins $`I`$ filters with FWHM seeing conditions ranging from 0$`\stackrel{}{\mathrm{.}}`$9 to 1$`\stackrel{}{\mathrm{.}}`$1. Multiple $`BVRI`$ exposures of the Landolt standard field Rubin 149 (Landolt 1992) were also taken on the 15<sup>th</sup> for use in photometric calibration. Table 1 presents a log of the CTIO observations. Figure 1 shows a 6 arcmin square exposure of CTQ 839 and nearby stars from one of the $`I`$ band frames.
The CCD frames were bias subtracted, trimmed, and flatfield corrected using the Vista reduction program. The flatfield frames consisted of twilight exposures taken on multiple nights of the CTIO observing run, and were cleaned of cosmic rays using โautocleanโ, a program written and kindly supplied by J. Tonry. As noted in ยง1, images of CTQ 839 are separable into two components in all filters (see Figure 2). The double images were therefore fit with two empirical point spread functions (PSFs) using a variant of the program DoPHOT (Schechter et al. 1993), designed to deal with close, point-like and extended objects (Schechter and Moore 1993). Star # 5 identified in Figure 1 provided the empirical PSF. Results for the relative positions and apparent magnitudes of the brighter and fainter components (denoted by A and B, respectively) are presented in Table 2. Here we present results from simultaneous fitting of magnitudes and relative positions of the two components. Magnitude solutions using fixed separations differ by $`<0.01`$ mag in all filters. One of the $`B`$ band frames (#54), which registered a cosmic ray detection $`1\mathrm{}`$ north of component B, was omitted from the analysis.
It can be seen that the A:B flux ratio exhibits a rather strong dependence with filter, dropping from -2.56 mag in $`B`$ band down to -1.85 mag at $`R`$ and $`I`$ wavelengths. The $`B`$ band separation of $`2\stackrel{}{\mathrm{.}}`$064 is also $``$ 0$`\stackrel{}{\mathrm{.}}`$03 smaller than the separations found at $`R`$ and $`I`$ wavelengths of 2$`\stackrel{}{\mathrm{.}}`$098 and 2$`\stackrel{}{\mathrm{.}}`$092, respectively. The reason for this difference in image separations at blue and red wavelengths is not immediately clear to us, although we have ruled out variations in the CCD scale as a possible source. Although the separations are consistent with gravitational lensing, the wavelength dependent flux ratio would require strong reddening and/or microlensing of the quasarโs light to conform with the lensing hypothesis. PSF subtraction of components A and B using stacked images for each filter showed no indication of a third component in the system, although deeper and redder searches for possible signs of a lensing galaxy are described in the following subsection.
The observations reported above used the 1.5 m telescope at CTIO operating at a focal length of $`f/13.5`$. The 1.5 m is a Ritchey-Chrรฉtien telescope, designed to be free of comatic aberration at $`f/7.5`$, but not at $`f/13.5`$. Observations were carried out at $`f/13.5`$ in order to make use of the smaller pixel scale at that $`f`$ ratio. The observations reported above therefore suffer from coma, which introduces an off-axis distortion in the shape of the PSF across the CCD chip. This effect grows with increasing distance from the center of the chip, and, given the relatively good seeing conditions during the observing run, can cause PSF magnitudes to systematically underestimate corresponding aperture magnitudes by as much as 0.1 mag for peripheral stars. In performing the PSF analysis described above, we were careful to choose a PSF star as close as possible to CTQ 839 ($``$ 30 arcseconds away) in order to minimize the effects of coma. The resulting PSF and aperture magnitudes for the combined flux from components A and B agree to within 0.035 mag in $`B`$, and better than 0.010 mag in $`R`$ and $`I`$.
For use with future observations, aperture magnitudes were determined for 8 field stars within a 4 arcminute radius from the target quasar. An aperture diameter of 9$`\stackrel{}{\mathrm{.}}`$6 was used. Observations were calibrated using the Rubin 149 standard field (Landolt 1992) mentioned above, with extinction coefficients taken from the 1990 CTIO Facilities Manual ($`k_B=0.22,k_V=0.11,k_R=0.08,k_I=0.04`$). These results are presented in Table 3, along with corresponding astrometric solutions for the selected reference stars.
### 2.2 Follow-up Optical and Infrared Imaging
In an effort to further probe the system for the possible presence of a lensing galaxy, follow-up $`R`$ and $`H`$ band observations were carried out within a few months of the original observations. On 23 December 1998, a series of six 10 minute $`R`$ band exposures of CTQ 839 were taken by one of us (E.C.) with the du Pont 2.5 m telescope at LCO. The Tek #5 detector set in the #3 gain position was employed, providing a gain of 3.0 e<sup>-</sup>/ADU, a readnoise of 7.0 e<sup>-</sup>, and scale of 0$`\stackrel{}{\mathrm{.}}`$2604 per pixel. Seeing conditions for the series of observations were slightly better than those taken at CTIO, with an average FWHM of 0$`\stackrel{}{\mathrm{.}}`$95. After bias-subtraction and flatfielding, the images were co-added using integer pixel shifts. The resulting stacked image was then reduced in the same manner as described above. Given the longer exposure times as compared to the CTIO data, star \# 5 became saturated on the CCD detector and a new star (# 4 in Figure 1) provided the empirical PSF.
Results from PSF analysis yield an A:B flux ratio of 5.64 and a separation of 2$`\stackrel{}{\mathrm{.}}`$092 at a PA of 160$`\stackrel{}{\mathrm{.}}`$8 E of N. Both of these results compare well with the CTIO $`R`$ band solutions presented in Table 2. In the two upper panels of Figure 2, we show an excised portion from the stacked image of CTQ 839 (left), along with residuals after PSF subtraction (right). (The bottom two panels show $`H`$ band observations taken two months later; see the following subsection). In the residual panels, tick marks indicate the centroid locations of components A and B as determined from the PSF fits. The orientation of the images are the same in all panels of the figure.
When ground-based observations of close separation, doubly lensed systems are fit with two PSFs, a characteristic residual pattern emerges after PSF subtraction. These patterns arise from using only two PSFs to model the light from both quasar images as well as the lensing galaxy, and consist of undulating regions of positive and negative residuals. For instance, one typical residual pattern consists of a โdivot-bump-divotโ undulation, as seen by Schechter et al. (1998) and Morgan et al. (1999). These types of patterns are not present in the upper right panel of Figure 2, and therefore argues against the presence of any significant third component in the system. (The residuals located around the center of component A, which do not show significant structure, are likely due to imperfections in component Aโs fit). In order to place a magnitude limit on this null result, we inserted a series of gaussian profiles of varying magnitudes into the stacked $`R`$ band image and investigated the residual pattern that emerged after fitting each system with two PSFs. The position of the gaussian profile was dictated by the singular isothermal sphere (SIS) model, that is, the center of the gaussian was placed collinear with the centroids of components A and B, with the ratio of relative separations from the two components given by the LCO $`R`$ band flux ratio. The FWHM of the profile was dictated by the average seeing conditions for the LCO run. The profileโs magnitude, starting at the same brightness as component B, was successively dimmed by 0.1 mag increments until the characteristic residual pattern was no longer unmistakable. We conclude that we would have confidently detected any third component in the system brighter than $`R=22.5`$ at the expected position for a lensing galaxy.
Following the null result in $`R`$ band, infrared observations of CTQ 839 in $`H`$ band (1.65 microns) were carried out by two of us (G.B. and I.T.) at LCO on 1999 February 5. The IRCAM infrared camera (Persson et al. 1992) mounted on the du Pont 2.5 m telescope was used at a scale of $`0\stackrel{}{\mathrm{.}}3478`$ per pixel. The total integration time on CTQ 839 was 3250 seconds, divided into 13 individually dithered frames of $`5\times 50`$ s each. The reduction procedures were carried out by one of us (S.E.P.) and followed closely those described in Persson et al. (1998). The 13 fully processed frames were combined into a final stacked image, from which photometry was obtained. Given the narrow (256 $`\times `$ 256) field of view of the detector, only star #5 as shown in Figure 1 was available to provide the empirical PSF. This star is rather bright at $`H=13.4`$, and as a consequence, the central pixel value for the object required a linearity correction at the $`5\%`$ level. The residual error associated with this correction affects aperture magnitudes for star #5 no larger than 0.01 mag, which is not critical for the analysis that follows.
Results from empirical PSF analysis yield an $`H`$ band A:B flux ratio of 8.42, which is 1.5 times larger than the $`R`$ and $`I`$ band results. An analysis using an analytical PSF, as described in Schechter et al. (1993), yields an A:B flux ratio of 8.30. The $`m_Am_B`$ magnitude differences between the empirical and analytical models therefore differ on the $`0.02`$ mag level. The separation between the two components as determined from empirical PSF fitting was 2$`\stackrel{}{\mathrm{.}}`$101 $`\pm `$ 0$`\stackrel{}{\mathrm{.}}`$020, which agrees with the separations found at $`R`$ and $`I`$ wavelengths. An excised portion of the $`H`$ band observation centered CTQ 839, along with residuals after fitting with two empirical PSFs, are shown in the bottom two panels of Figure 2. The residual image again shows no indication of a significant third component in the system. The clustering of positive residuals at the centroid of Aโs fit, which are of order $`5\%`$ of Aโs peak intensity, are consistent with the imperfection in the empirical PSF template described above. Using the identical procedure outlined for the LCO $`R`$ band data, we conclude that we would have confidently detected a third component at the expected position of a lensing galaxy brighter than $`H=17.4`$.
## 3 SPECTROSCOPY
Spectra of both components of CTQ 839 were obtained on 1998 December 27 by one of us (M.T.R.) with the 4.0 m telescope at CTIO. The R-C Spectrograph together with the Blue Air Schmidt camera and Loral 3 K $`\times `$ 1 K CCD were used. The wavelength scale for the observations was 1.205 ร
per pixel, with a wavelength range from 3670 to 7210 ร
, and a long, 1$`\mathrm{}`$ wide slit; seeing conditions were 1$`\stackrel{}{\mathrm{.}}`$3 FWHM. With the slit orientation placed perpendicular to the component separation, one 360 s exposure centered on component A and two 1800 s exposures centered on component B were taken. Three spectrophotometric standard stars from Baldwin and Stone (1984) were also observed during the night for flux calibration purposes. All spectra were bias-subtracted and flatfield corrected using standard IRAF procedures. The observations of CTQ 839 were carried out close to the zenith, with airmasses ranging from 1.001 to 1.010, so differential lightlosses due to atmospheric refraction were not a problem. However, with a separation distance smaller than twice the seeing disc, some contamination of the fainter componentโs spectra with light from component A was unavoidable. This contamination, which we have estimated to be $``$5% of Bโs raw spectra, is straightforward to compensate for assuming gaussian profiles and a knowledge of the seeing disc and slit characteristics. In Figure 3, we show the spectra of component A and the decomposed average spectra of component B, along with the identification of prominent emission features.
The spectra of A and B both show quasar emission profiles at similar redshifts. Both components exhibit appropriately redshifted Lyman $`\alpha `$ $`\lambda `$1216, N V $`\lambda `$1240, and C IV $`\lambda `$1549 emission features, while A also displays O I $`\lambda `$1304, Si IV $`\lambda `$1397 \+ O IV $`\lambda `$1402, and C III\] $`\lambda `$1909 emission lines as well. Table 4 lists the strongest emission features for both components, as well as redshift determinations based on gaussian fits to the peaks of the profiles. The redshifts of both spectra are consistent with a $`z=2.24`$ quasar. A cross-correlation between the two spectra yields relative redshifts that agree at the $`100\text{ km s}^1`$ level.
The quotient of the two spectra, shown in Figure 4, shows strong evidence for differences in the equivalent widths of related emission features. The prominent peaks present in the quotient spectrum, corresponding to the C III\], C IV, and Ly $`\alpha `$+N V emission features, are consistent with A having progressively stronger emission lines with respect its continuum than B does as one moves from the red to blue wavelengths. There is also an indication for a harder blue continuum in A than in B, longward of the Lyman $`\alpha `$ emission feature. As discussed further in ยง6, these spectral differences between the two components are difficult to reconcile under the lensing hypothesis.
## 4 RADIO OBSERVATIONS
In order to search for radio emission from CTQ 839, we first queried the NRAO VLA Sky Survey (NVSS) (Condon et al. 1998) at the position of the brighter optical component. The NVSS is a 1.4 GHz radio continuum survey of all the sky north of $`40^{}`$, carried out with the NRAO Very Large Array (VLA) in its D configuration. The FWHM resolution is 45 arcseconds and the quoted completeness limit is 2.5 mJy. However, no radio source was found within 3 arcminutes of the optical position.
A deeper probe for radio emission from CTQ 839 was performed by one of us (J.N.W.) on 1999 July 21 using the VLA. The search was carried out with a 15 minute integration at 8.4 GHz, while the VLA was in the A configuration. The FWHM of the synthesized beam was 0.5 arcseconds in the N/S direction and 0.2 arcseconds in the E/W direction. No significant sources of radio flux were detected within 5 arcseconds of the position of the brighter optical component of CTQ 839. The rms noise level in this field was 0.17 mJy per synthesized beam, so our observation rules out (at the 5$`\sigma `$ level) any sources of compact flux above 0.85 mJy. We therefore classify CTQ 839 as an $`O^2`$ quasar pair.
## 5 SIS MODEL AND INTERPRETATION
If CTQ 839 is a gravitational lens system, then the failure to detect a third component can place constraints on the characteristics of any lensing galaxy that may be present <sup>1</sup><sup>1</sup>1In this section, we will continue to refer to the โlensing galaxyโ, although we realize its existence is by no means conclusive.. In the following section, we use a simple SIS model to describe the galaxy potential in order to predict the lensing galaxyโs luminosity as a function of distance. Combined with the magnitude detection limits discussed in ยง2, we investigate the types of bounds that can be placed on the lens galaxy evolutionary type and redshift.
The SIS model is characterized by three parameters: two angular coordinates for the center of the potential, as well as the associated line-of-sight velocity dispersion $`\sigma `$ which measures the depth of the potential well. For a SIS model, the velocity dispersion of the lensing potential is related to the image separation $`\theta `$ by
$$\frac{\sigma ^2}{c^2}=\frac{D_S}{D_{LS}}\frac{\theta }{8\pi }$$
(1)
where $`D_S`$ and $`D_{LS}`$ are angular diameter distances from the observer to the source and from the lens to the source, respectively, and $`\theta `$ is measured in radians (see, for example, Narayan and Bartelmann 1998). We assume the galaxyโs central velocity dispersion is related to its $`B`$ band luminosity $`L`$ via a Faber-Jackson relationship of the form
$$\frac{L}{L_{}}=\left(\frac{\sigma }{\sigma _{}}\right)^\gamma ,$$
(2)
where, following Keeton, Kochanek, and Falco (1998), we adopt $`\sigma _{}=220\text{ km s}^1`$, $`\gamma =4.0`$ for early-type galaxies, and $`\sigma _{}=144\text{ km s}^1`$, $`\gamma =2.6`$ for late-type galaxies. $`L_{}`$ corresponds to a $`B`$ band magnitude of $`M_B^{}=19.7+5\mathrm{log}h`$, where the Hubble constant has been parameterized by $`H_o=100h\text{ km s}^1\text{ Mpc}^1`$. For a given redshift $`z_l`$ of the lensing galaxy, we can then estimate its cosmological distance modulus via
$$m_{AB}(\lambda _{obs})M_{AB}(\lambda _{rest})=5\mathrm{log}\frac{D_L}{10\text{ pc}}+7.5\mathrm{log}(1+z{}_{l}{}^{})$$
(3)
where $`D_L`$ is the angular diameter distance of the lensing galaxy. For the purpose of calculating $`M_{AB}(\lambda _{rest})`$, spectral energy distributions (SEDs) for both early- and late-type galaxies were obtained from Lilly (1997), which consisted of interpolation and extrapolation of the SEDs presented by Coleman, Wu, and Weedman (1980). The SEDs are then normalized to the Faber-Jackson luminosity at $`4400(1+z)`$ ร
, which yields the predicted AB magnitudes. Transformations to the $`BVRI`$ system from the $`AB`$ magnitudes were performed by adding -0.110, 0.011, 0.199, and 0.456, respectively, to the $`AB`$ magnitudes (Fukugita, Shimasaku, and Ichikawa 1995).
We have calculated predicted $`R`$ band magnitudes for the lensing galaxy as a function of lensing redshift for both an $`\mathrm{\Omega }_m=1,\mathrm{\Omega }_\mathrm{\Lambda }=0`$ Einstein-de Sitter universe and an $`\mathrm{\Omega }_m=0.3,\mathrm{\Omega }_\mathrm{\Lambda }=0.7`$ open universe (See Figure 5). For the Einstein-de Sitter cosmology, it can be seen that the late-type spiral model lies above (i.e., brighter than) the detection threshold by a full magnitude for the entire range of $`z_l`$, and makes it therefore an unlikely model for the lensing galaxy. For the same cosmology, the early-type elliptical model requires $`1.0z_l2.0`$ for consistency with the detection limit found in ยง2. At a redshift of $`z_l=1`$, the elliptical galaxy model is already rather luminous, with an intrinsic luminosity of $`5L_{}`$ (corresponding to a velocity dispersion of $`325`$ km s<sup>-1</sup>). We can estimate the likelihood of finding a lensing galaxy within the above redshift range using the procedures of Kochanek (1992). Using the critical lens radius of $`r=1\stackrel{}{\mathrm{.}}045`$ for CTQ 839, we compute a median redshift for the lensing galaxy of $`z=0.46`$, with a $`2\sigma `$ probability interval of $`0.11z0.93`$. Thus, under the lensing hypothesis, the lensing galaxy ought to have been seen in an Einstein-de Sitter cosmology in more than 95% of such cases.
The constraints on the existence of the lensing galaxy are far less stringent for the $`\mathrm{\Lambda }`$ dominated cosmology. For the $`\mathrm{\Omega }_m=0.3,\mathrm{\Omega }_\mathrm{\Lambda }=0.7`$ universe, consistency with the $`R`$ band detection limit from ยง2 requires $`0.7z_l2.1`$. The median redshift is found to be $`z=0.57`$ with a $`2\sigma `$ probability range of $`0.16z1.07`$, which does not significantly constrain the redshift of the lensing galaxy. Thus while the existence of the lensing galaxy is highly unlikely in an Einstein-de Sitter universe, the $`\mathrm{\Lambda }`$ dominated model cannot argue for or against the lensing hypothesis.
## 6 SUMMARY AND CONCLUSIONS
Although it may be attractive to explain CTQ 839 as a gravitational lens, there are clearly a number of characteristics of the system that make the lensing hypothesis less than convincing. For example, while the observed image separation of $`2\stackrel{}{\mathrm{.}}1`$ is typical of known double gravitational lens systems, the broadband A:B flux ratios (10.4:1 in $`B`$, 5.5:1 in $`R`$ and $`I`$, and 8.4:1 in $`H`$) exhibit a rather large variation with wavelength. Since the detection limit for a third component in the system is $``$ 22.5 in $`R`$, the smaller $`R`$ and $`I`$ band flux ratios is likely not flux augmentation of component B from an intervening galaxy. Also, if this was the case, the SEDs of early-type galaxies would predict an even smaller flux ratio in $`H`$ band, which is not observed. Extinction of component Bโs light by a line of sight absorber is a possible explanation, although such an absorber would have to preferentially absorb more flux at $`B`$ and $`H`$ wavelengths and less so at $`R`$.
Microlensing of quasar light by an intervening galaxy remains a possible explanation for the observed differences in flux ratios, although the situation is highly contrived. First, we note that the quotient spectra shown in Figure 4 exhibits an enhancement of blue continuum flux in component A as compared to component B shortward of the $`\lambda _{obs}5500`$ร
mark, which is consistent with microlensing of component Aโs light by stars in an intervening galaxy (Kayser et al. 1986). Such an effect has already been observed in at least one confirmed gravitational lens, HE 1104-1805 (Wisotzki et al. 1993). However, the observed differences in the line strengths of respective emission features for the two components are difficult to reconcile under the microlensing scenario. The line strengths of the emission features, which are thought to arise from a region roughly an order of magnitude larger than the continuum emitting region, ought not to be strongly affected by microlensing.
We therefore conclude that CTQ 839 is unlikely to be a gravitationally lensed system. The broadband flux differences, spectral dissimilarities, and failure to detect a lensing galaxy all argue against (although do not explicitly rule out) the gravitational lensing explanation for CTQ 839. If CTQ 839 is not a lens, it must be two separate quasars. The nearly identical redshifts derived from the spectra of the two components would then argue for a physical binary system. At a separation of 2$`\stackrel{}{\mathrm{.}}`$1 and a redshift of $`z=2.24`$, the projected separation of the system is 8.3 $`h^1`$ kpc ($`\mathrm{\Omega }_m=1`$), which would make CTQ 839 the smallest projected separation binary quasar currently known (Kochanek et al. 1998).
N.D.M. and P.L.S. gratefully acknowledge the support of the U.S. National Science Foundation through grant AST96-16866. J.N.W. thanks the Fannie and John Hertz Foundation for financial support. J.M. thanks FONDECYT, Chile, for support through grant 1980172. M.T.R. acknowledges partial support from FONDEYCT, Chile, through grant 19890659 and a Cรกtedral Presidencial (1996). |
warning/0001/physics0001057.html | ar5iv | text | # ON VARIATIONAL SOLUTION OF THE FOUR-BODY SANTILLI-SHILLADY MODEL OF ๐ปโ MOLECULE
## 1 Introduction
In this paper, we consider the four-body Santilli-Shillady isochemical model of $`H_2`$ molecule characterized by additional short-range attractive Hulten potential between the electrons. This potential is assumed to lead to bound state of electrons called isoelectronium. The restricted three-body Santilli-Shillady model (stable and point-like isoelectronium) of $`H_2`$ has been studied in ref. , in terms of exact solution. For the mass of isoelectronium $`M=2m_e`$, this solution implied much lower energy than the experimental one so we varied the mass and obtained that $`M=0.308381m_e`$ fits the experimental binding energy, $`E_{exper}[H_2]=1.174474\text{ a.u.}`$ up to six decimal places, although at bigger value of the internuclear distance, $`R=1.675828\text{ a.u.}`$ in contrast to $`R_{exper}[H_2]=1.4011\text{ a.u.}`$ We realize that the three-body model is capable to represent the binding energy but it is only some approximation to the four-body model, and one should study the general four-body hamiltonian of the Santilli-Shillady model as well.
In the present paper, we use Ritz variational approach to the four-body Santilli-Shillady isochemical model of $`H_2`$ molecule, i.e. without restriction that the isoelectronium is stable and point-like particle, in order to find the ground state energy and bond length of the $`H_2`$ molecule.
In Sec. 2, we analyze some features of the four-body Santilli-Shillady isochemical model of $`H_2`$ molecule.
In Sec. 3, we apply Ritz variational approach to the four-body Santilli-Shillady model of $`H_2`$ molecule. We calculate Coloumb integral for the cases of Hulten potential (Sec. 3.1.1), exponential screened Coloumb potential (Sec. 3.1.2), and Gaussian screened Coloumb potential (Sec. 3.1.3). Owing to Gegenbauer expansion, exchange integral has been calculated for the case of exponential screened potential, with some approximation made (Sec. 3.1.4). Exchange integrals for the Hulten potential and the Gaussian screened Coloumb potential have not been derived, and require more study. We present main details of calculations of the Coloumb and exchange integrals which have been appeared to be rather cumbersome, especially in the case of Hulten potential.
In Sec. 3.2, we make numerical fitting of the variational energy for the case of exponential screened Coloumb potential $`V_e`$. Also, we estimate the weight of the isoelectronium phase. However, we use all the important results of the analysis made for the Hulten potential $`V_h`$.
1) We conclude that the $`V_e`$-based model with the one-level isoelectronium is capable to fit the experimental data on $`H_2`$ molecule (both the binding energy $`E`$ and the bond length $`R`$). This is in confirmation of the results of numerical HFR approach (SASLOBE routine) to the $`V_g`$-based model of ref. .
2) One of the interesting implications of the Ritz variational approach to the Hulten potential case is that the correlation length parameter $`r_c`$, entering the Hulten potential, and, as a consequence, the variational energy, should run discrete set of values during the variation. In other words, this means that only some fixed values of the effective radius of the one-level isoelectronium are admitted, in the original Santilli-Shillady model, within the framework of the Ritz approach. This highly remarkable property is specific to the Hulten potential $`V_h`$ while it is absent in the $`V_e`$, or $`V_g`$-based models.
3) Also, we achieved an estimation of the weight of the isoelectronium phase for the case of $`V_e`$-based model which is appeared to be of the order of 1%โฆ6%. This weight has been estimated from the energy contribution, related to the exponential screened potential $`V_e`$, in comparison to the contribution related to the Coloumb potential.
4) Another general conclusion is that the effective radius of the isoelectronium $`r_c`$ should be less that $`0.25\text{ a.u.}`$
We note that the weight of the phase does not mean directly a time share between the two regimes, i.e., 1โฆ6% of time for the pure isoelectronium regime, and 99โฆ94% of time for the decoupled electrons regime. This means instead relative contribution to the total energy provided by the potential $`V_e`$ and by the usual Coloumb potential between the electrons, respectively. As a consequence, the weight of the isoelectronium phase, which can be thought of as a measure of stability of the isoelectronium, may be
1. Different from the obtained 1โฆ6% when calculated for some other characteristics of the molecule, e.g., for a relative contribution of the pure isoelectronium to the total magnetic moment of the $`H_2`$ molecule;
2. Different from the obtained 1โฆ6% for the case of the original Hulten potential $`V_h`$.
So, the result of the calculation made in this paper is not the final result implied by the general four-body Santilli-Shillady model of $`H_2`$ molecule since the latter model is based on the Hulten potential $`V_h`$. This paper can be viewed only as a preliminary study to it. However, we have made some essential advance in analyzing the original Hulten potential case (Sec. 3.1.1), which we have used in the $`V_e`$-based model.
Below, we describe the procedure used in Sec. 3 in a more detail. In Ritz variational approach, the main problem is to calculate analytically so called molecular integrals. The variational molecular energy, in which we are interested in, is expressed in terms of these integrals; see Eq.(3.2). These integrals arise when using some wave function basis (usually it is a simple hydrogen ground state wave functions) in the Schrรถdinger equation for the molecule. The main idea of the Ritz approach is to introduce parameters into the wave function, and vary them, together with the internuclear distance parameter $`R`$, to achieve a minimum of the molecular energy. In the case under study, we have two parameters, $`\gamma `$ and $`\rho `$, where $`\gamma `$ enters hydrogen-like ground state wave function (3.10), and $`\rho =\gamma R`$ measures internuclear distance. These parameters should be varied (analytically or numerically) in the final analytical expression of the molecular energy, after the calculation is made for the associated molecular integrals.
However, the four-body Santilli-Shillady model of $`H_2`$ molecule suggests additional, Hulten potential interaction between the electrons. The Hulten potential contains two parameters, $`V_0`$ and $`r_c`$, where $`V_0`$ is a general factor, and $`r_c`$ is a correlation length parameter which can be viewed as an effective radius of the isoelectronium; see Eq. (3.23). Thus, we have four parameters to be varied, $`\gamma `$, $`\rho `$, $`V_0`$, and $`r_c`$. The introducing of Hulten potential leads to modification of some molecular integrals, namely, of the Coloumb and exchange integrals; see Eqs. (3.5) and (3.7). The other molecular integrals remain the same as in the case of usual model of $`H_2`$, and we use the known analytical results for them. So, we should calculate the associated Coloumb and exchange integrals for the Hulten potential to get the variational energy analytically. In fact, calculating of these integrals, which are six-fold ones, constitutes the main problem here. Normally, Coloumb integral, which can be performed in bispherical coordinates, is much easier than the exchange one, which is performed in bishperoidal coordinates.
Calculation of the Coloumb integral for Hulten potential, $`V_h`$, appeared to be rather nontrivial (Sec. 3.1.1). Namely, we used bispherical coordinates, and have faced several special functions, such as polylogarithmic function, Riemann $`\zeta `$-function, digamma function, and Lerch function, during the calculation. Despite the fact that we see no essential obstacles to calculate this six-fold integral, we stopped the calculation after fifth step because sixth (the last) step assumes necessity to calculate it separately for each integer value of $`\lambda ^1(2\gamma r_c)^1`$, together with the need to handle very big number of terms. During the calculations, we were forced to use the condition that $`\lambda ^1`$ should take integer values in order to prevent divergency of the Coloumb integral for Hulten potential. Namely, some combination of terms containing Lerch functions gives a finite value only if this condition holds. This condition is specific to Hulten potential. Note also that we can not get general form of a final expression for the Coloumb integral for Hulten potential because Lerch functions entering the intermediate expression (after the fifth step, see Eq.(3.80)) can be integrated over only for a concrete numerical value of their third argument.
In order to proceed with the Santilli-Shillady approach, we invoke to two different simplified potentials, the exponential screened Coloumb potential, $`V_e`$, and the Gaussian screened Coloumb potential, $`V_g`$, instead of the Hulten potential $`V_h`$. They both mimic well Hulten potential at short and long range asymptotics, and each contains two parameters, for which we use the notation, $`A`$ and $`r_c`$. In order to reproduce the short range asymptotics of Hulten potential the parameter $`A`$ should have the value $`A=V_0r_c`$, for both the potentials. The Coloumb integrals for these two potentials have been calculated exactly (Secs. 3.1.2 and 3.1.3) owing to the fact that they are much simpler than the Hulten potential. Particularly, we note that the final expression of the Coloumb integral for $`V_g`$ contains only one special function, the error function $`\text{erf}(z)`$, while for $`V_e`$ it contains no special functions at all.
Having these results we turned next to the most hard part of work: the exchange integral. Usually, to calculate it one has to use bispheroidal coordinates, and needs in an expansion of the potential in some orthogonal polynomials, such as Legendre polynomials, in bispheroidal coordinates. Here, only the exponential screened potential $`V_e`$ is known to have such an expansion but it is formulated, however, in terms of bispherical coordinates (the Gegenbauer expansion). Accordingly, we calculated exactly the exchange integral for $`V_e`$, at zero internuclear separation, $`R=0`$, at which case one can use bispherical coordinates. After that, we recovered partially the $`R`$ dependence using the standard result for the exchange integral for Coloumb potential (Sugiuraโs result). Thus, we achieved some approximate expression of the exchange integral for the case of $`V_e`$. So, all the subsequent results correspond to the $`V_e`$-based model.
Inserting obtained $`V_e`$-based Coloumb and exchange integrals into the total molecular energy expression, we get the final analytical expression containing four parameters, $`\gamma `$, $`\rho `$, $`A`$, and $`r_c`$. Prior to going into details of the energy minimization for the $`V_e`$-based (approximate) model, we analyze the set of parameters, and the conditions which we derived in the original Hulten potential case.
(1) From the analysis of Hulten potential, we see (Sec. 2.1) that the existence of a bound state of two electrons (which is proper isoelectronium) leads to the following relationship between the parameters for the case of one energy level of the electron-electron system: $`V_0=\mathrm{}^2/(2mr_c^2)`$. So, using the above mentioned relation $`A=V_0r_c`$ we have $`A=1/r_c2\gamma /\lambda `$, in atomic units ($`\mathrm{}=m_e=c=1`$). Thus note that, in this paper, we confined our consideration to the case of one-level isoelectronium.
(2) From the analysis of the Coloumb integral for Hulten potential, we see (Sec. 3.1.1) that the condition, $`\lambda ^1=`$ integer number, should hold, and one can use it as well.
We use the above two conditions, coming from the Hulten potential analysis, in the energy minimization calculations for the case of our $`V_e`$-based model. The first condition diminishes the number of independent parameters by one (they become three, $`\gamma `$, $`\rho `$, and $`\lambda `$) while the second condition means a discretization of the $`\lambda `$ parameter, $`\lambda ^1=4,5,6,\mathrm{}`$ Here, we used the condition $`\lambda ^1>3`$ which we obtained during the calculation of the Coloumb integral for $`V_e`$.
With the above set up, we minimized the total molecular energy of the $`V_e`$-based model. Numerical analysis shows that the $`\lambda `$ dependence does not reveal any minimum, in the interval of interest, $`4<\lambda ^1<60`$, while we have a minimum of the energy at some values of $`\gamma `$ and $`\rho `$. So, we calculated the energy minima for different values of $`\lambda `$, in the interval of interest, $`4<\lambda ^1<60`$. Results are presented in Tables 2 and 3. One can see that the binding energy decreases with the increase of the parameter $`r_c`$, which corresponds to the effective radius of the isoelectronium.
The following remarks are in order.
(i) Note that the discrete character of $`r_c`$ does not mean that the isoelectronium is some kind of a multilevel system, with different effective radia of isoelectronium assigned to the levels. We remind that we treat the isoelectronium as one-level system due to the above mentioned relation $`V_0=\mathrm{}^2/(2mr_c^2)`$. In fact, this means that there is a set of one-level isoelectronia of different fixed effective radia from which we should select only one, to fit the experimental data.
(ii) The use of the exponential screened potential $`V_e`$ can only be treated as some approximation to the original Hulten potential, and, thus, to the original Santilli-Shillady model of $`H_2`$ molecule. So, the numerical results obtained in Sec. 3.2 are valid only within this approximation. Hulten potential makes a difference (one can see this, e.g., by comparing Sec. 3.1.1 and Sec. 3.1.2), and it is worth to be investigated more closely by, for example, combination of analytical and numerical methods.
(iii) The results obtained in ref. are based on the Gaussian screened Coloumb potential $`V_g`$ approximation, to which the present work gives support in the form of exact analytical calculation of the Coloumb integral for $`V_g`$ (Sec. 3.1.3). Also, the present work gives possibility to make a comparative analysis of ref. , due to some similarity of the used potentials, $`V_e`$ and $`V_g`$.
(iv) Both the Coloumb integrals, for $`V_e`$ and $`V_g`$, reveal a minimum in respect with $`\lambda =2\gamma r_c`$, i.e. in respect with $`r_c`$ (see Figures 6 and 9) since minimization in Ritz parameter $`\gamma `$ is made independently. In principle, this gives us an opportunity to minimize the total molecular energy $`E_{mol}`$ with respect to $`r_c`$. However, there are two reasons that we can not provide this minimization. First, these minima correspond to rather large values of $`r_c`$, namely, $`r_c1\text{ a.u.}`$ for $`V_e`$ (Fig. 6), and $`r_c>2\text{ a.u.}`$ for $`V_g`$ (Fig. 9). Of course, this is not an obstacle to do minimization but we note that we generally assume that the effective radius of the isoelectronium $`r_c`$ is much less than the internuclear distance, $`r_cR=R_{exper}[H_2]=1.4011\text{ a.u.}`$ Second, and the main, reason is that for the exponential screened potential case (Sec. 3.1.2) the parameter $`\lambda `$ should be less than 1/3 to provide convergency of the associated Coloumb integral. Typically, $`\gamma 1.2`$, from which we obtain the condition $`r_c=\lambda /2\gamma <0.2\text{ a.u.}`$ Also, for the Hulten potential case (Sec. 3.1.1), we obtained $`\lambda <1/2`$, and hence $`r_c<0.25\text{ a.u.}`$ This means that, in fact, it is impossible to reach finite minimum of the total molecular energy $`E_{mol}`$ in respect with $`r_c`$ since the Coloumb integrals blow up, at $`r_c>0.25\text{ a.u.}`$, leading thus to infinite total energy $`E_{mol}`$. So, in our approach we arrive at a strict theoretical conclusion that the effective radius of the isoelectronium $`r_c`$ must be less than $`0.25\text{ a.u.}`$ Clearly, this supports our assumption that $`r_c`$ is much less than the internuclear distance $`R`$.
## 2 Santilli-Shillady model and the barrier
In this Section, we consider the general four-body Santilli-Shillady model of $`H_2`$ molecule, in Born-Oppenheimer approximation (i.e. at fixed nuclei). Shrรถdinger equation for $`H_2`$ molecule with the additional short range attractive Hulten potential between the electrons is of the following form:
$`({\displaystyle \frac{\mathrm{}^2}{2m_1}}_1^2{\displaystyle \frac{\mathrm{}^2}{2m_2}}_2^2V_0{\displaystyle \frac{e^{r_{12}/r_c}}{1e^{r_{12}/r_c}}}+{\displaystyle \frac{e^2}{r_{12}}}`$ (2.1)
$`{\displaystyle \frac{e^2}{r_{1a}}}{\displaystyle \frac{e^2}{r_{2a}}}{\displaystyle \frac{e^2}{r_{1b}}}{\displaystyle \frac{e^2}{r_{2b}}}+{\displaystyle \frac{e^2}{R}})|\psi =E|\psi ,`$
where $`R`$ is distance between the nuclei $`a`$ and $`b`$.
Interaction between the two electrons in the model is due to the potential
$$V(r_{12})=V_C(r_{12})+V_h(r_{12})=\frac{e^2}{r_{12}}V_0\frac{e^{r_{12}/r_c}}{1e^{r_{12}/r_c}},$$
(2.2)
where $`r_{12}`$ is distance between the electrons, $`V_0`$ and $`r_c`$ are real positive parameters. Here, first term, $`V_C`$, is usual repulsive Coloumb potential, and the second term, $`V_h`$, is an attractive Hulten potential.
Extrema of $`V(r_{12})`$ are defined by the equation
$$V^{}(r_{12})=\frac{e^2}{r_{12}^2}+\frac{V_0}{r_c}\frac{e^{r_{12}/r_c}}{(e^{r_{12}/r_c}1)^2}=0.$$
(2.3)
In the limit $`r_{12}\mathrm{}`$, potential $`V(r_{12})e^2/r_{12}=V_C(r_{12})`$. Series expansion of $`V(r_{12})`$ at $`r_{12}0`$ is
$$V(r_{12})|_{r_{12}0}=\frac{e^2V_0r_c}{r_{12}}+\frac{V_0}{2}\frac{V_0}{12r_c}r_{12}+O(r_{12}^3).$$
(2.4)
In general, there is relationship of Hulten potential to Bernoulli polynomials $`B_n(x)`$. Namely, Bernoulli polynomials are defined due to
$$\frac{se^{xs}}{e^s1}=\underset{n=0}{\overset{\mathrm{}}{}}B_n(x)\frac{s^n}{n!},$$
(2.5)
and we can reproduce Hulten potential,
$$\frac{e^s}{1e^s}=\frac{1}{s}\underset{n=0}{\overset{\mathrm{}}{}}B_n(1)\frac{s^n}{n!},$$
(2.6)
taking $`s=r_{12}/r_c`$. First five Bernoulli coefficients are
$$B_0(1)=1,B_1(1)=\frac{1}{2},B_2(1)=\frac{1}{6},B_3(1)=0,B_4(1)=\frac{1}{30}.$$
(2.7)
Eq.(2.6) means expansion of Hulten potential with the use of Bernoulli coefficients.
Eq.(2.4) implies that to have an attraction near $`r_{12}=0`$, which is necessary for forming of isoelectronium, we should put the condition
$$V_0r_c>e^2.$$
(2.8)
We note that, in view of the asymptotics (2.4), $`Q=\sqrt{V_0r_c}`$ can be thought of as Hulten charge of the electrons.
Under this condition, $`V(r_{12})`$ has one maximum at the point defined by Eq.(2.3). This is the equilibrium point at which the Coloumb potential is equal to the Hulten potential. So, we have barrier ($`B`$) separating two asymptotic regions, ($`A`$) $`r0`$ and ($`C`$) $`r\mathrm{}`$, with Coloumb-like attraction and Coloumb-like repulsion, respectively.
In the region $`A`$, attractive Hulten potential $`V_h`$ dominates, and therefore two electrons form bound state (isoelectronium), while in the region $`C`$ Coloumb repulsion $`V_C`$ dominates, and they are separated. This separation is limited by the size of the neutral molecule. For example, assuming that $`H_2`$ molecule is in the ground state we have $`rr_{mol}=3.46\text{ bohrs}`$, where we have assumed that separation between the protons is $`R=1.46\text{ bohrs}=0.77\AA `$.
Existence of the bound state of the electrons and of the barrier $`B`$ is a novel feature provided by the model. The asymptotic states, in regions $`A`$ and $`C`$, pertube each other due to the barrier effect in region $`B`$.
### 2.1 Region $`A`$
In the case
$$V_0r_ce^2$$
(2.9)
we can ignore Coloumb repulsion $`V_C`$, and region $`A`$ is a Hulten region, $`|V_h||V_C|`$; see Eq.(2.4). Then, exact solution of one-particle Schrรถdinger equation with Hulten potential $`V_h`$, where wave function has the boundary conditions $`\psi (0)=0`$ and $`\psi (\mathrm{})=0`$ (see , problem 68), can be used to establish the relation between the parameters $`V_0`$ and $`r_c`$, and to estimate $`r_c`$.
Energy spectrum for Hulten potential is given by
$$E_n=V_0\left(\frac{\beta ^2n^2}{2n\beta }\right)^2,n=1,2,\mathrm{}.$$
(2.10)
where
$$\beta ^2=\frac{2mV_0}{\mathrm{}^2}r_c^2,$$
(2.11)
and $`m`$ is mass of the particle. Assuming that there is only one energy level, namely, ground state $`n=1`$, we obtain the condition
$$\beta ^2=1,$$
(2.12)
which can be rewritten as
$$r_c=\mathrm{}\sqrt{\frac{1}{2mV_0}}.$$
(2.13)
Note that this state is characterized by approximately zero energy, $`E_1=0`$, due to Eq.(2.10); strictly speaking, $`\beta ^2`$ must be bigger but close to 1 in Eq.(2.12).
We should to note that the number of energy levels for Hulten potential is always finite due to Eq.(2.10). Assumption that there are more than one energy levels in the bound state of two electrons, i.e. that $`\beta >1`$, leads to drastical decrease of ground level energy $`E_1<0`$, and relatively small increase of characteristic size of isoelectronium in the ground state.
As the conclusion, the model implies โquantizationโ of the distance between two electrons, $`r=r_{12}`$, namely, forming of relatively small quasiparticle (isoelectronium) characterized by total mass $`M=2m_e`$, charge $`q=2e`$, spin zero, $`s=0`$, and small size in the ground one-level state. This quasiparticle, as a strongly correlated system of two electrons, moves in the potential of two protons of $`H_2`$ molecule, and one can apply methods developed for $`H_2^+`$ ion, with electron replaced by isoelectronium, to calculate approximate energy spectrum of $`H_2`$ . However, this quasiparticle is not stable, being a quasi-stationary state, due to finite height and width of the barrier $`B`$. So, we must take into account effects of both regions $`B`$ and $`C`$ to obtain correct energy spectrum of $`H_2`$ molecule, within the framework of the model.
### 2.2 Region $`B`$
Quasiclassically, due to smooth shape of the barrier, and because of exponential decrease of wave functions inside the barrier, electrons are not much time in region $`B`$, so we can ignore this transient phase in subsequent consideration.
We should to point out that the existence of the bound state in the region $`A`$ and repulsion in the region $`C`$ unavoidably leads to existence of the barrier.
### 2.3 Region $`C`$
In general, region $`C`$ is infinite, $`r_C<r<\mathrm{}`$, where $`r_C`$ is the distance between two electrons at which the Hulten potential is much smaller than the Coloumb potential, $`|V_h||V_C|`$.
In this region, electrons are not strongly correlated, in comparison to that in region $`A`$. Here, correlation is due to usual overlapping, Coloumb repulsion, exchange effects, and Coloumb attraction to protons. Shortly, we have the usual set up as it for the standard model of $`H_2`$ molecule.
Discarding, for a moment, effects coming from the consideration of regions $`A`$ and $`B`$, we have finite motion of the electrons in region $`C`$. Namely, in the ground state of $`H_2`$, the distance between electrons is confined by $`r=r_{mol}=3.46\text{ bohrs}`$. We restrict consideration by the ground state of $`H_2`$ molecule.
Due to this finiteness of the region $`C`$, $`r<r_{mol}`$, two electrons on the same orbit have constant probability to penetrate the barrier to form strongly correlated system, isoelectronium, and vice verca.
### 2.4 Model of decay of isoelectronium
Below, we assume that the isoelectronium undergoes decay, and the resulting two electrons are separated by sufficiently large distance, in the final state. This leads us to consideration of the effective life-time of isoelectronium. To estimate the order of the life-time, we use ordinary formula for radioactive $`\alpha `$-decay since the potential $`V(r)`$ is of the same shape, with very sharp decrease at $`r<r_{max}`$ and Coloumb repulsion at $`r>r_{max}`$. This quisiclassical model is a crude approximation because in fact the electrons do not leave the molecule. Moreover, we have the two asymptotic regimes simultaneously, with some distribution of probability, and it would be more justified here to say on frequency of the decay-formation process. However, due to our assumption of small size of isoelectronium, in comparison to the molecule size, we can study an elementary process of decay separately, and use the notion of life-time.
Decay constant is
$$\lambda =\frac{\mathrm{}D_0}{2mr_{max}^2}\mathrm{exp}\left\{\frac{4\pi Ze^2}{\mathrm{}}\sqrt{\frac{m}{2E}}+\frac{4e}{\mathrm{}}\sqrt{Zmr_{max}}\right\},$$
(2.14)
where we put, in atomic units,
$$\mathrm{}=1,e=1,m=1/2,Z=1,r_{max}=0.048,E=1.$$
(2.15)
Here, $`E=1\text{ a.u.}=27.212\text{ eV}`$ is double kinetic energy of the electron on first Bohrโs orbit, $`a_0=0.529\AA `$, that corresponds approximately to maximal relative kinetic energy of two electrons in ground state of $`H_2`$, and $`m=1/2`$ is reduced mass of two electrons.
We obtain the following numerical estimation for the life-time of isoelectronium:
$$1/\lambda =D_01.610^{17}\mathrm{sec},$$
(2.16)
i.e. it is of the order of 1 atomic unit of time, $`\tau =2.4210^{17}\mathrm{sec}`$. For lower values of the relative energy $`E`$, we obtain longer lifetimes; see Table 2.
The quasiclassical model for decay we are using here is the following. Particle of reduced mass $`m=1/2`$ penetrate the barrier $`B`$. This means a decay of isoelectronium. In the center of mass of electrons system, electrons undergo Coloumb repulsion and move in opposite directions receiving equal speed so that at large distances, $`rr_{max}`$, each of them have some kinetic energy. This energy can be given approximate upper estimation using linear velocity of electron on first Bohrโs orbit, $`v=2.1910^6`$ cm/sec, since electrons are in the ground level of $`H_2`$ molecule (this is the effect of the nuclei). This upper estimation corresponds to assumption of zero velocity of the center of mass in respect to protons which we adopt here. Kinetic energy of the particle of reduced mass is then double kinetic energy of electron, in center of mass system.
As the conclusion, in the framework of the model, $`H_2`$ molecule can be viewed as a mixed state of $`H_2^+`$ ion like system, i.e. strongly correlated phase (Hulten phase), when electrons form isoelectronium, and standard model of $`H_2`$, i.e. weakly correlated phase (Coloumb phase), when electrons are separated by large distance, $`r>r_{max}`$. Note that, as it has been mentioned above, we ignore the transient phase (inside the barrier) in this consideration. Evidently, the (statistical) weight of each phase depends on the characteristics of the potential $`V(r_{12})`$.
For extremally high barrier, only one of the phases could be realized with some energy spectra in each phase, namely, either spectrum of $`H_2^+`$ ion like system (with electron replaced by isoelectronium), or usual spectrum of $`H_2`$ molecule (without Hulten potential), respectively.
For high but finite barrier, each phase receives perturbation, and their (ground) energy levels split to two levels corresponding to simultaneous realization of both the phases. Note that the value $`V_{max}`$ is indeed high, $`V_{max}500\text{ eV}`$, under the given values of the parameters.
In general, existence of the strongly correlated phase (isoelectronium) leads to increase of the predicted dissociation energy, $`D`$, of $`H_2`$ molecule. Indeed, the mutual infuence of the regions $`A`$ and $`C`$ decreases the ground energy level $`E`$ of $`H_2`$ due to the above mentioned splitting. The general formula for $`D`$ is
$$D=2E_0(E+\frac{1}{2}\mathrm{}\omega ),$$
(2.17)
where $`2E_0=1`$ is total energy of two separated $`H`$ atoms, and $`\frac{1}{2}\mathrm{}\omega `$ is zero mode energy of the protons in $`H_2`$. So, decreasing of $`E<0`$ causes increase of $`D`$.
It is remarkable to note that experimental data give dissociation energy $`D_{exper}[H_2]=4.45\text{ eV}`$ for $`H_2`$ molecule (see, e.g. and references therein) while theoretical predictions within the standard model are $`D=2.90\text{ eV}`$ (Heitler-London), $`D=3.75\text{ eV}`$ (Flugge), and $`D=4.37\text{ eV}`$ (Hylleraas). We observe that improvement of the variational approximation gives better fits but still it gives lower values (about 2% lower) partially due to the fact that variational technique used there predicts generally bigger value (upper limit) for the ground energy.
Below, we use the same Ritz variational technique as it had been used by Heitler, London and Hylleraas but the feature of the model is the existence of additional attractive short range potential between the electrons suggested by Santilli and Shillady.
## 3 Variational solution for ground state energy of $`H_2`$ molecule
In the limiting case of large distances between the nuclei, $`R\mathrm{}`$, we have the total wave function of the electrons in the form
$$|\psi =f(r_{a1})f(r_{b2})\pm f(r_{b1})f(r_{a2}),$$
(3.1)
where the first term corresponds to the case when electron 1 is placed close to nucleus $`a`$ and $`f(r_{a1})`$ is wave function of the corresponding separate $`H`$ atom while the second term corresponds to the case when electron 1 is placed close to nucleus $`b`$. Symmetrized combination ($`{}_{}{}^{}+_{}^{}`$ sign) corresponds to antiparallel spins of the electrons 1 and 2, and, as the result of the usual analysis, leads to attraction between the $`H`$ atoms. Below, we use this symmetrized representation of the total wave function as the approximation to exact wave function.
### 3.1 Analytical calculations
By using Ritz variational approach and representation (3.1), we obtain from the Schrรถdinger equation (2.1) the energy of $`H_2`$ molecule in the following form (cf. ),
$$E_{mol}=2\frac{๐+๐^{}๐ฎ}{1+๐ฎ^2}\frac{2(๐+๐ฎ)(๐^{}+^{})}{1+๐ฎ^2}+\frac{1}{R},$$
(3.2)
where
$$๐ฎ=๐vf^{}(r_{a1})f(r_{b1})$$
(3.3)
is overlap integral,
$$๐=๐v\frac{1}{r_{b1}}|f(r_{a1})|^2,$$
(3.4)
$$๐^{}=๐v_1๐v_2\left(\frac{1}{r_{12}}V_0\frac{e^{r_{12}/r_c}}{1e^{r_{12}/r_c}}\right)|f(r_{a1})|^2|f(r_{b2})|^2,$$
(3.5)
are Coloumb integrals,
$$=๐v\frac{1}{r_{a1}}f^{}(r_{a1})f(r_{b1}),$$
(3.6)
$$^{}=๐v_1๐v_2\left(\frac{1}{r_{12}}V_0\frac{e^{r_{12}/r_c}}{1e^{r_{12}/r_c}}\right)f^{}(r_{a1})f(r_{b1})f^{}(r_{a2})f(r_{b2})$$
(3.7)
are exchange integrals,
$$๐=๐vf^{}(r_{a1})\left(\frac{1}{2}_1^2\frac{1}{r_{a1}}\right)f(r_{a1})$$
(3.8)
and
$$๐^{}=๐vf^{}(r_{a1})\left(\frac{1}{2}_1^2\frac{1}{r_{b1}}\right)f(r_{b1}).$$
(3.9)
We use atomic units, $`e=1`$, $`m_1=m_2=m_e=1`$.
Quite natural choice is that the wave functions in Eq.(3.1) are taken in the form of hydrogen ground state wave function,
$$f(r)=\sqrt{\frac{\gamma ^3}{\pi }}e^{\gamma r},$$
(3.10)
where $`\gamma `$ is Ritz variational parameter ($`\gamma `$=1 for the proper hydrogen wave function), and $`r=r_{a1},r_{b1},r_{a2},r_{b2}`$. With the help of $`\gamma `$ we should make better approximation to an exact wave function of the ground state. Namely, we should calculate all the integrals presented above analytically, and then vary the parameters $`\gamma `$ away from the value $`\gamma =1`$ and $`R`$ in some appropriate region, say $`1<R<2`$, to minimize the energy (3.2). As the energy minimum will be identified the found value of the parameter $`R`$ corresponds to optimal distance between the nuclei. This value should be compared to the experimental value of $`R`$.
All the molecular integrals (3.3)-(3.9), except for the Hulten potential parts in (3.5) and (3.7), are wellknown and can be calculated exactly; see, e.g. . Namely, they are
$$๐ฎ=\left(1+\rho +\frac{1}{3}\rho ^2\right)e^\rho ,$$
(3.11)
$$๐๐_C=\frac{\gamma }{\rho }(1(1+\rho )e^{2\rho }),$$
(3.12)
$$๐_C^{}๐_{|V_0=0}^{}=\frac{\gamma }{\rho }\left(1(1+\frac{11}{8}\rho +\frac{3}{4}\rho ^2+\frac{1}{6}\rho ^3)e^{2\rho }\right),$$
(3.13)
$$_C=\gamma (1+\rho )e^\rho ,$$
(3.14)
$$_C^{}_{|V_0=0}^{}=\gamma \left(\frac{5}{8}+\frac{23}{20}\rho \frac{3}{5}\rho ^2\frac{1}{15}\rho ^3\right)e^{2\rho }+\frac{6\gamma }{5}\frac{h(\rho )}{\rho },$$
(3.15)
$$h(\rho )=๐ฎ^2(\rho )(\mathrm{ln}\rho +C)๐ฎ^2(\rho )E_1(4\rho )+2๐ฎ(\rho )๐ฎ(\rho )E_1(2\rho ),$$
(3.16)
$$E_1(\rho )=\underset{\rho }{\overset{\mathrm{}}{}}\frac{e^t}{t}๐t,$$
(3.17)
$$๐=\frac{1}{2}\gamma ^2+\gamma (\gamma 1),๐^{}=\frac{1}{2}\gamma ^2๐ฎ+\gamma (\gamma 1),$$
(3.18)
where $`C`$ is Euler constant, and we have denoted
$$\rho =\gamma R,$$
(3.19)
which can be taken as a second Ritz variational parameter in addition to $`\gamma `$. The most hard part of work here is the exchange integral (3.15), which was calculated for the first time by Sugiura (1927), and contains one special function, the exponential integral function $`E_1(\rho )`$.
Our problem is thus to calculate analytically the Hulten potential parts of the Coloumb integral (3.5) and of the exchange integral (3.7), and then vary all the Ritz variational parameters in order to minimize the ground state energy (3.2),
$$E_{mol}(\text{parameters})=\text{minimum}.$$
(3.20)
In general, we have four parameters in our problem, $`E_{mol}=E_{mol}(\gamma ,\rho ,V_0,r_c)`$, with the first two parameters characterizing inverse radius of electronic orbit and the internuclear distance, respectively, and the last two parameters coming from the Hulten potential. However, assuming that the isoelectronium is characterized by one energy level, i.e. $`\beta =1`$, we have the relation (2.13) between $`V_0`$ and $`r_c`$ so that we are left with three independent parameters, say, $`E_{mol}=E_{mol}(\gamma ,\rho ,r_c)`$. In fact, we have three independent parameters for any fixed number $`\beta `$ of the levels due to the general relation (2.11),
$$V_0=\frac{\beta ^2\mathrm{}^2}{2mr_c^2},\beta =1,2,\mathrm{}.$$
(3.21)
Behavior of the energy $`E_{mol}`$ as a function of $`\gamma `$ and $`\rho `$ is more or less clear owing to known variational analysis of the standard model of $`H_2`$ molecule. Namely, $`E_{mol}`$ reveals a local minimum at some values of $`\gamma `$ and $`\rho `$. Thus, we should closely analyze the $`r_c`$ dependence of the energy which is specific to the Santilli-Shillady model of $`H_2`$ molecule.
Below, we turn to the Coloumb integral for the Hulten potential.
#### 3.1.1 Coloumb integral for Hulten potential
To calculate the Hulten part of the Coloumb integral (3.5) we use spherical coordinates, $`(r_{b2},\theta _2,\phi _2)`$, when integrating over second electron, and $`(r_{b1},\theta _1,\phi _1)`$, when integrating over first electron.
The integral is
$$๐_h^{}=4\pi ^2\underset{0}{\overset{\pi }{}}d\theta _1\underset{0}{\overset{\mathrm{}}{}}dr_{b1}\underset{0}{\overset{\pi }{}}d\theta _2\underset{0}{\overset{\mathrm{}}{}}dr_{b2}V_h(r_{12})\left(\frac{\gamma ^3}{\pi }e^{2\gamma r_{b2}}r_{b2}^{}{}_{}{}^{2}\mathrm{sin}\theta _2\right)\times $$
(3.22)
$$\times \left(\frac{\gamma ^3}{\pi }e^{2\gamma \sqrt{r_{b1}^{}{}_{}{}^{2}+R^2r_{b1}^{}{}_{}{}^{2}R\mathrm{cos}\theta _1}}r_{b1}^{}{}_{}{}^{2}\mathrm{sin}\theta _1\right),$$
where Hulten potential is
$$V_h(r_{12})=V_0\frac{e^{\sqrt{r_{b2}^{}{}_{}{}^{2}+r_{b1}^{}{}_{}{}^{2}2r_{b2}r_{b1}\mathrm{cos}\theta _2}/r_c}}{1e^{\sqrt{r_{b2}^{}{}_{}{}^{2}+r_{b1}^{}{}_{}{}^{2}2r_{b2}r_{b1}\mathrm{cos}\theta _2}/r_c}}.$$
(3.23)
Here, we have used
$$r_{a1}=\sqrt{r_{b1}^{}{}_{}{}^{2}+R^2r_{b1}^{}{}_{}{}^{2}R\mathrm{cos}\theta _1},$$
$$r_{12}=\sqrt{r_{b2}^{}{}_{}{}^{2}+r_{b1}^{}{}_{}{}^{2}2r_{b2}r_{b1}\mathrm{cos}\theta _2},$$
and the fact that integrals over azimuthal angles $`\phi _1`$ and $`\phi _2`$ give us $`4\pi ^2`$.
First, we integrate over coordinates of second electron,
$$I=2\pi \underset{0}{\overset{\pi }{}}๐\theta _2\underset{0}{\overset{\mathrm{}}{}}๐r_{b2}V_h(r_{12})\left(\frac{\gamma ^3}{\pi }e^{2\gamma r_{b2}}r_{b2}^{}{}_{}{}^{2}\mathrm{sin}\theta _2\right).$$
(3.24)
Integration over $`\theta _2`$ gives us
$$I=\underset{0}{\overset{\mathrm{}}{}}๐r_{b2}(I_1+I_2+I_3+I_4+I_5),$$
(3.25)
where
$$I_1=4\gamma ^3e^{2\gamma r_{b2}}r_{b2}^{}{}_{}{}^{2},$$
(3.26)
$$I_2=2\gamma ^3r_c\frac{r_{b2}}{r_{b1}}\sqrt{(r_{b1}r_{b2})^2}e^{2\gamma r_{b2}}\mathrm{ln}(1e^{\sqrt{(r_{b1}r_{b2})^2}/r_c}),$$
(3.27)
$$I_3=2\gamma ^3r_c\frac{r_{b2}}{r_{b1}}\sqrt{(r_{b1}+r_{b2})^2}e^{2\gamma r_{b2}}\mathrm{ln}(1e^{\sqrt{(r_{b1}+r_{b2})^2}/r_c}),$$
(3.28)
$$I_4=2\gamma ^3r_{c}^{}{}_{}{}^{2}\frac{r_{b2}}{r_{b1}}e^{2\gamma r_{b2}}Li_2(e^{\sqrt{(r_{b1}r_{b2})^2}/r_c}),$$
(3.29)
$$I_5=2\gamma ^3r_{c}^{}{}_{}{}^{2}\frac{r_{b2}}{r_{b1}}e^{2\gamma r_{b2}}Li_2(e^{\sqrt{(r_{b1}+r_{b2})^2}/r_c}),$$
(3.30)
and
$$Li_2(z)=\underset{k=1}{\overset{\mathrm{}}{}}\frac{z^k}{k^2}=\underset{z}{\overset{0}{}}\frac{\mathrm{ln}(1t)}{t}๐t$$
(3.31)
is dilogarithm function.
Now, we turn to integrating over $`r_{b2}`$. For $`I_1`$ we have
$$\underset{0}{\overset{\mathrm{}}{}}๐r_{b2}I_1=1.$$
(3.32)
In $`I_2`$, we should keep $`(r_{b1}r_{b2})`$ to be positive so we write down two separate terms,
$$\underset{0}{\overset{\mathrm{}}{}}๐r_{b2}I_2=I_{21}+I_{22}\underset{0}{\overset{r_{b1}}{}}๐r_{b2}I_2(r_{b2}<r_{b1})+\underset{r_{b1}}{\overset{\mathrm{}}{}}๐r_{b2}I_2(r_{b2}>r_{b1}).$$
(3.33)
In these two integrals, $`I_{21}`$ and $`I_{22}`$, we change variable $`r_{b2}`$ to $`x`$ and $`y`$, respectively,
$$x=(r_{b1}r_{b2})/r_c,r_{b1}/r_c<x<0,y=(r_{b2}r_{b1})/r_c,0<y<\mathrm{},$$
(3.34)
in order to simplify integrating. In terms of these variables, we have
$$I_{21}=\underset{r_{b1}/r_c}{\overset{0}{}}๐x2\gamma ^3r_{c}^{}{}_{}{}^{3}(x\frac{r_c}{r_{b1}}x^2)e^{2\gamma (r_{b1}r_cx)}\mathrm{ln}(1e^x),$$
(3.35)
$$I_{22}=\underset{0}{\overset{\mathrm{}}{}}๐y2\gamma ^3r_{c}^{}{}_{}{}^{3}(y+\frac{r_c}{r_{b1}}y^2)e^{2\gamma (r_{b1}+r_cy)}\mathrm{ln}(1e^y).$$
(3.36)
We are unable to perform these integrals directly. To calculate these integrals we use method of differentiating in parameter. Namely, we use simpler integrals,
$$L_1=๐xe^{2\gamma r_cx}\mathrm{ln}(1e^x)$$
(3.37)
and
$$L_2=๐ye^{2\gamma r_cy}\mathrm{ln}(1e^y),$$
(3.38)
and differentiate them in parameter $`r_c`$ to reproduce $`I_{21}`$ and $`I_{22}`$. (One can use parameter $`\gamma `$ for this purpose, or introduce an independent parameter putting it to one after making calculations, with the same result.) Namely, by using definitions of $`L_1`$ and $`L_2`$ we have
$$I_{21}=2\gamma ^3r_{c}^{}{}_{}{}^{3}(\frac{1}{2}\frac{d}{dr_c}L_1\frac{r_c}{4r_{b1}}\frac{d^2}{dr_{c}^{}{}_{}{}^{2}}L_1)|_{x=r_{b1}/r_c}^{x=0},$$
(3.39)
$$I_{22}=2\gamma ^3r_{c}^{}{}_{}{}^{3}(\frac{1}{2}\frac{d}{dr_c}L_2\frac{r_c}{4r_{b1}}\frac{d^2}{dr_{c}^{}{}_{}{}^{2}}L_2)|_{y=0}^{y=\mathrm{}}.$$
(3.40)
Now, the problem is to calculate indefinite integrals, $`L_1`$ and $`L_2`$, which make basis for further algebraic calculations. After making the calculations, we have
$$L_1=\frac{1}{4\gamma ^2r_{c}^{}{}_{}{}^{2}}e^{2\gamma r_c}\left(2\gamma r_c(\mathrm{\Phi }(e^x,1,2\gamma r_c)+\mathrm{ln}(1e^x))1\right)$$
(3.41)
and
$$L_2=\frac{1}{4\gamma ^2r_{c}^{}{}_{}{}^{2}}e^{2\gamma r_c}\left(2\gamma r_c(\mathrm{\Phi }(e^y,1,2\gamma r_c)+\mathrm{ln}(1e^y))+1\right),$$
(3.42)
where
$$\mathrm{\Phi }(z,s,a)=\underset{k=0}{\overset{\mathrm{}}{}}\frac{z^k}{(a+k)^s},a+k0,$$
(3.43)
is Lerch function, which is a generalization of polylogarithm function $`Li_n(z)`$ and Riemann $`\zeta `$-function. Particularly, $`Li_2(z)=\mathrm{\Phi }(z,2,0)`$. Also, we note that the Lerch function arises when dealing with Fermi-Dirac distribution, e.g.,
$$\underset{0}{\overset{\mathrm{}}{}}๐k\frac{k^s}{e^{k\mu }+1}=e^\mu \mathrm{\Gamma }(s+1)\mathrm{\Phi }(e^\mu ,s+1,1).$$
(3.44)
Below, we will need in derivatives of Lerch function $`\mathrm{\Phi }(z,s,a)`$ in third argument. By using the definition (3.43) we obtain directly
$$\frac{d}{da}\mathrm{\Phi }(z,s,a)\mathrm{\Phi }^{}(z,s,a)=s\mathrm{\Phi }(z,s+1,a),$$
(3.45)
$$\frac{d^2}{da^2}\mathrm{\Phi }(z,s,a)\mathrm{\Phi }^{\prime \prime }(z,s,a)=s(s+1)\mathrm{\Phi }(z,s+2,a).$$
(3.46)
Inserting (3.41) and (3.42) into (3.39) and (3.40) we get
$$I_{21}=\frac{1}{4\gamma r_{b1}}(e^{2\gamma (r_{b1}r_cx)}(3+2\gamma (r_{b1}(2+\gamma r_{b1})r_cx+\gamma r_{c}^{}{}_{}{}^{2}x^2)$$
(3.47)
$$2\gamma r_c((1+\gamma (r_{b1}2(1+\gamma r_{b1})r_cx+2\gamma r_{c}^{}{}_{}{}^{2}x^2))[\mathrm{\Phi }(e^x,1,2\gamma r_c)+\mathrm{ln}(1e^x)]+$$
$$+2\gamma r_c((1+\gamma (r_{b1}2r_cx))\mathrm{\Phi }^{}(e^x,1,2\gamma r_c)+\gamma r_c\mathrm{\Phi }^{\prime \prime }(e^x,1,2\gamma r_c))))|_{x=r_{b1}/r_c}^{x=0},$$
$$I_{22}=\frac{1}{4\gamma r_{b1}}(e^{2\gamma (r_{b1}+r_cx)}(3+2\gamma (r_{b1}+(2+\gamma r_{b1})r_cx+\gamma r_{c}^{}{}_{}{}^{2}x^2)+$$
(3.48)
$$+2\gamma r_c((1+\gamma (r_{b1}+2(1+\gamma r_{b1})r_cx+2\gamma r_{c}^{}{}_{}{}^{2}x^2))[\mathrm{\Phi }(e^x,1,2\gamma r_c)+\mathrm{ln}(1e^x)]+$$
$$+2\gamma r_c((1+\gamma (r_{b1}+2r_cx))\mathrm{\Phi }^{}(e^x,1,2\gamma r_c)+\gamma r_c\mathrm{\Phi }^{\prime \prime }(e^x,1,2\gamma r_c))))|_{y=0}^{y=\mathrm{}}.$$
Now, we have to use the above derivatives (3.45) and (3.46) of Lerch function to obtain final expressions for $`I_{21}`$ and $`I_{22}`$. Then, we should take the limits $`xr_{b1}/r_c`$, $`x0`$, and $`y0`$, $`y\mathrm{}`$, respectively. The endpoints $`x=r_{b1}/r_c`$ and $`y=\mathrm{}`$ can be inserted easily, with the endpoint $`y=\mathrm{}`$ yielding zero, while the limits $`x0`$ and $`y0`$ require some care because of the presence of some divergent terms.
To collect all the terms, we sum up $`I_{21}`$ and $`(1)I_{22}`$ given by (3.47) and (3.48), put $`x=y`$, and take common limit $`x0`$, inserting $`x=0`$ for polynomial and exponential (welldefined) terms. We get
$$I_{21}I_{22}|_{x0}=$$
$$=\frac{1}{2r_{b1}}(r_ce^{2\gamma r_{b1}}(2\gamma r_c(1+\gamma r_{b1})[\mathrm{\Phi }(e^x,2,2\gamma r_c)\mathrm{\Phi }(e^x,2,2\gamma r_c)]+$$
(3.49)
$$+4\gamma ^2r_{c}^{}{}_{}{}^{2}[\mathrm{\Phi }(e^x,3,2\gamma r_c)+\mathrm{\Phi }(e^x,3,2\gamma r_c)]+$$
$$+(1+\gamma r_{b1})[\mathrm{\Phi }(e^x,1,2\gamma r_c)+\mathrm{\Phi }(e^x,1,2\gamma r_c)2\mathrm{ln}(1e^x)])|_{x0}$$
The limits of Lerch functions of second, $`\mathrm{\Phi }(e^x,2,\pm 2\gamma r_c)`$, and third,
$`\mathrm{\Phi }(e^x,3,\pm 2\gamma r_c)`$, order, at $`x0`$, are welldefined while each of the terms in
$$B(2\gamma r_c)\underset{x0}{lim}[\mathrm{\Phi }(e^x,1,2\gamma r_c)+\mathrm{\Phi }(e^x,1,2\gamma r_c)2\mathrm{ln}(1e^x)]$$
(3.50)
is divergent since Lerch function of first order, $`\mathrm{\Phi }(e^x,1,\pm 2\gamma r_c)`$, increases unboundedly at $`x0`$. We will analyze this limit below, to identify the condition at which the divergencies cancel each other. Now, we collect all the terms obtaining final result for the integral in the form
$$\underset{0}{\overset{\mathrm{}}{}}dr_{b2}I_2=\frac{1}{4r_{b1}}(\frac{1}{\gamma }(3+2\gamma r_{b1}+8\gamma ^3r_{c}^{}{}_{}{}^{3}\mathrm{\Phi }(e^{r_{b1}/r_c},3,2\gamma r_c)+$$
(3.51)
$$+2\gamma r_c(1\gamma r_c)[\mathrm{\Phi }(e^{r_{b1}/r_c},1,2\gamma r_c)+2\gamma r_c\mathrm{\Phi }(e^{r_{b1}/r_c},2,2\gamma r_c)+\mathrm{ln}(1e^{r_{b1}/r_c})])$$
$$2r_ce^{2\gamma r_{b1}}(1+\gamma r_{b1})\{B(2\gamma r_c)+2\gamma r_c[\zeta (2,2\gamma r_c)\zeta (2,2\gamma r_c)]+$$
$$+4\gamma ^2r_{c}^{}{}_{}{}^{2}[\zeta (3,2\gamma r_c)+\zeta (3,2\gamma r_c)]\}),$$
where
$$\zeta (s,a)=\underset{k=1}{\overset{\mathrm{}}{}}\frac{1}{(a+k)^s},a+k0,$$
(3.52)
is generalized Riemann $`\zeta `$-function. The values of $`\zeta (2,\pm 2\gamma r_c)`$ and $`\zeta (3,\pm 2\gamma r_c)`$ entering (3.51) are welldefined. For example, at $`\gamma =1.4`$ and $`r_c=0.0048`$, we have
$$\zeta (2,\pm 2\gamma r_c)5537,\zeta (3,\pm 2\gamma r_c)2462.$$
(3.53)
Now, we turn to close consideration of the limit (3.50) entering (3.51). Let us calculate it for the particular value $`2\gamma r_c=1/100`$. Using expansion of each term of $`B`$ around $`x=0`$, we obtain
$$B(\frac{1}{100})=\underset{s1}{lim}[100\frac{1}{\mathrm{\Gamma }(\frac{1}{100})}\left\{100\mathrm{\Gamma }(\frac{101}{100})(C+\mathrm{ln}(1s)+\psi (\frac{1}{100}))\right\}$$
(3.54)
$$\frac{1}{99\mathrm{\Gamma }(\frac{99}{100})}\left\{100\mathrm{\Gamma }(\frac{199}{100})(C+\mathrm{ln}(1s)+\psi (\frac{99}{100}))\right\}+2\mathrm{ln}(1s)+O(1s)],$$
where we have denoted, for brevity, $`s=e^x`$,
$$\psi (z)=\underset{n=0}{\overset{\mathrm{}}{}}\frac{1}{z+n}=\frac{\mathrm{\Gamma }^{}(z)}{\mathrm{\Gamma }(z)}$$
(3.55)
is digamma function, $`\mathrm{\Gamma }(z)`$ is Euler gamma function, and $`C`$ is Euler constant. Using elementary properties of gamma function we obtain from Eq.(3.54)
$$B(\frac{1}{100})=1002C\psi (\frac{1}{100})\psi (\frac{99}{100}),$$
(3.56)
so one can see that the logarithmic divergent terms cancel each other, and the limit is welldefined for $`2\gamma r_c=1/100`$. The same is true for any integer value of
$$k=\frac{1}{2\gamma r_c}$$
(3.57)
while at noninteger $`k`$ the limit $`B(\frac{1}{k})`$ blows up. Generalizing the above particular result (3.56), we can write down
$$B(\frac{1}{k})=k2C\psi (\frac{1}{k})\psi (1\frac{1}{k}),$$
(3.58)
for any integer $`k>2`$.
This highly remarkable result means that to have finite value of the Coloumb integral we should use the condition that $`\lambda ^1(2\gamma r_c)^1=k`$ is an integer number. Recalling that typically $`\gamma 1.5`$ and $`r_c0.01`$ we have the integer number $`k30`$.
Now, we turn to the next integral, $`I_3`$. It is similar to $`I_2`$ so that we present the final expression,
$$\underset{0}{\overset{\mathrm{}}{}}dr_{b2}I_3=\frac{1}{4\gamma r_{b2}}(3+2\gamma r_{b1}+2\gamma r_c(1+\gamma r_{b1})[\mathrm{\Phi }(e^{r_{b1}/r_c},1,2\gamma r_c)+\mathrm{ln}(1e^{r_{b1}/r_c})]$$
(3.59)
$$4\gamma ^2r_{c}^{}{}_{}{}^{2}(1+\gamma r_{b1})\mathrm{\Phi }(e^{r_{b1}/r_c},2,2\gamma r_c)+8\gamma ^3r_{c}^{}{}_{}{}^{3}\mathrm{\Phi }(e^{r_{b1}/r_c},3,2\gamma r_c)).$$
The integral $`I_4`$ is more complicated,
$$\underset{0}{\overset{\mathrm{}}{}}๐r_{b2}I_4=I_{41}+I_{42},$$
(3.60)
where
$$I_{41}=\underset{0}{\overset{r_{b1}}{}}๐r_{b2}2\gamma ^3r_{c}^{}{}_{}{}^{2}\frac{r_{b2}}{r_{b1}}e^{2\gamma r_{b2}}Li_2(e^{(r_{b1}r_{b2})/r_c}),$$
(3.61)
$$I_{42}=\underset{r_{b1}}{\overset{\mathrm{}}{}}๐r_{b2}2\gamma ^3r_{c}^{}{}_{}{}^{2}\frac{r_{b2}}{r_{b1}}e^{2\gamma r_{b2}}Li_2(e^{(r_{b2}r_{b1})/r_c}).$$
(3.62)
Introducing variables
$$x=(r_{b1}r_{b2})/r_c,y=(r_{b2}r_{b1})/r_c,$$
(3.63)
we rewrite the integrals in the form
$$I_{41}=\underset{r_{b1}/r_c}{\overset{0}{}}๐x2\gamma ^3r_{c}^{}{}_{}{}^{3}e^{2\gamma (r_{b1}r_cx)}[Li_2(e^x)\frac{r_c}{r_{b1}}xLi_2(e^x)],$$
(3.64)
$$I_{42}=\underset{0}{\overset{\mathrm{}}{}}๐y2\gamma ^3r_{c}^{}{}_{}{}^{3}e^{2\gamma (r_{b1}+r_cy)}[Li_2(e^y)+\frac{r_c}{r_{b1}}yLi_2(e^y)].$$
(3.65)
In the r.h.s. of $`I_{41}`$, the first term can be calculated directly in terms of Lerch function while the second term can be obtained from the first term by differentiating it in the parameter, for which we choose again $`r_c`$. Namely, the basic integral, which we will use to calculate $`I_{41}`$, is
$$M_1=\underset{x_0}{\overset{0}{}}๐xe^{2\gamma r_cx}Li_2(e^x),$$
(3.66)
for which we have
$$M_1=\frac{1}{24\gamma ^3r_{c}^{}{}_{}{}^{3}\mathrm{\Gamma }(2\gamma r_c)}(3(e^{2\gamma r_cx_0}1)\mathrm{\Gamma }(2\gamma r_c)+$$
(3.67)
$$+\mathrm{\Gamma }(1+2\gamma r_c)(\gamma r_c\pi ^23C3\psi (2\gamma r_c))$$
$$3e^{2\gamma r_cx_0}\mathrm{\Gamma }(1+2\gamma r_c)(\mathrm{\Phi }(e^{x_0},1,2\gamma r_c)+\mathrm{ln}(1e^{x_0})+2\gamma r_cLi_2(e^{x_0}))).$$
We use this result in the first term of $`I_{41}`$. Differentiating $`M_1`$ given by Eqs. (3.66) and (3.67) in $`r_c`$, we reproduce the second term of $`I_{41}`$, up to a factor. So, collecting these results and inserting $`x_0=r_{b1}/r_c`$ we obtain after some algebra
$$I_{41}=\frac{1}{12r_{b1}\mathrm{\Gamma }(1+2\gamma r_c)}(e^{2\gamma r_c}[r_c(9+4\pi ^2\gamma ^3r_{c}^{}{}_{}{}^{2}r_{b1})\mathrm{\Gamma }(2\gamma r_c)+$$
(3.68)
$$\mathrm{\Gamma }(1+2\gamma r_c)(6Cr_c3r_{b1}6C\gamma r_cr_{b1}\pi ^2\gamma r_{c}^{}{}_{}{}^{2}6(\gamma r_{b1}1)r_c\psi (2\gamma r_c)$$
$$6\gamma r_{c}^{}{}_{}{}^{2}\psi ^{}(2\gamma r_c))]+3(2r_{b1}\mathrm{\Gamma }(1+2\gamma r_c)3r_c\mathrm{\Gamma }(2\gamma r_c)+$$
$$+2r_c\mathrm{\Gamma }(1+2\gamma r_c)[(12\gamma r_{b1})\mathrm{\Phi }(e^{r_{b1}/r_c},1,2\gamma r_c)+\gamma r_c\mathrm{\Phi }(e^{r_{b1}/r_c},2,2\gamma r_c)+$$
$$+(12\gamma r_{b1})\mathrm{ln}(1e^{r_{b1}/r_c})+\gamma (14\gamma r_{b1})r_c\psi ^{}(e^{r_{b1}/r_c})]),$$
where $`\psi ^{}(z)=d\psi (z)/dz`$ is derivative of digamma function.
To calculate $`I_{42}`$ we use a similar method. However, care should be exerted when taking limit $`y0`$. The basic integral, which we will use to calculate $`I_{42}`$, is
$$M_2=๐ye^{2\gamma r_cy}Li_2(e^y),$$
(3.69)
where we have replaced $`yy`$ so that the endpoints will be due to $`0<y<\mathrm{}`$. The result for $`M_2`$ is
$$M_2=\frac{1}{8\gamma ^3r_{c}^{}{}_{}{}^{3}}e^{2\gamma r_cy}(1+2\gamma r_ce^y\mathrm{\Phi }(e^y,1,1+2\gamma r_c)+2\gamma r_c\mathrm{ln}(1e^y)$$
(3.70)
$$4\gamma ^2r_{c}^{}{}_{}{}^{2}Li_2(e^y)).$$
We should insert here the endpoints $`y=0`$ and $`y=\mathrm{}`$. In the limit $`y\mathrm{}`$, $`M_2`$ is zero. In the limit $`y0`$, we have
$$Li_2(e^y)|_{y0}=\frac{\pi ^2}{6}$$
(3.71)
and, assuming that $`k=1/(2\gamma r_c)`$ is an integer number,
$$\mathrm{\Phi }(e^y,1,1+2\gamma r_c)+\mathrm{ln}(1e^y)|_{y0}=(\frac{1}{2\gamma r_c}+C+\psi (2\gamma r_c)).$$
(3.72)
Thus,
$$M_2|_{y=0}^{y=\mathrm{}}=\frac{1}{12\gamma ^2r_{c}^{}{}_{}{}^{2}}(3C+\pi ^2\gamma r_c+3\psi (2\gamma r_c)),$$
(3.73)
for integer $`k`$. We should point out that, in the case of noninteger $`k`$, $`M_2`$ increases unboundedly at $`y0`$. Using this result in $`I_{42}`$, we obtain
$$I_{42}=\frac{r_c}{12r_{b1}}e^{2\gamma r_{b1}}(6C(1+\gamma r_{b1})+\pi ^2\gamma r_c+2\pi ^2\gamma ^2r_cr_{b1}+$$
(3.74)
$$+6(1+\gamma r_{b1})\psi (2\gamma r_c)6\gamma r_c\psi ^{}(2\gamma r_c)).$$
Summing up $`I_{41}`$ given by (3.68) and $`I_{42}`$ given by (3.74), we get
$$\underset{0}{\overset{\mathrm{}}{}}dr_{b2}I_4==\frac{1}{4r_{b1}\mathrm{\Gamma }(1+2\gamma r_c)}(3r_c\mathrm{\Gamma }(2\gamma r_c)+e^{2\gamma r_c}[r_{b1}\mathrm{\Gamma }(1+2\gamma r_c)+$$
(3.75)
$$+r_c(3\mathrm{\Gamma }(2\gamma r_c)+4\mathrm{\Gamma }(1+2\gamma r_c)((1+\gamma r_{b1})(C+\psi (2\gamma r_c)+\gamma r_c\psi ^{}(2\gamma r_c)))]$$
$$2r_c\mathrm{\Gamma }(1+2\gamma r_c)[\mathrm{\Phi }(e^{r_{b1}/r_c},1,2\gamma r_c)+\mathrm{ln}(1e^{r_{b1}/r_c})+\gamma r_c(\mathrm{\Phi }(e^{r_{b1}/r_c},2,2\gamma r_c)+$$
$$+Li_2(e^{r_{b1}/r_c}))]).$$
The integral $`I_5`$ is similar to $`I_4`$ so that we present the final expression,
$$\underset{0}{\overset{\mathrm{}}{}}dr_{b2}I_5=\frac{1}{8\gamma r_{b1}}(3+4\gamma r_c[e^{r_{b1}/r_c}(\mathrm{\Phi }(e^{r_{b1}/r_c},1,1+2\gamma r_c)+$$
(3.76)
$$+\mathrm{\Phi }(e^{r_{b1}/r_c},2,1+2\gamma r_c))+\mathrm{ln}(1e^{r_{b1}/r_c})\gamma r_c\psi ^{}(e^{r_{b1}/r_c})]).$$
Now, we are in a position to sum up all the calculated integrals $`I_1,\mathrm{},I_5`$, and obtain, due to (3.25), the following final expression for the Coloumb integral over coordinates of second electron,
$$I(r_{b1})=(\frac{1}{2}+\frac{5}{8\gamma r_{b1}})e^{2\gamma r_{b1}}+$$
(3.77)
$$+\frac{1}{2}\gamma r_c[\pi (1+\frac{1}{\gamma r_{b1}})\text{ctg}(2\gamma r_c\pi )e^{2\gamma r_{b1}}\frac{1}{\gamma r_{b1}}e^{r_{b1}/r_c}\mathrm{\Phi }(e^{r_{b1}/r_c},1,1+2\gamma r_c)+$$
$$+\mathrm{\Phi }(e^{r_{b1}/r_c},1,2\gamma r_c)\mathrm{\Phi }(e^{r_{b1}/r_c},1,2\gamma r_c)+\frac{1}{\gamma r_{b1}}\mathrm{\Phi }(e^{r_{b1}/r_c},1,2\gamma r_c)]+$$
$$+\gamma ^2r_{c}^{}{}_{}{}^{2}[\frac{1}{2\gamma r_{b1}}e^{r_{b1}/r_c}\mathrm{\Phi }(e^{r_{b1}/r_c},2,2\gamma r_c)\mathrm{\Phi }(e^{r_{b1}/r_c},2,2\gamma r_c)$$
$$\mathrm{\Phi }(e^{r_{b1}/r_c},2,2\gamma r_c)+\frac{1}{\gamma r_{b1}}(\frac{1}{2}\mathrm{\Phi }(e^{r_{b1}/r_c},2,2\gamma r_c)\mathrm{\Phi }(e^{r_{b1}/r_c},2,2\gamma r_c))+$$
$$+\frac{1}{\gamma r_{b1}}e^{2\gamma r_{b1}}\psi ^{}(2\gamma r_c)+e^{2\gamma r_{b1}}(1+\frac{1}{\gamma r_{b1}})(\zeta (2,2\gamma r_c)\zeta (2,2\gamma r_c))]+$$
$$+\frac{2}{\gamma r_{b1}}\gamma ^3r_{c}^{}{}_{}{}^{3}[\mathrm{\Phi }(e^{r_{b1}/r_c},3,2\gamma r_c)+\mathrm{\Phi }(e^{r_{b1}/r_c},3,2\gamma r_c)$$
$$e^{2\gamma r_{b1}}(\zeta (3,2\gamma r_c)+\zeta (3,2\gamma r_c))],$$
where we have collected the terms due to power degrees of $`r_c`$. It should be stressed that here $`(2\gamma r_c)^1`$ is assumed to be an integer number. The above expression represents the Hulten part of the electrostatic potential caused by charge distribution of the second electron.
Next step is to integrate (3.77) over the coordinates of first electron,
$$๐_h^{}=2\pi \underset{0}{\overset{\pi }{}}๐\theta _1\underset{0}{\overset{\mathrm{}}{}}๐r_{b1}I(r_{b1})\frac{\gamma ^3}{\pi }e^{2\gamma \sqrt{r_{b1}^{}{}_{}{}^{2}+R^2r_{b1}^{}{}_{}{}^{2}R\mathrm{cos}\theta _1}}r_{b1}^{}{}_{}{}^{2}\mathrm{sin}\theta _1.$$
(3.78)
Prior to that, we denote
$$\lambda =2\gamma r_c=\frac{1}{k},r=\gamma r_{b1},$$
(3.79)
and rewrite Eq. (3.77) in a more compact form,
$$I(r)=(\frac{1}{2}+\frac{5}{8r})e^{2r}+\frac{1}{4}\lambda [\pi (1+\frac{1}{r})\text{ctg}(\pi \lambda )e^{2r}\frac{1}{r}e^{2r/\lambda }\mathrm{\Phi }(e^{2r/\lambda },1,1+\lambda )+$$
(3.80)
$$+\mathrm{\Phi }(e^{2r/\lambda },1,\lambda )\mathrm{\Phi }(e^{2r/\lambda },1,\lambda )+\frac{1}{r}\mathrm{\Phi }(e^{2r/\lambda },1,\lambda )]+$$
$$+\frac{\lambda ^2}{4}[\frac{1}{2r}e^{2r/\lambda }\mathrm{\Phi }(e^{2r/\lambda },2,\lambda )\mathrm{\Phi }(e^{2r/\lambda },2,\lambda )\mathrm{\Phi }(e^{2r/\lambda },2,\lambda )+$$
$$\frac{1}{r}(\frac{1}{2}\mathrm{\Phi }(e^{2r/\lambda },2,\lambda )\mathrm{\Phi }(e^{2r/\lambda },2,\lambda )+e^{2r}\psi ^{}(\lambda ))+e^{2r}(1+\frac{1}{r})(\zeta (2,\lambda )\zeta (2,\lambda ))]$$
$$+\frac{\lambda ^3}{4r}\left[\mathrm{\Phi }(e^{2r/\lambda },3,\lambda )+\mathrm{\Phi }(e^{2r/\lambda },3,\lambda )e^{2r}(\zeta (3,\lambda )+\zeta (3,\lambda ))\right].$$
Since $`I(r)`$ does not depend on $`\theta _1`$ one can easily integrate over $`\theta _1`$ in Eq.(3.78), and then change variable $`r_{b1}`$ to $`r=\gamma r_{b1}`$, obtaining
$$๐_h^{}=\frac{1}{2\rho }\underset{0}{\overset{\mathrm{}}{}}๐rI(r)\left[(1+2\sqrt{(\rho r)^2})re^{2\sqrt{(\rho r)^2}}(1+2(\rho +r))re^{2(\rho +r)}\right],$$
(3.81)
where $`\rho =\gamma R`$. Again, we should use separate intervals to keep $`(\rho r)`$ to be positive, namely, we rewrite $`๐_h^{}`$ as
$$๐_h^{}=J_1+J_2+J_3,$$
(3.82)
where
$$J_1=\frac{1}{2\rho }\underset{0}{\overset{\rho }{}}๐rI(r)(1+2\rho 2r)re^{2(\rho r)},$$
(3.83)
$$J_2=\frac{1}{2\rho }\underset{\rho }{\overset{\mathrm{}}{}}๐rI(r)(1+2r2\rho )re^{2(r\rho )},$$
(3.84)
$$J_3=\frac{1}{2\rho }\underset{0}{\overset{\mathrm{}}{}}๐rI(r)(1+2\rho +2r)re^{2(\rho +r)}.$$
(3.85)
Now, we are ready to make integration over the last remaining variable, $`r`$, to obtain complete analytical expression of the Coloumb integral for the Hulten potential.
However, Lerch functions entering Eq.(3.80) make obstacle to do integral (3.82) for a general case because they have different functional form for different values of the parameter $`\lambda `$. So, each of the above integrals $`J_{1,2,3}`$ should be calculated independently for every numerical value of $`\lambda `$. Moreover, for the values of interest, e.g., $`\lambda =1/30`$, each Lerch function is expressed in the form of sum of elementary functions with too big number of nontrivial terms to handle them (incomplete Euler beta function arises here). So, the integral cannot be reliably calculated even for a single value of $`\lambda `$, within the interval of interest, $`\lambda =1/30,1/31,\mathrm{},1/100`$. Also, elementary analysis shows that we can not implement the assumption of small $`r_c`$ into Eq.(3.80), to use first order approximation in $`r_c`$. Indeed, Lerch functions in (3.80) contain $`r_c`$ both in first and third argument so that their asymptotics at $`r_c0`$ make no sense.
Thus, we stop here further calculation of the Coloumb integral $`๐_h^{}`$ getting, however, as our main result the fact that $`(2\gamma r_c)^1`$ should be integer number, in the variational approach to the model, to have finite energy of the ground state. We consider this as very interesting result deserving rather involved calculations made above.
Also, we have a detailed technical view on the problems which arise when dealing with molecular integrals with the Hulten potential. Practically, this means that there is a very little hope that the exchange integral (3.7), which is structurally much more complicated than the above considered Coloumb one, can be calculated exactly for the case of Hulten potential.
Because of these difficulties, below we use appropriate simplified potentials, instead of Hulten potential, to have some analytical set up for the variational analysis of the Santilli-Shillady model. Clearly, by this we go to some approximation to the original Santilli-Shillady model.
#### 3.1.2 Coloumb integral for exponential screened Coloumb potential
We use simple function to mimic Hulten potential. Namely, we approximate the general potential (2.2) by
$$V(r_{12})=V_C+V_e=\frac{e^2}{r_{12}}\frac{Ae^{r_{12}/r_c}}{r_{12}},$$
(3.86)
where $`A`$ and $`r_c`$ are positive parameters. It has similar behavior both at short and long distances. Indeed, at long distances, $`r_{12}\mathrm{}`$, we can ignore $`V_e`$ and the behavior is solely due to the Coloumb potential while its series expansion about the point $`r_{12}=0`$ (short distances) is
$$V(r_{12})_{|r_{12}0}=\frac{e^2A}{r_{12}}+\frac{A}{r_c}\frac{A}{2r_c}r_{12}+O(r_{12}^2).$$
(3.87)
Here, we should put $`A=V_0r_c`$ to have the same coefficient at $`r_{12}^1`$ in the $`r_{12}0`$ asymptotics as it is in the case of Hulten potential; see Eq.(2.4). Using Eq.(3.21) we have
$$A=V_0r_c=\frac{\beta ^2\mathrm{}^2}{2mr_c},\beta =1,2,\mathrm{},$$
(3.88)
where $`\beta `$ is a number of energy levels of isoelectronium. Taking $`\beta =1`$ we have, in atomic units ($`\mathrm{}=1`$, $`m=m_e/2=1/2`$),
$$A=\frac{1}{r_c}.$$
(3.89)
Below, we calculate the Coloumb integral (3.5), with the exponential screened Coloumb potential $`V_e`$ defined by Eq.(3.86),
$$๐_E^{}=๐v_1๐v_2\left(\frac{e^2}{r_{12}}\frac{Ae^{r_{12}/r_c}}{r_{12}}\right)|f(r_{a1})|^2|f(r_{b2})|^2,$$
(3.90)
Below, we present some details of calculation of the Coloumb integral (3.90). Apart from the case of Hulten potential considered in Sec. 3.1.1, it appears that this integral can be calculated in terms of elementary functions.
The integral we are calculating is
$$๐_e^{}=๐v_1๐v_2\frac{Ae^{r_{12}/r_c}}{r_{12}}|f(r_{a1})|^2|f(r_{b2})|^2,$$
(3.91)
where
$$f(r)=\sqrt{\frac{\gamma ^3}{\pi }}e^{\gamma r},$$
(3.92)
and $`dv_1`$ and $`dv_2`$ are volume elements for the first and second electron, respectively. We use spherical coordinates. In spherical coordinates $`(r_{b2},\theta _2,\phi _2)`$, with polar axis directed along the vector $`\stackrel{}{r}_{b1}`$, we have
$$r_{12}=\sqrt{r_{b1}^{}{}_{}{}^{2}+r_{b2}^{}{}_{}{}^{2}2r_{b1}r_{b2}\mathrm{cos}\theta _2}.$$
(3.93)
We use these coordinates when integrating over second electron. In spherical coordinates $`(r_{b1},\theta _1,\phi _1)`$, with polar axis directed along the vector $`\stackrel{}{R}`$, we have
$$r_{a1}=\sqrt{r_{b1}^{}{}_{}{}^{2}+R^22r_{b1}R\mathrm{cos}\theta _1}.$$
(3.94)
We use these coordinates when integrating over first electron.
First, we integrate over angular coordinates of second electron,
$$I_1=\underset{0}{\overset{2\pi }{}}๐\phi _2\underset{0}{\overset{\pi }{}}๐\theta _2\frac{Ae^{r_{12}/r_c}}{r_{12}}\frac{\gamma ^3}{\pi }e^{2\gamma r_{b2}}r_{b2}^{}{}_{}{}^{2}\mathrm{sin}\theta _2,$$
(3.95)
where $`r_{12}`$ is defined by (3.93). It is relatively easy to calculate this integral,
$$I_1=\frac{2A\gamma ^3r_c}{r_{b1}}e^{2\gamma r_{b2}}\left(e^{\sqrt{(r_{b2}r_{b1})^2}/r_c}e^{\sqrt{(r_{b2}+r_{b1})^2}/r_c}\right).$$
(3.96)
Further, integrating on radial coordinate $`r_{b2}`$ must be performed in separate intervals,
$$I_2=\underset{0}{\overset{r_{b1}}{}}๐r_{b2}I_1(r_{b2}<r_{b1})+\underset{r_{b1}}{\overset{\mathrm{}}{}}๐r_{b2}I_1(r_{b2}>r_{b1}),$$
(3.97)
where
$$\sqrt{(r_{b1}r_{b2})^2}=\{\begin{array}{cc}r_{b1}r_{b2},& r_{b2}<r_{b1},\\ r_{b2}r_{b1},& r_{b2}>r_{b1},\end{array}$$
(3.98)
with the result
$$I_2=\frac{4A\gamma ^3r_{c}^{}{}_{}{}^{2}(4\gamma r_{c}^{}{}_{}{}^{2}(e^{r_{b1}/r_c}e^{2\gamma r_{b1}})+r_{b1}e^{2\gamma r_{b1}}(14\gamma ^2r_{c}^{}{}_{}{}^{2}))}{r_{b1}(14\gamma ^2r_{c}^{}{}_{}{}^{2})^2}.$$
(3.99)
Now, we turn to integrating over coordinates of first electron, $`(r_{b1},\theta _1,\phi _1)`$,
$$I_3=\underset{0}{\overset{\pi }{}}๐\theta _1\underset{0}{\overset{2\pi }{}}๐\phi _1I_2\frac{\gamma ^3}{\pi }e^{2\gamma r_{a1}}r_{b1}^{}{}_{}{}^{2}\mathrm{sin}\theta _1,$$
(3.100)
where $`r_{a1}`$ is defined by (3.94). We obtain after tedious calculations
$$I_3=\frac{2A\gamma ^4r_{c}^{}{}_{}{}^{2}}{R(14\gamma ^2r_{c}^{}{}_{}{}^{2})^2}e^{2\gamma (\sqrt{(Rr_{b1})^2}+r_{b1}+\sqrt{(R+r_{b1})^2})r_{b1}/r_c}$$
(3.101)
$$\times \left[e^{2\gamma \sqrt{(R+r_{b1})^2}}2\gamma e^{2\gamma \sqrt{(R+r_{b1})^2}}\sqrt{(Rr_{b1})^2}+2\gamma e^{2\gamma \sqrt{(Rr_{b1})^2}}\sqrt{(R+r_{b1})^2}\right]$$
$$\times \left[e^{r_{b1}/r_c}(4\gamma (1+\gamma r_{b1})r_{c}^{}{}_{}{}^{2}r_{b1})4\gamma r_{c}^{}{}_{}{}^{2}e^{2\gamma r_{b1}}\right].$$
Again, we must further integrate in $`r_{b1}`$ by separate intervals,
$$I_4=\underset{0}{\overset{R}{}}๐r_{b1}I_3(r_{b1}<R)+\underset{R}{\overset{\mathrm{}}{}}๐r_{b1}I_3(r_{b1}>R)I_{41}+I_{42},$$
(3.102)
obtaining after rather tedious calculations
$$I_{41}=\frac{Ae^{6\gamma R}\gamma ^2r_{c}^{}{}_{}{}^{2}}{12R(14\gamma ^2r_{c}^{}{}_{}{}^{2})^4}\times $$
(3.103)
$$[96e^{2\gamma RR/r_c}\gamma ^3((4\gamma (2\gamma Rr_c+R+r_c)+1)(12\gamma r_c)^2+e^{4\gamma R}(2\gamma r_c+1)^2(4\gamma r_c1))r_{c}^{}{}_{}{}^{3}$$
$$3e^{4\gamma R}+3\gamma (64\gamma ^5(\gamma R(8\gamma R+13)+4)r_{c}^{}{}_{}{}^{6}+16\gamma ^3(\gamma R(24\gamma R+31)+9)r_{c}^{}{}_{}{}^{4}$$
$$4\gamma (\gamma R(24\gamma R+23)+6)r_{c}^{}{}_{}{}^{2}+R(8\gamma R+5))e^{4\gamma R}\gamma \times $$
$$(64\gamma ^5(\gamma R(4\gamma R(2\gamma R+9)+57)+36)r_{c}^{}{}_{}{}^{6}48\gamma ^3(\gamma R(4\gamma R(2\gamma R+7)+21)9)r_{c}^{}{}_{}{}^{4}$$
$$+12\gamma (\gamma R(4\gamma R(2\gamma R+5)+1)6)r_{c}^{}{}_{}{}^{2}R(4\gamma R(2\gamma R+3)3)+3))+3],$$
$$I_{42}=\frac{Ae^{6\gamma R}\gamma ^2r_{c}^{}{}_{}{}^{2}}{4R(12\gamma r_c)^2(1+2\gamma r_c)^4}\times $$
(3.104)
$$\times [1e^{4\gamma R}128\gamma ^6R^2r_{c}^{}{}_{}{}^{5}+\gamma ((53e^{4\gamma R})R4(e^{4\gamma R}1)r_c)+$$
$$+16\gamma ^5Rr_{c}^{}{}_{}{}^{3}\left(8R+(16e^{2\gamma RR/r_c}+3e^{4\gamma R}13)r_c\right)+$$
$$+16e^{R/r_c}\gamma ^4r_{c}^{}{}_{}{}^{3}((8e^{2\gamma R}+3e^{4\gamma +R/r_c}13e^{R/r_c}R4(e^{4\gamma R}1)(2e^{2\gamma R}e^{R/r_c})r_c)+$$
$$+4\gamma ^2\left(2R^2(3e^{4\gamma R}5)Rr_c+3(e^{4\gamma R}1)r_{c}^{}{}_{}{}^{2}\right)+$$
$$+32\gamma ^3r_c(R^2Rr_c+e^{R/r_c}(e^{4\gamma R}1)(2e^{R/r_c}e^{2\gamma R})r_{c}^{}{}_{}{}^{2}))].$$
In calculating $`I_{42}`$, we put the condition
$$6\gamma r_c<1,$$
(3.105)
which is necessary to prevent divergency at the endpoint $`r_{b1}=\mathrm{}`$. Collecting the above two integrals we obtain
$$I_4๐_e^{}=\frac{A\gamma ^3r_{c}^{}{}_{}{}^{2}}{6R(14\gamma ^2r_{c}^{}{}_{}{}^{2})^4}[e^{2\gamma R}(R(3+2\gamma R(3+2\gamma R))$$
(3.106)
$$+12\gamma ^2Rr_{c}^{}{}_{}{}^{2}(5+2\gamma R(5+2\gamma R))48\gamma ^4Rr_{c}^{}{}_{}{}^{4}(15+2\gamma R(7+2\gamma R))$$
$$+64\gamma ^5r_{c}^{}{}_{}{}^{6}(24+\gamma R(33+2\gamma R(9+2\gamma R))))1536\gamma ^5r_{c}^{}{}_{}{}^{6}e^{R/r_c}].$$
Thus, we have finally for the Coloumb integral for exponential screened Coloumb potential,
$$๐_e^{}=\frac{A\lambda ^2}{8(1\lambda ^2)^4}\frac{\gamma e^{2\rho }}{\rho }[(\rho +2\rho ^2+\frac{4}{3}\rho ^3)+3\lambda ^2(5\rho +10\rho ^2+4\rho ^3)$$
(3.107)
$$\lambda ^4(15\rho +14\rho ^2+4\rho ^3)+\lambda ^6(8+11\rho +6\rho ^2+\frac{4}{3}\rho ^38e^{2\rho \frac{2\rho }{\lambda }})].$$
Here, we have used notation $`\lambda =2\gamma r_c`$, and also $`\lambda <1/3`$ due to Eq.(3.105).
The total Coloumb integral is
$$๐_E^{}=๐_C^{}๐_e^{},$$
(3.108)
where $`๐_C^{}`$ is wellknown Coloumb potential part given by Eq.(3.13).
Below, we turn to the other potential, Gaussian screened Coloumb potential, considered by Santilli and Shillady . The Coloumb integral for this potential can be calculated exactly, and the result contains one special function, the error function $`\text{erf}(z)`$.
#### 3.1.3 Coloumb integral for Gaussian screened Coloumb potential
In this Section, we calculate the Coloumb integral for the case of Gaussian screened potential. Namely, we approximate the general potential (2.2) by
$$V(r_{12})=V_C+V_g=\frac{e^2}{r_{12}}\frac{Ae^{r_{12}^2/c}}{r_{12}},$$
(3.109)
where $`A`$ and $`c=r_{c}^{}{}_{}{}^{2}`$ are positive parameters. At long distances, $`r_{12}\mathrm{}`$, we can ignore $`V_g`$ while its series expansion about the point $`r_{12}=0`$ is
$$V(r_{12})_{|r_{12}0}=\frac{e^2A}{r_{12}}+\frac{A}{c}r_{12}+O(r_{12}^2).$$
(3.110)
Here, we should put $`A=V_0r_c`$ to have the same coefficient at $`r_{12}^1`$ in the $`r_{12}0`$ asymptotics as it is in the case of Hulten potential; see Eq.(2.4).
The Coloumb integral is
$$๐_G^{}=๐v_1๐v_2\left(\frac{e^2}{r_{12}}\frac{Ae^{r_{12}^2/c}}{r_{12}}\right)|f(r_{a1})|^2|f(r_{b2})|^2.$$
(3.111)
The integral we are calculating is
$$๐_g^{}=๐v_1๐v_2\frac{Ae^{r_{12}^2/c}}{r_{12}}|f(r_{a1})|^2|f(r_{b2})|^2,$$
(3.112)
where notation and coordinate system are due to Sec. 3.1.2. First, we integrate over angular coordinates of second electron,
$$I_1=\underset{0}{\overset{2\pi }{}}๐\phi _2\underset{0}{\overset{\pi }{}}๐\theta _2\frac{Ae^{r_{12}^2/c}}{r_{12}}\frac{\gamma ^3}{\pi }e^{2\gamma r_{b2}}r_{b2}^{}{}_{}{}^{2}\mathrm{sin}\theta _2,$$
(3.113)
where $`r_{12}`$ is defined by (3.93). We have
$$I_1=\frac{A\gamma ^3\sqrt{\pi c}e^{2\gamma r_{b2}}}{r_{b1}}\left(\text{erf}(\sqrt{\frac{(r_{b1}+r_{b2})^2}{c}})\text{erf}(\sqrt{\frac{(r_{b1}r_{b2})^2}{c}})\right),$$
(3.114)
where
$$\text{erf}(z)=\frac{2}{\sqrt{\pi }}\underset{0}{\overset{z}{}}e^{t^2}๐t$$
(3.115)
is error function. Further, integrating on radial coordinate $`r_{b2}`$ must be performed in separate intervals,
$$I_2=\underset{0}{\overset{r_{b1}}{}}๐r_{b2}I_1(r_{b2}<r_{b1})+\underset{r_{b1}}{\overset{\mathrm{}}{}}๐r_{b2}I_1(r_{b2}>r_{b1}),$$
(3.116)
where
$$\sqrt{(r_{b1}r_{b2})^2}=\{\begin{array}{cc}r_{b1}r_{b2},& r_{b2}<r_{b1},\\ r_{b2}r_{b1},& r_{b2}>r_{b1},\end{array}$$
(3.117)
with the result
$$I_2=\frac{A\gamma \sqrt{c}e^{2\gamma r_{b1}r_{b1}^{}{}_{}{}^{2}/c}}{4r_{b1}}(4\gamma \sqrt{c}(e^{r_{b1}^{}{}_{}{}^{2}/c}e^{2\gamma r_{b1})})$$
(3.118)
$$+\sqrt{\pi }e^{r_{b1}^{}{}_{}{}^{2}/c+c\gamma ^2}[(1+2\gamma (r_{b1}c\gamma ))(\text{erfc}(\frac{r_{b1}c\gamma }{\sqrt{c}})+2\text{erfc}(\sqrt{c}\gamma )2)$$
$$+e^{4\gamma r_{b1}}(2\gamma (r_{b1}+c\gamma )1)\text{erfc}(\frac{r_{b1}+c\gamma }{\sqrt{c}})]),$$
where $`\text{erfc}(z)=1\text{erf}(z)`$. Now, we turn to integrating over coordinates of first electron, $`(r_{b1},\theta _1,\phi _1)`$,
$$I_3=\underset{0}{\overset{\pi }{}}๐\theta _1\underset{0}{\overset{2\pi }{}}๐\phi _1I_2\frac{\gamma ^3}{\pi }e^{2\gamma r_{a1}}r_{b1}^{}{}_{}{}^{2}\mathrm{sin}\theta _1,$$
(3.119)
where $`r_{a1}`$ is defined by (3.94). We obtain after tedious calculations
$$I_3=\frac{A\sqrt{c}\gamma ^2}{8R}e^{\frac{r_{b1}^{}{}_{}{}^{2}}{c}2\gamma (\sqrt{(Rr_{b1})^2}+2r_{b1}+\sqrt{(R+r_{b1})^2})}$$
(3.120)
$$\times (e^{2\gamma (\sqrt{(Rr_{b1})^2}+r_{b1})}e^{2\gamma (\sqrt{(R+r_{b1})^2}+r_{b1})}2\gamma e^{2\gamma (\sqrt{(R+r_{b1})^2}+r_{b1})}\sqrt{(Rr_{b1})^2}$$
$$+2\gamma e^{2\gamma (\sqrt{(Rr_{b1})^2}+r_{b1})}\sqrt{(R+r_{b1})^2})$$
$$\times [\sqrt{\pi }e^{\frac{r_{b1}^{}{}_{}{}^{2}}{c}+c\gamma ^2}((1+2\gamma r_{b1}2c\gamma ^2)(2\text{erfc}(\gamma \sqrt{c})+\text{erfc}(\frac{r_{b1}\gamma c}{\sqrt{c}})2)$$
$$+(1+2\gamma r_{b1}+2c\gamma ^2)e^{4\gamma r_{b1}}\text{erfc}(\frac{r_{b1}+\gamma c}{\sqrt{c}}))].$$
Again, we must further integrate in $`r_{b1}`$ by separate intervals,
$$I_4=\underset{0}{\overset{R}{}}๐r_{b1}I_3(r_{b1}<R)+\underset{R}{\overset{\mathrm{}}{}}๐r_{b1}I_3(r_{b1}>R).$$
(3.121)
First, we replace the endpoint $`r_{b1}=\mathrm{}`$ by finite value $`r_{b1}=\mathrm{\Lambda }`$ to avoid divergencies at intermediate calculations. After straightforward but tedious calculations we obtain rather long expression so that we do not represent it here noting however that the following integrals are used during the calculations:
$$\text{erf}(z)๐z=\frac{e^{z^2}}{\sqrt{\pi }}+z\text{erf}(z),$$
(3.122)
$$z\text{erf}(z)๐z=\frac{ze^{z^2}}{2\sqrt{\pi }}\frac{1}{4}\text{erf}(z)+\frac{1}{2}z^2\text{erf}(z),$$
(3.123)
$$e^{az}\text{erf}(z)๐z=\frac{1}{a}e^{az}\text{erf}(z)+\frac{1}{a}e^{a^2/4}\text{erf}(\frac{a}{2}+z),$$
(3.124)
$$ze^{az}\text{erf}(z)๐z=\frac{1}{a\sqrt{\pi }}e^{azz^2}\frac{1}{a^2}e^{az}(1+az)\text{erf}(z)$$
(3.125)
$$\frac{1}{2a^2}(a^21)e^{a^2/4}\text{erf}(\frac{a}{2}+z),$$
$$e^{azbz^2}๐z=\frac{\sqrt{\pi }}{2\sqrt{b}}e^{a^2/(4b)}\text{erf}(\frac{a+2bz}{2\sqrt{b}}),$$
(3.126)
$$ze^{azbz^2}๐z=\frac{1}{2b}e^{azbz^2}\frac{a\sqrt{\pi }}{4b^{3/2}}e^{a^2/(4b)}\text{erf}(\frac{a+2bz}{2\sqrt{b}}).$$
(3.127)
Using $`lim_\mathrm{\Lambda }\mathrm{}\text{erf}(\mathrm{\Lambda })=1`$ and replacing welldefined exponentially decreasing terms by zero, we obtain some finite terms and big number (about fourty) of $`\mathrm{\Lambda }`$ dependent terms, which are unbounded at $`\mathrm{\Lambda }\mathrm{}`$. All the divergent terms totally cancel each other so the final expression turns out to be automatically finite.
As the result, we obtain the Coloumb integral for Gaussian screened Coloumb potential in the following form:
$$๐_g^{}=\frac{A\gamma \kappa e^{2\rho }}{96\rho }[(60+96\rho +48\rho ^2)\kappa +(32+48\rho )\kappa ^316\kappa ^5$$
(3.128)
$$+\left((60+16\rho ^2)\kappa 32\kappa ^3+16\kappa ^5\right)e^{2\rho \frac{\rho ^2}{\kappa ^2}}$$
$$+\sqrt{\pi }e^{\kappa ^2}((30\rho +8\rho ^336\rho \kappa ^2+24\rho \kappa ^4)(2\text{erf}(\kappa )\text{erfc}(\frac{\rho }{\kappa }\kappa )e^{4\rho }\text{erfc}(\frac{\rho }{\kappa }+\kappa ))$$
$$+(15+24\rho ^2(18+24\rho ^2)\kappa ^2+12\kappa ^48\kappa ^6)(2\text{erf}(\kappa )\text{erfc}(\frac{\rho }{\kappa }\kappa )+e^{4\rho }\text{erfc}(\frac{\rho }{\kappa }+\kappa )))],$$
where we have used notation
$$\kappa =\gamma \sqrt{c}=\gamma r_c=\frac{\lambda }{2}.$$
(3.129)
The total Coloumb integral is
$$๐_G^{}=๐_C^{}๐_g^{},$$
(3.130)
where $`๐_C^{}`$ is given by Eq.(3.13).
#### 3.1.4 Exchange integral
Our general remark is that all calculations for the above Coloumb integrals are made in spherical coordinates, which correspond to spherical symmetry of the charge distributions of both $`1s`$ electrons, $`|\psi (r_{a1})|^2`$ and $`|\psi (r_{a2})|^2`$, each moving around one nucleus. One can use prolate spheroidal coordinates, which are exploited sometimes when integrating over coordinates of last electron, but we have encountered the same problem of big number of terms in the intermediate expressions, with no advantage in comparison to the use of spherical coordinates.
Unlike to Coloumb integral, calculation of exchange integral should be made in the spheroidal coordinates, which correspond to spheroidal symmetry of charge distributions of the electrons, $`\psi ^{}(r_{a1})\psi (r_{b1})`$ and $`\psi ^{}(r_{a2})\psi (r_{b2})`$, each moving around both the nuclei, $`a`$ and $`b`$.
Calculation of the exchange integral,
$$^{}=๐v_1๐v_2V(r_{12})f^{}(r_{a1})f(r_{b1})f^{}(r_{a2})f(r_{b2}),$$
(3.131)
essentially depends on the form of the potential $`V(r_{12})`$ in the sense that the integration can be made only in spheroidal coordinates, $`(x_1,y_1,\phi _1)`$ and $`(x_2,y_2,\phi _2)`$, and one should use an expansion of $`V(r_{12})`$ in the associated Legendre polynomials.
For the usual Coloumb potential, $`V(r_{12})=r_{12}^1`$, it is rather long (about 12 pages to present the main details) and nontrivial calculation, where Neumann expansion in terms of associated Legendre polynomials, in spheroidal coordinates, is used (celebrated result by Sugiura, see Eq.(3.15)).
In general, any analytical square integrable function can be expanded in associated Legendre polynomials. However, in direct calculating of the expansion coefficients by means of integral of the function with Legendre polynomials, one meets serious problems even for simple functions. Practically, one uses, instead, properties of special functions to derive such expansions.
We mention that there is Gegenbauer expansion , having in a particular case the form
$$\frac{e^{ikr_{12}}}{r_{12}}=\frac{1}{r_1r_2}\underset{l=0}{\overset{\mathrm{}}{}}\sqrt{\frac{2l+1}{4\pi }}\frac{i}{k}j_l(kr_1)n_l^{(1)}(kr_2)Y_{l,0}(\theta _{12}),$$
(3.132)
where $`j_l(z)`$ and $`n_l^{(1)}(z)`$ are spherical Bessel and spherical Hankel functions of first kind, respectively, $`\theta _{12}`$ is angle between vectors $`\stackrel{}{r}_1`$ and $`\stackrel{}{r}_2`$, and $`r_1=|\stackrel{}{r}_1|`$, $`r_2=|\stackrel{}{r}_2|`$; $`r_1<r_2`$. Spherical harmonics $`Y_{l,0}(\theta _{12})`$ can be rewritten in terms of Legendre polynomials due to the summation theorem.
We note that this expansion can be used, at $`k=i/r_c`$, to reproduce exponential screened potential, $`V_e(r_{12})`$, and to calculate associated exchange integral (3.131) but, however, it concerns spherical (not spheroidal) coordinates, $`(r_1,\theta _1,\phi _1)`$ and $`(r_2,\theta _2,\phi _2)`$.
For Hulten potential $`V_h(r_{12})`$, exponential screened potential, $`V_e(r_{12})`$, and Gaussian screened potential, $`V_g(r_{12})`$, which are of interest in this paper, we have no such an expansion in spheroidal coordinates. To stress that this is not only the problem of changing coordinate system, we mention that the solution of usual 3-dimensional wave equation, $`\mathrm{\Delta }\psi +k^2\psi =0`$, is given by function $`e^{i\stackrel{}{k}\stackrel{}{r}}/r`$, in spherical coordinates, to which one can apply Gegenbauer expansion, while in spheroidal coordinates its solution is represented by complicated function containing infinite series of recurrent coefficients ; see also , Sec. 3.4. As the result, we have no possibility to calculate exactly exchange integrals for these non-Coloumb potentials.
In order to obtain approximate expression for the exchange integral for the case of the above non-Coloumb potentials, we make analysis of asymptotics of the standard exchange integral (i.e. that for the Coloumb potential), Eq.(3.15). It is easy to derive that
$$_{C}^{}{}_{|\rho \mathrm{}}{}^{}e^{2\rho },$$
(3.133)
at long distances between the nuclei, and
$$_{C}^{}{}_{|\rho =0}{}^{}=\frac{5}{8}\gamma ,$$
(3.134)
in the case of coinciding nuclei. At $`r_c^10`$, we should have the same asymptotics for exchange integral for each of the above non-Coloumb potentials because these potentials behave as Coloumb potential at $`r_c^10`$.
In both the limiting cases, $`\rho \mathrm{}`$ and $`\rho =0`$, the exchange integral for the non-Coloumb potentials is simplified, and one can use spherical coordinates since the two-center problem is reduced to one-center problem. We consider two limiting cases.
a) $`\rho =\mathrm{}`$ case.
This case is trivial because exchange integral tends to zero due to lack of overlapping of the wave functions of two $`H`$ atoms.
b) $`\rho =0`$ case.
In this case , we have $`r_{a1}=r_{b1}=r_1`$ and $`r_{a2}=r_{b2}=r_2`$ so that Eq.(3.131) becomes
$$^{}=๐v_1๐v_2V(r_{12})|f(r_1)|^2|f(r_2)|^2,$$
(3.135)
One can see that this is the case of $`He`$ atom with two electrons in the ground state. Evidently, in terms of our anzatz (3.1) we have complete overlapping of the wave functions.
Even the above mentioned simplification of the exchange integral and use of spherical coordinates does not enable us to calculate straightforwardly the integral (3.135) for the non-Coloumb potentials, $`V_h`$, $`V_e`$, or $`V_g`$; the integrands are still too complicated. This indicates that we should use expansion of these potentials in Legendre polynomials, in spherical coordinates, to perform the integrals. Only exponential screened potential $`V_e`$ is given such an expansion here. Namely, this is Gegenbauer expansion (3.132), owing to which we can calculate the exchange potential for exponential screened potential $`V_e`$, to which we turn below.
Exchange integral for the exponential screened Coloumb potential $`V_e`$, at $`\rho =0`$.
The integral is
$$_{E}^{}{}_{|\rho =0}{}^{}(_C^{}_e^{})_{|\rho =0}=\frac{5}{8}\gamma ๐v_1๐v_2\frac{Ae^{r_{12}/r_c}}{r_{12}}|f(r_1)|^2|f(r_2)|^2,$$
(3.136)
where we have used Eq.(3.134) for the usual Coloumb potential part of the integral. In the Gegenbauer expansion (3.132), we assume $`k=i/r_c`$ to reproduce the potential $`V_e(r_{12})`$. Since the wave functions $`f(r_1)`$ and $`f(r_2)`$ given by Eq.(3.10) do not depend on the angles, only $`l=0`$, $`m=0`$ term of the expansion (3.132) contributes to the exchange integral (3.136) due to orthogonality of Legendre polynomials. Using
$$j_0(z)=\mathrm{sin}z,n_0(z)=ie^{iz},Y_{0,0}=\sqrt{\frac{1}{4\pi }},$$
(3.137)
we thus have
$$\frac{Ae^{ikr_{12}}}{r_{12}}\{\begin{array}{cc}\frac{A}{kr_1r_2}\mathrm{sin}kr_1e^{ikr_2},& r_1<r_2,\\ \frac{A}{kr_1r_2}\mathrm{sin}kr_2e^{ikr_1},& r_1>r_2,\end{array}$$
(3.138)
Then the exchange integral (3.136) is written as
$$_{E}^{}{}_{|\rho =0}{}^{}=\frac{5}{8}\gamma \underset{0}{\overset{\mathrm{}}{}}4\pi r_2^2dr_2[\underset{0}{\overset{r_2}{}}4\pi r_1^2dr_1\frac{A}{kr_1r_2}\mathrm{sin}kr_1e^{ikr_2}\frac{\gamma ^3}{\pi }e^{2\gamma r_1}\frac{\gamma ^3}{\pi }e^{2\gamma r_2}$$
(3.139)
$$+\underset{r_2}{\overset{\mathrm{}}{}}4\pi r_1^2dr_1\frac{A}{kr_1r_2}\mathrm{sin}kr_2e^{ikr_1}\frac{\gamma ^3}{\pi }e^{2\gamma r_1}\frac{\gamma ^3}{\pi }e^{2\gamma r_2}],$$
where $`4\pi r_1^2`$ and $`4\pi r_2^2`$ are volume factors. The two above integrals over $`r_1`$ can be easily calculated, with the result
$$\frac{16A\gamma ^6r_2}{(k^2+4\gamma ^2)^2}[4\gamma e^{i(k+2i\gamma )r_2}+\frac{1}{k}e^{i(k+4i\gamma )}((k^24\gamma ^2)\mathrm{sin}kr_24k\gamma \mathrm{cos}kr_2$$
(3.140)
$$(k^2+4\gamma ^2)(k\mathrm{cos}kr_2+2\gamma \mathrm{sin}kr_2))]$$
and
$$\frac{16A\gamma ^6r_2}{k(k+2i\gamma )^2}\left(1+(2\gamma ik)\mathrm{sin}kr_2e^{i(k+4i\gamma )}\right).$$
(3.141)
Summing up these terms and integrating over $`r_2`$ we get after some algebra
$$_{E}^{}{}_{|\rho =0}{}^{}=\frac{5}{8}\gamma +\frac{A\gamma ^3}{2(k+2i\gamma )^4}(k^2+8ik\gamma 20\gamma ^2).$$
(3.142)
Inserting
$$k=\frac{i}{r_c},$$
(3.143)
to reproduce the potential $`V_e`$, and denoting $`\lambda =2\gamma r_c`$ we write down our final result,
$$_{E}^{}{}_{|\rho =0}{}^{}=\frac{5}{8}\gamma \frac{\gamma A\lambda ^2}{8(1+\lambda )^4}(1+4\lambda +5\lambda ^2).$$
(3.144)
Note that, at $`r_c^10`$, i.e. at $`\lambda \mathrm{}`$, we have
$$_{E}^{}{}_{|\rho =0}{}^{}=\frac{5}{8}\gamma \frac{5}{8}A\gamma $$
(3.145)
that is in agreement with the value (3.134). We should to emphasize here that Eq.(3.144) is exact result for the exchange integral $`_E^{}`$, at $`\rho =0`$.
Next step is to implement $`\rho `$ dependence into (3.144) following to natural criteria. To restore partially $`\rho `$ dependence in the exchange integral (3.144), we use exact result (3.15), and write down for the $`\rho `$ dependent exchange integral the following approximate expression:
$$_E^{}=_C^{}_e^{}_C^{}\frac{A\lambda ^2}{(1+\lambda )^4}(\frac{1}{8}+\frac{1}{2}\lambda +\frac{5}{8}\lambda ^2)\frac{8}{5}_C^{},$$
(3.146)
where $`_C^{}`$ is standard exact exchange integral for Coloumb potential given by Eq.(3.15) while the approximate $`\lambda `$ dependent part arised from our potential $`V_e`$.
We have a good accuracy of the approximation (3.146). Indeed, exchange integrals make sensible contribution to the total molecular energy at deep overlapping of the wave functions, $`S>0.5`$, and we have made calculation just for the case of complete overlapping, $`S=1`$, with the necessary asymptotic factor, $`e^{2\rho }`$, provided by $`_C^{}(\rho )`$. Note that at $`\lambda \mathrm{}`$, the term $`_e^{}`$ of Eq.(3.146) becomes $`A_C^{}`$, as it should be because at $`\lambda \mathrm{}`$ (no screening) we have $`V_eA/r_{12}`$. In addition, although there is no possibility to restore completely $`\rho `$ dependence for the second term in r.h.s. of Eq.(3.146), we have got information on $`\lambda `$ dependence, which is of most interest here.
### 3.2 Numerical calculations for the $`V_e`$-based model
In this Section, we consider the case of exponential screened potential $`V_e=Ae^{r_{12}/r_c}/r`$, for which we have calculated all the needed molecular integrals.
The $`H_2`$ molecule energy, due to Eq.(3.2), is written as
$$E_{mol}(\gamma ,R,A,r_c)=2\frac{๐+๐^{}๐ฎ}{1+๐ฎ^2}\frac{2(๐+๐ฎ)(๐_C^{}๐_e^{}+_C^{}_e^{})}{1+๐ฎ^2}+\frac{1}{R},$$
(3.147)
where the specific terms are the Coloumb integral $`๐_e^{}`$ given by Eq.(3.107) and the exchange integral $`_e^{}`$ given by (3.146). We should find extremum of $`E_{mol}`$ as a function of our basic parameters, $`\gamma `$, $`R`$, $`A`$, and $`r_c`$. We are using notation $`\rho =\gamma R`$ and $`\lambda =2\gamma r_c`$ so that our four parameters are $`\gamma `$, $`\rho `$, $`A`$, and $`\lambda `$. In general, the number of energy levels of isoelectronium can also be viewed as a parameter of the model. However, we restrict our consideration by the one-level case, $`\beta ^2=1`$; see Sec. 2.1.
#### 3.2.1 Minimization of the energy
First, we analyze the $`A`$ dependence of $`E_{mol}`$. Due to Eq.(3.89), for one-level isoelectronium we have $`A=r_c^1`$, that can be identically rewritten as
$$A=\frac{2\gamma }{\lambda }.$$
(3.148)
Thus the $`A`$ dependence converts to $`\gamma `$ and $`\lambda `$ dependence. This is the consequence of consideration of the Hulten potential interaction for the electron pair made in Sec. 2.1.
Second, we turn to $`\gamma `$ dependence. Due to (3.148), the $`A`$ dependent parts, $`๐_e^{}`$ and $`_e^{}`$, acquire additional $`\gamma `$ factor and thus become $`\gamma ^2`$ dependent. The other molecular integrals depend on $`\gamma `$ linearly so that we define accordingly,
$$\overline{๐}=\frac{1}{\gamma }๐,\overline{}=\frac{1}{\gamma },\overline{๐}_C^{}=\frac{1}{\gamma }๐_C^{},\overline{}_C^{}=\frac{1}{\gamma }_C^{},\overline{๐}_e^{}=\frac{1}{\gamma ^2}๐_e^{},\overline{}_e^{}=\frac{1}{\gamma ^2}_e^{}.$$
(3.149)
Inserting the computed integrals $`๐`$ and $`๐^{}`$ into (3.147) we have
$$E_{mol}(\gamma ,\rho ,\lambda )=a\gamma +b\gamma ^2,$$
(3.150)
where
$$a(\rho ,\lambda )=\frac{2+2\overline{๐}+4S\overline{}\overline{๐}_C^{}\overline{}_C^{}}{1+S^2}\frac{1}{\rho }$$
(3.151)
and
$$b(\rho ,\lambda )=\frac{S^212S\overline{}+\overline{๐}_e^{}+\overline{}_e^{}}{1+S^2}.$$
(3.152)
The value of $`\gamma `$ corresponding to an extremum of $`E_{mol}`$ is found from the equation $`dE_{mol}/d\gamma =0`$, which gives the optimal value
$$\gamma _{opt}=\frac{a}{2b}.$$
(3.153)
Inserting this into (3.150) we get the extremal value of $`E_{mol}`$,
$$E_{mol}(\rho ,\lambda )=\frac{a^2}{4b}.$$
(3.154)
Using definitions of $`a`$ and $`b`$ we have explicitly
$$\gamma _{opt}=\frac{12\rho +S^2+\rho (2\overline{๐}4S\overline{}+\overline{๐}_C^{}+\overline{๐}_C^{})}{2\rho (1+S^22S\overline{}_C^{}+\overline{๐}_e^{}+\overline{}_e^{})}$$
(3.155)
and
$$E_{mol}(\rho ,\lambda )=\frac{(12\rho +S^2+\rho (2\overline{๐}4S\overline{}+\overline{๐}_C^{}+\overline{๐}_C^{}))^2}{4\rho ^2(1+S^2)(1+S^22S\overline{}_C^{}+\overline{๐}_e^{}+\overline{}_e^{})}.$$
(3.156)
Next, we turn to the extremum in the parameter $`\rho `$. The $`\rho `$ dependence, as well as the $`\lambda `$ dependence, of $`E_{mol}`$ is essentially nonalgebraic so that we are forced to use numerical calculations.
It appears that the $`\lambda `$ dependence does not reveal any local energy minimum while the $`\rho `$ dependence does. Below, we use the condition, $`\lambda ^1=`$ integer number, obtained during the calculation of the Coloumb integral with Hulten potential $`V_h`$; see Eq.(3.57). Although there is obviously no necessity to keep this condition for the case of exponential screened potential $`V_e`$, we consider it as a prescription for allowed values of $`\lambda `$.
Since the $`\lambda `$ dependence of the energy has no minimum we can use fitting of the predicted energy $`E_{mol}(\lambda )`$ to the experimental value by varying $`\lambda `$. This allows us to estimate the value of the parameter $`\lambda `$, and thus the value of the effective radius of the isoelectronium $`r_c=\lambda /2\gamma _{opt}`$.
#### 3.2.2 Fitting of the energy and the bond length
The procedure is the following. We fix some numerical value of $`\lambda `$, and identify minimal value of $`E_{mol}(\rho ,\lambda )`$, given by Eq.(3.156), in respect with the parameter $`\rho `$. This gives us minimal energy and corresponding optimal value of $`\rho `$, at some fixed value of $`\lambda `$. Then, we calculate $`\gamma _{opt}`$ by using Eq.(3.155), and use obtained values of $`\rho _{opt}`$ and $`\gamma _{opt}`$ to calculate values of $`R_{opt}`$ and $`r_c`$.
We calculated minimal values of $`E_{mol}`$ in $`\rho `$, for a wide range of integer values of $`\lambda ^1`$. The results are presented in Tables 2 and 3, and Figures 15 and 16. One can see that the energy $`E_{mol}`$ decreases with the increase of $`r_c`$ (proportional to size of isoelectronium), as it was expected to be.
We note that all the presented values of $`E_{mol}`$ in Tables 2 and 3 are lower than that, $`E_{mol}^{var}=1.139\text{ a.u.}`$, obtained via two-parametric Ritz variational approach to the standard model of $`H_2`$ (see, e.g., ), which is the model without the assumption of short-range attractive potential between the electrons. This means that the $`V_e`$-based model gives better prediction than the one of the standard model, for any admitted value of the effective radius of isoelectronium $`r_c>0`$. Indeed, the standard prediction $`E_{mol}^{var}=1.139\text{ a.u.}`$ is much higher than the experimental value $`E_{exper}[H_2]=1.174474\text{ a.u.}`$
#### 3.2.3 The results of fitting
Best fit of the energy $`E_{mol}`$.
Due to Table 2 (see also Fig. 15), the experimental value, $`E_{exp}[H_2]=1.174\mathrm{}1.164\text{ a.u.}`$ (here we take 0.9% uncertainty of the experimental value) is fitted by
$$r_c=0.0833\mathrm{}0.0600\text{ a.u.},$$
(3.157)
i.e. $`\lambda =1/5\mathrm{}1/7`$, with the optimal distance, $`R_{opt}=1.3184\mathrm{}1.3441\text{ a.u.}`$ We see that the predicted $`R_{opt}`$ appeared to be about 6% less than the experimental value $`R_{exper}[H_2]=1.4011\text{ a.u.}`$ We assign this discrepancy to the approximation we have made for the exchange integral (3.146).
Below we fit $`R_{opt}`$, to estimate the associated minimal energy.
Best fit of the internuclear distance $`R`$.
Due to Table 2 (see also Fig. 16), the experimental value of the internuclear distance, $`R_{exp}=1.4011\text{ a.u.}`$, is fitted by $`r_c=0.0115\text{ a.u.}`$, with the corresponding minimal energy $`E_{min}=1.144\text{ a.u.}`$, which is about 3% bigger than the experimental value. Again, we assign this discrepancy to the approximation we have made for the exchange integral (3.146), and take
$$r_c=0.0115\text{ a.u.},$$
(3.158)
i.e. $`\lambda =1/37`$, as the result of our final fit noting that (a) in ref. the value $`r_c=0.0112\text{ a.u.}`$ has been used to make exact numerical fit of the energy, with corresponding $`R=1.40\text{ a.u.}`$, and (b) we have less discrepancy.
The weight of the pure isoelectronium phase.
To estimate the weight of the pure isoelectronium phase, which can be viewed as a measure of stability of the pure isoelectronium state, we use the above obtained fits and the fact that this phase makes contribution to the total molecular energy via the Coloumb and exchange integrals.
According to Eq.(3.147), the isoelectronium phase displays itself only by the term $`P_e|๐_e^{}(\gamma ,\rho ,\lambda )+_e^{}(\gamma ,\rho ,\lambda )|`$ while the Coloumb phase displays itself by the corresponding term $`P_C|๐_C^{}(\gamma ,\rho )+_C^{}(\gamma ,\rho )|`$. Putting the total sum $`P_C+P_e=1`$, i.e. $`P_C+P_e`$ is 100%, the weights are defined simply by
$$W_C=\frac{P_C}{P_C+P_e},W_e=\frac{P_e}{P_C+P_e},$$
(3.159)
The weight for the best fit of $`R`$.
At the values $`\lambda =1/37`$ (i.e. $`r_c=0.0115`$), $`\gamma =1.1706`$, and $`\rho =1.6320`$, for which we have minimal $`E_{mol}=1.144`$ and optimal $`R=1.40`$, we get the numerical values of the weights,
$$W_e=0.84\%$$
(3.160)
for the pure isoelectronium phase, and $`W_C=100\%W_e=99.16\%`$ for the Coloumb phase.
The weight for the best fit of $`E_{mol}`$.
At the values $`\lambda =1/5`$ (i.e. $`r_c=0.0833\text{ a.u.}`$), $`\gamma =1.2005`$, and $`\rho =1.5827`$, for which we have minimal $`E_{mol}=1.173\text{ a.u.}`$ and optimal $`R=1.318\text{ a.u.}`$, we obtain
$$W_e=6.16\%,W_C=93.84\%.$$
(3.161)
From the above two cases, one can see that the weight of pure isoelectronium phase is estimated to be
$$W_e1\mathrm{}6\%,$$
(3.162)
for the predicted variational energy $`E_{mol}=1.143\mathrm{}1.173\text{ a.u.}`$
The biggest possible weight.
Note that in our $`V_e`$-based model the biggest allowed value of $`\lambda `$ is $`\lambda =1/4`$ (i.e. $`r_c=0.1034`$) because $`\lambda <1/3`$, to avoid divergency of the Coloumb integral $`๐_e`$. For this value of $`\lambda `$, we obtain minimal $`E_{mol}=1.182\text{ a.u.}`$ and optimal $`R=1.297\text{ a.u.}`$ This value corresponds to the biggest possible weight of the pure isoelectronium phase,
$$W_e=7.32\%,$$
(3.163)
within our approximate model.
The following three remarks are in order.
(i) We consider the existence of this upper limit, $`W_e7.32\%`$, as a highly remarkable implication of our $`V_e`$-model noting however that it may be artifact of the use of the exponential screened Coloumb potential.
(ii) Another remarkable implication is due to the condition, $`\lambda ^1=`$ integer number, obtained for the case of Hulten potential. One can see from Table 2 that the energy $`E_{mol}`$ varies discretely with the discrete variation of $`\lambda ^1`$. This means that there is no possibility to make a โsmooth fitโ. For example, at $`\lambda =1/5`$, we have $`E_{mol}=1.173`$, and the nearest two values, $`\lambda =1/4`$ and $`\lambda =1/6`$, give us $`E_{mol}=1.182`$ and $`E_{mol}=1.167`$, respectively. Therefore, owing to the above condition the model becomes more predicitive.
(iii) Numerical calculation shows that the formal use of the exact Coloumb integral $`๐_g^{}`$, given by Eq.(3.128), of the Gaussian screened Coloumb potential, instead of $`๐_e^{}`$, in Eq.(3.147) gives us approximately the same fits. Namely, the best fit of the energy is achieved at $`\lambda =1/5`$, with $`r_c=0.1042`$, optimal $`R=1.323`$, and minimal $`E_{mol}=1.172`$. Also, the best fit of $`R=1.40`$ is at $`\lambda =1/29`$, for which $`r_c=0.0147`$ and minimal $`E_{mol}=1.144`$. Here, we have used the same exchange integral as it is for the case of exponential screened potential so these fits have been presented just for a comparison with our basic fits, and to check the results. Note that for the case of Gaussian screened Coloumb integral we have no restriction on the allowed values of $`\lambda `$. Analysis shows that, at big values of $`\lambda `$, e.g. at $`\lambda >4`$, the integral $`๐_g^{}`$, given by (3.128), rapidly oscillates in the region of small $`\rho `$ ($`\rho <0.5`$). This means that when the correlation length $`r_c`$ becomes comparable to the internuclear distance an effect of instability of the molecule arises. This can be viewed as a natural criterium to fix the upper limit of $`\lambda `$. Normally, we use the values $`\lambda <1`$, for which case there are no any oscillations of $`๐_g^{}`$ (see Fig. 9). |
warning/0001/hep-ph0001075.html | ar5iv | text | # String formation and chiral symmetry breaking in the heavyโlight quarkโantiquark system in QCD
## I Introduction
The nature of QCD string formation between static sources was studied on the lattice and analytically . From these investigations it was shown that the string consists of a predominantly colorโelectric longitudinal field. At the critical temperature $`T_c`$ this electric field disappears and and at the same time the deconfined phase with colorโmagnetic condensate sets in. This effect was predicted theoretically in and also seen in lattice measurements. At the same temperature Chiral Symmetry Breaking (CSB) for light quarks is found to disappear, which indicates that there is an intimate connection between the string formation and CSB. In the case of heavy quark systems CSB occurs due to the quark mass and confinement can be described as the area law of the Wilson loop. How CSB and confinement are explicitly realized for the light quark system and what equation describes its dynamics is an interesting and open problem.
It is the purpose of the present paper to study this issue in the simplest dynamical example โ in the system of one light quark and a heavy antiquark. This allows us to describe the dynamics of light quark (its propagator) in a gaugeโinvariant manner, while physically the light quark is expected to be confined at the end of a string connected to the static source. Applying the formalism of field correlators (FC) , we derive the effective quark Lagrangian, containing any number of quark operators multiplied by field correlators. To proceed further one can use the limit of large $`N_c`$ and write down the Dyson-Schwinger equations with the mass operator expressed through the Greenโs function. The resulting equations are nonlocal and nonlinear. It is not clear from the beginning how confinement and CSB would manifest themselves in the solution of these equations. Some hint was provided in Ref. using a relativistic WKB analysis , where it was shown that at large distances from the heavy source the dynamics of the light quark is described by the Dirac equation with a scalar linear confining interaction, which leads to CSB.
In this paper we examine the properties of the Greenโs function of the color singlet $`q\overline{Q}`$ system, where the antiquark is treated in the static limit. In section II we formulate the full form of nonlocal and nonlinear equations for the light quark propagator and its eigenfunctions and study the behaviour at all distances. We also take into account both perturbative and nonperturbative contributions to the interaction kernel. As a result our equations contain both a confining interaction and color Coulomb part. Similar to the heavy quark situation we argue that the string formation for low angular momentum is of a color-electric nature. Moreover, the confinement of the light quark to the heavy one is shown to be of Lorentz scalar type. In section III the resulting nonlinear equations are studied numerically. The energy spectrum and the structure of the low lying eigenfunctions are presented. We in particular study the $`B`$ and $`D`$ meson spectrum. We compare them with experimental data for $`B`$, $`D`$ mesons and results of other calculations, exploiting for this purpose the expansion of Heavy Quark Effective Theory (HQET). In section IV the chiral condensate of the light quarks $`\overline{q}q`$ is determined by taking the limit of the Greenโs function $`S(x,y)`$, with both $`x`$ and $`y`$ tending to zero. In this limit the heavy quark is turned off and the condensate $`\overline{q}qS(0,0)`$ should have a value not depending on the presence of heavy quark.
## II Derivation of equations
### A Dyson-Schwinger equations
In this section we give an outline of the procedure to obtain the Dyson-Schwinger equations for the color white $`q\overline{Q}`$ system. Our starting point is the gaugeโinvariant light quark Greenโs function $`S(x,y)`$ in the presence of a static heavy antiquark placed in the origin. In the static limit, the heavy quark can be treated as an external source. Assuming the Euclidean metric and letting $`T=(xy)_4`$, the heavy antiquark propagator in the modified Fock-Schwinger gauge is proportional to the parallel transporter,namely
$$S_{\overline{Q}}(A)=h(๐ฑ,๐ฒ)Pexp(ig_0^TA_4(๐ซ=0,\tau )๐\tau )$$
(1)
with
$$h(๐ฑ,๐ฒ)=\frac{i}{2}\delta ^{(3)}(๐ฑ๐ฒ)[(1+\gamma _4)e^{m_QT}\mathrm{\Theta }(T)+(TT,\gamma _4\gamma _4)].$$
This acts as a static source situated at the origin. As a result the proper limit for $`m_Q\mathrm{}`$ of the Greenโs function of the $`q\overline{Q}`$ system can be defined as
$$S(x,y)=\psi (x)\mathrm{\Pi }(x,y)\psi ^+(y).$$
(2)
Here we have to average over the gluon fields $`A`$ and light quark fields $`\psi `$, while $`\mathrm{\Pi }`$ contains the parallel transporters between the end points
$$\mathrm{\Pi }(x,y)\varphi (๐ฑ,x_4;0,x_4)\varphi (0,x_4;0,y_4)\varphi (0,y_4;๐ฒ,y_4)$$
with
$$\varphi (๐ฑ,x_4;๐ฒ,y_4)=Pexpig_x^yA_\mu ๐z_\mu $$
.
The averaging over the gluon fields $`A`$ has to be done over the perturbative and nonperturbative gluon fields $`a_\mu `$ and $`B_\mu `$ contributions respectively, where the total gluonic field $`A_\mu =B_\mu +a_\mu `$. We now apply the method of field correlators (FC), which was developed in a series of papers to derive the effective quark Lagrangian from QCD. Let us first consider the effects of averaging over nonperturbative (NP) field $`B_\mu `$. We may write for the partition function
$$Pe^{g{\scriptscriptstyle \psi ^+\widehat{B}(x)\psi ๐x}}_Be^{L_{eff}}=Pexp[\underset{n}{}\frac{g^n}{n!}d^4x_1\mathrm{}d^4x_nj(1)\mathrm{}j(n)B(1)\mathrm{}B(n)]$$
(3)
with $`j(n)j_{\mu _n}(x_n)=\psi ^+(x_n)\gamma _{\mu _n}\psi (x_n)`$ and $`B(n)=B_{\mu _n}(x_n)`$. To write the correlator $`B\mathrm{}B`$ for the gaugeโinvariant situation corresponding to the color white $`q\overline{Q}`$ system one can use the modified Fock-Schwinger gauge to express the correlator of $`B(x)`$ through FC:
$$N(1,\mathrm{}n)B(1)\mathrm{}B(n)๐x(1)\mathrm{}๐x(n)F(1)\mathrm{}F(n).$$
(4)
The effective interaction kernel in Eq. (3) can now be used to write a DysonโSchwingerโtype equation for the quark Greenโs function $`S`$. To simplify matter, one can consider the large $`N_c`$ limit, in which case the connected selfโenergy kernel $`M(x,y)`$ is obtained from Eq. (3) by replacing any pair of adjacent $`\psi `$โoperators by
$$\psi _{a\alpha }(x)\psi _{a\beta }^+(y)N_cS_{\alpha \beta }(x,y),$$
(5)
where $`a\alpha `$ are color and Lorentz index respectively (for details of derivation see ). As a result one obtains the equation for $`S(x,y)`$
$$(i/_xim)S(x,y)iM(x,z)S(z,y)d^4z=\delta ^{(4)}(xy),$$
(6)
where the kernel $`M`$ is expressed through $`N`$ as
$`iM(x,y)`$ $`=N_{\mu \nu }^{(2)}(x,y)\gamma _\mu S(x,y)\gamma _\nu +`$ (8)
$`{\displaystyle \underset{n=3}{\overset{\mathrm{}}{}}}{\displaystyle d^4x_2\mathrm{}d^4x_{n1}\gamma _{\mu _1}S(x,x_2)\gamma _{\mu _2}\mathrm{}\gamma _{\mu _{n1}}S(x_{n1},y)\gamma _{\mu _n}N_{\mu _1\mathrm{}\mu _n}^{(n)}(x,x_2\mathrm{}x_{n1},y)}.`$
The system of equations (6-7) is exact in the large $`N_c`$ limit and is well defined provided all NP correlators $`F(1)\mathrm{}F(n)`$ are known.
Evidence has been found in recent accurate lattice calculations of static potentials in different $`SU(3)`$ representations, that the contributions of the higher correlators $`F(1)\mathrm{}F(n)`$ for $`n>2`$ to the planar Wilson loop are small. In particular, these terms are found to contribute only around a few percent of the dominant Gaussian correlator . Hence, the Gaussian Stochastic Model, based on the lowest correlator $`F(1)F(2)`$ is expected to be a good approximation. In view of this we assume in this paper that the Gaussian approximation holds and we keep similar as in Ref. only the first term in Eq. (8). One can parametrize the Gaussian correlator according to as
$$F(1)F(2)=\frac{1}{N_c}trF_{\mu \lambda }(x)\varphi (x,0)F_{\nu \sigma }(0)\varphi (0,x)=D(x)(\delta _{\mu \nu }\delta _{\lambda \sigma }\delta _{\mu \sigma }\delta _{\nu \lambda })+\mathrm{\Delta }_{\mu \lambda \nu \sigma }^{(1)},$$
(9)
where only $`D(x)`$ is responsible for confinement and it contributes to string tension $`\sigma `$, while $`\mathrm{\Delta }^{(1)}`$ is a full derivative. As a consequence, the latter contributes to the perimeter of Wilson loop and $`\varphi (x,0)=Pexpig_0^xB_\mu ๐z_\mu `$. From this we find, that the $`N_{\mu \nu }^{(2)}`$ can explicitly be written in the gauge as
$$N_{\mu \nu }^{(2)}(x,y)=(\delta _{\mu \nu }\delta _{ik}\delta _{\mu k}\delta _{\nu i})_0^x๐u_i\alpha _\mu (u)_0^y๐v_k\alpha _\nu (v)D(uv),$$
(10)
where $`\alpha _4(u)=1,\alpha _i(u)=\frac{u_i}{x_i}`$.
In Eq. (10) only the nonperturbative confining piece of the Gaussian correlator (9) is retained, since the perturbative part and $`\mathrm{\Delta }^{(1)}`$ do not produce neither the string (confinement) nor CSB . In the case of a nonrotating string the terms in Eq. (7) with space components $`\gamma _k`$ are suppressed by powers of velocity of the endpoints of the string. In what follows we shall keep for simplicity only the component $`N_{44}^{(2)}J(x,y)`$ of $`N_{\mu \nu }^{(2)}`$. Hence the kernel $`M`$ is proportional to the FC of the colorโelectric field $`E_iF_{i4}`$. It is the dominant part of the string. Colorโmagnetic components are neglected in this first step and can be considered as a correction.
In this way one arrives at the system of equations where we keep the same notation $`M(x,y)`$ for the retained piece of the kernel
$$iM(x,y)=J(x,y)\gamma _4S(x,y)\gamma _4$$
(11)
$$(i/_xim)S(x,y)iM(x,z)S(z,y)d^4z=\delta ^{(4)}(xy).$$
(12)
### B Partial wave reduced equations
The kernel $`J(x,y)`$ in Eq. (11) depends on the time as $`\frac{x_4y_4}{T_g}`$. Therefore in the limit of $`T_g0`$ it becomes local in time. As a result, the Fourierโtransform of $`M`$ in the fourth coordinate does not depend on the momentum $`p_4`$ for $`T_g0`$. The corresponding eigenfunctions $`\psi _n(๐ฑ,p_4)=\psi _n(๐ฑ)`$ and eigenvalues $`\epsilon _n=\epsilon _n(p_4)`$ of the light quark in the white light-heavy configuration can readily be obtained from Eqs. (11-12) in the discussed instantaneous limit. We find
$$(\frac{๐ถ}{i}\frac{}{๐ฑ}+\beta m)\psi _n(\stackrel{}{x})+\beta M(๐ฑ,๐ณ)\psi _n(๐ณ)d^3๐ณ=\epsilon _n\psi _n(๐ฑ)$$
(13)
$$M(๐ฑ,๐ณ)=J(๐ฑ,๐ณ)\beta \mathrm{\Lambda }(๐ฑ,๐ณ),\mathrm{\Lambda }(๐ฑ,๐ณ)=\underset{k}{}\psi _k(x)sign\epsilon _k\psi _k^+(z)$$
(14)
The sphericalโspinor decomposition of $`\psi _n`$ for the total and orbital angular momentum channel $`j`$, $`l=j\pm \frac{1}{2}`$,
$$\psi _n(r)=\frac{1}{r}\left(\begin{array}{c}G_n(r)\mathrm{\Omega }_{jlM}\hfill \\ iF_n(r)\mathrm{\Omega }_{jl^{}M}\hfill \end{array}\right),l^{}=2jl$$
(15)
yields equations for the partial waves
$$\frac{dF_n}{dr}\frac{\kappa }{r}F_n+(\epsilon _nm)G_nM_{11}G_niM_{12}F_n=0$$
(16)
$$\frac{dG_n}{dr}+\frac{\kappa }{r}G_n(\epsilon _n+m)F_nM_{22}F_n+iM_{21}G_n=0$$
(17)
with $`\kappa =\pm (j+\frac{1}{2})`$. Clearly the kernels $`M_{ik}`$ are nonlocal in space, i.e.
$$M_{ik}G_n=_0^{\mathrm{}}M_{ik}^{jj}(r,r^{})G_n(r^{})rr^{}๐r^{}$$
(18)
with $`M_{ik}^{jj}=\mathrm{\Omega }_{jl_iM}|M_{ik}|\mathrm{\Omega }_{jl_kM}`$ and $`M_{ik}`$ is given by Eq. (14).
Equations (16-17) are invariant under the transformation
$$\epsilon _n\epsilon _n,\kappa \kappa ,G_nF_n$$
(19)
which also yields $`M_{11}M_{22},M_{12}M_{21}.`$ The symmetry (19) implies that the spectrum is symmetric in $`\epsilon _n\epsilon _n`$, which is a property of a scalar interaction. The Lorentz scalar nature of the confining interaction has the nice feature that it doesnot lead to instability problems in the Dirac equation . Moreover, nontrivial solutions of Eqs. (16-17), if they exist, signify spontaneous CSB.
We solve Eqs. (16-17), using the relativistic WKB approximation for the kernel $`M_{ik}`$. To simplify the calculations the Gaussian form for $`D(x)`$ was used (since all observables are integrals of FC, its explicit form is not essential at large distances, provided the FC have a finite range $`T_g`$ and it yields the same value of the string tension $`\sigma `$)
$$D(u)=D(0)exp(u^2/4T_g^2),D(0)=\frac{\sigma }{2\pi T_g^2}$$
(20)
with $`\sigma =0.2GeV^2,`$ and $`T_g=0.25fm`$, taken in accordance with the lattice measurements .
As the reference basis we take the WKB solutions of the Dirac equation for the local linear confining potential $`\sigma r`$, and the WKB computed kernels $`\stackrel{~}{M}`$ and $`\stackrel{~}{\mathrm{\Lambda }}(๐ฑ,๐ฒ)`$ in Eq. (14) are determined by explicit summation over eigenstates. It was checked by an independent calculation that the relativistic WKB procedure yields eigenvalues of the linear potential with accuracy better than one percent.
The general structure of $`M`$ and $`\stackrel{~}{M}`$ can be derived from Eqs. (14-15). One can write $`M`$ as a $`4\times 4`$ matrix as follows
$$M=M^{(0)}I+M^{(i)}\widehat{\sigma }_i+M^{(4)}\gamma _4+M_\gamma ^{(i)}\gamma _i,$$
(21)
where $`\gamma _i,\gamma _4`$ are usual Dirac matrices, $`i=1,2,3`$ and $`\widehat{\sigma }_i=\left(\begin{array}{cc}\sigma _i\hfill & 0\hfill \\ 0\hfill & \sigma _i\hfill \end{array}\right)`$.
The same representation holds for the WKB approximated $`\stackrel{~}{M}`$. From the WKB analysis we find that $`\stackrel{~}{M}^{(0)}`$ is the only growing kernel. It behaves asymptotically as
$$M^{(0)}(๐ฑ,๐ฒ)=\sigma \frac{|๐ฑ+๐ฒ|}{2}\stackrel{~}{\delta }^{(3)}(๐ฑ๐ฒ)$$
(22)
where $`\stackrel{~}{\delta }^{(3)}(๐ซ)`$ is a smeared $`\delta `$ โ function with the range of nonlocality decreasing asymptotically with growing $`|๐ฑ|`$, $`|๐ฒ|`$. The term $`M^{(i)}`$ is proportional to the angular momentum $`L`$ and asymptotically it behaves as $`O(1/x)`$. One can also prove that $`M^{(4)},M_\gamma ^{(i)}`$ do not grow at large $`x,y`$ . Hence in all problems where large distances are dominant one can consider only the first term in Eq. (21). Using this approximation we get for the kernel (14)
$$\stackrel{~}{M}=\stackrel{~}{M}^{(0)}(๐ฑ,๐ฒ)I=J(๐ฑ,๐ฒ)\stackrel{~}{\mathrm{\Lambda }}(๐ฑ,๐ฒ)I$$
(23)
with
$$J(๐ฑ,๐ฒ)=\sigma \frac{\mathrm{๐ฑ๐ฒ}}{\sqrt{\pi }}f(๐ฑ,๐ฒ),f(๐ฑ,๐ฒ)=_0^1๐s_0^1๐te^{\frac{(๐ฑs๐ฒt)^2}{4T_g^2}}$$
(24)
and where $`\stackrel{~}{\mathrm{\Lambda }}`$ is now a scalar quantity.
A curious feature of the considered equations is the way how the string connecting the light quark to the source is being created. Actually, if one takes only lowest partial waves inside the kernel $`M`$ (i.e. in $`\mathrm{\Lambda }(๐ฑ,๐ฒ)`$, Eq. (14)), then the effective potential in Eqs. (16-17) is not confining. If one however sums up over all angular states and radial excitations in $`\mathrm{\Lambda }`$, then the resulting $`\mathrm{\Lambda }(x,y)`$ is a smeared $`\delta `$โ function leading to the quasilocal confining kernel $`\stackrel{~}{M}`$. E.g. using eigenfunctions for the local case, $`\stackrel{~}{\mathrm{\Lambda }}`$ can be computed quasiclassically to be
$$\stackrel{~}{\mathrm{\Lambda }}(๐ฑ,๐ฒ)=\frac{\sigma ^2xy}{2\pi ^2}\frac{K_1(\sigma \sqrt{xy}\sqrt{(xy)^2+\theta ^2xy)}}{\sqrt{(xy)^2+\theta ^2xy}},๐ฑ๐ฒ=xy\mathrm{cos}\theta $$
(25)
In Eq. (25) one can clearly see that $`\stackrel{~}{\mathrm{\Lambda }}(x,y)`$ is a normalized smeared $`\delta `$โfunction, with smearing radius in $`|xy|`$ being $`\frac{1}{\sigma \sqrt{xy}}`$. For large distances it is nonvanishing only in the forward direction.
Insertion of this $`\stackrel{~}{\mathrm{\Lambda }}`$ into $`\stackrel{~}{M}`$, Eq. (23), produces linear confinement due to the kernel $`\stackrel{~}{\mathrm{\Lambda }}(๐ฑ,๐ฒ)`$, as is given by Eq. (25) (while simply averaging the contribution from each individual orbital in Eq. (14) over the angle between $`๐ฑ`$ and $`๐ฒ`$ would produce no confinement at all). The computed kernel $`\stackrel{~}{M}(x,y)`$ turns out to be nonlocal, but very close to the linear potential at large distances. Indeed, the effective localized potential defined as
$$V_{eff}(r)=M^{(0)}(r,x)rx๐x$$
(26)
approaches at very large distances, i.e. for $`\sigma ^{1/2}r>200`$, a linear dependence with a slope given by $`\sigma `$. However at shorter distances $`V_{eff}`$ looks also linear over a relatively large region with a slope almost the same as $`\sigma `$, reflecting the presence of a small local curvature. In particular, we find that in the region $`5<\sigma ^{1/2}r<20`$ the effective potential can reasonable well be described by $`V_0(r)=0.9\sigma r1.8\sigma ^{1/2}`$. In Fig. 1 are shown the results up to $`r=20`$ in units of $`\sigma ^{1/2}`$.
Since it is only the higher states and large distances which are important in the creation of this $`\delta `$ โ functionโtype behaviour of $`\mathrm{\Lambda }(x,y)`$, and since the WKB method does well for high states and at large distances, one can clearly conclude, that linear confinement should be obtained if one sums over all exact solutions of Eqs. (16-17). Hence this should be a property of the exact solution.
The property, that the kernel has a focussing effect in the forward direction can be used to get a somewhat simpler form. For this purpose we may also use
$$\stackrel{~}{\mathrm{\Lambda }}(๐ฑ,๐ฒ)=\frac{\sigma }{\pi ^2\sqrt{xy}}K_0(a)\delta (cos\theta _xcos\theta _y),a=\sigma \sqrt{xy}|xy|$$
(27)
The eigenvalues and eigenfunctions in Eqs. (16-17) have been determined using the kernel $`\stackrel{~}{M}(x,y)`$, given by Eqs. (25) and (27). Some results are shown in Table 1 and Fig. 2. In our present study we have taken $`m=0`$. Note, that $`\stackrel{~}{M}(x,y)`$ is approaching a local linear potential at large distances, $`x,y\begin{array}{c}>\hfill \\ \hfill \end{array}\sigma ^{1/2}`$, which justifies a posteriori our choice of the reference basis. Due to the nonlocality of the interaction the predicted spectrum is found to be different from that of the linear potential, valid at large distances. Moreover, comparing the level structures of the $`J=\frac{1}{2}`$ channel as obtained using the kernels (25) and (27) we see from Table I, that they are qualitatively very similar, corroborating that there is indeed a strong forward focussing effect in the quark propagator. From Fig. 2 we see that for all L values the higher radially excited levels are close to the predictions of linear potential $`V_0(r)=0.9\sigma r1.8\sigma ^{1/2}`$, in agreement with the fact that the interaction at large distance can indeed be described by a local linear potential. On the other hand the nonlocal kernel predictions for the low lying states clearly deviates strongly from those of the (shifted) linear potential. Hence the nonlocal nature of the force does affect the spectrum in an essential way.
The eigenfunctions for the nonlocal kernels look qualitatitively similar to the corresponding ones of the shifted linear potential. In Fig. 3 are shown the ground state and first excited state for the $`J=\frac{1}{2}`$ channels. Altough the differences are substantial for these low lying states, the agreement for higher excited states is considerably better. Moreover, we find that the large distances and high states of the WKB states agree well with the corresponding eigenfunctions.
### C Inclusion of perturbative exchanges
Till now we have considered only NP part of the gluonic field, $`B_\mu `$. In this section we include the perturbative part, $`a_\mu `$, and neglect for simplicity the interference terms. Therefore the effect of $`a_\mu `$ is accounted for in the appearance of an additional factor in the partition function (3), namely
$$Z=Z_{NP}Z_{pert},Z_{pert}=e^{g{\scriptscriptstyle ๐x\psi ^+\widehat{a}(x)\psi (x)}+ig{\scriptscriptstyle ๐z_4a_4(z_4)}}_a,$$
(28)
where the second term in the exponent of (28) corresponds to the interaction of the perturbative part of the gluon field with the static antiquark. We have used in (28), that due to the โtHooft identity one can average independently over $`B_\mu `$ and $`a_\mu `$.
The result of averaging yields a new additive term in $`L_{eff}`$, Eq. (3),
$$L_c=g๐x\psi ^+(x)\widehat{A}^{(c)}(x)\psi (x),$$
(29)
where we have defined
$$A_\mu ^{(c)}(x)=ig๐z_4<a_4(x)a_4(z_4)>=\delta _{\mu 4}\frac{(i)gC_2}{4\pi |๐ฑ|}.$$
(30)
The presence of $`L_c`$ in Eq. (3) does not influence the derivation of basic Eqs. (11-12). The only difference is that Eq. (12) assumes the form
$$(i\widehat{}g\widehat{A}^c(x)im)S(x,y)iM(x,z)S(z,y)d^4z=\delta ^{(4)}(xy).$$
(31)
Eq. (11) does not change and the kernel $`J(x,y)`$ contains as before only nonperturbative contributions. Note however that $`S(x,y)`$ in $`M(x,y)`$ in Eq. (11) now contains also perturbative gluon exchanges. This is a new type of interference of perturbative and NP terms, which appears irrespectively of our neglect of this interference within the averaging procedure over $`B_\mu `$ and $`a_\mu `$. In other words another class of diagrams is responsible for this interference.
Correspondingly in the static equations (13) one should replace
$$\beta m\beta m\frac{C_2\alpha _s}{|๐ณ|}$$
(32)
The equations for the partial waves (16-17) are modified due to the presence of the color Coulomb potential $`V(r)`$ in a simple way. Since
$$V(r)=\frac{C_2\alpha _s}{r}$$
(33)
is local and a Lorentz vector, it always appears in the combination $`\epsilon _nV(r)`$. Hence one has instead of Eqs. (16-17)
$$\frac{dF_\nu }{dr}\frac{\kappa }{r}F_\nu +(\epsilon _\nu V(r)m)G_\nu M_{11}GiM_{12}F=0,$$
(34)
$$\frac{dG_\nu }{dr}+\frac{\kappa }{r}G_\nu (\epsilon _\nu V(r)+m)F_\nu M_{22}F+iM_{21}G=0,$$
(35)
where we have denoted
$$M_{ik}\left(\begin{array}{c}G\hfill \\ F\hfill \end{array}\right)<\nu |M_{ik}|\nu ^{}>\left(\begin{array}{c}G_\nu ^{}(w)\hfill \\ F_\nu ^{}(w)\hfill \end{array}\right)rw๐w.$$
(36)
Here $`M_{ik}`$ is defined as in Eq. (14) and the matrix $`\mathrm{\Lambda }_{ik}`$ in Eq. (14) involves the sum over all states, including positive and negative $`\epsilon _n`$. There in section IIB we have exploited the symmetry (19). However Eqs. (34-35) are invariant under another transformation, namely
$$\epsilon _n\epsilon _n,V(r)V(r),\kappa \kappa ,G_nF_n.$$
(37)
Now the sum over negative $`\epsilon _n`$ can be expressed through the corresponding sum over positive $`\epsilon _n`$ with exchange $`G_nF_n`$ as before, but also with the inversion of sign of Coulomb interaction, i.e. Coulomb attraction for positive $`\epsilon _n`$ is replaced by Coulomb repulsion for negative $`\epsilon _n`$.
In what follows we shall denote wave functions of the positive energy states with repulsive Coulomb with the sign of tilde: $`\stackrel{~}{G}_\nu ,\stackrel{~}{F}_\nu `$. Then using (14) the matrix $`\beta \mathrm{\Lambda }_{ik}`$ can be written as a sum over only positive $`\epsilon _n`$ as follows
$$\beta \mathrm{\Lambda }_{ik}^{\mu \mu ^{}}=\frac{1}{xy}\underset{jlM,n>0}{}\left(\begin{array}{cc}G_\mu G_\mu ^{}^{}\stackrel{~}{F}_\mu \stackrel{~}{F}_\mu ^{}^{},\hfill & i(G_\mu F_\mu ^{}^{}\stackrel{~}{F}_\mu \stackrel{~}{G}_\mu ^{}^{})\hfill \\ i(F_\mu G_\mu ^{}^{}\stackrel{~}{G}_\mu \stackrel{~}{F}_\mu ^{}^{}),\hfill & \stackrel{~}{G}_\mu \stackrel{~}{G}_\mu ^{}^{}F_\mu F_\mu ^{}^{}\hfill \end{array}\right).$$
(38)
Since $`\beta \mathrm{\Lambda }`$ is exactly the combination which enters the mass matrix (14), one can list in (38) scalar and vector (proportional to $`\beta )`$ parts:
$$M=M_sI+M_v\beta +\mathrm{\Delta }M,$$
(39)
where $`\mathrm{\Delta }M`$ contains spin-dependent terms, which can be considered as in section IIB, while $`M_s`$, $`M_v`$ are
$$M_{s,v}=C\underset{jlM\mu ,n>0}{}[G_\mu G_\mu ^{}\stackrel{~}{F}_\mu \stackrel{~}{F}_\mu ^{}\pm (\stackrel{~}{G}_\mu \stackrel{~}{G}_\mu F_\mu F_\mu ^{})]$$
(40)
where
$$C=\frac{1}{4}\sqrt{\pi }T_gD(0)\frac{๐ฑ๐ฒ}{xy}f(๐ฑ,๐ฒ),f(๐ฑ,๐ฒ)=_0^1๐s_0^1๐texp\left(\frac{(๐ฑs๐ฒt)^2}{4T_g^2}\right).$$
(41)
From (40) it is clear that the vector part $`M_v`$ is only due to the presence of Coulomb interaction. Corrections at large distances due to the vector part can be treated again in the relativistic WKB. A rough estimate of $`M_v`$ at large $`r`$ yields
$$\frac{M_v}{M_s}\frac{\alpha _s}{\sigma r^2}$$
(42)
and hence can be neglected at large enough $`r`$.
## III Numerical solutions of equations and comparison to $`B`$, $`D`$ mesons
We have performed numerical studies of Eqs. (34-35) with the kernel (25) for different values of the quark mass $`m`$ and different values of $`T_g`$. To simplify calculations only the dominant part of the mass operator $`M_{ik}`$ was retained, i.e. $`\widehat{M}_{11}=M_{22}=M^{(0)}`$, while $`M_{12},M_{21}`$ have been neglected. For $`M^{(0)}`$ the representation (23) was used
$$M^{(0)}(๐ฑ,๐ฒ)=J(๐ฑ,๐ฒ)\stackrel{~}{\mathrm{\Lambda }}(๐ฑ,๐ฒ)I,$$
where $`\stackrel{~}{\mathrm{\Lambda }}`$ is taken to be the kernel (25). Results of our calculations for the ground state energy are listed in Table 2. One can see a rather sharp change of energy when $`\alpha _s`$ changes from 0 to 0.3 and when $`T_g`$ is changing from 0 to 0.25, while further increase of $`\alpha _s`$ or $`T_g`$ does not produce such a strong dependence.
Solutions of our equations (34-35) can be compared with physical states of $`B`$, $`D`$ and $`B_s,D_s`$ mesons. To this end one should have in mind that in Eqs. (34-35) the static approximation for the heavy quark $`b,c`$ was used, and hence all corrections $`O(1/m_Q^n)`$ with $`n1`$ are neglected.
One can exploit at this point the HQET expansion for the mass $`m_H`$ of heavyโlight boson
$$m_H=m_Q(1+\frac{\overline{\mathrm{\Lambda }}}{m_Q}+\frac{1}{2m_Q^2}(\lambda _1+d_H\lambda _2)+O(1/m_Q^3),$$
(43)
where $`\lambda _n`$ are free parameters, depending on dynamics, and $`d_H`$ is the hyperfine splitting parameter. It is clear from the preceding that eigenvalues of Eqs. (34-35) yield the value $`\overline{\mathrm{\Lambda }}`$, which depends on the quantum numbers of the state,
$$\overline{\mathrm{\Lambda }}(j,l,n_r)=\epsilon _n(j,l)$$
(44)
Consider now the results of the present approach, i.e. solutions of Diracโtype equations (34-35). In the local case $`(T_g0)`$ when the kernel $`M`$ reduces to the linear potential $`\sigma r`$, we have
$$\overline{\mathrm{\Lambda }}_D^{(loc)}=0.690GeV(\alpha _s=0,\sigma =0.18GeV^2)$$
(45)
and
$$\overline{\mathrm{\Lambda }}_D^{(loc)}=0.493GeV(\alpha _s=0.3,\sigma =0.18GeV^2).$$
(46)
This should be compared to the nonlocal case
$$\overline{\mathrm{\Lambda }}_D^{(nonloc)}=0.415GeV(\alpha _s=0,\sigma =0.18GeV^2)$$
(47)
and
$$\overline{\mathrm{\Lambda }}_D^{(nonloc)}=0.288GeV(\alpha _s=0.3,\sigma =0.18GeV^2).$$
(48)
These latter values are in general agreement with the results of the QCD heavyโflavour sum rules
$$\overline{\mathrm{\Lambda }}=0.57\pm 0.07GeV$$
(49)
and more recent analysis from semileptonic $`B`$ decays
$$\overline{\mathrm{\Lambda }}=0.39\pm 0.11GeV$$
(50)
Another interesting comparison is with the experimental values of the $`B`$โmeson mass (the term $`\lambda _2`$ in Eq. (43) can be determined from the $`B^{}B`$ mass difference). Using Eq. (48) and $`\overline{M}_B=\frac{3M_B^{}+M_B}{4}=5.312GeV`$ one can estimate (neglecting $`\lambda _1)`$ the pole mass of the $`b`$โquark to be $`m_b(pole)5.0GeV`$, which is in reasonable agreement with the analysis of the quarkonium spectra in .
A similar analysis can be done for the $`B_s`$ meson; the corresponding values for $`\overline{\mathrm{\Lambda }}_s`$ with $`m_s=0.15`$ and 0.20 $`GeV`$, are $`\overline{\mathrm{\Lambda }}_s\overline{\mathrm{\Lambda }}=0.084`$ and $`\overline{\mathrm{\Lambda }}_s\overline{\mathrm{\Lambda }}=0.115`$ for $`\alpha _s=0.3`$ . One can compare these values with the mass difference $`B_s,B^0,\mathrm{\Delta }M_s(B)=(0.090\pm 0.0038)GeV`$. These numbers for $`\overline{\mathrm{\Lambda }}`$ can be compared with those in Table 3, where also results of lattice calculations and of the constituent quark model (CQM) are given.
## IV Chiral condensate
As a check of CSB in our Eqs. (16-17) we have computed the chiral condensate, which can be expressed through the eigenfunctions as in (to simplify matter we disregard in this section perturbative contributions).
$$\overline{q}q=\frac{N_c}{2\pi }\underset{n=0}{\overset{\mathrm{}}{}}[(A_n^{})^2(B_n^+)^2],$$
(51)
where $`A_n^{}=(\frac{G_n(r)}{r})_{r=0},B_n^+=(\frac{F_n(r)}{r})_{r=0}`$, and $`G_n,F_n`$ refer to solutions with $`\kappa =1,l=0`$ and $`\kappa =+1,l=1`$ respectively. In the local linear potential case the values of $`A_n^{},B_n^+`$ have been computed in the WKB method and shown to yield a monotonically divergent series $`\overline{q}q=\frac{N_c}{2\pi }_n\frac{const}{\sqrt{n}}`$.
It can be argued (using Eqs. (8-9) from ), that the nonlocality of the kernel $`M`$ in spaceโtime, present by definition in Eq. (8) improves the convergence of the series and yields a finite result for $`\overline{q}q`$. We have found $`A_n^{},B_n^+`$ from the solutions of the nonlocal equations (16-17) with the kernel $`\stackrel{~}{M}`$ and compared them with the local case, when $`\stackrel{~}{M}`$ reduces to the local linear potential. Results are shown in Table 4.
One can see from the results, that in the nonlocal case the magnitude of $`s_n(A_n^{})^2(B_n^+)^2`$, is clearly diminished as compared to the reference local case, and is of reasonable order of magnitude. From the obtained sequence of $`s_n`$ we get that $`\overline{q}q=0.5\sigma ^{3/2}`$ and $`0.7\sigma ^{3/2}`$ in the nonlocal cases of the kernels (21) and (22) respectively. Adopting a value of $`\sigma =0.2GeV^2`$ we find $`\overline{q}q=(350MeV)^3`$ and $`(400MeV)^3`$ respectively, to be compared with the usually acceptable value of $`(250MeV)^3`$. However convergence is still slow as seen from Table 4 and the converged values are somewhat higher. We have checked that convergence is somewhat improved when one takes into account the intrinsic nonlocality of the kernel $`M`$ in $`๐ฑ,๐ฒ`$. To this end we have modified the kernel $`\stackrel{~}{M}`$ obtained from WKB analysis, replacing $`\stackrel{~}{\delta }`$ in Eq. (21) by a Gaussian factor
$$Nexp(\frac{(xy)^2}{a^2})\delta (cos\theta _xcos\theta _y)$$
(52)
and studied the sequences of $`s_n`$ as functions of the nonlocality range $`a`$. Results are shown in Table 4 for two values of $`a=0.3\sigma ^{1/2}`$ and $`0.5\sigma ^{1/2}`$. The strength $`N`$ is chosen such that numerically the slope of $`\sigma =0.2GeV^2`$ is reproduced for large distances. The condensate values varies in the considered region substantially, showing that effects of lonlocality are important.
The slope of the effective potential $`V_{eff}(r)`$, determined by $`N`$, strongly depends on $`a`$ for $`aT_g`$. We believe that the reason for this lies in the fact, that the chiral condensate $`\overline{q}q`$ depends crucially on the nonlocality both in time components of $`M(๐ฑ,๐ฒ;x_4,y_4)`$ and in spacial components. The first nonlocality was however disregarded in Eqs. (13-14), when the $`p_4`$ โ dependence was omitted in $`M`$ and $`\psi _n`$ (the static limit). It was indeed shown in Ref. , that taking this dependence into account significantly improves convergence of the sum in Eq. (51). The full account of this effect requires solution of timeโdependent Eqs. (11-12), which is numerically a much more difficult problem.
## V Conclusion
We have studied the confining and CSB properties in the system of one light quark and one static antiquark. The effective mass operator is written explicitly for large $`N_c`$, as a sum over vacuum field correlators. Keeping only the Gaussian field correlator, we have obtained a closed system of equations in the limit of large $`N_c`$. Our results support the presence of a Lorentz scalar linear confinement for the light quark, which signifies CSB for this system, and yield eigenfunctions and eigenvalues for the heavyโlight system containing both confinement and CSB.
As a direct evidence of CSB we have computed the chiral condensate, which appears to be of the correct sign and having the proper large $`N_c`$ dependence. Our result yields a reasonable order of magnitude of $`\overline{q}q`$, provided convergence of the sum is achieved. At this point it is useful to compare the CSB picture of the NJL model and our approach. In the NJL model confinement and string are absent and CSB may occur due to the condensation of $`q\overline{q}`$ pairs in the scalar channel. In our case, being the large $`N_c`$ approximation of the real QCD, a string is built up between light and heavy quark, which depends not only on light quark coordinates $`๐ฑ,๐ฒ`$, but also on the distance from them to the heavy antiquark. In the presence of confinement, the phenomenon of CSB is due to the spontaneous creation of the scalar string, which is forbidden by chiral symmetry.
Eigenvalues $`\epsilon _n`$ and eigenfunctions obtained numerically for lowest states, represent the leading contributions of the HQET expansion in powers of $`1/m_Q`$. Results for the energies $`\epsilon _n`$ in our method are compared of the lattice and QCD sum rule calculations, and also with experimental extraction of $`\epsilon _n=\overline{\mathrm{\Lambda }}(n)`$, showing an overall agreement with the B and D meson masses.
Acknowledgement
This work was started while one of the authors (Yu.S.) was a visitor at ITP Utrecht. The hospitality of the Institute and financial support by FOM are gratefully acknowledged. Yu.S. is grateful to I.M.Narodetsky for a useful discussion.
Table 1. Energy eigenvalues for Eqs. (13-14) with $`J=\frac{1}{2}`$ and the kernels, given by Eq. (25) and Eq. (27), as compared with those of the Dirac equation for the linear local potential $`V_{lin}(r)=r`$.
| | $`V=r`$ | | kernel (25) | | kernel (27) | |
| --- | --- | --- | --- | --- | --- | --- |
| n | $`\kappa =1`$ | $`\kappa =1`$ | $`\kappa =1`$ | $`\kappa =1`$ | $`\kappa =1`$ | $`\kappa =1`$ |
| 0 | 1.619 | 2.294 | 0.925 | 1.472 | 0.969 | 1.516 |
| 1 | 2.603 | 3.031 | 1.719 | 2.056 | 1.765 | 2.113 |
| 2 | 3.291 | 3.626 | 2.277 | 2.541 | 2.334 | 2.608 |
| 3 | 3.855 | 4.138 | 2.740 | 2.964 | 2.809 | 3.042 |
| 4 | 4.345 | 4.594 | 3.144 | 3.342 | 3.226 | 3.432 |
| 5 | 4.784 | 5.008 | 3.507 | 3.685 | 3.603 | 3.790 |
| 6 | 5.186 | 5.334 | 3.838 | 4.002 | 3.950 | 4.123 |
Table 2. Ground state energy eigenvalue (in units of $`\sqrt{\sigma }`$) for Eqs. (34-35) with $`\alpha _s=0`$, 0.3 and 0.39 and quark masses $`m=5MeV`$, $`0.15GeV`$ and $`0.2GeV`$ (upper, middle and lower entry) for different values of $`T_g`$, $`T_g=0`$, 0.25, 0.5 and 1 (in units of $`1/\sqrt{\sigma }`$).
| | $`T_g`$ | 0 | 0.25 | 0.5 | 1 |
| --- | --- | --- | --- | --- | --- |
| $`\alpha _s`$ | | | | | |
| | | 1.628 | 0.985 | 0.979 | 0.907 |
| 0 | | 1.886 | 1.225 | 1.217 | 1.145 |
| | | 1.978 | 1.314 | 1.305 | 1.233 |
| | | 1.163 | 0.684 | 0.679 | 0.628 |
| 0.3 | | 1.378 | 0.884 | 0.877 | 0.826 |
| | | 1.456 | 0.959 | 0.951 | 0.900 |
| | | 1.004 | 0.585 | 0.580 | 0.536 |
| 0.39 | | 1.201 | 0.768 | 0.761 | 0.717 |
| | | 1.272 | 0.837 | 0.830 | 0.786 |
Table 3. Energy eigenvalues $`\overline{\mathrm{\Lambda }}`$ of the heavyโlight system in the static heavy quark approximation obtained in different approaches.
| Refs. | Method | $`\overline{\mathrm{\Lambda }}(GeV)`$ |
| --- | --- | --- |
| 20 | QCD sum rules | $`>0.5`$ |
| 21 | QCD sum rules | $`0.4รท0.5`$ |
| 24 | Lattice | $`0.18\pm 0.03`$ |
| 22 | experim. | $`0.33\pm 0.11`$ |
| 25 | QCM | $`0.35`$ |
| 26 | QCM | $`0.5รท0.6`$ |
| 27 | Rel. QCM | $`0.386`$ |
| this work | Nonlin. Dirac Eq. | $`0.287`$ |
Table 4. The difference $`s_n=|A_n|^2|B_n|^2`$ for $`n=0,1,\mathrm{}6`$, in case of the nonlocal kernels (25) and (27) and corrected for a normalized Gaussian nonlocality (52) with a range of $`a`$ =0.3 and $`a`$=0.5. For comparison the results are shown for a local linear potential $`V_{lin}=r`$ .
| | $`A_n^2B_n^2`$ | | | | |
| --- | --- | --- | --- | --- | --- |
| n | $`V=r`$ | kernel (25) | kernel (27) | $`a`$=0.3 | $`a`$=0.5 |
| 0 | 0.79 | 0.42 | 0.50 | 0.15 | 0.23 |
| 1 | 0.51 | 0.21 | 0.34 | 0.04 | 0.12 |
| 2 | 0.41 | 0.12 | 0.19 | 0.02 | 0.10 |
| 3 | 0.35 | 0.10 | 0.16 | 0.02 | 0.09 |
| 4 | 0.31 | 0.08 | 0.11 | 0.01 | 0.09 |
| 5 | 0.27 | 0.07 | 0.11 | 0.01 | 0.08 |
| 6 | 0.26 | 0.06 | 0.09 | 0 | 0.07 | |
warning/0001/math-ph0001037.html | ar5iv | text | # Abel ODEs: Equivalence and Integrable Classes
## 1 Introduction
From some point of view, after Riccati type ODEs, the simplest first order ordinary differential equations (ODEs) are those having as right hand side (RHS) a third degree polynomial in the dependent variable, also called Abel type ODEs<sup>1</sup><sup>1</sup>1For convenience, in this work, by โAbel ODEsโ we mean Abel ODEs of first kind, since Abel ODEs of second kind can always be transformed into first kind by a simple change of variables.
$$y^{}=f_3y^3+f_2y^2+f_1y+f_0$$
(1)
where $`yy(x)`$, and $`f_0`$, $`f_1`$, $`f_2`$ and $`f_3`$ are analytic functions of $`x`$. As opposed to Riccati ODEs, for which integration strategies can be built around their equivalence to second order linear ODEs, Abel ODEs admit just a few available integration strategies, most of them based on the pioneering works by Abel, Liouville and Appell around 100 years ago . In those works it was shown that Abel ODEs can be organized into equivalence classes. Two Abel ODEs are defined to be equivalent if one can be obtained from the other through the transformation
$$\{x=F(t),y(x)=P(t)u(t)+Q(t)\}$$
(2)
where $`t`$ and $`u(t)`$ are respectively the new independent and dependent variables, and $`F`$, $`P`$ and $`Q`$ are arbitrary functions of $`t`$ satisfying $`F^{}0`$ and $`P0`$.
Integration strategies were then discussed in , around objects invariant under Eq.(2)<sup>2</sup><sup>2</sup>2The invariants change in form for $`F(t)t`$, but keep their value. See Eq.(5). (herein called the invariants) which can be built with the coefficients $`\{f_3,f_2,f_1,f_0\}`$ and their derivatives. To each class there corresponds a different set of values of these invariants, and actually any one of them (we shall pick one and call it the invariant) is enough to characterize a class. A simple integrable case happens when the invariant is constant<sup>3</sup><sup>3</sup>3There exists one invariant such that if it is constant then the other invariants are as well.; the solution to the ODE then follows straightforwardly in terms of quadratures, as explained in textbooks . On the contrary, when the invariant is not constant, just a few integrable cases are known and the formulation of solving strategies based on the equivalence between two such Abel ODEs, one of which is integrable, appears to be only partially explored in the literature, and not explored in general in computer algebra systems.
Having this in mind, this paper concerns Abel ODEs with non-constant invariant and presents:
1. A discussion and classification of the integrable Abel ODEs found both in Kamkeโs book and in the works from the late 19<sup>th</sup> and early 20<sup>th</sup> century by Abel, Appell, Liouville and other sources;
2. A set of new integrable Abel ODE classes - some depending on arbitrary parameters - derived from those aforementioned works;
3. An explicit method of verifying or denying the equivalence between two given Abel ODEs, one of which we know how to solve since it represents one of the above mentioned classes; and in the positive case, a way to determine the function parameters $`F`$, $`P`$ and $`Q`$ of the transformation Eq.(2) which maps one into the other;
4. A computational scheme to resolve the equivalence problem in the case of parameterized classes, including the determination of the value for the class parameter when the equivalence exists;
5. A set of computer algebra (Maple) routines implementing the algorithms presented in items (3) and (4) above, to systematically solve - in principle - any Abel ODE belonging to one of the classes, parameterized or not, presented here and for which a closed form solution is known (items (1) and (2) above).
Item (1) is interesting since the Abel ODEs shown in textbooks in general, including Kamkeโs book, are displayed without further classification, and in fact many of them belong to the same class. This classification in terms of invariant theory is also necessary in a computational scheme for solving Abel ODEs as the one being presented, and we have not found it in other references. The integrable classes mentioned in (2) are new to the best of our knowledge, although directly or indirectly derived from previous works. The formulation of the equivalence problem mentioned in (3) is the one given by Liouville in , is systematic and does not involve solving any auxiliary differential equations<sup>4</sup><sup>4</sup>4An approach somewhat similar to this one by Liouville is discussed in .. Concerning item (4), the idea can be viewed as a way of avoiding the untractable expressions which one would encounter when making direct use of Liouvilleโs strategy with parameterized classes. The strategy presented is applicable when there exists a solution for some numerical values of the parameter, or when this parameter is a rational function of other symbols entering the input ODE. Regarding item (5), the implementation presented here is, as far as we know, unique in computer algebra systems in its ability to solve non-constant invariant, parameterized or not, Abel ODE classes.
The paper is organized as follows. In sec. 2, the basic definitions and the classic formulation of the equivalence problem in terms of invariants is reviewed and shown to apply straightforwardly to the case of a non-parameterized class. In sec. 3 it is shown how these ideas can be complemented by taking advantage of computers to tackle the equivalence problem in the case of a parameterized class. Section 4 presents a classification of the integrable classes we have found in the literature with some additional comments as to their derivation. In sec. 5 new integrable Abel classes are presented. In sec. 6, a test-suite for the routines presented is discussed and statistics are shown describing the performance with this test-suite as well as with Kamkeโs first order examples. Finally, the conclusions contain some general remarks about this work and its possible extensions.
Additionally, we present in the Appendix a table listing the distinct Abel ODE classes that we have found, representative ODEs from each class, and their respective solutions.
## 2 Classical Theory for Abel ODEs
In general, Abel type ODEs can be studied using two related concepts: invariants and ODE equivalence classes. We define two Abel ODEs to be equivalent<sup>5</sup><sup>5</sup>5For a more formal definition of class see if one can be obtained from the other using a transformation of the form Eq.(2). The equivalence class containing a given ODE is then the set of all the ODEs equivalent to the given one. We note that although the infinitely many members of a class can be mapped into each other by using Eq.(2), there are also infinitely many disjoint Abel classes (Eq.(2) is not sufficient to map any Abel ODE into a given one).
To each class one can associate an infinite sequence of absolute invariants . To see this, consider two Abel ODEs, the first Eq.(1), the second obtained from Eq.(1) through the transformation Eq.(2)
$$u^{}=\stackrel{~}{f_3}u^3+\stackrel{~}{f_2}u^2+\stackrel{~}{f_1}u+\stackrel{~}{f_0}$$
(3)
where the coefficients $`\{\stackrel{~}{f_0},\stackrel{~}{f_1},\stackrel{~}{f_2},\stackrel{~}{f_3}\}`$, are related to the those of Eq.(1) by
$`\stackrel{~}{f_0}`$ $`=`$ $`{\displaystyle \frac{F^{^{}}\left(f_0(F)+f_1(F)Q+f_2(F)Q^2+f_3(F)Q^3\right)Q^{^{}}}{P}}`$
$`\stackrel{~}{f_1}`$ $`=`$ $`{\displaystyle \frac{P^{^{}}}{P}}F^{^{}}\left(f_1(F)+2f_2(F)Q+3f_3(F)Q^2\right)`$
$`\stackrel{~}{f_2}`$ $`=`$ $`PF^{^{}}\left(f_2(F)+3f_3(F)Q\right)`$
$`\stackrel{~}{f_3}`$ $`=`$ $`P^2F^{^{}}f_3(F)`$ (4)
Following , we call an absolute invariant of Eq.(1) a function $`I(f,x)`$ of the coefficients $`\{f_0,f_1,f_2,f_3\}`$ and their derivatives with respect to $`x`$ such that, for all $`\{F,P,Q\}`$ in Eq.(2),
$$I(\stackrel{~}{f},t)|_{\stackrel{~}{f}=\stackrel{~}{f}(f,t)}=I(f,x)|_{x=F(t)}$$
(5)
where $`\stackrel{~}{f}=\stackrel{~}{f}(f,t)`$ represents the coefficients $`\{\stackrel{~}{f_0}`$, $`\stackrel{~}{f_1}`$, $`\stackrel{~}{f_2}`$, $`\stackrel{~}{f_3}\}`$ and their derivatives with respect to $`t`$, expressed using Eq.(4).
Similarly, we call a relative invariant a function, say $`s`$, of the coefficients of Eq.(1) and their derivatives such that when changing variables using Eq.(2), the resulting expression is equal to the original one up to a factor, say $`\phi _s`$, dependent uniquely on the functions $`F`$, $`P`$, and $`Q`$ in Eq.(2) and independent of the coefficients themselves :
$$s(\stackrel{~}{f})|_{\stackrel{~}{f}=\stackrel{~}{f}(f,t)}=\phi _s(F,P,Q)s(f)|_{x=F(t)}$$
(6)
Liouville showed that in the case of Abel equations there is a relative invariant of weight 3
$$s_3f_0f_3^2+\frac{1}{3}\left(\frac{2f_2^3}{9}f_1f_2f_3+f_3f_2^{}f_2f_3^{}\right)$$
(7)
which can be used to recursively generate an infinite sequence of relative invariants $`s_{2m+1}`$ of weights $`2m+1`$ respectively<sup>6</sup><sup>6</sup>6In the case of $`s_3`$, $`\phi _{s_3}=(F^{}P)^3`$; the weight $`n`$ refers to the degree of $`\phi _{s_n}`$ with respect to $`(F^{}P)`$., through the formula
$$s_{2m+1}f_3s_{2m1}^{}\left(2m1\right)s_{2m1}\left(f_3^{^{}}+f_1f_3\frac{f_{2}^{}{}_{}{}^{2}}{3}\right)$$
(8)
As is clear from this definition, the product of two relative invariants respectively of weights $`n`$ and $`m`$ is a relative invariant of weight $`n+m`$, and by dividing any two relative invariants of equal weight one can generate an infinite sequence of absolute invariants
$$I_1=\frac{s_5^3}{s_3^5},I_2=\frac{s_7s_3}{s_5^2},I_3=\frac{s_9}{s_3^3},\text{etcโฆ}$$
(9)
In , Appell showed that this sequence can also be obtained from two basic absolute invariants - say $`J_1,J_0`$, by expressing $`J_1`$ as a function of $`J_0`$ and then differentiating the result with respect to $`J_0`$. As a consequence, if $`I_1`$ is constant then all the other ones will be too. This fact allows one to identify the constant character of the invariants in Eq.(9) by looking at just the first one. We note also that there are infinitely many different classes having $`I_1`$ constant, related to the infinitely many possible constant values $`I_1`$ can have.
### 2.1 Integration strategy
A description of a method of integration when the invariants are constant<sup>7</sup><sup>7</sup>7In it is also shown that in the constant invariant case the problem can also be formulated in terms of the symmetries of these ODEs. is found in the works by Abel , Liouville and Appell . In such a case, all members of the class can be systematically mapped into a separable first order ODE (representative of the class), by appropriately choosing $`F`$, $`P`$ and $`Q`$ in Eq.(2); see for instance and .
A quite different situation happens when $`I_1`$ is not constant. In such a case, relatively few integrable Abel ODEs are known, and the integration methods used to solve each of them depend in an essential way on non-invariant properties of the coefficients $`f`$. Those methods are then useless for solving the other infinitely many members of the same classes, unless one can solve the related equivalence problems; i.e., determining - when they exist -the values of $`F`$, $`P`$ and $`Q`$ in Eq.(2) linking two Abel ODE which belong to the same class.
### 2.2 Identifying an ODE as member of a given Abel ODE class
Consider two Abel ODEs; the first one given by Eq.(1), and a second one being of the same form, but with coefficients $`\stackrel{~}{f_0}`$, $`\stackrel{~}{f_1}`$, $`\stackrel{~}{f_2}`$ and $`\stackrel{~}{f_3}`$. The problem now is to determine whether the second Abel ODE can be obtained from Eq.(1) by changing variables using Eq.(2). This problem can be formulated by equating the coefficients between the transformed equation, obtained by applying the transformation Eq.(2) to Eq.(1), and the second Abel ODE, resulting in Eq.(4), which can be seen as an ODE system for $`\{F,P,Q\}`$. To solve this system, following Liouville , we first note that the absolute invariants corresponding to the two Abel ODEs donโt depend on $`P`$ or $`Q`$ (see previous section). Hence the function $`F`$ entering Eq.(2) can be obtained by just running an elimination process using two of these absolute invariants, for instance $`I_1=s_5^3/s_3^5`$ and $`I_2=s_5s_7/s_3^4`$:
$$0=\frac{\stackrel{~}{s_5}^3}{\stackrel{~}{s_3}^5}\frac{s_5^3}{s_3^5}|_{x=F(t)}0=\frac{\stackrel{~}{s_5}\stackrel{~}{s_7}}{\stackrel{~}{s_3}^4}\frac{s_5s_7}{s_3^4}|_{x=F(t)}$$
(10)
As discussed in , the existence of a common solution $`F(t)`$ to both equations above (such that $`F^{}0`$) is the necessary and sufficient condition for the existence of a transformation Eq.(2) relating the two Abel ODEs. Once $`F`$ is known, the system Eq.(4) becomes trivial in that $`P`$ and $`Q`$ can be re-expressed in terms of $`F`$ by performing fairly simple calculations. In the case of interest of this work - non-constant invariant<sup>8</sup><sup>8</sup>8When the invariant is non-constant, $`s_30`$. \- the resulting expressions are:
$$P(t)=\frac{F^{}\stackrel{~}{f_3}^2s_3}{f_{3}^{}{}_{}{}^{2}\stackrel{~}{s_3}}|_{x=F(t)}Q(t)=\frac{F^{}\stackrel{~}{f_2}\stackrel{~}{f_3}s_3f_2f_3\stackrel{~}{s_3}}{3f_{3}^{}{}_{}{}^{2}\stackrel{~}{s_3}}|_{x=F(t)}$$
(11)
where $`\{f_i,\stackrel{~}{f}_i\}`$ with $`i:03`$ are the coefficients of the two Abel equations, $`s_3`$ is the relative invariant Eq.(7) expressed in terms of $`f_i`$ and $`\stackrel{~}{s_3}=s_3|_{f_i=\stackrel{~}{f_i}}`$.
Concerning the explicit solution $`F(t)`$ for Eq.(10), we note that our interest in solving the equivalence problem is in that it leads directly to the solution of other members of an Abel class, when the solution to a representative of the class is known. In turn, all the solvable classes we are aware of have a representative with rational coefficients (see the Appendix), and hence also rational invariants<sup>9</sup><sup>9</sup>9On the other hand, there is no reason to expect that the second of the two Abel ODEs being tested for equivalence also has rational invariants. If however the invariants of both Abel ODEs are rational in $`t`$ and the coefficients are numbers, then it is also possible to determine $`F(t)`$ by performing a rational function decomposition as mentioned in .. Hence, assuming that one of the two Abel ODEs has rational coefficients and that Eq.(10) was obtained by applying Eq.(2) to it, the system Eq.(10) will always be polynomial in $`F(t)`$. In such a case, when a common solution $`F(t)`$ to both equations exists, the resultant between these polynomials will be zero ; i.e.: there will be a common factor, depending on $`F`$ and $`t`$ and representing the common solution, which can be obtained by calculating the greatest common divisor (GCD) between the two equations in Eq.(10). Conversely, if that GCD does not depend on $`F`$, a transformation Eq.(2) linking the input equation to Eq.(1) does not exist. That the dependence on $`F`$ of this GCD is a necessary condition for the existence of the desired transformation Eq.(2) is a consequence of the validity of Eq.(5) and hence the system (10). A proof of its sufficiency was given by Appell in .
The whole process just described to determine the equivalence between two given Abel ODEs, one of which is rational in $`x`$, can be summarized as follows:
1. Calculate two absolute invariants, set up the system Eq.(10), and calculate the GCD between the two equations;
2. When this GCD does not depend on $`F`$, the ODEs donโt belong to the same class; otherwise determine an explicit expression for $`F(t)`$ from the result of the GCD calculation;
3. Plug this value for $`F`$ into the formulas Eq.(11) to determine the values of $`P(t)`$ and $`Q(t)`$, arriving in this way at the transformation Eq.(2) mapping one Abel ODE into the other.
Example:
Consider the two non-constant invariant Abel ODEs
$$y^{}=\frac{1}{2(x+4)}\left(xy^3+y^2\right)$$
(12)
$$y^{}=\frac{\left(f^{}xf\right)}{2(f+3x)}\left(\left(xf\right)y^3+y^2\right)\frac{y}{x}$$
(13)
where in the above $`ff(x)`$ is an analytic (arbitrary) function. As in the typical situation one of these ODEs has invariants rational in $`x`$ and we know its solution; i.e. for Eq.(12) we have
$$C_1+\frac{\sqrt{y^2x4y1}}{y}+2\mathrm{arctan}\left(\frac{1+2y}{\sqrt{y^2x4y1}}\right)=0$$
(14)
where $`C_1`$ is an arbitrary constant. We would then like to determine whether there are functions $`\{F,P,Q\}`$ so that Eq.(12) transforms under Eq.(2) into Eq.(13), and if so, determine $`\{F,P,Q\}`$ and use them together with Eq.(14) to build the answer to Eq.(13). For this purpose, we start (step (1)) by computing the relative invariants $`s_3,s_5,s_7`$, leading to Eq.(10)
$$\begin{array}{ccc}0\hfill & =& \frac{\left(87ft+9f^2+184t^2\right)^3}{t\left(9f+31t\right)^5}\frac{\left(280+105F+9F^2\right)^3}{\left(40+9F\right)^5}\hfill \\ 0\hfill & =& \frac{\left(81f^3+1431f^2t+7185ft^2+10903t^3\right)\left(9f+31t\right)}{\left(87ft+9f^2+184t^2\right)^2}\frac{\left(1674F^2+10290F+81F^3+19600\right)\left(40+9F\right)}{\left(280+105F+9F^2\right)^2}\hfill \end{array}$$
(15)
where in the above $`FF(t)`$ is the function we are looking for and $`f`$ is taken at $`x=t`$. We recall that when Eq.(13) has non-constant invariant the denominators in above are not zero since $`s_30`$. Calculating the GCD between the numerators of the expressions above (step (2)) and equating this GCD to zero, we obtain<sup>10</sup><sup>10</sup>10We note that when $`f(t)`$ is an algebraic mapping involving varied analytic functions, Mapleโs procedures to simplify and put them in normal form may fail and consequently the GCD computation may not be successful.
$$27(t+tFf)=0$$
(16)
from where the common solution $`F(t)`$ to both equations is given by
$$F(t)=\frac{f(t)}{t}1$$
(17)
Substituting this value of $`F`$ into Eq.(11), a transformation of the form Eq.(2) mapping Eq.(12) into Eq.(13) is finally given by
$$\{x=\frac{f(t)}{t}1,y(x)=tu(t)\}$$
(18)
from where by changing variables in Eq.(14) using the transformation above and renaming the variables ($`tx,uy`$), the solution to Eq.(13) is obtained
$`C_1+{\displaystyle \frac{\sqrt{\left(\frac{f}{x}1\right)x^2y^24xy1}}{xy}}+2\mathrm{arctan}\left({\displaystyle \frac{\left(1+2xy\right)}{\sqrt{\left(\frac{f}{x}1\right)x^2y^24xy1}}}\right)=0`$ (19)
## 3 Parameterized Abel ODE classes
We formulate here the equivalence problem in the case of parameterized classes. By โparameterized classโ we mean an (Abel) ODE class depending on symbolic parameters which cannot be removed by changing variables using Eq.(2). The interest in parameterized solvable classes is clear: to each set of values of the parameters corresponds - roughly speaking - a different Abel class<sup>11</sup><sup>11</sup>11There may be particular different sets of values for which the resulting ODEs will nevertheless belong to the same class.. Hence, a formulation of the equivalence problem for parameterized classes enables one to solve all the members of infinitely many classes at once.
In order to simplify the discussion, we consider the problem of an Abel ODE class depending on just one parameter<sup>12</sup><sup>12</sup>12The integrable classes presented in the literature depend at most on one parameter (see sec. 4)., say $`๐`$. Also, we distinguish between two different types of problems: one is when the equivalence problem has a solution for a specific numerical value of $`๐`$; the other happens when to have a solution it is required that $`๐`$ assumes symbolic values, for instance in terms of other symbols entering the input ODE. We discuss first the numerical case, and in the next subsection we show how the symbolic case can be mapped into many numerical problems - when the parameter depends on other symbols in a rational manner - by using rational interpolation methods.
### 3.1 Solution for some numerical value of $`๐`$
To facilitate the exposition we present the discussion around a concrete example. Consider the equivalence problem between a given Abel ODE, for instance,
$$y^{}=8\frac{\left(1x^4x^8\right)y^3}{x^7}+4\frac{y^2}{x^4}+\frac{y}{x}$$
(20)
and the one presented in Abelโs memoires
$$y^{}=\frac{\left(๐x^4+x^2+1\right)y^3}{x^3}+y^2$$
(21)
If this equivalence exists, then it exists for a specific value of the parameter $`๐`$ since there is no solution for arbitrary $`๐`$ (the existence of such a solution would mean the class does not really depend on any parameter). Hence, the common solution $`F(t)`$ to the system Eq.(10) will not show up until the correct value of $`๐`$ is determined<sup>13</sup><sup>13</sup>13There may be more than one solution $`๐`$., invalidating the itemized algorithm of the previous section.
A natural alternative to this problem would be to take one more absolute invariant, for instance, $`s_3s_7/s_5^2`$, so that our system Eq.(10) becomes
$$0=\frac{\stackrel{~}{s_5}^3}{\stackrel{~}{s_3}^5}\frac{s_5^3}{s_3^5}|_{x=F(t)}0=\frac{\stackrel{~}{s_3}\stackrel{~}{s_7}}{\stackrel{~}{s_5}^2}\frac{s_3s_7}{s_5^2}|_{x=F(t)}0=\frac{\stackrel{~}{s_5}\stackrel{~}{s_7}}{\stackrel{~}{s_3}^4}\frac{s_5s_7}{s_3^4}|_{x=F(t)}$$
(22)
and search for a solution to this problem such that $`F^{}0,๐^{}=0`$. For that purpose, eliminate $`๐`$ from the first and second expressions above by taking the resultant with respect to $`๐`$, obtaining \- say - $`R_1`$. In the same way, eliminate $`๐`$ from the first and the third expressions of Eq.(22) obtaining $`R_2`$. Hence, when a solution exists, the resultant between $`R_1`$ and $`R_2`$ with respect to $`F`$ will vanish. In other words, the algorithm of the previous section will work if instead of performing the calculations over the expressions Eq.(10) we perform them over $`R_1`$ and $`R_2`$. The GCD between $`R_1`$ and $`R_2`$ will then return the factor depending on both $`F`$ and $`t`$, whose solution is the function $`F(t)`$ we are interested in.
This method, simple and correct in theory, unfortunately does not work in practice because the expressions tend to grow in size so much that the computation of the first of these three resultants may not be possible, even with a simple example such as the one shown above. The problem resides in the fact that multivariate GCDs and resultants are quite expensive operations for the current symbolic computation environments.
An alternative to this problem consists of reducing it to a sequence of bivariate GCD and resultant calculations, for which the available algorithms are relatively fast. The idea can be summarized as follows.
1. From the previous considerations, when a solution to the equivalence problem exists, the resultant between any two of the expressions in Eq.(22) will not vanish for any value of $`t`$, since we havenโt introduced the correct (unknown at this point) value of $`๐`$. Hence, if we insert in Eq.(22) a numerical value<sup>14</sup><sup>14</sup>14We note there may exist โinvalid evaluation pointsโ; roughly speaking to avoid this problem this evaluation point must not cancel any of the coefficients of the variables remaining in the system - see . for $`t`$ and calculate the GCD between any two of the resulting expressions, this GCD cannot contain any factor depending on $`F`$. This gives us a first โexistence conditionโ test for the solution before proceeding further;
2. When Eq.(22) evaluated at $`t=\text{number}`$ passed the test of the previous step, take two of the resulting three expressions and calculate their resultant with respect to $`F`$, obtaining, say, $`\stackrel{~}{R}_1`$. Then take a different pair and calculate their resultant with respect to $`F`$ again, obtaining, say, $`\stackrel{~}{R}_2`$. Neither of these resultants will vanish since the GCD calculations of the previous step showed no factor depending on $`F`$. Also, the calculation of $`\stackrel{~}{R}_1`$ and $`\stackrel{~}{R}_2`$ is now quite simpler since the expressions do not involve $`t`$;
3. Then if a solution to the problem exists, the GCD between $`\stackrel{~}{R}_1`$ and $`\stackrel{~}{R}_2`$ will yield a factor depending on $`๐`$; equating it to zero and solving it for $`๐`$ will give the common solution $`๐`$ for $`\stackrel{~}{R}_1`$ and $`\stackrel{~}{R}_2`$. More precisely, what we will get in this way is a set of candidates (including among them the correct value) for $`๐`$; not all of them will necessarily lead to a solution $`F(t)`$ to the original problem;
4. We now plug these candidates for $`๐`$ into Eq.(22), one at a time, receiving a system of three expressions involving again only two unknowns, now $`F`$ and $`t`$. If there is a common solution $`F(t)`$ to these expressions, the resultant with respect to F between any two of them will vanish. Hence, the GCD between those two expressions will contain a factor depending both on $`F`$ and $`t`$; equating this factor to zero and solving for $`F`$ leads to the solution $`F(t)`$.
Returning to our example of determining the equivalence between Eq.(20) and Eq.(21), the itemized procedure just outlined runs as follows.
According to step (1), $`t=0`$ is tried first, but it is found to be an invalid evaluation point. The next value of $`t`$ to try, $`t=1`$ turns out to be valid, so Eq.(22) was evaluated at $`t=1`$; the GCDs between any two of the three resulting expressions do not depend on $`F`$, so this first test for the โexistenceโ of a solution passed.
Continuing with step (2), the calculation of $`\stackrel{~}{R}_1`$ and $`\stackrel{~}{R}_2`$ is performed without problems concerning the size of the expressions.
The GCD of step (3) results in the three factors: $`36๐5`$, $`๐+1`$ and $`9๐2`$; equating them to zero and solving them for $`๐`$ we arrive at three candidates for $`๐`$.
In step (4), plugging each of these candidates one at a time into Eq.(22) and taking the GCD between two of the three resulting expressions we note that $`๐=5/36`$ does not lead to any factor depending on both $`F`$ and $`t`$, but $`๐=1`$ leads to such factor: $`F^2t^41`$. So that for $`๐=1`$ the problem admits two solutions: $`F=\pm 1/t^2`$.
Finally, by introducing $`F=1/t^2`$ into the formulas for $`P`$ and $`Q`$ Eq.(11) we arrive at the transformation of the form Eq.(2) mapping Eq.(21) into Eq.(20)
$$\left\{x=\frac{1}{t^2},y=2\frac{u(t)}{t}\right\}$$
(23)
and hence by applying the same change of variables to the answer of Eq.(21) and substituting $`C=1`$ we obtain the answer to Eq.(20).
A remark however is in order: if, in step (3) of the algorithm just described, the numeric candidates for $`๐`$ involve fractional powers of rational numbers, the Maple system may then enter not efficient expensive computations, exhausting the system resources before determining the expression $`F(t)`$ solving the problem in step (4). The root of this limitation seems to be in the absence in Maple of built-in normalization for such โnumeric radicalsโ<sup>15</sup><sup>15</sup>15A built-in normalization of radicals is implemented in the computer algebra system โMathematicaโ. For typical problems we tried where Maple exhausted the system resources trying to determine the solution $`F(t)`$, we exported the mathematical expressions involving radicals to Mathematica and noticed that $`F(t)`$ was determined in this other computer algebra system in reasonable time..
### 3.2 Solution when the parameter $`๐`$ is some rational function of other symbols
When $`๐`$ assumes symbolic values, for instance it depends on other symbols \- say $`\{\alpha \}`$ \- entering the input ODE, if this dependency is rational it is possible to map the determination of $`๐(\alpha )`$ into a sequence of problems having for solution a numerical value of $`๐`$. In turn each of these numerical problems can be tackled using the algorithm of the previous subsection. The idea consists of attributing numerical values to the symbols $`\{\alpha \}`$ entering the invariants in order to determine $`๐(\alpha )`$ by means of a rational interpolation. To simplify the presentation we first discuss the case when $`\{\alpha \}`$ consists of a single parameter, and then show how to extend the algorithm to the case in which $`\{\alpha \}`$ consists of many parameters by commenting on a concrete example. So when $`\alpha `$ consists of just one parameter this rational interpolation scheme is summarized as follows:
1. Take the system Eq.(22) and attribute a numerical value to $`\alpha `$ (check for possible wrong evaluation points), so the resulting system depends on just $`x`$, $`๐`$ and $`F`$;
2. Enter step (1) in the algorithm of the previous subsection and run all the steps:
1. If there is no solution for $`๐`$ and $`F`$ such that $`๐^{}=0`$ and $`F^{}0`$ then quit the process - the input ODE does not belong to this class, or the solution involves a non rational dependency of $`๐`$ on $`\{\alpha \}`$.
2. If however a solution for $`๐`$ and $`F`$ was found, record the values for $`๐`$ and $`\alpha `$ \- they represent a point of a curve $`๐(\alpha )`$;
3. Using the points recorded so far, interpolate $`๐`$ as a function of $`\alpha `$ and test if this interpolated value already solves the problem for an arbitrary $`\alpha `$<sup>16</sup><sup>16</sup>16This is done by restoring the symbol โ$`\alpha `$โ in the system obtained in step (1) and checking if the system is satisfied.:
1. If so, the problem has been solved;
2. Otherwise change the evaluation point of $`\alpha `$ and re-enter step (2) of this enumeration;
We note that the rational interpolation of $`๐(\alpha )`$ requires the knowledge a priori of the polynomial degrees in $`\alpha `$ of both numerator and denominator. That information is not available in advance, but we know that these two degrees sum to $`n1`$, where $`n`$ is the number of points being interpolated. So, when performing the test in step (3) and before going to step (3b) we actually test all possible different interpolations, starting with the maximum possible degree for the numerator and finishing with the maximum possible degree for the denominator.
Concerning the extension of this algorithm for the case when $`\{\alpha \}`$ involves more than one parameter, for instance $`\{a,b\}`$, this extension is easy and better illustrated with an example. Consider the equivalence problem between the Abel ODEs
$$y^{}=\frac{C\left(2x^2C2\right)y^33Cy^2+Cxy}{1x^2C}\text{and}y^{}=\frac{a\left(2x^2a2b\right)y^33ay^2b+abxy}{b\left(bx^2a\right)}$$
(24)
which solution is just the identity $`xx,yy`$ , but only exists when $`๐=\frac{a}{b}`$ (we choose the identity without loss of generality and so that the solving process is easy to follow). We want to determine $`๐=\frac{a}{b}`$ (rational function of two parameters) using the algorithm just described.
We start with step (1), take the system Eq.(22) corresponding to Eq.(24) and attribute numerical values for the first parameter, $`a`$, checking for possible wrong evaluation points - we end up evaluating this system at $`a=1`$. However, the system still depends on the second parameter, $`b`$, so we attribute numerical values to $`b`$ too - check for possible wrong evaluation points - and hence end up evaluating the system altogether at $`a=1,b=2`$. The resulting system now only depends on $`x`$, $`F`$ and $`๐`$.
So we enter step (2) (actually the whole algorithm of sec. 3.1) with this evaluated system of the previous step and detect that $`๐=1/2`$ leads to the solution $`F=x`$; so a solution $`๐^{}=0`$ and $`F^{}0`$ exists.
Hence, in step (3) we interpolate $`๐(b)`$ (so far we have just one point, so $`๐(b)=1/2`$) and test this value for arbitrary $`b`$, verifying that the interpolation is still incomplete: the system Eq.(22) with $`a=1`$ and arbitrary $`b`$ is not satisfied by $`๐=1/2,F=x`$.
We are then in step (3.3b); so attribute a new numerical value to $`b`$, hence evaluate the system Eq.(22) at $`a=1,b=3`$ and re-enter step (2) finding that $`๐=1/3`$ leads to $`F=x`$.
We record this new point (step (2.2b)) and interpolate - now using the two points obtained so far - our first interpolation is of degree 1 in $`b`$: $`๐(b)=(5b)/6`$. This interpolation however does not solve the problem for arbitrary $`b`$. So we increase by one the degree in $`b`$ of the denominator in the interpolation, resulting in $`๐(b)=1/b`$, and verify that this second interpolation indeed solves the system for arbitrary $`b`$ \- so the interpolation for $`b`$ at $`a=1`$ is complete.
We return then to the interpolation of $`๐(a,b)`$ with respect to $`a`$, record that for $`a=1`$ there is a solution $`๐(b)=1/b`$, and test this solution for arbitrary $`a`$ (step (3) with respect to $`a`$), verifying that the interpolation for $`a`$ is still incomplete.
Hence we re-enter step (1) with new evaluation points $`a=2,b=4`$, and re-start the process of determining $`๐(b)`$.
Following the same steps just described, for $`a=2`$ we find that $`๐(b)=2/b`$ solves the problem for arbitrary $`b`$, leading to the second point in the interpolation of $`๐(a,b)`$ with respect to $`a`$.
So we are now in step (3) again, and interpolating $`๐`$ using the two points obtained so far we find $`๐(a,b)=a/b`$ \- this value of $`๐(a,b)`$ is verified to satisfy the system Eq.(22) for arbitrary $`a`$, and thus the problem has been solved.
The algorithm just described, though expensive in computations, is successful in solving the equivalence problem in reasonable time for typical situations (see sec. 6). In this example, for instance, it took 12 seconds to: setup the invariants and the system Eq.(22), simplify this system to a normal form, run all the items of the algorithm just described to determine $`๐(a,b)`$ and $`F(t)`$, then determine $`P(t)`$ and $`Q(t)`$ according to Eq.(11), and finally return a solution to the second of the ODEs in Eq.(24).
#### 3.2.1 Remark on the existence of multiple solutions for the class parameter $`๐(\alpha )`$
The interpolation algorithm just described is valid provided there is only one solution curve $`๐(\alpha )`$; otherwise we may end up trying to interpolate $`๐(\alpha )`$ using points which belong to different solution curves, leading nowhere. In turn, the existence of many curves $`๐(\alpha )`$ solving a given problem is related to the existence of symmetries in the invariants (of the ODE representative of the class we want to match) entering Eq.(22). Concretely, if the mapping $`\{๐\kappa (๐),x\varphi (x,๐)\}`$ is a symmetry of these invariants, then if $`\{๐,F(x)\}`$ leads to a solution for the equivalence problem, consequently $`\{\kappa (๐),\varphi (F(x),๐)\}`$ will also lead to a (different) solution. A concrete example of this situation is discussed in sec. 6.1.
Concerning detecting this situation, we note that if the mapping $`\{๐\kappa (๐),x\varphi (x,๐)\}`$ is a symmetry of the invariants, then the inverse mapping is also a symmetry, and since these invariants are rational in both $`x`$ and $`๐`$, the form of such a symmetry mapping is
$$\kappa (๐)=\frac{a๐+b}{c๐+d},\varphi (x,๐)=\frac{f(๐)x+g(๐)}{h(๐)x+j(๐)}$$
(25)
that is, a fractional linear mapping, where {a, b, c, d} and $`\{f(๐),g(๐),h(๐),j(๐)\}`$ are in principle constants and functions to be determined. Regarding the use of interpolation methods, the problem happens when $`\kappa (๐)๐`$. A first manner of detecting this problem then consists of setting up the system of equations and inequations
$$I(x,๐)=I(x,๐)_{|_{๐=\kappa (๐)}^{x=\varphi (x,๐)}},\varphi =\frac{F}{G},F_{xx}=0,G_{xx}=0,\kappa (๐)๐$$
(26)
where $`I(x,๐)=s_5^3/s_3^5`$ is the first invariant and $`F`$, $`G`$ and $`\kappa `$ are the unknowns to be determined, and seeing if this system is consistent. This check for consistency can be performed by simplifying this system with respect to its integrability conditions - for this purpose we used the diffalg and RIF Maple packages. When the system has no solution, this fact is detected by these packages; otherwise the related symmetry of the invariants is obtained from the output of these packages directly.
## 4 Integrable Abel ODE classes found in the literature
This section is devoted to a compilation of integrable Abel ODE classes found in the literature. The compilation is not intended to be complete, but it nevertheless covers various of the usual references; mainly Kamkeโs and Murphyโs books , and the original works by Abel, Liouville and others on these subjects .
One of the noticeable things in these references is that the presentation of integrable cases lacks a classification in terms of their invariants. Consequently, many of these ODEs can actually be obtained from one another by means of Eq.(2), that is, they belong to the same class. Since part of this work consisted in writing computer routines addressing the equivalence problem, we performed this classification, and therefore present here a more compact collection of integrable Abel ODE classes, as opposed to just integrable ODEs. Classes not depending on parameters are labelled by numbers (e.g., Class 1), while those depending on parameters are labelled with letters (e.g., Class A). The solutions to the representatives of these classes are presented altogether in a table in the Appendix.
While revising the related literature we also noticed that various of the cases presented in books or papers are in fact particular cases of the integrable classes presented by Abel, Liouville and Appell in . In turn the methods they used to obtain new integrable classes seem to be forgotten or not mentioned elsewhere. So, it appeared reasonable to start by reviewing and analyzing selected parts of those works in this section, and then show in the next section how, starting from these ideas, additional integrable classes can be obtained.
The first large presentation of integrable cases is due to Abel himself in . His idea was to consider integrating factors of the form
$$\mu =\mathrm{e}^{r(x,y)}$$
(27)
for โAbelโ equations written in terms of two arbitrary functions $`p`$ and $`q`$ as:
$$\mathrm{\Phi }yy^{}+p(x)+q^{}(x)y=0$$
(28)
The first non-trivial case discussed in was found by taking $`r(x,y)`$ as quadratic in $`y`$:
$$\mu =\mathrm{e}^{(\alpha +\beta y+\gamma y^2)}$$
where $`\alpha `$, $`\beta `$ and $`\gamma `$ are arbitrary functions of $`x`$. Abel formulated this problem by applying Eulerโs operator to the total derivative $`\mu \mathrm{\Phi }`$, obtaining a system easily solvable for $`\alpha `$, $`\beta `$, $`\gamma `$ and $`p`$. The resulting Abel family has non-constant invariant and is shown in Abelโs memoires as depending on one arbitrary function $`q(x)`$ and two arbitrary constants $`C_i`$:
$$yy^{}\frac{q^{}}{2C_1q+C_2}+q^{}y$$
(29)
(for the corresponding integrating factor see ). Now, for the purpose of building computer routines addressing the equivalence problem, it is crucial to determine whether or not a given class depends on parameters since, as explained in sec. 3, in such a case the formulation of that problem is much more difficult. In the case of Eq.(29), the two parameters $`C_i`$ and the function $`q(x)`$, can be removed by first converting the ODE to first kind using $`y(x)=1/v(x)`$, and then employing a transformation of the form Eq.(2): $`\{x=F(t),v(x)=u(t)\sqrt{2C_1}\}`$, with $`F`$ implicitly defined by $`2C_1q(F)t\sqrt{2C_1}+C_2=0`$, arriving at a representative of the class simpler than Eq.(29),
$$y^{}=\frac{y^3}{x}+y^2$$
(30)
and showing that this class does not depend on parameters. It is then easy to verify that Eq.(30) is a particular case of a parameterized class<sup>17</sup><sup>17</sup>17Eq.(30) is obtained from Eq.(58) taking $`C=0`$ and changing variables $`\{x=it,y=iu(t)\}`$. derived from Appellโs work .
The next integrable case shown by Abel is obtained by considering for Eq.(28) an integrating factor of the form $`\mu =\mathrm{exp}(1/(\alpha +\beta y))`$. Proceeding as in the previous case, Abel arrived at another integrable ODE class with non-constant invariant, which however (see ) is a particular member of the parameterized class Eq.(33) shown by Abel in the same paper.
Constant Invariant case
Abel then considered an integrating factor of the form $`\mu =\left(\alpha +\beta y\right)^n`$. This ansatz does not lead to a non-constant invariant family. However, this is the first presentation we have found of a method for the constant invariant case. Liouville, and others after him, rediscovered this method, presented in Kamke as due to M. Chini , and in Murphyโs book as a change of variables mapping an Abel ODE into a separable one. A recent discussion of the symmetries of this constant invariant problem is found in .
Class โAโ depending on one arbitrary parameter
The next ansatz considered by Abel was
$$\mu =\left(A+y\right)^a\left(B+y\right)^by$$
(31)
where $`A(x)`$ and $`B(x)`$ are arbitrary functions and $`a`$ and $`b`$ are arbitrary constants. By taking $`b=a`$ Abel showed that a tractable integrable case results:
$$yy^{}+\frac{q^{}}{4q}\left(\left(q+2\frac{C_1}{q}\right)^2\frac{q^2}{a^2}\right)+q^{}y=0$$
(32)
The arbitrary function $`q(x)`$ can be removed together with the constant $`C_1`$ by rewriting this ODE in first kind format, and then appropriately choosing $`\{F,P,Q\}`$ in Eq.(2); so that a simpler representative of this class depending on only one parameter โ$`\alpha `$โ, is given by<sup>18</sup><sup>18</sup>18A representative of the same class of Eq.(33) is shown in as $`y^{}=\frac{4}{9x^3}\left(\left(x^2+1\right)^2cx^4\right)y^3+\frac{4y^2}{3}`$
$$y^{}=\left(\alpha x+\frac{1}{x}+\frac{1}{x^3}\right)y^3+y^2$$
(33)
Class 1
In , Halphen noted a connection between doubly-periodic elliptic functions and the Abel type ODE
$$y^{}=\frac{3y\left(1+y\right)4x}{x\left(8y1\right)}$$
(34)
which transforms into itself under infinitely many rational changes of variables, from where he was able to determine both a parametric and an algebraic solution for it (see the Appendix).
Class 2
In a paper by Liouville mostly dedicated to Abel equations, he discussed the integrable cases known at that time (1903), and presented some new ones. Liouville reviewed Abelโs work and considered for Eq.(28) an integrating factor of the form Eq.(27) with $`r(x,y)`$ cubic in $`y`$, arriving at the integrable family $`y^{}=6axy^2+3ay^3`$, depending on a parameter $`a`$. This parameter however can be removed by changing variables as in $`\{y=u(t)/\sqrt[3]{3a},x=t/\sqrt[3]{3a}\}`$ arriving at the integrable class free of parameters represented by
$$y^{}=2y^2x+y^3$$
(35)
Class โBโ depending on one arbitrary parameter
As a generalization of Eq.(35), in Liouville also presented the parameterized family
$$y^{}+\left(3mx^2+4m^2x+n\right)y^3+3xy^2=0$$
(36)
written in terms of two parameters $`m`$ and $`n`$ and which can be mapped into a Riccati ODE solvable in terms of special functions. Eq.(35) is a member of the class represented by Eq.(36) after setting $`m=0`$. However, when $`m=0`$, $`n`$ can be removed from Eq.(36) by changing variables $`\left\{x=t\sqrt[3]{n},y(x)=tu(t)/n^{2/3}\right\}`$, leading to a class without parameters - actually represented by Eq.(35). In turn, when $`m0`$, $`m`$ and $`n`$ can be โmergedโ by changing variables $`\left\{y=u(t)/m^2,x=mt\right\}`$ and introducing a new parameter $`a=n/m^3`$, resulting in
$$y^{}=\left(3x^2+4x+a\right)y^33xy^2$$
(37)
In summary, Eq.(36) is not a full 2-parameter class, but instead two classes represented by Eqs.(35) and (37), respectively depending on zero and one parameters. A simpler representative for this class and its solution are found in the Appendix.
Class 3
Still in Liouville pointed out that by interchanging the role between the dependent and independent variables in Eq.(35) one arrives at a different Abel integrable class. After rewriting this resulting ODE in first kind format and performing a change of variables of the form Eq.(2), a simpler representative of this integrable class is given by
$$y^{}=\frac{y^3}{4x^2}y^2$$
(38)
### 4.1 Integrable Abel ODE classes shown in Kamke and some others books
One of the most well known collection of (69) Abel ODEs is the one shown in Kamkeโs book. This collection however makes no distinction between constant or non-constant invariant cases, presents ODEs of the same class as different, and does not discuss what would be the representative for each class depending on the least number of parameters. A first classification for these Abel ODEs is then given by<sup>19</sup><sup>19</sup>19In this classification, by โtoo generalโ we mean: these ODEs cannot be solved without restricting the example to a concrete particular case. We excluded ODEs - like those numbered 230 and 232 - which are already of Bernoulli type. We note also that the ODEs shown in Kamke without solution can all be transformed into an Emden type second order ODE shown in Kamke as 6.74, for which only a general discussion is presented. In turn, a detailed discussion on the integrable cases of Emden type ODEs is found in .:
| Classification | ODE numbers as in Kamkeโs book |
| --- | --- |
| 4 are too general | 50, 219, 250, 269 |
| 40 constant invariant | 38, 41, 46, 49, 51, 188, 204, 213, 214, 215, 216, 218, 221, 222, 223, 224, 225, 226, 227, 228, 229, 231, 236, 238, 239, 243, 244, 245, 246, 247, 248, 249, 251, 252, 254, 255, 260, 261, 262, 264 |
| 24 non-constant invariant | 36, 37, 40, 42, 43, 45, 47, 48, 111, 145, 146, 147, 151, 169, 185, 203, 205, 206, 234, 235, 237, 253, 257, 265 |
| 10 shown without solution | 40, 47, 48, 203, 205, 206, 234, 237, 253, 265 |
| Table 1. First classification for the 69 Abel ODEs shown in Kamkeโs book. | |
As mentioned, all constant invariant ODEs can systematically be transformed into separable ODEs (see for instance Murphyโs book), so that the interesting subset is the one comprising 24 ODEs having non-constant invariants. We note also that 10 of these 24 ODEs are shown in the book without a solution, and in fact we were unable to solve any of 203, 205, 206, 234, 253 or 265, so that the number of integrable cases for us is 18.
From these 18 ODEs (and hence from the 69 Abel type Kamkeโs examples), only four - those numbered: 47, 185, 235 and 237- would really lead to additional integrable classes with respect to those presented in the works by Abel, Liouville and Appell. We note however that the examples 47, 185 and 237 are all members of Class โCโ (see Eq.(49)), which can be derived from the work by Abel \- even when it was not presented in the original work. So that the number of additional integrable classes presented in Kamke reduces to one, represented by the example 235. The classification and details are as follows.
Class 4
$$\left(xy+a\right)y^{}+by=0$$
(39)
This ODE (K 1.235) is presented in Kamke in terms of two arbitrary parameters $`\{a,b\}`$; then, a change of variables which transforms it into a linear ODE is shown. A simpler representative of this class - not depending on parameters - can be obtained by rewriting this equation in first kind format via $`\{x=t,y=\frac{1}{tu(t)}\frac{a}{t}\}`$ and then changing variables $`\{x=\frac{a}{tb},y=\frac{tu(t)}{a}\}`$, leading to
$$y^{}=y^3\frac{\left(x+1\right)}{x}y^2$$
(40)
Comments on Kamkeโs example 47
For the ODE
$$y^{}a\left(x^nx\right)y^3y^2=0$$
(41)
presented in Kamke as K 1.47, there is no solution shown in the book, but instead a suggestion of transforming the ODE into a second order one. We followed that suggestion and then ran a symmetry analysis, noticing that the resulting ODE will have two point symmetries if either $`\{a=\frac{2n+2}{9+6n+n^2}\}`$ or $`\{n=2,a=\frac{6}{25}\}`$, leading to two integrable classes not shown in the book. In the former case, from Eq.(41), we arrive at
$$y^{}+\frac{\left(2n+2\right)\left(x^nx\right)y^3}{9+n^2+6n}y^2=0$$
(42)
However, this ODE can be transformed into Eq.(49) by changing variables $`\{x=t^{\frac{2}{1n}},y=u(t)\frac{n+3}{2}t^{\frac{n+1}{n1}}\}`$ followed by $`n=\frac{a+2}{a2}`$, so that it belongs to Class C. In the same line, taking $`\{n=2,a=\frac{6}{25}\}`$ in Eq.(41), and changing variables $`\{x=\frac{t^21}{t^2},y=5/2u(t)t^3\}`$ one arrives at Eq.(49) with $`a=6`$, so that this second branch of Eq.(41) is also a member of Class C.
Comments on Kamkeโs example 237
$$x\left(y+a\right)y^{}+by+cx=0$$
(43)
This ODE (K 1.237) depending on three arbitrary parameters $`\{a,b,c\}`$, is presented in the book without a solution. We note however that changing $`\{xy,yx\}`$ leads to an ODE also of Abel type and in second kind format. Converting the latter to first kind format via $`\{x=t,y=\frac{1}{cu(t)}\frac{bt}{c}\}`$, replacing $`yy^{}`$ and running a symmetry analysis, the resulting second order ODE has two symmetries when $`a=2b`$, leading to an integrable case. Introducing $`a=2b`$ into Eq.(43), rewriting it in first kind format via $`\{x=t,y=\frac{1}{tu(t)}+2b\}`$ and changing variables $`\{x=\frac{b^2\left(t+4\right)}{2c},y=\frac{2cu(t)}{b^3\left(t+4\right)}\}`$ leads to a simpler representative of the class not depending on any parameters:
$$y^{}=\frac{xy^3+2y^2}{2(x+4)}$$
(44)
However, by changing variables $`\{x=4(1t^2)/t^2,y=u(t)t/2\}`$ one arrives at Eq.(49) again, this time with $`a=1/2`$, so that Eq.(44) is also member of Class C.
A classification for all these 18 non-constant invariant Kamke examples is then as follows<sup>20</sup><sup>20</sup>20Equations K.1.47, K.1.48 and K.1.237 belong to Class C for infinitely many - however particular - values of one of the two parameters (see Eq.(42)); we donโt know their solution for other values.
| Class 2 | Class 3 | Class 4 | Class A | Class B | Class C | Class D |
| --- | --- | --- | --- | --- | --- | --- |
| 36, 40 | 145, 147 | 235 | 257 | 42, 43 | 45, 47, 48, 151, 185, 237 | 37, 111, 146, 169 |
| Table 2. Classification for the 18 non-constant invariant solvable Abel ODEs in Kamkeโs book. | | | | | | |
where classes C and D are defined in sec. 5. In summary, all but one of Kamkeโs 58 solvable examples (18 non-constant invariant + 40 constant invariant) are particular cases of the integrable classes presented by Abel, Liouville and Appell in , or can be derived from there (those belonging to Classes C and D).
Another collection of Abel ODEs is found in the book by Murphy . After selecting those examples not having a constant invariant and for which a solution is shown in the book, we arrived at a set of nine ODEs, numbered in the book as: 78, 79, 80, 86, 275, 304, 345, 383 and 593. None of these ODEs represent an additional integrable class; their distribution among the classes discussed in this work is as follows
| Class 2 | Class 3 | Class B | Class C | Class D |
| --- | --- | --- | --- | --- |
| 78, 80 | 275 | 86 | 304, 383, 593 | 79, 345 |
| Table 3. Classification for the non-constant invariant solvable Abel ODEs in Murphyโs book. | | | | |
A wider collection of Abel ODEs than the one shown in Kamkeโs book is found in the book by Polyanin and Zaitsev . This book is rather new (1995) and covers a vast number of integrable ODE problems which we have not found in other books, hence making the examples attractive. On the other hand the Abel ODEs shown there are classified not according to their invariants but according to their form, and the origin of their solutions is not given. Apart from a main section consisting of four tables (82 Abel ODEs \- all derived from four basic ones), the book contains other sections illustrating mappings between Abel and higher order ODEs. The quantity of examples is large and the computational routines we prepared for the equivalence problem are not yet covering in full the case in which the parameters of the class may assume symbolic values. As a result we still donโt have a way to solve the equivalence problem for the whole set of integrable classes presented in . Our analysis of these Abel ODEs of is then still incomplete; consequently we restricted the presentation here to just a sample, constituted by the ODEs of the first of these four tables. These are 20 ODEs obtained from
$$yy^{}y=sx+Ax^m$$
(45)
by giving particular values to the parameters $`m`$ and $`s`$ ($`A`$ is kept arbitrary). These ODEs appear in section 1.3.1 of under the numbers: 1, 2, 10, 16, 19, 22, 23, 26, 27, 30, 32, 33, 45, 46, 47, 48, 53, 54, 55 and 56. We were not able to classify those numbered 27, 20, 48, 55 and 56. The distribution of the remaining ODEs, in the classes discussed in this work, is as follows:
| Constant invariant | Class 1 | Class 2 | Class 3 | Class C | Class D |
| --- | --- | --- | --- | --- | --- |
| 1, 2, 26 | 23 | 32 | 33 | 10, 19, 22, 45, 46, 47, 53, 54 | 16 |
| Table 4. Classification for 15 of the 20 Abel ODE examples of Table 1.1 of . | | | | | |
## 5 New integrable Abel ODE classes derived from previous works
Class โCโ depending on one arbitrary parameter
The form of the integrating factor studied by Abel actually leads to other integrable cases not mentioned in the original work . One of them is obtained by taking $`b=a`$ in Eq.(31), resulting in the ODE family<sup>21</sup><sup>21</sup>21$`n`$ in Eq.(46) is related to $`a`$ in Eq.(31) by $`n=1/(2a+1)`$
$$yy^{}q^{}y\frac{q^{}n^2\left(\frac{q}{n}+C_{1}^{}{}_{}{}^{2}\left(\frac{q}{n}\right)^{2n1}\right)}{\left(n+1\right)^2}=0$$
(46)
where $`n1`$. The function $`q(x)`$ and the parameter $`C_1`$ can be removed as done with Eq.(32), leading to
$$y^{}=n\left(xx^{2n1}\right)y^3\left(n+1\right)y^2$$
(47)
which is turned exact by means of the integrating factor
$$\mu =\frac{\left(1+\left(\left(x^2x^{2n}\right)y2x\right)y\right)^{\frac{n+1}{2n}}}{y^{\frac{2n1}{n}}}$$
(48)
A simpler representative of this class is obtained by changing variables $`\{y=u(t)t^{\frac{n}{n1}},x=t^{\frac{1}{1n}}\}`$, then introducing a new parameter by means of $`n=\frac{\alpha }{\alpha 2}`$, arriving at
$$y^{}=\frac{\alpha \left(1x^2\right)y^3}{2x}+\left(\alpha 1\right)y^2\frac{\alpha y}{2x}$$
(49)
Taking into account Eq.(48), an implicit solution for this class is given by
$$C_1+\frac{\alpha }{x}\left(1\frac{\left(1xy\right)^2}{y^2}\right)^{1/\alpha }2^{^{\frac{1xy}{y}}}\left(1z^2\right)^{\frac{1\alpha }{\alpha }}๐z=0$$
(50)
Class โDโ depending on one arbitrary parameter
In , Appell showed a series of integrable cases derived from the solutions to
$$u^{}=A(u)+B(u)t$$
(51)
By changing variables $`\{t=\frac{1}{y}\frac{A(x)}{B(x)},u=x\}`$, this ODE is transformed into the Abel ODE
$$y^{}=\frac{y^3}{B}\left(\frac{A}{B}\right)^{}y^2$$
(52)
where $`A`$ and $`B`$ are now functions of $`x`$. Any particular $`\{A,B\}`$ leading to a solvable case in Eq.(51) will then also lead to an integrable Abel ODE Eq.(52). Among the choices for $`\{A,B\}`$ considered in \- such that Eq.(51) results linear, homogeneous, or of Riccati type - only this mapping into Riccati type leads to something new. This case is obtained by taking
$$A=ax^2+bx+c,B=\alpha x^2+\beta x+\gamma $$
(53)
The related Abel ODE family, depending on six parameters $`\{a,b,c,\alpha ,\beta ,\gamma \}`$, is given by
$$y^{}=\frac{y^3}{\alpha x^2+\beta x+\gamma }y^2\frac{d}{dx}\left(\frac{ax^2+bx+c}{\alpha x^2+\beta x+\gamma }\right)$$
(54)
and its solution could be expressed in terms of the solution to the Riccati ODE
$$y^{}=\left(a+\alpha x\right)y^2+\left(b+\beta x\right)y+c+\gamma x$$
(55)
However, we were not able to solve this Riccati ODE for arbitrary values of the six parameters involved and in there is no indication of how that could be done. The alternative we then investigated is to consider the second order ODE obtained by replacing $`y=y^{}`$ in Eq.(55). That ODE has two point symmetries if and only if $`\alpha =0`$. With these symmetries we were able to solve that second order ODE, and hence Eq.(55) when $`\alpha =0`$. Concerning the related Abel family Eq.(54) - now depending on five parameters - an appropriate change of variables of the form Eq.(2)
$$\left\{x=\frac{t\sqrt{\beta }}{a}\frac{\gamma }{\beta },y=\sqrt{\beta }u(t)\right\}$$
(56)
followed by the introduction of a new parameter $`C`$ by means of
$$C=\frac{\left(\beta ^2c+\alpha \gamma ^2\right)a\alpha \beta \gamma b}{\beta ^2}$$
(57)
transforms Eq.(54) into a simpler representative for the class
$$y^{}=\frac{y^3}{x}\frac{\left(C+x^2\right)y^2}{x^2}$$
(58)
also showing that this class depends not on five but on one parameter<sup>22</sup><sup>22</sup>22We note that in this process we have made two implicit assumptions: $`a0`$ and $`\beta 0`$. To assure that the cases in Eq.(54) are covered by Eq.(58) we then also considered $`a=0`$ and $`\beta =0`$ separately, arriving at ODEs respectively members of the classes represented by Eq.(58) and Eq.(35).. It appeared of value to us also to determine the number of parameters on which Eq.(54) depends in the general case, that is before taking $`\alpha =0`$. For that purpose we searched for the appropriate changes of variables of the form Eq.(2) which would remove as many as possible of these parameters, requiring that both the change and its inverse are finite. We then considered the branches which become infinite for some particular values of the parameters $`\{a,b,c,\alpha ,\beta ,\gamma \}`$ entering the transformations found. The results are summarized as follows. If all these parameters are different from zero, introducing new parameters $`\{A,B,C,G\}`$ by means of
$`\alpha `$ $`=`$ $`{\displaystyle \frac{\beta ^2+4A^4}{4\gamma }}`$
$`b`$ $`=`$ $`{\displaystyle \frac{8\beta \gamma ^2aA^2B+C}{2\gamma A^2B\left(\beta ^2+4A^4\right)}}`$
$`c`$ $`=`$ $`{\displaystyle \frac{A^2BC+16\gamma ^2aA^6B+\beta C+4\beta ^2\gamma ^2aA^2B}{A^2B\left(\beta ^2+4A^4\right)^2}}`$
$`\gamma `$ $`=`$ $`{\displaystyle \frac{C}{2A^3BG\left(\beta ^2+4A^4\right)}}`$ (59)
followed by changing variables $`\{x=\frac{C\left(2tA^2\beta \right)}{\left(\beta ^2+4A^4\right)^2A^3BG},y=u(t)A\}`$ in the six-parameter Eq.(54), one arrives at a 2-parameter representative for the same class
$$y^{}=\frac{y^3}{x^2+1}+\frac{G\left(Bx+x^21\right)y^2}{\left(x^2+1\right)^2}$$
(60)
Now the case $`\alpha =0`$ was already shown to lead to Eq.(58), and all the other possible branches obtained from Eq.(54) by taking some of the other parameters equal to zero lead either to constant invariant families, or to members of the classes already discussed in this work<sup>23</sup><sup>23</sup>23There is a special case, when $`b=4\frac{\gamma \beta a}{\beta ^2+4A^4}`$, where the resulting Abel ODE can only be obtained from Eq.(60) by taking appropriate limits.
Three new classes not depending on parameters
While analyzing the works and Kamkeโs examples, a large number of symbolic experiments were performed, sometimes leading to intermediate results which with a bit more work appeared to be new integrable classes by themselves. This happened three times, resulting in classes 5, 6 and 7, for which representatives and solutions are given as follows:
Class 5
$$y^{}=\frac{\left(2x+3\right)\left(x+1\right)y^3}{2x^5}+\frac{\left(5x+8\right)y^2}{2x^3}$$
(61)
Solution:
$$C_1+\frac{\sqrt{A}}{\sqrt[4]{4\frac{\left(x+1\right)^2}{x^2A}+1}}+^{2\frac{x+1}{x\sqrt{A}}}\left(z^2+1\right)^{5/4}๐z=0$$
(62)
where $`A=\frac{4}{y}\frac{10}{x}\frac{6}{x^2}4`$.
Class 6
$$y^{}=\frac{y^3}{x^2\left(x1\right)^2}+\frac{\left(1xx^2\right)y^2}{x^2\left(x1\right)^2}$$
(63)
Solution:
$$C_1\mathrm{Ei}(1,\frac{y+x^2x}{xy\left(x1\right)})+\frac{\left(x1\right)y\mathrm{e}^{\frac{xyx^2}{xy\left(x1\right)}}}{x1+y}=0$$
(64)
where $`\mathrm{Ei}(n,x)={\displaystyle _1^{^{\mathrm{}}}}{\displaystyle \frac{\mathrm{e}^{xt}}{t^n}}๐t`$ is the exponential integral.
Class 7
$$y^{}=\frac{\left(4x^4+5x^2+1\right)y^3}{2x^3}+y^2+\frac{\left(14x^2\right)y}{2x\left(x^2+1\right)}$$
(65)
Solution:
$$C_1+2\frac{x+A}{\sqrt[4]{A^2+1}\left(Ax1\right)}+^A\left(z^2+1\right)^{5/4}๐z=0$$
(66)
where $`A={\displaystyle \frac{x2yx^43yx^2y}{x\left(x+yx^2+y\right)}}`$
## 6 Computer algebra routines, tests and performance
The two itemized algorithms described in sections 2.2 and 3 for solving the equivalence problem between two given Abel ODEs were implemented in Maple R5, in the framework of the ODEtools package . The implementation consists of various routines, mainly accomplishing the following:
1. determine whether a given Abel ODE belongs to one of the solvable classes described in the previous sections; in doing that, determine also the function $`F(t)`$ entering Eq.(2) and the value of the parameters in the case of a parameterized class;
2. use that information to determine the functions $`P(t)`$ and $`Q(t)`$ entering Eq.(2) and return a solution to the given ODE by means of changing variables in the solution available for the representative of the class (see the Appendix).
### 6.1 Representatives with simpler invariants for Classes A, C and D
While preparing the computational routines being presented, we noticed that: if on the one hand the solutions to the representatives for the parameterized classes A, C and D (Eqs.(33, 49, 58)) can be expressed in a relatively simple manner (see the Appendix), on the other hand, for each of these representatives, the form of the three invariants entering Eq.(22) is much simpler if an appropriate redefinition of the class parameter followed by a change of variables of the form Eq.(2) is performed. In turn, the complexity of these invariants is a relevant issue for a computer algebra implementation since simpler invariants lead to simpler GCD and resultant computations.
In the case of Class A (Eq.(33)), redefining the parameter as $`\alpha =5/36\kappa /3`$ \- where $`\kappa `$ is the new parameter - and changing variables $`\{x=\frac{\sqrt{12t6}}{2t1},y=\frac{\left(2t1\right)u(t)}{\sqrt{12t6}}\}`$ lead to the class representative
$$y^{}=\frac{\left(\left(3\kappa 2xx^2\right)y^39y^29y\right)}{9(2x1)}$$
(67)
and the first invariant $`s_5^3/s_3^5`$ for it is
$$\frac{\left(9\kappa ^2+\left(30x^26x+3\right)\kappa 6x^3+5x^4+7x^2\right)^3}{9\left(\kappa +x^2\right)^5}$$
(68)
as opposed to
$$729\frac{\left(27\alpha ^2x^8\left(72x^6+270x^4+15x^8\right)\alpha +54x^4+36x^2+135+15x^6+2x^8\right)^3}{x^4\left(2x^49\alpha x^4+9x^2+27\right)^5}$$
(69)
which is the same first invariant but calculated for Eq.(33). Actually a measure of the Maple computational length of the three invariants $`[s_5^3/s_3^5,s_3s_7/s_5^2,s_5s_7/s_3^4]`$ entering Eq.(22) shows the values when calculated on Eq.(67) above and when calculated on Eq.(33). Also, the degrees in $`x`$ of both numerator and denominator of Eq.(68) are lower than the corresponding degrees of Eq.(69).
A similar situation happens with Class C (Eq.(49)), where redefining the parameter as $`\alpha =\frac{\kappa 3}{\kappa }`$ and changing variables $`\left\{x=\frac{\sqrt{\left(\kappa +1\right)\left(3\kappa \right)\left(3t+4\right)t}}{\left(\kappa +1\right)t},y=\frac{t^2u(t)\kappa \left(\left(3t+4\right)\kappa +3t+4\right)}{\sqrt{\left(\kappa +1\right)\left(3\kappa \right)\left(3t+4\right)t}}\right\}`$ lead to a class representative a bit more complicated than Eq.(49):
$$y^{}=4\left((\kappa 2)x+\kappa 3\right)\kappa xy^3+6y^2\frac{\left(6\kappa x+5\kappa +3\right)y}{\left(3x+4\right)\kappa x}$$
(70)
for which, however, the computational length of the three invariants $`[s_5^3/s_3^5,s_3s_7/s_5^2,s_5s_7/s_3^4]`$ is as opposed to when calculated for Eq.(49). The degrees with respect to $`x`$ of the numerators and denominators of the invariants of Eq.(70) are also lower than those of the invariants of Eq.(49).
The same reduction in the complexity of the invariants is achieved for class D (Eq.(58)) by redefining the parameter via $`C=\frac{9}{8}\sqrt{\kappa }`$ and changing variables $`\{x=\frac{3\sqrt{2\left(1t^2\kappa \right)\sqrt{\kappa }}}{4(1t\sqrt{\kappa })},y=\sqrt{2\left(1t^2\kappa \right)\sqrt{\kappa }}u(t)\}`$, leading to
$$y^{}=\frac{\left(2x^2\kappa 2\right)\kappa y^33y^2\kappa +\kappa xy}{1\kappa x^2}$$
(71)
for which the first invariant, $`s_5^3/s_3^5`$, is given by<sup>24</sup><sup>24</sup>24The case $`\kappa =0`$ is treated separately.
$$\frac{\left(\left(2x^4+15x^3\right)\kappa ^2\left(4x^2+15x9\right)\kappa +2\right)^3}{4\kappa ^3\left(\kappa x^3x+1\right)^5}$$
(72)
as opposed to
$$729\frac{\left(C+x^2\right)^3\left(2C^4+8x^2C^3+\left(12x^4+15x^2\right)C^2+8x^6C+2x^815x^6+9x^4\right)^3}{\left(2C^3+6C^2x^2+\left(9x^2+6x^4\right)C+2x^69x^4\right)^5}$$
(73)
which is the same first invariant but calculated for the class representative Eq.(58). In Eq.(72), not only the degrees with respect to $`x`$ but also those with respect to the class parameter $`\kappa `$ are lower than the corresponding degrees in Eq.(73).
Moreover, for class D this change in the representative of the class also fixes a problem: the representative Eq.(58) is itself invariant under $`CC`$ followed by the change of variables
$$\left\{x=\frac{iC}{t},y=iu(t)\right\}$$
(74)
where $`i`$ is the imaginary unit, and hence Eq.(73) is invariant under $`\{CC,xiC/x\}`$. Consequently, if $`\{C,F(t),P(t),Q(t)\}`$ is a solution to the equivalence problem between a given ODE and Eq.(58), then $`\{C,\frac{iC}{F(t)},iP(t),Q(t)\}`$ is also a solution; this fact invalidates the use of the interpolation method described in sec. 3.2 with Eq.(58) since that method can only be used when the interpolated solution $`C(\alpha )`$ is unique.
### 6.2 Installation
The programs being presented have been written as one more step in the development of the ODEtools Maple package and hence are integrated to it and not distributed separately. To install the new Abel routines then what is necessary is to install ODEtools to run in the Maple environment by putting the related libraries (two files - maple.ind and maple.lib) in any directory - say DE\_libraries\_directory \- then opening Maple, and adding that directory to Mapleโs libname variable via
```
> libname := DE_libraries_directory, libname:
```
where โ`>`โ is the Maple prompt. This instruction automatically updates Mapleโs dsolve subroutines to make use of the new routines for solving Abel ODEs - no further steps are required. In this way, the new Abel ODE routines are automatically used by Mapleโs dsolve when the input ODE is of Abel type, as well as to solve higher order ODEs when they can be reduced to first order Abel ODEs members of the classes discussed in sec. 4 and 5.
Apart from this integration with dsolve, it is also possible to try just the routines being presented by giving to dsolve the extra argument `[Abel]`. For example (Kamkeโs ODE 37)
```
> ode[37] := diff(y(x),x)-y(x)^3-a*exp(x)*y(x)^2 = 0;
```
$$ode_{37}:=y^{}y^3a\mathrm{e}^xy^2=0$$
```
> dsolve(ode[37], [Abel]);
```
$$C_1+\frac{\mathrm{e}^{1/2\left(a\mathrm{e}^x+y^1\right)^2}}{a\mathrm{e}^x}+\frac{\sqrt{2\pi }}{2}\mathrm{erf}\left(\frac{\sqrt{2}}{2}\left(a\mathrm{e}^x+y^1\right)\right)=0$$
(75)
These implicit answers can also be tested in the Maple worksheet, interactively, using the standard Maple odetest command.
Due to the intrinsic complexity of Abel equations, the solution for most of the solvable classes is expressed in implicit form and in terms of elliptic integrals and special or hypergeometric functions. Then, to save the time Maple spends in trying to โintegrateโ these integrals or to โinvertโ these algebraic expressions it is frequently convenient to call dsolve with the optional extra arguments `โuseInt, implicitโ`, meaning: use Mapleโs inert `Int` and return the solution directly in implicit form<sup>25</sup><sup>25</sup>25The option implicit- is standard in Mapleโs dsolve and the implementation of the option useInt- comes with the routines for Abel ODEs presented here.. For example, for Kamkeโs 185
```
> ode[185] := x^7*diff(y(x),x)+2*(x^2+1)*y(x)^3+5*x^3*y(x)^2 = 0
```
$$ode_{185}:=x^7y^{}+2\left(x^2+1\right)y^3+5x^3y^2=0$$
the solution obtained using these extra arguments is:
```
> dsolve(ode[185], [Abel], useInt, implicit);
```
$$C_1+\frac{x}{\sqrt[4]{A^2+1}}+1/2^A\left(z^2+1\right)^{5/4}๐z=0$$
(76)
where $`A=x^1+\frac{x^2}{y}`$. The explicit computation of the integral above leads to a complicated expression with hypergeometric functions somehow obscuring the structure of the solution.
Concerning Mapleโs difficulty in solving the problem when the class parameter $`๐`$ involves radicals (see end of sec. 3.1), we implemented an environment variable controlling how hard the routines will work, in order to avoid exhausting system resources unless a hard trial is specifically requested. This environment variable is `_Env_odsolve_Abel_try_hard`, it can be assigned a positive integer from 1 to 5, and by default it is assigned to 4; the meaning of the possible values is as follows:
* if `_Env_odsolve_Abel_try_hard = 1` then the algorithm for parameterized Abel classes is disabled, and for non-parameterized classes only a restricted equivalence using $`\{x=t,y=P(t)u+Q\}`$, that is: with $`F(t)=t`$ in Eq.(2), is tried;
* if `_Env_odsolve_Abel_try_hard = 2` then for non-parameterized classes a full equivalence using Eq.(2) is tried;
* if `_Env_odsolve_Abel_try_hard = 3` then the algorithm for parameterized Abel classes is enabled but only for numerical solutions for the class parameter $`๐`$;
* if `_Env_odsolve_Abel_try_hard = 4` then for parameterized Abel classes both a numerical or a rational interpolation solution for the class parameter $`๐`$ are tried (symbolic variables in the coefficients are allowed but solutions with radicals are not computed) ;
* if `_Env_odsolve_Abel_try_hard = 5` then the algorithm for parameterized Abel classes will also compute solutions involving radicals for the class parameter $`๐`$.
Finally, these routines for Abel ODEs were programmed to provide extensive run-time information on the computations being performed through the Maple standard userinfo & infolevel scheme. For example, Kamkeโs ODE 43 belongs to class B (see Table 2.) and both the way how this is determined and the explicit values found for $`F`$, $`P`$, $`Q`$, and the parameter entering Eq.(37) can be seen by setting the infolevel as follows<sup>26</sup><sup>26</sup>26We kept few lines to illustrate the userinfo feature, and represented the missing ones by โโฆ.โ-.:
```
> infolevel[dsolve] := 4;
> ode[43] :=
```
$$ode_{43}:=y^{}+\left(3ax^2+4a^2x+b\right)y^3+3y^2x=0$$
```
> dsolve(ode[43], [Abel], implicit);
....
The relative invariant s3 is: -3*a*x^2+b-2*x^3
The first absolute invariant s5^3/s3^5 is: 108*(12*a^2*x^3-8*a*x*b + ...
The second absolute invariant s3*s7/s5^2 is: 1/3*(3*a*x^2-b+2*x^3) * ...
The third absolute invariant s5*s7/s3^4 is: 9*(372*a^3*x^4+450*a^2*x^5 + ...
....
....
-> ======================================
-> ...checking Abel class B (by Liouville)
Trying a = 0
Trying a = 1
Trying b = 0
Trying b = 1
-> Step 1: checking for a disqualifying factor on F after evaluating x
Trying x = 0
*** No disqualifying factor on F was found ***
-> Step 2: calculating resultants to eliminate F and get candidates for C
*** Candidates for C are [1, 4, 1/4], ***
-> Step 3: looking for a solution F depending on x
_____________________________
C = 1/4 leads to the solutions [{F = -1-3/2*x}]
Interpolated candidate for the class parameter C is: C = 1/4
General testing of the candidate C = 1/4 with arbitrary b
Interpolation is still incomplete; trying next value of b
_____________________________
Trying b = 2
....
....
General testing of the candidate C = -3/4*b+1 with arbitrary b
_____________________________
C = -3/4*b+1 leads to the solutions [{F = -1-3/2*x}]
General test of C = -3/4*b+1 passed OK;
interpolation for b in this level is complete.
....
....
General testing of the candidate C = (-3/4*b+a^3)/a^3 with arbitrary a
_____________________________
C = (-3/4*b+a^3)/a^3 leads to the solutions [{F = -1/2*(3*x+2*a)/a}]
General test of C = (-3/4*b+a^3)/a^3 passed OK;
interpolation for b in this level is complete.
_____________________________
Value of the Class parameter solving the problem is:
C = 1/4*(-3*b+4*a^3)/a^3
Inverse of the transformation solving the problem is:
{t = -1/2*(3*x+2*a)/a, u(t) = -2/3*a^2*y(x)}
Solution:
```
$$C_1+\frac{\left(\left(\frac{2a+3x}{2a}A_1\right)\mathrm{K}(A_1,\sqrt{A_2})\sqrt{A_2}\mathrm{K}(A_1+1,\sqrt{A_2})\right)}{\left(\left(\frac{2a+3x}{2a}A_1\right)\mathrm{I}(A_1,\sqrt{A_2})+\sqrt{A_2}\mathrm{I}(A_1+1,\sqrt{A_2})\right)}=0$$
(77)
where $`A_1=\frac{1}{2}\sqrt{\frac{3b+4a^3}{a^3}}`$, $`A_2=\frac{3b4a^3}{4a^3}+\frac{\left(2a+3x\right)^2}{4a^2}\frac{3}{2a^2y}`$, and I and K are respectively the modified Bessel functions of first and second kind.
### 6.3 Tests and Performance
The idea was to test these computational routines to confirm the correctness of the returned solutions as well as to obtain the classification presented in the previous sections for solvable Abel ODEs. The first testing arena was the 69 Abel examples found in Kamke plus the 9 solvable examples with non-constant invariant from Murphyโs book, plus the first 20 Abel ODE examples from mentioned at the end of sec. 4 \- totaling 98 Abel ODE examples. The routines passed these tests - the solutions obtained were confirmed to be correct using other symbolic computation tools interactively - and the resulting classification is that shown in Tables 1, 2, 3 and 4 of sec. 4.
The aforementioned test however involves only 40 โnon-constant invariant and solvableโ Abel ODE examples, and does not fully test the new routines. We have then set up a more thorough test, which can be taken as a test-suite for other computer algebra implementations of methods for solving Abel ODEs. The ideas behind this additional test-suite are summarized as follows:
$``$ Take the representatives of the seven Abel class not depending on parameters, discussed in sec. 4 and 5, and generate with each four more Abel ODEs of the same class by applying the different types of transformations:
1. The general case with $`F,P`$ and $`Q`$ arbitrary:
$$tr_1:=\{x=F(t),y(x)=P(t)u(t)+Q(t)\}$$
(78)
2. Rational transformation with symbolic coefficients $`a`$ and $`b`$:
$$tr_2:=\{x=\frac{a}{t}+bt,y(x)=\frac{u(t)}{t}+1\}$$
(79)
3. Non-rational transformation involving an elementary function and symbols:
$$tr_3:=\{x=\mathrm{e}^t+1+\frac{t}{a},y(x)=u(t)+1\}$$
(80)
4. Non-rational transformation involving abstract powers and symbols:
$$tr_4:=\{x=\frac{a}{t^n}+\frac{t}{b},y(x)=u(t)+1\}$$
(81)
For example, the first of these four Abel ODEs generated from Eq.(34) is given by:
$`y^{}`$ $`=`$ $`{\displaystyle \frac{F^{}P^2\left(3+F\right)y^3}{8F}}+{\displaystyle \frac{\left(\left(3F9\right)Q10\right)PF^{}y^2}{8F}}`$
$`+\left({\displaystyle \frac{\left(\left(3F9\right)Q^220Q3\right)F^{}}{8F}}{\displaystyle \frac{P^{}}{P}}\right)y+{\displaystyle \frac{\left(\left(F3\right)Q^310Q^23Q\right)F^{}}{8FP}}{\displaystyle \frac{Q^{}}{P}}`$
and the other three are obtained from this one by replacing $`F`$, $`P`$ and $`Q`$ by the values implied by Eqs. (79), (80) and (81). The scheme just outlined generates 28 more solvable Abel ODE examples with non-constant invariant (4 per class), and suffices for testing the solving of classes without parameters.
Concerning classes with parameters:
$``$ Take the representatives of the four Abel classes depending on parameters discussed in sec. 4 and 6.1, generate with each one four more Abel ODEs of the same class by applying the transformations Eq.(79) and Eq.(81), preceded by replacing the single parameter $`๐`$ entering each class representative by $`๐=2`$ and $`๐=\alpha /\beta `$.
This increases by 4 x 2 x 2 = 16 more ODE examples, and the solving of these examples tests both the scheme for numeric values of the parameter $`๐`$ as well as the case where $`๐`$ is a rational function of other symbols entering the given ODE.
The time spent by the routines being presented in solving all these 28 + 16 = 44 additional Abel ODEs with non-constant invariant, is summarized in Tables 5, 6 and 7 as follows:
| Class | transf. (i) | transf. (ii) | transf. (iii) | transf. (iv) |
| --- | --- | --- | --- | --- |
| 1; Eq.(34) | 0.255 sec. | 0.879 sec. | 0.687 sec. | 1.256 sec. |
| 2; Eq.(35) | 1.530 sec. | 1.900 sec. | 1.998 sec. | 2.620 sec. |
| 3; Eq.(38) | 0.277 sec. | 6.760 sec. | 5.623 sec. | 11.635 sec. |
| 4; Eq.(40) | 0.456 sec. | 7.126 sec. | 3.941 sec. | 15.511 sec. |
| 5; Eq.(61) | 0.610 sec. | 2.505 sec. | 3.082 sec. | 7.877 sec. |
| 6; Eq.(63) | 1.379 sec. | 120.952 sec. | 45.565 sec. | 282.103 sec. |
| 7; Eq.(65) | 0.896 sec. | 15.441 sec. | 24.586 sec. | 183.557 sec. |
| Table 5. Timings for 28 Abel ODEs with non-constant invariant - 7 classes free of parameters. | | | | |
| --- | --- | --- | --- | --- |
| Class parameter $`๐=2`$ | | |
| --- | --- | --- |
| Class | transf. (ii) | transf. (iv) |
| A; Eq.(67) | 25.546 sec. | 41.304 sec. |
| B; Eq.(37) | 29.396 sec. | 54.171 sec. |
| C; Eq.(70) | 18.648 sec. | 41.425 sec. |
| D; Eq.(71) | 36.641 sec. | 75.962 sec. |
| Table 6. Timings for 8 Abel ODEs with non-constant invariant - โnumericโ class parameter. | | |
| --- | --- | --- |
| Class parameter $`๐=\frac{\alpha }{\beta }`$ | | |
| --- | --- | --- |
| Class | transf. (ii) | transf. (iv) |
| A; Eq.(67) | 67.076 sec. | 156.425 sec. |
| B; Eq.(37) | 174.751 sec. | 491.210 sec. |
| C; Eq.(70) | 73.940 sec. | 174.761 sec. |
| D; Eq.(71) | 88.365 sec. | 177.314 sec. |
| Table 7. Timings for 8 Abel ODEs with non-constant invariant - โsymbolicโ class parameter. | | |
| --- | --- | --- |
Concerning a comparison of performances between the new routines and those available in other computer algebra systems (CAS), this appeared to us not justified in this case: roughly speaking none of these CAS return solutions for Abel ODEs with non-constant invariant. More precisely, from Mathematica 3.0, Macsyma 2.7, Maple 5.1 and Mupad 4.0, all of them failed<sup>27</sup><sup>27</sup>27For Kamkeโs ODE 257, Macsyma (2.7) returns a wrong answer in terms of $`y^{}`$. in solving any of the 18 Kamkeโs examples shown in Table 2. (and hence in solving any of the 44 ODEs of Tables 5, 6 and 7), except for Kamkeโs example 235 - it is an inverse linear ODE - and anyway none of them solved it after transforming it to first kind format.
### 6.4 Performance with the 1$`^{\mathrm{st}}`$order ODE examples from Kamke
Although the main purpose of this paper is to present a computational scheme for finding solutions to Abel ODEs, it is interesting to see how odsolve \- the ODE-solver of the ODEtools Maple package \- performs with the addition of these new routines. The performance with all of Kamkeโs 553 solvable examples<sup>28</sup><sup>28</sup>28We classified as unsolvable in general Kamkeโs examples 50, 55, 56, 74, 79, 82, 202, 219, 250, 269, 331, 370, 461, 503 and 576. after incorporating the computational routines presented in this paper is: 97% are solved. This performance is summarized as follows
| | | | Average time | |
| --- | --- | --- | --- | --- |
| Degree in $`y^{}`$ | ODEs | Solved | solved | fail |
| 1 | 350 | 337 | 3.2 sec. | 12.9 sec. |
| 2 | 145 | 140 | 8.8 sec. | 61.1 sec. |
| 3 | 27 | 26 | 7.2 sec. | 17 sec. |
| higher | 31 | 30 | 13.4 sec. | 25.2 sec. |
| Total: | 553 | 533 | $``$ 6 sec. | $``$ 20 sec. |
| Table 8. Kamkeโs first order ODEs, solved by odsolve: $`97\%`$ | | | | |
The number and classification of Kamkeโs 1$`^{\mathrm{st}}`$order ODEs still not solved by odsolve is now:
| Class | Kamkeโs numbering |
| --- | --- |
| rational | 452 |
| Riccati | 25 |
| NONE | 80, 81, 83, 87, 121, 128, 340, 367, 395, 460, 506, 510, 543, 572 |
| Table 9. Kamkeโs 1$`^{\mathrm{st}}`$order solvable ODEs for which odsolve fails: $`3\%`$ | |
| --- | --- |
where the Abel ODEs numbered in Kamkeโs book as 47, 48, 205, 206, 237 and 265 not presented in the tables above are known to be solvable only for specific values of their parameters - and for these values odsolve succeeds - and the ODEs 234 and 253 were not included since their solutions are not shown in the book or known to us.
## 7 Conclusions
In this paper, a first classification, according to invariant theory, of solvable non-constant invariant Abel ODEs found in the literature, was presented. Also, a set of new solvable classes, depending on one or no parameters, derived from the analysis of the works by Abel, Liouville and Appell , was shown. Computer algebra routines were then developed, in the framework of the Maple ODEtools package, to solve - in principle - any member of these classes by solving its related equivalence problem. The result is a concrete new tool for solving Abel type ODEs fully integrated with Mapleโs ODE-solver dsolve.
The classification shown has had the intention of giving a first step towards organizing in a single reference the integrable cases widely scattered throughout the literature. The derivation of new solvable parameterized classes from the works by Abel and Appell in the 19<sup>th</sup> century (Classes โCโ and โDโ) also showed that valuable information can still be obtained from these old papers. In fact, from Tables 1, 2, 3 and 4 in sec. 4, the larger number of integrable cases found in textbooks are particular members of this Class โCโ (Eq.(49)) - an integrable class derived from a case somehow disregarded in Abelโs Memoires .
As for the computer routines, the implementation presented here for solving the equivalence problem for parameterized classes proved to be a valuable tool in most of the Abel ODE examples we were able to collect. The Abel ODE routines here presented were recently integrated to the Maple computer algebra system and are already part of the upcoming Maple release 6.
On the other hand, we note the intrinsic limitation of this Abel ODE problem: most of the solutions can only be obtained in implicit form and in terms of quadratures; in turn, these integrals are usually elliptic integrals so that they cannot be expressed using elementary functions. Also the classification of integrable cases presented here is incomplete in that it is missing - at least - a more thorough analysis of the integrable cases presented in . We are presently working on this topic and expect to succeed in obtaining reportable results in the near future.
Acknowledgments
This work was partially supported by the State University of Rio de Janeiro (UERJ), Brazil and by the Symbolic Computation Group, Faculty of Mathematics, University of Waterloo, Ontario, Canada. The authors would like to thank K. von Bรผlow for a careful reading of this paper, and T. Kolokolnikov and A. Wittkopf for fruitful discussions.
## Appendix A<sup>29</sup><sup>29</sup>29The solution shown for the representative of class D is not valid when $`\alpha `$ is an integer, or when $`2\alpha `$ is a positive integer. In those cases, the solution of the associated Riccati equation Eq.(55) takes many different forms depending on the value of $`\alpha `$, which we found inconvenient to present here. $`\mathrm{Ei}(n,x)={\displaystyle _1^{^{\mathrm{}}}}{\displaystyle \frac{\mathrm{e}^{xt}}{t^n}}๐t`$ is the exponential integral, $`\mathrm{Ai}(x)`$ and $`\mathrm{Bi}(x)`$ are the Airy wave functions, $`\mathrm{K}(x)`$ and $`\mathrm{I}(x)`$ are the modified Bessel functions of the first and second kinds, respectively, and $`\mathrm{M}(x)`$ and $`\mathrm{W}(x)`$ are the Whittaker functions.
| Class | Representative equation and solution |
| --- | --- |
| 1 | $`y^{}=\frac{3y^23yx}{x\left(8y9\right)},`$ $`C_1+\frac{x^3\left(4x^2+\left(8y^236y+27\right)x+4y^44y^3\right)}{\left(x^2+2x\left(y^23y\right)+y^4\right)^3}=0`$ |
| 2 | $`y^{}=2y^2x+y^3,`$ $`C_1+\frac{x\mathrm{Ai}\left(x^2\frac{1}{y}\right)+\mathrm{Ai}(1,x^2\frac{1}{y})}{x\mathrm{Bi}\left(x^2\frac{1}{y}\right)+\mathrm{Bi}(1,x^2\frac{1}{y})}=0`$ |
| 3 | $`y^{}=\frac{y^3}{4x^2}y^2,`$ $`C_1+\frac{\left(x\frac{1}{y}\right)\mathrm{Ai}\left(\left(x\frac{1}{y}\right)^2\frac{1}{2x}\right)+\mathrm{Ai}(1,\left(x\frac{1}{y}\right)^2\frac{1}{2x})}{\left(x\frac{1}{y}\right)\mathrm{Bi}\left(\left(x\frac{1}{y}\right)^2\frac{1}{2x}\right)+\mathrm{Bi}(1,\left(x\frac{1}{y}\right)^2\frac{1}{2x})}=0`$ |
| 4 | $`y^{}=y^3\frac{x+1}{x}y^2,`$ $`C_1+\frac{1}{x}e^{\frac{1}{y}x}\mathrm{Ei}(1,x\frac{1}{y})=0`$ |
| 5 | $`y^{}=\frac{\left(2x+3\right)\left(x+1\right)y^3}{2x^5}+\frac{\left(5x+8\right)y^2}{2x^3},`$ $`C_1+\frac{\sqrt{A}}{\sqrt[4]{4\frac{\left(x+1\right)^2}{x^2A}+1}}+^{2\frac{x+1}{x\sqrt{A}}}\left(z^2+1\right)^{5/4}๐z=0,A=\frac{4}{y}\frac{10}{x}\frac{6}{x^2}4`$ |
| 6 | $`y^{}=\frac{y^3}{x^2\left(x1\right)^2}+\frac{\left(1xx^2\right)y^2}{x^2\left(x1\right)^2},`$ $`C_1\mathrm{Ei}(1,\frac{y+x^2x}{xy\left(x1\right)})+\frac{\left(x1\right)y\mathrm{e}^{\frac{xyx^2}{xy\left(x1\right)}}}{x1+y}=0`$ |
| 7 | $`y^{}=\frac{\left(4x^4+5x^2+1\right)y^3}{2x^3}+y^2+\frac{\left(14x^2\right)y}{2x\left(x^2+1\right)}`$ $`C_1+2\frac{x+A}{\sqrt[4]{A^2+1}\left(Ax1\right)}+^A\left(z^2+1\right)^{5/4}๐z=0A={\displaystyle \frac{x2yx^43yx^2y}{x\left(x+yx^2+y\right)}}`$ |
| A | $`y^{}=\left(\alpha x+\frac{1}{x}+\frac{1}{x^3}\right)y^3+y^2,`$ $`C_1+\frac{x^3}{y+x}\mathrm{exp}\left(^{\frac{yx^2}{y+x}}\frac{2dz}{z^2z\alpha z^3}\right)^{\frac{yx^2}{y+x}}\mathrm{exp}\left(\frac{2dz}{z^2z\alpha z^3}\right)๐z=0`$ |
| B | $`y^{}=2\left(x^2\alpha ^2\right)y^3+2\left(x+1\right)y^2,`$ $`C_1+\frac{\left(\alpha +x\right)\mathrm{K}(\alpha ,\sqrt{x^2+\frac{1}{y}\alpha ^2})+\sqrt{x^2+\frac{1}{y}\alpha ^2}\mathrm{K}(1+\alpha ,\sqrt{x^2+\frac{1}{y}\alpha ^2})}{\left(\alpha +x\right)\mathrm{I}(\alpha ,\sqrt{x^2+\frac{1}{y}\alpha ^2})\sqrt{x^2+\frac{1}{y}\alpha ^2}\mathrm{I}(1+\alpha ,\sqrt{x^2+\frac{1}{y}\alpha ^2})}=0`$ |
| C | $`y^{}=\frac{\alpha \left(1x^2\right)y^3}{2x}+\left(\alpha 1\right)y^2\frac{\alpha y}{2x},`$ $`C_1+\frac{\alpha }{x}\left(1\left(\frac{1}{y}x\right)^2\right)^{1/\alpha }2^{^{\frac{1xy}{y}}}\left(1z^2\right)^{\frac{1\alpha }{\alpha }}๐z=0`$ |
| D | $`y^{}=\frac{y^3}{x}\frac{\left(\alpha +x^2\right)y^2}{x^2},`$ $`C_1+\frac{\left(\alpha +1\right)\mathrm{M}(\frac{\alpha }{2}\frac{3}{4},\frac{1}{4},\frac{1}{2}\left(x\frac{\alpha }{x}\frac{1}{y}\right)^2)+\left(\frac{x}{y}x^2\right)\mathrm{M}(\frac{\alpha }{2}+\frac{1}{4},\frac{1}{4},\frac{1}{2}\left(x\frac{\alpha }{x}\frac{1}{y}\right)^2)}{\left(\alpha ^2+\alpha \right)\mathrm{W}(\frac{\alpha }{2}\frac{3}{4},\frac{1}{4},\frac{1}{2}\left(x\frac{\alpha }{x}\frac{1}{y}\right)^2)+2\left(\frac{x}{y}x^2\right)\mathrm{W}(\frac{\alpha }{2}+\frac{1}{4},\frac{1}{4},\frac{1}{2}\left(x\frac{\alpha }{x}\frac{1}{y}\right)^2)}=0`$ |
| Representative ODEs and their solutions for the Abel ODE classes presented in this work. | |
| --- | --- | |
warning/0001/hep-ph0001177.html | ar5iv | text | # Outโofโequilibrium dynamics of large-๐ ฯโด QFT in finite volume
## I Introduction
In the last few years a great deal of attention has been paid to the study of interacting quantum fields out of equilibrium. There are, in fact, many interesting physical situations in which the standard Sโmatrix approach cannot give sensible information about the behavior of the system, because it evolves through a series of highly excited states (i.e., states of finite energy density).
As an example consider any model of cosmological inflation: it is not possible to extract precise predictions on physical observables without including in the treatment the quantum backโreaction of the field on the spaceโtime geometry and on itself .
On the side of particle physics, the ultra-relativistic heavy-ion collisions, scheduled in the forthcoming years at CERNโSPS, BNLโRHIC and CERNโLHC, are supposed to produce hadron matter at very high densities and temperatures; in such a regime the usual approach based on particle scattering cannot be considered a good interpretative tool at all. To extract sensible information from the theory new computational schemes are necessary, that go beyond the simple Feynman diagram expansion. The use of resummation schemes, like the HartreeโFock (HF) approximation and the large $`N`$ limit (LN) , or the Hard Thermal Loop resummation for systems at finite temperature (HTL) , can be considered a first step in this direction. They, in fact, enforce a sum over an infinite subset of Feynman diagrams that are dominant in a given region of the parameter space, where the simple truncation of the usual perturbative series at finite order cannot give sensible answers.
Quite recently HF and LN have been used in order to clarify some dynamical aspects of the large $`N`$ $`\varphi ^4`$ theory. For the readerโs benefit and to better motivate this work, we give a very short summary of the conclusions reached in previous works: i) the early time evolution is dominated by a soโcalled โlinear regimeโ, during which the energy initially stored in one (or few) modes of the field is transferred to other modes via either parametric or spinodal unstabilities, resulting in a large particle production and a consequent dissipation for the initial condensate ; ii) the linear regime stops at a time scale $`t_1\mathrm{log}(\lambda ^1)`$ (where $`\lambda `$ is the quartic coupling constant), by which the effects of the quantum fluctuation become of the same order as the classical contribution and the dynamics turns completely non linear and non perturbative ; iii) after the time $`t_1`$ the relaxation occurs via power laws with anomalous dynamical exponent ; iv) the asymptotic particle distribution, obtained as the result of the copious particle production at the expenses of the โclassicalโ energy, is strongly non-thermal ; and finally, v) at very large time scale, $`t\sqrt{V}`$ (where $`V`$ stands for the volume of the system), the nonโperturbative and nonโlinear evolution might eventually produce the onset of a novel form of nonโequilibrium BoseโEinstein condensation of the longโwavelength Goldstone bosons usually present in the broken symmetry phase of the model . Another very interesting result in concerns the dynamical Maxwell construction, which reproduces the flat region of the effective potential in case of broken symmetry as asymptotic fixed points of the background evolution. Moreover, the LN approximation scheme has been used to follow the evolution of a initial state characterized by an occupied spherical shell in momentum space (a spherical โtsunamiโ ), around a particular momentum $`\left|\stackrel{}{k}_0\right|`$, with the following results: i) in a theory where the symmetry is spontaneously broken at zero density, if we start with a finite density initial state with restored symmetry, the spinodal instabilities lead to a dynamical symmetry breaking; ii) the evolution produces a re-arrangement of the particle distribution towards low momenta, signalling the onset of Bose condensation; iii) the equation of state of the asymptotic gas is ultra-relativistic (even if the distribution is not thermal) .
In this article we present a detailed study, in finite volume, of dynamical evolution out of equilibrium for the $`\mathrm{\Phi }^4`$ scalar field in the large $`N`$ limit. More precisely, we determine how such dynamics scales with the size of the periodic box containing the system in the case of uniform backgrounds. This is necessary to address questions like outโofโequilibrium symmetry breaking and dynamical BoseโEinstein condensation.
In section II we define the model in finite volume, giving all the relevant notations and definitions. We also stress the convexity of the effective potential as an exact result, valid for the full renormalized theory in any volume.
In section III we derive the large $`N`$ approximation of the $`O(N)`$invariant version of $`\lambda (\mathit{\varphi }^2)^2`$ model, according to the general rules of ref. . In this derivation it appears evident the essential property of the $`N\mathrm{}`$ limit of being a particular type of classical limit, so that it leads to a classical phase space, a classical hamiltonian with associated Hamiltonโs equations of motion \[see eqs. (16), (17) and (18)\]. We then minimize the hamiltonian function(al) and determine the conditions for massless Goldstone bosons (i.e. transverse fluctuations of the field) to form a BoseโEinstein condensate, delocalizing the vacuum field expectation value (see also ref. ). This necessarily requires that the width of the zeroโmode fluctuations becomes macroscopically large, that is of the order of the volume. Only when the background takes one of the extremal values proper of symmetry breaking the width of the zeroโmode fluctuations is of order $`L^{1/2}`$, as typical of a free massless spectrum.
The study of the lowest energy states of the model is needed for comparison with the results of the numerical simulations, which show that the zeroโmode width $`\sigma _0`$ stays microscopic (that is such that $`\sigma _0/`$volume$`0`$ when the volume diverges) whenever it starts from initial conditions in which it is microscopic. Our results, in fact, show clearly the presence of a time scale $`\tau _L`$, proportional to the linear size $`L`$ of the system, at which finite volume effects start to manifest. We shall give a very simple physical interpretation of this time scale in section III D. The important point is that after $`\tau _L`$ the zero mode amplitude starts decreasing, then enters an erratic evolution, but never grows macroscopically large. This result is at odd with the interpretation of the linear lateโtime growth of the zeroโmode width as a full dynamical BoseโEinstein condensation of Goldstone bosons, but is compatible with the โnovelโ form of BEC reported in .
In fact we do find that the size of the lowโlying widths at time $`\tau _L`$ is of order $`L`$, to be compared to the equilibrium situation where they would be of order $`L^0`$ in the massive case or of order $`L^{1/2}`$ in the massless case. Perhaps the denomination โmicroscopicโ should be reserved to this two possibilities. Therefore, since our initial condition are indeed microscopic in this restricted sense, we do observe in the outโofโequilibrium evolution a rapid transition to a different regime intermediate between the microscopic one and the macroscopic one characteristic of BoseโEinstein condensation. As we shall discuss more in detail later on, this fully agrees with the result found in , that the timeโdependent field correlations vanish at large separations more slowly than for equilibrium free massless fields (as $`r^1`$ rather than $`r^2`$), but definitely faster than the equilibrium broken symmetry phase characterized by constant correlations at large distances.
At any rate, when one considers microscopic initial conditions for the choice of bare mass which corresponds to broken symmetry, the role itself of symmetry breaking is not very clear in the large $`N`$ description of the outโofโequilibrium dynamics, making equally obscure the issues concerning the soโcalled quantum phase ordering . This is because the limit $`N\mathrm{}`$ is completely saturated by gaussian states, which might signal the onset of symmetry breaking only developing macroscopically large fluctuations. Since such fluctuations do not appear to be there, the meaning itself of symmetry breaking, as something happening as times goes on and accompanied by some kind of phase ordering, is quite unclear. We postpone to a companion work the discussion about the possibility of using more comprehensive approximation schemes, that include some nonโgaussian features of the complete theory. As far as the large $`N`$ approximation is concerned, we underline that an important limitation of our approach, as well as of those of the references mentioned above, is in any case the assumption of a uniform background. Nonetheless, phenomena like the asymptotic vanishing of the effective mass and the dynamical Maxwell construction, taking place in this contest of a uniform background and large $`N`$ expansion, are certainly very significant manifestations of symmetry breaking and in particular of the Goldstone theorem which applies when a continuous symmetry is broken.
Finally, in section IV we summarize the results presented in this article and we sketch some interesting open problems that we plan to study in forthcoming works.
## II Cutoff field theory
We consider the $`N`$component scalar field operator $`\mathit{\varphi }`$ in a $`D`$dimensional periodic box of size $`L`$ and write its Fourier expansion as customary
$$\mathit{\varphi }(x)=L^{D/2}\underset{k}{}\mathit{\varphi }_ke^{ikx},\mathit{\varphi }_k^{}=\mathit{\varphi }_k$$
with the wavevectors $`k`$ naturally quantized: $`k=(2\pi /L)n`$, $`n^D`$. The canonically conjugated momentum $`๐
`$ has a similar expansion
$$๐
(x)=L^{D/2}\underset{k}{}๐
_ke^{ikx},๐
_k^{}=๐
_k$$
with the commutation rules $`[\varphi _k^\alpha ,\pi _k^{}^\beta ]=i\delta _{kk^{}}^{(D)}\delta ^{\alpha \beta }`$. The introduction of a finite volume should be regarded as a regularization of the infrared properties of the model, which allows to โcountโ the different field modes and is needed especially in the case of broken symmetry. In fact, all the results we have summarized in section I, have been obtained simulating the system directly in infinite volume, where the evolution equations contain momentum integrals, that must be computed numerically by a proper, but nonetheless rather arbitrary, discretization in momentum space. Of course, the final result should be as insensitive as possible to the particular choice of the integration grid. In such a situation, the definition of a โzeroโ mode and the interpretation of its late time behavior might not be rigorous enough, unless, for some reason, it turns out that a particular mode requires a different treatment compared to the others. In order to understand this point, it is necessary to put the system in a finite volume (a box of size $`V`$); the periodic boundary conditions let us single out the zero mode in a rigorous way and thus we can carefully analyze its scaling properties with $`V`$ and get some information on the infinite volume limit.
To regularize also the ultraviolet behavior, we restrict the sums over wavevectors to the points lying within the $`D`$dimensional sphere of radius $`\mathrm{\Lambda }`$, that is $`k^2\mathrm{\Lambda }^2`$, with $`๐ฉ=\mathrm{\Lambda }L/2\pi `$ some large integer. Clearly we have reduced the original fieldโtheoretical problem to a quantumโmechanical framework with finitely many (of order $`๐ฉ^{D1}`$) degrees of freedom.
The $`\varphi ^4`$ Hamiltonian reads
$$\begin{array}{cc}\hfill H& =\frac{1}{2}d^Dx\left[๐
^2+(\mathit{\varphi })^2+m_\text{b}^2\mathit{\varphi }^2+\lambda _\text{b}(\mathit{\varphi }^2)^2\right]=\frac{1}{2}\underset{k}{}\left[๐
_k๐
_k+(k^2+m_\text{b}^2)\mathit{\varphi }_k\mathit{\varphi }_k\right]+\hfill \\ & +\frac{\lambda _\text{b}}{4L^D}\underset{k_1,k_2,k_3,k_4}{}(\mathit{\varphi }_{k_1}\mathit{\varphi }_{k_2})(\mathit{\varphi }_{k_3}\mathit{\varphi }_{k_4})\delta _{k_1+k_2+k_3+k_4,0}^{(D)}\hfill \end{array}$$
(1)
where $`m_\text{b}^2`$ and $`\lambda _\text{b}`$ should depend on the UV cutoff $`\mathrm{\Lambda }`$ in such a way to guarantee a finite limit $`\mathrm{\Lambda }\mathrm{}`$ for all observable quantities. As is known , this implies triviality (that is vanishing of renormalized vertex functions with more than two external lines) for $`D>3`$ and very likely also for $`D=3`$. In the latter case triviality is manifest in the oneโloop approximation and in large$`N`$ limit due to the Landau pole. For this reason we shall keep $`\mathrm{\Lambda }`$ finite and regard the $`\varphi ^4`$ model as an effective lowโenergy theory (here lowโenergy means practically all energies below Planckโs scale, due to the large value of the Landau pole for renormalized coupling constants of order one or less).
We shall work in the wavefunction representation where $`๐|\mathrm{\Psi }=\mathrm{\Psi }(๐)`$ and
$$(\mathit{\varphi }_0\mathrm{\Psi })(๐)=๐_0\mathrm{\Psi }(๐),(๐
_0\mathrm{\Psi })(๐)=i\frac{}{๐_0}\mathrm{\Psi }(๐)$$
while for $`k>0`$ (in lexicographic sense)
$$(\mathit{\varphi }_{\pm k}\mathrm{\Psi })(๐)=\frac{1}{\sqrt{2}}\left(๐_k\pm i๐_k\right)\mathrm{\Psi }(๐),(๐
_{\pm k}\mathrm{\Psi })(๐)=\frac{1}{\sqrt{2}}\left(i\frac{}{๐_k}\pm \frac{}{๐_k}\right)\mathrm{\Psi }(๐)$$
Notice that by construction the variables $`๐_k`$ are all real. Of course, when either one of the cutoffs are removed, the wave function $`\mathrm{\Psi }(๐)`$ acquires infinitely many arguments and becomes what is usually called a wavefunctional.
In practice, the problem of studying the dynamics of the $`\varphi ^4`$ field out of equilibrium consists now in trying to solve the time-dependent Schroedinger equation given an initial wavefunction $`\mathrm{\Psi }(๐,t=0)`$ that describes a state of the field far away from the vacuum. By this we mean a nonโstationary state that, in the infinite volume limit $`L\mathrm{}`$, would lay outside the particle Fock space constructed upon the vacuum. This approach could be generalized in a straightforward way to mixtures described by density matrices, as done, for instance, in . Here we shall restrict to pure states, for sake of simplicity and because all relevant aspects of the problem are already present in this case.
It is by now well known that perturbation theory is not suitable for the purpose stated above. Due to parametric resonances and/or spinodal instabilities there are modes of the field that grow exponentially in time until they produce nonโperturbative effects for any coupling constant, no matter how small. On the other hand, only few, by now standard, approximate nonโperturbative schemes are available for the $`\varphi ^4`$ theory, and to these we have to resort after all. We shall consider here only the large $`N`$ expansion to leading order, remanding to another work the definition of a time-dependent HartreeโFock (tdHF) approach (a generalization of the treatment given, for instance, in ). In fact these two methods are very closely related, as shown in , where several techniques to derive reasonable dynamical evolution equations for nonโequilibrium $`\varphi ^4`$ are compared.
We close this section by stressing that the introduction of both a UV and IR cutoff allows to easily derive the wellโknown rigorous result concerning the flatness of the effective potential. In fact $`V_{\text{eff}}(\overline{\varphi })`$ is a convex analytic function in a finite neighborhood of $`\overline{\varphi }=0`$, as long as the cutoffs are present, due to the uniqueness of the ground state. In the infrared limit $`L\mathrm{}`$, however, $`V_{\text{eff}}(\overline{\varphi })`$ might flatten around $`\overline{\varphi }=0`$. Of course this possibility would apply in case of spontaneous symmetry breaking, that is for a doubleโwell classical potential. This is a subtle and important point that will play a crucial role later on, even if the effective potential is relevant for the static properties of the model rather than the dynamical evolution out of equilibrium that interests us here. In fact, the dynamical evolution in QFT is governed by the CTP effective action and one might expect that, although nonโlocal in time, it asymptotically reduces to a multiple of the effective potential for trajectories of $`\overline{\varphi }(t)`$ with a fixed point at infinite time. In such case there should exist a oneโtoโone correspondence between fixed points and minima of the effective potential.
## III Large $`N`$ expansion at leading order
### A Definitions
In this section we consider the standard nonโperturbative approach to the $`\varphi ^4`$ model which is applicable also out of equilibrium, namely the large $`N`$ method as presented in . However we shall follow a different derivation which makes the gaussian nature of the $`N\mathrm{}`$ limit more explicit.
It is known that the theory described by the Hamiltonian (1) is well behaved for large $`N`$, provided that the quartic coupling constant $`\lambda _\text{b}`$ is rescaled with $`1/N`$. For example, it is possible to define a perturbation theory, based on the small expansion parameter $`1/N`$, in the framework of which one can compute any quantity at any chosen order in $`1/N`$. From the diagrammatic point of view, this procedure corresponds to a resummation of the usual perturbative series that automatically collects all the graphs of a given order in $`1/N`$ together . Moreover, it has been established since the early 80โs that the leading order approximation (that is the strict limit $`N\mathrm{}`$) is actually a classical limit , in the sense that there exists a classical system (i.e., a classical phase space, a Poisson bracket and a classical Hamiltonian) whose dynamics controls the evolution of all fundamental quantum observables, such as field correlation functions, in the $`N\mathrm{}`$ limit. For instance, from the absolute minimum of the classical Hamiltonian one reads the energy of the ground state, while the spectrum is given by the frequencies of small oscillations about this minimum, etc. etc.. We are here interested in finding an efficient and rapid way to compute the quantum evolution equations for some observables in the $`N\mathrm{}`$ limit, and we will see that this task is easily accomplished just by deriving the canonical Hamilton equations from the large $`N`$ classical Hamiltonian.
Following Yaffe , we write the quantum mechanical hamiltonian as
$$H=Nh(A,C)$$
(2)
in terms of the square matrices $`A`$, $`C`$ with operator entries ($`\mathit{\varpi }_k`$ is the canonical momentum conjugated to the real mode $`๐_k`$)
$$A_{kk^{}}=\frac{1}{N}๐_k๐_k^{},C_{kk^{}}=\frac{1}{N}\mathit{\varpi }_k\mathit{\varpi }_k^{}$$
These are example of โclassicalโ operators, whose two-point correlation functions factorize in the $`N\mathrm{}`$ limit. This can be shown by considering the coherent states
$$\mathrm{\Psi }_{z,q,p}(๐)=C(z)\mathrm{exp}\left[i\underset{k}{}๐_k๐_k\frac{1}{2N}\underset{kk^{}}{}z_{kk^{}}(๐_k๐_k)(๐_k^{}๐_k^{})\right]$$
(3)
where the complex symmetric matrix $`z`$ has a positive definite real part while $`๐_k`$ and $`๐_k`$ are real and coincide, respectively, with the coherent state expectation values of $`\mathit{\varpi }_k`$ and $`๐_k`$. As these parameters take all their possible values, the coherent states form an overcomplete set in the cutoff Hilbert space of the model. The crucial property which ensures factorization is that they become all orthogonal in the $`N\mathrm{}`$ limit. Moreover one can show that the coherent states parameters form a classical phase space with Poisson brackets
$$\{q_k^i,p_k^{}^j\}_{\mathrm{P}.\mathrm{B}.}=\delta _{kk^{}}\delta ^{ij},\{w_{kk^{}},v_{qq^{}}\}_{\mathrm{P}.\mathrm{B}.}=\delta _{kq}\delta _{k^{}q^{}}+\delta _{kq^{}}\delta _{k^{}q}$$
where $`w`$ and $`v`$ reparametrize $`z`$ as $`z=\frac{1}{2}w^1+iv`$. It is understood that the dimensionality of the vectors $`๐_k`$ and $`๐_k`$ is arbitrary but finite \[that is, only a finite number, say $`n`$, of pairs $`(\phi _k^i,\varpi _k^i)`$ may take a nonvanishing expectation value as $`N\mathrm{}`$\].
Once applied to the classical operators $`A_{kk^{}}`$ and $`C_{kk^{}}`$ the large $`N`$ factorization allow to obtain the classical hamiltonian by simply replacing $`A`$ and $`C`$ in eq. (2) by the coherent expectation values
$$A_{kk^{}}=๐_k๐_k^{}+w_{kk^{}},C_{kk^{}}=๐_k๐_k^{}+(vwv)_{kk^{}}+\frac{1}{4}(w^1)_{kk^{}}$$
In our situation, having assumed a uniform background expectation value for $`\mathit{\varphi }`$, we have $`๐_k=๐_k=0`$ for all $`k0`$; moreover, translation invariance implies that $`w`$ and $`v`$ are diagonal matrices, so that we may set
$$w_{kk^{}}=\sigma _k^2\delta _{kk^{}},v_{kk^{}}=\frac{s_k}{\sigma _k}\delta _{kk^{}}$$
in term of the canonical couples $`(\sigma _k,s_k)`$ which satisfy $`\{\sigma _k,s_k^{}\}_{\mathrm{P}.\mathrm{B}.}=\delta _{kk^{}}`$. Notice that the $`\sigma _k`$ are just the widths (rescaled by $`N^{1/2}`$) of the $`O(N)`$ symmetric and translation invariant gaussian coherent states.
Thus we find the classical hamiltonian
$$h_{\mathrm{cl}}=\frac{1}{2}(๐_0^2+m_\text{b}^2๐_0^2)+\frac{1}{2}\underset{k}{}\left[s_k^2+(k^2+m_\text{b}^2)\sigma _k^2+\frac{1}{4\sigma _k^2}\right]+\frac{\lambda _\text{b}}{4L^D}\left(๐_0^2+\underset{k}{}\sigma _k^2\right)^2$$
where by Hamiltonโs equations of motion $`๐_0=\dot{๐}_0`$ and $`s_k=\dot{\sigma }_k`$. The corresponding conserved energy density $`=L^Dh_{\mathrm{cl}}`$ may be written
$$\begin{array}{cc}\hfill & =๐ฏ+๐ฑ,๐ฏ=\frac{1}{2}\dot{\overline{\mathit{\varphi }}}^2+\frac{1}{2L^D}\underset{k}{}\dot{\sigma }_k^2\hfill \\ \hfill ๐ฑ& =\frac{1}{2L^D}\underset{k}{}\left(k^2\sigma _k^2+\frac{1}{4\sigma _k^2}\right)+V(\overline{\mathit{\varphi }}^2+\mathrm{\Sigma }),\mathrm{\Sigma }=\frac{1}{L^D}\underset{k}{}\sigma _k^2\hfill \end{array}$$
(4)
where $`\overline{\mathit{\varphi }}=L^{D/2}๐_0`$ and $`V`$ is the $`O(N)`$invariant quartic potential regarded as a function of $`\mathit{\varphi }^2`$, that is $`V(z)=\frac{1}{2}m_\text{b}^2z+\frac{1}{4}\lambda _\text{b}z^2`$. It is worth noticing that eq. (4) would apply as is to generic $`V(z)`$.
### B Static properties
Let us consider first the static aspects embodied in the effective potential $`V_{\text{eff}}(\overline{\mathit{\varphi }})`$, that is the minimum of the potential energy $`๐ฑ`$ at fixed $`\overline{\mathit{\varphi }}`$. We first define in a precise way the unbroken symmetry phase, in this large $`N`$ context, as the case when $`V_{\text{eff}}(\overline{\mathit{\varphi }})`$ has a unique minimum at $`\overline{\mathit{\varphi }}=0`$ in the limit of infinite volume. Minimizing $`๐ฑ`$ w.r.t. $`\sigma _k`$ yields
$$\begin{array}{cc}\hfill \sigma _k^2=\frac{1}{2\sqrt{k^2+M^2}},M^2& =m_\text{b}^2+2V^{}(\overline{\mathit{\varphi }}^2+\mathrm{\Sigma })\hfill \\ & =m_\text{b}^2+\lambda _\text{b}\overline{\mathit{\varphi }}^2+\frac{\lambda _\text{b}}{L^D}\underset{k}{}\frac{1}{2\sqrt{k^2+M^2}}\hfill \end{array}$$
(5)
that is the widths characteristic of a free theory with selfโconsistent mass $`M`$ fixed by the gap equation. By the assumption of unbroken symmetry, when $`\overline{\mathit{\varphi }}=0`$ and at infinite volume $`M`$ coincides with the equilibrium mass $`m`$ of the theory, that may be regarded as independent scale parameter. Since in the limit $`L\mathrm{}`$ sums are replaced by integrals
$$\mathrm{\Sigma }_{k^2\mathrm{\Lambda }^2}\frac{d^Dk}{(2\pi )^D}\sigma _k^2$$
we obtain the standard bare mass parameterization
$$m_\text{b}^2=m^2\lambda _\text{b}I_D(m^2,\mathrm{\Lambda }),I_D(z,\mathrm{\Lambda })_{k^2\mathrm{\Lambda }^2}\frac{d^Dk}{(2\pi )^D}\frac{1}{2\sqrt{k^2+z}}$$
(6)
and the renormalized gap equation
$$M^2=m^2+\lambda \overline{\varphi }^2+\lambda \left[I_D(M^2,\mathrm{\Lambda })I_D(m^2,\mathrm{\Lambda })\right]_{\text{finite}}$$
(7)
which implies, when $`D=3`$,
$$\lambda _\text{b}=\lambda \left(1\frac{\lambda }{8\pi ^2}\mathrm{log}\frac{2\mathrm{\Lambda }}{m\sqrt{e}}\right)^1$$
(8)
with a suitable choice of the finite part. No coupling constant renormalization occurs instead when $`D=1`$. The renormalized gap equation (7) may also be written quite concisely
$$\frac{M^2}{\widehat{\lambda }(M)}=\frac{m^2}{\widehat{\lambda }(m)}+\overline{\mathit{\varphi }}^2$$
(9)
in terms of the oneโloop running couplings constant
$$\widehat{\lambda }(\mu )=\lambda \left[1\frac{\lambda }{8\pi ^2}\mathrm{log}\frac{\mu }{m}\right]^1,\widehat{\lambda }(m)=\lambda ,\widehat{\lambda }(2\mathrm{\Lambda }e^{1/2})=\lambda _\text{b}$$
It is the Landau pole in $`\widehat{\lambda }(2\mathrm{\Lambda }e^{1/2})`$ that actually forbids the limit $`\mathrm{\Lambda }\mathrm{}`$. Hence we must keep the cutoff finite and smaller than $`\mathrm{\Lambda }_{\text{pole}}`$, so that the theory does retain a slight inverseโpower dependence on it. At any rate, there exists a very wide window where this dependence is indeed very weak for couplings of order one or less, since $`\mathrm{\Lambda }_{\text{pole}}=\frac{1}{2}m\mathrm{exp}(1/2+8\pi ^2/\lambda )m`$. Moreover, we see from eq. (9) that for $`\sqrt{\lambda }|\overline{\varphi }|`$ much smaller than the Landau pole there are two solutions for $`M`$, one โphysicalโ, always larger than $`m`$ and of the same order of $`m+\sqrt{\lambda }|\overline{\varphi }|`$, and one โunphysicalโ, close to the Landau pole.
One can now easily verify that the effective potential has indeed a unique minimum in $`\overline{\mathit{\varphi }}=0`$, as required. In fact, if we assign arbitrary $`\overline{\varphi }`$dependent values to the widths $`\sigma _k`$, (minus) the effective force reads
$$\frac{d}{d\overline{\varphi }^i}๐ฑ(\overline{\mathit{\varphi }},\{\sigma _k(\overline{\mathit{\varphi }})\})=M^2\overline{\varphi }^i+\underset{k}{}\frac{๐ฑ}{\sigma _k}\frac{d\sigma _k}{d\overline{\varphi }^i}$$
(10)
and reduces to $`M^2\overline{\varphi }^i`$ when the widths are extremal as in eq. (5); but $`M^2`$ is positive for unbroken symmetry and so $`\overline{\mathit{\varphi }}=0`$ is the unique minimum.
We define the symmetry as broken whenever the infinite volume $`V_{\text{eff}}`$ has more than one minimum. Of course, as long as $`L`$ is finite, $`V_{\text{eff}}`$ has a unique minimum in $`\overline{\mathit{\varphi }}=0`$, because of the uniqueness of the ground state in Quantum Mechanics, as already discussed in section II. Let us therefore proceed more formally and take the limit $`L\mathrm{}`$ directly on the potential energy $`๐ฑ`$. It reads
$$๐ฑ=\frac{1}{2}_{k^2\mathrm{\Lambda }^2}\frac{d^Dk}{(2\pi )^D}\left(k^2\sigma _k^2+\frac{1}{4\sigma _k^2}\right)+V(\overline{\mathit{\varphi }}^2+\mathrm{\Sigma }),\mathrm{\Sigma }=_{k^2\mathrm{\Lambda }^2}\frac{d^Dk}{(2\pi )^D}\sigma _k^2$$
where we write for convenience the treeโlevel potential $`V`$ in the positive definite form $`V(z)=\frac{1}{4}\lambda _\text{b}(z+m_\text{b}^2/\lambda _\text{b})^2`$. $`๐ฑ`$ is now the sum of two positive definite terms. Suppose there exists a configuration such that $`V(\overline{\mathit{\varphi }}^2+\mathrm{\Sigma })=0`$ and the first term in $`๐ฑ`$ is at its minimum. Then this is certainly the absolute minimum of $`๐ฑ`$. This configuration indeed exists at infinite volume when $`D=3`$:
$$\sigma _k^2=\frac{1}{2|k|},\overline{\mathit{\varphi }}^2=v^2,m_\text{b}^2=\lambda _\text{b}\left[v^2+I_3(0,\mathrm{\Lambda })\right]$$
(11)
where the nonnegative $`v`$ should be regarded as an independent parameter fixing the scale of the symmetry breaking. It replaces the mass parameter $`m`$ of the unbroken symmetry case: now the theory is massless in accordance with Goldstone theorem. On the contrary, if $`D=1`$ this configuration is not allowed due to the infrared divergences caused by the massless nature of the width spectrum. This is just the standard manifestation of MerminโWagnerโColeman theorem that forbids continuous symmetry breaking in a twoโdimensional spaceโtime .
At finite volumes we cannot minimize the first term in $`๐ฑ`$ since this requires $`\sigma _0`$ to diverge, making it impossible to keep $`V(\overline{\mathit{\varphi }}^2+\mathrm{\Sigma })=0`$. In fact we know that the uniqueness of the ground state with finitely many degrees of freedom implies the minimization equations (5) to hold always true with a $`M^2`$ strictly positive. Therefore, broken symmetry should manifest itself as the situation in which the equilibrium value of $`M^2`$ is a positive definite function of $`L`$ which vanishes in the $`L\mathrm{}`$ limit.
We can confirm this qualitative conclusion as follows. We assume that the bare mass has the form given in eq. (11) and that $`\overline{\mathit{\varphi }}^2=v^2`$ too. Minimizing the potential energy leads always to the massive spectrum, eq. (5), with the gap equation
$$\frac{M^2}{\lambda _\text{b}}=\frac{1}{2L^3M}+\frac{1}{2L^3}\underset{k0}{}\frac{1}{\sqrt{k^2+M^2}}\frac{\mathrm{\Lambda }^2}{8\pi ^2}$$
(12)
If $`M^2>0`$ does not vanish too fast for large volumes, or stays even finite, then the sum on the modes has a behavior similar to the corresponding infinite volume integral: there is a quadratic divergence that cancels the infinite volume contribution, and a logarithmic one that renormalizes the bare coupling. The direct computation of the integral would produce a term containing the $`M^2\mathrm{log}(\mathrm{\Lambda }/M)`$. This can be split into $`M^2[\mathrm{log}(\mathrm{\Lambda }/v)\mathrm{log}(M/v)]`$ by using $`v`$ as mass scale. The first term renormalizes the coupling correctly, while the second one vanishes if $`M^2`$ vanishes in the infinite volume limit.
When $`L\mathrm{}`$, the asymptotic solution of (12) reads
$$M=\left(\frac{\lambda }{2}\right)^{1/3}L^1+\mathrm{h}.\mathrm{o}.\mathrm{t}.$$
that indeed vanishes in the limit. Note also that the exponent is consistent with the assumption made above that $`M`$ vanishes slowly enough to approximate the sum over $`k0`$ with an integral with the same $`M`$.
Let us now consider a state whose field expectation value $`\overline{\mathit{\varphi }}^2`$ is different from $`v^2`$. If $`\overline{\mathit{\varphi }}^2>v`$, the minimization equations (5) leads to a positive squared mass spectrum for the fluctuations, with $`M^2`$ given selfโconsistently by the gap equation. On the contrary, as soon as $`\overline{\mathit{\varphi }}^2<v^2`$, one immediately see that a positive $`M^2`$ cannot solve the gap equation
$$M^2=\lambda _\text{b}\left(\overline{\mathit{\varphi }}^2v^2+\frac{\sigma _0^2}{L^3}+\frac{1}{2L^3}\underset{k0}{}\frac{1}{\sqrt{k^2+M^2}}\frac{\mathrm{\Lambda }^2}{8\pi ^2}\right)$$
if we insist on the requirement that $`\sigma _0`$ not be macroscopic. In fact, the r.h.s. of the previous equation is negative, no matter which positive value for the effective mass we choose, at least for $`L`$ large enough. But nothing prevent us to consider a static configuration for which the amplitude of the zero mode is macroscopically large (i.e. it rescales with the volume $`L^3`$). Actually, if we choose
$$\frac{\sigma _0^2}{L^3}=v^2\overline{\mathit{\varphi }}^2+\frac{1}{2L^3M}$$
we obtain the same equation as we did before and the same value for the potential, that is the minimum, in the limit $`L\mathrm{}`$. Note that at this level the effective mass $`M`$ needs not to have the same behavior in the $`L\mathrm{}`$ limit, but it is free of rescaling with a different power of $`L`$. We can be even more precise: we isolate the part of the potential that refers to the zero mode width $`\sigma _0`$ ($`\mathrm{\Sigma }^{}`$ does not contain the $`\sigma _0`$ contribution)
$$\frac{1}{2}\left[m_\text{b}^2+\lambda _\text{b}\left(\overline{\mathit{\varphi }}^2+\mathrm{\Sigma }^{}\right)\right]\frac{\sigma _0^2}{L^3}+\frac{\lambda _\text{b}}{4}\frac{\sigma _0^4}{L^6}+\frac{1}{8L^3\sigma _0^2}$$
and we minimize it, keeping $`\overline{\mathit{\varphi }}^2`$ fixed. The minimum is attained at $`t=\sigma _0^2/L^3`$ solution of the cubic equation
$$\lambda _\text{b}t^3+\alpha \lambda _\text{b}t^2\frac{1}{4}L^6=0$$
where $`\alpha =\overline{\mathit{\varphi }}^2v^2+\mathrm{\Sigma }^{}I_3(0,\mathrm{\Lambda })`$. Note that $`\lambda _\text{b}\alpha `$ depends on $`L`$ and it has a finite limit in infinite volume: $`\lambda (\overline{\mathit{\varphi }}^2v^2)`$. The solution of the cubic equation is
$$\lambda _\text{b}t=\lambda _\text{b}(v^2\overline{\mathit{\varphi }}^2)+\frac{1}{4}[L^3(v^2\overline{\mathit{\varphi }}^2)]^2+\mathrm{h}.\mathrm{o}.\mathrm{t}.$$
from which the effective mass can be identified as proportional to $`L^3`$. The stability equations for all the other modes can now be solved by a massive spectrum, in a much similar way as before.
Since $`\sigma _0`$ is now macroscopically large, the infinite volume limit of the $`\sigma _k`$ distribution (that gives a measure of the transverse fluctuations in the $`O(N)`$ model) develop a $`\delta `$like singularity, signalling a Bose condensation of the Goldstone bosons:
$$\sigma _k^2=(v^2\overline{\mathit{\varphi }}^2)\delta ^{(D)}(k)+\frac{1}{2k}$$
(13)
At the same time it is evident that the minimal potential energy is the same as when $`\overline{\mathit{\varphi }}^2=v^2`$, that is the effective potential flattens, in accord with the Maxwell construction.
Eq. (13) corresponds in configuration space to the $`2`$point correlation function
$$\underset{N\mathrm{}}{lim}\frac{\mathit{\varphi }(x)\mathit{\varphi }(y)}{N}=\overline{\mathit{\varphi }}^2+\frac{d^Dk}{(2\pi )^D}\sigma _k^2e^{ik(xy)}=C(\overline{\mathit{\varphi }}^2)+\mathrm{\Delta }_D(xy)$$
(14)
where $`\mathrm{\Delta }_D(xy)`$ is the massless freeโfield equalโtime correlator, while
$$C(\overline{\mathit{\varphi }}^2)=v^2\mathrm{\Theta }(v^2\overline{\mathit{\varphi }}^2)+\overline{\mathit{\varphi }}^2\mathrm{\Theta }(\overline{\mathit{\varphi }}^2v^2)=\text{max}(v^2,\overline{\mathit{\varphi }}^2)$$
(15)
This expression can be extended to unbroken symmetry by setting in that case $`C(\overline{\mathit{\varphi }}^2)=\overline{\mathit{\varphi }}^2`$.
Quite evidently, when eq. (15) holds, symmetry breaking can be inferred from the limit $`|xy|\mathrm{}`$, if clustering is assumed , since $`\mathrm{\Delta }_D(xy)`$ vanishes for large separations. Of course this contradicts the infinite volume limit of the finiteโvolume definition, $`\overline{\mathit{\varphi }}=lim_N\mathrm{}N^{1/2}\mathit{\varphi }(x)`$, except at the extremal points $`\overline{\mathit{\varphi }}^2=v^2`$.
In fact the $`L\mathrm{}`$ limit of the finite volume states with $`\overline{\mathit{\varphi }}^2<v^2`$ do violate clustering, because they are linear superpositions of vectors belonging to superselected sectors and therefore they are indistinguishable from statistical mixtures. We can give the following intuitive picture for large $`N`$. Consider any one of the superselected sectors based on a physical vacuum with $`\overline{\mathit{\varphi }}^2=v^2`$. By condensing a macroscopic number of transverse Goldstone bosons at zeroโmomentum, one can build coherent states with rotated $`\overline{\mathit{\varphi }}`$. By incoherently averaging over such rotated states one obtains new states with field expectation values shorter than $`v`$ by any prefixed amount. In the large $`N`$ approximation this averaging is necessarily uniform and is forced upon us by the residual $`O(N1)`$ symmetry.
### C Outโofโequilibrium dynamics
We now turn to the dynamics out of equilibrium in this large $`N`$ context. It is governed by the equations of motion derived from the total energy density $``$ in eq. (4), that is
$$\frac{d^2\overline{\mathit{\varphi }}}{dt^2}=M^2\overline{\mathit{\varphi }},\frac{d^2\sigma _k}{dt^2}=(k^2+M^2)\sigma _k+\frac{1}{4\sigma _k^3}$$
(16)
where the generally timeโdependent effective squared mass $`M^2`$ is given by
$$M^2=m^2+\lambda _\text{b}\left[\overline{\mathit{\varphi }}^2+\mathrm{\Sigma }I_D(m^2,\mathrm{\Lambda })\right]$$
(17)
in case of unbroken symmetry and
$$M^2=\lambda _\text{b}\left[\overline{\mathit{\varphi }}^2v^2+\mathrm{\Sigma }I_3(0,\mathrm{\Lambda })\right]$$
(18)
for broken symmetry in $`D=3`$.
At time zero, the specific choice of initial conditions for $`\sigma _k`$ that give the smallest energy contribution, that is
$$\dot{\sigma }_k=0,\sigma _k^2=\frac{1}{2\sqrt{k^2+M^2}}$$
(19)
turns eq. (17) into the usual gap equation (5). For any value of $`\overline{\varphi }`$ this equation has one solution smoothly connected to the value $`M=m`$ at $`\overline{\varphi }=0`$. Of course other initial conditions are possible. The only requirement is that the corresponding energy must differ from that of the ground state by an ultraviolet finite amount, as it occurs for the choice (19). In fact this is guaranteed by the gap equation itself, as evident from eq. (10): when the widths $`\sigma _k`$ are extremal the effective force is finite, and therefore so are all potential energy differences.
This simple argument needs a refinement in two respects.
Firstly, in case of symmetry breaking the formal energy minimization w.r.t. $`\sigma _k`$ leads always to eqs. (19), but these are acceptable initial conditions only if the gap equation that follows from eq. (18) in the $`L\mathrm{}`$ limit, namely
$$M^2=\lambda _\text{b}\left[\overline{\mathit{\varphi }}^2v^2+I_D(M^2,\mathrm{\Lambda })I_D(0,\mathrm{\Lambda })\right]$$
(20)
admits a nonnegative, physical solution for $`M^2`$.
Secondly, ultraviolet finiteness only requires that the sum over $`k`$ in eq. (10) be finite and this follows if eq. (19) holds at least for $`k`$ large enough, solving the issue raised in the first point: negative $`M^2`$ are allowed by imposing a new form of gap equation
$$M^2=\lambda _\text{b}\left[\overline{\mathit{\varphi }}^2v^2+\frac{1}{L^D}\underset{k^2<|M^2|}{}\sigma _k^2+\frac{1}{L^D}\underset{k^2>|M^2|}{}\frac{1}{2\sqrt{k^2|M^2|}}I_D(0,\mathrm{\Lambda })\right]$$
(21)
where all $`\sigma _k`$ with $`k^2<|M^2|`$ are kept free (but all by hypothesis microscopic) initial conditions. Of course there is no energy minimization in this case. To determine when this new form is required, we observe that, neglecting the inverseโpower corrections in the UV cutoff we may write eq. (20) in the following form
$$\frac{M^2}{\widehat{\lambda }(M)}=\overline{\mathit{\varphi }}^2v^2$$
(22)
There exists a positive solution $`M^2`$ smoothly connected to the ground state, $`\overline{\mathit{\varphi }}^2=v^2`$ and $`M^2=0`$, only provided $`\overline{\mathit{\varphi }}^2v^2`$. So, in the large $`N`$ limit, as soon as we start with $`\overline{\mathit{\varphi }}^2v^2`$, we cannot satisfy the gap equation with a positive value of $`M^2`$.
Once a definite choice of initial conditions is made, the system of differential equations (16), (17) or (18) can be solved numerically with standard integration algorithms. This has been already done by several authors , working directly in infinite volume, with the following general results. In the case of unbroken symmetry it has been established that the $`\sigma _k`$ corresponding to wavevectors $`k`$ in the soโcalled forbidden bands with parametric resonances grow exponentially in time until their growth is shut off by the backโreaction. For broken symmetry it is the region in $`k`$space with the spinodal instabilities caused by an initially negative $`M^2`$, whose widths grow exponentially before the backโreaction shutoff. After the shutoff time the effective mass tends to a positive constant for unbroken symmetry and to zero for broken symmetry (in D=3), so that the only width with a chance to keep growing indefinitely is $`\sigma _0`$ for broken symmetry.
Of course, in all these approaches the integration over modes in the backโreaction $`\mathrm{\Sigma }`$ cannot be done exactly and is always replaced by a discrete sum of a certain type, depending on the details of the algorithms. Hence there exists always an effective infrared cutoff, albeit too small to be detectable in the numerical outputs. A possible troublesome aspect of this is the proper identification of the zeroโmode width $`\sigma _0`$. Even if a (rather arbitrary) choice of discretization is made where a $`\sigma _0`$ appears, it is not really possible to determine whether during the exponential growth or after such width becomes of the order of the volume. Our aim is just to answer this question and therefore we perform our numerical evolution in finite volumes of several growing sizes. Remanding to the appendix for the details of our method, we summarize our results in the next subsection.
### D Numerical results
After a careful study in $`D=3`$ of the scaling behavior of the dynamics with respect to different values of $`L`$, the linear size of the system, we reached the following conclusion: there exist a $`L`$dependent time, that we denote by $`\tau _L`$, that splits the evolution in two parts; for $`t\tau _L`$, the behavior of the system does not differ appreciably from its counterpart at infinite volume, while finite volume effects abruptly alter the evolution as soon as $`t`$ exceeds $`\tau _L`$; moreover
* $`\tau _L`$ is proportional to the linear size of the box $`L`$ and so it rescales as the cubic root of the volume.
* $`\tau _L`$ does not depend on the value of the quartic coupling constant $`\lambda `$, at least in a first approximation.
The figures show the behavior of the width of the zero mode $`\sigma _0`$ (see Fig. 1), of the squared effective mass $`M^2`$ (see Fig. 2 ) and of the backโreaction $`\mathrm{\Sigma }`$ (see Fig. 3), in the more interesting case of broken symmetry. The initial conditions are chosen in several different ways (see the appendix for details), but correspond to a negative $`M^2`$ at early times with the initial widths all microscopic, that is at most of order $`L^{1/2}`$. This is particularly relevant for the zeroโmode width $`\sigma _0`$, which is instead macroscopic in the lowest energy state when $`\overline{\mathit{\varphi }}^2<v^2`$, as discussed above. As for the background, the figures are relative to the simplest case $`\overline{\mathit{\varphi }}=0=\dot{\overline{\mathit{\varphi }}}`$, but we have considered also initial conditions with $`\overline{\mathit{\varphi }}>0`$, reproducing the โdynamical Maxwell constructionโ observed in ref. . At any rate, for the purposes of this work, above all it is important to observe that, due to the quantum backโreaction, $`M^2`$ rapidly becomes positive, within the soโcalled spinodal time , and then, for times before $`\tau _L`$, the weakly dissipative regime takes place where $`M^2`$ oscillates around zero with amplitude decreasing as $`t^1`$ and a frequency fixed by the largest spinodal wavevector, in complete agreement with the infiniteโvolume results . Correspondingly, after the exponential growth until the spinodal time, the width of the zeroโmode grows on average linearly with time, reaching a maximum for $`t\tau _L`$. Precisely, $`\sigma _0`$ performs small amplitude oscillations with the same frequency of $`M^2`$ around a linear function of the form $`A+Bt`$, where $`A,B\lambda ^{1/2}`$ (see Fig. 4), confirming what already found in refs. ; then quite suddenly it turns down and enters long irregular Poincarรฉโlike cycles. Since the spinodal oscillation frequency does not depend appreciably on $`L`$, the curves of $`\sigma _0`$ at different values of $`L`$ are practically identical for $`t<\tau _L`$. After a certain number of complete oscillations, a number that scales linearly with $`L`$, a small change in the behavior of $`M^2`$ (see Fig. 5) determines an inversion in $`\sigma _0`$ (see Fig. 6), evidently because of a phase crossover between the two oscillation patterns. Shortly after $`\tau _L`$ dissipation practically stops as the oscillations of $`M^2`$ stop decreasing in amplitude and become more and more irregular, reflecting the same irregularity in the evolution of the widths.
We can give a straightforward physical interpretation for the presence of the time scale $`\tau _L`$. As shown in , long after the spinodal time $`t_1`$, the effective mass oscillates around zero with a decreasing amplitude and affects the quantum fluctuations in such a way that the equalโtime twoโpoint correlation function contains a timeโdependent nonโperturbative disturbance growing at twice the speed of light. This is interpreted in terms of large numbers of Goldstone bosons equally produced at any point in space (due to translation invariance) and radially propagating at the speed of light. This picture applies also at finite volumes, in the bulk, for volumes large enough. Hence, due to our periodic boundary conditions, after a time exactly equal to $`L/2`$ the forward wave front meets the backward wave front at the opposite point with respect to the source, and the propagating wave starts interfering with itself and heavily changes the dynamics with respect to that in infinite volume. This argument leads us to give the value of $`\pi `$ for the proportionality coefficient between $`\tau _L`$ and $`L/2\pi `$, prevision very well verified by the numerical results, as can be inferred by a look at the figures.
The main consequence of this scenario is that the linear growth of the zeroโmode width at infinite volume should not be interpreted as a standard form of BoseโEinstein Condensation (BEC), occurring with time, but should be consistently considered as โnovelโ form of dynamical BEC, as found by the authors of . In fact, if a macroscopic condensation were really there, the zero mode would develop a $`\delta `$ function in infinite volume, that would be announced by a width of the zero mode growing to values $`O(L^{3/2})`$ at any given size $`L`$. Now, while it is surely true that when we push $`L`$ to infinity, also the time $`\tau _L`$ tends to infinity, allowing the zero mode to grow indefinitely, it is also true that, at any fixed though arbitrarily large volume, the zero mode never reaches a width $`O(L^{3/2})`$, just because $`\tau _LL`$. In other words, if we start from initial conditions where $`\sigma _0`$ is microscopic, then it never becomes macroscopic later on.
On the other hand, looking at the behavior of the mode functions of momenta $`k=(2\pi /L)n`$ for $`n`$ fixed but for different values of $`L`$, one realizes that they obey a scaling similar to that observed for the zeroโmode: they oscillate in time with an amplitude and a period that are $`O(L)`$ (see fig. 7 and 8). Thus, each mode shows a behavior that is exactly half a way between a macroscopic amplitude \[i.e. $`O(L^{3/2})`$\] and a usual microscopic one \[i.e. at most $`O(L^{1/2})`$\]. This means that the spectrum of the quantum fluctuations at times of the order of the diverging volume can be interpreted as a massless spectrum of interacting Goldstone modes, because their power spectrum develops in the limit a $`1/k^2`$ singularity, rather than the $`1/k`$ pole typical of free massless modes. As a consequence the equalโtime field correlation function \[see eq. (14)\] will fall off as $`|๐๐|^1`$ for large separations smaller only than the diverging elapsed time. This is in accord with what found in , where the same conclusion where reached after a study of the correlation function for the scalar field in infinite volume.
The fact that each mode never becomes macroscopic, if it started microscopic, might be regarded as a manifestation of unitarity in the large $`N`$ approximation: an initial gaussian state with only microscopic widths satisfies clustering and clustering cannot be spoiled by a unitary time evolution. As a consequence, in the infiniteโvolume lateโtime dynamics, the zeroโmode width $`\sigma _0`$ does not play any special role and only the behavior of $`\sigma _k`$ as $`k0`$ is relevant. As already stated above, it turns out from our numerics as well as from refs. that this behavior is of a novel type characteristic both of the outโofโequilibrium dynamics and of the equilibrium finiteโtemperature theory , with $`\sigma _k1/k`$.
A final comment should be made about the periodic boundary conditions used for these simulations. This choice guarantees the translation invariance of the dynamics needed to consider a stable uniform background. If we had chosen other boundary conditions (Dirichlet or Neumann, for instance), the translation symmetry would have been broken and an uniform background would have become non-uniform pretty soon. Of course, we expect the bulk behavior to be independent of the particular choice for the boundary conditions in the infinite volume limit, even if a rigorous proof of this statement is still lacking.
## IV Discussion and perspectives
In this work we have presented a rather detailed study of the dynamical evolution out of equilibrium, in finite volume (a cubic box of size $`L`$ in $`3`$D), for the $`\varphi ^4`$ QFT in the large $`N`$ limit. For comparison, we have also analyzed some static characteristics of the theory both in unbroken and broken symmetry phases.
We have reached the conclusion, based on strong numerical evidence, that the linear growth of the zeroโmode quantum fluctuations, observed already in the large $`N`$ approach of refs. , may be consistently interpreted as a โnovelโ form of dynamical BoseโEinstein condensation, different from the traditional one in finite temperature field theory at equilibrium. In fact, in finite volume, $`\sigma _0`$ never grows to $`O(L^{3/2})`$ if it starts from a microscopic value, that is at most of order $`L^{1/2}`$. On the other hand all longโwavelength fluctuations rapidly become of order $`L`$, signalling a novel infrared behavior quite different from free massless fields at equilibrium \[recall that the large $`N`$ approximation is of mean field type, with no direct interaction among particle excitations\]. This is in agreement with the properties of the twoโpoint function determined in .
The numerical evidence for the linear dependence of $`\tau _L`$ on $`L`$ is very strong, and the qualitative argument given in the previous section clearly explains the physics that determines it. Nonetheless a solid analytic understanding of the detailed (quantitative) mechanism that produces the inversion of $`\dot{\sigma }_0`$ around $`\tau _L`$ and its subsequent irregular behavior, is, at least in our opinion, more difficult to obtain. One could use intuitive and generic arguments like the quantization of momentum in multiples of $`2\pi /L`$, but the evolution equations do not have any simple scaling behavior towards a universal form, when mass dimensions are expressed in multiples of $`2\pi /L`$ and time in multiples of $`L`$. Moreover, the qualitative form of the evolution depends heavily on our choice of initial conditions. In fact, before finite volumes effects show up, the trajectories of the quantum modes are rather complex but regular enough, having a small-scale quasi-periodic almost mode-independent motion within a large-scale quasi-periodic mode-dependent envelope, with a very delicate resonant equilibrium \[Cfr. Fig. 1 and 7\]. Apparently \[Cfr. Fig. 5 and 6\], it is a sudden small beat that causes the turn around of the zero-mode and of the other low-lying modes (with many thousands of coupled modes, it is very difficult for the delicate resonant equilibrium to fully come back ever again), but we think that a deeper comprehension of the nonโlinear coupled dynamics is needed in order to venture into a true analytic explanation.
On the other hand it is not difficult to understand why $`\tau _L`$ does not depend appreciably on the coupling constant: when finite-volume effects first come in, that is when the wave propagating at the speed of light first starts to interfere with itself, the quantum back-reaction $`\lambda \mathrm{\Sigma }`$ has settled on values of order 1, because the time $`\tau _LL/2`$ is much greater than the spinodal time $`t_1`$. The slope of the linear envelope of the zero mode does depend on $`\lambda `$ because it is fixed by the early exponential growth. Similarly, it is easy to realize that the numerical integrations of refs. over continuum momenta correspond roughly to an effective volume much larger than ours, so that the calculated evolution remained far away from the onset of finite-volume effects.
The main limitation of the large $`N`$ approximation, as far as the evolution of the widths $`\sigma _k`$ is concerned, is in its intrinsic gaussian nature. In fact, one might envisage a scenario in which, while gaussian fluctuations stay microscopic, nonโgaussian fluctuations grow in time to a macroscopic size. Therefore, in order to clarify this point and go beyond the gaussian approximation, we are going to consider, in a forthcoming work , a timeโdependent HF approximation capable in principle of describing the dynamics of nonโgaussian fluctuation of a single scalar field with $`\varphi ^4`$ interaction.
Another open question concerns the connection between the minima of the effective potential and the asymptotic values for the evolution of the background, within the simplest gaussian approximation. As already pointed out in , a dynamical Maxwell construction occurs for the $`O(N)`$ model in infinite volume and at leading order in $`1/N`$ in case of broken symmetry, in the sense that any value of the background within the spinodal region can be obtained as large time limit of the evolution starting from suitable initial conditions. It would be very enlightening if we could prove this โexperimentalโ result by first principles arguments, based on CTP formalism. Furthermore, preliminary numerical evidence suggests that something similar occurs also in the Hartree approximation for a single field, but a more thorough and detailed analysis is needed.
It would be interesting also to study the dynamical realization of the Goldstone paradigm, namely the asymptotic vanishing of the effective mass in the broken symmetry phases, in different models; this issue needs further study in the $`2D`$ case , where it is known that the Goldstone theorem is not valid.
## V Acknowledgements
C. D. thanks D. Boyanovsky, H. de Vega, R. Holman and M. Simionato for very interesting discussions. C. D. and E. M. thank MURST and INFN for financial support. Part of the results contained in this work has been presented by E. M. at the 1999 CRM Summer School on โTheoretical Physics at the End of the XXth Centuryโ, held in Banff (Alberta), Canada, June 27 - July 10, 1999. E. M. thanks CRM (Universitรฉ de Montrรจal) for partial financial support.
## A Details of the numerical analysis
We present here the precise form of the evolution equations for the field background and the quantum mode widths, which control the outโofโequilibrium dynamics of the $`\varphi ^4`$ model in finite volume at the leading order in the $`1/N`$ approach, as described in sections III C. We restrict here our attention to the tridimensional case.
Let us begin by noticing that each eigenvalue of the Laplacian operator in a $`3D`$ finite volume is of the form $`k_n^2=\left(\frac{2\pi }{L}\right)^2n`$, where $`n`$ is a nonโnegative integer obtained as the sum of three squared integers, $`n=n_x^2+n_y^2+n_z^2`$. Then we associate a degeneracy factor $`g_n`$ to each eigenvalue, representing the number of different ordered triples $`(n_x,n_y,n_z)`$ yielding the same $`n`$. One may verify that $`g_n`$ takes on the continuum value of $`4\pi k^2`$ in the infinite volume limit, where $`k=\left(\frac{2\pi }{L}\right)^2n`$ is kept fixed when $`L\mathrm{}`$.
Now, the system of coupled ordinary differential equations is, in case of the large $`N`$ approach,
$$\left[\frac{d^2}{dt^2}+M^2\right]\varphi =0,\left[\frac{d^2}{dt^2}+\left(\frac{2\pi }{L}\right)^2n+M^2\right]\sigma _n\frac{1}{4\sigma _n^3}=0$$
(A1)
where the index $`n`$ ranges from $`0`$ to $`๐ฉ^2`$, $`๐ฉ=\mathrm{\Lambda }L/2\pi `$ and $`M^2(t)`$ is defined by the eq. (17) in case of unbroken symmetry and by eq. (18) in case of broken symmetry. The backโreaction $`\mathrm{\Sigma }`$ reads, in the notations of this appendix
$$\mathrm{\Sigma }=\frac{1}{L^D}\underset{n=0}{\overset{๐ฉ^2}{}}g_n\sigma _n^2$$
Technically it is simpler to treat an equivalent set of equations, which are formally linear and do not contain the singular Heisenberg term $`\sigma _n^3`$. This is done by introducing the complex mode amplitudes $`z_n=\sigma _n\mathrm{exp}(i\theta _n)`$, where the phases $`\theta _n`$ satisfy $`\sigma _n^2\dot{\theta }_n=1`$. Then we find a discrete version of the equations studied for instance in ref ,
$$\left[\frac{d^2}{dt^2}+\left(\frac{2\pi }{L}\right)^2n+M^2\right]z_n=0,\mathrm{\Sigma }=\frac{1}{L^D}\underset{n=0}{\overset{๐ฉ^2}{}}g_n|z_n|^2$$
(A2)
subject to the Wronskian condition
$$z_n\dot{\overline{z_n}}\overline{z_n}\dot{z}_n=i$$
One realizes that the Heisenberg term in $`\sigma _n`$ corresponds to the centrifugal potential for the motion in the complex plane of $`z_n`$.
Let us now come back to the equations (A2). To solve these evolution equations, we have to choose suitable initial conditions respecting the Wronskian condition. In case of unbroken symmetry, once we have fixed the value of $`\varphi `$ and its first time derivative at initial time, the most natural way of fixing the initial conditions for the $`z_n`$ is to require that they minimize the energy at $`t=0`$. We can obviously fix the arbitrary phase in such a way to have a real initial value for the complex mode functions
$$z_n(0)=\frac{1}{\sqrt{2\mathrm{\Omega }_n}}\frac{dz_n}{dt}(0)=ฤฑ\sqrt{\frac{\mathrm{\Omega }_n}{2}}$$
where $`\mathrm{\Omega }_n=\sqrt{k_n^2+M^2(0)}`$. The initial squared effective mass $`M^2(t=0)`$, has to be determined self-consistently, by means of its definition (17).
In case of broken symmetry, the gap equation is a viable mean for fixing the initial conditions only when $`\varphi `$ lies outside the spinodal region \[see eq (22)\]; otherwise, the gap equation does not admit a positive solution for the squared effective mass. In that case, we have to resort to other methods, in order to choose the initial conditions. Following the discussion presented in III C, one possible choice is to set $`\sigma _k^2=\frac{1}{2\sqrt{k^2+|M^2|}}`$ for $`k^2<|M^2|`$ in eq. (21) and then solve the corresponding gap equation (21). An other acceptable choice would be to solve the gap equation (21), once we have set a massless spectrum for all the spinodal modes but the zero mode, which is started from an arbitrary, albeit microscopic, value.
There is actually a third possibility, that is in some sense half a way between the unbroken and broken symmetry case. We could allow for a time dependent bare mass, in such a way to simulate a sort of cooling down of the system. In order to do that, we could start with a unbroken symmetry bare potential (which fixes initial conditions naturally via the gap equation) and then turn to a broken symmetry one after a short interval of time. This evolution is achieved by a proper interpolation in time of the two inequivalent parameterizations of the bare mass, eqs. (6) and (11).
We looked for the influence this different choices could produce in the results and indeed they depend very little and only quantitatively from the choice of initial condition we make.
As far as the numerical algorithm is concerned, we used a $`4`$th order Runge-Kutta algorithm to solve the coupled differential equations (A2), performing the computations in boxes of linear size ranging from $`L=20\pi `$ to $`L=400\pi `$ and verifying the conservation of the Wronskian to order $`10^5`$. Typically, we have chosen values of $`๐ฉ`$ corresponding to the UV cutoff $`\mathrm{\Lambda }`$ equal to small multiples of $`m`$ for unbroken symmetry and of $`v\sqrt{\lambda }`$ for broken symmetry. In fact, the dynamics is very weakly sensitive to the presence of the ultraviolet modes, once the proper subtractions are performed. This is because only the modes inside the unstable (forbidden or spinodal) band grow exponentially fast, reaching soon non perturbative amplitudes (i.e. $`\lambda ^{1/2}`$), while the modes lying outside the unstable band remains perturbative, contributing very little to the quantum backโreaction and weakly affecting the overall dynamics. The unique precaution to take is that the initial conditions be such that the unstable band lay well within the cutoff. |
warning/0001/astro-ph0001133.html | ar5iv | text | # Giant protostellar outflows revealed by infrared imaging
## 1 Introduction
Star formation is known to be intimately connected with outflow activity, especially when the star is in its main accretion phase, still deeply embedded in its parent cloud (e.g. Bontemps et al. 1996). Many low-mass young stellar objects are found to drive well-collimated jets. Interestingly, the length-scales of these jets have increased with advances in detector technology, namely bigger imaging arrays. It now seems that flows from low-mass stars extending over several parsecs are no great exception (e.g. Bally & Devine 1994, 1997; Devine et al. 1997; Eislรถffel & Mundt 1997; Reipurth et al. 1997, 1998; Stanke et al. 1999); however, the great majority of them have been found at optical wavelengths.
In this paper we demonstrate that infrared imaging is also capable of revealing large scale flows. This is complementary progress, since infrared imaging is a much more efficient way of finding jets deep in molecular clouds and interacting with the clouds, as these are hidden from view at optical wavelengths (e.g. the infrared jets HH 211; McCaughrean et al. 1994; and HH 212; Zinnecker et al. 1998). Only a combination of infrared and optical imaging can fully reveal the true number of jets in a cloud and thus enable us to address questions concerning the role of outflows from young stars for the evolution of a cloud, e.g. in terms of momentum-injection to stabilize the cloud against collapse.
The data presented here are part of an infrared survey for embedded jets in the Orion A cloud. The full survey will cover an area of about 1 square degree with a sensitivity comparable to that of previous infrared observations of individual protostellar jets. Initial results from the survey were published by Stanke et al. (1998, Paper I in the following). The new results presented here confirm that infrared imaging is an efficient way to search large areas for embedded flows hidden from view at optical wavelengths, but which are bright in H<sub>2</sub> emission in the near-infrared.
## 2 Observations
The infrared data presented here were obtained during several observing runs (Jan. 10-11 1998, Oct. 23-26 1998, Dec. 5 1998) using the Omega Prime wide-field near-IR camera (Bizenberger et al. 1998; McCaughrean et al., in prep.) on the Calar Alto 3.5 m telescope. The camera uses a 1024$`\times `$1024 pixel HgCdTe array: at 0.4 arcsec per pixel, the field-of-view is 6.7 arcmin$`\times `$6.7 arcmin. Images were taken through a 1 % filter centred on the $`\mathrm{v}=1`$โ0 S(1) line of H<sub>2</sub> at 2.12 $`\mu `$m and a broad-band K filter (1.944โ2.292 $`\mu `$m) in order to discriminate line and continuum sources. The images presented here are excerpts of larger mosaics, each of them a combination of 36 overlapping images. In the 2.12 $`\mu `$m filter, the net integration time was 10 min in the central part of each mosaic and 5 min at its edges; at K, the corresponding integration times were 1 min and 30 sec. We reach a 5$`\sigma `$ limiting H<sub>2</sub> surface brightness of $`10^{18}`$ W m<sup>-2</sup> arcsec<sup>-2</sup> (K$``$17 for continuum point sources). Standard data reduction techniques were used to sky subtract, flat field, and mosaic the data (McCaughrean et al. 1994).
The data discussed in this paper are part of a much larger survey in Orion A split into a number of overlapping fields. We only assign designations to those H<sub>2</sub> features which are part of the outflows discussed here. We adopt a nomenclature similar to that used in Paper I, with an additional digit indicating the survey field. For example, SMZ 7-10 means H<sub>2</sub> feature number 10 in survey field 7. A complete catalogue of all features found during our survey is deferred to a future paper.
Millimetre continuum maps at 1.3 mm were taken of several of the most interesting new H<sub>2</sub> sources during an observing run in February 1999, using the MPIfR 37 channel bolometer array on the IRAM 30 m telescope on Pico Veleta. The details of these observations as well as the complete results will be discussed elsewhere; here we will only present data related to the flows discussed in this paper.
## 3 The HH 43/38 outflow
### 3.1 Previous studies
HH 43 and HH 38 were discovered via optical photography by Haro (1953; see Herbig 1974). CCD images of these objects were presented by Strom et al. (1986) and Eislรถffel & Mundt (1997), the latter discovering another feature, HH 43X, north-west of HH 43. Infrared studies of H<sub>2</sub> line emission from HH 43 have been reported in several papers (Schwartz et al. 1987, 1988; Zinnecker et al. 1989; Gredel 1994; Schwartz & Greene 1999). An optically invisible infrared source (HH 43-IRS1, IRAS 05357$-$0710) north-west of HH 43 was considered as the source of the HH 43/38 system (Cohen & Schwartz 1983, 1987; Reipurth & Aspin 1997). Cohen et al. (1984, 1985) presented far-infrared (47-130 $`\mu `$m) measurements of this source, showing it to be a cool low-luminosity object. At near-infrared wavelengths, the source was resolved into a double star with some associated nebulosity (Gredel 1994; Moneti & Reipurth 1995). Casali (1995) measured the K-band polarisation of the source as 3.78 % at a position angle of 46 degrees.
Searching for a molecular outflow associated with the HH objects, Edwards & Snell (1983, 1984) found no evidence for well-ordered supersonic molecular motion there, and while Morgan & Bally (1991) observed some evidence for high-velocity material at the position of IRAS 05357$-$0710, they later regarded this not to be a molecular outflow in a follow-up paper (Morgan et al. 1991).
HH 64 was found by Reipurth & Graham (1988). Cohen (1990) found a very red IRAS source (IRAS 05355$-$0709C) close to this HH-object, which they assumed to be its driving source. The region was searched for CS (1โ0) emission by Tatematsu et al. (1993) and Morata et al. (1997). Tatematsu et al. found two cores which appear to be at the positions of the IRAS sources: their core #98 corresponds to IRAS 05357$-$0710, and core #97 to IRAS 05355$-$0709C. Anglada et al. (1989) observed an ammonia condensation at the position of the latter core and suggested that the source responsible for the excitation of HH 64 as well as HH 43/38 might be embedded in this condensation. Below we show that this hypothesis was correct.
### 3.2 New observational results
Fig. 1 shows the region around HH 43/38 and to their north-west as we have imaged it. A long chain of H<sub>2</sub> features is clearly identified, delineating a well collimated jet. HH 38 (SMZ 7-14), HH 43 (SMZ 7-10), and HH 43X (SMZ 7-9) are seen to be part of the south-eastern lobe of this jet. A gap is evident between features SMZ 7-7 and SMZ 7-6. The latter features, along with HH 43X and HH 64, are seen to form two pairs lying roughly symmetrically around the middle of this gap. No further H<sub>2</sub> features (besides possibly some very faint traces of emission) are found to the north-west of HH 64 out to a distance of $``$8 arcmin from the middle of the gap, although it should be noted that the outermost part of the image is from data taken under rather bad weather conditions, leading to a nominally poorer detection limit; nevertheless, features with a H<sub>2</sub> brightness comparable to HH 43 can safely be excluded. We do not detect any continuum emission in the gap between SMZ 7-6 and 7-7 to a limiting magnitude of about K=17.5 (3$`\sigma `$ for a point source).
To the north-west of HH 43X we identify a cometary H<sub>2</sub> feature (SMZ 7-8, see Fig. 2) with a bright head to the north-west and a tail to the south-east, which is clearly displaced from the axis of the large flow. The tail points back to HH 43-IRS1, which is resolved into the components previously reported by Gredel (1994) and Moneti & Reipurth (1995). The diffuse nebulosity consists of two roughly symmetric lobes separated by a dark lane, which is oriented at an angle of $``$45 degrees east of north. The fainter star is located slightly displaced from the centre of symmetry of the nebulosity ($``$0.6โ0.9 arcsec to the north-west). The dark lane is about 2.5 arcsec wide and at least 4 arcsec long, corresponding to about 1100 AU and 1800 AU respectively at a distance of 450 pc.
We examined HIRES-processed IRAS maps of the area around HH 43, as shown in Fig. 3. HH 43-IRS1 (IRAS 05357$-$0710) is detected at all four IRAS wavelengths. Cohen (1990) found an additional source (IRAS 05355$-$0709C) on the coadded IRAS images, which is not in the Point Source Catalog, but is clearly seen in the 60 $`\mu `$m and 100 $`\mu `$m HIRES maps. From the 60 $`\mu `$m map (with 10 iterations of HIRES processing) we find this source to be located at $`\alpha =5^h37^m57.^s8`$, $`\delta =7\mathrm{ยฐ}07^{}03^{\prime \prime }`$ (J2000), which is about 30 arcsec north-east of the position given by Cohen (1990). However, the positional accuracy (as defined by the deconvolution) is probably not better than a few pixels, that is about 30 arcsec . IRAS 05355$-$0709C is fainter than IRAS 05357$-$0710 at 60 $`\mu `$m, but brighter at 100 $`\mu `$m, consistent with the fluxes given for these sources by Cohen (1990) and Cohen & Schwartz (1987).
In addition, a strong compact 1.3 mm source was found at $`\alpha =5^h37^m57.^s5`$, $`\delta =7\mathrm{ยฐ}07^{}00^{\prime \prime }`$ (J2000; positional accuracy $``$3 arcsec) with a flux of about 600 mJy in an 11 arcsec beam. This source is henceforth referred to as HH 43 MMS1. A 4 arcmin$`\times `$4.6 arcmin section of the 1.3 mm map is shown in Fig. 4 as a contour plot overlaid on the corresponding 2.12 $`\mu `$m infrared image. Some 32 arcsec south and 20 arcsec west of HH 43 MMS1 there is an east-west elongated ridge of emission, which we label HH 43 MMS2. Surrounding both features there appears to be substantial extended emission out to about 30 arcsec north-east and 100 arcsec to the west and south-west of HH 43 MMS1. No significant emission was detected from HH 43-IRS1, a result which is consistent with our detection limit of $``$50 mJy and the 1.3 mm flux of 47$`\pm `$14 mJy measured by Reipurth et al. (1993). The position of HH 43 MMS1 coincides with that of IRAS 05355$-$0709C to within a few arcseconds, and thus it is very likely the same source that is seen at 60 $`\mu `$m, 100 $`\mu `$m, and 1.3 mm.
### 3.3 Discussion
The new near-infrared data clearly show that the HH-objects 38, 43, and 64 are part of a single well-collimated flow. It extends over at least 11 arcmin in projection (from HH 38 to HH 64 only) equivalent to 1.4 pc at a distance of 450 pc. This projected length is probably close to the true length of the flow, since infrared spectroscopy of HH 43 indicates that the flow lies close to the plane of the sky (e.g. Schwartz & Greene 1999).
The driving source of this flow appears to be the newly discovered 1.3 mm object HH 43 MMS1, which is associated with the cold IRAS source 05355$``$0709C, embedded in a dense NH<sub>3</sub>/CS core (Anglada et al. 1989; Morata et al. 1997; Tatematsu et al. 1993), and appears to be a very good candidate for a Class 0 protostar (see Fig. 5). It seems reasonable to assume that the flow is intrinsically fully symmetric and extends out to similar distances on both sides of its source with a total length of $``$2 pc (the distance from the source to HH 38 is about 1 pc). Indeed, some very faint traces of emission might be present in our low S/N data north-west of HH 64, but this has to be confirmed. A similar break in symmetry is also observed in HH 212, the prototypical symmetric H<sub>2</sub> jet (Zinnecker et al. 1998), where the outermost bow shock is only visible in one jet lobe.
Identifying HH 43 MMS1 as the driving source also removes the difficulties in explaining the morphology of some features (see e.g. Eislรถffel & Mundt 1997): the south-eastern part of HH 43 as well as the bow-shock like HH 43X clearly do not point back towards HH 43-IRS1, but towards HH 43 MMS1. Thus HH 43-IRS1 seems to be unrelated to HH 43 and the associated infrared jet. HH 43 MMS1 is obviously deeply embedded, since it is only seen at the longest IRAS and millimetre wavelengths and is completely invisible in our K-band images. It is likely a Class 0 type object, as is suggested by the model SED (Fig. 5), which yields a ratio of $`L_{\mathrm{bol}}/L_{\mathrm{submm}}`$ of only 11 (note that the high-frequency fit in the two component curve in Fig. 5 is only constrained by upper limits, the derived $`L_{\mathrm{bol}}/L_{\mathrm{submm}}`$ of 11 thus is in fact an upper limit). HH 43 MMS1 thus easily fullfills the criterion for a classification as a Class 0 object ($`L_{\mathrm{bol}}/L_{\mathrm{submm}}<200`$) suggested by Andrรฉ et al. (1993). Additional (sub)millimetre continuum observations are needed to further constrain the SED of HH 43 MMS1 and to estimate its evolutionary stage.
Identifying HH 43 MMS1 as the driving source also removes the difficulties in explaining the morphology of some features (see e.g. Eislรถffel & Mundt 1997): the south-eastern part of HH 43 as well as the bow-shock like HH 43X clearly do not point back towards HH 43-IRS1, but towards HH 43 MMS1. Thus HH 43-IRS1 seems to be unrelated to HH 43 and the associated infrared jet. HH 43 MMS1 is obviously deeply embedded, since it is only seen at the longest IRAS and millimetre wavelengths and is completely invisible in our K-band images. It is likely a Class 0 type object, as is suggested by the model SED (Fig. 5), which yields a ratio of $`L_{\mathrm{bol}}/L_{\mathrm{submm}}`$ of only 11 (note that the high-frequency fit in the two component curve in Fig. 5 is only constrained by upper limits, the derived $`L_{\mathrm{bol}}/L_{\mathrm{submm}}`$ of 11 thus is in fact an upper limit). HH 43 MMS1 thus easily fullfills the criterion for a classification as a Class 0 object ($`L_{\mathrm{bol}}/L_{\mathrm{submm}}<200`$) suggested by Andrรฉ et al. (1993). Additional (sub)millimetre continuum observations are needed to further constrain the SED of HH 43 MMS1 and to estimate its evolutionary stage.
The K-band morphology of HH 43-IRS1 suggests that the southern star is surrounded by a disk-like structure or a flattened envelope which is seen close to edge on, and a tenuous, more spherically symmetric envelope. The star appears within the dark lane, located close to its north-western edge. This suggests that the north-western part of the disk is slightly tilted towards the observer. The disk is seen at a position angle of $``$45 degrees, in excellent agreement with the polarization angle derived by Casali (1995), assuming that light scattered in a tenuous envelope above a disk-like dust configuration will yield a polarization angle parallel to the disk plane (e.g. Elsรคsser & Staude 1978; Fischer et al. 1994, 1996). Since the disk is seen close to edge-on, the star suffers substantial extinction, which leads to its disappearance at shorter wavelengths (see Moneti & Reipurth 1995). The star to the north, which is the brighter component of the system in the K-band, also disappears at shorter wavelengths and is thus heavily extincted. It is probably located behind the star+disk system; whether it is physically related to the system is not clear. We cannot exclude the possibility that it is a background star, but since there is a dense cloud core associated with HH 43-IRS1, it might well be another young star embedded in the same core. The dust lane separating the lobes of the reflection nebula in HH 43-IRS1 seems to be much bigger than those observed in YSOs in the Taurus star forming region (Padgett et al. 1999) and the Orion Nebula silhouette disks (McCaughrean & OโDell 1996; McCaughrean et al. 1998).
The connecting line between SMZ 7-8 to the north-west and the star in the dark lane is at 85 degrees with respect to the dark lane, consistent with the usual finding that jets are perpendicular to the disks of their driving sources. The morphology of SMZ 7-8 is suggestive of a bow shock like structure heading away from the disk+star system. We thus regard the disk+star system as the source of a second jet which includes knot SMZ 7-8 (see Fig. 2), independent of the large HH 38/43/64 flow from HH 43 MMS1. There is no conclusive evidence for a counter jet in this system, although if we reflect the north-western bow symmetrically about the disk, a putative counter-jet would lie at the northern edge of HH 43, making it hard to see against this bright feature. Furthermore, the suggested inclination of the disk also implies that the north-western part of the jet is the blueshifted lobe, whereas the counter-lobe would be redshifted, running into the cloud, and thus possibly heavily obscured.
## 4 The L1641-S and L1641-S3 flow
### 4.1 Previous studies
IRAS 05380$-$0728 is located about 10 arcmin to the south and 34 arcmin to the east of the HH 43 region and is one of the more luminous young stellar objects in the L1641 cloud ($``$250โ370 L; Reipurth & Bally 1986; Cohen 1990), probably a low- to intermediate-mass star which may have undergone an FU Orionis type outburst (Strom & Strom 1993). Fig. 6 shows a 2.12$`\mu `$m image of the area around this source and its associated reflection nebulosities Re 50 and Re 50 N (Reipurth 1985; Strom et al. 1989; Heyer et al. 1990; Casali 1991; Chen & Tokunaga 1994; Hodapp 1994; Casali 1995; Colomรฉ et al. 1996; Reipurth & Aspin 1997). Reipurth & Bally (1986) and Fukui et al. (1986) independently discovered a bipolar high velocity molecular gas flow apparently from IRAS 05380$-$0728, the so-called L1641-S outflow. Further mapping of the high velocity gas in this flow presented by Morgan et al. (1991, flow MB 40) showed that the northern redshifted lobe of the flow describes a long curve to the north and west of the IRAS source position (see contour-overlay on the infrared image in Fig. 6). HH 65 is located within this lobe (Reipurth & Graham 1988). Far-infrared and submillimetre measurements for IRAS 05380$-$0728 have been presented by Reipurth et al. (1993), Colomรฉ et al. (1996), Chini et al. (1997), Zavagno et al. (1997), and Di Francesco et al. (1998); see also Cohen (1990). Morgan et al. (1990) found two radio continuum sources in the region surrounding Re 50 N; one is close to the IRAS and 2 $`\mu `$m source position of the central source of the Re 50 N nebulosity, while the second is displaced by 50 arcsec to the west. There is no NH<sub>3</sub> core associated with this second source (Wouterloot et al. 1988, 1989).
Another IRAS source (IRAS 05375$-$0731) is found about 3 arcmin south and 7.9 arcmin west of IRAS 05380$-$0728. It is associated with the molecular outflow L1641-S3 (Fukui 1988; Fukui et al. 1989). The distribution of high velocity molecular material around IRAS 05375$-$0731 has been mapped by Wilking et al. (1990) and Morgan et al. (1991, flow MB 41, see overlay of CO contours on the infrared image in Fig. 6). While Morgan et al. (1991) found a bipolar distribution, Wilking et al. (1990) found a superposition of blue- and redshifted gas, possibly indicating a flow lying close to the plane of the sky. Near-infrared images (Chen & Tokunaga 1994; Hodapp 1994; Strom et al. 1989) revealed a small group of rather faint stars close to the IRAS source position; it is not entirely clear if any of them are the direct counterpart to the IRAS source, but the reddest one was labeled as L1641-S3 IRS by Chen & Tokunaga. Also associated with this region are a radio continuum source (Morgan et al. 1990) and a water maser (Wouterloot & Walmsley 1986), although the latter was not redetected in a later study by Felli et al. (1992). The source has been detected by Price et al. (1983) as FIRSSE 101 in the far-infrared; further measurements in the far-infrared/submillimetre range are presented by Zavagno et al. (1997). Finally, there is an NH<sub>3</sub> core associated with IRAS 05375$-$0731 (Wouterloot et al. 1988, 1989)
### 4.2 New observational results
Fig. 6 shows a 17 arcmin$`\times `$17 arcmin 2.12 $`\mu `$m narrow band image of the L1641-S/L1641-S3 area, while Fig. 7 shows the same region after continuum subtraction. The main feature is a large system of H<sub>2</sub> knots and filaments to the north of Re 50 labelled SMZ 9-4, SMZ 9-5, and SMZ 9-6. HH 65 is seen to be part of the much bigger feature SMZ 9-6, including a bright, smoothly curving filamentary structure SMZ 9-6A to the north-west of HH 65. Further east, a number of fainter, more diffuse features (together labeled SMZ 9-5) define a curving path which first heads eastwards, then turns slightly to the south, and then more to the north again. Finally it ends in a multiple filamentary structure (SMZ 9-4), with the filaments curving in a similar way to the filament SMZ 9-6A. In the south-west corner of the image we find several faint, diffuse H<sub>2</sub> structures (SMZ 9-11, SMZ 9-12, SMZ 9-13, SMZ 9-14, and SMZ 9-15). Further faint structures are also visible to the west and south of Re 50 N. Finally, a few more H<sub>2</sub> features (SMZ 9-3) are found north-east and north-west of the brightest H<sub>2</sub> filament. The near-infrared source designated as L1641-S3 IRS appears extended in our images and is located at $`\alpha =5^h39^m55.^s8`$, $`\delta =7\mathrm{ยฐ}30^{}28^{\prime \prime }`$ (J2000).
Fig. 8 shows an overlay of a contour plot of a 1.3 mm map of the L1641-S3 outflow source region on the 2.12 $`\mu `$m image. A strong pointlike 1.3 mm source is found at $`\alpha =5^h39^m55.^s9`$, $`\delta =7\mathrm{ยฐ}30^{}28^{\prime \prime }`$ (J2000), coincident with L1641-S3 IRS within the positional errors of the 1.3 mm maps ($``$ 3 arcsec). A second fainter millimetre continuum source is found 20 arcsec to the west and 55 arcsec to the north. Further north, at about $`\alpha =5^h39^m57^s`$, $`\delta =7\mathrm{ยฐ}26^{}50^{\prime \prime }`$ (J2000), we find a large patch of extended emission (size about 4 arcmin$`\times `$2 arcmin). There appears to be some low-level extended emission 4 arcmin west of L1641-S3 MMS1 as well.
### 4.3 Discussion
As in the case of the HH 43 flow system, our new sensitive wide field infrared data suggest a revised picture for the L1641-S and L1641-S3 region. The dominating outflow seems not to be driven by the luminous source associated with Re 50 N, but rather by L1641-S3 IRS (IRAS 05375$-$0731). This flow includes the CO outflows L1641-S3 and the redshifted lobe of L1641-S, and is traced by infrared H<sub>2</sub> emission over much of its length.
The H<sub>2</sub> features SMZ 9-11 to 9-15 in the blueshifted lobe of L1641-S3 (which has not been mapped in CO), suggest a wide opening angle ($``$40 degrees) of the flow on this side. The redshifted lobe is defined by the redshifted molecular gas of the L1641-S3 and L1641-S flows (MB 41 and MB 40). As is seen in Fig. 6, these CO lobes appear to trace a single outflow lobe when placed alongside each other. The flow shows very strong bending. Close to the source, it is oriented at a position angle of about 60 degrees, then turns north to a position angle of about 30 degrees, then at about the location of HH 65 and the bright H<sub>2</sub> filament SMZ 9-6A (and also the peak in the redshifted CO emission), it bends strongly to an eastward flow direction at SMZ 9-5, before finally turning back north to roughly its original position angle of 60 degrees at SMZ 9-4. This behaviour is reminiscent of the wiggling of the HH 46/47 jet (e.g. Heathcote et al. 1996), albeit on a much larger scale; for example, the bright curving H<sub>2</sub> filaments in SMZ 9-6 mimic the H$`\alpha `$ wisps found in HH 46/47 at the positions of apparent bends in the jet. Heathcote et al. suggested that the gas does not really flow along the channel defined by the Herbig-Haro emission, but ballistically along the original direction, with the wiggles reflecting changes in the direction of ejection of the gas at the source. This is probably the case here as well. The direction of the flow at its end and close to the source appear to be the same, thus it might be delineating one period, assuming that the changes are periodic, e.g. caused by a companion star. Assuming a typical tangential flow velocity of about 200 km/s, the distance from the source to the outermost part of the flow (1.8 pc) translates to a period of about 9000 yr. If we assume that the wiggle of the jet is simply caused by periodic changes of the source position in a binary orbit (see e.g. Fendt & Zinnecker 1998), this period would imply a separation of the binary components of order 1000 AU or a few arcsec at the distance to Orion. However, the apparent widening of both outflow lobes with distance from the source and the large amplitude of the wiggles cannot be explained by simply shifting around the outflow source in any reasonable binary orbit, and rather suggest a precessing outflow than just a moving source. A precessing jet would be caused by a precessing disk, which would be expected in a binary system with non-coplanar disk and orbital planes. Following the arguments given by Terquem et al. (1999), a precession period of 9000 years implies a binary separation (very roughly) on the order 10 to 100 AU, corresponding to angular separations of the order 0.1 arcsec in Orion. It should be possible to resolve separations of a few arcsec or a few tenths of an arcsec with existing or future instruments in the mid-infrared and millimetre wavelength ranges, allowing a test of the above assumptions.
Alternatively, we may simply see a poorly collimated outflow, possibly also responsible for exciting the faint features to the west of Re 50 N and SMZ 9-3, with the brighter H<sub>2</sub> features only imitating a bending flow, or illuminating particular parts of a large outflow cavity. In this respect it is also interesting to note that the luminosity of the driving source L1641-S3 MMS1 is rather high ($`70L\mathrm{}`$, compared to $`5L\mathrm{}`$ in HH 43 MMS1), possibly suggesting that L1641-S3 MMS1 may be an intermediate-mass young stellar object. The L1641-S3 outflow may thus give further support to the suggestion that flows from intermediate- and high-mass protostars might be systematically less well collimated than those from their low-mass counterparts, as is seen e.g. in DR 21 (Davis & Smith 1996), G192.16-3.82 (Shepherd et al. 1998), the molecular outflow G75 C in the ON2 core (Shepherd et al. 1997), and HH 288 (McCaughrean & Dent, in prep.). Note however that Davis et al. (1998) find that flows from high mass protostars may nevertheless be driven by collimated jets like their low mass counterparts.
The driving source of the outflow appears to be L1641-S3 IRS = L1641-S3 MMS1. The spectral energy distribution of this source is suggestive of a Class 0 type object ($`L_{\mathrm{bol}}/L_{\mathrm{submm}}80`$โ90, see Fig. 9), although its detection at the shorter IRAS wavelengths and its association with a small near-infrared nebulosity suggests that it may be approaching the Class I stage.
## 5 The L1641-N outflow
In Paper I, we found evidence for a large scale H<sub>2</sub> jet from the embedded L1641-N cluster, with the more prominent features (SMZ 23) extending to the south of the cluster. Meanwhile, a chain of optically visible Herbig-Haro objects (HH 306-HH 310) has been found to the north of the cluster by Reipurth et al. (1998; see also Mader et al. 1999), which is probably the counterpart to the infrared H<sub>2</sub> flow. However, the northern Herbig-Haro flow terminates at a distance of 6.3 pc north of the cluster, whereas the southern infrared flow as delineated mainly by SMZ 23 was seen to extend only $``$1.8 pc south of the cluster.
In Fig. 10 we present a 2.12 $`\mu `$m image of the region south of the L1641-N cluster with H<sub>2</sub> features marked. We find a number of large H<sub>2</sub> features to the north-west, west, and south-west of V 380 Ori (SMZ 6-2, SMZ 6-4, and SMZ 6-16) which appear to be the continuation of the southern H<sub>2</sub> flow from L1641-N, beyond SMZ 23. Thus we suggest that the newly found features form the counterpart to the large northern Herbig-Haro flow. The flow heads due south at its origin in L 1641-N and then smoothly bends to the east down to SMZ 6-4. SMZ 6-16 is then located due south of SMZ 6-4, which makes its association with the flow somewhat uncertain; however, it might indicate a bending of the flow back to the original north-south direction. SMZ 6-16 has a bright rim at its southern edge and some more diffuse emission to the north, similar to SMZ 26B (HH 87/88), which is the southern terminating working surface in the HH 34 giant flow. Thus, SMZ 6-16 might indicate a large bow shock structure in a flow running from north to south. Including the newly discovered features, the southern infrared lobe of the L1641-N flow is now seen to extend over almost 30 arcmin ($``$4 pc projected length).
The region around V 380 Ori has been searched for high-velocity molecular gas by Edwards & Snell (1984) and Levreault (1988). Among other features, they found a large lobe of redshifted gas extending from about the region around V 380 Ori to the south (see also Morgan & Bally 1990; Morgan et al. 1991; flows MB 20/MB 21). No blueshifted counterflow was found. V 380 Ori was suggested as the driving source for this north-south oriented monopolar flow. In contrast, Corcoran & Ray (1995) suggested an east-west oriented outflow from V 380 Ori. The newly discovered H<sub>2</sub> features (including SMZ 6-16, somewhat off the track of the flow as suggested by the other H<sub>2</sub> features) appear superimposed on the north-south oriented redshifted CO outflow lobe found by Edwards & Snell (1984) and Levreault (1988). We suggest that this CO lobe is not related to V 380 Ori, but indeed another piece of the southern, redshifted part of the L1641-N outflow.
## 6 Summary and conclusions
Using data from our wide-field infrared imaging search for H<sub>2</sub> jets in the Orion A molecular cloud, we have demonstrated that parsec-scale jets can be found at infrared wavelengths. In particular, the previously known features HH 43, HH 38, and HH 64 are found to be connected by several newly discovered H<sub>2</sub> features into one large flow. A second large flow is found to consist of the previously known L1641-S3 CO outflow and the redshifted lobe of the L1641-S CO outflow containing HH 65. Again, the full extent of this flow is revealed by infrared imaging. In both cases, we also found deeply embedded millimetre continuum sources which we suggest to be Class 0 protostars driving the flows. Finally, we have shown that the southern lobe of the L1641-N flow, which is traced by H<sub>2</sub> emission, may extend over at least 4 pc, i.e. over a similar length as the northern lobe, which is traced by optically visible HH-objects. We suggest that the known molecular outflow lobe MB 20/21 (originally thought to be driven by V 380 Ori) is connected to this giant flow.
Besides the three flows discussed in this paper, other giant flows are known in Orion A. In addition to the prototype HH 34 system (Bally & Devine 1994; Eislรถffel & Mundt 1997; Devine et al. 1997), the HH 1/2 system probably belongs to this class of objects (Ogura 1995). Some flows in the OMC 2/3 region also seem to extend over large distances, including the HH41/295 flow (Reipurth et al. 1997) and the H<sub>2</sub> flow J of Yu et al. (1997). Evidence for further large-scale flows from the L1641-N cluster is reported by Reipurth et al. (1998) and Mader et al. (1999). The discovery of so many giant flows in a single cloud appears to indicate that giant flows are not exceptionally rare systems, but may indeed be more the rule than previously thought.
The redshifted lobes of the L1641-S3 and the L1641-N flows are both seen to be associated with substantial amounts of relatively slow moving molecular gas as traced by the CO emission over most of their length. The molecular gas has probably been entrained more or less in situ, not at the base of the flows. This implies that both flows plough through the dense gas of the cloud itself, not through the rather thin medium around the cloud like most of the optically-visible giant HH flows. Thus the flows from these two protostars alone are seen to affect a significant volume of the cloud. This gives additional support to the idea that flows from young low mass stars may play an important role in the dynamical evolution of a cloud and for the regulation of star formation, not only on small scales (disruption of the cloud cores associated with the protostar driving the flow), but also on large scales (injection of momentum, excitation of turbulence, stabilisation of the cloud against collapse). However, a more complete discussion of that point is deferred to a future paper based on the complete H<sub>2</sub> survey for protostellar jets in Orion.
The sources of the two newly recognized giant flows, HH 43 MMS1 and L1641-S3 MMS1, are both deeply embedded, bright at 1.3 mm, very red at IRAS wavelengths, and qualify as Class 0 protostars; only one of them (L1641-S3 MMS1) appears to be associated with a small, faint near-infrared nebulosity. Similar behaviour is found for most of the Orion parsec-scale flows, implying that the most pronounced large-scale Herbig-Haro and H<sub>2</sub> flows are driven by very young objects, typically infrared Class I and even Class 0 type objects. Thus the largest and presumably most energetic outflows are associated with the youngest objects, not the more evolved objects like optically-visible T Tauri stars.
###### Acknowledgements.
Many thanks are due to Frank Bertoldi, Ernst Kreysa, Karl Menten, Frรฉdรฉrique Motte, Bernd Weferling, and Robert Zylka for their help during the Pico Veleta observing run and first aid in reducing the 1.3 mm data. Thanks are also due to the referee, Chris Davis, for his comments which helped to strengthen some points. This work was supported by the *Deutsche Forschungsgemeinschaft, DFG* project number Zi 242/9-1. |
warning/0001/astro-ph0001526.html | ar5iv | text | # On the Existence of Differences in Luminosity between Horizontal Branch Stars in Globular Clusters and in the Field
## 1 Introduction
The well-known dichotomy existing between short and long distance scale as derived from old, Population II stars is still an unsolved problem, not yet settled even after the improvement, in distance determinations, due to the release of the Hipparcos Parallax Catalogue. In fact, while distances to globular clusters by main sequence fitting to local subdwarfs with parallaxes measured by Hipparcos, favour the long distance scale (see Gratton et al. 1997, Paper I, and Carretta et al. 2000, Paper II, for extensive updates and discussions on this topic), the statistical parallaxes of field RR Lyraes (one of the most commonly used galactic standard candle), based on Hipparcos proper motions (Fernley et al. 1998; Popowski and Gould, 1998, Tsujimoto et al., 1997) still lead to the short distance scale.
Following an alternative approach, Gratton (1998) used Hipparcos parallaxes for a sample of field metal-poor HB stars in order to directly calibrate these standard candles. Given the paucity of RR Lyrae variables within reasonable distances from the Sun, Hipparcos was able to measure useful parallaxes for only 3 variables, RR Lyrae itself and two additional stars. Uncertainties in the parallax determinations of the latter are however rather large. In order to increase the statistical significance of the sample, Gratton (1998) selected also red and blue HB field stars from various sources. His final sample consists of 20 stars with $`V<9`$ and 2 stars slightly fainter than this limit. For a consistent comparison of the results, metal abundances for the stars in the sample were put on the same metallicity scale used in Paper I and II. The mean weighted absolute magnitude found by Gratton (1998) with this procedure is $`M_V=+0.69\pm 0.10`$ (at average metallicity \[Fe/H\]=$`1.41`$), and brightens to $`M_V=+0.60\pm 0.12`$ (at average \[Fe/H\]=$`1.51`$) when HD17072, a suspected first ascent giant branch star (see however Carney, Lee & Habgood, 1998) is discarded from the sample. This latter value has been recently revised to $`M_V=+0.62\pm 0.11`$ (see Koen & Laney 1998) in order to account for the intrinsic scatter in the HB magnitudes when correcting for the Lutz-Kelker effect. The error bar given by this analysis is still large enough that a final choice between the short and long distance scale can not be made.
Gratton (1998), however pointed out how different distance determination methods seem to give consistent results as far as only $`field`$ HB stars or only globular cluster HB stars are considered. This argument led him to suggest that a real difference in luminosity of $`0.1รท0.2`$ mag, might actually exist between HB stars in globular cluster and in the general field, the cluster stars being brighter.
The hypothesis of an intrinsic difference between field and cluster RR Lyraes was immediately challenged by Catelan (1998; C98) who used the period-temperature distribution for both field and cluster variables at about fixed metallicity (in a metal-poor and a metal-rich regime) to show that GC and field RR Lyrae are essentially indistinguishable in the $`PT_{eq}`$ plane, thus concluding that there is no significant difference in luminosity between them.
In the present paper we try to have a deeper insight into this problem. Both the original targets in C98 analysis as well as a new, more homogeneous sample of cluster and field variables are used in order to achieve a quantitative estimate of the possible differences in luminosity between HB field and cluster stars, and to complement and refine the fully qualitative approach used by C98.
## 2 A reanalysis of C98 data
As a starting point, we have re-analyzed the original set of data considered by C98, kindly provided by the author. C98 dataset consists of 35 field RR Lyraes and 49 variables in 5 globular clusters (namely NGC 362, M5, M68, M15 and M92). Variables were selected by C98 in order to be of ab-type, and with light curves not affected by Blazhko effect. Metallicities were from Layden et al. (1996), and are therefore on Zinn & West (1984) metallicity scale, which is also adopted by C98 for the globular cluster variables.
According to C98 we used the empirical relation of Carney, Storm & Jones (1992a; CSJ) which gives the so-called โequilibrium temperatureโ T<sub>eq</sub> of a RR Lyrae variable as a function of the blue amplitude $`A_B`$, metal abundance \[Fe/H\] and pulsational period P. This relation provides a very tight logPโlogT<sub>eq</sub> between cluster and field variables (see figure 2 in C98), and hence seems to be the best confirmation of a similarity in their luminosities.
Following an approach similar to that of C98, we then derived T<sub>eq</sub> values for all stars in his sample. However, in order to quantify the supposed identity between field and cluster variables at fixed metal abundance we used the M15 variables as a reference. A quadratic relationship was fit to the M15 RR Lyraes in the logPโlogT<sub>eq</sub> plane (see Figure 1), and we derived for each variable the expected period if the star was to follow the M15 logPโlogT<sub>eq</sub> relation.
Differences between the actually observed (as quoted by C98) and the expected period were then computed for each star : $`\mathrm{\Delta }`$Ps $`=`$ dlogP<sub>oss-exp</sub>.
Figure 2 shows the run of $`\mathrm{\Delta }`$Ps $`vs`$ \[Fe/H\] for all stars in the original C98 sample. For the cluster variables we show the average value at the cluster metallicity.
A linear best-fit was drawn through the cluster data in order to evaluate how much field stars differ from the cluster variables. Admittedly, there is some hint that fitting by a second order polynomial could be a better representation of the data, however the lack of clusters around \[Fe/H\]$`=1.5รท1.8`$ in C98 original sample prevents further assessment of this issue. We then computed the distance of each field star from the fit representing the period-shift $`vs`$ \[Fe/H\] relation defined by the cluster variables.
The unweighted average derived from all 35 (both metal-poor and metal-rich) field stars is :
$`<\delta (\mathrm{\Delta }Ps)>`$ $`=`$ $`<\delta (dlogP_{star}dlogP_{fit,GC})>`$ (1)
$`=`$ $`+0.0108\pm 0.0042`$
with $`\sigma =0.0246`$. This relation is a more quantitative measure of the results achieved by C98, and simply tells us that the field variables have at least the same, or possibly even longer period shifts (within rather large internal errors) with respect to cluster variables of comparable metal abundance. This could be interpreted as a (weak) evidence for the field RR Lyraes to be $`brighter`$ than their cluster counterparts. A result a little unpalatable, since the discrepancy between distance scales could be accounted for only whether field RR Lyraes were fainter than their cluster counterparts. We should recall, however, that what the period shift between the two distributions (at fixed temperature and metallicity) actually measures is the the convolution of stellar mass $`and`$ luminosity effects, as in the classical equation of pulsation by van Albada & Baker (1971) (e.g. eq. 10 in CSJ or the more recent one by Bono et al. 1997). This point will be further discussed below.
### 2.1 The effect of changing dataset
Having settled the size of the effect we can expect, and of the related errors, we then explored how the observational parameters involved in the analysis could affect the derived result.
We thus repeated the analysis adopting different sets of parameters, metallicity scale (from Clementini et al. 1995, C95 or Blanco 1992 for field variables; and from Carretta & Gratton 1997, CG97, or Zinn & West 1984 for clusters) and light curve parameters (as the updated blue amplitudes for field stars by Nikolov et al. 1984).
These tests allowed us to ascertain that:
1. The effect that we want to highlight is (or could be) very subtle: in the best case we want to detect a difference of about 0.1-0.15 mag in the HB luminosities of cluster and field stars.
2. The internal errors alone are very likely about the same size of the effect we are looking for. Even parameters usually thought to be very reliable and simple to measure, as amplitudes and periods, should be carefully checked.
3. There is a strong suggestion that homogeneous data sets could greatly help to reduce systematic errors which may smear out real differences.
### 2.2 Is the pulsational approach the proper way to detect possible luminosity differences between field and cluster stars ?
There may be some procedural concerns about the use of the pulsational approach to detect a systematic difference in luminosity between field and cluster HB stars.
A first concern was pointed out by Catelan (1998), who suggested that an empirical calibration of T<sub>eq</sub> omitting the period term would be a better approach, with respect to the original CSJ calibration. In fact, there is a risk of being catched in a circular line of thought, entering with pulsational periods in order to derive temperatures and then using distributions of field and cluster stars in the T$`{}_{\mathrm{eq}}{}^{}`$logP plane to highlight a difference (shift) in the periods, to be interpreted as a luminosity difference. This can be seen also by a simple experiment suggested by the referee. If we decrease by 0.2 mag the luminosities of variables in Figure 1 in order to simulate a fainter sample, then the combination of the van Albada & Baker pulsation equation and of eq. (16) in CSJ acts to shift these fainter variables toward shorter periods and higher temperatures. The net effect is to transport the overall distribution in the T<sub>eq</sub> -logP plane along the relationship (and its extension to higher temperatures), so that no significant period shift can be detected between the original, observed sample of M 15 variables and the simulated faint one.
Unfortunately, the relationship for T<sub>eq</sub> derived by C98 with no period term is so scattered (see Figure 3 in C98), that no useful information can be derived, apart from a generic similarity of the field and cluster star distributions in the T$`{}_{\mathrm{eq}}{}^{}`$logP plane. Moreover, a reference relationship logP $`vs`$ log T<sub>eq</sub> seems not very easy to establish for any of the clusters in C98 list, judging from his Figure 3.
A point of further concern is that both in the original CSJ and in C98 analysis, the empirical calibrations of T<sub>eq</sub> are derived using only the field stars and then applied also to the cluster variables. The underlying assumption is that RR Lyrae stars share the same physical parameters, with no dependence on the environment in which they were born : the dense clusters or the much looser field. We believe that this is a dangerous procedure which may mimic spurious and/or mask real differences in the absolute magnitude of field and cluster variables.
## 3 An independent evaluation of C98 results
In order to quantitatively assess the analysis by C98 we need a more homogeneous sample of stars and a temperature calibration which can be applied independently to cluster and field RR Lyraes. We have thus selected a sample of field stars smaller in numerical size with respect to C98, but which can provide a tighter distribution, thanks to the higher degree of accuracy of their photometric data. Using new model atmospheres by Kurucz (1993) and semi-empirical colour-temperature calibrations, new temperatures were derived for all field and cluster variables, irrespective of their belonging to one or the other enviroment. Selected samples and their analysis are discussed in the following section.
### 3.1 Definition of a new sample, observables and derived temperatures
Our new sample of field stars consists of 16 ab-type RR Lyraes used by CSJ in Baade-Wesselink analyses (we disregarded RS Boo since it is presently known to be affected by Blazhko). The new targets satisfy all requisites listed in section 3.1 of C98 : being the variables usually used in the Baade-Wesselink approach to distances, they all have a very good coverage at each phase of light curve, periods and amplitudes are measured with much greater accuracy than for variables in generic surveys as those of Layden et al. (1996) and all have new homogeneous metallicity estimates from the new metallicity scale for the RR Lyraes variables of C95<sup>1</sup><sup>1</sup>1The metal abundance scale in C95 is fully consistent with the metallicity scale derived by CG97 for globular cluster giants (see discussion in CG97).
As in CSJ, we have not eliminated the supposedly evolved stars DX Del and SS Leo, since we are comparing sets of stars (field and clustersโ) which both are likely to include some evolved variables.
As for the cluster variables, we considered 19 RRab in M15 (the same used by C98), 41 RRab variables in M 3 with new CCD photometry from Carretta et al. (1998), 22 RRab in M 5 (Ripepi et al. 1998, private communication), 8 type-ab RR Lyrae with clean light curves in M 68 (Walker 1994), 6 RRab in M 92 (Carney et al. 1992b) and 18 RRab in NGC 6362 (Walker 1999, private communication). All observational datasets but those for M15 are from recent CCD $`B,V`$ photometry. However, we are confident that even if based on photographic observations, derived quantities for M15 variables well compare with the CCD data for the other clusters. In fact, none of our conclusions would be altered whether any other of the clusters in our sample was taken as a reference, even if we would have used NGC 6362, whose stars span all the relevant range in temperature. In particular, the period shift relations (see next section) are not sensitive to the choice of the cluster used to define the fiducial locus for determining the period shifts.
We used the magnitude-average color index $`(BV)_{mag}`$ as temperature indicator. As widely discussed in CSJ, there is no particular reason to believe the B$``$V to be the best colour index to reproduce the RR Lyraeโs equilibrium temperature. However, B,V photometry is presently available for a larger number of cluster variables, and, on the other hand, since this is a $`differential`$ analysis between cluster and field stars, the crucial point is to use the same colour (no matter which), for both kind of variables, in order to achieve the maximum degree of homogeneity.
$`<B><V>`$ colours (where $`<B>`$ and $`<V>`$ are intensity-averaged magnitudes), are available for all cluster variables, as well as for the field sample adopted by CSJ (column 6 of their Table 4).
Using the latter we derived a relation to transform the intensity-averaged $`<B><V>`$ colours into magnitude-averaged $`(BV)_{mag}`$ colours for all variables in our sample.
Figure 3 displays this colour conversion derived from the field Baade-Wesselink stars of CSJ (column 6 of their Table 4). The best-fitting relation is :
$`(BV)_{mag}(<B><V>)=`$
$`0.328(\pm 0.035)(<B><V>)+0.134(\pm 0.001)`$ (2)
$`r.m.s.`$ = 0.006, correlation coefficient $`r=0.92`$, based on 17 stars.
Although the two indices are not in a one-to-one correspondence, the above transformation allows us to obtain a set of colours mutually consistent for both the field and the cluster stars in our samples.
Temperatures were obtained for all stars using this colour and the semi-empirical procedure discussed at length in Gratton, Carretta & Castelli (1996). Briefly, the latest model atmospheres by Kurucz (1993) were tied (i.e. โcorrected withโ) to an empirical colour-temperature calibration for pop. I stars based on the Infrared Flux Method (see details in Gratton, Carretta & Castelli). The corrected models then were used to read from the observed de-reddened colours the temperatures at the gravity and metal abundance appropriate for each star.
A constant value of $`\mathrm{log}g=2.75`$ was adopted for all variables. This is the value generally adopted for RR Lyraes at minimum light and presently there is some indication that a larger value by 0.10-0.15 dex might be more appropriate for RR Lyraes at minimum light. However the results of our differential analysis are not affected by the adopted value of $`\mathrm{log}g`$ provided that the same value is adopted for both field and cluster variables<sup>2</sup><sup>2</sup>2Strictly speaking, gravity should be slighthly different in RR Lyraes of different luminosities and masses. However, changing gravity by 0.1 dex (corresponding to a difference of 25% in the mass and/or 0.25 mag in magnitude) only changes T<sub>eff</sub>โs by 10 K. The effect on the period shifts is only of 0.002 dex and can be neglected when compared with uncertainties e.g. on interstellar reddening. Metal abundances for the field stars were from C95. Reddening for M15 ($`E(BV)=0.09\pm 0.01)`$, M3 ($`E(BV)=0.02\pm 0.01)`$, M5 ($`E(BV)=0.035\pm 0.005)`$, M92 ($`E(BV)=0.025\pm 0.005)`$ and M 68 ($`E(BV)=0.04\pm 0.01`$) are fully consistent with the new reddening scale of Paper I (Gratton et al. 1997) and Paper II (Carretta et al. 2000). In turn, metal abundances for the cluster variables were slightly different from the values of the original CG97 scale. Taking into account the adopted reddenings, the temperatures for the cluster RR Lyraes were thus obtained using \[Fe/H\]$`=1.30`$ for M 3, \[Fe/H\]$`=2.14`$ for M15, \[Fe/H\]$`=1.95`$ for M 68, \[Fe/H\]$`=1.10`$ for M 5 and \[Fe/H\]$`=2.15`$ for M 92. For NGC 6362 (not included in the sample of Paper I and II), we adopt \[Fe/H\]=$``$0.96 dex from Carretta & Gratton (1997) and E(B$``$V)=0.08 mag from Brocato et al. (1999).
### 3.2 Analysis
The only assumption made so far is that for each variable, either in the field or in a cluster, a temperature can be defined using the latest Kurucz model atmospheres (empirically corrected) and that the temperatures so derived represent the ones the variables would have if they were static stars. Bearing in mind that we aim at a differential comparison, we may ask how much this assumption can be trusted.
As an estimate we can compare our newly derived temperatures with the equilibrium temperatures for field stars analyzed with the Baade-Wesselink method, listed in column 9 of Table 4 in CSJ.
This comparison is shown in Figure 4. Our temperatures are on average larger than the equilibrium temperatures of CSJ, the mean difference being $`<T_{\mathrm{eff},\mathrm{thispaper}}T_{\mathrm{eq},\mathrm{CSJ}}>=71\pm 15`$ K ($`r.m.s.=61`$ K, 17 stars), with no trend with temperature (or metal-abundance).
Taken at face value, this indicates that equilibrium temperatures for pulsating pop. II stars as the RR Lyraes, computed from infrared colours (as those usually employed in Baade-Wesselink analyses) and the old Kurucz (1979) model atmospheres, not corrected to empirical data, are in good agreement with effective temperature derived from intensity-averaged $`BV`$ colours using the new Kurucz (1993) and the semi-empirical procedure defined above. However, we caution that some of the original Baade-Wesselink analyses, from which values in table 4 of CSJ were taken, adopted a semi-empirical calibration.
As a comparison, we also computed differences for the same set of stars with the mean effective temperatures derived by McNamara (1997) using optical and near infrared colours and the temperature calibrations given by the new Kurucz models. In Figure 4 it is easy to see that in this case, although the average difference is not much larger than in the previous comparison ($`<`$T$`{}_{\mathrm{eff},\mathrm{us}}{}^{}T_{\mathrm{eq},\mathrm{McNam}}>=101\pm 16`$ K (r.m.s. = 67 K, 17 stars), a clear trend with temperature arises, likely due to the absence of semi-empirical corrections to the models employed by McNamara (1997).
However, we can be confident that our temperature scale i) is fully homogeneous for both field and cluster variables and ii) likely due to various, compensating effects, the derived temperature are not much far away from the so-called equilibrium temperature of a pulsating star, as discussed in CSJ.
Note that in our analysis we avoid introducing any spurious results due to the assumption that field and cluster variables have similar histories, as implicit in the CSJ and C98 analysis.
As before we used M15 as a reference and derived a linear relationship as the best fit in the the logPโlogT<sub>eff</sub> plane (a quadratic relation is no longer justified). Entering in this relation with the derived effective temperature for each stars, again we computed the difference between the actually observed and the expected period of each star, $`\mathrm{\Delta }`$Ps $`=`$ dlogP<sub>oss-exp</sub>. These differences are displayed in Figure 5 as a function of the metal abundances, taking for each cluster the unweighted mean of all cluster variables. In order to evaluate the relevance of systematic errors in the reddening scale (which affects the derived temperatures $`via`$ $`BV`$ colours) we repeated our analysis changing by $`\pm 0.02`$ the adopted reddening values for the cluster stars. The corresponding uncertainty in the derived $`\mathrm{\Delta }`$Ps $`=`$ dlogP<sub>oss-exp</sub> would be about 0.02.
As for systematic errors in the metallicity scale, the overall uncertainty cannot be reduced below 0.1 dex, as discussed by Carretta et al. (2000). We therefore adopted this figure as the possible systematic errors due to the adoption of CG97 scale.
A linear best fit regression through the data of the 16 B-W field stars in Figure 5 gives :
$$\mathrm{\Delta }Ps=0.0625[\mathrm{Fe}/\mathrm{H}]0.1141,$$
(3)
which is the bisector of direct and inverse relations, 16 stars, correlation coefficient of $`r=0.86`$. Nothing new in this relation, which slope is very similar to others obtained using field variables (e.g. Fernley 1993: -0.073).
On average, the cluster RR Lyraes are not too much far away from the mean locus defined by field variables in Figure 5. The impression by eye is of a hint for the RR Lyraes in clusters having slightly shorter periods than field variables of similar metallicity.
We then derived the same fit also for the average values defined by the cluster variables. Taking again the bisecant of relations obtained exchanging dependent and independent variables, we derived from the 6 clusters considered here:
$$\mathrm{\Delta }Ps=0.0319[\mathrm{Fe}/\mathrm{H}]0.0712,$$
(4)
with a correlation coefficient $`r=0.80`$.
This slope is much less than the value derived by the latest study of Sandage (1993). However, since we are mainly interested to obtain a figure for the actual differences between the distribution of field and cluster variables at fixed metallicity, we have computed the offset from each field RR Lyrae with respect to the fit defined by the cluster variables.
The unweighted average derived from all the 16 field stars is now:
$`<\delta (\mathrm{\Delta }Ps)>`$ $`=`$ $`<\delta (dlogP_{star}dlogP_{fit,GC})>`$ (5)
$`=`$ $`0.0067\pm 0.0069.`$
On the other side, when considering only field stars with \[Fe/H\] values in the metallicity range spanned by globular clusters (i.e. excluding from the average those more metal-rich than \[Fe/H\]$`>0.96`$) the above figure become $`<\delta (\mathrm{\Delta }Ps)>=0.0066\pm 0.0075`$, based on 10 stars.
Both these results and the attached error bars simply tell us that any difference between field and cluster variables has to be considered at best not very significant. In the following, we will use the average from Eq. (5), but the discussion would not change even using the other value.
## 4 Discussion
If we now combine the result obtained in the previous section with the pulsational equation of (e.g.) Van Albada & Baker (1971) we can simply write:
$$\mathrm{\Delta }logP=(0.0067\pm 0.0069)=0.84\mathrm{\Delta }logL0.68\mathrm{\Delta }logM,$$
(6)
where all differential quantities (periods, luminosities and masses) correspond to mean field $``$ cluster and are read at fixed temperature (and metallicity). Since, according to the pulsation equation the period depends on both luminosity and mass to understand the physical meaning of our results we can make two different assumptions concerning the mass of pulsating variables on the horizontal branch.
First we suppose that cluster and field variables were formed with identical masses and share similar histories and properties independently of the enviroment, and that the present mass is, for example, about 0.6 M, as the classical value for globular cluster HB stars.
Then from equation (6) one derives that $`\mathrm{\Delta }logL_{ficl}=0.008\pm 0.008`$, and, neglecting at first order the bolometric corrections, $`\mathrm{\Delta }`$ log$`M_V`$(fi-cl) $`=0.02\pm 0.02`$. This means that field variables are approximately as luminous as cluster variables of same metallicity. This value is only a twenty percent in magnitude of the effect ($`0.1`$ mag) required to explain the discrepancy between different distance indicators.
Given the above result, if we want that field variables are about 0.1 mag $`fainter`$ than cluster RR Lyraes, we must release the assumption of equal masses. If so, simple computations allow to obtain the values listed in Table 1.
It is then clear that if the discrepancy in distances calibrated using cluster and field variables is due to an intrinsic difference in luminosity of $`0.1`$ mag, between field and cluster HB stars, we should also postulate that HB stars in the more sparse field are about 0.05 M less massive than their cluster counterparts of similar temperature.
### 4.1 Comparison with Zero Age Horizontal Branch models
How this prediction derived from purely pulsational properties compares with evolutionary models of Zero Age Horizontal Branch (ZAHB) stars ? Is there a basic parameter (e.g. core mass or abundance of helium Y) which variation within plausible ranges could result into a difference of about 0.05 M ?
The theory of stellar evolution has by long time secured the notion that the enhancement of Y in a ZAHB model results in a star populating the HB at bluer colours (i.e. warmer temperatures) and at brighter luminosities than a model with similar structural parameters but lower Y abundance.
Actually, we are not interested in $`how`$ a HB star has gained a larger Y abundance, but mainly in the consequences that such enhancement could have. However, stars obviously arrive on the HB following a well defined evolutionary path, and it is known that a good candidate to give larger Y abundance is for instance the presence of deep extra-mixing whose onset is likely related to some non-standard mechanism (see e.g. Cavallo et al. 1998 and quoted references). In this scenario, both bluer colours and brighter luminosities are due to the higher level attained by the stars to the red giant tip, with consequent enhanced mass loss and helium core mass. The net result would thus be a star which locates on the blue part of the horizontal branch. In order to find such high-Y โclusterโ star at the same colour of a field RR Lyrae, i.e. inside the instability strip, the star should also have been born with a higher mass, so to spend a part of his helium-core burning as a pulsating variable.
To have a more quantitative insight into this scenario, we need a set of ZAHB models which explore systematically the variations of each structural parameter (mass, core mass, Y abundance) while keeping all the others fixed. Unfortunately, most of published studies assume a typical set of parameters and then follow the global $`evolution`$ of the stars, making it difficult, or almost impossible, to discriminate among different involved parameters, and the new models computed by Sweigart (1997) where enhancement of Y is explicitly treated are unfortunately still unpublished. The most complete set of classical ZAHB models available so far are those Sweigart & Gross (1976: SG76). In this respect, SG76 models are the tool we need; all subsequent improvements in the input physics are of minor relevance, in the present context, since we are mainly interested to study differential effects.
Results of interpolations in the SG76 models are listed in Table 2, where we report the variations obtained in the total ZAHB mass varying the core mass and the Y abundance, at fixed metallicity and temperature. Figures are derived holding also a constant difference of $`\mathrm{\Delta }logL=0.04`$, (i.e. $`0.1mag)`$.
Values in Table 2 read as the changes in M<sub>c</sub> and Y of a theoretical ZAHB cluster star in order to be i) brighter (by $`0.1`$ mag) and, at same time, ii) more massive than a theoretical field star.
It is easy to see that
1. if Y is enhanced at fixed core mass, the resulting enhancement in mass is negligible, if compared to the required 0.06 M. Moreover, the solution would be a little unpalatable, since it is well known that a higher Y implies higher luminosities of the star at the giant tip, but also an increase of the luminosities of the RR Lyraes, and hence in their periods (Sweigart 1997), at odd with present results. On the other side, a higher Y, if primordial, would result in lower luminosities at the red giant tip and in a smaller core mass (Sweigart, Greggio & Renzini 1990);
2. if both Y and M<sub>c</sub> are increased, a larger increase in the mass appears
3. finally, it seems that better results are obtained simply with an increased core mass. In fact, if we assume that cluster variables start their HB evolution with core masses 0.014 M larger than field stars (e.g. due to enhanced internal rotation, maybe driven by the denser environment of their formation and evolution) then it is possible for such stars to reach a difference of about 0.02 M. However, even in this case the mass enhancement is a factor 3 smaller than derived from the pulsational analysis.
### 4.2 Summary and conclusions
In this paper we explore the suggestion by Gratton (1998) that the existence of an intrinsic difference of about 0.1 mag between the luminosity of field and cluster HB stars could be responsible of the disagreement found between distance calibrations ultimately based on Hipparcos parallaxes.
Following an approach very similar in principle to that presented by Catelan (1998), we used photometric data of RR Lyrae variables in the field and in globular clusters to study their pulsational properties. Quantitative estimates of the amount of possible differences between the two kind of variables are given.
The differential comparison of the resultant period distributions leads to the following conclusions:
1. repeated trials using the original samples of variables adopted by C98 but different metallicity scales, or sets of light curve parameters, highlight that inhomogeneities in the data sets or intrinsic (even small) internal errors could blur a luminosity difference between field and cluster stars as derived from differences in their period distributions at fixed metallicity;
2. when a homogeneous metallicity scale for both field and cluster RR Lyraes is used, and consistent temperatures from optical colours and the new Kuruczโs model atmospheres (tied to empirical calibrations) are derived, more quantitative and stringent estimates are possible;
3. we confirm on the whole C98 findings: at fixed temperature and metal abundance the run of periods with \[Fe/H\] is essentially the same for both field and cluster RR Lyrae stars, even if there could be a small hint for cluster RR Lyrae having slightly shorter periods than field variables of the same metallicity;
4. our best estimate from 16 field RR Lyraes (with accurate parameters from Baade-Wesselink analyses) and 114 cluster RR Lyraes with recent and accurate photometry is that on average field stars have a difference of $`0.0067\pm 0.0069`$ in $`\mathrm{\Delta }`$ logP with respect to cluster variables. The statistical significance of this effect is then very scarce;
5. when combined with the classical pulsational equation by (e.g.) van Albada & Baker (1971), this difference formally implies that at fixed temperature and metallicity either the mass or the luminosity (or both) of field and cluster variables must be slightly different, however assuming that field and cluster RR Lyraes were born with same mass would result in a difference of only about two hundreth of a magnitude. On the other hand, in order to achieve a 0.1 mag difference in HB luminosity as suggested by Gratton (1998), the field variables should be about $`0.05M_{}`$ less massive than their cluster counterparts at same temperature and metal abundance.
Unfortunately, the determination of the mass of a star is still one of the most difficult problems in the observational astrophysics. In the case of pop. II pulsating stars, one could exploit the pulsation theory in order to derive mass estimates from periods; however, the exact value of the masses (and their run as a function of \[Fe/H\]) is still an unsettled and controversial issue.
Results from a simple explorations of SG76 ZAHB grids, discussed in Section 4.1, show that a larger core mass in the HB phase could give the larger masses required to explain, at least in part, the discrepancy in HB luminosity tentatively suggested to exist between field and cluster stars. From a theoretical point of view, a good candidate to give larger masses for stars born and evolved in the cluster dense environment is an enhanced Y abundance, possibly due to extra-mixing phenomena likely related to non-standard core rotation (see Cavallo et al. 1998 and references therein). Since evidences of deep mixing are only found in cluster giants and not in field stars (Gratton et al. 2000), the star birth-place and the density of the environment hosting its evolutionary life, in particular, could be responsible for the differences found between field and cluster objetcs.
However, according to Sweigart (1997), non-canonical helium-mixed models would result into an increase of the luminosity of the RR Lyrae variables, and hence in their pulsational periods. This is clearly at odd with C98 and our results, which find no or very little differences between the period distributions of field and cluster variables at fixed temperature.
ACKNOWLEDGEMENTS
We warmly thank Marcio Catelan for kindly providing his original data and a preprint of his paper in advance of publication. We also thank Alistair Walker for sending us mean colour and parameters for variables in NGC 6362 in machine ready form in advance of publication. It is a pleasure to thank Bernardo Salasnich and Leo Girardi for helpful discussions, as well the referee for her/his useful comments. |
warning/0001/astro-ph0001201.html | ar5iv | text | # A Long Observation of NGC 5548 by BeppoSAX: the High Energy Cut-off, Intrinsic Spectral Variability and a Truly Warm Absorber
## 1 Introduction
NGC 5548 is an extensively UV/X-ray studied, low redshift (z = 0.017), bright Seyfert 1 galaxy. Evidence for multiple spectral features in the 0.1-20 keV spectrum of NGC 5548 abound yet are still controversial: a soft excess when in a bright state (a 2-6 keV luminosity of $`L_{26}\genfrac{}{}{0pt}{}{_>}{^{}}2\times 10^{43}`$ erg s<sup>-1</sup> is the critical value in Branduardi-Raymont, 1986; Reynolds, 1997), an ionized absorber (Nandra et al., 1993, Reynolds, 1997, George et al., 1998), a Compton hump (Matsuoka et al., 1990, Piro et al., 1990, Nandra & Pounds, 1994), and a broad ($`\sigma =0.49_{0.14}^{+0.20}`$ keV) Fe K line at 6.48 keV (Fabian et al., 1994, Reynolds, 1997). This uncertainty is mainly due to the lack of simultaneous broad band observations of this variable AGN (Branduardi-Raymont, 1986, Done et al., 1995) with sufficient energy resolution and collecting area to clearly disentangle the many proposed components. BeppoSAX, with its unique array of co-pointed instruments provides a new hope of understanding this AGN. Here we present data from a long (8 day) observation with BeppoSAX that begins to fulfill these hopes.
In this paper we present the analysis of a long ($`8`$ day) BeppoSAX observation of NGC 5548. The observation was performed as a part of the BeppoSAX Core Program to study the broad-band (0.1โ200 keV) spectral variability of Seyfert galaxies.
## 2 Data reduction and analysis
NGC 5548 was observed continuously by BeppoSAX (Boella et al., 1997a) for 8 days (1997 August 14-22) with a net exposure of 314 ks. We reduced the LECS (Low Energy Concentrator Spectrometer, Parmar et al., 1997), MECS (Medium Energy Concentrator Spectrometer, Boella et al., 1997b) and PDS (Phoswich Detector System, Frontera et al., 1997) data following the standard reduction procedures. LECS, MECS and PDS data have been screened according to Fiore, Guainazzi & Grandi, 1999, to produce equalized event files. For the PDS data we used the fixed rise-time thresholds in the standard processing. Data from the 2 MECS units have been merged together to increase the signal to noise ratio.
The average 2-10 keV source intensity of $`3.5\times 10^{11}`$ erg cm<sup>-2</sup> s<sup>-1</sup> was $`50`$ % lower than that reported by Reynolds (1997) and only 6% lower than the GINGA mean (Nandra & Pounds, 1994). This corresponds to a 2-6 keV luminosity of $`2.8\times 10^{43}`$ erg s<sup>-1</sup> (using $`H_0=50`$ km s<sup>-1</sup> and $`q_0=0.1`$ throughout), just above the critical value determined by Branduardi-Raymont (1986, $`L_{26}=2\times 10^{43}`$ erg s<sup>-1</sup>). In Fig. 1 a,b,c,d we show the LECS, MECS, PDS and the total 0.1-200 keV source lightcurves. In the 0.1-3 keV energy range the source brightened during the central part of the observation by $`30\%`$, while it brightened only $`15\%`$ in the 2-10 keV band, before returning to the original intensity for the last $`100`$ ks. In contrast the high energy (13-200 keV) lightcurve was consistent with being constant over the whole observation. Thus variations of $`<15\%`$ were observed when averaged over the entire 0.1-200 keV band. In Fig. 1d (bottom panel) we show the MECS 1-3 keV to 3-10 keV hardness ratio, which confirms that the source became softer during the small flare.
The long exposure and the long timescale of the observed variability ($`100`$ ks) allowed us to perform detailed spectral and spectral variability studies. For this purpose we divided the observation into three parts, each with nearly constant intensity and with similar signal to noise (Figure 1, Table 1): two low state spectra at the same flux level from the first $`120`$ ks (L1) and the last 106 ks (L2), and a third, high state, spectrum (H) from the central 66 ks of the observation. We extracted spectra from the three instruments according to Fiore, Guainazzi & Grandi, 1999.
## 3 Model Independent Analysis
NGC 5548 experienced strong spectral variability during the BeppoSAX observation. Fig. 2 (upper panel) shows this in a modelโindependent way using ratios between the raw data for spectra H and L1. The ratio is a smoothly changing function of energy. In contrast the L1/L2 (Fig2, lower panel) ratio is consistent with no spectral variability between the first and the last part of the BeppoSAX observation. We note that there are known non-linear effects in the responses of the BeppoSAX detectors which could in principle invalidate the usefulness of the model-independent analysis based on the ratio of the raw data. However the extent of these effects is smaller than the amplitude of the features visible in Figure 2. For the two imaging detectors LECS and MECS, the non-linearities are $`\genfrac{}{}{0pt}{}{_<}{^{}}5`$% (MECS: Boella et al., 1997b; LECS: Parmar et al., 1997). For the PDS these effects are more complex because they involve Compton scattering of photons of E$`>100`$ keV in the phoswich (Frontera et al., 1997). These effects are therefore negligible below this energy for our purpose.
A result similar to that shown in the upper panel of Figure 2 was found by Done et al. (1995) in ROSAT PSPC data. They ascribed the observed variation to a changing soft component. However, the smoothness of the H/L1 ratio across three decades in X-ray energy makes that explanation unlikely. Similarly a change in the opacity or the column of an ionized absorber, which could only modify the emerging spectral shape between 0.5 and $`2`$ keV (Nicastro et al., 1999a), is ruled out.
A simple change in the emitted power-law (F(E) = A E, with A in erg s<sup>-1</sup> cm$`2`$ $`keV^1`$, at 1 keV) provides a better fit. To better address this point we show, superimposed on the data in the upper panel of Fig.2, 3 curves: (a) a dashed line corresponding to a simple 0.1-200 keV flux variation of $`+15\%`$, (b) a dotted line corresponding to a steepening (larger $`\alpha `$) of spectrum H compared to spectrum L1 by $`\mathrm{\Delta }\alpha =0.1`$ ( H/L1 = 1.36$`\times E^{0.1}`$, E in keV) with a pivot-point at $`13`$ keV and high energy cut-off fixed at 115 keV; (c) a solid curve showing a similar change in slope ($`\mathrm{\Delta }\alpha =0.19`$, pivot energy at $`6`$ keV), plus an increase of the e-folding energy of a high energy cutoff by $`\mathrm{\Delta }E_c=(E_c(L)E_c(H1))=785`$ keV ( H/L1 = 1.36$`\times E^{0.19}\left(\frac{e^{E/900}}{e^{E/115}}\right)`$, E in keV). These two models are both good representations of the whole 0.1-200 keV data.
Finally we note that there is no evidence of change at the energy of the iron line in either the H/L1 or L2/L1 ratios, suggesting that the line equivalent width and profile do not change during the whole observation.
## 4 The Integrated Low-State Spectrum
We have shown (ยง3) that NGC 5548 experienced spectral variability between L1 and H, but did not vary significantly between the first and the last part of the observation (L1 and L2). This allows us to sum spectra L1 and L2 to obtain the integrated low-state spectrum L = L1+L2 (see Tab. 1). The high quality of these data over the full 3 decade range of the BeppoSAX instruments allows us to perform a detailed spectral analysis on spectrum L, and to discuss in detail each individual component. We defer to ยง5 analysis of the spectral variability.
The integrated BeppoSAX spectrum L is shown in the upper panel of Fig.3, along with the best fitting power law model absorbed at low energy by cold gas (Model A) constrained to have a minimum hydrogen column density equal to the Galactic value: N$`{}_{H}{}^{}=1.7\times 10^{20}`$ cm<sup>-2</sup> (Stark et al., 1997). We allow the relative MECS to PDS normalization to vary by 5% around the value of 0.86, to account for the estimated systematic uncertainty (Fiore, Guainazzi & Grandi, 1999). Analogously we leave the LECS to MECS normalization free to vary over the range of acceptable values 0.7-1 (Fiore, Guainazzi & Grandi, 1999). Throughout the paper errors are quoted at a confidence level of 90% for 1 interesting parameter (i.e. $`\mathrm{\Delta }\chi ^2=2.7`$). The fit is very poor ($`\chi _r^2(dof)=3.06(112)`$). Five features are clearly visible in the ratio between the data and the best fit model (Fig. 3, lower panel): (1) a deficit of counts between $`0.6`$ and $`1.5`$ keV (Nandra et al., 1993); (2) an excess below 0.6 keV; (3) a narrow, line-like feature around 6 keV, at an energy consistent with the K$`\alpha `$ fluorescence line from almost neutral iron (Fabian et al., 1994); (4) a strong systematic excess above 10 keV (Matsuoka et al., 1990, Piro et al., 1990, Nandra & Pounds, 1994); (5) a marked deficit above 30 keV. The first four of these features have already been observed in separate narrow band X-ray spectra of NGC 5548, and interpreted in models involving the reprocessing of the nuclear radiation by both cold and โwarmโ matter close to the nuclear source: the low energy excess and deficit of counts can be explained by absorption by highly ionized gas obscuring the line of sight to the source (Nandra et al., 1993), and the narrow iron line at 6.4 keV (rest frame) and the high energy bump seen at $`1020`$ keV can be described by reflection off optically thick cold matter (Nandra & Pounds, 1994).
We therefore fitted a model (Model B, Table 2) including these extra components: a warm absorber (parameterized by the equivalent hydrogen column density N<sub>H</sub>, and the ionization parameter U, defined as the ratio of the number density of hydrogen ionizing photons to the number density of hydrogen), a Gaussian emission line (to model the FeK$`\alpha `$ emission line at 6.4 keV, rest frame) and a reflection hump (with same index as the intrinsic continuum and inclination angle $`i=30^o`$). Note that we do not include an additional soft component. Model B greatly improves the fit ($`\mathrm{\Delta }\chi ^2=189`$, with the addition of $`\mathrm{\Delta }\nu =6`$ free parameters: probability of exceeding $`F\mathrm{\Delta }\chi ^2/(\mathrm{\Delta }\nu \chi _r^2)`$ $`<<0.001`$), although it remains unacceptable ($`\chi _r^2(dof)=1.45(106)`$).
A large deviation above 10 keV is still present (Fig. 4a), along with a residual systematic structure between 0.4 and 0.7 keV. We added an exponential high energy cut-off to the power law and the reflection components, linking the two e-folding energies to the same value (Model C), producing a highly significant improvement in the fit (Fig. 4b, $`\mathrm{\Delta }\chi ^2=46.9`$, with the addition of 1 free parameter: probability of exceeding F $`<0.001`$). The fit is now acceptable ($`\chi _r^2(dof)=1.02(105)`$). However the residuals now show the presence of a narrow $`3\sigma `$ excess at $`0.450.65`$ keV (Fig. 4b). The uncertainty on the LECS calibration at 0.6 keV, is less than 5% (Orr et al., 1998), while the deviations between the LECS data and Model C at this energy are 10-20% (Fig. 4b) giving us confidence in the reality of this feature.
To model this $`0.450.65`$ keV excess, we introduced a second, narrow ($`\sigma =50`$ eV, fixed) Gaussian emission line (Model D), which improves the fit by $`\mathrm{\Delta }\chi ^2=6.2`$ (probability of exceeding F of 0.08, with the addition of two parameters). The fit is now excellent (Table 3), and the residuals are flat over the whole energy range (Fig. 4c). The best fit energy of the emission line is consistent with the energies of the OVII K$`\alpha _{1,2,3}`$ triplet transitions (E = 0.561, 0.569, 0.574 keV).
To check the consistency of our findings, we retrieved by the ASCA public archive a set of unpublished ASCA observations of NGC 5548 taken in 1996, and analyzed them to investigate the presence of a similar feature in the ASCA data. These data clearly confirm the suggestion of a $`0.6`$ keV emission feature arising from BeppoSAX data. We found that a narrow and highly significant excess of counts is left at the lower energy end of the ASCA-SIS, after fitting the 0.5-10 keV SIS+GIS data with Model B (Costantini, Nicastro & Elvis , in preparation). The energy of these feature, when modelling the data with an additional gaussian emission line, is consistent with that found with BeppoSAX.
Finally we tested a different model for the low energy spectral features seen in the BeppoSAX spectrum of NGC 5548. We explored the possibility that a high ionized reflector, rather than absorber, could account for both the excess of counts below $`0.6`$ keV, and the deficit between $`0.6`$ and $`1.5`$ keV (Ross & Fabian, 1993). We fitted the 0.1-6 keV LECS+MECS BeppoSAX data with a model consisting of an ionized reflector (Magdziarz & Zdziarski, 1995) and a gaussian at 0.54 keV. This model does not produce a good fit to the data ($`\chi _r^2(dof)=1.3(73)`$), and the residuals clearly still show a broad deficit of counts at the energies of the OVII and OVIII K-edges. We then rule out this possibility.
In Table 4 we summarize the components of models A-D, along with the statistical results.
### 4.1 Flat Intrinsic Continuum and High Energy Cut-off
The BeppoSAX โlow-stateโ data of NGC 5548 are well fitted by an intrinsic X-ray continuum consisting of a power law that requires an exponential cut-off, $`E_c=115_{27}^{+39}`$ keV. This is the second direct, well-constrained measurement of a high energy cut-off in a single AGN (after that of NGC 4151, also measured by BeppoSAX: Piro et al, 1998). In moderate band width data the power law spectral index and the e-folding energy of the high energy cut-off can be tightly correlated. BeppoSAX avoids this degeneracy, as shown in Fig. 5 (dashed lines).
The intrinsic power law spectral index (measured over the entire 0.1-200 keV band) is unusually flat ($`\mathrm{\Gamma }=1.59_{0.02}^{+0.03}`$). Previous measurements (EINSTEIN: Weaver, Arnaud & Mushotzky, 1995; GINGA: Nandra & Pounds, 1994; ASCA: Reynolds, 1997) of the intrinsic 2-20 keV continuum of NGC 5548 gave a steeper spectral index ($`\mathrm{\Gamma }1.92.0`$). This discrepancy is not due to the restricted energy range of those instruments, compared to BeppoSAX. We refitted the low state (L) BeppoSAX data (Model B), using only the ASCA SIS+GIS energy range (0.6-8 keV, Table 2, second column), and find a power law spectral index that is still flatter than that measured with ASCA ($`\mathrm{\Gamma }_{SAX:0.68keV}=1.66_{0.09}^{+0.07}`$ vs. $`\mathrm{\Gamma }_{ASCA}=1.89_{0.01}^{+0.02}`$, Reynolds, 1997), implying a real change of slope between the ASCA and BeppoSAX observations (taken 4 years apart), comparable with the change from L1 to H in the BeppoSAX data. We also note that the $`0.68`$ keV spectral index $`\mathrm{\Gamma }_{SAX:0.68keV}=1.66_{0.09}^{+0.07}`$, is fully consistent with the value measured over the entire BeppoSAX band: $`\mathrm{\Gamma }=1.59_{0.02}^{+0.03}`$. This does not allow us to rule out the presence of a curvature of the nuclear X-ray continuum even at energies lower than $`10`$ keV, and to discriminate (at those energies) between continuum models more complicated than a simple power law.
### 4.2 The Warm Absorber and the 0.54 keV Emission Feature
To investigate the low energy absorption features we fitted the data with a grid of single-zone photoionization models built with CLOUDY (V.90.04, Ferland, 1997), for 30 values of the ionization parameter U (-1.0 $`<`$ log U $`<`$ 1.9) and 30 values of the equivalent total hydrogen column density N<sub>H</sub> (21.0 $`<`$ log N<sub>H</sub> $`<`$ 22.9). We assume solar (Grevesse & Anders, 1989; Grevesse & Noels, 1993) abundances of the elements from H to Zn. These models include only the transmitted spectra. Emission from the gas is not considered here. We tested the dependence of these models on the electron density of the gas $`n_e`$, and verified that the relative abundance of the most abundant ion of the oxygen (OVII) varies by less than 3 % for changes of $`n_e`$ in the range $`10^110^{12}`$ cm<sup>-3</sup> (see also Nicastro et al., 1999a). We then assumed $`n_e=2\times 10^9`$ cm<sup>-3</sup>, the value needed if the ionized absorbing gas was in pressure equilibrium with the gas of the Broad Emission Line Regions of NGC 5548 (BELRs. See ยง6.3). We adopted as an ionizing continuum, the observed NGC 5548 Spectral Energy Distribution (SED. Mathur et al., 1995). In the 0.1-150 keV energy range we used the cut-off power law measured by BeppoSAX. The ionization parameter U is not a good absolute estimator of the ionization state of the X-ray absorbing gas. A more reliable indicator of the physical ionization state of the absorber is the relative ion abundance distribution, which is well defined by measuring the relative abundance of at least two abundant ions of the most abundant elements. We report values of U only for completeness and instead base our discussion on the relative ionic abundances in the gas (Table 3).
Model D gives a column density of the absorber of log N$`{}_{H}{}^{}=21.44\pm 0.12`$, consistent with both the ASCA estimate of log N$`{}_{H}{}^{}=21.51_{0.13}^{+0.09}`$ \[George et et al., 1998: from their best fitting Model B(i)\], and the total hydrogen column density of the UV absorber, and so consistent with the unified X-ray/UV absorber model proposed by Mathur et al. (1995, but see also the later results in Mathur et al., 1999). OVII and OVIII fractions (Table 3) are also consistent with previous X-ray (Reynolds, 1997, George et al., 1998) and UV (Mathur et al., 1995, 1999; Crenshaw & Kraemer, 1999) data.
The emission line energy ($`0.54_{0.06}^{+0.07}`$ keV) is consistent with the oscillator-strength-weighted energy of the strong OVII K$`\alpha _{1,2,3}`$, OVIII$`K\alpha _{1,2}`$ and OVII K$`\beta `$ transitions ($`<E>=0.60`$ keV, Verner et al., 1996, observed at 0.59 keV for the redshift of NGC 5548). No other obvious candidates are available (e.g. OVI at 0.08 keV, NVI at 0.43 keV, NeIX at 0.92 keV: Verner et al., 1996). OVII-OVIII are also the most abundant oxygen ions in an photoionized gas with log U=0.7, as in the warm absorber of Model D. This makes the ionized absorber gas an excellent candidate to produce the detected emission features at 0.54 keV. Emission lines from the most abundant ions will inevitably be produced in gas in photoionization equilibrium, however their intensities and equivalent widths depend strongly on the assumed geometry and gas dynamics (Netzer, 1993, 1996; Nicastro, Fiore, Matt & Elvis, 1999c, in preparation). In order to test this hypothesis quantitatively we built models for ionized absorbers which include the contributions of both gas emission and resonant absorption. The calculation of the resonant absorption features is carried out as described in Nicastro, Fiore & Matt (1999b), and we use CLOUDY (V.90.04, Ferland, 1997) to include the emission contribution (whose calculation includes resonant scattering of the emission lines). A detailed and quantitative presentation of these models is deferred to a forthcoming paper (Nicastro, Fiore, Matt & Elvis, 1999c). The equivalent width of the emission lines is much more sensitive than the ion relative abundance to the exact value of the electron density. We verified with CLOUDY that the equivalent width of the permitted OVII K$`\alpha `$ emission line varies by a factor $`2`$ for changes of $`n_e`$ in the range $`10^110^{12}`$ cm$`3`$, increasing monotonically up to densities of $`10^8`$ cm<sup>-3</sup>, and then dropping by $`50\%`$ at $`n_e=10^{12}`$ cm<sup>-3</sup>. As in our previous pure-absorption models, we assume here $`n_e=2\times 10^9`$ cm<sup>-3</sup>. We adopted a spherical geometry and the value of the covering factor $`f_c`$, as seen by the central source, which maximizes the emission from the gas. In a static configuration (no bulk motion of the gas) this value is $`f_c=0.5`$ (Netzer, 1993; Nicastro, Fiore, Matt & Elvis, 1999c), since higher values of $`f_c`$ would result in larger net reabsorption by the near side clouds. However if the gas is outflowing/inflowing the emission contribution increases monotonically as $`f_c`$ increases, and reach its maximum for $`f_c=1`$. We also accounted for โmicro-turbulenceโ ($`\sigma _v`$) and bulk motion ($`v_{bulk}`$) of the gas. The effect of micro-turbulence is to reduce the opacity at the center of the emission lines and to increase the efficiency of continuum pumping excitation of their upper levels, and so increasing the linesโ net intensities. The effect of a net bulk motion of the gas is that of shifting the absorption lines blueward or redward with respect to the corresponding emission lines (and so to increase the net equivalent width of the emission lines, if the lines are resolved). The tentative identification of the UV absorber component associated with the X-ray absorber of NGC 5548 (Mathur et al., 1999) gives $`v_{bulk}=+500`$ km s<sup>-1</sup>, $`\sigma _v=FWHM/(2\sqrt{ln2})=100`$ km s<sup>-1</sup>, which we use here. Smaller/larger values of the ratio $`(v_{bulk}/\sigma _v)`$ would result in smaller/larger net equivalent widths of the emission line, due to the increasing/decreasing negative contribution of the correspondent absorption transition. The maximum obtainable value of the net equivalent width of the emission lines, however, saturates for $`v_{bulk}/\sigma _v\genfrac{}{}{0pt}{}{_>}{^{}}2`$ (reaching 99% of its intrinsic maximum for $`v_{bulk}/\sigma _v\genfrac{}{}{0pt}{}{_>}{^{}}6/\sqrt{2}4.2`$).
The best fit photoionization equilibrium model (log U=0.70 and log N$`{}_{H}{}^{}=21.44`$: Model D) predicts three OVII K$`\alpha _{1,2,3}`$ emission lines at 0.561, 0.569, 0.574 keV, along with OVII K$`\beta `$ at 0.67 keV and OVIII K$`\alpha _{1,2}`$ at 0.65 keV (Figure 6, upper panel). These lines are the strongest emission features expected between 0.4 and 0.7 keV (Figure 6, upper panel). However the total equivalent width of all six lines (relative to the absorbed continuum) is $`\genfrac{}{}{0pt}{}{_<}{^{}}8`$ eV (at an oscillator-strength-weighted energy of $`<E>=0.60`$ keV. Figure 6, upper panel), insufficient to explain the measured strength of $`53_{37}^{+41}`$ eV. Since we maximized the emission line strength then, unless peculiar and asymmetric geometrical configurations are assumed (i.e., our line of sight to NGC 5548 has a lower gas column than other directions, and/or the symmetry is conical rather than spherical or cylindrical), we can rule out the possibility that the same gas in photoionization equilibrium is responsible for both the observed absorption and the emission features.
An alternative physical explanation is that the ionized absorbing gas has a higher temperature so that collisional ionization competes with photoionization (โhybridโ model: details are in Nicastro et al., 1999a). Because in this case the electron temperature lies nearer to the line excitation temperature, the emissivity from the single cloud of gas is enhanced, thus the emission line equivalent widths are larger, relative to pure photoionization equilibrium case (for fixed relative abundances of the oxygen ions). A model with the same $`N_H`$ as in the pure-photoionization case ( log N$`{}_{H}{}^{}=21.44`$) but the temperature of the gas raised to $`1.2\times 10^6`$ K and log U = -0.2 is shown in Figure 6 (lower panel; note the different scale in the vertical axes of the two panels). With these values the relative abundances of OVII and OVIII in the gas are $`n_{OVII}=0.60`$ and $`n_{OVIII}=0.35`$, similar to those obtained with the best fitting model D (Fig. 6, and see Tab. 3). The 0.60 keV blend of OVII K$`\alpha _{1,2,3}`$, OVIII K$`\alpha _{1,2}`$ and OVII K$`\beta `$ now has EW = 60 eV, seven times as strong as in the pure photoionization model, and fully consistent with the observed value in NGC 5548.
### 4.3 The Reflection Hump and the Iron Line
The broad band BeppoSAX data allow us to strongly constrain the relative amount of reflection, the energy, the width and the equivalent width of the iron line, and the primary continuum parameters (see ยง4) simultaneously.
The reflection component is strongly required by our broad band BeppoSAX data of NGC 5548. Eliminating this component from Model B, and refitting the data, gives $`\mathrm{\Delta }\chi ^2=23`$ (for one interesting parameter, corresponding to a probability of P$`<0.001`$).
The position of the Fe line (Tab. 3) is consistent with low-to-medium ionized iron (FeI to FeXVI), and its intrinsic width is $`<320`$ eV at a 99% confidence level (Tab. 3). We measure an equivalent width of EW(Fe) = $`127_{23}^{+30}`$ eV, consistent with the measured relative amount of reflection $`R=0.55_{0.17}^{+0.19}`$. Essentially due to the limited energy resolution of the MECS at 6 keV, we can not rule out a weak red wing to the iron line, although statistically it is not required (a fit with a disk-line model with all parameters except the index of the emissivity law and the chemical composition free to vary, gives $`\chi _r^2(dof)=1.0(100)`$, and: $`E=6.34_{0.18}^{+0.35}`$ keV, $`6<R_{in}<54`$, $`R_{out}>1800`$ โin units of gravitational radiiโ, $`i<39^o`$).
We checked the consistency of our results with those obtainable with narrower band instruments, comparing the 2-10 keV MECS data with the 2-10 keV GIS data of a 1996 ASCA observation of NGC 5548. In Figure 7 we show the 1, 2 and 3 $`\sigma `$ confidence levels for the two parameters R and $`\sigma (Fe)`$. Though both the parameters are better constrained in the MECS than in the GIS (due to the higher signal to noise of the BeppoSAX L spectrum), MECS and GIS data are fully consistent with each other.
## 5 Spectral variability: the โhigh-stateโ spectrum
Here we use the well-constrained spectral model D from ยง4 to investigate the spectral variability experienced by the source during the central part of the observation, H.
We fitted spectrum H with Model D, again leaving the relative MECS to PDS normalization free to vary within the 5% uncertanty around its estimated value of 0.86 (Fiore, Guainazzi & Grandi, 1999). We found that the amount of cold absorption was coincident with the Galactic value along the line of sight and that the warm absorber and the iron emission line parameters were consistent, within the errors, with those found fitting the spectrum L. We then fixed the neutral column of gas to the Galactic value, and the column density of the warm absorber and the energy of the iron emission line to the best fit values of log N$`{}_{H}{}^{}=21.44`$ and $`E=6.30`$ keV found fitting spectrum L, and refitted the data. The results are shown in Table 5. For comparison we report in the same table the corresponding values for the low-state spectrum (same as in table 3).
All the parameter values are consistent with those obtained for spectrum L, except for the slope of the intrinsic continuum power law, and the e-folding energy of the high-energy cut-off. The intrinsic continuum power law of spectrum H steepens by $`\mathrm{\Delta }\mathrm{\Gamma }=0.19_{0.05}^{+0.07}`$ compared to spectrum L, and the e-folding energy of the high energy cut-off is shifted to higher energy (Figure 5).
This finding supports the result of ยง3 that a $`\mathrm{\Delta }\mathrm{\Gamma }0.19`$ change of the primary power law spectral index, pivoting somewhere in the medium X-ray band, can entirely account for the observed spectral variability (Fig. 8a). As a further check we tested the hypothesis of changes in the parameter values of the other components of our model while keeping the slope of the intrinsic power law fixed to the spectrum L value of $`\mathrm{\Gamma }=1.59`$. We left all the other parameters free to vary. The $`\chi ^2`$ is now unacceptably high ($`\chi _r^2(dof)=1.45(107)`$) and the residuals clearly show the curvature of the intrinsic continuum along with many narrow deviations over the whole 0.1-200 keV band (Fig. 8b). We also tested the hypothesis of an additional soft variable component, which is not visible in spectrum L, but appears in spectrum H. We added a blackbody (Case a) or a second power law (Case b) to Model B, and refitted the 0.1-10 keV part of spectra L and H. In Case a we left all the continuum parameters free to vary for both spectra. In Case b we forced the slopes of the two power laws to have the same values in both the spectra, but left the normalizations free to vary between the two state spectra. Both these models provide a good description of the data ($`\chi _r^2(dof)=0.97(188)`$ and $`\chi _r^2(dof)=0.97(190)`$ for Cases a and b respectively) but the result is again a variation of a single component over the whole 0.1-10 keV band. In Case b the black body normalizations are pegged to zero in both spectra, and in Case a only one of power laws dominates the 0.1-10 keV band in both the L and H data, although it is the flat one in L and the steep one in H. Fixing the ratio of the two normalizations of one of the two power laws to the observed value of 1.15 (in the 2-10 keV band), gives an unacceptably high $`\chi ^2=1.59(191)`$. We conclude that the whole 0.1-200 keV power-law continuum spectral index of NGC 5548 underwent a significant variation during the 300 ks BeppoSAX observation, while the 2-10 keV luminosity varied by less than $`<15\%`$. This is the first direct evidence of a change of the slope of the power law continuum in this Seyfert galaxy.
We do not detect any change of the ionization state of the โwarm absorberโ, between the two-state spectra L and H. However, even accounting for both the changes in ionizing luminosity and in spectral shape, and in the best photoionization equilibrium case, we would expect changes in $`n_{OVII}`$ and $`n_{OVIII}`$ of only $`+10\%`$ and $`+3\%`$ respectively, well within the errors.
## 6 Discussion
### 6.1 Steepening of the Intrinsic Power Law
The 3 decade broad band of BeppoSAX has allowed us to detect a change in the primary continuum shape of NGC 5548, which is apparently correlated with slightly different intensity states. The slope of the intrinsic power law steepens by $`\mathrm{\Delta }\mathrm{\Gamma }=0.19`$ as the 0.1-3 keV source flux brightens by a factor of 1.3, during the central part of the observation. The 2-10 keV flux increases by a factor 1.11 only during the same event.
Spectral index variations are predicted by both thermal (Haardt, Maraschi & Ghisellini, 1997) and non-thermal (Svensson, 1996) models for the production of the X-ray spectrum in Seyfert 1, based on variations of the Compton optical depth in the electron and/or electron-positron region illuminated by the soft thermal photons emitted by the disk. However if these regions are pair-dominated (i.e., have large values of the compactness parameter), the optical depth variations are well correlated with conspicuous (an order of magnitude for $`\mathrm{\Delta }\mathrm{\Gamma }=0.2`$, Svensson, 1996), soft (disk photons) and hard (X-ray photons) luminosity variations (Haardt, Maraschi & Ghisellini, 1997, Svensson, 1996). These are not observed in our observation of NGC 5548.
Instead we observe $`\mathrm{\Delta }\mathrm{\Gamma }=0.19`$, associated to a small 0.1-3 keV luminosity change (30%). Fluctuations of the Compton optical depth unrelated to luminosity changes are expected in electron-dominated thermal coronae (i.e. with small values of the compactness). In fact, in this case a relationship between the luminosity and the optical depth can not be specified a priori (Haardt, Maraschi & Gisellini, 1997), and fluctuations of $`\tau `$, and so of $`\mathrm{\Gamma }`$, uncorrelated with luminosity changes are possible. Based on our data we can estimate a lower limit for the compactness in NGC 5548. The ionizing luminosity of the source in the low-state is $`L2.2\times 10^{44}`$ erg s<sup>-1</sup>, and the timescale over which the source change flux by a factor of 2 is greater than $`\mathrm{\Delta }T8`$ days. So the compactness must be $`l>(L/c\mathrm{\Delta }T)(\sigma _T/me_ec^3)0.2`$ (Done & Fabian, 1989). This is a weak limit for the estimated compactness in AGN ($`l1100`$, Svensson, 1996). Our data suggest then that a thermal, electron dominated (low compactness) corona is generating the X-ray power law in NGC 5548.
### 6.2 The High Energy Cut-Off
In the high signal to noise, low-and-flat state spectrum of NGC 5548, we clearly measure a steepening of the intrinsic spectral shape above $`50`$ keV and model the curvature of the continuum with a cut-off power law with e-folding energy of $`E_c=115_{27}^{+39}`$ keV. The evidence of a high energy cutoff is much weaker in the lower signal to noise, high-and-steep state spectrum extracted from the central part of the BeppoSAX observation. In this case we measure $`E_c>130`$ (at a confindence level of $`99\%`$), which suggests a shift of this component to higher energy as the intrinsic power law steepens.
Both thermal and non-thermal models predict a high energy break of the X-ray power law. In non-thermal models the steepening of the X-ray power law at high energy is due to energy losses by high energy photons which downscatter off relatively cold electrons. The energy of the break is then of the order of $`E_{break}m_ec^2/\tau _{pair}^25l^1`$ MeV (where $`\tau _{pair}^2`$ is the typical number of scatterings experienced by a single photon before escaping: Svensson, 1996). For $`E_{break}50170`$ keV, $`l`$ is in the range 100-30. In thermal models instead an exponential turnover is expected at an energy of $`E_c=2kT_{corona}`$, due to the cut-off of the Maxwellian distribution of the electrons in the corona (Haardt, Maraschi & Ghisellini, 1997). Our data do not allow us to distinguish between the two different expected spectral shapes.
Thermal models also predict a strict anticorrelation between the temperature of the corona (and hence $`E_c`$) and the steepness of the whole X-ray continuum. Our data would suggest the opposite trend. However, we note that the coronaโlike geometry proposed by Haardt (1993), produces an inverse Compton spectrum which is not (for E$`<<`$ kT) a single powerโlaw, but rather resembles a broken powerโlaw, with the break depending on T<sub>corona</sub> (Haardt, 1993, Petrucci et al., in preparation). A test of more accurate and selfconsistent models for the production of the X-ray continuum is deferred to a forthcoming paper (Petrucci et al., in preparation).
NGC 5548 is only the second Seyfert 1 galaxy for which a clear detection of a high energy cut-off with relatively low and well constrained e-folding energy (between 50 and 150 keV) has been reported. The other is NGC 4151 which also shows a very flat X-ray spectrum ($`\mathrm{\Gamma }=1.21.5`$, Piro et al., 1998). Larger samples of Seyfert 1s with broad-band X-ray data are needed to find clear answers to this issue.
### 6.3 A โTruly Warmโ Absorber/Emitter
The ionized absorber of NGC 5548 during the BeppoSAX observation has physical and geometrical properties in all respects similar to those derived by Mathur et al. (1999), based on their analysis of the ASCA and HST GHRS data of NGC 5548. The most abundant ions of the oxygen are He-like and H-like. We also found a marginal evidence for a sharp emission feature in the LECS spectrum, at an energy consistent with that of the oscillator-strength-weighted blend of OVII K$`\alpha _{1,2,3}`$, OVIII K$`\alpha _{1,2}`$ and OVII K$`\beta `$ emission lines. We modelled the emission feature adding a narrow gaussian to our model, and obtain an equivalent width of EW=$`53_{37}^{+41}`$. A similar feature was found in the ASCA spectrum of the warm absorber Seyfert 1 galaxy NGC 3783 (George, Turner & Netzer, 1995). In that case the column density of the ionized absorber was almost an order of magnitude higher than that of the ionized gas in NGC 5548. The authors concluded that the same photoionized material is responsible for the OVII and OVIII absorption and the emission line at $`0.6`$ keV. For NGC 5548 however, if the BeppoSAX detection of the OVII-OVIII K$`\alpha `$, K$`\beta `$ emission lines is correct (as suggested by the 1996 ASCA data of this source: Costantini, Nicastro & Elvis, in preparation), the implied 3$`\sigma `$ lower limit of their total equivalent width is EW$`>16`$ eV. Since (ยง4.2) our pure photoionization model found the maximum line strength (for each line) to be $`1.5`$ eV, the observed value is too large to be produced by the same gas which is responsible for the observed OVII and OVIII K edge absorption if photoionization equilibrium applies (and a spherical and symmetric geometry is assumed: Nicastro, Fiore, Matt & Elvis, 1999c), even when the blend of OVII K$`\alpha _{1,2,3}`$, K$`\beta `$ and OVIII K$`\alpha _{1,2}`$ is considered (EW$`{}_{TOT}{}^{}\genfrac{}{}{0pt}{}{_<}{^{}}8`$ eV).
Instead we propose that the gas is kept at a temperature higher than the equilibrium photoionization temperature (for example by mechanical heating). This strongly increases the emissivity from the gas, and would simultaneously let photoionization continue driving small and rapid changes of the ionization structure (see Nicastro et al., 1999a, for details). However the relative contibution from photoionization and collisional ionization in this gas is critical. A gas temperature of $`1.2\times 10^6`$ K is sufficient to produce OVII K$`\alpha _{1,2,3}`$ emission lines of total equivalent width of few tens of eV (with $`f_c=1`$), and so reproduce the observed quantity. At this temperature, if the ionization parameter departed markedly from log U = -0.2 (the value used earlier: see ยง4.2, โhybridโ model) then the ionization structure of the gas would no longer reproduce the observed optical depth of the OVII and OVIII K edges. To fit the observed features a value of log U lower than -0.2 must be combined with a temperature higher than $`1.2\times 10^6`$ K, and so with a higher value of the gas emission measure. However, this produces an excess of soft X-ray emission which is incompatible with our data. At the opposite extreme, values of log U higher than -0.2 must be accompanied by temperatures $`\genfrac{}{}{0pt}{}{_<}{^{}}1.2\times 10^6`$ K and so by an insufficient gas emissivity to explain the observed OVII-OVIII K$`\alpha ,\beta `$ equivalent width.
The tighteness of this constraints can be used to check the consistency between our model of a โtruly warmโ X-ray absorber/emitter and the hypothesis of pressure equilibrium of the X-ray absorber/emitter of NGC 5548 with the high ionization BELRs of this source. The condition of pressure equilibrium gives a density for the X-ray absorber of: $`n_{H_{WA}}=p_{BELR}/T_{WA}2\times 10^9`$ cm<sup>-3</sup> ($`p_{BELR}2\times 10^{15}`$ K cm<sup>-3</sup>, Krolik et al., 1991). Hence, using the ionizing luminosity of NGC 5548 as estimated by our data (L$`{}_{ion}{}^{}=2\times 10^{44}`$ erg s<sup>-1</sup>, which gives a rate of ionizing photons of Q$`{}_{ion}{}^{}=1.2\times 10^{54}`$ ph s<sup>-1</sup>), and the adopted value of log U = -0.2, we obtain a distance of the warm gas from the central source of $`R_{WA}20`$ lt-day, consistent with the estimated distance of the BELR (Wandel, Peterson & Malkan, 1999). Finally, assuming a BELR density of $`n_{H_{BELR}}=110\times 10^{10}`$ cm<sup>-3</sup> (Krolik et al., 1991, Ferland et al., 1992), the adopted values of U and $`T_{WA}`$ imply $`U_{BELR}=0.10.01`$, consistent with the estimates obtained for the BELR of NGC 5548 based on reverberation studies (Ferland et al., 1992). We then propose that NGC 5548 hosts a โtruly warmโ absorber/emitter in pressure equilibrium with the BELR clouds, at a distance of few lt-day from the central source. Variability studies of this component will be crucial to establish the validity of this model.
## 7 Conclusion
We analyzed the BeppoSAX data of a long observation of the Seyfert 1 galaxy NGC 5548, and presented the results of our analysis. In the following we summarize our main findings:
* The broad band of BeppoSAX allowed us to strongly constrain the shape of the intrinsic continuum of NGC 5548. We found that, for most of the duration of the observation, it is well described by an unusually flat, cut-off power law with $`<\mathrm{\Gamma }>=1.59`$, and $`E_c=115`$ keV, over the whole 0.1-200 keV band. No additional soft component is required by our data.
* A clear hump around 20-30 keV is present in the raw BeppoSAX data of NGC 5548, as well as an apparently narrow emission line feature at 6.3 keV. We interpreted both of these components as due to reprocessing of the primary radiation by cold matter (possibly the accretion disk) illuminated by the central source and covering $`50\%`$ of the sky seen by the primary radiation, consistent with previous observations.
* We performed time resolved spectral analysis of our data and detected a strong, highly energy dependent, spectral variability. Most of this spectral variability can be accounted for by a simple change of the intrinsic spectral slope, with no, or small, total luminosity changes. We can rule out changes being due to variations in the relative intensity of a putative soft component.
We showed that this behaviour is compatible with the X-ray spectrum being generated in a non-pair-dominated hot corona.
* We confirmed the presence of a warm absorber in NGC 5548, with physical properties fully consistent with those deduced by previous measurements. OVII and OVIII are also the most abundant oxygen ions in the ionized absorber.
* We found a marginal evidence for a narrow emission feature at $`0.6`$ keV, consistent with the energy of the oscillator-strength-weighted blend of the OVII K$`\alpha _{1,2,3}`$, OVIII K$`\alpha _{1,2}`$ and OVII K$`\beta `$ emission lines.
The 3$`\sigma `$ lower limit on the equivalent width of this emission line is EW$`>16`$ eV. The same ionized gas, if in photoionization equilibrium (and unless peculiar and asymmetric geometries are present), cannot be responsible for both the observed 0.5-2 keV absorption features and a such a strong feature.
* We propose, as an alternative, that the ionized absorbing gas is truly a warm absorber (at $`10^6`$ K), so that collisional ionization competes with photoionization. The temperature and ionization parameter suggested by our data, give a location for the warm gas coincident with the BELR in NGC 5548. We also propose that this gas is in pressure equilibrium with the BELR clouds of NGC 5548.
F.N would like to thank Smita Mathur for the usefull discussions. An anonymous referee provided helpful suggestions which improved the presentation of our data. We thank the SAX Scientific Data Center. This work was supported in part by NASA grants NAG5-6078 (LTSA), NAG5-3039 and NAG5-2476. |
warning/0001/quant-ph0001017.html | ar5iv | text | # Some conceptual issues involving probability in quantum mechanics It is a pleasure to dedicate this article to Arthur Fine. The subject of our paper is close to one of Arthurโs best known articles on the foundations of physics [Fine 1982].
## 1 Introduction
The issue of the completeness of quantum mechanics has been a subject of intense research for almost a century. One of the most influential papers is undoubtedly that of Eintein, Podolski and Rosen \[Einstein et al. 1935\], where after analyzing entangled two-particle states they concluded that quantum mechanics could not be considered a complete theory. In 1964 John Bell showed that not only was quantum mechanics incomplete but, if one wanted a complete description of reality that was local, one would obtain correlations that are incompatible with the ones predicted by quantum mechanics \[Bell 1987\]. This happens because some quantum mechanical states do not allow for the existence of joint probability distributions of all the possible outcomes of experiments. If a joint distribution exists, then one could consistently create a local hidden variable that would factor this distribution. The nonexistence of local hidden variables that would โcompleteโ quantum mechanics, hence the nonexistence of joint probability distributions, was verified experimentally in 1982 by Aspect, Dalibard and Roger \[Aspect at al. 1982\], when they showed, in a series of beautifully designed experiments, that an entangled photon state of the form
$$|\psi =\frac{1}{\sqrt{2}}(|+|+),$$
(1)
(where $`|+|+_A|_B`$ represents, for example, two photons $`A`$ and $`B`$ with helicity $`+1`$ and $`1,`$ respectively) violates the Clauser-Horne-Shimony-Holt form of Bellโs inequalities \[Clauser et al. 1969\], as predicted by quantum mechanical computations. More recently, Weihs et al. confirmed Aspectโs experiment with a truly random selection of the polarization angles, thus with a more strict nonlocality criteria satisfied \[Weihs et al. 1998\]. We note that the proof that the Clauser et al. form of Bellโs inequalities implies the existence of a joint probability distribution of the observable random variables is the mains result in \[Fine 1982\].
The nonexistence of joint probability distributions also comes into play in the consistent-history interpretation of quantum mechanics. In this interpretation, each sequence of properties for a given quantum mechanical system represents a possible history for this system, and a set of such histories is called a family of histories \[Gell-Mann and Hartle 1990\]. A family of *consistent* histories is one that has a joint probability distribution for all possible histories in this family, with the joint probability distribution defined as any probability measure on the space of all histories. One can easily show that quantum mechanics implies the nonexistence of such probability functions for some families of histories. Families of histories that do not have a joint probability distribution are called inconsistent histories.
Another important example, also related to the nonexistence of a joint probability distribution, is the famous Kochen-Specker theorem, that shows that a given hidden-variable theory that is consistent with the quantum mechanical results has to be contextual \[Kochen and Specker 1967\], i.e., the hidden variable has to depend on the values of the actual experimental settings, regardless of how far apart the actual components of the experiment are located (throughout this paper, we will use interchangeably the concepts of local and noncontextual hidden variables; for a detailed discussion, see \[Suppes and Zanotti 1976\] and \[DโEspagnat 1989\]).
More recently, a marriage between Bellโs inequalities and the Kochen-Specker theorem led to the Greenberger-Horne-Zeilinger (GHZ) theorem. The GHZ theorem shows that if one assumes that one can consistently assign values to the outcomes of a measurement before the measure is performed, a mathematical contradiction arises \[Greenberger et al. 1989\] โ once again, having a complete data table would allow us to compute the joint probability distribution, so we conclude that no joint distribution exists that is consistent with quantum mechanical results. In this paper, we propose the usage of nonmonotonic upper probabilities as a tool to derive consistent joint upper probabilities for the contextual hidden variables.
## 2 The GHZ Theorem
In 1989 Greenberger, Horne and Zeilinger (GHZ) proved that if the quantum mechanical predictions for entangled states are correct, then the assumption that there exist noncontextual hidden variables that can accommodate those predictions leads to contradictions \[Greenberger et al. 1989\]. Their proof of the incompatibility of noncontextual hidden variables with quantum mechanics is now known as the GHZ theorem. This theorem proposes a new test for quantum mechanics based on correlations between more than two particles. What makes the GHZ theorem distinct from Bellโs inequalities is the fact that they use only perfect correlations. The argument for the GHZ theorem, as stated by Mermin \[Mermin 1990a\], goes as follows. We start with a three-particle entangled state
$$|\psi =\frac{1}{\sqrt{2}}(|+_1|+_2|_3+|_1|_2|+_3),$$
(2)
where we use a notation similar to that of equation (1). This state is an eigenstate of the following spin operators:
$`\widehat{๐}`$ $`=`$ $`\widehat{\sigma }_{1x}\widehat{\sigma }_{2y}\widehat{\sigma }_{3y},\widehat{๐}=\widehat{\sigma }_{1y}\widehat{\sigma }_{2x}\widehat{\sigma }_{3y},`$ (3)
$`\widehat{๐}`$ $`=`$ $`\widehat{\sigma }_{1y}\widehat{\sigma }_{2y}\widehat{\sigma }_{3x},\widehat{๐}=\widehat{\sigma }_{1x}\widehat{\sigma }_{2x}\widehat{\sigma }_{3x}.`$ (4)
If we compute the expected values for the correlations above, we obtain at once that $`E(\widehat{๐})=E(\widehat{๐})=E(\widehat{๐})=1`$ and $`E(\widehat{๐})=1.`$ Let us now suppose that the value of the spin for each particle is dictated by a hidden variable $`\lambda `$, and let us call this value $`s_{ij}(\lambda ),`$ where $`i=1\mathrm{}3`$ and $`j=x,y.`$ Then, we have that
$`E(\widehat{๐}\widehat{๐}\widehat{๐})`$ $`=`$ $`(s_{1x}s_{2y}s_{3y})(s_{1y}s_{2x}s_{3y})(s_{1y}s_{2y}s_{3x})`$ (5)
$`=`$ $`s_{1x}s_{2x}s_{3x}(s_{1y}^2s_{2y}^2s_{3y}^2).`$ (6)
Since the $`s_{ij}(\lambda )`$ can only be $`1`$ or $`1,`$ we obtain
$$E(\widehat{๐}\widehat{๐}\widehat{๐})=s_{1x}s_{2x}s_{3x}=E(\widehat{๐}).$$
(7)
But (5) implies that $`E(\widehat{๐}\widehat{๐}\widehat{๐})=1`$ whereas (7) implies $`E(\widehat{๐}\widehat{๐}\widehat{๐})=E(\widehat{๐})=1,`$ a clear contradiction. It is clear from the above derivation that one could avoid contradictions if we allowed the value of $`\lambda `$ to depend on the experimental setup, i.e., if we allowed $`\lambda `$ to be a contextual hidden variable. In other words, what the GHZ theorem proves is that noncontextual hidden variables cannot reproduce quantum mechanical predictions.
This striking characteristic of GHZโs predictions, however, has a major problem. How can one verify experimentally predictions based on correlation-one statements, since experimentally one cannot obtain events perfectly correlated? This problem was also present on Bellโs original paper, where he considered cases where the correlations were one. To โavoid Bellโs experimentally unrealistic restrictionsโ, Clauser, Horne, Shimony and Holt \[Clauser et al. 1969\] derived a new set of inequalities that would take into account imperfections in the measurement process. However, Bellโs inequalities are quite different from the GHZ case, where it is necessary to have experimentally unrealistic perfect correlations. This can be seen from the following theorem (a version for a 4 particle entangled system is found in \[Suppes et al. 1998\]).
Let $`๐,`$ $`๐,`$ and $`๐`$ be three $`\pm 1`$ random variables and let
(i) $`E(๐)=E(๐)=E(๐)=1,`$
(ii) $`E(\mathrm{๐๐๐})=1,`$
then (i) and (ii) imply a contradiction.
*Proof:* By definition
$$E(๐)=P(a)P(\overline{a}),$$
(8)
where we use a notation where $`a`$ is $`๐=1`$, $`\overline{a}`$ is $`๐=1`$, and so on. Since $`0P(a),P(\overline{a})1`$, it follows at once from (i) that
$$P(a)=1$$
(9)
and similarly
$$P(b)=P(c)=1.$$
(10)
Using again the definition of expectation and the inequalities $`P(\overline{a}bc)P(\overline{a})=0,`$ etc., we have
$$\begin{array}{ccc}E(\mathrm{๐๐๐})\hfill & =\hfill & P(abc)+P(\overline{ab}c)+P(a\overline{bc})+P(\overline{a}b\overline{c})\hfill \\ & =\hfill & P(abc)[P(\overline{a}bc)+P(a\overline{b}c)+P(ab\overline{c})+P(\overline{a}\overline{b}\overline{c})]\hfill \\ & =\hfill & 1,\hfill \end{array}$$
(11)
from (9) and (10), since all but the first term on the right is 0, and thus by conservation of probability $`P(ABC)=1`$. But (11) contradicts (ii).
It is important to note that if we could measure all the random variables simultaneously, we would have a joint distribution. The existence of a joint probability distribution is a necessary and sufficient condition for the existence of a noncontextual hidden variable \[Suppes and Zanotti 1981\]. Hence, if the quantum mechanical GHZ correlations are obtained, then no noncontextual hidden variable exists. However, this abstract version of the GHZ theorem still involves probability-one statements. On the other hand, the correlations present in the GHZ state are so strong that even if we allow for experimental errors, the non-existence of a joint distribution can still be verified, as we show in the following theorem \[Barros and Suppes 2000\].
If $`๐,`$ $`๐,`$ and $`๐`$ are three $`\pm 1`$ random variables, a joint probability distribution exists for the given expectations $`E(๐),`$ $`E(๐),`$ $`E(๐),`$ and $`E(\mathrm{๐๐๐})`$ if and only if the following inequalities are satisfied:
$$2E(๐)+E(๐)+E(๐)E(\mathrm{๐๐๐})2,$$
(12)
$$2E(๐)+E(๐)+E(๐)+E(\mathrm{๐๐๐})2,$$
(13)
$$2E(๐)E(๐)+E(๐)+E(\mathrm{๐๐๐})2,$$
(14)
$$2E(๐)+E(๐)E(๐)+E(\mathrm{๐๐๐})2.$$
(15)
*Proof:* First we prove necessity. Let us assume that there is a joint probability distribution consisting of the eight atoms $`abc,`$ $`ab\overline{c},`$ $`a\overline{b}c,`$ $`\mathrm{}\overline{a}\overline{b}\overline{c}`$. Then,
$$E(๐)=P(a)P(\overline{a}),$$
where
$$P(a)=P(abc)+P(a\overline{b}c)+P(ab\overline{c})+P(a\overline{b}\overline{c}),$$
and
$$P(\overline{a})=P(\overline{a}bc)+P(\overline{a}\overline{b}c)+P(\overline{a}b\overline{c})+P(\overline{a}\overline{b}\overline{c}).$$
Similar equations hold for $`E(๐)`$ and $`E(๐).`$ For $`E(\mathrm{๐๐๐})`$ we obtain
$`E(\mathrm{๐๐๐})`$ $`=`$ $`P(\mathrm{๐๐๐}=1)P(\mathrm{๐๐๐}=1)`$
$`=`$ $`P(abc)+P(a\overline{b}\overline{c})++P(\overline{a}\overline{b}c)+P(\overline{a}b\overline{c})`$
$`[P(a\overline{b}c)+P(ab\overline{c})+P(\overline{a}bc)+P(\overline{a}\overline{b}\overline{c})].`$
Corresponding to the first inequality above, we now sum over the probability expressions for the expectations
$$F=E(๐)+E(๐)+E(๐)E(\mathrm{๐๐๐}),$$
and obtain the expression
$`F`$ $`=`$ $`2[P(abc)+P(\overline{a}bc)+P(a\overline{b}c)+P(ab\overline{c})]`$
$`2[P(\overline{a}\overline{b}\overline{c})+P(\overline{a}\overline{b}c)+P(\overline{a}b\overline{c})+P(a\overline{b}\overline{c})],`$
and since all the probabilities are nonnegative and sum to $`1`$, we infer at once inequality (12). The derivation of the other three inequalities is very similar.
To prove the converse, i.e., that these inequalities imply the existence of a joint probability distribution, is slightly more complicated. We restrict ourselves to the symmetric case
$$P(a)=P(b)=P(c)=p,$$
$$P(\mathrm{๐๐๐}=1)=q$$
and thus
$$E(๐)=E(๐)=E(๐)=2p1,$$
$$E(\mathrm{๐๐๐})=2q1.$$
In this case, (12) can be written as
$$03pq2,$$
while the other three inequalities yield just $`0p+q2`$. Let
$$x=P(\overline{a}bc)=P(a\overline{b}c)=P(ab\overline{c}),$$
$$y=P(\overline{a}\overline{b}c)=P(\overline{a}b\overline{c})=P(a\overline{b}\overline{c}),$$
$$z=P(abc),$$
and
$$w=P(\overline{a}\overline{b}\overline{c}).$$
It is easy to show that on the boundary $`3p=q`$ defined by the inequalities the values $`x=0,`$ $`y=q/3,`$ $`z=0,`$ $`w=1q`$ define a possible joint probability distribution, since $`3x+3y+z+w=1`$. On the other boundary, $`3p=q+2`$ a possible joint distribution is $`x=(1q)/3,`$ $`y=0,`$ $`z=q,`$ $`w=0`$. Then, for any values of $`q`$ and $`p`$ within the boundaries of the inequality we can take a linear combination of these distributions with weights $`(3pq)/2`$ and $`1(3pq)/2`$, chosen such that the weighed probabilities add to one, and obtain the joint probability distribution:
$`x`$ $`=`$ $`\left(1{\displaystyle \frac{3pq}{2}}\right){\displaystyle \frac{1q}{3}},`$
$`y`$ $`=`$ $`\left({\displaystyle \frac{3pq}{2}}\right){\displaystyle \frac{q}{3}},`$
$`z`$ $`=`$ $`\left(1{\displaystyle \frac{3pq}{2}}\right)q,`$
$`w`$ $`=`$ $`{\displaystyle \frac{3pq}{2}}\left(1q\right),`$
which proves that if the inequalities are satisfied a joint probability distribution exists, and therefore a noncontextual hidden variable as well, thus completing the proof. The generalization to the asymmetric case is tedious but straightforward.
As a consequence of the inequalities above, one can show that the correlations present in the GHZ state are so strong that even if we allow for experimental errors, the non-existence of a joint distribution can still be verified \[Barros and Suppes 2000\].
Let $`๐,`$ $`๐,`$ and $`๐`$ be three $`\pm 1`$ random variables such that
(i) $`E(๐)=E(๐)=E(๐)1ฯต`$,
(ii) $`E(\mathrm{๐๐๐})1+ฯต`$,
where $`ฯต`$ represents a decrease of the observed $`GHZ`$ correlations due to experimental errors. Then, there cannot exist a joint probability distribution of $`๐,`$ $`๐,`$ and $`๐`$ if
$$ฯต<\frac{1}{2}.$$
(16)
*Proof:* To see this, let us compute the value of $`F`$ define above. We obtain at once that
$$F=3(1ฯต)(1+ฯต).$$
But the observed correlations are only compatible with a noncontextual hidden variable theory if $`F2`$, hence $`ฯต<\frac{1}{2}.`$ Then, there cannot exist a joint probability distribution of $`๐,`$ $`๐,`$ and $`๐`$ satisfying (i) and (ii) if
$$ฯต<\frac{1}{2}.$$
(17)
From the inequality obtained above, it is clear that any experiment that obtains GHZ-type correlations stronger than $`0.5`$ cannot have a joint probability distribution. For example, the recent experiment made at Innsbruck \[Bouwemeester et al. 1999\] with three-photon entangled states supports the quantum mechanical result that no noncontextual hidden variable exists that explain their correlations \[Barros and Suppes 2000\]. Thus, with this reformulation of the GHZ theorem it is possible to use strong, yet imperfect, experimental correlations to prove that a noncontextual hidden-variable theory is incompatible with the experimental results.
## 3 Upper and Lower Probabilities and the GHZ theorem
We saw at the previous section that quantum mechanics does not allow, for some cases, the definition of a joint probability distribution for all the observables. However, if we weaken the probability axioms, it is possible to prove that one can find a consistent set of upper probabilities for the events \[Suppes and Zanotti 1991\]. Upper probabilities are defined in the following way. Let $`\mathrm{\Omega }`$ be a nonempty set, $`F`$ a boolean algebra on $`\mathrm{\Omega }`$ and $`P^{}`$ a real valued function on $`F.`$ Then the triple $`(\mathrm{\Omega },F,P^{})`$ is an *upper probability* if for all $`\xi _1`$ and $`\xi _2`$ in $`F`$ we have that
$`0P^{}(\xi _1)1,`$
$`P^{}(\mathrm{})=0,`$
$`P^{}(\mathrm{\Omega })=1,`$
and if $`\xi _1`$ and $`\xi _2`$ are disjoint, i.e. $`\xi _1\xi _2=\mathrm{},`$ then
$`P^{}(\xi _1\xi _2)P^{}(\xi _1)+P^{}(\xi _2).`$
As we can see, this last property weakens the standard axioms for probability, as one of the consequences of these axioms is that it may be true, for an upper probability, that
$$\xi _1\xi _2\text{ and }P^{}(\xi _1)>P^{}(\xi _2),$$
a quite nonstandard property. In a similar way, *lower probabilities* are defined as satisfying the triple $`(\mathrm{\Omega },F,P_{})`$ such that for all $`\xi _1`$ and $`\xi _2`$ in $`F`$ we have that
$`0P_{}(\xi _1)1,`$
$`P_{}(\mathrm{})=0,`$
$`P_{}(\mathrm{\Omega })=1,`$
and if $`\xi _1`$ and $`\xi _2`$ are disjoint, i.e. $`\xi _1\xi _2=\mathrm{},`$ then
$`P_{}(\xi _1\xi _2)P_{}(\xi _1)+P_{}(\xi _2).`$
Let us see how upper and lower probabilities can be used to obtain joint upper and lower probability distributions. We can start with the standard Bellโs variables $`๐,`$ $`๐`$ and $`๐,`$ where each random variable represents a different angles for the Stern-Gerlach apparatus (we follow the example in \[Suppes and Zanotti 1991\]). In the experimental setup used by Bell, a two-particle system with entangled spin state was used, and for that reason we can only measure two variables at the same time. However, since they are spin measurements, we have the constraint
$$P(๐=1)=P(๐=1)=P(๐=1)=\frac{1}{2}.$$
The question that Bell posed is whether we can fill the missing values of the data table in a way that is consistent with the correlations given by quantum mechanics for the pairs of variables, that is, $`E(\mathrm{๐๐}),`$ $`E(\mathrm{๐๐}),`$ $`E(\mathrm{๐๐}).`$ It is well known that for some sets of angles, the joint probability distribution of $`๐,`$ $`๐,`$ and $`๐`$ exists, while for other set of angles it does not exist. We can prove that the joint doesnโt exist in the following way. We start with the values for the correlations used by Bell:
$`E(\mathrm{๐๐})`$ $`=`$ $`{\displaystyle \frac{\sqrt{3}}{2}},`$ (18)
$`E(\mathrm{๐๐})`$ $`=`$ $`{\displaystyle \frac{\sqrt{3}}{2}},`$ (19)
$`E(\mathrm{๐๐})`$ $`=`$ $`{\displaystyle \frac{1}{2}}.`$ (20)
The correlations above correspond to the angles $`\widehat{\mathrm{๐๐}}=30^\text{o},`$ $`\widehat{\mathrm{๐๐}}=30^\text{o}`$ and $`\widehat{\mathrm{๐๐}}=60^\text{o}`$ for the detectors, and require that
$`E(\mathrm{๐๐})`$ $`=`$ $`E(\mathrm{๐๐}|๐=1)P(๐=1)+E(\mathrm{๐๐}|๐=1)P(๐=1),`$
$`E(\mathrm{๐๐})`$ $`=`$ $`E(\mathrm{๐๐}|๐=1)P(๐=1)+E(\mathrm{๐๐}|๐=1)P(๐=1),`$
$`E(\mathrm{๐๐})`$ $`=`$ $`E(\mathrm{๐๐}|๐=1)P(๐=1)+E(\mathrm{๐๐}|๐=1)P(๐=1),`$
which can be written as
$`2E(\mathrm{๐๐})`$ $`=`$ $`E(\mathrm{๐๐}|๐=1)+E(\mathrm{๐๐}|๐=1),`$ (21)
$`2E(\mathrm{๐๐})`$ $`=`$ $`E(\mathrm{๐๐}|๐=1)+E(\mathrm{๐๐}|๐=1),`$ (22)
$`2E(\mathrm{๐๐})`$ $`=`$ $`E(\mathrm{๐๐}|๐=1)+E(\mathrm{๐๐}|๐=1),`$ (23)
because $`P(๐=1)=P(๐=1),`$ etc. Symmetry requires that
$`E(\mathrm{๐๐}|๐=1)`$ $`=`$ $`E(\mathrm{๐๐}|๐=1),`$ (24)
$`E(\mathrm{๐๐}|๐=1)`$ $`=`$ $`E(\mathrm{๐๐}|๐=1)`$ (25)
and if we use the requirement that all probabilities must sum to one we have six equations and six unknown conditional expectations. It is easy to see that the system of linear equations (21)โ(25) does not have a solution for the correlations shown in (18), hence no joint probability distribution exists. What happened? The correlations are too strong for us to fill up a table with all the experimental results, including the ones that did not occur. One extreme example can be obtained if we use the extreme case of correlation one expectations, given by
$`E(\mathrm{๐๐})`$ $`=`$ $`1,`$
$`E(\mathrm{๐๐})`$ $`=`$ $`1,`$
$`E(\mathrm{๐๐})`$ $`=`$ $`1,`$
where once again no joint probability distribution exists.
What changes with upper probabilities? The system of linear equations (21) becomes a system of inequalities:
$`2E^{}(\mathrm{๐๐})`$ $``$ $`E^{}(\mathrm{๐๐}|๐=1)+E^{}(\mathrm{๐๐}|๐=1),`$ (26)
$`2E^{}(\mathrm{๐๐})`$ $``$ $`E^{}(\mathrm{๐๐}|๐=1)+E^{}(\mathrm{๐๐}|๐=1),`$ (27)
$`2E^{}(\mathrm{๐๐})`$ $``$ $`E^{}(\mathrm{๐๐}|๐=1)+E^{}(\mathrm{๐๐}|๐=1),`$ (28)
plus the symmetry
$`E^{}(\mathrm{๐๐}|๐=1)`$ $`=`$ $`E^{}(\mathrm{๐๐}|๐=1),`$ (29)
$`E^{}(\mathrm{๐๐}|๐=1)`$ $`=`$ $`E^{}(\mathrm{๐๐}|๐=1),`$ (30)
and the fact that the sum of all upper probabilities must be greater or equal than one. It is straightforward to obtain solutions to (26)โ(30), and then we can find upper probabilities that are consistent with the conditional expectations.
The following theorem shows that the GHZ theorem fail if we allow lower probabilities.
Let $`๐,`$ $`๐,`$ and $`๐`$ be three $`\pm 1`$ random variables and let
(i) $`E_{}(๐)=E(๐)=1,`$
(ii) $`E_{}(๐)=E(๐)=1,`$
(iii) $`E_{}(๐)=E(๐)=1,`$
(iv) $`E_{}(\mathrm{๐๐๐})=E(\mathrm{๐๐๐})=1.`$
Then, there exist a lower joint probability distribution that is compatible with (i)โ(iv).
*Proof:* We will prove this theorem by explicitly constructing a lower joint probability distribution. First, we note that
$$E_{}(๐)=P_{}(a)P_{}(\overline{a})=1,$$
$$E_{}(๐)=P_{}(b)P_{}(\overline{b})=1,$$
$$E_{}(๐)=P_{}(c)P_{}(\overline{c})=1,$$
and hence
$`P_{}(a)=1,`$ $`P_{}(\overline{a})=0,`$ (31)
$`P_{}(b)=1,`$ $`P_{}(\overline{b})=0,`$ (32)
$`P_{}(c)=1`$ $`P_{}(\overline{c})=0.`$ (33)
From the definition of lowers and from (31)โ(33) we have
$`P_{}(abc)+P_{}(a\overline{b}c)+P_{}(ab\overline{c})+P_{}(a\overline{b}\overline{c})`$ $``$ $`1,`$ (34)
$`P_{}(abc)+P_{}(\overline{a}bc)+P_{}(ab\overline{c})+P_{}(\overline{a}b\overline{c})`$ $``$ $`1,`$ (35)
$`P_{}(abc)+P_{}(\overline{a}bc)+P_{}(a\overline{b}c)+P_{}(\overline{a}\overline{b}c)`$ $``$ $`1,`$ (36)
and from (iv)
$`P_{}(abc)+P_{}(\overline{a}\overline{b}c)+P_{}(a\overline{b}\overline{c})+P_{}(\overline{a}b\overline{c})+`$ (37)
$`P_{}(\overline{a}bc)P_{}(a\overline{b}c)P_{}(ab\overline{c})P_{}(\overline{a}\overline{b}\overline{c})`$ $`=`$ $`1.`$ (38)
The lowers must also be superadditive in the whole probability space, and we have
$`P_{}(abc)+P_{}(\overline{a}\overline{b}c)+P_{}(a\overline{b}\overline{c})+P_{}(\overline{a}b\overline{c})+`$ (39)
$`P_{}(\overline{a}bc)+P_{}(a\overline{b}c)+P_{}(ab\overline{c})+P_{}(\overline{a}\overline{b}\overline{c})`$ $``$ $`1.`$ (40)
From (38) and (40) we have
$$P_{}(abc)=P_{}(\overline{a}\overline{b}c)=P_{}(a\overline{b}\overline{c})=P_{}(\overline{a}b\overline{c})=0$$
and the system reduces to
$`P_{}(a\overline{b}c)+P_{}(ab\overline{c})`$ $``$ $`1,`$ (41)
$`P_{}(\overline{a}bc)+P_{}(ab\overline{c})`$ $``$ $`1,`$ (42)
$`P_{}(\overline{a}bc)+P_{}(a\overline{b}c)`$ $``$ $`1,`$ (43)
$`P_{}(\overline{a}bc)+P_{}(a\overline{b}c)+P_{}(ab\overline{c})+P_{}(\overline{a}\overline{b}\overline{c})`$ $`=`$ $`1.`$ (44)
A possible solution for the system (41)โ(44) is
$`P_{}(\overline{a}bc)=P_{}(a\overline{b}c)=P_{}(ab\overline{c})`$ $`=`$ $`{\displaystyle \frac{1}{3}}`$
$`P_{}(\overline{a}\overline{b}\overline{c})`$ $`=`$ $`0,`$
as we wanted to prove. In a similar way, we have the following:
Let $`๐,`$ $`๐,`$ and $`๐`$ be three $`\pm 1`$ random variables and let
(i) $`E^{}(๐)=E(๐)=1,`$
(ii) $`E^{}(๐)=E(๐)=1,`$
(iii) $`E^{}(๐)=E(๐)=1,`$
(iv) $`E^{}(\mathrm{๐๐๐})=E(\mathrm{๐๐๐})=1.`$
Then, there exist an upper probability distribution that is compatible with (i)โ(iv).
*Proof:* Similar to the proof for the lower.
We note that the nonmonotonic upper and lower probabilities shown to exist in Theorems 3 and 4 do not, because of their nonmonotonicity, satisfy the usual definitional relation between upper and lower probabilities, for any event $`A`$:
$$P^{}(A)=1P_{}(\overline{A}).$$
## 4 Final Remarks
To apply the upper probabilities to the GHZ theorem, we gave a probabilistic random variable version of it. We then showed that, if we use upper probabilities, the GHZ theorem does not hold anymore, and hence the inconsistencies cannot be proved to exist for the upper probabilities. Such upper probabilities are a natural way to deal with contextual problems in statistics. Whether they lead to fruitful theoretical developments in a new direction is, however, an open question. |
warning/0001/math0001095.html | ar5iv | text | # Pentagon equation and matrix bialgebras
## 1 Introduction
This is the first article in the series of papers on algebraic constructions related to $`6j`$-symbols for finite fusion rules. The subject of this article can be considered as the case of one element fusion rule. Instead of starting with categorical setting (which is treated in the last section) we consider equivalent problem of describing bialgebra structures on matrix algebra and reformulate it as a problem of solving some equation closely connected with the so-called pentagon equation (see ). The set of solutions to this modified pentagon equation projects onto the set of solutions to ordinary pentagon equation. We extend Baaj and Skandalisโ description of unitary solutions to pentagon equation to the general case which says that any solution corresponds up to a multiplicity to some Hopf algebra. We give an algebraic description of solutions to modified pentagon equation. In particular, we establish the relation between Hopf-Galois algebras and solutions to modified pentagon equation when the corresponding solution to pentagon equation has multiplicity one.
The work was partially supported by RFBR grant no. 99-01-01144.
## Acknowledgment
The work was started during my visit of Macquarie University (Sydney, Australia) and was completed in Max-Planck-Institut fรผr Mathematik (Bonn, Germany). I would like to thank these institutions for hospitality and inspiring atmosphere. Special thanks to Prof. Ross Street who pointed out to me the references and explained his own work .
## 2 Matrix bialgebras and pentagon equation
In this section we give a description of bialgebra structures on the endomorphism algebra $`End(V)`$ of vector space $`V`$ in terms of some tensors. Remind that a bialgebra is an algebra $`E`$ together with a unital (identity preserving) homomorphism of algebras $`\mathrm{\Delta }:EEE`$ (comultiplication) satisfying the so-called coassociativity axiom: $`(\mathrm{\Delta }I)\mathrm{\Delta }=(I\mathrm{\Delta })\mathrm{\Delta }`$. We start with the following auxiliary statement.
###### Lemma 2.1
Let $`f:End(V)End(U)`$ be a unital homomorphism of endomorphism algebras. Then there exists a vector space $`W`$ and an isomorphism $`UVW`$. The composition $`\stackrel{~}{f}:End(V)End(VW)`$ of $`f`$ with the factorization isomorphism has a form $`\stackrel{~}{f}(x)=xI`$.
The centralizer of the subalgebra $`End(V)I`$ in the algebra $`End(VW)`$ coincides with $`1End(W)`$.
Proof:
The algebra $`End(V)`$ is simple. Hence any its module is a direct sum of simple one. The natural $`End(V)`$-module $`V`$ is simple. The homomorphism $`f`$ defines $`End(V)`$-module structure on vector space $`U`$. Now evaluation map $`VHom_{End(V)}(V,U)U`$ gives an isomorphism. Moreover this isomorphism is an isomorphism of $`End(V)`$-modules and the action of $`End(V)`$ on $`VHom_{End(V)}(V,U)`$ has the form $`x(vl)=x(v)l`$. Denoting $`Hom_{End(V)}(V,U)`$ by $`W`$ we get the first part of the statement.
The second part of the lemma follows from the identifications:
$$C_{End(VW)}(End(V)I)End_{End(V)}(VW)=End(W).$$
$`\mathrm{}`$
For a linear map $`F:VMVV`$ define the maps $`F_{12},F_{13},F_{23}`$ which are map $`F`$ acting on first and second (first and third, second and third) tensor functor of appropriate tensor product.
###### Theorem 2.2
Any bialgebra comultiplication $`\mathrm{\Delta }:End(V)End(V)End(V)`$ is defined by an isomorphism $`F:VMVV`$ (for some vector space $`M`$) such that
$$F_{12}F_{13}F_{23}F_{12}mod(1Aut(M^2))$$
(1)
in the following way
$$\mathrm{\Delta }_F(x)=F(x1)F^1,xEnd(V).$$
The bialgebras $`(End(V),\mathrm{\Delta }_F),(End(V),\mathrm{\Delta }_F^{})`$ defined by the automorphisms $`F,F^{}`$ are isomorphic iff there exist an isomorphisms $`f:VV,g:MM^{}`$ such that
$$(ff)F=F^{}(fg).$$
(2)
Proof:
Denoting $`Hom_{End(V)}(V,V^2)`$ by $`M`$ we get desired isomorphism (evaluation map):
$$F:VMVV$$
which by lemma 2.1 conjugates given inclusion $`\mathrm{\Delta }:End(V)End(V)End(V)`$ with the standard one $`End(V)End(V)End(M)`$ ($`x(x1)`$) so that $`\mathrm{\Delta }(x)=\mathrm{\Delta }_F(x)=F(x1)F^1`$.
Coassociativity of the comultiplication $`\mathrm{\Delta }_F`$ is equivalent to the congruence (1). Indeed, the equality between
$$(\mathrm{\Delta }_FI)\mathrm{\Delta }_F(x)=F_{12}F_{13}(x11)F_{13}^1F_{12}^1$$
and
$$(I\mathrm{\Delta }_F)\mathrm{\Delta }_F(x)=F_{23}F_{12}(x11)F_{12}^1F_{23}^1$$
means that $`F_{12}^1F_{23}^1F_{12}F_{13}`$ lies in the centralizer of $`End(V)11`$ in $`End(VMM)`$ which by lemma 2.1 coincides with $`1End(M^2)`$. $`\mathrm{}`$
We say that the pair of isomorphisms $`(F,\mathrm{\Phi }),F:VMVV,\mathrm{\Phi }Aut(M^2)`$ satisfies to modified pentagon equation on vector spaces $`V,M`$ if
$$F_{12}F_{13}\mathrm{\Phi }_{23}=F_{23}F_{12}.$$
(3)
For any solution $`(F,\mathrm{\Phi })`$ of modified pentagon equation automorphism $`F`$ satisfies to the congruence (1) and vise versa any solution $`F`$ of the congruence (1) defines the solution $`(F,\mathrm{\Phi })`$ of modified pentagon equation by
$$\mathrm{\Phi }_{23}=F_{13}^1F_{12}^1F_{23}F_{12}.$$
Two solution $`F,F^{}`$ of the congruence 1 lie in the same equivalence class of the relation (2)
$$(ff)F=F^{}(fg),\text{ for some }f,gAut(V)$$
if and only if for the pairs $`(F,\mathrm{\Phi }),(F^{},\mathrm{\Phi }^{})`$ we have
$$(ff)F=F^{}(fg),(gg)\mathrm{\Phi }=\mathrm{\Phi }^{}(gg)fAut(V),gAut(M).$$
Indeed,
$$(fff)F_{23}F_{12}=F_{23}^{}(ffg)F_{12}=F_{23}^{}F_{12}^{}(fgg)$$
coincides with
$$(fff)F_{12}F_{13}\mathrm{\Phi }_{23}=F_{12}^{}(fgf)F_{13}\mathrm{\Phi }_{23}=F_{12}^{}F_{13}^{}(fgg)\mathrm{\Phi }_{23}$$
which means that $`(gg)\mathrm{\Phi }=\mathrm{\Phi }^{}(gg)`$.
###### Proposition 2.3
If the pair $`F,\mathrm{\Phi }`$ satisfies to modified pentagon equation (3) then $`\mathrm{\Phi }`$ is a solution to pentagon equation
$$\mathrm{\Phi }_{12}\mathrm{\Phi }_{13}\mathrm{\Phi }_{23}=\mathrm{\Phi }_{23}\mathrm{\Phi }_{12}.$$
(4)
Proof:
Since $`F_{14}\mathrm{\Phi }_{24}F_{12}^1=F_{12}^1F_{24}`$ the expression
$$\mathrm{\Phi }_{23}\mathrm{\Phi }_{24}\mathrm{\Phi }_{34}=F_{14}^1\mathrm{\Phi }_{23}F_{14}\mathrm{\Phi }_{24}F_{12}^1\mathrm{\Phi }_{34}F_{12}$$
coincides with
$$F_{14}^1\mathrm{\Phi }_{23}F_{12}^1F_{24}\mathrm{\Phi }_{34}F_{12}=F_{14}^1\mathrm{\Phi }_{23}F_{12}^1F_{23}^1F_{23}F_{24}\mathrm{\Phi }_{34}F_{12}.$$
Which again using modified pentagon equations $`\mathrm{\Phi }_{23}F_{12}^1F_{23}^1=F_{13}^1F_{12}^1`$ and $`F_{23}F_{24}\mathrm{\Phi }_{34}=F_{34}F_{23}`$ can be transformed into
$$F_{14}^1F_{13}^1F_{12}^1F_{34}F_{23}F_{12}=F_{14}^1F_{13}^1F_{34}F_{12}^1F_{23}F_{12}=F_{14}^1F_{13}^1F_{34}F_{13}F_{13}^1F_{12}^1F_{23}F_{12},$$
which in his turn coincides with
$$\mathrm{\Phi }_{23}\mathrm{\Phi }_{12}$$
since $`F_{14}^1F_{13}^1F_{34}F_{13}=\mathrm{\Phi }_{23}`$ and $`F_{13}^1F_{12}^1F_{23}F_{12}=\mathrm{\Phi }_{12}`$. $`\mathrm{}`$
Note that the pair of identity operators $`(I,I)`$ is a solution to modified pentagon equation.
###### Example 2.4
Let $`\mathrm{\Phi }`$ be a solution of pentagon equation. Then $`(\mathrm{\Phi },\mathrm{\Phi })`$ is a solution to the modified pentagon equation.
The modified pentagon equation for the pair $`(\mathrm{\Phi },\mathrm{\Phi })`$ is an ordinary pentagon equation for $`\mathrm{\Phi }`$.$`\mathrm{}`$
Denote by $`t=t_V:VVVV`$ the permutation of tensor factors. If $`\mathrm{\Delta }`$ is a coproduct on the algebra $`E`$, then $`t_E\mathrm{\Delta }`$ is also a coproduct. Applying this to endomorphism algebra $`E=End(V)`$ we get the following operation on solutions to modified pentagon equation.
###### Example 2.5
Let $`(F,\mathrm{\Phi })`$ be a solution of modified pentagon equation. Then $`(tF,\mathrm{\Phi }_{21}^1)`$ is also a solution. In particular, for any solution $`\mathrm{\Phi }`$ to pentagon equation $`\mathrm{\Phi }_{21}^1`$ is also a solution.
Let us give the direct verification of this fact. First, conjugation modified pentagon equation for $`(F,\mathrm{\Phi })`$ with $`t_{23}`$ we get the following
$$F_{13}F_{12}\mathrm{\Phi }_{32}=F_{32}F_{13}\text{or}F_{32}F_{13}\mathrm{\Phi }_{32}^1=F_{13}F_{12}.$$
Then, since $`t_{12}t_{13}=t_{23}t_{12}`$ the left side of modified pentagon equation for $`(tF,\mathrm{\Phi }_{21}^1)`$
$$t_{12}F_{12}t_{13}F_{13}\mathrm{\Phi }_{32}^1=t_{12}t_{13}F_{32}F_{13}\mathrm{\Phi }_{32}^1$$
coincides with the right side
$$t_{23}F_{23}t_{12}F_{12}=t_{23}t_{12}F_{13}F_{12}.$$
$`\mathrm{}`$
Starting with the pair $`(\mathrm{\Phi }_{21}^1,\mathrm{\Phi }_{21}^1)`$ we get that $`(\mathrm{\Phi }^1t,\mathrm{\Phi })`$ is a solution to the modified pentagon equation for any solution of pentagon equation $`\mathrm{\Phi }`$. In particular, the pair $`(t,I)`$ is a solution to modified pentagon equation.
Tensor product of bialgebras $`(End(V),\mathrm{\Delta }_F),(End(V^{}),\mathrm{\Delta }_F^{})`$ corresponds to the following operation on solutions of modified pentagon equation. $`\mathrm{}`$
###### Example 2.6
Let $`(F,\mathrm{\Phi }),(\mathrm{\Phi }^{},F^{})`$ be a solutions of pentagon equation on $`(V,M)`$ and $`(V^{},M^{})`$ respectively. Then $`(t_{23}(F^{})t_{23},t_{23}(\mathrm{\Phi }\mathrm{\Phi }^{})t_{23})`$ is a solution of pentagon equation on $`(VV^{},MM^{})`$.
In particular, for identity solution $`(F,\mathrm{\Phi }^{})=(I,I)`$ the pair $`(t_{23}(FI)t_{23},t_{23}(\mathrm{\Phi }I)t_{23})`$ is a solution of pentagon equation.
As a partial case of previous example we have the following tensor product operation for solutions of pentagon equation.
###### Example 2.7
Let $`\mathrm{\Phi }Aut(V^2),\mathrm{\Phi }^{}Aut(V_{}^{}{}_{}{}^{2})`$ be a solutions of pentagon equation. Then $`t_{23}(\mathrm{\Phi }\mathrm{\Phi }^{})t_{23}Aut((VV^{})^2)`$ is a solution of pentagon equation.
In particular, for identity solution $`\mathrm{\Phi }^{}`$ the operator $`t_{23}(\mathrm{\Phi }I)t_{23}`$ is a solution of pentagon equation.
## 3 Hopf modules and examples of solutions to pentagon equation
Here we give a way of constructing solutions to pentagon equation using Hopf modules. In the slightly different form this construction appeared in the papers .
Let $`H`$ be a Hopf algebra. Hopf $`H`$-module is a vector space $`M`$ with a structure of left $`H`$-module
$$\mu _M:HMM,\mu _M(hm)=hm$$
and a structure of right $`H`$-comodule
$$\mathrm{\Delta }_M:MMH,$$
compatible in the following sense:
$$\mathrm{\Delta }_M(hm)=\mathrm{\Delta }_H(h)\mathrm{\Delta }_M(m)(H\text{-linearity}).$$
###### Proposition 3.1
Let $`H`$ be a Hopf algebra and $`M`$ be a Hopf $`H`$-module with multiplication $`\mu `$ and comultiplication $`\mathrm{\Delta }`$. Then the formula
$$\mathrm{\Phi }_M=(I\mu )(\mathrm{\Delta }I)$$
defines a solution to pentagon equation on vector space $`MM`$.
Proof:
We will use Sweedlerโs notation for comodule structure $`\mathrm{\Delta }_M(m)=_{(m)}m_{(0)}m_{(1)}`$. For example, the map $`\mathrm{\Phi }_M`$ has a form:
$$\mathrm{\Phi }(mn)=\underset{(m)}{}m_{(0)}m_{(1)}n.$$
It can be checked directly that the inverse map is given by the formula:
$$\mathrm{\Phi }^1(mn)=\underset{(m)}{}m_{(0)}S(m_{(1)})n.$$
For the sake of simplicity we will omit summation signs in verification of pentagon equation:
$$\mathrm{\Phi }_{12}\mathrm{\Phi }_{13}\mathrm{\Phi }_{23}(mnl)=\mathrm{\Phi }_{12}\mathrm{\Phi }_{13}(mn_{(0)}n_{(1)}l)=$$
$$\mathrm{\Phi }_{12}(m_{(0)}n_{(0)}m_{(1)}n_{(1)}l)=m_{(0)}m_{(1)}n_{(0)}m_{(2)}n_{(1)}l.$$
On the other side
$$\mathrm{\Phi }_{23}\mathrm{\Phi }_{12}(mnl)=\mathrm{\Phi }_{23}(m_{(0)}m_{(1)}nl)=m_{(0)}m_{(1)}n_{(0)}m_{(2)}n_{(1)}l.$$
$`\mathrm{}`$
###### Example 3.2
The Hopf algebra $`H`$ considered as a left module and a right comodule over itself is a Hopf module. We shall call this Hopf module trivial.
We can define Hopf module structure on any multiple $`MV`$ of Hopf module $`M`$. It can be checked directly that corresponding solutions of pentagon equations are related as follows:
$$\mathrm{\Phi }_{MV}=t_{23}(\mathrm{\Phi }_MV)t_{23}.$$
The next theorem is known as โFundamental theorem of Hopf modulesโ (see, for example, ). It states that up to a multiplication by vector space any Hopf module is isomorphic to trivial one. Here by a homomorphism of Hopf modules we mean a homomorphism preserving module and comodule structures. For a Hopf module $`M`$ denote by $`M_H=\{mM,\mathrm{\Delta }_M(m)=m1\}`$ the subspace of coinvariant elements.
###### Theorem 3.3
Let $`M`$ be a Hopf $`H`$ module. Then the map
$$HM_HM,hmhm$$
is an isomorphism of Hopf modules where Hopf module structure on $`HM_H`$ is induced by the natural Hopf module structure on $`H`$ (see example 3.2).
Hint of the proof:
Construct inverse map $`MHM_H`$ as $`m_{(m)}m_{(2)}S^1(m_{(1)})m_{(0)}`$ where $`S^1`$ is inverse map to the antipode $`S`$ of the Hopf algebra $`H`$. $`\mathrm{}`$
###### Corollary 3.4
For any Hopf $`H`$-module $`M`$ the solution of pentagon equation $`\mathrm{\Phi }_M`$ is isomorphic to $`\mathrm{\Phi }_HM_H`$.
## 4 Construction of solutions for modified pentagon equation
Here we give a way of constructing solutions to modified pentagon equation for a given solution of pentagon equation $`\mathrm{\Phi }`$ corresponding to Hopf $`H`$-module $`M`$. The initial data is more peculiar then in the case of pentagon equation and consists of:
right $`H`$-module coalgebra $`L`$ with structure maps
$$\mathrm{\Delta }_L:LLL,(\text{coalgebra comultiplication}),$$
$$\mu _L:LHL,(H\text{module structure});$$
right $`L`$-comodule $`V`$
$$\mathrm{\Delta }_V:VVL$$
and the pairing
$$\pi :LMV.$$
(5)
In addition to ordinary axioms of module coalgebra:
$$(\mathrm{\Delta }_LI)\mathrm{\Delta }_L=(I\mathrm{\Delta }_L)\mathrm{\Delta }_L:LLLL,$$
$$\mu _L(\mu _LI)=\mu _L(I\mu _H):LHHL,$$
$$\mathrm{\Delta }_L\mu _L=(\mu _L\mu _H)t_{23}(\mathrm{\Delta }_L\mathrm{\Delta }_H):LHLL$$
and of comodule
$$(\mathrm{\Delta }_VI)\mathrm{\Delta }_V=(I\mathrm{\Delta }_L)\mathrm{\Delta }_V:VVLL$$
we need following properties for the pairing 5:
$$\pi (\mu _LI)=\pi (I\mu _M):LHMV,$$
$$\mathrm{\Delta }_V\pi =(\pi \mu _L)t_{23}(\mathrm{\Delta }_L\mathrm{\Delta }_M):LMVL,$$
where $`\mu _M:HMM`$, $`\mathrm{\Delta }_M:MMH`$ are $`H`$-module and comodule structure on $`M`$ respectively. Note that the first property means that $`\pi `$ factors through tensor product over $`H`$ $`\pi :LML_HMV`$. Since $`MHM^H`$ as an $`H`$-module $`\pi `$is defined by the map $`\overline{\pi }:LM^HV`$. The second property says that $`\overline{\pi }`$ is a homomorphism of $`L`$-comodules where $`L`$-comodule structure on $`LM^H`$ comes from those on $`L`$.
Having all of these we can define a map $`F_V:VMVV`$ by $`F_V=(I\pi )(\mathrm{\Delta }_VI)`$.
###### Proposition 4.1
The map $`F_V`$ satisfies to modified pentagon equation $`F_{12}F_{13}\mathrm{\Phi }_{23}=F_{23}F_{12}`$, where $`\mathrm{\Phi }`$ is the solution of pentagon equation corresponding to Hopf $`H`$-module $`M`$.
Proof:
The proof is similar to the proof of proposition 3.1. $`\mathrm{}`$
Now we give some additional data providing invertibility of the map $`F_V`$.
###### Proposition 4.2
Suppose that our coalgebra is equipped with the counite $`\epsilon _L:Lk`$ and we are given by a pairing $`\nu :LVM`$ such that the compositions
$$\nu (I\pi )(\mathrm{\Delta }_LI):LMM,$$
$$\pi (I\nu )(\mathrm{\Delta }_LI):LVV$$
have a form $`\epsilon _LI`$. Then the map $`F_V`$ is invertible with the inverse $`F_V^1=(I\nu )(\mathrm{\Delta }_LI)`$.
Proof:
Consists of direct checking.$`\mathrm{}`$
Consider as an example the case when vector space $`M^H`$ is one dimensional.
Let $`L`$ be a right $`H`$-module coalgebra . Define
$$F_L=(I\mu _L)(\mathrm{\Delta }_LI):LHLL,$$
where $`\mu _L:LHL`$ is the action map and $`\mathrm{\Delta }_L:LLL`$ is the coalgebra structure. Right $`H`$-module coalgebra $`L`$ is called Galois if the map $`F_L`$ is an isomorphism.
###### Example 4.3
Then the pair $`(F_L,\mathrm{\Phi }_H)`$ is a solution of modified pentagon equation.
## 5 Solutions of pentagon equation
The theorem of Baaj and Skandalis (Theorem 4.10. of ) says that any finite dimensional unitary solution of pentagon equation corresponds to some Hopf module over some Hopf $`C^{}`$-algebra. In this section we give a proof this theorem without any unitary assumptions which works over arbitrary field. We basically follow the scheme of with some minor changes. Namely we use Hamilton-Cayleyโs theorem defining counite and antipode and the notion of Hopf module in formulation of the theorem.
In papers of G.Militaru the variant of the theorem of Baaj and Skandalis was proved for more general situation of arbitrary (not necessary invertible) solution of pentagon equation but the resulting Hopf module there is a Hopf module over bialgebra which does not posses an antipode in general.
According to the theorem 2.2 two solutions $`(\mathrm{\Phi },\mathrm{\Phi })`$ and $`(\mathrm{\Phi }^1t,\mathrm{\Phi })`$ of the modified pentagon equation define two comultiplications on the endomorphism algebra $`End(M)`$
$$\mathrm{\Delta }_l(x)=\mathrm{\Delta }_\mathrm{\Phi }(x)=\mathrm{\Phi }(x1)\mathrm{\Phi }^1,$$
$$\mathrm{\Delta }_r(x)=\mathrm{\Delta }_{\mathrm{\Phi }^1t}(x)=\mathrm{\Phi }^1(1x)\mathrm{\Phi },$$
which give two multiplications $`\mu _l,\mu _r`$ on the dual space $`End(M)^{}`$.
###### Lemma 5.1
The maps defined by
$$\lambda (\omega )=(\omega I)(\mathrm{\Phi }),\rho (\omega )=(I\omega )(\mathrm{\Phi })$$
are a homomorphisms of bialgebras
$$\lambda :(End(M)^{},\mu _r,\mathrm{\Delta })(End(M),,\mathrm{\Delta }_l),\rho :(End(M)^{},\mu _l,\mathrm{\Delta })(End(M),,\mathrm{\Delta }_r),$$
where $``$ is the composition of endomorphisms in $`End(M)`$ and $`\mathrm{\Delta }`$ denotes a comultiplication on $`End(M)^{}`$ dual to $``$.
Proof:
The homomorphism property for $`\lambda `$ follows from pentagon equation for $`\mathrm{\Phi }`$:
$$\lambda (\omega )\lambda (\omega ^{})=(\omega \omega ^{}I)(\mathrm{\Phi }_{13}\mathrm{\Phi }_{23})=(\omega \omega ^{}I)(\mathrm{\Phi }_{12}^1\mathrm{\Phi }_{23}\mathrm{\Phi }_{12})=\lambda (\omega _r\omega ^{}).$$
Analogously for $`\rho `$:
$$\rho (\omega )\rho (\omega ^{})=(I\omega \omega ^{})(\mathrm{\Phi }_{12}\mathrm{\Phi }_{13})=(I\omega \omega ^{})(\mathrm{\Phi }_{23}\mathrm{\Phi }_{12}\mathrm{\Phi }_{23}^1)=\rho (\omega _l\omega ^{}).$$
Now check that $`\lambda `$ is a homomorphism of coalgebras, that is $`\mathrm{\Delta }_l\lambda =(\lambda \lambda )\mathrm{\Delta }`$:
$$(\lambda \lambda )(\mathrm{\Delta }(\omega ))=(\mathrm{\Delta }(\omega )II)(\mathrm{\Phi }_{13}\mathrm{\Phi }_{24})=(\omega II)(\mathrm{\Phi }_{12}\mathrm{\Phi }_{13})=$$
$$(\omega II)(\mathrm{\Phi }_{23}\mathrm{\Phi }_{12}\mathrm{\Phi }_{23}^1)=\mathrm{\Delta }_l\lambda (\omega ).$$
Analogously, for $`\mathrm{\Delta }_r\rho =(\rho \rho )\mathrm{\Delta }_r`$:
$$(\rho \rho )(\mathrm{\Delta }(\omega ))=(II\mathrm{\Delta }_r(\omega ))(\mathrm{\Phi }_{13}\mathrm{\Phi }_{24})=(II\omega )(\mathrm{\Phi }_{13}\mathrm{\Phi }_{23})=$$
$$(II\omega )(\mathrm{\Phi }_{12}^1\mathrm{\Phi }_{23}\mathrm{\Phi }_{12})=\mathrm{\Delta }_r\rho (\omega ).$$
$`\mathrm{}`$
In particular, $`im(\lambda ),im(\rho )`$ are a bialgebras and subalgebras of $`End(M)`$. By the definition $`\mathrm{\Phi }im(\rho )im(\lambda )End(M)`$ and formula
$$(\rho (\omega ),\lambda (\omega ^{}))=(\omega ^{}\omega )(\mathrm{\Phi })$$
defines non-degenerated pairing $`(,):im(\rho )im(\lambda )k`$.
###### Lemma 5.2
The pairing $`(,)`$ is a pairing of bialgebras, e.g.
$$(\mathrm{\Delta }_r(\rho (\omega )),\lambda (\omega ^{})\lambda (\omega ^{\prime \prime }))=(\rho (\omega ),\lambda (\omega ^{})\lambda (\omega ^{\prime \prime }))$$
and
$$(\rho (\omega )\rho (\omega ^{}),\mathrm{\Delta }_l(\lambda (\omega ^{\prime \prime }))=(\rho (\omega )\rho (\omega ^{}),\lambda (\omega ^{\prime \prime })).$$
Proof:
The verification is direct:
$$(\mathrm{\Delta }_r(\rho (\omega )),\lambda (\omega ^{})\lambda (\omega ^{\prime \prime }))=((\rho \rho )(\mathrm{\Delta }(\omega )),\lambda (\omega ^{})\lambda (\omega ^{\prime \prime }))=$$
$$(\mathrm{\Delta }(\omega )\omega ^{}\omega ^{\prime \prime })(\mathrm{\Phi }_{13}\mathrm{\Phi }_{24})=(\omega \omega ^{}\omega ^{\prime \prime })(\mathrm{\Phi }_{12}\mathrm{\Phi }_{13})=(\omega \omega ^{}\omega ^{\prime \prime })(\mathrm{\Phi }_{23}\mathrm{\Phi }_{12}\mathrm{\Phi }_{23}^1)=$$
$$(\omega \omega ^{}_l\omega ^{\prime \prime })(\mathrm{\Phi })=(\rho (\omega ),\lambda (\omega ^{}_l\omega ^{\prime \prime }))=(\rho (\omega ),\lambda (\omega ^{})\lambda (\omega ^{\prime \prime }))$$
and
$$(\rho (\omega )\rho (\omega ^{}),\mathrm{\Delta }_l(\lambda (\omega ^{\prime \prime })))=(\rho (\omega )\rho (\omega ^{}),(\lambda \lambda )(\mathrm{\Delta }(\omega ^{\prime \prime })))=$$
$$(\omega \omega ^{}\mathrm{\Delta }(\omega ^{\prime \prime }))(\mathrm{\Phi }_{13}\mathrm{\Phi }_{24})=(\omega \omega ^{}\omega ^{\prime \prime })(\mathrm{\Phi }_{13}\mathrm{\Phi }_{23})=(\omega \omega ^{}\omega ^{\prime \prime })(\mathrm{\Phi }_{12}^1\mathrm{\Phi }_{12}\mathrm{\Phi }_{12})=$$
$$(\omega _r\omega ^{}\omega ^{\prime \prime })(\mathrm{\Phi })=(\rho (\omega _r\omega ^{}),\lambda (\omega ^{\prime \prime }))=(\rho (\omega )\rho (\omega ^{}),\lambda (\omega ^{\prime \prime })).$$
$`\mathrm{}`$
###### Lemma 5.3
Subalgebras $`im(\lambda ),im(\rho )End(M)`$ are unital, e.g. endomorphism $`I`$ belongs to $`im(\rho )`$ and $`im(\lambda )`$.
Proof:
Here we use the fact that $`M`$ is finite dimensional vector space.
Let $`\chi _\mathrm{\Phi }(t)k[t]`$ be the characteristic polynomial of $`\mathrm{\Phi }End(M^2)`$. Since $`\mathrm{\Phi }`$ is invertible $`\chi _\mathrm{\Phi }(0)0`$. By Hamilton-Cayleyโs theorem $`\chi _\mathrm{\Phi }(\mathrm{\Phi })=0`$. Hence we can represent identity as a linear combinations of powers of $`\mathrm{\Phi }`$. Namely, $`I=f(\mathrm{\Phi })im(\rho )im(\lambda )`$ where $`f(t)=det(\mathrm{\Phi })^1\chi _\mathrm{\Phi }(t)1`$. $`\mathrm{}`$
Now we construct a counite for the bialgebra $`im(\lambda )`$ using duality between $`im(\lambda )`$ and $`im(\rho )`$.
###### Proposition 5.4
Let $`\epsilon \rho ^1(I)`$, that is $`\rho (\epsilon )=I`$. Then the restriction of $`\epsilon `$ to $`im(\lambda )`$ is a homomorphism of algebras $`\epsilon :im(\lambda )k`$.
This homomorphism is a counite with respect to the coproduct $`\mathrm{\Delta }_l`$, that is
$$(\epsilon I)\mathrm{\Delta }_l=(I\epsilon )\mathrm{\Delta }_l=I.$$
Proof:
Note that for any $`\omega End(M)^{}`$
$$\epsilon (\lambda (\omega ))=(\omega \epsilon )(\mathrm{\Phi })=\omega (I)$$
and for any $`\omega ,\omega ^{}End(M)^{}`$
$$(\omega _r\omega ^{})(I)=(\omega \omega ^{})(\mathrm{\Phi }^1(II)\mathrm{\Phi })=(\omega \omega ^{})(II)=\omega (I)\omega ^{}(I).$$
Now
$$\epsilon (\lambda (\omega )\lambda (\omega ^{}))=\epsilon (\lambda (\omega _r\omega ^{}))=(\omega _r\omega ^{})(I)=\omega (I)\omega ^{}(I)=\epsilon (\lambda (\omega ))\epsilon (\lambda (\omega ^{})).$$
The counite axiom can be verified directly:
$$(\epsilon I)\mathrm{\Delta }(\lambda (\omega ))=(\epsilon I)(\lambda \lambda )(\mathrm{\Delta }(\omega ))=(\omega \epsilon I)(\mathrm{\Phi }_{12}\mathrm{\Phi }_{13})=(\omega I)(\mathrm{\Phi })=\lambda (\omega ).$$
Analogously,
$$(I\epsilon )\mathrm{\Delta }(\lambda (\omega ))=(I\epsilon )(\lambda \lambda )(\mathrm{\Delta }(\omega ))=(\omega I\epsilon )(\mathrm{\Phi }_{12}\mathrm{\Phi }_{13})=(\omega I)(\mathrm{\Phi })=\lambda (\omega ).$$
$`\mathrm{}`$
It follows by the uniqueness of a counite that all $`\epsilon \rho ^1(I)`$ give the same restriction to $`im(\lambda )`$.
The next step is to define the antipode of the bialgebra $`im(\lambda )`$.
###### Proposition 5.5
The map $`S:im(\lambda )im(\lambda )`$ given by $`S(\lambda (\omega ))=\sigma (\omega )=(\omega I)(\mathrm{\Phi }^1)`$ is an antipode for the bialgebra $`im(\lambda )`$.
Proof:
First we need to check the correctness of the definition. Note that $`\omega (im(\rho ))=0`$ for $`\omega ker(\lambda )`$. Indeed,
$$\omega (\rho (\omega ^{}))=(\omega \omega ^{})(\mathrm{\Phi })=0.$$
Since $`\mathrm{\Phi }^1`$ belongs to $`im(\rho )im(\lambda )`$ we have that $`(\omega I)(\mathrm{\Phi }^1)=0`$ for any $`\omega ker(\lambda )`$. This proves that the assertion $`S(\lambda (\omega ))=(\omega I)(\mathrm{\Phi }^1)`$ defines a map from $`im(\lambda )`$. To check that this is a map to $`im(\lambda )`$ is enough to note that since $`\mathrm{\Phi }^1im(\rho )im(\lambda )`$ any expression of the form $`(\omega I)(\mathrm{\Phi }^1)`$ lies in $`im(\lambda )`$.
Now we are ready to verify antipode axioms:
$$(IS)\mathrm{\Delta }_l(\lambda (\omega ))=(IS)(\lambda \lambda )(\mathrm{\Delta }(\omega ))=(\lambda \sigma )(\mathrm{\Delta }(\omega ))=$$
$$(\mathrm{\Delta }(\omega )II)(\mathrm{\Phi }_{13}\mathrm{\Phi }_{24}^1)=(\omega II)(\mathrm{\Phi }_{12}\mathrm{\Phi }_{13}^1)$$
so
$$\mu (IS)\mathrm{\Delta }_l(\lambda (\omega ))=(\omega I)(\mathrm{\Phi }_{12}\mathrm{\Phi }_{12}^1)=\omega (I)I.$$
Analogously,
$$(SI)\mathrm{\Delta }_l(\lambda (\omega ))=(SI)(\lambda \lambda )(\mathrm{\Delta }(\omega ))=(\sigma \lambda )(\mathrm{\Delta }(\omega ))=$$
$$(\mathrm{\Delta }(\omega )II)(\mathrm{\Phi }_{13}^1\mathrm{\Phi }_{24})=(\omega II)(\mathrm{\Phi }_{12}^1\mathrm{\Phi }_{13})$$
so
$$\mu (SI)\mathrm{\Delta }_l(\lambda (\omega ))=(\omega I)(\mathrm{\Phi }_{12}^1\mathrm{\Phi }_{12})=\omega (I)I.$$
$`\mathrm{}`$
We have proven that $`H=im(\lambda )`$ is a Hopf algebra. By the construction $`M`$ is a left module over $`H`$ and over dual Hopf algebra $`H^{}im(\rho )`$. In the next proposition we construct $`H`$-comodule structure on $`M`$ which corresponds to $`H^{}`$-module structure and prove that this structure is $`H`$-linear, so that $`M`$ is a Hopf $`H`$-module.
###### Proposition 5.6
The formula $`r_\mathrm{\Phi }(m)=\mathrm{\Phi }(mI)`$ defines right $`H`$-comodule structure on $`M`$.
This comodule structure satisfies to the equation:
$$r_\mathrm{\Phi }(\lambda (\omega )m)=\mathrm{\Delta }_l(\lambda (\omega ))r_\mathrm{\Phi }(m).$$
Proof:
Since $`\mathrm{\Phi }`$ lies in $`im(\rho )im(\lambda )`$ then $`\mathrm{\Phi }(vI)`$ belongs to $`Mim(\lambda )`$. Coassociativity of comodule structure is a consequence of pentagon equation:
$$(r_\mathrm{\Phi }I)r_\mathrm{\Phi }(m)=(\mathrm{\Phi }_{12}\mathrm{\Phi }_{13})(mII)=(\mathrm{\Phi }_{23}\mathrm{\Phi }_{12}\mathrm{\Phi }_{23}^1)(mII)=(I\mathrm{\Delta }_H)r_\mathrm{\Phi }(m).$$
Transforming left side of $`H`$-linearity equation:
$$r_\mathrm{\Phi }(\lambda (\omega )m)=(\omega II)(\mathrm{\Phi }_{23}\mathrm{\Phi }_{12})(ImI)=(\omega II)(\mathrm{\Phi }_{12}\mathrm{\Phi }_{13}\mathrm{\Phi }_{23})(ImI)$$
we can see that it coincides with $`\mathrm{\Delta }_l(\lambda (\omega ))r_\mathrm{\Phi }(m)`$. $`\mathrm{}`$
Now we are ready to prove the main theorem of this section (see theorem 4.10 from ) which says that up to multiplication by a vector space any solution of pentagon equation corresponds to some Hopf algebra.
###### Theorem 5.7
The solution to the pentagon equation $`\mathrm{\Phi }_M`$ corresponding to the Hopf $`H`$-module structure on $`M`$ coincides with $`\mathrm{\Phi }`$.
In particular, the multiplication map $`HM_HM`$ is an isomorphism and identifies $`\mathrm{\Phi }`$ with $`\mathrm{\Phi }_HM_H`$, where $`M_H=\{mM,\mathrm{\Phi }(mn)=mnnM\}`$.
Proof:
By the definition of Hopf $`H`$-module structure on $`M`$:
$$\mathrm{\Phi }_M(mn)=(I\mu _M)(\mathrm{\Delta }_MI)(mn)=\mathrm{\Phi }(mn).$$
Now by the corollary 3.4 the multiplication map $`HM_HM`$ is an isomorphism and identifies $`\mathrm{\Phi }`$ with $`\mathrm{\Phi }_HM_H`$, where $`M_H`$ is a space of coinvariants of the comodule structure $`\mathrm{\Delta }_M`$. Using the definition of $`\mathrm{\Delta }_M`$ we can see that $`M_H`$ coincides with the space $`\{mM,\mathrm{\Phi }(mn)=mnnM\}`$. $`\mathrm{}`$
## 6 Modified pentagon equation
In this section we apply methods of the previous section to solutions of modified pentagon equation.
###### Proposition 6.1
Let $`(F,\mathrm{\Phi })`$ be a solution of modified pentagon equation on $`(V,M)`$. Then the map $`\mathrm{\Delta }_{F,\mathrm{\Phi }}:Hom(M,V)Hom(M,V)Hom(M,V)`$ defined by
$$\mathrm{\Delta }_{F,\mathrm{\Phi }}(x)=F(x1)\mathrm{\Phi }^1$$
is a coassociative comultiplication.
The map
$$\lambda _F:(End(V)^{},\mathrm{\Delta })(Hom(M,V),\mathrm{\Delta }_{F,\mathrm{\Phi }})$$
is a homomorphism of coalgebras and the map
$$\rho _F:(Hom(M,V)^{},\mu _{F,\mathrm{\Phi }})End(V),\rho _F(\omega )=(I\omega )(F)$$
is a homomorphism of algebras, where $`\mu _{F,\mathrm{\Phi }}`$ is a multiplication on the dual space $`Hom(M,V)^{}`$ defined by $`\mathrm{\Delta }_{F,\mathrm{\Phi }}`$.
Proof:
The coassociativity of $`\mathrm{\Delta }_{F,\mathrm{\Phi }}`$ is a direct consequence of modified pentagon equation for $`(F,\mathrm{\Phi })`$ and pentagon equation for $`\mathrm{\Phi }`$ which being combined together state that
$$F_{12}^1F_{23}^1F_{12}F_{13}=\mathrm{\Phi }_{23}^1=\mathrm{\Phi }_{12}^1\mathrm{\Phi }_{23}^1\mathrm{\Phi }_{12}\mathrm{\Phi }_{13}.$$
This implies the coincidence between
$$(\mathrm{\Delta }_{F,\mathrm{\Phi }}I)\mathrm{\Delta }_{F,\mathrm{\Phi }}(x)=F_{12}F_{13}(x11)\mathrm{\Phi }_{13}^1\mathrm{\Phi }_{12}^1$$
and
$$(I\mathrm{\Delta }_{F,\mathrm{\Phi }})\mathrm{\Delta }_{F,\mathrm{\Phi }}(x)=F_{23}F_{12}(x11)\mathrm{\Phi }_{12}^1\mathrm{\Phi }_{23}^1.$$
The homomorphism properties for $`\lambda _F,\rho _F`$ also follow from the modified pentagon equation
$$(\lambda _F\lambda _F)(\mathrm{\Delta }(\omega ))=(\mathrm{\Delta }(\omega )II)(F_{13}F_{24})=(\omega II)(F_{12}F_{13})=$$
$$(\omega II)(F_{23}F_{12}\mathrm{\Phi }_{23}^1)=\mathrm{\Delta }_{F,\mathrm{\Phi }}\lambda (\omega ).$$
and
$$\rho _F(\omega )\rho _F(\omega ^{})=(I\omega \omega ^{})(F_{12}F_{13})=$$
$$(I\omega \omega ^{})(F_{23}F_{12}\mathrm{\Phi }_{23}^1)=\rho _F(\omega _{F,\mathrm{\Phi }}\omega ^{}).$$
$`\mathrm{}`$
In particular, $`im(\lambda _F)`$ is a coalgebra and $`im(\rho _F)`$ is an algebra. In the next proposition we prove that $`im(\lambda _F)`$ is an $`im(\lambda _\mathrm{\Phi })`$-module coalgebra and $`im(\rho _F)`$ is an $`im(\lambda _\mathrm{\Phi })`$-comodule algebra.
###### Proposition 6.2
The subspace $`im(\lambda _F)Hom(M,V)`$ is closed under right multiplication by elements from $`im(\lambda _\mathrm{\Phi })End(M)`$. Moreover, this $`im(\lambda _\mathrm{\Phi })`$-module structure on $`im(\lambda _F)`$ is compatible with comultiplication:
$$\mathrm{\Delta }_{F,\mathrm{\Phi }}(\lambda _F(\omega )\lambda _\mathrm{\Phi }(\omega ^{}))=\mathrm{\Delta }_{F,\mathrm{\Phi }}(\lambda _F(\omega ))\mathrm{\Delta }_\mathrm{\Phi }(\lambda _\mathrm{\Phi }(\omega ^{})).$$
Dually, comultiplication $`\mathrm{\Delta }_F`$ sends $`im(\rho _F)`$ into $`im(\rho _F)im(\rho _\mathrm{\Phi })`$ which gives a right $`im(\lambda _\mathrm{\Phi })`$-comodule structure on the algebra $`im(\rho _F)`$.
Proof:
$$\lambda _F(\omega )\lambda _\mathrm{\Phi }(\omega ^{})=(\omega \omega ^{}I)(F_{13}\mathrm{\Phi }_{23})=(\omega \omega ^{}I)(F_{12}^1F_{23}F_{12})=\lambda (\omega _{\overline{F}}\omega ^{}).$$
Compatibility with comultiplication follows from the equality
$$\mathrm{\Delta }_{F,\mathrm{\Phi }}(xy)=\mathrm{\Delta }_{F,\mathrm{\Phi }}(x)\mathrm{\Delta }_\mathrm{\Phi }(y),xHom(M,V),yEnd(M).$$
It follows from the equality
$$(\rho _F\rho _\mathrm{\Phi })(\mathrm{\Delta }(\omega ))=(II\mathrm{\Delta }(\omega ))(F_{13}\mathrm{\Phi }_{24})=(II\omega )(F_{13}\mathrm{\Phi }_{23})=$$
$$(II\omega )(F_{12}^1F_{23}F_{12})=\mathrm{\Delta }_F\rho _F(\omega )$$
that $`\mathrm{\Delta }_F(im(\rho _F))im(\rho _F)im(\rho _\mathrm{\Phi })`$. Hence algebra homomorphism $`\mathrm{\Delta }_F`$ defines $`im(\lambda _\mathrm{\Phi })`$-comodule structure. $`\mathrm{}`$
In the next proposition we construct $`im(\lambda _F)`$-comodule structure on $`V`$ which is dual to $`im(\rho _F)`$-module structure.
###### Proposition 6.3
The formula $`r_F(v)=F(vI)Vim(\lambda _F)`$ defines right $`im(\lambda _F)`$-comodule structure on $`V`$.
Proof:
Since $`F`$ lies in $`im(\rho _F)im(\lambda _F)`$ then $`F(vI)`$ belongs to $`Vim(\lambda _F)`$. Coassociativity of comodule structure is a consequence of pentagon equation:
$$(r_FI)r_F(v)=(F_{12}F_{13})(vII)=(F_{23}F_{12}\mathrm{\Phi }_{23}^1)(vII)=(I\mathrm{\Delta }_{F,\mathrm{\Phi }})r_F(v).$$
$`\mathrm{}`$
Since $`im(\lambda _F)`$ is a subspace of $`Hom(M,V)`$ we have a pairing
$$\mu _F:im(\lambda _F)MV,\lambda _F(\omega )m\lambda _F(\omega )m.$$
By the definition it has the following associativity property:
$$\lambda _F(\omega )(\lambda _\mathrm{\Phi }(\omega )m)=(\lambda _F(\omega )\lambda _\mathrm{\Phi }(\omega ))m,\omega End(V)^{},mM.$$
Next proposition shows compatibility of this pairing with the $`im(\lambda _\mathrm{\Phi })`$-comodule structure on $`M`$ and the $`im(\lambda _F)`$-comodule structure on $`V`$.
###### Proposition 6.4
The $`im(\lambda _\mathrm{\Phi })`$-comodule structure on $`M`$ and the $`im(\lambda _F)`$-comodule structure on $`V`$ are compatible in the following way
$$r_F(\lambda _F(\omega )m)=\mathrm{\Delta }_{F,\mathrm{\Phi }}(\lambda _F(\omega ))r_\mathrm{\Phi }(m),\omega End(V)^{},mM.$$
Proof:
Indeed,
$$r_F(\lambda _F(\omega )m)=(\omega II)(F_{23}F_{12})(ImI),$$
which by modified pentagon equation coincides with
$$(\omega II)(F_{12}F_{13}\mathrm{\Phi }_{23})(ImI)=(\mathrm{\Delta }(\omega )II)(F_{13}F_{24}\mathrm{\Phi }_{34})(ImI)=$$
$$(\lambda _F\lambda _F)(\mathrm{\Delta }(\omega ))r_\mathrm{\Phi }(m)=\mathrm{\Delta }_{F,\mathrm{\Phi }}(\lambda _F(\omega ))r_\mathrm{\Phi }(m).$$
$`\mathrm{}`$
Now we have all necessary properties for the pair $`(V,M)`$ and we are ready to prove main theorem of this section.
###### Theorem 6.5
The solution to the modified pentagon equation $`F_{(V,M)}`$ corresponding to the pair $`(V,M)`$ (see section 4) coincides with $`F`$.
Proof:
By the definition of $`F_{(V,M)}`$:
$$F_{(V,M)}(vm)=(I\mu _F)(r_FI)(vm)=F(vm).\mathrm{}$$
## 7 Categorical reformulation
We start with description of $`k`$-linear monoidal structures on the category of (finite dimensional) vector spaces $`๐ฑ\mathrm{๐๐๐ก}_k`$ over $`k`$. The standard reference for the theory of monoidal categories and Tannaka-Krein theory is . For the sake of completeness we give some basic definitions of the theory. A monoidal category is a category $`๐ข`$ with a bifunctor
$$:๐ข\times ๐ข๐ข(X,Y)XY$$
which is called tensor (or monoidal) product. This functor is supposed to be equipped with a functorial collection of isomorphisms (so-called associativity constraint)
$$\phi _{X,Y,Z}:X(YZ)(XY)ZX,Y,Z๐ข$$
satisfying the so-called pentagon axiom:
the diagram
$$\begin{array}{ccccc}X(Y(ZW))& \stackrel{\phi _{X,Y,ZW}}{}& (XY)(ZW)& \stackrel{\phi _{XY,Z,W}}{}& ((XY)Z)W\\ I_X\phi _{Y,Z,W}& & & & \phi _{X,Y,Z}I_W\\ X((YZ)W)& & \stackrel{\phi _{X,YZ,W}}{}& & (X(YZ))W\end{array}$$
(6)
is commutative for any objects $`X,Y,Z,W๐ข`$.
By the functoriality any tensor product functor which is just a $`k`$-bi-linear functor $`:๐ฑ\mathrm{๐๐๐ก}_k\times ๐ฑ\mathrm{๐๐๐ก}_k๐ฑ\mathrm{๐๐๐ก}_k`$ is defined by the vector space $`kk=M`$ (multiplicity space). Denote by $`_M`$ the tensor product functor corresponding to a given vector space $`M`$. We have $`U_1_MU_2=U_1MU_2`$ for arbitrary vector spaces $`U_i`$.
Now we establish the relation between associativity constraint for the tensor product $`_M`$ and pentagon equation.
###### Proposition 7.1
There is one-to-one correspondence between structures of monoidal category on the category of finite dimensional vector spaces and solution of pentagon equation.
Proof:
By functoriality any associativity constraint $`\phi `$ for $`_M`$ is given by the automorphism $`\mathrm{\Phi }Aut(M^2)`$:
$$MM=k_MM=k_M(k_Mk)\stackrel{\phi _{k,k,k}}{}(k_Mk)_Mk=M_Mk=MM$$
and has the following general form
$$\begin{array}{ccc}U_1_M(U_2_MU_3)& \stackrel{\phi _{U_1,U_2,U_3}}{}& (U_1_MU_2)_MU_3\\ ||& & ||\\ U_1_M(U_2MU_3)& & (U_1MU_2)_MU_3\\ ||& & ||\\ U_1MU_2MU_3& \stackrel{\mathrm{\Phi }_{24}}{}& U_1MU_2MU_3\end{array}$$
In particular,
$$\phi _{k_Mk,k,k}=\phi _{M,k,k}=\mathrm{\Phi }_{23},$$
$$\phi _{k,k_Mk,k}=\phi _{k,M,k}=\mathrm{\Phi }_{13},$$
$$\phi _{k,k,k_Mk}=\phi _{k,k,M}=\mathrm{\Phi }_{12}.$$
Combining it with
$$I_k_M\phi _{k,k,k}=\mathrm{\Phi }_{23},\phi _{k,k,k}_MI_k=\mathrm{\Phi }_{12}$$
we see that pentagon axiom (6) for $`\phi `$ is equivalent to the pentagon equation for $`\mathrm{\Phi }`$
$$\mathrm{\Phi }_{12}\mathrm{\Phi }_{13}\mathrm{\Phi }_{23}=\mathrm{\Phi }_{23}\mathrm{\Phi }_{12}.\mathrm{}$$
For a solution of pentagon equation $`\mathrm{\Phi }Aut(M^2)`$ denote by $`๐ข_\mathrm{\Phi }`$ the corresponding monoidal category structure on the category of finite dimensional vector spaces.
Now we discuss the relation between $`k`$-linear monoidal functors from the category $`๐ข_\mathrm{\Phi }`$ to vector spaces $`๐ฑ\mathrm{๐๐๐ก}_k`$ and modified pentagon equation. A monoidal functor between monoidal categories $`๐ข`$ and $``$ is a functor $`\omega :๐ข`$ , which is equipped with a functorial collection of isomorphisms (the so-called monoidal structure)
$$\omega _{X,Y}:\omega (XY)\omega (X)\omega (Y)X,Y๐ข,$$
for which the following diagram is commutative for any objects $`X,Y,Z๐ข`$
$$\begin{array}{ccccc}F(X(YZ))& \stackrel{\omega _{X,YZ}}{}& \omega (X)\omega (YZ)& \stackrel{I\omega _{Y,Z}}{}& \omega (X)(\omega (Y)\omega (Z))\\ \omega \left(\phi _{X,Y,Z}\right)& & & & \phi _{\omega \left(X\right),\omega \left(Y\right),\omega \left(Z\right)}\\ \omega ((XY)Z)& \stackrel{\omega _{XY,Z}}{}& \omega (XY)\omega (Z)& \stackrel{\omega _{X,Y}I}{}& (\omega (X)\omega (Y))\omega (Z)\end{array}$$
(7)
###### Proposition 7.2
There is one-to-one correspondence between $`k`$-linear monoidal functors $`๐ข_\mathrm{\Phi }๐ฑ\mathrm{๐๐๐ก}_k`$ and vector spaces $`V`$ with isomorphisms $`F:VMVV`$ satisfying modified pentagon equation $`F_{12}F_{13}\mathrm{\Phi }_{23}=F_{23}F_{12}.`$
Proof:
By functoriality any $`k`$-linear functor $`\omega `$ is defined by its evaluation $`\omega (k)=V`$. For arbitrary vector space $`U`$ it gives $`\omega (U)=VU`$. Monoidal structure of such a functor is given by the isomorphism $`F:VMVV`$:
$$VM=\omega (M)=\omega (k_Mk)\stackrel{\omega _{k,k}}{}\omega (k)\omega (k)=VV$$
and has the following general form
$$\begin{array}{ccc}\omega (U_1_MU_2)& \stackrel{\omega _{U_1,U_2}}{}& \omega (U_1)\omega (U_2)\\ ||& & ||\\ \omega (U_1MU_2)& & \omega (U_1)\omega (U_2)\\ ||& & ||\\ WU_1MU_2& \stackrel{F_{13}}{}& WU_1WU_2\end{array}$$
In particular,
$$\omega _{k_Mk,k}=\omega _{M,k}=F_{13},\omega _{k,k_Mk}=\omega _{k,M}=F_{12}.$$
Combining it with
$$I_{\omega (k)}\omega _{k,k}=I_VF=\omega _{23},$$
$$\omega _{k,k}I_{\omega (k)}=FI_V=F_{12}$$
and with
$$\omega (\varphi _{k,k,k})=\omega (\mathrm{\Phi })=\mathrm{\Phi }_{23}$$
we can see that compatibility axiom for monoidal structure (7) is equivalent to the modified pentagon equation for $`(F,\mathrm{\Phi })`$
$$F_{12}F_{13}\mathrm{\Phi }_{23}=F_{23}F_{12}.\mathrm{}$$
Denote by $`\omega _F`$ the functor $`๐ข_\mathrm{\Phi }๐ฑ\mathrm{๐๐๐ก}_k`$. General Tannaka-Krein theory predicts existence of bialgebra structure on the algebra of natural transformations $`End(\omega )`$ of any monoidal functor $`\omega `$ from a semisimple category to the category of vector spaces. It is not hard to verify that $`End(\omega _F)`$ coincides with the bialgebra $`(End(V),\mathrm{\Delta }_F)`$ defined in the first section.
In general, we say that the pair $`(V,F)`$ consisting of an object $`V`$ of some $`k`$-linear monoidal category $`๐ข`$ and an isomorphism $`v:VMVV`$ (for some vector space $`M`$) is an idempotent object in $`๐ข`$ if $`v`$ satisfies to modified pentagon equation
$$\begin{array}{ccccc}VMM& \stackrel{vI}{}& VVM& \stackrel{Iv}{}& VVV\\ I\mathrm{\Phi }& & & & \\ VMM& & & & vI\\ c_{V,M}I& & & & \\ MVM& \stackrel{Iv}{}& MVV& \stackrel{c_{M,V}I}{}& VMV\end{array}$$
or $`v_{12}v_{13}\mathrm{\Phi }_{23}=v_{23}v_{12}`$ for some automorphism $`\mathrm{\Phi }Aut(M^2)`$. For simplicity we omit brackets and associativity isomorphism in category $`๐ข`$. Then automatically (see proposition 2.3) $`\mathrm{\Phi }`$ is a solution to pentagon equation.
The examples of such objects are Hopf modules (see section 3 for definition). To clarify this we need the categorical characterization of Hopf modules.
Let $`๐ข`$ be a $`k`$-linear monoidal category and $`\omega :๐ข๐ฑ\mathrm{๐๐๐ก}_k`$ be a $`k`$-linear monoidal functor. We call the object $`E๐`$ integral with respect to $`\omega `$ if it is equipped with functorial collection of isomorphisms $`e_X:E\omega (X)EX`$ such that $`e_1=I`$ and
$$\begin{array}{ccc}E\omega (XY)& \stackrel{e_{XY}}{}& EXY\\ I\omega _{X,Y}& & e_XI\\ E\omega (X)\omega (Y)& & E\omega (X)Y\\ Ic_{\omega \left(X\right),\omega \left(Y\right)}& & c_{Y,\omega \left(X\right)}I\\ E\omega (Y)\omega (X)& \stackrel{e_YI}{}& EY\omega (X)\end{array}$$
(8)
Now we establish the relation between integral and idempotent objects.
###### Lemma 7.3
For any integral object $`(E,e)`$ the pair $`(E,e_E)`$ is an idempotent object with $`M=\omega (E)`$ and $`\mathrm{\Phi }=\omega (e_E)`$.
Proof:
The modified pentagon equation can be glued from two diagrams:
$$\begin{array}{ccc}E\omega (E)\omega (E)& \stackrel{e_E\omega (E)}{}& EE\omega (E)\\ I\omega _{E,\omega \left(E\right)}& & ||\\ E\omega (E\omega (E))& \stackrel{e_{E\omega (E)}}{}& EE\omega (E)\\ I\omega \left(e_E\right)& & Ie_E\\ E\omega (EE)& \stackrel{e_{EE}}{}& EEE\\ I\omega _{E,E}& & e_EI\\ E\omega (E)\omega (E)& & E\omega (E)E\\ Ic_{\omega \left(E\right),\omega \left(E\right)}& & c_{E,\omega \left(E\right)}I\\ E\omega (E)\omega (E)& \stackrel{e_EI}{}& EE\omega (E)\end{array}$$
One (bottom part) is the diagram 8 for $`X=Y=E`$ and the other (top part) is a functoriality diagram for collection $`e`$. To complete the proof it is enough to note that composition of three top left vertical arrows in the glued diagram coincides with $`I\mathrm{\Phi }`$. $`\mathrm{}`$
In particular, any integral object $`(E,e)`$ in the category $`๐ข`$ defines solution to pentagon equation $`\mathrm{\Phi }=\omega (e_E)`$ and monoidal functor $`๐ข_\mathrm{\Phi }๐ข`$ sending $`k`$ to $`E`$.
Now we describe integral objects in categories of modules over Hopf algebras. Denote by $`(H)`$ the category of modules over Hopf algebra $`H`$. The forgetful functor $`(H)๐ฑ\mathrm{๐๐๐ก}_k`$ has natural monoidal structure.
###### Proposition 7.4
Integral object in the category of modules over Hopf algebra with respect to forgetful functor are Hopf modules.
Sketch of the proof:
Let $`V`$ be a Hopf module. It can be checked directly that for any (left) $`H`$-module $`X`$ the map $`V\omega (X)VX`$ defined by $`vx_{(v)}v_{(0)}v_{(1)}x`$ is a structure of integral object.
Conversely, if $`(E,e)`$ is an integral object in $`(H)`$ then the composition
$$E\stackrel{Ii}{}E\omega (H)\stackrel{e_H}{}EH$$
is an $`H`$-linear comodule structure (Hopf module structure). Here we consider Hopf algebra $`H`$ as a module over itself; $`i:k\omega (H)`$ denotes the unital inclusion $`cc1`$. $`\mathrm{}`$
Combining the lemma and proposition above we can see that any Hopf module $`V`$ defines a solution to pentagon equation $`\mathrm{\Phi }`$ on $`M=\omega (V)`$ (which is of course one defined in section 3) and a monoidal functor $`๐ข_\mathrm{\Phi }(H)`$ sending $`k`$ to $`V`$. The categorical reformulation of theorem 5.7 says that for any solution $`\mathrm{\Phi }`$ of pentagon equation there is a monoidal functor from $`๐ข_\mathrm{\Phi }`$ to the category of modules $`(H)`$ over some Hopf algebra $`H`$ sending $`k`$ to some Hopf module $`V`$. We outline the direct proof of this fact. First of all we need to define the category $`(H)`$ in terms of $`\mathrm{\Phi }`$.
Following we call the pair $`(X,\mathrm{\Psi })`$ of a vector space $`X`$ and an isomorphism $`v:MXMX`$ a module over the solution to pentagon equation $`(M,\mathrm{\Phi })`$ if the following equality for operators on $`MMX`$ holds:
$$\mathrm{\Phi }_{12}\mathrm{\Psi }_{13}\mathrm{\Psi }_{23}=\mathrm{\Psi }_{23}\mathrm{\Phi }_{12}.$$
(9)
For example, the pair $`(M,\mathrm{\Phi })`$ is a module over itself and the pair $`(X,I_{MX})`$ is a module over $`(M,\mathrm{\Phi })`$ for any vector space $`X`$.
Note that if we are given by Hopf $`H`$-module structure on $`M`$ defining $`\mathrm{\Phi }`$ then any $`H`$-module becomes an $`(M,\mathrm{\Phi })`$-module by the formula $`\mathrm{\Psi }(mx)=_{(m)}m_{(0)}m_{(1)}x`$.
A morphism of $`(M,\mathrm{\Phi })`$-modules $`(X,\mathrm{\Psi })(X^{},\mathrm{\Psi }^{})`$ is a linear map $`f:XX^{}`$ such that
$$\mathrm{\Psi }^{}(If)=(If)\mathrm{\Psi }.$$
(10)
It was shown in (in much more general setting) that the category $`(M,\mathrm{\Phi })`$ of modules over $`(M,\mathrm{\Phi })`$ forms rigid monoidal category with tensor product defined by
$$(X,\mathrm{\Psi })(X^{},\mathrm{\Psi }^{})=(XX^{},\mathrm{\Psi }_{12}\mathrm{\Psi }_{}^{}{}_{13}{}^{}).$$
Forgetting of module structure gives a $`k`$-linear monoidal functor $`(M,\mathrm{\Phi })๐ฑ\mathrm{๐๐๐ก}_k`$. Tannaka-Krein theory decomposes this forgetful functor into an equivalence $`(M,\mathrm{\Phi })(H)`$ and a forgetful functor $`(H)๐ฑ\mathrm{๐๐๐ก}_k`$ for category $`(H)`$ of (finite-dimensional over $`k`$) modules over some Hopf algebra $`H`$. It is not hard to check that this Hopf algebra coincides with one constructed in section 5.
Now we concentrate on the role of Hopf bimodules in the picture. We prove that $`(M,\mathrm{\Phi })`$ has a natural structure of integral object in the category $`(M,\mathrm{\Phi })`$.
###### Proposition 7.5
For any $`(M,\mathrm{\Phi })`$-module $`(X,\mathrm{\Psi })`$ structural map $`\mathrm{\Psi }`$ defines the isomorphism of $`(M,\mathrm{\Phi })`$-modules
$$\mathrm{\Psi }:(M,\mathrm{\Phi })X(M,\mathrm{\Phi })(X,\mathrm{\Psi }).$$
In other words we have a functorial (in $`(X,\mathrm{\Psi })`$) collection of isomorphisms $`e_{(X,\mathrm{\Psi })}:\mathrm{\Psi }:(M,\mathrm{\Phi })\omega (X,\mathrm{\Psi })(M,\mathrm{\Phi })(X,\mathrm{\Psi }).`$
Moreover the diagrams of the form 8 are commutative for this collection.
Sketch of the proof:
Indeed, the structural maps of the tensor products $`(M,\mathrm{\Phi })X`$, $`(M,\mathrm{\Phi })(X,\mathrm{\Psi })`$ are $`\mathrm{\Phi }_{12}`$ and $`\mathrm{\Phi }_{12}\mathrm{\Psi }_{13}`$ correspondently. Hence we can consider the equation (9) as a homomorphism condition (10) for $`\mathrm{\Psi }`$.
The commutativity of the diagram 8 is equivalent to the decomposition of the equation
$$\mathrm{\Phi }_{12}\mathrm{\Psi }_{13}\mathrm{\Psi }_{}^{}{}_{14}{}^{}\mathrm{\Psi }_{23}\mathrm{\Psi }_{}^{}{}_{24}{}^{}=\mathrm{\Psi }_{23}\mathrm{\Psi }_{}^{}{}_{24}{}^{}\mathrm{\Phi }_{12}$$
(11)
into commutativity equation and equations of the form 9 (up to some common factors):
using $`\mathrm{\Psi }_{}^{}{}_{14}{}^{}\mathrm{\Psi }_{23}=\mathrm{\Psi }_{23}\mathrm{\Psi }_{}^{}{}_{14}{}^{}`$ we replace left side of (11) by $`\mathrm{\Phi }_{12}\mathrm{\Psi }_{13}\mathrm{\Psi }_{23}\mathrm{\Psi }_{}^{}{}_{14}{}^{}\mathrm{\Psi }_{}^{}{}_{24}{}^{}`$, then using $`\mathrm{\Phi }_{12}\mathrm{\Psi }_{13}\mathrm{\Psi }_{23}=\mathrm{\Psi }_{23}\mathrm{\Phi }_{12}`$ we rewrite the result as $`\mathrm{\Psi }_{23}\mathrm{\Phi }_{12}\mathrm{\Psi }_{}^{}{}_{14}{}^{}\mathrm{\Psi }_{}^{}{}_{24}{}^{}`$ and finally using $`\mathrm{\Phi }_{12}\mathrm{\Psi }_{}^{}{}_{13}{}^{}\mathrm{\Psi }_{}^{}{}_{23}{}^{}=\mathrm{\Psi }_{}^{}{}_{23}{}^{}\mathrm{\Phi }_{12}`$ we get the right side of (11). $`\mathrm{}`$
Combining lemma 7.3 with proposition 7.5 we get a monoidal functor from the category $`๐ข_\mathrm{\Phi }`$ to the category of modules $`(M,\mathrm{\Phi })`$ sending $`k`$ to $`(M,\mathrm{\Phi })`$. Tannaka-Krein equivalence $`(M,\mathrm{\Phi })(H)`$ identify integral object $`(M,\mathrm{\Phi })`$ with some Hopf $`H`$-module.
We finish this section reformulating connection between Hopf-Galois algebras and solutions to modified pentagon equation (and thus by proposition 7.2 monoidal functors from category $`๐ข_\mathrm{\Phi }`$ to vector spaces). It was proved in that any linear monoidal functor from category $`(H)`$ of (finite-dimensional over $`k`$) modules over some Hopf algebra $`H`$ to vector spaces corresponds to some Hopf Galois $`H`$-module coalgebra. Thus the connection between Galois coalgebras and monoidal functors from $`๐ข_\mathrm{\Phi }`$ to $`๐ฑ\mathrm{๐๐๐ก}_k`$ can be represented by commutative diagram of monoidal functors:
$$\begin{array}{ccc}๐ข(\mathrm{\Phi })& \stackrel{\omega _F}{}& ๐ฑ\mathrm{๐๐๐ก}_k\\ & & ||\\ (H)& \stackrel{\omega _L}{}& ๐ฑ\mathrm{๐๐๐ก}_k\end{array}$$
where $`\omega _L`$ is a monoidal functor corresponding to Galois coalgebra $`L`$ and $`\omega _F`$ is a monoidal functor corresponding to solution $`F`$ of modified pentagon equation which is defined by $`L`$ (see example 4.3). |
warning/0001/hep-ph0001166.html | ar5iv | text | # 1 Introduction
## 1 Introduction
Traditionally, the weakness of gravitational interactions at the scales accessible to particle physics experiments has been explained by postulating that the Planck scale at which gravity becomes strong is very high, $`M_{\text{Pl}}10^{19}`$ GeV. Below this scale, ordinary quantum field theory applies, but, when this scale is reached, one can observe the underlying quantum theory that incorporates quantum gravity. A disappointing feature of the traditional framework is that the enormously high value of the Planck scale prevents us from observing any effects of quantum gravity in laboratory experiments in the conceivable future, which means that the search for the quantum theory of gravity has to proceed without any experimental input. Recently Arkani-Hamed, Dimopoulos, and Dvali (ADD) have proposed an alternative to this pessimistic scenario. They have constructed models in which gravity becomes strong at a scale $`M`$ of order 1 TeV. They explain the apparent weakness of gravity at lower energies by the presence of compact dimensions with compactification radius $`RM^1`$. We will call these โlarge extra dimensionsโ. In this framework, gravity could have significant effects on particle interactions at the energies accessible to current experiments and observations .
So far, almost all work on the phenomenological implications of large extra dimensions has concentrated on the effects of real and virtual graviton emission. It is the basic assumption of the model that gravitons can move in the extra dimensions. Then the graviton quantum states will be characterized by a (quantized) momentum in the extra dimensions. The states with nonzero momentum are called Kaluza-Klein (KK) excitations; they can be described equivalently as massive spin-2 particles in 4 dimensions, with mass equal to the higher-dimensional momentum, which couple to Standard Model particles through a coupling to the energy-momentum tensor $`T^{\mu \nu }`$ with strength $`M_{\text{Pl}}^1`$. The sum over these states leads gravity to become strong at a scale $`MM_{\text{Pl}}`$ because the spectrum of KK excitations becomes exceedingly dense as the size $`R`$ of the compact dimensions is taken to be much larger than $`M^1`$.
Because the low-energy coupling of the KK excitations is model-independent, one can study processes in which gravitons are emitted into the extra dimensions in the context of a low-energy effective field theory. For collision energies much less than $`M`$, the cross sections for missing-energy signatures are not sensitive to the details of physics at the scale $`M`$. This fact allows one to obtain model-independent bounds on $`M`$. On the other hand, it means that the simple observation of graviton emission does not give information about the nature of the fundamental gravity theory.
The approach of low-energy effective field theory can also be applied to processes in which the KK excitations appear as virtual exchanges contributing to the scattering of Standard Model particles . In this case, the contribution of low-energy effective field theory is cutoff-dependent and of the same order as that from possible higher-dimension operators. In phenomenological analyses, the virtual KK exchange is typically represented as a dimension-8 contact interaction of the form $`T^{\mu \nu }T_{\mu \nu }`$ with a coefficient proportional to $`1/M^4`$. The precise value of this coefficient depends on the underlying model. It is also possible that this model could predict additional contact interactions with a different spin structure that could also be observed as corrections to Standard Model scattering processes. For these reasons, the virtual exchanges cannot be used to put lower bounds on $`M`$. On the other hand, the presence of high-spin contact interactions can produce impressive signals, and the measurement of the coefficients of these interactions can give new information on the fundamental theory.
The study of large extra dimensions differs from other phenomenological problems in that the underlying theory from which the low-energy effective description is derived is a theory of quantum gravity. This fact may bring in new and unforseen consequences. In particular, the only known framework that allows a self-consistent description of quantum gravity is string theory . But string theory is not simply a theory of quantum gravity. As an essential part of its structure, not only the gravitons but also the particles of the Standard Model must have an extended structure. This means that, in a string theory description, there will be additional modifications of Standard Model amplitudes due to string excitations which might compete with or even overwhelm the modifications due to graviton exchange.
In this paper, we will study the signatures of string theory in a simple toy model with large extra dimensions. The most important effects in this model come from the exchange of string Regge (SR) excitations of Standard Model particles. We will show that, in Standard Model scattering processes, contact interactions due to SR exchange produce their own characteristic effects in differential cross sections. We will also show that these typically dominate the effects due to KK exchange. In addition, the SR excitations can be directly produced as resonances. These effects have been discussed previously, but at a more qualitative level, by Lykken , and by Tye and collaborators . The effects of SR resonances have also been studied some time ago, in the context of composite models of quarks and leptons, by Bars and Hinchliffe .
The dominance of SR over KK effects is a generic feature of weakly-coupled string theory. It follows from the counting of coupling constants in string perturbation theory , which is illustrated in Figure 1. To model the ADD scenario, we consider open string theories which contain at low energy a set of Yang-Mills gauge bosons that can be identified with gauge bosons of the Standard Model. We denote the dimensionless Yang-Mills coupling by $`g`$. Figure 1(a) shows the string generalization of a Standard Model two-body scattering amplitude at order $`g^2`$. This amplitude coincides with the Standard Model expectation in the limit in which the center-of-mass energy is much lower than the string scale $`M_S`$ and, at higher energy, shows corrections proportional to powers of $`(s/M_S^2)`$. These are the effects of SR excitations. Figure 1(b) shows the leading string contribution to graviton emission. The graviton is a closed string state, and thus this process involves the closed-string coupling constant, which is of order $`g^2`$; the full amplitude is of order $`g^3`$. Figure 1(c) shows one contribution to the one-loop corrections to two-body scattering. This diagram is of order $`g^4`$. However, as Lovelace originally showed, this string diagram contains the graviton-exchange contribution when factorized as indicated in the figure. Thus, the exchange of gravitons and their KK excitations are suppressed with respect to SR exchange by a factor $`g^2`$ in the amplitude.
In this paper, we will flesh out the picture represented by Figure 1 using an illustrative toy string model. In Section 2, we will present this model, which uses scattering amplitudes on the 3-brane of weakly-coupled Type IIB string theory to describe a string version of Quantum Electrodynamics with electrons and photons. In Section 3, we will apply this model to compute the cross sections for Bhabha scattering and $`e^+e^{}\gamma \gamma `$ at high energy. In Section 4, we will discuss the phenomenological consequences of those results, both for contact interactions in high-energy scattering and for the direct observability of SR resonances. We will find a direct bound on the string scale of $`M_S>1`$ TeV. Translated into a bound on the fundamental quantum gravity scale, this becomes $`M>1.6`$ TeV. This bound is admittedly model-dependent, but it is also larger than any other current limit by more than a factor of two for the relevant case of 6 large extra dimensions.
In the remainder of the paper, we will discuss the more familiar signatures of large extra dimensions in string language. In Section 5, we will study the KK graviton emission process $`e^+e^{}\gamma G`$. In Section 6, we will discuss the effects of virtual KK graviton exchange through a detailed analysis of the process of $`\gamma \gamma `$ elastic scattering. This analysis will also allow us to derive the relation between the string scale and the fundamental quantum gravity scale. In Section 7, we will review the collider limits on large extra dimensions in the light of the new picture presented in this paper. Section 8 will present our conclusions. A series of appendices review formulae for the analysis of Bhabha scattering and present some of the more technical details of the string calculations.
A number of the topics considered in Sections 5 and 6 have recently been considered, from a slightly different point of view, in a paper of Dudas and Mourad . The phenomenological importance of SR resonances in models with a low string scale has been discussed briefly by Accomando, Antoniadis, and Benakli .
## 2 The model
In this paper, we would like to investigate the simplest model that illustrates the influence of string Regge (SR) excitations on physical cross sections. Thus, we will be content to study a simple embedding of the Quantum Electrodynamics of electrons and photons into string theory. This theory contains only one gauge group and only vectorlike couplings. More realistic string models with large extra dimensions have been constructed by Kakuzhadze, Tye, and Shiu , Antoniadis, Bachas, and Dudas , and Ibanez, Rabadan, and Uranga . These models are quite complicated. The added structure is inessential to the general phenomenological picture that we will present in this paper, though there are many model-dependent details that would be interesting to study.
With this motivation, we consider a very simple embedding of QED into Type IIB string theory. In this theory, there exists a stable BPS object, the D3-brane, which is a 4-dimensional hypersurface on which open strings may end. We will assume that the 10-dimensional space of the theory has 6 dimensions compactified on a torus with a periodicity $`2\pi R`$, and that $`N`$ coincident D3-branes are stretched out in the 4 extended dimensions. The massless states associated with open strings that end on the branes are described by an $`N=4`$ supersymmetric Yang-Mills theory with a gauge group $`U(N)`$. These states include gauge bosons $`A^{\mu a}`$, gauginos $`\stackrel{~}{g}^{ai}`$, and complex scalars $`\varphi ^a`$, where $`a`$ is an index of the adjoint representation of $`U(N)`$ and $`i`$ runs from 1 to 4. We will project this theory down to a $`U(1)`$ gauge theory with two massless Weyl fermions and identify the gauge boson and fermions of that theory with the photon and electron of QED.
We take the parameters of this theory to be the string scale $`M_S=\alpha _{}^{}{}_{}{}^{1/2}`$ and the (dimensionless) Yang-Mills coupling constant $`g`$, which we identify with a Standard Model gauge coupling. (Except for this definition of $`g`$, we adopt the conventions of ). Note that $`M_S`$ is directly observable: The SR resonances occur at masses $`M_n=\sqrt{n}M_S`$, for $`n=1,2,\mathrm{}`$.
The gravitational constant and other physical scales in the theory are derivable from $`M_S`$ and $`g`$. However, the relation involves one-loop calculations and is model-dependent, depending on the full spectrum of the theory. Quite generally in the ADD scenario, the Newton constant which represents the observed strength of gravity is given in terms of the fundamental gravitational scale $`M`$ by the relation
$$(4\pi G_N)^1=M^{n+2}R^n,$$
(1)
where the compact dimensions are taken to be flat and periodic with period $`2\pi R`$. Our toy model corresponds to the case $`n=6`$. In Section 6 we will present a simple but model-dependent computation of the relation between $`M`$ and string scale $`M_S`$. We will show that
$$\frac{M}{M_S}=\left(\frac{1}{\pi }\right)^{1/8}\alpha ^{1/4},$$
(2)
where $`\alpha =g^2/4\pi `$. Then, for two extreme choices,
$`\alpha =1/137`$ $``$ $`M/M_S=3.0;`$
$`\alpha =\alpha _s(\text{1 TeV})`$ $``$ $`M/M_S=1.6.`$ (3)
In scattering amplitudes involving virtual gravitons, the gravity scale will enter as $`M^4`$, and so the string and gravity effects will be well-separated in size. For future reference, the tension of the D3-brane is given by
$$\tau _3=\frac{1}{8\pi ^3}\alpha ^1M_S^4.$$
(4)
The relations in (3) illustrate the most problematical aspect of our analysis. The naive string constructions we will use in this paper require all of the Standard Model gauge couplings to be unified at the string scale. Proposals for splitting these couplings to realistic values using the vacuum expectation value of a string modulus field are given in . However, in this paper we will deal with the Standard Model interactions only one at a time.
The explicit embedding that we will use is the following: Consider the $`SU(2)`$ subgroup of $`U(N)`$ with generators
$$t^+=\frac{1}{\sqrt{2}}\left(\begin{array}{cc}0& 1\\ 0& 0\end{array}\right),t^{}=\frac{1}{\sqrt{2}}\left(\begin{array}{cc}0& 0\\ 1& 0\end{array}\right),t^3=\frac{1}{2}\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right).$$
(5)
(In general, we normalize $`SU(N)`$ generators to $`\text{tr}[t^a(t^b)^{}]=\frac{1}{2}\delta ^{ab}`$.) We can identify the left-handed electron $`e_L^{}`$, the left-handed positron $`e_L^+`$, and the photon $`A_\mu `$ as
$$e_L^{}=\stackrel{~}{g}^1,e_L^+=\stackrel{~}{g}^{+1},A_\mu =A_\mu ^3,$$
(6)
where the superscript denotes the matrix from (5) which would be used in computing the Chan-Paton factor. The three generators form a closed operator algebra, and in fact the tree amplitudes of $`N=4`$ super-Yang-Mills theory which have only these states on external lines also involve only these states on internal lines. In string theory, we can reduce the massless sector to this set of states by an appropriate orbifold projection . (For example, in a $`U(2)`$ theory, mod out by $`Z_2\times Z_3`$, where $`Z_2`$ is the center of $`SU(2)`$ and the internal indices $`i`$ are assigned the $`Z_3`$ phases $`1,\zeta ,\zeta ,\zeta `$, with $`\zeta =e^{2\pi i/3}`$). This gives an explicit prescription for computing tree-level string corrections to QED amplitudes. The electric charge of the electron is given by
$$e=g,$$
(7)
as one can determine from the commutator $`[t^+,t^3]`$. To compute loop corrections, we should properly extend this theory to a full modular-invariant string construction. Instead, for simplicity, we will use the content of the original $`N=4`$ supersymmetric theory to compute the loop diagram studied in Section 6.
Most of our analysis will be carried out at the tree level in string theory. A tree-level amplitude for a particular process actually depends only on whether that process involves open- or closed-string states and is otherwise independent of which weak coupling string theory it belongs to. Beyond this, it depends only on the correlation function of the vertex operators associated with the external particles for that process and is independent of the remainder of the string spectrum. If the tree amplitude for a process involves four particles from an $`N=4`$ supersymmetric string theory, the amplitude is identical whether the full theory has $`N=4`$ supersymmetry or is nonsupersymmetric. This identity is explicit when a nonsupersymmetric model is constructed as an orbifold of a supersymmetric theory and, in that situation, is a special case of the โinheritanceโ property of orbifolds. This identity is also familiar in field theory, where tree-level scattering amplitudes in QCD are computed by recognizing that they are identical to amplitudes in a supersymmetric generalization of QCD . Thus, the string corrections to tree-level Standard Model amplitudes that we will compute in this paper are actually valid for any situation in which the quarks and leptons come from the untwisted sector of an open string orbifold.
Our tree amplitudes are model-independent in another way. An alternative string construction of the ADD scenario would be to consider Type IIA string theory with 5 dimensions large and one dimension small. Then the ADD scenario would arise if the Standard Model particles were bound to a D4-brane wrapped around the small dimension. Similarly, one could consider $`n`$ large and $`(6n)`$ small dimensions, with a D$`(9n)`$-brane wrapped around the small dimensions. If the small dimensions are smaller than 1/TeV, all external states would necessarily carry zero momentum in these directions. Then actually the tree amplitudes derived in this paper would apply for any value of $`n`$. We should also note that while we assume the toroidal compactification of the extra dimensions here, we expect the results for scattering of open strings on the D brane in Sections 3 and 4 to remain valid for models with a warp factor in the bulk , provided that the bulk curvature is sufficiently small near the brane.
## 3 Stringy corrections to $`e^+e^{}\gamma \gamma `$ and Bhabha scattering
In this section, we will use our toy model to compute the effects of TeV scale strings on Bhabha scattering and $`\gamma \gamma `$ production in $`e^+e^{}`$ collisions. We will compute the leading-order scattering amplitudes in string theory, using the external states described in the previous section.
Tree amplitudes of open-string theory are given as sums of ordered amplitudes multiplied by group theory Chan-Paton factors . We consider amplitudes with all momenta directed inward. Let the ordered amplitude with external states $`(1,2,3,4)`$ be denoted $`g^2A(1,2,3,4)`$. Then the full scattering amplitude $`๐(1,2,3,4)`$ is given by
$`๐(1,2,3,4)`$ $`=`$ $`g^2A(1,2,3,4)\text{tr}[t^1t^2t^3t^4+t^4t^3t^2t^1]`$ (8)
$`+g^2A(1,3,2,4)\text{tr}[t^1t^3t^2t^4+t^4t^2t^3t^1]`$
$`+g^2A(1,2,4,3)\text{tr}[t^1t^2t^4t^3+t^3t^4t^2t^1].`$
To compute QED amplitudes with fixed external states, we would substitute for each $`t^i`$ the appropriate matrix from (5) (or, for outgoing states, the Hermitian conjugate matrix).
The field theory tree amplitudes of Yang-Mills theory can be cast into the same form , and it is useful to consider that case first. Only a subset of the possible 4-point ordered amplitudes are nonzero; those amplitudes are given in Figure 2. In this figure, a wavy external line denotes a gauge boson, and a straight external line denotes a fermion. The sign denotes the helicity (for states directed inward). The diagrams are presented with the $`s`$-channel vertical and the $`t`$-channel horizontal. Actually, the four amplitudes involving fermions can be derived from the two with only gauge bosons by the use of $`N=1`$ supersymmetry Ward identities, and these identities also imply the vanishing of the ordered amplitudes for helicity combinations not shown in the figure. The two 4-gauge boson amplitudes are related by $`N=2`$ supersymmetry. This is an example of the model-independence discussed at the end of the previous section.
It is straightforward to check that these formulae give the familiar QED tree amplitudes. For example, for $`e_L^{}e_R^+`$ elastic scattering, only the first line of (8) has a nonzero Chan-Paton factor and we find
$$๐(e_L^{}e_R^+e_L^{}e_R^+)=2e^2\frac{u^2}{st}=2e^2u\left(\frac{1}{s}+\frac{1}{t}\right),$$
(9)
with $`g=e`$. For $`e^+e^{}`$ annihilation to $`\gamma \gamma `$, all three terms contribute and we find, for example,
$$๐(e_L^{}e_R^+\gamma _L\gamma _R)=e^2\sqrt{\frac{u}{t}}\left[\frac{u}{s}+\frac{t}{s}1\right]=2e^2\sqrt{\frac{u}{t}}.$$
(10)
The generalization of the formulae in Figure 2 to string states on a D-brane is known to be quite simple : All of the amplitudes shown in the figure are multiplied by the common factor
$$๐ฎ(s,t)=\frac{\mathrm{\Gamma }(1\alpha ^{}s)\mathrm{\Gamma }(1\alpha ^{}t)}{\mathrm{\Gamma }(1\alpha ^{}s\alpha ^{}t)}.$$
(11)
This factor is essentially the original Veneziano amplitude . Before we apply this result, it will be useful to sketch its derivation.
In the model described in Section 2, the electron and photon states are massless states of open strings ending on the D3-brane. These states are described by the quantum theory of fluctuations of an open string in which the string fields have Neumann boundary conditions in the $`\mu =`$ 0โ3 directions and Dirichlet boundary conditions in the $`\mu =`$ 5โ10 directions. The string world surface has the topology of a disk, as shown in Figure 3(a). The scattering amplitudes are evaluated by mapping this surface onto a circle in the complex plane, as in Figure 3(b), and then into the upper half plane. External open string states are represented by operators, called โvertex operatorsโ, placed on the boundary, and group theory matrices $`t^a`$, the Chan-Paton factors. When the boundary is mapped to the real line, the vertex operators appear in a given order 1,2,3,4, and their correlation function gives the ordered amplitude $`A(1,2,3,4)`$ which appears in (8). By summing over all orderings, one builds up the complete formula for $`๐(1,2,3,4)`$.
The explicit formula for the 4-point ordered amplitude is
$$A(1,2,3,4)=\frac{1}{\alpha _{}^{}{}_{}{}^{2}}X^2_0^1๐x\underset{i=1}{\overset{4}{}}๐ฑ_{q_i}(x_i,k_i),$$
(12)
where $`๐ฑ_{q_i}(x_i)`$ is the vertex operator of the state $`i`$. The operators are placed on the real axis at $`x_i=0,x,1,X`$, with $`X`$ to be fixed and sent to $`\mathrm{}`$. The index $`q_i`$ denotes the superconformal charge, which for the disk amplitude is constrained by $`_iq_i=2`$.
A good way to account for the boundary conditions on the real line is to perform the โdoubling trickโ, which represents left-moving fields on the world-sheet by fields in the upper half plane and right-moving fields by their continuation to the lower half plane. Explicitly, let us split the worldsheet boson field into its holomorphic and antiholomorphic parts:
$$X^\mu (z,\overline{z})=X^\mu (z)+\overline{X}^\mu (\overline{z}).$$
(13)
The boundary conditions imposed on $`X(z)`$ and the worldsheet fermion field $`\psi (z)`$ on the real line are then
$$X^\mu (z)=\pm \overline{X}^\mu (\overline{z}),\psi ^\mu (z)=\pm \overline{\psi }^\mu (\overline{z}),$$
(14)
where the plus sign corresponds to $`\mu =03`$ (Neumann boundary conditions), and the minus sign to $`\mu =510`$ (Dirichlet boundary conditions.) The fields $`X(z)`$ and $`\psi (z)`$ are originally defined only on the upper half-plane, $`๐^+`$. We extend the definitions of these fields to the full plane by identifying
$$X^\mu (z)=\pm \overline{X}^\mu (z),\psi ^\mu (z)=\pm \overline{\psi }^\mu (z),z๐^{},$$
(15)
where the plus and minus signs again correspond to the Neumann and Dirichlet boundary conditions. With these definitions, the correlation functions of these fields are given by
$`X^\mu (w)X^\nu (z)`$ $`=`$ $`{\displaystyle \frac{\alpha ^{}}{2}}g^{\mu \nu }\mathrm{ln}|wz|,`$
$`\psi ^\mu (w)\psi ^\nu (z)`$ $`=`$ $`g^{\mu \nu }(wz)^1,`$ (16)
for any $`\mu `$ and $`\nu `$.
The open string vertex operators are built from the worldsheet boson and fermion fields $`X^\mu `$ and $`\psi ^\mu `$, the spin field $`\mathrm{\Theta }_\alpha `$, and the superconformal ghost field $`\varphi `$. We work in the space-time metric $`(,+,\mathrm{},+)`$, and define the conventional Mandelstam variables by $`s=2k_1k_2`$, $`t=2k_1k_4`$, and $`u=2k_1k_3`$. Then, for photons, the vertex operators with $`q=1`$ and $`q=0`$ take the form
$`๐ฑ_1^\mu (x,k)`$ $`=`$ $`(2\alpha ^{})^{1/2}e^\varphi \psi ^\mu e^{i2kX}(x),`$
$`๐ฑ_0^\mu (x,k)`$ $`=`$ $`2(iX^\mu +\alpha ^{}k\psi \psi ^\mu )e^{i2kX}(x).`$ (17)
These expressions are referred to, respectively, as the โ$`1`$ pictureโ and the โ0 pictureโ. The factor of 2 in the exponentials compensates for the replacement of the full $`X^\mu (z,\overline{z})`$ by its holomorphic part in (13). For fermions, the vertex operator with with $`q=1/2`$ (โ$`1/2`$ pictureโ) is
$$๐ฑ_{1/2}^\alpha (x,k)=2^{1/2}\alpha _{}^{}{}_{}{}^{3/4}e^{\varphi /2}\mathrm{\Theta }^\alpha e^{i2kX}(x).$$
(18)
Note that for open strings, the momenta and polarization tensors are required to be parallel to the D-brane, so all the fields that appear in the vertex operators (17) and (18) have Neumann boundary conditions. It is then not surprising that the result (11) is identical to the corresponding result in type I string theory.
The correlators required for the calculation are given by (16) and
$`e^{\varphi (w)}e^{\varphi (z)}`$ $`=`$ $`(wz)^1,`$
$`\mathrm{\Theta }_\alpha (w)\mathrm{\Theta }_\beta (z)`$ $`=`$ $`C_{\alpha \beta }(wz)^{5/4},`$ (19)
where $`C_{\alpha \beta }`$ is the charge conjugation matrix. Explicitly evaluating the expressions (12) with these vertex operators and correlators, one finds the expressions in Figure 2 multiplied by the form factor (11), as promised.
A check on the normalization of the 0 picture operator is given by the operator product relation
$`ฯต_2๐ฑ_1(x,k_2)ฯต_1๐ฑ_0(0,k_1)`$ $``$ (20)
$`\alpha ^{}x^{2k_1k_2\alpha ^{}1}\left\{ฯต_1ฯต_2(k_1k_2)_\mu +2ฯต_1k_2ฯต_{2\mu }2ฯต_2k_1ฯต_{1\mu }\right\}๐ฑ_1^\mu (0,k_1+k_2)+\mathrm{\Delta },`$
where $`\mathrm{\Delta }`$ is a total derivative in $`x`$. A similar relation holds for $`๐ฑ_0(x,k_2)๐ฑ_0(0,k_1)`$. When inserted into (8), these relations give the correct factorization to a pole in $`(k_1+k_2)^2`$ and the three-gluon vertex, as shown in Figure 4. The relative normalization of $`๐ฑ_0`$ and $`๐ฑ_1`$ is given by the picture-changing relation . Then comparison of the four-point amplitudes to those of Yang-Mills theory gives the normalization of (12).
To compare string amplitudes to Standard Model amplitudes, we are typically interested in the limit in which $`s`$, $`t`$, $`u`$ are much less than the string scale $`M_S=\alpha _{}^{}{}_{}{}^{1/2}`$. In this limit,
$$๐ฎ(s,t)=\left(1\frac{\pi ^2}{6}\frac{st}{M_S^4}+\mathrm{}\right).$$
(21)
It is interesting that, in the toy model, the leading corrections are proportional to $`M_S^4`$, corresponding to an operator of dimension 8. This is a consequence of the fact that the first higher-dimension operator with $`N=4`$ supersymmetry appears at dimension 8 . It is likely that in more general string models in which quarks and leptons appear from twisted sectors of the orbifold, the first string corrections would be proportional to $`M_S^2`$.
Now we can apply the form factor (11) to representative QED processes. For Bhabha scattering, only the first Chan-Paton factor is nonzero, and so we find
$`๐(e_L^{}e_R^+e_L^{}e_R^+)`$ $`=`$ $`2e^2{\displaystyle \frac{u^2}{st}}๐ฎ(s,t),`$
$`๐(e_L^{}e_R^+e_R^{}e_L^+)`$ $`=`$ $`2e^2{\displaystyle \frac{t}{s}}๐ฎ(s,t),`$
$`๐(e_L^{}e_L^+e_L^{}e_L^+)`$ $`=`$ $`2e^2{\displaystyle \frac{s}{t}}๐ฎ(s,t),`$ (22)
and the same results for the parity-reflected processes. In general, all helicity amplitudes for Bhabha scattering are given by their field theory expressions multiplied by $`๐ฎ(s,t)`$. This form factor has SR poles in the $`s`$\- and $`t`$-channels. A $`u`$-channel pole cannot appear, because the open string contains no states with electric charge $`\pm 2`$.
For $`e^+e^{}\gamma \gamma `$, the result is more complex. The string form factor appears in all three possible channels, and we find
$$๐(e_L^{}e_R^+\gamma _L\gamma _R)=e^2\sqrt{\frac{u}{t}}\left[\frac{u}{s}๐ฎ(s,t)+\frac{t}{s}๐ฎ(s,u)๐ฎ(t,u)\right].$$
(23)
The other nonzero helicity amplitudes are derived from this one by parity reflection and crossing. In particular, the amplitude for production of $`\gamma _R\gamma _R`$ remains zero. The amplitude (23) contains massive SR poles in all three channels.
## 4 String phenomenology at colliders
The expressions for stringy corrections that we have derived allow one to search for signals of string theory in collider experiments. In this section, we will discuss these explicit signatures of string theory. We begin by considering effects visible as contact interactions well below the string scale. We will then discuss direct observation of the string Regge excitations.
### 4.1 Contact interactions
Both two-photon production and Bhabha scattering have been studied at LEP 2 at the highest available energies. We consider first the case of two-photon production. Deviations from the Standard Model cross section have been analyzed by the LEP experiments in terms of Drellโs parametrization
$$\frac{d\sigma }{d\mathrm{cos}\theta }=\frac{d\sigma }{d\mathrm{cos}\theta }|_{SM}\left(1\pm \frac{2ut}{\mathrm{\Lambda }_\pm ^4}\right).$$
(24)
For the case of $`e^+e^{}\gamma \gamma `$, it is actually a general result that the first correction due to a higher-dimension operator comes from a unique dimension-8 operator. This operator is proportional to the cross term in $`T^{\mu \nu }T_{\mu \nu }`$, where $`T^{\mu \nu }`$ is the energy-momentum tensor of QED. Thus, Drellโs parametrization (24) should apply to any model of new physics at short distances.
To compare our string theory results to this expression, insert (21) into (23); this gives
$$๐(e_L^{}e_R^+\gamma _L\gamma _R)=2e^2\sqrt{\frac{u}{t}}\left[1+\frac{\pi ^2}{12}\frac{ut}{M_S^4}+\mathrm{}\right].$$
(25)
Squaring this expression, and noting that the correction is invariant to crossing $`tu`$, we can identify
$$\mathrm{\Lambda }_+=\left(12/\pi ^2\right)^{1/4}M_S.$$
(26)
The OPAL collaboration has reported a limit $`\mathrm{\Lambda }_+>304`$ GeV from measurements at 183 and 189 GeV in the center of mass. The ALEPH, DELPHI, and L3 collaborations have reported similar constraints . The OPAL result corresponds to a limit
$$M_S>290\text{GeV},\text{95\% conf.}$$
(27)
If we use the first line of (3) to convert this to a limit on the fundamental quantum gravity scale, we find $`M>870`$ GeV.
The comparison of string predictions to the data on Bhabha scattering brings in two new considerations. The first of these is that Bhabha scattering at energies above the $`Z^0`$ resonance includes $`Z^0`$ exchange as an important contribution, while the $`Z^0`$ was not a part of our string QED. To find a prescription for including both $`\gamma `$ and $`Z^0`$ exchange, we recall that all QED Bhabha scattering amplitudes are multiplied by the common form factor $`๐ฎ(s,t)`$. Thus, we suggest comparing the data on Bhabha scattering to the simple formula
$$\frac{d\sigma }{d\mathrm{cos}\theta }(e^{}e^+e^{}e^+)=\frac{d\sigma }{d\mathrm{cos}\theta }|_{SM}\left|๐ฎ(s,t)\right|^2.$$
(28)
This is essentially the assumption that the SR excitations of the photon and the $`Z^0`$ have the same spectrum, up to contributions of size $`m_Z^2`$ that we can ignore in computing their masses, and that the SR excitations of the $`Z^0`$ have the same polarization asymmetry as the $`Z^0`$ in their coupling to electrons.
The second complication for Bhabha scattering is that, unlike the case of $`e^+e^{}\gamma \gamma `$, there are many possible forms for the higher-dimension corrections to the Standard Model result. Already at dimension 6 there are three possible helicity-conserving operators, of which two are also parity-conserving. At dimension 8 there are 4 parity-conserving operators. Various combinations of these operators have been proposed as the basis for fits to Bhabha scattering data. It would be useful to review the most important models proposed previously and to compare them to (28).
For many years, Bhabha scattering has been of interest as the most sensitive probe of lepton substructure. The form proposed for deviations from the Standard Model prediction was the most general combination of helicity-conserving dimension-6 operators
$$\delta =\frac{4\pi }{2\mathrm{\Lambda }^2}\left[\eta _{LL}\overline{e}_L\gamma ^\mu e_L\overline{e}_L\gamma _\mu e_L+\eta _{RR}\overline{e}_R\gamma ^\mu e_R\overline{e}_R\gamma _\mu e_R+2\eta _{RL}\overline{e}_R\gamma ^\mu e_R\overline{e}_L\gamma _\mu e_L\right],$$
(29)
where the $`\eta _a`$ are $`\pm 1`$ or 0 and the mass scale $`\mathrm{\Lambda }`$ is taken to be the scale of compositeness.
With the recent interest in large extra dimensions and low-scale quantum gravity, Bhabha scattering has been reconsidered as a place to look for the effects of virtual KK graviton exchange. As we have remarked in the introduction, the effect of KK exchange is not reliably computable in low-energy effective field theory. Typically, this effect is modeled by introducting an appropriate contact interaction with an adjustible coefficient . In this paper we will follow Hewettโs convention by representing the effective Lagrangian for KK exchange as :
$$\delta =i\frac{4\lambda }{M_H^4}T^{\mu \nu }T_{\mu \nu },$$
(30)
where $`\lambda =\pm 1`$ and $`T^{\mu \nu }`$ is the full energy-momentum tensor of the model. Hewett writes the scale in this Lagrangian as $`M_S`$; we use the notation $`M_H`$ to distinguish this mass scale from the string scale .
It should be noted that the expressions (29) and (30) do not contain any powers of a small coupling constant. When these expressions are added to the Standard Model formulae, the higher-dimension operators compete with amplitudes that are of order $`g^2`$. This allows one to obtain very stringent bounds on the coefficient of the new operators. Bounds on the $`\mathrm{\Lambda }`$ parameters, for example, are typically a factor of 20 higher than the center-of-mass energy of the $`e^+e^{}`$ collisions being analyzed. The physical meaning of these bounds, however, depends on the relation between the coefficients in (29) and (30) and the predictions of the underlying fundamental theory. In Section 6, we will derive (30) from our toy string model and show that the coefficient is of order
$$\frac{1}{M_H^4}\frac{g^4}{M_S^4}.$$
(31)
Thus, (30) is parametrically suppressed with respect to the effects of SR exchange. This conclusion is generic when quantum gravity is represented by a weakly-coupled string theory, though perhaps in other models of quantum gravity (30) might be the dominant effect.
With this in mind, we will compare the models discussed above to an illustrative data set for Bhabha scattering at LEP 2. A complete analysis of the LEP 2 data is beyond the scope of this paper. For reference, we have listed the various expressions for the Bhabha scattering cross sections in these models in Appendix A.
The four LEP experiments have all announced preliminary results on the Bhabha scattering cross section at high energies and have used the results to put limits on 4-fermion contact interactions. In particular, the L3 experiment has published their data at 183 GeV in a form convenient for our analysis. In Figure 5, we compare this data to the formula (28) and to the analogous formulae derived from (29) and (30). The curves shown are the 95% confidence exclusion limits for the various models considered: for SR exchange, $`M_S>410`$ GeV, for KK exchange with $`\lambda =+1`$, $`M_H>830`$ GeV, for compositeness with VV contact interactions ($`\eta _{LL}=\eta _{RR}=\eta _{RL}=1`$) $`\mathrm{\Lambda }>8800`$ GeV, for compositeness with AA contact interactions, ($`\eta _{LL}=\eta _{RR}=\eta _{RL}=+1`$), $`\mathrm{\Lambda }>6700`$ GeV. In a weakly-coupled string theory, the dominant effect would come from $`M_S`$. Using the relation (3), the exclusion limit on $`M_S`$ derived from this data would correspond to a limit on the quantum gravity scale of $`M>1.2`$ TeV.
A similar analysis can be used to estimate the sensitivity of experiments at future, higher-energy $`e^+e^{}`$ colliders. As a guide, consider a linear $`e^+e^{}`$ collider running at a center of mass energy of 1 TeV. With a 100 fb<sup>-1</sup> data sample, the measurement of Bhabha scattering should be systematics limited. We consider a set of 8 measurements of the differential cross sections corresponding to the bin centers in Figure 5 and assume that each measurement is made to 3% accuracy and agrees with the Standard Model expectation. Then the 95% confidence exclusion limits for the four models just considered are: for SR exchange, $`M_S>3.1`$ TeV, for KK exchange with $`\lambda =+1`$, $`M_H>6.2`$ TeV, for compositeness with VV contact interactions $`\mathrm{\Lambda }>88`$ TeV, for compositeness with AA contact interactions, $`\mathrm{\Lambda }>62`$ TeV. The corresponding deviations from the Standard Model expectation are graphed as a function of $`\mathrm{cos}\theta `$ in Figure 6. Using (3), the limit on $`M_S`$ would translate to a limit $`M>9.3`$ TeV on the quantum gravity scale.
A remarkable feature of Figure 6 is that the four curves shown have very different shapes. If a deviation from the Standard Model is seen, then with higher statistics or higher energy it should be possible to determine which of these theories, if any, gives the correct description.
### 4.2 Resonances
Though theories based on contact interactions are limited to the first deviations from the Standard Model, our string theory formulae are valid at higher energies, and we can examine their characteristic features there. The most obvious property apparent in (11) is the presence of a sequence of $`s`$-channel poles at masses $`M_n=\sqrt{n}M_S`$, for $`n=1,2,\mathrm{}`$. It is interesting to explore the properties of the first resonances in some detail.
The stringy form factor $`๐ฎ(s,t)`$ has its first pole at $`s=M_S^2`$. Near this point, it has the form
$$๐ฎ(s,t)\frac{t}{sM_S^2}.$$
(32)
We can use (32) to find the first resonance in string QED tree amplitudes. The pole contributions are
$`๐(e_L^{}e_R^+e_L^{}e_R^+)`$ $`=`$ $`2e^2{\displaystyle \frac{u^2}{s^2}}{\displaystyle \frac{s}{sM_S^2}},๐(e_L^{}e_R^+e_R^{}e_L^+)=2e^2{\displaystyle \frac{t^2}{s^2}}{\displaystyle \frac{s}{sM_S^2}},`$
$`๐(e_L^{}e_R^+\gamma _L\gamma _R)`$ $`=`$ $`2e^2{\displaystyle \frac{u\sqrt{ut}}{s^2}}{\displaystyle \frac{s}{sM_S^2}},๐(\gamma _L\gamma _R\gamma _L\gamma _R)=2e^2{\displaystyle \frac{u^2}{s^2}}{\displaystyle \frac{s}{sM_S^2}},`$
$`๐(\gamma _R\gamma _R\gamma _R\gamma _R)`$ $`=`$ $`2e^2{\displaystyle \frac{s}{sM_S^2}},๐(e_R^{}e_R^+e_R^{}e_R^+)=2e^2{\displaystyle \frac{s}{sM_S^2}},`$
with equal results for the parity-reflected and time-reversed processes, and zero for all other possible reactions.
The properties of the first SR resonances can then be found by factorizing these expressions. They require four spin 0 resonances $`\gamma _{0i}`$, $`i=1,\mathrm{},4`$, one spin 1 resonance $`\gamma _1^{}`$ and one spin 2 resonance $`\gamma _2^{}`$. Four spin zero resonances are needed because the transition amplitudes between any pair of $`e_R^{}e_R^+`$, $`e_L^{}e_L^+`$, $`\gamma _R\gamma _R`$ and $`\gamma _L\gamma _L`$ vanish. The on-shell couplings of electron and photon pairs to the resonances are
| $`๐(\gamma _R\gamma _R\gamma _{01}^{})=\sqrt{2}eM_S,`$ | $`๐(e_L^{}e_R^+\gamma _1^{})=\sqrt{\frac{3}{2}}eM_Sฯต_{}^\mu ,`$ |
| --- | --- |
| $`๐(\gamma _L\gamma _L\gamma _{02}^{})=\sqrt{2}eM_S,`$ | $`๐(e_R^{}e_L^+\gamma _2^{})=\sqrt{\frac{1}{2}}eM_S\frac{1}{\sqrt{2}}[ฯต_+^\mu ฯต_0^\nu +ฯต_+^\nu ฯต_0^\mu ],`$ |
| $`๐(e_R^{}e_R^+\gamma _{03}^{})=\sqrt{2}eM_S,`$ | $`๐(e_L^{}e_R^+\gamma _2^{})=\sqrt{\frac{1}{2}}eM_S\frac{1}{\sqrt{2}}[ฯต_{}^\mu ฯต_0^\nu +ฯต_{}^\nu ฯต_0^\mu ],`$ |
| $`๐(e_L^{}e_L^+\gamma _{04}^{})=\sqrt{2}eM_S,`$ | $`๐(\gamma _L\gamma _R\gamma _2^{})=\sqrt{2}eM_Sฯต_{}^\mu ฯต_{}^\nu ,`$ |
| $`๐(e_R^{}e_L^+\gamma _1^{})=\sqrt{\frac{3}{2}}eM_Sฯต_+^\mu ,`$ | |
(36)
(37)
where, when the first particle moves in the $`+\widehat{3}`$ direction,
$$ฯต_+^\mu =\frac{1}{\sqrt{2}}(0,1,i,0)^\mu ,ฯต_{}^\mu =\frac{1}{\sqrt{2}}(0,1,i,0)^\mu ,ฯต_0^\mu =(0,0,0,1)^\mu .$$
(38)
Feynman rules which give rise to these expressions are listed in Figure 7.
From these expressions, we can compute the width of the resonances. For the scalar SR resonances,
$$\mathrm{\Gamma }_{01}=\mathrm{\Gamma }_{02}=\frac{\alpha }{4}M_S,\mathrm{\Gamma }_{03}=\mathrm{\Gamma }_{04}=\frac{\alpha }{2}M_S.$$
(39)
For the vector resonance,
$$\mathrm{\Gamma }_1=\frac{\alpha }{4}M_S,$$
(40)
with equal contributions from decays to $`e_R^{}e_L^+`$ and $`e_L^{}e_R^+`$. For the spin 2 resonance,
$$\mathrm{\Gamma }_2(e^+e^{})=\mathrm{\Gamma }_2(\gamma \gamma )=\frac{\alpha }{20}M_S,\mathrm{\Gamma }_2=\frac{\alpha }{10}M_S,$$
(41)
again with equal contributions from $`e_R^{}e_L^+`$ and $`e_L^{}e_R^+`$. The production cross sections can be derived from these formulae using, for example
$$\sigma (e^+e^{}\gamma _J^{})=4\pi ^2(2J+1)\frac{\mathrm{\Gamma }(\gamma _J^{}e^+e^{})}{M_S}\delta (sM_S^2).$$
(42)
In $`e^+e^{}`$ collisions, one currently has data available only up to 200 GeV. In quark-antiquark processes, however, collision energies up to 1 TeV and above are available in the Tevatron data. Thus, it is important to generalize this analysis to $`q\overline{q}`$ collisions so that we can ask whether the SR excitations of the gluon ought to have been seen at the Tevatron. We will now present our first attempt at a generalization of string QED to string QCD. Though this theory will not be completely satisfactory, it will at least allow us to estimate the bound on the string scale from the study of jets at the Tevatron.
Consider, then, a system of four D3-branes with a $`U(4)`$ gauge symmetry. Represent the gluons of QCD by the gauge bosons of $`SU(3)U(4)`$, that is, by $`3\times 3`$ Chan-Paton matrices $`t^a`$. Represent left-handed quarks and antiquarks of one flavor by the $`U(4)`$ matrices
$$(t^i)_{pq}=\frac{1}{\sqrt{2}}\delta _p^i\delta _q^4,(\overline{t}^i)_{pq}=\frac{1}{\sqrt{2}}\delta _q^i\delta _p^4.$$
(43)
Ideally, we would like to make an orbifold projection of the $`U(4)`$ theory onto a theory which contained only these quarks and gluons at the massless level. Unfortunately, this is not possible, because the commutator $`[t^i,\overline{t}^j]`$ includes not only a linear combination of the $`t^a`$ but also the $`U(1)`$ generator
$$t_4=\frac{1}{\sqrt{24}}\left(\begin{array}{cccc}1& & & \\ & 1& & \\ & & 1& \\ & & & 3\end{array}\right).$$
(44)
Thus, this massless $`U(1)`$ gauge boson will also appear in quark-quark scattering amplitudes.
Keeping this problem in mind, we compute the amplitude for $`q_L\overline{q}_R`$ scattering using (8). Only the first line has a nonzero Chan-Paton factor, which equals
$`\text{tr}[t^i\overline{t}^jt^k\overline{t}^{\mathrm{}}+\overline{t}^{\mathrm{}}t^k\overline{t}^jt^i]`$ $`=`$ $`{\displaystyle \frac{1}{4}}\left\{\delta ^{jk}\delta ^\mathrm{}i+\delta ^k\mathrm{}\delta ^{ij}\right\}`$ (45)
$`=`$ $`{\displaystyle \frac{1}{2}}\left\{(t^a)_{ji}(t^a)_\mathrm{}k+{\displaystyle \frac{2}{3}}\delta ^{ji}\delta ^\mathrm{}k\right\}.`$
In the last line, the first term corresponds to color octet exchange in the $`s`$-channel, and the second term to exchange of a $`U(1)`$ boson corresponding to the generator (44). To make our estimate, we will drop the $`U(1)`$ piece and then factorize the color octet piece of the amplitude as above. This gives
$$๐(q_L^i\overline{q}_R^jq_L^{\mathrm{}}\overline{q}_R^k)=2g^2\frac{u^2}{st}(t^a)_{ji}(t^a)_\mathrm{}k๐ฎ(s,t),$$
(46)
which implies:
$`๐(q_L^i\overline{q}_R^jg_1^a)`$ $`=`$ $`\sqrt{{\displaystyle \frac{3}{2}}}gM_S(t^a)_{ji}ฯต_{}^\mu ,`$
$`๐(q_L^i\overline{q}_R^jg_2^a)`$ $`=`$ $`\sqrt{{\displaystyle \frac{1}{2}}}gM_S(t^a)_{ji}{\displaystyle \frac{1}{\sqrt{2}}}[ฯต_{}^\mu ฯต_0^\nu +ฯต_{}^\nu ฯต_0^\mu ],`$ (47)
and similarly for $`q_R^i\overline{q}_L^j`$. The result is just what we would have obtained by replacing $`e`$ by $`g`$ and adding an $`SU(3)`$ color matrix in the Feynman rules of Figure 7. From these matrix elements, we can compute the production cross sections from unpolarized $`q\overline{q}`$ initial states:
$$\sigma (q\overline{q}g_1^{})=\frac{4\pi ^2\alpha _s}{3}\delta (sM_S^2),\sigma (q\overline{q}g_2^{})=\frac{4\pi ^2\alpha _s}{9}\delta (sM_S^2),$$
(48)
so that
$$\sigma (q\overline{q}g^{})=\frac{16\pi ^2\alpha _s}{9}\delta (sM_S^2).$$
(49)
The result (49) can be compared to the cross section for producing the axigluon and coloron , hypothetical massive vector or axial vector bosons that couple to $`q\overline{q}`$ with the QCD coupling strength. In either case, the cross section is
$$\sigma (q\overline{q}V)=\frac{16\pi ^2\alpha _s}{9}\delta (sM_S^2).$$
(50)
Then we can use experimental constraints on these objects to place a direct experimental bound on the string scale. A recent paper by the CDF collaboration has searched for the presence of a narrow resonance in the two-jet invariant mass distribution in $`p\overline{p}`$ collisions at the Tevatron . The CDF collaboration does not find evidence for such a resonance and puts a lower limit of 980 GeV (at 95% conf.) on the axigluon or coloron mass. Naively, we should have the same limit on $`M_S`$. Several uncertain factors appear in this comparison, however. On the negative side, the events with $`g_2^{}`$ have an angular distribution which is more peaked toward the beam axis, and so the acceptance for these events should be lower. The angular distribution for the $`g_1^{}`$ events is identical to that from the axigluon or coloron. On the positive side, we have ignored scalar gluon resonances and the production of $`g_1^{}`$ and $`g_2^{}`$ by gluons. Thus, we might say that the CDF limit constraints the string scale $`M_S`$ to be greater than approximately 1 TeV. If we convert this limit to a limit on the quantum gravity scale using the second line of (3), we find that $`M>1.6`$ TeV.
The sensitivity to SR resonances in quark and gluon scattering will increase dramatically when the LHC begins operation. The sensitivity of higher-energy hadron colliders to the axigluon was estimated some time ago by Bagger, Schmidt, and King . Scaling their results to the LHC energy, we expect that the LHC could put a limit of about 5 TeV on the axigluon mass, and a comparable limit on $`M_S`$. Using (3), this would correspond to a limit $`M>8`$ TeV. These values are sufficiently high that string resonances ought to be discovered at the LHC if the low quantum gravity scale is connected to the mechanism of electroweak symmetry breaking as suggested by ADD .
To conclude this section, we discuss what happens when we probe even higher energies, above the scale of the first SR resonance. When $`s>M_n^2`$, the expression (11) has a zero at $`t=(sM_n^2)`$. Thus, above the first resonance, there is one zero in $`\mathrm{cos}\theta `$, above the second resonance, there are two zeros, and so forth. This leads to an angular distribution of the sort produced by diffractive scattering. In Figure 8, we plot the differential cross section for Bhabha scattering, from (28), for a sequence of energies that interleave the SR resonances.
It is well-known from the old string literature that the differential cross sections at very high energy have the form of a narrow diffractive peak. Indeed, using Stirlingโs formula to evaluate $`๐ฎ(s,t)`$ in the limit $`s\mathrm{}`$ and fixed angle, we find
$$๐ฎ(s,t)\mathrm{exp}[\alpha ^{}sf(\mathrm{cos}\theta )],$$
(51)
where $`f(\theta )`$ is the positive function
$$f(c)=\frac{1+c}{2}\mathrm{log}\frac{1+c}{2}\frac{1c}{2}\mathrm{log}\frac{1c}{2}.$$
(52)
However, at intermediate energies, the large positive deviation in the backward direction is also an important part of the string signature. As $`\mathrm{cos}\theta 1`$,
$$\left|๐ฎ(s,t)\right|^2\left(\frac{\pi \alpha ^{}s}{\mathrm{sin}\pi \alpha ^{}s}\right)^2.$$
(53)
Thus for increasing $`s`$ there is a larger enhancement, but in a narrower region of backward angles.
## 5 Stringy corrections to $`e^+e^{}\gamma G`$
Our toy model includes the process of graviton emission in electron-positron annihilation, $`e^+e^{}\gamma G`$. This process gives a missing-energy signature which becomes significant when the center-of-mass energy of the annihilation approaches the gravitational scale $`M`$. The process has been used by the LEP 2 experiments to put constraints on the size of large extra dimensions. In this section, we study the stringy corrections to this process.
To begin, we recall that the leading contribution to this process at low energy is model-independent. The calculation uses only the fact that a gravitonโeven a KK excitationโcouples to the energy-momentum tensor of matter . The coupling has the usual 4-dimensional gravitational strength. From this, one finds that the polarized differential cross section for the process $`e_L^{}e_R^+\gamma G`$, for production of a given KK excitation of mass $`m`$, is given by
$`{\displaystyle \frac{d\sigma }{d\mathrm{cos}\theta }}|_{\text{ft}}`$ $`=`$ $`{\displaystyle \frac{\pi \alpha G_N}{1m^2/s}}[(1+\mathrm{cos}^2\theta )(1+\left({\displaystyle \frac{m^2}{s}}\right)^4)`$ (54)
$`+\left({\displaystyle \frac{13\mathrm{cos}^2\theta +4\mathrm{cos}^4\theta }{1\mathrm{cos}^2\theta }}\right){\displaystyle \frac{m^2}{s}}(1+\left({\displaystyle \frac{m^2}{s}}\right)^2)+6\mathrm{cos}^2\theta \left({\displaystyle \frac{m^2}{s}}\right)^2].`$
To obtain the full cross-section for graviton emission at a given collision energy, we need to sum over all the modes whose emission is kinematically allowed. The resulting cross-section behaves as $`\sigma s^{n/2}/M^{n+2}`$. This expression grows with $`s`$; if it were valid for all $`s`$, it would violate unitarity. We will see that string theory supplies an appropriate form factor to cut off this dependence.
In our stringy toy model, the graviton is a part of the closed string massless spectrum, while the electrons and photons are described by massless states of open strings. Therefore, to study the process $`e^+e^{}\gamma G`$ we consider the string scattering amplitude involving three open strings and a closed string. The calculation of this amplitude is very similar to the calculation of the four open-string scattering presented in Section 3. The amplitude is given by
$$(1,2,3,G)=gM(1,2,3,G)\text{tr}([t^1,t^2]t^3),$$
(55)
where we need to substitute for each $`t^i`$ the appropriate matrix from (5). To evaluate the ordered amplitude $`M(1,2,3,G)`$, we map the string worldsheet in Fig. 9 (a) onto a disc, and then into the upper half plane. The three open string vertex operators have to be placed on the boundary; the closed string vertex operator can sit anywhere inside the upper half plane. Then, the ordered amplitude is
$$M(1,2,3,G)=\frac{1}{\alpha _{}^{}{}_{}{}^{2}}X^2_{๐^+}d^2z\underset{i=1}{\overset{3}{}}๐ฑ_{q_i}(x_i,k_i)๐ฑ_{q_G}(z,\overline{z},k_G),$$
(56)
where $`๐ฑ_{q_i}(x_i)`$ is the vertex operator of the open string state $`i`$, and $`๐ฑ_{q_G}(z,\overline{z})`$ is the vertex operator of the graviton. The open string vertex operators are placed on the real axis at $`x_i=0,1,X`$, with $`X`$ to be fixed and sent to $`\mathrm{}`$. The integral is taken over the upper half plane $`๐^+`$. Just as in Section 3, we perform the doubling trick, extending the definitions of the fields to the full complex plane; then the open string vertex operators are given by (17) and (18). The closed string vertex operator in the 0 picture takes the form
$`๐ฑ_{0,0}^{\mu \lambda }(z,\overline{z},k)`$ $`=`$ $`{\displaystyle \frac{\kappa }{\pi \alpha ^{}}}D_\nu ^\lambda \left(X^\mu (z)+ik\psi (z)\psi ^\mu (z)\right)e^{ikX(z)}`$ (57)
$`\left(X^\nu (\overline{z})+iDk\psi (\overline{z})\psi ^\nu (\overline{z})\right)e^{iDkX(\overline{z})},`$
where $`D_\nu ^\mu =1`$ for $`\mu =\nu =\mathrm{0..3}`$, $`D_\nu ^\mu =1`$ for $`\mu =\nu =\mathrm{5..10}`$, and $`D_\nu ^\mu =0`$ for $`\mu \nu `$. Using these vertex operators and the correlation functions given in (16) and (19), the amplitude (56) can be evaluated. In this calculation, one encounters integrals of the form
$`I_0(a,b,c)`$ $`=`$ $`{\displaystyle _{๐^+}}d^2z|z|^a|1z|^b(z\overline{z})^c,`$
$`I_1(a,b,c)`$ $`=`$ $`{\displaystyle _{๐^+}}d^2z|z|^a|1z|^b(z\overline{z})^c(z+\overline{z}),`$ (58)
with arbitrary $`a,b,c.`$ Using the representation
$$|z|^a=\frac{1}{\mathrm{\Gamma }(a/2)}_0^{\mathrm{}}t^{a/21}e^{t|z|^2}๐t$$
(59)
these integrals can be evaluated. The results are
$`I_0(a,b,c)`$ $`=`$ $`(2i)^c{\displaystyle \frac{\sqrt{\pi }}{2}}\mathrm{\Gamma }\left(1(a+b+c)/2\right)`$
$`{\displaystyle \frac{\mathrm{\Gamma }\left((1+c)/2\right)\mathrm{\Gamma }\left(1+(b+c)/2\right)\mathrm{\Gamma }\left(1+(a+c)/2\right)}{\mathrm{\Gamma }(a/2)\mathrm{\Gamma }(b/2)\mathrm{\Gamma }\left(2+(a+b)/2+c\right)}};`$
$`I_1(a,b,c)`$ $`=`$ $`2{\displaystyle \frac{2+a+c}{4+a+b+2c}}I_0(a,b,c).`$ (60)
We find that the individual amplitudes contributing to (54) are all multiplied by a common factor
$`(s,t,u,m^2)`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{\pi }}}e^{(\mathrm{log}2)\alpha ^{}m^2}\mathrm{\Gamma }({\displaystyle \frac{1}{2}}{\displaystyle \frac{1}{2}}\alpha ^{}m^2)`$ (61)
$`{\displaystyle \frac{\mathrm{\Gamma }(1\frac{1}{2}\alpha ^{}s)\mathrm{\Gamma }(1\frac{1}{2}\alpha ^{}t)\mathrm{\Gamma }(1\frac{1}{2}\alpha ^{}u)}{\mathrm{\Gamma }(1+\frac{1}{2}\alpha ^{}(sm^2))\mathrm{\Gamma }(1+\frac{1}{2}\alpha ^{}(tm^2))\mathrm{\Gamma }(1+\frac{1}{2}\alpha ^{}(um^2))}}.`$
An analogous result holds for the process $`gggG`$: To obtain the string theory amplitude, we just multiply the field theory answer by the same prefactor (61). This result is in agreement with the calculation of Dudas and Mourad . We believe, but we have not been able to show, that the relation among amplitudes is a consequence of the $`N=4`$ supersymmetry of the underlying model. The field-theory cross section formula (54) is then modified by
$$\frac{d\sigma }{d\mathrm{cos}\theta }=\frac{d\sigma }{d\mathrm{cos}\theta }|_{\text{ft}}\left|(s,t,u,m^2)\right|^2.$$
(62)
The expression (61) has an interesting pole structure . The poles in the $`s`$ channel occur for $`s=2nM_S^2`$, and correspond to producing SR states with an even excitation number. The SR states with an odd excitation number cannot decay into a graviton and an open string massless state. On the other hand, these states can mix with the graviton, leading to the appearance of extra poles at $`m^2=(2n+1)M_S^2`$. These poles were also observed by Hashimoto and Klebanov in their calculation of the gluon-gluon-graviton vertex. Their presence is essential for the correct factorization properties of the form factor (61).
The form factor (61) expresses the way in which the amplitudes for KK graviton emission are cut off in all relevant high-energy limits. Assume that the kinematic variables are sufficiently far away from any of the poles in (61). (Near the poles, the effects of finite width of the resonances have to be taken into account. This is beyond the scope of our analysis here.) For the radiation of state of very high mass, we can evaluate $``$ at the threshold $`s=m^2`$, $`t=u=0`$, and then take $`m^2`$ large. Using Stirlingโs formula, we find
$$\mathrm{exp}[(\mathrm{log}2)\alpha ^{}m^2].$$
(63)
In the limit of fixed mass, $`s\mathrm{}`$, and fixed angle, we find
$$\mathrm{exp}[\alpha ^{}sf(\mathrm{cos}\theta )],$$
(64)
where $`f(c)`$ is the function defined in (52). In the high-energy limit in which $`s,t,u,m^2`$ all become large together, we find the more complicated formula
$$\mathrm{exp}[\frac{1}{2}\alpha ^{}sf(x,\mathrm{cos}\theta )],$$
(65)
where $`x=m^2/s`$ and $`f(x,c)`$ is given by
$`f(x,c)`$ $`=`$ $`x\mathrm{log}4x(1x){\displaystyle \frac{(1+c)}{2}}\mathrm{log}{\displaystyle \frac{(1+c)}{2}}(1x){\displaystyle \frac{(1c)}{2}}\mathrm{log}{\displaystyle \frac{(1c)}{2}}`$ (66)
$`({\displaystyle \frac{(1+c)}{2}}+x{\displaystyle \frac{(1c)}{2}})\mathrm{log}({\displaystyle \frac{(1+c)}{2}}+x{\displaystyle \frac{(1c)}{2}})`$
$`({\displaystyle \frac{(1c)}{2}}+x{\displaystyle \frac{(1+c)}{2}})\mathrm{log}({\displaystyle \frac{(1c)}{2}}+x{\displaystyle \frac{(1+c)}{2}}).`$
The function $`f(x,c)`$ is positive for the allowed values of $`c`$ and $`x`$, even though this property is not manifest in (66). Thus, the string correction (61) gives a form factor suppression in all hard-scattering regions.
Recently, Bando et al. have pointed out that high-mass graviton emission from a brane is suppressed by a form factor effect due to brane recoil. The formula they propose is
$$\mathrm{exp}[\frac{1}{2}\frac{\mathrm{\Lambda }_S^2}{\tau _3}m^2],$$
(67)
where $`\tau _3`$ is the brane tension and $`\mathrm{\Lambda }_S`$ is a cutoff scale which should be of order $`M_S`$. The expression in the exponent is smaller than that in (63) by a factor of order $`g_{YM}^2`$. In weak-coupling Type IIB string theory, brane recoil is described by the emission of scalars in the $`N=4`$ gauge multiplet associated with brane. With the orbifold projection described in Section 2, there is one scalar $`\varphi ^3`$ that survives and remains in the spectrum. This scalar does not couple to the QED state in the field theory limit, but it does couple through higher-dimension operators. However, these couplings are proportional to one factor of $`g_{YM}`$ in the amplitude for each $`\varphi ^3`$ emitted. These inelastic processes deplete the cross section for elastic $`G`$ emission without $`\varphi ^3`$ emission and should lead to a form factor suppression of the form $`\mathrm{exp}[cg_{YM}^2m^2/M_S^2]`$. This is in agreement with the result of . However, we see from (63) that there is a parametrically more important source for the form factor, the intrinsic non-pointlike nature of the states in string theory. We should note that the numerical coefficient in the formula (4) for the brane tension is quite small, so that effects of the size (67) might nevertheless be relevant.
In our study of open-string scattering, we saw that the form factor cutoff of string amplitudes is important only at very high energy. At energies of the order of the string scale, a much more important phenomenon is the enhancement of scattering cross sections through the effect of SR resonances. We have seen that the amplitudes for graviton emission contain the series of SR poles at $`s=2nM_S^2`$ and $`m^2=(2n+1)M_S^2`$. Thus, string theory predicts an enhancement of the rate for graviton emission processes such as $`e^+e^{}\gamma G`$ through resonant processes such as
$$e^+e^{}\gamma ^{}\gamma G,e^+e^{}\gamma \gamma _{1,2}^{}\gamma \gamma G.$$
(68)
Typically, the resonances would be seen more clearly in $`e^+e^{}`$ or $`q\overline{q}`$ elastic scattering. However, the resonant production of missing-energy events would be an important confirmation that the observed resonances were a manifestation of quantum gravity with large extra dimensions.
## 6 Stringy corrections to $`\gamma \gamma `$ scattering
In this section, we address the question of the relative strengths of the effective operators in the low energy theory mediated by virtual SR and KK exchanges. At the end of Section 1, we argued, on very general grounds, that in any weakly coupled string theory the SR-mediated operators are expected to dominate. Here, we will substantiate this claim by an explicit calculation.
### 6.1 Tree amplitude
It is important to note that, unlike renormalizable field theory, string theory gives a nonzero contribution to the $`\gamma \gamma `$ scattering amplitude at the tree level. To compute this amplitude, we follow the procedure outlined in Section 3. We find
$$๐(\gamma _R\gamma _R\gamma _R\gamma _R)=e^2s^2\left[\frac{1}{st}๐ฎ(s,t)+\frac{1}{su}๐ฎ(s,u)+\frac{1}{tu}๐ฎ(t,u)\right],$$
(69)
where $`๐ฎ(s,t)`$ is given by (11). The helicity amplitudes for $`\gamma _R\gamma _L\gamma _R\gamma _L`$ and $`\gamma _L\gamma _L\gamma _L\gamma _L`$ can be obtained from (69) by crossing. All other helicity amplitudes vanish.
The expression (69) must vanish in the field theory limit $`\alpha ^{}0`$. This is easily seen as a consequence of $`s+t+u=0`$. Using a higherโorder expansion of $`๐ฎ`$, as in (21), we obtain
$$๐(\gamma _R\gamma _R\gamma _R\gamma _R)=\frac{\pi ^2}{2}e^2\frac{s^2}{M_S^4}+\mathrm{},๐(\gamma _R\gamma _L\gamma _R\gamma _L)=\frac{\pi ^2}{2}e^2\frac{u^2}{M_S^4}+\mathrm{}.$$
(70)
This result can be compared to the $`\gamma \gamma \gamma \gamma `$ amplitude induced by KK graviton exchange. Using the effective Lagrangian (30), it is straightforward to see that
$$๐_{\mathrm{KK}}(\gamma _R\gamma _R\gamma _R\gamma _R)=16\frac{\lambda }{M_H^4}s^2,๐_{\mathrm{KK}}(\gamma _R\gamma _L\gamma _R\gamma _L)=16\frac{\lambda }{M_H^4}u^2.$$
(71)
These expressions have exactly the same form as (70), and this must be so, because there is only one gauge-invariant, parity-conserving dimension 8 operator which contributes to $`\gamma \gamma \gamma \gamma `$. However, the scale $`M_H`$ in (71) is different from the string scale that appears in (70). We have already remarked in Section 4 that the relation between $`M_S`$ and $`M_H`$ can be obtained explicitly in our string model, and that in a weakly-coupled string theory the effect of KK graviton exchanges (71) is subdominant to the SR exchanges (70). In the next section, we will derive that result.
### 6.2 Loop amplitude
In string theory, the graviton exchange proper arises at the next order in perturbation theory. The graviton is a closed-string state. It first appears in open-string perturbation theory through the 1-loop diagram shown in Figure 10 . In this section, we will compute this diagram and show that it contains a piece which has the form of the one-graviton exchange amplitude. Some other properties of this diagram have recently been analyzed in .
In the covariant formulation of string theory , the open string loop amplitude shown in Figure 10 is computed in terms of correlation functions of vertex operators placed on the two boundaries. It is convenient to conformally map the annulus shown in Figure 10(b) into a cylinder, represented by a rectangle in the complex plane
$$0\mathrm{}w\pi ,0\mathrm{}w2\pi t,$$
(72)
periodically connected with the identification $`ww+2\pi it`$. The boundaries of the annulus are mapped to the lines $`\mathrm{}w=0`$ and $`\mathrm{}w=\pi `$. The parameter $`t`$ is a modulus which must be integrated over the whole range $`0<t<\mathrm{}`$.
The complete four-point open string amplitude is a sum of ordered amplitudes in which the four vertex operators are placed on the boundaries in all possible ways. The open strings on a D-brane and the Type IIB closed strings are oriented, so we do not need to consider non-orientable worldsheets such as the Mobius strip. Thus,
$`๐_{1\mathrm{loop}}=`$ $`g^4A_\mathrm{p}(1,2,3,4)\text{tr}[t^1t^2t^3t^4]+\mathrm{perms}+`$ (73)
$`g^4A_{\mathrm{np}}(1,2;3,4)\text{tr}[t^1t^2]\text{tr}[t^3t^4]+\mathrm{perms}.`$ (75)
This equation is the analogue of the tree-level color decomposition in (8). Only the second line, the โnon-planarโ amplitude, has the correct color structure to represent graviton exchange. We will show that the first term in the second line, which we denote $`๐_{Gs}`$, contains the contribution of a virtual graviton exchanged in the $`s`$-channel.
The explicit expression for $`๐_{Gs}`$ is
$`๐_{Gs}=g^4\text{tr}[t^1t^2]\text{tr}[t^3t^4]{\displaystyle \frac{dt}{t}\left[\underset{i=1}{\overset{4}{}}_0^{2\pi t}๐y_i\right]}`$
$`Z_x^p{\displaystyle \underset{\lambda }{}}Z_\lambda {\displaystyle \underset{i=1}{\overset{4}{}}}ฯต_i๐ฑ_0(w_i,k_i)_\lambda ,`$ (76)
where $`Z_x^p`$ denotes the partition function of the worldsheet bosons $`X^\mu `$ and the anticommuting ghosts, and $`Z_\lambda `$ denotes the partition function of the worldsheet fermions $`\psi ^\mu `$ and the commuting ghosts. The expectation value is correspondingly assumed to be computed only from field contractions, excluding the partition functions. The parameter $`\lambda `$ denotes the periodicities of the worldsheet fermions. As we stated in Section 2, we will carry out our computations in this section in the original $`N=4`$ supersymmetric Type IIB theory. Thus, we will sum only over uniform periodic and antiperiodic boundary conditions for the world-sheet fermions around each of the two cycles. The vertex operators are placed at
$$w_1=iy_1,w_2=iy_2,w_3=\pi +iy_3,w_4=\pi +iy_4.$$
(77)
We will check the overall normalization of this expression in Section 6.3.
The easiest way to account for the boundary conditions on the worldsheet fermions is to extend their definitions to $`\pi \mathrm{}w2\pi `$. On this extended worldsheet, the fermions are holomorphic, and their possible periodicities and correlators are the same as for a torus with modulus $`it`$. For the worldsheet bosons, the boundary conditions can be described using the method of image charges. For the fields located on the boundary and satisfying Neumann boundary conditions the correlator is the same as that for a torus with modulus $`it`$, with an extra factor of 2 from the image fields. The correlators necessary for our calculation are listed explicitly in Appendix B.
For the computation of this section, we will be interested in the contribution to the amplitude from bosonic closed string states propagating up the cylinder. These states have fermions antiperiodic around the cylinder, that is, in the direction of $`\mathrm{}w`$. Both boundary conditions in the direction of $`\mathrm{}w`$ are needed to enforce the GSO projection . We will refer to the partition functions for the sectors antiperiodic in the imaginary direction and antiperiodic/periodic in the real direction as $`Z_A^A`$/$`Z_P^A`$ and use a similar notation for the correlation functions. In Section 6.3, we will also consider the contribution from bosonic open string states propagating around the cylinder. These states have fermions with boundary conditions antiperiodic in the real direction. The computation will involve the partition functions $`Z_A^A`$/$`Z_A^P`$ and the analogous correlators.
For the cylinder amplitude, the superconformal charges satisfy $`_iq_i=0`$. Thus, we will write all four vertex operators in the 0 picture. We will use the explicit form
$$๐ฑ_0^\mu (k_i)=(i\dot{X}^\mu +\alpha ^{}2k\psi \psi ^\mu )e^{ik_iX}(w_i,\overline{w_i}),$$
(78)
where the dot denotes a derivative along the boundary. Note that this expression is slightly different from (17) in that the $`X`$ field has not been split into holomorphic and antiholomorphic components.
The $`t`$ integration in (76) runs from 0 to $`\mathrm{}`$. However, this domain of integration can be separated into two regions. In the limit of small $`t`$, the cylinder becomes very long and the amplitude is dominated by light closed-string states. In the limit of large $`t`$, the cylinder becomes very narrow and the amplitude is dominated by light open-string states. The separation between these two regions is ambiguous, since only their sum is a well-defined gauge-invariant quantity. We parametrize this ambiguity by the integration cutoff $`t_0`$. Below we will show that the small-$`t`$ region reproduces the graviton exchange amplitudes (71), with $`M_H`$ related to the string scale and $`t_0`$. In this calculation, we will use the small-$`t`$ expansions of the partition functions and correlators. These expressions (given in Appendix B) are valid up to $`t\pi `$. This suggests that the natural value of the cutoff is $`t_0\pi `$. The expression we will derive for $`M_H`$ will depend on $`t_0`$. This simply makes clear that the loop diagrams of string theory also give other contributions to the dimension 8 terms of the effective Lagrangian. The most important point is that all of these contributions are subleading, suppressed by a power of $`g^2`$ relative to the SR contribution (69).
We now describe the evaluation of the graviton-exchange contribution in (76). For the moment, we consider a D$`p`$-brane with $`p`$ arbitrary; later we will specialize to the case $`p=3`$. Using the small-$`t`$ expressions of the partition functions and correlators given in Appendix B, we find the expression
$`๐_{Gs}`$ $`=`$ $`g^4\delta ^{12}\delta ^{34}4^{\alpha ^{}s}2^{(73p)/2}\pi ^{3p}\alpha _{}^{}{}_{}{}^{(7p)/2}{\displaystyle _0^{t_0}}๐tt^{(5p)/2}\mathrm{exp}({\displaystyle \frac{\alpha ^{}s}{2}}{\displaystyle \frac{\pi }{t}})`$ (79)
$`2\left[{\displaystyle \underset{i=1}{\overset{4}{}}}{\displaystyle _0^1}๐Y_i\right](\mathrm{sin}\pi Y_{12})^{\alpha ^{}s}(\mathrm{sin}\pi Y_{34})^{\alpha ^{}s}F(Y_i;ฯต_i,k_i)+\mathrm{\Delta },`$
where $`F`$ is a function of external momenta and kinematics which has no $`t`$ dependence, $`Y_i=y_i/2\pi t`$, $`Y_{ij}=Y_iY_j`$, and $`\mathrm{\Delta }`$ is the contribution to the integral from the large-$`t`$ region. Explicitly, the function $`F`$ is given by
$$F=C_1+C_2,$$
(80)
where
$`C_1`$ $`=`$ $`\left({\displaystyle \frac{1}{2\alpha ^{}}}\right)^2ฯต_1ฯต_2ฯต_3ฯต_4\mathrm{sin}^2\pi Y_{12}\mathrm{sin}^2\pi Y_{34},`$
$`C_2`$ $`=`$ $`k_1k_4(2k_1k_4ฯต_1ฯต_2ฯต_3ฯต_4+2ฯต_1ฯต_2(k_1ฯต_3k_3ฯต_4+ฯต_3k_4k_2ฯต_4)`$ (81)
$`+2ฯต_3ฯต_4(k_1ฯต_2ฯต_1k_3+ฯต_1k_2ฯต_2k_4)).`$
Since we are only interested in the $`s0`$ limit of the amplitude, in (81) we have dropped the terms which do not contribute in this limit.
The small-$`t`$ contribution in (79) factorizes into two integrals, the modulus intergal in the first line and the coordinate integral in the second line. The coordinate integral can be easily evaluated. In this calculation, one encounters two simple integrals,
$$I_1=_0^1๐Y_1_0^1๐Y_2(\mathrm{sin}\pi Y_{12})^{2\alpha ^{}s},$$
(82)
and
$$I_2=_0^1๐Y_1_0^1๐Y_2(\mathrm{sin}\pi Y_{12})^{\alpha ^{}s}.$$
(83)
Evaluating these integrals in the limit $`\alpha ^{}s0`$ yields $`I_1=0`$, $`I_2=1`$. Therefore, in this limit we have
$$2\left[\underset{i=1}{\overset{4}{}}_0^1๐Y_i\right](\mathrm{sin}\pi Y_{12})^{\alpha ^{}s}(\mathrm{sin}\pi Y_{34})^{\alpha ^{}s}F(Y_i;ฯต_i,k_i)=2C_2.$$
(84)
One can show that this expression is identical to the matrix element of the square of the photon energy-momentum tensor, $`T^{\mu \nu }(1,2)T_{\mu \nu }(3,4)`$. This means that in this limit, this process is accurately described by the effective Lagrangian (30). The integral over the modulus $`t`$ then determines the coefficient of this operator.
The modulus integral can be rewritten in a form reminiscent of a massive graviton propagator from field theory. To do this, we change the integration variable to $`v=1/t`$, and use the identity
$$v^{(p9)/2}=\left(\frac{\alpha ^{}}{2}\right)^{(9p)/2}d^{9p}m\mathrm{exp}(\frac{\pi \alpha ^{}m^2}{2}v).$$
(85)
Performing the $`v`$ integration, we find
$$_0^{t_0}๐tt^{(5p)/2}\mathrm{exp}(\frac{\alpha ^{}s}{2}\frac{\pi }{t})=\left(\frac{\alpha ^{}}{2}\right)^{(7p)/2}\frac{1}{\pi }d^{9p}m\frac{1}{sm^2}\mathrm{exp}\left(\frac{\pi \alpha ^{}(sm^2)}{2}v_0\right),$$
(86)
where $`v_0=1/t_0`$. When both $`s`$ and $`m^2`$ are small compared to $`1/\alpha ^{}`$, the integrand in (86) is just the field theory propagator. We have already pointed out that the virtual graviton exchange cannot be analyzed within effective field theory; technically, this results from the divergence of the KK mass integration in the region of high $`m`$. The integral in (86), however, is finite, due to the exponential suppression for $`\alpha ^{}m^21`$. This finite coefficient gives the scale $`M_H`$ in (30).
Evaluating the integral (86) for $`s=0`$ and assembling the pieces, we obtain as the leading term in the low-energy expansion of the small-$`t`$ integral of (79)
$$๐_{Gs}=g^4\delta ^{12}\delta ^{34}2^{(93p)/2}\pi ^{(133p)/2}(\pi v_0)^{(p7)/2}\alpha _{}^{}{}_{}{}^{(7p)/2}\frac{1}{7p}T^{\mu \nu }(1,2)T_{\mu \nu }(3,4)+\mathrm{}.$$
(87)
Now set $`p=3`$. The amplitude (87) can be reproduced by the effective Lagrangian (30), provided that we identify
$$\frac{8}{M_H^4}=g^4\frac{\pi ^2}{4}\frac{1}{M_S^4}(\pi v_0)^2,$$
(88)
and use $`\lambda =+1`$ in (30). As we have explained above, for a numerical estimate we can evaluate this expression with $`v_01/\pi `$. This gives
$$\frac{1}{M_H^4}\frac{\pi ^2}{32}\frac{g^4}{M_S^4}.$$
(89)
As expected, the relation is of the form (31), with an additional suppression from the numerical coefficient on the right-hand side. Substituting this value of $`M_H`$ into (71), we confirm that this contribution is subdominant with respect to the SR exchange amplitude (69).
### 6.3 Normalization
There is another reason that we must analyze the one-loop diagram, and that is to find the relation between the effective Newton constant or the gravity scale $`M`$ and the more fundamental string theory parameters $`g`$ and $`\alpha ^{}`$. We have already quoted this relation in (2). In this section, we will give the derivation. Once again, our analysis will be done for the toy case of an $`N=4`$ supersymmetric D-brane theory.
Our procedure is illustrated in Figure 11. We will first take the $`t\mathrm{}`$ limit of the cylinder and relate this to a loop diagram of Yang-Mills field theory. This will determine the normalization of the diagram. Then we will take the $`t0`$ limit to identify the graviton exchange. In this section, we will give what we consider the shortest route through this analysis, considering a two-point function in the first part of the calculation rather than a four-point function, and, in the second part, considering only one fairly simple structure in the gravitation interaction.
We thus consider first the $`t\mathrm{}`$ limit. In principle, we should study the four-point loop diagram. However, it is simpler to analyze the two-point function. The normalizations of these diagrams are related by considering the limit $`(k_1+k_2)^20`$, in which pairs of vertex operators factorize into single vertex operator insertions as shown in Figure 4. Through this relation, the normalization of (76) is equivalent to the following normalization of the planar two-point loop amplitude shown in Figure 11(a):
$$๐_2=g^2\text{tr}[t^1t^2]\text{tr}[1]\frac{dt}{t}\left[\underset{i=1}{\overset{2}{}}_0^{2\pi t}๐y_i\right]Z_x^p\underset{\lambda }{}Z_\lambda \underset{i=1}{\overset{2}{}}ฯต_i๐ฑ_0(w_i,k_i)_\lambda ,$$
(90)
where the notations are as in (76) and the two vertex operators are placed at $`w_1`$ and $`w_2`$ in (77).
It is simplest to concentrate on the structure
$$ฯต_1k_2ฯต_2k_1.$$
(91)
Looking back to the form (78), we see that this structure arises in two ways in the contraction of vertex operators, from the contraction of the two factors $`\dot{X}`$ with factors $`kX`$ in the exponentials, and from a contraction of the fermionic terms with one another. The correlators for $`X`$ and $`\psi `$ should be taken in the limit $`t\mathrm{}`$; the appropriate expressions are given in (128). In the two sectors corresponding to bosonic open string states, these terms give
$`\mathrm{\Pi }ฯต๐ฑ_A^A`$ $``$ $`ฯต_1k_2ฯต_2k_1\left[\alpha _{}^{}{}_{}{}^{2}(12Y)^24\alpha _{}^{}{}_{}{}^{2}(e^{\pi tY}+e^{\pi t(1Y)})^2\right],`$
$`\mathrm{\Pi }ฯต๐ฑ_A^P`$ $``$ $`ฯต_1k_2ฯต_2k_1\left[\alpha _{}^{}{}_{}{}^{2}(12Y)^24\alpha _{}^{}{}_{}{}^{2}(e^{\pi tY}e^{\pi t(1Y)})^2\right],`$ (92)
where $`Y=Y_{12}`$ and, for clarity, we have left off the expectation value of the exponentials. Restoring this factor, including the partition functions from (112), and making the cancellations between the two sectors, we find
$`๐_2`$ $`=`$ $`g^2N_C\delta ^{12}{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{dt}{t}}{\displaystyle \frac{64\pi ^2t^2\alpha _{}^{}{}_{}{}^{2}}{(8\pi ^2\alpha ^{}t)^{d/2}}}`$ (93)
$`{\displaystyle _0^1}dYฯต_1k_2ฯต_2k_1[(12Y)^21]\mathrm{exp}[\alpha ^{}k_1k_2(2\pi t)Y(1Y)],`$
where we have replaced $`(p+1)=d`$ and $`\text{tr}[1]=N_C`$. Now do the $`t`$ integral. For $`d`$ close to 4, we obtain,
$`๐_2`$ $`=`$ $`g^2N_C\delta ^{12}{\displaystyle \frac{8}{(4\pi )^2}}\mathrm{\Gamma }\left(2{\displaystyle \frac{d}{2}}\right)ฯต_1k_2ฯต_2k_1{\displaystyle _0^1}๐Y\left[(12Y)^21\right].`$ (94)
As Kaplunovsky pointed out for the analogous closed string calculation, this result can be matched to the computation of the one-loop two-point diagram in Yang-Mills theory in the background-field gauge. The required expressions are given in . The value of this diagram given there, summed over the bosonic content of the $`N=4`$ supersymmetric Yang-Mills theory (1 vector and 6 scalars), is
$$g^2N_C\delta ^{12}\frac{1}{(4\pi )^2}\mathrm{\Gamma }\left(2\frac{d}{2}\right)(ฯต_1ฯต_2k_1k_2ฯต_1k_2ฯต_2k_1)_0^1๐Y\left[8(12Y)^28\right].$$
(95)
In this expression, $`Y`$ is the Feynman parameter. The first term in the bracket comes from a spin-independent determinant, the second term from the spin operator. The expressions (94) and (95) match. Thus, the normalization assumed in (90) and in (76) is correct.
Now we turn to the $`t0`$ limit. Here it is simplest to extract the graviton exchange by considering the limit of high-energy scattering with low momentum transfer. That is, we set
$$k_2k_1,k_4k_3.$$
(96)
Then the usual graviton exchange diagram in four dimensions contains a term
$$๐=8\pi G_N\delta ^{12}\delta ^{34}(2k_1^\mu k_1^\nu )\frac{1}{s}(2k_{3\mu }k_{3\nu })=8\pi G_N\frac{t^2}{s}\delta ^{12}\delta ^{34},$$
(97)
where $`s=(k_1+k_2)^2=(k_3+k_4)^2`$ and $`t=(k_1+k_3)^2`$.
In the scattering amplitude of four vector bosons, this term has the structure
$$ฯต_1ฯต_2ฯต_3ฯต_4k_1k_3k_1k_3,$$
(98)
using (96) to replace $`k_2`$ and $`k_4`$.
After close examination of the various terms contributing to (76), one can see that, after the cancellation between the $`Z_A^A`$ and $`Z_P^A`$ sectors, there is only one source for a term of this structure. That is the contribution in which one takes only the fermionic term in each vertex operator (78) and contracts the $`ฯต\psi `$ operators on the same side of the cylinder and the $`k\psi `$ operators across the cylinder. The correlators needd are given in (122) and (127). There are two contractions of this type for each sector. When these two terms are added, all dependence on the $`Y_{ij}`$ cancels out. The contributions from the two sectors then add constructively. The sum of these terms gives
$$๐_{Gs}=(ฯต_1ฯต_2ฯต_3ฯต_4\delta ^{12}\delta ^{34}t^2)g^4\frac{2(2\pi \alpha ^{})^4}{(8\pi ^2\alpha ^{})^{(p+1)/2}}_0^{\mathrm{}}๐tt^{(5p)/2}\left[\frac{1}{2}e^{\pi /4t}\right]^{2\alpha ^{}s}.$$
(99)
One should be careful to note that the $`t`$ in the prefactor is the Mandelstam invariant, whereas the other factors $`t`$ represent the modulus of the cylinder.
This expression can be simplified by changing variables from $`t`$ to $`v=1/t`$ and then introducing the variable $`m`$ as in (85). Setting also $`p=3`$, we arrive at the expression
$$๐_{Gs}=(ฯต_1ฯต_2ฯต_3ฯต_4\delta ^{12}\delta ^{34}t^2)g^4\frac{\alpha _{}^{}{}_{}{}^{4}}{8\pi }d^6m\frac{1}{m^2s}.$$
(100)
We can convert the integral over $`m`$ to a discrete sum over KK states in the 6 large extra dimensions of periodicity $`2\pi R`$ by using the relation
$$R^6d^6m=\underset{m}{}.$$
(101)
Finally, we may pick off the term in the sum that corresponds to the massless graviton in four dimensions. We then identify
$$8\pi G_N=\frac{1}{8\pi }\alpha _{}^{}{}_{}{}^{4}g^4R^6.$$
(102)
Replacing $`G_N`$ with the fundamental quantum gravity scale $`M`$ according to (1), we find
$$M^8=\pi \alpha ^2\alpha _{}^{}{}_{}{}^{4},$$
(103)
which is equivalent to the promised relation (2).
## 7 Experimental constraints on the quantum gravity scale
It is useful to compare the constraints on the large extra dimension scenario that we have obtained in this paper through model-dependent string effects to more robust, model-independent constraints. In the introduction, we noted that previous constraints on large extra dimensions have come from two sources, searches for missing energy due to gravitation radiation into the extra dimensions, and searches for contact interactions due to KK graviton exchange. It has become clear in this paper that the possible contact interactions are model-dependent and may not be of purely gravitational origin. So the truly model-independent constraints come only from missing-energy experiments.
In Table 1, we summarize the most important present and future constraints on the quantum gravity scale $`M`$ from missing-energy searches. This table updates the table presented in and improves upon it in several important respects.
The first line of Table 1 gives the constraints obtained in from the consistency of the observed neutrino flux from the supernova SN1987A with the predictions of the stellar collapse models. This analysis puts an upper bound on the rate of energy loss through graviton emission. There exist some strong astrophysical bounds on the scale of quantum gravityโfor example, โbut these depend on assumptions about the cosmological scenario. The constraint from the supernova is different in character. Since we have a reasonable understanding of the composition of a supernova and of the conditions inside its core during collapse, it is possible to calculate the gravitional radiation expected in this process in an unambiguous way. The typical energy of the emitted gravitons is well below a TeV, and so the emission rate calculation uses only the model-independent low-energy limit of the gravitational coupling. It is argued in that, though there are uncertainties in the parameters of the supernova core, the bounds quoted should be accurate to better than a factor of 2. The bound for the case of two large extra dimensions ($`n=2`$) is surprisingly strong and must be taken seriously. We note that the values given in the remaining lines of the table are more precise 95% confidence exclusion limits available from accelerator experiments.
The second line of the table gives the constraints arising from the process $`e^+e^{}\gamma +`$ (missing) which have been announced by the ALEPH collaboration . Similar constraints on anomalous single photon production have been announced by the other LEP experiments .
The third line of the table is derived from a new search for events with one jet and missing $`E_T`$ presented by the CDF collaboration in . Of the five cuts on missing $`E_T`$ presented in this analysis, the analysis based on the cut $`E_T>200`$ GeV turns out to give the best sensitivity. We have applied the formulae in to convert the limit on the cross section to the quoted bounds on $`M`$. Note that these bounds are very close to the estimates in the โFuture Tevatronโ line of .
The fourth line of the table gives the reach of a 1 TeV $`e^+e^{}`$ linear collider as computed in . However, in the fifth line, the constraints given in Table 1 for the LHC are much stronger. This is the result of the observation, made in , that at the LHC there is a dramatic improvement in signal/background if one makes a very hard $`E_T`$ cut. It is advantageous to move this cut to as high a value as the statistics permit. The results shown here correspond to the analysis in applied to a cut at $`E_T>1000`$ GeV.
For the LHC search, one may worry that the effective field theory used to obtain the bounds in Table 1 breaks down for the collisions of the most energetic partons. In Section 5, we have derived the form factor which describes the modification of the cross sections at high energies due to string theory effects. We have shown that at very high energies, this form factor leads to exponential suppression of the signal cross section. One might expect that the sensitivity of the LHC searches will be somewhat lowered by this effect. However, it turns out that for values of the string scale in the few-TeV range, this effect does not significantly alter the signal rates at LHC. In fact, we find a relatively small effect for typical parton-parton center-of-mass energies and a dramatic enhancement when partons can combine to the SR resonances, due to processes analogous to (68) with an excited gluon or quark intermediate state. In the situation in which these states are present, they would also be seen as resonances in the two-jet invariant mass distribution. We conclude that in either case, whether the resonances are observed or not, the bounds in the last line of Table 1 would not be significantly decreased by stringy physics.
## 8 Conclusions
In this paper, we have studied the phenomenology of large extra dimensions for the situation in which quantum gravity is represented by a weakly-coupled string theory. We have found that, in this case, the signatures of large extra dimensions which have been considered in the literature up to now are overshadowed by genuine string effects. The first sign of new physics is found in string corrections to Standard Model two-body scattering cross sections, leading to contact interactions due to string resonances and to the dramatic appearance of these resonances at colliders. The fact that these resonances have not yet been observed allows us to put a lower bound on the string scale of about 1 TeV. The corresponding limit on the quantum gravity scale, $`M>1.6`$ TeV, is much stronger than that of any current accelerator experiment. The next generation of colliders should probe values of the string scale up to 5 TeV and values of the quantum gravity scale above 8 TeV.
The motivation for the idea of large extra dimensions in the work of Arkani-Hamed, Dimopoulos, and Dvali came from the possibility of a natural relation between the weak interaction scale and the scale of quantum gravity. If this possibility is indeed realized, the linear collider and the LHC will carry out experimental measurements of string physics. For many years, physicists have thought of strings as tiny objects and imagined that we could observe them in experiments only in some distant era. It seems now that this era could be close at hand.
ACKNOWLEDGEMENTS
We are grateful to Nima Arkani-Hamed for stimulating this investigation, and to Nicolas Toumbas, who collaborated with us in the early stages of this work. We thank Dimitri Bourilkov for a very useful correspondence concerning the LEP 2 Bhabha scattering data. We also thank Tom Banks, Hooman Davoudiasl, Savas Dimopoulos, Lance Dixon, Ian Hinchliffe, Ann Nelson, and Alex Pomarol for helpful discussions and the Institute for Theoretical Physics at UC Santa Barbara for hospitality. The work of SC was supported in part by the US National Science Foundation under contract PHYโ9870115; the work of MP and MEP was supported by the US Department of Energy under contract DEโAC03โ76SF00515.
## Appendix A Reference formulae for models of contact interactions
In this appendix, we give the explicit expressions for the contact-interaction corrections to Bhabha scattering that are compared in Figures 5 and 6. We also give the first contact-interaction corrections to the $`e^+e^{}\gamma \gamma `$ and $`\gamma \gamma \gamma \gamma `$ amplitudes.
The unpolarized cross section formula for Bhabha scattering can be written in the form
$$\frac{d\sigma }{d\mathrm{cos}\theta }=\frac{\pi \alpha ^2}{2s}\left[u^2(|A_{LL}|^2+|A_{RR}|^2)+2t^2|A_{RL,s}|^2+2s^2|A_{RL,t}|^2\right],$$
(104)
where
$`A_{LL}`$ $`=`$ $`{\displaystyle \frac{1}{s}}+{\displaystyle \frac{1}{t}}+{\displaystyle \frac{(\frac{1}{2}\mathrm{sin}^2\theta _w)^2}{\mathrm{sin}^2\theta _w\mathrm{cos}^2\theta _w}}\left({\displaystyle \frac{1}{sm_Z^2}}+{\displaystyle \frac{1}{tm_Z^2}}\right)+\mathrm{\Delta }_{LL}`$
$`A_{RR}`$ $`=`$ $`{\displaystyle \frac{1}{s}}+{\displaystyle \frac{1}{t}}+{\displaystyle \frac{\mathrm{sin}^2\theta _w}{\mathrm{cos}^2\theta _w}}\left({\displaystyle \frac{1}{sm_Z^2}}+{\displaystyle \frac{1}{tm_Z^2}}\right)+\mathrm{\Delta }_{RR}`$
$`A_{RL,s}`$ $`=`$ $`{\displaystyle \frac{1}{s}}{\displaystyle \frac{(\frac{1}{2}\mathrm{sin}^2\theta _w)}{\mathrm{cos}^2\theta _w}}{\displaystyle \frac{1}{sm_Z^2}}+\mathrm{\Delta }_{RL,s}`$
$`A_{RL,t}`$ $`=`$ $`{\displaystyle \frac{1}{t}}{\displaystyle \frac{(\frac{1}{2}\mathrm{sin}^2\theta _w)}{\mathrm{cos}^2\theta _w}}{\displaystyle \frac{1}{tm_Z^2}}+\mathrm{\Delta }_{RL,t}.`$ (105)
For KK graviton exchange parametrized by (30) ,
$`\mathrm{\Delta }_{LL}=\mathrm{\Delta }_{RR}`$ $`=`$ $`{\displaystyle \frac{\lambda }{\pi \alpha M_H^4}}\left[(u+{\displaystyle \frac{3}{4}}s)+(u+{\displaystyle \frac{3}{4}}t)\right]`$
$`\mathrm{\Delta }_{RL,s}`$ $`=`$ $`{\displaystyle \frac{\lambda }{\pi \alpha M_H^4}}(t+{\displaystyle \frac{3}{4}}s)`$
$`\mathrm{\Delta }_{RL,s}`$ $`=`$ $`{\displaystyle \frac{\lambda }{\pi \alpha M_H^4}}(s+{\displaystyle \frac{3}{4}}t).`$ (106)
For standard dimension-6 contact interactions ,
$`\mathrm{\Delta }_{LL}`$ $`=`$ $`2{\displaystyle \frac{\eta _{LL}}{\alpha \mathrm{\Lambda }^2}}`$
$`\mathrm{\Delta }_{RR}`$ $`=`$ $`2{\displaystyle \frac{\eta _{RR}}{\alpha \mathrm{\Lambda }^2}}`$
$`\mathrm{\Delta }_{RL,s}=\mathrm{\Delta }_{RL,t}`$ $`=`$ $`{\displaystyle \frac{\eta _{RL}}{\alpha \mathrm{\Lambda }^2}}.`$ (107)
The VV case corresponds to $`\eta _{LL}=\eta _{RR}=\eta _{RL}=\pm 1`$. The AA case corresponds to $`\eta _{LL}=\eta _{RR}=\eta _{RL}=\pm 1`$.
For the string model described in Sections 2 and 3, the corrections are more easily described by (28).
The expressions above are written in such a way that they can easily be pulled apart into cross sections for definite helicity initial and final states. At a high-energy linear collider with a polarized $`e^{}`$ beam, it is possible to resolve ambiguities in the relative contributions of the various $`\mathrm{\Delta }_i`$.
For completeness, we note also that the amplitude for $`e^+e^{}\gamma \gamma `$, which is given by (25) in our string model, takes the following form with KK graviton exchange parametrized by (30) :
$$๐(e_L^{}e_R^+\gamma _L\gamma _R)=2e^2\sqrt{\frac{u}{t}}\left[1+\frac{\lambda }{\pi \alpha M_H^4}ut\right].$$
(108)
Thus, in this model, we may identify Drellโs $`\mathrm{\Lambda }_\pm `$ parameter as
$$\mathrm{\Lambda }_\lambda =(\pi \alpha )^{1/4}M_H0.39M_H.$$
(109)
## Appendix B Ingredients needed for the one-loop calculation in Section 6
The partition functions for the cylinder with modulus $`it`$, with fermion periodicities required for our calculation in Section 6, are:
$`Z_x^p`$ $`=`$ $`(8\pi ^2\alpha ^{}t)^{(p+1)/2}\eta (it)^8;`$
$`Z_A^A`$ $`=`$ $`\left({\displaystyle \frac{\vartheta _{00}(0it)}{\eta (it)}}\right)^4;`$
$`Z_P^A`$ $`=`$ $`\left({\displaystyle \frac{\vartheta _{10}(0it)}{\eta (it)}}\right)^4;`$
$`Z_A^P`$ $`=`$ $`\left({\displaystyle \frac{\vartheta _{01}(0it)}{\eta (it)}}\right)^4.`$ (110)
Note that the zero-mode integration in the bosonic partition function, $`Z_x`$, was performed only in the directions transverse to the brane. It turns out that this is the only place in the calculation which depends on $`p`$. The small-$`t`$ expansions of the partition functions which we will use for the calculation in Section 6.2 are
$`Z_x^p`$ $`=`$ $`(8\pi ^2\alpha ^{})^{(p+1)/2}t^{(7p)/2}e^{2\pi /3t}+\mathrm{};`$
$`Z_A^A`$ $`=`$ $`e^{\pi /3t}(1+8e^{\pi /t}+\mathrm{});`$
$`Z_P^A`$ $`=`$ $`e^{\pi /3t}(18e^{\pi /t}+\mathrm{}).`$ (111)
In the calculation in Section 6.3, we will make use of the following large-$`t`$ expansions:
$`Z_x^p`$ $`=`$ $`(8\pi ^2\alpha ^{}t)^{(p+1)/2}e^{2\pi t/3}+\mathrm{};`$
$`Z_A^A`$ $`=`$ $`e^{\pi t/3}(1+8e^{\pi t}+\mathrm{});`$
$`Z_A^P`$ $`=`$ $`e^{\pi t/3}(18e^{\pi t}+\mathrm{}).`$ (112)
Here and below, we only keep the leading terms in the expansions of bosonic partition functions and correlators. For fermionic quantities, we keep the first subleading corrections, since in some cases the leading terms cancel after different sectors are combined.
We will also need the following correlation functions (all of them are understood to exclude the corresponding partition function):
$`X^\mu (w_i)X^\nu (w_j)`$ $`=`$ $`g^{\mu \nu }(\alpha ^{}\mathrm{log}|\vartheta _{11}({\displaystyle \frac{w_{ij}}{2\pi }}it)|^2+\alpha ^{}{\displaystyle \frac{(\mathrm{}w_{ij})^2}{2\pi t}});`$
$`\psi ^\mu (w_i)\psi ^\nu (w_j)_A^A`$ $`=`$ $`{\displaystyle \frac{g^{\mu \nu }}{2\pi }}{\displaystyle \frac{\vartheta _{00}\left(\frac{w_{ij}}{2\pi }it\right)}{\vartheta _{11}\left(\frac{w_{ij}}{2\pi }it\right)}}{\displaystyle \frac{_\nu \vartheta _{11}(0it)}{\vartheta _{00}(0it)}};`$
$`\psi ^\mu (w_i)\psi ^\nu (w_j)_P^A`$ $`=`$ $`{\displaystyle \frac{g^{\mu \nu }}{2\pi }}{\displaystyle \frac{\vartheta _{10}\left(\frac{w_{ij}}{2\pi }it\right)}{\vartheta _{11}\left(\frac{w_{ij}}{2\pi }it\right)}}{\displaystyle \frac{_\nu \vartheta _{11}(0it)}{\vartheta _{10}(0it)}};`$
$`\psi ^\mu (w_i)\psi ^\nu (w_j)_A^P`$ $`=`$ $`{\displaystyle \frac{g^{\mu \nu }}{2\pi }}{\displaystyle \frac{\vartheta _{01}\left(\frac{w_{ij}}{2\pi }it\right)}{\vartheta _{11}\left(\frac{w_{ij}}{2\pi }it\right)}}{\displaystyle \frac{_\nu \vartheta _{11}(0it)}{\vartheta _{01}(0it)}},`$ (113)
where $`w_{ij}=w_iw_j.`$ The fermionic correlators here are just the same as for a torus with modulus $`it`$; they are valid for arbitrary $`w_i`$โs. On the other hand, the bosonic correlator in the first line is only valid for the fields that are placed on the boundary and satisfy Neumann boundary conditions. It differs from a torus correlator by a factor of 2, which correctly takes into account the image charges in this case. This correlator is sufficient for our present calculation.
The small-$`t`$ expansions of the correlators (113) depend on whether the two fields are on the same side of the cylinder or not. We can write $`w_{ij}=\pi \mathrm{\Delta }_{ij}+2\pi iy_{ij}`$, where $`y_{ij}=y_iy_j`$, and $`\mathrm{\Delta }_{ij}=0`$ if $`i`$ and $`j`$ are on the same side of the cylinder, and 1 otherwise (this assumes, without loss of generality, that $`i>j`$.) The small-$`t`$ expansions for the case of $`\mathrm{\Delta }_{ij}=0`$ are,
$`X^\mu (w_i)X^\nu (w_j)`$ $`=`$ $`g^{\mu \nu }\alpha ^{}\left({\displaystyle \frac{\pi }{2t}}2\mathrm{log}2+\mathrm{log}t2\mathrm{log}\mathrm{sin}\pi Y_{ij}\right)+\mathrm{};`$ (114)
$`\dot{X}^\mu (w_i)X^\nu (w_j)`$ $`=`$ $`ig^{\mu \nu }{\displaystyle \frac{\alpha ^{}}{t}}\mathrm{cot}\pi Y_{ij}+\mathrm{};`$ (116)
$`\dot{X}^\mu (w_i)\dot{X}^\nu (w_j)`$ $`=`$ $`g^{\mu \nu }{\displaystyle \frac{\alpha ^{}}{2t^2}}{\displaystyle \frac{1}{\mathrm{sin}^2\pi Y_{ij}}}+\mathrm{};`$ (118)
$`\psi ^\mu (w_i)\psi ^\nu (w_j)_A^A`$ $`=`$ $`ig^{\mu \nu }{\displaystyle \frac{1}{2t}}{\displaystyle \frac{1}{\mathrm{sin}\pi Y_{ij}}}\left(14e^{\pi /t}\mathrm{sin}^2\pi Y_{ij}+\mathrm{}\right);`$ (120)
$`\psi ^\mu (w_i)\psi ^\nu (w_j)_P^A`$ $`=`$ $`ig^{\mu \nu }{\displaystyle \frac{1}{2t}}{\displaystyle \frac{1}{\mathrm{sin}\pi Y_{ij}}}\left(1+4e^{\pi /t}\mathrm{sin}^2\pi Y_{ij}+\mathrm{}\right),`$ (122)
where $`Y_{ij}=y_{ij}/t`$. For the case of $`\mathrm{\Delta }_{ij}=1`$ we get:
$`X^\mu (w_i)X^\nu (w_j)`$ $`=`$ $`g^{\mu \nu }\alpha ^{}\mathrm{log}t+\mathrm{};`$ (123)
$`\psi ^\mu (w_i)\psi ^\nu (w_j)_A^A`$ $`=`$ $`g^{\mu \nu }{\displaystyle \frac{2}{t}}e^{\pi /2t}\mathrm{cos}\pi Y_{ij}+\mathrm{};`$ (125)
$`\psi ^\mu (w_i)\psi ^\nu (w_j)_P^A`$ $`=`$ $`ig^{\mu \nu }{\displaystyle \frac{2}{t}}e^{\pi /2t}\mathrm{sin}\pi Y_{ij}+\mathrm{}`$ (127)
The other two correlators, $`<\dot{X}X>`$ and $`<\dot{X}\dot{X}>`$, are in this case suppressed by $`e^{\pi /t}`$ and do not play a role.
For the calculation in Section 6.3, we need the large-$`t`$ expansions of the correlators (113), with the fields on the same side of the cylinder. These are given by
$`X^\mu (w_i)X^\nu (w_j)`$ $`=`$ $`2\pi t\alpha ^{}g^{\mu \nu }Y_{ij}(1Y_{ij});`$
$`\dot{X}^\mu (w_i)X^\nu (w_j)`$ $`=`$ $`i\alpha ^{}g^{\mu \nu }(12Y_{ij});`$
$`\dot{X}^\mu (w_i)\dot{X}^\nu (w_j)`$ $`=`$ $`g^{\mu \nu }{\displaystyle \frac{\alpha ^{}}{\pi t}};`$
$`\psi ^\mu (w_i)\psi ^\nu (w_j)_A^A`$ $`=`$ $`ig^{\mu \nu }\left(e^{\pi tY_{ij}}+e^{\pi t(1Y_{ij})}\right);`$
$`\psi ^\mu (w_i)\psi ^\nu (w_j)_A^P`$ $`=`$ $`ig^{\mu \nu }\left(e^{\pi tY_{ij}}e^{\pi t(1Y_{ij})}\right).`$ (128) |
warning/0001/math0001117.html | ar5iv | text | # Untitled Document
WEIGHTED TRACES ON ALGEBRAS OF PSEUDO-DIFFERENTIAL OPERATORS
AND GEOMETRY ON LOOP GROUPS
January 2000
A.CARDONA\*, C.DUCOURTIOUX, J.P.MAGNOT, S.PAYCHA
Laboratoire de Mathรฉmatiques Appliquรฉes
Universitรฉ Blaise Pascal (Clermont II)
Complexe Universitaire des Cรฉzeaux
63177 Aubiรจre Cedex
(\*) and Universidad de Los Andes, Bogota, Colombia
cardona@ucfma.univ-bpclermont.fr
c.ducour@ucfma.univ-bpclermont.fr
magnot@ucfma.univ-bpclermont.fr
paycha@ucfma.univ-bpclermont.fr
Acknowledgements
The last author would like to thank Jรผrgen Jost for inviting her for a three months stay at the Max Planck Institut in den Naturwissenschaften during the fall of 1998 when a big part of this paper was written. It was completed during a stay at the Mathematics Institute of the University of Bonn supported by the von Humboldt Stiftung.
Abstract
Using weighted traces which are linear functionals of the type
$$Atr^Q(A):=\left(tr(AQ^z)z^1tr(AQ^z)\right)_{z=0}$$
defined on the whole algebra of (classical) pseudo-differential operators (P.D.O.s) and where $`Q`$ is some positive invertible elliptic operator, we investigate the geometry of loop groups in the light of the cohomology of pseudo-differential operators. We set up a geometric framework to study a class of infinite dimensional manifolds in which we recover some results on the geometry of loop groups, using again weighted traces. Along the way, we investigate properties of extensions of the Radul and Schwinger cocycles defined with the help of weighted traces.
Rรฉsumรฉ
A lโaide de traces pondรฉrรฉes qui sont des fonctionnelles linรฉaires du type:
$$Atr^Q(A):=\left(tr(AQ^z)z^1tr(AQ^z)\right)_{z=0}$$
dรฉfinies sur toute lโalgรจbre des opรฉrateurs pseudo-diffรฉrentiels classiques, $`Q`$ รฉtant un opรฉrateur elliptique inversible, on รฉtudie la gรฉomรฉtrie de lโespace des lacets ร la lumiรจre de la cohomologie des opรฉrateurs pseudo-diffรฉrentiels. On met en place un cadre gรฉomรฉtrique afin dโรฉtudier une classe de variรฉtรฉs de dimension infinie, cadre dans lequel on retrouve, toujours ร lโaide des traces pondรฉrรฉes, des rรฉsultats concernant la gรฉomรฉtrie des lacets. Ces traces pondรฉrรฉes nous permettent aussi dโรฉtendre la notion de cocycle de Radul et de Schwinger et dโen รฉtudier certaines propriรฉtรฉs.
Introduction
The paper is built up from two parts: the first one presents algebraic tools which are used in the second part to extend some geometric concepts to the infinite dimensional context. The approach to the geometry of loop groups we present in the second part uses weighted traces and it is new to our knowledge. Weighted traces offer a useful tool to prove both algebraic and geometric results, some of which had been proved elsewhere by other methods. Let us describe the contents of the first part of the paper (sections 1-4). The Lie algebra of interest in our context is the infinite dimensional Lie algebra of pseudo-differential operators (P.D.Os) acting on sections of some finite rank vector bundle $`E`$ based on a closed manifold $`M`$ (its dimension does not yet play a role at this stage). The Lie bracket is given by the operator bracket. It is well known that when the manifold is connected and of dimension strictly larger than $`1`$, the only trace on this algebra, i.e the only $`\mathrm{IC}`$ valued linear functional which satisfies the tracial property namely $`tr[A,B]=0`$, is the Wodzicki residue trace \[W\], see also \[K\] for a review. Here we shall consider linear functionals on this algebra which arise as zeta function regularized traces . They involve a weight given by a positive self-adjoint elliptic operator and in general depend on the choice of the weight hence the terminology weighted traces which we shall use here. One can find such โtracesโ (sometimes implicitely) in the literature on determinant bundles \[BF\] (in particular in the connection and the curvature) and more generally when investigating geometry in infinite dimensions (in particular for the notion of minimality of submanifolds \[MRT\],\[AP\]). Here we will be using them to define the Ricci curvature on current groups and first Chern form on loop groups. These weighted traces extend the so called canonical trace of Kontsevich and Vishik \[KV1\] defined on the subalgebra of P.D.Os with integer order that lie in the odd class (defined below) provided the dimension of the underlying manifold is odd. On this subalgebra (which contains ordinary differential operators) the weighted traces actually obey the tracial property $`tr[A,B]=0`$ which is not the case on the whole algebra of P.D.Os. Here, rather than searching for subalgebras on which the โweighted tracesโ actually are traces, we focus on the obstruction that prevents them from being traces on a bigger subalgebra of P.D.Os. It is measured by the coboundary of the weighted trace and yields a generalization of the well known Radul cocycle (see \[KK\], \[R\], \[M2\]). It is of infinite dimensional essence since it can be expressed in terms of a Wodzicki residue which is a purely infinite dimensional trace. With this idea in mind of studying infinite dimensional obstructions rather than restricting ourselves to subalgebras on which they vanish, using weighted traces we build up bilinear functionals on the whole algebra of classical P.D.Os. These restrict to the twisted Radul \[M2\] and Schwinger \[S\], \[M2\], \[CFNW\] cocycles on the (rather small) subalgebra of P.D.Os that lie in $`g_{res}`$, the algebra of bounded operators $`A`$ such that $`[A,ฯต]`$ is of order no larger than $`\frac{dimM}{2}`$ where $`ฯต`$ is the โsignโ of some self-adjoint elliptic operator. Finally, we investigate the link between our Schwinger functional and a generalization of another cocycle arising in the context of a central extension of the group $`G_{res}`$ of invertible operators in $`g_{res}`$. These relations boil down to well-known \[PS\] identifications of these cocycles on the restriction to $`g_{res}`$ for the obstruction to such identifications which is given in terms of Radul cocycles vanishing on this subalgebra. Let us now turn to a more geometric point of view which leads us to the second part of the paper (sections 5-9). We now consider families of pseudo-differential operators, extending the notion of weighted trace to such families. On top of the algebraic obstructions mentioned above, due to the dependence on the weight, there are obstructions of geometric type which arise when trying to generalize properties of classical geometric concepts to infinite dimensions. Current groups $`Map(M,G)`$ where $`M`$ is a manifold and $`G`$ is a Lie group, offer a first tractable example of infinite dimensional manifold. As spaces of classical paths in Wess-Zumino Witten models, they play an important role in quantum field theories. They have been the topic of many an investigation and particularly from a geometric point of view (e.g. \[DL\], \[F1,2\], \[P\], \[PS\], \[SP\], \[Wu\]). In a pioneering article \[F1\], Freed suggested a way to geeneralize to these groups some of the methods available to study finite dimensional Lie groups. In particular, he defined a Ricci curvature on current groups and a first Chern form on the $`H^{\frac{1}{2}}`$ based loop group which is Kรคhler. Other methods \[SW\] have since then been suggested, leading to the same expressions. In \[F1\] (see also \[F2\], \[SW\]) the author uses a โconditionedโ trace which involves taking a โtwo step traceโ, namely first the trace on the Lie algebra of the finite dimensional Lie group $`G`$ and then, when the operator obtained this way is trace-class, taking its trace. He shows that the (conditioned) first Chern form on the based loop group is proportional to the symplectic form, from which follows on one hand that it is closed and hence defines the first Chern class, and on the other hand that it is Kรคhler-Einstein. Here we suggest a more general approach to defining Ricci curvature and first Chern forms on a class of manifolds which includes the current groups mentioned above. Carried out (in a left invariant way ) to current Lie groups equipped with a weight given by a left invariant field of elliptic operators, the notion of weighted trace enables us to define the weighted Ricci curvature as a weighted pseudo-trace of $`R:Z\mathrm{\Omega }(Z,)`$ where $`\mathrm{\Omega }`$ is the curvature tensor and the weighted first Chern form as a weighted pseudo-trace of the curvature, provided the operators involved are classical P.D.Os. We express the weighted first Chern form in terms of a pull-back by the adjoint representation of a cocycle on $`g_{res}`$, thus relating the closed two form given by the Chern form with a closed two cochain. A well known result by Kuiper \[Ku\] shows that a Hilbert manifold is parallelisable since $`GL(H)`$ is contractible for a Hilbert space $`H`$ (the model space of the manifold). This might seem in conflict with the fact that the first Chern class does not vanish. But in our approach, the model space $`H`$ being a space of sections of some (finite rank) vector bundle $`E`$ based on a closed manifold $`M`$, the structure group $`GL(H)`$ reduces to a non contractible group, namely the group $`Ell_0^{}(M,E)`$ of zero order invertible elliptic operators acting on these sections. This opens a road to many questions, such as finding a criteria for the weighted first Chern form and higher order forms to be closed, investigating the holonomy group which we expect to be non trivial because of the reasons mentioned above. This article yields a geometric setting in which such questions make sense for a class of Hilbert manifolds beyong the example of loop groups. It confronts this approach with other approaches in the specific case of loop groups.
1. Weighted traces on the algebra of (classical) P.D.Os In \[KV 1\], the authors introduced a new trace type functional $`TR`$ on operators in $`PDO(M,E)`$ with order in $`\alpha _0+\mathrm{Z}\mathrm{Z}`$ where $`\alpha _0`$ is some non integer complex number called the canonical trace. Avoiding integer orders has to do with the fact that a positive homogeneous distribution on $`\mathrm{IR}^n/\{0\}`$ of non integer order can be extended in a unique way to a positive homogeneous distribution on $`\mathrm{IR}^n`$ \[H\]. In this section we first briefly recall the construction of this trace $`TR`$ and some of its properties. We use this trace as a tool to build linear functionals of pseudo-differential operators which we call โweighted tracesโ and to study some of their properties. Such functionals were also considered in \[MN\] to prove a pseudo-differential generalization of the Atiyah-Patodi-Singer index theorem. For the sake of self-containedness, we recall the proof of this fact. We refer the reader to Appendix A for notations and basic facts concerning pseudo-differential operators. $``$ The Kontsevich and Vishik canonical trace Let us first describe the general lines of the construction in a heuristic way. For a (classical) pseudo-differential operator locally given by:
$$Au(x):=_{\mathrm{IR}^n}a(x,\xi )\widehat{u}(\xi )๐\xi $$
where $`\sigma _A(x,\xi )`$ is the (locally defined) total symbol (see Appendix A), we would like to define
$$\mathrm{"}TR(A):=_M_{\mathrm{IR}^n}tr_x\sigma _A(x,\xi )๐\xi ๐vol(x)\mathrm{"}$$
which in general does not make sense since $`a(x,\xi )`$ typically has components of degree $`dimM`$. It however does make sense for an operator of order $`<dimM`$ and yields the ordinary trace. We should therefore find a way of only picking up the finite term in such an expression. Let us consider a classical P.D.O $`A`$ of order $`\alpha `$ with symbol $`\sigma =_{j=0}^N\mathrm{\Psi }\sigma _{\alpha j}+\sigma _{(N)}`$ ( see \[Sh\] for a discussion about such assumptions on the symbol) where $`\sigma _{\alpha j}S^{\alpha j}(\mathrm{IR}^n)`$, $`\sigma _{(N)}S^{\alpha N1}(\mathrm{IR}^n)`$ and where $`\mathrm{\Psi }`$ is a smooth function on $`\mathrm{IR}^n`$ which is zero in $`B(0,\frac{1}{4})`$ and equal to $`1`$ on $`\mathrm{IR}^nB(0,\frac{1}{2})`$. Since $`\sigma _{(N)}(\xi )=O(|\xi |^{\alpha N})`$ we have $`_{B(0,R)}\sigma _{(N)}(\xi )๐\xi =_{\mathrm{IR}^n}\sigma _{(N)}(\xi )๐\xi +O(R^{\alpha +nN})`$. On the other hand, splitting the integrals $`_{B(0,R)}\mathrm{\Psi }\sigma _{\alpha j}(\xi )๐\xi =_{B(0,1)}\mathrm{\Psi }\sigma _{\alpha j}(x,\xi )๐\xi +_{B(0,R)/B(0,1)}\mathrm{\Psi }\sigma _{\alpha j}(x,\xi )๐\xi `$, we use the fact that $`\sigma _{\alpha j}`$ is homogeneous of order $`\alpha j`$ to express the last integral. If $`\alpha `$ is an integer then there is an integer $`j_0`$ such that $`\alpha j_0+n=0`$ and we have for $`N>j_0`$:
$$\begin{array}{cc}& \underset{j=0}{\overset{N}{}}_{B(0,R)/B(0,1)}\mathrm{\Psi }\sigma _{\alpha j}(x,\xi )๐\xi =\underset{j=0}{\overset{N}{}}_1^R_{|\xi |=1}r^{\alpha j+n1}\sigma _{\alpha j}(x,\xi )๐\xi ๐r\hfill \\ & _R\mathrm{}\underset{j=0,n+\alpha j0}{\overset{N}{}}\frac{1}{\alpha +nj}R^{\alpha j+n}_{|\xi |=1}a_{\alpha j}(x,\xi )๐\xi +logR_{|\xi |=1}a_n(x,\xi )๐\xi +\text{constant term}.\hfill \end{array}$$
Finally this yields the existence of an asymptotic expansion in $`R+\mathrm{}`$ and of a constant (w.r.to $`R`$) $`C(a(x,))`$ such that
$$\begin{array}{cc}& _{B(0,R)}a(x,\xi )๐\xi \hfill \\ & _R\mathrm{}\underset{j=0,n+\alpha j0}{\overset{N}{}}\frac{1}{\alpha +nj}R^{\alpha j+n}_{|\xi |=1}\sigma _{\alpha j}(x,\xi )๐\xi +logR_{|\xi |=1}a_n(x,\xi )๐\xi +C(a(x,)).\hfill \end{array}$$
Because of the logarithmic term in $`R`$, one does not expect the finite part
$$f.p.\left(_{\mathrm{IR}^n}a(x,\xi )๐\xi \right)=LIM_R\mathrm{}_{B(0,R)}\sigma (x,\xi )๐\xi $$
to be invariant under a change of variable of $`\mathrm{IR}^n`$. However if the order $`\alpha `$ is not an integer, there is no logarithmic divergence and $`f.p._{\mathrm{IR}^n}\sigma (x,\xi )๐\xi `$ is independent of the local representation $`a(x,\xi )`$ of $`A`$: Proposition \[KV1, 2\] (see also \[Le\] in Prop. 5.2): Provided $`APDO(M,E)`$ has non integer order:
$$TR(A):=_Mtr_{E_x}(f.p.(_{\mathrm{IR}^n}\sigma _A(x,\xi )d\xi ))dvol(x)$$
$`(1.1)`$
where $`\sigma _A`$ is the symbol of $`A`$, is well-defined and satisfies the tracial property:
$$TR([A,B])=0APDO(M,E),BPDO(M,E),\text{such that}ord(A)+ord(B)\mathrm{Z}\mathrm{Z}.$$
Here $`tr_{E_x}`$ denotes the trace on the fibre $`E_x`$ of the bundle $`E`$ above the point $`x`$. The linear functional $`TR`$ coincides with the usual trace for P.D.Os of order with real part strictly smaller than minus the dimension of the underlying manifold the operators are acting on. $`TR`$ is in fact a trace functional i.e $`TR[A,B]=0`$ for any $`A,BPDO(M,E)`$ such that $`ordA+ordB\mathrm{IC}\mathrm{Z}\mathrm{Z}`$ (see Proposition 3.2 in \[KV2\]). $``$ A fundamental property of the canonical trace Following Kontsevich and Vishik, we shall call a local family $`A_zPDO(M,E)`$ with distribution kernels (locally) weakly holomorphic if the following conditons are satisfied: (i) The order $`\alpha _z`$ of $`A_z`$ is (locally) holomorphic in $`z`$, (ii)The kernel $`A_z(x,y)`$ of $`A_z`$ is (locally) holomorphic in $`z`$ for $`x,y`$ in disjoint local charts, (iii)Given any local chart $`U`$ on $`M`$, the homogeneous components $`\sigma _{A_z,\alpha _zj}(x,\frac{\xi }{|\xi |})`$ of the symbol $`\sigma _{A_z}(x,\xi )`$ are (locally) holomorphic functions in $`z`$ on the restriction to $`U`$ of the cotangent sphere bundle $`S^{}M`$. (iv) When $`x`$ and $`y`$ belong to a common local chart $`U`$, the difference between $`A_z(x,y)`$ and the truncated kernel $`_{j=0}^N\rho (|\xi |)\sigma _{A_z,\alpha _zj}(x,\xi )exp(ixy,\xi d\xi `$ of class $`C^{k(N)}`$ for some $`k(N)`$ increasing with $`N`$ tends to a (locally) holomorphic kernel on $`U\times U`$ when $`N\mathrm{}`$. In (iii) the topology on the space of symbols is given by the supremium norm of the symbol and its derivatives and in (iv) the convergence is to be understood in the sense of weak convergence of distributions \[KV1\].
The following property of the canonical trace plays a fundamental part in these notes: Fundamental property(see \[KV2\] Proposition 3.4 and \[KV1\] Th.3.13) For any (local weak) holomorphic family $`A(z)`$ of classical P.D.Os on $`M`$, $`zU\mathrm{IC}`$, $`ordA_z=\alpha (z)`$ where $`\alpha `$ is holomorphic and $`\alpha ^{}`$ does not vanish, the function $`TR(A_z)`$ is meromorphic with no more than simple poles at $`z=mU\mathrm{Z}\mathrm{Z}`$. $``$ Wodzicki residue Applying the fundamental property to the family $`A_z^Q:=AQ^z`$ where $`QEll_{ord>0}^{,+}(M,E)`$, $`APDO(M,E)`$ leads to the notion of Wodzicki residue \[W\] (see also \[K\] for a review):
$$res(A):=ordQRes_{z=0}TR(A_z^Q)$$
$`(1.2)`$
which is in fact independent of $`Q`$. This can be carried out for any operator $`QEll_{ord>0}^+(M,E)`$ which might not be injective, replacing it by the operator $`Q+P_Q`$ in the above formulas so that $`A_z^Q=A(Q+P_Q)^z`$ where $`P_Q`$ denotes the orthogonal projection of $`Q`$ onto its kernel which is finite dimensional since $`Q`$ is elliptic and the manifold closed. The projection is orthogonal for the inner product on the space $`C^{\mathrm{}}(M,E)`$ of smooth sections of $`E`$ induced by the hermitian structure $`,_x`$ on the fibre over $`xM`$ and the Riemannian volume measure $`\mu `$ on $`M`$:
$$\sigma ,\rho :=_M๐\mu (x)\sigma (x),\rho (x)_x\sigma ,\rho C^{\mathrm{}}(M,E)$$
The Wodzicki residue defines a trace functional on the algebra $`PDO(M,E)`$ of classical P.D.Os and vanishes for any classical P.D.O with non integer order. Since $`res`$ also vanishes on a smoothing operator, it induces a trace functional on the symbol algebra of $`PDO(M,E)`$ \[W\]. It is in fact the unique trace functional on $`PDO(M,E)`$ provided $`M`$ is connected and has dimension $`>1`$. An important feature of the Wodzicki residue we need to keep in mind here is that it vanishes on any trace-class operator as can easily be checked from the above definition. Kontsevich and Vishik (see \[KV2\] see (3.16)) furthermore show that, given a local weak holomorphic family $`A_z`$ of operators of order $`\alpha (z)`$, the poles of $`TR(A_z)`$ at entire points are expressed in terms of Wodzicki residues:
$$Res_{z=m}TR(A_z)=\frac{1}{\alpha ^{}(\alpha ^1(m))}res(\sigma (A_{\alpha ^{}1(m)}).$$
$`(1.3)`$
$``$ The weighted trace The fundamental property leads yet to another linear functional on the algebra $`PDO(M,E)`$ which is not a trace but interesting all the same because it does not vanish on trace class operators for which it coincides with the ordinary trace. Given $`QEll_{ord>0}^{,+}(M,E)`$ of order $`ordQ`$, we call the $`Q`$-weighted trace of an operator $`APDO(M,E)`$:
$$tr^Q(A):=\left[TR(AQ^z)\frac{1}{ordQz}res(A)\right]_{z=0}.$$
$`(1.4)`$
Here again, this extends to the case when $`Q`$ is not injective setting:
$$tr^Q(A):=\left[TR(A(Q+P_Q)^z)\frac{1}{ordQz}res(A)\right]_{z=0}$$
$`(1.5)`$
where as before $`P_Q`$ is the orthogonal projection onto the kernel of $`Q`$. Warning! To simplify notations, we shall often assume that $`Q`$ is injective subintending that when it is not, one should replace $`Q`$ by $`Q+P_Q`$. As we shall soon see, although refered to as weighted traces here, these functionals do not satisfy the tracial property $`tr^Q[A,B]=0`$ and hence do not deserve the name โtraceโ. We shall all the same keep to this abusive terminology which turns out to be very convenient. $``$ Dependence on the weight Unlike the Wodzicki residue, it generally depends on the choice of $`Q`$. The dependence is intrinsically infinite dimensional since it is measured in terms of a Wodzicki residue. Indeed, let $`Q_1,Q_2`$ be two operators in $`Ell_{ord>0}^{,+}(M,E)`$ with same order $`q`$, applying the fundamental property and $`()`$ to the (locally around zero) holomorphic family $`A(\frac{Q_1^zQ_2^z}{z})`$ of order $`\alpha (z):=ordAzq`$, we find (see Prop.2.2 in \[KV2\]):
$$\begin{array}{cc}\hfill tr^{Q_1}(A)tr^{Q_2}(A)& =\underset{z0}{lim}TR(A(\frac{Q_1^zQ_2^z}{z}))\hfill \\ & =q^1res(A(logQ_1logQ_2))\hfill \end{array}$$
$`(1.6)`$
which is well defined since $`logQ_1logQ_2`$ lies in $`PDO(M,E).`$ However when the underlying manifold is odd dimensional and for any odd-class classical P.D.O $`A`$ with integer order, the TR-generalized zeta function $`zTR(AQ^z)`$ is regular at $`0`$ and $`tr^Q(A)`$ obtained as the limit when $`z0`$ of these expressions is independent of the choice of $`Q`$ (see \[KV1\] Prop. 4.1 where it is denoted by $`Tr_{(1)}`$). This limit can be seen as an extension of the canonical trace to operators with integer orders. Although weighted traces are not tracial, they have a useful covariance property: Lemma 1 Let $`CPDO(M,E)`$ be injective and bounded. With the same notations as above, we have:
$$tr^{C^1QC}(A)=tr^Q(CAC^1).$$
$`(1.7)`$
Proof: For $`z\mathrm{IC}`$ with real part large enough, we have:
$$\begin{array}{cc}\hfill TR(A(C^1QC)^z)& =TR(AC^1Q^zC)\hfill \\ & =TR(CAC^1Q^z)\hfill \end{array}$$
where we have used that $`TR`$ is tracial. Taking the renormalized limit then yields the result. $``$
2. The Radul cocycle as a coboundary of the weighted trace In this section we investigate the coboundary of the weighted traces introduced above. We show how the cocycles obtained in this way relate to the Radul cocycle which arises in geometric quantization \[M2\]. In the context of regularized determinants, it is related to the multiplicative anomaly \[D\]. This cocycle was already investigated in \[MN\] where the authors express the coboundary of weighted traces in terms of a Wodzicki residue (see \[MN\] Lemma 13). We shall need some definitions of Lie algebra cohomology (see e.g. \[M1\] ). $``$ Lie algebra cohomology: Let $`L`$ be a Lie algebra and $`V`$ an $`L`$-module (the action of $`L`$ on $`V`$ is denoted by a โdotโ). A cochain of degree n ( or n-cochain) with values in $`V`$ is an antisymmetric multilinear map $`c:L\times L\times \mathrm{}\times L`$ (n times) $`V`$. Let $`C^n(L,V)`$ denote the space of all $`n`$-cochains and let us define the coboundary operator:
$$\delta :C^n(L,V)C^{n+1}(L,V)$$
$$\begin{array}{cc}\hfill (\delta c^n)(x_1,x_2,\mathrm{},x_{n+1})=& \underset{i<j}{}(1)^{i+j}c^n([x_i,x_j],x_1,\mathrm{},\widehat{x}_i,\mathrm{},\widehat{x}_j,\mathrm{},x_{n+1})\hfill \\ & +\underset{i=1}{\overset{n+1}{}}(1)^{i+1}x_ic^n(x_1,\mathrm{},x_{i1},x_{i+1},\mathrm{},x_{n+1})\hfill \end{array}$$
Taking $`V=\mathrm{IC}`$ with the trivial zero action of $`L`$ on $`\mathrm{IC}`$, $`xz:=0`$ for any $`xL`$, $`z\mathrm{IC}`$ we have :
$$(\delta c^n)(x_1,x_2,\mathrm{},x_{n+1})=\underset{i<j}{}(1)^{i+j}c^n([x_i,x_j],x_1,\mathrm{},\widehat{x}_i,\mathrm{},\widehat{x}_j,\mathrm{},x_{n+1}).$$
In particular $`\delta ^2=0`$ . An n-cocycle is a cochain of degree $`n`$ with vanishing coboundary. Let us denote by $`B^n(L,V)`$ the set of $`n`$-coboundaries, by $`Z^n(L,V)`$ the set of $`n`$ cocycles and let us call $`H^n(L,V):=Z^n(L,V)/B^n(L,V)`$ the $`n`$-th cohomology space. $``$ The weighted Radul cocycle on $`PDO(M,E)`$ We now apply this construction to $`L:=PDO(M,E)`$ and the $`1`$-cochain given by a weighted trace. Let $`QEll_{ord>0}^{,+}(M,E)`$. The $`Q`$-weighted traces $`tr^Q`$ do not satisfy the cyclicity property thus leading to a cocycle given by its coboundary:
$$c_R^Q(A,B):=\delta tr^Q(A,B)=tr^Q[A,B]A,BPDO(M,E)$$
$`(2.1)`$
which we call the $`Q`$-weighted Radul cocycle. This terminology is justified on the grounds of the following proposition which relates $`c_R^Q`$ to a more familiar expression of the classical Radul cocycle. Proposition 1 (i) For any $`APDO(M,E)`$ the operator $`[logQ,A]`$ lies in $`PDO(M,E)`$. (ii) For any $`A,BPDO(M,E)`$ we have:
$$c_R^Q(A,B)=\frac{1}{ordQ}res([logQ,A]B)=\underset{z0}{lim}TR(Q^z[A,B]).$$
$`(2.2)`$
Remark This relation expresses the algebraic obstruction preventing a weighted trace from being tracial in terms of a Wodzicki residue, which is a trace of purely infinite dimensional type. Proof: (i) We first check that $`[logQ,A]`$ lies in $`PDO(M,E)`$ if the order $`ordA`$ of $`A`$ is strictly positive.
$$\begin{array}{cc}\hfill [A,logQ]& =(AlogQlogQA)\hfill \\ & =A\left(logQ\frac{ordQ}{ordA}log(|A|+1)\right)+\left(\frac{ordQ}{ordA}log(|A|+1)logQ\right)A.\hfill \end{array}$$
Since the difference of two logarithms of PDOs of same order is a P.D.O, this proves that if the order of $`A`$ is strictly positive then $`[logQ,A]`$ is a P.D.O. Now if $`A`$ has negative order, let $`k=\frac{|ordA|+1}{ordQ}`$. Then setting $`P=Q^k`$ we have that $`P`$ is of order equal to $`|ordA|+1`$ and we have:
$$\begin{array}{cc}\hfill P(AlogQ(logQ)A)& =Q^kAlogQlogQQ^kA\hfill \\ & =[Q^kA,logQ],\hfill \end{array}$$
which by the previous results applied to $`Q^kA`$ which has strictly positive order, shows it is a P.D.O. (ii) For $`Rez`$ large enough we have:
$$\begin{array}{cc}\hfill TR(Q^z[A,B])& =TR([Q^z,A]B)+TR([A,Q^zB])\hfill \\ & =TR([Q^z,A]B)+TR([AQ^{\frac{z}{2}},Q^{\frac{z}{2}}B])\hfill \\ & =TR([Q^z,A]B).\hfill \end{array}$$
In the same way we have:
$$\begin{array}{cc}\hfill TR(Q^z[A,B])& =TR([A,B]Q^z)\hfill \\ & =TR(A[B,Q^z])+TR(AQ^zBBAQ^z)\hfill \\ & =TR(A[B,Q^z])+TR([AQ^{\frac{z}{2}},Q^{\frac{z}{2}}B])\hfill \\ & =TR(A[B,Q^z]).\hfill \end{array}$$
Appplying the fundamental property and $`()`$ to the family $`z^1[Q^z,A]B`$ of order $`\alpha (z)=zordQ+ordA+ordB`$ we find:
$$\begin{array}{cc}\hfill \underset{z0}{lim}TR([Q^z,A]B)& =Res_{z=0}TR(z^1[Q^z,A]B)\hfill \\ & =\frac{1}{ordQ}res\left(\frac{d}{dz}_{/z=0}([Q^z,A]B)\right)\hfill \\ & =\frac{1}{ordQ}res([logQ,A]B).\hfill \end{array}$$
In a similar way we have:
$$\underset{z0}{lim}TR(A[B,Q^z])=\frac{1}{ordQ}res(A[B,logQ]).$$
This proposition shows that the $`Q`$-weighted Radul cocycle generalizes the usual Radul cocycle obtained for $`\sigma (Q)=|\xi |`$ \[CFNW\] (see formula (24)),\[M2\], \[R\] (see formula (41)), \[KK\] (see formula (1)). In this particular case, the Radul cocycle was also considered in \[KV2\] in relation to multiplicative anomalies for determinants of elliptic operators (see also \[D\]) but it was later considered for more general operators $`Q`$ in \[MN\] and proved (see Lemma 13) to be a coboundary in the Hochschild cohomology of pseudo-differential operators. As a consequence of the above proposition, the weighted Radul cocycle vanishes on the algebra of odd-class P.D.Os with integer order whenever the underlying manifold is odd dimensional. To each $`\mathrm{IC}`$-valued cocycle $`c_R^Q`$ corresponds a central extension (see e.g \[M1\], \[Ki\], \[Rog\]) which we shall denote by $`PDO(M,E)^Q:=\{(A,\lambda ),APDO(M,E)\}`$ with Lie bracket:
$$[(A,\lambda ),(B,\mu )]^Q:=([A,B],c_R^Q(A,B)).$$
In other words we have the exact sequence of Lie algebras:
$$0\mathrm{IC}PDO(M,E)^QPDO(M,E)0$$
Two such extensions $`PDO(M,E)^{Q_1}`$ and $`PDO(M,E)^{Q_2}`$ are equivalent since for $`Q_1,Q_2Ell_{>0}^+(M,V)`$ the cocycles $`c_R^{Q_1}`$ and $`c_R^{Q_2}`$ are cohomologous. Indeed their difference
$$c_R^{Q_1}c_R^{Q_2}=res((logQ_1logQ_2)[,]))$$
is the coboundary of the $`1`$ cochain $`Ares((logQ_1logQ_2))`$. Since the difference of two logarithms is a zero order P.D.O, this latter $`1`$ cochain is a particular example of a family of cochains parametrized by the algebra $`PDO^0(M,E)`$ of operators in $`PDO(M,E)`$ of order $`zero`$:
$$res^P:=res(P)$$
$`(2.3)`$
where $`P`$ is a zero order P.D.O. The coboundary $`(A,B)res(P[A,B])`$ does not vanish in general. The case when $`P=ฯต(D)`$ is the sign of a self-adjoint operator $`D`$ gives rise to Mickelssonโs $`res^{}`$ linear functional.
3. A weighted Schwinger functional In this section $`D`$ denotes a self-adjoint elliptic operator acting on smooth sections of a vector bundle $`E`$ with strictly positive order. An example of such a bundle is given by the spinor bundle over an odd dimensional spin compact manifold $`M`$ and the operator $`D`$ by the Dirac operator on $`M`$. Adopting notations which are frequently used in the context of geometric quantization, let $`ฯต(D):=D+P_D(|D|+P_D)^1`$ denote the sign of $`D`$ which defines a classical P.D.O. of order $`0`$. The operator $`Q:=|D|`$ lies in $`Ell_{ord>0}^+(M,E)`$. $``$ A Schwinger functional We define the Schwinger functional:
$$\begin{array}{cc}\hfill PDO(M,E)& \mathrm{IC}\hfill \\ \hfill (A,B)& c_S^D(A,B):=\frac{1}{2}tr^{|D|}(ฯต(D)[ฯต(D),A][ฯต(D),B]).\hfill \end{array}$$
$`(3.1)`$
The terminology โSchwinger functionalโ is motivated by the fact (as we shall see later) that on some subalgebras of $`PDO(M,E)`$ it coincides with the usual Schwinger cocycle. A straightforward computation yields:
$$c_S^D(A,B)=tr^{|D|}([A,ฯต(D)]B)=tr^{|D|}(A[ฯต(D),B]).$$
Let us introduce the linear functional
$$\begin{array}{cc}\hfill PDO(M,E)& \mathrm{IC}\hfill \\ \hfill A& tr_ฯต^D(A):=tr^{|D|}(ฯต(D)A)\hfill \end{array}$$
which we call the signed weighted trace of $`A`$. As before the terminology โtraceโ is not appropriate here since this functional does not in general have vanishing coboundary. Proposition 3 Let $`๐ฎ`$ be a subalgebra of $`PDO(M,E)`$. The following conditions are equivalent (all the cocycles are Lie algebra cocycles):
1) $`c_S^D`$ is a $`2`$-cocycle on $`๐ฎ`$,
2) $`c_S^D`$ is an antisymmetric bilinear form on $`๐ฎ`$,
3) The bilinear map
$$(A,B)c_{TR}^D(A,B):=tr^{|D|}[ฯต(D)A,B]$$
defines a $`2`$-cocycle on $`๐ฎ`$,
4) $`c_{TR}^D`$ is an antisymmetric bilinear form on $`S`$,
5) For any $`A,B๐ฎ`$ we have:
$$res^{ฯต(D)}[A,[log|D|,B]]=0,$$
$`(3.2)`$
Provided one of these conditions is fufilled, the following relation holds:
$$c_{TR}^Dc_S^D=\delta tr_ฯต^D$$
so that the cocycles $`c_{TR}^D`$ and $`c_S^D`$ are cohomologous. Remarks 1) $`c_{TR}^D(A,B)`$ coincides with $`2c^{}(A,B)`$ defined in (2.11) of \[M2\]. 2) Part 5) of this proposition shows that the obstruction to the cocycle property of the various functionals involved arises as a residue $`res^{ฯต(D)}[A,[log|D|,B]]`$. Here again the obstruction is therefore purely infinite dimensional. The following definitions and the lemma below will be used in the proof. We shall set:
$$\stackrel{~}{c}_{TR}(A,B):=tr^{|D|}[Aฯต(D),B]$$
and
$$\overline{c}_{TR}:=\frac{c_{TR}^D+\stackrel{~}{c}_{TR}^D}{2}.$$
Lemma 2
$$c_{TR}^D(B,A)=\stackrel{~}{c}_{TR}^D(A,B)A,BPDO(M,E)$$
so that $`\overline{c}_{TR}^D`$ is antisymmetric. Moreover
$$\begin{array}{cc}\hfill \overline{c}_{TR}^D(A,B)& =\delta tr_ฯต^D(A,B)+\frac{1}{2}\left(tr^{|D|}([ฯต(D),B]A)Tr^{|D|}([A,ฯต(D)]B)\right)\hfill \\ & =\frac{1}{2}\delta tr_ฯต^D(A,B)+\frac{1}{2}tr^{|D|}\left(Aฯต(D)BBฯต(D)A\right)\hfill \end{array}$$
where $`\delta tr_ฯต^D`$ denotes the coboundary of the signed weighted trace in the Lie algebra cohomology. Proof: Let us first see how $`c_{TR}^D`$ transforms when exchanging $`A`$ and $`B`$.
$$\begin{array}{cc}\hfill c_{TR}^D(A,B)& =tr^{|D|}[ฯต(D)B,A]\hfill \\ & =tr^{|D|}(ฯต(D)BAAฯต(D)B)\hfill \\ & =tr^{|D|}(Aฯต(D)BBAฯต(D))\hfill \\ & =tr^{|D|}[Aฯต(D),B]\hfill \\ & =\stackrel{~}{c}_{TR}^D(A,B)\hfill \end{array}$$
From this it follows that $`\overline{c}_{TR}^D`$ is antisymmetric. The second statement then easily follows from Lemma 2 and the definition of $`\overline{c}_{TR}^D`$:
$$\begin{array}{cc}\hfill \overline{c}_{TR}^D(A,B)& =\frac{1}{2}\left(c_{TR}^D(A,B)c_{TR}^D(B,A)\right)\hfill \\ & =\frac{1}{2}(c_ฯต(A,B)c_ฯต(B,A))+\frac{1}{2}(tr^{|D|}([ฯต(D),B]Atr^{|D|}([ฯต(D),A]B))\hfill \\ & =c_ฯต^D(A,B)\frac{1}{2}(tr^{|D|}([ฯต(D),B]A+tr^{|D|}([A,ฯต(D)]B))\hfill \\ & =\frac{1}{2}c_ฯต^D(A,B)+\frac{1}{2}\left(tr^{|D|}(Aฯต(D)BBฯต(D)A)\right).\hfill \end{array}$$
Proof of proposition 3: $`2)4):`$ A straightforward computation yields:
$$\begin{array}{cc}\hfill c_S^D(A,B)& =c_S^D(B,A)\hfill \\ & tr^{|D|}(A[ฯต(D),B])=tr^{|D|}(B[ฯต(D),A])\hfill \\ & tr^{|D|}[Aฯต(D),B]=tr^{|D|}[A,Bฯต(D)]\hfill \\ & \stackrel{~}{c}_{TR}^D(A,B)=c_{TR}(A,B)\hfill \\ & c_{TR}^D(B,A)=c_{TR}(A,B)\hfill \end{array}$$
where we have used the result of Lemma 2. $`3)4):`$ The implication from left to right is clear since a $`2`$-cocycle is antisymmetric. Let us prove the other implication $`4)3)`$ which amounts to showing that when $`c_{TR}`$ is antisymmetric it defines a cocycle. Since $`c_{TR}^D`$ is antisymmetric, we have $`c_{TR}^D=\stackrel{~}{c}_{TR}^D=\overline{c}_{TR}^D`$. By Lemma 2, it is sufficient to show that $`\omega `$ defined by $`\omega (A,B)Tr^{|D|}(Aฯต(D)BBฯต(D)A)`$ has vanishing coboundary $`\delta \omega `$. A direct computation yields:
$$\delta \omega (A,B,C)=c_{TR}^D(A,[Bฯต(D),Cฯต(D)])+c_{TR}^D(B,[Cฯต(D),Aฯต(D)])+c_{TR}^D(C,[Aฯต(D),Bฯต(D)])$$
from which follows that:
$$\begin{array}{cc}\hfill \delta \omega (Aฯต(D),Bฯต(D),Cฯต(D))& =c_{TR}^D(Aฯต(D),[B,C])+c_{TR}^D(Bฯต(D),[C,A])+c_{TR}^D(Cฯต(D),[A,B])\hfill \\ & =\stackrel{~}{c}_{TR}^D(Aฯต(D),[B,C])+\stackrel{~}{c}_{TR}^D(Bฯต(D),[C,A])+\stackrel{~}{c}_{TR}^D(Cฯต(D),[A,B])\hfill \\ & =tr^{|D|}[A,[B,C]]+Tr^{|D|}[B,[C,A]]+Tr^{|D|}[C,[A,B]]\hfill \\ & =0\hfill \end{array}$$
Since any P.D.O $`A`$ can be written $`A=A_1ฯต(D)`$ where $`A_1Aฯต(D)`$ is a P.D.O, the result follows. $`4)5):`$ Since this computation is similar to the one used in the proof of Lemma 2 we shall skip some intermediate steps here. Let $`d`$ denote the order of $`D`$ and let us set $`l:=log|D|`$.
$$\begin{array}{cc}\hfill dc_{TR}^D(B,A)& =dc_R^{|D|}(ฯต(D)B,A)\hfill \\ & =res(ฯต(D)[l,B]A)\hfill \\ & =res(ฯต(D)A[l,B])res(ฯต(D)[A,[l,B]])\hfill \\ & =d\stackrel{~}{c}_{TR}^D(B,A)res(ฯต(D)[A,[l,B]])\hfill \\ & =dc_{TR}^D(A,B)res(ฯต(D)[A,[l,B]])\hfill \end{array}$$
hence
$$c_{TR}^D(B,A)+c_{TR}^D(A,B)=\frac{1}{d}res(\sigma (ฯต(D)[A,[l,B]]))$$
and $`c_{TR}`$ is antisymmetric if and only if condition (3.2) is satisfied. so that $`c_{TR}^D`$ is antisymmetric if and only if $`res(ฯต(D)[A,[l,B]])=0`$ for any $`A,B`$ in the algebra under consideration. $`1)2):`$ Only the implication from right to left is non trivial. To prove it, we use Lemma 2 once again by which we have:
$$\overline{c}_{TR}^D(A,B)=\frac{1}{2}\delta tr_ฯต^D(A,B)+\frac{1}{2}Tr^{|D|}\left(Aฯต(D)BBฯต(D)A\right).$$
On the other hand, since $`c_S^D`$ is antisymmetric, we have:
$$c_S^D(A,B)=\frac{1}{4}tr^{|D|}(ฯต(D)[[ฯต(D),A],[ฯต(D),B]].$$
Then a direct computation yields:
$$c_S^D(A,B)=\frac{1}{2}(\delta tr_ฯต^D)(A,B)+\frac{1}{2}tr^{|D|}\left(Aฯต(D)BBฯต(D)A\right).$$
Hence
$$\overline{c}_{TR}^Dc_S^D=\delta tr_ฯต^D$$
From this identity follows that, provided $`c_S^D`$ and $`c_{TR}^D`$ are cocycles, then they are cohomologous. $``$ The algebra $`PDO(M,E)_{res}^D`$ Let us introduce a subalgebra $`PDO(M,E)_{res}^D`$of $`PDO(M,E)`$ the definition of which is close to the algebra $`g_{res}`$ (they coincide up to the fact that we require the operators to be P.D.Os) which plays a substancial part in geometric quantization techniques (see e.g. \[PS\]):
$$\begin{array}{cc}\hfill PDO(M,E)_{res}^D& =\{APDO(M,E),ordA0ord([A,ฯต(D)])<\frac{dimM}{2}\}\hfill \\ & =g_{res}^DPDO(M,E)\hfill \end{array}$$
$`(3.3)`$
where as before $`dimM`$ is the dimension of the underlying manifold and where $`ordA`$ denotes the order of the operator $`A`$. Here $`g_{res}^D:=\{A(L^2(M,E)),[A,ฯต(D)]\text{ is Hilbert Schmidt}\}`$ where $`L^2(M,E)`$ is the closure of the space $`C^{\mathrm{}}(M,E)`$ of smooth sections of $`E`$ for the $`L^2`$ inner product induced by the hermitian structure on $`E`$ and the Riemannian volume measure on $`M`$ as above, $`(L^2(M,E))`$ denoting the algebra of bounded operators on this Hilbert space. An immediate consequence of Proposition 3 is the following Corollary: Corollary 1 Let $`๐ฎ`$ be a subalgebra of $`PDO(M,E)`$. If $`๐ฎ`$ is stable under the map $`A[log|D|,A]`$ i.e if
$$A๐ฎ[log|D|,A]๐ฎ$$
$`(3.4)`$
and if moreover
$$res(ฯต(D)[A,B])=0A,B๐ฎ$$
$`(3.5)`$
hold on the subalgebra $`S`$, then so does relation (R) and $`c_{TR}^D`$ and $`c_S^D`$ define cohomologous cocycles on $`๐ฎ`$. Corollary 2 On the algebra $`PDO(M,E)_{res}^D`$ the Schwinger functional coincides with the usual Schwinger cocycle and $`c_{TR}^D`$ with the usual twisted Radul cocycle and we have:
$$c_S^Dc_{TR}^D.$$
Proof: We need to check that $`PDO(M,E)_{res}^D`$ fulfills assumptions (3.4) and (3.5) of Corollary 1. As before we set $`l:=log|D|`$. Since $`[ฯต(D),[l,A]]=[l,[ฯต(D),A]]`$ , the order of $`[ฯต(D),[l,A]]`$ is the same as that of $`[ฯต(D),A]`$ so that if $`A`$ lies in $`PDO(M,E)_{res}^D`$, so does $`[l,A]`$. Hence assumption (3.4) is satisfied on $`PDO(M,E)_{res}^D`$. An easy computation yields:
$$res(ฯต(D)[A,B])=\frac{1}{2}res(ฯต(D)[[ฯต(D),A],[ฯต(D),B]])$$
which vanishes on $`PDO(M,E)_{res}^D`$ since it is the trace of an operator of degree strictly smaller than $`dimM`$, the operator $`ฯต(D)`$ being of order $`0`$ and the operators $`[ฯต(D),A]`$, $`[ฯต(D),B]`$ being both of order strictly smaller than $`\frac{dimM}{2}`$. Hence assumption (ii) is satisfied on $`PDO(M,E)_{res}^D`$. Corollary 1 implies that both $`c_S^D`$ and $`c_{TR}^D`$ are cocycles on $`PDO(M,E)_{res}^D`$ and that they are cohomologous. On $`PDO(M,E)_{res}^D`$ the cocycle $`c_S^D`$ reads:
$$c_S^D(A,B)=\frac{1}{2}tr(ฯต(D)[ฯต(D),A][ฯต(D),B])$$
where now $`tr`$ is an ordinary trace and it therefore coincides with the usual Schwinger cocycle see e.g (according to the author, the definition might change by a constant factor) \[CFNW\], \[PS\], \[M2\], \[S\]. The expression of $`c_{TR}^D`$ in terms of a residue obtained in section 1 (Proposition 1) yields
$$c_{TR}^D(A,B)=\frac{1}{ordD}res([log|D|,ฯต(D)A]B)$$
thus relating $`c_{TR}^D`$ with Mickelssons \[M2\] twisted Radul cocycle (they coincide up to a factor $`2`$).
4. Cocycles and group representations Given $`DEll_{ord>0}^{s.a}(M,E)`$, we have a natural polarization of the space $`HL^2(M,E)`$ given by:
$$H=H_+(D)H_{}(D)$$
where $`H_+(D)=\pi _+(D)(H)`$, $`H_{}(D)=\pi _{}(D)(H)`$, $`\pi _+(D)=\frac{1+ฯต(D)}{2}`$, $`\pi _{}(D)=\frac{ฯต(D)+Id}{2}`$. Notice that with this choice for $`ฯต(D)`$, $`\pi _+(D)`$ is $`1`$ on $`KerD`$ and $`\pi _{}(D)`$ vanishes on $`KerD`$. Of course we could have chosen the other convention namely $`\pi _{}(D)`$ to be $`1`$ on $`KerD`$ and $`\pi _+(D)`$ to vanish on $`KerD`$. In this polarization, let us write an operator $`APDO(M,E)`$ as a matrix:
$$A\left[\begin{array}{cc}A_{++}& A_+\\ A_+& A_{}\end{array}\right].$$
Since $`ฯต(D)=\left[\begin{array}{cc}1& 0\\ 0& 1\end{array}\right]`$, we have
$$[ฯต(D),A]=2\left[\begin{array}{cc}0& A_+\\ A_+& 0\end{array}\right].$$
The operator $`ฯต(D)`$ being a P.D.O, the operator $`\left[\begin{array}{cc}0& A_+\\ A_+& 0\end{array}\right]`$ is also P.D.O. For $`DEll_{ord>0}^{s.a}(M,E)`$ we introduce the following bilinear functional:
$$\begin{array}{cc}\hfill \lambda ^D:PDO(M,E)& \mathrm{IC}\hfill \\ \hfill (A,B)& tr^{|D|}([A_{++},B_{++}][A,B]_{++}).\hfill \end{array}$$
$`(4.1)`$
On $`PDO(M,E)_{res}^D`$ it coincides with the Lie algebra cocycle corresponding to a central extension of the group:
$$G_{res}^D\{AGL(H),A_+\text{ and }A_+\text{ are Hilbert-Schmidt}\}$$
$`(4.2)`$
as described in \[PS\] (6.6.5). The Schwinger functional $`c_S^D`$ not being antisymmetric in general (since according to proposition 3 it then becomes a cocycle), it is natural to introduce a mean Schwinger functional in a similar way to what we we did for the twisted Radul cocycle:
$$\overline{c}_S^D(A,B):=\frac{c_S^D(A,B)c_S^D(B,A)}{2}=\frac{1}{4}tr^{|D|}[[ฯต(D),A],[ฯต(D),B]].$$
Lemma 3 Let $`DEll_{ord>0}^{s.a}(M,E)`$, let $`APDO(M,E)`$, $`BPDO(M,E)`$. Then
$$\begin{array}{cc}\hfill 2\lambda ^D(A,B)& =2tr^{|D|}(B_+A_+A_+B_+)\hfill \\ & =\overline{c}_S^D(A,B)c_R^{|D|}(A_+,B_+)c_R^{|D|}(A_+,B_+)\hfill \end{array}$$
Furthermore $`\lambda ^D`$ is a cocycle on $`PDO(M,E)_{res}^D`$ and the following relation holds on $`PDO(M,E)_{res}^D`$:
$$\begin{array}{cc}\hfill \frac{1}{2}c_S^D(A,B)& =\lambda ^D(A,B)\hfill \\ & =TR(A_+B_+A_+B_+)\hfill \\ & =\frac{1}{4}TR\left(ฯต(D)[ฯต(D),A][ฯต(D),B]\right)\hfill \end{array}$$
confirming formula (6.6.6) of \[PS\]. Remark : Although the individual components $`A_{++},A_+,A_+,A_{}`$ of the two by two matrix of $`A`$ in the polar decomposition might not be P.D.Os, $`c_R^{|D|}(A_+,B_+)=tr^{|D|}[A_+,B_+]`$ and $`c_R^{|D|}(A_+,B_+)=tr^{|D|}[A_+,B_+]`$ are well defined since these are weighted traces of operator brackets which are P.D.Os as projections of operator brackets of P.D.Os of the type $`[[ฯต(D),A],[ฯต(D),B]]`$. Proof: Let us start with the general case. The first equality follows from a direct computation and the definition of the weighted Radul cocycle. The second equality follows from the fact that
$$ฯต(D)[[ฯต(D),A],[ฯต(D),B]]=4\left[\begin{array}{cc}A_+B_+B_+A_+& 0\\ 0& B_+A_+A_+B_+\end{array}\right]$$
since this yields:
$$\overline{c}_S^D(A,B)=2tr^{|D|}(A_+B_+B_+A_+)+c_R^{|D|}(B_+,A_+)+c_R^{|D|}(B_+,A_+).$$
When considering the case of $`PDO(M,E)_{res}^D`$, the Radul cocycles $`c_R^{|D|}(B_+,A_+)`$ and $`c_R^{|D|}(B_+,A_+)`$ vanish since $`B_+,A_+,B_+,A_+`$ are Hilbert-Schmidt and the weighted traces $`tr^{|D|}`$ become ordinary traces $`TR`$. This completes the proof.
Noting that $`[ฯต(D),A]=[ฯต(D),A]^{}`$ is equivalent to $`A_+=A_+^{}`$ for an operator $`C:H_+(D)H_{}(D)`$ we shall set:
$$j(C)\left[\begin{array}{cc}0& C^{}\\ C& 0\end{array}\right].$$
Lemma 4 Let $`DEll_{ord>0}^{s.a}(M,E)`$, let $`A:H_+(D)H_{}(D)`$, $`B:H_+(D)H_{}(D)`$ such that $`j(A)PDO(M,E)`$, and $`j(B)PDO(M,E)`$. Then
$$\begin{array}{cc}\hfill \overline{c}_S^D(j(A),j(B))& =\delta (tr_ฯต^D)(j(A),j(B))\hfill \\ & =2tr^{|D|}(B^{}AA^{}B)+c_R^{|D|}(A,B^{})+c_R^{|D|}(A^{},B)\hfill \end{array}$$
Proof: Since $`[ฯต(D),j(A)]=2j(A)`$ we have:
$$\begin{array}{cc}\hfill \overline{c}_S^D(j(A),j(B))& =\frac{1}{4}tr_ฯต^D([[ฯต(D),j(A)],[ฯต(D),j(B)]]\hfill \\ & =tr_ฯต^D([j(A),j(B)])\hfill \\ & =c_ฯต^D(j(A),j(B))\hfill \\ & =tr_ฯต^D\left[\begin{array}{cc}A^{}B+B^{}A& 0\\ 0& AB^{}+BA^{}\end{array}\right]\hfill \\ & =2tr^D(B^{}AA^{}B)+c_R^{|D|}(A,B^{})c_R^{|D|}(B,A^{})\hfill \\ & =2tr^D(B^{}AA^{}B)+c_R^{|D|}(A,B^{})+c_R^{|D|}(A^{},B)\hfill \end{array}$$
On the grounds of this proposition we set for two operators $`A,B`$ such that $`A^{}BPDO(M,E)`$:
$$\omega ^D(A,B)itr^{|D|}(A^{}BB^{}A).$$
$`\omega ^D`$ relates to the mean Schwinger functional $`\overline{c}_S`$ as follows: Corollary 3 Let $`A:H_+(D)H_{}(D)`$ and $`B:H_+(D)H_{}(D)`$ be operators such that $`j(A)`$ and $`j(B)`$ lie in $`PDO(M,E)`$. Then
$$\overline{c}_S^D(j(A),j(B))=2i\omega ^D(A,B)+c_R^{|D|}(A,B^{})+c_R^{|D|}(A^{},B).$$
Whenever $`A`$ and $`B`$ are Hilbert-Schmidt, then
$$\overline{c}_S^D(j(A),j(B))=2i\omega ^D(A,B)=2tr(A^{}BB^{}A)$$
where $`tr`$ is now an ordinary trace.
5. Geometry on weighted bundles $``$ Weighted vector bundles We shall say that a Hilbert space $`H`$ lies in the class $`๐`$ whenever there is a compact boundaryless compact smooth Riemannian manifold $`M`$, a finite rank smooth vector bundle $`E`$ based on $`M`$ and $`s>\frac{dimM}{2}`$ such that $`H=H^s(M,E)`$. Typically, letting $`G`$ be a Lie group and $`Lie(G)`$ be its Lie algebra, the Lie algebra $`H^s(M,Lie(G))`$ of the Hilbert current group $`H^s(M,G)`$ lies in $`๐`$. Let $`๐`$ be the class of Hilbert vector bundles $`X`$ based on a (possibly infinite dimensional) manifold $`X`$ with fibres modelled on a separable Hilbert space $`H`$ in the class $`๐`$ defined previously and with transition maps in $`PDO(M,E)`$ when $`H`$ is the space of sections in some Sobolev class of a vector bundle $`E`$ based on $`M`$. $`๐๐ณ`$ denotes the class of infinite dimensional manifolds $`X`$ with tangent bundle $`TX`$ in $`๐`$. Since the transition maps are bounded, they correspond to operators of order $`0`$ and since they are invertible, they are in fact elliptic operators of order zero so that they lie in $`Ell(M,E)`$. To illustrate this setting let us give examples of manifolds, resp. vector bundles in the class $`๐๐ณ`$, resp. $`๐`$. $``$ Examples i) Finite rank vector bundles lie in the class $`๐`$. To see this we take a manifold $`M=\{\}`$ reduced to a point $``$, the bundle $`E`$ to be trivial of the type $`\{\}\times \mathrm{IR}^d`$ (or $`\{\}\times \mathrm{IC}^d`$ if the bundle is complex). The transition functions belong to $`Ell(\{\},E)=Gl_d(\mathrm{IR})`$ ( or $`Gl_d(\mathrm{IC})`$ if the bundle is complex). ii) If $`G`$ denotes a Lie group and $`s>\frac{dimM}{2}`$, then the current group $`H^s(M,G)`$ is a Hilbert Lie group and can be equipped with a left invariant atlas $`\varphi _\gamma (u)(x):=exp_{\gamma (x)}(u(x))xM,\gamma H^s(M,G)`$ where $`exp_{\gamma (x)}`$ is the exponential coordinate chart at point $`\gamma (x)`$ induced by a left invariant Riemannian metric on $`G`$. The transition functions are given by multiplication operators which indeed are P.D.Os. iii) More generally, let $`N`$ be a Riemannian manifold, then the space $`H^s(M,N)`$ is a Hilbert manifold with tangent space at a point $`\gamma `$ given by $`H^s(M,\gamma ^{}TN)`$. This manifold, which is modelled on $`H^s(M,\mathrm{IR}^n)`$ where $`n`$ is the dimension of $`N`$, can be equipped with an atlas induced by the exponential map $`exp^N`$ on $`N`$ in a similar way to the above description, a local chart being of the type $`\varphi _\gamma (u)(x)=exp_{\gamma (x)}^N(u(x))`$. The transition functions are locally given by multiplication operators and hence define PDOโs. $``$ Bundles of operators Let $``$ be a $`C^{\mathrm{}}`$ vector bundle in the class $`๐`$ based on a manifold $`B`$ and let it be modelled on a separable Hilbert space $`H`$. For any $`bB`$ we shall denote by $`PDO_b()`$ the class of operators $`A_b`$ acting densely on the fibre $`_b`$ above $`b`$ such that for any local trivialisation $`\varphi :|_{U_b}U_b\times H`$ around $`b`$ the operator $`\varphi ^{\mathrm{}}A(b):=\varphi (b)A_b\varphi (b)^1`$ (where $`\varphi (b):_bH`$ (obtained after localizing it using smooth cut-off functions) is the isomorphism induced by the trivialization) lies in $`PDO(M,E)`$. This definition involves local charts but it is in fact independent of the choice of local chart. Indeed, for another local chart $`(U,\psi )`$ we have $`\psi ^{\mathrm{}}A(b)=\psi (b)\varphi (b)^1\varphi ^{\mathrm{}}A_b\varphi (b)\psi (b)^1`$, and since the transition functions $`\psi (b)\varphi (b)^1`$ are given by classical P.D.Os, the condition $`\varphi ^{\mathrm{}}APDO(M,E)`$ is independent of the choice of $`\varphi `$. In a similar way, the notion of order defined by the order of the P.D.O in the local chart, does not depend on the choice of local chart. We shall also denote by $`Ell_b()`$ the class of operators $`A_b`$ acting densely on $`_b`$ such that for any local trivialisation $`\varphi :/U_pU_p\times H`$ around $`b`$ the operator $`\varphi :/U_bU_b\times H`$ around $`b`$ the operator, $`(\varphi ^{\mathrm{}}A)(b)`$ lies in $`Ell(M,E)`$. Here again, the definition involves a choice of local chart but is in fact independent of that choice.Indeed, the principal symbol being multiplicative and using the characterization of ellipticity in terms of invertibility of the principal symbol, one easily checks that the condition $`\varphi ^{\mathrm{}}AEll(M,E)`$ is independent of the choice of $`\varphi `$. This gives rise to two bundles $`PDO():=_{bB}PDO_b()`$ and $`Ell():=_{bB}Ell_b()`$ with fibre at point $`b`$ given respectively by $`PDO_b()`$ and $`Ell_b()`$. Notice that when $``$ is a bundle of finite rank, we have $`PDO()=Hom()`$ and $`Ell()=GL()`$ since the underlying manifold $`M`$ reduces to a point and the vector bundle to a vector space, namely the model space of $``$. $``$ Weighted traces A weight on a smooth finite rank vector bundle in the class $`๐`$ is a smooth section of $`Ell()`$ of operators with constant order which is locally a positive self-adjoint (elliptic) operator. If $``$ has finite rank, it is simply given by a section of $`GL()`$ which locally is a positive self-adjoint operator and hence corresponds to the choice of a Riemannian metric. If the base manifold is either locally compact and a countable union of compacts or if it is a paracompact Hilbert manifold modelled on a separable Hilbert space, then it has a smooth partition of unity \[L\] and such a global section can be built up patching up local sections of $`Ell()`$, i.e maps from an open neighborhood in the base manifold $`B`$ to $`Ell(M,E)`$ (this construction is similar to the one that gives the existence of a Riemannian metric under the same conditions). Let $`๐`$ be modelled on $`H`$. We define the weighted pseudo-trace of a field of operators:
$$\begin{array}{cc}\hfill \mathrm{\Gamma }(PDO())& \mathrm{\Gamma }(X,\mathrm{IR})(\text{or}\mathrm{\Gamma }(X,\mathrm{IC}))\hfill \\ \hfill A& tr^Q(A)\hfill \end{array}$$
$`(5.1)`$
locally; in a local chart $`(U,\varphi )`$ it is defined by the $`\varphi ^{\mathrm{}}`$-weighted pseudo-trace $`tr^{\varphi ^{\mathrm{}}Q}(\varphi ^{\mathrm{}}A)`$ of $`\varphi ^{\mathrm{}}A`$. Although the definition involves a choice of local chart, because of the covariance property it is in fact independent of this choice. Indeed if $`\varphi `$ and $`\psi `$ are two local charts around $`x`$ and if we set $`C:=\psi \varphi ^1`$ in formula (1.7), we have:
$$tr^{\psi ^{\mathrm{}}Q}(\psi ^{\mathrm{}}A)=tr^{C\varphi ^{\mathrm{}}QC^1}(C\varphi ^{\mathrm{}}AC^1)=tr^{\varphi ^{\mathrm{}}Q}(\varphi ^{\mathrm{}}A).$$
Remark Looking back at definition (1.3), strictly speaking, in order to make sure the field $`P_Q`$ of orthogonal projections is smooth, we should assume that $`KerQ`$ has constant dimension. This is an artificial difficulty which comes from the specific construction of the zeta function renormalization since it involves taking complex powers of an invertible operator $`(Q+P_Q)^z`$. However, bearing in mind that the heat-kernel renormalization only involves the exponential $`e^{tQ}`$ and the fact that $`Lim_{z0}tr(A(Q+P_Q)^z)=Lim_{ฯต0}tr(Ae^{ฯตQ})`$ (which can be shown via the Mellin transform), we see that the jumps in the dimension of the kernel of $`Q`$ do not affect the renormalized trace. From now on, we shall assume that $`Q`$ is invertible, otherwise one just replaces $`Q`$ by $`Q^{}:=Q+P_Q`$ where $`P_Q`$ is the orthogonal projection onto the kernel of $`Q`$. $``$ Variations of weighted traces Proposition 4 Let $`(,Q)`$ be a weighted vector bundle equipped with a connection $``$ and let $`\alpha `$, $`\beta `$ be two $`PDO()`$ valued forms on the base manifold of $``$. 1)
$$tr^Q([\alpha ,\beta ])=\frac{1}{ordQ}res([logQ,\alpha ]\beta )$$
$`(5.2)`$
2) Provided both $`[,logQ]`$ and $`[,\alpha ]`$ are sections of $`PDO(),`$ we have
$$[,tr^Q](\alpha ):=dtr^Q(\alpha )=tr^Q([,\alpha ])=\frac{1}{ordQ}res(\alpha [,logQ]).$$
$`(5.3)`$
Here $`ordQ`$ denotes the (constant) order of the field of elliptic operators $`Q`$ . Proof (5.2 ) follows from Proposition 1 (compare with (2.2)) As in the proof of Proposition 1, to prove (5.3) we use the fundamental property of the canonical trace $`TR`$. Let us first consider a smooth (for the natural topology on classical P.D.Os induced by the topology of uniform convergence in all derivatives on the classical symbols) one parameter family of operators $`A_tPDO(M,E)`$ with constant order $`a`$ and a smooth one parameter family $`Q_tEll_{ord>0}^+(M,E)`$ (for the natural Frรฉchet topology on $`PDO(M,E)`$) with constant order $`q`$, $`t`$ varying in $`]0,1[`$. Applying formula (1.3) to $`A_z:=\frac{A_0(Q_t^zQ_0^z)}{z}`$ (in which case $`\alpha (z)=aqz`$) and then going to the limit when $`t0`$ yields:
$$\frac{d}{dt}_{/t=0}tr^{Q_t}(A_0)=\frac{1}{q}res(A_0\frac{d}{dt}_{/t=0}logQ_t).$$
let us now also consider a $`1`$-parameter smooth family $`(A_t)`$ of P.D.Os of constant order, then:
$$\frac{d}{dt}_{/t=0}tr^{Q_t}(A_t)=tr^{Q_0}(\frac{d}{dt}_{/t=0}A_t)\frac{1}{ordQ_0}res(A_0\frac{d}{dt}_{/t=0}logQ_t).$$
Similarly, given any local trivialization around a point $`b_0`$ in the base manifold $`B`$ of $``$, we have:
$$dtr^Q(\alpha )=tr^Q(d\alpha )\frac{1}{ordQ}res(\alpha dlogQ).$$
$`(5.4)`$
Let us write $`=d+\theta `$ in this local trivialization. Since by assumption $`[,\alpha ]`$ is a section of $`PDO()`$, we have that locally $`d\alpha +[\theta ,\alpha ]PDO(M,E)`$ and hence, since $`d\alpha PDO(M,E)`$ as the differential of a P.D.O, we conclude that $`[\theta ,\alpha ]`$ also lies in $`PDO(M,E)`$. Applying the first part of the lemma yields:
$$tr^Q([\theta ,\alpha ])=\frac{1}{ordQ}res([logQ,\theta ]\alpha ).$$
$`(5.5)`$
Combining (5.4) and (5.5) we find:
$$\begin{array}{cc}\hfill dtr^Q(\alpha )& =tr^Q(d\alpha )\frac{1}{ordQ}res(\alpha dlogQ)\hfill \\ & =tr^Q([,\alpha ])tr^Q([\theta ,\alpha ])\frac{1}{ordQ}res(\alpha dlogQ)\hfill \\ & =tr^Q([,\alpha ])+\frac{1}{ordQ}res([logQ,\theta ]\alpha )\frac{1}{ordQ}res(\alpha dlogQ)\hfill \\ & =tr^Q([,\alpha ])\frac{1}{ordQ}res(\alpha [,logQ]).\hfill \end{array}$$
$``$ Ricci curvature on weighted manifolds Let $`X๐๐ณ`$ be equipped with a Levi-Civita connection $`^X`$ such that for any tangent vector fields $`U,V`$ the map:
$$R^X(U,V):W\mathrm{\Omega }^s(W,U)VW\mathrm{\Gamma }(TX)$$
is a section of $`PDO(TX)`$ where $`\mathrm{\Omega }^X`$ denotes the curvature of $`^X`$. Then, given a weight $`Q`$ on $`X`$ we can define the $`Q`$-weighted Ricci curvature:
$$r_1^{X,Q}:=tr^Q(R^X).$$
$`(5.6)`$
Since $`tr^Q`$ coincides with the ordinary trace on trace-class operators if $`X`$ is finite dimensional, it then coincides with the ordinary Ricci curvature. $``$ Weighted first Chern form on Kรคhler manifolds Let us now assume $`X๐๐ณ`$ is Kรคhler and let $`^X`$ be the a Kรคhler connection, let $`Q`$ be a weight on $`X`$. In a similar way, provided that for any holomorphic tangent vector field $`U`$ and for any antiholomorphic tangent vector field $`\overline{V}`$ the map $`\mathrm{\Omega }^X(U,\overline{V})`$ is a section of $`PDO(TX)`$, we define the $`Q`$-weighted first Chern form:
$$r_1^{X,Q}(U,\overline{V}):=tr^Q(\mathrm{\Omega }^X(U,\overline{V}))$$
$`(5.7)`$
where $`\mathrm{\Omega }^X`$ denotes the curvature of $`^X`$. If $`X`$ is finite dimensional, it coincides with the ordinary first Chern form. Unlike the first Chern form on a finite dimensional manifold, it is not closed in general. This follows from the following proposition: Corollary 4 Let $`(X,^X)`$ be a Kรคhler manifold eqXipped with a weight $`Q`$ and let $`ordQ`$ be the (constant) order of $`Q`$. Then, provided $`[^X,logQ]`$ is a $`PDO(TX)`$ valued one form and provided the curvature $`\mathrm{\Omega }^X`$ is a $`PDO(TX)`$ valued two form, we have
$$dr_1^Q=\frac{1}{ordQ}res([^X,logQ]\mathrm{\Omega }^X).$$
$`(5.8)`$
Proof: Applying Proposition 4 to $`\alpha :=\mathrm{\Omega }^X`$, we find:
$$\begin{array}{cc}\hfill dtr^Q(\mathrm{\Omega }^X)& =tr^Q([^X,\mathrm{\Omega }^X])\frac{1}{ordQ}res(\mathrm{\Omega }^X[^X,logQ])\hfill \\ & =\frac{1}{ordQ}res(\mathrm{\Omega }^X[^X,logQ])\hfill \end{array}$$
where we have used the Bianchi identity $`[^X,\mathrm{\Omega }^X]=0`$. $``$ $``$ Remark From Kuiperโs results \[Ku\] on the contractibility of the unitary group of a separable Hilbert space, we know that the orthonormal frame bundle $`O()`$ of a hermitian vector bundle with fibres modelled on a Hilbert space, is topologically trivial. This is the case for the class of manifolds we are investigating, and from this topological triviality one might expect that whenever the first Chern class is closed, teh corresponding characteristic class should vanish. This is not the case as we shall see shortly in the case of current groups, but in our setting the non triviality seems to come from the fact that we restict ourselves to P.D.Os. Indeed, the holonomy bundle is a reduction of the frame bundle and for a manifold $`X`$ in the class $`๐๐ณ`$ (recall that we only allowed transition maps which were P.D.Os) modelled on the Hilbert space $`H:=H^s(M,E)`$ for some $`s>\frac{dimM}{2}`$, the structure group of the frame bundle is
$$GL^{Ell}(H):=GL(H)PDO(M,E)=GL(H)Ell(M,E)=Ell_0^{}(M,E)$$
$`(5.9)`$
where $`Ell_0^{}(M,E)`$ denotes the group of invertible zero order elliptic operators acting on sections of $`E`$. But it is a well-known fact that the pathwise connected component of identity of $`Ell_0^{}(M,E)`$ has a non trivial fundamental group isomorphic to $`K_0(S^{}M)`$ where $`S^{}M`$ is the unit sphere in the cotangent bundle of $`M`$ (see Appendix B).
6. Weighted Lie groups In this section, we apply the constructions and results of the previous section to the case of weighted Lie groups, thus preparing for the next section where we will specialize to current groups. Here $`๐ข`$ is an infinite dimensional Hilbert Lie group in the class $`๐๐ณ`$ with Lie algebra $`Lie(๐ข)=H^s(M,E)`$ (for some $`s>\frac{dimM}{2}`$ and some hermitian vector bundle $`E`$ based on some manifold $`M`$). $``$ Left invariant weights A natural weight on $`๐ข`$ is given by a left invariant field of operators
$$Q(\gamma ):=L_{\gamma }^{}{}_{}{}^{}Q_0L_{\gamma }^{}{}_{}{}^{1}\gamma ๐ข$$
$`(6.1)`$
where $`Q_0Ell_{ord>0}^+(M,E)`$ is any weight on the Lie algebra $`Lie(๐ข)`$ and where $`L_\gamma `$ denotes left multiplication. As a consequence of Proposition 4 we have: Corollary 5Let $``$ be a left invariant connection on $`๐ข`$ induced by a left invariant one form $`\theta _0Lie(๐ข)Hom(Lie(๐ข))`$. Then, given a left invariant $`p`$-form $`\omega `$ on $`๐ข`$ which we identify with $`\omega _0\mathrm{\Lambda }^p(Lie๐ข)PDO(M,E)`$, we have:
$$[,tr^Q](\omega ):=dtr^Q(\omega )tr^Q([,\omega ])=tr^{Q_0}([\theta _0,\omega _0])=\frac{1}{ordQ_0}res([logQ_0,\theta _0]\omega _0)$$
where the bracket is an operator bracket so that $`[\theta _0,\omega _0](X,X_1,\mathrm{},X_p)=[\theta _0(X),\omega _0(X_1,\mathrm{},X_p)].`$ Proof: Recall that for a left invariant $`p`$-form $`\alpha `$ on $`๐ข`$ we have:
$$d\alpha (X_1,\mathrm{},X_{p+1})=\underset{i<j}{}(1)^{i+j}\alpha ([X_i,X_j],X_1,\mathrm{},\widehat{X}_i,\mathrm{},\widehat{X}_j,\mathrm{},X_{p+1})$$
whereby the $`X_i^{}s`$ are left invariant vector fields. Applying this to $`\alpha =tr^Q(\omega )`$ and $`\alpha =\omega `$ and then taking the $`Q`$-weighted trace yields $`dtr^Q(\omega )=tr^Q(d\omega )`$. Hence
$$\begin{array}{cc}\hfill [,tr^Q](\omega )& :=dtr^Q(\omega )tr^Q([,\omega ])\hfill \\ & =tr^Q([\theta ,\omega ])\hfill \\ & =tr^{Q_0}([\theta _0,\omega _0])\hfill \\ & =\frac{1}{ordQ_0}res([logQ_0,\theta _0]\omega _0).\hfill \end{array}$$
$``$ Diagonal weights Let us now consider the particular case when $`E`$ is a trivial bundle $`E=M\times V`$ where $`V`$ is a finite dimensional hermitian vector space. We can restrict ourselves to (left invariant) diagonal weights, meaning by this that:
$$Q_0:=\overline{Q}_01_V$$
$`(6.2)`$
where $`1_V`$ is the identity opertor on $`V`$ so that (6.1) reads:
$$Q(\gamma )=L_\gamma _{}\left(\overline{Q}_01_V\right)L_\gamma _{}^1.$$
$`(6.3)`$
In that case, for a left invariant $`p`$-form induced by $`\omega _0`$ as in the above corollary we have:
$$\begin{array}{cc}\hfill tr^Q(\omega )& =tr^{Q_0}(\omega _0)\hfill \\ & =Lim_{z0}tr(\omega _0(Q_0+P_{Q_0})^z)\hfill \\ & =Lim_{z0}tr(\omega _0(\overline{Q}_0+P_{\overline{Q}_0})^z1)\hfill \\ & =Lim_{z0}tr((\overline{Q}_0+P_{\overline{Q}_0})^ztr_V(\omega _0))\hfill \\ & =tr^{\overline{Q}_0}(tr_V(\omega _0))\hfill \end{array}$$
$`(6.4)`$
where $`tr_V`$ denotes the finite dimensional trace on the Lie algebra $`V`$. Hence, provided $`tr_V(\omega _0)`$ is itself a trace-class operator (i.e a P.D.O of order $`<\frac{dimM}{2}`$) then
$$tr^Q(\omega )=tr(tr_V(\omega _0))$$
$`(6.5)`$
(where $`tr`$ is an ordinary trace) is independent of $`\overline{Q}_0`$. This is the type of situation we shall be working with in these notes. Since there is no weight dependence in that case, we can choose any operator $`\overline{Q}_0`$ and in particular the Laplace operator $`\mathrm{\Delta }`$ on the Riemannian manifold $`M`$.
7. The case of current groups We now specialize to the current group $`H^s(M,G)`$ ($`s>\frac{dimM}{2}`$) of $`H^s`$ maps from the compact Riemannian manifold $`M`$ to a semi-simple Lie group $`G`$ of compact type (which ensures that the Killing form is non degenerate and that the adjoint representation $`ad`$ on the Lie algebra is antisymmetric for this bilinear form). $`H^s(M,G)`$ is a Hilbert Lie group with Lie algebra $`H^s(M,Lie(G))`$ where $`Lie(G)`$ denotes the Lie algebra of $`G`$ so that the current group $`H^s(M,G)`$ is a manifold in the class $`๐๐ณ`$, the underlying finite rank vector bundle $`E`$ being trivial since $`E=M\times Lie(G)`$ and we equip $`H^s(M,G)`$ with the left invariant and diagonal weight $`Q`$ (see (6.2) with $`V=Lie(G)`$):
$$Q(\gamma ):=L_\gamma (\overline{Q}1_{Lie(G)})L_\gamma ^1=L_\gamma \left(\mathrm{\Delta }1_{Lie(G)}\right)L_\gamma ^1\gamma H^s(M,G).$$
$`(7.1)`$
$``$ A left invariant metric The operator $`Q_0:=\mathrm{\Delta }1_{Lie(G)}`$ is an elliptic operator of order $`2`$ acting densely on $`H^s(M,Lie(G))`$. It is positive for the scalar product
$$,_0:=_M๐vol(x)(,)_{Lie(G)}$$
$`(7.2)`$
where $`(,)_{Lie(G)}`$ is a scalar product on $`Lie(G)`$ given by minus the Killing form. $`H^s(M,G)`$ is equipped with a left invariant metric defined in terms of the following scalar product on $`H^s(M,Lie(G))`$:
$$,_0^s:=(Q_0+P_{Q_0})^{\frac{s}{2}},(Q_0+P_{Q_0})^{\frac{s}{2}}_0.$$
$`(7.3)`$
$``$ The Levi-Civita connection The corresponding the Levi-Civita connection is a left invariant connection $`^s=d+\theta ^s`$ where $`\theta ^s`$ is a left invariant $`Hom(H^s(M,Lie(G))`$ valued one form on $`H^s(M,G)`$ given by a map $`\theta _0^s:Lie(๐ข)Hom(Lie๐ข)`$ (see \[F1\] formula (1.9) up to a sign mistake):
$$\theta _0^s(U)=\frac{1}{2}\left(ad_U+(Q_0+P_{Q_0})^sad_U(Q_0+P_{Q_0})^s(Q_0+P_{Q_0})^sad_{(Q_0+P_{Q_0})^sU}\right)$$
$`(7.4)`$
where $`U`$ is an element of the Lie algebra $`H^s(M,Lie(G))`$. $``$ The curvature The curvature $`\mathrm{\Omega }^s`$ is a left invariant two form given by an element $`\mathrm{\Omega }_0^s\mathrm{\Lambda }^2(H^s(M,Lie(G))Hom(H^s(M,Lie(G))`$:
$$\mathrm{\Omega }_0^s(U,V)=\theta _0^s\theta _0^s(U,V)=[\theta ^s(U),\theta ^s(V)]\theta _0^s([U,V])U,VH^s(M,Lie(G)).$$
$`(7.5)`$
Warning We shall henceforth identify left invariant forms with forms on the Lie algebra $`H^s(M,Lie(G))`$ so that for left invariant fields $`X,Y`$ induced respectively by elements $`X_0,Y_0H^s(M,Lie(G))`$ we have $`\theta ^s(X)=\theta _0^s(X_0)`$ and $`\mathrm{\Omega }^s(X,Y)=\mathrm{\Omega }_0^s(X_0,Y_0)`$. D.Freed shows (\[F1\] prop. 1.14) that for smooth $`X,YH^s(M,Lie(G))`$ the map
$$R^s(X,Y):Z\mathrm{\Omega }^s(Z,X)Y$$
is a pseudo-differential operator (with Sobolev coefficients) of order $`max(1,2s)`$. Freed in fact proves the result for smooth $`X`$ and $`Y`$ but his proof easily extends to the case when $`X,YH^s(M,Lie(G))`$ since it involves a counting of the order of the operator $`R^s(X,Y)`$ which is independent of the degree of regularity of its coefficients. Notice that for $`s\frac{1}{2}\mathrm{IN}\{0\}`$, we have $`max(1,2s)=1`$. We further quote from \[F1\]: Lemma 5 For $`X,YH^s(M,Lie(G))`$ the operator $`tr_{Lie(G)}R^s(X,Y)`$ is a classical pseudo-differential operator of order $`2q`$ where $`q=min(1,2s)`$ and where as before $`tr_{Lie(G)}`$ denotes the finite dimensional trace on $`Lie(G)`$. Proof: We refer the reader to \[F1\] Proposition 1.18 where this result was proven in the case when $`X,YC^{\mathrm{}}(M,Lie(G))`$. $``$ $``$ The Ricci curvature Following the general procedure described in section 5, let us define a weighted Ricci curvature on the Lie group $`H^s(M,G)`$ and compare it to that of \[F1\]. For any diagonal weight $`Q`$, as a consequence of (6.5) we have:
$$Ricc^{s,Q}(X,Y):=tr^Q(R^s(X,Y))=tr^{\overline{Q}_0}\left(tr_{Lie(G)}(R^s(X,Y))\right).$$
$`(7.6)`$
Hence provided $`min(1,2s)>\frac{dimM}{2}`$, using Lemma 5 we can write:
$$Ricc^{s,Q}(X,Y)=tr(tr_{Lie(G)}(R^s(X,Y)))$$
$`(7.7)`$
wher $`tr`$ is now an ordinary trace, since in that case, $`tr_{Lie(G)}(R^s(X,Y))`$ is trace-class. This holds in particular for any loop group $`H^s(S^1,G)`$ with $`s>\frac{1}{2}`$. It is easy to check that in that case and for any left invariant weight $`Q`$ the Wodzicki residue $`res(R^s(X,Y))`$ of the operator $`R^s(X,Y)`$ vanishes and that the $`Q`$-weighted Ricci curvature is given by an ordinary limit:
$$Ricc^{s,Q}(X,Y)=lim_{z0}(TR(R^s(X,Y)Q^z))$$
$`(7.8)`$
however this limit depends on the choice of $`Q`$. $``$ The case when $`s`$ is an integer and $`M`$ is odd dimensional There are some cases for which the $`Q`$-dependence vanishes; we shall need a preliminary lemma to single them out. Lemma 6 For two left invariant vector fields $`X,Y`$ on $`H^s(M,G)`$ and $`s\mathrm{IN}`$ the operator $`R^s(X,Y)`$ lies in the odd-class. Proof From the expression of the Levi-Civita curvature (7.5) one sees that the operator $`R^s(X,Y)`$ is built up from compositions and linear combinations of operators in the odd class provided $`Q_0`$ is in the odd class, since $`ad_X`$ and $`adX`$ both lie in the odd class since they are built up from multiplication operators. Since for any integer $`s`$, the operator $`Q_0^s`$ does lie in the odd class and since the product of odd-class operators lies in the odd class, the result follows.
Proposition 5 When the underlying manifold is odd dimensional and $`s`$ is an integer, the (left invariant) weighted Ricci curvature on $`H^s(M,G)`$ equipped with the left invariant connection (7.4) is independent of the choice of the weight $`Q`$ among operators in the odd class and we have for any two left invariant vector fields $`X`$ and $`Y`$ on $`H^s(M,G)`$:
$$Ricc^{s,Q}(X,Y)=TR_{odd}(R^s(X,Y)).$$
$`(7.9)`$
Proof : As we saw in section 1 (formula (1.6)), the dependence on the choice of $`Q`$ is measured in terms of a residue since for two weights $`Q_1`$ and $`Q_2`$ we have:
$$tr^{Q_1}(R^s(X,Y))tr^{Q_2}(R^s(X,Y))=q^1\left(res(R^s(X,Y)(logQ_1logQ_2))\right)$$
where $`q`$ is the common order of $`Q_1`$ and $`Q_2`$. If both $`Q_1`$ and $`Q_2`$ lie in the odd-class so does $`logQ_1logQ_2`$ (see Proof of Proposition 4.1 in \[KV1\]), and since $`R^s(X,Y)`$ also lies in the odd class, the operator $`R^s(X,Y)(logQ_1logQ_2))`$ lies in the odd-class. Thus in odd dimensions its Wodzicki residue vanishes. $``$ There is a priori no reason for a similar property to hold for $`s=\frac{1}{2}`$, which is the case we shall be focusing on in the following. The $`Q`$-weighted Ricci curvature will in general depend on the choice of $`Q`$.
8. The case of the based loop group $`H_e^{\frac{1}{2}}(S^1,G)`$ We now specialize to the case when $`M=S^1`$ and $`s=\frac{1}{2}`$, applying the results of the previous section to a based loop group. Notice that for $`s=\frac{1}{2}`$, $`H^s`$ maps are not even continuous in that case, and in practice, one works with $`H^{\frac{1}{2}+ฯต},ฯต>0`$ in order to have continuous objects. The space $`H_e^s(S^1,G)H^s(S^1,G)`$ of $`G`$ valued loops with value $`e_G`$ at a given point, where $`e_G`$ is the identity element in $`G`$. It is a Hilbert manifold with Lie algebra the based loop algebra $`H_0^s(S^1,Lie(G))`$ of maps in the loop algebra which all coincide at point $`0`$. $`H_e^s(S^1,G)`$ is equipped with an almost complex structure, for which the metric is in fact Kรคhler when $`s=\frac{1}{2}`$ \[F1\]. We equip it as before with a left invariant diagonal weight $`Q`$, the operator $`Q`$ being the same one used to define the connection $`\theta ^s`$. $``$ An almost complex structure on $`H_e^{\frac{1}{2}}(S^1,G)`$ We introduce a Dirac operator $`\overline{D}_0=z\frac{d}{dz}=i\frac{d}{dt}`$ (with $`z=e^{it}`$) acting on $`C^{\mathrm{}}(S^1,\mathrm{IC})`$ and we set $`D_0:=\overline{D}_01_{Lie(G)}=z\frac{d}{dz}1_{Lie(G)}`$, $`D`$ denoting the left invariant field of operators generated by $`D_0`$. We shall choose $`Q_0:=D_0^2=\mathrm{\Delta }1_{Lie(G)}`$ where $`\mathrm{\Delta }`$ is the Laplacian on functions, to define the left invariant diagonal weight $`Q`$. $`D_0`$ is injective when restricted to based loops and the sign $`ฯต(D_0):=D_0|D_0|^1`$ of the Dirac operator is a pseudo-differential operator of order $`0`$ which yields a conjugation on $`H_0^{\frac{1}{2}}(S^1,Lie(G))`$ since $`ฯต(D_0)^2=1`$. We have the splitting:
$$H:=H_0^{\frac{1}{2}}(S^1,Lie(G))=H_+H_{}$$
$`(8.1)`$
where
$$H_+:=Ker(ฯต(D_0)1)=\pi _+\left(H_0^{\frac{1}{2}}(S^1,Lie(G))\right)$$
and
$$H_{}:=Ker(ฯต(D_0)+1)=\pi _{}\left(H_0^{\frac{1}{2}}(S^1,Lie(G))\right).$$
Here $`\pi _+`$ and $`\pi _+`$ are the orthogonal projections w.r. to the scalar product $`,_0^{\frac{1}{2}}`$ defined in (7.3) $`\pi _+:=\frac{ฯต(D_0)+1}{2}`$,$`\pi _{}:=\frac{ฯต(D_0)+1}{2}`$. The set $`\{u_n(t):=e^{int},n\mathrm{Z}\mathrm{Z}\}`$ yields a C.O.N.S of eigenvectors of $`D_0`$ corresponding to the set of eigenvalues $`\{\lambda _n:=n,n\mathrm{Z}\mathrm{Z}\}`$. Then the set $`\{u_n^+(t):=e^{int},n\mathrm{IN}\}`$ spans $`H_+`$ and the set $`\{u_n^{}(t):=e^{int},n\mathrm{IN}\}`$ spans $`H_{}`$. The map $`J_0`$ defined by $`J_0u_n^+:=iu_n^+`$, $`J_0u_n^{}:=iu_n^{}`$ or equivalently by $`J_0:=iฯต(D_0)`$ obeys the relation $`J_0^2=1`$ and yields a natural almost complex structure on $`H`$ for which the $`(1,0)`$ part $`H^{1,0}:=Ker(J+i)`$ coincides with $`H_+`$, the $`(0,1)`$ part $`H^{1,0}:=Ker(Ji)=H_+`$ with $`H_{}`$. By left invariance of the metric on $`H_0^{\frac{1}{2}}(S^1,Lie(G))`$, this gives rise to a left invariant almost complex structure $`J(\gamma ):=L_\gamma J_0L_\gamma ^1`$ on the Lie group $`H_e^{\frac{1}{2}}(S^1,G)`$. $``$ The Kรคhler connection on $`H_e^{\frac{1}{2}}(S^1,G)`$ $`H_e^{\frac{1}{2}}(S^1,G)`$ is equipped with a left invariant symplectic (hence closed see \[Pr\]) form:
$$\omega (X,Y):=_0^1X_0^{},Y_0๐t$$
where $`X,Y`$ are two left invariant vector fields generated by $`X_0,Y_0`$ and where the โprimeโ denotes derivation with respect to $`t`$. One can check that the invariant bilinear form given by $`B(X,Y):=\mathrm{\Omega }(X,JY)`$ yields back the $`H^{\frac{1}{2}}`$ Sobolev metric given by (7.3). The associated left invariant hermitian form reads:
$$,_0^{\frac{1}{2}}:=,ฯต(D_0)_0^{\frac{1}{2}}=D_0,_{L^2}.$$
$`(8.2)`$
where we have used the fact that $`ฯต(D_0)_0Q_0^{\frac{1}{2}}=ฯต(D_0)|D_0|=D_0`$ and where $`,,_{L^2}`$ denotes the $`L^2`$ hermitian product. Hence $`S`$ is a Kรคhler form on the manifold $`H^{\frac{1}{2}}(S^1,G)`$ equipped with the $`H^{\frac{1}{2}}`$ left invariant Riemannian metric given by (7.3) and the left invariant complex structure $`J`$ defined above. As a consequence, the Levi-Civita connection $`^{\frac{1}{2}}`$ on $`H_e^{\frac{1}{2}}(S^1,G)`$ is Kรคhlerian, meaning by this that $`[^{\frac{1}{2}},J]=0`$. Since $`^{\frac{1}{2}}`$ commutes with $`J`$, $`\theta ^{\frac{1}{2}}`$ defined in (7.4) (with $`s=\frac{1}{2}`$ ) stablizes each of the spaces $`H_+`$ and $`H_{}`$. Its restriction to $`H_+`$:
$$\varphi _0:=\left[\theta _0^{\frac{1}{2}}\right]_{++}$$
$`(8.3)`$
defines a left invariant the (complex) Kรคhler connection $`\varphi `$. In the next lemma we use the splitting $`H=H_+H_{}`$ to write any operator $`AHom(H_0^{\frac{1}{2}}(S^1,Lie(G))`$ as a matrix:
$$A=\left[\begin{array}{cc}A_{++}& A_+\\ A_+& A_{}\end{array}\right].$$
Following \[F1\] we introduce the Tรถplitz operators $`T_X:=(ad_X)_{++},XH`$. Lemma 7 For $`UH_+`$,
$$\varphi (U)=D^1T_UD.$$
For $`\overline{V}H_{}`$,
$$\varphi (\overline{V})=T_{\overline{V}}.$$
Proof: 1) For $`UH_+`$, we have $`|D_0|U=D_0U.`$ Hence, setting $`U=\alpha a`$, $`V:=\beta b`$, $`\alpha ,\beta H^{\frac{1}{2}}(S^1,\mathrm{IC})`$, $`a,bLie(G)`$, we have:
$$\begin{array}{cc}\hfill |D_0|^1ad_{|D_0|U}V& =|D_0|^1(D_0\alpha )\beta ad_ab\hfill \\ & =|D_0|^1D_0(\alpha \beta )ad_ab|D_0|^1\alpha D_0\beta ad_ab\hfill \\ & =|D_0|^1D_0ad_U|D_0|^1ad_UD_0\hfill \\ & =ฯต(D_0)ad_U|D_0|^1ad_UD_0.\hfill \end{array}$$
Inserting this into (7.4) yields:
$$\theta ^{\frac{1}{2}}(U)=\frac{1}{2}(ad_U+|D_0|^1ad_U|D_0|+|D_0|^1ad_UD_0ฯต(D_0)ad_U)$$
which, when restricted to $`H_+`$ and using (8.3) leads to:
$$\begin{array}{cc}\hfill \varphi _0(U)& =\pi _{}ad_{U}^{}{}_{|H_+}{}^{}+|D_0|^1ad_UD_{0}^{}{}_{|H_+}{}^{}\hfill \\ & =D_0^1T_UD_0\hfill \end{array}$$
since $`ad_U`$ stablizes $`H_+`$ for $`UH_+`$. 2) For $`\overline{V}H_{}`$, $`|D_0|\overline{V}=D_0\overline{V}`$ and hence in a similar manner as above we have:
$$|D_0|^1ad_{|D_0|\overline{V}}=|D_0|^1ad_{\overline{V}}D_0ฯต(D_0)ad_{\overline{V}}$$
which yields in turn:
$$\theta ^{\frac{1}{2}}(\overline{V})=\frac{1}{2}(ad_{\overline{V}}+ฯต(D)ad_{\overline{V}}|D|^1ad_{\overline{V}}D+|D|^1ad_{\overline{V}}|D|).$$
When restricted to $`H_+`$ and using (8.3), this reads:
$$\varphi (\overline{V})=\pi _+ad_{\overline{V}}^{}{}_{|_{H_+}}{}^{}=T_{\overline{V}}.$$
9. The first Chern form on $`H_e^{\frac{1}{2}}(S^1,G)`$ and the cohomology of $`PDO(S^1,Lie(G))`$ $``$ The first Chern form on $`H_e^{\frac{1}{2}}(S^1,G)`$ The curvature (which also stablizes $`H_+`$ ) is given as in (7.5) by:
$$\mathrm{\Omega }(X,\overline{Y}):=[\varphi (X),\varphi (\overline{Y})]\varphi ([X,\overline{Y}])=\left(\mathrm{\Omega }^{\frac{1}{2}}(X,\overline{Y})\right)^{1,0}.$$
$`(9.1)`$
From \[F1\] (see Remarks above Theorem 2.20 ), we also know that $`\mathrm{\Omega }(X,\overline{Y})`$ is a classical pseudo-differential operator of order $`1`$ and hence that it is not trace-class. However $`tr_{Lie(G)}\mathrm{\Omega }(X,\overline{Y})`$ being of order $`2`$ (see \[F1\] Remarks above Theorem 2.20), is trace-class. We now define the weighted first Chern form on $`H_e^{\frac{1}{2}}(S^1,G)`$ according to (5.7) where the weighted Kรคhler manifold here is $`(H_e^{\frac{1}{2}}(S^1,G),Q)`$:
$$r_1^Q:=tr^Q(\mathrm{\Omega })$$
Here the complex curvature is given by (9.1) and the trace is taken w.r. to the hermitian metric. If the weight is diagonal then
$$r_1^Q=tr(tr_{LieG}\mathrm{\Omega }).$$
This follows from (6.5) and shows our definition of weighted first Chern form coincides (with a good choice of the weight) with the โtwo step traceโ used by Freed \[F1\] to make sense of the trace of the curvature (see Theorem 2.20 in \[F1\]). In fact only the $`(1,0)`$ part $`Q^{1,0}:=Q_{++}`$ of $`Q`$ comes into play in the expression $`tr^Q(\mathrm{\Omega })`$ since $`\mathrm{\Omega }`$ is the $`(1,0)`$ part of $`\mathrm{\Omega }^{\frac{1}{2}}`$. Before we give an expression of the first Chern form, we need wome preliminary results. Lemma 8 For the diagonal weight $`Q`$ on $`H^{\frac{1}{2}}(S^1,G)`$ chosen as in (7.1), we have
$$tr^Q(\varphi (Z))=0ZH^{\frac{1}{2}}(S^1,Lie(G)).$$
$`(9.2)`$
Proof: It is sufficient to establish the formula for $`Z`$ of the type $`\gamma cH^s(M,\mathrm{IR})Lie(G)`$. If $`Z=\gamma cH_{}`$, then $`\varphi (Z)=(ad_Z)_{++}`$ and hence (compare with (6.5))
$$\begin{array}{cc}\hfill tr^Q(\varphi (Z))& =tr^Q((ad_Z)_{++})\hfill \\ & =Lim_{z0}tr((ad_Z)_{++}\overline{Q}_0^z1_{Lie(G)}))\hfill \\ & =Lim_{z0}tr\left((M_\gamma \overline{Q}_0^zad_c)_{++}\right)\hfill \\ & =tr^{\overline{Q}_0}(M_\gamma )_{++}tr_{LieG}(ad_c)\hfill \\ & =0\hfill \end{array}$$
since $`ad_c`$ is antisymmetric. $`M_\gamma `$ denotes the multiplication operator by $`\gamma `$ and $`Lim`$ the renormalized limit. Here we have used the fact that $`Q_0`$ stablizes $`H_+`$. If $`ZH_+`$, then $`\varphi (Z)=D^1T_ZD=D^1(ad_Z)_{++}D`$ and
$$tr^Q(D^1(ad_Z)_{++}D)=tr^Q((ad_Z)_{++})=0.$$
Proposition 6 Let $`H_e^{\frac{1}{2}}(S^1,G)`$ be equipped with a left invariant diagonal weight $`Q`$. The $`Q`$-weighted first Chern form is the pull-back by $`\varphi `$ of the $`Q_0`$ Radul cocycle $`c_R^{Q_0}`$ on $`PDO(S^1,Lie(G))`$ and coรฏncides with Freedโs conditioned first Chern form \[F1\]. For any $`X,YH_0^{\frac{1}{2}}(S^1,Lie(G))`$ induced by $`X_0H_+`$ and $`\overline{Y}_0H_{}`$ we have:
$$r_1^Q(X,\overline{Y})=c_R^{Q_0}(\varphi (X),\varphi (\overline{Y}))$$
$`(9.3)`$
where $`tr`$ denotes the ordinary trace and $`\lambda ^D`$ was defined in formula (4.1). Proof: The first term in the expression (9.1) of the curvature is a bracket of P.D.Os so that taking the weighted pseudo-trace yields $`tr^{Q_0}[\varphi (X),\varphi (\overline{Y})]=c_R^{Q_0}(\varphi (X),\varphi (\overline{Y}))`$. The trace of the second term in (9.1) vanishes by lemma 7 and we find
$$r_1^Q(X,\overline{Y})=tr^Q(\mathrm{\Omega }(X,\overline{Y}))=tr^{Q_0}([\varphi (X),\varphi (\overline{Y})]=c_R^{Q_0}(\varphi (X),\varphi (\overline{Y}))$$
as announced. $``$ Before we go on to the next proposition, let us recall various ways of expressing the Killing form on the Lie algebra $`Lie(G)`$:
$$\begin{array}{cc}\hfill a,b_{Lie(G)}& =tr_{Lie(G)}(ad_b^{}ad_a)\hfill \\ & =\underset{i=1}{\overset{d}{}}[a,c_i],[b,c_i]_{Lie(G)}\hfill \\ & =tr_{Lie(G)}[a,[b,]]\hfill \\ & =tr_{Lie(G)}[b,[a,]].\hfill \end{array}$$
where $`c_i,i=1,\mathrm{},d`$ varies in an O.N.B of $`Lie(G)`$. Proposition 7 The map:
$$\begin{array}{cc}\hfill ad:C^{\mathrm{}}(S^1,Lie(G))& C^{\mathrm{}}(S^1,Hom(Lie(G)))\hfill \\ \hfill X:=\underset{n\mathrm{Z}\mathrm{Z}}{}a_nz^n& ad_X=\underset{n\mathrm{Z}\mathrm{Z}}{}ad_{a_n}z^n\hfill \end{array}$$
extends to a map:
$$ad:H^{\frac{1}{2}}(S^1,Lie(G))g_{res}^D$$
with the notations of (3.3) replacing $`M`$ by $`S^1`$ and $`E`$ by the trivial bundle $`S^1\times Lie(G)`$. In particular, $`[ad_X]_+`$ and $`[ad_X]_+`$ are Hilbert-Schmidt operators on $`L^2(S^1,Lie(G))`$. Proof: (this proof is close to that of proposition 6.3.1 in chapter 6 of \[PS\]) For $`bL`$ , $`q\mathrm{Z}\mathrm{Z}`$ and $`X=_{n\mathrm{Z}\mathrm{Z}}a_nz^nC^{\mathrm{}}(S^1,Lie(G))`$ we have:
$$\begin{array}{cc}\hfill ad_X(bz^q)& =\underset{n\mathrm{Z}\mathrm{Z}}{}ad_{a_n}(b)z^{n+q}\hfill \\ & =\underset{p\mathrm{Z}\mathrm{Z}}{}ad_{a_{pq}}(b)z^p\hfill \end{array}$$
so that $`ad_X`$ is represented by an $`\mathrm{Z}\mathrm{Z}\times \mathrm{Z}\mathrm{Z}`$ matrix with entries $`ad_{X}^{}{}_{p,q}{}^{}=ad_{a_{pq}}Hom(Lie(G))`$. We compute the Hilbert-Schmidt norm of $`(ad_{X}^{}{}_{+}{}^{}`$ denoted by $`_{HS}`$. We shall use the fact that for $`aLie(G)`$ we have $`ad_a_{HS}^2=a^2`$ where this last norm is the one corresponding to the Killing form.
$$\begin{array}{cc}\hfill ad_{X}^{}{}_{+}{}^{}_{HS}^2& =\underset{q<0}{}\underset{i=1}{\overset{dimLie(G)}{}}(ad_X)_+b_iz^q,(ad_X)_+b_iz^q\hfill \\ & =\underset{p0,q<0}{}\underset{i=1}{\overset{dimLie(G)}{}}ad_{a_{pq}}(b_i)z^p,ad_{a_{pq}}(b_i)z^p\hfill \\ & =\underset{p0,q<0}{}\underset{i=1}{\overset{dimLie(G)}{}}ad_{a_{pq}}(b_i),ad_{a_{pq}}(b_i)_{Lie(G)}\hfill \\ & =\underset{p0,q<0}{}ad_{X}^{}{}_{p,q}{}^{}_{HS}^2\hfill \\ & =\underset{k>0}{}kad_{a}^{}{}_{k}{}^{}_{HS}^2\hfill \\ & =\underset{k>0}{}ka_k^2\hfill \end{array}$$
where $`(b_i)_{i=1,\mathrm{},dimLie(G))}`$ is an orthonormal basis of $`Lie(G)`$. In the same way, we have:
$$ad_{X}^{}{}_{+}{}^{}_{HS}^2=\underset{k<0}{}a_k^2|k|.$$
From the above expressions of the Hilbert-Schmidt norms of $`ad_{X}^{}{}_{+}{}^{}`$ and $`ad_{X}^{}{}_{+}{}^{}`$, it follows that these are Hilbert-Schmidt whenever $`_{k\mathrm{Z}\mathrm{Z}}|k|a_k^2`$ is finite. But this is the condition for $`X`$ to lie in $`H^{\frac{1}{2}}(S^1,L)`$. $``$ The first Chern form in terms of the Kรคhler form Recall that the exterior derivative of a form can be expressed in terms of a Wodzicki residue (see (5.8)). Unlike in the finite dimensional setting, here the weighted first Chern form $`r_1^Q`$ might therefore not be closed. The following proposition shows that it relates to two closed forms, the Kรคhler form $`\omega `$ on the one hand (this fact had already been shown by Freed) and the pull-back by the adjoint representation of the cocycle $`\lambda ^D`$ on the other hand. Lemma 9 The pull-back $`ad^{}\lambda ^D`$ of the cochain $`\lambda ^D`$ defined by (4.1) is closed on $`H^{\frac{1}{2}}(S^1,Lie(G))`$. Proof: By Proposition 7, we know that $`ad`$ takes its values in $`g_{res}^D`$. On the other hand $`\lambda _D`$ is indeed a cocycle on $`g_{res}^D`$. Combining these two facts yields, for three left invariant vector fields $`X,Y,Z`$:
$$\begin{array}{cc}\hfill \delta (ad^{}\lambda ^D)(X,Y,Z)& =ad^{}\lambda ^D([X,Y],Z)ad^{}\lambda ^D([Y,Z],X)+ad^{}\lambda ^D([Z,X],Y)\hfill \\ & =\delta \lambda ^D(ad_X,ad_Y,ad_Z)\hfill \\ & =0\hfill \end{array}$$
where $`\delta \lambda ^D`$ denotes the coboundary of $`\lambda ^D`$. Proposition 8 The first Chern form is given by:
$$r_1^Q(U,\overline{V})=ad^{}\lambda ^D(U,\overline{V})=i\omega (U,\overline{V})$$
for $`U,VH^{\frac{1}{2}}(S^1,Lie(G))`$ and it is a closed form. Here $`\omega `$ is the symplectic form on $`H_e^{\frac{1}{2}}(S^1,G)`$ defined in section 8. Proof: From the results of Proposition 7 combined with the description of $`\lambda ^D`$ given in Lemma 3, we know that $`\lambda ^D(ad_U,ad_{\overline{Y}})=tr(ad_{\overline{V}}^{}{}_{+}{}^{}ad_{U}^{}{}_{+}{}^{}ad_{U}^{}{}_{+}{}^{}ad_{\overline{V}}^{}{}_{+}{}^{})`$ is an ordinary trace and hence it reads:
$$\lambda ^D(ad_U,ad_{\overline{V}})=\underset{q\mathrm{IN},i=1,\mathrm{},d}{}\left(ad_{\overline{V}}^{}{}_{+}{}^{}ad_{U}^{}{}_{+}{}^{}z^qc_i,z^qc_iad_{U}^{}{}_{+}{}^{}ad_{\overline{V}}^{}{}_{+}{}^{}z^qc_i,z^qc_i\right)$$
where $`c_i,i=1,\mathrm{},d`$ varies in an O.N.B of $`Lie(G)`$. Since $`L^2(S^1,Lie(G))`$ is spanned by elements of the type $`z^na,n\mathrm{Z}\mathrm{Z},aLie(G)`$, it is sufficient to compute this last sum for $`U=z^na`$, $`\overline{V}=z^pb`$, $`a,bLie(G)`$, $`n\mathrm{Z}\mathrm{Z},p\mathrm{Z}\mathrm{Z}`$. For $`q\mathrm{IN},cLie(G)`$ we have:
$$(ad_U)_+(z^qc)=\begin{array}{cc}0& \text{ if }n+q>0\\ [a,c]z^{n+q}& \text{ if }n+q0\end{array}$$
and
$$(ad_{\overline{V}})_+(ad_U)_+(z^qc)=\begin{array}{cc}0& \text{ if }n+q>0\text{or}np+q<0\\ [b,[a,c]]z^{np+q}& \text{ if }n+q0\text{ and}np+q0.\end{array}$$
In the same way
$$(ad_{\overline{V}})_+(z^qc)=\begin{array}{cc}0& \text{ if }p+q>0\\ [b,c]z^{n+q}& \text{ if }p+q0\end{array}$$
and
$$(ad_U)_+(ad_V)_+(z^qc)=\begin{array}{cc}0& \text{ if }p+q>0\text{or}np+q<0\\ [a,[b,c]]z^{np+q}& \text{ if }p+q0\text{ and}np+q0.\end{array}$$
Finally this yields (since the only non vanishing terms correspond to $`np=0`$):
$$\begin{array}{cc}\hfill \lambda ^D(ad_U,ad_{\overline{V}})& =\underset{0<qn,i=1,\mathrm{},d}{}[a,[b,c_i]]z^q,c_iz^q\hfill \\ & =ntr_{Lie(G)}(c[a,[b,c]])\hfill \\ & =na,b_{Lie(G)}\hfill \\ & =i\omega (z^na,z^nb)\hfill \\ & =i\omega (U,\overline{V})\hfill \end{array}$$
which yields the result. A similar computation \[F1\] (see Theorem 2.20) shows that the weighted first Chern form can be expressed in terms of the symplectic form $`\omega `$ and yields the result.
Appendix A. Classical elliptic pseudo-differential operator This Appendix gives a brief presentation of the basic tools in our framework, namely classical pseudo-differential operators and particularly elliptic ones, their logarithms and their complex powers. Classical references are \[G\], \[Se\], \[Sh\]. $``$The symbol set Let $`U`$ be an open subset of $`\mathrm{IR}^d`$. Given $`\alpha \mathrm{IC}`$, let us denote by $`S^\alpha (U)`$ the set of complex valued smooth function
$$\begin{array}{cc}\hfill U\times \mathrm{IR}^d& \mathrm{IR}\hfill \\ \hfill (U,\xi )& \sigma (U,\xi )\hfill \end{array}$$
satisfiying the following property. Given any compact subset $`K`$ of $`U`$ and any two multiindices $`\gamma =(\gamma _1,\mathrm{},\gamma _d)\mathrm{IN}^d`$, $`\delta =(\delta _1,\mathrm{},\delta _d)\mathrm{IN}^d`$ , there exists a constant $`C_{\alpha ,\beta }(K)`$ such that
$$|D_x^\gamma D_\xi ^\delta \sigma (x,\xi )|C_{\gamma ,\delta }(K)(1+|\xi |)^{Re\alpha |\delta |}xK,\xi \mathrm{IR}^d.$$
An element of $`S^\alpha (x)`$ is called a symbol of order $`\alpha `$. Let $`S^m(x)`$ denote the set of symbols of order $`m`$. The principal part of the symbol $`\sigma S^\alpha (x)`$ (or principal symbol ) is defined as follows:
$$\sigma _\alpha (x,\xi )=\underset{t+\mathrm{}}{lim}\frac{\sigma (U,t\xi )}{t^\alpha }.$$
A smoothing symbol is a symbol in
$$S^{\mathrm{}}(x)\underset{k\mathrm{IN}}{}S^k(U)$$
and the relation
$$\sigma \stackrel{~}{\sigma }\sigma \stackrel{~}{\sigma }S^{\mathrm{}}(U)$$
defines an equivalence relation on $`S(x)`$. A symbol of order $`\alpha `$ is called a classical symbol if there exist $`\sigma _{\alpha j}S^{\alpha j}(x)`$, $`j\mathrm{IN}`$ such that:
$$\sigma (x,\xi )\underset{j=0}{\overset{\mathrm{}}{}}\sigma _{\alpha j}(x,\xi )$$
which are positively homogeneous, i.e
$$\sigma _{\alpha j}(x,t\xi )=t^{\alpha j}\sigma _{\alpha j}(x,\xi )t\mathrm{IR}^+.$$
Following Kontsevich and Vishik \[KV 1\], we shall say that a classical symbol lies in the odd-class if the positively homogeneous components $`\sigma _{\alpha j}`$ are moreover homogeneous i.e:
$$\sigma _{\alpha j}(x,t\xi )=t^{\alpha j}\sigma _{\alpha j}(x,\xi )t\mathrm{IR}.$$
$``$ From symbols to pseudo-differential operators To a symbol $`\sigma S^\alpha (U)`$ we can associate the pseudo-differential operator (P.D.O.) with symbol $`\sigma `$ defined by:
$$\begin{array}{cc}\hfill A:C_c^{\mathrm{}}(x)& \mathrm{IC}^{\mathrm{}}(\mathrm{IR}^d)\hfill \\ \hfill u& \left(UAu(x)=_{\mathrm{IR}^d}e^{i\xi xU}\sigma (U,\xi )\widehat{u}(\xi )๐\xi \right).\hfill \end{array}$$
where $`C_c^{\mathrm{}}(U)`$ denotes the space of complex valued smooth functions with compact support in $`U`$. The principal symbol of $`A`$ is given by the principal part $`\sigma _P(A)`$ of its symbol $`\sigma (A)`$. If the symbol is classical, we shall call the corresponding P.D.O classical. The set of classical P.D.Os of order $`\alpha `$ (resp. $`m`$) is denoted by $`PDO^\alpha (U)`$ (resp. $`PDO^m(U)`$). A smoothing P.D.O. is an operator in
$$PDO^{\mathrm{}}(U)\underset{k\mathrm{IN}}{}PDO^k(U)$$
and there is an exact sequence:
$$0PDO^{\mathrm{}}(U)PDO^m(U)S^m(U)0.$$
An ordinary differential operator of order $`m\mathrm{IN}`$ is defined by a polynomial symbol (the polynomial being of order $`m`$) in $`\xi `$:
$$\sigma (x,\xi )=\underset{j=0}{\overset{m}{}}a_k(x)\xi ^k.$$
A differential operator is local i.e $`u=0Au=0`$. But a pseudo-differential operator, because of the smearing produced by the Fourier transform, is not local. It is only pseudo-local i.e if $`u`$ is smooth on an open set $`U`$ then $`Pu`$ is also smooth on any open subset $`VU`$. The various classes of symbols introduced previously induce corresponding classes of pseudo-differential operators. A classical pseudo-differential is a P.D.O such that its symbol has components given by classical symbols and an odd-class classical P.D.O is a classical P.D.O such that its symbol has components given by symbols in the odd class. Notice that ordinary differential operators with integer order provide examples of P.D.Os in the odd class. $``$ pseudo-differential operators acting on sections of vector bundles The notion of pseudo-differential operator can be carried out to operators acting on sections of vector bundles. Let $`\pi _E:EM`$, $`\pi _F:FM`$ be two smooth vector bundles with rank $`r_E`$ and $`r_F`$ respectively based on a smooth manifold $`M`$ with dimension $`d`$. An operator
$$P:\mathrm{\Gamma }(M,E)\mathrm{\Gamma }(M,F)$$
acting from the space $`\mathrm{\Gamma }(M,E)`$ of smooth sections of $`E`$ to the space $`\mathrm{\Gamma }(M,F)`$ of smooth sections of $`F`$ is called a pseudo-differential operator of order $`\alpha `$ if given a neighborhood of any point $`mM`$, there is a local trivialization i.e a morphism:
$$\varphi :(U,U\times \mathrm{IC}^{r_E},U\times \mathrm{IC}^{r_F})(M,E,F)$$
where $`U`$ is an open subset of $`\mathrm{IR}^d`$, the induced linear map $`\varphi ^{\mathrm{}}A:C_c^{\mathrm{}}(U,U\times \mathrm{IC}^{r_E})C_c^{\mathrm{}}(U,U\times \mathrm{IC}^{r_F})`$ has a symbol $`\sigma (\varphi ^{\mathrm{}}A)C^{\mathrm{}}(U\times \mathrm{IR}^d)_{r_E,r_F}(\mathrm{IC})`$ with matrix components in $`S^\alpha (U)`$. Here $`_{k,l}`$ denotes the space of $`k\times l`$ matrices with coefficients in $`\mathrm{IC}`$. It is a classical P.D.O if $`\sigma (\varphi ^{\mathrm{}}A)`$ is a classical symbol. These definitions involve a choice of trivialization but can be shown to be independent of this choice. $`\sigma (\varphi ^{\mathrm{}}A)`$ is called the (formal) symbol of $`A`$ and is only defined locally. However its principal part $`\sigma _P(\varphi ^{\mathrm{}}A)`$ is independent of the choice of coordinate charts and is therefore defined globally. We shall denote it by $`\sigma _P(A).`$ It is called the principal symbol of the P.D.O. Let us denote by $`PDO^\alpha (M,E,F)`$ the space of all classical P.D.Os of order $`\alpha `$ and by $`PDO^m(M,E,F)`$ the space of all classical P.D.Os of order $`m`$. When $`E=F`$, we shall denote these spaces by $`PDO^\alpha (M,E)`$ (resp. $`PDO^m(M,E)`$ ) and when $`E`$ is the trivial bundle $`M\times \mathrm{IC}`$ by $`PDO^\alpha (M)`$ (resp.$`PDO^m(M)`$). The symbol set $`S^m(M,E,F)`$ is defined by the exact sequence:
$$0PDO^{\mathrm{}}(M,E,F)PDO^m(M,E,F)S^m(M,E,F)0$$
where $`PDO^{\mathrm{}}(M,E,F):=_{k0}PDO^k(M,E,F)`$. When $`M`$ is compact, their is a notion of product of two pseudo-differential operators and
$$PDO(M,E):=\underset{m\mathrm{Z}\mathrm{Z}}{}PDO^m(M,E)$$
defines an associative algebra. From now on we shall assume $`M`$ is a smooth compact manifold without boundary. $``$Admissible elliptic pseudo-differential operators Let $`E`$ and $`F`$ be as before two finite rank vector bundles based on a smooth manifold $`M`$. When $`r_E=r_F`$ a pseudo-differential operator $`P:\mathrm{\Gamma }(E)\mathrm{\Gamma }(F)`$ of order $`m`$ is called elliptic if its principal symbol $`\sigma _P^m(x,\xi )`$ is an invertible matrix for $`\xi 0`$. Let $`Ell(M,E)`$ denote the set of elliptic classical P.D.Os. Let us denote by $`Ell_{ord>0}^{}(M,E)`$ the class of invertible elliptic operators of strictly positive order. Since $`M`$ is compact the spectrum $`spec(A)`$ of such an operator consists of isolated eigenvalues with finite multiplicity \[Sh\]. There is therefore a disc $`D_R`$ of radius $`R>0`$ around the origin which does not contain any point of the spectrum. We shall say that $`A`$ has spectral cut $`L_\theta `$ if there is a ray $`L_\theta =\{\lambda \mathrm{IC},arg\lambda =\theta \}`$ in the complex plane which does not intersect the spectrum of $`A`$. Such an operator will be called admissible and we shall denote by $`Ell_{ord>0}^{,adm}(M,E)`$ the set of such admissible operators. Any invertible elliptic operator with strictly positive order such that the matrix given by its principal symbol has no eigenvalues in some non empty conical neighborhood $`\mathrm{\Lambda }`$ of a ray in the spectral plane is admissible since in that case at most a finite number of eigenvalues of the operator are contained in $`\mathrm{\Lambda }`$ \[Sh\]. Let us introduce some notations. $`Ell_{ord>0}^{s.a}(M,E)`$, resp. $`Ell_{ord>0}^+(M,E)`$ denotes the set of self-adjoint, resp. positive self-adjoint elliptic operators with strictly positive order. Adding an upper index $``$ restricts to injective operators so that $`Ell_{ord>0}^{s.a}(M,E)`$, resp. $`Ell_{ord>0}^+(M,E)`$ denotes the set of self-adjoint injective, resp. positive self-adjoint injective elliptic operators with strictly positive order and we have following inclusions:
$$Ell_{ord>0}^+(M,E)Ell_{ord>0}^{s.a}(M,E)Ell_{ord>0}^{,adm}(M,E).$$
$``$ Complex powers and logarithms of elliptic operators Let $`AEll_{>0}^{,adm}(M,E)`$ with spectral cut $`L_\theta `$. For $`Rez<0`$, the complex power $`A_\theta ^z`$ of $`A`$ is a bounded operator on any space $`H^s(M,E)`$ of sections of $`E`$ of Sobolev class $`H^s`$ defined by the contour integral:
$$A_\theta ^z=\frac{i}{2\pi }_{\mathrm{\Gamma }_\theta }\lambda ^z(A\lambda I)^1๐\lambda $$
where $`\mathrm{\Gamma }_\theta =\mathrm{\Gamma }_{1,\theta }\mathrm{\Gamma }_{2,\theta }\mathrm{\Gamma }_{3,\theta }`$ $`\mathrm{\Gamma }_{1,\theta }=\{\lambda =re^{i\theta },rR\}`$,$`\mathrm{\Gamma }_{2,\theta }=\{\lambda =Re^{i\varphi },\theta \varphi \theta \}`$, $`\mathrm{\Gamma }_{3,\theta }=\{\lambda =re^{i(\theta 2\pi )},rR\}`$. Here $`\lambda ^z=exp(zlog\lambda )`$ where $`log\lambda =log|\lambda |+i\theta `$ on $`\mathrm{\Gamma }_{1,\theta }`$ and $`log\lambda =log|\lambda |+i(\theta 2\pi )`$ on $`\mathrm{\Gamma }_{3,\theta }`$. This definition is independent of the choice of $`R`$ but depends on the choice of $`\theta `$ and yields for any $`z\mathrm{IC}`$ an elliptic operator $`A_\theta ^z`$ of order $`zord(A)`$. When $`z=k,k\mathrm{IN}`$, then $`A^z`$ coincides with $`A^k`$ of order $`kord(A)`$. When $`M`$ is Riemannian, $`E`$ is hermitian and $`A`$ is essentially self-adjoint, then $`A_\theta ^z`$ is independent of the choice of $`\theta `$ and coincides with the complex powers defined using spectral representation. In the following we shall focus on operators in $`Ell_{ord>0}^+(M,E)`$ in which case we shall use the principal branch of the logarithm, taking $`\theta =\pi `$ and simply drop the mention $`\theta `$. For arbitrary $`k\mathrm{Z}\mathrm{Z}`$, the map $`zA_\theta ^z`$ defines a holomorphic function from $`\{z\mathrm{IC},Rez<k\}`$ to the space of bounded linear maps $`(H^s(M,E)H^{skordA}(M,E))`$ for any $`s\mathrm{IR}`$ and we can set:
$$log_\theta A\left[\frac{}{z}A_\theta ^z\right]_{z=0}$$
which defines a (non classical) P.D.O operator of zero order and hence a bounded operator from $`H^s(M,E)`$ to $`H^{sฯต}(M,E)`$ for any $`ฯต>0`$ and any $`s\mathrm{IR}`$. In local coordinates $`(x,\xi )`$ on $`T^{}M`$, the symbol of $`log_\theta A`$ reads:
$$\sigma _{log_\theta A}(x,\xi )=ord(A)log|\xi |Id+\text{ a classical P.D.O symbol of order 0}.$$
Hence, although the logarithm of an injective elliptic classical pseudo-differential operator with admissible cut $`L_\theta `$ is not itself a classical pseudo-differential operator, for two operators $`AEll_{ord>0}^{,adm}(M,E)`$, $`BEll_{ord>0}^{,adm}(M,E)`$ admitting spectral cuts $`L_\theta `$ and $`L_\varphi `$:
$$\frac{log_\theta A}{ordA}\frac{log_\varphi B}{ordB}PDO^0(M,E).$$
Appendix B
In this appendix, we recall why the first fundamental group of $`Ell_0^{,0}(M,E)`$ is non trivial, where $`Ell_0^{,0}(M,E)`$ denotes the pathwise connected component of identity in the group of invertible zero order elliptic P.D.Os.
$`E`$ denotes a finite rank vector bundle based on a compact boundaryless Riemannian manifold $`M`$ equipped with a connection $``$. We keep the notations of Appendix A. To begin with, let us describe the topology on $`Ell_0^{}(M,E)`$. $``$A Frรฉchet structure on the algebra of P.D.Os and their symbols The space $`SPDO(M,E)`$ of symbols of classical pseudo-differential operators is a Frรฉchet space when equipped with the following family of semi-norms labelled by multiindices $`\gamma \mathrm{IN}^d`$, $`\delta \mathrm{IN}^d`$ and $`i\{1,\mathrm{},N\}`$, $`k\mathrm{IN}`$:
$$\sigma _{\gamma ,\delta ,k}:=max_i\left(sup_{y\overline{V_i},\xi \mathrm{IR}^d,\xi =1}D_y^\gamma D_\xi ^\delta \sigma _k(y,\xi )\right)$$
where $`\sigma _k`$ is the homogeneous component of order $`k`$, $`\{V_i\}_{i=1,\mathrm{},N}`$ is a finite open cover (with $`\overline{V}_i`$ compact) of $`M`$ associated to a partition of unity $`\{V_i,\xi _i\}_{i=1,\mathrm{},N}`$ on $`M`$ subordinated to some finite open covering $`\{U_i\}_{i=1,\mathrm{},N}`$. This topology combined with natural semi-norms on kernels of compact operators, induces a Frรฉchet structure on the space of classical pseudo-differential operators $`PDO(M,E)`$ via the identification of operators with their symbol up to a smoothing operator (see \[KV1\] section 3) $``$From elliptic operators to their principal symbols Let $`S_PEll_0^{}(M,E)`$ denote the group of principal symbols of operators in $`Ell_0^{}(M,E)`$ and let $`Ell_0^{,0}(M,E)`$, resp. $`S_PEll_0^{,0}(M,E)`$ be the pathwise connected component of identity of these topological spaces. Lemma
$$\mathrm{\Pi }_1(S_PEll_0^{,0}(M,E))=\mathrm{\Pi }_1(Ell_0^{,0}(M,E)).$$
Proof: We show that $`Ell_0^{,0}(M,E)`$ and $`S_PEll_0^{,0}(M,E)`$ have same homotopy type. The result then follows. Let
$$\begin{array}{cc}\hfill f:Ell_0^{,0}(M,E)& S_PEll_0^{,0}(M,E)\hfill \\ \hfill A& \sigma _P(A).\hfill \end{array}$$
The connection on $`E`$ and the metric on $`M`$ give a canonical way of assigning to a symbol $`\sigma `$, an operator $`Op(\sigma (A))(u)(x)`$ with same principal symbol (see e.g. \[BML\] page 188 and references therein):
$$Op(\sigma (A))(u)(x):=_{T_x^{}M}e^{ix\xi }\sigma (\xi )\widehat{u}(\xi )๐\xi $$
where the Fourier transform is defined using the exponential map on $`M`$ and parallel transport on $`E`$. This gives rise to a map:
$$\begin{array}{cc}\hfill g:S_PEll_0^{,0}(M,E)& Ell_0^{,0}(M,E)\hfill \\ \hfill \sigma & Op\left(\sigma \right).\hfill \end{array}$$
Then $`fg=Id`$. One does not expect that $`gf=Id`$ because $`A=Op(\sigma _P(A))(1+K)`$ for some (uniquely defined) compact operator $`K`$. However we have $`gfId`$ where $``$ denotes homotopy of maps. Indeed, we can build a map $`(A,s)F(A,s)`$ on $`Ell_0^{}(M,E)\times [0,1]`$ such that $`F(A,0)=gf(A)=Op\left(\sigma _P(A)\right)`$ and $`F(A,1)=A`$ for any $`AEll_0^{}(M,E)`$. Set
$$F(A,s):=(1s)Op\left(\sigma _P(A)\right)+sA=Op\left(\sigma _P(A)\right)(1+sK),$$
then $`F(A,s)Ell_0(M,E)`$ since $`\sigma _P(F(A,s))=\sigma _P\left(Op\left(\sigma _P(A)\right)\right)=\sigma _P(A)`$ and $`A`$ is elliptic of order zero. The operator $`1+sK`$ might not be invertible so let us show how we can modify it continuously into a family $`\stackrel{~}{F}(A,s)Ell_0^{}(M,E)`$. Let $`\gamma (s):=Id+K_s`$. Let us denote by $`S_K`$ the spectrum of $`K`$. Since $`K`$ can be extended into a compact operator of $`L^2(M,E)`$, $`S_K`$ is compact and consists of isolated points, up to 0 which may be an accumulation point. The operator $`Id+sK`$ is not invertible whenever $`\frac{1}{s}`$ is an eigenvalue of $`K`$. Since $`s[0,1]`$, we have $`\frac{1}{s}1`$. The set $`]\mathrm{},1]S_K`$ is finite, so $`\gamma (s)`$ is invertible, up to a finite number of $`s[0,1]`$. We can therefore modify $`\gamma `$ continously into some continuous path $`\stackrel{~}{\gamma }(s):=1+\varphi (s)K`$ with values in the set $`\{Id+Kinvertible:KPDO^1(M,E)\}`$, where $`PDO^\alpha (M,E)`$ denotes the subspace of pseudodifferential opertors with order $`\alpha `$. Setting $`\stackrel{~}{F}(A,s):=Op\left(\sigma _P(A)\right)\left(1+\varphi (s)K\right)`$ yields the result. $``$ The fundamental group $`\mathrm{\Pi }_1(Ell_0^{}(M,E))`$ We shall discuss the case when $`E`$ is trivial, but the results extend to the non trivial case using the fact that $`M`$ being compact, for large enough $`N`$, there is a vector bundle $`F`$ based on $`M`$ such that $`EF1_N`$ where $`1_N`$ is the trivial bundle of rank $`N`$ on $`M`$. Let $`๐`$ be a complex unitary Banach algebra and let $`N`$ be a positive integer large enough. Then (\[Ka\], chIII, Th1.25) the fundamental group $`\mathrm{\Pi }_1(GL_N(๐))`$ is isomorphic to $`K(๐)`$, where $`GL_N(๐)`$ is the group of invertible maps of $`๐^N`$ and $`K(๐)`$ is the Grothendieck group of finitely generated projective $`๐`$-modules. Applying that result to the Banach space $`C(S^{}M)`$ of continuous functions on $`S^{}M`$, we find that:
$$\mathrm{\Pi }_1(GL_N(C(S^{}M)))K_0(S^{}M):=K(C(S^{}M)).$$
Using the classical result (see e.g. \[Di\], vol. 3) that on a compact manifold a continuous function is homotopic to a smooth function and that two homotopic smooth functions are smoothly homotopic, we have:
$$\mathrm{\Pi }_1(GL_N(C^{\mathrm{}}(S^{}M)))=\mathrm{\Pi }_1(GL_N(C(S^{}M))).$$
It follows that for the trivial bundle $`1_N`$ based on $`M`$ of rank $`N`$ large enough:
$$\mathrm{\Pi }_1\left(S_PEll_0^{,0}(M,1_N)\right)\mathrm{\Pi }_1(GL_N(C(S^{}M)))K_0(S^{}M).$$
On the other hand, $`\mathrm{\Pi }_1(GL_N(C(S^{}M)))`$ being generated by one parameter subgroups $`exp(2\pi itp)`$, $`p:\left(C(S^{}M)\right)^N\left(C(S^{}M)\right)^N`$, such that $`p^2=p`$ \[Ka\], is non trivial so that $`\mathrm{\Pi }_1(Ell_0^{}(M,1_N))`$ is non trivial as announced. Remark: In fact, since the arcwise connected component of identity $`Ell^{,0}(M,E)`$ of the group $`Ell^{}(M,E)`$ of invertible elliptic operators has the same fundamental group as $`Ell_0^{,0}(M,E)`$ consisting of those that have zero order, this shows that $`\mathrm{\Pi }_1(Ell^{,0}(M,E))=\mathrm{\Pi }_1(Ell_0^{,0}(M,E))K_0(S^{}M)`$ is generated by one parameter subgroups $`exp(2\pi itP)`$ of the above type and hence non trivial.
REFERENCES
\[AP\] M.Arnaudon, S.Paycha, Regularisable and minimal orbits for group actions in infinite dimensions, Comm.Math.Phys. 191 , 641โ662 (1998)
\[BF\] J.M.Bismut, D.Freed,The Analysis of elliptic families I, Comm.Math.Phys. 106, 159โ176 (1986)
\[BGV\] N. Berline, E. Getzler, M. Vergne, Heat-kernels and Dirac operators, Springer Verlag 1991
\[CFNW\] M.Cederwall, G.Ferretti, B. Nilsson, A.Westerberg, Schwinger terms and Cohomology of pseudodifferential operators, Comm.Math.Phys. 175 , 203โ220 (1996)
\[CE\] H. Cartan, S. Eilenberg, Homological algebra, Princeton University Press (1956)
\[C1\] A. Connes,Non commutative differential geometry, Publications I.H.E.S , n.62 (1985) p. 257-360
\[C2\] A. Connes,Non commutative geometry, Academic Press, San Diego, 1994
\[D\] C.Ducourtioux,Ph.D thesis in preparation
\[Di\] J. Dieudonnรฉ, Elรฉments dโanalyse 3, Gauthier-Villars (1970)
\[DL\] B. Driver, T. Lohrenz, Log Sobolev Inequalities on loop groups , Journal of Functional Analysis 140, 381-438 (1996)
\[FLPTT\] N. Fagella, A. Lesne, S. Paycha, L. Tedeschini-Lalli, S.T. Tsou, Renormalization, Proceedings of a workshop, Paris 1997
\[F1\] D. Freed, The geometry of loop groups, Journ. Diff. Geom. 28, 223โ276 (1988)
\[F2\] D. Freed, Flag manifolds and infinite dimensional Kรคhler geometry, in โInfinite dimensional groups with applications, Ed. V. Kac, Math. Sci. Res. Inst. Publ., Vol 4, Springer, New York, 1985
\[G\] P. Gilkey, Invariance theory. The heat equation and the Atiyah-Singer index theorem, Publish or Perish (1984), second edition, CRC Press (1995)
\[H\] L.Hรถrmander, The analysis of linear partial differential operators, I. Grundlagen der mathematischen Wiss. 256, Springer Verlag (1983)
\[Ka\] M.Karoubi, $`K`$-theory (An introduction), Springer Verlag 1978
\[K\] Ch. Kassel, Le rรฉsidu non commutatif \[dโaprรจs Wodzicki\], Sรฉminaire Bourbaki 708 (1989)
\[Ki\] A. Kirillov, Elements of theory of representation, Berlin, Springer Verlag, 1976
\[KOS\] H.H Kuo, N. Obata, K. Saito, Lรฉvy Laplacians of generalized functions on a nuclear space,
\[KK\] O.S. Kravenchko, B.A. Khesin, A central extension of the algebra of pseudodifferential symbols, Funct.Anal.Appl. 25, 152โ154 (1991)
\[KV1\] M.Kontsevich, S. Vishik, Determinants of elliptic pseudodifferential operators, Max Planck Preprint (1994)
\[KV2\] M.Kontsevich, S. Vishik, Geometry of determinants of elliptic operators in Functional Analysis on the Eve of the 21st Century Vol. I (ed. S.Gindikin, J.Lepowski,R.L.Wilson) Progress in Mathematics (1994)
\[Ku\] N.H.Kuiper, The homotopy type of the unitary group of Hilbert spaces, Topology 3 (1965)
\[L\] S. Lang, An introduction to differentiable manifolds, Interscience, John Wiley (1967)
\[LM\] B.Lawson, M.L Michelsohn, Spin Geometry, Princeton Mathematical Series, 1989
\[Le\] M.Lesch, On the non commutative residue for pseudo-differential operators with log-polyhomogeneous symbols, Annals of global analysis and geometry 17 (1998)
\[MRT\] Y. Maeda, S. Rosenberg, P.TondeurThe curvature of gauge orbits in โ Global Analysis and Modern Mathematicsโ edited by K. Uhlenbeck, Publich or Perish (1994)
\[M\] J. Marsden, Applications of Global Analysis in Mathematical Physics, Publish or Perish, 1974
\[M1\] J.Mickelsson, Current algebras and groups,New York, Plenum Press, 1989
\[M2\] J.Mickelsson, Second Quantization, anomalies and group extensions, Lecture notes given at the โColloque sur les Mรฉthodes Gรฉomรฉtriques en physique, C.I.R.M, Luminy, June 1997
\[MN\] R. Melrose and V. Nestor, Homology of pseudo-differential operators I. manifolds with boundary, Preprint: funct-an/9606005, Oct. 98
\[Pa\] S. Paycha, Renormalized traces as a looking glass into infinite dimensional geometry; a proposal to extend concepts of Riemannian geometry to a class of Hilbert manifolds, Preprint 1999
\[Pr\] A. Pressley, The Energy flow on, the loop space of a compact Lie group, Journ. of the LOndon Math. Soc. Vol 26, 1982
\[PS\] A. Pressley, G. Segal, Loop groups, Oxford Univ.Press (1988)
\[R\] A.O.Radul, Lie algebras of differential operators, their central extensions, and W-algebras, Funct.Anal.Appl. 25, 25โ39(1991)
\[Rog\] C. Roger, extensions centrales dโalgebres et de groupes de Lie de dimension infinie, algebres de Virasoro et gรฉnรฉralisations, Prรฉpublications de lโinstitut Girard Desargues (1995)
\[S\] J. Schwinger, Field theory commutators, Phys. Rev. Lett. 3, 296-297 (1959)
\[Se\] R.T. Seeley, Complex powers of an elliptic operator, Singular integrals, Proc. Symp. Pure Math., Chicago, Amer.Math. Soc., Providence, 288-307 (1966)
\[Sh\] M.A. Shubin, pseudodifferential Operators and Spectral Theory, Springer Verlag 1987
\[SW\] M. Spera, T. Wurzbacher, Differential geometry of Grassmannian embedding of based loop groups, Preprint 1998
\[W\] M. Wodzicki, Non commutative residue in Lecture Notes in Mathematics 1289 Springer Verlag (1987)
\[Wu\] T. Wurzbacher, Symplectic geometry of the loop space of a Riemannian manifold, Journal of Geometry and Physics 16, 345-384 (1995) |
warning/0001/quant-ph0001059.html | ar5iv | text | # Geometric Phase, Curvature, and Extrapotentials in Constrained Quantum Systems
## I Introduction
We derive an effective Hamiltonian for a quantum system subject to an infinite restoring force. Though our results are quite general, we are motivated by several specific applications, such as stiff molecular bonds in rigid molecules and clusters of rigid molecules, molecular systems evolving along reaction paths, and electrons confined to quantum strip waveguides.
For comparison, consider first a classical system. We have in mind a system which initially occupies any position in the full configuration space (called the ambient space) but is subsequently confined to a submanifold (called the constraint manifold) by the introduction of a restoring force, which in a certain limit becomes infinite. Here, the Hamiltonian is simply the kinetic energy plus the constraining potential, which we assume is constant along the constraint manifold. Assuming the initial velocity is tangent to the constraint manifold, it is well known that the systemโs trajectory remains on the constraint manifold and that its motion is determined solely by the form of the kinetic energy tangent to the manifold . This kinetic energy, in turn, depends only on the Riemannian metric of the constraint manifold. Thus, the motion of the constrained classical system depends only on the internal metric of the constraint manifold and is independent of the ambient space, the embedding of the constraint manifold within the ambient space, or the details of the constraining potential. It is a remarkable fact, then, that for a quantum system this is no longer true. The constrained quantum Hamiltonian depends on the curvature of the ambient space, the external curvatures of the constraint manifold, and on properties of the constraining potential.
It has already been known for some time that a constrained quantum system โsensesโ the local neighborhood of the constraint manifold . As a simple example, consider a quantum system whose motion transverse to the constraint manifold is in the ground state. Due to the conservation of the transverse action, the constrained quantum system sees the transverse zero point energy as an effective potential on the constraint manifold. (We call this the adiabatic potential.) The adiabatic potential also appears classically if the initial velocity of the system has a nonzero component normal to the constraint manifold. Classically, the adiabatic potential can always be eliminated by choosing an initial velocity tangent to the constraint manifold. Quantum mechanically, however, due to the Heisenberg principle, the transverse action and hence the adiabatic potential can never be eliminated.
The present paper focuses on effects of the ambient space and constraining potential other than the adiabatic potential. Following da Costa, we assume that the constraining potential has the same form at each point of the constraint manifold. The adiabatic potential is therefore constant along the constraint manifold and can subsequently be ignored. (In Sect. VI C, we discuss briefly how a small amount of variation in the adiabatic potential can be accommodated.) In two noteworthy papers, da Costa, using this assumption, derived the effective Hamiltonian for a system of $`n`$ constrained point particles. This Hamiltonian contains two terms. The first is proportional to the Laplacian on the constraint manifold, and therefore depends only on the internal metric of the constraint manifold. The second, however, is an effective potential, called the extrapotential, which depends not only on the internal curvature, but also the external curvature of the constraint manifold. This extrapotential is of order $`\mathrm{}^2`$ and therefore vanishes in the classical (and semi-classical) limit. As a simple, yet illustrative, example, consider a system defined on $`^3`$ constrained to lie on a curve. For this system, the extrapotential is $`\mathrm{}^2/(8\rho ^2)`$, where $`\rho `$ is the radius of curvature. This result was obtained by da Costa ; the same result was obtained earlier by Marcus and Switkes et al. for curves in $`^2`$. Others have also studied this order $`\mathrm{}^2`$ extrapotential, including Jensen and Koppe and Kaplan, Maitra, and Heller . Since the extrapotential depends on the external curvature, it can never be derived from a procedure which quantizes the constrained classical system (which depends only on the internal curvature), an approach common in the literature of constrained quantum mechanics. (See, for example, the review of DeWitt .)
As mentioned above, once the constraining potential is defined at one point of the constraint manifold, the constraining potential at all other points must have the same form. However, this requirement does not completely determine the constraining potential at all points since the orientation of the potential is left unspecified. In other words, the equipotentials surrounding the constraint manifold can twist in some unspecified manner as one moves along the manifold.
Da Costa fixed the twisting ambiguity by imposing what we call a โno twistโ condition on the potential. Physically, this condition requires the restoring forces in the neighborhood of the constraint manifold to be normal to the manifold. It can be viewed as an extension of the fact that in classical mechanics nondissipative forces are normal to the constraint manifold at the point of the manifold itself. Da Costa astutely realized that if the no twist condition were violated, the motion transverse to the constraint manifold would be coupled to the motion tangent to the manifold and the Schrรถdinger equation would not separate.
For some submanifolds there exist no potentials which satisfy the no twist condition. In Ref. da Costa derived a local geometric criterion on the external curvature of a submanifold which was necessary and sufficient to determine the existence of a non-twisting potential. Unfortunately, several common examples of constrained systems do not satisfy this criterion. For example, consider a model of a polymer by $`n>2`$ point particles where the distances between each particle $`i`$ and its neighbor $`i+1`$ are fixed. (This model also applies to the double pendulum.) These systems all fail the criterion as does a system of $`n>2`$ point particles constrained to form a rigid body. Even if a given submanifold can have a non-twisting constraining potential, there is no guarantee that this potential is the one dictated by the physics of the system under consideration.
The principal objective of this paper is to derive an effective Hamiltonian for a constrained quantum system with arbitrary twisting of the potential. The presence of the potential twist leads to several qualitative changes in the Hamiltonian. First, the Hamiltonian is no longer a scalar operator, but rather a $`k\times k`$ matrix of operators acting on a $`k`$-dimensional vector-valued wave function defined over the constraint manifold. Here, $`k`$ is the degeneracy of the transverse modes with each component of the vector wave function representing a different transverse mode. Of course, if one chooses a nondegenerate transverse mode, the Hamiltonian reduces to a single component.
Perhaps the most interesting consequence of dropping the no twist condition is the emergence of a $`U(k)`$ gauge potential, or connection, in the constrained Hamiltonian. This gauge potential is a coupling between the twisting of the potential and the generalized angular momentum of the transverse modes. Modes with no such angular momentum are unaffected by the potential twist. This gauge potential is an example of geometric phase and is closely related to the phase originally introduced by Berry in the context of adiabatic transport of quantum states . It is interesting to note that the gauge potential is of order $`\mathrm{}^0`$ and therefore, like the adiabatic potential, is essentially of classical origin.
In addition to creating the gauge potential, the potential twist adds a term to the extrapotential. Unlike the extrapotential terms derived by da Costa, the potential twist term is not a scalar function, but a $`k\times k`$ matrix of such functions with possible off-diagonal terms coupling the degenerate transverse modes. The potential twist term depends on the standard deviation of the angular momentum of the transverse modes and thus disappears for angular momentum eigenstates.
In some applications, the ambient space may not be flat. For example, the internal space of a molecule with $`n>2`$ atoms is not flat . Constraining such a molecule to a reaction path therefore requires an analysis accounting for the ambient curvature. We therefore do not assume in this paper that the ambient space is flat. The effects of the ambient curvature are most notable as additional terms in the extrapotential, although it also modifies the curvature of the gauge potential.
This paper has the following organization. In Sect. II, we introduce many of the key concepts by a simple example: that of a system on $`^3`$ constrained to a curve. Section II is purely expository, containing no derivations. Section III briefly introduces the general problem. In Sect. IV we focus on the constraining potential. We take care to precisely define the requirement that it have the same form at all points of the constraint manifold. We also define a tensor which measures the twisting of the potential. In Sect. IV C we specify how the potential is to scale in $`ฯต`$, where $`ฯต0`$ represents an infinite constraining force. The main computations of the paper are in Sect. V in which we expand the kinetic energy in $`ฯต`$ and arrive at a preliminary expression, Eq. (42), for the constrained kinetic energy. Section V E is devoted to deriving various expressions for the extrapotential. In Sect. VI, we apply first order perturbation theory to transform to a set of degenerate transverse modes, thereby obtaining Eqs. (106) โ (109), which are the main results of the paper. Section VI C briefly discusses nonconstant constraining potentials. In Sect. VII, we study the geometric origins of the gauge potential and various related connections. We also compute their curvatures. Section VIII contains some special cases, including constraint manifolds of codimensions one and two, rotationally invariant constraining potentials, and harmonic constraining potentials. In Sect. VIII E, we show that the gauge potential vanishes for certain systems with reflection symmetry. Conclusions are in Sect. IX. There are three Appendices. Appendix A contains a very brief review of curves in $`^3`$. Appendix B is a review of the second fundamental form. Appendix C summarizes an expression we will need for the quantum kinetic energy.
## II A Simple Example: A Curve in $`^3`$
The ultimate objective of this paper is to constrain a quantum wave function, defined on an arbitrary manifold (the ambient space), to a (locally) arbitrary submanifold (the constraint manifold) via some general constraining potential. Before solving the full problem, however, it is instructive to consider a simple (though certainly non-trivial), concrete example of the constraining procedure. We present no derivations here; our results will be justified later in Sect. VIII B.
We consider a curve embedded in flat three-dimensional space $`^3`$ and parameterized by its arclength $`\alpha `$. (See Fig. 1.) Such a curve is characterized by its curvature $`\kappa `$ and torsion $`\tau `$. (See Appendix A.) We take this curve to be the axis of a quantum waveguide. That is, there is a tube enclosing the curve such that the potential is zero inside the tube and infinite outside. We assume the cross-section of the tube is constant along the curve. More precisely, if we cut the tube along a plane normal to the curve (called hereafter a normal plane), the cross-sectional shape of the tube is independent of where along the curve we cut. Two such cross-sections have the same shape if one can be rotated into the other. This rotational freedom permits the cross-sectional shape to twist as one moves along the curve, even if the curve itself is straight. The orientation of the cross-sectional shape is specified by two orthonormal vectors $`๐_1`$ and $`๐_2`$ chosen from each normal plane along the curve. The choice of orthonormal frame $`(๐_1,๐_2)`$ is such that the cross-sectional shape (with respect to this frame) is independent of $`\alpha `$. In Fig. 1, the cross-section is a triangle with no reflection symmetry. Such symmetry is nongeneric and can cause certain terms discussed below to vanish. (See Sect. VIII E.)
We assume that the transverse dimensions of the tube are small compared to the radius of curvature $`\rho =\kappa ^1`$ and the inverse torsion $`\tau ^1`$. We can then separate out the โfastโ transverse degrees of freedom and obtain an effective one-dimensional Hamiltonian in the โslowโ longitudinal, or tangential, coordinate $`\alpha `$. To accomplish this separation, we pick a transverse mode $`\chi (u^1,u^2)`$ of the waveguide. Here $`(u^1,u^2)`$ are the Cartesian coordinates in the normal plane taken with respect to the frame $`(๐_1,๐_2)`$; the quantities $`(u^1,u^2,\alpha )`$ thus coordinatize the tube. The transverse mode $`\chi (u^1,u^2)`$ is a normalized eigenfunction of the transverse Hamiltonian $`H_{}=(\pi _1^2+\pi _2^2)/2+V_{}(u^1,u^2)`$, where $`\pi _j=i\mathrm{}/u^j`$, $`j=1,2`$, and $`V_{}(u^1,u^2)`$ is the potential energy which defines the tube. The eigenvalue of $`H_{}`$ corresponding to $`\chi `$ is called the transverse energy. For simplicity, we assume that the transverse energy is nondegenerate.
To lowest order in the width of the tube, an eigenfunction $`\psi `$ of the wave guide has the form
$$\psi (u^1,u^2,\alpha )=\chi (u^1,u^2)\varphi (\alpha ).$$
(1)
As we take the limit where the transverse dimensions of the waveguide shrink to zero (keeping the quantum numbers of the transverse mode fixed), the transverse energy obviously tends toward infinity. However, due to the constancy of the cross-sectional shape, this transverse energy, though very large, is itself constant along the curve. We thus subtract it off, leaving a residual Hamiltonian $`H_{}`$, which we call the constrained Hamiltonian. The constrained Hamiltonian acts only on $`\varphi `$, resulting in the Schrรถdinger equation
$$H_{}\varphi =E_{}\varphi .$$
(2)
The principal objective of this paper is to determine the form of this constrained Hamiltonian.
As we will show later, the constrained Hamiltonian is not simply $`\pi _{}^2/2`$ where $`\pi _{}=i\mathrm{}/\alpha `$. Rather, there are effects from the curvature $`\kappa `$ and the rate at which the cross-sectional shape twists along the curve. To make this latter concept more precise, we define the potential twist $`S=๐_1(d๐_2/d\alpha )`$ which measures the rotation rate of the cross-sectional shape. The potential twist admits the following description. Let $`\theta `$ denote the angle between the principal normal $`\widehat{๐ง}`$ (see Appendix A) and the frame $`(๐_1,๐_2)`$, specifically, $`\widehat{๐ง}๐_1=\mathrm{cos}\theta `$, $`\widehat{๐ง}๐_2=\mathrm{sin}\theta `$. Let $`\omega =d\theta /d\alpha `$ denote the rotation rate of the frame $`(๐_1,๐_2)`$ with respect to $`\widehat{๐ง}`$. Then $`S`$ is related to $`\omega `$ and the torsion $`\tau `$ by $`S=\tau +\omega `$. Taking $`S=0`$, we obtain the case considered by da Costa in Ref. . We next define an angular momentum operator $`\mathrm{\Lambda }`$ in the tangential direction by $`\mathrm{\Lambda }=(u^1\pi _2u^2\pi _1)/2`$. The constrained Hamiltonian is then
$$H_{}=K_{}+V_{ex},$$
(3)
where
$`K_{}`$ $`=`$ $`{\displaystyle \frac{1}{2}}(\pi _{}+2S\mathrm{\Lambda })^2,`$ (4)
$`V_{ex}`$ $`=`$ $`{\displaystyle \frac{\mathrm{}^2}{8}}\kappa ^2+2S^2(\mathrm{\Lambda }^2\mathrm{\Lambda }^2),`$ (5)
and where the bracket notation $``$ denotes the expectation value with respect to the transverse mode $`\chi `$.
Observe that the tangential kinetic energy $`K_{}`$ departs from $`\pi _{}^2/2`$ due to the inclusion of the term $`2S\mathrm{\Lambda }`$, which couples the angular momentum of the transverse mode to the rate of potential twist. This term is a gauge coupling, a fact we explore further in Sect. VII. For now, we simply note that because the curve is one-dimensional, the gauge coupling can be removed from Eq. (4) by a gauge transformation. In the present context, a gauge transformation consists of changing the phase of the wave function $`\varphi `$. This transformation is not without its consequences, however, as it will obviously change the boundary conditions which $`\varphi `$ must satisfy. Also, we stress that if the constraint manifold has dimension greater than one, it will not in general be possible to remove the gauge coupling by a gauge transformation.
As a final observation on $`K_{}`$, notice that the gauge coupling is of order $`\mathrm{}^0`$, which indicates that it is essentially a classical quantity. This coupling should therefore appear in a classical theory of constraints which takes into account the potential twist.
Turning to the quantity $`V_{ex}`$, we note that it is a real-valued function of $`\alpha `$, containing no differential operators. For this reason, we call $`V_{ex}`$ an extrapotential. The extrapotential contains two terms, $`\mathrm{}^2\kappa ^2/8`$ and $`2S^2(\mathrm{\Lambda }^2\mathrm{\Lambda }^2)`$. The first of these was derived by da Costa for the case $`S=0`$ . It has the physical effect of attracting $`\varphi `$ to regions of high curvature, a fact which may produce curvature-induced bound states in the waveguide. Such bound states are of current interest and are reviewed by Duclos and Exner . The term $`\mathrm{}^2\kappa ^2/8`$ is of order $`\mathrm{}^2`$ and therefore disappears in the classical (and semi-classical) limit. The second term of $`V_{ex}`$, like the gauge coupling in $`K_{}`$, depends on both the potential twist $`S`$ and the angular momentum $`\mathrm{\Lambda }`$. Notice, however, that it is the standard deviation of the angular momentum which appears in $`V_{ex}`$. This means, for example, that the second term of $`V_{ex}`$ vanishes for transverse modes which are angular momentum eigenstates. It is interesting to observe that, in contrast to the first term, the second term of $`V_{ex}`$ has the physical effect of expelling the wave function $`\varphi `$ from regions of high twist $`S`$. Also, the second term is of order $`\mathrm{}^0`$, which means that, like the gauge coupling in $`K_{}`$, it survives the classical limit.
## III Introduction to the General Problem
We describe here how the setup in Sect. II is modified for the general problem. First, the ambient space in Sect. II was assumed to be $`^3`$. In the general problem, we allow the ambient space to be an arbitrary Riemannian manifold, which we denote by $`๐`$. The kinetic energy of the wave function $`\psi `$, defined over $`๐`$, is given by $`K=\mathrm{}^2\mathrm{}/2`$, where $`\mathrm{}`$ is the Laplacian on $`๐`$. Unlike Sect. II, the ambient space is not assumed to be flat, and, as we will discover, the curvature of the ambient space produces additional terms in $`V_{ex}`$.
Next, we constrain the wave function to lie in the vicinity of a (locally) arbitrary (embedded) submanifold $`๐`$ of $`๐`$ with dimension $`m`$ and codimension $`d`$. We call $`๐`$ the constraint manifold. In Sect. II, the constraint manifold was a one-dimensional curve. The curvature and torsion of this curve played a critical role in the analysis. The appropriate generalization of the curvature and torsion is the second fundamental form $`T`$, which is a rank three tensor. (See Appendix B.)
In Sect. II, the constraint was imposed by a hard-wall potential that was infinite outside of a tube and zero inside. We then took the limit in which the width of the tube went to $`0`$. In the general problem, we impose the constraint by an arbitrary potential $`V_{}`$, subject to a few reasonable conditions. This potential is defined on a set of coordinates transverse to the constraint manifold and, for this reason, is called the transverse (or constraining) potential. The transverse potential depends on a scaling parameter $`ฯต`$ which is analogous to the width parameter of the tube; the constraint is imposed by taking the limit $`ฯต`$ goes to $`0`$.
One of the conditions we do still require of $`V_{}`$ is that it be independent of the location on the constraint manifold. This condition, as well as a few other minor conditions, are explained fully in the next section.
## IV The Transverse Potential
### A Constancy of the Transverse Potential
In Sect. II, we defined the constraining potential by first specifying the form of the potential on a plane normal to the curve and then specifying the orientation of this potential at all points along the curve. For the general case, we use the same fundamental idea except that now, due to the curvature of the ambient space, we must take care to define how we generalize the concept of the normal plane.
It is useful to consider two separate but related spaces for a given point $`q`$ on the constraint manifold. The first is the linear space of all vectors normal to the constraint manifold. We call this the normal space at $`q`$ and denote it by $`N_q`$. The second space of interest is the submanifold of the ambient space formed by geodesics emanating from $`q`$ normal to the constraint manifold. We call this the transverse space at $`q`$ and denote it by $`๐ฐ_q`$. The spaces $`N_q`$ and $`๐ฐ_q`$ are related by the exponential map which takes a vector $`๐ฏN_q`$ into the point $`\mathrm{exp}๐ฏ๐`$. The point $`\mathrm{exp}๐ฏ๐`$ lies on the geodesic emanating from $`q`$ in the direction of $`๐ฏ`$; it lies at a distance $`|๐ฏ|=(๐ฏ,๐ฏ)^{1/2}`$ from $`q`$ along this geodesic. (We use the notation $`,`$ for the metric on $`๐`$.) Thus, we find $`๐ฐ_q=\mathrm{exp}N_q`$. We now modify our original definition of $`๐ฐ_q`$. If the geodesics emanating from the constraint manifold $`๐`$ in the neighborhood of $`q`$ flow to an arbitrary length, they will in general intersect one another. This can be witnessed even in the simple example of Sect. II. Thus, in defining $`๐ฐ_q`$, we assume that the geodesics flow for a small enough length to avoid such intersections and that this maximal length is independent of the point $`q`$ on the constraint manifold. In summary, then, we foliate a neighborhood (which we call the tubular neighborhood) of the constraint manifold $`๐`$ by the transverse spaces $`๐ฐ_q`$, which we have in turn related to the normal spaces $`N_q`$ by the exponential map. Using the exponential map to construct tubular neighborhoods in this fashion is a standard technique. For details, see, for example, Lang and Vanhecke .
Since we have identified normal vectors with points in the neighborhood of the constraint manifold, we view the transverse potential $`V_{}`$ as a function defined on the normal spaces. With this interpretation, we will require that $`V_{}`$, as a function of $`q`$ and the vectors in $`N_q`$, be independent of $`q`$. By independent, we really mean independent modulo $`SO(d)`$ rotations in $`N_q`$. Let us make this more precise. As in Sect. II, we specify the orientation of the transverse potential by an orthonormal basis $`๐_\mu `$, $`\mu =1,\mathrm{},d`$ of the normal space $`N_q`$. This basis forms a normal frame for the constraint manifold which we call the potential frame. For a given normal vector field $`๐ฎ`$, we introduce the components $`u^\mu `$, $`\mu =1,\mathrm{},d`$ with respect to $`๐_\mu `$. The quantities $`u^\mu `$ coordinatize both the normal space $`N_q`$ and the transverse space $`๐ฐ_q`$, for which they are commonly called Riemannian normal coordinates . We use sans serif for the list of coordinates $`๐=(u^1,\mathrm{},u^d)`$, reserving the bold notation $`๐ฎ`$ for the vector field. The neighborhood of $`๐`$ is therefore conveniently parameterized by $`(๐,q)`$. The heuristic constraint that $`V_{}`$ be independent of position on the constraint manifold can now be made precise by the following statement: the transverse potential $`V_{}(๐,q)`$ as a function of $`(๐,q)`$ is required to be independent of $`q`$.
In general, the construction of the parameters $`u^\mu `$ presented here is only possible locally on $`๐`$. That is, it may be impossible to define $`u^\mu `$ in the neighborhood of the whole constraint manifold simultaneously. The construction can break down in two ways. First, it may be impossible to construct a tubular neighborhood for the entire constraint manifold. One can see this even with the simple example of Sect. II. If the one-dimensional curve spirals in on itself, then the width of the tubular neighborhood is forced to go to $`0`$. (Recall that the width of the tubular neighborhood must be the same for all point on the constraint manifold.) Assuming that a tubular neighborhood does indeed exist for the manifold, there is still a second way in which the construction may break down. This occurs if there does not exist a potential frame $`๐_\mu `$ which is globally defined. (This happens when the normal bundle is nontrivial.) For example, let the ambient space be a Mรถbius strip and let the constraint manifold be a curve which wraps around the Mรถbius strip once. Clearly, there does not exist a normal frame for $`๐`$ which is defined globally. It is our viewpoint that these two obstacles (in particular the first) are not common in physical problems. Even if one did encounter a problem in which the $`u^\mu `$ were not definable globally, since the results of this paper involve only the local form of the Hamiltonian, they would still apply to such problems.
### B The Potential Twist Tensor
In this section, we generalize the potential twist $`S`$, of Sect. II, to a rank three potential twist tensor (also denoted $`S`$) defined for any $`q๐`$. For an arbitrary vector $`๐T_q๐`$, $`S_๐`$ is a linear map on $`T_q๐`$. (Here, $`T_q๐`$ is the $`(d+m)`$-dimensional tangent space of $`๐`$ at $`q`$.) Let $`๐ฑT_q๐`$ be an arbitrary vector tangent to $`๐`$. Then, we define
$$S_๐๐ฑ=0.$$
(6)
Now let $`๐ฏT_q๐`$ be an arbitrary vector normal to $`๐`$. We extend $`๐ฏ`$ to a vector field on $`๐`$ (defined in the neighborhood of $`q`$) by assuming that $`๐ฏ`$ is normal to $`๐`$ and furthermore that its components with respect to $`๐_\mu `$ are constant. We now complete the definition of $`S_๐`$ by prescribing
$$S_๐๐ฏ=P_{}_{P_{}๐}๐ฏ,$$
(7)
where $``$ is the Levi-Civita connection on $`๐`$ and $`P_{}`$ and $`P_{}`$ are the projection operators onto the normal and tangent spaces of $`๐`$ respectively. It is straightforward to verify that $`S`$ defined by Eqs. (6) and (7) is indeed a tensor.
Like the second fundamental form $`T`$ (see Appendix B), $`S`$ satisfies the antisymmetry property
$$๐,S_๐๐=๐,S_๐๐,$$
(8)
where $`๐,๐,๐T_q๐`$ are arbitrary. To prove the above equation, we need only consider the case $`๐=๐ฏN_q`$, $`๐=๐ฐN_q`$, and $`๐=๐ฑ`$ tangent to $`๐`$, since all other cases vanish. Since $`S`$ is a tensor, we may assume that $`๐ฏ`$ and $`๐ฐ`$ are vector fields and that their components with respect to $`๐_\mu `$ are constant. Since the frame $`๐_\mu `$ is orthonormal, $`๐ฏ,๐ฐ`$ is constant, and therefore Eq. (7) implies $`๐ฏ,S_๐ฑ๐ฐ=๐ฏ,_๐ฑ๐ฐ=_๐ฑ๐ฏ,๐ฐ=๐ฐ,S_๐ฑ๐ฏ`$.
### C Scaling of the Transverse Potential
In Sect. II, we imposed the constraint by taking the limit in which the width of the waveguide went to zero. Here, we describe a similar scaling procedure using, however, a more general transverse potential. Heuristically, we assume that $`V_{}(๐;ฯต)`$ depends on the scaling parameter $`ฯต`$ in such a way that the potential grows narrower and deeper as $`ฯต`$ tends toward $`0`$. To make this precise, we rescale the transverse potential in the following way
$$\stackrel{~}{V}_{}(\stackrel{~}{๐};ฯต)=ฯต^2V_{}(ฯต\stackrel{~}{๐};ฯต),$$
(9)
where
$$u^\mu =ฯต\stackrel{~}{u}^\mu .$$
(10)
We assume that $`\stackrel{~}{u}^\mu `$ has no dependence itself on $`ฯต`$ and that $`\stackrel{~}{V}_{}(\stackrel{~}{๐};ฯต)`$ is smooth in $`ฯต`$ at $`ฯต=0`$, by which we mean that $`\stackrel{~}{V}_{}(\stackrel{~}{๐};ฯต)`$ can be expanded as $`\stackrel{~}{V}_{}(\stackrel{~}{๐};ฯต)=\stackrel{~}{V}_{}^0(\stackrel{~}{๐})+ฯต\stackrel{~}{V}_{}^1(\stackrel{~}{๐})+ฯต^2\stackrel{~}{V}_{}^2(\stackrel{~}{๐})+..`$. We also assume that $`\stackrel{~}{V}_{}^0(\stackrel{~}{๐})`$ does not vanish. In Sect. VI B, we will make some very natural, further assumptions on the existence of bound states for the transverse potential and on the smoothness in $`ฯต`$ of the corresponding eigenenergies.
As a concrete example take $`\stackrel{~}{V}_{}(\stackrel{~}{๐};ฯต)=\stackrel{~}{V}_{}(\stackrel{~}{๐})`$ to be a finite-depth square well with no $`ฯต`$ dependence. Then $`V_{}(๐;ฯต)`$ is a finite-depth square well whose width scales as $`ฯต`$ and whose depth scales as $`1/ฯต^2`$. Of course, these scaling rules apply to any potential satisfying the conditions described above. They guarantee that the expectation value of $`u^\mu `$ with respect to a transverse mode scales as $`ฯต`$ (assuming a fixed quantum number for the transverse mode). This fact shows that the wave function becomes more and more localized in the vicinity of the constraint manifold as $`ฯต`$ tends toward $`0`$.
## V Expansion of the Kinetic Energy
The derivation of the constrained Hamiltonian (such as Eqs. (3) โ (5)) proceeds in two steps. The first is to expand the kinetic energy in powers of $`ฯต`$. The second is to transform to a basis of transverse modes and to apply a first order perturbation treatment to the expanded Hamiltonian. This section is devoted to the first step.
### A Definition of the Vielbein
We will express the kinetic energy in terms of a vielbein $`๐_a`$, $`a=1,\mathrm{},d+m`$, on $`๐`$. Appendix C gives the necessary background for this technique. To span the transverse dimensions, we take $`๐_\mu =/u^\mu `$, $`\mu =1,\mathrm{},d`$, where it is understood that, for the purpose of the partial derivative, the position $`q๐`$ is held fixed. In selecting vector fields to span the remaining dimensions, we first choose an arbitrary set of orthonormal vector fields $`๐_i`$, $`i=d+1,\mathrm{},d+m`$, defined over $`๐`$ and tangent to $`๐`$. We then use $`๐_\mu `$ to Lie transport these vector fields into the tubular neighborhood of $`๐`$. That is, we require the Lie derivatives with respect to $`๐_\mu `$ to vanish,
$$[๐_\mu ,๐_i]=0.$$
(11)
We use the following notational scheme in this paper. The indices $`a,b,c,\mathrm{}`$ range from $`1,\mathrm{},d+m`$ and label the basis vectors $`๐_a`$ and any components with respect to this basis. The indices $`\mu ,\nu ,\sigma ,\mathrm{}`$ range from $`1,\mathrm{},d`$ and label the vector fields $`๐_\mu =/u^\mu `$ and their related components. The indices $`i,j,k,\mathrm{}`$ range from $`d+1,\mathrm{},d+m`$ and label the vector fields $`๐_i`$ and their related components. Except where otherwise noted, we employ the convention that an index $`a,b,c,\mathrm{}`$, $`\mu ,\nu ,\sigma ,\mathrm{}`$, or $`i,j,k,\mathrm{}`$ is implicitly summed over when occurring twice in the same expression.
For future reference, we present some facts regarding the structure constants $`\beta _{ab}^c`$, defined by $`[๐_a,๐_b]=\beta _{ab}^c๐_c`$. First, Eq. (11) immediately yields
$$\beta _{\mu j}^c=\beta _{j\mu }^c=0.$$
(12)
Furthermore, since $`๐_\mu =/u^\mu `$ is a coordinate basis on the transverse spaces $`๐ฐ_q`$, we find $`\beta _{\mu \nu }^c=0`$. Combining this with Eq. (12), we have
$$\beta _{\mu b}^c=\beta _{b\mu }^c=0.$$
(13)
Next we note that
$$0=[๐_\mu ,[๐_i,๐_j]]=[๐_\mu ,\beta _{ij}^c๐_c]=[/u^\mu \beta _{ij}^c]๐_c,$$
(14)
where the first equality follows from the Jacobi identity and Eq. (11) and the third equality follows from Eq. (13). We use the bracket notation $`[]`$ in the final equality to emphasize that the differential operator acts only on the quantities inside the bracket. Since the $`๐_c`$ form a basis, we have
$$\frac{\beta _{ij}^c}{u^\mu }=0.$$
(15)
Furthermore, since the $`๐_i`$ are tangent to $`๐`$ when restricted to $`๐`$, we have $`\beta _{ij}^\sigma =0`$ on $`๐`$. Combining this fact with Eq. (15), we find that
$$\beta _{ij}^\sigma =0$$
(16)
within the tubular neighborhood. Thus, the distribution of vector fields $`๐_i`$ is integrable everywhere. (The submanifolds thus defined by Frobeniusโs Theorem are manifolds of constant $`V_{}`$.)
### B Transformation of the Kinetic Energy
As in Appendix C, the momentum operators are defined to be $`\pi _a=i\mathrm{}๐_a`$. They are not in general Hermitian since the Hermitian conjugate is given by Eq. (C4). The kinetic energy is given by $`K=\pi _a^{}G^{ab}\pi _b/2`$, where $`G_{ab}=๐_a,๐_b`$ are the components of the metric tensor and $`G^{ab}`$ is the inverse of $`G_{ab}`$.
Appendix C also provides the framework for scaling the quantum wave function by an arbitrary (strictly) positive function $`s:๐`$ (see Eq. (C5)) in order to modify the form of the kinetic energy. We apply this scaling formalism here, taking
$$s=G^{1/4},$$
(17)
where $`G=detG_{ab}`$. As mentioned in Appendix C, this scaling defines a new inner product of wave functions. We observe that the original inner product of two wave functions $`\phi `$ and $`\phi ^{}`$ is given by
$$\phi |\phi ^{}=\sqrt{G}\nu \phi ^{}(๐,q)\phi ^{}(๐,q),$$
(18)
where $`\nu `$ is the $`(d+m)`$-form
$$\nu =E^1\mathrm{}E^{(d+m)}=du^1\mathrm{}du^dE^{(d+1)}\mathrm{}E^{(d+m)}.$$
(19)
Here, $`E^a`$ is the basis of one-forms dual to the vielbein $`๐_a`$. We have also used the fact that $`E^\mu =du^\mu `$. (Be careful not to confuse the notation $`|`$ with $`,`$ which denotes the Riemannian metric on $`๐`$.) From Eq. (C6) we therefore find that the scaled inner product of two (scaled) wave functions $`\psi `$ and $`\psi ^{}`$ is
$$\psi |\psi ^{}_s=\nu \psi ^{}(๐,q)\psi ^{}(๐,q).$$
(20)
This scaled inner product gives rise to a scaled Hermitian conjugate, denoted $`(s)`$. Using Eqs. (C4), (C8), and (13), we find that $`\pi _\mu `$ is Hermitian with respect to the scaled Hermitian conjugate,
$$\pi _\mu ^{(s)}=\pi _\mu .$$
(21)
Furthermore, Eqs. (C4), (C8), and (12) give the (scaled) Hermitian conjugate of $`\pi _j`$ as
$$\pi _j^{(s)}=\pi _j+i\mathrm{}\beta _{jb}^b=\pi _j+i\mathrm{}\beta _{jk}^k.$$
(22)
We now restrict the momentum operator $`\pi _j`$ to $`๐`$. We write $`\pi _j|_0`$ to make this explicit; we use the notation $`|_0`$ for any quantity restricted to $`๐`$ since this corresponds to $`u^\mu =0`$. For present purposes, we consider the constraint manifold in its own right without being viewed as embedded in the ambient space. With this interpretation, the vector field $`\pi _j|_0`$ has a well-defined Hermitian conjugate which we denote by $`(\pi _j|_0)^{}`$. This Hermitian conjugate is given by Eq. (C4), where it is understood that the symbol $`G`$ now refers only to the determinant of the metric $`G_{ij}`$ on $`๐`$. However, since the basis $`๐_i|_0`$ is orthonormal, we have $`G=1`$, and hence
$$(\pi _j|_0)^{}=\pi _j|_0+i\mathrm{}\gamma _{jk}^k,$$
(23)
where the functions $`\gamma _{ij}^k`$, defined on $`๐`$, are the structure constants for $`๐_i|_0`$. Since $`[๐_i|_0,๐_j|_0]=[๐_i,๐_j]|_0`$, the structure constants $`\gamma _{ij}^k`$ are equal to $`\beta _{ij}^k|_0`$. Comparing Eq. (23) to Eq. (22), we now have the following convenient description for $`\pi _j^{(s)}`$ when restricted to $`๐`$
$$\left(\pi _j^{(s)}\right)|_0=(\pi _j|_0)^{}.$$
(24)
The scaled kinetic energy is given by Eq. (C9). Noting Eq. (21), we rewrite this as
$$K_s=\frac{1}{2}\left(\pi _\mu G^{\mu \nu }\pi _\nu +\pi _\mu G^{\mu j}\pi _j+\pi _i^{(s)}G^{i\nu }\pi _\nu +\pi _i^{(s)}G^{ij}\pi _j\right)+V_s$$
(25)
where,
$$V_s=\frac{1}{8}\left(\frac{1}{4}G^{ab}[\pi _a\mathrm{ln}G][\pi _b\mathrm{ln}G]+[\pi _a^{(s)}G^{ab}[\pi _b\mathrm{ln}G]]\right).$$
(26)
We will henceforth drop the $`s`$ index on $`K_s`$, $`(s)`$, and $`|_s`$, with the scaling being implicitly understood.
### C Expansion of the Kinetic Energy
In this section, we expand the kinetic energy Eq. (25) through order $`ฯต^0`$. Recall from Eq. (10) that $`u^\mu `$ is of order $`ฯต^1`$, and hence the momentum $`\pi _\mu =i\mathrm{}/u^\mu `$ is of order $`ฯต^1`$. From Eq. (11), we see that the momentum $`\pi _i`$ is of order $`ฯต^0`$. Furthermore, from Eq. (15), we see that $`\beta _{ij}^k`$ is of order $`ฯต^0`$, and combining this fact with Eq. (22) we find that $`\pi _i^{}`$ is of order $`ฯต^0`$. (Recall that the index โ$`(s)`$โ is now implicit.) These scaling properties imply that to expand Eq. (25) to order $`ฯต^0`$, we must expand $`V_s`$, $`G^{ij}`$, $`G^{i\mu }`$, and $`G^{\mu \nu }`$ to orders $`ฯต^0`$, $`ฯต^0`$, $`ฯต^1`$, and $`ฯต^2`$ respectively.
Since $`๐_a`$ is an orthonormal frame at $`u^\mu =0`$, we have the following identities
$`G_{ab}|_0`$ $`=`$ $`G^{ab}|_0=\delta _{ab},`$ (27)
$`G_{,\sigma }^{ab}|_0`$ $`=`$ $`G_{ab,\sigma }|_0,`$ (28)
where we use the notation โ$`,\sigma `$โ for the derivative $`/u^\sigma `$. Equations (27) and (28) yield the following expansions of $`G^{ab}`$,
$`G^{ij}(๐)`$ $`=`$ $`\delta _{ij}+O(ฯต),`$ (29)
$`G^{\mu j}(๐)`$ $`=`$ $`G_{,\sigma }^{\mu j}|_0u^\sigma +O(ฯต^2)=G_{\mu j,\sigma }|_0u^\sigma +O(ฯต^2),`$ (30)
$`G^{\mu \nu }(๐)`$ $`=`$ $`\delta _{\mu \nu }+G_{,\sigma }^{\mu \nu }|_0u^\sigma +{\displaystyle \frac{1}{2}}G_{,\sigma \tau }^{\mu \nu }|_0u^\sigma u^\tau +O(ฯต^3)`$ (31)
$`=`$ $`\delta _{\mu \nu }G_{\mu \nu ,\sigma }|_0u^\sigma +{\displaystyle \frac{1}{2}}(G_{\mu \nu ,\sigma \tau }+2G_{\mu a,\sigma }G_{a\nu ,\tau })|_0u^\sigma u^\tau +O(ฯต^3).`$ (32)
The derivatives of the metric $`G_{ab}`$ appearing above are conveniently expressed in terms of the potential twist tensor $`S`$ and the Riemannian curvature $`R`$ on $`๐`$. To see this, we first introduce the components of $`S`$, $`T`$, and $`R`$ via
$`S_{abc}`$ $`=`$ $`๐_a,S_{๐_c}๐_b,`$ (33)
$`T_{abc}`$ $`=`$ $`๐_a,T_{๐_c}๐_b,`$ (34)
$`R_{abcd}`$ $`=`$ $`๐_a,R_{๐_c๐_d}๐_b,`$ (35)
where $`T`$ is included for completeness and for future reference. We then have the following identities
$`G_{\mu j,\sigma }|_0`$ $`=`$ $`S_{\mu \sigma j}|_0,`$ (36)
$`G_{\mu \nu ,\sigma }|_0`$ $`=`$ $`0,`$ (37)
$`G_{\mu \nu ,\sigma \tau }|_0`$ $`=`$ $`{\displaystyle \frac{1}{3}}\left(R_{\mu \sigma \nu \tau }+R_{\nu \sigma \mu \tau }\right)|_0.`$ (38)
In actuality, the $`|_0`$ notation on $`S_{\mu \sigma j}`$ is redundant since $`S`$ is only defined on $`๐`$, but we will make use of this notation as a convenient reminder. We will derive Eqs. (36) โ (38) momentarily, but for now we insert them into Eqs. (29) โ (32) to obtain
$`G^{ij}(๐)`$ $`=`$ $`\delta _{ij}+O(ฯต),`$ (39)
$`G^{\mu j}(๐)`$ $`=`$ $`S_{\mu \sigma j}|_0u^\sigma +O(ฯต^2),`$ (40)
$`G^{\mu \nu }(๐)`$ $`=`$ $`\delta _{\mu \nu }+\left({\displaystyle \frac{1}{3}}R_{\mu \sigma \nu \tau }+S_{\mu \sigma k}S_{\nu \tau k}\right)|_0u^\sigma u^\tau +O(ฯต^3),`$ (41)
where we have used the well-known symmetry of the Riemannian curvature $`R_{abcd}=R_{cdab}`$. We next insert Eqs. (39) โ (41) into Eq. (25) to arrive at the main result of this section,
$$K=K_{}+K_{}^p+K_R+V_{ex}^p+O(ฯต),$$
(42)
where
$`K_{}`$ $`=`$ $`{\displaystyle \frac{1}{2}}\pi ^\mu |_0\pi _\mu ,`$ (43)
$`K_{}^p`$ $`=`$ $`{\displaystyle \frac{1}{2}}(\pi ^i|_0+S^{\mu \nu i}|_0\mathrm{\Lambda }_{\mu \nu })^{}(\pi _i+S_i^{\sigma \tau }|_0\mathrm{\Lambda }_{\sigma \tau }),`$ (44)
$`K_R`$ $`=`$ $`{\displaystyle \frac{1}{6}}R^{\mu \nu \sigma \tau }|_0\mathrm{\Lambda }_{\mu \nu }\mathrm{\Lambda }_{\sigma \tau },`$ (45)
$`V_{ex}^p`$ $`=`$ $`V_s|_0.`$ (46)
We have taken advantage of the antisymmetry property Eq. (8), $`S_{\mu \nu i}=S_{\nu \mu i}`$, and the well-known antisymmetry relations $`R_{\mu \nu \sigma \tau }=R_{\nu \mu \sigma \tau }=R_{\mu \nu \tau \sigma }`$ to introduce the operators
$$\mathrm{\Lambda }_{\mu \nu }=\frac{1}{2}(u_\mu \pi _\nu u_\nu \pi _\mu )=\frac{1}{2}(\pi _\nu u_\mu \pi _\mu u_\nu ),$$
(47)
which are generalized angular momentum operators acting on the transverse space. That is, they generate $`SO(d)`$ rotations in the transverse space. They are the generalization of the angular momentum $`\mathrm{\Lambda }`$ defined in Sect. II.
In Eqs. (43) โ (45), we have employed the standard practice of raising tensor indices by contraction with $`G^{ab}`$. Thus, $`\pi ^\mu =G^{\mu a}\pi _a`$, $`S^{\mu \nu i}=G^{\mu a}G^{\nu b}G^{ic}S_{abc}`$, etc. However, since $`G^{ab}|_0=\delta _{ab}`$, the raised components and lowered components of any tensor evaluated at $`u^\mu =0`$ are actually equal. One could, therefore, equally well have written Eqs. (43) โ (45) with all components lowered. The purpose of using raised components is simply to express these equations in manifestly covariant form.
We now mention a few facts concerning the Hermitian conjugate which we used to derive Eq. (44). First, notice from Eq. (21) that $`\mathrm{\Lambda }_{\mu \nu }^{}=\mathrm{\Lambda }_{\mu \nu }`$. Also notice that since $`S^{\mu \nu i}|_0`$ has no dependence on $`u^\mu `$, $`S^{\mu \nu i}|_0`$ and $`\mathrm{\Lambda }_{\mu \nu }`$ commute. This means in particular that the Hermitian conjugate in Eq. (44) may be applied to the $`\pi ^i|_0`$ term alone. Finally, we used Eq. (24) to relate $`\left(\pi _i|_0\right)^{}`$ to $`(\pi _i^{})|_0`$. Notice from Eq. (23) that if $`๐_i|_0`$ is a coordinate basis on $`๐`$ then the Hermitian conjugate may be dispensed with altogether.
We call the terms $`K_{}`$, $`K_{}^p`$ and $`K_R`$ appearing in Eq. (42) the transverse kinetic energy, the (preliminary) tangential kinetic energy, and the curvature energy respectively. The last term $`V_{ex}^p`$, being a scalar, nondifferential operator, we call the (preliminary) extrapotential. The three terms $`K_{}^p`$, $`K_R`$, and $`V_{ex}^p`$ are all order $`ฯต^0`$. The transverse kinetic energy $`K_{}`$ is of order $`ฯต^2`$ and therefore blows up as $`ฯต`$ shrinks to $`0`$. The energy associated with this term will therefore be subtracted off with the remaining three terms giving rise to the residual kinetic energy. Notice that each of the four terms in Eq. (42) is Hermitian with respect to the (scaled) Hermitian conjugate.
### D Proof of Identities (36) โ (38)
We return now to justify Eqs. (36) โ (38). Considering Eq. (36), we have
$`G_{\mu j,\sigma }|_0`$ $`=`$ $`\left(_{๐_\sigma }G_{\mu j}\right)|_0=\left(_{๐_\sigma }๐_\mu ,๐_j\right)|_0`$ (48)
$`=`$ $`_{๐_\sigma }๐_\mu ,๐_j|_0+๐_\mu ,_{๐_\sigma }๐_j|_0,`$ (49)
where in the first equality, we replaced the coordinate derivative by the covariant derivative, treating $`G_{\mu j}`$ as a scalar function. The second equality is the definition of $`G_{\mu j}`$, and the third equality follows from the Leibniz rule and the vanishing of the metric tensor under covariant differentiation. We next define the vector
$$๐_{\sigma \mu }=_{๐_\sigma }๐_\mu ,$$
(50)
which we now demonstrate vanishes on $`๐`$. To prove this, we first show that it is everywhere symmetric in $`\mu `$ and $`\sigma `$ by using the general formula
$$_๐๐_๐๐=[๐,๐],$$
(51)
where $`๐`$ and $`๐`$ are arbitrary vector fields on $`๐`$. By substituting $`๐=๐_\sigma `$ and $`๐=๐_\mu `$ and recalling that $`[๐_\mu ,๐_\sigma ]=0`$, we find that $`๐_{\sigma \mu }=๐_{\mu \sigma }`$. Since $`๐_{\sigma \mu }`$ is symmetric, it vanishes if and only if $`v^\sigma v^\mu ๐_{\sigma \mu }=0`$ for an arbitrary list of (constant) real numbers $`v^\sigma `$. For such an arbitrary list, we define the vector field $`๐ฏ=v^\sigma ๐_\sigma `$ over $`๐`$. Since $`v^\mu /u^\sigma =0`$, we see from Eq. (50) that $`v^\sigma v^\mu ๐_{\sigma \mu }=_๐ฏ๐ฏ`$. Since the quantities $`u^\mu `$ are defined via geodesic flow away from $`๐`$, an integral curve of $`๐ฏ`$ which passes through $`๐`$ is itself a geodesic. By the geodesic equation, $`(_๐ฏ๐ฏ)|_0=0`$. Thus, $`v^\sigma v^\mu ๐_{\sigma \mu }|_0=0`$ and hence
$$\left(_{๐_\sigma }๐_\mu \right)|_0=0.$$
(52)
We return to Eq. (49) and write
$$G_{\mu j,\sigma }|_0=๐_\mu ,_{๐_\sigma }๐_j|_0=๐_\mu ,_{๐_j}๐_\sigma |_0=S_{\mu \sigma j}|_0,$$
(53)
where the first equality follows from Eq. (52), the second from Eqs. (51) and (11), and the third from Eqs. (33) and (7).
To prove Eqs. (37) and (38), we fix a point $`q`$ on the constraint manifold and restrict our attention to a single transverse space $`๐ฐ_q`$ which we temporarily forget is embedded in $`๐`$. Recall that the vectors $`๐_\mu `$ are tangent to $`๐ฐ_q`$ and $`G_{\mu \nu }=๐_\mu ,๐_\nu `$ is the metric tensor on $`๐ฐ_q`$. Furthermore, the coordinates $`u^\mu `$ are Riemannian normal coordinates on $`๐ฐ_q`$ and it is well-known that the expansion of the metric to second order in the Riemannian normal coordinates is
$$G_{\mu \nu }(๐)=\delta _{\mu \nu }\frac{1}{3}\overline{R}_{\mu \sigma \nu \tau }|_0u^\sigma u^\tau +\mathrm{},$$
(54)
where $`\overline{R}`$ is the Riemannian curvature of the transverse space $`๐ฐ_q`$. The vanishing in Eq. (54) of the term linear in $`๐`$ proves Eq. (37). Similarly, the quadratic term in Eq. (54) yields
$$G_{\mu \nu ,\sigma \tau }|_0=\frac{1}{3}\left(\overline{R}_{\mu \sigma \nu \tau }+\overline{R}_{\nu \sigma \mu \tau }\right)|_0.$$
(55)
To complete the proof of Eq. (38), we must prove that the components $`\overline{R}_{\mu \sigma \nu \tau }|_0`$ of the Riemannian curvature on $`๐ฐ_q`$ agree with the components $`R_{\mu \sigma \nu \tau }|_0`$ of the Riemannian curvature on $`๐`$. To prove this, we use the Gauss relation given by Eq. (B10) and which we reexpress here in component form
$$R_{\mu \sigma \nu \tau }=\overline{R}_{\mu \sigma \nu \tau }+\overline{T}_{\sigma \nu }^a\overline{T}_{a\mu \tau }\overline{T}_{\sigma \tau }^a\overline{T}_{a\mu \nu }.$$
(56)
Since we are applying the Gauss equation to the submanifold $`๐ฐ_q`$ instead of $`๐`$, we place an overbar on the symbols for the second fundamental form and the Riemannian curvature. Here, $`\overline{T}`$ is the second fundamental form of $`๐ฐ_q`$. Recall that $`P_{}`$ and $`P_{}`$ were defined to be respectively the tangent and normal projection operators onto $`๐`$. We extend the definition of these operators for $`u^\mu `$ not equal to $`0`$ by defining $`P_{}`$ and $`P_{}`$ to be the normal and tangent projection operators respectively onto $`๐ฐ_q`$. With this definition, the second fundamental form $`\overline{T}`$ is given by (see Eq. (B1))
$$\overline{T}_๐๐=P_{}_{P_{}๐}P_{}๐+P_{}_{P_{}๐}P_{}๐,$$
(57)
where $`๐`$ and $`๐`$ are arbitrary vector fields over $`๐ฐ_q`$ which are tangent to $`๐`$. Since $`๐_\mu `$ is tangent to $`๐ฐ_q`$ everywhere, we have
$$\overline{T}_{\sigma \mu \nu }=๐_\sigma ,\overline{T}_{๐_\nu }๐_\mu =0.$$
(58)
Furthermore, since $`๐_i`$ is normal to $`๐ฐ_q`$ at $`u^\mu =0`$, we have
$$\overline{T}_{i\mu \nu }|_0=๐_i,\overline{T}_{๐_\nu }๐_\mu |_0=๐_i,_{๐_\nu }๐_\mu |_0=0,$$
(59)
where the last equality follows from Eq. (52). Combining Eqs. (58) and (59) yields $`\overline{T}_{\mu \nu }^a|_0=\overline{T}_{a\mu \nu }|_0=0`$ from which follows, using Eq. (56), $`\overline{R}_{\mu \sigma \nu \tau }|_0=R_{\mu \sigma \nu \tau }|_0`$. This concludes the proof of Eq. (38).
### E Computation of the Extrapotential
In this section, we analyze the extrapotential $`V_{ex}^p`$ formed by evaluating Eq. (26) at $`u^\mu =0`$. As we will see, $`V_{ex}^p`$ may be expressed solely in terms of the second fundamental form $`T`$ of the constraint manifold and the Riemannian curvature $`R`$ of $`๐`$ evaluated on $`๐`$ with no dependence on the potential twist $`S`$. Specifically, we will derive the following manifestly covariant form
$$V_{ex}^p=\frac{\mathrm{}^2}{8}\left(2T^{i\mu j}T_{i\mu j}T_i^{i\mu }T_{\mu j}^j+2R_{i\mu }^{i\mu }+\frac{2}{3}R_{\mu \nu }^{\mu \nu }\right)|_0.$$
(60)
Setting $`R_{abcd}=0`$, the above equation agrees with da Costa (Ref. , Eq. (33)). Da Costa also assumes that $`S=0`$. Since we do not make this assumption, Eq. (60) is a generalization of da Costaโs result to both the case of nonzero Riemannian curvature in the ambient space and nonzero twist of the potential.
There are several other convenient forms for $`V_{ex}^p`$. We first introduce the following notation
$``$ $`=`$ $`R_{ab}^{ab}|_0,`$ (61)
$`_{}`$ $`=`$ $`R_{\mu \nu }^{\mu \nu }|_0,`$ (62)
$`_{}`$ $`=`$ $`R_{ij}^{ij}|_0,`$ (63)
$`\widehat{}`$ $`=`$ $`\widehat{R}_{ij}^{ij},`$ (64)
$`๐ฏ^2`$ $`=`$ $`{\displaystyle \frac{1}{2}}T^{abc}T_{abc}=T^{i\mu j}T_{i\mu j}=T^{\mu ij}T_{\mu ij},`$ (65)
$`^2`$ $`=`$ $`T_a^{ab}T_{bc}^c=T_i^{i\mu }T_{\mu j}^j=T_i^{\mu i}T_{\mu j}^j,`$ (66)
where we use Eqs. (B4), (B6), and (B7) in Eqs. (65) and (66). The quantities $``$ and $`\widehat{}`$ are the scalar curvatures on $`๐`$ and $`๐`$ respectively. The quantity $``$ is called the mean curvature. Using the fact that $`=_{}+_{}+2R_{i\mu }^{i\mu }|_0`$, we rewrite Eq. (60) as
$$V_{ex}^p=\frac{\mathrm{}^2}{8}\left(2๐ฏ^2^2+_{}\frac{1}{3}_{}\right).$$
(67)
Furthermore, the Gauss Eq. (B10) yields
$$๐ฏ^2=^2\widehat{}+_{},$$
(68)
from which we find
$`V_{ex}^p`$ $`=`$ $`{\displaystyle \frac{\mathrm{}^2}{8}}\left(๐ฏ^2\widehat{}+{\displaystyle \frac{1}{3}}_{}\right)`$ (69)
$`=`$ $`{\displaystyle \frac{\mathrm{}^2}{8}}\left(^22\widehat{}++_{}{\displaystyle \frac{1}{3}}_{}\right).`$ (70)
Assuming the tensor $`R=0`$, Eq. (70) agrees with Ref. , Eq. (36).
The remainder of this section is devoted to the derivation of Eq. (60). Considering the first term of Eq. (26), we observe that since $`G_{ab}|_0=\delta _{ab}`$, we find
$`[\pi _i\mathrm{ln}G]|_0`$ $`=`$ $`0,`$ (71)
$`[\pi _\mu \mathrm{ln}G]|_0`$ $`=`$ $`i\mathrm{}(G^{ab}G_{ba,\mu })|_0=i\mathrm{}G_{aa,\mu }|_0=i\mathrm{}G_{jj,\mu }|_0,`$ (72)
where we have used Eq. (37) in the last step of Eq. (72). Equations (71) and (72) yield
$$\frac{1}{4}\left(G^{ab}[\pi _a\mathrm{ln}G][\pi _b\mathrm{ln}G]\right)|_0=\frac{\mathrm{}^2}{4}\left(G_{ii,\mu }G_{jj,\mu }\right)|_0.$$
(73)
Considering the second term of Eq. (26), we note
$$[\pi _a^{}G^{ab}[\pi _b\mathrm{ln}G]]|_0=\left([\pi _a^{},G^{ab}][\pi _b\mathrm{ln}G]+G^{ab}[\pi _a^{}[\pi _b\mathrm{ln}G]]\right)|_0.$$
(74)
The first term of Eq. (74) vanishes from Eqs. (21), (22), (40), (41), and (71). The second term evaluates to
$`\left(G^{ab}[\pi _a^{}[\pi _b\mathrm{ln}G]]\right)|_0`$ $`=`$ $`[\pi _\mu [\pi _\mu \mathrm{ln}G]]|_0=\mathrm{}^2\left(G_{ab,\mu }G_{ab,\mu }G_{aa,\mu \mu }\right)|_0`$ (75)
$`=`$ $`\mathrm{}^2\left(2S_{\mu \nu i}S_{\mu \nu i}+G_{ij,\mu }G_{ij,\mu }+{\displaystyle \frac{2}{3}}R_{\mu \nu \mu \nu }G_{ii,\mu \mu }\right)|_0,`$ (76)
where the first equality follows from Eqs. (21), (22), (27), and (71), the second equality is a straightforward computation, and the third equality results from Eqs. (36) โ (38). Collecting the preceding results, we find
$$V_{ex}^p=\frac{\mathrm{}^2}{8}\left(\frac{1}{4}G_{ii,\mu }G_{jj,\mu }+G_{ij,\mu }G_{ij,\mu }G_{ii,\mu \mu }+2S_{\mu \nu i}S_{\mu \nu i}+\frac{2}{3}R_{\mu \nu \mu \nu }\right)|_0.$$
(77)
The various derivatives of $`G_{ij}`$ appearing in the above may be reexpressed using the following identities, to be derived momentarily,
$`G_{ij,\mu }|_0`$ $`=`$ $`2T_{i\mu j}|_0,`$ (78)
$`G_{ij,\mu \nu }|_0`$ $`=`$ $`(T_{a\mu i}T_{a\nu j}+T_{a\mu j}T_{a\nu i}+S_{a\mu i}S_{a\nu j}+S_{a\mu j}S_{a\nu i}R_{i\mu j\nu }R_{j\mu i\nu })|_0.`$ (79)
Upon inserting Eqs. (78) and (79) into Eq. (77) one obtains Eq. (60).
Considering Eq. (78), it follows from
$$G_{ij,\mu }|_0=(_{๐_\mu }๐_i,๐_j)|_0=_{๐_i}๐_\mu ,๐_j|_0+๐_i,_{๐_j}๐_\mu |_0=2T_{i\mu j}|_0,$$
(80)
where in the second equality we used the Leibniz rule and interchanged the derivatives by virtue of Eqs. (51) and (11). The final equality follows from the definition of the second fundamental form Eq. (B1) and Eqs. (B2) and (B3).
Considering Eq. (79), we have
$`G_{ij,\mu \nu }|_0`$ $`=`$ $`\left(_{๐_\nu }_{๐_\mu }๐_i,๐_j\right)|_0`$ (81)
$`=`$ $`_{๐_i}๐_\mu ,_{๐_j}๐_\nu |_0+_{๐_i}๐_\nu ,_{๐_j}๐_\mu |_0`$ (83)
$`+_{๐_\nu }_{๐_i}๐_\mu ,๐_j|_0+๐_i,_{๐_\nu }_{๐_j}๐_\mu |_0.`$
In the second equality, we again applied the Leibniz rule and interchanged derivatives by virtue of Eqs. (51) and (11). We next note that the covariant derivative of $`๐_\mu `$ by $`๐_i`$ is given by
$$\left(_{๐_i}๐_\mu \right)|_0=\left(P_{}_{๐_i}๐_\mu +P_{}_{๐_i}๐_\mu \right)|_0=\left(T_{๐_i}๐_\mu +S_{๐_i}๐_\mu \right)|_0,$$
(84)
where the first equality follows from the fact that $`P_{}+P_{}`$ is the identity and the second from the definitions Eqs. (B1) and (7) and the fact that $`๐_\mu `$ is normal to $`๐`$. We also observe from Eq. (52) that $`\left(_{๐_i}_{๐_\nu }๐_\mu \right)|_0=0`$, and therefore
$$\left(_{๐_\nu }_{๐_i}๐_\mu \right)|_0=\left(R_{๐_\nu ๐_i}๐_\mu \right)|_0,$$
(85)
where we have used Eqs. (11) and (B8). Inserting Eqs. (84) and (85) into Eq. (83) yields the desired result Eq. (79).
## VI The Constrained Hamiltonian
In Sect. V we expanded the kinetic energy in $`ฯต`$, obtaining two terms. One term, the transverse kinetic energy, is of order $`ฯต^2`$; the other term is of order $`ฯต^0`$. In this section, we apply (degenerate) first order perturbation theory to derive a constrained Hamiltonian for the eigenenergies. In doing so, we introduce the transverse modes characterizing the wave function away from the constraint manifold.
### A Rescaling by $`ฯต`$ and the expansion of the Hamiltonian
By adding the potential energy $`V_{}(๐)`$ to the kinetic energy Eq. (42), we have the following Hamiltonian
$$H=H_{}+H_{}+O(ฯต),$$
(86)
where
$`H_{}`$ $`=`$ $`K_{}+V_{},`$ (87)
$`H_{}^p`$ $`=`$ $`K_{}^p+K_R+V_{ex}^p,`$ (88)
are called the transverse and (preliminary) tangential Hamiltonians respectively.
In order to clarify the subsequent perturbation analysis, we explicitly exhibit the $`ฯต`$ dependence of various quantities by rescaling them in $`ฯต`$. To begin, we repeat the previous definition Eq. (10) of the rescaled quantities $`\stackrel{~}{u}^\mu `$ and also define rescaled momenta $`\stackrel{~}{\pi }_\mu `$,
$`u^\mu `$ $`=`$ $`ฯต\stackrel{~}{u}^\mu ,`$ (89)
$`\pi _\mu `$ $`=`$ $`{\displaystyle \frac{1}{ฯต}}\stackrel{~}{\pi }_\mu .`$ (90)
Notice that both $`\stackrel{~}{u}^\mu `$ and $`\stackrel{~}{\pi }_\mu `$ scale as $`ฯต^0`$. In general, the scaled version of a quantity (denoted with a tilde) is defined such that the lowest order nonvanishing term of its expansion in $`ฯต`$ is of order $`ฯต^0`$. Thus, for a quantity homogeneous in $`ฯต`$, the scaled version is independent of $`ฯต`$. For convenience, we repeat the definition Eq. (9) of the rescaled potential energy $`\stackrel{~}{V}_{}`$ and also define a rescaled transverse kinetic energy and transverse Hamiltonian
$`\stackrel{~}{V}_{}(\stackrel{~}{๐};ฯต)`$ $`=`$ $`ฯต^2V_{}(๐;ฯต)=ฯต^2V_{}(ฯต\stackrel{~}{๐};ฯต),`$ (91)
$`\stackrel{~}{K}_{}`$ $`=`$ $`ฯต^2K_{}={\displaystyle \frac{1}{2}}\stackrel{~}{\pi }_\mu \stackrel{~}{\pi }_\mu ,`$ (92)
$`\stackrel{~}{H}_{}(ฯต)`$ $`=`$ $`ฯต^2H_{}(ฯต)=\stackrel{~}{K}_{}+\stackrel{~}{V}_{}(\stackrel{~}{๐};ฯต).`$ (93)
By our previous assumptions in Sect. IV C, $`\stackrel{~}{V}_{}(\stackrel{~}{๐};ฯต)`$ is smooth in $`ฯต`$ and does not vanish at $`ฯต=0`$. As for $`\stackrel{~}{K}_{}`$, it is clearly independent of $`ฯต`$. Thus, $`\stackrel{~}{H}_{}(ฯต)`$ is smooth in $`ฯต`$ at $`ฯต=0`$; its lowest order term is order $`ฯต^0`$, but depending on $`\stackrel{~}{V}_{}`$, it may have higher order terms as well. Recall that $`K_{}^p`$, $`K_R`$, $`V_{ex}^p`$, and $`H_{}^p`$ are already independent of $`ฯต`$ and therefore need no further scaling. For notational continuity, however, we nevertheless define
$`\stackrel{~}{K}_{}^p`$ $`=`$ $`K_{}^p,`$ (94)
$`\stackrel{~}{K}_R`$ $`=`$ $`K_R,`$ (95)
$`\stackrel{~}{V}_{ex}^p`$ $`=`$ $`V_{ex}^p,`$ (96)
$`\stackrel{~}{H}_{}^p`$ $`=`$ $`H_{}^p.`$ (97)
We rescale the full Hamiltonian by defining
$$\stackrel{~}{H}(ฯต)=ฯต^2H(ฯต)=\stackrel{~}{H}_{}(ฯต)+ฯต^2\stackrel{~}{H}_{}^p+O(ฯต^3).$$
(98)
In a typical Taylor series expansion of $`\stackrel{~}{H}(ฯต)`$, we would remove the order $`ฯต`$ term from $`\stackrel{~}{H}_{}(ฯต)`$ and leave it as a separate term. We would also combine the order $`ฯต^2`$ term of $`\stackrel{~}{H}_{}(ฯต)`$ with the tangential Hamiltonian $`ฯต^2\stackrel{~}{H}_{}^p`$. Here, however, we wish to keep the $`ฯต`$ and $`ฯต^2`$ terms together in the transverse Hamiltonian $`\stackrel{~}{H}_{}(ฯต)`$. We therefore define a new perturbation parameter $`\kappa =ฯต^2`$ and rewrite Eq. (98) as
$$\stackrel{~}{H}(ฯต,\kappa )=\stackrel{~}{H}_{}(ฯต)+\kappa \stackrel{~}{H}_{}^p+O(ฯต^3).$$
(99)
Our objective is to find the eigenvalues of $`\stackrel{~}{H}`$ through order $`ฯต^2`$. Viewing $`ฯต`$ and $`\kappa `$ as formally independent in Eq. (99), our objective becomes finding the eigenvalues of $`\stackrel{~}{H}`$ through second order in $`ฯต`$ and first order in $`\kappa `$. Our procedure is to assume that the eigenvalues of $`H_{}(ฯต)`$ can be solved exactly (or at least through order $`ฯต^2`$) and then apply first order perturbation theory in $`\kappa `$. To simplify notation, we drop the $`ฯต`$ dependence (but not $`\kappa `$ dependence) for the duration of the derivation.
### B Transformation to the Transverse Modes
The zeroth order term (in $`\kappa `$) of $`\stackrel{~}{H}(\kappa )`$ is the transverse Hamiltonian $`\stackrel{~}{H}_{}`$, which has the form
$$\stackrel{~}{H}_{}=\frac{\mathrm{}^2}{2}\frac{}{\stackrel{~}{u}^\mu }\frac{}{\stackrel{~}{u}^\mu }+\stackrel{~}{V}_{}(\stackrel{~}{๐}).$$
(100)
Since $`\stackrel{~}{H}_{}`$ depends only on the quantities $`\stackrel{~}{u}^\mu `$, we may restrict its domain to functions of $`\stackrel{~}{u}^\mu `$ alone. For the moment we adopt this understanding for the domain of $`\stackrel{~}{H}_{}`$. We pick an eigenvalue $`\stackrel{~}{E}_{}`$ (the transverse energy) of $`\stackrel{~}{H}_{}`$ with finite multiplicity $`k`$ and bounded eigenstates. We call these eigenstates the transverse modes (with energy $`\stackrel{~}{E}_{}`$). We let $`\chi _n(\stackrel{~}{๐})`$, $`n=1,\mathrm{},k`$, denote an orthonormal basis of these transverse modes. By orthonormal, we mean
$$\chi _n|\chi _n^{}_๐=๐u^1\mathrm{}du^d\chi _n^{}(\stackrel{~}{๐})\chi _n^{}(\stackrel{~}{๐})=\delta _{nn^{}},$$
(101)
where the $`๐`$ subscript indicates integration only over the variables $`u^\mu `$ as opposed to the full $`(d+m)`$-form $`\nu `$ in Eq. (19).
We now adopt the understanding that $`\stackrel{~}{H}_{}`$ acts on wave functions of both $`\stackrel{~}{u}^\mu `$ and $`q`$. Such an eigenfunction with eigenvalue $`\stackrel{~}{E}_{}`$ has the general form
$$\psi (\stackrel{~}{๐},q)=\underset{n=1}{\overset{k}{}}\chi _n(\stackrel{~}{๐})\varphi _n(q),$$
(102)
where the functions $`\varphi _n(q)`$ are arbitrary. We therefore identify an eigenfunction $`\psi (\stackrel{~}{๐},q)`$, having eigenvalue $`\stackrel{~}{E}_{}`$, with the $`k`$ functions $`\varphi _n(q)`$. Notice that with our current understanding for the domain of the operator $`\stackrel{~}{H}_{}`$, $`\stackrel{~}{E}_{}`$ is a degenerate eigenvalue, even for $`k=1`$.
Recall the steps involved in first order degenerate perturbation theory. First, one either proves or assumes that the desired eigenvalue and eigenfunction is analytic in the perturbation parameter $`\kappa `$. (Here, we simply assume this fact.) Next, one determines the zeroth order energy and zeroth order eigenstates. Then, one considers the operator formed by restricting the first order term of the Hamiltonian to the space of zeroth order eigenstates. The first order corrections to the energy are the eigenvalues of this restricted operator. In the present case, the zeroth order energy is $`\stackrel{~}{E}_{}`$, and the zeroth order eigenstates are given by Eq. (102). The first order correction to the Hamiltonian is $`\kappa \stackrel{~}{H}_{}^p`$. Denoting the first order correction to the energy by $`\kappa \stackrel{~}{E}_{}`$, the eigenvalue equation for $`\stackrel{~}{E}_{}`$ is
$$\underset{n^{}=1}{\overset{k}{}}\left(\stackrel{~}{H}_{}\right)_{nn^{}}\varphi _n^{}=\stackrel{~}{E}_{}\varphi _n,$$
(103)
where the $`(\stackrel{~}{H}_{})_{nn^{}}`$ are the differential operators
$$\left(\stackrel{~}{H}_{}\right)_{nn^{}}=\chi _n|\stackrel{~}{H}_{}^p\chi _n^{}_๐=๐u^1\mathrm{}du^d\chi _n^{}(\stackrel{~}{๐})\left(\stackrel{~}{H}_{}^p\chi _n^{}\right)(\stackrel{~}{๐}).$$
(104)
We call $`(\stackrel{~}{H}_{})_{nn^{}}`$ the constrained, or tangential, Hamiltonian.
We now recall that $`\kappa =ฯต^2`$ and reintroduce the explicit $`ฯต`$ dependence. Summarizing our analysis thus far, we have shown that an eigenvalue of $`\stackrel{~}{H}(ฯต)`$ through order $`ฯต^2`$ is given by $`\stackrel{~}{E}_{}(ฯต)+ฯต^2\stackrel{~}{E}_{}(ฯต)`$, where $`\stackrel{~}{E}_{}(ฯต)`$ and $`\stackrel{~}{E}_{}(ฯต)`$ are eigenvalues of $`\stackrel{~}{H}_{}(ฯต)`$ and $`(\stackrel{~}{H}_{})_{nn^{}}(ฯต)`$. Of course, assuming smoothness in $`ฯต`$, it is sufficient to solve for $`\stackrel{~}{E}_{}(ฯต)`$ and $`\stackrel{~}{E}_{}(ฯต)`$ through orders $`ฯต^2`$ and $`ฯต^0`$ respectively. We will therefore only require $`\stackrel{~}{E}_{}(ฯต)`$ and $`(\stackrel{~}{H}_{})_{nn^{}}(ฯต)`$ evaluated at $`ฯต=0`$, which we denote by $`\stackrel{~}{E}_{}`$ and $`(\stackrel{~}{H}_{})_{nn^{}}`$ respectively. Also, by virtue of Eq. (104), we assume for the remainder of the paper that the transverse modes $`\chi _n(\stackrel{~}{๐})`$ are only order $`ฯต^0`$ eigenfunctions of $`\stackrel{~}{H}_{}(ฯต)`$.
We view Eq. (103) as a $`k`$-dimensional vector wave equation for a vector wave function defined over the constraint manifold. We introduce the bold notation $`\mathit{\varphi }(q)`$ for the vector wave function with components $`\varphi _n(q)`$ and the sans serif notation $`\stackrel{~}{๐ง}_{}`$ for the matrix of differential operators with components $`(\stackrel{~}{H}_{})_{nn^{}}`$. Equation (103) can therefore be written more compactly as
$$๐ง_{}\mathit{\varphi }=E_{}\mathit{\varphi }.$$
(105)
Having completed the perturbation analysis, we have dropped the tildes from $`๐ง_{}`$ and its eigenvalue $`E_{}`$. Using Eqs. (88), (44), (45) and a little algebra, we express $`๐ง_{}`$ as
$$๐ง_{}=๐ช_{}+๐ต_{ex},$$
(106)
where
$`๐ช_{}`$ $`=`$ $`{\displaystyle \frac{1}{2}}(\pi ^i|_0๐จ+S^{\mu \nu i}|_0\mathsf{\Lambda }_{\mu \nu })^{}(\pi _i๐จ+S_i^{\sigma \tau }|_0\mathsf{\Lambda }_{\sigma \tau }),`$ (107)
$`๐ต_{ex}`$ $`=`$ $`V_{ex}^p๐จ+\left({\displaystyle \frac{1}{2}}S^{\mu \nu i}S_i^{\sigma \tau }+{\displaystyle \frac{1}{6}}R^{\mu \nu \sigma \tau }\right)|_0\mathsf{\Lambda }_{\mu \nu \sigma \tau }^{(2)}\left({\displaystyle \frac{1}{2}}S^{\mu \nu i}S_i^{\sigma \tau }\right)|_0\mathsf{\Lambda }_{\mu \nu }\mathsf{\Lambda }_{\sigma \tau }`$ (108)
$`=`$ $`V_{ex}^p๐จ+\left({\displaystyle \frac{1}{2}}S^{\mu \nu i}S_i^{\sigma \tau }\right)|_0\left(\mathsf{\Lambda }_{\mu \nu \sigma \tau }^{(2)}\mathsf{\Lambda }_{\mu \nu }\mathsf{\Lambda }_{\sigma \tau }\right)+{\displaystyle \frac{1}{6}}R^{\mu \nu \sigma \tau }|_0\mathsf{\Lambda }_{\mu \nu \sigma \tau }^{(2)},`$ (109)
and where $`๐จ`$ is the $`k\times k`$ identity matrix and $`\mathsf{\Lambda }_{\mu \nu }`$ and $`\mathsf{\Lambda }_{\mu \nu \sigma \tau }^2`$ are the $`k\times k`$ matrices having the following components respectively
$`(\mathrm{\Lambda }_{\mu \nu })_{nn^{}}`$ $`=`$ $`\chi _n|\mathrm{\Lambda }_{\mu \nu }\chi _n^{}_๐,`$ (110)
$`(\mathrm{\Lambda }_{\mu \nu \sigma \tau }^{(2)})_{nn^{}}`$ $`=`$ $`\chi _n|\mathrm{\Lambda }_{\mu \nu }\mathrm{\Lambda }_{\sigma \tau }\chi _n^{}_๐.`$ (111)
Equations (106) โ (109) encapsulate the main result of this paper. Specifically, the constrained Hamiltonian $`๐ง_{}`$ is a $`k\times k`$ matrix of differential operators. It is the residual Hamiltonian remaining after the infinite transverse energy $`E_{}`$ is subtracted off. The kinetic energy $`๐ช_{}`$, which we call the (final) tangential kinetic energy, differs from the โstandardโ kinetic energy in that it has a gauge potential term. Physically, the gauge potential couples the tangential momenta to the generalized angular momentum of the transverse modes. The quantity $`๐ต_{ex}`$ is a $`k\times k`$ matrix of nondifferential operators which we call the (final) extrapotential. Notice that any possible off-diagonal coupling in $`๐ง_{}`$ is due to the angular momentum of the transverse modes. The preliminary tangential kinetic energy $`K_{}^p`$, extrapotential $`V_{ex}^p`$, and tangential Hamiltonian $`H_{}^p`$ are distinguished from the (final) tangential kinetic energy $`๐ช_{}`$, extrapotential $`๐ต_{ex}`$, and tangential Hamiltonian $`๐ง_{}`$ by the โ$`p`$โ superscript. We often drop the โpreliminaryโ and โfinalโ modifiers when referring to these terms, relying on their symbols and context to make our precise meaning clear.
In the event of a nondegenerate transverse mode, that is $`k=1`$, $`๐ง_{}`$ becomes a scalar wave operator $`H_{}`$ acting on scalar wave functions defined over the constraint manifold. In this case, we see the emergence of Eqs. (3) โ (5) presented in Sect. II. The exact derivation of these equations from the more general Eqs. (106) โ (109) will be presented in Sect. VIII B.
### C Nonconstant Transverse Potentials
Up to now, we have assumed that the transverse potential $`V_{}(๐)`$ is constant (modulo $`SO(d)`$ rotations) along the constraint manifold $`๐`$. For some physical systems this assumption holds exactly due to some symmetry on the ambient space, such as $`SO(3)`$ rotations in the case of a rigid body. However, for other systems, this assumption may be only approximately satisfied; the constraining potential may in fact vary along the constraint manifold. This is true, for example, of a molecule evolving along a reaction path; there is no symmetry dictating that the frequencies of the small transverse vibrations be constant. The purpose of this section is to illustrate how small variations in the transverse potential may be easily included within our formalism.
The key idea is to only allow dependence on $`q`$ at order $`ฯต^2`$. Specifically, we assume the transverse potential can be expanded as
$$\stackrel{~}{V}_{}(\stackrel{~}{๐},q;ฯต)=\stackrel{~}{V}_{}^0(\stackrel{~}{๐})+ฯต\stackrel{~}{V}_{}^1(\stackrel{~}{๐})+ฯต^2\stackrel{~}{V}_{}^2(\stackrel{~}{๐},q)+O(ฯต^3).$$
(112)
Applying this expansion to Eq. (93), an eigenvalue $`\stackrel{~}{E}_{}`$ of $`\stackrel{~}{H}_{}`$ (assuming analyticity in $`ฯต`$) can be expanded as
$$\stackrel{~}{E}_{}(q;ฯต)=\stackrel{~}{E}_{}^0+ฯต\stackrel{~}{E}_{}^1+ฯต^2\stackrel{~}{E}_{}^2(q)+O(ฯต^3).$$
(113)
The first two terms of $`E_{}=\stackrel{~}{E}_{}/ฯต^2`$ blow up as $`ฯต`$ goes to $`0`$. However, these two terms are constant in $`q`$ and may thus be subtracted off. The next order term $`\stackrel{~}{E}_{}^2(q)`$ does depend on $`q`$ and is of the same order in $`ฯต`$ as $`๐ง_{}`$. Thus, $`\stackrel{~}{E}_{}^2(q)`$ may be combined with the extrapotential $`๐ต_{ex}`$ in $`๐ง_{}`$ to form the effective potential
$$๐ต_{ef}(q)=๐ต_{ex}(q)+\stackrel{~}{E}_{}^2(q)๐จ.$$
(114)
This is the only modification which needs to be made to our formalism. Notice that the transverse modes $`\chi _n(๐)`$ need not be modified since they are defined to be only order $`ฯต^0`$ eigenfunctions of $`H_{}`$ and hence are unaffected by the term $`\stackrel{~}{E}_{}^2(q)`$.
## VII Analysis of Connections
Both the preliminary and the final tangential kinetic energies $`K_{}^p`$ and $`๐ช_{}`$ exhibit a gauge potential proportional to $`S_{abc}`$. In this section, we study the geometric origins of these gauge potentials and compute their curvatures.
We begin by reviewing the connection on normal vector fields over $`๐`$. We note that many equivalent definitions exist for the general concept of a connection. For the purposes of this paper, a connection is taken to be a covariant derivative operator which acts on some space of vector fields. For more background, see any of a number of standard references . For the remainder of this section, $`๐ฏ`$ is an arbitrary normal vector field over $`๐`$ and $`๐ฑ`$ and $`๐ฒ`$ are arbitrary tangent vector fields over $`๐`$. The normal connection $`^N`$ is defined by
$$_๐ฑ^N๐ฏ=P_{}_๐ฑ๐ฏ=P_{}_๐ฑP_{}๐ฏ,$$
(115)
where $``$ is the Levi-Civita connection on $`๐`$. Notice that $`_๐ฑ^N๐ฏ`$ is itself a normal vector field. The curvature of $`^N`$, denoted $`B^N`$, is computed to be
$`B_{\mathrm{๐ฑ๐ฒ}}^N๐ฏ`$ $`=`$ $`\left(_๐ฑ^N_๐ฒ^N_๐ฒ^N_๐ฑ^N_{[๐ฑ,๐ฒ]}^N\right)๐ฏ`$ (116)
$`=`$ $`P_{}\left(_๐ฑP_{}_๐ฒ_๐ฒP_{}_๐ฑ_{[๐ฑ,๐ฒ]}\right)P_{}๐ฏ`$ (117)
$`=`$ $`P_{}\left(R_{\mathrm{๐ฑ๐ฒ}}_๐ฑP_{}_๐ฒ+_๐ฒP_{}_๐ฑ\right)P_{}๐ฏ=P_{}\left(R_{\mathrm{๐ฑ๐ฒ}}T_๐ฑT_๐ฒ+T_๐ฒT_๐ฑ\right)P_{}๐ฏ,`$ (118)
where the first equality is simply the definition of the curvature, the second follows from Eq. (115), the third from noting $`P_{}=IP_{}`$ and Eq. (B8), and the forth from Eq. (B1). As expected, the curvature depends only on the nature of the embedding of $`๐`$ (via the tensor $`T`$) and on the curvature of $`๐`$. If we assume that the tensors $`B^N`$ and $`R`$ vanish, then we obtain the class of embeddings considered by da Costa . For such embeddings, one can choose a potential frame with vanishing twist, thus eliminating coupling between the transverse modes. (This follows from Eq. (121) below and the fact that for vanishing curvature, one can always find a frame for which the gauge potential vanishes.) Also, for a non-twisting potential frame, the submanifolds of constant potential are orthogonal to the transverse spaces $`๐ฐ_q`$. Hence, at all points the restoring force is directed inward tangent to the $`๐ฐ_q`$.
It is instructive to compute the gauge potential explicitly for the connection $`^N`$. For this computation we first choose an arbitrary orthonormal frame (not necessarily the potential frame) $`๐_\mu `$, $`\mu =1,\mathrm{},d`$, for each normal space $`N_q`$. We denote the components of an arbitrary normal vector field $`๐ฏ`$ with respect to $`๐_\mu `$ by $`v^\mu `$. Then, the components of $`_{๐_i}^N๐ฏ`$ are given by
$$(_{๐_i}^N๐ฏ)^\mu =๐_iv^\mu +(A_i^N)_\nu ^\mu v^\nu ,$$
(119)
where we have defined the gauge potential
$$(A_i^N)_{\mu \nu }=๐_\mu ,_{๐_i}^N๐_\nu =๐_\mu ,_{๐_i}๐_\nu .$$
(120)
Due to the orthonormality of the $`๐_\mu `$, $`(A_i^N)_{\mu \nu }`$ is antisymmetric in $`\mu `$ and $`\nu `$. The gauge potential can therefore be viewed as a one-form on $`๐`$ with values in the Lie algebra $`so(d)`$, which contains all antisymmetric $`d\times d`$ matrices. If we choose $`๐_\mu =๐_\mu `$, we recognize from Eq. (7) that the gauge potential is related to the potential twist tensor by
$$(A_i^N)_{\mu \nu }=S_{\mu \nu i}.$$
(121)
This result will be import below for analyzing $`K_{}^p`$ and $`๐ช_{}`$.
We now consider a function $`\psi (๐ฏ,q)`$, such as the quantum wave function, defined in the neighborhood of $`๐`$. (We use the bold notation $`๐ฏ`$ instead of sans serif used earlier because we wish to emphasize the dependence of $`\psi `$ on the normal vector and not on its components with respect to a given frame, such as the potential frame.) The connection $`^N`$ which acts on normal vector fields gives rise to another connection $`^p`$ which acts on the function $`\psi (๐ฏ,q)`$. In order to define $`(_๐ฑ^p\psi )(๐ฏ,q)`$, we first choose a path $`q^{}(\alpha )`$ such that $`q^{}(0)=q`$ and $`(dq^{}/d\alpha )(0)=๐ฑ`$. We then denote by $`๐ฏ^{}(\alpha )`$ the unique normal vector at each point $`q^{}(\alpha )`$ satisfying $`๐ฏ^{}(0)=๐ฏ`$ and $`(_{d/d\alpha }^N๐ฏ^{})(\alpha )=0`$. Then, the connection $`^p`$ is defined by
$$(_๐ฑ^p\psi )(๐ฏ,q)=\frac{d}{d\alpha }|_{\alpha =0}\psi \mathbf{(}๐ฏ^{}(\alpha ),q^{}(\alpha )\mathbf{)}.$$
(122)
The transverse kinetic energy $`K_{}^p`$ can be directly related to the covariant derivative $`^p`$. To do this, it is useful to compute the gauge potential of $`^p`$ explicitly. As before, we consider an orthonormal frame $`๐_\mu `$ and denote the components of $`๐ฏ`$ by $`v^\mu `$. Then, the function $`\psi (๐ฏ,q)`$ can also be interpreted as a function of $`(๐,q)`$, where $`๐=(v^1,\mathrm{},v^d)`$ is the collection of components. We therefore have
$`(_{๐_i}^p\psi )(๐ฏ,q)`$ $`=`$ $`{\displaystyle \frac{d}{d\alpha }}|_{\alpha =0}\psi \mathbf{(}๐^{}(\alpha ),q^{}(\alpha )\mathbf{)}={\displaystyle \frac{d}{d\alpha }}|_{\alpha =0}\psi \mathbf{(}๐^{}(\alpha ),q\mathbf{)}+{\displaystyle \frac{d}{d\alpha }}|_{\alpha =0}\psi \mathbf{(}๐,q^{}(\alpha )\mathbf{)}`$ (123)
$`=`$ $`(๐_iv_{}^{}{}_{}{}^{\mu })(q){\displaystyle \frac{\psi }{v^\mu }}(๐ฏ,q)+(๐_i\psi )\mathbf{(}๐ฏ,q\mathbf{)},`$ (124)
where in the third equality, the derivatives $`/v^\mu `$ and $`๐_i`$ are understood to have $`q`$ and $`v^\nu `$ held fixed respectively. From Eq. (119) and the condition $`_{๐_i}^N๐ฏ^{}=0`$, we find
$$๐_iv_{}^{}{}_{}{}^{\mu }=(A_i^N)_\nu ^\mu v_{}^{}{}_{}{}^{\nu }.$$
(125)
Inserting this result into Eq. (124) yields
$$(_{๐_i}^p\psi )(๐ฏ,q)=\left[\mathbf{(}๐_i+A_i^p\mathbf{)}\psi \right](๐ฏ,q),$$
(126)
where
$$A_i^p=(A_i^N)^{\mu \nu }\mathrm{\Omega }_{\mu \nu },$$
(127)
and where we have used the antisymmetry of $`(A_i^N)_{\mu \nu }`$ to introduce the operator
$$\mathrm{\Omega }_{\mu \nu }=\frac{1}{2}\left(v_\mu \frac{}{v^\nu }v_\nu \frac{}{v^\mu }\right).$$
(128)
Obviously if $`๐_\mu =๐_\mu `$, then $`\mathrm{\Lambda }_{\mu \nu }=i\mathrm{}\mathrm{\Omega }_{\mu \nu }`$. The relevance of $`^p`$ for $`K_{}^p`$ is now clear. By choosing $`๐_\mu =๐_\mu `$ and applying Eqs. (121), (126), and (127) to Eq. (44), we see that
$$K_{}^p=\frac{\mathrm{}^2}{2}(_{๐_i}^p)^{}_{๐_i}^p.$$
(129)
Thus the preliminary tangential kinetic energy is just proportional to the Laplacian defined in terms of the connection $`^p`$. (Compare Eq. (129) to Eq. (C2).)
Considering Eq. (127), we see that the two gauge potentials $`(A_i^N)_{\mu \nu }`$ and $`A_i^p`$ differ only in their representation of $`so(d)`$. For $`(A_i^N)_{\mu \nu }`$, we use a representation by $`d\times d`$ antisymmetric matrices, whereas for $`A_i^p`$ we use a representation by the operators $`\mathrm{\Omega }_{\mu \nu }`$. Therefore, the curvature of the connections $`^N`$ and $`^p`$ are also related by simply switching the representation of $`so(d)`$. Hence the curvature $`B^p`$ of $`^p`$ is
$`B_{\mathrm{๐ฑ๐ฒ}}^p\psi `$ $`=`$ $`\left(_๐ฑ^p_๐ฒ^p_๐ฒ^p_๐ฑ^p_{[๐ฑ,๐ฒ]}^p\right)\psi =(B_{\mathrm{๐ฑ๐ฒ}}^N)^{\mu \nu }\mathrm{\Omega }_{\mu \nu }\psi `$ (130)
$`=`$ $`\left(R_{\mathrm{๐ฑ๐ฒ}}T_๐ฑT_๐ฒ+T_๐ฒT_๐ฑ\right)^{\mu \nu }\mathrm{\Omega }_{\mu \nu }\psi .`$ (131)
We now consider a $`k`$-dimensional vector-valued function $`\mathit{\varphi }(q)`$ with components $`\varphi _n(q)`$. The connection $`^p`$ induces a connection $`^{}`$ on $`\mathit{\varphi }`$ by the formula
$$(_๐ฑ^{}\mathit{\varphi })_n=\chi _n|_๐ฑ^p\underset{n^{}=1}{\overset{k}{}}\chi _n^{}\varphi _n^{}_๐.$$
(132)
The tangential kinetic energy $`๐ช_{}`$ is closely related to the connection $`^{}`$ as we now show. We take the orthonormal frame $`๐_\mu `$ to be $`๐_\mu `$, and we recall that $`\chi _n(๐)`$ is a function of $`u^\mu `$ alone and $`\mathit{\varphi }(q)`$ is a function of $`q`$ alone. Then applying Eqs. (121), (126), and (127), we find
$`_๐ฑ^p\chi _n`$ $`=`$ $`(S_๐ฑ)^{\mu \nu }\mathrm{\Omega }_{\mu \nu }\chi _n,`$ (133)
$`_๐ฑ^p\varphi _n`$ $`=`$ $`๐ฑ\varphi _n,`$ (134)
where $`(S_๐ฑ)_{\mu \nu }=๐_\mu ,S_๐ฑ๐_\nu `$. Then Eq. (132) yields
$`_๐ฑ^{}`$ $`=`$ $`๐ฑ๐จ+๐ _๐ฑ^{},`$ (135)
$`๐ _๐ฑ^{}`$ $`=`$ $`(S_๐ฑ)^{\mu \nu }\mathsf{\Omega }_{\mu \nu },`$ (136)
where $`\mathsf{\Omega }_{\mu \nu }`$ is the $`k\times k`$ matrix with components
$$(\mathrm{\Omega }_{\mu \nu })_{nn^{}}=\chi _n|\mathrm{\Omega }_{\mu \nu }\chi _n^{}_๐.$$
(137)
From Eq. (133), we note that the components of $`๐ _๐ฑ^{}`$ can also be written as
$$(A_๐ฑ^{})_{nn^{}}=\chi _n|_๐ฑ^p\chi _n^{}_๐.$$
(138)
Equations (135) and (136) show that the tangential kinetic energy $`๐ช_{}`$, Eq. (107), is given by
$$๐ช_{}=\frac{\mathrm{}^2}{2}(_{๐_i}^{})^{}_{๐_i}^{},$$
(139)
analogous to Eq. (129) for $`K_{}^p`$.
The connection $`^{}`$ is closely related to the adiabatic transport of quantum states and the associated geometric phase due to Berry . If a set of $`k`$ degenerate quantum states $`\xi _n(\eta )`$, $`n=1,\mathrm{},k`$, depending smoothly on a set of $`m`$ external parameters $`\eta =(\eta _1,\mathrm{},\eta _m)`$, is subject to an adiabatic variation $`\eta (\alpha )`$ of these parameters, then the $`\xi _n(\alpha )=\xi _n(\eta (\alpha ))`$ satisfy $`\xi _n|d\xi _n^{}/d\alpha =_i\xi _n|\xi _n^{}/\eta _i(d\eta _i/d\alpha )=0`$. Simon recognized that this condition defines a connection
$$_{/\eta _i}^B=\frac{}{\eta _i}๐จ+๐ _i^B$$
(140)
acting on the vector-valued wave function $`๐=(\xi _1,\mathrm{},\xi _k)`$ parameterized by $`\eta `$. The gauge potential $`๐ _i^B`$ is a $`k\times k`$ matrix with components
$$(A_i^B)_{nn^{}}=\xi _n|\frac{\xi _n^{}}{\eta _i}.$$
(141)
If the parameters $`\eta _i`$ are themselves quantized, then the momentum conjugate to $`\eta _i`$ is not simply $`i\mathrm{}(/\eta _i)๐จ`$, but rather $`i\mathrm{}_{/\eta _i}^B=i\mathrm{}(/\eta _i๐จ+๐ _i^B)`$. This situation applies, for example, to the Born-Oppenheimer theory of molecules, wherein the parameters $`\eta _i`$ describe the positions of the nuclei and the $`\xi _n`$ represent the quantum state of the electrons. For the constrained quantum systems considered in this paper, the ordering in $`ฯต`$ adiabatically separates the transverse modes $`\chi _n`$ (analogous to the $`\xi _n`$) from the motion along the constraint manifold (analogous to the space of $`\eta _i`$). Therefore, the gauge potential $`๐ _๐ฑ^{}`$ occurring in Eq. (135) is essentially the same as Berryโs gauge potential $`๐ _i^B`$ occurring in Eq. (140). We say โessentially the sameโ because the coordinate derivative $`/\eta _i`$ of Eq. (141) has been replaced by the covariant derivative $`^p`$ of Eq. (138), this covariant derivative being the geometrically natural connection for the transverse modes.
We next compute the curvature of the connection $`^{}`$. In terms of the gauge potential $`๐ ^{}=(S_๐ฑ)^{\mu \nu }\mathsf{\Omega }_{\mu \nu }`$, we have
$$๐ก_{\mathrm{๐ฑ๐ฒ}}^{}=(d๐ ^{})(๐ฑ,๐ฒ)+[๐ _๐ฑ^{},๐ _๐ฒ^{}]=(dS^{\mu \nu })(๐ฑ,๐ฒ)\mathsf{\Omega }_{\mu \nu }+(S_๐ฑ)^{\mu \nu }(S_๐ฒ)^{\sigma \tau }[\mathsf{\Omega }_{\mu \nu },\mathsf{\Omega }_{\sigma \tau }],$$
(142)
where $`dS^{\mu \nu }`$ is the exterior derivative of $`S^{\mu \nu }`$, viewed as a one-form over $`๐`$. We determine $`dS^{\mu \nu }`$ from the formula Eq. (131) for the curvature $`B^p`$. We first note
$$B_{\mathrm{๐ฑ๐ฒ}}^p=(dA^p)(๐ฑ,๐ฒ)+[A_๐ฑ^p,A_๐ฒ^p]=(dS^{\mu \nu })(๐ฑ,๐ฒ)\mathrm{\Omega }_{\mu \nu }+(S_๐ฑ)^{\mu \nu }(S_๐ฒ)^{\sigma \tau }[\mathrm{\Omega }_{\mu \nu },\mathrm{\Omega }_{\sigma \tau }],$$
(143)
where we have used Eqs. (121) and (127). It is straightforward to verify that the $`\mathrm{\Omega }_{\mu \nu }`$ satisfy the following commutation relations,
$$[\mathrm{\Omega }_{\mu \nu },\mathrm{\Omega }_{\sigma \tau }]=\frac{1}{2}(\delta _{\mu \sigma }\mathrm{\Omega }_{\tau \nu }+\delta _{\nu \tau }\mathrm{\Omega }_{\sigma \mu }+\delta _{\mu \tau }\mathrm{\Omega }_{\nu \sigma }+\delta _{\nu \sigma }\mathrm{\Omega }_{\mu \tau }),$$
(144)
and hence Eq. (143) reduces to
$$B_{\mathrm{๐ฑ๐ฒ}}^p=[(dS^{\mu \nu })(๐ฑ,๐ฒ)+(S_๐ฑS_๐ฒS_๐ฒS_๐ฑ)^{\mu \nu }]\mathrm{\Omega }_{\mu \nu }.$$
(145)
Combining this equation with Eq. (131) produces
$$(dS^{\mu \nu })(๐ฑ,๐ฒ)=(R_{\mathrm{๐ฑ๐ฒ}}T_๐ฑT_๐ฒ+T_๐ฒT_๐ฑS_๐ฑS_๐ฒ+S_๐ฒS_๐ฑ)^{\mu \nu }.$$
(146)
Combining Eq. (146) in turn with Eq. (142), we arrive at the following useful formula for the curvature of $`^{}`$,
$$๐ก_{\mathrm{๐ฑ๐ฒ}}^{}=(R_{\mathrm{๐ฑ๐ฒ}}T_๐ฑT_๐ฒ+T_๐ฒT_๐ฑS_๐ฑS_๐ฒ+S_๐ฒS_๐ฑ)^{\mu \nu }\mathsf{\Omega }_{\mu \nu }+(S_๐ฑ)^{\mu \nu }(S_๐ฒ)^{\sigma \tau }[\mathsf{\Omega }_{\mu \nu },\mathsf{\Omega }_{\sigma \tau }].$$
(147)
Using the commutation relations Eq. (144) the above equation can be recast as
$`B_{\mathrm{๐ฑ๐ฒ}}^{}`$ $`=`$ $`\left(R_{\mathrm{๐ฑ๐ฒ}}T_๐ฑT_๐ฒ+T_๐ฒT_๐ฑ\right)^{\mu \nu }\mathsf{\Omega }_{\mu \nu }`$ (149)
$`+[(S_๐ฑ)^{\mu \nu }(S_๐ฒ)^{\sigma \tau }(S_๐ฒ)^{\mu \nu }(S_๐ฑ)^{\sigma \tau }](\mathsf{\Omega }_{\mu \nu }\mathsf{\Omega }_{\sigma \tau }\mathsf{\Omega }_{\mu \nu \sigma \tau }^{(2)}),`$
where $`\mathsf{\Omega }_{\mu \nu \sigma \tau }^{(2)}`$ is the $`k\times k`$ matrix with components
$$(\mathrm{\Omega }_{\mu \nu \sigma \tau }^{(2)})_{nn^{}}=\chi _n|\mathrm{\Omega }_{\mu \nu }\mathrm{\Omega }_{\sigma \tau }\chi _n^{}.$$
(150)
## VIII Specific Cases and Examples
We consider several concrete examples to help clarify the general theory.
### A Codimension One Case
We assume here that the codimension of the constraint manifold is $`d=1`$. Since there is only one normal direction, we expect the potential twist to vanish. Indeed, this follows from the antisymmetry property $`S_{\mu \nu i}=S_{\nu \mu i}`$ (Eq. (8)) and the fact that $`\mu =\nu =1`$. Similarly, the normal components of the Riemannian curvature $`R_{\mu \nu \sigma \tau }`$ also vanish due to the well-known antisymmetry property $`R_{abcd}=R_{bacd}=R_{abdc}`$. From this fact follows $`_{}=R_{\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}11}}^{11}|_0=0`$ and $`=_{}`$. The expressions for $`๐ฏ^2`$ and $`^2`$ can also be simplified by introducing the rank two symmetric tensor $`W`$ defined on vectors tangent to $`๐`$ and with components $`W_j^i=T_{\mathrm{\hspace{0.33em}\hspace{0.33em}1}j}^i`$. (This tensor is often called the Weingarten map.) Then $`๐ฏ^2=\text{Tr}(W^2)`$ and $`^2=(\text{Tr}W)^2`$. Hence, the tangential Hamiltonian Eq. (106) becomes
$`๐ง_{}`$ $`=`$ $`๐ช_{}+๐ต_{ex},`$ (151)
$`๐ช_{}`$ $`=`$ $`{\displaystyle \frac{1}{2}}\pi ^i|_{0}^{}{}_{}{}^{}\pi _i๐จ,`$ (152)
$`๐ต_{ex}`$ $`=`$ $`V_{ex}^p๐จ={\displaystyle \frac{\mathrm{}^2}{8}}\left(๐ฏ^2\widehat{}+_{}\right)๐จ`$ (153)
$`=`$ $`{\displaystyle \frac{\mathrm{}^2}{8}}\left(^22\widehat{}+2_{}\right)๐จ={\displaystyle \frac{\mathrm{}^2}{8}}\left(2๐ฏ^2^2\right)๐จ,`$ (154)
where we have used Eqs. (67), (69), and (70). Notice that the tangential kinetic energy is proportional to the standard Laplacian on $`๐`$. All reference to $`\mathrm{\Lambda }_{\mu \nu }`$ has vanished, and hence all coupling between the degenerate transverse modes has been eliminated. The $`k`$-dimensional Schrรถdinger equation therefore separates into $`k`$ independent scalar Schrรถdinger equations.
We consider the case where the ambient space $`๐`$ is a flat two-dimensional space and the constraint manifold $`๐`$ is a curve in that space. Then, we note that $`\widehat{}==_{}=0`$. Furthermore, the second fundamental form, or equivalently the Weingarten map, has only one nonzero component. We denote this component by $`W=W_{ii}=\kappa =1/\rho `$, where $`\kappa `$ is the external curvature and $`\rho `$ is the radius of curvature. Then, the extrapotential is
$$V_{ex}^p=\frac{\mathrm{}^2}{8}\frac{1}{\rho ^2}=\frac{\mathrm{}^2}{8}\kappa ^2.$$
(155)
As in Sect. II the sign on $`V_{ex}^p`$ is such that $`\varphi `$ is attracted to regions of high curvature. This extrapotential was derived earlier by Marcus and Switkes, Russell, and Skinner .
We next consider the case where $`๐`$ is a flat three-dimensional space and $`๐`$ is a two-dimensional surface. We still have that $`=_{}=0`$. Furthermore, the eigenvalues of the second rank two-dimensional tensor $`W_{ij}`$ are $`\kappa _1=1/\rho _1`$ and $`\kappa _2=1/\rho _2`$, where $`\rho _1`$ and $`\rho _2`$ are the two external radii of curvature. Then the extrapotential $`V_{ex}^p`$ is conveniently written
$$V_{ex}^p=\frac{\mathrm{}^2}{8}\left[2\text{Tr}(W^2)(\text{Tr}W)^2\right]=\frac{\mathrm{}^2}{8}\left(\frac{1}{\rho _1}\frac{1}{\rho _2}\right)^2=\frac{\mathrm{}^2}{8}\left(\kappa _1\kappa _2\right)^2.$$
(156)
This result was previously derived by Jensen and Koppe as well as da Costa .
### B Codimension Two Case
We assume here that the codimension of the constraint manifold is $`d=2`$. This allows us to define the quantities $`S_i`$, $`\mathsf{\Lambda }`$, and $`\mathsf{\Lambda }^{(2)}`$ by
$`S_{\mu \nu i}`$ $`=`$ $`S_iฯต_{\mu \nu },`$ (157)
$`\mathsf{\Lambda }_{\mu \nu }`$ $`=`$ $`\mathsf{\Lambda }ฯต_{\mu \nu },`$ (158)
$`\mathsf{\Lambda }_{\mu \nu \sigma \tau }^{(2)}`$ $`=`$ $`\mathsf{\Lambda }^{(2)}ฯต_{\mu \nu }ฯต_{\sigma \tau },`$ (159)
where $`ฯต_{\mu \nu }`$ is the $`2\times 2`$ antisymmetric tensor with $`ฯต_{12}=ฯต_{21}=1`$. Furthermore, we have
$$R_{\mu \nu \sigma \tau }=\frac{1}{2}_{}ฯต_{\mu \nu }ฯต_{\sigma \tau }.$$
(160)
We express the tangential Hamiltonian Eq. (106) as
$`๐ง_{}`$ $`=`$ $`๐ช_{}+๐ต_{ex},`$ (161)
$`๐ช_{}`$ $`=`$ $`{\displaystyle \frac{1}{2}}(\pi ^i|_0๐จ+2S^i|_0\mathsf{\Lambda })^{}(\pi _i๐จ+2S_i\mathsf{\Lambda }),`$ (162)
$`๐ต_{ex}`$ $`=`$ $`V_{ex}^p๐จ+(2S^iS_i)|_0(\mathsf{\Lambda }^{(2)}\mathsf{\Lambda }^2)+{\displaystyle \frac{1}{3}}_{}|_0\mathsf{\Lambda }^2.`$ (163)
We consider the case of Sect. II where $`๐`$ is a flat three-dimensional space and $`๐`$ is a one-dimensional curve. First, we note $`\widehat{}==_{}=_{}=0`$. Next, we denote the single component of tangential momentum by $`\pi _{}=\pi _i`$. Since $`๐`$ is one dimensional, $`\pi _{}=i\mathrm{}/\alpha =\pi _{}^{}`$, where $`\alpha `$ is the geodesic length. Furthermore, the potential twist is determined by the sole component $`๐ฎ=S_i`$. The second fundamental form can be identified with a normal vector $`T^\mu =T^{i\mu i}`$ of magnitude $`\kappa =1/\rho `$. Hence, $`^2=T_i^{i\mu }T_{\mu j}^j=T^\mu T_\mu =\kappa ^2=1/\rho ^2`$. Using Eq. (70), Eqs. (162) and (163) therefore simplify to
$`๐ช_{}`$ $`=`$ $`{\displaystyle \frac{1}{2}}(\pi _{}๐จ+2๐ฎ\mathsf{\Lambda })^2,`$ (164)
$`๐ต_{ex}`$ $`=`$ $`{\displaystyle \frac{\mathrm{}^2}{8}}\kappa ^2๐จ+(2๐ฎ^2)(\mathsf{\Lambda }^{(2)}\mathsf{\Lambda }^2).`$ (165)
Assuming a single nondegenerate transverse mode, Eqs. (164) and (165) yield Eqs. (4) and (5).
### C Rotationally Invariant Transverse Potential
In this section, we assume the transverse potential $`V_{}(๐)`$ is rotationally invariant, depending only on the radius $`u=(u^\mu u^\mu )^{1/2}`$ in the normal space. The potential frame $`๐_\mu `$ can therefore be any orthonormal frame we like. This freedom in the choice of potential frame produces a large range of possible potential twist tensors $`S`$, with the actual choice of $`S`$ being simply a matter of convention. The Hamiltonian $`๐ง_{}`$ in Eq. (106), however, should be independent (up to a rescaling of the wave function $`\mathit{\varphi }`$) of any such conventions. In the remainder of this section, we show explicitly how the dependence on $`S`$ drops out of $`๐ง_{}`$ under the assumption of rotational invariance.
First, we observe that the transverse Hamiltonian Eq. (87) has the form
$$H_{}=\frac{\mathrm{}^2}{2}\frac{1}{u^{d1}}\frac{}{u}u^{d1}\frac{}{u}+\frac{\mathrm{\Lambda }^2}{2u^2}+V_{}(u),$$
(166)
where $`\mathrm{\Lambda }^2`$ is the Casimir operator
$$\mathrm{\Lambda }^2=\mathrm{\Lambda }_{\mu \nu }\mathrm{\Lambda }^{\mu \nu }.$$
(167)
Therefore, an eigenfunction $`\chi _n`$ of $`H_{}`$ is necessarily an eigenfunction of $`\mathrm{\Lambda }^2`$. We denote by $`\chi _n^\lambda `$ such an eigenfunction whose $`\mathrm{\Lambda }^2`$ eigenvalue is $`\lambda `$. A basic fact concerning the eigenspaces of the Casimir $`\mathrm{\Lambda }^2`$ is that they block diagonalize the generators $`\mathrm{\Lambda }_{\mu \nu }`$. That is, $`\chi _n^\lambda |\mathrm{\Lambda }_{\mu \nu }\chi _n^{}^\lambda ^{}_๐=0`$ if $`\lambda \lambda ^{}`$.<sup>*</sup><sup>*</sup>*This follows quickly from $`[\mathrm{\Lambda }^2,\mathrm{\Lambda }_{\mu \nu }]=0`$. Note, $`(\lambda \lambda ^{})\chi _n^\lambda |\mathrm{\Lambda }_{\mu \nu }\chi _n^{}^\lambda ^{}_๐=\chi _n^\lambda |[\mathrm{\Lambda }^2,\mathrm{\Lambda }_{\mu \nu }]\chi _n^{}^\lambda ^{}_๐=0`$. Based on the definitions Eqs. (110) and (111) for $`\mathsf{\Lambda }_{\mu \nu \sigma \tau }^{(2)}`$ and $`\mathsf{\Lambda }_{\mu \nu }`$, this fact implies that for the space of transverse modes for a given $`E_{}`$, $`\mathsf{\Lambda }_{\mu \nu \sigma \tau }^{(2)}=\mathsf{\Lambda }_{\mu \nu }\mathsf{\Lambda }_{\sigma \tau }`$. Similarly, $`\mathsf{\Omega }_{\mu \nu \sigma \tau }^{(2)}=\mathsf{\Omega }_{\mu \nu }\mathsf{\Omega }_{\sigma \tau }`$. We therefore see from Eq. (109) that all $`S`$ dependence drops out of $`๐ต_{ex}`$,
$$๐ต_{ex}=V_{ex}^p๐จ+\frac{1}{6}R^{\mu \nu \sigma \tau }|_0\mathsf{\Lambda }_{\mu \nu }\mathsf{\Lambda }_{\sigma \tau }.$$
(168)
Considering $`๐ช_{}`$, even though Eq. (107) is written in terms of the potential twist $`S`$, we showed in Sect. VII (specifically Eq. (139)) that $`๐ช_{}`$ can be expressed in terms of the Laplacian associated with the connection $`^{}`$. From Eq. (149) and the results above, we see that the curvature $`B^{}`$ of this connection is independent of $`S`$,
$$B_{\mathrm{๐ฑ๐ฒ}}^{}=\left(R_{\mathrm{๐ฑ๐ฒ}}T_๐ฑT_๐ฒ+T_๐ฒT_๐ฑ\right)^{\mu \nu }\mathsf{\Omega }_{\mu \nu }.$$
(169)
Now if two connections $`^{}`$ and $`_{}^{}{}_{}{}^{}`$ have the same curvature, then their associated Laplacians can only differ by a rescaling of the wave function. Hence, the Hamiltonian $`๐ง_{}`$ for different choices of the potential twist $`S`$ can at most differ by such a rescaling.
### D Harmonic Transverse Potentials
We assume that the transverse potential is quadratic in the $`u^\mu `$
$$V_{}(๐;ฯต)=\underset{\mu }{}\frac{1}{2}(\omega _\mu (ฯต))^2u^\mu u^\mu $$
(170)
and that the oscillation frequencies depend on $`ฯต`$ via $`\omega _\mu (ฯต)=\stackrel{~}{\omega }_\mu /ฯต^2`$, with $`\stackrel{~}{\omega }_\mu `$ being independent of $`ฯต`$. (For clarity, we make summation over the indices $`\mu `$, $`\nu `$, $`\sigma `$, โฆ explicit in this section.) We introduce the standard machinery of raising, lowering, and number operators for each degree of freedom,
$`a_\mu `$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2\mathrm{}}}}\left(\sqrt{\omega _\mu }u_\mu +i{\displaystyle \frac{\pi _\mu }{\sqrt{\omega _\mu }}}\right),`$ (171)
$`u^\mu `$ $`=`$ $`\sqrt{{\displaystyle \frac{\mathrm{}}{2\omega _\mu }}}\left(a_\mu +a_\mu ^{}\right),`$ (172)
$`\pi _\mu `$ $`=`$ $`i\sqrt{{\displaystyle \frac{\mathrm{}\omega _\mu }{2}}}\left(a_\mu a_\mu ^{}\right),`$ (173)
$`N_\mu `$ $`=`$ $`a_\mu ^{}a_\mu ,`$ (174)
$`[a_\mu ,a_\nu ^{}]`$ $`=`$ $`\delta _{\mu \nu }.`$ (175)
Notice that $`a_\mu `$ and $`N_\mu `$ scale as $`ฯต^0`$. The transverse Hamiltonians $`H_{}`$ and $`\stackrel{~}{H}_{}`$ have the usual form
$`H_{}(ฯต)`$ $`=`$ $`{\displaystyle \underset{\mu }{}}\mathrm{}\omega _\mu (ฯต)\left(N_\mu +{\displaystyle \frac{1}{2}}\right),`$ (176)
$`\stackrel{~}{H}_{}`$ $`=`$ $`{\displaystyle \underset{\mu }{}}\mathrm{}\stackrel{~}{\omega }_\mu \left(N_\mu +{\displaystyle \frac{1}{2}}\right),`$ (177)
and the transverse modes can therefore be labeled by the number of quanta $`n_\mu `$ in each degree of freedom $`\mu `$. We denote such a mode by $`\chi _๐`$ where $`๐=(n_1,\mathrm{},n_d)`$. Inserting Eqs. (172) and (173) into Eq. (47) yields
$$\mathrm{\Lambda }_{\mu \nu }=\frac{i\mathrm{}}{4\sqrt{\omega _\mu \omega _\nu }}\left[(\omega _\mu \omega _\nu )(a_\mu a_\nu a_\mu ^{}a_\nu ^{})+(\omega _\mu +\omega _\nu )(a_\nu ^{}a_\mu a_\mu ^{}a_\nu )\right],$$
(178)
from which one quickly sees
$$\chi _๐|\mathrm{\Lambda }_{\mu \nu }\chi _๐_๐=0.$$
(179)
A significantly more involved computation yields
$$\chi _๐|\mathrm{\Lambda }_{\mu \nu }\mathrm{\Lambda }_{\sigma \tau }\chi _๐_๐=\frac{\mathrm{}^2}{8}\left[2\left(n_\mu +\frac{1}{2}\right)\left(n_\nu +\frac{1}{2}\right)\frac{\omega _\mu ^2+\omega _\nu ^2}{\omega _\mu \omega _\nu }1\right](\delta _{\mu \tau }\delta _{\nu \sigma }\delta _{\mu \sigma }\delta _{\nu \tau }).$$
(180)
We now assume that $`\chi =\chi _๐`$ is a nondegenerate transverse mode. The tangential Hamiltonian Eq. (106) is therefore a scalar operator. Using Eqs. (179) and (180), $`H_{}`$ is
$`H_{}`$ $`=`$ $`K_{}+V_{ex},`$ (181)
$`K_{}`$ $`=`$ $`{\displaystyle \frac{1}{2}}\pi ^i|_{0}^{}{}_{}{}^{}\pi _i,`$ (182)
$`V_{ex}`$ $`=`$ $`V_{ex}^p{\displaystyle \frac{\mathrm{}^2}{8}}{\displaystyle \underset{\mu \nu }{}}\left(S^{\mu \nu i}S_{\mu \nu i}+{\displaystyle \frac{1}{3}}R_{\mu \nu }^{\mu \nu }\right)|_0\left[12\left(n_\mu +{\displaystyle \frac{1}{2}}\right)\left(n_\nu +{\displaystyle \frac{1}{2}}\right){\displaystyle \frac{\omega _\mu ^2+\omega _\nu ^2}{\omega _\mu \omega _\nu }}\right].`$ (183)
The most striking aspect of the above equations is that, due to the vanishing of $`\chi |\mathrm{\Lambda }_{\mu \nu }\chi _๐`$, the tangential kinetic energy $`K_{}`$ is proportional to the standard Laplacian on the constraint manifold. Thus, all of the effects of external curvature and potential twist are contained in the extrapotential $`V_{ex}`$.
### E Potentials with Reflection Symmetry
The vanishing of $`\chi |\mathrm{\Lambda }_{\mu \nu }\chi _๐`$ (and hence the potential twist as well) from $`K_{}`$ in Eq. (182) follows from general considerations of reflection symmetry, and therefore occurs for a large class of symmetric potentials.
Let $`๐ฐO(d)`$ be a reflection acting on the transverse coordinates $`๐=(u^1,\mathrm{},u^d)`$, and assume that, for a given $`\sigma `$, $`u^\sigma `$ is mapped to $`u^\sigma `$ and all other coordinates remain fixed. Thus, $`๐ฐ=๐ฐ^1=๐ฐ^{}`$. Furthermore, assume that $`V_{}(๐)`$ is invariant under the action of $`๐ฐ`$, that is $`V_{}(\mathrm{๐ฐ๐})=V_{}(๐)`$. The reflection $`๐ฐ`$ also has an induced action on the transverse modes, which we denote by $`Q`$ and which is given by $`(Q\chi _n)(๐)=\chi _n(๐ฐ^1๐)`$. Due to the symmetry of $`V_{}`$, $`Q`$ commutes with $`H_{}`$,
$$[Q,H_{}]=0.$$
(184)
Furthermore, the following are easily verified
$`Qu^\sigma Q^{}`$ $`=`$ $`u^\sigma ,`$ (185)
$`Q\pi _\sigma Q^{}`$ $`=`$ $`\pi _\sigma ,`$ (186)
$`Q\mathrm{\Lambda }_{\sigma \mu }Q^{}`$ $`=`$ $`\mathrm{\Lambda }_{\sigma \mu }\text{for all }\mu .`$ (187)
We now consider a single nondegenerate transverse mode, denoted simply by $`\chi `$. Due to Eq. (184), $`\chi `$ must also be an eigenfunction of $`Q`$ with eigenvalue either $`+1`$ or $`1`$ (since $`Q^2=I`$). Combining these facts with Eq. (187) and recalling $`Q=Q^{}`$, we have
$$\chi |\mathrm{\Lambda }_{\sigma \mu }\chi _๐=Q\chi |\mathrm{\Lambda }_{\sigma \mu }Q\chi _๐=\chi _n|Q\mathrm{\Lambda }_{\sigma \mu }Q^{}\chi _n_๐=\chi |\mathrm{\Lambda }_{\sigma \mu }\chi _๐,$$
(188)
and hence
$$\chi |\mathrm{\Lambda }_{\sigma \mu }\chi _๐=0\text{for all }\mu .$$
(189)
If the potential $`V_{}(๐)`$ is symmetric with respect to at least $`d1`$ such $`๐ฐ`$ reflections, possessing $`d1`$ distinct and orthogonal reflection axes $`u^\sigma `$, then $`\chi |\mathrm{\Lambda }_{\sigma \mu }\chi _๐`$ vanishes for all $`\mu ,\nu =1,\mathrm{},d`$. For such highly symmetric potentials, $`K_{}`$ is again given by Eq. (182) and the only effect of the potential twist is to be found in $`V_{ex}`$. This is the case for such common potentials as the simple harmonic oscillator, analyzed in the last section, as well as the $`d`$-dimensional square well. Note that this analysis says nothing about the off-diagonal terms of $`(\mathrm{\Lambda }_{\mu \nu })_{nn^{}}`$ for a system with degenerate transverse modes; for such systems, there may indeed be a nonvanishing gauge potential.
## IX Conclusions
We have rigorously derived the effective Hamiltonian of a constrained quantum system by considering the limit as the restoring force becomes infinite. In doing so, we have been careful to avoid unnecessary assumptions on the curvature of the ambient space, the form of the constraint manifold, and the manner of the constraining potential. This general approach yields important new terms in the effective potential $`๐ต_{ex}`$, as outlined in Sects. V E and VI, as well as a gauge potential in the tangential kinetic energy $`๐ช_{}`$, as outlined in Sects. VI and VII. Furthermore, this general approach allows our theory to be applied to several examples of physical importance. These examples include reaction paths for molecular reaction and scattering problems, twisted quantum waveguides, the double pendulum, and models of polymers by rigid constraints.
Perhaps the most important example of a constrained quantum system is the quantum rigid body. Though we lack space to include the analysis here, we have successfully applied our theory to this case. Physically, we have in mind such systems as semirigid molecules. If we assume that the standard Born-Oppenheimer ordering for semirigid molecules is valid, then our constrained Hamiltonian reproduces (through the lowest three orders in the Born-Oppenheimer ordering parameter) the standard results for the rotation-vibration energy levels of a semirigid molecule. (See, for example, Papouลกek and Aliev .) For such molecules the gauge potential term in $`K_{}`$ vanishes due to the harmonic form of the constraining potential. (We assume a nondegenerate vibrational state; see Sect. VIII D.) For this reason, a more interesting example would be one in which the standard semirigid analysis breaks down. This occurs, for example, in rigid clusters of molecules held together by van der Waals forces. For these systems, the gauge potential will not in general disappear and should have measurable effects on the rotation-vibration spectrum. We will pursue these issues in future publications.
## X Acknowledgments
The author wishes to acknowledge Jerry Marsden and Alan Weinstein, who were instrumental in the initial motivation of this problem. The author is also especially grateful to Robert Littlejohn, for many extended discussions and thoughtful insight, and to Michael Mรผller for his careful review of the manuscript. This work was supported by the Engineering Research Program of the Office of Basic Energy Sciences at the U. S. Department of Energy under Contract No. DE-AC03-76SF00098.
## A A Brief Review of Curves in $`^3`$
We cite a few important facts about curves in $`^3`$ which we need in the body of the paper. For greater depth, see, for example, Spivak . Consider a curve $`๐ฑ(\alpha )`$ in $`^3`$. The parameterization of the curve is given by $`\alpha `$ which measures the arclength along the curve. Hence the tangent vector $`\widehat{๐ญ}=d๐ฑ/d\alpha `$ is of unit length. We denote the principal normal and the binormal by $`\widehat{๐ง}`$ and $`\widehat{๐}`$, respectively. They are given by
$`\widehat{๐ง}`$ $`=`$ $`{\displaystyle \frac{d\widehat{๐ญ}/d\alpha }{|d\widehat{๐ญ}/d\alpha |}},`$ (A1)
$`\widehat{๐}`$ $`=`$ $`\widehat{๐ญ}\mathbf{\times }\widehat{๐ง}.`$ (A2)
The vectors $`(\widehat{๐ญ},\widehat{๐ง},\widehat{๐})`$ form an orthonormal right-handed frame. The derivatives of this frame are given by the famous Serret-Frenet formulas which may be summarized as
$$\frac{d}{d\alpha }\left[\begin{array}{c}\widehat{๐ญ}\\ \widehat{๐ง}\\ \widehat{๐}\end{array}\right]=\left[\begin{array}{ccc}0& \kappa & 0\\ \kappa & 0& \tau \\ 0& \tau & 0\end{array}\right]\left[\begin{array}{c}\widehat{๐ญ}\\ \widehat{๐ง}\\ \widehat{๐}\end{array}\right],$$
(A3)
where $`\kappa (\alpha )`$ and $`\tau (\alpha )`$ are called the curvature and torsion respectively. The curvature and torsion have units of reciprocal length. The reciprocal of $`\kappa `$ is the radius of curvature $`\rho =\kappa ^1`$.
## B The Second Fundamental Form
The external curvature of a submanifold $`๐`$ embedded in a manifold $`๐`$ is conveniently specified by a rank three tensor $`T`$ called the second fundamental form. Since the second fundamental form is of critical importance in the body of this paper, we briefly review a few of its relevant properties. For greater detail, see Refs. .
Throughout this appendix, $`๐`$, $`๐`$, $`๐`$ denote arbitrary vector fields tangent to $`๐`$ and defined over $`๐`$; $`๐ฐ`$, $`๐ฑ`$, $`๐ฒ`$, $`๐ณ`$ denote vector fields tangent to $`๐`$; and $`๐ฏ`$ denotes a vector field normal to $`๐`$. The second fundamental form applied to $`๐`$ and $`๐`$, denoted $`T_๐๐`$, is a vector field defined by
$$T_๐๐=P_{}_{P_{}๐}P_{}๐+P_{}_{P_{}๐}P_{}๐,$$
(B1)
where $``$ is the Levi-Civita connection on $`๐`$ and $`P_{}`$ and $`P_{}`$ are respectively the tangent and normal projection operators of $`๐`$ Our definition of the second fundamental form differs in the choice of domain and range from that in Ref. . We follow the definition of Ref. .. It is straightforward to verify that $`T`$ is in fact a tensor. Furthermore, the second fundamental form satisfies the identities
$`๐,T_๐๐`$ $`=`$ $`๐,T_๐๐,`$ (B2)
$`T_๐ฑ๐ฒ`$ $`=`$ $`T_๐ฒ๐ฑ,`$ (B3)
where $`,`$ denotes the Riemannian metric on $`๐`$. In terms of the components $`T_{abc}=๐_a,T_{๐_c}๐_b`$ introduced in Sect. V C, we have
$`T_{abc}`$ $`=`$ $`T_{bac},`$ (B4)
$`T_{aij}`$ $`=`$ $`T_{aji},`$ (B5)
$`T_{ab\mu }`$ $`=`$ $`0,`$ (B6)
$`T_{\mu \nu a}`$ $`=`$ $`T_{ija}=0,`$ (B7)
where the first two equations are simply component forms for Eqs. (B2) and (B3) and the last two follow easily from Eq. (B1).
In Sects. V D and V E, we need the Gauss equation, a well-known identity relating the second fundamental form $`T`$, the Riemannian curvature $`\widehat{R}`$ of $`๐`$, and the Riemannian curvature $`R`$ of $`๐`$. The Riemannian curvatures are defined by
$`R_{\mathrm{๐๐}}๐`$ $`=`$ $`(_๐_๐_๐_๐_{[๐,๐]})๐,`$ (B8)
$`\widehat{R}_{\mathrm{๐ฑ๐ฒ}}๐ณ`$ $`=`$ $`(\widehat{}_๐ฑ\widehat{}_๐ฒ\widehat{}_๐ฒ\widehat{}_๐ฑ\widehat{}_{[๐ฑ,๐ฒ]})๐ณ,`$ (B9)
where $`\widehat{}`$ denotes the Levi-Civita connection on $`๐`$. The Gauss equation is then
$$๐ฐ,R_{\mathrm{๐ฑ๐ฒ}}๐ณ=๐ฐ,\widehat{R}_{\mathrm{๐ฑ๐ฒ}}๐ณ+T_๐ฑ๐ณ,T_๐ฒ๐ฐT_๐ฒ๐ณ,T_๐ฑ๐ฐ.$$
(B10)
## C The Quantum Kinetic Energy with Respect to a Vielbein
We present two expressions for the kinetic energy of a quantum system on a Riemannian manifold of dimension $`n`$. These expressions differ in the scaling of the quantum wave function. We refer the reader to earlier related analyses for derivations and discussion.
We express the kinetic energy in terms of a vielbein. By a vielbein on a Riemannian manifold, we mean a set of vector fields $`๐_a`$, $`a=1,\mathrm{},n`$, forming a basis of each tangent space. The structure constants $`\beta _{ab}^c`$ of the vielbein are defined by
$$[๐_a,๐_b]=\beta _{ab}^c๐_c,$$
(C1)
where $`[,]`$ denotes the Lie bracket. The structure constants vanish if and only if the vielbein is a coordinate basis, that is if and only if there exists a set of coordinates $`x^a`$ such that $`๐_a=/x^a`$. We denote the components of the Riemannian metric with respect to the vielbein by $`G_{ab}`$ and the inverse of $`G_{ab}`$ by $`G^{ab}`$.
We define the kinetic energy of the quantum system by $`K=\mathrm{}^2\mathrm{}/2`$, where $`\mathrm{}`$ is the Laplacian. In terms of the vielbein, the kinetic energy is ,
$$K=\frac{\mathrm{}^2}{2}๐_a^{}G^{ab}๐_b=\frac{1}{2}\pi _a^{}G^{ab}\pi _b,$$
(C2)
where
$$\pi _a=i\mathrm{}๐_a,$$
(C3)
are the momentum operators. In the above $``$ denotes the Hermitian conjugate. In general, the momenta $`\pi _a`$ are not Hermitian. They do, however, satisfy the following useful identity
$$\pi _a^{}=\pi _a+\left[\pi _a\mathrm{ln}\sqrt{G}\right]+i\mathrm{}\beta _{ab}^b,$$
(C4)
where $`G=detG_{ab}`$. The bracket notation in Eq. (C4) indicates that the quantity inside the brackets is a scalar; that is, $`\pi _a`$ acts only on $`\mathrm{ln}\sqrt{G}`$.
Often it is useful to scale the original wave function $`\phi `$ by some real positive function $`s`$ to form a new wave function $`\psi `$,
$$\psi =s\phi .$$
(C5)
Such a scaling produces a new kinetic energy operator acting on the new wave function $`\psi `$. By conveniently choosing the scale factor $`s`$, the new kinetic energy may acquire a more convenient form than the old kinetic energy. To demonstrate how the kinetic energy transforms, we first observe that the scaled wave functions have a different inner product than the unscaled wave functions. Denoting the unscaled inner product by $`|`$, the scaled inner product $`|_s`$ is defined by
$$\psi |\psi ^{}_s=\frac{1}{s}\psi |\frac{1}{s}\psi ^{},$$
(C6)
for arbitrary wave functions $`\psi `$ and $`\psi ^{}`$. This scaled inner product in turn defines a scaled Hermitian conjugate $`A^{(s)}`$ of an operator $`A`$. Specifically,
$$A^{(s)}=s^2A^{}\frac{1}{s^2}.$$
(C7)
Applying Eq. (C7) to Eq. (C4), we find
$$\pi _a^{(s)}=\pi _a^{}2[\pi _a\mathrm{ln}s].$$
(C8)
The scaling of the wave function transforms the kinetic energy operator $`K`$ into $`K_s=sK(1/s)`$. It can be shown that $`K_s`$ reduces to
$$K_s=\frac{1}{2}\pi _a^{(s)}G^{ab}\pi _b+V_s,$$
(C9)
where
$`V_s`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left(G^{ab}[\pi _a\mathrm{ln}s][\pi _b\mathrm{ln}s]+[\pi _a^{(s)}G^{ab}[\pi _b\mathrm{ln}s]]\right)`$ (C10)
$`=`$ $`{\displaystyle \frac{1}{2}}\left(G^{ab}[\pi _a\mathrm{ln}s][\pi _b\mathrm{ln}s][\pi _a^{}G^{ab}[\pi _b\mathrm{ln}s]]\right).`$ (C11) |
warning/0001/cond-mat0001041.html | ar5iv | text | # Evidence for a transition from a bulk Meissner-state to a spontaneous vortex phase in RuSr2GdCu2O8 from DC magnetisation measurements
## I <br>Figure Captions
Figure 1: Zero-field-cooled (zfc) volume susceptibility, $`\chi _{V,}`$ at 6.5 Oe of the pure sample A (solid line) and the Zn-substituted sample C (dotted line). Inset: Susceptibility of sample A around the SC transition, T$`{}_{c}{}^{}=45`$ K, shown on an enlarged scale.
Figure 2: Field-cooled (fc) volume susceptibility, $`\chi _V,`$ (a) of the pure sample A at 2.5, 6.5, 10, 20, 35, 50, 100 and 500 Oe (solid lines) and the Zn-substituted sample C at 2.5 and 100 Oe (dotted lines); (b) of sample A at 2.5, 1.5, 0.75 and 0.5 Oe (solid lines) and sample C at 0.5 Oe (dotted line).
Figure 3: Low temperature fc curve at 6.5 Oe for the pure sample B which has the same T$`{}_{c}{}^{}=45K`$ as sample A but has been prepared under slightly different conditions as noted in the text. The Meissner-phase forms at significantly lower temperature T$`{}_{}{}^{\mathrm{๐๐ }}`$16 K. Note the thermal hysteresis of $`\chi _V`$ around T<sup>ms</sup> which is absent if the sample is only cooled to T=17 K (crosses) and subsequently warmed (open circles).
Figure 4: Thermal hysteresis of the fc data of sample A at 35, 50, 100, 250 and 500 Oe. The solid lines (dotted lines) show $`\chi _V`$ upon cooling (warming). Arrows indicate the direction of the temperature change. At 100 Oe two hysteresis curves for cooling to 2 and 4 K are shown by the thick and thin dotted lines, respectively. |
warning/0001/hep-ph0001146.html | ar5iv | text | # I Introduction
## I Introduction
Recently charge symmetry violation (CSV) in the parton distributions of the nucleon has attracted great interest. It has been generally believed that charge symmetry (CS) was highly respected in the nucleon system. Most low energy experiments have shown that CS is satisfied within about $`1\%`$ in reaction amplitudes , and most high energy tests are also consistent with the charge symmetry, although generally with less precision than the low energy tests (for a recent review see and references therein). However some theoretical calculations have suggested that the CSV in the valence quark distributions may be as large as $`5\%10\%`$ which is rather large compared with the low-energy results. There have been proposed a number of experiments in which CSV may be observed .
A serious challenge to CS has come from the comparison of the $`F_2`$ structure functions measured in charged and uncharged lepton deep inelastic scattering performed by Boros, Londergan and Thomas . A significantly larger CSV than the expectations of both theory and other experiments was found for the $`s`$ and $`\overline{s}`$ distributions in the low-$`x`$ region ($`x<0.1`$). Any unexpected large CSV will greatly affect our understanding of non-perturbative dynamics and hadronic structure , and also the extraction of $`sin^2\theta _W`$ from neutrino scattering . More recently, Boros, Steffens, Londergan and Thomas performed a similar analysis to Ref. with improved corrections for nuclear shadowing and the charm threshold in the neutrino data. They found that the data (including the low-$`x`$ region where a large discrepancy was found in ) are consistent with charge symmetry within experimental errors and the present uncertainty in the strange quark distribution of the nucleon.
Thus suggestions of any large CSV in the parton distributions of the nucleon result from theoretical calculations . Most of these theoretical calculations are performed using a quark model, such as the MIT bag model , Los Alamos Potential Model , or a model independent version . The quark model calculations are based on a quark-spectator (quark-diquark) picture of the nucleon in deep inelastic reactions, which is questionable in the low-$`x`$ region. Hence the quark model predictions for the low-$`x`$ behaviour of CSV are not very reliable. For the sea quark content, quark model calculations involve spectator states containing four quarks, which is also unreliable in the low-$`x`$ region as the mass parameter for these four-quark (or three-quark one-antiquark) states is less well-determined than that for the diquark states. CSV has also been estimated using the light-cone baryon-meson fluctuation model . However, this calculation is mostly qualitative โ the quantitative calculations are highly dependant on the model parameters. More theoretical study on CSV from a different point of view than the quark model will be worthwhile.
In this paper, we point out that the CSV in both the valence quark and sea quark has the same non-perturbative source, and that the meson cloud model (MCM) โ which has been successfully applied to the study of the sea quark content of nucleon (including SU(2) flavour asymmetry , and $`s`$ and $`\overline{s}`$ content of the nucleon ) โ can provide a natural explanation of CSV in the valence quark and sea quark distributions of the nucleon. We shall make an alternative investigation of CSV in the parton distributions of the nucleon by using the meson cloud model instead of the more commonly used quark model. Our calculations for the CSV are significantly different from the quark model predictions.
## II Charge symmetry violation in the meson cloud model
Charge symmetry results from the $`180^o`$-rotation invariance of the strong Hamiltonian about the $`2`$-axis in isospace . At the quark level charge symmetry implies the invariance of a system under the interchange of up and down quarks. For the valence and sea parton distributions, this results in the following relations:
$`u_v^p(x)=d_v^n(x),`$ $`d_v^p(x)=u_v^n(x),`$ (1)
$`\overline{u}^p(x)=\overline{d}^n(x),`$ $`\overline{d}^p(x)=\overline{u}^n(x),`$ (2)
$`s^p(x)=s^n(x),`$ $`\overline{s}^p(x)=\overline{s}^n(x).`$ (3)
The charge symmetry violation in the parton distributions of the nucleon can be โmeasuredโ via the quantities:
$`\delta d_v(x)=d_v^p(x)u_v^n(x),`$ $`\delta u_v(x)=u_v^p(x)d_v^n(x),`$ (4)
$`\delta \overline{d}(x)=\overline{d}^p(x)\overline{u}^n(x),`$ $`\delta \overline{u}(x)=\overline{u}^p(x)\overline{d}^n(x),`$ (5)
$`\delta s(x)=s^p(x)s^n(x),`$ $`\delta \overline{s}(x)=\overline{s}^p(x)\overline{s}^n(x).`$ (6)
Before entering detailed calculation, it is helpful to break down the parton distribution in the nucleon into three parts: the parton distribution in the bare nucleon, the perturbative contribution, and the non-perturbative contribution i.e.
$`d^p=d_{\mathrm{bare}}^p+d_{\mathrm{per}}^p+d_{\mathrm{non}}^p,`$ $`u^p=u_{\mathrm{bare}}^p+u_{\mathrm{per}}^p+u_{\mathrm{non}}^p,`$ (7)
$`\overline{d}^p=\overline{d}_{\mathrm{per}}^p+\overline{d}_{\mathrm{non}}^p,`$ $`\overline{u}^p=\overline{u}_{\mathrm{per}}^p+\overline{u}_{\mathrm{non}}^p,`$ (8)
$`s^p=s_{\mathrm{per}}^p+s_{\mathrm{non}}^p,`$ $`\overline{s}^p=\overline{s}_{\mathrm{per}}^p+\overline{s}_{\mathrm{non}}^p.`$ (9)
Similar relations exist for the parton distribution of the neutron. We expect that the bare part obeys the charge symmetry
$`d_{\mathrm{bare}}^p=u_{\mathrm{bare}}^n,u_{\mathrm{bare}}^p=d_{\mathrm{bare}}^n.`$ (10)
The perturbative sea is produced in a very short time via gluon splitting, thus we expect the perturbative sea to also be SU(2) flavour symmetric,
$`\overline{d}_{\mathrm{per}}^p=\overline{u}_{\mathrm{per}}^p,\overline{u}_{\mathrm{per}}^p=\overline{d}_{\mathrm{per}}^n,`$ (11)
quark-antiquark symmetric,
$`q_{\mathrm{per}}^{p,n}=\overline{q}_{\mathrm{per}}^{p,n},(q=u,d,s)`$ (12)
and charge symmetric,
$`d_{\mathrm{per}}^p=u_{\mathrm{per}}^n,`$ $`u_{\mathrm{per}}^p=d_{\mathrm{per}}^n,`$ (13)
$`\overline{d}_{\mathrm{per}}^p=\overline{u}_{\mathrm{per}}^n,`$ $`\overline{u}_{\mathrm{per}}^p=\overline{d}_{\mathrm{per}}^n,`$ (14)
$`s_{\mathrm{per}}^p=s_{\mathrm{per}}^n,`$ $`\overline{s}_{\mathrm{per}}^p=\overline{s}_{\mathrm{per}}^n.`$ (15)
From the bare parton distribution being charge symmetric \[Eq. (10)\] and the perturbative sea being quark-antiquark symmetric \[Eq. (12)\] and charge symmetric \[Eqs. (13) and (14)\] we have the CS violating valence distributions
$`\delta d_v`$ $`=`$ $`(d_{\mathrm{non}}^p\overline{d}_{\mathrm{non}}^p)(u_{\mathrm{non}}^n\overline{u}_{\mathrm{non}}^n),`$ (16)
$`\delta u_v`$ $`=`$ $`(u_{\mathrm{non}}^p\overline{u}_{\mathrm{non}}^p)(d_{\mathrm{non}}^n\overline{d}_{\mathrm{non}}^n).`$ (17)
Using the charge symmetry of the perturbative sea \[Eqs. (13)-(15)\] we can obtain the CS violating sea distributions
$`\delta \overline{d}=\overline{d}_{\mathrm{non}}^p\overline{u}_{\mathrm{non}}^n,`$ $`\delta \overline{u}=\overline{u}_{\mathrm{non}}^p\overline{d}_{\mathrm{non}}^n,`$ (18)
$`\delta s=s_{\mathrm{non}}^ps_{\mathrm{non}}^n,`$ $`\delta \overline{s}=\overline{s}_{\mathrm{non}}^p\overline{s}_{\mathrm{non}}^n.`$ (19)
Thus the charge symmetry violation in both the valence and the sea distributions has a non-perturbative origin.
The meson cloud model (MCM) is a model of the non-perturbative contribution to the quark distributions of the nucleon. It can provide natural explanations of the flavour asymmetry in the nucleon sea and quark-antiquark asymmetry in the nucleon. The essential point of the MCM is that the nucleon can fluctuate into different baryon-meson Fock states,
$`|N_{\mathrm{physical}}=Z|N_{\mathrm{bare}}+{\displaystyle \underset{BM}{}}{\displaystyle \underset{\lambda \lambda ^{}}{}}{\displaystyle ๐yd^2๐ค_{}\varphi _{BM}^{\lambda \lambda ^{}}(y,k_{}^2)|B^\lambda (y,๐ค_{});M^\lambda ^{}(1y,๐ค_{})}`$ (20)
where $`Z`$ is the wave function renormalization constant, $`\varphi _{BM}^{\lambda \lambda ^{}}(y,k_{}^2)`$ is the wave function of the Fock state containing a baryon ($`B`$) with longitudinal momentum fraction $`y`$, transverse momentum $`๐ค_{}`$, and helicity $`\lambda `$, and a meson ($`M`$) with momentum fraction $`1y`$, transverse momentum $`๐ค_{}`$, and helicity $`\lambda ^{}`$. The model assumes that the lifetime of a virtual baryon-meson Fock state is much larger than the interaction time in the deep inelastic or Drell-Yan process, thus the contribution from the virtual baryon-meson Fock states to the quark and anti-quark distributions of the nucleon can be written as convolutions
$`q_{\mathrm{non}}(x)`$ $`=`$ $`{\displaystyle \underset{BM}{}}\left[{\displaystyle _x^1}{\displaystyle \frac{dy}{y}}f_{BM/N}(y)q^B({\displaystyle \frac{x}{y}})+{\displaystyle _x^1}{\displaystyle \frac{dy}{y}}f_{MB/N}(1y)q^M({\displaystyle \frac{x}{y}})\right],`$ (21)
$`\overline{q}_{\mathrm{non}}(x)`$ $`=`$ $`{\displaystyle \underset{BM}{}}{\displaystyle _x^1}{\displaystyle \frac{dy}{y}}f_{MB/N}(1y)\overline{q}^M({\displaystyle \frac{x}{y}}),`$ (22)
where $`f_{BM/N}(y)=f_{MB/N}(1y)`$ is fluctuation function which gives the probability for the nucleon fluctuating into a virtual $`BM`$ state
$`f_{BM/N}(y)={\displaystyle _0^{\mathrm{}}}๐k_{}^2\left|\varphi _{BM}(y,k_{}^2)\right|^2.`$ (23)
As the proton and neutron form an SU(2) isospin doublet, the baryons and mesons in their respective virtual Fock states differ only in their carried charge. If we neglect the mass differences among these baryons and mesons the fluctuation functions for the proton and neutron will be the same. Thus the contributions to the parton distribution of the nucleon from these fluctuations will be isospin symmetric (charge symmetric). As is well known, the electromagnetic interaction induces mass differences among these baryons and mesons. If we take into account these mass differences, the probabilities for the corresponding fluctuations of proton and neutron will be different and thus the contributions to the parton distributions of the proton and neutron will be different, which results in CSV in the parton distributions of the nucleon. Thus the MCM can provide a natural explanation of CSV in the parton distributions of the nucleon. Although it is argued from the quark model calculations that the electromagnetic effect does not play a significant role in the calculation of CSV in the parton distributions, it is worthwhile to study this effect using a different theoretical picture.
For the CSV in the up and down quark distributions, we consider the fluctuations $`NN\pi `$ and $`N\mathrm{\Delta }\pi `$, but neglect the other fluctuations such as $`NN(\mathrm{\Delta })\rho `$, $`NN\eta (\omega )`$ and $`N\mathrm{\Delta }\eta (\omega )`$ due to the higher masses of the involved mesons. Thus the fluctuations we consider include:
$`p(uud)n(udd)+\pi ^+(u\overline{d}),`$ $`n(udd)p(uud)+\pi ^{}(\overline{u}d),`$ (24)
$`p(uud)\mathrm{\Delta }^0(udd)+\pi ^+(u\overline{d}),`$ $`n(udd)\mathrm{\Delta }^+(uud)+\pi ^{}(\overline{u}d),`$ (25)
$`p(uud)p(uud)+\pi ^0([u\overline{u}+d\overline{d}]/\sqrt{2}),`$ $`n(udd)n(udd)+\pi ^0([u\overline{u}+d\overline{d}]/\sqrt{2}),`$ (26)
$`p(uud)\mathrm{\Delta }^+(uud)+\pi ^0([u\overline{u}+d\overline{d}]/\sqrt{2}),`$ $`n(udd)\mathrm{\Delta }^0(udd)+\pi ^0([u\overline{u}+d\overline{d}]/\sqrt{2}),`$ (27)
$`p(uud)\mathrm{\Delta }^{++}(uuu)+\pi ^{}(\overline{u}d),`$ $`n(udd)\mathrm{\Delta }^{}(ddd)+\pi ^+(u\overline{d}).`$ (28)
For the CSV in the strange and anti-strange quark distributions we consider the following fluctuations
$`p(uud)\mathrm{\Lambda }(uds)+K^+(u\overline{s}),`$ $`n(udd)\mathrm{\Lambda }(uds)+K^0(d\overline{s}),`$ (29)
$`p(uud)\mathrm{\Sigma }^0(uds)+K^+(u\overline{s}),`$ $`n(udd)\mathrm{\Sigma }^0(uds)+K^0(d\overline{s}),`$ (30)
$`p(uud)\mathrm{\Sigma }^+(uus)+K^0(d\overline{s}),`$ $`n(udd)\mathrm{\Sigma }^{}(dds)+K^+(u\overline{s}).`$ (31)
Since we consider the CSV between the proton and neutron, we should not neglect the mass difference between the proton and neutron, $`m_pm_n=1.3\mathrm{MeV}`$, and those among the baryon and meson multiplets ,
$`m_\mathrm{\Delta }^{}m_{\mathrm{\Delta }^0}`$ $`=`$ $`m_{\mathrm{\Delta }^0}m_{\mathrm{\Delta }^+}=m_{\mathrm{\Delta }^+}m_{\mathrm{\Delta }^{++}}=1.3\mathrm{MeV},`$ (32)
$`m_\mathrm{\Sigma }^{}m_{\mathrm{\Sigma }^0}`$ $`=`$ $`4.8\mathrm{MeV},`$ (33)
$`m_{\mathrm{\Sigma }^0}m_{\mathrm{\Sigma }^+}`$ $`=`$ $`3.3\mathrm{MeV},`$ (34)
$`m_{\pi ^\pm }m_{\pi ^0}`$ $`=`$ $`4.6\mathrm{MeV},`$ (35)
$`m_{K^0}m_{K^+}`$ $`=`$ $`4.0\mathrm{MeV}.`$ (36)
The probabilities of various fluctuations can be calculated using the effective Lagrangian and time-ordered perturbation theory in the infinite momentum frame. For the fluctuations $`NN\pi `$, $`N\mathrm{\Lambda }K`$ and $`N\mathrm{\Sigma }K`$, the fluctuation functions can be expressed as
$`f_{BM/N}(y)=C_1{\displaystyle \frac{g_{NBM}^2}{16\pi ^2}}{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{dk_{}^2}{y(1y)}}{\displaystyle \frac{G_{BM}^2(M_{BM}^2)}{(m_N^2m_{BM}^2)^2}}{\displaystyle \frac{k_{}^2+(ym_Nm_B)^2}{y}},`$ (37)
where $`y`$ is the longitudinal momentum fraction of the baryon $`B`$, $`g_{NBM}`$ is the effective coupling constant, $`m_{BM}^2`$ is the invariant mass squared of the $`BM`$ Fock state,
$`m_{BM}^2={\displaystyle \frac{m_B^2+k_{}^2}{y}}+{\displaystyle \frac{m_\pi ^2+k_{}^2}{1y}},`$ (38)
and $`G_{BM}`$ is the phenomenological form factor, for which we adopt the exponential form
$`G_{BM}(y,k_{}^2)=\mathrm{exp}\left[{\displaystyle \frac{m_N^2m_{BM}^2(y,k_{}^2)}{2\mathrm{\Lambda }^2}}\right].`$ (39)
$`\mathrm{\Lambda }`$ is a cut-off parameter which can be taken as $`\mathrm{\Lambda }=1.08\mathrm{GeV}`$ for all fluctuations involving octet baryons and pseudoscalar or vector mesons . The effective coupling constants are taken to be $`g_{NN\pi }=13.07`$ , $`g_{N\mathrm{\Lambda }K}=13.12`$ and $`g_{N\mathrm{\Sigma }K}=6.82`$ . The coefficient $`C_1`$ in Eq. (37) comes from the Clebsch-Gordan coefficients for the fluctuations of different isospin multiplets. $`C_1=1`$ for $`f_{p\pi ^0/p}`$, $`f_{n\pi ^0/n}`$, $`f_{\mathrm{\Lambda }K^+/p}`$, $`f_{\mathrm{\Lambda }K^0/n}`$, $`f_{\mathrm{\Sigma }^0K^+/p}`$ and $`f_{\mathrm{\Sigma }^{}K^0/n}`$, and $`C_1=2`$ for $`f_{n\pi ^+/p}`$, $`f_{p\pi ^{}/n}`$, $`f_{\mathrm{\Sigma }^+K^0/p}`$, and $`f_{\mathrm{\Sigma }^{}K^+/n}`$. For the fluctuation $`N\mathrm{\Delta }\pi `$ we have
$`f_{\mathrm{\Delta }\pi /N}(y)=C_2{\displaystyle \frac{g^2}{16\pi ^2}}{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{dk_{}^2}{y(1y)}}{\displaystyle \frac{G_{\mathrm{\Delta }\pi }^2(M_{\mathrm{\Delta }\pi }^2)}{(m_N^2m_{\mathrm{\Delta }\pi }^2)^2}}{\displaystyle \frac{[k_{}^2+(m_\mathrm{\Delta }ym_N)^2][k_{}^2+(m_\mathrm{\Delta }+ym_N)^2]^2}{6m_\mathrm{\Delta }^2y^3}},`$ (40)
where $`g=11.8\mathrm{GeV}^1`$ , $`C_2=1`$ for $`f_{\mathrm{\Delta }^{++}\pi ^{}/p}`$ and $`f_{\mathrm{\Delta }^{}\pi ^+/n}`$, $`C_2=2/3`$ for $`f_{\mathrm{\Delta }^+\pi ^0/p}`$ and $`f_{\mathrm{\Delta }^0\pi ^0/n}`$, and $`C_2=1/3`$ for $`f_{\mathrm{\Delta }^0\pi ^+/p}`$ and $`f_{\mathrm{\Delta }^+\pi ^{}/p}`$. We adopt the exponential form \[Eq. (39)\] for the form factor and the cut-off parameter was taken to be $`\mathrm{\Lambda }=0.98\mathrm{GeV}`$ .
In the meson cloud model the non-perturbative contribution to the quark and the anti-quark distributions in the nucleon sea come from the quarks and anti-quarks of the baryons ($`N,\mathrm{\Lambda },\mathrm{\Sigma }`$) and mesons ($`\pi ,K`$) in the virtual baryon-meson Fock states. So we need the parton distributions in the involved baryons and mesons as input. For the parton distribution in the pion, we employ the parameterization given by Glรผck, Reya, and Stratmann (GRS98) and we neglect the sea content in the meson, that is,
$`\overline{d}^{\pi ^+}=u^{\pi ^+}`$ $`=`$ $`\overline{u}^\pi ^{}=d^\pi ^{}={\displaystyle \frac{1}{2}}v^\pi ,`$ (41)
$`\overline{u}^{\pi ^0}=u^{\pi ^0}`$ $`=`$ $`\overline{d}^{\pi ^0}=d^{\pi ^0}={\displaystyle \frac{1}{4}}v^\pi ,`$ (42)
$`v^\pi (x,\mu _{\mathrm{NLO}}^2)`$ $`=`$ $`1.052x^{0.495}(1+0.357\sqrt{x})(1x)^{0.365},`$ (43)
at scale $`\mu _{\mathrm{NLO}}^2=0.34`$ GeV<sup>2</sup>. For the $`\overline{s}`$ distribution in the $`K^+`$ and $`K^0`$ we use the GRS98 parameterization
$`\overline{s}^{K^+}(x,\mu _{\mathrm{NLO}}^2)=\overline{s}^{K^0}(x,\mu _{\mathrm{NLO}}^2)=\left[10.540(1x)^{0.17}\right]v^\pi (x,\mu _{\mathrm{NLO}}^2)`$ (44)
at scale $`\mu _{\mathrm{NLO}}^2=0.34`$ GeV<sup>2</sup>. For the quark distributions in the bare baryons, we first use the up and down quark distributions in the proton given by Glรผck, Reya, and Vogt (GRV98) ,
$`d^p(x,\mu _{\mathrm{NLO}}^2)`$ $`=`$ $`0.624(1x)u^p(x,\mu _{\mathrm{NLO}}^2)`$ (45)
$`u^p(x,\mu _{\mathrm{NLO}}^2)`$ $`=`$ $`0.632x^{0.57}(1x)^{3.09}(1+18.2x),`$ (46)
at scale $`\mu _{\mathrm{NLO}}^2=0.40`$ GeV<sup>2</sup>, then relate these to the distributions in the other baryons via the relations
$`d^n=u^{\mathrm{\Delta }^+}=d^{\mathrm{\Delta }^0}=u^p,`$ $`u^n=d^{\mathrm{\Delta }^+}=u^{\mathrm{\Delta }^0}=d^p,`$ (47)
$`u^{\mathrm{\Delta }^{++}}=u^p+d^p,`$ $`d^\mathrm{\Delta }^{}=u^p+d^p,`$ (48)
$`s^\mathrm{\Lambda }=s^{\mathrm{\Sigma }^+}=s^{\mathrm{\Sigma }^0}=s^\mathrm{\Sigma }^{}`$ $`=`$ $`{\displaystyle \frac{1}{2}}u^p.`$ (49)
We evolve the distributions to the scale $`Q^2=4`$ GeV<sup>2</sup> using the program of Miyama and Kumano in which the evolution equation is solved numerically using a brute-force method. We found that at $`Q^2=4`$ GeV<sup>2</sup> the parton distributions we need ($`v^\pi (x,Q^2)`$, $`\overline{s}^{\overline{K}^0}(x,Q^2)`$ $`u^p(x,Q^2)`$ and $`d^p(x,Q^2)`$) can be parametrized using the following form
$`q(x,Q^2)=ax^b(1x)^c(1+d\sqrt{x}+ex)`$ (50)
with the parameters given in Table 1. We estimate the uncertainty in solving the evolution equations numerically and parametrizating the parton distribution in the form of Eq. (50) to be about $`2\%`$ in the $`x`$-region which we are interested in ie $`x>10^3`$.
The final expressions for the CSV in the valence parton distributions are given by:
$`x\delta d_v`$ $`=`$ $`{\displaystyle _0^x}dy{\displaystyle \frac{x}{y}}\{[f_{n\pi ^+/p}(y)f_{p\pi ^{}/n}(y)]u^p({\displaystyle \frac{x}{y}})`$ (57)
$`\left[f_{n\pi ^+/p}(1y)f_{p\pi ^{}/n}(1y)\right]{\displaystyle \frac{1}{2}}v^\pi ({\displaystyle \frac{x}{y}})`$
$`+\left[f_{p\pi ^0/p}(y)f_{n\pi ^0/n}(y)\right]d^p({\displaystyle \frac{x}{y}})`$
$`+\left[f_{\mathrm{\Delta }^0\pi ^+/p}(y)f_{\mathrm{\Delta }^+\pi ^{}/n}(y)\right]u^p({\displaystyle \frac{x}{y}})`$
$`\left[f_{\mathrm{\Delta }^0\pi ^+/p}(1y)f_{\mathrm{\Delta }^+\pi ^{}/n}(1y)\right]{\displaystyle \frac{1}{2}}v^\pi ({\displaystyle \frac{x}{y}})`$
$`+\left[f_{\mathrm{\Delta }^+\pi ^0/p}(y)f_{\mathrm{\Delta }^0\pi ^0/n}(y)\right]d^p({\displaystyle \frac{x}{y}})`$
$`+[f_{\mathrm{\Delta }^{++}\pi ^{}/p}(1y)f_{\mathrm{\Delta }^{}\pi ^+/n}(1y)]{\displaystyle \frac{1}{2}}v^\pi ({\displaystyle \frac{x}{y}})\},`$
$`x\delta u_v`$ $`=`$ $`{\displaystyle _0^x}dy{\displaystyle \frac{x}{y}}\{[f_{n\pi ^+/p}(y)f_{p\pi ^{}/n}(y)]d^p({\displaystyle \frac{x}{y}})`$ (65)
$`\left[f_{n\pi ^+/p}(1y)f_{p\pi ^{}/n}(1y)\right]{\displaystyle \frac{1}{2}}v^\pi ({\displaystyle \frac{x}{y}})`$
$`+\left[f_{p\pi ^0/p}(y)f_{n\pi ^0/n}(y)\right]u^p({\displaystyle \frac{x}{y}})`$
$`+\left[f_{\mathrm{\Delta }^0\pi ^+/p}(y)f_{\mathrm{\Delta }^+\pi ^{}/n}(y)\right]d^p({\displaystyle \frac{x}{y}})`$
$`\left[f_{\mathrm{\Delta }^0\pi ^+/p}(1y)f_{\mathrm{\Delta }^+\pi ^{}/n}(1y)\right]{\displaystyle \frac{1}{2}}v^\pi ({\displaystyle \frac{x}{y}})`$
$`+\left[f_{\mathrm{\Delta }^+\pi ^0/p}(y)f_{\mathrm{\Delta }^0\pi ^0/n}(y)\right]u^p({\displaystyle \frac{x}{y}})`$
$`+\left[f_{\mathrm{\Delta }^{++}\pi ^{}/p}(y)f_{\mathrm{\Delta }^{}\pi ^+/n}(y)\right](u^p+d^p)`$
$`+[f_{\mathrm{\Delta }^{++}\pi ^{}/p}(1y)f_{\mathrm{\Delta }^{}\pi ^+/n}(1y)]{\displaystyle \frac{1}{2}}v^\pi ({\displaystyle \frac{x}{y}})\}.`$
For CSV in the sea we obtain
$`x\delta \overline{d}`$ $`=`$ $`{\displaystyle _0^x}dy{\displaystyle \frac{x}{y}}\{f_{n\pi ^+/p}(1y)f_{p\pi ^{}/n}(1y)+{\displaystyle \frac{1}{2}}[f_{p\pi ^0/p}(1y)f_{n\pi ^0/n}(1y)]`$ (68)
$`+f_{\mathrm{\Delta }^0\pi ^+/p}(1y)f_{\mathrm{\Delta }^+\pi ^{}/n}(1y)`$
$`+{\displaystyle \frac{1}{2}}[f_{\mathrm{\Delta }^+\pi ^0/p}(1y)f_{\mathrm{\Delta }^0\pi ^0/n}(1y)]\}{\displaystyle \frac{1}{2}}v^\pi ({\displaystyle \frac{x}{y}}),`$
$`x\delta \overline{u}`$ $`=`$ $`{\displaystyle _0^x}dy{\displaystyle \frac{x}{y}}\{{\displaystyle \frac{1}{2}}[f_{p\pi ^0/p}(1y)f_{n\pi ^0/n}(1y)]+{\displaystyle \frac{1}{2}}[f_{\mathrm{\Delta }^+\pi ^0/p}(1y)f_{\mathrm{\Delta }^0\pi ^0/n}(1y)]`$ (70)
$`+f_{\mathrm{\Delta }^{++}\pi ^{}/p}(1y)f_{\mathrm{\Delta }^{}\pi ^+/n}(1y)\}{\displaystyle \frac{1}{2}}v^\pi ({\displaystyle \frac{x}{y}}),`$
$`x\delta s`$ $`=`$ $`{\displaystyle _0^x}dy{\displaystyle \frac{x}{y}}\{f_{\mathrm{\Lambda }K^+/p}(y)f_{\mathrm{\Lambda }K^0/n}(y)+f_{\mathrm{\Sigma }^0K^+/p}(y)f_{\mathrm{\Sigma }^0K^0/n}(y)`$ (72)
$`+f_{\mathrm{\Sigma }^+K^0/p}(y)f_{\mathrm{\Sigma }^{}K^+/n}(y)\}{\displaystyle \frac{1}{2}}u^p({\displaystyle \frac{x}{y}})`$
$`x\delta \overline{s}`$ $`=`$ $`{\displaystyle _0^x}dy{\displaystyle \frac{x}{y}}\{f_{\mathrm{\Lambda }K^+/p}(1y)f_{\mathrm{\Lambda }K^0/n}(1y)+f_{\mathrm{\Sigma }^0K^+/p}(1y)f_{\mathrm{\Sigma }^0K^0/n}(1y)`$ (74)
$`+f_{\mathrm{\Sigma }^+K^0/p}(1y)f_{\mathrm{\Sigma }^{}K^+/n}(1y)\}{\displaystyle \frac{1}{2}}\overline{s}^K({\displaystyle \frac{x}{y}})`$
## III Result and discussion
From Eqs. (57) - (74) we can see explicitly that the differences among various fluctuation functions such as $`f_{n\pi ^+/p}`$ and $`f_{p\pi ^{}/n}`$ result in the CSV in the parton distributions of the nucleon. We plot these differences in Figs. 1-3. It can be seen that the difference in the fluctuation $`NN\pi `$ ($`f_{n\pi ^+/p}f_{p\pi ^{}/n}`$, โฆ) is much larger than that in the fluctuation $`N\mathrm{\Delta }\pi `$ ($`f_{\mathrm{\Delta }^{++}\pi ^{}/p}f_{\mathrm{\Delta }^+\pi ^+/n}`$, โฆ) and the latter is much bigger than that in the fluctuation $`N\mathrm{\Lambda }K`$ ($`f_{\mathrm{\Lambda }\pi ^+/p}f_{\mathrm{\Lambda }K^0/n}`$, โฆ). Thus the CSV in the valence and sea up and down quarks should be the same order, and both larger than the CSV in the $`s`$ and $`\overline{s}`$ distributions. The difference $`f_{n\pi ^+/p}f_{p\pi ^{}/n}`$ is much larger than $`f_{p\pi ^0/p}f_{n\pi ^0/n}`$, thus the CSV in the sea of the minority quark flavor ($`\delta \overline{d}`$) will be much larger than that of the majority quark flavor ($`\delta \overline{u}`$) due to the absence of $`(f_{n\pi ^+/p}f_{p\pi ^{}/n})`$ term in $`\delta \overline{u}`$. The probabilities of the various fluctuations can be obtained by integrating the corresponding fluctuation functions. We find the probabilities of the dominant fluctuations to be
$`P(pn\pi ^+)=0.202`$ $`P(np\pi ^{})=0.205,`$ (75)
$`P(p\mathrm{\Delta }^{++}\pi ^{})=0.0481`$ $`P(n\mathrm{\Delta }^{}\pi ^+)=0.0475,`$ (76)
$`P(p\mathrm{\Lambda }K^+)=0.0127`$ $`P(n\mathrm{\Lambda }K^0)=0.0125,`$ (77)
that is there is about a $`1\%`$ excess of fluctuations $`np\pi ^{}`$ over $`pn\pi ^+`$ and $`p\mathrm{\Delta }^{++}\pi ^{}`$ over $`n\mathrm{\Delta }^{}\pi ^+`$, and about $`2\%`$ excess of $`p\mathrm{\Lambda }K^+`$ over $`n\mathrm{\Lambda }K^0`$.
We present our results for the CSV in the valence quark sector ($`x\delta d_v`$ and $`x\delta u_v`$) in Fig. 4. We find that $`x\delta d_v`$ and $`x\delta u_v`$ have similar shape and both are negative, which is quite different from the quark model prediction of $`x\delta d_v`$ being positive for most values of $`x`$ . Furthermore, our numerical results are about 10% of the quark model estimation . It has been argued that although the absolute values of $`\delta d_v`$ and $`\delta u_v`$ are small, the ratio $`R_{\mathrm{min}}=\delta d_v/d_v^p`$ may be much larger than the ratio $`R_{\mathrm{maj}}=\delta u_v/u_v^p`$ in the large-$`x`$ region since the $`d_v^p(x)/u_v^p(x)1/2`$ as $`x0`$, and values as large as $`5\%10\%`$ have been obtained for the ratio $`\delta d_v/d_v^p`$. No such large-$`x`$ enhancement appears in our calculation for both ratios. We find that the ratio $`\delta d_v/d_v^p`$ exhibits a maximum about $`0.2\%`$ at $`x=0.1`$ while the ratio $`\delta u_v/u_v^p`$ diverges as $`x0`$ but is smaller than $`0.3\%`$ in the region of $`x>0.02`$. The numerical results for the CSV in the sea quark ($`x\delta \overline{d}`$, $`x\delta \overline{u}`$, $`x\delta s`$ and $`x\delta \overline{s}`$) are given Fig. 5. We find that $`x\delta \overline{d}`$ has the largest CSV and that $`x\delta d_v`$ and $`x\delta u_v`$ are of similar magnitude, which is consistent with our expectation from the analysis of the fluctuation functions. Our prediction for the $`x\delta \overline{d}`$ being negative is opposite to the positive theoretical prediction in . Our calculation for the low-$`x`$ behaviours of $`x\delta \overline{d}`$, $`x\delta \overline{u}`$, $`x\delta s`$ and $`x\delta \overline{s}`$ are also quite different from the quark model prediction โ the quark model predicts that these quantities diverge as $`x0`$ while our calculations show all these CSV distributions go to $`0`$ as $`x0`$. We did not find any significant large CSV in the sea quark distribution of the nucleon, which is consistent with the most recent phenomenological analysis . We would like to emphasise that instead of the quark model we adopt a totally different model, the meson cloud model, to calculate the charge symmetry violation in the parton distributions of the nucleon. The quark model calculation in the small-$`x`$ region is not very reliable since the quark-diquark picture that is employed breaks down in this region. The meson cloud model is suitable in the study of the CSV in the parton distribution of the nucleon since it has the same non-perturbative origin as the $`\overline{d}/\overline{u}`$ asymmetry in the proton.
## IV Criticism of Quark Model Calculations
We have already mentioned a few of the difficulties with CSV calculations using quark models. A recent paper by Benesh and Londergan attempted to avoid any quark model specifics and relate possible CSV in the valence quark distributions to the measured valence distributions. Starting from the parton model expression for a quark distribution
$`q(x)`$ $`=`$ $`p^+{\displaystyle \underset{n}{}}\delta (p^+(1x)p_n^+)\left|n|\mathrm{\Psi }(0)|p\right|^2`$ (78)
where the intermediate state $`|n`$ has 4-momentum $`p_n`$ and the \+ components of momenta are defined by $`k^+=k^0+k^z`$, and then making the assumption that the intermediate state can be modelled by a diquark system with definite mass $`M_d`$, Benesh and Londergan investigate the consequences for CSV of varying $`M_d`$. Following the Adelaide group , we can attempt to determine the dependence of the quark distribution on $`M_d`$. Assuming that the modulus squared of the wavefunction for the struck quark in the nucleon is symmetric about the z-axis, we can use the delta function to perform the integration over transverse diquark momenta
$`{\displaystyle ๐๐ฉ_n\delta (p^+(1x)p_n^+)}`$ $`=`$ $`2\pi {\displaystyle _{p_{min}}^{\mathrm{}}}๐p_np_n`$ (79)
where
$`p_{min}`$ $`=`$ $`\left|{\displaystyle \frac{M^2(1x)^2M_d^2}{2M(1x)}}\right|`$ (80)
$`p_T`$ $`=`$ $`2M(1x)\sqrt{M_d^2+๐ฉ_n^2}M^2(1x)^2M_d^2`$ (81)
and $`M`$ is the nucleon mass. Therefore we obtain the quark distribution in the form
$`q(x)`$ $`=`$ $`{\displaystyle _{p_{min}(x,M_d)}^{\mathrm{}}}๐p_ng(p_n)`$ (82)
where $`g(k)`$ only depends on the magnitude of the 3-momentum (this is not true in the case of spin dependent quark distributions), and we have reminded ourselves that $`p_{min}`$ is a function of $`x`$ and $`M_d`$. Thus all the $`M_d`$ dependence of the quark distribution is in the lower limit of the integral. Use of the fundamental theorem of calculus then gives
$`{\displaystyle \frac{q(x)}{M_d}}`$ $`=`$ $`{\displaystyle \frac{q(x)}{x}}{\displaystyle \frac{p_{min}}{M_d}}/{\displaystyle \frac{p_{min}}{x}}`$ (83)
$`=`$ $`{\displaystyle \frac{2M_d(1x)}{M^2(1x)^2+M_d^2}}{\displaystyle \frac{q(x)}{x}}.`$ (84)
This expression is similar to that of Benesh and Londergan , except that in their case the $`/x`$ operator acts on the product of the kinematic factor $`2M_d(1x)/(M^2(1x)^2+M_d^2)`$ and the original quark distribution. The derivation of reference differs from ours in that they make a variation in $`M_d`$ under the integral in equation (78), then evaluate the integral over $`๐ฉ_n`$ by ignoring any transverse momenta of the diquark. However in our expression, all transverse momenta have been properly integrated over (in the parton model the transverse momentum of the struck quark vanishes).
Benesh and Londergan then use the idea of Close and Thomas that the quark model $`SU(4)`$ spin-isospin symmetry is broken by the color hyperfine interaction. The hyperfine interaction leads to a splitting in the masses of the spin-0 and spin-1 diquark states and hence to a difference between the up and down valence distributions:
$`u_v(x)`$ $`=`$ $`{\displaystyle \frac{3}{2}}q_v^s(x)+{\displaystyle \frac{1}{2}}q_v^t(x)`$ (85)
$`d_v(x)`$ $`=`$ $`q_v^t(x)`$ (86)
where the superscripts $`s,t`$ refer to singlet and triplet diquark states respectively. If the $`N\mathrm{\Delta }`$ mass splitting is also caused by the color hyperfine interaction, then the shifts in the singlet and triplet diquark masses are found to be
$`\delta _{hf}M_d^t`$ $`=`$ $`{\displaystyle \frac{1}{3}}\delta _{hf}M_d^s=+50\mathrm{M}\mathrm{e}\mathrm{V}.`$ (87)
By now expanding $`q_v^s(x,M_d)`$ and $`q_v^t(x,M_d)`$ in Taylor series to first order in $`\delta M_d`$ about the symmetry point $`M_d=M_d^0,q_v^s(x,M_d^0)=q_v^t(x,M_d^0)`$ Benesh and Londergan obtain the shift in the triplet quark distribution
$`\delta _{hf}q_v^t(x)`$ $`=`$ $`{\displaystyle \frac{1}{6}}(2d_v(x)u_v(x)).`$ (88)
Now as the CSV at the quark level comes from quark mass and electromagnetic effects, both of which are iso-vector, the only mass shift is in $`M_d^t`$, and to first order the shift in the triplet quark distribution will be proportional to that from the hyperfine interaction
$`\delta _{CSV}q_v^t(x)`$ $`=`$ $`{\displaystyle \frac{\delta _{CSV}M_d^t}{\delta _{hf}M_d^t}}{\displaystyle \frac{2d_v(x)u_v(x)}{6}}.`$ (89)
The main difficulty with this argument is that it is entirely based on first order shifts in the quark distributions. However the second order terms can be estimated, and they are of similar magnitude to the first order terms. Expanding the quark distributions to second order in $`\delta M_d`$ about the symmetry point we have
$`q_v(x,M_d^0+\delta M_d)`$ $`=`$ $`q_v(x,M_d^0)+\delta M_d{\displaystyle \frac{q_v(x,M_d)}{M_d}}\left|{}_{M_d=M_d^0}{}^{}+{\displaystyle \frac{1}{2}}(\delta M_d)^2{\displaystyle \frac{^2q_v(x,M_d)}{(M_d)^2}}\right|_{M_d=M_d^0}`$ (90)
where the partial derivatives on the right hand side can be evaluated using (84). This then gives for the hyperfine shift in the triplet quark distribution
$`\delta _{hf}q_v^t(x)`$ $`=`$ $`{\displaystyle \frac{1}{3}}\delta _{hf}q_v^s(x)+2(\delta M_d^t)^2{\displaystyle \frac{^2q_v(x,M_d)}{(M_d)^2}}|_{M_d=M_d^0}`$ (91)
$`=`$ $`{\displaystyle \frac{1}{6}}(2d_v(x)u_v(x))+{\displaystyle \frac{3}{2}}(\delta M_d^t)^2{\displaystyle \frac{^2q_v(x,M_d)}{(M_d)^2}}|{}_{M_d=M_d^0}{}^{}.`$ (92)
As an estimate of the second order term we can approximate $`q_v(x,M_d^0)`$ by $`d_vx`$ or $`(u_v(x)+d_v(x))/3`$ and use one of the well-known parametrizations of the quark valence distributions . We also take our value for $`M_d^0`$ to lie in the range $`(0.650.85)M`$, which is the range suggested by quark models, though the results are not very sensitive to the value of $`M_d^0`$ in this range. In Fig. 6 we compare the first and second order terms for $`\delta _{hf}q_v^t(x)`$. We can see that our estimate of the second order term is of comparable magnitude to the first order term over most of the $`x`$-range. Indeed for low $`x`$ it is larger, showing the unreliability of quark model calculations in this region. For $`x>0.2`$ the second order term is of opposite sign to the first order term, which indicates that the first order estimate of the shift in the quark distribution is too large in this region. This in turn implies that the estimate of CSV induced shifts in the quark distributions in this region are also too large. These conclusions are not much influenced by the choice of valence quark parametrization, the value chosen for $`M_d^0`$, or whether we use Benesh and Londerganโs expression for the dependence of the quark distribution on $`M_d`$ rather than equation (84). The reason for these conclusions not being greatly influenced by the choice of expression for $`q(x)/M_d`$ is that, with the quark distributions used, the highest order derivative term in $`x`$ always dominates. This is a consequence of the divergences in the valence quark distributions near $`x=0`$, $`q_v(x)x^{0.5}`$ in all cases.
## V Summary
Although it has been generally assumed that charge symmetry was highly respected in the nucleon system, there have been some phenomenological analysis and theoretical calculations about the possible extent of CSV in the parton distributions. Any unexpected large CSV will greatly affect our understanding on the non-perturbative dynamics and hadronic structure, and the extraction of $`sin^2\theta _W`$ from neutrino scattering. Up to now most theoretical attempts to calculate the CSV in the parton distributions are based on the quark model and employ the quark-diquark model. In this paper we point out that CSV in both the valence and sea quark distributions of the nucleon can arise from the non-perturbative dynamics of the nucleon. We present an alternative analysis of CSV in the parton distributions employing the meson cloud model, which has previously been successful in the study of the flavour asymmetry and the quark-antiquark asymmetry of the nucleon. In the meson cloud model the proton and neutron may fluctuate into hadron-meson Fock states in which the hadrons and mesons are in different charged states respectively. As we consider the mass differences among these hadrons and mesons, the probabilities of proton and neutron fluctuating into the corresponding Fock states will be different. Thus the non-perturbative contributions to the valence and sea quarks distributions will be different, which naturally leads to the CSV in both the valence and sea distributions of the nucleon. Our predictions for the CSV in the valence sector and sea sector are both different from the quark model calculations. We also point out the deficiencies of quark model based calculations of CSV in the parton distributions. In particular the quark-diquark picture is inadequate at low-$`x`$, and in the medium-$`x`$ region the use of a first order shift in the parton distributions must be questioned, as higher order shifts are of similar magnitude. The coming experimental information on the parton distributions of the nucleon and more theoretical studies on this issue will examine these calculations.
## Acknowledgments
This work was partially supported by the Massey Postdoctoral Foundation, New Zealand. We are grateful to Prof. A. W. Thomas for useful discussions. F. G. Cao would like to thank the Special Research Center for the Subatomic Structure of Matter for its hospitality.
Table 1. Parameters in Eq. (50) at $`Q^2=4`$ GeV<sup>2</sup>.
| | $`a`$ | $`b`$ | $`c`$ | $`d`$ | $`e`$ |
| --- | --- | --- | --- | --- | --- |
| $`v^\pi (x,Q^2)`$ | $`1.712`$ | $`0.518`$ | $`1.182`$ | $`0.836`$ | $`0.972`$ |
| $`\overline{s}^K(x,Q^2)`$ | $`0.803`$ | $`0.516`$ | $`1.306`$ | $`0.762`$ | $`0.957`$ |
| $`u^p(x,Q^2)`$ | $`1.029`$ | $`0.572`$ | $`3.933`$ | $`1.550`$ | $`6.033`$ |
| $`d^p(x,Q^2)`$ | $`0.615`$ | $`0.575`$ | $`5.096`$ | $`1.102`$ | $`6.773`$ |
## Figure Captions
Fig. 1. The differences between fluctuation functions $`f_{N\pi /p}`$ and $`f_{N\pi /n}`$.
Fig. 2. The differences between fluctuation functions $`f_{\mathrm{\Delta }\pi /p}`$ and $`f_{\mathrm{\Delta }\pi /n}`$.
Fig. 3. The differences between fluctuation functions $`f_{\mathrm{\Lambda }(\mathrm{\Sigma })K/p}`$ and $`f_{\mathrm{\Lambda }(\mathrm{\Sigma })K/n}`$.
Fig. 4 The charge symmetry violation in the valence quark sector. $`R_{\mathrm{min}}=\delta d_v/d_v^p`$ and $`R_{\mathrm{maj}}=\delta u_v/u_v^p`$.
Fig. 5 The charge symmetry violation in the sea quark sector.
Fig. 6 The first and second order shifts in the triplet quark distribution caused by the color hyperfine interaction. The solid curve is the first order shift $`(2d_v(x)u_v(x))/6`$ calculated using the parametrizations of reference . The dashed curve is the second order shift estimated using $`q(x)=(d_v(x)+u_v(x))/3`$, $`M_d^0=0.75M`$, and a mass shift of 50 MeV for the triplet diquark state. |
warning/0001/physics0001064.html | ar5iv | text | # Space-time geometry of quantum dielectrics
## I Introduction
A moving dielectric medium appears to light as an effective gravitational field . The medium alters the way in which an electromagnetic field perceives space and time, formulated most concisely in Gordonโs effective spaceโtime metric
$$g_{\alpha \beta }^F=g_{\alpha \beta }+\left(\frac{1}{\epsilon \mu }1\right)u_\alpha u_\beta .$$
(1)
We allow for a backโground metric $`g_{\alpha \beta }`$, mostly to have the convenience of choosing arbitrary coordinates, but also for the possible inclusion of a genuine gravitational field. Gordonโs metric (1) depends on the dielectric properties of the medium, on the permittivity $`\epsilon `$ and on the magnetic permeability $`\mu `$, as well as on the fourโdimensional flow $`u^\alpha `$ of the medium (the local fourโvelocity). The product $`\epsilon \mu `$ is the square of the refractive index and the prefactor $`1(\epsilon \mu )^1`$ is known as Fresnelโs dragging coefficient (in Fresnelโs days the part of the ether that the moving medium is able to drag ). In the limit of geometrical optics , light rays are zeroโgeodesic lines with respect to Gordonโs metric . In the special case of a medium at rest, this result is equivalent to Fermatโs principle and to the formulation of geometrical optics as a nonโEuclidean geometry in space .
Light sees dielectric matter as an effective spaceโtime metric. How does matter see light? In atom optics , the traditional role of light and matter is reversed: Atomic deโBroglie waves are subject to atomโoptical instruments made of light. Light acts on matter waves in a similar way as matter acts on light. This paper indicates that an atomic matter wave experiences an electromagnetic field as the effective metric
$$g_{\alpha \beta }^A=\left(1a_F\right)g_{\alpha \beta }bT_{\alpha \beta }^F$$
(2)
with
$$a=\frac{1}{mc^2\rho }(\epsilon +\frac{1}{\mu }2),b=\frac{1}{mc^2\rho }(\epsilon \frac{1}{\mu }).$$
(3)
Here $`_F`$ is the Lagrangian of the free electromagnetic field, defined in Eq. (8), and $`T_{\alpha \beta }^F`$ is the freeโfield energyโmomentum tensor (10). As usual, $`c`$ denotes the speed of light in vacuum and $`m`$ is the mass of a single dielectric atom. In the definition (3), $`\rho `$ can be regarded as the probability density of the atomic deโBroglie wave, for most practical purposes. (Strictly speaking, $`mc^2\rho `$ describes the total enthalpy density of the matter wave, including the rest energy as the lionโs share.) Throughout this paper we employ SI units and use the LandauโLifshitz convention of general relativity (with the exception of using greek spaceโtime and latin space indices). To derive the result (2) with the dielectric parameters (3) we postulate that the interaction between light and matter takes on the general form of a metric. Then we demonstrate the consistency of this idea with previous knowledge, and in particular with Gordonโs metric (1).
The metric (2) indicates that the energyโmomentum of light curves directly the spaceโtime of a dielectric matter wave. Under normal circumstances the deviation from the backโground geometry is very small, see Eqs. (2) and (3), because the ratio between the electromagnetic energy and the atomic rest energy $`mc^2`$ is typically an extremely small number. In the Newtonian limit of general relativity , the gravitational correction to a flat Minkowski spaceโtime is tiny as well, because the correction is proportional to the ratio between the potential energy and $`mc^2`$ of a test particle. For weak gravitational fields and low testโparticle velocities, general relativity is an equivalent formulation of Newtonian physics that agrees in all predicted effects and yet establishes a radically different physical interpretation. Similarly, given the current state of the art in atom optics, the idea that light curves the spaceโtime for matter waves is an equivalent formulation of the known light forces, i.e. of the dipole force and of the recently investigated Rรถntgen interaction . However, one can conceive of significantly enhancing the dielectric properties of matter waves using similar methods as in the spectacular demonstrations of slow light . Loosely speaking, a large effective dielectric constant $`\epsilon `$ could counteract the rest energy $`mc^2`$ in the relations (3). In this way one could use light to build atomโoptical analogues of astronomical objects on Earth, for example a black hole made of light.
## II Electromagnetic fields
### A Field tensors
Let us first agree on the definitions of the principal electromagnetic quantities in SI units in general relativity. We employ the spaceโtime coordinates $`x^\mu =(ct,๐ฑ)`$. The electromagnetic fourโpotential is
$$A_\nu =(U,c๐).$$
(4)
The electromagnetic fieldโstrength tensor is constructed as
$$F_{\mu \nu }D_\mu A_\nu D_\nu A_\mu =_\mu A_\nu _\nu A_\mu $$
(5)
using the covariant derivatives $`D_\mu `$ with respect to the backโground metric $`g_{\mu \nu }`$. As is well known , in the definition (5) of $`F_{\mu \nu }`$ on a possibly curved spaceโtime, we have been able to replace the $`D_\mu `$ by ordinary partial derivatives $`_\mu /x^\mu `$. The fieldโstrength tensor reads in localโgalilean coordinates (in a local Minkowski frame)
$$F_{\mu \nu }=\left(\begin{array}{cccc}0& E_x& E_y& E_z\\ E_x& 0& cB_z& cB_y\\ E_y& cB_z& 0& cB_x\\ E_z& cB_y& cB_x& 0\end{array}\right).$$
(6)
It will become useful at a later stage of this enterprise to introduce a fourโdimensional formulation $`H^{\mu \nu }`$ of the dielectric $`๐`$ and $`๐`$ fields,
$$H^{\mu \nu }=\left(\begin{array}{cccc}0& D_x& D_y& D_z\\ D_x& 0& H_z/c& H_y/c\\ D_y& H_z/c& 0& H_x/c\\ D_z& H_y/c& H_x/c& 0\end{array}\right),$$
(7)
here defined in localโgalilean coordinates.
### B Quadratic field tensors
In dielectric media, induced atomic dipoles constitute an interaction between light and matter that is quadratic in the electromagnetic fieldโstrength tensor . Let us therefore list a set of linearly independent secondโrank tensors that are quadratic in $`F_{\mu \nu }`$. The most elementary one is the product of the metric tensor $`g_{\mu \nu }`$ with the scalar Lagrangian $`_F`$ of the free electromagnetic field . This Lagrangian is
$$_F=\frac{\epsilon _0}{4}F_{\alpha \beta }F^{\alpha \beta }=\frac{\epsilon _0}{4}g^{\alpha \alpha ^{}}g^{\beta \beta \mathrm{`}}F_{\alpha \beta }F_{\alpha ^{}\beta ^{}},$$
(8)
or, in localโgalilean coordinates,
$$_F=\frac{\epsilon _0}{2}\left(E^2c^2B^2\right).$$
(9)
Another quadratic secondโrank tensor is the free electromagnetic energyโmomentum tensor
$$T_{\mu \nu }^F=\epsilon _0F_{\mu \alpha }g^{\alpha \beta }F_{\beta \nu }_Fg_{\mu \nu },$$
(10)
or, in localโgalilean coordinates,
$$T_{\mu \nu }^F=\left(\begin{array}{cc}I& ๐/c\\ ๐/c& \sigma \end{array}\right),T_F^{\mu \nu }=\left(\begin{array}{cc}I& ๐/c\\ ๐/c& \sigma \end{array}\right)$$
(11)
with
$`I`$ $`=`$ $`{\displaystyle \frac{\epsilon _0}{2}}(E^2+c^2B^2),๐=\epsilon _0c^2๐๐,`$ (12)
$`\sigma `$ $`=`$ $`\epsilon _0\left[\left({\displaystyle \frac{E^2}{2}}+{\displaystyle \frac{c^2B^2}{2}}\right)\mathrm{๐}๐๐c^2๐๐\right].`$ (13)
Here $`I`$ denotes the intensity, $`๐`$ is the Poynting vector, and $`\sigma `$ is Maxwellโs stress tensor. The symbols $``$ and $``$ denote the threeโdimensional vector and tensor product, respectively.
We can only form secondโrank tensors from $`F_{\alpha \beta }F^{\alpha ^{}\beta ^{}}`$ by some contraction. Consequently, the linear combinations of the two elementary tensors $`_Fg_{\mu \nu }`$ and $`T_{\mu \nu }^F`$ form the complete class of secondโrank tensors that are quadratic in the field strengths $`F_{\mu \nu }`$.
## III Classical atoms
### A Postulates
Consider a classical atom in an electromagnetic field. The atom is pointโlike, has a mass $`m`$, and can sustain induced electric and magnetic dipoles. In the restframe of the atom the dipoles respond to the square of the electric field strength, $`E^2`$, and to the magnetic $`B^2`$, respectively. How does a dielectric atom experience the electromagnetic field when the atom is moving?
Let us postulate that the atom sees the field as an effective metric. Consequently, according to general relativity , the action $`S_0`$ of the atom is
$$S_0=mcds,ds^2=g_{\mu \nu }^Adx^\mu dx^\nu .$$
(14)
Let us further postulate that the metric of the atom, $`g_{\mu \nu }^A`$, is quadratic in the electromagnetic field strengths. Any metric is a secondโrank tensor. Hence, we obtain from Sec. IIB the general form (2) mentioned in the Introduction.
### B Properties
A metric of the structure (2) has nice mathematical properties. In particular, the contravariant metric tensor $`g_A^{\mu \nu }`$ (the inverse of $`g_{\mu \nu }^A`$) takes on a simple analytic expression,
$$\sqrt{g_A}g_A^{\mu \nu }=\sqrt{g}\left[\left(1a_F\right)g^{\mu \nu }+bT_F^{\mu \nu }\right]$$
(15)
with
$$g_A\mathrm{det}(g_{\mu \nu }^A)$$
(16)
and
$$\sqrt{g_A}=\sqrt{g}\left[\left(1a_F\right)^2\frac{b^2}{4}T_{\alpha \beta }^FT_F^{\alpha \beta }\right],$$
(17)
as one verifies in localโgalilean coordinates, with the relation $`T_{\alpha \beta }^FT_F^{\alpha \beta }=\epsilon _0^2[(E^2c^2B^2)^2+4c^2(๐๐)^2]`$.
### C Nonโrelativistic limit
So far, we have not seen how the metric theory (2) and (14) is related to the model of a moving induced dipole. Let us consider the nonโrelativistic limit of velocities low compared with the speed of light. This limit corresponds to a motion in an inertial frame close to a restframe coโmoving with the atom. We also regard the electromagnetic field energy to be weak compared with the atomic rest energy $`mc^2`$. We neglect any genuine gravitational field, and obtain in cartesian coordinates
$`ds`$ $`=`$ $`\sqrt{\left(1a_F\right)\left(c^2dt^2d๐ฑ^2\right)bT_{\mu \nu }^Fdx^\mu dx^\nu }`$ (18)
$``$ $`\sqrt{c^2dt^2d๐ฑ^2\left(a_F+bT_{00}^F\right)c^2dt^2}`$ (19)
$``$ $`\left(1{\displaystyle \frac{v^2}{2c^2}}{\displaystyle \frac{a_F+bT_{00}^F}{2}}\right)cdt`$ (20)
with $`๐ฏ=d๐ฑ/dt`$. Consequently, we can write the action $`S_0`$ as
$$S_0=mc๐s\left(mc^2+L_0\right)๐t$$
(21)
with the nonโrelativistic Lagrangian
$$L_0=\frac{m}{2}v^2+\frac{\alpha _E}{2}E^2+\frac{\alpha _B}{2}c^2B^2$$
(22)
and
$`\alpha _E={\displaystyle \frac{a+b}{2}}\epsilon _0mc^2`$ , $`\alpha _B={\displaystyle \frac{ba}{2}}\epsilon _0mc^2,`$ (23)
$`a={\displaystyle \frac{\alpha _E\alpha _B}{\epsilon _0mc^2}}`$ , $`b={\displaystyle \frac{\alpha _E+\alpha _B}{\epsilon _0mc^2}}.`$ (24)
The Lagrangian $`L_0`$ describes indeed a nonโrelativistic atom with electric and magnetic polarizibility $`\alpha _E`$ and $`\alpha _B`$, respectively. In this way we have verified that the metric theory (2) and (14) agrees with the physical picture of traveling dipoles and, simultaneously, we have been able to express the coefficients $`a`$ and $`b`$ of the metric (2) in terms of atomic quantities.
## IV Matter waves
### A Postulate
Gordon has shown that an electromagnetic field experiences dielectric matter as the effective metric (1). Here we postulate that also the opposite is true: A dielectric matter wave sees the electromagnetic field as a metric, and in particular as the metric (2) that we have motivated for traveling dipoles in Sec. III. We demonstrate the consistency of this idea with Gordonโs theory in Sec. V. Let us model the matter wave as, fittingly, a complex KleinโGordon scalar $`\psi `$ in an effectively curved spaceโtime. The action $`S_A`$ of the atom wave $`\psi `$ is
$$S_A=_A\sqrt{g}d^4x$$
(25)
in terms of the KleinโGordon Lagrangian
$`_A`$ $`=`$ $`\sqrt{{\displaystyle \frac{g_A}{g}}}\left[{\displaystyle \frac{1}{2m}}g_A^{\mu \nu }\left(i\mathrm{}_\mu \psi ^{}\right)\left(i\mathrm{}_\nu \psi \right){\displaystyle \frac{mc^2}{2}}\psi ^{}\psi \right]`$ (26)
$`=`$ $`\sqrt{{\displaystyle \frac{g_A}{g}}}\left[{\displaystyle \frac{\mathrm{}^2}{2m}}(D_A^\mu \psi ^{})(D_\mu ^A\psi ){\displaystyle \frac{mc^2}{2}}\psi ^{}\psi \right]`$ (27)
where we have employed the covariant derivatives $`D_\mu ^A`$ with respect to the effective metric (2). The action (25) is minimal if the matter wave $`\psi `$ obeys the KleinโGordon equation
$$D_\mu ^AD_A^\mu \psi +\frac{m^2c^2}{\mathrm{}^2}\psi =0,$$
(28)
or, written explicitly ,
$$\frac{1}{\sqrt{g_A}}_\mu \left(\sqrt{g_A}g_A^{\mu \nu }_\nu \psi \right)+\frac{m^2c^2}{\mathrm{}^2}\psi =0.$$
(29)
Equation (29) together with the functions (15) and (17) and the parameters (24) describes how atomic matter waves respond to electromagnetic fields.
### B Rรถntgen limit
Let us prove explicitly that the KleinโGordon Lagrangian (27) contains the known light forces in the limit of relatively low velocities (compared with $`c`$) and of weak fields (compared with $`mc^2`$). We separate from the atomic wave function $`\psi `$ the notorious rapid oscillations due to the rest energy $`mc^2`$ by defining
$$\phi \psi \mathrm{exp}\left(i\frac{mc^2}{\mathrm{}}t\right).$$
(30)
We neglect gravity and obtain in cartesian coordinates
$`_A`$ $``$ $`\sqrt{g_A}[{\displaystyle \frac{1}{2}}g_A^{00}(mc^2\phi ^{}\phi +i\mathrm{}\phi ^{}\dot{\phi }i\mathrm{}\dot{\phi }^{}\phi )`$ (33)
$`+{\displaystyle \frac{i\mathrm{}c}{2}}g_A^{0k}\left(\phi ^{}_k\phi \phi _k\phi ^{}\right)`$
$`{\displaystyle \frac{\mathrm{}^2}{2m}}g_A^{kl}(_k\phi ^{})(_l\phi ){\displaystyle \frac{mc^2}{2}}\phi ^{}\phi ]`$
$``$ $`{\displaystyle \frac{mc^2}{2}}\phi ^{}\phi \left(1a_F+bT_F^{00}\right)+{\displaystyle \frac{i\mathrm{}}{2}}\left(\phi ^{}\dot{\phi }\dot{\phi }^{}\phi \right)`$ (36)
$`+{\displaystyle \frac{i\mathrm{}c}{2}}bT_F^{0k}\left(\phi ^{}_k\phi \phi _k\phi ^{}\right)`$
$`{\displaystyle \frac{\mathrm{}^2}{2m}}(\phi ^{})(\phi ){\displaystyle \frac{mc^2}{2}}\left(12a_F\right)\phi ^{}\phi .`$
$`=`$ $`{\displaystyle \frac{i\mathrm{}}{2}}\left(\phi ^{}\dot{\phi }\dot{\phi }^{}\phi \right){\displaystyle \frac{\mathrm{}^2}{2m}}(\phi ^{})(\phi )`$ (39)
$`+\left({\displaystyle \frac{\alpha _E}{2}}E^2+{\displaystyle \frac{\alpha _B}{2}}c^2B^2\right)\phi ^{}\phi `$
$`+{\displaystyle \frac{\alpha _E+\alpha _B}{2m}}\left(๐๐\right)i\mathrm{}\left(\phi ^{}\phi \phi \phi ^{}\right).`$
This result agrees with the Rรถntgen Lagrangian of Ref. in the limit of weak fields and, consequently, describes indeed the known nonโresonant light forces including the Rรถntgen interaction .
### C Dielectric flow
Accelerated by light forces, an atomic matter wave will form a probability current that appears as a dielectric flow. Let us calculate the flow from the phase $`S`$ of the wave function,
$$\psi =|\psi |e^{iS}.$$
(40)
We introduce
$$w^\mu \frac{\mathrm{}}{mc}g_A^{\mu \nu }_\nu S,$$
(41)
and obtain from the KleinโGordon equation (29) the conservation law of the fourโdimensional probability current,
$$D_\mu ^A\left(|\psi |^2w^\mu \right)=\frac{1}{\sqrt{g_A}}_\mu \left(\sqrt{g_A}|\psi |^2w^\mu \right)=0.$$
(42)
In the absence of electromagnetic forces, $`w^\mu `$ describes the local fourโvelocity of a free matter wave. In the presence of a field, we introduce the dielectric flow $`u^\mu `$ by normalizing $`w^\mu `$ to unity with respect to the backโground metric $`g_{\mu \nu }`$,
$$u^\mu \frac{w^\mu }{w},w\sqrt{g_{\mu \nu }w^\mu w^\nu }.$$
(43)
We define two densities, $`\varrho `$ and $`\rho `$, as
$$\varrho |\psi |^2w\sqrt{\frac{g_A}{g}},\rho \varrho w.$$
(44)
We obtain from the conservation law (42)
$$\frac{1}{\sqrt{g}}_\mu \left(\sqrt{g}\varrho u^\mu \right)=D_\mu \left(\varrho u^\mu \right)=0.$$
(45)
Consequently, $`\varrho `$ is the scalar probability density of the atomic deโBroglie wave. For most practical purposes the two densities $`\varrho `$ and $`\rho `$ are identical, because $`w`$ is unity to a very good approximation. The difference between $`\varrho `$ and $`\rho `$ is subtle: In Sec. VE we show that $`mc^2\rho `$ is the total enthalpy density of the dielectric matter wave, with the restโenergy density $`mc^2\varrho `$ as the lionโs share.
### D Hydrodynamic limit
As has been mentioned, the objective of this paper is the proof that the metric interaction (2) between matter waves and light is compatible with the known theory of dielectrics . When a matter wave or, more likely, a macroscopic condensate of identical matter waves reaches the status of a dielectric it behaves like a quantum fluid. In this macroscopic limit the deโBroglie density varies over significantly larger ranges than the deโBroglie wave length (the same applies to frequencies), and a hydrodynamic approach has become extremely successful . Let us approximate
$$i\mathrm{}_\nu \psi \psi \mathrm{}_\nu S.$$
(46)
We obtain from the KleinโGordon Lagrangian (27) the hydrodynamic approximation
$$_A=\sqrt{\frac{g_A}{g}}|\psi |^2\left[\frac{\mathrm{}^2}{2m}g_A^{\mu \nu }(_\mu S)(_\nu S)\frac{mc^2}{2}\right].$$
(47)
Let us consider the EulerโLagrange equations derived from the hydrodynamic Lagrangian (47). We obtain from the $`_\mu S`$ dependence of $`_A`$ the dielectric flow (45) and from a variation with respect to $`|\psi |^2`$ the dielectric HamiltonโJacobi equation
$$g_A^{\mu \nu }(_\mu S)(_\nu S)=\frac{m^2c^2}{\mathrm{}^2},$$
(48)
or, in terms of the fourโvector $`w^\mu `$ of Eq. (41),
$$g_{\mu \nu }^Aw^\mu w^\nu =1.$$
(49)
In the hydrodynamic limit the $`w^\mu `$ vector represents a fourโvelocity that is normalized with respect to the effective metric (2). We also see that the hydrodynamic Lagrangian (47) vanishes at the actual minimum that corresponds to the physical behavior of a dielectric matter wave.
## V Quantum dielectrics
### A Actio et reactio
In the previous section we considered a dielectric matter wave in a given electromagnetic field. Gordon studied the opposite extreme โ an electromagnetic field in a given dielectric medium. Let us address here an intermediate regime of actio et reactio where light acts on matter as well as matter acts on light. Such a physical regime, characterizing a quantum dielectric, occurs for example when a BoseโEinstein condensate of an alkali vapor interacts nonโresonantly with light . If we were able to arrive at Gordonโs metric (1) from our starting point (2) we were inclined to take this as evidence that our approach is right.
To include the dynamics of the electromagnetic field we add the freeโfield Lagrangian $`_F`$ to the atomic $`_A`$ in hydrodynamic approximation (47),
$$=_F+_A,$$
(50)
and regard the electromagnetic field as a dynamic object that is subject to the principle of least action. We could also easily include other interactions by additional terms in $`_A`$ such as the atomic collisions within a BoseโEinstein condensate by a GrossโPitaevskii term. Let us consider the field variation
$`\delta _F`$ $`=`$ $`\delta _F_F+\sqrt{{\displaystyle \frac{g_A}{g}}}|\psi |^2{\displaystyle \frac{\mathrm{}^2}{2m}}(_\mu S)(_\nu S)\delta _Fg_A^{\mu \nu }`$ (52)
$`+\sqrt{{\displaystyle \frac{g}{g_A}}}_A\delta _F\sqrt{{\displaystyle \frac{g_A}{g}}}.`$
As has been mentioned in Sec. IVD, the atomic Lagrangian $`_A`$ vanishes at the minimum of the action, in the hydrodynamic limit. We utilize that
$$\delta _Fg_A^{\mu \nu }=g_A^{\mu \alpha }g_A^{\nu \beta }\delta _Fg_{\alpha \beta }^A,$$
(53)
and obtain, using Eqs. (41-44),
$$\delta _F=\delta _F_F\frac{mc^2}{2}\rho u^\alpha u^\beta \delta _Fg_{\alpha \beta }^A.$$
(54)
The variation of the Lagrangian with respect to the field determines via the EulerโLagrange equations the field dynamics. Can we cast $`\delta _F`$ in the role of a dielectric?
### B Effective Lagrangian
The principal mathematical artifice of this paper is an effective Lagrangian that is designed to agree with $``$ under field variations, and that describes a dielectric medium,
$$_{\mathrm{EFF}}_F+\frac{mc^2}{2}\rho \left(g_{\alpha \beta }g_{\alpha \beta }^A\right)u^\alpha u^\beta $$
(55)
with
$$\delta _F=\delta _F_{\mathrm{EFF}}.$$
(56)
Note that the two field variations in the relation (56) differ in a subtle way: On the leftโhand side, $`\delta _F`$ abbreviates the total variation with respect to the electromagnetic field, whereas on the rightโhand side of Eq. (56) we treat $`\epsilon `$, $`\mu `$, and $`u^\alpha `$ as being fixed, despite their hidden dependence on the field due to the relations (41-44).
We show explicitly in Sec. VD that $`_{\mathrm{EFF}}`$ is indeed the desired Lagrangian of light in a dielectric medium. Here we note that $`_{\mathrm{EFF}}`$ may metamorphose into a multitude of forms. For example, we introduce the permittivity $`\epsilon `$ and the magnetic permeability $`\mu `$ in terms of elementary atomic quantities and in accordance with the parameters (3) mentioned in the Introduction
$$\epsilon =1+\frac{\alpha _E}{\epsilon _0}\rho ,\frac{1}{\mu }=1\frac{\alpha _B}{\epsilon _0}\rho .$$
(57)
In this way we obtain directly from Eqs. (2) and (3)
$$_{\mathrm{EFF}}=\frac{1}{2}\left[\left(\epsilon +\frac{1}{\mu }\right)_F+\left(\epsilon \frac{1}{\mu }\right)u^\alpha u^\beta T_{\alpha \beta }^F\right].$$
(58)
We can also express the effective Lagrangian as
$$_{\mathrm{EFF}}=\frac{1}{\mu }_F+\epsilon _0\frac{\epsilon \mu 1}{2\mu }F_{\alpha ^{}\beta ^{}}F_{\alpha \beta }u^\alpha u^\alpha ^{}g^{\beta \beta ^{}},$$
(59)
due to the definition (10) of the freeโfield energyโmomentum tensor, or we may perform further manipulations, utilizing the relations
$`F_{\alpha ^{}\beta ^{}}`$ $`F_{\alpha \beta }u^\alpha u^\alpha ^{}g^{\beta \beta ^{}}=F_{\alpha ^{}\beta ^{}}F_{\alpha \beta }g^{\alpha \alpha ^{}}u^\beta u^\beta ^{},`$ (60)
$`F_{\alpha ^{}\beta ^{}}`$ $`F_{\alpha \beta }u^\alpha u^\alpha ^{}u^\beta u^\beta ^{}=0,`$ (61)
due to the symmetry of the backโground metric $`g^{\alpha \beta }`$ and the antiโsymmetry of the fieldโstrength tensor $`F_{\alpha \beta }`$.
### C Gordonโs metric
Quite remarkably, one can express the effective Lagrangian in the form
$$_{\mathrm{EFF}}=\frac{\epsilon _0}{4\mu }F_{\alpha \beta }F^{(\alpha )(\beta )}$$
(62)
with
$$F^{(\alpha )(\beta )}g_F^{\alpha \alpha ^{}}g_F^{\beta \beta ^{}}F_{\alpha ^{}\beta ^{}}$$
(63)
and
$$g_F^{\alpha \beta }=g^{\alpha \beta }+(\epsilon \mu 1)u^\alpha u^\beta .$$
(64)
The effective Lagrangian appears as the free electromagnetic Lagrangian in a curved spaceโtime with metric (64). A short exercise proves that $`g_F^{\alpha \beta }`$ is the inverse of $`g_{\alpha \beta }^F`$, i.e., as the notation suggests it, the contravariant metric tensor with respect to the covariant $`g_{\alpha \beta }^F`$. Consequently, we have indeed arrived at Gordonโs spaceโtime geometry of light in moving media, starting from our metric (2), which supports the validity of our postulates.
Note that Gordonโs spaceโtime geometry is not completely perfect . The metrics (1) and (64) depend only on the square of the refractive index, $`\epsilon \mu `$, whereas a dielectric medium is characterized by two dielectric constants $`\epsilon `$ and $`\mu `$, in general. What is the imperfection in the Lagrangian (62)? In order to describe a density in general relativity, and in particular a Lagrangian density, we must consider the determinant of the metric that describes the scaling of space and time. Gordon calculated the determinant by employing coโmoving medium coordinates, with the result
$$g_F\mathrm{det}\left(g_{\alpha \beta }^F\right)=\frac{g}{\epsilon \mu }.$$
(65)
Hence we obtain the effective action
$`S_{\mathrm{EFF}}`$ $`=`$ $`{\displaystyle _{\mathrm{EFF}}\sqrt{g}d^4x}`$ (66)
$`=`$ $`{\displaystyle \frac{\epsilon _0}{4}}{\displaystyle \sqrt{\frac{\epsilon }{\mu }}F_{\alpha \beta }F^{(\alpha )(\beta )}\sqrt{g_F}d^4x}`$ (67)
that may deviate from the perfect
$$S_F=\frac{\epsilon _0}{4}F_{\alpha \beta }F^{(\alpha )(\beta )}\sqrt{g_F}d^4x$$
(68)
when $`\epsilon /\mu `$ varies significantly. However, when the density profile of the quantum liquid varies smoothly compared with the wave length of light we can neglect the variation of $`\epsilon /\mu `$. Ultracold atoms or BoseโEinstein condensates are usually in this regime that is also compatible with the hydrodynamic behavior of the quantum liquid.
### D Maxwellโs equations
The first group of Maxwellโs equations follows from the structure (5) of the fieldโstrength tensor $`F_{\mu \nu }`$. The EulerโLagrange equations of the effective Lagrangian (62) yield the second group ,
$$D_\alpha H^{\alpha \beta }=0\text{or}_\alpha \left(\sqrt{g}H^{\alpha \beta }\right)=0$$
(69)
with the constitutive equations
$$H^{\alpha \beta }=\frac{ฯต_0}{\mu }F^{(\alpha )(\beta )}.$$
(70)
In localโgalilean coordinates we can represent $`H^{\alpha \beta }`$ in terms (7) of the dielectric $`๐`$ and $`๐`$ fields in SI units. In this way we find yet another physically meaningful expression for the effective Lagrangian,
$$_{\mathrm{EFF}}=\frac{1}{4}F_{\alpha \beta }H^{\alpha \beta }=\frac{๐๐}{2}\frac{๐๐}{2},$$
(71)
which is indeed the explicit form of the Lagrangian for the electromagnetic field in a linear dielectric.
Equation (70) is equivalent to Minkowskiโs constitutive equations in a moving medium . In the limit of low velocities we recover the familiar relations $`๐=\epsilon _0\epsilon ๐`$ and $`\mu ๐=\epsilon _0c^2๐`$, and, via Eq. (57),
$$๐(\epsilon _0+\alpha _E\varrho )๐,๐(\epsilon _0\alpha _B\varrho )c^2๐,$$
(72)
assuming a weak field when $`\rho \varrho `$. Relativistic firstโorder corrections lead to the constitutive equations derived in Ref. that describe, for example, the Rรถntgen effect or lead to Fresnelโs light drag measured in Fizeauโs experiment .
In case of a smooth dielectric density we can regard $`\epsilon /\mu `$ as a constant, and obtain from Maxwellโs equations
$$_\alpha \left(\sqrt{g_F}F^{(\alpha )(\beta )}\right)=0\text{or}D_\alpha ^FF^{(\alpha )(\beta )}=0.$$
(73)
Light experiences the quantum dielectric as the spaceโtime metric (1), i.e. as an effective gravitational field.
### E Energyโmomentum tensor
According to Antoci and Mihich Gordon has already settled the notorious debate about Minkowskiโs versus Abrahamโs energyโmomentum tensor in Abrahamโs favor. However, in his paper , Gordon assumed the dielectric properties of the medium $`\epsilon `$, $`\mu `$, and $`u^\alpha `$, as preassigned quantities. Having done so, the derived energyโmomentum tensor is valid if and only if the dielectric quantities are constants, i.e. in the case of a uniform medium, because the conservation of energy and momentum presupposes the homogeneity of spaceโtime, according to Noetherโs theorem. If one tries to determine the energy and momentum of the electromagnetic field in an inhomogeneous medium one must not consider the dielectric properties as given functions, but rather as being generated by a physical object, such as the quantum dielectric studied in this paper. In short, one should take into account actio et reactio, and in particular the back action of the medium (an effect seen experimentally ). Does Abrahamโs tensor have significance beyond uniform media?
Let us determine the energyโmomentum tensor via the royal road of general relativity, as a variation of the Lagrangian with respect to the backโground metric ,
$$T^{\mu \nu }=\frac{2}{\sqrt{g}}\frac{\delta \left(\sqrt{g}\right)}{\delta g_{\mu \nu }}=2\frac{\delta }{\delta g_{\mu \nu }}g^{\mu \nu }.$$
(74)
A metric variation $`\delta _g`$ of the Lagrangian gives, in analogy with Eq. (54) and the considerations in Sec. VB,
$`\delta _g`$ $`=`$ $`\delta _g_F{\displaystyle \frac{mc^2}{2}}\rho u^\alpha u^\beta \delta _gg_{\alpha \beta }^A`$ (75)
$`=`$ $`\delta _g_{\mathrm{EFF}}{\displaystyle \frac{mc^2}{2}}\rho u^\alpha u^\beta \delta _gg_{\alpha \beta }.`$ (76)
We recall that $`_A`$ vanishes in the hydrodynamic limit. Consequently, we arrive at the total energyโmomentum tensor in the form
$$T^{\mu \nu }=2\frac{\delta _{\mathrm{EFF}}}{\delta g_{\mu \nu }}_Fg^{\mu \nu }+mc^2\rho u^\mu u^\nu .$$
(77)
We represent this expression as the sum
$$T^{\mu \nu }=T_A^{\mu \nu }+T_{\mathrm{EFF}}^{\mu \nu }$$
(78)
with the atomic component
$`T_A^{\mu \nu }`$ $`=`$ $`mc^2\rho u^\mu u^\nu pg^{\mu \nu },`$ (79)
$`p`$ $`=`$ $`_F_{\mathrm{EFF}}={\displaystyle \frac{1}{4}}F_{\alpha \beta }\left(H^{\alpha \beta }\epsilon _0F^{\alpha \beta }\right),`$ (80)
and
$$T_{\mathrm{EFF}}^{\mu \nu }=2\frac{\delta _{\mathrm{EFF}}}{\delta g_{\mu \nu }}_{\mathrm{EFF}}g^{\mu \nu }.$$
(81)
We are inclined to interpret the tensor (81) as the effective energyโmomentum tensor of the electromagnetic field in the presence of a dielectric medium.
The atomic tensor (79) appears as the energyโmomentum of a fluid under the dielectric pressure (80). In the limit of low flow velocities the pressure approaches $`\epsilon _0(\alpha _EE^2+\alpha _Bc^2B^2)\varrho /2`$, according to Eqs. (71) and (72). In this limit, atomic dipoles with positive $`\alpha _E`$ and $`\alpha _B`$ are attracted towards increasing field intensities. We also see from the atomic energyโmomentum tensor (79) that a dielectric fluid possesses the total enthalpy density $`mc^2\rho =mc^2w\varrho `$, including the relativistic rest energy. In this way we have found an interpretation for the density $`\rho `$ that appears at the prominent place (3). To calculate the enthalpy, we express the effective Lagrangian (55) in terms of the norm $`w`$. We use the definition (43) of the fourโvelocity $`u^\alpha `$ and the normalization (49) of the $`w^\alpha `$, and obtain
$$p=_F_{\mathrm{EFF}}=\frac{mc^2\varrho }{2}\left(\frac{1}{w}w\right),$$
(82)
or, by inversion,
$$mc^2\rho =mc^2w\varrho =\sqrt{m^2c^4\varrho ^2+p^2}p.$$
(83)
This equation describes how the enthalpy density depends on the pressure and on the dielectric density. On the other hand, Eq. (80) quantifies the pressure that depends on the dielectric density and flow, and on the electromagnetic field as an external quantity. We may interpret the two formulas (80) and (83) as the equations of state for the quantum dielectric. The density of the fluidโs internal energy is the difference between enthalpy density and pressure
$$ฯต=\sqrt{m^2c^4\varrho ^2+p^2}2p.$$
(84)
We see that the internal energy approaches $`mc^2+\epsilon _0(\alpha _EE^2+\alpha _Bc^2B^2)`$ in the limit of a slow flow and a low dielectric pressure. Atomic dipoles with positive $`\alpha _E`$ and $`\alpha _B`$ seem to gain internal energy in the presence of an electromagnetic field.
Let us turn to the energyโmomentum tensor of the field. The effective Lagrangian $`_{\mathrm{EFF}}`$ characterizes a medium with preassigned dielectric functions $`\epsilon `$ and $`\mu `$, i.e. Gordonโs case . Consequently , the effective energyโmomentum tensor of the electromagnetic field is Abrahamโs
$$T_{\mathrm{EFF}}^{\mu \nu }=T_{\mathrm{Ab}}^{\mu \nu }=T_{\mathrm{Mk}}^{\mu \nu }(\epsilon \mu 1)u^\mu \mathrm{\Omega }^\nu ,$$
(85)
with Minkowskiโs tensor ,
$$T_{\mathrm{Mk}}^{\mu \nu }=H^{\mu \alpha }F_{\alpha \beta }g^{\beta \nu }+\frac{1}{4}H^{\alpha \beta }F_{\alpha \beta }g^{\mu \nu },$$
(86)
corrected by the Ruhstrahl
$$\mathrm{\Omega }^\nu =F_{\alpha \alpha ^{}}u^\alpha ^{}u_\beta \left(H^{\alpha \beta }u^\nu +H^{\beta \nu }u^\alpha +H^{\nu \alpha }u^\beta \right).$$
(87)
In locally coโmoving galilean coordinates or in a medium at rest, the spatial component of the Ruhstrahl is proportional to the Poynting vector (hence the name),
$$\mathrm{\Omega }^\nu =(0,\frac{๐๐}{c}).$$
(88)
In this case the effective energyโmomentum tensor of the field takes the form
$$T_{\mathrm{Ab}}^{\mu \nu }=\left(\begin{array}{cc}I& ๐/c\\ ๐/c& \sigma \end{array}\right)$$
(89)
with intensity $`I`$, Poynting vector $`๐`$, and stress tensor $`\sigma `$
$`I`$ $`=`$ $`{\displaystyle \frac{๐๐}{2}}+{\displaystyle \frac{๐๐}{2}},๐=๐๐,`$ (90)
$`\sigma `$ $`=`$ $`\left({\displaystyle \frac{๐๐}{2}}+{\displaystyle \frac{๐๐}{2}}\right)\mathrm{๐}๐๐๐๐.`$ (91)
We see that Abrahamโs tensor describes indeed the effective energyโmomentum of the electromagnetic field, even in the general case of a nonโuniform medium that is able to move under the pressure of light forces.
## VI Credo
Light experiences dielectric matter as an effective gravitational field and matter experiences light as a form of gravity as well. Light and matter see each other as dual spaceโtime metrics, a unique model in field theory, to the knowledge of the author. We have solidified this mental picture by postulating the idea and demonstrating its striking consistency with the theory of dielectrics . It would be interesting to see whether our model can be derived directly from first principles. In passing, we have determined the energyโmomentum tensor that governs actio et reactio of electromagnetic fields in quantum dielectrics. The tensor is Abrahamโs plus the energyโmomentum of the medium characterized by a dielectric pressure and an enthalpy density.
Our idea may serve as a guiding line for understanding the effects of slow light on matter waves. Here one can conceive of creating light fields that appear to atoms as quasiโastronomical objects. The holy grail in this field would be the creation of a black hole made of light.
Light and matter interact with each other as if both were gravitational fields, and light and matter are genuine quantum fields in Nature. A distinct quantum regime of dielectrics has been prepared in the laboratories where BoseโEinstein condensates of alkali vapors interact nonโresonantly with light quanta, but has never been viewed as an analogue of quantum gravity, to the knowledge of the author. Sound in superfluids and in alkali BoseโEinstein condensates has been considered as a quantum field in a curved spaceโtime, as being able to emit the acoustic analogue of Hawking radiation . However, the quantum sound still propagates in a classical medium, in contrast to light quanta in a quantum dielectric. In many respects, we have reasons to hope that BoseโEinstein condensates may serve as testable prototype models for quantum gravity.
## Acknowledgements
I am very grateful to Sir Michael Berry, Ignacio Cirac, Carsten Henkel, Susanne Klein, Rodney Loudon, Paul Piwnicki, Stig Stenholm, and Martin Wilkens for conversations on moving media. I acknowledge the generous support of the Alexander von Humboldt Foundation and of the Gรถran Gustafsson Stiftelse. |
warning/0001/hep-ph0001252.html | ar5iv | text | # 1 Introduction
## 1 Introduction
We canโt say that the complete knowledge on beauty quark pair production is obtained now, since the data we have on $`b\overline{b}`$ production cross sections are not yet sufficient. In order to monitor the model ideas on the phenomenon it seems useful to revise once in a while the collected data. Recently the results of several experiments carried out at two energies of colliding protons, 630 GeV and 1.8 TeV, are represented in the literature.
In this article the two of the models are compared: on the one hand - the phenomenological Quark-Gluon String Model , based on idea of hadronic amplitude duality and on the theory of supercritical Pomeron , on the other hand - the wide-spread Monte Carlo code PYTHIA , which includes the results of QCD perturbative diagram calculations.
The production cross sections arising with energy is a fact which was widely discussed in recent studies . The theory of supercritical Pomeron postulates the rising as $`s^{\mathrm{\Delta }_P\left(0\right)}`$, where $`\mathrm{\Delta }_P\left(0\right)=\alpha _P\left(0\right)1`$, $`\alpha _P\left(0\right)`$ is the intercept of Regge trajectory of Pomeron.
The energy behavior of the production cross section in QCD perturbative approach are provided by the choise of gluon structure functions of interacting hadrons. Most of those functions, which are accepted now for MC simulations of high energy collisions, are built to approximate the recent data from HERA measured up to $`x=10^4`$. It should be noticed, that all known gluon structure functions which satisfied this recent data, can be taken for the modelling of $`b\overline{b}`$ production at LHC, because the value $`10^4`$ is the very region of x attributed for B meson production at 14 TeV due to the following estimation: $`2m_B/\left(14TeV\right)10^4`$. One of those appropriate gluon distributions is CTEQ structure function, which is involved into PYTHIA code as default distribution.
## 2 Parameters defining the B-meson production cross sections in QGSM.
The major parameter of QGSM which defines the cross section dependence on energy is $`\mathrm{\Delta }_P\left(0\right)_{eff}`$, which have to be less than the same value for one Pomeron exchange, because multipomeron diagrams or branches should be taken into account in the calculations. This parameter depends on the mean value of transverse momenta transmitted in the process. Therefore the energy dependences of different mass particle production cross section must be led by different $`\mathrm{\Delta }_P\left(0\right)_{eff}`$.
The $`\mathrm{\Delta }_P\left(0\right)\left(Q^2\right)`$-dependence can be illustrated well with the data obtained in H1 experiment at HERA . The Pomeron exchange plays an important role in electron-proton collisions too. The multyhadron production takes place in this process due to the Pomeron exchange between photon and beam proton. The $`\mathrm{\Delta }\left(Q^2\right)`$ data are shown in Fig. , where each point was extracted from measured $`F_2(x,Q^2)`$ function by the approximation with the simple dependence: $`F_2x^{\mathrm{\Delta }\left(Q^2\right)}`$.
It should be noticed that the theoretical curve was calculated under the assumption of one Pomeron exchange which doesnโt include the branches as against the proton-proton interaction. The QGSM scheme for heavy meson production have to be similar to the one Pomeron exchange pattern due to the energy conservation reasons. So we can take for the $`\mathrm{\Delta }_{eff}`$ the value 0,3 corresponding to the H1 data approximation at $`Q^2=\left(2m_B\right)^2`$. Itโs worth beeng stressed here, that this value differs from $`\mathrm{\Delta }_{eff}`$=0.12 which was chosen for light quark meson production in early papers .
Another Regge trajectory parameter is important for model description of inclusive cross sections of B meson pair production, itโs $`\alpha _\mathrm{{\rm Y}}\left(0\right)`$, the intercept of $`b\overline{b}`$-trajectory. It provides the increasing cross section at the energy region close after the B pair production threshould. There are various opinions about the value of this parameter. From the point of view of QGSM it may vary in the range of -16$`รท`$0 . The other athors prefere to take $`\alpha _\mathrm{{\rm Y}}\left(0\right)=9`$ .
The value $`\alpha _\mathrm{{\rm Y}}\left(0\right)=16`$ will be taken here for to estimate the upper limit of growing cross section when it increases rapidly after the threshould. The parameter discussed above exists in the functions of fragmentation of qurk-gluon strings into each sort of B-mesons. Those functions are written in QGSM according to the rules fulfilled by the Regge asympthotics .
For example, the function for d quark string fragmentation into $`B^+`$ contains the following factors:
$$๐_d^{B^+}\left(z\right)=\frac{a_0^B}{z}\left(1z\right)^{\alpha _\mathrm{{\rm Y}}\left(0\right)\alpha _\mathrm{{\rm Y}}\left(0\right)+\lambda }\left(1+a_1^Bz^2\right),$$
where $`a_0^B`$ is the density parameter for fragmentation of quark-gluon string into B-mesons. The $`a_1^D`$ is the parameter of string fragmentation asymmetry introduced in to provide a transition between probabilities of the B production at z$``$0 and z$``$1.The value $`a_1^B`$ can be of the order 10 and actually doesnโt impact on the value of B production cross section at energies higher than 1.8 TeV.
The calculations of $`p_{}`$ spectra of produced hadrons are also available in the framework of QGSM, as it had been done already in for $`\pi `$-mesons. The spectra can be described up to the momenta of order few GeV/c in this substantially nonperturbative model . The distributions were of exponential character at low $`p_{}`$ in this approach. Therefore the transverse momenta distributions for heavy flavor particles was not elaborated in this model.
## 3 PYTHIA machinery
The version PYTHIA 5.7 was taken to calculate the spectra of B-mesons at three energies of colliding protons: 630 GeV, 1,8 TeV and 14 TeV. The CTEQ gluon structure function are used in this version to discribe the increasing cross sections. On the one hand the process of production of such heavy quarks as b is good enough for beeing described by perturbative QCD diagram with gluon-gluon fusion. On the other hand, more and more low x gluons are involved into this process at energy rising and cross section becomes dependent on the accuracy of gluon structure function measured at low x.
As it was mentioned in Introduction, we have precise data on $`F_2`$ due to HERA experiments up to $`x10^4`$, what is enough for the calculation of $`b\overline{b}`$ production at LHC energies. Such a way CTEQ structure functions have to provide the right description of increasing cross sections of $`b\overline{b}`$ pair production.
However, b quarks can be obtained not only in gluon fusion process. Two additional ways exist to produce $`b\overline{b}`$ pair, they are: gluon splitting $`\mathrm{๐๐}\mathrm{๐๐}`$, where gluons gives $`b\overline{b}`$ pair in the next- to-leading order of corrections, and heavy flavor exitation $`Q_igQ_ig`$. In PYTHIA this subprocesses are taken into account with massless matrix elements. It is a problem how to sum the resulting distributions from such different deposits.
It makes the $`p_{}`$-spectra at 1.8 Tev comparable with the data obtained in CDF experiment ( see Fig.). At the same time there is not good description of UA1 data . It looks like the $`p_{}`$ spectra were increased with additional fractions only by a factor and there is not any difference between the patterns of spectra for different subprocesses. It leads to rather flat form of transverse momentum distributions in PYTHIA and to small total cross section of B production at various energies.
## 4 Comparison of cross section energy dependences
The resulting energy dependences of production cross section are shown on Fig. for PYTHIA program and for QGS model as well.
As it was mentioned above the QGSM curve is highest as it is possible in this model after the normalisation to the CDF cross section. But latter was obtained due to PYTHIA (see Fig. ).Thus the point where both curves are crossing is rather conventional and depends on the form of $`p_{}`$-spectra at low transverse momenta accepted in PYTHIA.
## 5 PYTHIA and QGSM predictions for the asymmetry between $`B^0/\overline{B^0}`$ spectra
It would be interesting to consider the leading effect in the spectra of B-mesons at various energies. The valuable asymmetry between $`B^0`$\- and $`\overline{B^0}`$-meson spectra at LHC energy will be important for CP violation measurements.The recent prediction of the y dependence of such leading/nonleading asymmetry provided with PYTHIA simulations gives zero value of A(y) in wide range of y at the central region (see Fig. ).
The A(y) dependence in fragmentation region $`x_F`$ 1 contradicts with all similar asymmetry measurements for D-meson spectra . The intersection of inclusive spectra of different type of B-mesons gives the asymmetry passing trough zero at some $`y_0`$, while the measured spectra of leading particles are always higher then nonleading particle spectra, so the asymmetry is positive value. In opposite to PYTHIA predictions, the asymmetry calculated in the framework of QGSM is rising function up to $`x_F`$ 1. This behavior is usually peculiar for the string approach because of so called โbeam dragโ effect. The valuable asymmetry in central region given in QGSM prediction is not enough small for not to be taken into account at CP violation measurements. It looks important to consider both these predictions in details and to discuss the probability of nonzero asymmetry in the production spectra at LHC energy.
## 6 Conclusions
We have compared two approaches for the understanding of the heavy flavored particle production: one of them is mostly perturbative and another one is nonperturbative at all. This comparison shows that some different suggestion has to be done for low transverse momenta distributions of B-mesons to put into agreement the both model predictions at LHC energy. The contradicting dependences for $`B^0/\overline{B^0}`$ asymmetry in the B meson production spectra might be important for the CP violation measurements.
## 7 Acknowledgments
I would like to express my gratitude to A.B.Kaidalov, A.Kharchilava and S.Baranov for numerous discussions. This work was supported by DFG grant $`N^{}`$ 436 RUS 113/332/O(R). |
warning/0001/cond-mat0001081.html | ar5iv | text | # Transforming Gaussian diffusion into fractional, a generalized law of large numbers approach
## I Introduction
The continuous time random walk (CTRW), introduced by Montroll and Weiss , describes different types of diffusion processes including the standard Gaussian diffusion, sub diffusion and Lรฉvy walks . The CTRW has been a useful tool in many diverse fields and over the last three decades e.g., it was used to describe transport in disorder medium and in low dimensional chaotic systems . More recently fractional kinetic equations were investigated as a tool describing phenomenologically anomalous diffusion . In these equations fractional derivatives replace ordinary integer derivatives in the standard integer kinetic equation. It is believed that fractional calculus can be used to model non integer diffusion phenomena. Schneider and Wyss have formulated the fractional diffusion equation (see Sec. III) and under certain conditions the fractional diffusion equation describes the asymptotic large time behavior of the decoupled sub-diffusive CTRW (with $`r^2t^\delta `$; $`\delta <1`$). Thus some what like ordinary random walks which can be described asymptoticly by the diffusion equation, so can the CTRW be described by fractional diffusion equation.
Recently, Metzler, Barkai and Klafter have investigated fractional FokkerโPlanck equation (FFPE) to describe an anomalous sub-diffusion motion in an external nonโlinear force field. In the absence of the external force field the FFPE reduces to the fractional diffusion equation. The FFPE is an asymptotic equation which extends the CTRW to include the effect of an external force field. The CTRW itself was not designed for a description of a particle in external force field thus the FFPE describes new type of behavior not explored in depth yet.
The main result in this paper concerns a transformation of ordinary Gaussian diffusion into fractional diffusion . Let $`P_\alpha (๐ซ,t)`$ be the propagator (Greenโs function) of fractional FokkerโPlanck equation (a special case is the fractional diffusion equation), here $`0<\alpha <1`$ is the fractional exponent and $`\alpha =1`$ is the standard case described by ordinary Fokker-Planck equation. Then the solution, $`P_\alpha (๐ซ,t)`$, is found based on the transformation
$$P_\alpha (๐ซ,t)=_0^{\mathrm{}}\left[\frac{dN(s,t)}{ds}\right]P_1(๐ซ,s)๐s,$$
(1)
and
$$N(s,t)=1L_\alpha \left(\frac{t}{s^{1/\alpha }}\right)$$
(2)
denotes the โinverseโ one sided Lรฉvy stable distribution (i.e., $`L_\alpha (x)`$ is the one sided Lรฉvy stable distribution). For example consider the force free case. The solution of the FFPE $`P_\alpha (๐ซ,t)`$, with initial conditions concentrated on the origin, is found by transforming $`P_1(๐ซ,s)`$ and the transformed function is the well known Gaussian solution of the integer diffusion equation with initial conditions concentrated on the origin. The transformation Eq. (1) besides its practical value for finding solutions of the FFPE also explains its meaning and relation to the CTRW (see details below).
We list previous work on the integral transformation Eq. (1)
1 Bouchaud and Georges for the asymptotic long time limit of the CTRW, then $`P_1(๐ซ,s)`$ is Gaussian,
2 Zumofen and Klafter for CTRW on a lattice, then $`P_1(๐ซ,s)`$ is solution of an ordinary random walk on a lattice,
3 Saichev and Zaslavsky for one dimensional solution of fractional diffusion equation, they have also considered an extension which includes Lรฉvy flights,
4 Barkai and Silbey for the fractional Ornstein-Uhlenbeck process.
We generalize these results for the dynamics described by the FFPE, and investigate in greater detail Eq. (1) in the context of CTRW and fractional diffusion equation in dimensions $`d=1,2,3`$.
This paper is organized as follows. In Sec. (II) we follow and derive the transformation Eq. (1) based on the CTRW. We find an exact solution to the CTRW valid for short and long times and compare this solution with the result obtained from Eq. (1). In Sec. (III) we consider the fractional diffusion equation. We investigate its integral solution Eq. (1) as well as the Fox function solution found in . In Sec. (IV) we show how to obtain solutions of the FFPE using Eq. (1). We give as examples the motion of a particle in a uniform and harmonic force fields. In Sec. (V) a brief summary is given.
## II Continuous Time Random Walk
In the decoupled version of the continuous time random walk (CTRW), a random walker hops from site to site and at each site it is trapped for a random time . For this well known model two independent probability densities describe the random walk. The first is $`\psi (t)`$, the probability density function (PDF) of the independent identically distributed (IID) pausing times between successive steps. The second is the PDF $`f\left(r\right)`$ for the IID displacements of the random walker at each step. Thus the CTRW describes a process for which the particle is trapped on the origin for time $`\tau _1`$, it then jumps to $`๐ซ_\mathrm{๐}`$, it is trapped on $`๐ซ_\mathrm{๐}`$ for time $`\tau _2`$ and then it jumps to a new location, the process is then renewed. In what follows we assume $`f(r)`$ has a finite variance and a zero mean. The asymptotic behavior of the decoupled CTRW is well investigated and we now summarize some results from the CTRW literature which are of relevance to our work.
Let $`P(r,t)`$ be the PDF of finding the CTRW particle at $`r`$ at time $`t`$. Let $`N_{CT}(s,t)`$ be the probability that $`s`$ steps are made in the time interval $`(0,t)`$ so clearly
$$\underset{s=0}{\overset{\mathrm{}}{}}N_{CT}(s,t)=1$$
(3)
and the subscript CT denoted the CTRW. Because the model is decoupled
$$P(r,t)=\underset{s=0}{\overset{\mathrm{}}{}}N_{CT}(s,t)W(r,s),$$
(4)
and $`W(r,s)`$ is the probability density that a particle has reached $`r`$ after $`s`$ steps. In what follows we shall use the Fourier $`(๐ซ๐ค)`$ and Laplace $`(tu)`$ transforms, we use the convention that the argument of a function indicates in which space the function is defined, e.g.,
$$P(r,u)=\underset{s=0}{\overset{\mathrm{}}{}}N_{CT}(s,u)W(r,s),$$
(5)
is the Laplace transform of $`P(r,t)`$.
$`W(r,s)`$ will generally depend on $`f(r)`$, however we are interested only in the large time behavior of $`P(r,t)`$ meaning that only contributions from large $`s`$ are important. From standard central limit theorem we know
$$W(r,s)_s\mathrm{}G(r,s)=\frac{1}{\left(4\pi s\right)^{d/2}}\mathrm{exp}\left(\frac{r^2}{4s}\right).$$
(6)
We have used the assumption that the system is unbiased and use convenient units. For most systems and for large times $`N_{CT}(s,t)`$ is concentrated on $`s=t/\tau _{av}`$ and $`\tau _{av}=_0^{\mathrm{}}t\psi (t)๐t`$ is the averaged pausing time. In this case $`P(r,t)`$ will become Gaussian when $`t\mathrm{}`$. When $`N_{CT}(s,t)`$ is broad a non-Gaussian behavior is found. This is the case when $`\tau _{av}=\mathrm{}`$ or in other words when the Laplace transform of $`\psi (t)`$; $`\psi (u)`$ behaves as
$$\psi \left(u\right)1u^\alpha +\mathrm{}0<\alpha 1,$$
(7)
and we have used convenient units. Shlesinger showed that in this case an anomalous sub-diffusive behavior is found, $`r^2t^\alpha `$. Because the steps are independent and based on convolution theorem of Laplace transform
$$N_{CT}(s,u)=$$
$$\frac{1\psi \left(u\right)}{u}\mathrm{exp}\left\{s\mathrm{ln}\left[\psi \left(u\right)\right]\right\}$$
$$u^{\alpha 1}\mathrm{exp}\left(su^\alpha \right),$$
(8)
and $`N_{CT}(s,u)`$ is the Laplace transform of $`N_{CT}(s,t)`$.
Following we replace the summation in Eq. (5) with integration and use Eq. (6) to find
$$P(r,u)_0^{\mathrm{}}n(s,u)G(r,s)๐s$$
(9)
and according to Eq. (8)
$$n(s,u)u^{\alpha 1}\mathrm{exp}\left(su^\alpha \right).$$
(10)
Using the inverse Laplace transform we find
$$P(r,t)_0^{\mathrm{}}n(s,t)G(r,s)๐s$$
(11)
and $`n(s,t)`$ is the inverse Laplace transform of $`n(s,u)`$
$$n(s,t)=\frac{1}{\alpha }\frac{t}{s^{1+1/\alpha }}l_\alpha \left(\frac{t}{s^{1/\alpha }}\right)$$
(12)
$`l_\alpha \left(z\right)`$ in Eq. (12) is a one sided Lรฉvy stable probability density whose Laplace transform is
$$l_\alpha \left(u\right)=_0^{\mathrm{}}e^{ux}l_\alpha \left(x\right)๐x=e^{u^\alpha }.$$
(13)
According to Eq. (11) the large time behavior of the CTRW solution is reached using an integral transformation of the Gaussian solution of ordinary diffusion processes \[i.e., of $`G(r,s)`$\]. As we shall see similar transformations can be used to solve the FFPE and in particular the fractional diffusion equation.
The kernel $`n(s,t)`$ is a non negative PDF normalized according to
$$_0^{\mathrm{}}n(s,t)๐s=1,$$
(14)
it replaces the CTRW probability $`N_{CT}(s,t)`$. Notice that the PDF $`n(s,t)`$, like $`N_{CT}(s,t)`$, is independent of the dimensionality of the problem $`d`$. Some properties of $`n(s,t)`$ are given in . All moments of $`n(s,t)`$ are finite and are given later in Eq. (87). It is easy to show that $`n(s,t)=[dN(s,t)/ds]`$ with $`N(s,t)`$ defined in Eq. (2). $`n(s,t)`$ is called the โinverseโ one sided Lรฉvy stable probability density .
When $`\alpha =1`$, we find $`n(s,t)=\delta (st)`$ and hence $`P(r,t)`$ is Gaussian. This is expected since when $`\alpha =1`$ the first moment of pausing times $`\tau _{av}`$ is finite, therefore the law of large numbers is valid, and hence we expect that the number of steps $`s`$ in the random walk scheme will follow $`st/\tau _{av}`$. When $`\alpha <1`$ the law of large numbers is not valid and instead the random number of steps $`s`$ is described by $`n(s,t)`$. Thus the transformation, Eq. (11), has a meaning of a generalized law of large numbers.
Eq. (12) gives the kernel $`n(s,t)`$ in terms of a one sided Lรฉvy stable density. Schneider has expressed Lรฉvy stable densities in terms of a Fox H function
$$l_\alpha (z)=\frac{1}{\alpha z^2}H_{1,1}^{1,0}\left[z^1|\begin{array}{c}(1,1)\\ (1/\alpha ,1/\alpha )\end{array}\right].$$
(15)
Asymptotic behaviors of the one sided Lรฉvy density can be found in Fellerโs book or based on the asymptotic behaviors of the H Fox function . For the cases $`\alpha =1/3,1/2,2/3`$, closed form equations in terms of known functions, may be found in (notice that points out relevant errors in the literature on Lรฉvy stable densities).
Important for our purposes is the result obtained by Tunaley already in $`1974`$ which expressed the asymptotic behavior of the CTRW solution $`P(r,t)`$, in terms of its FourierโLaplace transform, as shown in
$$P(๐ค,u)\frac{u^{\alpha 1}}{u^\alpha +๐ค^2}.$$
(16)
In Appendix A we verify that Eq. (16) is indeed the FourierโLaplace transform of Eq. (11). The inversion of equation (16) was accomplished by Tunaley in one dimension and by Schneider and Wyss in dimensions two and three (see more details below).
### A CTRW Solution
Let us consider an example of the (decoupled) CTRW process. The solution of the CTRW in $`๐ค,๐ฎ`$ space is
$$P(๐ค,u)=\frac{1\psi (u)}{u}\frac{1}{1\psi (u)f(๐ค)}.$$
(17)
Usually CTRW solutions are found based on numerical inverse FourierโLaplace transform of Eq. (17). Here we find an exact solution of the CTRW process for a special choice of $`f(r)`$ and $`\psi (t)`$. Our solution is an infinite sum of well known functions. We assume the PDF of jump times $`\psi (t)`$ to be one sided Lรฉvy stable density with $`\psi (u)=\mathrm{exp}(u^\alpha )`$ . Displacements are assumed to be Gaussian and then
$$W(r,s)=\frac{1}{\left(4\pi s\right)^{d/2}}\mathrm{exp}\left(\frac{r^2}{4s}\right).$$
(18)
is exact, not only asymptotic. For this choice of PDFs the solution of the CTRW can be found explicitly. We use
$$N_{CT}(s,u)=\frac{1\mathrm{exp}(u^\alpha )}{u}\mathrm{exp}\left(su^\alpha \right)$$
(19)
and the convolution theorem of Laplace transform to find
$$N_0(t)=1L_\alpha \left(t\right)$$
$$N_{CT}(s,t)=L_\alpha \left(\frac{t}{s^{1/\alpha }}\right)L_\alpha \left(\frac{t}{(s+1)^{1/\alpha }}\right)$$
(20)
and
$$L_\alpha \left(t\right)=_0^tl_\alpha (t)๐t$$
(21)
is the one sided Lรฉvy stable distribution. Inserting Eqs. (18), (20) in Eq. (4) we find
$$P(r,t)=\left[1L_\alpha \left(t\right)\right]\delta \left(r\right)+$$
$$\underset{s=1}{\overset{\mathrm{}}{}}\{L_\alpha \left[\frac{t}{s^{1/\alpha }}\right]L_\alpha \left[\frac{t}{\left(s+1\right)^{1/\alpha }}\right]\}\times $$
$$\frac{1}{\left(4\pi s\right)^{d/2}}\mathrm{exp}\left(\frac{r^2}{4s}\right).$$
(22)
The first term on the right hand side describes random walks for which the particle did not leave the origin within the observation time $`t`$, the other terms describe random walks where the number of steps is $`s`$. In Fig. 1 we show the solution of the CTRW process in a scaling form. We consider a three dimensional case, $`\alpha =1/2`$ and use
$$L_{1/2}\left(t\right)=\frac{1}{\sqrt{\pi }}\mathrm{\Gamma }[1/2,1/(4t)].$$
(23)
Here $`\mathrm{\Gamma }(,)`$ denotes the incomplete Gamma function. The figure shows $`r^3P(r,t)`$ versus the scaling variable $`\xi =r^2/t^\alpha `$. For large times $`t`$ the solution converges to the asymptotic solution found based on the integral transformation Eq. (11).
Let us calculate the Cartesian moments
$$M(2m_1,\mathrm{},2m_d)=$$
$$_{\mathrm{}}^{\mathrm{}}๐x_1\mathrm{}_{\mathrm{}}^{\mathrm{}}๐x_dx_1^{2m_1}\mathrm{}x_d^{2m_d}P(x_1,\mathrm{},x_d,t)$$
(24)
with non negative integers $`m_1,m_2,\mathrm{}`$. Clearly the odd moments are equal zero and from normalization $`M(0,\mathrm{},0)=1`$. In Appendix B we find
$$M(2m_1,\mathrm{},2m_d)=$$
$$C_{m,d}\underset{s=1}{\overset{\mathrm{}}{}}\left\{L_\alpha \left[\frac{t}{s^{1/\alpha }}\right]L_\alpha \left[\frac{t}{\left(s+1\right)^{1/\alpha }}\right]\right\}s^m$$
(25)
with
$$C_{m,d}=\frac{2^{2m}}{\pi ^{d/2}}\mathrm{\Pi }_{i=1}^d\mathrm{\Gamma }\left(m_i+\frac{1}{2}\right)$$
(26)
and $`m=_i^dm_i>0`$. In the Appendix we also show that for $`t\mathrm{}`$
$$M(2m_1,\mathrm{},2m_d)$$
$$C_{m,d}\mathrm{\Gamma }(1+m)\frac{t^{\alpha m}}{\mathrm{\Gamma }\left(1+\alpha m\right)}.$$
(27)
To derive Eq. (27) we used the small $`u`$ expansion of the Laplace transform of Eq. (25) and Tauberian theorem. Later we shall show that the moments in Eq. (27) are identical to the moments obtained directly from the integral transformation Eq. (11). Hence our interpretation of Eq. (11) as the asymptotic solution of the CTRW is justified, \[however our derivation of Eq. (27), is based on a specific choice of $`\psi (t)`$ and $`f(r)`$\].
## III Fractional Diffusion Equation
The fractional diffusion equation describes the asymptotic behavior of the CTRW Eq. (11). Balakrishnan has derived a fractional diffusion process based upon a generalization of Brownian motion in one dimension. Schneider and Wyss have formulated the following fractional diffusion equation describing this process
$$\frac{P(r,t)}{t}=_0D_t^{1\alpha }^2P(r,t),$$
(28)
and $`0<\alpha <1`$. The fractional Riemann Lioville derivative $`{}_{0}{}^{}D_{t}^{1\alpha }`$ in Eq. (28) is defined
$${}_{0}{}^{}D_{t}^{1\alpha }Z\left(t\right)=\frac{1}{\mathrm{\Gamma }(\alpha )}\frac{}{t}_0^t๐t^{}\frac{Z(t^{})}{(tt^{})^{1\alpha }}.$$
(29)
The FourierโLaplace solution of the fractional diffusion equation, with initial conditions $`P(๐ซ,t=0)=\delta (๐ซ)`$,
$$P(๐ค,u)=\frac{u^{\alpha 1}}{u^\alpha +๐ค^2}$$
(30)
is identical to the right hand side of Eq. (16). Therefore the integral solution of fractional diffusion equation is
$$P(r,t)=_0^{\mathrm{}}n(s,t)G(r,s)๐s.$$
(31)
Thus solution Eq. (11) which was only asymptotic in the CTRW framework is exact within the fractional diffusion equation approach. The integral solution Eq. (31) for $`d=1`$ was investigated by Saichev and Zaslavsky, we investigate also the cases $`d=2,3`$ which exhibit behaviors different than the one dimensional case.
As mentioned, Schneider and Wyss have inverted Eq. (30) finding the solution to the fractional diffusion equation in terms of rather formal Fox function. Later we shall show that moments of $`P(r,t)`$ calculated based on the integral solution Eq. (31) in dimensions $`d=1,2,3`$ are identical to the moments calculated based upon the Fox function solution. In this sense we show that the two solutions are identical.
At this stage we have only shown that the two approaches, asymptotic CTRW and fractional diffusion equation are identical. The reader might be wondering why should we bother with the fractional equation if the CTRW approach gives identical results. The situation is some what similar to the standard diffusion equation which predicts a Gaussian evolution. The diffusion equation $`\alpha =1`$ is an asymptotic equation describing much more general random walks. The main advantage of the diffusion equation over a random walk approach is its simplicity. Solutions with special boundary conditions (reflecting/absorbing) are relatively simple and usually capture the essence of the more complex random walks. Another extension of the diffusion equation is the diffusion in external field as described by FokkerโPlanck equations, such an extension for random walks is cumbersome. The same is true for the fractional diffusion equation. It can serve as a phenomenological tool describing anomalous diffusion. As we shall show in the next section we can include the effects of an external field. On the other hand CTRW by definition is not built to consider an external field.
### A Integral Solution
We first notice that the integral solution Eq. (31) shows that $`P(r,t)`$ is normalized and non-negative. The normalization is easily seen with the help of Eq. (14) and the non negativity of $`P(r,t)`$ is evident because both $`n(s,t)`$ and $`G(r,s)`$ are non negative.
Calculation of spherical moments $`r^n(t)`$ is now considered. The calculation follows two steps, the first is to calculate $`r^n(s)`$ for Gaussian diffusion (or find the Gaussian moments in a text book) then using the transformation defined in Eq. (31) we find the moments $`r^n(t)`$ for the fractional case. More precisely in Laplace $`u`$ space
$$r^n\left(u\right)=_0^{\mathrm{}}๐sn(s,u)\left[G(r,s)r^n๐๐ซ\right],$$
(32)
the Gaussian spherical moments are
$$G(r,s)r^n๐๐ซ=\mathrm{\Omega }_d\mathrm{\Gamma }\left(\frac{n+d}{2}\right)2^ns^{n/2}$$
(33)
with $`\mathrm{\Omega }_1=1/\sqrt{\pi }`$, $`\mathrm{\Omega }_2=1`$ and $`\mathrm{\Omega }_3=2/\sqrt{\pi }`$. The moments for the fractional case are found using Eq. (32)
$$r^n\left(u\right)=$$
$$\mathrm{\Omega }_d2^n\mathrm{\Gamma }\left(\frac{n+d}{2}\right)_0^{\mathrm{}}๐ss^{n/2}n(s,u)=$$
$$\mathrm{\Omega }_d2^n\mathrm{\Gamma }\left(\frac{n+d}{2}\right)\mathrm{\Gamma }(1+n/2)u^{1\alpha n/2}.$$
(34)
Using the inverse Laplace transformation we find
$$r^n\left(t\right)=\mathrm{\Omega }_d\frac{\mathrm{\Gamma }\left(\frac{n+d}{2}\right)2^n\mathrm{\Gamma }\left(1+n/2\right)}{\mathrm{\Gamma }\left(1+\alpha n/2\right)}t^{\alpha n/2}.$$
(35)
For $`n=2`$
$$r^2\left(t\right)=\frac{2dt^\alpha }{\mathrm{\Gamma }\left(1+\alpha \right)}.$$
(36)
When $`\alpha =1`$ the moments in Eq. (35) are identical to the Gaussian moments in Eq. (33).
In Appendix C we calculate the Cartesian moments defined in Eq. (24). We find
$$M(2m_1,\mathrm{},2m_d)=C_{m,d}\frac{\mathrm{\Gamma }\left(1+m\right)}{\mathrm{\Gamma }\left(1+\alpha m\right)}t^{\alpha m}$$
(37)
which is identical to the right hand side of equation (27). Thus the moments of the integral solution of fractional diffusion equation are identical to the asymptotic behavior of the moments of the CTRW found in the previous section.
Eq. (37) was also found in based on Fox function solution of fractional diffusion equation (see details next subsection). Since the integral solution Eq. (31) gives identical moments to those found using the Fox function solution one can assume that the two solutions are identical. It would be interesting to show in a more direct way that the Fox function solution is identical to the integral solution. In Appendix D we use a theorem on Fox functions and attempt to give such a proof. Unfortunately we fail. The interested reader is referred to Appendix D.
### B Fox Function Solution
In , the Mellin transform was used to find the solution of the fractional diffusion equation in terms of Fox H function
$$P(r,t)=\alpha ^1\pi ^{d/2}r^d\times $$
$$H_{12}^{20}\left(2^{2/\alpha }r^{2/\alpha }t^1|\begin{array}{c}(1,1)\hfill \\ (d/2,1/\alpha ),(1,1/\alpha )\hfill \end{array}\right)$$
(38)
The asymptotic expression for this solution is
$$P(r,t)\kappa ^\alpha r^d\xi ^{\frac{d}{2\left(2\alpha \right)}}\mathrm{exp}\left(\lambda _1\xi ^{\frac{1}{2\alpha }}\right),$$
(39)
where $`\xi r^2/t^\alpha `$ is the scaling variable,
$$\kappa ^\alpha =\pi ^{d/2}2^{\frac{d}{2\alpha }}\left(2\alpha \right)^{1/2}\alpha ^{[\alpha (d+1)/21]/(2\alpha )]}$$
(40)
and
$$\lambda _1=\left(2\alpha \right)\alpha ^{\alpha /\left(2\alpha \right)}2^{2/\left(2\alpha \right)}.$$
(41)
Eq. (39) is valid for $`\xi >>1`$. For a brief introduction to Fox functions and its application see .
The behavior for $`\xi <<1`$ is found in Appendix E based on the asymptotic expansion of the $`H`$ function . For dimensions $`d=1`$ we find
$$P(r,t)=\frac{1}{2t^{\alpha /2}}\underset{n=0}{\overset{\mathrm{}}{}}\frac{\left(1\right)^n\xi ^{n/2}}{n!\mathrm{\Gamma }\left[1\alpha (n+1)/2\right]}$$
(42)
and for $`d=3`$
$$P(r,t)=\frac{1}{4\pi t^{3\alpha /2}\xi ^{1/2}}\underset{n=0}{\overset{\mathrm{}}{}}\frac{\left(1\right)^n\xi ^{n/2}}{n!\mathrm{\Gamma }\left[1\alpha (1+n/2)\right]}.$$
(43)
The leading terms in these expansions are for $`d=1`$
$$P(r,t)=\frac{1}{2\mathrm{\Gamma }\left(1\alpha /2\right)}\frac{1}{t^{\alpha /2}}\mathrm{}$$
(44)
and for $`d=3`$
$$P(r,t)=\frac{1}{4\pi \mathrm{\Gamma }\left(1\alpha \right)}\frac{r^1}{t^\alpha }+\mathrm{}.$$
(45)
We see that for $`d=1`$ and $`\alpha =1`$, $`P(r=0,t)=1/(2\sqrt{\pi t})`$, as expected for this normal case. We also see that for $`d=3`$, $`\alpha 1`$ and when $`r0`$ the solution diverges like $`P(r,t)1/r`$. This behavior is not unphysical and $`P(r,t)`$ is normalized according to $`4\pi P(r,t)r^2๐r=1`$.
For $`d=2`$ the asymptotic expansion of the Fox function is not known (see more details in Appendix E). Using Eq. (31) one can show that for $`d=2`$, and $`\xi <<1`$
$$P(r,t)\frac{1}{\pi \mathrm{\Gamma }(1\alpha )t^\alpha }\mathrm{ln}\left[t^{\alpha /2}/r\right].$$
(46)
Eqs. (44-46) were derived independently by A. I. Saichev .
The CTRW behavior on the origin is different than what we have found for the fractional diffusion equation Eqs. (44-46). Within the CTRW on the decay on the origin is described with the first term on the right hand side of Eq. (22), $`\left[1L_\alpha \left(t\right)\right]\delta \left(r\right)`$. This term describes random walks for which the particle is trapped on its initial location during the time of observation $`t`$. This CTRW term decays like $`t^\alpha \delta \left(r\right)`$ for long times and no such singular term is found in the solution of the fractional. In dimensions $`d=2,3`$ the decay of the CTRW singular term is as slow as the decay of $`P(0,t)`$ found from the fractional diffusion equation. Hence on the origin the fractional diffusion approximation does not work well. In contrast, for normal random walks, the singular term decays exponentially with time and then the diffusion approximation works well already after an exponentially short time.
### C Example
Here we use the integral solution of the fractional diffusion equation for a specific example. The Fox function solution are not tabulated and hence from a practical point of view one has to use numerical methods to find the solution. Also the convergence of the asymptotic solution for $`\xi <<1`$ is not expected to be fast and hence the integral solution is of special practical use.
We consider the case $`\alpha =1/2`$ and $`d=3`$. Using
$$l_{1/2}\left(z\right)=\frac{1}{2\sqrt{\pi }}z^{3/2}\mathrm{exp}\left[1/\left(4z\right)\right]$$
(47)
with $`z>0`$. We find the integral solution $`P(r,t)`$ Eq. (31) using numerical integration. We have used Mathematica which gave all the numerical results without difficulty. In Fig. 2 we show $`P(r,t)`$ versus r on a semi log plot and for different times $`t`$. Close to the origin $`r=0`$ we observe a sharp increase of $`P(r,t)`$ as predicted in Eq. (45).
More detailed behavior of $`P(r,t)`$ is presented in Fig. 3 where we show $`rP(r,t)`$ versus $`r`$. In Fig. 3 we also exhibit linear curves based on the asymptotic expansion Eq. (43) which predicts
$$rP(r,t)C_1(t)rC_2(t)$$
(48)
where
$$C_1(t)=\frac{1}{4\pi ^{3/2}t^{1/2}}$$
(49)
and $`C_2(t)=1/[4\pi \mathrm{\Gamma }(1/4)t^{3/4}]`$. Eq. (48) is valid when $`\xi =r^2/t^{1/2}<<1`$ and as expected we see in Fig. 3 that for this case the numerical integral solution and the asymptotic solution (48) agree well.
Finally in Fig. 4 we show our results in a scaling form, similar to the way we presented the CTRW results in Fig. 1 which shows the CTRW solution for both short and long times. We present $`r^3P(r,t)`$ versus $`\xi `$ for different choices of time $`t`$. We observe collapse of all curves, both for shorter and longer times, onto one master curve. We also show the asymptotic behaviors $`\xi >>1`$ and $`\xi <<1`$, Eq. (39) and Eq. (43) respectively. The numerical integral solution Eq. (31) agrees well with the asymptotic behaviors in the appropriate regimes. Comparing Fig. (1) and Fig. (4) we see that the solution of the fractional diffusion equation approximates the exact solution of the CTRW very well when $`t\mathrm{}`$ and $`r0`$.
## IV Fractional FokkerโPlanck Equation
Let us consider the fractional FokkerโPlanck equation (FFPE) describing the stochastic evolution of a test particle under combined influence of external force field $`F(x)`$ and a thermal heat bath. The equation reads
$$\dot{P}(x,t)=K_\alpha {}_{0}{}^{}D_{t}^{1\alpha }\widehat{L}_{fp}P(x,t)$$
(50)
and
$$\widehat{L}_{fp}=\frac{}{x}\frac{F(x)}{k_bT}+\frac{^2}{x^2}$$
(51)
is the FokkerโPlanck operator. $`K_\alpha `$ is a generalized diffusion constant and $`T`$ is temperature. The fractional derivative $`{}_{0}{}^{}D_{t}^{1\alpha }`$ was defined in Eq. (29). We consider the one dimensional case and extensions to higher dimensions are straight forward. When $`F(x)=0`$ the FFPE Eq. (50) reduces to the fractional diffusion equation (28). When $`\alpha =1`$ we recover the ordinary Fokker-Planck equation. The stationary solution of the FFPE is the Boltzmann distribution and the equation is compatible with linear response theory .
In this section we find an integral solution of the FFPE. We show that the solution is normalized and non negative, an issue not discussed in . Without loss of generality we set $`K_\alpha =1`$. We consider initial condition $`P(x,t=0)=\delta (xx_0)`$.
First let us show that the solution is normalized. Integrating Eq. (50) with respect to $`x`$ and using the boundary conditions
$$P(x,t)|_{x=\pm \mathrm{}}=\frac{P(x,t)}{x}|_{x=\pm \mathrm{}}=0$$
(52)
we find
$$_{\mathrm{}}^{\mathrm{}}\dot{P}(x,t)๐x=0,$$
(53)
meaning that normalization is conserved as it should.
We write the solution of the FFPE Eq. (50) in terms of an integral of a product of two functions
$$P(x,t)=_0^{\mathrm{}}๐s\stackrel{~}{n}(s,t)P_1(x,s)$$
(54)
where
$$\frac{P_1(x,s)}{s}=\widehat{L}_{FP}P_1(x,s).$$
(55)
$`P_1(x,s)`$ is a normalized solution of the ordinary Fokker-Planck equation with initial conditions $`P_1(x,s=0)=\delta (xx_0)`$. Methods of solution of Eq. (55) are given in Riskenโs book . In what follows we shall prove that $`\stackrel{~}{n}(s,t)=n(s,t)`$, defined in Eq. (12).
We use the Laplace transform of Eq. (54) and normalization condition of $`P(x,t)`$ to show
$$_0^{\mathrm{}}\stackrel{~}{n}(s,u)๐s=1/u.$$
(56)
Hence $`\stackrel{~}{n}(s,t)`$ is normalized according to $`_0^{\mathrm{}}\stackrel{~}{n}(s,t)๐s=1`$. The Laplace transform of Eq. (50) reads
$$uP(x,u)\delta (xx_0)=u^{1\alpha }\widehat{L}_{fp}P(x,u),$$
(57)
inserting Eq. (54) in Eq. (57) we find
$$u_0^{\mathrm{}}\stackrel{~}{n}(s,u)P_1(x,s)๐s\delta (xx_0)=$$
$$u^{1\alpha }_0^{\mathrm{}}\stackrel{~}{n}(s,u)\widehat{L}_{fp}P_1(x,s)๐s,$$
(58)
integrating by parts using Eq. (55) we find
$$u_0^{\mathrm{}}\stackrel{~}{n}(s,u)P_1(x,s)๐s\delta (xx_0)=$$
$$u^{1\alpha }\left[\stackrel{~}{n}(\mathrm{},u)P_1\left(x,s=\mathrm{}\right)\stackrel{~}{n}(0,u)P_1(x,0)\right]$$
$$u^{1\alpha }_0^{\mathrm{}}\left[\frac{}{s}\stackrel{~}{n}(s,u)\right]P_1(x,s)๐s.$$
(59)
From Eq. (56) $`\stackrel{~}{n}(\mathrm{},u)=0`$ and since $`P_1(x,0)=\delta (xx_0)`$ we may rewrite Eq. (59)
$$_0^{\mathrm{}}\left\{u\stackrel{~}{n}(s,u)+u^{1\alpha }\left[\frac{}{s}\stackrel{~}{n}(s,u)\right]\right\}P_1(x,s)๐s=$$
$$\left[1u^{1\alpha }\stackrel{~}{n}(0,u)\right]\delta (xx_0).$$
(60)
Eq. (60) is solved once both of its sides are equal zero; therefore two conditions must be satisfied, the first
$$\stackrel{~}{n}(0,u)=u^{\alpha 1},$$
(61)
and the second
$$\frac{}{s}\stackrel{~}{n}(s,u)=u^\alpha \stackrel{~}{n}(s,u).$$
(62)
The solution of Eq. (62) with initial condition Eq. (61) is
$$\stackrel{~}{n}(s,u)=u^{\alpha 1}\mathrm{exp}\left(u^\alpha s\right).$$
(63)
Thus $`\stackrel{~}{n}(s,u)=n(s,u)`$ found in the context of the solution of CTRW Eq. (10).
We see that the integral solution of the FFPE has a similar structure as the solutions of the fractional diffusion equation or equivalently of the asymptotic CTRW. The solution is Eq. (54) with $`n(s,t)`$ defined in Eq. (12) and $`P_1(x,s)`$ being the solution of corresponding ordinary Fokker-Planck equation. It is easy to understand that the solution, $`P(x,t)`$ is normalized and non negative because both $`P_1(x,t)`$ and $`n(s,t)`$ are normalized PDFs.
In a different method of solution of the fractional FokkerโPlanck equation was investigated. It can be shown that $`P(x,t)`$ can be expanded in an eigen function expansion, which is similar to the standard eigen function expansion of solution of ordinary FokkerโPlanck equation, however eigen modes decay according to MittagโLeffler relaxation instead of the standard exponential decay of the modes in the integer FokkerโPlanck equation. It can be shown that the integral solution investigated here is identical to the eigen function expansion in .
### A Example 1, Biased Fractional Wiener Process
Consider the biased fractional diffusion process defined with a generalized diffusion coefficient $`K_\alpha `$ and a uniform force field $`F(x)=F>0`$. For this case the mean displacement grows slower than linear with time according to
$$x(t)=\frac{FK_\alpha t^\alpha }{k_bT\mathrm{\Gamma }(1+\alpha )}.$$
(64)
The well known solution of the ordinary Fokker-Planck equation is
$$P_1(x,s)=\frac{1}{4\pi s}\mathrm{exp}\left[\frac{\left(x\stackrel{~}{F}s\right)^2}{4s}\right].$$
(65)
Eq. (65) describes a biased Wiener process and $`\stackrel{~}{F}=F/(k_bT)`$. The solution $`P(x,t)`$ for the fractional case is found using the transformation Eq. (54). In Laplace $`u`$ space the solution is
$$P(x,u)=$$
$$\frac{\stackrel{~}{F}u^{\alpha 1}\tau ^\alpha }{\sqrt{1+4(u\tau )^\alpha }}\mathrm{exp}\left[\frac{\stackrel{~}{F}(x\sqrt{1+4(u\tau )^\alpha }|x|)}{2}\right]$$
(66)
and $`\tau ^\alpha =1/(\stackrel{~}{F}K_\alpha )^2`$. For $`(\tau u)^\alpha <<1`$, corresponding to the long time behavior of the solution $`P(x,t)`$, we find
$$P(x,u)\{\begin{array}{cc}\stackrel{~}{F}u^{\alpha 1}\tau ^\alpha \mathrm{exp}\left(\stackrel{~}{F}\tau ^\alpha u^\alpha x\right)& x>0\\ & \\ \stackrel{~}{F}u^{\alpha 1}\tau ^\alpha \mathrm{exp}\left(\stackrel{~}{F}|x|\stackrel{~}{F}|x|\tau ^\alpha u^\alpha \right)& x0.\end{array}$$
(67)
Since $`_{\mathrm{}}^0P(x,u)u^{\alpha 1}`$ and hence according to Tauberian theorem $`_{\mathrm{}}^0P(x,t)1/t^\alpha `$ we have $`P(x,t)=0`$ for $`x<0`$ when $`t\mathrm{}`$. Therefore, using the inverse Laplace transform of Eq. (67) the asymptotic behavior of $`P(x,t)`$ is
$$\underset{t\mathrm{}}{lim}P(x,t)=\{\begin{array}{cc}\frac{1}{\alpha A^{1/\alpha }}\frac{t}{x^{1+1/\alpha }}l_\alpha \left(\frac{t}{A^{1/\alpha }x^{1/\alpha }}\right)& 0<x\\ & \\ 0& 0>x,\end{array}$$
(68)
and $`A=\stackrel{~}{F}\tau ^\alpha `$. Integrating Eq. (68) we find the distribution function
$$\underset{t\mathrm{}}{lim}_{\mathrm{}}^xP(x,t)๐x=1L_\alpha \left(\frac{t}{A^{1/\alpha }x^{1/\alpha }}\right),$$
(69)
valid for $`x>0`$. Eq. (69) was derived also in based on the biased CTRW, thus as expected the solution of the fractional Fokker-Planck equation converges to the solution of the CTRW in the limit of large $`t`$.
On the origin one can use Tauberian theorem to show
$$P(0,t)\frac{A}{\mathrm{\Gamma }(1\alpha )}t^\alpha $$
(70)
valid for long times. For the case $`F=0`$ we have found $`P(0,t)t^{\alpha /2}`$ so as expected the decay on the origin is faster for the biased case since particles are drifting away from the origin.
In Fig. 5 we present the solution for the case $`\alpha =1/2`$, then
$$P(x,t)=$$
$$\frac{1}{\sqrt{tK_{1/2}^2\pi }}_0^{\mathrm{}}๐s\frac{1}{\sqrt{4\pi s}}\mathrm{exp}\left[\frac{s^2}{4K_{1/2}^2t}\frac{(x\stackrel{~}{F}s)^2}{4s}\right]$$
(71)
which is evaluated numerically. For large times we have
$$P(x,t)\frac{A}{\sqrt{\pi t}}\mathrm{exp}\left[\frac{A^2x^2}{4t}\right]$$
(72)
for $`0<x`$. As seen in the figure the exact result exhibits a strong sensitivity on initial condition and the maximum of $`P(x,t)`$ is located on $`x=0`$. This is different than ordinary diffusion process in which the maximum of $`P(x,t)`$ is on $`x(t)`$. The curves in Fig. 5 are similar to those observed by Scher and Montroll based on lattice simulation of CTRW and also by Weissman et al who investigated biased CTRW using an analytical approach. The FFPE solution presented here is much simpler than the CTRW solution, still it captures all the important features of the more complex CTRW result.
### B Example 2, the Fractional Ornstein-Uhlenbeck Process
We consider as a second example the fractional OrnsteinโUhlenbeck (OU) process, namely the motion of the test particle in harmonic oscillator. This case cannot be analyzed using the CTRW in a direct way. The CTRW formalism considers only uniformly biased random walks and the fractional processes in non-uniform fields are not uniformly biased. We consider $`\alpha =1/2`$, $`F(x)/k_bT=x`$ and use the well known solution of the ordinary OU process
$$P_1(x,s)=\frac{1}{\sqrt{2\pi \left(1e^{2s}\right)}}\mathrm{exp}\left[\frac{\left(xx_0e^s\right)^2}{2\left(1e^{2s}\right)}\right].$$
(73)
The solution of the fractional OU process is then found using numerical integration of Eq. (54) using Eqs. (47) and (73). Our results are presented in Fig. 6. We have considered an initial condition $`x_0=1/2`$ and we observe a strong dependence of the solution on the initial condition. A cusp on $`x=x_0`$ is observed for all times $`t`$, thus the initial condition has a strong influence on the solution. The solution approaches the stationary Gaussian shape slowly in a power law way and the solution deviates from Gaussian for any finite time. Unlike the ordinary Gaussian OU process, the maximum of $`P(x,t)`$ is not on the average $`x(t)`$ but rather the maximum is for short times located on the initial condition.
Like the ordinary OU process the fractional OU process has a special role. The ordinary OU process describes two types of behaviors, the first is an over damped motion of a particle in harmonic potential, the second is the velocity of a Brownian particle modeled by the Langevin equation, the latter is the basis for the Kramers equation. In a similar way the fractional OU describes an over damped and anomalous motion in Harmonic potential considered in this section and in , it can also be used to model the velocity of a particle exhibiting a Lรฉvy walk type of motion . The fractional OU process is the basis of the fractional Kramers equation introduced recently by Barkai and Silbey , this equation describes super diffusion while in this work we have considered sub diffusion.
## V Summary
Fractional diffusion equation is an asymptotic equation which predicts the behavior of the decoupled continuous time random walk in the sub diffusive regime. The fractional FokkerโPlanck equation considers such a sub-diffusive motion in an external force field and close to thermal equilibrium. In this work we have considered an integral transformation which gives the solution of the fractional FokkerโPlanck equation in terms of solution of ordinary FokkerโPlanck equation. Solution of ordinary FokkerโPlanck equation can be found based on different analytical and numerical methods . The integral transformation describes also the long time behavior of the CTRW in dimension $`d=1,2,3`$. Thus the transformation maps Gaussian diffusion onto fractional diffusion, and it can serve as a practical tool for finding solution of certain fractional kinetic equations.
## VI Acknowledgments
EB thanks A. I. Saichev and G. M. Zaslavsky for correspondence and J. Klafter, R. Metzler and G. Zumofen for discussions.
## VII Appendix A
We rewrite Eq. (16)
$$P(๐ค,u)u^{\alpha 1}_0^{\mathrm{}}e^{s\left(u^\alpha +๐ค^2\right)}๐s,$$
(74)
the inverse Fourier transform of Eq. (74) is
$$P(๐ซ,u)u^{\alpha 1}^1\left\{_0^{\mathrm{}}e^{s\left(u^\alpha +๐ค^2\right)}๐s\right\},$$
(75)
and $`^1`$ is the inverse Fourier transform. Changing the order of integration over parameter $`s`$ and the operation $`^1`$ (this is later justified by the identity of moments of the integral solution and that found in ) we find using Eq. (8)
$$P(๐ซ,u)_0^{\mathrm{}}n(s,u)\frac{1}{\left(4\pi s\right)^{d/2}}\mathrm{exp}\left(r^2/4s\right)๐s.$$
(76)
Applying the inverse Laplace transform to this equation, changing the order of integrations over $`s`$ and the inverse Laplace operation, we find Eq. (11).
## VIII Appendix B
We investigate the Cartesian moments of the CTRW. These moments are
$$M(2m_1,\mathrm{},2m_d)=$$
$$_{\mathrm{}}^{\mathrm{}}dx_1\mathrm{}_{\mathrm{}}^{\mathrm{}}dx_dx_1^{2m_1}x_2^{m_2}\mathrm{}x_d^{2m_d}\times $$
$$\underset{s=1}{\overset{\mathrm{}}{}}\{L_\alpha \left[\frac{t}{s^{1/\alpha }}\right]L_\alpha \left[\frac{t}{\left(s+1\right)^{1/\alpha }}\right]\}\times $$
$$\frac{1}{\left(4\pi s\right)^{d/2}}\mathrm{exp}\left(\frac{x_1^2+\mathrm{}x_d^2}{4s}\right).$$
(77)
Changing order of integration and summation, we use the identity
$$_{\mathrm{}}^{\mathrm{}}dx_1\mathrm{}_{\mathrm{}}^{\mathrm{}}dx_d\frac{1}{\left(4\pi s\right)^{d/2}}\mathrm{exp}(\frac{x_1^2+\mathrm{}x_d^2}{4s})\times $$
$$x_1^{2m_1}\mathrm{}x_d^{2m_d}=C_{m,d}s^m$$
(78)
were $`C_{m,d}`$ is defined in Eq. (26) Inserting Eq. (78) in Eq. (77) we find Eq. (25).
The Laplace transform of Eq. (25) is
$$M(2m_1,\mathrm{},2m_d)=C_{m,d}\frac{1e^{u^\alpha }}{u}\underset{s=1}{\overset{\mathrm{}}{}}e^{su^\alpha }s^m.$$
(79)
We use
$$\underset{s=1}{\overset{\mathrm{}}{}}e^{su^\alpha }s^m=(1)^m\left(\frac{d}{dx}\right)^m\underset{s=1}{\overset{\mathrm{}}{}}e^{xs}|_{x=u^\alpha }=$$
$$\left(1\right)^m\left(\frac{d}{dx}\right)^m\frac{e^x}{1e^x}|_{x=u^\alpha },$$
(80)
and for small $`u^\alpha `$ we find
$$M(2m_1,\mathrm{},2m_d)C_{m,d}\mathrm{\Gamma }(m+1)u^{1\alpha m}.$$
(81)
Applying Tauberian theorem \[i.e., inverting Eq. (81)\] we find Eq. (27).
## IX Appendix C
We consider the moments
$$M(2m_1,\mathrm{},2m_d)=$$
$$_{\mathrm{}}^{\mathrm{}}๐x_1\mathrm{}_{\mathrm{}}^{\mathrm{}}๐x_d\left[_0^{\mathrm{}}n(s,t)G(๐ซ,s)\right]x_1^{2m_1}\mathrm{}x_d^{2m_d},$$
(82)
changing the order of integration over $`s`$ and the $`d`$ dimensional integration, using Eq. (78) we find
$$M(2m_1,\mathrm{},2m_d)=$$
$$C_{m,d}_0^{\mathrm{}}s^mn(s,t)๐s.$$
(83)
were $`C_{m,d}`$ is defined in Eq. (26) and $`m=_{i=1}^dm_i`$, Using Eq. (12) the integral in Eq. (83) is
$$I_m(t)_0^{\mathrm{}}s^mn(s,t)๐s=$$
$$\frac{1}{\alpha }_0^{\mathrm{}}\frac{1}{t^\alpha }\left(\frac{t}{s^{1/\alpha }}\right)^{\alpha +1}l_\alpha \left(t/s^{1/\alpha }\right)s^m๐s.$$
(84)
Notice that $`I_m(t)`$ is the $`m`$ moment of $`n(s,t)`$. Changing the integration variable, in Eq. (84), according to $`y=t/s^{1/\alpha }`$ we find
$$I_m(t)=t^{\alpha m}_0^{\mathrm{}}y^{\alpha m}l_\alpha \left(y\right)๐y,$$
(85)
the calculation of the negative moment in Eq. (85) is straight forward using Laplace transform technique , we find
$$I_m(t)=\frac{t^{\alpha m}}{\mathrm{\Gamma }\left(\alpha m\right)}_0^{\mathrm{}}u^{\alpha m1}l_\alpha \left(u\right)๐u$$
(86)
with $`l_\alpha \left(u\right)=\mathrm{exp}(u^\alpha )`$ being the Laplace transform of the one sided Lรฉvy density. Therefor we find
$$I_m(t)=\frac{\mathrm{\Gamma }\left(1+m\right)}{\mathrm{\Gamma }\left(1+\alpha m\right)}t^{\alpha m}.$$
(87)
Inserting Eq. (87) in Eq. (83) we find the result Eq. (37). Notice that when $`\alpha =1`$, $`I_m(t)=t^m`$, this is expected since for this case $`n(s,t)=\delta (st)`$.
## X Appendix D
Integral formulas involving the product of two H Fox functions are a helpful tool with which H functions can be represented in terms of known functions. According to equation $`(\mathrm{5.1.2})`$
$$_0^{\mathrm{}}dx\{x^{\eta 1}H_{p,q}^{m,n}\left[zx^\sigma \right|\begin{array}{c}(a_j,\alpha _j)_{1,p}\hfill \\ (b_j,\beta _j)_{1,q}\hfill \end{array}]$$
$$\times H_{P,Q}^{M,N}\left[sx\right|\begin{array}{c}(c_j,\gamma _j)_{1,P}\hfill \\ (d_j,\delta _j)_{1,Q}\hfill \end{array}]\}=$$
$$s^\eta H_{p+P,q+Q}^{m+M,n+N}\left[zx^\sigma \right|$$
$$\begin{array}{c}(a_j,\alpha _j)_{1,n},(c_j+\eta \gamma _j,\sigma \gamma _j)_{1,P},(a_j,\alpha _j)_{n+1,p}\hfill \\ (b_j,\beta _j)_{1,m},(d_j+\eta \delta _j,\sigma \delta _j)_{1,Q},(b_j,\beta )_{m+1,q},\hfill \end{array}],$$
(88)
provided that seven conditions are satisfied, $`\sigma >0`$,
$$\delta =\underset{j=1}{\overset{q}{}}\beta _j\underset{j=1}{\overset{p}{}}\alpha _j>0,$$
(89)
$$\delta ^{}=\underset{j=1}{\overset{Q}{}}\delta _j\underset{j=1}{\overset{P}{}}\gamma _j>0,$$
(90)
$$A=\underset{j=1}{\overset{n}{}}\alpha _j\underset{j=n+1}{\overset{p}{}}\alpha _j+\underset{j=1}{\overset{m}{}}\beta _j\underset{j=m+1}{\overset{q}{}}\beta _j>0,$$
(91)
$$A^{}=\underset{j=1}{\overset{M}{}}\delta _j\underset{j=M+1}{\overset{Q}{}}\delta _j+\underset{j=1}{\overset{N}{}}\gamma _j\underset{j=N+1}{\overset{P}{}}\gamma _j>0,$$
(92)
$$\sigma \begin{array}{c}\\ \text{max}\\ 1jn\end{array}\left[\left(a_j1\right)/\alpha _j\right]\begin{array}{c}\\ \text{min}\\ 1jM\end{array}\left[d_j/\delta _j\right]<\eta $$
(93)
and
$$\eta <\sigma \begin{array}{c}\\ \text{min}\\ 1jm\end{array}\left[b_j/\beta _j\right]\begin{array}{c}\\ \text{max}\\ 1jN\end{array}\left[\left(c_j1\right)/\gamma _j\right].$$
(94)
We have considered the case when all parameters are real, for the case when parameters may become complex see .
We express the integral solution Eq. (31), in terms of an integral of a product of two Fox H functions, after some rearrangements and with the use of Eq. (15) and the following identity
$$G(s,r)=\pi ^{d/2}r^dH_{0,1}^{1,0}\left[\frac{r^2}{4s}\right|\begin{array}{c}\hfill \\ (d/2,1)\hfill \end{array}]$$
(95)
we find
$$P(r,t)=_0^{\mathrm{}}n(s,t)G(r,s)๐s=$$
$$\alpha ^2\pi ^{d/2}r^d\{$$
$$_0^{\mathrm{}}dxx^{11/\alpha }H_{1,1}^{1,0}\left[x^{1/\alpha }\right|\begin{array}{c}(1,1)\hfill \\ (1/\alpha ,1/\alpha )\hfill \end{array}]\times $$
$$H_{0,1}^{1,0}\left[\frac{r^2}{4t^\alpha }x|\begin{array}{c}\hfill \\ (d/2,1)\hfill \end{array}\right].$$
(96)
Using Eq. (88) one can hope to prove that the integral solution Eq. (31) and the Fox Function solution Eq. (38) are identical, provided of course that all conditions Eqs. (89-94) are satisfied. Unfortunately the integral identity Eq. (88) cannot be used for this aim because condition Eq. (94) is not satisfied, inserting the fractional parameters in Eq. (94) we find
$$1/\alpha <1/\alpha $$
(97)
which shows that the conditions are not fulfilled.
It is worthwhile mentioning that all the other conditions (89-93) are fulfilled and that it is possible to prove that the integral solution and the Fox function solution are identical if we replace $`<`$ with $``$ in Eq. (94). Using Eq. (88), the identity
$$\alpha ^1H_{1,2}^{2,0}\left[x^{1/\alpha }|\begin{array}{c}(0,1)\hfill \\ (0,1/\alpha ),(d/2,1/\alpha )\hfill \end{array}\right]=$$
$$H_{1,2}^{2,0}\left[x^{1/\alpha }|\begin{array}{c}(1,1)\hfill \\ (d/2,1/\alpha ),(1,1/\alpha )\hfill \end{array}\right]=$$
(98)
which can be proven based upon the definition of the Fox function, and
$$z^\sigma H_{p,q}^{m,n}\left[z|\begin{array}{c}(a_j,\alpha _j)_{1,p}\hfill \\ (b_j,\beta _j)_{1,q}\hfill \end{array}\right]=$$
$$H_{p,q}^{m,n}\left[z|\begin{array}{c}(a_j,\sigma \alpha _j)_{1,p}\hfill \\ (b_j,\sigma \beta _j)_{1,q}\hfill \end{array}\right]$$
(99)
we find
$$P(r,t)=_0^{\mathrm{}}n(s,t)G(r,s)๐s=$$
$$\alpha ^1\pi ^{d/2}r^d\times $$
$$H_{12}^{20}\left(2^{2/\alpha }r^{2/\alpha }t^1|\begin{array}{c}(1,1)\hfill \\ (d/2,1/\alpha ),(1,1/\alpha )\hfill \end{array}\right)$$
(100)
which is the Fox function solution of Eq. (38). We emphasize that this equation was not proven in this Appendix because condition in Eq. (94) was not satisfied.
## XI Appendix E
The Fox function is represented as
$$H_{p,q}^{m,n}\left[x|\begin{array}{c}(a_1,\alpha _1)\mathrm{}(a_p,\alpha _p)\hfill \\ (b_1,\beta _1)\mathrm{}(b_q,\beta _q)\hfill \end{array}\right],$$
(101)
and for our choice of parameters
$$m=2,n=0,p=1,q=2$$
$$a_1=1,\alpha _1=1$$
$$b_1=d/2,\beta _1=1/\alpha ,b_2=1,\beta _2=1/\alpha $$
(102)
The asymptotic expansion of the H Fox function, for $`0<x<<1`$, is defined when two conditions are satisfied . The first is
$$\delta =\underset{j=1}{\overset{q}{}}\beta _j\underset{j=1}{\overset{p}{}}\alpha _j>0,$$
(103)
and for the case $`\delta =0`$ see . For our case, defined by the parameters in Eq. (102), $`\delta =2/\alpha 1>0`$ when $`0<\alpha <1`$. The second condition is
$$\beta _h\left(b_j+\mathrm{\Lambda }\right)\beta _j\left(b_h+k\right)$$
(104)
for
$$jh;j=h=1,\mathrm{}m;\mathrm{\Lambda },k=0,1,2,\mathrm{}.$$
(105)
Using Eq. (102) condition Eq. (105) reads
$$\frac{1}{\alpha }\left(1+\mathrm{\Lambda }\right)\frac{1}{\alpha }\left(d/2+k\right)$$
(106)
therefore the condition is satisfied for dimensions $`d=1`$ and $`d=3`$ but not for $`d=2`$. When conditions are satisfied
$$H_{p,q}^{m,n}\left[x|\begin{array}{c}(a_1,\alpha _1)\mathrm{}(a_p,\alpha _p)\hfill \\ (b_1,\beta _1)\mathrm{}(b_q,\beta _q)\hfill \end{array}\right]=$$
$$\underset{h=1}{\overset{m}{}}\underset{k=0}{\overset{\mathrm{}}{}}\mathrm{\Pi }_{j=1,jh}^m\mathrm{\Gamma }(b_j\beta _j\xi _{h,k})\mathrm{\Pi }_{j=1}^n\mathrm{\Gamma }(1a_j+\alpha _j\xi _{h,k})\times $$
$$(1)^k\left(x\right)^{\xi _{h,k}}[\mathrm{\Pi }_{j=m+1}^q\mathrm{\Gamma }(1b_j+\beta _j\xi _{h,k})$$
$$\mathrm{\Pi }_{j=n+1}^p\mathrm{\Gamma }(a_j\alpha _j\xi _{h,k})k!\beta _h]^1$$
(107)
where
$$\xi _{h,k}=\left(b_h+k\right)/\beta _h.$$
(108)
Using Eq. (107) we find
$$H_{1,2}^{2,0}\left[x|\begin{array}{c}(1,1)\hfill \\ (d/2,1/\alpha ),(1,1/\alpha )\hfill \end{array}\right]=$$
$$\alpha \{\underset{k=0}{\overset{\mathrm{}}{}}\frac{\mathrm{\Gamma }\left(1d/2k\right)\left(1\right)^kx^{\alpha (d/2+k)}}{\mathrm{\Gamma }\left(1\alpha d/2\alpha k\right)k!}$$
$$+\underset{k=0}{\overset{\mathrm{}}{}}\frac{\mathrm{\Gamma }\left(d/21k\right)\left(1\right)^kx^{\alpha (1+k)}}{\mathrm{\Gamma }\left(1\alpha \alpha k\right)k!}\}.$$
(109)
Using some simple manipulations we find our results Eqs. (42) and (43). |
warning/0001/hep-th0001020.html | ar5iv | text | # 1 Introduction
## 1 Introduction
Physicists seldom define their terms. So although I know roughly what a moduli space is, and the sort of thing one does with it in physics, I was not really very sure of what exactly it is. So I asked Frances (Kirwan), just as the Balliol College (where participants were lodged) porters did when they also wanted to know what a moduli space was. I have always taken it to be some sort of useful parameter space, convenient in the sense that mathematicians have already worked out all its properties (at least in the classical cases). But Frances told me something much more significantโshe describes it as a parameter space in the nicest possible way.
So in the next 55 minutes or so, I shall try to give you a rough picture of how physicists have made use of this nice concept of a parameter space. We should note, however, that it is far from a one-way traffic. Much of the tremendous progress in 4-manifold theory, and a large part of it is done here, came about by studying certain moduli spaces occurring in mathematical physics.
A few notes of warning, however, are in place. For a hard-nosed or pragmatic physicist, (A) spacetime $`X`$ has 4 dimensions, 3 space and 1 time, with an indefinite metric. By an indefinite metric I mean that the quadratic form giving the metric is not positive definite, so that two distinct points in spacetime can be null-separated. In fact, distances along light-paths are always zero. For him (or her) also (B) spacetime is by and large like $`^4`$, that is, (i) flat, (ii) looking more or less the same in all directions, (iii) real, and (iv) more or less infinite in all its 4 directions and hence non-compact.
On the other hand, algebraic geometry is more about Riemannian manifolds and the best results are almost always obtained for the compact case. In order to make contact, the concept of spacetime has to be modified in several significant ways.
1. One considers definite metrics, a process known as euclideanization. Then many nice things happen. In particular, the wave operator
$$\mathrm{}=\frac{^2}{t^2}\frac{^2}{x^2}\frac{^2}{y^2}\frac{^2}{z^2}$$
which is hyperbolic, becomes the 4-dimensional Laplacian
$$^2=\frac{^2}{t^2}+\frac{^2}{x^2}+\frac{^2}{y^2}+\frac{^2}{z^2}$$
which is elliptic, and for elliptic operators there are all sorts of good results like the index theorems. Euclideanization is done in the following: Self-dual YangโMills theory, instantons, monopoles, SeibergโWitten theory, strings, โฆ.
2. Alternatively, one complexifies spacetime, and then the question of definite or indefinite metric disappears. In this case, one can use powerful complex manifold techniques including twistor theory. This is also where supersymmetry comes in mathematically. Moreoever, by a change of point of view (see later), Riemann surfaces also play an important role. Complexification is done in superstrings, supersymmetric YangโMills theory, $`M`$-theory, โฆ.
3. One also changes the topology of spacetime by compactifying some or all of its directions. In some cases, this is only a mild change, amounting to imposing certain decay properties at infinity (see later). In other cases, this gives rise to important symmetries of the theory. Compactification is done in instantons, superstrings, $`M`$-theory, โฆ.
4. One either changes the number of spacetime dimensions or re-interprets some of them as other degrees of freedom. This dimensional change is done in strings, superstrings, monopoles, $`M`$-theory, โฆ.
At first sight, these modifications look drastic. The hope is that they somehow reflect important properties of the real physical world, and that the nice results we have do not disappear on us once we know how to undo the modifications. Surprisingly, the (largely unknown) mathematics underlying real 4-dimensional spacetime looks at present quite intractable!
## 2 YangโMills theory (Gauge theory)
Unlike most of the other theories I shall mention, YangโMills theory is an experimentally โprovenโ theory. In fact, it is generally believed, even by hard-nosed or pragmatic physicists, that YangโMills theory is the basis of all of particle physics. From the physics point of view, YangโMills theory is the correct framework to encode the invariance of particle theory under the action of a symmetry groupโthe gauge group $`G`$โat each spacetime point. For example, let $`\psi (x)`$ be the wave-function of a quantum particle. Then the physical system is invariant under the action of the group:
$$\psi (x)\mathrm{\Lambda }(x)\psi (x),\mathrm{\Lambda }(x)G.$$
This invariance is known as gauge invariance. Now the groups that are most relevant to particle physics are $`U(1),SU(2),SU(3)`$. However, we shall come across other groups as well. But for simplicity, we shall take $`G=SU(2)`$, unless otherwise stated.
There is an additional ingredient in many favoured gauge theories, namely supersymmetry. This is a symmetry relating two kinds of particles: bosons (e.g. a photon) with integral spin and fermions (e.g. an electron) with half-integral spin. Spin is a kind of internal angular momentum which is inherently quantum mechanical. Since bosons and fermions in general behave quite differently (e.g. they obey different statistics), this symmetry is not observed in nature. However, one can imagine this symmetry holding for example at ultra-high energies. What makes this symmetry theoretically interesting is that many theories simplify and often become complex analytic with this extra symmetry, making much of the underlying mathematics accessible. Also the complex analyticity links such theories with most studies of moduli spaces.
Mathematically, YangโMills theory can be modelled (in the simplest case) by a principal bundle $`P`$ (see Figure 1)
together with a connection on it. I remind you that, roughly speaking, a principal bundle is a manifold $`P`$ with a projection $`\pi `$ onto a base space $`X`$, and a right action by the structure group $`G`$. In general, the base space can be any smooth manifold, but here we consider only the case of spacetime $`X`$. Above each point $`xX`$, the inverse image (called the fibre) $`\pi ^1(x)`$ is homeomorphic to $`G`$. The total space $`P`$ is locally a product, in the sense that $`X`$ is covered by open set $`U_\alpha `$ and $`\pi ^1(U_\alpha )`$ is homeomorphic to $`U_\alpha \times G`$. A connection $`A`$ is a 1-form on $`P`$ with values in the Lie algebra $`๐ค`$ of $`G`$, satisfying certain conditions and giving a prescription for differentiating vectors and tensors on $`X`$. It combines with the usual exterior derivative $`d`$ to give the covariant exterior derivative $`d_A`$:
$$d_A=d+A$$
in such a way as to preserve gauge invariance.
Next we need the curvature 2-form:
$$F_A=dA+AA(F_{\mu \nu }=_\nu A_\mu _\mu A_\nu +ig[A_\mu ,A_\nu ]).$$
The second formula (in brackets) is the same as the first one, but written in local coordinates, or โwith indicesโ, where $`\mu =0,1,2,3.`$
Since $`dimX=4`$ (for the moment, anyway), we have the Hodge star operator which takes 2-forms to 2-forms:
$`:`$ $`\mathrm{\Omega }^2`$ $`\mathrm{\Omega }^2`$
$`F_A`$ $`{}_{}{}^{}F_{A}^{}.`$
In local coordinates, this can be written as
$${}_{}{}^{}F_{\mu \nu }^{}=\frac{1}{2}ฯต_{\mu \nu \rho \sigma }F^{\rho \sigma },$$
where $`ฯต_{\mu \nu \rho \sigma }`$ is a completely skew symbol defined by $`ฯต_{0123}=1`$. Notice that
$`()^2`$ $`=`$ $`+1\mathrm{in}\mathrm{euclidean}\mathrm{metric}`$
$`()^2`$ $`=`$ $`1\mathrm{in}\mathrm{Minkowskian}\mathrm{metric}.`$
YangโMills theory is given by the YangโMills action or functional
$$S(A)=\frac{1}{8\pi ^2}_Xtr(F_A{}_{}{}^{}F_{A}^{})=\frac{1}{8\pi ^2}F_A^2.$$
The curvature satisfies:
$`d_AF_A`$ $`=`$ $`0(\mathrm{Bianchi}\mathrm{identity})`$
$`d_A{}_{}{}^{}F_{A}^{}`$ $`=`$ $`0(\mathrm{Yang}\mathrm{Mills}\mathrm{equation}).`$
These are the classical equations for YangโMills theory. Notice that the first one is an identity from differential geometry, and the second one comes from the first variation of the action.
The space of connections $`๐`$ is an affine space, but we are really interested in connections modulo gauge equivalence. Two connections $`A,A^{}`$ are gauge equivalent if they are โgauge transformsโ of each other:
$$A^{}=\mathrm{\Lambda }^1A\mathrm{\Lambda }+\mathrm{\Lambda }^1d\mathrm{\Lambda }.$$
In other words, $`\mathrm{\Lambda }(x)G`$, $`\mathrm{\Lambda }`$ is a fibre-preserving automorphism of $`P`$ invariant under the action of $`G`$. We shall use the symbol $`๐ข`$ for the group of gauge transformations $`\mathrm{\Lambda }`$.
So we come to our first, most basic, moduli space
$$\overline{}=๐/๐ข.$$
It is in general infinite-dimensional with complicated topology.
We shall be interested in various subspaces or refinements of $`\overline{}`$.
One theoretical use of $`\overline{}`$ itself is in (the euclidean formulation of) quantum field theory, where with the Feynman path integral approach, one has to consider the integral of the exponential of the YangโMills action over $`\overline{}`$:
$$_\overline{}e^{S(A)}.$$
But this integral is very difficult to define in general!
The moduli space $`\overline{}`$ has a singular set which represents the reducible connections, which are connections with holonomy group $`HG`$ such that the centralizer of $`H`$ properly contains the centre of $`G`$. We say then that the connection reduces to $`H`$. The complement $``$ of this singular set is dense in $`\overline{}`$, and represents the irreducible connections. For $`G=SU(2)`$, near an irreducible connection $`\overline{}`$ is smooth, but reducible connections lead to cone-like singularities in $`\overline{}`$.
### 2.1 Instantons
Recall that $`G=SU(2)`$. Bundles $`P`$ over $`X`$ are classified by the second Chern class of the associated rank 2 vector bundle $`E`$ (cf. Rosa-Maria Mirรณ-Roigโs talk):
$$k=c_2(E)[X]=\frac{1}{8\pi ^2}_XtrF_A^2.$$
We say that a connection $`A`$ is self-dual (or anti-self-dual) if its curvature $`F_A`$ satisfies
$$F_A={}_{}{}^{}F_{A}^{}(\mathrm{resp}.F_A={}_{}{}^{}F_{A}^{}).$$
Then given any connection $`A`$, we can decompose the corresponding curvature $`F_A`$ into its self-dual and and anti-self-dual parts:
$$F_A=F_A^++F_A^{}.$$
In the context of YangโMills theory a self-dual connection is called an instanton<sup>1</sup><sup>1</sup>1It is a matter of convention whether one so defines a self-dual or anti-self-dual connection.:
$$F_A={}_{}{}^{}F_{A}^{}F_A^{}=0.$$
In this case,
$$\mathrm{Bianchi}\mathrm{identity}\mathrm{Yang}\mathrm{Mills}\mathrm{equation}.$$
In other words, a self-dual connection is automatically a classical solution.
Now we have
$`S(A)`$ $`=`$ $`{\displaystyle \frac{1}{8\pi ^2}}{\displaystyle _X}|F_A^+|^2+|F_A^{}|^2`$
$`k`$ $`=`$ $`{\displaystyle \frac{1}{8\pi ^2}}{\displaystyle _X}|F_A^+|^2|F_A^{}|^2.`$
Hence one has immediately
$$S(A)k,$$
and
$$S(A)=kF_A^{}=0.$$
So a self-dual connection gives an absolute minimum for the action. The integer $`k`$ is known as the instanton number.
Warning: Nontrivial self-dual connections exist only when $`X`$ is either euclidean or complex.
The mathematical magic of instantons is that instead of solving the second order YangโMills equations we have only the first order self-duality equation to deal with. These connections can actually be constructed using euclidean twistor methods without explicitly solving any equations (cf. Tatiana Ivanovaโs talk).
Physically, the presence of instanton contribution in the path integral allows tunnelling between different vacua (i.e. lowest energy states) of the relevant YangโMills theory (namely quantum chromodynamics for strong interactions or QCD). This role of the instantons can be compared to lower-dimensional objects such as โsolitonsโ or topological defects called โkinksโ which connect up two different states at infinity (see Figure 2). The two phenomena are quite similar, since โtunnellingโ means a quantum particle can penetrate a potential barrier which a classical particle cannot go through, thus connecting two classically separate states.
The effect of instantons is โnon-perturbativeโ in the sense that such an effect cannot be obtained as a term in a power series expansion of $`g`$ the coupling constant (which is measure of the โstrengthโ of the interaction under consideration, and which appears for example in the nonlinear term of the curvature form $`F_{\mu \nu }`$). This is a direct manifestation of the fact that instantons are topological in nature and cannot be obtained by any โlocalโ considerations such as power series expansions.
Since in euclidean space the YangโMills equations are elliptic, and concentrating on irreducible connections gets rid of zero eigenvalues, one can use the index theorem to count the โformal dimensionโ of instanton moduli space. Typically the smooth part of the moduli space will have this formal dimension as its actual dimension. For example,
$$X=S^4,dim_{}(_{I,k})=8k3.$$
Uhlenbeck has given a unique compactification of $`_I`$, the union for all $`k`$. For more details about instanton moduli spaces, I again refer you to Tatiana Ivanovaโs talk.
### 2.2 Monopoles
Recall $`G=SU(2)`$.
Consider a YangโMills theory with a scalar field (called Higgs field) $`\varphi `$, together with a potential term $`V(\varphi )`$ which is added to the YangโMills action. Suppose further that
$$V(\varphi _0)=\mathrm{minimum}\mathrm{for}|\varphi _0|0,$$
and that $`V(\varphi )`$ is invariant under a subgroup $`U(1)SU(2)`$. Then for those connections of $`P`$ which are reducible to this $`U(1)`$ subgroup, we can for certain purposes concentrate on this โresidual gauge symmetryโ and have a $`U(1)`$ gauge theory. If we interpret this $`U(1)`$ as Maxwellโs theory of electromagnetism, then a non-trivial reduction of $`P`$ can be regarded as a magnetic monopole. The magnetic charge $`k`$ is given by the first Chern class of the reduced bundle. In fact we have the following exact sequence which gives us an isomorphism:
$$\begin{array}{ccccccc}\pi _2(SU(2))& & \pi _2(SU(2)/U(1))& \stackrel{}{}& \pi _1(U(1))& & \pi _1(SU(2))\\ & & & & & & \\ 0& & & & & & 0\end{array}$$
Unlike the original magnetic monopole considered by Dirac, these โt HooftโPolyakov monopoles have finite energy and are the soliton solutions of the field equations corresponding to the action:
$$S(A,\varphi )=S(A)+D\varphi ^2+\lambda (1|\varphi |^2)^2,$$
where the last term is the usual form of the potential $`V(\varphi )`$. From this we get the YangโMillsโHiggs equations (YMH):
$`D_AF`$ $`=`$ $`0,`$
$`D_A{}_{}{}^{}F`$ $`=`$ $`[\varphi ,D_A\varphi ],`$
$`D_A{}_{}{}^{}D_{A}^{}\varphi `$ $`=`$ $`2\lambda \varphi (|\varphi |^21).`$
Now we specialize to a certain limit, the Prasad-Somerfeld limit: $`V(\varphi )=0`$, but $`|\varphi |1`$ at infinity. Then the YangโMillsโHiggs system becomes:
$`D_AF`$ $`=`$ $`0,`$
$`D_A{}_{}{}^{}F`$ $`=`$ $`[\varphi ,D_A\varphi ],`$
$`D_A{}_{}{}^{}D_{A}^{}\varphi `$ $`=`$ $`0.`$
Consider next a YangโMills theory in euclidean $`^4`$, invariant under $`x_4`$-translations. Then we can write
$$A=A_1dx_1+A_2dx_2+A_3dx_3+\varphi dx_4,$$
where $`A_1,A_2,A_3,\varphi `$ are Lie algebra-valued functions on $`^3`$. The action can be written as
$$S(A)=F_A^2=F^2+D\varphi ^2,$$
where now $`F`$ is the curvature of the connections in 3 dimensions:
$$A^{}=A_1dx_1+A_2dx_2+A_3dx_3,$$
and $`D`$ is the corresponding 3-dimensional covariant derivative. In this way, we can make the following identification since the actions for the two theories are identical:
$$\mathrm{YMH}\mathrm{on}^3\mathrm{dimensionally}\mathrm{reduced}\mathrm{YM}\mathrm{on}^4.$$
In this case,
$$F_A={}_{}{}^{}F_{A}^{}\mathrm{first}2\mathrm{YMH}.$$
Hence a solution to the Bogomolny equation
$$F={}_{}{}^{}D_{A}^{}\varphi $$
gives a solution of YMH. These are known as โstatic monopolesโ.
The moduli spaces $`_k`$ corresponding to a given charge $`k`$ are well studied, at least for $`k=1,2`$. The translation group $`^3`$ acts freely on $`_k`$, so does an overall phase factor $`S^1`$. Dividing these out we get the reduced monopole moduli spaces $`_k^0,dim_{}=4k4`$. Taking the $`k`$-fold covers, one obtains:
$$\stackrel{~}{}_k^3\times S^1\times \stackrel{~}{}_k^0.$$
The special case of $`k=2`$ has been studied by Atiyah and Hitchin as an entirely novel way of obtaining the scattering properties of two monopoles, using a metric on $`_2^0`$ they discovered, and assuming (with Manton) that geodesic motion on it describes adiabatic motion of the two monopoles. This is the most direct use that I know of of moduli space for deriving something akin to dynamics!
### 2.3 Topological field theory
I wish just to mention a class of quantum field theories called topological quantum field theories (TQFT), where the observables (correlation functions) depend only on the global features of the space on which these theories are defined, and are independent of the metric (which, however, may appear in the classical theory). Atiyah gave an axiomatic approach to these, but there are so many local experts here that I do not feel justified in expanding on that!
Instead, I shall just indicate the role of moduli space in Wittenโs approach. Starting with a moduli space $``$ one can get fields, equations and symmetries of the theory. Witten postulates the existence of certain operators $`๐ช_i`$ corresponding to cohomology classes $`\eta _i`$ of $``$ such that
$$๐ช_1\mathrm{}๐ช_n=_{}\eta _1\mathrm{}\eta _n,$$
where $`\mathrm{}`$ denotes the correlation function of the operators. Hence he obtains these correlation functions as intersection numbers of $``$, using Donaldson theory. So in a sense the TQFT is entirely defined by $``$.
The observables called correlation functions can best be understood in the case of, for example, a 2-point function in statistical mechanics. This is the probability, given particle 1, of finding particle 2 at another fixed location.
To go into any further details about TQFT would require more detailed knowledge both of quantum field theory and supersymmetry. These would lead us unfortunately too far from the context of this workshop.
### 2.4 SeibergโWitten theory
Recall that a spin structure on $`X`$ is a lift of the structure group of the tangent bundle of $`X`$ from $`SO(4)`$ to its double cover Spin(4)$`SU(2)\times SU(2)`$. Because of this isomorphism, one can represent a spin structure more concretely as a pair of complex 2-plane bundles $`S^+,S^{}X`$, each with structure group $`SU(2)`$. A slightly more general concept is a spin<sup>c</sup> structure over $`X`$, which is given by a pair of vector bundles $`W^+,W^{}`$ over $`X`$ with an isomorphism for the second exterior powers
$$\mathrm{\Lambda }^2W^+=\mathrm{\Lambda }^2W^{}=L,\mathrm{say},$$
such that one has locally
$$W^\pm =S^1L^{\frac{1}{2}},$$
where $`L^{\frac{1}{2}}`$ is a local square root of $`L:L^{\frac{1}{2}}L^{\frac{1}{2}}=L`$.
Given a spin<sup>c</sup> manifold $`X`$, the SeibergโWitten equations (SW) are written for a system consisting of 1) a unitary connection $`A`$ on $`L=\mathrm{\Lambda }^2W^\pm `$, and 2) $`\psi `$ a section of $`W^+`$. Then these equations are:
$`D_A\psi `$ $`=`$ $`0`$
$`F_A^+`$ $`=`$ $`\tau (\psi ,\psi ),`$
where $`\tau `$ is a sesquilinear map $`\tau :W^+\times W^+\mathrm{\Lambda }^+`$.
The SeibergโWitten equations (SW) can be obtained from varying the following functional:
$$E(A,\psi )=_X|D_A\psi |^2+|F_A^++\tau (\psi ,\psi )|^2+R^2/8+2\pi ^2c_1(L)^2,$$
where $`R`$ is the scalar curvature of $`X`$ and $`c_1(L)`$ is the first Chern class of $`L`$. Notice that the last two terms depend only on $`X`$ and $`L`$, so that solutions of SW are absolute minima of $`E`$ on the given bundle $`L`$.
The relevant moduli space here is the space $``$ of all irreducible solution pairs $`(A,\psi )`$, modulo gauge transformations. The SeibergโWitten invariants are then homology classes of $``$, independent of the metric on $`X`$. These invariants prove very useful in 4-manifold theory. In particular, Seiberg and Witten give a โphysicistโs proofโ that the instanton invariants of certain 4-manifolds (namely with $`b^+>1`$, where $`b^+`$ is the dimension of the space of self-dual harmonic forms) can be expressed in terms of the SeibergโWitten invariants.
From the quantum field theory point of view, the importance of SeibergโWitten theory lies in the concept of duality. In a modified version of YangโMills theory, called $`N=2`$ supersymmetric YangโMills theory, the quantum field theory is described by a scale parameter $`t`$ and a complex parameter $`u`$ (here supersymmetry is essential). In the limit $`t\mathrm{}`$, the theory is described by an analytic function $`\tau `$ of $`u`$. If $`b^+(X)>1`$, then $`\tau `$ is modular (in the classical sense) with respect to the action of $`SL(2,)`$. This means in particular that a theory with parameter $`u`$ is related to a theory with parameter $`u^1`$ in a definite and known way. The transformation $`uu^1`$ corresponds to changing the coupling constant to its inverse. Hence for the magnetic monopoles of the theory this represents a duality transformation: from electric with coupling $`e`$ to magnetic with coupling $`\stackrel{~}{e}`$ and vice versa, since Diracโs quantization condition states that $`e\stackrel{~}{e}=1`$ in suitable units. By relating a โstrongly coupledโ theory to a โweakly coupledโ theory, one can hope to obtain results on the former by performing perturbative calculations (which are meaningless when coupling is strong) in the latter. By inspecting their moduli spaces one is often able to identify pairs of dually related theories.
## 3 String and related theories
I shall be extremely brief about these theories. The reason is, apart from my own obvious ignorance, that they are considerably more complicated than gauge theories and require much more knowledge not only of quantum physics but also of algebraic geometry than can reasonably be dealt with in this workshop. My aim here is just to give a taste of some immensely active areas of research in mathematical physics in recent years where moduli spaces play an important role.
The gist of string theory is that the fundamental objects under study are not point-like particles as in gauge field theories but 1-dimensional extended strings. These strings are really the microscopic quantum analogues of violin strings: they move in space and they also vibrate. The equation of motion of a free string can be obtained from an action which is similar to that for a massless free particle. In the latter case we have
$$S_0=๐\tau \eta _{\mu \nu }\frac{dx^\mu }{d\tau }\frac{dx^\nu }{d\tau }$$
which is just the length of the โworldlineโ in spacetime $`X`$ traced out by the particle as it travels through space. Here $`\eta _{\mu \nu }`$ is the metric on $`X`$ and $`x^\mu `$ are the coordinates of the particle. For the string the free action is the area of the โworldsheetโ (with coordinates $`\sigma ,\tau `$) traced out by the 1-dimensional string in spacetime $`X`$:
$$S_1=๐\sigma ๐\tau \eta ^{\alpha \beta }\eta _{\mu \nu }_\alpha x^\mu _\beta x^\nu ,$$
where the indices $`\alpha ,\beta =0,1`$ refer to the worldsheet. Varying $`S_1`$ with respect to $`x`$ gives simply the 2-dimensional wave equation:
$$\left(\frac{^2}{\tau ^2}\frac{^2}{\sigma ^2}\right)x^\mu =0.$$
We see that in this context spacetime coordinates can be regarded as fields on the 2-dimensional surface which is the worldsheet.
Interaction between strings are given by the joining and splitting of strings so that the resultant worldsheet can be visualized, on euclideanization, as a Riemann surface $`\mathrm{\Sigma }`$ with a given genus (see Figure 3).
For example, a hole in $`\mathrm{\Sigma }`$ can be obtained by one closed string splitting into two and then joining together again. In fact, a useful way of looking at string theory is to think of it as being given by an embedding $`f`$ of a Riemann surface $`\mathrm{\Sigma }`$ into spacetime $`X`$ (Figure 4).
### 3.1 Conformal field theory
We have written the action $`S_1`$ for a free string in terms of a particular parametrization of $`\mathrm{\Sigma }`$, but obviously the physics ought to be invariant under reparametrization. The group of reparametrization on $`\mathrm{\Sigma }`$ is the infinite-dimensional conformal group, and that is the symmetry group of string theory.
On the other hand, on a given Riemann surface $`\mathrm{\Sigma }`$ one can consider certain field theories which have this invariance. These are called conformal field theories (CFT) and play important roles in statistical mechanics and critical phenomena (e.g. phase change), when the theories become independent of the length scale (so that quantities are defined only up to conformal transformations).
The concept of moduli plays an important role in CFT. In fact, the original idea of modulus is defined for Riemann surfaces (see talk by Frances Kirwan). So a torus $`T^2`$ has one modulus $`\tau `$ (see Figure 5).
The conformal structure of $`T^2`$ is invariant under the action of the modular group $`SL(2,)`$ on $`\tau `$.
CFT are often studied for their own sake, but as far as string theories are concerned their use lies in the fact that they are the terms in a first-quantized, perturbative formulation of string theory. Schematically, one can think of string theory as the โsum over $`g`$โ of CFT on Riemann surfaces of genus $`g`$. Unfortunately, this โsummationโ has never yet been given a precise meaning. What provides some hope that the problem may be tractable is the fact that the infinite-dimensional integral $`e^{S_1(x)}`$ occurring in the path integral formalism can be reduced to one on the moduli space of the Riemann surface, which is finite-dimensional.
### 3.2 Various string theories
Up to now I have been carefully vague about the nature of spacetime $`X`$ in string theory. It turns out that to get a consistent, first-quantized theory, one needs $`X`$ to have 26 dimensions! If we modify the theory by adding supersymmetry to produce a superstring theory, then $`dimX=10`$. However, this potentially disastrous requirement has been turned to good use to produce interesting theories in 4 dimensions, as we now briefly sketch.
We shall concentrate on the supersymmetric version as being the more favoured by string theorists, in that we now assume $`dimX=10`$. Imagine that one can compactify 6 of these 10 dimensions so that
$$XK\times ^4$$
with $`K`$ a compact 6-dimensional space, and moreover that the size of $`K`$ is small. Since the length is an inverse measure of energy, this means that to observers of low energy (such as us) spacetime will just look 4-dimensional and the other 6 dimensions are curled up so tight we cannot see them. The often-quoted example is that a water pipe looks like a thin line from a distance.
Not only that, the symmetries of $`X`$ can be factored into that of $`^4`$ (the usual ones) and that of $`K`$. The latter can then be interpreted as the internal symmetries of YangโMills theory. In fact, the choice of $`K`$ is dictated by which gauge symmetry one wants.
There are in all 5 string theories. A string can be open (homeomorphic to an interval) or closed (homeomorphic to a circle). An open string theory is called Type I. For closed strings, depending in the boundary conditions one imposes, one has Type IIA or Type IIB. If one combines both the usual and the supersymmetric versions one obtains the heterotic string, with gauge group (after suitable compactification) either $`E_8\times E_8`$ or $`SO(32)`$. The $`E_8\times E_8`$ heterotic string is particularly favoured as being able to include various YangโMills theories which are important in particle physics.
### 3.3 $`M`$ Theory
One can generalize the 1-dimensional strings to higher-dimensional objects called โmembranesโ; similarly superstrings to โsupermembranesโ. The study of these last objects have become particularly fashionable, especially after the introduction of something called $`M`$-theory.
Now supersymmetry can also be made into a local gauge theory which is then called supergravity. It was shown some time ago that in supergravity, $`dimX11`$, so 11-dimensional supergravity was studied as being in some sense a unique theory.
$`M`$-theory is perceived as an 11-dimensional supergravity theory, where the 11-dimensional manifold $`X`$ can be variously compactified to give different superstring theories. Moreover, solitonic solutions are found which are supermembranes. By examining the moduli of these solutions one can connect pairs of underlying string theories. For example, reminiscent of the SeibergโWitten duality and using the modular transformations on the modulus $`\tau `$ of the torus (in one of the compactifications of $`X`$), one can connect the two different versions of the heterotic string. In fact, by using both compactification and duality one finds that $`M`$-theory can give rise to all the 5 superstring theories mentioned above. So in some sense, all the 5 are equivalent and one can imagine that they are just different perturbative expansions of the same underlying $`M`$-theory.
Most recently, Maldecena suggested that $`M`$-theory on compactification on a particular 5-dimensional manifold (called anti-de Sitter space), including all its gravitational interactions, may be described by a (non-gravitational) YangโMills theory on the boundary of $`X`$ which happens to be 4-dimensional Minkowski space (i.e. flat spacetime). This opens up some new vistas in the field.
Although progress is made in an almost day-to-day basis, we are still waiting for a fuller description, perhaps even a definition, of $`M`$-theory. Meanwhile, it has generated a lot of interest and especially intense study into the various moduli spaces that occur.
## 4 Conclusions
I have endeavoured to describe a few pysical theories in which moduli space plays an important role. However, I must say that the success in the reverse direction is more spectacularโusing YangโMills moduli spaces (in different specializations) to understand 4-manifolds, following Donaldson, Kronheimer and many others. At the beginning I have explained why the success in physics is more restricted. Nevertheless, there are many high points:
1. Self-dual YangโMills $``$ instantons $``$ vacuum structure of QCD.
2. Monopole moduli spaces $``$ identification of pairs of dual theories in SeibergโWitten scheme $``$ hope for possibility of practical computations in quantum field theory.
3. Classification of conformal field theories $``$ application of theoretical statistical mechanics.
4. Identifying moduli spaces to connect up the different string theories $``$ leading to a unification in 11 dimensions?
But for lack of time and expertise, I have omitted many other areas of mathematical physics being actively pursued at present in which moduli spaces play significant roles.
## References
The following is only a small selection of articles that I have used in preparing this talk. They are in no way even representative.
1. Self-duality in 4-dimension Riemannian geometry, M.F. Atiyah, N.J. Hitchin and I.M. Singer, Proc. Roy. Soc. A362 (1978) 425-461.
2. The SeibergโWitten equations and 4-manifold topology, S.K. Donaldson, Bull. AMS 33 (1996) 45-70.
3. The first chapter of Superstring theory, M.B. Green, J.H. Schwarz and E. Witten, Cambridge University Press, 1987.
4. A laymanโs guide to $`M`$-theory, M.J. Duff, hep-th/9805177, talk delivered at the Abdus Salam Memorial Meeting, ICTP, Trieste, November 1997. |
warning/0001/cond-mat0001280.html | ar5iv | text | # Precision microwave dielectric and magnetic susceptibility measurements of correlated electronic materials using superconducting cavities
## I Introduction
The continuing discovery of new electronic materials calls for new methods of measuring their electric and magnetic properties. Microwave cavity perturbation techniques have proved to be very useful for the study of transport dynamics at microwave frequencies, in materials such as semiconductors, magnetic ferrites and exotic materials such as Charge and Spin Density Waves .
In all of these previous studies normal metal cavities were used. To study the (then) newly discovered high temperature superconductors (HTS), the use of superconducting cavities was introduced by Sridhar and Kennedy. The reduction in background absorption by a factor of $`10^4`$ from a normal metal cavity enabled the measurement of absorption in small, single crystal superconductors and thin films. The surface impedance $`\stackrel{~}{Z}_s=R_siX_s`$ was obtained in terms of changes of the cavity parameters : the shift in frequency $`\delta f`$ and quality factor $`Q`$. Subsequently the concept of the โhot fingerโ technique introduced in has been used in measurements in other laboratories also with the purpose of studying HTS .
In this paper, we present a reanalysis of the cavity perturbation technique, and describe a new application utilizing superconducting microwave cavities, to study dynamic electric and magnetic susceptibilities of strongly correlated electronic materials. We focus on the configuration where the sample is placed at a microwave magnetic field maximum of the $`TE_{011}`$ mode.
1. We introduce an electromagnetic susceptibility $`\stackrel{~}{\zeta }=\zeta ^{}+i\zeta ^{\prime \prime }`$, which provides a useful framework to discuss the results of the microwave measurements. We use $`\stackrel{~}{\zeta }_H`$ to note the case where the sample is measured in a microwave magnetic field (e.g. in the $`TE_{011}`$ mode), and $`\stackrel{~}{\zeta }_E`$ when the sample is placed in a microwave electric field (e.g. in the $`TM_{010}`$ mode). Depending on sample properties, the measured parameter $`\stackrel{~}{\zeta }`$ can be related to the sample magnetic permeability ($`\stackrel{~}{\mu }=1+\stackrel{~}{\chi }_M`$) and dielectric permittivity ($`\stackrel{~}{\epsilon }=1+\stackrel{~}{\chi }_P`$), where $`\stackrel{~}{\chi }_M(\stackrel{~}{\chi }_E)`$ are the magnetic(electric) susceptibilities, the conductivity $`\stackrel{~}{\sigma }`$ and the surface impedance $`Z_s`$. These various limits are discussed in detail in the paper.
2. For highly insulating samples with $`\stackrel{~}{\epsilon }1`$, the technique is a very sensitive method of measuring the magnetic susceptibility, since $`\stackrel{~}{\zeta }\stackrel{~}{\chi }_M=\chi _M^{}+i\chi _M^{\prime \prime }`$. The sensitivity of this technique is compared with others, and it is shown that the microwave method, when superconducting cavities are used, can equal or even exceed that of a dc SQUID for relative changes in susceptibility, such as with changing $`T`$. It also yields results on samples (typically mm-sized) in which comparable ac susceptibility measurements do not have sufficient sensitivity. As an example of this technique we show that it yields information on magnetodynamics in a spin chain material $`Sr_2CuO_3`$.
3. When the sample conductivity or dielectric constant is substantial, the measurements are dominated by these parameters. For insulating samples with even moderate dielectric constants $`\epsilon ^{}`$, the experiments are a direct measurement of $`\stackrel{~}{\epsilon }=\epsilon ^{}+i\epsilon ^{\prime \prime }`$. Thus we are able to measure $`\stackrel{~}{\epsilon }`$ even though the sample is placed in a microwave magnetic field maximum. (In fact the $`H_\omega `$ field measurements of $`\stackrel{~}{\epsilon }`$ have an advantage over $`E_\omega `$ measurements as they are not subject to the so-called depolarization peak). We describe an inversion procedure to obtain the complex dielectric constant $`\stackrel{~}{\epsilon }+i\stackrel{~}{\sigma }/\omega \epsilon _o`$ from the measured data. A spectacular example of the dielectric measurements is the observation of a dielectric loss peak in $`\epsilon ^{\prime \prime }(T)`$ due to dielectric relaxation in the spin ladder compound $`Sr_{14}Cu_{24}O_{41}`$.
4. For sufficiently large $`\epsilon ^{}`$, dimensional resonances can occur when the microwave essentially enter into the sample. An striking example of this is presented in data on $`SrTiO_3`$.
5. When the conductivity $`\sigma `$ is appreciable, it can lead to an eddy current contribution resulting in a peak in absorption with increasing conductivity. (This is the magnetic analog of the so-called depolarization peak for $`E_\omega `$ field measurements). For large conductivity the results tend to the surface impedance limit. This is the limit used in previous measurements of the surface impedance of metals and superconductors. This paper presents a unified approach which encompasses both the insulating and highly metallic limits.
The cavity perturbation method discussed here yields unique information on spin and charge dynamics at short time scales between Neutron Scattering and NMR and $`\mu SR`$, and has led to the observation of some unique phenomena in quantum magnets, dielectrics and superconductors.
## II Description of apparatus and measurement technique
A right cylindrical cavity (inner radius $`7/8`$ $`inch`$ and axial length $`1`$ $`inch`$) was made of pure Niobium ($`Nb`$), which is a superconductor below $`T_c=9.2K`$. The cavity was fabricated in three pieces: two end plates with the needed holes and one center ring. The top plate has a center pumping hole ($`3.56`$ $`mm`$ diameter), and two coupling holes ($`3.56`$ $`mm`$ diameter), the bottom plate has one centrally located hole ($`6.7`$ $`mm`$ diameter), through which the sample is inserted into the cavity. The $`TE_{011}`$ mode is degenerate with the $`TM_{111}`$ mode. As $`TE_{011}`$ is the desired operating mode, the diameter of these coupling holes was chosen to provide enough perturbation to split the two modes more than $`40MHz`$ apart. The high quality $`Nb`$ stock was carefully machined at very low speed to the needed shape and then polished without lubricant, which would otherwise cause oxidation on the $`Nb`$ surface. Each piece was then annealed, and the grains, which grew due to annealing, vary from sub millimeter size to roughly $`4`$ $`mm`$ diameter. The three-piece cavity was tightly held by a stainless steel assembly consisting of a top ring, a center piece for alignment and a bottom ring. The whole resonator was then mounted in an alignment frame, supported on the top by a stainless steel dewar probe ($`10.16`$ $`cm`$ diameter and $`1`$ $`m`$ long) and, on the bottom, with a sealed copper cup ($`10.16`$ $`cm`$ diameter and $`10.16`$ $`cm`$ long) with a removable bottom copper plate. Indium seals were used so that the entire assembly were vacuum tight. Superconducting operation of the cavity was accomplished using a bath of liquid $`{}_{4}{}^{}He`$.
A small piece of sample was mounted on the top of a sapphire rod ($`1.56mm`$ diameter $`\times `$ $`52mm`$ long) using very little Apiezon-N grease. The anisotropic response of the sample can be measured by mounting the sample in different orientations with respect to the applied microwave field for a given mode, as shown in the set-up diagram (Fig. 1). The sapphire rod with sample was inserted into the cavity, along its axis from the bottom, such that the sample is stationed exactly at the center of the cavity. Support and adjustment of the sapphire rod was provided by a copper tube ($`20mm`$ long), the overlap between copper tube and sapphire rod is adjustable and finally fixed with GE-Varnish to guarantee good thermal contact. The copper tube was brazed at the end to a $`6.35`$ $`mm`$ diameter stainless tube (wall thickness $`0.15mm`$), and the stainless tube was brazed to the bottom copper plate to form a thermal path to the bottom plate which is in contact with liquid $`{}_{4}{}^{}He`$.
To heat the sample to higher temperature, a $`50`$ $`\mathrm{\Omega }`$ heating coil ($`0.1`$ $`mm`$ Nichrome wire of $`6.5`$ $`\mathrm{\Omega }/ft.`$) was wound around the copper tube, and the control of the sample temperature was accomplished using an external temperature controller (Lake Shore DRC 82C), with a Silicon Diode temperature sensor (Lake Shore DT 470) which is attached to the sapphire rod outside the cavity. Another temperature sensor is put in the Helium chamber to monitor the bath temperature.
Microwaves were generated with a HP8510B network analyzer, a HP8341B synthesized sweeper and a HP8516A reflection/transmission test set, and coupled into and out of the resonator from the top, through two adjustable $`50\mathrm{\Omega }`$ coaxial lines, each terminated in a loop. One very useful feature of the design is the ability to vary the coupling to the resonator by moving the lines in and out along the axis of the resonator. Thus it is possible to achieve critical coupling and weak coupling over a wide range of the resonator quality factor $`Q`$ ($`10^4`$ $``$ $`10^8`$). For fixed coupling, input microwave power can be easily varied and the nonlinear effect of the some samples can be observed. The resonant cavity operated at desired $`TE_{011}`$ mode has the highest quality factor $`Q`$ about $`2\times 10^8`$ at $`2K`$ bath temperature.
## III Electrodynamic basis of the measurement
A small sample of volume $`V_s`$ placed in a resonant cavity causes the resonant frequency $`f`$ and quality $`Q`$ factor to change by a small amount $`\delta f`$. Assuming the shift in frequency is much smaller than the resonant frequency, $`\delta ff`$, the change in cavity parameters can be expressed as
$$\frac{\delta \stackrel{~}{f}}{f}\frac{(\stackrel{~}{\mu }1)\mu _o}{4U}\stackrel{}{H}\stackrel{}{H}_o๐V_s+\frac{(\stackrel{~}{\epsilon }1)\epsilon _o}{4U}\stackrel{}{E}\stackrel{}{E}_o๐V_s$$
(1)
where the complex frequency shift $`\delta \stackrel{~}{f}=\delta fi\mathrm{\Delta }f`$. $`\delta ff_sf_c`$ and $`\mathrm{\Delta }f\mathrm{\Delta }f_s\mathrm{\Delta }f_c`$ are the changes in the resonant frequency $`f`$ and the resonance width $`\mathrm{\Delta }f`$ respectively with (subscript $`s`$) and without (subscript $`c`$) the sample. The resonance width is related to the cavity $`Q`$ factor by $`\mathrm{\Delta }f=f/2Q`$. $`U`$ is the energy stored in the cavity of the resonant mode. $`(\stackrel{}{H}_o,\stackrel{}{E}_o)`$ and $`(\stackrel{}{H},\stackrel{}{E})`$ are the cavity field configurations before and after the sample perturbation. A time dependence $`e^{i\omega t}`$ is assumed. $`\stackrel{~}{\epsilon }`$ and $`\stackrel{~}{\mu }`$ are the complex permittivity and permeability. We define the magnetic susceptibility as $`\stackrel{~}{\chi }_M=\stackrel{~}{\mu }1=\chi _M^{}+i\chi _M^{\prime \prime }`$ and the dielectric susceptibility as $`\stackrel{~}{\chi }_P=\stackrel{~}{\epsilon }1=\chi _P^{}+i\chi _P^{\prime \prime }`$.
It is convenient to discuss experimental results in terms of an effective dynamic or electromagnetic susceptibility $`\zeta `$,
$$\delta \stackrel{~}{f}g\stackrel{~}{\zeta }g(\zeta ^{}+i\zeta ^{\prime \prime })$$
(2)
where $`g`$ is a sample geometrical factor, which is specific to the mode geometry and sample shape. Under appropriate conditions, $`\stackrel{~}{\zeta }`$ can be directly associated with the conventional magnetic $`\stackrel{~}{\chi }_M`$ or dielectric $`\stackrel{~}{\chi }_P`$ susceptibilities, as will be shown below.
To proceed further requires additional assumptions. Various approximations have been made, called the โQuasistaticโ (QS), โExtended Quasistaticโ (EQS) and Spherical Wave (SW) analysis, depending on the approximation used to obtain the fields $`(\stackrel{}{H},\stackrel{}{E})`$. An extensive analysis was carried out by Brodwin and Parsons (BP) , which covers essentially all the regimes needed for the experimental measurement discussed here. In the following we use BP and analyze the various regimes.
## IV Spherical sample in $`TE_{011}`$ mode
The $`TE_{011}`$ configuration is well suited as a probe of the microwave response of materials because of the very high $`Q`$โs achievable in this mode. In the present experiments, the sample is located at the center of the cavity on the axis. In this location we have maximum uniform axial magnetic field $`\stackrel{}{H}`$ and zero electric field $`\stackrel{}{E}`$. (See Fig.1 for spatial profiles of the $`\stackrel{}{H}`$ and $`\stackrel{}{E}`$ fields). In the following we use the analysis of BP, details of which are given in the appendix.
The geometrical factor $`g`$ of a spherical sample is given as
$$g=\frac{f}{J_0^2(\beta _{01}^{^{}}r_o)\left[1+\left(\frac{\pi }{L\beta _{01}^{}}\right)^2\right]}\frac{V_s}{V_c}$$
(3)
where $`V_c`$ is the volume of the empty cavity. $`\beta _{01}^{^{}}r_o`$ is the first root of Bessel function $`J_0^{^{}}(\beta r_o)=0`$. Using the cavity inner radius $`r_o=7/8`$ $`inch`$, and axial length $`L=1`$ $`inch`$, we get $`g`$ $`1.036\times 10^{15}V_s`$ $`[`$m$`{}_{}{}^{3}`$sec.$`{}_{}{}^{1}]`$, where $`V_s`$ is the sample volume.
The important parameters that define the analysis are the wave vector inside and outside the sample : $`k_o=\omega /c`$ and $`k=k_o\sqrt{\stackrel{~}{\epsilon }+i\stackrel{~}{\sigma }/\omega \epsilon _o}`$. The full-wave analysis yields in principle (see Appendix A), results of the frequency shift due to sample perturbation for a large range of sample sizes and material properties. However in all cases of experimental interest, the sample size is much smaller than the cavity dimensions, so that the condition $`k_oa1`$ is rigorously satisfied. For example, if $`a=1mm`$ and the measuring frequency is $`10GHz`$, then $`k_oa0.2`$. In this limit, we obtain
$$\stackrel{~}{\zeta }_H=\frac{3}{2}\left(\frac{(2\stackrel{~}{\mu }+1)j_1(ka)\mathrm{sin}(ka)}{(\stackrel{~}{\mu }1)j_1(ka)+\mathrm{sin}(ka)}\right)$$
(4)
We use the subscript $`H`$ to denote that the $`EM`$ susceptibility $`\stackrel{~}{\zeta }`$ is being measured with an applied microwave magnetic field $`\stackrel{}{H}_\omega `$.
This general form is in principle valid for arbitrary $`ka`$ which is determined by material properties $`\stackrel{~}{\mu }`$, $`\stackrel{~}{\epsilon }`$ and $`\stackrel{~}{\sigma }`$. However in this form it is not very useful. It is therefore necessary to consider the different limits of this expression. Below we discuss the various limits and their applicability.
### A: Magnetic permeability and susceptibility measurements
More generally the result in this limit can be written as :
$$\stackrel{~}{\zeta }_H=3\frac{\stackrel{~}{\mu }1}{\stackrel{~}{\mu }+2}+\frac{9}{10}\left[\frac{\stackrel{~}{\mu }^26\stackrel{~}{\mu }+4}{\left(\stackrel{~}{\mu }+2\right)^2}\left(k_oa\right)^2+\frac{\stackrel{~}{\mu }}{\left(\stackrel{~}{\mu }+2\right)^2}\left(ka\right)^2\right].$$
(5)
Clearly the experiment measures $`\stackrel{~}{\mu }`$ only if the second term is negligible. This may be possible in ferromagnetic samples where $`\mu ^{}1`$ provided the spins continue to respond at microwave frequencies. For weakly paramagnetic samples, we have
$$\stackrel{~}{\zeta }_H=\stackrel{~}{\chi }_M\text{}(k_oa)^2\stackrel{~}{\chi }_P\stackrel{~}{\chi }_M\text{ }$$
(6)
This limit is only achieved provided the sample is highly insulating and the dielectric constant is nearly $`1`$.
#### A:.1 Sensitivity and accuracy of magnetic susceptibility measurements
Having established the relationship between magnetic susceptibility $`\stackrel{~}{\chi }_M=\chi _M^{}+i\chi _M^{\prime \prime }`$ and measured electromagnetic susceptibility $`\stackrel{~}{\zeta }_H`$ in Eq. 6, we can estimate the measurement sensitivity of the technique. Clearly the sensitivity is associated with both the size of samples and the cavity resonant frequency $`f`$. The bigger the sample size is, the higher the sensitivity is, as seen from Eq.2, 3, 6, provided we still retain the small perturbation limit. Assuming a typical small sample has the dimension of $`V_s1\times 1\times 0.5mm^3`$, as in our experiment, we can detect the frequency shift $`\delta f`$ and the absorption width $`\mathrm{\Delta }f`$ as small as $`1Hz`$ in a resonant frequency of $`10^{10}Hz`$. This results in a sensitivity limit of $`\delta \zeta _H^{}10^6`$ and hence $`\delta \chi _M^{}10^6`$. For comparison, Table 1 lists the sensitivities of some commonly used techniques for magnetic susceptibility $`\chi `$ measurements. In these measurements, $`\chi `$ generally has the form
$$\chi \frac{M}{V_sH}$$
(7)
where $`M`$ is the magnetic moment in \[$`Am^2`$\], $`H`$ is the applied magnetic field in \[$`Am^1`$\]. If the same sample with volume $`V_s`$ is used for all these measurements, assuming an applied field of $`H10^5A/m`$ corresponding to typical microwave fields, we can compare their sensitivities, as listed in Table 1. Note that Eq. 6 gives the magnetic volume susceptibility $`\stackrel{~}{\chi }_M=dM/dH`$ in unit of SI or MKS (dimensionless). In CGS, it is usually expressed in \[$`emucm^3`$\]. To convert from CGS to SI, a conversion multiplying factor of $`4\pi `$ is used.
The table shows that the hot finger cavity perturbation technique undoubtedly has one of the highest measurement sensitivities available. While other methods may require relatively large sample size and large applied field $`H`$, these are not required in the microwave measurements. However this high sensitivity is achieved only for relative changes, such as for instance with varying temperature. The precision for absolute measurements is much less due to small uncertainties in sample location.
### B: Lossy Dielectric, Permittivity and Surface Impedance Measurements
For even moderate conductivity and dielectric constants, the magnetic contribution is overwhelmed by the dielectric and conductivity contributions. Taking $`\stackrel{~}{\mu }1`$, in the limit $`\stackrel{~}{\chi }_M(k_oa)^2\stackrel{~}{\chi }_P`$ , we have :
$$\stackrel{~}{\zeta }_H=\frac{3}{2}\left(1\frac{3}{(ka)^2}+\frac{3\mathrm{cot}ka}{ka}\right)$$
(8)
#### B:.1 Dielectric permittivity and susceptibility measurements
The small $`ka`$ limit of this results leads directly to a measurement of the dielectric permittivity or susceptibility:
$`\stackrel{~}{\zeta }_H`$ $``$ $`{\displaystyle \frac{1}{10}}(k_oa)^2(\stackrel{~}{\epsilon }+i\stackrel{~}{\sigma }/\omega \epsilon _o1);\stackrel{~}{\chi }_M(k_oa)^2\stackrel{~}{\chi }_P\text{ , }ka1`$
$``$ $`{\displaystyle \frac{1}{10}}(k_oa)^2\stackrel{~}{\chi }_P;\stackrel{~}{\sigma }=0,\stackrel{~}{\chi }_M(k_oa)^2\stackrel{~}{\chi }_P\text{ , }ka1`$
A surprising conclusion is that one can measure the dielectric properties even though the sample is placed in a pure microwave magnetic field. We emphasize that this conclusion has nothing to do with the spatial variation of the $`E`$-field near the cavity axis. It simply arises from the wave equation and holds, within geometric factors, even in a homogenous magnetic field and with zero electric field, such as can be achieved in a split ring resonator .
This method of measuring $`\stackrel{~}{\epsilon }`$ has one important advantage over $`E`$-field cavity perturbation measurements. The measured quantity is directly proportional to $`\stackrel{~}{\epsilon }1`$ and holds even when $`\stackrel{~}{\epsilon }1`$ so long as $`(k_oa)^2(\stackrel{~}{\epsilon }1)<1`$, while in the $`E`$-field method the measured frequency shifts are proportional to $`(\stackrel{~}{\epsilon }1)/(\stackrel{~}{\epsilon }+2)`$, due to so-called depolarization effects (see Appendix B), and can obscure the direct interpretation of the results.
For general $`\stackrel{~}{\epsilon }`$ Eq.8 can be inverted to obtain $`\stackrel{~}{\epsilon }`$ from the measured $`\stackrel{~}{\zeta }_H`$. Examples of such inversions are presented later.
Note that the dielectric permittivity measured is that appropriate to the plane perpendicular to the direction of the magnetic field $`\stackrel{}{H}_\omega `$. This is the direction of the displacement currents, and also the induced conduction currents. If the response in the plane is anisotropic then the measured $`\stackrel{~}{\epsilon }`$ will be an appropriate mixture of the responses in the different axes in the perpendicular plane. This must be viewed as a drawback compared to the E-field method, where in principle the response along each axis can be measured using a needle shaped specimen.
#### B:.2 Surface Impedance measurements (skin depth or eddy current limit)
The other useful limit is for a highly conducting material, where $`ka=(1+i)a/\delta =(1+i)a\sqrt{\mu _o\omega \sigma }`$. The skin depth $`\delta `$ $`=1/\sqrt{\mu _o\sigma \omega }`$ $`a`$, hence
$$\stackrel{~}{\zeta }_H\zeta _H\mathrm{}^{}=\frac{3}{\mu _o\omega a}(X_s+iR_s);\text{when }\stackrel{~}{\chi }_M0\text{}a\delta \text{}ka1\text{ }$$
(10)
It is useful to reference the data to the complete diamagnetic result $`\zeta _H\mathrm{}^{}=1.5`$ for a sphere. Thus in this limit the data are a direct measure of the surface impedance
$$\stackrel{~}{Z}_s=R_siX_s=\sqrt{\frac{i\omega \mu _o}{\stackrel{~}{\sigma }}}$$
(11)
The normalization factor $`3/(\mu _o\omega a)`$ is specific to the spherical sample and $`TE_{011}`$ mode geometries. Note that in this limit the measured data are $`1/\sqrt{\sigma }`$.
It is worth noting that this result (Eq.11) is also valid for complex conductivity $`\stackrel{~}{\sigma }=\sigma _1+i\sigma _2`$ such as for a superconductor. The above treatment assumes that displacement current effects are negligible. If they are also present and can be represented in terms of a dielectric constant $`\stackrel{~}{\epsilon }`$, then we can also write
$$\stackrel{~}{Z}_s=R_siX_s=\sqrt{\frac{i\omega \mu _o}{\stackrel{~}{\sigma }i\omega \stackrel{~}{\epsilon }}}$$
(12)
#### B:.3 Conductivity (Eddy Current) Peaks and Dielectric Loss Peaks
As noted above, the measured changes in the cavity resonance parameters expressed here in terms of the electromagnetic susceptibility $`\stackrel{~}{\zeta }_H`$ change from a $`\sigma `$ dependence for small $`\sigma `$ to a $`1/\sqrt{\sigma }`$ dependence for large $`\sigma `$. Thus as $`\sigma `$ is varied this results in a peak in the absorption or in $`\zeta _H^{\prime \prime }`$, accompanied by a change of state of $`\zeta _H^{}`$ from $`0`$ to $`\zeta _H\mathrm{}^{}=1.5`$, as shown in Fig.2. This conductivity or eddy current peak is similar to the depolarization peak observed in $`E`$-field measurements. Of course the location of the conductivity peak is determined by both the conductivity and the sample dimensions.
In certain materials, particularly the oxides, there are dielectric loss peaks intrinsic to the material, arising from a dielectric constant $`\stackrel{~}{\epsilon }=\epsilon ^{}+i\epsilon ^{\prime \prime }=\epsilon (0)/(1+i\omega \tau )`$. Usually $`\tau `$ is a strong function of temperature $`T`$, and hence when $`T`$ is varied, a peak in $`\epsilon ^{\prime \prime }(T)`$ occurs at a peak temperature $`T_p`$ where $`\omega \tau (T_p)=1`$. Since $`\tau (T)`$ increases with decreasing $`T`$, this peak shifts to lower peak temperatures $`T_p`$ when the measurement frequency is decreased. Since $`\stackrel{~}{\zeta }_H`$is proportional to $`\stackrel{~}{\epsilon }`$ in the appropriate limit, a peak will be observed in $`\zeta _H^{\prime \prime }`$ also as $`T`$ is varied.
In such materials $`\sigma (T)`$ is a also a strong function of $`T`$ and typically is semiconducting : $`\sigma (T)=\sigma _oexp(T_{s0}/T)`$. Under such conditions, the experimental data will display two peaks, one a dielectric loss peak and the other a conductivity peak, as $`T`$ is varied. When the measuring frequency $`\omega `$ is reduced, the dielectric loss peak will move to lower $`T`$ while the conductivity peak will move to higher $`T`$, i.e. the peaks move apart on the $`T`$ axis with decreasing $`\omega `$. A specific example of a dielectric loss peak in the spin ladder material $`Sr_{14}Cu_{24}O_{41}`$ is discussed later.
#### B:.4 Dimensional Resonances
A remarkable prediction of Eq.8 is the occurrence of dimensional resonances when the dielectric constant varies strongly. This is shown in Fig.3 (a) and (b). The resonances occur whenever $`ka=(n+1/2)\pi `$ and are quite sharp. They correspond to situations where the electromagnetic field essentially resonates inside the sample, just like a dielectric resonator. We have observed such resonances in $`SrTiO_3`$, which can be viewed as a quantum paraelectric with transition temperature at $`T=0`$, and in which material $`\epsilon ^{}`$ increases rapidly with decreasing $`T`$ to values approaching several thousands. Results are discussed later.
## V Experimental Procedures
In the experiment, we first carry out a background run to measure the resonance frequency $`f_c(T)`$and the width $`\mathrm{\Delta }f_c(T)`$ of the empty cavity as a function of $`T`$. Then the sample is inserted in and corresponding parameters $`f_s(T)`$ and $`\mathrm{\Delta }f_s(T)`$ are measured. $`\stackrel{~}{\zeta }_H=\zeta _H^{}(T)+i\zeta _H^{^{\prime \prime }}(T)`$ is obtained using
$`\zeta _H^{}(T)`$ $`=`$ $`{\displaystyle \frac{1}{g}}(f_s(T)f_c(T))`$ (13)
$`\zeta _H^{\prime \prime }(T)`$ $`=`$ $`{\displaystyle \frac{1}{g}}(\mathrm{\Delta }f_s(T)\mathrm{\Delta }f_c(T)).`$
$`g`$ is given by Eq. 3. In practice, while relative changes $`\delta f_s(T)=f_s(T)f_s(T_{ref})`$ or $`\delta f_c(T)=f_c(T)f_c(T_{ref})`$ referred to a reference temperature $`T_{ref}`$ can be measured with extremely high precision, there can be larger errors in the absolute value of $`f_s(T)f_c(T)`$. For this reason we represent the data as
$$\zeta _H^{}(T)=\frac{1}{g}\left[(\delta f_s(T)\delta f_c(T))+\delta f(T_{ref})\right]$$
(14)
In many cases, the background correction $`\delta f_c(T)`$ can be negligible. It is convenient to present the data as $`\delta \zeta _H^{}(T)=\zeta _H^{}(T)\zeta _H^{}(T_{ref})`$ instead of $`\zeta _H^{}(T)`$. To get the absolute value of $`\zeta _H^{}`$, calibration can be made by putting the sample into the cavity to measure $`f_s`$ and then immediately taking the sample out to measure $`f_c`$ at a fixed temperature, and thus obtain $`\delta f(T_{ref})`$. For many samples, (e.g. see $`Sr_{14}Cu_{24}O_{41}`$ later), $`\zeta _H^{}(T_{ref})\zeta _H^{}(T)`$ particularly at high $`T`$, so that in these cases, $`\delta \zeta _H^{}\zeta _H^{}`$.
### A: Inversion of experimental $`\stackrel{~}{\zeta }`$ or $`\stackrel{~}{Z}_s`$ data to obtain $`\stackrel{~}{\epsilon }`$ and $`\stackrel{~}{\sigma }`$
The next key step is to obtain the fundamental material property, the sample dielectric function $`\stackrel{~}{\epsilon }`$ or conductivity $`\stackrel{~}{\sigma }`$, from the experimental data represented either as $`\stackrel{~}{\zeta }`$ or the surface impedance $`\stackrel{~}{Z}_s`$. Two approaches are possible here :
1. A direct inversion of Eq.8 for $`\stackrel{~}{\zeta }(T)`$ data or Eq.11 for $`\stackrel{~}{Z}_s(T)`$ data to extract $`\stackrel{~}{\sigma }(T)i\omega \stackrel{~}{\epsilon }(T)`$,
2. Modelling of $`\stackrel{~}{\sigma }(T)i\omega \stackrel{~}{\epsilon }(T)`$ to quantitatively match the $`\stackrel{~}{\zeta }(T)`$ data using Eq.8 or $`\stackrel{~}{Z}_s(T)`$ data using Eq.11.
We discuss both these procedures below.
#### A:.1 Inversion of equations
We have successfully solved Eq.8 to obtain $`\stackrel{~}{z}=ka=k_oa\sqrt{\stackrel{~}{\epsilon }+i\stackrel{~}{\sigma }/\omega \epsilon _o}`$ using the subroutine FSOLVE in MATLAB. The success of the solution depends crucially on the values of $`\stackrel{~}{\zeta }`$ or equivalently $`\stackrel{~}{z}`$. For values of $`\zeta ^{},\zeta ^{\prime \prime }1`$, which is well in the QS or EQS limits, the solution is very accurate and yields the sample $`\stackrel{~}{\sigma }i\omega \stackrel{~}{\epsilon }`$ with ease. In this limit $`\stackrel{~}{z}1`$, corresponding to typical dielectric constants $`\epsilon ^{}1000`$ (for the sample and cavity sizes discussed in this paper) and not too small $`\epsilon ^{\prime \prime }`$. Thus for lossy dielectrics, the results for $`\stackrel{~}{\epsilon }`$ can be easily obtained. The results of such a solution for the material $`Sr_{14}Cu_{24}O_{41}`$ are discussed later in this paper. Results on several other materials which have similar properties, such as $`La_{5/3}Sr_{1/3}NiO_4`$, $`YBa_2Cu_3O_{6.0}`$ and $`\mathrm{Pr}Ba_2Cu_3O_{7.0}`$, are described in previous and forthcoming papers .
Great care must be exercised in two regimes of parameter values :
1. when $`\zeta ^{}1`$, and $`\zeta ^{\prime \prime }1`$, which corresponds to $`ka1`$ ( $`\epsilon ^{}>1000`$ for the conditions of the experiments in this paper). Here the resonances of $`\mathrm{cot}(z)`$ enter when $`ka=(n+1)\pi /2`$, leading to dimensional resonances discussed in other sections.
2. the metallic limit, when $`\zeta ^{}1.5`$, and $`\zeta ^{\prime \prime }1`$. In this limit it is more appropriate to use the surface impedance limit Eq.11 rather than Eq.8.
The principal difficulty in the above two limits is that there are many nearby minima of the underlying function, and the program quickly converges to spurious solutions. Future work will focus on this important problem.
#### A:.2 Modelling the conductivity and dielectric constant to match the data
Even if the solution procedure is successful and the material $`\stackrel{~}{\sigma }i\omega \stackrel{~}{\epsilon }`$ is successfully extracted, a quantitative understanding of the experimental results for $`\stackrel{~}{\sigma }i\omega \stackrel{~}{\epsilon }`$ requires a model. Where the solution is not easily attained due to the difficulties mentioned above, we have found it necessary to bypass the solution procedure and instead use model calculations of $`\stackrel{~}{\sigma }i\omega \stackrel{~}{\epsilon }`$ to describe the $`\stackrel{~}{\zeta }`$ data using Eq.8 and Eq.11.
## VI Experimental Results
We describe below measurements on three different materials all in single crystal form. These crystals have typical dimensions of $`1\times 1\times 0.5mm^3`$ and have been extensively characterized by a vast array of measurements: dc resistivity, dc SQUID susceptibility, XRD, neutron scattering and high pressure studies. Structural studies of the single crystals show that of all of these measurements indicate single phase, high quality crystals.
### A: Magnetodynamics in the spin chain material $`Sr_2CuO_3`$
$`Sr_2CuO_3`$ single crystals were prepared by the floating zone technique. It is an insulator in a large range of temperature and there is only possesses linear $`CuO`$ chains and is regard as an ideal one-dimensional spin $`1/2`$ chain. In the measurement, the sample is mounted in such way that the microwave field $`H_\omega //\widehat{c}`$-axis. Fig.4 (a) shows the plot of $`\delta \zeta _H^{}(T)\zeta _H^{}(T)\zeta _H^{}(2K)`$ vs. $`T`$ and $`\zeta _H^{\prime \prime }(T)`$ vs. $`T`$ for $`Sr_2CuO_3`$. $`\delta \zeta _H^{}`$ shows a monotonic increase with temperature $`T`$, and $`\zeta _H^{\prime \prime }`$ has insignificant changes from $`6K`$ to $`260K`$. These results are consistent with the DC magnetic susceptibility measurements , as shown in dashed line of Fig.4 (a), and indicate that the sample perturbation effect is in the magnetic susceptibility limit $`(k_oa)^2\stackrel{~}{\chi }_P\stackrel{~}{\chi }_M`$, so that $`\stackrel{~}{\zeta }_H\stackrel{~}{\chi }_M`$. Thus in this material we are essentially measuring the magnetic susceptibility.
At low temperatures, additional features are observed in both $`\delta \zeta _H^{}`$ and $`\zeta _H^{\prime \prime }`$, as shown in Fig.4 (b) and (c). These peaks are microwave signatures of the 3D Heisenberg AFM transition at $`T_N5K`$. The very high sensitivity of the technique utilizing a superconducting cavity is evident from the data in Fig.4.
### B: Dielectric Loss Peaks in the spin ladder material $`Sr_{14}Cu_{24}O_{41}`$
There is increasing interest for studying spin/ladder compounds because superconductivity can be obtained in $`Ca`$ doped $`Sr_{14}Cu_{24}O_{41}`$ under high pressures. In Fig. 5, we show the results of $`\stackrel{~}{\zeta }(T)`$ in the case of $`H_\omega //\widehat{c}`$-axis for $`Sr_{14}Cu_{24}O_{41}`$. The striking feature of the data is the rapid drop with decreasing $`T`$ in $`\delta \zeta _H^{\prime \prime }(T)`$ below approximately $`200K`$, accompanied by a relatively sharp peak in $`\zeta _H^{\prime \prime }(T)`$ at $`T170K`$, which is not seen in the DC magnetic susceptibility measurement (Fig.5 (b)). The extraordinary dynamic range (over $`4`$ orders of magnitude in $`\stackrel{~}{\zeta }_H`$) of the superconducting cavity enables us to see an additional peak at low $`T`$ in Fig.5 (b) (the semilog plot of (a) data). Although a similar peak is also observed in $`\chi _{dc}(T)`$, the magnitude is about $`10`$ times smaller than $`\zeta _H^{\prime \prime }(T)`$. At high temperatures, the measured $`\zeta _H^{}/\chi _{dc}10^3`$. Thus in this material the dielectric contributions dominate, i.e. $`(k_oa)^2\stackrel{~}{\chi }_P\stackrel{~}{\chi }_M`$, and we are thus measuring the dielectric constant.
Fig.5 (c) shows the dielectric constant $`\epsilon ^{}`$ and $`\epsilon ^{\prime \prime }`$ obtained from the measured $`\stackrel{~}{\zeta }`$ data and inverting Eq.8. The loss peak in $`\epsilon ^{\prime \prime }`$ is clearly evident, and is accompanied by a change of state of $`\epsilon ^{}`$. These data indicate an essentially pure dielectric relaxation process in this spin ladder material, arising from the presence of charges due to doping.
The dielectric mode is well described by a Cole-Davidson form $`\stackrel{~}{\epsilon }(\omega ,T)=\epsilon (0)/[1+i\omega \tau (T)]^\beta `$, with $`\epsilon (0)=27`$, $`\beta =0.6`$ and an activated relaxation time $`\tau (T)=1.6\times 10^{16}\mathrm{exp}(T_{\tau 0}/T)`$ \[$`\mathrm{sec}.]`$, with an activation energy $`T_{\tau 0}=2000K`$. When the relaxation rate $`\tau ^1(T)`$ varies rapidly with $`T`$ and crosses the measurement frequency $`\omega `$, a peak occurs at $`T_p`$, where $`\omega \tau (T_p)=1`$, as shown in Fig.5. In this material the relaxation time $`\tau (T)`$ appears to follow the conductivity $`\sigma (T)`$, indicating that the free carriers determine the polarization relaxation. Extensive details of the polarization dynamics in this material and in the related $`Sr_{14x}Ca_xCu_{24}O_{41}`$ family are discussed in a forthcoming publication .
### C: Dimensional resonances in $`SrTiO_3`$
One of the striking predictions of the above analysis is the occurrence of dimensional resonances discussed in an earlier section. These resonances occur when the dielectric constant is so large that the condition $`ka=(n+1/2)\pi `$ is satisfied. We have experimentally observed such resonances in single crystal samples of $`SrTiO_3`$ measured in a $`TE_{011}`$ cavity. The single crystal samples were purchased from Aesar Mfg. Co.. The experimental data are shown in Fig.3 (c) and (d), where $`\delta \zeta _H^{}(T)`$ and $`\zeta _H^{\prime \prime }(T)`$ are shown as a function of $`T`$ for a sample with dimensions $`0.5\times 0.5\times 0.5mm^3`$. The data clearly show resonances as a function of $`T`$. In this material $`\epsilon ^{}`$ increases strongly with decreasing $`T`$ approaching values of nearly $`1000`$. The experimental data shown in Fig.3 (c) and (d) are quantitatively consistent with the behavior in (a) and (b).
Inversion of Eq.8 shows a weakly $`T`$-dependent $`\epsilon ^{}850`$ between $`250K`$ and about $`75K`$. This value is entirely consistent with other measurements . However at lower temperatures the inversion of the $`\stackrel{~}{\zeta }_H`$ data to obtain $`\stackrel{~}{\epsilon }`$ is problematical because of the dimensional resonances. Here as noted before the solutions do not appear to be unique and spurious solutions are found.
An important parameter for microwave applications is the microwave loss in $`SrTiO_3`$. We find that $`\epsilon ^{\prime \prime }(T)`$ varies between $`0.10.25`$ in the temperature region between $`250K`$ and about $`75K`$ where smooth solutions of $`\epsilon ^{}850`$ are obtained. We also note that the raw experimental data for $`\stackrel{~}{\zeta }`$ indicate that the biggest resonances occur at $`62K`$ and $`37K`$ which is exactly where dielectric anomalies have been reported in lower temperature measurements .
### D: Conclusion
We have shown that a careful analysis of cavity perturbation methods, combined with the use of superconducting cavities, leads to a powerful method of measuring transport properties at microwave frequencies. The method can lead to exceptionally high sensitivities for the material properties.
A surprising result is that dielectric constants can be measured even though the sample is placed in a microwave magnetic field. One consequence of this conclusion is that many such experiments which use samples in microwave magnetic fields, such as non-resonant microwave absorption measurements , should be carefully analyzed for the influence of dielectric properties, and not just the magnetic properties.
The resulting microwave measurements show new dynamic phenomena with time scales corresponding to the GHz frequency ranges which is not seen in static dc SQUID susceptibility measurements. The microwave measurements yield information on dynamics at time scales $`10^{11}`$ sec., comparable to NS, but shorter than NMR and NQR ($`10^7`$ sec.) and $`\mu SR`$ ($`10^8`$ sec.), and are a sensitive probe of charge dynamics novel electronic materials. These results will lead us to a new perspective of how to understand other cuprates. In future publications, we will discuss the results of measurements on low dimensional spin systems and high temperature superconductors.
We thank A. Revcolevschi for providing samples of $`Sr_2CuO_3`$ and $`Sr_{14}Cu_{24}O_{41}`$, and R. S. Markiewicz for useful discussions. This research was supported by NSF-9711910, AFOSR-F30602-95-2-0011 and ONR-N00014-00-1-0002.
## Appendix A Spherical sample in magnetic field maximum of TE<sub>011</sub> mode
Brodwin and Parsons treated a spherical homogeneous sample with radius $`a`$ in a resonant cavity when the restrictions $`ka1`$ and $`k_oa1`$ are removed. They use a method developed by Stratton in which the electric and magnetic fields inside and outside the perturbing sample are expressed as expansions of spherical vector potential functions.
The field configurations of the $`TE_{011}`$ mode in cylindrical coordinate ($`\widehat{r},\widehat{\phi },\widehat{z})`$ are expressed as
$`\stackrel{}{H}(r,\phi ,z)`$ $`=`$ $`H_o{\displaystyle \frac{\pi }{\beta _{01}^{^{}}L}}J_1(\beta _{01}^{^{}}r)\mathrm{cos}({\displaystyle \frac{\pi z}{L}})\widehat{r}+H_oJ_0(\beta _{01}^{^{}}r)\mathrm{sin}({\displaystyle \frac{\pi z}{L}})\widehat{z}`$ (15)
$`\stackrel{}{E}(r,\phi ,z)`$ $`=`$ $`H_o{\displaystyle \frac{i\omega \mu _o}{\beta _{01}^{^{}}}}J_1(\beta _{01}^{^{}}r)\mathrm{sin}({\displaystyle \frac{\pi z}{L}})\widehat{\phi }`$
where $`H_o`$ is the maximum magnetic field in the center of the cavity, $`r_o`$ is the radius of the cavity, $`L`$ is the cavity axial length, and $`\beta _{01}^{^{}}r_o`$ is the first root of the Bessel function $`J_0^{}(\beta r_o)=0`$. A time dependence $`e^{i\omega t}`$ is assumed in $`\stackrel{}{H}`$ and $`\stackrel{}{E}`$. When a small sample with radius $`a`$ is put inside the cavity, the complex frequency shift $`\delta \stackrel{~}{\omega }`$ for $`TE_{011}`$ mode is given by the following expression
$$\frac{\delta \stackrel{~}{\omega }}{\omega }=\frac{i9\eta \mathrm{sin}^2\alpha }{2J_0^2(\beta _{01}^{}r_o)}\underset{n=1}{\overset{\mathrm{}}{}}\frac{2(2n+1)}{3n(n+1)}\left[\frac{P_n^{}(\mathrm{cos}\alpha )}{\mathrm{sin}\alpha }\right]^2\delta _{0n}\left(\frac{a_n^r}{\rho ^3}\right)$$
(16)
where $`\beta =\sqrt{k_o^2h^2}=\kappa _o\mathrm{sin}\alpha `$, $`h=\pi /L`$, and $`\eta =V_s/V_c`$ is the filling factor. $`a_n^r`$ is the coefficient corresponding to the reflected (scattered) field and is given by the following expression with $`\rho =k_oa`$ and $`N\rho =ka`$.
$$a_n^r=\frac{\stackrel{~}{\mu }j_n(N\rho )[\rho j_n(\rho )]^{^{}}+j_n(\rho )[N\rho j_n(N\rho )]^{^{}}}{\stackrel{~}{\mu }j_n(N\rho )[\rho h_n^1(\rho )]^{^{}}h_n^1(\rho )[N\rho j_n(N\rho )]^{^{}}}$$
(17)
In the following we define the sample geometrical factor $`\gamma `$ as
$$\gamma =\frac{\eta \mathrm{sin}^2\alpha }{J_0^2(\beta _{01}^{}r_o)}$$
(18)
Considering the cavity resonant frequency $`\omega _{mnp}=c\sqrt{\beta _{mn}^{}_{}{}^{}2+p^2\pi ^2/L^2}`$, $`\gamma `$ for the $`TE_{011}`$ mode can be rewritten as
$$\gamma =\frac{\eta }{J_0^2(\beta _{01}^{}r_o)\left[1+\left(\frac{\pi }{\beta _{01}^{}L}\right)^2\right]}$$
(19)
This series in Eq.16 is rapidly convergent for samples with diameters less than $`\lambda /2\pi ,`$ therefore the leading term give a good approximation for the frequency shift.
$$\frac{\delta \stackrel{~}{\omega }}{\omega }=\frac{i9\gamma a_1^r}{2\rho ^3}$$
(20)
Using the spherical Bessel functions: $`j_1(\rho )=[\mathrm{sin}(\rho )\rho \mathrm{cos}(\rho )]/\rho ^2`$, $`h_1^{(1)}(\rho )=e^{i\rho }(\rho +i)/\rho ^2`$, and $`[\rho j_1(\rho )]^{^{}}=[\mathrm{sin}(\rho )\rho \mathrm{cos}(\rho )]/\rho ^2+\mathrm{sin}(\rho )`$, we can examine the results of $`\delta \stackrel{~}{\omega }`$ in various limits.
### A: Extended Quasistatic Limit $`k_oa1`$
In this approximation, $`j_1(\rho )=\rho /3\rho ^3/30`$, $`h_1^{(1)}(\rho )=i/\rho ^2i/2+\rho /3+O(3)`$, and $`[\rho j_1(\rho )]^{^{}}=2\rho /32\rho ^3/15`$, $`[\rho h_1^{(1)}(\rho )]^{^{}}=i/\rho ^2i/2+2\rho /3+O(3)`$, Eq.16 can be written as
$$\frac{\delta \stackrel{~}{\omega }}{\omega }=\frac{3\gamma }{2}\left[\frac{2\stackrel{~}{\mu }j_1(N\rho )[N\rho j_1(N\rho )]^{^{}}}{\stackrel{~}{\mu }j_1(N\rho )+[N\rho j_1(N\rho )]^{^{}}}\right]+A_1(k_oa)^2$$
(21)
where $`A_1`$ is coefficient of the second order term of $`k_oa`$ and given by:
$`A_1`$ $`=`$ $`{\displaystyle \frac{3\gamma }{20}}\left[{\displaystyle \frac{4\stackrel{~}{\mu }j_1(N\rho )\left[N\rho j_1(N\rho )\right]^{^{}}}{\stackrel{~}{\mu }j_1(N\rho )+\left[N\rho j_1(N\rho )\right]^{^{}}}}\right]`$
$`{\displaystyle \frac{3\gamma }{4}}\left[{\displaystyle \frac{\left(2\stackrel{~}{\mu }j_1(N\rho )\left[N\rho j_1(N\rho )\right]^{^{}}\right)\left(\stackrel{~}{\mu }j_1(N\rho )\left[N\rho j_1(N\rho )\right]^{^{}}\right)}{\left(\stackrel{~}{\mu }j_1(N\rho )+\left[N\rho j_1(N\rho )\right]^{^{}}\right)^2}}\right]`$
#### A:.1 Quasistatic Limit $`k_oa1`$
Considering $`N\rho =ka1`$ the frequency shift $`\delta \stackrel{~}{\omega }`$ can be reduced to:
$$\frac{\delta \stackrel{~}{\omega }}{\omega }=3\gamma \frac{\stackrel{~}{\mu }1}{\stackrel{~}{\mu }+2}\frac{9\gamma }{10}\left[\frac{\stackrel{~}{\mu }^26\stackrel{~}{\mu }+4}{\left(\stackrel{~}{\mu }+2\right)^2}\left(k_oa\right)^2+\frac{\stackrel{~}{\mu }}{\left(\stackrel{~}{\mu }+2\right)^2}\left(ka\right)^2\right]$$
(23)
If $`\stackrel{~}{\mu }=1+\stackrel{~}{\chi }_m1`$, in addition to $`k_oa`$, $`ka1`$, then the above equation reduces to
$`{\displaystyle \frac{\delta \stackrel{~}{\omega }}{\omega }}`$ $``$ $`\gamma (\stackrel{~}{\mu }1)`$
$`=`$ $`\gamma \stackrel{~}{\chi }_M=\gamma (\chi _M^{^{}}+i\chi _M^{^{\prime \prime }})`$
Here the frequency shift $`\delta \stackrel{~}{\omega }`$ is a measurement of the complex magnetic susceptibility $`\stackrel{~}{\chi }_M`$.
#### A:.2 Pure conductor: Eddy current or skin depth limit: $`\stackrel{~}{\mu }=1`$, $`\stackrel{~}{\epsilon }=1`$, $`\stackrel{~}{\sigma }=\sigma `$
In this limit $`ka=(1+i)a/\delta ,`$ where $`\delta `$ $`=1/\sqrt{\mu _o\stackrel{~}{\sigma }\omega }`$ is the skin depth. Retaining the first order in the series of Eq.21, we obtain the complex frequency shift $`\delta \stackrel{~}{\omega }`$:
$`{\displaystyle \frac{\delta \stackrel{~}{\omega }}{\omega }}`$ $`=`$ $`{\displaystyle \frac{3\gamma }{2}}\left[{\displaystyle \frac{2j_1(N\rho )[N\rho j_1(N\rho )]^{^{}}}{j_1(N\rho )+[N\rho j_1(N\rho )]^{^{}}}}\right]`$
$``$ $`{\displaystyle \frac{3\gamma }{2}}\left[1{\displaystyle \frac{3}{(ka)^2}}+{\displaystyle \frac{3\mathrm{cot}ka}{ka}}\right]`$
By using $`\mathrm{cot}(x+iy)=\mathrm{sin}2x/(\mathrm{cosh}2y\mathrm{cos}2x)i\mathrm{sinh}2y/(\mathrm{cosh}2y\mathrm{cos}2x)`$, we obtain the following expressions:
$`\mathrm{R}e\left({\displaystyle \frac{\delta \stackrel{~}{\omega }}{\omega }}\right)`$ $`=`$ $`{\displaystyle \frac{3\gamma }{2}}\left[1{\displaystyle \frac{3}{2}}\left({\displaystyle \frac{\delta }{a}}\right)\left({\displaystyle \frac{\mathrm{sinh}\frac{2\delta }{a}\mathrm{sin}\frac{2\delta }{a}}{\mathrm{cosh}\frac{2\delta }{a}\mathrm{cos}\frac{2\delta }{a}}}\right)\right]`$ (26)
$`\mathrm{I}m\left({\displaystyle \frac{\delta \stackrel{~}{\omega }}{\omega }}\right)`$ $`=`$ $`{\displaystyle \frac{9\gamma }{4}}\left({\displaystyle \frac{\delta }{a}}\right)^2\left[1\left({\displaystyle \frac{a}{\delta }}\right)\left({\displaystyle \frac{\mathrm{sinh}\frac{2\delta }{a}+\mathrm{sin}\frac{2\delta }{a}}{\mathrm{cosh}\frac{2\delta }{a}\mathrm{cos}\frac{2\delta }{a}}}\right)\right]`$
In the low frequency limit where $`\delta a`$ the above formulas become:
$`\mathrm{R}e\left({\displaystyle \frac{\delta \stackrel{~}{\omega }}{\omega }}\right)`$ $`=`$ $`{\displaystyle \frac{4\gamma }{105}}\left({\displaystyle \frac{\delta }{a}}\right)^4`$ (27)
$`\mathrm{I}m\left({\displaystyle \frac{\delta \stackrel{~}{\omega }}{\omega }}\right)`$ $`=`$ $`{\displaystyle \frac{\gamma }{5}}\left({\displaystyle \frac{\delta }{a}}\right)^2`$
In the high frequency limit where $`\delta a`$ we obtain the expressions:
$`\mathrm{R}e\left({\displaystyle \frac{\delta \stackrel{~}{\omega }}{\omega }}\right)`$ $`=`$ $`{\displaystyle \frac{3\gamma }{2}}{\displaystyle \frac{9\gamma }{4}}\left({\displaystyle \frac{\delta }{a}}\right)`$ (28)
$`\mathrm{I}m\left({\displaystyle \frac{\delta \stackrel{~}{\omega }}{\omega }}\right)`$ $`=`$ $`{\displaystyle \frac{9\gamma }{4}}\left({\displaystyle \frac{\delta }{a}}\right)`$
Therefore the complex frequency shift $`\delta \stackrel{~}{\omega }`$ could be written in terms of surface impedance $`Z_s=R_siX_s`$:
$$\frac{\delta \stackrel{~}{\omega }}{\omega }=\frac{3\gamma }{2}\left[1\frac{3}{\omega \mu _oa}\left(X_s+iR_s\right)\right]$$
(29)
with $`R_s=X_s=\sqrt{\omega \mu _o/2\sigma }`$.
#### A:.3 Lossy Dielectric : $`\stackrel{~}{\mu }=1,`$ $`\stackrel{~}{\epsilon }=\epsilon ^{}+i\epsilon ^{\prime \prime }`$, $`\stackrel{~}{\sigma }=\sigma `$.
In this case, $`k^2=(\omega /c)^2(\stackrel{~}{\epsilon }+i\stackrel{~}{\sigma }/\omega \epsilon _o)`$, the frequency shift has a similar expression with the one derived for a perfect conductor but in this case the real and the imaginary part of the wave vector are not equal.
$$\frac{\delta \stackrel{~}{\omega }}{\omega }=\frac{3\gamma }{2}\left[1\frac{3}{(ka)^2}+\frac{3\mathrm{cot}ka}{ka}\right]$$
(30)
In the limit where $`ka1`$, the above equation can be written as
$`{\displaystyle \frac{\delta \stackrel{~}{\omega }}{\omega }}`$ $``$ $`{\displaystyle \frac{\gamma }{10}}(k_oa)^2(\stackrel{~}{\epsilon }+i{\displaystyle \frac{\stackrel{~}{\sigma }}{\omega \epsilon _o}}1)`$
$``$ $`{\displaystyle \frac{\gamma }{10}}(k_oa)^2\stackrel{~}{\chi }_P\text{ ; \hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}(when }\stackrel{~}{\sigma }=0\text{)}`$
where $`\stackrel{~}{\chi }_P\stackrel{~}{\epsilon }1=\epsilon ^{}1+i\epsilon ^{\prime \prime }=\chi _P^{}+i\chi _P^{\prime \prime }`$. Here the frequency shift $`\delta \stackrel{~}{\omega }`$ is a measurement of the complex dielectric susceptibility $`\stackrel{~}{\chi }_P`$ when $`\stackrel{~}{\sigma }=0`$.
## Appendix B Sample in TM<sub>110</sub> Electric Field Maximum
Although we have focussed on the $`TE_{011}`$ mode, it is also possible to carry out measurements using the $`TM_{110}`$ mode. For a sample placed in the cavity center at the microwave electric field maximum, the frequency shift is :
$$\frac{\delta \stackrel{~}{\omega }}{\omega }=\frac{i9\eta }{4J_1^2(\beta _{01}^{}r_o)}\underset{n=1}{\overset{\mathrm{}}{}}\frac{2(2n+1)}{3n(n+1)}\left[P_n^{}(0)\right]^2\delta _{0n}\left(\frac{b_n^r}{\rho ^3}\right)$$
(32)
with the reflection coefficient :
$$b_n^r=\frac{\stackrel{~}{\epsilon }j_n(N\rho )[\rho j_n(\rho )]^{^{}}+j_n(\rho )[N\rho j_n(N\rho )]^{^{}}}{\stackrel{~}{\epsilon }j_n(N\rho )[\rho h_n^1(\rho )]^{^{}}h_n^1(\rho )[N\rho j_n(N\rho )]^{^{}}}$$
(33)
In the first order the Eq.32 becomes:
$$\frac{\delta \stackrel{~}{\omega }}{\omega }=\frac{3\gamma ^{^{}}}{2}\left(\frac{2\stackrel{~}{\epsilon }j_1(N\rho )[N\rho j_1(N\rho )]^{^{}}}{\stackrel{~}{\epsilon }j_1(N\rho )+[N\rho j_1(N\rho )]^{^{}}}\right)$$
(34)
with a new geometrical factor $`\gamma ^{^{}}`$ given by:
$$\gamma ^{^{}}=\frac{\eta }{2J_1^2(\kappa _{01}r_o)}$$
(35)
where $`\kappa _{01}r_o`$ is the first root of the Bessel function $`J_0(\kappa r_o)=0`$
In the limit where $`ka1`$ the frequency shift is:
$$\frac{\delta \stackrel{~}{\omega }}{\omega }=3\gamma ^{^{}}\frac{\stackrel{~}{\epsilon }1}{\stackrel{~}{\epsilon }+2}+O(2)$$
(36) |
warning/0001/astro-ph0001310.html | ar5iv | text | # Detection of an X-Ray Hot Region in the Virgo Cluster of Galaxies with ASCA
## 1 Introduction
Recent X-ray observations are revealing significant large-scale variations of temperature and surface brightness in many clusters, providing evidence that clusters are evolving. Hydrodynamic simulations show that clusters recently formed through mergers should indicate a complex temperature structure, and become more regular with time (e.g. Roettiger, Burns, & Loken 1993; Takizawa 1999). Thus, spatial distributions of the temperature of the intracluster medium (ICM) provide important clues about the dynamical evolution and the present state of the cluster.
In this letter, we perform a detailed investigation on the temperature structure of the ICM in the Virgo cluster, based on extensive mapping observations with ASCA (Tanaka, Inoue, & Holt 1994). This nearest rich cluster enables ASCA to perform spatially resolved spectroscopy with moderate spatial resolution, and hot-gas properties can be studied in both galaxy scales ($`<100`$ kpc) and in the whole cluster scale ($`>1`$ Mpc). The Virgo cluster is thought to be a dynamically young system as recognized from its irregular structure in the optical and X-ray bands. Thus, the present mapping study of the cluster should provide us with valuable information to investigate the on-going heating process in the ICM.
We assume the distance to the Virgo cluster to be 20 Mpc (e.g. Federspiel et al. 1998), hence $`1^{}`$ angular separation at the cluster corresponds to 5.8 kpc. The solar number abundance of Fe relative to H is taken as $`4.68\times 10^5`$ (Anders & Grevesse 1989) throughout this letter.
## 2 Observation and Analysis
### 2.1 Observation
The mapping observations of the Virgo cluster have been carried out in December 1996 to December 1998, with 28 pointings and a total exposure time of $`500`$ ksec (Matsumoto et al. 1999; Ohashi et al. 1999; Yamasaki et al. 1999). Together with the data in the archive, the area covered with ASCA in the Virgo cluster is $`10`$ deg<sup>2</sup>. Figure 1 shows the ASCA observed regions overlayed on the X-ray contours with ROSAT (Bรถhringer et al. 1994). The radius of the circles is $`22^{}`$ corresponding to GIS field of view (Makishima et al. 1996; Ohashi et al. 1996).
We selected the GIS data observed with the telescope elevation angle from the Earth rim $`>5^{}`$, and the data taken with unstable attitude after maneuvers were discarded. Flare-like events due to the background fluctuation were also excluded (Ishisaki 1996). The cosmic X-ray background (CXB) was estimated from the archival data taken during 1993โ1994 (Ikebe 1995), and the long-term variability of the non X-ray background of the GIS (Ishisaki 1996) was corrected for.
### 2.2 Image Analysis
To derive the pure ICM component, contaminating X-ray sources have to be excluded. We carried out a source detection analysis developed for the CXB study by Ueda (Ueda et al. 1999), who dealt with the complicated detector response in a systematic way including the position- and energy-dependence of the point spread function (PSF) of the ASCA X-ray telescope. Pointings containing bright sources such as M87, M49, A1553, and A1541 were excluded from the analysis, leaving 29 pointings to be analyzed (indicated by blue circles in Figure 1). We adopt a rather low flux level in detecting the source candidates, $`3\sigma `$ above the background in $`0.77`$ keV band, since our interest is in the remaining diffuse component. The analysis detected 231 source candidates with X-ray flux $`1\times 10^{13}`$ erg cm<sup>-2</sup> s<sup>-1</sup> in the $`210`$ keV band and all of them have been masked out from the mapping data (see Figure 1). The mask regions centered on the candidate positions have radii depending on the source flux, since brighter sources affect wider regions due to the image spread by the PSF effect. The mask radius is determined where the surface brightness due to the source drops to less than 5% of the ICM level.
To examine spatially resolved spectral features, we need to know the surface brightness distribution in the whole cluster to estimate the amount of the stray light, which consists of photons generated outside the field of view (e.g. Honda et al. 1996). Fortunately, the RASS (ROSAT All-Sky Survey, Bรถhringer et al. 1994; Voges et al. 1996) data is available for this purpose, and we can produce the template of the brightness profile. In the RASS image, diffuse soft X-ray emission, which possibly comes from the rim of LOOP I (e.g. Raymond 1984; Egger & Aschenbach 1995), is present in the direction of the Virgo cluster (Snowden et al. 1995). Therefore, we use the PSPC data only above 0.9 keV to exclude possible soft X-ray contamination. Spatially uniform background, obtained from a blank sky region, was subtracted from the Virgo RASS data. Based on this template image, a ray-tracing simulation (Tsusaka et al. 1995) for the GIS observation was carried out assuming a uniform temperature of 2.0 keV and a metallicity of 0.2 solar, which are the typical parameters in the Virgo field (e.g. Koyama, Takano, & Tawara 1991; Matsumoto et al. 1999).
As a result, we found that the intermediate region between M87 and M49 was almost free from stray light from M87 and M49. The contaminating flux from these 2 galaxies is less than a few %. Considering the complex spectral structures (such as temperature and abundance gradients) in these galaxies, this makes the data analysis for the intermediate region much easier.
### 2.3 Spectral Analysis
To derive spectral parameters (such as temperature and surface brightness) in a region, we have to know parameters in the surrounding regions to evaluate the contamination of stray light. We carried out a first-order estimation of the spectral parameters, adopting an analysis method developed by Honda et al. (1996). This is performed by fitting individual spectra with a modified response functions that partly compensates the effect of the stray light. The RASS image obtained in the previous section is used to estimate the amount of the stray light.
A spectral analysis has been performed for each pointed region in the $`0.78`$ keV band with a Raymond-Smith model (Raymond & Smith 1977, hereafter R-S model). The interstellar absorption $`N_\mathrm{H}`$ is fixed to the Galactic value ($`1.72.5\times 10^{20}`$ cm<sup>-2</sup>). The temperature distribution derived from the GIS spectral fits is shown in Figure 2, with a color-coded plot of temperature in the left panel and a plot as a function of distance from M87 in the right panel, respectively.
As shown in Figure 2, the temperature around M87 is $`2.5`$ keV and slightly decreases to $`2.0`$ keV at $`1^{}`$ away from M87 in the northwest region. In the south region, the average temperature at a distance of $`2^{}`$ from M87 is still $`2.5`$ keV, with a large scatter from 1.8 keV to 3.4 keV. The metal abundance is poorly constrained in most of the regions because of low photon statistics. We only mention that the best-fit values suggest that the metal abundance in the general cluster regions is around 0.2 solar with a scatter of about $`\pm 0.2`$ solar from position to position. If we fitted the spectra with free absorption, the data generally require that no absorption (even the Galactic $`N_\mathrm{H}`$) is present. This suggests existence of an additional soft component below $`1`$ keV, which may be the foreground emission of the Galactic soft X-rays.
## 3 The Hot Region
As shown in Figure 2, three regions W1, W2 and W3 along the โemission bridgeโ between M87 and M49 show the ICM temperatures rising to $`3`$ keV. To improve the statistics, the three spectra are combined (hereafter called W123) because individual fits indicate statistically the same temperature. For comparison, the spectra for regions E1, E2, and E3 (hereafter E123), just to the east of W123, are also combined. Both W123 and E123 regions are elongated in parallel to the โemission bridgeโ, and the distance from M87 and M49 is almost the same. Errors in the contaminating spectra from nearby bright sources (i.e. A1553, NGC 4325, A1541, and QSO 1225+089) and those in the remaining fluxes of masked-out sources are the major origin of the systematic error for temperatures in W123 and E123. This error is found to be less than 0.2 keV, and its effect works on the two temperatures in the same sense: thus keeping the temperature difference the same.
The single-temperature model gives a poor fit in the energy range $`0.78.0`$ keV for the 2 spectra W123 and E123. As shown in Table 1, the best-fit results are $`\chi ^2/\nu =45.2/21`$ and $`31.1/21`$ for W123 and E123, respectively. This is mainly due to excess emission below 1 keV in both spectra. A two-component (R-S and a soft thermal bremsstrahlung) model improves the fit to $`\chi ^2/\nu =21.8/20`$ and $`\chi ^2/\nu =27.2/20`$ for W123 and E123, respectively (see Figure 3). The additional thermal bremsstrahlung component yields the best-fit temperature $`kT=0.20.3`$ keV with $`F_\mathrm{X}0.8\times 10^{15}`$ erg cm<sup>-2</sup> s<sup>-1</sup> arcmin<sup>-2</sup> for $`0.52`$ keV for both W123 and E123 spectra. These values are close to the ROSAT result ($`kT=0.15`$ keV and $`F_\mathrm{X}=0.6\times 10^{15}`$ ergs cm<sup>-2</sup> s<sup>-1</sup> arcmin<sup>-2</sup> in $`0.52`$ keV band, Irwin & Sarazin (1996)), supporting the view that the soft emission is due to the Galactic hot interstellar medium. These results indicate that the region W123 has a significantly high temperature of $`4`$ keV, while E123, just in the east of W123, shows $`2`$ keV which is the typical temperature of the Virgo cluster.
Based on these results, we neglect the energy range below 2 keV and look into the pure ICM component in the region W123. We also subtracted contaminating photons, which come from the surrounding region of W123, assuming the temperature and metallicity of the surrounding ICM are 2 keV and 0.2 solar, respectively. Fixing the abundance to 0.2 solar, acceptable fits with R-S model are obtained with $`\chi ^2/\nu =10.5/11`$ (Table 1). The ray-tracing simulation gives a systematic error for the stray-light intensity by $`30\%+10\%`$ as estimated from offset observations of the Crab nebula (Ishisaki 1996). The systematic error due to a fluctuation of the CXB flux is $`10`$% for the GIS field of view, and the uncertainty in the estimation of the non X-ray background level is $`6`$%. Including all these errors, we can conclude that the temperature in W123 is still higher than that in E123 with more than 90% confidence. The total flux of the hot region in $`210`$ keV band is ($`9.210.3`$) $`\times 10^{12}`$ erg cm<sup>-2</sup> s<sup>-1</sup> at the 90% confidence limit, and the emission measure $`n^2๐l`$ is estimated to be ($`3.45.0`$) $`\times 10^{16}`$ cm<sup>-5</sup> assuming an extent of the hot region to be $`4\times 10^3`$ arcmin<sup>2</sup>.
So far, the โhotโ emission has been assumed to have a thermal spectrum. However, the data also allow non-thermal (power-law) models. Spectral fit for the $`28`$ keV hot region data (W123) with a power-law model gives an acceptable result of $`\chi ^2/\nu =9.9/11`$ with a photon index between $`1.72.3`$ at the 90% confidence (see Table 1). Since no significant Fe-K line is seen in the W123 spectrum with $`EW821`$ eV for a 6.7 keV line at the 90% confidence, the diffuse non-thermal emission remains as a possibility from the ASCA observations.
## 4 Discussion
The previous Ginga observations have suggested a temperature rise in the ICM from M87 to M49 (Takano 1990; Koyama, Takano, & Tawara 1991). However, ROSAT data showed no such evidence (Bรถhringer et al. 1994) and implied a possibility that the non-imaging Ginga data were contaminated by background sources. The extensive mapping observations from ASCA have shown the correct temperature structure in the Virgo cluster for the first time and unambiguously detected an unusual โhotโ region in the ICM. This detection provides a clear evidence that the Virgo cluster is a young system in which a local gas heating is taking place now in the cluster outskirts.
The emission measure of the hot region W123 obtained in the previous section gives a rough estimate of the gas density $`n`$ to be of the order of $`1\times 10^4`$ cm<sup>-3</sup>. Here, we assume that the line-of-sight depth of the hot component is $`300`$ kpc which is the same order as the projected length of the region. The internal thermal energy of the hot component, $`E_{\mathrm{th}}=VnkT`$ where $`V`$ and $`T`$ are volume and temperature, is calculated as $`10^{60}`$ ergs. This level of energy is orders of magnitude lower than the kinetic energy involved in a typical subcluster merger ($`10^{6364}`$ ergs). This means that if there is a bulk motion of gas in this region with $`v1000`$ km s<sup>-1</sup>, caused by an infall of galaxies or a small group of galaxies, then it can supply enough energy to heat up the gas to the observed temperature.
One might feel some difficulty to heat up a localized spot by a merger, however, such a local effect could be produced by the merging of a subclump of the irregular M49 subcluster. Time scale for thermal conduction is roughly estimated as $`t_{\mathrm{cond}}8\times 10^8`$ yr for the gas density in the hot region and the scale length of the temperature gradient to be 500 kpc. If the gas moving with $`v1000`$ km s<sup>-1</sup> receives some heat input, the hot region would become elongated by $`800`$ kpc because of the slow heat conduction. This situation could be related with the observed north-south elongation of the hot region.
Honda et al. (1996) reported temperature variation of the ICM in the Coma cluster, and found a remarkable hot region ($`11`$ keV) which is distinct from the average temperature of the whole cluster ($`8`$ keV). This hot region is located at $`40^{}`$ (1.6 Mpc) offset from the cluster center, and has an angular extent of $`20^{}`$ radius. We can roughly estimate the extra internal energy in the Coma hot region as $`8\times 10^{61}`$ ergs, which is nearly 2 orders of magnitude higher than that in the Virgo case.
Irwin & Sarazin (1996) discuss that M49 is moving supersonically ($`v1300`$ km s<sup>-1</sup>) in the Virgo ICM toward the direction of M87. The gravitational mass of M49 subcluster is estimated as $`8.7\times 10^{13}M_{}`$ from the ROSAT observation (Schindler, Binggeli, & Bรถhringer 1999). Then the kinetic energy of the M49 subcluster is roughly estimated as $`1\times 10^{63}`$ ergs, which is sufficiently large to heat up the hot-region gas. Using Rankine-Hugoniot jump condition (e.g. Shu 1992), the Mach number of the shock wave to heat up 2 keV gas to $`4`$ keV should be $`2`$. Since the sound velocity of the 2 keV gas is $`700`$ km s<sup>-1</sup>, the required velocity is close to the Irwin & Sarazin result.
Hard X-ray ($`kT10`$ keV) emission from clusters of galaxies was reported from previous observations (e.g. Fusco-Femiano et al. 1999 for Coma cluster), and the existence of non-thermal emission has been suggested. For the hot region detected here, we cannot confirm whether the origin of the hard emission is thermal or non-thermal, because of the lack of observational evidences. If we assume that relativistic electrons are produced by first-order Fermi acceleration, the momentum spectrum of the electrons is described as $`N(p)=N_0p^\mu `$. Here, $`\mu =(r+2)/(r1)`$ and $`r`$ is a ratio of the shock compression. For the shock with a Mach number $`2`$, the exponent is implied as $`\mu 3.3`$. In such a steep spectra, an energy loss due to nonthermal bremsstrahlung dominates the inverse Compton loss (Sarazin & Kempner 1999). The electrons lose their energy through Coulomb loss, whose time scale is estimated as $`t_{\mathrm{Coul}}3\times 10^8\gamma `$ yr. This is similar to the $`t_{\mathrm{cond}}`$ estimated above.
Above considerations suggest that in both thermal and non-thermal cases, the extra energy built up in the hot region would dissipate away within about 1 Gyr, due to thermal conduction or Coulomb loss. Therefore, it seems likely that the energy supply into the hot region has started only within the past 1 Gyr, or alternatively a long continuous supply of energy has been occurring here over a cosmological time scale. The infall of the M49 subcluster can supply energy into ICM for a very long time and is probably connected with the local gas heating as detected in the Virgo cluster.
We thank Y. Ueda and Y. Ishisaki for their support of source detection analysis and background estimation. Stimulating discussion with T. Reiprich, C. Sarazin, K. Masai, S. Okamura and M. Takizawa are also acknowledged. K. K. acknowledges hospitality in MPE and support from the Japan Science and Technology Corporation (JST). This work is partly supported by the Grants-in Aid of the Ministry of Education, Science, Sports and Culture of Japan, 08404010. |
warning/0001/quant-ph0001105.html | ar5iv | text | # 1 Introduction
## 1 Introduction
Possessing superposition and entanglement properties, quantum information has revealed many interesting features which classical information has no counterpart. Unlike classical information, quantum information cannot be duplicated, i.e. unknown quantum state cannot be copied exactly . However a universal quantum (approximate) cloner has been introduced which takes an unknown quantum state and generates multiple copies, with fidelity 5/6 in case of two copies as outputs regardless of an input. Another feature of quantum information has been discovered recently where it was shown that more quantum information can be gained from two anti-parallel spins than from two parallel ones, i.e. one can measure the spin direction $`|๐ง`$ with better fidelity when two qubits are in anti-parallel, $`|๐ง,๐ง`$, than in parallel, $`|๐ง,๐ง`$.
In this paper, we consider a universal quantum anti-cloner which takes an unknown quantum state just as in quantum cloner but its output as one with the same copy while the second one with opposite spin direction to the input state. For the Bloch vector, an input $`๐ง`$, quantum anti-cloner would have the input as $`\frac{1}{2}(\mathrm{๐}+๐ง\sigma )`$, then it generates two outputs, $`\frac{1}{2}(\mathrm{๐}+\eta ๐ง\sigma )`$ and $`\frac{1}{2}(\mathrm{๐}\eta ๐ง\sigma )`$, where $`0\eta 1`$ is the shrinking factor and the fidelity is defined as $`F=๐ง|\rho ^{(\mathrm{o}ut)}|๐ง=\frac{1}{2}(1+\eta )`$. If spin flipping were allowed then anti-cloner would have the same fidelity as the regular cloner since one could clone first then flip the spin of the second copy. However spin flipping of an unknown state is not allowed in quantum mechanics. Consider a spin-flipping of an unknown state,
$$\left(\begin{array}{c}e^{\frac{i\phi }{2}}\mathrm{cos}\frac{\theta }{2}\\ e^{\frac{i\phi }{2}}\mathrm{sin}\frac{\theta }{2}\end{array}\right)\left(\begin{array}{c}e^{\frac{i\phi }{2}}\mathrm{sin}\frac{\theta }{2}\\ e^{\frac{i\phi }{2}}\mathrm{cos}\frac{\theta }{2}\end{array}\right)$$
(1)
This transformation can be done only by an anti-unitary operation where the anti-unitary transformation, $`V`$, satisfies the following two conditions
$`(i)`$ $`|\psi |\varphi |=|\psi ^{}|\varphi ^{}|`$
$`(ii)`$ $`V\left(a|0+b|1\right)=a^{}V|0+b^{}V|1`$ (2)
where $`V|\psi |\psi ^{}`$ and $`V|\varphi |\varphi ^{}`$.
In sect. 2, we derive a unitary transformation for a optimal universal quantum anti-cloner where we obtain 2/3 for fidelity. This value is equal to the fidelity of measurement which is for a given single unknown state, how precisely one can determine its state . In sect. 3, we show that the quantum state can be anti-cloned exactly with non-zero probability. In case of two states, the probability of exact anti-cloning is higher than the probability of distinguishing between the two states. We conclude with discussions on further prospects on related issues.
## 2 Universal quantum anti-cloning
In this section, we study the unitary transformation with optimal fidelity for a universal quantum anti-cloner. Let us consider an input state $`|๐ง=\alpha |0+\beta |1`$ such that for an input density matrix,
$$\rho ^{(\mathrm{i}n)}=\left(\begin{array}{cc}|\alpha |^2& \alpha \beta ^{}\\ \beta \alpha ^{}& |\beta |^2\end{array}\right)=\left(\begin{array}{cc}1+n_z& n_xin_y\\ n_x+in_y& 1n_z\end{array}\right)$$
(3)
the output density matrix yields the first particle same as the input while the second one with opposite spin direction as follows
$$\rho _1^{(\mathrm{o}ut)}=\frac{\mathrm{๐}+\eta ๐ง\sigma }{2}=\frac{1}{2}\left(\begin{array}{cc}1+\eta n_z& \eta (n_xin_y)\\ \eta (n_x+in_y)& 1\eta n_z\end{array}\right)$$
(4)
$$\rho _2^{(\mathrm{o}ut)}=\frac{\mathrm{๐}\eta ๐ง\sigma }{2}=\frac{1}{2}\left(\begin{array}{cc}1\eta n_z& \eta (n_xin_y)\\ \eta (n_x+in_y)& 1+\eta n_z\end{array}\right)$$
(5)
We want to consider the constraints in order to satisfy the output density matrices (4,5) with maximum fidelity, i.e. $`\eta `$. The conditions (4) and (5) imply that the two output density matrices as symmetric except its spin direction which are opposite to each other. We also impose the universality constraint that the fidelity deos not depend on the input state $`|๐ง`$.
Let us consider the following general transformation,
$`|0|Q_{23}`$ $``$ $`a|00|A+b|01|B+c|10|C+d|11|D`$
$`|1|Q_{23}`$ $``$ $`\stackrel{~}{a}|11|\stackrel{~}{A}+\stackrel{~}{b}|10|\stackrel{~}{B}+\stackrel{~}{c}|01|\stackrel{~}{C}+\stackrel{~}{d}|00|\stackrel{~}{D}`$ (6)
where $`|Q_{23}`$ is the state to be anti-cloned and the initial ancilla state and the ancillas, $`|A,\mathrm{},|\stackrel{~}{D}`$, are normalised but not necessarily orthogonal. After following the transformation (6) for the input state $`|๐ง`$, we have the following reduced density matrices after tracing out 23 and 13, respectively,
$`\rho _1`$ $`=`$ $`\{(|a|^2+|b|^2)|\alpha |^2+(a\stackrel{~}{d}^{}\stackrel{~}{D}|A+b\stackrel{~}{c}^{}\stackrel{~}{C}|B)\alpha \beta ^{}`$ (7)
$`+`$ $`(\stackrel{~}{c}b^{}B|\stackrel{~}{C}+\stackrel{~}{d}a^{}A|\stackrel{~}{D})\beta \alpha ^{}+(|\stackrel{~}{c}|^2+|\stackrel{~}{d}|^2)|\beta |^2\}|00|`$
$`+`$ $`\{(ac^{}C|A+bd^{}D|B)|\alpha |^2+(a\stackrel{~}{b}^{}\stackrel{~}{B}|A+b\stackrel{~}{a}^{}\stackrel{~}{A}|B)\alpha \beta ^{}`$
$`+`$ $`(\stackrel{~}{c}d^{}D|\stackrel{~}{C}+\stackrel{~}{d}c^{}C|\stackrel{~}{D})\beta \alpha ^{}+(\stackrel{~}{c}\stackrel{~}{a}^{}\stackrel{~}{A}|\stackrel{~}{C}+\stackrel{~}{d}\stackrel{~}{b}^{}\stackrel{~}{B}|\stackrel{~}{D})|\beta |^2\}|01|`$
$`+`$ $`\{(ca^{}A|C+db^{}B|D)|\alpha |^2+(c\stackrel{~}{d}^{}\stackrel{~}{D}|C+d\stackrel{~}{c}^{}\stackrel{~}{C}|D)\alpha \beta ^{}`$
$`+`$ $`(\stackrel{~}{a}b^{}B|\stackrel{~}{A}+\stackrel{~}{b}a^{}A|\stackrel{~}{B})\beta \alpha ^{}+(\stackrel{~}{a}\stackrel{~}{c}^{}\stackrel{~}{C}|\stackrel{~}{A}+\stackrel{~}{b}\stackrel{~}{d}^{}\stackrel{~}{D}|\stackrel{~}{B})\beta ^2\}|10|`$
$`+`$ $`\{(|c|^2+|d|^2)\alpha ^2+(c\stackrel{~}{b}^{}\stackrel{~}{B}|C+d\stackrel{~}{a}^{}\stackrel{~}{A}|D)\alpha \beta ^{}`$
$`+`$ $`(\stackrel{~}{a}d^{}D|\stackrel{~}{A}+\stackrel{~}{b}c^{}C|\stackrel{~}{B})\beta \alpha ^{}+(|\stackrel{~}{a}|^2+|\stackrel{~}{b}|^2)|\beta |^2\}|11|`$
and
$`\rho _2`$ $`=`$ $`\{(|a|^2+|c|^2)|\alpha |^2+(a\stackrel{~}{d}^{}\stackrel{~}{D}|A+c\stackrel{~}{b}^{}\stackrel{~}{B}|C)\alpha \beta ^{}`$ (8)
$`+`$ $`(\stackrel{~}{b}c^{}C|\stackrel{~}{B}+\stackrel{~}{d}a^{}A|\stackrel{~}{D})\beta \alpha ^{}+(|\stackrel{~}{b}|^2+|\stackrel{~}{d}|^2)|\beta |^2\}|00|`$
$`+`$ $`\{(ab^{}B|A+cd^{}D|C)|\alpha |^2+(a\stackrel{~}{c}^{}\stackrel{~}{C}|A+c\stackrel{~}{a}^{}\stackrel{~}{A}|C)\alpha \beta ^{}`$
$`+`$ $`(\stackrel{~}{b}d^{}D|\stackrel{~}{B}+\stackrel{~}{d}b^{}B|\stackrel{~}{D})\beta \alpha ^{}+(\stackrel{~}{b}\stackrel{~}{a}^{}\stackrel{~}{A}|\stackrel{~}{B}+\stackrel{~}{d}\stackrel{~}{c}^{}\stackrel{~}{C}|\stackrel{~}{D})|\beta |^2\}|01`$
$`+`$ $`\{(ba^{}A|B+dc^{}C|D)|\alpha |^2+(b\stackrel{~}{d}^{}\stackrel{~}{D}|B+d\stackrel{~}{b}^{}\stackrel{~}{B}|D)\alpha \beta ^{}`$
$`+`$ $`(\stackrel{~}{a}c^{}C|\stackrel{~}{A}+\stackrel{~}{c}a^{}A|\stackrel{~}{C})\beta \alpha ^{}+(\stackrel{~}{a}\stackrel{~}{b}^{}\stackrel{~}{B}|\stackrel{~}{A}+\stackrel{~}{c}\stackrel{~}{d}^{}\stackrel{~}{D}|\stackrel{~}{C})|\beta |^2\}|10|`$
$`+`$ $`\{(|b|^2+|d|^2)|\alpha |^2+(b\stackrel{~}{c}^{}\stackrel{~}{C}|B+d\stackrel{~}{a}^{}\stackrel{~}{A}|D)\alpha \beta ^{}`$
$`+`$ $`(\stackrel{~}{a}d^{}D|\stackrel{~}{A}+\stackrel{~}{c}b^{}B|\stackrel{~}{C})\beta \alpha ^{}+(|\stackrel{~}{a}|^2+|\stackrel{~}{c}|^2)|\beta |^2\}|11|`$
We want to consider constraints for $`\rho _1`$ and $`\rho _2`$ in (7,8) to be same as $`\rho _1^{(\mathrm{o}ut)}`$ and $`\rho _2^{(out)}`$ in (4,5) with maximum value for $`\eta `$. Let us write the coefficients as follows
$$a=|a|e^{i\delta _a},b=|b|e^{i\delta _b},c=|c|e^{i\delta _c},d=|d|e^{i\delta _d}$$
(9)
and likewise for tilded cases. Also we could write $`A|B=|A|B|e^{i\delta _{AB}}`$ and others are similarly defined. First, there are normalisation conditions to be satisfied for the transformation (6),
$`|a|^2+|b|^2+|c|^2+|d|^2`$ $`=`$ $`1`$
$`|\stackrel{~}{a}|^2+|\stackrel{~}{b}|^2+|\stackrel{~}{c}|^2+|\stackrel{~}{d}|^2`$ $`=`$ $`1`$ (10)
and the orthogonality
$$a^{}\stackrel{~}{d}A|\stackrel{~}{D}+c^{}\stackrel{~}{b}C|\stackrel{~}{B}+b^{}\stackrel{~}{c}B|\stackrel{~}{C}+d^{}\stackrel{~}{a}D|\stackrel{~}{A}=0$$
(11)
Comparing $`n_z`$ terms in $`\rho _1^{(\mathrm{o}ut)}`$ and $`\rho _2^{(\mathrm{o}ut)}`$, we get the following constraints from (7,8)
$$|a|=|d|,|\stackrel{~}{a}|=|\stackrel{~}{d}|$$
(12)
$$a\stackrel{~}{d}^{}\stackrel{~}{D}|A+b\stackrel{~}{c}^{}\stackrel{~}{C}|Bc\stackrel{~}{b}^{}\stackrel{~}{B}|Cd\stackrel{~}{a}^{}\stackrel{~}{A}|D=0$$
(13)
and
$$\eta =|b|^2|c|^2=2|b|^2+2|a|^21$$
(14)
where the last relation in (14) results from (10) and (12) Next, Comparing $`n_x`$ and $`n_y`$ terms yields
$`\eta `$ $`=`$ $`\mathrm{R}e[a^{}\stackrel{~}{b}A|\stackrel{~}{B}+b^{}\stackrel{~}{a}B|\stackrel{~}{A}]`$ (15)
$`=`$ $`\mathrm{R}e[\stackrel{~}{c}a^{}A|\stackrel{~}{C}+\stackrel{~}{a}c^{}C|\stackrel{~}{A}]`$ (16)
and also the following must be satisfied.
$`\mathrm{I}m[a^{}\stackrel{~}{b}A|\stackrel{~}{B}+b^{}\stackrel{~}{a}B|\stackrel{~}{A}]=0`$ (17)
$`\mathrm{I}m[\stackrel{~}{c}a^{}A|\stackrel{~}{C}+\stackrel{~}{a}c^{}C|\stackrel{~}{A}]=0`$ (18)
$`b\stackrel{~}{d}^{}\stackrel{~}{D}|B+d\stackrel{~}{b}^{}\stackrel{~}{B}|D=0`$ (19)
$`ca^{}A|C+db^{}B|D=0`$ (20)
$`\stackrel{~}{a}\stackrel{~}{c}^{}\stackrel{~}{C}|\stackrel{~}{A}+\stackrel{~}{b}\stackrel{~}{d}^{}\stackrel{~}{D}|\stackrel{~}{B}=0`$ (21)
$`c\stackrel{~}{d}^{}\stackrel{~}{D}|C+d\stackrel{~}{c}^{}\stackrel{~}{C}|D=0`$ (22)
$`a^{}bA|B+c^{}dC|D=0`$ (23)
$`\stackrel{~}{b}^{}\stackrel{~}{a}\stackrel{~}{B}|\stackrel{~}{A}+\stackrel{~}{d}^{}\stackrel{~}{c}\stackrel{~}{D}|\stackrel{~}{C}=0`$ (24)
For the transformation (6), we could also impose the constraint such that the output reduced density matrices do not change under $`|0|1`$, then the following is true,
$$|a|=|\stackrel{~}{a}|,|b|=|\stackrel{~}{b}|,|c|=|\stackrel{~}{c}|$$
(25)
From (14,15,16)
$`\eta `$ $`=`$ $`|a||b|\mathrm{R}e[e^{i(\delta _a\delta _{\stackrel{~}{b}}+\delta _{A\stackrel{~}{B}})}|A|\stackrel{~}{B}|+e^{i(\delta _b\delta _{\stackrel{~}{a}}+\delta _{B\stackrel{~}{A}})}|B|\stackrel{~}{A}|]`$ (26)
$`=`$ $`|a||c|\mathrm{R}e[e^{i(\delta _{\stackrel{~}{c}}\delta _a+\delta _{A\stackrel{~}{C}})}|A|\stackrel{~}{C}|+e^{i(\delta _{\stackrel{~}{a}}\delta _c+\delta _{C\stackrel{~}{A}})}|C|\stackrel{~}{A}|]`$ (27)
then the maximum $`\eta `$ can be obtained when Re part in (27) is maximum, i.e. 2. Therefore with (14), following conditions can be obtained,
$$|a|^2+|c|^2=\frac{1\eta }{2},|a||c|=\frac{\eta }{2}$$
(28)
then
$`(|a||c|)^2`$ $`=`$ $`|a|^2+|c|^22|a||c|`$ (29)
$`=`$ $`{\displaystyle \frac{1\eta }{2}}\eta 0`$
$``$ $`\eta {\displaystyle \frac{1}{3}}`$
Therefore the maximum of $`\eta `$ is 1/3. For $`\eta =\frac{1}{3}`$,
$`|\stackrel{~}{a}|=|a|`$ $`=`$ $`\sqrt{{\displaystyle \frac{1}{6}}},|\stackrel{~}{b}|=|b|=\sqrt{{\displaystyle \frac{1}{2}}}`$ (30)
$`|\stackrel{~}{c}|=|c|`$ $`=`$ $`\sqrt{{\displaystyle \frac{1}{6}}},|\stackrel{~}{d}|=|d|=\sqrt{{\displaystyle \frac{1}{6}}}`$ (31)
and the minus sign for (27) and (26) can be satisfied with the following phase choice.
$$\delta _c=\delta _{\stackrel{~}{c}}=\pi ,\delta _b=\delta _{\stackrel{~}{b}}=\mathrm{cos}^1(\frac{1}{\sqrt{3}})$$
(32)
while all other phases, $`\delta _a,\mathrm{},\delta _{A\stackrel{~}{B}},\mathrm{}`$, vanish. The ancillas satisfying the constraint (11),(13),(17-24) can be of the following form,
$`|A`$ $`=`$ $`(1,0,0,0),|\stackrel{~}{A}=(0,1,0,0)`$ (33)
$`|B`$ $`=`$ $`(0,1,0,0),|\stackrel{~}{B}=(1,0,0,0)`$ (34)
$`|C`$ $`=`$ $`(0,1,0,0),|\stackrel{~}{C}=(1,0,0,0)`$ (35)
$`|D`$ $`=`$ $`(0,0,1,0),|\stackrel{~}{D}=(0,0,0,1)`$ (36)
with the usual basis $`|00,|01,|10,|11`$. Then the fidelity $`\frac{2}{3}`$ can be obtained with the following transformation
$`|0|Q`$ $``$ $`\sqrt{{\displaystyle \frac{1}{6}}}|0000+\sqrt{{\displaystyle \frac{1}{2}}}\mathrm{e}xp[i\mathrm{cos}^1({\displaystyle \frac{1}{\sqrt{3}}})]|0101\sqrt{{\displaystyle \frac{1}{6}}}|1001+\sqrt{{\displaystyle \frac{1}{6}}}|1110`$ (37)
$`|1|Q`$ $``$ $`\sqrt{{\displaystyle \frac{1}{6}}}|1101+\sqrt{{\displaystyle \frac{1}{2}}}\mathrm{e}xp[i\mathrm{cos}^1({\displaystyle \frac{1}{\sqrt{3}}})]|1000\sqrt{{\displaystyle \frac{1}{6}}}|0100+\sqrt{{\displaystyle \frac{1}{6}}}|0011`$ (38)
Note that the fidelity of universal anti-cloner, $`F_{OACM}`$, is same as the measurement fidelity, which is 2/3. Hence one way to implement optimal anti-cloning is to measure the unknown input state and prepare two qubits with opposite spin directions. It is also implied that fidelity of spin flipping also should be bounded from below by 2/3, i.e. $`F_{OSFM}F_{SFM}^{}=\frac{2}{3}`$, since after anti-cloning, one can throw away the first qubit and will be left with the second qubit which has opposite direction to the input state. Therefore, the following holds,
$$F_{OACM}F_{OSFM}$$
(39)
In <sup>3</sup><sup>3</sup>3 In , the term Universal NOT-gate was used rather than spin flipping, it was claimed that the optimal spin flipping is achieved with 2/3 fidelity including classical measurement. Due to this classical information, one can prepare additional qubit as the original input state (i.e. opposite to the output) which implies $`F_{OACM}F_{ACM}^{}=\frac{2}{3}`$, therefore
$$F_{OACM}F_{OSFM}$$
(40)
Therefore, equality between $`F_{OACM}`$ and $`F_{OSFM}`$ holds.
## 3 Probabilistic quantum anti-cloning
There is another type of imperfect cloning, a probabilistic cloner. Duan and Guo showed that there can be a unitary transformation such that linearly independent states can be cloned perfectily, with non-zero probability. Can anti-cloning be done probabilistically, i.e. can we find a unitary transformation such that $`|๐ฆ_i|0\stackrel{\mathrm{P}rob0}{}|๐ฆ_i|๐ฆ_i,i=1,\mathrm{},n`$, can be achieved. In order to show it, we follow Duan and Guoโs method with the following transformation,
$$U(|๐ฆ_i|0|Y_0)=\sqrt{f}|๐ฆ_i|๐ฆ_i|Y_0+\underset{j=1}{\overset{n}{}}a_{ij}|Q_{12}^{(j)}|Y_j$$
(41)
where $`|Y_0`$ and $`|Y_j`$ are orthonormal probe, such that whether cloning was successful or failed can be known, and $`|Q_{12}^{(j)}`$ are normalised. Taking inner product of (41), we get
$$\left[๐ฆ_i|๐ฆ_j\right]=f\left[๐ฆ_i|๐ฆ_j๐ฆ_i|๐ฆ_j\right]+[a_{ij}][a_{ji}^{}]$$
(42)
where we take $`[]`$ to be a matrix. For any $`n`$-vector $`๐ค=(k_1,\mathrm{},k_n)`$, we can write $`๐ค[๐ฆ_i|๐ฆ_j]๐ค^{}=K|K`$ where $`|Kk_1|๐ฆ_1+\mathrm{}+k_n|๐ฆ_n`$. Since $`|K`$ is a quantum state (linear combination of $`|๐ฆ_i`$s), its norm is always greater than or equal to zero. It is zero only when $`|K`$ itself is zero. If $`|๐ฆ_1,\mathrm{},|๐ฆ_n`$ are linearly independent, then $`|K`$ is never zero for any $`n`$-vector $`(k_1,\mathrm{},k_n)`$. Therefore when $`|๐ฆ_i`$ are linearly independent, $`[๐ฆ_i|๐ฆ_j]`$ is positive definite. Due to continuity, $`[๐ฆ_i|๐ฆ_j]f[๐ฆ_i|๐ฆ_j๐ฆ_i|๐ฆ_j]`$ is also positive definite with sufficiently small $`f`$. Therefore $`[๐ฆ_i|๐ฆ_jf[๐ฆ_i|๐ฆ_j๐ฆ_i|๐ฆ_j]`$ can be diagonalised and $`[a_{ij}][a_{ji}^{}]`$ can be chosen such that (42) is satisfied. Therefore there exists a unitary operator $`U`$ such that (41) is satisfied.
Consider the following general unitary transformation,
$`U\left(|๐ฆ_1\right)`$ $`=`$ $`\sqrt{f}|๐ฆ_1|๐ฆ_1|Y_0+\sqrt{1f}|Q|Y_1`$
$`U\left(|๐ฆ_2\right)`$ $`=`$ $`\sqrt{f}|๐ฆ_2|๐ฆ_2|Y_0+\sqrt{1f}|Q|Y_1`$ (43)
where $`|Y_0`$ and $`|Y_1`$ are orthonormal and $`|Q`$ are normalised. Then a cloning efficiency $`f`$ for probabilistic quantum anti-cloner can be obtained as follows,
$$f\frac{1|๐ฆ_1|๐ฆ_2|}{1|๐ฆ_1|๐ฆ_2||๐ฆ_1|๐ฆ_2|}=\frac{1|๐ฆ_1|๐ฆ_2|}{1|๐ฆ_1|๐ฆ_2|^2}$$
(44)
The equality in (44) holds if $`๐ฆ_1|๐ฆ_2`$ and $`๐ฆ_1|๐ฆ_2`$ are real and positive which can be achieved by redefining these states by multiplying them by a phase. Therefore, the probabilistic anti-cloner has the same efficiency as in the Duan and Guoโs regular cloner. One can see that the above probabilistic quantum anti-cloner can be generalised to clone $`\mu =(L,M)`$ copies for $`L`$ regular copies and $`M`$ copies of opposite spin direction. For $`n=2`$ case, as $`\mu \mathrm{}`$, the bound (44) approaches the probability of distinguishability given by $`1|๐ฆ_1|๐ฆ_2|`$ for given two states $`|๐ฆ_1`$ and $`|๐ฆ_2`$ . In , it was shown that the no-signalling condition restricts the number of states that can be cloned in a given Hilbert space. Following the same argument, one can show that if PQACM can clone $`N+1`$ or more states in a $`N`$-dimensional Hilbert space,then faster-than-light signalling can be achieved. Therefore no-signalling condition imposes a constraint such that probabilistic quantum anti-cloner cannot clone more than $`N`$ states.
Let us consider the following simple example. We take $`|๐ฆ_1=|0`$, $`|๐ฆ_2=\mathrm{cos}\theta |0+\mathrm{sin}\theta |1`$, $`|Y_0=|0`$, $`|Y_1=|1`$ , $`|Q_{12}=|00`$ then with maximum efficiency (44) of
$$f=\frac{1\mathrm{cos}\theta }{1\mathrm{cos}^2\theta }$$
(45)
we can find the unitary operator $`U=_{i=1}^8|N_iM_i|`$ where $`|M_1=|000,|M_2=|001`$, $`\mathrm{}`$ ,$`|M_8=|111`$ and $`|N_i`$โs are as follows
$$|N_1=\frac{1}{\sqrt{1+\mathrm{cos}\theta }}|010+\frac{\sqrt{\mathrm{cos}\theta }}{\sqrt{1+\mathrm{cos}\theta }}|001$$
(46)
$`|N_2`$ $`=`$ $`{\displaystyle \frac{\mathrm{cos}\theta }{\sqrt{1+\mathrm{cos}\theta }}}|000+{\displaystyle \frac{\mathrm{cos}\theta (\mathrm{cos}\theta 1)}{\mathrm{sin}\theta \sqrt{1+\mathrm{cos}\theta }}}|010{\displaystyle \frac{\mathrm{sin}\theta }{\sqrt{1+\mathrm{cos}\theta }}}|100`$ (47)
$`+{\displaystyle \frac{\mathrm{cos}\theta }{\sqrt{1+\mathrm{cos}\theta }}}|110+{\displaystyle \frac{\sqrt{\mathrm{cos}\theta }(1\mathrm{cos}\theta )}{\mathrm{sin}\theta \sqrt{1+\mathrm{cos}\theta }}}|001`$
and $`|N_3,\mathrm{},|N_8`$ are chosen as orthonormal states to (46) and (47). One can see $`UU^{}=U^{}U=\mathrm{๐}`$ and can easily check that $`U`$ yields $`|0|1`$ with the maximum efficiency given in (44).
With a similar argument, one can show the spin flipping, $`|๐ฆ|๐ฆ`$, can be done probabilistically, i.e.
$`U|๐ฆ_1|B_0`$ $`=`$ $`\sqrt{F}|๐ฆ_1|B_1+\sqrt{1\xi _1}|Q`$
$`U|๐ฆ_2|B_0`$ $`=`$ $`\sqrt{F}|๐ฆ_2|B_2+\sqrt{1\xi _2}|Q`$ (48)
can be shown to exist. When $`|B_1`$ and $`|B_2`$ are orthogonal, one can identify $`|๐ฆ_1`$ and $`|๐ฆ_2`$ and can prepare as many states as one wants and its efficiency bound is same as distinguishability between the two states.
## 4 Discussions
We have considered two types of quantum cloning for two anti-parallel outputs. In probabilistic cloning, for two input states, the anti-cloning efficiency is higher than the efficiency of distinguishing between the two states. On the other hand, in case of deterministic cloning, the fidelity of universal anti-cloner and the fidelity of measurement are above equal to 2/3. In other words, one could measure the input state and prepare two anti-parallel qubits (or as many as one wants) and this would have the same fidelity as in universal anti-cloner. The question of why universal anti-cloner has the same fidelity as the fidelity of measurement does not seem to have an immediate explanation.
In , it was shown that, unlike as in the classical case, quantum conditional entropy, which is the information about B which cannot be gained by measuring A, can have negative values. This negativity of entropy has been puzzling and its exact physical meaning has been questioned. In an anology with particle physics, it has been suggested that anti-qubits may be useful in describing quantum information processes where anti-qubits were introduced as qubits traveling backward in time . |
warning/0001/nlin0001072.html | ar5iv | text | # 1 Introduction
## 1 Introduction
In this paper we consider four classes of difference and differential-difference equations, which contain, in particular, the Toda lattice introduced by Toda , the relativistic Toda lattice introduced by Ruijsenaars , their generalizations introduced by Yamilov and Suris and their discretizations introduced by Hirota and Suris . The systematic account is given of the method proposed in the papers for studying of these classes of equations, and some classification results are presented. The main idea of the method can be described in a few words.
For the given lattice equation $`e[q]=0`$ we define, in a certain way, the pair of transformations $`T_+:qQ`$ and $`T_{}:q\stackrel{~}{Q}.`$ In general, variables $`Q`$ and $`\stackrel{~}{Q}`$ satisfy the distinct equations. We say that the lattice $`e[q]=0`$ admits the duality transformations $`T_\pm `$ if both variables $`Q,`$ $`\stackrel{~}{Q}`$ satisfy the same equation $`E[Q]=0`$ which is called the dual equation. Obviously, the composition of the duality transformations can be used for reproducing of solutions. Denoting iterations of the transformations $`T_l=T_{}^1T_+`$ and $`T_l^+=T_+T_{}^1`$ by superscript $`l`$ one obtains the commutative diagram displayed on the Figure 1.
In other words, transformations $`T_l`$ and $`T_l^+`$ define the Bรคcklund transformations for the given lattice and its dual. This explains the connection between duality transformations and integrability since existence of the Bรคcklund transformation is an indispensable feature of any integrable system. Quite analogously, the Bรคcklund transformations for the KdV and mKdV equations are obtained by the composition of two slightly different Miura maps and the Schlesinger transformation for PII is constructed from two substitutions into PXXXIV.
The requirement that the variables $`Q`$ and $`\stackrel{~}{Q}`$ satisfy the same equation is stringent enough and allows to distinguish effectively the integrable relativistic lattices. However, it does not work for the subclass of nonrelativistic lattices, which are characterized by the property $`T_+=T_{}.`$ In contrast to the relativistic case, the duality transformation is now irrelevant to integrability and cannot be used for classification. Nevertheless, this subclass does not require special treatment since it arises as a by-product of already obtained results. Namely, it turns out that the iterations of the transformation $`T_l`$ for the relativistic lattice admitting duality transformation are described by some lattice of the Toda type and the same is true for their discrete analogues.
The above scheme is applied to the discrete relativistic lattices in the Section 2. In the Section 2.1 we introduce the notions of the duality transformation and the dual equation, and prove that the dual equation belongs to the same class. In the Section 2.3 we study the Bรคcklund transformation $`T_l`$ in more details and prove that it is equivalent to some discrete lattice of the Toda type. The classification of integrable equations is performed in the Section 2.2, several multifield generalizations are presented in the Section 4. The Section 3 devoted to the lattices of the relativistic Toda type is, in fact, exact continuous double of the Section 2. Joint of both discrete and continuous theories is performed in the Section 3.4.
## 2 The lattices of the discrete relativistic Toda type
This section deals with the difference equations on the real or complex variable $`q`$ defined on the two-dimensional integer lattice. As a rule, we use abridged notation $`q=q_{mn},`$ $`q_{ij}=q_{m+i,n+j}.`$ The shift operators on the first and second subscripts will be denoted $`T_m`$ and $`T_n`$ respectively. Next, it is convenient to introduce notation for the differences (see Figure 2)
$$x=qq_{1,0},y=qq_{0,1},z=qq_{1,1}.$$
Obviously, one of these differences can be expressed through the other two, e.g. $`z=x+y_{1,0}=y+x_{0,1},`$ and the following identity is valid
$$(T_m1)y_{0,1}=(T_n1)x_{1,0}.$$
(2.1)
The uppercase letters $`Q,X,Y,Z`$ (reserved for the dual variable and its differences) are used in quite similar manner.
### 2.1 Duality transformations
The lattices of the discrete relativistic Toda type
$$(T_m1)f(x)+(T_n1)g(y)+(T_mT_n1)h(z)=0$$
(2.2)
are the Euler equations for the Lagrangians of the form
$$=\underset{m,n}{}(a(x)+b(y)+c(z))$$
(2.3)
where $`f=a^{},`$ $`g=b^{},`$ $`h=c^{}.`$ In this section we assume that $`f^{}g^{}h^{}0`$ in order to eliminate the case of the nonrelativistic lattices, so that equation (2.2) involves 7 nodes of the lattice as shown on the Figure 2. It is clear that the roles of the shifts $`T_m,T_n`$ and $`T_m^1T_n^1`$ in (2.2) are equal (actually, one can assume them as the generators of the regular hexagonal lattice).
Equation (2.2) can be rewritten in two equivalent forms of the momentum conservation law:
$$(T_m1)(f(x)+h(z_{0,1}))+(T_n1)(g(y)+h(z))=0$$
$$(T_m1)(f(x)+h(z))+(T_n1)(g(y)+h(z_{1,0}))=0.$$
This allows to introduce the pair of transformations acting on the differences:
$$T_+:(X,Y_{1,0})=T(x,y_{0,1}),T_{}:(\stackrel{~}{X}_{0,1},\stackrel{~}{Y})=T(x_{1,0},y)$$
(2.4)
where the mapping $`T:(x,y)(X,Y)`$ is given by the formulae
$$X=g(y)+h(x+y),Y=f(x)h(x+y).$$
(2.5)
Obviously, the variables $`Q`$ and $`\stackrel{~}{Q}`$ corresponding to $`X,Y`$ and $`\stackrel{~}{X},\stackrel{~}{Y}`$ are defined only up to the addition of an arbitrary constant. However, the equations for these variables contain only differences and can be derived by solving (2.4), (2.5) with respect to $`x,y`$ and using the identity (2.1). Generally, these equations are distinct from each other.
###### Definition 1.
The lattice (2.2) admits the duality transformations (2.4), (2.5) if the mapping $`T`$ is invertible and both variables $`Q,`$ $`\stackrel{~}{Q}`$ satisfy the same lattice which is called dual to (2.2).
The following Theorem characterizes equations admitting duality transformations in terms of the mapping $`T.`$ It also demonstrates that the duality transformations do not lead out off the class (2.2). The immediate corollary is that the equation which is dual to the dual equation coincides with the original one.
###### Theorem 1.
Equation (2.2) admits duality transformations if and only if the inverse of (2.5) is of the form
$$x=G(Y)+H(X+Y),y=F(X)H(X+Y).$$
(2.6)
In this case the dual equation is of the form
$$(T_m1)F(X)+(T_n1)G(Y)+(T_mT_n1)H(Z)=0.$$
(2.7)
* Let $`T^1`$ be of the stated form, then one can easily check that the elimination of $`x,y`$ by means of the identity (2.1) brings to equation (2.7) for both transformations (2.4).
Conversely, assume that equations for $`Q`$ and $`\stackrel{~}{Q}`$ coincide. Let $`T^1`$ be of the form
$$x=\mathrm{\Phi }(X,Y),y=\mathrm{\Psi }(X,Y)$$
then inverses of the transformations (2.4) are given by the formulae (the tilde in the second one is omitted)
$`T_+^1:`$ $`x=\mathrm{\Phi }(X,Y_{1,0}),y_{0,1}=\mathrm{\Psi }(X,Y_{1,0}),`$
$`T_{}^1:`$ $`x_{1,0}=\mathrm{\Phi }(X_{0,1},Y),y=\mathrm{\Psi }(X_{0,1},Y).`$
The identity (2.1) yields the equations
$$\mathrm{\Phi }(X_{1,1},Y_{0,1})\mathrm{\Phi }(X_{1,0},Y)\mathrm{\Psi }(X_{1,0},Y)+\mathrm{\Psi }(X,Y_{1,0})=0,$$
$$\mathrm{\Phi }(X,Y_{0,1})\mathrm{\Phi }(X_{0,1},Y)\mathrm{\Psi }(X_{1,0},Y_{1,1})+\mathrm{\Psi }(X,Y_{0,1})=0$$
which must be equivalent to each other. In order to compare these equations, rewrite the last one in the form
$$\mathrm{\Phi }(X,Y_{0,1})\mathrm{\Phi }(X+Y_{1,0}Y,Y)\mathrm{\Psi }(X_{1,0},X_{1,1}+Y_{0,1}X_{1,0})+\mathrm{\Psi }(X,Y_{0,1})=0,$$
so that both equations contain the variables $`X_{1,1},`$ $`X_{1,0},`$ $`X,`$ $`Y_{0,1},`$ $`Y,`$ $`Y_{1,0}.`$ Now let us consider $`X_{1,1}`$ as function on the rest variables from this set, then
$`{\displaystyle \frac{X_{1,1}}{Y_{0,1}}}`$ $`={\displaystyle \frac{_{Y_{0,1}}\mathrm{\Phi }(X_{1,1},Y_{0,1})}{_{X_{1,1}}\mathrm{\Phi }(X_{1,1},Y_{0,1})}}`$
$`={\displaystyle \frac{_{Y_{0,1}}[\mathrm{\Phi }(X,Y_{0,1})\mathrm{\Psi }(X_{1,0},X_{1,1}+Y_{0,1}X_{1,0})+\mathrm{\Psi }(X,Y_{0,1})]}{_{X_{1,1}}\mathrm{\Psi }(X_{1,0},X_{1,1}+Y_{0,1}X_{1,0})}}.`$
The last equality must be satisfied identically. Differentiating it with respect to $`X`$ and $`X_{1,0}`$ yields
$$_X_{Y_{0,1}}(\mathrm{\Phi }(X,Y_{0,1})+\mathrm{\Psi }(X,Y_{0,1}))=0,_{X_{1,0}}_{Y_{0,1}}\mathrm{\Psi }(X_{1,0},X_{1,1}+Y_{0,1}X_{1,0})=0$$
and hence
$$\mathrm{\Phi }(X,Y)+\mathrm{\Psi }(X,Y)=G(Y)F(X),\mathrm{\Psi }(X,Y)=K(X)H(X+Y).$$
(2.8)
Analogously, differentiating the relation
$`{\displaystyle \frac{Y_{1,0}}{X}}`$ $`={\displaystyle \frac{_X\mathrm{\Psi }(X,Y_{1,0})}{_{Y_{1,0}}\mathrm{\Psi }(X,Y_{1,0})}}`$
$`={\displaystyle \frac{_X[\mathrm{\Phi }(X,Y_{0,1})\mathrm{\Phi }(X+Y_{1,0}Y,Y)+\mathrm{\Psi }(X,Y_{0,1})]}{_{Y_{1,0}}\mathrm{\Phi }(X+Y_{1,0}Y,Y)}}`$
with respect to $`Y`$ yields
$$_X_Y\mathrm{\Phi }(X+Y_{1,0}Y,Y)=0\mathrm{\Phi }(X,Y)=L(Y)+M(X+Y).$$
Comparing with (2.8) completes the proof. โ
### 2.2 Classification theorem
The Definition 1 turns out to be severe enough and allows to obtain the finite list of integrable equations (2.2). In virtue of the Theorem 1 it is sufficient to find all functions $`f,g,h`$ such that inverse of the transformation (2.5) is given by (2.6). This means that the Jacobian $`\mathrm{\Delta }=f^{}g^{}+g^{}h^{}+h^{}f^{}`$ of the map (2.5) must be nonzero and the following identities must hold
$$x_{XY}+y_{XY}=0,x_{XY}=x_{XX},y_{XY}=y_{YY}.$$
These three relations are equivalent. Indeed, the Jacobi matrix is
$$\left(\begin{array}{cc}x_X& y_X\\ x_Y& y_Y\end{array}\right)=\frac{1}{\mathrm{\Delta }}\left(\begin{array}{cc}h^{}& f^{}+h^{}\\ g^{}h^{}& h^{}\end{array}\right)$$
that is $`x_X=y_Y.`$ Straightforward computation proves that functions $`f(x),`$ $`g(y),`$ $`h(x+y)`$ must satisfy the equation
$$(g^{}+h^{})\frac{f^{\prime \prime }}{f^{}}+(f^{}+h^{})\frac{g^{\prime \prime }}{g^{}}=(f^{}+g^{})\frac{h^{\prime \prime }}{h^{}}.$$
The designations $`f^{}=1/u,`$ $`g^{}=1/v,`$ $`h^{}=1/w`$ rewrite it in more convenient form
$$[v(y)+w(x+y)]u^{}(x)+[u(x)+w(x+y)]v^{}(y)=[u(x)+v(y)]w^{}(x+y).$$
(2.9)
The classification problem is reduced to solving of this functional equation. Of course, functions $`u,v,w`$ can be multiplied by an arbitrary constant and the linear transformation
$$\stackrel{~}{x}=c(xx_0),\stackrel{~}{y}=c(yy_0)\stackrel{~}{q}_{m,n}=c(q_{m,n}mx_0ny_0const)$$
can be applied, as well as the permutation of the $`x,y,z`$ axes.
At first let us consider some degenerate cases. Assume that two of three functions are constant, say $`u`$ and $`v.`$ Then (2.9) yields that either $`w`$ is constant as well, or $`u=v`$ and $`w`$ is arbitrary. In the first case the equation (2.2) is linear and in the second one it is of the form
$$\alpha (q_{1,0}+q_{1,0}q_{0,1}q_{0,1})+h(q_{1,1}q)h(qq_{1,1})=0$$
and admits โintegrationโ:
$$\alpha (q_{m+1,n}q_{m,n+1})+h(q_{m+1,n+1}q_{m,n})=c_{mn}.$$
Other degenerate case corresponds to the vanishing of the Jacobian, what is equivalent to $`u+v+w=0`$ and implies that all three functions are linear. In this case one can prove that equation (2.2) can be reduced to the equation on the variable $`p=y_{0,1}/x.`$ Further on we will not consider these degenerate cases.
###### Lemma 2.
The functions $`u,v,w`$ satisfy the equations
$$(u^{})^2=\delta u^2+2\alpha u+\epsilon ,(v^{})^2=\delta v^2+2\beta v+\epsilon ,(w^{})^2=\delta w^2+2\gamma w+\epsilon .$$
(2.10)
* At first prove that functions $`u(x)`$ and $`v(y)`$ satisfy the equation
$$(u^{\prime \prime }v^{\prime \prime })(u+v)(u^{})^2+(v^{})^2=k(uv),k=const.$$
(2.11)
Let us eliminate $`w`$ from (2.9). Applying the operator $`_x_y`$ one obtains the linear system on $`w,w^{}:`$
$$\left(\begin{array}{cc}u^{}+v^{}& uv\\ u^{\prime \prime }v^{\prime \prime }& v^{}u^{}\end{array}\right)\left(\begin{array}{c}w+u\\ w^{}\end{array}\right)=(uv)\left(\begin{array}{c}u^{}\\ u^{\prime \prime }\end{array}\right).$$
Its determinant $`\mathrm{\Delta }`$ is exactly the left hand side of the equation (2.11). If it is identically zero then (2.11) is proved, otherwise one finds
$$w+u=\frac{uv}{\mathrm{\Delta }}(uu^{\prime \prime }(u^{})^2+u^{\prime \prime }v+u^{}v^{}),w^{}=\frac{uv}{\mathrm{\Delta }}(u^{\prime \prime }v^{}+u^{}v^{\prime \prime })$$
and consequently $`((uv)/\mathrm{\Delta })_y(uu^{\prime \prime }(u^{})^2+u^{\prime \prime }v+u^{}v^{})=0.`$ Assume that the expression in the second bracket vanishes. If $`u^{}0`$ then $`v^{}=u^{\prime \prime }v/u^{}+u^{}uu^{\prime \prime }/u^{},`$ $`v^{\prime \prime }=u^{\prime \prime }v^{}/u^{},`$ but then, as one easily checks, $`\mathrm{\Delta }=0.`$ If $`u^{}=0`$ then $`w+u=0,`$ that is we come to the degenerate solution excluded above. Therefore $`((uv)/\mathrm{\Delta })_y=0.`$ Due to the symmetry between $`u`$ and $`v`$ one can prove analogously $`((uv)/\mathrm{\Delta })_x=0`$ and obtain (2.11).
Further on, rewriting (2.11) in the form
$$\left(\frac{u^{}}{u+v}\right)_x=\frac{v^{\prime \prime }+k}{u+v}\frac{(v^{})^2+2kv}{(u+v)^2},$$
multiplying by $`u^{}/(u+v)`$ and integrating with respect to $`x`$ yield
$$(u^{})^2=\delta (y)(u+v)^22(v^{\prime \prime }+k)(u+v)+(v^{})^2+2kv.$$
Replacing $`u`$ and $`v`$ one obtains
$$(v^{})^2=\stackrel{~}{\delta }(x)(u+v)^22(u^{\prime \prime }+k)(u+v)+(u^{})^2+2ku.$$
Subtracting one equation from another and using (2.11) one obtains $`\stackrel{~}{\delta }=\delta =const.`$ Summing and dividing by $`u+v`$ give $`u^{\prime \prime }+v^{\prime \prime }=\delta (u+v)k.`$ The separation of the variables yields $`(u^{})^2=\delta u^2+2\alpha u+\epsilon ,`$ $`(v^{})^2=\delta v^2+2\beta v+\stackrel{~}{\epsilon },`$ where $`\alpha +\beta =k,`$ and substitution into (2.11) proves $`\epsilon =\stackrel{~}{\epsilon }.`$ The last of the equations (2.10) is obtained in virtue of the symmetry of $`x,y,z`$ axes. โ
It is clear that the solutions of equations (2.10) must satisfy also some additional relations. However their analysis is not in principle difficult, and the direct examination of all solutions brings to the following list.
###### Theorem 3.
The equations (2.2) admitting duality transformations are exhausted, up to the changes $`\stackrel{~}{q}_{m,n}=c(q_{m,n}mx_0ny_0)`$ and permutations of $`x,y,z`$ axes, by the following sets of the functions $`f,g,h.`$ In formulae (A), (B), (C) the parameters are constrained by relation $`\lambda +\mu +\nu =0,`$ and in (I) by relation $`\lambda \mu \nu =1.`$
$$\begin{array}{ccccc}(A)\hfill & & f=\frac{\mu }{x},\hfill & g=\frac{\nu }{y},\hfill & h=\frac{\lambda }{z},\hfill \\ (B)\hfill & & f=\mu \mathrm{coth}x,\hfill & g=\nu \mathrm{coth}y,\hfill & h=\lambda \mathrm{coth}z,\hfill \\ (C)\hfill & & f=\frac{1}{2}\mathrm{log}\frac{x+\mu }{x\mu },\hfill & g=\frac{1}{2}\mathrm{log}\frac{y+\nu }{y\nu },\hfill & h=\frac{1}{2}\mathrm{log}\frac{z+\lambda }{z\lambda },\hfill \\ (D)\hfill & & f=\mathrm{log}x,\hfill & g=\mathrm{log}y,\hfill & h=\mathrm{log}(11/z),\hfill \\ (E)\hfill & & f=e^x1,\hfill & g=e^y,\hfill & h=\frac{1}{1+e^z},\hfill \\ (F)\hfill & & f=\mathrm{log}(e^x1),\hfill & g=\mathrm{log}(e^y1),\hfill & h=\mathrm{log}(e^z1),\hfill \\ (G)\hfill & & f=\mathrm{log}(e^x1),\hfill & g=\mathrm{log}(e^y1),\hfill & h=z,\hfill \\ (H)\hfill & & f=\mathrm{log}(\lambda ^1(e^x+1)),\hfill & g=\mathrm{log}(e^y1),\hfill & h=\mathrm{log}\frac{e^z+\lambda }{e^z+1},\hfill \\ (I)\hfill & & f=\mathrm{log}\frac{\mu e^x+1}{e^x+\mu },\hfill & g=\mathrm{log}\frac{\nu e^y+1}{e^y+\nu },\hfill & h=\mathrm{log}\frac{\lambda e^z+1}{e^z+\lambda }.\hfill \end{array}$$
The duality transformations (2.4), (2.5) link together the equations corresponding to solutions (B) and (C), (D) and (E), (F) and (G), while the equations corresponding to solutions (A), (H) and (I) are self-dual. โ
Notice, that the cases (A) and (B) are connected by the point transformation $`q=\mathrm{exp}(2\stackrel{~}{q}).`$ It is explained by the fact that the Lagrangian $`(\mu \mathrm{log}x+\nu \mathrm{log}y+\lambda \mathrm{log}z)`$ of the equation (2.2), (A) is invariant under the dilations $`qCq`$ as well as under the shifts $`qq+C.`$ On the other hand, the inversions $`qq/(1Cq)`$ preserve the Lagrangian as well but the change $`q=1/\stackrel{~}{q}`$ which maps this group into the shift group does not bring to a new equation.
### 2.3 The lattices of the discrete Toda type
The method proposed in the Section 2.1 does not work for the lattices of the discrete Toda type which correspond to the case $`h=0`$ (this is equivalent to $`f^{}g^{}h^{}=0,`$ without loss of generality). Indeed, in this case transformations $`T_+`$ and $`T_{}`$ coincide for arbitrary $`f,g`$ and classification becomes impossible. However, we can dispense with it since these lattices arise as a by-product of already obtained results for the discrete relativistic lattices. Namely, we will demonstrate, using only few basic formulae from the Section 2.1 and without any complicated calculations, that the iterations of the Bรคcklund transformation $`T_l=T_{}^1T_+`$ are described by some discrete lattice of the Toda type. Hence we automatically obtain some list of integrable lattices, see Theorem 4 below. Probably, this list is exhaustive (cf. ), but unfortunately I do not know any classification results, like Yamilovโs Theorem 8, which can be compared with this list.
Let us consider some lattice (2.2) admitting duality transformations and denote iterations of $`T_l`$ by superscript $`l,`$ in such a way that tilde in the formula (2.4) corresponds to the value $`l1.`$ It is possible to rewrite equations (2.2), (2.7) in terms of the mixed variables $`x,X.`$ Indeed, one obtains directly from (2.4), (2.5) the relations
$$X_{0,1}=g(y)+h(x_{0,1}+y),X^1=g(y_{0,1})+h(x_{1,1}+y_{0,1})$$
and therefore equation (2.2) is equivalent to
$$(T_m1)f(x)+X^1X_{0,1}=0.$$
Analogously, in virtue of (2.6) equation (2.7) is equivalent to
$$(T_m1)F(X)+x_{1,1}x_{1,0}^1=0.$$
โIntegratingโ this equation with respect to $`m`$ (recall that $`x_{1,0}=(T_m1)q`$) one obtains $`X=\phi ((T_lT_n)q+c)`$ where function $`\phi `$ is inverse of $`F.`$ The constant $`c`$ does not depend on $`m,`$ but may depend on $`l,n,`$ and it can be set to zero without lost of generality by means of appropriate shift of the variables $`q.`$ Then eliminating $`X`$ from the previous equation brings to the lattice of the discrete Toda type
$$(T_m1)f(x)(T_lT_n^11)\phi ((1T_l^1T_n)q)=0.$$
(2.12)
So, we have already proved that the Bรคcklund transformation for the discrete relativistic lattice is governed by some discrete nonrelativistic lattice. However, the complete picture is even more rich: it turns out that the 3-dimensional lattice generated by the shifts $`T_l,`$ $`T_m,`$ $`T_n`$ contains 3 instances of the discrete nonrelativistic lattices and 4 instances of the discrete relativistic lattices. In order to see this let us remind that the roles of all shifts in equation (2.2) are equal and therefore we can repeat the above calculation starting from the other set of mixed variables. More precisely, rewriting equations (2.2) and (2.7) in terms of $`y,Y`$ or $`z,Z`$ one obtains the equations
$$(T_n1)g(y)+Y_{1,0}^1Y=0,(T_n1)G(Y)+y_{0,1}y_{1,1}^1=0,$$
$$(T_mT_n1)h(z)+ZZ^1=0,(T_mT_n1)H(Z)+z_{1,1}^1z_{1,1}=0$$
and further elimination of $`Y`$ and $`Z`$ yields equations
$$(T_n1)g(y)(T_lT_m1)\psi ((1T_l^1T_m^1)q)=0,$$
(2.13)
$$(T_mT_n1)h(z)+(T_l1)\eta ((T_l^11)q)=0$$
(2.14)
where functions $`\psi `$ and $`\eta `$ are inverses of $`G`$ and $`H`$ respectively.
The placement of the variables involved in equations (2.12), (2.13) and (2.14) is shown at the Figure 3. At this picture the variables $`q`$ involved in the original equation (2.2) lie in the plane $`(mn)`$ and are marked by black symbols, while the variables obtained by means of the Bรคcklund transformations are marked by the white ones. Equation (2.12) constraints two black and two white circles (and, of course, the centre of the cube), squares and triangles correspond to (2.13) and (2.14).
Next, subtracting from the equation (2.2) two of three equations (2.12), (2.13) and (2.14) we obtain some discrete relativistic lattice again, for example subtracting of (2.13), (2.14) yields equation
$$(T_m1)f(x)(T_l1)\eta ((T_l^11)q)+(T_lT_m1)\psi ((1T_l^1T_m^1)q)=0$$
which involves variables in the plane $`(lm).`$ Two other choices bring to the discrete relativistic lattices in the plane $`(ln)`$ and in the plane spanned over the vectors $`(0,1,1)`$ and $`(1,1,0).`$
In conclusion of this section we present the list of the discrete lattices of the Toda type which is obtained by direct examination of the equations listed in the Theorem 3.
###### Theorem 4.
The equations (2.12), (2.13), (2.14) are equivalent, up to the renaming of the shifts and the linear changes $`\stackrel{~}{q}_{m,n}=c(q_{m,n}mx_0ny_0),`$ to the lattices
$$(T_m1)\frac{1}{x}=(T_n1)\frac{1}{y},$$
$$(T_m1)e^x=(T_n1)e^y,$$
$$(T_m1)\frac{1}{e^x1}=(T_n1)\frac{1}{e^y1},$$
$$(T_m1)\mathrm{log}x=(T_n1)\mathrm{log}y,$$
$$(T_m1)\mathrm{log}(11/x)=(T_n1)\mathrm{log}(11/y),$$
$$(T_m1)\mathrm{log}(e^x1)=(T_n1)\mathrm{log}(e^y1),$$
$$(T_m1)x=(T_n1)\mathrm{log}(e^y1),$$
$$(T_m1)\mathrm{log}\left(\frac{e^x\lambda }{e^x1}\right)=(T_n1)\mathrm{log}\left(\frac{e^y\lambda }{e^y1}\right).$$
## 3 The lattices of the relativistic Toda type
Now let us consider the differential-difference equations on the variable $`q_n(t).`$ As before we will omit the subscript $`n:`$ $`q=q_n,`$ $`q_i=q_{n+i}`$ and use uppercase letters for the dual variables. Instead of differences $`x,y,z`$ we consider the quantities
$$p=\dot{q},y=qq_1$$
which satisfy identity
$$D_ty_1=(T_n1)p.$$
(3.1)
### 3.1 Duality transformations
The lattices of the relativistic Toda type
$$\dot{p}=r(p)(h(y_1)p_1h(y)p_1+g(y_1)g(y)),\dot{y}=pp_1$$
(3.2)
are the Euler equations for the Lagrangians of the form
$$=๐t\underset{n}{}(a(p)b(y)c(y)p),$$
(3.3)
where $`r=1/a^{\prime \prime },`$ $`g=b^{},`$ $`h=c^{}.`$ In this section we assume that the nondegeneracy condition $`h0,`$ $`|g^{}|+|h^{}|0`$ is fulfilled.
The term $`c(y)p`$ of the Lagrangian is equivalent, up to the total divergence, to $`c(y_1)p`$ and this results in two equivalent forms of the momentum conservation law:
$$D_t(a^{}(p)c(y_1))=(T_n1)(h(y)p_1+g(y))$$
$$D_t(a^{}(p)c(y))=(T_n1)(h(y)p+g(y)).$$
This allows to define the pair of transformations
$$T_+:(P,Y)=T(p,y_1),T_{}:(\stackrel{~}{P}_1,\stackrel{~}{Y})=T(p,y)$$
(3.4)
where mapping $`T:(p,y)(P,Y)`$ is defined by the formula
$$P=h(y)p+g(y),Y=a^{}(p)c(y).$$
(3.5)
Notice, that the lattices (3.2) are the continuous limit of the discrete relativistic lattices (2.2). Indeed, let us consider the family of the functionals of the form (2.3)
$$_\epsilon =\underset{m,n}{}(\epsilon a(x_{m,n}/\epsilon )\epsilon b(y_{m,n})+k(y_{m,n})k(z_{m,n})),$$
and let $`k^{}=c,`$ $`q_{m,n}=q_n(t),`$ $`t=m\epsilon .`$ It is easy to see that passage to the limit $`\epsilon 0`$ brings exactly to the Lagrangian (3.3). Transformations (3.4) also are obtained from (2.4) by this limit.
###### Definition 2.
The lattice (3.2) admits the duality transformations (3.4), (3.5) if the mapping $`T`$ is invertible and both variables $`Q,`$ $`\stackrel{~}{Q}`$ satisfy the same lattice which is called dual to (3.2).
Notice that functions $`a,b,c`$ are defined up to some linear transformations and therefore mapping (3.5) is not uniquely defined. However, it is easy to see that this arbitrariness corresponds to the linear changes of the variables $`P,Y`$ and all lattices dual to the given equation (3.2) are equivalent under the changes $`Q_n\alpha Q_n+\beta n+\gamma t.`$
###### Theorem 5.
The lattice (3.2) admits duality transformations if and only if the inverse of the mapping (3.5) is of the form
$$p=H(Y)P+G(Y),y=A^{}(P)C(Y),H=C^{}.$$
(3.6)
In this case the dual equation is of the form ($`R=1/A^{\prime \prime }`$)
$$\dot{P}=R(P)(H(Y_1)P_1H(Y)P_1+G(Y_1)G(Y)),\dot{Y}=PP_1.$$
(3.7)
* Let $`T^1`$ be of the form (3.6). Then elimination of $`p,y`$ from (3.4) by means of identity (3.1) brings to equation (3.7) in both cases.
In order to prove inverse statement we need to compare equations on the variables $`Q`$ and $`\stackrel{~}{Q}.`$ Straightforward calculation proves that the variables $`P,Y`$ satisfy the equations of the form
$$\dot{P}=r(p)(h(y_1)(PP_1)+\mathrm{\Delta }(p,y_1)(p_1p)),\dot{Y}=PP_1,$$
(3.8)
where $`\mathrm{\Delta }(p,y)=h^2(y)+\frac{1}{r(p)}(h^{}(y)p+g^{}(y))`$ and $`p_1,p,y_1`$ have to be expressed in terms of $`P_1,P,Y_1,Y.`$ Therefore, the dual equation must be linear in $`P_1.`$ Analogously, considering equation on $`\stackrel{~}{Q}`$ one proves that the dual equation must be linear in $`P_1.`$ Since in (3.8) only $`p_1`$ depends on $`P_1,`$ hence mapping $`T`$ must satisfy the condition $`^2p/P^2=0.`$ This proves the first formula in (3.6). The second one follows from the property $`y/Y=p/P`$ which is evident from the structure of the Jacobi matrix
$$\left(\begin{array}{cc}p/P& y/P\\ p/Y& y/Y\end{array}\right)=\frac{1}{\mathrm{\Delta }(p,y)}\left(\begin{array}{cc}h(y)& 1/r(p)\\ h^{}(y)p+g^{}(y)& h(y)\end{array}\right)$$
(3.9)
where $`\mathrm{\Delta }0`$ by Definition 2. โ
As in the discrete case we see that the original lattice is dual for its dual and the composition $`T_{}^1T_+`$ defines the Bรคcklund transformation for (3.2) (see Figure 2). We will study this Bรคcklund transformation in the Section 3.3.
### 3.2 Classification theorem
###### Theorem 6.
The lattices (3.2) admitting the duality transformations are characterized by the following equations for the coefficients:
$$\begin{array}{cc}& r=r_2p^2+r_1p+r_0,\hfill \\ & g^{}=r_2g^2+R_1g+R_0r_0h^2,h^{}=2r_2ghr_1h^2+R_1h.\hfill \end{array}$$
(3.10)
The coefficients of the dual lattice satisfy equations
$$\begin{array}{cc}& R=r_2P^2+R_1P+R_0,\hfill \\ & G^{}=r_2G^2+r_1G+r_0R_0H^2,H^{}=2r_2GHR_1H^2+r_1H.\hfill \end{array}$$
(3.11)
* Straightforward calculation proves that necessary and sufficient condition $`^2p/P^2=0`$ is equivalent to the following relation involving the functions $`r(p),g(y),h(y):`$
$$h(h^{\prime \prime }p+g^{\prime \prime })h^{}(h^{}p+g^{})+2rh^2h^{}r^{}h^2(h^{}p+g^{})=0.$$
(3.12)
The second derivative of (3.12) with respect to $`p`$ is $`r^{\prime \prime \prime }(h^{}p+g^{})=0.`$ By the condition of nondegeneracy, this implies that $`r`$ is a polynomial and its degree is less than 3. Dividing (3.12) by $`h^2`$ and integrating the result with respect to $`y`$ one obtains
$$h^{}p+g^{}+2rh^2r^{}h(hp+g)=h(\alpha +R_1p)$$
where $`\alpha `$ and $`R_1`$ are some constants. Let $`r=r_2p^2+r_1p+r_0,`$ then collecting the coefficients of $`p`$ in this relation gives the system
$$g^{}=\alpha h+r_1gh2r_0h^2,h^{}=2r_2ghr_1h^2+R_1h$$
which is equivalent to (3.10) modulo common first integral
$$r_2g^2+R_1g+R_0r_1gh+r_0h^2=\alpha h.$$
(3.13)
The formula for $`R`$ can be easily obtained from the relation $`R=\mathrm{\Delta }(p,y)r.`$ Thus, equations for $`g`$ and $`h`$ are uniquely defined by the coefficients of the polynomials $`r`$ and $`R.`$ Since due to the Theorem 5 the duality relation is symmetric, hence equations for $`G`$ and $`H`$ can be written automatically. โ
#### Remark.
The question about value of the first integral for the system (3.11) is a bit more complicated. However, this value can be calculated by sequential elimination of $`g,h`$ and $`p`$ from the formula (3.13) using relations $`g=Php,`$ $`h=RH/r`$ and $`p=HP+G.`$ This brings to relation
$$r_2G^2+r_1G+r_0R_1GH+R_0H^2=\alpha H,$$
that is, the value of the integration constant $`\alpha `$ for the dual lattice is the same. Notice, that actually role of this constant is very important since it may depend on $`n.`$ This brings to the integrable lattices containing an arbitrary parameter in each node (it plays role of the discrete spectrum when constructing solitons). For sake of simplicity we will not consider this generalization.
By use of the first integral (3.13) the system (3.10) is reduced to the equation
$$(h^{})^2=(r_1^24r_2r_0)h^4+(4\alpha r_22R_1r_1)h^3+(R_1^24r_2R_0)h^2$$
(3.14)
which is solved in elementary functions. Consideration of all the possible choices of parameters and the branches of solutions brings to the following result.
###### Theorem 7.
Up to the transformations $`q_n\alpha q_n+\beta t+\gamma n,`$ $`t\delta t,`$ the nondegenerate lattices (3.2) admitting the duality transformations are exhausted by the list $`(\dot{y}=pp_1):`$
$`(a)`$ $`\dot{p}=p_1e^{y_1}p_1e^ye^{2y_1}+e^{2y},`$
$`(b)`$ $`\dot{p}=p\left({\displaystyle \frac{p_1}{y_1}}{\displaystyle \frac{p_1}{y}}+y_1y\right),`$
$`(c_{\mu ,\nu })`$ $`\dot{p}=p\left({\displaystyle \frac{p_1}{1+\mu e^{y_1}}}{\displaystyle \frac{p_1}{1+\mu e^y}}+\nu (e^{y_1}e^y)\right),`$
$`(d)`$ $`\dot{p}=p(p+1)\left({\displaystyle \frac{p_1}{y_1}}{\displaystyle \frac{p_1}{y}}\right),`$
$`(e_\mu )`$ $`\dot{p}=p(p\mu )\left({\displaystyle \frac{p_1}{\mu +e^{y_1}}}{\displaystyle \frac{p_1}{\mu +e^y}}\right),`$
$`(f_\mu )`$ $`\dot{p}=(p^2+\mu )\left({\displaystyle \frac{p_1y_1}{\mu +y_1^2}}{\displaystyle \frac{p_1y}{\mu +y^2}}\right),`$
$`(g_\mu )`$ $`\dot{p}={\displaystyle \frac{1}{2}}(p^2+1\mu ^2)\left({\displaystyle \frac{p_1\mathrm{sinh}y_1}{\mu +\mathrm{cosh}y_1}}{\displaystyle \frac{p_1\mathrm{sinh}y}{\mu +\mathrm{cosh}y}}\right).`$
The duality transformations (3.4), (3.5) link together equations (a) and (b), (d) and (e<sub>0</sub>), (f<sub>ฮผ</sub>) for $`\mu 0`$ and (g<sub>ยฑ1</sub>), while the rest equations are self-dual. โ
### 3.3 The lattices of the Toda type
As in the discrete case, the subclass of the Toda type lattices
$$\dot{p}=r(p)(f(y_1)f(y)),\dot{y}=pp_1$$
(3.15)
must be considered separately, since transformations $`T_+`$ and $`T_{}`$ coincide. It should be mentioned that in this case the duality transformation was introduced by Toda . It is given by the formula
$$P=f(y_1),Y=a^{}(p),a^{\prime \prime }=1/r$$
and the coefficients of the dual lattice are defined by the formulae $`f^{}=R(f),`$ $`F(a^{}(p))=p.`$ For example, the Toda lattice $`\ddot{q}=\mathrm{exp}(q_1q)\mathrm{exp}(qq_1)`$ is dual to the lattice $`\ddot{Q}=\dot{Q}(Q_12QQ_1).`$
In contrast to the relativistic case, the duality transformation is irrelevant to integrability and cannot be used for the classification of the lattices (3.15). For the first time this problem was solved by Yamilov in the framework of the symmetry approach.
###### Theorem 8 (Yamilov, ).
The Toda type lattice (3.15) admits the higher symmetries iff
$$r(p)=r_2p^2+r_1p+r_0,f^{}=r_2f^2+R_1f+R_0.$$
(3.16)
This result demonstrates that integrable lattices (3.15) can be obtained from the integrable lattices of the relativistic Toda type by passing to the limit $`h=0`$ in equations (3.10). More interesting link between this two classes of equations can be established along the same arguments as in the Section 2.3.
Let us denote iterations of transformation $`T_l=T_{}^1T_+`$ by the superscript $`l`$ and let tilde in the formula (3.4) corresponds to the value $`l1.`$ Then formulae (3.4), (3.5), (3.6) take form
$`P_1=h(y)p_1+g(y),P^1=h(y_1)p_1+g(y_1)`$ (3.17)
$`p^1=H(Y)P_1+G(Y),p=H(Y)P+G(Y)`$ (3.18)
and therefore equations (3.2), (3.7) can be rewritten as a coupled lattice of the Volterra type
$$\dot{p}=r(p)(P^1P_1),\dot{P}=R(P)(p_1p^1).$$
Assuming
$$P=f(q_1q^1),f^{}=R(f)$$
(3.19)
we obtain the Toda type lattice
$$\ddot{q}=r(\dot{q})(f(q_1^1q)f(qq_1^1))$$
(3.20)
for the variables situated along the line $`l+n=const.`$ Since functions $`r,R`$ were described in the Section 3, we immediately repeat the Yamilovโs result (3.16).
### 3.4 Nonlinear superposition principle
Recall, that in the discrete case the 3-dimensional lattice generated by the shifts $`T_l,`$ $`T_m,`$ $`T_n`$ contains 4 instances of the discrete relativistic lattices and 3 instances of nonrelativistic ones (see Figure 3). In the continuous case the picture is more bare: the variables $`q_{ln}`$ form the 2-dimensional lattice containing 2 instances of the relativistics lattices, 1 instance of the Toda type lattices and 1 instance of the discrete Toda type lattices, as shown on the Figure 4.
The shift $`T_n`$ corresponds to the lattice (3.2) and the diagonal shift $`T_l^1T_n`$ corresponds to the lattice (3.20), as was proved in the previous section. The following Theorem demonstrates that the condition of the commutativity of these shifts is equivalent to some discrete lattice of the Toda type which can be interpreted as the nonlinear superposition principle of the equations (3.2) and (3.20).
###### Theorem 9.
The variables $`q,q^{\pm 1},q_{\pm 1}`$ are related by equation of the form
$$(T_l1)c(q^1q+\delta )+(T_n1)c(qq_1)=0.$$
(3.21)
* The function $`\phi `$ inverse to $`A^{}`$ satisfies equation $`\phi ^{}=R(\phi ),`$ that is $`f(y)=\phi (y+\epsilon ).`$ Therefore
$$C(Y)=A^{}(P)y_1=A^{}(f(q_1q^1))q_1+q=qq^1+\epsilon ,$$
that is $`Y=s(qq^1+\epsilon )`$ where $`s`$ is the inverse function to $`C.`$ It is easy to prove that $`s^{}`$ satisfies the same equation (3.14) as $`h=c^{},`$ that is $`s(y)=c(y+\stackrel{~}{\epsilon })+const.`$
Next, eliminating $`p`$ from the formulae
$$Y=a^{}(p)c(y_1),Y^1=a^{}(p)c(y)$$
brings to $`Y^1Y=(T_n1)c(y)`$ and this completes the proof. โ
It is clear that parameter $`\delta `$ can be set to zero by the proper choice of the function $`f`$ in formula (3.19) or, equivalently, by the shift $`q_n^lq_n^l+\delta l.`$ Therefore, the terms in (3.21) are symmetric and the shifts $`T_l`$ and $`T_n`$ plays the equal roles, although their origin was different. This suggests that $`T_l`$ corresponds to some relativistic lattice as well. Indeed, eliminating $`P`$ from the equation $`\dot{p}=r(p)(P^1P_1)`$ by means of the relations (3.18) brings to equation
$$\dot{p}=r(p)\left(\frac{p^1}{H(Y)}+\frac{p^1}{H(Y^1)}+\frac{G(Y)}{H(Y)}\frac{G(Y^1)}{H(Y^1)}\right)$$
and since $`Y=c(qq^1+\delta )`$ we obtain a relativistic lattice again. The coefficients of this equation coincide with $`h,g`$ up to the shift of argument.
### 3.5 Example: Heisenberg chain
In the previous sections the discrete symmetries of integrable lattices were studied. Now we will briefly discuss the continuous symmetries. I restrict myself by example of the equation (f<sub>0</sub>)<sup>1</sup><sup>1</sup>1These results were obtained in collaboration with Professors A.B. Shabat and A.P. Veselov.. This choice is motivated by the link between (f<sub>0</sub>) and the Heisenberg chain which became very popular due to its applications in discrete geometry . Remind that this model reads
$$s_t=as\times \left(\frac{s_1}{1+s,s_1}+\frac{s_1}{1+s,s_1}\right)+b\left(\frac{s_1+s}{1+s,s_1}\frac{s+s_1}{1+s,s_1}\right)$$
(3.22)
where $`s^3,`$ $`s,s=1,`$ $`,`$ and $`\times `$ denote standard scalar and vector products respectively, $`a`$ and $`b`$ are arbitrary constants. Up to the author knowledge this equation was introduced by Sklyanin in the case $`b=0`$ and by Ragnisco and Santini in the general case. The continuous limit of the Sklyanin lattice is the Heisenberg model
$$s_t=s\times s_{xx},s,s=1.$$
It was noticed that $`r`$-matrices for these models coincide and, more generally, this property can be accepted as a definition of correct discretization for a given continuous equation .
In order to obtain (f<sub>0</sub>) let us consider the complexification $`s^3s^3`$ and the stereographic projection
$$s=S(u,v)=\frac{1}{uv}(1uv,i+iuv,u+v).$$
(3.23)
It is convenient to represent the flow (3.22) for arbitrary set of parameters $`a,b`$ as a linear combination of the flows corresponding to the sets $`a=i,b=\pm 1.`$ These flows are given by the following formulae in terms of the variables $`u,v:`$
$$u_{t_+}=\frac{(u_1u)(uv)}{u_1v},v_{t_+}=\frac{(uv)(vv_1)}{uv_1},$$
(3.24)
$$u_t_{}=\frac{(u_1u)(uv)}{u_1v},v_t_{}=\frac{(uv)(vv_1)}{uv_1}.$$
(3.25)
The lattice (3.24) appeared in for the first time.
Next, notice that elimination of $`P`$ in virtue of (3.19) brings the formulae (3.17) to the form
$$\dot{q}=\frac{f(q_1q^1)g(q_1q)}{h(q_1q)},\dot{q}^1=\frac{f(qq_1^1)g(q^1q_1^1)}{h(q^1q_1^1)}.$$
It is easy to see that this system corresponding to the case (f<sub>0</sub>) ($`f(y)=g(y)=1/y,`$ $`h(y)=1/y^2`$) coincide with (3.24) if we assume $`q=u,`$ $`q^1=v`$ and $`t=t_+.`$ Since the lattice (3.25) is obtained by reflection $`nn`$ we immediately come to the following statement.
###### Proposition 10.
Both variables $`u`$ and $`v`$ satisfy the lattices
$$q_{t_\pm t_\pm }=q_{t_\pm }^2\left(\frac{q_{\pm 1,t_\pm }}{(q_{\pm 1}q)^2}\frac{q_{1,t_\pm }}{(qq_1)^2}\frac{1}{q_{\pm 1}q}+\frac{1}{qq_1}\right)$$
in virtue of the equations (3.24), (3.25).
The lattices (3.24), (3.25) are Hamiltonian. Their Hamiltonians are
$$H_+=\underset{n}{}\mathrm{log}\frac{u_1v}{uv},H_{}=\underset{n}{}\mathrm{log}\frac{uv_1}{uv}$$
respectively and the Poisson brackets are of the form
$$\{u_m,v_n\}=(u_nv_n)^2\delta _{mn},\{u_m,u_n\}=\{v_m,v_n\}=0.$$
(3.26)
The next proposition can be proved by straightforward calculations. It demonstrates that the lattices (3.24), (3.25) are the symmetries of each other and the shift $`(u_n,v_n)(u_{n+1},v_{n+1})`$ defines the Bรคcklund transformation for some hyperbolic system.
###### Proposition 11.
The Hamiltonians $`H_+`$ and $`H_{}`$ are in involution, the vector fields $`_{t_+}`$ and $`_t_{}`$ commute and the variables $`u_n,v_n`$ satisfy the system
$$u_{t_+t_{}}=\frac{2u_{t_+}u_t_{}}{uv}u_{t_+}u_t_{},v_{t_+t_{}}=\frac{2v_{t_+}v_t_{}}{vu}+v_{t_+}+v_t_{}.$$
(3.27)
In terms of the vector (3.23) this system reads
$$s_{t_+t_{}}+s_{t_+},s_t_{}s+is\times (s_{t_+}+s_t_{})=0,s,s=1.$$
Commutativity of the flows $`_{t_+},`$ $`_t_{}`$ allows to construct compatible zero curvature representations
$$W_{t_+}=U_1^+WWU^+,W_t_{}=U_1^{}WWU^{},$$
so that $`trW_N\mathrm{}W_1`$ generates the common first integrals of the both lattices under the periodic boundary conditions $`u_N=u_0,`$ $`v_N=v_0.`$ The matrices $`W,U^\pm `$ are given by the formulae
$$W=\lambda I+P(u,v),U^+=\frac{1}{2\lambda }(P(u,v_1)P(v_1,u)),$$
$$U^{}=\frac{1}{2(\lambda +1)}(P(v,u_1)P(u_1,v))$$
where $`P`$ denotes the projector
$$P(u,v)=\frac{1}{uv}\left(\begin{array}{cc}v& uv\\ 1& u\end{array}\right).$$
The Poisson structure (3.26) can be written in the $`r`$-matrix form
$$\{W_m(\lambda )\underset{,}{}W_n(\mu )\}=\delta _{mn}[r,W_m(\lambda )W_n(\mu )]$$
with the same $`r`$-matrix as for the Heisenberg model:
$$r=\frac{1}{\lambda \mu }\left(\begin{array}{cccc}1& 0& 0& 0\\ 0& 0& 1& 0\\ 0& 1& 0& 0\\ 0& 0& 0& 1\end{array}\right).$$
Equation (3.27) gives an example of hyperbolic symmetry. Any lattice (3.2) of the relativistic Toda type admitting duality transformations possesses also evolution symmetries, the simplest representative is given in the following Theorem.
###### Theorem 12 ().
The Lagrangian (3.3) admits the variational symmetry of the form
$$q_\tau =r(p)(h(y_1)p_1+h(y)p_1+g(y_1)+g(y))+R_1p^2$$
(3.28)
if and only if the coefficients $`r,g,h`$ satisfy the system (3.10).
Obviously, one can rewrite this symmetry as the partial differential equation if use the lattice (3.2) itself and assume $`q=u,`$ $`q_1=v:`$
$`u_\tau +u_{tt}`$ $`=2r(u_t)(h(vu)v_t+g(vu))+R_1u_t^2,`$
$`v_\tau v_{tt}`$ $`=2r(v_t)(h(vu)u_t+g(vu))+R_1v_t^2.`$
Returning to the system (3.24) one obtains the symmetry ($`t=t_+`$)
$$\begin{array}{cc}\hfill iu_\tau u_{tt}& =2u_t^2\left(\frac{v_t}{(uv)^2}\frac{1}{uv}\right),\hfill \\ \hfill iv_\tau +v_{tt}& =2v_t^2\left(\frac{u_t}{(uv)^2}\frac{1}{uv}\right)\hfill \end{array}$$
(3.29)
which is equivalent, in the geometrical terms (3.23), to the modified Heisenberg model
$$s_\tau =s\times s_{tt}\frac{i}{2}s_ts_t,s_t,s,s=1.$$
(3.30)
So we have proved that the flows defined by the lattice (3.22) at $`a=i,b=1`$ and equation (3.30) commute. One can reformulate this result as follows:
###### Proposition 13.
The shift $`ss_1`$ in the Heisenberg chain (3.22) for $`a=i,b=1`$ defines the $`t`$-part of the Bรคcklund transformation for equation (3.30).
Notice, that according to (3.24) this shift is equivalent to solving of Riccati equation on the variable $`v_1:`$
$$u_t=\frac{(u_1u)(uv)}{u_1v},v_{1,t}=\frac{(u_1v_1)(v_1v)}{u_1v}.$$
Another choice of dependent variables allows to rewrite (3.29) in the form
$$iu_{1,\tau }=u_{1,tt}\frac{2u_{1,t}^2}{u_1v},iv_\tau =v_{tt}\frac{2v_t^2}{u_1v}.$$
This system is equivalent to the Heisenberg model $`\sigma _\tau =\sigma \times \sigma _{tt},`$ $`\sigma ,\sigma =1`$ in terms of the vector $`\sigma =S(u_1,v),`$ which is related with the vectors (3.23) by the formula
$$\sigma =\frac{s_1+sis_1\times s}{1+s_1,s}.$$
## 4 Multifield examples
In the previous sections we assumed that the variable $`q`$ was scalar ($`q๐ฝ,`$ $`๐ฝ=,`$). The brief analysis proves that the Theorems 1 and 5 remain valid for the vector case $`q๐ฝ^N`$ as well. Thus, the multifield generalizations can be obtained by finding of the mappings $`T:๐ฝ^{2N}๐ฝ^{2N}`$ with the special structure described in these Theorems. This problem is much more difficult than in the scalar case and the classification of such mappings is far from completeness.
However it is not difficult to find some particular examples. The most simple are the generalizations of the discrete Heisenberg equation (A) and the Heisenberg lattice (f<sub>0</sub>) related to the Jordan triple systems. It was Svinolupov who recognized the role of the Jordan algebraic structures in the theory of integrable systems for the first time. I provide only necessary information about the Jordan triple systems, the interested reader can find more in and references therein.
### 4.1 Jordan triple systems
The ternary algebra $`J`$ with multiplication $`\{\}:J^3J`$ is called Jordan triple system if the following identities hold
$$\{abc\}=\{cba\},$$
(4.1)
$$\{ab\{cde\}\}\{cd\{abe\}\}=\{\{cba\}de\}\{c\{bad\}e\}.$$
(4.2)
Some consequences are:
$$\begin{array}{cc}& \{ab\{aca\}\}=\{a\{bac\}a\}=\{\{aba\}ca\},\{a\{bab\}c\}=\{\{aba\}bc\},\hfill \\ & 2\{\{abc\}bd\}=\{a\{bcb\}d\}+\{c\{bab\}d\}.\hfill \end{array}$$
(4.3)
Calculations are simplified by use of the linear operators $`L_{ab},M_{ab},M_a:JJ`$ defined for arbitrary elements $`a,bJ`$ as follows:
$$L_{ab}(c)=\{abc\},M_{ab}(c)=\{acb\},M_a(c)=M_{aa}(c)=\{aca\}.$$
Notice, that identity (4.2) is equivalent to
$$[L_{ab},L_{cd}]=L_{\{cba\}d}L_{c\{bad\}}.$$
(4.4)
We are especially interested in rational expressions. They are build from the inverse elements which are defined as $`a^1=M_a^1(a).`$ In some Jordan triple systems the operator $`M_a`$ is degenerate. In such cases the notion of inverse element can be partly substituted by the notion of a deformation vector which is the solution of the system $`b/a=M_b`$ . However, for sake of simplicity, we shall consider only the cases when $`detM_a0`$ almost everywhere in $`J.`$ The Jordan triple system with this property are called Jordan triple system with invertible elements.
The following Lemma demonstrates that some properties of $`a^1`$ are the same as in a ring.
###### Lemma 14.
Let the operator $`M_a`$ be invertible then
$$M_{a^1}=M_a^1,(a^1)^1=a,(a^1)_x=M_a^1(a_x)$$
(4.5)
and the following rules of cancellation of $`a`$ and $`a^1`$ are valid
$$\begin{array}{cc}& \{aa^1b\}=b,\{a^1\{acb\}a^1\}=\{cba^1\},\hfill \\ & \{a\{a^1ca^1\}b\}=\{ca^1b\},\{ba\{a^1ca^1\}\}=\{bca^1\}.\hfill \end{array}$$
(4.6)
* Let $`bJ,`$ $`c=M_a(b),`$ $`d=M_{a^1}(c).`$ One obtains using (4.3)
$`M_a(d)`$ $`=\{a\{a^1ca^1\}a\}=2\{aa^1\{aa^1c\}\}\{a\{a^1aa^1\}c\}`$
$`=2\{aa^1\{aa^1\{aba\}\}\}\{\{aa^1a\}a^1c\}=\{aa^1\{aba\}\}`$
$`=\{aba\}=M_a(b),`$
and therefore $`b=d.`$ Arbitrariness of $`b`$ implies $`M_{a^1}M_a=I.`$
Further on, $`(a^1)^1=M_{a^1}^1(a^1)=M_a(a^1)=a.`$
For arbitrary $`b`$
$$\{aa^1b\}=\{aa^1\{aM_a^1(b)a\}\}=\{\{aa^1a\}M_a^1(b)a\}=\{aM_a^1(b)a\}=b.$$
Taking this into account when differentiating relation $`a=\{aa^1a\}`$ one proves the last formula in (4.5).
In virtue of (4.4) and relation $`L_{aa^1}=I`$ which is already proved one obtains
$$L_{M_{a^1}(c)a}=L_{a^1\{ca^1a\}}+[L_{a^1c},L_{a^1a}]=L_{a^1c}+[L_{a^1c},I]=L_{a^1c},$$
and the last formula in (4.6) is proved. The next to the last is equivalent (use operator notation). Finally, (4.2) implies
$$\{a^1\{acb\}a^1\}=2\{a^1bc\}L_{M_{a^1}(b)a}(c)=\{a^1bc\}.\mathit{}$$
Now we can prove the formula (โharmonic meanโ)
$$(a^1+b^1)^1=\{a(a+b)^1b\}.$$
(4.7)
Let us denote $`a+b=c^1,`$ then
$`M_{(a^1+b^1)}(\{acb\})=`$ $`\{a^1\{acb\}a^1\}+\{b^1\{acb\}b^1\}`$
$`+2\{a^1\{ac(c^1a)\}b^1\}`$
$`=`$ $`\{cba^1\}\{b^1ac\}+2b^1.`$
Symmetrization on $`a`$ and $`b`$ gives (4.7).
Analogue of the Killing form in the Jordan triple system is the scalar product $`a,b=trL_{ab}.`$ Relation (4.4) implies the invariance property
$$\{abc\},d=a,\{bcd\}$$
(4.8)
of this product. Further on we assume that it is also symmetric and nondegenerate. Notice that if the element $`a`$ is invertible then the equalities $`L_{ab}=M_aM_{a^1b},`$ $`L_{ba}=M_{a^1b}M_a`$ imply $`a,b=b,a.`$ Hence, in virtue of continuity, in the Jordan triple systems with invertible elements the symmetry property always holds.
Examples below together with the reductions $`a=\pm a^\tau `$ of the Examples 1 and 2 (with $`M=N`$) exhaust all simple Jordan triple systems aside from two exceptional ones.
#### Example 1.
The linear space $`J`$ of $`N\times N`$ matrices becomes the Jordan triple system with respect to the triple product defined by means of the standard matrix multiplication as follows
$$\{abc\}=\frac{1}{2}(abc+cba).$$
The operator $`M_a`$ is invertible iff $`deta0,`$ that is almost everywhere. The element $`a^1`$ coincides with inverse matrix. The subspaces of symmetric and skewsymmetric matrices are Jordan triple systems as well. However, in the Jordan triple system of the skewsymmetric matrices of odd order the operator $`M_a`$ is degenerate for all $`a.`$ The scalar product is $`a,b=trab.`$
#### Example 2.
Previous example admits generalisation for $`N\times M`$ matrices:
$$\{abc\}=\frac{1}{2}(ab^\tau c+cb^\tau a),$$
where <sup>ฯ</sup> denotes transposition. In particular, if $`M=1`$ then $`J`$ turns into the $`N`$-dimensional vector space with multiplication
$$\{abc\}=\frac{1}{2}(a,bc+c,ba)$$
where $`,`$ denotes the standard scalar product. However, the operator $`M_a`$ is not invertible for $`MN`$.
#### Example 3.
More interesting triple product in the $`N`$-dimensional vector space is given by
$$\{abc\}=a,bc+c,baa,cb.$$
The scalar product in $`J`$ coincides with the standard one. Operator $`M_a,`$ its inverse and vector $`a^1`$ are defined by formulae
$$M_a(b)=2a,baa,ab,M_a^1=a,a^2M_a,a^1=a,a^1a.$$
### 4.2 Jordan analogues of Heisenberg equations
The lattice (2.2), (A) of discrete relativistic Toda type admits literal generalization
$$\mu (T_m1)x^1+\nu (T_n1)y^1+\lambda (T_mT_n1)z^1=0,\lambda +\mu +\nu =0$$
(4.9)
for arbitrary Jordan triple system with invertible elements. Indeed, using relation (4.7) one can prove that the mapping $`T:J^2J^2`$ given by the formulae
$$X=\nu y^1+\lambda (x+y)^1,Y=\mu x^1\lambda (x+y)^1$$
coincide with its inverse and therefore the duality transformations (2.4) map equation (4.9) into itself. The nonrelativistic analogue of this equation corresponding to the set of parameters $`\mu =1,`$ $`\nu =1,`$ $`\lambda =0`$ can be obtained along the same scheme as in the Section 2.3.
Equation (4.9) is the Euler equation for the Lagrangian
$$=\underset{m,n}{}(\mu \mathrm{log}detM_x+\nu \mathrm{log}detM_y+\lambda \mathrm{log}detM_z).$$
In order to prove this, note that if $`f(u)=\frac{1}{2}\mathrm{log}detM_u`$ then, in virtue of Lemma 14,
$$\frac{f}{u},v=\frac{d}{d\epsilon }f(u+\epsilon v)|_{\epsilon =0}=tr(M_u^1M_{uv})=trL_{u^1v}=u^1,v$$
and therefore $`f/u=u^1.`$
Analogously, the lattice (f<sub>0</sub>) of the relativistic Toda type admits generalization
$$\dot{y}=pp_1,\dot{p}=M_p(M_{y_1}^1(p_1)M_y^1(p_1)y_1^1+y^1)$$
corresponding to the Lagrangian
$$=๐t\underset{n}{}(\frac{1}{2}\mathrm{log}detM_pp,y^1\frac{1}{2}\mathrm{log}detM_y).$$
Indeed, if $`f(u)=u^1,a`$ then
$$\frac{d}{d\epsilon }f(u+\epsilon v)|_{\epsilon =0}=M_u^1(v),a=M_u^1(a),v$$
and therefore $`f/u=M_u^1(a).`$
The duality transformations (3.4) are defined by the mapping $`T`$ of the form
$$Y=p^1y^1,P=y^1M_y^1(p)$$
which is involutive. Applying the scheme of the Section 3.3 one obtains the Toda type lattice
$$\ddot{q}=M_{\dot{q}}((q_1q)^1(qq_1)^1)$$
which is equivalent to the Jordan Volterra lattice
$$\dot{p}=M_p(PP_1),\dot{P}=M_P(p_1p).$$
#### Acknowledgements.
Author thanks Professors B.A. Kupershmidt and A.B. Shabat for useful discussions. This work was supported by the grant # 99-01-00431 of the Russian Foundation for Basic Research. |
warning/0001/hep-ph0001250.html | ar5iv | text | # Electrons, New Physics, and the Future of Parity-Violation
## I Introduction
In addition to celebrating the silver anniversary this year of electron scattering at the MIT-Bates Laboratory, we may also mark the passing of 25 years for another sub-field of physics: parity-violation (PV) in semi-leptonic neutral current interactions. Since the MIT-Bates Laboratory has made important contributions to the field of neutral current PV, it seems highly appropriate to consider the future of the field at this Symposium. Paul Souder has discussed in detail the history of neutral current PV in electron scattering, and Besty Beise has summarized the present program of strange-quark searches here at MIT-Bates, the Jefferson Lab, and Mainz. Consequently, I will focus on the future: where the field might go once the current round of parity-violating electron scattering (PVES) experiments are completed. I will also broaden the topic to include PV in atoms. Historically, atomic parity-violation (APV) has been at the forefront of the field, and it will undoubtedly continue to hold such a position in the future. In discussing the future, I hope to convey the following three points: (a) the forefront of neutral current PV will consist of searches for physics beyond the Standard Model; (b) APV and PVES can play complementary roles in this search for โnew physicsโ; and (c) parity-violating, low-energy semi-leptonic processesand high-energy collider searches can, in principle, provide complementary insights as to what may lie beyond the Standard Model (SM). My guess is that this situation will persist for the better part of the next decade, until the LHC begins to produce significant physics results.
Before considering the next decade, it is useful to look back briefly a the last quarter century. One may trace the birth of this field to the Bouchiats, who proposed in 1974 that studying PV atomic processes might produce evidence for the weak neutral currents of the SM in the semi-leptonic sector Bou74 . The Bouchiats suggested a clever technique for enhancing the signal for these tiny neutral currents so that they might be observed in table top experiments. This technique, called โStark mixingโ, relies on the interference of a Stark-induced mixing of opposite parity states in an atom and the mixing caused by weak neutral currents. In effect, the Stark-induced amplitude functions as a lever arm to magnify the importance of the neutral current amplitude. The importance of this idea cannot be overstated. Following the Bouchiatsโ proposal, a number of groups endeavored to search for weak neutral currents in APV, using either the Stark-mixing idea or by studying the rotation of plane-polarized light as it passes through a gas of atoms (see, e.g., Ref. Bud98 and references therein). In fact, the recent, very precise result for cesium APV reported by the Boulder group was obtained using a variation on the Bouchiatsโ original Stark-mixing idea Woo97 . The result of these APV experiments has been to confirm the SM prediction for the structure of the weak neutral current in the low-energy domain at the few percent level. Given the scope of effort involved in testing the SM in high-energy collider experiments, the results of the APV measurements represent a significant triumph for table top physics.
Among noteworthy collider experiments are those involving semi-leptonic PVES. Results from the SLAC deep-inelastic PVES experiment on deuterium were reported in the late 1970โs Pre78 . These results also confirmed the structure of the semi-leptonic weak neutral currents of the SM and yielded a value for the weak mixing angle with nine uncertainty. About a decade later, the collaboration at Mainz reported results on a quasi-elastic PVES experiment involving a <sup>8</sup>Be target Hei89 . This experiment tested a different combination of the neutral current parameters than tested by the SLAC experiment. Shortly after the appearance of the Mainz result, the results of the elastic PVES experiment on <sup>12</sup>C performed at Bates were repoted Sou90 . Again, the results of the carbon experiment complemented those from quasi-elastic and deep inelastic measurements and confirmed the predictions of the SM. As discussed in more detail by Paul Souder, an important benefit of these PVES experiments was the development of experimental expertise and technology that is crucial to the sucess of the present program and the future prospects of PVES.
Turning back to APV, the Boulder groupโs result for cesium dominates the present landscape. The group reports an experimental error of less than 0.4 %. As with the earlier APV and PVES experiments, the goal of the cesium measurement was to test the SM. The cesium results deviate from the SM prediction by about 1.5%, representing a 2.5$`\sigma `$ difference. The potentially serious consequences of this deviation call for a repeat measurement. To that end, the Bouchiat group in Paris is currently involved in another Stark-mixing experiment with cesium, although the experimental uncertainty is not projected to be as small as in the Boulder measurement.
Over the last decade, the emphasis of PVES has shifted away from SM tests to the study of hadron structure. As Paul Souder and Besty Beise discussed, a well-defined program of measurements to determine the nucleonโs strange quark vector current form factors is underway PVES89 . Instead of studying the structure of the lepton-quark weak neutral current interaction, these experiments rely on the present knowledge of that interaction in order to learn something new about the sea-quark structure of the nucleon. Results from the MIT-Bates backward angle experiments on the proton and deuterium have been reported by the SAMPLE collaboration Mue97 , and the results of a forward angle measurement have been published by the HAPPEX collaboration at the Jefferson Lab Ani98 . The list of approved strange-quark experiments also includes the G0 experiment at the Jefferson, a Hall-C experiment on <sup>4</sup>He, and an experiment on the proton at the MAMI facility in Mainz PVES89 . The HAPPEX collaboration has also been approved to run another forward angle proton measurement at $`Q^2`$ similar to that of the SAMPLE experiment. In addition, the G0 detector will be used to measure the axial vector $`N\mathrm{\Delta }`$ transition form factor.
For both PVES and APV, the next generation of experiments are on the horizon. The groups in Seattle and Berkeley have undertaken measurements of APV observables for several atoms along the chain of isotopes. As I discuss below, ratios of such observables are less sensitive to atomic theory uncertainties than is the APV observable for a single isotope. One hopes that such measurements may provide an even more precise tool for uncovering new physics than the Boulder cesium experiment. In order to realize this goal, however, one requires a new level of insight into nuclear structure than required for the interpretation of a single isotope APV measurement. In the case of PVES, the interest of future experiments seems to be returning to studying the weak neutral current interaction at the elementary fermion level. To that end, a purely leptonic experiment involving PV Mรถller scattering has been approved for SLAC SLAC97 . Similarly, a letter of intent to perform a precise, forward angle PV $`\stackrel{}{e}p`$ experiment at the Jefferson Lab has appeared LOI99 . Finally, the Jefferson Lab PAC is considering a proposal to carry out elastic PVES with a <sup>208</sup>Pb target Mic99 . This experiment would provide the most precise information we have to date on the distribution of neutrons in a nucleus, something of considerable interest to nuclear structure physicists. At the same time, the <sup>208</sup>Pb experiment may provide enough nuclear structure information to help with the interpretation of the APV isotope ratio studies in terms of new electroweak physics. In this respect, the lead experiment would solidify a unique marriage of table top and collider efforts having important consequences for atomic, nuclear, and particle physics.
In the remainder of this discussion, I consider these future APV and PVES experiments in detail. First, I review the motiviation for searching for new physics at low-energies. I subsequently review the basics of the relevant PV observables and show how precise measurements of these observables can provide a window on physics at the TeV scale. I give a few examples of new physics scenarios that can be tested by low-energy PV and consider a possible connection with nuclear $`\beta `$-decay. Finally, I discuss the relationship between the APV isotope ratio studies, the nuclear neutron distribution $`\rho _n(r)`$, and the PVES experiment on <sup>208</sup>Pb. For an in-depth discussion of these issues, I refer the reader to Refs. MRM99 ; Mus94
## II Searching for new physics
Although there exist a plethora of data confirming the electroweak sector of the Standard Model at the few $`\times 0.1\%`$ level, there also exist strong conceptual reasons to believe that the SM is only a piece of some larger framework. A nice perspective from which to view the reasons for this belief is the so-called high-energy desert. The high-energy desert is the region in mass scale ranging from the weak scale $`M_{WEAK}250`$ GeV up to the Planck scale $`M_P1/\sqrt{8\pi G_{NEWTON}}=2.4\times 10^{18}`$ GeV. The conceptual shortcomings of the SM appear at both edges of this desert. First, at the high-energy end, the SM does not appear to produce unification of the electroweak and strong interactions at any scale. If one perturbatively runs the SU(3$`)_C`$, SU(2$`)_L`$, and U(1$`)_Y`$ couplings up from the weak scale, they never meet at a common point. This lack of unification is undesireable, particularly if one believes a common framework ought to describe the electroweak, strong, and gravitational interactions.
At the low-energy ($`\mu <<M_{WEAK}`$) edge ofthe desert, the SM is similarly less than satisfying. The most obvious shortcoming is the presence of 19 independent parameters (in the limit of zero neutrino mass) which must be determined from experiment. In addition, the violation of discrete symmetries, such as parity and CP, is put in by hand. The SM does not explain why nature violates these symmetries; it simply incorporates them into a unified framework. Similarly, the quantization of electric charge must be put in by hand; it does not follow naturally (at tree-level in the theory) as does, say, the quantization of isospin charge Moh92 . A particularly serious challenge for the SM is to account for the wide range of mass scales in the SM spectrum. A related aspect of this โhierarchy problemโ has to do with quadratic divergences appearing in the renormalization of the Higgs mass. The presence of these divergences lead one to wonder why the Higgs mass should turn out to be at or below the weak scale without the aid of some fine tuning of electroweak parameters. In short, the SM leaves open many questions regarding the various mass scales governing low-energy physics.
Despite the phenomenological successes of the SM, then, one has good reason to believe there must exist some larger framework which contains the SU(3$`)_C\times `$SU(2$`)_L\times `$U(1$`)_Y`$ theory and which, presmumably, provides answers to the conceptual puzzles of the SM. Hence, there exists intense interest these days in the search for new physics. In considering what such new physics might be, one faces two broad questions: (a) Which new physics scenarios are most viable, both conceptually and phenomenologically? (b) What are the mass scales associated with a given scenario? In the remainder of this discussion, I will illustrate the insight parity-violating processes involving electrons might play.
## III PV observables
The basic quantity of interest in considering neutral current PV is the so-called weak charge, $`Q_W`$. This quantity is the weak neutral current analog of the EM charge. It gives the strength of the vector current coupling of $`Z^0`$-boson to an elementary fermion or system of fermions. For our purposes, it is useful to write the weak charge as
$$Q_W=Q_W(\text{SM})+\mathrm{\Delta }Q_W(\text{NEW})+\mathrm{\Delta }Q_W(\text{MB}),$$
(1)
where the first term, $`Q_W(\text{SM})`$ is the contribution to the weak charge from the SM. This contribution can be computed precisely and compared with an experimental value for $`Q_W`$. Any significant deviation would signal non-zero values for the remaining terms. Of these, $`\mathrm{\Delta }Q_W(\text{NEW})`$ represents contributions from possible physics beyond the SM, while $`\mathrm{\Delta }Q_W(\text{MB})`$ denotes contributions from convnetional many-body effects, such as strong interactions among quarks which interfere with the $`Z^0`$-quark interaction. The extent to which we can reliably compute the latter determines the confidence with which we can learn about $`\mathrm{\Delta }Q_W(\text{NEW})`$ from a given measurement.
Presently, the most precise determination of $`Q_W`$ has been obtained with APV in cesium. In an APV process, the weak neutral current interaction between the electron and nucleus generates a PV atomic Hamiltonian which mixes states of opposite parity in the atom:
$$_W^{PV}=_W^{PV}(\text{NSID})+_W^{PV}(\text{NSD}).$$
(2)
Here, โNSIDโ and โNSDโ denote, respectively, nuclear spin-independent and nuclear spin-dependent components of the interaction. The for mer arises from the product of axial vector electron and vector nuclear currents, whereas the latter arises from a $`V(e)\times A(\text{nucleus})`$ structure. These terms can be separated by measuring PV transitions between different hyperfine levels. The NSID term contains $`Q_W`$. The physics of the NSD term, which includes the effects of the nuclear anapole moment, is also interesting, though I will not consider it further here (for a general discussion, see Ref. Mus91 ).
As pointed out by the Bouchiats, the small parity-mixing effects caused by $`_W^{PV}`$ can be enhanced by applying an electric field, which also causes states of opposite parity to mix. Reversing the direction of the applied field can isolate terms in the transition rate which depend on the interference of the Stark and weak interaction amplitudes. In the end, one extracts a ratio such as
$$|A_{PV}|/|A_{STARK}|=\xi Q_W,$$
(3)
where $`\xi `$ is an atomic structure-dependent constant that must be computed by atomic theorists. Thus, any errors associated with atomic theory will propagate into uncertainties in $`Q_W`$ (we might associate these conventional, atomic structure uncertainties with $`\mathrm{\Delta }Q_W(\text{MB})`$ ). In fact, the dominant uncertainty in the present value for $`Q_W`$ of cesium is from atomic theory.
An alternate method for determining $`Q_W`$ is with PVES. In PVES, one scatters longitudinally polarized electrons from a target and compares the cross sections when the electron helicity is flipped. Any non-zero difference results from an interference of the PV neutral current electron-nucleus scattering amplitude and the more familiar, parity-conserving electromagnetic amplitude. The observable of interest in this case is the โleft-rightโ asymmetry
$$A_{LR}=\frac{N_+N_{}}{N_++N_{}}=a_0Q^2\left\{\frac{Q_W}{Q_{EM}}+F(q)\right\}.$$
(4)
Here, $`N_\pm `$ denote the number of electrons detected for a given helicity of the incident beam; $`a_0`$ is a constant whose scale is set by the Fermi and EM fine structure constants; $`Q^2`$ is the square of the momentum transfer; $`Q_{EM}`$ is the electromagnetic charge of the target; and $`F(q)`$ is a term which depends on hadronic or nuclear form factors. In principle, one can separate the effects of $`F(q)`$ from those of $`Q_W`$ by exploiting the kinematic dependence of the former. The goal of the present strange-quark program is to determined the contribution made by strange quarks to $`F(q)`$.
It is interesting to compare present and prospective determinations of $`Q_W`$ with those of other low-energy electroweak observables. In Table I I list several of interest.
The top line in Table I gives the present limits on the agreement of the cesium weak charge with the SM predicition. The Boulder group finds 2.5$`\sigma `$ deviation (about 1.5%) from the SM value. The following rows give the expected precision on the weak charge of the electron expected in the SLAC Mรถller experiment and the weak charge of the proton in a prospective Jefferson Lab experiment. It is worth noting that the electron and proton weak charges are suppressed at tree level by $`(14\mathrm{sin}^2\theta _W)0.1`$; the electron weak charge is further suppressed by SM radiative corrections Cza96 . Crudely speaking, then, a 10% determination of the proton or electron weak charge is equivalent to a 1% determination of the weak charge of the cesium atom. The fourth line gives the expected precision for the isotope ratio measurements at Berkeley and Seattle. Note that the prospective experimental error is much smaller than the present theoretical uncertainty โ a point I address at the end of this discussion.
The other entries in Table I include the anomalous magnetic moment of the muon, the permanent electric dipole moments (EDMโs) of the electron, neutron, and mercury atom; and the square of the $`ud`$ matrix element of the CKM matrix. Thus far, one has no evidence of a non-vanishing permanent EDM or of a muon anomalous moment which differs from the SM prediction. In the case of $`|V_{ud}|^2`$, however, an average over the results of nine superallowed, Fermi nuclear $`\beta `$-decays yields a deviation from the requirements of CKM unitarity at the 2.2$`\sigma `$ level ( about 0.3%) Tow95 ; Hag96 . It is intriguing that both semi-leptonic observables โ $`Q_W`$ from cesium APV and $`|V_{ud}|^2`$ from superallowed $`\beta `$-decay โ have the same relative sign for the experimental deviation from the SM prediction. If this discrepancy is due to new physics, this common sign may point to a common new physics scenario, as I discuss below.
## IV PV and new physics
Before considering specific scenarios for physics beyond the SM, it is useful to consider the generic sensitivity of PV observable to such scenarios. In doing so, I follow the discussion of Ref. MRM99 restrict my attention to those scenarios which generate new effective, four-fermion interactions. Specifically, I write the PV fermion-fermion interaction as
$$=_{S.M.}^{PV}+_{NEW}^{PV},$$
(5)
where
$$_{S.M.}^{PV}=\frac{G_F}{2\sqrt{2}}g_A^e\overline{e}\gamma _\mu \gamma _5e\underset{f}{}g_V^f\overline{f}\gamma ^\mu f$$
(6)
gives the SM contribution and
$$_{NEW}^{PV}=\frac{4\pi \kappa ^2}{\mathrm{\Lambda }^2}\overline{e}\gamma _\mu \gamma _5e\underset{f}{}h_V^f\overline{f}\gamma ^\mu f$$
(7)
is the contribution from some new physics. Here, $`g_A^e`$ axial vector electron-$`Z^0`$ coupling and $`g_V^f`$ is the vector current coupling of the $`Z^0`$ to fermion $`f`$. In Eq. (7), $`\mathrm{\Lambda }`$ denotes the mass scale associated with the new physics and $`\kappa ^2`$ parameterizes the overall strength of the interaction. The $`h_V^f`$ give the scenario-specific couplings of the electron axial vector cu rrent to the vector current of fermion $`f`$. If the SM interaction in Eq. (6) determines the SM value of $`Q_W`$, the the fractional shift induced by the new interaction in Eq. (7) is
$$\frac{\mathrm{\Delta }Q_W}{Q_W(\text{SM})}=\frac{8\sqrt{2}\pi }{\mathrm{\Lambda }^2G_F},$$
(8)
assuming $`g_A^eg_V^f`$ and $`h_V^f`$ have commensurate magnitudes. If an experiment is sensitive to shifts on the order of $`\mathrm{\Delta }Q_W/Q_W(\text{SM})0.01`$, then Eq. (8) implies one is probing new physics at the $`\mathrm{\Lambda }20\kappa `$ TeV scale. For new physics of a strong-interaction character, one expects $`\kappa ^21`$, while for new gauge interactions one expects $`\kappa ^2\alpha `$. In either case, high-precision PV measurements are incredibly powerful probes of physics at the TeV scale.
It is instructive to consider how these general features apply in the case of specific new physics scenarios. One of the most interesting such scenarios is that of extended gauge symmetry. The basic of extended gauge symmetry is that the SM group structure is embedded in some larger group $`G`$. The full symmetry of $`G`$ may break down spontaneously at one or more scales $`M_X`$ above the weak scale, leaving the SU(3$`)_C\times `$SU(2$`)_L\times `$U(1$`)_Y`$ symmetry of the SM intact at $`M_{WEAK}`$. In principle, the gauge bosons associated with the additional symmetries of $`G`$ will acquire masses commensurate with the symmetry breaking scales $`M_X`$. If one of these scales is not too much larger than $`M_{WEAK}`$, then the additional massive gauge bosons could generate small effects in low-energy processes.
In addition to its phenomenological implications, extended gauge symmetry can provide resolution to some of the rough edges of the SM. For example, if $`G`$ contains an SU(2$`)_R`$ subgroup, then one has a natural explanation for PV at low-energies. At some high scale, one has exact parity symmetry. However, if the scale of symmetry breaking associated with the right-handed sector is much larger than $`M_{WEAK}`$, the right-handed gauge bosons will be too heavy to compete effectively with the SM gauge bosons, so that low-energy processes favor the left-handed sector. Similarly, the electromagnetic charge can appear as a generator of $`G`$, in which case its quantization is natural. Even the apparent lack of SM coupling unification can be resolved by extended gauge symmetry. The presence of additional symmetry breaking scales implies that the running of the couplings will change as one crosses each scale. Thus, there exists sufficient room within different extended gauge group scenarios to bring about coupling unification near the expected grand unified scale.
Here, I concentrate on the neutral current phenomenology of extended gauge symmetry. Specifically, I consider a scenario in which spontaneous symmetry breaking of $`G`$ yields a second neutral gauge boson $`Z^{}`$ with mass not too different from the weak scale. To make life simple, I also consider the case in which this $`Z^{}`$ does not mix with the SM $`Z^0`$. If it did mix, its effects would show up strongly in the $`Z`$-pole observables. In fact, the latter severely constrain the mass of a $`Z^{}`$ that does mix with the $`Z^0`$ Lan95 . In the language of Eq. (7), we have for this scenario $`\kappa ^2=\alpha ^{}`$, the fine-structure constant associated with the $`Z^{}`$ interaction; $`\mathrm{\Lambda }=M_Z^{}`$; and the $`h_V^f`$ to be specified by a particular scenario.
Given the experimental precisions listed in Table I, how sensitive would the different measurements be to extended gauge symmetry-induced new interactions? A detailed summary is given in Ref. MRM99 . Here, I quote a few illustrative examples. Extended gauge symmetry scenarios which fit naturally into the framework of heterotic strings live in a group called E<sub>6</sub>. The factors of E<sub>6</sub> include two U(1) groups called U(1$`)_\chi `$ and U(1$`)_\psi `$. The neutral gauge boson associated with the U(1$`)_\chi `$ would show up particularly strongly in low-energy PV if it had a sufficiently low mass; the $`Z_\psi `$, on the other hand, does not contribute to PV amplitudes at tree-level. Let $`G_\chi `$ denotes the Fermi constant associated with the interactions of the $`Z_\chi `$. We may characterize the sensitivity of various PV observable in terms of the ratio $`r_\chi =G_\chi /G_F`$. The present cesium APV is able to discern effects of the scale $`r_\chi 0.003`$ or larger. The sensitivities of the SLAC Mรถller experiment, the proposed Jefferson Lab PV $`ep`$ experiment, and the isotope ratio measurements are comparable. We can turn this statement about Fermi constants into mass limits by assuming the break down of E<sub>6</sub> to the SM $`\times `$ U(1$`)_\chi `$ occurs in one step, so that the coupling associated with the new U(1) group is maximal. In this case, the cesium APV, isotope ratio, and PVES measurements would probe $`M_{Z_\chi }`$ at about the one TeV level or better. In contrast, the sensitivity of the cesium measurement to a neutral right-handed gauge boson vastly exceeds the corresponding sensitivities of the isotope ratio and PVES measurements. Thus, the use of different low-energy PV measurements could prove useful in sorting out among competing scenarios.
It is also interesting to compare the sensitivities of low-energy PV and high-energy collider experiments. In terms of mass limits, the โreachโ of the present and prospective PV experiments exceeds that of the Tevatron by almost a factor of two. Even an up-graded Tevatron (Tev33) would only achieve comparable sensitivities. One must wait until the LHC has taken sufficient data before the PV sensitivities will be surpassed. In fact, the information provided by colliders and the PV measurements is complementary. The colliders are primarily sensitive to the mass scale associated with the new gauge boson relatively insensitive to the coupling strength $`g^{}`$ or detailed structure of the fermion-$`Z^{}`$ coupling. The PV observables, in contrast, depend on $`(g^{}/M_Z^{})^2`$ ($`\kappa /\mathrm{\Lambda }`$ in the language of Eq. (7)) and on the effective couplings fermion-$`Z^{}`$ couplings ($`h_V^f`$ in Eq. (7)).
To illustrate, I again consider E<sub>6</sub> theories Lon86 . The phenomenology of neutral E<sub>6</sub> gauge bosons is essentially governed by three parameters: $`M_Z^{}`$; a parameter $`\lambda _g`$ which governs the overall coupling strength $`g^{}`$ and whose value depends on the number of symmetry breaking steps leading to a massive $`Z^{}`$; and an โextendedโ weak mixing angle $`\varphi `$ which describes the structure of the additional โlow-energyโ U(1) group. Specifically, if $`Z_\chi `$ and $`Z_\psi `$ are the gauge bosons associated with the U(1$`)_\chi `$ and U(1$`)_\psi `$ groups, respectively, then a general neutral E<sub>6</sub> gauge boson can be written as
$$Z^{}=\mathrm{cos}\varphi Z_\psi +\mathrm{sin}\varphi Z_\chi .$$
(9)
The couplings $`h_V^f`$ of this $`Z^{}`$ to electrons and light quarks are given by
$`h_V^u`$ $`=`$ $`0`$ (10)
$`h_V^d=h_V^e`$ $`=`$ $`\left[\mathrm{sin}^2\varphi \sqrt{15}\mathrm{sin}\varphi \mathrm{cos}\varphi /3\right]/20.`$ (11)
Note that for $`\varphi =0`$ or $`\pi `$, $`Z^{}=Z_\psi `$ and all of the PV couplings vanish. The d-quark and electron couplings also vanish for $`\varphi =\varphi _c=\mathrm{tan}^1(\sqrt{5/3})`$ and have opposite signs for $`\varphi `$ on either side of $`\varphi _c`$. Thus, the net effect of the $`Z^{}`$ on $`Q_W`$ can be either positive or negative, depending on the value of $`\varphi `$. The present present cesium APV results favor $`\varphi >\varphi _c`$, if an E<sub>6</sub> gauge boson is responsible for the observed deviation from the SM value for $`Q_W`$. This kind of information about the structure of the extended gauge sector is difficult to obtain from high-energy collider limits.
It is also amusing to combine information obtained from colliders and low-energy experiments. To do so, letโs assume the E<sub>6</sub> gauge boson is responsible for the deviation of the cesium $`Q_W`$ from the SM value (about a two $`\sigma `$ effect). Under this assumption, one has a relationship between $`M_Z^{}`$, $`\lambda _g`$, and $`\varphi `$. A second condition derives from the CDF lower bounds, which are roughly 600 GeV with little dependence on the value of $`\varphi `$. Combining the two pieces of information, one obtains
$$600\text{GeV}\genfrac{}{}{0pt}{}{<}{}M_Z^{}\genfrac{}{}{0pt}{}{<}{}1.15\lambda _g\text{TeV},$$
(12)
where $`\lambda _g1`$. This range is already rather narrow. If a future up-graded Tevatron found no evidence for extra neutral gauge bosons with a mass less than about one TeV, then a low-mass $`Z^{}`$ would be ruled out as the culprit behind the cesium APV result.
Another popular extension of the Standard Model is supersymmetry. The literature on SUSY extensions of the SM is legion, so I will not discuss SUSY models in detail. The appeal of SUSY includes its solution to the hierarchy problem associated with mass renormalization. In addition, the gauge couplings in the minimal supersymmetric standard model (MSSM) unify at the GUT scale when run perturbatively up from the weak scale. Whether this coupling unification is fortuitous or reflects deeper physics can be debated. It is, nevertheless, intriguing. One important characteristic of the MSSM as far as low-energy phenomenology is concerned involves a quantity called R-parity. The R-parity quantum number is defined as
$$P_R=(1)^{3(BL)+2S},$$
(13)
where $`B`$, $`L`$, and $`S`$ denote the baryon number, lepton number, and spin, respectively, of a given particle. Every SM particle has $`P_R=1`$ while each superpartner has $`P_R=1`$. The MSSM conserves total $`P_R`$, which implies that every interaction involves an even number of superpartners. As a result, superpartners cannot appear in low-energy processes involving SM particles at tree-level. They only contribute through loops. Their effects are correspondgingly suppressed by loop factors, making them hard to see at low-energies.
It is possible, however, to write down simple extensions of the MSSM in which $`P_R`$ is not conserved. In such $`B`$ and/or $`L`$-violating theories, superpartner effects can appear at tree-level. To illustrate, consider a purely leptonic R parity-violating SUSY model. The relevant Lagrangian is Bar89
$$_{RPV}=\lambda _{ijk}(\stackrel{~}{e}_R^k)^{}(\overline{\nu }_L^i)^ce_L^j+\text{h.c.},$$
(14)
where $`\stackrel{~}{e}_R^k`$ denotes the bosonic superpartner of a right-handed charged lepton of generation $`k`$ (the other superscripts denote generation). Since the interaction contains three leptons, $`L`$ (and $`P_R`$) are not conserved. Tree-level exchange of the $`\stackrel{~}{e}_R^k`$ between lepton currents can generate new four-fermion effective interactions, such as the following interaction relevant to $`\mu `$-decay:
$$_{EFF}=(\lambda _{12k}/\sqrt{2}M_{\varphi _{kR}^e})^2\overline{e}_L\gamma _\alpha \nu _L^e\overline{\nu }_L^\mu \gamma ^\alpha \mu _L.$$
(15)
The interaction of Eq. (14) may provide a partial explanation for both the cesium APV result and the apparent CKM unitarity violation inferred from the superallowed $`\beta `$-decays. The reason has to do with the Fermi constant. Both the $`\beta `$-decay amplitude and the PV amplitude of Eq. (6) are written in terms of the Fermi constant. The reason is that these amplitudes depend on $`g^2/M_W^2`$, which can be related to the Fermi constant as measured in $`\mu `$-decay. At tree-level, this relationship is given by
$$\frac{g^2}{8M_W^2}=\frac{G_F}{\sqrt{2}}.$$
(16)
Because of the precision with which $`\mu `$-decay is measured, Eq. (16) must be modified to account for electroweak radiative corrections:
$$\frac{g^2}{8M_W^2}(1+\mathrm{\Delta }r)=\frac{G_\mu }{\sqrt{2}},$$
(17)
where $`\mathrm{\Delta }r`$ contains the radiative corrections. Suppose now some new physics, such as the interaction of Eq. (15), contributes to $`\mu `$-decay. Then one must further modify Eq. (17) as
$$\frac{g^2}{8M_W^2}(1+\mathrm{\Delta }r+\mathrm{\Delta }_\mu ^{NEW})=\frac{G_\mu }{\sqrt{2}},$$
(18)
where $`\mathrm{\Delta }_\mu ^{NEW}`$ gives the corrections from the new interaction. When writing down the amplitude for $`\beta `$-decay or PV, one needs $`g^2/M_W^2`$ in terms of $`G_\mu `$:
$$\frac{g^2}{8M_W^2}=\frac{G_\mu }{\sqrt{2}}(1\mathrm{\Delta }r\mathrm{\Delta }_\mu ^{NEW})$$
(19)
to first order in the small corrections.
To make contact with the semi-leptonic observables, it is useful to consider the effective Fermi constants $`G_F^\beta `$ and $`G_F^{PV}`$ which govern them. In terms of other quantities, these effective Fermi constants are
$`G_F^\beta `$ $`=`$ $`G_\mu (1\mathrm{\Delta }r+\mathrm{\Delta }r_\beta \mathrm{\Delta }_\mu ^{NEW}+\mathrm{\Delta }_\beta ^{NEW})|V_{ud}|^2`$ (20)
$`G_F^{PV}`$ $`=`$ $`G_\mu (1\mathrm{\Delta }r+\mathrm{\Delta }r_{PV}\mathrm{\Delta }_\mu ^{NEW}+\mathrm{\Delta }_{PV}^{NEW})Q_W,`$ (21)
where $`\mathrm{\Delta }r_\beta `$ and $`\mathrm{\Delta }r_{PV}`$ denote SM radiative corrections to the $`\beta `$-decay and PV amplitudes, respectively, and $`\mathrm{\Delta }_\beta ^{NEW}`$ and $`\mathrm{\Delta }_{PV}^{NEW}`$ are the corresponding contributions from new interactions.
The results of from the superallowed decays and cesium APV imply
$`G_F^{\beta ,EX}/G_F^{\beta ,SM}`$ $`<`$ $`1`$ (22)
$`G_F^{PV,EX}/G_F^{PV,SM}`$ $`<`$ $`1`$ (23)
where the $`EX`$ and $`SM`$ superscripts denote the experimental and SM values, respectively. From Eq. (20), we see that if the new physics contributions vanish, one obtains the conventional interpretation of the experimental results:
$`|V_{ud}|_{EX}^2/|V_{ud}|_{SM}^2`$ $`<`$ $`1`$ (24)
$`Q_W^{EX}/Q_W^{SM}`$ $`<`$ $`1.`$ (25)
However, an equally acceptable explanation is to assume $`|V_{ud}|^2`$ and $`Q_W`$ assume their SM values and that
$`\mathrm{\Delta }_\beta ^{NEW}\mathrm{\Delta }_\mu ^{NEW}`$ $`<`$ $`1`$ (26)
$`\mathrm{\Delta }_{PV}^{NEW}\mathrm{\Delta }_\mu ^{NEW}`$ $`<`$ $`1.`$ (27)
In particular, if both $`\mathrm{\Delta }_\beta ^{NEW}`$ and $`\mathrm{\Delta }_{PV}^{NEW}`$ vanish and if $`\mathrm{\Delta }_\mu ^{NEW}>0`$, the measured effective Fermi constants in $`\beta `$-decay and cesium APV would be smaller in magnitude than the SM predictions.
The R parity-violating interaction of Eq. (15) generates just such a positive value for $`\mathrm{\Delta }_\mu ^{NEW}`$:
$$\mathrm{\Delta }_\mu ^{NEW}=\frac{\lambda _{12k}^2}{4\sqrt{2}G_\mu M_{\varphi _{kR}^e}^2}.$$
(28)
Using the present experimental results and Eq. (20) one obtains
$$\lambda _{12k}=(0.027\pm 0.007)(M_{\stackrel{~}{e}_k}/100\text{GeV})$$
(29)
from superallowed decays and
$$\lambda _{12k}=(0.13\pm 0.05)(M_{\stackrel{~}{e}_k}/100\text{GeV})$$
(30)
from cesium APV. Although these results differ by more than one $`\sigma `$, one should keep in mind that the cesium result is the first PV result to differ from the SM, whereas the superallowed results depend on an average of $`ft`$ values for nine different decays, several of which have been measured more than once. In short, the precise magnitude of the deviation leading to Eq. (30) may not be as robust as that observed in $`\beta `$-decay. The primary point here is that the magnitudes of the results in Eqs. (29-30) are not too distinct, and the signs of the observed deviations are both consistent with the R parity-violating effects in Eqs. (15) and (28). It will be interesting to see whether future electron PV experiments also produce deviations from the SM predictions consistent with this SUSY scenario<sup>1</sup><sup>1</sup>1Another constraint on R parity-violating SUSY may be obtained from relations among electroweak parameters. The constraints imposed by these relations on some types of new physics have been analyzed in Ref. Mar99 . The corresponding SUSY constraints will be discussed in a forthcoming publication.
## V interpretation issues and neutron distributions
In general, the interpretation of precision, low-energy measurements raises thorny issues not relevant to high-energy measurements. The PV processes discussed here are no exception. To illustrate, I consider the interpretation of atomic PV. As noted above, the dominant error in the cesium weak charge comes from atomic theory. Although this theory error appears to have been reduced in light of new measurements of parity-conserving atomic transitions, it is questionable whether further reductions can be achieved. A clever strategy for evading this atomic structure uncertainty is to measure ratios of APV observables along an isotope chain. A representative ratio is
$$=\frac{A_{PV}^{NSID}(N^{})A_{PV}^{NSID}(N)}{A_{PV}^{NSID}(N^{})+A_{PV}^{NSID}(N)},$$
(31)
where $`A_{PV}^{NSID}(N)`$ is an APV nuclear spin-independent observable for an atom with neutron number $`N`$. Since the atomic electronic structure contributions $`A_{PV}^{NSID}(N)`$ and $`A_{PV}^{NSID}(N^{})`$ are relatively constant (for a given $`Z`$), the atomic structure-dependence drops out of the ratio $``$ and one has
$$\frac{Q_W(N^{})Q_W(N)}{Q_W(N)+Q_W(N^{})}SM(1+\delta _{}),$$
(32)
where $`_{SM}`$ is the value of the ratio in the SM.
The correction $`\delta _{}`$ contains contributions from possible new physics. As first pointed out by Fortson, Wilets, and Pang, however, there is also a second effect due to the variation of the neutron density $`\rho _n(r)`$ along the isotope chain For90 . To get an idea of the relative importance of these two contributions, one can model the nucleus as a sphere of constant neutron and proton density out to radii $`R_N`$ and $`R_P`$, respectively. In this case, one has
$$\delta _{}\left(\frac{2Z}{N+N^{}}\right)\mathrm{\Delta }Q_W^P\left(\frac{N^{}}{\mathrm{\Delta }N}\right)(Z\alpha )^2(3/7)\delta (\mathrm{\Delta }X_N),$$
(33)
where $`\mathrm{\Delta }Q_W^P`$ is the shift in the protonโs weak charge due to new physics,
$$\mathrm{\Delta }X_N=\frac{R_N^{}R_N}{R_P}$$
(34)
is the shift in the mean square neutron radius (relative to the proton radius) along the isotope chain, and $`\delta (\mathrm{\Delta }X_N)`$ is the uncertainty in this shift.
Several features of Eq. (33) are worth noting. First, the shift in the ratio $``$ due to new physics depends primarily on the shift in the weak charge of the proton. The shift in the weak charge of the neutron largely cancels out of the ratio, to first order in small shifts. Whereas the weak charge of a single isotope is slightly more sensitive $`\mathrm{\Delta }Q_W^N`$ than to $`Q_W^P`$, the sensitivity of $``$ to new physics is dominated by $`\mathrm{\Delta }Q_W^P`$. Second, the dependence of $``$ on variations in neutron radii along the isotope shift is enhanced by a factor of $`N^{}/\mathrm{\Delta }N`$. For a heavy atom like cesium or barium, for example, this enhancement factor can be on the order of 5. Thus, if one is going to use APV isotope ratio measurements to learn about $`\mathrm{\Delta }Q_W^P`$, one must have extremely precise knowledge of the shift in neutron radii.
At present, there exist no high-precision experimental determinations of the neutron radii of heavy nuclei. Consequently, nuclear theory must be used to determine the second term on the RHS of Eq. (33). To set the scale of the level of accuracy nuclear theory must achieve to make the isotope ratio measurements useful, supposed we require the uncertainty in the neutron radius term to be as small as the prospective experimental uncertainty in the value of $``$, namely, 0.1 %. Pollock Pol92 and Chen and Vogel Che93 ; Vog94 have analyzed the nuclear model spread in $`\mathrm{\Delta }X_N`$; from their analyses, we learn that nuclear theory is at least a factor of two away from achieivng the requisite precision (for a summary of the theoretical situation, see Ref. MRM99 ). In principle, this presents a stumbling block for the isotope ratio program.
There exist two strategies for overcoming this difficutly. One is to perform a direct measurement of $`\mathrm{\Delta }Q_W^P`$ using PVES from a proton target. From Eq. (4), we may write the proton asymmetry as
$$A_{LR}(^1\text{H})=a_0Q^2[Q_W^P+F^p(q)],$$
(35)
where $`Q_W^P`$ is the proton weak charge. The form factor term $`F^p(q)`$ vanishes in the forward angle limit. Thus, by going to forward angle kinematics, the $`Q_W^P`$ can be separated from $`F^p(q)`$. The form factor term is presently under study in the strange quark experiments. Upon completion of the strange quark program, this term should be known with sufficient precision over a large enough kinematic range to afford a precise separation of $`Q_W^P`$ in a future, forward angle measurement. A letter of intent for such a measurement has recently been issued LOI99 . The proposed measurement would employ a re-configured G0 apparatus in order to reach suffienct forward angle kinematics. It is hoped that this measurement will yield a 3-5% determination of $`Q_W^P`$. This level of precision would be comparable to a 0.1-0.2% determinatio of $``$, if the interpretation of the latter were not clouded by $`\rho _n(r)`$ uncertainties.
A second for getting around the $`\rho _n(r)`$ problem in Eq. (33) involves measuring the neutron distribution of a heavy nucleus using PVES. It is possible that a sufficiently precise determination of $`\rho _n(r)`$ on a single isotope would sufficiently constrain nuclear theory that the nuclear model-dependence in the isotope shifts, $`\delta (\mathrm{\Delta }X_N)`$ would be reduced to an acceptable level. The idea for using PVES to determine $`\rho _n(r)`$ was first suggested by Donnelly, Dubach, and Sick Don89 . These authors noted that the $`Z^0`$ preferentially sees neutrons over protons, since at tree-level in the SM, $`Q_W^P=14\mathrm{sin}^2\theta _W0.1`$ whereas $`Q_W^N=1`$. Thus, the PV asymmetry for scattering from a heavy nucleus should be quite sensitive to the neutron distribution. To illustrate this idea, consider PVES from a $`(J^\pi ,T)=(0^+,0)`$ nucleus. The asymmetry has the form Don89 ; Mus94
$$\left[\frac{4\sqrt{2}\pi \alpha }{G_F|Q^2|}\right]A_{LR}=Q_W^P+Q_W^N\frac{d^3xj_0(qx)\rho _n(\stackrel{}{x})}{d^3xj_0(qx)\rho _p(\stackrel{}{x})}.$$
(36)
Since $`\rho _p(\stackrel{}{x})`$ is typically known with very high accuracy, the PV asymmetry essentially becomes a โmeterโ of $`\rho _n(q)`$. This idea is being exploited in a proposal before the Jefferson Lab PAC Mic99 .
It goes without saying that a precise determination of $`\rho _n(q)`$ for any heavy nucleus is of fundamental interest for nuclear structure physics. From this standpoint alone, the investment of effort in making the measurement is well-justified. It remains to be seen, however, whether the information gleaned from a precise determination of $`\rho _n(q)`$ for <sup>208</sup>Pb at one or two kinematic points will suffice to reduce the nuclear structure uncertainty in Eq. (33). For example, it is unlikely that lead atoms will be used in the APV isotope ratios. The isotopes of Ba and Yb are currently under study in Seattle and Berkeley. Moreover, the interpretation of $``$ requires knowledge of $`\rho _n(r)`$ in more detail than implied by the simplified expression in Eq. (33). Whether knowledge of the momentum-space distribution at a few points will supply the necessary details about $`\rho _n(r)`$ is an open question. Finally, the constraints which knowledge of $`\rho _n(r)`$ for a single isotope would place on calculations of isotope shifts has yet to be quantified. In short, there exist several challenges for nuclear theory in making a PVES determination of $`\rho _n(q)`$ useful for the APV isotope ratios (for a recent discussion of these issues, see Ref. Hor99 ). From this standpoint, a measurement of the PV $`\stackrel{}{e}p`$ asymmetry provides a cleaner and more direct window on $`\mathrm{\Delta }Q_W^P`$.
## VI Conclusions
The field of parity-violation with electrons has made tremendous strides in 25 years. I hope this discussion has convinced the reader that its future prospects are just as exciting as its history. For the next decade at least, it is likely that PV with electrons will provide one of the most powerful probes of new physics at the TeV scale, complementing information to be gained from high-energy collider experiments. At the same time, it will remain a focal point for interdisciplinary activity, bringing together insights from particle, nuclear, and atomic physics. One may only speculate as to the new insights PV with electrons will provide for each field by the time a Bates-35 celebration is planned.
ACKNOWLEDGEMENTS
It is a pleasure to thank W.J. Marciano, D. Budker, R. Carlini, J.M. Finn, E.N. Fortson, S.J. Pollock, and P. Souder for useful discussions and S.J. Puglia for assistance in preparing the manuscript. This work was supported in part under U.S. Department of Energy contract #DE-AC05-84ER40150 and a National Science Foundation Young Investigator Award. |
warning/0001/physics0001039.html | ar5iv | text | # Untitled Document
Localized Superluminal Solutions to Maxwell Equations propagating along a normal-sized waveguide$`^{()}`$ footnotetext: $`^{()}`$ Work partially supported by CAPES (Brazil), and by INFN, MURST and CNR (Italy).
Michel Zamboni Rached
Dep.to de Fโisica, Universidade Estadual de Campinas, SP, Brazil.
and
Erasmo Recami
Facoltร di Ingegneria, Universitร Statale di Bergamo, Dalmine (BG), Italy;
INFNโSezione di Milano, Milan, Italy; and
DMOโFEEC and CCS, State University of Campinas, Campinas, S.P., Brazil.
Abstract โ We show that localized (non-evanescent) solutions to Maxwell equations exist, which propagate without distortion along normal waveguides with Superluminal speed.
PACS nos.: 03.50.De ; 41.20.Jb ; 03.30.+p ; 03.40.Kf ; 14.80.-j .
Keywords: Wave-guides; Localized solutions to Maxwell equations; Superluminal waves; Bessel beams; Limited-dispersion beams; Electromagnetic wavelets; X-shaped waves; Evanescent waves; Electromagnetism; Microwaves; Optics; Classical physics; General physics; Special relativity
1. โ Introduction: Localized solutions to the wave equations
Since 1915 Bateman showed that Maxwell equations admit (besides of the ordinary planewave solutions, endowed in vacuum with speed $`c`$) of wavelet-type solutions, endowed in vacuum with group-velocities $`0vc`$. But Batemanโs work went practically unnoticed. Only few authors, as Barut et al. followed such a research line; incidentally, Barut et al. constructed even a wavelet-type solution travelling with Superluminal group-velocity $`v>c`$.
In recent times, however, many authors discussed the fact that all (homogeneous) wave equations admit solutions with $`0<v<\mathrm{}`$: see, e.g., Donnelly & Ziolkowski, Esposito, Vaz & Rodrigues. Most of those authors confined themselves to investigate (sub- or Super-luminal) localized non-dispersive solutions in vacuum: namely, those solutions that were called โundistorted progressive wavesโ by Courant & Hilbert. Among localized solutions, the most interesting appeared to be the so-called โX-shapedโ waves, which โpredicted even by Special Relativity in its extended versionโ had been mathematically constructed by Lu & Greenleaf for acoustic waves, and by Ziolkowski et al., and later Recami, for electromagnetism.
Let us recall that such โX-shapedโ localized solutions are Superluminal (i.e., travel with $`v>c`$ in the vacuum) in the electromagnetic case; and are โsuper-sonicโ (i.e., travel with a speed larger than the sound-speed in the medium) in the acoustic case. The first authors to produce X-shaped waves experimentally were Lu & Greenleaf for acoustics, and Saari et al. for optics.
Notwithstanding all that work, still it is not yet well understood what solutions (let us now confine ourselves to Maxwell equations and to electromagnetic waves) have to enter into the play in many experiments.
2. โ About evanescent waves
Most of the experimental results, actually, did not refer to the abovementioned localized, sub- or Super-luminal, solutions, which in vacuum are expected to propagate rigidly (or almost rigidly, when suitably truncated). The experiments most after fashion are, on the contrary, those measuring the group-velocity of evanescent waves\[cf., e.g., refs.11,12\]. In fact, both Quantum Mechanics and Special Relativity had predicted tunnelling wavepackets (tunnelling photons too) and/or evanescent waves to be Superluminal.
For instance, experiments with evanescent waves travelling down an undersized waveguide revealed that evanescent modes are endowed with Superluminal group-velocities.
A problem arises in connection with the experiment with two โbarriersโ 1 and 2 (i.e., segments of undersized waveguide). In fact, it has been found that for suitable frequency bands the wave coming out from barrier 1 goes on with practically infinite speed, crossing the intermediate normal-sized waveguide 3 in zero time. Even if this can be theoretically understood by looking at the relevant transfer function (see the computer simulations, based on Maxwell equations only, in refs.), it is natural to wonder what are the solutions of Maxwell equations that can travel with Superluminal speed in a normal waveguide (where one normally meets ordinary propagating โand not evanescentโ modes)โฆ
Namely, the dispersion relation in undersized guides is $`\omega ^2k^2=\mathrm{\Omega }^2`$, so that the standard formula $`v\mathrm{d}\omega /\mathrm{d}k`$ yields a $`v>c`$ group-velocity. However, in normal guides the dispersion relation becomes $`\omega ^2k^2=+\mathrm{\Omega }^2`$, so that the same formula yields values $`v<c`$ only.
We are going to show that actually localized solutions to Maxwell equations propagating with $`v>c`$ do exist even in normal waveguides; but their group-velocity $`v`$ cannot be given<sup>#1</sup> by the approximate footnotetext: <sup>#1</sup> Let us recall that the group-velocity is well defined only when the pulse has a clear bump in space; but it can be calculated by the approximate, elementary relation $`v\mathrm{d}\omega /\mathrm{d}k`$ only when some extra conditions are satisfied (namely, when $`\omega `$ as a function of $`k`$ is also clearly bumped). formula $`v\mathrm{d}\omega /\mathrm{d}k`$. One of the main motivations of the present note is just contributing to the clarification of this question.
3. โ About some localized solutions to Maxwell equations.
Let us start by considering localized solutions to Maxwell equations in vacuum. A theorem by Lu et al. showed how to start from a solution holding in the plane $`(x,y)`$ for constructing a threedimensional solution rigidly moving along the $`z`$-axis with Superluminal velocity $`v`$. Namely, let us assume that $`\psi (๐;t)`$, with $`๐(x,y)`$, is a solution of the 2-dimensional homogeneous wave equation:
$`\left(_x^2+_y^2\frac{1}{c^2}_t^2\right)\psi (๐;t)=0.`$ (1)
By applying the transformation $`๐๐\mathrm{sin}\theta `$; $`tt(\mathrm{cos}\theta /c)z`$, the angle $`\theta `$ being fixed, with $`0<\theta <\pi /2`$, one gets that $`\psi (๐\mathrm{sin}\theta ;t(\mathrm{cos}\theta /c)z)`$ is a solution to the threedimensional homogeneous wave-equation
$`\left(\mathbf{}^2\frac{1}{c^2}_t^2\right)\psi (๐\mathrm{sin}\theta ;t\frac{\mathrm{cos}\theta }{c}z)=0.`$ (2)
The mentioned theorem holds for the vacuum case, and in general is not valid when introducing boundary conditions. However we discovered that, in the case of a bidimensional solution $`\psi `$ valid on a circular domain of the $`(x,y)`$ plane, such that $`\psi =0`$ for $`|๐|=0`$, the transformation above leads us to a (three-dimensional) localized solution rigidly travelling with Superluminal speed $`v=c/\mathrm{cos}\theta `$ inside a cylindrical waveguide; even if the waveguide radius $`r`$ will be no longer $`a`$, but $`r=a/\mathrm{sin}\theta >a`$. We can therefore obtain an undistorted Superluminal solution propagating down cylindrical (metallic) waveguides for each (2-dimensional) solution valid on a circular domain. Let us recall that, as well-known, any solution to the scalar wave equation corresponds to solutions of the (vectorial) Maxwell equations (cf., e.g., ref. and refs. therein).
For simplicity, let us put the origin O at the center of the circular domain $`๐`$, and choose a 2-dimensional solution that be axially symmetric $`\psi (\rho ;t)`$, with $`\rho =|๐|`$, and with the initial conditions $`\psi (\rho ;t=0)=\varphi (\rho )`$, and $`\psi /t=\xi (\rho )`$ at $`t=0`$.
Notice that, because of the transformations
$`\rho \rho \mathrm{sin}\theta `$ (3a)
$`tt{\displaystyle \frac{\mathrm{cos}\theta }{c}}z,`$ (3b)
the more the initial $`\psi (\rho ;t)`$ is localized at $`t=0`$, the more the (threedimensional) wave $`\psi (\rho \mathrm{sin}\theta ;t(\mathrm{cos}\theta /c)z`$ will be localized around $`z=vt`$. It should be also emphasized that, because of transformation (3b), the velocity $`c`$ goes into the velocity $`v=c/cos\theta >c`$.
Let us start with the formal choice
$`\varphi (\rho )={\displaystyle \frac{\delta (\rho )}{\rho }};\xi (\rho )0.`$ (4)
In cylindrical coordinates the wave equation (1) becomes
$`\left({\displaystyle \frac{1}{\rho }}_\rho \rho _\rho {\displaystyle \frac{1}{c^2}}_t^2\right)\psi (\rho ;t)=0,`$ (1โ)
which exhibits the assumed axial symmetry. Looking for factorized solutions of the type $`\psi (\rho ;t)=R(\rho )T(t)`$, one gets the equations $`_t^2T=\omega ^2T`$ and $`(\rho ^1_\rho +_\rho ^2+\omega ^2/c^2)R=0`$, where the โseparation constantโ $`\omega `$ is a real parameter, which yield the solutions
$`T=A\mathrm{cos}\omega t+B\mathrm{sin}\omega t`$
(5)
$`R=CJ_0({\displaystyle \frac{\omega }{c}}\rho ),`$
where quantities $`A,B,C`$ are real constants, and $`J_0`$ is the ordinary zero-order Bessel function (we disregarded the analogous solution $`Y_0(\omega \rho /c)`$ since it diverges for $`\rho =0`$). Finally, by imposing the boundary condition $`\psi =0`$ at $`\rho =a`$, one arrives at the base solutions
$`\psi (\rho ;t)=J_0({\displaystyle \frac{k_n}{a}}\rho )\left(A_n\mathrm{cos}\omega _nt+B_n\mathrm{sin}\omega _nt\right);k{\displaystyle \frac{\omega }{c}}a,`$ (6)
the roots of the Bessel function being
$$k_n=\frac{\omega _na}{c}.$$
The general solution for our bidimensional problem (with our boundary conditions) will therefore be the Fourier-type series
$`\mathrm{\Psi }_{2\mathrm{D}}(\rho ;t)=_{n=1}^{\mathrm{}}J_0({\displaystyle \frac{k_n}{a}}\rho )\left(A_n\mathrm{cos}\omega _nt+B_n\mathrm{sin}\omega _nt\right).`$ (7)
The initial conditions (4) imply that $`A_nJ_0(k_n\rho /a)=\delta (\rho )/\rho `$, and $`B_nJ_0(k_n\rho /a)=\mathrm{\hspace{0.33em}0}`$, so that all $`B_n`$ must vanish, while $`A_n=\mathrm{\hspace{0.33em}2}[a^2J_1^2(k_n)]^1`$; and eventually one gets:
$`\mathrm{\Psi }_{2\mathrm{D}}(\rho ;t)=_{n=1}^{\mathrm{}}\left({\displaystyle \frac{2}{a^2J_1^2(k_n)}}\right)J_0({\displaystyle \frac{k_n}{a}}\rho )\mathrm{cos}\omega _nt.`$ (8) ,
where $`\omega _n=k_nc/a`$.
Let us explicitly notice that we can pass from such a formal solution to more physical ones, just by considering a finite number $`N`$ of terms. In fact, each partial expansion will satisfy (besides the boundary condition) the second initial condition $`_t\psi =0`$ for $`t=0`$, while the first initial condition gets the form $`\varphi (\rho )=f(\rho )`$, where $`f(\rho )`$ will be a (well) localized function, but no longer a delta-type function. Actually, the โlocalizationโ of $`\varphi (\rho )`$ increases with increasing $`N`$. We shall come back to this point below.
4. โ Localized waves propagating Superluminally down (normal-sized) waveguides.
We have now to apply transformations (3) to solution (8), in order to pass to threedimensional waves propagating along a cylindrical (metallic) waveguide with radius $`r=a/\mathrm{sin}\theta `$. We obtain that Maxwell equations admit in such a case the solutions
$`\mathrm{\Psi }_{3\mathrm{D}}(\rho ,z;t)=_{n=1}^{\mathrm{}}\left({\displaystyle \frac{2}{a^2J_1^2(k_n)}}\right)J_0({\displaystyle \frac{k_n}{a}}\rho \mathrm{sin}\theta )\mathrm{cos}\left[{\displaystyle \frac{k_n\mathrm{cos}\theta }{a}}(z{\displaystyle \frac{c}{\mathrm{cos}\theta }}t)\right]`$ (9)
where $`\omega _n=k_nc/a`$, which are sums over different propagating modes.
Such solutions propagate, down the waveguide, rigidly with Superluminal velocity<sup>#2</sup> $`v=c/\mathrm{cos}\theta `$. Therefore, (non-evanescent) footnotetext: <sup>#2</sup> Let us stress that each eq.(9) represents a multimodal (but localized) propagation, as if the geometric dispersion compensated for the multimodal dispersion. solutions to Maxwell equations exist, that are waves propagating undistorted along normal waveguides with Superluminal speed (even if in normal-sized waveguides the dispersion relation for each mode, i.e. for each term of the Fourier-Bessel expansion, is the ordinary โsubluminalโ one, $`\omega ^2/c^2k_z^2=+\mathrm{\Omega }^2)`$.
It is interesting that our Superluminal solutions travel rigidly down the waveguide: this is at variance with what happens for truncated (Superluminal) solutions\[7-10\], which travel almost rigidly only along their finite โfield depthโ and then abruptly decay.
Finally, let us consider a finite number of terms in eq.(8), at $`t=0`$. We made a few numerical evaluations: let us consider the results for $`N=22`$ (however, similar results can be already obtained, e.g., for $`N=10`$). The first initial condition of eq.(4), then, is no longer a delta function, but results to be the (bumped) bidimensional wave represented in Fig.1.
The threedimensional wave, eq.(9), corresponding to it, i.e., with the same finite number $`N=22`$ of terms, is depicted in Fig.2. It is still an exact solution of the wave equation, for a metallic (normal-sized) waveguide with radius $`r=a/\mathrm{sin}\theta `$, propagating rigidly with Superluminal group-velocity $`v=c/\mathrm{cos}\theta `$; moreover, it is now a physical solution. In Fig.2 one can see its central portion, while in Fig.3 it is shown the space profile along $`z`$, for $`t=\mathrm{const}.`$, of such a propagating wave.
Acknowledgements โ The authors are grateful to Flavio Fontana (Pirelli Cavi, Italy) for having suggested the problem, and to Hugo E. Hernรกndez-Figueroa (Fac. of Electric Engineering, UNICAMP) and Amr Shaarawi (Cairo University) for continuous scientific collaboration. Thanks are also due to Antรดnio Chaves Maia Neto for his kind help in the numerical evaluations, and to Franco Bassani, Carlo Becchi, Rodolfo Bonifacio, Ray Chiao, Gianni Degli Antoni, Roberto Garavaglia, Gershon Kurizki, Giuseppe Marchesini, Marcello Pignanelli, Andrea Salanti, Abraham Steinberg and Jacobus Swart for stimulating discussions.
Figure Captions
Fig.1 โ Shape of the bidimensional solution of the wave equation valid on the circular domain $`\rho a;a=0.1\mathrm{mm}`$ of the $`(x,y)`$ plane, for $`t=0`$, corresponding to the sum of $`N=22`$ terms in the expansion (8). It is no longer a delta function, but it is still very well peaked. By choosing it as the initial condition, instead of the first one of eqs.(4), one gets the threedimensional wave depicted in Figs.2 and 3. The normalization condition is such that $`|\mathrm{\Psi }_{2\mathrm{D}}(\rho =0;t=0)|^2=\mathrm{\hspace{0.33em}1}`$.
Fig.2 โ The (very well localized) threedimensional wave corresponding to the initial, bidimensional choice in Fig.1. It propagates rigidly (along the normal-sized circular waveguide with radius $`r=a/\mathrm{sin}\theta `$) with Superluminal speed $`v=c/\mathrm{cos}\theta `$. Quantity $`\eta `$ is defined as $`\eta (z\frac{c}{\mathrm{cos}\theta }t)`$. The normalization condition is such that $`|\mathrm{\Psi }_{3\mathrm{D}}(\rho =0;\eta =0)|^2=\mathrm{\hspace{0.33em}1}`$.
Fig.3 โ The shape along $`z`$, at $`t=0`$, of the threedimensional wave whose main peak is shown in Fig.2.
References
H.Bateman: Electrical and Optical Wave Motion (Cambridge Univ.Press; Cambridge, 1915).
A.O.Barut and H.C.Chandola: Phys. Lett. A180 (1993) 5. See also A.O.Barut: Phys. Lett. A189 (1994) 277, and A.O.Barut et al.: refs..
A.O.Barut and A.Grant: Found. Phys. Lett. 3 (1990) 303; A.O.Barut and A.J.Bracken: Found. Phys. 22 (1992) 1267. See also refs. below.
R.Donnelly and R.W.Ziolkowski: Proc. Royal Soc. London A440 (1993) 541 \[cf. also I.M.Besieris, A.M.Shaarawi and R.W.Ziolkowski: J. Math. Phys. 30 (1989) 1254\]; S.Esposito: Phys. Lett A225 (1997) 203; W.A.Rodrigues Jr. and J.Vaz Jr., Adv. Appl. Cliff. Alg. S-7 (1997) 457.
See, e.g., E.Recami: โClassical tachyons and possible applications,โ Rivista Nuovo Cimento 9 (1986), issue no.6, pp.1-178; and refs. therein.
Jian-yu Lu and J.F.Greenleaf: IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control 39 (1992) 19.
R.W.Ziolkowski, I.M.Besieris and A.M.Shaarawi: J. Opt. Soc. Am. A10 (1993) 75.
E.Recami: โOn localized โX-shapedโ Superluminal solutions to Maxwell equationsโ, Physica A 252 (1998) 586.
Jian-yu Lu and J.F.Greenleaf: IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control 39 (1992) 441.
P.Saari and K.Reivelt: โEvidence of X-shaped propagation-invariant localized light wavesโ, Phys. Rev. Lett. 79 (1997) 4135. See also H.Sรตnajalg, M.Rรคtsep and P.Saari: Opt. Lett. 22 (1997) 310; Laser Phys. 7 (1997) 32).
A.M.Steinberg, P.G.Kwiat and R.Y.Chiao: Phys. Rev. Lett. 71 (1993) 708, and refs. therein; Scient. Am. 269 (1993) issue no.2, p.38. Cf. also R.Y.Chiao, A.E.Kozhekin, G.Kurizki: Phys. Rev. Lett. 77 (1996) 1254; Phys. Rev. A53 (1996) 586.
A.Enders and G.Nimtz: J. de Physique-I 2 (1992) 1693; 3 (1993) 1089; 4 (1994) 1; H.M.Brodowsky, W.Heitmann and G.Nimtz: J. de Physique-I 4 (1994) 565; Phys. Lett. A222 (1996) 125; Phys. Lett. A196 (1994) 154; G.Nimtz and W.Heitmann: Prog. Quant. Electr. 21 (1997) 81.
See V.S.Olkhovsky and E.Recami: Phys. Reports 214 (1992) 339, and refs. therein; V.S.Olkhovsky et al.: J. de Physique-I 5 (1995) 1351-1365; T.E.Hartman: J. Appl. Phys. 33 (1962) 3427.
Cf. A.P.L.Barbero, H.E.Hernรกndez-Figueroa and E.Recami: โOn the propagation speed of evanescent modesโ \[LANL Archives # physics/9811001\], submitted for pub., and refs. therein. Cf. also E.Recami, H.E.Hernรกndez F., and A.P.L.Barbero: Ann. der Phys. 7 (1998) 764.
G.Nimtz, A.Enders and H.Spieker: J. de Physique-I 4 (1994) 565; โPhotonic tunnelling experiments: Superluminal tunnellingโ, in Wave and Particle in Light and Matter โ Proceedings of the Trani Workshop, Italy, Sept.1992, ed. by A.van der Merwe and A.Garuccio (Plenum; New York, 1993).
H.M.Brodowsky, W.Heitmann and G.Nimtz: Phys. Lett. A222 (1996) 125.
R.Garavaglia: Thesis work (Dip. Sc. Informazione, Universitร statale di Milano; Milan, 1998; G.Degli Antoni and E.Recami supervisors).
E.Recami and F.Fontana: โSpecial Relativity and Superluminal motionsโ, submitted for publication.
J.-y.Lu, H.-h.Zou and J.F.Greenleaf: IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency Control 42 (1995) 850-853. |
warning/0001/cond-mat0001030.html | ar5iv | text | # Relaxation processes in harmonic glasses?
## Abstract
A relaxation process, with the associated phenomenology of sound attenuation and sound velocity dispersion, is found in a simulated harmonic Lennard-Jones glass. We propose to identify this process with the so called microscopic (or instantaneous) relaxation process observed in real glasses and supercooled liquids. A model based on the memory function approach accounts for the observation, and allows to relate to each others: 1) the characteristic time and strength of this process, 2) the low frequency limit of the dynamic structure factor of the glass, and 3) the high frequency sound attenuation coefficient, with its observed quadratic dependence on the momentum transfer.
In a recent paper W. Gรถtze and M.R. Mayr adapted the Mode Coupling Theory to glassy phase. This theory, as shown numerically for a hard-spheres model, accounts for many of the features found in the dynamic structure factor, $`S(Q,\omega )`$, of glasses. Among these: i) The existence of propagating excitations (sound waves), with an almost linear momentum transfer ($`Q`$) dependence of their excitation energy $`\mathrm{\Omega }(Q)`$, up to $`Q`$ values that are a significant fraction of $`Q_o`$, the maximum of the static structure factor $`S(Q)`$. ii) The quadratic dependence on $`Q`$ of the excitations broadening $`\mathrm{\Gamma }(Q)`$, and therefore of sound attenuation. iii) The temperature insensitivity of $`\mathrm{\Gamma }(Q)`$. iv) The development in the $`S(Q,\omega )`$ at large $`Q`$ ($`Q/Q_o`$$``$0.3) of a secondary excitation band, at frequencies below the Brillouin peak. The theory also predicts two features not yet experimentally detected: a) A positive dispersion of the sound velocity and b) an intensity โgapโ in the low frequency region of the $`S(Q,\omega )`$. In spite of the success of this approach, it is still of great interest to investigate the physical origin of these phenomena, and, in particular, whether they are related to the topological disorder and/or to the anharmonicity of the interatomic potential.
In this Letter we report a Molecular Dynamics (MD) simulation study of the $`Q`$ dependence of the sound velocity in a model monatomic Lennard-Jones glass in the harmonic approximation. We show that even in a harmonic glass, by increasing $`Q`$ there is a positive dispersion of the sound velocity, thus proving one of the prediction of the MCT theory , and relating this phenomenon to the topological disorder. Since this dispersion is similar to that found in presence of a relaxation process, we attempt to apply a generalized Langevin equation with an effective memory function approach to describe the density fluctuations dynamics. This formalism allows to account for the ubiquitous $`Q^2`$-dependence of the high frequency sound absorption observed in many glasses by experimenys and by MD simulations . These results suggest to identify this process with the one referred to as microscopic or instantaneous relaxation in real systems.
The investigated systems consist of $`N`$=2048, 10976 and 32000 argon atoms interacting via a (6-12) Lennard-Jones potential ($`ฯต`$=125.2 K, $`\sigma `$=3.405 A). A standard microcanonical MD simulation, performed at decreasing temperatures in the normal liquid phase, is followed by a fast quench ($`\dot{T}`$$``$ 10<sup>12</sup> K/s) of the slightly supercooled liquid down to $``$5 K . Starting from the glass configuration at $`T`$=5 K, the atomic trajectories $`\overline{r}_i(t)`$ (here $`i`$=$`1\mathrm{}N`$ is the particle label) are followed and stored for subsequent analysis, and the โinherentโ configuration $`\{\overline{x}_i\}_{i=1..N}`$ at $`T`$=0 K is calculated by the steepest descent method. Two different procedures have been used to derive the $`S(Q,\omega )`$: (i) The trajectories calculated from MD in the glass are used to compute the corresponding time correlation function and (ii) The Normal Mode Analysis (NMA) is applied to the inherent configuration. This last procedure works in the harmonic approximation, which is obtained by retaining only the quadratic term of the interaction potential. From the calculation of the dynamical matrix $`๐`$, one computes the eigenvalues, $`\omega _p`$ ($`p`$=$`\mathrm{1..3}N`$ is the mode index) and the eigenvectors ($`\overline{e}_p(i)`$). For large size samples (where a direct diagonalization of $`๐`$ is not feasible) the $`S(Q,\omega )`$ is obtained by the method of moments .
In the MD runs, the dynamics structure factor is calculated from the numerical time-Fourier transform of the intermediate scattering function $`F(Q,t)`$, defined as:
$$F(Q,t)=1/N\mathrm{\Sigma }_{ij}\mathrm{exp}(i\overline{Q}\overline{r}_i(t))\mathrm{exp}(i\overline{Q}\overline{r}_j(0)).$$
(1)
In the harmonic framework and in the classical limit, the one excitation approximation of the dynamic structure factor $`S^{^{(1)}}(Q,\omega )`$ is obtained by the normal mode expansion of the atomic displacements:
$$\overline{r}_i(t)=\overline{x}_i+\sqrt{K_BT/M}\mathrm{\Sigma }_p\overline{e}_p(i)A_p(t)/\sqrt{\omega _p}.$$
(2)
$`A_p(t)`$ is the amplitude of the $`p`$-th normal mode and is characterized by $`|A_p(t)|^2=1`$. This gives:
$$S^{^{(1)}}(Q,\omega )=(K_BTQ^2/M\omega ^2)\mathrm{\Sigma }_pE_p(Q)\delta (\omega \omega _p),$$
(3)
where we have introduced the spatial power spectrum of the longitudinal component of the eigenvectors, $`E_p(Q)`$:
$$E_p(Q)=|\mathrm{\Sigma }_i(\widehat{Q}\overline{e}_p(i))\mathrm{exp}(i\overline{Q}\overline{x}_i)|^2.$$
(4)
Here, $`\widehat{Q}`$=$`\overline{Q}/|Q|`$ and the Debye-Waller factor has been neglected. Differently from the second ($`\omega ^2S(Q,\omega )๐\omega `$= $`\omega ^2S^{^{(1)}}(Q,\omega )๐\omega `$=$`K_BTQ^2/M`$) and higher moments sum rules, the zeroth moment sum rule for $`S(Q,\omega )`$, $`S(Q,\omega )๐\omega `$=$`S(Q)`$, does not hold for $`S^{^{(1)}}(Q,\omega )`$, as in this function the elastic intensity is missing. Rather, it is useful to define the โinelasticโ contribution to $`S(Q)`$: $`S__i(Q)`$=$`S^{^{(1)}}(Q,\omega )๐\omega `$.
Selected examples of the $`S(Q,\omega )`$ calculated for different size systems, and with different methods are reported in Fig. 1. As there is a trivial dependence of the $`S^{^{(1)}}(Q,\omega )`$ on $`T`$, here we report this quantity multiplied by the factor $`(M/K_BT)`$. Figure 1a shows the $`S(Q,\omega )`$ in the low $`Q`$ range, as calculated from the MD runs for the $`N`$=32000 particles system, while Fig. 1b shows the intermediate $`Q`$ range, in the case of the harmonic approximation for the $`N`$=2048 system. As a check of consistency, the inset of Fig. 1b shows the $`Q`$ dependence of the second moment of $`(M/K_BT)S(Q,\omega )`$ (+) which, according to the sum rule, should be $`Q^2`$ (full line). The inset of Fig. 1a shows, at two $`Q`$ values, the comparison of the $`S(Q,\omega )`$ ($`N`$=2048) calculated either from the MD runs (full line) or in the harmonic approximation ($``$). The $`S(Q,\omega )`$, as derived with the two different methods, are equivalent in the whole considered $`Q`$ range. This indicates that the Newtonian dynamics at $`T`$=5 K is truly harmonic, and that the results obtained by MD at this $`T`$ and NMA can be interchanged among each other.
The general features of the $`S(Q,\omega )`$ reported in Fig. 1 are: i) A Brillouin peak, dispersing and becoming broader with increasing $`Q`$, dominates the spectrum up to $`Q`$$``$10 nm<sup>-1</sup> ($`Q/Q_o`$$``$0.45). ii) At larger $`Q`$ values a second peak, observed at frequencies below the Brillouin peak, starts to dominate the data. This secondary peak has been already detected in many systems: 1) in liquid water, both experimentally and by MD , where it has been interpreted as a signature of the transverse dynamics; 2) in vitreous silica, by MD, where it has been interpreted either in terms of a transverse dynamics or as an evidence of the Boson peak ; and 3) in a hard sphere glass, where it has been theoretically predicted . In this respect, it is worth to underline the striking similarity between the $`S(Q,\omega )`$ reported here and those of Fig. 5 of Ref. . Finally, iii), the $`S(Q,\omega )`$ up to $`Q`$$``$5 nm<sup>-1</sup> shows a nearly constant and $`Q`$-independent intensity below the Brillouin peak frequency. This plateau, whose value is $`lim_{\omega 0}(M/K_BT)S(Q,\omega )`$= $`A_0`$$``$0.03 10<sup>-18</sup> s<sup>3</sup>/m<sup>2</sup>, can be identified with the plateau observed in the constant $`\omega `$ cuts of the $`S(Q,\omega )`$ at $`Q`$ larger than the Brillouin peak in simulated , calculated (Fig 18 in Ref. ) and measured systems.
Figure 2 shows the position of the maxima ($`\mathrm{\Omega }__C(Q)`$) of the current spectra $`C_L(Q,\omega )`$ (=$`\omega ^2/Q^2S(Q,\omega )`$) as a function of $`Q`$ in the investigated $`Q`$ range. The inset of Fig. 2 shows the peak position $`\mathrm{\Omega }(Q)`$ and the broadening $`\mathrm{\Gamma }(Q)`$ of the $`S(Q,\omega )`$ in the low $`Q`$ region, as measured directly from the spectra of Fig. 1a. The $`Q`$ dependence of these parameters agrees with the behavior observed in all the glasses investigated so far : a linear (square) dependece of $`\mathrm{\Omega }(Q)`$ ($`\mathrm{\Gamma }(Q)`$). The behavior of $`\mathrm{\Omega }__C(Q)`$ closely resembles that of the acoustic phonon branches in crystals, i. e. the almost linear behavior in the small $`Q`$ region, a maximum around $`Q_o/2`$, and a minimum around $`Q_o`$. Most importantly one can clearly detect a positive dispersion of the sound velocity, as highlighted by the dashed line in Fig. 2. This dispersion is better seen in Fig. 3, where the apparent sound velocity, $`v(Q)`$=$`\mathrm{\Omega }__C(Q)/Q`$, is reported. The quantity $`v(Q)`$ (full dots) undergoes a transition from the โlow frequencyโ sound velocity $`v_o(Q)`$ towards the infinite frequency sound velocity $`v_{\mathrm{}}(Q)`$. Here $`v_o(Q)`$ is calculated as $`v_o(Q)`$=$`\sqrt{K_BT/MS_i(Q)}`$ (dashed line), and $`v_{\mathrm{}}(Q)`$ is calculated either as the fourth moment of the calculated $`S(Q,\omega )`$ (open points) or from the expression of the fourth moment in terms of both the pair correlation function and the interaction potential (full line) . The velocity dispersion is observed up to $`Q`$$``$5 nm<sup>-1</sup>, i. e. $`\mathrm{\Omega }__C`$$``$35 cm<sup>-1</sup>, and an opposite dispersion is observed in the region approaching $`Q_o`$. These observations recall of a typicall relaxation scenario. When a relaxation process with characteristic time $`\tau `$ is active in the system, the transition from $`v_o`$ to $`v_{\mathrm{}}`$ takes place when the condition $`\omega \tau `$=1 is fulfilled. In the present case, considering that the first transition is at at $`Q`$$``$2 nm<sup>-1</sup> where $`\mathrm{\Omega }__C`$$``$15 cm<sup>-1</sup>, the value of $`\tau `$ results to be around 0.3 ps. The second and third transitions between $`v_o`$ and $`v_{\mathrm{}}`$ are observed just below and above $`Q_o`$ as a consequence of the slowing down of the dynamics (deGennes narrowing) around $`Q_o`$. Consequently the whole behavior of $`v(Q)`$, as reported in Fig. 3, can be qualitatively understood in terms of a relaxation process with a characteristic time of $`\tau `$$``$0.3 ps.
The identification of a such a relaxation process in an harmonic system, suggests to use the formalism that describes the density correlators $`\varphi (Q,t)`$=$`F(Q,t)/S(Q)`$ through its generalized Langevin equation :
$$\ddot{\varphi }(Q,t)+\omega _o^2\varphi (Q,t)+_o^tm(Q,tt^{})\dot{\varphi }(Q,t^{})๐t=0$$
(5)
where $`\omega _o^2`$=$`K_BTQ^2/MS(Q)`$ and $`m(Q,t)`$ is the โmemory functionโ. This equation has been rigorously derived for ergodic systems, but, as recently shown in the framework of the MCT , it can be still applied in the non-ergodic glassy phase making the the following substitutions: $`\varphi (Q,t)\varphi ^{}(Q,t)`$=$`(\varphi (Q,t)f__Q)/(1f__Q)`$, $`\omega _o^2(Q)\omega _{o\mu }^2`$ $``$$`\omega _\mathrm{}\alpha ^2`$=$`K_BTQ^2/MS_i(Q)`$ and $`m(Q,t)m_\mu (Q,t)`$=$`m(Q,t)\omega _o^2f__Q/(1f__Q)`$. Therefore, the โvibrationalโ dynamics of interest is now described by the correlators $`\varphi ^{}(Q,t)`$, which is obtained subtracting from $`\varphi (Q,t)`$ the long time plateau level, whose value is the non-ergodicity parameter $`f__Q`$. It is worth to note the subtraction of the constant term $`S(Q)f__Q`$ from F(Q,t) is equivalent to neglect the elastic contribution in the $`S(Q,\omega )`$, and that $`F^{}(Q,t)`$=$`S_i(Q)\varphi ^{}(Q,t)`$ is the Fourier transform of the $`S^{^{(1)}}(Q,\omega )`$ as defined in Eq. (3). The whole dynamic behavior of $`\varphi ^{}(Q,t)`$ is now contained in the microscopic contribution to the memory function $`m_\mu (Q,t)`$. In the case of a harmonic system, the function $`m_\mu (Q,t)`$ can be explicitly calculated from the eigenvalues and eigenvectors of the system. Indeed, the Laplace transform (indicated by hats) of Eq. (5) (after the previously indicated substitutions) and a straightforward algebra gives:
$$\widehat{m}_\mu (Q,s)=[\widehat{\varphi }^{}(Q,s)[s^2+\omega _{o\mu }^2]s]\left[1s\widehat{\varphi }^{}(Q,s)\right]^1.$$
(6)
Then, from an inverse Fourier and a subsequent Laplace transform of Eq. (3), it is easy to get an explicit expression for $`\widehat{\varphi }^{}(Q,s)`$ to be inserted in Eq. (6). This gives:
$$\widehat{m}_\mu (Q,s)=[\mathrm{\Sigma }\frac{E_p(Q)}{\omega _p^2}\frac{s}{s^2+\omega _p^2}]\left[\mathrm{\Sigma }\frac{E_p(Q)}{\omega _p^2}\mathrm{\Sigma }\frac{E_p(Q)}{s^2+\omega _p^2}\right]^1s$$
(7)
This equation provides an explicit expression of the memory function in terms of the system eigenstates. Considering that $`m_\mu (Q,t)`$ is mainly characterized by parameters as its initial value $`\mathrm{\Delta }__Q^2`$ and total area $`\mathrm{\Gamma }__Q`$ and therefore by a decaying time-scale $`\tau __Q`$$``$$`\mathrm{\Gamma }__Q/\mathrm{\Delta }__Q^2`$ , one can show that $`\widehat{m}(Q,s`$$``$$`\mathrm{})`$=$`\mathrm{\Delta }__Q^2/s`$ and $`\widehat{m}(Q,s`$$``$0)=$`\mathrm{\Gamma }__Q`$ . Inserting these limiting values into Eq. (7) one obtains:
$`\mathrm{\Delta }__Q^2`$ $`=`$ $`\left[\mathrm{\Sigma }_pE_p(Q)\omega _p^2\right]\left[\mathrm{\Sigma }_pE_p(Q)\omega _p^2\right]^1`$ (8)
$`\mathrm{\Gamma }__Q`$ $`=`$ $`\left[\mathrm{\Sigma }_p{\displaystyle \frac{E_p(Q)}{\omega _p^2}}\right]^2\underset{s0}{lim}\left[\mathrm{\Sigma }_p{\displaystyle \frac{E_p(Q)}{\omega _p^2}}{\displaystyle \frac{s}{s^2+\omega _p^2}}\right]`$ (9)
It is now easy to identify, through the explicit expression of the zeroth and fourth moments of Eq. (3) that $`\mathrm{\Sigma }_pE_p(Q)\omega _p^2`$=$`v_{\mathrm{}}^2Q^2`$ and $`(\mathrm{\Sigma }_pE_p(Q)\omega _p^2)^1`$=$`v_o^2Q^2`$, confirming that $`\mathrm{\Delta }__Q^2`$=$`(v_{\mathrm{}}^2v_o^2)Q^2`$. The determination of $`\mathrm{\Gamma }__Q`$ which, being the area of the memory function, coincides with the Brillouin broadening in the $`\omega \tau `$$`<<`$1 limit is slightly more involved. Using the representation $`\delta (x)=1/\pi lim_{s0}s/(s^2+x^2)`$, in Eq. (3) one sees that:
$$S^{^{(1)}}(Q,\omega )=\frac{K_BTQ^2}{\pi M}\underset{s0}{lim}\mathrm{\Sigma }_p\frac{E_p(Q)}{\omega _p^2}\frac{s}{s^2+(\omega \omega _p)^2},$$
(10)
and comparing of Eqs. (10) and (9) one gets :
$$\mathrm{\Gamma }__Q=v_o^4Q^2\frac{\pi M}{K_BT}\underset{\omega 0}{lim}S(Q,\omega ).$$
(11)
Similarly:
$$\tau __Q\frac{v_o^4}{(v_{\mathrm{}}^2v_o^2)}\frac{\pi M}{K_BT}\underset{\omega 0}{lim}S(Q,\omega ).$$
(12)
Considering that $`lim_{\omega 0}S(Q,\omega )`$ is $`Q`$-independent (see Fig. 1), these expressions give account of the observation in the low $`Q`$ region -where the $`Q`$ dependence of $`v_o`$ and $`v_{\mathrm{}}`$ is negligible- that the Brillouin peak broadening is proportional to $`Q^2`$ and the $`\tau `$ is $`Q`$-independent. As a check of internal consistency, using the value of $`A_o`$ previously reported, one finds $`\tau `$$``$0.6 ps and $`\mathrm{\Gamma }__Q`$\[cm<sup>-1</sup>\]= 1.35$`Q^2`$\[nm<sup>-1</sup>\]. The value of $`\tau __Q`$ overestimates the one deduced from Fig. 3, and this probably due to the rough expression of $`\tau `$ as $`\mathrm{\Gamma }__Q/\mathrm{\Delta }__Q^2`$. On the contrary, as shown in the inset of Fig. 2, $`\mathrm{\Gamma }__Q`$ from Eq. (11) is in excellent agreement with the one directly derived from the width of the $`S(Q,\omega )`$.
It is worth to discuss the microscopic origin of the observed relaxation process. A relaxation process can be pictured as the macroscopic manifestation of microscopic phenomena associated with the existence of channels by which the energy stored in a given โmodeโ relaxes towards other degrees of freedom. The $`S(Q^{},\omega )`$, through the fluctuation-dissipation relation, reflects the time evolution of the energy initially stored ($`t`$=$`t_o`$) in a Plane Wave (PW) of wavelength $`2\pi /Q^{}`$. As the PW is not an eigenstate of the disordered system, at $`t>t_o`$ there will be a transfer of amplitude from this PW towards other PWs of different $`Q`$ values. This process is controlled by the difference bewteen the considered PW and the normal modes of the topologically disordered glassy structure. This energy flow takes place on the time scale $`\tau `$ as derived from Eq. (12), and gives rise to the observed relaxation process phenomenology. Consequently, one can speculate that this process is the instantaneous or microscopic process empirically introduced to explain the $`S(Q,\omega )`$ measured in real glasses and liquids by Brillouin light and x-ray Scattering .
We thanks W. Gรถtze for helpful discussions and critical readings of the manuscript. |
warning/0001/hep-th0001004.html | ar5iv | text | # Moving Mirrors and Thermodynamic Paradoxes
## I Introduction
Almost twentyโfive years ago, Davies and Fulling , following a suggestion of DeWitt (see also ), introduced the โmoving mirrorโ models: massless scalar quantum fields in twoโdimensional Minkowski space responding to perfectly reflecting boundaries. Such models have been of inestimable value in clarifying conceptual issues raised by more complicated theories; most notably, there are connections between moving mirror models and the Hawking process.
There are still aspects of the movingโmirror models which are not satisfactorily understood. The most important of these are the thermodynamic paradoxes, which seem to be consequences of basic features of the models, and so unavoidable in them. Consider for simplicity the case of a nonโrelativistic mirror with position $`q(t)`$. Then, assuming the field was initially in the vacuum state (and that in the far past the mirror was stationary), the expected energy on the right (respectively, left) of the mirror is
$$E_{\genfrac{}{}{0pt}{}{\mathrm{right}}{\mathrm{left}}}=(12\pi )^1(\mathrm{}/c)\ddot{q}+O(1/c^2).$$
(1)
Now we come to the key point. It follows immediately that the total expected energy in the field is
$$E_{\mathrm{left}}+E_{\mathrm{right}}=0+O(1/c^2),$$
(2)
so (to lowest order) no energy is required to move the mirror. This is extraordinary. The leading effect of the mirrorโs motion is to separate the vacuum into packets of positive and negative expected energy, at no energetic cost. If these expectationโvalues can be regarded as classical energies, then we have a direct violation of the second law of thermodynamics. One can easily construct paradoxes based on this, and in fact Davies described a perpetualโmotion machine which turns around this idea . (See also .)
One might raise some objections to specific elements of Daviesโs proposal, and indeed various workers have done so (mainly concentrating on the problems of absorbing the negative energy packets ; see also the earlier paper ). Still, it seems hard to avoid the central point: if one can split the vacuum into positiveโ and negativeโenergy parts, with negligible energetic input, isnโt one violating the second law? Even setting aside possible difficulties associated with managing or absorbing the negativeโenergy packets, couldnโt one drive perpetualโmotion machines by simply keeping the positive energies produced by the mirrorโs motion, leaving the negativeโenergy packets to go their ways?
The aim of the present paper is to resolve this point. I shall show that the approximation that has been made, that the mirrorโs trajectory can be treated classically, is invalid for the purpose of computing the necessary energy transfers. The limitations of validity of the classical model are reached before effects of energyโexchange between it and the quantum field can be computed. Thus, insofar as the mirror can be treated classically, there is no violation of the second law.
Going beyond this classical model, I shall consider a model with a nonโrelativistic, firstโquantized mirror moving in an external potential. At least where the potential is quadratic, it will be shown that the measurement of the energy packets is always accompanied by a much larger spread in the mirrorโs energy. This means that attempts to measure the vacuum field energies cause quantum fluctuations in the mirrorโs state, fluctuations that cannot be ignored for the purposes of understanding the energy transfers between the field and the mirror.
It should be emphasized that the present work indicates that any โsemiclassicalโ attempt to model the quantum backโreaction on the mirror is invalid for the purposes of modeling the energyโtransfers that would occur in attempts to measure the vacuum field energy density. This is because semiclassical approximations (which give the backโreaction of the quantum field on the mirror in terms of expectations) are precisely those which assume that fluctuations in the mirrorโs state are negligible, and this is just what fails here.
The present results fit well with those of a related investigation, by Parentani . He introduced a model with a secondโquantized mirror in a linear external potential. He was able to show (with certain approximations) that the forward quanta would decohere. This is because their states become correlated with that of the mirror. The general lesson to be drawn from the models, then, is that the entanglement of the mirrorโs state with that of the field can be a dominant effect, and the entanglement can involve fluctuations in the mirrorโs energy larger than the fieldโs energy.
One can view the present work as estimating the magnitudes of the effects of quantum fluctuations in the mirrorโs state on the energyโtransfers in the system; the effects are large enough to invalidate the externalโfield approximation. However, to go beyond this negative conclusion, and analyze in detail what does happen in the energy transfers, is another issue. I shall argue below that even this model is probably inadequate for a satisfactory understanding of these issues, and it will be necessary to pass to a theory where the internal degrees of freedom of the mirror (and the scattering of virtual field quanta from these) are accounted for. This is surprising and perhaps disconcerting: one would have thought that a quantum field responding to a slow, heavy mirror could be analyzed without needing to account for the mirrorโs structure as a system of quantum fields. But if one wants to understand the vacuum energies, such an analysis seems necessary.
Although such a sophisticated model will ultimately be necessary, there are good reasons for considering the nonโrelativistic firstโquantized pointโparticle model, at least initially. The most important one is that there is little ambiguity in defining it, whereas to go beyond it requires many choices. (The more sophisticated models require one to make assumptions amounting to a choice of dispersive susceptibility $`\chi (\omega )`$, and there is functional freedom in doing so.) The nonโrelativistic mirror, by contrast, can be a firstโquantized point particle, and there is little ambiguity in how to proceed. Thus the pointโparticle results, while more limited, are at least clearly modelโindependent.
Another reason for starting with the firstโquantized pointโparticle model is that the quantum measurement issues can be analyzed at a fairly elementary level. Finally, the model is perhaps of some interest beyond the present paper. The very fact that it is of limited validity can be turned to advantage, because higher relativistic corrections can be ignored and a great deal of its structure can be worked out explicitly.
In Section II, the DaviesโFulling models are reviewed. This section may be read rapidly, but should not be skipped. No details of the calculations are given, but the physical basis of the renormalization and some of the limitations on the validity of the model are discussed in IIB. These limitations figure essentially in later arguments. Section III briefly derives the firstโquantized mirror model. Section IV gives the main analysis of the measurement of field vacuum energy and its limitations. The last section summarizes the main conclusions.
In most places, particularly in estimates of the magnitudes of energies, factors of $`c`$ and $`\mathrm{}`$ are given explicitly. However, factors of $`c`$ have been omitted in a few places (advanced and retarded coordinates, etc.), where they would make the appearance of the equations unnecessarily complicated.
## II A Classical Moving Mirror
In this section, I shall review the standard treatment of a massless field influenced by a moving mirror in twoโdimensional Minkowski space . No details of standard computations will be given; the emphasis will be on the physical assumptions and consequences.
### A Basic Formalism and Results
Let $`(t,x)`$ be coordinates on twoโdimensional Minkowski space with metric $`ds^2=dt^2dx^2`$. We introduce retarded and advanced null coordinates by $`u=tx`$ and $`v=t+x`$ as usual, and vectors $`l^a=_v`$, $`n^a=_u`$. It is convenient to regard the trajectory of the mirror as given by $`v=V(u)`$ or $`u=U(v)`$. We assume that the trajectory is timelike and is asymoptotically stationary in the past.
We consider a massless scalar field. Any solution to the field equation can be written locally as $`\varphi (u,v)=f(u)+g(v)`$. The mirror is considered to enforce the boundary condition $`\varphi (u,V(u))=0`$. Thus we must have $`f(u)=g(V(u))`$. We shall write
$$\varphi =f(u)f(U(v)),v<V(u)$$
(3)
(to the left of the mirror) and
$$\varphi =g(v)g(V(u)),v>V(u)$$
(4)
(to the right). Thus the symbol $`f`$ will only be used for fields on the left, and $`g`$ only for fields on the right. Then the functions $`f`$ and $`g`$ can be considered data at $`^{}`$ for the field. We may also interpret these equations at the operator level; then $`\widehat{f}`$ and $`\widehat{g}`$ are the โinโ operators.
The stressโenergy operator is
$$\widehat{T}_{ab}=:\widehat{T}_{ab}:+\widehat{T}_{ab}^{\mathrm{ren}},$$
(5)
where the colons stand for normal ordering and $`\widehat{T}_{ab}^{\mathrm{ren}}`$ is the renormalized vacuum expectation value. This last is defined by pointโsplitting. One starts with the formal expression
$$\widehat{T}_{ab}^{\mathrm{formal}}=\left(\delta _a^p\delta _b^q(1/2)g_{ab}g^{pq}\right)_p\widehat{\varphi }|_{(u_1,v_1)}_q\widehat{\varphi }|_{(u_2,v_2)},$$
(6)
and considers the limit as $`(u_2,v_2)(u_1,v_1)`$. The expectation value $`0|\widehat{T}_{ab}^{\mathrm{formal}}|0`$ contains two terms: a divergent one which is independent of position, and a finite term. It is the finite term which is $`\widehat{T}_{ab}^{\mathrm{ren}}`$. The divergent term, present even in Minkowski space, is the โunrenormalized stressโenergy of the Minkowski vacuum.โ The result is
$$\widehat{T}_{ab}^{\mathrm{ren}}=(12\pi )^1\mathrm{}\left(\frac{3}{4}\left(\frac{V^{\prime \prime }}{V^{}}\right)^2\frac{1}{2}\frac{V^{\prime \prime \prime }}{V^{}}\right)l_al_b$$
(7)
on the right. (On the left, one has an expression of the same form, with $`U`$ replacing $`V`$ and $`n_a`$ replacing $`l_a`$.) In the limit of nonโrelativistic motion, with the trajectory given by $`x=q(t)`$, we have
$$\widehat{T}_{ab}^{\mathrm{ren}}=(12\pi )^1(\mathrm{}/c^2)\left((1+\dot{q})_t^3q+3\ddot{q}^2\dot{q}\right)l_al_b+\mathrm{}.$$
(8)
(We have given as many terms as we shall need later.) This is to be evaluated at the time $`t^{}`$ such that $`(tt^{},xq(t^{}))`$ is null futureโpointing.
From these formulas, one can derive expressions for the expected renormalized energy in the field to the left and the right of the mirror:
$`E_{\mathrm{right}}`$ $`=`$ $`(12\pi )^1(\mathrm{}/c^2){\displaystyle _{q(t)}^{\mathrm{}}}\left(_t^3q+3\ddot{q}^2\dot{q}\right)๐x`$ (9)
$`=`$ $`(12\pi )^1(\mathrm{}/c){\displaystyle _{\mathrm{}}^t}\left(_t^3q+3\ddot{q}^2\dot{q}\right)๐t^{}`$ (10)
$`=`$ $`(12\pi )^1(\mathrm{}/c)\ddot{q}+\mathrm{}.`$ (11)
(An integration by parts can be used to justify discarding the second term when passing to the last line.) On the left of the mirror, one has
$$E_{\mathrm{left}}=+(12\pi )^1(\mathrm{}/c)\ddot{q}+\mathrm{}.$$
(12)
Thus, to lowest order, no total expected energy is produced, but the mirrorโs motion effects a separation of the vacuum energy into positive and negative terms. The leading nontrivial contribution to the total expected energy in the field is of order $`\dot{x}`$ (that is, $`v/c`$) smaller; it is
$`E_{\mathrm{total}}`$ $`=`$ $`(6\pi )^1(\mathrm{}/c^2){\displaystyle _{\mathrm{}}^t}_t^3q\dot{q}dt^{}`$ (13)
$`=`$ $`(6\pi )^1(\mathrm{}/c^2)\left[\ddot{q}\dot{q}+{\displaystyle _{\mathrm{}}^t}(\ddot{q})^2๐t^{}\right].`$ (14)
Thus the total energy put into the field must be positive, if the motion is asymptotically inertial.
### B The Renormalization
While all of the foregoing is standard, one must remember that we do not at present have a firstโprinciples understanding of the infinite vacuum energy density and (therefore) of its renormalization. While the standard computation of $`\widehat{T}_{ab}^{\mathrm{ren}}`$ will be accepted here (within a regime of applicability to be discussed shortly), since the interpretation of this quantity is critical to the physics of the mirror, it is appropriate to discuss what has been done carefully. These points are important:
(a) The โoperator partโ of $`\widehat{T}_{ab}`$ โ that is, the operator modulo the addition of cโnumber terms like $`\widehat{T}_{ab}^{\mathrm{ren}}`$ โ is determined by the equation of motion for the fields, and so is unambiguously defined irrespective of the renormalization. In other words, different choices of renormalization can only affect the cโnumber terms.
(b) It is not trivial that the theory is renormalizable. The idealized perfectโreflector nature of the boundary causes a great deal of cancellation of ultraviolet contributions to the stressโenergy in the neighborhood of the mirror. In a more realistic model, one would expect dispersive effects to disturb these cancellations. This would lead to terms which were formally divergent as one approached the mirror (although the theory itself would break down as one approached within a distance of the order of the skin depth of the mirror). Cf. ref. .
A related point is that we have ignored whatever internal physics the mirror has which causes it to reflect. For an actual (electromagnetic) mirror, there are ions and conduction electrons whose contributions to the electromagnetic stressโenergy outside the mirror might not be ignorable.
(c) Consider for the moment replacing the perfectly reflecting mirror by a more realistic model, where one has a mirror with a dispersive susceptibility tending rapidly to zero beyond some cutโoff frequency $`\omega _\mathrm{p}`$. The effect of this would be to introduce a frequencyโdependent potential term into the equation of motion, or equivalently, in coordinate space, a convolution of $`\widehat{\varphi }`$ with the Fourier transform of that potential. This term would act like a perfect reflector on field modes of frequencies $`\omega \omega _\mathrm{p}`$, but the structure of the potential would become important at scales $`\omega \omega _\mathrm{p}`$.
In such a model, the mirror will act like a classical reflector of lowโfrequency modes only as long as the time scale defined by its acceleration is significantly larger than $`1/\omega _\mathrm{p}`$. If the acceleration is greater, we must take into account the mixing of lowโfrequency and highโfrequency modes due to the mirrorโs motion.
What this means for the present paper is that the computation of $`\widehat{T}_{ab}^{\mathrm{ren}}`$ is only credible as long as the inverse time scales over which $`\dot{q}`$ changes are much less that the plasma frequency $`\omega _\mathrm{p}`$ of the mirror. In particular, we must have $`|\ddot{q}|\omega _\mathrm{p}|\dot{q}|`$ or we are not justified in using the standard formula, equation (7), and its consequences, equations (8)โ(14).
(d) The usual procedure is to take the points $`(u_1,v_1)`$ and $`(u_2,v_2)`$ separated by a small imaginary timelike interval. This has the effect of introducing an ultraviolet cutโoff. This is attractive, because one can then argue that the justification of the procedure is that real experiments only probe an object up to a finite frequency. Also this procedure ascribes to Minkowski spaceโtime a (divergent) positive expected energy density, whereas realโseparated points give rise to negative energy densities.
However, this procedure requires one to consider the worldโline $`V(u+i\delta u)`$ at complex points as well, and it is hard to give a physical interpretation of this. If $`V`$ is analytic, of course, one has a clear candidate definition for $`V(u+i\delta u)`$. However, even in this case $`V(u+i\delta u)`$ depends nonโlocally on the real trajectory $`V(u)`$. It is in particular hard to see how to reconcile oneโs notions of causality (being able to change $`V(u)`$ freely in the future of $`u=u_0`$, irrespective of its behavior in the past) with the requirement of analyticity.
In practice, this point is usually ignored, and $`V(u+i\delta u)`$ simply represented by a Taylor series whose convergence is not questioned. We remark that if $`V(u)`$ is not analytic but Cauchyโs formula is used to provide a candidate definition for $`V(u+i\delta u)`$, the cโnumber term $`\widehat{T}_{ab}^{\mathrm{ren}}`$ becomes divergent; the theory is not renormalizable.
We shall not pursue this question of how or whether the standard renormalization is justified. Still, it is a point which is not really satisfatorily understood.
## III A FirstโQuantized Mirror
In order to estimate the effects of quantum fluctuations in the state of the mirror on the energy exchanges between it and the field, we must quantize the mirror. We shall consider a simple model, in which the mirror is considered to be heavy and its motion nonโrelativistic. Then the mirror may be treated as a firstโquantized particle. Let us begin by anticipating the limitations of this model.
(a) If the mirrorโs mass is $`m`$, then the model will only be valid for field modes of frequencies $`mc^2/\mathrm{}`$. The mass provides an effective ultraviolet cutโoff.
(b) The model can accurately predict dynamics only for a finite time. This is because eventually relativistic corrections to phases become significant. Correspondingly, there will be a limit to the accuracy of the energy levels predicted by the model.
(c) It will be most important to recall that a firstโquantized model is only valid at length scales greater than the Compton wavelength $`\mathrm{}/(mc)`$. At smaller length scales, attempts to measure the position of the particle require localized energies large enough that pair creation (here, of quanta of the โmirrorโ field) becomes nonโnegligible, and this precisely means that the firstโquantized model breaks down. This means that the position operator $`q`$ of the mirror only has a wellโdefined correspondence with physical reality on greater length scales.
(d) Even on the oneโparticle Hilbert space, relativistic corrections make the inner product $`\psi |\psi `$ nonโlocal with a length scale of order $`\mathrm{}/(mc)`$. This means that, as far as the oneโparticle model makes sense, the quantum observable $`q`$ always has a spread of at least the order of the Compton wavelength.
The general strategy will be to first consider the mirror as classical and moving in a specified external potential $`๐ฑ(q)`$, and then promote the mirrorโs position $`q`$ and momentum $`p`$ to quantum operators. The Hamiltonian of the mirror is just $`p^2/(2m)+๐ฑ(q)`$, so the main work involved is to compute the Hamiltonian of the field.
In fact, for the purposes of the present paper, it is only really necessary to compute the contributions to the vacuum energy part of the Hamiltonian: the normalโordered terms are not needed. Still, we shall give these terms, for the purpose of making clear just what the model is. The dynamical consequences of the terms will be investigated elsewhere.
We begin by working out the contributions to the field Hamiltonian at $`t=0`$ from the left and the right of the mirror. (The choice $`t=0`$ is of course conventional; other choices of time slice will be related by unitary transformations.) We have
$$\widehat{H}_{\genfrac{}{}{0pt}{}{\mathrm{right}}{\mathrm{left}}}=:\widehat{H}_{\genfrac{}{}{0pt}{}{\mathrm{right}}{\mathrm{left}}}:(12\pi )^1\ddot{q}.$$
(15)
Using the mirrorโs equation of motion, we will replace $`\ddot{q}`$ by $`(1/m)๐ฑ^{}`$. For the normalโordered terms, we have
$$:\widehat{H}_{\mathrm{right}}:=_{q(0)}^{\mathrm{}}:\left(\widehat{g}^{}(x)\right)^2+\left(V^{}(x)\widehat{g}^{}(V(x))\right)^2:dx,$$
(16)
with $`:\widehat{H}_{\mathrm{left}}:`$ given by a similar expression. We now reโwrite the contribution from the second term, in two steps. We have
$$_{q(0)}^{\mathrm{}}:\left(V^{}(x)\widehat{g}^{}(V(x))\right)^2:dx=_{\mathrm{}}^{q(0)}:\widehat{g}^{}(v)^2:V^{}dv.$$
(17)
Since $`V^{}`$ is a perturbation of unity, we split off a term where $`V^{}`$ is replaced by unity, and combine it with the first term in $`:\widehat{H}_{\mathrm{right}}:`$ to give the Hamiltonian of a free field (in the presence of a fixed mirror) plus a perturbation, which is
$`{\displaystyle _{\mathrm{}}^{q(0)}}`$ $`:\widehat{g}^{}(v)^2:(V^{}1)dv`$ (19)
$`=2{\displaystyle _{\mathrm{}}^0}:\widehat{g}^{}(t+q(t))^2:\dot{q}(t)dt.`$
Thus we have
$$:\widehat{H}_{\mathrm{right}}:=:\widehat{H}_{\mathrm{right},\mathrm{fixed}}:+:\widehat{H}_{\mathrm{right},\mathrm{pert}}:,$$
(20)
where
$$:\widehat{H}_{\mathrm{right},\mathrm{fixed}}:=_{\mathrm{}}^{\mathrm{}}:\widehat{g}^{}(x)^2:dx$$
(21)
and
$$:\widehat{H}_{\mathrm{right},\mathrm{pert}}:=2_{\mathrm{}}^0:\widehat{g}^{}(t+q(t))^2:\dot{q}(t)dt.$$
(22)
This term is already of order $`v/c`$. Thus we may compute $`q(t)`$ and $`\dot{q}(t)=p(t)/m`$ to the required accuracy from the mirrorโs Hamiltonian
$$\widehat{H}_{\mathrm{mirror}}=p^2/(2m)+๐ฑ.$$
(23)
Using this, choosing a Hermitian factorโordering, and abusing notation by keeping the same symbol for the Hamiltonian with quantized $`p`$ and $`q`$, we find
$`:\widehat{H}_{\mathrm{right},\mathrm{pert}}:=m^1{\displaystyle _{\mathrm{}}^0}e^{i\widehat{H}_{\mathrm{mirror}}t}`$ (24)
$`\times (:\widehat{g}^{}(t+q(0))^2:p(0)+p(0):\widehat{g}^{}(t+q(0))^2:)`$ (25)
$`\times e^{i\widehat{H}_{\mathrm{mirror}}t}dt.`$ (26)
This is the final expression for the normalโordered part of the correction to the freeโfield Hamiltonian in the model with firstโquantized mirror. As mentioned above, we do not really need this explicit form in what follows, but present it for the purposes of defining the model.
Before analyzing how passage to this model affects the paradoxes of the classical mirror, a few comments about the modelโs structure are in order. One can regard this model as a perturbation of a stationary classical mirror, the perturbation parameter being $`m^1`$. Adopting this point of view, one can ask how the eigenstates of the classical mirror are affected by taking into account its finite mass. The integrals over the halfโline in equation (24) will contain creation$``$creation and annihilation$``$annihilation operators, and these will result in a โdressingโ of the states. In particular, the vacuum state will be dressed with twoโparticle contributions. The mirror, too, will be affected by the operators $`p`$ and $`q`$; the dressing will contribute states which in the unperturbed theory would be excited.
## IV Measurement of the Vacuum Energy
We now take up the question of how well the vacuum energy on either side of the mirror can be measured, and to what extent those measurements are compatible with the treatment of the mirror as a classical object.
Throughout this section, we consider the measurement of $`\widehat{H}_{\mathrm{right}}`$. This means a measurement is made of the field energy on the entire halfโspace to the right of the mirror. (Of course, this is for many purposes an idealization. In many cases, one would consider the energyโcontent over a fixed region of space, and restrict the mirror to be on one side of that. However, such analyses are cumbersome and will not be attempted here.) We also assume that the field is initially in the vacuum state.
### A The Classical Model
In this subsection, we shall assume that the mirror is in a state which can be wellโmodeled by a classical trajectory. Thus we may assume that at any time $`t`$ the mirrorโs position and momentum may be measured to classical accuracies $`\mathrm{\Delta }q`$ and $`\mathrm{\Delta }p`$ which are larger than the spreads in the corresponding quantum observables. Then there is a classical limit to the accuracy to which the mirrorโs energy is known:
$$\mathrm{\Delta }H_{\mathrm{mirror}}\frac{p}{m}\mathrm{\Delta }p+๐ฑ^{}(q)\mathrm{\Delta }q.$$
(27)
In particular, the limit of the accuracy in the energy due to the classical uncertainty in position must satisfy
$$|\mathrm{\Delta }H_{\mathrm{mirror}}||๐ฑ^{}\mathrm{\Delta }q|.$$
(28)
However, note that $`\mathrm{\Delta }q`$ must be far larger than the mirrorโs Compton wavelength for the mirror to be in a classical state. Thus we have
$$|\mathrm{\Delta }H_{\mathrm{mirror}}|\widehat{H}_{\mathrm{right}}.$$
(29)
In other words, to the extent that the classical model of the mirrorโs trajectory is credible, the lack of accuracy in knowledge of the mirrorโs energy must be far larger than the vacuum energy in the field. This means that while the mirrorโs motion splits the vacuum into energy packets of opposite signs, the uncertainty in the energetic cost of this separation is far larger than the magnitude of the separation itself. Thus there is no detectable violation of the second law.
Note too that this means an attempt to consider a term like $`\widehat{H}_{\mathrm{right}}`$ as a semiclassical contribution to the mirrorโs energy is misguided. It is not of itself wrong, but it is a correction far below the scale at which any classical treatment of the mirror is valid.
### B The FirstโQuantized Model
In the previous subsection, we saw that energies of the scale $`\widehat{H}_{\mathrm{right}}`$ were far below contributions which could be meaningfully treated by a classical model of the mirror. This of course suggests that we must pass to a quantized mirror to understand the energyโtransfers between it and the field. I shall do so here, using the firstโquantized model, but I shall not attempt a full analysis of the problem. This is partly because of technical difficulties in the firstโquantized model (as I shall explain), but there is a deeper reason.
We saw in the previous subsection that the mirrorโs Compton wavelength entered in limiting the validity of the classical model. This length is the scale at which a firstโquantized treatment breaks down, so we may expect that even the firstโquantized model will be inadequate. This is indeed the case, as will be discussed below. However, the analysis of the firstโquantized mirror will uncover a new physical effect in the energetics, and so we take it up here.
We have $`\widehat{H}_{\mathrm{right}}=(12\pi m)^1๐ฑ^{}(q)`$. This means a measurement of the vacuum energy is essentially a measurement of $`q`$. (A strictly linear potential $`๐ฑ=`$const$`q`$ is excluded for several reasons. The most important of these is that the corresponding classical trajectories would not obey the boundary conditions necessary for the derivation of the formulas for $`\widehat{T}_{ab}^{\mathrm{ren}}`$.) A measurement of $`q`$ is always made with a quantum uncertainty, and insofar as the firstโquantized model is valid the spread in the quantum observable must be larger than the Compton wavelength:
$$\mathrm{\Delta }q\mathrm{}/(mc).$$
(30)
Note that while the symbol used ($`\mathrm{\Delta }q`$) is the same as in the previous subsection, the meaning here is different. Here $`\mathrm{\Delta }q`$ represents not just a lack of knowledge or of measurement resolution, but the spread of the components of the wave function with respect to the spectral resolution of the operator $`q`$.
The spread in the mirrorโs potential energy is
$$\mathrm{\Delta }๐ฑ๐ฑ^{}\mathrm{\Delta }q๐ฑ^{}(\mathrm{}/(mc)),$$
(31)
which is far larger than the vacuum energy. This suggests the relation
$$\mathrm{\Delta }E_{\mathrm{mirror}}\widehat{H}_{\mathrm{right}},$$
(32)
that is, the spread in the mirrorโs energy must be larger than the vacuum energy in the field. This would mean that the vacuum energy could not be usefully separated from the mirrorโs own energy, and indeed the vacuum energy would have to be considered as part of the constitutive energy of the mirror.
Of course, the relation (32) has not been established rigorously, because we have neglected possible cancellations between the spreads of the mirrorโs kinetic energy and its potential energy. A careful argument would seem to be technically very difficult, especially as we have made essentially no restriction on the potential. However, the physical conclusion โ that the spread in the mirrorโs energy is large compared to the vacuum energy โ seems suggestive eneough that it is worthwhile to raise as a generic possibility.
We can establish the relation (32) in the case of a mirror moving in a quadratic potential $`๐ฑ=kq^2/2`$. In this case, the vacuum energy is $`(12\pi m)^1kq`$, so a measurement of this is precisely a measurement of $`q`$. A reliable measurement of this energy therefore requires a measurement with nominal value $`\overline{q}`$ and spread $`\mathrm{\Delta }q`$ related by $`\mathrm{\Delta }q/\overline{q}<1`$. Now, since $`[\widehat{H}_{\mathrm{mirror}},q]=i\mathrm{}p/m`$ we must have $`\mathrm{\Delta }E\mathrm{\Delta }q\mathrm{}\overline{p}/m`$. On th other hand, we have $`\mathrm{\Delta }p\mathrm{}/\mathrm{\Delta }q`$ and thus $`\overline{p}\mathrm{}/\overline{q}`$. Thus the mirrorโs energy spread
$`\mathrm{\Delta }E`$ $`\mathrm{}/(m\overline{q}\mathrm{\Delta }q)`$ (37)
$`\mathrm{}/(m\overline{q}^2)`$
$`(\mathrm{}/m)(mc/\mathrm{})(1/\overline{q})`$
$`(\mathrm{}/m)(k\overline{q}^2/\mathrm{}c)(1/\overline{q})`$
$`=(\mathrm{}/mc)k\overline{q},`$
which is the vacuum energy.
### C Limitation of the FirstโQuantized Model
The treatment of the mirror as a firstโquantized particle is only accurate within certain regimes. An important limitation is that a real mirror does not reflect all frequencies perfectly, but becomes transparent to sufficiently high modes.
To understand this, let $`\omega _\mathrm{p}`$ be a โplasmaโ frequency, giving a scale beyond which the mirror becomes essentially transparent. Associated with this is a โskin depthโ $`c/\omega _\mathrm{p}`$ to which modes penetrate before being reflected. The mirrorโs position, as a reflecting surface, is not defined more accurately than this skin depth. This means that the model is only credible insofar as it depends on spatial resolutions $`c/\omega _\mathrm{p}`$. Now for a realistic mirror we must have
$$\mathrm{}\omega _\mathrm{p}mc^2,$$
(38)
that is, the plasma energy should be less than the rest energy, or equivalently, the skin depth should be much larger than the Compton wavelength. However, the vacuum energy on one side of the field is
$$\left|(12\pi )^1๐ฑ^{}\frac{\mathrm{}}{mc}\right|(12\pi )^1\left|๐ฑ\left(q+\frac{\mathrm{}}{mc}\right)๐ฑ(q)\right|$$
(39)
is a measurement of the difference of potential energies over the Compton wavelength $`\mathrm{}/(mc)c/\omega _\mathrm{p}`$. Thus this energy difference is well below the ambiguity in the mirrorโs Hamiltonian
$$\widehat{H}_{\mathrm{mirror}}=\frac{p^2}{2m}+๐ฑ.$$
(40)
We see that the firstโquantized model is not accurate enough to determine whether there are exchanges of energy between the mirror and the field of the same scale as the vacuum energy. The limitation is in the treatment of the mirror as perfectly reflecting, and the neglect of whatever internal physics of the mirror gives rise to that refelection. Presumably, an accurate model will require passing beyond this.
## V Summary and Implications
### A Summary
I have reโexamined the โmoving mirrorโ models of Davies and Fulling, giving attention to their limits of validity in computing energy transfers between the mirror and the vacuum energy of the field. Insofar as the mirror can be modeled by a classical point particle, we find that the lack of accuracy in its energy far exceeds the vacuum energy. This means that, while the motion of the mirror splits the vacuum into positive and negative energy packets, the magnitudes of those energies are far below the uncertainty in the mirrorโs energy. Thus no violation of the second law arises.
Moving beyond the classical model to firstโquantized mirror, we were able to show, at least in the case of a quadratic external potential, that the quantum spread in the mirrorโs energy must be greater than the fieldโs vacuum energy. This would mean that a measurement of the field energy would necessarily drive the mirror into a superposition of energy states, with width greater than the vacuum energy. For more general potentials, we gave suggestive but not conclusive arguments for the same conclusion.
These conclusions are consonant with the results of Parentani . With a different although related purpose, he investigated a secondโquantized mirror model with certain approximations, and found correlations between the mirror state and those of forwardโscattered quanta. Taken together, the two models show: (a) that when vacuum field energies are measured, the entanglement of the mirrorโs state with that of the field may be a dominant effect; and (b) the entanglement may involve fluctuations in the mirror energy grater than the fieldโs vacuum energy.
Even the model of the mirror by firstโquantized point particle turned out to be insufficiently accurate for quantitative analysis of the energy transfers between the mirror and the field. We found that in order to reliably compute mirror energies to a resolution of the order of the vacuum energy, one will need to take into account the finite reflectivity of the mirror, and its structure on scales of the order of its skin depth.
We found too evidence for another limitation on energy measurements, deeper than that set by the finite reflectivity. At every point where we used the skin depth to restrict the limits of measurability of energies, we also used the Compton wavelength $`\mathrm{}/(mc)c/\omega _\mathrm{p}`$ of the mirror. The Compton wavelength is the scale at which the firstโquantized, nonโrelativistic model of a particle (here, the mirror) breaks down (irrespective of its reflective properties). The appearance of this scale seems to indicate that in a deep way we must confront the infinitely many degrees of freedom of the mirror as a secondโquantized object, before we will be able to have a satisfactory understanding of the transfers of energy between it and the field.
These requirements to pass to a very deep model of the mirror in order to reliably study the energetics of the system must be considered a surprise, because for a long time it has been assumed that at least for sufficiently heavy mirrors a classical model should be valid. However, we see that not only does this fail, but even a firstโquantized mirror model is not sufficiently refined to make positive predictions.
### B Implications
These results have serious implications for attempts to understand the backโreaction of the quantum field on the mirror. Even the first conclusion, that such backโreaction effects are below the range of applicability of the classical model, is important. It shows that any attempt to treat these backโreaction effects semiclassically is misguided, because the backโreactions are far smaller than the scales at which the classical model can be trusted anyway.
The second result (that the spread in the mirrorโs energy must be greater than the vacuum energy) shows that the semiclassical model is not merely insufficiently refined: its basic assumption is wholly misdirected. A semiclassical approximation precisely assumes that the quantum fluctuations in the mirrorโs state are negligible: but the opposite is the case.
It is hoped that the features uncovered in this simple model will be guides to the analysis of more complicated and realistic field theories. In particular, a main motivation for this work was as a warmโup for an analysis of similar issues in the Hawking process, where a wholly convincing understanding of the backreaction has yet to be reached. |
warning/0001/hep-ph0001095.html | ar5iv | text | # I Introduction
## I Introduction
The fascinating possibility that the ground state of QCD at finite quark density can behave as a color superconductor has attracted much attention recently . Of particular interest is the symmetry breaking pattern that may arise for $`N_f=3`$ light quark flavors. As first shown by Alford, Rajagopal and Wilczek , the diquark condensates can lock the color and flavor symmetry transformations (Color-Flavor-Locking or CFL for brief). In the limit of massless quarks, these condensates spontaneously break both color and chiral symmetries. The eight gluons become massive through the Higgs mechanism while eight plus two Goldstone bosons are leftover. The two octets of states are analogous to the octets of vector and light pseudoscalar mesons in vacuum and Schรคfer and Wilczek have conjectured that there might exist some sort of continuity between the properties of QCD at zero density and in the CFL phase. The two other massless excitations are the Goldstone modes associated with the spontaneous breaking of the baryon number ($`U(1)_B`$) and axial ($`U(1)_A`$) symmetries. Because of the axial anomaly, the latter is not a true symmetry of QCD. However, because instantons effects are small at large densities, the associated mode can be treated as a true Goldstone mode as a first approximation. Although there are significant differences, to which we shall come back later, the situation at high density is analogous to considering the limit of large numbers of colors $`N_c`$ in vacuum , in which $`U(1)_A`$ breaking effects are $`1/N_c`$ suppressed.
For energies which are small compared to the gap of the superconducting CFL phase, the dynamics of these Goldstone modes is most conveniently described with the help of an effective lagrangian. As suggested by the symmetry breaking pattern in the CFL phase at large densities, this effective theory is analogous to Chiral Perturbation Theory ($`\chi `$PT) in the large $`N_c`$ limit of QCD in vacuum . For large densities, the leading order effective lagrangian invariant under $`U(3)_L\times U(3)_R`$ flavor symmetry has been constructed in some recent works . In particular, Son and Stephanov have shown how the parameters of the lagrangian can be computed at large densities by matching to the underlying microscopic theory. In the present note we make a first attempt to extend these works to lower density regimes, taking into account the effects of instanton-induced interactions. By using the power counting rules of $`\chi `$PT we construct the effective lagrangians up to order $`E^4`$ in an energy expansion. We make a particular emphasis on the meson spectrum and compute their masses as function of the quark chemical potential $`\mu `$. For moderate densities, $`\mu <\mathrm{\hspace{0.33em}10}^4`$ MeV, the gauge coupling grows large and instanton effects become important. We give numerical values for the masses of the mesons simply assuming that the analytical expressions for the gap and condensates, which have been computed at weak gauge coupling, also hold at strong coupling. Our calculations suggests that some of the Goldstone modes cannot exist as low energy excitations for values of $`\mu <1000`$ MeV, as their masses are very close to the unstability threshold $`2\mathrm{\Delta }`$, where $`\mathrm{\Delta }`$ is the gap in the CFL phase. These conclusions depend on the value of the color superconducting gap as estimated in the literature, and of a number of approximations that we have made.
In the next section, we write down the general low energy effective action for the pseudoscalar Goldstone modes up to $`E^4`$ in a low energy expansion and estimate the couplings that are relevant for the meson mass spectrum. In Sec. III we give some numbers for the meson masses at low densities, $`\mu <\mathrm{\hspace{0.33em}2500}`$ MeV and finally draw some conclusions.
## II Effective chiral lagrangian in the CFL phase
We follow Gatto and Casualboni and Son and Stephanov to construct the effective lagrangian for the low energy excitations of the CLF phase of QCD. The ground state is characterized by the two diquark condensates
$$X^{ia}ฯต^{ijk}ฯต^{abc}\psi _L^{bj}\psi _L^{ck}^{},Y^{ia}ฯต^{ijk}ฯต^{abc}\psi _R^{bj}\psi _R^{ck}^{},$$
(1)
where $`a,b,c`$ denote color indices, while $`i,j,k`$ refer to flavor ones. Under an $`SU(3)_c\times SU(3)_L\times SU(3)_R`$ the condensates transform as
$$XU_LXU_c^{},YU_RYU_c^{},$$
(2)
and as
$$Xe^{2i\alpha }e^{2i\beta }X,Ye^{2i\alpha }e^{2i\beta }Y,$$
(3)
under $`U(1)_A`$ and $`U(1)_B`$ transformations defined as
$$\psi _Le^{i(\alpha +\beta )}\psi _L,\psi _Re^{i(\alpha +\beta )}\psi _R.$$
(4)
One can factor out the norm of the condensates and consider the unitary matrices $`X`$ and $`Y`$. The slow variations of the phases of these matrices then correspond to the low energy excitations. Altogether these are $`9+9=18`$ degrees of freedom: $`8`$ will be absorbed by the gluons through the Higgs mechanism, which leaves $`10`$ true low energy excitations. At low energies the gluons decouple from the theory, as they are heavy degrees of freedom. It is convenient to collect all the Goldstone modes in the unitary matrix
$$\mathrm{\Sigma }=XY^{},$$
(5)
which is singlet of $`SU(3)_c`$ and $`U(1)_B`$ and transforms as
$$\mathrm{\Sigma }e^{4i\alpha }U_L\mathrm{\Sigma }U_R^{},$$
(6)
under $`SU(3)_L\times SU(3)_R\times U(1)_A`$. This leaves apart the low energy excitation that emerges from the spontaneous breaking of baryon number symmetry $`U(1)_B`$. Because baryon number is an exact global symmetry of QCD, this Goldstone mode is always massless in the CFL phase, independent of the quark masses. As our focus here is on the effect of chiral symmetry breaking by finite quark masses, to simplify our discussion we will simply drop this degree of freedom in the sequel.
To proceed we parametrise the unitary matrix $`\mathrm{\Sigma }`$ as
$$\mathrm{\Sigma }=\mathrm{exp}\left(i\frac{\mathrm{\Phi }}{f_\pi }\right),$$
(7)
where $`\mathrm{\Phi }=\varphi ^AT^A`$, $`A=1,\mathrm{}9`$. The $`T^A`$ for $`A=1,\mathrm{},8`$ are the Gell-Mann generators of $`SU(3)`$ and $`T^9=\sqrt{2/3}\mathbf{\hspace{0.17em}1}_3`$, all normalized to $`\mathrm{Tr}(T^AT^B)=2\delta ^{AB}`$. We use the same nomenclature as in vacuum for the nonet of Goldstone modes,
$$\mathrm{\Phi }=\left(\begin{array}{ccc}\pi _0+\frac{1}{\sqrt{3}}\left(\eta _8+\sqrt{2}\frac{f_\pi }{f_{\eta _0}}\eta _0\right)& \sqrt{2}\pi ^+& \sqrt{2}K^+\\ \sqrt{2}\pi ^{}& \pi _0+\frac{1}{\sqrt{3}}\left(\eta _8+\sqrt{2}\frac{f_\pi }{f_{\eta _0}}\eta _0\right)& \sqrt{2}K_0\\ \sqrt{2}K^{}& \sqrt{2}\overline{K}_0& \frac{1}{\sqrt{3}}\left(2\eta _8+\sqrt{2}\frac{f_\pi }{f_{\eta _0}}\eta _0\right)\end{array}\right).$$
(8)
In (8), $`f_\pi `$ and $`f_{\eta _0}`$ are the decay constants at finite density respectively of the octet of mesons and of the singlet $`\eta ^0`$. A priori there is no reason for these to be equal. By matching to the microscopic theory, Son and Stephanov have found
$$f_\pi ^2=\frac{218\mathrm{log}2}{18}\frac{\mu ^2}{2\pi ^2},f_{\eta _0}^2=\frac{3}{4}\frac{\mu ^2}{2\pi ^2},$$
(9)
where $`\mu `$ is the quark chemical potential. These expressions are valid at large densities (i.e. small gauge coupling). From (9), the ratio $`f_\pi /f_{\eta _0}1.07`$ is close to one. This result is reminiscent of the $`OZI`$ rule often invoked to hold in vacuum. At smaller densities, the gauge coupling grows large and the ratio could significantly depart from unity because of instanton effects.
The quark mass term in the microscopic lagrangian
$$\mathrm{\Delta }=\overline{\psi }_L\psi _R+h.c.$$
(10)
breaks chiral symmetry. Its effects can be introduced in the effective lagrangian by treating the mass matrix as an external field with vacuum expectation value
$$=\text{diag}(m_u,m_d,m_s),$$
(11)
where $`m_u,m_d`$ and $`m_s`$ refer to the up, down and strange quark masses, respectively. The spurion field then transforms as
$$e^{2i\alpha }U_L^{}U_R,$$
(12)
under $`SU(3)_L\times SU(3)_R\times U(1)_A`$.
To take into account the effects of the $`U(1)_A`$ anomaly, we also allow for the presence of the $`\theta `$ term in the microscopic lagrangian,
$$\mathrm{\Delta }_\theta =\theta \frac{g^2}{32\pi ^2}F_{\mu \nu }^A\stackrel{~}{F}^{\mu \nu ,A}.$$
(13)
As for the quark mass term, we will treat $`\theta `$ as an external field with vanishing expectation value which, to account for the variation of the quark measure, transforms as
$$\theta \theta 2N_f\alpha \theta 6\alpha $$
(14)
under $`U(1)_A`$ transformations. Then any arbitrary function of the combination
$$X=\theta \frac{i}{2}\mathrm{Tr}\mathrm{log}\mathrm{\Sigma }\theta +\sqrt{\frac{3}{2}}\frac{\eta _0}{f_{\eta _0}},$$
(15)
is invariant under $`SU(3)_L\times SU(3)_R\times U(1)_A`$ transformations.
With these ingredients, we are almost ready to construct the low energy effective lagrangian. One last issue is that of power counting. In QCD in vacuum, because $`M_\pi ^2m_q`$, the expansion is in powers of the pion external energy or momenta and quark masses and $`E^2p^2m_q`$. Similarly, at large $`N_c`$, $`U(1)_A`$ breaking effects are counted as $`E^21/N_c`$, because $`M_{\eta _0}^21/N_c`$. In the CFL phase of QCD on the other hand, the power counting depends very much on the density. At finite densities, instantons effects are screened $`(\mathrm{\Lambda }/\mu )^\alpha `$, where $`\mathrm{\Lambda }200`$ MeV is the scale of QCD, $`\mu `$ is the quark chemical potential and $`\alpha `$ is some positive constant which depends on the process under consideration. This has two consequences at large densities. The first is that $`U(1)_A`$ is essentially a good symmetry and the $`\eta _0`$ is approximately massless in the chiral limit. Then, because the two condensates break chiral symmetry only through the intermediate of color transformations, there is an approximate $`Z_2^L\times Z_2^R`$ symmetry which acts independently on the left and right-handed quarks . At high densities, this implies that the leading contribution to the mass of the Goldstone modes is $`๐ช(m_q^2)`$. The natural power counting rule at high densities is then $`E^2m_q^2`$ because $`M_\pi ^2m_q^2`$. At the moderate densities that could eventually be of interest for heavy ion collisions or for neutron stars, instantons effects are likely to be non-negligible. As $`Z_2^L\times Z_2^R`$ symmetry is broken to the diagonal $`Z_2`$ by instantons, a meson mass term $`M_\pi ^2m_q`$ is allowed . If instanton effects are dominant, the power counting is that of vacuum, $`E^2m_q`$. For convenience, we will adopt the counting rules of $`\chi `$PT as in vacuum and comment when necessary about the differences that arise in the CFL phase of QCD.
A final remark concerning the region of applicability of the energy expansion. In vacuum, the low energy expansion is valid for $`E^2<f_\pi ^2`$. In the CFL phase however, $`f_\pi \mu `$, which is independent and much larger than the gap ($`\mathrm{\Delta }\mu `$) at large densities (see (9)). However, the effective theory must break down for $`E2\mathrm{\Delta }`$, which is the energy necessary to excite one quasiparticle pair out of the superconducting ground state, and the low energy expansion is only valid for the more restrictive range $`E^2M_\pi ^2<\mathrm{\hspace{0.33em}4}\mathrm{\Delta }^2`$ There is a formal analogy between this behavior and that of $`\chi `$PT in vacuum in the large $`N_c`$ limit. A priori the expansion is valid for $`E<f_\pi `$. But because $`f_\pi ^2N_c\mathrm{\Lambda }^2`$ grows large as $`N_c\mathrm{}`$ and as something is bound to happen for $`E\mathrm{\Lambda }200`$ MeV, which is the scale of quark confinement, the low energy expansion is limited to $`E<\mathrm{\Lambda }`$..
At order $`E^2`$, the most general lagrangian compatible with the symmetries of the condensates is
$`_2`$ $`=`$ $`V_1(X)\mathrm{Tr}\left(_\mu \mathrm{\Sigma }^\mu \mathrm{\Sigma }^{}\right)+\{V_2(X)\mathrm{Tr}\left(\mathrm{\Sigma }^{}\right)e^{i\theta }+h.c.\}`$ (16)
$`+`$ $`V_3(X)\left(\mathrm{Tr}\mathrm{\Sigma }_\mu \mathrm{\Sigma }^{}\right)^2+V_4(X)\left(\mathrm{Tr}\mathrm{\Sigma }_\mu \mathrm{\Sigma }^{}\right)^\mu \theta +V_5(X)_\mu \theta ^\mu \theta .`$ (17)
We have written (16) using a compact notation. Since Lorentz invariance is broken at finite density (only a $`O(3)`$ symmetry is preserved) all the four-vectors should be split into temporal and spatial components, and the functions $`V_i`$ multiplying those spatial or temporal components are not forced to be the same by symmetry arguments. For example, the first term in (16) should read
$$V_1(X)\mathrm{Tr}\left(_\mu \mathrm{\Sigma }^\mu \mathrm{\Sigma }^{}\right)V_{1,t}(X)\mathrm{Tr}\left(_0\mathrm{\Sigma }_0\mathrm{\Sigma }^{}\right)V_{1,s}(X)\mathrm{Tr}\left(_i\mathrm{\Sigma }_i\mathrm{\Sigma }^{}\right).$$
(18)
The same situation occurs for the remaining terms and functions $`V_3,V_4,V_5`$ of $`X`$. All the couplings in (16) can be a priori arbitrary functions of $`X`$ (15). At large densities, $`U(1)_A`$ symmetry breaking effects are exponentially suppressed and the couplings only depend on the chemical potential $`\mu `$.
To reproduce the standard normalization of the meson kinetic terms, we impose
$$V_{1,t}(0)=\frac{f_\pi ^2}{4}.$$
(19)
The ratio $`V_{1,s}(0)/V_{1,t}(0)=v^2`$ is the velocity squared of the Goldstone bosons. At large densities, Son and Stephanov have found that $`v`$ is equal to the speed of sound $`1/\sqrt{3}`$ for all the low energy modes, including the baryon Goldstone mode.
The operators in (16) are not all independent. The last three terms can be transformed into each other with a field redefinition, $`\eta _0/f_{\eta _0}\eta _0/f_{\eta _0}+\kappa \theta `$. Using this freedom, we can choose to set $`V_4(0)`$ to zero. With this choice, the last operator becomes irrelevant for the meson spectrum and can be discarded.
The operator that is left, with coupling $`V_3(X)`$, contributes to the difference between $`f_\pi `$ and $`f_{\eta _0}`$. At high densities, using (9),
$$V_{3,t}(0)=\frac{f_\pi ^2f_{\eta _0}^2}{12}0.01f_\pi ^2,$$
(20)
which is small compared to $`f_\pi ^2`$. At moderate densities, $`V_{3,t}`$ could receive large contributions from instantons as is manifest from the mixing with $`V_4`$ and $`V_5`$.
Consider now the mass term in (16). This term is analogous to the leading mass term in $`\chi `$PT in vacuum. The only difference is the occurrence of the phase $`\theta `$, which is absent in vacuum. This is because at zero density the condensate that breaks chiral symmetry is $`\overline{\psi }_L\psi _R`$ which transforms like $`^{}`$ under $`SU(3)_L\times SU(3)_R\times U(1)_A`$. In (16) the presence of the $`\theta `$ in the effective lagrangian is the trademark of a one instanton effect; in the CFL phase, a one instanton process can be saturated by closing its six external quark legs with the insertion of one left-handed diquark, one right-handed diquark and one chiral condensate, $`(\psi _L\psi _L)(\overline{\psi }_R\overline{\psi }_R)(\overline{\psi }_R\psi _L)`$. As in vacuum, a non-zero chiral condensate $`\overline{\psi }\psi `$ leads to $`M_\pi ^2m_q\overline{\psi }\psi `$. Instanton effects are small and can be reliably computed at high densities $`\mu \mathrm{\Lambda }`$. In particular, Schรคfer has obtained
$`\overline{\psi }\psi `$ $`=`$ $`\overline{u}u+\overline{d}d+\overline{s}s`$ (21)
$``$ $`2\left({\displaystyle \frac{\mu ^2}{2\pi ^2}}\right){\displaystyle \frac{3\sqrt{2}\pi }{g(\mu )}}{\displaystyle \frac{18}{5}}G(\mu )\varphi _A^2(\mu ),`$ (22)
where $`g(\mu )`$ is the gauge coupling estimated at the scale $`\mu `$ using the one-loop beta function. The factor $`G(\mu )`$ is the one instanton weight integrated over all instanton sizes $`\rho `$, which are peaked around $`\rho \mu ^1`$ at finite density,
$$G(\mu )0.26\mathrm{\Lambda }^5((\beta _0\mathrm{log}(\mu /\mathrm{\Lambda }))^{2N_c}\left(\frac{\mathrm{\Lambda }}{\mu }\right)^{\beta _0+5}.$$
(23)
We have used the running of $`g(\mu )`$ at one-loop, $`\mathrm{\Lambda }=200`$ MeV and $`\beta _0=11/3N_c2/3N_f9`$ is the first coefficient of the QCD beta function. Finally, $`\varphi _A(\mu )\psi _L\psi _L\psi _R\psi _R`$ is the diquark condensate in the CFL phase of QCD,
$$\varphi _A(\mu )2\left(\frac{\mu ^2}{2\pi ^2}\right)\left(\frac{3\sqrt{2}\pi }{g(\mu )}\right)\mathrm{\Delta }(\mu ),$$
(24)
and
$$\mathrm{\Delta }(\mu )=b_0^{}512\pi ^4(2/N_f)^{5/2}g(\mu )^5\mu \mathrm{exp}\left(\frac{3\pi ^2}{\sqrt{2}g(\mu )}\right),$$
(25)
is the color superconducting gap. The factor $`b_0^{}`$ is unknown but expected to be $`๐ช(1)`$ . At large densities $`\mu \mathrm{\Lambda }`$, instantons are suppressed, $`\overline{\psi }\psi `$ goes to zero and the mass term of (16) vanishes. At the moderate densities that could be of interest for heavy ion collisions or neutron stars, this instanton effect is however not negligible. Using the Gell-Mann-Oakes-Renner relation, and $`\overline{\psi }\psi =\overline{u}u+\overline{d}d+\overline{s}s3\overline{u}u`$, we find
$$V_2(0)=\frac{1}{6}\overline{\psi }\psi .$$
(26)
At order $`E^4`$, there are a few more operators<sup>ยง</sup><sup>ยง</sup>ยงAt this order, one could also consider adding a Wess-Zumino-Witten term .,
$`_4`$ $`=`$ $`K_1(X)\left[\mathrm{Tr}\left(_\mu \mathrm{\Sigma }^\mu \mathrm{\Sigma }^{}\right)\right]^2+K_2(X)\mathrm{Tr}\left(_\mu \mathrm{\Sigma }_\nu \mathrm{\Sigma }^{}\right)\mathrm{Tr}\left(^\mu \mathrm{\Sigma }^\nu \mathrm{\Sigma }^{}\right)`$ (27)
$`+`$ $`K_3(X)\mathrm{Tr}\left(_\mu \mathrm{\Sigma }^\mu \mathrm{\Sigma }^{}_\nu \mathrm{\Sigma }^\nu \mathrm{\Sigma }^{}\right)`$ (28)
$`+`$ $`\{K_4(X)\mathrm{Tr}\left(_\mu \mathrm{\Sigma }^\mu \mathrm{\Sigma }^{}\right)\mathrm{Tr}\left(\mathrm{\Sigma }^{}\right)e^{i\theta }+h.c.\}+\{K_5(X)\mathrm{Tr}\left(_\mu \mathrm{\Sigma }^\mu \mathrm{\Sigma }^{}\mathrm{\Sigma }^{}\right)e^{i\theta }+h.c.\}`$ (29)
$`+`$ $`\{K_6(X)det(\mathrm{\Sigma })\mathrm{Tr}\left(\mathrm{\Sigma }^{}\right)\mathrm{Tr}\left(\mathrm{\Sigma }^{}\right)+h.c.\}+K_7(X)\mathrm{Tr}\left(\mathrm{\Sigma }^{}\right)\mathrm{Tr}\left(^{}\mathrm{\Sigma }\right)`$ (30)
$`+`$ $`\{K_8(X)det(\mathrm{\Sigma })\mathrm{Tr}\left(\mathrm{\Sigma }^{}\mathrm{\Sigma }^{}\right)+h.c.\}.`$ (31)
Again, we have used a compact notation for the terms in $`K_i`$, $`i=1\mathrm{}5`$ which should be split into temporal and spatial components.
Like the term $`V_2`$ in (16), the terms $`K_4`$ and $`K_5`$ depend explicitly on $`\theta `$ and so are exponentially suppressed at high densities. Similarly, at large chemical potential $`\mu `$, the remaining functions $`K_i(X)`$ reduce to constants $`K_i(X=0;\mu )`$. In this limit, the mass pattern of the Goldstone bosons will be determined by the couplings $`K_6`$, $`K_7`$ and $`K_8`$ in (27). As first shown by Son and Stephanov , at large densities, these coupling constants can be computed by matching to the underlying microscopic theory.
In principle, a systematic and non-ambiguous strategy to compute the coefficients of the effective theory is to use the background field technique , introducing external sources and symmetry breaking order parameters and integrating out the quark fields. At high densities, gluon exchange is suppressed and this amounts to a quark one-loop calculation, which would fix the coefficients of all the operators to arbitrary order in the meson fields. A slightly different strategy has been advocated in : set the meson fields to zero, $`\mathrm{\Sigma }=\mathrm{๐}_3`$ in the operators of (27), and compare the shift in ground state energy induced by non-zero quark masses in both the effective and microscopic theories. This approach is a priori ambiguous, as different operators could contribute to the shift in ground state energy. In practice, to order $`E^4`$, enough constraints can be derived to completely fix the couplings $`K_6`$, $`K_7`$ and $`K_8`$. The diagrams in the full microscopic theory are those of Fig.1. The rhs diagram can only contribute to $`K_6`$ and $`K_8`$. The lhs diagram can contribute to $`K_7`$. These couplings can be fixed by considering two different quark mass patterns,
$$_1=m\mathrm{๐}_3\text{and}_2=\mathrm{diag}(0,0,m_s)=\frac{m}{\sqrt{3}}\lambda ^8+\frac{m}{\sqrt{6}}\lambda ^9.$$
(32)
In the effective theory, these respective choices lead to the following shifts in ground state energy density
$`\mathrm{\Delta }\epsilon _1`$ $`=`$ $`\{(9m^2K_6+h.c.)+9m^2K_7+(3m^2K_8+h.c.)\},`$ (33)
$`\mathrm{\Delta }\epsilon _2`$ $`=`$ $`\{(m_s^2K_6+h.c.)+m_s^2K_7+(m_s^2K_8+h.c.)\}.`$ (34)
A special feature of a quark mass term is that it couples fermions and antifermions. At finite density, quark mass effects are then suppressed by the necessity to excite antiparticles with characteristic momentum $`2\mu `$ . A simple way to see this is to consider the effective theory for the fermion excitations near the surface of the Fermi sea. In momentum space, the free fermion lagrangian is
$$=\overline{\psi }(\gamma ^\mu p_\mu +\mu \gamma ^0)\psi m\overline{\psi }\psi .$$
(35)
If we consider the fermion mass term as a perturbation and decompose the fermions field $`\psi `$ into positive and negative energy components $`\psi _+`$ and $`\psi _{}`$, using the projectors
$$\mathrm{\Lambda }^\pm =\frac{1}{2}(1\pm \alpha \widehat{๐ช}),$$
(36)
and $`\mathrm{\Lambda }^\pm \psi _\pm =\psi _\pm `$, the lagrangian becomes
$$=\psi _+^{}(q_0|\stackrel{}{q}|+\mu )\psi _++\psi _{}^{}(q_0+|\stackrel{}{q}|+\mu )\psi _{}m(\psi _+^{}\psi _{}+\psi _{}^{}\psi _+).$$
(37)
This shows that the fermion mass term couples particles and antiparticles. If we integrate out the antifermions, we get at leading order
$$\psi _+^{}(q_0q_{})\psi _+\frac{m^2}{\mu }\psi _+^{}\psi _+$$
(38)
where $`q_{}|\stackrel{}{q}|\mu `$. This shows that quark mass effects are suppressed at high densities, as one could have expected. Note that the modified mass term $`m^2/\mu `$ has the same structure as a chemical potential term. This is a relevant operator which modifies the shape of the Fermi surface .
The calculation of the l.h.s. diagram of Fig. 1 gives
$$\frac{\mathrm{\Delta }\epsilon _1}{\mathrm{\Delta }\epsilon _2}=3\frac{m^2}{m_s^2},$$
(39)
to $`๐ช(\frac{m_q^2\mathrm{\Delta }^4}{\mu ^4})`$, where $`\mathrm{\Delta }`$ stands for generic gap and/or antigaps. This diagram simply gives a trivial shift of the vacuum energy, which in the effective theory corresponds to a constant (independent of $`\mathrm{\Sigma }`$) invariant operator
$$\mathrm{\Delta }\text{Tr}(^{}),$$
(40)
and, consequently, $`K_7=0`$.
Estimates of the r.h.s. diagram of Fig. 1 indicate that
$$K_{6,8}\frac{\mathrm{\Delta }\overline{\mathrm{\Delta }}}{\mu ^2}\mathrm{log}(\mathrm{\Delta }/\mu ),$$
(41)
where $`\mathrm{\Delta }`$ is a generic expression for a quark gap and $`\overline{\mathrm{\Delta }}`$ is an antigap. The dependence on antigaps arises from the structure of the r.h.s. diagram with both particles and antiparticles propagators. Unfortunately, the expression of the antigap is not known. Preliminary estimates have revealed that it is gauge-dependent . It is however a physical quantity (pole of the antiquark quasi-particles and holes), which should be gauge-invariant on quasi-particles mass-shell. At weak coupling, the antigap is presumably much smaller than the gap. Here, we will assume that $`K_{6,8}0`$ at large and moderate densities.
To be complete we should also include the contribution of instantons to the mass of the singlet meson $`\eta _0`$, which enters the effective lagrangian through a function of $`X`$ alone,
$$_0=V_0(X),$$
(42)
This piece is $`E^0`$ as it involves no derivatives of the meson fields. Expanding to second order in $`X`$ gives
$$M_{\eta _0}^2=\frac{3}{2}\frac{V_0^{^{\prime \prime }}(0)}{f_{\eta _0}^2}.$$
(43)
The constant $`V_0^{^{\prime \prime }}`$ (and more generally the whole function $`V(X)`$) could in principle be computed at large densities using instanton calculus. Unfortunately, the result is not known. However, at large to moderate densities we might expect this contribution to be suppressed.
For three flavors, the two remaining legs of a one instanton contribution to $`M_{\eta _0}^2`$ can only be closed with the insertion of either a quark mass term or a chiral condensate $`\overline{\psi }\psi `$. In the chiral limit, the latter only arises through another instanton process. On dimensional grounds we would expect
$$M_{\eta _0}^2\mu ^4G(\mu )|\overline{\psi }\psi |$$
(44)
in the chiral limit, but from such a rough estimate, it is not reasonable to infer whether $`M_{\eta _0}`$ ever grows large at low densities.
## III Meson mass pattern
In this section we give numerical estimates of the pseudoscalar mesons as function of the quark chemical potential.
At very large densities, instanton effects are suppressed: meson masses are linear in the quark masses and presumably very small. At lower densities, a non-zero $`\overline{\psi }\psi `$ condensate induced by instantons introduce contributions to the meson masses which are proportional to $`m_q^{1/2}`$. Here, we will estimate this last effect as it is dominant in the low density regime of the theory. We will use the expression of the chiral condensate and the gap as computed at weak coupling (21) and we will assume that the corrections are small even at moderate densities $`\mu 600`$ MeV.
The masses of the charged pions and kaons deduced from (16) read
$$M_{\pi ^\pm }^2=\frac{2V_2}{f_\pi ^2}\left(m_u+m_d\right),M_{K^\pm }^2=\frac{2V_2}{f_\pi ^2}\left(m_u+m_s\right),M_{K^0,\overline{K}^0}^2=\frac{2V_2}{f_\pi ^2}\left(m_d+m_s\right),$$
(45)
In the limit of no $`U(1)_A`$ breaking effects, the neutral mesons $`\pi ^0,\eta _8`$ and $`\eta _0`$ are strongly mixed. Their mass matrix reads
$$\frac{2V_2}{f_\pi ^2}\left(\begin{array}{ccc}m_u+m_d& \frac{m_um_d}{\sqrt{3}}& \frac{f_\pi }{f_{\eta _0}}\frac{\sqrt{2}(m_um_d)}{\sqrt{3}}\\ \frac{2(m_um_d)}{\sqrt{3}}& \frac{m_u+m_d+4m_s}{3}& \frac{f_\pi }{f_{\eta _0}}\frac{\sqrt{2}(m_u+m_d2m_s)}{\sqrt{3}}\\ \frac{f_\pi }{f_{\eta _0}}\frac{\sqrt{2}(m_um_d)}{\sqrt{3}}& \frac{f_\pi }{f_{\eta _0}}\frac{\sqrt{2}(m_u+m_d2m_s)}{\sqrt{3}}& \frac{f_\pi ^2}{f_{\eta _0}^2}\frac{2(m_u+m_d+m_s)}{3}\end{array}\right).$$
(46)
We have assumed that the contribution from the quark masses and the chiral condensate is parametrically larger than the two-instantons contribution to the mass of $`\eta _0`$. As the latter effect, if important, would increase the mass of $`\eta _0`$, our numerical estimates should be considered as lower bounds on the mass of $`\eta ^{}`$. Neglecting the difference between $`f_\pi `$ and $`f_{\eta _0}`$ we get the mass eigenvalues characteristic of ideal mixing,
$$M_{\overline{u}u}^2=\frac{2V_2}{f_\pi ^2}\mathrm{\hspace{0.17em}2}m_u,M_{\overline{d}d}^2=\frac{2V_2}{f_\pi ^2}\mathrm{\hspace{0.17em}2}m_d,M_{\overline{s}s}^2=\frac{2V_2}{f_\pi ^2}\mathrm{\hspace{0.17em}2}m_s.$$
(47)
To give an idea of the numerical values of the meson masses at $`\mu 600`$ MeV, we take $`m_u=4`$ MeV, $`m_d=7`$ MeV and $`m_s=150`$ and evaluate the value of the chiral condensate for $`\mathrm{\Lambda }=200`$ MeV and $`b_0^{}=1`$, to find
$`M_{\pi ^\pm }^230\mathrm{MeV},M_{K^\pm }^2112\mathrm{MeV},M_{K^0,\overline{K}^0}^2113\mathrm{MeV},`$ (48)
$`M_{\overline{u}u}^226\mathrm{MeV},M_{\overline{d}d}^234\mathrm{MeV},M_{\overline{s}s}^271\mathrm{MeV}.`$ (49)
A more accurate analysis of the meson masses would require to take into account terms which are proportional to $`m_q^2`$. However, in the low density regime, and due to the smallness of $`K_6`$ and $`K_8`$, these represent negligeable corrections to the above estimates.
Because the expression for the chiral condensate involves the square of the gap, which is only known up to a factor $`b_0^{}`$ assumed to be $`๐ช(1)`$ , there is some theoretical uncertainty in the meson masses at low densities and the values quoted are at most indicative. At this order, this uncertainty can be cancelled by considering the mass over gap ratios. These ratios are independent of the gap and, moreover, tell us whether the mesons can exist as low energy excitations in the CFL phase at chemical potential $`\mu `$. The ratios of the charged and neutral meson masses over two times the gap are shown in Figs. 4 and 5. For the strange particles, the ratio is larger than one or very close to the instability threshold unless $`\mu >1`$ GeV, which suggests that these particles do not exist as stable low energy excitations at low densities.
## IV Conclusions
We have considered the effective lagrangian for the Goldstone modes in the Color-Flavor-Locking phase of QCD at high densities, up to $`E^4`$ in a low energy expansion $`E<\mathrm{\hspace{0.33em}2}\mathrm{\Delta }`$. A tentative analysis of the meson mass pattern that emerges from this lagrangian, including instanton effects, suggests that the kaons and the pure strange neutral meson may be absent from the spectrum for moderate densities. Their masses are significantly larger than two times the gap for $`\mu <\mathrm{\hspace{0.33em}1}`$ GeV. For the pions, the situation is much better as there their masses are significantly smaller than the instability threshold for all densities.
Obviously, there is room for improvements. Our approach has been far from systematic and we have made some ad hoc assumptions. Although the effects are likely to be small at moderate densities, it is of much interest to gain a better understanding of the quadratic quark mass effects . A calculation of the contribution of instantons to the mass of $`\eta _0`$ in the chiral limit would also prove to be most useful. If a spontaneously broken, approximate $`U(2)_L\times U(2)_R`$ flavor symmetry survives in the CFL phase of QCD at moderate densities, it could also be interesting to determine the potential for the $`\eta _0`$, $`V_0(X)`$. The effective theory for the low energy excitations should be very much analogous to the large $`N_c`$ lagrangian of Di Vecchia and Veneziano and Witten .
## Acknowledgements
We are grateful to Deog Ki Hong, Stephen Hsu, Thomas Schรคfer, Rob Pisarski and Dirk Rischke for useful comments. |
warning/0001/hep-th0001009.html | ar5iv | text | # Gluon Pair Production From Space-Time Dependent Chromofield
## I Introduction
Over the years there have been several investigations on the production of charged particles from a classical electro-magnetic field, a phenomenon which was discovered nearly fifty years ago by Schwinger . By now, the production of electron-positron pairs from the abelian field is extensively studied both theoretically and experimentally . The subject of quark/anti-quark and gluon pair production from the non-abelian field is relatively new and is not fully solved. It might be important for the production of the quark-gluon plasma (QGP) in the laboratory by high energy heavy-ion collisions. Lattice QCD predicts the existence of such a state of matter at high temperatures ($``$ 200 MeV) and densities . In high energy heavy-ion collisions at RHIC and LHC the receding nuclei might produce a strong chromofield which would then polarize the QCD vacuum and produce quark/anti-quark pairs and gluons. These produced quarks and gluons collide with each other to form a thermalized quark-gluon plasma. The space-time evolution of the quark-gluon plasma in the presence of a background chromofield is studied by solving relativistic non-abelian transport equation of quarks and gluons with all the dynamical effects taken into account. As color is a dynamical variable in the non-abelian theory, the relativistic non-abelian transport equation for quark and gluon is
$$\left[p_\mu ^\mu +gQ^aF_{\mu \nu }^ap^\nu _p^\mu +gf^{abc}Q^aA_\mu ^bp^\mu _Q^c\right]f(x,p,Q)=C(x,p,Q)+S(x,p,Q).$$
(1)
Here $`f(x,p,Q)`$ is the single-particle distribution function of the parton in the 14 dimensional extended phase space which includes coordinate, momentum and color in SU(3). The first term on the LHS of Eq. (1) corresponds to the usual convective flow, the second term is the non-Abelian version of the Lorentz-force term and the third term corresponds to the precession of the color charge, as described by Wongโs equation . $`C`$ and $`S`$ on the RHS of Eq. (1) are the collision and the source terms, respectively. Note that there are separate transport equations for quarks, antiquarks and gluons . The source term $`S`$ contains the detailed information about the production of quarks and gluons from the chromofield. It is defined as the probability W for the parton production per unit time per unit volume of the phase space. Hence the space-time evolution of the quark-gluon plasma in a high energy heavy-ion collision crucially depends on how quarks and gluons are produced from the chromofield .
The production of fermion pairs from the field is studied in two different cases: 1) from a constant, uniform electric field and 2) from a space-time dependent field. The fermion pair production from a constant, uniform field is computed by Schwinger. This is an exact one-loop nonperturbative result . This result can also be understood in terms of a semiclassical tunneling across the mass gap . However, a space-time dependent field $`A`$ can directly excite the negative energy particles to levels above the mass gap, by a perturbative mechanism $`Aq\overline{q}`$, without recourse to any tunneling or barrier penetration mechanism . The particle production from this mechanism is important in high energy heavy-ion collisions. As shown by numerical studies , the chromofield acquires a space-time dependence as soon as it starts producing partons and so Schwingerโs non-perturbative formula for particle production from a constant field is not applicable. In this case the treatment of parton production from a space-time dependent chromofield is necessary. It can be mentioned here that, so far the transport equation is solved with parton productions from a constant field taken into account because gluon pair production from the space-time dependent chromofield is not computed. Aim of this paper is to calculate the probability for the process $`Agg`$ which is similar to the process $`Aq\overline{q}`$. Quark and gluon production from a space-time dependent chromofield is needed to study the production and equilibration of a quark-gluon plasma in ultra relativistic heavy-ion collisions at RHIC and LHC.
The production of $`q\overline{q}`$ pairs from a space-time dependent non-abelian field is almost the same to that of the production of $`e^+e^{}`$ pairs from the abelian field. This is because, apart from the color factors, the interaction of the quantized Dirac field with the classical potential is the same in both cases. All the methods used to obtain the probability for the production of the $`e^+e^{}`$ pair from a space-time dependent Maxwell field can be applied to obtain the probability for the production of the $`q\overline{q}`$ pair from a space-time dependent Yang-Mills field. On the other hand, the production of gluon pairs from a space-time dependent Yang-Mills field is not straight forward and there is no counter part to this in the abelian theory. In contrast to abelian theory the quantized Yang-Mills field interacts with the classical non-abelian potential. Because of this interaction there is gluon production from the QCD vacuum in the presence of an external chromofield. This phenomenon is absent in the Maxwell theory.
We again mention here that the fermion pair production from the space-time dependent field is studied in the literature but gluon production from space-time dependent chromofield is not studied so far. This is because a consistent theory involving the interaction of the gluons with the classical chromofield is not available in the conventional theory of QCD. In the case of fermions one knows exactly the theory for the interaction of fermions with the classical field. This is given by the Dirac equation (see section II). However, a consistent theory of gluons in the presence of external classical chromofield is not available in the conventional QCD. This problem is addressed properly within the background field method of QCD which was introduced by DeWitt and โt Hooft. We will briefly mention the main differences between the background field method of QCD and the conventional method of QCD in section III. Unlike in conventional QCD, the Feynman diagrams obtained in the background field method contain classical chromofields and gluons. We will use the Feynman rules derived in the background field method of QCD to obtain the probability for the processes $`Aq\overline{q},gg`$ via vacuum polarization. The study of the production and evolution of the quark-gluon plasma at RHIC and LHC by solving the relativistic non-abelian transport equation with parton production from a space-time dependent chromofield will be undertaken in the future.
This paper is organized as follows. We compute the probability for the process $`Aq\overline{q}`$ in section II. The probability for the process $`Agg`$ is computed in section III. A brief description about future research and conclusions can be found in section IV.
## II Probability for the process $`A_{cl}Q\overline{Q}`$
The production of $`q\overline{q}`$ pairs from a non-abelian field via vacuum polarization is simillar to that of the production of $`e^+e^{}`$ pairs from the abelian field. This is because the interaction lagrangian of the quantized Dirac field with the classical gauge potential is similar in both the cases. The amplitude for the lowest order process $`Ae^+e^{}`$ which contributes to the production of $`e^+e^{}`$ pair from a space-time dependent classical abelian field $`A^\mu `$ is given by
$$M=<k_1,k_2|S^{(1)}|0>=ie\overline{u}(k_1)\gamma _\mu A^\mu (K)v(k_2).$$
(2)
In the above expression $`k_1`$, $`k_2`$ are the four momenta of the produced electron and positron respectively and $`A(K)`$ is the Fourier transform of the space-time dependent field $`A^\mu (x)`$ with $`K=k_1+k_2`$.
The probability for pair production is:
$$W^{(1)}=\frac{d^3k_1}{(2\pi )^32k_1^0}\frac{d^3k_2}{(2\pi )^32k_2^0}d^4K(2\pi )^4\delta ^{(4)}(Kk_1k_2)T$$
(3)
where
$$T=\mathrm{\Sigma }_{spin}|M|^2.$$
(4)
Performing the spin sum one obtains
$`W_{e^+e^{}}^{(1)}={\displaystyle \frac{\alpha }{3}}{\displaystyle _{K^2>4m_e^2}}d^4K(1{\displaystyle \frac{4m_e^2}{K^2}})^{1/2}(1+{\displaystyle \frac{2m_e^2}{K^2}})[KA(K)KA(K)`$ (5)
$`K^2A(K)A(K)].`$ (6)
In the above expression $`\alpha =\frac{e^2}{4\pi }`$ is the coupling constant. Using
$$F^{\mu \nu }(K)=i(K^\mu A^\nu K^\nu A^\mu )$$
(7)
one obtains
$$W_{e^+e^{}}^{(1)}=\frac{\alpha }{3}_{K^2>4m_e^2}d^4K(1\frac{4m_e^2}{K^2})^{1/2}(1+\frac{2m_e^2}{K^2})[|E(K)|^2|B(K)|^2].$$
(8)
This is the probability for the production of the $`e^+e^{}`$ pair from a space-time dependent abelian field. This result was obtained for the first time by Schwinger (see Eq. (6.33) in ).
Now we proceed to compute the probability for the production of a $`q\overline{q}`$ pair from a space-time dependent non-abelian field. The amplitude for the lowest order process $`Aq\overline{q}`$ which contributes to $`q\overline{q}`$ pair production from a space-time dependent classical non-abelian field $`A^{a\mu }`$ is given by
$$M=ig\overline{u}^i(k_1)\gamma _\mu T_{ij}^aA^{a\mu }(K)v^j(k_2).$$
(9)
In the above expression $`k_1`$ and $`k_2`$ are the four momenta of the quark and anti-quark respectively and $`A^{a\mu }(K)`$ is the Fourier transform of the space-time dependent non-abelian field $`A^{a\mu }(x)`$ with $`K=k_1+k_2`$. $`T_{ij}^a`$ are the generators in the fundamental representation, $`i,j`$ are the color indices for the quarks and $`a`$ is the color index of the non-abelian field.
Following the above procedure and choosing SU(3) gauge group one obtains
$`W_{q\overline{q}}^{(1)}=tr[T^aT^b]{\displaystyle \frac{\alpha _s}{3}}{\displaystyle _{K^2>4m_q^2}}d^4K(1{\displaystyle \frac{4m_q^2}{K^2}})^{1/2}(1+{\displaystyle \frac{2m_q^2}{K^2}})[KA^a(K)KA^b(K)`$ (10)
$`K^2A^a(K)A^b(K)].`$ (11)
In the above expression $`\alpha _s=\frac{g^2}{4\pi }`$. Using $`tr[T^aT^b]=\frac{1}{2}\delta ^{ab}`$ one obtains
$`W_{q\overline{q}}^{(1)}={\displaystyle \frac{\alpha _s}{6}}{\displaystyle _{K^2>4m_q^2}}d^4K(1{\displaystyle \frac{4m_q^2}{K^2}})^{1/2}(1+{\displaystyle \frac{2m_q^2}{K^2}})[KA^a(K)KA^a(K)`$ (12)
$`K^2A^a(K)A^a(K)].`$ (13)
This is the probability for the production of a $`q\overline{q}`$ pair from a space-time dependent chromofield to the lowest order in the coupling constant $`\alpha _s`$. It can be noted that Eq. (13) is valid only for a single flavor of quarks. This equation is simillar to Eq. (6) in the abelian theory.
## III Probability for the process $`Agg`$
Now we proceed to compute the probability for the process $`A_{cl}gg`$ via vacuum polarization. The amplitude for this process (see Fig. 1(a)) is given by
$$M=ฯต^{b\nu }(k_1)V_{\mu \nu \lambda }^{abc}(K,k_1,k_2)A^{a\mu }(K)ฯต^{c\lambda }(k_2).$$
(14)
In the above expression $`k_1`$, $`k_2`$ are the four momenta of the produced gluons, $`A^{a\mu }(K)`$ is the Fourier transform of the non-abelian field $`A^{a\mu }(x)`$ and $`V_{\mu \nu \lambda }^{abc}(K,k_1,k_2)`$ is the vertex involving a single classical field and two gluons with $`K=k_1+k_2`$. In the conventional method of QCD such vertices which involve classical field and gluons are not calculated. Hence, the gluon production from a space-time dependent chromofield is not available. However, such a calculation is possible in the background field method of QCD which was introduced by DeWitt and โt Hooft . This is because the Feynman diagrams involving a classical chromofield and gluons are obtained in the background field method of QCD. We use the Feynman rules obtained by the background field method of QCD to compute the probability for the process $`A_{cl}gg`$ via vacuum polarization. First of all we will briefly describe the main differences between conventional QCD and the background field method of QCD before studying the above process.
In the conventional theory of QCD the generating function is
$$Z[J]=[dA]๐etM_Gexp(i[S[A]\frac{1}{2\alpha }GG+JA]),$$
(15)
where we have not included the quark part for simplicity. In the above expression the gauge field action is
$$S[A]=\frac{1}{4}d^4x(F^{a\mu \nu })^2$$
(16)
with
$$F^{a\mu \nu }=^\mu A^{a\nu }^\nu A^{a\mu }+gf^{abc}A^{b\mu }A^{c\nu }$$
(17)
for the group with structure constants $`f^{abc}`$. The other two terms are
$$JA=d^4xJ_\mu ^aA^{a\mu }$$
(18)
and
$$GG=d^4xG^aG^a$$
(19)
with $`G^a`$ being the gauge fixing term. A typical choice for the gauge fixing term in QCD is
$$G^a=_\mu A^{a\mu }.$$
(20)
The matrix element of $`(M_G(x,y))`$ is given by
$$(M_G(x,y))^{ab}=\frac{\delta (G^a(x))}{\delta \theta ^b(y)}$$
(21)
where $`\frac{\delta (G^a(x))}{\delta \theta ^b(y)}`$ is the derivative of the gauge fixing term under the infinitesimal gauge transformation
$$\delta A^{a\mu }=f^{abc}\theta ^bA^{c\mu }+\frac{1}{g}^\mu \theta ^a.$$
(22)
The $`detM_G`$ in the generating functional is written as functional integral over the Faddeev-Popov ghost field $`\chi ^a`$. We mention here that any physical quantity calculated in this method is gauge invariant and independent of the particular gauge chosen. However, there is still a problem of gauge invariance with some of the quantities like off-shell Greenโs functions or divergent counter terms. This problem arises because in order to quantize the theory one must fix a gauge. This means that the total lagrangian we actually use in conventional QCD, consisting of the classical lagrangian (which is explicitly gauge invariant) plus a gauge-fixing and ghost terms, is not gauge invariant. The background field method is a technique which allows one to fix a gauge in quantizing the theory without losing explicit gauge invariance which is present at the classical level of the gauge field theory. In the background field approach, one arranges things so that explicit gauge invariance is still present once gauge fixing and ghost terms are added. In this way Greenโs functions obey the naive Ward identities of gauge invariance and the unphysical quantities like divergent counterterms becomes gauge invariant. However, for our purpose of gluon productions from the space-time dependent chromofield to the leading order we do not need to address all these issues here. The details about the explicit gauge invariance of QCD in the presence of background chromofield can be found in . In our study we require Feynman rules involving the classical chromofield and gluons. For this reason we briefly outline the procedure of background field method in QCD and present the generating functional which generates the Feynman rules involving gluons, ghosts and the classical chromofields.
In the background field method of QCD the gauge fixing term is given by
$$G^a=^\mu A_q^{a\mu }+gf^{abc}A_{cl}^{b\mu }A_q^{c\mu },$$
(23)
which depends on the classical background field $`A_{cl}^{a\mu }`$. The variable of the integration in the functional integral is the quantum gauge field $`A_q^{a\mu }`$ and, following โt Hooft the background field is not coupled to the external source $`J`$. The generating functional depends on $`J`$ and $`A_{cl}`$ and is given by
$$Z[J,A_{cl}]=[dA_q]๐etM_Gexp(i[S[A_q+A_{cl}]\frac{1}{2\alpha }GG+JA_q]).$$
(24)
The matrix element of $`M_G`$ is given by
$$(M_G(x,y))^{ab}=\frac{\delta (G^a(x))}{\delta \theta ^b(y)}$$
(25)
where $`\frac{\delta (G^a(x))}{\delta \theta ^b(y)}`$ is the derivative of the gauge fixing term under the infinitesimal gauge transformation
$$\delta A_q^{a\mu }=f^{abc}\theta ^b(A_q^{c\mu }+A_{cl}^{c\mu })+\frac{1}{g}^\mu \theta ^a.$$
(26)
Writing $`detM_G`$ as functional integral over the ghost field one finds
$`Z[J,A_{cl},\xi ,\xi ^{}]={\displaystyle }[dA_q][d\chi ][d\chi ^{}]exp(i[S[A_q+A_{cl}]+S_{ghost}{\displaystyle \frac{1}{2\alpha }}GG+JA_q`$ (27)
$`+\chi ^{}\xi +\xi ^{}\chi ]).`$ (28)
where $`\xi `$ and $`\xi ^{}`$ are source functions for the ghosts and
$`S_{ghost}={\displaystyle }d^4x\chi _a^{}[\mathrm{}^2\delta ^{ab}g\stackrel{}{}_\mu f^{abc}(A_{cl}^{c\mu }+A_q^{c\mu })+gf^{abc}A^{c\mu }\stackrel{}{}_\mu `$ (29)
$`+g^2f^{ace}f^{edb}A_{cl\mu }^c(A_{cl}^{d\mu }+A_q^{d\mu })]\chi _b.`$ (30)
The Feynman rules for QCD in the presence of a classical background chromofield are constructed from the generating functional which is given by Eq. (28) with the gauge fixing term given by Eq. (23). We use the Feynman rules obtained in the background field method to compute the probability for the process $`A_{cl}gg`$. Within this method the vertex $`V_{\mu \nu \lambda }^{abc}(K,k_1,k_2)`$, involving a single classical chromofield and two gluons which appear in the amplitude (See Fig. 1(a) and Eq. (14)) in Feynman gauge is given by
$$V_{\mu \nu \lambda }^{abc}(K,k_1,k_2)=gf^{abc}[2g_{\mu \nu }K_\lambda 2g_{\mu \lambda }K_\nu g_{\nu \lambda }k_\mu ],$$
(31)
where $`K=k_1+k_2`$ and $`k=k_1k_2`$. The above vertex is different from the three gluon vertex usually used in conventional QCD.
Using Eq. (14) in (4) we find
$$T_{gl}=\frac{1}{4}\mathrm{\Sigma }_{spin}A^{a\mu }(K)A^{a^{}\mu ^{}}(K)V_{\mu \nu \lambda }^{abc}(K,k)V_{\mu ^{}\nu ^{}\lambda ^{}}^{a^{}b^{}c^{}}(K,k)ฯต^{b\nu }(k_1)ฯต^{b^{}\nu ^{}}(k_1)ฯต^{c\lambda }(k_2)ฯต^{c^{}\lambda ^{}}(k_2),$$
(32)
where we have used $`A^{}(K)=A(K)`$. The factor $`\frac{1}{4}`$ is the weight factor given in the Feynman amplitude in order to obtain a correct gauge field and gauge-fixing action. To obtain correct and physical results we have to use the appropriate projection operators for the transverse polarization states of the gluons. For this purpose we proceed as follows. First of all we use the polarization sum
$$\mathrm{\Sigma }_{spin}ฯต_1^\nu ฯต_1^\nu ^{}=\mathrm{\Sigma }_{spin}ฯต_2^\nu ฯต_2^\nu ^{}=g^{\nu \nu ^{}}$$
(33)
and then substract the corresponding ghost contributions (see Fig. 1(b)).
Using Eq. (31), (33) and performing the calculation we find
$`T_{gl}={\displaystyle \frac{1}{4}}`$ $`A^{a\mu }(K)A^{a^{}\mu ^{}}(K)V_{\mu \nu \lambda }^{abc}(K,k)V_{\mu ^{}\nu ^{}\lambda ^{}}^{a^{}bc}(K,k)g^{\nu \nu ^{}}g^{\lambda \lambda ^{}}=Ng^2[4k_1k_2A^a(K)A^a(K)`$ (36)
$`3k_1A^a(K)k_2A^a(K)3k_2A^a(K)k_1A^a(K)k_1A^a(K)k_1A^a(K)`$
$`k_2A^a(K)k_2A^a(K)],`$
where we have used $`f^{abc}f^{a^{}bc}=N\delta ^{aa^{}}`$ in SU(N) gauge group.
Now we proceed to compute the probability for the process $`A_{cl}gg`$. To obtain the probability for the above process we require that $`K^2=(k_1+k_2)^2>0`$ with $`K^0>0`$. We recall that for real gluons $`k_1^2=k_2^2=0`$. With the above requirements we proceed to perform the integration in Eq. (3) using $`T_{gl}`$ from Eq. (36). As gluons are similar particles we multiply a factor $`\frac{1}{2}`$ in the phase space and find
$`W_{gl}^{(1)}={\displaystyle \frac{1}{4}}{\displaystyle \frac{Ng^2}{32\pi ^2}}{\displaystyle _{K^2>0}}d^4K\theta (K^0){\displaystyle \frac{d^3k_1}{k_1^0}\frac{d^3k_2}{k_2^0}\delta ^{(4)}(Kk_1k_2)A_\mu ^a(K)A_\nu ^a(K)}`$ (37)
$`[8g^{\mu \nu }K^28K^\mu K^\nu +4k^\mu k^\nu ]`$ (38)
where $`k=k_1k_2`$.
Now, we evaluate the corresponding ghost diagrams which have to be substracted from the above result. The amplitude for the ghost part (see Fig. $`1_b`$) is given by
$$M_{gh}^{bc}=gf^{abc}k^\mu A_\mu ^a.$$
(39)
so that
$$T_{gh}=\frac{1}{4}M_{gh}^{bc}M_{gh}^{bc}{}_{}{}^{}$$
(40)
Using Eq. (40) in (3) we find
$`W_{gh}^{(1)}={\displaystyle \frac{1}{4}}{\displaystyle \frac{Ng^2}{16\pi ^2}}{\displaystyle _{K^2>0}}d^4K\theta (K^0){\displaystyle \frac{d^3k_1}{k_1^0}\frac{d^3k_2}{k_2^0}\delta ^{(4)}(Kk_1k_2)A_\mu ^a(K)A_\nu ^a(K)[k^\mu k^\nu ]}`$ (41)
Substracting Eq. (41) from (38) we find
$`W_{Agg}^{(1)}={\displaystyle \frac{1}{4}}{\displaystyle \frac{Ng^2}{32\pi ^2}}{\displaystyle _{K^2>0}}d^4K\theta (K^0){\displaystyle \frac{d^3k_1}{k_1^0}\frac{d^3k_2}{k_2^0}\delta ^{(4)}(Kk_1k_2)A_\mu ^a(K)A_\nu ^a(K)}`$ (42)
$`[8g^{\mu \nu }8K^\mu K^\nu +2k^\mu k^\nu ]`$ (43)
To perform the integration in the above equation we proceed as follows. First of all using
$$\frac{d^3k_2}{2k_2^0}=d^4k_2\delta (k_2^2)\theta (k_2^0)$$
(44)
we check that
$$\frac{d^3k_1}{k_1^0}\frac{d^3k_2}{k_2^0}\delta ^{(4)}(Kk_1k_2)g^{\mu \nu }=2\pi g^{\mu \nu }.$$
(45)
Using Eq. (45) we find
$`{\displaystyle \frac{d^3k_1}{k_1^0}\frac{d^3k_2}{k_2^0}\delta ^{(4)}(Kk_1k_2)k_1^\mu k_2^\nu }={\displaystyle \frac{d^3k_1}{k_1^0}\frac{d^3k_2}{k_2^0}\delta ^{(4)}(Kk_1k_2)k_2^\mu k_1^\nu }`$ (46)
$`={\displaystyle \frac{\pi }{6}}[K^2g^{\mu \nu }+2K^\mu K^\nu ],`$ (47)
and
$`{\displaystyle \frac{d^3k_1}{k_1^0}\frac{d^3k_2}{k_2^0}\delta ^{(4)}(Kk_1k_2)k_1^\mu k_1^\nu }={\displaystyle \frac{d^3k_1}{k_1^0}\frac{d^3k_2}{k_2^0}\delta ^{(4)}(Kk_1k_2)k_2^\mu k_2^\nu }`$ (48)
$`={\displaystyle \frac{\pi }{6}}[K^2g^{\mu \nu }+4K^\mu K^\nu ].`$ (49)
With the help of the above relations the integration in Eq. (43) can be easily performed. Using (45), (47) and (49) we obtain from Eq. (43)
$$W_{Agg}^{(1)}=\frac{11N\alpha _s}{24}_{K^2>0}d^4K\theta (K^0)[K^2A^a(K)A^a(K)KA^a(K)KA^a(K)]$$
(50)
In the above expression the repeated indices are summed from $`a=1,\mathrm{}(N^21)`$. For SU(3) gauge group we obtain
$$W_{Agg}^{(1)}=\frac{11\alpha _s}{8}_{K^2>0}d^4K\theta (K^0)[K^2A^a(K)A^a(K)KA^a(K)KA^a(K)]$$
(51)
This is the probability computed from the process $`Agg`$ via vacuum polarization. It can be seen that the above expression is transverse with respect to the momentum of the field $`K`$, i.e, when $`A^\mu `$ is replaced by $`K^\mu `$ we get $`W_{Agg}^{(1)}=0`$. This result is a part of the expression for the total probability (including higher order terms in $`gA`$) for the production of gluon pairs.
## IV Conclusions
Using the background field method of QCD, we have computed the probability for the processes $`Aq\overline{q},gg`$ via vacuum polarization, in the presence of a space-time dependent chromofield $`A`$. These processes are similar to $`Ae^+e^{}`$ in QED. For any arbitary classical non-abelian chromofield one has to check the gauge invariance of the result with respect to the general non-abelian local gauge transformation which is known as type I gauge transformation . Under this gauge transformation the classical chromofield transform like:
$$A_{cl}^\mu UA_{cl}^\mu U^1\frac{i}{g}(_\mu U)U^1.$$
(52)
The gluonic and ghost fields transform like:
$$A_q^\mu UA_q^\mu U^1,$$
(53)
and
$$\chi ^{}U\chi U^1$$
(54)
respectively.
In these situations one has to add the results of the higher order processes (for the gluon pair production) to the leading order results obtained in this paper. The gluon pair production amplitude which is obtained from the first order $`S`$ matrix contains higher order terms in $`gA`$. This is due to the non-abelian nature of the gluonic and classical chromofield. The result of the gluon pair production probability which contains both leading order and higher order terms in $`gA`$ (but first order in $`S`$ matrix) will be reported elsewhere . In this paper we have presented only the result from the leading order processes $`Agg`$. This result is a part of the expression for the total probability of the gluon production from a space-time dependent chromofield . Quark and gluon production from a space-time dependent chromofield will play a crucial role in the production and equilibration of the quark-gluon plasma in ultra relativistic heavy-ion collsions. Our main goal is to study the production and equilibration of the quark-gluon plasma by solving relativistic non-abelian transport equations for partons (see Eq. (1)) with parton production from a space-time dependent chromofield taken into account. Such work requires extensive numerical computation and needs a separate publication.
###### Acknowledgements.
We thank Dennis D. Dietrich for useful discussions, reading of the manuscript and drawing the Feynman diagrams. We also thank Dr. Qun Wang and Dr. Chung-Wen Kao for useful discussions. G.C.N. acknowledges the financial support from Alexander von Humboldt Foundation. |
warning/0001/hep-ex0001037.html | ar5iv | text | # HIGGS SEARCHES AT LEP2 WITH THE ALEPH DETECTOR
## 1 Introduction
A neutral Higgs boson is required to complete the particle spectrum of the standard model. Fits to electroweak data $`^\mathrm{?}`$ from LEP, SLC and Tevatron indicate that if the Higgs particle exists it has most probably a mass around 100 GeV$`/c^2`$. The situation at LEP, which is summarized in Fig. 1 (left), shows that we are getting more and more sensitive to this mass range.
All mass exclusion limits given in this report are at 95% confidence level. To get an idea of the sensitivity of a specific analysis the expected mass exclusion limit $`^\mathrm{?}`$ from gedanken experiments usually is also quoted. All the search analyses use cut-based selections and/or neural network methods. We make extensive use of Monte Carlo simulations for the neural net training and the setting of the selection cuts. No details of the preliminary analyses are given, however, the references always point to the corresponding latest published analysis.
## 2 The Standard Model Higgs Boson
At LEP2, the dominant Higgs production mechanism is the so-called Higgs-strahlung process $`\mathrm{e}^+\mathrm{e}^{}`$$``$ Z$``$ HZ. There are also minor contributions from processes where the Higgs boson is generated through the fusion of two electroweak gauge bosons. In the mass range accessible at LEP2, the Higgs boson decays in 92% of the cases into two quark jets (91% of which are $`\mathrm{b}\overline{\mathrm{b}}`$) and in about 8% of the cases into $`\tau ^+\tau ^{}`$. Combining these numbers with the branching fractions of the Z boson we obtain the branching fractions of HZ into the following four characteristic event topologies:
* four jet channel: $`\mathrm{q}^{}\overline{\mathrm{q}}^{}`$$`\mathrm{q}\overline{\mathrm{q}}`$ (64.6%)
* missing energy channel: H$`\nu \overline{\nu }`$ (20.0%)
* lepton channel: H$`\mathrm{}^+\mathrm{}^{}`$, $`\mathrm{}`$ = e or $`\mu `$ (6.7%)
* $`\tau `$โchannel: either H or Z decays to $`\tau ^+\tau ^{}`$ (8.7%)
Except for the lepton channel, a high b-quark content is required for the quark jets coming from Higgs boson decay. In all channels the invariant mass of the decay products of the Z must be compatible with $`m_\mathrm{Z}`$. The results of the HZ analyses $`^\mathrm{?}`$ applied to data taken with the ALEPH detector $`^\mathrm{?}`$$`^,`$$`^\mathrm{?}`$ at 189 GeV are summarized in Table 1.
The analyses for the four individual channels are optimized in such a way that the combination of these analyses gives the best global HZ searches analysis. The definition used here for โbest analysisโ is: the analysis which gives the highest expected Higgs mass exclusion limit under the hypothesis that there is no signal $`^\mathrm{?}`$.
No signal is observed in 250 $`\mathrm{pb}^1`$ of data (cf. $`m_\mathrm{h}`$ distribution in Fig. 2) which allows to exclude a standard model Higgs boson with a mass less than 90.4 GeV$`/c^2`$, as shown in Fig. 2 (right). The expected limit is 93.4 GeV$`/c^2`$.
## 3 Neutral Higgs Bosons of the MSSM
The particle spectrum of the Higgs sector of the minimal supersymmetric extension of the standard model (MSSM) consists of five physical states, two CP-even neutral bosons h and H (with mixing angle $`\alpha `$ and masses $`m_\mathrm{h}`$ $`<`$ $`m_\mathrm{H}`$), one CP-odd neutral boson A and two charged bosons H<sup>ยฑ</sup>. At tree level, the Higgs sector can be parameterized by two independent parameters, e.g. $`m_\mathrm{h}`$ and $`\mathrm{tan}\beta `$ = $`v_2/v_1`$, the ratio of the vacuum expectation values of the two Higgs doublets. Only the h and A bosons are within the reach of LEP2. They are expected to be produced in Z decays via the Higgs-strahlung process Z$``$ hZ, with a cross section proportional to $`\mathrm{sin}^2(\beta \alpha )`$, and via the associated pair production Z$``$ hA, with a cross section proportional to $`\mathrm{cos}^2(\beta \alpha )`$.
These two processes are complementary, in the sense that if one cross section is maximal the other is minimal and vice versa. Thus both processes must be searched for. For the first process, the results from the hZ searches (Sec. 2) can be used. The second process is searched for in the following two decay channels:
* hA $``$ $`\mathrm{b}\overline{\mathrm{b}}`$$`\mathrm{b}\overline{\mathrm{b}}`$ (85%)
* hA $``$ $`\mathrm{b}\overline{\mathrm{b}}`$$`\tau ^+\tau ^{}`$, $`\tau ^+\tau ^{}`$$`\mathrm{b}\overline{\mathrm{b}}`$ (15%)
The performance of the two analyses $`^\mathrm{?}`$ is reported in Table 2.
No signal is observed in the data, as can be seen in Fig. 3 (left). This allows to set an upper limit on the cross section for hA production as a function of $`m_\mathrm{h}`$, which implies an upper limit on $`\mathrm{cos}^2(\beta \alpha )`$ as a function of $`m_\mathrm{h}`$. On the other hand, the hZ searches interpreted in this way result in an upper limit on $`\mathrm{sin}^2(\beta \alpha )`$ as a function of $`m_\mathrm{h}`$. Combining the hZ and hA searches in an optimal way $`^\mathrm{?}`$ leads to an excluded region in the two-dimensional parameter space $`[`$$`m_\mathrm{h}`$, $`\mathrm{sin}^2(\beta \alpha )]`$, which usually is translated to the $`[`$$`m_\mathrm{h}`$, $`\mathrm{tan}\beta ]`$ plane, as shown in Fig. 3 (right). For $`\mathrm{tan}\beta 1`$ we find that ALEPH data up to centre-of-mass energies of 189 GeV exclude the h and A Higgs bosons of the MSSM with masses less than 80.8 and 81.2 GeV$`/c^2`$, respectively, i.e. the neutral Higgs bosons must be heavier than the W boson if $`\mathrm{tan}\beta 1`$.
## 4 Charged Higgs Bosons
In the MSSM charged Higgs bosons are heavier than W bosons. This restriction does not hold for general two doublet models, which are very attractive theoretically because of the absence of flavor changing neutral currents and the relation $`m_\mathrm{W}`$ = $`m_\mathrm{Z}`$ $`\mathrm{cos}\theta _W`$ holding at tree level. The H<sup>ยฑ</sup> has the same decay modes as the W<sup>ยฑ</sup>, but since its coupling is proportional to the charged fermion masses, it dominantly decays into the heaviest energetically allowed fermion pair of the quark and lepton families. Whether H<sup>ยฑ</sup> decay preferentially into quarks or leptons depends on other parameters of the model. Therefore, the search for pair-produced H<sup>+</sup>H<sup>-</sup> is performed in the following three channels:
* leptonic channel: $`\tau ^+\nu _\tau \tau ^{}\overline{\nu }_\tau `$
* mixed channel: $`\mathrm{c}\overline{\mathrm{s}}\tau ^{}\overline{\nu }_\tau `$
* hadronic or four jet channel: $`\mathrm{c}\overline{\mathrm{s}}\mathrm{s}\overline{\mathrm{c}}`$
The performance of these analyses $`^\mathrm{?}`$ is summarized in Table 3.
The search for charged Higgs bosons in the three final states $`\tau ^+\nu _\tau \tau ^{}\overline{\nu }_\tau `$, $`\mathrm{c}\overline{\mathrm{s}}\tau ^{}\overline{\nu }_\tau `$ and $`\mathrm{c}\overline{\mathrm{s}}\mathrm{s}\overline{\mathrm{c}}`$ has been performed using 175 $`\mathrm{pb}^1`$ of ALEPH data collected at $`\sqrt{s}`$ = 189 GeV. No evidence of Higgs boson production was found (Fig. 4 left) and mass limits were set as a function of the branching ratio $`(\mathrm{H}^+\tau ^+\nu _\tau )`$. The result of the combination of the three analyses is displayed in Fig. 4 (right) where the curves corresponding to expected and observed confidence levels of 95% exclusion are drawn. As can be seen from this figure, charged Higgs bosons with masses below 62.5 GeV$`/c^2`$ are excluded at 95% C.L. independently of $`(\mathrm{H}^+\tau ^+\nu _\tau )`$, where the expected mass exclusion limit is 68.5 GeV$`/c^2`$.
## 5 Invisible Higgs Boson Decays
Many extensions of the standard model allow for the Higgs boson to decay invisibly, e.g. into a pair of lightest neutralinos when the neutralino $`\chi `$ is light enough. Since the Higgs boson is produced through the Higgs-strahlung process hZ, this leads to the following two event topologies:
* a pair of acoplanar leptons, when Z $``$ $`\mathrm{e}^+\mathrm{e}^{}`$ or Z $``$ $`\mu ^+\mu ^{}`$
* a pair of acoplanar jets, when the Z decays hadronically
where the acoplanarity is defined as the azimuthal angle between the two lepton or jet directions. Two analyses $`^\mathrm{?}`$ are designed for the two channels. When they are applied to ALEPH data taken at $`\sqrt{s}`$ = 189 GeV, 33 candidates are found, in agreement with 33.6 events expected from all background processes (cf. Table 4). The distributions of the reconstructed Higgs boson masses are shown in Fig. 5 (left).
Quite generally, the production cross section for invisible Higgs boson decay can be parameterized as $`\xi ^2\sigma _{\mathrm{SM}}`$($`\mathrm{e}^+\mathrm{e}^{}`$$``$ hZ), where $`\xi ^2`$ is a model dependent factor ranging from 0 to 1. One way of presenting the result of the negative searches is to calculate the 95% C.L. level upper limit on the production cross section of the invisibly decaying Higgs boson in units of $`\sigma _{\mathrm{SM}}`$($`\mathrm{e}^+\mathrm{e}^{}`$$``$ hZ) as a function of $`m_\mathrm{h}`$, which is shown in Fig. 5 (right).
## 6 Summary
In this report, an overview has been given of the present status of Higgs boson searches with the ALEPH detector. The results are mainly based on the analysis of data taken at 189 GeV centre-of-mass energy. When the data taken at lower energies are included in the analyses the 95% confidence level exclusion limits improve slightly.
For the standard model Higgs boson we find $`m_\mathrm{H}`$ $`>`$ 90.4 GeV$`/c^2`$ at 95% C.L. using 250 $`\mathrm{pb}^1`$ of data taken at $`\sqrt{s}`$ = 161, 172, 183 and 189 GeV. This means that a Higgs boson with a mass equal to the Z mass just cannot be excluded at 95% C.L. with ALEPH data taken up to the year 1998.
Including all lower energy data, for the neutral Higgs bosons of the MSSM we obtain the limits $`m_\mathrm{h}`$ $`>`$ 80.8 GeV$`/c^2`$ and $`m_\mathrm{A}`$ $`>`$ 81.2 GeV$`/c^2`$ (valid for all values of $`\mathrm{tan}\beta 1`$).
From the analysis of 175 $`\mathrm{pb}^1`$ of ALEPH data collected in 1998, the charged Higgs bosons of general two doublet models are excluded below 62.5 GeV$`/c^2`$ independent of their decay mode.
Finally, using the same data, invisibly decaying Higgs bosons are searched for. For a production cross section equal to that of the standard model Higgs boson, i.e. $`\xi ^2=1`$, masses below 92.8 GeV$`/c^2`$ are excluded.
## Acknowledgments
I would like to thank all my colleagues from the ALEPH Higgs Task Force for the continuous effort to produce all these results.
## References |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.